1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * from nfs_vnops.c 8.16 (Berkeley) 5/27/95 35 */ 36 37 #include <sys/cdefs.h> 38 /* 39 * vnode op calls for Sun NFS version 2, 3 and 4 40 */ 41 42 #include "opt_inet.h" 43 44 #include <sys/param.h> 45 #include <sys/kernel.h> 46 #include <sys/systm.h> 47 #include <sys/resourcevar.h> 48 #include <sys/proc.h> 49 #include <sys/mount.h> 50 #include <sys/bio.h> 51 #include <sys/buf.h> 52 #include <sys/extattr.h> 53 #include <sys/filio.h> 54 #include <sys/jail.h> 55 #include <sys/malloc.h> 56 #include <sys/mbuf.h> 57 #include <sys/namei.h> 58 #include <sys/socket.h> 59 #include <sys/vnode.h> 60 #include <sys/dirent.h> 61 #include <sys/fcntl.h> 62 #include <sys/lockf.h> 63 #include <sys/stat.h> 64 #include <sys/sysctl.h> 65 #include <sys/signalvar.h> 66 67 #include <vm/vm.h> 68 #include <vm/vm_extern.h> 69 #include <vm/vm_object.h> 70 #include <vm/vnode_pager.h> 71 72 #include <fs/nfs/nfsport.h> 73 #include <fs/nfsclient/nfsnode.h> 74 #include <fs/nfsclient/nfsmount.h> 75 #include <fs/nfsclient/nfs.h> 76 #include <fs/nfsclient/nfs_kdtrace.h> 77 78 #include <net/if.h> 79 #include <netinet/in.h> 80 #include <netinet/in_var.h> 81 82 #include <nfs/nfs_lock.h> 83 84 #ifdef KDTRACE_HOOKS 85 #include <sys/dtrace_bsd.h> 86 87 dtrace_nfsclient_accesscache_flush_probe_func_t 88 dtrace_nfscl_accesscache_flush_done_probe; 89 uint32_t nfscl_accesscache_flush_done_id; 90 91 dtrace_nfsclient_accesscache_get_probe_func_t 92 dtrace_nfscl_accesscache_get_hit_probe, 93 dtrace_nfscl_accesscache_get_miss_probe; 94 uint32_t nfscl_accesscache_get_hit_id; 95 uint32_t nfscl_accesscache_get_miss_id; 96 97 dtrace_nfsclient_accesscache_load_probe_func_t 98 dtrace_nfscl_accesscache_load_done_probe; 99 uint32_t nfscl_accesscache_load_done_id; 100 #endif /* !KDTRACE_HOOKS */ 101 102 /* Defs */ 103 #define TRUE 1 104 #define FALSE 0 105 106 extern struct nfsstatsv1 nfsstatsv1; 107 extern int nfsrv_useacl; 108 extern int nfscl_debuglevel; 109 MALLOC_DECLARE(M_NEWNFSREQ); 110 111 static vop_read_t nfsfifo_read; 112 static vop_write_t nfsfifo_write; 113 static vop_close_t nfsfifo_close; 114 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *, 115 struct thread *); 116 static vop_lookup_t nfs_lookup; 117 static vop_create_t nfs_create; 118 static vop_mknod_t nfs_mknod; 119 static vop_open_t nfs_open; 120 static vop_pathconf_t nfs_pathconf; 121 static vop_close_t nfs_close; 122 static vop_access_t nfs_access; 123 static vop_getattr_t nfs_getattr; 124 static vop_setattr_t nfs_setattr; 125 static vop_read_t nfs_read; 126 static vop_fsync_t nfs_fsync; 127 static vop_remove_t nfs_remove; 128 static vop_link_t nfs_link; 129 static vop_rename_t nfs_rename; 130 static vop_mkdir_t nfs_mkdir; 131 static vop_rmdir_t nfs_rmdir; 132 static vop_symlink_t nfs_symlink; 133 static vop_readdir_t nfs_readdir; 134 static vop_strategy_t nfs_strategy; 135 static int nfs_lookitup(struct vnode *, char *, int, 136 struct ucred *, struct thread *, struct nfsnode **); 137 static int nfs_sillyrename(struct vnode *, struct vnode *, 138 struct componentname *); 139 static vop_access_t nfsspec_access; 140 static vop_readlink_t nfs_readlink; 141 static vop_print_t nfs_print; 142 static vop_advlock_t nfs_advlock; 143 static vop_advlockasync_t nfs_advlockasync; 144 static vop_getacl_t nfs_getacl; 145 static vop_setacl_t nfs_setacl; 146 static vop_advise_t nfs_advise; 147 static vop_allocate_t nfs_allocate; 148 static vop_deallocate_t nfs_deallocate; 149 static vop_copy_file_range_t nfs_copy_file_range; 150 static vop_ioctl_t nfs_ioctl; 151 static vop_getextattr_t nfs_getextattr; 152 static vop_setextattr_t nfs_setextattr; 153 static vop_listextattr_t nfs_listextattr; 154 static vop_deleteextattr_t nfs_deleteextattr; 155 static vop_lock1_t nfs_lock; 156 157 /* 158 * Global vfs data structures for nfs 159 */ 160 161 static struct vop_vector newnfs_vnodeops_nosig = { 162 .vop_default = &default_vnodeops, 163 .vop_access = nfs_access, 164 .vop_advlock = nfs_advlock, 165 .vop_advlockasync = nfs_advlockasync, 166 .vop_close = nfs_close, 167 .vop_create = nfs_create, 168 .vop_fsync = nfs_fsync, 169 .vop_getattr = nfs_getattr, 170 .vop_getpages = ncl_getpages, 171 .vop_putpages = ncl_putpages, 172 .vop_inactive = ncl_inactive, 173 .vop_link = nfs_link, 174 .vop_lock1 = nfs_lock, 175 .vop_lookup = nfs_lookup, 176 .vop_mkdir = nfs_mkdir, 177 .vop_mknod = nfs_mknod, 178 .vop_open = nfs_open, 179 .vop_pathconf = nfs_pathconf, 180 .vop_print = nfs_print, 181 .vop_read = nfs_read, 182 .vop_readdir = nfs_readdir, 183 .vop_readlink = nfs_readlink, 184 .vop_reclaim = ncl_reclaim, 185 .vop_remove = nfs_remove, 186 .vop_rename = nfs_rename, 187 .vop_rmdir = nfs_rmdir, 188 .vop_setattr = nfs_setattr, 189 .vop_strategy = nfs_strategy, 190 .vop_symlink = nfs_symlink, 191 .vop_write = ncl_write, 192 .vop_getacl = nfs_getacl, 193 .vop_setacl = nfs_setacl, 194 .vop_advise = nfs_advise, 195 .vop_allocate = nfs_allocate, 196 .vop_deallocate = nfs_deallocate, 197 .vop_copy_file_range = nfs_copy_file_range, 198 .vop_ioctl = nfs_ioctl, 199 .vop_getextattr = nfs_getextattr, 200 .vop_setextattr = nfs_setextattr, 201 .vop_listextattr = nfs_listextattr, 202 .vop_deleteextattr = nfs_deleteextattr, 203 }; 204 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops_nosig); 205 206 static int 207 nfs_vnodeops_bypass(struct vop_generic_args *a) 208 { 209 210 return (vop_sigdefer(&newnfs_vnodeops_nosig, a)); 211 } 212 213 struct vop_vector newnfs_vnodeops = { 214 .vop_default = &default_vnodeops, 215 .vop_bypass = nfs_vnodeops_bypass, 216 }; 217 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops); 218 219 static struct vop_vector newnfs_fifoops_nosig = { 220 .vop_default = &fifo_specops, 221 .vop_access = nfsspec_access, 222 .vop_close = nfsfifo_close, 223 .vop_fsync = nfs_fsync, 224 .vop_getattr = nfs_getattr, 225 .vop_inactive = ncl_inactive, 226 .vop_pathconf = nfs_pathconf, 227 .vop_print = nfs_print, 228 .vop_read = nfsfifo_read, 229 .vop_reclaim = ncl_reclaim, 230 .vop_setattr = nfs_setattr, 231 .vop_write = nfsfifo_write, 232 }; 233 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops_nosig); 234 235 static int 236 nfs_fifoops_bypass(struct vop_generic_args *a) 237 { 238 239 return (vop_sigdefer(&newnfs_fifoops_nosig, a)); 240 } 241 242 struct vop_vector newnfs_fifoops = { 243 .vop_default = &default_vnodeops, 244 .vop_bypass = nfs_fifoops_bypass, 245 }; 246 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops); 247 248 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, 249 struct componentname *cnp, struct vattr *vap); 250 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 251 int namelen, struct ucred *cred, struct thread *td); 252 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, 253 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp, 254 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td); 255 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp, 256 struct componentname *scnp, struct sillyrename *sp); 257 258 /* 259 * Global variables 260 */ 261 SYSCTL_DECL(_vfs_nfs); 262 263 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO; 264 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 265 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 266 267 static int nfs_prime_access_cache = 0; 268 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW, 269 &nfs_prime_access_cache, 0, 270 "Prime NFS ACCESS cache when fetching attributes"); 271 272 static int newnfs_commit_on_close = 0; 273 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW, 274 &newnfs_commit_on_close, 0, "write+commit on close, else only write"); 275 276 static int nfs_clean_pages_on_close = 1; 277 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW, 278 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close"); 279 280 int newnfs_directio_enable = 0; 281 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW, 282 &newnfs_directio_enable, 0, "Enable NFS directio"); 283 284 int nfs_keep_dirty_on_error; 285 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW, 286 &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned"); 287 288 /* 289 * This sysctl allows other processes to mmap a file that has been opened 290 * O_DIRECT by a process. In general, having processes mmap the file while 291 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow 292 * this by default to prevent DoS attacks - to prevent a malicious user from 293 * opening up files O_DIRECT preventing other users from mmap'ing these 294 * files. "Protected" environments where stricter consistency guarantees are 295 * required can disable this knob. The process that opened the file O_DIRECT 296 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not 297 * meaningful. 298 */ 299 int newnfs_directio_allow_mmap = 1; 300 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW, 301 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens"); 302 303 static uint64_t nfs_maxalloclen = 64 * 1024 * 1024; 304 SYSCTL_U64(_vfs_nfs, OID_AUTO, maxalloclen, CTLFLAG_RW, 305 &nfs_maxalloclen, 0, "NFS max allocate/deallocate length"); 306 307 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \ 308 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \ 309 | NFSACCESS_DELETE | NFSACCESS_LOOKUP) 310 311 /* 312 * SMP Locking Note : 313 * The list of locks after the description of the lock is the ordering 314 * of other locks acquired with the lock held. 315 * np->n_mtx : Protects the fields in the nfsnode. 316 VM Object Lock 317 VI_MTX (acquired indirectly) 318 * nmp->nm_mtx : Protects the fields in the nfsmount. 319 rep->r_mtx 320 * ncl_iod_mutex : Global lock, protects shared nfsiod state. 321 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list. 322 nmp->nm_mtx 323 rep->r_mtx 324 * rep->r_mtx : Protects the fields in an nfsreq. 325 */ 326 327 static int 328 nfs_lock(struct vop_lock1_args *ap) 329 { 330 struct vnode *vp; 331 struct nfsnode *np; 332 u_quad_t nsize; 333 int error, lktype; 334 bool onfault; 335 336 vp = ap->a_vp; 337 lktype = ap->a_flags & LK_TYPE_MASK; 338 error = VOP_LOCK1_APV(&default_vnodeops, ap); 339 if (error != 0 || vp->v_op != &newnfs_vnodeops) 340 return (error); 341 np = VTONFS(vp); 342 if (np == NULL) 343 return (0); 344 NFSLOCKNODE(np); 345 if ((np->n_flag & NVNSETSZSKIP) == 0 || (lktype != LK_SHARED && 346 lktype != LK_EXCLUSIVE && lktype != LK_UPGRADE && 347 lktype != LK_TRYUPGRADE)) { 348 NFSUNLOCKNODE(np); 349 return (0); 350 } 351 onfault = (ap->a_flags & LK_EATTR_MASK) == LK_NOWAIT && 352 (ap->a_flags & LK_INIT_MASK) == LK_CANRECURSE && 353 (lktype == LK_SHARED || lktype == LK_EXCLUSIVE); 354 if (onfault && vp->v_vnlock->lk_recurse == 0) { 355 /* 356 * Force retry in vm_fault(), to make the lock request 357 * sleepable, which allows us to piggy-back the 358 * sleepable call to vnode_pager_setsize(). 359 */ 360 NFSUNLOCKNODE(np); 361 VOP_UNLOCK(vp); 362 return (EBUSY); 363 } 364 if ((ap->a_flags & LK_NOWAIT) != 0 || 365 (lktype == LK_SHARED && vp->v_vnlock->lk_recurse > 0)) { 366 NFSUNLOCKNODE(np); 367 return (0); 368 } 369 if (lktype == LK_SHARED) { 370 NFSUNLOCKNODE(np); 371 VOP_UNLOCK(vp); 372 ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK); 373 ap->a_flags |= LK_EXCLUSIVE; 374 error = VOP_LOCK1_APV(&default_vnodeops, ap); 375 if (error != 0 || vp->v_op != &newnfs_vnodeops) 376 return (error); 377 if (vp->v_data == NULL) 378 goto downgrade; 379 MPASS(vp->v_data == np); 380 NFSLOCKNODE(np); 381 if ((np->n_flag & NVNSETSZSKIP) == 0) { 382 NFSUNLOCKNODE(np); 383 goto downgrade; 384 } 385 } 386 np->n_flag &= ~NVNSETSZSKIP; 387 nsize = np->n_size; 388 NFSUNLOCKNODE(np); 389 vnode_pager_setsize(vp, nsize); 390 downgrade: 391 if (lktype == LK_SHARED) { 392 ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK); 393 ap->a_flags |= LK_DOWNGRADE; 394 (void)VOP_LOCK1_APV(&default_vnodeops, ap); 395 } 396 return (0); 397 } 398 399 static int 400 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td, 401 struct ucred *cred, u_int32_t *retmode) 402 { 403 int error = 0, attrflag, i, lrupos; 404 u_int32_t rmode; 405 struct nfsnode *np = VTONFS(vp); 406 struct nfsvattr nfsva; 407 408 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag, 409 &rmode); 410 if (attrflag) 411 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 412 if (!error) { 413 lrupos = 0; 414 NFSLOCKNODE(np); 415 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 416 if (np->n_accesscache[i].uid == cred->cr_uid) { 417 np->n_accesscache[i].mode = rmode; 418 np->n_accesscache[i].stamp = time_second; 419 break; 420 } 421 if (i > 0 && np->n_accesscache[i].stamp < 422 np->n_accesscache[lrupos].stamp) 423 lrupos = i; 424 } 425 if (i == NFS_ACCESSCACHESIZE) { 426 np->n_accesscache[lrupos].uid = cred->cr_uid; 427 np->n_accesscache[lrupos].mode = rmode; 428 np->n_accesscache[lrupos].stamp = time_second; 429 } 430 NFSUNLOCKNODE(np); 431 if (retmode != NULL) 432 *retmode = rmode; 433 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0); 434 } else if (NFS_ISV4(vp)) { 435 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 436 } 437 #ifdef KDTRACE_HOOKS 438 if (error != 0) 439 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0, 440 error); 441 #endif 442 return (error); 443 } 444 445 /* 446 * nfs access vnode op. 447 * For nfs version 2, just return ok. File accesses may fail later. 448 * For nfs version 3, use the access rpc to check accessibility. If file modes 449 * are changed on the server, accesses might still fail later. 450 */ 451 static int 452 nfs_access(struct vop_access_args *ap) 453 { 454 struct vnode *vp = ap->a_vp; 455 int error = 0, i, gotahit; 456 u_int32_t mode, wmode, rmode; 457 int v34 = NFS_ISV34(vp); 458 struct nfsnode *np = VTONFS(vp); 459 460 /* 461 * Disallow write attempts on filesystems mounted read-only; 462 * unless the file is a socket, fifo, or a block or character 463 * device resident on the filesystem. 464 */ 465 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS | 466 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL | 467 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) { 468 switch (vp->v_type) { 469 case VREG: 470 case VDIR: 471 case VLNK: 472 return (EROFS); 473 default: 474 break; 475 } 476 } 477 /* 478 * For nfs v3 or v4, check to see if we have done this recently, and if 479 * so return our cached result instead of making an ACCESS call. 480 * If not, do an access rpc, otherwise you are stuck emulating 481 * ufs_access() locally using the vattr. This may not be correct, 482 * since the server may apply other access criteria such as 483 * client uid-->server uid mapping that we do not know about. 484 */ 485 if (v34) { 486 if (ap->a_accmode & VREAD) 487 mode = NFSACCESS_READ; 488 else 489 mode = 0; 490 if (vp->v_type != VDIR) { 491 if (ap->a_accmode & VWRITE) 492 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 493 if (ap->a_accmode & VAPPEND) 494 mode |= NFSACCESS_EXTEND; 495 if (ap->a_accmode & VEXEC) 496 mode |= NFSACCESS_EXECUTE; 497 if (ap->a_accmode & VDELETE) 498 mode |= NFSACCESS_DELETE; 499 } else { 500 if (ap->a_accmode & VWRITE) 501 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 502 if (ap->a_accmode & VAPPEND) 503 mode |= NFSACCESS_EXTEND; 504 if (ap->a_accmode & VEXEC) 505 mode |= NFSACCESS_LOOKUP; 506 if (ap->a_accmode & VDELETE) 507 mode |= NFSACCESS_DELETE; 508 if (ap->a_accmode & VDELETE_CHILD) 509 mode |= NFSACCESS_MODIFY; 510 } 511 /* XXX safety belt, only make blanket request if caching */ 512 if (nfsaccess_cache_timeout > 0) { 513 wmode = NFSACCESS_READ | NFSACCESS_MODIFY | 514 NFSACCESS_EXTEND | NFSACCESS_EXECUTE | 515 NFSACCESS_DELETE | NFSACCESS_LOOKUP; 516 } else { 517 wmode = mode; 518 } 519 520 /* 521 * Does our cached result allow us to give a definite yes to 522 * this request? 523 */ 524 gotahit = 0; 525 NFSLOCKNODE(np); 526 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 527 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) { 528 if (time_second < (np->n_accesscache[i].stamp 529 + nfsaccess_cache_timeout) && 530 (np->n_accesscache[i].mode & mode) == mode) { 531 NFSINCRGLOBAL(nfsstatsv1.accesscache_hits); 532 gotahit = 1; 533 } 534 break; 535 } 536 } 537 NFSUNLOCKNODE(np); 538 #ifdef KDTRACE_HOOKS 539 if (gotahit != 0) 540 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp, 541 ap->a_cred->cr_uid, mode); 542 else 543 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp, 544 ap->a_cred->cr_uid, mode); 545 #endif 546 if (gotahit == 0) { 547 /* 548 * Either a no, or a don't know. Go to the wire. 549 */ 550 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses); 551 error = nfs34_access_otw(vp, wmode, ap->a_td, 552 ap->a_cred, &rmode); 553 if (!error && 554 (rmode & mode) != mode) 555 error = EACCES; 556 } 557 return (error); 558 } else { 559 if ((error = nfsspec_access(ap)) != 0) { 560 return (error); 561 } 562 /* 563 * Attempt to prevent a mapped root from accessing a file 564 * which it shouldn't. We try to read a byte from the file 565 * if the user is root and the file is not zero length. 566 * After calling nfsspec_access, we should have the correct 567 * file size cached. 568 */ 569 NFSLOCKNODE(np); 570 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD) 571 && VTONFS(vp)->n_size > 0) { 572 struct iovec aiov; 573 struct uio auio; 574 char buf[1]; 575 576 NFSUNLOCKNODE(np); 577 aiov.iov_base = buf; 578 aiov.iov_len = 1; 579 auio.uio_iov = &aiov; 580 auio.uio_iovcnt = 1; 581 auio.uio_offset = 0; 582 auio.uio_resid = 1; 583 auio.uio_segflg = UIO_SYSSPACE; 584 auio.uio_rw = UIO_READ; 585 auio.uio_td = ap->a_td; 586 587 if (vp->v_type == VREG) 588 error = ncl_readrpc(vp, &auio, ap->a_cred); 589 else if (vp->v_type == VDIR) { 590 char* bp; 591 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 592 aiov.iov_base = bp; 593 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 594 error = ncl_readdirrpc(vp, &auio, ap->a_cred, 595 ap->a_td); 596 free(bp, M_TEMP); 597 } else if (vp->v_type == VLNK) 598 error = ncl_readlinkrpc(vp, &auio, ap->a_cred); 599 else 600 error = EACCES; 601 } else 602 NFSUNLOCKNODE(np); 603 return (error); 604 } 605 } 606 607 /* 608 * nfs open vnode op 609 * Check to see if the type is ok 610 * and that deletion is not in progress. 611 * For paged in text files, you will need to flush the page cache 612 * if consistency is lost. 613 */ 614 /* ARGSUSED */ 615 static int 616 nfs_open(struct vop_open_args *ap) 617 { 618 struct vnode *vp = ap->a_vp; 619 struct nfsnode *np = VTONFS(vp); 620 struct vattr vattr; 621 int error; 622 int fmode = ap->a_mode; 623 struct ucred *cred; 624 vm_object_t obj; 625 626 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) 627 return (EOPNOTSUPP); 628 629 /* 630 * For NFSv4, we need to do the Open Op before cache validation, 631 * so that we conform to RFC3530 Sec. 9.3.1. 632 */ 633 if (NFS_ISV4(vp)) { 634 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td); 635 if (error) { 636 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 637 (gid_t)0); 638 return (error); 639 } 640 } 641 642 /* 643 * Now, if this Open will be doing reading, re-validate/flush the 644 * cache, so that Close/Open coherency is maintained. 645 */ 646 NFSLOCKNODE(np); 647 if (np->n_flag & NMODIFIED) { 648 NFSUNLOCKNODE(np); 649 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 650 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 651 if (VN_IS_DOOMED(vp)) 652 return (EBADF); 653 } 654 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 655 if (error == EINTR || error == EIO) { 656 if (NFS_ISV4(vp)) 657 (void) nfsrpc_close(vp, 0, ap->a_td); 658 return (error); 659 } 660 NFSLOCKNODE(np); 661 np->n_attrstamp = 0; 662 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 663 if (vp->v_type == VDIR) 664 np->n_direofoffset = 0; 665 NFSUNLOCKNODE(np); 666 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 667 if (error) { 668 if (NFS_ISV4(vp)) 669 (void) nfsrpc_close(vp, 0, ap->a_td); 670 return (error); 671 } 672 NFSLOCKNODE(np); 673 np->n_mtime = vattr.va_mtime; 674 if (NFS_ISV4(vp)) 675 np->n_change = vattr.va_filerev; 676 } else { 677 NFSUNLOCKNODE(np); 678 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 679 if (error) { 680 if (NFS_ISV4(vp)) 681 (void) nfsrpc_close(vp, 0, ap->a_td); 682 return (error); 683 } 684 NFSLOCKNODE(np); 685 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) || 686 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 687 if (vp->v_type == VDIR) 688 np->n_direofoffset = 0; 689 NFSUNLOCKNODE(np); 690 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 691 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 692 if (VN_IS_DOOMED(vp)) 693 return (EBADF); 694 } 695 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 696 if (error == EINTR || error == EIO) { 697 if (NFS_ISV4(vp)) 698 (void) nfsrpc_close(vp, 0, ap->a_td); 699 return (error); 700 } 701 NFSLOCKNODE(np); 702 np->n_mtime = vattr.va_mtime; 703 if (NFS_ISV4(vp)) 704 np->n_change = vattr.va_filerev; 705 } 706 } 707 708 /* 709 * If the object has >= 1 O_DIRECT active opens, we disable caching. 710 */ 711 if (newnfs_directio_enable && (fmode & O_DIRECT) && 712 (vp->v_type == VREG)) { 713 if (np->n_directio_opens == 0) { 714 NFSUNLOCKNODE(np); 715 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 716 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 717 if (VN_IS_DOOMED(vp)) 718 return (EBADF); 719 } 720 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 721 if (error) { 722 if (NFS_ISV4(vp)) 723 (void) nfsrpc_close(vp, 0, ap->a_td); 724 return (error); 725 } 726 NFSLOCKNODE(np); 727 np->n_flag |= NNONCACHE; 728 } 729 np->n_directio_opens++; 730 } 731 732 /* If opened for writing via NFSv4.1 or later, mark that for pNFS. */ 733 if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0) 734 np->n_flag |= NWRITEOPENED; 735 736 /* 737 * If this is an open for writing, capture a reference to the 738 * credentials, so they can be used by ncl_putpages(). Using 739 * these write credentials is preferable to the credentials of 740 * whatever thread happens to be doing the VOP_PUTPAGES() since 741 * the write RPCs are less likely to fail with EACCES. 742 */ 743 if ((fmode & FWRITE) != 0) { 744 cred = np->n_writecred; 745 np->n_writecred = crhold(ap->a_cred); 746 } else 747 cred = NULL; 748 NFSUNLOCKNODE(np); 749 750 if (cred != NULL) 751 crfree(cred); 752 vnode_create_vobject(vp, vattr.va_size, ap->a_td); 753 754 /* 755 * If the text file has been mmap'd, flush any dirty pages to the 756 * buffer cache and then... 757 * Make sure all writes are pushed to the NFS server. If this is not 758 * done, the modify time of the file can change while the text 759 * file is being executed. This will cause the process that is 760 * executing the text file to be terminated. 761 */ 762 if (vp->v_writecount <= -1) { 763 if ((obj = vp->v_object) != NULL && 764 vm_object_mightbedirty(obj)) { 765 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 766 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 767 if (VN_IS_DOOMED(vp)) 768 return (EBADF); 769 } 770 vnode_pager_clean_sync(vp); 771 } 772 773 /* Now, flush the buffer cache. */ 774 ncl_flush(vp, MNT_WAIT, curthread, 0, 0); 775 776 /* And, finally, make sure that n_mtime is up to date. */ 777 np = VTONFS(vp); 778 NFSLOCKNODE(np); 779 np->n_mtime = np->n_vattr.na_mtime; 780 NFSUNLOCKNODE(np); 781 } 782 return (0); 783 } 784 785 /* 786 * nfs close vnode op 787 * What an NFS client should do upon close after writing is a debatable issue. 788 * Most NFS clients push delayed writes to the server upon close, basically for 789 * two reasons: 790 * 1 - So that any write errors may be reported back to the client process 791 * doing the close system call. By far the two most likely errors are 792 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 793 * 2 - To put a worst case upper bound on cache inconsistency between 794 * multiple clients for the file. 795 * There is also a consistency problem for Version 2 of the protocol w.r.t. 796 * not being able to tell if other clients are writing a file concurrently, 797 * since there is no way of knowing if the changed modify time in the reply 798 * is only due to the write for this client. 799 * (NFS Version 3 provides weak cache consistency data in the reply that 800 * should be sufficient to detect and handle this case.) 801 * 802 * The current code does the following: 803 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 804 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 805 * or commit them (this satisfies 1 and 2 except for the 806 * case where the server crashes after this close but 807 * before the commit RPC, which is felt to be "good 808 * enough". Changing the last argument to ncl_flush() to 809 * a 1 would force a commit operation, if it is felt a 810 * commit is necessary now. 811 * for NFS Version 4 - flush the dirty buffers and commit them, if 812 * nfscl_mustflush() says this is necessary. 813 * It is necessary if there is no write delegation held, 814 * in order to satisfy open/close coherency. 815 * If the file isn't cached on local stable storage, 816 * it may be necessary in order to detect "out of space" 817 * errors from the server, if the write delegation 818 * issued by the server doesn't allow the file to grow. 819 */ 820 /* ARGSUSED */ 821 static int 822 nfs_close(struct vop_close_args *ap) 823 { 824 struct vnode *vp = ap->a_vp; 825 struct nfsnode *np = VTONFS(vp); 826 struct nfsvattr nfsva; 827 struct ucred *cred; 828 int error = 0, ret, localcred = 0; 829 int fmode = ap->a_fflag; 830 831 if (NFSCL_FORCEDISM(vp->v_mount)) 832 return (0); 833 /* 834 * During shutdown, a_cred isn't valid, so just use root. 835 */ 836 if (ap->a_cred == NOCRED) { 837 cred = newnfs_getcred(); 838 localcred = 1; 839 } else { 840 cred = ap->a_cred; 841 } 842 if (vp->v_type == VREG) { 843 /* 844 * Examine and clean dirty pages, regardless of NMODIFIED. 845 * This closes a major hole in close-to-open consistency. 846 * We want to push out all dirty pages (and buffers) on 847 * close, regardless of whether they were dirtied by 848 * mmap'ed writes or via write(). 849 */ 850 if (nfs_clean_pages_on_close && vp->v_object) { 851 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 852 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 853 if (VN_IS_DOOMED(vp) && ap->a_fflag != FNONBLOCK) 854 return (EBADF); 855 } 856 vnode_pager_clean_async(vp); 857 } 858 NFSLOCKNODE(np); 859 if (np->n_flag & NMODIFIED) { 860 NFSUNLOCKNODE(np); 861 if (NFS_ISV3(vp)) { 862 /* 863 * Under NFSv3 we have dirty buffers to dispose of. We 864 * must flush them to the NFS server. We have the option 865 * of waiting all the way through the commit rpc or just 866 * waiting for the initial write. The default is to only 867 * wait through the initial write so the data is in the 868 * server's cache, which is roughly similar to the state 869 * a standard disk subsystem leaves the file in on close(). 870 * 871 * We cannot clear the NMODIFIED bit in np->n_flag due to 872 * potential races with other processes, and certainly 873 * cannot clear it if we don't commit. 874 * These races occur when there is no longer the old 875 * traditional vnode locking implemented for Vnode Ops. 876 */ 877 int cm = newnfs_commit_on_close ? 1 : 0; 878 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 879 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 880 if (VN_IS_DOOMED(vp) && ap->a_fflag != FNONBLOCK) 881 return (EBADF); 882 } 883 error = ncl_flush(vp, MNT_WAIT, ap->a_td, cm, 0); 884 /* np->n_flag &= ~NMODIFIED; */ 885 } else if (NFS_ISV4(vp)) { 886 if (nfscl_mustflush(vp) != 0) { 887 int cm = newnfs_commit_on_close ? 1 : 0; 888 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 889 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 890 if (VN_IS_DOOMED(vp) && ap->a_fflag != 891 FNONBLOCK) 892 return (EBADF); 893 } 894 error = ncl_flush(vp, MNT_WAIT, ap->a_td, 895 cm, 0); 896 /* 897 * as above w.r.t races when clearing 898 * NMODIFIED. 899 * np->n_flag &= ~NMODIFIED; 900 */ 901 } 902 } else { 903 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 904 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 905 if (VN_IS_DOOMED(vp) && ap->a_fflag != 906 FNONBLOCK) 907 return (EBADF); 908 } 909 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 910 } 911 NFSLOCKNODE(np); 912 } 913 /* 914 * Invalidate the attribute cache in all cases. 915 * An open is going to fetch fresh attrs any way, other procs 916 * on this node that have file open will be forced to do an 917 * otw attr fetch, but this is safe. 918 * --> A user found that their RPC count dropped by 20% when 919 * this was commented out and I can't see any requirement 920 * for it, so I've disabled it when negative lookups are 921 * enabled. (What does this have to do with negative lookup 922 * caching? Well nothing, except it was reported by the 923 * same user that needed negative lookup caching and I wanted 924 * there to be a way to disable it to see if it 925 * is the cause of some caching/coherency issue that might 926 * crop up.) 927 */ 928 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) { 929 np->n_attrstamp = 0; 930 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 931 } 932 if (np->n_flag & NWRITEERR) { 933 np->n_flag &= ~NWRITEERR; 934 error = np->n_error; 935 } 936 NFSUNLOCKNODE(np); 937 } 938 939 if (NFS_ISV4(vp)) { 940 /* 941 * Get attributes so "change" is up to date. 942 */ 943 if (error == 0 && nfscl_nodeleg(vp, 0) != 0 && 944 vp->v_type == VREG && 945 (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) { 946 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva); 947 if (!ret) { 948 np->n_change = nfsva.na_filerev; 949 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 950 0, 0); 951 } 952 } 953 954 /* 955 * and do the close. 956 */ 957 ret = nfsrpc_close(vp, 0, ap->a_td); 958 if (!error && ret) 959 error = ret; 960 if (error) 961 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 962 (gid_t)0); 963 } 964 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) { 965 NFSLOCKNODE(np); 966 KASSERT((np->n_directio_opens > 0), 967 ("nfs_close: unexpectedly value (0) of n_directio_opens\n")); 968 np->n_directio_opens--; 969 if (np->n_directio_opens == 0) 970 np->n_flag &= ~NNONCACHE; 971 NFSUNLOCKNODE(np); 972 } 973 if (localcred) 974 NFSFREECRED(cred); 975 return (error); 976 } 977 978 /* 979 * nfs getattr call from vfs. 980 */ 981 static int 982 nfs_getattr(struct vop_getattr_args *ap) 983 { 984 struct vnode *vp = ap->a_vp; 985 struct thread *td = curthread; /* XXX */ 986 struct nfsnode *np = VTONFS(vp); 987 int error = 0; 988 struct nfsvattr nfsva; 989 struct vattr *vap = ap->a_vap; 990 struct vattr vattr; 991 struct nfsmount *nmp; 992 993 nmp = VFSTONFS(vp->v_mount); 994 /* 995 * Update local times for special files. 996 */ 997 NFSLOCKNODE(np); 998 if (np->n_flag & (NACC | NUPD)) 999 np->n_flag |= NCHG; 1000 NFSUNLOCKNODE(np); 1001 /* 1002 * First look in the cache. 1003 * For "syskrb5" mounts, nm_fhsize might still be zero and 1004 * cached attributes should be ignored. 1005 */ 1006 if (nmp->nm_fhsize > 0 && ncl_getattrcache(vp, &vattr) == 0) { 1007 ncl_copy_vattr(vap, &vattr); 1008 1009 /* 1010 * Get the local modify time for the case of a write 1011 * delegation. 1012 */ 1013 nfscl_deleggetmodtime(vp, &vap->va_mtime); 1014 return (0); 1015 } 1016 1017 if (NFS_ISV34(vp) && nfs_prime_access_cache && 1018 nfsaccess_cache_timeout > 0) { 1019 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses); 1020 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL); 1021 if (ncl_getattrcache(vp, ap->a_vap) == 0) { 1022 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime); 1023 return (0); 1024 } 1025 } 1026 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva); 1027 if (!error) 1028 error = nfscl_loadattrcache(&vp, &nfsva, vap, 0, 0); 1029 if (!error) { 1030 /* 1031 * Get the local modify time for the case of a write 1032 * delegation. 1033 */ 1034 nfscl_deleggetmodtime(vp, &vap->va_mtime); 1035 } else if (NFS_ISV4(vp)) { 1036 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1037 } 1038 return (error); 1039 } 1040 1041 /* 1042 * nfs setattr call. 1043 */ 1044 static int 1045 nfs_setattr(struct vop_setattr_args *ap) 1046 { 1047 struct vnode *vp = ap->a_vp; 1048 struct nfsnode *np = VTONFS(vp); 1049 struct thread *td = curthread; /* XXX */ 1050 struct vattr *vap = ap->a_vap; 1051 int error = 0; 1052 u_quad_t tsize; 1053 struct timespec ts; 1054 1055 #ifndef nolint 1056 tsize = (u_quad_t)0; 1057 #endif 1058 1059 /* 1060 * Setting of flags and marking of atimes are not supported. 1061 */ 1062 if (vap->va_flags != VNOVAL) 1063 return (EOPNOTSUPP); 1064 1065 /* 1066 * Disallow write attempts if the filesystem is mounted read-only. 1067 */ 1068 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 1069 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 1070 vap->va_mtime.tv_sec != VNOVAL || 1071 vap->va_birthtime.tv_sec != VNOVAL || 1072 vap->va_mode != (mode_t)VNOVAL) && 1073 (vp->v_mount->mnt_flag & MNT_RDONLY)) 1074 return (EROFS); 1075 if (vap->va_size != VNOVAL) { 1076 switch (vp->v_type) { 1077 case VDIR: 1078 return (EISDIR); 1079 case VCHR: 1080 case VBLK: 1081 case VSOCK: 1082 case VFIFO: 1083 if (vap->va_mtime.tv_sec == VNOVAL && 1084 vap->va_atime.tv_sec == VNOVAL && 1085 vap->va_birthtime.tv_sec == VNOVAL && 1086 vap->va_mode == (mode_t)VNOVAL && 1087 vap->va_uid == (uid_t)VNOVAL && 1088 vap->va_gid == (gid_t)VNOVAL) 1089 return (0); 1090 vap->va_size = VNOVAL; 1091 break; 1092 default: 1093 /* 1094 * Disallow write attempts if the filesystem is 1095 * mounted read-only. 1096 */ 1097 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1098 return (EROFS); 1099 /* 1100 * We run vnode_pager_setsize() early (why?), 1101 * we must set np->n_size now to avoid vinvalbuf 1102 * V_SAVE races that might setsize a lower 1103 * value. 1104 */ 1105 NFSLOCKNODE(np); 1106 tsize = np->n_size; 1107 NFSUNLOCKNODE(np); 1108 error = ncl_meta_setsize(vp, td, vap->va_size); 1109 NFSLOCKNODE(np); 1110 if (np->n_flag & NMODIFIED) { 1111 tsize = np->n_size; 1112 NFSUNLOCKNODE(np); 1113 error = ncl_vinvalbuf(vp, vap->va_size == 0 ? 1114 0 : V_SAVE, td, 1); 1115 if (error != 0) { 1116 vnode_pager_setsize(vp, tsize); 1117 return (error); 1118 } 1119 /* 1120 * Call nfscl_delegmodtime() to set the modify time 1121 * locally, as required. 1122 */ 1123 nfscl_delegmodtime(vp, NULL); 1124 } else 1125 NFSUNLOCKNODE(np); 1126 /* 1127 * np->n_size has already been set to vap->va_size 1128 * in ncl_meta_setsize(). We must set it again since 1129 * nfs_loadattrcache() could be called through 1130 * ncl_meta_setsize() and could modify np->n_size. 1131 */ 1132 NFSLOCKNODE(np); 1133 np->n_vattr.na_size = np->n_size = vap->va_size; 1134 NFSUNLOCKNODE(np); 1135 } 1136 } else { 1137 NFSLOCKNODE(np); 1138 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) && 1139 (np->n_flag & NMODIFIED) && vp->v_type == VREG) { 1140 NFSUNLOCKNODE(np); 1141 error = ncl_vinvalbuf(vp, V_SAVE, td, 1); 1142 if (error == EINTR || error == EIO) 1143 return (error); 1144 } else 1145 NFSUNLOCKNODE(np); 1146 } 1147 error = nfs_setattrrpc(vp, vap, ap->a_cred, td); 1148 if (vap->va_size != VNOVAL) { 1149 if (error == 0) { 1150 nanouptime(&ts); 1151 NFSLOCKNODE(np); 1152 np->n_localmodtime = ts; 1153 NFSUNLOCKNODE(np); 1154 } else { 1155 NFSLOCKNODE(np); 1156 np->n_size = np->n_vattr.na_size = tsize; 1157 vnode_pager_setsize(vp, tsize); 1158 NFSUNLOCKNODE(np); 1159 } 1160 } 1161 if (vap->va_mtime.tv_sec != VNOVAL && error == 0) 1162 nfscl_delegmodtime(vp, &vap->va_mtime); 1163 return (error); 1164 } 1165 1166 /* 1167 * Do an nfs setattr rpc. 1168 */ 1169 static int 1170 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred, 1171 struct thread *td) 1172 { 1173 struct nfsnode *np = VTONFS(vp); 1174 int error, ret, attrflag, i; 1175 struct nfsvattr nfsva; 1176 1177 if (NFS_ISV34(vp)) { 1178 NFSLOCKNODE(np); 1179 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) 1180 np->n_accesscache[i].stamp = 0; 1181 np->n_flag |= NDELEGMOD; 1182 NFSUNLOCKNODE(np); 1183 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp); 1184 } 1185 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag); 1186 if (attrflag) { 1187 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 1188 if (ret && !error) 1189 error = ret; 1190 } 1191 if (error && NFS_ISV4(vp)) 1192 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid); 1193 return (error); 1194 } 1195 1196 /* 1197 * nfs lookup call, one step at a time... 1198 * First look in cache 1199 * If not found, unlock the directory nfsnode and do the rpc 1200 */ 1201 static int 1202 nfs_lookup(struct vop_lookup_args *ap) 1203 { 1204 struct componentname *cnp = ap->a_cnp; 1205 struct vnode *dvp = ap->a_dvp; 1206 struct vnode **vpp = ap->a_vpp; 1207 struct mount *mp = dvp->v_mount; 1208 int flags = cnp->cn_flags; 1209 struct vnode *newvp; 1210 struct nfsmount *nmp; 1211 struct nfsnode *np, *newnp; 1212 int error = 0, attrflag, dattrflag, ltype, ncticks; 1213 struct thread *td = curthread; 1214 struct nfsfh *nfhp; 1215 struct nfsvattr dnfsva, nfsva; 1216 struct vattr vattr; 1217 struct timespec nctime, ts; 1218 uint32_t openmode; 1219 1220 *vpp = NULLVP; 1221 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) && 1222 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 1223 return (EROFS); 1224 if (dvp->v_type != VDIR) 1225 return (ENOTDIR); 1226 nmp = VFSTONFS(mp); 1227 np = VTONFS(dvp); 1228 1229 /* For NFSv4, wait until any remove is done. */ 1230 NFSLOCKNODE(np); 1231 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) { 1232 np->n_flag |= NREMOVEWANT; 1233 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0); 1234 } 1235 NFSUNLOCKNODE(np); 1236 1237 error = vn_dir_check_exec(dvp, cnp); 1238 if (error != 0) 1239 return (error); 1240 error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks); 1241 if (error > 0 && error != ENOENT) 1242 return (error); 1243 if (error == -1) { 1244 /* 1245 * Lookups of "." are special and always return the 1246 * current directory. cache_lookup() already handles 1247 * associated locking bookkeeping, etc. 1248 */ 1249 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 1250 return (0); 1251 } 1252 1253 /* 1254 * We only accept a positive hit in the cache if the 1255 * change time of the file matches our cached copy. 1256 * Otherwise, we discard the cache entry and fallback 1257 * to doing a lookup RPC. We also only trust cache 1258 * entries for less than nm_nametimeo seconds. 1259 * 1260 * To better handle stale file handles and attributes, 1261 * clear the attribute cache of this node if it is a 1262 * leaf component, part of an open() call, and not 1263 * locally modified before fetching the attributes. 1264 * This should allow stale file handles to be detected 1265 * here where we can fall back to a LOOKUP RPC to 1266 * recover rather than having nfs_open() detect the 1267 * stale file handle and failing open(2) with ESTALE. 1268 */ 1269 newvp = *vpp; 1270 newnp = VTONFS(newvp); 1271 if (!(nmp->nm_flag & NFSMNT_NOCTO) && 1272 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1273 !(newnp->n_flag & NMODIFIED)) { 1274 NFSLOCKNODE(newnp); 1275 newnp->n_attrstamp = 0; 1276 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1277 NFSUNLOCKNODE(newnp); 1278 } 1279 if (nfscl_nodeleg(newvp, 0) == 0 || 1280 ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) && 1281 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 && 1282 timespeccmp(&vattr.va_ctime, &nctime, ==))) { 1283 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits); 1284 return (0); 1285 } 1286 cache_purge(newvp); 1287 if (dvp != newvp) 1288 vput(newvp); 1289 else 1290 vrele(newvp); 1291 *vpp = NULLVP; 1292 } else if (error == ENOENT) { 1293 if (VN_IS_DOOMED(dvp)) 1294 return (ENOENT); 1295 /* 1296 * We only accept a negative hit in the cache if the 1297 * modification time of the parent directory matches 1298 * the cached copy in the name cache entry. 1299 * Otherwise, we discard all of the negative cache 1300 * entries for this directory. We also only trust 1301 * negative cache entries for up to nm_negnametimeo 1302 * seconds. 1303 */ 1304 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) && 1305 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 && 1306 timespeccmp(&vattr.va_mtime, &nctime, ==)) { 1307 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits); 1308 return (ENOENT); 1309 } 1310 cache_purge_negative(dvp); 1311 } 1312 1313 openmode = 0; 1314 #if 0 1315 /* 1316 * The use of LookupOpen breaks some builds. It is disabled 1317 * until that is fixed. 1318 */ 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 #endif 1339 1340 newvp = NULLVP; 1341 NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses); 1342 nanouptime(&ts); 1343 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1344 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1345 openmode); 1346 if (dattrflag) 1347 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 1348 if (error) { 1349 if (newvp != NULLVP) { 1350 vput(newvp); 1351 *vpp = NULLVP; 1352 } 1353 1354 if (error != ENOENT) { 1355 if (NFS_ISV4(dvp)) 1356 error = nfscl_maperr(td, error, (uid_t)0, 1357 (gid_t)0); 1358 return (error); 1359 } 1360 1361 /* The requested file was not found. */ 1362 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1363 (flags & ISLASTCN)) { 1364 /* 1365 * XXX: UFS does a full VOP_ACCESS(dvp, 1366 * VWRITE) here instead of just checking 1367 * MNT_RDONLY. 1368 */ 1369 if (mp->mnt_flag & MNT_RDONLY) 1370 return (EROFS); 1371 return (EJUSTRETURN); 1372 } 1373 1374 if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) { 1375 /* 1376 * Cache the modification time of the parent 1377 * directory from the post-op attributes in 1378 * the name cache entry. The negative cache 1379 * entry will be ignored once the directory 1380 * has changed. Don't bother adding the entry 1381 * if the directory has already changed. 1382 */ 1383 NFSLOCKNODE(np); 1384 if (timespeccmp(&np->n_vattr.na_mtime, 1385 &dnfsva.na_mtime, ==)) { 1386 NFSUNLOCKNODE(np); 1387 cache_enter_time(dvp, NULL, cnp, 1388 &dnfsva.na_mtime, NULL); 1389 } else 1390 NFSUNLOCKNODE(np); 1391 } 1392 return (ENOENT); 1393 } 1394 1395 /* 1396 * Handle RENAME case... 1397 */ 1398 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 1399 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1400 free(nfhp, M_NFSFH); 1401 return (EISDIR); 1402 } 1403 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, LK_EXCLUSIVE); 1404 if (error) 1405 return (error); 1406 newvp = NFSTOV(np); 1407 /* 1408 * If n_localmodtime >= time before RPC, then 1409 * a file modification operation, such as 1410 * VOP_SETATTR() of size, has occurred while 1411 * the Lookup RPC and acquisition of the vnode 1412 * happened. As such, the attributes might 1413 * be stale, with possibly an incorrect size. 1414 */ 1415 NFSLOCKNODE(np); 1416 if (timespecisset(&np->n_localmodtime) && 1417 timespeccmp(&np->n_localmodtime, &ts, >=)) { 1418 NFSCL_DEBUG(4, "nfs_lookup: rename localmod " 1419 "stale attributes\n"); 1420 attrflag = 0; 1421 } 1422 NFSUNLOCKNODE(np); 1423 if (attrflag) 1424 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1425 *vpp = newvp; 1426 return (0); 1427 } 1428 1429 if (flags & ISDOTDOT) { 1430 ltype = NFSVOPISLOCKED(dvp); 1431 error = vfs_busy(mp, MBF_NOWAIT); 1432 if (error != 0) { 1433 vfs_ref(mp); 1434 NFSVOPUNLOCK(dvp); 1435 error = vfs_busy(mp, 0); 1436 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1437 vfs_rel(mp); 1438 if (error == 0 && VN_IS_DOOMED(dvp)) { 1439 vfs_unbusy(mp); 1440 error = ENOENT; 1441 } 1442 if (error != 0) 1443 return (error); 1444 } 1445 NFSVOPUNLOCK(dvp); 1446 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, 1447 cnp->cn_lkflags); 1448 if (error == 0) 1449 newvp = NFSTOV(np); 1450 vfs_unbusy(mp); 1451 if (newvp != dvp) 1452 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1453 if (VN_IS_DOOMED(dvp)) { 1454 if (error == 0) { 1455 if (newvp == dvp) 1456 vrele(newvp); 1457 else 1458 vput(newvp); 1459 } 1460 error = ENOENT; 1461 } 1462 if (error != 0) 1463 return (error); 1464 if (attrflag) 1465 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1466 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1467 free(nfhp, M_NFSFH); 1468 VREF(dvp); 1469 newvp = dvp; 1470 if (attrflag) 1471 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1472 } else { 1473 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, 1474 cnp->cn_lkflags); 1475 if (error) 1476 return (error); 1477 newvp = NFSTOV(np); 1478 /* 1479 * If n_localmodtime >= time before RPC, then 1480 * a file modification operation, such as 1481 * VOP_SETATTR() of size, has occurred while 1482 * the Lookup RPC and acquisition of the vnode 1483 * happened. As such, the attributes might 1484 * be stale, with possibly an incorrect size. 1485 */ 1486 NFSLOCKNODE(np); 1487 if (timespecisset(&np->n_localmodtime) && 1488 timespeccmp(&np->n_localmodtime, &ts, >=)) { 1489 NFSCL_DEBUG(4, "nfs_lookup: localmod " 1490 "stale attributes\n"); 1491 attrflag = 0; 1492 } 1493 NFSUNLOCKNODE(np); 1494 if (attrflag) 1495 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1496 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1497 !(np->n_flag & NMODIFIED)) { 1498 /* 1499 * Flush the attribute cache when opening a 1500 * leaf node to ensure that fresh attributes 1501 * are fetched in nfs_open() since we did not 1502 * fetch attributes from the LOOKUP reply. 1503 */ 1504 NFSLOCKNODE(np); 1505 np->n_attrstamp = 0; 1506 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1507 NFSUNLOCKNODE(np); 1508 } 1509 } 1510 if ((cnp->cn_flags & MAKEENTRY) && dvp != newvp && 1511 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) && 1512 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0)) 1513 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 1514 newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime); 1515 *vpp = newvp; 1516 return (0); 1517 } 1518 1519 /* 1520 * nfs read call. 1521 * Just call ncl_bioread() to do the work. 1522 */ 1523 static int 1524 nfs_read(struct vop_read_args *ap) 1525 { 1526 struct vnode *vp = ap->a_vp; 1527 1528 switch (vp->v_type) { 1529 case VREG: 1530 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred)); 1531 case VDIR: 1532 return (EISDIR); 1533 default: 1534 return (EOPNOTSUPP); 1535 } 1536 } 1537 1538 /* 1539 * nfs readlink call 1540 */ 1541 static int 1542 nfs_readlink(struct vop_readlink_args *ap) 1543 { 1544 struct vnode *vp = ap->a_vp; 1545 1546 if (vp->v_type != VLNK) 1547 return (EINVAL); 1548 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred)); 1549 } 1550 1551 /* 1552 * Do a readlink rpc. 1553 * Called by ncl_doio() from below the buffer cache. 1554 */ 1555 int 1556 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1557 { 1558 int error, ret, attrflag; 1559 struct nfsvattr nfsva; 1560 1561 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva, 1562 &attrflag); 1563 if (attrflag) { 1564 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 1565 if (ret && !error) 1566 error = ret; 1567 } 1568 if (error && NFS_ISV4(vp)) 1569 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1570 return (error); 1571 } 1572 1573 /* 1574 * nfs read rpc call 1575 * Ditto above 1576 */ 1577 int 1578 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1579 { 1580 int error, ret, attrflag; 1581 struct nfsvattr nfsva; 1582 struct nfsmount *nmp; 1583 1584 nmp = VFSTONFS(vp->v_mount); 1585 error = EIO; 1586 attrflag = 0; 1587 if (NFSHASPNFS(nmp)) 1588 error = nfscl_doiods(vp, uiop, NULL, NULL, 1589 NFSV4OPEN_ACCESSREAD, 0, cred, uiop->uio_td); 1590 NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error); 1591 if (error != 0 && error != EFAULT) 1592 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, 1593 &attrflag); 1594 if (attrflag) { 1595 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 1596 if (ret && !error) 1597 error = ret; 1598 } 1599 if (error && NFS_ISV4(vp)) 1600 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1601 return (error); 1602 } 1603 1604 /* 1605 * nfs write call 1606 */ 1607 int 1608 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 1609 int *iomode, int *must_commit, int called_from_strategy, int ioflag) 1610 { 1611 struct nfsvattr nfsva; 1612 int error, attrflag, ret; 1613 struct nfsmount *nmp; 1614 1615 nmp = VFSTONFS(vp->v_mount); 1616 error = EIO; 1617 attrflag = 0; 1618 if (NFSHASPNFS(nmp)) 1619 error = nfscl_doiods(vp, uiop, iomode, must_commit, 1620 NFSV4OPEN_ACCESSWRITE, 0, cred, uiop->uio_td); 1621 NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error); 1622 if (error != 0 && error != EFAULT) 1623 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred, 1624 uiop->uio_td, &nfsva, &attrflag, called_from_strategy, 1625 ioflag); 1626 if (attrflag) { 1627 if (VTONFS(vp)->n_flag & ND_NFSV4) 1628 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 1, 1); 1629 else 1630 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 1631 if (ret && !error) 1632 error = ret; 1633 } 1634 if (DOINGASYNC(vp)) 1635 *iomode = NFSWRITE_FILESYNC; 1636 if (error && NFS_ISV4(vp)) 1637 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1638 return (error); 1639 } 1640 1641 /* 1642 * nfs mknod rpc 1643 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1644 * mode set to specify the file type and the size field for rdev. 1645 */ 1646 static int 1647 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1648 struct vattr *vap) 1649 { 1650 struct nfsvattr nfsva, dnfsva; 1651 struct vnode *newvp = NULL; 1652 struct nfsnode *np = NULL, *dnp; 1653 struct nfsfh *nfhp; 1654 struct vattr vattr; 1655 int error = 0, attrflag, dattrflag; 1656 u_int32_t rdev; 1657 1658 if (vap->va_type == VCHR || vap->va_type == VBLK) 1659 rdev = vap->va_rdev; 1660 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1661 rdev = 0xffffffff; 1662 else 1663 return (EOPNOTSUPP); 1664 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1665 return (error); 1666 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap, 1667 rdev, vap->va_type, cnp->cn_cred, curthread, &dnfsva, 1668 &nfsva, &nfhp, &attrflag, &dattrflag); 1669 if (!error) { 1670 if (!nfhp) 1671 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1672 cnp->cn_namelen, cnp->cn_cred, curthread, 1673 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 0); 1674 if (nfhp) 1675 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1676 curthread, &np, LK_EXCLUSIVE); 1677 } 1678 if (dattrflag) 1679 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 1680 if (!error) { 1681 newvp = NFSTOV(np); 1682 if (attrflag != 0) { 1683 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1684 if (error != 0) 1685 vput(newvp); 1686 } 1687 } 1688 if (!error) { 1689 *vpp = newvp; 1690 } else if (NFS_ISV4(dvp)) { 1691 error = nfscl_maperr(curthread, error, vap->va_uid, 1692 vap->va_gid); 1693 } 1694 dnp = VTONFS(dvp); 1695 NFSLOCKNODE(dnp); 1696 dnp->n_flag |= NMODIFIED; 1697 if (!dattrflag) { 1698 dnp->n_attrstamp = 0; 1699 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1700 } 1701 NFSUNLOCKNODE(dnp); 1702 return (error); 1703 } 1704 1705 /* 1706 * nfs mknod vop 1707 * just call nfs_mknodrpc() to do the work. 1708 */ 1709 /* ARGSUSED */ 1710 static int 1711 nfs_mknod(struct vop_mknod_args *ap) 1712 { 1713 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap)); 1714 } 1715 1716 static struct mtx nfs_cverf_mtx; 1717 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex", 1718 MTX_DEF); 1719 1720 static nfsquad_t 1721 nfs_get_cverf(void) 1722 { 1723 static nfsquad_t cverf; 1724 nfsquad_t ret; 1725 static int cverf_initialized = 0; 1726 1727 mtx_lock(&nfs_cverf_mtx); 1728 if (cverf_initialized == 0) { 1729 cverf.lval[0] = arc4random(); 1730 cverf.lval[1] = arc4random(); 1731 cverf_initialized = 1; 1732 } else 1733 cverf.qval++; 1734 ret = cverf; 1735 mtx_unlock(&nfs_cverf_mtx); 1736 1737 return (ret); 1738 } 1739 1740 /* 1741 * nfs file create call 1742 */ 1743 static int 1744 nfs_create(struct vop_create_args *ap) 1745 { 1746 struct vnode *dvp = ap->a_dvp; 1747 struct vattr *vap = ap->a_vap; 1748 struct componentname *cnp = ap->a_cnp; 1749 struct nfsnode *np = NULL, *dnp; 1750 struct vnode *newvp = NULL; 1751 struct nfsmount *nmp; 1752 struct nfsvattr dnfsva, nfsva; 1753 struct nfsfh *nfhp; 1754 nfsquad_t cverf; 1755 int error = 0, attrflag, dattrflag, fmode = 0; 1756 struct vattr vattr; 1757 1758 /* 1759 * Oops, not for me.. 1760 */ 1761 if (vap->va_type == VSOCK) 1762 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1763 1764 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1765 return (error); 1766 if (vap->va_vaflags & VA_EXCLUSIVE) 1767 fmode |= O_EXCL; 1768 dnp = VTONFS(dvp); 1769 nmp = VFSTONFS(dvp->v_mount); 1770 again: 1771 /* For NFSv4, wait until any remove is done. */ 1772 NFSLOCKNODE(dnp); 1773 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) { 1774 dnp->n_flag |= NREMOVEWANT; 1775 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0); 1776 } 1777 NFSUNLOCKNODE(dnp); 1778 1779 cverf = nfs_get_cverf(); 1780 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1781 vap, cverf, fmode, cnp->cn_cred, curthread, &dnfsva, &nfsva, 1782 &nfhp, &attrflag, &dattrflag); 1783 if (!error) { 1784 if (nfhp == NULL) 1785 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1786 cnp->cn_namelen, cnp->cn_cred, curthread, 1787 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 0); 1788 if (nfhp != NULL) 1789 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1790 curthread, &np, LK_EXCLUSIVE); 1791 } 1792 if (dattrflag) 1793 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 1794 if (!error) { 1795 newvp = NFSTOV(np); 1796 if (attrflag == 0) 1797 error = nfsrpc_getattr(newvp, cnp->cn_cred, curthread, 1798 &nfsva); 1799 if (error == 0) 1800 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1801 } 1802 if (error) { 1803 if (newvp != NULL) { 1804 vput(newvp); 1805 newvp = NULL; 1806 } 1807 if (NFS_ISV34(dvp) && (fmode & O_EXCL) && 1808 error == NFSERR_NOTSUPP) { 1809 fmode &= ~O_EXCL; 1810 goto again; 1811 } 1812 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) { 1813 if (nfscl_checksattr(vap, &nfsva)) { 1814 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred, 1815 curthread, &nfsva, &attrflag); 1816 if (error && (vap->va_uid != (uid_t)VNOVAL || 1817 vap->va_gid != (gid_t)VNOVAL)) { 1818 /* try again without setting uid/gid */ 1819 vap->va_uid = (uid_t)VNOVAL; 1820 vap->va_gid = (uid_t)VNOVAL; 1821 error = nfsrpc_setattr(newvp, vap, NULL, 1822 cnp->cn_cred, curthread, &nfsva, &attrflag); 1823 } 1824 if (attrflag) 1825 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 1826 0, 1); 1827 if (error != 0) 1828 vput(newvp); 1829 } 1830 } 1831 if (!error) { 1832 if ((cnp->cn_flags & MAKEENTRY) && attrflag) { 1833 if (dvp != newvp) 1834 cache_enter_time(dvp, newvp, cnp, 1835 &nfsva.na_ctime, NULL); 1836 else 1837 printf("nfs_create: bogus NFS server returned " 1838 "the directory as the new file object\n"); 1839 } 1840 *ap->a_vpp = newvp; 1841 } else if (NFS_ISV4(dvp)) { 1842 error = nfscl_maperr(curthread, error, vap->va_uid, 1843 vap->va_gid); 1844 } 1845 NFSLOCKNODE(dnp); 1846 dnp->n_flag |= NMODIFIED; 1847 if (!dattrflag) { 1848 dnp->n_attrstamp = 0; 1849 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1850 } 1851 NFSUNLOCKNODE(dnp); 1852 return (error); 1853 } 1854 1855 /* 1856 * nfs file remove call 1857 * To try and make nfs semantics closer to ufs semantics, a file that has 1858 * other processes using the vnode is renamed instead of removed and then 1859 * removed later on the last close. 1860 * - If v_usecount > 1 1861 * If a rename is not already in the works 1862 * call nfs_sillyrename() to set it up 1863 * else 1864 * do the remove rpc 1865 */ 1866 static int 1867 nfs_remove(struct vop_remove_args *ap) 1868 { 1869 struct vnode *vp = ap->a_vp; 1870 struct vnode *dvp = ap->a_dvp; 1871 struct componentname *cnp = ap->a_cnp; 1872 struct nfsnode *np = VTONFS(vp); 1873 int error = 0; 1874 struct vattr vattr; 1875 1876 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount")); 1877 if (vp->v_type == VDIR) 1878 error = EPERM; 1879 else if (vrefcnt(vp) == 1 || (np->n_sillyrename && 1880 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 1881 vattr.va_nlink > 1)) { 1882 /* 1883 * Purge the name cache so that the chance of a lookup for 1884 * the name succeeding while the remove is in progress is 1885 * minimized. Without node locking it can still happen, such 1886 * that an I/O op returns ESTALE, but since you get this if 1887 * another host removes the file.. 1888 */ 1889 cache_purge(vp); 1890 /* 1891 * throw away biocache buffers, mainly to avoid 1892 * unnecessary delayed writes later. 1893 */ 1894 error = ncl_vinvalbuf(vp, 0, curthread, 1); 1895 if (error != EINTR && error != EIO) 1896 /* Do the rpc */ 1897 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr, 1898 cnp->cn_namelen, cnp->cn_cred, curthread); 1899 /* 1900 * Kludge City: If the first reply to the remove rpc is lost.. 1901 * the reply to the retransmitted request will be ENOENT 1902 * since the file was in fact removed 1903 * Therefore, we cheat and return success. 1904 */ 1905 if (error == ENOENT) 1906 error = 0; 1907 } else if (!np->n_sillyrename) 1908 error = nfs_sillyrename(dvp, vp, cnp); 1909 NFSLOCKNODE(np); 1910 np->n_attrstamp = 0; 1911 NFSUNLOCKNODE(np); 1912 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 1913 return (error); 1914 } 1915 1916 /* 1917 * nfs file remove rpc called from nfs_inactive 1918 */ 1919 int 1920 ncl_removeit(struct sillyrename *sp, struct vnode *vp) 1921 { 1922 /* 1923 * Make sure that the directory vnode is still valid. 1924 * XXX we should lock sp->s_dvp here. 1925 */ 1926 if (sp->s_dvp->v_type == VBAD) 1927 return (0); 1928 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen, 1929 sp->s_cred, NULL)); 1930 } 1931 1932 /* 1933 * Nfs remove rpc, called from nfs_remove() and ncl_removeit(). 1934 */ 1935 static int 1936 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 1937 int namelen, struct ucred *cred, struct thread *td) 1938 { 1939 struct nfsvattr dnfsva; 1940 struct nfsnode *dnp = VTONFS(dvp); 1941 int error = 0, dattrflag; 1942 1943 NFSLOCKNODE(dnp); 1944 dnp->n_flag |= NREMOVEINPROG; 1945 NFSUNLOCKNODE(dnp); 1946 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva, 1947 &dattrflag); 1948 NFSLOCKNODE(dnp); 1949 if ((dnp->n_flag & NREMOVEWANT)) { 1950 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG); 1951 NFSUNLOCKNODE(dnp); 1952 wakeup((caddr_t)dnp); 1953 } else { 1954 dnp->n_flag &= ~NREMOVEINPROG; 1955 NFSUNLOCKNODE(dnp); 1956 } 1957 if (dattrflag) 1958 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 1959 NFSLOCKNODE(dnp); 1960 dnp->n_flag |= NMODIFIED; 1961 if (!dattrflag) { 1962 dnp->n_attrstamp = 0; 1963 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1964 } 1965 NFSUNLOCKNODE(dnp); 1966 if (error && NFS_ISV4(dvp)) 1967 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1968 return (error); 1969 } 1970 1971 /* 1972 * nfs file rename call 1973 */ 1974 static int 1975 nfs_rename(struct vop_rename_args *ap) 1976 { 1977 struct vnode *fvp = ap->a_fvp; 1978 struct vnode *tvp = ap->a_tvp; 1979 struct vnode *fdvp = ap->a_fdvp; 1980 struct vnode *tdvp = ap->a_tdvp; 1981 struct componentname *tcnp = ap->a_tcnp; 1982 struct componentname *fcnp = ap->a_fcnp; 1983 struct nfsnode *fnp = VTONFS(ap->a_fvp); 1984 struct nfsnode *tdnp = VTONFS(ap->a_tdvp); 1985 struct nfsv4node *newv4 = NULL; 1986 int error; 1987 1988 /* Check for cross-device rename */ 1989 if ((fvp->v_mount != tdvp->v_mount) || 1990 (tvp && (fvp->v_mount != tvp->v_mount))) { 1991 error = EXDEV; 1992 goto out; 1993 } 1994 1995 if (fvp == tvp) { 1996 printf("nfs_rename: fvp == tvp (can't happen)\n"); 1997 error = 0; 1998 goto out; 1999 } 2000 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0) 2001 goto out; 2002 2003 /* 2004 * We have to flush B_DELWRI data prior to renaming 2005 * the file. If we don't, the delayed-write buffers 2006 * can be flushed out later after the file has gone stale 2007 * under NFSV3. NFSV2 does not have this problem because 2008 * ( as far as I can tell ) it flushes dirty buffers more 2009 * often. 2010 * 2011 * Skip the rename operation if the fsync fails, this can happen 2012 * due to the server's volume being full, when we pushed out data 2013 * that was written back to our cache earlier. Not checking for 2014 * this condition can result in potential (silent) data loss. 2015 */ 2016 error = VOP_FSYNC(fvp, MNT_WAIT, curthread); 2017 NFSVOPUNLOCK(fvp); 2018 if (!error && tvp) 2019 error = VOP_FSYNC(tvp, MNT_WAIT, curthread); 2020 if (error) 2021 goto out; 2022 2023 /* 2024 * If the tvp exists and is in use, sillyrename it before doing the 2025 * rename of the new file over it. 2026 * XXX Can't sillyrename a directory. 2027 */ 2028 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename && 2029 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 2030 vput(tvp); 2031 tvp = NULL; 2032 } 2033 2034 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen, 2035 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 2036 curthread); 2037 2038 if (error == 0 && NFS_ISV4(tdvp)) { 2039 /* 2040 * For NFSv4, check to see if it is the same name and 2041 * replace the name, if it is different. 2042 */ 2043 newv4 = malloc( 2044 sizeof (struct nfsv4node) + 2045 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1, 2046 M_NFSV4NODE, M_WAITOK); 2047 NFSLOCKNODE(tdnp); 2048 NFSLOCKNODE(fnp); 2049 if (fnp->n_v4 != NULL && fvp->v_type == VREG && 2050 (fnp->n_v4->n4_namelen != tcnp->cn_namelen || 2051 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4), 2052 tcnp->cn_namelen) || 2053 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen || 2054 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 2055 tdnp->n_fhp->nfh_len))) { 2056 free(fnp->n_v4, M_NFSV4NODE); 2057 fnp->n_v4 = newv4; 2058 newv4 = NULL; 2059 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len; 2060 fnp->n_v4->n4_namelen = tcnp->cn_namelen; 2061 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 2062 tdnp->n_fhp->nfh_len); 2063 NFSBCOPY(tcnp->cn_nameptr, 2064 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen); 2065 } 2066 NFSUNLOCKNODE(tdnp); 2067 NFSUNLOCKNODE(fnp); 2068 if (newv4 != NULL) 2069 free(newv4, M_NFSV4NODE); 2070 } 2071 2072 if (fvp->v_type == VDIR) { 2073 if (tvp != NULL && tvp->v_type == VDIR) 2074 cache_purge(tdvp); 2075 cache_purge(fdvp); 2076 } 2077 2078 out: 2079 if (tdvp == tvp) 2080 vrele(tdvp); 2081 else 2082 vput(tdvp); 2083 if (tvp) 2084 vput(tvp); 2085 vrele(fdvp); 2086 vrele(fvp); 2087 /* 2088 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 2089 */ 2090 if (error == ENOENT) 2091 error = 0; 2092 return (error); 2093 } 2094 2095 /* 2096 * nfs file rename rpc called from nfs_remove() above 2097 */ 2098 static int 2099 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp, 2100 struct sillyrename *sp) 2101 { 2102 2103 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen, 2104 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred, 2105 curthread)); 2106 } 2107 2108 /* 2109 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 2110 */ 2111 static int 2112 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr, 2113 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr, 2114 int tnamelen, struct ucred *cred, struct thread *td) 2115 { 2116 struct nfsvattr fnfsva, tnfsva; 2117 struct nfsnode *fdnp = VTONFS(fdvp); 2118 struct nfsnode *tdnp = VTONFS(tdvp); 2119 int error = 0, fattrflag, tattrflag; 2120 2121 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp, 2122 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag, 2123 &tattrflag); 2124 NFSLOCKNODE(fdnp); 2125 fdnp->n_flag |= NMODIFIED; 2126 if (fattrflag != 0) { 2127 NFSUNLOCKNODE(fdnp); 2128 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, 0, 1); 2129 } else { 2130 fdnp->n_attrstamp = 0; 2131 NFSUNLOCKNODE(fdnp); 2132 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp); 2133 } 2134 NFSLOCKNODE(tdnp); 2135 tdnp->n_flag |= NMODIFIED; 2136 if (tattrflag != 0) { 2137 NFSUNLOCKNODE(tdnp); 2138 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, 0, 1); 2139 } else { 2140 tdnp->n_attrstamp = 0; 2141 NFSUNLOCKNODE(tdnp); 2142 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 2143 } 2144 if (error && NFS_ISV4(fdvp)) 2145 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2146 return (error); 2147 } 2148 2149 /* 2150 * nfs hard link create call 2151 */ 2152 static int 2153 nfs_link(struct vop_link_args *ap) 2154 { 2155 struct vnode *vp = ap->a_vp; 2156 struct vnode *tdvp = ap->a_tdvp; 2157 struct componentname *cnp = ap->a_cnp; 2158 struct nfsnode *np, *tdnp; 2159 struct nfsvattr nfsva, dnfsva; 2160 int error = 0, attrflag, dattrflag; 2161 2162 /* 2163 * Push all writes to the server, so that the attribute cache 2164 * doesn't get "out of sync" with the server. 2165 * XXX There should be a better way! 2166 */ 2167 VOP_FSYNC(vp, MNT_WAIT, curthread); 2168 2169 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 2170 cnp->cn_cred, curthread, &dnfsva, &nfsva, &attrflag, &dattrflag); 2171 tdnp = VTONFS(tdvp); 2172 NFSLOCKNODE(tdnp); 2173 tdnp->n_flag |= NMODIFIED; 2174 if (dattrflag != 0) { 2175 NFSUNLOCKNODE(tdnp); 2176 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, 0, 1); 2177 } else { 2178 tdnp->n_attrstamp = 0; 2179 NFSUNLOCKNODE(tdnp); 2180 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 2181 } 2182 if (attrflag) 2183 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 2184 else { 2185 np = VTONFS(vp); 2186 NFSLOCKNODE(np); 2187 np->n_attrstamp = 0; 2188 NFSUNLOCKNODE(np); 2189 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 2190 } 2191 /* 2192 * If negative lookup caching is enabled, I might as well 2193 * add an entry for this node. Not necessary for correctness, 2194 * but if negative caching is enabled, then the system 2195 * must care about lookup caching hit rate, so... 2196 */ 2197 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 && 2198 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 2199 if (tdvp != vp) 2200 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL); 2201 else 2202 printf("nfs_link: bogus NFS server returned " 2203 "the directory as the new link\n"); 2204 } 2205 if (error && NFS_ISV4(vp)) 2206 error = nfscl_maperr(curthread, error, (uid_t)0, 2207 (gid_t)0); 2208 return (error); 2209 } 2210 2211 /* 2212 * nfs symbolic link create call 2213 */ 2214 static int 2215 nfs_symlink(struct vop_symlink_args *ap) 2216 { 2217 struct vnode *dvp = ap->a_dvp; 2218 struct vattr *vap = ap->a_vap; 2219 struct componentname *cnp = ap->a_cnp; 2220 struct nfsvattr nfsva, dnfsva; 2221 struct nfsfh *nfhp; 2222 struct nfsnode *np = NULL, *dnp; 2223 struct vnode *newvp = NULL; 2224 int error = 0, attrflag, dattrflag, ret; 2225 2226 vap->va_type = VLNK; 2227 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2228 ap->a_target, vap, cnp->cn_cred, curthread, &dnfsva, 2229 &nfsva, &nfhp, &attrflag, &dattrflag); 2230 if (nfhp) { 2231 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, curthread, 2232 &np, LK_EXCLUSIVE); 2233 if (!ret) 2234 newvp = NFSTOV(np); 2235 else if (!error) 2236 error = ret; 2237 } 2238 if (newvp != NULL) { 2239 if (attrflag) 2240 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 2241 } else if (!error) { 2242 /* 2243 * If we do not have an error and we could not extract the 2244 * newvp from the response due to the request being NFSv2, we 2245 * have to do a lookup in order to obtain a newvp to return. 2246 */ 2247 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2248 cnp->cn_cred, curthread, &np); 2249 if (!error) 2250 newvp = NFSTOV(np); 2251 } 2252 if (error) { 2253 if (newvp) 2254 vput(newvp); 2255 if (NFS_ISV4(dvp)) 2256 error = nfscl_maperr(curthread, error, 2257 vap->va_uid, vap->va_gid); 2258 } else { 2259 *ap->a_vpp = newvp; 2260 } 2261 2262 dnp = VTONFS(dvp); 2263 NFSLOCKNODE(dnp); 2264 dnp->n_flag |= NMODIFIED; 2265 if (dattrflag != 0) { 2266 NFSUNLOCKNODE(dnp); 2267 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 2268 } else { 2269 dnp->n_attrstamp = 0; 2270 NFSUNLOCKNODE(dnp); 2271 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2272 } 2273 /* 2274 * If negative lookup caching is enabled, I might as well 2275 * add an entry for this node. Not necessary for correctness, 2276 * but if negative caching is enabled, then the system 2277 * must care about lookup caching hit rate, so... 2278 */ 2279 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2280 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 2281 if (dvp != newvp) 2282 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 2283 NULL); 2284 else 2285 printf("nfs_symlink: bogus NFS server returned " 2286 "the directory as the new file object\n"); 2287 } 2288 return (error); 2289 } 2290 2291 /* 2292 * nfs make dir call 2293 */ 2294 static int 2295 nfs_mkdir(struct vop_mkdir_args *ap) 2296 { 2297 struct vnode *dvp = ap->a_dvp; 2298 struct vattr *vap = ap->a_vap; 2299 struct componentname *cnp = ap->a_cnp; 2300 struct nfsnode *np = NULL, *dnp; 2301 struct vnode *newvp = NULL; 2302 struct vattr vattr; 2303 struct nfsfh *nfhp; 2304 struct nfsvattr nfsva, dnfsva; 2305 int error = 0, attrflag, dattrflag, ret; 2306 2307 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) 2308 return (error); 2309 vap->va_type = VDIR; 2310 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2311 vap, cnp->cn_cred, curthread, &dnfsva, &nfsva, &nfhp, 2312 &attrflag, &dattrflag); 2313 dnp = VTONFS(dvp); 2314 NFSLOCKNODE(dnp); 2315 dnp->n_flag |= NMODIFIED; 2316 if (dattrflag != 0) { 2317 NFSUNLOCKNODE(dnp); 2318 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 2319 } else { 2320 dnp->n_attrstamp = 0; 2321 NFSUNLOCKNODE(dnp); 2322 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2323 } 2324 if (nfhp) { 2325 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, curthread, 2326 &np, LK_EXCLUSIVE); 2327 if (!ret) { 2328 newvp = NFSTOV(np); 2329 if (attrflag) 2330 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 2331 0, 1); 2332 } else if (!error) 2333 error = ret; 2334 } 2335 if (!error && newvp == NULL) { 2336 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2337 cnp->cn_cred, curthread, &np); 2338 if (!error) { 2339 newvp = NFSTOV(np); 2340 if (newvp->v_type != VDIR) 2341 error = EEXIST; 2342 } 2343 } 2344 if (error) { 2345 if (newvp) 2346 vput(newvp); 2347 if (NFS_ISV4(dvp)) 2348 error = nfscl_maperr(curthread, error, 2349 vap->va_uid, vap->va_gid); 2350 } else { 2351 /* 2352 * If negative lookup caching is enabled, I might as well 2353 * add an entry for this node. Not necessary for correctness, 2354 * but if negative caching is enabled, then the system 2355 * must care about lookup caching hit rate, so... 2356 */ 2357 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2358 (cnp->cn_flags & MAKEENTRY) && 2359 attrflag != 0 && dattrflag != 0) { 2360 if (dvp != newvp) 2361 cache_enter_time(dvp, newvp, cnp, 2362 &nfsva.na_ctime, &dnfsva.na_ctime); 2363 else 2364 printf("nfs_mkdir: bogus NFS server returned " 2365 "the directory that the directory was " 2366 "created in as the new file object\n"); 2367 } 2368 *ap->a_vpp = newvp; 2369 } 2370 return (error); 2371 } 2372 2373 /* 2374 * nfs remove directory call 2375 */ 2376 static int 2377 nfs_rmdir(struct vop_rmdir_args *ap) 2378 { 2379 struct vnode *vp = ap->a_vp; 2380 struct vnode *dvp = ap->a_dvp; 2381 struct componentname *cnp = ap->a_cnp; 2382 struct nfsnode *dnp; 2383 struct nfsvattr dnfsva; 2384 int error, dattrflag; 2385 2386 if (dvp == vp) 2387 return (EINVAL); 2388 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2389 cnp->cn_cred, curthread, &dnfsva, &dattrflag); 2390 dnp = VTONFS(dvp); 2391 NFSLOCKNODE(dnp); 2392 dnp->n_flag |= NMODIFIED; 2393 if (dattrflag != 0) { 2394 NFSUNLOCKNODE(dnp); 2395 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 2396 } else { 2397 dnp->n_attrstamp = 0; 2398 NFSUNLOCKNODE(dnp); 2399 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2400 } 2401 2402 cache_purge(dvp); 2403 cache_purge(vp); 2404 if (error && NFS_ISV4(dvp)) 2405 error = nfscl_maperr(curthread, error, (uid_t)0, 2406 (gid_t)0); 2407 /* 2408 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2409 */ 2410 if (error == ENOENT) 2411 error = 0; 2412 return (error); 2413 } 2414 2415 /* 2416 * nfs readdir call 2417 */ 2418 static int 2419 nfs_readdir(struct vop_readdir_args *ap) 2420 { 2421 struct vnode *vp = ap->a_vp; 2422 struct nfsnode *np = VTONFS(vp); 2423 struct uio *uio = ap->a_uio; 2424 ssize_t tresid, left; 2425 int error = 0; 2426 struct vattr vattr; 2427 2428 if (ap->a_eofflag != NULL) 2429 *ap->a_eofflag = 0; 2430 if (vp->v_type != VDIR) 2431 return(EPERM); 2432 2433 /* 2434 * First, check for hit on the EOF offset cache 2435 */ 2436 NFSLOCKNODE(np); 2437 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2438 (np->n_flag & NMODIFIED) == 0) { 2439 NFSUNLOCKNODE(np); 2440 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) { 2441 NFSLOCKNODE(np); 2442 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) || 2443 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 2444 NFSUNLOCKNODE(np); 2445 NFSINCRGLOBAL(nfsstatsv1.direofcache_hits); 2446 if (ap->a_eofflag != NULL) 2447 *ap->a_eofflag = 1; 2448 return (0); 2449 } else 2450 NFSUNLOCKNODE(np); 2451 } 2452 } else 2453 NFSUNLOCKNODE(np); 2454 2455 /* 2456 * NFS always guarantees that directory entries don't straddle 2457 * DIRBLKSIZ boundaries. As such, we need to limit the size 2458 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial 2459 * directory entry. 2460 */ 2461 left = uio->uio_resid % DIRBLKSIZ; 2462 if (left == uio->uio_resid) 2463 return (EINVAL); 2464 uio->uio_resid -= left; 2465 2466 /* 2467 * For readdirplus, if starting to read the directory, 2468 * purge the name cache, since it will be reloaded by 2469 * this directory read. 2470 * This removes potentially stale name cache entries. 2471 */ 2472 if (uio->uio_offset == 0 && 2473 (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_RDIRPLUS) != 0) 2474 cache_purge(vp); 2475 2476 /* 2477 * Call ncl_bioread() to do the real work. 2478 */ 2479 tresid = uio->uio_resid; 2480 error = ncl_bioread(vp, uio, 0, ap->a_cred); 2481 2482 if (!error && uio->uio_resid == tresid) { 2483 NFSINCRGLOBAL(nfsstatsv1.direofcache_misses); 2484 if (ap->a_eofflag != NULL) 2485 *ap->a_eofflag = 1; 2486 } 2487 2488 /* Add the partial DIRBLKSIZ (left) back in. */ 2489 uio->uio_resid += left; 2490 return (error); 2491 } 2492 2493 /* 2494 * Readdir rpc call. 2495 * Called from below the buffer cache by ncl_doio(). 2496 */ 2497 int 2498 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2499 struct thread *td) 2500 { 2501 struct nfsvattr nfsva; 2502 nfsuint64 *cookiep, cookie; 2503 struct nfsnode *dnp = VTONFS(vp); 2504 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2505 int error = 0, eof, attrflag; 2506 2507 KASSERT(uiop->uio_iovcnt == 1 && 2508 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2509 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2510 ("nfs readdirrpc bad uio")); 2511 2512 /* 2513 * If there is no cookie, assume directory was stale. 2514 */ 2515 ncl_dircookie_lock(dnp); 2516 NFSUNLOCKNODE(dnp); 2517 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2518 if (cookiep) { 2519 cookie = *cookiep; 2520 ncl_dircookie_unlock(dnp); 2521 } else { 2522 ncl_dircookie_unlock(dnp); 2523 return (NFSERR_BAD_COOKIE); 2524 } 2525 2526 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2527 (void)ncl_fsinfo(nmp, vp, cred, td); 2528 2529 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva, 2530 &attrflag, &eof); 2531 if (attrflag) 2532 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 2533 2534 if (!error) { 2535 /* 2536 * We are now either at the end of the directory or have filled 2537 * the block. 2538 */ 2539 if (eof) { 2540 NFSLOCKNODE(dnp); 2541 dnp->n_direofoffset = uiop->uio_offset; 2542 NFSUNLOCKNODE(dnp); 2543 } else { 2544 if (uiop->uio_resid > 0) 2545 printf("EEK! readdirrpc resid > 0\n"); 2546 ncl_dircookie_lock(dnp); 2547 NFSUNLOCKNODE(dnp); 2548 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2549 *cookiep = cookie; 2550 ncl_dircookie_unlock(dnp); 2551 } 2552 } else if (NFS_ISV4(vp)) { 2553 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2554 } 2555 return (error); 2556 } 2557 2558 /* 2559 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc(). 2560 */ 2561 int 2562 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2563 struct thread *td) 2564 { 2565 struct nfsvattr nfsva; 2566 nfsuint64 *cookiep, cookie; 2567 struct nfsnode *dnp = VTONFS(vp); 2568 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2569 int error = 0, attrflag, eof; 2570 2571 KASSERT(uiop->uio_iovcnt == 1 && 2572 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2573 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2574 ("nfs readdirplusrpc bad uio")); 2575 2576 /* 2577 * If there is no cookie, assume directory was stale. 2578 */ 2579 ncl_dircookie_lock(dnp); 2580 NFSUNLOCKNODE(dnp); 2581 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2582 if (cookiep) { 2583 cookie = *cookiep; 2584 ncl_dircookie_unlock(dnp); 2585 } else { 2586 ncl_dircookie_unlock(dnp); 2587 return (NFSERR_BAD_COOKIE); 2588 } 2589 2590 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2591 (void)ncl_fsinfo(nmp, vp, cred, td); 2592 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva, 2593 &attrflag, &eof); 2594 if (attrflag) 2595 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 2596 2597 if (!error) { 2598 /* 2599 * We are now either at end of the directory or have filled the 2600 * the block. 2601 */ 2602 if (eof) { 2603 NFSLOCKNODE(dnp); 2604 dnp->n_direofoffset = uiop->uio_offset; 2605 NFSUNLOCKNODE(dnp); 2606 } else { 2607 if (uiop->uio_resid > 0) 2608 printf("EEK! readdirplusrpc resid > 0\n"); 2609 ncl_dircookie_lock(dnp); 2610 NFSUNLOCKNODE(dnp); 2611 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2612 *cookiep = cookie; 2613 ncl_dircookie_unlock(dnp); 2614 } 2615 } else if (NFS_ISV4(vp)) { 2616 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2617 } 2618 return (error); 2619 } 2620 2621 /* 2622 * Silly rename. To make the NFS filesystem that is stateless look a little 2623 * more like the "ufs" a remove of an active vnode is translated to a rename 2624 * to a funny looking filename that is removed by nfs_inactive on the 2625 * nfsnode. There is the potential for another process on a different client 2626 * to create the same funny name between the nfs_lookitup() fails and the 2627 * nfs_rename() completes, but... 2628 */ 2629 static int 2630 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2631 { 2632 struct sillyrename *sp; 2633 struct nfsnode *np; 2634 int error; 2635 short pid; 2636 unsigned int lticks; 2637 2638 cache_purge(dvp); 2639 np = VTONFS(vp); 2640 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir")); 2641 sp = malloc(sizeof (struct sillyrename), 2642 M_NEWNFSREQ, M_WAITOK); 2643 sp->s_cred = crhold(cnp->cn_cred); 2644 sp->s_dvp = dvp; 2645 VREF(dvp); 2646 2647 /* 2648 * Fudge together a funny name. 2649 * Changing the format of the funny name to accommodate more 2650 * sillynames per directory. 2651 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is 2652 * CPU ticks since boot. 2653 */ 2654 pid = curthread->td_proc->p_pid; 2655 lticks = (unsigned int)ticks; 2656 for ( ; ; ) { 2657 sp->s_namlen = sprintf(sp->s_name, 2658 ".nfs.%08x.%04x4.4", lticks, 2659 pid); 2660 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2661 curthread, NULL)) 2662 break; 2663 lticks++; 2664 } 2665 error = nfs_renameit(dvp, vp, cnp, sp); 2666 if (error) 2667 goto bad; 2668 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2669 curthread, &np); 2670 np->n_sillyrename = sp; 2671 return (0); 2672 bad: 2673 vrele(sp->s_dvp); 2674 crfree(sp->s_cred); 2675 free(sp, M_NEWNFSREQ); 2676 return (error); 2677 } 2678 2679 /* 2680 * Look up a file name and optionally either update the file handle or 2681 * allocate an nfsnode, depending on the value of npp. 2682 * npp == NULL --> just do the lookup 2683 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2684 * handled too 2685 * *npp != NULL --> update the file handle in the vnode 2686 */ 2687 static int 2688 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred, 2689 struct thread *td, struct nfsnode **npp) 2690 { 2691 struct vnode *newvp = NULL, *vp; 2692 struct nfsnode *np, *dnp = VTONFS(dvp); 2693 struct nfsfh *nfhp, *onfhp; 2694 struct nfsvattr nfsva, dnfsva; 2695 struct componentname cn; 2696 int error = 0, attrflag, dattrflag; 2697 u_int hash; 2698 struct timespec ts; 2699 2700 nanouptime(&ts); 2701 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva, 2702 &nfhp, &attrflag, &dattrflag, 0); 2703 if (dattrflag) 2704 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 2705 if (npp && !error) { 2706 if (*npp != NULL) { 2707 np = *npp; 2708 vp = NFSTOV(np); 2709 /* 2710 * For NFSv4, check to see if it is the same name and 2711 * replace the name, if it is different. 2712 */ 2713 if (np->n_v4 != NULL && nfsva.na_type == VREG && 2714 (np->n_v4->n4_namelen != len || 2715 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) || 2716 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || 2717 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2718 dnp->n_fhp->nfh_len))) { 2719 free(np->n_v4, M_NFSV4NODE); 2720 np->n_v4 = malloc( 2721 sizeof (struct nfsv4node) + 2722 dnp->n_fhp->nfh_len + len - 1, 2723 M_NFSV4NODE, M_WAITOK); 2724 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; 2725 np->n_v4->n4_namelen = len; 2726 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2727 dnp->n_fhp->nfh_len); 2728 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len); 2729 } 2730 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, 2731 FNV1_32_INIT); 2732 onfhp = np->n_fhp; 2733 /* 2734 * Rehash node for new file handle. 2735 */ 2736 vfs_hash_rehash(vp, hash); 2737 np->n_fhp = nfhp; 2738 if (onfhp != NULL) 2739 free(onfhp, M_NFSFH); 2740 newvp = NFSTOV(np); 2741 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) { 2742 free(nfhp, M_NFSFH); 2743 VREF(dvp); 2744 newvp = dvp; 2745 } else { 2746 cn.cn_nameptr = name; 2747 cn.cn_namelen = len; 2748 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td, 2749 &np, LK_EXCLUSIVE); 2750 if (error) 2751 return (error); 2752 newvp = NFSTOV(np); 2753 /* 2754 * If n_localmodtime >= time before RPC, then 2755 * a file modification operation, such as 2756 * VOP_SETATTR() of size, has occurred while 2757 * the Lookup RPC and acquisition of the vnode 2758 * happened. As such, the attributes might 2759 * be stale, with possibly an incorrect size. 2760 */ 2761 NFSLOCKNODE(np); 2762 if (timespecisset(&np->n_localmodtime) && 2763 timespeccmp(&np->n_localmodtime, &ts, >=)) { 2764 NFSCL_DEBUG(4, "nfs_lookitup: localmod " 2765 "stale attributes\n"); 2766 attrflag = 0; 2767 } 2768 NFSUNLOCKNODE(np); 2769 } 2770 if (!attrflag && *npp == NULL) { 2771 if (newvp == dvp) 2772 vrele(newvp); 2773 else 2774 vput(newvp); 2775 return (ENOENT); 2776 } 2777 if (attrflag) 2778 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 2779 } 2780 if (npp && *npp == NULL) { 2781 if (error) { 2782 if (newvp) { 2783 if (newvp == dvp) 2784 vrele(newvp); 2785 else 2786 vput(newvp); 2787 } 2788 } else 2789 *npp = np; 2790 } 2791 if (error && NFS_ISV4(dvp)) 2792 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2793 return (error); 2794 } 2795 2796 /* 2797 * Nfs Version 3 and 4 commit rpc 2798 */ 2799 int 2800 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred, 2801 struct thread *td) 2802 { 2803 struct nfsvattr nfsva; 2804 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2805 struct nfsnode *np; 2806 struct uio uio; 2807 int error, attrflag; 2808 2809 np = VTONFS(vp); 2810 error = EIO; 2811 attrflag = 0; 2812 if (NFSHASPNFS(nmp) && (np->n_flag & NDSCOMMIT) != 0) { 2813 uio.uio_offset = offset; 2814 uio.uio_resid = cnt; 2815 error = nfscl_doiods(vp, &uio, NULL, NULL, 2816 NFSV4OPEN_ACCESSWRITE, 1, cred, td); 2817 if (error != 0) { 2818 NFSLOCKNODE(np); 2819 np->n_flag &= ~NDSCOMMIT; 2820 NFSUNLOCKNODE(np); 2821 } 2822 } 2823 if (error != 0) { 2824 mtx_lock(&nmp->nm_mtx); 2825 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) { 2826 mtx_unlock(&nmp->nm_mtx); 2827 return (0); 2828 } 2829 mtx_unlock(&nmp->nm_mtx); 2830 error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva, 2831 &attrflag); 2832 } 2833 if (attrflag != 0) 2834 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 2835 if (error != 0 && NFS_ISV4(vp)) 2836 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2837 return (error); 2838 } 2839 2840 /* 2841 * Strategy routine. 2842 * For async requests when nfsiod(s) are running, queue the request by 2843 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the 2844 * request. 2845 */ 2846 static int 2847 nfs_strategy(struct vop_strategy_args *ap) 2848 { 2849 struct buf *bp; 2850 struct vnode *vp; 2851 struct ucred *cr; 2852 2853 bp = ap->a_bp; 2854 vp = ap->a_vp; 2855 KASSERT(bp->b_vp == vp, ("missing b_getvp")); 2856 KASSERT(!(bp->b_flags & B_DONE), 2857 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); 2858 2859 if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno) 2860 bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize / 2861 DEV_BSIZE); 2862 if (bp->b_iocmd == BIO_READ) 2863 cr = bp->b_rcred; 2864 else 2865 cr = bp->b_wcred; 2866 2867 /* 2868 * If the op is asynchronous and an i/o daemon is waiting 2869 * queue the request, wake it up and wait for completion 2870 * otherwise just do it ourselves. 2871 */ 2872 if ((bp->b_flags & B_ASYNC) == 0 || 2873 ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread)) 2874 (void) ncl_doio(vp, bp, cr, curthread, 1); 2875 return (0); 2876 } 2877 2878 /* 2879 * fsync vnode op. Just call ncl_flush() with commit == 1. 2880 */ 2881 /* ARGSUSED */ 2882 static int 2883 nfs_fsync(struct vop_fsync_args *ap) 2884 { 2885 2886 if (ap->a_vp->v_type != VREG) { 2887 /* 2888 * For NFS, metadata is changed synchronously on the server, 2889 * so there is nothing to flush. Also, ncl_flush() clears 2890 * the NMODIFIED flag and that shouldn't be done here for 2891 * directories. 2892 */ 2893 return (0); 2894 } 2895 return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0)); 2896 } 2897 2898 /* 2899 * Flush all the blocks associated with a vnode. 2900 * Walk through the buffer pool and push any dirty pages 2901 * associated with the vnode. 2902 * If the called_from_renewthread argument is TRUE, it has been called 2903 * from the NFSv4 renew thread and, as such, cannot block indefinitely 2904 * waiting for a buffer write to complete. 2905 */ 2906 int 2907 ncl_flush(struct vnode *vp, int waitfor, struct thread *td, 2908 int commit, int called_from_renewthread) 2909 { 2910 struct nfsnode *np = VTONFS(vp); 2911 struct buf *bp; 2912 int i; 2913 struct buf *nbp; 2914 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2915 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos; 2916 int passone = 1, trycnt = 0; 2917 u_quad_t off, endoff, toff; 2918 struct ucred* wcred = NULL; 2919 struct buf **bvec = NULL; 2920 struct bufobj *bo; 2921 #ifndef NFS_COMMITBVECSIZ 2922 #define NFS_COMMITBVECSIZ 20 2923 #endif 2924 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ]; 2925 u_int bvecsize = 0, bveccount; 2926 struct timespec ts; 2927 2928 if (called_from_renewthread != 0) 2929 slptimeo = hz; 2930 if (nmp->nm_flag & NFSMNT_INT) 2931 slpflag = PCATCH; 2932 if (!commit) 2933 passone = 0; 2934 bo = &vp->v_bufobj; 2935 /* 2936 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the 2937 * server, but has not been committed to stable storage on the server 2938 * yet. On the first pass, the byte range is worked out and the commit 2939 * rpc is done. On the second pass, bwrite() is called to do the 2940 * job. 2941 */ 2942 again: 2943 off = (u_quad_t)-1; 2944 endoff = 0; 2945 bvecpos = 0; 2946 if (NFS_ISV34(vp) && commit) { 2947 if (bvec != NULL && bvec != bvec_on_stack) 2948 free(bvec, M_TEMP); 2949 /* 2950 * Count up how many buffers waiting for a commit. 2951 */ 2952 bveccount = 0; 2953 BO_LOCK(bo); 2954 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2955 if (!BUF_ISLOCKED(bp) && 2956 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) 2957 == (B_DELWRI | B_NEEDCOMMIT)) 2958 bveccount++; 2959 } 2960 /* 2961 * Allocate space to remember the list of bufs to commit. It is 2962 * important to use M_NOWAIT here to avoid a race with nfs_write. 2963 * If we can't get memory (for whatever reason), we will end up 2964 * committing the buffers one-by-one in the loop below. 2965 */ 2966 if (bveccount > NFS_COMMITBVECSIZ) { 2967 /* 2968 * Release the vnode interlock to avoid a lock 2969 * order reversal. 2970 */ 2971 BO_UNLOCK(bo); 2972 bvec = (struct buf **) 2973 malloc(bveccount * sizeof(struct buf *), 2974 M_TEMP, M_NOWAIT); 2975 BO_LOCK(bo); 2976 if (bvec == NULL) { 2977 bvec = bvec_on_stack; 2978 bvecsize = NFS_COMMITBVECSIZ; 2979 } else 2980 bvecsize = bveccount; 2981 } else { 2982 bvec = bvec_on_stack; 2983 bvecsize = NFS_COMMITBVECSIZ; 2984 } 2985 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2986 if (bvecpos >= bvecsize) 2987 break; 2988 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2989 nbp = TAILQ_NEXT(bp, b_bobufs); 2990 continue; 2991 } 2992 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 2993 (B_DELWRI | B_NEEDCOMMIT)) { 2994 BUF_UNLOCK(bp); 2995 nbp = TAILQ_NEXT(bp, b_bobufs); 2996 continue; 2997 } 2998 BO_UNLOCK(bo); 2999 bremfree(bp); 3000 /* 3001 * Work out if all buffers are using the same cred 3002 * so we can deal with them all with one commit. 3003 * 3004 * NOTE: we are not clearing B_DONE here, so we have 3005 * to do it later on in this routine if we intend to 3006 * initiate I/O on the bp. 3007 * 3008 * Note: to avoid loopback deadlocks, we do not 3009 * assign b_runningbufspace. 3010 */ 3011 if (wcred == NULL) 3012 wcred = bp->b_wcred; 3013 else if (wcred != bp->b_wcred) 3014 wcred = NOCRED; 3015 vfs_busy_pages(bp, 0); 3016 3017 BO_LOCK(bo); 3018 /* 3019 * bp is protected by being locked, but nbp is not 3020 * and vfs_busy_pages() may sleep. We have to 3021 * recalculate nbp. 3022 */ 3023 nbp = TAILQ_NEXT(bp, b_bobufs); 3024 3025 /* 3026 * A list of these buffers is kept so that the 3027 * second loop knows which buffers have actually 3028 * been committed. This is necessary, since there 3029 * may be a race between the commit rpc and new 3030 * uncommitted writes on the file. 3031 */ 3032 bvec[bvecpos++] = bp; 3033 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 3034 bp->b_dirtyoff; 3035 if (toff < off) 3036 off = toff; 3037 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); 3038 if (toff > endoff) 3039 endoff = toff; 3040 } 3041 BO_UNLOCK(bo); 3042 } 3043 if (bvecpos > 0) { 3044 /* 3045 * Commit data on the server, as required. 3046 * If all bufs are using the same wcred, then use that with 3047 * one call for all of them, otherwise commit each one 3048 * separately. 3049 */ 3050 if (wcred != NOCRED) 3051 retv = ncl_commit(vp, off, (int)(endoff - off), 3052 wcred, td); 3053 else { 3054 retv = 0; 3055 for (i = 0; i < bvecpos; i++) { 3056 off_t off, size; 3057 bp = bvec[i]; 3058 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 3059 bp->b_dirtyoff; 3060 size = (u_quad_t)(bp->b_dirtyend 3061 - bp->b_dirtyoff); 3062 retv = ncl_commit(vp, off, (int)size, 3063 bp->b_wcred, td); 3064 if (retv) break; 3065 } 3066 } 3067 3068 if (retv == NFSERR_STALEWRITEVERF) 3069 ncl_clearcommit(vp->v_mount); 3070 3071 /* 3072 * Now, either mark the blocks I/O done or mark the 3073 * blocks dirty, depending on whether the commit 3074 * succeeded. 3075 */ 3076 for (i = 0; i < bvecpos; i++) { 3077 bp = bvec[i]; 3078 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 3079 if (!NFSCL_FORCEDISM(vp->v_mount) && retv) { 3080 /* 3081 * Error, leave B_DELWRI intact 3082 */ 3083 vfs_unbusy_pages(bp); 3084 brelse(bp); 3085 } else { 3086 /* 3087 * Success, remove B_DELWRI ( bundirty() ). 3088 * 3089 * b_dirtyoff/b_dirtyend seem to be NFS 3090 * specific. We should probably move that 3091 * into bundirty(). XXX 3092 */ 3093 bufobj_wref(bo); 3094 bp->b_flags |= B_ASYNC; 3095 bundirty(bp); 3096 bp->b_flags &= ~B_DONE; 3097 bp->b_ioflags &= ~BIO_ERROR; 3098 bp->b_dirtyoff = bp->b_dirtyend = 0; 3099 bufdone(bp); 3100 } 3101 } 3102 } 3103 3104 /* 3105 * Start/do any write(s) that are required. 3106 */ 3107 loop: 3108 BO_LOCK(bo); 3109 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 3110 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 3111 if (waitfor != MNT_WAIT || passone) 3112 continue; 3113 3114 error = BUF_TIMELOCK(bp, 3115 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 3116 BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo); 3117 if (error == 0) { 3118 BUF_UNLOCK(bp); 3119 goto loop; 3120 } 3121 if (error == ENOLCK) { 3122 error = 0; 3123 goto loop; 3124 } 3125 if (called_from_renewthread != 0) { 3126 /* 3127 * Return EIO so the flush will be retried 3128 * later. 3129 */ 3130 error = EIO; 3131 goto done; 3132 } 3133 if (newnfs_sigintr(nmp, td)) { 3134 error = EINTR; 3135 goto done; 3136 } 3137 if (slpflag == PCATCH) { 3138 slpflag = 0; 3139 slptimeo = 2 * hz; 3140 } 3141 goto loop; 3142 } 3143 if ((bp->b_flags & B_DELWRI) == 0) 3144 panic("nfs_fsync: not dirty"); 3145 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) { 3146 BUF_UNLOCK(bp); 3147 continue; 3148 } 3149 BO_UNLOCK(bo); 3150 bremfree(bp); 3151 bp->b_flags |= B_ASYNC; 3152 bwrite(bp); 3153 if (newnfs_sigintr(nmp, td)) { 3154 error = EINTR; 3155 goto done; 3156 } 3157 goto loop; 3158 } 3159 if (passone) { 3160 passone = 0; 3161 BO_UNLOCK(bo); 3162 goto again; 3163 } 3164 if (waitfor == MNT_WAIT) { 3165 while (bo->bo_numoutput) { 3166 error = bufobj_wwait(bo, slpflag, slptimeo); 3167 if (error) { 3168 BO_UNLOCK(bo); 3169 if (called_from_renewthread != 0) { 3170 /* 3171 * Return EIO so that the flush will be 3172 * retried later. 3173 */ 3174 error = EIO; 3175 goto done; 3176 } 3177 error = newnfs_sigintr(nmp, td); 3178 if (error) 3179 goto done; 3180 if (slpflag == PCATCH) { 3181 slpflag = 0; 3182 slptimeo = 2 * hz; 3183 } 3184 BO_LOCK(bo); 3185 } 3186 } 3187 if (bo->bo_dirty.bv_cnt != 0 && commit) { 3188 BO_UNLOCK(bo); 3189 goto loop; 3190 } 3191 /* 3192 * Wait for all the async IO requests to drain 3193 */ 3194 BO_UNLOCK(bo); 3195 } else 3196 BO_UNLOCK(bo); 3197 if (NFSHASPNFS(nmp)) { 3198 nfscl_layoutcommit(vp, td); 3199 /* 3200 * Invalidate the attribute cache, since writes to a DS 3201 * won't update the size attribute. 3202 */ 3203 NFSLOCKNODE(np); 3204 np->n_attrstamp = 0; 3205 } else 3206 NFSLOCKNODE(np); 3207 if (np->n_flag & NWRITEERR) { 3208 error = np->n_error; 3209 np->n_flag &= ~NWRITEERR; 3210 } 3211 if (commit && bo->bo_dirty.bv_cnt == 0 && 3212 bo->bo_numoutput == 0) 3213 np->n_flag &= ~NMODIFIED; 3214 NFSUNLOCKNODE(np); 3215 done: 3216 if (bvec != NULL && bvec != bvec_on_stack) 3217 free(bvec, M_TEMP); 3218 if (error == 0 && commit != 0 && waitfor == MNT_WAIT && 3219 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0)) { 3220 if (trycnt++ < 5) { 3221 /* try, try again... */ 3222 passone = 1; 3223 wcred = NULL; 3224 bvec = NULL; 3225 bvecsize = 0; 3226 goto again; 3227 } 3228 vn_printf(vp, "ncl_flush failed"); 3229 error = called_from_renewthread != 0 ? EIO : EBUSY; 3230 } 3231 if (error == 0) { 3232 nanouptime(&ts); 3233 NFSLOCKNODE(np); 3234 np->n_localmodtime = ts; 3235 NFSUNLOCKNODE(np); 3236 } 3237 return (error); 3238 } 3239 3240 /* 3241 * NFS advisory byte-level locks. 3242 */ 3243 static int 3244 nfs_advlock(struct vop_advlock_args *ap) 3245 { 3246 struct vnode *vp = ap->a_vp; 3247 struct ucred *cred; 3248 struct nfsnode *np = VTONFS(ap->a_vp); 3249 struct proc *p = (struct proc *)ap->a_id; 3250 struct thread *td = curthread; /* XXX */ 3251 struct vattr va; 3252 int ret, error; 3253 u_quad_t size; 3254 struct nfsmount *nmp; 3255 3256 error = NFSVOPLOCK(vp, LK_SHARED); 3257 if (error != 0) 3258 return (EBADF); 3259 nmp = VFSTONFS(vp->v_mount); 3260 if (!NFS_ISV4(vp) || (nmp->nm_flag & NFSMNT_NOLOCKD) != 0) { 3261 if ((nmp->nm_flag & NFSMNT_NOLOCKD) != 0) { 3262 size = np->n_size; 3263 NFSVOPUNLOCK(vp); 3264 error = lf_advlock(ap, &(vp->v_lockf), size); 3265 } else { 3266 if (nfs_advlock_p != NULL) 3267 error = nfs_advlock_p(ap); 3268 else { 3269 NFSVOPUNLOCK(vp); 3270 error = ENOLCK; 3271 } 3272 } 3273 if (error == 0 && ap->a_op == F_SETLK) { 3274 error = NFSVOPLOCK(vp, LK_SHARED); 3275 if (error == 0) { 3276 /* Mark that a file lock has been acquired. */ 3277 NFSLOCKNODE(np); 3278 np->n_flag |= NHASBEENLOCKED; 3279 NFSUNLOCKNODE(np); 3280 NFSVOPUNLOCK(vp); 3281 } 3282 } 3283 return (error); 3284 } else if ((ap->a_flags & (F_POSIX | F_FLOCK)) != 0) { 3285 if (vp->v_type != VREG) { 3286 error = EINVAL; 3287 goto out; 3288 } 3289 if ((ap->a_flags & F_POSIX) != 0) 3290 cred = p->p_ucred; 3291 else 3292 cred = td->td_ucred; 3293 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 3294 if (VN_IS_DOOMED(vp)) { 3295 error = EBADF; 3296 goto out; 3297 } 3298 3299 /* 3300 * If this is unlocking a write locked region, flush and 3301 * commit them before unlocking. This is required by 3302 * RFC3530 Sec. 9.3.2. 3303 */ 3304 if (ap->a_op == F_UNLCK && 3305 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id, 3306 ap->a_flags)) 3307 (void) ncl_flush(vp, MNT_WAIT, td, 1, 0); 3308 3309 /* 3310 * Mark NFS node as might have acquired a lock. 3311 * This is separate from NHASBEENLOCKED, because it must 3312 * be done before the nfsrpc_advlock() call, which might 3313 * add a nfscllock structure to the client state. 3314 * It is used to check for the case where a nfscllock 3315 * state structure cannot exist for the file. 3316 * Only done for "oneopenown" NFSv4.1/4.2 mounts. 3317 */ 3318 if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp)) { 3319 NFSLOCKNODE(np); 3320 np->n_flag |= NMIGHTBELOCKED; 3321 NFSUNLOCKNODE(np); 3322 } 3323 3324 /* 3325 * Loop around doing the lock op, while a blocking lock 3326 * must wait for the lock op to succeed. 3327 */ 3328 do { 3329 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op, 3330 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags); 3331 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3332 ap->a_op == F_SETLK) { 3333 NFSVOPUNLOCK(vp); 3334 error = nfs_catnap(PZERO | PCATCH, ret, 3335 "ncladvl"); 3336 if (error) 3337 return (EINTR); 3338 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 3339 if (VN_IS_DOOMED(vp)) { 3340 error = EBADF; 3341 goto out; 3342 } 3343 } 3344 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3345 ap->a_op == F_SETLK); 3346 if (ret == NFSERR_DENIED) { 3347 error = EAGAIN; 3348 goto out; 3349 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) { 3350 error = ret; 3351 goto out; 3352 } else if (ret != 0) { 3353 error = EACCES; 3354 goto out; 3355 } 3356 3357 /* 3358 * Now, if we just got a lock, invalidate data in the buffer 3359 * cache, as required, so that the coherency conforms with 3360 * RFC3530 Sec. 9.3.2. 3361 */ 3362 if (ap->a_op == F_SETLK) { 3363 if ((np->n_flag & NMODIFIED) == 0) { 3364 np->n_attrstamp = 0; 3365 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3366 ret = VOP_GETATTR(vp, &va, cred); 3367 } 3368 if ((np->n_flag & NMODIFIED) || ret || 3369 np->n_change != va.va_filerev) { 3370 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1); 3371 np->n_attrstamp = 0; 3372 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3373 ret = VOP_GETATTR(vp, &va, cred); 3374 if (!ret) { 3375 np->n_mtime = va.va_mtime; 3376 np->n_change = va.va_filerev; 3377 } 3378 } 3379 /* Mark that a file lock has been acquired. */ 3380 NFSLOCKNODE(np); 3381 np->n_flag |= NHASBEENLOCKED; 3382 NFSUNLOCKNODE(np); 3383 } 3384 } else 3385 error = EOPNOTSUPP; 3386 out: 3387 NFSVOPUNLOCK(vp); 3388 return (error); 3389 } 3390 3391 /* 3392 * NFS advisory byte-level locks. 3393 */ 3394 static int 3395 nfs_advlockasync(struct vop_advlockasync_args *ap) 3396 { 3397 struct vnode *vp = ap->a_vp; 3398 u_quad_t size; 3399 int error; 3400 3401 error = NFSVOPLOCK(vp, LK_SHARED); 3402 if (error) 3403 return (error); 3404 if (NFS_ISV4(vp)) { 3405 NFSVOPUNLOCK(vp); 3406 return (EOPNOTSUPP); 3407 } 3408 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3409 size = VTONFS(vp)->n_size; 3410 NFSVOPUNLOCK(vp); 3411 error = lf_advlockasync(ap, &(vp->v_lockf), size); 3412 } else { 3413 NFSVOPUNLOCK(vp); 3414 error = EOPNOTSUPP; 3415 } 3416 return (error); 3417 } 3418 3419 /* 3420 * Print out the contents of an nfsnode. 3421 */ 3422 static int 3423 nfs_print(struct vop_print_args *ap) 3424 { 3425 struct vnode *vp = ap->a_vp; 3426 struct nfsnode *np = VTONFS(vp); 3427 3428 printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid, 3429 (uintmax_t)np->n_vattr.na_fsid); 3430 if (vp->v_type == VFIFO) 3431 fifo_printinfo(vp); 3432 printf("\n"); 3433 return (0); 3434 } 3435 3436 /* 3437 * nfs special file access vnode op. 3438 * Essentially just get vattr and then imitate iaccess() since the device is 3439 * local to the client. 3440 */ 3441 static int 3442 nfsspec_access(struct vop_access_args *ap) 3443 { 3444 struct vattr *vap; 3445 struct ucred *cred = ap->a_cred; 3446 struct vnode *vp = ap->a_vp; 3447 accmode_t accmode = ap->a_accmode; 3448 struct vattr vattr; 3449 int error; 3450 3451 /* 3452 * Disallow write attempts on filesystems mounted read-only; 3453 * unless the file is a socket, fifo, or a block or character 3454 * device resident on the filesystem. 3455 */ 3456 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3457 switch (vp->v_type) { 3458 case VREG: 3459 case VDIR: 3460 case VLNK: 3461 return (EROFS); 3462 default: 3463 break; 3464 } 3465 } 3466 vap = &vattr; 3467 error = VOP_GETATTR(vp, vap, cred); 3468 if (error) 3469 goto out; 3470 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid, 3471 accmode, cred); 3472 out: 3473 return error; 3474 } 3475 3476 /* 3477 * Read wrapper for fifos. 3478 */ 3479 static int 3480 nfsfifo_read(struct vop_read_args *ap) 3481 { 3482 struct nfsnode *np = VTONFS(ap->a_vp); 3483 int error; 3484 3485 /* 3486 * Set access flag. 3487 */ 3488 NFSLOCKNODE(np); 3489 np->n_flag |= NACC; 3490 vfs_timestamp(&np->n_atim); 3491 NFSUNLOCKNODE(np); 3492 error = fifo_specops.vop_read(ap); 3493 return error; 3494 } 3495 3496 /* 3497 * Write wrapper for fifos. 3498 */ 3499 static int 3500 nfsfifo_write(struct vop_write_args *ap) 3501 { 3502 struct nfsnode *np = VTONFS(ap->a_vp); 3503 3504 /* 3505 * Set update flag. 3506 */ 3507 NFSLOCKNODE(np); 3508 np->n_flag |= NUPD; 3509 vfs_timestamp(&np->n_mtim); 3510 NFSUNLOCKNODE(np); 3511 return(fifo_specops.vop_write(ap)); 3512 } 3513 3514 /* 3515 * Close wrapper for fifos. 3516 * 3517 * Update the times on the nfsnode then do fifo close. 3518 */ 3519 static int 3520 nfsfifo_close(struct vop_close_args *ap) 3521 { 3522 struct vnode *vp = ap->a_vp; 3523 struct nfsnode *np = VTONFS(vp); 3524 struct vattr vattr; 3525 struct timespec ts; 3526 3527 NFSLOCKNODE(np); 3528 if (np->n_flag & (NACC | NUPD)) { 3529 vfs_timestamp(&ts); 3530 if (np->n_flag & NACC) 3531 np->n_atim = ts; 3532 if (np->n_flag & NUPD) 3533 np->n_mtim = ts; 3534 np->n_flag |= NCHG; 3535 if (vrefcnt(vp) == 1 && 3536 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3537 VATTR_NULL(&vattr); 3538 if (np->n_flag & NACC) 3539 vattr.va_atime = np->n_atim; 3540 if (np->n_flag & NUPD) 3541 vattr.va_mtime = np->n_mtim; 3542 NFSUNLOCKNODE(np); 3543 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3544 goto out; 3545 } 3546 } 3547 NFSUNLOCKNODE(np); 3548 out: 3549 return (fifo_specops.vop_close(ap)); 3550 } 3551 3552 static int 3553 nfs_getacl(struct vop_getacl_args *ap) 3554 { 3555 int error; 3556 3557 if (ap->a_type != ACL_TYPE_NFS4) 3558 return (EOPNOTSUPP); 3559 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp); 3560 if (error > NFSERR_STALE) { 3561 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3562 error = EPERM; 3563 } 3564 return (error); 3565 } 3566 3567 static int 3568 nfs_setacl(struct vop_setacl_args *ap) 3569 { 3570 int error; 3571 3572 if (ap->a_type != ACL_TYPE_NFS4) 3573 return (EOPNOTSUPP); 3574 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp); 3575 if (error > NFSERR_STALE) { 3576 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3577 error = EPERM; 3578 } 3579 return (error); 3580 } 3581 3582 /* 3583 * VOP_ADVISE for NFS. 3584 * Just return 0 for any errors, since it is just a hint. 3585 */ 3586 static int 3587 nfs_advise(struct vop_advise_args *ap) 3588 { 3589 struct thread *td = curthread; 3590 struct nfsmount *nmp; 3591 uint64_t len; 3592 int error; 3593 3594 /* 3595 * First do vop_stdadvise() to handle the buffer cache. 3596 */ 3597 error = vop_stdadvise(ap); 3598 if (error != 0) 3599 return (error); 3600 if (ap->a_start < 0 || ap->a_end < 0) 3601 return (0); 3602 if (ap->a_end == OFF_MAX) 3603 len = 0; 3604 else if (ap->a_end < ap->a_start) 3605 return (0); 3606 else 3607 len = ap->a_end - ap->a_start + 1; 3608 nmp = VFSTONFS(ap->a_vp->v_mount); 3609 mtx_lock(&nmp->nm_mtx); 3610 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 3611 (NFSHASPNFS(nmp) && (nmp->nm_privflag & NFSMNTP_IOADVISETHRUMDS) == 3612 0) || (nmp->nm_privflag & NFSMNTP_NOADVISE) != 0) { 3613 mtx_unlock(&nmp->nm_mtx); 3614 return (0); 3615 } 3616 mtx_unlock(&nmp->nm_mtx); 3617 error = nfsrpc_advise(ap->a_vp, ap->a_start, len, ap->a_advice, 3618 td->td_ucred, td); 3619 if (error == NFSERR_NOTSUPP) { 3620 mtx_lock(&nmp->nm_mtx); 3621 nmp->nm_privflag |= NFSMNTP_NOADVISE; 3622 mtx_unlock(&nmp->nm_mtx); 3623 } 3624 return (0); 3625 } 3626 3627 /* 3628 * nfs allocate call 3629 */ 3630 static int 3631 nfs_allocate(struct vop_allocate_args *ap) 3632 { 3633 struct vnode *vp = ap->a_vp; 3634 struct thread *td = curthread; 3635 struct nfsvattr nfsva; 3636 struct nfsmount *nmp; 3637 struct nfsnode *np; 3638 off_t alen; 3639 int attrflag, error, ret; 3640 struct timespec ts; 3641 struct uio io; 3642 3643 attrflag = 0; 3644 nmp = VFSTONFS(vp->v_mount); 3645 np = VTONFS(vp); 3646 mtx_lock(&nmp->nm_mtx); 3647 if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION && 3648 (nmp->nm_privflag & NFSMNTP_NOALLOCATE) == 0) { 3649 mtx_unlock(&nmp->nm_mtx); 3650 alen = *ap->a_len; 3651 if ((uint64_t)alen > nfs_maxalloclen) 3652 alen = nfs_maxalloclen; 3653 3654 /* Check the file size limit. */ 3655 io.uio_offset = *ap->a_offset; 3656 io.uio_resid = alen; 3657 error = vn_rlimit_fsize(vp, &io, td); 3658 3659 /* 3660 * Flush first to ensure that the allocate adds to the 3661 * file's allocation on the server. 3662 */ 3663 if (error == 0) { 3664 vnode_pager_clean_sync(vp); 3665 error = ncl_flush(vp, MNT_WAIT, td, 1, 0); 3666 } 3667 if (error == 0) 3668 error = nfsrpc_allocate(vp, *ap->a_offset, alen, 3669 &nfsva, &attrflag, ap->a_cred, td); 3670 if (error == 0) { 3671 *ap->a_offset += alen; 3672 *ap->a_len -= alen; 3673 nanouptime(&ts); 3674 NFSLOCKNODE(np); 3675 np->n_localmodtime = ts; 3676 NFSUNLOCKNODE(np); 3677 } else if (error == NFSERR_NOTSUPP) { 3678 mtx_lock(&nmp->nm_mtx); 3679 nmp->nm_privflag |= NFSMNTP_NOALLOCATE; 3680 mtx_unlock(&nmp->nm_mtx); 3681 error = EINVAL; 3682 } 3683 } else { 3684 mtx_unlock(&nmp->nm_mtx); 3685 error = EINVAL; 3686 } 3687 if (attrflag != 0) { 3688 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 3689 if (error == 0 && ret != 0) 3690 error = ret; 3691 } 3692 if (error != 0) 3693 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 3694 return (error); 3695 } 3696 3697 /* 3698 * nfs deallocate call 3699 */ 3700 static int 3701 nfs_deallocate(struct vop_deallocate_args *ap) 3702 { 3703 struct vnode *vp = ap->a_vp; 3704 struct thread *td = curthread; 3705 struct nfsvattr nfsva; 3706 struct nfsmount *nmp; 3707 struct nfsnode *np; 3708 off_t tlen, mlen; 3709 int attrflag, error, ret; 3710 bool clipped; 3711 struct timespec ts; 3712 3713 error = 0; 3714 attrflag = 0; 3715 nmp = VFSTONFS(vp->v_mount); 3716 np = VTONFS(vp); 3717 mtx_lock(&nmp->nm_mtx); 3718 if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION && 3719 (nmp->nm_privflag & NFSMNTP_NODEALLOCATE) == 0) { 3720 mtx_unlock(&nmp->nm_mtx); 3721 tlen = omin(OFF_MAX - *ap->a_offset, *ap->a_len); 3722 NFSCL_DEBUG(4, "dealloc: off=%jd len=%jd maxfilesize=%ju\n", 3723 (intmax_t)*ap->a_offset, (intmax_t)tlen, 3724 (uintmax_t)nmp->nm_maxfilesize); 3725 if ((uint64_t)*ap->a_offset >= nmp->nm_maxfilesize) { 3726 /* Avoid EFBIG error return from the NFSv4.2 server. */ 3727 *ap->a_len = 0; 3728 return (0); 3729 } 3730 clipped = false; 3731 if ((uint64_t)*ap->a_offset + tlen > nmp->nm_maxfilesize) 3732 tlen = nmp->nm_maxfilesize - *ap->a_offset; 3733 if ((uint64_t)*ap->a_offset < np->n_size) { 3734 /* Limit the len to nfs_maxalloclen before EOF. */ 3735 mlen = omin((off_t)np->n_size - *ap->a_offset, tlen); 3736 if ((uint64_t)mlen > nfs_maxalloclen) { 3737 NFSCL_DEBUG(4, "dealloc: tlen maxalloclen\n"); 3738 tlen = nfs_maxalloclen; 3739 clipped = true; 3740 } 3741 } 3742 if (error == 0) 3743 error = ncl_vinvalbuf(vp, V_SAVE, td, 1); 3744 if (error == 0) { 3745 vnode_pager_purge_range(vp, *ap->a_offset, 3746 *ap->a_offset + tlen); 3747 error = nfsrpc_deallocate(vp, *ap->a_offset, tlen, 3748 &nfsva, &attrflag, ap->a_cred, td); 3749 NFSCL_DEBUG(4, "dealloc: rpc=%d\n", error); 3750 } 3751 if (error == 0) { 3752 NFSCL_DEBUG(4, "dealloc: attrflag=%d na_size=%ju\n", 3753 attrflag, (uintmax_t)nfsva.na_size); 3754 nanouptime(&ts); 3755 NFSLOCKNODE(np); 3756 np->n_localmodtime = ts; 3757 NFSUNLOCKNODE(np); 3758 if (attrflag != 0) { 3759 if ((uint64_t)*ap->a_offset < nfsva.na_size) 3760 *ap->a_offset += omin((off_t) 3761 nfsva.na_size - *ap->a_offset, 3762 tlen); 3763 } 3764 if (clipped && tlen < *ap->a_len) 3765 *ap->a_len -= tlen; 3766 else 3767 *ap->a_len = 0; 3768 } else if (error == NFSERR_NOTSUPP) { 3769 mtx_lock(&nmp->nm_mtx); 3770 nmp->nm_privflag |= NFSMNTP_NODEALLOCATE; 3771 mtx_unlock(&nmp->nm_mtx); 3772 } 3773 } else { 3774 mtx_unlock(&nmp->nm_mtx); 3775 error = EIO; 3776 } 3777 /* 3778 * If the NFS server cannot perform the Deallocate operation, just call 3779 * vop_stddeallocate() to perform it. 3780 */ 3781 if (error != 0 && error != NFSERR_FBIG && error != NFSERR_INVAL) { 3782 error = vop_stddeallocate(ap); 3783 NFSCL_DEBUG(4, "dealloc: stddeallocate=%d\n", error); 3784 } 3785 if (attrflag != 0) { 3786 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 3787 if (error == 0 && ret != 0) 3788 error = ret; 3789 } 3790 if (error != 0) 3791 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 3792 return (error); 3793 } 3794 3795 /* 3796 * nfs copy_file_range call 3797 */ 3798 static int 3799 nfs_copy_file_range(struct vop_copy_file_range_args *ap) 3800 { 3801 struct vnode *invp = ap->a_invp; 3802 struct vnode *outvp = ap->a_outvp; 3803 struct mount *mp; 3804 vm_object_t invp_obj; 3805 struct nfsvattr innfsva, outnfsva; 3806 struct vattr va, *vap; 3807 struct uio io; 3808 struct nfsmount *nmp; 3809 size_t len, len2; 3810 ssize_t r; 3811 int error, inattrflag, outattrflag, ret, ret2, invp_lock; 3812 off_t inoff, outoff; 3813 bool consecutive, must_commit, tryoutcred; 3814 3815 /* 3816 * NFSv4.2 Copy is not permitted for infile == outfile. 3817 * TODO: copy_file_range() between multiple NFS mountpoints 3818 */ 3819 if (invp == outvp || invp->v_mount != outvp->v_mount) { 3820 generic_copy: 3821 return (ENOSYS); 3822 } 3823 3824 invp_lock = LK_SHARED; 3825 relock: 3826 3827 /* Lock both vnodes, avoiding risk of deadlock. */ 3828 do { 3829 mp = NULL; 3830 error = vn_start_write(outvp, &mp, V_WAIT); 3831 if (error == 0) { 3832 error = vn_lock(outvp, LK_EXCLUSIVE); 3833 if (error == 0) { 3834 error = vn_lock(invp, invp_lock | LK_NOWAIT); 3835 if (error == 0) 3836 break; 3837 VOP_UNLOCK(outvp); 3838 if (mp != NULL) 3839 vn_finished_write(mp); 3840 mp = NULL; 3841 error = vn_lock(invp, invp_lock); 3842 if (error == 0) 3843 VOP_UNLOCK(invp); 3844 } 3845 } 3846 if (mp != NULL) 3847 vn_finished_write(mp); 3848 } while (error == 0); 3849 if (error != 0) 3850 return (error); 3851 3852 /* 3853 * More reasons to avoid nfs copy: not NFSv4.2, or explicitly 3854 * disabled. 3855 */ 3856 nmp = VFSTONFS(invp->v_mount); 3857 mtx_lock(&nmp->nm_mtx); 3858 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 3859 (nmp->nm_privflag & NFSMNTP_NOCOPY) != 0) { 3860 mtx_unlock(&nmp->nm_mtx); 3861 VOP_UNLOCK(invp); 3862 VOP_UNLOCK(outvp); 3863 if (mp != NULL) 3864 vn_finished_write(mp); 3865 goto generic_copy; 3866 } 3867 mtx_unlock(&nmp->nm_mtx); 3868 3869 /* 3870 * Do the vn_rlimit_fsize() check. Should this be above the VOP layer? 3871 */ 3872 io.uio_offset = *ap->a_outoffp; 3873 io.uio_resid = *ap->a_lenp; 3874 error = vn_rlimit_fsizex(outvp, &io, 0, &r, ap->a_fsizetd); 3875 *ap->a_lenp = io.uio_resid; 3876 /* 3877 * No need to call vn_rlimit_fsizex_res before return, since the uio is 3878 * local. 3879 */ 3880 3881 /* 3882 * Flush the input file so that the data is up to date before 3883 * the copy. Flush writes for the output file so that they 3884 * do not overwrite the data copied to the output file by the Copy. 3885 * Set the commit argument for both flushes so that the data is on 3886 * stable storage before the Copy RPC. This is done in case the 3887 * server reboots during the Copy and needs to be redone. 3888 */ 3889 if (error == 0) { 3890 invp_obj = invp->v_object; 3891 if (invp_obj != NULL && vm_object_mightbedirty(invp_obj)) { 3892 if (invp_lock != LK_EXCLUSIVE) { 3893 invp_lock = LK_EXCLUSIVE; 3894 VOP_UNLOCK(invp); 3895 VOP_UNLOCK(outvp); 3896 if (mp != NULL) 3897 vn_finished_write(mp); 3898 goto relock; 3899 } 3900 vnode_pager_clean_sync(invp); 3901 } 3902 error = ncl_flush(invp, MNT_WAIT, curthread, 1, 0); 3903 } 3904 if (error == 0) 3905 error = ncl_vinvalbuf(outvp, V_SAVE, curthread, 0); 3906 3907 /* Do the actual NFSv4.2 RPC. */ 3908 ret = ret2 = 0; 3909 len = *ap->a_lenp; 3910 mtx_lock(&nmp->nm_mtx); 3911 if ((nmp->nm_privflag & NFSMNTP_NOCONSECUTIVE) == 0) 3912 consecutive = true; 3913 else 3914 consecutive = false; 3915 mtx_unlock(&nmp->nm_mtx); 3916 inoff = *ap->a_inoffp; 3917 outoff = *ap->a_outoffp; 3918 tryoutcred = true; 3919 must_commit = false; 3920 if (error == 0) { 3921 vap = &VTONFS(invp)->n_vattr.na_vattr; 3922 error = VOP_GETATTR(invp, vap, ap->a_incred); 3923 if (error == 0) { 3924 /* 3925 * Clip "len" at va_size so that RFC compliant servers 3926 * will not reply NFSERR_INVAL. 3927 * Setting "len == 0" for the RPC would be preferred, 3928 * but some Linux servers do not support that. 3929 * If the len is being set to 0, do a Setattr RPC to 3930 * set the server's atime. This behaviour was the 3931 * preferred one for the FreeBSD "collective". 3932 */ 3933 if (inoff >= vap->va_size) { 3934 *ap->a_lenp = len = 0; 3935 if ((nmp->nm_mountp->mnt_flag & MNT_NOATIME) == 3936 0) { 3937 VATTR_NULL(&va); 3938 va.va_atime.tv_sec = 0; 3939 va.va_atime.tv_nsec = 0; 3940 va.va_vaflags = VA_UTIMES_NULL; 3941 inattrflag = 0; 3942 error = nfsrpc_setattr(invp, &va, NULL, 3943 ap->a_incred, curthread, &innfsva, 3944 &inattrflag); 3945 if (inattrflag != 0) 3946 ret = nfscl_loadattrcache(&invp, 3947 &innfsva, NULL, 0, 1); 3948 if (error == 0 && ret != 0) 3949 error = ret; 3950 } 3951 } else if (inoff + len > vap->va_size) 3952 *ap->a_lenp = len = vap->va_size - inoff; 3953 } else 3954 error = 0; 3955 } 3956 3957 /* 3958 * len will be set to 0 upon a successful Copy RPC. 3959 * As such, this only loops when the Copy RPC needs to be retried. 3960 */ 3961 while (len > 0 && error == 0) { 3962 inattrflag = outattrflag = 0; 3963 len2 = len; 3964 if (tryoutcred) 3965 error = nfsrpc_copy_file_range(invp, ap->a_inoffp, 3966 outvp, ap->a_outoffp, &len2, ap->a_flags, 3967 &inattrflag, &innfsva, &outattrflag, &outnfsva, 3968 ap->a_outcred, consecutive, &must_commit); 3969 else 3970 error = nfsrpc_copy_file_range(invp, ap->a_inoffp, 3971 outvp, ap->a_outoffp, &len2, ap->a_flags, 3972 &inattrflag, &innfsva, &outattrflag, &outnfsva, 3973 ap->a_incred, consecutive, &must_commit); 3974 if (inattrflag != 0) 3975 ret = nfscl_loadattrcache(&invp, &innfsva, NULL, 0, 1); 3976 if (outattrflag != 0) 3977 ret2 = nfscl_loadattrcache(&outvp, &outnfsva, NULL, 3978 1, 1); 3979 if (error == 0) { 3980 if (consecutive == false) { 3981 if (len2 == len) { 3982 mtx_lock(&nmp->nm_mtx); 3983 nmp->nm_privflag |= 3984 NFSMNTP_NOCONSECUTIVE; 3985 mtx_unlock(&nmp->nm_mtx); 3986 } else 3987 error = NFSERR_OFFLOADNOREQS; 3988 } 3989 *ap->a_lenp = len2; 3990 len = 0; 3991 if (len2 > 0 && must_commit && error == 0) 3992 error = ncl_commit(outvp, outoff, *ap->a_lenp, 3993 ap->a_outcred, curthread); 3994 if (error == 0 && ret != 0) 3995 error = ret; 3996 if (error == 0 && ret2 != 0) 3997 error = ret2; 3998 } else if (error == NFSERR_OFFLOADNOREQS && consecutive) { 3999 /* 4000 * Try consecutive == false, which is ok only if all 4001 * bytes are copied. 4002 * If only some bytes were copied when consecutive 4003 * is false, there is no way to know which bytes 4004 * still need to be written. 4005 */ 4006 consecutive = false; 4007 error = 0; 4008 } else if (error == NFSERR_ACCES && tryoutcred) { 4009 /* Try again with incred. */ 4010 tryoutcred = false; 4011 error = 0; 4012 } 4013 if (error == NFSERR_STALEWRITEVERF) { 4014 /* 4015 * Server rebooted, so do it all again. 4016 */ 4017 *ap->a_inoffp = inoff; 4018 *ap->a_outoffp = outoff; 4019 len = *ap->a_lenp; 4020 must_commit = false; 4021 error = 0; 4022 } 4023 } 4024 VOP_UNLOCK(invp); 4025 VOP_UNLOCK(outvp); 4026 if (mp != NULL) 4027 vn_finished_write(mp); 4028 if (error == NFSERR_NOTSUPP || error == NFSERR_OFFLOADNOREQS || 4029 error == NFSERR_ACCES) { 4030 /* 4031 * Unlike the NFSv4.2 Copy, vn_generic_copy_file_range() can 4032 * use a_incred for the read and a_outcred for the write, so 4033 * try this for NFSERR_ACCES failures for the Copy. 4034 * For NFSERR_NOTSUPP and NFSERR_OFFLOADNOREQS, the Copy can 4035 * never succeed, so disable it. 4036 */ 4037 if (error != NFSERR_ACCES) { 4038 /* Can never do Copy on this mount. */ 4039 mtx_lock(&nmp->nm_mtx); 4040 nmp->nm_privflag |= NFSMNTP_NOCOPY; 4041 mtx_unlock(&nmp->nm_mtx); 4042 } 4043 *ap->a_inoffp = inoff; 4044 *ap->a_outoffp = outoff; 4045 error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp, 4046 ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags, 4047 ap->a_incred, ap->a_outcred, ap->a_fsizetd); 4048 } else if (error != 0) 4049 *ap->a_lenp = 0; 4050 4051 if (error != 0) 4052 error = nfscl_maperr(curthread, error, (uid_t)0, (gid_t)0); 4053 return (error); 4054 } 4055 4056 /* 4057 * nfs ioctl call 4058 */ 4059 static int 4060 nfs_ioctl(struct vop_ioctl_args *ap) 4061 { 4062 struct vnode *vp = ap->a_vp; 4063 struct nfsvattr nfsva; 4064 struct nfsmount *nmp; 4065 int attrflag, content, error, ret; 4066 bool eof = false; /* shut up compiler. */ 4067 4068 /* Do the actual NFSv4.2 RPC. */ 4069 switch (ap->a_command) { 4070 case FIOSEEKDATA: 4071 content = NFSV4CONTENT_DATA; 4072 break; 4073 case FIOSEEKHOLE: 4074 content = NFSV4CONTENT_HOLE; 4075 break; 4076 default: 4077 return (ENOTTY); 4078 } 4079 4080 error = vn_lock(vp, LK_EXCLUSIVE); 4081 if (error != 0) 4082 return (EBADF); 4083 4084 if (vp->v_type != VREG) { 4085 VOP_UNLOCK(vp); 4086 return (ENOTTY); 4087 } 4088 nmp = VFSTONFS(vp->v_mount); 4089 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION) { 4090 VOP_UNLOCK(vp); 4091 error = vop_stdioctl(ap); 4092 return (error); 4093 } 4094 4095 attrflag = 0; 4096 if (*((off_t *)ap->a_data) >= VTONFS(vp)->n_size) 4097 error = ENXIO; 4098 else { 4099 /* 4100 * Flush all writes, so that the server is up to date. 4101 * Although a Commit is not required, the commit argument 4102 * is set so that, for a pNFS File/Flexible File Layout 4103 * server, the LayoutCommit will be done to ensure the file 4104 * size is up to date on the Metadata Server. 4105 */ 4106 4107 vnode_pager_clean_sync(vp); 4108 error = ncl_flush(vp, MNT_WAIT, ap->a_td, 1, 0); 4109 if (error == 0) 4110 error = nfsrpc_seek(vp, (off_t *)ap->a_data, &eof, 4111 content, ap->a_cred, &nfsva, &attrflag); 4112 /* If at eof for FIOSEEKDATA, return ENXIO. */ 4113 if (eof && error == 0 && content == NFSV4CONTENT_DATA) 4114 error = ENXIO; 4115 } 4116 if (attrflag != 0) { 4117 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4118 if (error == 0 && ret != 0) 4119 error = ret; 4120 } 4121 NFSVOPUNLOCK(vp); 4122 4123 if (error != 0) 4124 error = ENXIO; 4125 return (error); 4126 } 4127 4128 /* 4129 * nfs getextattr call 4130 */ 4131 static int 4132 nfs_getextattr(struct vop_getextattr_args *ap) 4133 { 4134 struct vnode *vp = ap->a_vp; 4135 struct nfsmount *nmp; 4136 struct ucred *cred; 4137 struct thread *td = ap->a_td; 4138 struct nfsvattr nfsva; 4139 ssize_t len; 4140 int attrflag, error, ret; 4141 4142 nmp = VFSTONFS(vp->v_mount); 4143 mtx_lock(&nmp->nm_mtx); 4144 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 4145 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 || 4146 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) { 4147 mtx_unlock(&nmp->nm_mtx); 4148 return (EOPNOTSUPP); 4149 } 4150 mtx_unlock(&nmp->nm_mtx); 4151 4152 cred = ap->a_cred; 4153 if (cred == NULL) 4154 cred = td->td_ucred; 4155 /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */ 4156 attrflag = 0; 4157 error = nfsrpc_getextattr(vp, ap->a_name, ap->a_uio, &len, &nfsva, 4158 &attrflag, cred, td); 4159 if (attrflag != 0) { 4160 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4161 if (error == 0 && ret != 0) 4162 error = ret; 4163 } 4164 if (error == 0 && ap->a_size != NULL) 4165 *ap->a_size = len; 4166 4167 switch (error) { 4168 case NFSERR_NOTSUPP: 4169 case NFSERR_OPILLEGAL: 4170 mtx_lock(&nmp->nm_mtx); 4171 nmp->nm_privflag |= NFSMNTP_NOXATTR; 4172 mtx_unlock(&nmp->nm_mtx); 4173 error = EOPNOTSUPP; 4174 break; 4175 case NFSERR_NOXATTR: 4176 case NFSERR_XATTR2BIG: 4177 error = ENOATTR; 4178 break; 4179 default: 4180 error = nfscl_maperr(td, error, 0, 0); 4181 break; 4182 } 4183 return (error); 4184 } 4185 4186 /* 4187 * nfs setextattr call 4188 */ 4189 static int 4190 nfs_setextattr(struct vop_setextattr_args *ap) 4191 { 4192 struct vnode *vp = ap->a_vp; 4193 struct nfsmount *nmp; 4194 struct ucred *cred; 4195 struct thread *td = ap->a_td; 4196 struct nfsvattr nfsva; 4197 int attrflag, error, ret; 4198 4199 nmp = VFSTONFS(vp->v_mount); 4200 mtx_lock(&nmp->nm_mtx); 4201 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 4202 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 || 4203 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) { 4204 mtx_unlock(&nmp->nm_mtx); 4205 return (EOPNOTSUPP); 4206 } 4207 mtx_unlock(&nmp->nm_mtx); 4208 4209 if (ap->a_uio->uio_resid < 0) 4210 return (EINVAL); 4211 cred = ap->a_cred; 4212 if (cred == NULL) 4213 cred = td->td_ucred; 4214 /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */ 4215 attrflag = 0; 4216 error = nfsrpc_setextattr(vp, ap->a_name, ap->a_uio, &nfsva, 4217 &attrflag, cred, td); 4218 if (attrflag != 0) { 4219 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4220 if (error == 0 && ret != 0) 4221 error = ret; 4222 } 4223 4224 switch (error) { 4225 case NFSERR_NOTSUPP: 4226 case NFSERR_OPILLEGAL: 4227 mtx_lock(&nmp->nm_mtx); 4228 nmp->nm_privflag |= NFSMNTP_NOXATTR; 4229 mtx_unlock(&nmp->nm_mtx); 4230 error = EOPNOTSUPP; 4231 break; 4232 case NFSERR_NOXATTR: 4233 case NFSERR_XATTR2BIG: 4234 error = ENOATTR; 4235 break; 4236 default: 4237 error = nfscl_maperr(td, error, 0, 0); 4238 break; 4239 } 4240 return (error); 4241 } 4242 4243 /* 4244 * nfs listextattr call 4245 */ 4246 static int 4247 nfs_listextattr(struct vop_listextattr_args *ap) 4248 { 4249 struct vnode *vp = ap->a_vp; 4250 struct nfsmount *nmp; 4251 struct ucred *cred; 4252 struct thread *td = ap->a_td; 4253 struct nfsvattr nfsva; 4254 size_t len, len2; 4255 uint64_t cookie; 4256 int attrflag, error, ret; 4257 bool eof; 4258 4259 nmp = VFSTONFS(vp->v_mount); 4260 mtx_lock(&nmp->nm_mtx); 4261 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 4262 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 || 4263 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) { 4264 mtx_unlock(&nmp->nm_mtx); 4265 return (EOPNOTSUPP); 4266 } 4267 mtx_unlock(&nmp->nm_mtx); 4268 4269 cred = ap->a_cred; 4270 if (cred == NULL) 4271 cred = td->td_ucred; 4272 4273 /* Loop around doing List Extended Attribute RPCs. */ 4274 eof = false; 4275 cookie = 0; 4276 len2 = 0; 4277 error = 0; 4278 while (!eof && error == 0) { 4279 len = nmp->nm_rsize; 4280 attrflag = 0; 4281 error = nfsrpc_listextattr(vp, &cookie, ap->a_uio, &len, &eof, 4282 &nfsva, &attrflag, cred, td); 4283 if (attrflag != 0) { 4284 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4285 if (error == 0 && ret != 0) 4286 error = ret; 4287 } 4288 if (error == 0) { 4289 len2 += len; 4290 if (len2 > SSIZE_MAX) 4291 error = ENOATTR; 4292 } 4293 } 4294 if (error == 0 && ap->a_size != NULL) 4295 *ap->a_size = len2; 4296 4297 switch (error) { 4298 case NFSERR_NOTSUPP: 4299 case NFSERR_OPILLEGAL: 4300 mtx_lock(&nmp->nm_mtx); 4301 nmp->nm_privflag |= NFSMNTP_NOXATTR; 4302 mtx_unlock(&nmp->nm_mtx); 4303 error = EOPNOTSUPP; 4304 break; 4305 case NFSERR_NOXATTR: 4306 case NFSERR_XATTR2BIG: 4307 error = ENOATTR; 4308 break; 4309 default: 4310 error = nfscl_maperr(td, error, 0, 0); 4311 break; 4312 } 4313 return (error); 4314 } 4315 4316 /* 4317 * nfs setextattr call 4318 */ 4319 static int 4320 nfs_deleteextattr(struct vop_deleteextattr_args *ap) 4321 { 4322 struct vnode *vp = ap->a_vp; 4323 struct nfsmount *nmp; 4324 struct nfsvattr nfsva; 4325 int attrflag, error, ret; 4326 4327 nmp = VFSTONFS(vp->v_mount); 4328 mtx_lock(&nmp->nm_mtx); 4329 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 4330 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 || 4331 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) { 4332 mtx_unlock(&nmp->nm_mtx); 4333 return (EOPNOTSUPP); 4334 } 4335 mtx_unlock(&nmp->nm_mtx); 4336 4337 /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */ 4338 attrflag = 0; 4339 error = nfsrpc_rmextattr(vp, ap->a_name, &nfsva, &attrflag, ap->a_cred, 4340 ap->a_td); 4341 if (attrflag != 0) { 4342 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4343 if (error == 0 && ret != 0) 4344 error = ret; 4345 } 4346 4347 switch (error) { 4348 case NFSERR_NOTSUPP: 4349 case NFSERR_OPILLEGAL: 4350 mtx_lock(&nmp->nm_mtx); 4351 nmp->nm_privflag |= NFSMNTP_NOXATTR; 4352 mtx_unlock(&nmp->nm_mtx); 4353 error = EOPNOTSUPP; 4354 break; 4355 case NFSERR_NOXATTR: 4356 case NFSERR_XATTR2BIG: 4357 error = ENOATTR; 4358 break; 4359 default: 4360 error = nfscl_maperr(ap->a_td, error, 0, 0); 4361 break; 4362 } 4363 return (error); 4364 } 4365 4366 /* 4367 * Return POSIX pathconf information applicable to nfs filesystems. 4368 */ 4369 static int 4370 nfs_pathconf(struct vop_pathconf_args *ap) 4371 { 4372 struct nfsv3_pathconf pc; 4373 struct nfsvattr nfsva; 4374 struct vnode *vp = ap->a_vp; 4375 struct nfsmount *nmp; 4376 struct thread *td = curthread; 4377 off_t off; 4378 bool eof; 4379 int attrflag, error; 4380 4381 if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX || 4382 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED || 4383 ap->a_name == _PC_NO_TRUNC)) || 4384 (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) { 4385 /* 4386 * Since only the above 4 a_names are returned by the NFSv3 4387 * Pathconf RPC, there is no point in doing it for others. 4388 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can 4389 * be used for _PC_NFS4_ACL as well. 4390 */ 4391 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva, 4392 &attrflag); 4393 if (attrflag != 0) 4394 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4395 if (error != 0) 4396 return (error); 4397 } else { 4398 /* 4399 * For NFSv2 (or NFSv3 when not one of the above 4 a_names), 4400 * just fake them. 4401 */ 4402 pc.pc_linkmax = NFS_LINK_MAX; 4403 pc.pc_namemax = NFS_MAXNAMLEN; 4404 pc.pc_notrunc = 1; 4405 pc.pc_chownrestricted = 1; 4406 pc.pc_caseinsensitive = 0; 4407 pc.pc_casepreserving = 1; 4408 error = 0; 4409 } 4410 switch (ap->a_name) { 4411 case _PC_LINK_MAX: 4412 #ifdef _LP64 4413 *ap->a_retval = pc.pc_linkmax; 4414 #else 4415 *ap->a_retval = MIN(LONG_MAX, pc.pc_linkmax); 4416 #endif 4417 break; 4418 case _PC_NAME_MAX: 4419 *ap->a_retval = pc.pc_namemax; 4420 break; 4421 case _PC_PIPE_BUF: 4422 if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO) 4423 *ap->a_retval = PIPE_BUF; 4424 else 4425 error = EINVAL; 4426 break; 4427 case _PC_CHOWN_RESTRICTED: 4428 *ap->a_retval = pc.pc_chownrestricted; 4429 break; 4430 case _PC_NO_TRUNC: 4431 *ap->a_retval = pc.pc_notrunc; 4432 break; 4433 case _PC_ACL_NFS4: 4434 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 && 4435 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL)) 4436 *ap->a_retval = 1; 4437 else 4438 *ap->a_retval = 0; 4439 break; 4440 case _PC_ACL_PATH_MAX: 4441 if (NFS_ISV4(vp)) 4442 *ap->a_retval = ACL_MAX_ENTRIES; 4443 else 4444 *ap->a_retval = 3; 4445 break; 4446 case _PC_PRIO_IO: 4447 *ap->a_retval = 0; 4448 break; 4449 case _PC_SYNC_IO: 4450 *ap->a_retval = 0; 4451 break; 4452 case _PC_ALLOC_SIZE_MIN: 4453 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize; 4454 break; 4455 case _PC_FILESIZEBITS: 4456 if (NFS_ISV34(vp)) 4457 *ap->a_retval = 64; 4458 else 4459 *ap->a_retval = 32; 4460 break; 4461 case _PC_REC_INCR_XFER_SIZE: 4462 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 4463 break; 4464 case _PC_REC_MAX_XFER_SIZE: 4465 *ap->a_retval = -1; /* means ``unlimited'' */ 4466 break; 4467 case _PC_REC_MIN_XFER_SIZE: 4468 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 4469 break; 4470 case _PC_REC_XFER_ALIGN: 4471 *ap->a_retval = PAGE_SIZE; 4472 break; 4473 case _PC_SYMLINK_MAX: 4474 *ap->a_retval = NFS_MAXPATHLEN; 4475 break; 4476 case _PC_MIN_HOLE_SIZE: 4477 /* Only some NFSv4.2 servers support Seek for Holes. */ 4478 *ap->a_retval = 0; 4479 nmp = VFSTONFS(vp->v_mount); 4480 if (NFS_ISV4(vp) && nmp->nm_minorvers == NFSV42_MINORVERSION) { 4481 /* 4482 * NFSv4.2 doesn't have an attribute for hole size, 4483 * so all we can do is see if the Seek operation is 4484 * supported and then use f_iosize as a "best guess". 4485 */ 4486 mtx_lock(&nmp->nm_mtx); 4487 if ((nmp->nm_privflag & NFSMNTP_SEEKTESTED) == 0) { 4488 mtx_unlock(&nmp->nm_mtx); 4489 off = 0; 4490 attrflag = 0; 4491 error = nfsrpc_seek(vp, &off, &eof, 4492 NFSV4CONTENT_HOLE, td->td_ucred, &nfsva, 4493 &attrflag); 4494 if (attrflag != 0) 4495 (void) nfscl_loadattrcache(&vp, &nfsva, 4496 NULL, 0, 1); 4497 mtx_lock(&nmp->nm_mtx); 4498 if (error == NFSERR_NOTSUPP) 4499 nmp->nm_privflag |= NFSMNTP_SEEKTESTED; 4500 else 4501 nmp->nm_privflag |= NFSMNTP_SEEKTESTED | 4502 NFSMNTP_SEEK; 4503 error = 0; 4504 } 4505 if ((nmp->nm_privflag & NFSMNTP_SEEK) != 0) 4506 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 4507 mtx_unlock(&nmp->nm_mtx); 4508 } 4509 break; 4510 4511 default: 4512 error = vop_stdpathconf(ap); 4513 break; 4514 } 4515 return (error); 4516 } 4517