1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * from nfs_vnops.c 8.16 (Berkeley) 5/27/95 35 */ 36 37 #include <sys/cdefs.h> 38 __FBSDID("$FreeBSD$"); 39 40 /* 41 * vnode op calls for Sun NFS version 2, 3 and 4 42 */ 43 44 #include "opt_inet.h" 45 46 #include <sys/param.h> 47 #include <sys/kernel.h> 48 #include <sys/systm.h> 49 #include <sys/resourcevar.h> 50 #include <sys/proc.h> 51 #include <sys/mount.h> 52 #include <sys/bio.h> 53 #include <sys/buf.h> 54 #include <sys/extattr.h> 55 #include <sys/filio.h> 56 #include <sys/jail.h> 57 #include <sys/malloc.h> 58 #include <sys/mbuf.h> 59 #include <sys/namei.h> 60 #include <sys/socket.h> 61 #include <sys/vnode.h> 62 #include <sys/dirent.h> 63 #include <sys/fcntl.h> 64 #include <sys/lockf.h> 65 #include <sys/stat.h> 66 #include <sys/sysctl.h> 67 #include <sys/signalvar.h> 68 69 #include <vm/vm.h> 70 #include <vm/vm_extern.h> 71 #include <vm/vm_object.h> 72 73 #include <fs/nfs/nfsport.h> 74 #include <fs/nfsclient/nfsnode.h> 75 #include <fs/nfsclient/nfsmount.h> 76 #include <fs/nfsclient/nfs.h> 77 #include <fs/nfsclient/nfs_kdtrace.h> 78 79 #include <net/if.h> 80 #include <netinet/in.h> 81 #include <netinet/in_var.h> 82 83 #include <nfs/nfs_lock.h> 84 85 #ifdef KDTRACE_HOOKS 86 #include <sys/dtrace_bsd.h> 87 88 dtrace_nfsclient_accesscache_flush_probe_func_t 89 dtrace_nfscl_accesscache_flush_done_probe; 90 uint32_t nfscl_accesscache_flush_done_id; 91 92 dtrace_nfsclient_accesscache_get_probe_func_t 93 dtrace_nfscl_accesscache_get_hit_probe, 94 dtrace_nfscl_accesscache_get_miss_probe; 95 uint32_t nfscl_accesscache_get_hit_id; 96 uint32_t nfscl_accesscache_get_miss_id; 97 98 dtrace_nfsclient_accesscache_load_probe_func_t 99 dtrace_nfscl_accesscache_load_done_probe; 100 uint32_t nfscl_accesscache_load_done_id; 101 #endif /* !KDTRACE_HOOKS */ 102 103 /* Defs */ 104 #define TRUE 1 105 #define FALSE 0 106 107 extern struct nfsstatsv1 nfsstatsv1; 108 extern int nfsrv_useacl; 109 extern int nfscl_debuglevel; 110 MALLOC_DECLARE(M_NEWNFSREQ); 111 112 static vop_read_t nfsfifo_read; 113 static vop_write_t nfsfifo_write; 114 static vop_close_t nfsfifo_close; 115 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *, 116 struct thread *); 117 static vop_lookup_t nfs_lookup; 118 static vop_create_t nfs_create; 119 static vop_mknod_t nfs_mknod; 120 static vop_open_t nfs_open; 121 static vop_pathconf_t nfs_pathconf; 122 static vop_close_t nfs_close; 123 static vop_access_t nfs_access; 124 static vop_getattr_t nfs_getattr; 125 static vop_setattr_t nfs_setattr; 126 static vop_read_t nfs_read; 127 static vop_fsync_t nfs_fsync; 128 static vop_remove_t nfs_remove; 129 static vop_link_t nfs_link; 130 static vop_rename_t nfs_rename; 131 static vop_mkdir_t nfs_mkdir; 132 static vop_rmdir_t nfs_rmdir; 133 static vop_symlink_t nfs_symlink; 134 static vop_readdir_t nfs_readdir; 135 static vop_strategy_t nfs_strategy; 136 static int nfs_lookitup(struct vnode *, char *, int, 137 struct ucred *, struct thread *, struct nfsnode **); 138 static int nfs_sillyrename(struct vnode *, struct vnode *, 139 struct componentname *); 140 static vop_access_t nfsspec_access; 141 static vop_readlink_t nfs_readlink; 142 static vop_print_t nfs_print; 143 static vop_advlock_t nfs_advlock; 144 static vop_advlockasync_t nfs_advlockasync; 145 static vop_getacl_t nfs_getacl; 146 static vop_setacl_t nfs_setacl; 147 static vop_advise_t nfs_advise; 148 static vop_allocate_t nfs_allocate; 149 static vop_deallocate_t nfs_deallocate; 150 static vop_copy_file_range_t nfs_copy_file_range; 151 static vop_ioctl_t nfs_ioctl; 152 static vop_getextattr_t nfs_getextattr; 153 static vop_setextattr_t nfs_setextattr; 154 static vop_listextattr_t nfs_listextattr; 155 static vop_deleteextattr_t nfs_deleteextattr; 156 static vop_lock1_t nfs_lock; 157 158 /* 159 * Global vfs data structures for nfs 160 */ 161 162 static struct vop_vector newnfs_vnodeops_nosig = { 163 .vop_default = &default_vnodeops, 164 .vop_access = nfs_access, 165 .vop_advlock = nfs_advlock, 166 .vop_advlockasync = nfs_advlockasync, 167 .vop_close = nfs_close, 168 .vop_create = nfs_create, 169 .vop_fsync = nfs_fsync, 170 .vop_getattr = nfs_getattr, 171 .vop_getpages = ncl_getpages, 172 .vop_putpages = ncl_putpages, 173 .vop_inactive = ncl_inactive, 174 .vop_link = nfs_link, 175 .vop_lock1 = nfs_lock, 176 .vop_lookup = nfs_lookup, 177 .vop_mkdir = nfs_mkdir, 178 .vop_mknod = nfs_mknod, 179 .vop_open = nfs_open, 180 .vop_pathconf = nfs_pathconf, 181 .vop_print = nfs_print, 182 .vop_read = nfs_read, 183 .vop_readdir = nfs_readdir, 184 .vop_readlink = nfs_readlink, 185 .vop_reclaim = ncl_reclaim, 186 .vop_remove = nfs_remove, 187 .vop_rename = nfs_rename, 188 .vop_rmdir = nfs_rmdir, 189 .vop_setattr = nfs_setattr, 190 .vop_strategy = nfs_strategy, 191 .vop_symlink = nfs_symlink, 192 .vop_write = ncl_write, 193 .vop_getacl = nfs_getacl, 194 .vop_setacl = nfs_setacl, 195 .vop_advise = nfs_advise, 196 .vop_allocate = nfs_allocate, 197 .vop_deallocate = nfs_deallocate, 198 .vop_copy_file_range = nfs_copy_file_range, 199 .vop_ioctl = nfs_ioctl, 200 .vop_getextattr = nfs_getextattr, 201 .vop_setextattr = nfs_setextattr, 202 .vop_listextattr = nfs_listextattr, 203 .vop_deleteextattr = nfs_deleteextattr, 204 }; 205 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops_nosig); 206 207 static int 208 nfs_vnodeops_bypass(struct vop_generic_args *a) 209 { 210 211 return (vop_sigdefer(&newnfs_vnodeops_nosig, a)); 212 } 213 214 struct vop_vector newnfs_vnodeops = { 215 .vop_default = &default_vnodeops, 216 .vop_bypass = nfs_vnodeops_bypass, 217 }; 218 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops); 219 220 static struct vop_vector newnfs_fifoops_nosig = { 221 .vop_default = &fifo_specops, 222 .vop_access = nfsspec_access, 223 .vop_close = nfsfifo_close, 224 .vop_fsync = nfs_fsync, 225 .vop_getattr = nfs_getattr, 226 .vop_inactive = ncl_inactive, 227 .vop_pathconf = nfs_pathconf, 228 .vop_print = nfs_print, 229 .vop_read = nfsfifo_read, 230 .vop_reclaim = ncl_reclaim, 231 .vop_setattr = nfs_setattr, 232 .vop_write = nfsfifo_write, 233 }; 234 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops_nosig); 235 236 static int 237 nfs_fifoops_bypass(struct vop_generic_args *a) 238 { 239 240 return (vop_sigdefer(&newnfs_fifoops_nosig, a)); 241 } 242 243 struct vop_vector newnfs_fifoops = { 244 .vop_default = &default_vnodeops, 245 .vop_bypass = nfs_fifoops_bypass, 246 }; 247 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops); 248 249 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, 250 struct componentname *cnp, struct vattr *vap); 251 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 252 int namelen, struct ucred *cred, struct thread *td); 253 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, 254 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp, 255 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td); 256 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp, 257 struct componentname *scnp, struct sillyrename *sp); 258 259 /* 260 * Global variables 261 */ 262 SYSCTL_DECL(_vfs_nfs); 263 264 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO; 265 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 266 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 267 268 static int nfs_prime_access_cache = 0; 269 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW, 270 &nfs_prime_access_cache, 0, 271 "Prime NFS ACCESS cache when fetching attributes"); 272 273 static int newnfs_commit_on_close = 0; 274 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW, 275 &newnfs_commit_on_close, 0, "write+commit on close, else only write"); 276 277 static int nfs_clean_pages_on_close = 1; 278 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW, 279 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close"); 280 281 int newnfs_directio_enable = 0; 282 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW, 283 &newnfs_directio_enable, 0, "Enable NFS directio"); 284 285 int nfs_keep_dirty_on_error; 286 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW, 287 &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned"); 288 289 /* 290 * This sysctl allows other processes to mmap a file that has been opened 291 * O_DIRECT by a process. In general, having processes mmap the file while 292 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow 293 * this by default to prevent DoS attacks - to prevent a malicious user from 294 * opening up files O_DIRECT preventing other users from mmap'ing these 295 * files. "Protected" environments where stricter consistency guarantees are 296 * required can disable this knob. The process that opened the file O_DIRECT 297 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not 298 * meaningful. 299 */ 300 int newnfs_directio_allow_mmap = 1; 301 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW, 302 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens"); 303 304 static uint64_t nfs_maxalloclen = 64 * 1024 * 1024; 305 SYSCTL_U64(_vfs_nfs, OID_AUTO, maxalloclen, CTLFLAG_RW, 306 &nfs_maxalloclen, 0, "NFS max allocate/deallocate length"); 307 308 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \ 309 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \ 310 | NFSACCESS_DELETE | NFSACCESS_LOOKUP) 311 312 /* 313 * SMP Locking Note : 314 * The list of locks after the description of the lock is the ordering 315 * of other locks acquired with the lock held. 316 * np->n_mtx : Protects the fields in the nfsnode. 317 VM Object Lock 318 VI_MTX (acquired indirectly) 319 * nmp->nm_mtx : Protects the fields in the nfsmount. 320 rep->r_mtx 321 * ncl_iod_mutex : Global lock, protects shared nfsiod state. 322 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list. 323 nmp->nm_mtx 324 rep->r_mtx 325 * rep->r_mtx : Protects the fields in an nfsreq. 326 */ 327 328 static int 329 nfs_lock(struct vop_lock1_args *ap) 330 { 331 struct vnode *vp; 332 struct nfsnode *np; 333 u_quad_t nsize; 334 int error, lktype; 335 bool onfault; 336 337 vp = ap->a_vp; 338 lktype = ap->a_flags & LK_TYPE_MASK; 339 error = VOP_LOCK1_APV(&default_vnodeops, ap); 340 if (error != 0 || vp->v_op != &newnfs_vnodeops) 341 return (error); 342 np = VTONFS(vp); 343 if (np == NULL) 344 return (0); 345 NFSLOCKNODE(np); 346 if ((np->n_flag & NVNSETSZSKIP) == 0 || (lktype != LK_SHARED && 347 lktype != LK_EXCLUSIVE && lktype != LK_UPGRADE && 348 lktype != LK_TRYUPGRADE)) { 349 NFSUNLOCKNODE(np); 350 return (0); 351 } 352 onfault = (ap->a_flags & LK_EATTR_MASK) == LK_NOWAIT && 353 (ap->a_flags & LK_INIT_MASK) == LK_CANRECURSE && 354 (lktype == LK_SHARED || lktype == LK_EXCLUSIVE); 355 if (onfault && vp->v_vnlock->lk_recurse == 0) { 356 /* 357 * Force retry in vm_fault(), to make the lock request 358 * sleepable, which allows us to piggy-back the 359 * sleepable call to vnode_pager_setsize(). 360 */ 361 NFSUNLOCKNODE(np); 362 VOP_UNLOCK(vp); 363 return (EBUSY); 364 } 365 if ((ap->a_flags & LK_NOWAIT) != 0 || 366 (lktype == LK_SHARED && vp->v_vnlock->lk_recurse > 0)) { 367 NFSUNLOCKNODE(np); 368 return (0); 369 } 370 if (lktype == LK_SHARED) { 371 NFSUNLOCKNODE(np); 372 VOP_UNLOCK(vp); 373 ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK); 374 ap->a_flags |= LK_EXCLUSIVE; 375 error = VOP_LOCK1_APV(&default_vnodeops, ap); 376 if (error != 0 || vp->v_op != &newnfs_vnodeops) 377 return (error); 378 if (vp->v_data == NULL) 379 goto downgrade; 380 MPASS(vp->v_data == np); 381 NFSLOCKNODE(np); 382 if ((np->n_flag & NVNSETSZSKIP) == 0) { 383 NFSUNLOCKNODE(np); 384 goto downgrade; 385 } 386 } 387 np->n_flag &= ~NVNSETSZSKIP; 388 nsize = np->n_size; 389 NFSUNLOCKNODE(np); 390 vnode_pager_setsize(vp, nsize); 391 downgrade: 392 if (lktype == LK_SHARED) { 393 ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK); 394 ap->a_flags |= LK_DOWNGRADE; 395 (void)VOP_LOCK1_APV(&default_vnodeops, ap); 396 } 397 return (0); 398 } 399 400 static int 401 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td, 402 struct ucred *cred, u_int32_t *retmode) 403 { 404 int error = 0, attrflag, i, lrupos; 405 u_int32_t rmode; 406 struct nfsnode *np = VTONFS(vp); 407 struct nfsvattr nfsva; 408 409 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag, 410 &rmode, NULL); 411 if (attrflag) 412 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 413 if (!error) { 414 lrupos = 0; 415 NFSLOCKNODE(np); 416 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 417 if (np->n_accesscache[i].uid == cred->cr_uid) { 418 np->n_accesscache[i].mode = rmode; 419 np->n_accesscache[i].stamp = time_second; 420 break; 421 } 422 if (i > 0 && np->n_accesscache[i].stamp < 423 np->n_accesscache[lrupos].stamp) 424 lrupos = i; 425 } 426 if (i == NFS_ACCESSCACHESIZE) { 427 np->n_accesscache[lrupos].uid = cred->cr_uid; 428 np->n_accesscache[lrupos].mode = rmode; 429 np->n_accesscache[lrupos].stamp = time_second; 430 } 431 NFSUNLOCKNODE(np); 432 if (retmode != NULL) 433 *retmode = rmode; 434 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0); 435 } else if (NFS_ISV4(vp)) { 436 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 437 } 438 #ifdef KDTRACE_HOOKS 439 if (error != 0) 440 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0, 441 error); 442 #endif 443 return (error); 444 } 445 446 /* 447 * nfs access vnode op. 448 * For nfs version 2, just return ok. File accesses may fail later. 449 * For nfs version 3, use the access rpc to check accessibility. If file modes 450 * are changed on the server, accesses might still fail later. 451 */ 452 static int 453 nfs_access(struct vop_access_args *ap) 454 { 455 struct vnode *vp = ap->a_vp; 456 int error = 0, i, gotahit; 457 u_int32_t mode, wmode, rmode; 458 int v34 = NFS_ISV34(vp); 459 struct nfsnode *np = VTONFS(vp); 460 461 /* 462 * Disallow write attempts on filesystems mounted read-only; 463 * unless the file is a socket, fifo, or a block or character 464 * device resident on the filesystem. 465 */ 466 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS | 467 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL | 468 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) { 469 switch (vp->v_type) { 470 case VREG: 471 case VDIR: 472 case VLNK: 473 return (EROFS); 474 default: 475 break; 476 } 477 } 478 /* 479 * For nfs v3 or v4, check to see if we have done this recently, and if 480 * so return our cached result instead of making an ACCESS call. 481 * If not, do an access rpc, otherwise you are stuck emulating 482 * ufs_access() locally using the vattr. This may not be correct, 483 * since the server may apply other access criteria such as 484 * client uid-->server uid mapping that we do not know about. 485 */ 486 if (v34) { 487 if (ap->a_accmode & VREAD) 488 mode = NFSACCESS_READ; 489 else 490 mode = 0; 491 if (vp->v_type != VDIR) { 492 if (ap->a_accmode & VWRITE) 493 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 494 if (ap->a_accmode & VAPPEND) 495 mode |= NFSACCESS_EXTEND; 496 if (ap->a_accmode & VEXEC) 497 mode |= NFSACCESS_EXECUTE; 498 if (ap->a_accmode & VDELETE) 499 mode |= NFSACCESS_DELETE; 500 } else { 501 if (ap->a_accmode & VWRITE) 502 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 503 if (ap->a_accmode & VAPPEND) 504 mode |= NFSACCESS_EXTEND; 505 if (ap->a_accmode & VEXEC) 506 mode |= NFSACCESS_LOOKUP; 507 if (ap->a_accmode & VDELETE) 508 mode |= NFSACCESS_DELETE; 509 if (ap->a_accmode & VDELETE_CHILD) 510 mode |= NFSACCESS_MODIFY; 511 } 512 /* XXX safety belt, only make blanket request if caching */ 513 if (nfsaccess_cache_timeout > 0) { 514 wmode = NFSACCESS_READ | NFSACCESS_MODIFY | 515 NFSACCESS_EXTEND | NFSACCESS_EXECUTE | 516 NFSACCESS_DELETE | NFSACCESS_LOOKUP; 517 } else { 518 wmode = mode; 519 } 520 521 /* 522 * Does our cached result allow us to give a definite yes to 523 * this request? 524 */ 525 gotahit = 0; 526 NFSLOCKNODE(np); 527 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 528 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) { 529 if (time_second < (np->n_accesscache[i].stamp 530 + nfsaccess_cache_timeout) && 531 (np->n_accesscache[i].mode & mode) == mode) { 532 NFSINCRGLOBAL(nfsstatsv1.accesscache_hits); 533 gotahit = 1; 534 } 535 break; 536 } 537 } 538 NFSUNLOCKNODE(np); 539 #ifdef KDTRACE_HOOKS 540 if (gotahit != 0) 541 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp, 542 ap->a_cred->cr_uid, mode); 543 else 544 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp, 545 ap->a_cred->cr_uid, mode); 546 #endif 547 if (gotahit == 0) { 548 /* 549 * Either a no, or a don't know. Go to the wire. 550 */ 551 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses); 552 error = nfs34_access_otw(vp, wmode, ap->a_td, 553 ap->a_cred, &rmode); 554 if (!error && 555 (rmode & mode) != mode) 556 error = EACCES; 557 } 558 return (error); 559 } else { 560 if ((error = nfsspec_access(ap)) != 0) { 561 return (error); 562 } 563 /* 564 * Attempt to prevent a mapped root from accessing a file 565 * which it shouldn't. We try to read a byte from the file 566 * if the user is root and the file is not zero length. 567 * After calling nfsspec_access, we should have the correct 568 * file size cached. 569 */ 570 NFSLOCKNODE(np); 571 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD) 572 && VTONFS(vp)->n_size > 0) { 573 struct iovec aiov; 574 struct uio auio; 575 char buf[1]; 576 577 NFSUNLOCKNODE(np); 578 aiov.iov_base = buf; 579 aiov.iov_len = 1; 580 auio.uio_iov = &aiov; 581 auio.uio_iovcnt = 1; 582 auio.uio_offset = 0; 583 auio.uio_resid = 1; 584 auio.uio_segflg = UIO_SYSSPACE; 585 auio.uio_rw = UIO_READ; 586 auio.uio_td = ap->a_td; 587 588 if (vp->v_type == VREG) 589 error = ncl_readrpc(vp, &auio, ap->a_cred); 590 else if (vp->v_type == VDIR) { 591 char* bp; 592 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 593 aiov.iov_base = bp; 594 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 595 error = ncl_readdirrpc(vp, &auio, ap->a_cred, 596 ap->a_td); 597 free(bp, M_TEMP); 598 } else if (vp->v_type == VLNK) 599 error = ncl_readlinkrpc(vp, &auio, ap->a_cred); 600 else 601 error = EACCES; 602 } else 603 NFSUNLOCKNODE(np); 604 return (error); 605 } 606 } 607 608 /* 609 * nfs open vnode op 610 * Check to see if the type is ok 611 * and that deletion is not in progress. 612 * For paged in text files, you will need to flush the page cache 613 * if consistency is lost. 614 */ 615 /* ARGSUSED */ 616 static int 617 nfs_open(struct vop_open_args *ap) 618 { 619 struct vnode *vp = ap->a_vp; 620 struct nfsnode *np = VTONFS(vp); 621 struct vattr vattr; 622 int error; 623 int fmode = ap->a_mode; 624 struct ucred *cred; 625 vm_object_t obj; 626 627 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) 628 return (EOPNOTSUPP); 629 630 /* 631 * For NFSv4, we need to do the Open Op before cache validation, 632 * so that we conform to RFC3530 Sec. 9.3.1. 633 */ 634 if (NFS_ISV4(vp)) { 635 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td); 636 if (error) { 637 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 638 (gid_t)0); 639 return (error); 640 } 641 } 642 643 /* 644 * Now, if this Open will be doing reading, re-validate/flush the 645 * cache, so that Close/Open coherency is maintained. 646 */ 647 NFSLOCKNODE(np); 648 if (np->n_flag & NMODIFIED) { 649 NFSUNLOCKNODE(np); 650 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 651 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 652 if (VN_IS_DOOMED(vp)) 653 return (EBADF); 654 } 655 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 656 if (error == EINTR || error == EIO) { 657 if (NFS_ISV4(vp)) 658 (void) nfsrpc_close(vp, 0, ap->a_td); 659 return (error); 660 } 661 NFSLOCKNODE(np); 662 np->n_attrstamp = 0; 663 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 664 if (vp->v_type == VDIR) 665 np->n_direofoffset = 0; 666 NFSUNLOCKNODE(np); 667 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 668 if (error) { 669 if (NFS_ISV4(vp)) 670 (void) nfsrpc_close(vp, 0, ap->a_td); 671 return (error); 672 } 673 NFSLOCKNODE(np); 674 np->n_mtime = vattr.va_mtime; 675 if (NFS_ISV4(vp)) 676 np->n_change = vattr.va_filerev; 677 } else { 678 NFSUNLOCKNODE(np); 679 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 680 if (error) { 681 if (NFS_ISV4(vp)) 682 (void) nfsrpc_close(vp, 0, ap->a_td); 683 return (error); 684 } 685 NFSLOCKNODE(np); 686 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) || 687 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 688 if (vp->v_type == VDIR) 689 np->n_direofoffset = 0; 690 NFSUNLOCKNODE(np); 691 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 692 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 693 if (VN_IS_DOOMED(vp)) 694 return (EBADF); 695 } 696 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 697 if (error == EINTR || error == EIO) { 698 if (NFS_ISV4(vp)) 699 (void) nfsrpc_close(vp, 0, ap->a_td); 700 return (error); 701 } 702 NFSLOCKNODE(np); 703 np->n_mtime = vattr.va_mtime; 704 if (NFS_ISV4(vp)) 705 np->n_change = vattr.va_filerev; 706 } 707 } 708 709 /* 710 * If the object has >= 1 O_DIRECT active opens, we disable caching. 711 */ 712 if (newnfs_directio_enable && (fmode & O_DIRECT) && 713 (vp->v_type == VREG)) { 714 if (np->n_directio_opens == 0) { 715 NFSUNLOCKNODE(np); 716 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 717 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 718 if (VN_IS_DOOMED(vp)) 719 return (EBADF); 720 } 721 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 722 if (error) { 723 if (NFS_ISV4(vp)) 724 (void) nfsrpc_close(vp, 0, ap->a_td); 725 return (error); 726 } 727 NFSLOCKNODE(np); 728 np->n_flag |= NNONCACHE; 729 } 730 np->n_directio_opens++; 731 } 732 733 /* If opened for writing via NFSv4.1 or later, mark that for pNFS. */ 734 if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0) 735 np->n_flag |= NWRITEOPENED; 736 737 /* 738 * If this is an open for writing, capture a reference to the 739 * credentials, so they can be used by ncl_putpages(). Using 740 * these write credentials is preferable to the credentials of 741 * whatever thread happens to be doing the VOP_PUTPAGES() since 742 * the write RPCs are less likely to fail with EACCES. 743 */ 744 if ((fmode & FWRITE) != 0) { 745 cred = np->n_writecred; 746 np->n_writecred = crhold(ap->a_cred); 747 } else 748 cred = NULL; 749 NFSUNLOCKNODE(np); 750 751 if (cred != NULL) 752 crfree(cred); 753 vnode_create_vobject(vp, vattr.va_size, ap->a_td); 754 755 /* 756 * If the text file has been mmap'd, flush any dirty pages to the 757 * buffer cache and then... 758 * Make sure all writes are pushed to the NFS server. If this is not 759 * done, the modify time of the file can change while the text 760 * file is being executed. This will cause the process that is 761 * executing the text file to be terminated. 762 */ 763 if (vp->v_writecount <= -1) { 764 if ((obj = vp->v_object) != NULL && 765 vm_object_mightbedirty(obj)) { 766 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 767 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 768 if (VN_IS_DOOMED(vp)) 769 return (EBADF); 770 } 771 VM_OBJECT_WLOCK(obj); 772 vm_object_page_clean(obj, 0, 0, OBJPC_SYNC); 773 VM_OBJECT_WUNLOCK(obj); 774 } 775 776 /* Now, flush the buffer cache. */ 777 ncl_flush(vp, MNT_WAIT, curthread, 0, 0); 778 779 /* And, finally, make sure that n_mtime is up to date. */ 780 np = VTONFS(vp); 781 NFSLOCKNODE(np); 782 np->n_mtime = np->n_vattr.na_mtime; 783 NFSUNLOCKNODE(np); 784 } 785 return (0); 786 } 787 788 /* 789 * nfs close vnode op 790 * What an NFS client should do upon close after writing is a debatable issue. 791 * Most NFS clients push delayed writes to the server upon close, basically for 792 * two reasons: 793 * 1 - So that any write errors may be reported back to the client process 794 * doing the close system call. By far the two most likely errors are 795 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 796 * 2 - To put a worst case upper bound on cache inconsistency between 797 * multiple clients for the file. 798 * There is also a consistency problem for Version 2 of the protocol w.r.t. 799 * not being able to tell if other clients are writing a file concurrently, 800 * since there is no way of knowing if the changed modify time in the reply 801 * is only due to the write for this client. 802 * (NFS Version 3 provides weak cache consistency data in the reply that 803 * should be sufficient to detect and handle this case.) 804 * 805 * The current code does the following: 806 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 807 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 808 * or commit them (this satisfies 1 and 2 except for the 809 * case where the server crashes after this close but 810 * before the commit RPC, which is felt to be "good 811 * enough". Changing the last argument to ncl_flush() to 812 * a 1 would force a commit operation, if it is felt a 813 * commit is necessary now. 814 * for NFS Version 4 - flush the dirty buffers and commit them, if 815 * nfscl_mustflush() says this is necessary. 816 * It is necessary if there is no write delegation held, 817 * in order to satisfy open/close coherency. 818 * If the file isn't cached on local stable storage, 819 * it may be necessary in order to detect "out of space" 820 * errors from the server, if the write delegation 821 * issued by the server doesn't allow the file to grow. 822 */ 823 /* ARGSUSED */ 824 static int 825 nfs_close(struct vop_close_args *ap) 826 { 827 struct vnode *vp = ap->a_vp; 828 struct nfsnode *np = VTONFS(vp); 829 struct nfsvattr nfsva; 830 struct ucred *cred; 831 int error = 0, ret, localcred = 0; 832 int fmode = ap->a_fflag; 833 834 if (NFSCL_FORCEDISM(vp->v_mount)) 835 return (0); 836 /* 837 * During shutdown, a_cred isn't valid, so just use root. 838 */ 839 if (ap->a_cred == NOCRED) { 840 cred = newnfs_getcred(); 841 localcred = 1; 842 } else { 843 cred = ap->a_cred; 844 } 845 if (vp->v_type == VREG) { 846 /* 847 * Examine and clean dirty pages, regardless of NMODIFIED. 848 * This closes a major hole in close-to-open consistency. 849 * We want to push out all dirty pages (and buffers) on 850 * close, regardless of whether they were dirtied by 851 * mmap'ed writes or via write(). 852 */ 853 if (nfs_clean_pages_on_close && vp->v_object) { 854 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 855 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 856 if (VN_IS_DOOMED(vp) && ap->a_fflag != FNONBLOCK) 857 return (EBADF); 858 } 859 VM_OBJECT_WLOCK(vp->v_object); 860 vm_object_page_clean(vp->v_object, 0, 0, 0); 861 VM_OBJECT_WUNLOCK(vp->v_object); 862 } 863 NFSLOCKNODE(np); 864 if (np->n_flag & NMODIFIED) { 865 NFSUNLOCKNODE(np); 866 if (NFS_ISV3(vp)) { 867 /* 868 * Under NFSv3 we have dirty buffers to dispose of. We 869 * must flush them to the NFS server. We have the option 870 * of waiting all the way through the commit rpc or just 871 * waiting for the initial write. The default is to only 872 * wait through the initial write so the data is in the 873 * server's cache, which is roughly similar to the state 874 * a standard disk subsystem leaves the file in on close(). 875 * 876 * We cannot clear the NMODIFIED bit in np->n_flag due to 877 * potential races with other processes, and certainly 878 * cannot clear it if we don't commit. 879 * These races occur when there is no longer the old 880 * traditional vnode locking implemented for Vnode Ops. 881 */ 882 int cm = newnfs_commit_on_close ? 1 : 0; 883 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 884 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 885 if (VN_IS_DOOMED(vp) && ap->a_fflag != FNONBLOCK) 886 return (EBADF); 887 } 888 error = ncl_flush(vp, MNT_WAIT, ap->a_td, cm, 0); 889 /* np->n_flag &= ~NMODIFIED; */ 890 } else if (NFS_ISV4(vp)) { 891 if (nfscl_mustflush(vp) != 0) { 892 int cm = newnfs_commit_on_close ? 1 : 0; 893 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 894 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 895 if (VN_IS_DOOMED(vp) && ap->a_fflag != 896 FNONBLOCK) 897 return (EBADF); 898 } 899 error = ncl_flush(vp, MNT_WAIT, ap->a_td, 900 cm, 0); 901 /* 902 * as above w.r.t races when clearing 903 * NMODIFIED. 904 * np->n_flag &= ~NMODIFIED; 905 */ 906 } 907 } else { 908 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 909 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 910 if (VN_IS_DOOMED(vp) && ap->a_fflag != 911 FNONBLOCK) 912 return (EBADF); 913 } 914 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 915 } 916 NFSLOCKNODE(np); 917 } 918 /* 919 * Invalidate the attribute cache in all cases. 920 * An open is going to fetch fresh attrs any way, other procs 921 * on this node that have file open will be forced to do an 922 * otw attr fetch, but this is safe. 923 * --> A user found that their RPC count dropped by 20% when 924 * this was commented out and I can't see any requirement 925 * for it, so I've disabled it when negative lookups are 926 * enabled. (What does this have to do with negative lookup 927 * caching? Well nothing, except it was reported by the 928 * same user that needed negative lookup caching and I wanted 929 * there to be a way to disable it to see if it 930 * is the cause of some caching/coherency issue that might 931 * crop up.) 932 */ 933 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) { 934 np->n_attrstamp = 0; 935 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 936 } 937 if (np->n_flag & NWRITEERR) { 938 np->n_flag &= ~NWRITEERR; 939 error = np->n_error; 940 } 941 NFSUNLOCKNODE(np); 942 } 943 944 if (NFS_ISV4(vp)) { 945 /* 946 * Get attributes so "change" is up to date. 947 */ 948 if (error == 0 && nfscl_mustflush(vp) != 0 && 949 vp->v_type == VREG && 950 (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) { 951 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva, 952 NULL); 953 if (!ret) { 954 np->n_change = nfsva.na_filerev; 955 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 956 0, 0); 957 } 958 } 959 960 /* 961 * and do the close. 962 */ 963 ret = nfsrpc_close(vp, 0, ap->a_td); 964 if (!error && ret) 965 error = ret; 966 if (error) 967 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 968 (gid_t)0); 969 } 970 if (newnfs_directio_enable) 971 KASSERT((np->n_directio_asyncwr == 0), 972 ("nfs_close: dirty unflushed (%d) directio buffers\n", 973 np->n_directio_asyncwr)); 974 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) { 975 NFSLOCKNODE(np); 976 KASSERT((np->n_directio_opens > 0), 977 ("nfs_close: unexpectedly value (0) of n_directio_opens\n")); 978 np->n_directio_opens--; 979 if (np->n_directio_opens == 0) 980 np->n_flag &= ~NNONCACHE; 981 NFSUNLOCKNODE(np); 982 } 983 if (localcred) 984 NFSFREECRED(cred); 985 return (error); 986 } 987 988 /* 989 * nfs getattr call from vfs. 990 */ 991 static int 992 nfs_getattr(struct vop_getattr_args *ap) 993 { 994 struct vnode *vp = ap->a_vp; 995 struct thread *td = curthread; /* XXX */ 996 struct nfsnode *np = VTONFS(vp); 997 int error = 0; 998 struct nfsvattr nfsva; 999 struct vattr *vap = ap->a_vap; 1000 struct vattr vattr; 1001 1002 /* 1003 * Update local times for special files. 1004 */ 1005 NFSLOCKNODE(np); 1006 if (np->n_flag & (NACC | NUPD)) 1007 np->n_flag |= NCHG; 1008 NFSUNLOCKNODE(np); 1009 /* 1010 * First look in the cache. 1011 */ 1012 if (ncl_getattrcache(vp, &vattr) == 0) { 1013 ncl_copy_vattr(vap, &vattr); 1014 1015 /* 1016 * Get the local modify time for the case of a write 1017 * delegation. 1018 */ 1019 nfscl_deleggetmodtime(vp, &vap->va_mtime); 1020 return (0); 1021 } 1022 1023 if (NFS_ISV34(vp) && nfs_prime_access_cache && 1024 nfsaccess_cache_timeout > 0) { 1025 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses); 1026 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL); 1027 if (ncl_getattrcache(vp, ap->a_vap) == 0) { 1028 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime); 1029 return (0); 1030 } 1031 } 1032 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL); 1033 if (!error) 1034 error = nfscl_loadattrcache(&vp, &nfsva, vap, 0, 0); 1035 if (!error) { 1036 /* 1037 * Get the local modify time for the case of a write 1038 * delegation. 1039 */ 1040 nfscl_deleggetmodtime(vp, &vap->va_mtime); 1041 } else if (NFS_ISV4(vp)) { 1042 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1043 } 1044 return (error); 1045 } 1046 1047 /* 1048 * nfs setattr call. 1049 */ 1050 static int 1051 nfs_setattr(struct vop_setattr_args *ap) 1052 { 1053 struct vnode *vp = ap->a_vp; 1054 struct nfsnode *np = VTONFS(vp); 1055 struct thread *td = curthread; /* XXX */ 1056 struct vattr *vap = ap->a_vap; 1057 int error = 0; 1058 u_quad_t tsize; 1059 struct timespec ts; 1060 1061 #ifndef nolint 1062 tsize = (u_quad_t)0; 1063 #endif 1064 1065 /* 1066 * Setting of flags and marking of atimes are not supported. 1067 */ 1068 if (vap->va_flags != VNOVAL) 1069 return (EOPNOTSUPP); 1070 1071 /* 1072 * Disallow write attempts if the filesystem is mounted read-only. 1073 */ 1074 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 1075 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 1076 vap->va_mtime.tv_sec != VNOVAL || 1077 vap->va_birthtime.tv_sec != VNOVAL || 1078 vap->va_mode != (mode_t)VNOVAL) && 1079 (vp->v_mount->mnt_flag & MNT_RDONLY)) 1080 return (EROFS); 1081 if (vap->va_size != VNOVAL) { 1082 switch (vp->v_type) { 1083 case VDIR: 1084 return (EISDIR); 1085 case VCHR: 1086 case VBLK: 1087 case VSOCK: 1088 case VFIFO: 1089 if (vap->va_mtime.tv_sec == VNOVAL && 1090 vap->va_atime.tv_sec == VNOVAL && 1091 vap->va_birthtime.tv_sec == VNOVAL && 1092 vap->va_mode == (mode_t)VNOVAL && 1093 vap->va_uid == (uid_t)VNOVAL && 1094 vap->va_gid == (gid_t)VNOVAL) 1095 return (0); 1096 vap->va_size = VNOVAL; 1097 break; 1098 default: 1099 /* 1100 * Disallow write attempts if the filesystem is 1101 * mounted read-only. 1102 */ 1103 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1104 return (EROFS); 1105 /* 1106 * We run vnode_pager_setsize() early (why?), 1107 * we must set np->n_size now to avoid vinvalbuf 1108 * V_SAVE races that might setsize a lower 1109 * value. 1110 */ 1111 NFSLOCKNODE(np); 1112 tsize = np->n_size; 1113 NFSUNLOCKNODE(np); 1114 error = ncl_meta_setsize(vp, td, vap->va_size); 1115 NFSLOCKNODE(np); 1116 if (np->n_flag & NMODIFIED) { 1117 tsize = np->n_size; 1118 NFSUNLOCKNODE(np); 1119 error = ncl_vinvalbuf(vp, vap->va_size == 0 ? 1120 0 : V_SAVE, td, 1); 1121 if (error != 0) { 1122 vnode_pager_setsize(vp, tsize); 1123 return (error); 1124 } 1125 /* 1126 * Call nfscl_delegmodtime() to set the modify time 1127 * locally, as required. 1128 */ 1129 nfscl_delegmodtime(vp); 1130 } else 1131 NFSUNLOCKNODE(np); 1132 /* 1133 * np->n_size has already been set to vap->va_size 1134 * in ncl_meta_setsize(). We must set it again since 1135 * nfs_loadattrcache() could be called through 1136 * ncl_meta_setsize() and could modify np->n_size. 1137 */ 1138 NFSLOCKNODE(np); 1139 np->n_vattr.na_size = np->n_size = vap->va_size; 1140 NFSUNLOCKNODE(np); 1141 } 1142 } else { 1143 NFSLOCKNODE(np); 1144 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) && 1145 (np->n_flag & NMODIFIED) && vp->v_type == VREG) { 1146 NFSUNLOCKNODE(np); 1147 error = ncl_vinvalbuf(vp, V_SAVE, td, 1); 1148 if (error == EINTR || error == EIO) 1149 return (error); 1150 } else 1151 NFSUNLOCKNODE(np); 1152 } 1153 error = nfs_setattrrpc(vp, vap, ap->a_cred, td); 1154 if (vap->va_size != VNOVAL) { 1155 if (error == 0) { 1156 nanouptime(&ts); 1157 NFSLOCKNODE(np); 1158 np->n_localmodtime = ts; 1159 NFSUNLOCKNODE(np); 1160 } else { 1161 NFSLOCKNODE(np); 1162 np->n_size = np->n_vattr.na_size = tsize; 1163 vnode_pager_setsize(vp, tsize); 1164 NFSUNLOCKNODE(np); 1165 } 1166 } 1167 return (error); 1168 } 1169 1170 /* 1171 * Do an nfs setattr rpc. 1172 */ 1173 static int 1174 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred, 1175 struct thread *td) 1176 { 1177 struct nfsnode *np = VTONFS(vp); 1178 int error, ret, attrflag, i; 1179 struct nfsvattr nfsva; 1180 1181 if (NFS_ISV34(vp)) { 1182 NFSLOCKNODE(np); 1183 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) 1184 np->n_accesscache[i].stamp = 0; 1185 np->n_flag |= NDELEGMOD; 1186 NFSUNLOCKNODE(np); 1187 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp); 1188 } 1189 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag, 1190 NULL); 1191 if (attrflag) { 1192 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 1193 if (ret && !error) 1194 error = ret; 1195 } 1196 if (error && NFS_ISV4(vp)) 1197 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid); 1198 return (error); 1199 } 1200 1201 /* 1202 * nfs lookup call, one step at a time... 1203 * First look in cache 1204 * If not found, unlock the directory nfsnode and do the rpc 1205 */ 1206 static int 1207 nfs_lookup(struct vop_lookup_args *ap) 1208 { 1209 struct componentname *cnp = ap->a_cnp; 1210 struct vnode *dvp = ap->a_dvp; 1211 struct vnode **vpp = ap->a_vpp; 1212 struct mount *mp = dvp->v_mount; 1213 int flags = cnp->cn_flags; 1214 struct vnode *newvp; 1215 struct nfsmount *nmp; 1216 struct nfsnode *np, *newnp; 1217 int error = 0, attrflag, dattrflag, ltype, ncticks; 1218 struct thread *td = curthread; 1219 struct nfsfh *nfhp; 1220 struct nfsvattr dnfsva, nfsva; 1221 struct vattr vattr; 1222 struct timespec nctime, ts; 1223 uint32_t openmode; 1224 1225 *vpp = NULLVP; 1226 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) && 1227 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 1228 return (EROFS); 1229 if (dvp->v_type != VDIR) 1230 return (ENOTDIR); 1231 nmp = VFSTONFS(mp); 1232 np = VTONFS(dvp); 1233 1234 /* For NFSv4, wait until any remove is done. */ 1235 NFSLOCKNODE(np); 1236 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) { 1237 np->n_flag |= NREMOVEWANT; 1238 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0); 1239 } 1240 NFSUNLOCKNODE(np); 1241 1242 error = vn_dir_check_exec(dvp, cnp); 1243 if (error != 0) 1244 return (error); 1245 error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks); 1246 if (error > 0 && error != ENOENT) 1247 return (error); 1248 if (error == -1) { 1249 /* 1250 * Lookups of "." are special and always return the 1251 * current directory. cache_lookup() already handles 1252 * associated locking bookkeeping, etc. 1253 */ 1254 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 1255 /* XXX: Is this really correct? */ 1256 if (cnp->cn_nameiop != LOOKUP && 1257 (flags & ISLASTCN)) 1258 cnp->cn_flags |= SAVENAME; 1259 return (0); 1260 } 1261 1262 /* 1263 * We only accept a positive hit in the cache if the 1264 * change time of the file matches our cached copy. 1265 * Otherwise, we discard the cache entry and fallback 1266 * to doing a lookup RPC. We also only trust cache 1267 * entries for less than nm_nametimeo seconds. 1268 * 1269 * To better handle stale file handles and attributes, 1270 * clear the attribute cache of this node if it is a 1271 * leaf component, part of an open() call, and not 1272 * locally modified before fetching the attributes. 1273 * This should allow stale file handles to be detected 1274 * here where we can fall back to a LOOKUP RPC to 1275 * recover rather than having nfs_open() detect the 1276 * stale file handle and failing open(2) with ESTALE. 1277 */ 1278 newvp = *vpp; 1279 newnp = VTONFS(newvp); 1280 if (!(nmp->nm_flag & NFSMNT_NOCTO) && 1281 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1282 !(newnp->n_flag & NMODIFIED)) { 1283 NFSLOCKNODE(newnp); 1284 newnp->n_attrstamp = 0; 1285 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1286 NFSUNLOCKNODE(newnp); 1287 } 1288 if (nfscl_nodeleg(newvp, 0) == 0 || 1289 ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) && 1290 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 && 1291 timespeccmp(&vattr.va_ctime, &nctime, ==))) { 1292 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits); 1293 if (cnp->cn_nameiop != LOOKUP && 1294 (flags & ISLASTCN)) 1295 cnp->cn_flags |= SAVENAME; 1296 return (0); 1297 } 1298 cache_purge(newvp); 1299 if (dvp != newvp) 1300 vput(newvp); 1301 else 1302 vrele(newvp); 1303 *vpp = NULLVP; 1304 } else if (error == ENOENT) { 1305 if (VN_IS_DOOMED(dvp)) 1306 return (ENOENT); 1307 /* 1308 * We only accept a negative hit in the cache if the 1309 * modification time of the parent directory matches 1310 * the cached copy in the name cache entry. 1311 * Otherwise, we discard all of the negative cache 1312 * entries for this directory. We also only trust 1313 * negative cache entries for up to nm_negnametimeo 1314 * seconds. 1315 */ 1316 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) && 1317 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 && 1318 timespeccmp(&vattr.va_mtime, &nctime, ==)) { 1319 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits); 1320 return (ENOENT); 1321 } 1322 cache_purge_negative(dvp); 1323 } 1324 1325 openmode = 0; 1326 #if 0 1327 /* 1328 * The use of LookupOpen breaks some builds. It is disabled 1329 * until that is fixed. 1330 */ 1331 /* 1332 * If this an NFSv4.1/4.2 mount using the "oneopenown" mount 1333 * option, it is possible to do the Open operation in the same 1334 * compound as Lookup, so long as delegations are not being 1335 * issued. This saves doing a separate RPC for Open. 1336 * For pnfs, do not do this, since the Open+LayoutGet will 1337 * be needed as a separate RPC. 1338 */ 1339 NFSLOCKMNT(nmp); 1340 if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp) && !NFSHASPNFS(nmp) && 1341 (nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0 && 1342 (!NFSMNT_RDONLY(mp) || (flags & OPENWRITE) == 0) && 1343 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN)) { 1344 if ((flags & OPENREAD) != 0) 1345 openmode |= NFSV4OPEN_ACCESSREAD; 1346 if ((flags & OPENWRITE) != 0) 1347 openmode |= NFSV4OPEN_ACCESSWRITE; 1348 } 1349 NFSUNLOCKMNT(nmp); 1350 #endif 1351 1352 newvp = NULLVP; 1353 NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses); 1354 nanouptime(&ts); 1355 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1356 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1357 NULL, openmode); 1358 if (dattrflag) 1359 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 1360 if (error) { 1361 if (newvp != NULLVP) { 1362 vput(newvp); 1363 *vpp = NULLVP; 1364 } 1365 1366 if (error != ENOENT) { 1367 if (NFS_ISV4(dvp)) 1368 error = nfscl_maperr(td, error, (uid_t)0, 1369 (gid_t)0); 1370 return (error); 1371 } 1372 1373 /* The requested file was not found. */ 1374 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1375 (flags & ISLASTCN)) { 1376 /* 1377 * XXX: UFS does a full VOP_ACCESS(dvp, 1378 * VWRITE) here instead of just checking 1379 * MNT_RDONLY. 1380 */ 1381 if (mp->mnt_flag & MNT_RDONLY) 1382 return (EROFS); 1383 cnp->cn_flags |= SAVENAME; 1384 return (EJUSTRETURN); 1385 } 1386 1387 if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) { 1388 /* 1389 * Cache the modification time of the parent 1390 * directory from the post-op attributes in 1391 * the name cache entry. The negative cache 1392 * entry will be ignored once the directory 1393 * has changed. Don't bother adding the entry 1394 * if the directory has already changed. 1395 */ 1396 NFSLOCKNODE(np); 1397 if (timespeccmp(&np->n_vattr.na_mtime, 1398 &dnfsva.na_mtime, ==)) { 1399 NFSUNLOCKNODE(np); 1400 cache_enter_time(dvp, NULL, cnp, 1401 &dnfsva.na_mtime, NULL); 1402 } else 1403 NFSUNLOCKNODE(np); 1404 } 1405 return (ENOENT); 1406 } 1407 1408 /* 1409 * Handle RENAME case... 1410 */ 1411 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 1412 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1413 free(nfhp, M_NFSFH); 1414 return (EISDIR); 1415 } 1416 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, LK_EXCLUSIVE); 1417 if (error) 1418 return (error); 1419 newvp = NFSTOV(np); 1420 /* 1421 * If n_localmodtime >= time before RPC, then 1422 * a file modification operation, such as 1423 * VOP_SETATTR() of size, has occurred while 1424 * the Lookup RPC and acquisition of the vnode 1425 * happened. As such, the attributes might 1426 * be stale, with possibly an incorrect size. 1427 */ 1428 NFSLOCKNODE(np); 1429 if (timespecisset(&np->n_localmodtime) && 1430 timespeccmp(&np->n_localmodtime, &ts, >=)) { 1431 NFSCL_DEBUG(4, "nfs_lookup: rename localmod " 1432 "stale attributes\n"); 1433 attrflag = 0; 1434 } 1435 NFSUNLOCKNODE(np); 1436 if (attrflag) 1437 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1438 *vpp = newvp; 1439 cnp->cn_flags |= SAVENAME; 1440 return (0); 1441 } 1442 1443 if (flags & ISDOTDOT) { 1444 ltype = NFSVOPISLOCKED(dvp); 1445 error = vfs_busy(mp, MBF_NOWAIT); 1446 if (error != 0) { 1447 vfs_ref(mp); 1448 NFSVOPUNLOCK(dvp); 1449 error = vfs_busy(mp, 0); 1450 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1451 vfs_rel(mp); 1452 if (error == 0 && VN_IS_DOOMED(dvp)) { 1453 vfs_unbusy(mp); 1454 error = ENOENT; 1455 } 1456 if (error != 0) 1457 return (error); 1458 } 1459 NFSVOPUNLOCK(dvp); 1460 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, 1461 cnp->cn_lkflags); 1462 if (error == 0) 1463 newvp = NFSTOV(np); 1464 vfs_unbusy(mp); 1465 if (newvp != dvp) 1466 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1467 if (VN_IS_DOOMED(dvp)) { 1468 if (error == 0) { 1469 if (newvp == dvp) 1470 vrele(newvp); 1471 else 1472 vput(newvp); 1473 } 1474 error = ENOENT; 1475 } 1476 if (error != 0) 1477 return (error); 1478 if (attrflag) 1479 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1480 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1481 free(nfhp, M_NFSFH); 1482 VREF(dvp); 1483 newvp = dvp; 1484 if (attrflag) 1485 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1486 } else { 1487 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, 1488 cnp->cn_lkflags); 1489 if (error) 1490 return (error); 1491 newvp = NFSTOV(np); 1492 /* 1493 * If n_localmodtime >= time before RPC, then 1494 * a file modification operation, such as 1495 * VOP_SETATTR() of size, has occurred while 1496 * the Lookup RPC and acquisition of the vnode 1497 * happened. As such, the attributes might 1498 * be stale, with possibly an incorrect size. 1499 */ 1500 NFSLOCKNODE(np); 1501 if (timespecisset(&np->n_localmodtime) && 1502 timespeccmp(&np->n_localmodtime, &ts, >=)) { 1503 NFSCL_DEBUG(4, "nfs_lookup: localmod " 1504 "stale attributes\n"); 1505 attrflag = 0; 1506 } 1507 NFSUNLOCKNODE(np); 1508 if (attrflag) 1509 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1510 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1511 !(np->n_flag & NMODIFIED)) { 1512 /* 1513 * Flush the attribute cache when opening a 1514 * leaf node to ensure that fresh attributes 1515 * are fetched in nfs_open() since we did not 1516 * fetch attributes from the LOOKUP reply. 1517 */ 1518 NFSLOCKNODE(np); 1519 np->n_attrstamp = 0; 1520 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1521 NFSUNLOCKNODE(np); 1522 } 1523 } 1524 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1525 cnp->cn_flags |= SAVENAME; 1526 if ((cnp->cn_flags & MAKEENTRY) && dvp != newvp && 1527 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) && 1528 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0)) 1529 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 1530 newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime); 1531 *vpp = newvp; 1532 return (0); 1533 } 1534 1535 /* 1536 * nfs read call. 1537 * Just call ncl_bioread() to do the work. 1538 */ 1539 static int 1540 nfs_read(struct vop_read_args *ap) 1541 { 1542 struct vnode *vp = ap->a_vp; 1543 1544 switch (vp->v_type) { 1545 case VREG: 1546 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred)); 1547 case VDIR: 1548 return (EISDIR); 1549 default: 1550 return (EOPNOTSUPP); 1551 } 1552 } 1553 1554 /* 1555 * nfs readlink call 1556 */ 1557 static int 1558 nfs_readlink(struct vop_readlink_args *ap) 1559 { 1560 struct vnode *vp = ap->a_vp; 1561 1562 if (vp->v_type != VLNK) 1563 return (EINVAL); 1564 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred)); 1565 } 1566 1567 /* 1568 * Do a readlink rpc. 1569 * Called by ncl_doio() from below the buffer cache. 1570 */ 1571 int 1572 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1573 { 1574 int error, ret, attrflag; 1575 struct nfsvattr nfsva; 1576 1577 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva, 1578 &attrflag, NULL); 1579 if (attrflag) { 1580 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 1581 if (ret && !error) 1582 error = ret; 1583 } 1584 if (error && NFS_ISV4(vp)) 1585 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1586 return (error); 1587 } 1588 1589 /* 1590 * nfs read rpc call 1591 * Ditto above 1592 */ 1593 int 1594 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1595 { 1596 int error, ret, attrflag; 1597 struct nfsvattr nfsva; 1598 struct nfsmount *nmp; 1599 1600 nmp = VFSTONFS(vp->v_mount); 1601 error = EIO; 1602 attrflag = 0; 1603 if (NFSHASPNFS(nmp)) 1604 error = nfscl_doiods(vp, uiop, NULL, NULL, 1605 NFSV4OPEN_ACCESSREAD, 0, cred, uiop->uio_td); 1606 NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error); 1607 if (error != 0) 1608 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, 1609 &attrflag, NULL); 1610 if (attrflag) { 1611 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 1612 if (ret && !error) 1613 error = ret; 1614 } 1615 if (error && NFS_ISV4(vp)) 1616 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1617 return (error); 1618 } 1619 1620 /* 1621 * nfs write call 1622 */ 1623 int 1624 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 1625 int *iomode, int *must_commit, int called_from_strategy, int ioflag) 1626 { 1627 struct nfsvattr nfsva; 1628 int error, attrflag, ret; 1629 struct nfsmount *nmp; 1630 1631 nmp = VFSTONFS(vp->v_mount); 1632 error = EIO; 1633 attrflag = 0; 1634 if (NFSHASPNFS(nmp)) 1635 error = nfscl_doiods(vp, uiop, iomode, must_commit, 1636 NFSV4OPEN_ACCESSWRITE, 0, cred, uiop->uio_td); 1637 NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error); 1638 if (error != 0) 1639 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred, 1640 uiop->uio_td, &nfsva, &attrflag, called_from_strategy, 1641 ioflag); 1642 if (attrflag) { 1643 if (VTONFS(vp)->n_flag & ND_NFSV4) 1644 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 1, 1); 1645 else 1646 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 1647 if (ret && !error) 1648 error = ret; 1649 } 1650 if (DOINGASYNC(vp)) 1651 *iomode = NFSWRITE_FILESYNC; 1652 if (error && NFS_ISV4(vp)) 1653 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1654 return (error); 1655 } 1656 1657 /* 1658 * nfs mknod rpc 1659 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1660 * mode set to specify the file type and the size field for rdev. 1661 */ 1662 static int 1663 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1664 struct vattr *vap) 1665 { 1666 struct nfsvattr nfsva, dnfsva; 1667 struct vnode *newvp = NULL; 1668 struct nfsnode *np = NULL, *dnp; 1669 struct nfsfh *nfhp; 1670 struct vattr vattr; 1671 int error = 0, attrflag, dattrflag; 1672 u_int32_t rdev; 1673 1674 if (vap->va_type == VCHR || vap->va_type == VBLK) 1675 rdev = vap->va_rdev; 1676 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1677 rdev = 0xffffffff; 1678 else 1679 return (EOPNOTSUPP); 1680 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1681 return (error); 1682 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap, 1683 rdev, vap->va_type, cnp->cn_cred, curthread, &dnfsva, 1684 &nfsva, &nfhp, &attrflag, &dattrflag, NULL); 1685 if (!error) { 1686 if (!nfhp) 1687 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1688 cnp->cn_namelen, cnp->cn_cred, curthread, 1689 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1690 NULL, 0); 1691 if (nfhp) 1692 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1693 curthread, &np, LK_EXCLUSIVE); 1694 } 1695 if (dattrflag) 1696 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 1697 if (!error) { 1698 newvp = NFSTOV(np); 1699 if (attrflag != 0) { 1700 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1701 if (error != 0) 1702 vput(newvp); 1703 } 1704 } 1705 if (!error) { 1706 *vpp = newvp; 1707 } else if (NFS_ISV4(dvp)) { 1708 error = nfscl_maperr(curthread, error, vap->va_uid, 1709 vap->va_gid); 1710 } 1711 dnp = VTONFS(dvp); 1712 NFSLOCKNODE(dnp); 1713 dnp->n_flag |= NMODIFIED; 1714 if (!dattrflag) { 1715 dnp->n_attrstamp = 0; 1716 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1717 } 1718 NFSUNLOCKNODE(dnp); 1719 return (error); 1720 } 1721 1722 /* 1723 * nfs mknod vop 1724 * just call nfs_mknodrpc() to do the work. 1725 */ 1726 /* ARGSUSED */ 1727 static int 1728 nfs_mknod(struct vop_mknod_args *ap) 1729 { 1730 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap)); 1731 } 1732 1733 static struct mtx nfs_cverf_mtx; 1734 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex", 1735 MTX_DEF); 1736 1737 static nfsquad_t 1738 nfs_get_cverf(void) 1739 { 1740 static nfsquad_t cverf; 1741 nfsquad_t ret; 1742 static int cverf_initialized = 0; 1743 1744 mtx_lock(&nfs_cverf_mtx); 1745 if (cverf_initialized == 0) { 1746 cverf.lval[0] = arc4random(); 1747 cverf.lval[1] = arc4random(); 1748 cverf_initialized = 1; 1749 } else 1750 cverf.qval++; 1751 ret = cverf; 1752 mtx_unlock(&nfs_cverf_mtx); 1753 1754 return (ret); 1755 } 1756 1757 /* 1758 * nfs file create call 1759 */ 1760 static int 1761 nfs_create(struct vop_create_args *ap) 1762 { 1763 struct vnode *dvp = ap->a_dvp; 1764 struct vattr *vap = ap->a_vap; 1765 struct componentname *cnp = ap->a_cnp; 1766 struct nfsnode *np = NULL, *dnp; 1767 struct vnode *newvp = NULL; 1768 struct nfsmount *nmp; 1769 struct nfsvattr dnfsva, nfsva; 1770 struct nfsfh *nfhp; 1771 nfsquad_t cverf; 1772 int error = 0, attrflag, dattrflag, fmode = 0; 1773 struct vattr vattr; 1774 1775 /* 1776 * Oops, not for me.. 1777 */ 1778 if (vap->va_type == VSOCK) 1779 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1780 1781 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1782 return (error); 1783 if (vap->va_vaflags & VA_EXCLUSIVE) 1784 fmode |= O_EXCL; 1785 dnp = VTONFS(dvp); 1786 nmp = VFSTONFS(dvp->v_mount); 1787 again: 1788 /* For NFSv4, wait until any remove is done. */ 1789 NFSLOCKNODE(dnp); 1790 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) { 1791 dnp->n_flag |= NREMOVEWANT; 1792 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0); 1793 } 1794 NFSUNLOCKNODE(dnp); 1795 1796 cverf = nfs_get_cverf(); 1797 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1798 vap, cverf, fmode, cnp->cn_cred, curthread, &dnfsva, &nfsva, 1799 &nfhp, &attrflag, &dattrflag, NULL); 1800 if (!error) { 1801 if (nfhp == NULL) 1802 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1803 cnp->cn_namelen, cnp->cn_cred, curthread, 1804 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1805 NULL, 0); 1806 if (nfhp != NULL) 1807 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1808 curthread, &np, LK_EXCLUSIVE); 1809 } 1810 if (dattrflag) 1811 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 1812 if (!error) { 1813 newvp = NFSTOV(np); 1814 if (attrflag == 0) 1815 error = nfsrpc_getattr(newvp, cnp->cn_cred, 1816 curthread, &nfsva, NULL); 1817 if (error == 0) 1818 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 1819 } 1820 if (error) { 1821 if (newvp != NULL) { 1822 vput(newvp); 1823 newvp = NULL; 1824 } 1825 if (NFS_ISV34(dvp) && (fmode & O_EXCL) && 1826 error == NFSERR_NOTSUPP) { 1827 fmode &= ~O_EXCL; 1828 goto again; 1829 } 1830 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) { 1831 if (nfscl_checksattr(vap, &nfsva)) { 1832 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred, 1833 curthread, &nfsva, &attrflag, NULL); 1834 if (error && (vap->va_uid != (uid_t)VNOVAL || 1835 vap->va_gid != (gid_t)VNOVAL)) { 1836 /* try again without setting uid/gid */ 1837 vap->va_uid = (uid_t)VNOVAL; 1838 vap->va_gid = (uid_t)VNOVAL; 1839 error = nfsrpc_setattr(newvp, vap, NULL, 1840 cnp->cn_cred, curthread, &nfsva, 1841 &attrflag, NULL); 1842 } 1843 if (attrflag) 1844 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 1845 0, 1); 1846 if (error != 0) 1847 vput(newvp); 1848 } 1849 } 1850 if (!error) { 1851 if ((cnp->cn_flags & MAKEENTRY) && attrflag) { 1852 if (dvp != newvp) 1853 cache_enter_time(dvp, newvp, cnp, 1854 &nfsva.na_ctime, NULL); 1855 else 1856 printf("nfs_create: bogus NFS server returned " 1857 "the directory as the new file object\n"); 1858 } 1859 *ap->a_vpp = newvp; 1860 } else if (NFS_ISV4(dvp)) { 1861 error = nfscl_maperr(curthread, error, vap->va_uid, 1862 vap->va_gid); 1863 } 1864 NFSLOCKNODE(dnp); 1865 dnp->n_flag |= NMODIFIED; 1866 if (!dattrflag) { 1867 dnp->n_attrstamp = 0; 1868 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1869 } 1870 NFSUNLOCKNODE(dnp); 1871 return (error); 1872 } 1873 1874 /* 1875 * nfs file remove call 1876 * To try and make nfs semantics closer to ufs semantics, a file that has 1877 * other processes using the vnode is renamed instead of removed and then 1878 * removed later on the last close. 1879 * - If v_usecount > 1 1880 * If a rename is not already in the works 1881 * call nfs_sillyrename() to set it up 1882 * else 1883 * do the remove rpc 1884 */ 1885 static int 1886 nfs_remove(struct vop_remove_args *ap) 1887 { 1888 struct vnode *vp = ap->a_vp; 1889 struct vnode *dvp = ap->a_dvp; 1890 struct componentname *cnp = ap->a_cnp; 1891 struct nfsnode *np = VTONFS(vp); 1892 int error = 0; 1893 struct vattr vattr; 1894 1895 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name")); 1896 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount")); 1897 if (vp->v_type == VDIR) 1898 error = EPERM; 1899 else if (vrefcnt(vp) == 1 || (np->n_sillyrename && 1900 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 1901 vattr.va_nlink > 1)) { 1902 /* 1903 * Purge the name cache so that the chance of a lookup for 1904 * the name succeeding while the remove is in progress is 1905 * minimized. Without node locking it can still happen, such 1906 * that an I/O op returns ESTALE, but since you get this if 1907 * another host removes the file.. 1908 */ 1909 cache_purge(vp); 1910 /* 1911 * throw away biocache buffers, mainly to avoid 1912 * unnecessary delayed writes later. 1913 */ 1914 error = ncl_vinvalbuf(vp, 0, curthread, 1); 1915 if (error != EINTR && error != EIO) 1916 /* Do the rpc */ 1917 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr, 1918 cnp->cn_namelen, cnp->cn_cred, curthread); 1919 /* 1920 * Kludge City: If the first reply to the remove rpc is lost.. 1921 * the reply to the retransmitted request will be ENOENT 1922 * since the file was in fact removed 1923 * Therefore, we cheat and return success. 1924 */ 1925 if (error == ENOENT) 1926 error = 0; 1927 } else if (!np->n_sillyrename) 1928 error = nfs_sillyrename(dvp, vp, cnp); 1929 NFSLOCKNODE(np); 1930 np->n_attrstamp = 0; 1931 NFSUNLOCKNODE(np); 1932 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 1933 return (error); 1934 } 1935 1936 /* 1937 * nfs file remove rpc called from nfs_inactive 1938 */ 1939 int 1940 ncl_removeit(struct sillyrename *sp, struct vnode *vp) 1941 { 1942 /* 1943 * Make sure that the directory vnode is still valid. 1944 * XXX we should lock sp->s_dvp here. 1945 */ 1946 if (sp->s_dvp->v_type == VBAD) 1947 return (0); 1948 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen, 1949 sp->s_cred, NULL)); 1950 } 1951 1952 /* 1953 * Nfs remove rpc, called from nfs_remove() and ncl_removeit(). 1954 */ 1955 static int 1956 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 1957 int namelen, struct ucred *cred, struct thread *td) 1958 { 1959 struct nfsvattr dnfsva; 1960 struct nfsnode *dnp = VTONFS(dvp); 1961 int error = 0, dattrflag; 1962 1963 NFSLOCKNODE(dnp); 1964 dnp->n_flag |= NREMOVEINPROG; 1965 NFSUNLOCKNODE(dnp); 1966 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva, 1967 &dattrflag, NULL); 1968 NFSLOCKNODE(dnp); 1969 if ((dnp->n_flag & NREMOVEWANT)) { 1970 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG); 1971 NFSUNLOCKNODE(dnp); 1972 wakeup((caddr_t)dnp); 1973 } else { 1974 dnp->n_flag &= ~NREMOVEINPROG; 1975 NFSUNLOCKNODE(dnp); 1976 } 1977 if (dattrflag) 1978 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 1979 NFSLOCKNODE(dnp); 1980 dnp->n_flag |= NMODIFIED; 1981 if (!dattrflag) { 1982 dnp->n_attrstamp = 0; 1983 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1984 } 1985 NFSUNLOCKNODE(dnp); 1986 if (error && NFS_ISV4(dvp)) 1987 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1988 return (error); 1989 } 1990 1991 /* 1992 * nfs file rename call 1993 */ 1994 static int 1995 nfs_rename(struct vop_rename_args *ap) 1996 { 1997 struct vnode *fvp = ap->a_fvp; 1998 struct vnode *tvp = ap->a_tvp; 1999 struct vnode *fdvp = ap->a_fdvp; 2000 struct vnode *tdvp = ap->a_tdvp; 2001 struct componentname *tcnp = ap->a_tcnp; 2002 struct componentname *fcnp = ap->a_fcnp; 2003 struct nfsnode *fnp = VTONFS(ap->a_fvp); 2004 struct nfsnode *tdnp = VTONFS(ap->a_tdvp); 2005 struct nfsv4node *newv4 = NULL; 2006 int error; 2007 2008 KASSERT((tcnp->cn_flags & HASBUF) != 0 && 2009 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name")); 2010 /* Check for cross-device rename */ 2011 if ((fvp->v_mount != tdvp->v_mount) || 2012 (tvp && (fvp->v_mount != tvp->v_mount))) { 2013 error = EXDEV; 2014 goto out; 2015 } 2016 2017 if (fvp == tvp) { 2018 printf("nfs_rename: fvp == tvp (can't happen)\n"); 2019 error = 0; 2020 goto out; 2021 } 2022 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0) 2023 goto out; 2024 2025 /* 2026 * We have to flush B_DELWRI data prior to renaming 2027 * the file. If we don't, the delayed-write buffers 2028 * can be flushed out later after the file has gone stale 2029 * under NFSV3. NFSV2 does not have this problem because 2030 * ( as far as I can tell ) it flushes dirty buffers more 2031 * often. 2032 * 2033 * Skip the rename operation if the fsync fails, this can happen 2034 * due to the server's volume being full, when we pushed out data 2035 * that was written back to our cache earlier. Not checking for 2036 * this condition can result in potential (silent) data loss. 2037 */ 2038 error = VOP_FSYNC(fvp, MNT_WAIT, curthread); 2039 NFSVOPUNLOCK(fvp); 2040 if (!error && tvp) 2041 error = VOP_FSYNC(tvp, MNT_WAIT, curthread); 2042 if (error) 2043 goto out; 2044 2045 /* 2046 * If the tvp exists and is in use, sillyrename it before doing the 2047 * rename of the new file over it. 2048 * XXX Can't sillyrename a directory. 2049 */ 2050 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename && 2051 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 2052 vput(tvp); 2053 tvp = NULL; 2054 } 2055 2056 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen, 2057 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 2058 curthread); 2059 2060 if (error == 0 && NFS_ISV4(tdvp)) { 2061 /* 2062 * For NFSv4, check to see if it is the same name and 2063 * replace the name, if it is different. 2064 */ 2065 newv4 = malloc( 2066 sizeof (struct nfsv4node) + 2067 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1, 2068 M_NFSV4NODE, M_WAITOK); 2069 NFSLOCKNODE(tdnp); 2070 NFSLOCKNODE(fnp); 2071 if (fnp->n_v4 != NULL && fvp->v_type == VREG && 2072 (fnp->n_v4->n4_namelen != tcnp->cn_namelen || 2073 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4), 2074 tcnp->cn_namelen) || 2075 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen || 2076 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 2077 tdnp->n_fhp->nfh_len))) { 2078 #ifdef notdef 2079 { char nnn[100]; int nnnl; 2080 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99; 2081 bcopy(tcnp->cn_nameptr, nnn, nnnl); 2082 nnn[nnnl] = '\0'; 2083 printf("ren replace=%s\n",nnn); 2084 } 2085 #endif 2086 free(fnp->n_v4, M_NFSV4NODE); 2087 fnp->n_v4 = newv4; 2088 newv4 = NULL; 2089 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len; 2090 fnp->n_v4->n4_namelen = tcnp->cn_namelen; 2091 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 2092 tdnp->n_fhp->nfh_len); 2093 NFSBCOPY(tcnp->cn_nameptr, 2094 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen); 2095 } 2096 NFSUNLOCKNODE(tdnp); 2097 NFSUNLOCKNODE(fnp); 2098 if (newv4 != NULL) 2099 free(newv4, M_NFSV4NODE); 2100 } 2101 2102 if (fvp->v_type == VDIR) { 2103 if (tvp != NULL && tvp->v_type == VDIR) 2104 cache_purge(tdvp); 2105 cache_purge(fdvp); 2106 } 2107 2108 out: 2109 if (tdvp == tvp) 2110 vrele(tdvp); 2111 else 2112 vput(tdvp); 2113 if (tvp) 2114 vput(tvp); 2115 vrele(fdvp); 2116 vrele(fvp); 2117 /* 2118 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 2119 */ 2120 if (error == ENOENT) 2121 error = 0; 2122 return (error); 2123 } 2124 2125 /* 2126 * nfs file rename rpc called from nfs_remove() above 2127 */ 2128 static int 2129 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp, 2130 struct sillyrename *sp) 2131 { 2132 2133 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen, 2134 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred, 2135 curthread)); 2136 } 2137 2138 /* 2139 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 2140 */ 2141 static int 2142 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr, 2143 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr, 2144 int tnamelen, struct ucred *cred, struct thread *td) 2145 { 2146 struct nfsvattr fnfsva, tnfsva; 2147 struct nfsnode *fdnp = VTONFS(fdvp); 2148 struct nfsnode *tdnp = VTONFS(tdvp); 2149 int error = 0, fattrflag, tattrflag; 2150 2151 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp, 2152 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag, 2153 &tattrflag, NULL, NULL); 2154 NFSLOCKNODE(fdnp); 2155 fdnp->n_flag |= NMODIFIED; 2156 if (fattrflag != 0) { 2157 NFSUNLOCKNODE(fdnp); 2158 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, 0, 1); 2159 } else { 2160 fdnp->n_attrstamp = 0; 2161 NFSUNLOCKNODE(fdnp); 2162 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp); 2163 } 2164 NFSLOCKNODE(tdnp); 2165 tdnp->n_flag |= NMODIFIED; 2166 if (tattrflag != 0) { 2167 NFSUNLOCKNODE(tdnp); 2168 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, 0, 1); 2169 } else { 2170 tdnp->n_attrstamp = 0; 2171 NFSUNLOCKNODE(tdnp); 2172 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 2173 } 2174 if (error && NFS_ISV4(fdvp)) 2175 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2176 return (error); 2177 } 2178 2179 /* 2180 * nfs hard link create call 2181 */ 2182 static int 2183 nfs_link(struct vop_link_args *ap) 2184 { 2185 struct vnode *vp = ap->a_vp; 2186 struct vnode *tdvp = ap->a_tdvp; 2187 struct componentname *cnp = ap->a_cnp; 2188 struct nfsnode *np, *tdnp; 2189 struct nfsvattr nfsva, dnfsva; 2190 int error = 0, attrflag, dattrflag; 2191 2192 /* 2193 * Push all writes to the server, so that the attribute cache 2194 * doesn't get "out of sync" with the server. 2195 * XXX There should be a better way! 2196 */ 2197 VOP_FSYNC(vp, MNT_WAIT, curthread); 2198 2199 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 2200 cnp->cn_cred, curthread, &dnfsva, &nfsva, &attrflag, 2201 &dattrflag, NULL); 2202 tdnp = VTONFS(tdvp); 2203 NFSLOCKNODE(tdnp); 2204 tdnp->n_flag |= NMODIFIED; 2205 if (dattrflag != 0) { 2206 NFSUNLOCKNODE(tdnp); 2207 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, 0, 1); 2208 } else { 2209 tdnp->n_attrstamp = 0; 2210 NFSUNLOCKNODE(tdnp); 2211 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 2212 } 2213 if (attrflag) 2214 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 2215 else { 2216 np = VTONFS(vp); 2217 NFSLOCKNODE(np); 2218 np->n_attrstamp = 0; 2219 NFSUNLOCKNODE(np); 2220 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 2221 } 2222 /* 2223 * If negative lookup caching is enabled, I might as well 2224 * add an entry for this node. Not necessary for correctness, 2225 * but if negative caching is enabled, then the system 2226 * must care about lookup caching hit rate, so... 2227 */ 2228 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 && 2229 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 2230 if (tdvp != vp) 2231 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL); 2232 else 2233 printf("nfs_link: bogus NFS server returned " 2234 "the directory as the new link\n"); 2235 } 2236 if (error && NFS_ISV4(vp)) 2237 error = nfscl_maperr(curthread, error, (uid_t)0, 2238 (gid_t)0); 2239 return (error); 2240 } 2241 2242 /* 2243 * nfs symbolic link create call 2244 */ 2245 static int 2246 nfs_symlink(struct vop_symlink_args *ap) 2247 { 2248 struct vnode *dvp = ap->a_dvp; 2249 struct vattr *vap = ap->a_vap; 2250 struct componentname *cnp = ap->a_cnp; 2251 struct nfsvattr nfsva, dnfsva; 2252 struct nfsfh *nfhp; 2253 struct nfsnode *np = NULL, *dnp; 2254 struct vnode *newvp = NULL; 2255 int error = 0, attrflag, dattrflag, ret; 2256 2257 vap->va_type = VLNK; 2258 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2259 ap->a_target, vap, cnp->cn_cred, curthread, &dnfsva, 2260 &nfsva, &nfhp, &attrflag, &dattrflag, NULL); 2261 if (nfhp) { 2262 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, curthread, 2263 &np, LK_EXCLUSIVE); 2264 if (!ret) 2265 newvp = NFSTOV(np); 2266 else if (!error) 2267 error = ret; 2268 } 2269 if (newvp != NULL) { 2270 if (attrflag) 2271 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 2272 } else if (!error) { 2273 /* 2274 * If we do not have an error and we could not extract the 2275 * newvp from the response due to the request being NFSv2, we 2276 * have to do a lookup in order to obtain a newvp to return. 2277 */ 2278 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2279 cnp->cn_cred, curthread, &np); 2280 if (!error) 2281 newvp = NFSTOV(np); 2282 } 2283 if (error) { 2284 if (newvp) 2285 vput(newvp); 2286 if (NFS_ISV4(dvp)) 2287 error = nfscl_maperr(curthread, error, 2288 vap->va_uid, vap->va_gid); 2289 } else { 2290 *ap->a_vpp = newvp; 2291 } 2292 2293 dnp = VTONFS(dvp); 2294 NFSLOCKNODE(dnp); 2295 dnp->n_flag |= NMODIFIED; 2296 if (dattrflag != 0) { 2297 NFSUNLOCKNODE(dnp); 2298 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 2299 } else { 2300 dnp->n_attrstamp = 0; 2301 NFSUNLOCKNODE(dnp); 2302 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2303 } 2304 /* 2305 * If negative lookup caching is enabled, I might as well 2306 * add an entry for this node. Not necessary for correctness, 2307 * but if negative caching is enabled, then the system 2308 * must care about lookup caching hit rate, so... 2309 */ 2310 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2311 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 2312 if (dvp != newvp) 2313 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 2314 NULL); 2315 else 2316 printf("nfs_symlink: bogus NFS server returned " 2317 "the directory as the new file object\n"); 2318 } 2319 return (error); 2320 } 2321 2322 /* 2323 * nfs make dir call 2324 */ 2325 static int 2326 nfs_mkdir(struct vop_mkdir_args *ap) 2327 { 2328 struct vnode *dvp = ap->a_dvp; 2329 struct vattr *vap = ap->a_vap; 2330 struct componentname *cnp = ap->a_cnp; 2331 struct nfsnode *np = NULL, *dnp; 2332 struct vnode *newvp = NULL; 2333 struct vattr vattr; 2334 struct nfsfh *nfhp; 2335 struct nfsvattr nfsva, dnfsva; 2336 int error = 0, attrflag, dattrflag, ret; 2337 2338 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) 2339 return (error); 2340 vap->va_type = VDIR; 2341 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2342 vap, cnp->cn_cred, curthread, &dnfsva, &nfsva, &nfhp, 2343 &attrflag, &dattrflag, NULL); 2344 dnp = VTONFS(dvp); 2345 NFSLOCKNODE(dnp); 2346 dnp->n_flag |= NMODIFIED; 2347 if (dattrflag != 0) { 2348 NFSUNLOCKNODE(dnp); 2349 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 2350 } else { 2351 dnp->n_attrstamp = 0; 2352 NFSUNLOCKNODE(dnp); 2353 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2354 } 2355 if (nfhp) { 2356 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, curthread, 2357 &np, LK_EXCLUSIVE); 2358 if (!ret) { 2359 newvp = NFSTOV(np); 2360 if (attrflag) 2361 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 2362 0, 1); 2363 } else if (!error) 2364 error = ret; 2365 } 2366 if (!error && newvp == NULL) { 2367 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2368 cnp->cn_cred, curthread, &np); 2369 if (!error) { 2370 newvp = NFSTOV(np); 2371 if (newvp->v_type != VDIR) 2372 error = EEXIST; 2373 } 2374 } 2375 if (error) { 2376 if (newvp) 2377 vput(newvp); 2378 if (NFS_ISV4(dvp)) 2379 error = nfscl_maperr(curthread, error, 2380 vap->va_uid, vap->va_gid); 2381 } else { 2382 /* 2383 * If negative lookup caching is enabled, I might as well 2384 * add an entry for this node. Not necessary for correctness, 2385 * but if negative caching is enabled, then the system 2386 * must care about lookup caching hit rate, so... 2387 */ 2388 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2389 (cnp->cn_flags & MAKEENTRY) && 2390 attrflag != 0 && dattrflag != 0) { 2391 if (dvp != newvp) 2392 cache_enter_time(dvp, newvp, cnp, 2393 &nfsva.na_ctime, &dnfsva.na_ctime); 2394 else 2395 printf("nfs_mkdir: bogus NFS server returned " 2396 "the directory that the directory was " 2397 "created in as the new file object\n"); 2398 } 2399 *ap->a_vpp = newvp; 2400 } 2401 return (error); 2402 } 2403 2404 /* 2405 * nfs remove directory call 2406 */ 2407 static int 2408 nfs_rmdir(struct vop_rmdir_args *ap) 2409 { 2410 struct vnode *vp = ap->a_vp; 2411 struct vnode *dvp = ap->a_dvp; 2412 struct componentname *cnp = ap->a_cnp; 2413 struct nfsnode *dnp; 2414 struct nfsvattr dnfsva; 2415 int error, dattrflag; 2416 2417 if (dvp == vp) 2418 return (EINVAL); 2419 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2420 cnp->cn_cred, curthread, &dnfsva, &dattrflag, NULL); 2421 dnp = VTONFS(dvp); 2422 NFSLOCKNODE(dnp); 2423 dnp->n_flag |= NMODIFIED; 2424 if (dattrflag != 0) { 2425 NFSUNLOCKNODE(dnp); 2426 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 2427 } else { 2428 dnp->n_attrstamp = 0; 2429 NFSUNLOCKNODE(dnp); 2430 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2431 } 2432 2433 cache_purge(dvp); 2434 cache_purge(vp); 2435 if (error && NFS_ISV4(dvp)) 2436 error = nfscl_maperr(curthread, error, (uid_t)0, 2437 (gid_t)0); 2438 /* 2439 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2440 */ 2441 if (error == ENOENT) 2442 error = 0; 2443 return (error); 2444 } 2445 2446 /* 2447 * nfs readdir call 2448 */ 2449 static int 2450 nfs_readdir(struct vop_readdir_args *ap) 2451 { 2452 struct vnode *vp = ap->a_vp; 2453 struct nfsnode *np = VTONFS(vp); 2454 struct uio *uio = ap->a_uio; 2455 ssize_t tresid, left; 2456 int error = 0; 2457 struct vattr vattr; 2458 2459 if (ap->a_eofflag != NULL) 2460 *ap->a_eofflag = 0; 2461 if (vp->v_type != VDIR) 2462 return(EPERM); 2463 2464 /* 2465 * First, check for hit on the EOF offset cache 2466 */ 2467 NFSLOCKNODE(np); 2468 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2469 (np->n_flag & NMODIFIED) == 0) { 2470 NFSUNLOCKNODE(np); 2471 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) { 2472 NFSLOCKNODE(np); 2473 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) || 2474 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 2475 NFSUNLOCKNODE(np); 2476 NFSINCRGLOBAL(nfsstatsv1.direofcache_hits); 2477 if (ap->a_eofflag != NULL) 2478 *ap->a_eofflag = 1; 2479 return (0); 2480 } else 2481 NFSUNLOCKNODE(np); 2482 } 2483 } else 2484 NFSUNLOCKNODE(np); 2485 2486 /* 2487 * NFS always guarantees that directory entries don't straddle 2488 * DIRBLKSIZ boundaries. As such, we need to limit the size 2489 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial 2490 * directory entry. 2491 */ 2492 left = uio->uio_resid % DIRBLKSIZ; 2493 if (left == uio->uio_resid) 2494 return (EINVAL); 2495 uio->uio_resid -= left; 2496 2497 /* 2498 * Call ncl_bioread() to do the real work. 2499 */ 2500 tresid = uio->uio_resid; 2501 error = ncl_bioread(vp, uio, 0, ap->a_cred); 2502 2503 if (!error && uio->uio_resid == tresid) { 2504 NFSINCRGLOBAL(nfsstatsv1.direofcache_misses); 2505 if (ap->a_eofflag != NULL) 2506 *ap->a_eofflag = 1; 2507 } 2508 2509 /* Add the partial DIRBLKSIZ (left) back in. */ 2510 uio->uio_resid += left; 2511 return (error); 2512 } 2513 2514 /* 2515 * Readdir rpc call. 2516 * Called from below the buffer cache by ncl_doio(). 2517 */ 2518 int 2519 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2520 struct thread *td) 2521 { 2522 struct nfsvattr nfsva; 2523 nfsuint64 *cookiep, cookie; 2524 struct nfsnode *dnp = VTONFS(vp); 2525 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2526 int error = 0, eof, attrflag; 2527 2528 KASSERT(uiop->uio_iovcnt == 1 && 2529 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2530 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2531 ("nfs readdirrpc bad uio")); 2532 2533 /* 2534 * If there is no cookie, assume directory was stale. 2535 */ 2536 ncl_dircookie_lock(dnp); 2537 NFSUNLOCKNODE(dnp); 2538 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2539 if (cookiep) { 2540 cookie = *cookiep; 2541 ncl_dircookie_unlock(dnp); 2542 } else { 2543 ncl_dircookie_unlock(dnp); 2544 return (NFSERR_BAD_COOKIE); 2545 } 2546 2547 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2548 (void)ncl_fsinfo(nmp, vp, cred, td); 2549 2550 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva, 2551 &attrflag, &eof, NULL); 2552 if (attrflag) 2553 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 2554 2555 if (!error) { 2556 /* 2557 * We are now either at the end of the directory or have filled 2558 * the block. 2559 */ 2560 if (eof) { 2561 NFSLOCKNODE(dnp); 2562 dnp->n_direofoffset = uiop->uio_offset; 2563 NFSUNLOCKNODE(dnp); 2564 } else { 2565 if (uiop->uio_resid > 0) 2566 printf("EEK! readdirrpc resid > 0\n"); 2567 ncl_dircookie_lock(dnp); 2568 NFSUNLOCKNODE(dnp); 2569 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2570 *cookiep = cookie; 2571 ncl_dircookie_unlock(dnp); 2572 } 2573 } else if (NFS_ISV4(vp)) { 2574 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2575 } 2576 return (error); 2577 } 2578 2579 /* 2580 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc(). 2581 */ 2582 int 2583 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2584 struct thread *td) 2585 { 2586 struct nfsvattr nfsva; 2587 nfsuint64 *cookiep, cookie; 2588 struct nfsnode *dnp = VTONFS(vp); 2589 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2590 int error = 0, attrflag, eof; 2591 2592 KASSERT(uiop->uio_iovcnt == 1 && 2593 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2594 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2595 ("nfs readdirplusrpc bad uio")); 2596 2597 /* 2598 * If there is no cookie, assume directory was stale. 2599 */ 2600 ncl_dircookie_lock(dnp); 2601 NFSUNLOCKNODE(dnp); 2602 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2603 if (cookiep) { 2604 cookie = *cookiep; 2605 ncl_dircookie_unlock(dnp); 2606 } else { 2607 ncl_dircookie_unlock(dnp); 2608 return (NFSERR_BAD_COOKIE); 2609 } 2610 2611 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2612 (void)ncl_fsinfo(nmp, vp, cred, td); 2613 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva, 2614 &attrflag, &eof, NULL); 2615 if (attrflag) 2616 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 2617 2618 if (!error) { 2619 /* 2620 * We are now either at end of the directory or have filled the 2621 * the block. 2622 */ 2623 if (eof) { 2624 NFSLOCKNODE(dnp); 2625 dnp->n_direofoffset = uiop->uio_offset; 2626 NFSUNLOCKNODE(dnp); 2627 } else { 2628 if (uiop->uio_resid > 0) 2629 printf("EEK! readdirplusrpc resid > 0\n"); 2630 ncl_dircookie_lock(dnp); 2631 NFSUNLOCKNODE(dnp); 2632 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2633 *cookiep = cookie; 2634 ncl_dircookie_unlock(dnp); 2635 } 2636 } else if (NFS_ISV4(vp)) { 2637 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2638 } 2639 return (error); 2640 } 2641 2642 /* 2643 * Silly rename. To make the NFS filesystem that is stateless look a little 2644 * more like the "ufs" a remove of an active vnode is translated to a rename 2645 * to a funny looking filename that is removed by nfs_inactive on the 2646 * nfsnode. There is the potential for another process on a different client 2647 * to create the same funny name between the nfs_lookitup() fails and the 2648 * nfs_rename() completes, but... 2649 */ 2650 static int 2651 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2652 { 2653 struct sillyrename *sp; 2654 struct nfsnode *np; 2655 int error; 2656 short pid; 2657 unsigned int lticks; 2658 2659 cache_purge(dvp); 2660 np = VTONFS(vp); 2661 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir")); 2662 sp = malloc(sizeof (struct sillyrename), 2663 M_NEWNFSREQ, M_WAITOK); 2664 sp->s_cred = crhold(cnp->cn_cred); 2665 sp->s_dvp = dvp; 2666 VREF(dvp); 2667 2668 /* 2669 * Fudge together a funny name. 2670 * Changing the format of the funny name to accommodate more 2671 * sillynames per directory. 2672 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is 2673 * CPU ticks since boot. 2674 */ 2675 pid = curthread->td_proc->p_pid; 2676 lticks = (unsigned int)ticks; 2677 for ( ; ; ) { 2678 sp->s_namlen = sprintf(sp->s_name, 2679 ".nfs.%08x.%04x4.4", lticks, 2680 pid); 2681 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2682 curthread, NULL)) 2683 break; 2684 lticks++; 2685 } 2686 error = nfs_renameit(dvp, vp, cnp, sp); 2687 if (error) 2688 goto bad; 2689 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2690 curthread, &np); 2691 np->n_sillyrename = sp; 2692 return (0); 2693 bad: 2694 vrele(sp->s_dvp); 2695 crfree(sp->s_cred); 2696 free(sp, M_NEWNFSREQ); 2697 return (error); 2698 } 2699 2700 /* 2701 * Look up a file name and optionally either update the file handle or 2702 * allocate an nfsnode, depending on the value of npp. 2703 * npp == NULL --> just do the lookup 2704 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2705 * handled too 2706 * *npp != NULL --> update the file handle in the vnode 2707 */ 2708 static int 2709 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred, 2710 struct thread *td, struct nfsnode **npp) 2711 { 2712 struct vnode *newvp = NULL, *vp; 2713 struct nfsnode *np, *dnp = VTONFS(dvp); 2714 struct nfsfh *nfhp, *onfhp; 2715 struct nfsvattr nfsva, dnfsva; 2716 struct componentname cn; 2717 int error = 0, attrflag, dattrflag; 2718 u_int hash; 2719 struct timespec ts; 2720 2721 nanouptime(&ts); 2722 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva, 2723 &nfhp, &attrflag, &dattrflag, NULL, 0); 2724 if (dattrflag) 2725 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1); 2726 if (npp && !error) { 2727 if (*npp != NULL) { 2728 np = *npp; 2729 vp = NFSTOV(np); 2730 /* 2731 * For NFSv4, check to see if it is the same name and 2732 * replace the name, if it is different. 2733 */ 2734 if (np->n_v4 != NULL && nfsva.na_type == VREG && 2735 (np->n_v4->n4_namelen != len || 2736 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) || 2737 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || 2738 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2739 dnp->n_fhp->nfh_len))) { 2740 #ifdef notdef 2741 { char nnn[100]; int nnnl; 2742 nnnl = (len < 100) ? len : 99; 2743 bcopy(name, nnn, nnnl); 2744 nnn[nnnl] = '\0'; 2745 printf("replace=%s\n",nnn); 2746 } 2747 #endif 2748 free(np->n_v4, M_NFSV4NODE); 2749 np->n_v4 = malloc( 2750 sizeof (struct nfsv4node) + 2751 dnp->n_fhp->nfh_len + len - 1, 2752 M_NFSV4NODE, M_WAITOK); 2753 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; 2754 np->n_v4->n4_namelen = len; 2755 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2756 dnp->n_fhp->nfh_len); 2757 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len); 2758 } 2759 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, 2760 FNV1_32_INIT); 2761 onfhp = np->n_fhp; 2762 /* 2763 * Rehash node for new file handle. 2764 */ 2765 vfs_hash_rehash(vp, hash); 2766 np->n_fhp = nfhp; 2767 if (onfhp != NULL) 2768 free(onfhp, M_NFSFH); 2769 newvp = NFSTOV(np); 2770 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) { 2771 free(nfhp, M_NFSFH); 2772 VREF(dvp); 2773 newvp = dvp; 2774 } else { 2775 cn.cn_nameptr = name; 2776 cn.cn_namelen = len; 2777 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td, 2778 &np, LK_EXCLUSIVE); 2779 if (error) 2780 return (error); 2781 newvp = NFSTOV(np); 2782 /* 2783 * If n_localmodtime >= time before RPC, then 2784 * a file modification operation, such as 2785 * VOP_SETATTR() of size, has occurred while 2786 * the Lookup RPC and acquisition of the vnode 2787 * happened. As such, the attributes might 2788 * be stale, with possibly an incorrect size. 2789 */ 2790 NFSLOCKNODE(np); 2791 if (timespecisset(&np->n_localmodtime) && 2792 timespeccmp(&np->n_localmodtime, &ts, >=)) { 2793 NFSCL_DEBUG(4, "nfs_lookitup: localmod " 2794 "stale attributes\n"); 2795 attrflag = 0; 2796 } 2797 NFSUNLOCKNODE(np); 2798 } 2799 if (!attrflag && *npp == NULL) { 2800 if (newvp == dvp) 2801 vrele(newvp); 2802 else 2803 vput(newvp); 2804 return (ENOENT); 2805 } 2806 if (attrflag) 2807 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1); 2808 } 2809 if (npp && *npp == NULL) { 2810 if (error) { 2811 if (newvp) { 2812 if (newvp == dvp) 2813 vrele(newvp); 2814 else 2815 vput(newvp); 2816 } 2817 } else 2818 *npp = np; 2819 } 2820 if (error && NFS_ISV4(dvp)) 2821 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2822 return (error); 2823 } 2824 2825 /* 2826 * Nfs Version 3 and 4 commit rpc 2827 */ 2828 int 2829 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred, 2830 struct thread *td) 2831 { 2832 struct nfsvattr nfsva; 2833 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2834 struct nfsnode *np; 2835 struct uio uio; 2836 int error, attrflag; 2837 2838 np = VTONFS(vp); 2839 error = EIO; 2840 attrflag = 0; 2841 if (NFSHASPNFS(nmp) && (np->n_flag & NDSCOMMIT) != 0) { 2842 uio.uio_offset = offset; 2843 uio.uio_resid = cnt; 2844 error = nfscl_doiods(vp, &uio, NULL, NULL, 2845 NFSV4OPEN_ACCESSWRITE, 1, cred, td); 2846 if (error != 0) { 2847 NFSLOCKNODE(np); 2848 np->n_flag &= ~NDSCOMMIT; 2849 NFSUNLOCKNODE(np); 2850 } 2851 } 2852 if (error != 0) { 2853 mtx_lock(&nmp->nm_mtx); 2854 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) { 2855 mtx_unlock(&nmp->nm_mtx); 2856 return (0); 2857 } 2858 mtx_unlock(&nmp->nm_mtx); 2859 error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva, 2860 &attrflag, NULL); 2861 } 2862 if (attrflag != 0) 2863 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 2864 if (error != 0 && NFS_ISV4(vp)) 2865 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2866 return (error); 2867 } 2868 2869 /* 2870 * Strategy routine. 2871 * For async requests when nfsiod(s) are running, queue the request by 2872 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the 2873 * request. 2874 */ 2875 static int 2876 nfs_strategy(struct vop_strategy_args *ap) 2877 { 2878 struct buf *bp; 2879 struct vnode *vp; 2880 struct ucred *cr; 2881 2882 bp = ap->a_bp; 2883 vp = ap->a_vp; 2884 KASSERT(bp->b_vp == vp, ("missing b_getvp")); 2885 KASSERT(!(bp->b_flags & B_DONE), 2886 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); 2887 2888 if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno) 2889 bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize / 2890 DEV_BSIZE); 2891 if (bp->b_iocmd == BIO_READ) 2892 cr = bp->b_rcred; 2893 else 2894 cr = bp->b_wcred; 2895 2896 /* 2897 * If the op is asynchronous and an i/o daemon is waiting 2898 * queue the request, wake it up and wait for completion 2899 * otherwise just do it ourselves. 2900 */ 2901 if ((bp->b_flags & B_ASYNC) == 0 || 2902 ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread)) 2903 (void) ncl_doio(vp, bp, cr, curthread, 1); 2904 return (0); 2905 } 2906 2907 /* 2908 * fsync vnode op. Just call ncl_flush() with commit == 1. 2909 */ 2910 /* ARGSUSED */ 2911 static int 2912 nfs_fsync(struct vop_fsync_args *ap) 2913 { 2914 2915 if (ap->a_vp->v_type != VREG) { 2916 /* 2917 * For NFS, metadata is changed synchronously on the server, 2918 * so there is nothing to flush. Also, ncl_flush() clears 2919 * the NMODIFIED flag and that shouldn't be done here for 2920 * directories. 2921 */ 2922 return (0); 2923 } 2924 return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0)); 2925 } 2926 2927 /* 2928 * Flush all the blocks associated with a vnode. 2929 * Walk through the buffer pool and push any dirty pages 2930 * associated with the vnode. 2931 * If the called_from_renewthread argument is TRUE, it has been called 2932 * from the NFSv4 renew thread and, as such, cannot block indefinitely 2933 * waiting for a buffer write to complete. 2934 */ 2935 int 2936 ncl_flush(struct vnode *vp, int waitfor, struct thread *td, 2937 int commit, int called_from_renewthread) 2938 { 2939 struct nfsnode *np = VTONFS(vp); 2940 struct buf *bp; 2941 int i; 2942 struct buf *nbp; 2943 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2944 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos; 2945 int passone = 1, trycnt = 0; 2946 u_quad_t off, endoff, toff; 2947 struct ucred* wcred = NULL; 2948 struct buf **bvec = NULL; 2949 struct bufobj *bo; 2950 #ifndef NFS_COMMITBVECSIZ 2951 #define NFS_COMMITBVECSIZ 20 2952 #endif 2953 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ]; 2954 u_int bvecsize = 0, bveccount; 2955 struct timespec ts; 2956 2957 if (called_from_renewthread != 0) 2958 slptimeo = hz; 2959 if (nmp->nm_flag & NFSMNT_INT) 2960 slpflag = PCATCH; 2961 if (!commit) 2962 passone = 0; 2963 bo = &vp->v_bufobj; 2964 /* 2965 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the 2966 * server, but has not been committed to stable storage on the server 2967 * yet. On the first pass, the byte range is worked out and the commit 2968 * rpc is done. On the second pass, ncl_writebp() is called to do the 2969 * job. 2970 */ 2971 again: 2972 off = (u_quad_t)-1; 2973 endoff = 0; 2974 bvecpos = 0; 2975 if (NFS_ISV34(vp) && commit) { 2976 if (bvec != NULL && bvec != bvec_on_stack) 2977 free(bvec, M_TEMP); 2978 /* 2979 * Count up how many buffers waiting for a commit. 2980 */ 2981 bveccount = 0; 2982 BO_LOCK(bo); 2983 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2984 if (!BUF_ISLOCKED(bp) && 2985 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) 2986 == (B_DELWRI | B_NEEDCOMMIT)) 2987 bveccount++; 2988 } 2989 /* 2990 * Allocate space to remember the list of bufs to commit. It is 2991 * important to use M_NOWAIT here to avoid a race with nfs_write. 2992 * If we can't get memory (for whatever reason), we will end up 2993 * committing the buffers one-by-one in the loop below. 2994 */ 2995 if (bveccount > NFS_COMMITBVECSIZ) { 2996 /* 2997 * Release the vnode interlock to avoid a lock 2998 * order reversal. 2999 */ 3000 BO_UNLOCK(bo); 3001 bvec = (struct buf **) 3002 malloc(bveccount * sizeof(struct buf *), 3003 M_TEMP, M_NOWAIT); 3004 BO_LOCK(bo); 3005 if (bvec == NULL) { 3006 bvec = bvec_on_stack; 3007 bvecsize = NFS_COMMITBVECSIZ; 3008 } else 3009 bvecsize = bveccount; 3010 } else { 3011 bvec = bvec_on_stack; 3012 bvecsize = NFS_COMMITBVECSIZ; 3013 } 3014 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 3015 if (bvecpos >= bvecsize) 3016 break; 3017 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 3018 nbp = TAILQ_NEXT(bp, b_bobufs); 3019 continue; 3020 } 3021 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 3022 (B_DELWRI | B_NEEDCOMMIT)) { 3023 BUF_UNLOCK(bp); 3024 nbp = TAILQ_NEXT(bp, b_bobufs); 3025 continue; 3026 } 3027 BO_UNLOCK(bo); 3028 bremfree(bp); 3029 /* 3030 * Work out if all buffers are using the same cred 3031 * so we can deal with them all with one commit. 3032 * 3033 * NOTE: we are not clearing B_DONE here, so we have 3034 * to do it later on in this routine if we intend to 3035 * initiate I/O on the bp. 3036 * 3037 * Note: to avoid loopback deadlocks, we do not 3038 * assign b_runningbufspace. 3039 */ 3040 if (wcred == NULL) 3041 wcred = bp->b_wcred; 3042 else if (wcred != bp->b_wcred) 3043 wcred = NOCRED; 3044 vfs_busy_pages(bp, 1); 3045 3046 BO_LOCK(bo); 3047 /* 3048 * bp is protected by being locked, but nbp is not 3049 * and vfs_busy_pages() may sleep. We have to 3050 * recalculate nbp. 3051 */ 3052 nbp = TAILQ_NEXT(bp, b_bobufs); 3053 3054 /* 3055 * A list of these buffers is kept so that the 3056 * second loop knows which buffers have actually 3057 * been committed. This is necessary, since there 3058 * may be a race between the commit rpc and new 3059 * uncommitted writes on the file. 3060 */ 3061 bvec[bvecpos++] = bp; 3062 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 3063 bp->b_dirtyoff; 3064 if (toff < off) 3065 off = toff; 3066 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); 3067 if (toff > endoff) 3068 endoff = toff; 3069 } 3070 BO_UNLOCK(bo); 3071 } 3072 if (bvecpos > 0) { 3073 /* 3074 * Commit data on the server, as required. 3075 * If all bufs are using the same wcred, then use that with 3076 * one call for all of them, otherwise commit each one 3077 * separately. 3078 */ 3079 if (wcred != NOCRED) 3080 retv = ncl_commit(vp, off, (int)(endoff - off), 3081 wcred, td); 3082 else { 3083 retv = 0; 3084 for (i = 0; i < bvecpos; i++) { 3085 off_t off, size; 3086 bp = bvec[i]; 3087 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 3088 bp->b_dirtyoff; 3089 size = (u_quad_t)(bp->b_dirtyend 3090 - bp->b_dirtyoff); 3091 retv = ncl_commit(vp, off, (int)size, 3092 bp->b_wcred, td); 3093 if (retv) break; 3094 } 3095 } 3096 3097 if (retv == NFSERR_STALEWRITEVERF) 3098 ncl_clearcommit(vp->v_mount); 3099 3100 /* 3101 * Now, either mark the blocks I/O done or mark the 3102 * blocks dirty, depending on whether the commit 3103 * succeeded. 3104 */ 3105 for (i = 0; i < bvecpos; i++) { 3106 bp = bvec[i]; 3107 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 3108 if (!NFSCL_FORCEDISM(vp->v_mount) && retv) { 3109 /* 3110 * Error, leave B_DELWRI intact 3111 */ 3112 vfs_unbusy_pages(bp); 3113 brelse(bp); 3114 } else { 3115 /* 3116 * Success, remove B_DELWRI ( bundirty() ). 3117 * 3118 * b_dirtyoff/b_dirtyend seem to be NFS 3119 * specific. We should probably move that 3120 * into bundirty(). XXX 3121 */ 3122 bufobj_wref(bo); 3123 bp->b_flags |= B_ASYNC; 3124 bundirty(bp); 3125 bp->b_flags &= ~B_DONE; 3126 bp->b_ioflags &= ~BIO_ERROR; 3127 bp->b_dirtyoff = bp->b_dirtyend = 0; 3128 bufdone(bp); 3129 } 3130 } 3131 } 3132 3133 /* 3134 * Start/do any write(s) that are required. 3135 */ 3136 loop: 3137 BO_LOCK(bo); 3138 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 3139 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 3140 if (waitfor != MNT_WAIT || passone) 3141 continue; 3142 3143 error = BUF_TIMELOCK(bp, 3144 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 3145 BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo); 3146 if (error == 0) { 3147 BUF_UNLOCK(bp); 3148 goto loop; 3149 } 3150 if (error == ENOLCK) { 3151 error = 0; 3152 goto loop; 3153 } 3154 if (called_from_renewthread != 0) { 3155 /* 3156 * Return EIO so the flush will be retried 3157 * later. 3158 */ 3159 error = EIO; 3160 goto done; 3161 } 3162 if (newnfs_sigintr(nmp, td)) { 3163 error = EINTR; 3164 goto done; 3165 } 3166 if (slpflag == PCATCH) { 3167 slpflag = 0; 3168 slptimeo = 2 * hz; 3169 } 3170 goto loop; 3171 } 3172 if ((bp->b_flags & B_DELWRI) == 0) 3173 panic("nfs_fsync: not dirty"); 3174 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) { 3175 BUF_UNLOCK(bp); 3176 continue; 3177 } 3178 BO_UNLOCK(bo); 3179 bremfree(bp); 3180 bp->b_flags |= B_ASYNC; 3181 bwrite(bp); 3182 if (newnfs_sigintr(nmp, td)) { 3183 error = EINTR; 3184 goto done; 3185 } 3186 goto loop; 3187 } 3188 if (passone) { 3189 passone = 0; 3190 BO_UNLOCK(bo); 3191 goto again; 3192 } 3193 if (waitfor == MNT_WAIT) { 3194 while (bo->bo_numoutput) { 3195 error = bufobj_wwait(bo, slpflag, slptimeo); 3196 if (error) { 3197 BO_UNLOCK(bo); 3198 if (called_from_renewthread != 0) { 3199 /* 3200 * Return EIO so that the flush will be 3201 * retried later. 3202 */ 3203 error = EIO; 3204 goto done; 3205 } 3206 error = newnfs_sigintr(nmp, td); 3207 if (error) 3208 goto done; 3209 if (slpflag == PCATCH) { 3210 slpflag = 0; 3211 slptimeo = 2 * hz; 3212 } 3213 BO_LOCK(bo); 3214 } 3215 } 3216 if (bo->bo_dirty.bv_cnt != 0 && commit) { 3217 BO_UNLOCK(bo); 3218 goto loop; 3219 } 3220 /* 3221 * Wait for all the async IO requests to drain 3222 */ 3223 BO_UNLOCK(bo); 3224 NFSLOCKNODE(np); 3225 while (np->n_directio_asyncwr > 0) { 3226 np->n_flag |= NFSYNCWAIT; 3227 error = newnfs_msleep(td, &np->n_directio_asyncwr, 3228 &np->n_mtx, slpflag | (PRIBIO + 1), 3229 "nfsfsync", 0); 3230 if (error) { 3231 if (newnfs_sigintr(nmp, td)) { 3232 NFSUNLOCKNODE(np); 3233 error = EINTR; 3234 goto done; 3235 } 3236 } 3237 } 3238 NFSUNLOCKNODE(np); 3239 } else 3240 BO_UNLOCK(bo); 3241 if (NFSHASPNFS(nmp)) { 3242 nfscl_layoutcommit(vp, td); 3243 /* 3244 * Invalidate the attribute cache, since writes to a DS 3245 * won't update the size attribute. 3246 */ 3247 NFSLOCKNODE(np); 3248 np->n_attrstamp = 0; 3249 } else 3250 NFSLOCKNODE(np); 3251 if (np->n_flag & NWRITEERR) { 3252 error = np->n_error; 3253 np->n_flag &= ~NWRITEERR; 3254 } 3255 if (commit && bo->bo_dirty.bv_cnt == 0 && 3256 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0) 3257 np->n_flag &= ~NMODIFIED; 3258 NFSUNLOCKNODE(np); 3259 done: 3260 if (bvec != NULL && bvec != bvec_on_stack) 3261 free(bvec, M_TEMP); 3262 if (error == 0 && commit != 0 && waitfor == MNT_WAIT && 3263 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 || 3264 np->n_directio_asyncwr != 0)) { 3265 if (trycnt++ < 5) { 3266 /* try, try again... */ 3267 passone = 1; 3268 wcred = NULL; 3269 bvec = NULL; 3270 bvecsize = 0; 3271 goto again; 3272 } 3273 vn_printf(vp, "ncl_flush failed"); 3274 error = called_from_renewthread != 0 ? EIO : EBUSY; 3275 } 3276 if (error == 0) { 3277 nanouptime(&ts); 3278 NFSLOCKNODE(np); 3279 np->n_localmodtime = ts; 3280 NFSUNLOCKNODE(np); 3281 } 3282 return (error); 3283 } 3284 3285 /* 3286 * NFS advisory byte-level locks. 3287 */ 3288 static int 3289 nfs_advlock(struct vop_advlock_args *ap) 3290 { 3291 struct vnode *vp = ap->a_vp; 3292 struct ucred *cred; 3293 struct nfsnode *np = VTONFS(ap->a_vp); 3294 struct proc *p = (struct proc *)ap->a_id; 3295 struct thread *td = curthread; /* XXX */ 3296 struct vattr va; 3297 int ret, error; 3298 u_quad_t size; 3299 struct nfsmount *nmp; 3300 3301 error = NFSVOPLOCK(vp, LK_SHARED); 3302 if (error != 0) 3303 return (EBADF); 3304 if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) { 3305 if (vp->v_type != VREG) { 3306 error = EINVAL; 3307 goto out; 3308 } 3309 if ((ap->a_flags & F_POSIX) != 0) 3310 cred = p->p_ucred; 3311 else 3312 cred = td->td_ucred; 3313 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 3314 if (VN_IS_DOOMED(vp)) { 3315 error = EBADF; 3316 goto out; 3317 } 3318 3319 /* 3320 * If this is unlocking a write locked region, flush and 3321 * commit them before unlocking. This is required by 3322 * RFC3530 Sec. 9.3.2. 3323 */ 3324 if (ap->a_op == F_UNLCK && 3325 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id, 3326 ap->a_flags)) 3327 (void) ncl_flush(vp, MNT_WAIT, td, 1, 0); 3328 3329 /* 3330 * Mark NFS node as might have acquired a lock. 3331 * This is separate from NHASBEENLOCKED, because it must 3332 * be done before the nfsrpc_advlock() call, which might 3333 * add a nfscllock structure to the client state. 3334 * It is used to check for the case where a nfscllock 3335 * state structure cannot exist for the file. 3336 * Only done for "oneopenown" NFSv4.1/4.2 mounts. 3337 */ 3338 nmp = VFSTONFS(vp->v_mount); 3339 if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp)) { 3340 NFSLOCKNODE(np); 3341 np->n_flag |= NMIGHTBELOCKED; 3342 NFSUNLOCKNODE(np); 3343 } 3344 3345 /* 3346 * Loop around doing the lock op, while a blocking lock 3347 * must wait for the lock op to succeed. 3348 */ 3349 do { 3350 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op, 3351 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags); 3352 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3353 ap->a_op == F_SETLK) { 3354 NFSVOPUNLOCK(vp); 3355 error = nfs_catnap(PZERO | PCATCH, ret, 3356 "ncladvl"); 3357 if (error) 3358 return (EINTR); 3359 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 3360 if (VN_IS_DOOMED(vp)) { 3361 error = EBADF; 3362 goto out; 3363 } 3364 } 3365 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3366 ap->a_op == F_SETLK); 3367 if (ret == NFSERR_DENIED) { 3368 error = EAGAIN; 3369 goto out; 3370 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) { 3371 error = ret; 3372 goto out; 3373 } else if (ret != 0) { 3374 error = EACCES; 3375 goto out; 3376 } 3377 3378 /* 3379 * Now, if we just got a lock, invalidate data in the buffer 3380 * cache, as required, so that the coherency conforms with 3381 * RFC3530 Sec. 9.3.2. 3382 */ 3383 if (ap->a_op == F_SETLK) { 3384 if ((np->n_flag & NMODIFIED) == 0) { 3385 np->n_attrstamp = 0; 3386 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3387 ret = VOP_GETATTR(vp, &va, cred); 3388 } 3389 if ((np->n_flag & NMODIFIED) || ret || 3390 np->n_change != va.va_filerev) { 3391 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1); 3392 np->n_attrstamp = 0; 3393 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3394 ret = VOP_GETATTR(vp, &va, cred); 3395 if (!ret) { 3396 np->n_mtime = va.va_mtime; 3397 np->n_change = va.va_filerev; 3398 } 3399 } 3400 /* Mark that a file lock has been acquired. */ 3401 NFSLOCKNODE(np); 3402 np->n_flag |= NHASBEENLOCKED; 3403 NFSUNLOCKNODE(np); 3404 } 3405 } else if (!NFS_ISV4(vp)) { 3406 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3407 size = VTONFS(vp)->n_size; 3408 NFSVOPUNLOCK(vp); 3409 error = lf_advlock(ap, &(vp->v_lockf), size); 3410 } else { 3411 if (nfs_advlock_p != NULL) 3412 error = nfs_advlock_p(ap); 3413 else { 3414 NFSVOPUNLOCK(vp); 3415 error = ENOLCK; 3416 } 3417 } 3418 if (error == 0 && ap->a_op == F_SETLK) { 3419 error = NFSVOPLOCK(vp, LK_SHARED); 3420 if (error == 0) { 3421 /* Mark that a file lock has been acquired. */ 3422 NFSLOCKNODE(np); 3423 np->n_flag |= NHASBEENLOCKED; 3424 NFSUNLOCKNODE(np); 3425 NFSVOPUNLOCK(vp); 3426 } 3427 } 3428 return (error); 3429 } else 3430 error = EOPNOTSUPP; 3431 out: 3432 NFSVOPUNLOCK(vp); 3433 return (error); 3434 } 3435 3436 /* 3437 * NFS advisory byte-level locks. 3438 */ 3439 static int 3440 nfs_advlockasync(struct vop_advlockasync_args *ap) 3441 { 3442 struct vnode *vp = ap->a_vp; 3443 u_quad_t size; 3444 int error; 3445 3446 if (NFS_ISV4(vp)) 3447 return (EOPNOTSUPP); 3448 error = NFSVOPLOCK(vp, LK_SHARED); 3449 if (error) 3450 return (error); 3451 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3452 size = VTONFS(vp)->n_size; 3453 NFSVOPUNLOCK(vp); 3454 error = lf_advlockasync(ap, &(vp->v_lockf), size); 3455 } else { 3456 NFSVOPUNLOCK(vp); 3457 error = EOPNOTSUPP; 3458 } 3459 return (error); 3460 } 3461 3462 /* 3463 * Print out the contents of an nfsnode. 3464 */ 3465 static int 3466 nfs_print(struct vop_print_args *ap) 3467 { 3468 struct vnode *vp = ap->a_vp; 3469 struct nfsnode *np = VTONFS(vp); 3470 3471 printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid, 3472 (uintmax_t)np->n_vattr.na_fsid); 3473 if (vp->v_type == VFIFO) 3474 fifo_printinfo(vp); 3475 printf("\n"); 3476 return (0); 3477 } 3478 3479 /* 3480 * This is the "real" nfs::bwrite(struct buf*). 3481 * We set B_CACHE if this is a VMIO buffer. 3482 */ 3483 int 3484 ncl_writebp(struct buf *bp, int force __unused, struct thread *td) 3485 { 3486 int oldflags, rtval; 3487 3488 if (bp->b_flags & B_INVAL) { 3489 brelse(bp); 3490 return (0); 3491 } 3492 3493 oldflags = bp->b_flags; 3494 bp->b_flags |= B_CACHE; 3495 3496 /* 3497 * Undirty the bp. We will redirty it later if the I/O fails. 3498 */ 3499 bundirty(bp); 3500 bp->b_flags &= ~B_DONE; 3501 bp->b_ioflags &= ~BIO_ERROR; 3502 bp->b_iocmd = BIO_WRITE; 3503 3504 bufobj_wref(bp->b_bufobj); 3505 curthread->td_ru.ru_oublock++; 3506 3507 /* 3508 * Note: to avoid loopback deadlocks, we do not 3509 * assign b_runningbufspace. 3510 */ 3511 vfs_busy_pages(bp, 1); 3512 3513 BUF_KERNPROC(bp); 3514 bp->b_iooffset = dbtob(bp->b_blkno); 3515 bstrategy(bp); 3516 3517 if ((oldflags & B_ASYNC) != 0) 3518 return (0); 3519 3520 rtval = bufwait(bp); 3521 if (oldflags & B_DELWRI) 3522 reassignbuf(bp); 3523 brelse(bp); 3524 return (rtval); 3525 } 3526 3527 /* 3528 * nfs special file access vnode op. 3529 * Essentially just get vattr and then imitate iaccess() since the device is 3530 * local to the client. 3531 */ 3532 static int 3533 nfsspec_access(struct vop_access_args *ap) 3534 { 3535 struct vattr *vap; 3536 struct ucred *cred = ap->a_cred; 3537 struct vnode *vp = ap->a_vp; 3538 accmode_t accmode = ap->a_accmode; 3539 struct vattr vattr; 3540 int error; 3541 3542 /* 3543 * Disallow write attempts on filesystems mounted read-only; 3544 * unless the file is a socket, fifo, or a block or character 3545 * device resident on the filesystem. 3546 */ 3547 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3548 switch (vp->v_type) { 3549 case VREG: 3550 case VDIR: 3551 case VLNK: 3552 return (EROFS); 3553 default: 3554 break; 3555 } 3556 } 3557 vap = &vattr; 3558 error = VOP_GETATTR(vp, vap, cred); 3559 if (error) 3560 goto out; 3561 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid, 3562 accmode, cred); 3563 out: 3564 return error; 3565 } 3566 3567 /* 3568 * Read wrapper for fifos. 3569 */ 3570 static int 3571 nfsfifo_read(struct vop_read_args *ap) 3572 { 3573 struct nfsnode *np = VTONFS(ap->a_vp); 3574 int error; 3575 3576 /* 3577 * Set access flag. 3578 */ 3579 NFSLOCKNODE(np); 3580 np->n_flag |= NACC; 3581 vfs_timestamp(&np->n_atim); 3582 NFSUNLOCKNODE(np); 3583 error = fifo_specops.vop_read(ap); 3584 return error; 3585 } 3586 3587 /* 3588 * Write wrapper for fifos. 3589 */ 3590 static int 3591 nfsfifo_write(struct vop_write_args *ap) 3592 { 3593 struct nfsnode *np = VTONFS(ap->a_vp); 3594 3595 /* 3596 * Set update flag. 3597 */ 3598 NFSLOCKNODE(np); 3599 np->n_flag |= NUPD; 3600 vfs_timestamp(&np->n_mtim); 3601 NFSUNLOCKNODE(np); 3602 return(fifo_specops.vop_write(ap)); 3603 } 3604 3605 /* 3606 * Close wrapper for fifos. 3607 * 3608 * Update the times on the nfsnode then do fifo close. 3609 */ 3610 static int 3611 nfsfifo_close(struct vop_close_args *ap) 3612 { 3613 struct vnode *vp = ap->a_vp; 3614 struct nfsnode *np = VTONFS(vp); 3615 struct vattr vattr; 3616 struct timespec ts; 3617 3618 NFSLOCKNODE(np); 3619 if (np->n_flag & (NACC | NUPD)) { 3620 vfs_timestamp(&ts); 3621 if (np->n_flag & NACC) 3622 np->n_atim = ts; 3623 if (np->n_flag & NUPD) 3624 np->n_mtim = ts; 3625 np->n_flag |= NCHG; 3626 if (vrefcnt(vp) == 1 && 3627 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3628 VATTR_NULL(&vattr); 3629 if (np->n_flag & NACC) 3630 vattr.va_atime = np->n_atim; 3631 if (np->n_flag & NUPD) 3632 vattr.va_mtime = np->n_mtim; 3633 NFSUNLOCKNODE(np); 3634 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3635 goto out; 3636 } 3637 } 3638 NFSUNLOCKNODE(np); 3639 out: 3640 return (fifo_specops.vop_close(ap)); 3641 } 3642 3643 /* 3644 * Just call ncl_writebp() with the force argument set to 1. 3645 * 3646 * NOTE: B_DONE may or may not be set in a_bp on call. 3647 */ 3648 static int 3649 nfs_bwrite(struct buf *bp) 3650 { 3651 3652 return (ncl_writebp(bp, 1, curthread)); 3653 } 3654 3655 struct buf_ops buf_ops_newnfs = { 3656 .bop_name = "buf_ops_nfs", 3657 .bop_write = nfs_bwrite, 3658 .bop_strategy = bufstrategy, 3659 .bop_sync = bufsync, 3660 .bop_bdflush = bufbdflush, 3661 }; 3662 3663 static int 3664 nfs_getacl(struct vop_getacl_args *ap) 3665 { 3666 int error; 3667 3668 if (ap->a_type != ACL_TYPE_NFS4) 3669 return (EOPNOTSUPP); 3670 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3671 NULL); 3672 if (error > NFSERR_STALE) { 3673 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3674 error = EPERM; 3675 } 3676 return (error); 3677 } 3678 3679 static int 3680 nfs_setacl(struct vop_setacl_args *ap) 3681 { 3682 int error; 3683 3684 if (ap->a_type != ACL_TYPE_NFS4) 3685 return (EOPNOTSUPP); 3686 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3687 NULL); 3688 if (error > NFSERR_STALE) { 3689 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3690 error = EPERM; 3691 } 3692 return (error); 3693 } 3694 3695 /* 3696 * VOP_ADVISE for NFS. 3697 * Just return 0 for any errors, since it is just a hint. 3698 */ 3699 static int 3700 nfs_advise(struct vop_advise_args *ap) 3701 { 3702 struct thread *td = curthread; 3703 struct nfsmount *nmp; 3704 uint64_t len; 3705 int error; 3706 3707 /* 3708 * First do vop_stdadvise() to handle the buffer cache. 3709 */ 3710 error = vop_stdadvise(ap); 3711 if (error != 0) 3712 return (error); 3713 if (ap->a_start < 0 || ap->a_end < 0) 3714 return (0); 3715 if (ap->a_end == OFF_MAX) 3716 len = 0; 3717 else if (ap->a_end < ap->a_start) 3718 return (0); 3719 else 3720 len = ap->a_end - ap->a_start + 1; 3721 nmp = VFSTONFS(ap->a_vp->v_mount); 3722 mtx_lock(&nmp->nm_mtx); 3723 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 3724 (NFSHASPNFS(nmp) && (nmp->nm_privflag & NFSMNTP_IOADVISETHRUMDS) == 3725 0) || (nmp->nm_privflag & NFSMNTP_NOADVISE) != 0) { 3726 mtx_unlock(&nmp->nm_mtx); 3727 return (0); 3728 } 3729 mtx_unlock(&nmp->nm_mtx); 3730 error = nfsrpc_advise(ap->a_vp, ap->a_start, len, ap->a_advice, 3731 td->td_ucred, td); 3732 if (error == NFSERR_NOTSUPP) { 3733 mtx_lock(&nmp->nm_mtx); 3734 nmp->nm_privflag |= NFSMNTP_NOADVISE; 3735 mtx_unlock(&nmp->nm_mtx); 3736 } 3737 return (0); 3738 } 3739 3740 /* 3741 * nfs allocate call 3742 */ 3743 static int 3744 nfs_allocate(struct vop_allocate_args *ap) 3745 { 3746 struct vnode *vp = ap->a_vp; 3747 struct thread *td = curthread; 3748 struct nfsvattr nfsva; 3749 struct nfsmount *nmp; 3750 struct nfsnode *np; 3751 off_t alen; 3752 int attrflag, error, ret; 3753 struct timespec ts; 3754 struct uio io; 3755 3756 attrflag = 0; 3757 nmp = VFSTONFS(vp->v_mount); 3758 np = VTONFS(vp); 3759 mtx_lock(&nmp->nm_mtx); 3760 if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION && 3761 (nmp->nm_privflag & NFSMNTP_NOALLOCATE) == 0) { 3762 mtx_unlock(&nmp->nm_mtx); 3763 alen = *ap->a_len; 3764 if ((uint64_t)alen > nfs_maxalloclen) 3765 alen = nfs_maxalloclen; 3766 3767 /* Check the file size limit. */ 3768 io.uio_offset = *ap->a_offset; 3769 io.uio_resid = alen; 3770 error = vn_rlimit_fsize(vp, &io, td); 3771 3772 /* 3773 * Flush first to ensure that the allocate adds to the 3774 * file's allocation on the server. 3775 */ 3776 if (error == 0) 3777 error = ncl_flush(vp, MNT_WAIT, td, 1, 0); 3778 if (error == 0) 3779 error = nfsrpc_allocate(vp, *ap->a_offset, alen, 3780 &nfsva, &attrflag, ap->a_cred, td, NULL); 3781 if (error == 0) { 3782 *ap->a_offset += alen; 3783 *ap->a_len -= alen; 3784 nanouptime(&ts); 3785 NFSLOCKNODE(np); 3786 np->n_localmodtime = ts; 3787 NFSUNLOCKNODE(np); 3788 } else if (error == NFSERR_NOTSUPP) { 3789 mtx_lock(&nmp->nm_mtx); 3790 nmp->nm_privflag |= NFSMNTP_NOALLOCATE; 3791 mtx_unlock(&nmp->nm_mtx); 3792 error = EINVAL; 3793 } 3794 } else { 3795 mtx_unlock(&nmp->nm_mtx); 3796 error = EINVAL; 3797 } 3798 if (attrflag != 0) { 3799 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 3800 if (error == 0 && ret != 0) 3801 error = ret; 3802 } 3803 if (error != 0) 3804 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 3805 return (error); 3806 } 3807 3808 /* 3809 * nfs deallocate call 3810 */ 3811 static int 3812 nfs_deallocate(struct vop_deallocate_args *ap) 3813 { 3814 struct vnode *vp = ap->a_vp; 3815 struct thread *td = curthread; 3816 struct nfsvattr nfsva; 3817 struct nfsmount *nmp; 3818 struct nfsnode *np; 3819 off_t tlen, mlen; 3820 int attrflag, error, ret; 3821 bool clipped; 3822 struct timespec ts; 3823 3824 error = 0; 3825 attrflag = 0; 3826 nmp = VFSTONFS(vp->v_mount); 3827 np = VTONFS(vp); 3828 mtx_lock(&nmp->nm_mtx); 3829 if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION && 3830 (nmp->nm_privflag & NFSMNTP_NODEALLOCATE) == 0) { 3831 mtx_unlock(&nmp->nm_mtx); 3832 tlen = omin(OFF_MAX - *ap->a_offset, *ap->a_len); 3833 NFSCL_DEBUG(4, "dealloc: off=%jd len=%jd maxfilesize=%ju\n", 3834 (intmax_t)*ap->a_offset, (intmax_t)tlen, 3835 (uintmax_t)nmp->nm_maxfilesize); 3836 if ((uint64_t)*ap->a_offset >= nmp->nm_maxfilesize) { 3837 /* Avoid EFBIG error return from the NFSv4.2 server. */ 3838 *ap->a_len = 0; 3839 return (0); 3840 } 3841 clipped = false; 3842 if ((uint64_t)*ap->a_offset + tlen > nmp->nm_maxfilesize) 3843 tlen = nmp->nm_maxfilesize - *ap->a_offset; 3844 if ((uint64_t)*ap->a_offset < np->n_size) { 3845 /* Limit the len to nfs_maxalloclen before EOF. */ 3846 mlen = omin((off_t)np->n_size - *ap->a_offset, tlen); 3847 if ((uint64_t)mlen > nfs_maxalloclen) { 3848 NFSCL_DEBUG(4, "dealloc: tlen maxalloclen\n"); 3849 tlen = nfs_maxalloclen; 3850 clipped = true; 3851 } 3852 } 3853 if (error == 0) 3854 error = ncl_vinvalbuf(vp, V_SAVE, td, 1); 3855 if (error == 0) { 3856 vnode_pager_purge_range(vp, *ap->a_offset, 3857 *ap->a_offset + tlen); 3858 error = nfsrpc_deallocate(vp, *ap->a_offset, tlen, 3859 &nfsva, &attrflag, ap->a_cred, td, NULL); 3860 NFSCL_DEBUG(4, "dealloc: rpc=%d\n", error); 3861 } 3862 if (error == 0) { 3863 NFSCL_DEBUG(4, "dealloc: attrflag=%d na_size=%ju\n", 3864 attrflag, (uintmax_t)nfsva.na_size); 3865 nanouptime(&ts); 3866 NFSLOCKNODE(np); 3867 np->n_localmodtime = ts; 3868 NFSUNLOCKNODE(np); 3869 if (attrflag != 0) { 3870 if ((uint64_t)*ap->a_offset < nfsva.na_size) 3871 *ap->a_offset += omin((off_t) 3872 nfsva.na_size - *ap->a_offset, 3873 tlen); 3874 } 3875 if (clipped && tlen < *ap->a_len) 3876 *ap->a_len -= tlen; 3877 else 3878 *ap->a_len = 0; 3879 } else if (error == NFSERR_NOTSUPP) { 3880 mtx_lock(&nmp->nm_mtx); 3881 nmp->nm_privflag |= NFSMNTP_NODEALLOCATE; 3882 mtx_unlock(&nmp->nm_mtx); 3883 } 3884 } else { 3885 mtx_unlock(&nmp->nm_mtx); 3886 error = EIO; 3887 } 3888 /* 3889 * If the NFS server cannot perform the Deallocate operation, just call 3890 * vop_stddeallocate() to perform it. 3891 */ 3892 if (error != 0 && error != NFSERR_FBIG && error != NFSERR_INVAL) { 3893 error = vop_stddeallocate(ap); 3894 NFSCL_DEBUG(4, "dealloc: stddeallocate=%d\n", error); 3895 } 3896 if (attrflag != 0) { 3897 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 3898 if (error == 0 && ret != 0) 3899 error = ret; 3900 } 3901 if (error != 0) 3902 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 3903 return (error); 3904 } 3905 3906 /* 3907 * nfs copy_file_range call 3908 */ 3909 static int 3910 nfs_copy_file_range(struct vop_copy_file_range_args *ap) 3911 { 3912 struct vnode *invp = ap->a_invp; 3913 struct vnode *outvp = ap->a_outvp; 3914 struct mount *mp; 3915 struct nfsvattr innfsva, outnfsva; 3916 struct vattr *vap; 3917 struct uio io; 3918 struct nfsmount *nmp; 3919 size_t len, len2; 3920 int error, inattrflag, outattrflag, ret, ret2; 3921 off_t inoff, outoff; 3922 bool consecutive, must_commit, tryoutcred; 3923 3924 ret = ret2 = 0; 3925 nmp = VFSTONFS(invp->v_mount); 3926 mtx_lock(&nmp->nm_mtx); 3927 /* NFSv4.2 Copy is not permitted for infile == outfile. */ 3928 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 3929 (nmp->nm_privflag & NFSMNTP_NOCOPY) != 0 || invp == outvp) { 3930 mtx_unlock(&nmp->nm_mtx); 3931 error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp, 3932 ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags, 3933 ap->a_incred, ap->a_outcred, ap->a_fsizetd); 3934 return (error); 3935 } 3936 mtx_unlock(&nmp->nm_mtx); 3937 3938 /* Lock both vnodes, avoiding risk of deadlock. */ 3939 do { 3940 mp = NULL; 3941 error = vn_start_write(outvp, &mp, V_WAIT); 3942 if (error == 0) { 3943 error = vn_lock(outvp, LK_EXCLUSIVE); 3944 if (error == 0) { 3945 error = vn_lock(invp, LK_SHARED | LK_NOWAIT); 3946 if (error == 0) 3947 break; 3948 VOP_UNLOCK(outvp); 3949 if (mp != NULL) 3950 vn_finished_write(mp); 3951 mp = NULL; 3952 error = vn_lock(invp, LK_SHARED); 3953 if (error == 0) 3954 VOP_UNLOCK(invp); 3955 } 3956 } 3957 if (mp != NULL) 3958 vn_finished_write(mp); 3959 } while (error == 0); 3960 if (error != 0) 3961 return (error); 3962 3963 /* 3964 * Do the vn_rlimit_fsize() check. Should this be above the VOP layer? 3965 */ 3966 io.uio_offset = *ap->a_outoffp; 3967 io.uio_resid = *ap->a_lenp; 3968 error = vn_rlimit_fsize(outvp, &io, ap->a_fsizetd); 3969 3970 /* 3971 * Flush the input file so that the data is up to date before 3972 * the copy. Flush writes for the output file so that they 3973 * do not overwrite the data copied to the output file by the Copy. 3974 * Set the commit argument for both flushes so that the data is on 3975 * stable storage before the Copy RPC. This is done in case the 3976 * server reboots during the Copy and needs to be redone. 3977 */ 3978 if (error == 0) 3979 error = ncl_flush(invp, MNT_WAIT, curthread, 1, 0); 3980 if (error == 0) 3981 error = ncl_flush(outvp, MNT_WAIT, curthread, 1, 0); 3982 3983 /* Do the actual NFSv4.2 RPC. */ 3984 len = *ap->a_lenp; 3985 mtx_lock(&nmp->nm_mtx); 3986 if ((nmp->nm_privflag & NFSMNTP_NOCONSECUTIVE) == 0) 3987 consecutive = true; 3988 else 3989 consecutive = false; 3990 mtx_unlock(&nmp->nm_mtx); 3991 inoff = *ap->a_inoffp; 3992 outoff = *ap->a_outoffp; 3993 tryoutcred = true; 3994 must_commit = false; 3995 if (error == 0) { 3996 vap = &VTONFS(invp)->n_vattr.na_vattr; 3997 error = VOP_GETATTR(invp, vap, ap->a_incred); 3998 if (error == 0) { 3999 /* 4000 * Clip "len" at va_size so that RFC compliant servers 4001 * will not reply NFSERR_INVAL. 4002 * Setting "len == 0" for the RPC would be preferred, 4003 * but some Linux servers do not support that. 4004 */ 4005 if (inoff >= vap->va_size) 4006 *ap->a_lenp = len = 0; 4007 else if (inoff + len > vap->va_size) 4008 *ap->a_lenp = len = vap->va_size - inoff; 4009 } else 4010 error = 0; 4011 } 4012 4013 /* 4014 * len will be set to 0 upon a successful Copy RPC. 4015 * As such, this only loops when the Copy RPC needs to be retried. 4016 */ 4017 while (len > 0 && error == 0) { 4018 inattrflag = outattrflag = 0; 4019 len2 = len; 4020 if (tryoutcred) 4021 error = nfsrpc_copy_file_range(invp, ap->a_inoffp, 4022 outvp, ap->a_outoffp, &len2, ap->a_flags, 4023 &inattrflag, &innfsva, &outattrflag, &outnfsva, 4024 ap->a_outcred, consecutive, &must_commit); 4025 else 4026 error = nfsrpc_copy_file_range(invp, ap->a_inoffp, 4027 outvp, ap->a_outoffp, &len2, ap->a_flags, 4028 &inattrflag, &innfsva, &outattrflag, &outnfsva, 4029 ap->a_incred, consecutive, &must_commit); 4030 if (inattrflag != 0) 4031 ret = nfscl_loadattrcache(&invp, &innfsva, NULL, 0, 1); 4032 if (outattrflag != 0) 4033 ret2 = nfscl_loadattrcache(&outvp, &outnfsva, NULL, 4034 1, 1); 4035 if (error == 0) { 4036 if (consecutive == false) { 4037 if (len2 == len) { 4038 mtx_lock(&nmp->nm_mtx); 4039 nmp->nm_privflag |= 4040 NFSMNTP_NOCONSECUTIVE; 4041 mtx_unlock(&nmp->nm_mtx); 4042 } else 4043 error = NFSERR_OFFLOADNOREQS; 4044 } 4045 *ap->a_lenp = len2; 4046 len = 0; 4047 if (len2 > 0 && must_commit && error == 0) 4048 error = ncl_commit(outvp, outoff, *ap->a_lenp, 4049 ap->a_outcred, curthread); 4050 if (error == 0 && ret != 0) 4051 error = ret; 4052 if (error == 0 && ret2 != 0) 4053 error = ret2; 4054 } else if (error == NFSERR_OFFLOADNOREQS && consecutive) { 4055 /* 4056 * Try consecutive == false, which is ok only if all 4057 * bytes are copied. 4058 * If only some bytes were copied when consecutive 4059 * is false, there is no way to know which bytes 4060 * still need to be written. 4061 */ 4062 consecutive = false; 4063 error = 0; 4064 } else if (error == NFSERR_ACCES && tryoutcred) { 4065 /* Try again with incred. */ 4066 tryoutcred = false; 4067 error = 0; 4068 } 4069 if (error == NFSERR_STALEWRITEVERF) { 4070 /* 4071 * Server rebooted, so do it all again. 4072 */ 4073 *ap->a_inoffp = inoff; 4074 *ap->a_outoffp = outoff; 4075 len = *ap->a_lenp; 4076 must_commit = false; 4077 error = 0; 4078 } 4079 } 4080 VOP_UNLOCK(invp); 4081 VOP_UNLOCK(outvp); 4082 if (mp != NULL) 4083 vn_finished_write(mp); 4084 if (error == NFSERR_NOTSUPP || error == NFSERR_OFFLOADNOREQS || 4085 error == NFSERR_ACCES) { 4086 /* 4087 * Unlike the NFSv4.2 Copy, vn_generic_copy_file_range() can 4088 * use a_incred for the read and a_outcred for the write, so 4089 * try this for NFSERR_ACCES failures for the Copy. 4090 * For NFSERR_NOTSUPP and NFSERR_OFFLOADNOREQS, the Copy can 4091 * never succeed, so disable it. 4092 */ 4093 if (error != NFSERR_ACCES) { 4094 /* Can never do Copy on this mount. */ 4095 mtx_lock(&nmp->nm_mtx); 4096 nmp->nm_privflag |= NFSMNTP_NOCOPY; 4097 mtx_unlock(&nmp->nm_mtx); 4098 } 4099 *ap->a_inoffp = inoff; 4100 *ap->a_outoffp = outoff; 4101 error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp, 4102 ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags, 4103 ap->a_incred, ap->a_outcred, ap->a_fsizetd); 4104 } else if (error != 0) 4105 *ap->a_lenp = 0; 4106 4107 if (error != 0) 4108 error = nfscl_maperr(curthread, error, (uid_t)0, (gid_t)0); 4109 return (error); 4110 } 4111 4112 /* 4113 * nfs ioctl call 4114 */ 4115 static int 4116 nfs_ioctl(struct vop_ioctl_args *ap) 4117 { 4118 struct vnode *vp = ap->a_vp; 4119 struct nfsvattr nfsva; 4120 struct nfsmount *nmp; 4121 int attrflag, content, error, ret; 4122 bool eof = false; /* shut up compiler. */ 4123 4124 if (vp->v_type != VREG) 4125 return (ENOTTY); 4126 nmp = VFSTONFS(vp->v_mount); 4127 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION) { 4128 error = vop_stdioctl(ap); 4129 return (error); 4130 } 4131 4132 /* Do the actual NFSv4.2 RPC. */ 4133 switch (ap->a_command) { 4134 case FIOSEEKDATA: 4135 content = NFSV4CONTENT_DATA; 4136 break; 4137 case FIOSEEKHOLE: 4138 content = NFSV4CONTENT_HOLE; 4139 break; 4140 default: 4141 return (ENOTTY); 4142 } 4143 4144 error = vn_lock(vp, LK_SHARED); 4145 if (error != 0) 4146 return (EBADF); 4147 attrflag = 0; 4148 if (*((off_t *)ap->a_data) >= VTONFS(vp)->n_size) 4149 error = ENXIO; 4150 else { 4151 /* 4152 * Flush all writes, so that the server is up to date. 4153 * Although a Commit is not required, the commit argument 4154 * is set so that, for a pNFS File/Flexible File Layout 4155 * server, the LayoutCommit will be done to ensure the file 4156 * size is up to date on the Metadata Server. 4157 */ 4158 error = ncl_flush(vp, MNT_WAIT, ap->a_td, 1, 0); 4159 if (error == 0) 4160 error = nfsrpc_seek(vp, (off_t *)ap->a_data, &eof, 4161 content, ap->a_cred, &nfsva, &attrflag); 4162 /* If at eof for FIOSEEKDATA, return ENXIO. */ 4163 if (eof && error == 0 && content == NFSV4CONTENT_DATA) 4164 error = ENXIO; 4165 } 4166 if (attrflag != 0) { 4167 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4168 if (error == 0 && ret != 0) 4169 error = ret; 4170 } 4171 NFSVOPUNLOCK(vp); 4172 4173 if (error != 0) 4174 error = ENXIO; 4175 return (error); 4176 } 4177 4178 /* 4179 * nfs getextattr call 4180 */ 4181 static int 4182 nfs_getextattr(struct vop_getextattr_args *ap) 4183 { 4184 struct vnode *vp = ap->a_vp; 4185 struct nfsmount *nmp; 4186 struct ucred *cred; 4187 struct thread *td = ap->a_td; 4188 struct nfsvattr nfsva; 4189 ssize_t len; 4190 int attrflag, error, ret; 4191 4192 nmp = VFSTONFS(vp->v_mount); 4193 mtx_lock(&nmp->nm_mtx); 4194 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 4195 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 || 4196 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) { 4197 mtx_unlock(&nmp->nm_mtx); 4198 return (EOPNOTSUPP); 4199 } 4200 mtx_unlock(&nmp->nm_mtx); 4201 4202 cred = ap->a_cred; 4203 if (cred == NULL) 4204 cred = td->td_ucred; 4205 /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */ 4206 attrflag = 0; 4207 error = nfsrpc_getextattr(vp, ap->a_name, ap->a_uio, &len, &nfsva, 4208 &attrflag, cred, td); 4209 if (attrflag != 0) { 4210 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4211 if (error == 0 && ret != 0) 4212 error = ret; 4213 } 4214 if (error == 0 && ap->a_size != NULL) 4215 *ap->a_size = len; 4216 4217 switch (error) { 4218 case NFSERR_NOTSUPP: 4219 case NFSERR_OPILLEGAL: 4220 mtx_lock(&nmp->nm_mtx); 4221 nmp->nm_privflag |= NFSMNTP_NOXATTR; 4222 mtx_unlock(&nmp->nm_mtx); 4223 error = EOPNOTSUPP; 4224 break; 4225 case NFSERR_NOXATTR: 4226 case NFSERR_XATTR2BIG: 4227 error = ENOATTR; 4228 break; 4229 default: 4230 error = nfscl_maperr(td, error, 0, 0); 4231 break; 4232 } 4233 return (error); 4234 } 4235 4236 /* 4237 * nfs setextattr call 4238 */ 4239 static int 4240 nfs_setextattr(struct vop_setextattr_args *ap) 4241 { 4242 struct vnode *vp = ap->a_vp; 4243 struct nfsmount *nmp; 4244 struct ucred *cred; 4245 struct thread *td = ap->a_td; 4246 struct nfsvattr nfsva; 4247 int attrflag, error, ret; 4248 4249 nmp = VFSTONFS(vp->v_mount); 4250 mtx_lock(&nmp->nm_mtx); 4251 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 4252 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 || 4253 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) { 4254 mtx_unlock(&nmp->nm_mtx); 4255 return (EOPNOTSUPP); 4256 } 4257 mtx_unlock(&nmp->nm_mtx); 4258 4259 if (ap->a_uio->uio_resid < 0) 4260 return (EINVAL); 4261 cred = ap->a_cred; 4262 if (cred == NULL) 4263 cred = td->td_ucred; 4264 /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */ 4265 attrflag = 0; 4266 error = nfsrpc_setextattr(vp, ap->a_name, ap->a_uio, &nfsva, 4267 &attrflag, cred, td); 4268 if (attrflag != 0) { 4269 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4270 if (error == 0 && ret != 0) 4271 error = ret; 4272 } 4273 4274 switch (error) { 4275 case NFSERR_NOTSUPP: 4276 case NFSERR_OPILLEGAL: 4277 mtx_lock(&nmp->nm_mtx); 4278 nmp->nm_privflag |= NFSMNTP_NOXATTR; 4279 mtx_unlock(&nmp->nm_mtx); 4280 error = EOPNOTSUPP; 4281 break; 4282 case NFSERR_NOXATTR: 4283 case NFSERR_XATTR2BIG: 4284 error = ENOATTR; 4285 break; 4286 default: 4287 error = nfscl_maperr(td, error, 0, 0); 4288 break; 4289 } 4290 return (error); 4291 } 4292 4293 /* 4294 * nfs listextattr call 4295 */ 4296 static int 4297 nfs_listextattr(struct vop_listextattr_args *ap) 4298 { 4299 struct vnode *vp = ap->a_vp; 4300 struct nfsmount *nmp; 4301 struct ucred *cred; 4302 struct thread *td = ap->a_td; 4303 struct nfsvattr nfsva; 4304 size_t len, len2; 4305 uint64_t cookie; 4306 int attrflag, error, ret; 4307 bool eof; 4308 4309 nmp = VFSTONFS(vp->v_mount); 4310 mtx_lock(&nmp->nm_mtx); 4311 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 4312 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 || 4313 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) { 4314 mtx_unlock(&nmp->nm_mtx); 4315 return (EOPNOTSUPP); 4316 } 4317 mtx_unlock(&nmp->nm_mtx); 4318 4319 cred = ap->a_cred; 4320 if (cred == NULL) 4321 cred = td->td_ucred; 4322 4323 /* Loop around doing List Extended Attribute RPCs. */ 4324 eof = false; 4325 cookie = 0; 4326 len2 = 0; 4327 error = 0; 4328 while (!eof && error == 0) { 4329 len = nmp->nm_rsize; 4330 attrflag = 0; 4331 error = nfsrpc_listextattr(vp, &cookie, ap->a_uio, &len, &eof, 4332 &nfsva, &attrflag, cred, td); 4333 if (attrflag != 0) { 4334 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4335 if (error == 0 && ret != 0) 4336 error = ret; 4337 } 4338 if (error == 0) { 4339 len2 += len; 4340 if (len2 > SSIZE_MAX) 4341 error = ENOATTR; 4342 } 4343 } 4344 if (error == 0 && ap->a_size != NULL) 4345 *ap->a_size = len2; 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(td, error, 0, 0); 4361 break; 4362 } 4363 return (error); 4364 } 4365 4366 /* 4367 * nfs setextattr call 4368 */ 4369 static int 4370 nfs_deleteextattr(struct vop_deleteextattr_args *ap) 4371 { 4372 struct vnode *vp = ap->a_vp; 4373 struct nfsmount *nmp; 4374 struct nfsvattr nfsva; 4375 int attrflag, error, ret; 4376 4377 nmp = VFSTONFS(vp->v_mount); 4378 mtx_lock(&nmp->nm_mtx); 4379 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || 4380 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 || 4381 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) { 4382 mtx_unlock(&nmp->nm_mtx); 4383 return (EOPNOTSUPP); 4384 } 4385 mtx_unlock(&nmp->nm_mtx); 4386 4387 /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */ 4388 attrflag = 0; 4389 error = nfsrpc_rmextattr(vp, ap->a_name, &nfsva, &attrflag, ap->a_cred, 4390 ap->a_td); 4391 if (attrflag != 0) { 4392 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4393 if (error == 0 && ret != 0) 4394 error = ret; 4395 } 4396 4397 switch (error) { 4398 case NFSERR_NOTSUPP: 4399 case NFSERR_OPILLEGAL: 4400 mtx_lock(&nmp->nm_mtx); 4401 nmp->nm_privflag |= NFSMNTP_NOXATTR; 4402 mtx_unlock(&nmp->nm_mtx); 4403 error = EOPNOTSUPP; 4404 break; 4405 case NFSERR_NOXATTR: 4406 case NFSERR_XATTR2BIG: 4407 error = ENOATTR; 4408 break; 4409 default: 4410 error = nfscl_maperr(ap->a_td, error, 0, 0); 4411 break; 4412 } 4413 return (error); 4414 } 4415 4416 /* 4417 * Return POSIX pathconf information applicable to nfs filesystems. 4418 */ 4419 static int 4420 nfs_pathconf(struct vop_pathconf_args *ap) 4421 { 4422 struct nfsv3_pathconf pc; 4423 struct nfsvattr nfsva; 4424 struct vnode *vp = ap->a_vp; 4425 struct nfsmount *nmp; 4426 struct thread *td = curthread; 4427 off_t off; 4428 bool eof; 4429 int attrflag, error; 4430 4431 if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX || 4432 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED || 4433 ap->a_name == _PC_NO_TRUNC)) || 4434 (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) { 4435 /* 4436 * Since only the above 4 a_names are returned by the NFSv3 4437 * Pathconf RPC, there is no point in doing it for others. 4438 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can 4439 * be used for _PC_NFS4_ACL as well. 4440 */ 4441 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva, 4442 &attrflag, NULL); 4443 if (attrflag != 0) 4444 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1); 4445 if (error != 0) 4446 return (error); 4447 } else { 4448 /* 4449 * For NFSv2 (or NFSv3 when not one of the above 4 a_names), 4450 * just fake them. 4451 */ 4452 pc.pc_linkmax = NFS_LINK_MAX; 4453 pc.pc_namemax = NFS_MAXNAMLEN; 4454 pc.pc_notrunc = 1; 4455 pc.pc_chownrestricted = 1; 4456 pc.pc_caseinsensitive = 0; 4457 pc.pc_casepreserving = 1; 4458 error = 0; 4459 } 4460 switch (ap->a_name) { 4461 case _PC_LINK_MAX: 4462 #ifdef _LP64 4463 *ap->a_retval = pc.pc_linkmax; 4464 #else 4465 *ap->a_retval = MIN(LONG_MAX, pc.pc_linkmax); 4466 #endif 4467 break; 4468 case _PC_NAME_MAX: 4469 *ap->a_retval = pc.pc_namemax; 4470 break; 4471 case _PC_PIPE_BUF: 4472 if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO) 4473 *ap->a_retval = PIPE_BUF; 4474 else 4475 error = EINVAL; 4476 break; 4477 case _PC_CHOWN_RESTRICTED: 4478 *ap->a_retval = pc.pc_chownrestricted; 4479 break; 4480 case _PC_NO_TRUNC: 4481 *ap->a_retval = pc.pc_notrunc; 4482 break; 4483 case _PC_ACL_NFS4: 4484 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 && 4485 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL)) 4486 *ap->a_retval = 1; 4487 else 4488 *ap->a_retval = 0; 4489 break; 4490 case _PC_ACL_PATH_MAX: 4491 if (NFS_ISV4(vp)) 4492 *ap->a_retval = ACL_MAX_ENTRIES; 4493 else 4494 *ap->a_retval = 3; 4495 break; 4496 case _PC_PRIO_IO: 4497 *ap->a_retval = 0; 4498 break; 4499 case _PC_SYNC_IO: 4500 *ap->a_retval = 0; 4501 break; 4502 case _PC_ALLOC_SIZE_MIN: 4503 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize; 4504 break; 4505 case _PC_FILESIZEBITS: 4506 if (NFS_ISV34(vp)) 4507 *ap->a_retval = 64; 4508 else 4509 *ap->a_retval = 32; 4510 break; 4511 case _PC_REC_INCR_XFER_SIZE: 4512 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 4513 break; 4514 case _PC_REC_MAX_XFER_SIZE: 4515 *ap->a_retval = -1; /* means ``unlimited'' */ 4516 break; 4517 case _PC_REC_MIN_XFER_SIZE: 4518 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 4519 break; 4520 case _PC_REC_XFER_ALIGN: 4521 *ap->a_retval = PAGE_SIZE; 4522 break; 4523 case _PC_SYMLINK_MAX: 4524 *ap->a_retval = NFS_MAXPATHLEN; 4525 break; 4526 case _PC_MIN_HOLE_SIZE: 4527 /* Only some NFSv4.2 servers support Seek for Holes. */ 4528 *ap->a_retval = 0; 4529 nmp = VFSTONFS(vp->v_mount); 4530 if (NFS_ISV4(vp) && nmp->nm_minorvers == NFSV42_MINORVERSION) { 4531 /* 4532 * NFSv4.2 doesn't have an attribute for hole size, 4533 * so all we can do is see if the Seek operation is 4534 * supported and then use f_iosize as a "best guess". 4535 */ 4536 mtx_lock(&nmp->nm_mtx); 4537 if ((nmp->nm_privflag & NFSMNTP_SEEKTESTED) == 0) { 4538 mtx_unlock(&nmp->nm_mtx); 4539 off = 0; 4540 attrflag = 0; 4541 error = nfsrpc_seek(vp, &off, &eof, 4542 NFSV4CONTENT_HOLE, td->td_ucred, &nfsva, 4543 &attrflag); 4544 if (attrflag != 0) 4545 nfscl_loadattrcache(&vp, &nfsva, 4546 NULL, 0, 1); 4547 mtx_lock(&nmp->nm_mtx); 4548 if (error == NFSERR_NOTSUPP) 4549 nmp->nm_privflag |= NFSMNTP_SEEKTESTED; 4550 else 4551 nmp->nm_privflag |= NFSMNTP_SEEKTESTED | 4552 NFSMNTP_SEEK; 4553 error = 0; 4554 } 4555 if ((nmp->nm_privflag & NFSMNTP_SEEK) != 0) 4556 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 4557 mtx_unlock(&nmp->nm_mtx); 4558 } 4559 break; 4560 4561 default: 4562 error = vop_stdpathconf(ap); 4563 break; 4564 } 4565 return (error); 4566 } 4567