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/jail.h> 55 #include <sys/malloc.h> 56 #include <sys/mbuf.h> 57 #include <sys/namei.h> 58 #include <sys/socket.h> 59 #include <sys/vnode.h> 60 #include <sys/dirent.h> 61 #include <sys/fcntl.h> 62 #include <sys/lockf.h> 63 #include <sys/stat.h> 64 #include <sys/sysctl.h> 65 #include <sys/signalvar.h> 66 67 #include <vm/vm.h> 68 #include <vm/vm_extern.h> 69 #include <vm/vm_object.h> 70 71 #include <fs/nfs/nfsport.h> 72 #include <fs/nfsclient/nfsnode.h> 73 #include <fs/nfsclient/nfsmount.h> 74 #include <fs/nfsclient/nfs.h> 75 #include <fs/nfsclient/nfs_kdtrace.h> 76 77 #include <net/if.h> 78 #include <netinet/in.h> 79 #include <netinet/in_var.h> 80 81 #include <nfs/nfs_lock.h> 82 83 #ifdef KDTRACE_HOOKS 84 #include <sys/dtrace_bsd.h> 85 86 dtrace_nfsclient_accesscache_flush_probe_func_t 87 dtrace_nfscl_accesscache_flush_done_probe; 88 uint32_t nfscl_accesscache_flush_done_id; 89 90 dtrace_nfsclient_accesscache_get_probe_func_t 91 dtrace_nfscl_accesscache_get_hit_probe, 92 dtrace_nfscl_accesscache_get_miss_probe; 93 uint32_t nfscl_accesscache_get_hit_id; 94 uint32_t nfscl_accesscache_get_miss_id; 95 96 dtrace_nfsclient_accesscache_load_probe_func_t 97 dtrace_nfscl_accesscache_load_done_probe; 98 uint32_t nfscl_accesscache_load_done_id; 99 #endif /* !KDTRACE_HOOKS */ 100 101 /* Defs */ 102 #define TRUE 1 103 #define FALSE 0 104 105 extern struct nfsstatsv1 nfsstatsv1; 106 extern int nfsrv_useacl; 107 extern int nfscl_debuglevel; 108 MALLOC_DECLARE(M_NEWNFSREQ); 109 110 static vop_read_t nfsfifo_read; 111 static vop_write_t nfsfifo_write; 112 static vop_close_t nfsfifo_close; 113 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *, 114 struct thread *); 115 static vop_lookup_t nfs_lookup; 116 static vop_create_t nfs_create; 117 static vop_mknod_t nfs_mknod; 118 static vop_open_t nfs_open; 119 static vop_pathconf_t nfs_pathconf; 120 static vop_close_t nfs_close; 121 static vop_access_t nfs_access; 122 static vop_getattr_t nfs_getattr; 123 static vop_setattr_t nfs_setattr; 124 static vop_read_t nfs_read; 125 static vop_fsync_t nfs_fsync; 126 static vop_remove_t nfs_remove; 127 static vop_link_t nfs_link; 128 static vop_rename_t nfs_rename; 129 static vop_mkdir_t nfs_mkdir; 130 static vop_rmdir_t nfs_rmdir; 131 static vop_symlink_t nfs_symlink; 132 static vop_readdir_t nfs_readdir; 133 static vop_strategy_t nfs_strategy; 134 static int nfs_lookitup(struct vnode *, char *, int, 135 struct ucred *, struct thread *, struct nfsnode **); 136 static int nfs_sillyrename(struct vnode *, struct vnode *, 137 struct componentname *); 138 static vop_access_t nfsspec_access; 139 static vop_readlink_t nfs_readlink; 140 static vop_print_t nfs_print; 141 static vop_advlock_t nfs_advlock; 142 static vop_advlockasync_t nfs_advlockasync; 143 static vop_getacl_t nfs_getacl; 144 static vop_setacl_t nfs_setacl; 145 static vop_set_text_t nfs_set_text; 146 147 /* 148 * Global vfs data structures for nfs 149 */ 150 struct vop_vector newnfs_vnodeops = { 151 .vop_default = &default_vnodeops, 152 .vop_access = nfs_access, 153 .vop_advlock = nfs_advlock, 154 .vop_advlockasync = nfs_advlockasync, 155 .vop_close = nfs_close, 156 .vop_create = nfs_create, 157 .vop_fsync = nfs_fsync, 158 .vop_getattr = nfs_getattr, 159 .vop_getpages = ncl_getpages, 160 .vop_putpages = ncl_putpages, 161 .vop_inactive = ncl_inactive, 162 .vop_link = nfs_link, 163 .vop_lookup = nfs_lookup, 164 .vop_mkdir = nfs_mkdir, 165 .vop_mknod = nfs_mknod, 166 .vop_open = nfs_open, 167 .vop_pathconf = nfs_pathconf, 168 .vop_print = nfs_print, 169 .vop_read = nfs_read, 170 .vop_readdir = nfs_readdir, 171 .vop_readlink = nfs_readlink, 172 .vop_reclaim = ncl_reclaim, 173 .vop_remove = nfs_remove, 174 .vop_rename = nfs_rename, 175 .vop_rmdir = nfs_rmdir, 176 .vop_setattr = nfs_setattr, 177 .vop_strategy = nfs_strategy, 178 .vop_symlink = nfs_symlink, 179 .vop_write = ncl_write, 180 .vop_getacl = nfs_getacl, 181 .vop_setacl = nfs_setacl, 182 .vop_set_text = nfs_set_text, 183 }; 184 185 struct vop_vector newnfs_fifoops = { 186 .vop_default = &fifo_specops, 187 .vop_access = nfsspec_access, 188 .vop_close = nfsfifo_close, 189 .vop_fsync = nfs_fsync, 190 .vop_getattr = nfs_getattr, 191 .vop_inactive = ncl_inactive, 192 .vop_pathconf = nfs_pathconf, 193 .vop_print = nfs_print, 194 .vop_read = nfsfifo_read, 195 .vop_reclaim = ncl_reclaim, 196 .vop_setattr = nfs_setattr, 197 .vop_write = nfsfifo_write, 198 }; 199 200 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, 201 struct componentname *cnp, struct vattr *vap); 202 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 203 int namelen, struct ucred *cred, struct thread *td); 204 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, 205 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp, 206 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td); 207 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp, 208 struct componentname *scnp, struct sillyrename *sp); 209 210 /* 211 * Global variables 212 */ 213 SYSCTL_DECL(_vfs_nfs); 214 215 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO; 216 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 217 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 218 219 static int nfs_prime_access_cache = 0; 220 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW, 221 &nfs_prime_access_cache, 0, 222 "Prime NFS ACCESS cache when fetching attributes"); 223 224 static int newnfs_commit_on_close = 0; 225 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW, 226 &newnfs_commit_on_close, 0, "write+commit on close, else only write"); 227 228 static int nfs_clean_pages_on_close = 1; 229 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW, 230 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close"); 231 232 int newnfs_directio_enable = 0; 233 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW, 234 &newnfs_directio_enable, 0, "Enable NFS directio"); 235 236 int nfs_keep_dirty_on_error; 237 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW, 238 &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned"); 239 240 /* 241 * This sysctl allows other processes to mmap a file that has been opened 242 * O_DIRECT by a process. In general, having processes mmap the file while 243 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow 244 * this by default to prevent DoS attacks - to prevent a malicious user from 245 * opening up files O_DIRECT preventing other users from mmap'ing these 246 * files. "Protected" environments where stricter consistency guarantees are 247 * required can disable this knob. The process that opened the file O_DIRECT 248 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not 249 * meaningful. 250 */ 251 int newnfs_directio_allow_mmap = 1; 252 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW, 253 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens"); 254 255 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \ 256 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \ 257 | NFSACCESS_DELETE | NFSACCESS_LOOKUP) 258 259 /* 260 * SMP Locking Note : 261 * The list of locks after the description of the lock is the ordering 262 * of other locks acquired with the lock held. 263 * np->n_mtx : Protects the fields in the nfsnode. 264 VM Object Lock 265 VI_MTX (acquired indirectly) 266 * nmp->nm_mtx : Protects the fields in the nfsmount. 267 rep->r_mtx 268 * ncl_iod_mutex : Global lock, protects shared nfsiod state. 269 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list. 270 nmp->nm_mtx 271 rep->r_mtx 272 * rep->r_mtx : Protects the fields in an nfsreq. 273 */ 274 275 static int 276 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td, 277 struct ucred *cred, u_int32_t *retmode) 278 { 279 int error = 0, attrflag, i, lrupos; 280 u_int32_t rmode; 281 struct nfsnode *np = VTONFS(vp); 282 struct nfsvattr nfsva; 283 284 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag, 285 &rmode, NULL); 286 if (attrflag) 287 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 288 if (!error) { 289 lrupos = 0; 290 mtx_lock(&np->n_mtx); 291 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 292 if (np->n_accesscache[i].uid == cred->cr_uid) { 293 np->n_accesscache[i].mode = rmode; 294 np->n_accesscache[i].stamp = time_second; 295 break; 296 } 297 if (i > 0 && np->n_accesscache[i].stamp < 298 np->n_accesscache[lrupos].stamp) 299 lrupos = i; 300 } 301 if (i == NFS_ACCESSCACHESIZE) { 302 np->n_accesscache[lrupos].uid = cred->cr_uid; 303 np->n_accesscache[lrupos].mode = rmode; 304 np->n_accesscache[lrupos].stamp = time_second; 305 } 306 mtx_unlock(&np->n_mtx); 307 if (retmode != NULL) 308 *retmode = rmode; 309 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0); 310 } else if (NFS_ISV4(vp)) { 311 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 312 } 313 #ifdef KDTRACE_HOOKS 314 if (error != 0) 315 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0, 316 error); 317 #endif 318 return (error); 319 } 320 321 /* 322 * nfs access vnode op. 323 * For nfs version 2, just return ok. File accesses may fail later. 324 * For nfs version 3, use the access rpc to check accessibility. If file modes 325 * are changed on the server, accesses might still fail later. 326 */ 327 static int 328 nfs_access(struct vop_access_args *ap) 329 { 330 struct vnode *vp = ap->a_vp; 331 int error = 0, i, gotahit; 332 u_int32_t mode, wmode, rmode; 333 int v34 = NFS_ISV34(vp); 334 struct nfsnode *np = VTONFS(vp); 335 336 /* 337 * Disallow write attempts on filesystems mounted read-only; 338 * unless the file is a socket, fifo, or a block or character 339 * device resident on the filesystem. 340 */ 341 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS | 342 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL | 343 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) { 344 switch (vp->v_type) { 345 case VREG: 346 case VDIR: 347 case VLNK: 348 return (EROFS); 349 default: 350 break; 351 } 352 } 353 /* 354 * For nfs v3 or v4, check to see if we have done this recently, and if 355 * so return our cached result instead of making an ACCESS call. 356 * If not, do an access rpc, otherwise you are stuck emulating 357 * ufs_access() locally using the vattr. This may not be correct, 358 * since the server may apply other access criteria such as 359 * client uid-->server uid mapping that we do not know about. 360 */ 361 if (v34) { 362 if (ap->a_accmode & VREAD) 363 mode = NFSACCESS_READ; 364 else 365 mode = 0; 366 if (vp->v_type != VDIR) { 367 if (ap->a_accmode & VWRITE) 368 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 369 if (ap->a_accmode & VAPPEND) 370 mode |= NFSACCESS_EXTEND; 371 if (ap->a_accmode & VEXEC) 372 mode |= NFSACCESS_EXECUTE; 373 if (ap->a_accmode & VDELETE) 374 mode |= NFSACCESS_DELETE; 375 } else { 376 if (ap->a_accmode & VWRITE) 377 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 378 if (ap->a_accmode & VAPPEND) 379 mode |= NFSACCESS_EXTEND; 380 if (ap->a_accmode & VEXEC) 381 mode |= NFSACCESS_LOOKUP; 382 if (ap->a_accmode & VDELETE) 383 mode |= NFSACCESS_DELETE; 384 if (ap->a_accmode & VDELETE_CHILD) 385 mode |= NFSACCESS_MODIFY; 386 } 387 /* XXX safety belt, only make blanket request if caching */ 388 if (nfsaccess_cache_timeout > 0) { 389 wmode = NFSACCESS_READ | NFSACCESS_MODIFY | 390 NFSACCESS_EXTEND | NFSACCESS_EXECUTE | 391 NFSACCESS_DELETE | NFSACCESS_LOOKUP; 392 } else { 393 wmode = mode; 394 } 395 396 /* 397 * Does our cached result allow us to give a definite yes to 398 * this request? 399 */ 400 gotahit = 0; 401 mtx_lock(&np->n_mtx); 402 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 403 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) { 404 if (time_second < (np->n_accesscache[i].stamp 405 + nfsaccess_cache_timeout) && 406 (np->n_accesscache[i].mode & mode) == mode) { 407 NFSINCRGLOBAL(nfsstatsv1.accesscache_hits); 408 gotahit = 1; 409 } 410 break; 411 } 412 } 413 mtx_unlock(&np->n_mtx); 414 #ifdef KDTRACE_HOOKS 415 if (gotahit != 0) 416 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp, 417 ap->a_cred->cr_uid, mode); 418 else 419 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp, 420 ap->a_cred->cr_uid, mode); 421 #endif 422 if (gotahit == 0) { 423 /* 424 * Either a no, or a don't know. Go to the wire. 425 */ 426 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses); 427 error = nfs34_access_otw(vp, wmode, ap->a_td, 428 ap->a_cred, &rmode); 429 if (!error && 430 (rmode & mode) != mode) 431 error = EACCES; 432 } 433 return (error); 434 } else { 435 if ((error = nfsspec_access(ap)) != 0) { 436 return (error); 437 } 438 /* 439 * Attempt to prevent a mapped root from accessing a file 440 * which it shouldn't. We try to read a byte from the file 441 * if the user is root and the file is not zero length. 442 * After calling nfsspec_access, we should have the correct 443 * file size cached. 444 */ 445 mtx_lock(&np->n_mtx); 446 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD) 447 && VTONFS(vp)->n_size > 0) { 448 struct iovec aiov; 449 struct uio auio; 450 char buf[1]; 451 452 mtx_unlock(&np->n_mtx); 453 aiov.iov_base = buf; 454 aiov.iov_len = 1; 455 auio.uio_iov = &aiov; 456 auio.uio_iovcnt = 1; 457 auio.uio_offset = 0; 458 auio.uio_resid = 1; 459 auio.uio_segflg = UIO_SYSSPACE; 460 auio.uio_rw = UIO_READ; 461 auio.uio_td = ap->a_td; 462 463 if (vp->v_type == VREG) 464 error = ncl_readrpc(vp, &auio, ap->a_cred); 465 else if (vp->v_type == VDIR) { 466 char* bp; 467 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 468 aiov.iov_base = bp; 469 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 470 error = ncl_readdirrpc(vp, &auio, ap->a_cred, 471 ap->a_td); 472 free(bp, M_TEMP); 473 } else if (vp->v_type == VLNK) 474 error = ncl_readlinkrpc(vp, &auio, ap->a_cred); 475 else 476 error = EACCES; 477 } else 478 mtx_unlock(&np->n_mtx); 479 return (error); 480 } 481 } 482 483 484 /* 485 * nfs open vnode op 486 * Check to see if the type is ok 487 * and that deletion is not in progress. 488 * For paged in text files, you will need to flush the page cache 489 * if consistency is lost. 490 */ 491 /* ARGSUSED */ 492 static int 493 nfs_open(struct vop_open_args *ap) 494 { 495 struct vnode *vp = ap->a_vp; 496 struct nfsnode *np = VTONFS(vp); 497 struct vattr vattr; 498 int error; 499 int fmode = ap->a_mode; 500 struct ucred *cred; 501 502 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) 503 return (EOPNOTSUPP); 504 505 /* 506 * For NFSv4, we need to do the Open Op before cache validation, 507 * so that we conform to RFC3530 Sec. 9.3.1. 508 */ 509 if (NFS_ISV4(vp)) { 510 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td); 511 if (error) { 512 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 513 (gid_t)0); 514 return (error); 515 } 516 } 517 518 /* 519 * Now, if this Open will be doing reading, re-validate/flush the 520 * cache, so that Close/Open coherency is maintained. 521 */ 522 mtx_lock(&np->n_mtx); 523 if (np->n_flag & NMODIFIED) { 524 mtx_unlock(&np->n_mtx); 525 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 526 if (error == EINTR || error == EIO) { 527 if (NFS_ISV4(vp)) 528 (void) nfsrpc_close(vp, 0, ap->a_td); 529 return (error); 530 } 531 mtx_lock(&np->n_mtx); 532 np->n_attrstamp = 0; 533 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 534 if (vp->v_type == VDIR) 535 np->n_direofoffset = 0; 536 mtx_unlock(&np->n_mtx); 537 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 538 if (error) { 539 if (NFS_ISV4(vp)) 540 (void) nfsrpc_close(vp, 0, ap->a_td); 541 return (error); 542 } 543 mtx_lock(&np->n_mtx); 544 np->n_mtime = vattr.va_mtime; 545 if (NFS_ISV4(vp)) 546 np->n_change = vattr.va_filerev; 547 } else { 548 mtx_unlock(&np->n_mtx); 549 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 550 if (error) { 551 if (NFS_ISV4(vp)) 552 (void) nfsrpc_close(vp, 0, ap->a_td); 553 return (error); 554 } 555 mtx_lock(&np->n_mtx); 556 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) || 557 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 558 if (vp->v_type == VDIR) 559 np->n_direofoffset = 0; 560 mtx_unlock(&np->n_mtx); 561 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 562 if (error == EINTR || error == EIO) { 563 if (NFS_ISV4(vp)) 564 (void) nfsrpc_close(vp, 0, ap->a_td); 565 return (error); 566 } 567 mtx_lock(&np->n_mtx); 568 np->n_mtime = vattr.va_mtime; 569 if (NFS_ISV4(vp)) 570 np->n_change = vattr.va_filerev; 571 } 572 } 573 574 /* 575 * If the object has >= 1 O_DIRECT active opens, we disable caching. 576 */ 577 if (newnfs_directio_enable && (fmode & O_DIRECT) && 578 (vp->v_type == VREG)) { 579 if (np->n_directio_opens == 0) { 580 mtx_unlock(&np->n_mtx); 581 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 582 if (error) { 583 if (NFS_ISV4(vp)) 584 (void) nfsrpc_close(vp, 0, ap->a_td); 585 return (error); 586 } 587 mtx_lock(&np->n_mtx); 588 np->n_flag |= NNONCACHE; 589 } 590 np->n_directio_opens++; 591 } 592 593 /* If opened for writing via NFSv4.1 or later, mark that for pNFS. */ 594 if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0) 595 np->n_flag |= NWRITEOPENED; 596 597 /* 598 * If this is an open for writing, capture a reference to the 599 * credentials, so they can be used by ncl_putpages(). Using 600 * these write credentials is preferable to the credentials of 601 * whatever thread happens to be doing the VOP_PUTPAGES() since 602 * the write RPCs are less likely to fail with EACCES. 603 */ 604 if ((fmode & FWRITE) != 0) { 605 cred = np->n_writecred; 606 np->n_writecred = crhold(ap->a_cred); 607 } else 608 cred = NULL; 609 mtx_unlock(&np->n_mtx); 610 611 if (cred != NULL) 612 crfree(cred); 613 vnode_create_vobject(vp, vattr.va_size, ap->a_td); 614 return (0); 615 } 616 617 /* 618 * nfs close vnode op 619 * What an NFS client should do upon close after writing is a debatable issue. 620 * Most NFS clients push delayed writes to the server upon close, basically for 621 * two reasons: 622 * 1 - So that any write errors may be reported back to the client process 623 * doing the close system call. By far the two most likely errors are 624 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 625 * 2 - To put a worst case upper bound on cache inconsistency between 626 * multiple clients for the file. 627 * There is also a consistency problem for Version 2 of the protocol w.r.t. 628 * not being able to tell if other clients are writing a file concurrently, 629 * since there is no way of knowing if the changed modify time in the reply 630 * is only due to the write for this client. 631 * (NFS Version 3 provides weak cache consistency data in the reply that 632 * should be sufficient to detect and handle this case.) 633 * 634 * The current code does the following: 635 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 636 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 637 * or commit them (this satisfies 1 and 2 except for the 638 * case where the server crashes after this close but 639 * before the commit RPC, which is felt to be "good 640 * enough". Changing the last argument to ncl_flush() to 641 * a 1 would force a commit operation, if it is felt a 642 * commit is necessary now. 643 * for NFS Version 4 - flush the dirty buffers and commit them, if 644 * nfscl_mustflush() says this is necessary. 645 * It is necessary if there is no write delegation held, 646 * in order to satisfy open/close coherency. 647 * If the file isn't cached on local stable storage, 648 * it may be necessary in order to detect "out of space" 649 * errors from the server, if the write delegation 650 * issued by the server doesn't allow the file to grow. 651 */ 652 /* ARGSUSED */ 653 static int 654 nfs_close(struct vop_close_args *ap) 655 { 656 struct vnode *vp = ap->a_vp; 657 struct nfsnode *np = VTONFS(vp); 658 struct nfsvattr nfsva; 659 struct ucred *cred; 660 int error = 0, ret, localcred = 0; 661 int fmode = ap->a_fflag; 662 663 if (NFSCL_FORCEDISM(vp->v_mount)) 664 return (0); 665 /* 666 * During shutdown, a_cred isn't valid, so just use root. 667 */ 668 if (ap->a_cred == NOCRED) { 669 cred = newnfs_getcred(); 670 localcred = 1; 671 } else { 672 cred = ap->a_cred; 673 } 674 if (vp->v_type == VREG) { 675 /* 676 * Examine and clean dirty pages, regardless of NMODIFIED. 677 * This closes a major hole in close-to-open consistency. 678 * We want to push out all dirty pages (and buffers) on 679 * close, regardless of whether they were dirtied by 680 * mmap'ed writes or via write(). 681 */ 682 if (nfs_clean_pages_on_close && vp->v_object) { 683 VM_OBJECT_WLOCK(vp->v_object); 684 vm_object_page_clean(vp->v_object, 0, 0, 0); 685 VM_OBJECT_WUNLOCK(vp->v_object); 686 } 687 mtx_lock(&np->n_mtx); 688 if (np->n_flag & NMODIFIED) { 689 mtx_unlock(&np->n_mtx); 690 if (NFS_ISV3(vp)) { 691 /* 692 * Under NFSv3 we have dirty buffers to dispose of. We 693 * must flush them to the NFS server. We have the option 694 * of waiting all the way through the commit rpc or just 695 * waiting for the initial write. The default is to only 696 * wait through the initial write so the data is in the 697 * server's cache, which is roughly similar to the state 698 * a standard disk subsystem leaves the file in on close(). 699 * 700 * We cannot clear the NMODIFIED bit in np->n_flag due to 701 * potential races with other processes, and certainly 702 * cannot clear it if we don't commit. 703 * These races occur when there is no longer the old 704 * traditional vnode locking implemented for Vnode Ops. 705 */ 706 int cm = newnfs_commit_on_close ? 1 : 0; 707 error = ncl_flush(vp, MNT_WAIT, ap->a_td, cm, 0); 708 /* np->n_flag &= ~NMODIFIED; */ 709 } else if (NFS_ISV4(vp)) { 710 if (nfscl_mustflush(vp) != 0) { 711 int cm = newnfs_commit_on_close ? 1 : 0; 712 error = ncl_flush(vp, MNT_WAIT, ap->a_td, 713 cm, 0); 714 /* 715 * as above w.r.t races when clearing 716 * NMODIFIED. 717 * np->n_flag &= ~NMODIFIED; 718 */ 719 } 720 } else { 721 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 722 } 723 mtx_lock(&np->n_mtx); 724 } 725 /* 726 * Invalidate the attribute cache in all cases. 727 * An open is going to fetch fresh attrs any way, other procs 728 * on this node that have file open will be forced to do an 729 * otw attr fetch, but this is safe. 730 * --> A user found that their RPC count dropped by 20% when 731 * this was commented out and I can't see any requirement 732 * for it, so I've disabled it when negative lookups are 733 * enabled. (What does this have to do with negative lookup 734 * caching? Well nothing, except it was reported by the 735 * same user that needed negative lookup caching and I wanted 736 * there to be a way to disable it to see if it 737 * is the cause of some caching/coherency issue that might 738 * crop up.) 739 */ 740 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) { 741 np->n_attrstamp = 0; 742 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 743 } 744 if (np->n_flag & NWRITEERR) { 745 np->n_flag &= ~NWRITEERR; 746 error = np->n_error; 747 } 748 mtx_unlock(&np->n_mtx); 749 } 750 751 if (NFS_ISV4(vp)) { 752 /* 753 * Get attributes so "change" is up to date. 754 */ 755 if (error == 0 && nfscl_mustflush(vp) != 0 && 756 vp->v_type == VREG && 757 (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) { 758 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva, 759 NULL); 760 if (!ret) { 761 np->n_change = nfsva.na_filerev; 762 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 763 NULL, 0, 0); 764 } 765 } 766 767 /* 768 * and do the close. 769 */ 770 ret = nfsrpc_close(vp, 0, ap->a_td); 771 if (!error && ret) 772 error = ret; 773 if (error) 774 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 775 (gid_t)0); 776 } 777 if (newnfs_directio_enable) 778 KASSERT((np->n_directio_asyncwr == 0), 779 ("nfs_close: dirty unflushed (%d) directio buffers\n", 780 np->n_directio_asyncwr)); 781 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) { 782 mtx_lock(&np->n_mtx); 783 KASSERT((np->n_directio_opens > 0), 784 ("nfs_close: unexpectedly value (0) of n_directio_opens\n")); 785 np->n_directio_opens--; 786 if (np->n_directio_opens == 0) 787 np->n_flag &= ~NNONCACHE; 788 mtx_unlock(&np->n_mtx); 789 } 790 if (localcred) 791 NFSFREECRED(cred); 792 return (error); 793 } 794 795 /* 796 * nfs getattr call from vfs. 797 */ 798 static int 799 nfs_getattr(struct vop_getattr_args *ap) 800 { 801 struct vnode *vp = ap->a_vp; 802 struct thread *td = curthread; /* XXX */ 803 struct nfsnode *np = VTONFS(vp); 804 int error = 0; 805 struct nfsvattr nfsva; 806 struct vattr *vap = ap->a_vap; 807 struct vattr vattr; 808 809 /* 810 * Update local times for special files. 811 */ 812 mtx_lock(&np->n_mtx); 813 if (np->n_flag & (NACC | NUPD)) 814 np->n_flag |= NCHG; 815 mtx_unlock(&np->n_mtx); 816 /* 817 * First look in the cache. 818 */ 819 if (ncl_getattrcache(vp, &vattr) == 0) { 820 vap->va_type = vattr.va_type; 821 vap->va_mode = vattr.va_mode; 822 vap->va_nlink = vattr.va_nlink; 823 vap->va_uid = vattr.va_uid; 824 vap->va_gid = vattr.va_gid; 825 vap->va_fsid = vattr.va_fsid; 826 vap->va_fileid = vattr.va_fileid; 827 vap->va_size = vattr.va_size; 828 vap->va_blocksize = vattr.va_blocksize; 829 vap->va_atime = vattr.va_atime; 830 vap->va_mtime = vattr.va_mtime; 831 vap->va_ctime = vattr.va_ctime; 832 vap->va_gen = vattr.va_gen; 833 vap->va_flags = vattr.va_flags; 834 vap->va_rdev = vattr.va_rdev; 835 vap->va_bytes = vattr.va_bytes; 836 vap->va_filerev = vattr.va_filerev; 837 /* 838 * Get the local modify time for the case of a write 839 * delegation. 840 */ 841 nfscl_deleggetmodtime(vp, &vap->va_mtime); 842 return (0); 843 } 844 845 if (NFS_ISV34(vp) && nfs_prime_access_cache && 846 nfsaccess_cache_timeout > 0) { 847 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses); 848 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL); 849 if (ncl_getattrcache(vp, ap->a_vap) == 0) { 850 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime); 851 return (0); 852 } 853 } 854 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL); 855 if (!error) 856 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0); 857 if (!error) { 858 /* 859 * Get the local modify time for the case of a write 860 * delegation. 861 */ 862 nfscl_deleggetmodtime(vp, &vap->va_mtime); 863 } else if (NFS_ISV4(vp)) { 864 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 865 } 866 return (error); 867 } 868 869 /* 870 * nfs setattr call. 871 */ 872 static int 873 nfs_setattr(struct vop_setattr_args *ap) 874 { 875 struct vnode *vp = ap->a_vp; 876 struct nfsnode *np = VTONFS(vp); 877 struct thread *td = curthread; /* XXX */ 878 struct vattr *vap = ap->a_vap; 879 int error = 0; 880 u_quad_t tsize; 881 882 #ifndef nolint 883 tsize = (u_quad_t)0; 884 #endif 885 886 /* 887 * Setting of flags and marking of atimes are not supported. 888 */ 889 if (vap->va_flags != VNOVAL) 890 return (EOPNOTSUPP); 891 892 /* 893 * Disallow write attempts if the filesystem is mounted read-only. 894 */ 895 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 896 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 897 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 898 (vp->v_mount->mnt_flag & MNT_RDONLY)) 899 return (EROFS); 900 if (vap->va_size != VNOVAL) { 901 switch (vp->v_type) { 902 case VDIR: 903 return (EISDIR); 904 case VCHR: 905 case VBLK: 906 case VSOCK: 907 case VFIFO: 908 if (vap->va_mtime.tv_sec == VNOVAL && 909 vap->va_atime.tv_sec == VNOVAL && 910 vap->va_mode == (mode_t)VNOVAL && 911 vap->va_uid == (uid_t)VNOVAL && 912 vap->va_gid == (gid_t)VNOVAL) 913 return (0); 914 vap->va_size = VNOVAL; 915 break; 916 default: 917 /* 918 * Disallow write attempts if the filesystem is 919 * mounted read-only. 920 */ 921 if (vp->v_mount->mnt_flag & MNT_RDONLY) 922 return (EROFS); 923 /* 924 * We run vnode_pager_setsize() early (why?), 925 * we must set np->n_size now to avoid vinvalbuf 926 * V_SAVE races that might setsize a lower 927 * value. 928 */ 929 mtx_lock(&np->n_mtx); 930 tsize = np->n_size; 931 mtx_unlock(&np->n_mtx); 932 error = ncl_meta_setsize(vp, ap->a_cred, td, 933 vap->va_size); 934 mtx_lock(&np->n_mtx); 935 if (np->n_flag & NMODIFIED) { 936 tsize = np->n_size; 937 mtx_unlock(&np->n_mtx); 938 error = ncl_vinvalbuf(vp, vap->va_size == 0 ? 939 0 : V_SAVE, td, 1); 940 if (error != 0) { 941 vnode_pager_setsize(vp, tsize); 942 return (error); 943 } 944 /* 945 * Call nfscl_delegmodtime() to set the modify time 946 * locally, as required. 947 */ 948 nfscl_delegmodtime(vp); 949 } else 950 mtx_unlock(&np->n_mtx); 951 /* 952 * np->n_size has already been set to vap->va_size 953 * in ncl_meta_setsize(). We must set it again since 954 * nfs_loadattrcache() could be called through 955 * ncl_meta_setsize() and could modify np->n_size. 956 */ 957 mtx_lock(&np->n_mtx); 958 np->n_vattr.na_size = np->n_size = vap->va_size; 959 mtx_unlock(&np->n_mtx); 960 } 961 } else { 962 mtx_lock(&np->n_mtx); 963 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) && 964 (np->n_flag & NMODIFIED) && vp->v_type == VREG) { 965 mtx_unlock(&np->n_mtx); 966 error = ncl_vinvalbuf(vp, V_SAVE, td, 1); 967 if (error == EINTR || error == EIO) 968 return (error); 969 } else 970 mtx_unlock(&np->n_mtx); 971 } 972 error = nfs_setattrrpc(vp, vap, ap->a_cred, td); 973 if (error && vap->va_size != VNOVAL) { 974 mtx_lock(&np->n_mtx); 975 np->n_size = np->n_vattr.na_size = tsize; 976 vnode_pager_setsize(vp, tsize); 977 mtx_unlock(&np->n_mtx); 978 } 979 return (error); 980 } 981 982 /* 983 * Do an nfs setattr rpc. 984 */ 985 static int 986 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred, 987 struct thread *td) 988 { 989 struct nfsnode *np = VTONFS(vp); 990 int error, ret, attrflag, i; 991 struct nfsvattr nfsva; 992 993 if (NFS_ISV34(vp)) { 994 mtx_lock(&np->n_mtx); 995 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) 996 np->n_accesscache[i].stamp = 0; 997 np->n_flag |= NDELEGMOD; 998 mtx_unlock(&np->n_mtx); 999 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp); 1000 } 1001 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag, 1002 NULL); 1003 if (attrflag) { 1004 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1005 if (ret && !error) 1006 error = ret; 1007 } 1008 if (error && NFS_ISV4(vp)) 1009 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid); 1010 return (error); 1011 } 1012 1013 /* 1014 * nfs lookup call, one step at a time... 1015 * First look in cache 1016 * If not found, unlock the directory nfsnode and do the rpc 1017 */ 1018 static int 1019 nfs_lookup(struct vop_lookup_args *ap) 1020 { 1021 struct componentname *cnp = ap->a_cnp; 1022 struct vnode *dvp = ap->a_dvp; 1023 struct vnode **vpp = ap->a_vpp; 1024 struct mount *mp = dvp->v_mount; 1025 int flags = cnp->cn_flags; 1026 struct vnode *newvp; 1027 struct nfsmount *nmp; 1028 struct nfsnode *np, *newnp; 1029 int error = 0, attrflag, dattrflag, ltype, ncticks; 1030 struct thread *td = cnp->cn_thread; 1031 struct nfsfh *nfhp; 1032 struct nfsvattr dnfsva, nfsva; 1033 struct vattr vattr; 1034 struct timespec nctime; 1035 1036 *vpp = NULLVP; 1037 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) && 1038 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 1039 return (EROFS); 1040 if (dvp->v_type != VDIR) 1041 return (ENOTDIR); 1042 nmp = VFSTONFS(mp); 1043 np = VTONFS(dvp); 1044 1045 /* For NFSv4, wait until any remove is done. */ 1046 mtx_lock(&np->n_mtx); 1047 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) { 1048 np->n_flag |= NREMOVEWANT; 1049 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0); 1050 } 1051 mtx_unlock(&np->n_mtx); 1052 1053 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) 1054 return (error); 1055 error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks); 1056 if (error > 0 && error != ENOENT) 1057 return (error); 1058 if (error == -1) { 1059 /* 1060 * Lookups of "." are special and always return the 1061 * current directory. cache_lookup() already handles 1062 * associated locking bookkeeping, etc. 1063 */ 1064 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 1065 /* XXX: Is this really correct? */ 1066 if (cnp->cn_nameiop != LOOKUP && 1067 (flags & ISLASTCN)) 1068 cnp->cn_flags |= SAVENAME; 1069 return (0); 1070 } 1071 1072 /* 1073 * We only accept a positive hit in the cache if the 1074 * change time of the file matches our cached copy. 1075 * Otherwise, we discard the cache entry and fallback 1076 * to doing a lookup RPC. We also only trust cache 1077 * entries for less than nm_nametimeo seconds. 1078 * 1079 * To better handle stale file handles and attributes, 1080 * clear the attribute cache of this node if it is a 1081 * leaf component, part of an open() call, and not 1082 * locally modified before fetching the attributes. 1083 * This should allow stale file handles to be detected 1084 * here where we can fall back to a LOOKUP RPC to 1085 * recover rather than having nfs_open() detect the 1086 * stale file handle and failing open(2) with ESTALE. 1087 */ 1088 newvp = *vpp; 1089 newnp = VTONFS(newvp); 1090 if (!(nmp->nm_flag & NFSMNT_NOCTO) && 1091 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1092 !(newnp->n_flag & NMODIFIED)) { 1093 mtx_lock(&newnp->n_mtx); 1094 newnp->n_attrstamp = 0; 1095 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1096 mtx_unlock(&newnp->n_mtx); 1097 } 1098 if (nfscl_nodeleg(newvp, 0) == 0 || 1099 ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) && 1100 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 && 1101 timespeccmp(&vattr.va_ctime, &nctime, ==))) { 1102 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits); 1103 if (cnp->cn_nameiop != LOOKUP && 1104 (flags & ISLASTCN)) 1105 cnp->cn_flags |= SAVENAME; 1106 return (0); 1107 } 1108 cache_purge(newvp); 1109 if (dvp != newvp) 1110 vput(newvp); 1111 else 1112 vrele(newvp); 1113 *vpp = NULLVP; 1114 } else if (error == ENOENT) { 1115 if (dvp->v_iflag & VI_DOOMED) 1116 return (ENOENT); 1117 /* 1118 * We only accept a negative hit in the cache if the 1119 * modification time of the parent directory matches 1120 * the cached copy in the name cache entry. 1121 * Otherwise, we discard all of the negative cache 1122 * entries for this directory. We also only trust 1123 * negative cache entries for up to nm_negnametimeo 1124 * seconds. 1125 */ 1126 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) && 1127 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 && 1128 timespeccmp(&vattr.va_mtime, &nctime, ==)) { 1129 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits); 1130 return (ENOENT); 1131 } 1132 cache_purge_negative(dvp); 1133 } 1134 1135 error = 0; 1136 newvp = NULLVP; 1137 NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses); 1138 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1139 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1140 NULL); 1141 if (dattrflag) 1142 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1143 if (error) { 1144 if (newvp != NULLVP) { 1145 vput(newvp); 1146 *vpp = NULLVP; 1147 } 1148 1149 if (error != ENOENT) { 1150 if (NFS_ISV4(dvp)) 1151 error = nfscl_maperr(td, error, (uid_t)0, 1152 (gid_t)0); 1153 return (error); 1154 } 1155 1156 /* The requested file was not found. */ 1157 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1158 (flags & ISLASTCN)) { 1159 /* 1160 * XXX: UFS does a full VOP_ACCESS(dvp, 1161 * VWRITE) here instead of just checking 1162 * MNT_RDONLY. 1163 */ 1164 if (mp->mnt_flag & MNT_RDONLY) 1165 return (EROFS); 1166 cnp->cn_flags |= SAVENAME; 1167 return (EJUSTRETURN); 1168 } 1169 1170 if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) { 1171 /* 1172 * Cache the modification time of the parent 1173 * directory from the post-op attributes in 1174 * the name cache entry. The negative cache 1175 * entry will be ignored once the directory 1176 * has changed. Don't bother adding the entry 1177 * if the directory has already changed. 1178 */ 1179 mtx_lock(&np->n_mtx); 1180 if (timespeccmp(&np->n_vattr.na_mtime, 1181 &dnfsva.na_mtime, ==)) { 1182 mtx_unlock(&np->n_mtx); 1183 cache_enter_time(dvp, NULL, cnp, 1184 &dnfsva.na_mtime, NULL); 1185 } else 1186 mtx_unlock(&np->n_mtx); 1187 } 1188 return (ENOENT); 1189 } 1190 1191 /* 1192 * Handle RENAME case... 1193 */ 1194 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 1195 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1196 free(nfhp, M_NFSFH); 1197 return (EISDIR); 1198 } 1199 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1200 LK_EXCLUSIVE); 1201 if (error) 1202 return (error); 1203 newvp = NFSTOV(np); 1204 if (attrflag) 1205 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1206 0, 1); 1207 *vpp = newvp; 1208 cnp->cn_flags |= SAVENAME; 1209 return (0); 1210 } 1211 1212 if (flags & ISDOTDOT) { 1213 ltype = NFSVOPISLOCKED(dvp); 1214 error = vfs_busy(mp, MBF_NOWAIT); 1215 if (error != 0) { 1216 vfs_ref(mp); 1217 NFSVOPUNLOCK(dvp, 0); 1218 error = vfs_busy(mp, 0); 1219 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1220 vfs_rel(mp); 1221 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) { 1222 vfs_unbusy(mp); 1223 error = ENOENT; 1224 } 1225 if (error != 0) 1226 return (error); 1227 } 1228 NFSVOPUNLOCK(dvp, 0); 1229 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1230 cnp->cn_lkflags); 1231 if (error == 0) 1232 newvp = NFSTOV(np); 1233 vfs_unbusy(mp); 1234 if (newvp != dvp) 1235 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1236 if (dvp->v_iflag & VI_DOOMED) { 1237 if (error == 0) { 1238 if (newvp == dvp) 1239 vrele(newvp); 1240 else 1241 vput(newvp); 1242 } 1243 error = ENOENT; 1244 } 1245 if (error != 0) 1246 return (error); 1247 if (attrflag) 1248 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1249 0, 1); 1250 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1251 free(nfhp, M_NFSFH); 1252 VREF(dvp); 1253 newvp = dvp; 1254 if (attrflag) 1255 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1256 0, 1); 1257 } else { 1258 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1259 cnp->cn_lkflags); 1260 if (error) 1261 return (error); 1262 newvp = NFSTOV(np); 1263 if (attrflag) 1264 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1265 0, 1); 1266 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1267 !(np->n_flag & NMODIFIED)) { 1268 /* 1269 * Flush the attribute cache when opening a 1270 * leaf node to ensure that fresh attributes 1271 * are fetched in nfs_open() since we did not 1272 * fetch attributes from the LOOKUP reply. 1273 */ 1274 mtx_lock(&np->n_mtx); 1275 np->n_attrstamp = 0; 1276 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1277 mtx_unlock(&np->n_mtx); 1278 } 1279 } 1280 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1281 cnp->cn_flags |= SAVENAME; 1282 if ((cnp->cn_flags & MAKEENTRY) && 1283 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) && 1284 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0)) 1285 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 1286 newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime); 1287 *vpp = newvp; 1288 return (0); 1289 } 1290 1291 /* 1292 * nfs read call. 1293 * Just call ncl_bioread() to do the work. 1294 */ 1295 static int 1296 nfs_read(struct vop_read_args *ap) 1297 { 1298 struct vnode *vp = ap->a_vp; 1299 1300 switch (vp->v_type) { 1301 case VREG: 1302 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred)); 1303 case VDIR: 1304 return (EISDIR); 1305 default: 1306 return (EOPNOTSUPP); 1307 } 1308 } 1309 1310 /* 1311 * nfs readlink call 1312 */ 1313 static int 1314 nfs_readlink(struct vop_readlink_args *ap) 1315 { 1316 struct vnode *vp = ap->a_vp; 1317 1318 if (vp->v_type != VLNK) 1319 return (EINVAL); 1320 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred)); 1321 } 1322 1323 /* 1324 * Do a readlink rpc. 1325 * Called by ncl_doio() from below the buffer cache. 1326 */ 1327 int 1328 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1329 { 1330 int error, ret, attrflag; 1331 struct nfsvattr nfsva; 1332 1333 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva, 1334 &attrflag, NULL); 1335 if (attrflag) { 1336 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1337 if (ret && !error) 1338 error = ret; 1339 } 1340 if (error && NFS_ISV4(vp)) 1341 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1342 return (error); 1343 } 1344 1345 /* 1346 * nfs read rpc call 1347 * Ditto above 1348 */ 1349 int 1350 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1351 { 1352 int error, ret, attrflag; 1353 struct nfsvattr nfsva; 1354 struct nfsmount *nmp; 1355 1356 nmp = VFSTONFS(vnode_mount(vp)); 1357 error = EIO; 1358 attrflag = 0; 1359 if (NFSHASPNFS(nmp)) 1360 error = nfscl_doiods(vp, uiop, NULL, NULL, 1361 NFSV4OPEN_ACCESSREAD, 0, cred, uiop->uio_td); 1362 NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error); 1363 if (error != 0) 1364 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, 1365 &attrflag, NULL); 1366 if (attrflag) { 1367 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1368 if (ret && !error) 1369 error = ret; 1370 } 1371 if (error && NFS_ISV4(vp)) 1372 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1373 return (error); 1374 } 1375 1376 /* 1377 * nfs write call 1378 */ 1379 int 1380 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 1381 int *iomode, int *must_commit, int called_from_strategy) 1382 { 1383 struct nfsvattr nfsva; 1384 int error, attrflag, ret; 1385 struct nfsmount *nmp; 1386 1387 nmp = VFSTONFS(vnode_mount(vp)); 1388 error = EIO; 1389 attrflag = 0; 1390 if (NFSHASPNFS(nmp)) 1391 error = nfscl_doiods(vp, uiop, iomode, must_commit, 1392 NFSV4OPEN_ACCESSWRITE, 0, cred, uiop->uio_td); 1393 NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error); 1394 if (error != 0) 1395 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred, 1396 uiop->uio_td, &nfsva, &attrflag, NULL, 1397 called_from_strategy); 1398 if (attrflag) { 1399 if (VTONFS(vp)->n_flag & ND_NFSV4) 1400 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1, 1401 1); 1402 else 1403 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1404 1); 1405 if (ret && !error) 1406 error = ret; 1407 } 1408 if (DOINGASYNC(vp)) 1409 *iomode = NFSWRITE_FILESYNC; 1410 if (error && NFS_ISV4(vp)) 1411 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1412 return (error); 1413 } 1414 1415 /* 1416 * nfs mknod rpc 1417 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1418 * mode set to specify the file type and the size field for rdev. 1419 */ 1420 static int 1421 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1422 struct vattr *vap) 1423 { 1424 struct nfsvattr nfsva, dnfsva; 1425 struct vnode *newvp = NULL; 1426 struct nfsnode *np = NULL, *dnp; 1427 struct nfsfh *nfhp; 1428 struct vattr vattr; 1429 int error = 0, attrflag, dattrflag; 1430 u_int32_t rdev; 1431 1432 if (vap->va_type == VCHR || vap->va_type == VBLK) 1433 rdev = vap->va_rdev; 1434 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1435 rdev = 0xffffffff; 1436 else 1437 return (EOPNOTSUPP); 1438 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1439 return (error); 1440 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap, 1441 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva, 1442 &nfsva, &nfhp, &attrflag, &dattrflag, NULL); 1443 if (!error) { 1444 if (!nfhp) 1445 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1446 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, 1447 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1448 NULL); 1449 if (nfhp) 1450 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1451 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); 1452 } 1453 if (dattrflag) 1454 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1455 if (!error) { 1456 newvp = NFSTOV(np); 1457 if (attrflag != 0) { 1458 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1459 0, 1); 1460 if (error != 0) 1461 vput(newvp); 1462 } 1463 } 1464 if (!error) { 1465 *vpp = newvp; 1466 } else if (NFS_ISV4(dvp)) { 1467 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, 1468 vap->va_gid); 1469 } 1470 dnp = VTONFS(dvp); 1471 mtx_lock(&dnp->n_mtx); 1472 dnp->n_flag |= NMODIFIED; 1473 if (!dattrflag) { 1474 dnp->n_attrstamp = 0; 1475 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1476 } 1477 mtx_unlock(&dnp->n_mtx); 1478 return (error); 1479 } 1480 1481 /* 1482 * nfs mknod vop 1483 * just call nfs_mknodrpc() to do the work. 1484 */ 1485 /* ARGSUSED */ 1486 static int 1487 nfs_mknod(struct vop_mknod_args *ap) 1488 { 1489 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap)); 1490 } 1491 1492 static struct mtx nfs_cverf_mtx; 1493 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex", 1494 MTX_DEF); 1495 1496 static nfsquad_t 1497 nfs_get_cverf(void) 1498 { 1499 static nfsquad_t cverf; 1500 nfsquad_t ret; 1501 static int cverf_initialized = 0; 1502 1503 mtx_lock(&nfs_cverf_mtx); 1504 if (cverf_initialized == 0) { 1505 cverf.lval[0] = arc4random(); 1506 cverf.lval[1] = arc4random(); 1507 cverf_initialized = 1; 1508 } else 1509 cverf.qval++; 1510 ret = cverf; 1511 mtx_unlock(&nfs_cverf_mtx); 1512 1513 return (ret); 1514 } 1515 1516 /* 1517 * nfs file create call 1518 */ 1519 static int 1520 nfs_create(struct vop_create_args *ap) 1521 { 1522 struct vnode *dvp = ap->a_dvp; 1523 struct vattr *vap = ap->a_vap; 1524 struct componentname *cnp = ap->a_cnp; 1525 struct nfsnode *np = NULL, *dnp; 1526 struct vnode *newvp = NULL; 1527 struct nfsmount *nmp; 1528 struct nfsvattr dnfsva, nfsva; 1529 struct nfsfh *nfhp; 1530 nfsquad_t cverf; 1531 int error = 0, attrflag, dattrflag, fmode = 0; 1532 struct vattr vattr; 1533 1534 /* 1535 * Oops, not for me.. 1536 */ 1537 if (vap->va_type == VSOCK) 1538 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1539 1540 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1541 return (error); 1542 if (vap->va_vaflags & VA_EXCLUSIVE) 1543 fmode |= O_EXCL; 1544 dnp = VTONFS(dvp); 1545 nmp = VFSTONFS(vnode_mount(dvp)); 1546 again: 1547 /* For NFSv4, wait until any remove is done. */ 1548 mtx_lock(&dnp->n_mtx); 1549 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) { 1550 dnp->n_flag |= NREMOVEWANT; 1551 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0); 1552 } 1553 mtx_unlock(&dnp->n_mtx); 1554 1555 cverf = nfs_get_cverf(); 1556 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1557 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, 1558 &nfhp, &attrflag, &dattrflag, NULL); 1559 if (!error) { 1560 if (nfhp == NULL) 1561 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1562 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, 1563 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1564 NULL); 1565 if (nfhp != NULL) 1566 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1567 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); 1568 } 1569 if (dattrflag) 1570 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1571 if (!error) { 1572 newvp = NFSTOV(np); 1573 if (attrflag == 0) 1574 error = nfsrpc_getattr(newvp, cnp->cn_cred, 1575 cnp->cn_thread, &nfsva, NULL); 1576 if (error == 0) 1577 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1578 0, 1); 1579 } 1580 if (error) { 1581 if (newvp != NULL) { 1582 vput(newvp); 1583 newvp = NULL; 1584 } 1585 if (NFS_ISV34(dvp) && (fmode & O_EXCL) && 1586 error == NFSERR_NOTSUPP) { 1587 fmode &= ~O_EXCL; 1588 goto again; 1589 } 1590 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) { 1591 if (nfscl_checksattr(vap, &nfsva)) { 1592 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred, 1593 cnp->cn_thread, &nfsva, &attrflag, NULL); 1594 if (error && (vap->va_uid != (uid_t)VNOVAL || 1595 vap->va_gid != (gid_t)VNOVAL)) { 1596 /* try again without setting uid/gid */ 1597 vap->va_uid = (uid_t)VNOVAL; 1598 vap->va_gid = (uid_t)VNOVAL; 1599 error = nfsrpc_setattr(newvp, vap, NULL, 1600 cnp->cn_cred, cnp->cn_thread, &nfsva, 1601 &attrflag, NULL); 1602 } 1603 if (attrflag) 1604 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 1605 NULL, 0, 1); 1606 if (error != 0) 1607 vput(newvp); 1608 } 1609 } 1610 if (!error) { 1611 if ((cnp->cn_flags & MAKEENTRY) && attrflag) 1612 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 1613 NULL); 1614 *ap->a_vpp = newvp; 1615 } else if (NFS_ISV4(dvp)) { 1616 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, 1617 vap->va_gid); 1618 } 1619 mtx_lock(&dnp->n_mtx); 1620 dnp->n_flag |= NMODIFIED; 1621 if (!dattrflag) { 1622 dnp->n_attrstamp = 0; 1623 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1624 } 1625 mtx_unlock(&dnp->n_mtx); 1626 return (error); 1627 } 1628 1629 /* 1630 * nfs file remove call 1631 * To try and make nfs semantics closer to ufs semantics, a file that has 1632 * other processes using the vnode is renamed instead of removed and then 1633 * removed later on the last close. 1634 * - If v_usecount > 1 1635 * If a rename is not already in the works 1636 * call nfs_sillyrename() to set it up 1637 * else 1638 * do the remove rpc 1639 */ 1640 static int 1641 nfs_remove(struct vop_remove_args *ap) 1642 { 1643 struct vnode *vp = ap->a_vp; 1644 struct vnode *dvp = ap->a_dvp; 1645 struct componentname *cnp = ap->a_cnp; 1646 struct nfsnode *np = VTONFS(vp); 1647 int error = 0; 1648 struct vattr vattr; 1649 1650 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name")); 1651 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount")); 1652 if (vp->v_type == VDIR) 1653 error = EPERM; 1654 else if (vrefcnt(vp) == 1 || (np->n_sillyrename && 1655 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 1656 vattr.va_nlink > 1)) { 1657 /* 1658 * Purge the name cache so that the chance of a lookup for 1659 * the name succeeding while the remove is in progress is 1660 * minimized. Without node locking it can still happen, such 1661 * that an I/O op returns ESTALE, but since you get this if 1662 * another host removes the file.. 1663 */ 1664 cache_purge(vp); 1665 /* 1666 * throw away biocache buffers, mainly to avoid 1667 * unnecessary delayed writes later. 1668 */ 1669 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1); 1670 if (error != EINTR && error != EIO) 1671 /* Do the rpc */ 1672 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr, 1673 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread); 1674 /* 1675 * Kludge City: If the first reply to the remove rpc is lost.. 1676 * the reply to the retransmitted request will be ENOENT 1677 * since the file was in fact removed 1678 * Therefore, we cheat and return success. 1679 */ 1680 if (error == ENOENT) 1681 error = 0; 1682 } else if (!np->n_sillyrename) 1683 error = nfs_sillyrename(dvp, vp, cnp); 1684 mtx_lock(&np->n_mtx); 1685 np->n_attrstamp = 0; 1686 mtx_unlock(&np->n_mtx); 1687 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 1688 return (error); 1689 } 1690 1691 /* 1692 * nfs file remove rpc called from nfs_inactive 1693 */ 1694 int 1695 ncl_removeit(struct sillyrename *sp, struct vnode *vp) 1696 { 1697 /* 1698 * Make sure that the directory vnode is still valid. 1699 * XXX we should lock sp->s_dvp here. 1700 */ 1701 if (sp->s_dvp->v_type == VBAD) 1702 return (0); 1703 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen, 1704 sp->s_cred, NULL)); 1705 } 1706 1707 /* 1708 * Nfs remove rpc, called from nfs_remove() and ncl_removeit(). 1709 */ 1710 static int 1711 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 1712 int namelen, struct ucred *cred, struct thread *td) 1713 { 1714 struct nfsvattr dnfsva; 1715 struct nfsnode *dnp = VTONFS(dvp); 1716 int error = 0, dattrflag; 1717 1718 mtx_lock(&dnp->n_mtx); 1719 dnp->n_flag |= NREMOVEINPROG; 1720 mtx_unlock(&dnp->n_mtx); 1721 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva, 1722 &dattrflag, NULL); 1723 mtx_lock(&dnp->n_mtx); 1724 if ((dnp->n_flag & NREMOVEWANT)) { 1725 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG); 1726 mtx_unlock(&dnp->n_mtx); 1727 wakeup((caddr_t)dnp); 1728 } else { 1729 dnp->n_flag &= ~NREMOVEINPROG; 1730 mtx_unlock(&dnp->n_mtx); 1731 } 1732 if (dattrflag) 1733 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1734 mtx_lock(&dnp->n_mtx); 1735 dnp->n_flag |= NMODIFIED; 1736 if (!dattrflag) { 1737 dnp->n_attrstamp = 0; 1738 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1739 } 1740 mtx_unlock(&dnp->n_mtx); 1741 if (error && NFS_ISV4(dvp)) 1742 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1743 return (error); 1744 } 1745 1746 /* 1747 * nfs file rename call 1748 */ 1749 static int 1750 nfs_rename(struct vop_rename_args *ap) 1751 { 1752 struct vnode *fvp = ap->a_fvp; 1753 struct vnode *tvp = ap->a_tvp; 1754 struct vnode *fdvp = ap->a_fdvp; 1755 struct vnode *tdvp = ap->a_tdvp; 1756 struct componentname *tcnp = ap->a_tcnp; 1757 struct componentname *fcnp = ap->a_fcnp; 1758 struct nfsnode *fnp = VTONFS(ap->a_fvp); 1759 struct nfsnode *tdnp = VTONFS(ap->a_tdvp); 1760 struct nfsv4node *newv4 = NULL; 1761 int error; 1762 1763 KASSERT((tcnp->cn_flags & HASBUF) != 0 && 1764 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name")); 1765 /* Check for cross-device rename */ 1766 if ((fvp->v_mount != tdvp->v_mount) || 1767 (tvp && (fvp->v_mount != tvp->v_mount))) { 1768 error = EXDEV; 1769 goto out; 1770 } 1771 1772 if (fvp == tvp) { 1773 printf("nfs_rename: fvp == tvp (can't happen)\n"); 1774 error = 0; 1775 goto out; 1776 } 1777 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0) 1778 goto out; 1779 1780 /* 1781 * We have to flush B_DELWRI data prior to renaming 1782 * the file. If we don't, the delayed-write buffers 1783 * can be flushed out later after the file has gone stale 1784 * under NFSV3. NFSV2 does not have this problem because 1785 * ( as far as I can tell ) it flushes dirty buffers more 1786 * often. 1787 * 1788 * Skip the rename operation if the fsync fails, this can happen 1789 * due to the server's volume being full, when we pushed out data 1790 * that was written back to our cache earlier. Not checking for 1791 * this condition can result in potential (silent) data loss. 1792 */ 1793 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread); 1794 NFSVOPUNLOCK(fvp, 0); 1795 if (!error && tvp) 1796 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread); 1797 if (error) 1798 goto out; 1799 1800 /* 1801 * If the tvp exists and is in use, sillyrename it before doing the 1802 * rename of the new file over it. 1803 * XXX Can't sillyrename a directory. 1804 */ 1805 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename && 1806 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1807 vput(tvp); 1808 tvp = NULL; 1809 } 1810 1811 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1812 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1813 tcnp->cn_thread); 1814 1815 if (error == 0 && NFS_ISV4(tdvp)) { 1816 /* 1817 * For NFSv4, check to see if it is the same name and 1818 * replace the name, if it is different. 1819 */ 1820 newv4 = malloc( 1821 sizeof (struct nfsv4node) + 1822 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1, 1823 M_NFSV4NODE, M_WAITOK); 1824 mtx_lock(&tdnp->n_mtx); 1825 mtx_lock(&fnp->n_mtx); 1826 if (fnp->n_v4 != NULL && fvp->v_type == VREG && 1827 (fnp->n_v4->n4_namelen != tcnp->cn_namelen || 1828 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4), 1829 tcnp->cn_namelen) || 1830 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen || 1831 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 1832 tdnp->n_fhp->nfh_len))) { 1833 #ifdef notdef 1834 { char nnn[100]; int nnnl; 1835 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99; 1836 bcopy(tcnp->cn_nameptr, nnn, nnnl); 1837 nnn[nnnl] = '\0'; 1838 printf("ren replace=%s\n",nnn); 1839 } 1840 #endif 1841 free(fnp->n_v4, M_NFSV4NODE); 1842 fnp->n_v4 = newv4; 1843 newv4 = NULL; 1844 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len; 1845 fnp->n_v4->n4_namelen = tcnp->cn_namelen; 1846 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 1847 tdnp->n_fhp->nfh_len); 1848 NFSBCOPY(tcnp->cn_nameptr, 1849 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen); 1850 } 1851 mtx_unlock(&tdnp->n_mtx); 1852 mtx_unlock(&fnp->n_mtx); 1853 if (newv4 != NULL) 1854 free(newv4, M_NFSV4NODE); 1855 } 1856 1857 if (fvp->v_type == VDIR) { 1858 if (tvp != NULL && tvp->v_type == VDIR) 1859 cache_purge(tdvp); 1860 cache_purge(fdvp); 1861 } 1862 1863 out: 1864 if (tdvp == tvp) 1865 vrele(tdvp); 1866 else 1867 vput(tdvp); 1868 if (tvp) 1869 vput(tvp); 1870 vrele(fdvp); 1871 vrele(fvp); 1872 /* 1873 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1874 */ 1875 if (error == ENOENT) 1876 error = 0; 1877 return (error); 1878 } 1879 1880 /* 1881 * nfs file rename rpc called from nfs_remove() above 1882 */ 1883 static int 1884 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp, 1885 struct sillyrename *sp) 1886 { 1887 1888 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen, 1889 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred, 1890 scnp->cn_thread)); 1891 } 1892 1893 /* 1894 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1895 */ 1896 static int 1897 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr, 1898 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr, 1899 int tnamelen, struct ucred *cred, struct thread *td) 1900 { 1901 struct nfsvattr fnfsva, tnfsva; 1902 struct nfsnode *fdnp = VTONFS(fdvp); 1903 struct nfsnode *tdnp = VTONFS(tdvp); 1904 int error = 0, fattrflag, tattrflag; 1905 1906 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp, 1907 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag, 1908 &tattrflag, NULL, NULL); 1909 mtx_lock(&fdnp->n_mtx); 1910 fdnp->n_flag |= NMODIFIED; 1911 if (fattrflag != 0) { 1912 mtx_unlock(&fdnp->n_mtx); 1913 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1); 1914 } else { 1915 fdnp->n_attrstamp = 0; 1916 mtx_unlock(&fdnp->n_mtx); 1917 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp); 1918 } 1919 mtx_lock(&tdnp->n_mtx); 1920 tdnp->n_flag |= NMODIFIED; 1921 if (tattrflag != 0) { 1922 mtx_unlock(&tdnp->n_mtx); 1923 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1); 1924 } else { 1925 tdnp->n_attrstamp = 0; 1926 mtx_unlock(&tdnp->n_mtx); 1927 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 1928 } 1929 if (error && NFS_ISV4(fdvp)) 1930 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1931 return (error); 1932 } 1933 1934 /* 1935 * nfs hard link create call 1936 */ 1937 static int 1938 nfs_link(struct vop_link_args *ap) 1939 { 1940 struct vnode *vp = ap->a_vp; 1941 struct vnode *tdvp = ap->a_tdvp; 1942 struct componentname *cnp = ap->a_cnp; 1943 struct nfsnode *np, *tdnp; 1944 struct nfsvattr nfsva, dnfsva; 1945 int error = 0, attrflag, dattrflag; 1946 1947 /* 1948 * Push all writes to the server, so that the attribute cache 1949 * doesn't get "out of sync" with the server. 1950 * XXX There should be a better way! 1951 */ 1952 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread); 1953 1954 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 1955 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag, 1956 &dattrflag, NULL); 1957 tdnp = VTONFS(tdvp); 1958 mtx_lock(&tdnp->n_mtx); 1959 tdnp->n_flag |= NMODIFIED; 1960 if (dattrflag != 0) { 1961 mtx_unlock(&tdnp->n_mtx); 1962 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1); 1963 } else { 1964 tdnp->n_attrstamp = 0; 1965 mtx_unlock(&tdnp->n_mtx); 1966 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 1967 } 1968 if (attrflag) 1969 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1970 else { 1971 np = VTONFS(vp); 1972 mtx_lock(&np->n_mtx); 1973 np->n_attrstamp = 0; 1974 mtx_unlock(&np->n_mtx); 1975 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 1976 } 1977 /* 1978 * If negative lookup caching is enabled, I might as well 1979 * add an entry for this node. Not necessary for correctness, 1980 * but if negative caching is enabled, then the system 1981 * must care about lookup caching hit rate, so... 1982 */ 1983 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 && 1984 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 1985 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL); 1986 } 1987 if (error && NFS_ISV4(vp)) 1988 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, 1989 (gid_t)0); 1990 return (error); 1991 } 1992 1993 /* 1994 * nfs symbolic link create call 1995 */ 1996 static int 1997 nfs_symlink(struct vop_symlink_args *ap) 1998 { 1999 struct vnode *dvp = ap->a_dvp; 2000 struct vattr *vap = ap->a_vap; 2001 struct componentname *cnp = ap->a_cnp; 2002 struct nfsvattr nfsva, dnfsva; 2003 struct nfsfh *nfhp; 2004 struct nfsnode *np = NULL, *dnp; 2005 struct vnode *newvp = NULL; 2006 int error = 0, attrflag, dattrflag, ret; 2007 2008 vap->va_type = VLNK; 2009 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2010 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, 2011 &nfsva, &nfhp, &attrflag, &dattrflag, NULL); 2012 if (nfhp) { 2013 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, 2014 &np, NULL, LK_EXCLUSIVE); 2015 if (!ret) 2016 newvp = NFSTOV(np); 2017 else if (!error) 2018 error = ret; 2019 } 2020 if (newvp != NULL) { 2021 if (attrflag) 2022 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 2023 0, 1); 2024 } else if (!error) { 2025 /* 2026 * If we do not have an error and we could not extract the 2027 * newvp from the response due to the request being NFSv2, we 2028 * have to do a lookup in order to obtain a newvp to return. 2029 */ 2030 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2031 cnp->cn_cred, cnp->cn_thread, &np); 2032 if (!error) 2033 newvp = NFSTOV(np); 2034 } 2035 if (error) { 2036 if (newvp) 2037 vput(newvp); 2038 if (NFS_ISV4(dvp)) 2039 error = nfscl_maperr(cnp->cn_thread, error, 2040 vap->va_uid, vap->va_gid); 2041 } else { 2042 *ap->a_vpp = newvp; 2043 } 2044 2045 dnp = VTONFS(dvp); 2046 mtx_lock(&dnp->n_mtx); 2047 dnp->n_flag |= NMODIFIED; 2048 if (dattrflag != 0) { 2049 mtx_unlock(&dnp->n_mtx); 2050 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2051 } else { 2052 dnp->n_attrstamp = 0; 2053 mtx_unlock(&dnp->n_mtx); 2054 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2055 } 2056 /* 2057 * If negative lookup caching is enabled, I might as well 2058 * add an entry for this node. Not necessary for correctness, 2059 * but if negative caching is enabled, then the system 2060 * must care about lookup caching hit rate, so... 2061 */ 2062 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2063 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 2064 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL); 2065 } 2066 return (error); 2067 } 2068 2069 /* 2070 * nfs make dir call 2071 */ 2072 static int 2073 nfs_mkdir(struct vop_mkdir_args *ap) 2074 { 2075 struct vnode *dvp = ap->a_dvp; 2076 struct vattr *vap = ap->a_vap; 2077 struct componentname *cnp = ap->a_cnp; 2078 struct nfsnode *np = NULL, *dnp; 2079 struct vnode *newvp = NULL; 2080 struct vattr vattr; 2081 struct nfsfh *nfhp; 2082 struct nfsvattr nfsva, dnfsva; 2083 int error = 0, attrflag, dattrflag, ret; 2084 2085 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) 2086 return (error); 2087 vap->va_type = VDIR; 2088 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2089 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp, 2090 &attrflag, &dattrflag, NULL); 2091 dnp = VTONFS(dvp); 2092 mtx_lock(&dnp->n_mtx); 2093 dnp->n_flag |= NMODIFIED; 2094 if (dattrflag != 0) { 2095 mtx_unlock(&dnp->n_mtx); 2096 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2097 } else { 2098 dnp->n_attrstamp = 0; 2099 mtx_unlock(&dnp->n_mtx); 2100 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2101 } 2102 if (nfhp) { 2103 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, 2104 &np, NULL, LK_EXCLUSIVE); 2105 if (!ret) { 2106 newvp = NFSTOV(np); 2107 if (attrflag) 2108 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 2109 NULL, 0, 1); 2110 } else if (!error) 2111 error = ret; 2112 } 2113 if (!error && newvp == NULL) { 2114 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2115 cnp->cn_cred, cnp->cn_thread, &np); 2116 if (!error) { 2117 newvp = NFSTOV(np); 2118 if (newvp->v_type != VDIR) 2119 error = EEXIST; 2120 } 2121 } 2122 if (error) { 2123 if (newvp) 2124 vput(newvp); 2125 if (NFS_ISV4(dvp)) 2126 error = nfscl_maperr(cnp->cn_thread, error, 2127 vap->va_uid, vap->va_gid); 2128 } else { 2129 /* 2130 * If negative lookup caching is enabled, I might as well 2131 * add an entry for this node. Not necessary for correctness, 2132 * but if negative caching is enabled, then the system 2133 * must care about lookup caching hit rate, so... 2134 */ 2135 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2136 (cnp->cn_flags & MAKEENTRY) && 2137 attrflag != 0 && dattrflag != 0) 2138 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 2139 &dnfsva.na_ctime); 2140 *ap->a_vpp = newvp; 2141 } 2142 return (error); 2143 } 2144 2145 /* 2146 * nfs remove directory call 2147 */ 2148 static int 2149 nfs_rmdir(struct vop_rmdir_args *ap) 2150 { 2151 struct vnode *vp = ap->a_vp; 2152 struct vnode *dvp = ap->a_dvp; 2153 struct componentname *cnp = ap->a_cnp; 2154 struct nfsnode *dnp; 2155 struct nfsvattr dnfsva; 2156 int error, dattrflag; 2157 2158 if (dvp == vp) 2159 return (EINVAL); 2160 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2161 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL); 2162 dnp = VTONFS(dvp); 2163 mtx_lock(&dnp->n_mtx); 2164 dnp->n_flag |= NMODIFIED; 2165 if (dattrflag != 0) { 2166 mtx_unlock(&dnp->n_mtx); 2167 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2168 } else { 2169 dnp->n_attrstamp = 0; 2170 mtx_unlock(&dnp->n_mtx); 2171 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2172 } 2173 2174 cache_purge(dvp); 2175 cache_purge(vp); 2176 if (error && NFS_ISV4(dvp)) 2177 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, 2178 (gid_t)0); 2179 /* 2180 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2181 */ 2182 if (error == ENOENT) 2183 error = 0; 2184 return (error); 2185 } 2186 2187 /* 2188 * nfs readdir call 2189 */ 2190 static int 2191 nfs_readdir(struct vop_readdir_args *ap) 2192 { 2193 struct vnode *vp = ap->a_vp; 2194 struct nfsnode *np = VTONFS(vp); 2195 struct uio *uio = ap->a_uio; 2196 ssize_t tresid, left; 2197 int error = 0; 2198 struct vattr vattr; 2199 2200 if (ap->a_eofflag != NULL) 2201 *ap->a_eofflag = 0; 2202 if (vp->v_type != VDIR) 2203 return(EPERM); 2204 2205 /* 2206 * First, check for hit on the EOF offset cache 2207 */ 2208 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2209 (np->n_flag & NMODIFIED) == 0) { 2210 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) { 2211 mtx_lock(&np->n_mtx); 2212 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) || 2213 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 2214 mtx_unlock(&np->n_mtx); 2215 NFSINCRGLOBAL(nfsstatsv1.direofcache_hits); 2216 if (ap->a_eofflag != NULL) 2217 *ap->a_eofflag = 1; 2218 return (0); 2219 } else 2220 mtx_unlock(&np->n_mtx); 2221 } 2222 } 2223 2224 /* 2225 * NFS always guarantees that directory entries don't straddle 2226 * DIRBLKSIZ boundaries. As such, we need to limit the size 2227 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial 2228 * directory entry. 2229 */ 2230 left = uio->uio_resid % DIRBLKSIZ; 2231 if (left == uio->uio_resid) 2232 return (EINVAL); 2233 uio->uio_resid -= left; 2234 2235 /* 2236 * Call ncl_bioread() to do the real work. 2237 */ 2238 tresid = uio->uio_resid; 2239 error = ncl_bioread(vp, uio, 0, ap->a_cred); 2240 2241 if (!error && uio->uio_resid == tresid) { 2242 NFSINCRGLOBAL(nfsstatsv1.direofcache_misses); 2243 if (ap->a_eofflag != NULL) 2244 *ap->a_eofflag = 1; 2245 } 2246 2247 /* Add the partial DIRBLKSIZ (left) back in. */ 2248 uio->uio_resid += left; 2249 return (error); 2250 } 2251 2252 /* 2253 * Readdir rpc call. 2254 * Called from below the buffer cache by ncl_doio(). 2255 */ 2256 int 2257 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2258 struct thread *td) 2259 { 2260 struct nfsvattr nfsva; 2261 nfsuint64 *cookiep, cookie; 2262 struct nfsnode *dnp = VTONFS(vp); 2263 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2264 int error = 0, eof, attrflag; 2265 2266 KASSERT(uiop->uio_iovcnt == 1 && 2267 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2268 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2269 ("nfs readdirrpc bad uio")); 2270 2271 /* 2272 * If there is no cookie, assume directory was stale. 2273 */ 2274 ncl_dircookie_lock(dnp); 2275 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2276 if (cookiep) { 2277 cookie = *cookiep; 2278 ncl_dircookie_unlock(dnp); 2279 } else { 2280 ncl_dircookie_unlock(dnp); 2281 return (NFSERR_BAD_COOKIE); 2282 } 2283 2284 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2285 (void)ncl_fsinfo(nmp, vp, cred, td); 2286 2287 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva, 2288 &attrflag, &eof, NULL); 2289 if (attrflag) 2290 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2291 2292 if (!error) { 2293 /* 2294 * We are now either at the end of the directory or have filled 2295 * the block. 2296 */ 2297 if (eof) 2298 dnp->n_direofoffset = uiop->uio_offset; 2299 else { 2300 if (uiop->uio_resid > 0) 2301 printf("EEK! readdirrpc resid > 0\n"); 2302 ncl_dircookie_lock(dnp); 2303 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2304 *cookiep = cookie; 2305 ncl_dircookie_unlock(dnp); 2306 } 2307 } else if (NFS_ISV4(vp)) { 2308 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2309 } 2310 return (error); 2311 } 2312 2313 /* 2314 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc(). 2315 */ 2316 int 2317 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2318 struct thread *td) 2319 { 2320 struct nfsvattr nfsva; 2321 nfsuint64 *cookiep, cookie; 2322 struct nfsnode *dnp = VTONFS(vp); 2323 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2324 int error = 0, attrflag, eof; 2325 2326 KASSERT(uiop->uio_iovcnt == 1 && 2327 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2328 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2329 ("nfs readdirplusrpc bad uio")); 2330 2331 /* 2332 * If there is no cookie, assume directory was stale. 2333 */ 2334 ncl_dircookie_lock(dnp); 2335 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2336 if (cookiep) { 2337 cookie = *cookiep; 2338 ncl_dircookie_unlock(dnp); 2339 } else { 2340 ncl_dircookie_unlock(dnp); 2341 return (NFSERR_BAD_COOKIE); 2342 } 2343 2344 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2345 (void)ncl_fsinfo(nmp, vp, cred, td); 2346 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva, 2347 &attrflag, &eof, NULL); 2348 if (attrflag) 2349 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2350 2351 if (!error) { 2352 /* 2353 * We are now either at end of the directory or have filled the 2354 * the block. 2355 */ 2356 if (eof) 2357 dnp->n_direofoffset = uiop->uio_offset; 2358 else { 2359 if (uiop->uio_resid > 0) 2360 printf("EEK! readdirplusrpc resid > 0\n"); 2361 ncl_dircookie_lock(dnp); 2362 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2363 *cookiep = cookie; 2364 ncl_dircookie_unlock(dnp); 2365 } 2366 } else if (NFS_ISV4(vp)) { 2367 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2368 } 2369 return (error); 2370 } 2371 2372 /* 2373 * Silly rename. To make the NFS filesystem that is stateless look a little 2374 * more like the "ufs" a remove of an active vnode is translated to a rename 2375 * to a funny looking filename that is removed by nfs_inactive on the 2376 * nfsnode. There is the potential for another process on a different client 2377 * to create the same funny name between the nfs_lookitup() fails and the 2378 * nfs_rename() completes, but... 2379 */ 2380 static int 2381 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2382 { 2383 struct sillyrename *sp; 2384 struct nfsnode *np; 2385 int error; 2386 short pid; 2387 unsigned int lticks; 2388 2389 cache_purge(dvp); 2390 np = VTONFS(vp); 2391 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir")); 2392 sp = malloc(sizeof (struct sillyrename), 2393 M_NEWNFSREQ, M_WAITOK); 2394 sp->s_cred = crhold(cnp->cn_cred); 2395 sp->s_dvp = dvp; 2396 VREF(dvp); 2397 2398 /* 2399 * Fudge together a funny name. 2400 * Changing the format of the funny name to accommodate more 2401 * sillynames per directory. 2402 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is 2403 * CPU ticks since boot. 2404 */ 2405 pid = cnp->cn_thread->td_proc->p_pid; 2406 lticks = (unsigned int)ticks; 2407 for ( ; ; ) { 2408 sp->s_namlen = sprintf(sp->s_name, 2409 ".nfs.%08x.%04x4.4", lticks, 2410 pid); 2411 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2412 cnp->cn_thread, NULL)) 2413 break; 2414 lticks++; 2415 } 2416 error = nfs_renameit(dvp, vp, cnp, sp); 2417 if (error) 2418 goto bad; 2419 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2420 cnp->cn_thread, &np); 2421 np->n_sillyrename = sp; 2422 return (0); 2423 bad: 2424 vrele(sp->s_dvp); 2425 crfree(sp->s_cred); 2426 free(sp, M_NEWNFSREQ); 2427 return (error); 2428 } 2429 2430 /* 2431 * Look up a file name and optionally either update the file handle or 2432 * allocate an nfsnode, depending on the value of npp. 2433 * npp == NULL --> just do the lookup 2434 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2435 * handled too 2436 * *npp != NULL --> update the file handle in the vnode 2437 */ 2438 static int 2439 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred, 2440 struct thread *td, struct nfsnode **npp) 2441 { 2442 struct vnode *newvp = NULL, *vp; 2443 struct nfsnode *np, *dnp = VTONFS(dvp); 2444 struct nfsfh *nfhp, *onfhp; 2445 struct nfsvattr nfsva, dnfsva; 2446 struct componentname cn; 2447 int error = 0, attrflag, dattrflag; 2448 u_int hash; 2449 2450 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva, 2451 &nfhp, &attrflag, &dattrflag, NULL); 2452 if (dattrflag) 2453 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2454 if (npp && !error) { 2455 if (*npp != NULL) { 2456 np = *npp; 2457 vp = NFSTOV(np); 2458 /* 2459 * For NFSv4, check to see if it is the same name and 2460 * replace the name, if it is different. 2461 */ 2462 if (np->n_v4 != NULL && nfsva.na_type == VREG && 2463 (np->n_v4->n4_namelen != len || 2464 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) || 2465 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || 2466 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2467 dnp->n_fhp->nfh_len))) { 2468 #ifdef notdef 2469 { char nnn[100]; int nnnl; 2470 nnnl = (len < 100) ? len : 99; 2471 bcopy(name, nnn, nnnl); 2472 nnn[nnnl] = '\0'; 2473 printf("replace=%s\n",nnn); 2474 } 2475 #endif 2476 free(np->n_v4, M_NFSV4NODE); 2477 np->n_v4 = malloc( 2478 sizeof (struct nfsv4node) + 2479 dnp->n_fhp->nfh_len + len - 1, 2480 M_NFSV4NODE, M_WAITOK); 2481 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; 2482 np->n_v4->n4_namelen = len; 2483 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2484 dnp->n_fhp->nfh_len); 2485 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len); 2486 } 2487 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, 2488 FNV1_32_INIT); 2489 onfhp = np->n_fhp; 2490 /* 2491 * Rehash node for new file handle. 2492 */ 2493 vfs_hash_rehash(vp, hash); 2494 np->n_fhp = nfhp; 2495 if (onfhp != NULL) 2496 free(onfhp, M_NFSFH); 2497 newvp = NFSTOV(np); 2498 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) { 2499 free(nfhp, M_NFSFH); 2500 VREF(dvp); 2501 newvp = dvp; 2502 } else { 2503 cn.cn_nameptr = name; 2504 cn.cn_namelen = len; 2505 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td, 2506 &np, NULL, LK_EXCLUSIVE); 2507 if (error) 2508 return (error); 2509 newvp = NFSTOV(np); 2510 } 2511 if (!attrflag && *npp == NULL) { 2512 if (newvp == dvp) 2513 vrele(newvp); 2514 else 2515 vput(newvp); 2516 return (ENOENT); 2517 } 2518 if (attrflag) 2519 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 2520 0, 1); 2521 } 2522 if (npp && *npp == NULL) { 2523 if (error) { 2524 if (newvp) { 2525 if (newvp == dvp) 2526 vrele(newvp); 2527 else 2528 vput(newvp); 2529 } 2530 } else 2531 *npp = np; 2532 } 2533 if (error && NFS_ISV4(dvp)) 2534 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2535 return (error); 2536 } 2537 2538 /* 2539 * Nfs Version 3 and 4 commit rpc 2540 */ 2541 int 2542 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred, 2543 struct thread *td) 2544 { 2545 struct nfsvattr nfsva; 2546 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2547 struct nfsnode *np; 2548 struct uio uio; 2549 int error, attrflag; 2550 2551 np = VTONFS(vp); 2552 error = EIO; 2553 attrflag = 0; 2554 if (NFSHASPNFS(nmp) && (np->n_flag & NDSCOMMIT) != 0) { 2555 uio.uio_offset = offset; 2556 uio.uio_resid = cnt; 2557 error = nfscl_doiods(vp, &uio, NULL, NULL, 2558 NFSV4OPEN_ACCESSWRITE, 1, cred, td); 2559 if (error != 0) { 2560 mtx_lock(&np->n_mtx); 2561 np->n_flag &= ~NDSCOMMIT; 2562 mtx_unlock(&np->n_mtx); 2563 } 2564 } 2565 if (error != 0) { 2566 mtx_lock(&nmp->nm_mtx); 2567 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) { 2568 mtx_unlock(&nmp->nm_mtx); 2569 return (0); 2570 } 2571 mtx_unlock(&nmp->nm_mtx); 2572 error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva, 2573 &attrflag, NULL); 2574 } 2575 if (attrflag != 0) 2576 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 2577 0, 1); 2578 if (error != 0 && NFS_ISV4(vp)) 2579 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2580 return (error); 2581 } 2582 2583 /* 2584 * Strategy routine. 2585 * For async requests when nfsiod(s) are running, queue the request by 2586 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the 2587 * request. 2588 */ 2589 static int 2590 nfs_strategy(struct vop_strategy_args *ap) 2591 { 2592 struct buf *bp; 2593 struct vnode *vp; 2594 struct ucred *cr; 2595 2596 bp = ap->a_bp; 2597 vp = ap->a_vp; 2598 KASSERT(bp->b_vp == vp, ("missing b_getvp")); 2599 KASSERT(!(bp->b_flags & B_DONE), 2600 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); 2601 BUF_ASSERT_HELD(bp); 2602 2603 if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno) 2604 bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize / 2605 DEV_BSIZE); 2606 if (bp->b_iocmd == BIO_READ) 2607 cr = bp->b_rcred; 2608 else 2609 cr = bp->b_wcred; 2610 2611 /* 2612 * If the op is asynchronous and an i/o daemon is waiting 2613 * queue the request, wake it up and wait for completion 2614 * otherwise just do it ourselves. 2615 */ 2616 if ((bp->b_flags & B_ASYNC) == 0 || 2617 ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread)) 2618 (void) ncl_doio(vp, bp, cr, curthread, 1); 2619 return (0); 2620 } 2621 2622 /* 2623 * fsync vnode op. Just call ncl_flush() with commit == 1. 2624 */ 2625 /* ARGSUSED */ 2626 static int 2627 nfs_fsync(struct vop_fsync_args *ap) 2628 { 2629 2630 if (ap->a_vp->v_type != VREG) { 2631 /* 2632 * For NFS, metadata is changed synchronously on the server, 2633 * so there is nothing to flush. Also, ncl_flush() clears 2634 * the NMODIFIED flag and that shouldn't be done here for 2635 * directories. 2636 */ 2637 return (0); 2638 } 2639 return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0)); 2640 } 2641 2642 /* 2643 * Flush all the blocks associated with a vnode. 2644 * Walk through the buffer pool and push any dirty pages 2645 * associated with the vnode. 2646 * If the called_from_renewthread argument is TRUE, it has been called 2647 * from the NFSv4 renew thread and, as such, cannot block indefinitely 2648 * waiting for a buffer write to complete. 2649 */ 2650 int 2651 ncl_flush(struct vnode *vp, int waitfor, struct thread *td, 2652 int commit, int called_from_renewthread) 2653 { 2654 struct nfsnode *np = VTONFS(vp); 2655 struct buf *bp; 2656 int i; 2657 struct buf *nbp; 2658 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2659 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos; 2660 int passone = 1, trycnt = 0; 2661 u_quad_t off, endoff, toff; 2662 struct ucred* wcred = NULL; 2663 struct buf **bvec = NULL; 2664 struct bufobj *bo; 2665 #ifndef NFS_COMMITBVECSIZ 2666 #define NFS_COMMITBVECSIZ 20 2667 #endif 2668 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ]; 2669 u_int bvecsize = 0, bveccount; 2670 2671 if (called_from_renewthread != 0) 2672 slptimeo = hz; 2673 if (nmp->nm_flag & NFSMNT_INT) 2674 slpflag = PCATCH; 2675 if (!commit) 2676 passone = 0; 2677 bo = &vp->v_bufobj; 2678 /* 2679 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the 2680 * server, but has not been committed to stable storage on the server 2681 * yet. On the first pass, the byte range is worked out and the commit 2682 * rpc is done. On the second pass, ncl_writebp() is called to do the 2683 * job. 2684 */ 2685 again: 2686 off = (u_quad_t)-1; 2687 endoff = 0; 2688 bvecpos = 0; 2689 if (NFS_ISV34(vp) && commit) { 2690 if (bvec != NULL && bvec != bvec_on_stack) 2691 free(bvec, M_TEMP); 2692 /* 2693 * Count up how many buffers waiting for a commit. 2694 */ 2695 bveccount = 0; 2696 BO_LOCK(bo); 2697 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2698 if (!BUF_ISLOCKED(bp) && 2699 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) 2700 == (B_DELWRI | B_NEEDCOMMIT)) 2701 bveccount++; 2702 } 2703 /* 2704 * Allocate space to remember the list of bufs to commit. It is 2705 * important to use M_NOWAIT here to avoid a race with nfs_write. 2706 * If we can't get memory (for whatever reason), we will end up 2707 * committing the buffers one-by-one in the loop below. 2708 */ 2709 if (bveccount > NFS_COMMITBVECSIZ) { 2710 /* 2711 * Release the vnode interlock to avoid a lock 2712 * order reversal. 2713 */ 2714 BO_UNLOCK(bo); 2715 bvec = (struct buf **) 2716 malloc(bveccount * sizeof(struct buf *), 2717 M_TEMP, M_NOWAIT); 2718 BO_LOCK(bo); 2719 if (bvec == NULL) { 2720 bvec = bvec_on_stack; 2721 bvecsize = NFS_COMMITBVECSIZ; 2722 } else 2723 bvecsize = bveccount; 2724 } else { 2725 bvec = bvec_on_stack; 2726 bvecsize = NFS_COMMITBVECSIZ; 2727 } 2728 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2729 if (bvecpos >= bvecsize) 2730 break; 2731 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2732 nbp = TAILQ_NEXT(bp, b_bobufs); 2733 continue; 2734 } 2735 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 2736 (B_DELWRI | B_NEEDCOMMIT)) { 2737 BUF_UNLOCK(bp); 2738 nbp = TAILQ_NEXT(bp, b_bobufs); 2739 continue; 2740 } 2741 BO_UNLOCK(bo); 2742 bremfree(bp); 2743 /* 2744 * Work out if all buffers are using the same cred 2745 * so we can deal with them all with one commit. 2746 * 2747 * NOTE: we are not clearing B_DONE here, so we have 2748 * to do it later on in this routine if we intend to 2749 * initiate I/O on the bp. 2750 * 2751 * Note: to avoid loopback deadlocks, we do not 2752 * assign b_runningbufspace. 2753 */ 2754 if (wcred == NULL) 2755 wcred = bp->b_wcred; 2756 else if (wcred != bp->b_wcred) 2757 wcred = NOCRED; 2758 vfs_busy_pages(bp, 1); 2759 2760 BO_LOCK(bo); 2761 /* 2762 * bp is protected by being locked, but nbp is not 2763 * and vfs_busy_pages() may sleep. We have to 2764 * recalculate nbp. 2765 */ 2766 nbp = TAILQ_NEXT(bp, b_bobufs); 2767 2768 /* 2769 * A list of these buffers is kept so that the 2770 * second loop knows which buffers have actually 2771 * been committed. This is necessary, since there 2772 * may be a race between the commit rpc and new 2773 * uncommitted writes on the file. 2774 */ 2775 bvec[bvecpos++] = bp; 2776 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2777 bp->b_dirtyoff; 2778 if (toff < off) 2779 off = toff; 2780 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); 2781 if (toff > endoff) 2782 endoff = toff; 2783 } 2784 BO_UNLOCK(bo); 2785 } 2786 if (bvecpos > 0) { 2787 /* 2788 * Commit data on the server, as required. 2789 * If all bufs are using the same wcred, then use that with 2790 * one call for all of them, otherwise commit each one 2791 * separately. 2792 */ 2793 if (wcred != NOCRED) 2794 retv = ncl_commit(vp, off, (int)(endoff - off), 2795 wcred, td); 2796 else { 2797 retv = 0; 2798 for (i = 0; i < bvecpos; i++) { 2799 off_t off, size; 2800 bp = bvec[i]; 2801 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2802 bp->b_dirtyoff; 2803 size = (u_quad_t)(bp->b_dirtyend 2804 - bp->b_dirtyoff); 2805 retv = ncl_commit(vp, off, (int)size, 2806 bp->b_wcred, td); 2807 if (retv) break; 2808 } 2809 } 2810 2811 if (retv == NFSERR_STALEWRITEVERF) 2812 ncl_clearcommit(vp->v_mount); 2813 2814 /* 2815 * Now, either mark the blocks I/O done or mark the 2816 * blocks dirty, depending on whether the commit 2817 * succeeded. 2818 */ 2819 for (i = 0; i < bvecpos; i++) { 2820 bp = bvec[i]; 2821 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 2822 if (retv) { 2823 /* 2824 * Error, leave B_DELWRI intact 2825 */ 2826 vfs_unbusy_pages(bp); 2827 brelse(bp); 2828 } else { 2829 /* 2830 * Success, remove B_DELWRI ( bundirty() ). 2831 * 2832 * b_dirtyoff/b_dirtyend seem to be NFS 2833 * specific. We should probably move that 2834 * into bundirty(). XXX 2835 */ 2836 bufobj_wref(bo); 2837 bp->b_flags |= B_ASYNC; 2838 bundirty(bp); 2839 bp->b_flags &= ~B_DONE; 2840 bp->b_ioflags &= ~BIO_ERROR; 2841 bp->b_dirtyoff = bp->b_dirtyend = 0; 2842 bufdone(bp); 2843 } 2844 } 2845 } 2846 2847 /* 2848 * Start/do any write(s) that are required. 2849 */ 2850 loop: 2851 BO_LOCK(bo); 2852 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2853 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2854 if (waitfor != MNT_WAIT || passone) 2855 continue; 2856 2857 error = BUF_TIMELOCK(bp, 2858 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 2859 BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo); 2860 if (error == 0) { 2861 BUF_UNLOCK(bp); 2862 goto loop; 2863 } 2864 if (error == ENOLCK) { 2865 error = 0; 2866 goto loop; 2867 } 2868 if (called_from_renewthread != 0) { 2869 /* 2870 * Return EIO so the flush will be retried 2871 * later. 2872 */ 2873 error = EIO; 2874 goto done; 2875 } 2876 if (newnfs_sigintr(nmp, td)) { 2877 error = EINTR; 2878 goto done; 2879 } 2880 if (slpflag == PCATCH) { 2881 slpflag = 0; 2882 slptimeo = 2 * hz; 2883 } 2884 goto loop; 2885 } 2886 if ((bp->b_flags & B_DELWRI) == 0) 2887 panic("nfs_fsync: not dirty"); 2888 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) { 2889 BUF_UNLOCK(bp); 2890 continue; 2891 } 2892 BO_UNLOCK(bo); 2893 bremfree(bp); 2894 if (passone || !commit) 2895 bp->b_flags |= B_ASYNC; 2896 else 2897 bp->b_flags |= B_ASYNC; 2898 bwrite(bp); 2899 if (newnfs_sigintr(nmp, td)) { 2900 error = EINTR; 2901 goto done; 2902 } 2903 goto loop; 2904 } 2905 if (passone) { 2906 passone = 0; 2907 BO_UNLOCK(bo); 2908 goto again; 2909 } 2910 if (waitfor == MNT_WAIT) { 2911 while (bo->bo_numoutput) { 2912 error = bufobj_wwait(bo, slpflag, slptimeo); 2913 if (error) { 2914 BO_UNLOCK(bo); 2915 if (called_from_renewthread != 0) { 2916 /* 2917 * Return EIO so that the flush will be 2918 * retried later. 2919 */ 2920 error = EIO; 2921 goto done; 2922 } 2923 error = newnfs_sigintr(nmp, td); 2924 if (error) 2925 goto done; 2926 if (slpflag == PCATCH) { 2927 slpflag = 0; 2928 slptimeo = 2 * hz; 2929 } 2930 BO_LOCK(bo); 2931 } 2932 } 2933 if (bo->bo_dirty.bv_cnt != 0 && commit) { 2934 BO_UNLOCK(bo); 2935 goto loop; 2936 } 2937 /* 2938 * Wait for all the async IO requests to drain 2939 */ 2940 BO_UNLOCK(bo); 2941 mtx_lock(&np->n_mtx); 2942 while (np->n_directio_asyncwr > 0) { 2943 np->n_flag |= NFSYNCWAIT; 2944 error = newnfs_msleep(td, &np->n_directio_asyncwr, 2945 &np->n_mtx, slpflag | (PRIBIO + 1), 2946 "nfsfsync", 0); 2947 if (error) { 2948 if (newnfs_sigintr(nmp, td)) { 2949 mtx_unlock(&np->n_mtx); 2950 error = EINTR; 2951 goto done; 2952 } 2953 } 2954 } 2955 mtx_unlock(&np->n_mtx); 2956 } else 2957 BO_UNLOCK(bo); 2958 if (NFSHASPNFS(nmp)) { 2959 nfscl_layoutcommit(vp, td); 2960 /* 2961 * Invalidate the attribute cache, since writes to a DS 2962 * won't update the size attribute. 2963 */ 2964 mtx_lock(&np->n_mtx); 2965 np->n_attrstamp = 0; 2966 } else 2967 mtx_lock(&np->n_mtx); 2968 if (np->n_flag & NWRITEERR) { 2969 error = np->n_error; 2970 np->n_flag &= ~NWRITEERR; 2971 } 2972 if (commit && bo->bo_dirty.bv_cnt == 0 && 2973 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0) 2974 np->n_flag &= ~NMODIFIED; 2975 mtx_unlock(&np->n_mtx); 2976 done: 2977 if (bvec != NULL && bvec != bvec_on_stack) 2978 free(bvec, M_TEMP); 2979 if (error == 0 && commit != 0 && waitfor == MNT_WAIT && 2980 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 || 2981 np->n_directio_asyncwr != 0)) { 2982 if (trycnt++ < 5) { 2983 /* try, try again... */ 2984 passone = 1; 2985 wcred = NULL; 2986 bvec = NULL; 2987 bvecsize = 0; 2988 goto again; 2989 } 2990 vn_printf(vp, "ncl_flush failed"); 2991 error = called_from_renewthread != 0 ? EIO : EBUSY; 2992 } 2993 return (error); 2994 } 2995 2996 /* 2997 * NFS advisory byte-level locks. 2998 */ 2999 static int 3000 nfs_advlock(struct vop_advlock_args *ap) 3001 { 3002 struct vnode *vp = ap->a_vp; 3003 struct ucred *cred; 3004 struct nfsnode *np = VTONFS(ap->a_vp); 3005 struct proc *p = (struct proc *)ap->a_id; 3006 struct thread *td = curthread; /* XXX */ 3007 struct vattr va; 3008 int ret, error = EOPNOTSUPP; 3009 u_quad_t size; 3010 3011 if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) { 3012 if (vp->v_type != VREG) 3013 return (EINVAL); 3014 if ((ap->a_flags & F_POSIX) != 0) 3015 cred = p->p_ucred; 3016 else 3017 cred = td->td_ucred; 3018 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 3019 if (vp->v_iflag & VI_DOOMED) { 3020 NFSVOPUNLOCK(vp, 0); 3021 return (EBADF); 3022 } 3023 3024 /* 3025 * If this is unlocking a write locked region, flush and 3026 * commit them before unlocking. This is required by 3027 * RFC3530 Sec. 9.3.2. 3028 */ 3029 if (ap->a_op == F_UNLCK && 3030 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id, 3031 ap->a_flags)) 3032 (void) ncl_flush(vp, MNT_WAIT, td, 1, 0); 3033 3034 /* 3035 * Loop around doing the lock op, while a blocking lock 3036 * must wait for the lock op to succeed. 3037 */ 3038 do { 3039 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op, 3040 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags); 3041 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3042 ap->a_op == F_SETLK) { 3043 NFSVOPUNLOCK(vp, 0); 3044 error = nfs_catnap(PZERO | PCATCH, ret, 3045 "ncladvl"); 3046 if (error) 3047 return (EINTR); 3048 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 3049 if (vp->v_iflag & VI_DOOMED) { 3050 NFSVOPUNLOCK(vp, 0); 3051 return (EBADF); 3052 } 3053 } 3054 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3055 ap->a_op == F_SETLK); 3056 if (ret == NFSERR_DENIED) { 3057 NFSVOPUNLOCK(vp, 0); 3058 return (EAGAIN); 3059 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) { 3060 NFSVOPUNLOCK(vp, 0); 3061 return (ret); 3062 } else if (ret != 0) { 3063 NFSVOPUNLOCK(vp, 0); 3064 return (EACCES); 3065 } 3066 3067 /* 3068 * Now, if we just got a lock, invalidate data in the buffer 3069 * cache, as required, so that the coherency conforms with 3070 * RFC3530 Sec. 9.3.2. 3071 */ 3072 if (ap->a_op == F_SETLK) { 3073 if ((np->n_flag & NMODIFIED) == 0) { 3074 np->n_attrstamp = 0; 3075 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3076 ret = VOP_GETATTR(vp, &va, cred); 3077 } 3078 if ((np->n_flag & NMODIFIED) || ret || 3079 np->n_change != va.va_filerev) { 3080 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1); 3081 np->n_attrstamp = 0; 3082 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3083 ret = VOP_GETATTR(vp, &va, cred); 3084 if (!ret) { 3085 np->n_mtime = va.va_mtime; 3086 np->n_change = va.va_filerev; 3087 } 3088 } 3089 /* Mark that a file lock has been acquired. */ 3090 mtx_lock(&np->n_mtx); 3091 np->n_flag |= NHASBEENLOCKED; 3092 mtx_unlock(&np->n_mtx); 3093 } 3094 NFSVOPUNLOCK(vp, 0); 3095 return (0); 3096 } else if (!NFS_ISV4(vp)) { 3097 error = NFSVOPLOCK(vp, LK_SHARED); 3098 if (error) 3099 return (error); 3100 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3101 size = VTONFS(vp)->n_size; 3102 NFSVOPUNLOCK(vp, 0); 3103 error = lf_advlock(ap, &(vp->v_lockf), size); 3104 } else { 3105 if (nfs_advlock_p != NULL) 3106 error = nfs_advlock_p(ap); 3107 else { 3108 NFSVOPUNLOCK(vp, 0); 3109 error = ENOLCK; 3110 } 3111 } 3112 if (error == 0 && ap->a_op == F_SETLK) { 3113 error = NFSVOPLOCK(vp, LK_SHARED); 3114 if (error == 0) { 3115 /* Mark that a file lock has been acquired. */ 3116 mtx_lock(&np->n_mtx); 3117 np->n_flag |= NHASBEENLOCKED; 3118 mtx_unlock(&np->n_mtx); 3119 NFSVOPUNLOCK(vp, 0); 3120 } 3121 } 3122 } 3123 return (error); 3124 } 3125 3126 /* 3127 * NFS advisory byte-level locks. 3128 */ 3129 static int 3130 nfs_advlockasync(struct vop_advlockasync_args *ap) 3131 { 3132 struct vnode *vp = ap->a_vp; 3133 u_quad_t size; 3134 int error; 3135 3136 if (NFS_ISV4(vp)) 3137 return (EOPNOTSUPP); 3138 error = NFSVOPLOCK(vp, LK_SHARED); 3139 if (error) 3140 return (error); 3141 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3142 size = VTONFS(vp)->n_size; 3143 NFSVOPUNLOCK(vp, 0); 3144 error = lf_advlockasync(ap, &(vp->v_lockf), size); 3145 } else { 3146 NFSVOPUNLOCK(vp, 0); 3147 error = EOPNOTSUPP; 3148 } 3149 return (error); 3150 } 3151 3152 /* 3153 * Print out the contents of an nfsnode. 3154 */ 3155 static int 3156 nfs_print(struct vop_print_args *ap) 3157 { 3158 struct vnode *vp = ap->a_vp; 3159 struct nfsnode *np = VTONFS(vp); 3160 3161 printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid, 3162 (uintmax_t)np->n_vattr.na_fsid); 3163 if (vp->v_type == VFIFO) 3164 fifo_printinfo(vp); 3165 printf("\n"); 3166 return (0); 3167 } 3168 3169 /* 3170 * This is the "real" nfs::bwrite(struct buf*). 3171 * We set B_CACHE if this is a VMIO buffer. 3172 */ 3173 int 3174 ncl_writebp(struct buf *bp, int force __unused, struct thread *td) 3175 { 3176 int oldflags, rtval; 3177 3178 BUF_ASSERT_HELD(bp); 3179 3180 if (bp->b_flags & B_INVAL) { 3181 brelse(bp); 3182 return (0); 3183 } 3184 3185 oldflags = bp->b_flags; 3186 bp->b_flags |= B_CACHE; 3187 3188 /* 3189 * Undirty the bp. We will redirty it later if the I/O fails. 3190 */ 3191 bundirty(bp); 3192 bp->b_flags &= ~B_DONE; 3193 bp->b_ioflags &= ~BIO_ERROR; 3194 bp->b_iocmd = BIO_WRITE; 3195 3196 bufobj_wref(bp->b_bufobj); 3197 curthread->td_ru.ru_oublock++; 3198 3199 /* 3200 * Note: to avoid loopback deadlocks, we do not 3201 * assign b_runningbufspace. 3202 */ 3203 vfs_busy_pages(bp, 1); 3204 3205 BUF_KERNPROC(bp); 3206 bp->b_iooffset = dbtob(bp->b_blkno); 3207 bstrategy(bp); 3208 3209 if ((oldflags & B_ASYNC) != 0) 3210 return (0); 3211 3212 rtval = bufwait(bp); 3213 if (oldflags & B_DELWRI) 3214 reassignbuf(bp); 3215 brelse(bp); 3216 return (rtval); 3217 } 3218 3219 /* 3220 * nfs special file access vnode op. 3221 * Essentially just get vattr and then imitate iaccess() since the device is 3222 * local to the client. 3223 */ 3224 static int 3225 nfsspec_access(struct vop_access_args *ap) 3226 { 3227 struct vattr *vap; 3228 struct ucred *cred = ap->a_cred; 3229 struct vnode *vp = ap->a_vp; 3230 accmode_t accmode = ap->a_accmode; 3231 struct vattr vattr; 3232 int error; 3233 3234 /* 3235 * Disallow write attempts on filesystems mounted read-only; 3236 * unless the file is a socket, fifo, or a block or character 3237 * device resident on the filesystem. 3238 */ 3239 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3240 switch (vp->v_type) { 3241 case VREG: 3242 case VDIR: 3243 case VLNK: 3244 return (EROFS); 3245 default: 3246 break; 3247 } 3248 } 3249 vap = &vattr; 3250 error = VOP_GETATTR(vp, vap, cred); 3251 if (error) 3252 goto out; 3253 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid, 3254 accmode, cred, NULL); 3255 out: 3256 return error; 3257 } 3258 3259 /* 3260 * Read wrapper for fifos. 3261 */ 3262 static int 3263 nfsfifo_read(struct vop_read_args *ap) 3264 { 3265 struct nfsnode *np = VTONFS(ap->a_vp); 3266 int error; 3267 3268 /* 3269 * Set access flag. 3270 */ 3271 mtx_lock(&np->n_mtx); 3272 np->n_flag |= NACC; 3273 vfs_timestamp(&np->n_atim); 3274 mtx_unlock(&np->n_mtx); 3275 error = fifo_specops.vop_read(ap); 3276 return error; 3277 } 3278 3279 /* 3280 * Write wrapper for fifos. 3281 */ 3282 static int 3283 nfsfifo_write(struct vop_write_args *ap) 3284 { 3285 struct nfsnode *np = VTONFS(ap->a_vp); 3286 3287 /* 3288 * Set update flag. 3289 */ 3290 mtx_lock(&np->n_mtx); 3291 np->n_flag |= NUPD; 3292 vfs_timestamp(&np->n_mtim); 3293 mtx_unlock(&np->n_mtx); 3294 return(fifo_specops.vop_write(ap)); 3295 } 3296 3297 /* 3298 * Close wrapper for fifos. 3299 * 3300 * Update the times on the nfsnode then do fifo close. 3301 */ 3302 static int 3303 nfsfifo_close(struct vop_close_args *ap) 3304 { 3305 struct vnode *vp = ap->a_vp; 3306 struct nfsnode *np = VTONFS(vp); 3307 struct vattr vattr; 3308 struct timespec ts; 3309 3310 mtx_lock(&np->n_mtx); 3311 if (np->n_flag & (NACC | NUPD)) { 3312 vfs_timestamp(&ts); 3313 if (np->n_flag & NACC) 3314 np->n_atim = ts; 3315 if (np->n_flag & NUPD) 3316 np->n_mtim = ts; 3317 np->n_flag |= NCHG; 3318 if (vrefcnt(vp) == 1 && 3319 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3320 VATTR_NULL(&vattr); 3321 if (np->n_flag & NACC) 3322 vattr.va_atime = np->n_atim; 3323 if (np->n_flag & NUPD) 3324 vattr.va_mtime = np->n_mtim; 3325 mtx_unlock(&np->n_mtx); 3326 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3327 goto out; 3328 } 3329 } 3330 mtx_unlock(&np->n_mtx); 3331 out: 3332 return (fifo_specops.vop_close(ap)); 3333 } 3334 3335 /* 3336 * Just call ncl_writebp() with the force argument set to 1. 3337 * 3338 * NOTE: B_DONE may or may not be set in a_bp on call. 3339 */ 3340 static int 3341 nfs_bwrite(struct buf *bp) 3342 { 3343 3344 return (ncl_writebp(bp, 1, curthread)); 3345 } 3346 3347 struct buf_ops buf_ops_newnfs = { 3348 .bop_name = "buf_ops_nfs", 3349 .bop_write = nfs_bwrite, 3350 .bop_strategy = bufstrategy, 3351 .bop_sync = bufsync, 3352 .bop_bdflush = bufbdflush, 3353 }; 3354 3355 static int 3356 nfs_getacl(struct vop_getacl_args *ap) 3357 { 3358 int error; 3359 3360 if (ap->a_type != ACL_TYPE_NFS4) 3361 return (EOPNOTSUPP); 3362 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3363 NULL); 3364 if (error > NFSERR_STALE) { 3365 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3366 error = EPERM; 3367 } 3368 return (error); 3369 } 3370 3371 static int 3372 nfs_setacl(struct vop_setacl_args *ap) 3373 { 3374 int error; 3375 3376 if (ap->a_type != ACL_TYPE_NFS4) 3377 return (EOPNOTSUPP); 3378 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3379 NULL); 3380 if (error > NFSERR_STALE) { 3381 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3382 error = EPERM; 3383 } 3384 return (error); 3385 } 3386 3387 static int 3388 nfs_set_text(struct vop_set_text_args *ap) 3389 { 3390 struct vnode *vp = ap->a_vp; 3391 struct nfsnode *np; 3392 3393 /* 3394 * If the text file has been mmap'd, flush any dirty pages to the 3395 * buffer cache and then... 3396 * Make sure all writes are pushed to the NFS server. If this is not 3397 * done, the modify time of the file can change while the text 3398 * file is being executed. This will cause the process that is 3399 * executing the text file to be terminated. 3400 */ 3401 if (vp->v_object != NULL) { 3402 VM_OBJECT_WLOCK(vp->v_object); 3403 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC); 3404 VM_OBJECT_WUNLOCK(vp->v_object); 3405 } 3406 3407 /* Now, flush the buffer cache. */ 3408 ncl_flush(vp, MNT_WAIT, curthread, 0, 0); 3409 3410 /* And, finally, make sure that n_mtime is up to date. */ 3411 np = VTONFS(vp); 3412 mtx_lock(&np->n_mtx); 3413 np->n_mtime = np->n_vattr.na_mtime; 3414 mtx_unlock(&np->n_mtx); 3415 3416 vp->v_vflag |= VV_TEXT; 3417 return (0); 3418 } 3419 3420 /* 3421 * Return POSIX pathconf information applicable to nfs filesystems. 3422 */ 3423 static int 3424 nfs_pathconf(struct vop_pathconf_args *ap) 3425 { 3426 struct nfsv3_pathconf pc; 3427 struct nfsvattr nfsva; 3428 struct vnode *vp = ap->a_vp; 3429 struct thread *td = curthread; 3430 int attrflag, error; 3431 3432 if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX || 3433 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED || 3434 ap->a_name == _PC_NO_TRUNC)) || 3435 (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) { 3436 /* 3437 * Since only the above 4 a_names are returned by the NFSv3 3438 * Pathconf RPC, there is no point in doing it for others. 3439 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can 3440 * be used for _PC_NFS4_ACL as well. 3441 */ 3442 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva, 3443 &attrflag, NULL); 3444 if (attrflag != 0) 3445 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 3446 1); 3447 if (error != 0) 3448 return (error); 3449 } else { 3450 /* 3451 * For NFSv2 (or NFSv3 when not one of the above 4 a_names), 3452 * just fake them. 3453 */ 3454 pc.pc_linkmax = NFS_LINK_MAX; 3455 pc.pc_namemax = NFS_MAXNAMLEN; 3456 pc.pc_notrunc = 1; 3457 pc.pc_chownrestricted = 1; 3458 pc.pc_caseinsensitive = 0; 3459 pc.pc_casepreserving = 1; 3460 error = 0; 3461 } 3462 switch (ap->a_name) { 3463 case _PC_LINK_MAX: 3464 #ifdef _LP64 3465 *ap->a_retval = pc.pc_linkmax; 3466 #else 3467 *ap->a_retval = MIN(LONG_MAX, pc.pc_linkmax); 3468 #endif 3469 break; 3470 case _PC_NAME_MAX: 3471 *ap->a_retval = pc.pc_namemax; 3472 break; 3473 case _PC_PIPE_BUF: 3474 if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO) 3475 *ap->a_retval = PIPE_BUF; 3476 else 3477 error = EINVAL; 3478 break; 3479 case _PC_CHOWN_RESTRICTED: 3480 *ap->a_retval = pc.pc_chownrestricted; 3481 break; 3482 case _PC_NO_TRUNC: 3483 *ap->a_retval = pc.pc_notrunc; 3484 break; 3485 case _PC_ACL_EXTENDED: 3486 *ap->a_retval = 0; 3487 break; 3488 case _PC_ACL_NFS4: 3489 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 && 3490 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL)) 3491 *ap->a_retval = 1; 3492 else 3493 *ap->a_retval = 0; 3494 break; 3495 case _PC_ACL_PATH_MAX: 3496 if (NFS_ISV4(vp)) 3497 *ap->a_retval = ACL_MAX_ENTRIES; 3498 else 3499 *ap->a_retval = 3; 3500 break; 3501 case _PC_MAC_PRESENT: 3502 *ap->a_retval = 0; 3503 break; 3504 case _PC_PRIO_IO: 3505 *ap->a_retval = 0; 3506 break; 3507 case _PC_SYNC_IO: 3508 *ap->a_retval = 0; 3509 break; 3510 case _PC_ALLOC_SIZE_MIN: 3511 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize; 3512 break; 3513 case _PC_FILESIZEBITS: 3514 if (NFS_ISV34(vp)) 3515 *ap->a_retval = 64; 3516 else 3517 *ap->a_retval = 32; 3518 break; 3519 case _PC_REC_INCR_XFER_SIZE: 3520 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 3521 break; 3522 case _PC_REC_MAX_XFER_SIZE: 3523 *ap->a_retval = -1; /* means ``unlimited'' */ 3524 break; 3525 case _PC_REC_MIN_XFER_SIZE: 3526 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 3527 break; 3528 case _PC_REC_XFER_ALIGN: 3529 *ap->a_retval = PAGE_SIZE; 3530 break; 3531 case _PC_SYMLINK_MAX: 3532 *ap->a_retval = NFS_MAXPATHLEN; 3533 break; 3534 3535 default: 3536 error = vop_stdpathconf(ap); 3537 break; 3538 } 3539 return (error); 3540 } 3541 3542