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; 145 146 /* 147 * Global vfs data structures for nfs 148 */ 149 150 static struct vop_vector newnfs_vnodeops_nosig = { 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 }; 183 184 static int 185 nfs_vnodeops_bypass(struct vop_generic_args *a) 186 { 187 188 return (vop_sigdefer(&newnfs_vnodeops_nosig, a)); 189 } 190 191 struct vop_vector newnfs_vnodeops = { 192 .vop_default = &default_vnodeops, 193 .vop_bypass = nfs_vnodeops_bypass, 194 }; 195 196 static struct vop_vector newnfs_fifoops_nosig = { 197 .vop_default = &fifo_specops, 198 .vop_access = nfsspec_access, 199 .vop_close = nfsfifo_close, 200 .vop_fsync = nfs_fsync, 201 .vop_getattr = nfs_getattr, 202 .vop_inactive = ncl_inactive, 203 .vop_pathconf = nfs_pathconf, 204 .vop_print = nfs_print, 205 .vop_read = nfsfifo_read, 206 .vop_reclaim = ncl_reclaim, 207 .vop_setattr = nfs_setattr, 208 .vop_write = nfsfifo_write, 209 }; 210 211 static int 212 nfs_fifoops_bypass(struct vop_generic_args *a) 213 { 214 215 return (vop_sigdefer(&newnfs_fifoops_nosig, a)); 216 } 217 218 struct vop_vector newnfs_fifoops = { 219 .vop_default = &default_vnodeops, 220 .vop_bypass = nfs_fifoops_bypass, 221 }; 222 223 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, 224 struct componentname *cnp, struct vattr *vap); 225 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 226 int namelen, struct ucred *cred, struct thread *td); 227 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, 228 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp, 229 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td); 230 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp, 231 struct componentname *scnp, struct sillyrename *sp); 232 233 /* 234 * Global variables 235 */ 236 SYSCTL_DECL(_vfs_nfs); 237 238 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO; 239 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 240 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 241 242 static int nfs_prime_access_cache = 0; 243 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW, 244 &nfs_prime_access_cache, 0, 245 "Prime NFS ACCESS cache when fetching attributes"); 246 247 static int newnfs_commit_on_close = 0; 248 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW, 249 &newnfs_commit_on_close, 0, "write+commit on close, else only write"); 250 251 static int nfs_clean_pages_on_close = 1; 252 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW, 253 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close"); 254 255 int newnfs_directio_enable = 0; 256 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW, 257 &newnfs_directio_enable, 0, "Enable NFS directio"); 258 259 int nfs_keep_dirty_on_error; 260 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW, 261 &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned"); 262 263 /* 264 * This sysctl allows other processes to mmap a file that has been opened 265 * O_DIRECT by a process. In general, having processes mmap the file while 266 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow 267 * this by default to prevent DoS attacks - to prevent a malicious user from 268 * opening up files O_DIRECT preventing other users from mmap'ing these 269 * files. "Protected" environments where stricter consistency guarantees are 270 * required can disable this knob. The process that opened the file O_DIRECT 271 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not 272 * meaningful. 273 */ 274 int newnfs_directio_allow_mmap = 1; 275 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW, 276 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens"); 277 278 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \ 279 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \ 280 | NFSACCESS_DELETE | NFSACCESS_LOOKUP) 281 282 /* 283 * SMP Locking Note : 284 * The list of locks after the description of the lock is the ordering 285 * of other locks acquired with the lock held. 286 * np->n_mtx : Protects the fields in the nfsnode. 287 VM Object Lock 288 VI_MTX (acquired indirectly) 289 * nmp->nm_mtx : Protects the fields in the nfsmount. 290 rep->r_mtx 291 * ncl_iod_mutex : Global lock, protects shared nfsiod state. 292 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list. 293 nmp->nm_mtx 294 rep->r_mtx 295 * rep->r_mtx : Protects the fields in an nfsreq. 296 */ 297 298 static int 299 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td, 300 struct ucred *cred, u_int32_t *retmode) 301 { 302 int error = 0, attrflag, i, lrupos; 303 u_int32_t rmode; 304 struct nfsnode *np = VTONFS(vp); 305 struct nfsvattr nfsva; 306 307 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag, 308 &rmode, NULL); 309 if (attrflag) 310 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 311 if (!error) { 312 lrupos = 0; 313 mtx_lock(&np->n_mtx); 314 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 315 if (np->n_accesscache[i].uid == cred->cr_uid) { 316 np->n_accesscache[i].mode = rmode; 317 np->n_accesscache[i].stamp = time_second; 318 break; 319 } 320 if (i > 0 && np->n_accesscache[i].stamp < 321 np->n_accesscache[lrupos].stamp) 322 lrupos = i; 323 } 324 if (i == NFS_ACCESSCACHESIZE) { 325 np->n_accesscache[lrupos].uid = cred->cr_uid; 326 np->n_accesscache[lrupos].mode = rmode; 327 np->n_accesscache[lrupos].stamp = time_second; 328 } 329 mtx_unlock(&np->n_mtx); 330 if (retmode != NULL) 331 *retmode = rmode; 332 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0); 333 } else if (NFS_ISV4(vp)) { 334 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 335 } 336 #ifdef KDTRACE_HOOKS 337 if (error != 0) 338 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0, 339 error); 340 #endif 341 return (error); 342 } 343 344 /* 345 * nfs access vnode op. 346 * For nfs version 2, just return ok. File accesses may fail later. 347 * For nfs version 3, use the access rpc to check accessibility. If file modes 348 * are changed on the server, accesses might still fail later. 349 */ 350 static int 351 nfs_access(struct vop_access_args *ap) 352 { 353 struct vnode *vp = ap->a_vp; 354 int error = 0, i, gotahit; 355 u_int32_t mode, wmode, rmode; 356 int v34 = NFS_ISV34(vp); 357 struct nfsnode *np = VTONFS(vp); 358 359 /* 360 * Disallow write attempts on filesystems mounted read-only; 361 * unless the file is a socket, fifo, or a block or character 362 * device resident on the filesystem. 363 */ 364 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS | 365 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL | 366 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) { 367 switch (vp->v_type) { 368 case VREG: 369 case VDIR: 370 case VLNK: 371 return (EROFS); 372 default: 373 break; 374 } 375 } 376 /* 377 * For nfs v3 or v4, check to see if we have done this recently, and if 378 * so return our cached result instead of making an ACCESS call. 379 * If not, do an access rpc, otherwise you are stuck emulating 380 * ufs_access() locally using the vattr. This may not be correct, 381 * since the server may apply other access criteria such as 382 * client uid-->server uid mapping that we do not know about. 383 */ 384 if (v34) { 385 if (ap->a_accmode & VREAD) 386 mode = NFSACCESS_READ; 387 else 388 mode = 0; 389 if (vp->v_type != VDIR) { 390 if (ap->a_accmode & VWRITE) 391 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 392 if (ap->a_accmode & VAPPEND) 393 mode |= NFSACCESS_EXTEND; 394 if (ap->a_accmode & VEXEC) 395 mode |= NFSACCESS_EXECUTE; 396 if (ap->a_accmode & VDELETE) 397 mode |= NFSACCESS_DELETE; 398 } else { 399 if (ap->a_accmode & VWRITE) 400 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 401 if (ap->a_accmode & VAPPEND) 402 mode |= NFSACCESS_EXTEND; 403 if (ap->a_accmode & VEXEC) 404 mode |= NFSACCESS_LOOKUP; 405 if (ap->a_accmode & VDELETE) 406 mode |= NFSACCESS_DELETE; 407 if (ap->a_accmode & VDELETE_CHILD) 408 mode |= NFSACCESS_MODIFY; 409 } 410 /* XXX safety belt, only make blanket request if caching */ 411 if (nfsaccess_cache_timeout > 0) { 412 wmode = NFSACCESS_READ | NFSACCESS_MODIFY | 413 NFSACCESS_EXTEND | NFSACCESS_EXECUTE | 414 NFSACCESS_DELETE | NFSACCESS_LOOKUP; 415 } else { 416 wmode = mode; 417 } 418 419 /* 420 * Does our cached result allow us to give a definite yes to 421 * this request? 422 */ 423 gotahit = 0; 424 mtx_lock(&np->n_mtx); 425 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 426 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) { 427 if (time_second < (np->n_accesscache[i].stamp 428 + nfsaccess_cache_timeout) && 429 (np->n_accesscache[i].mode & mode) == mode) { 430 NFSINCRGLOBAL(nfsstatsv1.accesscache_hits); 431 gotahit = 1; 432 } 433 break; 434 } 435 } 436 mtx_unlock(&np->n_mtx); 437 #ifdef KDTRACE_HOOKS 438 if (gotahit != 0) 439 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp, 440 ap->a_cred->cr_uid, mode); 441 else 442 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp, 443 ap->a_cred->cr_uid, mode); 444 #endif 445 if (gotahit == 0) { 446 /* 447 * Either a no, or a don't know. Go to the wire. 448 */ 449 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses); 450 error = nfs34_access_otw(vp, wmode, ap->a_td, 451 ap->a_cred, &rmode); 452 if (!error && 453 (rmode & mode) != mode) 454 error = EACCES; 455 } 456 return (error); 457 } else { 458 if ((error = nfsspec_access(ap)) != 0) { 459 return (error); 460 } 461 /* 462 * Attempt to prevent a mapped root from accessing a file 463 * which it shouldn't. We try to read a byte from the file 464 * if the user is root and the file is not zero length. 465 * After calling nfsspec_access, we should have the correct 466 * file size cached. 467 */ 468 mtx_lock(&np->n_mtx); 469 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD) 470 && VTONFS(vp)->n_size > 0) { 471 struct iovec aiov; 472 struct uio auio; 473 char buf[1]; 474 475 mtx_unlock(&np->n_mtx); 476 aiov.iov_base = buf; 477 aiov.iov_len = 1; 478 auio.uio_iov = &aiov; 479 auio.uio_iovcnt = 1; 480 auio.uio_offset = 0; 481 auio.uio_resid = 1; 482 auio.uio_segflg = UIO_SYSSPACE; 483 auio.uio_rw = UIO_READ; 484 auio.uio_td = ap->a_td; 485 486 if (vp->v_type == VREG) 487 error = ncl_readrpc(vp, &auio, ap->a_cred); 488 else if (vp->v_type == VDIR) { 489 char* bp; 490 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 491 aiov.iov_base = bp; 492 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 493 error = ncl_readdirrpc(vp, &auio, ap->a_cred, 494 ap->a_td); 495 free(bp, M_TEMP); 496 } else if (vp->v_type == VLNK) 497 error = ncl_readlinkrpc(vp, &auio, ap->a_cred); 498 else 499 error = EACCES; 500 } else 501 mtx_unlock(&np->n_mtx); 502 return (error); 503 } 504 } 505 506 507 /* 508 * nfs open vnode op 509 * Check to see if the type is ok 510 * and that deletion is not in progress. 511 * For paged in text files, you will need to flush the page cache 512 * if consistency is lost. 513 */ 514 /* ARGSUSED */ 515 static int 516 nfs_open(struct vop_open_args *ap) 517 { 518 struct vnode *vp = ap->a_vp; 519 struct nfsnode *np = VTONFS(vp); 520 struct vattr vattr; 521 int error; 522 int fmode = ap->a_mode; 523 struct ucred *cred; 524 vm_object_t obj; 525 526 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) 527 return (EOPNOTSUPP); 528 529 /* 530 * For NFSv4, we need to do the Open Op before cache validation, 531 * so that we conform to RFC3530 Sec. 9.3.1. 532 */ 533 if (NFS_ISV4(vp)) { 534 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td); 535 if (error) { 536 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 537 (gid_t)0); 538 return (error); 539 } 540 } 541 542 /* 543 * Now, if this Open will be doing reading, re-validate/flush the 544 * cache, so that Close/Open coherency is maintained. 545 */ 546 mtx_lock(&np->n_mtx); 547 if (np->n_flag & NMODIFIED) { 548 mtx_unlock(&np->n_mtx); 549 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 550 if (error == EINTR || error == EIO) { 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 np->n_attrstamp = 0; 557 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 558 if (vp->v_type == VDIR) 559 np->n_direofoffset = 0; 560 mtx_unlock(&np->n_mtx); 561 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 562 if (error) { 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 } else { 572 mtx_unlock(&np->n_mtx); 573 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 574 if (error) { 575 if (NFS_ISV4(vp)) 576 (void) nfsrpc_close(vp, 0, ap->a_td); 577 return (error); 578 } 579 mtx_lock(&np->n_mtx); 580 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) || 581 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 582 if (vp->v_type == VDIR) 583 np->n_direofoffset = 0; 584 mtx_unlock(&np->n_mtx); 585 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 586 if (error == EINTR || error == EIO) { 587 if (NFS_ISV4(vp)) 588 (void) nfsrpc_close(vp, 0, ap->a_td); 589 return (error); 590 } 591 mtx_lock(&np->n_mtx); 592 np->n_mtime = vattr.va_mtime; 593 if (NFS_ISV4(vp)) 594 np->n_change = vattr.va_filerev; 595 } 596 } 597 598 /* 599 * If the object has >= 1 O_DIRECT active opens, we disable caching. 600 */ 601 if (newnfs_directio_enable && (fmode & O_DIRECT) && 602 (vp->v_type == VREG)) { 603 if (np->n_directio_opens == 0) { 604 mtx_unlock(&np->n_mtx); 605 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 606 if (error) { 607 if (NFS_ISV4(vp)) 608 (void) nfsrpc_close(vp, 0, ap->a_td); 609 return (error); 610 } 611 mtx_lock(&np->n_mtx); 612 np->n_flag |= NNONCACHE; 613 } 614 np->n_directio_opens++; 615 } 616 617 /* If opened for writing via NFSv4.1 or later, mark that for pNFS. */ 618 if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0) 619 np->n_flag |= NWRITEOPENED; 620 621 /* 622 * If this is an open for writing, capture a reference to the 623 * credentials, so they can be used by ncl_putpages(). Using 624 * these write credentials is preferable to the credentials of 625 * whatever thread happens to be doing the VOP_PUTPAGES() since 626 * the write RPCs are less likely to fail with EACCES. 627 */ 628 if ((fmode & FWRITE) != 0) { 629 cred = np->n_writecred; 630 np->n_writecred = crhold(ap->a_cred); 631 } else 632 cred = NULL; 633 mtx_unlock(&np->n_mtx); 634 635 if (cred != NULL) 636 crfree(cred); 637 vnode_create_vobject(vp, vattr.va_size, ap->a_td); 638 639 /* 640 * If the text file has been mmap'd, flush any dirty pages to the 641 * buffer cache and then... 642 * Make sure all writes are pushed to the NFS server. If this is not 643 * done, the modify time of the file can change while the text 644 * file is being executed. This will cause the process that is 645 * executing the text file to be terminated. 646 */ 647 if (vp->v_writecount <= -1) { 648 if ((obj = vp->v_object) != NULL && 649 (obj->flags & OBJ_MIGHTBEDIRTY) != 0) { 650 VM_OBJECT_WLOCK(obj); 651 vm_object_page_clean(obj, 0, 0, OBJPC_SYNC); 652 VM_OBJECT_WUNLOCK(obj); 653 } 654 655 /* Now, flush the buffer cache. */ 656 ncl_flush(vp, MNT_WAIT, curthread, 0, 0); 657 658 /* And, finally, make sure that n_mtime is up to date. */ 659 np = VTONFS(vp); 660 mtx_lock(&np->n_mtx); 661 np->n_mtime = np->n_vattr.na_mtime; 662 mtx_unlock(&np->n_mtx); 663 } 664 return (0); 665 } 666 667 /* 668 * nfs close vnode op 669 * What an NFS client should do upon close after writing is a debatable issue. 670 * Most NFS clients push delayed writes to the server upon close, basically for 671 * two reasons: 672 * 1 - So that any write errors may be reported back to the client process 673 * doing the close system call. By far the two most likely errors are 674 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 675 * 2 - To put a worst case upper bound on cache inconsistency between 676 * multiple clients for the file. 677 * There is also a consistency problem for Version 2 of the protocol w.r.t. 678 * not being able to tell if other clients are writing a file concurrently, 679 * since there is no way of knowing if the changed modify time in the reply 680 * is only due to the write for this client. 681 * (NFS Version 3 provides weak cache consistency data in the reply that 682 * should be sufficient to detect and handle this case.) 683 * 684 * The current code does the following: 685 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 686 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 687 * or commit them (this satisfies 1 and 2 except for the 688 * case where the server crashes after this close but 689 * before the commit RPC, which is felt to be "good 690 * enough". Changing the last argument to ncl_flush() to 691 * a 1 would force a commit operation, if it is felt a 692 * commit is necessary now. 693 * for NFS Version 4 - flush the dirty buffers and commit them, if 694 * nfscl_mustflush() says this is necessary. 695 * It is necessary if there is no write delegation held, 696 * in order to satisfy open/close coherency. 697 * If the file isn't cached on local stable storage, 698 * it may be necessary in order to detect "out of space" 699 * errors from the server, if the write delegation 700 * issued by the server doesn't allow the file to grow. 701 */ 702 /* ARGSUSED */ 703 static int 704 nfs_close(struct vop_close_args *ap) 705 { 706 struct vnode *vp = ap->a_vp; 707 struct nfsnode *np = VTONFS(vp); 708 struct nfsvattr nfsva; 709 struct ucred *cred; 710 int error = 0, ret, localcred = 0; 711 int fmode = ap->a_fflag; 712 713 if (NFSCL_FORCEDISM(vp->v_mount)) 714 return (0); 715 /* 716 * During shutdown, a_cred isn't valid, so just use root. 717 */ 718 if (ap->a_cred == NOCRED) { 719 cred = newnfs_getcred(); 720 localcred = 1; 721 } else { 722 cred = ap->a_cred; 723 } 724 if (vp->v_type == VREG) { 725 /* 726 * Examine and clean dirty pages, regardless of NMODIFIED. 727 * This closes a major hole in close-to-open consistency. 728 * We want to push out all dirty pages (and buffers) on 729 * close, regardless of whether they were dirtied by 730 * mmap'ed writes or via write(). 731 */ 732 if (nfs_clean_pages_on_close && vp->v_object) { 733 VM_OBJECT_WLOCK(vp->v_object); 734 vm_object_page_clean(vp->v_object, 0, 0, 0); 735 VM_OBJECT_WUNLOCK(vp->v_object); 736 } 737 mtx_lock(&np->n_mtx); 738 if (np->n_flag & NMODIFIED) { 739 mtx_unlock(&np->n_mtx); 740 if (NFS_ISV3(vp)) { 741 /* 742 * Under NFSv3 we have dirty buffers to dispose of. We 743 * must flush them to the NFS server. We have the option 744 * of waiting all the way through the commit rpc or just 745 * waiting for the initial write. The default is to only 746 * wait through the initial write so the data is in the 747 * server's cache, which is roughly similar to the state 748 * a standard disk subsystem leaves the file in on close(). 749 * 750 * We cannot clear the NMODIFIED bit in np->n_flag due to 751 * potential races with other processes, and certainly 752 * cannot clear it if we don't commit. 753 * These races occur when there is no longer the old 754 * traditional vnode locking implemented for Vnode Ops. 755 */ 756 int cm = newnfs_commit_on_close ? 1 : 0; 757 error = ncl_flush(vp, MNT_WAIT, ap->a_td, cm, 0); 758 /* np->n_flag &= ~NMODIFIED; */ 759 } else if (NFS_ISV4(vp)) { 760 if (nfscl_mustflush(vp) != 0) { 761 int cm = newnfs_commit_on_close ? 1 : 0; 762 error = ncl_flush(vp, MNT_WAIT, ap->a_td, 763 cm, 0); 764 /* 765 * as above w.r.t races when clearing 766 * NMODIFIED. 767 * np->n_flag &= ~NMODIFIED; 768 */ 769 } 770 } else { 771 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 772 } 773 mtx_lock(&np->n_mtx); 774 } 775 /* 776 * Invalidate the attribute cache in all cases. 777 * An open is going to fetch fresh attrs any way, other procs 778 * on this node that have file open will be forced to do an 779 * otw attr fetch, but this is safe. 780 * --> A user found that their RPC count dropped by 20% when 781 * this was commented out and I can't see any requirement 782 * for it, so I've disabled it when negative lookups are 783 * enabled. (What does this have to do with negative lookup 784 * caching? Well nothing, except it was reported by the 785 * same user that needed negative lookup caching and I wanted 786 * there to be a way to disable it to see if it 787 * is the cause of some caching/coherency issue that might 788 * crop up.) 789 */ 790 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) { 791 np->n_attrstamp = 0; 792 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 793 } 794 if (np->n_flag & NWRITEERR) { 795 np->n_flag &= ~NWRITEERR; 796 error = np->n_error; 797 } 798 mtx_unlock(&np->n_mtx); 799 } 800 801 if (NFS_ISV4(vp)) { 802 /* 803 * Get attributes so "change" is up to date. 804 */ 805 if (error == 0 && nfscl_mustflush(vp) != 0 && 806 vp->v_type == VREG && 807 (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) { 808 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva, 809 NULL); 810 if (!ret) { 811 np->n_change = nfsva.na_filerev; 812 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 813 NULL, 0, 0); 814 } 815 } 816 817 /* 818 * and do the close. 819 */ 820 ret = nfsrpc_close(vp, 0, ap->a_td); 821 if (!error && ret) 822 error = ret; 823 if (error) 824 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 825 (gid_t)0); 826 } 827 if (newnfs_directio_enable) 828 KASSERT((np->n_directio_asyncwr == 0), 829 ("nfs_close: dirty unflushed (%d) directio buffers\n", 830 np->n_directio_asyncwr)); 831 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) { 832 mtx_lock(&np->n_mtx); 833 KASSERT((np->n_directio_opens > 0), 834 ("nfs_close: unexpectedly value (0) of n_directio_opens\n")); 835 np->n_directio_opens--; 836 if (np->n_directio_opens == 0) 837 np->n_flag &= ~NNONCACHE; 838 mtx_unlock(&np->n_mtx); 839 } 840 if (localcred) 841 NFSFREECRED(cred); 842 return (error); 843 } 844 845 /* 846 * nfs getattr call from vfs. 847 */ 848 static int 849 nfs_getattr(struct vop_getattr_args *ap) 850 { 851 struct vnode *vp = ap->a_vp; 852 struct thread *td = curthread; /* XXX */ 853 struct nfsnode *np = VTONFS(vp); 854 int error = 0; 855 struct nfsvattr nfsva; 856 struct vattr *vap = ap->a_vap; 857 struct vattr vattr; 858 859 /* 860 * Update local times for special files. 861 */ 862 mtx_lock(&np->n_mtx); 863 if (np->n_flag & (NACC | NUPD)) 864 np->n_flag |= NCHG; 865 mtx_unlock(&np->n_mtx); 866 /* 867 * First look in the cache. 868 */ 869 if (ncl_getattrcache(vp, &vattr) == 0) { 870 vap->va_type = vattr.va_type; 871 vap->va_mode = vattr.va_mode; 872 vap->va_nlink = vattr.va_nlink; 873 vap->va_uid = vattr.va_uid; 874 vap->va_gid = vattr.va_gid; 875 vap->va_fsid = vattr.va_fsid; 876 vap->va_fileid = vattr.va_fileid; 877 vap->va_size = vattr.va_size; 878 vap->va_blocksize = vattr.va_blocksize; 879 vap->va_atime = vattr.va_atime; 880 vap->va_mtime = vattr.va_mtime; 881 vap->va_ctime = vattr.va_ctime; 882 vap->va_gen = vattr.va_gen; 883 vap->va_flags = vattr.va_flags; 884 vap->va_rdev = vattr.va_rdev; 885 vap->va_bytes = vattr.va_bytes; 886 vap->va_filerev = vattr.va_filerev; 887 /* 888 * Get the local modify time for the case of a write 889 * delegation. 890 */ 891 nfscl_deleggetmodtime(vp, &vap->va_mtime); 892 return (0); 893 } 894 895 if (NFS_ISV34(vp) && nfs_prime_access_cache && 896 nfsaccess_cache_timeout > 0) { 897 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses); 898 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL); 899 if (ncl_getattrcache(vp, ap->a_vap) == 0) { 900 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime); 901 return (0); 902 } 903 } 904 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL); 905 if (!error) 906 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0); 907 if (!error) { 908 /* 909 * Get the local modify time for the case of a write 910 * delegation. 911 */ 912 nfscl_deleggetmodtime(vp, &vap->va_mtime); 913 } else if (NFS_ISV4(vp)) { 914 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 915 } 916 return (error); 917 } 918 919 /* 920 * nfs setattr call. 921 */ 922 static int 923 nfs_setattr(struct vop_setattr_args *ap) 924 { 925 struct vnode *vp = ap->a_vp; 926 struct nfsnode *np = VTONFS(vp); 927 struct thread *td = curthread; /* XXX */ 928 struct vattr *vap = ap->a_vap; 929 int error = 0; 930 u_quad_t tsize; 931 932 #ifndef nolint 933 tsize = (u_quad_t)0; 934 #endif 935 936 /* 937 * Setting of flags and marking of atimes are not supported. 938 */ 939 if (vap->va_flags != VNOVAL) 940 return (EOPNOTSUPP); 941 942 /* 943 * Disallow write attempts if the filesystem is mounted read-only. 944 */ 945 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 946 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 947 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 948 (vp->v_mount->mnt_flag & MNT_RDONLY)) 949 return (EROFS); 950 if (vap->va_size != VNOVAL) { 951 switch (vp->v_type) { 952 case VDIR: 953 return (EISDIR); 954 case VCHR: 955 case VBLK: 956 case VSOCK: 957 case VFIFO: 958 if (vap->va_mtime.tv_sec == VNOVAL && 959 vap->va_atime.tv_sec == VNOVAL && 960 vap->va_mode == (mode_t)VNOVAL && 961 vap->va_uid == (uid_t)VNOVAL && 962 vap->va_gid == (gid_t)VNOVAL) 963 return (0); 964 vap->va_size = VNOVAL; 965 break; 966 default: 967 /* 968 * Disallow write attempts if the filesystem is 969 * mounted read-only. 970 */ 971 if (vp->v_mount->mnt_flag & MNT_RDONLY) 972 return (EROFS); 973 /* 974 * We run vnode_pager_setsize() early (why?), 975 * we must set np->n_size now to avoid vinvalbuf 976 * V_SAVE races that might setsize a lower 977 * value. 978 */ 979 mtx_lock(&np->n_mtx); 980 tsize = np->n_size; 981 mtx_unlock(&np->n_mtx); 982 error = ncl_meta_setsize(vp, ap->a_cred, td, 983 vap->va_size); 984 mtx_lock(&np->n_mtx); 985 if (np->n_flag & NMODIFIED) { 986 tsize = np->n_size; 987 mtx_unlock(&np->n_mtx); 988 error = ncl_vinvalbuf(vp, vap->va_size == 0 ? 989 0 : V_SAVE, td, 1); 990 if (error != 0) { 991 vnode_pager_setsize(vp, tsize); 992 return (error); 993 } 994 /* 995 * Call nfscl_delegmodtime() to set the modify time 996 * locally, as required. 997 */ 998 nfscl_delegmodtime(vp); 999 } else 1000 mtx_unlock(&np->n_mtx); 1001 /* 1002 * np->n_size has already been set to vap->va_size 1003 * in ncl_meta_setsize(). We must set it again since 1004 * nfs_loadattrcache() could be called through 1005 * ncl_meta_setsize() and could modify np->n_size. 1006 */ 1007 mtx_lock(&np->n_mtx); 1008 np->n_vattr.na_size = np->n_size = vap->va_size; 1009 mtx_unlock(&np->n_mtx); 1010 } 1011 } else { 1012 mtx_lock(&np->n_mtx); 1013 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) && 1014 (np->n_flag & NMODIFIED) && vp->v_type == VREG) { 1015 mtx_unlock(&np->n_mtx); 1016 error = ncl_vinvalbuf(vp, V_SAVE, td, 1); 1017 if (error == EINTR || error == EIO) 1018 return (error); 1019 } else 1020 mtx_unlock(&np->n_mtx); 1021 } 1022 error = nfs_setattrrpc(vp, vap, ap->a_cred, td); 1023 if (error && vap->va_size != VNOVAL) { 1024 mtx_lock(&np->n_mtx); 1025 np->n_size = np->n_vattr.na_size = tsize; 1026 vnode_pager_setsize(vp, tsize); 1027 mtx_unlock(&np->n_mtx); 1028 } 1029 return (error); 1030 } 1031 1032 /* 1033 * Do an nfs setattr rpc. 1034 */ 1035 static int 1036 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred, 1037 struct thread *td) 1038 { 1039 struct nfsnode *np = VTONFS(vp); 1040 int error, ret, attrflag, i; 1041 struct nfsvattr nfsva; 1042 1043 if (NFS_ISV34(vp)) { 1044 mtx_lock(&np->n_mtx); 1045 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) 1046 np->n_accesscache[i].stamp = 0; 1047 np->n_flag |= NDELEGMOD; 1048 mtx_unlock(&np->n_mtx); 1049 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp); 1050 } 1051 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag, 1052 NULL); 1053 if (attrflag) { 1054 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1055 if (ret && !error) 1056 error = ret; 1057 } 1058 if (error && NFS_ISV4(vp)) 1059 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid); 1060 return (error); 1061 } 1062 1063 /* 1064 * nfs lookup call, one step at a time... 1065 * First look in cache 1066 * If not found, unlock the directory nfsnode and do the rpc 1067 */ 1068 static int 1069 nfs_lookup(struct vop_lookup_args *ap) 1070 { 1071 struct componentname *cnp = ap->a_cnp; 1072 struct vnode *dvp = ap->a_dvp; 1073 struct vnode **vpp = ap->a_vpp; 1074 struct mount *mp = dvp->v_mount; 1075 int flags = cnp->cn_flags; 1076 struct vnode *newvp; 1077 struct nfsmount *nmp; 1078 struct nfsnode *np, *newnp; 1079 int error = 0, attrflag, dattrflag, ltype, ncticks; 1080 struct thread *td = cnp->cn_thread; 1081 struct nfsfh *nfhp; 1082 struct nfsvattr dnfsva, nfsva; 1083 struct vattr vattr; 1084 struct timespec nctime; 1085 1086 *vpp = NULLVP; 1087 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) && 1088 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 1089 return (EROFS); 1090 if (dvp->v_type != VDIR) 1091 return (ENOTDIR); 1092 nmp = VFSTONFS(mp); 1093 np = VTONFS(dvp); 1094 1095 /* For NFSv4, wait until any remove is done. */ 1096 mtx_lock(&np->n_mtx); 1097 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) { 1098 np->n_flag |= NREMOVEWANT; 1099 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0); 1100 } 1101 mtx_unlock(&np->n_mtx); 1102 1103 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) 1104 return (error); 1105 error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks); 1106 if (error > 0 && error != ENOENT) 1107 return (error); 1108 if (error == -1) { 1109 /* 1110 * Lookups of "." are special and always return the 1111 * current directory. cache_lookup() already handles 1112 * associated locking bookkeeping, etc. 1113 */ 1114 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 1115 /* XXX: Is this really correct? */ 1116 if (cnp->cn_nameiop != LOOKUP && 1117 (flags & ISLASTCN)) 1118 cnp->cn_flags |= SAVENAME; 1119 return (0); 1120 } 1121 1122 /* 1123 * We only accept a positive hit in the cache if the 1124 * change time of the file matches our cached copy. 1125 * Otherwise, we discard the cache entry and fallback 1126 * to doing a lookup RPC. We also only trust cache 1127 * entries for less than nm_nametimeo seconds. 1128 * 1129 * To better handle stale file handles and attributes, 1130 * clear the attribute cache of this node if it is a 1131 * leaf component, part of an open() call, and not 1132 * locally modified before fetching the attributes. 1133 * This should allow stale file handles to be detected 1134 * here where we can fall back to a LOOKUP RPC to 1135 * recover rather than having nfs_open() detect the 1136 * stale file handle and failing open(2) with ESTALE. 1137 */ 1138 newvp = *vpp; 1139 newnp = VTONFS(newvp); 1140 if (!(nmp->nm_flag & NFSMNT_NOCTO) && 1141 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1142 !(newnp->n_flag & NMODIFIED)) { 1143 mtx_lock(&newnp->n_mtx); 1144 newnp->n_attrstamp = 0; 1145 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1146 mtx_unlock(&newnp->n_mtx); 1147 } 1148 if (nfscl_nodeleg(newvp, 0) == 0 || 1149 ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) && 1150 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 && 1151 timespeccmp(&vattr.va_ctime, &nctime, ==))) { 1152 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits); 1153 if (cnp->cn_nameiop != LOOKUP && 1154 (flags & ISLASTCN)) 1155 cnp->cn_flags |= SAVENAME; 1156 return (0); 1157 } 1158 cache_purge(newvp); 1159 if (dvp != newvp) 1160 vput(newvp); 1161 else 1162 vrele(newvp); 1163 *vpp = NULLVP; 1164 } else if (error == ENOENT) { 1165 if (dvp->v_iflag & VI_DOOMED) 1166 return (ENOENT); 1167 /* 1168 * We only accept a negative hit in the cache if the 1169 * modification time of the parent directory matches 1170 * the cached copy in the name cache entry. 1171 * Otherwise, we discard all of the negative cache 1172 * entries for this directory. We also only trust 1173 * negative cache entries for up to nm_negnametimeo 1174 * seconds. 1175 */ 1176 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) && 1177 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 && 1178 timespeccmp(&vattr.va_mtime, &nctime, ==)) { 1179 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits); 1180 return (ENOENT); 1181 } 1182 cache_purge_negative(dvp); 1183 } 1184 1185 error = 0; 1186 newvp = NULLVP; 1187 NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses); 1188 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1189 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1190 NULL); 1191 if (dattrflag) 1192 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1193 if (error) { 1194 if (newvp != NULLVP) { 1195 vput(newvp); 1196 *vpp = NULLVP; 1197 } 1198 1199 if (error != ENOENT) { 1200 if (NFS_ISV4(dvp)) 1201 error = nfscl_maperr(td, error, (uid_t)0, 1202 (gid_t)0); 1203 return (error); 1204 } 1205 1206 /* The requested file was not found. */ 1207 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1208 (flags & ISLASTCN)) { 1209 /* 1210 * XXX: UFS does a full VOP_ACCESS(dvp, 1211 * VWRITE) here instead of just checking 1212 * MNT_RDONLY. 1213 */ 1214 if (mp->mnt_flag & MNT_RDONLY) 1215 return (EROFS); 1216 cnp->cn_flags |= SAVENAME; 1217 return (EJUSTRETURN); 1218 } 1219 1220 if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) { 1221 /* 1222 * Cache the modification time of the parent 1223 * directory from the post-op attributes in 1224 * the name cache entry. The negative cache 1225 * entry will be ignored once the directory 1226 * has changed. Don't bother adding the entry 1227 * if the directory has already changed. 1228 */ 1229 mtx_lock(&np->n_mtx); 1230 if (timespeccmp(&np->n_vattr.na_mtime, 1231 &dnfsva.na_mtime, ==)) { 1232 mtx_unlock(&np->n_mtx); 1233 cache_enter_time(dvp, NULL, cnp, 1234 &dnfsva.na_mtime, NULL); 1235 } else 1236 mtx_unlock(&np->n_mtx); 1237 } 1238 return (ENOENT); 1239 } 1240 1241 /* 1242 * Handle RENAME case... 1243 */ 1244 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 1245 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1246 free(nfhp, M_NFSFH); 1247 return (EISDIR); 1248 } 1249 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1250 LK_EXCLUSIVE); 1251 if (error) 1252 return (error); 1253 newvp = NFSTOV(np); 1254 if (attrflag) 1255 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1256 0, 1); 1257 *vpp = newvp; 1258 cnp->cn_flags |= SAVENAME; 1259 return (0); 1260 } 1261 1262 if (flags & ISDOTDOT) { 1263 ltype = NFSVOPISLOCKED(dvp); 1264 error = vfs_busy(mp, MBF_NOWAIT); 1265 if (error != 0) { 1266 vfs_ref(mp); 1267 NFSVOPUNLOCK(dvp, 0); 1268 error = vfs_busy(mp, 0); 1269 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1270 vfs_rel(mp); 1271 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) { 1272 vfs_unbusy(mp); 1273 error = ENOENT; 1274 } 1275 if (error != 0) 1276 return (error); 1277 } 1278 NFSVOPUNLOCK(dvp, 0); 1279 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1280 cnp->cn_lkflags); 1281 if (error == 0) 1282 newvp = NFSTOV(np); 1283 vfs_unbusy(mp); 1284 if (newvp != dvp) 1285 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1286 if (dvp->v_iflag & VI_DOOMED) { 1287 if (error == 0) { 1288 if (newvp == dvp) 1289 vrele(newvp); 1290 else 1291 vput(newvp); 1292 } 1293 error = ENOENT; 1294 } 1295 if (error != 0) 1296 return (error); 1297 if (attrflag) 1298 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1299 0, 1); 1300 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1301 free(nfhp, M_NFSFH); 1302 VREF(dvp); 1303 newvp = dvp; 1304 if (attrflag) 1305 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1306 0, 1); 1307 } else { 1308 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1309 cnp->cn_lkflags); 1310 if (error) 1311 return (error); 1312 newvp = NFSTOV(np); 1313 if (attrflag) 1314 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1315 0, 1); 1316 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1317 !(np->n_flag & NMODIFIED)) { 1318 /* 1319 * Flush the attribute cache when opening a 1320 * leaf node to ensure that fresh attributes 1321 * are fetched in nfs_open() since we did not 1322 * fetch attributes from the LOOKUP reply. 1323 */ 1324 mtx_lock(&np->n_mtx); 1325 np->n_attrstamp = 0; 1326 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1327 mtx_unlock(&np->n_mtx); 1328 } 1329 } 1330 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1331 cnp->cn_flags |= SAVENAME; 1332 if ((cnp->cn_flags & MAKEENTRY) && 1333 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) && 1334 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0)) 1335 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 1336 newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime); 1337 *vpp = newvp; 1338 return (0); 1339 } 1340 1341 /* 1342 * nfs read call. 1343 * Just call ncl_bioread() to do the work. 1344 */ 1345 static int 1346 nfs_read(struct vop_read_args *ap) 1347 { 1348 struct vnode *vp = ap->a_vp; 1349 1350 switch (vp->v_type) { 1351 case VREG: 1352 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred)); 1353 case VDIR: 1354 return (EISDIR); 1355 default: 1356 return (EOPNOTSUPP); 1357 } 1358 } 1359 1360 /* 1361 * nfs readlink call 1362 */ 1363 static int 1364 nfs_readlink(struct vop_readlink_args *ap) 1365 { 1366 struct vnode *vp = ap->a_vp; 1367 1368 if (vp->v_type != VLNK) 1369 return (EINVAL); 1370 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred)); 1371 } 1372 1373 /* 1374 * Do a readlink rpc. 1375 * Called by ncl_doio() from below the buffer cache. 1376 */ 1377 int 1378 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1379 { 1380 int error, ret, attrflag; 1381 struct nfsvattr nfsva; 1382 1383 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva, 1384 &attrflag, NULL); 1385 if (attrflag) { 1386 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1387 if (ret && !error) 1388 error = ret; 1389 } 1390 if (error && NFS_ISV4(vp)) 1391 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1392 return (error); 1393 } 1394 1395 /* 1396 * nfs read rpc call 1397 * Ditto above 1398 */ 1399 int 1400 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1401 { 1402 int error, ret, attrflag; 1403 struct nfsvattr nfsva; 1404 struct nfsmount *nmp; 1405 1406 nmp = VFSTONFS(vnode_mount(vp)); 1407 error = EIO; 1408 attrflag = 0; 1409 if (NFSHASPNFS(nmp)) 1410 error = nfscl_doiods(vp, uiop, NULL, NULL, 1411 NFSV4OPEN_ACCESSREAD, 0, cred, uiop->uio_td); 1412 NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error); 1413 if (error != 0) 1414 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, 1415 &attrflag, NULL); 1416 if (attrflag) { 1417 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1418 if (ret && !error) 1419 error = ret; 1420 } 1421 if (error && NFS_ISV4(vp)) 1422 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1423 return (error); 1424 } 1425 1426 /* 1427 * nfs write call 1428 */ 1429 int 1430 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 1431 int *iomode, int *must_commit, int called_from_strategy) 1432 { 1433 struct nfsvattr nfsva; 1434 int error, attrflag, ret; 1435 struct nfsmount *nmp; 1436 1437 nmp = VFSTONFS(vnode_mount(vp)); 1438 error = EIO; 1439 attrflag = 0; 1440 if (NFSHASPNFS(nmp)) 1441 error = nfscl_doiods(vp, uiop, iomode, must_commit, 1442 NFSV4OPEN_ACCESSWRITE, 0, cred, uiop->uio_td); 1443 NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error); 1444 if (error != 0) 1445 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred, 1446 uiop->uio_td, &nfsva, &attrflag, NULL, 1447 called_from_strategy); 1448 if (attrflag) { 1449 if (VTONFS(vp)->n_flag & ND_NFSV4) 1450 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1, 1451 1); 1452 else 1453 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1454 1); 1455 if (ret && !error) 1456 error = ret; 1457 } 1458 if (DOINGASYNC(vp)) 1459 *iomode = NFSWRITE_FILESYNC; 1460 if (error && NFS_ISV4(vp)) 1461 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1462 return (error); 1463 } 1464 1465 /* 1466 * nfs mknod rpc 1467 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1468 * mode set to specify the file type and the size field for rdev. 1469 */ 1470 static int 1471 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1472 struct vattr *vap) 1473 { 1474 struct nfsvattr nfsva, dnfsva; 1475 struct vnode *newvp = NULL; 1476 struct nfsnode *np = NULL, *dnp; 1477 struct nfsfh *nfhp; 1478 struct vattr vattr; 1479 int error = 0, attrflag, dattrflag; 1480 u_int32_t rdev; 1481 1482 if (vap->va_type == VCHR || vap->va_type == VBLK) 1483 rdev = vap->va_rdev; 1484 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1485 rdev = 0xffffffff; 1486 else 1487 return (EOPNOTSUPP); 1488 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1489 return (error); 1490 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap, 1491 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva, 1492 &nfsva, &nfhp, &attrflag, &dattrflag, NULL); 1493 if (!error) { 1494 if (!nfhp) 1495 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1496 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, 1497 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1498 NULL); 1499 if (nfhp) 1500 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1501 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); 1502 } 1503 if (dattrflag) 1504 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1505 if (!error) { 1506 newvp = NFSTOV(np); 1507 if (attrflag != 0) { 1508 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1509 0, 1); 1510 if (error != 0) 1511 vput(newvp); 1512 } 1513 } 1514 if (!error) { 1515 *vpp = newvp; 1516 } else if (NFS_ISV4(dvp)) { 1517 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, 1518 vap->va_gid); 1519 } 1520 dnp = VTONFS(dvp); 1521 mtx_lock(&dnp->n_mtx); 1522 dnp->n_flag |= NMODIFIED; 1523 if (!dattrflag) { 1524 dnp->n_attrstamp = 0; 1525 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1526 } 1527 mtx_unlock(&dnp->n_mtx); 1528 return (error); 1529 } 1530 1531 /* 1532 * nfs mknod vop 1533 * just call nfs_mknodrpc() to do the work. 1534 */ 1535 /* ARGSUSED */ 1536 static int 1537 nfs_mknod(struct vop_mknod_args *ap) 1538 { 1539 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap)); 1540 } 1541 1542 static struct mtx nfs_cverf_mtx; 1543 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex", 1544 MTX_DEF); 1545 1546 static nfsquad_t 1547 nfs_get_cverf(void) 1548 { 1549 static nfsquad_t cverf; 1550 nfsquad_t ret; 1551 static int cverf_initialized = 0; 1552 1553 mtx_lock(&nfs_cverf_mtx); 1554 if (cverf_initialized == 0) { 1555 cverf.lval[0] = arc4random(); 1556 cverf.lval[1] = arc4random(); 1557 cverf_initialized = 1; 1558 } else 1559 cverf.qval++; 1560 ret = cverf; 1561 mtx_unlock(&nfs_cverf_mtx); 1562 1563 return (ret); 1564 } 1565 1566 /* 1567 * nfs file create call 1568 */ 1569 static int 1570 nfs_create(struct vop_create_args *ap) 1571 { 1572 struct vnode *dvp = ap->a_dvp; 1573 struct vattr *vap = ap->a_vap; 1574 struct componentname *cnp = ap->a_cnp; 1575 struct nfsnode *np = NULL, *dnp; 1576 struct vnode *newvp = NULL; 1577 struct nfsmount *nmp; 1578 struct nfsvattr dnfsva, nfsva; 1579 struct nfsfh *nfhp; 1580 nfsquad_t cverf; 1581 int error = 0, attrflag, dattrflag, fmode = 0; 1582 struct vattr vattr; 1583 1584 /* 1585 * Oops, not for me.. 1586 */ 1587 if (vap->va_type == VSOCK) 1588 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1589 1590 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1591 return (error); 1592 if (vap->va_vaflags & VA_EXCLUSIVE) 1593 fmode |= O_EXCL; 1594 dnp = VTONFS(dvp); 1595 nmp = VFSTONFS(vnode_mount(dvp)); 1596 again: 1597 /* For NFSv4, wait until any remove is done. */ 1598 mtx_lock(&dnp->n_mtx); 1599 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) { 1600 dnp->n_flag |= NREMOVEWANT; 1601 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0); 1602 } 1603 mtx_unlock(&dnp->n_mtx); 1604 1605 cverf = nfs_get_cverf(); 1606 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1607 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, 1608 &nfhp, &attrflag, &dattrflag, NULL); 1609 if (!error) { 1610 if (nfhp == NULL) 1611 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1612 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, 1613 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1614 NULL); 1615 if (nfhp != NULL) 1616 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1617 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); 1618 } 1619 if (dattrflag) 1620 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1621 if (!error) { 1622 newvp = NFSTOV(np); 1623 if (attrflag == 0) 1624 error = nfsrpc_getattr(newvp, cnp->cn_cred, 1625 cnp->cn_thread, &nfsva, NULL); 1626 if (error == 0) 1627 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1628 0, 1); 1629 } 1630 if (error) { 1631 if (newvp != NULL) { 1632 vput(newvp); 1633 newvp = NULL; 1634 } 1635 if (NFS_ISV34(dvp) && (fmode & O_EXCL) && 1636 error == NFSERR_NOTSUPP) { 1637 fmode &= ~O_EXCL; 1638 goto again; 1639 } 1640 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) { 1641 if (nfscl_checksattr(vap, &nfsva)) { 1642 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred, 1643 cnp->cn_thread, &nfsva, &attrflag, NULL); 1644 if (error && (vap->va_uid != (uid_t)VNOVAL || 1645 vap->va_gid != (gid_t)VNOVAL)) { 1646 /* try again without setting uid/gid */ 1647 vap->va_uid = (uid_t)VNOVAL; 1648 vap->va_gid = (uid_t)VNOVAL; 1649 error = nfsrpc_setattr(newvp, vap, NULL, 1650 cnp->cn_cred, cnp->cn_thread, &nfsva, 1651 &attrflag, NULL); 1652 } 1653 if (attrflag) 1654 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 1655 NULL, 0, 1); 1656 if (error != 0) 1657 vput(newvp); 1658 } 1659 } 1660 if (!error) { 1661 if ((cnp->cn_flags & MAKEENTRY) && attrflag) 1662 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 1663 NULL); 1664 *ap->a_vpp = newvp; 1665 } else if (NFS_ISV4(dvp)) { 1666 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, 1667 vap->va_gid); 1668 } 1669 mtx_lock(&dnp->n_mtx); 1670 dnp->n_flag |= NMODIFIED; 1671 if (!dattrflag) { 1672 dnp->n_attrstamp = 0; 1673 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1674 } 1675 mtx_unlock(&dnp->n_mtx); 1676 return (error); 1677 } 1678 1679 /* 1680 * nfs file remove call 1681 * To try and make nfs semantics closer to ufs semantics, a file that has 1682 * other processes using the vnode is renamed instead of removed and then 1683 * removed later on the last close. 1684 * - If v_usecount > 1 1685 * If a rename is not already in the works 1686 * call nfs_sillyrename() to set it up 1687 * else 1688 * do the remove rpc 1689 */ 1690 static int 1691 nfs_remove(struct vop_remove_args *ap) 1692 { 1693 struct vnode *vp = ap->a_vp; 1694 struct vnode *dvp = ap->a_dvp; 1695 struct componentname *cnp = ap->a_cnp; 1696 struct nfsnode *np = VTONFS(vp); 1697 int error = 0; 1698 struct vattr vattr; 1699 1700 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name")); 1701 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount")); 1702 if (vp->v_type == VDIR) 1703 error = EPERM; 1704 else if (vrefcnt(vp) == 1 || (np->n_sillyrename && 1705 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 1706 vattr.va_nlink > 1)) { 1707 /* 1708 * Purge the name cache so that the chance of a lookup for 1709 * the name succeeding while the remove is in progress is 1710 * minimized. Without node locking it can still happen, such 1711 * that an I/O op returns ESTALE, but since you get this if 1712 * another host removes the file.. 1713 */ 1714 cache_purge(vp); 1715 /* 1716 * throw away biocache buffers, mainly to avoid 1717 * unnecessary delayed writes later. 1718 */ 1719 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1); 1720 if (error != EINTR && error != EIO) 1721 /* Do the rpc */ 1722 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr, 1723 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread); 1724 /* 1725 * Kludge City: If the first reply to the remove rpc is lost.. 1726 * the reply to the retransmitted request will be ENOENT 1727 * since the file was in fact removed 1728 * Therefore, we cheat and return success. 1729 */ 1730 if (error == ENOENT) 1731 error = 0; 1732 } else if (!np->n_sillyrename) 1733 error = nfs_sillyrename(dvp, vp, cnp); 1734 mtx_lock(&np->n_mtx); 1735 np->n_attrstamp = 0; 1736 mtx_unlock(&np->n_mtx); 1737 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 1738 return (error); 1739 } 1740 1741 /* 1742 * nfs file remove rpc called from nfs_inactive 1743 */ 1744 int 1745 ncl_removeit(struct sillyrename *sp, struct vnode *vp) 1746 { 1747 /* 1748 * Make sure that the directory vnode is still valid. 1749 * XXX we should lock sp->s_dvp here. 1750 */ 1751 if (sp->s_dvp->v_type == VBAD) 1752 return (0); 1753 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen, 1754 sp->s_cred, NULL)); 1755 } 1756 1757 /* 1758 * Nfs remove rpc, called from nfs_remove() and ncl_removeit(). 1759 */ 1760 static int 1761 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 1762 int namelen, struct ucred *cred, struct thread *td) 1763 { 1764 struct nfsvattr dnfsva; 1765 struct nfsnode *dnp = VTONFS(dvp); 1766 int error = 0, dattrflag; 1767 1768 mtx_lock(&dnp->n_mtx); 1769 dnp->n_flag |= NREMOVEINPROG; 1770 mtx_unlock(&dnp->n_mtx); 1771 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva, 1772 &dattrflag, NULL); 1773 mtx_lock(&dnp->n_mtx); 1774 if ((dnp->n_flag & NREMOVEWANT)) { 1775 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG); 1776 mtx_unlock(&dnp->n_mtx); 1777 wakeup((caddr_t)dnp); 1778 } else { 1779 dnp->n_flag &= ~NREMOVEINPROG; 1780 mtx_unlock(&dnp->n_mtx); 1781 } 1782 if (dattrflag) 1783 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1784 mtx_lock(&dnp->n_mtx); 1785 dnp->n_flag |= NMODIFIED; 1786 if (!dattrflag) { 1787 dnp->n_attrstamp = 0; 1788 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1789 } 1790 mtx_unlock(&dnp->n_mtx); 1791 if (error && NFS_ISV4(dvp)) 1792 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1793 return (error); 1794 } 1795 1796 /* 1797 * nfs file rename call 1798 */ 1799 static int 1800 nfs_rename(struct vop_rename_args *ap) 1801 { 1802 struct vnode *fvp = ap->a_fvp; 1803 struct vnode *tvp = ap->a_tvp; 1804 struct vnode *fdvp = ap->a_fdvp; 1805 struct vnode *tdvp = ap->a_tdvp; 1806 struct componentname *tcnp = ap->a_tcnp; 1807 struct componentname *fcnp = ap->a_fcnp; 1808 struct nfsnode *fnp = VTONFS(ap->a_fvp); 1809 struct nfsnode *tdnp = VTONFS(ap->a_tdvp); 1810 struct nfsv4node *newv4 = NULL; 1811 int error; 1812 1813 KASSERT((tcnp->cn_flags & HASBUF) != 0 && 1814 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name")); 1815 /* Check for cross-device rename */ 1816 if ((fvp->v_mount != tdvp->v_mount) || 1817 (tvp && (fvp->v_mount != tvp->v_mount))) { 1818 error = EXDEV; 1819 goto out; 1820 } 1821 1822 if (fvp == tvp) { 1823 printf("nfs_rename: fvp == tvp (can't happen)\n"); 1824 error = 0; 1825 goto out; 1826 } 1827 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0) 1828 goto out; 1829 1830 /* 1831 * We have to flush B_DELWRI data prior to renaming 1832 * the file. If we don't, the delayed-write buffers 1833 * can be flushed out later after the file has gone stale 1834 * under NFSV3. NFSV2 does not have this problem because 1835 * ( as far as I can tell ) it flushes dirty buffers more 1836 * often. 1837 * 1838 * Skip the rename operation if the fsync fails, this can happen 1839 * due to the server's volume being full, when we pushed out data 1840 * that was written back to our cache earlier. Not checking for 1841 * this condition can result in potential (silent) data loss. 1842 */ 1843 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread); 1844 NFSVOPUNLOCK(fvp, 0); 1845 if (!error && tvp) 1846 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread); 1847 if (error) 1848 goto out; 1849 1850 /* 1851 * If the tvp exists and is in use, sillyrename it before doing the 1852 * rename of the new file over it. 1853 * XXX Can't sillyrename a directory. 1854 */ 1855 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename && 1856 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1857 vput(tvp); 1858 tvp = NULL; 1859 } 1860 1861 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1862 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1863 tcnp->cn_thread); 1864 1865 if (error == 0 && NFS_ISV4(tdvp)) { 1866 /* 1867 * For NFSv4, check to see if it is the same name and 1868 * replace the name, if it is different. 1869 */ 1870 newv4 = malloc( 1871 sizeof (struct nfsv4node) + 1872 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1, 1873 M_NFSV4NODE, M_WAITOK); 1874 mtx_lock(&tdnp->n_mtx); 1875 mtx_lock(&fnp->n_mtx); 1876 if (fnp->n_v4 != NULL && fvp->v_type == VREG && 1877 (fnp->n_v4->n4_namelen != tcnp->cn_namelen || 1878 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4), 1879 tcnp->cn_namelen) || 1880 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen || 1881 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 1882 tdnp->n_fhp->nfh_len))) { 1883 #ifdef notdef 1884 { char nnn[100]; int nnnl; 1885 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99; 1886 bcopy(tcnp->cn_nameptr, nnn, nnnl); 1887 nnn[nnnl] = '\0'; 1888 printf("ren replace=%s\n",nnn); 1889 } 1890 #endif 1891 free(fnp->n_v4, M_NFSV4NODE); 1892 fnp->n_v4 = newv4; 1893 newv4 = NULL; 1894 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len; 1895 fnp->n_v4->n4_namelen = tcnp->cn_namelen; 1896 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 1897 tdnp->n_fhp->nfh_len); 1898 NFSBCOPY(tcnp->cn_nameptr, 1899 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen); 1900 } 1901 mtx_unlock(&tdnp->n_mtx); 1902 mtx_unlock(&fnp->n_mtx); 1903 if (newv4 != NULL) 1904 free(newv4, M_NFSV4NODE); 1905 } 1906 1907 if (fvp->v_type == VDIR) { 1908 if (tvp != NULL && tvp->v_type == VDIR) 1909 cache_purge(tdvp); 1910 cache_purge(fdvp); 1911 } 1912 1913 out: 1914 if (tdvp == tvp) 1915 vrele(tdvp); 1916 else 1917 vput(tdvp); 1918 if (tvp) 1919 vput(tvp); 1920 vrele(fdvp); 1921 vrele(fvp); 1922 /* 1923 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1924 */ 1925 if (error == ENOENT) 1926 error = 0; 1927 return (error); 1928 } 1929 1930 /* 1931 * nfs file rename rpc called from nfs_remove() above 1932 */ 1933 static int 1934 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp, 1935 struct sillyrename *sp) 1936 { 1937 1938 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen, 1939 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred, 1940 scnp->cn_thread)); 1941 } 1942 1943 /* 1944 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1945 */ 1946 static int 1947 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr, 1948 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr, 1949 int tnamelen, struct ucred *cred, struct thread *td) 1950 { 1951 struct nfsvattr fnfsva, tnfsva; 1952 struct nfsnode *fdnp = VTONFS(fdvp); 1953 struct nfsnode *tdnp = VTONFS(tdvp); 1954 int error = 0, fattrflag, tattrflag; 1955 1956 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp, 1957 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag, 1958 &tattrflag, NULL, NULL); 1959 mtx_lock(&fdnp->n_mtx); 1960 fdnp->n_flag |= NMODIFIED; 1961 if (fattrflag != 0) { 1962 mtx_unlock(&fdnp->n_mtx); 1963 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1); 1964 } else { 1965 fdnp->n_attrstamp = 0; 1966 mtx_unlock(&fdnp->n_mtx); 1967 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp); 1968 } 1969 mtx_lock(&tdnp->n_mtx); 1970 tdnp->n_flag |= NMODIFIED; 1971 if (tattrflag != 0) { 1972 mtx_unlock(&tdnp->n_mtx); 1973 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1); 1974 } else { 1975 tdnp->n_attrstamp = 0; 1976 mtx_unlock(&tdnp->n_mtx); 1977 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 1978 } 1979 if (error && NFS_ISV4(fdvp)) 1980 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1981 return (error); 1982 } 1983 1984 /* 1985 * nfs hard link create call 1986 */ 1987 static int 1988 nfs_link(struct vop_link_args *ap) 1989 { 1990 struct vnode *vp = ap->a_vp; 1991 struct vnode *tdvp = ap->a_tdvp; 1992 struct componentname *cnp = ap->a_cnp; 1993 struct nfsnode *np, *tdnp; 1994 struct nfsvattr nfsva, dnfsva; 1995 int error = 0, attrflag, dattrflag; 1996 1997 /* 1998 * Push all writes to the server, so that the attribute cache 1999 * doesn't get "out of sync" with the server. 2000 * XXX There should be a better way! 2001 */ 2002 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread); 2003 2004 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 2005 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag, 2006 &dattrflag, NULL); 2007 tdnp = VTONFS(tdvp); 2008 mtx_lock(&tdnp->n_mtx); 2009 tdnp->n_flag |= NMODIFIED; 2010 if (dattrflag != 0) { 2011 mtx_unlock(&tdnp->n_mtx); 2012 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1); 2013 } else { 2014 tdnp->n_attrstamp = 0; 2015 mtx_unlock(&tdnp->n_mtx); 2016 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 2017 } 2018 if (attrflag) 2019 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2020 else { 2021 np = VTONFS(vp); 2022 mtx_lock(&np->n_mtx); 2023 np->n_attrstamp = 0; 2024 mtx_unlock(&np->n_mtx); 2025 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 2026 } 2027 /* 2028 * If negative lookup caching is enabled, I might as well 2029 * add an entry for this node. Not necessary for correctness, 2030 * but if negative caching is enabled, then the system 2031 * must care about lookup caching hit rate, so... 2032 */ 2033 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 && 2034 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 2035 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL); 2036 } 2037 if (error && NFS_ISV4(vp)) 2038 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, 2039 (gid_t)0); 2040 return (error); 2041 } 2042 2043 /* 2044 * nfs symbolic link create call 2045 */ 2046 static int 2047 nfs_symlink(struct vop_symlink_args *ap) 2048 { 2049 struct vnode *dvp = ap->a_dvp; 2050 struct vattr *vap = ap->a_vap; 2051 struct componentname *cnp = ap->a_cnp; 2052 struct nfsvattr nfsva, dnfsva; 2053 struct nfsfh *nfhp; 2054 struct nfsnode *np = NULL, *dnp; 2055 struct vnode *newvp = NULL; 2056 int error = 0, attrflag, dattrflag, ret; 2057 2058 vap->va_type = VLNK; 2059 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2060 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, 2061 &nfsva, &nfhp, &attrflag, &dattrflag, NULL); 2062 if (nfhp) { 2063 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, 2064 &np, NULL, LK_EXCLUSIVE); 2065 if (!ret) 2066 newvp = NFSTOV(np); 2067 else if (!error) 2068 error = ret; 2069 } 2070 if (newvp != NULL) { 2071 if (attrflag) 2072 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 2073 0, 1); 2074 } else if (!error) { 2075 /* 2076 * If we do not have an error and we could not extract the 2077 * newvp from the response due to the request being NFSv2, we 2078 * have to do a lookup in order to obtain a newvp to return. 2079 */ 2080 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2081 cnp->cn_cred, cnp->cn_thread, &np); 2082 if (!error) 2083 newvp = NFSTOV(np); 2084 } 2085 if (error) { 2086 if (newvp) 2087 vput(newvp); 2088 if (NFS_ISV4(dvp)) 2089 error = nfscl_maperr(cnp->cn_thread, error, 2090 vap->va_uid, vap->va_gid); 2091 } else { 2092 *ap->a_vpp = newvp; 2093 } 2094 2095 dnp = VTONFS(dvp); 2096 mtx_lock(&dnp->n_mtx); 2097 dnp->n_flag |= NMODIFIED; 2098 if (dattrflag != 0) { 2099 mtx_unlock(&dnp->n_mtx); 2100 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2101 } else { 2102 dnp->n_attrstamp = 0; 2103 mtx_unlock(&dnp->n_mtx); 2104 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2105 } 2106 /* 2107 * If negative lookup caching is enabled, I might as well 2108 * add an entry for this node. Not necessary for correctness, 2109 * but if negative caching is enabled, then the system 2110 * must care about lookup caching hit rate, so... 2111 */ 2112 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2113 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 2114 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL); 2115 } 2116 return (error); 2117 } 2118 2119 /* 2120 * nfs make dir call 2121 */ 2122 static int 2123 nfs_mkdir(struct vop_mkdir_args *ap) 2124 { 2125 struct vnode *dvp = ap->a_dvp; 2126 struct vattr *vap = ap->a_vap; 2127 struct componentname *cnp = ap->a_cnp; 2128 struct nfsnode *np = NULL, *dnp; 2129 struct vnode *newvp = NULL; 2130 struct vattr vattr; 2131 struct nfsfh *nfhp; 2132 struct nfsvattr nfsva, dnfsva; 2133 int error = 0, attrflag, dattrflag, ret; 2134 2135 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) 2136 return (error); 2137 vap->va_type = VDIR; 2138 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2139 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp, 2140 &attrflag, &dattrflag, NULL); 2141 dnp = VTONFS(dvp); 2142 mtx_lock(&dnp->n_mtx); 2143 dnp->n_flag |= NMODIFIED; 2144 if (dattrflag != 0) { 2145 mtx_unlock(&dnp->n_mtx); 2146 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2147 } else { 2148 dnp->n_attrstamp = 0; 2149 mtx_unlock(&dnp->n_mtx); 2150 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2151 } 2152 if (nfhp) { 2153 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, 2154 &np, NULL, LK_EXCLUSIVE); 2155 if (!ret) { 2156 newvp = NFSTOV(np); 2157 if (attrflag) 2158 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 2159 NULL, 0, 1); 2160 } else if (!error) 2161 error = ret; 2162 } 2163 if (!error && newvp == NULL) { 2164 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2165 cnp->cn_cred, cnp->cn_thread, &np); 2166 if (!error) { 2167 newvp = NFSTOV(np); 2168 if (newvp->v_type != VDIR) 2169 error = EEXIST; 2170 } 2171 } 2172 if (error) { 2173 if (newvp) 2174 vput(newvp); 2175 if (NFS_ISV4(dvp)) 2176 error = nfscl_maperr(cnp->cn_thread, error, 2177 vap->va_uid, vap->va_gid); 2178 } else { 2179 /* 2180 * If negative lookup caching is enabled, I might as well 2181 * add an entry for this node. Not necessary for correctness, 2182 * but if negative caching is enabled, then the system 2183 * must care about lookup caching hit rate, so... 2184 */ 2185 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2186 (cnp->cn_flags & MAKEENTRY) && 2187 attrflag != 0 && dattrflag != 0) 2188 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 2189 &dnfsva.na_ctime); 2190 *ap->a_vpp = newvp; 2191 } 2192 return (error); 2193 } 2194 2195 /* 2196 * nfs remove directory call 2197 */ 2198 static int 2199 nfs_rmdir(struct vop_rmdir_args *ap) 2200 { 2201 struct vnode *vp = ap->a_vp; 2202 struct vnode *dvp = ap->a_dvp; 2203 struct componentname *cnp = ap->a_cnp; 2204 struct nfsnode *dnp; 2205 struct nfsvattr dnfsva; 2206 int error, dattrflag; 2207 2208 if (dvp == vp) 2209 return (EINVAL); 2210 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2211 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL); 2212 dnp = VTONFS(dvp); 2213 mtx_lock(&dnp->n_mtx); 2214 dnp->n_flag |= NMODIFIED; 2215 if (dattrflag != 0) { 2216 mtx_unlock(&dnp->n_mtx); 2217 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2218 } else { 2219 dnp->n_attrstamp = 0; 2220 mtx_unlock(&dnp->n_mtx); 2221 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2222 } 2223 2224 cache_purge(dvp); 2225 cache_purge(vp); 2226 if (error && NFS_ISV4(dvp)) 2227 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, 2228 (gid_t)0); 2229 /* 2230 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2231 */ 2232 if (error == ENOENT) 2233 error = 0; 2234 return (error); 2235 } 2236 2237 /* 2238 * nfs readdir call 2239 */ 2240 static int 2241 nfs_readdir(struct vop_readdir_args *ap) 2242 { 2243 struct vnode *vp = ap->a_vp; 2244 struct nfsnode *np = VTONFS(vp); 2245 struct uio *uio = ap->a_uio; 2246 ssize_t tresid, left; 2247 int error = 0; 2248 struct vattr vattr; 2249 2250 if (ap->a_eofflag != NULL) 2251 *ap->a_eofflag = 0; 2252 if (vp->v_type != VDIR) 2253 return(EPERM); 2254 2255 /* 2256 * First, check for hit on the EOF offset cache 2257 */ 2258 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2259 (np->n_flag & NMODIFIED) == 0) { 2260 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) { 2261 mtx_lock(&np->n_mtx); 2262 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) || 2263 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 2264 mtx_unlock(&np->n_mtx); 2265 NFSINCRGLOBAL(nfsstatsv1.direofcache_hits); 2266 if (ap->a_eofflag != NULL) 2267 *ap->a_eofflag = 1; 2268 return (0); 2269 } else 2270 mtx_unlock(&np->n_mtx); 2271 } 2272 } 2273 2274 /* 2275 * NFS always guarantees that directory entries don't straddle 2276 * DIRBLKSIZ boundaries. As such, we need to limit the size 2277 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial 2278 * directory entry. 2279 */ 2280 left = uio->uio_resid % DIRBLKSIZ; 2281 if (left == uio->uio_resid) 2282 return (EINVAL); 2283 uio->uio_resid -= left; 2284 2285 /* 2286 * Call ncl_bioread() to do the real work. 2287 */ 2288 tresid = uio->uio_resid; 2289 error = ncl_bioread(vp, uio, 0, ap->a_cred); 2290 2291 if (!error && uio->uio_resid == tresid) { 2292 NFSINCRGLOBAL(nfsstatsv1.direofcache_misses); 2293 if (ap->a_eofflag != NULL) 2294 *ap->a_eofflag = 1; 2295 } 2296 2297 /* Add the partial DIRBLKSIZ (left) back in. */ 2298 uio->uio_resid += left; 2299 return (error); 2300 } 2301 2302 /* 2303 * Readdir rpc call. 2304 * Called from below the buffer cache by ncl_doio(). 2305 */ 2306 int 2307 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2308 struct thread *td) 2309 { 2310 struct nfsvattr nfsva; 2311 nfsuint64 *cookiep, cookie; 2312 struct nfsnode *dnp = VTONFS(vp); 2313 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2314 int error = 0, eof, attrflag; 2315 2316 KASSERT(uiop->uio_iovcnt == 1 && 2317 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2318 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2319 ("nfs readdirrpc bad uio")); 2320 2321 /* 2322 * If there is no cookie, assume directory was stale. 2323 */ 2324 ncl_dircookie_lock(dnp); 2325 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2326 if (cookiep) { 2327 cookie = *cookiep; 2328 ncl_dircookie_unlock(dnp); 2329 } else { 2330 ncl_dircookie_unlock(dnp); 2331 return (NFSERR_BAD_COOKIE); 2332 } 2333 2334 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2335 (void)ncl_fsinfo(nmp, vp, cred, td); 2336 2337 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva, 2338 &attrflag, &eof, NULL); 2339 if (attrflag) 2340 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2341 2342 if (!error) { 2343 /* 2344 * We are now either at the end of the directory or have filled 2345 * the block. 2346 */ 2347 if (eof) 2348 dnp->n_direofoffset = uiop->uio_offset; 2349 else { 2350 if (uiop->uio_resid > 0) 2351 printf("EEK! readdirrpc resid > 0\n"); 2352 ncl_dircookie_lock(dnp); 2353 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2354 *cookiep = cookie; 2355 ncl_dircookie_unlock(dnp); 2356 } 2357 } else if (NFS_ISV4(vp)) { 2358 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2359 } 2360 return (error); 2361 } 2362 2363 /* 2364 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc(). 2365 */ 2366 int 2367 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2368 struct thread *td) 2369 { 2370 struct nfsvattr nfsva; 2371 nfsuint64 *cookiep, cookie; 2372 struct nfsnode *dnp = VTONFS(vp); 2373 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2374 int error = 0, attrflag, eof; 2375 2376 KASSERT(uiop->uio_iovcnt == 1 && 2377 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2378 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2379 ("nfs readdirplusrpc bad uio")); 2380 2381 /* 2382 * If there is no cookie, assume directory was stale. 2383 */ 2384 ncl_dircookie_lock(dnp); 2385 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2386 if (cookiep) { 2387 cookie = *cookiep; 2388 ncl_dircookie_unlock(dnp); 2389 } else { 2390 ncl_dircookie_unlock(dnp); 2391 return (NFSERR_BAD_COOKIE); 2392 } 2393 2394 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2395 (void)ncl_fsinfo(nmp, vp, cred, td); 2396 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva, 2397 &attrflag, &eof, NULL); 2398 if (attrflag) 2399 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2400 2401 if (!error) { 2402 /* 2403 * We are now either at end of the directory or have filled the 2404 * the block. 2405 */ 2406 if (eof) 2407 dnp->n_direofoffset = uiop->uio_offset; 2408 else { 2409 if (uiop->uio_resid > 0) 2410 printf("EEK! readdirplusrpc resid > 0\n"); 2411 ncl_dircookie_lock(dnp); 2412 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2413 *cookiep = cookie; 2414 ncl_dircookie_unlock(dnp); 2415 } 2416 } else if (NFS_ISV4(vp)) { 2417 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2418 } 2419 return (error); 2420 } 2421 2422 /* 2423 * Silly rename. To make the NFS filesystem that is stateless look a little 2424 * more like the "ufs" a remove of an active vnode is translated to a rename 2425 * to a funny looking filename that is removed by nfs_inactive on the 2426 * nfsnode. There is the potential for another process on a different client 2427 * to create the same funny name between the nfs_lookitup() fails and the 2428 * nfs_rename() completes, but... 2429 */ 2430 static int 2431 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2432 { 2433 struct sillyrename *sp; 2434 struct nfsnode *np; 2435 int error; 2436 short pid; 2437 unsigned int lticks; 2438 2439 cache_purge(dvp); 2440 np = VTONFS(vp); 2441 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir")); 2442 sp = malloc(sizeof (struct sillyrename), 2443 M_NEWNFSREQ, M_WAITOK); 2444 sp->s_cred = crhold(cnp->cn_cred); 2445 sp->s_dvp = dvp; 2446 VREF(dvp); 2447 2448 /* 2449 * Fudge together a funny name. 2450 * Changing the format of the funny name to accommodate more 2451 * sillynames per directory. 2452 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is 2453 * CPU ticks since boot. 2454 */ 2455 pid = cnp->cn_thread->td_proc->p_pid; 2456 lticks = (unsigned int)ticks; 2457 for ( ; ; ) { 2458 sp->s_namlen = sprintf(sp->s_name, 2459 ".nfs.%08x.%04x4.4", lticks, 2460 pid); 2461 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2462 cnp->cn_thread, NULL)) 2463 break; 2464 lticks++; 2465 } 2466 error = nfs_renameit(dvp, vp, cnp, sp); 2467 if (error) 2468 goto bad; 2469 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2470 cnp->cn_thread, &np); 2471 np->n_sillyrename = sp; 2472 return (0); 2473 bad: 2474 vrele(sp->s_dvp); 2475 crfree(sp->s_cred); 2476 free(sp, M_NEWNFSREQ); 2477 return (error); 2478 } 2479 2480 /* 2481 * Look up a file name and optionally either update the file handle or 2482 * allocate an nfsnode, depending on the value of npp. 2483 * npp == NULL --> just do the lookup 2484 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2485 * handled too 2486 * *npp != NULL --> update the file handle in the vnode 2487 */ 2488 static int 2489 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred, 2490 struct thread *td, struct nfsnode **npp) 2491 { 2492 struct vnode *newvp = NULL, *vp; 2493 struct nfsnode *np, *dnp = VTONFS(dvp); 2494 struct nfsfh *nfhp, *onfhp; 2495 struct nfsvattr nfsva, dnfsva; 2496 struct componentname cn; 2497 int error = 0, attrflag, dattrflag; 2498 u_int hash; 2499 2500 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva, 2501 &nfhp, &attrflag, &dattrflag, NULL); 2502 if (dattrflag) 2503 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2504 if (npp && !error) { 2505 if (*npp != NULL) { 2506 np = *npp; 2507 vp = NFSTOV(np); 2508 /* 2509 * For NFSv4, check to see if it is the same name and 2510 * replace the name, if it is different. 2511 */ 2512 if (np->n_v4 != NULL && nfsva.na_type == VREG && 2513 (np->n_v4->n4_namelen != len || 2514 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) || 2515 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || 2516 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2517 dnp->n_fhp->nfh_len))) { 2518 #ifdef notdef 2519 { char nnn[100]; int nnnl; 2520 nnnl = (len < 100) ? len : 99; 2521 bcopy(name, nnn, nnnl); 2522 nnn[nnnl] = '\0'; 2523 printf("replace=%s\n",nnn); 2524 } 2525 #endif 2526 free(np->n_v4, M_NFSV4NODE); 2527 np->n_v4 = malloc( 2528 sizeof (struct nfsv4node) + 2529 dnp->n_fhp->nfh_len + len - 1, 2530 M_NFSV4NODE, M_WAITOK); 2531 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; 2532 np->n_v4->n4_namelen = len; 2533 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2534 dnp->n_fhp->nfh_len); 2535 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len); 2536 } 2537 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, 2538 FNV1_32_INIT); 2539 onfhp = np->n_fhp; 2540 /* 2541 * Rehash node for new file handle. 2542 */ 2543 vfs_hash_rehash(vp, hash); 2544 np->n_fhp = nfhp; 2545 if (onfhp != NULL) 2546 free(onfhp, M_NFSFH); 2547 newvp = NFSTOV(np); 2548 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) { 2549 free(nfhp, M_NFSFH); 2550 VREF(dvp); 2551 newvp = dvp; 2552 } else { 2553 cn.cn_nameptr = name; 2554 cn.cn_namelen = len; 2555 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td, 2556 &np, NULL, LK_EXCLUSIVE); 2557 if (error) 2558 return (error); 2559 newvp = NFSTOV(np); 2560 } 2561 if (!attrflag && *npp == NULL) { 2562 if (newvp == dvp) 2563 vrele(newvp); 2564 else 2565 vput(newvp); 2566 return (ENOENT); 2567 } 2568 if (attrflag) 2569 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 2570 0, 1); 2571 } 2572 if (npp && *npp == NULL) { 2573 if (error) { 2574 if (newvp) { 2575 if (newvp == dvp) 2576 vrele(newvp); 2577 else 2578 vput(newvp); 2579 } 2580 } else 2581 *npp = np; 2582 } 2583 if (error && NFS_ISV4(dvp)) 2584 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2585 return (error); 2586 } 2587 2588 /* 2589 * Nfs Version 3 and 4 commit rpc 2590 */ 2591 int 2592 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred, 2593 struct thread *td) 2594 { 2595 struct nfsvattr nfsva; 2596 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2597 struct nfsnode *np; 2598 struct uio uio; 2599 int error, attrflag; 2600 2601 np = VTONFS(vp); 2602 error = EIO; 2603 attrflag = 0; 2604 if (NFSHASPNFS(nmp) && (np->n_flag & NDSCOMMIT) != 0) { 2605 uio.uio_offset = offset; 2606 uio.uio_resid = cnt; 2607 error = nfscl_doiods(vp, &uio, NULL, NULL, 2608 NFSV4OPEN_ACCESSWRITE, 1, cred, td); 2609 if (error != 0) { 2610 mtx_lock(&np->n_mtx); 2611 np->n_flag &= ~NDSCOMMIT; 2612 mtx_unlock(&np->n_mtx); 2613 } 2614 } 2615 if (error != 0) { 2616 mtx_lock(&nmp->nm_mtx); 2617 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) { 2618 mtx_unlock(&nmp->nm_mtx); 2619 return (0); 2620 } 2621 mtx_unlock(&nmp->nm_mtx); 2622 error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva, 2623 &attrflag, NULL); 2624 } 2625 if (attrflag != 0) 2626 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 2627 0, 1); 2628 if (error != 0 && NFS_ISV4(vp)) 2629 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2630 return (error); 2631 } 2632 2633 /* 2634 * Strategy routine. 2635 * For async requests when nfsiod(s) are running, queue the request by 2636 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the 2637 * request. 2638 */ 2639 static int 2640 nfs_strategy(struct vop_strategy_args *ap) 2641 { 2642 struct buf *bp; 2643 struct vnode *vp; 2644 struct ucred *cr; 2645 2646 bp = ap->a_bp; 2647 vp = ap->a_vp; 2648 KASSERT(bp->b_vp == vp, ("missing b_getvp")); 2649 KASSERT(!(bp->b_flags & B_DONE), 2650 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); 2651 BUF_ASSERT_HELD(bp); 2652 2653 if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno) 2654 bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize / 2655 DEV_BSIZE); 2656 if (bp->b_iocmd == BIO_READ) 2657 cr = bp->b_rcred; 2658 else 2659 cr = bp->b_wcred; 2660 2661 /* 2662 * If the op is asynchronous and an i/o daemon is waiting 2663 * queue the request, wake it up and wait for completion 2664 * otherwise just do it ourselves. 2665 */ 2666 if ((bp->b_flags & B_ASYNC) == 0 || 2667 ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread)) 2668 (void) ncl_doio(vp, bp, cr, curthread, 1); 2669 return (0); 2670 } 2671 2672 /* 2673 * fsync vnode op. Just call ncl_flush() with commit == 1. 2674 */ 2675 /* ARGSUSED */ 2676 static int 2677 nfs_fsync(struct vop_fsync_args *ap) 2678 { 2679 2680 if (ap->a_vp->v_type != VREG) { 2681 /* 2682 * For NFS, metadata is changed synchronously on the server, 2683 * so there is nothing to flush. Also, ncl_flush() clears 2684 * the NMODIFIED flag and that shouldn't be done here for 2685 * directories. 2686 */ 2687 return (0); 2688 } 2689 return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0)); 2690 } 2691 2692 /* 2693 * Flush all the blocks associated with a vnode. 2694 * Walk through the buffer pool and push any dirty pages 2695 * associated with the vnode. 2696 * If the called_from_renewthread argument is TRUE, it has been called 2697 * from the NFSv4 renew thread and, as such, cannot block indefinitely 2698 * waiting for a buffer write to complete. 2699 */ 2700 int 2701 ncl_flush(struct vnode *vp, int waitfor, struct thread *td, 2702 int commit, int called_from_renewthread) 2703 { 2704 struct nfsnode *np = VTONFS(vp); 2705 struct buf *bp; 2706 int i; 2707 struct buf *nbp; 2708 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2709 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos; 2710 int passone = 1, trycnt = 0; 2711 u_quad_t off, endoff, toff; 2712 struct ucred* wcred = NULL; 2713 struct buf **bvec = NULL; 2714 struct bufobj *bo; 2715 #ifndef NFS_COMMITBVECSIZ 2716 #define NFS_COMMITBVECSIZ 20 2717 #endif 2718 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ]; 2719 u_int bvecsize = 0, bveccount; 2720 2721 if (called_from_renewthread != 0) 2722 slptimeo = hz; 2723 if (nmp->nm_flag & NFSMNT_INT) 2724 slpflag = PCATCH; 2725 if (!commit) 2726 passone = 0; 2727 bo = &vp->v_bufobj; 2728 /* 2729 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the 2730 * server, but has not been committed to stable storage on the server 2731 * yet. On the first pass, the byte range is worked out and the commit 2732 * rpc is done. On the second pass, ncl_writebp() is called to do the 2733 * job. 2734 */ 2735 again: 2736 off = (u_quad_t)-1; 2737 endoff = 0; 2738 bvecpos = 0; 2739 if (NFS_ISV34(vp) && commit) { 2740 if (bvec != NULL && bvec != bvec_on_stack) 2741 free(bvec, M_TEMP); 2742 /* 2743 * Count up how many buffers waiting for a commit. 2744 */ 2745 bveccount = 0; 2746 BO_LOCK(bo); 2747 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2748 if (!BUF_ISLOCKED(bp) && 2749 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) 2750 == (B_DELWRI | B_NEEDCOMMIT)) 2751 bveccount++; 2752 } 2753 /* 2754 * Allocate space to remember the list of bufs to commit. It is 2755 * important to use M_NOWAIT here to avoid a race with nfs_write. 2756 * If we can't get memory (for whatever reason), we will end up 2757 * committing the buffers one-by-one in the loop below. 2758 */ 2759 if (bveccount > NFS_COMMITBVECSIZ) { 2760 /* 2761 * Release the vnode interlock to avoid a lock 2762 * order reversal. 2763 */ 2764 BO_UNLOCK(bo); 2765 bvec = (struct buf **) 2766 malloc(bveccount * sizeof(struct buf *), 2767 M_TEMP, M_NOWAIT); 2768 BO_LOCK(bo); 2769 if (bvec == NULL) { 2770 bvec = bvec_on_stack; 2771 bvecsize = NFS_COMMITBVECSIZ; 2772 } else 2773 bvecsize = bveccount; 2774 } else { 2775 bvec = bvec_on_stack; 2776 bvecsize = NFS_COMMITBVECSIZ; 2777 } 2778 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2779 if (bvecpos >= bvecsize) 2780 break; 2781 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2782 nbp = TAILQ_NEXT(bp, b_bobufs); 2783 continue; 2784 } 2785 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 2786 (B_DELWRI | B_NEEDCOMMIT)) { 2787 BUF_UNLOCK(bp); 2788 nbp = TAILQ_NEXT(bp, b_bobufs); 2789 continue; 2790 } 2791 BO_UNLOCK(bo); 2792 bremfree(bp); 2793 /* 2794 * Work out if all buffers are using the same cred 2795 * so we can deal with them all with one commit. 2796 * 2797 * NOTE: we are not clearing B_DONE here, so we have 2798 * to do it later on in this routine if we intend to 2799 * initiate I/O on the bp. 2800 * 2801 * Note: to avoid loopback deadlocks, we do not 2802 * assign b_runningbufspace. 2803 */ 2804 if (wcred == NULL) 2805 wcred = bp->b_wcred; 2806 else if (wcred != bp->b_wcred) 2807 wcred = NOCRED; 2808 vfs_busy_pages(bp, 1); 2809 2810 BO_LOCK(bo); 2811 /* 2812 * bp is protected by being locked, but nbp is not 2813 * and vfs_busy_pages() may sleep. We have to 2814 * recalculate nbp. 2815 */ 2816 nbp = TAILQ_NEXT(bp, b_bobufs); 2817 2818 /* 2819 * A list of these buffers is kept so that the 2820 * second loop knows which buffers have actually 2821 * been committed. This is necessary, since there 2822 * may be a race between the commit rpc and new 2823 * uncommitted writes on the file. 2824 */ 2825 bvec[bvecpos++] = bp; 2826 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2827 bp->b_dirtyoff; 2828 if (toff < off) 2829 off = toff; 2830 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); 2831 if (toff > endoff) 2832 endoff = toff; 2833 } 2834 BO_UNLOCK(bo); 2835 } 2836 if (bvecpos > 0) { 2837 /* 2838 * Commit data on the server, as required. 2839 * If all bufs are using the same wcred, then use that with 2840 * one call for all of them, otherwise commit each one 2841 * separately. 2842 */ 2843 if (wcred != NOCRED) 2844 retv = ncl_commit(vp, off, (int)(endoff - off), 2845 wcred, td); 2846 else { 2847 retv = 0; 2848 for (i = 0; i < bvecpos; i++) { 2849 off_t off, size; 2850 bp = bvec[i]; 2851 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2852 bp->b_dirtyoff; 2853 size = (u_quad_t)(bp->b_dirtyend 2854 - bp->b_dirtyoff); 2855 retv = ncl_commit(vp, off, (int)size, 2856 bp->b_wcred, td); 2857 if (retv) break; 2858 } 2859 } 2860 2861 if (retv == NFSERR_STALEWRITEVERF) 2862 ncl_clearcommit(vp->v_mount); 2863 2864 /* 2865 * Now, either mark the blocks I/O done or mark the 2866 * blocks dirty, depending on whether the commit 2867 * succeeded. 2868 */ 2869 for (i = 0; i < bvecpos; i++) { 2870 bp = bvec[i]; 2871 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 2872 if (retv) { 2873 /* 2874 * Error, leave B_DELWRI intact 2875 */ 2876 vfs_unbusy_pages(bp); 2877 brelse(bp); 2878 } else { 2879 /* 2880 * Success, remove B_DELWRI ( bundirty() ). 2881 * 2882 * b_dirtyoff/b_dirtyend seem to be NFS 2883 * specific. We should probably move that 2884 * into bundirty(). XXX 2885 */ 2886 bufobj_wref(bo); 2887 bp->b_flags |= B_ASYNC; 2888 bundirty(bp); 2889 bp->b_flags &= ~B_DONE; 2890 bp->b_ioflags &= ~BIO_ERROR; 2891 bp->b_dirtyoff = bp->b_dirtyend = 0; 2892 bufdone(bp); 2893 } 2894 } 2895 } 2896 2897 /* 2898 * Start/do any write(s) that are required. 2899 */ 2900 loop: 2901 BO_LOCK(bo); 2902 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2903 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2904 if (waitfor != MNT_WAIT || passone) 2905 continue; 2906 2907 error = BUF_TIMELOCK(bp, 2908 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 2909 BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo); 2910 if (error == 0) { 2911 BUF_UNLOCK(bp); 2912 goto loop; 2913 } 2914 if (error == ENOLCK) { 2915 error = 0; 2916 goto loop; 2917 } 2918 if (called_from_renewthread != 0) { 2919 /* 2920 * Return EIO so the flush will be retried 2921 * later. 2922 */ 2923 error = EIO; 2924 goto done; 2925 } 2926 if (newnfs_sigintr(nmp, td)) { 2927 error = EINTR; 2928 goto done; 2929 } 2930 if (slpflag == PCATCH) { 2931 slpflag = 0; 2932 slptimeo = 2 * hz; 2933 } 2934 goto loop; 2935 } 2936 if ((bp->b_flags & B_DELWRI) == 0) 2937 panic("nfs_fsync: not dirty"); 2938 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) { 2939 BUF_UNLOCK(bp); 2940 continue; 2941 } 2942 BO_UNLOCK(bo); 2943 bremfree(bp); 2944 if (passone || !commit) 2945 bp->b_flags |= B_ASYNC; 2946 else 2947 bp->b_flags |= B_ASYNC; 2948 bwrite(bp); 2949 if (newnfs_sigintr(nmp, td)) { 2950 error = EINTR; 2951 goto done; 2952 } 2953 goto loop; 2954 } 2955 if (passone) { 2956 passone = 0; 2957 BO_UNLOCK(bo); 2958 goto again; 2959 } 2960 if (waitfor == MNT_WAIT) { 2961 while (bo->bo_numoutput) { 2962 error = bufobj_wwait(bo, slpflag, slptimeo); 2963 if (error) { 2964 BO_UNLOCK(bo); 2965 if (called_from_renewthread != 0) { 2966 /* 2967 * Return EIO so that the flush will be 2968 * retried later. 2969 */ 2970 error = EIO; 2971 goto done; 2972 } 2973 error = newnfs_sigintr(nmp, td); 2974 if (error) 2975 goto done; 2976 if (slpflag == PCATCH) { 2977 slpflag = 0; 2978 slptimeo = 2 * hz; 2979 } 2980 BO_LOCK(bo); 2981 } 2982 } 2983 if (bo->bo_dirty.bv_cnt != 0 && commit) { 2984 BO_UNLOCK(bo); 2985 goto loop; 2986 } 2987 /* 2988 * Wait for all the async IO requests to drain 2989 */ 2990 BO_UNLOCK(bo); 2991 mtx_lock(&np->n_mtx); 2992 while (np->n_directio_asyncwr > 0) { 2993 np->n_flag |= NFSYNCWAIT; 2994 error = newnfs_msleep(td, &np->n_directio_asyncwr, 2995 &np->n_mtx, slpflag | (PRIBIO + 1), 2996 "nfsfsync", 0); 2997 if (error) { 2998 if (newnfs_sigintr(nmp, td)) { 2999 mtx_unlock(&np->n_mtx); 3000 error = EINTR; 3001 goto done; 3002 } 3003 } 3004 } 3005 mtx_unlock(&np->n_mtx); 3006 } else 3007 BO_UNLOCK(bo); 3008 if (NFSHASPNFS(nmp)) { 3009 nfscl_layoutcommit(vp, td); 3010 /* 3011 * Invalidate the attribute cache, since writes to a DS 3012 * won't update the size attribute. 3013 */ 3014 mtx_lock(&np->n_mtx); 3015 np->n_attrstamp = 0; 3016 } else 3017 mtx_lock(&np->n_mtx); 3018 if (np->n_flag & NWRITEERR) { 3019 error = np->n_error; 3020 np->n_flag &= ~NWRITEERR; 3021 } 3022 if (commit && bo->bo_dirty.bv_cnt == 0 && 3023 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0) 3024 np->n_flag &= ~NMODIFIED; 3025 mtx_unlock(&np->n_mtx); 3026 done: 3027 if (bvec != NULL && bvec != bvec_on_stack) 3028 free(bvec, M_TEMP); 3029 if (error == 0 && commit != 0 && waitfor == MNT_WAIT && 3030 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 || 3031 np->n_directio_asyncwr != 0)) { 3032 if (trycnt++ < 5) { 3033 /* try, try again... */ 3034 passone = 1; 3035 wcred = NULL; 3036 bvec = NULL; 3037 bvecsize = 0; 3038 goto again; 3039 } 3040 vn_printf(vp, "ncl_flush failed"); 3041 error = called_from_renewthread != 0 ? EIO : EBUSY; 3042 } 3043 return (error); 3044 } 3045 3046 /* 3047 * NFS advisory byte-level locks. 3048 */ 3049 static int 3050 nfs_advlock(struct vop_advlock_args *ap) 3051 { 3052 struct vnode *vp = ap->a_vp; 3053 struct ucred *cred; 3054 struct nfsnode *np = VTONFS(ap->a_vp); 3055 struct proc *p = (struct proc *)ap->a_id; 3056 struct thread *td = curthread; /* XXX */ 3057 struct vattr va; 3058 int ret, error; 3059 u_quad_t size; 3060 3061 error = NFSVOPLOCK(vp, LK_SHARED); 3062 if (error != 0) 3063 return (EBADF); 3064 if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) { 3065 if (vp->v_type != VREG) { 3066 error = EINVAL; 3067 goto out; 3068 } 3069 if ((ap->a_flags & F_POSIX) != 0) 3070 cred = p->p_ucred; 3071 else 3072 cred = td->td_ucred; 3073 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 3074 if (vp->v_iflag & VI_DOOMED) { 3075 error = EBADF; 3076 goto out; 3077 } 3078 3079 /* 3080 * If this is unlocking a write locked region, flush and 3081 * commit them before unlocking. This is required by 3082 * RFC3530 Sec. 9.3.2. 3083 */ 3084 if (ap->a_op == F_UNLCK && 3085 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id, 3086 ap->a_flags)) 3087 (void) ncl_flush(vp, MNT_WAIT, td, 1, 0); 3088 3089 /* 3090 * Loop around doing the lock op, while a blocking lock 3091 * must wait for the lock op to succeed. 3092 */ 3093 do { 3094 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op, 3095 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags); 3096 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3097 ap->a_op == F_SETLK) { 3098 NFSVOPUNLOCK(vp, 0); 3099 error = nfs_catnap(PZERO | PCATCH, ret, 3100 "ncladvl"); 3101 if (error) 3102 return (EINTR); 3103 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 3104 if (vp->v_iflag & VI_DOOMED) { 3105 error = EBADF; 3106 goto out; 3107 } 3108 } 3109 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3110 ap->a_op == F_SETLK); 3111 if (ret == NFSERR_DENIED) { 3112 error = EAGAIN; 3113 goto out; 3114 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) { 3115 error = ret; 3116 goto out; 3117 } else if (ret != 0) { 3118 error = EACCES; 3119 goto out; 3120 } 3121 3122 /* 3123 * Now, if we just got a lock, invalidate data in the buffer 3124 * cache, as required, so that the coherency conforms with 3125 * RFC3530 Sec. 9.3.2. 3126 */ 3127 if (ap->a_op == F_SETLK) { 3128 if ((np->n_flag & NMODIFIED) == 0) { 3129 np->n_attrstamp = 0; 3130 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3131 ret = VOP_GETATTR(vp, &va, cred); 3132 } 3133 if ((np->n_flag & NMODIFIED) || ret || 3134 np->n_change != va.va_filerev) { 3135 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1); 3136 np->n_attrstamp = 0; 3137 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3138 ret = VOP_GETATTR(vp, &va, cred); 3139 if (!ret) { 3140 np->n_mtime = va.va_mtime; 3141 np->n_change = va.va_filerev; 3142 } 3143 } 3144 /* Mark that a file lock has been acquired. */ 3145 mtx_lock(&np->n_mtx); 3146 np->n_flag |= NHASBEENLOCKED; 3147 mtx_unlock(&np->n_mtx); 3148 } 3149 } else if (!NFS_ISV4(vp)) { 3150 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3151 size = VTONFS(vp)->n_size; 3152 NFSVOPUNLOCK(vp, 0); 3153 error = lf_advlock(ap, &(vp->v_lockf), size); 3154 } else { 3155 if (nfs_advlock_p != NULL) 3156 error = nfs_advlock_p(ap); 3157 else { 3158 NFSVOPUNLOCK(vp, 0); 3159 error = ENOLCK; 3160 } 3161 } 3162 if (error == 0 && ap->a_op == F_SETLK) { 3163 error = NFSVOPLOCK(vp, LK_SHARED); 3164 if (error == 0) { 3165 /* Mark that a file lock has been acquired. */ 3166 mtx_lock(&np->n_mtx); 3167 np->n_flag |= NHASBEENLOCKED; 3168 mtx_unlock(&np->n_mtx); 3169 NFSVOPUNLOCK(vp, 0); 3170 } 3171 } 3172 return (error); 3173 } else 3174 error = EOPNOTSUPP; 3175 out: 3176 NFSVOPUNLOCK(vp, 0); 3177 return (error); 3178 } 3179 3180 /* 3181 * NFS advisory byte-level locks. 3182 */ 3183 static int 3184 nfs_advlockasync(struct vop_advlockasync_args *ap) 3185 { 3186 struct vnode *vp = ap->a_vp; 3187 u_quad_t size; 3188 int error; 3189 3190 if (NFS_ISV4(vp)) 3191 return (EOPNOTSUPP); 3192 error = NFSVOPLOCK(vp, LK_SHARED); 3193 if (error) 3194 return (error); 3195 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3196 size = VTONFS(vp)->n_size; 3197 NFSVOPUNLOCK(vp, 0); 3198 error = lf_advlockasync(ap, &(vp->v_lockf), size); 3199 } else { 3200 NFSVOPUNLOCK(vp, 0); 3201 error = EOPNOTSUPP; 3202 } 3203 return (error); 3204 } 3205 3206 /* 3207 * Print out the contents of an nfsnode. 3208 */ 3209 static int 3210 nfs_print(struct vop_print_args *ap) 3211 { 3212 struct vnode *vp = ap->a_vp; 3213 struct nfsnode *np = VTONFS(vp); 3214 3215 printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid, 3216 (uintmax_t)np->n_vattr.na_fsid); 3217 if (vp->v_type == VFIFO) 3218 fifo_printinfo(vp); 3219 printf("\n"); 3220 return (0); 3221 } 3222 3223 /* 3224 * This is the "real" nfs::bwrite(struct buf*). 3225 * We set B_CACHE if this is a VMIO buffer. 3226 */ 3227 int 3228 ncl_writebp(struct buf *bp, int force __unused, struct thread *td) 3229 { 3230 int oldflags, rtval; 3231 3232 BUF_ASSERT_HELD(bp); 3233 3234 if (bp->b_flags & B_INVAL) { 3235 brelse(bp); 3236 return (0); 3237 } 3238 3239 oldflags = bp->b_flags; 3240 bp->b_flags |= B_CACHE; 3241 3242 /* 3243 * Undirty the bp. We will redirty it later if the I/O fails. 3244 */ 3245 bundirty(bp); 3246 bp->b_flags &= ~B_DONE; 3247 bp->b_ioflags &= ~BIO_ERROR; 3248 bp->b_iocmd = BIO_WRITE; 3249 3250 bufobj_wref(bp->b_bufobj); 3251 curthread->td_ru.ru_oublock++; 3252 3253 /* 3254 * Note: to avoid loopback deadlocks, we do not 3255 * assign b_runningbufspace. 3256 */ 3257 vfs_busy_pages(bp, 1); 3258 3259 BUF_KERNPROC(bp); 3260 bp->b_iooffset = dbtob(bp->b_blkno); 3261 bstrategy(bp); 3262 3263 if ((oldflags & B_ASYNC) != 0) 3264 return (0); 3265 3266 rtval = bufwait(bp); 3267 if (oldflags & B_DELWRI) 3268 reassignbuf(bp); 3269 brelse(bp); 3270 return (rtval); 3271 } 3272 3273 /* 3274 * nfs special file access vnode op. 3275 * Essentially just get vattr and then imitate iaccess() since the device is 3276 * local to the client. 3277 */ 3278 static int 3279 nfsspec_access(struct vop_access_args *ap) 3280 { 3281 struct vattr *vap; 3282 struct ucred *cred = ap->a_cred; 3283 struct vnode *vp = ap->a_vp; 3284 accmode_t accmode = ap->a_accmode; 3285 struct vattr vattr; 3286 int error; 3287 3288 /* 3289 * Disallow write attempts on filesystems mounted read-only; 3290 * unless the file is a socket, fifo, or a block or character 3291 * device resident on the filesystem. 3292 */ 3293 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3294 switch (vp->v_type) { 3295 case VREG: 3296 case VDIR: 3297 case VLNK: 3298 return (EROFS); 3299 default: 3300 break; 3301 } 3302 } 3303 vap = &vattr; 3304 error = VOP_GETATTR(vp, vap, cred); 3305 if (error) 3306 goto out; 3307 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid, 3308 accmode, cred, NULL); 3309 out: 3310 return error; 3311 } 3312 3313 /* 3314 * Read wrapper for fifos. 3315 */ 3316 static int 3317 nfsfifo_read(struct vop_read_args *ap) 3318 { 3319 struct nfsnode *np = VTONFS(ap->a_vp); 3320 int error; 3321 3322 /* 3323 * Set access flag. 3324 */ 3325 mtx_lock(&np->n_mtx); 3326 np->n_flag |= NACC; 3327 vfs_timestamp(&np->n_atim); 3328 mtx_unlock(&np->n_mtx); 3329 error = fifo_specops.vop_read(ap); 3330 return error; 3331 } 3332 3333 /* 3334 * Write wrapper for fifos. 3335 */ 3336 static int 3337 nfsfifo_write(struct vop_write_args *ap) 3338 { 3339 struct nfsnode *np = VTONFS(ap->a_vp); 3340 3341 /* 3342 * Set update flag. 3343 */ 3344 mtx_lock(&np->n_mtx); 3345 np->n_flag |= NUPD; 3346 vfs_timestamp(&np->n_mtim); 3347 mtx_unlock(&np->n_mtx); 3348 return(fifo_specops.vop_write(ap)); 3349 } 3350 3351 /* 3352 * Close wrapper for fifos. 3353 * 3354 * Update the times on the nfsnode then do fifo close. 3355 */ 3356 static int 3357 nfsfifo_close(struct vop_close_args *ap) 3358 { 3359 struct vnode *vp = ap->a_vp; 3360 struct nfsnode *np = VTONFS(vp); 3361 struct vattr vattr; 3362 struct timespec ts; 3363 3364 mtx_lock(&np->n_mtx); 3365 if (np->n_flag & (NACC | NUPD)) { 3366 vfs_timestamp(&ts); 3367 if (np->n_flag & NACC) 3368 np->n_atim = ts; 3369 if (np->n_flag & NUPD) 3370 np->n_mtim = ts; 3371 np->n_flag |= NCHG; 3372 if (vrefcnt(vp) == 1 && 3373 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3374 VATTR_NULL(&vattr); 3375 if (np->n_flag & NACC) 3376 vattr.va_atime = np->n_atim; 3377 if (np->n_flag & NUPD) 3378 vattr.va_mtime = np->n_mtim; 3379 mtx_unlock(&np->n_mtx); 3380 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3381 goto out; 3382 } 3383 } 3384 mtx_unlock(&np->n_mtx); 3385 out: 3386 return (fifo_specops.vop_close(ap)); 3387 } 3388 3389 /* 3390 * Just call ncl_writebp() with the force argument set to 1. 3391 * 3392 * NOTE: B_DONE may or may not be set in a_bp on call. 3393 */ 3394 static int 3395 nfs_bwrite(struct buf *bp) 3396 { 3397 3398 return (ncl_writebp(bp, 1, curthread)); 3399 } 3400 3401 struct buf_ops buf_ops_newnfs = { 3402 .bop_name = "buf_ops_nfs", 3403 .bop_write = nfs_bwrite, 3404 .bop_strategy = bufstrategy, 3405 .bop_sync = bufsync, 3406 .bop_bdflush = bufbdflush, 3407 }; 3408 3409 static int 3410 nfs_getacl(struct vop_getacl_args *ap) 3411 { 3412 int error; 3413 3414 if (ap->a_type != ACL_TYPE_NFS4) 3415 return (EOPNOTSUPP); 3416 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3417 NULL); 3418 if (error > NFSERR_STALE) { 3419 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3420 error = EPERM; 3421 } 3422 return (error); 3423 } 3424 3425 static int 3426 nfs_setacl(struct vop_setacl_args *ap) 3427 { 3428 int error; 3429 3430 if (ap->a_type != ACL_TYPE_NFS4) 3431 return (EOPNOTSUPP); 3432 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3433 NULL); 3434 if (error > NFSERR_STALE) { 3435 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3436 error = EPERM; 3437 } 3438 return (error); 3439 } 3440 3441 /* 3442 * Return POSIX pathconf information applicable to nfs filesystems. 3443 */ 3444 static int 3445 nfs_pathconf(struct vop_pathconf_args *ap) 3446 { 3447 struct nfsv3_pathconf pc; 3448 struct nfsvattr nfsva; 3449 struct vnode *vp = ap->a_vp; 3450 struct thread *td = curthread; 3451 int attrflag, error; 3452 3453 if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX || 3454 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED || 3455 ap->a_name == _PC_NO_TRUNC)) || 3456 (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) { 3457 /* 3458 * Since only the above 4 a_names are returned by the NFSv3 3459 * Pathconf RPC, there is no point in doing it for others. 3460 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can 3461 * be used for _PC_NFS4_ACL as well. 3462 */ 3463 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva, 3464 &attrflag, NULL); 3465 if (attrflag != 0) 3466 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 3467 1); 3468 if (error != 0) 3469 return (error); 3470 } else { 3471 /* 3472 * For NFSv2 (or NFSv3 when not one of the above 4 a_names), 3473 * just fake them. 3474 */ 3475 pc.pc_linkmax = NFS_LINK_MAX; 3476 pc.pc_namemax = NFS_MAXNAMLEN; 3477 pc.pc_notrunc = 1; 3478 pc.pc_chownrestricted = 1; 3479 pc.pc_caseinsensitive = 0; 3480 pc.pc_casepreserving = 1; 3481 error = 0; 3482 } 3483 switch (ap->a_name) { 3484 case _PC_LINK_MAX: 3485 #ifdef _LP64 3486 *ap->a_retval = pc.pc_linkmax; 3487 #else 3488 *ap->a_retval = MIN(LONG_MAX, pc.pc_linkmax); 3489 #endif 3490 break; 3491 case _PC_NAME_MAX: 3492 *ap->a_retval = pc.pc_namemax; 3493 break; 3494 case _PC_PIPE_BUF: 3495 if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO) 3496 *ap->a_retval = PIPE_BUF; 3497 else 3498 error = EINVAL; 3499 break; 3500 case _PC_CHOWN_RESTRICTED: 3501 *ap->a_retval = pc.pc_chownrestricted; 3502 break; 3503 case _PC_NO_TRUNC: 3504 *ap->a_retval = pc.pc_notrunc; 3505 break; 3506 case _PC_ACL_NFS4: 3507 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 && 3508 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL)) 3509 *ap->a_retval = 1; 3510 else 3511 *ap->a_retval = 0; 3512 break; 3513 case _PC_ACL_PATH_MAX: 3514 if (NFS_ISV4(vp)) 3515 *ap->a_retval = ACL_MAX_ENTRIES; 3516 else 3517 *ap->a_retval = 3; 3518 break; 3519 case _PC_PRIO_IO: 3520 *ap->a_retval = 0; 3521 break; 3522 case _PC_SYNC_IO: 3523 *ap->a_retval = 0; 3524 break; 3525 case _PC_ALLOC_SIZE_MIN: 3526 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize; 3527 break; 3528 case _PC_FILESIZEBITS: 3529 if (NFS_ISV34(vp)) 3530 *ap->a_retval = 64; 3531 else 3532 *ap->a_retval = 32; 3533 break; 3534 case _PC_REC_INCR_XFER_SIZE: 3535 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 3536 break; 3537 case _PC_REC_MAX_XFER_SIZE: 3538 *ap->a_retval = -1; /* means ``unlimited'' */ 3539 break; 3540 case _PC_REC_MIN_XFER_SIZE: 3541 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 3542 break; 3543 case _PC_REC_XFER_ALIGN: 3544 *ap->a_retval = PAGE_SIZE; 3545 break; 3546 case _PC_SYMLINK_MAX: 3547 *ap->a_retval = NFS_MAXPATHLEN; 3548 break; 3549 3550 default: 3551 error = vop_stdpathconf(ap); 3552 break; 3553 } 3554 return (error); 3555 } 3556 3557