nfs_clvnops.c revision 221973
1/*- 2 * Copyright (c) 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * Rick Macklem at The University of Guelph. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 4. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * from nfs_vnops.c 8.16 (Berkeley) 5/27/95 33 */ 34 35#include <sys/cdefs.h> 36__FBSDID("$FreeBSD: head/sys/fs/nfsclient/nfs_clvnops.c 221973 2011-05-15 20:52:43Z rmacklem $"); 37 38/* 39 * vnode op calls for Sun NFS version 2, 3 and 4 40 */ 41 42#include "opt_inet.h" 43 44#include <sys/param.h> 45#include <sys/kernel.h> 46#include <sys/systm.h> 47#include <sys/resourcevar.h> 48#include <sys/proc.h> 49#include <sys/mount.h> 50#include <sys/bio.h> 51#include <sys/buf.h> 52#include <sys/jail.h> 53#include <sys/malloc.h> 54#include <sys/mbuf.h> 55#include <sys/namei.h> 56#include <sys/socket.h> 57#include <sys/vnode.h> 58#include <sys/dirent.h> 59#include <sys/fcntl.h> 60#include <sys/lockf.h> 61#include <sys/stat.h> 62#include <sys/sysctl.h> 63#include <sys/signalvar.h> 64 65#include <vm/vm.h> 66#include <vm/vm_object.h> 67#include <vm/vm_extern.h> 68#include <vm/vm_object.h> 69 70#include <fs/nfs/nfsport.h> 71#include <fs/nfsclient/nfsnode.h> 72#include <fs/nfsclient/nfsmount.h> 73#include <fs/nfsclient/nfs.h> 74 75#include <net/if.h> 76#include <netinet/in.h> 77#include <netinet/in_var.h> 78 79#include <nfs/nfs_lock.h> 80 81/* Defs */ 82#define TRUE 1 83#define FALSE 0 84 85extern struct nfsstats newnfsstats; 86extern int nfsrv_useacl; 87MALLOC_DECLARE(M_NEWNFSREQ); 88 89/* 90 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these 91 * calls are not in getblk() and brelse() so that they would not be necessary 92 * here. 93 */ 94#ifndef B_VMIO 95#define vfs_busy_pages(bp, f) 96#endif 97 98static vop_read_t nfsfifo_read; 99static vop_write_t nfsfifo_write; 100static vop_close_t nfsfifo_close; 101static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *, 102 struct thread *); 103static vop_lookup_t nfs_lookup; 104static vop_create_t nfs_create; 105static vop_mknod_t nfs_mknod; 106static vop_open_t nfs_open; 107static vop_pathconf_t nfs_pathconf; 108static vop_close_t nfs_close; 109static vop_access_t nfs_access; 110static vop_getattr_t nfs_getattr; 111static vop_setattr_t nfs_setattr; 112static vop_read_t nfs_read; 113static vop_fsync_t nfs_fsync; 114static vop_remove_t nfs_remove; 115static vop_link_t nfs_link; 116static vop_rename_t nfs_rename; 117static vop_mkdir_t nfs_mkdir; 118static vop_rmdir_t nfs_rmdir; 119static vop_symlink_t nfs_symlink; 120static vop_readdir_t nfs_readdir; 121static vop_strategy_t nfs_strategy; 122static vop_lock1_t nfs_lock1; 123static int nfs_lookitup(struct vnode *, char *, int, 124 struct ucred *, struct thread *, struct nfsnode **); 125static int nfs_sillyrename(struct vnode *, struct vnode *, 126 struct componentname *); 127static vop_access_t nfsspec_access; 128static vop_readlink_t nfs_readlink; 129static vop_print_t nfs_print; 130static vop_advlock_t nfs_advlock; 131static vop_advlockasync_t nfs_advlockasync; 132static vop_getacl_t nfs_getacl; 133static vop_setacl_t nfs_setacl; 134 135/* 136 * Global vfs data structures for nfs 137 */ 138struct vop_vector newnfs_vnodeops = { 139 .vop_default = &default_vnodeops, 140 .vop_access = nfs_access, 141 .vop_advlock = nfs_advlock, 142 .vop_advlockasync = nfs_advlockasync, 143 .vop_close = nfs_close, 144 .vop_create = nfs_create, 145 .vop_fsync = nfs_fsync, 146 .vop_getattr = nfs_getattr, 147 .vop_getpages = ncl_getpages, 148 .vop_putpages = ncl_putpages, 149 .vop_inactive = ncl_inactive, 150 .vop_link = nfs_link, 151 .vop_lock1 = nfs_lock1, 152 .vop_lookup = nfs_lookup, 153 .vop_mkdir = nfs_mkdir, 154 .vop_mknod = nfs_mknod, 155 .vop_open = nfs_open, 156 .vop_pathconf = nfs_pathconf, 157 .vop_print = nfs_print, 158 .vop_read = nfs_read, 159 .vop_readdir = nfs_readdir, 160 .vop_readlink = nfs_readlink, 161 .vop_reclaim = ncl_reclaim, 162 .vop_remove = nfs_remove, 163 .vop_rename = nfs_rename, 164 .vop_rmdir = nfs_rmdir, 165 .vop_setattr = nfs_setattr, 166 .vop_strategy = nfs_strategy, 167 .vop_symlink = nfs_symlink, 168 .vop_write = ncl_write, 169 .vop_getacl = nfs_getacl, 170 .vop_setacl = nfs_setacl, 171}; 172 173struct vop_vector newnfs_fifoops = { 174 .vop_default = &fifo_specops, 175 .vop_access = nfsspec_access, 176 .vop_close = nfsfifo_close, 177 .vop_fsync = nfs_fsync, 178 .vop_getattr = nfs_getattr, 179 .vop_inactive = ncl_inactive, 180 .vop_print = nfs_print, 181 .vop_read = nfsfifo_read, 182 .vop_reclaim = ncl_reclaim, 183 .vop_setattr = nfs_setattr, 184 .vop_write = nfsfifo_write, 185}; 186 187static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, 188 struct componentname *cnp, struct vattr *vap); 189static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 190 int namelen, struct ucred *cred, struct thread *td); 191static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, 192 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp, 193 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td); 194static int nfs_renameit(struct vnode *sdvp, struct vnode *svp, 195 struct componentname *scnp, struct sillyrename *sp); 196 197/* 198 * Global variables 199 */ 200#define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1)) 201 202SYSCTL_DECL(_vfs_nfs); 203 204static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO; 205SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 206 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 207 208static int nfs_prime_access_cache = 0; 209SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW, 210 &nfs_prime_access_cache, 0, 211 "Prime NFS ACCESS cache when fetching attributes"); 212 213static int newnfs_commit_on_close = 0; 214SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW, 215 &newnfs_commit_on_close, 0, "write+commit on close, else only write"); 216 217static int nfs_clean_pages_on_close = 1; 218SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW, 219 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close"); 220 221int newnfs_directio_enable = 0; 222SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW, 223 &newnfs_directio_enable, 0, "Enable NFS directio"); 224 225/* 226 * This sysctl allows other processes to mmap a file that has been opened 227 * O_DIRECT by a process. In general, having processes mmap the file while 228 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow 229 * this by default to prevent DoS attacks - to prevent a malicious user from 230 * opening up files O_DIRECT preventing other users from mmap'ing these 231 * files. "Protected" environments where stricter consistency guarantees are 232 * required can disable this knob. The process that opened the file O_DIRECT 233 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not 234 * meaningful. 235 */ 236int newnfs_directio_allow_mmap = 1; 237SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW, 238 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens"); 239 240#if 0 241SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD, 242 &newnfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count"); 243 244SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD, 245 &newnfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count"); 246#endif 247 248#define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \ 249 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \ 250 | NFSACCESS_DELETE | NFSACCESS_LOOKUP) 251 252/* 253 * SMP Locking Note : 254 * The list of locks after the description of the lock is the ordering 255 * of other locks acquired with the lock held. 256 * np->n_mtx : Protects the fields in the nfsnode. 257 VM Object Lock 258 VI_MTX (acquired indirectly) 259 * nmp->nm_mtx : Protects the fields in the nfsmount. 260 rep->r_mtx 261 * ncl_iod_mutex : Global lock, protects shared nfsiod state. 262 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list. 263 nmp->nm_mtx 264 rep->r_mtx 265 * rep->r_mtx : Protects the fields in an nfsreq. 266 */ 267 268static int 269nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td, 270 struct ucred *cred, u_int32_t *retmode) 271{ 272 int error = 0, attrflag, i, lrupos; 273 u_int32_t rmode; 274 struct nfsnode *np = VTONFS(vp); 275 struct nfsvattr nfsva; 276 277 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag, 278 &rmode, NULL); 279 if (attrflag) 280 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 281 if (!error) { 282 lrupos = 0; 283 mtx_lock(&np->n_mtx); 284 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 285 if (np->n_accesscache[i].uid == cred->cr_uid) { 286 np->n_accesscache[i].mode = rmode; 287 np->n_accesscache[i].stamp = time_second; 288 break; 289 } 290 if (i > 0 && np->n_accesscache[i].stamp < 291 np->n_accesscache[lrupos].stamp) 292 lrupos = i; 293 } 294 if (i == NFS_ACCESSCACHESIZE) { 295 np->n_accesscache[lrupos].uid = cred->cr_uid; 296 np->n_accesscache[lrupos].mode = rmode; 297 np->n_accesscache[lrupos].stamp = time_second; 298 } 299 mtx_unlock(&np->n_mtx); 300 if (retmode != NULL) 301 *retmode = rmode; 302 } else if (NFS_ISV4(vp)) { 303 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 304 } 305 return (error); 306} 307 308/* 309 * nfs access vnode op. 310 * For nfs version 2, just return ok. File accesses may fail later. 311 * For nfs version 3, use the access rpc to check accessibility. If file modes 312 * are changed on the server, accesses might still fail later. 313 */ 314static int 315nfs_access(struct vop_access_args *ap) 316{ 317 struct vnode *vp = ap->a_vp; 318 int error = 0, i, gotahit; 319 u_int32_t mode, wmode, rmode; 320 int v34 = NFS_ISV34(vp); 321 struct nfsnode *np = VTONFS(vp); 322 323 /* 324 * Disallow write attempts on filesystems mounted read-only; 325 * unless the file is a socket, fifo, or a block or character 326 * device resident on the filesystem. 327 */ 328 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS | 329 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL | 330 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) { 331 switch (vp->v_type) { 332 case VREG: 333 case VDIR: 334 case VLNK: 335 return (EROFS); 336 default: 337 break; 338 } 339 } 340 /* 341 * For nfs v3 or v4, check to see if we have done this recently, and if 342 * so return our cached result instead of making an ACCESS call. 343 * If not, do an access rpc, otherwise you are stuck emulating 344 * ufs_access() locally using the vattr. This may not be correct, 345 * since the server may apply other access criteria such as 346 * client uid-->server uid mapping that we do not know about. 347 */ 348 if (v34) { 349 if (ap->a_accmode & VREAD) 350 mode = NFSACCESS_READ; 351 else 352 mode = 0; 353 if (vp->v_type != VDIR) { 354 if (ap->a_accmode & VWRITE) 355 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 356 if (ap->a_accmode & VAPPEND) 357 mode |= NFSACCESS_EXTEND; 358 if (ap->a_accmode & VEXEC) 359 mode |= NFSACCESS_EXECUTE; 360 if (ap->a_accmode & VDELETE) 361 mode |= NFSACCESS_DELETE; 362 } else { 363 if (ap->a_accmode & VWRITE) 364 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 365 if (ap->a_accmode & VAPPEND) 366 mode |= NFSACCESS_EXTEND; 367 if (ap->a_accmode & VEXEC) 368 mode |= NFSACCESS_LOOKUP; 369 if (ap->a_accmode & VDELETE) 370 mode |= NFSACCESS_DELETE; 371 if (ap->a_accmode & VDELETE_CHILD) 372 mode |= NFSACCESS_MODIFY; 373 } 374 /* XXX safety belt, only make blanket request if caching */ 375 if (nfsaccess_cache_timeout > 0) { 376 wmode = NFSACCESS_READ | NFSACCESS_MODIFY | 377 NFSACCESS_EXTEND | NFSACCESS_EXECUTE | 378 NFSACCESS_DELETE | NFSACCESS_LOOKUP; 379 } else { 380 wmode = mode; 381 } 382 383 /* 384 * Does our cached result allow us to give a definite yes to 385 * this request? 386 */ 387 gotahit = 0; 388 mtx_lock(&np->n_mtx); 389 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 390 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) { 391 if (time_second < (np->n_accesscache[i].stamp 392 + nfsaccess_cache_timeout) && 393 (np->n_accesscache[i].mode & mode) == mode) { 394 NFSINCRGLOBAL(newnfsstats.accesscache_hits); 395 gotahit = 1; 396 } 397 break; 398 } 399 } 400 mtx_unlock(&np->n_mtx); 401 if (gotahit == 0) { 402 /* 403 * Either a no, or a don't know. Go to the wire. 404 */ 405 NFSINCRGLOBAL(newnfsstats.accesscache_misses); 406 error = nfs34_access_otw(vp, wmode, ap->a_td, 407 ap->a_cred, &rmode); 408 if (!error && 409 (rmode & mode) != mode) 410 error = EACCES; 411 } 412 return (error); 413 } else { 414 if ((error = nfsspec_access(ap)) != 0) { 415 return (error); 416 } 417 /* 418 * Attempt to prevent a mapped root from accessing a file 419 * which it shouldn't. We try to read a byte from the file 420 * if the user is root and the file is not zero length. 421 * After calling nfsspec_access, we should have the correct 422 * file size cached. 423 */ 424 mtx_lock(&np->n_mtx); 425 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD) 426 && VTONFS(vp)->n_size > 0) { 427 struct iovec aiov; 428 struct uio auio; 429 char buf[1]; 430 431 mtx_unlock(&np->n_mtx); 432 aiov.iov_base = buf; 433 aiov.iov_len = 1; 434 auio.uio_iov = &aiov; 435 auio.uio_iovcnt = 1; 436 auio.uio_offset = 0; 437 auio.uio_resid = 1; 438 auio.uio_segflg = UIO_SYSSPACE; 439 auio.uio_rw = UIO_READ; 440 auio.uio_td = ap->a_td; 441 442 if (vp->v_type == VREG) 443 error = ncl_readrpc(vp, &auio, ap->a_cred); 444 else if (vp->v_type == VDIR) { 445 char* bp; 446 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 447 aiov.iov_base = bp; 448 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 449 error = ncl_readdirrpc(vp, &auio, ap->a_cred, 450 ap->a_td); 451 free(bp, M_TEMP); 452 } else if (vp->v_type == VLNK) 453 error = ncl_readlinkrpc(vp, &auio, ap->a_cred); 454 else 455 error = EACCES; 456 } else 457 mtx_unlock(&np->n_mtx); 458 return (error); 459 } 460} 461 462 463/* 464 * nfs open vnode op 465 * Check to see if the type is ok 466 * and that deletion is not in progress. 467 * For paged in text files, you will need to flush the page cache 468 * if consistency is lost. 469 */ 470/* ARGSUSED */ 471static int 472nfs_open(struct vop_open_args *ap) 473{ 474 struct vnode *vp = ap->a_vp; 475 struct nfsnode *np = VTONFS(vp); 476 struct vattr vattr; 477 int error; 478 int fmode = ap->a_mode; 479 480 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) 481 return (EOPNOTSUPP); 482 483 /* 484 * For NFSv4, we need to do the Open Op before cache validation, 485 * so that we conform to RFC3530 Sec. 9.3.1. 486 */ 487 if (NFS_ISV4(vp)) { 488 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td); 489 if (error) { 490 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 491 (gid_t)0); 492 return (error); 493 } 494 } 495 496 /* 497 * Now, if this Open will be doing reading, re-validate/flush the 498 * cache, so that Close/Open coherency is maintained. 499 */ 500 mtx_lock(&np->n_mtx); 501 if (np->n_flag & NMODIFIED) { 502 mtx_unlock(&np->n_mtx); 503 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 504 if (error == EINTR || error == EIO) { 505 if (NFS_ISV4(vp)) 506 (void) nfsrpc_close(vp, 0, ap->a_td); 507 return (error); 508 } 509 mtx_lock(&np->n_mtx); 510 np->n_attrstamp = 0; 511 if (vp->v_type == VDIR) 512 np->n_direofoffset = 0; 513 mtx_unlock(&np->n_mtx); 514 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 515 if (error) { 516 if (NFS_ISV4(vp)) 517 (void) nfsrpc_close(vp, 0, ap->a_td); 518 return (error); 519 } 520 mtx_lock(&np->n_mtx); 521 np->n_mtime = vattr.va_mtime; 522 if (NFS_ISV4(vp)) 523 np->n_change = vattr.va_filerev; 524 } else { 525 mtx_unlock(&np->n_mtx); 526 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 527 if (error) { 528 if (NFS_ISV4(vp)) 529 (void) nfsrpc_close(vp, 0, ap->a_td); 530 return (error); 531 } 532 mtx_lock(&np->n_mtx); 533 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) || 534 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 535 if (vp->v_type == VDIR) 536 np->n_direofoffset = 0; 537 mtx_unlock(&np->n_mtx); 538 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 539 if (error == EINTR || error == EIO) { 540 if (NFS_ISV4(vp)) 541 (void) nfsrpc_close(vp, 0, ap->a_td); 542 return (error); 543 } 544 mtx_lock(&np->n_mtx); 545 np->n_mtime = vattr.va_mtime; 546 if (NFS_ISV4(vp)) 547 np->n_change = vattr.va_filerev; 548 } 549 } 550 551 /* 552 * If the object has >= 1 O_DIRECT active opens, we disable caching. 553 */ 554 if (newnfs_directio_enable && (fmode & O_DIRECT) && 555 (vp->v_type == VREG)) { 556 if (np->n_directio_opens == 0) { 557 mtx_unlock(&np->n_mtx); 558 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 559 if (error) { 560 if (NFS_ISV4(vp)) 561 (void) nfsrpc_close(vp, 0, ap->a_td); 562 return (error); 563 } 564 mtx_lock(&np->n_mtx); 565 np->n_flag |= NNONCACHE; 566 } 567 np->n_directio_opens++; 568 } 569 mtx_unlock(&np->n_mtx); 570 vnode_create_vobject(vp, vattr.va_size, ap->a_td); 571 return (0); 572} 573 574/* 575 * nfs close vnode op 576 * What an NFS client should do upon close after writing is a debatable issue. 577 * Most NFS clients push delayed writes to the server upon close, basically for 578 * two reasons: 579 * 1 - So that any write errors may be reported back to the client process 580 * doing the close system call. By far the two most likely errors are 581 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 582 * 2 - To put a worst case upper bound on cache inconsistency between 583 * multiple clients for the file. 584 * There is also a consistency problem for Version 2 of the protocol w.r.t. 585 * not being able to tell if other clients are writing a file concurrently, 586 * since there is no way of knowing if the changed modify time in the reply 587 * is only due to the write for this client. 588 * (NFS Version 3 provides weak cache consistency data in the reply that 589 * should be sufficient to detect and handle this case.) 590 * 591 * The current code does the following: 592 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 593 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 594 * or commit them (this satisfies 1 and 2 except for the 595 * case where the server crashes after this close but 596 * before the commit RPC, which is felt to be "good 597 * enough". Changing the last argument to ncl_flush() to 598 * a 1 would force a commit operation, if it is felt a 599 * commit is necessary now. 600 * for NFS Version 4 - flush the dirty buffers and commit them, if 601 * nfscl_mustflush() says this is necessary. 602 * It is necessary if there is no write delegation held, 603 * in order to satisfy open/close coherency. 604 * If the file isn't cached on local stable storage, 605 * it may be necessary in order to detect "out of space" 606 * errors from the server, if the write delegation 607 * issued by the server doesn't allow the file to grow. 608 */ 609/* ARGSUSED */ 610static int 611nfs_close(struct vop_close_args *ap) 612{ 613 struct vnode *vp = ap->a_vp; 614 struct nfsnode *np = VTONFS(vp); 615 struct nfsvattr nfsva; 616 struct ucred *cred; 617 int error = 0, ret, localcred = 0; 618 int fmode = ap->a_fflag; 619 620 if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF)) 621 return (0); 622 /* 623 * During shutdown, a_cred isn't valid, so just use root. 624 */ 625 if (ap->a_cred == NOCRED) { 626 cred = newnfs_getcred(); 627 localcred = 1; 628 } else { 629 cred = ap->a_cred; 630 } 631 if (vp->v_type == VREG) { 632 /* 633 * Examine and clean dirty pages, regardless of NMODIFIED. 634 * This closes a major hole in close-to-open consistency. 635 * We want to push out all dirty pages (and buffers) on 636 * close, regardless of whether they were dirtied by 637 * mmap'ed writes or via write(). 638 */ 639 if (nfs_clean_pages_on_close && vp->v_object) { 640 VM_OBJECT_LOCK(vp->v_object); 641 vm_object_page_clean(vp->v_object, 0, 0, 0); 642 VM_OBJECT_UNLOCK(vp->v_object); 643 } 644 mtx_lock(&np->n_mtx); 645 if (np->n_flag & NMODIFIED) { 646 mtx_unlock(&np->n_mtx); 647 if (NFS_ISV3(vp)) { 648 /* 649 * Under NFSv3 we have dirty buffers to dispose of. We 650 * must flush them to the NFS server. We have the option 651 * of waiting all the way through the commit rpc or just 652 * waiting for the initial write. The default is to only 653 * wait through the initial write so the data is in the 654 * server's cache, which is roughly similar to the state 655 * a standard disk subsystem leaves the file in on close(). 656 * 657 * We cannot clear the NMODIFIED bit in np->n_flag due to 658 * potential races with other processes, and certainly 659 * cannot clear it if we don't commit. 660 * These races occur when there is no longer the old 661 * traditional vnode locking implemented for Vnode Ops. 662 */ 663 int cm = newnfs_commit_on_close ? 1 : 0; 664 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm, 0); 665 /* np->n_flag &= ~NMODIFIED; */ 666 } else if (NFS_ISV4(vp)) { 667 if (nfscl_mustflush(vp) != 0) { 668 int cm = newnfs_commit_on_close ? 1 : 0; 669 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, 670 cm, 0); 671 /* 672 * as above w.r.t races when clearing 673 * NMODIFIED. 674 * np->n_flag &= ~NMODIFIED; 675 */ 676 } 677 } else 678 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 679 mtx_lock(&np->n_mtx); 680 } 681 /* 682 * Invalidate the attribute cache in all cases. 683 * An open is going to fetch fresh attrs any way, other procs 684 * on this node that have file open will be forced to do an 685 * otw attr fetch, but this is safe. 686 * --> A user found that their RPC count dropped by 20% when 687 * this was commented out and I can't see any requirement 688 * for it, so I've disabled it when negative lookups are 689 * enabled. (What does this have to do with negative lookup 690 * caching? Well nothing, except it was reported by the 691 * same user that needed negative lookup caching and I wanted 692 * there to be a way to disable it to see if it 693 * is the cause of some caching/coherency issue that might 694 * crop up.) 695 */ 696 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) 697 np->n_attrstamp = 0; 698 if (np->n_flag & NWRITEERR) { 699 np->n_flag &= ~NWRITEERR; 700 error = np->n_error; 701 } 702 mtx_unlock(&np->n_mtx); 703 } 704 705 if (NFS_ISV4(vp)) { 706 /* 707 * Get attributes so "change" is up to date. 708 */ 709 if (error == 0 && nfscl_mustflush(vp) != 0) { 710 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva, 711 NULL); 712 if (!ret) { 713 np->n_change = nfsva.na_filerev; 714 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 715 NULL, 0, 0); 716 } 717 } 718 719 /* 720 * and do the close. 721 */ 722 ret = nfsrpc_close(vp, 0, ap->a_td); 723 if (!error && ret) 724 error = ret; 725 if (error) 726 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 727 (gid_t)0); 728 } 729 if (newnfs_directio_enable) 730 KASSERT((np->n_directio_asyncwr == 0), 731 ("nfs_close: dirty unflushed (%d) directio buffers\n", 732 np->n_directio_asyncwr)); 733 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) { 734 mtx_lock(&np->n_mtx); 735 KASSERT((np->n_directio_opens > 0), 736 ("nfs_close: unexpectedly value (0) of n_directio_opens\n")); 737 np->n_directio_opens--; 738 if (np->n_directio_opens == 0) 739 np->n_flag &= ~NNONCACHE; 740 mtx_unlock(&np->n_mtx); 741 } 742 if (localcred) 743 NFSFREECRED(cred); 744 return (error); 745} 746 747/* 748 * nfs getattr call from vfs. 749 */ 750static int 751nfs_getattr(struct vop_getattr_args *ap) 752{ 753 struct vnode *vp = ap->a_vp; 754 struct thread *td = curthread; /* XXX */ 755 struct nfsnode *np = VTONFS(vp); 756 int error = 0; 757 struct nfsvattr nfsva; 758 struct vattr *vap = ap->a_vap; 759 struct vattr vattr; 760 761 /* 762 * Update local times for special files. 763 */ 764 mtx_lock(&np->n_mtx); 765 if (np->n_flag & (NACC | NUPD)) 766 np->n_flag |= NCHG; 767 mtx_unlock(&np->n_mtx); 768 /* 769 * First look in the cache. 770 */ 771 if (ncl_getattrcache(vp, &vattr) == 0) { 772 vap->va_type = vattr.va_type; 773 vap->va_mode = vattr.va_mode; 774 vap->va_nlink = vattr.va_nlink; 775 vap->va_uid = vattr.va_uid; 776 vap->va_gid = vattr.va_gid; 777 vap->va_fsid = vattr.va_fsid; 778 vap->va_fileid = vattr.va_fileid; 779 vap->va_size = vattr.va_size; 780 vap->va_blocksize = vattr.va_blocksize; 781 vap->va_atime = vattr.va_atime; 782 vap->va_mtime = vattr.va_mtime; 783 vap->va_ctime = vattr.va_ctime; 784 vap->va_gen = vattr.va_gen; 785 vap->va_flags = vattr.va_flags; 786 vap->va_rdev = vattr.va_rdev; 787 vap->va_bytes = vattr.va_bytes; 788 vap->va_filerev = vattr.va_filerev; 789 /* 790 * Get the local modify time for the case of a write 791 * delegation. 792 */ 793 nfscl_deleggetmodtime(vp, &vap->va_mtime); 794 return (0); 795 } 796 797 if (NFS_ISV34(vp) && nfs_prime_access_cache && 798 nfsaccess_cache_timeout > 0) { 799 NFSINCRGLOBAL(newnfsstats.accesscache_misses); 800 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL); 801 if (ncl_getattrcache(vp, ap->a_vap) == 0) { 802 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime); 803 return (0); 804 } 805 } 806 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL); 807 if (!error) 808 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0); 809 if (!error) { 810 /* 811 * Get the local modify time for the case of a write 812 * delegation. 813 */ 814 nfscl_deleggetmodtime(vp, &vap->va_mtime); 815 } else if (NFS_ISV4(vp)) { 816 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 817 } 818 return (error); 819} 820 821/* 822 * nfs setattr call. 823 */ 824static int 825nfs_setattr(struct vop_setattr_args *ap) 826{ 827 struct vnode *vp = ap->a_vp; 828 struct nfsnode *np = VTONFS(vp); 829 struct thread *td = curthread; /* XXX */ 830 struct vattr *vap = ap->a_vap; 831 int error = 0; 832 u_quad_t tsize; 833 834#ifndef nolint 835 tsize = (u_quad_t)0; 836#endif 837 838 /* 839 * Setting of flags and marking of atimes are not supported. 840 */ 841 if (vap->va_flags != VNOVAL) 842 return (EOPNOTSUPP); 843 844 /* 845 * Disallow write attempts if the filesystem is mounted read-only. 846 */ 847 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 848 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 849 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 850 (vp->v_mount->mnt_flag & MNT_RDONLY)) 851 return (EROFS); 852 if (vap->va_size != VNOVAL) { 853 switch (vp->v_type) { 854 case VDIR: 855 return (EISDIR); 856 case VCHR: 857 case VBLK: 858 case VSOCK: 859 case VFIFO: 860 if (vap->va_mtime.tv_sec == VNOVAL && 861 vap->va_atime.tv_sec == VNOVAL && 862 vap->va_mode == (mode_t)VNOVAL && 863 vap->va_uid == (uid_t)VNOVAL && 864 vap->va_gid == (gid_t)VNOVAL) 865 return (0); 866 vap->va_size = VNOVAL; 867 break; 868 default: 869 /* 870 * Disallow write attempts if the filesystem is 871 * mounted read-only. 872 */ 873 if (vp->v_mount->mnt_flag & MNT_RDONLY) 874 return (EROFS); 875 /* 876 * We run vnode_pager_setsize() early (why?), 877 * we must set np->n_size now to avoid vinvalbuf 878 * V_SAVE races that might setsize a lower 879 * value. 880 */ 881 mtx_lock(&np->n_mtx); 882 tsize = np->n_size; 883 mtx_unlock(&np->n_mtx); 884 error = ncl_meta_setsize(vp, ap->a_cred, td, 885 vap->va_size); 886 mtx_lock(&np->n_mtx); 887 if (np->n_flag & NMODIFIED) { 888 tsize = np->n_size; 889 mtx_unlock(&np->n_mtx); 890 if (vap->va_size == 0) 891 error = ncl_vinvalbuf(vp, 0, td, 1); 892 else 893 error = ncl_vinvalbuf(vp, V_SAVE, td, 1); 894 if (error) { 895 vnode_pager_setsize(vp, tsize); 896 return (error); 897 } 898 /* 899 * Call nfscl_delegmodtime() to set the modify time 900 * locally, as required. 901 */ 902 nfscl_delegmodtime(vp); 903 } else 904 mtx_unlock(&np->n_mtx); 905 /* 906 * np->n_size has already been set to vap->va_size 907 * in ncl_meta_setsize(). We must set it again since 908 * nfs_loadattrcache() could be called through 909 * ncl_meta_setsize() and could modify np->n_size. 910 */ 911 mtx_lock(&np->n_mtx); 912 np->n_vattr.na_size = np->n_size = vap->va_size; 913 mtx_unlock(&np->n_mtx); 914 }; 915 } else { 916 mtx_lock(&np->n_mtx); 917 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) && 918 (np->n_flag & NMODIFIED) && vp->v_type == VREG) { 919 mtx_unlock(&np->n_mtx); 920 if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 && 921 (error == EINTR || error == EIO)) 922 return (error); 923 } else 924 mtx_unlock(&np->n_mtx); 925 } 926 error = nfs_setattrrpc(vp, vap, ap->a_cred, td); 927 if (error && vap->va_size != VNOVAL) { 928 mtx_lock(&np->n_mtx); 929 np->n_size = np->n_vattr.na_size = tsize; 930 vnode_pager_setsize(vp, tsize); 931 mtx_unlock(&np->n_mtx); 932 } 933 return (error); 934} 935 936/* 937 * Do an nfs setattr rpc. 938 */ 939static int 940nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred, 941 struct thread *td) 942{ 943 struct nfsnode *np = VTONFS(vp); 944 int error, ret, attrflag, i; 945 struct nfsvattr nfsva; 946 947 if (NFS_ISV34(vp)) { 948 mtx_lock(&np->n_mtx); 949 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) 950 np->n_accesscache[i].stamp = 0; 951 np->n_flag |= NDELEGMOD; 952 mtx_unlock(&np->n_mtx); 953 } 954 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag, 955 NULL); 956 if (attrflag) { 957 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 958 if (ret && !error) 959 error = ret; 960 } 961 if (error && NFS_ISV4(vp)) 962 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid); 963 return (error); 964} 965 966/* 967 * nfs lookup call, one step at a time... 968 * First look in cache 969 * If not found, unlock the directory nfsnode and do the rpc 970 */ 971static int 972nfs_lookup(struct vop_lookup_args *ap) 973{ 974 struct componentname *cnp = ap->a_cnp; 975 struct vnode *dvp = ap->a_dvp; 976 struct vnode **vpp = ap->a_vpp; 977 struct mount *mp = dvp->v_mount; 978 int flags = cnp->cn_flags; 979 struct vnode *newvp; 980 struct nfsmount *nmp; 981 struct nfsnode *np, *newnp; 982 int error = 0, attrflag, dattrflag, ltype; 983 struct thread *td = cnp->cn_thread; 984 struct nfsfh *nfhp; 985 struct nfsvattr dnfsva, nfsva; 986 struct vattr vattr; 987 struct timespec dmtime; 988 989 *vpp = NULLVP; 990 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) && 991 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 992 return (EROFS); 993 if (dvp->v_type != VDIR) 994 return (ENOTDIR); 995 nmp = VFSTONFS(mp); 996 np = VTONFS(dvp); 997 998 /* For NFSv4, wait until any remove is done. */ 999 mtx_lock(&np->n_mtx); 1000 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) { 1001 np->n_flag |= NREMOVEWANT; 1002 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0); 1003 } 1004 mtx_unlock(&np->n_mtx); 1005 1006 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) 1007 return (error); 1008 error = cache_lookup(dvp, vpp, cnp); 1009 if (error > 0 && error != ENOENT) 1010 return (error); 1011 if (error == -1) { 1012 /* 1013 * We only accept a positive hit in the cache if the 1014 * change time of the file matches our cached copy. 1015 * Otherwise, we discard the cache entry and fallback 1016 * to doing a lookup RPC. 1017 * 1018 * To better handle stale file handles and attributes, 1019 * clear the attribute cache of this node if it is a 1020 * leaf component, part of an open() call, and not 1021 * locally modified before fetching the attributes. 1022 * This should allow stale file handles to be detected 1023 * here where we can fall back to a LOOKUP RPC to 1024 * recover rather than having nfs_open() detect the 1025 * stale file handle and failing open(2) with ESTALE. 1026 */ 1027 newvp = *vpp; 1028 newnp = VTONFS(newvp); 1029 if (!(nmp->nm_flag & NFSMNT_NOCTO) && 1030 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1031 !(newnp->n_flag & NMODIFIED)) { 1032 mtx_lock(&newnp->n_mtx); 1033 newnp->n_attrstamp = 0; 1034 mtx_unlock(&newnp->n_mtx); 1035 } 1036 if (nfscl_nodeleg(newvp, 0) == 0 || 1037 (VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 && 1038 timespeccmp(&vattr.va_ctime, &newnp->n_ctime, ==))) { 1039 NFSINCRGLOBAL(newnfsstats.lookupcache_hits); 1040 if (cnp->cn_nameiop != LOOKUP && 1041 (flags & ISLASTCN)) 1042 cnp->cn_flags |= SAVENAME; 1043 return (0); 1044 } 1045 cache_purge(newvp); 1046 if (dvp != newvp) 1047 vput(newvp); 1048 else 1049 vrele(newvp); 1050 *vpp = NULLVP; 1051 } else if (error == ENOENT) { 1052 if (dvp->v_iflag & VI_DOOMED) 1053 return (ENOENT); 1054 /* 1055 * We only accept a negative hit in the cache if the 1056 * modification time of the parent directory matches 1057 * our cached copy. Otherwise, we discard all of the 1058 * negative cache entries for this directory. We also 1059 * only trust -ve cache entries for less than 1060 * nm_negative_namecache_timeout seconds. 1061 */ 1062 if ((u_int)(ticks - np->n_dmtime_ticks) < 1063 (nmp->nm_negnametimeo * hz) && 1064 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 && 1065 timespeccmp(&vattr.va_mtime, &np->n_dmtime, ==)) { 1066 NFSINCRGLOBAL(newnfsstats.lookupcache_hits); 1067 return (ENOENT); 1068 } 1069 cache_purge_negative(dvp); 1070 mtx_lock(&np->n_mtx); 1071 timespecclear(&np->n_dmtime); 1072 mtx_unlock(&np->n_mtx); 1073 } 1074 1075 /* 1076 * Cache the modification time of the parent directory in case 1077 * the lookup fails and results in adding the first negative 1078 * name cache entry for the directory. Since this is reading 1079 * a single time_t, don't bother with locking. The 1080 * modification time may be a bit stale, but it must be read 1081 * before performing the lookup RPC to prevent a race where 1082 * another lookup updates the timestamp on the directory after 1083 * the lookup RPC has been performed on the server but before 1084 * n_dmtime is set at the end of this function. 1085 */ 1086 dmtime = np->n_vattr.na_mtime; 1087 error = 0; 1088 newvp = NULLVP; 1089 NFSINCRGLOBAL(newnfsstats.lookupcache_misses); 1090 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1091 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1092 NULL); 1093 if (dattrflag) 1094 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1095 if (error) { 1096 if (newvp != NULLVP) { 1097 vput(newvp); 1098 *vpp = NULLVP; 1099 } 1100 1101 if (error != ENOENT) { 1102 if (NFS_ISV4(dvp)) 1103 error = nfscl_maperr(td, error, (uid_t)0, 1104 (gid_t)0); 1105 return (error); 1106 } 1107 1108 /* The requested file was not found. */ 1109 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1110 (flags & ISLASTCN)) { 1111 /* 1112 * XXX: UFS does a full VOP_ACCESS(dvp, 1113 * VWRITE) here instead of just checking 1114 * MNT_RDONLY. 1115 */ 1116 if (mp->mnt_flag & MNT_RDONLY) 1117 return (EROFS); 1118 cnp->cn_flags |= SAVENAME; 1119 return (EJUSTRETURN); 1120 } 1121 1122 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE) { 1123 /* 1124 * Maintain n_dmtime as the modification time 1125 * of the parent directory when the oldest -ve 1126 * name cache entry for this directory was 1127 * added. If a -ve cache entry has already 1128 * been added with a newer modification time 1129 * by a concurrent lookup, then don't bother 1130 * adding a cache entry. The modification 1131 * time of the directory might have changed 1132 * due to the file this lookup failed to find 1133 * being created. In that case a subsequent 1134 * lookup would incorrectly use the entry 1135 * added here instead of doing an extra 1136 * lookup. 1137 */ 1138 mtx_lock(&np->n_mtx); 1139 if (timespeccmp(&np->n_dmtime, &dmtime, <=)) { 1140 if (!timespecisset(&np->n_dmtime)) { 1141 np->n_dmtime = dmtime; 1142 np->n_dmtime_ticks = ticks; 1143 } 1144 mtx_unlock(&np->n_mtx); 1145 cache_enter(dvp, NULL, cnp); 1146 } else 1147 mtx_unlock(&np->n_mtx); 1148 } 1149 return (ENOENT); 1150 } 1151 1152 /* 1153 * Handle RENAME case... 1154 */ 1155 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 1156 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1157 FREE((caddr_t)nfhp, M_NFSFH); 1158 return (EISDIR); 1159 } 1160 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1161 LK_EXCLUSIVE); 1162 if (error) 1163 return (error); 1164 newvp = NFSTOV(np); 1165 if (attrflag) 1166 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1167 0, 1); 1168 *vpp = newvp; 1169 cnp->cn_flags |= SAVENAME; 1170 return (0); 1171 } 1172 1173 if (flags & ISDOTDOT) { 1174 ltype = VOP_ISLOCKED(dvp); 1175 error = vfs_busy(mp, MBF_NOWAIT); 1176 if (error != 0) { 1177 vfs_ref(mp); 1178 VOP_UNLOCK(dvp, 0); 1179 error = vfs_busy(mp, 0); 1180 vn_lock(dvp, ltype | LK_RETRY); 1181 vfs_rel(mp); 1182 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) { 1183 vfs_unbusy(mp); 1184 error = ENOENT; 1185 } 1186 if (error != 0) 1187 return (error); 1188 } 1189 VOP_UNLOCK(dvp, 0); 1190 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1191 cnp->cn_lkflags); 1192 if (error == 0) 1193 newvp = NFSTOV(np); 1194 vfs_unbusy(mp); 1195 if (newvp != dvp) 1196 vn_lock(dvp, ltype | LK_RETRY); 1197 if (dvp->v_iflag & VI_DOOMED) { 1198 if (error == 0) { 1199 if (newvp == dvp) 1200 vrele(newvp); 1201 else 1202 vput(newvp); 1203 } 1204 error = ENOENT; 1205 } 1206 if (error != 0) 1207 return (error); 1208 if (attrflag) 1209 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1210 0, 1); 1211 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1212 FREE((caddr_t)nfhp, M_NFSFH); 1213 VREF(dvp); 1214 newvp = dvp; 1215 if (attrflag) 1216 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1217 0, 1); 1218 } else { 1219 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1220 cnp->cn_lkflags); 1221 if (error) 1222 return (error); 1223 newvp = NFSTOV(np); 1224 if (attrflag) 1225 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1226 0, 1); 1227 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1228 !(np->n_flag & NMODIFIED)) { 1229 /* 1230 * Flush the attribute cache when opening a 1231 * leaf node to ensure that fresh attributes 1232 * are fetched in nfs_open() since we did not 1233 * fetch attributes from the LOOKUP reply. 1234 */ 1235 mtx_lock(&np->n_mtx); 1236 np->n_attrstamp = 0; 1237 mtx_unlock(&np->n_mtx); 1238 } 1239 } 1240 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1241 cnp->cn_flags |= SAVENAME; 1242 if ((cnp->cn_flags & MAKEENTRY) && 1243 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) { 1244 np->n_ctime = np->n_vattr.na_vattr.va_ctime; 1245 cache_enter(dvp, newvp, cnp); 1246 } 1247 *vpp = newvp; 1248 return (0); 1249} 1250 1251/* 1252 * nfs read call. 1253 * Just call ncl_bioread() to do the work. 1254 */ 1255static int 1256nfs_read(struct vop_read_args *ap) 1257{ 1258 struct vnode *vp = ap->a_vp; 1259 1260 switch (vp->v_type) { 1261 case VREG: 1262 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred)); 1263 case VDIR: 1264 return (EISDIR); 1265 default: 1266 return (EOPNOTSUPP); 1267 } 1268} 1269 1270/* 1271 * nfs readlink call 1272 */ 1273static int 1274nfs_readlink(struct vop_readlink_args *ap) 1275{ 1276 struct vnode *vp = ap->a_vp; 1277 1278 if (vp->v_type != VLNK) 1279 return (EINVAL); 1280 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred)); 1281} 1282 1283/* 1284 * Do a readlink rpc. 1285 * Called by ncl_doio() from below the buffer cache. 1286 */ 1287int 1288ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1289{ 1290 int error, ret, attrflag; 1291 struct nfsvattr nfsva; 1292 1293 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva, 1294 &attrflag, NULL); 1295 if (attrflag) { 1296 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1297 if (ret && !error) 1298 error = ret; 1299 } 1300 if (error && NFS_ISV4(vp)) 1301 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1302 return (error); 1303} 1304 1305/* 1306 * nfs read rpc call 1307 * Ditto above 1308 */ 1309int 1310ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1311{ 1312 int error, ret, attrflag; 1313 struct nfsvattr nfsva; 1314 1315 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, &attrflag, 1316 NULL); 1317 if (attrflag) { 1318 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1319 if (ret && !error) 1320 error = ret; 1321 } 1322 if (error && NFS_ISV4(vp)) 1323 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1324 return (error); 1325} 1326 1327/* 1328 * nfs write call 1329 */ 1330int 1331ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 1332 int *iomode, int *must_commit, int called_from_strategy) 1333{ 1334 struct nfsvattr nfsva; 1335 int error = 0, attrflag, ret; 1336 u_char verf[NFSX_VERF]; 1337 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1338 1339 *must_commit = 0; 1340 error = nfsrpc_write(vp, uiop, iomode, verf, cred, 1341 uiop->uio_td, &nfsva, &attrflag, NULL, called_from_strategy); 1342 NFSLOCKMNT(nmp); 1343 if (!error && NFSHASWRITEVERF(nmp) && 1344 NFSBCMP(verf, nmp->nm_verf, NFSX_VERF)) { 1345 *must_commit = 1; 1346 NFSBCOPY(verf, nmp->nm_verf, NFSX_VERF); 1347 } 1348 NFSUNLOCKMNT(nmp); 1349 if (attrflag) { 1350 if (VTONFS(vp)->n_flag & ND_NFSV4) 1351 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1, 1352 1); 1353 else 1354 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1355 1); 1356 if (ret && !error) 1357 error = ret; 1358 } 1359 if (vp->v_mount->mnt_kern_flag & MNTK_ASYNC) 1360 *iomode = NFSWRITE_FILESYNC; 1361 if (error && NFS_ISV4(vp)) 1362 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1363 return (error); 1364} 1365 1366/* 1367 * nfs mknod rpc 1368 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1369 * mode set to specify the file type and the size field for rdev. 1370 */ 1371static int 1372nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1373 struct vattr *vap) 1374{ 1375 struct nfsvattr nfsva, dnfsva; 1376 struct vnode *newvp = NULL; 1377 struct nfsnode *np = NULL, *dnp; 1378 struct nfsfh *nfhp; 1379 struct vattr vattr; 1380 int error = 0, attrflag, dattrflag; 1381 u_int32_t rdev; 1382 1383 if (vap->va_type == VCHR || vap->va_type == VBLK) 1384 rdev = vap->va_rdev; 1385 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1386 rdev = 0xffffffff; 1387 else 1388 return (EOPNOTSUPP); 1389 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1390 return (error); 1391 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap, 1392 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva, 1393 &nfsva, &nfhp, &attrflag, &dattrflag, NULL); 1394 if (!error) { 1395 if (!nfhp) 1396 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1397 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, 1398 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1399 NULL); 1400 if (nfhp) 1401 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1402 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); 1403 } 1404 if (dattrflag) 1405 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1406 if (!error) { 1407 newvp = NFSTOV(np); 1408 if (attrflag != 0) { 1409 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1410 0, 1); 1411 if (error != 0) 1412 vput(newvp); 1413 } 1414 } 1415 if (!error) { 1416 if ((cnp->cn_flags & MAKEENTRY)) 1417 cache_enter(dvp, newvp, cnp); 1418 *vpp = newvp; 1419 } else if (NFS_ISV4(dvp)) { 1420 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, 1421 vap->va_gid); 1422 } 1423 dnp = VTONFS(dvp); 1424 mtx_lock(&dnp->n_mtx); 1425 dnp->n_flag |= NMODIFIED; 1426 if (!dattrflag) 1427 dnp->n_attrstamp = 0; 1428 mtx_unlock(&dnp->n_mtx); 1429 return (error); 1430} 1431 1432/* 1433 * nfs mknod vop 1434 * just call nfs_mknodrpc() to do the work. 1435 */ 1436/* ARGSUSED */ 1437static int 1438nfs_mknod(struct vop_mknod_args *ap) 1439{ 1440 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap)); 1441} 1442 1443static struct mtx nfs_cverf_mtx; 1444MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex", 1445 MTX_DEF); 1446 1447static nfsquad_t 1448nfs_get_cverf(void) 1449{ 1450 static nfsquad_t cverf; 1451 nfsquad_t ret; 1452 static int cverf_initialized = 0; 1453 1454 mtx_lock(&nfs_cverf_mtx); 1455 if (cverf_initialized == 0) { 1456 cverf.lval[0] = arc4random(); 1457 cverf.lval[1] = arc4random(); 1458 cverf_initialized = 1; 1459 } else 1460 cverf.qval++; 1461 ret = cverf; 1462 mtx_unlock(&nfs_cverf_mtx); 1463 1464 return (ret); 1465} 1466 1467/* 1468 * nfs file create call 1469 */ 1470static int 1471nfs_create(struct vop_create_args *ap) 1472{ 1473 struct vnode *dvp = ap->a_dvp; 1474 struct vattr *vap = ap->a_vap; 1475 struct componentname *cnp = ap->a_cnp; 1476 struct nfsnode *np = NULL, *dnp; 1477 struct vnode *newvp = NULL; 1478 struct nfsmount *nmp; 1479 struct nfsvattr dnfsva, nfsva; 1480 struct nfsfh *nfhp; 1481 nfsquad_t cverf; 1482 int error = 0, attrflag, dattrflag, fmode = 0; 1483 struct vattr vattr; 1484 1485 /* 1486 * Oops, not for me.. 1487 */ 1488 if (vap->va_type == VSOCK) 1489 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1490 1491 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1492 return (error); 1493 if (vap->va_vaflags & VA_EXCLUSIVE) 1494 fmode |= O_EXCL; 1495 dnp = VTONFS(dvp); 1496 nmp = VFSTONFS(vnode_mount(dvp)); 1497again: 1498 /* For NFSv4, wait until any remove is done. */ 1499 mtx_lock(&dnp->n_mtx); 1500 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) { 1501 dnp->n_flag |= NREMOVEWANT; 1502 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0); 1503 } 1504 mtx_unlock(&dnp->n_mtx); 1505 1506 cverf = nfs_get_cverf(); 1507 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1508 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, 1509 &nfhp, &attrflag, &dattrflag, NULL); 1510 if (!error) { 1511 if (nfhp == NULL) 1512 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1513 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, 1514 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1515 NULL); 1516 if (nfhp != NULL) 1517 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1518 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); 1519 } 1520 if (dattrflag) 1521 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1522 if (!error) { 1523 newvp = NFSTOV(np); 1524 if (attrflag) 1525 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1526 0, 1); 1527 } 1528 if (error) { 1529 if (newvp != NULL) { 1530 vput(newvp); 1531 newvp = NULL; 1532 } 1533 if (NFS_ISV34(dvp) && (fmode & O_EXCL) && 1534 error == NFSERR_NOTSUPP) { 1535 fmode &= ~O_EXCL; 1536 goto again; 1537 } 1538 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) { 1539 if (nfscl_checksattr(vap, &nfsva)) { 1540 /* 1541 * We are normally called with only a partially 1542 * initialized VAP. Since the NFSv3 spec says that 1543 * the server may use the file attributes to 1544 * store the verifier, the spec requires us to do a 1545 * SETATTR RPC. FreeBSD servers store the verifier in 1546 * atime, but we can't really assume that all servers 1547 * will so we ensure that our SETATTR sets both atime 1548 * and mtime. 1549 */ 1550 if (vap->va_mtime.tv_sec == VNOVAL) 1551 vfs_timestamp(&vap->va_mtime); 1552 if (vap->va_atime.tv_sec == VNOVAL) 1553 vap->va_atime = vap->va_mtime; 1554 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred, 1555 cnp->cn_thread, &nfsva, &attrflag, NULL); 1556 if (error && (vap->va_uid != (uid_t)VNOVAL || 1557 vap->va_gid != (gid_t)VNOVAL)) { 1558 /* try again without setting uid/gid */ 1559 vap->va_uid = (uid_t)VNOVAL; 1560 vap->va_gid = (uid_t)VNOVAL; 1561 error = nfsrpc_setattr(newvp, vap, NULL, 1562 cnp->cn_cred, cnp->cn_thread, &nfsva, 1563 &attrflag, NULL); 1564 } 1565 if (attrflag) 1566 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 1567 NULL, 0, 1); 1568 } 1569 } 1570 if (!error) { 1571 if (cnp->cn_flags & MAKEENTRY) 1572 cache_enter(dvp, newvp, cnp); 1573 *ap->a_vpp = newvp; 1574 } else if (NFS_ISV4(dvp)) { 1575 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, 1576 vap->va_gid); 1577 } 1578 mtx_lock(&dnp->n_mtx); 1579 dnp->n_flag |= NMODIFIED; 1580 if (!dattrflag) 1581 dnp->n_attrstamp = 0; 1582 mtx_unlock(&dnp->n_mtx); 1583 return (error); 1584} 1585 1586/* 1587 * nfs file remove call 1588 * To try and make nfs semantics closer to ufs semantics, a file that has 1589 * other processes using the vnode is renamed instead of removed and then 1590 * removed later on the last close. 1591 * - If v_usecount > 1 1592 * If a rename is not already in the works 1593 * call nfs_sillyrename() to set it up 1594 * else 1595 * do the remove rpc 1596 */ 1597static int 1598nfs_remove(struct vop_remove_args *ap) 1599{ 1600 struct vnode *vp = ap->a_vp; 1601 struct vnode *dvp = ap->a_dvp; 1602 struct componentname *cnp = ap->a_cnp; 1603 struct nfsnode *np = VTONFS(vp); 1604 int error = 0; 1605 struct vattr vattr; 1606 1607 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name")); 1608 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount")); 1609 if (vp->v_type == VDIR) 1610 error = EPERM; 1611 else if (vrefcnt(vp) == 1 || (np->n_sillyrename && 1612 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 1613 vattr.va_nlink > 1)) { 1614 /* 1615 * Purge the name cache so that the chance of a lookup for 1616 * the name succeeding while the remove is in progress is 1617 * minimized. Without node locking it can still happen, such 1618 * that an I/O op returns ESTALE, but since you get this if 1619 * another host removes the file.. 1620 */ 1621 cache_purge(vp); 1622 /* 1623 * throw away biocache buffers, mainly to avoid 1624 * unnecessary delayed writes later. 1625 */ 1626 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1); 1627 /* Do the rpc */ 1628 if (error != EINTR && error != EIO) 1629 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr, 1630 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread); 1631 /* 1632 * Kludge City: If the first reply to the remove rpc is lost.. 1633 * the reply to the retransmitted request will be ENOENT 1634 * since the file was in fact removed 1635 * Therefore, we cheat and return success. 1636 */ 1637 if (error == ENOENT) 1638 error = 0; 1639 } else if (!np->n_sillyrename) 1640 error = nfs_sillyrename(dvp, vp, cnp); 1641 mtx_lock(&np->n_mtx); 1642 np->n_attrstamp = 0; 1643 mtx_unlock(&np->n_mtx); 1644 return (error); 1645} 1646 1647/* 1648 * nfs file remove rpc called from nfs_inactive 1649 */ 1650int 1651ncl_removeit(struct sillyrename *sp, struct vnode *vp) 1652{ 1653 /* 1654 * Make sure that the directory vnode is still valid. 1655 * XXX we should lock sp->s_dvp here. 1656 */ 1657 if (sp->s_dvp->v_type == VBAD) 1658 return (0); 1659 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen, 1660 sp->s_cred, NULL)); 1661} 1662 1663/* 1664 * Nfs remove rpc, called from nfs_remove() and ncl_removeit(). 1665 */ 1666static int 1667nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 1668 int namelen, struct ucred *cred, struct thread *td) 1669{ 1670 struct nfsvattr dnfsva; 1671 struct nfsnode *dnp = VTONFS(dvp); 1672 int error = 0, dattrflag; 1673 1674 mtx_lock(&dnp->n_mtx); 1675 dnp->n_flag |= NREMOVEINPROG; 1676 mtx_unlock(&dnp->n_mtx); 1677 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva, 1678 &dattrflag, NULL); 1679 mtx_lock(&dnp->n_mtx); 1680 if ((dnp->n_flag & NREMOVEWANT)) { 1681 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG); 1682 mtx_unlock(&dnp->n_mtx); 1683 wakeup((caddr_t)dnp); 1684 } else { 1685 dnp->n_flag &= ~NREMOVEINPROG; 1686 mtx_unlock(&dnp->n_mtx); 1687 } 1688 if (dattrflag) 1689 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1690 mtx_lock(&dnp->n_mtx); 1691 dnp->n_flag |= NMODIFIED; 1692 if (!dattrflag) 1693 dnp->n_attrstamp = 0; 1694 mtx_unlock(&dnp->n_mtx); 1695 if (error && NFS_ISV4(dvp)) 1696 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1697 return (error); 1698} 1699 1700/* 1701 * nfs file rename call 1702 */ 1703static int 1704nfs_rename(struct vop_rename_args *ap) 1705{ 1706 struct vnode *fvp = ap->a_fvp; 1707 struct vnode *tvp = ap->a_tvp; 1708 struct vnode *fdvp = ap->a_fdvp; 1709 struct vnode *tdvp = ap->a_tdvp; 1710 struct componentname *tcnp = ap->a_tcnp; 1711 struct componentname *fcnp = ap->a_fcnp; 1712 struct nfsnode *fnp = VTONFS(ap->a_fvp); 1713 struct nfsnode *tdnp = VTONFS(ap->a_tdvp); 1714 struct nfsv4node *newv4 = NULL; 1715 int error; 1716 1717 KASSERT((tcnp->cn_flags & HASBUF) != 0 && 1718 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name")); 1719 /* Check for cross-device rename */ 1720 if ((fvp->v_mount != tdvp->v_mount) || 1721 (tvp && (fvp->v_mount != tvp->v_mount))) { 1722 error = EXDEV; 1723 goto out; 1724 } 1725 1726 if (fvp == tvp) { 1727 ncl_printf("nfs_rename: fvp == tvp (can't happen)\n"); 1728 error = 0; 1729 goto out; 1730 } 1731 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0) 1732 goto out; 1733 1734 /* 1735 * We have to flush B_DELWRI data prior to renaming 1736 * the file. If we don't, the delayed-write buffers 1737 * can be flushed out later after the file has gone stale 1738 * under NFSV3. NFSV2 does not have this problem because 1739 * ( as far as I can tell ) it flushes dirty buffers more 1740 * often. 1741 * 1742 * Skip the rename operation if the fsync fails, this can happen 1743 * due to the server's volume being full, when we pushed out data 1744 * that was written back to our cache earlier. Not checking for 1745 * this condition can result in potential (silent) data loss. 1746 */ 1747 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread); 1748 VOP_UNLOCK(fvp, 0); 1749 if (!error && tvp) 1750 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread); 1751 if (error) 1752 goto out; 1753 1754 /* 1755 * If the tvp exists and is in use, sillyrename it before doing the 1756 * rename of the new file over it. 1757 * XXX Can't sillyrename a directory. 1758 */ 1759 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename && 1760 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1761 vput(tvp); 1762 tvp = NULL; 1763 } 1764 1765 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1766 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1767 tcnp->cn_thread); 1768 1769 if (error == 0 && NFS_ISV4(tdvp)) { 1770 /* 1771 * For NFSv4, check to see if it is the same name and 1772 * replace the name, if it is different. 1773 */ 1774 MALLOC(newv4, struct nfsv4node *, 1775 sizeof (struct nfsv4node) + 1776 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1, 1777 M_NFSV4NODE, M_WAITOK); 1778 mtx_lock(&tdnp->n_mtx); 1779 mtx_lock(&fnp->n_mtx); 1780 if (fnp->n_v4 != NULL && fvp->v_type == VREG && 1781 (fnp->n_v4->n4_namelen != tcnp->cn_namelen || 1782 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4), 1783 tcnp->cn_namelen) || 1784 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen || 1785 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 1786 tdnp->n_fhp->nfh_len))) { 1787#ifdef notdef 1788{ char nnn[100]; int nnnl; 1789nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99; 1790bcopy(tcnp->cn_nameptr, nnn, nnnl); 1791nnn[nnnl] = '\0'; 1792printf("ren replace=%s\n",nnn); 1793} 1794#endif 1795 FREE((caddr_t)fnp->n_v4, M_NFSV4NODE); 1796 fnp->n_v4 = newv4; 1797 newv4 = NULL; 1798 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len; 1799 fnp->n_v4->n4_namelen = tcnp->cn_namelen; 1800 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 1801 tdnp->n_fhp->nfh_len); 1802 NFSBCOPY(tcnp->cn_nameptr, 1803 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen); 1804 } 1805 mtx_unlock(&tdnp->n_mtx); 1806 mtx_unlock(&fnp->n_mtx); 1807 if (newv4 != NULL) 1808 FREE((caddr_t)newv4, M_NFSV4NODE); 1809 } 1810 1811 if (fvp->v_type == VDIR) { 1812 if (tvp != NULL && tvp->v_type == VDIR) 1813 cache_purge(tdvp); 1814 cache_purge(fdvp); 1815 } 1816 1817out: 1818 if (tdvp == tvp) 1819 vrele(tdvp); 1820 else 1821 vput(tdvp); 1822 if (tvp) 1823 vput(tvp); 1824 vrele(fdvp); 1825 vrele(fvp); 1826 /* 1827 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1828 */ 1829 if (error == ENOENT) 1830 error = 0; 1831 return (error); 1832} 1833 1834/* 1835 * nfs file rename rpc called from nfs_remove() above 1836 */ 1837static int 1838nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp, 1839 struct sillyrename *sp) 1840{ 1841 1842 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen, 1843 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred, 1844 scnp->cn_thread)); 1845} 1846 1847/* 1848 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1849 */ 1850static int 1851nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr, 1852 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr, 1853 int tnamelen, struct ucred *cred, struct thread *td) 1854{ 1855 struct nfsvattr fnfsva, tnfsva; 1856 struct nfsnode *fdnp = VTONFS(fdvp); 1857 struct nfsnode *tdnp = VTONFS(tdvp); 1858 int error = 0, fattrflag, tattrflag; 1859 1860 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp, 1861 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag, 1862 &tattrflag, NULL, NULL); 1863 mtx_lock(&fdnp->n_mtx); 1864 fdnp->n_flag |= NMODIFIED; 1865 if (fattrflag != 0) { 1866 mtx_unlock(&fdnp->n_mtx); 1867 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1); 1868 } else { 1869 fdnp->n_attrstamp = 0; 1870 mtx_unlock(&fdnp->n_mtx); 1871 } 1872 mtx_lock(&tdnp->n_mtx); 1873 tdnp->n_flag |= NMODIFIED; 1874 if (tattrflag != 0) { 1875 mtx_unlock(&tdnp->n_mtx); 1876 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1); 1877 } else { 1878 tdnp->n_attrstamp = 0; 1879 mtx_unlock(&tdnp->n_mtx); 1880 } 1881 if (error && NFS_ISV4(fdvp)) 1882 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1883 return (error); 1884} 1885 1886/* 1887 * nfs hard link create call 1888 */ 1889static int 1890nfs_link(struct vop_link_args *ap) 1891{ 1892 struct vnode *vp = ap->a_vp; 1893 struct vnode *tdvp = ap->a_tdvp; 1894 struct componentname *cnp = ap->a_cnp; 1895 struct nfsnode *np, *tdnp; 1896 struct nfsvattr nfsva, dnfsva; 1897 int error = 0, attrflag, dattrflag; 1898 1899 if (vp->v_mount != tdvp->v_mount) { 1900 return (EXDEV); 1901 } 1902 1903 /* 1904 * Push all writes to the server, so that the attribute cache 1905 * doesn't get "out of sync" with the server. 1906 * XXX There should be a better way! 1907 */ 1908 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread); 1909 1910 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 1911 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag, 1912 &dattrflag, NULL); 1913 tdnp = VTONFS(tdvp); 1914 mtx_lock(&tdnp->n_mtx); 1915 tdnp->n_flag |= NMODIFIED; 1916 if (dattrflag != 0) { 1917 mtx_unlock(&tdnp->n_mtx); 1918 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1); 1919 } else { 1920 tdnp->n_attrstamp = 0; 1921 mtx_unlock(&tdnp->n_mtx); 1922 } 1923 if (attrflag) 1924 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1925 else { 1926 np = VTONFS(vp); 1927 mtx_lock(&np->n_mtx); 1928 np->n_attrstamp = 0; 1929 mtx_unlock(&np->n_mtx); 1930 } 1931 /* 1932 * If negative lookup caching is enabled, I might as well 1933 * add an entry for this node. Not necessary for correctness, 1934 * but if negative caching is enabled, then the system 1935 * must care about lookup caching hit rate, so... 1936 */ 1937 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 && 1938 (cnp->cn_flags & MAKEENTRY)) 1939 cache_enter(tdvp, vp, cnp); 1940 if (error && NFS_ISV4(vp)) 1941 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, 1942 (gid_t)0); 1943 return (error); 1944} 1945 1946/* 1947 * nfs symbolic link create call 1948 */ 1949static int 1950nfs_symlink(struct vop_symlink_args *ap) 1951{ 1952 struct vnode *dvp = ap->a_dvp; 1953 struct vattr *vap = ap->a_vap; 1954 struct componentname *cnp = ap->a_cnp; 1955 struct nfsvattr nfsva, dnfsva; 1956 struct nfsfh *nfhp; 1957 struct nfsnode *np = NULL, *dnp; 1958 struct vnode *newvp = NULL; 1959 int error = 0, attrflag, dattrflag, ret; 1960 1961 vap->va_type = VLNK; 1962 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1963 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, 1964 &nfsva, &nfhp, &attrflag, &dattrflag, NULL); 1965 if (nfhp) { 1966 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, 1967 &np, NULL, LK_EXCLUSIVE); 1968 if (!ret) 1969 newvp = NFSTOV(np); 1970 else if (!error) 1971 error = ret; 1972 } 1973 if (newvp != NULL) { 1974 if (attrflag) 1975 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1976 0, 1); 1977 } else if (!error) { 1978 /* 1979 * If we do not have an error and we could not extract the 1980 * newvp from the response due to the request being NFSv2, we 1981 * have to do a lookup in order to obtain a newvp to return. 1982 */ 1983 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1984 cnp->cn_cred, cnp->cn_thread, &np); 1985 if (!error) 1986 newvp = NFSTOV(np); 1987 } 1988 if (error) { 1989 if (newvp) 1990 vput(newvp); 1991 if (NFS_ISV4(dvp)) 1992 error = nfscl_maperr(cnp->cn_thread, error, 1993 vap->va_uid, vap->va_gid); 1994 } else { 1995 /* 1996 * If negative lookup caching is enabled, I might as well 1997 * add an entry for this node. Not necessary for correctness, 1998 * but if negative caching is enabled, then the system 1999 * must care about lookup caching hit rate, so... 2000 */ 2001 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2002 (cnp->cn_flags & MAKEENTRY)) 2003 cache_enter(dvp, newvp, cnp); 2004 *ap->a_vpp = newvp; 2005 } 2006 2007 dnp = VTONFS(dvp); 2008 mtx_lock(&dnp->n_mtx); 2009 dnp->n_flag |= NMODIFIED; 2010 if (dattrflag != 0) { 2011 mtx_unlock(&dnp->n_mtx); 2012 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2013 } else { 2014 dnp->n_attrstamp = 0; 2015 mtx_unlock(&dnp->n_mtx); 2016 } 2017 return (error); 2018} 2019 2020/* 2021 * nfs make dir call 2022 */ 2023static int 2024nfs_mkdir(struct vop_mkdir_args *ap) 2025{ 2026 struct vnode *dvp = ap->a_dvp; 2027 struct vattr *vap = ap->a_vap; 2028 struct componentname *cnp = ap->a_cnp; 2029 struct nfsnode *np = NULL, *dnp; 2030 struct vnode *newvp = NULL; 2031 struct vattr vattr; 2032 struct nfsfh *nfhp; 2033 struct nfsvattr nfsva, dnfsva; 2034 int error = 0, attrflag, dattrflag, ret; 2035 2036 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) 2037 return (error); 2038 vap->va_type = VDIR; 2039 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2040 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp, 2041 &attrflag, &dattrflag, NULL); 2042 dnp = VTONFS(dvp); 2043 mtx_lock(&dnp->n_mtx); 2044 dnp->n_flag |= NMODIFIED; 2045 if (dattrflag != 0) { 2046 mtx_unlock(&dnp->n_mtx); 2047 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2048 } else { 2049 dnp->n_attrstamp = 0; 2050 mtx_unlock(&dnp->n_mtx); 2051 } 2052 if (nfhp) { 2053 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, 2054 &np, NULL, LK_EXCLUSIVE); 2055 if (!ret) { 2056 newvp = NFSTOV(np); 2057 if (attrflag) 2058 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 2059 NULL, 0, 1); 2060 } else if (!error) 2061 error = ret; 2062 } 2063 if (!error && newvp == NULL) { 2064 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2065 cnp->cn_cred, cnp->cn_thread, &np); 2066 if (!error) { 2067 newvp = NFSTOV(np); 2068 if (newvp->v_type != VDIR) 2069 error = EEXIST; 2070 } 2071 } 2072 if (error) { 2073 if (newvp) 2074 vput(newvp); 2075 if (NFS_ISV4(dvp)) 2076 error = nfscl_maperr(cnp->cn_thread, error, 2077 vap->va_uid, vap->va_gid); 2078 } else { 2079 /* 2080 * If negative lookup caching is enabled, I might as well 2081 * add an entry for this node. Not necessary for correctness, 2082 * but if negative caching is enabled, then the system 2083 * must care about lookup caching hit rate, so... 2084 */ 2085 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2086 (cnp->cn_flags & MAKEENTRY)) 2087 cache_enter(dvp, newvp, cnp); 2088 *ap->a_vpp = newvp; 2089 } 2090 return (error); 2091} 2092 2093/* 2094 * nfs remove directory call 2095 */ 2096static int 2097nfs_rmdir(struct vop_rmdir_args *ap) 2098{ 2099 struct vnode *vp = ap->a_vp; 2100 struct vnode *dvp = ap->a_dvp; 2101 struct componentname *cnp = ap->a_cnp; 2102 struct nfsnode *dnp; 2103 struct nfsvattr dnfsva; 2104 int error, dattrflag; 2105 2106 if (dvp == vp) 2107 return (EINVAL); 2108 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2109 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL); 2110 dnp = VTONFS(dvp); 2111 mtx_lock(&dnp->n_mtx); 2112 dnp->n_flag |= NMODIFIED; 2113 if (dattrflag != 0) { 2114 mtx_unlock(&dnp->n_mtx); 2115 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2116 } else { 2117 dnp->n_attrstamp = 0; 2118 mtx_unlock(&dnp->n_mtx); 2119 } 2120 2121 cache_purge(dvp); 2122 cache_purge(vp); 2123 if (error && NFS_ISV4(dvp)) 2124 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, 2125 (gid_t)0); 2126 /* 2127 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2128 */ 2129 if (error == ENOENT) 2130 error = 0; 2131 return (error); 2132} 2133 2134/* 2135 * nfs readdir call 2136 */ 2137static int 2138nfs_readdir(struct vop_readdir_args *ap) 2139{ 2140 struct vnode *vp = ap->a_vp; 2141 struct nfsnode *np = VTONFS(vp); 2142 struct uio *uio = ap->a_uio; 2143 int tresid, error = 0; 2144 struct vattr vattr; 2145 2146 if (vp->v_type != VDIR) 2147 return(EPERM); 2148 2149 /* 2150 * First, check for hit on the EOF offset cache 2151 */ 2152 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2153 (np->n_flag & NMODIFIED) == 0) { 2154 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) { 2155 mtx_lock(&np->n_mtx); 2156 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) || 2157 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 2158 mtx_unlock(&np->n_mtx); 2159 NFSINCRGLOBAL(newnfsstats.direofcache_hits); 2160 return (0); 2161 } else 2162 mtx_unlock(&np->n_mtx); 2163 } 2164 } 2165 2166 /* 2167 * Call ncl_bioread() to do the real work. 2168 */ 2169 tresid = uio->uio_resid; 2170 error = ncl_bioread(vp, uio, 0, ap->a_cred); 2171 2172 if (!error && uio->uio_resid == tresid) 2173 NFSINCRGLOBAL(newnfsstats.direofcache_misses); 2174 return (error); 2175} 2176 2177/* 2178 * Readdir rpc call. 2179 * Called from below the buffer cache by ncl_doio(). 2180 */ 2181int 2182ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2183 struct thread *td) 2184{ 2185 struct nfsvattr nfsva; 2186 nfsuint64 *cookiep, cookie; 2187 struct nfsnode *dnp = VTONFS(vp); 2188 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2189 int error = 0, eof, attrflag; 2190 2191 KASSERT(uiop->uio_iovcnt == 1 && 2192 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2193 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2194 ("nfs readdirrpc bad uio")); 2195 2196 /* 2197 * If there is no cookie, assume directory was stale. 2198 */ 2199 ncl_dircookie_lock(dnp); 2200 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2201 if (cookiep) { 2202 cookie = *cookiep; 2203 ncl_dircookie_unlock(dnp); 2204 } else { 2205 ncl_dircookie_unlock(dnp); 2206 return (NFSERR_BAD_COOKIE); 2207 } 2208 2209 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2210 (void)ncl_fsinfo(nmp, vp, cred, td); 2211 2212 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva, 2213 &attrflag, &eof, NULL); 2214 if (attrflag) 2215 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2216 2217 if (!error) { 2218 /* 2219 * We are now either at the end of the directory or have filled 2220 * the block. 2221 */ 2222 if (eof) 2223 dnp->n_direofoffset = uiop->uio_offset; 2224 else { 2225 if (uiop->uio_resid > 0) 2226 ncl_printf("EEK! readdirrpc resid > 0\n"); 2227 ncl_dircookie_lock(dnp); 2228 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2229 *cookiep = cookie; 2230 ncl_dircookie_unlock(dnp); 2231 } 2232 } else if (NFS_ISV4(vp)) { 2233 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2234 } 2235 return (error); 2236} 2237 2238/* 2239 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc(). 2240 */ 2241int 2242ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2243 struct thread *td) 2244{ 2245 struct nfsvattr nfsva; 2246 nfsuint64 *cookiep, cookie; 2247 struct nfsnode *dnp = VTONFS(vp); 2248 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2249 int error = 0, attrflag, eof; 2250 2251 KASSERT(uiop->uio_iovcnt == 1 && 2252 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2253 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2254 ("nfs readdirplusrpc bad uio")); 2255 2256 /* 2257 * If there is no cookie, assume directory was stale. 2258 */ 2259 ncl_dircookie_lock(dnp); 2260 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2261 if (cookiep) { 2262 cookie = *cookiep; 2263 ncl_dircookie_unlock(dnp); 2264 } else { 2265 ncl_dircookie_unlock(dnp); 2266 return (NFSERR_BAD_COOKIE); 2267 } 2268 2269 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2270 (void)ncl_fsinfo(nmp, vp, cred, td); 2271 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva, 2272 &attrflag, &eof, NULL); 2273 if (attrflag) 2274 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2275 2276 if (!error) { 2277 /* 2278 * We are now either at end of the directory or have filled the 2279 * the block. 2280 */ 2281 if (eof) 2282 dnp->n_direofoffset = uiop->uio_offset; 2283 else { 2284 if (uiop->uio_resid > 0) 2285 ncl_printf("EEK! readdirplusrpc resid > 0\n"); 2286 ncl_dircookie_lock(dnp); 2287 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2288 *cookiep = cookie; 2289 ncl_dircookie_unlock(dnp); 2290 } 2291 } else if (NFS_ISV4(vp)) { 2292 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2293 } 2294 return (error); 2295} 2296 2297/* 2298 * Silly rename. To make the NFS filesystem that is stateless look a little 2299 * more like the "ufs" a remove of an active vnode is translated to a rename 2300 * to a funny looking filename that is removed by nfs_inactive on the 2301 * nfsnode. There is the potential for another process on a different client 2302 * to create the same funny name between the nfs_lookitup() fails and the 2303 * nfs_rename() completes, but... 2304 */ 2305static int 2306nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2307{ 2308 struct sillyrename *sp; 2309 struct nfsnode *np; 2310 int error; 2311 short pid; 2312 unsigned int lticks; 2313 2314 cache_purge(dvp); 2315 np = VTONFS(vp); 2316 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir")); 2317 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2318 M_NEWNFSREQ, M_WAITOK); 2319 sp->s_cred = crhold(cnp->cn_cred); 2320 sp->s_dvp = dvp; 2321 VREF(dvp); 2322 2323 /* 2324 * Fudge together a funny name. 2325 * Changing the format of the funny name to accomodate more 2326 * sillynames per directory. 2327 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is 2328 * CPU ticks since boot. 2329 */ 2330 pid = cnp->cn_thread->td_proc->p_pid; 2331 lticks = (unsigned int)ticks; 2332 for ( ; ; ) { 2333 sp->s_namlen = sprintf(sp->s_name, 2334 ".nfs.%08x.%04x4.4", lticks, 2335 pid); 2336 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2337 cnp->cn_thread, NULL)) 2338 break; 2339 lticks++; 2340 } 2341 error = nfs_renameit(dvp, vp, cnp, sp); 2342 if (error) 2343 goto bad; 2344 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2345 cnp->cn_thread, &np); 2346 np->n_sillyrename = sp; 2347 return (0); 2348bad: 2349 vrele(sp->s_dvp); 2350 crfree(sp->s_cred); 2351 free((caddr_t)sp, M_NEWNFSREQ); 2352 return (error); 2353} 2354 2355/* 2356 * Look up a file name and optionally either update the file handle or 2357 * allocate an nfsnode, depending on the value of npp. 2358 * npp == NULL --> just do the lookup 2359 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2360 * handled too 2361 * *npp != NULL --> update the file handle in the vnode 2362 */ 2363static int 2364nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred, 2365 struct thread *td, struct nfsnode **npp) 2366{ 2367 struct vnode *newvp = NULL, *vp; 2368 struct nfsnode *np, *dnp = VTONFS(dvp); 2369 struct nfsfh *nfhp, *onfhp; 2370 struct nfsvattr nfsva, dnfsva; 2371 struct componentname cn; 2372 int error = 0, attrflag, dattrflag; 2373 u_int hash; 2374 2375 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva, 2376 &nfhp, &attrflag, &dattrflag, NULL); 2377 if (dattrflag) 2378 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2379 if (npp && !error) { 2380 if (*npp != NULL) { 2381 np = *npp; 2382 vp = NFSTOV(np); 2383 /* 2384 * For NFSv4, check to see if it is the same name and 2385 * replace the name, if it is different. 2386 */ 2387 if (np->n_v4 != NULL && nfsva.na_type == VREG && 2388 (np->n_v4->n4_namelen != len || 2389 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) || 2390 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || 2391 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2392 dnp->n_fhp->nfh_len))) { 2393#ifdef notdef 2394{ char nnn[100]; int nnnl; 2395nnnl = (len < 100) ? len : 99; 2396bcopy(name, nnn, nnnl); 2397nnn[nnnl] = '\0'; 2398printf("replace=%s\n",nnn); 2399} 2400#endif 2401 FREE((caddr_t)np->n_v4, M_NFSV4NODE); 2402 MALLOC(np->n_v4, struct nfsv4node *, 2403 sizeof (struct nfsv4node) + 2404 dnp->n_fhp->nfh_len + len - 1, 2405 M_NFSV4NODE, M_WAITOK); 2406 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; 2407 np->n_v4->n4_namelen = len; 2408 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2409 dnp->n_fhp->nfh_len); 2410 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len); 2411 } 2412 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, 2413 FNV1_32_INIT); 2414 onfhp = np->n_fhp; 2415 /* 2416 * Rehash node for new file handle. 2417 */ 2418 vfs_hash_rehash(vp, hash); 2419 np->n_fhp = nfhp; 2420 if (onfhp != NULL) 2421 FREE((caddr_t)onfhp, M_NFSFH); 2422 newvp = NFSTOV(np); 2423 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) { 2424 FREE((caddr_t)nfhp, M_NFSFH); 2425 VREF(dvp); 2426 newvp = dvp; 2427 } else { 2428 cn.cn_nameptr = name; 2429 cn.cn_namelen = len; 2430 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td, 2431 &np, NULL, LK_EXCLUSIVE); 2432 if (error) 2433 return (error); 2434 newvp = NFSTOV(np); 2435 } 2436 if (!attrflag && *npp == NULL) { 2437 if (newvp == dvp) 2438 vrele(newvp); 2439 else 2440 vput(newvp); 2441 return (ENOENT); 2442 } 2443 if (attrflag) 2444 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 2445 0, 1); 2446 } 2447 if (npp && *npp == NULL) { 2448 if (error) { 2449 if (newvp) { 2450 if (newvp == dvp) 2451 vrele(newvp); 2452 else 2453 vput(newvp); 2454 } 2455 } else 2456 *npp = np; 2457 } 2458 if (error && NFS_ISV4(dvp)) 2459 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2460 return (error); 2461} 2462 2463/* 2464 * Nfs Version 3 and 4 commit rpc 2465 */ 2466int 2467ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred, 2468 struct thread *td) 2469{ 2470 struct nfsvattr nfsva; 2471 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2472 int error, attrflag; 2473 u_char verf[NFSX_VERF]; 2474 2475 mtx_lock(&nmp->nm_mtx); 2476 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) { 2477 mtx_unlock(&nmp->nm_mtx); 2478 return (0); 2479 } 2480 mtx_unlock(&nmp->nm_mtx); 2481 error = nfsrpc_commit(vp, offset, cnt, cred, td, verf, &nfsva, 2482 &attrflag, NULL); 2483 if (!error) { 2484 if (NFSBCMP((caddr_t)nmp->nm_verf, verf, NFSX_VERF)) { 2485 NFSBCOPY(verf, (caddr_t)nmp->nm_verf, NFSX_VERF); 2486 error = NFSERR_STALEWRITEVERF; 2487 } 2488 if (!error && attrflag) 2489 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 2490 0, 1); 2491 } else if (NFS_ISV4(vp)) { 2492 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2493 } 2494 return (error); 2495} 2496 2497/* 2498 * Strategy routine. 2499 * For async requests when nfsiod(s) are running, queue the request by 2500 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the 2501 * request. 2502 */ 2503static int 2504nfs_strategy(struct vop_strategy_args *ap) 2505{ 2506 struct buf *bp = ap->a_bp; 2507 struct ucred *cr; 2508 2509 KASSERT(!(bp->b_flags & B_DONE), 2510 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); 2511 BUF_ASSERT_HELD(bp); 2512 2513 if (bp->b_iocmd == BIO_READ) 2514 cr = bp->b_rcred; 2515 else 2516 cr = bp->b_wcred; 2517 2518 /* 2519 * If the op is asynchronous and an i/o daemon is waiting 2520 * queue the request, wake it up and wait for completion 2521 * otherwise just do it ourselves. 2522 */ 2523 if ((bp->b_flags & B_ASYNC) == 0 || 2524 ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread)) 2525 (void) ncl_doio(ap->a_vp, bp, cr, curthread, 1); 2526 return (0); 2527} 2528 2529/* 2530 * fsync vnode op. Just call ncl_flush() with commit == 1. 2531 */ 2532/* ARGSUSED */ 2533static int 2534nfs_fsync(struct vop_fsync_args *ap) 2535{ 2536 return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1, 0)); 2537} 2538 2539/* 2540 * Flush all the blocks associated with a vnode. 2541 * Walk through the buffer pool and push any dirty pages 2542 * associated with the vnode. 2543 * If the called_from_renewthread argument is TRUE, it has been called 2544 * from the NFSv4 renew thread and, as such, cannot block indefinitely 2545 * waiting for a buffer write to complete. 2546 */ 2547int 2548ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td, 2549 int commit, int called_from_renewthread) 2550{ 2551 struct nfsnode *np = VTONFS(vp); 2552 struct buf *bp; 2553 int i; 2554 struct buf *nbp; 2555 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2556 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos; 2557 int passone = 1, trycnt = 0; 2558 u_quad_t off, endoff, toff; 2559 struct ucred* wcred = NULL; 2560 struct buf **bvec = NULL; 2561 struct bufobj *bo; 2562#ifndef NFS_COMMITBVECSIZ 2563#define NFS_COMMITBVECSIZ 20 2564#endif 2565 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ]; 2566 int bvecsize = 0, bveccount; 2567 2568 if (called_from_renewthread != 0) 2569 slptimeo = hz; 2570 if (nmp->nm_flag & NFSMNT_INT) 2571 slpflag = NFS_PCATCH; 2572 if (!commit) 2573 passone = 0; 2574 bo = &vp->v_bufobj; 2575 /* 2576 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the 2577 * server, but has not been committed to stable storage on the server 2578 * yet. On the first pass, the byte range is worked out and the commit 2579 * rpc is done. On the second pass, ncl_writebp() is called to do the 2580 * job. 2581 */ 2582again: 2583 off = (u_quad_t)-1; 2584 endoff = 0; 2585 bvecpos = 0; 2586 if (NFS_ISV34(vp) && commit) { 2587 if (bvec != NULL && bvec != bvec_on_stack) 2588 free(bvec, M_TEMP); 2589 /* 2590 * Count up how many buffers waiting for a commit. 2591 */ 2592 bveccount = 0; 2593 BO_LOCK(bo); 2594 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2595 if (!BUF_ISLOCKED(bp) && 2596 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) 2597 == (B_DELWRI | B_NEEDCOMMIT)) 2598 bveccount++; 2599 } 2600 /* 2601 * Allocate space to remember the list of bufs to commit. It is 2602 * important to use M_NOWAIT here to avoid a race with nfs_write. 2603 * If we can't get memory (for whatever reason), we will end up 2604 * committing the buffers one-by-one in the loop below. 2605 */ 2606 if (bveccount > NFS_COMMITBVECSIZ) { 2607 /* 2608 * Release the vnode interlock to avoid a lock 2609 * order reversal. 2610 */ 2611 BO_UNLOCK(bo); 2612 bvec = (struct buf **) 2613 malloc(bveccount * sizeof(struct buf *), 2614 M_TEMP, M_NOWAIT); 2615 BO_LOCK(bo); 2616 if (bvec == NULL) { 2617 bvec = bvec_on_stack; 2618 bvecsize = NFS_COMMITBVECSIZ; 2619 } else 2620 bvecsize = bveccount; 2621 } else { 2622 bvec = bvec_on_stack; 2623 bvecsize = NFS_COMMITBVECSIZ; 2624 } 2625 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2626 if (bvecpos >= bvecsize) 2627 break; 2628 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2629 nbp = TAILQ_NEXT(bp, b_bobufs); 2630 continue; 2631 } 2632 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 2633 (B_DELWRI | B_NEEDCOMMIT)) { 2634 BUF_UNLOCK(bp); 2635 nbp = TAILQ_NEXT(bp, b_bobufs); 2636 continue; 2637 } 2638 BO_UNLOCK(bo); 2639 bremfree(bp); 2640 /* 2641 * Work out if all buffers are using the same cred 2642 * so we can deal with them all with one commit. 2643 * 2644 * NOTE: we are not clearing B_DONE here, so we have 2645 * to do it later on in this routine if we intend to 2646 * initiate I/O on the bp. 2647 * 2648 * Note: to avoid loopback deadlocks, we do not 2649 * assign b_runningbufspace. 2650 */ 2651 if (wcred == NULL) 2652 wcred = bp->b_wcred; 2653 else if (wcred != bp->b_wcred) 2654 wcred = NOCRED; 2655 vfs_busy_pages(bp, 1); 2656 2657 BO_LOCK(bo); 2658 /* 2659 * bp is protected by being locked, but nbp is not 2660 * and vfs_busy_pages() may sleep. We have to 2661 * recalculate nbp. 2662 */ 2663 nbp = TAILQ_NEXT(bp, b_bobufs); 2664 2665 /* 2666 * A list of these buffers is kept so that the 2667 * second loop knows which buffers have actually 2668 * been committed. This is necessary, since there 2669 * may be a race between the commit rpc and new 2670 * uncommitted writes on the file. 2671 */ 2672 bvec[bvecpos++] = bp; 2673 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2674 bp->b_dirtyoff; 2675 if (toff < off) 2676 off = toff; 2677 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); 2678 if (toff > endoff) 2679 endoff = toff; 2680 } 2681 BO_UNLOCK(bo); 2682 } 2683 if (bvecpos > 0) { 2684 /* 2685 * Commit data on the server, as required. 2686 * If all bufs are using the same wcred, then use that with 2687 * one call for all of them, otherwise commit each one 2688 * separately. 2689 */ 2690 if (wcred != NOCRED) 2691 retv = ncl_commit(vp, off, (int)(endoff - off), 2692 wcred, td); 2693 else { 2694 retv = 0; 2695 for (i = 0; i < bvecpos; i++) { 2696 off_t off, size; 2697 bp = bvec[i]; 2698 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2699 bp->b_dirtyoff; 2700 size = (u_quad_t)(bp->b_dirtyend 2701 - bp->b_dirtyoff); 2702 retv = ncl_commit(vp, off, (int)size, 2703 bp->b_wcred, td); 2704 if (retv) break; 2705 } 2706 } 2707 2708 if (retv == NFSERR_STALEWRITEVERF) 2709 ncl_clearcommit(vp->v_mount); 2710 2711 /* 2712 * Now, either mark the blocks I/O done or mark the 2713 * blocks dirty, depending on whether the commit 2714 * succeeded. 2715 */ 2716 for (i = 0; i < bvecpos; i++) { 2717 bp = bvec[i]; 2718 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 2719 if (retv) { 2720 /* 2721 * Error, leave B_DELWRI intact 2722 */ 2723 vfs_unbusy_pages(bp); 2724 brelse(bp); 2725 } else { 2726 /* 2727 * Success, remove B_DELWRI ( bundirty() ). 2728 * 2729 * b_dirtyoff/b_dirtyend seem to be NFS 2730 * specific. We should probably move that 2731 * into bundirty(). XXX 2732 */ 2733 bufobj_wref(bo); 2734 bp->b_flags |= B_ASYNC; 2735 bundirty(bp); 2736 bp->b_flags &= ~B_DONE; 2737 bp->b_ioflags &= ~BIO_ERROR; 2738 bp->b_dirtyoff = bp->b_dirtyend = 0; 2739 bufdone(bp); 2740 } 2741 } 2742 } 2743 2744 /* 2745 * Start/do any write(s) that are required. 2746 */ 2747loop: 2748 BO_LOCK(bo); 2749 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2750 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2751 if (waitfor != MNT_WAIT || passone) 2752 continue; 2753 2754 error = BUF_TIMELOCK(bp, 2755 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 2756 BO_MTX(bo), "nfsfsync", slpflag, slptimeo); 2757 if (error == 0) { 2758 BUF_UNLOCK(bp); 2759 goto loop; 2760 } 2761 if (error == ENOLCK) { 2762 error = 0; 2763 goto loop; 2764 } 2765 if (called_from_renewthread != 0) { 2766 /* 2767 * Return EIO so the flush will be retried 2768 * later. 2769 */ 2770 error = EIO; 2771 goto done; 2772 } 2773 if (newnfs_sigintr(nmp, td)) { 2774 error = EINTR; 2775 goto done; 2776 } 2777 if (slpflag & PCATCH) { 2778 slpflag = 0; 2779 slptimeo = 2 * hz; 2780 } 2781 goto loop; 2782 } 2783 if ((bp->b_flags & B_DELWRI) == 0) 2784 panic("nfs_fsync: not dirty"); 2785 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) { 2786 BUF_UNLOCK(bp); 2787 continue; 2788 } 2789 BO_UNLOCK(bo); 2790 bremfree(bp); 2791 if (passone || !commit) 2792 bp->b_flags |= B_ASYNC; 2793 else 2794 bp->b_flags |= B_ASYNC; 2795 bwrite(bp); 2796 if (newnfs_sigintr(nmp, td)) { 2797 error = EINTR; 2798 goto done; 2799 } 2800 goto loop; 2801 } 2802 if (passone) { 2803 passone = 0; 2804 BO_UNLOCK(bo); 2805 goto again; 2806 } 2807 if (waitfor == MNT_WAIT) { 2808 while (bo->bo_numoutput) { 2809 error = bufobj_wwait(bo, slpflag, slptimeo); 2810 if (error) { 2811 BO_UNLOCK(bo); 2812 if (called_from_renewthread != 0) { 2813 /* 2814 * Return EIO so that the flush will be 2815 * retried later. 2816 */ 2817 error = EIO; 2818 goto done; 2819 } 2820 error = newnfs_sigintr(nmp, td); 2821 if (error) 2822 goto done; 2823 if (slpflag & PCATCH) { 2824 slpflag = 0; 2825 slptimeo = 2 * hz; 2826 } 2827 BO_LOCK(bo); 2828 } 2829 } 2830 if (bo->bo_dirty.bv_cnt != 0 && commit) { 2831 BO_UNLOCK(bo); 2832 goto loop; 2833 } 2834 /* 2835 * Wait for all the async IO requests to drain 2836 */ 2837 BO_UNLOCK(bo); 2838 mtx_lock(&np->n_mtx); 2839 while (np->n_directio_asyncwr > 0) { 2840 np->n_flag |= NFSYNCWAIT; 2841 error = newnfs_msleep(td, &np->n_directio_asyncwr, 2842 &np->n_mtx, slpflag | (PRIBIO + 1), 2843 "nfsfsync", 0); 2844 if (error) { 2845 if (newnfs_sigintr(nmp, td)) { 2846 mtx_unlock(&np->n_mtx); 2847 error = EINTR; 2848 goto done; 2849 } 2850 } 2851 } 2852 mtx_unlock(&np->n_mtx); 2853 } else 2854 BO_UNLOCK(bo); 2855 mtx_lock(&np->n_mtx); 2856 if (np->n_flag & NWRITEERR) { 2857 error = np->n_error; 2858 np->n_flag &= ~NWRITEERR; 2859 } 2860 if (commit && bo->bo_dirty.bv_cnt == 0 && 2861 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0) 2862 np->n_flag &= ~NMODIFIED; 2863 mtx_unlock(&np->n_mtx); 2864done: 2865 if (bvec != NULL && bvec != bvec_on_stack) 2866 free(bvec, M_TEMP); 2867 if (error == 0 && commit != 0 && waitfor == MNT_WAIT && 2868 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 || 2869 np->n_directio_asyncwr != 0) && trycnt++ < 5) { 2870 /* try, try again... */ 2871 passone = 1; 2872 wcred = NULL; 2873 bvec = NULL; 2874 bvecsize = 0; 2875printf("try%d\n", trycnt); 2876 goto again; 2877 } 2878 return (error); 2879} 2880 2881/* 2882 * NFS advisory byte-level locks. 2883 */ 2884static int 2885nfs_advlock(struct vop_advlock_args *ap) 2886{ 2887 struct vnode *vp = ap->a_vp; 2888 struct ucred *cred; 2889 struct nfsnode *np = VTONFS(ap->a_vp); 2890 struct proc *p = (struct proc *)ap->a_id; 2891 struct thread *td = curthread; /* XXX */ 2892 struct vattr va; 2893 int ret, error = EOPNOTSUPP; 2894 u_quad_t size; 2895 2896 if (NFS_ISV4(vp) && (ap->a_flags & F_POSIX)) { 2897 cred = p->p_ucred; 2898 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2899 if (vp->v_iflag & VI_DOOMED) { 2900 VOP_UNLOCK(vp, 0); 2901 return (EBADF); 2902 } 2903 2904 /* 2905 * If this is unlocking a write locked region, flush and 2906 * commit them before unlocking. This is required by 2907 * RFC3530 Sec. 9.3.2. 2908 */ 2909 if (ap->a_op == F_UNLCK && 2910 nfscl_checkwritelocked(vp, ap->a_fl, cred, td)) 2911 (void) ncl_flush(vp, MNT_WAIT, cred, td, 1, 0); 2912 2913 /* 2914 * Loop around doing the lock op, while a blocking lock 2915 * must wait for the lock op to succeed. 2916 */ 2917 do { 2918 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op, 2919 ap->a_fl, 0, cred, td); 2920 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 2921 ap->a_op == F_SETLK) { 2922 VOP_UNLOCK(vp, 0); 2923 error = nfs_catnap(PZERO | PCATCH, ret, 2924 "ncladvl"); 2925 if (error) 2926 return (EINTR); 2927 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2928 if (vp->v_iflag & VI_DOOMED) { 2929 VOP_UNLOCK(vp, 0); 2930 return (EBADF); 2931 } 2932 } 2933 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 2934 ap->a_op == F_SETLK); 2935 if (ret == NFSERR_DENIED) { 2936 VOP_UNLOCK(vp, 0); 2937 return (EAGAIN); 2938 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) { 2939 VOP_UNLOCK(vp, 0); 2940 return (ret); 2941 } else if (ret != 0) { 2942 VOP_UNLOCK(vp, 0); 2943 return (EACCES); 2944 } 2945 2946 /* 2947 * Now, if we just got a lock, invalidate data in the buffer 2948 * cache, as required, so that the coherency conforms with 2949 * RFC3530 Sec. 9.3.2. 2950 */ 2951 if (ap->a_op == F_SETLK) { 2952 if ((np->n_flag & NMODIFIED) == 0) { 2953 np->n_attrstamp = 0; 2954 ret = VOP_GETATTR(vp, &va, cred); 2955 } 2956 if ((np->n_flag & NMODIFIED) || ret || 2957 np->n_change != va.va_filerev) { 2958 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1); 2959 np->n_attrstamp = 0; 2960 ret = VOP_GETATTR(vp, &va, cred); 2961 if (!ret) { 2962 np->n_mtime = va.va_mtime; 2963 np->n_change = va.va_filerev; 2964 } 2965 } 2966 } 2967 VOP_UNLOCK(vp, 0); 2968 return (0); 2969 } else if (!NFS_ISV4(vp)) { 2970 error = vn_lock(vp, LK_SHARED); 2971 if (error) 2972 return (error); 2973 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 2974 size = VTONFS(vp)->n_size; 2975 VOP_UNLOCK(vp, 0); 2976 error = lf_advlock(ap, &(vp->v_lockf), size); 2977 } else { 2978 if (nfs_advlock_p != NULL) 2979 error = nfs_advlock_p(ap); 2980 else { 2981 VOP_UNLOCK(vp, 0); 2982 error = ENOLCK; 2983 } 2984 } 2985 } 2986 return (error); 2987} 2988 2989/* 2990 * NFS advisory byte-level locks. 2991 */ 2992static int 2993nfs_advlockasync(struct vop_advlockasync_args *ap) 2994{ 2995 struct vnode *vp = ap->a_vp; 2996 u_quad_t size; 2997 int error; 2998 2999 if (NFS_ISV4(vp)) 3000 return (EOPNOTSUPP); 3001 error = vn_lock(vp, LK_SHARED); 3002 if (error) 3003 return (error); 3004 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3005 size = VTONFS(vp)->n_size; 3006 VOP_UNLOCK(vp, 0); 3007 error = lf_advlockasync(ap, &(vp->v_lockf), size); 3008 } else { 3009 VOP_UNLOCK(vp, 0); 3010 error = EOPNOTSUPP; 3011 } 3012 return (error); 3013} 3014 3015/* 3016 * Print out the contents of an nfsnode. 3017 */ 3018static int 3019nfs_print(struct vop_print_args *ap) 3020{ 3021 struct vnode *vp = ap->a_vp; 3022 struct nfsnode *np = VTONFS(vp); 3023 3024 ncl_printf("\tfileid %ld fsid 0x%x", 3025 np->n_vattr.na_fileid, np->n_vattr.na_fsid); 3026 if (vp->v_type == VFIFO) 3027 fifo_printinfo(vp); 3028 printf("\n"); 3029 return (0); 3030} 3031 3032/* 3033 * This is the "real" nfs::bwrite(struct buf*). 3034 * We set B_CACHE if this is a VMIO buffer. 3035 */ 3036int 3037ncl_writebp(struct buf *bp, int force __unused, struct thread *td) 3038{ 3039 int s; 3040 int oldflags = bp->b_flags; 3041#if 0 3042 int retv = 1; 3043 off_t off; 3044#endif 3045 3046 BUF_ASSERT_HELD(bp); 3047 3048 if (bp->b_flags & B_INVAL) { 3049 brelse(bp); 3050 return(0); 3051 } 3052 3053 bp->b_flags |= B_CACHE; 3054 3055 /* 3056 * Undirty the bp. We will redirty it later if the I/O fails. 3057 */ 3058 3059 s = splbio(); 3060 bundirty(bp); 3061 bp->b_flags &= ~B_DONE; 3062 bp->b_ioflags &= ~BIO_ERROR; 3063 bp->b_iocmd = BIO_WRITE; 3064 3065 bufobj_wref(bp->b_bufobj); 3066 curthread->td_ru.ru_oublock++; 3067 splx(s); 3068 3069 /* 3070 * Note: to avoid loopback deadlocks, we do not 3071 * assign b_runningbufspace. 3072 */ 3073 vfs_busy_pages(bp, 1); 3074 3075 BUF_KERNPROC(bp); 3076 bp->b_iooffset = dbtob(bp->b_blkno); 3077 bstrategy(bp); 3078 3079 if( (oldflags & B_ASYNC) == 0) { 3080 int rtval = bufwait(bp); 3081 3082 if (oldflags & B_DELWRI) { 3083 s = splbio(); 3084 reassignbuf(bp); 3085 splx(s); 3086 } 3087 brelse(bp); 3088 return (rtval); 3089 } 3090 3091 return (0); 3092} 3093 3094/* 3095 * nfs special file access vnode op. 3096 * Essentially just get vattr and then imitate iaccess() since the device is 3097 * local to the client. 3098 */ 3099static int 3100nfsspec_access(struct vop_access_args *ap) 3101{ 3102 struct vattr *vap; 3103 struct ucred *cred = ap->a_cred; 3104 struct vnode *vp = ap->a_vp; 3105 accmode_t accmode = ap->a_accmode; 3106 struct vattr vattr; 3107 int error; 3108 3109 /* 3110 * Disallow write attempts on filesystems mounted read-only; 3111 * unless the file is a socket, fifo, or a block or character 3112 * device resident on the filesystem. 3113 */ 3114 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3115 switch (vp->v_type) { 3116 case VREG: 3117 case VDIR: 3118 case VLNK: 3119 return (EROFS); 3120 default: 3121 break; 3122 } 3123 } 3124 vap = &vattr; 3125 error = VOP_GETATTR(vp, vap, cred); 3126 if (error) 3127 goto out; 3128 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid, 3129 accmode, cred, NULL); 3130out: 3131 return error; 3132} 3133 3134/* 3135 * Read wrapper for fifos. 3136 */ 3137static int 3138nfsfifo_read(struct vop_read_args *ap) 3139{ 3140 struct nfsnode *np = VTONFS(ap->a_vp); 3141 int error; 3142 3143 /* 3144 * Set access flag. 3145 */ 3146 mtx_lock(&np->n_mtx); 3147 np->n_flag |= NACC; 3148 getnanotime(&np->n_atim); 3149 mtx_unlock(&np->n_mtx); 3150 error = fifo_specops.vop_read(ap); 3151 return error; 3152} 3153 3154/* 3155 * Write wrapper for fifos. 3156 */ 3157static int 3158nfsfifo_write(struct vop_write_args *ap) 3159{ 3160 struct nfsnode *np = VTONFS(ap->a_vp); 3161 3162 /* 3163 * Set update flag. 3164 */ 3165 mtx_lock(&np->n_mtx); 3166 np->n_flag |= NUPD; 3167 getnanotime(&np->n_mtim); 3168 mtx_unlock(&np->n_mtx); 3169 return(fifo_specops.vop_write(ap)); 3170} 3171 3172/* 3173 * Close wrapper for fifos. 3174 * 3175 * Update the times on the nfsnode then do fifo close. 3176 */ 3177static int 3178nfsfifo_close(struct vop_close_args *ap) 3179{ 3180 struct vnode *vp = ap->a_vp; 3181 struct nfsnode *np = VTONFS(vp); 3182 struct vattr vattr; 3183 struct timespec ts; 3184 3185 mtx_lock(&np->n_mtx); 3186 if (np->n_flag & (NACC | NUPD)) { 3187 getnanotime(&ts); 3188 if (np->n_flag & NACC) 3189 np->n_atim = ts; 3190 if (np->n_flag & NUPD) 3191 np->n_mtim = ts; 3192 np->n_flag |= NCHG; 3193 if (vrefcnt(vp) == 1 && 3194 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3195 VATTR_NULL(&vattr); 3196 if (np->n_flag & NACC) 3197 vattr.va_atime = np->n_atim; 3198 if (np->n_flag & NUPD) 3199 vattr.va_mtime = np->n_mtim; 3200 mtx_unlock(&np->n_mtx); 3201 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3202 goto out; 3203 } 3204 } 3205 mtx_unlock(&np->n_mtx); 3206out: 3207 return (fifo_specops.vop_close(ap)); 3208} 3209 3210/* 3211 * Just call ncl_writebp() with the force argument set to 1. 3212 * 3213 * NOTE: B_DONE may or may not be set in a_bp on call. 3214 */ 3215static int 3216nfs_bwrite(struct buf *bp) 3217{ 3218 3219 return (ncl_writebp(bp, 1, curthread)); 3220} 3221 3222struct buf_ops buf_ops_newnfs = { 3223 .bop_name = "buf_ops_nfs", 3224 .bop_write = nfs_bwrite, 3225 .bop_strategy = bufstrategy, 3226 .bop_sync = bufsync, 3227 .bop_bdflush = bufbdflush, 3228}; 3229 3230/* 3231 * Cloned from vop_stdlock(), and then the ugly hack added. 3232 */ 3233static int 3234nfs_lock1(struct vop_lock1_args *ap) 3235{ 3236 struct vnode *vp = ap->a_vp; 3237 int error = 0; 3238 3239 /* 3240 * Since vfs_hash_get() calls vget() and it will no longer work 3241 * for FreeBSD8 with flags == 0, I can only think of this horrible 3242 * hack to work around it. I call vfs_hash_get() with LK_EXCLOTHER 3243 * and then handle it here. All I want for this case is a v_usecount 3244 * on the vnode to use for recovery, while another thread might 3245 * hold a lock on the vnode. I have the other threads blocked, so 3246 * there isn't any race problem. 3247 */ 3248 if ((ap->a_flags & LK_TYPE_MASK) == LK_EXCLOTHER) { 3249 if ((ap->a_flags & LK_INTERLOCK) == 0) 3250 panic("ncllock1"); 3251 if ((vp->v_iflag & VI_DOOMED)) 3252 error = ENOENT; 3253 VI_UNLOCK(vp); 3254 return (error); 3255 } 3256 return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp), 3257 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file, 3258 ap->a_line)); 3259} 3260 3261static int 3262nfs_getacl(struct vop_getacl_args *ap) 3263{ 3264 int error; 3265 3266 if (ap->a_type != ACL_TYPE_NFS4) 3267 return (EOPNOTSUPP); 3268 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3269 NULL); 3270 if (error > NFSERR_STALE) { 3271 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3272 error = EPERM; 3273 } 3274 return (error); 3275} 3276 3277static int 3278nfs_setacl(struct vop_setacl_args *ap) 3279{ 3280 int error; 3281 3282 if (ap->a_type != ACL_TYPE_NFS4) 3283 return (EOPNOTSUPP); 3284 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3285 NULL); 3286 if (error > NFSERR_STALE) { 3287 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3288 error = EPERM; 3289 } 3290 return (error); 3291} 3292 3293/* 3294 * Return POSIX pathconf information applicable to nfs filesystems. 3295 */ 3296static int 3297nfs_pathconf(struct vop_pathconf_args *ap) 3298{ 3299 struct nfsv3_pathconf pc; 3300 struct nfsvattr nfsva; 3301 struct vnode *vp = ap->a_vp; 3302 struct thread *td = curthread; 3303 int attrflag, error; 3304 3305 if (NFS_ISV34(vp)) { 3306 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva, 3307 &attrflag, NULL); 3308 if (attrflag != 0) 3309 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 3310 1); 3311 if (error != 0) 3312 return (error); 3313 } else { 3314 /* For NFSv2, just fake them. */ 3315 pc.pc_linkmax = LINK_MAX; 3316 pc.pc_namemax = NFS_MAXNAMLEN; 3317 pc.pc_notrunc = 1; 3318 pc.pc_chownrestricted = 1; 3319 pc.pc_caseinsensitive = 0; 3320 pc.pc_casepreserving = 1; 3321 error = 0; 3322 } 3323 switch (ap->a_name) { 3324 case _PC_LINK_MAX: 3325 *ap->a_retval = pc.pc_linkmax; 3326 break; 3327 case _PC_NAME_MAX: 3328 *ap->a_retval = pc.pc_namemax; 3329 break; 3330 case _PC_PATH_MAX: 3331 *ap->a_retval = PATH_MAX; 3332 break; 3333 case _PC_PIPE_BUF: 3334 *ap->a_retval = PIPE_BUF; 3335 break; 3336 case _PC_CHOWN_RESTRICTED: 3337 *ap->a_retval = pc.pc_chownrestricted; 3338 break; 3339 case _PC_NO_TRUNC: 3340 *ap->a_retval = pc.pc_notrunc; 3341 break; 3342 case _PC_ACL_EXTENDED: 3343 *ap->a_retval = 0; 3344 break; 3345 case _PC_ACL_NFS4: 3346 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 && 3347 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL)) 3348 *ap->a_retval = 1; 3349 else 3350 *ap->a_retval = 0; 3351 break; 3352 case _PC_ACL_PATH_MAX: 3353 if (NFS_ISV4(vp)) 3354 *ap->a_retval = ACL_MAX_ENTRIES; 3355 else 3356 *ap->a_retval = 3; 3357 break; 3358 case _PC_MAC_PRESENT: 3359 *ap->a_retval = 0; 3360 break; 3361 case _PC_ASYNC_IO: 3362 /* _PC_ASYNC_IO should have been handled by upper layers. */ 3363 KASSERT(0, ("_PC_ASYNC_IO should not get here")); 3364 error = EINVAL; 3365 break; 3366 case _PC_PRIO_IO: 3367 *ap->a_retval = 0; 3368 break; 3369 case _PC_SYNC_IO: 3370 *ap->a_retval = 0; 3371 break; 3372 case _PC_ALLOC_SIZE_MIN: 3373 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize; 3374 break; 3375 case _PC_FILESIZEBITS: 3376 if (NFS_ISV34(vp)) 3377 *ap->a_retval = 64; 3378 else 3379 *ap->a_retval = 32; 3380 break; 3381 case _PC_REC_INCR_XFER_SIZE: 3382 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 3383 break; 3384 case _PC_REC_MAX_XFER_SIZE: 3385 *ap->a_retval = -1; /* means ``unlimited'' */ 3386 break; 3387 case _PC_REC_MIN_XFER_SIZE: 3388 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 3389 break; 3390 case _PC_REC_XFER_ALIGN: 3391 *ap->a_retval = PAGE_SIZE; 3392 break; 3393 case _PC_SYMLINK_MAX: 3394 *ap->a_retval = NFS_MAXPATHLEN; 3395 break; 3396 3397 default: 3398 error = EINVAL; 3399 break; 3400 } 3401 return (error); 3402} 3403 3404