nfs_clvnops.c revision 220763
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 220763 2011-04-18 00:41:23Z 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_newnfs); 203 204static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO; 205SYSCTL_INT(_vfs_newnfs, 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_newnfs, 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_newnfs, 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_newnfs, 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_newnfs, 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_newnfs, 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_newnfs, OID_AUTO, access_cache_hits, CTLFLAG_RD, 242 &newnfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count"); 243 244SYSCTL_INT(_vfs_newnfs, 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 ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1030 !(newnp->n_flag & NMODIFIED)) { 1031 mtx_lock(&newnp->n_mtx); 1032 newnp->n_attrstamp = 0; 1033 mtx_unlock(&newnp->n_mtx); 1034 } 1035 if (nfscl_nodeleg(newvp, 0) == 0 || 1036 (VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 && 1037 timespeccmp(&vattr.va_ctime, &newnp->n_ctime, ==))) { 1038 NFSINCRGLOBAL(newnfsstats.lookupcache_hits); 1039 if (cnp->cn_nameiop != LOOKUP && 1040 (flags & ISLASTCN)) 1041 cnp->cn_flags |= SAVENAME; 1042 return (0); 1043 } 1044 cache_purge(newvp); 1045 if (dvp != newvp) 1046 vput(newvp); 1047 else 1048 vrele(newvp); 1049 *vpp = NULLVP; 1050 } else if (error == ENOENT) { 1051 if (dvp->v_iflag & VI_DOOMED) 1052 return (ENOENT); 1053 /* 1054 * We only accept a negative hit in the cache if the 1055 * modification time of the parent directory matches 1056 * our cached copy. Otherwise, we discard all of the 1057 * negative cache entries for this directory. We also 1058 * only trust -ve cache entries for less than 1059 * nm_negative_namecache_timeout seconds. 1060 */ 1061 if ((u_int)(ticks - np->n_dmtime_ticks) < 1062 (nmp->nm_negnametimeo * hz) && 1063 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 && 1064 timespeccmp(&vattr.va_mtime, &np->n_dmtime, ==)) { 1065 NFSINCRGLOBAL(newnfsstats.lookupcache_hits); 1066 return (ENOENT); 1067 } 1068 cache_purge_negative(dvp); 1069 mtx_lock(&np->n_mtx); 1070 timespecclear(&np->n_dmtime); 1071 mtx_unlock(&np->n_mtx); 1072 } 1073 1074 /* 1075 * Cache the modification time of the parent directory in case 1076 * the lookup fails and results in adding the first negative 1077 * name cache entry for the directory. Since this is reading 1078 * a single time_t, don't bother with locking. The 1079 * modification time may be a bit stale, but it must be read 1080 * before performing the lookup RPC to prevent a race where 1081 * another lookup updates the timestamp on the directory after 1082 * the lookup RPC has been performed on the server but before 1083 * n_dmtime is set at the end of this function. 1084 */ 1085 dmtime = np->n_vattr.na_mtime; 1086 error = 0; 1087 newvp = NULLVP; 1088 NFSINCRGLOBAL(newnfsstats.lookupcache_misses); 1089 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1090 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1091 NULL); 1092 if (dattrflag) 1093 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1094 if (error) { 1095 if (newvp != NULLVP) { 1096 vput(newvp); 1097 *vpp = NULLVP; 1098 } 1099 1100 if (error != ENOENT) { 1101 if (NFS_ISV4(dvp)) 1102 error = nfscl_maperr(td, error, (uid_t)0, 1103 (gid_t)0); 1104 return (error); 1105 } 1106 1107 /* The requested file was not found. */ 1108 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1109 (flags & ISLASTCN)) { 1110 /* 1111 * XXX: UFS does a full VOP_ACCESS(dvp, 1112 * VWRITE) here instead of just checking 1113 * MNT_RDONLY. 1114 */ 1115 if (mp->mnt_flag & MNT_RDONLY) 1116 return (EROFS); 1117 cnp->cn_flags |= SAVENAME; 1118 return (EJUSTRETURN); 1119 } 1120 1121 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE) { 1122 /* 1123 * Maintain n_dmtime as the modification time 1124 * of the parent directory when the oldest -ve 1125 * name cache entry for this directory was 1126 * added. If a -ve cache entry has already 1127 * been added with a newer modification time 1128 * by a concurrent lookup, then don't bother 1129 * adding a cache entry. The modification 1130 * time of the directory might have changed 1131 * due to the file this lookup failed to find 1132 * being created. In that case a subsequent 1133 * lookup would incorrectly use the entry 1134 * added here instead of doing an extra 1135 * lookup. 1136 */ 1137 mtx_lock(&np->n_mtx); 1138 if (timespeccmp(&np->n_dmtime, &dmtime, <=)) { 1139 if (!timespecisset(&np->n_dmtime)) { 1140 np->n_dmtime = dmtime; 1141 np->n_dmtime_ticks = ticks; 1142 } 1143 mtx_unlock(&np->n_mtx); 1144 cache_enter(dvp, NULL, cnp); 1145 } else 1146 mtx_unlock(&np->n_mtx); 1147 } 1148 return (ENOENT); 1149 } 1150 1151 /* 1152 * Handle RENAME case... 1153 */ 1154 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 1155 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1156 FREE((caddr_t)nfhp, M_NFSFH); 1157 return (EISDIR); 1158 } 1159 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1160 LK_EXCLUSIVE); 1161 if (error) 1162 return (error); 1163 newvp = NFSTOV(np); 1164 if (attrflag) 1165 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1166 0, 1); 1167 *vpp = newvp; 1168 cnp->cn_flags |= SAVENAME; 1169 return (0); 1170 } 1171 1172 if (flags & ISDOTDOT) { 1173 ltype = VOP_ISLOCKED(dvp); 1174 error = vfs_busy(mp, MBF_NOWAIT); 1175 if (error != 0) { 1176 vfs_ref(mp); 1177 VOP_UNLOCK(dvp, 0); 1178 error = vfs_busy(mp, 0); 1179 vn_lock(dvp, ltype | LK_RETRY); 1180 vfs_rel(mp); 1181 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) { 1182 vfs_unbusy(mp); 1183 error = ENOENT; 1184 } 1185 if (error != 0) 1186 return (error); 1187 } 1188 VOP_UNLOCK(dvp, 0); 1189 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1190 cnp->cn_lkflags); 1191 if (error == 0) 1192 newvp = NFSTOV(np); 1193 vfs_unbusy(mp); 1194 if (newvp != dvp) 1195 vn_lock(dvp, ltype | LK_RETRY); 1196 if (dvp->v_iflag & VI_DOOMED) { 1197 if (error == 0) { 1198 if (newvp == dvp) 1199 vrele(newvp); 1200 else 1201 vput(newvp); 1202 } 1203 error = ENOENT; 1204 } 1205 if (error != 0) 1206 return (error); 1207 if (attrflag) 1208 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1209 0, 1); 1210 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1211 FREE((caddr_t)nfhp, M_NFSFH); 1212 VREF(dvp); 1213 newvp = dvp; 1214 if (attrflag) 1215 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1216 0, 1); 1217 } else { 1218 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1219 cnp->cn_lkflags); 1220 if (error) 1221 return (error); 1222 newvp = NFSTOV(np); 1223 if (attrflag) 1224 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1225 0, 1); 1226 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1227 !(np->n_flag & NMODIFIED)) { 1228 /* 1229 * Flush the attribute cache when opening a 1230 * leaf node to ensure that fresh attributes 1231 * are fetched in nfs_open() since we did not 1232 * fetch attributes from the LOOKUP reply. 1233 */ 1234 mtx_lock(&np->n_mtx); 1235 np->n_attrstamp = 0; 1236 mtx_unlock(&np->n_mtx); 1237 } 1238 } 1239 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1240 cnp->cn_flags |= SAVENAME; 1241 if ((cnp->cn_flags & MAKEENTRY) && 1242 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) { 1243 np->n_ctime = np->n_vattr.na_vattr.va_ctime; 1244 cache_enter(dvp, newvp, cnp); 1245 } 1246 *vpp = newvp; 1247 return (0); 1248} 1249 1250/* 1251 * nfs read call. 1252 * Just call ncl_bioread() to do the work. 1253 */ 1254static int 1255nfs_read(struct vop_read_args *ap) 1256{ 1257 struct vnode *vp = ap->a_vp; 1258 1259 switch (vp->v_type) { 1260 case VREG: 1261 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred)); 1262 case VDIR: 1263 return (EISDIR); 1264 default: 1265 return (EOPNOTSUPP); 1266 } 1267} 1268 1269/* 1270 * nfs readlink call 1271 */ 1272static int 1273nfs_readlink(struct vop_readlink_args *ap) 1274{ 1275 struct vnode *vp = ap->a_vp; 1276 1277 if (vp->v_type != VLNK) 1278 return (EINVAL); 1279 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred)); 1280} 1281 1282/* 1283 * Do a readlink rpc. 1284 * Called by ncl_doio() from below the buffer cache. 1285 */ 1286int 1287ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1288{ 1289 int error, ret, attrflag; 1290 struct nfsvattr nfsva; 1291 1292 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva, 1293 &attrflag, NULL); 1294 if (attrflag) { 1295 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1296 if (ret && !error) 1297 error = ret; 1298 } 1299 if (error && NFS_ISV4(vp)) 1300 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1301 return (error); 1302} 1303 1304/* 1305 * nfs read rpc call 1306 * Ditto above 1307 */ 1308int 1309ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1310{ 1311 int error, ret, attrflag; 1312 struct nfsvattr nfsva; 1313 1314 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, &attrflag, 1315 NULL); 1316 if (attrflag) { 1317 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1318 if (ret && !error) 1319 error = ret; 1320 } 1321 if (error && NFS_ISV4(vp)) 1322 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1323 return (error); 1324} 1325 1326/* 1327 * nfs write call 1328 */ 1329int 1330ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 1331 int *iomode, int *must_commit, int called_from_strategy) 1332{ 1333 struct nfsvattr nfsva; 1334 int error = 0, attrflag, ret; 1335 u_char verf[NFSX_VERF]; 1336 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1337 1338 *must_commit = 0; 1339 error = nfsrpc_write(vp, uiop, iomode, verf, cred, 1340 uiop->uio_td, &nfsva, &attrflag, NULL, called_from_strategy); 1341 NFSLOCKMNT(nmp); 1342 if (!error && NFSHASWRITEVERF(nmp) && 1343 NFSBCMP(verf, nmp->nm_verf, NFSX_VERF)) { 1344 *must_commit = 1; 1345 NFSBCOPY(verf, nmp->nm_verf, NFSX_VERF); 1346 } 1347 NFSUNLOCKMNT(nmp); 1348 if (attrflag) { 1349 if (VTONFS(vp)->n_flag & ND_NFSV4) 1350 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1, 1351 1); 1352 else 1353 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1354 1); 1355 if (ret && !error) 1356 error = ret; 1357 } 1358 if (vp->v_mount->mnt_kern_flag & MNTK_ASYNC) 1359 *iomode = NFSWRITE_FILESYNC; 1360 if (error && NFS_ISV4(vp)) 1361 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1362 return (error); 1363} 1364 1365/* 1366 * nfs mknod rpc 1367 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1368 * mode set to specify the file type and the size field for rdev. 1369 */ 1370static int 1371nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1372 struct vattr *vap) 1373{ 1374 struct nfsvattr nfsva, dnfsva; 1375 struct vnode *newvp = NULL; 1376 struct nfsnode *np = NULL, *dnp; 1377 struct nfsfh *nfhp; 1378 struct vattr vattr; 1379 int error = 0, attrflag, dattrflag; 1380 u_int32_t rdev; 1381 1382 if (vap->va_type == VCHR || vap->va_type == VBLK) 1383 rdev = vap->va_rdev; 1384 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1385 rdev = 0xffffffff; 1386 else 1387 return (EOPNOTSUPP); 1388 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1389 return (error); 1390 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap, 1391 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva, 1392 &nfsva, &nfhp, &attrflag, &dattrflag, NULL); 1393 if (!error) { 1394 if (!nfhp) 1395 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1396 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, 1397 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1398 NULL); 1399 if (nfhp) 1400 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1401 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); 1402 } 1403 if (dattrflag) 1404 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1405 if (!error) { 1406 newvp = NFSTOV(np); 1407 if (attrflag != 0) { 1408 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1409 0, 1); 1410 if (error != 0) 1411 vput(newvp); 1412 } 1413 } 1414 if (!error) { 1415 if ((cnp->cn_flags & MAKEENTRY)) 1416 cache_enter(dvp, newvp, cnp); 1417 *vpp = newvp; 1418 } else if (NFS_ISV4(dvp)) { 1419 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, 1420 vap->va_gid); 1421 } 1422 dnp = VTONFS(dvp); 1423 mtx_lock(&dnp->n_mtx); 1424 dnp->n_flag |= NMODIFIED; 1425 if (!dattrflag) 1426 dnp->n_attrstamp = 0; 1427 mtx_unlock(&dnp->n_mtx); 1428 return (error); 1429} 1430 1431/* 1432 * nfs mknod vop 1433 * just call nfs_mknodrpc() to do the work. 1434 */ 1435/* ARGSUSED */ 1436static int 1437nfs_mknod(struct vop_mknod_args *ap) 1438{ 1439 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap)); 1440} 1441 1442static struct mtx nfs_cverf_mtx; 1443MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex", 1444 MTX_DEF); 1445 1446static nfsquad_t 1447nfs_get_cverf(void) 1448{ 1449 static nfsquad_t cverf; 1450 nfsquad_t ret; 1451 static int cverf_initialized = 0; 1452 1453 mtx_lock(&nfs_cverf_mtx); 1454 if (cverf_initialized == 0) { 1455 cverf.lval[0] = arc4random(); 1456 cverf.lval[1] = arc4random(); 1457 cverf_initialized = 1; 1458 } else 1459 cverf.qval++; 1460 ret = cverf; 1461 mtx_unlock(&nfs_cverf_mtx); 1462 1463 return (ret); 1464} 1465 1466/* 1467 * nfs file create call 1468 */ 1469static int 1470nfs_create(struct vop_create_args *ap) 1471{ 1472 struct vnode *dvp = ap->a_dvp; 1473 struct vattr *vap = ap->a_vap; 1474 struct componentname *cnp = ap->a_cnp; 1475 struct nfsnode *np = NULL, *dnp; 1476 struct vnode *newvp = NULL; 1477 struct nfsmount *nmp; 1478 struct nfsvattr dnfsva, nfsva; 1479 struct nfsfh *nfhp; 1480 nfsquad_t cverf; 1481 int error = 0, attrflag, dattrflag, fmode = 0; 1482 struct vattr vattr; 1483 1484 /* 1485 * Oops, not for me.. 1486 */ 1487 if (vap->va_type == VSOCK) 1488 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1489 1490 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1491 return (error); 1492 if (vap->va_vaflags & VA_EXCLUSIVE) 1493 fmode |= O_EXCL; 1494 dnp = VTONFS(dvp); 1495 nmp = VFSTONFS(vnode_mount(dvp)); 1496again: 1497 /* For NFSv4, wait until any remove is done. */ 1498 mtx_lock(&dnp->n_mtx); 1499 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) { 1500 dnp->n_flag |= NREMOVEWANT; 1501 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0); 1502 } 1503 mtx_unlock(&dnp->n_mtx); 1504 1505 cverf = nfs_get_cverf(); 1506 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1507 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, 1508 &nfhp, &attrflag, &dattrflag, NULL); 1509 if (!error) { 1510 if (nfhp == NULL) 1511 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1512 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, 1513 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1514 NULL); 1515 if (nfhp != NULL) 1516 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1517 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); 1518 } 1519 if (dattrflag) 1520 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1521 if (!error) { 1522 newvp = NFSTOV(np); 1523 if (attrflag) 1524 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1525 0, 1); 1526 } 1527 if (error) { 1528 if (newvp != NULL) { 1529 vput(newvp); 1530 newvp = NULL; 1531 } 1532 if (NFS_ISV34(dvp) && (fmode & O_EXCL) && 1533 error == NFSERR_NOTSUPP) { 1534 fmode &= ~O_EXCL; 1535 goto again; 1536 } 1537 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) { 1538 if (nfscl_checksattr(vap, &nfsva)) { 1539 /* 1540 * We are normally called with only a partially 1541 * initialized VAP. Since the NFSv3 spec says that 1542 * the server may use the file attributes to 1543 * store the verifier, the spec requires us to do a 1544 * SETATTR RPC. FreeBSD servers store the verifier in 1545 * atime, but we can't really assume that all servers 1546 * will so we ensure that our SETATTR sets both atime 1547 * and mtime. 1548 */ 1549 if (vap->va_mtime.tv_sec == VNOVAL) 1550 vfs_timestamp(&vap->va_mtime); 1551 if (vap->va_atime.tv_sec == VNOVAL) 1552 vap->va_atime = vap->va_mtime; 1553 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred, 1554 cnp->cn_thread, &nfsva, &attrflag, NULL); 1555 if (error && (vap->va_uid != (uid_t)VNOVAL || 1556 vap->va_gid != (gid_t)VNOVAL)) { 1557 /* try again without setting uid/gid */ 1558 vap->va_uid = (uid_t)VNOVAL; 1559 vap->va_gid = (uid_t)VNOVAL; 1560 error = nfsrpc_setattr(newvp, vap, NULL, 1561 cnp->cn_cred, cnp->cn_thread, &nfsva, 1562 &attrflag, NULL); 1563 } 1564 if (attrflag) 1565 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 1566 NULL, 0, 1); 1567 } 1568 } 1569 if (!error) { 1570 if (cnp->cn_flags & MAKEENTRY) 1571 cache_enter(dvp, newvp, cnp); 1572 *ap->a_vpp = newvp; 1573 } else if (NFS_ISV4(dvp)) { 1574 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, 1575 vap->va_gid); 1576 } 1577 mtx_lock(&dnp->n_mtx); 1578 dnp->n_flag |= NMODIFIED; 1579 if (!dattrflag) 1580 dnp->n_attrstamp = 0; 1581 mtx_unlock(&dnp->n_mtx); 1582 return (error); 1583} 1584 1585/* 1586 * nfs file remove call 1587 * To try and make nfs semantics closer to ufs semantics, a file that has 1588 * other processes using the vnode is renamed instead of removed and then 1589 * removed later on the last close. 1590 * - If v_usecount > 1 1591 * If a rename is not already in the works 1592 * call nfs_sillyrename() to set it up 1593 * else 1594 * do the remove rpc 1595 */ 1596static int 1597nfs_remove(struct vop_remove_args *ap) 1598{ 1599 struct vnode *vp = ap->a_vp; 1600 struct vnode *dvp = ap->a_dvp; 1601 struct componentname *cnp = ap->a_cnp; 1602 struct nfsnode *np = VTONFS(vp); 1603 int error = 0; 1604 struct vattr vattr; 1605 1606 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name")); 1607 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount")); 1608 if (vp->v_type == VDIR) 1609 error = EPERM; 1610 else if (vrefcnt(vp) == 1 || (np->n_sillyrename && 1611 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 1612 vattr.va_nlink > 1)) { 1613 /* 1614 * Purge the name cache so that the chance of a lookup for 1615 * the name succeeding while the remove is in progress is 1616 * minimized. Without node locking it can still happen, such 1617 * that an I/O op returns ESTALE, but since you get this if 1618 * another host removes the file.. 1619 */ 1620 cache_purge(vp); 1621 /* 1622 * throw away biocache buffers, mainly to avoid 1623 * unnecessary delayed writes later. 1624 */ 1625 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1); 1626 /* Do the rpc */ 1627 if (error != EINTR && error != EIO) 1628 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr, 1629 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread); 1630 /* 1631 * Kludge City: If the first reply to the remove rpc is lost.. 1632 * the reply to the retransmitted request will be ENOENT 1633 * since the file was in fact removed 1634 * Therefore, we cheat and return success. 1635 */ 1636 if (error == ENOENT) 1637 error = 0; 1638 } else if (!np->n_sillyrename) 1639 error = nfs_sillyrename(dvp, vp, cnp); 1640 mtx_lock(&np->n_mtx); 1641 np->n_attrstamp = 0; 1642 mtx_unlock(&np->n_mtx); 1643 return (error); 1644} 1645 1646/* 1647 * nfs file remove rpc called from nfs_inactive 1648 */ 1649int 1650ncl_removeit(struct sillyrename *sp, struct vnode *vp) 1651{ 1652 /* 1653 * Make sure that the directory vnode is still valid. 1654 * XXX we should lock sp->s_dvp here. 1655 */ 1656 if (sp->s_dvp->v_type == VBAD) 1657 return (0); 1658 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen, 1659 sp->s_cred, NULL)); 1660} 1661 1662/* 1663 * Nfs remove rpc, called from nfs_remove() and ncl_removeit(). 1664 */ 1665static int 1666nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 1667 int namelen, struct ucred *cred, struct thread *td) 1668{ 1669 struct nfsvattr dnfsva; 1670 struct nfsnode *dnp = VTONFS(dvp); 1671 int error = 0, dattrflag; 1672 1673 mtx_lock(&dnp->n_mtx); 1674 dnp->n_flag |= NREMOVEINPROG; 1675 mtx_unlock(&dnp->n_mtx); 1676 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva, 1677 &dattrflag, NULL); 1678 mtx_lock(&dnp->n_mtx); 1679 if ((dnp->n_flag & NREMOVEWANT)) { 1680 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG); 1681 mtx_unlock(&dnp->n_mtx); 1682 wakeup((caddr_t)dnp); 1683 } else { 1684 dnp->n_flag &= ~NREMOVEINPROG; 1685 mtx_unlock(&dnp->n_mtx); 1686 } 1687 if (dattrflag) 1688 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1689 mtx_lock(&dnp->n_mtx); 1690 dnp->n_flag |= NMODIFIED; 1691 if (!dattrflag) 1692 dnp->n_attrstamp = 0; 1693 mtx_unlock(&dnp->n_mtx); 1694 if (error && NFS_ISV4(dvp)) 1695 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1696 return (error); 1697} 1698 1699/* 1700 * nfs file rename call 1701 */ 1702static int 1703nfs_rename(struct vop_rename_args *ap) 1704{ 1705 struct vnode *fvp = ap->a_fvp; 1706 struct vnode *tvp = ap->a_tvp; 1707 struct vnode *fdvp = ap->a_fdvp; 1708 struct vnode *tdvp = ap->a_tdvp; 1709 struct componentname *tcnp = ap->a_tcnp; 1710 struct componentname *fcnp = ap->a_fcnp; 1711 struct nfsnode *fnp = VTONFS(ap->a_fvp); 1712 struct nfsnode *tdnp = VTONFS(ap->a_tdvp); 1713 struct nfsv4node *newv4 = NULL; 1714 int error; 1715 1716 KASSERT((tcnp->cn_flags & HASBUF) != 0 && 1717 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name")); 1718 /* Check for cross-device rename */ 1719 if ((fvp->v_mount != tdvp->v_mount) || 1720 (tvp && (fvp->v_mount != tvp->v_mount))) { 1721 error = EXDEV; 1722 goto out; 1723 } 1724 1725 if (fvp == tvp) { 1726 ncl_printf("nfs_rename: fvp == tvp (can't happen)\n"); 1727 error = 0; 1728 goto out; 1729 } 1730 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0) 1731 goto out; 1732 1733 /* 1734 * We have to flush B_DELWRI data prior to renaming 1735 * the file. If we don't, the delayed-write buffers 1736 * can be flushed out later after the file has gone stale 1737 * under NFSV3. NFSV2 does not have this problem because 1738 * ( as far as I can tell ) it flushes dirty buffers more 1739 * often. 1740 * 1741 * Skip the rename operation if the fsync fails, this can happen 1742 * due to the server's volume being full, when we pushed out data 1743 * that was written back to our cache earlier. Not checking for 1744 * this condition can result in potential (silent) data loss. 1745 */ 1746 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread); 1747 VOP_UNLOCK(fvp, 0); 1748 if (!error && tvp) 1749 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread); 1750 if (error) 1751 goto out; 1752 1753 /* 1754 * If the tvp exists and is in use, sillyrename it before doing the 1755 * rename of the new file over it. 1756 * XXX Can't sillyrename a directory. 1757 */ 1758 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename && 1759 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1760 vput(tvp); 1761 tvp = NULL; 1762 } 1763 1764 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1765 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1766 tcnp->cn_thread); 1767 1768 if (error == 0 && NFS_ISV4(tdvp)) { 1769 /* 1770 * For NFSv4, check to see if it is the same name and 1771 * replace the name, if it is different. 1772 */ 1773 MALLOC(newv4, struct nfsv4node *, 1774 sizeof (struct nfsv4node) + 1775 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1, 1776 M_NFSV4NODE, M_WAITOK); 1777 mtx_lock(&tdnp->n_mtx); 1778 mtx_lock(&fnp->n_mtx); 1779 if (fnp->n_v4 != NULL && fvp->v_type == VREG && 1780 (fnp->n_v4->n4_namelen != tcnp->cn_namelen || 1781 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4), 1782 tcnp->cn_namelen) || 1783 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen || 1784 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 1785 tdnp->n_fhp->nfh_len))) { 1786#ifdef notdef 1787{ char nnn[100]; int nnnl; 1788nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99; 1789bcopy(tcnp->cn_nameptr, nnn, nnnl); 1790nnn[nnnl] = '\0'; 1791printf("ren replace=%s\n",nnn); 1792} 1793#endif 1794 FREE((caddr_t)fnp->n_v4, M_NFSV4NODE); 1795 fnp->n_v4 = newv4; 1796 newv4 = NULL; 1797 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len; 1798 fnp->n_v4->n4_namelen = tcnp->cn_namelen; 1799 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 1800 tdnp->n_fhp->nfh_len); 1801 NFSBCOPY(tcnp->cn_nameptr, 1802 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen); 1803 } 1804 mtx_unlock(&tdnp->n_mtx); 1805 mtx_unlock(&fnp->n_mtx); 1806 if (newv4 != NULL) 1807 FREE((caddr_t)newv4, M_NFSV4NODE); 1808 } 1809 1810 if (fvp->v_type == VDIR) { 1811 if (tvp != NULL && tvp->v_type == VDIR) 1812 cache_purge(tdvp); 1813 cache_purge(fdvp); 1814 } 1815 1816out: 1817 if (tdvp == tvp) 1818 vrele(tdvp); 1819 else 1820 vput(tdvp); 1821 if (tvp) 1822 vput(tvp); 1823 vrele(fdvp); 1824 vrele(fvp); 1825 /* 1826 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1827 */ 1828 if (error == ENOENT) 1829 error = 0; 1830 return (error); 1831} 1832 1833/* 1834 * nfs file rename rpc called from nfs_remove() above 1835 */ 1836static int 1837nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp, 1838 struct sillyrename *sp) 1839{ 1840 1841 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen, 1842 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred, 1843 scnp->cn_thread)); 1844} 1845 1846/* 1847 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1848 */ 1849static int 1850nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr, 1851 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr, 1852 int tnamelen, struct ucred *cred, struct thread *td) 1853{ 1854 struct nfsvattr fnfsva, tnfsva; 1855 struct nfsnode *fdnp = VTONFS(fdvp); 1856 struct nfsnode *tdnp = VTONFS(tdvp); 1857 int error = 0, fattrflag, tattrflag; 1858 1859 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp, 1860 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag, 1861 &tattrflag, NULL, NULL); 1862 mtx_lock(&fdnp->n_mtx); 1863 fdnp->n_flag |= NMODIFIED; 1864 if (fattrflag != 0) { 1865 mtx_unlock(&fdnp->n_mtx); 1866 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1); 1867 } else { 1868 fdnp->n_attrstamp = 0; 1869 mtx_unlock(&fdnp->n_mtx); 1870 } 1871 mtx_lock(&tdnp->n_mtx); 1872 tdnp->n_flag |= NMODIFIED; 1873 if (tattrflag != 0) { 1874 mtx_unlock(&tdnp->n_mtx); 1875 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1); 1876 } else { 1877 tdnp->n_attrstamp = 0; 1878 mtx_unlock(&tdnp->n_mtx); 1879 } 1880 if (error && NFS_ISV4(fdvp)) 1881 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1882 return (error); 1883} 1884 1885/* 1886 * nfs hard link create call 1887 */ 1888static int 1889nfs_link(struct vop_link_args *ap) 1890{ 1891 struct vnode *vp = ap->a_vp; 1892 struct vnode *tdvp = ap->a_tdvp; 1893 struct componentname *cnp = ap->a_cnp; 1894 struct nfsnode *np, *tdnp; 1895 struct nfsvattr nfsva, dnfsva; 1896 int error = 0, attrflag, dattrflag; 1897 1898 if (vp->v_mount != tdvp->v_mount) { 1899 return (EXDEV); 1900 } 1901 1902 /* 1903 * Push all writes to the server, so that the attribute cache 1904 * doesn't get "out of sync" with the server. 1905 * XXX There should be a better way! 1906 */ 1907 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread); 1908 1909 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 1910 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag, 1911 &dattrflag, NULL); 1912 tdnp = VTONFS(tdvp); 1913 mtx_lock(&tdnp->n_mtx); 1914 tdnp->n_flag |= NMODIFIED; 1915 if (dattrflag != 0) { 1916 mtx_unlock(&tdnp->n_mtx); 1917 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1); 1918 } else { 1919 tdnp->n_attrstamp = 0; 1920 mtx_unlock(&tdnp->n_mtx); 1921 } 1922 if (attrflag) 1923 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1924 else { 1925 np = VTONFS(vp); 1926 mtx_lock(&np->n_mtx); 1927 np->n_attrstamp = 0; 1928 mtx_unlock(&np->n_mtx); 1929 } 1930 /* 1931 * If negative lookup caching is enabled, I might as well 1932 * add an entry for this node. Not necessary for correctness, 1933 * but if negative caching is enabled, then the system 1934 * must care about lookup caching hit rate, so... 1935 */ 1936 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 && 1937 (cnp->cn_flags & MAKEENTRY)) 1938 cache_enter(tdvp, vp, cnp); 1939 if (error && NFS_ISV4(vp)) 1940 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, 1941 (gid_t)0); 1942 return (error); 1943} 1944 1945/* 1946 * nfs symbolic link create call 1947 */ 1948static int 1949nfs_symlink(struct vop_symlink_args *ap) 1950{ 1951 struct vnode *dvp = ap->a_dvp; 1952 struct vattr *vap = ap->a_vap; 1953 struct componentname *cnp = ap->a_cnp; 1954 struct nfsvattr nfsva, dnfsva; 1955 struct nfsfh *nfhp; 1956 struct nfsnode *np = NULL, *dnp; 1957 struct vnode *newvp = NULL; 1958 int error = 0, attrflag, dattrflag, ret; 1959 1960 vap->va_type = VLNK; 1961 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1962 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, 1963 &nfsva, &nfhp, &attrflag, &dattrflag, NULL); 1964 if (nfhp) { 1965 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, 1966 &np, NULL, LK_EXCLUSIVE); 1967 if (!ret) 1968 newvp = NFSTOV(np); 1969 else if (!error) 1970 error = ret; 1971 } 1972 if (newvp != NULL) { 1973 if (attrflag) 1974 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1975 0, 1); 1976 } else if (!error) { 1977 /* 1978 * If we do not have an error and we could not extract the 1979 * newvp from the response due to the request being NFSv2, we 1980 * have to do a lookup in order to obtain a newvp to return. 1981 */ 1982 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1983 cnp->cn_cred, cnp->cn_thread, &np); 1984 if (!error) 1985 newvp = NFSTOV(np); 1986 } 1987 if (error) { 1988 if (newvp) 1989 vput(newvp); 1990 if (NFS_ISV4(dvp)) 1991 error = nfscl_maperr(cnp->cn_thread, error, 1992 vap->va_uid, vap->va_gid); 1993 } else { 1994 /* 1995 * If negative lookup caching is enabled, I might as well 1996 * add an entry for this node. Not necessary for correctness, 1997 * but if negative caching is enabled, then the system 1998 * must care about lookup caching hit rate, so... 1999 */ 2000 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2001 (cnp->cn_flags & MAKEENTRY)) 2002 cache_enter(dvp, newvp, cnp); 2003 *ap->a_vpp = newvp; 2004 } 2005 2006 dnp = VTONFS(dvp); 2007 mtx_lock(&dnp->n_mtx); 2008 dnp->n_flag |= NMODIFIED; 2009 if (dattrflag != 0) { 2010 mtx_unlock(&dnp->n_mtx); 2011 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2012 } else { 2013 dnp->n_attrstamp = 0; 2014 mtx_unlock(&dnp->n_mtx); 2015 } 2016 return (error); 2017} 2018 2019/* 2020 * nfs make dir call 2021 */ 2022static int 2023nfs_mkdir(struct vop_mkdir_args *ap) 2024{ 2025 struct vnode *dvp = ap->a_dvp; 2026 struct vattr *vap = ap->a_vap; 2027 struct componentname *cnp = ap->a_cnp; 2028 struct nfsnode *np = NULL, *dnp; 2029 struct vnode *newvp = NULL; 2030 struct vattr vattr; 2031 struct nfsfh *nfhp; 2032 struct nfsvattr nfsva, dnfsva; 2033 int error = 0, attrflag, dattrflag, ret; 2034 2035 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) 2036 return (error); 2037 vap->va_type = VDIR; 2038 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2039 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp, 2040 &attrflag, &dattrflag, NULL); 2041 dnp = VTONFS(dvp); 2042 mtx_lock(&dnp->n_mtx); 2043 dnp->n_flag |= NMODIFIED; 2044 if (dattrflag != 0) { 2045 mtx_unlock(&dnp->n_mtx); 2046 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2047 } else { 2048 dnp->n_attrstamp = 0; 2049 mtx_unlock(&dnp->n_mtx); 2050 } 2051 if (nfhp) { 2052 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, 2053 &np, NULL, LK_EXCLUSIVE); 2054 if (!ret) { 2055 newvp = NFSTOV(np); 2056 if (attrflag) 2057 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 2058 NULL, 0, 1); 2059 } else if (!error) 2060 error = ret; 2061 } 2062 if (!error && newvp == NULL) { 2063 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2064 cnp->cn_cred, cnp->cn_thread, &np); 2065 if (!error) { 2066 newvp = NFSTOV(np); 2067 if (newvp->v_type != VDIR) 2068 error = EEXIST; 2069 } 2070 } 2071 if (error) { 2072 if (newvp) 2073 vput(newvp); 2074 if (NFS_ISV4(dvp)) 2075 error = nfscl_maperr(cnp->cn_thread, error, 2076 vap->va_uid, vap->va_gid); 2077 } else { 2078 /* 2079 * If negative lookup caching is enabled, I might as well 2080 * add an entry for this node. Not necessary for correctness, 2081 * but if negative caching is enabled, then the system 2082 * must care about lookup caching hit rate, so... 2083 */ 2084 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2085 (cnp->cn_flags & MAKEENTRY)) 2086 cache_enter(dvp, newvp, cnp); 2087 *ap->a_vpp = newvp; 2088 } 2089 return (error); 2090} 2091 2092/* 2093 * nfs remove directory call 2094 */ 2095static int 2096nfs_rmdir(struct vop_rmdir_args *ap) 2097{ 2098 struct vnode *vp = ap->a_vp; 2099 struct vnode *dvp = ap->a_dvp; 2100 struct componentname *cnp = ap->a_cnp; 2101 struct nfsnode *dnp; 2102 struct nfsvattr dnfsva; 2103 int error, dattrflag; 2104 2105 if (dvp == vp) 2106 return (EINVAL); 2107 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2108 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL); 2109 dnp = VTONFS(dvp); 2110 mtx_lock(&dnp->n_mtx); 2111 dnp->n_flag |= NMODIFIED; 2112 if (dattrflag != 0) { 2113 mtx_unlock(&dnp->n_mtx); 2114 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2115 } else { 2116 dnp->n_attrstamp = 0; 2117 mtx_unlock(&dnp->n_mtx); 2118 } 2119 2120 cache_purge(dvp); 2121 cache_purge(vp); 2122 if (error && NFS_ISV4(dvp)) 2123 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, 2124 (gid_t)0); 2125 /* 2126 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2127 */ 2128 if (error == ENOENT) 2129 error = 0; 2130 return (error); 2131} 2132 2133/* 2134 * nfs readdir call 2135 */ 2136static int 2137nfs_readdir(struct vop_readdir_args *ap) 2138{ 2139 struct vnode *vp = ap->a_vp; 2140 struct nfsnode *np = VTONFS(vp); 2141 struct uio *uio = ap->a_uio; 2142 int tresid, error = 0; 2143 struct vattr vattr; 2144 2145 if (vp->v_type != VDIR) 2146 return(EPERM); 2147 2148 /* 2149 * First, check for hit on the EOF offset cache 2150 */ 2151 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2152 (np->n_flag & NMODIFIED) == 0) { 2153 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) { 2154 mtx_lock(&np->n_mtx); 2155 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) || 2156 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 2157 mtx_unlock(&np->n_mtx); 2158 NFSINCRGLOBAL(newnfsstats.direofcache_hits); 2159 return (0); 2160 } else 2161 mtx_unlock(&np->n_mtx); 2162 } 2163 } 2164 2165 /* 2166 * Call ncl_bioread() to do the real work. 2167 */ 2168 tresid = uio->uio_resid; 2169 error = ncl_bioread(vp, uio, 0, ap->a_cred); 2170 2171 if (!error && uio->uio_resid == tresid) 2172 NFSINCRGLOBAL(newnfsstats.direofcache_misses); 2173 return (error); 2174} 2175 2176/* 2177 * Readdir rpc call. 2178 * Called from below the buffer cache by ncl_doio(). 2179 */ 2180int 2181ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2182 struct thread *td) 2183{ 2184 struct nfsvattr nfsva; 2185 nfsuint64 *cookiep, cookie; 2186 struct nfsnode *dnp = VTONFS(vp); 2187 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2188 int error = 0, eof, attrflag; 2189 2190 KASSERT(uiop->uio_iovcnt == 1 && 2191 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2192 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2193 ("nfs readdirrpc bad uio")); 2194 2195 /* 2196 * If there is no cookie, assume directory was stale. 2197 */ 2198 ncl_dircookie_lock(dnp); 2199 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2200 if (cookiep) { 2201 cookie = *cookiep; 2202 ncl_dircookie_unlock(dnp); 2203 } else { 2204 ncl_dircookie_unlock(dnp); 2205 return (NFSERR_BAD_COOKIE); 2206 } 2207 2208 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2209 (void)ncl_fsinfo(nmp, vp, cred, td); 2210 2211 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva, 2212 &attrflag, &eof, NULL); 2213 if (attrflag) 2214 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2215 2216 if (!error) { 2217 /* 2218 * We are now either at the end of the directory or have filled 2219 * the block. 2220 */ 2221 if (eof) 2222 dnp->n_direofoffset = uiop->uio_offset; 2223 else { 2224 if (uiop->uio_resid > 0) 2225 ncl_printf("EEK! readdirrpc resid > 0\n"); 2226 ncl_dircookie_lock(dnp); 2227 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2228 *cookiep = cookie; 2229 ncl_dircookie_unlock(dnp); 2230 } 2231 } else if (NFS_ISV4(vp)) { 2232 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2233 } 2234 return (error); 2235} 2236 2237/* 2238 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc(). 2239 */ 2240int 2241ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2242 struct thread *td) 2243{ 2244 struct nfsvattr nfsva; 2245 nfsuint64 *cookiep, cookie; 2246 struct nfsnode *dnp = VTONFS(vp); 2247 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2248 int error = 0, attrflag, eof; 2249 2250 KASSERT(uiop->uio_iovcnt == 1 && 2251 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2252 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2253 ("nfs readdirplusrpc bad uio")); 2254 2255 /* 2256 * If there is no cookie, assume directory was stale. 2257 */ 2258 ncl_dircookie_lock(dnp); 2259 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2260 if (cookiep) { 2261 cookie = *cookiep; 2262 ncl_dircookie_unlock(dnp); 2263 } else { 2264 ncl_dircookie_unlock(dnp); 2265 return (NFSERR_BAD_COOKIE); 2266 } 2267 2268 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2269 (void)ncl_fsinfo(nmp, vp, cred, td); 2270 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva, 2271 &attrflag, &eof, NULL); 2272 if (attrflag) 2273 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2274 2275 if (!error) { 2276 /* 2277 * We are now either at end of the directory or have filled the 2278 * the block. 2279 */ 2280 if (eof) 2281 dnp->n_direofoffset = uiop->uio_offset; 2282 else { 2283 if (uiop->uio_resid > 0) 2284 ncl_printf("EEK! readdirplusrpc resid > 0\n"); 2285 ncl_dircookie_lock(dnp); 2286 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2287 *cookiep = cookie; 2288 ncl_dircookie_unlock(dnp); 2289 } 2290 } else if (NFS_ISV4(vp)) { 2291 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2292 } 2293 return (error); 2294} 2295 2296/* 2297 * Silly rename. To make the NFS filesystem that is stateless look a little 2298 * more like the "ufs" a remove of an active vnode is translated to a rename 2299 * to a funny looking filename that is removed by nfs_inactive on the 2300 * nfsnode. There is the potential for another process on a different client 2301 * to create the same funny name between the nfs_lookitup() fails and the 2302 * nfs_rename() completes, but... 2303 */ 2304static int 2305nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2306{ 2307 struct sillyrename *sp; 2308 struct nfsnode *np; 2309 int error; 2310 short pid; 2311 unsigned int lticks; 2312 2313 cache_purge(dvp); 2314 np = VTONFS(vp); 2315 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir")); 2316 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2317 M_NEWNFSREQ, M_WAITOK); 2318 sp->s_cred = crhold(cnp->cn_cred); 2319 sp->s_dvp = dvp; 2320 VREF(dvp); 2321 2322 /* 2323 * Fudge together a funny name. 2324 * Changing the format of the funny name to accomodate more 2325 * sillynames per directory. 2326 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is 2327 * CPU ticks since boot. 2328 */ 2329 pid = cnp->cn_thread->td_proc->p_pid; 2330 lticks = (unsigned int)ticks; 2331 for ( ; ; ) { 2332 sp->s_namlen = sprintf(sp->s_name, 2333 ".nfs.%08x.%04x4.4", lticks, 2334 pid); 2335 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2336 cnp->cn_thread, NULL)) 2337 break; 2338 lticks++; 2339 } 2340 error = nfs_renameit(dvp, vp, cnp, sp); 2341 if (error) 2342 goto bad; 2343 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2344 cnp->cn_thread, &np); 2345 np->n_sillyrename = sp; 2346 return (0); 2347bad: 2348 vrele(sp->s_dvp); 2349 crfree(sp->s_cred); 2350 free((caddr_t)sp, M_NEWNFSREQ); 2351 return (error); 2352} 2353 2354/* 2355 * Look up a file name and optionally either update the file handle or 2356 * allocate an nfsnode, depending on the value of npp. 2357 * npp == NULL --> just do the lookup 2358 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2359 * handled too 2360 * *npp != NULL --> update the file handle in the vnode 2361 */ 2362static int 2363nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred, 2364 struct thread *td, struct nfsnode **npp) 2365{ 2366 struct vnode *newvp = NULL, *vp; 2367 struct nfsnode *np, *dnp = VTONFS(dvp); 2368 struct nfsfh *nfhp, *onfhp; 2369 struct nfsvattr nfsva, dnfsva; 2370 struct componentname cn; 2371 int error = 0, attrflag, dattrflag; 2372 u_int hash; 2373 2374 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva, 2375 &nfhp, &attrflag, &dattrflag, NULL); 2376 if (dattrflag) 2377 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2378 if (npp && !error) { 2379 if (*npp != NULL) { 2380 np = *npp; 2381 vp = NFSTOV(np); 2382 /* 2383 * For NFSv4, check to see if it is the same name and 2384 * replace the name, if it is different. 2385 */ 2386 if (np->n_v4 != NULL && nfsva.na_type == VREG && 2387 (np->n_v4->n4_namelen != len || 2388 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) || 2389 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || 2390 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2391 dnp->n_fhp->nfh_len))) { 2392#ifdef notdef 2393{ char nnn[100]; int nnnl; 2394nnnl = (len < 100) ? len : 99; 2395bcopy(name, nnn, nnnl); 2396nnn[nnnl] = '\0'; 2397printf("replace=%s\n",nnn); 2398} 2399#endif 2400 FREE((caddr_t)np->n_v4, M_NFSV4NODE); 2401 MALLOC(np->n_v4, struct nfsv4node *, 2402 sizeof (struct nfsv4node) + 2403 dnp->n_fhp->nfh_len + len - 1, 2404 M_NFSV4NODE, M_WAITOK); 2405 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; 2406 np->n_v4->n4_namelen = len; 2407 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2408 dnp->n_fhp->nfh_len); 2409 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len); 2410 } 2411 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, 2412 FNV1_32_INIT); 2413 onfhp = np->n_fhp; 2414 /* 2415 * Rehash node for new file handle. 2416 */ 2417 vfs_hash_rehash(vp, hash); 2418 np->n_fhp = nfhp; 2419 if (onfhp != NULL) 2420 FREE((caddr_t)onfhp, M_NFSFH); 2421 newvp = NFSTOV(np); 2422 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) { 2423 FREE((caddr_t)nfhp, M_NFSFH); 2424 VREF(dvp); 2425 newvp = dvp; 2426 } else { 2427 cn.cn_nameptr = name; 2428 cn.cn_namelen = len; 2429 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td, 2430 &np, NULL, LK_EXCLUSIVE); 2431 if (error) 2432 return (error); 2433 newvp = NFSTOV(np); 2434 } 2435 if (!attrflag && *npp == NULL) { 2436 vrele(newvp); 2437 return (ENOENT); 2438 } 2439 if (attrflag) 2440 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 2441 0, 1); 2442 } 2443 if (npp && *npp == NULL) { 2444 if (error) { 2445 if (newvp) { 2446 if (newvp == dvp) 2447 vrele(newvp); 2448 else 2449 vput(newvp); 2450 } 2451 } else 2452 *npp = np; 2453 } 2454 if (error && NFS_ISV4(dvp)) 2455 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2456 return (error); 2457} 2458 2459/* 2460 * Nfs Version 3 and 4 commit rpc 2461 */ 2462int 2463ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred, 2464 struct thread *td) 2465{ 2466 struct nfsvattr nfsva; 2467 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2468 int error, attrflag; 2469 u_char verf[NFSX_VERF]; 2470 2471 mtx_lock(&nmp->nm_mtx); 2472 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) { 2473 mtx_unlock(&nmp->nm_mtx); 2474 return (0); 2475 } 2476 mtx_unlock(&nmp->nm_mtx); 2477 error = nfsrpc_commit(vp, offset, cnt, cred, td, verf, &nfsva, 2478 &attrflag, NULL); 2479 if (!error) { 2480 if (NFSBCMP((caddr_t)nmp->nm_verf, verf, NFSX_VERF)) { 2481 NFSBCOPY(verf, (caddr_t)nmp->nm_verf, NFSX_VERF); 2482 error = NFSERR_STALEWRITEVERF; 2483 } 2484 if (!error && attrflag) 2485 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 2486 0, 1); 2487 } else if (NFS_ISV4(vp)) { 2488 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2489 } 2490 return (error); 2491} 2492 2493/* 2494 * Strategy routine. 2495 * For async requests when nfsiod(s) are running, queue the request by 2496 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the 2497 * request. 2498 */ 2499static int 2500nfs_strategy(struct vop_strategy_args *ap) 2501{ 2502 struct buf *bp = ap->a_bp; 2503 struct ucred *cr; 2504 2505 KASSERT(!(bp->b_flags & B_DONE), 2506 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); 2507 BUF_ASSERT_HELD(bp); 2508 2509 if (bp->b_iocmd == BIO_READ) 2510 cr = bp->b_rcred; 2511 else 2512 cr = bp->b_wcred; 2513 2514 /* 2515 * If the op is asynchronous and an i/o daemon is waiting 2516 * queue the request, wake it up and wait for completion 2517 * otherwise just do it ourselves. 2518 */ 2519 if ((bp->b_flags & B_ASYNC) == 0 || 2520 ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread)) 2521 (void) ncl_doio(ap->a_vp, bp, cr, curthread, 1); 2522 return (0); 2523} 2524 2525/* 2526 * fsync vnode op. Just call ncl_flush() with commit == 1. 2527 */ 2528/* ARGSUSED */ 2529static int 2530nfs_fsync(struct vop_fsync_args *ap) 2531{ 2532 return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1, 0)); 2533} 2534 2535/* 2536 * Flush all the blocks associated with a vnode. 2537 * Walk through the buffer pool and push any dirty pages 2538 * associated with the vnode. 2539 * If the called_from_renewthread argument is TRUE, it has been called 2540 * from the NFSv4 renew thread and, as such, cannot block indefinitely 2541 * waiting for a buffer write to complete. 2542 */ 2543int 2544ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td, 2545 int commit, int called_from_renewthread) 2546{ 2547 struct nfsnode *np = VTONFS(vp); 2548 struct buf *bp; 2549 int i; 2550 struct buf *nbp; 2551 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2552 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos; 2553 int passone = 1, trycnt = 0; 2554 u_quad_t off, endoff, toff; 2555 struct ucred* wcred = NULL; 2556 struct buf **bvec = NULL; 2557 struct bufobj *bo; 2558#ifndef NFS_COMMITBVECSIZ 2559#define NFS_COMMITBVECSIZ 20 2560#endif 2561 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ]; 2562 int bvecsize = 0, bveccount; 2563 2564 if (called_from_renewthread != 0) 2565 slptimeo = hz; 2566 if (nmp->nm_flag & NFSMNT_INT) 2567 slpflag = NFS_PCATCH; 2568 if (!commit) 2569 passone = 0; 2570 bo = &vp->v_bufobj; 2571 /* 2572 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the 2573 * server, but has not been committed to stable storage on the server 2574 * yet. On the first pass, the byte range is worked out and the commit 2575 * rpc is done. On the second pass, ncl_writebp() is called to do the 2576 * job. 2577 */ 2578again: 2579 off = (u_quad_t)-1; 2580 endoff = 0; 2581 bvecpos = 0; 2582 if (NFS_ISV34(vp) && commit) { 2583 if (bvec != NULL && bvec != bvec_on_stack) 2584 free(bvec, M_TEMP); 2585 /* 2586 * Count up how many buffers waiting for a commit. 2587 */ 2588 bveccount = 0; 2589 BO_LOCK(bo); 2590 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2591 if (!BUF_ISLOCKED(bp) && 2592 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) 2593 == (B_DELWRI | B_NEEDCOMMIT)) 2594 bveccount++; 2595 } 2596 /* 2597 * Allocate space to remember the list of bufs to commit. It is 2598 * important to use M_NOWAIT here to avoid a race with nfs_write. 2599 * If we can't get memory (for whatever reason), we will end up 2600 * committing the buffers one-by-one in the loop below. 2601 */ 2602 if (bveccount > NFS_COMMITBVECSIZ) { 2603 /* 2604 * Release the vnode interlock to avoid a lock 2605 * order reversal. 2606 */ 2607 BO_UNLOCK(bo); 2608 bvec = (struct buf **) 2609 malloc(bveccount * sizeof(struct buf *), 2610 M_TEMP, M_NOWAIT); 2611 BO_LOCK(bo); 2612 if (bvec == NULL) { 2613 bvec = bvec_on_stack; 2614 bvecsize = NFS_COMMITBVECSIZ; 2615 } else 2616 bvecsize = bveccount; 2617 } else { 2618 bvec = bvec_on_stack; 2619 bvecsize = NFS_COMMITBVECSIZ; 2620 } 2621 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2622 if (bvecpos >= bvecsize) 2623 break; 2624 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2625 nbp = TAILQ_NEXT(bp, b_bobufs); 2626 continue; 2627 } 2628 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 2629 (B_DELWRI | B_NEEDCOMMIT)) { 2630 BUF_UNLOCK(bp); 2631 nbp = TAILQ_NEXT(bp, b_bobufs); 2632 continue; 2633 } 2634 BO_UNLOCK(bo); 2635 bremfree(bp); 2636 /* 2637 * Work out if all buffers are using the same cred 2638 * so we can deal with them all with one commit. 2639 * 2640 * NOTE: we are not clearing B_DONE here, so we have 2641 * to do it later on in this routine if we intend to 2642 * initiate I/O on the bp. 2643 * 2644 * Note: to avoid loopback deadlocks, we do not 2645 * assign b_runningbufspace. 2646 */ 2647 if (wcred == NULL) 2648 wcred = bp->b_wcred; 2649 else if (wcred != bp->b_wcred) 2650 wcred = NOCRED; 2651 vfs_busy_pages(bp, 1); 2652 2653 BO_LOCK(bo); 2654 /* 2655 * bp is protected by being locked, but nbp is not 2656 * and vfs_busy_pages() may sleep. We have to 2657 * recalculate nbp. 2658 */ 2659 nbp = TAILQ_NEXT(bp, b_bobufs); 2660 2661 /* 2662 * A list of these buffers is kept so that the 2663 * second loop knows which buffers have actually 2664 * been committed. This is necessary, since there 2665 * may be a race between the commit rpc and new 2666 * uncommitted writes on the file. 2667 */ 2668 bvec[bvecpos++] = bp; 2669 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2670 bp->b_dirtyoff; 2671 if (toff < off) 2672 off = toff; 2673 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); 2674 if (toff > endoff) 2675 endoff = toff; 2676 } 2677 BO_UNLOCK(bo); 2678 } 2679 if (bvecpos > 0) { 2680 /* 2681 * Commit data on the server, as required. 2682 * If all bufs are using the same wcred, then use that with 2683 * one call for all of them, otherwise commit each one 2684 * separately. 2685 */ 2686 if (wcred != NOCRED) 2687 retv = ncl_commit(vp, off, (int)(endoff - off), 2688 wcred, td); 2689 else { 2690 retv = 0; 2691 for (i = 0; i < bvecpos; i++) { 2692 off_t off, size; 2693 bp = bvec[i]; 2694 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2695 bp->b_dirtyoff; 2696 size = (u_quad_t)(bp->b_dirtyend 2697 - bp->b_dirtyoff); 2698 retv = ncl_commit(vp, off, (int)size, 2699 bp->b_wcred, td); 2700 if (retv) break; 2701 } 2702 } 2703 2704 if (retv == NFSERR_STALEWRITEVERF) 2705 ncl_clearcommit(vp->v_mount); 2706 2707 /* 2708 * Now, either mark the blocks I/O done or mark the 2709 * blocks dirty, depending on whether the commit 2710 * succeeded. 2711 */ 2712 for (i = 0; i < bvecpos; i++) { 2713 bp = bvec[i]; 2714 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 2715 if (retv) { 2716 /* 2717 * Error, leave B_DELWRI intact 2718 */ 2719 vfs_unbusy_pages(bp); 2720 brelse(bp); 2721 } else { 2722 /* 2723 * Success, remove B_DELWRI ( bundirty() ). 2724 * 2725 * b_dirtyoff/b_dirtyend seem to be NFS 2726 * specific. We should probably move that 2727 * into bundirty(). XXX 2728 */ 2729 bufobj_wref(bo); 2730 bp->b_flags |= B_ASYNC; 2731 bundirty(bp); 2732 bp->b_flags &= ~B_DONE; 2733 bp->b_ioflags &= ~BIO_ERROR; 2734 bp->b_dirtyoff = bp->b_dirtyend = 0; 2735 bufdone(bp); 2736 } 2737 } 2738 } 2739 2740 /* 2741 * Start/do any write(s) that are required. 2742 */ 2743loop: 2744 BO_LOCK(bo); 2745 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2746 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2747 if (waitfor != MNT_WAIT || passone) 2748 continue; 2749 2750 error = BUF_TIMELOCK(bp, 2751 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 2752 BO_MTX(bo), "nfsfsync", slpflag, slptimeo); 2753 if (error == 0) { 2754 BUF_UNLOCK(bp); 2755 goto loop; 2756 } 2757 if (error == ENOLCK) { 2758 error = 0; 2759 goto loop; 2760 } 2761 if (called_from_renewthread != 0) { 2762 /* 2763 * Return EIO so the flush will be retried 2764 * later. 2765 */ 2766 error = EIO; 2767 goto done; 2768 } 2769 if (newnfs_sigintr(nmp, td)) { 2770 error = EINTR; 2771 goto done; 2772 } 2773 if (slpflag & PCATCH) { 2774 slpflag = 0; 2775 slptimeo = 2 * hz; 2776 } 2777 goto loop; 2778 } 2779 if ((bp->b_flags & B_DELWRI) == 0) 2780 panic("nfs_fsync: not dirty"); 2781 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) { 2782 BUF_UNLOCK(bp); 2783 continue; 2784 } 2785 BO_UNLOCK(bo); 2786 bremfree(bp); 2787 if (passone || !commit) 2788 bp->b_flags |= B_ASYNC; 2789 else 2790 bp->b_flags |= B_ASYNC; 2791 bwrite(bp); 2792 if (newnfs_sigintr(nmp, td)) { 2793 error = EINTR; 2794 goto done; 2795 } 2796 goto loop; 2797 } 2798 if (passone) { 2799 passone = 0; 2800 BO_UNLOCK(bo); 2801 goto again; 2802 } 2803 if (waitfor == MNT_WAIT) { 2804 while (bo->bo_numoutput) { 2805 error = bufobj_wwait(bo, slpflag, slptimeo); 2806 if (error) { 2807 BO_UNLOCK(bo); 2808 if (called_from_renewthread != 0) { 2809 /* 2810 * Return EIO so that the flush will be 2811 * retried later. 2812 */ 2813 error = EIO; 2814 goto done; 2815 } 2816 error = newnfs_sigintr(nmp, td); 2817 if (error) 2818 goto done; 2819 if (slpflag & PCATCH) { 2820 slpflag = 0; 2821 slptimeo = 2 * hz; 2822 } 2823 BO_LOCK(bo); 2824 } 2825 } 2826 if (bo->bo_dirty.bv_cnt != 0 && commit) { 2827 BO_UNLOCK(bo); 2828 goto loop; 2829 } 2830 /* 2831 * Wait for all the async IO requests to drain 2832 */ 2833 BO_UNLOCK(bo); 2834 mtx_lock(&np->n_mtx); 2835 while (np->n_directio_asyncwr > 0) { 2836 np->n_flag |= NFSYNCWAIT; 2837 error = newnfs_msleep(td, &np->n_directio_asyncwr, 2838 &np->n_mtx, slpflag | (PRIBIO + 1), 2839 "nfsfsync", 0); 2840 if (error) { 2841 if (newnfs_sigintr(nmp, td)) { 2842 mtx_unlock(&np->n_mtx); 2843 error = EINTR; 2844 goto done; 2845 } 2846 } 2847 } 2848 mtx_unlock(&np->n_mtx); 2849 } else 2850 BO_UNLOCK(bo); 2851 mtx_lock(&np->n_mtx); 2852 if (np->n_flag & NWRITEERR) { 2853 error = np->n_error; 2854 np->n_flag &= ~NWRITEERR; 2855 } 2856 if (commit && bo->bo_dirty.bv_cnt == 0 && 2857 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0) 2858 np->n_flag &= ~NMODIFIED; 2859 mtx_unlock(&np->n_mtx); 2860done: 2861 if (bvec != NULL && bvec != bvec_on_stack) 2862 free(bvec, M_TEMP); 2863 if (error == 0 && commit != 0 && waitfor == MNT_WAIT && 2864 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 || 2865 np->n_directio_asyncwr != 0) && trycnt++ < 5) { 2866 /* try, try again... */ 2867 passone = 1; 2868 wcred = NULL; 2869 bvec = NULL; 2870 bvecsize = 0; 2871printf("try%d\n", trycnt); 2872 goto again; 2873 } 2874 return (error); 2875} 2876 2877/* 2878 * NFS advisory byte-level locks. 2879 */ 2880static int 2881nfs_advlock(struct vop_advlock_args *ap) 2882{ 2883 struct vnode *vp = ap->a_vp; 2884 struct ucred *cred; 2885 struct nfsnode *np = VTONFS(ap->a_vp); 2886 struct proc *p = (struct proc *)ap->a_id; 2887 struct thread *td = curthread; /* XXX */ 2888 struct vattr va; 2889 int ret, error = EOPNOTSUPP; 2890 u_quad_t size; 2891 2892 if (NFS_ISV4(vp) && (ap->a_flags & F_POSIX)) { 2893 cred = p->p_ucred; 2894 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2895 if (vp->v_iflag & VI_DOOMED) { 2896 VOP_UNLOCK(vp, 0); 2897 return (EBADF); 2898 } 2899 2900 /* 2901 * If this is unlocking a write locked region, flush and 2902 * commit them before unlocking. This is required by 2903 * RFC3530 Sec. 9.3.2. 2904 */ 2905 if (ap->a_op == F_UNLCK && 2906 nfscl_checkwritelocked(vp, ap->a_fl, cred, td)) 2907 (void) ncl_flush(vp, MNT_WAIT, cred, td, 1, 0); 2908 2909 /* 2910 * Loop around doing the lock op, while a blocking lock 2911 * must wait for the lock op to succeed. 2912 */ 2913 do { 2914 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op, 2915 ap->a_fl, 0, cred, td); 2916 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 2917 ap->a_op == F_SETLK) { 2918 VOP_UNLOCK(vp, 0); 2919 error = nfs_catnap(PZERO | PCATCH, ret, 2920 "ncladvl"); 2921 if (error) 2922 return (EINTR); 2923 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2924 if (vp->v_iflag & VI_DOOMED) { 2925 VOP_UNLOCK(vp, 0); 2926 return (EBADF); 2927 } 2928 } 2929 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 2930 ap->a_op == F_SETLK); 2931 if (ret == NFSERR_DENIED) { 2932 VOP_UNLOCK(vp, 0); 2933 return (EAGAIN); 2934 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) { 2935 VOP_UNLOCK(vp, 0); 2936 return (ret); 2937 } else if (ret != 0) { 2938 VOP_UNLOCK(vp, 0); 2939 return (EACCES); 2940 } 2941 2942 /* 2943 * Now, if we just got a lock, invalidate data in the buffer 2944 * cache, as required, so that the coherency conforms with 2945 * RFC3530 Sec. 9.3.2. 2946 */ 2947 if (ap->a_op == F_SETLK) { 2948 if ((np->n_flag & NMODIFIED) == 0) { 2949 np->n_attrstamp = 0; 2950 ret = VOP_GETATTR(vp, &va, cred); 2951 } 2952 if ((np->n_flag & NMODIFIED) || ret || 2953 np->n_change != va.va_filerev) { 2954 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1); 2955 np->n_attrstamp = 0; 2956 ret = VOP_GETATTR(vp, &va, cred); 2957 if (!ret) { 2958 np->n_mtime = va.va_mtime; 2959 np->n_change = va.va_filerev; 2960 } 2961 } 2962 } 2963 VOP_UNLOCK(vp, 0); 2964 return (0); 2965 } else if (!NFS_ISV4(vp)) { 2966 error = vn_lock(vp, LK_SHARED); 2967 if (error) 2968 return (error); 2969 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 2970 size = VTONFS(vp)->n_size; 2971 VOP_UNLOCK(vp, 0); 2972 error = lf_advlock(ap, &(vp->v_lockf), size); 2973 } else { 2974 if (nfs_advlock_p != NULL) 2975 error = nfs_advlock_p(ap); 2976 else { 2977 VOP_UNLOCK(vp, 0); 2978 error = ENOLCK; 2979 } 2980 } 2981 } 2982 return (error); 2983} 2984 2985/* 2986 * NFS advisory byte-level locks. 2987 */ 2988static int 2989nfs_advlockasync(struct vop_advlockasync_args *ap) 2990{ 2991 struct vnode *vp = ap->a_vp; 2992 u_quad_t size; 2993 int error; 2994 2995 if (NFS_ISV4(vp)) 2996 return (EOPNOTSUPP); 2997 error = vn_lock(vp, LK_SHARED); 2998 if (error) 2999 return (error); 3000 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3001 size = VTONFS(vp)->n_size; 3002 VOP_UNLOCK(vp, 0); 3003 error = lf_advlockasync(ap, &(vp->v_lockf), size); 3004 } else { 3005 VOP_UNLOCK(vp, 0); 3006 error = EOPNOTSUPP; 3007 } 3008 return (error); 3009} 3010 3011/* 3012 * Print out the contents of an nfsnode. 3013 */ 3014static int 3015nfs_print(struct vop_print_args *ap) 3016{ 3017 struct vnode *vp = ap->a_vp; 3018 struct nfsnode *np = VTONFS(vp); 3019 3020 ncl_printf("\tfileid %ld fsid 0x%x", 3021 np->n_vattr.na_fileid, np->n_vattr.na_fsid); 3022 if (vp->v_type == VFIFO) 3023 fifo_printinfo(vp); 3024 printf("\n"); 3025 return (0); 3026} 3027 3028/* 3029 * This is the "real" nfs::bwrite(struct buf*). 3030 * We set B_CACHE if this is a VMIO buffer. 3031 */ 3032int 3033ncl_writebp(struct buf *bp, int force __unused, struct thread *td) 3034{ 3035 int s; 3036 int oldflags = bp->b_flags; 3037#if 0 3038 int retv = 1; 3039 off_t off; 3040#endif 3041 3042 BUF_ASSERT_HELD(bp); 3043 3044 if (bp->b_flags & B_INVAL) { 3045 brelse(bp); 3046 return(0); 3047 } 3048 3049 bp->b_flags |= B_CACHE; 3050 3051 /* 3052 * Undirty the bp. We will redirty it later if the I/O fails. 3053 */ 3054 3055 s = splbio(); 3056 bundirty(bp); 3057 bp->b_flags &= ~B_DONE; 3058 bp->b_ioflags &= ~BIO_ERROR; 3059 bp->b_iocmd = BIO_WRITE; 3060 3061 bufobj_wref(bp->b_bufobj); 3062 curthread->td_ru.ru_oublock++; 3063 splx(s); 3064 3065 /* 3066 * Note: to avoid loopback deadlocks, we do not 3067 * assign b_runningbufspace. 3068 */ 3069 vfs_busy_pages(bp, 1); 3070 3071 BUF_KERNPROC(bp); 3072 bp->b_iooffset = dbtob(bp->b_blkno); 3073 bstrategy(bp); 3074 3075 if( (oldflags & B_ASYNC) == 0) { 3076 int rtval = bufwait(bp); 3077 3078 if (oldflags & B_DELWRI) { 3079 s = splbio(); 3080 reassignbuf(bp); 3081 splx(s); 3082 } 3083 brelse(bp); 3084 return (rtval); 3085 } 3086 3087 return (0); 3088} 3089 3090/* 3091 * nfs special file access vnode op. 3092 * Essentially just get vattr and then imitate iaccess() since the device is 3093 * local to the client. 3094 */ 3095static int 3096nfsspec_access(struct vop_access_args *ap) 3097{ 3098 struct vattr *vap; 3099 struct ucred *cred = ap->a_cred; 3100 struct vnode *vp = ap->a_vp; 3101 accmode_t accmode = ap->a_accmode; 3102 struct vattr vattr; 3103 int error; 3104 3105 /* 3106 * Disallow write attempts on filesystems mounted read-only; 3107 * unless the file is a socket, fifo, or a block or character 3108 * device resident on the filesystem. 3109 */ 3110 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3111 switch (vp->v_type) { 3112 case VREG: 3113 case VDIR: 3114 case VLNK: 3115 return (EROFS); 3116 default: 3117 break; 3118 } 3119 } 3120 vap = &vattr; 3121 error = VOP_GETATTR(vp, vap, cred); 3122 if (error) 3123 goto out; 3124 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid, 3125 accmode, cred, NULL); 3126out: 3127 return error; 3128} 3129 3130/* 3131 * Read wrapper for fifos. 3132 */ 3133static int 3134nfsfifo_read(struct vop_read_args *ap) 3135{ 3136 struct nfsnode *np = VTONFS(ap->a_vp); 3137 int error; 3138 3139 /* 3140 * Set access flag. 3141 */ 3142 mtx_lock(&np->n_mtx); 3143 np->n_flag |= NACC; 3144 getnanotime(&np->n_atim); 3145 mtx_unlock(&np->n_mtx); 3146 error = fifo_specops.vop_read(ap); 3147 return error; 3148} 3149 3150/* 3151 * Write wrapper for fifos. 3152 */ 3153static int 3154nfsfifo_write(struct vop_write_args *ap) 3155{ 3156 struct nfsnode *np = VTONFS(ap->a_vp); 3157 3158 /* 3159 * Set update flag. 3160 */ 3161 mtx_lock(&np->n_mtx); 3162 np->n_flag |= NUPD; 3163 getnanotime(&np->n_mtim); 3164 mtx_unlock(&np->n_mtx); 3165 return(fifo_specops.vop_write(ap)); 3166} 3167 3168/* 3169 * Close wrapper for fifos. 3170 * 3171 * Update the times on the nfsnode then do fifo close. 3172 */ 3173static int 3174nfsfifo_close(struct vop_close_args *ap) 3175{ 3176 struct vnode *vp = ap->a_vp; 3177 struct nfsnode *np = VTONFS(vp); 3178 struct vattr vattr; 3179 struct timespec ts; 3180 3181 mtx_lock(&np->n_mtx); 3182 if (np->n_flag & (NACC | NUPD)) { 3183 getnanotime(&ts); 3184 if (np->n_flag & NACC) 3185 np->n_atim = ts; 3186 if (np->n_flag & NUPD) 3187 np->n_mtim = ts; 3188 np->n_flag |= NCHG; 3189 if (vrefcnt(vp) == 1 && 3190 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3191 VATTR_NULL(&vattr); 3192 if (np->n_flag & NACC) 3193 vattr.va_atime = np->n_atim; 3194 if (np->n_flag & NUPD) 3195 vattr.va_mtime = np->n_mtim; 3196 mtx_unlock(&np->n_mtx); 3197 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3198 goto out; 3199 } 3200 } 3201 mtx_unlock(&np->n_mtx); 3202out: 3203 return (fifo_specops.vop_close(ap)); 3204} 3205 3206/* 3207 * Just call ncl_writebp() with the force argument set to 1. 3208 * 3209 * NOTE: B_DONE may or may not be set in a_bp on call. 3210 */ 3211static int 3212nfs_bwrite(struct buf *bp) 3213{ 3214 3215 return (ncl_writebp(bp, 1, curthread)); 3216} 3217 3218struct buf_ops buf_ops_newnfs = { 3219 .bop_name = "buf_ops_nfs", 3220 .bop_write = nfs_bwrite, 3221 .bop_strategy = bufstrategy, 3222 .bop_sync = bufsync, 3223 .bop_bdflush = bufbdflush, 3224}; 3225 3226/* 3227 * Cloned from vop_stdlock(), and then the ugly hack added. 3228 */ 3229static int 3230nfs_lock1(struct vop_lock1_args *ap) 3231{ 3232 struct vnode *vp = ap->a_vp; 3233 int error = 0; 3234 3235 /* 3236 * Since vfs_hash_get() calls vget() and it will no longer work 3237 * for FreeBSD8 with flags == 0, I can only think of this horrible 3238 * hack to work around it. I call vfs_hash_get() with LK_EXCLOTHER 3239 * and then handle it here. All I want for this case is a v_usecount 3240 * on the vnode to use for recovery, while another thread might 3241 * hold a lock on the vnode. I have the other threads blocked, so 3242 * there isn't any race problem. 3243 */ 3244 if ((ap->a_flags & LK_TYPE_MASK) == LK_EXCLOTHER) { 3245 if ((ap->a_flags & LK_INTERLOCK) == 0) 3246 panic("ncllock1"); 3247 if ((vp->v_iflag & VI_DOOMED)) 3248 error = ENOENT; 3249 VI_UNLOCK(vp); 3250 return (error); 3251 } 3252 return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp), 3253 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file, 3254 ap->a_line)); 3255} 3256 3257static int 3258nfs_getacl(struct vop_getacl_args *ap) 3259{ 3260 int error; 3261 3262 if (ap->a_type != ACL_TYPE_NFS4) 3263 return (EOPNOTSUPP); 3264 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3265 NULL); 3266 if (error > NFSERR_STALE) { 3267 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3268 error = EPERM; 3269 } 3270 return (error); 3271} 3272 3273static int 3274nfs_setacl(struct vop_setacl_args *ap) 3275{ 3276 int error; 3277 3278 if (ap->a_type != ACL_TYPE_NFS4) 3279 return (EOPNOTSUPP); 3280 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3281 NULL); 3282 if (error > NFSERR_STALE) { 3283 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3284 error = EPERM; 3285 } 3286 return (error); 3287} 3288 3289/* 3290 * Return POSIX pathconf information applicable to nfs filesystems. 3291 */ 3292static int 3293nfs_pathconf(struct vop_pathconf_args *ap) 3294{ 3295 struct nfsv3_pathconf pc; 3296 struct nfsvattr nfsva; 3297 struct vnode *vp = ap->a_vp; 3298 struct thread *td = curthread; 3299 int attrflag, error; 3300 3301 if (NFS_ISV34(vp)) { 3302 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva, 3303 &attrflag, NULL); 3304 if (attrflag != 0) 3305 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 3306 1); 3307 if (error != 0) 3308 return (error); 3309 } else { 3310 /* For NFSv2, just fake them. */ 3311 pc.pc_linkmax = LINK_MAX; 3312 pc.pc_namemax = NFS_MAXNAMLEN; 3313 pc.pc_notrunc = 1; 3314 pc.pc_chownrestricted = 1; 3315 pc.pc_caseinsensitive = 0; 3316 pc.pc_casepreserving = 1; 3317 error = 0; 3318 } 3319 switch (ap->a_name) { 3320 case _PC_LINK_MAX: 3321 *ap->a_retval = pc.pc_linkmax; 3322 break; 3323 case _PC_NAME_MAX: 3324 *ap->a_retval = pc.pc_namemax; 3325 break; 3326 case _PC_PATH_MAX: 3327 *ap->a_retval = PATH_MAX; 3328 break; 3329 case _PC_PIPE_BUF: 3330 *ap->a_retval = PIPE_BUF; 3331 break; 3332 case _PC_CHOWN_RESTRICTED: 3333 *ap->a_retval = pc.pc_chownrestricted; 3334 break; 3335 case _PC_NO_TRUNC: 3336 *ap->a_retval = pc.pc_notrunc; 3337 break; 3338 case _PC_ACL_EXTENDED: 3339 *ap->a_retval = 0; 3340 break; 3341 case _PC_ACL_NFS4: 3342 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 && 3343 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL)) 3344 *ap->a_retval = 1; 3345 else 3346 *ap->a_retval = 0; 3347 break; 3348 case _PC_ACL_PATH_MAX: 3349 if (NFS_ISV4(vp)) 3350 *ap->a_retval = ACL_MAX_ENTRIES; 3351 else 3352 *ap->a_retval = 3; 3353 break; 3354 case _PC_MAC_PRESENT: 3355 *ap->a_retval = 0; 3356 break; 3357 case _PC_ASYNC_IO: 3358 /* _PC_ASYNC_IO should have been handled by upper layers. */ 3359 KASSERT(0, ("_PC_ASYNC_IO should not get here")); 3360 error = EINVAL; 3361 break; 3362 case _PC_PRIO_IO: 3363 *ap->a_retval = 0; 3364 break; 3365 case _PC_SYNC_IO: 3366 *ap->a_retval = 0; 3367 break; 3368 case _PC_ALLOC_SIZE_MIN: 3369 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize; 3370 break; 3371 case _PC_FILESIZEBITS: 3372 if (NFS_ISV34(vp)) 3373 *ap->a_retval = 64; 3374 else 3375 *ap->a_retval = 32; 3376 break; 3377 case _PC_REC_INCR_XFER_SIZE: 3378 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 3379 break; 3380 case _PC_REC_MAX_XFER_SIZE: 3381 *ap->a_retval = -1; /* means ``unlimited'' */ 3382 break; 3383 case _PC_REC_MIN_XFER_SIZE: 3384 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 3385 break; 3386 case _PC_REC_XFER_ALIGN: 3387 *ap->a_retval = PAGE_SIZE; 3388 break; 3389 case _PC_SYMLINK_MAX: 3390 *ap->a_retval = NFS_MAXPATHLEN; 3391 break; 3392 3393 default: 3394 error = EINVAL; 3395 break; 3396 } 3397 return (error); 3398} 3399 3400