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 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by the University of 19 * California, Berkeley and its contributors. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)nfs_vnops.c 8.16 (Berkeley) 5/27/95 37 * $FreeBSD: src/sys/nfs/nfs_vnops.c,v 1.150.2.5 2001/12/20 19:56:28 dillon Exp $ 38 * $DragonFly: src/sys/vfs/nfs/nfs_vnops.c,v 1.34 2004/10/07 10:03:03 dillon Exp $ 39 */ 40 41 42 /* 43 * vnode op calls for Sun NFS version 2 and 3 44 */ 45 46 #include "opt_inet.h" 47 48 #include <sys/param.h> 49 #include <sys/kernel.h> 50 #include <sys/systm.h> 51 #include <sys/resourcevar.h> 52 #include <sys/proc.h> 53 #include <sys/mount.h> 54 #include <sys/buf.h> 55 #include <sys/malloc.h> 56 #include <sys/mbuf.h> 57 #include <sys/namei.h> 58 #include <sys/socket.h> 59 #include <sys/vnode.h> 60 #include <sys/dirent.h> 61 #include <sys/fcntl.h> 62 #include <sys/lockf.h> 63 #include <sys/stat.h> 64 #include <sys/sysctl.h> 65 #include <sys/conf.h> 66 67 #include <vm/vm.h> 68 #include <vm/vm_extern.h> 69 #include <vm/vm_zone.h> 70 71 #include <sys/buf2.h> 72 73 #include <vfs/fifofs/fifo.h> 74 75 #include "rpcv2.h" 76 #include "nfsproto.h" 77 #include "nfs.h" 78 #include "nfsmount.h" 79 #include "nfsnode.h" 80 #include "xdr_subs.h" 81 #include "nfsm_subs.h" 82 #include "nqnfs.h" 83 84 #include <net/if.h> 85 #include <netinet/in.h> 86 #include <netinet/in_var.h> 87 88 /* Defs */ 89 #define TRUE 1 90 #define FALSE 0 91 92 /* 93 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these 94 * calls are not in getblk() and brelse() so that they would not be necessary 95 * here. 96 */ 97 #ifndef B_VMIO 98 #define vfs_busy_pages(bp, f) 99 #endif 100 101 static int nfsspec_read (struct vop_read_args *); 102 static int nfsspec_write (struct vop_write_args *); 103 static int nfsfifo_read (struct vop_read_args *); 104 static int nfsfifo_write (struct vop_write_args *); 105 static int nfsspec_close (struct vop_close_args *); 106 static int nfsfifo_close (struct vop_close_args *); 107 #define nfs_poll vop_nopoll 108 static int nfs_flush (struct vnode *,int,struct thread *,int); 109 static int nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *); 110 static int nfs_lookup (struct vop_lookup_args *); 111 static int nfs_create (struct vop_create_args *); 112 static int nfs_mknod (struct vop_mknod_args *); 113 static int nfs_open (struct vop_open_args *); 114 static int nfs_close (struct vop_close_args *); 115 static int nfs_access (struct vop_access_args *); 116 static int nfs_getattr (struct vop_getattr_args *); 117 static int nfs_setattr (struct vop_setattr_args *); 118 static int nfs_read (struct vop_read_args *); 119 static int nfs_mmap (struct vop_mmap_args *); 120 static int nfs_fsync (struct vop_fsync_args *); 121 static int nfs_remove (struct vop_remove_args *); 122 static int nfs_link (struct vop_link_args *); 123 static int nfs_rename (struct vop_rename_args *); 124 static int nfs_mkdir (struct vop_mkdir_args *); 125 static int nfs_rmdir (struct vop_rmdir_args *); 126 static int nfs_symlink (struct vop_symlink_args *); 127 static int nfs_readdir (struct vop_readdir_args *); 128 static int nfs_bmap (struct vop_bmap_args *); 129 static int nfs_strategy (struct vop_strategy_args *); 130 static int nfs_lookitup (struct vnode *, const char *, int, 131 struct ucred *, struct thread *, struct nfsnode **); 132 static int nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *); 133 static int nfsspec_access (struct vop_access_args *); 134 static int nfs_readlink (struct vop_readlink_args *); 135 static int nfs_print (struct vop_print_args *); 136 static int nfs_advlock (struct vop_advlock_args *); 137 static int nfs_bwrite (struct vop_bwrite_args *); 138 /* 139 * Global vfs data structures for nfs 140 */ 141 struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = { 142 { &vop_default_desc, vop_defaultop }, 143 { &vop_access_desc, (void *) nfs_access }, 144 { &vop_advlock_desc, (void *) nfs_advlock }, 145 { &vop_bmap_desc, (void *) nfs_bmap }, 146 { &vop_bwrite_desc, (void *) nfs_bwrite }, 147 { &vop_close_desc, (void *) nfs_close }, 148 { &vop_create_desc, (void *) nfs_create }, 149 { &vop_fsync_desc, (void *) nfs_fsync }, 150 { &vop_getattr_desc, (void *) nfs_getattr }, 151 { &vop_getpages_desc, (void *) nfs_getpages }, 152 { &vop_putpages_desc, (void *) nfs_putpages }, 153 { &vop_inactive_desc, (void *) nfs_inactive }, 154 { &vop_islocked_desc, (void *) vop_stdislocked }, 155 { &vop_lease_desc, vop_null }, 156 { &vop_link_desc, (void *) nfs_link }, 157 { &vop_lock_desc, (void *) vop_stdlock }, 158 { &vop_lookup_desc, (void *) nfs_lookup }, 159 { &vop_mkdir_desc, (void *) nfs_mkdir }, 160 { &vop_mknod_desc, (void *) nfs_mknod }, 161 { &vop_mmap_desc, (void *) nfs_mmap }, 162 { &vop_open_desc, (void *) nfs_open }, 163 { &vop_poll_desc, (void *) nfs_poll }, 164 { &vop_print_desc, (void *) nfs_print }, 165 { &vop_read_desc, (void *) nfs_read }, 166 { &vop_readdir_desc, (void *) nfs_readdir }, 167 { &vop_readlink_desc, (void *) nfs_readlink }, 168 { &vop_reclaim_desc, (void *) nfs_reclaim }, 169 { &vop_remove_desc, (void *) nfs_remove }, 170 { &vop_rename_desc, (void *) nfs_rename }, 171 { &vop_rmdir_desc, (void *) nfs_rmdir }, 172 { &vop_setattr_desc, (void *) nfs_setattr }, 173 { &vop_strategy_desc, (void *) nfs_strategy }, 174 { &vop_symlink_desc, (void *) nfs_symlink }, 175 { &vop_unlock_desc, (void *) vop_stdunlock }, 176 { &vop_write_desc, (void *) nfs_write }, 177 { NULL, NULL } 178 }; 179 180 /* 181 * Special device vnode ops 182 */ 183 struct vnodeopv_entry_desc nfsv2_specop_entries[] = { 184 { &vop_default_desc, (void *) spec_vnoperate }, 185 { &vop_access_desc, (void *) nfsspec_access }, 186 { &vop_close_desc, (void *) nfsspec_close }, 187 { &vop_fsync_desc, (void *) nfs_fsync }, 188 { &vop_getattr_desc, (void *) nfs_getattr }, 189 { &vop_inactive_desc, (void *) nfs_inactive }, 190 { &vop_islocked_desc, (void *) vop_stdislocked }, 191 { &vop_lock_desc, (void *) vop_stdlock }, 192 { &vop_print_desc, (void *) nfs_print }, 193 { &vop_read_desc, (void *) nfsspec_read }, 194 { &vop_reclaim_desc, (void *) nfs_reclaim }, 195 { &vop_setattr_desc, (void *) nfs_setattr }, 196 { &vop_unlock_desc, (void *) vop_stdunlock }, 197 { &vop_write_desc, (void *) nfsspec_write }, 198 { NULL, NULL } 199 }; 200 201 struct vnodeopv_entry_desc nfsv2_fifoop_entries[] = { 202 { &vop_default_desc, (void *) fifo_vnoperate }, 203 { &vop_access_desc, (void *) nfsspec_access }, 204 { &vop_close_desc, (void *) nfsfifo_close }, 205 { &vop_fsync_desc, (void *) nfs_fsync }, 206 { &vop_getattr_desc, (void *) nfs_getattr }, 207 { &vop_inactive_desc, (void *) nfs_inactive }, 208 { &vop_islocked_desc, (void *) vop_stdislocked }, 209 { &vop_lock_desc, (void *) vop_stdlock }, 210 { &vop_print_desc, (void *) nfs_print }, 211 { &vop_read_desc, (void *) nfsfifo_read }, 212 { &vop_reclaim_desc, (void *) nfs_reclaim }, 213 { &vop_setattr_desc, (void *) nfs_setattr }, 214 { &vop_unlock_desc, (void *) vop_stdunlock }, 215 { &vop_write_desc, (void *) nfsfifo_write }, 216 { NULL, NULL } 217 }; 218 219 static int nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp, 220 struct componentname *cnp, 221 struct vattr *vap); 222 static int nfs_removerpc (struct vnode *dvp, const char *name, 223 int namelen, 224 struct ucred *cred, struct thread *td); 225 static int nfs_renamerpc (struct vnode *fdvp, const char *fnameptr, 226 int fnamelen, struct vnode *tdvp, 227 const char *tnameptr, int tnamelen, 228 struct ucred *cred, struct thread *td); 229 static int nfs_renameit (struct vnode *sdvp, 230 struct componentname *scnp, 231 struct sillyrename *sp); 232 233 /* 234 * Global variables 235 */ 236 extern u_int32_t nfs_true, nfs_false; 237 extern u_int32_t nfs_xdrneg1; 238 extern struct nfsstats nfsstats; 239 extern nfstype nfsv3_type[9]; 240 struct thread *nfs_iodwant[NFS_MAXASYNCDAEMON]; 241 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON]; 242 int nfs_numasync = 0; 243 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1)) 244 245 SYSCTL_DECL(_vfs_nfs); 246 247 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO; 248 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 249 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 250 251 static int nfsneg_cache_timeout = NFS_MINATTRTIMO; 252 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW, 253 &nfsneg_cache_timeout, 0, "NFS NEGATIVE ACCESS cache timeout"); 254 255 static int nfsv3_commit_on_close = 0; 256 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW, 257 &nfsv3_commit_on_close, 0, "write+commit on close, else only write"); 258 #if 0 259 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD, 260 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count"); 261 262 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD, 263 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count"); 264 #endif 265 266 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \ 267 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \ 268 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP) 269 static int 270 nfs3_access_otw(struct vnode *vp, int wmode, 271 struct thread *td, struct ucred *cred) 272 { 273 const int v3 = 1; 274 u_int32_t *tl; 275 int error = 0, attrflag; 276 277 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 278 caddr_t bpos, dpos, cp2; 279 int32_t t1, t2; 280 caddr_t cp; 281 u_int32_t rmode; 282 struct nfsnode *np = VTONFS(vp); 283 284 nfsstats.rpccnt[NFSPROC_ACCESS]++; 285 nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 286 nfsm_fhtom(vp, v3); 287 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 288 *tl = txdr_unsigned(wmode); 289 nfsm_request(vp, NFSPROC_ACCESS, td, cred); 290 nfsm_postop_attr(vp, attrflag); 291 if (!error) { 292 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 293 rmode = fxdr_unsigned(u_int32_t, *tl); 294 np->n_mode = rmode; 295 np->n_modeuid = cred->cr_uid; 296 np->n_modestamp = mycpu->gd_time_seconds; 297 } 298 m_freem(mrep); 299 nfsmout: 300 return error; 301 } 302 303 /* 304 * nfs access vnode op. 305 * For nfs version 2, just return ok. File accesses may fail later. 306 * For nfs version 3, use the access rpc to check accessibility. If file modes 307 * are changed on the server, accesses might still fail later. 308 * 309 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 310 * struct thread *a_td) 311 */ 312 static int 313 nfs_access(struct vop_access_args *ap) 314 { 315 struct vnode *vp = ap->a_vp; 316 int error = 0; 317 u_int32_t mode, wmode; 318 int v3 = NFS_ISV3(vp); 319 struct nfsnode *np = VTONFS(vp); 320 321 /* 322 * Disallow write attempts on filesystems mounted read-only; 323 * unless the file is a socket, fifo, or a block or character 324 * device resident on the filesystem. 325 */ 326 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 327 switch (vp->v_type) { 328 case VREG: 329 case VDIR: 330 case VLNK: 331 return (EROFS); 332 default: 333 break; 334 } 335 } 336 /* 337 * For nfs v3, check to see if we have done this recently, and if 338 * so return our cached result instead of making an ACCESS call. 339 * If not, do an access rpc, otherwise you are stuck emulating 340 * ufs_access() locally using the vattr. This may not be correct, 341 * since the server may apply other access criteria such as 342 * client uid-->server uid mapping that we do not know about. 343 */ 344 if (v3) { 345 if (ap->a_mode & VREAD) 346 mode = NFSV3ACCESS_READ; 347 else 348 mode = 0; 349 if (vp->v_type != VDIR) { 350 if (ap->a_mode & VWRITE) 351 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 352 if (ap->a_mode & VEXEC) 353 mode |= NFSV3ACCESS_EXECUTE; 354 } else { 355 if (ap->a_mode & VWRITE) 356 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 357 NFSV3ACCESS_DELETE); 358 if (ap->a_mode & VEXEC) 359 mode |= NFSV3ACCESS_LOOKUP; 360 } 361 /* XXX safety belt, only make blanket request if caching */ 362 if (nfsaccess_cache_timeout > 0) { 363 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY | 364 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE | 365 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP; 366 } else { 367 wmode = mode; 368 } 369 370 /* 371 * Does our cached result allow us to give a definite yes to 372 * this request? 373 */ 374 if ((mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) && 375 (ap->a_cred->cr_uid == np->n_modeuid) && 376 ((np->n_mode & mode) == mode)) { 377 nfsstats.accesscache_hits++; 378 } else { 379 /* 380 * Either a no, or a don't know. Go to the wire. 381 */ 382 nfsstats.accesscache_misses++; 383 error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred); 384 if (!error) { 385 if ((np->n_mode & mode) != mode) { 386 error = EACCES; 387 } 388 } 389 } 390 } else { 391 if ((error = nfsspec_access(ap)) != 0) 392 return (error); 393 394 /* 395 * Attempt to prevent a mapped root from accessing a file 396 * which it shouldn't. We try to read a byte from the file 397 * if the user is root and the file is not zero length. 398 * After calling nfsspec_access, we should have the correct 399 * file size cached. 400 */ 401 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD) 402 && VTONFS(vp)->n_size > 0) { 403 struct iovec aiov; 404 struct uio auio; 405 char buf[1]; 406 407 aiov.iov_base = buf; 408 aiov.iov_len = 1; 409 auio.uio_iov = &aiov; 410 auio.uio_iovcnt = 1; 411 auio.uio_offset = 0; 412 auio.uio_resid = 1; 413 auio.uio_segflg = UIO_SYSSPACE; 414 auio.uio_rw = UIO_READ; 415 auio.uio_td = ap->a_td; 416 417 if (vp->v_type == VREG) { 418 error = nfs_readrpc(vp, &auio); 419 } else if (vp->v_type == VDIR) { 420 char* bp; 421 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 422 aiov.iov_base = bp; 423 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 424 error = nfs_readdirrpc(vp, &auio); 425 free(bp, M_TEMP); 426 } else if (vp->v_type == VLNK) { 427 error = nfs_readlinkrpc(vp, &auio); 428 } else { 429 error = EACCES; 430 } 431 } 432 } 433 /* 434 * [re]record creds for reading and/or writing if access 435 * was granted. Assume the NFS server will grant read access 436 * for execute requests. 437 */ 438 if (error == 0) { 439 if ((ap->a_mode & (VREAD|VEXEC)) && ap->a_cred != np->n_rucred) { 440 crhold(ap->a_cred); 441 if (np->n_rucred) 442 crfree(np->n_rucred); 443 np->n_rucred = ap->a_cred; 444 } 445 if ((ap->a_mode & VWRITE) && ap->a_cred != np->n_wucred) { 446 crhold(ap->a_cred); 447 if (np->n_wucred) 448 crfree(np->n_wucred); 449 np->n_wucred = ap->a_cred; 450 } 451 } 452 return(error); 453 } 454 455 /* 456 * nfs open vnode op 457 * Check to see if the type is ok 458 * and that deletion is not in progress. 459 * For paged in text files, you will need to flush the page cache 460 * if consistency is lost. 461 * 462 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 463 * struct thread *a_td) 464 */ 465 /* ARGSUSED */ 466 static int 467 nfs_open(struct vop_open_args *ap) 468 { 469 struct vnode *vp = ap->a_vp; 470 struct nfsnode *np = VTONFS(vp); 471 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 472 struct vattr vattr; 473 int error; 474 475 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 476 #ifdef DIAGNOSTIC 477 printf("open eacces vtyp=%d\n",vp->v_type); 478 #endif 479 return (EOPNOTSUPP); 480 } 481 /* 482 * Get a valid lease. If cached data is stale, flush it. 483 */ 484 if (nmp->nm_flag & NFSMNT_NQNFS) { 485 if (NQNFS_CKINVALID(vp, np, ND_READ)) { 486 do { 487 error = nqnfs_getlease(vp, ND_READ, ap->a_td); 488 } while (error == NQNFS_EXPIRED); 489 if (error) 490 return (error); 491 if (np->n_lrev != np->n_brev || 492 (np->n_flag & NQNFSNONCACHE)) { 493 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) 494 == EINTR) { 495 return (error); 496 } 497 np->n_brev = np->n_lrev; 498 } 499 } 500 } else { 501 if (np->n_flag & NMODIFIED) { 502 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) 503 == EINTR) { 504 return (error); 505 } 506 np->n_attrstamp = 0; 507 if (vp->v_type == VDIR) 508 np->n_direofoffset = 0; 509 error = VOP_GETATTR(vp, &vattr, ap->a_td); 510 if (error) 511 return (error); 512 np->n_mtime = vattr.va_mtime.tv_sec; 513 } else { 514 error = VOP_GETATTR(vp, &vattr, ap->a_td); 515 if (error) 516 return (error); 517 if (np->n_mtime != vattr.va_mtime.tv_sec) { 518 if (vp->v_type == VDIR) 519 np->n_direofoffset = 0; 520 if ((error = nfs_vinvalbuf(vp, V_SAVE, 521 ap->a_td, 1)) == EINTR) { 522 return (error); 523 } 524 np->n_mtime = vattr.va_mtime.tv_sec; 525 } 526 } 527 } 528 if ((nmp->nm_flag & NFSMNT_NQNFS) == 0) 529 np->n_attrstamp = 0; /* For Open/Close consistency */ 530 return (0); 531 } 532 533 /* 534 * nfs close vnode op 535 * What an NFS client should do upon close after writing is a debatable issue. 536 * Most NFS clients push delayed writes to the server upon close, basically for 537 * two reasons: 538 * 1 - So that any write errors may be reported back to the client process 539 * doing the close system call. By far the two most likely errors are 540 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 541 * 2 - To put a worst case upper bound on cache inconsistency between 542 * multiple clients for the file. 543 * There is also a consistency problem for Version 2 of the protocol w.r.t. 544 * not being able to tell if other clients are writing a file concurrently, 545 * since there is no way of knowing if the changed modify time in the reply 546 * is only due to the write for this client. 547 * (NFS Version 3 provides weak cache consistency data in the reply that 548 * should be sufficient to detect and handle this case.) 549 * 550 * The current code does the following: 551 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 552 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 553 * or commit them (this satisfies 1 and 2 except for the 554 * case where the server crashes after this close but 555 * before the commit RPC, which is felt to be "good 556 * enough". Changing the last argument to nfs_flush() to 557 * a 1 would force a commit operation, if it is felt a 558 * commit is necessary now. 559 * for NQNFS - do nothing now, since 2 is dealt with via leases and 560 * 1 should be dealt with via an fsync() system call for 561 * cases where write errors are important. 562 * 563 * nfs_close(struct vnodeop_desc *a_desc, struct vnode *a_vp, int a_fflag, 564 * struct ucred *a_cred, struct thread *a_td) 565 */ 566 /* ARGSUSED */ 567 static int 568 nfs_close(struct vop_close_args *ap) 569 { 570 struct vnode *vp = ap->a_vp; 571 struct nfsnode *np = VTONFS(vp); 572 int error = 0; 573 574 if (vp->v_type == VREG) { 575 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 && 576 (np->n_flag & NMODIFIED)) { 577 if (NFS_ISV3(vp)) { 578 /* 579 * Under NFSv3 we have dirty buffers to dispose of. We 580 * must flush them to the NFS server. We have the option 581 * of waiting all the way through the commit rpc or just 582 * waiting for the initial write. The default is to only 583 * wait through the initial write so the data is in the 584 * server's cache, which is roughly similar to the state 585 * a standard disk subsystem leaves the file in on close(). 586 * 587 * We cannot clear the NMODIFIED bit in np->n_flag due to 588 * potential races with other processes, and certainly 589 * cannot clear it if we don't commit. 590 */ 591 int cm = nfsv3_commit_on_close ? 1 : 0; 592 error = nfs_flush(vp, MNT_WAIT, ap->a_td, cm); 593 /* np->n_flag &= ~NMODIFIED; */ 594 } else { 595 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 596 } 597 np->n_attrstamp = 0; 598 } 599 if (np->n_flag & NWRITEERR) { 600 np->n_flag &= ~NWRITEERR; 601 error = np->n_error; 602 } 603 } 604 return (error); 605 } 606 607 /* 608 * nfs getattr call from vfs. 609 * 610 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred, 611 * struct thread *a_td) 612 */ 613 static int 614 nfs_getattr(struct vop_getattr_args *ap) 615 { 616 struct vnode *vp = ap->a_vp; 617 struct nfsnode *np = VTONFS(vp); 618 caddr_t cp; 619 u_int32_t *tl; 620 int32_t t1, t2; 621 caddr_t bpos, dpos; 622 int error = 0; 623 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 624 int v3 = NFS_ISV3(vp); 625 626 /* 627 * Update local times for special files. 628 */ 629 if (np->n_flag & (NACC | NUPD)) 630 np->n_flag |= NCHG; 631 /* 632 * First look in the cache. 633 */ 634 if (nfs_getattrcache(vp, ap->a_vap) == 0) 635 return (0); 636 637 if (v3 && nfsaccess_cache_timeout > 0) { 638 nfsstats.accesscache_misses++; 639 nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, nfs_vpcred(vp, ND_CHECK)); 640 if (nfs_getattrcache(vp, ap->a_vap) == 0) 641 return (0); 642 } 643 644 nfsstats.rpccnt[NFSPROC_GETATTR]++; 645 nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3)); 646 nfsm_fhtom(vp, v3); 647 nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, nfs_vpcred(vp, ND_CHECK)); 648 if (!error) { 649 nfsm_loadattr(vp, ap->a_vap); 650 } 651 m_freem(mrep); 652 nfsmout: 653 return (error); 654 } 655 656 /* 657 * nfs setattr call. 658 * 659 * nfs_setattr(struct vnodeop_desc *a_desc, struct vnode *a_vp, 660 * struct vattr *a_vap, struct ucred *a_cred, 661 * struct thread *a_td) 662 */ 663 static int 664 nfs_setattr(struct vop_setattr_args *ap) 665 { 666 struct vnode *vp = ap->a_vp; 667 struct nfsnode *np = VTONFS(vp); 668 struct vattr *vap = ap->a_vap; 669 int error = 0; 670 u_quad_t tsize; 671 672 #ifndef nolint 673 tsize = (u_quad_t)0; 674 #endif 675 676 /* 677 * Setting of flags is not supported. 678 */ 679 if (vap->va_flags != VNOVAL) 680 return (EOPNOTSUPP); 681 682 /* 683 * Disallow write attempts if the filesystem is mounted read-only. 684 */ 685 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 686 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 687 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 688 (vp->v_mount->mnt_flag & MNT_RDONLY)) 689 return (EROFS); 690 if (vap->va_size != VNOVAL) { 691 switch (vp->v_type) { 692 case VDIR: 693 return (EISDIR); 694 case VCHR: 695 case VBLK: 696 case VSOCK: 697 case VFIFO: 698 if (vap->va_mtime.tv_sec == VNOVAL && 699 vap->va_atime.tv_sec == VNOVAL && 700 vap->va_mode == (mode_t)VNOVAL && 701 vap->va_uid == (uid_t)VNOVAL && 702 vap->va_gid == (gid_t)VNOVAL) 703 return (0); 704 vap->va_size = VNOVAL; 705 break; 706 default: 707 /* 708 * Disallow write attempts if the filesystem is 709 * mounted read-only. 710 */ 711 if (vp->v_mount->mnt_flag & MNT_RDONLY) 712 return (EROFS); 713 714 /* 715 * We run vnode_pager_setsize() early (why?), 716 * we must set np->n_size now to avoid vinvalbuf 717 * V_SAVE races that might setsize a lower 718 * value. 719 */ 720 721 tsize = np->n_size; 722 error = nfs_meta_setsize(vp, ap->a_td, vap->va_size); 723 724 if (np->n_flag & NMODIFIED) { 725 if (vap->va_size == 0) 726 error = nfs_vinvalbuf(vp, 0, ap->a_td, 1); 727 else 728 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 729 if (error) { 730 np->n_size = tsize; 731 vnode_pager_setsize(vp, np->n_size); 732 return (error); 733 } 734 } 735 /* 736 * np->n_size has already been set to vap->va_size 737 * in nfs_meta_setsize(). We must set it again since 738 * nfs_loadattrcache() could be called through 739 * nfs_meta_setsize() and could modify np->n_size. 740 * 741 * (note that nfs_loadattrcache() will have called 742 * vnode_pager_setsize() for us in that case). 743 */ 744 np->n_vattr.va_size = np->n_size = vap->va_size; 745 }; 746 } else if ((vap->va_mtime.tv_sec != VNOVAL || 747 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NMODIFIED) && 748 vp->v_type == VREG && 749 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) == EINTR) 750 return (error); 751 error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td); 752 if (error && vap->va_size != VNOVAL) { 753 np->n_size = np->n_vattr.va_size = tsize; 754 vnode_pager_setsize(vp, np->n_size); 755 } 756 return (error); 757 } 758 759 /* 760 * Do an nfs setattr rpc. 761 */ 762 static int 763 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, 764 struct ucred *cred, struct thread *td) 765 { 766 struct nfsv2_sattr *sp; 767 struct nfsnode *np = VTONFS(vp); 768 caddr_t cp; 769 int32_t t1, t2; 770 caddr_t bpos, dpos, cp2; 771 u_int32_t *tl; 772 int error = 0, wccflag = NFSV3_WCCRATTR; 773 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 774 int v3 = NFS_ISV3(vp); 775 776 nfsstats.rpccnt[NFSPROC_SETATTR]++; 777 nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 778 nfsm_fhtom(vp, v3); 779 if (v3) { 780 nfsm_v3attrbuild(vap, TRUE); 781 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 782 *tl = nfs_false; 783 } else { 784 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 785 if (vap->va_mode == (mode_t)VNOVAL) 786 sp->sa_mode = nfs_xdrneg1; 787 else 788 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 789 if (vap->va_uid == (uid_t)VNOVAL) 790 sp->sa_uid = nfs_xdrneg1; 791 else 792 sp->sa_uid = txdr_unsigned(vap->va_uid); 793 if (vap->va_gid == (gid_t)VNOVAL) 794 sp->sa_gid = nfs_xdrneg1; 795 else 796 sp->sa_gid = txdr_unsigned(vap->va_gid); 797 sp->sa_size = txdr_unsigned(vap->va_size); 798 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 799 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 800 } 801 nfsm_request(vp, NFSPROC_SETATTR, td, cred); 802 if (v3) { 803 np->n_modestamp = 0; 804 nfsm_wcc_data(vp, wccflag); 805 } else 806 nfsm_loadattr(vp, (struct vattr *)0); 807 m_freem(mrep); 808 nfsmout: 809 return (error); 810 } 811 812 /* 813 * 'cached' nfs directory lookup 814 * 815 * nfs_lookup(struct vnodeop_desc *a_desc, struct vnode *a_dvp, 816 * struct vnode **a_vpp, struct componentname *a_cnp) 817 */ 818 static int 819 nfs_lookup(struct vop_lookup_args *ap) 820 { 821 struct componentname *cnp = ap->a_cnp; 822 struct vnode *dvp = ap->a_dvp; 823 struct vnode **vpp = ap->a_vpp; 824 int flags = cnp->cn_flags; 825 struct vnode *newvp; 826 u_int32_t *tl; 827 caddr_t cp; 828 int32_t t1, t2; 829 struct nfsmount *nmp; 830 caddr_t bpos, dpos, cp2; 831 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 832 long len; 833 nfsfh_t *fhp; 834 struct nfsnode *np; 835 int lockparent, wantparent, error = 0, attrflag, fhsize; 836 int v3 = NFS_ISV3(dvp); 837 struct thread *td = cnp->cn_td; 838 839 /* 840 * Read-only mount check and directory check. 841 */ 842 *vpp = NULLVP; 843 if ((flags & CNP_ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 844 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)) 845 return (EROFS); 846 847 if (dvp->v_type != VDIR) 848 return (ENOTDIR); 849 850 /* 851 * Look it up in the cache. Note that ENOENT is only returned if we 852 * previously entered a negative hit (see later on). The additional 853 * nfsneg_cache_timeout check causes previously cached results to 854 * be instantly ignored if the negative caching is turned off. 855 */ 856 lockparent = flags & CNP_LOCKPARENT; 857 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT); 858 nmp = VFSTONFS(dvp->v_mount); 859 np = VTONFS(dvp); 860 error = cache_lookup(dvp, vpp, cnp); 861 if (error != 0) { 862 struct vattr vattr; 863 int vpid; 864 865 if (error == ENOENT) { 866 if (nfsneg_cache_timeout) { 867 *vpp = NULLVP; 868 return (error); 869 } 870 goto miss; 871 } 872 if (error > 0) { 873 printf("nfs_lookup: %*.*s weird error %d\n", 874 (int)cnp->cn_namelen, (int)cnp->cn_namelen, 875 cnp->cn_nameptr, error); 876 *vpp = NULLVP; 877 return (error); 878 } 879 880 /* 881 * At this point we have a cache hit (error should be -1). 882 * The vnode returned in *vpp will be referenced but not 883 * locked. 884 */ 885 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) { 886 vrele(*vpp); 887 *vpp = NULLVP; 888 return (error); 889 } 890 891 newvp = *vpp; 892 vpid = newvp->v_id; 893 /* 894 * See the comment starting `Step through' in ufs/ufs_lookup.c 895 * for an explanation of the locking protocol 896 */ 897 if (dvp == newvp) { 898 /* newvp already ref'd from lookup */ 899 error = 0; 900 } else if (flags & CNP_ISDOTDOT) { 901 VOP_UNLOCK(dvp, NULL, 0, td); 902 cnp->cn_flags |= CNP_PDIRUNLOCK; 903 error = vget(newvp, NULL, LK_EXCLUSIVE, td); 904 vrele(newvp); /* get rid of ref from lookup */ 905 if (!error && lockparent && (flags & CNP_ISLASTCN)) { 906 error = vn_lock(dvp, NULL, LK_EXCLUSIVE, td); 907 if (error == 0) 908 cnp->cn_flags &= ~CNP_PDIRUNLOCK; 909 } 910 } else { 911 error = vget(newvp, NULL, LK_EXCLUSIVE, td); 912 vrele(newvp); /* get rid of ref from lookup */ 913 if (!lockparent || error || !(flags & CNP_ISLASTCN)) { 914 VOP_UNLOCK(dvp, NULL, 0, td); 915 cnp->cn_flags |= CNP_PDIRUNLOCK; 916 } 917 } 918 if (!error) { 919 /* 920 * Attempt to do a better job synchronizing our cache 921 * to the NFS server by checking the vnode against 922 * the nfs-only cache via VOP_GETATTR(). 923 * 924 * WARNING! An old ctime check has been removed. We 925 * can't just willy-nilly purge a directory vnode that 926 * might have children in the new VFS scheme. The 927 * ctime check was bogus anyway. 928 */ 929 if (vpid == newvp->v_id) { 930 if (VOP_GETATTR(newvp, &vattr, td) == 0) { 931 nfsstats.lookupcache_hits++; 932 if (cnp->cn_nameiop != NAMEI_LOOKUP && 933 (flags & CNP_ISLASTCN)) 934 cnp->cn_flags |= CNP_SAVENAME; 935 return (0); 936 } 937 cache_purge(newvp); 938 } 939 vput(newvp); 940 if (lockparent && dvp != newvp && (flags & CNP_ISLASTCN)) { 941 VOP_UNLOCK(dvp, NULL, 0, td); 942 cnp->cn_flags |= CNP_PDIRUNLOCK; 943 } 944 } 945 error = vn_lock(dvp, NULL, LK_EXCLUSIVE, td); 946 if (error == 0) 947 cnp->cn_flags &= ~CNP_PDIRUNLOCK; 948 *vpp = NULLVP; 949 if (error) 950 return (error); 951 } 952 953 miss: 954 /* 955 * Cache miss, go the wire. 956 */ 957 error = 0; 958 newvp = NULLVP; 959 nfsstats.lookupcache_misses++; 960 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 961 len = cnp->cn_namelen; 962 nfsm_reqhead(dvp, NFSPROC_LOOKUP, 963 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 964 nfsm_fhtom(dvp, v3); 965 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 966 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_td, cnp->cn_cred); 967 if (error) { 968 /* 969 * Cache negatve lookups to reduce NFS traffic, but use 970 * a fast timeout. 971 */ 972 if (error == ENOENT && 973 (cnp->cn_flags & CNP_MAKEENTRY) && 974 cnp->cn_nameiop == NAMEI_LOOKUP && 975 nfsneg_cache_timeout) { 976 int toval = nfsneg_cache_timeout * hz; 977 if (cnp->cn_flags & CNP_CACHETIMEOUT) { 978 if (cnp->cn_timeout > toval) 979 cnp->cn_timeout = toval; 980 } else { 981 cnp->cn_flags |= CNP_CACHETIMEOUT; 982 cnp->cn_timeout = toval; 983 } 984 cache_enter(dvp, NULL, cnp); 985 } 986 nfsm_postop_attr(dvp, attrflag); 987 m_freem(mrep); 988 goto nfsmout; 989 } 990 nfsm_getfh(fhp, fhsize, v3); 991 992 /* 993 * Handle RENAME case... 994 */ 995 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent && (flags & CNP_ISLASTCN)) { 996 if (NFS_CMPFH(np, fhp, fhsize)) { 997 m_freem(mrep); 998 return (EISDIR); 999 } 1000 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1001 if (error) { 1002 m_freem(mrep); 1003 return (error); 1004 } 1005 newvp = NFSTOV(np); 1006 if (v3) { 1007 nfsm_postop_attr(newvp, attrflag); 1008 nfsm_postop_attr(dvp, attrflag); 1009 } else 1010 nfsm_loadattr(newvp, (struct vattr *)0); 1011 *vpp = newvp; 1012 m_freem(mrep); 1013 cnp->cn_flags |= CNP_SAVENAME; 1014 if (!lockparent) { 1015 VOP_UNLOCK(dvp, NULL, 0, td); 1016 cnp->cn_flags |= CNP_PDIRUNLOCK; 1017 } 1018 return (0); 1019 } 1020 1021 if (flags & CNP_ISDOTDOT) { 1022 VOP_UNLOCK(dvp, NULL, 0, td); 1023 cnp->cn_flags |= CNP_PDIRUNLOCK; 1024 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1025 if (error) { 1026 vn_lock(dvp, NULL, LK_EXCLUSIVE | LK_RETRY, td); 1027 cnp->cn_flags &= ~CNP_PDIRUNLOCK; 1028 return (error); /* NOTE: return error from nget */ 1029 } 1030 newvp = NFSTOV(np); 1031 if (lockparent && (flags & CNP_ISLASTCN)) { 1032 error = vn_lock(dvp, NULL, LK_EXCLUSIVE, td); 1033 if (error) { 1034 vput(newvp); 1035 return (error); 1036 } 1037 cnp->cn_flags |= CNP_PDIRUNLOCK; 1038 } 1039 } else if (NFS_CMPFH(np, fhp, fhsize)) { 1040 vref(dvp); 1041 newvp = dvp; 1042 } else { 1043 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1044 if (error) { 1045 m_freem(mrep); 1046 return (error); 1047 } 1048 if (!lockparent || !(flags & CNP_ISLASTCN)) { 1049 VOP_UNLOCK(dvp, NULL, 0, td); 1050 cnp->cn_flags |= CNP_PDIRUNLOCK; 1051 } 1052 newvp = NFSTOV(np); 1053 } 1054 if (v3) { 1055 nfsm_postop_attr(newvp, attrflag); 1056 nfsm_postop_attr(dvp, attrflag); 1057 } else 1058 nfsm_loadattr(newvp, (struct vattr *)0); 1059 if (cnp->cn_nameiop != NAMEI_LOOKUP && (flags & CNP_ISLASTCN)) 1060 cnp->cn_flags |= CNP_SAVENAME; 1061 if ((cnp->cn_flags & CNP_MAKEENTRY) && 1062 (cnp->cn_nameiop != NAMEI_DELETE || !(flags & CNP_ISLASTCN))) { 1063 np->n_ctime = np->n_vattr.va_ctime.tv_sec; 1064 cache_enter(dvp, newvp, cnp); 1065 } 1066 *vpp = newvp; 1067 m_freem(mrep); 1068 nfsmout: 1069 if (error) { 1070 if (newvp != NULLVP) { 1071 vrele(newvp); 1072 *vpp = NULLVP; 1073 } 1074 if ((cnp->cn_nameiop == NAMEI_CREATE || cnp->cn_nameiop == NAMEI_RENAME) && 1075 (flags & CNP_ISLASTCN) && error == ENOENT) { 1076 if (!lockparent) { 1077 VOP_UNLOCK(dvp, NULL, 0, td); 1078 cnp->cn_flags |= CNP_PDIRUNLOCK; 1079 } 1080 if (dvp->v_mount->mnt_flag & MNT_RDONLY) 1081 error = EROFS; 1082 else 1083 error = EJUSTRETURN; 1084 } 1085 if (cnp->cn_nameiop != NAMEI_LOOKUP && (flags & CNP_ISLASTCN)) 1086 cnp->cn_flags |= CNP_SAVENAME; 1087 } 1088 return (error); 1089 } 1090 1091 /* 1092 * nfs read call. 1093 * Just call nfs_bioread() to do the work. 1094 * 1095 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1096 * struct ucred *a_cred) 1097 */ 1098 static int 1099 nfs_read(struct vop_read_args *ap) 1100 { 1101 struct vnode *vp = ap->a_vp; 1102 1103 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag)); 1104 switch (vp->v_type) { 1105 case VREG: 1106 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag)); 1107 case VDIR: 1108 return (EISDIR); 1109 default: 1110 return EOPNOTSUPP; 1111 } 1112 } 1113 1114 /* 1115 * nfs readlink call 1116 * 1117 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 1118 */ 1119 static int 1120 nfs_readlink(struct vop_readlink_args *ap) 1121 { 1122 struct vnode *vp = ap->a_vp; 1123 1124 if (vp->v_type != VLNK) 1125 return (EINVAL); 1126 return (nfs_bioread(vp, ap->a_uio, 0)); 1127 } 1128 1129 /* 1130 * Do a readlink rpc. 1131 * Called by nfs_doio() from below the buffer cache. 1132 */ 1133 int 1134 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop) 1135 { 1136 u_int32_t *tl; 1137 caddr_t cp; 1138 int32_t t1, t2; 1139 caddr_t bpos, dpos, cp2; 1140 int error = 0, len, attrflag; 1141 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1142 int v3 = NFS_ISV3(vp); 1143 1144 nfsstats.rpccnt[NFSPROC_READLINK]++; 1145 nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3)); 1146 nfsm_fhtom(vp, v3); 1147 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, nfs_vpcred(vp, ND_CHECK)); 1148 if (v3) 1149 nfsm_postop_attr(vp, attrflag); 1150 if (!error) { 1151 nfsm_strsiz(len, NFS_MAXPATHLEN); 1152 if (len == NFS_MAXPATHLEN) { 1153 struct nfsnode *np = VTONFS(vp); 1154 if (np->n_size && np->n_size < NFS_MAXPATHLEN) 1155 len = np->n_size; 1156 } 1157 nfsm_mtouio(uiop, len); 1158 } 1159 m_freem(mrep); 1160 nfsmout: 1161 return (error); 1162 } 1163 1164 /* 1165 * nfs read rpc call 1166 * Ditto above 1167 */ 1168 int 1169 nfs_readrpc(struct vnode *vp, struct uio *uiop) 1170 { 1171 u_int32_t *tl; 1172 caddr_t cp; 1173 int32_t t1, t2; 1174 caddr_t bpos, dpos, cp2; 1175 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1176 struct nfsmount *nmp; 1177 int error = 0, len, retlen, tsiz, eof, attrflag; 1178 int v3 = NFS_ISV3(vp); 1179 1180 #ifndef nolint 1181 eof = 0; 1182 #endif 1183 nmp = VFSTONFS(vp->v_mount); 1184 tsiz = uiop->uio_resid; 1185 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1186 return (EFBIG); 1187 while (tsiz > 0) { 1188 nfsstats.rpccnt[NFSPROC_READ]++; 1189 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1190 nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1191 nfsm_fhtom(vp, v3); 1192 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3); 1193 if (v3) { 1194 txdr_hyper(uiop->uio_offset, tl); 1195 *(tl + 2) = txdr_unsigned(len); 1196 } else { 1197 *tl++ = txdr_unsigned(uiop->uio_offset); 1198 *tl++ = txdr_unsigned(len); 1199 *tl = 0; 1200 } 1201 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 1202 if (v3) { 1203 nfsm_postop_attr(vp, attrflag); 1204 if (error) { 1205 m_freem(mrep); 1206 goto nfsmout; 1207 } 1208 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1209 eof = fxdr_unsigned(int, *(tl + 1)); 1210 } else 1211 nfsm_loadattr(vp, (struct vattr *)0); 1212 nfsm_strsiz(retlen, nmp->nm_rsize); 1213 nfsm_mtouio(uiop, retlen); 1214 m_freem(mrep); 1215 tsiz -= retlen; 1216 if (v3) { 1217 if (eof || retlen == 0) { 1218 tsiz = 0; 1219 } 1220 } else if (retlen < len) { 1221 tsiz = 0; 1222 } 1223 } 1224 nfsmout: 1225 return (error); 1226 } 1227 1228 /* 1229 * nfs write call 1230 */ 1231 int 1232 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, int *must_commit) 1233 { 1234 u_int32_t *tl; 1235 caddr_t cp; 1236 int32_t t1, t2, backup; 1237 caddr_t bpos, dpos, cp2; 1238 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1239 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1240 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit; 1241 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC; 1242 1243 #ifndef DIAGNOSTIC 1244 if (uiop->uio_iovcnt != 1) 1245 panic("nfs: writerpc iovcnt > 1"); 1246 #endif 1247 *must_commit = 0; 1248 tsiz = uiop->uio_resid; 1249 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1250 return (EFBIG); 1251 while (tsiz > 0) { 1252 nfsstats.rpccnt[NFSPROC_WRITE]++; 1253 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz; 1254 nfsm_reqhead(vp, NFSPROC_WRITE, 1255 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len)); 1256 nfsm_fhtom(vp, v3); 1257 if (v3) { 1258 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1259 txdr_hyper(uiop->uio_offset, tl); 1260 tl += 2; 1261 *tl++ = txdr_unsigned(len); 1262 *tl++ = txdr_unsigned(*iomode); 1263 *tl = txdr_unsigned(len); 1264 } else { 1265 u_int32_t x; 1266 1267 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1268 /* Set both "begin" and "current" to non-garbage. */ 1269 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1270 *tl++ = x; /* "begin offset" */ 1271 *tl++ = x; /* "current offset" */ 1272 x = txdr_unsigned(len); 1273 *tl++ = x; /* total to this offset */ 1274 *tl = x; /* size of this write */ 1275 } 1276 nfsm_uiotom(uiop, len); 1277 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, nfs_vpcred(vp, ND_WRITE)); 1278 if (v3) { 1279 wccflag = NFSV3_WCCCHK; 1280 nfsm_wcc_data(vp, wccflag); 1281 if (!error) { 1282 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED 1283 + NFSX_V3WRITEVERF); 1284 rlen = fxdr_unsigned(int, *tl++); 1285 if (rlen == 0) { 1286 error = NFSERR_IO; 1287 m_freem(mrep); 1288 break; 1289 } else if (rlen < len) { 1290 backup = len - rlen; 1291 uiop->uio_iov->iov_base -= backup; 1292 uiop->uio_iov->iov_len += backup; 1293 uiop->uio_offset -= backup; 1294 uiop->uio_resid += backup; 1295 len = rlen; 1296 } 1297 commit = fxdr_unsigned(int, *tl++); 1298 1299 /* 1300 * Return the lowest committment level 1301 * obtained by any of the RPCs. 1302 */ 1303 if (committed == NFSV3WRITE_FILESYNC) 1304 committed = commit; 1305 else if (committed == NFSV3WRITE_DATASYNC && 1306 commit == NFSV3WRITE_UNSTABLE) 1307 committed = commit; 1308 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){ 1309 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1310 NFSX_V3WRITEVERF); 1311 nmp->nm_state |= NFSSTA_HASWRITEVERF; 1312 } else if (bcmp((caddr_t)tl, 1313 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) { 1314 *must_commit = 1; 1315 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1316 NFSX_V3WRITEVERF); 1317 } 1318 } 1319 } else 1320 nfsm_loadattr(vp, (struct vattr *)0); 1321 if (wccflag) 1322 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime.tv_sec; 1323 m_freem(mrep); 1324 if (error) 1325 break; 1326 tsiz -= len; 1327 } 1328 nfsmout: 1329 if (vp->v_mount->mnt_flag & MNT_ASYNC) 1330 committed = NFSV3WRITE_FILESYNC; 1331 *iomode = committed; 1332 if (error) 1333 uiop->uio_resid = tsiz; 1334 return (error); 1335 } 1336 1337 /* 1338 * nfs mknod rpc 1339 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1340 * mode set to specify the file type and the size field for rdev. 1341 */ 1342 static int 1343 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1344 struct vattr *vap) 1345 { 1346 struct nfsv2_sattr *sp; 1347 u_int32_t *tl; 1348 caddr_t cp; 1349 int32_t t1, t2; 1350 struct vnode *newvp = (struct vnode *)0; 1351 struct nfsnode *np = (struct nfsnode *)0; 1352 struct vattr vattr; 1353 char *cp2; 1354 caddr_t bpos, dpos; 1355 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1356 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1357 u_int32_t rdev; 1358 int v3 = NFS_ISV3(dvp); 1359 1360 if (vap->va_type == VCHR || vap->va_type == VBLK) 1361 rdev = txdr_unsigned(vap->va_rdev); 1362 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1363 rdev = nfs_xdrneg1; 1364 else { 1365 return (EOPNOTSUPP); 1366 } 1367 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) { 1368 return (error); 1369 } 1370 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1371 nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1372 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1373 nfsm_fhtom(dvp, v3); 1374 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1375 if (v3) { 1376 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1377 *tl++ = vtonfsv3_type(vap->va_type); 1378 nfsm_v3attrbuild(vap, FALSE); 1379 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1380 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1381 *tl++ = txdr_unsigned(umajor(vap->va_rdev)); 1382 *tl = txdr_unsigned(uminor(vap->va_rdev)); 1383 } 1384 } else { 1385 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1386 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1387 sp->sa_uid = nfs_xdrneg1; 1388 sp->sa_gid = nfs_xdrneg1; 1389 sp->sa_size = rdev; 1390 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1391 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1392 } 1393 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_td, cnp->cn_cred); 1394 if (!error) { 1395 nfsm_mtofh(dvp, newvp, v3, gotvp); 1396 if (!gotvp) { 1397 if (newvp) { 1398 vput(newvp); 1399 newvp = (struct vnode *)0; 1400 } 1401 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1402 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1403 if (!error) 1404 newvp = NFSTOV(np); 1405 } 1406 } 1407 if (v3) 1408 nfsm_wcc_data(dvp, wccflag); 1409 m_freem(mrep); 1410 nfsmout: 1411 if (error) { 1412 if (newvp) 1413 vput(newvp); 1414 } else { 1415 if (cnp->cn_flags & CNP_MAKEENTRY) 1416 cache_enter(dvp, newvp, cnp); 1417 *vpp = newvp; 1418 } 1419 VTONFS(dvp)->n_flag |= NMODIFIED; 1420 if (!wccflag) 1421 VTONFS(dvp)->n_attrstamp = 0; 1422 return (error); 1423 } 1424 1425 /* 1426 * nfs mknod vop 1427 * just call nfs_mknodrpc() to do the work. 1428 * 1429 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp, 1430 * struct componentname *a_cnp, struct vattr *a_vap) 1431 */ 1432 /* ARGSUSED */ 1433 static int 1434 nfs_mknod(struct vop_mknod_args *ap) 1435 { 1436 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap); 1437 } 1438 1439 static u_long create_verf; 1440 /* 1441 * nfs file create call 1442 * 1443 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp, 1444 * struct componentname *a_cnp, struct vattr *a_vap) 1445 */ 1446 static int 1447 nfs_create(struct vop_create_args *ap) 1448 { 1449 struct vnode *dvp = ap->a_dvp; 1450 struct vattr *vap = ap->a_vap; 1451 struct componentname *cnp = ap->a_cnp; 1452 struct nfsv2_sattr *sp; 1453 u_int32_t *tl; 1454 caddr_t cp; 1455 int32_t t1, t2; 1456 struct nfsnode *np = (struct nfsnode *)0; 1457 struct vnode *newvp = (struct vnode *)0; 1458 caddr_t bpos, dpos, cp2; 1459 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0; 1460 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1461 struct vattr vattr; 1462 int v3 = NFS_ISV3(dvp); 1463 1464 /* 1465 * Oops, not for me.. 1466 */ 1467 if (vap->va_type == VSOCK) 1468 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1469 1470 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) { 1471 return (error); 1472 } 1473 if (vap->va_vaflags & VA_EXCLUSIVE) 1474 fmode |= O_EXCL; 1475 again: 1476 nfsstats.rpccnt[NFSPROC_CREATE]++; 1477 nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1478 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1479 nfsm_fhtom(dvp, v3); 1480 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1481 if (v3) { 1482 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1483 if (fmode & O_EXCL) { 1484 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1485 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1486 #ifdef INET 1487 if (!TAILQ_EMPTY(&in_ifaddrhead)) 1488 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr; 1489 else 1490 #endif 1491 *tl++ = create_verf; 1492 *tl = ++create_verf; 1493 } else { 1494 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED); 1495 nfsm_v3attrbuild(vap, FALSE); 1496 } 1497 } else { 1498 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1499 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1500 sp->sa_uid = nfs_xdrneg1; 1501 sp->sa_gid = nfs_xdrneg1; 1502 sp->sa_size = 0; 1503 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1504 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1505 } 1506 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_td, cnp->cn_cred); 1507 if (!error) { 1508 nfsm_mtofh(dvp, newvp, v3, gotvp); 1509 if (!gotvp) { 1510 if (newvp) { 1511 vput(newvp); 1512 newvp = (struct vnode *)0; 1513 } 1514 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1515 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1516 if (!error) 1517 newvp = NFSTOV(np); 1518 } 1519 } 1520 if (v3) 1521 nfsm_wcc_data(dvp, wccflag); 1522 m_freem(mrep); 1523 nfsmout: 1524 if (error) { 1525 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) { 1526 fmode &= ~O_EXCL; 1527 goto again; 1528 } 1529 if (newvp) 1530 vput(newvp); 1531 } else if (v3 && (fmode & O_EXCL)) { 1532 /* 1533 * We are normally called with only a partially initialized 1534 * VAP. Since the NFSv3 spec says that server may use the 1535 * file attributes to store the verifier, the spec requires 1536 * us to do a SETATTR RPC. FreeBSD servers store the verifier 1537 * in atime, but we can't really assume that all servers will 1538 * so we ensure that our SETATTR sets both atime and mtime. 1539 */ 1540 if (vap->va_mtime.tv_sec == VNOVAL) 1541 vfs_timestamp(&vap->va_mtime); 1542 if (vap->va_atime.tv_sec == VNOVAL) 1543 vap->va_atime = vap->va_mtime; 1544 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td); 1545 } 1546 if (!error) { 1547 if (cnp->cn_flags & CNP_MAKEENTRY) 1548 cache_enter(dvp, newvp, cnp); 1549 /* 1550 * The new np may have enough info for access 1551 * checks, make sure rucred and wucred are 1552 * initialized for read and write rpc's. 1553 */ 1554 np = VTONFS(newvp); 1555 if (np->n_rucred == NULL) 1556 np->n_rucred = crhold(cnp->cn_cred); 1557 if (np->n_wucred == NULL) 1558 np->n_wucred = crhold(cnp->cn_cred); 1559 *ap->a_vpp = newvp; 1560 } 1561 VTONFS(dvp)->n_flag |= NMODIFIED; 1562 if (!wccflag) 1563 VTONFS(dvp)->n_attrstamp = 0; 1564 return (error); 1565 } 1566 1567 /* 1568 * nfs file remove call 1569 * To try and make nfs semantics closer to ufs semantics, a file that has 1570 * other processes using the vnode is renamed instead of removed and then 1571 * removed later on the last close. 1572 * - If v_usecount > 1 1573 * If a rename is not already in the works 1574 * call nfs_sillyrename() to set it up 1575 * else 1576 * do the remove rpc 1577 * 1578 * nfs_remove(struct vnodeop_desc *a_desc, struct vnode *a_dvp, 1579 * struct vnode *a_vp, struct componentname *a_cnp) 1580 */ 1581 static int 1582 nfs_remove(struct vop_remove_args *ap) 1583 { 1584 struct vnode *vp = ap->a_vp; 1585 struct vnode *dvp = ap->a_dvp; 1586 struct componentname *cnp = ap->a_cnp; 1587 struct nfsnode *np = VTONFS(vp); 1588 int error = 0; 1589 struct vattr vattr; 1590 1591 #ifndef DIAGNOSTIC 1592 if ((cnp->cn_flags & CNP_HASBUF) == 0) 1593 panic("nfs_remove: no name"); 1594 if (vp->v_usecount < 1) 1595 panic("nfs_remove: bad v_usecount"); 1596 #endif 1597 if (vp->v_type == VDIR) 1598 error = EPERM; 1599 else if (vp->v_usecount == 1 || (np->n_sillyrename && 1600 VOP_GETATTR(vp, &vattr, cnp->cn_td) == 0 && 1601 vattr.va_nlink > 1)) { 1602 /* 1603 * Purge the name cache so that the chance of a lookup for 1604 * the name succeeding while the remove is in progress is 1605 * minimized. Without node locking it can still happen, such 1606 * that an I/O op returns ESTALE, but since you get this if 1607 * another host removes the file.. 1608 */ 1609 cache_purge(vp); 1610 /* 1611 * throw away biocache buffers, mainly to avoid 1612 * unnecessary delayed writes later. 1613 */ 1614 error = nfs_vinvalbuf(vp, 0, cnp->cn_td, 1); 1615 /* Do the rpc */ 1616 if (error != EINTR) 1617 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1618 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td); 1619 /* 1620 * Kludge City: If the first reply to the remove rpc is lost.. 1621 * the reply to the retransmitted request will be ENOENT 1622 * since the file was in fact removed 1623 * Therefore, we cheat and return success. 1624 */ 1625 if (error == ENOENT) 1626 error = 0; 1627 } else if (!np->n_sillyrename) 1628 error = nfs_sillyrename(dvp, vp, cnp); 1629 np->n_attrstamp = 0; 1630 return (error); 1631 } 1632 1633 /* 1634 * nfs file remove rpc called from nfs_inactive 1635 */ 1636 int 1637 nfs_removeit(struct sillyrename *sp) 1638 { 1639 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, 1640 sp->s_cred, NULL)); 1641 } 1642 1643 /* 1644 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1645 */ 1646 static int 1647 nfs_removerpc(struct vnode *dvp, const char *name, int namelen, 1648 struct ucred *cred, struct thread *td) 1649 { 1650 u_int32_t *tl; 1651 caddr_t cp; 1652 int32_t t1, t2; 1653 caddr_t bpos, dpos, cp2; 1654 int error = 0, wccflag = NFSV3_WCCRATTR; 1655 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1656 int v3 = NFS_ISV3(dvp); 1657 1658 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1659 nfsm_reqhead(dvp, NFSPROC_REMOVE, 1660 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1661 nfsm_fhtom(dvp, v3); 1662 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1663 nfsm_request(dvp, NFSPROC_REMOVE, td, cred); 1664 if (v3) 1665 nfsm_wcc_data(dvp, wccflag); 1666 m_freem(mrep); 1667 nfsmout: 1668 VTONFS(dvp)->n_flag |= NMODIFIED; 1669 if (!wccflag) 1670 VTONFS(dvp)->n_attrstamp = 0; 1671 return (error); 1672 } 1673 1674 /* 1675 * nfs file rename call 1676 * 1677 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp, 1678 * struct componentname *a_fcnp, struct vnode *a_tdvp, 1679 * struct vnode *a_tvp, struct componentname *a_tcnp) 1680 */ 1681 static int 1682 nfs_rename(struct vop_rename_args *ap) 1683 { 1684 struct vnode *fvp = ap->a_fvp; 1685 struct vnode *tvp = ap->a_tvp; 1686 struct vnode *fdvp = ap->a_fdvp; 1687 struct vnode *tdvp = ap->a_tdvp; 1688 struct componentname *tcnp = ap->a_tcnp; 1689 struct componentname *fcnp = ap->a_fcnp; 1690 int error; 1691 1692 #ifndef DIAGNOSTIC 1693 if ((tcnp->cn_flags & CNP_HASBUF) == 0 || 1694 (fcnp->cn_flags & CNP_HASBUF) == 0) 1695 panic("nfs_rename: no name"); 1696 #endif 1697 /* Check for cross-device rename */ 1698 if ((fvp->v_mount != tdvp->v_mount) || 1699 (tvp && (fvp->v_mount != tvp->v_mount))) { 1700 error = EXDEV; 1701 goto out; 1702 } 1703 1704 /* 1705 * We have to flush B_DELWRI data prior to renaming 1706 * the file. If we don't, the delayed-write buffers 1707 * can be flushed out later after the file has gone stale 1708 * under NFSV3. NFSV2 does not have this problem because 1709 * ( as far as I can tell ) it flushes dirty buffers more 1710 * often. 1711 */ 1712 1713 VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_td); 1714 if (tvp) 1715 VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_td); 1716 1717 /* 1718 * If the tvp exists and is in use, sillyrename it before doing the 1719 * rename of the new file over it. 1720 * XXX Can't sillyrename a directory. 1721 */ 1722 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename && 1723 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1724 cache_purge(tvp); 1725 vput(tvp); 1726 tvp = NULL; 1727 } 1728 1729 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1730 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1731 tcnp->cn_td); 1732 1733 cache_purge(fvp); 1734 #if 0 1735 if (fvp->v_type == VDIR) { 1736 if (tvp != NULL && tvp->v_type == VDIR) 1737 cache_purge(tdvp); 1738 cache_purge(fdvp); 1739 } 1740 #endif 1741 1742 out: 1743 if (tdvp == tvp) 1744 vrele(tdvp); 1745 else 1746 vput(tdvp); 1747 if (tvp) 1748 vput(tvp); 1749 vrele(fdvp); 1750 vrele(fvp); 1751 /* 1752 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1753 */ 1754 if (error == ENOENT) 1755 error = 0; 1756 return (error); 1757 } 1758 1759 /* 1760 * nfs file rename rpc called from nfs_remove() above 1761 */ 1762 static int 1763 nfs_renameit(struct vnode *sdvp, struct componentname *scnp, 1764 struct sillyrename *sp) 1765 { 1766 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1767 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td)); 1768 } 1769 1770 /* 1771 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1772 */ 1773 static int 1774 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen, 1775 struct vnode *tdvp, const char *tnameptr, int tnamelen, 1776 struct ucred *cred, struct thread *td) 1777 { 1778 u_int32_t *tl; 1779 caddr_t cp; 1780 int32_t t1, t2; 1781 caddr_t bpos, dpos, cp2; 1782 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1783 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1784 int v3 = NFS_ISV3(fdvp); 1785 1786 nfsstats.rpccnt[NFSPROC_RENAME]++; 1787 nfsm_reqhead(fdvp, NFSPROC_RENAME, 1788 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 1789 nfsm_rndup(tnamelen)); 1790 nfsm_fhtom(fdvp, v3); 1791 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 1792 nfsm_fhtom(tdvp, v3); 1793 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 1794 nfsm_request(fdvp, NFSPROC_RENAME, td, cred); 1795 if (v3) { 1796 nfsm_wcc_data(fdvp, fwccflag); 1797 nfsm_wcc_data(tdvp, twccflag); 1798 } 1799 m_freem(mrep); 1800 nfsmout: 1801 VTONFS(fdvp)->n_flag |= NMODIFIED; 1802 VTONFS(tdvp)->n_flag |= NMODIFIED; 1803 if (!fwccflag) 1804 VTONFS(fdvp)->n_attrstamp = 0; 1805 if (!twccflag) 1806 VTONFS(tdvp)->n_attrstamp = 0; 1807 return (error); 1808 } 1809 1810 /* 1811 * nfs hard link create call 1812 * 1813 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp, 1814 * struct componentname *a_cnp) 1815 */ 1816 static int 1817 nfs_link(struct vop_link_args *ap) 1818 { 1819 struct vnode *vp = ap->a_vp; 1820 struct vnode *tdvp = ap->a_tdvp; 1821 struct componentname *cnp = ap->a_cnp; 1822 u_int32_t *tl; 1823 caddr_t cp; 1824 int32_t t1, t2; 1825 caddr_t bpos, dpos, cp2; 1826 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 1827 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1828 int v3; 1829 1830 if (vp->v_mount != tdvp->v_mount) { 1831 return (EXDEV); 1832 } 1833 1834 /* 1835 * Push all writes to the server, so that the attribute cache 1836 * doesn't get "out of sync" with the server. 1837 * XXX There should be a better way! 1838 */ 1839 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_td); 1840 1841 v3 = NFS_ISV3(vp); 1842 nfsstats.rpccnt[NFSPROC_LINK]++; 1843 nfsm_reqhead(vp, NFSPROC_LINK, 1844 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 1845 nfsm_fhtom(vp, v3); 1846 nfsm_fhtom(tdvp, v3); 1847 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1848 nfsm_request(vp, NFSPROC_LINK, cnp->cn_td, cnp->cn_cred); 1849 if (v3) { 1850 nfsm_postop_attr(vp, attrflag); 1851 nfsm_wcc_data(tdvp, wccflag); 1852 } 1853 m_freem(mrep); 1854 nfsmout: 1855 VTONFS(tdvp)->n_flag |= NMODIFIED; 1856 if (!attrflag) 1857 VTONFS(vp)->n_attrstamp = 0; 1858 if (!wccflag) 1859 VTONFS(tdvp)->n_attrstamp = 0; 1860 /* 1861 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 1862 */ 1863 if (error == EEXIST) 1864 error = 0; 1865 return (error); 1866 } 1867 1868 /* 1869 * nfs symbolic link create call 1870 * 1871 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp, 1872 * struct componentname *a_cnp, struct vattr *a_vap, 1873 * char *a_target) 1874 */ 1875 static int 1876 nfs_symlink(struct vop_symlink_args *ap) 1877 { 1878 struct vnode *dvp = ap->a_dvp; 1879 struct vattr *vap = ap->a_vap; 1880 struct componentname *cnp = ap->a_cnp; 1881 struct nfsv2_sattr *sp; 1882 u_int32_t *tl; 1883 caddr_t cp; 1884 int32_t t1, t2; 1885 caddr_t bpos, dpos, cp2; 1886 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 1887 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1888 struct vnode *newvp = (struct vnode *)0; 1889 int v3 = NFS_ISV3(dvp); 1890 1891 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 1892 slen = strlen(ap->a_target); 1893 nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 1894 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 1895 nfsm_fhtom(dvp, v3); 1896 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1897 if (v3) { 1898 nfsm_v3attrbuild(vap, FALSE); 1899 } 1900 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 1901 if (!v3) { 1902 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1903 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 1904 sp->sa_uid = nfs_xdrneg1; 1905 sp->sa_gid = nfs_xdrneg1; 1906 sp->sa_size = nfs_xdrneg1; 1907 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1908 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1909 } 1910 1911 /* 1912 * Issue the NFS request and get the rpc response. 1913 * 1914 * Only NFSv3 responses returning an error of 0 actually return 1915 * a file handle that can be converted into newvp without having 1916 * to do an extra lookup rpc. 1917 */ 1918 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_td, cnp->cn_cred); 1919 if (v3) { 1920 if (error == 0) 1921 nfsm_mtofh(dvp, newvp, v3, gotvp); 1922 nfsm_wcc_data(dvp, wccflag); 1923 } 1924 1925 /* 1926 * out code jumps -> here, mrep is also freed. 1927 */ 1928 1929 m_freem(mrep); 1930 nfsmout: 1931 1932 /* 1933 * If we get an EEXIST error, silently convert it to no-error 1934 * in case of an NFS retry. 1935 */ 1936 if (error == EEXIST) 1937 error = 0; 1938 1939 /* 1940 * If we do not have (or no longer have) an error, and we could 1941 * not extract the newvp from the response due to the request being 1942 * NFSv2 or the error being EEXIST. We have to do a lookup in order 1943 * to obtain a newvp to return. 1944 */ 1945 if (error == 0 && newvp == NULL) { 1946 struct nfsnode *np = NULL; 1947 1948 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1949 cnp->cn_cred, cnp->cn_td, &np); 1950 if (!error) 1951 newvp = NFSTOV(np); 1952 } 1953 if (error) { 1954 if (newvp) 1955 vput(newvp); 1956 } else { 1957 *ap->a_vpp = newvp; 1958 } 1959 VTONFS(dvp)->n_flag |= NMODIFIED; 1960 if (!wccflag) 1961 VTONFS(dvp)->n_attrstamp = 0; 1962 return (error); 1963 } 1964 1965 /* 1966 * nfs make dir call 1967 * 1968 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp, 1969 * struct componentname *a_cnp, struct vattr *a_vap) 1970 */ 1971 static int 1972 nfs_mkdir(struct vop_mkdir_args *ap) 1973 { 1974 struct vnode *dvp = ap->a_dvp; 1975 struct vattr *vap = ap->a_vap; 1976 struct componentname *cnp = ap->a_cnp; 1977 struct nfsv2_sattr *sp; 1978 u_int32_t *tl; 1979 caddr_t cp; 1980 int32_t t1, t2; 1981 int len; 1982 struct nfsnode *np = (struct nfsnode *)0; 1983 struct vnode *newvp = (struct vnode *)0; 1984 caddr_t bpos, dpos, cp2; 1985 int error = 0, wccflag = NFSV3_WCCRATTR; 1986 int gotvp = 0; 1987 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1988 struct vattr vattr; 1989 int v3 = NFS_ISV3(dvp); 1990 1991 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) { 1992 return (error); 1993 } 1994 len = cnp->cn_namelen; 1995 nfsstats.rpccnt[NFSPROC_MKDIR]++; 1996 nfsm_reqhead(dvp, NFSPROC_MKDIR, 1997 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 1998 nfsm_fhtom(dvp, v3); 1999 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 2000 if (v3) { 2001 nfsm_v3attrbuild(vap, FALSE); 2002 } else { 2003 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2004 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2005 sp->sa_uid = nfs_xdrneg1; 2006 sp->sa_gid = nfs_xdrneg1; 2007 sp->sa_size = nfs_xdrneg1; 2008 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2009 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2010 } 2011 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_td, cnp->cn_cred); 2012 if (!error) 2013 nfsm_mtofh(dvp, newvp, v3, gotvp); 2014 if (v3) 2015 nfsm_wcc_data(dvp, wccflag); 2016 m_freem(mrep); 2017 nfsmout: 2018 VTONFS(dvp)->n_flag |= NMODIFIED; 2019 if (!wccflag) 2020 VTONFS(dvp)->n_attrstamp = 0; 2021 /* 2022 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2023 * if we can succeed in looking up the directory. 2024 */ 2025 if (error == EEXIST || (!error && !gotvp)) { 2026 if (newvp) { 2027 vrele(newvp); 2028 newvp = (struct vnode *)0; 2029 } 2030 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2031 cnp->cn_td, &np); 2032 if (!error) { 2033 newvp = NFSTOV(np); 2034 if (newvp->v_type != VDIR) 2035 error = EEXIST; 2036 } 2037 } 2038 if (error) { 2039 if (newvp) 2040 vrele(newvp); 2041 } else 2042 *ap->a_vpp = newvp; 2043 return (error); 2044 } 2045 2046 /* 2047 * nfs remove directory call 2048 * 2049 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp, 2050 * struct componentname *a_cnp) 2051 */ 2052 static int 2053 nfs_rmdir(struct vop_rmdir_args *ap) 2054 { 2055 struct vnode *vp = ap->a_vp; 2056 struct vnode *dvp = ap->a_dvp; 2057 struct componentname *cnp = ap->a_cnp; 2058 u_int32_t *tl; 2059 caddr_t cp; 2060 int32_t t1, t2; 2061 caddr_t bpos, dpos, cp2; 2062 int error = 0, wccflag = NFSV3_WCCRATTR; 2063 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2064 int v3 = NFS_ISV3(dvp); 2065 2066 if (dvp == vp) 2067 return (EINVAL); 2068 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2069 nfsm_reqhead(dvp, NFSPROC_RMDIR, 2070 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2071 nfsm_fhtom(dvp, v3); 2072 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2073 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_td, cnp->cn_cred); 2074 if (v3) 2075 nfsm_wcc_data(dvp, wccflag); 2076 m_freem(mrep); 2077 nfsmout: 2078 VTONFS(dvp)->n_flag |= NMODIFIED; 2079 if (!wccflag) 2080 VTONFS(dvp)->n_attrstamp = 0; 2081 #if 0 2082 cache_purge(dvp); 2083 #endif 2084 cache_purge(vp); 2085 /* 2086 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2087 */ 2088 if (error == ENOENT) 2089 error = 0; 2090 return (error); 2091 } 2092 2093 /* 2094 * nfs readdir call 2095 * 2096 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 2097 */ 2098 static int 2099 nfs_readdir(struct vop_readdir_args *ap) 2100 { 2101 struct vnode *vp = ap->a_vp; 2102 struct nfsnode *np = VTONFS(vp); 2103 struct uio *uio = ap->a_uio; 2104 int tresid, error; 2105 struct vattr vattr; 2106 2107 if (vp->v_type != VDIR) 2108 return (EPERM); 2109 /* 2110 * First, check for hit on the EOF offset cache 2111 */ 2112 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2113 (np->n_flag & NMODIFIED) == 0) { 2114 if (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) { 2115 if (NQNFS_CKCACHABLE(vp, ND_READ)) { 2116 nfsstats.direofcache_hits++; 2117 return (0); 2118 } 2119 } else if (VOP_GETATTR(vp, &vattr, uio->uio_td) == 0 && 2120 np->n_mtime == vattr.va_mtime.tv_sec) { 2121 nfsstats.direofcache_hits++; 2122 return (0); 2123 } 2124 } 2125 2126 /* 2127 * Call nfs_bioread() to do the real work. 2128 */ 2129 tresid = uio->uio_resid; 2130 error = nfs_bioread(vp, uio, 0); 2131 2132 if (!error && uio->uio_resid == tresid) 2133 nfsstats.direofcache_misses++; 2134 return (error); 2135 } 2136 2137 /* 2138 * Readdir rpc call. 2139 * Called from below the buffer cache by nfs_doio(). 2140 */ 2141 int 2142 nfs_readdirrpc(struct vnode *vp, struct uio *uiop) 2143 { 2144 int len, left; 2145 struct dirent *dp = NULL; 2146 u_int32_t *tl; 2147 caddr_t cp; 2148 int32_t t1, t2; 2149 nfsuint64 *cookiep; 2150 caddr_t bpos, dpos, cp2; 2151 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2152 nfsuint64 cookie; 2153 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2154 struct nfsnode *dnp = VTONFS(vp); 2155 u_quad_t fileno; 2156 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1; 2157 int attrflag; 2158 int v3 = NFS_ISV3(vp); 2159 2160 #ifndef DIAGNOSTIC 2161 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2162 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2163 panic("nfs readdirrpc bad uio"); 2164 #endif 2165 2166 /* 2167 * If there is no cookie, assume directory was stale. 2168 */ 2169 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2170 if (cookiep) 2171 cookie = *cookiep; 2172 else 2173 return (NFSERR_BAD_COOKIE); 2174 /* 2175 * Loop around doing readdir rpc's of size nm_readdirsize 2176 * truncated to a multiple of DIRBLKSIZ. 2177 * The stopping criteria is EOF or buffer full. 2178 */ 2179 while (more_dirs && bigenough) { 2180 nfsstats.rpccnt[NFSPROC_READDIR]++; 2181 nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) + 2182 NFSX_READDIR(v3)); 2183 nfsm_fhtom(vp, v3); 2184 if (v3) { 2185 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2186 *tl++ = cookie.nfsuquad[0]; 2187 *tl++ = cookie.nfsuquad[1]; 2188 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2189 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2190 } else { 2191 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2192 *tl++ = cookie.nfsuquad[0]; 2193 } 2194 *tl = txdr_unsigned(nmp->nm_readdirsize); 2195 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 2196 if (v3) { 2197 nfsm_postop_attr(vp, attrflag); 2198 if (!error) { 2199 nfsm_dissect(tl, u_int32_t *, 2200 2 * NFSX_UNSIGNED); 2201 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2202 dnp->n_cookieverf.nfsuquad[1] = *tl; 2203 } else { 2204 m_freem(mrep); 2205 goto nfsmout; 2206 } 2207 } 2208 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2209 more_dirs = fxdr_unsigned(int, *tl); 2210 2211 /* loop thru the dir entries, doctoring them to 4bsd form */ 2212 while (more_dirs && bigenough) { 2213 if (v3) { 2214 nfsm_dissect(tl, u_int32_t *, 2215 3 * NFSX_UNSIGNED); 2216 fileno = fxdr_hyper(tl); 2217 len = fxdr_unsigned(int, *(tl + 2)); 2218 } else { 2219 nfsm_dissect(tl, u_int32_t *, 2220 2 * NFSX_UNSIGNED); 2221 fileno = fxdr_unsigned(u_quad_t, *tl++); 2222 len = fxdr_unsigned(int, *tl); 2223 } 2224 if (len <= 0 || len > NFS_MAXNAMLEN) { 2225 error = EBADRPC; 2226 m_freem(mrep); 2227 goto nfsmout; 2228 } 2229 tlen = nfsm_rndup(len); 2230 if (tlen == len) 2231 tlen += 4; /* To ensure null termination */ 2232 left = DIRBLKSIZ - blksiz; 2233 if ((tlen + DIRHDSIZ) > left) { 2234 dp->d_reclen += left; 2235 uiop->uio_iov->iov_base += left; 2236 uiop->uio_iov->iov_len -= left; 2237 uiop->uio_offset += left; 2238 uiop->uio_resid -= left; 2239 blksiz = 0; 2240 } 2241 if ((tlen + DIRHDSIZ) > uiop->uio_resid) 2242 bigenough = 0; 2243 if (bigenough) { 2244 dp = (struct dirent *)uiop->uio_iov->iov_base; 2245 dp->d_fileno = (int)fileno; 2246 dp->d_namlen = len; 2247 dp->d_reclen = tlen + DIRHDSIZ; 2248 dp->d_type = DT_UNKNOWN; 2249 blksiz += dp->d_reclen; 2250 if (blksiz == DIRBLKSIZ) 2251 blksiz = 0; 2252 uiop->uio_offset += DIRHDSIZ; 2253 uiop->uio_resid -= DIRHDSIZ; 2254 uiop->uio_iov->iov_base += DIRHDSIZ; 2255 uiop->uio_iov->iov_len -= DIRHDSIZ; 2256 nfsm_mtouio(uiop, len); 2257 cp = uiop->uio_iov->iov_base; 2258 tlen -= len; 2259 *cp = '\0'; /* null terminate */ 2260 uiop->uio_iov->iov_base += tlen; 2261 uiop->uio_iov->iov_len -= tlen; 2262 uiop->uio_offset += tlen; 2263 uiop->uio_resid -= tlen; 2264 } else 2265 nfsm_adv(nfsm_rndup(len)); 2266 if (v3) { 2267 nfsm_dissect(tl, u_int32_t *, 2268 3 * NFSX_UNSIGNED); 2269 } else { 2270 nfsm_dissect(tl, u_int32_t *, 2271 2 * NFSX_UNSIGNED); 2272 } 2273 if (bigenough) { 2274 cookie.nfsuquad[0] = *tl++; 2275 if (v3) 2276 cookie.nfsuquad[1] = *tl++; 2277 } else if (v3) 2278 tl += 2; 2279 else 2280 tl++; 2281 more_dirs = fxdr_unsigned(int, *tl); 2282 } 2283 /* 2284 * If at end of rpc data, get the eof boolean 2285 */ 2286 if (!more_dirs) { 2287 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2288 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2289 } 2290 m_freem(mrep); 2291 } 2292 /* 2293 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2294 * by increasing d_reclen for the last record. 2295 */ 2296 if (blksiz > 0) { 2297 left = DIRBLKSIZ - blksiz; 2298 dp->d_reclen += left; 2299 uiop->uio_iov->iov_base += left; 2300 uiop->uio_iov->iov_len -= left; 2301 uiop->uio_offset += left; 2302 uiop->uio_resid -= left; 2303 } 2304 2305 /* 2306 * We are now either at the end of the directory or have filled the 2307 * block. 2308 */ 2309 if (bigenough) 2310 dnp->n_direofoffset = uiop->uio_offset; 2311 else { 2312 if (uiop->uio_resid > 0) 2313 printf("EEK! readdirrpc resid > 0\n"); 2314 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2315 *cookiep = cookie; 2316 } 2317 nfsmout: 2318 return (error); 2319 } 2320 2321 /* 2322 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2323 */ 2324 int 2325 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop) 2326 { 2327 int len, left; 2328 struct dirent *dp; 2329 u_int32_t *tl; 2330 caddr_t cp; 2331 int32_t t1, t2; 2332 struct vnode *newvp; 2333 nfsuint64 *cookiep; 2334 caddr_t bpos, dpos, cp2, dpossav1, dpossav2; 2335 struct mbuf *mreq, *mrep, *md, *mb, *mb2, *mdsav1, *mdsav2; 2336 struct nameidata nami, *ndp = &nami; 2337 struct componentname *cnp = &ndp->ni_cnd; 2338 nfsuint64 cookie; 2339 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2340 struct nfsnode *dnp = VTONFS(vp), *np; 2341 nfsfh_t *fhp; 2342 u_quad_t fileno; 2343 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2344 int attrflag, fhsize; 2345 2346 #ifndef nolint 2347 dp = (struct dirent *)0; 2348 #endif 2349 #ifndef DIAGNOSTIC 2350 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2351 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2352 panic("nfs readdirplusrpc bad uio"); 2353 #endif 2354 ndp->ni_dvp = vp; 2355 newvp = NULLVP; 2356 2357 /* 2358 * If there is no cookie, assume directory was stale. 2359 */ 2360 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2361 if (cookiep) 2362 cookie = *cookiep; 2363 else 2364 return (NFSERR_BAD_COOKIE); 2365 /* 2366 * Loop around doing readdir rpc's of size nm_readdirsize 2367 * truncated to a multiple of DIRBLKSIZ. 2368 * The stopping criteria is EOF or buffer full. 2369 */ 2370 while (more_dirs && bigenough) { 2371 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2372 nfsm_reqhead(vp, NFSPROC_READDIRPLUS, 2373 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2374 nfsm_fhtom(vp, 1); 2375 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2376 *tl++ = cookie.nfsuquad[0]; 2377 *tl++ = cookie.nfsuquad[1]; 2378 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2379 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2380 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2381 *tl = txdr_unsigned(nmp->nm_rsize); 2382 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 2383 nfsm_postop_attr(vp, attrflag); 2384 if (error) { 2385 m_freem(mrep); 2386 goto nfsmout; 2387 } 2388 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2389 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2390 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2391 more_dirs = fxdr_unsigned(int, *tl); 2392 2393 /* loop thru the dir entries, doctoring them to 4bsd form */ 2394 while (more_dirs && bigenough) { 2395 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2396 fileno = fxdr_hyper(tl); 2397 len = fxdr_unsigned(int, *(tl + 2)); 2398 if (len <= 0 || len > NFS_MAXNAMLEN) { 2399 error = EBADRPC; 2400 m_freem(mrep); 2401 goto nfsmout; 2402 } 2403 tlen = nfsm_rndup(len); 2404 if (tlen == len) 2405 tlen += 4; /* To ensure null termination*/ 2406 left = DIRBLKSIZ - blksiz; 2407 if ((tlen + DIRHDSIZ) > left) { 2408 dp->d_reclen += left; 2409 uiop->uio_iov->iov_base += left; 2410 uiop->uio_iov->iov_len -= left; 2411 uiop->uio_offset += left; 2412 uiop->uio_resid -= left; 2413 blksiz = 0; 2414 } 2415 if ((tlen + DIRHDSIZ) > uiop->uio_resid) 2416 bigenough = 0; 2417 if (bigenough) { 2418 dp = (struct dirent *)uiop->uio_iov->iov_base; 2419 dp->d_fileno = (int)fileno; 2420 dp->d_namlen = len; 2421 dp->d_reclen = tlen + DIRHDSIZ; 2422 dp->d_type = DT_UNKNOWN; 2423 blksiz += dp->d_reclen; 2424 if (blksiz == DIRBLKSIZ) 2425 blksiz = 0; 2426 uiop->uio_offset += DIRHDSIZ; 2427 uiop->uio_resid -= DIRHDSIZ; 2428 uiop->uio_iov->iov_base += DIRHDSIZ; 2429 uiop->uio_iov->iov_len -= DIRHDSIZ; 2430 cnp->cn_nameptr = uiop->uio_iov->iov_base; 2431 cnp->cn_namelen = len; 2432 nfsm_mtouio(uiop, len); 2433 cp = uiop->uio_iov->iov_base; 2434 tlen -= len; 2435 *cp = '\0'; 2436 uiop->uio_iov->iov_base += tlen; 2437 uiop->uio_iov->iov_len -= tlen; 2438 uiop->uio_offset += tlen; 2439 uiop->uio_resid -= tlen; 2440 } else 2441 nfsm_adv(nfsm_rndup(len)); 2442 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2443 if (bigenough) { 2444 cookie.nfsuquad[0] = *tl++; 2445 cookie.nfsuquad[1] = *tl++; 2446 } else 2447 tl += 2; 2448 2449 /* 2450 * Since the attributes are before the file handle 2451 * (sigh), we must skip over the attributes and then 2452 * come back and get them. 2453 */ 2454 attrflag = fxdr_unsigned(int, *tl); 2455 if (attrflag) { 2456 dpossav1 = dpos; 2457 mdsav1 = md; 2458 nfsm_adv(NFSX_V3FATTR); 2459 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2460 doit = fxdr_unsigned(int, *tl); 2461 if (doit) { 2462 nfsm_getfh(fhp, fhsize, 1); 2463 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2464 vref(vp); 2465 newvp = vp; 2466 np = dnp; 2467 } else { 2468 error = nfs_nget(vp->v_mount, fhp, 2469 fhsize, &np); 2470 if (error) 2471 doit = 0; 2472 else 2473 newvp = NFSTOV(np); 2474 } 2475 } 2476 if (doit && bigenough) { 2477 dpossav2 = dpos; 2478 dpos = dpossav1; 2479 mdsav2 = md; 2480 md = mdsav1; 2481 nfsm_loadattr(newvp, (struct vattr *)0); 2482 dpos = dpossav2; 2483 md = mdsav2; 2484 dp->d_type = 2485 IFTODT(VTTOIF(np->n_vattr.va_type)); 2486 ndp->ni_vp = newvp; 2487 cache_enter(ndp->ni_dvp, ndp->ni_vp, cnp); 2488 } 2489 } else { 2490 /* Just skip over the file handle */ 2491 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2492 i = fxdr_unsigned(int, *tl); 2493 nfsm_adv(nfsm_rndup(i)); 2494 } 2495 if (newvp != NULLVP) { 2496 if (newvp == vp) 2497 vrele(newvp); 2498 else 2499 vput(newvp); 2500 newvp = NULLVP; 2501 } 2502 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2503 more_dirs = fxdr_unsigned(int, *tl); 2504 } 2505 /* 2506 * If at end of rpc data, get the eof boolean 2507 */ 2508 if (!more_dirs) { 2509 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2510 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2511 } 2512 m_freem(mrep); 2513 } 2514 /* 2515 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2516 * by increasing d_reclen for the last record. 2517 */ 2518 if (blksiz > 0) { 2519 left = DIRBLKSIZ - blksiz; 2520 dp->d_reclen += left; 2521 uiop->uio_iov->iov_base += left; 2522 uiop->uio_iov->iov_len -= left; 2523 uiop->uio_offset += left; 2524 uiop->uio_resid -= left; 2525 } 2526 2527 /* 2528 * We are now either at the end of the directory or have filled the 2529 * block. 2530 */ 2531 if (bigenough) 2532 dnp->n_direofoffset = uiop->uio_offset; 2533 else { 2534 if (uiop->uio_resid > 0) 2535 printf("EEK! readdirplusrpc resid > 0\n"); 2536 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2537 *cookiep = cookie; 2538 } 2539 nfsmout: 2540 if (newvp != NULLVP) { 2541 if (newvp == vp) 2542 vrele(newvp); 2543 else 2544 vput(newvp); 2545 newvp = NULLVP; 2546 } 2547 return (error); 2548 } 2549 2550 /* 2551 * Silly rename. To make the NFS filesystem that is stateless look a little 2552 * more like the "ufs" a remove of an active vnode is translated to a rename 2553 * to a funny looking filename that is removed by nfs_inactive on the 2554 * nfsnode. There is the potential for another process on a different client 2555 * to create the same funny name between the nfs_lookitup() fails and the 2556 * nfs_rename() completes, but... 2557 */ 2558 static int 2559 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2560 { 2561 struct sillyrename *sp; 2562 struct nfsnode *np; 2563 int error; 2564 2565 /* 2566 * We previously purged dvp instead of vp. I don't know why, it 2567 * completely destroys performance. We can't do it anyway with the 2568 * new VFS API since we would be breaking the namecache topology. 2569 */ 2570 cache_purge(vp); 2571 np = VTONFS(vp); 2572 #ifndef DIAGNOSTIC 2573 if (vp->v_type == VDIR) 2574 panic("nfs: sillyrename dir"); 2575 #endif 2576 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2577 M_NFSREQ, M_WAITOK); 2578 sp->s_cred = crdup(cnp->cn_cred); 2579 sp->s_dvp = dvp; 2580 vref(dvp); 2581 2582 /* Fudge together a funny name */ 2583 sp->s_namlen = sprintf(sp->s_name, ".nfsA%08x4.4", (int)cnp->cn_td); 2584 2585 /* Try lookitups until we get one that isn't there */ 2586 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2587 cnp->cn_td, (struct nfsnode **)0) == 0) { 2588 sp->s_name[4]++; 2589 if (sp->s_name[4] > 'z') { 2590 error = EINVAL; 2591 goto bad; 2592 } 2593 } 2594 error = nfs_renameit(dvp, cnp, sp); 2595 if (error) 2596 goto bad; 2597 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2598 cnp->cn_td, &np); 2599 np->n_sillyrename = sp; 2600 return (0); 2601 bad: 2602 vrele(sp->s_dvp); 2603 crfree(sp->s_cred); 2604 free((caddr_t)sp, M_NFSREQ); 2605 return (error); 2606 } 2607 2608 /* 2609 * Look up a file name and optionally either update the file handle or 2610 * allocate an nfsnode, depending on the value of npp. 2611 * npp == NULL --> just do the lookup 2612 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2613 * handled too 2614 * *npp != NULL --> update the file handle in the vnode 2615 */ 2616 static int 2617 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred, 2618 struct thread *td, struct nfsnode **npp) 2619 { 2620 u_int32_t *tl; 2621 caddr_t cp; 2622 int32_t t1, t2; 2623 struct vnode *newvp = (struct vnode *)0; 2624 struct nfsnode *np, *dnp = VTONFS(dvp); 2625 caddr_t bpos, dpos, cp2; 2626 int error = 0, fhlen, attrflag; 2627 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2628 nfsfh_t *nfhp; 2629 int v3 = NFS_ISV3(dvp); 2630 2631 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2632 nfsm_reqhead(dvp, NFSPROC_LOOKUP, 2633 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2634 nfsm_fhtom(dvp, v3); 2635 nfsm_strtom(name, len, NFS_MAXNAMLEN); 2636 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred); 2637 if (npp && !error) { 2638 nfsm_getfh(nfhp, fhlen, v3); 2639 if (*npp) { 2640 np = *npp; 2641 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2642 free((caddr_t)np->n_fhp, M_NFSBIGFH); 2643 np->n_fhp = &np->n_fh; 2644 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH) 2645 np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK); 2646 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen); 2647 np->n_fhsize = fhlen; 2648 newvp = NFSTOV(np); 2649 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2650 vref(dvp); 2651 newvp = dvp; 2652 } else { 2653 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 2654 if (error) { 2655 m_freem(mrep); 2656 return (error); 2657 } 2658 newvp = NFSTOV(np); 2659 } 2660 if (v3) { 2661 nfsm_postop_attr(newvp, attrflag); 2662 if (!attrflag && *npp == NULL) { 2663 m_freem(mrep); 2664 if (newvp == dvp) 2665 vrele(newvp); 2666 else 2667 vput(newvp); 2668 return (ENOENT); 2669 } 2670 } else 2671 nfsm_loadattr(newvp, (struct vattr *)0); 2672 } 2673 m_freem(mrep); 2674 nfsmout: 2675 if (npp && *npp == NULL) { 2676 if (error) { 2677 if (newvp) { 2678 if (newvp == dvp) 2679 vrele(newvp); 2680 else 2681 vput(newvp); 2682 } 2683 } else 2684 *npp = np; 2685 } 2686 return (error); 2687 } 2688 2689 /* 2690 * Nfs Version 3 commit rpc 2691 */ 2692 int 2693 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td) 2694 { 2695 caddr_t cp; 2696 u_int32_t *tl; 2697 int32_t t1, t2; 2698 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2699 caddr_t bpos, dpos, cp2; 2700 int error = 0, wccflag = NFSV3_WCCRATTR; 2701 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2702 2703 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) 2704 return (0); 2705 nfsstats.rpccnt[NFSPROC_COMMIT]++; 2706 nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1)); 2707 nfsm_fhtom(vp, 1); 2708 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2709 txdr_hyper(offset, tl); 2710 tl += 2; 2711 *tl = txdr_unsigned(cnt); 2712 nfsm_request(vp, NFSPROC_COMMIT, td, nfs_vpcred(vp, ND_WRITE)); 2713 nfsm_wcc_data(vp, wccflag); 2714 if (!error) { 2715 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 2716 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl, 2717 NFSX_V3WRITEVERF)) { 2718 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 2719 NFSX_V3WRITEVERF); 2720 error = NFSERR_STALEWRITEVERF; 2721 } 2722 } 2723 m_freem(mrep); 2724 nfsmout: 2725 return (error); 2726 } 2727 2728 /* 2729 * Kludge City.. 2730 * - make nfs_bmap() essentially a no-op that does no translation 2731 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 2732 * (Maybe I could use the process's page mapping, but I was concerned that 2733 * Kernel Write might not be enabled and also figured copyout() would do 2734 * a lot more work than bcopy() and also it currently happens in the 2735 * context of the swapper process (2). 2736 * 2737 * nfs_bmap(struct vnode *a_vp, daddr_t a_bn, struct vnode **a_vpp, 2738 * daddr_t *a_bnp, int *a_runp, int *a_runb) 2739 */ 2740 static int 2741 nfs_bmap(struct vop_bmap_args *ap) 2742 { 2743 struct vnode *vp = ap->a_vp; 2744 2745 if (ap->a_vpp != NULL) 2746 *ap->a_vpp = vp; 2747 if (ap->a_bnp != NULL) 2748 *ap->a_bnp = ap->a_bn * btodb(vp->v_mount->mnt_stat.f_iosize); 2749 if (ap->a_runp != NULL) 2750 *ap->a_runp = 0; 2751 if (ap->a_runb != NULL) 2752 *ap->a_runb = 0; 2753 return (0); 2754 } 2755 2756 /* 2757 * Strategy routine. 2758 * For async requests when nfsiod(s) are running, queue the request by 2759 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 2760 * request. 2761 */ 2762 static int 2763 nfs_strategy(struct vop_strategy_args *ap) 2764 { 2765 struct buf *bp = ap->a_bp; 2766 struct thread *td; 2767 int error = 0; 2768 2769 KASSERT(!(bp->b_flags & B_DONE), ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); 2770 KASSERT(BUF_REFCNT(bp) > 0, ("nfs_strategy: buffer %p not locked", bp)); 2771 2772 if (bp->b_flags & B_PHYS) 2773 panic("nfs physio"); 2774 2775 if (bp->b_flags & B_ASYNC) 2776 td = NULL; 2777 else 2778 td = curthread; /* XXX */ 2779 2780 /* 2781 * If the op is asynchronous and an i/o daemon is waiting 2782 * queue the request, wake it up and wait for completion 2783 * otherwise just do it ourselves. 2784 */ 2785 if ((bp->b_flags & B_ASYNC) == 0 || 2786 nfs_asyncio(bp, td)) 2787 error = nfs_doio(bp, td); 2788 return (error); 2789 } 2790 2791 /* 2792 * Mmap a file 2793 * 2794 * NB Currently unsupported. 2795 * 2796 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred, 2797 * struct thread *a_td) 2798 */ 2799 /* ARGSUSED */ 2800 static int 2801 nfs_mmap(struct vop_mmap_args *ap) 2802 { 2803 return (EINVAL); 2804 } 2805 2806 /* 2807 * fsync vnode op. Just call nfs_flush() with commit == 1. 2808 * 2809 * nfs_fsync(struct vnodeop_desc *a_desc, struct vnode *a_vp, 2810 * struct ucred * a_cred, int a_waitfor, struct thread *a_td) 2811 */ 2812 /* ARGSUSED */ 2813 static int 2814 nfs_fsync(struct vop_fsync_args *ap) 2815 { 2816 return (nfs_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1)); 2817 } 2818 2819 /* 2820 * Flush all the blocks associated with a vnode. 2821 * Walk through the buffer pool and push any dirty pages 2822 * associated with the vnode. 2823 */ 2824 static int 2825 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit) 2826 { 2827 struct nfsnode *np = VTONFS(vp); 2828 struct buf *bp; 2829 int i; 2830 struct buf *nbp; 2831 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2832 int s, error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos; 2833 int passone = 1; 2834 u_quad_t off, endoff, toff; 2835 struct buf **bvec = NULL; 2836 #ifndef NFS_COMMITBVECSIZ 2837 #define NFS_COMMITBVECSIZ 20 2838 #endif 2839 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ]; 2840 int bvecsize = 0, bveccount; 2841 2842 if (nmp->nm_flag & NFSMNT_INT) 2843 slpflag = PCATCH; 2844 if (!commit) 2845 passone = 0; 2846 /* 2847 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the 2848 * server, but nas not been committed to stable storage on the server 2849 * yet. On the first pass, the byte range is worked out and the commit 2850 * rpc is done. On the second pass, nfs_writebp() is called to do the 2851 * job. 2852 */ 2853 again: 2854 off = (u_quad_t)-1; 2855 endoff = 0; 2856 bvecpos = 0; 2857 if (NFS_ISV3(vp) && commit) { 2858 s = splbio(); 2859 /* 2860 * Count up how many buffers waiting for a commit. 2861 */ 2862 bveccount = 0; 2863 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 2864 nbp = TAILQ_NEXT(bp, b_vnbufs); 2865 if (BUF_REFCNT(bp) == 0 && 2866 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) 2867 == (B_DELWRI | B_NEEDCOMMIT)) 2868 bveccount++; 2869 } 2870 /* 2871 * Allocate space to remember the list of bufs to commit. It is 2872 * important to use M_NOWAIT here to avoid a race with nfs_write. 2873 * If we can't get memory (for whatever reason), we will end up 2874 * committing the buffers one-by-one in the loop below. 2875 */ 2876 if (bvec != NULL && bvec != bvec_on_stack) 2877 free(bvec, M_TEMP); 2878 if (bveccount > NFS_COMMITBVECSIZ) { 2879 bvec = (struct buf **) 2880 malloc(bveccount * sizeof(struct buf *), 2881 M_TEMP, M_NOWAIT); 2882 if (bvec == NULL) { 2883 bvec = bvec_on_stack; 2884 bvecsize = NFS_COMMITBVECSIZ; 2885 } else 2886 bvecsize = bveccount; 2887 } else { 2888 bvec = bvec_on_stack; 2889 bvecsize = NFS_COMMITBVECSIZ; 2890 } 2891 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 2892 nbp = TAILQ_NEXT(bp, b_vnbufs); 2893 if (bvecpos >= bvecsize) 2894 break; 2895 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 2896 (B_DELWRI | B_NEEDCOMMIT) || 2897 BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) 2898 continue; 2899 bremfree(bp); 2900 /* 2901 * NOTE: we are not clearing B_DONE here, so we have 2902 * to do it later on in this routine if we intend to 2903 * initiate I/O on the bp. 2904 * 2905 * Note: to avoid loopback deadlocks, we do not 2906 * assign b_runningbufspace. 2907 */ 2908 bp->b_flags |= B_WRITEINPROG; 2909 vfs_busy_pages(bp, 1); 2910 2911 /* 2912 * bp is protected by being locked, but nbp is not 2913 * and vfs_busy_pages() may sleep. We have to 2914 * recalculate nbp. 2915 */ 2916 nbp = TAILQ_NEXT(bp, b_vnbufs); 2917 2918 /* 2919 * A list of these buffers is kept so that the 2920 * second loop knows which buffers have actually 2921 * been committed. This is necessary, since there 2922 * may be a race between the commit rpc and new 2923 * uncommitted writes on the file. 2924 */ 2925 bvec[bvecpos++] = bp; 2926 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2927 bp->b_dirtyoff; 2928 if (toff < off) 2929 off = toff; 2930 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); 2931 if (toff > endoff) 2932 endoff = toff; 2933 } 2934 splx(s); 2935 } 2936 if (bvecpos > 0) { 2937 /* 2938 * Commit data on the server, as required. Note that 2939 * nfs_commit will use the vnode's cred for the commit. 2940 */ 2941 retv = nfs_commit(vp, off, (int)(endoff - off), td); 2942 2943 if (retv == NFSERR_STALEWRITEVERF) 2944 nfs_clearcommit(vp->v_mount); 2945 2946 /* 2947 * Now, either mark the blocks I/O done or mark the 2948 * blocks dirty, depending on whether the commit 2949 * succeeded. 2950 */ 2951 for (i = 0; i < bvecpos; i++) { 2952 bp = bvec[i]; 2953 bp->b_flags &= ~(B_NEEDCOMMIT | B_WRITEINPROG | B_CLUSTEROK); 2954 if (retv) { 2955 /* 2956 * Error, leave B_DELWRI intact 2957 */ 2958 vfs_unbusy_pages(bp); 2959 brelse(bp); 2960 } else { 2961 /* 2962 * Success, remove B_DELWRI ( bundirty() ). 2963 * 2964 * b_dirtyoff/b_dirtyend seem to be NFS 2965 * specific. We should probably move that 2966 * into bundirty(). XXX 2967 */ 2968 s = splbio(); 2969 vp->v_numoutput++; 2970 bp->b_flags |= B_ASYNC; 2971 bundirty(bp); 2972 bp->b_flags &= ~(B_READ|B_DONE|B_ERROR); 2973 bp->b_dirtyoff = bp->b_dirtyend = 0; 2974 splx(s); 2975 biodone(bp); 2976 } 2977 } 2978 } 2979 2980 /* 2981 * Start/do any write(s) that are required. 2982 */ 2983 loop: 2984 s = splbio(); 2985 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 2986 nbp = TAILQ_NEXT(bp, b_vnbufs); 2987 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 2988 if (waitfor != MNT_WAIT || passone) 2989 continue; 2990 error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL, 2991 "nfsfsync", slpflag, slptimeo); 2992 splx(s); 2993 if (error == 0) 2994 panic("nfs_fsync: inconsistent lock"); 2995 if (error == ENOLCK) 2996 goto loop; 2997 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) { 2998 error = EINTR; 2999 goto done; 3000 } 3001 if (slpflag == PCATCH) { 3002 slpflag = 0; 3003 slptimeo = 2 * hz; 3004 } 3005 goto loop; 3006 } 3007 if ((bp->b_flags & B_DELWRI) == 0) 3008 panic("nfs_fsync: not dirty"); 3009 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) { 3010 BUF_UNLOCK(bp); 3011 continue; 3012 } 3013 bremfree(bp); 3014 if (passone || !commit) 3015 bp->b_flags |= B_ASYNC; 3016 else 3017 bp->b_flags |= B_ASYNC | B_WRITEINPROG; 3018 splx(s); 3019 VOP_BWRITE(bp->b_vp, bp); 3020 goto loop; 3021 } 3022 splx(s); 3023 if (passone) { 3024 passone = 0; 3025 goto again; 3026 } 3027 if (waitfor == MNT_WAIT) { 3028 while (vp->v_numoutput) { 3029 vp->v_flag |= VBWAIT; 3030 error = tsleep((caddr_t)&vp->v_numoutput, 3031 slpflag, "nfsfsync", slptimeo); 3032 if (error) { 3033 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) { 3034 error = EINTR; 3035 goto done; 3036 } 3037 if (slpflag == PCATCH) { 3038 slpflag = 0; 3039 slptimeo = 2 * hz; 3040 } 3041 } 3042 } 3043 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) && commit) { 3044 goto loop; 3045 } 3046 } 3047 if (np->n_flag & NWRITEERR) { 3048 error = np->n_error; 3049 np->n_flag &= ~NWRITEERR; 3050 } 3051 done: 3052 if (bvec != NULL && bvec != bvec_on_stack) 3053 free(bvec, M_TEMP); 3054 return (error); 3055 } 3056 3057 /* 3058 * NFS advisory byte-level locks. 3059 * Currently unsupported. 3060 * 3061 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl, 3062 * int a_flags) 3063 */ 3064 static int 3065 nfs_advlock(struct vop_advlock_args *ap) 3066 { 3067 struct nfsnode *np = VTONFS(ap->a_vp); 3068 3069 /* 3070 * The following kludge is to allow diskless support to work 3071 * until a real NFS lockd is implemented. Basically, just pretend 3072 * that this is a local lock. 3073 */ 3074 return (lf_advlock(ap, &(np->n_lockf), np->n_size)); 3075 } 3076 3077 /* 3078 * Print out the contents of an nfsnode. 3079 * 3080 * nfs_print(struct vnode *a_vp) 3081 */ 3082 static int 3083 nfs_print(struct vop_print_args *ap) 3084 { 3085 struct vnode *vp = ap->a_vp; 3086 struct nfsnode *np = VTONFS(vp); 3087 3088 printf("tag VT_NFS, fileid %ld fsid 0x%x", 3089 np->n_vattr.va_fileid, np->n_vattr.va_fsid); 3090 if (vp->v_type == VFIFO) 3091 fifo_printinfo(vp); 3092 printf("\n"); 3093 return (0); 3094 } 3095 3096 /* 3097 * Just call nfs_writebp() with the force argument set to 1. 3098 * 3099 * NOTE: B_DONE may or may not be set in a_bp on call. 3100 * 3101 * nfs_bwrite(struct vnode *a_bp) 3102 */ 3103 static int 3104 nfs_bwrite(struct vop_bwrite_args *ap) 3105 { 3106 return (nfs_writebp(ap->a_bp, 1, curthread)); 3107 } 3108 3109 /* 3110 * This is a clone of vn_bwrite(), except that B_WRITEINPROG isn't set unless 3111 * the force flag is one and it also handles the B_NEEDCOMMIT flag. We set 3112 * B_CACHE if this is a VMIO buffer. 3113 */ 3114 int 3115 nfs_writebp(struct buf *bp, int force, struct thread *td) 3116 { 3117 int s; 3118 int oldflags = bp->b_flags; 3119 #if 0 3120 int retv = 1; 3121 off_t off; 3122 #endif 3123 3124 if (BUF_REFCNT(bp) == 0) 3125 panic("bwrite: buffer is not locked???"); 3126 3127 if (bp->b_flags & B_INVAL) { 3128 brelse(bp); 3129 return(0); 3130 } 3131 3132 bp->b_flags |= B_CACHE; 3133 3134 /* 3135 * Undirty the bp. We will redirty it later if the I/O fails. 3136 */ 3137 3138 s = splbio(); 3139 bundirty(bp); 3140 bp->b_flags &= ~(B_READ|B_DONE|B_ERROR); 3141 3142 bp->b_vp->v_numoutput++; 3143 splx(s); 3144 3145 /* 3146 * Note: to avoid loopback deadlocks, we do not 3147 * assign b_runningbufspace. 3148 */ 3149 vfs_busy_pages(bp, 1); 3150 3151 if (force) 3152 bp->b_flags |= B_WRITEINPROG; 3153 BUF_KERNPROC(bp); 3154 VOP_STRATEGY(bp->b_vp, bp); 3155 3156 if( (oldflags & B_ASYNC) == 0) { 3157 int rtval = biowait(bp); 3158 3159 if (oldflags & B_DELWRI) { 3160 s = splbio(); 3161 reassignbuf(bp, bp->b_vp); 3162 splx(s); 3163 } 3164 3165 brelse(bp); 3166 return (rtval); 3167 } 3168 3169 return (0); 3170 } 3171 3172 /* 3173 * nfs special file access vnode op. 3174 * Essentially just get vattr and then imitate iaccess() since the device is 3175 * local to the client. 3176 * 3177 * nfsspec_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 3178 * struct thread *a_td) 3179 */ 3180 static int 3181 nfsspec_access(struct vop_access_args *ap) 3182 { 3183 struct vattr *vap; 3184 gid_t *gp; 3185 struct ucred *cred = ap->a_cred; 3186 struct vnode *vp = ap->a_vp; 3187 mode_t mode = ap->a_mode; 3188 struct vattr vattr; 3189 int i; 3190 int error; 3191 3192 /* 3193 * Disallow write attempts on filesystems mounted read-only; 3194 * unless the file is a socket, fifo, or a block or character 3195 * device resident on the filesystem. 3196 */ 3197 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3198 switch (vp->v_type) { 3199 case VREG: 3200 case VDIR: 3201 case VLNK: 3202 return (EROFS); 3203 default: 3204 break; 3205 } 3206 } 3207 /* 3208 * If you're the super-user, 3209 * you always get access. 3210 */ 3211 if (cred->cr_uid == 0) 3212 return (0); 3213 vap = &vattr; 3214 error = VOP_GETATTR(vp, vap, ap->a_td); 3215 if (error) 3216 return (error); 3217 /* 3218 * Access check is based on only one of owner, group, public. 3219 * If not owner, then check group. If not a member of the 3220 * group, then check public access. 3221 */ 3222 if (cred->cr_uid != vap->va_uid) { 3223 mode >>= 3; 3224 gp = cred->cr_groups; 3225 for (i = 0; i < cred->cr_ngroups; i++, gp++) 3226 if (vap->va_gid == *gp) 3227 goto found; 3228 mode >>= 3; 3229 found: 3230 ; 3231 } 3232 error = (vap->va_mode & mode) == mode ? 0 : EACCES; 3233 return (error); 3234 } 3235 3236 /* 3237 * Read wrapper for special devices. 3238 * 3239 * nfsspec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3240 * struct ucred *a_cred) 3241 */ 3242 static int 3243 nfsspec_read(struct vop_read_args *ap) 3244 { 3245 struct nfsnode *np = VTONFS(ap->a_vp); 3246 3247 /* 3248 * Set access flag. 3249 */ 3250 np->n_flag |= NACC; 3251 getnanotime(&np->n_atim); 3252 return (VOCALL(spec_vnode_vops, &ap->a_head)); 3253 } 3254 3255 /* 3256 * Write wrapper for special devices. 3257 * 3258 * nfsspec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3259 * struct ucred *a_cred) 3260 */ 3261 static int 3262 nfsspec_write(struct vop_write_args *ap) 3263 { 3264 struct nfsnode *np = VTONFS(ap->a_vp); 3265 3266 /* 3267 * Set update flag. 3268 */ 3269 np->n_flag |= NUPD; 3270 getnanotime(&np->n_mtim); 3271 return (VOCALL(spec_vnode_vops, &ap->a_head)); 3272 } 3273 3274 /* 3275 * Close wrapper for special devices. 3276 * 3277 * Update the times on the nfsnode then do device close. 3278 * 3279 * nfsspec_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred, 3280 * struct thread *a_td) 3281 */ 3282 static int 3283 nfsspec_close(struct vop_close_args *ap) 3284 { 3285 struct vnode *vp = ap->a_vp; 3286 struct nfsnode *np = VTONFS(vp); 3287 struct vattr vattr; 3288 3289 if (np->n_flag & (NACC | NUPD)) { 3290 np->n_flag |= NCHG; 3291 if (vp->v_usecount == 1 && 3292 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3293 VATTR_NULL(&vattr); 3294 if (np->n_flag & NACC) 3295 vattr.va_atime = np->n_atim; 3296 if (np->n_flag & NUPD) 3297 vattr.va_mtime = np->n_mtim; 3298 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE), ap->a_td); 3299 } 3300 } 3301 return (VOCALL(spec_vnode_vops, &ap->a_head)); 3302 } 3303 3304 /* 3305 * Read wrapper for fifos. 3306 * 3307 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3308 * struct ucred *a_cred) 3309 */ 3310 static int 3311 nfsfifo_read(struct vop_read_args *ap) 3312 { 3313 struct nfsnode *np = VTONFS(ap->a_vp); 3314 3315 /* 3316 * Set access flag. 3317 */ 3318 np->n_flag |= NACC; 3319 getnanotime(&np->n_atim); 3320 return (VOCALL(fifo_vnode_vops, &ap->a_head)); 3321 } 3322 3323 /* 3324 * Write wrapper for fifos. 3325 * 3326 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3327 * struct ucred *a_cred) 3328 */ 3329 static int 3330 nfsfifo_write(struct vop_write_args *ap) 3331 { 3332 struct nfsnode *np = VTONFS(ap->a_vp); 3333 3334 /* 3335 * Set update flag. 3336 */ 3337 np->n_flag |= NUPD; 3338 getnanotime(&np->n_mtim); 3339 return (VOCALL(fifo_vnode_vops, &ap->a_head)); 3340 } 3341 3342 /* 3343 * Close wrapper for fifos. 3344 * 3345 * Update the times on the nfsnode then do fifo close. 3346 * 3347 * nfsfifo_close(struct vnode *a_vp, int a_fflag, struct thread *a_td) 3348 */ 3349 static int 3350 nfsfifo_close(struct vop_close_args *ap) 3351 { 3352 struct vnode *vp = ap->a_vp; 3353 struct nfsnode *np = VTONFS(vp); 3354 struct vattr vattr; 3355 struct timespec ts; 3356 3357 if (np->n_flag & (NACC | NUPD)) { 3358 getnanotime(&ts); 3359 if (np->n_flag & NACC) 3360 np->n_atim = ts; 3361 if (np->n_flag & NUPD) 3362 np->n_mtim = ts; 3363 np->n_flag |= NCHG; 3364 if (vp->v_usecount == 1 && 3365 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3366 VATTR_NULL(&vattr); 3367 if (np->n_flag & NACC) 3368 vattr.va_atime = np->n_atim; 3369 if (np->n_flag & NUPD) 3370 vattr.va_mtime = np->n_mtim; 3371 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE), ap->a_td); 3372 } 3373 } 3374 return (VOCALL(fifo_vnode_vops, &ap->a_head)); 3375 } 3376 3377