Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
FreeBSD/Linux Kernel Cross Reference
sys/nfsclient/nfs_vnops.c
Version:
- FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
SearchContext: - none - 3 - 10
1 /*- 2 * Copyright (c) 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * Rick Macklem at The University of Guelph. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 4. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * @(#)nfs_vnops.c 8.16 (Berkeley) 5/27/95 33 */ 34 35 #include <sys/cdefs.h> 36 __FBSDID("$FreeBSD: src/sys/nfsclient/nfs_vnops.c,v 1.220.2.4 2006/04/18 05:18:47 jon Exp $"); 37 38 /* 39 * vnode op calls for Sun NFS version 2 and 3 40 */ 41 42 #include "opt_inet.h" 43 44 #include <sys/param.h> 45 #include <sys/kernel.h> 46 #include <sys/systm.h> 47 #include <sys/resourcevar.h> 48 #include <sys/proc.h> 49 #include <sys/mount.h> 50 #include <sys/bio.h> 51 #include <sys/buf.h> 52 #include <sys/malloc.h> 53 #include <sys/mbuf.h> 54 #include <sys/namei.h> 55 #include <sys/socket.h> 56 #include <sys/vnode.h> 57 #include <sys/dirent.h> 58 #include <sys/fcntl.h> 59 #include <sys/lockf.h> 60 #include <sys/stat.h> 61 #include <sys/sysctl.h> 62 63 #include <vm/vm.h> 64 #include <vm/vm_extern.h> 65 66 #include <fs/fifofs/fifo.h> 67 68 #include <rpc/rpcclnt.h> 69 70 #include <nfs/rpcv2.h> 71 #include <nfs/nfsproto.h> 72 #include <nfsclient/nfs.h> 73 #include <nfsclient/nfsnode.h> 74 #include <nfsclient/nfsmount.h> 75 #include <nfsclient/nfs_lock.h> 76 #include <nfs/xdr_subs.h> 77 #include <nfsclient/nfsm_subs.h> 78 79 #include <net/if.h> 80 #include <netinet/in.h> 81 #include <netinet/in_var.h> 82 83 /* Defs */ 84 #define TRUE 1 85 #define FALSE 0 86 87 /* 88 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these 89 * calls are not in getblk() and brelse() so that they would not be necessary 90 * here. 91 */ 92 #ifndef B_VMIO 93 #define vfs_busy_pages(bp, f) 94 #endif 95 96 static int nfsspec_read(struct vop_read_args *); 97 static int nfsspec_write(struct vop_write_args *); 98 static int nfsfifo_read(struct vop_read_args *); 99 static int nfsfifo_write(struct vop_write_args *); 100 static int nfsspec_close(struct vop_close_args *); 101 static int nfsfifo_close(struct vop_close_args *); 102 static int nfs_flush(struct vnode *, struct ucred *, int, struct thread *, 103 int); 104 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *, 105 struct thread *); 106 static int nfs_lookup(struct vop_lookup_args *); 107 static int nfs_create(struct vop_create_args *); 108 static int nfs_mknod(struct vop_mknod_args *); 109 static int nfs_open(struct vop_open_args *); 110 static int nfs_close(struct vop_close_args *); 111 static int nfs_access(struct vop_access_args *); 112 static int nfs_getattr(struct vop_getattr_args *); 113 static int nfs_setattr(struct vop_setattr_args *); 114 static int nfs_read(struct vop_read_args *); 115 static int nfs_fsync(struct vop_fsync_args *); 116 static int nfs_remove(struct vop_remove_args *); 117 static int nfs_link(struct vop_link_args *); 118 static int nfs_rename(struct vop_rename_args *); 119 static int nfs_mkdir(struct vop_mkdir_args *); 120 static int nfs_rmdir(struct vop_rmdir_args *); 121 static int nfs_symlink(struct vop_symlink_args *); 122 static int nfs_readdir(struct vop_readdir_args *); 123 static int nfs_strategy(struct vop_strategy_args *); 124 static int nfs_lookitup(struct vnode *, const char *, int, 125 struct ucred *, struct thread *, struct nfsnode **); 126 static int nfs_sillyrename(struct vnode *, struct vnode *, 127 struct componentname *); 128 static int nfsspec_access(struct vop_access_args *); 129 static int nfs_readlink(struct vop_readlink_args *); 130 static int nfs_print(struct vop_print_args *); 131 static int nfs_advlock(struct vop_advlock_args *); 132 133 /* 134 * Global vfs data structures for nfs 135 */ 136 vop_t **nfs_vnodeop_p; 137 static struct vnodeopv_entry_desc nfs_vnodeop_entries[] = { 138 { &vop_default_desc, (vop_t *) vop_defaultop }, 139 { &vop_access_desc, (vop_t *) nfs_access }, 140 { &vop_advlock_desc, (vop_t *) nfs_advlock }, 141 { &vop_close_desc, (vop_t *) nfs_close }, 142 { &vop_create_desc, (vop_t *) nfs_create }, 143 { &vop_fsync_desc, (vop_t *) nfs_fsync }, 144 { &vop_getattr_desc, (vop_t *) nfs_getattr }, 145 { &vop_getpages_desc, (vop_t *) nfs_getpages }, 146 { &vop_putpages_desc, (vop_t *) nfs_putpages }, 147 { &vop_inactive_desc, (vop_t *) nfs_inactive }, 148 { &vop_lease_desc, (vop_t *) vop_null }, 149 { &vop_link_desc, (vop_t *) nfs_link }, 150 { &vop_lookup_desc, (vop_t *) nfs_lookup }, 151 { &vop_mkdir_desc, (vop_t *) nfs_mkdir }, 152 { &vop_mknod_desc, (vop_t *) nfs_mknod }, 153 { &vop_open_desc, (vop_t *) nfs_open }, 154 { &vop_print_desc, (vop_t *) nfs_print }, 155 { &vop_read_desc, (vop_t *) nfs_read }, 156 { &vop_readdir_desc, (vop_t *) nfs_readdir }, 157 { &vop_readlink_desc, (vop_t *) nfs_readlink }, 158 { &vop_reclaim_desc, (vop_t *) nfs_reclaim }, 159 { &vop_remove_desc, (vop_t *) nfs_remove }, 160 { &vop_rename_desc, (vop_t *) nfs_rename }, 161 { &vop_rmdir_desc, (vop_t *) nfs_rmdir }, 162 { &vop_setattr_desc, (vop_t *) nfs_setattr }, 163 { &vop_strategy_desc, (vop_t *) nfs_strategy }, 164 { &vop_symlink_desc, (vop_t *) nfs_symlink }, 165 { &vop_write_desc, (vop_t *) nfs_write }, 166 { NULL, NULL } 167 }; 168 static struct vnodeopv_desc nfs_vnodeop_opv_desc = 169 { &nfs_vnodeop_p, nfs_vnodeop_entries }; 170 VNODEOP_SET(nfs_vnodeop_opv_desc); 171 172 /* 173 * Special device vnode ops 174 */ 175 vop_t **spec_nfsnodeop_p; 176 static struct vnodeopv_entry_desc nfs_specop_entries[] = { 177 { &vop_default_desc, (vop_t *) spec_vnoperate }, 178 { &vop_access_desc, (vop_t *) nfsspec_access }, 179 { &vop_close_desc, (vop_t *) nfsspec_close }, 180 { &vop_fsync_desc, (vop_t *) nfs_fsync }, 181 { &vop_getattr_desc, (vop_t *) nfs_getattr }, 182 { &vop_inactive_desc, (vop_t *) nfs_inactive }, 183 { &vop_print_desc, (vop_t *) nfs_print }, 184 { &vop_read_desc, (vop_t *) nfsspec_read }, 185 { &vop_reclaim_desc, (vop_t *) nfs_reclaim }, 186 { &vop_setattr_desc, (vop_t *) nfs_setattr }, 187 { &vop_write_desc, (vop_t *) nfsspec_write }, 188 { NULL, NULL } 189 }; 190 static struct vnodeopv_desc spec_nfsnodeop_opv_desc = 191 { &spec_nfsnodeop_p, nfs_specop_entries }; 192 VNODEOP_SET(spec_nfsnodeop_opv_desc); 193 194 vop_t **fifo_nfsnodeop_p; 195 static struct vnodeopv_entry_desc nfs_fifoop_entries[] = { 196 { &vop_default_desc, (vop_t *) fifo_vnoperate }, 197 { &vop_access_desc, (vop_t *) nfsspec_access }, 198 { &vop_close_desc, (vop_t *) nfsfifo_close }, 199 { &vop_fsync_desc, (vop_t *) nfs_fsync }, 200 { &vop_getattr_desc, (vop_t *) nfs_getattr }, 201 { &vop_inactive_desc, (vop_t *) nfs_inactive }, 202 { &vop_print_desc, (vop_t *) nfs_print }, 203 { &vop_read_desc, (vop_t *) nfsfifo_read }, 204 { &vop_reclaim_desc, (vop_t *) nfs_reclaim }, 205 { &vop_setattr_desc, (vop_t *) nfs_setattr }, 206 { &vop_write_desc, (vop_t *) nfsfifo_write }, 207 { NULL, NULL } 208 }; 209 static struct vnodeopv_desc fifo_nfsnodeop_opv_desc = 210 { &fifo_nfsnodeop_p, nfs_fifoop_entries }; 211 VNODEOP_SET(fifo_nfsnodeop_opv_desc); 212 213 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, 214 struct componentname *cnp, struct vattr *vap); 215 static int nfs_removerpc(struct vnode *dvp, const char *name, int namelen, 216 struct ucred *cred, struct thread *td); 217 static int nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, 218 int fnamelen, struct vnode *tdvp, 219 const char *tnameptr, int tnamelen, 220 struct ucred *cred, struct thread *td); 221 static int nfs_renameit(struct vnode *sdvp, struct componentname *scnp, 222 struct sillyrename *sp); 223 224 /* 225 * Global variables 226 */ 227 struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON]; 228 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON]; 229 int nfs_numasync = 0; 230 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1)) 231 232 SYSCTL_DECL(_vfs_nfs); 233 234 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO; 235 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 236 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 237 238 static int nfsv3_commit_on_close = 0; 239 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW, 240 &nfsv3_commit_on_close, 0, "write+commit on close, else only write"); 241 #if 0 242 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD, 243 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count"); 244 245 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD, 246 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count"); 247 #endif 248 249 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \ 250 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \ 251 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP) 252 static int 253 nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td, 254 struct ucred *cred) 255 { 256 const int v3 = 1; 257 u_int32_t *tl; 258 int error = 0, attrflag; 259 260 struct mbuf *mreq, *mrep, *md, *mb; 261 caddr_t bpos, dpos; 262 u_int32_t rmode; 263 struct nfsnode *np = VTONFS(vp); 264 265 nfsstats.rpccnt[NFSPROC_ACCESS]++; 266 mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 267 mb = mreq; 268 bpos = mtod(mb, caddr_t); 269 nfsm_fhtom(vp, v3); 270 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED); 271 *tl = txdr_unsigned(wmode); 272 nfsm_request(vp, NFSPROC_ACCESS, td, cred); 273 nfsm_postop_attr(vp, attrflag); 274 if (!error) { 275 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); 276 rmode = fxdr_unsigned(u_int32_t, *tl); 277 np->n_mode = rmode; 278 np->n_modeuid = cred->cr_uid; 279 np->n_modestamp = time_second; 280 } 281 m_freem(mrep); 282 nfsmout: 283 return (error); 284 } 285 286 /* 287 * nfs access vnode op. 288 * For nfs version 2, just return ok. File accesses may fail later. 289 * For nfs version 3, use the access rpc to check accessibility. If file modes 290 * are changed on the server, accesses might still fail later. 291 */ 292 static int 293 nfs_access(struct vop_access_args *ap) 294 { 295 struct vnode *vp = ap->a_vp; 296 int error = 0; 297 u_int32_t mode, wmode; 298 int v3 = NFS_ISV3(vp); 299 struct nfsnode *np = VTONFS(vp); 300 301 /* 302 * Disallow write attempts on filesystems mounted read-only; 303 * unless the file is a socket, fifo, or a block or character 304 * device resident on the filesystem. 305 */ 306 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 307 switch (vp->v_type) { 308 case VREG: 309 case VDIR: 310 case VLNK: 311 return (EROFS); 312 default: 313 break; 314 } 315 } 316 /* 317 * For nfs v3, check to see if we have done this recently, and if 318 * so return our cached result instead of making an ACCESS call. 319 * If not, do an access rpc, otherwise you are stuck emulating 320 * ufs_access() locally using the vattr. This may not be correct, 321 * since the server may apply other access criteria such as 322 * client uid-->server uid mapping that we do not know about. 323 */ 324 if (v3) { 325 if (ap->a_mode & VREAD) 326 mode = NFSV3ACCESS_READ; 327 else 328 mode = 0; 329 if (vp->v_type != VDIR) { 330 if (ap->a_mode & VWRITE) 331 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 332 if (ap->a_mode & VEXEC) 333 mode |= NFSV3ACCESS_EXECUTE; 334 } else { 335 if (ap->a_mode & VWRITE) 336 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 337 NFSV3ACCESS_DELETE); 338 if (ap->a_mode & VEXEC) 339 mode |= NFSV3ACCESS_LOOKUP; 340 } 341 /* XXX safety belt, only make blanket request if caching */ 342 if (nfsaccess_cache_timeout > 0) { 343 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY | 344 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE | 345 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP; 346 } else { 347 wmode = mode; 348 } 349 350 /* 351 * Does our cached result allow us to give a definite yes to 352 * this request? 353 */ 354 if ((time_second < (np->n_modestamp + nfsaccess_cache_timeout)) && 355 (ap->a_cred->cr_uid == np->n_modeuid) && 356 ((np->n_mode & mode) == mode)) { 357 nfsstats.accesscache_hits++; 358 } else { 359 /* 360 * Either a no, or a don't know. Go to the wire. 361 */ 362 nfsstats.accesscache_misses++; 363 error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred); 364 if (!error) { 365 if ((np->n_mode & mode) != mode) { 366 error = EACCES; 367 } 368 } 369 } 370 return (error); 371 } else { 372 if ((error = nfsspec_access(ap)) != 0) 373 return (error); 374 375 /* 376 * Attempt to prevent a mapped root from accessing a file 377 * which it shouldn't. We try to read a byte from the file 378 * if the user is root and the file is not zero length. 379 * After calling nfsspec_access, we should have the correct 380 * file size cached. 381 */ 382 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD) 383 && VTONFS(vp)->n_size > 0) { 384 struct iovec aiov; 385 struct uio auio; 386 char buf[1]; 387 388 aiov.iov_base = buf; 389 aiov.iov_len = 1; 390 auio.uio_iov = &aiov; 391 auio.uio_iovcnt = 1; 392 auio.uio_offset = 0; 393 auio.uio_resid = 1; 394 auio.uio_segflg = UIO_SYSSPACE; 395 auio.uio_rw = UIO_READ; 396 auio.uio_td = ap->a_td; 397 398 if (vp->v_type == VREG) 399 error = nfs_readrpc(vp, &auio, ap->a_cred); 400 else if (vp->v_type == VDIR) { 401 char* bp; 402 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 403 aiov.iov_base = bp; 404 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 405 error = nfs_readdirrpc(vp, &auio, ap->a_cred); 406 free(bp, M_TEMP); 407 } else if (vp->v_type == VLNK) 408 error = nfs_readlinkrpc(vp, &auio, ap->a_cred); 409 else 410 error = EACCES; 411 } 412 return (error); 413 } 414 } 415 416 /* 417 * nfs open vnode op 418 * Check to see if the type is ok 419 * and that deletion is not in progress. 420 * For paged in text files, you will need to flush the page cache 421 * if consistency is lost. 422 */ 423 /* ARGSUSED */ 424 static int 425 nfs_open(struct vop_open_args *ap) 426 { 427 struct vnode *vp = ap->a_vp; 428 struct nfsnode *np = VTONFS(vp); 429 struct vattr vattr; 430 int error; 431 432 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 433 #ifdef DIAGNOSTIC 434 printf("open eacces vtyp=%d\n", vp->v_type); 435 #endif 436 return (EOPNOTSUPP); 437 } 438 /* 439 * Get a valid lease. If cached data is stale, flush it. 440 */ 441 if (np->n_flag & NMODIFIED) { 442 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_td, 1); 443 if (error == EINTR || error == EIO) 444 return (error); 445 np->n_attrstamp = 0; 446 if (vp->v_type == VDIR) 447 np->n_direofoffset = 0; 448 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td); 449 if (error) 450 return (error); 451 np->n_mtime = vattr.va_mtime.tv_sec; 452 } else { 453 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td); 454 if (error) 455 return (error); 456 if (np->n_mtime != vattr.va_mtime.tv_sec) { 457 if (vp->v_type == VDIR) 458 np->n_direofoffset = 0; 459 error = nfs_vinvalbuf(vp, V_SAVE, 460 ap->a_cred, ap->a_td, 1); 461 if (error == EINTR || error == EIO) 462 return (error); 463 np->n_mtime = vattr.va_mtime.tv_sec; 464 } 465 } 466 np->n_attrstamp = 0; /* For Open/Close consistency */ 467 return (0); 468 } 469 470 /* 471 * nfs close vnode op 472 * What an NFS client should do upon close after writing is a debatable issue. 473 * Most NFS clients push delayed writes to the server upon close, basically for 474 * two reasons: 475 * 1 - So that any write errors may be reported back to the client process 476 * doing the close system call. By far the two most likely errors are 477 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 478 * 2 - To put a worst case upper bound on cache inconsistency between 479 * multiple clients for the file. 480 * There is also a consistency problem for Version 2 of the protocol w.r.t. 481 * not being able to tell if other clients are writing a file concurrently, 482 * since there is no way of knowing if the changed modify time in the reply 483 * is only due to the write for this client. 484 * (NFS Version 3 provides weak cache consistency data in the reply that 485 * should be sufficient to detect and handle this case.) 486 * 487 * The current code does the following: 488 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 489 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 490 * or commit them (this satisfies 1 and 2 except for the 491 * case where the server crashes after this close but 492 * before the commit RPC, which is felt to be "good 493 * enough". Changing the last argument to nfs_flush() to 494 * a 1 would force a commit operation, if it is felt a 495 * commit is necessary now. 496 */ 497 /* ARGSUSED */ 498 static int 499 nfs_close(struct vop_close_args *ap) 500 { 501 struct vnode *vp = ap->a_vp; 502 struct nfsnode *np = VTONFS(vp); 503 int error = 0; 504 505 if (vp->v_type == VREG) { 506 if (np->n_flag & NMODIFIED) { 507 if (NFS_ISV3(vp)) { 508 /* 509 * Under NFSv3 we have dirty buffers to dispose of. We 510 * must flush them to the NFS server. We have the option 511 * of waiting all the way through the commit rpc or just 512 * waiting for the initial write. The default is to only 513 * wait through the initial write so the data is in the 514 * server's cache, which is roughly similar to the state 515 * a standard disk subsystem leaves the file in on close(). 516 * 517 * We cannot clear the NMODIFIED bit in np->n_flag due to 518 * potential races with other processes, and certainly 519 * cannot clear it if we don't commit. 520 */ 521 int cm = nfsv3_commit_on_close ? 1 : 0; 522 error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_td, cm); 523 /* np->n_flag &= ~NMODIFIED; */ 524 } else { 525 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, ap->a_td); 526 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_td, 1); 527 VOP_UNLOCK(vp, 0, ap->a_td); 528 } 529 np->n_attrstamp = 0; 530 } 531 if (np->n_flag & NWRITEERR) { 532 np->n_flag &= ~NWRITEERR; 533 error = np->n_error; 534 } 535 } 536 return (error); 537 } 538 539 /* 540 * nfs getattr call from vfs. 541 */ 542 static int 543 nfs_getattr(struct vop_getattr_args *ap) 544 { 545 struct vnode *vp = ap->a_vp; 546 struct nfsnode *np = VTONFS(vp); 547 caddr_t bpos, dpos; 548 int error = 0; 549 struct mbuf *mreq, *mrep, *md, *mb; 550 int v3 = NFS_ISV3(vp); 551 552 /* 553 * Update local times for special files. 554 */ 555 if (np->n_flag & (NACC | NUPD)) 556 np->n_flag |= NCHG; 557 /* 558 * First look in the cache. 559 */ 560 if (nfs_getattrcache(vp, ap->a_vap) == 0) 561 return (0); 562 563 if (v3 && nfsaccess_cache_timeout > 0) { 564 nfsstats.accesscache_misses++; 565 nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, ap->a_cred); 566 if (nfs_getattrcache(vp, ap->a_vap) == 0) 567 return (0); 568 } 569 570 nfsstats.rpccnt[NFSPROC_GETATTR]++; 571 mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3)); 572 mb = mreq; 573 bpos = mtod(mb, caddr_t); 574 nfsm_fhtom(vp, v3); 575 nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, ap->a_cred); 576 if (!error) { 577 nfsm_loadattr(vp, ap->a_vap); 578 } 579 m_freem(mrep); 580 nfsmout: 581 return (error); 582 } 583 584 /* 585 * nfs setattr call. 586 */ 587 static int 588 nfs_setattr(struct vop_setattr_args *ap) 589 { 590 struct vnode *vp = ap->a_vp; 591 struct nfsnode *np = VTONFS(vp); 592 struct vattr *vap = ap->a_vap; 593 int error = 0; 594 u_quad_t tsize; 595 596 #ifndef nolint 597 tsize = (u_quad_t)0; 598 #endif 599 600 /* 601 * Setting of flags is not supported. 602 */ 603 if (vap->va_flags != VNOVAL) 604 return (EOPNOTSUPP); 605 606 /* 607 * Disallow write attempts if the filesystem is mounted read-only. 608 */ 609 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 610 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 611 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 612 (vp->v_mount->mnt_flag & MNT_RDONLY)) 613 return (EROFS); 614 if (vap->va_size != VNOVAL) { 615 switch (vp->v_type) { 616 case VDIR: 617 return (EISDIR); 618 case VCHR: 619 case VBLK: 620 case VSOCK: 621 case VFIFO: 622 if (vap->va_mtime.tv_sec == VNOVAL && 623 vap->va_atime.tv_sec == VNOVAL && 624 vap->va_mode == (mode_t)VNOVAL && 625 vap->va_uid == (uid_t)VNOVAL && 626 vap->va_gid == (gid_t)VNOVAL) 627 return (0); 628 vap->va_size = VNOVAL; 629 break; 630 default: 631 /* 632 * Disallow write attempts if the filesystem is 633 * mounted read-only. 634 */ 635 if (vp->v_mount->mnt_flag & MNT_RDONLY) 636 return (EROFS); 637 638 /* 639 * We run vnode_pager_setsize() early (why?), 640 * we must set np->n_size now to avoid vinvalbuf 641 * V_SAVE races that might setsize a lower 642 * value. 643 */ 644 645 tsize = np->n_size; 646 error = nfs_meta_setsize(vp, ap->a_cred, 647 ap->a_td, vap->va_size); 648 649 if (np->n_flag & NMODIFIED) { 650 if (vap->va_size == 0) 651 error = nfs_vinvalbuf(vp, 0, 652 ap->a_cred, ap->a_td, 1); 653 else 654 error = nfs_vinvalbuf(vp, V_SAVE, 655 ap->a_cred, ap->a_td, 1); 656 if (error) { 657 vnode_pager_setsize(vp, np->n_size); 658 return (error); 659 } 660 } 661 /* 662 * np->n_size has already been set to vap->va_size 663 * in nfs_meta_setsize(). We must set it again since 664 * nfs_loadattrcache() could be called through 665 * nfs_meta_setsize() and could modify np->n_size. 666 */ 667 np->n_vattr.va_size = np->n_size = vap->va_size; 668 }; 669 } else if ((vap->va_mtime.tv_sec != VNOVAL || 670 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NMODIFIED) && 671 vp->v_type == VREG && 672 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, 673 ap->a_td, 1)) != 0 && (error == EINTR || error == EIO)) 674 return (error); 675 error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td); 676 if (error && vap->va_size != VNOVAL) { 677 np->n_size = np->n_vattr.va_size = tsize; 678 vnode_pager_setsize(vp, np->n_size); 679 } 680 return (error); 681 } 682 683 /* 684 * Do an nfs setattr rpc. 685 */ 686 static int 687 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred, 688 struct thread *td) 689 { 690 struct nfsv2_sattr *sp; 691 struct nfsnode *np = VTONFS(vp); 692 caddr_t bpos, dpos; 693 u_int32_t *tl; 694 int error = 0, wccflag = NFSV3_WCCRATTR; 695 struct mbuf *mreq, *mrep, *md, *mb; 696 int v3 = NFS_ISV3(vp); 697 698 nfsstats.rpccnt[NFSPROC_SETATTR]++; 699 mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 700 mb = mreq; 701 bpos = mtod(mb, caddr_t); 702 nfsm_fhtom(vp, v3); 703 if (v3) { 704 nfsm_v3attrbuild(vap, TRUE); 705 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED); 706 *tl = nfs_false; 707 } else { 708 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR); 709 if (vap->va_mode == (mode_t)VNOVAL) 710 sp->sa_mode = nfs_xdrneg1; 711 else 712 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 713 if (vap->va_uid == (uid_t)VNOVAL) 714 sp->sa_uid = nfs_xdrneg1; 715 else 716 sp->sa_uid = txdr_unsigned(vap->va_uid); 717 if (vap->va_gid == (gid_t)VNOVAL) 718 sp->sa_gid = nfs_xdrneg1; 719 else 720 sp->sa_gid = txdr_unsigned(vap->va_gid); 721 sp->sa_size = txdr_unsigned(vap->va_size); 722 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 723 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 724 } 725 nfsm_request(vp, NFSPROC_SETATTR, td, cred); 726 if (v3) { 727 np->n_modestamp = 0; 728 nfsm_wcc_data(vp, wccflag); 729 } else 730 nfsm_loadattr(vp, NULL); 731 m_freem(mrep); 732 nfsmout: 733 return (error); 734 } 735 736 /* 737 * nfs lookup call, one step at a time... 738 * First look in cache 739 * If not found, unlock the directory nfsnode and do the rpc 740 */ 741 static int 742 nfs_lookup(struct vop_lookup_args *ap) 743 { 744 struct componentname *cnp = ap->a_cnp; 745 struct vnode *dvp = ap->a_dvp; 746 struct vnode **vpp = ap->a_vpp; 747 int flags = cnp->cn_flags; 748 struct vnode *newvp; 749 struct nfsmount *nmp; 750 caddr_t bpos, dpos; 751 struct mbuf *mreq, *mrep, *md, *mb; 752 long len; 753 nfsfh_t *fhp; 754 struct nfsnode *np; 755 int lockparent, wantparent, error = 0, attrflag, fhsize; 756 int v3 = NFS_ISV3(dvp); 757 struct thread *td = cnp->cn_thread; 758 759 *vpp = NULLVP; 760 cnp->cn_flags &= ~PDIRUNLOCK; 761 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 762 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 763 return (EROFS); 764 if (dvp->v_type != VDIR) 765 return (ENOTDIR); 766 lockparent = flags & LOCKPARENT; 767 wantparent = flags & (LOCKPARENT|WANTPARENT); 768 nmp = VFSTONFS(dvp->v_mount); 769 np = VTONFS(dvp); 770 if ((error = cache_lookup(dvp, vpp, cnp)) && error != ENOENT) { 771 struct vattr vattr; 772 int vpid; 773 774 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) { 775 *vpp = NULLVP; 776 return (error); 777 } 778 779 vhold(*vpp); 780 newvp = *vpp; 781 vpid = newvp->v_id; 782 /* 783 * See the comment starting `Step through' in ufs/ufs_lookup.c 784 * for an explanation of the locking protocol 785 */ 786 if (dvp == newvp) { 787 VREF(newvp); 788 error = 0; 789 } else if (flags & ISDOTDOT) { 790 VOP_UNLOCK(dvp, 0, td); 791 cnp->cn_flags |= PDIRUNLOCK; 792 error = vget(newvp, LK_EXCLUSIVE, td); 793 if (!error && lockparent && (flags & ISLASTCN)) { 794 error = vn_lock(dvp, LK_EXCLUSIVE, td); 795 if (error == 0) 796 cnp->cn_flags &= ~PDIRUNLOCK; 797 } 798 } else { 799 error = vget(newvp, LK_EXCLUSIVE, td); 800 if (!lockparent || error || !(flags & ISLASTCN)) { 801 VOP_UNLOCK(dvp, 0, td); 802 cnp->cn_flags |= PDIRUNLOCK; 803 } 804 } 805 if (!error) { 806 if (vpid == newvp->v_id) { 807 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, td) 808 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) { 809 nfsstats.lookupcache_hits++; 810 if (cnp->cn_nameiop != LOOKUP && 811 (flags & ISLASTCN)) 812 cnp->cn_flags |= SAVENAME; 813 vdrop(newvp); 814 return (0); 815 } 816 cache_purge(newvp); 817 } 818 vput(newvp); 819 if (lockparent && dvp != newvp && (flags & ISLASTCN)) 820 VOP_UNLOCK(dvp, 0, td); 821 } 822 vdrop(newvp); 823 error = vn_lock(dvp, LK_EXCLUSIVE, td); 824 *vpp = NULLVP; 825 if (error) { 826 cnp->cn_flags |= PDIRUNLOCK; 827 return (error); 828 } 829 cnp->cn_flags &= ~PDIRUNLOCK; 830 } 831 error = 0; 832 newvp = NULLVP; 833 nfsstats.lookupcache_misses++; 834 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 835 len = cnp->cn_namelen; 836 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP, 837 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 838 mb = mreq; 839 bpos = mtod(mb, caddr_t); 840 nfsm_fhtom(dvp, v3); 841 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 842 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred); 843 if (error) { 844 if (v3) { 845 nfsm_postop_attr(dvp, attrflag); 846 m_freem(mrep); 847 } 848 goto nfsmout; 849 } 850 nfsm_getfh(fhp, fhsize, v3); 851 852 /* 853 * Handle RENAME case... 854 */ 855 if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) { 856 if (NFS_CMPFH(np, fhp, fhsize)) { 857 m_freem(mrep); 858 return (EISDIR); 859 } 860 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 861 if (error) { 862 m_freem(mrep); 863 return (error); 864 } 865 newvp = NFSTOV(np); 866 if (v3) { 867 nfsm_postop_attr(newvp, attrflag); 868 nfsm_postop_attr(dvp, attrflag); 869 } else 870 nfsm_loadattr(newvp, NULL); 871 *vpp = newvp; 872 m_freem(mrep); 873 cnp->cn_flags |= SAVENAME; 874 if (!lockparent) { 875 VOP_UNLOCK(dvp, 0, td); 876 cnp->cn_flags |= PDIRUNLOCK; 877 } 878 return (0); 879 } 880 881 if (flags & ISDOTDOT) { 882 VOP_UNLOCK(dvp, 0, td); 883 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 884 if (error) { 885 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td); 886 return (error); 887 } 888 newvp = NFSTOV(np); 889 if (lockparent && (flags & ISLASTCN)) { 890 error = vn_lock(dvp, LK_EXCLUSIVE, td); 891 if (error) { 892 cnp->cn_flags |= PDIRUNLOCK; 893 vput(newvp); 894 return (error); 895 } 896 } else 897 cnp->cn_flags |= PDIRUNLOCK; 898 } else if (NFS_CMPFH(np, fhp, fhsize)) { 899 VREF(dvp); 900 newvp = dvp; 901 } else { 902 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 903 if (error) { 904 m_freem(mrep); 905 return (error); 906 } 907 if (!lockparent || !(flags & ISLASTCN)) { 908 cnp->cn_flags |= PDIRUNLOCK; 909 VOP_UNLOCK(dvp, 0, td); 910 } 911 newvp = NFSTOV(np); 912 } 913 if (v3) { 914 nfsm_postop_attr(newvp, attrflag); 915 nfsm_postop_attr(dvp, attrflag); 916 } else 917 nfsm_loadattr(newvp, NULL); 918 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 919 cnp->cn_flags |= SAVENAME; 920 if ((cnp->cn_flags & MAKEENTRY) && 921 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) { 922 np->n_ctime = np->n_vattr.va_ctime.tv_sec; 923 cache_enter(dvp, newvp, cnp); 924 } 925 *vpp = newvp; 926 m_freem(mrep); 927 nfsmout: 928 if (error) { 929 if (newvp != NULLVP) { 930 vrele(newvp); 931 *vpp = NULLVP; 932 } 933 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 934 (flags & ISLASTCN) && error == ENOENT) { 935 if (!lockparent) { 936 VOP_UNLOCK(dvp, 0, td); 937 cnp->cn_flags |= PDIRUNLOCK; 938 } 939 if (dvp->v_mount->mnt_flag & MNT_RDONLY) 940 error = EROFS; 941 else 942 error = EJUSTRETURN; 943 } 944 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 945 cnp->cn_flags |= SAVENAME; 946 } 947 return (error); 948 } 949 950 /* 951 * nfs read call. 952 * Just call nfs_bioread() to do the work. 953 */ 954 static int 955 nfs_read(struct vop_read_args *ap) 956 { 957 struct vnode *vp = ap->a_vp; 958 959 switch (vp->v_type) { 960 case VREG: 961 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred)); 962 case VDIR: 963 return (EISDIR); 964 default: 965 return (EOPNOTSUPP); 966 } 967 } 968 969 /* 970 * nfs readlink call 971 */ 972 static int 973 nfs_readlink(struct vop_readlink_args *ap) 974 { 975 struct vnode *vp = ap->a_vp; 976 977 if (vp->v_type != VLNK) 978 return (EINVAL); 979 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred)); 980 } 981 982 /* 983 * Do a readlink rpc. 984 * Called by nfs_doio() from below the buffer cache. 985 */ 986 int 987 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 988 { 989 caddr_t bpos, dpos; 990 int error = 0, len, attrflag; 991 struct mbuf *mreq, *mrep, *md, *mb; 992 int v3 = NFS_ISV3(vp); 993 994 nfsstats.rpccnt[NFSPROC_READLINK]++; 995 mreq = nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3)); 996 mb = mreq; 997 bpos = mtod(mb, caddr_t); 998 nfsm_fhtom(vp, v3); 999 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred); 1000 if (v3) 1001 nfsm_postop_attr(vp, attrflag); 1002 if (!error) { 1003 nfsm_strsiz(len, NFS_MAXPATHLEN); 1004 if (len == NFS_MAXPATHLEN) { 1005 struct nfsnode *np = VTONFS(vp); 1006 if (np->n_size && np->n_size < NFS_MAXPATHLEN) 1007 len = np->n_size; 1008 } 1009 nfsm_mtouio(uiop, len); 1010 } 1011 m_freem(mrep); 1012 nfsmout: 1013 return (error); 1014 } 1015 1016 /* 1017 * nfs read rpc call 1018 * Ditto above 1019 */ 1020 int 1021 nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1022 { 1023 u_int32_t *tl; 1024 caddr_t bpos, dpos; 1025 struct mbuf *mreq, *mrep, *md, *mb; 1026 struct nfsmount *nmp; 1027 int error = 0, len, retlen, tsiz, eof, attrflag; 1028 int v3 = NFS_ISV3(vp); 1029 1030 #ifndef nolint 1031 eof = 0; 1032 #endif 1033 nmp = VFSTONFS(vp->v_mount); 1034 tsiz = uiop->uio_resid; 1035 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1036 return (EFBIG); 1037 while (tsiz > 0) { 1038 nfsstats.rpccnt[NFSPROC_READ]++; 1039 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1040 mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1041 mb = mreq; 1042 bpos = mtod(mb, caddr_t); 1043 nfsm_fhtom(vp, v3); 1044 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3); 1045 if (v3) { 1046 txdr_hyper(uiop->uio_offset, tl); 1047 *(tl + 2) = txdr_unsigned(len); 1048 } else { 1049 *tl++ = txdr_unsigned(uiop->uio_offset); 1050 *tl++ = txdr_unsigned(len); 1051 *tl = 0; 1052 } 1053 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred); 1054 if (v3) { 1055 nfsm_postop_attr(vp, attrflag); 1056 if (error) { 1057 m_freem(mrep); 1058 goto nfsmout; 1059 } 1060 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED); 1061 eof = fxdr_unsigned(int, *(tl + 1)); 1062 } else 1063 nfsm_loadattr(vp, NULL); 1064 nfsm_strsiz(retlen, nmp->nm_rsize); 1065 nfsm_mtouio(uiop, retlen); 1066 m_freem(mrep); 1067 tsiz -= retlen; 1068 if (v3) { 1069 if (eof || retlen == 0) { 1070 tsiz = 0; 1071 } 1072 } else if (retlen < len) { 1073 tsiz = 0; 1074 } 1075 } 1076 nfsmout: 1077 return (error); 1078 } 1079 1080 /* 1081 * nfs write call 1082 */ 1083 int 1084 nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 1085 int *iomode, int *must_commit) 1086 { 1087 u_int32_t *tl; 1088 int32_t backup; 1089 caddr_t bpos, dpos; 1090 struct mbuf *mreq, *mrep, *md, *mb; 1091 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1092 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit; 1093 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC; 1094 1095 #ifndef DIAGNOSTIC 1096 if (uiop->uio_iovcnt != 1) 1097 panic("nfs: writerpc iovcnt > 1"); 1098 #endif 1099 *must_commit = 0; 1100 tsiz = uiop->uio_resid; 1101 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1102 return (EFBIG); 1103 while (tsiz > 0) { 1104 nfsstats.rpccnt[NFSPROC_WRITE]++; 1105 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz; 1106 mreq = nfsm_reqhead(vp, NFSPROC_WRITE, 1107 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len)); 1108 mb = mreq; 1109 bpos = mtod(mb, caddr_t); 1110 nfsm_fhtom(vp, v3); 1111 if (v3) { 1112 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED); 1113 txdr_hyper(uiop->uio_offset, tl); 1114 tl += 2; 1115 *tl++ = txdr_unsigned(len); 1116 *tl++ = txdr_unsigned(*iomode); 1117 *tl = txdr_unsigned(len); 1118 } else { 1119 u_int32_t x; 1120 1121 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED); 1122 /* Set both "begin" and "current" to non-garbage. */ 1123 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1124 *tl++ = x; /* "begin offset" */ 1125 *tl++ = x; /* "current offset" */ 1126 x = txdr_unsigned(len); 1127 *tl++ = x; /* total to this offset */ 1128 *tl = x; /* size of this write */ 1129 } 1130 nfsm_uiotom(uiop, len); 1131 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred); 1132 if (v3) { 1133 wccflag = NFSV3_WCCCHK; 1134 nfsm_wcc_data(vp, wccflag); 1135 if (!error) { 1136 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED 1137 + NFSX_V3WRITEVERF); 1138 rlen = fxdr_unsigned(int, *tl++); 1139 if (rlen == 0) { 1140 error = NFSERR_IO; 1141 m_freem(mrep); 1142 break; 1143 } else if (rlen < len) { 1144 backup = len - rlen; 1145 uiop->uio_iov->iov_base = 1146 (char *)uiop->uio_iov->iov_base - 1147 backup; 1148 uiop->uio_iov->iov_len += backup; 1149 uiop->uio_offset -= backup; 1150 uiop->uio_resid += backup; 1151 len = rlen; 1152 } 1153 commit = fxdr_unsigned(int, *tl++); 1154 1155 /* 1156 * Return the lowest committment level 1157 * obtained by any of the RPCs. 1158 */ 1159 if (committed == NFSV3WRITE_FILESYNC) 1160 committed = commit; 1161 else if (committed == NFSV3WRITE_DATASYNC && 1162 commit == NFSV3WRITE_UNSTABLE) 1163 committed = commit; 1164 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){ 1165 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1166 NFSX_V3WRITEVERF); 1167 nmp->nm_state |= NFSSTA_HASWRITEVERF; 1168 } else if (bcmp((caddr_t)tl, 1169 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) { 1170 *must_commit = 1; 1171 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1172 NFSX_V3WRITEVERF); 1173 } 1174 } 1175 } else 1176 nfsm_loadattr(vp, NULL); 1177 if (wccflag) 1178 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime.tv_sec; 1179 m_freem(mrep); 1180 if (error) 1181 break; 1182 tsiz -= len; 1183 } 1184 nfsmout: 1185 if (vp->v_mount->mnt_flag & MNT_ASYNC) 1186 committed = NFSV3WRITE_FILESYNC; 1187 *iomode = committed; 1188 if (error) 1189 uiop->uio_resid = tsiz; 1190 return (error); 1191 } 1192 1193 /* 1194 * nfs mknod rpc 1195 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1196 * mode set to specify the file type and the size field for rdev. 1197 */ 1198 static int 1199 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1200 struct vattr *vap) 1201 { 1202 struct nfsv2_sattr *sp; 1203 u_int32_t *tl; 1204 struct vnode *newvp = NULL; 1205 struct nfsnode *np = NULL; 1206 struct vattr vattr; 1207 caddr_t bpos, dpos; 1208 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1209 struct mbuf *mreq, *mrep, *md, *mb; 1210 u_int32_t rdev; 1211 int v3 = NFS_ISV3(dvp); 1212 1213 if (vap->va_type == VCHR || vap->va_type == VBLK) 1214 rdev = txdr_unsigned(vap->va_rdev); 1215 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1216 rdev = nfs_xdrneg1; 1217 else { 1218 return (EOPNOTSUPP); 1219 } 1220 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) { 1221 return (error); 1222 } 1223 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1224 mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1225 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1226 mb = mreq; 1227 bpos = mtod(mb, caddr_t); 1228 nfsm_fhtom(dvp, v3); 1229 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1230 if (v3) { 1231 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED); 1232 *tl++ = vtonfsv3_type(vap->va_type); 1233 nfsm_v3attrbuild(vap, FALSE); 1234 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1235 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED); 1236 *tl++ = txdr_unsigned(umajor(vap->va_rdev)); 1237 *tl = txdr_unsigned(uminor(vap->va_rdev)); 1238 } 1239 } else { 1240 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR); 1241 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1242 sp->sa_uid = nfs_xdrneg1; 1243 sp->sa_gid = nfs_xdrneg1; 1244 sp->sa_size = rdev; 1245 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1246 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1247 } 1248 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred); 1249 if (!error) { 1250 nfsm_mtofh(dvp, newvp, v3, gotvp); 1251 if (!gotvp) { 1252 if (newvp) { 1253 vput(newvp); 1254 newvp = NULL; 1255 } 1256 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1257 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np); 1258 if (!error) 1259 newvp = NFSTOV(np); 1260 } 1261 } 1262 if (v3) 1263 nfsm_wcc_data(dvp, wccflag); 1264 m_freem(mrep); 1265 nfsmout: 1266 if (error) { 1267 if (newvp) 1268 vput(newvp); 1269 } else { 1270 if (cnp->cn_flags & MAKEENTRY) 1271 cache_enter(dvp, newvp, cnp); 1272 *vpp = newvp; 1273 } 1274 VTONFS(dvp)->n_flag |= NMODIFIED; 1275 if (!wccflag) 1276 VTONFS(dvp)->n_attrstamp = 0; 1277 return (error); 1278 } 1279 1280 /* 1281 * nfs mknod vop 1282 * just call nfs_mknodrpc() to do the work. 1283 */ 1284 /* ARGSUSED */ 1285 static int 1286 nfs_mknod(struct vop_mknod_args *ap) 1287 { 1288 1289 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap)); 1290 } 1291 1292 static u_long create_verf; 1293 /* 1294 * nfs file create call 1295 */ 1296 static int 1297 nfs_create(struct vop_create_args *ap) 1298 { 1299 struct vnode *dvp = ap->a_dvp; 1300 struct vattr *vap = ap->a_vap; 1301 struct componentname *cnp = ap->a_cnp; 1302 struct nfsv2_sattr *sp; 1303 u_int32_t *tl; 1304 struct nfsnode *np = NULL; 1305 struct vnode *newvp = NULL; 1306 caddr_t bpos, dpos; 1307 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0; 1308 struct mbuf *mreq, *mrep, *md, *mb; 1309 struct vattr vattr; 1310 int v3 = NFS_ISV3(dvp); 1311 1312 /* 1313 * Oops, not for me.. 1314 */ 1315 if (vap->va_type == VSOCK) 1316 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1317 1318 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) { 1319 return (error); 1320 } 1321 if (vap->va_vaflags & VA_EXCLUSIVE) 1322 fmode |= O_EXCL; 1323 again: 1324 nfsstats.rpccnt[NFSPROC_CREATE]++; 1325 mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1326 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1327 mb = mreq; 1328 bpos = mtod(mb, caddr_t); 1329 nfsm_fhtom(dvp, v3); 1330 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1331 if (v3) { 1332 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED); 1333 if (fmode & O_EXCL) { 1334 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1335 tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF); 1336 #ifdef INET 1337 if (!TAILQ_EMPTY(&in_ifaddrhead)) 1338 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr; 1339 else 1340 #endif 1341 *tl++ = create_verf; 1342 *tl = ++create_verf; 1343 } else { 1344 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED); 1345 nfsm_v3attrbuild(vap, FALSE); 1346 } 1347 } else { 1348 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR); 1349 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1350 sp->sa_uid = nfs_xdrneg1; 1351 sp->sa_gid = nfs_xdrneg1; 1352 sp->sa_size = 0; 1353 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1354 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1355 } 1356 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred); 1357 if (!error) { 1358 nfsm_mtofh(dvp, newvp, v3, gotvp); 1359 if (!gotvp) { 1360 if (newvp) { 1361 vput(newvp); 1362 newvp = NULL; 1363 } 1364 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1365 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np); 1366 if (!error) 1367 newvp = NFSTOV(np); 1368 } 1369 } 1370 if (v3) 1371 nfsm_wcc_data(dvp, wccflag); 1372 m_freem(mrep); 1373 nfsmout: 1374 if (error) { 1375 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) { 1376 fmode &= ~O_EXCL; 1377 goto again; 1378 } 1379 if (newvp) 1380 vput(newvp); 1381 } else if (v3 && (fmode & O_EXCL)) { 1382 /* 1383 * We are normally called with only a partially initialized 1384 * VAP. Since the NFSv3 spec says that server may use the 1385 * file attributes to store the verifier, the spec requires 1386 * us to do a SETATTR RPC. FreeBSD servers store the verifier 1387 * in atime, but we can't really assume that all servers will 1388 * so we ensure that our SETATTR sets both atime and mtime. 1389 */ 1390 if (vap->va_mtime.tv_sec == VNOVAL) 1391 vfs_timestamp(&vap->va_mtime); 1392 if (vap->va_atime.tv_sec == VNOVAL) 1393 vap->va_atime = vap->va_mtime; 1394 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_thread); 1395 } 1396 if (!error) { 1397 if (cnp->cn_flags & MAKEENTRY) 1398 cache_enter(dvp, newvp, cnp); 1399 *ap->a_vpp = newvp; 1400 } 1401 VTONFS(dvp)->n_flag |= NMODIFIED; 1402 if (!wccflag) 1403 VTONFS(dvp)->n_attrstamp = 0; 1404 return (error); 1405 } 1406 1407 /* 1408 * nfs file remove call 1409 * To try and make nfs semantics closer to ufs semantics, a file that has 1410 * other processes using the vnode is renamed instead of removed and then 1411 * removed later on the last close. 1412 * - If v_usecount > 1 1413 * If a rename is not already in the works 1414 * call nfs_sillyrename() to set it up 1415 * else 1416 * do the remove rpc 1417 */ 1418 static int 1419 nfs_remove(struct vop_remove_args *ap) 1420 { 1421 struct vnode *vp = ap->a_vp; 1422 struct vnode *dvp = ap->a_dvp; 1423 struct componentname *cnp = ap->a_cnp; 1424 struct nfsnode *np = VTONFS(vp); 1425 int error = 0; 1426 struct vattr vattr; 1427 1428 #ifndef DIAGNOSTIC 1429 if ((cnp->cn_flags & HASBUF) == 0) 1430 panic("nfs_remove: no name"); 1431 if (vrefcnt(vp) < 1) 1432 panic("nfs_remove: bad v_usecount"); 1433 #endif 1434 if (vp->v_type == VDIR) 1435 error = EPERM; 1436 else if (vrefcnt(vp) == 1 || (np->n_sillyrename && 1437 VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_thread) == 0 && 1438 vattr.va_nlink > 1)) { 1439 /* 1440 * Purge the name cache so that the chance of a lookup for 1441 * the name succeeding while the remove is in progress is 1442 * minimized. Without node locking it can still happen, such 1443 * that an I/O op returns ESTALE, but since you get this if 1444 * another host removes the file.. 1445 */ 1446 cache_purge(vp); 1447 /* 1448 * throw away biocache buffers, mainly to avoid 1449 * unnecessary delayed writes later. 1450 */ 1451 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_thread, 1); 1452 /* Do the rpc */ 1453 if (error != EINTR && error != EIO) 1454 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1455 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread); 1456 /* 1457 * Kludge City: If the first reply to the remove rpc is lost.. 1458 * the reply to the retransmitted request will be ENOENT 1459 * since the file was in fact removed 1460 * Therefore, we cheat and return success. 1461 */ 1462 if (error == ENOENT) 1463 error = 0; 1464 } else if (!np->n_sillyrename) 1465 error = nfs_sillyrename(dvp, vp, cnp); 1466 np->n_attrstamp = 0; 1467 return (error); 1468 } 1469 1470 /* 1471 * nfs file remove rpc called from nfs_inactive 1472 */ 1473 int 1474 nfs_removeit(struct sillyrename *sp) 1475 { 1476 1477 /* 1478 * Make sure that the directory vnode is still valid. 1479 * XXX we should lock sp->s_dvp here. 1480 */ 1481 if (sp->s_dvp->v_type == VBAD) 1482 return (0); 1483 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred, 1484 NULL)); 1485 } 1486 1487 /* 1488 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1489 */ 1490 static int 1491 nfs_removerpc(struct vnode *dvp, const char *name, int namelen, 1492 struct ucred *cred, struct thread *td) 1493 { 1494 caddr_t bpos, dpos; 1495 int error = 0, wccflag = NFSV3_WCCRATTR; 1496 struct mbuf *mreq, *mrep, *md, *mb; 1497 int v3 = NFS_ISV3(dvp); 1498 1499 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1500 mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE, 1501 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1502 mb = mreq; 1503 bpos = mtod(mb, caddr_t); 1504 nfsm_fhtom(dvp, v3); 1505 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1506 nfsm_request(dvp, NFSPROC_REMOVE, td, cred); 1507 if (v3) 1508 nfsm_wcc_data(dvp, wccflag); 1509 m_freem(mrep); 1510 nfsmout: 1511 VTONFS(dvp)->n_flag |= NMODIFIED; 1512 if (!wccflag) 1513 VTONFS(dvp)->n_attrstamp = 0; 1514 return (error); 1515 } 1516 1517 /* 1518 * nfs file rename call 1519 */ 1520 static int 1521 nfs_rename(struct vop_rename_args *ap) 1522 { 1523 struct vnode *fvp = ap->a_fvp; 1524 struct vnode *tvp = ap->a_tvp; 1525 struct vnode *fdvp = ap->a_fdvp; 1526 struct vnode *tdvp = ap->a_tdvp; 1527 struct componentname *tcnp = ap->a_tcnp; 1528 struct componentname *fcnp = ap->a_fcnp; 1529 int error; 1530 1531 #ifndef DIAGNOSTIC 1532 if ((tcnp->cn_flags & HASBUF) == 0 || 1533 (fcnp->cn_flags & HASBUF) == 0) 1534 panic("nfs_rename: no name"); 1535 #endif 1536 /* Check for cross-device rename */ 1537 if ((fvp->v_mount != tdvp->v_mount) || 1538 (tvp && (fvp->v_mount != tvp->v_mount))) { 1539 error = EXDEV; 1540 goto out; 1541 } 1542 1543 if (fvp == tvp) { 1544 printf("nfs_rename: fvp == tvp (can't happen)\n"); 1545 error = 0; 1546 goto out; 1547 } 1548 if ((error = vn_lock(fvp, LK_EXCLUSIVE, fcnp->cn_thread)) != 0) 1549 goto out; 1550 1551 /* 1552 * We have to flush B_DELWRI data prior to renaming 1553 * the file. If we don't, the delayed-write buffers 1554 * can be flushed out later after the file has gone stale 1555 * under NFSV3. NFSV2 does not have this problem because 1556 * ( as far as I can tell ) it flushes dirty buffers more 1557 * often. 1558 */ 1559 VOP_FSYNC(fvp, fcnp->cn_cred, MNT_WAIT, fcnp->cn_thread); 1560 VOP_UNLOCK(fvp, 0, fcnp->cn_thread); 1561 if (tvp) 1562 VOP_FSYNC(tvp, tcnp->cn_cred, MNT_WAIT, tcnp->cn_thread); 1563 1564 /* 1565 * If the tvp exists and is in use, sillyrename it before doing the 1566 * rename of the new file over it. 1567 * XXX Can't sillyrename a directory. 1568 */ 1569 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename && 1570 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1571 vput(tvp); 1572 tvp = NULL; 1573 } 1574 1575 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1576 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1577 tcnp->cn_thread); 1578 1579 if (fvp->v_type == VDIR) { 1580 if (tvp != NULL && tvp->v_type == VDIR) 1581 cache_purge(tdvp); 1582 cache_purge(fdvp); 1583 } 1584 1585 out: 1586 if (tdvp == tvp) 1587 vrele(tdvp); 1588 else 1589 vput(tdvp); 1590 if (tvp) 1591 vput(tvp); 1592 vrele(fdvp); 1593 vrele(fvp); 1594 /* 1595 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1596 */ 1597 if (error == ENOENT) 1598 error = 0; 1599 return (error); 1600 } 1601 1602 /* 1603 * nfs file rename rpc called from nfs_remove() above 1604 */ 1605 static int 1606 nfs_renameit(struct vnode *sdvp, struct componentname *scnp, 1607 struct sillyrename *sp) 1608 { 1609 1610 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp, 1611 sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread)); 1612 } 1613 1614 /* 1615 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1616 */ 1617 static int 1618 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen, 1619 struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred, 1620 struct thread *td) 1621 { 1622 caddr_t bpos, dpos; 1623 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1624 struct mbuf *mreq, *mrep, *md, *mb; 1625 int v3 = NFS_ISV3(fdvp); 1626 1627 nfsstats.rpccnt[NFSPROC_RENAME]++; 1628 mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME, 1629 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 1630 nfsm_rndup(tnamelen)); 1631 mb = mreq; 1632 bpos = mtod(mb, caddr_t); 1633 nfsm_fhtom(fdvp, v3); 1634 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 1635 nfsm_fhtom(tdvp, v3); 1636 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 1637 nfsm_request(fdvp, NFSPROC_RENAME, td, cred); 1638 if (v3) { 1639 nfsm_wcc_data(fdvp, fwccflag); 1640 nfsm_wcc_data(tdvp, twccflag); 1641 } 1642 m_freem(mrep); 1643 nfsmout: 1644 VTONFS(fdvp)->n_flag |= NMODIFIED; 1645 VTONFS(tdvp)->n_flag |= NMODIFIED; 1646 if (!fwccflag) 1647 VTONFS(fdvp)->n_attrstamp = 0; 1648 if (!twccflag) 1649 VTONFS(tdvp)->n_attrstamp = 0; 1650 return (error); 1651 } 1652 1653 /* 1654 * nfs hard link create call 1655 */ 1656 static int 1657 nfs_link(struct vop_link_args *ap) 1658 { 1659 struct vnode *vp = ap->a_vp; 1660 struct vnode *tdvp = ap->a_tdvp; 1661 struct componentname *cnp = ap->a_cnp; 1662 caddr_t bpos, dpos; 1663 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 1664 struct mbuf *mreq, *mrep, *md, *mb; 1665 int v3; 1666 1667 if (vp->v_mount != tdvp->v_mount) { 1668 return (EXDEV); 1669 } 1670 1671 /* 1672 * Push all writes to the server, so that the attribute cache 1673 * doesn't get "out of sync" with the server. 1674 * XXX There should be a better way! 1675 */ 1676 VOP_FSYNC(vp, cnp->cn_cred, MNT_WAIT, cnp->cn_thread); 1677 1678 v3 = NFS_ISV3(vp); 1679 nfsstats.rpccnt[NFSPROC_LINK]++; 1680 mreq = nfsm_reqhead(vp, NFSPROC_LINK, 1681 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 1682 mb = mreq; 1683 bpos = mtod(mb, caddr_t); 1684 nfsm_fhtom(vp, v3); 1685 nfsm_fhtom(tdvp, v3); 1686 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1687 nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred); 1688 if (v3) { 1689 nfsm_postop_attr(vp, attrflag); 1690 nfsm_wcc_data(tdvp, wccflag); 1691 } 1692 m_freem(mrep); 1693 nfsmout: 1694 VTONFS(tdvp)->n_flag |= NMODIFIED; 1695 if (!attrflag) 1696 VTONFS(vp)->n_attrstamp = 0; 1697 if (!wccflag) 1698 VTONFS(tdvp)->n_attrstamp = 0; 1699 /* 1700 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 1701 */ 1702 if (error == EEXIST) 1703 error = 0; 1704 return (error); 1705 } 1706 1707 /* 1708 * nfs symbolic link create call 1709 */ 1710 static int 1711 nfs_symlink(struct vop_symlink_args *ap) 1712 { 1713 struct vnode *dvp = ap->a_dvp; 1714 struct vattr *vap = ap->a_vap; 1715 struct componentname *cnp = ap->a_cnp; 1716 struct nfsv2_sattr *sp; 1717 caddr_t bpos, dpos; 1718 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 1719 struct mbuf *mreq, *mrep, *md, *mb; 1720 struct vnode *newvp = NULL; 1721 int v3 = NFS_ISV3(dvp); 1722 1723 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 1724 slen = strlen(ap->a_target); 1725 mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 1726 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 1727 mb = mreq; 1728 bpos = mtod(mb, caddr_t); 1729 nfsm_fhtom(dvp, v3); 1730 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1731 if (v3) { 1732 nfsm_v3attrbuild(vap, FALSE); 1733 } 1734 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 1735 if (!v3) { 1736 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR); 1737 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 1738 sp->sa_uid = nfs_xdrneg1; 1739 sp->sa_gid = nfs_xdrneg1; 1740 sp->sa_size = nfs_xdrneg1; 1741 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1742 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1743 } 1744 1745 /* 1746 * Issue the NFS request and get the rpc response. 1747 * 1748 * Only NFSv3 responses returning an error of 0 actually return 1749 * a file handle that can be converted into newvp without having 1750 * to do an extra lookup rpc. 1751 */ 1752 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred); 1753 if (v3) { 1754 if (error == 0) 1755 nfsm_mtofh(dvp, newvp, v3, gotvp); 1756 nfsm_wcc_data(dvp, wccflag); 1757 } 1758 1759 /* 1760 * out code jumps -> here, mrep is also freed. 1761 */ 1762 1763 m_freem(mrep); 1764 nfsmout: 1765 1766 /* 1767 * If we get an EEXIST error, silently convert it to no-error 1768 * in case of an NFS retry. 1769 */ 1770 if (error == EEXIST) 1771 error = 0; 1772 1773 /* 1774 * If we do not have (or no longer have) an error, and we could 1775 * not extract the newvp from the response due to the request being 1776 * NFSv2 or the error being EEXIST. We have to do a lookup in order 1777 * to obtain a newvp to return. 1778 */ 1779 if (error == 0 && newvp == NULL) { 1780 struct nfsnode *np = NULL; 1781 1782 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1783 cnp->cn_cred, cnp->cn_thread, &np); 1784 if (!error) 1785 newvp = NFSTOV(np); 1786 } 1787 if (error) { 1788 if (newvp) 1789 vput(newvp); 1790 } else { 1791 *ap->a_vpp = newvp; 1792 } 1793 VTONFS(dvp)->n_flag |= NMODIFIED; 1794 if (!wccflag) 1795 VTONFS(dvp)->n_attrstamp = 0; 1796 return (error); 1797 } 1798 1799 /* 1800 * nfs make dir call 1801 */ 1802 static int 1803 nfs_mkdir(struct vop_mkdir_args *ap) 1804 { 1805 struct vnode *dvp = ap->a_dvp; 1806 struct vattr *vap = ap->a_vap; 1807 struct componentname *cnp = ap->a_cnp; 1808 struct nfsv2_sattr *sp; 1809 int len; 1810 struct nfsnode *np = NULL; 1811 struct vnode *newvp = NULL; 1812 caddr_t bpos, dpos; 1813 int error = 0, wccflag = NFSV3_WCCRATTR; 1814 int gotvp = 0; 1815 struct mbuf *mreq, *mrep, *md, *mb; 1816 struct vattr vattr; 1817 int v3 = NFS_ISV3(dvp); 1818 1819 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) { 1820 return (error); 1821 } 1822 len = cnp->cn_namelen; 1823 nfsstats.rpccnt[NFSPROC_MKDIR]++; 1824 mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR, 1825 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 1826 mb = mreq; 1827 bpos = mtod(mb, caddr_t); 1828 nfsm_fhtom(dvp, v3); 1829 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 1830 if (v3) { 1831 nfsm_v3attrbuild(vap, FALSE); 1832 } else { 1833 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR); 1834 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 1835 sp->sa_uid = nfs_xdrneg1; 1836 sp->sa_gid = nfs_xdrneg1; 1837 sp->sa_size = nfs_xdrneg1; 1838 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1839 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1840 } 1841 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred); 1842 if (!error) 1843 nfsm_mtofh(dvp, newvp, v3, gotvp); 1844 if (v3) 1845 nfsm_wcc_data(dvp, wccflag); 1846 m_freem(mrep); 1847 nfsmout: 1848 VTONFS(dvp)->n_flag |= NMODIFIED; 1849 if (!wccflag) 1850 VTONFS(dvp)->n_attrstamp = 0; 1851 /* 1852 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 1853 * if we can succeed in looking up the directory. 1854 */ 1855 if (error == EEXIST || (!error && !gotvp)) { 1856 if (newvp) { 1857 vrele(newvp); 1858 newvp = NULL; 1859 } 1860 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 1861 cnp->cn_thread, &np); 1862 if (!error) { 1863 newvp = NFSTOV(np); 1864 if (newvp->v_type != VDIR) 1865 error = EEXIST; 1866 } 1867 } 1868 if (error) { 1869 if (newvp) 1870 vrele(newvp); 1871 } else 1872 *ap->a_vpp = newvp; 1873 return (error); 1874 } 1875 1876 /* 1877 * nfs remove directory call 1878 */ 1879 static int 1880 nfs_rmdir(struct vop_rmdir_args *ap) 1881 { 1882 struct vnode *vp = ap->a_vp; 1883 struct vnode *dvp = ap->a_dvp; 1884 struct componentname *cnp = ap->a_cnp; 1885 caddr_t bpos, dpos; 1886 int error = 0, wccflag = NFSV3_WCCRATTR; 1887 struct mbuf *mreq, *mrep, *md, *mb; 1888 int v3 = NFS_ISV3(dvp); 1889 1890 if (dvp == vp) 1891 return (EINVAL); 1892 nfsstats.rpccnt[NFSPROC_RMDIR]++; 1893 mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR, 1894 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 1895 mb = mreq; 1896 bpos = mtod(mb, caddr_t); 1897 nfsm_fhtom(dvp, v3); 1898 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1899 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred); 1900 if (v3) 1901 nfsm_wcc_data(dvp, wccflag); 1902 m_freem(mrep); 1903 nfsmout: 1904 VTONFS(dvp)->n_flag |= NMODIFIED; 1905 if (!wccflag) 1906 VTONFS(dvp)->n_attrstamp = 0; 1907 cache_purge(dvp); 1908 cache_purge(vp); 1909 /* 1910 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 1911 */ 1912 if (error == ENOENT) 1913 error = 0; 1914 return (error); 1915 } 1916 1917 /* 1918 * nfs readdir call 1919 */ 1920 static int 1921 nfs_readdir(struct vop_readdir_args *ap) 1922 { 1923 struct vnode *vp = ap->a_vp; 1924 struct nfsnode *np = VTONFS(vp); 1925 struct uio *uio = ap->a_uio; 1926 int tresid, error; 1927 struct vattr vattr; 1928 1929 if (vp->v_type != VDIR) 1930 return (EPERM); 1931 /* 1932 * First, check for hit on the EOF offset cache 1933 */ 1934 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 1935 (np->n_flag & NMODIFIED) == 0) { 1936 if (VOP_GETATTR(vp, &vattr, ap->a_cred, uio->uio_td) == 0 && 1937 np->n_mtime == vattr.va_mtime.tv_sec) { 1938 nfsstats.direofcache_hits++; 1939 return (0); 1940 } 1941 } 1942 1943 /* 1944 * Call nfs_bioread() to do the real work. 1945 */ 1946 tresid = uio->uio_resid; 1947 error = nfs_bioread(vp, uio, 0, ap->a_cred); 1948 1949 if (!error && uio->uio_resid == tresid) 1950 nfsstats.direofcache_misses++; 1951 return (error); 1952 } 1953 1954 /* 1955 * Readdir rpc call. 1956 * Called from below the buffer cache by nfs_doio(). 1957 */ 1958 int 1959 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1960 { 1961 int len, left; 1962 struct dirent *dp = NULL; 1963 u_int32_t *tl; 1964 caddr_t cp; 1965 nfsuint64 *cookiep; 1966 caddr_t bpos, dpos; 1967 struct mbuf *mreq, *mrep, *md, *mb; 1968 nfsuint64 cookie; 1969 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1970 struct nfsnode *dnp = VTONFS(vp); 1971 u_quad_t fileno; 1972 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1; 1973 int attrflag; 1974 int v3 = NFS_ISV3(vp); 1975 1976 #ifndef DIAGNOSTIC 1977 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 1978 (uiop->uio_resid & (DIRBLKSIZ - 1))) 1979 panic("nfs readdirrpc bad uio"); 1980 #endif 1981 1982 /* 1983 * If there is no cookie, assume directory was stale. 1984 */ 1985 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 1986 if (cookiep) 1987 cookie = *cookiep; 1988 else 1989 return (NFSERR_BAD_COOKIE); 1990 /* 1991 * Loop around doing readdir rpc's of size nm_readdirsize 1992 * truncated to a multiple of DIRBLKSIZ. 1993 * The stopping criteria is EOF or buffer full. 1994 */ 1995 while (more_dirs && bigenough) { 1996 nfsstats.rpccnt[NFSPROC_READDIR]++; 1997 mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) + 1998 NFSX_READDIR(v3)); 1999 mb = mreq; 2000 bpos = mtod(mb, caddr_t); 2001 nfsm_fhtom(vp, v3); 2002 if (v3) { 2003 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED); 2004 *tl++ = cookie.nfsuquad[0]; 2005 *tl++ = cookie.nfsuquad[1]; 2006 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2007 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2008 } else { 2009 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED); 2010 *tl++ = cookie.nfsuquad[0]; 2011 } 2012 *tl = txdr_unsigned(nmp->nm_readdirsize); 2013 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred); 2014 if (v3) { 2015 nfsm_postop_attr(vp, attrflag); 2016 if (!error) { 2017 tl = nfsm_dissect(u_int32_t *, 2018 2 * NFSX_UNSIGNED); 2019 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2020 dnp->n_cookieverf.nfsuquad[1] = *tl; 2021 } else { 2022 m_freem(mrep); 2023 goto nfsmout; 2024 } 2025 } 2026 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); 2027 more_dirs = fxdr_unsigned(int, *tl); 2028 2029 /* loop thru the dir entries, doctoring them to 4bsd form */ 2030 while (more_dirs && bigenough) { 2031 if (v3) { 2032 tl = nfsm_dissect(u_int32_t *, 2033 3 * NFSX_UNSIGNED); 2034 fileno = fxdr_hyper(tl); 2035 len = fxdr_unsigned(int, *(tl + 2)); 2036 } else { 2037 tl = nfsm_dissect(u_int32_t *, 2038 2 * NFSX_UNSIGNED); 2039 fileno = fxdr_unsigned(u_quad_t, *tl++); 2040 len = fxdr_unsigned(int, *tl); 2041 } 2042 if (len <= 0 || len > NFS_MAXNAMLEN) { 2043 error = EBADRPC; 2044 m_freem(mrep); 2045 goto nfsmout; 2046 } 2047 tlen = nfsm_rndup(len); 2048 if (tlen == len) 2049 tlen += 4; /* To ensure null termination */ 2050 left = DIRBLKSIZ - blksiz; 2051 if ((tlen + DIRHDSIZ) > left) { 2052 dp->d_reclen += left; 2053 uiop->uio_iov->iov_base = 2054 (char *)uiop->uio_iov->iov_base + left; 2055 uiop->uio_iov->iov_len -= left; 2056 uiop->uio_offset += left; 2057 uiop->uio_resid -= left; 2058 blksiz = 0; 2059 } 2060 if ((tlen + DIRHDSIZ) > uiop->uio_resid) 2061 bigenough = 0; 2062 if (bigenough) { 2063 dp = (struct dirent *)uiop->uio_iov->iov_base; 2064 dp->d_fileno = (int)fileno; 2065 dp->d_namlen = len; 2066 dp->d_reclen = tlen + DIRHDSIZ; 2067 dp->d_type = DT_UNKNOWN; 2068 blksiz += dp->d_reclen; 2069 if (blksiz == DIRBLKSIZ) 2070 blksiz = 0; 2071 uiop->uio_offset += DIRHDSIZ; 2072 uiop->uio_resid -= DIRHDSIZ; 2073 uiop->uio_iov->iov_base = 2074 (char *)uiop->uio_iov->iov_base + DIRHDSIZ; 2075 uiop->uio_iov->iov_len -= DIRHDSIZ; 2076 nfsm_mtouio(uiop, len); 2077 cp = uiop->uio_iov->iov_base; 2078 tlen -= len; 2079 *cp = '\0'; /* null terminate */ 2080 uiop->uio_iov->iov_base = 2081 (char *)uiop->uio_iov->iov_base + tlen; 2082 uiop->uio_iov->iov_len -= tlen; 2083 uiop->uio_offset += tlen; 2084 uiop->uio_resid -= tlen; 2085 } else 2086 nfsm_adv(nfsm_rndup(len)); 2087 if (v3) { 2088 tl = nfsm_dissect(u_int32_t *, 2089 3 * NFSX_UNSIGNED); 2090 } else { 2091 tl = nfsm_dissect(u_int32_t *, 2092 2 * NFSX_UNSIGNED); 2093 } 2094 if (bigenough) { 2095 cookie.nfsuquad[0] = *tl++; 2096 if (v3) 2097 cookie.nfsuquad[1] = *tl++; 2098 } else if (v3) 2099 tl += 2; 2100 else 2101 tl++; 2102 more_dirs = fxdr_unsigned(int, *tl); 2103 } 2104 /* 2105 * If at end of rpc data, get the eof boolean 2106 */ 2107 if (!more_dirs) { 2108 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); 2109 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2110 } 2111 m_freem(mrep); 2112 } 2113 /* 2114 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2115 * by increasing d_reclen for the last record. 2116 */ 2117 if (blksiz > 0) { 2118 left = DIRBLKSIZ - blksiz; 2119 dp->d_reclen += left; 2120 uiop->uio_iov->iov_base = 2121 (char *)uiop->uio_iov->iov_base + left; 2122 uiop->uio_iov->iov_len -= left; 2123 uiop->uio_offset += left; 2124 uiop->uio_resid -= left; 2125 } 2126 2127 /* 2128 * We are now either at the end of the directory or have filled the 2129 * block. 2130 */ 2131 if (bigenough) 2132 dnp->n_direofoffset = uiop->uio_offset; 2133 else { 2134 if (uiop->uio_resid > 0) 2135 printf("EEK! readdirrpc resid > 0\n"); 2136 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2137 *cookiep = cookie; 2138 } 2139 nfsmout: 2140 return (error); 2141 } 2142 2143 /* 2144 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2145 */ 2146 int 2147 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 2148 { 2149 int len, left; 2150 struct dirent *dp; 2151 u_int32_t *tl; 2152 caddr_t cp; 2153 struct vnode *newvp; 2154 nfsuint64 *cookiep; 2155 caddr_t bpos, dpos, dpossav1, dpossav2; 2156 struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2; 2157 struct nameidata nami, *ndp = &nami; 2158 struct componentname *cnp = &ndp->ni_cnd; 2159 nfsuint64 cookie; 2160 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2161 struct nfsnode *dnp = VTONFS(vp), *np; 2162 nfsfh_t *fhp; 2163 u_quad_t fileno; 2164 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2165 int attrflag, fhsize; 2166 2167 #ifndef nolint 2168 dp = NULL; 2169 #endif 2170 #ifndef DIAGNOSTIC 2171 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2172 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2173 panic("nfs readdirplusrpc bad uio"); 2174 #endif 2175 ndp->ni_dvp = vp; 2176 newvp = NULLVP; 2177 2178 /* 2179 * If there is no cookie, assume directory was stale. 2180 */ 2181 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2182 if (cookiep) 2183 cookie = *cookiep; 2184 else 2185 return (NFSERR_BAD_COOKIE); 2186 /* 2187 * Loop around doing readdir rpc's of size nm_readdirsize 2188 * truncated to a multiple of DIRBLKSIZ. 2189 * The stopping criteria is EOF or buffer full. 2190 */ 2191 while (more_dirs && bigenough) { 2192 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2193 mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS, 2194 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2195 mb = mreq; 2196 bpos = mtod(mb, caddr_t); 2197 nfsm_fhtom(vp, 1); 2198 tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED); 2199 *tl++ = cookie.nfsuquad[0]; 2200 *tl++ = cookie.nfsuquad[1]; 2201 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2202 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2203 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2204 *tl = txdr_unsigned(nmp->nm_rsize); 2205 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred); 2206 nfsm_postop_attr(vp, attrflag); 2207 if (error) { 2208 m_freem(mrep); 2209 goto nfsmout; 2210 } 2211 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED); 2212 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2213 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2214 more_dirs = fxdr_unsigned(int, *tl); 2215 2216 /* loop thru the dir entries, doctoring them to 4bsd form */ 2217 while (more_dirs && bigenough) { 2218 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED); 2219 fileno = fxdr_hyper(tl); 2220 len = fxdr_unsigned(int, *(tl + 2)); 2221 if (len <= 0 || len > NFS_MAXNAMLEN) { 2222 error = EBADRPC; 2223 m_freem(mrep); 2224 goto nfsmout; 2225 } 2226 tlen = nfsm_rndup(len); 2227 if (tlen == len) 2228 tlen += 4; /* To ensure null termination*/ 2229 left = DIRBLKSIZ - blksiz; 2230 if ((tlen + DIRHDSIZ) > left) { 2231 dp->d_reclen += left; 2232 uiop->uio_iov->iov_base = 2233 (char *)uiop->uio_iov->iov_base + left; 2234 uiop->uio_iov->iov_len -= left; 2235 uiop->uio_offset += left; 2236 uiop->uio_resid -= left; 2237 blksiz = 0; 2238 } 2239 if ((tlen + DIRHDSIZ) > uiop->uio_resid) 2240 bigenough = 0; 2241 if (bigenough) { 2242 dp = (struct dirent *)uiop->uio_iov->iov_base; 2243 dp->d_fileno = (int)fileno; 2244 dp->d_namlen = len; 2245 dp->d_reclen = tlen + DIRHDSIZ; 2246 dp->d_type = DT_UNKNOWN; 2247 blksiz += dp->d_reclen; 2248 if (blksiz == DIRBLKSIZ) 2249 blksiz = 0; 2250 uiop->uio_offset += DIRHDSIZ; 2251 uiop->uio_resid -= DIRHDSIZ; 2252 uiop->uio_iov->iov_base = 2253 (char *)uiop->uio_iov->iov_base + DIRHDSIZ; 2254 uiop->uio_iov->iov_len -= DIRHDSIZ; 2255 cnp->cn_nameptr = uiop->uio_iov->iov_base; 2256 cnp->cn_namelen = len; 2257 nfsm_mtouio(uiop, len); 2258 cp = uiop->uio_iov->iov_base; 2259 tlen -= len; 2260 *cp = '\0'; 2261 uiop->uio_iov->iov_base = 2262 (char *)uiop->uio_iov->iov_base + tlen; 2263 uiop->uio_iov->iov_len -= tlen; 2264 uiop->uio_offset += tlen; 2265 uiop->uio_resid -= tlen; 2266 } else 2267 nfsm_adv(nfsm_rndup(len)); 2268 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED); 2269 if (bigenough) { 2270 cookie.nfsuquad[0] = *tl++; 2271 cookie.nfsuquad[1] = *tl++; 2272 } else 2273 tl += 2; 2274 2275 /* 2276 * Since the attributes are before the file handle 2277 * (sigh), we must skip over the attributes and then 2278 * come back and get them. 2279 */ 2280 attrflag = fxdr_unsigned(int, *tl); 2281 if (attrflag) { 2282 dpossav1 = dpos; 2283 mdsav1 = md; 2284 nfsm_adv(NFSX_V3FATTR); 2285 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); 2286 doit = fxdr_unsigned(int, *tl); 2287 if (doit) { 2288 nfsm_getfh(fhp, fhsize, 1); 2289 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2290 VREF(vp); 2291 newvp = vp; 2292 np = dnp; 2293 } else { 2294 error = nfs_nget(vp->v_mount, fhp, 2295 fhsize, &np); 2296 if (error) 2297 doit = 0; 2298 else 2299 newvp = NFSTOV(np); 2300 } 2301 } 2302 if (doit && bigenough) { 2303 dpossav2 = dpos; 2304 dpos = dpossav1; 2305 mdsav2 = md; 2306 md = mdsav1; 2307 nfsm_loadattr(newvp, NULL); 2308 dpos = dpossav2; 2309 md = mdsav2; 2310 dp->d_type = 2311 IFTODT(VTTOIF(np->n_vattr.va_type)); 2312 ndp->ni_vp = newvp; 2313 cache_enter(ndp->ni_dvp, ndp->ni_vp, cnp); 2314 } 2315 } else { 2316 /* Just skip over the file handle */ 2317 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); 2318 i = fxdr_unsigned(int, *tl); 2319 if (i) { 2320 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); 2321 fhsize = fxdr_unsigned(int, *tl); 2322 nfsm_adv(nfsm_rndup(fhsize)); 2323 } 2324 } 2325 if (newvp != NULLVP) { 2326 if (newvp == vp) 2327 vrele(newvp); 2328 else 2329 vput(newvp); 2330 newvp = NULLVP; 2331 } 2332 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); 2333 more_dirs = fxdr_unsigned(int, *tl); 2334 } 2335 /* 2336 * If at end of rpc data, get the eof boolean 2337 */ 2338 if (!more_dirs) { 2339 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); 2340 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2341 } 2342 m_freem(mrep); 2343 } 2344 /* 2345 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2346 * by increasing d_reclen for the last record. 2347 */ 2348 if (blksiz > 0) { 2349 left = DIRBLKSIZ - blksiz; 2350 dp->d_reclen += left; 2351 uiop->uio_iov->iov_base = 2352 (char *)uiop->uio_iov->iov_base + left; 2353 uiop->uio_iov->iov_len -= left; 2354 uiop->uio_offset += left; 2355 uiop->uio_resid -= left; 2356 } 2357 2358 /* 2359 * We are now either at the end of the directory or have filled the 2360 * block. 2361 */ 2362 if (bigenough) 2363 dnp->n_direofoffset = uiop->uio_offset; 2364 else { 2365 if (uiop->uio_resid > 0) 2366 printf("EEK! readdirplusrpc resid > 0\n"); 2367 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2368 *cookiep = cookie; 2369 } 2370 nfsmout: 2371 if (newvp != NULLVP) { 2372 if (newvp == vp) 2373 vrele(newvp); 2374 else 2375 vput(newvp); 2376 newvp = NULLVP; 2377 } 2378 return (error); 2379 } 2380 2381 /* 2382 * Silly rename. To make the NFS filesystem that is stateless look a little 2383 * more like the "ufs" a remove of an active vnode is translated to a rename 2384 * to a funny looking filename that is removed by nfs_inactive on the 2385 * nfsnode. There is the potential for another process on a different client 2386 * to create the same funny name between the nfs_lookitup() fails and the 2387 * nfs_rename() completes, but... 2388 */ 2389 static int 2390 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2391 { 2392 struct sillyrename *sp; 2393 struct nfsnode *np; 2394 int error; 2395 short pid; 2396 2397 cache_purge(dvp); 2398 np = VTONFS(vp); 2399 #ifndef DIAGNOSTIC 2400 if (vp->v_type == VDIR) 2401 panic("nfs: sillyrename dir"); 2402 #endif 2403 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2404 M_NFSREQ, M_WAITOK); 2405 sp->s_cred = crhold(cnp->cn_cred); 2406 sp->s_dvp = dvp; 2407 sp->s_removeit = nfs_removeit; 2408 VREF(dvp); 2409 2410 /* Fudge together a funny name */ 2411 pid = cnp->cn_thread->td_proc->p_pid; 2412 sp->s_namlen = sprintf(sp->s_name, ".nfsA%04x4.4", pid); 2413 2414 /* Try lookitups until we get one that isn't there */ 2415 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2416 cnp->cn_thread, NULL) == 0) { 2417 sp->s_name[4]++; 2418 if (sp->s_name[4] > 'z') { 2419 error = EINVAL; 2420 goto bad; 2421 } 2422 } 2423 error = nfs_renameit(dvp, cnp, sp); 2424 if (error) 2425 goto bad; 2426 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2427 cnp->cn_thread, &np); 2428 np->n_sillyrename = sp; 2429 return (0); 2430 bad: 2431 vrele(sp->s_dvp); 2432 crfree(sp->s_cred); 2433 free((caddr_t)sp, M_NFSREQ); 2434 return (error); 2435 } 2436 2437 /* 2438 * Look up a file name and optionally either update the file handle or 2439 * allocate an nfsnode, depending on the value of npp. 2440 * npp == NULL --> just do the lookup 2441 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2442 * handled too 2443 * *npp != NULL --> update the file handle in the vnode 2444 */ 2445 static int 2446 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred, 2447 struct thread *td, struct nfsnode **npp) 2448 { 2449 struct vnode *newvp = NULL; 2450 struct nfsnode *np, *dnp = VTONFS(dvp); 2451 caddr_t bpos, dpos; 2452 int error = 0, fhlen, attrflag; 2453 struct mbuf *mreq, *mrep, *md, *mb; 2454 nfsfh_t *nfhp; 2455 int v3 = NFS_ISV3(dvp); 2456 2457 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2458 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP, 2459 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2460 mb = mreq; 2461 bpos = mtod(mb, caddr_t); 2462 nfsm_fhtom(dvp, v3); 2463 nfsm_strtom(name, len, NFS_MAXNAMLEN); 2464 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred); 2465 if (npp && !error) { 2466 nfsm_getfh(nfhp, fhlen, v3); 2467 if (*npp) { 2468 np = *npp; 2469 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2470 free((caddr_t)np->n_fhp, M_NFSBIGFH); 2471 np->n_fhp = &np->n_fh; 2472 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH) 2473 np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK); 2474 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen); 2475 np->n_fhsize = fhlen; 2476 newvp = NFSTOV(np); 2477 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2478 VREF(dvp); 2479 newvp = dvp; 2480 } else { 2481 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 2482 if (error) { 2483 m_freem(mrep); 2484 return (error); 2485 } 2486 newvp = NFSTOV(np); 2487 } 2488 if (v3) { 2489 nfsm_postop_attr(newvp, attrflag); 2490 if (!attrflag && *npp == NULL) { 2491 m_freem(mrep); 2492 if (newvp == dvp) 2493 vrele(newvp); 2494 else 2495 vput(newvp); 2496 return (ENOENT); 2497 } 2498 } else 2499 nfsm_loadattr(newvp, NULL); 2500 } 2501 m_freem(mrep); 2502 nfsmout: 2503 if (npp && *npp == NULL) { 2504 if (error) { 2505 if (newvp) { 2506 if (newvp == dvp) 2507 vrele(newvp); 2508 else 2509 vput(newvp); 2510 } 2511 } else 2512 *npp = np; 2513 } 2514 return (error); 2515 } 2516 2517 /* 2518 * Nfs Version 3 commit rpc 2519 */ 2520 int 2521 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred, 2522 struct thread *td) 2523 { 2524 u_int32_t *tl; 2525 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2526 caddr_t bpos, dpos; 2527 int error = 0, wccflag = NFSV3_WCCRATTR; 2528 struct mbuf *mreq, *mrep, *md, *mb; 2529 2530 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) 2531 return (0); 2532 nfsstats.rpccnt[NFSPROC_COMMIT]++; 2533 mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1)); 2534 mb = mreq; 2535 bpos = mtod(mb, caddr_t); 2536 nfsm_fhtom(vp, 1); 2537 tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED); 2538 txdr_hyper(offset, tl); 2539 tl += 2; 2540 *tl = txdr_unsigned(cnt); 2541 nfsm_request(vp, NFSPROC_COMMIT, td, cred); 2542 nfsm_wcc_data(vp, wccflag); 2543 if (!error) { 2544 tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF); 2545 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl, 2546 NFSX_V3WRITEVERF)) { 2547 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 2548 NFSX_V3WRITEVERF); 2549 error = NFSERR_STALEWRITEVERF; 2550 } 2551 } 2552 m_freem(mrep); 2553 nfsmout: 2554 return (error); 2555 } 2556 2557 /* 2558 * Strategy routine. 2559 * For async requests when nfsiod(s) are running, queue the request by 2560 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 2561 * request. 2562 */ 2563 static int 2564 nfs_strategy(struct vop_strategy_args *ap) 2565 { 2566 struct buf *bp = ap->a_bp; 2567 struct ucred *cr; 2568 struct thread *td; 2569 int error = 0; 2570 2571 KASSERT(ap->a_vp == ap->a_bp->b_vp, ("%s(%p != %p)", 2572 __func__, ap->a_vp, ap->a_bp->b_vp)); 2573 KASSERT(!(bp->b_flags & B_DONE), ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); 2574 KASSERT(BUF_REFCNT(bp) > 0, ("nfs_strategy: buffer %p not locked", bp)); 2575 2576 if (bp->b_flags & B_ASYNC) 2577 td = NULL; 2578 else 2579 td = curthread; /* XXX */ 2580 2581 if (bp->b_iocmd == BIO_READ) 2582 cr = bp->b_rcred; 2583 else 2584 cr = bp->b_wcred; 2585 2586 /* 2587 * If the op is asynchronous and an i/o daemon is waiting 2588 * queue the request, wake it up and wait for completion 2589 * otherwise just do it ourselves. 2590 */ 2591 if ((bp->b_flags & B_ASYNC) == 0 || 2592 nfs_asyncio(bp, NOCRED, td)) 2593 error = nfs_doio(bp, cr, td); 2594 return (error); 2595 } 2596 2597 /* 2598 * fsync vnode op. Just call nfs_flush() with commit == 1. 2599 */ 2600 /* ARGSUSED */ 2601 static int 2602 nfs_fsync(struct vop_fsync_args *ap) 2603 { 2604 2605 return (nfs_flush(ap->a_vp, ap->a_cred, ap->a_waitfor, ap->a_td, 1)); 2606 } 2607 2608 /* 2609 * Flush all the blocks associated with a vnode. 2610 * Walk through the buffer pool and push any dirty pages 2611 * associated with the vnode. 2612 */ 2613 static int 2614 nfs_flush(struct vnode *vp, struct ucred *cred, int waitfor, struct thread *td, 2615 int commit) 2616 { 2617 struct nfsnode *np = VTONFS(vp); 2618 struct buf *bp; 2619 int i; 2620 struct buf *nbp; 2621 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2622 int s, error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos; 2623 int passone = 1; 2624 u_quad_t off, endoff, toff; 2625 struct ucred* wcred = NULL; 2626 struct buf **bvec = NULL; 2627 #ifndef NFS_COMMITBVECSIZ 2628 #define NFS_COMMITBVECSIZ 20 2629 #endif 2630 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ]; 2631 int bvecsize = 0, bveccount; 2632 2633 if (nmp->nm_flag & NFSMNT_INT) 2634 slpflag = PCATCH; 2635 if (!commit) 2636 passone = 0; 2637 /* 2638 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the 2639 * server, but has not been committed to stable storage on the server 2640 * yet. On the first pass, the byte range is worked out and the commit 2641 * rpc is done. On the second pass, nfs_writebp() is called to do the 2642 * job. 2643 */ 2644 again: 2645 off = (u_quad_t)-1; 2646 endoff = 0; 2647 bvecpos = 0; 2648 if (NFS_ISV3(vp) && commit) { 2649 s = splbio(); 2650 if (bvec != NULL && bvec != bvec_on_stack) 2651 free(bvec, M_TEMP); 2652 /* 2653 * Count up how many buffers waiting for a commit. 2654 */ 2655 bveccount = 0; 2656 VI_LOCK(vp); 2657 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 2658 nbp = TAILQ_NEXT(bp, b_vnbufs); 2659 if (BUF_REFCNT(bp) == 0 && 2660 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) 2661 == (B_DELWRI | B_NEEDCOMMIT)) 2662 bveccount++; 2663 } 2664 /* 2665 * Allocate space to remember the list of bufs to commit. It is 2666 * important to use M_NOWAIT here to avoid a race with nfs_write. 2667 * If we can't get memory (for whatever reason), we will end up 2668 * committing the buffers one-by-one in the loop below. 2669 */ 2670 if (bveccount > NFS_COMMITBVECSIZ) { 2671 /* 2672 * Release the vnode interlock to avoid a lock 2673 * order reversal. 2674 */ 2675 VI_UNLOCK(vp); 2676 bvec = (struct buf **) 2677 malloc(bveccount * sizeof(struct buf *), 2678 M_TEMP, M_NOWAIT); 2679 VI_LOCK(vp); 2680 if (bvec == NULL) { 2681 bvec = bvec_on_stack; 2682 bvecsize = NFS_COMMITBVECSIZ; 2683 } else 2684 bvecsize = bveccount; 2685 } else { 2686 bvec = bvec_on_stack; 2687 bvecsize = NFS_COMMITBVECSIZ; 2688 } 2689 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 2690 if (bvecpos >= bvecsize) 2691 break; 2692 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2693 nbp = TAILQ_NEXT(bp, b_vnbufs); 2694 continue; 2695 } 2696 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 2697 (B_DELWRI | B_NEEDCOMMIT)) { 2698 BUF_UNLOCK(bp); 2699 nbp = TAILQ_NEXT(bp, b_vnbufs); 2700 continue; 2701 } 2702 VI_UNLOCK(vp); 2703 bremfree(bp); 2704 /* 2705 * Work out if all buffers are using the same cred 2706 * so we can deal with them all with one commit. 2707 * 2708 * NOTE: we are not clearing B_DONE here, so we have 2709 * to do it later on in this routine if we intend to 2710 * initiate I/O on the bp. 2711 * 2712 * Note: to avoid loopback deadlocks, we do not 2713 * assign b_runningbufspace. 2714 */ 2715 if (wcred == NULL) 2716 wcred = bp->b_wcred; 2717 else if (wcred != bp->b_wcred) 2718 wcred = NOCRED; 2719 bp->b_flags |= B_WRITEINPROG; 2720 vfs_busy_pages(bp, 1); 2721 2722 VI_LOCK(vp); 2723 /* 2724 * bp is protected by being locked, but nbp is not 2725 * and vfs_busy_pages() may sleep. We have to 2726 * recalculate nbp. 2727 */ 2728 nbp = TAILQ_NEXT(bp, b_vnbufs); 2729 2730 /* 2731 * A list of these buffers is kept so that the 2732 * second loop knows which buffers have actually 2733 * been committed. This is necessary, since there 2734 * may be a race between the commit rpc and new 2735 * uncommitted writes on the file. 2736 */ 2737 bvec[bvecpos++] = bp; 2738 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2739 bp->b_dirtyoff; 2740 if (toff < off) 2741 off = toff; 2742 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); 2743 if (toff > endoff) 2744 endoff = toff; 2745 } 2746 splx(s); 2747 VI_UNLOCK(vp); 2748 } 2749 if (bvecpos > 0) { 2750 /* 2751 * Commit data on the server, as required. 2752 * If all bufs are using the same wcred, then use that with 2753 * one call for all of them, otherwise commit each one 2754 * separately. 2755 */ 2756 if (wcred != NOCRED) 2757 retv = nfs_commit(vp, off, (int)(endoff - off), 2758 wcred, td); 2759 else { 2760 retv = 0; 2761 for (i = 0; i < bvecpos; i++) { 2762 off_t off, size; 2763 bp = bvec[i]; 2764 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2765 bp->b_dirtyoff; 2766 size = (u_quad_t)(bp->b_dirtyend 2767 - bp->b_dirtyoff); 2768 retv = nfs_commit(vp, off, (int)size, 2769 bp->b_wcred, td); 2770 if (retv) break; 2771 } 2772 } 2773 2774 if (retv == NFSERR_STALEWRITEVERF) 2775 nfs_clearcommit(vp->v_mount); 2776 2777 /* 2778 * Now, either mark the blocks I/O done or mark the 2779 * blocks dirty, depending on whether the commit 2780 * succeeded. 2781 */ 2782 for (i = 0; i < bvecpos; i++) { 2783 bp = bvec[i]; 2784 bp->b_flags &= ~(B_NEEDCOMMIT | B_WRITEINPROG | B_CLUSTEROK); 2785 if (retv) { 2786 /* 2787 * Error, leave B_DELWRI intact 2788 */ 2789 vfs_unbusy_pages(bp); 2790 brelse(bp); 2791 } else { 2792 /* 2793 * Success, remove B_DELWRI ( bundirty() ). 2794 * 2795 * b_dirtyoff/b_dirtyend seem to be NFS 2796 * specific. We should probably move that 2797 * into bundirty(). XXX 2798 */ 2799 s = splbio(); 2800 VI_LOCK(vp); 2801 vp->v_numoutput++; 2802 VI_UNLOCK(vp); 2803 bp->b_flags |= B_ASYNC; 2804 bundirty(bp); 2805 bp->b_flags &= ~B_DONE; 2806 bp->b_ioflags &= ~BIO_ERROR; 2807 bp->b_dirtyoff = bp->b_dirtyend = 0; 2808 splx(s); 2809 bufdone(bp); 2810 } 2811 } 2812 } 2813 2814 /* 2815 * Start/do any write(s) that are required. 2816 */ 2817 loop: 2818 s = splbio(); 2819 VI_LOCK(vp); 2820 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 2821 nbp = TAILQ_NEXT(bp, b_vnbufs); 2822 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2823 if (waitfor != MNT_WAIT || passone) 2824 continue; 2825 2826 error = BUF_TIMELOCK(bp, 2827 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 2828 VI_MTX(vp), "nfsfsync", slpflag, slptimeo); 2829 splx(s); 2830 if (error == 0) 2831 panic("nfs_fsync: inconsistent lock"); 2832 if (error == ENOLCK) 2833 goto loop; 2834 error = nfs_sigintr(nmp, NULL, td); 2835 if (error) 2836 goto done; 2837 if (slpflag == PCATCH) { 2838 slpflag = 0; 2839 slptimeo = 2 * hz; 2840 } 2841 goto loop; 2842 } 2843 if ((bp->b_flags & B_DELWRI) == 0) 2844 panic("nfs_fsync: not dirty"); 2845 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) { 2846 BUF_UNLOCK(bp); 2847 continue; 2848 } 2849 VI_UNLOCK(vp); 2850 bremfree(bp); 2851 if (passone || !commit) 2852 bp->b_flags |= B_ASYNC; 2853 else 2854 bp->b_flags |= B_ASYNC | B_WRITEINPROG; 2855 splx(s); 2856 bwrite(bp); 2857 goto loop; 2858 } 2859 splx(s); 2860 if (passone) { 2861 passone = 0; 2862 VI_UNLOCK(vp); 2863 goto again; 2864 } 2865 if (waitfor == MNT_WAIT) { 2866 while (vp->v_numoutput) { 2867 vp->v_iflag |= VI_BWAIT; 2868 error = msleep((caddr_t)&vp->v_numoutput, VI_MTX(vp), 2869 slpflag | (PRIBIO + 1), "nfsfsync", slptimeo); 2870 if (error) { 2871 VI_UNLOCK(vp); 2872 error = nfs_sigintr(nmp, NULL, td); 2873 if (error) 2874 goto done; 2875 if (slpflag == PCATCH) { 2876 slpflag = 0; 2877 slptimeo = 2 * hz; 2878 } 2879 VI_LOCK(vp); 2880 } 2881 } 2882 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) && commit) { 2883 VI_UNLOCK(vp); 2884 goto loop; 2885 } 2886 } 2887 VI_UNLOCK(vp); 2888 if (np->n_flag & NWRITEERR) { 2889 error = np->n_error; 2890 np->n_flag &= ~NWRITEERR; 2891 } 2892 done: 2893 if (bvec != NULL && bvec != bvec_on_stack) 2894 free(bvec, M_TEMP); 2895 return (error); 2896 } 2897 2898 /* 2899 * NFS advisory byte-level locks. 2900 */ 2901 static int 2902 nfs_advlock(struct vop_advlock_args *ap) 2903 { 2904 2905 if ((VFSTONFS(ap->a_vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 2906 struct nfsnode *np = VTONFS(ap->a_vp); 2907 2908 return (lf_advlock(ap, &(np->n_lockf), np->n_size)); 2909 } 2910 return (nfs_dolock(ap)); 2911 } 2912 2913 /* 2914 * Print out the contents of an nfsnode. 2915 */ 2916 static int 2917 nfs_print(struct vop_print_args *ap) 2918 { 2919 struct vnode *vp = ap->a_vp; 2920 struct nfsnode *np = VTONFS(vp); 2921 2922 printf("\tfileid %ld fsid 0x%x", 2923 np->n_vattr.va_fileid, np->n_vattr.va_fsid); 2924 if (vp->v_type == VFIFO) 2925 fifo_printinfo(vp); 2926 printf("\n"); 2927 return (0); 2928 } 2929 2930 /* 2931 * This is the "real" nfs::bwrite(struct buf*). 2932 * B_WRITEINPROG isn't set unless the force flag is one and it 2933 * handles the B_NEEDCOMMIT flag. 2934 * We set B_CACHE if this is a VMIO buffer. 2935 */ 2936 int 2937 nfs_writebp(struct buf *bp, int force, struct thread *td) 2938 { 2939 int s; 2940 int oldflags = bp->b_flags; 2941 #if 0 2942 int retv = 1; 2943 off_t off; 2944 #endif 2945 2946 if (BUF_REFCNT(bp) == 0) 2947 panic("bwrite: buffer is not locked???"); 2948 2949 if (bp->b_flags & B_INVAL) { 2950 brelse(bp); 2951 return(0); 2952 } 2953 2954 bp->b_flags |= B_CACHE; 2955 2956 /* 2957 * Undirty the bp. We will redirty it later if the I/O fails. 2958 */ 2959 2960 s = splbio(); 2961 bundirty(bp); 2962 bp->b_flags &= ~B_DONE; 2963 bp->b_ioflags &= ~BIO_ERROR; 2964 bp->b_iocmd = BIO_WRITE; 2965 2966 VI_LOCK(bp->b_vp); 2967 bp->b_vp->v_numoutput++; 2968 VI_UNLOCK(bp->b_vp); 2969 curthread->td_proc->p_stats->p_ru.ru_oublock++; 2970 splx(s); 2971 2972 /* 2973 * Note: to avoid loopback deadlocks, we do not 2974 * assign b_runningbufspace. 2975 */ 2976 vfs_busy_pages(bp, 1); 2977 2978 if (force) 2979 bp->b_flags |= B_WRITEINPROG; 2980 BUF_KERNPROC(bp); 2981 bp->b_iooffset = dbtob(bp->b_blkno); 2982 VOP_STRATEGY(bp->b_vp, bp); 2983 2984 if( (oldflags & B_ASYNC) == 0) { 2985 int rtval = bufwait(bp); 2986 2987 if (oldflags & B_DELWRI) { 2988 s = splbio(); 2989 reassignbuf(bp); 2990 splx(s); 2991 } 2992 2993 brelse(bp); 2994 return (rtval); 2995 } 2996 2997 return (0); 2998 } 2999 3000 /* 3001 * nfs special file access vnode op. 3002 * Essentially just get vattr and then imitate iaccess() since the device is 3003 * local to the client. 3004 */ 3005 static int 3006 nfsspec_access(struct vop_access_args *ap) 3007 { 3008 struct vattr *vap; 3009 struct ucred *cred = ap->a_cred; 3010 struct vnode *vp = ap->a_vp; 3011 mode_t mode = ap->a_mode; 3012 struct vattr vattr; 3013 int error; 3014 3015 /* 3016 * Disallow write attempts on filesystems mounted read-only; 3017 * unless the file is a socket, fifo, or a block or character 3018 * device resident on the filesystem. 3019 */ 3020 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3021 switch (vp->v_type) { 3022 case VREG: 3023 case VDIR: 3024 case VLNK: 3025 return (EROFS); 3026 default: 3027 break; 3028 } 3029 } 3030 vap = &vattr; 3031 error = VOP_GETATTR(vp, vap, cred, ap->a_td); 3032 if (error) 3033 return (error); 3034 return (vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid, 3035 mode, cred, NULL)); 3036 } 3037 3038 /* 3039 * Read wrapper for special devices. 3040 */ 3041 static int 3042 nfsspec_read(struct vop_read_args *ap) 3043 { 3044 struct nfsnode *np = VTONFS(ap->a_vp); 3045 3046 /* 3047 * Set access flag. 3048 */ 3049 np->n_flag |= NACC; 3050 getnanotime(&np->n_atim); 3051 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap)); 3052 } 3053 3054 /* 3055 * Write wrapper for special devices. 3056 */ 3057 static int 3058 nfsspec_write(struct vop_write_args *ap) 3059 { 3060 struct nfsnode *np = VTONFS(ap->a_vp); 3061 3062 /* 3063 * Set update flag. 3064 */ 3065 np->n_flag |= NUPD; 3066 getnanotime(&np->n_mtim); 3067 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap)); 3068 } 3069 3070 /* 3071 * Close wrapper for special devices. 3072 * 3073 * Update the times on the nfsnode then do device close. 3074 */ 3075 static int 3076 nfsspec_close(struct vop_close_args *ap) 3077 { 3078 struct vnode *vp = ap->a_vp; 3079 struct nfsnode *np = VTONFS(vp); 3080 struct vattr vattr; 3081 3082 if (np->n_flag & (NACC | NUPD)) { 3083 np->n_flag |= NCHG; 3084 if (vrefcnt(vp) == 1 && 3085 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3086 VATTR_NULL(&vattr); 3087 if (np->n_flag & NACC) 3088 vattr.va_atime = np->n_atim; 3089 if (np->n_flag & NUPD) 3090 vattr.va_mtime = np->n_mtim; 3091 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_td); 3092 } 3093 } 3094 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap)); 3095 } 3096 3097 /* 3098 * Read wrapper for fifos. 3099 */ 3100 static int 3101 nfsfifo_read(struct vop_read_args *ap) 3102 { 3103 struct nfsnode *np = VTONFS(ap->a_vp); 3104 3105 /* 3106 * Set access flag. 3107 */ 3108 np->n_flag |= NACC; 3109 getnanotime(&np->n_atim); 3110 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap)); 3111 } 3112 3113 /* 3114 * Write wrapper for fifos. 3115 */ 3116 static int 3117 nfsfifo_write(struct vop_write_args *ap) 3118 { 3119 struct nfsnode *np = VTONFS(ap->a_vp); 3120 3121 /* 3122 * Set update flag. 3123 */ 3124 np->n_flag |= NUPD; 3125 getnanotime(&np->n_mtim); 3126 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap)); 3127 } 3128 3129 /* 3130 * Close wrapper for fifos. 3131 * 3132 * Update the times on the nfsnode then do fifo close. 3133 */ 3134 static int 3135 nfsfifo_close(struct vop_close_args *ap) 3136 { 3137 struct vnode *vp = ap->a_vp; 3138 struct nfsnode *np = VTONFS(vp); 3139 struct vattr vattr; 3140 struct timespec ts; 3141 3142 if (np->n_flag & (NACC | NUPD)) { 3143 getnanotime(&ts); 3144 if (np->n_flag & NACC) 3145 np->n_atim = ts; 3146 if (np->n_flag & NUPD) 3147 np->n_mtim = ts; 3148 np->n_flag |= NCHG; 3149 if (vrefcnt(vp) == 1 && 3150 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3151 VATTR_NULL(&vattr); 3152 if (np->n_flag & NACC) 3153 vattr.va_atime = np->n_atim; 3154 if (np->n_flag & NUPD) 3155 vattr.va_mtime = np->n_mtim; 3156 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, ap->a_td); 3157 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_td); 3158 VOP_UNLOCK(vp, 0, ap->a_td); 3159 } 3160 } 3161 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap)); 3162 } 3163
Cache object: 7b043f2b9d1c89b70e1f5b6b515c97d9
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]