nfs_vnops.c revision 232821
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: head/sys/nfsclient/nfs_vnops.c 232821 2012-03-11 12:19:58Z kib $"); 37 38/* 39 * vnode op calls for Sun NFS version 2 and 3 40 */ 41 42#include "opt_inet.h" 43#include "opt_kdtrace.h" 44 45#include <sys/param.h> 46#include <sys/kernel.h> 47#include <sys/systm.h> 48#include <sys/resourcevar.h> 49#include <sys/proc.h> 50#include <sys/mount.h> 51#include <sys/bio.h> 52#include <sys/buf.h> 53#include <sys/jail.h> 54#include <sys/malloc.h> 55#include <sys/mbuf.h> 56#include <sys/namei.h> 57#include <sys/socket.h> 58#include <sys/vnode.h> 59#include <sys/dirent.h> 60#include <sys/fcntl.h> 61#include <sys/lockf.h> 62#include <sys/stat.h> 63#include <sys/sysctl.h> 64#include <sys/signalvar.h> 65 66#include <vm/vm.h> 67#include <vm/vm_extern.h> 68#include <vm/vm_object.h> 69 70#include <nfs/nfsproto.h> 71#include <nfsclient/nfs.h> 72#include <nfsclient/nfsnode.h> 73#include <nfsclient/nfsmount.h> 74#include <nfs/nfs_kdtrace.h> 75#include <nfs/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#include <machine/stdarg.h> 84 85#ifdef KDTRACE_HOOKS 86#include <sys/dtrace_bsd.h> 87 88dtrace_nfsclient_accesscache_flush_probe_func_t 89 dtrace_nfsclient_accesscache_flush_done_probe; 90uint32_t nfsclient_accesscache_flush_done_id; 91 92dtrace_nfsclient_accesscache_get_probe_func_t 93 dtrace_nfsclient_accesscache_get_hit_probe, 94 dtrace_nfsclient_accesscache_get_miss_probe; 95uint32_t nfsclient_accesscache_get_hit_id; 96uint32_t nfsclient_accesscache_get_miss_id; 97 98dtrace_nfsclient_accesscache_load_probe_func_t 99 dtrace_nfsclient_accesscache_load_done_probe; 100uint32_t nfsclient_accesscache_load_done_id; 101#endif /* !KDTRACE_HOOKS */ 102 103/* Defs */ 104#define TRUE 1 105#define FALSE 0 106 107/* 108 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these 109 * calls are not in getblk() and brelse() so that they would not be necessary 110 * here. 111 */ 112#ifndef B_VMIO 113#define vfs_busy_pages(bp, f) 114#endif 115 116static vop_read_t nfsfifo_read; 117static vop_write_t nfsfifo_write; 118static vop_close_t nfsfifo_close; 119static int nfs_flush(struct vnode *, int, int); 120static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *); 121static vop_lookup_t nfs_lookup; 122static vop_create_t nfs_create; 123static vop_mknod_t nfs_mknod; 124static vop_open_t nfs_open; 125static vop_close_t nfs_close; 126static vop_access_t nfs_access; 127static vop_getattr_t nfs_getattr; 128static vop_setattr_t nfs_setattr; 129static vop_read_t nfs_read; 130static vop_fsync_t nfs_fsync; 131static vop_remove_t nfs_remove; 132static vop_link_t nfs_link; 133static vop_rename_t nfs_rename; 134static vop_mkdir_t nfs_mkdir; 135static vop_rmdir_t nfs_rmdir; 136static vop_symlink_t nfs_symlink; 137static vop_readdir_t nfs_readdir; 138static vop_strategy_t nfs_strategy; 139static int nfs_lookitup(struct vnode *, const char *, int, 140 struct ucred *, struct thread *, struct nfsnode **); 141static int nfs_sillyrename(struct vnode *, struct vnode *, 142 struct componentname *); 143static vop_access_t nfsspec_access; 144static vop_readlink_t nfs_readlink; 145static vop_print_t nfs_print; 146static vop_advlock_t nfs_advlock; 147static vop_advlockasync_t nfs_advlockasync; 148 149/* 150 * Global vfs data structures for nfs 151 */ 152struct vop_vector nfs_vnodeops = { 153 .vop_default = &default_vnodeops, 154 .vop_access = nfs_access, 155 .vop_advlock = nfs_advlock, 156 .vop_advlockasync = nfs_advlockasync, 157 .vop_close = nfs_close, 158 .vop_create = nfs_create, 159 .vop_fsync = nfs_fsync, 160 .vop_getattr = nfs_getattr, 161 .vop_getpages = nfs_getpages, 162 .vop_putpages = nfs_putpages, 163 .vop_inactive = nfs_inactive, 164 .vop_link = nfs_link, 165 .vop_lookup = nfs_lookup, 166 .vop_mkdir = nfs_mkdir, 167 .vop_mknod = nfs_mknod, 168 .vop_open = nfs_open, 169 .vop_print = nfs_print, 170 .vop_read = nfs_read, 171 .vop_readdir = nfs_readdir, 172 .vop_readlink = nfs_readlink, 173 .vop_reclaim = nfs_reclaim, 174 .vop_remove = nfs_remove, 175 .vop_rename = nfs_rename, 176 .vop_rmdir = nfs_rmdir, 177 .vop_setattr = nfs_setattr, 178 .vop_strategy = nfs_strategy, 179 .vop_symlink = nfs_symlink, 180 .vop_write = nfs_write, 181}; 182 183struct vop_vector nfs_fifoops = { 184 .vop_default = &fifo_specops, 185 .vop_access = nfsspec_access, 186 .vop_close = nfsfifo_close, 187 .vop_fsync = nfs_fsync, 188 .vop_getattr = nfs_getattr, 189 .vop_inactive = nfs_inactive, 190 .vop_print = nfs_print, 191 .vop_read = nfsfifo_read, 192 .vop_reclaim = nfs_reclaim, 193 .vop_setattr = nfs_setattr, 194 .vop_write = nfsfifo_write, 195}; 196 197static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, 198 struct componentname *cnp, struct vattr *vap); 199static int nfs_removerpc(struct vnode *dvp, const char *name, int namelen, 200 struct ucred *cred, struct thread *td); 201static int nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, 202 int fnamelen, struct vnode *tdvp, 203 const char *tnameptr, int tnamelen, 204 struct ucred *cred, struct thread *td); 205static int nfs_renameit(struct vnode *sdvp, struct componentname *scnp, 206 struct sillyrename *sp); 207 208/* 209 * Global variables 210 */ 211struct mtx nfs_iod_mtx; 212enum nfsiod_state nfs_iodwant[NFS_MAXASYNCDAEMON]; 213struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON]; 214int nfs_numasync = 0; 215#define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1)) 216 217SYSCTL_DECL(_vfs_oldnfs); 218 219static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO; 220SYSCTL_INT(_vfs_oldnfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 221 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 222 223static int nfs_prime_access_cache = 0; 224SYSCTL_INT(_vfs_oldnfs, OID_AUTO, prime_access_cache, CTLFLAG_RW, 225 &nfs_prime_access_cache, 0, 226 "Prime NFS ACCESS cache when fetching attributes"); 227 228static int nfsv3_commit_on_close = 0; 229SYSCTL_INT(_vfs_oldnfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW, 230 &nfsv3_commit_on_close, 0, "write+commit on close, else only write"); 231 232static int nfs_clean_pages_on_close = 1; 233SYSCTL_INT(_vfs_oldnfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW, 234 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close"); 235 236int nfs_directio_enable = 0; 237SYSCTL_INT(_vfs_oldnfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW, 238 &nfs_directio_enable, 0, "Enable NFS directio"); 239 240/* 241 * This sysctl allows other processes to mmap a file that has been opened 242 * O_DIRECT by a process. In general, having processes mmap the file while 243 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow 244 * this by default to prevent DoS attacks - to prevent a malicious user from 245 * opening up files O_DIRECT preventing other users from mmap'ing these 246 * files. "Protected" environments where stricter consistency guarantees are 247 * required can disable this knob. The process that opened the file O_DIRECT 248 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not 249 * meaningful. 250 */ 251int nfs_directio_allow_mmap = 1; 252SYSCTL_INT(_vfs_oldnfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW, 253 &nfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens"); 254 255#if 0 256SYSCTL_INT(_vfs_oldnfs, OID_AUTO, access_cache_hits, CTLFLAG_RD, 257 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count"); 258 259SYSCTL_INT(_vfs_oldnfs, OID_AUTO, access_cache_misses, CTLFLAG_RD, 260 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count"); 261#endif 262 263#define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \ 264 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \ 265 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP) 266 267/* 268 * SMP Locking Note : 269 * The list of locks after the description of the lock is the ordering 270 * of other locks acquired with the lock held. 271 * np->n_mtx : Protects the fields in the nfsnode. 272 VM Object Lock 273 VI_MTX (acquired indirectly) 274 * nmp->nm_mtx : Protects the fields in the nfsmount. 275 rep->r_mtx 276 * nfs_iod_mtx : Global lock, protects shared nfsiod state. 277 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list. 278 nmp->nm_mtx 279 rep->r_mtx 280 * rep->r_mtx : Protects the fields in an nfsreq. 281 */ 282 283static int 284nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td, 285 struct ucred *cred, uint32_t *retmode) 286{ 287 const int v3 = 1; 288 u_int32_t *tl; 289 int error = 0, attrflag, i, lrupos; 290 291 struct mbuf *mreq, *mrep, *md, *mb; 292 caddr_t bpos, dpos; 293 u_int32_t rmode; 294 struct nfsnode *np = VTONFS(vp); 295 296 nfsstats.rpccnt[NFSPROC_ACCESS]++; 297 mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 298 mb = mreq; 299 bpos = mtod(mb, caddr_t); 300 nfsm_fhtom(vp, v3); 301 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED); 302 *tl = txdr_unsigned(wmode); 303 nfsm_request(vp, NFSPROC_ACCESS, td, cred); 304 nfsm_postop_attr(vp, attrflag); 305 if (!error) { 306 lrupos = 0; 307 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); 308 rmode = fxdr_unsigned(u_int32_t, *tl); 309 mtx_lock(&np->n_mtx); 310 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 311 if (np->n_accesscache[i].uid == cred->cr_uid) { 312 np->n_accesscache[i].mode = rmode; 313 np->n_accesscache[i].stamp = time_second; 314 break; 315 } 316 if (i > 0 && np->n_accesscache[i].stamp < 317 np->n_accesscache[lrupos].stamp) 318 lrupos = i; 319 } 320 if (i == NFS_ACCESSCACHESIZE) { 321 np->n_accesscache[lrupos].uid = cred->cr_uid; 322 np->n_accesscache[lrupos].mode = rmode; 323 np->n_accesscache[lrupos].stamp = time_second; 324 } 325 mtx_unlock(&np->n_mtx); 326 if (retmode != NULL) 327 *retmode = rmode; 328 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0); 329 } 330 m_freem(mrep); 331nfsmout: 332#ifdef KDTRACE_HOOKS 333 if (error) { 334 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0, 335 error); 336 } 337#endif 338 return (error); 339} 340 341/* 342 * nfs access vnode op. 343 * For nfs version 2, just return ok. File accesses may fail later. 344 * For nfs version 3, use the access rpc to check accessibility. If file modes 345 * are changed on the server, accesses might still fail later. 346 */ 347static int 348nfs_access(struct vop_access_args *ap) 349{ 350 struct vnode *vp = ap->a_vp; 351 int error = 0, i, gotahit; 352 u_int32_t mode, rmode, wmode; 353 int v3 = NFS_ISV3(vp); 354 struct nfsnode *np = VTONFS(vp); 355 356 /* 357 * Disallow write attempts on filesystems mounted read-only; 358 * unless the file is a socket, fifo, or a block or character 359 * device resident on the filesystem. 360 */ 361 if ((ap->a_accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 362 switch (vp->v_type) { 363 case VREG: 364 case VDIR: 365 case VLNK: 366 return (EROFS); 367 default: 368 break; 369 } 370 } 371 /* 372 * For nfs v3, check to see if we have done this recently, and if 373 * so return our cached result instead of making an ACCESS call. 374 * If not, do an access rpc, otherwise you are stuck emulating 375 * ufs_access() locally using the vattr. This may not be correct, 376 * since the server may apply other access criteria such as 377 * client uid-->server uid mapping that we do not know about. 378 */ 379 if (v3) { 380 if (ap->a_accmode & VREAD) 381 mode = NFSV3ACCESS_READ; 382 else 383 mode = 0; 384 if (vp->v_type != VDIR) { 385 if (ap->a_accmode & VWRITE) 386 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 387 if (ap->a_accmode & VEXEC) 388 mode |= NFSV3ACCESS_EXECUTE; 389 } else { 390 if (ap->a_accmode & VWRITE) 391 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 392 NFSV3ACCESS_DELETE); 393 if (ap->a_accmode & VEXEC) 394 mode |= NFSV3ACCESS_LOOKUP; 395 } 396 /* XXX safety belt, only make blanket request if caching */ 397 if (nfsaccess_cache_timeout > 0) { 398 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY | 399 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE | 400 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP; 401 } else { 402 wmode = mode; 403 } 404 405 /* 406 * Does our cached result allow us to give a definite yes to 407 * this request? 408 */ 409 gotahit = 0; 410 mtx_lock(&np->n_mtx); 411 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 412 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) { 413 if (time_second < (np->n_accesscache[i].stamp + 414 nfsaccess_cache_timeout) && 415 (np->n_accesscache[i].mode & mode) == mode) { 416 nfsstats.accesscache_hits++; 417 gotahit = 1; 418 } 419 break; 420 } 421 } 422 mtx_unlock(&np->n_mtx); 423#ifdef KDTRACE_HOOKS 424 if (gotahit) 425 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp, 426 ap->a_cred->cr_uid, mode); 427 else 428 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp, 429 ap->a_cred->cr_uid, mode); 430#endif 431 if (gotahit == 0) { 432 /* 433 * Either a no, or a don't know. Go to the wire. 434 */ 435 nfsstats.accesscache_misses++; 436 error = nfs3_access_otw(vp, wmode, ap->a_td, ap->a_cred, 437 &rmode); 438 if (!error) { 439 if ((rmode & mode) != mode) 440 error = EACCES; 441 } 442 } 443 return (error); 444 } else { 445 if ((error = nfsspec_access(ap)) != 0) { 446 return (error); 447 } 448 /* 449 * Attempt to prevent a mapped root from accessing a file 450 * which it shouldn't. We try to read a byte from the file 451 * if the user is root and the file is not zero length. 452 * After calling nfsspec_access, we should have the correct 453 * file size cached. 454 */ 455 mtx_lock(&np->n_mtx); 456 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD) 457 && VTONFS(vp)->n_size > 0) { 458 struct iovec aiov; 459 struct uio auio; 460 char buf[1]; 461 462 mtx_unlock(&np->n_mtx); 463 aiov.iov_base = buf; 464 aiov.iov_len = 1; 465 auio.uio_iov = &aiov; 466 auio.uio_iovcnt = 1; 467 auio.uio_offset = 0; 468 auio.uio_resid = 1; 469 auio.uio_segflg = UIO_SYSSPACE; 470 auio.uio_rw = UIO_READ; 471 auio.uio_td = ap->a_td; 472 473 if (vp->v_type == VREG) 474 error = nfs_readrpc(vp, &auio, ap->a_cred); 475 else if (vp->v_type == VDIR) { 476 char* bp; 477 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 478 aiov.iov_base = bp; 479 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 480 error = nfs_readdirrpc(vp, &auio, ap->a_cred); 481 free(bp, M_TEMP); 482 } else if (vp->v_type == VLNK) 483 error = nfs_readlinkrpc(vp, &auio, ap->a_cred); 484 else 485 error = EACCES; 486 } else 487 mtx_unlock(&np->n_mtx); 488 return (error); 489 } 490} 491 492int nfs_otw_getattr_avoid = 0; 493 494/* 495 * nfs open vnode op 496 * Check to see if the type is ok 497 * and that deletion is not in progress. 498 * For paged in text files, you will need to flush the page cache 499 * if consistency is lost. 500 */ 501/* ARGSUSED */ 502static int 503nfs_open(struct vop_open_args *ap) 504{ 505 struct vnode *vp = ap->a_vp; 506 struct nfsnode *np = VTONFS(vp); 507 struct vattr vattr; 508 int error; 509 int fmode = ap->a_mode; 510 511 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) 512 return (EOPNOTSUPP); 513 514 /* 515 * Get a valid lease. If cached data is stale, flush it. 516 */ 517 mtx_lock(&np->n_mtx); 518 if (np->n_flag & NMODIFIED) { 519 mtx_unlock(&np->n_mtx); 520 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 521 if (error == EINTR || error == EIO) 522 return (error); 523 mtx_lock(&np->n_mtx); 524 np->n_attrstamp = 0; 525 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 526 if (vp->v_type == VDIR) 527 np->n_direofoffset = 0; 528 mtx_unlock(&np->n_mtx); 529 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 530 if (error) 531 return (error); 532 mtx_lock(&np->n_mtx); 533 np->n_mtime = vattr.va_mtime; 534 } else { 535 mtx_unlock(&np->n_mtx); 536 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 537 if (error) 538 return (error); 539 mtx_lock(&np->n_mtx); 540 if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 541 if (vp->v_type == VDIR) 542 np->n_direofoffset = 0; 543 mtx_unlock(&np->n_mtx); 544 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 545 if (error == EINTR || error == EIO) { 546 return (error); 547 } 548 mtx_lock(&np->n_mtx); 549 np->n_mtime = vattr.va_mtime; 550 } 551 } 552 /* 553 * If the object has >= 1 O_DIRECT active opens, we disable caching. 554 */ 555 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) { 556 if (np->n_directio_opens == 0) { 557 mtx_unlock(&np->n_mtx); 558 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 559 if (error) 560 return (error); 561 mtx_lock(&np->n_mtx); 562 np->n_flag |= NNONCACHE; 563 } 564 np->n_directio_opens++; 565 } 566 mtx_unlock(&np->n_mtx); 567 vnode_create_vobject(vp, vattr.va_size, ap->a_td); 568 return (0); 569} 570 571/* 572 * nfs close vnode op 573 * What an NFS client should do upon close after writing is a debatable issue. 574 * Most NFS clients push delayed writes to the server upon close, basically for 575 * two reasons: 576 * 1 - So that any write errors may be reported back to the client process 577 * doing the close system call. By far the two most likely errors are 578 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 579 * 2 - To put a worst case upper bound on cache inconsistency between 580 * multiple clients for the file. 581 * There is also a consistency problem for Version 2 of the protocol w.r.t. 582 * not being able to tell if other clients are writing a file concurrently, 583 * since there is no way of knowing if the changed modify time in the reply 584 * is only due to the write for this client. 585 * (NFS Version 3 provides weak cache consistency data in the reply that 586 * should be sufficient to detect and handle this case.) 587 * 588 * The current code does the following: 589 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 590 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 591 * or commit them (this satisfies 1 and 2 except for the 592 * case where the server crashes after this close but 593 * before the commit RPC, which is felt to be "good 594 * enough". Changing the last argument to nfs_flush() to 595 * a 1 would force a commit operation, if it is felt a 596 * commit is necessary now. 597 */ 598/* ARGSUSED */ 599static int 600nfs_close(struct vop_close_args *ap) 601{ 602 struct vnode *vp = ap->a_vp; 603 struct nfsnode *np = VTONFS(vp); 604 int error = 0; 605 int fmode = ap->a_fflag; 606 607 if (vp->v_type == VREG) { 608 /* 609 * Examine and clean dirty pages, regardless of NMODIFIED. 610 * This closes a major hole in close-to-open consistency. 611 * We want to push out all dirty pages (and buffers) on 612 * close, regardless of whether they were dirtied by 613 * mmap'ed writes or via write(). 614 */ 615 if (nfs_clean_pages_on_close && vp->v_object) { 616 VM_OBJECT_LOCK(vp->v_object); 617 vm_object_page_clean(vp->v_object, 0, 0, 0); 618 VM_OBJECT_UNLOCK(vp->v_object); 619 } 620 mtx_lock(&np->n_mtx); 621 if (np->n_flag & NMODIFIED) { 622 mtx_unlock(&np->n_mtx); 623 if (NFS_ISV3(vp)) { 624 /* 625 * Under NFSv3 we have dirty buffers to dispose of. We 626 * must flush them to the NFS server. We have the option 627 * of waiting all the way through the commit rpc or just 628 * waiting for the initial write. The default is to only 629 * wait through the initial write so the data is in the 630 * server's cache, which is roughly similar to the state 631 * a standard disk subsystem leaves the file in on close(). 632 * 633 * We cannot clear the NMODIFIED bit in np->n_flag due to 634 * potential races with other processes, and certainly 635 * cannot clear it if we don't commit. 636 */ 637 int cm = nfsv3_commit_on_close ? 1 : 0; 638 error = nfs_flush(vp, MNT_WAIT, cm); 639 /* np->n_flag &= ~NMODIFIED; */ 640 } else 641 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 642 mtx_lock(&np->n_mtx); 643 } 644 if (np->n_flag & NWRITEERR) { 645 np->n_flag &= ~NWRITEERR; 646 error = np->n_error; 647 } 648 mtx_unlock(&np->n_mtx); 649 } 650 if (nfs_directio_enable) 651 KASSERT((np->n_directio_asyncwr == 0), 652 ("nfs_close: dirty unflushed (%d) directio buffers\n", 653 np->n_directio_asyncwr)); 654 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) { 655 mtx_lock(&np->n_mtx); 656 KASSERT((np->n_directio_opens > 0), 657 ("nfs_close: unexpectedly value (0) of n_directio_opens\n")); 658 np->n_directio_opens--; 659 if (np->n_directio_opens == 0) 660 np->n_flag &= ~NNONCACHE; 661 mtx_unlock(&np->n_mtx); 662 } 663 return (error); 664} 665 666/* 667 * nfs getattr call from vfs. 668 */ 669static int 670nfs_getattr(struct vop_getattr_args *ap) 671{ 672 struct vnode *vp = ap->a_vp; 673 struct nfsnode *np = VTONFS(vp); 674 struct thread *td = curthread; 675 struct vattr *vap = ap->a_vap; 676 struct vattr vattr; 677 caddr_t bpos, dpos; 678 int error = 0; 679 struct mbuf *mreq, *mrep, *md, *mb; 680 int v3 = NFS_ISV3(vp); 681 682 /* 683 * Update local times for special files. 684 */ 685 mtx_lock(&np->n_mtx); 686 if (np->n_flag & (NACC | NUPD)) 687 np->n_flag |= NCHG; 688 mtx_unlock(&np->n_mtx); 689 /* 690 * First look in the cache. 691 */ 692 if (nfs_getattrcache(vp, &vattr) == 0) 693 goto nfsmout; 694 if (v3 && nfs_prime_access_cache && nfsaccess_cache_timeout > 0) { 695 nfsstats.accesscache_misses++; 696 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, ap->a_cred, NULL); 697 if (nfs_getattrcache(vp, &vattr) == 0) 698 goto nfsmout; 699 } 700 nfsstats.rpccnt[NFSPROC_GETATTR]++; 701 mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3)); 702 mb = mreq; 703 bpos = mtod(mb, caddr_t); 704 nfsm_fhtom(vp, v3); 705 nfsm_request(vp, NFSPROC_GETATTR, td, ap->a_cred); 706 if (!error) { 707 nfsm_loadattr(vp, &vattr); 708 } 709 m_freem(mrep); 710nfsmout: 711 vap->va_type = vattr.va_type; 712 vap->va_mode = vattr.va_mode; 713 vap->va_nlink = vattr.va_nlink; 714 vap->va_uid = vattr.va_uid; 715 vap->va_gid = vattr.va_gid; 716 vap->va_fsid = vattr.va_fsid; 717 vap->va_fileid = vattr.va_fileid; 718 vap->va_size = vattr.va_size; 719 vap->va_blocksize = vattr.va_blocksize; 720 vap->va_atime = vattr.va_atime; 721 vap->va_mtime = vattr.va_mtime; 722 vap->va_ctime = vattr.va_ctime; 723 vap->va_gen = vattr.va_gen; 724 vap->va_flags = vattr.va_flags; 725 vap->va_rdev = vattr.va_rdev; 726 vap->va_bytes = vattr.va_bytes; 727 vap->va_filerev = vattr.va_filerev; 728 729 return (error); 730} 731 732/* 733 * nfs setattr call. 734 */ 735static int 736nfs_setattr(struct vop_setattr_args *ap) 737{ 738 struct vnode *vp = ap->a_vp; 739 struct nfsnode *np = VTONFS(vp); 740 struct vattr *vap = ap->a_vap; 741 struct thread *td = curthread; 742 int error = 0; 743 u_quad_t tsize; 744 745#ifndef nolint 746 tsize = (u_quad_t)0; 747#endif 748 749 /* 750 * Setting of flags is not supported. 751 */ 752 if (vap->va_flags != VNOVAL) 753 return (EOPNOTSUPP); 754 755 /* 756 * Disallow write attempts if the filesystem is mounted read-only. 757 */ 758 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 759 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 760 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 761 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 762 error = EROFS; 763 goto out; 764 } 765 if (vap->va_size != VNOVAL) { 766 switch (vp->v_type) { 767 case VDIR: 768 return (EISDIR); 769 case VCHR: 770 case VBLK: 771 case VSOCK: 772 case VFIFO: 773 if (vap->va_mtime.tv_sec == VNOVAL && 774 vap->va_atime.tv_sec == VNOVAL && 775 vap->va_mode == (mode_t)VNOVAL && 776 vap->va_uid == (uid_t)VNOVAL && 777 vap->va_gid == (gid_t)VNOVAL) 778 return (0); 779 vap->va_size = VNOVAL; 780 break; 781 default: 782 /* 783 * Disallow write attempts if the filesystem is 784 * mounted read-only. 785 */ 786 if (vp->v_mount->mnt_flag & MNT_RDONLY) 787 return (EROFS); 788 /* 789 * We run vnode_pager_setsize() early (why?), 790 * we must set np->n_size now to avoid vinvalbuf 791 * V_SAVE races that might setsize a lower 792 * value. 793 */ 794 mtx_lock(&np->n_mtx); 795 tsize = np->n_size; 796 mtx_unlock(&np->n_mtx); 797 error = nfs_meta_setsize(vp, ap->a_cred, td, 798 vap->va_size); 799 mtx_lock(&np->n_mtx); 800 if (np->n_flag & NMODIFIED) { 801 tsize = np->n_size; 802 mtx_unlock(&np->n_mtx); 803 if (vap->va_size == 0) 804 error = nfs_vinvalbuf(vp, 0, td, 1); 805 else 806 error = nfs_vinvalbuf(vp, V_SAVE, td, 1); 807 if (error) { 808 vnode_pager_setsize(vp, tsize); 809 goto out; 810 } 811 } else 812 mtx_unlock(&np->n_mtx); 813 /* 814 * np->n_size has already been set to vap->va_size 815 * in nfs_meta_setsize(). We must set it again since 816 * nfs_loadattrcache() could be called through 817 * nfs_meta_setsize() and could modify np->n_size. 818 */ 819 mtx_lock(&np->n_mtx); 820 np->n_vattr.va_size = np->n_size = vap->va_size; 821 mtx_unlock(&np->n_mtx); 822 }; 823 } else { 824 mtx_lock(&np->n_mtx); 825 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) && 826 (np->n_flag & NMODIFIED) && vp->v_type == VREG) { 827 mtx_unlock(&np->n_mtx); 828 if ((error = nfs_vinvalbuf(vp, V_SAVE, td, 1)) != 0 && 829 (error == EINTR || error == EIO)) 830 return error; 831 } else 832 mtx_unlock(&np->n_mtx); 833 } 834 error = nfs_setattrrpc(vp, vap, ap->a_cred); 835 if (error && vap->va_size != VNOVAL) { 836 mtx_lock(&np->n_mtx); 837 np->n_size = np->n_vattr.va_size = tsize; 838 vnode_pager_setsize(vp, tsize); 839 mtx_unlock(&np->n_mtx); 840 } 841out: 842 return (error); 843} 844 845/* 846 * Do an nfs setattr rpc. 847 */ 848static int 849nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred) 850{ 851 struct nfsv2_sattr *sp; 852 struct nfsnode *np = VTONFS(vp); 853 caddr_t bpos, dpos; 854 u_int32_t *tl; 855 int error = 0, i, wccflag = NFSV3_WCCRATTR; 856 struct mbuf *mreq, *mrep, *md, *mb; 857 int v3 = NFS_ISV3(vp); 858 859 nfsstats.rpccnt[NFSPROC_SETATTR]++; 860 mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 861 mb = mreq; 862 bpos = mtod(mb, caddr_t); 863 nfsm_fhtom(vp, v3); 864 if (v3) { 865 nfsm_v3attrbuild(vap, TRUE); 866 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED); 867 *tl = nfs_false; 868 } else { 869 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR); 870 if (vap->va_mode == (mode_t)VNOVAL) 871 sp->sa_mode = nfs_xdrneg1; 872 else 873 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 874 if (vap->va_uid == (uid_t)VNOVAL) 875 sp->sa_uid = nfs_xdrneg1; 876 else 877 sp->sa_uid = txdr_unsigned(vap->va_uid); 878 if (vap->va_gid == (gid_t)VNOVAL) 879 sp->sa_gid = nfs_xdrneg1; 880 else 881 sp->sa_gid = txdr_unsigned(vap->va_gid); 882 sp->sa_size = txdr_unsigned(vap->va_size); 883 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 884 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 885 } 886 nfsm_request(vp, NFSPROC_SETATTR, curthread, cred); 887 if (v3) { 888 mtx_lock(&np->n_mtx); 889 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) 890 np->n_accesscache[i].stamp = 0; 891 mtx_unlock(&np->n_mtx); 892 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp); 893 nfsm_wcc_data(vp, wccflag); 894 } else 895 nfsm_loadattr(vp, NULL); 896 m_freem(mrep); 897nfsmout: 898 return (error); 899} 900 901/* 902 * nfs lookup call, one step at a time... 903 * First look in cache 904 * If not found, unlock the directory nfsnode and do the rpc 905 */ 906static int 907nfs_lookup(struct vop_lookup_args *ap) 908{ 909 struct componentname *cnp = ap->a_cnp; 910 struct vnode *dvp = ap->a_dvp; 911 struct vnode **vpp = ap->a_vpp; 912 struct mount *mp = dvp->v_mount; 913 struct vattr dvattr, vattr; 914 struct timespec nctime; 915 int flags = cnp->cn_flags; 916 struct vnode *newvp; 917 struct nfsmount *nmp; 918 caddr_t bpos, dpos; 919 struct mbuf *mreq, *mrep, *md, *mb; 920 long len; 921 nfsfh_t *fhp; 922 struct nfsnode *np, *newnp; 923 int error = 0, attrflag, dattrflag, fhsize, ltype, ncticks; 924 int v3 = NFS_ISV3(dvp); 925 struct thread *td = cnp->cn_thread; 926 927 *vpp = NULLVP; 928 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) && 929 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 930 return (EROFS); 931 if (dvp->v_type != VDIR) 932 return (ENOTDIR); 933 nmp = VFSTONFS(mp); 934 np = VTONFS(dvp); 935 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) { 936 *vpp = NULLVP; 937 return (error); 938 } 939 error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks); 940 if (error > 0 && error != ENOENT) 941 return (error); 942 if (error == -1) { 943 /* 944 * Lookups of "." are special and always return the 945 * current directory. cache_lookup() already handles 946 * associated locking bookkeeping, etc. 947 */ 948 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 949 /* XXX: Is this really correct? */ 950 if (cnp->cn_nameiop != LOOKUP && 951 (flags & ISLASTCN)) 952 cnp->cn_flags |= SAVENAME; 953 return (0); 954 } 955 956 /* 957 * We only accept a positive hit in the cache if the 958 * change time of the file matches our cached copy. 959 * Otherwise, we discard the cache entry and fallback 960 * to doing a lookup RPC. We also only trust cache 961 * entries for less than nm_nametimeo seconds. 962 * 963 * To better handle stale file handles and attributes, 964 * clear the attribute cache of this node if it is a 965 * leaf component, part of an open() call, and not 966 * locally modified before fetching the attributes. 967 * This should allow stale file handles to be detected 968 * here where we can fall back to a LOOKUP RPC to 969 * recover rather than having nfs_open() detect the 970 * stale file handle and failing open(2) with ESTALE. 971 */ 972 newvp = *vpp; 973 newnp = VTONFS(newvp); 974 if (!(nmp->nm_flag & NFSMNT_NOCTO) && 975 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 976 !(newnp->n_flag & NMODIFIED)) { 977 mtx_lock(&newnp->n_mtx); 978 newnp->n_attrstamp = 0; 979 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 980 mtx_unlock(&newnp->n_mtx); 981 } 982 if ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) && 983 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 && 984 timespeccmp(&vattr.va_ctime, &nctime, ==)) { 985 nfsstats.lookupcache_hits++; 986 if (cnp->cn_nameiop != LOOKUP && 987 (flags & ISLASTCN)) 988 cnp->cn_flags |= SAVENAME; 989 return (0); 990 } 991 cache_purge(newvp); 992 if (dvp != newvp) 993 vput(newvp); 994 else 995 vrele(newvp); 996 *vpp = NULLVP; 997 } else if (error == ENOENT) { 998 if (dvp->v_iflag & VI_DOOMED) 999 return (ENOENT); 1000 /* 1001 * We only accept a negative hit in the cache if the 1002 * modification time of the parent directory matches 1003 * the cached copy in the name cache entry. 1004 * Otherwise, we discard all of the negative cache 1005 * entries for this directory. We also only trust 1006 * negative cache entries for up to nm_negnametimeo 1007 * seconds. 1008 */ 1009 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) && 1010 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 && 1011 timespeccmp(&vattr.va_mtime, &nctime, ==)) { 1012 nfsstats.lookupcache_hits++; 1013 return (ENOENT); 1014 } 1015 cache_purge_negative(dvp); 1016 } 1017 1018 attrflag = dattrflag = 0; 1019 error = 0; 1020 newvp = NULLVP; 1021 nfsstats.lookupcache_misses++; 1022 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 1023 len = cnp->cn_namelen; 1024 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP, 1025 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 1026 mb = mreq; 1027 bpos = mtod(mb, caddr_t); 1028 nfsm_fhtom(dvp, v3); 1029 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 1030 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred); 1031 if (error) { 1032 if (v3) { 1033 nfsm_postop_attr_va(dvp, dattrflag, &vattr); 1034 m_freem(mrep); 1035 } 1036 goto nfsmout; 1037 } 1038 nfsm_getfh(fhp, fhsize, v3); 1039 1040 /* 1041 * Handle RENAME case... 1042 */ 1043 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 1044 if (NFS_CMPFH(np, fhp, fhsize)) { 1045 m_freem(mrep); 1046 return (EISDIR); 1047 } 1048 error = nfs_nget(mp, fhp, fhsize, &np, LK_EXCLUSIVE); 1049 if (error) { 1050 m_freem(mrep); 1051 return (error); 1052 } 1053 newvp = NFSTOV(np); 1054 if (v3) { 1055 nfsm_postop_attr(newvp, attrflag); 1056 nfsm_postop_attr(dvp, attrflag); 1057 } else 1058 nfsm_loadattr(newvp, NULL); 1059 *vpp = newvp; 1060 m_freem(mrep); 1061 cnp->cn_flags |= SAVENAME; 1062 return (0); 1063 } 1064 1065 if (flags & ISDOTDOT) { 1066 ltype = VOP_ISLOCKED(dvp); 1067 error = vfs_busy(mp, MBF_NOWAIT); 1068 if (error != 0) { 1069 vfs_ref(mp); 1070 VOP_UNLOCK(dvp, 0); 1071 error = vfs_busy(mp, 0); 1072 vn_lock(dvp, ltype | LK_RETRY); 1073 vfs_rel(mp); 1074 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) { 1075 vfs_unbusy(mp); 1076 error = ENOENT; 1077 } 1078 if (error != 0) { 1079 m_freem(mrep); 1080 return (error); 1081 } 1082 } 1083 VOP_UNLOCK(dvp, 0); 1084 error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags); 1085 if (error == 0) 1086 newvp = NFSTOV(np); 1087 vfs_unbusy(mp); 1088 if (newvp != dvp) 1089 vn_lock(dvp, ltype | LK_RETRY); 1090 if (dvp->v_iflag & VI_DOOMED) { 1091 if (error == 0) { 1092 if (newvp == dvp) 1093 vrele(newvp); 1094 else 1095 vput(newvp); 1096 } 1097 error = ENOENT; 1098 } 1099 if (error) { 1100 m_freem(mrep); 1101 return (error); 1102 } 1103 } else if (NFS_CMPFH(np, fhp, fhsize)) { 1104 VREF(dvp); 1105 newvp = dvp; 1106 } else { 1107 error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags); 1108 if (error) { 1109 m_freem(mrep); 1110 return (error); 1111 } 1112 newvp = NFSTOV(np); 1113 1114 /* 1115 * Flush the attribute cache when opening a leaf node 1116 * to ensure that fresh attributes are fetched in 1117 * nfs_open() if we are unable to fetch attributes 1118 * from the LOOKUP reply. 1119 */ 1120 if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1121 !(np->n_flag & NMODIFIED)) { 1122 mtx_lock(&np->n_mtx); 1123 np->n_attrstamp = 0; 1124 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1125 mtx_unlock(&np->n_mtx); 1126 } 1127 } 1128 if (v3) { 1129 nfsm_postop_attr_va(newvp, attrflag, &vattr); 1130 nfsm_postop_attr_va(dvp, dattrflag, &dvattr); 1131 } else { 1132 nfsm_loadattr(newvp, &vattr); 1133 attrflag = 1; 1134 } 1135 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1136 cnp->cn_flags |= SAVENAME; 1137 if ((cnp->cn_flags & MAKEENTRY) && 1138 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) && 1139 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0)) 1140 cache_enter_time(dvp, newvp, cnp, &vattr.va_ctime, 1141 newvp->v_type != VDIR ? NULL : &dvattr.va_ctime); 1142 *vpp = newvp; 1143 m_freem(mrep); 1144nfsmout: 1145 if (error) { 1146 if (newvp != NULLVP) { 1147 vput(newvp); 1148 *vpp = NULLVP; 1149 } 1150 1151 if (error != ENOENT) 1152 goto done; 1153 1154 /* The requested file was not found. */ 1155 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1156 (flags & ISLASTCN)) { 1157 /* 1158 * XXX: UFS does a full VOP_ACCESS(dvp, 1159 * VWRITE) here instead of just checking 1160 * MNT_RDONLY. 1161 */ 1162 if (mp->mnt_flag & MNT_RDONLY) 1163 return (EROFS); 1164 cnp->cn_flags |= SAVENAME; 1165 return (EJUSTRETURN); 1166 } 1167 1168 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE && 1169 dattrflag) { 1170 /* 1171 * Cache the modification time of the parent 1172 * directory from the post-op attributes in 1173 * the name cache entry. The negative cache 1174 * entry will be ignored once the directory 1175 * has changed. Don't bother adding the entry 1176 * if the directory has already changed. 1177 */ 1178 mtx_lock(&np->n_mtx); 1179 if (timespeccmp(&np->n_vattr.va_mtime, 1180 &vattr.va_mtime, ==)) { 1181 mtx_unlock(&np->n_mtx); 1182 cache_enter_time(dvp, NULL, cnp, 1183 &vattr.va_mtime, NULL); 1184 } else 1185 mtx_unlock(&np->n_mtx); 1186 } 1187 return (ENOENT); 1188 } 1189done: 1190 return (error); 1191} 1192 1193/* 1194 * nfs read call. 1195 * Just call nfs_bioread() to do the work. 1196 */ 1197static int 1198nfs_read(struct vop_read_args *ap) 1199{ 1200 struct vnode *vp = ap->a_vp; 1201 1202 switch (vp->v_type) { 1203 case VREG: 1204 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred)); 1205 case VDIR: 1206 return (EISDIR); 1207 default: 1208 return (EOPNOTSUPP); 1209 } 1210} 1211 1212/* 1213 * nfs readlink call 1214 */ 1215static int 1216nfs_readlink(struct vop_readlink_args *ap) 1217{ 1218 struct vnode *vp = ap->a_vp; 1219 1220 if (vp->v_type != VLNK) 1221 return (EINVAL); 1222 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred)); 1223} 1224 1225/* 1226 * Do a readlink rpc. 1227 * Called by nfs_doio() from below the buffer cache. 1228 */ 1229int 1230nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1231{ 1232 caddr_t bpos, dpos; 1233 int error = 0, len, attrflag; 1234 struct mbuf *mreq, *mrep, *md, *mb; 1235 int v3 = NFS_ISV3(vp); 1236 1237 nfsstats.rpccnt[NFSPROC_READLINK]++; 1238 mreq = nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3)); 1239 mb = mreq; 1240 bpos = mtod(mb, caddr_t); 1241 nfsm_fhtom(vp, v3); 1242 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred); 1243 if (v3) 1244 nfsm_postop_attr(vp, attrflag); 1245 if (!error) { 1246 nfsm_strsiz(len, NFS_MAXPATHLEN); 1247 if (len == NFS_MAXPATHLEN) { 1248 struct nfsnode *np = VTONFS(vp); 1249 mtx_lock(&np->n_mtx); 1250 if (np->n_size && np->n_size < NFS_MAXPATHLEN) 1251 len = np->n_size; 1252 mtx_unlock(&np->n_mtx); 1253 } 1254 nfsm_mtouio(uiop, len); 1255 } 1256 m_freem(mrep); 1257nfsmout: 1258 return (error); 1259} 1260 1261/* 1262 * nfs read rpc call 1263 * Ditto above 1264 */ 1265int 1266nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1267{ 1268 u_int32_t *tl; 1269 caddr_t bpos, dpos; 1270 struct mbuf *mreq, *mrep, *md, *mb; 1271 struct nfsmount *nmp; 1272 off_t end; 1273 int error = 0, len, retlen, tsiz, eof, attrflag; 1274 int v3 = NFS_ISV3(vp); 1275 int rsize; 1276 1277#ifndef nolint 1278 eof = 0; 1279#endif 1280 nmp = VFSTONFS(vp->v_mount); 1281 tsiz = uiop->uio_resid; 1282 mtx_lock(&nmp->nm_mtx); 1283 end = uiop->uio_offset + tsiz; 1284 if (end > nmp->nm_maxfilesize || end < uiop->uio_offset) { 1285 mtx_unlock(&nmp->nm_mtx); 1286 return (EFBIG); 1287 } 1288 rsize = nmp->nm_rsize; 1289 mtx_unlock(&nmp->nm_mtx); 1290 while (tsiz > 0) { 1291 nfsstats.rpccnt[NFSPROC_READ]++; 1292 len = (tsiz > rsize) ? rsize : tsiz; 1293 mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1294 mb = mreq; 1295 bpos = mtod(mb, caddr_t); 1296 nfsm_fhtom(vp, v3); 1297 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3); 1298 if (v3) { 1299 txdr_hyper(uiop->uio_offset, tl); 1300 *(tl + 2) = txdr_unsigned(len); 1301 } else { 1302 *tl++ = txdr_unsigned(uiop->uio_offset); 1303 *tl++ = txdr_unsigned(len); 1304 *tl = 0; 1305 } 1306 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred); 1307 if (v3) { 1308 nfsm_postop_attr(vp, attrflag); 1309 if (error) { 1310 m_freem(mrep); 1311 goto nfsmout; 1312 } 1313 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED); 1314 eof = fxdr_unsigned(int, *(tl + 1)); 1315 } else { 1316 nfsm_loadattr(vp, NULL); 1317 } 1318 nfsm_strsiz(retlen, rsize); 1319 nfsm_mtouio(uiop, retlen); 1320 m_freem(mrep); 1321 tsiz -= retlen; 1322 if (v3) { 1323 if (eof || retlen == 0) { 1324 tsiz = 0; 1325 } 1326 } else if (retlen < len) { 1327 tsiz = 0; 1328 } 1329 } 1330nfsmout: 1331 return (error); 1332} 1333 1334/* 1335 * nfs write call 1336 */ 1337int 1338nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 1339 int *iomode, int *must_commit) 1340{ 1341 u_int32_t *tl; 1342 int32_t backup; 1343 caddr_t bpos, dpos; 1344 struct mbuf *mreq, *mrep, *md, *mb; 1345 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1346 off_t end; 1347 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit; 1348 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC; 1349 int wsize; 1350 1351 KASSERT(uiop->uio_iovcnt == 1, ("nfs: writerpc iovcnt > 1")); 1352 *must_commit = 0; 1353 tsiz = uiop->uio_resid; 1354 mtx_lock(&nmp->nm_mtx); 1355 end = uiop->uio_offset + tsiz; 1356 if (end > nmp->nm_maxfilesize || end < uiop->uio_offset) { 1357 mtx_unlock(&nmp->nm_mtx); 1358 return (EFBIG); 1359 } 1360 wsize = nmp->nm_wsize; 1361 mtx_unlock(&nmp->nm_mtx); 1362 while (tsiz > 0) { 1363 nfsstats.rpccnt[NFSPROC_WRITE]++; 1364 len = (tsiz > wsize) ? wsize : tsiz; 1365 mreq = nfsm_reqhead(vp, NFSPROC_WRITE, 1366 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len)); 1367 mb = mreq; 1368 bpos = mtod(mb, caddr_t); 1369 nfsm_fhtom(vp, v3); 1370 if (v3) { 1371 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED); 1372 txdr_hyper(uiop->uio_offset, tl); 1373 tl += 2; 1374 *tl++ = txdr_unsigned(len); 1375 *tl++ = txdr_unsigned(*iomode); 1376 *tl = txdr_unsigned(len); 1377 } else { 1378 u_int32_t x; 1379 1380 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED); 1381 /* Set both "begin" and "current" to non-garbage. */ 1382 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1383 *tl++ = x; /* "begin offset" */ 1384 *tl++ = x; /* "current offset" */ 1385 x = txdr_unsigned(len); 1386 *tl++ = x; /* total to this offset */ 1387 *tl = x; /* size of this write */ 1388 } 1389 nfsm_uiotom(uiop, len); 1390 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred); 1391 if (v3) { 1392 wccflag = NFSV3_WCCCHK; 1393 nfsm_wcc_data(vp, wccflag); 1394 if (!error) { 1395 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED 1396 + NFSX_V3WRITEVERF); 1397 rlen = fxdr_unsigned(int, *tl++); 1398 if (rlen == 0) { 1399 error = NFSERR_IO; 1400 m_freem(mrep); 1401 break; 1402 } else if (rlen < len) { 1403 backup = len - rlen; 1404 uiop->uio_iov->iov_base = 1405 (char *)uiop->uio_iov->iov_base - 1406 backup; 1407 uiop->uio_iov->iov_len += backup; 1408 uiop->uio_offset -= backup; 1409 uiop->uio_resid += backup; 1410 len = rlen; 1411 } 1412 commit = fxdr_unsigned(int, *tl++); 1413 1414 /* 1415 * Return the lowest committment level 1416 * obtained by any of the RPCs. 1417 */ 1418 if (committed == NFSV3WRITE_FILESYNC) 1419 committed = commit; 1420 else if (committed == NFSV3WRITE_DATASYNC && 1421 commit == NFSV3WRITE_UNSTABLE) 1422 committed = commit; 1423 mtx_lock(&nmp->nm_mtx); 1424 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){ 1425 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1426 NFSX_V3WRITEVERF); 1427 nmp->nm_state |= NFSSTA_HASWRITEVERF; 1428 } else if (bcmp((caddr_t)tl, 1429 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) { 1430 *must_commit = 1; 1431 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1432 NFSX_V3WRITEVERF); 1433 } 1434 mtx_unlock(&nmp->nm_mtx); 1435 } 1436 } else { 1437 nfsm_loadattr(vp, NULL); 1438 } 1439 if (wccflag) { 1440 mtx_lock(&(VTONFS(vp))->n_mtx); 1441 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime; 1442 mtx_unlock(&(VTONFS(vp))->n_mtx); 1443 } 1444 m_freem(mrep); 1445 if (error) 1446 break; 1447 tsiz -= len; 1448 } 1449nfsmout: 1450 if (DOINGASYNC(vp)) 1451 committed = NFSV3WRITE_FILESYNC; 1452 *iomode = committed; 1453 if (error) 1454 uiop->uio_resid = tsiz; 1455 return (error); 1456} 1457 1458/* 1459 * nfs mknod rpc 1460 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1461 * mode set to specify the file type and the size field for rdev. 1462 */ 1463static int 1464nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1465 struct vattr *vap) 1466{ 1467 struct nfsv2_sattr *sp; 1468 u_int32_t *tl; 1469 struct vnode *newvp = NULL; 1470 struct nfsnode *np = NULL; 1471 struct vattr vattr; 1472 caddr_t bpos, dpos; 1473 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1474 struct mbuf *mreq, *mrep, *md, *mb; 1475 u_int32_t rdev; 1476 int v3 = NFS_ISV3(dvp); 1477 1478 if (vap->va_type == VCHR || vap->va_type == VBLK) 1479 rdev = txdr_unsigned(vap->va_rdev); 1480 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1481 rdev = nfs_xdrneg1; 1482 else { 1483 return (EOPNOTSUPP); 1484 } 1485 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) 1486 return (error); 1487 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1488 mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1489 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1490 mb = mreq; 1491 bpos = mtod(mb, caddr_t); 1492 nfsm_fhtom(dvp, v3); 1493 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1494 if (v3) { 1495 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED); 1496 *tl++ = vtonfsv3_type(vap->va_type); 1497 nfsm_v3attrbuild(vap, FALSE); 1498 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1499 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED); 1500 *tl++ = txdr_unsigned(major(vap->va_rdev)); 1501 *tl = txdr_unsigned(minor(vap->va_rdev)); 1502 } 1503 } else { 1504 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR); 1505 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1506 sp->sa_uid = nfs_xdrneg1; 1507 sp->sa_gid = nfs_xdrneg1; 1508 sp->sa_size = rdev; 1509 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1510 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1511 } 1512 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred); 1513 if (!error) { 1514 nfsm_mtofh(dvp, newvp, v3, gotvp); 1515 if (!gotvp) { 1516 if (newvp) { 1517 vput(newvp); 1518 newvp = NULL; 1519 } 1520 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1521 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np); 1522 if (!error) 1523 newvp = NFSTOV(np); 1524 } 1525 } 1526 if (v3) 1527 nfsm_wcc_data(dvp, wccflag); 1528 m_freem(mrep); 1529nfsmout: 1530 if (error) { 1531 if (newvp) 1532 vput(newvp); 1533 } else { 1534 *vpp = newvp; 1535 } 1536 mtx_lock(&(VTONFS(dvp))->n_mtx); 1537 VTONFS(dvp)->n_flag |= NMODIFIED; 1538 if (!wccflag) { 1539 VTONFS(dvp)->n_attrstamp = 0; 1540 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1541 } 1542 mtx_unlock(&(VTONFS(dvp))->n_mtx); 1543 return (error); 1544} 1545 1546/* 1547 * nfs mknod vop 1548 * just call nfs_mknodrpc() to do the work. 1549 */ 1550/* ARGSUSED */ 1551static int 1552nfs_mknod(struct vop_mknod_args *ap) 1553{ 1554 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap)); 1555} 1556 1557static u_long create_verf; 1558/* 1559 * nfs file create call 1560 */ 1561static int 1562nfs_create(struct vop_create_args *ap) 1563{ 1564 struct vnode *dvp = ap->a_dvp; 1565 struct vattr *vap = ap->a_vap; 1566 struct componentname *cnp = ap->a_cnp; 1567 struct nfsv2_sattr *sp; 1568 u_int32_t *tl; 1569 struct nfsnode *np = NULL; 1570 struct vnode *newvp = NULL; 1571 caddr_t bpos, dpos; 1572 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0; 1573 struct mbuf *mreq, *mrep, *md, *mb; 1574 struct vattr vattr; 1575 int v3 = NFS_ISV3(dvp); 1576 1577 /* 1578 * Oops, not for me.. 1579 */ 1580 if (vap->va_type == VSOCK) { 1581 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap); 1582 return (error); 1583 } 1584 1585 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) { 1586 return (error); 1587 } 1588 if (vap->va_vaflags & VA_EXCLUSIVE) 1589 fmode |= O_EXCL; 1590again: 1591 nfsstats.rpccnt[NFSPROC_CREATE]++; 1592 mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1593 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1594 mb = mreq; 1595 bpos = mtod(mb, caddr_t); 1596 nfsm_fhtom(dvp, v3); 1597 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1598 if (v3) { 1599 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED); 1600 if (fmode & O_EXCL) { 1601 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1602 tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF); 1603#ifdef INET 1604 CURVNET_SET(CRED_TO_VNET(cnp->cn_cred)); 1605 IN_IFADDR_RLOCK(); 1606 if (!TAILQ_EMPTY(&V_in_ifaddrhead)) 1607 *tl++ = IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr.s_addr; 1608 else 1609#endif 1610 *tl++ = create_verf; 1611#ifdef INET 1612 IN_IFADDR_RUNLOCK(); 1613 CURVNET_RESTORE(); 1614#endif 1615 *tl = ++create_verf; 1616 } else { 1617 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED); 1618 nfsm_v3attrbuild(vap, FALSE); 1619 } 1620 } else { 1621 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR); 1622 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1623 sp->sa_uid = nfs_xdrneg1; 1624 sp->sa_gid = nfs_xdrneg1; 1625 sp->sa_size = 0; 1626 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1627 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1628 } 1629 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred); 1630 if (!error) { 1631 nfsm_mtofh(dvp, newvp, v3, gotvp); 1632 if (!gotvp) { 1633 if (newvp) { 1634 vput(newvp); 1635 newvp = NULL; 1636 } 1637 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1638 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np); 1639 if (!error) 1640 newvp = NFSTOV(np); 1641 } 1642 } 1643 if (v3) 1644 nfsm_wcc_data(dvp, wccflag); 1645 m_freem(mrep); 1646nfsmout: 1647 if (error) { 1648 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) { 1649 fmode &= ~O_EXCL; 1650 goto again; 1651 } 1652 if (newvp) 1653 vput(newvp); 1654 } else if (v3 && (fmode & O_EXCL)) { 1655 /* 1656 * We are normally called with only a partially initialized 1657 * VAP. Since the NFSv3 spec says that server may use the 1658 * file attributes to store the verifier, the spec requires 1659 * us to do a SETATTR RPC. FreeBSD servers store the verifier 1660 * in atime, but we can't really assume that all servers will 1661 * so we ensure that our SETATTR sets both atime and mtime. 1662 */ 1663 if (vap->va_mtime.tv_sec == VNOVAL) 1664 vfs_timestamp(&vap->va_mtime); 1665 if (vap->va_atime.tv_sec == VNOVAL) 1666 vap->va_atime = vap->va_mtime; 1667 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred); 1668 if (error) 1669 vput(newvp); 1670 } 1671 if (!error) { 1672 *ap->a_vpp = newvp; 1673 } 1674 mtx_lock(&(VTONFS(dvp))->n_mtx); 1675 VTONFS(dvp)->n_flag |= NMODIFIED; 1676 if (!wccflag) { 1677 VTONFS(dvp)->n_attrstamp = 0; 1678 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1679 } 1680 mtx_unlock(&(VTONFS(dvp))->n_mtx); 1681 return (error); 1682} 1683 1684/* 1685 * nfs file remove call 1686 * To try and make nfs semantics closer to ufs semantics, a file that has 1687 * other processes using the vnode is renamed instead of removed and then 1688 * removed later on the last close. 1689 * - If v_usecount > 1 1690 * If a rename is not already in the works 1691 * call nfs_sillyrename() to set it up 1692 * else 1693 * do the remove rpc 1694 */ 1695static int 1696nfs_remove(struct vop_remove_args *ap) 1697{ 1698 struct vnode *vp = ap->a_vp; 1699 struct vnode *dvp = ap->a_dvp; 1700 struct componentname *cnp = ap->a_cnp; 1701 struct nfsnode *np = VTONFS(vp); 1702 int error = 0; 1703 struct vattr vattr; 1704 1705 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name")); 1706 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount")); 1707 if (vp->v_type == VDIR) 1708 error = EPERM; 1709 else if (vrefcnt(vp) == 1 || (np->n_sillyrename && 1710 !VOP_GETATTR(vp, &vattr, cnp->cn_cred) && vattr.va_nlink > 1)) { 1711 /* 1712 * Purge the name cache so that the chance of a lookup for 1713 * the name succeeding while the remove is in progress is 1714 * minimized. Without node locking it can still happen, such 1715 * that an I/O op returns ESTALE, but since you get this if 1716 * another host removes the file.. 1717 */ 1718 cache_purge(vp); 1719 /* 1720 * throw away biocache buffers, mainly to avoid 1721 * unnecessary delayed writes later. 1722 */ 1723 error = nfs_vinvalbuf(vp, 0, cnp->cn_thread, 1); 1724 /* Do the rpc */ 1725 if (error != EINTR && error != EIO) 1726 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1727 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread); 1728 /* 1729 * Kludge City: If the first reply to the remove rpc is lost.. 1730 * the reply to the retransmitted request will be ENOENT 1731 * since the file was in fact removed 1732 * Therefore, we cheat and return success. 1733 */ 1734 if (error == ENOENT) 1735 error = 0; 1736 } else if (!np->n_sillyrename) 1737 error = nfs_sillyrename(dvp, vp, cnp); 1738 mtx_lock(&np->n_mtx); 1739 np->n_attrstamp = 0; 1740 mtx_unlock(&np->n_mtx); 1741 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 1742 return (error); 1743} 1744 1745/* 1746 * nfs file remove rpc called from nfs_inactive 1747 */ 1748int 1749nfs_removeit(struct sillyrename *sp) 1750{ 1751 /* 1752 * Make sure that the directory vnode is still valid. 1753 * XXX we should lock sp->s_dvp here. 1754 */ 1755 if (sp->s_dvp->v_type == VBAD) 1756 return (0); 1757 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred, 1758 NULL)); 1759} 1760 1761/* 1762 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1763 */ 1764static int 1765nfs_removerpc(struct vnode *dvp, const char *name, int namelen, 1766 struct ucred *cred, struct thread *td) 1767{ 1768 caddr_t bpos, dpos; 1769 int error = 0, wccflag = NFSV3_WCCRATTR; 1770 struct mbuf *mreq, *mrep, *md, *mb; 1771 int v3 = NFS_ISV3(dvp); 1772 1773 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1774 mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE, 1775 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1776 mb = mreq; 1777 bpos = mtod(mb, caddr_t); 1778 nfsm_fhtom(dvp, v3); 1779 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1780 nfsm_request(dvp, NFSPROC_REMOVE, td, cred); 1781 if (v3) 1782 nfsm_wcc_data(dvp, wccflag); 1783 m_freem(mrep); 1784nfsmout: 1785 mtx_lock(&(VTONFS(dvp))->n_mtx); 1786 VTONFS(dvp)->n_flag |= NMODIFIED; 1787 if (!wccflag) { 1788 VTONFS(dvp)->n_attrstamp = 0; 1789 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1790 } 1791 mtx_unlock(&(VTONFS(dvp))->n_mtx); 1792 return (error); 1793} 1794 1795/* 1796 * nfs file rename call 1797 */ 1798static int 1799nfs_rename(struct vop_rename_args *ap) 1800{ 1801 struct vnode *fvp = ap->a_fvp; 1802 struct vnode *tvp = ap->a_tvp; 1803 struct vnode *fdvp = ap->a_fdvp; 1804 struct vnode *tdvp = ap->a_tdvp; 1805 struct componentname *tcnp = ap->a_tcnp; 1806 struct componentname *fcnp = ap->a_fcnp; 1807 int error; 1808 1809 KASSERT((tcnp->cn_flags & HASBUF) != 0 && 1810 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name")); 1811 /* Check for cross-device rename */ 1812 if ((fvp->v_mount != tdvp->v_mount) || 1813 (tvp && (fvp->v_mount != tvp->v_mount))) { 1814 error = EXDEV; 1815 goto out; 1816 } 1817 1818 if (fvp == tvp) { 1819 nfs_printf("nfs_rename: fvp == tvp (can't happen)\n"); 1820 error = 0; 1821 goto out; 1822 } 1823 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0) 1824 goto out; 1825 1826 /* 1827 * We have to flush B_DELWRI data prior to renaming 1828 * the file. If we don't, the delayed-write buffers 1829 * can be flushed out later after the file has gone stale 1830 * under NFSV3. NFSV2 does not have this problem because 1831 * ( as far as I can tell ) it flushes dirty buffers more 1832 * often. 1833 * 1834 * Skip the rename operation if the fsync fails, this can happen 1835 * due to the server's volume being full, when we pushed out data 1836 * that was written back to our cache earlier. Not checking for 1837 * this condition can result in potential (silent) data loss. 1838 */ 1839 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread); 1840 VOP_UNLOCK(fvp, 0); 1841 if (!error && tvp) 1842 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread); 1843 if (error) 1844 goto out; 1845 1846 /* 1847 * If the tvp exists and is in use, sillyrename it before doing the 1848 * rename of the new file over it. 1849 * XXX Can't sillyrename a directory. 1850 */ 1851 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename && 1852 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1853 vput(tvp); 1854 tvp = NULL; 1855 } 1856 1857 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1858 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1859 tcnp->cn_thread); 1860 1861 if (fvp->v_type == VDIR) { 1862 if (tvp != NULL && tvp->v_type == VDIR) 1863 cache_purge(tdvp); 1864 cache_purge(fdvp); 1865 } 1866 1867out: 1868 if (tdvp == tvp) 1869 vrele(tdvp); 1870 else 1871 vput(tdvp); 1872 if (tvp) 1873 vput(tvp); 1874 vrele(fdvp); 1875 vrele(fvp); 1876 /* 1877 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1878 */ 1879 if (error == ENOENT) 1880 error = 0; 1881 return (error); 1882} 1883 1884/* 1885 * nfs file rename rpc called from nfs_remove() above 1886 */ 1887static int 1888nfs_renameit(struct vnode *sdvp, struct componentname *scnp, 1889 struct sillyrename *sp) 1890{ 1891 1892 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp, 1893 sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread)); 1894} 1895 1896/* 1897 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1898 */ 1899static int 1900nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen, 1901 struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred, 1902 struct thread *td) 1903{ 1904 caddr_t bpos, dpos; 1905 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1906 struct mbuf *mreq, *mrep, *md, *mb; 1907 int v3 = NFS_ISV3(fdvp); 1908 1909 nfsstats.rpccnt[NFSPROC_RENAME]++; 1910 mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME, 1911 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 1912 nfsm_rndup(tnamelen)); 1913 mb = mreq; 1914 bpos = mtod(mb, caddr_t); 1915 nfsm_fhtom(fdvp, v3); 1916 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 1917 nfsm_fhtom(tdvp, v3); 1918 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 1919 nfsm_request(fdvp, NFSPROC_RENAME, td, cred); 1920 if (v3) { 1921 nfsm_wcc_data(fdvp, fwccflag); 1922 nfsm_wcc_data(tdvp, twccflag); 1923 } 1924 m_freem(mrep); 1925nfsmout: 1926 mtx_lock(&(VTONFS(fdvp))->n_mtx); 1927 VTONFS(fdvp)->n_flag |= NMODIFIED; 1928 mtx_unlock(&(VTONFS(fdvp))->n_mtx); 1929 mtx_lock(&(VTONFS(tdvp))->n_mtx); 1930 VTONFS(tdvp)->n_flag |= NMODIFIED; 1931 mtx_unlock(&(VTONFS(tdvp))->n_mtx); 1932 if (!fwccflag) { 1933 VTONFS(fdvp)->n_attrstamp = 0; 1934 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp); 1935 } 1936 if (!twccflag) { 1937 VTONFS(tdvp)->n_attrstamp = 0; 1938 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 1939 } 1940 return (error); 1941} 1942 1943/* 1944 * nfs hard link create call 1945 */ 1946static int 1947nfs_link(struct vop_link_args *ap) 1948{ 1949 struct vnode *vp = ap->a_vp; 1950 struct vnode *tdvp = ap->a_tdvp; 1951 struct componentname *cnp = ap->a_cnp; 1952 caddr_t bpos, dpos; 1953 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 1954 struct mbuf *mreq, *mrep, *md, *mb; 1955 int v3; 1956 1957 if (vp->v_mount != tdvp->v_mount) { 1958 return (EXDEV); 1959 } 1960 1961 /* 1962 * Push all writes to the server, so that the attribute cache 1963 * doesn't get "out of sync" with the server. 1964 * XXX There should be a better way! 1965 */ 1966 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread); 1967 1968 v3 = NFS_ISV3(vp); 1969 nfsstats.rpccnt[NFSPROC_LINK]++; 1970 mreq = nfsm_reqhead(vp, NFSPROC_LINK, 1971 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 1972 mb = mreq; 1973 bpos = mtod(mb, caddr_t); 1974 nfsm_fhtom(vp, v3); 1975 nfsm_fhtom(tdvp, v3); 1976 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1977 nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred); 1978 if (v3) { 1979 nfsm_postop_attr(vp, attrflag); 1980 nfsm_wcc_data(tdvp, wccflag); 1981 } 1982 m_freem(mrep); 1983nfsmout: 1984 mtx_lock(&(VTONFS(tdvp))->n_mtx); 1985 VTONFS(tdvp)->n_flag |= NMODIFIED; 1986 mtx_unlock(&(VTONFS(tdvp))->n_mtx); 1987 if (!attrflag) { 1988 VTONFS(vp)->n_attrstamp = 0; 1989 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 1990 } 1991 if (!wccflag) { 1992 VTONFS(tdvp)->n_attrstamp = 0; 1993 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 1994 } 1995 return (error); 1996} 1997 1998/* 1999 * nfs symbolic link create call 2000 */ 2001static int 2002nfs_symlink(struct vop_symlink_args *ap) 2003{ 2004 struct vnode *dvp = ap->a_dvp; 2005 struct vattr *vap = ap->a_vap; 2006 struct componentname *cnp = ap->a_cnp; 2007 struct nfsv2_sattr *sp; 2008 caddr_t bpos, dpos; 2009 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 2010 struct mbuf *mreq, *mrep, *md, *mb; 2011 struct vnode *newvp = NULL; 2012 int v3 = NFS_ISV3(dvp); 2013 2014 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 2015 slen = strlen(ap->a_target); 2016 mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 2017 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 2018 mb = mreq; 2019 bpos = mtod(mb, caddr_t); 2020 nfsm_fhtom(dvp, v3); 2021 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2022 if (v3) { 2023 nfsm_v3attrbuild(vap, FALSE); 2024 } 2025 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 2026 if (!v3) { 2027 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR); 2028 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 2029 sp->sa_uid = nfs_xdrneg1; 2030 sp->sa_gid = nfs_xdrneg1; 2031 sp->sa_size = nfs_xdrneg1; 2032 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2033 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2034 } 2035 2036 /* 2037 * Issue the NFS request and get the rpc response. 2038 * 2039 * Only NFSv3 responses returning an error of 0 actually return 2040 * a file handle that can be converted into newvp without having 2041 * to do an extra lookup rpc. 2042 */ 2043 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred); 2044 if (v3) { 2045 if (error == 0) 2046 nfsm_mtofh(dvp, newvp, v3, gotvp); 2047 nfsm_wcc_data(dvp, wccflag); 2048 } 2049 2050 /* 2051 * out code jumps -> here, mrep is also freed. 2052 */ 2053 2054 m_freem(mrep); 2055nfsmout: 2056 2057 /* 2058 * If we do not have an error and we could not extract the newvp from 2059 * the response due to the request being NFSv2, we have to do a 2060 * lookup in order to obtain a newvp to return. 2061 */ 2062 if (error == 0 && newvp == NULL) { 2063 struct nfsnode *np = NULL; 2064 2065 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2066 cnp->cn_cred, cnp->cn_thread, &np); 2067 if (!error) 2068 newvp = NFSTOV(np); 2069 } 2070 if (error) { 2071 if (newvp) 2072 vput(newvp); 2073 } else { 2074 *ap->a_vpp = newvp; 2075 } 2076 mtx_lock(&(VTONFS(dvp))->n_mtx); 2077 VTONFS(dvp)->n_flag |= NMODIFIED; 2078 mtx_unlock(&(VTONFS(dvp))->n_mtx); 2079 if (!wccflag) { 2080 VTONFS(dvp)->n_attrstamp = 0; 2081 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2082 } 2083 return (error); 2084} 2085 2086/* 2087 * nfs make dir call 2088 */ 2089static int 2090nfs_mkdir(struct vop_mkdir_args *ap) 2091{ 2092 struct vnode *dvp = ap->a_dvp; 2093 struct vattr *vap = ap->a_vap; 2094 struct componentname *cnp = ap->a_cnp; 2095 struct nfsv2_sattr *sp; 2096 int len; 2097 struct nfsnode *np = NULL; 2098 struct vnode *newvp = NULL; 2099 caddr_t bpos, dpos; 2100 int error = 0, wccflag = NFSV3_WCCRATTR; 2101 int gotvp = 0; 2102 struct mbuf *mreq, *mrep, *md, *mb; 2103 struct vattr vattr; 2104 int v3 = NFS_ISV3(dvp); 2105 2106 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) 2107 return (error); 2108 len = cnp->cn_namelen; 2109 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2110 mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR, 2111 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 2112 mb = mreq; 2113 bpos = mtod(mb, caddr_t); 2114 nfsm_fhtom(dvp, v3); 2115 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 2116 if (v3) { 2117 nfsm_v3attrbuild(vap, FALSE); 2118 } else { 2119 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR); 2120 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2121 sp->sa_uid = nfs_xdrneg1; 2122 sp->sa_gid = nfs_xdrneg1; 2123 sp->sa_size = nfs_xdrneg1; 2124 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2125 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2126 } 2127 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred); 2128 if (!error) 2129 nfsm_mtofh(dvp, newvp, v3, gotvp); 2130 if (v3) 2131 nfsm_wcc_data(dvp, wccflag); 2132 m_freem(mrep); 2133nfsmout: 2134 mtx_lock(&(VTONFS(dvp))->n_mtx); 2135 VTONFS(dvp)->n_flag |= NMODIFIED; 2136 mtx_unlock(&(VTONFS(dvp))->n_mtx); 2137 if (!wccflag) { 2138 VTONFS(dvp)->n_attrstamp = 0; 2139 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2140 } 2141 if (error == 0 && newvp == NULL) { 2142 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2143 cnp->cn_thread, &np); 2144 if (!error) { 2145 newvp = NFSTOV(np); 2146 if (newvp->v_type != VDIR) 2147 error = EEXIST; 2148 } 2149 } 2150 if (error) { 2151 if (newvp) 2152 vput(newvp); 2153 } else 2154 *ap->a_vpp = newvp; 2155 return (error); 2156} 2157 2158/* 2159 * nfs remove directory call 2160 */ 2161static int 2162nfs_rmdir(struct vop_rmdir_args *ap) 2163{ 2164 struct vnode *vp = ap->a_vp; 2165 struct vnode *dvp = ap->a_dvp; 2166 struct componentname *cnp = ap->a_cnp; 2167 caddr_t bpos, dpos; 2168 int error = 0, wccflag = NFSV3_WCCRATTR; 2169 struct mbuf *mreq, *mrep, *md, *mb; 2170 int v3 = NFS_ISV3(dvp); 2171 2172 if (dvp == vp) 2173 return (EINVAL); 2174 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2175 mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR, 2176 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2177 mb = mreq; 2178 bpos = mtod(mb, caddr_t); 2179 nfsm_fhtom(dvp, v3); 2180 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2181 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred); 2182 if (v3) 2183 nfsm_wcc_data(dvp, wccflag); 2184 m_freem(mrep); 2185nfsmout: 2186 mtx_lock(&(VTONFS(dvp))->n_mtx); 2187 VTONFS(dvp)->n_flag |= NMODIFIED; 2188 mtx_unlock(&(VTONFS(dvp))->n_mtx); 2189 if (!wccflag) { 2190 VTONFS(dvp)->n_attrstamp = 0; 2191 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2192 } 2193 cache_purge(dvp); 2194 cache_purge(vp); 2195 /* 2196 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2197 */ 2198 if (error == ENOENT) 2199 error = 0; 2200 return (error); 2201} 2202 2203/* 2204 * nfs readdir call 2205 */ 2206static int 2207nfs_readdir(struct vop_readdir_args *ap) 2208{ 2209 struct vnode *vp = ap->a_vp; 2210 struct nfsnode *np = VTONFS(vp); 2211 struct uio *uio = ap->a_uio; 2212 int tresid, error = 0; 2213 struct vattr vattr; 2214 2215 if (vp->v_type != VDIR) 2216 return(EPERM); 2217 2218 /* 2219 * First, check for hit on the EOF offset cache 2220 */ 2221 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2222 (np->n_flag & NMODIFIED) == 0) { 2223 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) { 2224 mtx_lock(&np->n_mtx); 2225 if (!NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 2226 mtx_unlock(&np->n_mtx); 2227 nfsstats.direofcache_hits++; 2228 goto out; 2229 } else 2230 mtx_unlock(&np->n_mtx); 2231 } 2232 } 2233 2234 /* 2235 * Call nfs_bioread() to do the real work. 2236 */ 2237 tresid = uio->uio_resid; 2238 error = nfs_bioread(vp, uio, 0, ap->a_cred); 2239 2240 if (!error && uio->uio_resid == tresid) { 2241 nfsstats.direofcache_misses++; 2242 } 2243out: 2244 return (error); 2245} 2246 2247/* 2248 * Readdir rpc call. 2249 * Called from below the buffer cache by nfs_doio(). 2250 */ 2251int 2252nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 2253{ 2254 int len, left; 2255 struct dirent *dp = NULL; 2256 u_int32_t *tl; 2257 caddr_t cp; 2258 nfsuint64 *cookiep; 2259 caddr_t bpos, dpos; 2260 struct mbuf *mreq, *mrep, *md, *mb; 2261 nfsuint64 cookie; 2262 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2263 struct nfsnode *dnp = VTONFS(vp); 2264 u_quad_t fileno; 2265 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1; 2266 int attrflag; 2267 int v3 = NFS_ISV3(vp); 2268 2269 KASSERT(uiop->uio_iovcnt == 1 && 2270 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2271 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2272 ("nfs readdirrpc bad uio")); 2273 2274 /* 2275 * If there is no cookie, assume directory was stale. 2276 */ 2277 nfs_dircookie_lock(dnp); 2278 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2279 if (cookiep) { 2280 cookie = *cookiep; 2281 nfs_dircookie_unlock(dnp); 2282 } else { 2283 nfs_dircookie_unlock(dnp); 2284 return (NFSERR_BAD_COOKIE); 2285 } 2286 2287 /* 2288 * Loop around doing readdir rpc's of size nm_readdirsize 2289 * truncated to a multiple of DIRBLKSIZ. 2290 * The stopping criteria is EOF or buffer full. 2291 */ 2292 while (more_dirs && bigenough) { 2293 nfsstats.rpccnt[NFSPROC_READDIR]++; 2294 mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) + 2295 NFSX_READDIR(v3)); 2296 mb = mreq; 2297 bpos = mtod(mb, caddr_t); 2298 nfsm_fhtom(vp, v3); 2299 if (v3) { 2300 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED); 2301 *tl++ = cookie.nfsuquad[0]; 2302 *tl++ = cookie.nfsuquad[1]; 2303 mtx_lock(&dnp->n_mtx); 2304 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2305 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2306 mtx_unlock(&dnp->n_mtx); 2307 } else { 2308 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED); 2309 *tl++ = cookie.nfsuquad[0]; 2310 } 2311 *tl = txdr_unsigned(nmp->nm_readdirsize); 2312 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred); 2313 if (v3) { 2314 nfsm_postop_attr(vp, attrflag); 2315 if (!error) { 2316 tl = nfsm_dissect(u_int32_t *, 2317 2 * NFSX_UNSIGNED); 2318 mtx_lock(&dnp->n_mtx); 2319 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2320 dnp->n_cookieverf.nfsuquad[1] = *tl; 2321 mtx_unlock(&dnp->n_mtx); 2322 } else { 2323 m_freem(mrep); 2324 goto nfsmout; 2325 } 2326 } 2327 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); 2328 more_dirs = fxdr_unsigned(int, *tl); 2329 2330 /* loop thru the dir entries, doctoring them to 4bsd form */ 2331 while (more_dirs && bigenough) { 2332 if (v3) { 2333 tl = nfsm_dissect(u_int32_t *, 2334 3 * NFSX_UNSIGNED); 2335 fileno = fxdr_hyper(tl); 2336 len = fxdr_unsigned(int, *(tl + 2)); 2337 } else { 2338 tl = nfsm_dissect(u_int32_t *, 2339 2 * NFSX_UNSIGNED); 2340 fileno = fxdr_unsigned(u_quad_t, *tl++); 2341 len = fxdr_unsigned(int, *tl); 2342 } 2343 if (len <= 0 || len > NFS_MAXNAMLEN) { 2344 error = EBADRPC; 2345 m_freem(mrep); 2346 goto nfsmout; 2347 } 2348 tlen = nfsm_rndup(len); 2349 if (tlen == len) 2350 tlen += 4; /* To ensure null termination */ 2351 left = DIRBLKSIZ - blksiz; 2352 if ((tlen + DIRHDSIZ) > left) { 2353 dp->d_reclen += left; 2354 uiop->uio_iov->iov_base = 2355 (char *)uiop->uio_iov->iov_base + left; 2356 uiop->uio_iov->iov_len -= left; 2357 uiop->uio_offset += left; 2358 uiop->uio_resid -= left; 2359 blksiz = 0; 2360 } 2361 if ((tlen + DIRHDSIZ) > uiop->uio_resid) 2362 bigenough = 0; 2363 if (bigenough) { 2364 dp = (struct dirent *)uiop->uio_iov->iov_base; 2365 dp->d_fileno = (int)fileno; 2366 dp->d_namlen = len; 2367 dp->d_reclen = tlen + DIRHDSIZ; 2368 dp->d_type = DT_UNKNOWN; 2369 blksiz += dp->d_reclen; 2370 if (blksiz == DIRBLKSIZ) 2371 blksiz = 0; 2372 uiop->uio_offset += DIRHDSIZ; 2373 uiop->uio_resid -= DIRHDSIZ; 2374 uiop->uio_iov->iov_base = 2375 (char *)uiop->uio_iov->iov_base + DIRHDSIZ; 2376 uiop->uio_iov->iov_len -= DIRHDSIZ; 2377 nfsm_mtouio(uiop, len); 2378 cp = uiop->uio_iov->iov_base; 2379 tlen -= len; 2380 *cp = '\0'; /* null terminate */ 2381 uiop->uio_iov->iov_base = 2382 (char *)uiop->uio_iov->iov_base + tlen; 2383 uiop->uio_iov->iov_len -= tlen; 2384 uiop->uio_offset += tlen; 2385 uiop->uio_resid -= tlen; 2386 } else 2387 nfsm_adv(nfsm_rndup(len)); 2388 if (v3) { 2389 tl = nfsm_dissect(u_int32_t *, 2390 3 * NFSX_UNSIGNED); 2391 } else { 2392 tl = nfsm_dissect(u_int32_t *, 2393 2 * NFSX_UNSIGNED); 2394 } 2395 if (bigenough) { 2396 cookie.nfsuquad[0] = *tl++; 2397 if (v3) 2398 cookie.nfsuquad[1] = *tl++; 2399 } else if (v3) 2400 tl += 2; 2401 else 2402 tl++; 2403 more_dirs = fxdr_unsigned(int, *tl); 2404 } 2405 /* 2406 * If at end of rpc data, get the eof boolean 2407 */ 2408 if (!more_dirs) { 2409 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); 2410 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2411 } 2412 m_freem(mrep); 2413 } 2414 /* 2415 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2416 * by increasing d_reclen for the last record. 2417 */ 2418 if (blksiz > 0) { 2419 left = DIRBLKSIZ - blksiz; 2420 dp->d_reclen += left; 2421 uiop->uio_iov->iov_base = 2422 (char *)uiop->uio_iov->iov_base + left; 2423 uiop->uio_iov->iov_len -= left; 2424 uiop->uio_offset += left; 2425 uiop->uio_resid -= left; 2426 } 2427 2428 /* 2429 * We are now either at the end of the directory or have filled the 2430 * block. 2431 */ 2432 if (bigenough) 2433 dnp->n_direofoffset = uiop->uio_offset; 2434 else { 2435 if (uiop->uio_resid > 0) 2436 nfs_printf("EEK! readdirrpc resid > 0\n"); 2437 nfs_dircookie_lock(dnp); 2438 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2439 *cookiep = cookie; 2440 nfs_dircookie_unlock(dnp); 2441 } 2442nfsmout: 2443 return (error); 2444} 2445 2446/* 2447 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2448 */ 2449int 2450nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 2451{ 2452 int len, left; 2453 struct dirent *dp; 2454 u_int32_t *tl; 2455 caddr_t cp; 2456 struct vnode *newvp; 2457 nfsuint64 *cookiep; 2458 caddr_t bpos, dpos, dpossav1, dpossav2; 2459 struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2; 2460 struct nameidata nami, *ndp = &nami; 2461 struct componentname *cnp = &ndp->ni_cnd; 2462 nfsuint64 cookie; 2463 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2464 struct nfsnode *dnp = VTONFS(vp), *np; 2465 struct vattr vattr, dvattr; 2466 nfsfh_t *fhp; 2467 u_quad_t fileno; 2468 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2469 int attrflag, dattrflag, fhsize; 2470 2471#ifndef nolint 2472 dp = NULL; 2473#endif 2474 KASSERT(uiop->uio_iovcnt == 1 && 2475 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2476 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2477 ("nfs readdirplusrpc bad uio")); 2478 ndp->ni_dvp = vp; 2479 newvp = NULLVP; 2480 2481 /* 2482 * If there is no cookie, assume directory was stale. 2483 */ 2484 nfs_dircookie_lock(dnp); 2485 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2486 if (cookiep) { 2487 cookie = *cookiep; 2488 nfs_dircookie_unlock(dnp); 2489 } else { 2490 nfs_dircookie_unlock(dnp); 2491 return (NFSERR_BAD_COOKIE); 2492 } 2493 /* 2494 * Loop around doing readdir rpc's of size nm_readdirsize 2495 * truncated to a multiple of DIRBLKSIZ. 2496 * The stopping criteria is EOF or buffer full. 2497 */ 2498 while (more_dirs && bigenough) { 2499 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2500 mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS, 2501 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2502 mb = mreq; 2503 bpos = mtod(mb, caddr_t); 2504 nfsm_fhtom(vp, 1); 2505 tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED); 2506 *tl++ = cookie.nfsuquad[0]; 2507 *tl++ = cookie.nfsuquad[1]; 2508 mtx_lock(&dnp->n_mtx); 2509 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2510 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2511 mtx_unlock(&dnp->n_mtx); 2512 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2513 *tl = txdr_unsigned(nmp->nm_rsize); 2514 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred); 2515 nfsm_postop_attr_va(vp, dattrflag, &dvattr); 2516 if (error) { 2517 m_freem(mrep); 2518 goto nfsmout; 2519 } 2520 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED); 2521 mtx_lock(&dnp->n_mtx); 2522 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2523 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2524 mtx_unlock(&dnp->n_mtx); 2525 more_dirs = fxdr_unsigned(int, *tl); 2526 2527 /* loop thru the dir entries, doctoring them to 4bsd form */ 2528 while (more_dirs && bigenough) { 2529 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED); 2530 fileno = fxdr_hyper(tl); 2531 len = fxdr_unsigned(int, *(tl + 2)); 2532 if (len <= 0 || len > NFS_MAXNAMLEN) { 2533 error = EBADRPC; 2534 m_freem(mrep); 2535 goto nfsmout; 2536 } 2537 tlen = nfsm_rndup(len); 2538 if (tlen == len) 2539 tlen += 4; /* To ensure null termination*/ 2540 left = DIRBLKSIZ - blksiz; 2541 if ((tlen + DIRHDSIZ) > left) { 2542 dp->d_reclen += left; 2543 uiop->uio_iov->iov_base = 2544 (char *)uiop->uio_iov->iov_base + left; 2545 uiop->uio_iov->iov_len -= left; 2546 uiop->uio_offset += left; 2547 uiop->uio_resid -= left; 2548 blksiz = 0; 2549 } 2550 if ((tlen + DIRHDSIZ) > uiop->uio_resid) 2551 bigenough = 0; 2552 if (bigenough) { 2553 dp = (struct dirent *)uiop->uio_iov->iov_base; 2554 dp->d_fileno = (int)fileno; 2555 dp->d_namlen = len; 2556 dp->d_reclen = tlen + DIRHDSIZ; 2557 dp->d_type = DT_UNKNOWN; 2558 blksiz += dp->d_reclen; 2559 if (blksiz == DIRBLKSIZ) 2560 blksiz = 0; 2561 uiop->uio_offset += DIRHDSIZ; 2562 uiop->uio_resid -= DIRHDSIZ; 2563 uiop->uio_iov->iov_base = 2564 (char *)uiop->uio_iov->iov_base + DIRHDSIZ; 2565 uiop->uio_iov->iov_len -= DIRHDSIZ; 2566 cnp->cn_nameptr = uiop->uio_iov->iov_base; 2567 cnp->cn_namelen = len; 2568 nfsm_mtouio(uiop, len); 2569 cp = uiop->uio_iov->iov_base; 2570 tlen -= len; 2571 *cp = '\0'; 2572 uiop->uio_iov->iov_base = 2573 (char *)uiop->uio_iov->iov_base + tlen; 2574 uiop->uio_iov->iov_len -= tlen; 2575 uiop->uio_offset += tlen; 2576 uiop->uio_resid -= tlen; 2577 } else 2578 nfsm_adv(nfsm_rndup(len)); 2579 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED); 2580 if (bigenough) { 2581 cookie.nfsuquad[0] = *tl++; 2582 cookie.nfsuquad[1] = *tl++; 2583 } else 2584 tl += 2; 2585 2586 /* 2587 * Since the attributes are before the file handle 2588 * (sigh), we must skip over the attributes and then 2589 * come back and get them. 2590 */ 2591 attrflag = fxdr_unsigned(int, *tl); 2592 if (attrflag) { 2593 dpossav1 = dpos; 2594 mdsav1 = md; 2595 nfsm_adv(NFSX_V3FATTR); 2596 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); 2597 doit = fxdr_unsigned(int, *tl); 2598 /* 2599 * Skip loading the attrs for "..". There's a 2600 * race between loading the attrs here and 2601 * lookups that look for the directory currently 2602 * being read (in the parent). We try to acquire 2603 * the exclusive lock on ".." here, owning the 2604 * lock on the directory being read. Lookup will 2605 * hold the lock on ".." and try to acquire the 2606 * lock on the directory being read. 2607 * 2608 * There are other ways of fixing this, one would 2609 * be to do a trylock on the ".." vnode and skip 2610 * loading the attrs on ".." if it happens to be 2611 * locked by another process. But skipping the 2612 * attrload on ".." seems the easiest option. 2613 */ 2614 if (strcmp(dp->d_name, "..") == 0) { 2615 doit = 0; 2616 /* 2617 * We've already skipped over the attrs, 2618 * skip over the filehandle. And store d_type 2619 * as VDIR. 2620 */ 2621 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); 2622 i = fxdr_unsigned(int, *tl); 2623 nfsm_adv(nfsm_rndup(i)); 2624 dp->d_type = IFTODT(VTTOIF(VDIR)); 2625 } 2626 if (doit) { 2627 nfsm_getfh(fhp, fhsize, 1); 2628 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2629 VREF(vp); 2630 newvp = vp; 2631 np = dnp; 2632 } else { 2633 error = nfs_nget(vp->v_mount, fhp, 2634 fhsize, &np, LK_EXCLUSIVE); 2635 if (error) 2636 doit = 0; 2637 else 2638 newvp = NFSTOV(np); 2639 } 2640 } 2641 if (doit && bigenough) { 2642 dpossav2 = dpos; 2643 dpos = dpossav1; 2644 mdsav2 = md; 2645 md = mdsav1; 2646 nfsm_loadattr(newvp, &vattr); 2647 dpos = dpossav2; 2648 md = mdsav2; 2649 dp->d_type = IFTODT(VTTOIF(vattr.va_type)); 2650 ndp->ni_vp = newvp; 2651 if (newvp->v_type != VDIR || dattrflag != 0) 2652 cache_enter_time(ndp->ni_dvp, ndp->ni_vp, 2653 cnp, &vattr.va_ctime, 2654 newvp->v_type != VDIR ? NULL : 2655 &dvattr.va_ctime); 2656 } 2657 } else { 2658 /* Just skip over the file handle */ 2659 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); 2660 i = fxdr_unsigned(int, *tl); 2661 if (i) { 2662 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); 2663 fhsize = fxdr_unsigned(int, *tl); 2664 nfsm_adv(nfsm_rndup(fhsize)); 2665 } 2666 } 2667 if (newvp != NULLVP) { 2668 if (newvp == vp) 2669 vrele(newvp); 2670 else 2671 vput(newvp); 2672 newvp = NULLVP; 2673 } 2674 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); 2675 more_dirs = fxdr_unsigned(int, *tl); 2676 } 2677 /* 2678 * If at end of rpc data, get the eof boolean 2679 */ 2680 if (!more_dirs) { 2681 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); 2682 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2683 } 2684 m_freem(mrep); 2685 } 2686 /* 2687 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2688 * by increasing d_reclen for the last record. 2689 */ 2690 if (blksiz > 0) { 2691 left = DIRBLKSIZ - blksiz; 2692 dp->d_reclen += left; 2693 uiop->uio_iov->iov_base = 2694 (char *)uiop->uio_iov->iov_base + left; 2695 uiop->uio_iov->iov_len -= left; 2696 uiop->uio_offset += left; 2697 uiop->uio_resid -= left; 2698 } 2699 2700 /* 2701 * We are now either at the end of the directory or have filled the 2702 * block. 2703 */ 2704 if (bigenough) 2705 dnp->n_direofoffset = uiop->uio_offset; 2706 else { 2707 if (uiop->uio_resid > 0) 2708 nfs_printf("EEK! readdirplusrpc resid > 0\n"); 2709 nfs_dircookie_lock(dnp); 2710 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2711 *cookiep = cookie; 2712 nfs_dircookie_unlock(dnp); 2713 } 2714nfsmout: 2715 if (newvp != NULLVP) { 2716 if (newvp == vp) 2717 vrele(newvp); 2718 else 2719 vput(newvp); 2720 newvp = NULLVP; 2721 } 2722 return (error); 2723} 2724 2725/* 2726 * Silly rename. To make the NFS filesystem that is stateless look a little 2727 * more like the "ufs" a remove of an active vnode is translated to a rename 2728 * to a funny looking filename that is removed by nfs_inactive on the 2729 * nfsnode. There is the potential for another process on a different client 2730 * to create the same funny name between the nfs_lookitup() fails and the 2731 * nfs_rename() completes, but... 2732 */ 2733static int 2734nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2735{ 2736 struct sillyrename *sp; 2737 struct nfsnode *np; 2738 int error; 2739 short pid; 2740 unsigned int lticks; 2741 2742 cache_purge(dvp); 2743 np = VTONFS(vp); 2744 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir")); 2745 sp = malloc(sizeof (struct sillyrename), 2746 M_NFSREQ, M_WAITOK); 2747 sp->s_cred = crhold(cnp->cn_cred); 2748 sp->s_dvp = dvp; 2749 sp->s_removeit = nfs_removeit; 2750 VREF(dvp); 2751 2752 /* 2753 * Fudge together a funny name. 2754 * Changing the format of the funny name to accomodate more 2755 * sillynames per directory. 2756 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is 2757 * CPU ticks since boot. 2758 */ 2759 pid = cnp->cn_thread->td_proc->p_pid; 2760 lticks = (unsigned int)ticks; 2761 for ( ; ; ) { 2762 sp->s_namlen = sprintf(sp->s_name, 2763 ".nfs.%08x.%04x4.4", lticks, 2764 pid); 2765 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2766 cnp->cn_thread, NULL)) 2767 break; 2768 lticks++; 2769 } 2770 error = nfs_renameit(dvp, cnp, sp); 2771 if (error) 2772 goto bad; 2773 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2774 cnp->cn_thread, &np); 2775 np->n_sillyrename = sp; 2776 return (0); 2777bad: 2778 vrele(sp->s_dvp); 2779 crfree(sp->s_cred); 2780 free((caddr_t)sp, M_NFSREQ); 2781 return (error); 2782} 2783 2784/* 2785 * Look up a file name and optionally either update the file handle or 2786 * allocate an nfsnode, depending on the value of npp. 2787 * npp == NULL --> just do the lookup 2788 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2789 * handled too 2790 * *npp != NULL --> update the file handle in the vnode 2791 */ 2792static int 2793nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred, 2794 struct thread *td, struct nfsnode **npp) 2795{ 2796 struct vnode *newvp = NULL; 2797 struct nfsnode *np, *dnp = VTONFS(dvp); 2798 caddr_t bpos, dpos; 2799 int error = 0, fhlen, attrflag; 2800 struct mbuf *mreq, *mrep, *md, *mb; 2801 nfsfh_t *nfhp; 2802 int v3 = NFS_ISV3(dvp); 2803 2804 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2805 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP, 2806 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2807 mb = mreq; 2808 bpos = mtod(mb, caddr_t); 2809 nfsm_fhtom(dvp, v3); 2810 nfsm_strtom(name, len, NFS_MAXNAMLEN); 2811 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred); 2812 if (npp && !error) { 2813 nfsm_getfh(nfhp, fhlen, v3); 2814 if (*npp) { 2815 np = *npp; 2816 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2817 free((caddr_t)np->n_fhp, M_NFSBIGFH); 2818 np->n_fhp = &np->n_fh; 2819 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH) 2820 np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK); 2821 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen); 2822 np->n_fhsize = fhlen; 2823 newvp = NFSTOV(np); 2824 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2825 VREF(dvp); 2826 newvp = dvp; 2827 } else { 2828 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, LK_EXCLUSIVE); 2829 if (error) { 2830 m_freem(mrep); 2831 return (error); 2832 } 2833 newvp = NFSTOV(np); 2834 } 2835 if (v3) { 2836 nfsm_postop_attr(newvp, attrflag); 2837 if (!attrflag && *npp == NULL) { 2838 m_freem(mrep); 2839 if (newvp == dvp) 2840 vrele(newvp); 2841 else 2842 vput(newvp); 2843 return (ENOENT); 2844 } 2845 } else 2846 nfsm_loadattr(newvp, NULL); 2847 } 2848 m_freem(mrep); 2849nfsmout: 2850 if (npp && *npp == NULL) { 2851 if (error) { 2852 if (newvp) { 2853 if (newvp == dvp) 2854 vrele(newvp); 2855 else 2856 vput(newvp); 2857 } 2858 } else 2859 *npp = np; 2860 } 2861 return (error); 2862} 2863 2864/* 2865 * Nfs Version 3 commit rpc 2866 */ 2867int 2868nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred, 2869 struct thread *td) 2870{ 2871 u_int32_t *tl; 2872 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2873 caddr_t bpos, dpos; 2874 int error = 0, wccflag = NFSV3_WCCRATTR; 2875 struct mbuf *mreq, *mrep, *md, *mb; 2876 2877 mtx_lock(&nmp->nm_mtx); 2878 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) { 2879 mtx_unlock(&nmp->nm_mtx); 2880 return (0); 2881 } 2882 mtx_unlock(&nmp->nm_mtx); 2883 nfsstats.rpccnt[NFSPROC_COMMIT]++; 2884 mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1)); 2885 mb = mreq; 2886 bpos = mtod(mb, caddr_t); 2887 nfsm_fhtom(vp, 1); 2888 tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED); 2889 txdr_hyper(offset, tl); 2890 tl += 2; 2891 *tl = txdr_unsigned(cnt); 2892 nfsm_request(vp, NFSPROC_COMMIT, td, cred); 2893 nfsm_wcc_data(vp, wccflag); 2894 if (!error) { 2895 tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF); 2896 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl, 2897 NFSX_V3WRITEVERF)) { 2898 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 2899 NFSX_V3WRITEVERF); 2900 error = NFSERR_STALEWRITEVERF; 2901 } 2902 } 2903 m_freem(mrep); 2904nfsmout: 2905 return (error); 2906} 2907 2908/* 2909 * Strategy routine. 2910 * For async requests when nfsiod(s) are running, queue the request by 2911 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 2912 * request. 2913 */ 2914static int 2915nfs_strategy(struct vop_strategy_args *ap) 2916{ 2917 struct buf *bp = ap->a_bp; 2918 struct ucred *cr; 2919 2920 KASSERT(!(bp->b_flags & B_DONE), 2921 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); 2922 BUF_ASSERT_HELD(bp); 2923 2924 if (bp->b_iocmd == BIO_READ) 2925 cr = bp->b_rcred; 2926 else 2927 cr = bp->b_wcred; 2928 2929 /* 2930 * If the op is asynchronous and an i/o daemon is waiting 2931 * queue the request, wake it up and wait for completion 2932 * otherwise just do it ourselves. 2933 */ 2934 if ((bp->b_flags & B_ASYNC) == 0 || 2935 nfs_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread)) 2936 (void)nfs_doio(ap->a_vp, bp, cr, curthread); 2937 return (0); 2938} 2939 2940/* 2941 * fsync vnode op. Just call nfs_flush() with commit == 1. 2942 */ 2943/* ARGSUSED */ 2944static int 2945nfs_fsync(struct vop_fsync_args *ap) 2946{ 2947 2948 return (nfs_flush(ap->a_vp, ap->a_waitfor, 1)); 2949} 2950 2951/* 2952 * Flush all the blocks associated with a vnode. 2953 * Walk through the buffer pool and push any dirty pages 2954 * associated with the vnode. 2955 */ 2956static int 2957nfs_flush(struct vnode *vp, int waitfor, int commit) 2958{ 2959 struct nfsnode *np = VTONFS(vp); 2960 struct buf *bp; 2961 int i; 2962 struct buf *nbp; 2963 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2964 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos; 2965 int passone = 1; 2966 u_quad_t off, endoff, toff; 2967 struct ucred* wcred = NULL; 2968 struct buf **bvec = NULL; 2969 struct bufobj *bo; 2970 struct thread *td = curthread; 2971#ifndef NFS_COMMITBVECSIZ 2972#define NFS_COMMITBVECSIZ 20 2973#endif 2974 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ]; 2975 int bvecsize = 0, bveccount; 2976 2977 if (nmp->nm_flag & NFSMNT_INT) 2978 slpflag = NFS_PCATCH; 2979 if (!commit) 2980 passone = 0; 2981 bo = &vp->v_bufobj; 2982 /* 2983 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the 2984 * server, but has not been committed to stable storage on the server 2985 * yet. On the first pass, the byte range is worked out and the commit 2986 * rpc is done. On the second pass, nfs_writebp() is called to do the 2987 * job. 2988 */ 2989again: 2990 off = (u_quad_t)-1; 2991 endoff = 0; 2992 bvecpos = 0; 2993 if (NFS_ISV3(vp) && commit) { 2994 if (bvec != NULL && bvec != bvec_on_stack) 2995 free(bvec, M_TEMP); 2996 /* 2997 * Count up how many buffers waiting for a commit. 2998 */ 2999 bveccount = 0; 3000 BO_LOCK(bo); 3001 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 3002 if (!BUF_ISLOCKED(bp) && 3003 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) 3004 == (B_DELWRI | B_NEEDCOMMIT)) 3005 bveccount++; 3006 } 3007 /* 3008 * Allocate space to remember the list of bufs to commit. It is 3009 * important to use M_NOWAIT here to avoid a race with nfs_write. 3010 * If we can't get memory (for whatever reason), we will end up 3011 * committing the buffers one-by-one in the loop below. 3012 */ 3013 if (bveccount > NFS_COMMITBVECSIZ) { 3014 /* 3015 * Release the vnode interlock to avoid a lock 3016 * order reversal. 3017 */ 3018 BO_UNLOCK(bo); 3019 bvec = (struct buf **) 3020 malloc(bveccount * sizeof(struct buf *), 3021 M_TEMP, M_NOWAIT); 3022 BO_LOCK(bo); 3023 if (bvec == NULL) { 3024 bvec = bvec_on_stack; 3025 bvecsize = NFS_COMMITBVECSIZ; 3026 } else 3027 bvecsize = bveccount; 3028 } else { 3029 bvec = bvec_on_stack; 3030 bvecsize = NFS_COMMITBVECSIZ; 3031 } 3032 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 3033 if (bvecpos >= bvecsize) 3034 break; 3035 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 3036 nbp = TAILQ_NEXT(bp, b_bobufs); 3037 continue; 3038 } 3039 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 3040 (B_DELWRI | B_NEEDCOMMIT)) { 3041 BUF_UNLOCK(bp); 3042 nbp = TAILQ_NEXT(bp, b_bobufs); 3043 continue; 3044 } 3045 BO_UNLOCK(bo); 3046 bremfree(bp); 3047 /* 3048 * Work out if all buffers are using the same cred 3049 * so we can deal with them all with one commit. 3050 * 3051 * NOTE: we are not clearing B_DONE here, so we have 3052 * to do it later on in this routine if we intend to 3053 * initiate I/O on the bp. 3054 * 3055 * Note: to avoid loopback deadlocks, we do not 3056 * assign b_runningbufspace. 3057 */ 3058 if (wcred == NULL) 3059 wcred = bp->b_wcred; 3060 else if (wcred != bp->b_wcred) 3061 wcred = NOCRED; 3062 vfs_busy_pages(bp, 1); 3063 3064 BO_LOCK(bo); 3065 /* 3066 * bp is protected by being locked, but nbp is not 3067 * and vfs_busy_pages() may sleep. We have to 3068 * recalculate nbp. 3069 */ 3070 nbp = TAILQ_NEXT(bp, b_bobufs); 3071 3072 /* 3073 * A list of these buffers is kept so that the 3074 * second loop knows which buffers have actually 3075 * been committed. This is necessary, since there 3076 * may be a race between the commit rpc and new 3077 * uncommitted writes on the file. 3078 */ 3079 bvec[bvecpos++] = bp; 3080 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 3081 bp->b_dirtyoff; 3082 if (toff < off) 3083 off = toff; 3084 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); 3085 if (toff > endoff) 3086 endoff = toff; 3087 } 3088 BO_UNLOCK(bo); 3089 } 3090 if (bvecpos > 0) { 3091 /* 3092 * Commit data on the server, as required. 3093 * If all bufs are using the same wcred, then use that with 3094 * one call for all of them, otherwise commit each one 3095 * separately. 3096 */ 3097 if (wcred != NOCRED) 3098 retv = nfs_commit(vp, off, (int)(endoff - off), 3099 wcred, td); 3100 else { 3101 retv = 0; 3102 for (i = 0; i < bvecpos; i++) { 3103 off_t off, size; 3104 bp = bvec[i]; 3105 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 3106 bp->b_dirtyoff; 3107 size = (u_quad_t)(bp->b_dirtyend 3108 - bp->b_dirtyoff); 3109 retv = nfs_commit(vp, off, (int)size, 3110 bp->b_wcred, td); 3111 if (retv) break; 3112 } 3113 } 3114 3115 if (retv == NFSERR_STALEWRITEVERF) 3116 nfs_clearcommit(vp->v_mount); 3117 3118 /* 3119 * Now, either mark the blocks I/O done or mark the 3120 * blocks dirty, depending on whether the commit 3121 * succeeded. 3122 */ 3123 for (i = 0; i < bvecpos; i++) { 3124 bp = bvec[i]; 3125 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 3126 if (retv) { 3127 /* 3128 * Error, leave B_DELWRI intact 3129 */ 3130 vfs_unbusy_pages(bp); 3131 brelse(bp); 3132 } else { 3133 /* 3134 * Success, remove B_DELWRI ( bundirty() ). 3135 * 3136 * b_dirtyoff/b_dirtyend seem to be NFS 3137 * specific. We should probably move that 3138 * into bundirty(). XXX 3139 */ 3140 bufobj_wref(bo); 3141 bp->b_flags |= B_ASYNC; 3142 bundirty(bp); 3143 bp->b_flags &= ~B_DONE; 3144 bp->b_ioflags &= ~BIO_ERROR; 3145 bp->b_dirtyoff = bp->b_dirtyend = 0; 3146 bufdone(bp); 3147 } 3148 } 3149 } 3150 3151 /* 3152 * Start/do any write(s) that are required. 3153 */ 3154loop: 3155 BO_LOCK(bo); 3156 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 3157 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 3158 if (waitfor != MNT_WAIT || passone) 3159 continue; 3160 3161 error = BUF_TIMELOCK(bp, 3162 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 3163 BO_MTX(bo), "nfsfsync", slpflag, slptimeo); 3164 if (error == 0) { 3165 BUF_UNLOCK(bp); 3166 goto loop; 3167 } 3168 if (error == ENOLCK) { 3169 error = 0; 3170 goto loop; 3171 } 3172 if (nfs_sigintr(nmp, td)) { 3173 error = EINTR; 3174 goto done; 3175 } 3176 if (slpflag & PCATCH) { 3177 slpflag = 0; 3178 slptimeo = 2 * hz; 3179 } 3180 goto loop; 3181 } 3182 if ((bp->b_flags & B_DELWRI) == 0) 3183 panic("nfs_fsync: not dirty"); 3184 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) { 3185 BUF_UNLOCK(bp); 3186 continue; 3187 } 3188 BO_UNLOCK(bo); 3189 bremfree(bp); 3190 if (passone || !commit) 3191 bp->b_flags |= B_ASYNC; 3192 else 3193 bp->b_flags |= B_ASYNC; 3194 bwrite(bp); 3195 if (nfs_sigintr(nmp, td)) { 3196 error = EINTR; 3197 goto done; 3198 } 3199 goto loop; 3200 } 3201 if (passone) { 3202 passone = 0; 3203 BO_UNLOCK(bo); 3204 goto again; 3205 } 3206 if (waitfor == MNT_WAIT) { 3207 while (bo->bo_numoutput) { 3208 error = bufobj_wwait(bo, slpflag, slptimeo); 3209 if (error) { 3210 BO_UNLOCK(bo); 3211 error = nfs_sigintr(nmp, td); 3212 if (error) 3213 goto done; 3214 if (slpflag & PCATCH) { 3215 slpflag = 0; 3216 slptimeo = 2 * hz; 3217 } 3218 BO_LOCK(bo); 3219 } 3220 } 3221 if (bo->bo_dirty.bv_cnt != 0 && commit) { 3222 BO_UNLOCK(bo); 3223 goto loop; 3224 } 3225 /* 3226 * Wait for all the async IO requests to drain 3227 */ 3228 BO_UNLOCK(bo); 3229 mtx_lock(&np->n_mtx); 3230 while (np->n_directio_asyncwr > 0) { 3231 np->n_flag |= NFSYNCWAIT; 3232 error = nfs_msleep(td, (caddr_t)&np->n_directio_asyncwr, 3233 &np->n_mtx, slpflag | (PRIBIO + 1), 3234 "nfsfsync", 0); 3235 if (error) { 3236 if (nfs_sigintr(nmp, td)) { 3237 mtx_unlock(&np->n_mtx); 3238 error = EINTR; 3239 goto done; 3240 } 3241 } 3242 } 3243 mtx_unlock(&np->n_mtx); 3244 } else 3245 BO_UNLOCK(bo); 3246 mtx_lock(&np->n_mtx); 3247 if (np->n_flag & NWRITEERR) { 3248 error = np->n_error; 3249 np->n_flag &= ~NWRITEERR; 3250 } 3251 if (commit && bo->bo_dirty.bv_cnt == 0 && 3252 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0) 3253 np->n_flag &= ~NMODIFIED; 3254 mtx_unlock(&np->n_mtx); 3255done: 3256 if (bvec != NULL && bvec != bvec_on_stack) 3257 free(bvec, M_TEMP); 3258 return (error); 3259} 3260 3261/* 3262 * NFS advisory byte-level locks. 3263 */ 3264static int 3265nfs_advlock(struct vop_advlock_args *ap) 3266{ 3267 struct vnode *vp = ap->a_vp; 3268 u_quad_t size; 3269 int error; 3270 3271 error = vn_lock(vp, LK_SHARED); 3272 if (error) 3273 return (error); 3274 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3275 size = VTONFS(vp)->n_size; 3276 VOP_UNLOCK(vp, 0); 3277 error = lf_advlock(ap, &(vp->v_lockf), size); 3278 } else { 3279 if (nfs_advlock_p) 3280 error = nfs_advlock_p(ap); 3281 else 3282 error = ENOLCK; 3283 } 3284 3285 return (error); 3286} 3287 3288/* 3289 * NFS advisory byte-level locks. 3290 */ 3291static int 3292nfs_advlockasync(struct vop_advlockasync_args *ap) 3293{ 3294 struct vnode *vp = ap->a_vp; 3295 u_quad_t size; 3296 int error; 3297 3298 error = vn_lock(vp, LK_SHARED); 3299 if (error) 3300 return (error); 3301 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3302 size = VTONFS(vp)->n_size; 3303 VOP_UNLOCK(vp, 0); 3304 error = lf_advlockasync(ap, &(vp->v_lockf), size); 3305 } else { 3306 VOP_UNLOCK(vp, 0); 3307 error = EOPNOTSUPP; 3308 } 3309 return (error); 3310} 3311 3312/* 3313 * Print out the contents of an nfsnode. 3314 */ 3315static int 3316nfs_print(struct vop_print_args *ap) 3317{ 3318 struct vnode *vp = ap->a_vp; 3319 struct nfsnode *np = VTONFS(vp); 3320 3321 nfs_printf("\tfileid %ld fsid 0x%x", 3322 np->n_vattr.va_fileid, np->n_vattr.va_fsid); 3323 if (vp->v_type == VFIFO) 3324 fifo_printinfo(vp); 3325 printf("\n"); 3326 return (0); 3327} 3328 3329/* 3330 * This is the "real" nfs::bwrite(struct buf*). 3331 * We set B_CACHE if this is a VMIO buffer. 3332 */ 3333int 3334nfs_writebp(struct buf *bp, int force __unused, struct thread *td) 3335{ 3336 int s; 3337 int oldflags = bp->b_flags; 3338#if 0 3339 int retv = 1; 3340 off_t off; 3341#endif 3342 3343 BUF_ASSERT_HELD(bp); 3344 3345 if (bp->b_flags & B_INVAL) { 3346 brelse(bp); 3347 return(0); 3348 } 3349 3350 bp->b_flags |= B_CACHE; 3351 3352 /* 3353 * Undirty the bp. We will redirty it later if the I/O fails. 3354 */ 3355 3356 s = splbio(); 3357 bundirty(bp); 3358 bp->b_flags &= ~B_DONE; 3359 bp->b_ioflags &= ~BIO_ERROR; 3360 bp->b_iocmd = BIO_WRITE; 3361 3362 bufobj_wref(bp->b_bufobj); 3363 curthread->td_ru.ru_oublock++; 3364 splx(s); 3365 3366 /* 3367 * Note: to avoid loopback deadlocks, we do not 3368 * assign b_runningbufspace. 3369 */ 3370 vfs_busy_pages(bp, 1); 3371 3372 BUF_KERNPROC(bp); 3373 bp->b_iooffset = dbtob(bp->b_blkno); 3374 bstrategy(bp); 3375 3376 if( (oldflags & B_ASYNC) == 0) { 3377 int rtval = bufwait(bp); 3378 3379 if (oldflags & B_DELWRI) { 3380 s = splbio(); 3381 reassignbuf(bp); 3382 splx(s); 3383 } 3384 brelse(bp); 3385 return (rtval); 3386 } 3387 3388 return (0); 3389} 3390 3391/* 3392 * nfs special file access vnode op. 3393 * Essentially just get vattr and then imitate iaccess() since the device is 3394 * local to the client. 3395 */ 3396static int 3397nfsspec_access(struct vop_access_args *ap) 3398{ 3399 struct vattr *vap; 3400 struct ucred *cred = ap->a_cred; 3401 struct vnode *vp = ap->a_vp; 3402 accmode_t accmode = ap->a_accmode; 3403 struct vattr vattr; 3404 int error; 3405 3406 /* 3407 * Disallow write attempts on filesystems mounted read-only; 3408 * unless the file is a socket, fifo, or a block or character 3409 * device resident on the filesystem. 3410 */ 3411 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3412 switch (vp->v_type) { 3413 case VREG: 3414 case VDIR: 3415 case VLNK: 3416 return (EROFS); 3417 default: 3418 break; 3419 } 3420 } 3421 vap = &vattr; 3422 error = VOP_GETATTR(vp, vap, cred); 3423 if (error) 3424 goto out; 3425 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid, 3426 accmode, cred, NULL); 3427out: 3428 return error; 3429} 3430 3431/* 3432 * Read wrapper for fifos. 3433 */ 3434static int 3435nfsfifo_read(struct vop_read_args *ap) 3436{ 3437 struct nfsnode *np = VTONFS(ap->a_vp); 3438 int error; 3439 3440 /* 3441 * Set access flag. 3442 */ 3443 mtx_lock(&np->n_mtx); 3444 np->n_flag |= NACC; 3445 getnanotime(&np->n_atim); 3446 mtx_unlock(&np->n_mtx); 3447 error = fifo_specops.vop_read(ap); 3448 return error; 3449} 3450 3451/* 3452 * Write wrapper for fifos. 3453 */ 3454static int 3455nfsfifo_write(struct vop_write_args *ap) 3456{ 3457 struct nfsnode *np = VTONFS(ap->a_vp); 3458 3459 /* 3460 * Set update flag. 3461 */ 3462 mtx_lock(&np->n_mtx); 3463 np->n_flag |= NUPD; 3464 getnanotime(&np->n_mtim); 3465 mtx_unlock(&np->n_mtx); 3466 return(fifo_specops.vop_write(ap)); 3467} 3468 3469/* 3470 * Close wrapper for fifos. 3471 * 3472 * Update the times on the nfsnode then do fifo close. 3473 */ 3474static int 3475nfsfifo_close(struct vop_close_args *ap) 3476{ 3477 struct vnode *vp = ap->a_vp; 3478 struct nfsnode *np = VTONFS(vp); 3479 struct vattr vattr; 3480 struct timespec ts; 3481 3482 mtx_lock(&np->n_mtx); 3483 if (np->n_flag & (NACC | NUPD)) { 3484 getnanotime(&ts); 3485 if (np->n_flag & NACC) 3486 np->n_atim = ts; 3487 if (np->n_flag & NUPD) 3488 np->n_mtim = ts; 3489 np->n_flag |= NCHG; 3490 if (vrefcnt(vp) == 1 && 3491 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3492 VATTR_NULL(&vattr); 3493 if (np->n_flag & NACC) 3494 vattr.va_atime = np->n_atim; 3495 if (np->n_flag & NUPD) 3496 vattr.va_mtime = np->n_mtim; 3497 mtx_unlock(&np->n_mtx); 3498 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3499 goto out; 3500 } 3501 } 3502 mtx_unlock(&np->n_mtx); 3503out: 3504 return (fifo_specops.vop_close(ap)); 3505} 3506 3507/* 3508 * Just call nfs_writebp() with the force argument set to 1. 3509 * 3510 * NOTE: B_DONE may or may not be set in a_bp on call. 3511 */ 3512static int 3513nfs_bwrite(struct buf *bp) 3514{ 3515 3516 return (nfs_writebp(bp, 1, curthread)); 3517} 3518 3519struct buf_ops buf_ops_nfs = { 3520 .bop_name = "buf_ops_nfs", 3521 .bop_write = nfs_bwrite, 3522 .bop_strategy = bufstrategy, 3523 .bop_sync = bufsync, 3524 .bop_bdflush = bufbdflush, 3525}; 3526