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("$FreeBSD$"); 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 <fs/fifofs/fifo.h> 71 72#include <nfs/nfsproto.h> 73#include <nfsclient/nfs.h> 74#include <nfsclient/nfsnode.h> 75#include <nfsclient/nfsmount.h> 76#include <nfs/nfs_kdtrace.h> 77#include <nfs/nfs_lock.h> 78#include <nfs/xdr_subs.h> 79#include <nfsclient/nfsm_subs.h> 80 81#include <net/if.h> 82#include <netinet/in.h> 83#include <netinet/in_var.h> 84 85#include <machine/stdarg.h> 86 87#ifdef KDTRACE_HOOKS 88#include <sys/dtrace_bsd.h> 89 90dtrace_nfsclient_accesscache_flush_probe_func_t 91 dtrace_nfsclient_accesscache_flush_done_probe; 92uint32_t nfsclient_accesscache_flush_done_id; 93 94dtrace_nfsclient_accesscache_get_probe_func_t 95 dtrace_nfsclient_accesscache_get_hit_probe, 96 dtrace_nfsclient_accesscache_get_miss_probe; 97uint32_t nfsclient_accesscache_get_hit_id; 98uint32_t nfsclient_accesscache_get_miss_id; 99 100dtrace_nfsclient_accesscache_load_probe_func_t 101 dtrace_nfsclient_accesscache_load_done_probe; 102uint32_t nfsclient_accesscache_load_done_id; 103#endif /* !KDTRACE_HOOKS */ 104 105/* Defs */ 106#define TRUE 1 107#define FALSE 0 108 109/* 110 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these 111 * calls are not in getblk() and brelse() so that they would not be necessary 112 * here. 113 */ 114#ifndef B_VMIO 115#define vfs_busy_pages(bp, f) 116#endif 117 118static vop_read_t nfsfifo_read; 119static vop_write_t nfsfifo_write; 120static vop_close_t nfsfifo_close; 121static int nfs_flush(struct vnode *, int, int); 122static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *); 123static vop_lookup_t nfs_lookup; 124static vop_create_t nfs_create; 125static vop_mknod_t nfs_mknod; 126static vop_open_t nfs_open; 127static vop_close_t nfs_close; 128static vop_access_t nfs_access; 129static vop_getattr_t nfs_getattr; 130static vop_setattr_t nfs_setattr; 131static vop_read_t nfs_read; 132static vop_fsync_t nfs_fsync; 133static vop_remove_t nfs_remove; 134static vop_link_t nfs_link; 135static vop_rename_t nfs_rename; 136static vop_mkdir_t nfs_mkdir; 137static vop_rmdir_t nfs_rmdir; 138static vop_symlink_t nfs_symlink; 139static vop_readdir_t nfs_readdir; 140static vop_strategy_t nfs_strategy; 141static int nfs_lookitup(struct vnode *, const char *, int, 142 struct ucred *, struct thread *, struct nfsnode **); 143static int nfs_sillyrename(struct vnode *, struct vnode *, 144 struct componentname *); 145static vop_access_t nfsspec_access; 146static vop_readlink_t nfs_readlink; 147static vop_print_t nfs_print; 148static vop_advlock_t nfs_advlock; 149static vop_advlockasync_t nfs_advlockasync; 150 151/* 152 * Global vfs data structures for nfs 153 */ 154struct vop_vector nfs_vnodeops = { 155 .vop_default = &default_vnodeops, 156 .vop_access = nfs_access, 157 .vop_advlock = nfs_advlock, 158 .vop_advlockasync = nfs_advlockasync, 159 .vop_close = nfs_close, 160 .vop_create = nfs_create, 161 .vop_fsync = nfs_fsync, 162 .vop_getattr = nfs_getattr, 163 .vop_getpages = nfs_getpages, 164 .vop_putpages = nfs_putpages, 165 .vop_inactive = nfs_inactive, 166 .vop_link = nfs_link, 167 .vop_lookup = nfs_lookup, 168 .vop_mkdir = nfs_mkdir, 169 .vop_mknod = nfs_mknod, 170 .vop_open = nfs_open, 171 .vop_print = nfs_print, 172 .vop_read = nfs_read, 173 .vop_readdir = nfs_readdir, 174 .vop_readlink = nfs_readlink, 175 .vop_reclaim = nfs_reclaim, 176 .vop_remove = nfs_remove, 177 .vop_rename = nfs_rename, 178 .vop_rmdir = nfs_rmdir, 179 .vop_setattr = nfs_setattr, 180 .vop_strategy = nfs_strategy, 181 .vop_symlink = nfs_symlink, 182 .vop_write = nfs_write, 183}; 184 185struct vop_vector nfs_fifoops = { 186 .vop_default = &fifo_specops, 187 .vop_access = nfsspec_access, 188 .vop_close = nfsfifo_close, 189 .vop_fsync = nfs_fsync, 190 .vop_getattr = nfs_getattr, 191 .vop_inactive = nfs_inactive, 192 .vop_print = nfs_print, 193 .vop_read = nfsfifo_read, 194 .vop_reclaim = nfs_reclaim, 195 .vop_setattr = nfs_setattr, 196 .vop_write = nfsfifo_write, 197}; 198 199static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, 200 struct componentname *cnp, struct vattr *vap); 201static int nfs_removerpc(struct vnode *dvp, const char *name, int namelen, 202 struct ucred *cred, struct thread *td); 203static int nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, 204 int fnamelen, struct vnode *tdvp, 205 const char *tnameptr, int tnamelen, 206 struct ucred *cred, struct thread *td); 207static int nfs_renameit(struct vnode *sdvp, struct componentname *scnp, 208 struct sillyrename *sp); 209 210/* 211 * Global variables 212 */ 213struct mtx nfs_iod_mtx; 214enum nfsiod_state nfs_iodwant[NFS_MAXASYNCDAEMON]; 215struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON]; 216int nfs_numasync = 0; 217#define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1)) 218 219SYSCTL_DECL(_vfs_oldnfs); 220 221static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO; 222SYSCTL_INT(_vfs_oldnfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 223 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 224 225static int nfs_prime_access_cache = 0; 226SYSCTL_INT(_vfs_oldnfs, OID_AUTO, prime_access_cache, CTLFLAG_RW, 227 &nfs_prime_access_cache, 0, 228 "Prime NFS ACCESS cache when fetching attributes"); 229 230static int nfsv3_commit_on_close = 0; 231SYSCTL_INT(_vfs_oldnfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW, 232 &nfsv3_commit_on_close, 0, "write+commit on close, else only write"); 233 234static int nfs_clean_pages_on_close = 1; 235SYSCTL_INT(_vfs_oldnfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW, 236 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close"); 237 238int nfs_directio_enable = 0; 239SYSCTL_INT(_vfs_oldnfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW, 240 &nfs_directio_enable, 0, "Enable NFS directio"); 241 242/* 243 * This sysctl allows other processes to mmap a file that has been opened 244 * O_DIRECT by a process. In general, having processes mmap the file while 245 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow 246 * this by default to prevent DoS attacks - to prevent a malicious user from 247 * opening up files O_DIRECT preventing other users from mmap'ing these 248 * files. "Protected" environments where stricter consistency guarantees are 249 * required can disable this knob. The process that opened the file O_DIRECT 250 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not 251 * meaningful. 252 */ 253int nfs_directio_allow_mmap = 1; 254SYSCTL_INT(_vfs_oldnfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW, 255 &nfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens"); 256 257#if 0 258SYSCTL_INT(_vfs_oldnfs, OID_AUTO, access_cache_hits, CTLFLAG_RD, 259 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count"); 260 261SYSCTL_INT(_vfs_oldnfs, OID_AUTO, access_cache_misses, CTLFLAG_RD, 262 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count"); 263#endif 264 265#define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \ 266 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \ 267 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP) 268 269/* 270 * SMP Locking Note : 271 * The list of locks after the description of the lock is the ordering 272 * of other locks acquired with the lock held. 273 * np->n_mtx : Protects the fields in the nfsnode. 274 VM Object Lock 275 VI_MTX (acquired indirectly) 276 * nmp->nm_mtx : Protects the fields in the nfsmount. 277 rep->r_mtx 278 * nfs_iod_mtx : Global lock, protects shared nfsiod state. 279 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list. 280 nmp->nm_mtx 281 rep->r_mtx 282 * rep->r_mtx : Protects the fields in an nfsreq. 283 */ 284 285static int 286nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td, 287 struct ucred *cred, uint32_t *retmode) 288{ 289 const int v3 = 1; 290 u_int32_t *tl; 291 int error = 0, attrflag, i, lrupos; 292 293 struct mbuf *mreq, *mrep, *md, *mb; 294 caddr_t bpos, dpos; 295 u_int32_t rmode; 296 struct nfsnode *np = VTONFS(vp); 297 298 nfsstats.rpccnt[NFSPROC_ACCESS]++; 299 mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 300 mb = mreq; 301 bpos = mtod(mb, caddr_t); 302 nfsm_fhtom(vp, v3); 303 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED); 304 *tl = txdr_unsigned(wmode); 305 nfsm_request(vp, NFSPROC_ACCESS, td, cred); 306 nfsm_postop_attr(vp, attrflag); 307 if (!error) { 308 lrupos = 0; 309 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); 310 rmode = fxdr_unsigned(u_int32_t, *tl); 311 mtx_lock(&np->n_mtx); 312 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 313 if (np->n_accesscache[i].uid == cred->cr_uid) { 314 np->n_accesscache[i].mode = rmode; 315 np->n_accesscache[i].stamp = time_second; 316 break; 317 } 318 if (i > 0 && np->n_accesscache[i].stamp < 319 np->n_accesscache[lrupos].stamp) 320 lrupos = i; 321 } 322 if (i == NFS_ACCESSCACHESIZE) { 323 np->n_accesscache[lrupos].uid = cred->cr_uid; 324 np->n_accesscache[lrupos].mode = rmode; 325 np->n_accesscache[lrupos].stamp = time_second; 326 } 327 mtx_unlock(&np->n_mtx); 328 if (retmode != NULL) 329 *retmode = rmode; 330 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0); 331 } 332 m_freem(mrep); 333nfsmout: 334#ifdef KDTRACE_HOOKS 335 if (error) { 336 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0, 337 error); 338 } 339#endif 340 return (error); 341} 342 343/* 344 * nfs access vnode op. 345 * For nfs version 2, just return ok. File accesses may fail later. 346 * For nfs version 3, use the access rpc to check accessibility. If file modes 347 * are changed on the server, accesses might still fail later. 348 */ 349static int 350nfs_access(struct vop_access_args *ap) 351{ 352 struct vnode *vp = ap->a_vp; 353 int error = 0, i, gotahit; 354 u_int32_t mode, rmode, wmode; 355 int v3 = NFS_ISV3(vp); 356 struct nfsnode *np = VTONFS(vp); 357 358 /* 359 * Disallow write attempts on filesystems mounted read-only; 360 * unless the file is a socket, fifo, or a block or character 361 * device resident on the filesystem. 362 */ 363 if ((ap->a_accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 364 switch (vp->v_type) { 365 case VREG: 366 case VDIR: 367 case VLNK: 368 return (EROFS); 369 default: 370 break; 371 } 372 } 373 /* 374 * For nfs v3, check to see if we have done this recently, and if 375 * so return our cached result instead of making an ACCESS call. 376 * If not, do an access rpc, otherwise you are stuck emulating 377 * ufs_access() locally using the vattr. This may not be correct, 378 * since the server may apply other access criteria such as 379 * client uid-->server uid mapping that we do not know about. 380 */ 381 if (v3) { 382 if (ap->a_accmode & VREAD) 383 mode = NFSV3ACCESS_READ; 384 else 385 mode = 0; 386 if (vp->v_type != VDIR) { 387 if (ap->a_accmode & VWRITE) 388 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 389 if (ap->a_accmode & VEXEC) 390 mode |= NFSV3ACCESS_EXECUTE; 391 } else { 392 if (ap->a_accmode & VWRITE) 393 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 394 NFSV3ACCESS_DELETE); 395 if (ap->a_accmode & VEXEC) 396 mode |= NFSV3ACCESS_LOOKUP; 397 } 398 /* XXX safety belt, only make blanket request if caching */ 399 if (nfsaccess_cache_timeout > 0) { 400 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY | 401 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE | 402 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP; 403 } else { 404 wmode = mode; 405 } 406 407 /* 408 * Does our cached result allow us to give a definite yes to 409 * this request? 410 */ 411 gotahit = 0; 412 mtx_lock(&np->n_mtx); 413 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 414 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) { 415 if (time_second < (np->n_accesscache[i].stamp + 416 nfsaccess_cache_timeout) && 417 (np->n_accesscache[i].mode & mode) == mode) { 418 nfsstats.accesscache_hits++; 419 gotahit = 1; 420 } 421 break; 422 } 423 } 424 mtx_unlock(&np->n_mtx); 425#ifdef KDTRACE_HOOKS 426 if (gotahit) 427 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp, 428 ap->a_cred->cr_uid, mode); 429 else 430 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp, 431 ap->a_cred->cr_uid, mode); 432#endif 433 if (gotahit == 0) { 434 /* 435 * Either a no, or a don't know. Go to the wire. 436 */ 437 nfsstats.accesscache_misses++; 438 error = nfs3_access_otw(vp, wmode, ap->a_td, ap->a_cred, 439 &rmode); 440 if (!error) { 441 if ((rmode & mode) != mode) 442 error = EACCES; 443 } 444 } 445 return (error); 446 } else { 447 if ((error = nfsspec_access(ap)) != 0) { 448 return (error); 449 } 450 /* 451 * Attempt to prevent a mapped root from accessing a file 452 * which it shouldn't. We try to read a byte from the file 453 * if the user is root and the file is not zero length. 454 * After calling nfsspec_access, we should have the correct 455 * file size cached. 456 */ 457 mtx_lock(&np->n_mtx); 458 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD) 459 && VTONFS(vp)->n_size > 0) { 460 struct iovec aiov; 461 struct uio auio; 462 char buf[1]; 463 464 mtx_unlock(&np->n_mtx); 465 aiov.iov_base = buf; 466 aiov.iov_len = 1; 467 auio.uio_iov = &aiov; 468 auio.uio_iovcnt = 1; 469 auio.uio_offset = 0; 470 auio.uio_resid = 1; 471 auio.uio_segflg = UIO_SYSSPACE; 472 auio.uio_rw = UIO_READ; 473 auio.uio_td = ap->a_td; 474 475 if (vp->v_type == VREG) 476 error = nfs_readrpc(vp, &auio, ap->a_cred); 477 else if (vp->v_type == VDIR) { 478 char* bp; 479 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 480 aiov.iov_base = bp; 481 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 482 error = nfs_readdirrpc(vp, &auio, ap->a_cred); 483 free(bp, M_TEMP); 484 } else if (vp->v_type == VLNK) 485 error = nfs_readlinkrpc(vp, &auio, ap->a_cred); 486 else 487 error = EACCES; 488 } else 489 mtx_unlock(&np->n_mtx); 490 return (error); 491 } 492} 493 494int nfs_otw_getattr_avoid = 0; 495 496/* 497 * nfs open vnode op 498 * Check to see if the type is ok 499 * and that deletion is not in progress. 500 * For paged in text files, you will need to flush the page cache 501 * if consistency is lost. 502 */ 503/* ARGSUSED */ 504static int 505nfs_open(struct vop_open_args *ap) 506{ 507 struct vnode *vp = ap->a_vp; 508 struct nfsnode *np = VTONFS(vp); 509 struct vattr vattr; 510 int error; 511 int fmode = ap->a_mode; 512 struct ucred *cred; 513 514 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) 515 return (EOPNOTSUPP); 516 517 /* 518 * Get a valid lease. If cached data is stale, flush it. 519 */ 520 mtx_lock(&np->n_mtx); 521 if (np->n_flag & NMODIFIED) { 522 mtx_unlock(&np->n_mtx); 523 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 524 if (error == EINTR || error == EIO) 525 return (error); 526 mtx_lock(&np->n_mtx); 527 np->n_attrstamp = 0; 528 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 529 if (vp->v_type == VDIR) 530 np->n_direofoffset = 0; 531 mtx_unlock(&np->n_mtx); 532 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 533 if (error) 534 return (error); 535 mtx_lock(&np->n_mtx); 536 np->n_mtime = vattr.va_mtime; 537 } else { 538 mtx_unlock(&np->n_mtx); 539 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 540 if (error) 541 return (error); 542 mtx_lock(&np->n_mtx); 543 if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 544 if (vp->v_type == VDIR) 545 np->n_direofoffset = 0; 546 mtx_unlock(&np->n_mtx); 547 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 548 if (error == EINTR || error == EIO) { 549 return (error); 550 } 551 mtx_lock(&np->n_mtx); 552 np->n_mtime = vattr.va_mtime; 553 } 554 } 555 /* 556 * If the object has >= 1 O_DIRECT active opens, we disable caching. 557 */ 558 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) { 559 if (np->n_directio_opens == 0) { 560 mtx_unlock(&np->n_mtx); 561 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 562 if (error) 563 return (error); 564 mtx_lock(&np->n_mtx); 565 np->n_flag |= NNONCACHE; 566 } 567 np->n_directio_opens++; 568 } 569 570 /* 571 * If this is an open for writing, capture a reference to the 572 * credentials, so they can be used by nfs_putpages(). Using 573 * these write credentials is preferable to the credentials of 574 * whatever thread happens to be doing the VOP_PUTPAGES() since 575 * the write RPCs are less likely to fail with EACCES. 576 */ 577 if ((fmode & FWRITE) != 0) { 578 cred = np->n_writecred; 579 np->n_writecred = crhold(ap->a_cred); 580 } else 581 cred = NULL; 582 mtx_unlock(&np->n_mtx); 583 if (cred != NULL) 584 crfree(cred); 585 vnode_create_vobject(vp, vattr.va_size, ap->a_td); 586 return (0); 587} 588 589/* 590 * nfs close vnode op 591 * What an NFS client should do upon close after writing is a debatable issue. 592 * Most NFS clients push delayed writes to the server upon close, basically for 593 * two reasons: 594 * 1 - So that any write errors may be reported back to the client process 595 * doing the close system call. By far the two most likely errors are 596 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 597 * 2 - To put a worst case upper bound on cache inconsistency between 598 * multiple clients for the file. 599 * There is also a consistency problem for Version 2 of the protocol w.r.t. 600 * not being able to tell if other clients are writing a file concurrently, 601 * since there is no way of knowing if the changed modify time in the reply 602 * is only due to the write for this client. 603 * (NFS Version 3 provides weak cache consistency data in the reply that 604 * should be sufficient to detect and handle this case.) 605 * 606 * The current code does the following: 607 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 608 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 609 * or commit them (this satisfies 1 and 2 except for the 610 * case where the server crashes after this close but 611 * before the commit RPC, which is felt to be "good 612 * enough". Changing the last argument to nfs_flush() to 613 * a 1 would force a commit operation, if it is felt a 614 * commit is necessary now. 615 */ 616/* ARGSUSED */ 617static int 618nfs_close(struct vop_close_args *ap) 619{ 620 struct vnode *vp = ap->a_vp; 621 struct nfsnode *np = VTONFS(vp); 622 int error = 0; 623 int fmode = ap->a_fflag; 624 625 if (vp->v_type == VREG) { 626 /* 627 * Examine and clean dirty pages, regardless of NMODIFIED. 628 * This closes a major hole in close-to-open consistency. 629 * We want to push out all dirty pages (and buffers) on 630 * close, regardless of whether they were dirtied by 631 * mmap'ed writes or via write(). 632 */ 633 if (nfs_clean_pages_on_close && vp->v_object) { 634 VM_OBJECT_LOCK(vp->v_object); 635 vm_object_page_clean(vp->v_object, 0, 0, 0); 636 VM_OBJECT_UNLOCK(vp->v_object); 637 } 638 mtx_lock(&np->n_mtx); 639 if (np->n_flag & NMODIFIED) { 640 mtx_unlock(&np->n_mtx); 641 if (NFS_ISV3(vp)) { 642 /* 643 * Under NFSv3 we have dirty buffers to dispose of. We 644 * must flush them to the NFS server. We have the option 645 * of waiting all the way through the commit rpc or just 646 * waiting for the initial write. The default is to only 647 * wait through the initial write so the data is in the 648 * server's cache, which is roughly similar to the state 649 * a standard disk subsystem leaves the file in on close(). 650 * 651 * We cannot clear the NMODIFIED bit in np->n_flag due to 652 * potential races with other processes, and certainly 653 * cannot clear it if we don't commit. 654 */ 655 int cm = nfsv3_commit_on_close ? 1 : 0; 656 error = nfs_flush(vp, MNT_WAIT, cm); 657 /* np->n_flag &= ~NMODIFIED; */ 658 } else 659 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 660 mtx_lock(&np->n_mtx); 661 } 662 if (np->n_flag & NWRITEERR) { 663 np->n_flag &= ~NWRITEERR; 664 error = np->n_error; 665 } 666 mtx_unlock(&np->n_mtx); 667 } 668 if (nfs_directio_enable) 669 KASSERT((np->n_directio_asyncwr == 0), 670 ("nfs_close: dirty unflushed (%d) directio buffers\n", 671 np->n_directio_asyncwr)); 672 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) { 673 mtx_lock(&np->n_mtx); 674 KASSERT((np->n_directio_opens > 0), 675 ("nfs_close: unexpectedly value (0) of n_directio_opens\n")); 676 np->n_directio_opens--; 677 if (np->n_directio_opens == 0) 678 np->n_flag &= ~NNONCACHE; 679 mtx_unlock(&np->n_mtx); 680 } 681 return (error); 682} 683 684/* 685 * nfs getattr call from vfs. 686 */ 687static int 688nfs_getattr(struct vop_getattr_args *ap) 689{ 690 struct vnode *vp = ap->a_vp; 691 struct nfsnode *np = VTONFS(vp); 692 struct thread *td = curthread; 693 struct vattr *vap = ap->a_vap; 694 struct vattr vattr; 695 caddr_t bpos, dpos; 696 int error = 0; 697 struct mbuf *mreq, *mrep, *md, *mb; 698 int v3 = NFS_ISV3(vp); 699 700 /* 701 * Update local times for special files. 702 */ 703 mtx_lock(&np->n_mtx); 704 if (np->n_flag & (NACC | NUPD)) 705 np->n_flag |= NCHG; 706 mtx_unlock(&np->n_mtx); 707 /* 708 * First look in the cache. 709 */ 710 if (nfs_getattrcache(vp, &vattr) == 0) 711 goto nfsmout; 712 if (v3 && nfs_prime_access_cache && nfsaccess_cache_timeout > 0) { 713 nfsstats.accesscache_misses++; 714 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, ap->a_cred, NULL); 715 if (nfs_getattrcache(vp, &vattr) == 0) 716 goto nfsmout; 717 } 718 nfsstats.rpccnt[NFSPROC_GETATTR]++; 719 mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3)); 720 mb = mreq; 721 bpos = mtod(mb, caddr_t); 722 nfsm_fhtom(vp, v3); 723 nfsm_request(vp, NFSPROC_GETATTR, td, ap->a_cred); 724 if (!error) { 725 nfsm_loadattr(vp, &vattr); 726 } 727 m_freem(mrep); 728nfsmout: 729 vap->va_type = vattr.va_type; 730 vap->va_mode = vattr.va_mode; 731 vap->va_nlink = vattr.va_nlink; 732 vap->va_uid = vattr.va_uid; 733 vap->va_gid = vattr.va_gid; 734 vap->va_fsid = vattr.va_fsid; 735 vap->va_fileid = vattr.va_fileid; 736 vap->va_size = vattr.va_size; 737 vap->va_blocksize = vattr.va_blocksize; 738 vap->va_atime = vattr.va_atime; 739 vap->va_mtime = vattr.va_mtime; 740 vap->va_ctime = vattr.va_ctime; 741 vap->va_gen = vattr.va_gen; 742 vap->va_flags = vattr.va_flags; 743 vap->va_rdev = vattr.va_rdev; 744 vap->va_bytes = vattr.va_bytes; 745 vap->va_filerev = vattr.va_filerev; 746 747 return (error); 748} 749 750/* 751 * nfs setattr call. 752 */ 753static int 754nfs_setattr(struct vop_setattr_args *ap) 755{ 756 struct vnode *vp = ap->a_vp; 757 struct nfsnode *np = VTONFS(vp); 758 struct vattr *vap = ap->a_vap; 759 struct thread *td = curthread; 760 int error = 0; 761 u_quad_t tsize; 762 763#ifndef nolint 764 tsize = (u_quad_t)0; 765#endif 766 767 /* 768 * Setting of flags is not supported. 769 */ 770 if (vap->va_flags != VNOVAL) 771 return (EOPNOTSUPP); 772 773 /* 774 * Disallow write attempts if the filesystem is mounted read-only. 775 */ 776 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 777 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 778 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 779 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 780 error = EROFS; 781 goto out; 782 } 783 if (vap->va_size != VNOVAL) { 784 switch (vp->v_type) { 785 case VDIR: 786 return (EISDIR); 787 case VCHR: 788 case VBLK: 789 case VSOCK: 790 case VFIFO: 791 if (vap->va_mtime.tv_sec == VNOVAL && 792 vap->va_atime.tv_sec == VNOVAL && 793 vap->va_mode == (mode_t)VNOVAL && 794 vap->va_uid == (uid_t)VNOVAL && 795 vap->va_gid == (gid_t)VNOVAL) 796 return (0); 797 vap->va_size = VNOVAL; 798 break; 799 default: 800 /* 801 * Disallow write attempts if the filesystem is 802 * mounted read-only. 803 */ 804 if (vp->v_mount->mnt_flag & MNT_RDONLY) 805 return (EROFS); 806 /* 807 * We run vnode_pager_setsize() early (why?), 808 * we must set np->n_size now to avoid vinvalbuf 809 * V_SAVE races that might setsize a lower 810 * value. 811 */ 812 mtx_lock(&np->n_mtx); 813 tsize = np->n_size; 814 mtx_unlock(&np->n_mtx); 815 error = nfs_meta_setsize(vp, ap->a_cred, td, 816 vap->va_size); 817 mtx_lock(&np->n_mtx); 818 if (np->n_flag & NMODIFIED) { 819 tsize = np->n_size; 820 mtx_unlock(&np->n_mtx); 821 if (vap->va_size == 0) 822 error = nfs_vinvalbuf(vp, 0, td, 1); 823 else 824 error = nfs_vinvalbuf(vp, V_SAVE, td, 1); 825 if (error) { 826 vnode_pager_setsize(vp, tsize); 827 goto out; 828 } 829 } else 830 mtx_unlock(&np->n_mtx); 831 /* 832 * np->n_size has already been set to vap->va_size 833 * in nfs_meta_setsize(). We must set it again since 834 * nfs_loadattrcache() could be called through 835 * nfs_meta_setsize() and could modify np->n_size. 836 */ 837 mtx_lock(&np->n_mtx); 838 np->n_vattr.va_size = np->n_size = vap->va_size; 839 mtx_unlock(&np->n_mtx); 840 }; 841 } else { 842 mtx_lock(&np->n_mtx); 843 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) && 844 (np->n_flag & NMODIFIED) && vp->v_type == VREG) { 845 mtx_unlock(&np->n_mtx); 846 if ((error = nfs_vinvalbuf(vp, V_SAVE, td, 1)) != 0 && 847 (error == EINTR || error == EIO)) 848 return error; 849 } else 850 mtx_unlock(&np->n_mtx); 851 } 852 error = nfs_setattrrpc(vp, vap, ap->a_cred); 853 if (error && vap->va_size != VNOVAL) { 854 mtx_lock(&np->n_mtx); 855 np->n_size = np->n_vattr.va_size = tsize; 856 vnode_pager_setsize(vp, tsize); 857 mtx_unlock(&np->n_mtx); 858 } 859out: 860 return (error); 861} 862 863/* 864 * Do an nfs setattr rpc. 865 */ 866static int 867nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred) 868{ 869 struct nfsv2_sattr *sp; 870 struct nfsnode *np = VTONFS(vp); 871 caddr_t bpos, dpos; 872 u_int32_t *tl; 873 int error = 0, i, wccflag = NFSV3_WCCRATTR; 874 struct mbuf *mreq, *mrep, *md, *mb; 875 int v3 = NFS_ISV3(vp); 876 877 nfsstats.rpccnt[NFSPROC_SETATTR]++; 878 mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 879 mb = mreq; 880 bpos = mtod(mb, caddr_t); 881 nfsm_fhtom(vp, v3); 882 if (v3) { 883 nfsm_v3attrbuild(vap, TRUE); 884 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED); 885 *tl = nfs_false; 886 } else { 887 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR); 888 if (vap->va_mode == (mode_t)VNOVAL) 889 sp->sa_mode = nfs_xdrneg1; 890 else 891 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 892 if (vap->va_uid == (uid_t)VNOVAL) 893 sp->sa_uid = nfs_xdrneg1; 894 else 895 sp->sa_uid = txdr_unsigned(vap->va_uid); 896 if (vap->va_gid == (gid_t)VNOVAL) 897 sp->sa_gid = nfs_xdrneg1; 898 else 899 sp->sa_gid = txdr_unsigned(vap->va_gid); 900 sp->sa_size = txdr_unsigned(vap->va_size); 901 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 902 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 903 } 904 nfsm_request(vp, NFSPROC_SETATTR, curthread, cred); 905 if (v3) { 906 mtx_lock(&np->n_mtx); 907 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) 908 np->n_accesscache[i].stamp = 0; 909 mtx_unlock(&np->n_mtx); 910 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp); 911 nfsm_wcc_data(vp, wccflag); 912 } else 913 nfsm_loadattr(vp, NULL); 914 m_freem(mrep); 915nfsmout: 916 return (error); 917} 918 919/* 920 * nfs lookup call, one step at a time... 921 * First look in cache 922 * If not found, unlock the directory nfsnode and do the rpc 923 */ 924static int 925nfs_lookup(struct vop_lookup_args *ap) 926{ 927 struct componentname *cnp = ap->a_cnp; 928 struct vnode *dvp = ap->a_dvp; 929 struct vnode **vpp = ap->a_vpp; 930 struct mount *mp = dvp->v_mount; 931 struct vattr dvattr, vattr; 932 struct timespec nctime; 933 int flags = cnp->cn_flags; 934 struct vnode *newvp; 935 struct nfsmount *nmp; 936 caddr_t bpos, dpos; 937 struct mbuf *mreq, *mrep, *md, *mb; 938 long len; 939 nfsfh_t *fhp; 940 struct nfsnode *np, *newnp; 941 int error = 0, attrflag, dattrflag, fhsize, ltype, ncticks; 942 int v3 = NFS_ISV3(dvp); 943 struct thread *td = cnp->cn_thread; 944 945 *vpp = NULLVP; 946 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) && 947 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 948 return (EROFS); 949 if (dvp->v_type != VDIR) 950 return (ENOTDIR); 951 nmp = VFSTONFS(mp); 952 np = VTONFS(dvp); 953 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) { 954 *vpp = NULLVP; 955 return (error); 956 } 957 error = cache_lookup_times(dvp, vpp, cnp, &nctime, &ncticks); 958 if (error > 0 && error != ENOENT) 959 return (error); 960 if (error == -1) { 961 /* 962 * Lookups of "." are special and always return the 963 * current directory. cache_lookup() already handles 964 * associated locking bookkeeping, etc. 965 */ 966 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 967 /* XXX: Is this really correct? */ 968 if (cnp->cn_nameiop != LOOKUP && 969 (flags & ISLASTCN)) 970 cnp->cn_flags |= SAVENAME; 971 return (0); 972 } 973 974 /* 975 * We only accept a positive hit in the cache if the 976 * change time of the file matches our cached copy. 977 * Otherwise, we discard the cache entry and fallback 978 * to doing a lookup RPC. We also only trust cache 979 * entries for less than nm_nametimeo seconds. 980 * 981 * To better handle stale file handles and attributes, 982 * clear the attribute cache of this node if it is a 983 * leaf component, part of an open() call, and not 984 * locally modified before fetching the attributes. 985 * This should allow stale file handles to be detected 986 * here where we can fall back to a LOOKUP RPC to 987 * recover rather than having nfs_open() detect the 988 * stale file handle and failing open(2) with ESTALE. 989 */ 990 newvp = *vpp; 991 newnp = VTONFS(newvp); 992 if (!(nmp->nm_flag & NFSMNT_NOCTO) && 993 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 994 !(newnp->n_flag & NMODIFIED)) { 995 mtx_lock(&newnp->n_mtx); 996 newnp->n_attrstamp = 0; 997 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 998 mtx_unlock(&newnp->n_mtx); 999 } 1000 if ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) && 1001 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 && 1002 timespeccmp(&vattr.va_ctime, &nctime, ==)) { 1003 nfsstats.lookupcache_hits++; 1004 if (cnp->cn_nameiop != LOOKUP && 1005 (flags & ISLASTCN)) 1006 cnp->cn_flags |= SAVENAME; 1007 return (0); 1008 } 1009 cache_purge(newvp); 1010 if (dvp != newvp) 1011 vput(newvp); 1012 else 1013 vrele(newvp); 1014 *vpp = NULLVP; 1015 } else if (error == ENOENT) { 1016 if (dvp->v_iflag & VI_DOOMED) 1017 return (ENOENT); 1018 /* 1019 * We only accept a negative hit in the cache if the 1020 * modification time of the parent directory matches 1021 * the cached copy in the name cache entry. 1022 * Otherwise, we discard all of the negative cache 1023 * entries for this directory. We also only trust 1024 * negative cache entries for up to nm_negnametimeo 1025 * seconds. 1026 */ 1027 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) && 1028 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 && 1029 timespeccmp(&vattr.va_mtime, &nctime, ==)) { 1030 nfsstats.lookupcache_hits++; 1031 return (ENOENT); 1032 } 1033 cache_purge_negative(dvp); 1034 } 1035 1036 attrflag = dattrflag = 0; 1037 error = 0; 1038 newvp = NULLVP; 1039 nfsstats.lookupcache_misses++; 1040 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 1041 len = cnp->cn_namelen; 1042 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP, 1043 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 1044 mb = mreq; 1045 bpos = mtod(mb, caddr_t); 1046 nfsm_fhtom(dvp, v3); 1047 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 1048 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred); 1049 if (error) { 1050 if (v3) { 1051 nfsm_postop_attr_va(dvp, dattrflag, &vattr); 1052 m_freem(mrep); 1053 } 1054 goto nfsmout; 1055 } 1056 nfsm_getfh(fhp, fhsize, v3); 1057 1058 /* 1059 * Handle RENAME case... 1060 */ 1061 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 1062 if (NFS_CMPFH(np, fhp, fhsize)) { 1063 m_freem(mrep); 1064 return (EISDIR); 1065 } 1066 error = nfs_nget(mp, fhp, fhsize, &np, LK_EXCLUSIVE); 1067 if (error) { 1068 m_freem(mrep); 1069 return (error); 1070 } 1071 newvp = NFSTOV(np); 1072 if (v3) { 1073 nfsm_postop_attr(newvp, attrflag); 1074 nfsm_postop_attr(dvp, attrflag); 1075 } else 1076 nfsm_loadattr(newvp, NULL); 1077 *vpp = newvp; 1078 m_freem(mrep); 1079 cnp->cn_flags |= SAVENAME; 1080 return (0); 1081 } 1082 1083 if (flags & ISDOTDOT) { 1084 ltype = VOP_ISLOCKED(dvp); 1085 error = vfs_busy(mp, MBF_NOWAIT); 1086 if (error != 0) { 1087 vfs_ref(mp); 1088 VOP_UNLOCK(dvp, 0); 1089 error = vfs_busy(mp, 0); 1090 vn_lock(dvp, ltype | LK_RETRY); 1091 vfs_rel(mp); 1092 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) { 1093 vfs_unbusy(mp); 1094 error = ENOENT; 1095 } 1096 if (error != 0) { 1097 m_freem(mrep); 1098 return (error); 1099 } 1100 } 1101 VOP_UNLOCK(dvp, 0); 1102 error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags); 1103 if (error == 0) 1104 newvp = NFSTOV(np); 1105 vfs_unbusy(mp); 1106 if (newvp != dvp) 1107 vn_lock(dvp, ltype | LK_RETRY); 1108 if (dvp->v_iflag & VI_DOOMED) { 1109 if (error == 0) { 1110 if (newvp == dvp) 1111 vrele(newvp); 1112 else 1113 vput(newvp); 1114 } 1115 error = ENOENT; 1116 } 1117 if (error) { 1118 m_freem(mrep); 1119 return (error); 1120 } 1121 } else if (NFS_CMPFH(np, fhp, fhsize)) { 1122 VREF(dvp); 1123 newvp = dvp; 1124 } else { 1125 error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags); 1126 if (error) { 1127 m_freem(mrep); 1128 return (error); 1129 } 1130 newvp = NFSTOV(np); 1131 1132 /* 1133 * Flush the attribute cache when opening a leaf node 1134 * to ensure that fresh attributes are fetched in 1135 * nfs_open() if we are unable to fetch attributes 1136 * from the LOOKUP reply. 1137 */ 1138 if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1139 !(np->n_flag & NMODIFIED)) { 1140 mtx_lock(&np->n_mtx); 1141 np->n_attrstamp = 0; 1142 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1143 mtx_unlock(&np->n_mtx); 1144 } 1145 } 1146 if (v3) { 1147 nfsm_postop_attr_va(newvp, attrflag, &vattr); 1148 nfsm_postop_attr_va(dvp, dattrflag, &dvattr); 1149 } else { 1150 nfsm_loadattr(newvp, &vattr); 1151 attrflag = 1; 1152 } 1153 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1154 cnp->cn_flags |= SAVENAME; 1155 if ((cnp->cn_flags & MAKEENTRY) && 1156 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) && 1157 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0)) 1158 cache_enter_time(dvp, newvp, cnp, &vattr.va_ctime, 1159 newvp->v_type != VDIR ? NULL : &dvattr.va_ctime); 1160 *vpp = newvp; 1161 m_freem(mrep); 1162nfsmout: 1163 if (error) { 1164 if (newvp != NULLVP) { 1165 vput(newvp); 1166 *vpp = NULLVP; 1167 } 1168 1169 if (error != ENOENT) 1170 goto done; 1171 1172 /* The requested file was not found. */ 1173 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1174 (flags & ISLASTCN)) { 1175 /* 1176 * XXX: UFS does a full VOP_ACCESS(dvp, 1177 * VWRITE) here instead of just checking 1178 * MNT_RDONLY. 1179 */ 1180 if (mp->mnt_flag & MNT_RDONLY) 1181 return (EROFS); 1182 cnp->cn_flags |= SAVENAME; 1183 return (EJUSTRETURN); 1184 } 1185 1186 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE && 1187 dattrflag) { 1188 /* 1189 * Cache the modification time of the parent 1190 * directory from the post-op attributes in 1191 * the name cache entry. The negative cache 1192 * entry will be ignored once the directory 1193 * has changed. Don't bother adding the entry 1194 * if the directory has already changed. 1195 */ 1196 mtx_lock(&np->n_mtx); 1197 if (timespeccmp(&np->n_vattr.va_mtime, 1198 &vattr.va_mtime, ==)) { 1199 mtx_unlock(&np->n_mtx); 1200 cache_enter_time(dvp, NULL, cnp, 1201 &vattr.va_mtime, NULL); 1202 } else 1203 mtx_unlock(&np->n_mtx); 1204 } 1205 return (ENOENT); 1206 } 1207done: 1208 return (error); 1209} 1210 1211/* 1212 * nfs read call. 1213 * Just call nfs_bioread() to do the work. 1214 */ 1215static int 1216nfs_read(struct vop_read_args *ap) 1217{ 1218 struct vnode *vp = ap->a_vp; 1219 1220 switch (vp->v_type) { 1221 case VREG: 1222 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred)); 1223 case VDIR: 1224 return (EISDIR); 1225 default: 1226 return (EOPNOTSUPP); 1227 } 1228} 1229 1230/* 1231 * nfs readlink call 1232 */ 1233static int 1234nfs_readlink(struct vop_readlink_args *ap) 1235{ 1236 struct vnode *vp = ap->a_vp; 1237 1238 if (vp->v_type != VLNK) 1239 return (EINVAL); 1240 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred)); 1241} 1242 1243/* 1244 * Do a readlink rpc. 1245 * Called by nfs_doio() from below the buffer cache. 1246 */ 1247int 1248nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1249{ 1250 caddr_t bpos, dpos; 1251 int error = 0, len, attrflag; 1252 struct mbuf *mreq, *mrep, *md, *mb; 1253 int v3 = NFS_ISV3(vp); 1254 1255 nfsstats.rpccnt[NFSPROC_READLINK]++; 1256 mreq = nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3)); 1257 mb = mreq; 1258 bpos = mtod(mb, caddr_t); 1259 nfsm_fhtom(vp, v3); 1260 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred); 1261 if (v3) 1262 nfsm_postop_attr(vp, attrflag); 1263 if (!error) { 1264 nfsm_strsiz(len, NFS_MAXPATHLEN); 1265 if (len == NFS_MAXPATHLEN) { 1266 struct nfsnode *np = VTONFS(vp); 1267 mtx_lock(&np->n_mtx); 1268 if (np->n_size && np->n_size < NFS_MAXPATHLEN) 1269 len = np->n_size; 1270 mtx_unlock(&np->n_mtx); 1271 } 1272 nfsm_mtouio(uiop, len); 1273 } 1274 m_freem(mrep); 1275nfsmout: 1276 return (error); 1277} 1278 1279/* 1280 * nfs read rpc call 1281 * Ditto above 1282 */ 1283int 1284nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1285{ 1286 u_int32_t *tl; 1287 caddr_t bpos, dpos; 1288 struct mbuf *mreq, *mrep, *md, *mb; 1289 struct nfsmount *nmp; 1290 off_t end; 1291 int error = 0, len, retlen, tsiz, eof, attrflag; 1292 int v3 = NFS_ISV3(vp); 1293 int rsize; 1294 1295#ifndef nolint 1296 eof = 0; 1297#endif 1298 nmp = VFSTONFS(vp->v_mount); 1299 tsiz = uiop->uio_resid; 1300 mtx_lock(&nmp->nm_mtx); 1301 end = uiop->uio_offset + tsiz; 1302 if (end > nmp->nm_maxfilesize || end < uiop->uio_offset) { 1303 mtx_unlock(&nmp->nm_mtx); 1304 return (EFBIG); 1305 } 1306 rsize = nmp->nm_rsize; 1307 mtx_unlock(&nmp->nm_mtx); 1308 while (tsiz > 0) { 1309 nfsstats.rpccnt[NFSPROC_READ]++; 1310 len = (tsiz > rsize) ? rsize : tsiz; 1311 mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1312 mb = mreq; 1313 bpos = mtod(mb, caddr_t); 1314 nfsm_fhtom(vp, v3); 1315 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3); 1316 if (v3) { 1317 txdr_hyper(uiop->uio_offset, tl); 1318 *(tl + 2) = txdr_unsigned(len); 1319 } else { 1320 *tl++ = txdr_unsigned(uiop->uio_offset); 1321 *tl++ = txdr_unsigned(len); 1322 *tl = 0; 1323 } 1324 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred); 1325 if (v3) { 1326 nfsm_postop_attr(vp, attrflag); 1327 if (error) { 1328 m_freem(mrep); 1329 goto nfsmout; 1330 } 1331 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED); 1332 eof = fxdr_unsigned(int, *(tl + 1)); 1333 } else { 1334 nfsm_loadattr(vp, NULL); 1335 } 1336 nfsm_strsiz(retlen, rsize); 1337 nfsm_mtouio(uiop, retlen); 1338 m_freem(mrep); 1339 tsiz -= retlen; 1340 if (v3) { 1341 if (eof || retlen == 0) { 1342 tsiz = 0; 1343 } 1344 } else if (retlen < len) { 1345 tsiz = 0; 1346 } 1347 } 1348nfsmout: 1349 return (error); 1350} 1351 1352/* 1353 * nfs write call 1354 */ 1355int 1356nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 1357 int *iomode, int *must_commit) 1358{ 1359 u_int32_t *tl; 1360 int32_t backup; 1361 caddr_t bpos, dpos; 1362 struct mbuf *mreq, *mrep, *md, *mb; 1363 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1364 off_t end; 1365 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit; 1366 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC; 1367 int wsize; 1368 1369 KASSERT(uiop->uio_iovcnt == 1, ("nfs: writerpc iovcnt > 1")); 1370 *must_commit = 0; 1371 tsiz = uiop->uio_resid; 1372 mtx_lock(&nmp->nm_mtx); 1373 end = uiop->uio_offset + tsiz; 1374 if (end > nmp->nm_maxfilesize || end < uiop->uio_offset) { 1375 mtx_unlock(&nmp->nm_mtx); 1376 return (EFBIG); 1377 } 1378 wsize = nmp->nm_wsize; 1379 mtx_unlock(&nmp->nm_mtx); 1380 while (tsiz > 0) { 1381 nfsstats.rpccnt[NFSPROC_WRITE]++; 1382 len = (tsiz > wsize) ? wsize : tsiz; 1383 mreq = nfsm_reqhead(vp, NFSPROC_WRITE, 1384 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len)); 1385 mb = mreq; 1386 bpos = mtod(mb, caddr_t); 1387 nfsm_fhtom(vp, v3); 1388 if (v3) { 1389 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED); 1390 txdr_hyper(uiop->uio_offset, tl); 1391 tl += 2; 1392 *tl++ = txdr_unsigned(len); 1393 *tl++ = txdr_unsigned(*iomode); 1394 *tl = txdr_unsigned(len); 1395 } else { 1396 u_int32_t x; 1397 1398 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED); 1399 /* Set both "begin" and "current" to non-garbage. */ 1400 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1401 *tl++ = x; /* "begin offset" */ 1402 *tl++ = x; /* "current offset" */ 1403 x = txdr_unsigned(len); 1404 *tl++ = x; /* total to this offset */ 1405 *tl = x; /* size of this write */ 1406 } 1407 nfsm_uiotom(uiop, len); 1408 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred); 1409 if (v3) { 1410 wccflag = NFSV3_WCCCHK; 1411 nfsm_wcc_data(vp, wccflag); 1412 if (!error) { 1413 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED 1414 + NFSX_V3WRITEVERF); 1415 rlen = fxdr_unsigned(int, *tl++); 1416 if (rlen == 0) { 1417 error = NFSERR_IO; 1418 m_freem(mrep); 1419 break; 1420 } else if (rlen < len) { 1421 backup = len - rlen; 1422 uiop->uio_iov->iov_base = 1423 (char *)uiop->uio_iov->iov_base - 1424 backup; 1425 uiop->uio_iov->iov_len += backup; 1426 uiop->uio_offset -= backup; 1427 uiop->uio_resid += backup; 1428 len = rlen; 1429 } 1430 commit = fxdr_unsigned(int, *tl++); 1431 1432 /* 1433 * Return the lowest committment level 1434 * obtained by any of the RPCs. 1435 */ 1436 if (committed == NFSV3WRITE_FILESYNC) 1437 committed = commit; 1438 else if (committed == NFSV3WRITE_DATASYNC && 1439 commit == NFSV3WRITE_UNSTABLE) 1440 committed = commit; 1441 mtx_lock(&nmp->nm_mtx); 1442 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){ 1443 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1444 NFSX_V3WRITEVERF); 1445 nmp->nm_state |= NFSSTA_HASWRITEVERF; 1446 } else if (bcmp((caddr_t)tl, 1447 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) { 1448 *must_commit = 1; 1449 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1450 NFSX_V3WRITEVERF); 1451 } 1452 mtx_unlock(&nmp->nm_mtx); 1453 } 1454 } else { 1455 nfsm_loadattr(vp, NULL); 1456 } 1457 if (wccflag) { 1458 mtx_lock(&(VTONFS(vp))->n_mtx); 1459 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime; 1460 mtx_unlock(&(VTONFS(vp))->n_mtx); 1461 } 1462 m_freem(mrep); 1463 if (error) 1464 break; 1465 tsiz -= len; 1466 } 1467nfsmout: 1468 if (DOINGASYNC(vp)) 1469 committed = NFSV3WRITE_FILESYNC; 1470 *iomode = committed; 1471 if (error) 1472 uiop->uio_resid = tsiz; 1473 return (error); 1474} 1475 1476/* 1477 * nfs mknod rpc 1478 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1479 * mode set to specify the file type and the size field for rdev. 1480 */ 1481static int 1482nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1483 struct vattr *vap) 1484{ 1485 struct nfsv2_sattr *sp; 1486 u_int32_t *tl; 1487 struct vnode *newvp = NULL; 1488 struct nfsnode *np = NULL; 1489 struct vattr vattr; 1490 caddr_t bpos, dpos; 1491 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1492 struct mbuf *mreq, *mrep, *md, *mb; 1493 u_int32_t rdev; 1494 int v3 = NFS_ISV3(dvp); 1495 1496 if (vap->va_type == VCHR || vap->va_type == VBLK) 1497 rdev = txdr_unsigned(vap->va_rdev); 1498 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1499 rdev = nfs_xdrneg1; 1500 else { 1501 return (EOPNOTSUPP); 1502 } 1503 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) 1504 return (error); 1505 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1506 mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1507 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1508 mb = mreq; 1509 bpos = mtod(mb, caddr_t); 1510 nfsm_fhtom(dvp, v3); 1511 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1512 if (v3) { 1513 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED); 1514 *tl++ = vtonfsv3_type(vap->va_type); 1515 nfsm_v3attrbuild(vap, FALSE); 1516 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1517 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED); 1518 *tl++ = txdr_unsigned(major(vap->va_rdev)); 1519 *tl = txdr_unsigned(minor(vap->va_rdev)); 1520 } 1521 } else { 1522 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR); 1523 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1524 sp->sa_uid = nfs_xdrneg1; 1525 sp->sa_gid = nfs_xdrneg1; 1526 sp->sa_size = rdev; 1527 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1528 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1529 } 1530 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred); 1531 if (!error) { 1532 nfsm_mtofh(dvp, newvp, v3, gotvp); 1533 if (!gotvp) { 1534 if (newvp) { 1535 vput(newvp); 1536 newvp = NULL; 1537 } 1538 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1539 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np); 1540 if (!error) 1541 newvp = NFSTOV(np); 1542 } 1543 } 1544 if (v3) 1545 nfsm_wcc_data(dvp, wccflag); 1546 m_freem(mrep); 1547nfsmout: 1548 if (error) { 1549 if (newvp) 1550 vput(newvp); 1551 } else { 1552 *vpp = newvp; 1553 } 1554 mtx_lock(&(VTONFS(dvp))->n_mtx); 1555 VTONFS(dvp)->n_flag |= NMODIFIED; 1556 if (!wccflag) { 1557 VTONFS(dvp)->n_attrstamp = 0; 1558 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1559 } 1560 mtx_unlock(&(VTONFS(dvp))->n_mtx); 1561 return (error); 1562} 1563 1564/* 1565 * nfs mknod vop 1566 * just call nfs_mknodrpc() to do the work. 1567 */ 1568/* ARGSUSED */ 1569static int 1570nfs_mknod(struct vop_mknod_args *ap) 1571{ 1572 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap)); 1573} 1574 1575static u_long create_verf; 1576/* 1577 * nfs file create call 1578 */ 1579static int 1580nfs_create(struct vop_create_args *ap) 1581{ 1582 struct vnode *dvp = ap->a_dvp; 1583 struct vattr *vap = ap->a_vap; 1584 struct componentname *cnp = ap->a_cnp; 1585 struct nfsv2_sattr *sp; 1586 u_int32_t *tl; 1587 struct nfsnode *np = NULL; 1588 struct vnode *newvp = NULL; 1589 caddr_t bpos, dpos; 1590 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0; 1591 struct mbuf *mreq, *mrep, *md, *mb; 1592 struct vattr vattr; 1593 int v3 = NFS_ISV3(dvp); 1594 1595 /* 1596 * Oops, not for me.. 1597 */ 1598 if (vap->va_type == VSOCK) { 1599 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap); 1600 return (error); 1601 } 1602 1603 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) { 1604 return (error); 1605 } 1606 if (vap->va_vaflags & VA_EXCLUSIVE) 1607 fmode |= O_EXCL; 1608again: 1609 nfsstats.rpccnt[NFSPROC_CREATE]++; 1610 mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1611 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1612 mb = mreq; 1613 bpos = mtod(mb, caddr_t); 1614 nfsm_fhtom(dvp, v3); 1615 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1616 if (v3) { 1617 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED); 1618 if (fmode & O_EXCL) { 1619 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1620 tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF); 1621#ifdef INET 1622 CURVNET_SET(CRED_TO_VNET(cnp->cn_cred)); 1623 IN_IFADDR_RLOCK(); 1624 if (!TAILQ_EMPTY(&V_in_ifaddrhead)) 1625 *tl++ = IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr.s_addr; 1626 else 1627#endif 1628 *tl++ = create_verf; 1629#ifdef INET 1630 IN_IFADDR_RUNLOCK(); 1631 CURVNET_RESTORE(); 1632#endif 1633 *tl = ++create_verf; 1634 } else { 1635 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED); 1636 nfsm_v3attrbuild(vap, FALSE); 1637 } 1638 } else { 1639 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR); 1640 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1641 sp->sa_uid = nfs_xdrneg1; 1642 sp->sa_gid = nfs_xdrneg1; 1643 sp->sa_size = 0; 1644 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1645 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1646 } 1647 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred); 1648 if (!error) { 1649 nfsm_mtofh(dvp, newvp, v3, gotvp); 1650 if (!gotvp) { 1651 if (newvp) { 1652 vput(newvp); 1653 newvp = NULL; 1654 } 1655 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1656 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np); 1657 if (!error) 1658 newvp = NFSTOV(np); 1659 } 1660 } 1661 if (v3) 1662 nfsm_wcc_data(dvp, wccflag); 1663 m_freem(mrep); 1664nfsmout: 1665 if (error) { 1666 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) { 1667 fmode &= ~O_EXCL; 1668 goto again; 1669 } 1670 if (newvp) 1671 vput(newvp); 1672 } else if (v3 && (fmode & O_EXCL)) { 1673 /* 1674 * We are normally called with only a partially initialized 1675 * VAP. Since the NFSv3 spec says that server may use the 1676 * file attributes to store the verifier, the spec requires 1677 * us to do a SETATTR RPC. FreeBSD servers store the verifier 1678 * in atime, but we can't really assume that all servers will 1679 * so we ensure that our SETATTR sets both atime and mtime. 1680 */ 1681 if (vap->va_mtime.tv_sec == VNOVAL) 1682 vfs_timestamp(&vap->va_mtime); 1683 if (vap->va_atime.tv_sec == VNOVAL) 1684 vap->va_atime = vap->va_mtime; 1685 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred); 1686 if (error) 1687 vput(newvp); 1688 } 1689 if (!error) { 1690 *ap->a_vpp = newvp; 1691 } 1692 mtx_lock(&(VTONFS(dvp))->n_mtx); 1693 VTONFS(dvp)->n_flag |= NMODIFIED; 1694 if (!wccflag) { 1695 VTONFS(dvp)->n_attrstamp = 0; 1696 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1697 } 1698 mtx_unlock(&(VTONFS(dvp))->n_mtx); 1699 return (error); 1700} 1701 1702/* 1703 * nfs file remove call 1704 * To try and make nfs semantics closer to ufs semantics, a file that has 1705 * other processes using the vnode is renamed instead of removed and then 1706 * removed later on the last close. 1707 * - If v_usecount > 1 1708 * If a rename is not already in the works 1709 * call nfs_sillyrename() to set it up 1710 * else 1711 * do the remove rpc 1712 */ 1713static int 1714nfs_remove(struct vop_remove_args *ap) 1715{ 1716 struct vnode *vp = ap->a_vp; 1717 struct vnode *dvp = ap->a_dvp; 1718 struct componentname *cnp = ap->a_cnp; 1719 struct nfsnode *np = VTONFS(vp); 1720 int error = 0; 1721 struct vattr vattr; 1722 1723 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name")); 1724 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount")); 1725 if (vp->v_type == VDIR) 1726 error = EPERM; 1727 else if (vrefcnt(vp) == 1 || (np->n_sillyrename && 1728 !VOP_GETATTR(vp, &vattr, cnp->cn_cred) && vattr.va_nlink > 1)) { 1729 /* 1730 * Purge the name cache so that the chance of a lookup for 1731 * the name succeeding while the remove is in progress is 1732 * minimized. Without node locking it can still happen, such 1733 * that an I/O op returns ESTALE, but since you get this if 1734 * another host removes the file.. 1735 */ 1736 cache_purge(vp); 1737 /* 1738 * throw away biocache buffers, mainly to avoid 1739 * unnecessary delayed writes later. 1740 */ 1741 error = nfs_vinvalbuf(vp, 0, cnp->cn_thread, 1); 1742 /* Do the rpc */ 1743 if (error != EINTR && error != EIO) 1744 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1745 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread); 1746 /* 1747 * Kludge City: If the first reply to the remove rpc is lost.. 1748 * the reply to the retransmitted request will be ENOENT 1749 * since the file was in fact removed 1750 * Therefore, we cheat and return success. 1751 */ 1752 if (error == ENOENT) 1753 error = 0; 1754 } else if (!np->n_sillyrename) 1755 error = nfs_sillyrename(dvp, vp, cnp); 1756 mtx_lock(&np->n_mtx); 1757 np->n_attrstamp = 0; 1758 mtx_unlock(&np->n_mtx); 1759 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 1760 return (error); 1761} 1762 1763/* 1764 * nfs file remove rpc called from nfs_inactive 1765 */ 1766int 1767nfs_removeit(struct sillyrename *sp) 1768{ 1769 /* 1770 * Make sure that the directory vnode is still valid. 1771 * XXX we should lock sp->s_dvp here. 1772 */ 1773 if (sp->s_dvp->v_type == VBAD) 1774 return (0); 1775 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred, 1776 NULL)); 1777} 1778 1779/* 1780 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1781 */ 1782static int 1783nfs_removerpc(struct vnode *dvp, const char *name, int namelen, 1784 struct ucred *cred, struct thread *td) 1785{ 1786 caddr_t bpos, dpos; 1787 int error = 0, wccflag = NFSV3_WCCRATTR; 1788 struct mbuf *mreq, *mrep, *md, *mb; 1789 int v3 = NFS_ISV3(dvp); 1790 1791 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1792 mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE, 1793 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1794 mb = mreq; 1795 bpos = mtod(mb, caddr_t); 1796 nfsm_fhtom(dvp, v3); 1797 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1798 nfsm_request(dvp, NFSPROC_REMOVE, td, cred); 1799 if (v3) 1800 nfsm_wcc_data(dvp, wccflag); 1801 m_freem(mrep); 1802nfsmout: 1803 mtx_lock(&(VTONFS(dvp))->n_mtx); 1804 VTONFS(dvp)->n_flag |= NMODIFIED; 1805 if (!wccflag) { 1806 VTONFS(dvp)->n_attrstamp = 0; 1807 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1808 } 1809 mtx_unlock(&(VTONFS(dvp))->n_mtx); 1810 return (error); 1811} 1812 1813/* 1814 * nfs file rename call 1815 */ 1816static int 1817nfs_rename(struct vop_rename_args *ap) 1818{ 1819 struct vnode *fvp = ap->a_fvp; 1820 struct vnode *tvp = ap->a_tvp; 1821 struct vnode *fdvp = ap->a_fdvp; 1822 struct vnode *tdvp = ap->a_tdvp; 1823 struct componentname *tcnp = ap->a_tcnp; 1824 struct componentname *fcnp = ap->a_fcnp; 1825 int error; 1826 1827 KASSERT((tcnp->cn_flags & HASBUF) != 0 && 1828 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name")); 1829 /* Check for cross-device rename */ 1830 if ((fvp->v_mount != tdvp->v_mount) || 1831 (tvp && (fvp->v_mount != tvp->v_mount))) { 1832 error = EXDEV; 1833 goto out; 1834 } 1835 1836 if (fvp == tvp) { 1837 nfs_printf("nfs_rename: fvp == tvp (can't happen)\n"); 1838 error = 0; 1839 goto out; 1840 } 1841 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0) 1842 goto out; 1843 1844 /* 1845 * We have to flush B_DELWRI data prior to renaming 1846 * the file. If we don't, the delayed-write buffers 1847 * can be flushed out later after the file has gone stale 1848 * under NFSV3. NFSV2 does not have this problem because 1849 * ( as far as I can tell ) it flushes dirty buffers more 1850 * often. 1851 * 1852 * Skip the rename operation if the fsync fails, this can happen 1853 * due to the server's volume being full, when we pushed out data 1854 * that was written back to our cache earlier. Not checking for 1855 * this condition can result in potential (silent) data loss. 1856 */ 1857 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread); 1858 VOP_UNLOCK(fvp, 0); 1859 if (!error && tvp) 1860 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread); 1861 if (error) 1862 goto out; 1863 1864 /* 1865 * If the tvp exists and is in use, sillyrename it before doing the 1866 * rename of the new file over it. 1867 * XXX Can't sillyrename a directory. 1868 */ 1869 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename && 1870 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1871 vput(tvp); 1872 tvp = NULL; 1873 } 1874 1875 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1876 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1877 tcnp->cn_thread); 1878 1879 if (fvp->v_type == VDIR) { 1880 if (tvp != NULL && tvp->v_type == VDIR) 1881 cache_purge(tdvp); 1882 cache_purge(fdvp); 1883 } 1884 1885out: 1886 if (tdvp == tvp) 1887 vrele(tdvp); 1888 else 1889 vput(tdvp); 1890 if (tvp) 1891 vput(tvp); 1892 vrele(fdvp); 1893 vrele(fvp); 1894 /* 1895 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1896 */ 1897 if (error == ENOENT) 1898 error = 0; 1899 return (error); 1900} 1901 1902/* 1903 * nfs file rename rpc called from nfs_remove() above 1904 */ 1905static int 1906nfs_renameit(struct vnode *sdvp, struct componentname *scnp, 1907 struct sillyrename *sp) 1908{ 1909 1910 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp, 1911 sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread)); 1912} 1913 1914/* 1915 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1916 */ 1917static int 1918nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen, 1919 struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred, 1920 struct thread *td) 1921{ 1922 caddr_t bpos, dpos; 1923 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1924 struct mbuf *mreq, *mrep, *md, *mb; 1925 int v3 = NFS_ISV3(fdvp); 1926 1927 nfsstats.rpccnt[NFSPROC_RENAME]++; 1928 mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME, 1929 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 1930 nfsm_rndup(tnamelen)); 1931 mb = mreq; 1932 bpos = mtod(mb, caddr_t); 1933 nfsm_fhtom(fdvp, v3); 1934 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 1935 nfsm_fhtom(tdvp, v3); 1936 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 1937 nfsm_request(fdvp, NFSPROC_RENAME, td, cred); 1938 if (v3) { 1939 nfsm_wcc_data(fdvp, fwccflag); 1940 nfsm_wcc_data(tdvp, twccflag); 1941 } 1942 m_freem(mrep); 1943nfsmout: 1944 mtx_lock(&(VTONFS(fdvp))->n_mtx); 1945 VTONFS(fdvp)->n_flag |= NMODIFIED; 1946 mtx_unlock(&(VTONFS(fdvp))->n_mtx); 1947 mtx_lock(&(VTONFS(tdvp))->n_mtx); 1948 VTONFS(tdvp)->n_flag |= NMODIFIED; 1949 mtx_unlock(&(VTONFS(tdvp))->n_mtx); 1950 if (!fwccflag) { 1951 VTONFS(fdvp)->n_attrstamp = 0; 1952 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp); 1953 } 1954 if (!twccflag) { 1955 VTONFS(tdvp)->n_attrstamp = 0; 1956 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 1957 } 1958 return (error); 1959} 1960 1961/* 1962 * nfs hard link create call 1963 */ 1964static int 1965nfs_link(struct vop_link_args *ap) 1966{ 1967 struct vnode *vp = ap->a_vp; 1968 struct vnode *tdvp = ap->a_tdvp; 1969 struct componentname *cnp = ap->a_cnp; 1970 caddr_t bpos, dpos; 1971 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 1972 struct mbuf *mreq, *mrep, *md, *mb; 1973 int v3; 1974 1975 if (vp->v_mount != tdvp->v_mount) { 1976 return (EXDEV); 1977 } 1978 1979 /* 1980 * Push all writes to the server, so that the attribute cache 1981 * doesn't get "out of sync" with the server. 1982 * XXX There should be a better way! 1983 */ 1984 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread); 1985 1986 v3 = NFS_ISV3(vp); 1987 nfsstats.rpccnt[NFSPROC_LINK]++; 1988 mreq = nfsm_reqhead(vp, NFSPROC_LINK, 1989 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 1990 mb = mreq; 1991 bpos = mtod(mb, caddr_t); 1992 nfsm_fhtom(vp, v3); 1993 nfsm_fhtom(tdvp, v3); 1994 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1995 nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred); 1996 if (v3) { 1997 nfsm_postop_attr(vp, attrflag); 1998 nfsm_wcc_data(tdvp, wccflag); 1999 } 2000 m_freem(mrep); 2001nfsmout: 2002 mtx_lock(&(VTONFS(tdvp))->n_mtx); 2003 VTONFS(tdvp)->n_flag |= NMODIFIED; 2004 mtx_unlock(&(VTONFS(tdvp))->n_mtx); 2005 if (!attrflag) { 2006 VTONFS(vp)->n_attrstamp = 0; 2007 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 2008 } 2009 if (!wccflag) { 2010 VTONFS(tdvp)->n_attrstamp = 0; 2011 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 2012 } 2013 return (error); 2014} 2015 2016/* 2017 * nfs symbolic link create call 2018 */ 2019static int 2020nfs_symlink(struct vop_symlink_args *ap) 2021{ 2022 struct vnode *dvp = ap->a_dvp; 2023 struct vattr *vap = ap->a_vap; 2024 struct componentname *cnp = ap->a_cnp; 2025 struct nfsv2_sattr *sp; 2026 caddr_t bpos, dpos; 2027 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 2028 struct mbuf *mreq, *mrep, *md, *mb; 2029 struct vnode *newvp = NULL; 2030 int v3 = NFS_ISV3(dvp); 2031 2032 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 2033 slen = strlen(ap->a_target); 2034 mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 2035 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 2036 mb = mreq; 2037 bpos = mtod(mb, caddr_t); 2038 nfsm_fhtom(dvp, v3); 2039 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2040 if (v3) { 2041 nfsm_v3attrbuild(vap, FALSE); 2042 } 2043 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 2044 if (!v3) { 2045 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR); 2046 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 2047 sp->sa_uid = nfs_xdrneg1; 2048 sp->sa_gid = nfs_xdrneg1; 2049 sp->sa_size = nfs_xdrneg1; 2050 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2051 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2052 } 2053 2054 /* 2055 * Issue the NFS request and get the rpc response. 2056 * 2057 * Only NFSv3 responses returning an error of 0 actually return 2058 * a file handle that can be converted into newvp without having 2059 * to do an extra lookup rpc. 2060 */ 2061 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred); 2062 if (v3) { 2063 if (error == 0) 2064 nfsm_mtofh(dvp, newvp, v3, gotvp); 2065 nfsm_wcc_data(dvp, wccflag); 2066 } 2067 2068 /* 2069 * out code jumps -> here, mrep is also freed. 2070 */ 2071 2072 m_freem(mrep); 2073nfsmout: 2074 2075 /* 2076 * If we do not have an error and we could not extract the newvp from 2077 * the response due to the request being NFSv2, we have to do a 2078 * lookup in order to obtain a newvp to return. 2079 */ 2080 if (error == 0 && newvp == NULL) { 2081 struct nfsnode *np = NULL; 2082 2083 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2084 cnp->cn_cred, cnp->cn_thread, &np); 2085 if (!error) 2086 newvp = NFSTOV(np); 2087 } 2088 if (error) { 2089 if (newvp) 2090 vput(newvp); 2091 } else { 2092 *ap->a_vpp = newvp; 2093 } 2094 mtx_lock(&(VTONFS(dvp))->n_mtx); 2095 VTONFS(dvp)->n_flag |= NMODIFIED; 2096 mtx_unlock(&(VTONFS(dvp))->n_mtx); 2097 if (!wccflag) { 2098 VTONFS(dvp)->n_attrstamp = 0; 2099 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2100 } 2101 return (error); 2102} 2103 2104/* 2105 * nfs make dir call 2106 */ 2107static int 2108nfs_mkdir(struct vop_mkdir_args *ap) 2109{ 2110 struct vnode *dvp = ap->a_dvp; 2111 struct vattr *vap = ap->a_vap; 2112 struct componentname *cnp = ap->a_cnp; 2113 struct nfsv2_sattr *sp; 2114 int len; 2115 struct nfsnode *np = NULL; 2116 struct vnode *newvp = NULL; 2117 caddr_t bpos, dpos; 2118 int error = 0, wccflag = NFSV3_WCCRATTR; 2119 int gotvp = 0; 2120 struct mbuf *mreq, *mrep, *md, *mb; 2121 struct vattr vattr; 2122 int v3 = NFS_ISV3(dvp); 2123 2124 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) 2125 return (error); 2126 len = cnp->cn_namelen; 2127 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2128 mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR, 2129 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 2130 mb = mreq; 2131 bpos = mtod(mb, caddr_t); 2132 nfsm_fhtom(dvp, v3); 2133 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 2134 if (v3) { 2135 nfsm_v3attrbuild(vap, FALSE); 2136 } else { 2137 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR); 2138 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2139 sp->sa_uid = nfs_xdrneg1; 2140 sp->sa_gid = nfs_xdrneg1; 2141 sp->sa_size = nfs_xdrneg1; 2142 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2143 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2144 } 2145 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred); 2146 if (!error) 2147 nfsm_mtofh(dvp, newvp, v3, gotvp); 2148 if (v3) 2149 nfsm_wcc_data(dvp, wccflag); 2150 m_freem(mrep); 2151nfsmout: 2152 mtx_lock(&(VTONFS(dvp))->n_mtx); 2153 VTONFS(dvp)->n_flag |= NMODIFIED; 2154 mtx_unlock(&(VTONFS(dvp))->n_mtx); 2155 if (!wccflag) { 2156 VTONFS(dvp)->n_attrstamp = 0; 2157 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2158 } 2159 if (error == 0 && newvp == NULL) { 2160 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2161 cnp->cn_thread, &np); 2162 if (!error) { 2163 newvp = NFSTOV(np); 2164 if (newvp->v_type != VDIR) 2165 error = EEXIST; 2166 } 2167 } 2168 if (error) { 2169 if (newvp) 2170 vput(newvp); 2171 } else 2172 *ap->a_vpp = newvp; 2173 return (error); 2174} 2175 2176/* 2177 * nfs remove directory call 2178 */ 2179static int 2180nfs_rmdir(struct vop_rmdir_args *ap) 2181{ 2182 struct vnode *vp = ap->a_vp; 2183 struct vnode *dvp = ap->a_dvp; 2184 struct componentname *cnp = ap->a_cnp; 2185 caddr_t bpos, dpos; 2186 int error = 0, wccflag = NFSV3_WCCRATTR; 2187 struct mbuf *mreq, *mrep, *md, *mb; 2188 int v3 = NFS_ISV3(dvp); 2189 2190 if (dvp == vp) 2191 return (EINVAL); 2192 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2193 mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR, 2194 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2195 mb = mreq; 2196 bpos = mtod(mb, caddr_t); 2197 nfsm_fhtom(dvp, v3); 2198 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2199 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred); 2200 if (v3) 2201 nfsm_wcc_data(dvp, wccflag); 2202 m_freem(mrep); 2203nfsmout: 2204 mtx_lock(&(VTONFS(dvp))->n_mtx); 2205 VTONFS(dvp)->n_flag |= NMODIFIED; 2206 mtx_unlock(&(VTONFS(dvp))->n_mtx); 2207 if (!wccflag) { 2208 VTONFS(dvp)->n_attrstamp = 0; 2209 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2210 } 2211 cache_purge(dvp); 2212 cache_purge(vp); 2213 /* 2214 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2215 */ 2216 if (error == ENOENT) 2217 error = 0; 2218 return (error); 2219} 2220 2221/* 2222 * nfs readdir call 2223 */ 2224static int 2225nfs_readdir(struct vop_readdir_args *ap) 2226{ 2227 struct vnode *vp = ap->a_vp; 2228 struct nfsnode *np = VTONFS(vp); 2229 struct uio *uio = ap->a_uio; 2230 int tresid, error = 0; 2231 struct vattr vattr; 2232 2233 if (vp->v_type != VDIR) 2234 return(EPERM); 2235 2236 /* 2237 * First, check for hit on the EOF offset cache 2238 */ 2239 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2240 (np->n_flag & NMODIFIED) == 0) { 2241 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) { 2242 mtx_lock(&np->n_mtx); 2243 if (!NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 2244 mtx_unlock(&np->n_mtx); 2245 nfsstats.direofcache_hits++; 2246 goto out; 2247 } else 2248 mtx_unlock(&np->n_mtx); 2249 } 2250 } 2251 2252 /* 2253 * Call nfs_bioread() to do the real work. 2254 */ 2255 tresid = uio->uio_resid; 2256 error = nfs_bioread(vp, uio, 0, ap->a_cred); 2257 2258 if (!error && uio->uio_resid == tresid) { 2259 nfsstats.direofcache_misses++; 2260 } 2261out: 2262 return (error); 2263} 2264 2265/* 2266 * Readdir rpc call. 2267 * Called from below the buffer cache by nfs_doio(). 2268 */ 2269int 2270nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 2271{ 2272 int len, left; 2273 struct dirent *dp = NULL; 2274 u_int32_t *tl; 2275 caddr_t cp; 2276 nfsuint64 *cookiep; 2277 caddr_t bpos, dpos; 2278 struct mbuf *mreq, *mrep, *md, *mb; 2279 nfsuint64 cookie; 2280 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2281 struct nfsnode *dnp = VTONFS(vp); 2282 u_quad_t fileno; 2283 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1; 2284 int attrflag; 2285 int v3 = NFS_ISV3(vp); 2286 2287 KASSERT(uiop->uio_iovcnt == 1 && 2288 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2289 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2290 ("nfs readdirrpc bad uio")); 2291 2292 /* 2293 * If there is no cookie, assume directory was stale. 2294 */ 2295 nfs_dircookie_lock(dnp); 2296 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2297 if (cookiep) { 2298 cookie = *cookiep; 2299 nfs_dircookie_unlock(dnp); 2300 } else { 2301 nfs_dircookie_unlock(dnp); 2302 return (NFSERR_BAD_COOKIE); 2303 } 2304 2305 /* 2306 * Loop around doing readdir rpc's of size nm_readdirsize 2307 * truncated to a multiple of DIRBLKSIZ. 2308 * The stopping criteria is EOF or buffer full. 2309 */ 2310 while (more_dirs && bigenough) { 2311 nfsstats.rpccnt[NFSPROC_READDIR]++; 2312 mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) + 2313 NFSX_READDIR(v3)); 2314 mb = mreq; 2315 bpos = mtod(mb, caddr_t); 2316 nfsm_fhtom(vp, v3); 2317 if (v3) { 2318 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED); 2319 *tl++ = cookie.nfsuquad[0]; 2320 *tl++ = cookie.nfsuquad[1]; 2321 mtx_lock(&dnp->n_mtx); 2322 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2323 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2324 mtx_unlock(&dnp->n_mtx); 2325 } else { 2326 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED); 2327 *tl++ = cookie.nfsuquad[0]; 2328 } 2329 *tl = txdr_unsigned(nmp->nm_readdirsize); 2330 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred); 2331 if (v3) { 2332 nfsm_postop_attr(vp, attrflag); 2333 if (!error) { 2334 tl = nfsm_dissect(u_int32_t *, 2335 2 * NFSX_UNSIGNED); 2336 mtx_lock(&dnp->n_mtx); 2337 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2338 dnp->n_cookieverf.nfsuquad[1] = *tl; 2339 mtx_unlock(&dnp->n_mtx); 2340 } else { 2341 m_freem(mrep); 2342 goto nfsmout; 2343 } 2344 } 2345 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); 2346 more_dirs = fxdr_unsigned(int, *tl); 2347 2348 /* loop thru the dir entries, doctoring them to 4bsd form */ 2349 while (more_dirs && bigenough) { 2350 if (v3) { 2351 tl = nfsm_dissect(u_int32_t *, 2352 3 * NFSX_UNSIGNED); 2353 fileno = fxdr_hyper(tl); 2354 len = fxdr_unsigned(int, *(tl + 2)); 2355 } else { 2356 tl = nfsm_dissect(u_int32_t *, 2357 2 * NFSX_UNSIGNED); 2358 fileno = fxdr_unsigned(u_quad_t, *tl++); 2359 len = fxdr_unsigned(int, *tl); 2360 } 2361 if (len <= 0 || len > NFS_MAXNAMLEN) { 2362 error = EBADRPC; 2363 m_freem(mrep); 2364 goto nfsmout; 2365 } 2366 tlen = nfsm_rndup(len); 2367 if (tlen == len) 2368 tlen += 4; /* To ensure null termination */ 2369 left = DIRBLKSIZ - blksiz; 2370 if ((tlen + DIRHDSIZ) > left) { 2371 dp->d_reclen += left; 2372 uiop->uio_iov->iov_base = 2373 (char *)uiop->uio_iov->iov_base + left; 2374 uiop->uio_iov->iov_len -= left; 2375 uiop->uio_offset += left; 2376 uiop->uio_resid -= left; 2377 blksiz = 0; 2378 } 2379 if ((tlen + DIRHDSIZ) > uiop->uio_resid) 2380 bigenough = 0; 2381 if (bigenough) { 2382 dp = (struct dirent *)uiop->uio_iov->iov_base; 2383 dp->d_fileno = (int)fileno; 2384 dp->d_namlen = len; 2385 dp->d_reclen = tlen + DIRHDSIZ; 2386 dp->d_type = DT_UNKNOWN; 2387 blksiz += dp->d_reclen; 2388 if (blksiz == DIRBLKSIZ) 2389 blksiz = 0; 2390 uiop->uio_offset += DIRHDSIZ; 2391 uiop->uio_resid -= DIRHDSIZ; 2392 uiop->uio_iov->iov_base = 2393 (char *)uiop->uio_iov->iov_base + DIRHDSIZ; 2394 uiop->uio_iov->iov_len -= DIRHDSIZ; 2395 nfsm_mtouio(uiop, len); 2396 cp = uiop->uio_iov->iov_base; 2397 tlen -= len; 2398 *cp = '\0'; /* null terminate */ 2399 uiop->uio_iov->iov_base = 2400 (char *)uiop->uio_iov->iov_base + tlen; 2401 uiop->uio_iov->iov_len -= tlen; 2402 uiop->uio_offset += tlen; 2403 uiop->uio_resid -= tlen; 2404 } else 2405 nfsm_adv(nfsm_rndup(len)); 2406 if (v3) { 2407 tl = nfsm_dissect(u_int32_t *, 2408 3 * NFSX_UNSIGNED); 2409 } else { 2410 tl = nfsm_dissect(u_int32_t *, 2411 2 * NFSX_UNSIGNED); 2412 } 2413 if (bigenough) { 2414 cookie.nfsuquad[0] = *tl++; 2415 if (v3) 2416 cookie.nfsuquad[1] = *tl++; 2417 } else if (v3) 2418 tl += 2; 2419 else 2420 tl++; 2421 more_dirs = fxdr_unsigned(int, *tl); 2422 } 2423 /* 2424 * If at end of rpc data, get the eof boolean 2425 */ 2426 if (!more_dirs) { 2427 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); 2428 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2429 } 2430 m_freem(mrep); 2431 } 2432 /* 2433 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2434 * by increasing d_reclen for the last record. 2435 */ 2436 if (blksiz > 0) { 2437 left = DIRBLKSIZ - blksiz; 2438 dp->d_reclen += left; 2439 uiop->uio_iov->iov_base = 2440 (char *)uiop->uio_iov->iov_base + left; 2441 uiop->uio_iov->iov_len -= left; 2442 uiop->uio_offset += left; 2443 uiop->uio_resid -= left; 2444 } 2445 2446 /* 2447 * We are now either at the end of the directory or have filled the 2448 * block. 2449 */ 2450 if (bigenough) 2451 dnp->n_direofoffset = uiop->uio_offset; 2452 else { 2453 if (uiop->uio_resid > 0) 2454 nfs_printf("EEK! readdirrpc resid > 0\n"); 2455 nfs_dircookie_lock(dnp); 2456 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2457 *cookiep = cookie; 2458 nfs_dircookie_unlock(dnp); 2459 } 2460nfsmout: 2461 return (error); 2462} 2463 2464/* 2465 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2466 */ 2467int 2468nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 2469{ 2470 int len, left; 2471 struct dirent *dp; 2472 u_int32_t *tl; 2473 caddr_t cp; 2474 struct vnode *newvp; 2475 nfsuint64 *cookiep; 2476 caddr_t bpos, dpos, dpossav1, dpossav2; 2477 struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2; 2478 struct nameidata nami, *ndp = &nami; 2479 struct componentname *cnp = &ndp->ni_cnd; 2480 nfsuint64 cookie; 2481 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2482 struct nfsnode *dnp = VTONFS(vp), *np; 2483 struct vattr vattr, dvattr; 2484 nfsfh_t *fhp; 2485 u_quad_t fileno; 2486 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2487 int attrflag, dattrflag, fhsize; 2488 2489#ifndef nolint 2490 dp = NULL; 2491#endif 2492 KASSERT(uiop->uio_iovcnt == 1 && 2493 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2494 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2495 ("nfs readdirplusrpc bad uio")); 2496 ndp->ni_dvp = vp; 2497 newvp = NULLVP; 2498 2499 /* 2500 * If there is no cookie, assume directory was stale. 2501 */ 2502 nfs_dircookie_lock(dnp); 2503 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2504 if (cookiep) { 2505 cookie = *cookiep; 2506 nfs_dircookie_unlock(dnp); 2507 } else { 2508 nfs_dircookie_unlock(dnp); 2509 return (NFSERR_BAD_COOKIE); 2510 } 2511 /* 2512 * Loop around doing readdir rpc's of size nm_readdirsize 2513 * truncated to a multiple of DIRBLKSIZ. 2514 * The stopping criteria is EOF or buffer full. 2515 */ 2516 while (more_dirs && bigenough) { 2517 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2518 mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS, 2519 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2520 mb = mreq; 2521 bpos = mtod(mb, caddr_t); 2522 nfsm_fhtom(vp, 1); 2523 tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED); 2524 *tl++ = cookie.nfsuquad[0]; 2525 *tl++ = cookie.nfsuquad[1]; 2526 mtx_lock(&dnp->n_mtx); 2527 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2528 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2529 mtx_unlock(&dnp->n_mtx); 2530 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2531 *tl = txdr_unsigned(nmp->nm_rsize); 2532 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred); 2533 nfsm_postop_attr_va(vp, dattrflag, &dvattr); 2534 if (error) { 2535 m_freem(mrep); 2536 goto nfsmout; 2537 } 2538 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED); 2539 mtx_lock(&dnp->n_mtx); 2540 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2541 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2542 mtx_unlock(&dnp->n_mtx); 2543 more_dirs = fxdr_unsigned(int, *tl); 2544 2545 /* loop thru the dir entries, doctoring them to 4bsd form */ 2546 while (more_dirs && bigenough) { 2547 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED); 2548 fileno = fxdr_hyper(tl); 2549 len = fxdr_unsigned(int, *(tl + 2)); 2550 if (len <= 0 || len > NFS_MAXNAMLEN) { 2551 error = EBADRPC; 2552 m_freem(mrep); 2553 goto nfsmout; 2554 } 2555 tlen = nfsm_rndup(len); 2556 if (tlen == len) 2557 tlen += 4; /* To ensure null termination*/ 2558 left = DIRBLKSIZ - blksiz; 2559 if ((tlen + DIRHDSIZ) > left) { 2560 dp->d_reclen += left; 2561 uiop->uio_iov->iov_base = 2562 (char *)uiop->uio_iov->iov_base + left; 2563 uiop->uio_iov->iov_len -= left; 2564 uiop->uio_offset += left; 2565 uiop->uio_resid -= left; 2566 blksiz = 0; 2567 } 2568 if ((tlen + DIRHDSIZ) > uiop->uio_resid) 2569 bigenough = 0; 2570 if (bigenough) { 2571 dp = (struct dirent *)uiop->uio_iov->iov_base; 2572 dp->d_fileno = (int)fileno; 2573 dp->d_namlen = len; 2574 dp->d_reclen = tlen + DIRHDSIZ; 2575 dp->d_type = DT_UNKNOWN; 2576 blksiz += dp->d_reclen; 2577 if (blksiz == DIRBLKSIZ) 2578 blksiz = 0; 2579 uiop->uio_offset += DIRHDSIZ; 2580 uiop->uio_resid -= DIRHDSIZ; 2581 uiop->uio_iov->iov_base = 2582 (char *)uiop->uio_iov->iov_base + DIRHDSIZ; 2583 uiop->uio_iov->iov_len -= DIRHDSIZ; 2584 cnp->cn_nameptr = uiop->uio_iov->iov_base; 2585 cnp->cn_namelen = len; 2586 nfsm_mtouio(uiop, len); 2587 cp = uiop->uio_iov->iov_base; 2588 tlen -= len; 2589 *cp = '\0'; 2590 uiop->uio_iov->iov_base = 2591 (char *)uiop->uio_iov->iov_base + tlen; 2592 uiop->uio_iov->iov_len -= tlen; 2593 uiop->uio_offset += tlen; 2594 uiop->uio_resid -= tlen; 2595 } else 2596 nfsm_adv(nfsm_rndup(len)); 2597 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED); 2598 if (bigenough) { 2599 cookie.nfsuquad[0] = *tl++; 2600 cookie.nfsuquad[1] = *tl++; 2601 } else 2602 tl += 2; 2603 2604 /* 2605 * Since the attributes are before the file handle 2606 * (sigh), we must skip over the attributes and then 2607 * come back and get them. 2608 */ 2609 attrflag = fxdr_unsigned(int, *tl); 2610 if (attrflag) { 2611 dpossav1 = dpos; 2612 mdsav1 = md; 2613 nfsm_adv(NFSX_V3FATTR); 2614 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); 2615 doit = fxdr_unsigned(int, *tl); 2616 /* 2617 * Skip loading the attrs for "..". There's a 2618 * race between loading the attrs here and 2619 * lookups that look for the directory currently 2620 * being read (in the parent). We try to acquire 2621 * the exclusive lock on ".." here, owning the 2622 * lock on the directory being read. Lookup will 2623 * hold the lock on ".." and try to acquire the 2624 * lock on the directory being read. 2625 * 2626 * There are other ways of fixing this, one would 2627 * be to do a trylock on the ".." vnode and skip 2628 * loading the attrs on ".." if it happens to be 2629 * locked by another process. But skipping the 2630 * attrload on ".." seems the easiest option. 2631 */ 2632 if (strcmp(dp->d_name, "..") == 0) { 2633 doit = 0; 2634 /* 2635 * We've already skipped over the attrs, 2636 * skip over the filehandle. And store d_type 2637 * as VDIR. 2638 */ 2639 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); 2640 i = fxdr_unsigned(int, *tl); 2641 nfsm_adv(nfsm_rndup(i)); 2642 dp->d_type = IFTODT(VTTOIF(VDIR)); 2643 } 2644 if (doit) { 2645 nfsm_getfh(fhp, fhsize, 1); 2646 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2647 VREF(vp); 2648 newvp = vp; 2649 np = dnp; 2650 } else { 2651 error = nfs_nget(vp->v_mount, fhp, 2652 fhsize, &np, LK_EXCLUSIVE); 2653 if (error) 2654 doit = 0; 2655 else 2656 newvp = NFSTOV(np); 2657 } 2658 } 2659 if (doit && bigenough) { 2660 dpossav2 = dpos; 2661 dpos = dpossav1; 2662 mdsav2 = md; 2663 md = mdsav1; 2664 nfsm_loadattr(newvp, &vattr); 2665 dpos = dpossav2; 2666 md = mdsav2; 2667 dp->d_type = IFTODT(VTTOIF(vattr.va_type)); 2668 ndp->ni_vp = newvp; 2669 if (newvp->v_type != VDIR || dattrflag != 0) 2670 cache_enter_time(ndp->ni_dvp, ndp->ni_vp, 2671 cnp, &vattr.va_ctime, 2672 newvp->v_type != VDIR ? NULL : 2673 &dvattr.va_ctime); 2674 } 2675 } else { 2676 /* Just skip over the file handle */ 2677 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); 2678 i = fxdr_unsigned(int, *tl); 2679 if (i) { 2680 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); 2681 fhsize = fxdr_unsigned(int, *tl); 2682 nfsm_adv(nfsm_rndup(fhsize)); 2683 } 2684 } 2685 if (newvp != NULLVP) { 2686 if (newvp == vp) 2687 vrele(newvp); 2688 else 2689 vput(newvp); 2690 newvp = NULLVP; 2691 } 2692 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); 2693 more_dirs = fxdr_unsigned(int, *tl); 2694 } 2695 /* 2696 * If at end of rpc data, get the eof boolean 2697 */ 2698 if (!more_dirs) { 2699 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); 2700 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2701 } 2702 m_freem(mrep); 2703 } 2704 /* 2705 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2706 * by increasing d_reclen for the last record. 2707 */ 2708 if (blksiz > 0) { 2709 left = DIRBLKSIZ - blksiz; 2710 dp->d_reclen += left; 2711 uiop->uio_iov->iov_base = 2712 (char *)uiop->uio_iov->iov_base + left; 2713 uiop->uio_iov->iov_len -= left; 2714 uiop->uio_offset += left; 2715 uiop->uio_resid -= left; 2716 } 2717 2718 /* 2719 * We are now either at the end of the directory or have filled the 2720 * block. 2721 */ 2722 if (bigenough) 2723 dnp->n_direofoffset = uiop->uio_offset; 2724 else { 2725 if (uiop->uio_resid > 0) 2726 nfs_printf("EEK! readdirplusrpc resid > 0\n"); 2727 nfs_dircookie_lock(dnp); 2728 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2729 *cookiep = cookie; 2730 nfs_dircookie_unlock(dnp); 2731 } 2732nfsmout: 2733 if (newvp != NULLVP) { 2734 if (newvp == vp) 2735 vrele(newvp); 2736 else 2737 vput(newvp); 2738 newvp = NULLVP; 2739 } 2740 return (error); 2741} 2742 2743/* 2744 * Silly rename. To make the NFS filesystem that is stateless look a little 2745 * more like the "ufs" a remove of an active vnode is translated to a rename 2746 * to a funny looking filename that is removed by nfs_inactive on the 2747 * nfsnode. There is the potential for another process on a different client 2748 * to create the same funny name between the nfs_lookitup() fails and the 2749 * nfs_rename() completes, but... 2750 */ 2751static int 2752nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2753{ 2754 struct sillyrename *sp; 2755 struct nfsnode *np; 2756 int error; 2757 short pid; 2758 unsigned int lticks; 2759 2760 cache_purge(dvp); 2761 np = VTONFS(vp); 2762 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir")); 2763 sp = malloc(sizeof (struct sillyrename), 2764 M_NFSREQ, M_WAITOK); 2765 sp->s_cred = crhold(cnp->cn_cred); 2766 sp->s_dvp = dvp; 2767 sp->s_removeit = nfs_removeit; 2768 VREF(dvp); 2769 2770 /* 2771 * Fudge together a funny name. 2772 * Changing the format of the funny name to accomodate more 2773 * sillynames per directory. 2774 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is 2775 * CPU ticks since boot. 2776 */ 2777 pid = cnp->cn_thread->td_proc->p_pid; 2778 lticks = (unsigned int)ticks; 2779 for ( ; ; ) { 2780 sp->s_namlen = sprintf(sp->s_name, 2781 ".nfs.%08x.%04x4.4", lticks, 2782 pid); 2783 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2784 cnp->cn_thread, NULL)) 2785 break; 2786 lticks++; 2787 } 2788 error = nfs_renameit(dvp, cnp, sp); 2789 if (error) 2790 goto bad; 2791 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2792 cnp->cn_thread, &np); 2793 np->n_sillyrename = sp; 2794 return (0); 2795bad: 2796 vrele(sp->s_dvp); 2797 crfree(sp->s_cred); 2798 free((caddr_t)sp, M_NFSREQ); 2799 return (error); 2800} 2801 2802/* 2803 * Look up a file name and optionally either update the file handle or 2804 * allocate an nfsnode, depending on the value of npp. 2805 * npp == NULL --> just do the lookup 2806 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2807 * handled too 2808 * *npp != NULL --> update the file handle in the vnode 2809 */ 2810static int 2811nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred, 2812 struct thread *td, struct nfsnode **npp) 2813{ 2814 struct vnode *newvp = NULL; 2815 struct nfsnode *np, *dnp = VTONFS(dvp); 2816 caddr_t bpos, dpos; 2817 int error = 0, fhlen, attrflag; 2818 struct mbuf *mreq, *mrep, *md, *mb; 2819 nfsfh_t *nfhp; 2820 int v3 = NFS_ISV3(dvp); 2821 2822 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2823 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP, 2824 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2825 mb = mreq; 2826 bpos = mtod(mb, caddr_t); 2827 nfsm_fhtom(dvp, v3); 2828 nfsm_strtom(name, len, NFS_MAXNAMLEN); 2829 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred); 2830 if (npp && !error) { 2831 nfsm_getfh(nfhp, fhlen, v3); 2832 if (*npp) { 2833 np = *npp; 2834 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2835 free((caddr_t)np->n_fhp, M_NFSBIGFH); 2836 np->n_fhp = &np->n_fh; 2837 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH) 2838 np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK); 2839 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen); 2840 np->n_fhsize = fhlen; 2841 newvp = NFSTOV(np); 2842 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2843 VREF(dvp); 2844 newvp = dvp; 2845 } else { 2846 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, LK_EXCLUSIVE); 2847 if (error) { 2848 m_freem(mrep); 2849 return (error); 2850 } 2851 newvp = NFSTOV(np); 2852 } 2853 if (v3) { 2854 nfsm_postop_attr(newvp, attrflag); 2855 if (!attrflag && *npp == NULL) { 2856 m_freem(mrep); 2857 if (newvp == dvp) 2858 vrele(newvp); 2859 else 2860 vput(newvp); 2861 return (ENOENT); 2862 } 2863 } else 2864 nfsm_loadattr(newvp, NULL); 2865 } 2866 m_freem(mrep); 2867nfsmout: 2868 if (npp && *npp == NULL) { 2869 if (error) { 2870 if (newvp) { 2871 if (newvp == dvp) 2872 vrele(newvp); 2873 else 2874 vput(newvp); 2875 } 2876 } else 2877 *npp = np; 2878 } 2879 return (error); 2880} 2881 2882/* 2883 * Nfs Version 3 commit rpc 2884 */ 2885int 2886nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred, 2887 struct thread *td) 2888{ 2889 u_int32_t *tl; 2890 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2891 caddr_t bpos, dpos; 2892 int error = 0, wccflag = NFSV3_WCCRATTR; 2893 struct mbuf *mreq, *mrep, *md, *mb; 2894 2895 mtx_lock(&nmp->nm_mtx); 2896 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) { 2897 mtx_unlock(&nmp->nm_mtx); 2898 return (0); 2899 } 2900 mtx_unlock(&nmp->nm_mtx); 2901 nfsstats.rpccnt[NFSPROC_COMMIT]++; 2902 mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1)); 2903 mb = mreq; 2904 bpos = mtod(mb, caddr_t); 2905 nfsm_fhtom(vp, 1); 2906 tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED); 2907 txdr_hyper(offset, tl); 2908 tl += 2; 2909 *tl = txdr_unsigned(cnt); 2910 nfsm_request(vp, NFSPROC_COMMIT, td, cred); 2911 nfsm_wcc_data(vp, wccflag); 2912 if (!error) { 2913 tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF); 2914 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl, 2915 NFSX_V3WRITEVERF)) { 2916 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 2917 NFSX_V3WRITEVERF); 2918 error = NFSERR_STALEWRITEVERF; 2919 } 2920 } 2921 m_freem(mrep); 2922nfsmout: 2923 return (error); 2924} 2925 2926/* 2927 * Strategy routine. 2928 * For async requests when nfsiod(s) are running, queue the request by 2929 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 2930 * request. 2931 */ 2932static int 2933nfs_strategy(struct vop_strategy_args *ap) 2934{ 2935 struct buf *bp = ap->a_bp; 2936 struct ucred *cr; 2937 2938 KASSERT(!(bp->b_flags & B_DONE), 2939 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); 2940 BUF_ASSERT_HELD(bp); 2941 2942 if (bp->b_iocmd == BIO_READ) 2943 cr = bp->b_rcred; 2944 else 2945 cr = bp->b_wcred; 2946 2947 /* 2948 * If the op is asynchronous and an i/o daemon is waiting 2949 * queue the request, wake it up and wait for completion 2950 * otherwise just do it ourselves. 2951 */ 2952 if ((bp->b_flags & B_ASYNC) == 0 || 2953 nfs_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread)) 2954 (void)nfs_doio(ap->a_vp, bp, cr, curthread); 2955 return (0); 2956} 2957 2958/* 2959 * fsync vnode op. Just call nfs_flush() with commit == 1. 2960 */ 2961/* ARGSUSED */ 2962static int 2963nfs_fsync(struct vop_fsync_args *ap) 2964{ 2965 2966 return (nfs_flush(ap->a_vp, ap->a_waitfor, 1)); 2967} 2968 2969/* 2970 * Flush all the blocks associated with a vnode. 2971 * Walk through the buffer pool and push any dirty pages 2972 * associated with the vnode. 2973 */ 2974static int 2975nfs_flush(struct vnode *vp, int waitfor, int commit) 2976{ 2977 struct nfsnode *np = VTONFS(vp); 2978 struct buf *bp; 2979 int i; 2980 struct buf *nbp; 2981 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2982 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos; 2983 int passone = 1; 2984 u_quad_t off, endoff, toff; 2985 struct ucred* wcred = NULL; 2986 struct buf **bvec = NULL; 2987 struct bufobj *bo; 2988 struct thread *td = curthread; 2989#ifndef NFS_COMMITBVECSIZ 2990#define NFS_COMMITBVECSIZ 20 2991#endif 2992 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ]; 2993 int bvecsize = 0, bveccount; 2994 2995 if (nmp->nm_flag & NFSMNT_INT) 2996 slpflag = NFS_PCATCH; 2997 if (!commit) 2998 passone = 0; 2999 bo = &vp->v_bufobj; 3000 /* 3001 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the 3002 * server, but has not been committed to stable storage on the server 3003 * yet. On the first pass, the byte range is worked out and the commit 3004 * rpc is done. On the second pass, nfs_writebp() is called to do the 3005 * job. 3006 */ 3007again: 3008 off = (u_quad_t)-1; 3009 endoff = 0; 3010 bvecpos = 0; 3011 if (NFS_ISV3(vp) && commit) { 3012 if (bvec != NULL && bvec != bvec_on_stack) 3013 free(bvec, M_TEMP); 3014 /* 3015 * Count up how many buffers waiting for a commit. 3016 */ 3017 bveccount = 0; 3018 BO_LOCK(bo); 3019 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 3020 if (!BUF_ISLOCKED(bp) && 3021 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) 3022 == (B_DELWRI | B_NEEDCOMMIT)) 3023 bveccount++; 3024 } 3025 /* 3026 * Allocate space to remember the list of bufs to commit. It is 3027 * important to use M_NOWAIT here to avoid a race with nfs_write. 3028 * If we can't get memory (for whatever reason), we will end up 3029 * committing the buffers one-by-one in the loop below. 3030 */ 3031 if (bveccount > NFS_COMMITBVECSIZ) { 3032 /* 3033 * Release the vnode interlock to avoid a lock 3034 * order reversal. 3035 */ 3036 BO_UNLOCK(bo); 3037 bvec = (struct buf **) 3038 malloc(bveccount * sizeof(struct buf *), 3039 M_TEMP, M_NOWAIT); 3040 BO_LOCK(bo); 3041 if (bvec == NULL) { 3042 bvec = bvec_on_stack; 3043 bvecsize = NFS_COMMITBVECSIZ; 3044 } else 3045 bvecsize = bveccount; 3046 } else { 3047 bvec = bvec_on_stack; 3048 bvecsize = NFS_COMMITBVECSIZ; 3049 } 3050 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 3051 if (bvecpos >= bvecsize) 3052 break; 3053 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 3054 nbp = TAILQ_NEXT(bp, b_bobufs); 3055 continue; 3056 } 3057 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 3058 (B_DELWRI | B_NEEDCOMMIT)) { 3059 BUF_UNLOCK(bp); 3060 nbp = TAILQ_NEXT(bp, b_bobufs); 3061 continue; 3062 } 3063 BO_UNLOCK(bo); 3064 bremfree(bp); 3065 /* 3066 * Work out if all buffers are using the same cred 3067 * so we can deal with them all with one commit. 3068 * 3069 * NOTE: we are not clearing B_DONE here, so we have 3070 * to do it later on in this routine if we intend to 3071 * initiate I/O on the bp. 3072 * 3073 * Note: to avoid loopback deadlocks, we do not 3074 * assign b_runningbufspace. 3075 */ 3076 if (wcred == NULL) 3077 wcred = bp->b_wcred; 3078 else if (wcred != bp->b_wcred) 3079 wcred = NOCRED; 3080 vfs_busy_pages(bp, 1); 3081 3082 BO_LOCK(bo); 3083 /* 3084 * bp is protected by being locked, but nbp is not 3085 * and vfs_busy_pages() may sleep. We have to 3086 * recalculate nbp. 3087 */ 3088 nbp = TAILQ_NEXT(bp, b_bobufs); 3089 3090 /* 3091 * A list of these buffers is kept so that the 3092 * second loop knows which buffers have actually 3093 * been committed. This is necessary, since there 3094 * may be a race between the commit rpc and new 3095 * uncommitted writes on the file. 3096 */ 3097 bvec[bvecpos++] = bp; 3098 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 3099 bp->b_dirtyoff; 3100 if (toff < off) 3101 off = toff; 3102 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); 3103 if (toff > endoff) 3104 endoff = toff; 3105 } 3106 BO_UNLOCK(bo); 3107 } 3108 if (bvecpos > 0) { 3109 /* 3110 * Commit data on the server, as required. 3111 * If all bufs are using the same wcred, then use that with 3112 * one call for all of them, otherwise commit each one 3113 * separately. 3114 */ 3115 if (wcred != NOCRED) 3116 retv = nfs_commit(vp, off, (int)(endoff - off), 3117 wcred, td); 3118 else { 3119 retv = 0; 3120 for (i = 0; i < bvecpos; i++) { 3121 off_t off, size; 3122 bp = bvec[i]; 3123 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 3124 bp->b_dirtyoff; 3125 size = (u_quad_t)(bp->b_dirtyend 3126 - bp->b_dirtyoff); 3127 retv = nfs_commit(vp, off, (int)size, 3128 bp->b_wcred, td); 3129 if (retv) break; 3130 } 3131 } 3132 3133 if (retv == NFSERR_STALEWRITEVERF) 3134 nfs_clearcommit(vp->v_mount); 3135 3136 /* 3137 * Now, either mark the blocks I/O done or mark the 3138 * blocks dirty, depending on whether the commit 3139 * succeeded. 3140 */ 3141 for (i = 0; i < bvecpos; i++) { 3142 bp = bvec[i]; 3143 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 3144 if (retv) { 3145 /* 3146 * Error, leave B_DELWRI intact 3147 */ 3148 vfs_unbusy_pages(bp); 3149 brelse(bp); 3150 } else { 3151 /* 3152 * Success, remove B_DELWRI ( bundirty() ). 3153 * 3154 * b_dirtyoff/b_dirtyend seem to be NFS 3155 * specific. We should probably move that 3156 * into bundirty(). XXX 3157 */ 3158 bufobj_wref(bo); 3159 bp->b_flags |= B_ASYNC; 3160 bundirty(bp); 3161 bp->b_flags &= ~B_DONE; 3162 bp->b_ioflags &= ~BIO_ERROR; 3163 bp->b_dirtyoff = bp->b_dirtyend = 0; 3164 bufdone(bp); 3165 } 3166 } 3167 } 3168 3169 /* 3170 * Start/do any write(s) that are required. 3171 */ 3172loop: 3173 BO_LOCK(bo); 3174 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 3175 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 3176 if (waitfor != MNT_WAIT || passone) 3177 continue; 3178 3179 error = BUF_TIMELOCK(bp, 3180 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 3181 BO_MTX(bo), "nfsfsync", slpflag, slptimeo); 3182 if (error == 0) { 3183 BUF_UNLOCK(bp); 3184 goto loop; 3185 } 3186 if (error == ENOLCK) { 3187 error = 0; 3188 goto loop; 3189 } 3190 if (nfs_sigintr(nmp, td)) { 3191 error = EINTR; 3192 goto done; 3193 } 3194 if (slpflag & PCATCH) { 3195 slpflag = 0; 3196 slptimeo = 2 * hz; 3197 } 3198 goto loop; 3199 } 3200 if ((bp->b_flags & B_DELWRI) == 0) 3201 panic("nfs_fsync: not dirty"); 3202 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) { 3203 BUF_UNLOCK(bp); 3204 continue; 3205 } 3206 BO_UNLOCK(bo); 3207 bremfree(bp); 3208 if (passone || !commit) 3209 bp->b_flags |= B_ASYNC; 3210 else 3211 bp->b_flags |= B_ASYNC; 3212 bwrite(bp); 3213 if (nfs_sigintr(nmp, td)) { 3214 error = EINTR; 3215 goto done; 3216 } 3217 goto loop; 3218 } 3219 if (passone) { 3220 passone = 0; 3221 BO_UNLOCK(bo); 3222 goto again; 3223 } 3224 if (waitfor == MNT_WAIT) { 3225 while (bo->bo_numoutput) { 3226 error = bufobj_wwait(bo, slpflag, slptimeo); 3227 if (error) { 3228 BO_UNLOCK(bo); 3229 error = nfs_sigintr(nmp, td); 3230 if (error) 3231 goto done; 3232 if (slpflag & PCATCH) { 3233 slpflag = 0; 3234 slptimeo = 2 * hz; 3235 } 3236 BO_LOCK(bo); 3237 } 3238 } 3239 if (bo->bo_dirty.bv_cnt != 0 && commit) { 3240 BO_UNLOCK(bo); 3241 goto loop; 3242 } 3243 /* 3244 * Wait for all the async IO requests to drain 3245 */ 3246 BO_UNLOCK(bo); 3247 mtx_lock(&np->n_mtx); 3248 while (np->n_directio_asyncwr > 0) { 3249 np->n_flag |= NFSYNCWAIT; 3250 error = nfs_msleep(td, (caddr_t)&np->n_directio_asyncwr, 3251 &np->n_mtx, slpflag | (PRIBIO + 1), 3252 "nfsfsync", 0); 3253 if (error) { 3254 if (nfs_sigintr(nmp, td)) { 3255 mtx_unlock(&np->n_mtx); 3256 error = EINTR; 3257 goto done; 3258 } 3259 } 3260 } 3261 mtx_unlock(&np->n_mtx); 3262 } else 3263 BO_UNLOCK(bo); 3264 mtx_lock(&np->n_mtx); 3265 if (np->n_flag & NWRITEERR) { 3266 error = np->n_error; 3267 np->n_flag &= ~NWRITEERR; 3268 } 3269 if (commit && bo->bo_dirty.bv_cnt == 0 && 3270 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0) 3271 np->n_flag &= ~NMODIFIED; 3272 mtx_unlock(&np->n_mtx); 3273done: 3274 if (bvec != NULL && bvec != bvec_on_stack) 3275 free(bvec, M_TEMP); 3276 return (error); 3277} 3278 3279/* 3280 * NFS advisory byte-level locks. 3281 */ 3282static int 3283nfs_advlock(struct vop_advlock_args *ap) 3284{ 3285 struct vnode *vp = ap->a_vp; 3286 u_quad_t size; 3287 int error; 3288 3289 error = vn_lock(vp, LK_SHARED); 3290 if (error) 3291 return (error); 3292 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3293 size = VTONFS(vp)->n_size; 3294 VOP_UNLOCK(vp, 0); 3295 error = lf_advlock(ap, &(vp->v_lockf), size); 3296 } else { 3297 if (nfs_advlock_p) 3298 error = nfs_advlock_p(ap); 3299 else 3300 error = ENOLCK; 3301 } 3302 3303 return (error); 3304} 3305 3306/* 3307 * NFS advisory byte-level locks. 3308 */ 3309static int 3310nfs_advlockasync(struct vop_advlockasync_args *ap) 3311{ 3312 struct vnode *vp = ap->a_vp; 3313 u_quad_t size; 3314 int error; 3315 3316 error = vn_lock(vp, LK_SHARED); 3317 if (error) 3318 return (error); 3319 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3320 size = VTONFS(vp)->n_size; 3321 VOP_UNLOCK(vp, 0); 3322 error = lf_advlockasync(ap, &(vp->v_lockf), size); 3323 } else { 3324 VOP_UNLOCK(vp, 0); 3325 error = EOPNOTSUPP; 3326 } 3327 return (error); 3328} 3329 3330/* 3331 * Print out the contents of an nfsnode. 3332 */ 3333static int 3334nfs_print(struct vop_print_args *ap) 3335{ 3336 struct vnode *vp = ap->a_vp; 3337 struct nfsnode *np = VTONFS(vp); 3338 3339 nfs_printf("\tfileid %ld fsid 0x%x", 3340 np->n_vattr.va_fileid, np->n_vattr.va_fsid); 3341 if (vp->v_type == VFIFO) 3342 fifo_printinfo(vp); 3343 printf("\n"); 3344 return (0); 3345} 3346 3347/* 3348 * This is the "real" nfs::bwrite(struct buf*). 3349 * We set B_CACHE if this is a VMIO buffer. 3350 */ 3351int 3352nfs_writebp(struct buf *bp, int force __unused, struct thread *td) 3353{ 3354 int s; 3355 int oldflags = bp->b_flags; 3356#if 0 3357 int retv = 1; 3358 off_t off; 3359#endif 3360 3361 BUF_ASSERT_HELD(bp); 3362 3363 if (bp->b_flags & B_INVAL) { 3364 brelse(bp); 3365 return(0); 3366 } 3367 3368 bp->b_flags |= B_CACHE; 3369 3370 /* 3371 * Undirty the bp. We will redirty it later if the I/O fails. 3372 */ 3373 3374 s = splbio(); 3375 bundirty(bp); 3376 bp->b_flags &= ~B_DONE; 3377 bp->b_ioflags &= ~BIO_ERROR; 3378 bp->b_iocmd = BIO_WRITE; 3379 3380 bufobj_wref(bp->b_bufobj); 3381 curthread->td_ru.ru_oublock++; 3382 splx(s); 3383 3384 /* 3385 * Note: to avoid loopback deadlocks, we do not 3386 * assign b_runningbufspace. 3387 */ 3388 vfs_busy_pages(bp, 1); 3389 3390 BUF_KERNPROC(bp); 3391 bp->b_iooffset = dbtob(bp->b_blkno); 3392 bstrategy(bp); 3393 3394 if( (oldflags & B_ASYNC) == 0) { 3395 int rtval = bufwait(bp); 3396 3397 if (oldflags & B_DELWRI) { 3398 s = splbio(); 3399 reassignbuf(bp); 3400 splx(s); 3401 } 3402 brelse(bp); 3403 return (rtval); 3404 } 3405 3406 return (0); 3407} 3408 3409/* 3410 * nfs special file access vnode op. 3411 * Essentially just get vattr and then imitate iaccess() since the device is 3412 * local to the client. 3413 */ 3414static int 3415nfsspec_access(struct vop_access_args *ap) 3416{ 3417 struct vattr *vap; 3418 struct ucred *cred = ap->a_cred; 3419 struct vnode *vp = ap->a_vp; 3420 accmode_t accmode = ap->a_accmode; 3421 struct vattr vattr; 3422 int error; 3423 3424 /* 3425 * Disallow write attempts on filesystems mounted read-only; 3426 * unless the file is a socket, fifo, or a block or character 3427 * device resident on the filesystem. 3428 */ 3429 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3430 switch (vp->v_type) { 3431 case VREG: 3432 case VDIR: 3433 case VLNK: 3434 return (EROFS); 3435 default: 3436 break; 3437 } 3438 } 3439 vap = &vattr; 3440 error = VOP_GETATTR(vp, vap, cred); 3441 if (error) 3442 goto out; 3443 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid, 3444 accmode, cred, NULL); 3445out: 3446 return error; 3447} 3448 3449/* 3450 * Read wrapper for fifos. 3451 */ 3452static int 3453nfsfifo_read(struct vop_read_args *ap) 3454{ 3455 struct nfsnode *np = VTONFS(ap->a_vp); 3456 int error; 3457 3458 /* 3459 * Set access flag. 3460 */ 3461 mtx_lock(&np->n_mtx); 3462 np->n_flag |= NACC; 3463 vfs_timestamp(&np->n_atim); 3464 mtx_unlock(&np->n_mtx); 3465 error = fifo_specops.vop_read(ap); 3466 return error; 3467} 3468 3469/* 3470 * Write wrapper for fifos. 3471 */ 3472static int 3473nfsfifo_write(struct vop_write_args *ap) 3474{ 3475 struct nfsnode *np = VTONFS(ap->a_vp); 3476 3477 /* 3478 * Set update flag. 3479 */ 3480 mtx_lock(&np->n_mtx); 3481 np->n_flag |= NUPD; 3482 vfs_timestamp(&np->n_mtim); 3483 mtx_unlock(&np->n_mtx); 3484 return(fifo_specops.vop_write(ap)); 3485} 3486 3487/* 3488 * Close wrapper for fifos. 3489 * 3490 * Update the times on the nfsnode then do fifo close. 3491 */ 3492static int 3493nfsfifo_close(struct vop_close_args *ap) 3494{ 3495 struct vnode *vp = ap->a_vp; 3496 struct nfsnode *np = VTONFS(vp); 3497 struct vattr vattr; 3498 struct timespec ts; 3499 3500 mtx_lock(&np->n_mtx); 3501 if (np->n_flag & (NACC | NUPD)) { 3502 vfs_timestamp(&ts); 3503 if (np->n_flag & NACC) 3504 np->n_atim = ts; 3505 if (np->n_flag & NUPD) 3506 np->n_mtim = ts; 3507 np->n_flag |= NCHG; 3508 if (vrefcnt(vp) == 1 && 3509 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3510 VATTR_NULL(&vattr); 3511 if (np->n_flag & NACC) 3512 vattr.va_atime = np->n_atim; 3513 if (np->n_flag & NUPD) 3514 vattr.va_mtime = np->n_mtim; 3515 mtx_unlock(&np->n_mtx); 3516 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3517 goto out; 3518 } 3519 } 3520 mtx_unlock(&np->n_mtx); 3521out: 3522 return (fifo_specops.vop_close(ap)); 3523} 3524 3525/* 3526 * Just call nfs_writebp() with the force argument set to 1. 3527 * 3528 * NOTE: B_DONE may or may not be set in a_bp on call. 3529 */ 3530static int 3531nfs_bwrite(struct buf *bp) 3532{ 3533 3534 return (nfs_writebp(bp, 1, curthread)); 3535} 3536 3537struct buf_ops buf_ops_nfs = { 3538 .bop_name = "buf_ops_nfs", 3539 .bop_write = nfs_bwrite, 3540 .bop_strategy = bufstrategy, 3541 .bop_sync = bufsync, 3542 .bop_bdflush = bufbdflush, 3543}; 3544