nfs_clvnops.c revision 230552
1/*- 2 * Copyright (c) 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * Rick Macklem at The University of Guelph. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 4. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * from nfs_vnops.c 8.16 (Berkeley) 5/27/95 33 */ 34 35#include <sys/cdefs.h> 36__FBSDID("$FreeBSD: head/sys/fs/nfsclient/nfs_clvnops.c 230552 2012-01-25 20:48:20Z kib $"); 37 38/* 39 * vnode op calls for Sun NFS version 2, 3 and 4 40 */ 41 42#include "opt_kdtrace.h" 43#include "opt_inet.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/nfs/nfsport.h> 71#include <fs/nfsclient/nfsnode.h> 72#include <fs/nfsclient/nfsmount.h> 73#include <fs/nfsclient/nfs.h> 74#include <fs/nfsclient/nfs_kdtrace.h> 75 76#include <net/if.h> 77#include <netinet/in.h> 78#include <netinet/in_var.h> 79 80#include <nfs/nfs_lock.h> 81 82#ifdef KDTRACE_HOOKS 83#include <sys/dtrace_bsd.h> 84 85dtrace_nfsclient_accesscache_flush_probe_func_t 86 dtrace_nfscl_accesscache_flush_done_probe; 87uint32_t nfscl_accesscache_flush_done_id; 88 89dtrace_nfsclient_accesscache_get_probe_func_t 90 dtrace_nfscl_accesscache_get_hit_probe, 91 dtrace_nfscl_accesscache_get_miss_probe; 92uint32_t nfscl_accesscache_get_hit_id; 93uint32_t nfscl_accesscache_get_miss_id; 94 95dtrace_nfsclient_accesscache_load_probe_func_t 96 dtrace_nfscl_accesscache_load_done_probe; 97uint32_t nfscl_accesscache_load_done_id; 98#endif /* !KDTRACE_HOOKS */ 99 100/* Defs */ 101#define TRUE 1 102#define FALSE 0 103 104extern struct nfsstats newnfsstats; 105extern int nfsrv_useacl; 106MALLOC_DECLARE(M_NEWNFSREQ); 107 108/* 109 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these 110 * calls are not in getblk() and brelse() so that they would not be necessary 111 * here. 112 */ 113#ifndef B_VMIO 114#define vfs_busy_pages(bp, f) 115#endif 116 117static vop_read_t nfsfifo_read; 118static vop_write_t nfsfifo_write; 119static vop_close_t nfsfifo_close; 120static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *, 121 struct thread *); 122static vop_lookup_t nfs_lookup; 123static vop_create_t nfs_create; 124static vop_mknod_t nfs_mknod; 125static vop_open_t nfs_open; 126static vop_pathconf_t nfs_pathconf; 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 vop_lock1_t nfs_lock1; 142static int nfs_lookitup(struct vnode *, char *, int, 143 struct ucred *, struct thread *, struct nfsnode **); 144static int nfs_sillyrename(struct vnode *, struct vnode *, 145 struct componentname *); 146static vop_access_t nfsspec_access; 147static vop_readlink_t nfs_readlink; 148static vop_print_t nfs_print; 149static vop_advlock_t nfs_advlock; 150static vop_advlockasync_t nfs_advlockasync; 151static vop_getacl_t nfs_getacl; 152static vop_setacl_t nfs_setacl; 153 154/* 155 * Global vfs data structures for nfs 156 */ 157struct vop_vector newnfs_vnodeops = { 158 .vop_default = &default_vnodeops, 159 .vop_access = nfs_access, 160 .vop_advlock = nfs_advlock, 161 .vop_advlockasync = nfs_advlockasync, 162 .vop_close = nfs_close, 163 .vop_create = nfs_create, 164 .vop_fsync = nfs_fsync, 165 .vop_getattr = nfs_getattr, 166 .vop_getpages = ncl_getpages, 167 .vop_putpages = ncl_putpages, 168 .vop_inactive = ncl_inactive, 169 .vop_link = nfs_link, 170 .vop_lock1 = nfs_lock1, 171 .vop_lookup = nfs_lookup, 172 .vop_mkdir = nfs_mkdir, 173 .vop_mknod = nfs_mknod, 174 .vop_open = nfs_open, 175 .vop_pathconf = nfs_pathconf, 176 .vop_print = nfs_print, 177 .vop_read = nfs_read, 178 .vop_readdir = nfs_readdir, 179 .vop_readlink = nfs_readlink, 180 .vop_reclaim = ncl_reclaim, 181 .vop_remove = nfs_remove, 182 .vop_rename = nfs_rename, 183 .vop_rmdir = nfs_rmdir, 184 .vop_setattr = nfs_setattr, 185 .vop_strategy = nfs_strategy, 186 .vop_symlink = nfs_symlink, 187 .vop_write = ncl_write, 188 .vop_getacl = nfs_getacl, 189 .vop_setacl = nfs_setacl, 190}; 191 192struct vop_vector newnfs_fifoops = { 193 .vop_default = &fifo_specops, 194 .vop_access = nfsspec_access, 195 .vop_close = nfsfifo_close, 196 .vop_fsync = nfs_fsync, 197 .vop_getattr = nfs_getattr, 198 .vop_inactive = ncl_inactive, 199 .vop_print = nfs_print, 200 .vop_read = nfsfifo_read, 201 .vop_reclaim = ncl_reclaim, 202 .vop_setattr = nfs_setattr, 203 .vop_write = nfsfifo_write, 204}; 205 206static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, 207 struct componentname *cnp, struct vattr *vap); 208static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 209 int namelen, struct ucred *cred, struct thread *td); 210static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, 211 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp, 212 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td); 213static int nfs_renameit(struct vnode *sdvp, struct vnode *svp, 214 struct componentname *scnp, struct sillyrename *sp); 215 216/* 217 * Global variables 218 */ 219#define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1)) 220 221SYSCTL_DECL(_vfs_nfs); 222 223static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO; 224SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 225 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 226 227static int nfs_prime_access_cache = 0; 228SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW, 229 &nfs_prime_access_cache, 0, 230 "Prime NFS ACCESS cache when fetching attributes"); 231 232static int newnfs_commit_on_close = 0; 233SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW, 234 &newnfs_commit_on_close, 0, "write+commit on close, else only write"); 235 236static int nfs_clean_pages_on_close = 1; 237SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW, 238 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close"); 239 240int newnfs_directio_enable = 0; 241SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW, 242 &newnfs_directio_enable, 0, "Enable NFS directio"); 243 244/* 245 * This sysctl allows other processes to mmap a file that has been opened 246 * O_DIRECT by a process. In general, having processes mmap the file while 247 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow 248 * this by default to prevent DoS attacks - to prevent a malicious user from 249 * opening up files O_DIRECT preventing other users from mmap'ing these 250 * files. "Protected" environments where stricter consistency guarantees are 251 * required can disable this knob. The process that opened the file O_DIRECT 252 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not 253 * meaningful. 254 */ 255int newnfs_directio_allow_mmap = 1; 256SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW, 257 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens"); 258 259#if 0 260SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD, 261 &newnfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count"); 262 263SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD, 264 &newnfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count"); 265#endif 266 267#define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \ 268 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \ 269 | NFSACCESS_DELETE | NFSACCESS_LOOKUP) 270 271/* 272 * SMP Locking Note : 273 * The list of locks after the description of the lock is the ordering 274 * of other locks acquired with the lock held. 275 * np->n_mtx : Protects the fields in the nfsnode. 276 VM Object Lock 277 VI_MTX (acquired indirectly) 278 * nmp->nm_mtx : Protects the fields in the nfsmount. 279 rep->r_mtx 280 * ncl_iod_mutex : Global lock, protects shared nfsiod state. 281 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list. 282 nmp->nm_mtx 283 rep->r_mtx 284 * rep->r_mtx : Protects the fields in an nfsreq. 285 */ 286 287static int 288nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td, 289 struct ucred *cred, u_int32_t *retmode) 290{ 291 int error = 0, attrflag, i, lrupos; 292 u_int32_t rmode; 293 struct nfsnode *np = VTONFS(vp); 294 struct nfsvattr nfsva; 295 296 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag, 297 &rmode, NULL); 298 if (attrflag) 299 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 300 if (!error) { 301 lrupos = 0; 302 mtx_lock(&np->n_mtx); 303 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 304 if (np->n_accesscache[i].uid == cred->cr_uid) { 305 np->n_accesscache[i].mode = rmode; 306 np->n_accesscache[i].stamp = time_second; 307 break; 308 } 309 if (i > 0 && np->n_accesscache[i].stamp < 310 np->n_accesscache[lrupos].stamp) 311 lrupos = i; 312 } 313 if (i == NFS_ACCESSCACHESIZE) { 314 np->n_accesscache[lrupos].uid = cred->cr_uid; 315 np->n_accesscache[lrupos].mode = rmode; 316 np->n_accesscache[lrupos].stamp = time_second; 317 } 318 mtx_unlock(&np->n_mtx); 319 if (retmode != NULL) 320 *retmode = rmode; 321 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0); 322 } else if (NFS_ISV4(vp)) { 323 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 324 } 325#ifdef KDTRACE_HOOKS 326 if (error != 0) 327 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0, 328 error); 329#endif 330 return (error); 331} 332 333/* 334 * nfs access vnode op. 335 * For nfs version 2, just return ok. File accesses may fail later. 336 * For nfs version 3, use the access rpc to check accessibility. If file modes 337 * are changed on the server, accesses might still fail later. 338 */ 339static int 340nfs_access(struct vop_access_args *ap) 341{ 342 struct vnode *vp = ap->a_vp; 343 int error = 0, i, gotahit; 344 u_int32_t mode, wmode, rmode; 345 int v34 = NFS_ISV34(vp); 346 struct nfsnode *np = VTONFS(vp); 347 348 /* 349 * Disallow write attempts on filesystems mounted read-only; 350 * unless the file is a socket, fifo, or a block or character 351 * device resident on the filesystem. 352 */ 353 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS | 354 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL | 355 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) { 356 switch (vp->v_type) { 357 case VREG: 358 case VDIR: 359 case VLNK: 360 return (EROFS); 361 default: 362 break; 363 } 364 } 365 /* 366 * For nfs v3 or v4, check to see if we have done this recently, and if 367 * so return our cached result instead of making an ACCESS call. 368 * If not, do an access rpc, otherwise you are stuck emulating 369 * ufs_access() locally using the vattr. This may not be correct, 370 * since the server may apply other access criteria such as 371 * client uid-->server uid mapping that we do not know about. 372 */ 373 if (v34) { 374 if (ap->a_accmode & VREAD) 375 mode = NFSACCESS_READ; 376 else 377 mode = 0; 378 if (vp->v_type != VDIR) { 379 if (ap->a_accmode & VWRITE) 380 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 381 if (ap->a_accmode & VAPPEND) 382 mode |= NFSACCESS_EXTEND; 383 if (ap->a_accmode & VEXEC) 384 mode |= NFSACCESS_EXECUTE; 385 if (ap->a_accmode & VDELETE) 386 mode |= NFSACCESS_DELETE; 387 } else { 388 if (ap->a_accmode & VWRITE) 389 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 390 if (ap->a_accmode & VAPPEND) 391 mode |= NFSACCESS_EXTEND; 392 if (ap->a_accmode & VEXEC) 393 mode |= NFSACCESS_LOOKUP; 394 if (ap->a_accmode & VDELETE) 395 mode |= NFSACCESS_DELETE; 396 if (ap->a_accmode & VDELETE_CHILD) 397 mode |= NFSACCESS_MODIFY; 398 } 399 /* XXX safety belt, only make blanket request if caching */ 400 if (nfsaccess_cache_timeout > 0) { 401 wmode = NFSACCESS_READ | NFSACCESS_MODIFY | 402 NFSACCESS_EXTEND | NFSACCESS_EXECUTE | 403 NFSACCESS_DELETE | NFSACCESS_LOOKUP; 404 } else { 405 wmode = mode; 406 } 407 408 /* 409 * Does our cached result allow us to give a definite yes to 410 * this request? 411 */ 412 gotahit = 0; 413 mtx_lock(&np->n_mtx); 414 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 415 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) { 416 if (time_second < (np->n_accesscache[i].stamp 417 + nfsaccess_cache_timeout) && 418 (np->n_accesscache[i].mode & mode) == mode) { 419 NFSINCRGLOBAL(newnfsstats.accesscache_hits); 420 gotahit = 1; 421 } 422 break; 423 } 424 } 425 mtx_unlock(&np->n_mtx); 426#ifdef KDTRACE_HOOKS 427 if (gotahit != 0) 428 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp, 429 ap->a_cred->cr_uid, mode); 430 else 431 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp, 432 ap->a_cred->cr_uid, mode); 433#endif 434 if (gotahit == 0) { 435 /* 436 * Either a no, or a don't know. Go to the wire. 437 */ 438 NFSINCRGLOBAL(newnfsstats.accesscache_misses); 439 error = nfs34_access_otw(vp, wmode, ap->a_td, 440 ap->a_cred, &rmode); 441 if (!error && 442 (rmode & mode) != mode) 443 error = EACCES; 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 = ncl_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 = ncl_readdirrpc(vp, &auio, ap->a_cred, 483 ap->a_td); 484 free(bp, M_TEMP); 485 } else if (vp->v_type == VLNK) 486 error = ncl_readlinkrpc(vp, &auio, ap->a_cred); 487 else 488 error = EACCES; 489 } else 490 mtx_unlock(&np->n_mtx); 491 return (error); 492 } 493} 494 495 496/* 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; 512 513 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) 514 return (EOPNOTSUPP); 515 516 /* 517 * For NFSv4, we need to do the Open Op before cache validation, 518 * so that we conform to RFC3530 Sec. 9.3.1. 519 */ 520 if (NFS_ISV4(vp)) { 521 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td); 522 if (error) { 523 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 524 (gid_t)0); 525 return (error); 526 } 527 } 528 529 /* 530 * Now, if this Open will be doing reading, re-validate/flush the 531 * cache, so that Close/Open coherency is maintained. 532 */ 533 mtx_lock(&np->n_mtx); 534 if (np->n_flag & NMODIFIED) { 535 mtx_unlock(&np->n_mtx); 536 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 537 if (error == EINTR || error == EIO) { 538 if (NFS_ISV4(vp)) 539 (void) nfsrpc_close(vp, 0, ap->a_td); 540 return (error); 541 } 542 mtx_lock(&np->n_mtx); 543 np->n_attrstamp = 0; 544 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 545 if (vp->v_type == VDIR) 546 np->n_direofoffset = 0; 547 mtx_unlock(&np->n_mtx); 548 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 549 if (error) { 550 if (NFS_ISV4(vp)) 551 (void) nfsrpc_close(vp, 0, ap->a_td); 552 return (error); 553 } 554 mtx_lock(&np->n_mtx); 555 np->n_mtime = vattr.va_mtime; 556 if (NFS_ISV4(vp)) 557 np->n_change = vattr.va_filerev; 558 } else { 559 mtx_unlock(&np->n_mtx); 560 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 561 if (error) { 562 if (NFS_ISV4(vp)) 563 (void) nfsrpc_close(vp, 0, ap->a_td); 564 return (error); 565 } 566 mtx_lock(&np->n_mtx); 567 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) || 568 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 569 if (vp->v_type == VDIR) 570 np->n_direofoffset = 0; 571 mtx_unlock(&np->n_mtx); 572 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 573 if (error == EINTR || error == EIO) { 574 if (NFS_ISV4(vp)) 575 (void) nfsrpc_close(vp, 0, ap->a_td); 576 return (error); 577 } 578 mtx_lock(&np->n_mtx); 579 np->n_mtime = vattr.va_mtime; 580 if (NFS_ISV4(vp)) 581 np->n_change = vattr.va_filerev; 582 } 583 } 584 585 /* 586 * If the object has >= 1 O_DIRECT active opens, we disable caching. 587 */ 588 if (newnfs_directio_enable && (fmode & O_DIRECT) && 589 (vp->v_type == VREG)) { 590 if (np->n_directio_opens == 0) { 591 mtx_unlock(&np->n_mtx); 592 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 593 if (error) { 594 if (NFS_ISV4(vp)) 595 (void) nfsrpc_close(vp, 0, ap->a_td); 596 return (error); 597 } 598 mtx_lock(&np->n_mtx); 599 np->n_flag |= NNONCACHE; 600 } 601 np->n_directio_opens++; 602 } 603 mtx_unlock(&np->n_mtx); 604 vnode_create_vobject(vp, vattr.va_size, ap->a_td); 605 return (0); 606} 607 608/* 609 * nfs close vnode op 610 * What an NFS client should do upon close after writing is a debatable issue. 611 * Most NFS clients push delayed writes to the server upon close, basically for 612 * two reasons: 613 * 1 - So that any write errors may be reported back to the client process 614 * doing the close system call. By far the two most likely errors are 615 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 616 * 2 - To put a worst case upper bound on cache inconsistency between 617 * multiple clients for the file. 618 * There is also a consistency problem for Version 2 of the protocol w.r.t. 619 * not being able to tell if other clients are writing a file concurrently, 620 * since there is no way of knowing if the changed modify time in the reply 621 * is only due to the write for this client. 622 * (NFS Version 3 provides weak cache consistency data in the reply that 623 * should be sufficient to detect and handle this case.) 624 * 625 * The current code does the following: 626 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 627 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 628 * or commit them (this satisfies 1 and 2 except for the 629 * case where the server crashes after this close but 630 * before the commit RPC, which is felt to be "good 631 * enough". Changing the last argument to ncl_flush() to 632 * a 1 would force a commit operation, if it is felt a 633 * commit is necessary now. 634 * for NFS Version 4 - flush the dirty buffers and commit them, if 635 * nfscl_mustflush() says this is necessary. 636 * It is necessary if there is no write delegation held, 637 * in order to satisfy open/close coherency. 638 * If the file isn't cached on local stable storage, 639 * it may be necessary in order to detect "out of space" 640 * errors from the server, if the write delegation 641 * issued by the server doesn't allow the file to grow. 642 */ 643/* ARGSUSED */ 644static int 645nfs_close(struct vop_close_args *ap) 646{ 647 struct vnode *vp = ap->a_vp; 648 struct nfsnode *np = VTONFS(vp); 649 struct nfsvattr nfsva; 650 struct ucred *cred; 651 int error = 0, ret, localcred = 0; 652 int fmode = ap->a_fflag; 653 654 if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF)) 655 return (0); 656 /* 657 * During shutdown, a_cred isn't valid, so just use root. 658 */ 659 if (ap->a_cred == NOCRED) { 660 cred = newnfs_getcred(); 661 localcred = 1; 662 } else { 663 cred = ap->a_cred; 664 } 665 if (vp->v_type == VREG) { 666 /* 667 * Examine and clean dirty pages, regardless of NMODIFIED. 668 * This closes a major hole in close-to-open consistency. 669 * We want to push out all dirty pages (and buffers) on 670 * close, regardless of whether they were dirtied by 671 * mmap'ed writes or via write(). 672 */ 673 if (nfs_clean_pages_on_close && vp->v_object) { 674 VM_OBJECT_LOCK(vp->v_object); 675 vm_object_page_clean(vp->v_object, 0, 0, 0); 676 VM_OBJECT_UNLOCK(vp->v_object); 677 } 678 mtx_lock(&np->n_mtx); 679 if (np->n_flag & NMODIFIED) { 680 mtx_unlock(&np->n_mtx); 681 if (NFS_ISV3(vp)) { 682 /* 683 * Under NFSv3 we have dirty buffers to dispose of. We 684 * must flush them to the NFS server. We have the option 685 * of waiting all the way through the commit rpc or just 686 * waiting for the initial write. The default is to only 687 * wait through the initial write so the data is in the 688 * server's cache, which is roughly similar to the state 689 * a standard disk subsystem leaves the file in on close(). 690 * 691 * We cannot clear the NMODIFIED bit in np->n_flag due to 692 * potential races with other processes, and certainly 693 * cannot clear it if we don't commit. 694 * These races occur when there is no longer the old 695 * traditional vnode locking implemented for Vnode Ops. 696 */ 697 int cm = newnfs_commit_on_close ? 1 : 0; 698 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm, 0); 699 /* np->n_flag &= ~NMODIFIED; */ 700 } else if (NFS_ISV4(vp)) { 701 if (nfscl_mustflush(vp) != 0) { 702 int cm = newnfs_commit_on_close ? 1 : 0; 703 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, 704 cm, 0); 705 /* 706 * as above w.r.t races when clearing 707 * NMODIFIED. 708 * np->n_flag &= ~NMODIFIED; 709 */ 710 } 711 } else 712 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 713 mtx_lock(&np->n_mtx); 714 } 715 /* 716 * Invalidate the attribute cache in all cases. 717 * An open is going to fetch fresh attrs any way, other procs 718 * on this node that have file open will be forced to do an 719 * otw attr fetch, but this is safe. 720 * --> A user found that their RPC count dropped by 20% when 721 * this was commented out and I can't see any requirement 722 * for it, so I've disabled it when negative lookups are 723 * enabled. (What does this have to do with negative lookup 724 * caching? Well nothing, except it was reported by the 725 * same user that needed negative lookup caching and I wanted 726 * there to be a way to disable it to see if it 727 * is the cause of some caching/coherency issue that might 728 * crop up.) 729 */ 730 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) { 731 np->n_attrstamp = 0; 732 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 733 } 734 if (np->n_flag & NWRITEERR) { 735 np->n_flag &= ~NWRITEERR; 736 error = np->n_error; 737 } 738 mtx_unlock(&np->n_mtx); 739 } 740 741 if (NFS_ISV4(vp)) { 742 /* 743 * Get attributes so "change" is up to date. 744 */ 745 if (error == 0 && nfscl_mustflush(vp) != 0) { 746 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva, 747 NULL); 748 if (!ret) { 749 np->n_change = nfsva.na_filerev; 750 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 751 NULL, 0, 0); 752 } 753 } 754 755 /* 756 * and do the close. 757 */ 758 ret = nfsrpc_close(vp, 0, ap->a_td); 759 if (!error && ret) 760 error = ret; 761 if (error) 762 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 763 (gid_t)0); 764 } 765 if (newnfs_directio_enable) 766 KASSERT((np->n_directio_asyncwr == 0), 767 ("nfs_close: dirty unflushed (%d) directio buffers\n", 768 np->n_directio_asyncwr)); 769 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) { 770 mtx_lock(&np->n_mtx); 771 KASSERT((np->n_directio_opens > 0), 772 ("nfs_close: unexpectedly value (0) of n_directio_opens\n")); 773 np->n_directio_opens--; 774 if (np->n_directio_opens == 0) 775 np->n_flag &= ~NNONCACHE; 776 mtx_unlock(&np->n_mtx); 777 } 778 if (localcred) 779 NFSFREECRED(cred); 780 return (error); 781} 782 783/* 784 * nfs getattr call from vfs. 785 */ 786static int 787nfs_getattr(struct vop_getattr_args *ap) 788{ 789 struct vnode *vp = ap->a_vp; 790 struct thread *td = curthread; /* XXX */ 791 struct nfsnode *np = VTONFS(vp); 792 int error = 0; 793 struct nfsvattr nfsva; 794 struct vattr *vap = ap->a_vap; 795 struct vattr vattr; 796 797 /* 798 * Update local times for special files. 799 */ 800 mtx_lock(&np->n_mtx); 801 if (np->n_flag & (NACC | NUPD)) 802 np->n_flag |= NCHG; 803 mtx_unlock(&np->n_mtx); 804 /* 805 * First look in the cache. 806 */ 807 if (ncl_getattrcache(vp, &vattr) == 0) { 808 vap->va_type = vattr.va_type; 809 vap->va_mode = vattr.va_mode; 810 vap->va_nlink = vattr.va_nlink; 811 vap->va_uid = vattr.va_uid; 812 vap->va_gid = vattr.va_gid; 813 vap->va_fsid = vattr.va_fsid; 814 vap->va_fileid = vattr.va_fileid; 815 vap->va_size = vattr.va_size; 816 vap->va_blocksize = vattr.va_blocksize; 817 vap->va_atime = vattr.va_atime; 818 vap->va_mtime = vattr.va_mtime; 819 vap->va_ctime = vattr.va_ctime; 820 vap->va_gen = vattr.va_gen; 821 vap->va_flags = vattr.va_flags; 822 vap->va_rdev = vattr.va_rdev; 823 vap->va_bytes = vattr.va_bytes; 824 vap->va_filerev = vattr.va_filerev; 825 /* 826 * Get the local modify time for the case of a write 827 * delegation. 828 */ 829 nfscl_deleggetmodtime(vp, &vap->va_mtime); 830 return (0); 831 } 832 833 if (NFS_ISV34(vp) && nfs_prime_access_cache && 834 nfsaccess_cache_timeout > 0) { 835 NFSINCRGLOBAL(newnfsstats.accesscache_misses); 836 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL); 837 if (ncl_getattrcache(vp, ap->a_vap) == 0) { 838 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime); 839 return (0); 840 } 841 } 842 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL); 843 if (!error) 844 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0); 845 if (!error) { 846 /* 847 * Get the local modify time for the case of a write 848 * delegation. 849 */ 850 nfscl_deleggetmodtime(vp, &vap->va_mtime); 851 } else if (NFS_ISV4(vp)) { 852 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 853 } 854 return (error); 855} 856 857/* 858 * nfs setattr call. 859 */ 860static int 861nfs_setattr(struct vop_setattr_args *ap) 862{ 863 struct vnode *vp = ap->a_vp; 864 struct nfsnode *np = VTONFS(vp); 865 struct thread *td = curthread; /* XXX */ 866 struct vattr *vap = ap->a_vap; 867 int error = 0; 868 u_quad_t tsize; 869 870#ifndef nolint 871 tsize = (u_quad_t)0; 872#endif 873 874 /* 875 * Setting of flags and marking of atimes are not supported. 876 */ 877 if (vap->va_flags != VNOVAL) 878 return (EOPNOTSUPP); 879 880 /* 881 * Disallow write attempts if the filesystem is mounted read-only. 882 */ 883 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 884 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 885 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 886 (vp->v_mount->mnt_flag & MNT_RDONLY)) 887 return (EROFS); 888 if (vap->va_size != VNOVAL) { 889 switch (vp->v_type) { 890 case VDIR: 891 return (EISDIR); 892 case VCHR: 893 case VBLK: 894 case VSOCK: 895 case VFIFO: 896 if (vap->va_mtime.tv_sec == VNOVAL && 897 vap->va_atime.tv_sec == VNOVAL && 898 vap->va_mode == (mode_t)VNOVAL && 899 vap->va_uid == (uid_t)VNOVAL && 900 vap->va_gid == (gid_t)VNOVAL) 901 return (0); 902 vap->va_size = VNOVAL; 903 break; 904 default: 905 /* 906 * Disallow write attempts if the filesystem is 907 * mounted read-only. 908 */ 909 if (vp->v_mount->mnt_flag & MNT_RDONLY) 910 return (EROFS); 911 /* 912 * We run vnode_pager_setsize() early (why?), 913 * we must set np->n_size now to avoid vinvalbuf 914 * V_SAVE races that might setsize a lower 915 * value. 916 */ 917 mtx_lock(&np->n_mtx); 918 tsize = np->n_size; 919 mtx_unlock(&np->n_mtx); 920 error = ncl_meta_setsize(vp, ap->a_cred, td, 921 vap->va_size); 922 mtx_lock(&np->n_mtx); 923 if (np->n_flag & NMODIFIED) { 924 tsize = np->n_size; 925 mtx_unlock(&np->n_mtx); 926 if (vap->va_size == 0) 927 error = ncl_vinvalbuf(vp, 0, td, 1); 928 else 929 error = ncl_vinvalbuf(vp, V_SAVE, td, 1); 930 if (error) { 931 vnode_pager_setsize(vp, tsize); 932 return (error); 933 } 934 /* 935 * Call nfscl_delegmodtime() to set the modify time 936 * locally, as required. 937 */ 938 nfscl_delegmodtime(vp); 939 } else 940 mtx_unlock(&np->n_mtx); 941 /* 942 * np->n_size has already been set to vap->va_size 943 * in ncl_meta_setsize(). We must set it again since 944 * nfs_loadattrcache() could be called through 945 * ncl_meta_setsize() and could modify np->n_size. 946 */ 947 mtx_lock(&np->n_mtx); 948 np->n_vattr.na_size = np->n_size = vap->va_size; 949 mtx_unlock(&np->n_mtx); 950 }; 951 } else { 952 mtx_lock(&np->n_mtx); 953 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) && 954 (np->n_flag & NMODIFIED) && vp->v_type == VREG) { 955 mtx_unlock(&np->n_mtx); 956 if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 && 957 (error == EINTR || error == EIO)) 958 return (error); 959 } else 960 mtx_unlock(&np->n_mtx); 961 } 962 error = nfs_setattrrpc(vp, vap, ap->a_cred, td); 963 if (error && vap->va_size != VNOVAL) { 964 mtx_lock(&np->n_mtx); 965 np->n_size = np->n_vattr.na_size = tsize; 966 vnode_pager_setsize(vp, tsize); 967 mtx_unlock(&np->n_mtx); 968 } 969 return (error); 970} 971 972/* 973 * Do an nfs setattr rpc. 974 */ 975static int 976nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred, 977 struct thread *td) 978{ 979 struct nfsnode *np = VTONFS(vp); 980 int error, ret, attrflag, i; 981 struct nfsvattr nfsva; 982 983 if (NFS_ISV34(vp)) { 984 mtx_lock(&np->n_mtx); 985 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) 986 np->n_accesscache[i].stamp = 0; 987 np->n_flag |= NDELEGMOD; 988 mtx_unlock(&np->n_mtx); 989 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp); 990 } 991 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag, 992 NULL); 993 if (attrflag) { 994 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 995 if (ret && !error) 996 error = ret; 997 } 998 if (error && NFS_ISV4(vp)) 999 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid); 1000 return (error); 1001} 1002 1003/* 1004 * nfs lookup call, one step at a time... 1005 * First look in cache 1006 * If not found, unlock the directory nfsnode and do the rpc 1007 */ 1008static int 1009nfs_lookup(struct vop_lookup_args *ap) 1010{ 1011 struct componentname *cnp = ap->a_cnp; 1012 struct vnode *dvp = ap->a_dvp; 1013 struct vnode **vpp = ap->a_vpp; 1014 struct mount *mp = dvp->v_mount; 1015 int flags = cnp->cn_flags; 1016 struct vnode *newvp; 1017 struct nfsmount *nmp; 1018 struct nfsnode *np, *newnp; 1019 int error = 0, attrflag, dattrflag, ltype, ncticks; 1020 struct thread *td = cnp->cn_thread; 1021 struct nfsfh *nfhp; 1022 struct nfsvattr dnfsva, nfsva; 1023 struct vattr vattr; 1024 struct timespec nctime; 1025 1026 *vpp = NULLVP; 1027 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) && 1028 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 1029 return (EROFS); 1030 if (dvp->v_type != VDIR) 1031 return (ENOTDIR); 1032 nmp = VFSTONFS(mp); 1033 np = VTONFS(dvp); 1034 1035 /* For NFSv4, wait until any remove is done. */ 1036 mtx_lock(&np->n_mtx); 1037 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) { 1038 np->n_flag |= NREMOVEWANT; 1039 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0); 1040 } 1041 mtx_unlock(&np->n_mtx); 1042 1043 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) 1044 return (error); 1045 error = cache_lookup_times(dvp, vpp, cnp, &nctime, &ncticks); 1046 if (error > 0 && error != ENOENT) 1047 return (error); 1048 if (error == -1) { 1049 /* 1050 * Lookups of "." are special and always return the 1051 * current directory. cache_lookup() already handles 1052 * associated locking bookkeeping, etc. 1053 */ 1054 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 1055 /* XXX: Is this really correct? */ 1056 if (cnp->cn_nameiop != LOOKUP && 1057 (flags & ISLASTCN)) 1058 cnp->cn_flags |= SAVENAME; 1059 return (0); 1060 } 1061 1062 /* 1063 * We only accept a positive hit in the cache if the 1064 * change time of the file matches our cached copy. 1065 * Otherwise, we discard the cache entry and fallback 1066 * to doing a lookup RPC. We also only trust cache 1067 * entries for less than nm_nametimeo seconds. 1068 * 1069 * To better handle stale file handles and attributes, 1070 * clear the attribute cache of this node if it is a 1071 * leaf component, part of an open() call, and not 1072 * locally modified before fetching the attributes. 1073 * This should allow stale file handles to be detected 1074 * here where we can fall back to a LOOKUP RPC to 1075 * recover rather than having nfs_open() detect the 1076 * stale file handle and failing open(2) with ESTALE. 1077 */ 1078 newvp = *vpp; 1079 newnp = VTONFS(newvp); 1080 if (!(nmp->nm_flag & NFSMNT_NOCTO) && 1081 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1082 !(newnp->n_flag & NMODIFIED)) { 1083 mtx_lock(&newnp->n_mtx); 1084 newnp->n_attrstamp = 0; 1085 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1086 mtx_unlock(&newnp->n_mtx); 1087 } 1088 if (nfscl_nodeleg(newvp, 0) == 0 || 1089 ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) && 1090 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 && 1091 timespeccmp(&vattr.va_ctime, &nctime, ==))) { 1092 NFSINCRGLOBAL(newnfsstats.lookupcache_hits); 1093 if (cnp->cn_nameiop != LOOKUP && 1094 (flags & ISLASTCN)) 1095 cnp->cn_flags |= SAVENAME; 1096 return (0); 1097 } 1098 cache_purge(newvp); 1099 if (dvp != newvp) 1100 vput(newvp); 1101 else 1102 vrele(newvp); 1103 *vpp = NULLVP; 1104 } else if (error == ENOENT) { 1105 if (dvp->v_iflag & VI_DOOMED) 1106 return (ENOENT); 1107 /* 1108 * We only accept a negative hit in the cache if the 1109 * modification time of the parent directory matches 1110 * the cached copy in the name cache entry. 1111 * Otherwise, we discard all of the negative cache 1112 * entries for this directory. We also only trust 1113 * negative cache entries for up to nm_negnametimeo 1114 * seconds. 1115 */ 1116 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) && 1117 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 && 1118 timespeccmp(&vattr.va_mtime, &nctime, ==)) { 1119 NFSINCRGLOBAL(newnfsstats.lookupcache_hits); 1120 return (ENOENT); 1121 } 1122 cache_purge_negative(dvp); 1123 } 1124 1125 error = 0; 1126 newvp = NULLVP; 1127 NFSINCRGLOBAL(newnfsstats.lookupcache_misses); 1128 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1129 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1130 NULL); 1131 if (dattrflag) 1132 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1133 if (error) { 1134 if (newvp != NULLVP) { 1135 vput(newvp); 1136 *vpp = NULLVP; 1137 } 1138 1139 if (error != ENOENT) { 1140 if (NFS_ISV4(dvp)) 1141 error = nfscl_maperr(td, error, (uid_t)0, 1142 (gid_t)0); 1143 return (error); 1144 } 1145 1146 /* The requested file was not found. */ 1147 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1148 (flags & ISLASTCN)) { 1149 /* 1150 * XXX: UFS does a full VOP_ACCESS(dvp, 1151 * VWRITE) here instead of just checking 1152 * MNT_RDONLY. 1153 */ 1154 if (mp->mnt_flag & MNT_RDONLY) 1155 return (EROFS); 1156 cnp->cn_flags |= SAVENAME; 1157 return (EJUSTRETURN); 1158 } 1159 1160 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE && 1161 dattrflag) { 1162 /* 1163 * Cache the modification time of the parent 1164 * directory from the post-op attributes in 1165 * the name cache entry. The negative cache 1166 * entry will be ignored once the directory 1167 * has changed. Don't bother adding the entry 1168 * if the directory has already changed. 1169 */ 1170 mtx_lock(&np->n_mtx); 1171 if (timespeccmp(&np->n_vattr.na_mtime, 1172 &dnfsva.na_mtime, ==)) { 1173 mtx_unlock(&np->n_mtx); 1174 cache_enter_time(dvp, NULL, cnp, 1175 &dnfsva.na_mtime); 1176 } else 1177 mtx_unlock(&np->n_mtx); 1178 } 1179 return (ENOENT); 1180 } 1181 1182 /* 1183 * Handle RENAME case... 1184 */ 1185 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 1186 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1187 FREE((caddr_t)nfhp, M_NFSFH); 1188 return (EISDIR); 1189 } 1190 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1191 LK_EXCLUSIVE); 1192 if (error) 1193 return (error); 1194 newvp = NFSTOV(np); 1195 if (attrflag) 1196 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1197 0, 1); 1198 *vpp = newvp; 1199 cnp->cn_flags |= SAVENAME; 1200 return (0); 1201 } 1202 1203 if (flags & ISDOTDOT) { 1204 ltype = NFSVOPISLOCKED(dvp); 1205 error = vfs_busy(mp, MBF_NOWAIT); 1206 if (error != 0) { 1207 vfs_ref(mp); 1208 NFSVOPUNLOCK(dvp, 0); 1209 error = vfs_busy(mp, 0); 1210 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1211 vfs_rel(mp); 1212 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) { 1213 vfs_unbusy(mp); 1214 error = ENOENT; 1215 } 1216 if (error != 0) 1217 return (error); 1218 } 1219 NFSVOPUNLOCK(dvp, 0); 1220 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1221 cnp->cn_lkflags); 1222 if (error == 0) 1223 newvp = NFSTOV(np); 1224 vfs_unbusy(mp); 1225 if (newvp != dvp) 1226 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1227 if (dvp->v_iflag & VI_DOOMED) { 1228 if (error == 0) { 1229 if (newvp == dvp) 1230 vrele(newvp); 1231 else 1232 vput(newvp); 1233 } 1234 error = ENOENT; 1235 } 1236 if (error != 0) 1237 return (error); 1238 if (attrflag) 1239 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1240 0, 1); 1241 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1242 FREE((caddr_t)nfhp, M_NFSFH); 1243 VREF(dvp); 1244 newvp = dvp; 1245 if (attrflag) 1246 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1247 0, 1); 1248 } else { 1249 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1250 cnp->cn_lkflags); 1251 if (error) 1252 return (error); 1253 newvp = NFSTOV(np); 1254 if (attrflag) 1255 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1256 0, 1); 1257 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1258 !(np->n_flag & NMODIFIED)) { 1259 /* 1260 * Flush the attribute cache when opening a 1261 * leaf node to ensure that fresh attributes 1262 * are fetched in nfs_open() since we did not 1263 * fetch attributes from the LOOKUP reply. 1264 */ 1265 mtx_lock(&np->n_mtx); 1266 np->n_attrstamp = 0; 1267 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1268 mtx_unlock(&np->n_mtx); 1269 } 1270 } 1271 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1272 cnp->cn_flags |= SAVENAME; 1273 if ((cnp->cn_flags & MAKEENTRY) && 1274 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) && attrflag) { 1275 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime); 1276 } 1277 *vpp = newvp; 1278 return (0); 1279} 1280 1281/* 1282 * nfs read call. 1283 * Just call ncl_bioread() to do the work. 1284 */ 1285static int 1286nfs_read(struct vop_read_args *ap) 1287{ 1288 struct vnode *vp = ap->a_vp; 1289 1290 switch (vp->v_type) { 1291 case VREG: 1292 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred)); 1293 case VDIR: 1294 return (EISDIR); 1295 default: 1296 return (EOPNOTSUPP); 1297 } 1298} 1299 1300/* 1301 * nfs readlink call 1302 */ 1303static int 1304nfs_readlink(struct vop_readlink_args *ap) 1305{ 1306 struct vnode *vp = ap->a_vp; 1307 1308 if (vp->v_type != VLNK) 1309 return (EINVAL); 1310 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred)); 1311} 1312 1313/* 1314 * Do a readlink rpc. 1315 * Called by ncl_doio() from below the buffer cache. 1316 */ 1317int 1318ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1319{ 1320 int error, ret, attrflag; 1321 struct nfsvattr nfsva; 1322 1323 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva, 1324 &attrflag, NULL); 1325 if (attrflag) { 1326 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1327 if (ret && !error) 1328 error = ret; 1329 } 1330 if (error && NFS_ISV4(vp)) 1331 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1332 return (error); 1333} 1334 1335/* 1336 * nfs read rpc call 1337 * Ditto above 1338 */ 1339int 1340ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1341{ 1342 int error, ret, attrflag; 1343 struct nfsvattr nfsva; 1344 1345 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, &attrflag, 1346 NULL); 1347 if (attrflag) { 1348 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1349 if (ret && !error) 1350 error = ret; 1351 } 1352 if (error && NFS_ISV4(vp)) 1353 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1354 return (error); 1355} 1356 1357/* 1358 * nfs write call 1359 */ 1360int 1361ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 1362 int *iomode, int *must_commit, int called_from_strategy) 1363{ 1364 struct nfsvattr nfsva; 1365 int error = 0, attrflag, ret; 1366 1367 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred, 1368 uiop->uio_td, &nfsva, &attrflag, NULL, called_from_strategy); 1369 if (attrflag) { 1370 if (VTONFS(vp)->n_flag & ND_NFSV4) 1371 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1, 1372 1); 1373 else 1374 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1375 1); 1376 if (ret && !error) 1377 error = ret; 1378 } 1379 if (vp->v_mount->mnt_kern_flag & MNTK_ASYNC) 1380 *iomode = NFSWRITE_FILESYNC; 1381 if (error && NFS_ISV4(vp)) 1382 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1383 return (error); 1384} 1385 1386/* 1387 * nfs mknod rpc 1388 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1389 * mode set to specify the file type and the size field for rdev. 1390 */ 1391static int 1392nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1393 struct vattr *vap) 1394{ 1395 struct nfsvattr nfsva, dnfsva; 1396 struct vnode *newvp = NULL; 1397 struct nfsnode *np = NULL, *dnp; 1398 struct nfsfh *nfhp; 1399 struct vattr vattr; 1400 int error = 0, attrflag, dattrflag; 1401 u_int32_t rdev; 1402 1403 if (vap->va_type == VCHR || vap->va_type == VBLK) 1404 rdev = vap->va_rdev; 1405 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1406 rdev = 0xffffffff; 1407 else 1408 return (EOPNOTSUPP); 1409 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1410 return (error); 1411 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap, 1412 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva, 1413 &nfsva, &nfhp, &attrflag, &dattrflag, NULL); 1414 if (!error) { 1415 if (!nfhp) 1416 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1417 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, 1418 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1419 NULL); 1420 if (nfhp) 1421 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1422 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); 1423 } 1424 if (dattrflag) 1425 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1426 if (!error) { 1427 newvp = NFSTOV(np); 1428 if (attrflag != 0) { 1429 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1430 0, 1); 1431 if (error != 0) 1432 vput(newvp); 1433 } 1434 } 1435 if (!error) { 1436 *vpp = newvp; 1437 } else if (NFS_ISV4(dvp)) { 1438 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, 1439 vap->va_gid); 1440 } 1441 dnp = VTONFS(dvp); 1442 mtx_lock(&dnp->n_mtx); 1443 dnp->n_flag |= NMODIFIED; 1444 if (!dattrflag) { 1445 dnp->n_attrstamp = 0; 1446 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1447 } 1448 mtx_unlock(&dnp->n_mtx); 1449 return (error); 1450} 1451 1452/* 1453 * nfs mknod vop 1454 * just call nfs_mknodrpc() to do the work. 1455 */ 1456/* ARGSUSED */ 1457static int 1458nfs_mknod(struct vop_mknod_args *ap) 1459{ 1460 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap)); 1461} 1462 1463static struct mtx nfs_cverf_mtx; 1464MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex", 1465 MTX_DEF); 1466 1467static nfsquad_t 1468nfs_get_cverf(void) 1469{ 1470 static nfsquad_t cverf; 1471 nfsquad_t ret; 1472 static int cverf_initialized = 0; 1473 1474 mtx_lock(&nfs_cverf_mtx); 1475 if (cverf_initialized == 0) { 1476 cverf.lval[0] = arc4random(); 1477 cverf.lval[1] = arc4random(); 1478 cverf_initialized = 1; 1479 } else 1480 cverf.qval++; 1481 ret = cverf; 1482 mtx_unlock(&nfs_cverf_mtx); 1483 1484 return (ret); 1485} 1486 1487/* 1488 * nfs file create call 1489 */ 1490static int 1491nfs_create(struct vop_create_args *ap) 1492{ 1493 struct vnode *dvp = ap->a_dvp; 1494 struct vattr *vap = ap->a_vap; 1495 struct componentname *cnp = ap->a_cnp; 1496 struct nfsnode *np = NULL, *dnp; 1497 struct vnode *newvp = NULL; 1498 struct nfsmount *nmp; 1499 struct nfsvattr dnfsva, nfsva; 1500 struct nfsfh *nfhp; 1501 nfsquad_t cverf; 1502 int error = 0, attrflag, dattrflag, fmode = 0; 1503 struct vattr vattr; 1504 1505 /* 1506 * Oops, not for me.. 1507 */ 1508 if (vap->va_type == VSOCK) 1509 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1510 1511 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1512 return (error); 1513 if (vap->va_vaflags & VA_EXCLUSIVE) 1514 fmode |= O_EXCL; 1515 dnp = VTONFS(dvp); 1516 nmp = VFSTONFS(vnode_mount(dvp)); 1517again: 1518 /* For NFSv4, wait until any remove is done. */ 1519 mtx_lock(&dnp->n_mtx); 1520 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) { 1521 dnp->n_flag |= NREMOVEWANT; 1522 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0); 1523 } 1524 mtx_unlock(&dnp->n_mtx); 1525 1526 cverf = nfs_get_cverf(); 1527 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1528 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, 1529 &nfhp, &attrflag, &dattrflag, NULL); 1530 if (!error) { 1531 if (nfhp == NULL) 1532 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1533 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, 1534 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1535 NULL); 1536 if (nfhp != NULL) 1537 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1538 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); 1539 } 1540 if (dattrflag) 1541 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1542 if (!error) { 1543 newvp = NFSTOV(np); 1544 if (attrflag) 1545 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1546 0, 1); 1547 } 1548 if (error) { 1549 if (newvp != NULL) { 1550 vput(newvp); 1551 newvp = NULL; 1552 } 1553 if (NFS_ISV34(dvp) && (fmode & O_EXCL) && 1554 error == NFSERR_NOTSUPP) { 1555 fmode &= ~O_EXCL; 1556 goto again; 1557 } 1558 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) { 1559 if (nfscl_checksattr(vap, &nfsva)) { 1560 /* 1561 * We are normally called with only a partially 1562 * initialized VAP. Since the NFSv3 spec says that 1563 * the server may use the file attributes to 1564 * store the verifier, the spec requires us to do a 1565 * SETATTR RPC. FreeBSD servers store the verifier in 1566 * atime, but we can't really assume that all servers 1567 * will so we ensure that our SETATTR sets both atime 1568 * and mtime. 1569 */ 1570 if (vap->va_mtime.tv_sec == VNOVAL) 1571 vfs_timestamp(&vap->va_mtime); 1572 if (vap->va_atime.tv_sec == VNOVAL) 1573 vap->va_atime = vap->va_mtime; 1574 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred, 1575 cnp->cn_thread, &nfsva, &attrflag, NULL); 1576 if (error && (vap->va_uid != (uid_t)VNOVAL || 1577 vap->va_gid != (gid_t)VNOVAL)) { 1578 /* try again without setting uid/gid */ 1579 vap->va_uid = (uid_t)VNOVAL; 1580 vap->va_gid = (uid_t)VNOVAL; 1581 error = nfsrpc_setattr(newvp, vap, NULL, 1582 cnp->cn_cred, cnp->cn_thread, &nfsva, 1583 &attrflag, NULL); 1584 } 1585 if (attrflag) 1586 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 1587 NULL, 0, 1); 1588 if (error != 0) 1589 vput(newvp); 1590 } 1591 } 1592 if (!error) { 1593 if ((cnp->cn_flags & MAKEENTRY) && attrflag) 1594 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime); 1595 *ap->a_vpp = newvp; 1596 } else if (NFS_ISV4(dvp)) { 1597 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, 1598 vap->va_gid); 1599 } 1600 mtx_lock(&dnp->n_mtx); 1601 dnp->n_flag |= NMODIFIED; 1602 if (!dattrflag) { 1603 dnp->n_attrstamp = 0; 1604 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1605 } 1606 mtx_unlock(&dnp->n_mtx); 1607 return (error); 1608} 1609 1610/* 1611 * nfs file remove call 1612 * To try and make nfs semantics closer to ufs semantics, a file that has 1613 * other processes using the vnode is renamed instead of removed and then 1614 * removed later on the last close. 1615 * - If v_usecount > 1 1616 * If a rename is not already in the works 1617 * call nfs_sillyrename() to set it up 1618 * else 1619 * do the remove rpc 1620 */ 1621static int 1622nfs_remove(struct vop_remove_args *ap) 1623{ 1624 struct vnode *vp = ap->a_vp; 1625 struct vnode *dvp = ap->a_dvp; 1626 struct componentname *cnp = ap->a_cnp; 1627 struct nfsnode *np = VTONFS(vp); 1628 int error = 0; 1629 struct vattr vattr; 1630 1631 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name")); 1632 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount")); 1633 if (vp->v_type == VDIR) 1634 error = EPERM; 1635 else if (vrefcnt(vp) == 1 || (np->n_sillyrename && 1636 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 1637 vattr.va_nlink > 1)) { 1638 /* 1639 * Purge the name cache so that the chance of a lookup for 1640 * the name succeeding while the remove is in progress is 1641 * minimized. Without node locking it can still happen, such 1642 * that an I/O op returns ESTALE, but since you get this if 1643 * another host removes the file.. 1644 */ 1645 cache_purge(vp); 1646 /* 1647 * throw away biocache buffers, mainly to avoid 1648 * unnecessary delayed writes later. 1649 */ 1650 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1); 1651 /* Do the rpc */ 1652 if (error != EINTR && error != EIO) 1653 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr, 1654 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread); 1655 /* 1656 * Kludge City: If the first reply to the remove rpc is lost.. 1657 * the reply to the retransmitted request will be ENOENT 1658 * since the file was in fact removed 1659 * Therefore, we cheat and return success. 1660 */ 1661 if (error == ENOENT) 1662 error = 0; 1663 } else if (!np->n_sillyrename) 1664 error = nfs_sillyrename(dvp, vp, cnp); 1665 mtx_lock(&np->n_mtx); 1666 np->n_attrstamp = 0; 1667 mtx_unlock(&np->n_mtx); 1668 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 1669 return (error); 1670} 1671 1672/* 1673 * nfs file remove rpc called from nfs_inactive 1674 */ 1675int 1676ncl_removeit(struct sillyrename *sp, struct vnode *vp) 1677{ 1678 /* 1679 * Make sure that the directory vnode is still valid. 1680 * XXX we should lock sp->s_dvp here. 1681 */ 1682 if (sp->s_dvp->v_type == VBAD) 1683 return (0); 1684 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen, 1685 sp->s_cred, NULL)); 1686} 1687 1688/* 1689 * Nfs remove rpc, called from nfs_remove() and ncl_removeit(). 1690 */ 1691static int 1692nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 1693 int namelen, struct ucred *cred, struct thread *td) 1694{ 1695 struct nfsvattr dnfsva; 1696 struct nfsnode *dnp = VTONFS(dvp); 1697 int error = 0, dattrflag; 1698 1699 mtx_lock(&dnp->n_mtx); 1700 dnp->n_flag |= NREMOVEINPROG; 1701 mtx_unlock(&dnp->n_mtx); 1702 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva, 1703 &dattrflag, NULL); 1704 mtx_lock(&dnp->n_mtx); 1705 if ((dnp->n_flag & NREMOVEWANT)) { 1706 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG); 1707 mtx_unlock(&dnp->n_mtx); 1708 wakeup((caddr_t)dnp); 1709 } else { 1710 dnp->n_flag &= ~NREMOVEINPROG; 1711 mtx_unlock(&dnp->n_mtx); 1712 } 1713 if (dattrflag) 1714 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1715 mtx_lock(&dnp->n_mtx); 1716 dnp->n_flag |= NMODIFIED; 1717 if (!dattrflag) { 1718 dnp->n_attrstamp = 0; 1719 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1720 } 1721 mtx_unlock(&dnp->n_mtx); 1722 if (error && NFS_ISV4(dvp)) 1723 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1724 return (error); 1725} 1726 1727/* 1728 * nfs file rename call 1729 */ 1730static int 1731nfs_rename(struct vop_rename_args *ap) 1732{ 1733 struct vnode *fvp = ap->a_fvp; 1734 struct vnode *tvp = ap->a_tvp; 1735 struct vnode *fdvp = ap->a_fdvp; 1736 struct vnode *tdvp = ap->a_tdvp; 1737 struct componentname *tcnp = ap->a_tcnp; 1738 struct componentname *fcnp = ap->a_fcnp; 1739 struct nfsnode *fnp = VTONFS(ap->a_fvp); 1740 struct nfsnode *tdnp = VTONFS(ap->a_tdvp); 1741 struct nfsv4node *newv4 = NULL; 1742 int error; 1743 1744 KASSERT((tcnp->cn_flags & HASBUF) != 0 && 1745 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name")); 1746 /* Check for cross-device rename */ 1747 if ((fvp->v_mount != tdvp->v_mount) || 1748 (tvp && (fvp->v_mount != tvp->v_mount))) { 1749 error = EXDEV; 1750 goto out; 1751 } 1752 1753 if (fvp == tvp) { 1754 ncl_printf("nfs_rename: fvp == tvp (can't happen)\n"); 1755 error = 0; 1756 goto out; 1757 } 1758 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0) 1759 goto out; 1760 1761 /* 1762 * We have to flush B_DELWRI data prior to renaming 1763 * the file. If we don't, the delayed-write buffers 1764 * can be flushed out later after the file has gone stale 1765 * under NFSV3. NFSV2 does not have this problem because 1766 * ( as far as I can tell ) it flushes dirty buffers more 1767 * often. 1768 * 1769 * Skip the rename operation if the fsync fails, this can happen 1770 * due to the server's volume being full, when we pushed out data 1771 * that was written back to our cache earlier. Not checking for 1772 * this condition can result in potential (silent) data loss. 1773 */ 1774 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread); 1775 NFSVOPUNLOCK(fvp, 0); 1776 if (!error && tvp) 1777 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread); 1778 if (error) 1779 goto out; 1780 1781 /* 1782 * If the tvp exists and is in use, sillyrename it before doing the 1783 * rename of the new file over it. 1784 * XXX Can't sillyrename a directory. 1785 */ 1786 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename && 1787 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1788 vput(tvp); 1789 tvp = NULL; 1790 } 1791 1792 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1793 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1794 tcnp->cn_thread); 1795 1796 if (error == 0 && NFS_ISV4(tdvp)) { 1797 /* 1798 * For NFSv4, check to see if it is the same name and 1799 * replace the name, if it is different. 1800 */ 1801 MALLOC(newv4, struct nfsv4node *, 1802 sizeof (struct nfsv4node) + 1803 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1, 1804 M_NFSV4NODE, M_WAITOK); 1805 mtx_lock(&tdnp->n_mtx); 1806 mtx_lock(&fnp->n_mtx); 1807 if (fnp->n_v4 != NULL && fvp->v_type == VREG && 1808 (fnp->n_v4->n4_namelen != tcnp->cn_namelen || 1809 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4), 1810 tcnp->cn_namelen) || 1811 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen || 1812 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 1813 tdnp->n_fhp->nfh_len))) { 1814#ifdef notdef 1815{ char nnn[100]; int nnnl; 1816nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99; 1817bcopy(tcnp->cn_nameptr, nnn, nnnl); 1818nnn[nnnl] = '\0'; 1819printf("ren replace=%s\n",nnn); 1820} 1821#endif 1822 FREE((caddr_t)fnp->n_v4, M_NFSV4NODE); 1823 fnp->n_v4 = newv4; 1824 newv4 = NULL; 1825 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len; 1826 fnp->n_v4->n4_namelen = tcnp->cn_namelen; 1827 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 1828 tdnp->n_fhp->nfh_len); 1829 NFSBCOPY(tcnp->cn_nameptr, 1830 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen); 1831 } 1832 mtx_unlock(&tdnp->n_mtx); 1833 mtx_unlock(&fnp->n_mtx); 1834 if (newv4 != NULL) 1835 FREE((caddr_t)newv4, M_NFSV4NODE); 1836 } 1837 1838 if (fvp->v_type == VDIR) { 1839 if (tvp != NULL && tvp->v_type == VDIR) 1840 cache_purge(tdvp); 1841 cache_purge(fdvp); 1842 } 1843 1844out: 1845 if (tdvp == tvp) 1846 vrele(tdvp); 1847 else 1848 vput(tdvp); 1849 if (tvp) 1850 vput(tvp); 1851 vrele(fdvp); 1852 vrele(fvp); 1853 /* 1854 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1855 */ 1856 if (error == ENOENT) 1857 error = 0; 1858 return (error); 1859} 1860 1861/* 1862 * nfs file rename rpc called from nfs_remove() above 1863 */ 1864static int 1865nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp, 1866 struct sillyrename *sp) 1867{ 1868 1869 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen, 1870 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred, 1871 scnp->cn_thread)); 1872} 1873 1874/* 1875 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1876 */ 1877static int 1878nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr, 1879 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr, 1880 int tnamelen, struct ucred *cred, struct thread *td) 1881{ 1882 struct nfsvattr fnfsva, tnfsva; 1883 struct nfsnode *fdnp = VTONFS(fdvp); 1884 struct nfsnode *tdnp = VTONFS(tdvp); 1885 int error = 0, fattrflag, tattrflag; 1886 1887 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp, 1888 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag, 1889 &tattrflag, NULL, NULL); 1890 mtx_lock(&fdnp->n_mtx); 1891 fdnp->n_flag |= NMODIFIED; 1892 if (fattrflag != 0) { 1893 mtx_unlock(&fdnp->n_mtx); 1894 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1); 1895 } else { 1896 fdnp->n_attrstamp = 0; 1897 mtx_unlock(&fdnp->n_mtx); 1898 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp); 1899 } 1900 mtx_lock(&tdnp->n_mtx); 1901 tdnp->n_flag |= NMODIFIED; 1902 if (tattrflag != 0) { 1903 mtx_unlock(&tdnp->n_mtx); 1904 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1); 1905 } else { 1906 tdnp->n_attrstamp = 0; 1907 mtx_unlock(&tdnp->n_mtx); 1908 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 1909 } 1910 if (error && NFS_ISV4(fdvp)) 1911 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1912 return (error); 1913} 1914 1915/* 1916 * nfs hard link create call 1917 */ 1918static int 1919nfs_link(struct vop_link_args *ap) 1920{ 1921 struct vnode *vp = ap->a_vp; 1922 struct vnode *tdvp = ap->a_tdvp; 1923 struct componentname *cnp = ap->a_cnp; 1924 struct nfsnode *np, *tdnp; 1925 struct nfsvattr nfsva, dnfsva; 1926 int error = 0, attrflag, dattrflag; 1927 1928 if (vp->v_mount != tdvp->v_mount) { 1929 return (EXDEV); 1930 } 1931 1932 /* 1933 * Push all writes to the server, so that the attribute cache 1934 * doesn't get "out of sync" with the server. 1935 * XXX There should be a better way! 1936 */ 1937 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread); 1938 1939 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 1940 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag, 1941 &dattrflag, NULL); 1942 tdnp = VTONFS(tdvp); 1943 mtx_lock(&tdnp->n_mtx); 1944 tdnp->n_flag |= NMODIFIED; 1945 if (dattrflag != 0) { 1946 mtx_unlock(&tdnp->n_mtx); 1947 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1); 1948 } else { 1949 tdnp->n_attrstamp = 0; 1950 mtx_unlock(&tdnp->n_mtx); 1951 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 1952 } 1953 if (attrflag) 1954 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1955 else { 1956 np = VTONFS(vp); 1957 mtx_lock(&np->n_mtx); 1958 np->n_attrstamp = 0; 1959 mtx_unlock(&np->n_mtx); 1960 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 1961 } 1962 /* 1963 * If negative lookup caching is enabled, I might as well 1964 * add an entry for this node. Not necessary for correctness, 1965 * but if negative caching is enabled, then the system 1966 * must care about lookup caching hit rate, so... 1967 */ 1968 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 && 1969 (cnp->cn_flags & MAKEENTRY) && dattrflag) { 1970 cache_enter_time(tdvp, vp, cnp, &dnfsva.na_mtime); 1971 } 1972 if (error && NFS_ISV4(vp)) 1973 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, 1974 (gid_t)0); 1975 return (error); 1976} 1977 1978/* 1979 * nfs symbolic link create call 1980 */ 1981static int 1982nfs_symlink(struct vop_symlink_args *ap) 1983{ 1984 struct vnode *dvp = ap->a_dvp; 1985 struct vattr *vap = ap->a_vap; 1986 struct componentname *cnp = ap->a_cnp; 1987 struct nfsvattr nfsva, dnfsva; 1988 struct nfsfh *nfhp; 1989 struct nfsnode *np = NULL, *dnp; 1990 struct vnode *newvp = NULL; 1991 int error = 0, attrflag, dattrflag, ret; 1992 1993 vap->va_type = VLNK; 1994 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1995 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, 1996 &nfsva, &nfhp, &attrflag, &dattrflag, NULL); 1997 if (nfhp) { 1998 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, 1999 &np, NULL, LK_EXCLUSIVE); 2000 if (!ret) 2001 newvp = NFSTOV(np); 2002 else if (!error) 2003 error = ret; 2004 } 2005 if (newvp != NULL) { 2006 if (attrflag) 2007 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 2008 0, 1); 2009 } else if (!error) { 2010 /* 2011 * If we do not have an error and we could not extract the 2012 * newvp from the response due to the request being NFSv2, we 2013 * have to do a lookup in order to obtain a newvp to return. 2014 */ 2015 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2016 cnp->cn_cred, cnp->cn_thread, &np); 2017 if (!error) 2018 newvp = NFSTOV(np); 2019 } 2020 if (error) { 2021 if (newvp) 2022 vput(newvp); 2023 if (NFS_ISV4(dvp)) 2024 error = nfscl_maperr(cnp->cn_thread, error, 2025 vap->va_uid, vap->va_gid); 2026 } else { 2027 *ap->a_vpp = newvp; 2028 } 2029 2030 dnp = VTONFS(dvp); 2031 mtx_lock(&dnp->n_mtx); 2032 dnp->n_flag |= NMODIFIED; 2033 if (dattrflag != 0) { 2034 mtx_unlock(&dnp->n_mtx); 2035 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2036 /* 2037 * If negative lookup caching is enabled, I might as well 2038 * add an entry for this node. Not necessary for correctness, 2039 * but if negative caching is enabled, then the system 2040 * must care about lookup caching hit rate, so... 2041 */ 2042 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2043 (cnp->cn_flags & MAKEENTRY)) { 2044 cache_enter_time(dvp, newvp, cnp, &dnfsva.na_mtime); 2045 } 2046 } else { 2047 dnp->n_attrstamp = 0; 2048 mtx_unlock(&dnp->n_mtx); 2049 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2050 } 2051 return (error); 2052} 2053 2054/* 2055 * nfs make dir call 2056 */ 2057static int 2058nfs_mkdir(struct vop_mkdir_args *ap) 2059{ 2060 struct vnode *dvp = ap->a_dvp; 2061 struct vattr *vap = ap->a_vap; 2062 struct componentname *cnp = ap->a_cnp; 2063 struct nfsnode *np = NULL, *dnp; 2064 struct vnode *newvp = NULL; 2065 struct vattr vattr; 2066 struct nfsfh *nfhp; 2067 struct nfsvattr nfsva, dnfsva; 2068 int error = 0, attrflag, dattrflag, ret; 2069 2070 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) 2071 return (error); 2072 vap->va_type = VDIR; 2073 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2074 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp, 2075 &attrflag, &dattrflag, NULL); 2076 dnp = VTONFS(dvp); 2077 mtx_lock(&dnp->n_mtx); 2078 dnp->n_flag |= NMODIFIED; 2079 if (dattrflag != 0) { 2080 mtx_unlock(&dnp->n_mtx); 2081 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2082 } else { 2083 dnp->n_attrstamp = 0; 2084 mtx_unlock(&dnp->n_mtx); 2085 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2086 } 2087 if (nfhp) { 2088 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, 2089 &np, NULL, LK_EXCLUSIVE); 2090 if (!ret) { 2091 newvp = NFSTOV(np); 2092 if (attrflag) 2093 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 2094 NULL, 0, 1); 2095 } else if (!error) 2096 error = ret; 2097 } 2098 if (!error && newvp == NULL) { 2099 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2100 cnp->cn_cred, cnp->cn_thread, &np); 2101 if (!error) { 2102 newvp = NFSTOV(np); 2103 if (newvp->v_type != VDIR) 2104 error = EEXIST; 2105 } 2106 } 2107 if (error) { 2108 if (newvp) 2109 vput(newvp); 2110 if (NFS_ISV4(dvp)) 2111 error = nfscl_maperr(cnp->cn_thread, error, 2112 vap->va_uid, vap->va_gid); 2113 } else { 2114 /* 2115 * If negative lookup caching is enabled, I might as well 2116 * add an entry for this node. Not necessary for correctness, 2117 * but if negative caching is enabled, then the system 2118 * must care about lookup caching hit rate, so... 2119 */ 2120 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2121 (cnp->cn_flags & MAKEENTRY) && dattrflag) { 2122 cache_enter_time(dvp, newvp, cnp, &dnfsva.na_mtime); 2123 } 2124 *ap->a_vpp = newvp; 2125 } 2126 return (error); 2127} 2128 2129/* 2130 * nfs remove directory call 2131 */ 2132static int 2133nfs_rmdir(struct vop_rmdir_args *ap) 2134{ 2135 struct vnode *vp = ap->a_vp; 2136 struct vnode *dvp = ap->a_dvp; 2137 struct componentname *cnp = ap->a_cnp; 2138 struct nfsnode *dnp; 2139 struct nfsvattr dnfsva; 2140 int error, dattrflag; 2141 2142 if (dvp == vp) 2143 return (EINVAL); 2144 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2145 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL); 2146 dnp = VTONFS(dvp); 2147 mtx_lock(&dnp->n_mtx); 2148 dnp->n_flag |= NMODIFIED; 2149 if (dattrflag != 0) { 2150 mtx_unlock(&dnp->n_mtx); 2151 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2152 } else { 2153 dnp->n_attrstamp = 0; 2154 mtx_unlock(&dnp->n_mtx); 2155 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2156 } 2157 2158 cache_purge(dvp); 2159 cache_purge(vp); 2160 if (error && NFS_ISV4(dvp)) 2161 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, 2162 (gid_t)0); 2163 /* 2164 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2165 */ 2166 if (error == ENOENT) 2167 error = 0; 2168 return (error); 2169} 2170 2171/* 2172 * nfs readdir call 2173 */ 2174static int 2175nfs_readdir(struct vop_readdir_args *ap) 2176{ 2177 struct vnode *vp = ap->a_vp; 2178 struct nfsnode *np = VTONFS(vp); 2179 struct uio *uio = ap->a_uio; 2180 int tresid, error = 0; 2181 struct vattr vattr; 2182 2183 if (vp->v_type != VDIR) 2184 return(EPERM); 2185 2186 /* 2187 * First, check for hit on the EOF offset cache 2188 */ 2189 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2190 (np->n_flag & NMODIFIED) == 0) { 2191 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) { 2192 mtx_lock(&np->n_mtx); 2193 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) || 2194 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 2195 mtx_unlock(&np->n_mtx); 2196 NFSINCRGLOBAL(newnfsstats.direofcache_hits); 2197 return (0); 2198 } else 2199 mtx_unlock(&np->n_mtx); 2200 } 2201 } 2202 2203 /* 2204 * Call ncl_bioread() to do the real work. 2205 */ 2206 tresid = uio->uio_resid; 2207 error = ncl_bioread(vp, uio, 0, ap->a_cred); 2208 2209 if (!error && uio->uio_resid == tresid) 2210 NFSINCRGLOBAL(newnfsstats.direofcache_misses); 2211 return (error); 2212} 2213 2214/* 2215 * Readdir rpc call. 2216 * Called from below the buffer cache by ncl_doio(). 2217 */ 2218int 2219ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2220 struct thread *td) 2221{ 2222 struct nfsvattr nfsva; 2223 nfsuint64 *cookiep, cookie; 2224 struct nfsnode *dnp = VTONFS(vp); 2225 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2226 int error = 0, eof, attrflag; 2227 2228 KASSERT(uiop->uio_iovcnt == 1 && 2229 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2230 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2231 ("nfs readdirrpc bad uio")); 2232 2233 /* 2234 * If there is no cookie, assume directory was stale. 2235 */ 2236 ncl_dircookie_lock(dnp); 2237 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2238 if (cookiep) { 2239 cookie = *cookiep; 2240 ncl_dircookie_unlock(dnp); 2241 } else { 2242 ncl_dircookie_unlock(dnp); 2243 return (NFSERR_BAD_COOKIE); 2244 } 2245 2246 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2247 (void)ncl_fsinfo(nmp, vp, cred, td); 2248 2249 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva, 2250 &attrflag, &eof, NULL); 2251 if (attrflag) 2252 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2253 2254 if (!error) { 2255 /* 2256 * We are now either at the end of the directory or have filled 2257 * the block. 2258 */ 2259 if (eof) 2260 dnp->n_direofoffset = uiop->uio_offset; 2261 else { 2262 if (uiop->uio_resid > 0) 2263 ncl_printf("EEK! readdirrpc resid > 0\n"); 2264 ncl_dircookie_lock(dnp); 2265 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2266 *cookiep = cookie; 2267 ncl_dircookie_unlock(dnp); 2268 } 2269 } else if (NFS_ISV4(vp)) { 2270 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2271 } 2272 return (error); 2273} 2274 2275/* 2276 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc(). 2277 */ 2278int 2279ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2280 struct thread *td) 2281{ 2282 struct nfsvattr nfsva; 2283 nfsuint64 *cookiep, cookie; 2284 struct nfsnode *dnp = VTONFS(vp); 2285 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2286 int error = 0, attrflag, eof; 2287 2288 KASSERT(uiop->uio_iovcnt == 1 && 2289 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2290 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2291 ("nfs readdirplusrpc bad uio")); 2292 2293 /* 2294 * If there is no cookie, assume directory was stale. 2295 */ 2296 ncl_dircookie_lock(dnp); 2297 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2298 if (cookiep) { 2299 cookie = *cookiep; 2300 ncl_dircookie_unlock(dnp); 2301 } else { 2302 ncl_dircookie_unlock(dnp); 2303 return (NFSERR_BAD_COOKIE); 2304 } 2305 2306 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2307 (void)ncl_fsinfo(nmp, vp, cred, td); 2308 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva, 2309 &attrflag, &eof, NULL); 2310 if (attrflag) 2311 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2312 2313 if (!error) { 2314 /* 2315 * We are now either at end of the directory or have filled the 2316 * the block. 2317 */ 2318 if (eof) 2319 dnp->n_direofoffset = uiop->uio_offset; 2320 else { 2321 if (uiop->uio_resid > 0) 2322 ncl_printf("EEK! readdirplusrpc resid > 0\n"); 2323 ncl_dircookie_lock(dnp); 2324 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2325 *cookiep = cookie; 2326 ncl_dircookie_unlock(dnp); 2327 } 2328 } else if (NFS_ISV4(vp)) { 2329 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2330 } 2331 return (error); 2332} 2333 2334/* 2335 * Silly rename. To make the NFS filesystem that is stateless look a little 2336 * more like the "ufs" a remove of an active vnode is translated to a rename 2337 * to a funny looking filename that is removed by nfs_inactive on the 2338 * nfsnode. There is the potential for another process on a different client 2339 * to create the same funny name between the nfs_lookitup() fails and the 2340 * nfs_rename() completes, but... 2341 */ 2342static int 2343nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2344{ 2345 struct sillyrename *sp; 2346 struct nfsnode *np; 2347 int error; 2348 short pid; 2349 unsigned int lticks; 2350 2351 cache_purge(dvp); 2352 np = VTONFS(vp); 2353 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir")); 2354 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2355 M_NEWNFSREQ, M_WAITOK); 2356 sp->s_cred = crhold(cnp->cn_cred); 2357 sp->s_dvp = dvp; 2358 VREF(dvp); 2359 2360 /* 2361 * Fudge together a funny name. 2362 * Changing the format of the funny name to accomodate more 2363 * sillynames per directory. 2364 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is 2365 * CPU ticks since boot. 2366 */ 2367 pid = cnp->cn_thread->td_proc->p_pid; 2368 lticks = (unsigned int)ticks; 2369 for ( ; ; ) { 2370 sp->s_namlen = sprintf(sp->s_name, 2371 ".nfs.%08x.%04x4.4", lticks, 2372 pid); 2373 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2374 cnp->cn_thread, NULL)) 2375 break; 2376 lticks++; 2377 } 2378 error = nfs_renameit(dvp, vp, cnp, sp); 2379 if (error) 2380 goto bad; 2381 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2382 cnp->cn_thread, &np); 2383 np->n_sillyrename = sp; 2384 return (0); 2385bad: 2386 vrele(sp->s_dvp); 2387 crfree(sp->s_cred); 2388 free((caddr_t)sp, M_NEWNFSREQ); 2389 return (error); 2390} 2391 2392/* 2393 * Look up a file name and optionally either update the file handle or 2394 * allocate an nfsnode, depending on the value of npp. 2395 * npp == NULL --> just do the lookup 2396 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2397 * handled too 2398 * *npp != NULL --> update the file handle in the vnode 2399 */ 2400static int 2401nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred, 2402 struct thread *td, struct nfsnode **npp) 2403{ 2404 struct vnode *newvp = NULL, *vp; 2405 struct nfsnode *np, *dnp = VTONFS(dvp); 2406 struct nfsfh *nfhp, *onfhp; 2407 struct nfsvattr nfsva, dnfsva; 2408 struct componentname cn; 2409 int error = 0, attrflag, dattrflag; 2410 u_int hash; 2411 2412 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva, 2413 &nfhp, &attrflag, &dattrflag, NULL); 2414 if (dattrflag) 2415 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2416 if (npp && !error) { 2417 if (*npp != NULL) { 2418 np = *npp; 2419 vp = NFSTOV(np); 2420 /* 2421 * For NFSv4, check to see if it is the same name and 2422 * replace the name, if it is different. 2423 */ 2424 if (np->n_v4 != NULL && nfsva.na_type == VREG && 2425 (np->n_v4->n4_namelen != len || 2426 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) || 2427 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || 2428 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2429 dnp->n_fhp->nfh_len))) { 2430#ifdef notdef 2431{ char nnn[100]; int nnnl; 2432nnnl = (len < 100) ? len : 99; 2433bcopy(name, nnn, nnnl); 2434nnn[nnnl] = '\0'; 2435printf("replace=%s\n",nnn); 2436} 2437#endif 2438 FREE((caddr_t)np->n_v4, M_NFSV4NODE); 2439 MALLOC(np->n_v4, struct nfsv4node *, 2440 sizeof (struct nfsv4node) + 2441 dnp->n_fhp->nfh_len + len - 1, 2442 M_NFSV4NODE, M_WAITOK); 2443 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; 2444 np->n_v4->n4_namelen = len; 2445 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2446 dnp->n_fhp->nfh_len); 2447 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len); 2448 } 2449 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, 2450 FNV1_32_INIT); 2451 onfhp = np->n_fhp; 2452 /* 2453 * Rehash node for new file handle. 2454 */ 2455 vfs_hash_rehash(vp, hash); 2456 np->n_fhp = nfhp; 2457 if (onfhp != NULL) 2458 FREE((caddr_t)onfhp, M_NFSFH); 2459 newvp = NFSTOV(np); 2460 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) { 2461 FREE((caddr_t)nfhp, M_NFSFH); 2462 VREF(dvp); 2463 newvp = dvp; 2464 } else { 2465 cn.cn_nameptr = name; 2466 cn.cn_namelen = len; 2467 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td, 2468 &np, NULL, LK_EXCLUSIVE); 2469 if (error) 2470 return (error); 2471 newvp = NFSTOV(np); 2472 } 2473 if (!attrflag && *npp == NULL) { 2474 if (newvp == dvp) 2475 vrele(newvp); 2476 else 2477 vput(newvp); 2478 return (ENOENT); 2479 } 2480 if (attrflag) 2481 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 2482 0, 1); 2483 } 2484 if (npp && *npp == NULL) { 2485 if (error) { 2486 if (newvp) { 2487 if (newvp == dvp) 2488 vrele(newvp); 2489 else 2490 vput(newvp); 2491 } 2492 } else 2493 *npp = np; 2494 } 2495 if (error && NFS_ISV4(dvp)) 2496 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2497 return (error); 2498} 2499 2500/* 2501 * Nfs Version 3 and 4 commit rpc 2502 */ 2503int 2504ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred, 2505 struct thread *td) 2506{ 2507 struct nfsvattr nfsva; 2508 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2509 int error, attrflag; 2510 u_char verf[NFSX_VERF]; 2511 2512 mtx_lock(&nmp->nm_mtx); 2513 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) { 2514 mtx_unlock(&nmp->nm_mtx); 2515 return (0); 2516 } 2517 mtx_unlock(&nmp->nm_mtx); 2518 error = nfsrpc_commit(vp, offset, cnt, cred, td, verf, &nfsva, 2519 &attrflag, NULL); 2520 if (!error) { 2521 mtx_lock(&nmp->nm_mtx); 2522 if (NFSBCMP((caddr_t)nmp->nm_verf, verf, NFSX_VERF)) { 2523 NFSBCOPY(verf, (caddr_t)nmp->nm_verf, NFSX_VERF); 2524 error = NFSERR_STALEWRITEVERF; 2525 } 2526 mtx_unlock(&nmp->nm_mtx); 2527 if (!error && attrflag) 2528 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 2529 0, 1); 2530 } else if (NFS_ISV4(vp)) { 2531 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2532 } 2533 return (error); 2534} 2535 2536/* 2537 * Strategy routine. 2538 * For async requests when nfsiod(s) are running, queue the request by 2539 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the 2540 * request. 2541 */ 2542static int 2543nfs_strategy(struct vop_strategy_args *ap) 2544{ 2545 struct buf *bp = ap->a_bp; 2546 struct ucred *cr; 2547 2548 KASSERT(!(bp->b_flags & B_DONE), 2549 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); 2550 BUF_ASSERT_HELD(bp); 2551 2552 if (bp->b_iocmd == BIO_READ) 2553 cr = bp->b_rcred; 2554 else 2555 cr = bp->b_wcred; 2556 2557 /* 2558 * If the op is asynchronous and an i/o daemon is waiting 2559 * queue the request, wake it up and wait for completion 2560 * otherwise just do it ourselves. 2561 */ 2562 if ((bp->b_flags & B_ASYNC) == 0 || 2563 ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread)) 2564 (void) ncl_doio(ap->a_vp, bp, cr, curthread, 1); 2565 return (0); 2566} 2567 2568/* 2569 * fsync vnode op. Just call ncl_flush() with commit == 1. 2570 */ 2571/* ARGSUSED */ 2572static int 2573nfs_fsync(struct vop_fsync_args *ap) 2574{ 2575 2576 if (ap->a_vp->v_type != VREG) { 2577 /* 2578 * For NFS, metadata is changed synchronously on the server, 2579 * so there is nothing to flush. Also, ncl_flush() clears 2580 * the NMODIFIED flag and that shouldn't be done here for 2581 * directories. 2582 */ 2583 return (0); 2584 } 2585 return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1, 0)); 2586} 2587 2588/* 2589 * Flush all the blocks associated with a vnode. 2590 * Walk through the buffer pool and push any dirty pages 2591 * associated with the vnode. 2592 * If the called_from_renewthread argument is TRUE, it has been called 2593 * from the NFSv4 renew thread and, as such, cannot block indefinitely 2594 * waiting for a buffer write to complete. 2595 */ 2596int 2597ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td, 2598 int commit, int called_from_renewthread) 2599{ 2600 struct nfsnode *np = VTONFS(vp); 2601 struct buf *bp; 2602 int i; 2603 struct buf *nbp; 2604 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2605 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos; 2606 int passone = 1, trycnt = 0; 2607 u_quad_t off, endoff, toff; 2608 struct ucred* wcred = NULL; 2609 struct buf **bvec = NULL; 2610 struct bufobj *bo; 2611#ifndef NFS_COMMITBVECSIZ 2612#define NFS_COMMITBVECSIZ 20 2613#endif 2614 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ]; 2615 int bvecsize = 0, bveccount; 2616 2617 if (called_from_renewthread != 0) 2618 slptimeo = hz; 2619 if (nmp->nm_flag & NFSMNT_INT) 2620 slpflag = NFS_PCATCH; 2621 if (!commit) 2622 passone = 0; 2623 bo = &vp->v_bufobj; 2624 /* 2625 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the 2626 * server, but has not been committed to stable storage on the server 2627 * yet. On the first pass, the byte range is worked out and the commit 2628 * rpc is done. On the second pass, ncl_writebp() is called to do the 2629 * job. 2630 */ 2631again: 2632 off = (u_quad_t)-1; 2633 endoff = 0; 2634 bvecpos = 0; 2635 if (NFS_ISV34(vp) && commit) { 2636 if (bvec != NULL && bvec != bvec_on_stack) 2637 free(bvec, M_TEMP); 2638 /* 2639 * Count up how many buffers waiting for a commit. 2640 */ 2641 bveccount = 0; 2642 BO_LOCK(bo); 2643 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2644 if (!BUF_ISLOCKED(bp) && 2645 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) 2646 == (B_DELWRI | B_NEEDCOMMIT)) 2647 bveccount++; 2648 } 2649 /* 2650 * Allocate space to remember the list of bufs to commit. It is 2651 * important to use M_NOWAIT here to avoid a race with nfs_write. 2652 * If we can't get memory (for whatever reason), we will end up 2653 * committing the buffers one-by-one in the loop below. 2654 */ 2655 if (bveccount > NFS_COMMITBVECSIZ) { 2656 /* 2657 * Release the vnode interlock to avoid a lock 2658 * order reversal. 2659 */ 2660 BO_UNLOCK(bo); 2661 bvec = (struct buf **) 2662 malloc(bveccount * sizeof(struct buf *), 2663 M_TEMP, M_NOWAIT); 2664 BO_LOCK(bo); 2665 if (bvec == NULL) { 2666 bvec = bvec_on_stack; 2667 bvecsize = NFS_COMMITBVECSIZ; 2668 } else 2669 bvecsize = bveccount; 2670 } else { 2671 bvec = bvec_on_stack; 2672 bvecsize = NFS_COMMITBVECSIZ; 2673 } 2674 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2675 if (bvecpos >= bvecsize) 2676 break; 2677 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2678 nbp = TAILQ_NEXT(bp, b_bobufs); 2679 continue; 2680 } 2681 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 2682 (B_DELWRI | B_NEEDCOMMIT)) { 2683 BUF_UNLOCK(bp); 2684 nbp = TAILQ_NEXT(bp, b_bobufs); 2685 continue; 2686 } 2687 BO_UNLOCK(bo); 2688 bremfree(bp); 2689 /* 2690 * Work out if all buffers are using the same cred 2691 * so we can deal with them all with one commit. 2692 * 2693 * NOTE: we are not clearing B_DONE here, so we have 2694 * to do it later on in this routine if we intend to 2695 * initiate I/O on the bp. 2696 * 2697 * Note: to avoid loopback deadlocks, we do not 2698 * assign b_runningbufspace. 2699 */ 2700 if (wcred == NULL) 2701 wcred = bp->b_wcred; 2702 else if (wcred != bp->b_wcred) 2703 wcred = NOCRED; 2704 vfs_busy_pages(bp, 1); 2705 2706 BO_LOCK(bo); 2707 /* 2708 * bp is protected by being locked, but nbp is not 2709 * and vfs_busy_pages() may sleep. We have to 2710 * recalculate nbp. 2711 */ 2712 nbp = TAILQ_NEXT(bp, b_bobufs); 2713 2714 /* 2715 * A list of these buffers is kept so that the 2716 * second loop knows which buffers have actually 2717 * been committed. This is necessary, since there 2718 * may be a race between the commit rpc and new 2719 * uncommitted writes on the file. 2720 */ 2721 bvec[bvecpos++] = bp; 2722 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2723 bp->b_dirtyoff; 2724 if (toff < off) 2725 off = toff; 2726 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); 2727 if (toff > endoff) 2728 endoff = toff; 2729 } 2730 BO_UNLOCK(bo); 2731 } 2732 if (bvecpos > 0) { 2733 /* 2734 * Commit data on the server, as required. 2735 * If all bufs are using the same wcred, then use that with 2736 * one call for all of them, otherwise commit each one 2737 * separately. 2738 */ 2739 if (wcred != NOCRED) 2740 retv = ncl_commit(vp, off, (int)(endoff - off), 2741 wcred, td); 2742 else { 2743 retv = 0; 2744 for (i = 0; i < bvecpos; i++) { 2745 off_t off, size; 2746 bp = bvec[i]; 2747 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2748 bp->b_dirtyoff; 2749 size = (u_quad_t)(bp->b_dirtyend 2750 - bp->b_dirtyoff); 2751 retv = ncl_commit(vp, off, (int)size, 2752 bp->b_wcred, td); 2753 if (retv) break; 2754 } 2755 } 2756 2757 if (retv == NFSERR_STALEWRITEVERF) 2758 ncl_clearcommit(vp->v_mount); 2759 2760 /* 2761 * Now, either mark the blocks I/O done or mark the 2762 * blocks dirty, depending on whether the commit 2763 * succeeded. 2764 */ 2765 for (i = 0; i < bvecpos; i++) { 2766 bp = bvec[i]; 2767 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 2768 if (retv) { 2769 /* 2770 * Error, leave B_DELWRI intact 2771 */ 2772 vfs_unbusy_pages(bp); 2773 brelse(bp); 2774 } else { 2775 /* 2776 * Success, remove B_DELWRI ( bundirty() ). 2777 * 2778 * b_dirtyoff/b_dirtyend seem to be NFS 2779 * specific. We should probably move that 2780 * into bundirty(). XXX 2781 */ 2782 bufobj_wref(bo); 2783 bp->b_flags |= B_ASYNC; 2784 bundirty(bp); 2785 bp->b_flags &= ~B_DONE; 2786 bp->b_ioflags &= ~BIO_ERROR; 2787 bp->b_dirtyoff = bp->b_dirtyend = 0; 2788 bufdone(bp); 2789 } 2790 } 2791 } 2792 2793 /* 2794 * Start/do any write(s) that are required. 2795 */ 2796loop: 2797 BO_LOCK(bo); 2798 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2799 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2800 if (waitfor != MNT_WAIT || passone) 2801 continue; 2802 2803 error = BUF_TIMELOCK(bp, 2804 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 2805 BO_MTX(bo), "nfsfsync", slpflag, slptimeo); 2806 if (error == 0) { 2807 BUF_UNLOCK(bp); 2808 goto loop; 2809 } 2810 if (error == ENOLCK) { 2811 error = 0; 2812 goto loop; 2813 } 2814 if (called_from_renewthread != 0) { 2815 /* 2816 * Return EIO so the flush will be retried 2817 * later. 2818 */ 2819 error = EIO; 2820 goto done; 2821 } 2822 if (newnfs_sigintr(nmp, td)) { 2823 error = EINTR; 2824 goto done; 2825 } 2826 if (slpflag & PCATCH) { 2827 slpflag = 0; 2828 slptimeo = 2 * hz; 2829 } 2830 goto loop; 2831 } 2832 if ((bp->b_flags & B_DELWRI) == 0) 2833 panic("nfs_fsync: not dirty"); 2834 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) { 2835 BUF_UNLOCK(bp); 2836 continue; 2837 } 2838 BO_UNLOCK(bo); 2839 bremfree(bp); 2840 if (passone || !commit) 2841 bp->b_flags |= B_ASYNC; 2842 else 2843 bp->b_flags |= B_ASYNC; 2844 bwrite(bp); 2845 if (newnfs_sigintr(nmp, td)) { 2846 error = EINTR; 2847 goto done; 2848 } 2849 goto loop; 2850 } 2851 if (passone) { 2852 passone = 0; 2853 BO_UNLOCK(bo); 2854 goto again; 2855 } 2856 if (waitfor == MNT_WAIT) { 2857 while (bo->bo_numoutput) { 2858 error = bufobj_wwait(bo, slpflag, slptimeo); 2859 if (error) { 2860 BO_UNLOCK(bo); 2861 if (called_from_renewthread != 0) { 2862 /* 2863 * Return EIO so that the flush will be 2864 * retried later. 2865 */ 2866 error = EIO; 2867 goto done; 2868 } 2869 error = newnfs_sigintr(nmp, td); 2870 if (error) 2871 goto done; 2872 if (slpflag & PCATCH) { 2873 slpflag = 0; 2874 slptimeo = 2 * hz; 2875 } 2876 BO_LOCK(bo); 2877 } 2878 } 2879 if (bo->bo_dirty.bv_cnt != 0 && commit) { 2880 BO_UNLOCK(bo); 2881 goto loop; 2882 } 2883 /* 2884 * Wait for all the async IO requests to drain 2885 */ 2886 BO_UNLOCK(bo); 2887 mtx_lock(&np->n_mtx); 2888 while (np->n_directio_asyncwr > 0) { 2889 np->n_flag |= NFSYNCWAIT; 2890 error = newnfs_msleep(td, &np->n_directio_asyncwr, 2891 &np->n_mtx, slpflag | (PRIBIO + 1), 2892 "nfsfsync", 0); 2893 if (error) { 2894 if (newnfs_sigintr(nmp, td)) { 2895 mtx_unlock(&np->n_mtx); 2896 error = EINTR; 2897 goto done; 2898 } 2899 } 2900 } 2901 mtx_unlock(&np->n_mtx); 2902 } else 2903 BO_UNLOCK(bo); 2904 mtx_lock(&np->n_mtx); 2905 if (np->n_flag & NWRITEERR) { 2906 error = np->n_error; 2907 np->n_flag &= ~NWRITEERR; 2908 } 2909 if (commit && bo->bo_dirty.bv_cnt == 0 && 2910 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0) 2911 np->n_flag &= ~NMODIFIED; 2912 mtx_unlock(&np->n_mtx); 2913done: 2914 if (bvec != NULL && bvec != bvec_on_stack) 2915 free(bvec, M_TEMP); 2916 if (error == 0 && commit != 0 && waitfor == MNT_WAIT && 2917 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 || 2918 np->n_directio_asyncwr != 0) && trycnt++ < 5) { 2919 /* try, try again... */ 2920 passone = 1; 2921 wcred = NULL; 2922 bvec = NULL; 2923 bvecsize = 0; 2924printf("try%d\n", trycnt); 2925 goto again; 2926 } 2927 return (error); 2928} 2929 2930/* 2931 * NFS advisory byte-level locks. 2932 */ 2933static int 2934nfs_advlock(struct vop_advlock_args *ap) 2935{ 2936 struct vnode *vp = ap->a_vp; 2937 struct ucred *cred; 2938 struct nfsnode *np = VTONFS(ap->a_vp); 2939 struct proc *p = (struct proc *)ap->a_id; 2940 struct thread *td = curthread; /* XXX */ 2941 struct vattr va; 2942 int ret, error = EOPNOTSUPP; 2943 u_quad_t size; 2944 2945 if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) { 2946 if (vp->v_type != VREG) 2947 return (EINVAL); 2948 if ((ap->a_flags & F_POSIX) != 0) 2949 cred = p->p_ucred; 2950 else 2951 cred = td->td_ucred; 2952 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 2953 if (vp->v_iflag & VI_DOOMED) { 2954 NFSVOPUNLOCK(vp, 0); 2955 return (EBADF); 2956 } 2957 2958 /* 2959 * If this is unlocking a write locked region, flush and 2960 * commit them before unlocking. This is required by 2961 * RFC3530 Sec. 9.3.2. 2962 */ 2963 if (ap->a_op == F_UNLCK && 2964 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id, 2965 ap->a_flags)) 2966 (void) ncl_flush(vp, MNT_WAIT, cred, td, 1, 0); 2967 2968 /* 2969 * Loop around doing the lock op, while a blocking lock 2970 * must wait for the lock op to succeed. 2971 */ 2972 do { 2973 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op, 2974 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags); 2975 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 2976 ap->a_op == F_SETLK) { 2977 NFSVOPUNLOCK(vp, 0); 2978 error = nfs_catnap(PZERO | PCATCH, ret, 2979 "ncladvl"); 2980 if (error) 2981 return (EINTR); 2982 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 2983 if (vp->v_iflag & VI_DOOMED) { 2984 NFSVOPUNLOCK(vp, 0); 2985 return (EBADF); 2986 } 2987 } 2988 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 2989 ap->a_op == F_SETLK); 2990 if (ret == NFSERR_DENIED) { 2991 NFSVOPUNLOCK(vp, 0); 2992 return (EAGAIN); 2993 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) { 2994 NFSVOPUNLOCK(vp, 0); 2995 return (ret); 2996 } else if (ret != 0) { 2997 NFSVOPUNLOCK(vp, 0); 2998 return (EACCES); 2999 } 3000 3001 /* 3002 * Now, if we just got a lock, invalidate data in the buffer 3003 * cache, as required, so that the coherency conforms with 3004 * RFC3530 Sec. 9.3.2. 3005 */ 3006 if (ap->a_op == F_SETLK) { 3007 if ((np->n_flag & NMODIFIED) == 0) { 3008 np->n_attrstamp = 0; 3009 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3010 ret = VOP_GETATTR(vp, &va, cred); 3011 } 3012 if ((np->n_flag & NMODIFIED) || ret || 3013 np->n_change != va.va_filerev) { 3014 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1); 3015 np->n_attrstamp = 0; 3016 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3017 ret = VOP_GETATTR(vp, &va, cred); 3018 if (!ret) { 3019 np->n_mtime = va.va_mtime; 3020 np->n_change = va.va_filerev; 3021 } 3022 } 3023 } 3024 NFSVOPUNLOCK(vp, 0); 3025 return (0); 3026 } else if (!NFS_ISV4(vp)) { 3027 error = NFSVOPLOCK(vp, LK_SHARED); 3028 if (error) 3029 return (error); 3030 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3031 size = VTONFS(vp)->n_size; 3032 NFSVOPUNLOCK(vp, 0); 3033 error = lf_advlock(ap, &(vp->v_lockf), size); 3034 } else { 3035 if (nfs_advlock_p != NULL) 3036 error = nfs_advlock_p(ap); 3037 else { 3038 NFSVOPUNLOCK(vp, 0); 3039 error = ENOLCK; 3040 } 3041 } 3042 } 3043 return (error); 3044} 3045 3046/* 3047 * NFS advisory byte-level locks. 3048 */ 3049static int 3050nfs_advlockasync(struct vop_advlockasync_args *ap) 3051{ 3052 struct vnode *vp = ap->a_vp; 3053 u_quad_t size; 3054 int error; 3055 3056 if (NFS_ISV4(vp)) 3057 return (EOPNOTSUPP); 3058 error = NFSVOPLOCK(vp, LK_SHARED); 3059 if (error) 3060 return (error); 3061 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3062 size = VTONFS(vp)->n_size; 3063 NFSVOPUNLOCK(vp, 0); 3064 error = lf_advlockasync(ap, &(vp->v_lockf), size); 3065 } else { 3066 NFSVOPUNLOCK(vp, 0); 3067 error = EOPNOTSUPP; 3068 } 3069 return (error); 3070} 3071 3072/* 3073 * Print out the contents of an nfsnode. 3074 */ 3075static int 3076nfs_print(struct vop_print_args *ap) 3077{ 3078 struct vnode *vp = ap->a_vp; 3079 struct nfsnode *np = VTONFS(vp); 3080 3081 ncl_printf("\tfileid %ld fsid 0x%x", 3082 np->n_vattr.na_fileid, np->n_vattr.na_fsid); 3083 if (vp->v_type == VFIFO) 3084 fifo_printinfo(vp); 3085 printf("\n"); 3086 return (0); 3087} 3088 3089/* 3090 * This is the "real" nfs::bwrite(struct buf*). 3091 * We set B_CACHE if this is a VMIO buffer. 3092 */ 3093int 3094ncl_writebp(struct buf *bp, int force __unused, struct thread *td) 3095{ 3096 int s; 3097 int oldflags = bp->b_flags; 3098#if 0 3099 int retv = 1; 3100 off_t off; 3101#endif 3102 3103 BUF_ASSERT_HELD(bp); 3104 3105 if (bp->b_flags & B_INVAL) { 3106 brelse(bp); 3107 return(0); 3108 } 3109 3110 bp->b_flags |= B_CACHE; 3111 3112 /* 3113 * Undirty the bp. We will redirty it later if the I/O fails. 3114 */ 3115 3116 s = splbio(); 3117 bundirty(bp); 3118 bp->b_flags &= ~B_DONE; 3119 bp->b_ioflags &= ~BIO_ERROR; 3120 bp->b_iocmd = BIO_WRITE; 3121 3122 bufobj_wref(bp->b_bufobj); 3123 curthread->td_ru.ru_oublock++; 3124 splx(s); 3125 3126 /* 3127 * Note: to avoid loopback deadlocks, we do not 3128 * assign b_runningbufspace. 3129 */ 3130 vfs_busy_pages(bp, 1); 3131 3132 BUF_KERNPROC(bp); 3133 bp->b_iooffset = dbtob(bp->b_blkno); 3134 bstrategy(bp); 3135 3136 if( (oldflags & B_ASYNC) == 0) { 3137 int rtval = bufwait(bp); 3138 3139 if (oldflags & B_DELWRI) { 3140 s = splbio(); 3141 reassignbuf(bp); 3142 splx(s); 3143 } 3144 brelse(bp); 3145 return (rtval); 3146 } 3147 3148 return (0); 3149} 3150 3151/* 3152 * nfs special file access vnode op. 3153 * Essentially just get vattr and then imitate iaccess() since the device is 3154 * local to the client. 3155 */ 3156static int 3157nfsspec_access(struct vop_access_args *ap) 3158{ 3159 struct vattr *vap; 3160 struct ucred *cred = ap->a_cred; 3161 struct vnode *vp = ap->a_vp; 3162 accmode_t accmode = ap->a_accmode; 3163 struct vattr vattr; 3164 int error; 3165 3166 /* 3167 * Disallow write attempts on filesystems mounted read-only; 3168 * unless the file is a socket, fifo, or a block or character 3169 * device resident on the filesystem. 3170 */ 3171 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3172 switch (vp->v_type) { 3173 case VREG: 3174 case VDIR: 3175 case VLNK: 3176 return (EROFS); 3177 default: 3178 break; 3179 } 3180 } 3181 vap = &vattr; 3182 error = VOP_GETATTR(vp, vap, cred); 3183 if (error) 3184 goto out; 3185 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid, 3186 accmode, cred, NULL); 3187out: 3188 return error; 3189} 3190 3191/* 3192 * Read wrapper for fifos. 3193 */ 3194static int 3195nfsfifo_read(struct vop_read_args *ap) 3196{ 3197 struct nfsnode *np = VTONFS(ap->a_vp); 3198 int error; 3199 3200 /* 3201 * Set access flag. 3202 */ 3203 mtx_lock(&np->n_mtx); 3204 np->n_flag |= NACC; 3205 getnanotime(&np->n_atim); 3206 mtx_unlock(&np->n_mtx); 3207 error = fifo_specops.vop_read(ap); 3208 return error; 3209} 3210 3211/* 3212 * Write wrapper for fifos. 3213 */ 3214static int 3215nfsfifo_write(struct vop_write_args *ap) 3216{ 3217 struct nfsnode *np = VTONFS(ap->a_vp); 3218 3219 /* 3220 * Set update flag. 3221 */ 3222 mtx_lock(&np->n_mtx); 3223 np->n_flag |= NUPD; 3224 getnanotime(&np->n_mtim); 3225 mtx_unlock(&np->n_mtx); 3226 return(fifo_specops.vop_write(ap)); 3227} 3228 3229/* 3230 * Close wrapper for fifos. 3231 * 3232 * Update the times on the nfsnode then do fifo close. 3233 */ 3234static int 3235nfsfifo_close(struct vop_close_args *ap) 3236{ 3237 struct vnode *vp = ap->a_vp; 3238 struct nfsnode *np = VTONFS(vp); 3239 struct vattr vattr; 3240 struct timespec ts; 3241 3242 mtx_lock(&np->n_mtx); 3243 if (np->n_flag & (NACC | NUPD)) { 3244 getnanotime(&ts); 3245 if (np->n_flag & NACC) 3246 np->n_atim = ts; 3247 if (np->n_flag & NUPD) 3248 np->n_mtim = ts; 3249 np->n_flag |= NCHG; 3250 if (vrefcnt(vp) == 1 && 3251 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3252 VATTR_NULL(&vattr); 3253 if (np->n_flag & NACC) 3254 vattr.va_atime = np->n_atim; 3255 if (np->n_flag & NUPD) 3256 vattr.va_mtime = np->n_mtim; 3257 mtx_unlock(&np->n_mtx); 3258 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3259 goto out; 3260 } 3261 } 3262 mtx_unlock(&np->n_mtx); 3263out: 3264 return (fifo_specops.vop_close(ap)); 3265} 3266 3267/* 3268 * Just call ncl_writebp() with the force argument set to 1. 3269 * 3270 * NOTE: B_DONE may or may not be set in a_bp on call. 3271 */ 3272static int 3273nfs_bwrite(struct buf *bp) 3274{ 3275 3276 return (ncl_writebp(bp, 1, curthread)); 3277} 3278 3279struct buf_ops buf_ops_newnfs = { 3280 .bop_name = "buf_ops_nfs", 3281 .bop_write = nfs_bwrite, 3282 .bop_strategy = bufstrategy, 3283 .bop_sync = bufsync, 3284 .bop_bdflush = bufbdflush, 3285}; 3286 3287/* 3288 * Cloned from vop_stdlock(), and then the ugly hack added. 3289 */ 3290static int 3291nfs_lock1(struct vop_lock1_args *ap) 3292{ 3293 struct vnode *vp = ap->a_vp; 3294 int error = 0; 3295 3296 /* 3297 * Since vfs_hash_get() calls vget() and it will no longer work 3298 * for FreeBSD8 with flags == 0, I can only think of this horrible 3299 * hack to work around it. I call vfs_hash_get() with LK_EXCLOTHER 3300 * and then handle it here. All I want for this case is a v_usecount 3301 * on the vnode to use for recovery, while another thread might 3302 * hold a lock on the vnode. I have the other threads blocked, so 3303 * there isn't any race problem. 3304 */ 3305 if ((ap->a_flags & LK_TYPE_MASK) == LK_EXCLOTHER) { 3306 if ((ap->a_flags & LK_INTERLOCK) == 0) 3307 panic("ncllock1"); 3308 if ((vp->v_iflag & VI_DOOMED)) 3309 error = ENOENT; 3310 VI_UNLOCK(vp); 3311 return (error); 3312 } 3313 return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp), 3314 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file, 3315 ap->a_line)); 3316} 3317 3318static int 3319nfs_getacl(struct vop_getacl_args *ap) 3320{ 3321 int error; 3322 3323 if (ap->a_type != ACL_TYPE_NFS4) 3324 return (EOPNOTSUPP); 3325 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3326 NULL); 3327 if (error > NFSERR_STALE) { 3328 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3329 error = EPERM; 3330 } 3331 return (error); 3332} 3333 3334static int 3335nfs_setacl(struct vop_setacl_args *ap) 3336{ 3337 int error; 3338 3339 if (ap->a_type != ACL_TYPE_NFS4) 3340 return (EOPNOTSUPP); 3341 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3342 NULL); 3343 if (error > NFSERR_STALE) { 3344 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3345 error = EPERM; 3346 } 3347 return (error); 3348} 3349 3350/* 3351 * Return POSIX pathconf information applicable to nfs filesystems. 3352 */ 3353static int 3354nfs_pathconf(struct vop_pathconf_args *ap) 3355{ 3356 struct nfsv3_pathconf pc; 3357 struct nfsvattr nfsva; 3358 struct vnode *vp = ap->a_vp; 3359 struct thread *td = curthread; 3360 int attrflag, error; 3361 3362 if (NFS_ISV4(vp) || (NFS_ISV3(vp) && (ap->a_name == _PC_LINK_MAX || 3363 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED || 3364 ap->a_name == _PC_NO_TRUNC))) { 3365 /* 3366 * Since only the above 4 a_names are returned by the NFSv3 3367 * Pathconf RPC, there is no point in doing it for others. 3368 */ 3369 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva, 3370 &attrflag, NULL); 3371 if (attrflag != 0) 3372 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 3373 1); 3374 if (error != 0) 3375 return (error); 3376 } else { 3377 /* 3378 * For NFSv2 (or NFSv3 when not one of the above 4 a_names), 3379 * just fake them. 3380 */ 3381 pc.pc_linkmax = LINK_MAX; 3382 pc.pc_namemax = NFS_MAXNAMLEN; 3383 pc.pc_notrunc = 1; 3384 pc.pc_chownrestricted = 1; 3385 pc.pc_caseinsensitive = 0; 3386 pc.pc_casepreserving = 1; 3387 error = 0; 3388 } 3389 switch (ap->a_name) { 3390 case _PC_LINK_MAX: 3391 *ap->a_retval = pc.pc_linkmax; 3392 break; 3393 case _PC_NAME_MAX: 3394 *ap->a_retval = pc.pc_namemax; 3395 break; 3396 case _PC_PATH_MAX: 3397 *ap->a_retval = PATH_MAX; 3398 break; 3399 case _PC_PIPE_BUF: 3400 *ap->a_retval = PIPE_BUF; 3401 break; 3402 case _PC_CHOWN_RESTRICTED: 3403 *ap->a_retval = pc.pc_chownrestricted; 3404 break; 3405 case _PC_NO_TRUNC: 3406 *ap->a_retval = pc.pc_notrunc; 3407 break; 3408 case _PC_ACL_EXTENDED: 3409 *ap->a_retval = 0; 3410 break; 3411 case _PC_ACL_NFS4: 3412 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 && 3413 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL)) 3414 *ap->a_retval = 1; 3415 else 3416 *ap->a_retval = 0; 3417 break; 3418 case _PC_ACL_PATH_MAX: 3419 if (NFS_ISV4(vp)) 3420 *ap->a_retval = ACL_MAX_ENTRIES; 3421 else 3422 *ap->a_retval = 3; 3423 break; 3424 case _PC_MAC_PRESENT: 3425 *ap->a_retval = 0; 3426 break; 3427 case _PC_ASYNC_IO: 3428 /* _PC_ASYNC_IO should have been handled by upper layers. */ 3429 KASSERT(0, ("_PC_ASYNC_IO should not get here")); 3430 error = EINVAL; 3431 break; 3432 case _PC_PRIO_IO: 3433 *ap->a_retval = 0; 3434 break; 3435 case _PC_SYNC_IO: 3436 *ap->a_retval = 0; 3437 break; 3438 case _PC_ALLOC_SIZE_MIN: 3439 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize; 3440 break; 3441 case _PC_FILESIZEBITS: 3442 if (NFS_ISV34(vp)) 3443 *ap->a_retval = 64; 3444 else 3445 *ap->a_retval = 32; 3446 break; 3447 case _PC_REC_INCR_XFER_SIZE: 3448 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 3449 break; 3450 case _PC_REC_MAX_XFER_SIZE: 3451 *ap->a_retval = -1; /* means ``unlimited'' */ 3452 break; 3453 case _PC_REC_MIN_XFER_SIZE: 3454 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 3455 break; 3456 case _PC_REC_XFER_ALIGN: 3457 *ap->a_retval = PAGE_SIZE; 3458 break; 3459 case _PC_SYMLINK_MAX: 3460 *ap->a_retval = NFS_MAXPATHLEN; 3461 break; 3462 3463 default: 3464 error = EINVAL; 3465 break; 3466 } 3467 return (error); 3468} 3469 3470