nfs_clvnops.c revision 276438
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: stable/10/sys/fs/nfsclient/nfs_clvnops.c 276438 2014-12-31 00:44:11Z rmacklem $"); 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; 106extern int nfscl_debuglevel; 107MALLOC_DECLARE(M_NEWNFSREQ); 108 109/* 110 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these 111 * calls are not in getblk() and brelse() so that they would not be necessary 112 * here. 113 */ 114#ifndef B_VMIO 115#define vfs_busy_pages(bp, f) 116#endif 117 118static vop_read_t nfsfifo_read; 119static vop_write_t nfsfifo_write; 120static vop_close_t nfsfifo_close; 121static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *, 122 struct thread *); 123static vop_lookup_t nfs_lookup; 124static vop_create_t nfs_create; 125static vop_mknod_t nfs_mknod; 126static vop_open_t nfs_open; 127static vop_pathconf_t nfs_pathconf; 128static vop_close_t nfs_close; 129static vop_access_t nfs_access; 130static vop_getattr_t nfs_getattr; 131static vop_setattr_t nfs_setattr; 132static vop_read_t nfs_read; 133static vop_fsync_t nfs_fsync; 134static vop_remove_t nfs_remove; 135static vop_link_t nfs_link; 136static vop_rename_t nfs_rename; 137static vop_mkdir_t nfs_mkdir; 138static vop_rmdir_t nfs_rmdir; 139static vop_symlink_t nfs_symlink; 140static vop_readdir_t nfs_readdir; 141static vop_strategy_t nfs_strategy; 142static vop_lock1_t nfs_lock1; 143static int nfs_lookitup(struct vnode *, char *, int, 144 struct ucred *, struct thread *, struct nfsnode **); 145static int nfs_sillyrename(struct vnode *, struct vnode *, 146 struct componentname *); 147static vop_access_t nfsspec_access; 148static vop_readlink_t nfs_readlink; 149static vop_print_t nfs_print; 150static vop_advlock_t nfs_advlock; 151static vop_advlockasync_t nfs_advlockasync; 152static vop_getacl_t nfs_getacl; 153static vop_setacl_t nfs_setacl; 154 155/* 156 * Global vfs data structures for nfs 157 */ 158struct vop_vector newnfs_vnodeops = { 159 .vop_default = &default_vnodeops, 160 .vop_access = nfs_access, 161 .vop_advlock = nfs_advlock, 162 .vop_advlockasync = nfs_advlockasync, 163 .vop_close = nfs_close, 164 .vop_create = nfs_create, 165 .vop_fsync = nfs_fsync, 166 .vop_getattr = nfs_getattr, 167 .vop_getpages = ncl_getpages, 168 .vop_putpages = ncl_putpages, 169 .vop_inactive = ncl_inactive, 170 .vop_link = nfs_link, 171 .vop_lock1 = nfs_lock1, 172 .vop_lookup = nfs_lookup, 173 .vop_mkdir = nfs_mkdir, 174 .vop_mknod = nfs_mknod, 175 .vop_open = nfs_open, 176 .vop_pathconf = nfs_pathconf, 177 .vop_print = nfs_print, 178 .vop_read = nfs_read, 179 .vop_readdir = nfs_readdir, 180 .vop_readlink = nfs_readlink, 181 .vop_reclaim = ncl_reclaim, 182 .vop_remove = nfs_remove, 183 .vop_rename = nfs_rename, 184 .vop_rmdir = nfs_rmdir, 185 .vop_setattr = nfs_setattr, 186 .vop_strategy = nfs_strategy, 187 .vop_symlink = nfs_symlink, 188 .vop_write = ncl_write, 189 .vop_getacl = nfs_getacl, 190 .vop_setacl = nfs_setacl, 191}; 192 193struct vop_vector newnfs_fifoops = { 194 .vop_default = &fifo_specops, 195 .vop_access = nfsspec_access, 196 .vop_close = nfsfifo_close, 197 .vop_fsync = nfs_fsync, 198 .vop_getattr = nfs_getattr, 199 .vop_inactive = ncl_inactive, 200 .vop_print = nfs_print, 201 .vop_read = nfsfifo_read, 202 .vop_reclaim = ncl_reclaim, 203 .vop_setattr = nfs_setattr, 204 .vop_write = nfsfifo_write, 205}; 206 207static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, 208 struct componentname *cnp, struct vattr *vap); 209static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 210 int namelen, struct ucred *cred, struct thread *td); 211static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, 212 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp, 213 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td); 214static int nfs_renameit(struct vnode *sdvp, struct vnode *svp, 215 struct componentname *scnp, struct sillyrename *sp); 216 217/* 218 * Global variables 219 */ 220#define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1)) 221 222SYSCTL_DECL(_vfs_nfs); 223 224static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO; 225SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 226 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 227 228static int nfs_prime_access_cache = 0; 229SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW, 230 &nfs_prime_access_cache, 0, 231 "Prime NFS ACCESS cache when fetching attributes"); 232 233static int newnfs_commit_on_close = 0; 234SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW, 235 &newnfs_commit_on_close, 0, "write+commit on close, else only write"); 236 237static int nfs_clean_pages_on_close = 1; 238SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW, 239 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close"); 240 241int newnfs_directio_enable = 0; 242SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW, 243 &newnfs_directio_enable, 0, "Enable NFS directio"); 244 245int nfs_keep_dirty_on_error; 246SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW, 247 &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned"); 248 249/* 250 * This sysctl allows other processes to mmap a file that has been opened 251 * O_DIRECT by a process. In general, having processes mmap the file while 252 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow 253 * this by default to prevent DoS attacks - to prevent a malicious user from 254 * opening up files O_DIRECT preventing other users from mmap'ing these 255 * files. "Protected" environments where stricter consistency guarantees are 256 * required can disable this knob. The process that opened the file O_DIRECT 257 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not 258 * meaningful. 259 */ 260int newnfs_directio_allow_mmap = 1; 261SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW, 262 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens"); 263 264#if 0 265SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD, 266 &newnfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count"); 267 268SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD, 269 &newnfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count"); 270#endif 271 272#define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \ 273 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \ 274 | NFSACCESS_DELETE | NFSACCESS_LOOKUP) 275 276/* 277 * SMP Locking Note : 278 * The list of locks after the description of the lock is the ordering 279 * of other locks acquired with the lock held. 280 * np->n_mtx : Protects the fields in the nfsnode. 281 VM Object Lock 282 VI_MTX (acquired indirectly) 283 * nmp->nm_mtx : Protects the fields in the nfsmount. 284 rep->r_mtx 285 * ncl_iod_mutex : Global lock, protects shared nfsiod state. 286 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list. 287 nmp->nm_mtx 288 rep->r_mtx 289 * rep->r_mtx : Protects the fields in an nfsreq. 290 */ 291 292static int 293nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td, 294 struct ucred *cred, u_int32_t *retmode) 295{ 296 int error = 0, attrflag, i, lrupos; 297 u_int32_t rmode; 298 struct nfsnode *np = VTONFS(vp); 299 struct nfsvattr nfsva; 300 301 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag, 302 &rmode, NULL); 303 if (attrflag) 304 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 305 if (!error) { 306 lrupos = 0; 307 mtx_lock(&np->n_mtx); 308 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 309 if (np->n_accesscache[i].uid == cred->cr_uid) { 310 np->n_accesscache[i].mode = rmode; 311 np->n_accesscache[i].stamp = time_second; 312 break; 313 } 314 if (i > 0 && np->n_accesscache[i].stamp < 315 np->n_accesscache[lrupos].stamp) 316 lrupos = i; 317 } 318 if (i == NFS_ACCESSCACHESIZE) { 319 np->n_accesscache[lrupos].uid = cred->cr_uid; 320 np->n_accesscache[lrupos].mode = rmode; 321 np->n_accesscache[lrupos].stamp = time_second; 322 } 323 mtx_unlock(&np->n_mtx); 324 if (retmode != NULL) 325 *retmode = rmode; 326 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0); 327 } else if (NFS_ISV4(vp)) { 328 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 329 } 330#ifdef KDTRACE_HOOKS 331 if (error != 0) 332 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0, 333 error); 334#endif 335 return (error); 336} 337 338/* 339 * nfs access vnode op. 340 * For nfs version 2, just return ok. File accesses may fail later. 341 * For nfs version 3, use the access rpc to check accessibility. If file modes 342 * are changed on the server, accesses might still fail later. 343 */ 344static int 345nfs_access(struct vop_access_args *ap) 346{ 347 struct vnode *vp = ap->a_vp; 348 int error = 0, i, gotahit; 349 u_int32_t mode, wmode, rmode; 350 int v34 = NFS_ISV34(vp); 351 struct nfsnode *np = VTONFS(vp); 352 353 /* 354 * Disallow write attempts on filesystems mounted read-only; 355 * unless the file is a socket, fifo, or a block or character 356 * device resident on the filesystem. 357 */ 358 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS | 359 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL | 360 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) { 361 switch (vp->v_type) { 362 case VREG: 363 case VDIR: 364 case VLNK: 365 return (EROFS); 366 default: 367 break; 368 } 369 } 370 /* 371 * For nfs v3 or v4, check to see if we have done this recently, and if 372 * so return our cached result instead of making an ACCESS call. 373 * If not, do an access rpc, otherwise you are stuck emulating 374 * ufs_access() locally using the vattr. This may not be correct, 375 * since the server may apply other access criteria such as 376 * client uid-->server uid mapping that we do not know about. 377 */ 378 if (v34) { 379 if (ap->a_accmode & VREAD) 380 mode = NFSACCESS_READ; 381 else 382 mode = 0; 383 if (vp->v_type != VDIR) { 384 if (ap->a_accmode & VWRITE) 385 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 386 if (ap->a_accmode & VAPPEND) 387 mode |= NFSACCESS_EXTEND; 388 if (ap->a_accmode & VEXEC) 389 mode |= NFSACCESS_EXECUTE; 390 if (ap->a_accmode & VDELETE) 391 mode |= NFSACCESS_DELETE; 392 } else { 393 if (ap->a_accmode & VWRITE) 394 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 395 if (ap->a_accmode & VAPPEND) 396 mode |= NFSACCESS_EXTEND; 397 if (ap->a_accmode & VEXEC) 398 mode |= NFSACCESS_LOOKUP; 399 if (ap->a_accmode & VDELETE) 400 mode |= NFSACCESS_DELETE; 401 if (ap->a_accmode & VDELETE_CHILD) 402 mode |= NFSACCESS_MODIFY; 403 } 404 /* XXX safety belt, only make blanket request if caching */ 405 if (nfsaccess_cache_timeout > 0) { 406 wmode = NFSACCESS_READ | NFSACCESS_MODIFY | 407 NFSACCESS_EXTEND | NFSACCESS_EXECUTE | 408 NFSACCESS_DELETE | NFSACCESS_LOOKUP; 409 } else { 410 wmode = mode; 411 } 412 413 /* 414 * Does our cached result allow us to give a definite yes to 415 * this request? 416 */ 417 gotahit = 0; 418 mtx_lock(&np->n_mtx); 419 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 420 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) { 421 if (time_second < (np->n_accesscache[i].stamp 422 + nfsaccess_cache_timeout) && 423 (np->n_accesscache[i].mode & mode) == mode) { 424 NFSINCRGLOBAL(newnfsstats.accesscache_hits); 425 gotahit = 1; 426 } 427 break; 428 } 429 } 430 mtx_unlock(&np->n_mtx); 431#ifdef KDTRACE_HOOKS 432 if (gotahit != 0) 433 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp, 434 ap->a_cred->cr_uid, mode); 435 else 436 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp, 437 ap->a_cred->cr_uid, mode); 438#endif 439 if (gotahit == 0) { 440 /* 441 * Either a no, or a don't know. Go to the wire. 442 */ 443 NFSINCRGLOBAL(newnfsstats.accesscache_misses); 444 error = nfs34_access_otw(vp, wmode, ap->a_td, 445 ap->a_cred, &rmode); 446 if (!error && 447 (rmode & mode) != mode) 448 error = EACCES; 449 } 450 return (error); 451 } else { 452 if ((error = nfsspec_access(ap)) != 0) { 453 return (error); 454 } 455 /* 456 * Attempt to prevent a mapped root from accessing a file 457 * which it shouldn't. We try to read a byte from the file 458 * if the user is root and the file is not zero length. 459 * After calling nfsspec_access, we should have the correct 460 * file size cached. 461 */ 462 mtx_lock(&np->n_mtx); 463 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD) 464 && VTONFS(vp)->n_size > 0) { 465 struct iovec aiov; 466 struct uio auio; 467 char buf[1]; 468 469 mtx_unlock(&np->n_mtx); 470 aiov.iov_base = buf; 471 aiov.iov_len = 1; 472 auio.uio_iov = &aiov; 473 auio.uio_iovcnt = 1; 474 auio.uio_offset = 0; 475 auio.uio_resid = 1; 476 auio.uio_segflg = UIO_SYSSPACE; 477 auio.uio_rw = UIO_READ; 478 auio.uio_td = ap->a_td; 479 480 if (vp->v_type == VREG) 481 error = ncl_readrpc(vp, &auio, ap->a_cred); 482 else if (vp->v_type == VDIR) { 483 char* bp; 484 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 485 aiov.iov_base = bp; 486 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 487 error = ncl_readdirrpc(vp, &auio, ap->a_cred, 488 ap->a_td); 489 free(bp, M_TEMP); 490 } else if (vp->v_type == VLNK) 491 error = ncl_readlinkrpc(vp, &auio, ap->a_cred); 492 else 493 error = EACCES; 494 } else 495 mtx_unlock(&np->n_mtx); 496 return (error); 497 } 498} 499 500 501/* 502 * nfs open vnode op 503 * Check to see if the type is ok 504 * and that deletion is not in progress. 505 * For paged in text files, you will need to flush the page cache 506 * if consistency is lost. 507 */ 508/* ARGSUSED */ 509static int 510nfs_open(struct vop_open_args *ap) 511{ 512 struct vnode *vp = ap->a_vp; 513 struct nfsnode *np = VTONFS(vp); 514 struct vattr vattr; 515 int error; 516 int fmode = ap->a_mode; 517 struct ucred *cred; 518 519 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) 520 return (EOPNOTSUPP); 521 522 /* 523 * For NFSv4, we need to do the Open Op before cache validation, 524 * so that we conform to RFC3530 Sec. 9.3.1. 525 */ 526 if (NFS_ISV4(vp)) { 527 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td); 528 if (error) { 529 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 530 (gid_t)0); 531 return (error); 532 } 533 } 534 535 /* 536 * Now, if this Open will be doing reading, re-validate/flush the 537 * cache, so that Close/Open coherency is maintained. 538 */ 539 mtx_lock(&np->n_mtx); 540 if (np->n_flag & NMODIFIED) { 541 mtx_unlock(&np->n_mtx); 542 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 543 if (error == EINTR || error == EIO) { 544 if (NFS_ISV4(vp)) 545 (void) nfsrpc_close(vp, 0, ap->a_td); 546 return (error); 547 } 548 mtx_lock(&np->n_mtx); 549 np->n_attrstamp = 0; 550 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 551 if (vp->v_type == VDIR) 552 np->n_direofoffset = 0; 553 mtx_unlock(&np->n_mtx); 554 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 555 if (error) { 556 if (NFS_ISV4(vp)) 557 (void) nfsrpc_close(vp, 0, ap->a_td); 558 return (error); 559 } 560 mtx_lock(&np->n_mtx); 561 np->n_mtime = vattr.va_mtime; 562 if (NFS_ISV4(vp)) 563 np->n_change = vattr.va_filerev; 564 } else { 565 mtx_unlock(&np->n_mtx); 566 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 567 if (error) { 568 if (NFS_ISV4(vp)) 569 (void) nfsrpc_close(vp, 0, ap->a_td); 570 return (error); 571 } 572 mtx_lock(&np->n_mtx); 573 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) || 574 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 575 if (vp->v_type == VDIR) 576 np->n_direofoffset = 0; 577 mtx_unlock(&np->n_mtx); 578 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 579 if (error == EINTR || error == EIO) { 580 if (NFS_ISV4(vp)) 581 (void) nfsrpc_close(vp, 0, ap->a_td); 582 return (error); 583 } 584 mtx_lock(&np->n_mtx); 585 np->n_mtime = vattr.va_mtime; 586 if (NFS_ISV4(vp)) 587 np->n_change = vattr.va_filerev; 588 } 589 } 590 591 /* 592 * If the object has >= 1 O_DIRECT active opens, we disable caching. 593 */ 594 if (newnfs_directio_enable && (fmode & O_DIRECT) && 595 (vp->v_type == VREG)) { 596 if (np->n_directio_opens == 0) { 597 mtx_unlock(&np->n_mtx); 598 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 599 if (error) { 600 if (NFS_ISV4(vp)) 601 (void) nfsrpc_close(vp, 0, ap->a_td); 602 return (error); 603 } 604 mtx_lock(&np->n_mtx); 605 np->n_flag |= NNONCACHE; 606 } 607 np->n_directio_opens++; 608 } 609 610 /* If opened for writing via NFSv4.1 or later, mark that for pNFS. */ 611 if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0) 612 np->n_flag |= NWRITEOPENED; 613 614 /* 615 * If this is an open for writing, capture a reference to the 616 * credentials, so they can be used by ncl_putpages(). Using 617 * these write credentials is preferable to the credentials of 618 * whatever thread happens to be doing the VOP_PUTPAGES() since 619 * the write RPCs are less likely to fail with EACCES. 620 */ 621 if ((fmode & FWRITE) != 0) { 622 cred = np->n_writecred; 623 np->n_writecred = crhold(ap->a_cred); 624 } else 625 cred = NULL; 626 mtx_unlock(&np->n_mtx); 627 628 if (cred != NULL) 629 crfree(cred); 630 vnode_create_vobject(vp, vattr.va_size, ap->a_td); 631 return (0); 632} 633 634/* 635 * nfs close vnode op 636 * What an NFS client should do upon close after writing is a debatable issue. 637 * Most NFS clients push delayed writes to the server upon close, basically for 638 * two reasons: 639 * 1 - So that any write errors may be reported back to the client process 640 * doing the close system call. By far the two most likely errors are 641 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 642 * 2 - To put a worst case upper bound on cache inconsistency between 643 * multiple clients for the file. 644 * There is also a consistency problem for Version 2 of the protocol w.r.t. 645 * not being able to tell if other clients are writing a file concurrently, 646 * since there is no way of knowing if the changed modify time in the reply 647 * is only due to the write for this client. 648 * (NFS Version 3 provides weak cache consistency data in the reply that 649 * should be sufficient to detect and handle this case.) 650 * 651 * The current code does the following: 652 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 653 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 654 * or commit them (this satisfies 1 and 2 except for the 655 * case where the server crashes after this close but 656 * before the commit RPC, which is felt to be "good 657 * enough". Changing the last argument to ncl_flush() to 658 * a 1 would force a commit operation, if it is felt a 659 * commit is necessary now. 660 * for NFS Version 4 - flush the dirty buffers and commit them, if 661 * nfscl_mustflush() says this is necessary. 662 * It is necessary if there is no write delegation held, 663 * in order to satisfy open/close coherency. 664 * If the file isn't cached on local stable storage, 665 * it may be necessary in order to detect "out of space" 666 * errors from the server, if the write delegation 667 * issued by the server doesn't allow the file to grow. 668 */ 669/* ARGSUSED */ 670static int 671nfs_close(struct vop_close_args *ap) 672{ 673 struct vnode *vp = ap->a_vp; 674 struct nfsnode *np = VTONFS(vp); 675 struct nfsvattr nfsva; 676 struct ucred *cred; 677 int error = 0, ret, localcred = 0; 678 int fmode = ap->a_fflag; 679 680 if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF)) 681 return (0); 682 /* 683 * During shutdown, a_cred isn't valid, so just use root. 684 */ 685 if (ap->a_cred == NOCRED) { 686 cred = newnfs_getcred(); 687 localcred = 1; 688 } else { 689 cred = ap->a_cred; 690 } 691 if (vp->v_type == VREG) { 692 /* 693 * Examine and clean dirty pages, regardless of NMODIFIED. 694 * This closes a major hole in close-to-open consistency. 695 * We want to push out all dirty pages (and buffers) on 696 * close, regardless of whether they were dirtied by 697 * mmap'ed writes or via write(). 698 */ 699 if (nfs_clean_pages_on_close && vp->v_object) { 700 VM_OBJECT_WLOCK(vp->v_object); 701 vm_object_page_clean(vp->v_object, 0, 0, 0); 702 VM_OBJECT_WUNLOCK(vp->v_object); 703 } 704 mtx_lock(&np->n_mtx); 705 if (np->n_flag & NMODIFIED) { 706 mtx_unlock(&np->n_mtx); 707 if (NFS_ISV3(vp)) { 708 /* 709 * Under NFSv3 we have dirty buffers to dispose of. We 710 * must flush them to the NFS server. We have the option 711 * of waiting all the way through the commit rpc or just 712 * waiting for the initial write. The default is to only 713 * wait through the initial write so the data is in the 714 * server's cache, which is roughly similar to the state 715 * a standard disk subsystem leaves the file in on close(). 716 * 717 * We cannot clear the NMODIFIED bit in np->n_flag due to 718 * potential races with other processes, and certainly 719 * cannot clear it if we don't commit. 720 * These races occur when there is no longer the old 721 * traditional vnode locking implemented for Vnode Ops. 722 */ 723 int cm = newnfs_commit_on_close ? 1 : 0; 724 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm, 0); 725 /* np->n_flag &= ~NMODIFIED; */ 726 } else if (NFS_ISV4(vp)) { 727 if (nfscl_mustflush(vp) != 0) { 728 int cm = newnfs_commit_on_close ? 1 : 0; 729 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, 730 cm, 0); 731 /* 732 * as above w.r.t races when clearing 733 * NMODIFIED. 734 * np->n_flag &= ~NMODIFIED; 735 */ 736 } 737 } else 738 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 739 mtx_lock(&np->n_mtx); 740 } 741 /* 742 * Invalidate the attribute cache in all cases. 743 * An open is going to fetch fresh attrs any way, other procs 744 * on this node that have file open will be forced to do an 745 * otw attr fetch, but this is safe. 746 * --> A user found that their RPC count dropped by 20% when 747 * this was commented out and I can't see any requirement 748 * for it, so I've disabled it when negative lookups are 749 * enabled. (What does this have to do with negative lookup 750 * caching? Well nothing, except it was reported by the 751 * same user that needed negative lookup caching and I wanted 752 * there to be a way to disable it to see if it 753 * is the cause of some caching/coherency issue that might 754 * crop up.) 755 */ 756 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) { 757 np->n_attrstamp = 0; 758 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 759 } 760 if (np->n_flag & NWRITEERR) { 761 np->n_flag &= ~NWRITEERR; 762 error = np->n_error; 763 } 764 mtx_unlock(&np->n_mtx); 765 } 766 767 if (NFS_ISV4(vp)) { 768 /* 769 * Get attributes so "change" is up to date. 770 */ 771 if (error == 0 && nfscl_mustflush(vp) != 0 && 772 vp->v_type == VREG && 773 (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) { 774 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva, 775 NULL); 776 if (!ret) { 777 np->n_change = nfsva.na_filerev; 778 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 779 NULL, 0, 0); 780 } 781 } 782 783 /* 784 * and do the close. 785 */ 786 ret = nfsrpc_close(vp, 0, ap->a_td); 787 if (!error && ret) 788 error = ret; 789 if (error) 790 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 791 (gid_t)0); 792 } 793 if (newnfs_directio_enable) 794 KASSERT((np->n_directio_asyncwr == 0), 795 ("nfs_close: dirty unflushed (%d) directio buffers\n", 796 np->n_directio_asyncwr)); 797 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) { 798 mtx_lock(&np->n_mtx); 799 KASSERT((np->n_directio_opens > 0), 800 ("nfs_close: unexpectedly value (0) of n_directio_opens\n")); 801 np->n_directio_opens--; 802 if (np->n_directio_opens == 0) 803 np->n_flag &= ~NNONCACHE; 804 mtx_unlock(&np->n_mtx); 805 } 806 if (localcred) 807 NFSFREECRED(cred); 808 return (error); 809} 810 811/* 812 * nfs getattr call from vfs. 813 */ 814static int 815nfs_getattr(struct vop_getattr_args *ap) 816{ 817 struct vnode *vp = ap->a_vp; 818 struct thread *td = curthread; /* XXX */ 819 struct nfsnode *np = VTONFS(vp); 820 int error = 0; 821 struct nfsvattr nfsva; 822 struct vattr *vap = ap->a_vap; 823 struct vattr vattr; 824 825 /* 826 * Update local times for special files. 827 */ 828 mtx_lock(&np->n_mtx); 829 if (np->n_flag & (NACC | NUPD)) 830 np->n_flag |= NCHG; 831 mtx_unlock(&np->n_mtx); 832 /* 833 * First look in the cache. 834 */ 835 if (ncl_getattrcache(vp, &vattr) == 0) { 836 vap->va_type = vattr.va_type; 837 vap->va_mode = vattr.va_mode; 838 vap->va_nlink = vattr.va_nlink; 839 vap->va_uid = vattr.va_uid; 840 vap->va_gid = vattr.va_gid; 841 vap->va_fsid = vattr.va_fsid; 842 vap->va_fileid = vattr.va_fileid; 843 vap->va_size = vattr.va_size; 844 vap->va_blocksize = vattr.va_blocksize; 845 vap->va_atime = vattr.va_atime; 846 vap->va_mtime = vattr.va_mtime; 847 vap->va_ctime = vattr.va_ctime; 848 vap->va_gen = vattr.va_gen; 849 vap->va_flags = vattr.va_flags; 850 vap->va_rdev = vattr.va_rdev; 851 vap->va_bytes = vattr.va_bytes; 852 vap->va_filerev = vattr.va_filerev; 853 /* 854 * Get the local modify time for the case of a write 855 * delegation. 856 */ 857 nfscl_deleggetmodtime(vp, &vap->va_mtime); 858 return (0); 859 } 860 861 if (NFS_ISV34(vp) && nfs_prime_access_cache && 862 nfsaccess_cache_timeout > 0) { 863 NFSINCRGLOBAL(newnfsstats.accesscache_misses); 864 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL); 865 if (ncl_getattrcache(vp, ap->a_vap) == 0) { 866 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime); 867 return (0); 868 } 869 } 870 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL); 871 if (!error) 872 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0); 873 if (!error) { 874 /* 875 * Get the local modify time for the case of a write 876 * delegation. 877 */ 878 nfscl_deleggetmodtime(vp, &vap->va_mtime); 879 } else if (NFS_ISV4(vp)) { 880 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 881 } 882 return (error); 883} 884 885/* 886 * nfs setattr call. 887 */ 888static int 889nfs_setattr(struct vop_setattr_args *ap) 890{ 891 struct vnode *vp = ap->a_vp; 892 struct nfsnode *np = VTONFS(vp); 893 struct thread *td = curthread; /* XXX */ 894 struct vattr *vap = ap->a_vap; 895 int error = 0; 896 u_quad_t tsize; 897 898#ifndef nolint 899 tsize = (u_quad_t)0; 900#endif 901 902 /* 903 * Setting of flags and marking of atimes are not supported. 904 */ 905 if (vap->va_flags != VNOVAL) 906 return (EOPNOTSUPP); 907 908 /* 909 * Disallow write attempts if the filesystem is mounted read-only. 910 */ 911 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 912 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 913 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 914 (vp->v_mount->mnt_flag & MNT_RDONLY)) 915 return (EROFS); 916 if (vap->va_size != VNOVAL) { 917 switch (vp->v_type) { 918 case VDIR: 919 return (EISDIR); 920 case VCHR: 921 case VBLK: 922 case VSOCK: 923 case VFIFO: 924 if (vap->va_mtime.tv_sec == VNOVAL && 925 vap->va_atime.tv_sec == VNOVAL && 926 vap->va_mode == (mode_t)VNOVAL && 927 vap->va_uid == (uid_t)VNOVAL && 928 vap->va_gid == (gid_t)VNOVAL) 929 return (0); 930 vap->va_size = VNOVAL; 931 break; 932 default: 933 /* 934 * Disallow write attempts if the filesystem is 935 * mounted read-only. 936 */ 937 if (vp->v_mount->mnt_flag & MNT_RDONLY) 938 return (EROFS); 939 /* 940 * We run vnode_pager_setsize() early (why?), 941 * we must set np->n_size now to avoid vinvalbuf 942 * V_SAVE races that might setsize a lower 943 * value. 944 */ 945 mtx_lock(&np->n_mtx); 946 tsize = np->n_size; 947 mtx_unlock(&np->n_mtx); 948 error = ncl_meta_setsize(vp, ap->a_cred, td, 949 vap->va_size); 950 mtx_lock(&np->n_mtx); 951 if (np->n_flag & NMODIFIED) { 952 tsize = np->n_size; 953 mtx_unlock(&np->n_mtx); 954 if (vap->va_size == 0) 955 error = ncl_vinvalbuf(vp, 0, td, 1); 956 else 957 error = ncl_vinvalbuf(vp, V_SAVE, td, 1); 958 if (error) { 959 vnode_pager_setsize(vp, tsize); 960 return (error); 961 } 962 /* 963 * Call nfscl_delegmodtime() to set the modify time 964 * locally, as required. 965 */ 966 nfscl_delegmodtime(vp); 967 } else 968 mtx_unlock(&np->n_mtx); 969 /* 970 * np->n_size has already been set to vap->va_size 971 * in ncl_meta_setsize(). We must set it again since 972 * nfs_loadattrcache() could be called through 973 * ncl_meta_setsize() and could modify np->n_size. 974 */ 975 mtx_lock(&np->n_mtx); 976 np->n_vattr.na_size = np->n_size = vap->va_size; 977 mtx_unlock(&np->n_mtx); 978 }; 979 } else { 980 mtx_lock(&np->n_mtx); 981 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) && 982 (np->n_flag & NMODIFIED) && vp->v_type == VREG) { 983 mtx_unlock(&np->n_mtx); 984 if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 && 985 (error == EINTR || error == EIO)) 986 return (error); 987 } else 988 mtx_unlock(&np->n_mtx); 989 } 990 error = nfs_setattrrpc(vp, vap, ap->a_cred, td); 991 if (error && vap->va_size != VNOVAL) { 992 mtx_lock(&np->n_mtx); 993 np->n_size = np->n_vattr.na_size = tsize; 994 vnode_pager_setsize(vp, tsize); 995 mtx_unlock(&np->n_mtx); 996 } 997 return (error); 998} 999 1000/* 1001 * Do an nfs setattr rpc. 1002 */ 1003static int 1004nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred, 1005 struct thread *td) 1006{ 1007 struct nfsnode *np = VTONFS(vp); 1008 int error, ret, attrflag, i; 1009 struct nfsvattr nfsva; 1010 1011 if (NFS_ISV34(vp)) { 1012 mtx_lock(&np->n_mtx); 1013 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) 1014 np->n_accesscache[i].stamp = 0; 1015 np->n_flag |= NDELEGMOD; 1016 mtx_unlock(&np->n_mtx); 1017 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp); 1018 } 1019 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag, 1020 NULL); 1021 if (attrflag) { 1022 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1023 if (ret && !error) 1024 error = ret; 1025 } 1026 if (error && NFS_ISV4(vp)) 1027 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid); 1028 return (error); 1029} 1030 1031/* 1032 * nfs lookup call, one step at a time... 1033 * First look in cache 1034 * If not found, unlock the directory nfsnode and do the rpc 1035 */ 1036static int 1037nfs_lookup(struct vop_lookup_args *ap) 1038{ 1039 struct componentname *cnp = ap->a_cnp; 1040 struct vnode *dvp = ap->a_dvp; 1041 struct vnode **vpp = ap->a_vpp; 1042 struct mount *mp = dvp->v_mount; 1043 int flags = cnp->cn_flags; 1044 struct vnode *newvp; 1045 struct nfsmount *nmp; 1046 struct nfsnode *np, *newnp; 1047 int error = 0, attrflag, dattrflag, ltype, ncticks; 1048 struct thread *td = cnp->cn_thread; 1049 struct nfsfh *nfhp; 1050 struct nfsvattr dnfsva, nfsva; 1051 struct vattr vattr; 1052 struct timespec nctime; 1053 1054 *vpp = NULLVP; 1055 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) && 1056 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 1057 return (EROFS); 1058 if (dvp->v_type != VDIR) 1059 return (ENOTDIR); 1060 nmp = VFSTONFS(mp); 1061 np = VTONFS(dvp); 1062 1063 /* For NFSv4, wait until any remove is done. */ 1064 mtx_lock(&np->n_mtx); 1065 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) { 1066 np->n_flag |= NREMOVEWANT; 1067 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0); 1068 } 1069 mtx_unlock(&np->n_mtx); 1070 1071 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) 1072 return (error); 1073 error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks); 1074 if (error > 0 && error != ENOENT) 1075 return (error); 1076 if (error == -1) { 1077 /* 1078 * Lookups of "." are special and always return the 1079 * current directory. cache_lookup() already handles 1080 * associated locking bookkeeping, etc. 1081 */ 1082 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 1083 /* XXX: Is this really correct? */ 1084 if (cnp->cn_nameiop != LOOKUP && 1085 (flags & ISLASTCN)) 1086 cnp->cn_flags |= SAVENAME; 1087 return (0); 1088 } 1089 1090 /* 1091 * We only accept a positive hit in the cache if the 1092 * change time of the file matches our cached copy. 1093 * Otherwise, we discard the cache entry and fallback 1094 * to doing a lookup RPC. We also only trust cache 1095 * entries for less than nm_nametimeo seconds. 1096 * 1097 * To better handle stale file handles and attributes, 1098 * clear the attribute cache of this node if it is a 1099 * leaf component, part of an open() call, and not 1100 * locally modified before fetching the attributes. 1101 * This should allow stale file handles to be detected 1102 * here where we can fall back to a LOOKUP RPC to 1103 * recover rather than having nfs_open() detect the 1104 * stale file handle and failing open(2) with ESTALE. 1105 */ 1106 newvp = *vpp; 1107 newnp = VTONFS(newvp); 1108 if (!(nmp->nm_flag & NFSMNT_NOCTO) && 1109 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1110 !(newnp->n_flag & NMODIFIED)) { 1111 mtx_lock(&newnp->n_mtx); 1112 newnp->n_attrstamp = 0; 1113 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1114 mtx_unlock(&newnp->n_mtx); 1115 } 1116 if (nfscl_nodeleg(newvp, 0) == 0 || 1117 ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) && 1118 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 && 1119 timespeccmp(&vattr.va_ctime, &nctime, ==))) { 1120 NFSINCRGLOBAL(newnfsstats.lookupcache_hits); 1121 if (cnp->cn_nameiop != LOOKUP && 1122 (flags & ISLASTCN)) 1123 cnp->cn_flags |= SAVENAME; 1124 return (0); 1125 } 1126 cache_purge(newvp); 1127 if (dvp != newvp) 1128 vput(newvp); 1129 else 1130 vrele(newvp); 1131 *vpp = NULLVP; 1132 } else if (error == ENOENT) { 1133 if (dvp->v_iflag & VI_DOOMED) 1134 return (ENOENT); 1135 /* 1136 * We only accept a negative hit in the cache if the 1137 * modification time of the parent directory matches 1138 * the cached copy in the name cache entry. 1139 * Otherwise, we discard all of the negative cache 1140 * entries for this directory. We also only trust 1141 * negative cache entries for up to nm_negnametimeo 1142 * seconds. 1143 */ 1144 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) && 1145 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 && 1146 timespeccmp(&vattr.va_mtime, &nctime, ==)) { 1147 NFSINCRGLOBAL(newnfsstats.lookupcache_hits); 1148 return (ENOENT); 1149 } 1150 cache_purge_negative(dvp); 1151 } 1152 1153 error = 0; 1154 newvp = NULLVP; 1155 NFSINCRGLOBAL(newnfsstats.lookupcache_misses); 1156 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1157 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1158 NULL); 1159 if (dattrflag) 1160 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1161 if (error) { 1162 if (newvp != NULLVP) { 1163 vput(newvp); 1164 *vpp = NULLVP; 1165 } 1166 1167 if (error != ENOENT) { 1168 if (NFS_ISV4(dvp)) 1169 error = nfscl_maperr(td, error, (uid_t)0, 1170 (gid_t)0); 1171 return (error); 1172 } 1173 1174 /* The requested file was not found. */ 1175 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1176 (flags & ISLASTCN)) { 1177 /* 1178 * XXX: UFS does a full VOP_ACCESS(dvp, 1179 * VWRITE) here instead of just checking 1180 * MNT_RDONLY. 1181 */ 1182 if (mp->mnt_flag & MNT_RDONLY) 1183 return (EROFS); 1184 cnp->cn_flags |= SAVENAME; 1185 return (EJUSTRETURN); 1186 } 1187 1188 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE && 1189 dattrflag) { 1190 /* 1191 * Cache the modification time of the parent 1192 * directory from the post-op attributes in 1193 * the name cache entry. The negative cache 1194 * entry will be ignored once the directory 1195 * has changed. Don't bother adding the entry 1196 * if the directory has already changed. 1197 */ 1198 mtx_lock(&np->n_mtx); 1199 if (timespeccmp(&np->n_vattr.na_mtime, 1200 &dnfsva.na_mtime, ==)) { 1201 mtx_unlock(&np->n_mtx); 1202 cache_enter_time(dvp, NULL, cnp, 1203 &dnfsva.na_mtime, NULL); 1204 } else 1205 mtx_unlock(&np->n_mtx); 1206 } 1207 return (ENOENT); 1208 } 1209 1210 /* 1211 * Handle RENAME case... 1212 */ 1213 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 1214 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1215 FREE((caddr_t)nfhp, M_NFSFH); 1216 return (EISDIR); 1217 } 1218 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1219 LK_EXCLUSIVE); 1220 if (error) 1221 return (error); 1222 newvp = NFSTOV(np); 1223 if (attrflag) 1224 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1225 0, 1); 1226 *vpp = newvp; 1227 cnp->cn_flags |= SAVENAME; 1228 return (0); 1229 } 1230 1231 if (flags & ISDOTDOT) { 1232 ltype = NFSVOPISLOCKED(dvp); 1233 error = vfs_busy(mp, MBF_NOWAIT); 1234 if (error != 0) { 1235 vfs_ref(mp); 1236 NFSVOPUNLOCK(dvp, 0); 1237 error = vfs_busy(mp, 0); 1238 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1239 vfs_rel(mp); 1240 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) { 1241 vfs_unbusy(mp); 1242 error = ENOENT; 1243 } 1244 if (error != 0) 1245 return (error); 1246 } 1247 NFSVOPUNLOCK(dvp, 0); 1248 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1249 cnp->cn_lkflags); 1250 if (error == 0) 1251 newvp = NFSTOV(np); 1252 vfs_unbusy(mp); 1253 if (newvp != dvp) 1254 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1255 if (dvp->v_iflag & VI_DOOMED) { 1256 if (error == 0) { 1257 if (newvp == dvp) 1258 vrele(newvp); 1259 else 1260 vput(newvp); 1261 } 1262 error = ENOENT; 1263 } 1264 if (error != 0) 1265 return (error); 1266 if (attrflag) 1267 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1268 0, 1); 1269 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1270 FREE((caddr_t)nfhp, M_NFSFH); 1271 VREF(dvp); 1272 newvp = dvp; 1273 if (attrflag) 1274 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1275 0, 1); 1276 } else { 1277 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1278 cnp->cn_lkflags); 1279 if (error) 1280 return (error); 1281 newvp = NFSTOV(np); 1282 if (attrflag) 1283 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1284 0, 1); 1285 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1286 !(np->n_flag & NMODIFIED)) { 1287 /* 1288 * Flush the attribute cache when opening a 1289 * leaf node to ensure that fresh attributes 1290 * are fetched in nfs_open() since we did not 1291 * fetch attributes from the LOOKUP reply. 1292 */ 1293 mtx_lock(&np->n_mtx); 1294 np->n_attrstamp = 0; 1295 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1296 mtx_unlock(&np->n_mtx); 1297 } 1298 } 1299 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1300 cnp->cn_flags |= SAVENAME; 1301 if ((cnp->cn_flags & MAKEENTRY) && 1302 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) && 1303 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0)) 1304 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 1305 newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime); 1306 *vpp = newvp; 1307 return (0); 1308} 1309 1310/* 1311 * nfs read call. 1312 * Just call ncl_bioread() to do the work. 1313 */ 1314static int 1315nfs_read(struct vop_read_args *ap) 1316{ 1317 struct vnode *vp = ap->a_vp; 1318 1319 switch (vp->v_type) { 1320 case VREG: 1321 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred)); 1322 case VDIR: 1323 return (EISDIR); 1324 default: 1325 return (EOPNOTSUPP); 1326 } 1327} 1328 1329/* 1330 * nfs readlink call 1331 */ 1332static int 1333nfs_readlink(struct vop_readlink_args *ap) 1334{ 1335 struct vnode *vp = ap->a_vp; 1336 1337 if (vp->v_type != VLNK) 1338 return (EINVAL); 1339 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred)); 1340} 1341 1342/* 1343 * Do a readlink rpc. 1344 * Called by ncl_doio() from below the buffer cache. 1345 */ 1346int 1347ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1348{ 1349 int error, ret, attrflag; 1350 struct nfsvattr nfsva; 1351 1352 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva, 1353 &attrflag, NULL); 1354 if (attrflag) { 1355 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1356 if (ret && !error) 1357 error = ret; 1358 } 1359 if (error && NFS_ISV4(vp)) 1360 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1361 return (error); 1362} 1363 1364/* 1365 * nfs read rpc call 1366 * Ditto above 1367 */ 1368int 1369ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1370{ 1371 int error, ret, attrflag; 1372 struct nfsvattr nfsva; 1373 struct nfsmount *nmp; 1374 1375 nmp = VFSTONFS(vnode_mount(vp)); 1376 error = EIO; 1377 attrflag = 0; 1378 if (NFSHASPNFS(nmp)) 1379 error = nfscl_doiods(vp, uiop, NULL, NULL, 1380 NFSV4OPEN_ACCESSREAD, cred, uiop->uio_td); 1381 NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error); 1382 if (error != 0) 1383 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, 1384 &attrflag, NULL); 1385 if (attrflag) { 1386 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1387 if (ret && !error) 1388 error = ret; 1389 } 1390 if (error && NFS_ISV4(vp)) 1391 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1392 return (error); 1393} 1394 1395/* 1396 * nfs write call 1397 */ 1398int 1399ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 1400 int *iomode, int *must_commit, int called_from_strategy) 1401{ 1402 struct nfsvattr nfsva; 1403 int error, attrflag, ret; 1404 struct nfsmount *nmp; 1405 1406 nmp = VFSTONFS(vnode_mount(vp)); 1407 error = EIO; 1408 attrflag = 0; 1409 if (NFSHASPNFS(nmp)) 1410 error = nfscl_doiods(vp, uiop, iomode, must_commit, 1411 NFSV4OPEN_ACCESSWRITE, cred, uiop->uio_td); 1412 NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error); 1413 if (error != 0) 1414 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred, 1415 uiop->uio_td, &nfsva, &attrflag, NULL, 1416 called_from_strategy); 1417 if (attrflag) { 1418 if (VTONFS(vp)->n_flag & ND_NFSV4) 1419 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1, 1420 1); 1421 else 1422 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1423 1); 1424 if (ret && !error) 1425 error = ret; 1426 } 1427 if (DOINGASYNC(vp)) 1428 *iomode = NFSWRITE_FILESYNC; 1429 if (error && NFS_ISV4(vp)) 1430 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1431 return (error); 1432} 1433 1434/* 1435 * nfs mknod rpc 1436 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1437 * mode set to specify the file type and the size field for rdev. 1438 */ 1439static int 1440nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1441 struct vattr *vap) 1442{ 1443 struct nfsvattr nfsva, dnfsva; 1444 struct vnode *newvp = NULL; 1445 struct nfsnode *np = NULL, *dnp; 1446 struct nfsfh *nfhp; 1447 struct vattr vattr; 1448 int error = 0, attrflag, dattrflag; 1449 u_int32_t rdev; 1450 1451 if (vap->va_type == VCHR || vap->va_type == VBLK) 1452 rdev = vap->va_rdev; 1453 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1454 rdev = 0xffffffff; 1455 else 1456 return (EOPNOTSUPP); 1457 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1458 return (error); 1459 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap, 1460 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva, 1461 &nfsva, &nfhp, &attrflag, &dattrflag, NULL); 1462 if (!error) { 1463 if (!nfhp) 1464 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1465 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, 1466 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1467 NULL); 1468 if (nfhp) 1469 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1470 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); 1471 } 1472 if (dattrflag) 1473 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1474 if (!error) { 1475 newvp = NFSTOV(np); 1476 if (attrflag != 0) { 1477 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1478 0, 1); 1479 if (error != 0) 1480 vput(newvp); 1481 } 1482 } 1483 if (!error) { 1484 *vpp = newvp; 1485 } else if (NFS_ISV4(dvp)) { 1486 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, 1487 vap->va_gid); 1488 } 1489 dnp = VTONFS(dvp); 1490 mtx_lock(&dnp->n_mtx); 1491 dnp->n_flag |= NMODIFIED; 1492 if (!dattrflag) { 1493 dnp->n_attrstamp = 0; 1494 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1495 } 1496 mtx_unlock(&dnp->n_mtx); 1497 return (error); 1498} 1499 1500/* 1501 * nfs mknod vop 1502 * just call nfs_mknodrpc() to do the work. 1503 */ 1504/* ARGSUSED */ 1505static int 1506nfs_mknod(struct vop_mknod_args *ap) 1507{ 1508 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap)); 1509} 1510 1511static struct mtx nfs_cverf_mtx; 1512MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex", 1513 MTX_DEF); 1514 1515static nfsquad_t 1516nfs_get_cverf(void) 1517{ 1518 static nfsquad_t cverf; 1519 nfsquad_t ret; 1520 static int cverf_initialized = 0; 1521 1522 mtx_lock(&nfs_cverf_mtx); 1523 if (cverf_initialized == 0) { 1524 cverf.lval[0] = arc4random(); 1525 cverf.lval[1] = arc4random(); 1526 cverf_initialized = 1; 1527 } else 1528 cverf.qval++; 1529 ret = cverf; 1530 mtx_unlock(&nfs_cverf_mtx); 1531 1532 return (ret); 1533} 1534 1535/* 1536 * nfs file create call 1537 */ 1538static int 1539nfs_create(struct vop_create_args *ap) 1540{ 1541 struct vnode *dvp = ap->a_dvp; 1542 struct vattr *vap = ap->a_vap; 1543 struct componentname *cnp = ap->a_cnp; 1544 struct nfsnode *np = NULL, *dnp; 1545 struct vnode *newvp = NULL; 1546 struct nfsmount *nmp; 1547 struct nfsvattr dnfsva, nfsva; 1548 struct nfsfh *nfhp; 1549 nfsquad_t cverf; 1550 int error = 0, attrflag, dattrflag, fmode = 0; 1551 struct vattr vattr; 1552 1553 /* 1554 * Oops, not for me.. 1555 */ 1556 if (vap->va_type == VSOCK) 1557 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1558 1559 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1560 return (error); 1561 if (vap->va_vaflags & VA_EXCLUSIVE) 1562 fmode |= O_EXCL; 1563 dnp = VTONFS(dvp); 1564 nmp = VFSTONFS(vnode_mount(dvp)); 1565again: 1566 /* For NFSv4, wait until any remove is done. */ 1567 mtx_lock(&dnp->n_mtx); 1568 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) { 1569 dnp->n_flag |= NREMOVEWANT; 1570 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0); 1571 } 1572 mtx_unlock(&dnp->n_mtx); 1573 1574 cverf = nfs_get_cverf(); 1575 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1576 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, 1577 &nfhp, &attrflag, &dattrflag, NULL); 1578 if (!error) { 1579 if (nfhp == NULL) 1580 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1581 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, 1582 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1583 NULL); 1584 if (nfhp != NULL) 1585 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1586 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); 1587 } 1588 if (dattrflag) 1589 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1590 if (!error) { 1591 newvp = NFSTOV(np); 1592 if (attrflag == 0) 1593 error = nfsrpc_getattr(newvp, cnp->cn_cred, 1594 cnp->cn_thread, &nfsva, NULL); 1595 if (error == 0) 1596 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1597 0, 1); 1598 } 1599 if (error) { 1600 if (newvp != NULL) { 1601 vput(newvp); 1602 newvp = NULL; 1603 } 1604 if (NFS_ISV34(dvp) && (fmode & O_EXCL) && 1605 error == NFSERR_NOTSUPP) { 1606 fmode &= ~O_EXCL; 1607 goto again; 1608 } 1609 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) { 1610 if (nfscl_checksattr(vap, &nfsva)) { 1611 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred, 1612 cnp->cn_thread, &nfsva, &attrflag, NULL); 1613 if (error && (vap->va_uid != (uid_t)VNOVAL || 1614 vap->va_gid != (gid_t)VNOVAL)) { 1615 /* try again without setting uid/gid */ 1616 vap->va_uid = (uid_t)VNOVAL; 1617 vap->va_gid = (uid_t)VNOVAL; 1618 error = nfsrpc_setattr(newvp, vap, NULL, 1619 cnp->cn_cred, cnp->cn_thread, &nfsva, 1620 &attrflag, NULL); 1621 } 1622 if (attrflag) 1623 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 1624 NULL, 0, 1); 1625 if (error != 0) 1626 vput(newvp); 1627 } 1628 } 1629 if (!error) { 1630 if ((cnp->cn_flags & MAKEENTRY) && attrflag) 1631 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 1632 NULL); 1633 *ap->a_vpp = newvp; 1634 } else if (NFS_ISV4(dvp)) { 1635 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, 1636 vap->va_gid); 1637 } 1638 mtx_lock(&dnp->n_mtx); 1639 dnp->n_flag |= NMODIFIED; 1640 if (!dattrflag) { 1641 dnp->n_attrstamp = 0; 1642 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1643 } 1644 mtx_unlock(&dnp->n_mtx); 1645 return (error); 1646} 1647 1648/* 1649 * nfs file remove call 1650 * To try and make nfs semantics closer to ufs semantics, a file that has 1651 * other processes using the vnode is renamed instead of removed and then 1652 * removed later on the last close. 1653 * - If v_usecount > 1 1654 * If a rename is not already in the works 1655 * call nfs_sillyrename() to set it up 1656 * else 1657 * do the remove rpc 1658 */ 1659static int 1660nfs_remove(struct vop_remove_args *ap) 1661{ 1662 struct vnode *vp = ap->a_vp; 1663 struct vnode *dvp = ap->a_dvp; 1664 struct componentname *cnp = ap->a_cnp; 1665 struct nfsnode *np = VTONFS(vp); 1666 int error = 0; 1667 struct vattr vattr; 1668 1669 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name")); 1670 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount")); 1671 if (vp->v_type == VDIR) 1672 error = EPERM; 1673 else if (vrefcnt(vp) == 1 || (np->n_sillyrename && 1674 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 1675 vattr.va_nlink > 1)) { 1676 /* 1677 * Purge the name cache so that the chance of a lookup for 1678 * the name succeeding while the remove is in progress is 1679 * minimized. Without node locking it can still happen, such 1680 * that an I/O op returns ESTALE, but since you get this if 1681 * another host removes the file.. 1682 */ 1683 cache_purge(vp); 1684 /* 1685 * throw away biocache buffers, mainly to avoid 1686 * unnecessary delayed writes later. 1687 */ 1688 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1); 1689 /* Do the rpc */ 1690 if (error != EINTR && error != EIO) 1691 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr, 1692 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread); 1693 /* 1694 * Kludge City: If the first reply to the remove rpc is lost.. 1695 * the reply to the retransmitted request will be ENOENT 1696 * since the file was in fact removed 1697 * Therefore, we cheat and return success. 1698 */ 1699 if (error == ENOENT) 1700 error = 0; 1701 } else if (!np->n_sillyrename) 1702 error = nfs_sillyrename(dvp, vp, cnp); 1703 mtx_lock(&np->n_mtx); 1704 np->n_attrstamp = 0; 1705 mtx_unlock(&np->n_mtx); 1706 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 1707 return (error); 1708} 1709 1710/* 1711 * nfs file remove rpc called from nfs_inactive 1712 */ 1713int 1714ncl_removeit(struct sillyrename *sp, struct vnode *vp) 1715{ 1716 /* 1717 * Make sure that the directory vnode is still valid. 1718 * XXX we should lock sp->s_dvp here. 1719 */ 1720 if (sp->s_dvp->v_type == VBAD) 1721 return (0); 1722 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen, 1723 sp->s_cred, NULL)); 1724} 1725 1726/* 1727 * Nfs remove rpc, called from nfs_remove() and ncl_removeit(). 1728 */ 1729static int 1730nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 1731 int namelen, struct ucred *cred, struct thread *td) 1732{ 1733 struct nfsvattr dnfsva; 1734 struct nfsnode *dnp = VTONFS(dvp); 1735 int error = 0, dattrflag; 1736 1737 mtx_lock(&dnp->n_mtx); 1738 dnp->n_flag |= NREMOVEINPROG; 1739 mtx_unlock(&dnp->n_mtx); 1740 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva, 1741 &dattrflag, NULL); 1742 mtx_lock(&dnp->n_mtx); 1743 if ((dnp->n_flag & NREMOVEWANT)) { 1744 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG); 1745 mtx_unlock(&dnp->n_mtx); 1746 wakeup((caddr_t)dnp); 1747 } else { 1748 dnp->n_flag &= ~NREMOVEINPROG; 1749 mtx_unlock(&dnp->n_mtx); 1750 } 1751 if (dattrflag) 1752 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1753 mtx_lock(&dnp->n_mtx); 1754 dnp->n_flag |= NMODIFIED; 1755 if (!dattrflag) { 1756 dnp->n_attrstamp = 0; 1757 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1758 } 1759 mtx_unlock(&dnp->n_mtx); 1760 if (error && NFS_ISV4(dvp)) 1761 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1762 return (error); 1763} 1764 1765/* 1766 * nfs file rename call 1767 */ 1768static int 1769nfs_rename(struct vop_rename_args *ap) 1770{ 1771 struct vnode *fvp = ap->a_fvp; 1772 struct vnode *tvp = ap->a_tvp; 1773 struct vnode *fdvp = ap->a_fdvp; 1774 struct vnode *tdvp = ap->a_tdvp; 1775 struct componentname *tcnp = ap->a_tcnp; 1776 struct componentname *fcnp = ap->a_fcnp; 1777 struct nfsnode *fnp = VTONFS(ap->a_fvp); 1778 struct nfsnode *tdnp = VTONFS(ap->a_tdvp); 1779 struct nfsv4node *newv4 = NULL; 1780 int error; 1781 1782 KASSERT((tcnp->cn_flags & HASBUF) != 0 && 1783 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name")); 1784 /* Check for cross-device rename */ 1785 if ((fvp->v_mount != tdvp->v_mount) || 1786 (tvp && (fvp->v_mount != tvp->v_mount))) { 1787 error = EXDEV; 1788 goto out; 1789 } 1790 1791 if (fvp == tvp) { 1792 ncl_printf("nfs_rename: fvp == tvp (can't happen)\n"); 1793 error = 0; 1794 goto out; 1795 } 1796 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0) 1797 goto out; 1798 1799 /* 1800 * We have to flush B_DELWRI data prior to renaming 1801 * the file. If we don't, the delayed-write buffers 1802 * can be flushed out later after the file has gone stale 1803 * under NFSV3. NFSV2 does not have this problem because 1804 * ( as far as I can tell ) it flushes dirty buffers more 1805 * often. 1806 * 1807 * Skip the rename operation if the fsync fails, this can happen 1808 * due to the server's volume being full, when we pushed out data 1809 * that was written back to our cache earlier. Not checking for 1810 * this condition can result in potential (silent) data loss. 1811 */ 1812 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread); 1813 NFSVOPUNLOCK(fvp, 0); 1814 if (!error && tvp) 1815 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread); 1816 if (error) 1817 goto out; 1818 1819 /* 1820 * If the tvp exists and is in use, sillyrename it before doing the 1821 * rename of the new file over it. 1822 * XXX Can't sillyrename a directory. 1823 */ 1824 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename && 1825 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1826 vput(tvp); 1827 tvp = NULL; 1828 } 1829 1830 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1831 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1832 tcnp->cn_thread); 1833 1834 if (error == 0 && NFS_ISV4(tdvp)) { 1835 /* 1836 * For NFSv4, check to see if it is the same name and 1837 * replace the name, if it is different. 1838 */ 1839 MALLOC(newv4, struct nfsv4node *, 1840 sizeof (struct nfsv4node) + 1841 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1, 1842 M_NFSV4NODE, M_WAITOK); 1843 mtx_lock(&tdnp->n_mtx); 1844 mtx_lock(&fnp->n_mtx); 1845 if (fnp->n_v4 != NULL && fvp->v_type == VREG && 1846 (fnp->n_v4->n4_namelen != tcnp->cn_namelen || 1847 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4), 1848 tcnp->cn_namelen) || 1849 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen || 1850 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 1851 tdnp->n_fhp->nfh_len))) { 1852#ifdef notdef 1853{ char nnn[100]; int nnnl; 1854nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99; 1855bcopy(tcnp->cn_nameptr, nnn, nnnl); 1856nnn[nnnl] = '\0'; 1857printf("ren replace=%s\n",nnn); 1858} 1859#endif 1860 FREE((caddr_t)fnp->n_v4, M_NFSV4NODE); 1861 fnp->n_v4 = newv4; 1862 newv4 = NULL; 1863 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len; 1864 fnp->n_v4->n4_namelen = tcnp->cn_namelen; 1865 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 1866 tdnp->n_fhp->nfh_len); 1867 NFSBCOPY(tcnp->cn_nameptr, 1868 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen); 1869 } 1870 mtx_unlock(&tdnp->n_mtx); 1871 mtx_unlock(&fnp->n_mtx); 1872 if (newv4 != NULL) 1873 FREE((caddr_t)newv4, M_NFSV4NODE); 1874 } 1875 1876 if (fvp->v_type == VDIR) { 1877 if (tvp != NULL && tvp->v_type == VDIR) 1878 cache_purge(tdvp); 1879 cache_purge(fdvp); 1880 } 1881 1882out: 1883 if (tdvp == tvp) 1884 vrele(tdvp); 1885 else 1886 vput(tdvp); 1887 if (tvp) 1888 vput(tvp); 1889 vrele(fdvp); 1890 vrele(fvp); 1891 /* 1892 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1893 */ 1894 if (error == ENOENT) 1895 error = 0; 1896 return (error); 1897} 1898 1899/* 1900 * nfs file rename rpc called from nfs_remove() above 1901 */ 1902static int 1903nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp, 1904 struct sillyrename *sp) 1905{ 1906 1907 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen, 1908 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred, 1909 scnp->cn_thread)); 1910} 1911 1912/* 1913 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1914 */ 1915static int 1916nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr, 1917 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr, 1918 int tnamelen, struct ucred *cred, struct thread *td) 1919{ 1920 struct nfsvattr fnfsva, tnfsva; 1921 struct nfsnode *fdnp = VTONFS(fdvp); 1922 struct nfsnode *tdnp = VTONFS(tdvp); 1923 int error = 0, fattrflag, tattrflag; 1924 1925 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp, 1926 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag, 1927 &tattrflag, NULL, NULL); 1928 mtx_lock(&fdnp->n_mtx); 1929 fdnp->n_flag |= NMODIFIED; 1930 if (fattrflag != 0) { 1931 mtx_unlock(&fdnp->n_mtx); 1932 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1); 1933 } else { 1934 fdnp->n_attrstamp = 0; 1935 mtx_unlock(&fdnp->n_mtx); 1936 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp); 1937 } 1938 mtx_lock(&tdnp->n_mtx); 1939 tdnp->n_flag |= NMODIFIED; 1940 if (tattrflag != 0) { 1941 mtx_unlock(&tdnp->n_mtx); 1942 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1); 1943 } else { 1944 tdnp->n_attrstamp = 0; 1945 mtx_unlock(&tdnp->n_mtx); 1946 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 1947 } 1948 if (error && NFS_ISV4(fdvp)) 1949 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1950 return (error); 1951} 1952 1953/* 1954 * nfs hard link create call 1955 */ 1956static int 1957nfs_link(struct vop_link_args *ap) 1958{ 1959 struct vnode *vp = ap->a_vp; 1960 struct vnode *tdvp = ap->a_tdvp; 1961 struct componentname *cnp = ap->a_cnp; 1962 struct nfsnode *np, *tdnp; 1963 struct nfsvattr nfsva, dnfsva; 1964 int error = 0, attrflag, dattrflag; 1965 1966 /* 1967 * Push all writes to the server, so that the attribute cache 1968 * doesn't get "out of sync" with the server. 1969 * XXX There should be a better way! 1970 */ 1971 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread); 1972 1973 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 1974 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag, 1975 &dattrflag, NULL); 1976 tdnp = VTONFS(tdvp); 1977 mtx_lock(&tdnp->n_mtx); 1978 tdnp->n_flag |= NMODIFIED; 1979 if (dattrflag != 0) { 1980 mtx_unlock(&tdnp->n_mtx); 1981 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1); 1982 } else { 1983 tdnp->n_attrstamp = 0; 1984 mtx_unlock(&tdnp->n_mtx); 1985 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 1986 } 1987 if (attrflag) 1988 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1989 else { 1990 np = VTONFS(vp); 1991 mtx_lock(&np->n_mtx); 1992 np->n_attrstamp = 0; 1993 mtx_unlock(&np->n_mtx); 1994 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 1995 } 1996 /* 1997 * If negative lookup caching is enabled, I might as well 1998 * add an entry for this node. Not necessary for correctness, 1999 * but if negative caching is enabled, then the system 2000 * must care about lookup caching hit rate, so... 2001 */ 2002 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 && 2003 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 2004 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL); 2005 } 2006 if (error && NFS_ISV4(vp)) 2007 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, 2008 (gid_t)0); 2009 return (error); 2010} 2011 2012/* 2013 * nfs symbolic link create call 2014 */ 2015static int 2016nfs_symlink(struct vop_symlink_args *ap) 2017{ 2018 struct vnode *dvp = ap->a_dvp; 2019 struct vattr *vap = ap->a_vap; 2020 struct componentname *cnp = ap->a_cnp; 2021 struct nfsvattr nfsva, dnfsva; 2022 struct nfsfh *nfhp; 2023 struct nfsnode *np = NULL, *dnp; 2024 struct vnode *newvp = NULL; 2025 int error = 0, attrflag, dattrflag, ret; 2026 2027 vap->va_type = VLNK; 2028 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2029 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, 2030 &nfsva, &nfhp, &attrflag, &dattrflag, NULL); 2031 if (nfhp) { 2032 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, 2033 &np, NULL, LK_EXCLUSIVE); 2034 if (!ret) 2035 newvp = NFSTOV(np); 2036 else if (!error) 2037 error = ret; 2038 } 2039 if (newvp != NULL) { 2040 if (attrflag) 2041 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 2042 0, 1); 2043 } else if (!error) { 2044 /* 2045 * If we do not have an error and we could not extract the 2046 * newvp from the response due to the request being NFSv2, we 2047 * have to do a lookup in order to obtain a newvp to return. 2048 */ 2049 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2050 cnp->cn_cred, cnp->cn_thread, &np); 2051 if (!error) 2052 newvp = NFSTOV(np); 2053 } 2054 if (error) { 2055 if (newvp) 2056 vput(newvp); 2057 if (NFS_ISV4(dvp)) 2058 error = nfscl_maperr(cnp->cn_thread, error, 2059 vap->va_uid, vap->va_gid); 2060 } else { 2061 *ap->a_vpp = newvp; 2062 } 2063 2064 dnp = VTONFS(dvp); 2065 mtx_lock(&dnp->n_mtx); 2066 dnp->n_flag |= NMODIFIED; 2067 if (dattrflag != 0) { 2068 mtx_unlock(&dnp->n_mtx); 2069 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2070 } else { 2071 dnp->n_attrstamp = 0; 2072 mtx_unlock(&dnp->n_mtx); 2073 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2074 } 2075 /* 2076 * If negative lookup caching is enabled, I might as well 2077 * add an entry for this node. Not necessary for correctness, 2078 * but if negative caching is enabled, then the system 2079 * must care about lookup caching hit rate, so... 2080 */ 2081 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2082 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 2083 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL); 2084 } 2085 return (error); 2086} 2087 2088/* 2089 * nfs make dir call 2090 */ 2091static int 2092nfs_mkdir(struct vop_mkdir_args *ap) 2093{ 2094 struct vnode *dvp = ap->a_dvp; 2095 struct vattr *vap = ap->a_vap; 2096 struct componentname *cnp = ap->a_cnp; 2097 struct nfsnode *np = NULL, *dnp; 2098 struct vnode *newvp = NULL; 2099 struct vattr vattr; 2100 struct nfsfh *nfhp; 2101 struct nfsvattr nfsva, dnfsva; 2102 int error = 0, attrflag, dattrflag, ret; 2103 2104 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) 2105 return (error); 2106 vap->va_type = VDIR; 2107 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2108 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp, 2109 &attrflag, &dattrflag, NULL); 2110 dnp = VTONFS(dvp); 2111 mtx_lock(&dnp->n_mtx); 2112 dnp->n_flag |= NMODIFIED; 2113 if (dattrflag != 0) { 2114 mtx_unlock(&dnp->n_mtx); 2115 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2116 } else { 2117 dnp->n_attrstamp = 0; 2118 mtx_unlock(&dnp->n_mtx); 2119 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2120 } 2121 if (nfhp) { 2122 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, 2123 &np, NULL, LK_EXCLUSIVE); 2124 if (!ret) { 2125 newvp = NFSTOV(np); 2126 if (attrflag) 2127 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 2128 NULL, 0, 1); 2129 } else if (!error) 2130 error = ret; 2131 } 2132 if (!error && newvp == NULL) { 2133 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2134 cnp->cn_cred, cnp->cn_thread, &np); 2135 if (!error) { 2136 newvp = NFSTOV(np); 2137 if (newvp->v_type != VDIR) 2138 error = EEXIST; 2139 } 2140 } 2141 if (error) { 2142 if (newvp) 2143 vput(newvp); 2144 if (NFS_ISV4(dvp)) 2145 error = nfscl_maperr(cnp->cn_thread, error, 2146 vap->va_uid, vap->va_gid); 2147 } else { 2148 /* 2149 * If negative lookup caching is enabled, I might as well 2150 * add an entry for this node. Not necessary for correctness, 2151 * but if negative caching is enabled, then the system 2152 * must care about lookup caching hit rate, so... 2153 */ 2154 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2155 (cnp->cn_flags & MAKEENTRY) && 2156 attrflag != 0 && dattrflag != 0) 2157 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 2158 &dnfsva.na_ctime); 2159 *ap->a_vpp = newvp; 2160 } 2161 return (error); 2162} 2163 2164/* 2165 * nfs remove directory call 2166 */ 2167static int 2168nfs_rmdir(struct vop_rmdir_args *ap) 2169{ 2170 struct vnode *vp = ap->a_vp; 2171 struct vnode *dvp = ap->a_dvp; 2172 struct componentname *cnp = ap->a_cnp; 2173 struct nfsnode *dnp; 2174 struct nfsvattr dnfsva; 2175 int error, dattrflag; 2176 2177 if (dvp == vp) 2178 return (EINVAL); 2179 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2180 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL); 2181 dnp = VTONFS(dvp); 2182 mtx_lock(&dnp->n_mtx); 2183 dnp->n_flag |= NMODIFIED; 2184 if (dattrflag != 0) { 2185 mtx_unlock(&dnp->n_mtx); 2186 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2187 } else { 2188 dnp->n_attrstamp = 0; 2189 mtx_unlock(&dnp->n_mtx); 2190 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2191 } 2192 2193 cache_purge(dvp); 2194 cache_purge(vp); 2195 if (error && NFS_ISV4(dvp)) 2196 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, 2197 (gid_t)0); 2198 /* 2199 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2200 */ 2201 if (error == ENOENT) 2202 error = 0; 2203 return (error); 2204} 2205 2206/* 2207 * nfs readdir call 2208 */ 2209static int 2210nfs_readdir(struct vop_readdir_args *ap) 2211{ 2212 struct vnode *vp = ap->a_vp; 2213 struct nfsnode *np = VTONFS(vp); 2214 struct uio *uio = ap->a_uio; 2215 ssize_t tresid; 2216 int error = 0; 2217 struct vattr vattr; 2218 2219 if (ap->a_eofflag != NULL) 2220 *ap->a_eofflag = 0; 2221 if (vp->v_type != VDIR) 2222 return(EPERM); 2223 2224 /* 2225 * First, check for hit on the EOF offset cache 2226 */ 2227 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2228 (np->n_flag & NMODIFIED) == 0) { 2229 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) { 2230 mtx_lock(&np->n_mtx); 2231 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) || 2232 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 2233 mtx_unlock(&np->n_mtx); 2234 NFSINCRGLOBAL(newnfsstats.direofcache_hits); 2235 if (ap->a_eofflag != NULL) 2236 *ap->a_eofflag = 1; 2237 return (0); 2238 } else 2239 mtx_unlock(&np->n_mtx); 2240 } 2241 } 2242 2243 /* 2244 * Call ncl_bioread() to do the real work. 2245 */ 2246 tresid = uio->uio_resid; 2247 error = ncl_bioread(vp, uio, 0, ap->a_cred); 2248 2249 if (!error && uio->uio_resid == tresid) { 2250 NFSINCRGLOBAL(newnfsstats.direofcache_misses); 2251 if (ap->a_eofflag != NULL) 2252 *ap->a_eofflag = 1; 2253 } 2254 return (error); 2255} 2256 2257/* 2258 * Readdir rpc call. 2259 * Called from below the buffer cache by ncl_doio(). 2260 */ 2261int 2262ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2263 struct thread *td) 2264{ 2265 struct nfsvattr nfsva; 2266 nfsuint64 *cookiep, cookie; 2267 struct nfsnode *dnp = VTONFS(vp); 2268 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2269 int error = 0, eof, attrflag; 2270 2271 KASSERT(uiop->uio_iovcnt == 1 && 2272 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2273 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2274 ("nfs readdirrpc bad uio")); 2275 2276 /* 2277 * If there is no cookie, assume directory was stale. 2278 */ 2279 ncl_dircookie_lock(dnp); 2280 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2281 if (cookiep) { 2282 cookie = *cookiep; 2283 ncl_dircookie_unlock(dnp); 2284 } else { 2285 ncl_dircookie_unlock(dnp); 2286 return (NFSERR_BAD_COOKIE); 2287 } 2288 2289 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2290 (void)ncl_fsinfo(nmp, vp, cred, td); 2291 2292 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva, 2293 &attrflag, &eof, NULL); 2294 if (attrflag) 2295 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2296 2297 if (!error) { 2298 /* 2299 * We are now either at the end of the directory or have filled 2300 * the block. 2301 */ 2302 if (eof) 2303 dnp->n_direofoffset = uiop->uio_offset; 2304 else { 2305 if (uiop->uio_resid > 0) 2306 ncl_printf("EEK! readdirrpc resid > 0\n"); 2307 ncl_dircookie_lock(dnp); 2308 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2309 *cookiep = cookie; 2310 ncl_dircookie_unlock(dnp); 2311 } 2312 } else if (NFS_ISV4(vp)) { 2313 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2314 } 2315 return (error); 2316} 2317 2318/* 2319 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc(). 2320 */ 2321int 2322ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2323 struct thread *td) 2324{ 2325 struct nfsvattr nfsva; 2326 nfsuint64 *cookiep, cookie; 2327 struct nfsnode *dnp = VTONFS(vp); 2328 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2329 int error = 0, attrflag, eof; 2330 2331 KASSERT(uiop->uio_iovcnt == 1 && 2332 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2333 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2334 ("nfs readdirplusrpc bad uio")); 2335 2336 /* 2337 * If there is no cookie, assume directory was stale. 2338 */ 2339 ncl_dircookie_lock(dnp); 2340 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2341 if (cookiep) { 2342 cookie = *cookiep; 2343 ncl_dircookie_unlock(dnp); 2344 } else { 2345 ncl_dircookie_unlock(dnp); 2346 return (NFSERR_BAD_COOKIE); 2347 } 2348 2349 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2350 (void)ncl_fsinfo(nmp, vp, cred, td); 2351 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva, 2352 &attrflag, &eof, NULL); 2353 if (attrflag) 2354 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2355 2356 if (!error) { 2357 /* 2358 * We are now either at end of the directory or have filled the 2359 * the block. 2360 */ 2361 if (eof) 2362 dnp->n_direofoffset = uiop->uio_offset; 2363 else { 2364 if (uiop->uio_resid > 0) 2365 ncl_printf("EEK! readdirplusrpc resid > 0\n"); 2366 ncl_dircookie_lock(dnp); 2367 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2368 *cookiep = cookie; 2369 ncl_dircookie_unlock(dnp); 2370 } 2371 } else if (NFS_ISV4(vp)) { 2372 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2373 } 2374 return (error); 2375} 2376 2377/* 2378 * Silly rename. To make the NFS filesystem that is stateless look a little 2379 * more like the "ufs" a remove of an active vnode is translated to a rename 2380 * to a funny looking filename that is removed by nfs_inactive on the 2381 * nfsnode. There is the potential for another process on a different client 2382 * to create the same funny name between the nfs_lookitup() fails and the 2383 * nfs_rename() completes, but... 2384 */ 2385static int 2386nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2387{ 2388 struct sillyrename *sp; 2389 struct nfsnode *np; 2390 int error; 2391 short pid; 2392 unsigned int lticks; 2393 2394 cache_purge(dvp); 2395 np = VTONFS(vp); 2396 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir")); 2397 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2398 M_NEWNFSREQ, M_WAITOK); 2399 sp->s_cred = crhold(cnp->cn_cred); 2400 sp->s_dvp = dvp; 2401 VREF(dvp); 2402 2403 /* 2404 * Fudge together a funny name. 2405 * Changing the format of the funny name to accomodate more 2406 * sillynames per directory. 2407 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is 2408 * CPU ticks since boot. 2409 */ 2410 pid = cnp->cn_thread->td_proc->p_pid; 2411 lticks = (unsigned int)ticks; 2412 for ( ; ; ) { 2413 sp->s_namlen = sprintf(sp->s_name, 2414 ".nfs.%08x.%04x4.4", lticks, 2415 pid); 2416 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2417 cnp->cn_thread, NULL)) 2418 break; 2419 lticks++; 2420 } 2421 error = nfs_renameit(dvp, vp, cnp, sp); 2422 if (error) 2423 goto bad; 2424 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2425 cnp->cn_thread, &np); 2426 np->n_sillyrename = sp; 2427 return (0); 2428bad: 2429 vrele(sp->s_dvp); 2430 crfree(sp->s_cred); 2431 free((caddr_t)sp, M_NEWNFSREQ); 2432 return (error); 2433} 2434 2435/* 2436 * Look up a file name and optionally either update the file handle or 2437 * allocate an nfsnode, depending on the value of npp. 2438 * npp == NULL --> just do the lookup 2439 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2440 * handled too 2441 * *npp != NULL --> update the file handle in the vnode 2442 */ 2443static int 2444nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred, 2445 struct thread *td, struct nfsnode **npp) 2446{ 2447 struct vnode *newvp = NULL, *vp; 2448 struct nfsnode *np, *dnp = VTONFS(dvp); 2449 struct nfsfh *nfhp, *onfhp; 2450 struct nfsvattr nfsva, dnfsva; 2451 struct componentname cn; 2452 int error = 0, attrflag, dattrflag; 2453 u_int hash; 2454 2455 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva, 2456 &nfhp, &attrflag, &dattrflag, NULL); 2457 if (dattrflag) 2458 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2459 if (npp && !error) { 2460 if (*npp != NULL) { 2461 np = *npp; 2462 vp = NFSTOV(np); 2463 /* 2464 * For NFSv4, check to see if it is the same name and 2465 * replace the name, if it is different. 2466 */ 2467 if (np->n_v4 != NULL && nfsva.na_type == VREG && 2468 (np->n_v4->n4_namelen != len || 2469 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) || 2470 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || 2471 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2472 dnp->n_fhp->nfh_len))) { 2473#ifdef notdef 2474{ char nnn[100]; int nnnl; 2475nnnl = (len < 100) ? len : 99; 2476bcopy(name, nnn, nnnl); 2477nnn[nnnl] = '\0'; 2478printf("replace=%s\n",nnn); 2479} 2480#endif 2481 FREE((caddr_t)np->n_v4, M_NFSV4NODE); 2482 MALLOC(np->n_v4, struct nfsv4node *, 2483 sizeof (struct nfsv4node) + 2484 dnp->n_fhp->nfh_len + len - 1, 2485 M_NFSV4NODE, M_WAITOK); 2486 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; 2487 np->n_v4->n4_namelen = len; 2488 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2489 dnp->n_fhp->nfh_len); 2490 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len); 2491 } 2492 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, 2493 FNV1_32_INIT); 2494 onfhp = np->n_fhp; 2495 /* 2496 * Rehash node for new file handle. 2497 */ 2498 vfs_hash_rehash(vp, hash); 2499 np->n_fhp = nfhp; 2500 if (onfhp != NULL) 2501 FREE((caddr_t)onfhp, M_NFSFH); 2502 newvp = NFSTOV(np); 2503 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) { 2504 FREE((caddr_t)nfhp, M_NFSFH); 2505 VREF(dvp); 2506 newvp = dvp; 2507 } else { 2508 cn.cn_nameptr = name; 2509 cn.cn_namelen = len; 2510 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td, 2511 &np, NULL, LK_EXCLUSIVE); 2512 if (error) 2513 return (error); 2514 newvp = NFSTOV(np); 2515 } 2516 if (!attrflag && *npp == NULL) { 2517 if (newvp == dvp) 2518 vrele(newvp); 2519 else 2520 vput(newvp); 2521 return (ENOENT); 2522 } 2523 if (attrflag) 2524 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 2525 0, 1); 2526 } 2527 if (npp && *npp == NULL) { 2528 if (error) { 2529 if (newvp) { 2530 if (newvp == dvp) 2531 vrele(newvp); 2532 else 2533 vput(newvp); 2534 } 2535 } else 2536 *npp = np; 2537 } 2538 if (error && NFS_ISV4(dvp)) 2539 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2540 return (error); 2541} 2542 2543/* 2544 * Nfs Version 3 and 4 commit rpc 2545 */ 2546int 2547ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred, 2548 struct thread *td) 2549{ 2550 struct nfsvattr nfsva; 2551 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2552 int error, attrflag; 2553 2554 mtx_lock(&nmp->nm_mtx); 2555 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) { 2556 mtx_unlock(&nmp->nm_mtx); 2557 return (0); 2558 } 2559 mtx_unlock(&nmp->nm_mtx); 2560 error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva, 2561 &attrflag, NULL); 2562 if (attrflag != 0) 2563 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 2564 0, 1); 2565 if (error != 0 && NFS_ISV4(vp)) 2566 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2567 return (error); 2568} 2569 2570/* 2571 * Strategy routine. 2572 * For async requests when nfsiod(s) are running, queue the request by 2573 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the 2574 * request. 2575 */ 2576static int 2577nfs_strategy(struct vop_strategy_args *ap) 2578{ 2579 struct buf *bp = ap->a_bp; 2580 struct ucred *cr; 2581 2582 KASSERT(!(bp->b_flags & B_DONE), 2583 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); 2584 BUF_ASSERT_HELD(bp); 2585 2586 if (bp->b_iocmd == BIO_READ) 2587 cr = bp->b_rcred; 2588 else 2589 cr = bp->b_wcred; 2590 2591 /* 2592 * If the op is asynchronous and an i/o daemon is waiting 2593 * queue the request, wake it up and wait for completion 2594 * otherwise just do it ourselves. 2595 */ 2596 if ((bp->b_flags & B_ASYNC) == 0 || 2597 ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread)) 2598 (void) ncl_doio(ap->a_vp, bp, cr, curthread, 1); 2599 return (0); 2600} 2601 2602/* 2603 * fsync vnode op. Just call ncl_flush() with commit == 1. 2604 */ 2605/* ARGSUSED */ 2606static int 2607nfs_fsync(struct vop_fsync_args *ap) 2608{ 2609 2610 if (ap->a_vp->v_type != VREG) { 2611 /* 2612 * For NFS, metadata is changed synchronously on the server, 2613 * so there is nothing to flush. Also, ncl_flush() clears 2614 * the NMODIFIED flag and that shouldn't be done here for 2615 * directories. 2616 */ 2617 return (0); 2618 } 2619 return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1, 0)); 2620} 2621 2622/* 2623 * Flush all the blocks associated with a vnode. 2624 * Walk through the buffer pool and push any dirty pages 2625 * associated with the vnode. 2626 * If the called_from_renewthread argument is TRUE, it has been called 2627 * from the NFSv4 renew thread and, as such, cannot block indefinitely 2628 * waiting for a buffer write to complete. 2629 */ 2630int 2631ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td, 2632 int commit, int called_from_renewthread) 2633{ 2634 struct nfsnode *np = VTONFS(vp); 2635 struct buf *bp; 2636 int i; 2637 struct buf *nbp; 2638 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2639 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos; 2640 int passone = 1, trycnt = 0; 2641 u_quad_t off, endoff, toff; 2642 struct ucred* wcred = NULL; 2643 struct buf **bvec = NULL; 2644 struct bufobj *bo; 2645#ifndef NFS_COMMITBVECSIZ 2646#define NFS_COMMITBVECSIZ 20 2647#endif 2648 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ]; 2649 int bvecsize = 0, bveccount; 2650 2651 if (called_from_renewthread != 0) 2652 slptimeo = hz; 2653 if (nmp->nm_flag & NFSMNT_INT) 2654 slpflag = PCATCH; 2655 if (!commit) 2656 passone = 0; 2657 bo = &vp->v_bufobj; 2658 /* 2659 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the 2660 * server, but has not been committed to stable storage on the server 2661 * yet. On the first pass, the byte range is worked out and the commit 2662 * rpc is done. On the second pass, ncl_writebp() is called to do the 2663 * job. 2664 */ 2665again: 2666 off = (u_quad_t)-1; 2667 endoff = 0; 2668 bvecpos = 0; 2669 if (NFS_ISV34(vp) && commit) { 2670 if (bvec != NULL && bvec != bvec_on_stack) 2671 free(bvec, M_TEMP); 2672 /* 2673 * Count up how many buffers waiting for a commit. 2674 */ 2675 bveccount = 0; 2676 BO_LOCK(bo); 2677 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2678 if (!BUF_ISLOCKED(bp) && 2679 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) 2680 == (B_DELWRI | B_NEEDCOMMIT)) 2681 bveccount++; 2682 } 2683 /* 2684 * Allocate space to remember the list of bufs to commit. It is 2685 * important to use M_NOWAIT here to avoid a race with nfs_write. 2686 * If we can't get memory (for whatever reason), we will end up 2687 * committing the buffers one-by-one in the loop below. 2688 */ 2689 if (bveccount > NFS_COMMITBVECSIZ) { 2690 /* 2691 * Release the vnode interlock to avoid a lock 2692 * order reversal. 2693 */ 2694 BO_UNLOCK(bo); 2695 bvec = (struct buf **) 2696 malloc(bveccount * sizeof(struct buf *), 2697 M_TEMP, M_NOWAIT); 2698 BO_LOCK(bo); 2699 if (bvec == NULL) { 2700 bvec = bvec_on_stack; 2701 bvecsize = NFS_COMMITBVECSIZ; 2702 } else 2703 bvecsize = bveccount; 2704 } else { 2705 bvec = bvec_on_stack; 2706 bvecsize = NFS_COMMITBVECSIZ; 2707 } 2708 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2709 if (bvecpos >= bvecsize) 2710 break; 2711 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2712 nbp = TAILQ_NEXT(bp, b_bobufs); 2713 continue; 2714 } 2715 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 2716 (B_DELWRI | B_NEEDCOMMIT)) { 2717 BUF_UNLOCK(bp); 2718 nbp = TAILQ_NEXT(bp, b_bobufs); 2719 continue; 2720 } 2721 BO_UNLOCK(bo); 2722 bremfree(bp); 2723 /* 2724 * Work out if all buffers are using the same cred 2725 * so we can deal with them all with one commit. 2726 * 2727 * NOTE: we are not clearing B_DONE here, so we have 2728 * to do it later on in this routine if we intend to 2729 * initiate I/O on the bp. 2730 * 2731 * Note: to avoid loopback deadlocks, we do not 2732 * assign b_runningbufspace. 2733 */ 2734 if (wcred == NULL) 2735 wcred = bp->b_wcred; 2736 else if (wcred != bp->b_wcred) 2737 wcred = NOCRED; 2738 vfs_busy_pages(bp, 1); 2739 2740 BO_LOCK(bo); 2741 /* 2742 * bp is protected by being locked, but nbp is not 2743 * and vfs_busy_pages() may sleep. We have to 2744 * recalculate nbp. 2745 */ 2746 nbp = TAILQ_NEXT(bp, b_bobufs); 2747 2748 /* 2749 * A list of these buffers is kept so that the 2750 * second loop knows which buffers have actually 2751 * been committed. This is necessary, since there 2752 * may be a race between the commit rpc and new 2753 * uncommitted writes on the file. 2754 */ 2755 bvec[bvecpos++] = bp; 2756 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2757 bp->b_dirtyoff; 2758 if (toff < off) 2759 off = toff; 2760 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); 2761 if (toff > endoff) 2762 endoff = toff; 2763 } 2764 BO_UNLOCK(bo); 2765 } 2766 if (bvecpos > 0) { 2767 /* 2768 * Commit data on the server, as required. 2769 * If all bufs are using the same wcred, then use that with 2770 * one call for all of them, otherwise commit each one 2771 * separately. 2772 */ 2773 if (wcred != NOCRED) 2774 retv = ncl_commit(vp, off, (int)(endoff - off), 2775 wcred, td); 2776 else { 2777 retv = 0; 2778 for (i = 0; i < bvecpos; i++) { 2779 off_t off, size; 2780 bp = bvec[i]; 2781 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2782 bp->b_dirtyoff; 2783 size = (u_quad_t)(bp->b_dirtyend 2784 - bp->b_dirtyoff); 2785 retv = ncl_commit(vp, off, (int)size, 2786 bp->b_wcred, td); 2787 if (retv) break; 2788 } 2789 } 2790 2791 if (retv == NFSERR_STALEWRITEVERF) 2792 ncl_clearcommit(vp->v_mount); 2793 2794 /* 2795 * Now, either mark the blocks I/O done or mark the 2796 * blocks dirty, depending on whether the commit 2797 * succeeded. 2798 */ 2799 for (i = 0; i < bvecpos; i++) { 2800 bp = bvec[i]; 2801 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 2802 if (retv) { 2803 /* 2804 * Error, leave B_DELWRI intact 2805 */ 2806 vfs_unbusy_pages(bp); 2807 brelse(bp); 2808 } else { 2809 /* 2810 * Success, remove B_DELWRI ( bundirty() ). 2811 * 2812 * b_dirtyoff/b_dirtyend seem to be NFS 2813 * specific. We should probably move that 2814 * into bundirty(). XXX 2815 */ 2816 bufobj_wref(bo); 2817 bp->b_flags |= B_ASYNC; 2818 bundirty(bp); 2819 bp->b_flags &= ~B_DONE; 2820 bp->b_ioflags &= ~BIO_ERROR; 2821 bp->b_dirtyoff = bp->b_dirtyend = 0; 2822 bufdone(bp); 2823 } 2824 } 2825 } 2826 2827 /* 2828 * Start/do any write(s) that are required. 2829 */ 2830loop: 2831 BO_LOCK(bo); 2832 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2833 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2834 if (waitfor != MNT_WAIT || passone) 2835 continue; 2836 2837 error = BUF_TIMELOCK(bp, 2838 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 2839 BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo); 2840 if (error == 0) { 2841 BUF_UNLOCK(bp); 2842 goto loop; 2843 } 2844 if (error == ENOLCK) { 2845 error = 0; 2846 goto loop; 2847 } 2848 if (called_from_renewthread != 0) { 2849 /* 2850 * Return EIO so the flush will be retried 2851 * later. 2852 */ 2853 error = EIO; 2854 goto done; 2855 } 2856 if (newnfs_sigintr(nmp, td)) { 2857 error = EINTR; 2858 goto done; 2859 } 2860 if (slpflag == PCATCH) { 2861 slpflag = 0; 2862 slptimeo = 2 * hz; 2863 } 2864 goto loop; 2865 } 2866 if ((bp->b_flags & B_DELWRI) == 0) 2867 panic("nfs_fsync: not dirty"); 2868 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) { 2869 BUF_UNLOCK(bp); 2870 continue; 2871 } 2872 BO_UNLOCK(bo); 2873 bremfree(bp); 2874 if (passone || !commit) 2875 bp->b_flags |= B_ASYNC; 2876 else 2877 bp->b_flags |= B_ASYNC; 2878 bwrite(bp); 2879 if (newnfs_sigintr(nmp, td)) { 2880 error = EINTR; 2881 goto done; 2882 } 2883 goto loop; 2884 } 2885 if (passone) { 2886 passone = 0; 2887 BO_UNLOCK(bo); 2888 goto again; 2889 } 2890 if (waitfor == MNT_WAIT) { 2891 while (bo->bo_numoutput) { 2892 error = bufobj_wwait(bo, slpflag, slptimeo); 2893 if (error) { 2894 BO_UNLOCK(bo); 2895 if (called_from_renewthread != 0) { 2896 /* 2897 * Return EIO so that the flush will be 2898 * retried later. 2899 */ 2900 error = EIO; 2901 goto done; 2902 } 2903 error = newnfs_sigintr(nmp, td); 2904 if (error) 2905 goto done; 2906 if (slpflag == PCATCH) { 2907 slpflag = 0; 2908 slptimeo = 2 * hz; 2909 } 2910 BO_LOCK(bo); 2911 } 2912 } 2913 if (bo->bo_dirty.bv_cnt != 0 && commit) { 2914 BO_UNLOCK(bo); 2915 goto loop; 2916 } 2917 /* 2918 * Wait for all the async IO requests to drain 2919 */ 2920 BO_UNLOCK(bo); 2921 mtx_lock(&np->n_mtx); 2922 while (np->n_directio_asyncwr > 0) { 2923 np->n_flag |= NFSYNCWAIT; 2924 error = newnfs_msleep(td, &np->n_directio_asyncwr, 2925 &np->n_mtx, slpflag | (PRIBIO + 1), 2926 "nfsfsync", 0); 2927 if (error) { 2928 if (newnfs_sigintr(nmp, td)) { 2929 mtx_unlock(&np->n_mtx); 2930 error = EINTR; 2931 goto done; 2932 } 2933 } 2934 } 2935 mtx_unlock(&np->n_mtx); 2936 } else 2937 BO_UNLOCK(bo); 2938 if (NFSHASPNFS(nmp)) { 2939 nfscl_layoutcommit(vp, td); 2940 /* 2941 * Invalidate the attribute cache, since writes to a DS 2942 * won't update the size attribute. 2943 */ 2944 mtx_lock(&np->n_mtx); 2945 np->n_attrstamp = 0; 2946 } else 2947 mtx_lock(&np->n_mtx); 2948 if (np->n_flag & NWRITEERR) { 2949 error = np->n_error; 2950 np->n_flag &= ~NWRITEERR; 2951 } 2952 if (commit && bo->bo_dirty.bv_cnt == 0 && 2953 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0) 2954 np->n_flag &= ~NMODIFIED; 2955 mtx_unlock(&np->n_mtx); 2956done: 2957 if (bvec != NULL && bvec != bvec_on_stack) 2958 free(bvec, M_TEMP); 2959 if (error == 0 && commit != 0 && waitfor == MNT_WAIT && 2960 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 || 2961 np->n_directio_asyncwr != 0) && trycnt++ < 5) { 2962 /* try, try again... */ 2963 passone = 1; 2964 wcred = NULL; 2965 bvec = NULL; 2966 bvecsize = 0; 2967printf("try%d\n", trycnt); 2968 goto again; 2969 } 2970 return (error); 2971} 2972 2973/* 2974 * NFS advisory byte-level locks. 2975 */ 2976static int 2977nfs_advlock(struct vop_advlock_args *ap) 2978{ 2979 struct vnode *vp = ap->a_vp; 2980 struct ucred *cred; 2981 struct nfsnode *np = VTONFS(ap->a_vp); 2982 struct proc *p = (struct proc *)ap->a_id; 2983 struct thread *td = curthread; /* XXX */ 2984 struct vattr va; 2985 int ret, error = EOPNOTSUPP; 2986 u_quad_t size; 2987 2988 if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) { 2989 if (vp->v_type != VREG) 2990 return (EINVAL); 2991 if ((ap->a_flags & F_POSIX) != 0) 2992 cred = p->p_ucred; 2993 else 2994 cred = td->td_ucred; 2995 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 2996 if (vp->v_iflag & VI_DOOMED) { 2997 NFSVOPUNLOCK(vp, 0); 2998 return (EBADF); 2999 } 3000 3001 /* 3002 * If this is unlocking a write locked region, flush and 3003 * commit them before unlocking. This is required by 3004 * RFC3530 Sec. 9.3.2. 3005 */ 3006 if (ap->a_op == F_UNLCK && 3007 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id, 3008 ap->a_flags)) 3009 (void) ncl_flush(vp, MNT_WAIT, cred, td, 1, 0); 3010 3011 /* 3012 * Loop around doing the lock op, while a blocking lock 3013 * must wait for the lock op to succeed. 3014 */ 3015 do { 3016 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op, 3017 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags); 3018 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3019 ap->a_op == F_SETLK) { 3020 NFSVOPUNLOCK(vp, 0); 3021 error = nfs_catnap(PZERO | PCATCH, ret, 3022 "ncladvl"); 3023 if (error) 3024 return (EINTR); 3025 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 3026 if (vp->v_iflag & VI_DOOMED) { 3027 NFSVOPUNLOCK(vp, 0); 3028 return (EBADF); 3029 } 3030 } 3031 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3032 ap->a_op == F_SETLK); 3033 if (ret == NFSERR_DENIED) { 3034 NFSVOPUNLOCK(vp, 0); 3035 return (EAGAIN); 3036 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) { 3037 NFSVOPUNLOCK(vp, 0); 3038 return (ret); 3039 } else if (ret != 0) { 3040 NFSVOPUNLOCK(vp, 0); 3041 return (EACCES); 3042 } 3043 3044 /* 3045 * Now, if we just got a lock, invalidate data in the buffer 3046 * cache, as required, so that the coherency conforms with 3047 * RFC3530 Sec. 9.3.2. 3048 */ 3049 if (ap->a_op == F_SETLK) { 3050 if ((np->n_flag & NMODIFIED) == 0) { 3051 np->n_attrstamp = 0; 3052 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3053 ret = VOP_GETATTR(vp, &va, cred); 3054 } 3055 if ((np->n_flag & NMODIFIED) || ret || 3056 np->n_change != va.va_filerev) { 3057 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1); 3058 np->n_attrstamp = 0; 3059 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3060 ret = VOP_GETATTR(vp, &va, cred); 3061 if (!ret) { 3062 np->n_mtime = va.va_mtime; 3063 np->n_change = va.va_filerev; 3064 } 3065 } 3066 /* Mark that a file lock has been acquired. */ 3067 mtx_lock(&np->n_mtx); 3068 np->n_flag |= NHASBEENLOCKED; 3069 mtx_unlock(&np->n_mtx); 3070 } 3071 NFSVOPUNLOCK(vp, 0); 3072 return (0); 3073 } else if (!NFS_ISV4(vp)) { 3074 error = NFSVOPLOCK(vp, LK_SHARED); 3075 if (error) 3076 return (error); 3077 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3078 size = VTONFS(vp)->n_size; 3079 NFSVOPUNLOCK(vp, 0); 3080 error = lf_advlock(ap, &(vp->v_lockf), size); 3081 } else { 3082 if (nfs_advlock_p != NULL) 3083 error = nfs_advlock_p(ap); 3084 else { 3085 NFSVOPUNLOCK(vp, 0); 3086 error = ENOLCK; 3087 } 3088 } 3089 if (error == 0 && ap->a_op == F_SETLK) { 3090 /* Mark that a file lock has been acquired. */ 3091 mtx_lock(&np->n_mtx); 3092 np->n_flag |= NHASBEENLOCKED; 3093 mtx_unlock(&np->n_mtx); 3094 } 3095 } 3096 return (error); 3097} 3098 3099/* 3100 * NFS advisory byte-level locks. 3101 */ 3102static int 3103nfs_advlockasync(struct vop_advlockasync_args *ap) 3104{ 3105 struct vnode *vp = ap->a_vp; 3106 u_quad_t size; 3107 int error; 3108 3109 if (NFS_ISV4(vp)) 3110 return (EOPNOTSUPP); 3111 error = NFSVOPLOCK(vp, LK_SHARED); 3112 if (error) 3113 return (error); 3114 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3115 size = VTONFS(vp)->n_size; 3116 NFSVOPUNLOCK(vp, 0); 3117 error = lf_advlockasync(ap, &(vp->v_lockf), size); 3118 } else { 3119 NFSVOPUNLOCK(vp, 0); 3120 error = EOPNOTSUPP; 3121 } 3122 return (error); 3123} 3124 3125/* 3126 * Print out the contents of an nfsnode. 3127 */ 3128static int 3129nfs_print(struct vop_print_args *ap) 3130{ 3131 struct vnode *vp = ap->a_vp; 3132 struct nfsnode *np = VTONFS(vp); 3133 3134 ncl_printf("\tfileid %ld fsid 0x%x", 3135 np->n_vattr.na_fileid, np->n_vattr.na_fsid); 3136 if (vp->v_type == VFIFO) 3137 fifo_printinfo(vp); 3138 printf("\n"); 3139 return (0); 3140} 3141 3142/* 3143 * This is the "real" nfs::bwrite(struct buf*). 3144 * We set B_CACHE if this is a VMIO buffer. 3145 */ 3146int 3147ncl_writebp(struct buf *bp, int force __unused, struct thread *td) 3148{ 3149 int s; 3150 int oldflags = bp->b_flags; 3151#if 0 3152 int retv = 1; 3153 off_t off; 3154#endif 3155 3156 BUF_ASSERT_HELD(bp); 3157 3158 if (bp->b_flags & B_INVAL) { 3159 brelse(bp); 3160 return(0); 3161 } 3162 3163 bp->b_flags |= B_CACHE; 3164 3165 /* 3166 * Undirty the bp. We will redirty it later if the I/O fails. 3167 */ 3168 3169 s = splbio(); 3170 bundirty(bp); 3171 bp->b_flags &= ~B_DONE; 3172 bp->b_ioflags &= ~BIO_ERROR; 3173 bp->b_iocmd = BIO_WRITE; 3174 3175 bufobj_wref(bp->b_bufobj); 3176 curthread->td_ru.ru_oublock++; 3177 splx(s); 3178 3179 /* 3180 * Note: to avoid loopback deadlocks, we do not 3181 * assign b_runningbufspace. 3182 */ 3183 vfs_busy_pages(bp, 1); 3184 3185 BUF_KERNPROC(bp); 3186 bp->b_iooffset = dbtob(bp->b_blkno); 3187 bstrategy(bp); 3188 3189 if( (oldflags & B_ASYNC) == 0) { 3190 int rtval = bufwait(bp); 3191 3192 if (oldflags & B_DELWRI) { 3193 s = splbio(); 3194 reassignbuf(bp); 3195 splx(s); 3196 } 3197 brelse(bp); 3198 return (rtval); 3199 } 3200 3201 return (0); 3202} 3203 3204/* 3205 * nfs special file access vnode op. 3206 * Essentially just get vattr and then imitate iaccess() since the device is 3207 * local to the client. 3208 */ 3209static int 3210nfsspec_access(struct vop_access_args *ap) 3211{ 3212 struct vattr *vap; 3213 struct ucred *cred = ap->a_cred; 3214 struct vnode *vp = ap->a_vp; 3215 accmode_t accmode = ap->a_accmode; 3216 struct vattr vattr; 3217 int error; 3218 3219 /* 3220 * Disallow write attempts on filesystems mounted read-only; 3221 * unless the file is a socket, fifo, or a block or character 3222 * device resident on the filesystem. 3223 */ 3224 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3225 switch (vp->v_type) { 3226 case VREG: 3227 case VDIR: 3228 case VLNK: 3229 return (EROFS); 3230 default: 3231 break; 3232 } 3233 } 3234 vap = &vattr; 3235 error = VOP_GETATTR(vp, vap, cred); 3236 if (error) 3237 goto out; 3238 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid, 3239 accmode, cred, NULL); 3240out: 3241 return error; 3242} 3243 3244/* 3245 * Read wrapper for fifos. 3246 */ 3247static int 3248nfsfifo_read(struct vop_read_args *ap) 3249{ 3250 struct nfsnode *np = VTONFS(ap->a_vp); 3251 int error; 3252 3253 /* 3254 * Set access flag. 3255 */ 3256 mtx_lock(&np->n_mtx); 3257 np->n_flag |= NACC; 3258 vfs_timestamp(&np->n_atim); 3259 mtx_unlock(&np->n_mtx); 3260 error = fifo_specops.vop_read(ap); 3261 return error; 3262} 3263 3264/* 3265 * Write wrapper for fifos. 3266 */ 3267static int 3268nfsfifo_write(struct vop_write_args *ap) 3269{ 3270 struct nfsnode *np = VTONFS(ap->a_vp); 3271 3272 /* 3273 * Set update flag. 3274 */ 3275 mtx_lock(&np->n_mtx); 3276 np->n_flag |= NUPD; 3277 vfs_timestamp(&np->n_mtim); 3278 mtx_unlock(&np->n_mtx); 3279 return(fifo_specops.vop_write(ap)); 3280} 3281 3282/* 3283 * Close wrapper for fifos. 3284 * 3285 * Update the times on the nfsnode then do fifo close. 3286 */ 3287static int 3288nfsfifo_close(struct vop_close_args *ap) 3289{ 3290 struct vnode *vp = ap->a_vp; 3291 struct nfsnode *np = VTONFS(vp); 3292 struct vattr vattr; 3293 struct timespec ts; 3294 3295 mtx_lock(&np->n_mtx); 3296 if (np->n_flag & (NACC | NUPD)) { 3297 vfs_timestamp(&ts); 3298 if (np->n_flag & NACC) 3299 np->n_atim = ts; 3300 if (np->n_flag & NUPD) 3301 np->n_mtim = ts; 3302 np->n_flag |= NCHG; 3303 if (vrefcnt(vp) == 1 && 3304 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3305 VATTR_NULL(&vattr); 3306 if (np->n_flag & NACC) 3307 vattr.va_atime = np->n_atim; 3308 if (np->n_flag & NUPD) 3309 vattr.va_mtime = np->n_mtim; 3310 mtx_unlock(&np->n_mtx); 3311 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3312 goto out; 3313 } 3314 } 3315 mtx_unlock(&np->n_mtx); 3316out: 3317 return (fifo_specops.vop_close(ap)); 3318} 3319 3320/* 3321 * Just call ncl_writebp() with the force argument set to 1. 3322 * 3323 * NOTE: B_DONE may or may not be set in a_bp on call. 3324 */ 3325static int 3326nfs_bwrite(struct buf *bp) 3327{ 3328 3329 return (ncl_writebp(bp, 1, curthread)); 3330} 3331 3332struct buf_ops buf_ops_newnfs = { 3333 .bop_name = "buf_ops_nfs", 3334 .bop_write = nfs_bwrite, 3335 .bop_strategy = bufstrategy, 3336 .bop_sync = bufsync, 3337 .bop_bdflush = bufbdflush, 3338}; 3339 3340/* 3341 * Cloned from vop_stdlock(), and then the ugly hack added. 3342 */ 3343static int 3344nfs_lock1(struct vop_lock1_args *ap) 3345{ 3346 struct vnode *vp = ap->a_vp; 3347 int error = 0; 3348 3349 /* 3350 * Since vfs_hash_get() calls vget() and it will no longer work 3351 * for FreeBSD8 with flags == 0, I can only think of this horrible 3352 * hack to work around it. I call vfs_hash_get() with LK_EXCLOTHER 3353 * and then handle it here. All I want for this case is a v_usecount 3354 * on the vnode to use for recovery, while another thread might 3355 * hold a lock on the vnode. I have the other threads blocked, so 3356 * there isn't any race problem. 3357 */ 3358 if ((ap->a_flags & LK_TYPE_MASK) == LK_EXCLOTHER) { 3359 if ((ap->a_flags & LK_INTERLOCK) == 0) 3360 panic("ncllock1"); 3361 if ((vp->v_iflag & VI_DOOMED)) 3362 error = ENOENT; 3363 VI_UNLOCK(vp); 3364 return (error); 3365 } 3366 return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp), 3367 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file, 3368 ap->a_line)); 3369} 3370 3371static int 3372nfs_getacl(struct vop_getacl_args *ap) 3373{ 3374 int error; 3375 3376 if (ap->a_type != ACL_TYPE_NFS4) 3377 return (EOPNOTSUPP); 3378 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3379 NULL); 3380 if (error > NFSERR_STALE) { 3381 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3382 error = EPERM; 3383 } 3384 return (error); 3385} 3386 3387static int 3388nfs_setacl(struct vop_setacl_args *ap) 3389{ 3390 int error; 3391 3392 if (ap->a_type != ACL_TYPE_NFS4) 3393 return (EOPNOTSUPP); 3394 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3395 NULL); 3396 if (error > NFSERR_STALE) { 3397 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3398 error = EPERM; 3399 } 3400 return (error); 3401} 3402 3403/* 3404 * Return POSIX pathconf information applicable to nfs filesystems. 3405 */ 3406static int 3407nfs_pathconf(struct vop_pathconf_args *ap) 3408{ 3409 struct nfsv3_pathconf pc; 3410 struct nfsvattr nfsva; 3411 struct vnode *vp = ap->a_vp; 3412 struct thread *td = curthread; 3413 int attrflag, error; 3414 3415 if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX || 3416 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED || 3417 ap->a_name == _PC_NO_TRUNC)) || 3418 (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) { 3419 /* 3420 * Since only the above 4 a_names are returned by the NFSv3 3421 * Pathconf RPC, there is no point in doing it for others. 3422 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can 3423 * be used for _PC_NFS4_ACL as well. 3424 */ 3425 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva, 3426 &attrflag, NULL); 3427 if (attrflag != 0) 3428 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 3429 1); 3430 if (error != 0) 3431 return (error); 3432 } else { 3433 /* 3434 * For NFSv2 (or NFSv3 when not one of the above 4 a_names), 3435 * just fake them. 3436 */ 3437 pc.pc_linkmax = LINK_MAX; 3438 pc.pc_namemax = NFS_MAXNAMLEN; 3439 pc.pc_notrunc = 1; 3440 pc.pc_chownrestricted = 1; 3441 pc.pc_caseinsensitive = 0; 3442 pc.pc_casepreserving = 1; 3443 error = 0; 3444 } 3445 switch (ap->a_name) { 3446 case _PC_LINK_MAX: 3447 *ap->a_retval = pc.pc_linkmax; 3448 break; 3449 case _PC_NAME_MAX: 3450 *ap->a_retval = pc.pc_namemax; 3451 break; 3452 case _PC_PATH_MAX: 3453 *ap->a_retval = PATH_MAX; 3454 break; 3455 case _PC_PIPE_BUF: 3456 *ap->a_retval = PIPE_BUF; 3457 break; 3458 case _PC_CHOWN_RESTRICTED: 3459 *ap->a_retval = pc.pc_chownrestricted; 3460 break; 3461 case _PC_NO_TRUNC: 3462 *ap->a_retval = pc.pc_notrunc; 3463 break; 3464 case _PC_ACL_EXTENDED: 3465 *ap->a_retval = 0; 3466 break; 3467 case _PC_ACL_NFS4: 3468 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 && 3469 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL)) 3470 *ap->a_retval = 1; 3471 else 3472 *ap->a_retval = 0; 3473 break; 3474 case _PC_ACL_PATH_MAX: 3475 if (NFS_ISV4(vp)) 3476 *ap->a_retval = ACL_MAX_ENTRIES; 3477 else 3478 *ap->a_retval = 3; 3479 break; 3480 case _PC_MAC_PRESENT: 3481 *ap->a_retval = 0; 3482 break; 3483 case _PC_ASYNC_IO: 3484 /* _PC_ASYNC_IO should have been handled by upper layers. */ 3485 KASSERT(0, ("_PC_ASYNC_IO should not get here")); 3486 error = EINVAL; 3487 break; 3488 case _PC_PRIO_IO: 3489 *ap->a_retval = 0; 3490 break; 3491 case _PC_SYNC_IO: 3492 *ap->a_retval = 0; 3493 break; 3494 case _PC_ALLOC_SIZE_MIN: 3495 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize; 3496 break; 3497 case _PC_FILESIZEBITS: 3498 if (NFS_ISV34(vp)) 3499 *ap->a_retval = 64; 3500 else 3501 *ap->a_retval = 32; 3502 break; 3503 case _PC_REC_INCR_XFER_SIZE: 3504 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 3505 break; 3506 case _PC_REC_MAX_XFER_SIZE: 3507 *ap->a_retval = -1; /* means ``unlimited'' */ 3508 break; 3509 case _PC_REC_MIN_XFER_SIZE: 3510 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 3511 break; 3512 case _PC_REC_XFER_ALIGN: 3513 *ap->a_retval = PAGE_SIZE; 3514 break; 3515 case _PC_SYMLINK_MAX: 3516 *ap->a_retval = NFS_MAXPATHLEN; 3517 break; 3518 3519 default: 3520 error = EINVAL; 3521 break; 3522 } 3523 return (error); 3524} 3525 3526