nfs_clvnops.c revision 1.3
1 /* $NetBSD: nfs_clvnops.c,v 1.3 2016/11/18 08:31:30 pgoyette Exp $ */ 2 /*- 3 * Copyright (c) 1989, 1993 4 * The Regents of the University of California. All rights reserved. 5 * 6 * This code is derived from software contributed to Berkeley by 7 * Rick Macklem at The University of Guelph. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * from nfs_vnops.c 8.16 (Berkeley) 5/27/95 34 */ 35 36 #include <sys/cdefs.h> 37 /* __FBSDID("FreeBSD: head/sys/fs/nfsclient/nfs_clvnops.c 304026 2016-08-12 22:44:59Z rmacklem "); */ 38 __RCSID("$NetBSD: nfs_clvnops.c,v 1.3 2016/11/18 08:31:30 pgoyette Exp $"); 39 40 /* 41 * vnode op calls for Sun NFS version 2, 3 and 4 42 */ 43 44 #include "opt_inet.h" 45 46 #include <sys/param.h> 47 #include <sys/kernel.h> 48 #include <sys/systm.h> 49 #include <sys/resourcevar.h> 50 #include <sys/proc.h> 51 #include <sys/mount.h> 52 #include <sys/bio.h> 53 #include <sys/buf.h> 54 #include <sys/jail.h> 55 #include <sys/malloc.h> 56 #include <sys/mbuf.h> 57 #include <sys/namei.h> 58 #include <sys/socket.h> 59 #include <sys/vnode.h> 60 #include <sys/dirent.h> 61 #include <sys/fcntl.h> 62 #include <sys/lockf.h> 63 #include <sys/stat.h> 64 #include <sys/sysctl.h> 65 #include <sys/signalvar.h> 66 67 #include <vm/vm.h> 68 #include <vm/vm_extern.h> 69 #include <vm/vm_object.h> 70 71 #include <fs/nfs/nfsport.h> 72 #include <fs/nfsclient/nfsnode.h> 73 #include <fs/nfsclient/nfsmount.h> 74 #include <fs/nfsclient/nfs.h> 75 #include <fs/nfsclient/nfs_kdtrace.h> 76 77 #include <net/if.h> 78 #include <netinet/in.h> 79 #include <netinet/in_var.h> 80 81 #include <nfs/nfs_lock.h> 82 83 #ifdef KDTRACE_HOOKS 84 #include <sys/dtrace_bsd.h> 85 86 dtrace_nfsclient_accesscache_flush_probe_func_t 87 dtrace_nfscl_accesscache_flush_done_probe; 88 uint32_t nfscl_accesscache_flush_done_id; 89 90 dtrace_nfsclient_accesscache_get_probe_func_t 91 dtrace_nfscl_accesscache_get_hit_probe, 92 dtrace_nfscl_accesscache_get_miss_probe; 93 uint32_t nfscl_accesscache_get_hit_id; 94 uint32_t nfscl_accesscache_get_miss_id; 95 96 dtrace_nfsclient_accesscache_load_probe_func_t 97 dtrace_nfscl_accesscache_load_done_probe; 98 uint32_t nfscl_accesscache_load_done_id; 99 #endif /* !KDTRACE_HOOKS */ 100 101 /* Defs */ 102 #define TRUE 1 103 #define FALSE 0 104 105 extern struct nfsstatsv1 nfsstatsv1; 106 extern int nfsrv_useacl; 107 extern int nfscl_debuglevel; 108 MALLOC_DECLARE(M_NEWNFSREQ); 109 110 /* 111 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these 112 * calls are not in getblk() and brelse() so that they would not be necessary 113 * here. 114 */ 115 #ifndef B_VMIO 116 #define vfs_busy_pages(bp, f) 117 #endif 118 119 static vop_read_t nfsfifo_read; 120 static vop_write_t nfsfifo_write; 121 static vop_close_t nfsfifo_close; 122 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *, 123 struct thread *); 124 static vop_lookup_t nfs_lookup; 125 static vop_create_t nfs_create; 126 static vop_mknod_t nfs_mknod; 127 static vop_open_t nfs_open; 128 static vop_pathconf_t nfs_pathconf; 129 static vop_close_t nfs_close; 130 static vop_access_t nfs_access; 131 static vop_getattr_t nfs_getattr; 132 static vop_setattr_t nfs_setattr; 133 static vop_read_t nfs_read; 134 static vop_fsync_t nfs_fsync; 135 static vop_remove_t nfs_remove; 136 static vop_link_t nfs_link; 137 static vop_rename_t nfs_rename; 138 static vop_mkdir_t nfs_mkdir; 139 static vop_rmdir_t nfs_rmdir; 140 static vop_symlink_t nfs_symlink; 141 static vop_readdir_t nfs_readdir; 142 static vop_strategy_t nfs_strategy; 143 static int nfs_lookitup(struct vnode *, char *, int, 144 struct ucred *, struct thread *, struct nfsnode **); 145 static int nfs_sillyrename(struct vnode *, struct vnode *, 146 struct componentname *); 147 static vop_access_t nfsspec_access; 148 static vop_readlink_t nfs_readlink; 149 static vop_print_t nfs_print; 150 static vop_advlock_t nfs_advlock; 151 static vop_advlockasync_t nfs_advlockasync; 152 static vop_getacl_t nfs_getacl; 153 static vop_setacl_t nfs_setacl; 154 155 /* 156 * Global vfs data structures for nfs 157 */ 158 struct 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_lookup = nfs_lookup, 172 .vop_mkdir = nfs_mkdir, 173 .vop_mknod = nfs_mknod, 174 .vop_open = nfs_open, 175 .vop_pathconf = nfs_pathconf, 176 .vop_print = nfs_print, 177 .vop_read = nfs_read, 178 .vop_readdir = nfs_readdir, 179 .vop_readlink = nfs_readlink, 180 .vop_reclaim = ncl_reclaim, 181 .vop_remove = nfs_remove, 182 .vop_rename = nfs_rename, 183 .vop_rmdir = nfs_rmdir, 184 .vop_setattr = nfs_setattr, 185 .vop_strategy = nfs_strategy, 186 .vop_symlink = nfs_symlink, 187 .vop_write = ncl_write, 188 .vop_getacl = nfs_getacl, 189 .vop_setacl = nfs_setacl, 190 }; 191 192 struct vop_vector newnfs_fifoops = { 193 .vop_default = &fifo_specops, 194 .vop_access = nfsspec_access, 195 .vop_close = nfsfifo_close, 196 .vop_fsync = nfs_fsync, 197 .vop_getattr = nfs_getattr, 198 .vop_inactive = ncl_inactive, 199 .vop_print = nfs_print, 200 .vop_read = nfsfifo_read, 201 .vop_reclaim = ncl_reclaim, 202 .vop_setattr = nfs_setattr, 203 .vop_write = nfsfifo_write, 204 }; 205 206 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, 207 struct componentname *cnp, struct vattr *vap); 208 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 209 int namelen, struct ucred *cred, struct thread *td); 210 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, 211 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp, 212 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td); 213 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp, 214 struct componentname *scnp, struct sillyrename *sp); 215 216 /* 217 * Global variables 218 */ 219 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1)) 220 221 SYSCTL_DECL(_vfs_nfs); 222 223 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO; 224 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 225 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 226 227 static int nfs_prime_access_cache = 0; 228 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW, 229 &nfs_prime_access_cache, 0, 230 "Prime NFS ACCESS cache when fetching attributes"); 231 232 static int newnfs_commit_on_close = 0; 233 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW, 234 &newnfs_commit_on_close, 0, "write+commit on close, else only write"); 235 236 static int nfs_clean_pages_on_close = 1; 237 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW, 238 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close"); 239 240 int newnfs_directio_enable = 0; 241 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW, 242 &newnfs_directio_enable, 0, "Enable NFS directio"); 243 244 int nfs_keep_dirty_on_error; 245 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW, 246 &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned"); 247 248 /* 249 * This sysctl allows other processes to mmap a file that has been opened 250 * O_DIRECT by a process. In general, having processes mmap the file while 251 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow 252 * this by default to prevent DoS attacks - to prevent a malicious user from 253 * opening up files O_DIRECT preventing other users from mmap'ing these 254 * files. "Protected" environments where stricter consistency guarantees are 255 * required can disable this knob. The process that opened the file O_DIRECT 256 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not 257 * meaningful. 258 */ 259 int newnfs_directio_allow_mmap = 1; 260 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW, 261 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens"); 262 263 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \ 264 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \ 265 | NFSACCESS_DELETE | NFSACCESS_LOOKUP) 266 267 /* 268 * SMP Locking Note : 269 * The list of locks after the description of the lock is the ordering 270 * of other locks acquired with the lock held. 271 * np->n_mtx : Protects the fields in the nfsnode. 272 VM Object Lock 273 VI_MTX (acquired indirectly) 274 * nmp->nm_mtx : Protects the fields in the nfsmount. 275 rep->r_mtx 276 * ncl_iod_mutex : Global lock, protects shared nfsiod state. 277 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list. 278 nmp->nm_mtx 279 rep->r_mtx 280 * rep->r_mtx : Protects the fields in an nfsreq. 281 */ 282 283 static int 284 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td, 285 struct ucred *cred, u_int32_t *retmode) 286 { 287 int error = 0, attrflag, i, lrupos; 288 u_int32_t rmode; 289 struct nfsnode *np = VTONFS(vp); 290 struct nfsvattr nfsva; 291 292 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag, 293 &rmode, NULL); 294 if (attrflag) 295 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 296 if (!error) { 297 lrupos = 0; 298 mtx_lock(&np->n_mtx); 299 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 300 if (np->n_accesscache[i].uid == cred->cr_uid) { 301 np->n_accesscache[i].mode = rmode; 302 np->n_accesscache[i].stamp = time_second; 303 break; 304 } 305 if (i > 0 && np->n_accesscache[i].stamp < 306 np->n_accesscache[lrupos].stamp) 307 lrupos = i; 308 } 309 if (i == NFS_ACCESSCACHESIZE) { 310 np->n_accesscache[lrupos].uid = cred->cr_uid; 311 np->n_accesscache[lrupos].mode = rmode; 312 np->n_accesscache[lrupos].stamp = time_second; 313 } 314 mtx_unlock(&np->n_mtx); 315 if (retmode != NULL) 316 *retmode = rmode; 317 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0); 318 } else if (NFS_ISV4(vp)) { 319 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 320 } 321 #ifdef KDTRACE_HOOKS 322 if (error != 0) 323 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0, 324 error); 325 #endif 326 return (error); 327 } 328 329 /* 330 * nfs access vnode op. 331 * For nfs version 2, just return ok. File accesses may fail later. 332 * For nfs version 3, use the access rpc to check accessibility. If file modes 333 * are changed on the server, accesses might still fail later. 334 */ 335 static int 336 nfs_access(struct vop_access_args *ap) 337 { 338 struct vnode *vp = ap->a_vp; 339 int error = 0, i, gotahit; 340 u_int32_t mode, wmode, rmode; 341 int v34 = NFS_ISV34(vp); 342 struct nfsnode *np = VTONFS(vp); 343 344 /* 345 * Disallow write attempts on filesystems mounted read-only; 346 * unless the file is a socket, fifo, or a block or character 347 * device resident on the filesystem. 348 */ 349 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS | 350 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL | 351 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) { 352 switch (vp->v_type) { 353 case VREG: 354 case VDIR: 355 case VLNK: 356 return (EROFS); 357 default: 358 break; 359 } 360 } 361 /* 362 * For nfs v3 or v4, check to see if we have done this recently, and if 363 * so return our cached result instead of making an ACCESS call. 364 * If not, do an access rpc, otherwise you are stuck emulating 365 * ufs_access() locally using the vattr. This may not be correct, 366 * since the server may apply other access criteria such as 367 * client uid-->server uid mapping that we do not know about. 368 */ 369 if (v34) { 370 if (ap->a_accmode & VREAD) 371 mode = NFSACCESS_READ; 372 else 373 mode = 0; 374 if (vp->v_type != VDIR) { 375 if (ap->a_accmode & VWRITE) 376 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 377 if (ap->a_accmode & VAPPEND) 378 mode |= NFSACCESS_EXTEND; 379 if (ap->a_accmode & VEXEC) 380 mode |= NFSACCESS_EXECUTE; 381 if (ap->a_accmode & VDELETE) 382 mode |= NFSACCESS_DELETE; 383 } else { 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_LOOKUP; 390 if (ap->a_accmode & VDELETE) 391 mode |= NFSACCESS_DELETE; 392 if (ap->a_accmode & VDELETE_CHILD) 393 mode |= NFSACCESS_MODIFY; 394 } 395 /* XXX safety belt, only make blanket request if caching */ 396 if (nfsaccess_cache_timeout > 0) { 397 wmode = NFSACCESS_READ | NFSACCESS_MODIFY | 398 NFSACCESS_EXTEND | NFSACCESS_EXECUTE | 399 NFSACCESS_DELETE | NFSACCESS_LOOKUP; 400 } else { 401 wmode = mode; 402 } 403 404 /* 405 * Does our cached result allow us to give a definite yes to 406 * this request? 407 */ 408 gotahit = 0; 409 mtx_lock(&np->n_mtx); 410 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 411 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) { 412 if (time_second < (np->n_accesscache[i].stamp 413 + nfsaccess_cache_timeout) && 414 (np->n_accesscache[i].mode & mode) == mode) { 415 NFSINCRGLOBAL(nfsstatsv1.accesscache_hits); 416 gotahit = 1; 417 } 418 break; 419 } 420 } 421 mtx_unlock(&np->n_mtx); 422 #ifdef KDTRACE_HOOKS 423 if (gotahit != 0) 424 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp, 425 ap->a_cred->cr_uid, mode); 426 else 427 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp, 428 ap->a_cred->cr_uid, mode); 429 #endif 430 if (gotahit == 0) { 431 /* 432 * Either a no, or a don't know. Go to the wire. 433 */ 434 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses); 435 error = nfs34_access_otw(vp, wmode, ap->a_td, 436 ap->a_cred, &rmode); 437 if (!error && 438 (rmode & mode) != mode) 439 error = EACCES; 440 } 441 return (error); 442 } else { 443 if ((error = nfsspec_access(ap)) != 0) { 444 return (error); 445 } 446 /* 447 * Attempt to prevent a mapped root from accessing a file 448 * which it shouldn't. We try to read a byte from the file 449 * if the user is root and the file is not zero length. 450 * After calling nfsspec_access, we should have the correct 451 * file size cached. 452 */ 453 mtx_lock(&np->n_mtx); 454 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD) 455 && VTONFS(vp)->n_size > 0) { 456 struct iovec aiov; 457 struct uio auio; 458 char buf[1]; 459 460 mtx_unlock(&np->n_mtx); 461 aiov.iov_base = buf; 462 aiov.iov_len = 1; 463 auio.uio_iov = &aiov; 464 auio.uio_iovcnt = 1; 465 auio.uio_offset = 0; 466 auio.uio_resid = 1; 467 auio.uio_segflg = UIO_SYSSPACE; 468 auio.uio_rw = UIO_READ; 469 auio.uio_td = ap->a_td; 470 471 if (vp->v_type == VREG) 472 error = ncl_readrpc(vp, &auio, ap->a_cred); 473 else if (vp->v_type == VDIR) { 474 char* bp; 475 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 476 aiov.iov_base = bp; 477 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 478 error = ncl_readdirrpc(vp, &auio, ap->a_cred, 479 ap->a_td); 480 free(bp, M_TEMP); 481 } else if (vp->v_type == VLNK) 482 error = ncl_readlinkrpc(vp, &auio, ap->a_cred); 483 else 484 error = EACCES; 485 } else 486 mtx_unlock(&np->n_mtx); 487 return (error); 488 } 489 } 490 491 492 /* 493 * nfs open vnode op 494 * Check to see if the type is ok 495 * and that deletion is not in progress. 496 * For paged in text files, you will need to flush the page cache 497 * if consistency is lost. 498 */ 499 /* ARGSUSED */ 500 static int 501 nfs_open(struct vop_open_args *ap) 502 { 503 struct vnode *vp = ap->a_vp; 504 struct nfsnode *np = VTONFS(vp); 505 struct vattr vattr; 506 int error; 507 int fmode = ap->a_mode; 508 struct ucred *cred; 509 510 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) 511 return (EOPNOTSUPP); 512 513 /* 514 * For NFSv4, we need to do the Open Op before cache validation, 515 * so that we conform to RFC3530 Sec. 9.3.1. 516 */ 517 if (NFS_ISV4(vp)) { 518 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td); 519 if (error) { 520 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 521 (gid_t)0); 522 return (error); 523 } 524 } 525 526 /* 527 * Now, if this Open will be doing reading, re-validate/flush the 528 * cache, so that Close/Open coherency is maintained. 529 */ 530 mtx_lock(&np->n_mtx); 531 if (np->n_flag & NMODIFIED) { 532 mtx_unlock(&np->n_mtx); 533 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 534 if (error == EINTR || error == EIO) { 535 if (NFS_ISV4(vp)) 536 (void) nfsrpc_close(vp, 0, ap->a_td); 537 return (error); 538 } 539 mtx_lock(&np->n_mtx); 540 np->n_attrstamp = 0; 541 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 542 if (vp->v_type == VDIR) 543 np->n_direofoffset = 0; 544 mtx_unlock(&np->n_mtx); 545 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 546 if (error) { 547 if (NFS_ISV4(vp)) 548 (void) nfsrpc_close(vp, 0, ap->a_td); 549 return (error); 550 } 551 mtx_lock(&np->n_mtx); 552 np->n_mtime = vattr.va_mtime; 553 if (NFS_ISV4(vp)) 554 np->n_change = vattr.va_filerev; 555 } else { 556 mtx_unlock(&np->n_mtx); 557 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 558 if (error) { 559 if (NFS_ISV4(vp)) 560 (void) nfsrpc_close(vp, 0, ap->a_td); 561 return (error); 562 } 563 mtx_lock(&np->n_mtx); 564 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) || 565 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 566 if (vp->v_type == VDIR) 567 np->n_direofoffset = 0; 568 mtx_unlock(&np->n_mtx); 569 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 570 if (error == EINTR || error == EIO) { 571 if (NFS_ISV4(vp)) 572 (void) nfsrpc_close(vp, 0, ap->a_td); 573 return (error); 574 } 575 mtx_lock(&np->n_mtx); 576 np->n_mtime = vattr.va_mtime; 577 if (NFS_ISV4(vp)) 578 np->n_change = vattr.va_filerev; 579 } 580 } 581 582 /* 583 * If the object has >= 1 O_DIRECT active opens, we disable caching. 584 */ 585 if (newnfs_directio_enable && (fmode & O_DIRECT) && 586 (vp->v_type == VREG)) { 587 if (np->n_directio_opens == 0) { 588 mtx_unlock(&np->n_mtx); 589 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 590 if (error) { 591 if (NFS_ISV4(vp)) 592 (void) nfsrpc_close(vp, 0, ap->a_td); 593 return (error); 594 } 595 mtx_lock(&np->n_mtx); 596 np->n_flag |= NNONCACHE; 597 } 598 np->n_directio_opens++; 599 } 600 601 /* If opened for writing via NFSv4.1 or later, mark that for pNFS. */ 602 if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0) 603 np->n_flag |= NWRITEOPENED; 604 605 /* 606 * If this is an open for writing, capture a reference to the 607 * credentials, so they can be used by ncl_putpages(). Using 608 * these write credentials is preferable to the credentials of 609 * whatever thread happens to be doing the VOP_PUTPAGES() since 610 * the write RPCs are less likely to fail with EACCES. 611 */ 612 if ((fmode & FWRITE) != 0) { 613 cred = np->n_writecred; 614 np->n_writecred = crhold(ap->a_cred); 615 } else 616 cred = NULL; 617 mtx_unlock(&np->n_mtx); 618 619 if (cred != NULL) 620 crfree(cred); 621 vnode_create_vobject(vp, vattr.va_size, ap->a_td); 622 return (0); 623 } 624 625 /* 626 * nfs close vnode op 627 * What an NFS client should do upon close after writing is a debatable issue. 628 * Most NFS clients push delayed writes to the server upon close, basically for 629 * two reasons: 630 * 1 - So that any write errors may be reported back to the client process 631 * doing the close system call. By far the two most likely errors are 632 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 633 * 2 - To put a worst case upper bound on cache inconsistency between 634 * multiple clients for the file. 635 * There is also a consistency problem for Version 2 of the protocol w.r.t. 636 * not being able to tell if other clients are writing a file concurrently, 637 * since there is no way of knowing if the changed modify time in the reply 638 * is only due to the write for this client. 639 * (NFS Version 3 provides weak cache consistency data in the reply that 640 * should be sufficient to detect and handle this case.) 641 * 642 * The current code does the following: 643 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 644 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 645 * or commit them (this satisfies 1 and 2 except for the 646 * case where the server crashes after this close but 647 * before the commit RPC, which is felt to be "good 648 * enough". Changing the last argument to ncl_flush() to 649 * a 1 would force a commit operation, if it is felt a 650 * commit is necessary now. 651 * for NFS Version 4 - flush the dirty buffers and commit them, if 652 * nfscl_mustflush() says this is necessary. 653 * It is necessary if there is no write delegation held, 654 * in order to satisfy open/close coherency. 655 * If the file isn't cached on local stable storage, 656 * it may be necessary in order to detect "out of space" 657 * errors from the server, if the write delegation 658 * issued by the server doesn't allow the file to grow. 659 */ 660 /* ARGSUSED */ 661 static int 662 nfs_close(struct vop_close_args *ap) 663 { 664 struct vnode *vp = ap->a_vp; 665 struct nfsnode *np = VTONFS(vp); 666 struct nfsvattr nfsva; 667 struct ucred *cred; 668 int error = 0, ret, localcred = 0; 669 int fmode = ap->a_fflag; 670 671 if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF)) 672 return (0); 673 /* 674 * During shutdown, a_cred isn't valid, so just use root. 675 */ 676 if (ap->a_cred == NOCRED) { 677 cred = newnfs_getcred(); 678 localcred = 1; 679 } else { 680 cred = ap->a_cred; 681 } 682 if (vp->v_type == VREG) { 683 /* 684 * Examine and clean dirty pages, regardless of NMODIFIED. 685 * This closes a major hole in close-to-open consistency. 686 * We want to push out all dirty pages (and buffers) on 687 * close, regardless of whether they were dirtied by 688 * mmap'ed writes or via write(). 689 */ 690 if (nfs_clean_pages_on_close && vp->v_object) { 691 VM_OBJECT_WLOCK(vp->v_object); 692 vm_object_page_clean(vp->v_object, 0, 0, 0); 693 VM_OBJECT_WUNLOCK(vp->v_object); 694 } 695 mtx_lock(&np->n_mtx); 696 if (np->n_flag & NMODIFIED) { 697 mtx_unlock(&np->n_mtx); 698 if (NFS_ISV3(vp)) { 699 /* 700 * Under NFSv3 we have dirty buffers to dispose of. We 701 * must flush them to the NFS server. We have the option 702 * of waiting all the way through the commit rpc or just 703 * waiting for the initial write. The default is to only 704 * wait through the initial write so the data is in the 705 * server's cache, which is roughly similar to the state 706 * a standard disk subsystem leaves the file in on close(). 707 * 708 * We cannot clear the NMODIFIED bit in np->n_flag due to 709 * potential races with other processes, and certainly 710 * cannot clear it if we don't commit. 711 * These races occur when there is no longer the old 712 * traditional vnode locking implemented for Vnode Ops. 713 */ 714 int cm = newnfs_commit_on_close ? 1 : 0; 715 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm, 0); 716 /* np->n_flag &= ~NMODIFIED; */ 717 } else if (NFS_ISV4(vp)) { 718 if (nfscl_mustflush(vp) != 0) { 719 int cm = newnfs_commit_on_close ? 1 : 0; 720 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, 721 cm, 0); 722 /* 723 * as above w.r.t races when clearing 724 * NMODIFIED. 725 * np->n_flag &= ~NMODIFIED; 726 */ 727 } 728 } else 729 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 730 mtx_lock(&np->n_mtx); 731 } 732 /* 733 * Invalidate the attribute cache in all cases. 734 * An open is going to fetch fresh attrs any way, other procs 735 * on this node that have file open will be forced to do an 736 * otw attr fetch, but this is safe. 737 * --> A user found that their RPC count dropped by 20% when 738 * this was commented out and I can't see any requirement 739 * for it, so I've disabled it when negative lookups are 740 * enabled. (What does this have to do with negative lookup 741 * caching? Well nothing, except it was reported by the 742 * same user that needed negative lookup caching and I wanted 743 * there to be a way to disable it to see if it 744 * is the cause of some caching/coherency issue that might 745 * crop up.) 746 */ 747 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) { 748 np->n_attrstamp = 0; 749 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 750 } 751 if (np->n_flag & NWRITEERR) { 752 np->n_flag &= ~NWRITEERR; 753 error = np->n_error; 754 } 755 mtx_unlock(&np->n_mtx); 756 } 757 758 if (NFS_ISV4(vp)) { 759 /* 760 * Get attributes so "change" is up to date. 761 */ 762 if (error == 0 && nfscl_mustflush(vp) != 0 && 763 vp->v_type == VREG && 764 (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) { 765 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva, 766 NULL); 767 if (!ret) { 768 np->n_change = nfsva.na_filerev; 769 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 770 NULL, 0, 0); 771 } 772 } 773 774 /* 775 * and do the close. 776 */ 777 ret = nfsrpc_close(vp, 0, ap->a_td); 778 if (!error && ret) 779 error = ret; 780 if (error) 781 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 782 (gid_t)0); 783 } 784 if (newnfs_directio_enable) 785 KASSERT((np->n_directio_asyncwr == 0), 786 ("nfs_close: dirty unflushed (%d) directio buffers\n", 787 np->n_directio_asyncwr)); 788 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) { 789 mtx_lock(&np->n_mtx); 790 KASSERT((np->n_directio_opens > 0), 791 ("nfs_close: unexpectedly value (0) of n_directio_opens\n")); 792 np->n_directio_opens--; 793 if (np->n_directio_opens == 0) 794 np->n_flag &= ~NNONCACHE; 795 mtx_unlock(&np->n_mtx); 796 } 797 if (localcred) 798 NFSFREECRED(cred); 799 return (error); 800 } 801 802 /* 803 * nfs getattr call from vfs. 804 */ 805 static int 806 nfs_getattr(struct vop_getattr_args *ap) 807 { 808 struct vnode *vp = ap->a_vp; 809 struct thread *td = curthread; /* XXX */ 810 struct nfsnode *np = VTONFS(vp); 811 int error = 0; 812 struct nfsvattr nfsva; 813 struct vattr *vap = ap->a_vap; 814 struct vattr vattr; 815 816 /* 817 * Update local times for special files. 818 */ 819 mtx_lock(&np->n_mtx); 820 if (np->n_flag & (NACC | NUPD)) 821 np->n_flag |= NCHG; 822 mtx_unlock(&np->n_mtx); 823 /* 824 * First look in the cache. 825 */ 826 if (ncl_getattrcache(vp, &vattr) == 0) { 827 vap->va_type = vattr.va_type; 828 vap->va_mode = vattr.va_mode; 829 vap->va_nlink = vattr.va_nlink; 830 vap->va_uid = vattr.va_uid; 831 vap->va_gid = vattr.va_gid; 832 vap->va_fsid = vattr.va_fsid; 833 vap->va_fileid = vattr.va_fileid; 834 vap->va_size = vattr.va_size; 835 vap->va_blocksize = vattr.va_blocksize; 836 vap->va_atime = vattr.va_atime; 837 vap->va_mtime = vattr.va_mtime; 838 vap->va_ctime = vattr.va_ctime; 839 vap->va_gen = vattr.va_gen; 840 vap->va_flags = vattr.va_flags; 841 vap->va_rdev = vattr.va_rdev; 842 vap->va_bytes = vattr.va_bytes; 843 vap->va_filerev = vattr.va_filerev; 844 /* 845 * Get the local modify time for the case of a write 846 * delegation. 847 */ 848 nfscl_deleggetmodtime(vp, &vap->va_mtime); 849 return (0); 850 } 851 852 if (NFS_ISV34(vp) && nfs_prime_access_cache && 853 nfsaccess_cache_timeout > 0) { 854 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses); 855 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL); 856 if (ncl_getattrcache(vp, ap->a_vap) == 0) { 857 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime); 858 return (0); 859 } 860 } 861 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL); 862 if (!error) 863 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0); 864 if (!error) { 865 /* 866 * Get the local modify time for the case of a write 867 * delegation. 868 */ 869 nfscl_deleggetmodtime(vp, &vap->va_mtime); 870 } else if (NFS_ISV4(vp)) { 871 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 872 } 873 return (error); 874 } 875 876 /* 877 * nfs setattr call. 878 */ 879 static int 880 nfs_setattr(struct vop_setattr_args *ap) 881 { 882 struct vnode *vp = ap->a_vp; 883 struct nfsnode *np = VTONFS(vp); 884 struct thread *td = curthread; /* XXX */ 885 struct vattr *vap = ap->a_vap; 886 int error = 0; 887 u_quad_t tsize; 888 889 #ifndef nolint 890 tsize = (u_quad_t)0; 891 #endif 892 893 /* 894 * Setting of flags and marking of atimes are not supported. 895 */ 896 if (vap->va_flags != VNOVAL) 897 return (EOPNOTSUPP); 898 899 /* 900 * Disallow write attempts if the filesystem is mounted read-only. 901 */ 902 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 903 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 904 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 905 (vp->v_mount->mnt_flag & MNT_RDONLY)) 906 return (EROFS); 907 if (vap->va_size != VNOVAL) { 908 switch (vp->v_type) { 909 case VDIR: 910 return (EISDIR); 911 case VCHR: 912 case VBLK: 913 case VSOCK: 914 case VFIFO: 915 if (vap->va_mtime.tv_sec == VNOVAL && 916 vap->va_atime.tv_sec == VNOVAL && 917 vap->va_mode == (mode_t)VNOVAL && 918 vap->va_uid == (uid_t)VNOVAL && 919 vap->va_gid == (gid_t)VNOVAL) 920 return (0); 921 vap->va_size = VNOVAL; 922 break; 923 default: 924 /* 925 * Disallow write attempts if the filesystem is 926 * mounted read-only. 927 */ 928 if (vp->v_mount->mnt_flag & MNT_RDONLY) 929 return (EROFS); 930 /* 931 * We run vnode_pager_setsize() early (why?), 932 * we must set np->n_size now to avoid vinvalbuf 933 * V_SAVE races that might setsize a lower 934 * value. 935 */ 936 mtx_lock(&np->n_mtx); 937 tsize = np->n_size; 938 mtx_unlock(&np->n_mtx); 939 error = ncl_meta_setsize(vp, ap->a_cred, td, 940 vap->va_size); 941 mtx_lock(&np->n_mtx); 942 if (np->n_flag & NMODIFIED) { 943 tsize = np->n_size; 944 mtx_unlock(&np->n_mtx); 945 if (vap->va_size == 0) 946 error = ncl_vinvalbuf(vp, 0, td, 1); 947 else 948 error = ncl_vinvalbuf(vp, V_SAVE, td, 1); 949 if (error) { 950 vnode_pager_setsize(vp, tsize); 951 return (error); 952 } 953 /* 954 * Call nfscl_delegmodtime() to set the modify time 955 * locally, as required. 956 */ 957 nfscl_delegmodtime(vp); 958 } else 959 mtx_unlock(&np->n_mtx); 960 /* 961 * np->n_size has already been set to vap->va_size 962 * in ncl_meta_setsize(). We must set it again since 963 * nfs_loadattrcache() could be called through 964 * ncl_meta_setsize() and could modify np->n_size. 965 */ 966 mtx_lock(&np->n_mtx); 967 np->n_vattr.na_size = np->n_size = vap->va_size; 968 mtx_unlock(&np->n_mtx); 969 } 970 } else { 971 mtx_lock(&np->n_mtx); 972 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) && 973 (np->n_flag & NMODIFIED) && vp->v_type == VREG) { 974 mtx_unlock(&np->n_mtx); 975 if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 && 976 (error == EINTR || error == EIO)) 977 return (error); 978 } else 979 mtx_unlock(&np->n_mtx); 980 } 981 error = nfs_setattrrpc(vp, vap, ap->a_cred, td); 982 if (error && vap->va_size != VNOVAL) { 983 mtx_lock(&np->n_mtx); 984 np->n_size = np->n_vattr.na_size = tsize; 985 vnode_pager_setsize(vp, tsize); 986 mtx_unlock(&np->n_mtx); 987 } 988 return (error); 989 } 990 991 /* 992 * Do an nfs setattr rpc. 993 */ 994 static int 995 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred, 996 struct thread *td) 997 { 998 struct nfsnode *np = VTONFS(vp); 999 int error, ret, attrflag, i; 1000 struct nfsvattr nfsva; 1001 1002 if (NFS_ISV34(vp)) { 1003 mtx_lock(&np->n_mtx); 1004 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) 1005 np->n_accesscache[i].stamp = 0; 1006 np->n_flag |= NDELEGMOD; 1007 mtx_unlock(&np->n_mtx); 1008 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp); 1009 } 1010 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag, 1011 NULL); 1012 if (attrflag) { 1013 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1014 if (ret && !error) 1015 error = ret; 1016 } 1017 if (error && NFS_ISV4(vp)) 1018 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid); 1019 return (error); 1020 } 1021 1022 /* 1023 * nfs lookup call, one step at a time... 1024 * First look in cache 1025 * If not found, unlock the directory nfsnode and do the rpc 1026 */ 1027 static int 1028 nfs_lookup(struct vop_lookup_args *ap) 1029 { 1030 struct componentname *cnp = ap->a_cnp; 1031 struct vnode *dvp = ap->a_dvp; 1032 struct vnode **vpp = ap->a_vpp; 1033 struct mount *mp = dvp->v_mount; 1034 int flags = cnp->cn_flags; 1035 struct vnode *newvp; 1036 struct nfsmount *nmp; 1037 struct nfsnode *np, *newnp; 1038 int error = 0, attrflag, dattrflag, ltype, ncticks; 1039 struct thread *td = cnp->cn_thread; 1040 struct nfsfh *nfhp; 1041 struct nfsvattr dnfsva, nfsva; 1042 struct vattr vattr; 1043 struct timespec nctime; 1044 1045 *vpp = NULLVP; 1046 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) && 1047 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 1048 return (EROFS); 1049 if (dvp->v_type != VDIR) 1050 return (ENOTDIR); 1051 nmp = VFSTONFS(mp); 1052 np = VTONFS(dvp); 1053 1054 /* For NFSv4, wait until any remove is done. */ 1055 mtx_lock(&np->n_mtx); 1056 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) { 1057 np->n_flag |= NREMOVEWANT; 1058 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0); 1059 } 1060 mtx_unlock(&np->n_mtx); 1061 1062 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) 1063 return (error); 1064 error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks); 1065 if (error > 0 && error != ENOENT) 1066 return (error); 1067 if (error == -1) { 1068 /* 1069 * Lookups of "." are special and always return the 1070 * current directory. cache_lookup() already handles 1071 * associated locking bookkeeping, etc. 1072 */ 1073 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 1074 /* XXX: Is this really correct? */ 1075 if (cnp->cn_nameiop != LOOKUP && 1076 (flags & ISLASTCN)) 1077 cnp->cn_flags |= SAVENAME; 1078 return (0); 1079 } 1080 1081 /* 1082 * We only accept a positive hit in the cache if the 1083 * change time of the file matches our cached copy. 1084 * Otherwise, we discard the cache entry and fallback 1085 * to doing a lookup RPC. We also only trust cache 1086 * entries for less than nm_nametimeo seconds. 1087 * 1088 * To better handle stale file handles and attributes, 1089 * clear the attribute cache of this node if it is a 1090 * leaf component, part of an open() call, and not 1091 * locally modified before fetching the attributes. 1092 * This should allow stale file handles to be detected 1093 * here where we can fall back to a LOOKUP RPC to 1094 * recover rather than having nfs_open() detect the 1095 * stale file handle and failing open(2) with ESTALE. 1096 */ 1097 newvp = *vpp; 1098 newnp = VTONFS(newvp); 1099 if (!(nmp->nm_flag & NFSMNT_NOCTO) && 1100 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1101 !(newnp->n_flag & NMODIFIED)) { 1102 mtx_lock(&newnp->n_mtx); 1103 newnp->n_attrstamp = 0; 1104 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1105 mtx_unlock(&newnp->n_mtx); 1106 } 1107 if (nfscl_nodeleg(newvp, 0) == 0 || 1108 ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) && 1109 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 && 1110 timespeccmp(&vattr.va_ctime, &nctime, ==))) { 1111 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits); 1112 if (cnp->cn_nameiop != LOOKUP && 1113 (flags & ISLASTCN)) 1114 cnp->cn_flags |= SAVENAME; 1115 return (0); 1116 } 1117 cache_purge(newvp); 1118 if (dvp != newvp) 1119 vput(newvp); 1120 else 1121 vrele(newvp); 1122 *vpp = NULLVP; 1123 } else if (error == ENOENT) { 1124 if (dvp->v_iflag & VI_DOOMED) 1125 return (ENOENT); 1126 /* 1127 * We only accept a negative hit in the cache if the 1128 * modification time of the parent directory matches 1129 * the cached copy in the name cache entry. 1130 * Otherwise, we discard all of the negative cache 1131 * entries for this directory. We also only trust 1132 * negative cache entries for up to nm_negnametimeo 1133 * seconds. 1134 */ 1135 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) && 1136 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 && 1137 timespeccmp(&vattr.va_mtime, &nctime, ==)) { 1138 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits); 1139 return (ENOENT); 1140 } 1141 cache_purge_negative(dvp); 1142 } 1143 1144 error = 0; 1145 newvp = NULLVP; 1146 NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses); 1147 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1148 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1149 NULL); 1150 if (dattrflag) 1151 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1152 if (error) { 1153 if (newvp != NULLVP) { 1154 vput(newvp); 1155 *vpp = NULLVP; 1156 } 1157 1158 if (error != ENOENT) { 1159 if (NFS_ISV4(dvp)) 1160 error = nfscl_maperr(td, error, (uid_t)0, 1161 (gid_t)0); 1162 return (error); 1163 } 1164 1165 /* The requested file was not found. */ 1166 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1167 (flags & ISLASTCN)) { 1168 /* 1169 * XXX: UFS does a full VOP_ACCESS(dvp, 1170 * VWRITE) here instead of just checking 1171 * MNT_RDONLY. 1172 */ 1173 if (mp->mnt_flag & MNT_RDONLY) 1174 return (EROFS); 1175 cnp->cn_flags |= SAVENAME; 1176 return (EJUSTRETURN); 1177 } 1178 1179 if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) { 1180 /* 1181 * Cache the modification time of the parent 1182 * directory from the post-op attributes in 1183 * the name cache entry. The negative cache 1184 * entry will be ignored once the directory 1185 * has changed. Don't bother adding the entry 1186 * if the directory has already changed. 1187 */ 1188 mtx_lock(&np->n_mtx); 1189 if (timespeccmp(&np->n_vattr.na_mtime, 1190 &dnfsva.na_mtime, ==)) { 1191 mtx_unlock(&np->n_mtx); 1192 cache_enter_time(dvp, NULL, cnp, 1193 &dnfsva.na_mtime, NULL); 1194 } else 1195 mtx_unlock(&np->n_mtx); 1196 } 1197 return (ENOENT); 1198 } 1199 1200 /* 1201 * Handle RENAME case... 1202 */ 1203 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 1204 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1205 FREE((caddr_t)nfhp, M_NFSFH); 1206 return (EISDIR); 1207 } 1208 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1209 LK_EXCLUSIVE); 1210 if (error) 1211 return (error); 1212 newvp = NFSTOV(np); 1213 if (attrflag) 1214 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1215 0, 1); 1216 *vpp = newvp; 1217 cnp->cn_flags |= SAVENAME; 1218 return (0); 1219 } 1220 1221 if (flags & ISDOTDOT) { 1222 ltype = NFSVOPISLOCKED(dvp); 1223 error = vfs_busy(mp, MBF_NOWAIT); 1224 if (error != 0) { 1225 vfs_ref(mp); 1226 NFSVOPUNLOCK(dvp, 0); 1227 error = vfs_busy(mp, 0); 1228 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1229 vfs_rel(mp); 1230 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) { 1231 vfs_unbusy(mp); 1232 error = ENOENT; 1233 } 1234 if (error != 0) 1235 return (error); 1236 } 1237 NFSVOPUNLOCK(dvp, 0); 1238 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1239 cnp->cn_lkflags); 1240 if (error == 0) 1241 newvp = NFSTOV(np); 1242 vfs_unbusy(mp); 1243 if (newvp != dvp) 1244 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1245 if (dvp->v_iflag & VI_DOOMED) { 1246 if (error == 0) { 1247 if (newvp == dvp) 1248 vrele(newvp); 1249 else 1250 vput(newvp); 1251 } 1252 error = ENOENT; 1253 } 1254 if (error != 0) 1255 return (error); 1256 if (attrflag) 1257 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1258 0, 1); 1259 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1260 FREE((caddr_t)nfhp, M_NFSFH); 1261 VREF(dvp); 1262 newvp = dvp; 1263 if (attrflag) 1264 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1265 0, 1); 1266 } else { 1267 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1268 cnp->cn_lkflags); 1269 if (error) 1270 return (error); 1271 newvp = NFSTOV(np); 1272 if (attrflag) 1273 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1274 0, 1); 1275 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1276 !(np->n_flag & NMODIFIED)) { 1277 /* 1278 * Flush the attribute cache when opening a 1279 * leaf node to ensure that fresh attributes 1280 * are fetched in nfs_open() since we did not 1281 * fetch attributes from the LOOKUP reply. 1282 */ 1283 mtx_lock(&np->n_mtx); 1284 np->n_attrstamp = 0; 1285 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1286 mtx_unlock(&np->n_mtx); 1287 } 1288 } 1289 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1290 cnp->cn_flags |= SAVENAME; 1291 if ((cnp->cn_flags & MAKEENTRY) && 1292 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) && 1293 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0)) 1294 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 1295 newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime); 1296 *vpp = newvp; 1297 return (0); 1298 } 1299 1300 /* 1301 * nfs read call. 1302 * Just call ncl_bioread() to do the work. 1303 */ 1304 static int 1305 nfs_read(struct vop_read_args *ap) 1306 { 1307 struct vnode *vp = ap->a_vp; 1308 1309 switch (vp->v_type) { 1310 case VREG: 1311 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred)); 1312 case VDIR: 1313 return (EISDIR); 1314 default: 1315 return (EOPNOTSUPP); 1316 } 1317 } 1318 1319 /* 1320 * nfs readlink call 1321 */ 1322 static int 1323 nfs_readlink(struct vop_readlink_args *ap) 1324 { 1325 struct vnode *vp = ap->a_vp; 1326 1327 if (vp->v_type != VLNK) 1328 return (EINVAL); 1329 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred)); 1330 } 1331 1332 /* 1333 * Do a readlink rpc. 1334 * Called by ncl_doio() from below the buffer cache. 1335 */ 1336 int 1337 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1338 { 1339 int error, ret, attrflag; 1340 struct nfsvattr nfsva; 1341 1342 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva, 1343 &attrflag, NULL); 1344 if (attrflag) { 1345 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1346 if (ret && !error) 1347 error = ret; 1348 } 1349 if (error && NFS_ISV4(vp)) 1350 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1351 return (error); 1352 } 1353 1354 /* 1355 * nfs read rpc call 1356 * Ditto above 1357 */ 1358 int 1359 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1360 { 1361 int error, ret, attrflag; 1362 struct nfsvattr nfsva; 1363 struct nfsmount *nmp; 1364 1365 nmp = VFSTONFS(vnode_mount(vp)); 1366 error = EIO; 1367 attrflag = 0; 1368 if (NFSHASPNFS(nmp)) 1369 error = nfscl_doiods(vp, uiop, NULL, NULL, 1370 NFSV4OPEN_ACCESSREAD, cred, uiop->uio_td); 1371 NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error); 1372 if (error != 0) 1373 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, 1374 &attrflag, NULL); 1375 if (attrflag) { 1376 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1377 if (ret && !error) 1378 error = ret; 1379 } 1380 if (error && NFS_ISV4(vp)) 1381 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1382 return (error); 1383 } 1384 1385 /* 1386 * nfs write call 1387 */ 1388 int 1389 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 1390 int *iomode, int *must_commit, int called_from_strategy) 1391 { 1392 struct nfsvattr nfsva; 1393 int error, attrflag, ret; 1394 struct nfsmount *nmp; 1395 1396 nmp = VFSTONFS(vnode_mount(vp)); 1397 error = EIO; 1398 attrflag = 0; 1399 if (NFSHASPNFS(nmp)) 1400 error = nfscl_doiods(vp, uiop, iomode, must_commit, 1401 NFSV4OPEN_ACCESSWRITE, cred, uiop->uio_td); 1402 NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error); 1403 if (error != 0) 1404 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred, 1405 uiop->uio_td, &nfsva, &attrflag, NULL, 1406 called_from_strategy); 1407 if (attrflag) { 1408 if (VTONFS(vp)->n_flag & ND_NFSV4) 1409 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1, 1410 1); 1411 else 1412 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1413 1); 1414 if (ret && !error) 1415 error = ret; 1416 } 1417 if (DOINGASYNC(vp)) 1418 *iomode = NFSWRITE_FILESYNC; 1419 if (error && NFS_ISV4(vp)) 1420 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1421 return (error); 1422 } 1423 1424 /* 1425 * nfs mknod rpc 1426 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1427 * mode set to specify the file type and the size field for rdev. 1428 */ 1429 static int 1430 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1431 struct vattr *vap) 1432 { 1433 struct nfsvattr nfsva, dnfsva; 1434 struct vnode *newvp = NULL; 1435 struct nfsnode *np = NULL, *dnp; 1436 struct nfsfh *nfhp; 1437 struct vattr vattr; 1438 int error = 0, attrflag, dattrflag; 1439 u_int32_t rdev; 1440 1441 if (vap->va_type == VCHR || vap->va_type == VBLK) 1442 rdev = vap->va_rdev; 1443 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1444 rdev = 0xffffffff; 1445 else 1446 return (EOPNOTSUPP); 1447 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1448 return (error); 1449 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap, 1450 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva, 1451 &nfsva, &nfhp, &attrflag, &dattrflag, NULL); 1452 if (!error) { 1453 if (!nfhp) 1454 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1455 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, 1456 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1457 NULL); 1458 if (nfhp) 1459 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1460 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); 1461 } 1462 if (dattrflag) 1463 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1464 if (!error) { 1465 newvp = NFSTOV(np); 1466 if (attrflag != 0) { 1467 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1468 0, 1); 1469 if (error != 0) 1470 vput(newvp); 1471 } 1472 } 1473 if (!error) { 1474 *vpp = newvp; 1475 } else if (NFS_ISV4(dvp)) { 1476 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, 1477 vap->va_gid); 1478 } 1479 dnp = VTONFS(dvp); 1480 mtx_lock(&dnp->n_mtx); 1481 dnp->n_flag |= NMODIFIED; 1482 if (!dattrflag) { 1483 dnp->n_attrstamp = 0; 1484 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1485 } 1486 mtx_unlock(&dnp->n_mtx); 1487 return (error); 1488 } 1489 1490 /* 1491 * nfs mknod vop 1492 * just call nfs_mknodrpc() to do the work. 1493 */ 1494 /* ARGSUSED */ 1495 static int 1496 nfs_mknod(struct vop_mknod_args *ap) 1497 { 1498 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap)); 1499 } 1500 1501 static struct mtx nfs_cverf_mtx; 1502 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex", 1503 MTX_DEF); 1504 1505 static nfsquad_t 1506 nfs_get_cverf(void) 1507 { 1508 static nfsquad_t cverf; 1509 nfsquad_t ret; 1510 static int cverf_initialized = 0; 1511 1512 mtx_lock(&nfs_cverf_mtx); 1513 if (cverf_initialized == 0) { 1514 cverf.lval[0] = arc4random(); 1515 cverf.lval[1] = arc4random(); 1516 cverf_initialized = 1; 1517 } else 1518 cverf.qval++; 1519 ret = cverf; 1520 mtx_unlock(&nfs_cverf_mtx); 1521 1522 return (ret); 1523 } 1524 1525 /* 1526 * nfs file create call 1527 */ 1528 static int 1529 nfs_create(struct vop_create_args *ap) 1530 { 1531 struct vnode *dvp = ap->a_dvp; 1532 struct vattr *vap = ap->a_vap; 1533 struct componentname *cnp = ap->a_cnp; 1534 struct nfsnode *np = NULL, *dnp; 1535 struct vnode *newvp = NULL; 1536 struct nfsmount *nmp; 1537 struct nfsvattr dnfsva, nfsva; 1538 struct nfsfh *nfhp; 1539 nfsquad_t cverf; 1540 int error = 0, attrflag, dattrflag, fmode = 0; 1541 struct vattr vattr; 1542 1543 /* 1544 * Oops, not for me.. 1545 */ 1546 if (vap->va_type == VSOCK) 1547 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1548 1549 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1550 return (error); 1551 if (vap->va_vaflags & VA_EXCLUSIVE) 1552 fmode |= O_EXCL; 1553 dnp = VTONFS(dvp); 1554 nmp = VFSTONFS(vnode_mount(dvp)); 1555 again: 1556 /* For NFSv4, wait until any remove is done. */ 1557 mtx_lock(&dnp->n_mtx); 1558 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) { 1559 dnp->n_flag |= NREMOVEWANT; 1560 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0); 1561 } 1562 mtx_unlock(&dnp->n_mtx); 1563 1564 cverf = nfs_get_cverf(); 1565 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1566 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, 1567 &nfhp, &attrflag, &dattrflag, NULL); 1568 if (!error) { 1569 if (nfhp == NULL) 1570 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1571 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, 1572 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1573 NULL); 1574 if (nfhp != NULL) 1575 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1576 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); 1577 } 1578 if (dattrflag) 1579 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1580 if (!error) { 1581 newvp = NFSTOV(np); 1582 if (attrflag == 0) 1583 error = nfsrpc_getattr(newvp, cnp->cn_cred, 1584 cnp->cn_thread, &nfsva, NULL); 1585 if (error == 0) 1586 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1587 0, 1); 1588 } 1589 if (error) { 1590 if (newvp != NULL) { 1591 vput(newvp); 1592 newvp = NULL; 1593 } 1594 if (NFS_ISV34(dvp) && (fmode & O_EXCL) && 1595 error == NFSERR_NOTSUPP) { 1596 fmode &= ~O_EXCL; 1597 goto again; 1598 } 1599 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) { 1600 if (nfscl_checksattr(vap, &nfsva)) { 1601 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred, 1602 cnp->cn_thread, &nfsva, &attrflag, NULL); 1603 if (error && (vap->va_uid != (uid_t)VNOVAL || 1604 vap->va_gid != (gid_t)VNOVAL)) { 1605 /* try again without setting uid/gid */ 1606 vap->va_uid = (uid_t)VNOVAL; 1607 vap->va_gid = (uid_t)VNOVAL; 1608 error = nfsrpc_setattr(newvp, vap, NULL, 1609 cnp->cn_cred, cnp->cn_thread, &nfsva, 1610 &attrflag, NULL); 1611 } 1612 if (attrflag) 1613 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 1614 NULL, 0, 1); 1615 if (error != 0) 1616 vput(newvp); 1617 } 1618 } 1619 if (!error) { 1620 if ((cnp->cn_flags & MAKEENTRY) && attrflag) 1621 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 1622 NULL); 1623 *ap->a_vpp = newvp; 1624 } else if (NFS_ISV4(dvp)) { 1625 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, 1626 vap->va_gid); 1627 } 1628 mtx_lock(&dnp->n_mtx); 1629 dnp->n_flag |= NMODIFIED; 1630 if (!dattrflag) { 1631 dnp->n_attrstamp = 0; 1632 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1633 } 1634 mtx_unlock(&dnp->n_mtx); 1635 return (error); 1636 } 1637 1638 /* 1639 * nfs file remove call 1640 * To try and make nfs semantics closer to ufs semantics, a file that has 1641 * other processes using the vnode is renamed instead of removed and then 1642 * removed later on the last close. 1643 * - If v_usecount > 1 1644 * If a rename is not already in the works 1645 * call nfs_sillyrename() to set it up 1646 * else 1647 * do the remove rpc 1648 */ 1649 static int 1650 nfs_remove(struct vop_remove_args *ap) 1651 { 1652 struct vnode *vp = ap->a_vp; 1653 struct vnode *dvp = ap->a_dvp; 1654 struct componentname *cnp = ap->a_cnp; 1655 struct nfsnode *np = VTONFS(vp); 1656 int error = 0; 1657 struct vattr vattr; 1658 1659 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name")); 1660 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount")); 1661 if (vp->v_type == VDIR) 1662 error = EPERM; 1663 else if (vrefcnt(vp) == 1 || (np->n_sillyrename && 1664 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 1665 vattr.va_nlink > 1)) { 1666 /* 1667 * Purge the name cache so that the chance of a lookup for 1668 * the name succeeding while the remove is in progress is 1669 * minimized. Without node locking it can still happen, such 1670 * that an I/O op returns ESTALE, but since you get this if 1671 * another host removes the file.. 1672 */ 1673 cache_purge(vp); 1674 /* 1675 * throw away biocache buffers, mainly to avoid 1676 * unnecessary delayed writes later. 1677 */ 1678 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1); 1679 /* Do the rpc */ 1680 if (error != EINTR && error != EIO) 1681 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr, 1682 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread); 1683 /* 1684 * Kludge City: If the first reply to the remove rpc is lost.. 1685 * the reply to the retransmitted request will be ENOENT 1686 * since the file was in fact removed 1687 * Therefore, we cheat and return success. 1688 */ 1689 if (error == ENOENT) 1690 error = 0; 1691 } else if (!np->n_sillyrename) 1692 error = nfs_sillyrename(dvp, vp, cnp); 1693 mtx_lock(&np->n_mtx); 1694 np->n_attrstamp = 0; 1695 mtx_unlock(&np->n_mtx); 1696 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 1697 return (error); 1698 } 1699 1700 /* 1701 * nfs file remove rpc called from nfs_inactive 1702 */ 1703 int 1704 ncl_removeit(struct sillyrename *sp, struct vnode *vp) 1705 { 1706 /* 1707 * Make sure that the directory vnode is still valid. 1708 * XXX we should lock sp->s_dvp here. 1709 */ 1710 if (sp->s_dvp->v_type == VBAD) 1711 return (0); 1712 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen, 1713 sp->s_cred, NULL)); 1714 } 1715 1716 /* 1717 * Nfs remove rpc, called from nfs_remove() and ncl_removeit(). 1718 */ 1719 static int 1720 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 1721 int namelen, struct ucred *cred, struct thread *td) 1722 { 1723 struct nfsvattr dnfsva; 1724 struct nfsnode *dnp = VTONFS(dvp); 1725 int error = 0, dattrflag; 1726 1727 mtx_lock(&dnp->n_mtx); 1728 dnp->n_flag |= NREMOVEINPROG; 1729 mtx_unlock(&dnp->n_mtx); 1730 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva, 1731 &dattrflag, NULL); 1732 mtx_lock(&dnp->n_mtx); 1733 if ((dnp->n_flag & NREMOVEWANT)) { 1734 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG); 1735 mtx_unlock(&dnp->n_mtx); 1736 wakeup((caddr_t)dnp); 1737 } else { 1738 dnp->n_flag &= ~NREMOVEINPROG; 1739 mtx_unlock(&dnp->n_mtx); 1740 } 1741 if (dattrflag) 1742 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1743 mtx_lock(&dnp->n_mtx); 1744 dnp->n_flag |= NMODIFIED; 1745 if (!dattrflag) { 1746 dnp->n_attrstamp = 0; 1747 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1748 } 1749 mtx_unlock(&dnp->n_mtx); 1750 if (error && NFS_ISV4(dvp)) 1751 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1752 return (error); 1753 } 1754 1755 /* 1756 * nfs file rename call 1757 */ 1758 static int 1759 nfs_rename(struct vop_rename_args *ap) 1760 { 1761 struct vnode *fvp = ap->a_fvp; 1762 struct vnode *tvp = ap->a_tvp; 1763 struct vnode *fdvp = ap->a_fdvp; 1764 struct vnode *tdvp = ap->a_tdvp; 1765 struct componentname *tcnp = ap->a_tcnp; 1766 struct componentname *fcnp = ap->a_fcnp; 1767 struct nfsnode *fnp = VTONFS(ap->a_fvp); 1768 struct nfsnode *tdnp = VTONFS(ap->a_tdvp); 1769 struct nfsv4node *newv4 = NULL; 1770 int error; 1771 1772 KASSERT((tcnp->cn_flags & HASBUF) != 0 && 1773 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name")); 1774 /* Check for cross-device rename */ 1775 if ((fvp->v_mount != tdvp->v_mount) || 1776 (tvp && (fvp->v_mount != tvp->v_mount))) { 1777 error = EXDEV; 1778 goto out; 1779 } 1780 1781 if (fvp == tvp) { 1782 printf("nfs_rename: fvp == tvp (can't happen)\n"); 1783 error = 0; 1784 goto out; 1785 } 1786 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0) 1787 goto out; 1788 1789 /* 1790 * We have to flush B_DELWRI data prior to renaming 1791 * the file. If we don't, the delayed-write buffers 1792 * can be flushed out later after the file has gone stale 1793 * under NFSV3. NFSV2 does not have this problem because 1794 * ( as far as I can tell ) it flushes dirty buffers more 1795 * often. 1796 * 1797 * Skip the rename operation if the fsync fails, this can happen 1798 * due to the server's volume being full, when we pushed out data 1799 * that was written back to our cache earlier. Not checking for 1800 * this condition can result in potential (silent) data loss. 1801 */ 1802 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread); 1803 NFSVOPUNLOCK(fvp, 0); 1804 if (!error && tvp) 1805 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread); 1806 if (error) 1807 goto out; 1808 1809 /* 1810 * If the tvp exists and is in use, sillyrename it before doing the 1811 * rename of the new file over it. 1812 * XXX Can't sillyrename a directory. 1813 */ 1814 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename && 1815 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1816 vput(tvp); 1817 tvp = NULL; 1818 } 1819 1820 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1821 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1822 tcnp->cn_thread); 1823 1824 if (error == 0 && NFS_ISV4(tdvp)) { 1825 /* 1826 * For NFSv4, check to see if it is the same name and 1827 * replace the name, if it is different. 1828 */ 1829 MALLOC(newv4, struct nfsv4node *, 1830 sizeof (struct nfsv4node) + 1831 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1, 1832 M_NFSV4NODE, M_WAITOK); 1833 mtx_lock(&tdnp->n_mtx); 1834 mtx_lock(&fnp->n_mtx); 1835 if (fnp->n_v4 != NULL && fvp->v_type == VREG && 1836 (fnp->n_v4->n4_namelen != tcnp->cn_namelen || 1837 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4), 1838 tcnp->cn_namelen) || 1839 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen || 1840 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 1841 tdnp->n_fhp->nfh_len))) { 1842 #ifdef notdef 1843 { char nnn[100]; int nnnl; 1844 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99; 1845 bcopy(tcnp->cn_nameptr, nnn, nnnl); 1846 nnn[nnnl] = '\0'; 1847 printf("ren replace=%s\n",nnn); 1848 } 1849 #endif 1850 FREE((caddr_t)fnp->n_v4, M_NFSV4NODE); 1851 fnp->n_v4 = newv4; 1852 newv4 = NULL; 1853 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len; 1854 fnp->n_v4->n4_namelen = tcnp->cn_namelen; 1855 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 1856 tdnp->n_fhp->nfh_len); 1857 NFSBCOPY(tcnp->cn_nameptr, 1858 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen); 1859 } 1860 mtx_unlock(&tdnp->n_mtx); 1861 mtx_unlock(&fnp->n_mtx); 1862 if (newv4 != NULL) 1863 FREE((caddr_t)newv4, M_NFSV4NODE); 1864 } 1865 1866 if (fvp->v_type == VDIR) { 1867 if (tvp != NULL && tvp->v_type == VDIR) 1868 cache_purge(tdvp); 1869 cache_purge(fdvp); 1870 } 1871 1872 out: 1873 if (tdvp == tvp) 1874 vrele(tdvp); 1875 else 1876 vput(tdvp); 1877 if (tvp) 1878 vput(tvp); 1879 vrele(fdvp); 1880 vrele(fvp); 1881 /* 1882 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1883 */ 1884 if (error == ENOENT) 1885 error = 0; 1886 return (error); 1887 } 1888 1889 /* 1890 * nfs file rename rpc called from nfs_remove() above 1891 */ 1892 static int 1893 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp, 1894 struct sillyrename *sp) 1895 { 1896 1897 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen, 1898 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred, 1899 scnp->cn_thread)); 1900 } 1901 1902 /* 1903 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1904 */ 1905 static int 1906 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr, 1907 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr, 1908 int tnamelen, struct ucred *cred, struct thread *td) 1909 { 1910 struct nfsvattr fnfsva, tnfsva; 1911 struct nfsnode *fdnp = VTONFS(fdvp); 1912 struct nfsnode *tdnp = VTONFS(tdvp); 1913 int error = 0, fattrflag, tattrflag; 1914 1915 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp, 1916 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag, 1917 &tattrflag, NULL, NULL); 1918 mtx_lock(&fdnp->n_mtx); 1919 fdnp->n_flag |= NMODIFIED; 1920 if (fattrflag != 0) { 1921 mtx_unlock(&fdnp->n_mtx); 1922 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1); 1923 } else { 1924 fdnp->n_attrstamp = 0; 1925 mtx_unlock(&fdnp->n_mtx); 1926 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp); 1927 } 1928 mtx_lock(&tdnp->n_mtx); 1929 tdnp->n_flag |= NMODIFIED; 1930 if (tattrflag != 0) { 1931 mtx_unlock(&tdnp->n_mtx); 1932 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1); 1933 } else { 1934 tdnp->n_attrstamp = 0; 1935 mtx_unlock(&tdnp->n_mtx); 1936 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 1937 } 1938 if (error && NFS_ISV4(fdvp)) 1939 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1940 return (error); 1941 } 1942 1943 /* 1944 * nfs hard link create call 1945 */ 1946 static int 1947 nfs_link(struct vop_link_args *ap) 1948 { 1949 struct vnode *vp = ap->a_vp; 1950 struct vnode *tdvp = ap->a_tdvp; 1951 struct componentname *cnp = ap->a_cnp; 1952 struct nfsnode *np, *tdnp; 1953 struct nfsvattr nfsva, dnfsva; 1954 int error = 0, attrflag, dattrflag; 1955 1956 /* 1957 * Push all writes to the server, so that the attribute cache 1958 * doesn't get "out of sync" with the server. 1959 * XXX There should be a better way! 1960 */ 1961 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread); 1962 1963 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 1964 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag, 1965 &dattrflag, NULL); 1966 tdnp = VTONFS(tdvp); 1967 mtx_lock(&tdnp->n_mtx); 1968 tdnp->n_flag |= NMODIFIED; 1969 if (dattrflag != 0) { 1970 mtx_unlock(&tdnp->n_mtx); 1971 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1); 1972 } else { 1973 tdnp->n_attrstamp = 0; 1974 mtx_unlock(&tdnp->n_mtx); 1975 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 1976 } 1977 if (attrflag) 1978 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1979 else { 1980 np = VTONFS(vp); 1981 mtx_lock(&np->n_mtx); 1982 np->n_attrstamp = 0; 1983 mtx_unlock(&np->n_mtx); 1984 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 1985 } 1986 /* 1987 * If negative lookup caching is enabled, I might as well 1988 * add an entry for this node. Not necessary for correctness, 1989 * but if negative caching is enabled, then the system 1990 * must care about lookup caching hit rate, so... 1991 */ 1992 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 && 1993 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 1994 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL); 1995 } 1996 if (error && NFS_ISV4(vp)) 1997 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, 1998 (gid_t)0); 1999 return (error); 2000 } 2001 2002 /* 2003 * nfs symbolic link create call 2004 */ 2005 static int 2006 nfs_symlink(struct vop_symlink_args *ap) 2007 { 2008 struct vnode *dvp = ap->a_dvp; 2009 struct vattr *vap = ap->a_vap; 2010 struct componentname *cnp = ap->a_cnp; 2011 struct nfsvattr nfsva, dnfsva; 2012 struct nfsfh *nfhp; 2013 struct nfsnode *np = NULL, *dnp; 2014 struct vnode *newvp = NULL; 2015 int error = 0, attrflag, dattrflag, ret; 2016 2017 vap->va_type = VLNK; 2018 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2019 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, 2020 &nfsva, &nfhp, &attrflag, &dattrflag, NULL); 2021 if (nfhp) { 2022 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, 2023 &np, NULL, LK_EXCLUSIVE); 2024 if (!ret) 2025 newvp = NFSTOV(np); 2026 else if (!error) 2027 error = ret; 2028 } 2029 if (newvp != NULL) { 2030 if (attrflag) 2031 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 2032 0, 1); 2033 } else if (!error) { 2034 /* 2035 * If we do not have an error and we could not extract the 2036 * newvp from the response due to the request being NFSv2, we 2037 * have to do a lookup in order to obtain a newvp to return. 2038 */ 2039 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2040 cnp->cn_cred, cnp->cn_thread, &np); 2041 if (!error) 2042 newvp = NFSTOV(np); 2043 } 2044 if (error) { 2045 if (newvp) 2046 vput(newvp); 2047 if (NFS_ISV4(dvp)) 2048 error = nfscl_maperr(cnp->cn_thread, error, 2049 vap->va_uid, vap->va_gid); 2050 } else { 2051 *ap->a_vpp = newvp; 2052 } 2053 2054 dnp = VTONFS(dvp); 2055 mtx_lock(&dnp->n_mtx); 2056 dnp->n_flag |= NMODIFIED; 2057 if (dattrflag != 0) { 2058 mtx_unlock(&dnp->n_mtx); 2059 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2060 } else { 2061 dnp->n_attrstamp = 0; 2062 mtx_unlock(&dnp->n_mtx); 2063 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2064 } 2065 /* 2066 * If negative lookup caching is enabled, I might as well 2067 * add an entry for this node. Not necessary for correctness, 2068 * but if negative caching is enabled, then the system 2069 * must care about lookup caching hit rate, so... 2070 */ 2071 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2072 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 2073 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL); 2074 } 2075 return (error); 2076 } 2077 2078 /* 2079 * nfs make dir call 2080 */ 2081 static int 2082 nfs_mkdir(struct vop_mkdir_args *ap) 2083 { 2084 struct vnode *dvp = ap->a_dvp; 2085 struct vattr *vap = ap->a_vap; 2086 struct componentname *cnp = ap->a_cnp; 2087 struct nfsnode *np = NULL, *dnp; 2088 struct vnode *newvp = NULL; 2089 struct vattr vattr; 2090 struct nfsfh *nfhp; 2091 struct nfsvattr nfsva, dnfsva; 2092 int error = 0, attrflag, dattrflag, ret; 2093 2094 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) 2095 return (error); 2096 vap->va_type = VDIR; 2097 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2098 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp, 2099 &attrflag, &dattrflag, NULL); 2100 dnp = VTONFS(dvp); 2101 mtx_lock(&dnp->n_mtx); 2102 dnp->n_flag |= NMODIFIED; 2103 if (dattrflag != 0) { 2104 mtx_unlock(&dnp->n_mtx); 2105 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2106 } else { 2107 dnp->n_attrstamp = 0; 2108 mtx_unlock(&dnp->n_mtx); 2109 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2110 } 2111 if (nfhp) { 2112 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, 2113 &np, NULL, LK_EXCLUSIVE); 2114 if (!ret) { 2115 newvp = NFSTOV(np); 2116 if (attrflag) 2117 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 2118 NULL, 0, 1); 2119 } else if (!error) 2120 error = ret; 2121 } 2122 if (!error && newvp == NULL) { 2123 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2124 cnp->cn_cred, cnp->cn_thread, &np); 2125 if (!error) { 2126 newvp = NFSTOV(np); 2127 if (newvp->v_type != VDIR) 2128 error = EEXIST; 2129 } 2130 } 2131 if (error) { 2132 if (newvp) 2133 vput(newvp); 2134 if (NFS_ISV4(dvp)) 2135 error = nfscl_maperr(cnp->cn_thread, error, 2136 vap->va_uid, vap->va_gid); 2137 } else { 2138 /* 2139 * If negative lookup caching is enabled, I might as well 2140 * add an entry for this node. Not necessary for correctness, 2141 * but if negative caching is enabled, then the system 2142 * must care about lookup caching hit rate, so... 2143 */ 2144 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2145 (cnp->cn_flags & MAKEENTRY) && 2146 attrflag != 0 && dattrflag != 0) 2147 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 2148 &dnfsva.na_ctime); 2149 *ap->a_vpp = newvp; 2150 } 2151 return (error); 2152 } 2153 2154 /* 2155 * nfs remove directory call 2156 */ 2157 static int 2158 nfs_rmdir(struct vop_rmdir_args *ap) 2159 { 2160 struct vnode *vp = ap->a_vp; 2161 struct vnode *dvp = ap->a_dvp; 2162 struct componentname *cnp = ap->a_cnp; 2163 struct nfsnode *dnp; 2164 struct nfsvattr dnfsva; 2165 int error, dattrflag; 2166 2167 if (dvp == vp) 2168 return (EINVAL); 2169 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2170 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL); 2171 dnp = VTONFS(dvp); 2172 mtx_lock(&dnp->n_mtx); 2173 dnp->n_flag |= NMODIFIED; 2174 if (dattrflag != 0) { 2175 mtx_unlock(&dnp->n_mtx); 2176 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2177 } else { 2178 dnp->n_attrstamp = 0; 2179 mtx_unlock(&dnp->n_mtx); 2180 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2181 } 2182 2183 cache_purge(dvp); 2184 cache_purge(vp); 2185 if (error && NFS_ISV4(dvp)) 2186 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, 2187 (gid_t)0); 2188 /* 2189 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2190 */ 2191 if (error == ENOENT) 2192 error = 0; 2193 return (error); 2194 } 2195 2196 /* 2197 * nfs readdir call 2198 */ 2199 static int 2200 nfs_readdir(struct vop_readdir_args *ap) 2201 { 2202 struct vnode *vp = ap->a_vp; 2203 struct nfsnode *np = VTONFS(vp); 2204 struct uio *uio = ap->a_uio; 2205 ssize_t tresid, left; 2206 int error = 0; 2207 struct vattr vattr; 2208 2209 if (ap->a_eofflag != NULL) 2210 *ap->a_eofflag = 0; 2211 if (vp->v_type != VDIR) 2212 return(EPERM); 2213 2214 /* 2215 * First, check for hit on the EOF offset cache 2216 */ 2217 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2218 (np->n_flag & NMODIFIED) == 0) { 2219 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) { 2220 mtx_lock(&np->n_mtx); 2221 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) || 2222 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 2223 mtx_unlock(&np->n_mtx); 2224 NFSINCRGLOBAL(nfsstatsv1.direofcache_hits); 2225 if (ap->a_eofflag != NULL) 2226 *ap->a_eofflag = 1; 2227 return (0); 2228 } else 2229 mtx_unlock(&np->n_mtx); 2230 } 2231 } 2232 2233 /* 2234 * NFS always guarantees that directory entries don't straddle 2235 * DIRBLKSIZ boundaries. As such, we need to limit the size 2236 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial 2237 * directory entry. 2238 */ 2239 left = uio->uio_resid % DIRBLKSIZ; 2240 if (left == uio->uio_resid) 2241 return (EINVAL); 2242 uio->uio_resid -= left; 2243 2244 /* 2245 * Call ncl_bioread() to do the real work. 2246 */ 2247 tresid = uio->uio_resid; 2248 error = ncl_bioread(vp, uio, 0, ap->a_cred); 2249 2250 if (!error && uio->uio_resid == tresid) { 2251 NFSINCRGLOBAL(nfsstatsv1.direofcache_misses); 2252 if (ap->a_eofflag != NULL) 2253 *ap->a_eofflag = 1; 2254 } 2255 2256 /* Add the partial DIRBLKSIZ (left) back in. */ 2257 uio->uio_resid += left; 2258 return (error); 2259 } 2260 2261 /* 2262 * Readdir rpc call. 2263 * Called from below the buffer cache by ncl_doio(). 2264 */ 2265 int 2266 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2267 struct thread *td) 2268 { 2269 struct nfsvattr nfsva; 2270 nfsuint64 *cookiep, cookie; 2271 struct nfsnode *dnp = VTONFS(vp); 2272 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2273 int error = 0, eof, attrflag; 2274 2275 KASSERT(uiop->uio_iovcnt == 1 && 2276 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2277 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2278 ("nfs readdirrpc bad uio")); 2279 2280 /* 2281 * If there is no cookie, assume directory was stale. 2282 */ 2283 ncl_dircookie_lock(dnp); 2284 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2285 if (cookiep) { 2286 cookie = *cookiep; 2287 ncl_dircookie_unlock(dnp); 2288 } else { 2289 ncl_dircookie_unlock(dnp); 2290 return (NFSERR_BAD_COOKIE); 2291 } 2292 2293 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2294 (void)ncl_fsinfo(nmp, vp, cred, td); 2295 2296 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva, 2297 &attrflag, &eof, NULL); 2298 if (attrflag) 2299 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2300 2301 if (!error) { 2302 /* 2303 * We are now either at the end of the directory or have filled 2304 * the block. 2305 */ 2306 if (eof) 2307 dnp->n_direofoffset = uiop->uio_offset; 2308 else { 2309 if (uiop->uio_resid > 0) 2310 printf("EEK! readdirrpc resid > 0\n"); 2311 ncl_dircookie_lock(dnp); 2312 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2313 *cookiep = cookie; 2314 ncl_dircookie_unlock(dnp); 2315 } 2316 } else if (NFS_ISV4(vp)) { 2317 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2318 } 2319 return (error); 2320 } 2321 2322 /* 2323 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc(). 2324 */ 2325 int 2326 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2327 struct thread *td) 2328 { 2329 struct nfsvattr nfsva; 2330 nfsuint64 *cookiep, cookie; 2331 struct nfsnode *dnp = VTONFS(vp); 2332 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2333 int error = 0, attrflag, eof; 2334 2335 KASSERT(uiop->uio_iovcnt == 1 && 2336 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2337 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2338 ("nfs readdirplusrpc bad uio")); 2339 2340 /* 2341 * If there is no cookie, assume directory was stale. 2342 */ 2343 ncl_dircookie_lock(dnp); 2344 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2345 if (cookiep) { 2346 cookie = *cookiep; 2347 ncl_dircookie_unlock(dnp); 2348 } else { 2349 ncl_dircookie_unlock(dnp); 2350 return (NFSERR_BAD_COOKIE); 2351 } 2352 2353 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2354 (void)ncl_fsinfo(nmp, vp, cred, td); 2355 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva, 2356 &attrflag, &eof, NULL); 2357 if (attrflag) 2358 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2359 2360 if (!error) { 2361 /* 2362 * We are now either at end of the directory or have filled the 2363 * the block. 2364 */ 2365 if (eof) 2366 dnp->n_direofoffset = uiop->uio_offset; 2367 else { 2368 if (uiop->uio_resid > 0) 2369 printf("EEK! readdirplusrpc resid > 0\n"); 2370 ncl_dircookie_lock(dnp); 2371 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2372 *cookiep = cookie; 2373 ncl_dircookie_unlock(dnp); 2374 } 2375 } else if (NFS_ISV4(vp)) { 2376 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2377 } 2378 return (error); 2379 } 2380 2381 /* 2382 * Silly rename. To make the NFS filesystem that is stateless look a little 2383 * more like the "ufs" a remove of an active vnode is translated to a rename 2384 * to a funny looking filename that is removed by nfs_inactive on the 2385 * nfsnode. There is the potential for another process on a different client 2386 * to create the same funny name between the nfs_lookitup() fails and the 2387 * nfs_rename() completes, but... 2388 */ 2389 static int 2390 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2391 { 2392 struct sillyrename *sp; 2393 struct nfsnode *np; 2394 int error; 2395 short pid; 2396 unsigned int lticks; 2397 2398 cache_purge(dvp); 2399 np = VTONFS(vp); 2400 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir")); 2401 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2402 M_NEWNFSREQ, M_WAITOK); 2403 sp->s_cred = crhold(cnp->cn_cred); 2404 sp->s_dvp = dvp; 2405 VREF(dvp); 2406 2407 /* 2408 * Fudge together a funny name. 2409 * Changing the format of the funny name to accommodate more 2410 * sillynames per directory. 2411 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is 2412 * CPU ticks since boot. 2413 */ 2414 pid = cnp->cn_thread->td_proc->p_pid; 2415 lticks = (unsigned int)ticks; 2416 for ( ; ; ) { 2417 sp->s_namlen = snprintf(sp->s_name, sizeof(sp->s_name), 2418 ".nfs.%08x.%04x4.4", lticks, 2419 pid); 2420 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2421 cnp->cn_thread, NULL)) 2422 break; 2423 lticks++; 2424 } 2425 error = nfs_renameit(dvp, vp, cnp, sp); 2426 if (error) 2427 goto bad; 2428 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2429 cnp->cn_thread, &np); 2430 np->n_sillyrename = sp; 2431 return (0); 2432 bad: 2433 vrele(sp->s_dvp); 2434 crfree(sp->s_cred); 2435 free((caddr_t)sp, M_NEWNFSREQ); 2436 return (error); 2437 } 2438 2439 /* 2440 * Look up a file name and optionally either update the file handle or 2441 * allocate an nfsnode, depending on the value of npp. 2442 * npp == NULL --> just do the lookup 2443 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2444 * handled too 2445 * *npp != NULL --> update the file handle in the vnode 2446 */ 2447 static int 2448 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred, 2449 struct thread *td, struct nfsnode **npp) 2450 { 2451 struct vnode *newvp = NULL, *vp; 2452 struct nfsnode *np, *dnp = VTONFS(dvp); 2453 struct nfsfh *nfhp, *onfhp; 2454 struct nfsvattr nfsva, dnfsva; 2455 struct componentname cn; 2456 int error = 0, attrflag, dattrflag; 2457 u_int hash; 2458 2459 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva, 2460 &nfhp, &attrflag, &dattrflag, NULL); 2461 if (dattrflag) 2462 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2463 if (npp && !error) { 2464 if (*npp != NULL) { 2465 np = *npp; 2466 vp = NFSTOV(np); 2467 /* 2468 * For NFSv4, check to see if it is the same name and 2469 * replace the name, if it is different. 2470 */ 2471 if (np->n_v4 != NULL && nfsva.na_type == VREG && 2472 (np->n_v4->n4_namelen != len || 2473 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) || 2474 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || 2475 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2476 dnp->n_fhp->nfh_len))) { 2477 #ifdef notdef 2478 { char nnn[100]; int nnnl; 2479 nnnl = (len < 100) ? len : 99; 2480 bcopy(name, nnn, nnnl); 2481 nnn[nnnl] = '\0'; 2482 printf("replace=%s\n",nnn); 2483 } 2484 #endif 2485 FREE((caddr_t)np->n_v4, M_NFSV4NODE); 2486 MALLOC(np->n_v4, struct nfsv4node *, 2487 sizeof (struct nfsv4node) + 2488 dnp->n_fhp->nfh_len + len - 1, 2489 M_NFSV4NODE, M_WAITOK); 2490 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; 2491 np->n_v4->n4_namelen = len; 2492 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2493 dnp->n_fhp->nfh_len); 2494 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len); 2495 } 2496 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, 2497 FNV1_32_INIT); 2498 onfhp = np->n_fhp; 2499 /* 2500 * Rehash node for new file handle. 2501 */ 2502 vfs_hash_rehash(vp, hash); 2503 np->n_fhp = nfhp; 2504 if (onfhp != NULL) 2505 FREE((caddr_t)onfhp, M_NFSFH); 2506 newvp = NFSTOV(np); 2507 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) { 2508 FREE((caddr_t)nfhp, M_NFSFH); 2509 VREF(dvp); 2510 newvp = dvp; 2511 } else { 2512 cn.cn_nameptr = name; 2513 cn.cn_namelen = len; 2514 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td, 2515 &np, NULL, LK_EXCLUSIVE); 2516 if (error) 2517 return (error); 2518 newvp = NFSTOV(np); 2519 } 2520 if (!attrflag && *npp == NULL) { 2521 if (newvp == dvp) 2522 vrele(newvp); 2523 else 2524 vput(newvp); 2525 return (ENOENT); 2526 } 2527 if (attrflag) 2528 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 2529 0, 1); 2530 } 2531 if (npp && *npp == NULL) { 2532 if (error) { 2533 if (newvp) { 2534 if (newvp == dvp) 2535 vrele(newvp); 2536 else 2537 vput(newvp); 2538 } 2539 } else 2540 *npp = np; 2541 } 2542 if (error && NFS_ISV4(dvp)) 2543 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2544 return (error); 2545 } 2546 2547 /* 2548 * Nfs Version 3 and 4 commit rpc 2549 */ 2550 int 2551 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred, 2552 struct thread *td) 2553 { 2554 struct nfsvattr nfsva; 2555 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2556 int error, attrflag; 2557 2558 mtx_lock(&nmp->nm_mtx); 2559 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) { 2560 mtx_unlock(&nmp->nm_mtx); 2561 return (0); 2562 } 2563 mtx_unlock(&nmp->nm_mtx); 2564 error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva, 2565 &attrflag, NULL); 2566 if (attrflag != 0) 2567 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 2568 0, 1); 2569 if (error != 0 && NFS_ISV4(vp)) 2570 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2571 return (error); 2572 } 2573 2574 /* 2575 * Strategy routine. 2576 * For async requests when nfsiod(s) are running, queue the request by 2577 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the 2578 * request. 2579 */ 2580 static int 2581 nfs_strategy(struct vop_strategy_args *ap) 2582 { 2583 struct buf *bp = ap->a_bp; 2584 struct ucred *cr; 2585 2586 KASSERT(!(bp->b_flags & B_DONE), 2587 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); 2588 BUF_ASSERT_HELD(bp); 2589 2590 if (bp->b_iocmd == BIO_READ) 2591 cr = bp->b_rcred; 2592 else 2593 cr = bp->b_wcred; 2594 2595 /* 2596 * If the op is asynchronous and an i/o daemon is waiting 2597 * queue the request, wake it up and wait for completion 2598 * otherwise just do it ourselves. 2599 */ 2600 if ((bp->b_flags & B_ASYNC) == 0 || 2601 ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread)) 2602 (void) ncl_doio(ap->a_vp, bp, cr, curthread, 1); 2603 return (0); 2604 } 2605 2606 /* 2607 * fsync vnode op. Just call ncl_flush() with commit == 1. 2608 */ 2609 /* ARGSUSED */ 2610 static int 2611 nfs_fsync(struct vop_fsync_args *ap) 2612 { 2613 2614 if (ap->a_vp->v_type != VREG) { 2615 /* 2616 * For NFS, metadata is changed synchronously on the server, 2617 * so there is nothing to flush. Also, ncl_flush() clears 2618 * the NMODIFIED flag and that shouldn't be done here for 2619 * directories. 2620 */ 2621 return (0); 2622 } 2623 return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1, 0)); 2624 } 2625 2626 /* 2627 * Flush all the blocks associated with a vnode. 2628 * Walk through the buffer pool and push any dirty pages 2629 * associated with the vnode. 2630 * If the called_from_renewthread argument is TRUE, it has been called 2631 * from the NFSv4 renew thread and, as such, cannot block indefinitely 2632 * waiting for a buffer write to complete. 2633 */ 2634 int 2635 ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td, 2636 int commit, int called_from_renewthread) 2637 { 2638 struct nfsnode *np = VTONFS(vp); 2639 struct buf *bp; 2640 int i; 2641 struct buf *nbp; 2642 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2643 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos; 2644 int passone = 1, trycnt = 0; 2645 u_quad_t off, endoff, toff; 2646 struct ucred* wcred = NULL; 2647 struct buf **bvec = NULL; 2648 struct bufobj *bo; 2649 #ifndef NFS_COMMITBVECSIZ 2650 #define NFS_COMMITBVECSIZ 20 2651 #endif 2652 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ]; 2653 int bvecsize = 0, bveccount; 2654 2655 if (called_from_renewthread != 0) 2656 slptimeo = hz; 2657 if (nmp->nm_flag & NFSMNT_INT) 2658 slpflag = PCATCH; 2659 if (!commit) 2660 passone = 0; 2661 bo = &vp->v_bufobj; 2662 /* 2663 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the 2664 * server, but has not been committed to stable storage on the server 2665 * yet. On the first pass, the byte range is worked out and the commit 2666 * rpc is done. On the second pass, ncl_writebp() is called to do the 2667 * job. 2668 */ 2669 again: 2670 off = (u_quad_t)-1; 2671 endoff = 0; 2672 bvecpos = 0; 2673 if (NFS_ISV34(vp) && commit) { 2674 if (bvec != NULL && bvec != bvec_on_stack) 2675 free(bvec, M_TEMP); 2676 /* 2677 * Count up how many buffers waiting for a commit. 2678 */ 2679 bveccount = 0; 2680 BO_LOCK(bo); 2681 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2682 if (!BUF_ISLOCKED(bp) && 2683 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) 2684 == (B_DELWRI | B_NEEDCOMMIT)) 2685 bveccount++; 2686 } 2687 /* 2688 * Allocate space to remember the list of bufs to commit. It is 2689 * important to use M_NOWAIT here to avoid a race with nfs_write. 2690 * If we can't get memory (for whatever reason), we will end up 2691 * committing the buffers one-by-one in the loop below. 2692 */ 2693 if (bveccount > NFS_COMMITBVECSIZ) { 2694 /* 2695 * Release the vnode interlock to avoid a lock 2696 * order reversal. 2697 */ 2698 BO_UNLOCK(bo); 2699 bvec = (struct buf **) 2700 malloc(bveccount * sizeof(struct buf *), 2701 M_TEMP, M_NOWAIT); 2702 BO_LOCK(bo); 2703 if (bvec == NULL) { 2704 bvec = bvec_on_stack; 2705 bvecsize = NFS_COMMITBVECSIZ; 2706 } else 2707 bvecsize = bveccount; 2708 } else { 2709 bvec = bvec_on_stack; 2710 bvecsize = NFS_COMMITBVECSIZ; 2711 } 2712 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2713 if (bvecpos >= bvecsize) 2714 break; 2715 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2716 nbp = TAILQ_NEXT(bp, b_bobufs); 2717 continue; 2718 } 2719 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 2720 (B_DELWRI | B_NEEDCOMMIT)) { 2721 BUF_UNLOCK(bp); 2722 nbp = TAILQ_NEXT(bp, b_bobufs); 2723 continue; 2724 } 2725 BO_UNLOCK(bo); 2726 bremfree(bp); 2727 /* 2728 * Work out if all buffers are using the same cred 2729 * so we can deal with them all with one commit. 2730 * 2731 * NOTE: we are not clearing B_DONE here, so we have 2732 * to do it later on in this routine if we intend to 2733 * initiate I/O on the bp. 2734 * 2735 * Note: to avoid loopback deadlocks, we do not 2736 * assign b_runningbufspace. 2737 */ 2738 if (wcred == NULL) 2739 wcred = bp->b_wcred; 2740 else if (wcred != bp->b_wcred) 2741 wcred = NOCRED; 2742 vfs_busy_pages(bp, 1); 2743 2744 BO_LOCK(bo); 2745 /* 2746 * bp is protected by being locked, but nbp is not 2747 * and vfs_busy_pages() may sleep. We have to 2748 * recalculate nbp. 2749 */ 2750 nbp = TAILQ_NEXT(bp, b_bobufs); 2751 2752 /* 2753 * A list of these buffers is kept so that the 2754 * second loop knows which buffers have actually 2755 * been committed. This is necessary, since there 2756 * may be a race between the commit rpc and new 2757 * uncommitted writes on the file. 2758 */ 2759 bvec[bvecpos++] = bp; 2760 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2761 bp->b_dirtyoff; 2762 if (toff < off) 2763 off = toff; 2764 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); 2765 if (toff > endoff) 2766 endoff = toff; 2767 } 2768 BO_UNLOCK(bo); 2769 } 2770 if (bvecpos > 0) { 2771 /* 2772 * Commit data on the server, as required. 2773 * If all bufs are using the same wcred, then use that with 2774 * one call for all of them, otherwise commit each one 2775 * separately. 2776 */ 2777 if (wcred != NOCRED) 2778 retv = ncl_commit(vp, off, (int)(endoff - off), 2779 wcred, td); 2780 else { 2781 retv = 0; 2782 for (i = 0; i < bvecpos; i++) { 2783 off_t off, size; 2784 bp = bvec[i]; 2785 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2786 bp->b_dirtyoff; 2787 size = (u_quad_t)(bp->b_dirtyend 2788 - bp->b_dirtyoff); 2789 retv = ncl_commit(vp, off, (int)size, 2790 bp->b_wcred, td); 2791 if (retv) break; 2792 } 2793 } 2794 2795 if (retv == NFSERR_STALEWRITEVERF) 2796 ncl_clearcommit(vp->v_mount); 2797 2798 /* 2799 * Now, either mark the blocks I/O done or mark the 2800 * blocks dirty, depending on whether the commit 2801 * succeeded. 2802 */ 2803 for (i = 0; i < bvecpos; i++) { 2804 bp = bvec[i]; 2805 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 2806 if (retv) { 2807 /* 2808 * Error, leave B_DELWRI intact 2809 */ 2810 vfs_unbusy_pages(bp); 2811 brelse(bp); 2812 } else { 2813 /* 2814 * Success, remove B_DELWRI ( bundirty() ). 2815 * 2816 * b_dirtyoff/b_dirtyend seem to be NFS 2817 * specific. We should probably move that 2818 * into bundirty(). XXX 2819 */ 2820 bufobj_wref(bo); 2821 bp->b_flags |= B_ASYNC; 2822 bundirty(bp); 2823 bp->b_flags &= ~B_DONE; 2824 bp->b_ioflags &= ~BIO_ERROR; 2825 bp->b_dirtyoff = bp->b_dirtyend = 0; 2826 bufdone(bp); 2827 } 2828 } 2829 } 2830 2831 /* 2832 * Start/do any write(s) that are required. 2833 */ 2834 loop: 2835 BO_LOCK(bo); 2836 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2837 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2838 if (waitfor != MNT_WAIT || passone) 2839 continue; 2840 2841 error = BUF_TIMELOCK(bp, 2842 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 2843 BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo); 2844 if (error == 0) { 2845 BUF_UNLOCK(bp); 2846 goto loop; 2847 } 2848 if (error == ENOLCK) { 2849 error = 0; 2850 goto loop; 2851 } 2852 if (called_from_renewthread != 0) { 2853 /* 2854 * Return EIO so the flush will be retried 2855 * later. 2856 */ 2857 error = EIO; 2858 goto done; 2859 } 2860 if (newnfs_sigintr(nmp, td)) { 2861 error = EINTR; 2862 goto done; 2863 } 2864 if (slpflag == PCATCH) { 2865 slpflag = 0; 2866 slptimeo = 2 * hz; 2867 } 2868 goto loop; 2869 } 2870 if ((bp->b_flags & B_DELWRI) == 0) 2871 panic("nfs_fsync: not dirty"); 2872 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) { 2873 BUF_UNLOCK(bp); 2874 continue; 2875 } 2876 BO_UNLOCK(bo); 2877 bremfree(bp); 2878 if (passone || !commit) 2879 bp->b_flags |= B_ASYNC; 2880 else 2881 bp->b_flags |= B_ASYNC; 2882 bwrite(bp); 2883 if (newnfs_sigintr(nmp, td)) { 2884 error = EINTR; 2885 goto done; 2886 } 2887 goto loop; 2888 } 2889 if (passone) { 2890 passone = 0; 2891 BO_UNLOCK(bo); 2892 goto again; 2893 } 2894 if (waitfor == MNT_WAIT) { 2895 while (bo->bo_numoutput) { 2896 error = bufobj_wwait(bo, slpflag, slptimeo); 2897 if (error) { 2898 BO_UNLOCK(bo); 2899 if (called_from_renewthread != 0) { 2900 /* 2901 * Return EIO so that the flush will be 2902 * retried later. 2903 */ 2904 error = EIO; 2905 goto done; 2906 } 2907 error = newnfs_sigintr(nmp, td); 2908 if (error) 2909 goto done; 2910 if (slpflag == PCATCH) { 2911 slpflag = 0; 2912 slptimeo = 2 * hz; 2913 } 2914 BO_LOCK(bo); 2915 } 2916 } 2917 if (bo->bo_dirty.bv_cnt != 0 && commit) { 2918 BO_UNLOCK(bo); 2919 goto loop; 2920 } 2921 /* 2922 * Wait for all the async IO requests to drain 2923 */ 2924 BO_UNLOCK(bo); 2925 mtx_lock(&np->n_mtx); 2926 while (np->n_directio_asyncwr > 0) { 2927 np->n_flag |= NFSYNCWAIT; 2928 error = newnfs_msleep(td, &np->n_directio_asyncwr, 2929 &np->n_mtx, slpflag | (PRIBIO + 1), 2930 "nfsfsync", 0); 2931 if (error) { 2932 if (newnfs_sigintr(nmp, td)) { 2933 mtx_unlock(&np->n_mtx); 2934 error = EINTR; 2935 goto done; 2936 } 2937 } 2938 } 2939 mtx_unlock(&np->n_mtx); 2940 } else 2941 BO_UNLOCK(bo); 2942 if (NFSHASPNFS(nmp)) { 2943 nfscl_layoutcommit(vp, td); 2944 /* 2945 * Invalidate the attribute cache, since writes to a DS 2946 * won't update the size attribute. 2947 */ 2948 mtx_lock(&np->n_mtx); 2949 np->n_attrstamp = 0; 2950 } else 2951 mtx_lock(&np->n_mtx); 2952 if (np->n_flag & NWRITEERR) { 2953 error = np->n_error; 2954 np->n_flag &= ~NWRITEERR; 2955 } 2956 if (commit && bo->bo_dirty.bv_cnt == 0 && 2957 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0) 2958 np->n_flag &= ~NMODIFIED; 2959 mtx_unlock(&np->n_mtx); 2960 done: 2961 if (bvec != NULL && bvec != bvec_on_stack) 2962 free(bvec, M_TEMP); 2963 if (error == 0 && commit != 0 && waitfor == MNT_WAIT && 2964 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 || 2965 np->n_directio_asyncwr != 0) && trycnt++ < 5) { 2966 /* try, try again... */ 2967 passone = 1; 2968 wcred = NULL; 2969 bvec = NULL; 2970 bvecsize = 0; 2971 printf("try%d\n", trycnt); 2972 goto again; 2973 } 2974 return (error); 2975 } 2976 2977 /* 2978 * NFS advisory byte-level locks. 2979 */ 2980 static int 2981 nfs_advlock(struct vop_advlock_args *ap) 2982 { 2983 struct vnode *vp = ap->a_vp; 2984 struct ucred *cred; 2985 struct nfsnode *np = VTONFS(ap->a_vp); 2986 struct proc *p = (struct proc *)ap->a_id; 2987 struct thread *td = curthread; /* XXX */ 2988 struct vattr va; 2989 int ret, error = EOPNOTSUPP; 2990 u_quad_t size; 2991 2992 if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) { 2993 if (vp->v_type != VREG) 2994 return (EINVAL); 2995 if ((ap->a_flags & F_POSIX) != 0) 2996 cred = p->p_ucred; 2997 else 2998 cred = td->td_ucred; 2999 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 3000 if (vp->v_iflag & VI_DOOMED) { 3001 NFSVOPUNLOCK(vp, 0); 3002 return (EBADF); 3003 } 3004 3005 /* 3006 * If this is unlocking a write locked region, flush and 3007 * commit them before unlocking. This is required by 3008 * RFC3530 Sec. 9.3.2. 3009 */ 3010 if (ap->a_op == F_UNLCK && 3011 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id, 3012 ap->a_flags)) 3013 (void) ncl_flush(vp, MNT_WAIT, cred, td, 1, 0); 3014 3015 /* 3016 * Loop around doing the lock op, while a blocking lock 3017 * must wait for the lock op to succeed. 3018 */ 3019 do { 3020 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op, 3021 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags); 3022 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3023 ap->a_op == F_SETLK) { 3024 NFSVOPUNLOCK(vp, 0); 3025 error = nfs_catnap(PZERO | PCATCH, ret, 3026 "ncladvl"); 3027 if (error) 3028 return (EINTR); 3029 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 3030 if (vp->v_iflag & VI_DOOMED) { 3031 NFSVOPUNLOCK(vp, 0); 3032 return (EBADF); 3033 } 3034 } 3035 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3036 ap->a_op == F_SETLK); 3037 if (ret == NFSERR_DENIED) { 3038 NFSVOPUNLOCK(vp, 0); 3039 return (EAGAIN); 3040 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) { 3041 NFSVOPUNLOCK(vp, 0); 3042 return (ret); 3043 } else if (ret != 0) { 3044 NFSVOPUNLOCK(vp, 0); 3045 return (EACCES); 3046 } 3047 3048 /* 3049 * Now, if we just got a lock, invalidate data in the buffer 3050 * cache, as required, so that the coherency conforms with 3051 * RFC3530 Sec. 9.3.2. 3052 */ 3053 if (ap->a_op == F_SETLK) { 3054 if ((np->n_flag & NMODIFIED) == 0) { 3055 np->n_attrstamp = 0; 3056 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3057 ret = VOP_GETATTR(vp, &va, cred); 3058 } 3059 if ((np->n_flag & NMODIFIED) || ret || 3060 np->n_change != va.va_filerev) { 3061 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1); 3062 np->n_attrstamp = 0; 3063 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3064 ret = VOP_GETATTR(vp, &va, cred); 3065 if (!ret) { 3066 np->n_mtime = va.va_mtime; 3067 np->n_change = va.va_filerev; 3068 } 3069 } 3070 /* Mark that a file lock has been acquired. */ 3071 mtx_lock(&np->n_mtx); 3072 np->n_flag |= NHASBEENLOCKED; 3073 mtx_unlock(&np->n_mtx); 3074 } 3075 NFSVOPUNLOCK(vp, 0); 3076 return (0); 3077 } else if (!NFS_ISV4(vp)) { 3078 error = NFSVOPLOCK(vp, LK_SHARED); 3079 if (error) 3080 return (error); 3081 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3082 size = VTONFS(vp)->n_size; 3083 NFSVOPUNLOCK(vp, 0); 3084 error = lf_advlock(ap, &(vp->v_lockf), size); 3085 } else { 3086 if (nfs_advlock_p != NULL) 3087 error = nfs_advlock_p(ap); 3088 else { 3089 NFSVOPUNLOCK(vp, 0); 3090 error = ENOLCK; 3091 } 3092 } 3093 if (error == 0 && ap->a_op == F_SETLK) { 3094 error = NFSVOPLOCK(vp, LK_SHARED); 3095 if (error == 0) { 3096 /* Mark that a file lock has been acquired. */ 3097 mtx_lock(&np->n_mtx); 3098 np->n_flag |= NHASBEENLOCKED; 3099 mtx_unlock(&np->n_mtx); 3100 NFSVOPUNLOCK(vp, 0); 3101 } 3102 } 3103 } 3104 return (error); 3105 } 3106 3107 /* 3108 * NFS advisory byte-level locks. 3109 */ 3110 static int 3111 nfs_advlockasync(struct vop_advlockasync_args *ap) 3112 { 3113 struct vnode *vp = ap->a_vp; 3114 u_quad_t size; 3115 int error; 3116 3117 if (NFS_ISV4(vp)) 3118 return (EOPNOTSUPP); 3119 error = NFSVOPLOCK(vp, LK_SHARED); 3120 if (error) 3121 return (error); 3122 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3123 size = VTONFS(vp)->n_size; 3124 NFSVOPUNLOCK(vp, 0); 3125 error = lf_advlockasync(ap, &(vp->v_lockf), size); 3126 } else { 3127 NFSVOPUNLOCK(vp, 0); 3128 error = EOPNOTSUPP; 3129 } 3130 return (error); 3131 } 3132 3133 /* 3134 * Print out the contents of an nfsnode. 3135 */ 3136 static int 3137 nfs_print(struct vop_print_args *ap) 3138 { 3139 struct vnode *vp = ap->a_vp; 3140 struct nfsnode *np = VTONFS(vp); 3141 3142 printf("\tfileid %ld fsid 0x%x", np->n_vattr.na_fileid, 3143 np->n_vattr.na_fsid); 3144 if (vp->v_type == VFIFO) 3145 fifo_printinfo(vp); 3146 printf("\n"); 3147 return (0); 3148 } 3149 3150 /* 3151 * This is the "real" nfs::bwrite(struct buf*). 3152 * We set B_CACHE if this is a VMIO buffer. 3153 */ 3154 int 3155 ncl_writebp(struct buf *bp, int force __unused, struct thread *td) 3156 { 3157 int s; 3158 int oldflags = bp->b_flags; 3159 #if 0 3160 int retv = 1; 3161 off_t off; 3162 #endif 3163 3164 BUF_ASSERT_HELD(bp); 3165 3166 if (bp->b_flags & B_INVAL) { 3167 brelse(bp); 3168 return(0); 3169 } 3170 3171 bp->b_flags |= B_CACHE; 3172 3173 /* 3174 * Undirty the bp. We will redirty it later if the I/O fails. 3175 */ 3176 3177 s = splbio(); 3178 bundirty(bp); 3179 bp->b_flags &= ~B_DONE; 3180 bp->b_ioflags &= ~BIO_ERROR; 3181 bp->b_iocmd = BIO_WRITE; 3182 3183 bufobj_wref(bp->b_bufobj); 3184 curthread->td_ru.ru_oublock++; 3185 splx(s); 3186 3187 /* 3188 * Note: to avoid loopback deadlocks, we do not 3189 * assign b_runningbufspace. 3190 */ 3191 vfs_busy_pages(bp, 1); 3192 3193 BUF_KERNPROC(bp); 3194 bp->b_iooffset = dbtob(bp->b_blkno); 3195 bstrategy(bp); 3196 3197 if( (oldflags & B_ASYNC) == 0) { 3198 int rtval = bufwait(bp); 3199 3200 if (oldflags & B_DELWRI) { 3201 s = splbio(); 3202 reassignbuf(bp); 3203 splx(s); 3204 } 3205 brelse(bp); 3206 return (rtval); 3207 } 3208 3209 return (0); 3210 } 3211 3212 /* 3213 * nfs special file access vnode op. 3214 * Essentially just get vattr and then imitate iaccess() since the device is 3215 * local to the client. 3216 */ 3217 static int 3218 nfsspec_access(struct vop_access_args *ap) 3219 { 3220 struct vattr *vap; 3221 struct ucred *cred = ap->a_cred; 3222 struct vnode *vp = ap->a_vp; 3223 accmode_t accmode = ap->a_accmode; 3224 struct vattr vattr; 3225 int error; 3226 3227 /* 3228 * Disallow write attempts on filesystems mounted read-only; 3229 * unless the file is a socket, fifo, or a block or character 3230 * device resident on the filesystem. 3231 */ 3232 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3233 switch (vp->v_type) { 3234 case VREG: 3235 case VDIR: 3236 case VLNK: 3237 return (EROFS); 3238 default: 3239 break; 3240 } 3241 } 3242 vap = &vattr; 3243 error = VOP_GETATTR(vp, vap, cred); 3244 if (error) 3245 goto out; 3246 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid, 3247 accmode, cred, NULL); 3248 out: 3249 return error; 3250 } 3251 3252 /* 3253 * Read wrapper for fifos. 3254 */ 3255 static int 3256 nfsfifo_read(struct vop_read_args *ap) 3257 { 3258 struct nfsnode *np = VTONFS(ap->a_vp); 3259 int error; 3260 3261 /* 3262 * Set access flag. 3263 */ 3264 mtx_lock(&np->n_mtx); 3265 np->n_flag |= NACC; 3266 vfs_timestamp(&np->n_atim); 3267 mtx_unlock(&np->n_mtx); 3268 error = fifo_specops.vop_read(ap); 3269 return error; 3270 } 3271 3272 /* 3273 * Write wrapper for fifos. 3274 */ 3275 static int 3276 nfsfifo_write(struct vop_write_args *ap) 3277 { 3278 struct nfsnode *np = VTONFS(ap->a_vp); 3279 3280 /* 3281 * Set update flag. 3282 */ 3283 mtx_lock(&np->n_mtx); 3284 np->n_flag |= NUPD; 3285 vfs_timestamp(&np->n_mtim); 3286 mtx_unlock(&np->n_mtx); 3287 return(fifo_specops.vop_write(ap)); 3288 } 3289 3290 /* 3291 * Close wrapper for fifos. 3292 * 3293 * Update the times on the nfsnode then do fifo close. 3294 */ 3295 static int 3296 nfsfifo_close(struct vop_close_args *ap) 3297 { 3298 struct vnode *vp = ap->a_vp; 3299 struct nfsnode *np = VTONFS(vp); 3300 struct vattr vattr; 3301 struct timespec ts; 3302 3303 mtx_lock(&np->n_mtx); 3304 if (np->n_flag & (NACC | NUPD)) { 3305 vfs_timestamp(&ts); 3306 if (np->n_flag & NACC) 3307 np->n_atim = ts; 3308 if (np->n_flag & NUPD) 3309 np->n_mtim = ts; 3310 np->n_flag |= NCHG; 3311 if (vrefcnt(vp) == 1 && 3312 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3313 VATTR_NULL(&vattr); 3314 if (np->n_flag & NACC) 3315 vattr.va_atime = np->n_atim; 3316 if (np->n_flag & NUPD) 3317 vattr.va_mtime = np->n_mtim; 3318 mtx_unlock(&np->n_mtx); 3319 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3320 goto out; 3321 } 3322 } 3323 mtx_unlock(&np->n_mtx); 3324 out: 3325 return (fifo_specops.vop_close(ap)); 3326 } 3327 3328 /* 3329 * Just call ncl_writebp() with the force argument set to 1. 3330 * 3331 * NOTE: B_DONE may or may not be set in a_bp on call. 3332 */ 3333 static int 3334 nfs_bwrite(struct buf *bp) 3335 { 3336 3337 return (ncl_writebp(bp, 1, curthread)); 3338 } 3339 3340 struct buf_ops buf_ops_newnfs = { 3341 .bop_name = "buf_ops_nfs", 3342 .bop_write = nfs_bwrite, 3343 .bop_strategy = bufstrategy, 3344 .bop_sync = bufsync, 3345 .bop_bdflush = bufbdflush, 3346 }; 3347 3348 static int 3349 nfs_getacl(struct vop_getacl_args *ap) 3350 { 3351 int error; 3352 3353 if (ap->a_type != ACL_TYPE_NFS4) 3354 return (EOPNOTSUPP); 3355 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3356 NULL); 3357 if (error > NFSERR_STALE) { 3358 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3359 error = EPERM; 3360 } 3361 return (error); 3362 } 3363 3364 static int 3365 nfs_setacl(struct vop_setacl_args *ap) 3366 { 3367 int error; 3368 3369 if (ap->a_type != ACL_TYPE_NFS4) 3370 return (EOPNOTSUPP); 3371 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3372 NULL); 3373 if (error > NFSERR_STALE) { 3374 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3375 error = EPERM; 3376 } 3377 return (error); 3378 } 3379 3380 /* 3381 * Return POSIX pathconf information applicable to nfs filesystems. 3382 */ 3383 static int 3384 nfs_pathconf(struct vop_pathconf_args *ap) 3385 { 3386 struct nfsv3_pathconf pc; 3387 struct nfsvattr nfsva; 3388 struct vnode *vp = ap->a_vp; 3389 struct thread *td = curthread; 3390 int attrflag, error; 3391 3392 if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX || 3393 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED || 3394 ap->a_name == _PC_NO_TRUNC)) || 3395 (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) { 3396 /* 3397 * Since only the above 4 a_names are returned by the NFSv3 3398 * Pathconf RPC, there is no point in doing it for others. 3399 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can 3400 * be used for _PC_NFS4_ACL as well. 3401 */ 3402 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva, 3403 &attrflag, NULL); 3404 if (attrflag != 0) 3405 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 3406 1); 3407 if (error != 0) 3408 return (error); 3409 } else { 3410 /* 3411 * For NFSv2 (or NFSv3 when not one of the above 4 a_names), 3412 * just fake them. 3413 */ 3414 pc.pc_linkmax = LINK_MAX; 3415 pc.pc_namemax = NFS_MAXNAMLEN; 3416 pc.pc_notrunc = 1; 3417 pc.pc_chownrestricted = 1; 3418 pc.pc_caseinsensitive = 0; 3419 pc.pc_casepreserving = 1; 3420 error = 0; 3421 } 3422 switch (ap->a_name) { 3423 case _PC_LINK_MAX: 3424 *ap->a_retval = pc.pc_linkmax; 3425 break; 3426 case _PC_NAME_MAX: 3427 *ap->a_retval = pc.pc_namemax; 3428 break; 3429 case _PC_PATH_MAX: 3430 *ap->a_retval = PATH_MAX; 3431 break; 3432 case _PC_PIPE_BUF: 3433 *ap->a_retval = PIPE_BUF; 3434 break; 3435 case _PC_CHOWN_RESTRICTED: 3436 *ap->a_retval = pc.pc_chownrestricted; 3437 break; 3438 case _PC_NO_TRUNC: 3439 *ap->a_retval = pc.pc_notrunc; 3440 break; 3441 case _PC_ACL_EXTENDED: 3442 *ap->a_retval = 0; 3443 break; 3444 case _PC_ACL_NFS4: 3445 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 && 3446 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL)) 3447 *ap->a_retval = 1; 3448 else 3449 *ap->a_retval = 0; 3450 break; 3451 case _PC_ACL_PATH_MAX: 3452 if (NFS_ISV4(vp)) 3453 *ap->a_retval = ACL_MAX_ENTRIES; 3454 else 3455 *ap->a_retval = 3; 3456 break; 3457 case _PC_MAC_PRESENT: 3458 *ap->a_retval = 0; 3459 break; 3460 case _PC_ASYNC_IO: 3461 /* _PC_ASYNC_IO should have been handled by upper layers. */ 3462 KASSERT(0, ("_PC_ASYNC_IO should not get here")); 3463 error = EINVAL; 3464 break; 3465 case _PC_PRIO_IO: 3466 *ap->a_retval = 0; 3467 break; 3468 case _PC_SYNC_IO: 3469 *ap->a_retval = 0; 3470 break; 3471 case _PC_ALLOC_SIZE_MIN: 3472 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize; 3473 break; 3474 case _PC_FILESIZEBITS: 3475 if (NFS_ISV34(vp)) 3476 *ap->a_retval = 64; 3477 else 3478 *ap->a_retval = 32; 3479 break; 3480 case _PC_REC_INCR_XFER_SIZE: 3481 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 3482 break; 3483 case _PC_REC_MAX_XFER_SIZE: 3484 *ap->a_retval = -1; /* means ``unlimited'' */ 3485 break; 3486 case _PC_REC_MIN_XFER_SIZE: 3487 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 3488 break; 3489 case _PC_REC_XFER_ALIGN: 3490 *ap->a_retval = PAGE_SIZE; 3491 break; 3492 case _PC_SYMLINK_MAX: 3493 *ap->a_retval = NFS_MAXPATHLEN; 3494 break; 3495 3496 default: 3497 error = EINVAL; 3498 break; 3499 } 3500 return (error); 3501 } 3502 3503
Indexes created Sat Oct 11 19:10:01 GMT 2025