1 /*- 2 * Copyright (c) 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * Rick Macklem at The University of Guelph. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 4. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * from nfs_vnops.c 8.16 (Berkeley) 5/27/95 33 */ 34 35 #include <sys/cdefs.h> 36 __FBSDID("$FreeBSD$"); 37 38 /* 39 * vnode op calls for Sun NFS version 2, 3 and 4 40 */ 41 42 #include "opt_inet.h" 43 44 #include <sys/param.h> 45 #include <sys/kernel.h> 46 #include <sys/systm.h> 47 #include <sys/resourcevar.h> 48 #include <sys/proc.h> 49 #include <sys/mount.h> 50 #include <sys/bio.h> 51 #include <sys/buf.h> 52 #include <sys/jail.h> 53 #include <sys/malloc.h> 54 #include <sys/mbuf.h> 55 #include <sys/namei.h> 56 #include <sys/socket.h> 57 #include <sys/vnode.h> 58 #include <sys/dirent.h> 59 #include <sys/fcntl.h> 60 #include <sys/lockf.h> 61 #include <sys/stat.h> 62 #include <sys/sysctl.h> 63 #include <sys/signalvar.h> 64 65 #include <vm/vm.h> 66 #include <vm/vm_extern.h> 67 #include <vm/vm_object.h> 68 69 #include <fs/nfs/nfsport.h> 70 #include <fs/nfsclient/nfsnode.h> 71 #include <fs/nfsclient/nfsmount.h> 72 #include <fs/nfsclient/nfs.h> 73 #include <fs/nfsclient/nfs_kdtrace.h> 74 75 #include <net/if.h> 76 #include <netinet/in.h> 77 #include <netinet/in_var.h> 78 79 #include <nfs/nfs_lock.h> 80 81 #ifdef KDTRACE_HOOKS 82 #include <sys/dtrace_bsd.h> 83 84 dtrace_nfsclient_accesscache_flush_probe_func_t 85 dtrace_nfscl_accesscache_flush_done_probe; 86 uint32_t nfscl_accesscache_flush_done_id; 87 88 dtrace_nfsclient_accesscache_get_probe_func_t 89 dtrace_nfscl_accesscache_get_hit_probe, 90 dtrace_nfscl_accesscache_get_miss_probe; 91 uint32_t nfscl_accesscache_get_hit_id; 92 uint32_t nfscl_accesscache_get_miss_id; 93 94 dtrace_nfsclient_accesscache_load_probe_func_t 95 dtrace_nfscl_accesscache_load_done_probe; 96 uint32_t nfscl_accesscache_load_done_id; 97 #endif /* !KDTRACE_HOOKS */ 98 99 /* Defs */ 100 #define TRUE 1 101 #define FALSE 0 102 103 extern struct nfsstatsv1 nfsstatsv1; 104 extern int nfsrv_useacl; 105 extern int nfscl_debuglevel; 106 MALLOC_DECLARE(M_NEWNFSREQ); 107 108 static vop_read_t nfsfifo_read; 109 static vop_write_t nfsfifo_write; 110 static vop_close_t nfsfifo_close; 111 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *, 112 struct thread *); 113 static vop_lookup_t nfs_lookup; 114 static vop_create_t nfs_create; 115 static vop_mknod_t nfs_mknod; 116 static vop_open_t nfs_open; 117 static vop_pathconf_t nfs_pathconf; 118 static vop_close_t nfs_close; 119 static vop_access_t nfs_access; 120 static vop_getattr_t nfs_getattr; 121 static vop_setattr_t nfs_setattr; 122 static vop_read_t nfs_read; 123 static vop_fsync_t nfs_fsync; 124 static vop_remove_t nfs_remove; 125 static vop_link_t nfs_link; 126 static vop_rename_t nfs_rename; 127 static vop_mkdir_t nfs_mkdir; 128 static vop_rmdir_t nfs_rmdir; 129 static vop_symlink_t nfs_symlink; 130 static vop_readdir_t nfs_readdir; 131 static vop_strategy_t nfs_strategy; 132 static int nfs_lookitup(struct vnode *, char *, int, 133 struct ucred *, struct thread *, struct nfsnode **); 134 static int nfs_sillyrename(struct vnode *, struct vnode *, 135 struct componentname *); 136 static vop_access_t nfsspec_access; 137 static vop_readlink_t nfs_readlink; 138 static vop_print_t nfs_print; 139 static vop_advlock_t nfs_advlock; 140 static vop_advlockasync_t nfs_advlockasync; 141 static vop_getacl_t nfs_getacl; 142 static vop_setacl_t nfs_setacl; 143 144 /* 145 * Global vfs data structures for nfs 146 */ 147 struct vop_vector newnfs_vnodeops = { 148 .vop_default = &default_vnodeops, 149 .vop_access = nfs_access, 150 .vop_advlock = nfs_advlock, 151 .vop_advlockasync = nfs_advlockasync, 152 .vop_close = nfs_close, 153 .vop_create = nfs_create, 154 .vop_fsync = nfs_fsync, 155 .vop_getattr = nfs_getattr, 156 .vop_getpages = ncl_getpages, 157 .vop_putpages = ncl_putpages, 158 .vop_inactive = ncl_inactive, 159 .vop_link = nfs_link, 160 .vop_lookup = nfs_lookup, 161 .vop_mkdir = nfs_mkdir, 162 .vop_mknod = nfs_mknod, 163 .vop_open = nfs_open, 164 .vop_pathconf = nfs_pathconf, 165 .vop_print = nfs_print, 166 .vop_read = nfs_read, 167 .vop_readdir = nfs_readdir, 168 .vop_readlink = nfs_readlink, 169 .vop_reclaim = ncl_reclaim, 170 .vop_remove = nfs_remove, 171 .vop_rename = nfs_rename, 172 .vop_rmdir = nfs_rmdir, 173 .vop_setattr = nfs_setattr, 174 .vop_strategy = nfs_strategy, 175 .vop_symlink = nfs_symlink, 176 .vop_write = ncl_write, 177 .vop_getacl = nfs_getacl, 178 .vop_setacl = nfs_setacl, 179 }; 180 181 struct vop_vector newnfs_fifoops = { 182 .vop_default = &fifo_specops, 183 .vop_access = nfsspec_access, 184 .vop_close = nfsfifo_close, 185 .vop_fsync = nfs_fsync, 186 .vop_getattr = nfs_getattr, 187 .vop_inactive = ncl_inactive, 188 .vop_print = nfs_print, 189 .vop_read = nfsfifo_read, 190 .vop_reclaim = ncl_reclaim, 191 .vop_setattr = nfs_setattr, 192 .vop_write = nfsfifo_write, 193 }; 194 195 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, 196 struct componentname *cnp, struct vattr *vap); 197 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 198 int namelen, struct ucred *cred, struct thread *td); 199 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, 200 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp, 201 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td); 202 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp, 203 struct componentname *scnp, struct sillyrename *sp); 204 205 /* 206 * Global variables 207 */ 208 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1)) 209 210 SYSCTL_DECL(_vfs_nfs); 211 212 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO; 213 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 214 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 215 216 static int nfs_prime_access_cache = 0; 217 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW, 218 &nfs_prime_access_cache, 0, 219 "Prime NFS ACCESS cache when fetching attributes"); 220 221 static int newnfs_commit_on_close = 0; 222 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW, 223 &newnfs_commit_on_close, 0, "write+commit on close, else only write"); 224 225 static int nfs_clean_pages_on_close = 1; 226 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW, 227 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close"); 228 229 int newnfs_directio_enable = 0; 230 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW, 231 &newnfs_directio_enable, 0, "Enable NFS directio"); 232 233 int nfs_keep_dirty_on_error; 234 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW, 235 &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned"); 236 237 /* 238 * This sysctl allows other processes to mmap a file that has been opened 239 * O_DIRECT by a process. In general, having processes mmap the file while 240 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow 241 * this by default to prevent DoS attacks - to prevent a malicious user from 242 * opening up files O_DIRECT preventing other users from mmap'ing these 243 * files. "Protected" environments where stricter consistency guarantees are 244 * required can disable this knob. The process that opened the file O_DIRECT 245 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not 246 * meaningful. 247 */ 248 int newnfs_directio_allow_mmap = 1; 249 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW, 250 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens"); 251 252 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \ 253 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \ 254 | NFSACCESS_DELETE | NFSACCESS_LOOKUP) 255 256 /* 257 * SMP Locking Note : 258 * The list of locks after the description of the lock is the ordering 259 * of other locks acquired with the lock held. 260 * np->n_mtx : Protects the fields in the nfsnode. 261 VM Object Lock 262 VI_MTX (acquired indirectly) 263 * nmp->nm_mtx : Protects the fields in the nfsmount. 264 rep->r_mtx 265 * ncl_iod_mutex : Global lock, protects shared nfsiod state. 266 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list. 267 nmp->nm_mtx 268 rep->r_mtx 269 * rep->r_mtx : Protects the fields in an nfsreq. 270 */ 271 272 static int 273 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td, 274 struct ucred *cred, u_int32_t *retmode) 275 { 276 int error = 0, attrflag, i, lrupos; 277 u_int32_t rmode; 278 struct nfsnode *np = VTONFS(vp); 279 struct nfsvattr nfsva; 280 281 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag, 282 &rmode, NULL); 283 if (attrflag) 284 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 285 if (!error) { 286 lrupos = 0; 287 mtx_lock(&np->n_mtx); 288 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 289 if (np->n_accesscache[i].uid == cred->cr_uid) { 290 np->n_accesscache[i].mode = rmode; 291 np->n_accesscache[i].stamp = time_second; 292 break; 293 } 294 if (i > 0 && np->n_accesscache[i].stamp < 295 np->n_accesscache[lrupos].stamp) 296 lrupos = i; 297 } 298 if (i == NFS_ACCESSCACHESIZE) { 299 np->n_accesscache[lrupos].uid = cred->cr_uid; 300 np->n_accesscache[lrupos].mode = rmode; 301 np->n_accesscache[lrupos].stamp = time_second; 302 } 303 mtx_unlock(&np->n_mtx); 304 if (retmode != NULL) 305 *retmode = rmode; 306 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0); 307 } else if (NFS_ISV4(vp)) { 308 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 309 } 310 #ifdef KDTRACE_HOOKS 311 if (error != 0) 312 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0, 313 error); 314 #endif 315 return (error); 316 } 317 318 /* 319 * nfs access vnode op. 320 * For nfs version 2, just return ok. File accesses may fail later. 321 * For nfs version 3, use the access rpc to check accessibility. If file modes 322 * are changed on the server, accesses might still fail later. 323 */ 324 static int 325 nfs_access(struct vop_access_args *ap) 326 { 327 struct vnode *vp = ap->a_vp; 328 int error = 0, i, gotahit; 329 u_int32_t mode, wmode, rmode; 330 int v34 = NFS_ISV34(vp); 331 struct nfsnode *np = VTONFS(vp); 332 333 /* 334 * Disallow write attempts on filesystems mounted read-only; 335 * unless the file is a socket, fifo, or a block or character 336 * device resident on the filesystem. 337 */ 338 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS | 339 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL | 340 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) { 341 switch (vp->v_type) { 342 case VREG: 343 case VDIR: 344 case VLNK: 345 return (EROFS); 346 default: 347 break; 348 } 349 } 350 /* 351 * For nfs v3 or v4, check to see if we have done this recently, and if 352 * so return our cached result instead of making an ACCESS call. 353 * If not, do an access rpc, otherwise you are stuck emulating 354 * ufs_access() locally using the vattr. This may not be correct, 355 * since the server may apply other access criteria such as 356 * client uid-->server uid mapping that we do not know about. 357 */ 358 if (v34) { 359 if (ap->a_accmode & VREAD) 360 mode = NFSACCESS_READ; 361 else 362 mode = 0; 363 if (vp->v_type != VDIR) { 364 if (ap->a_accmode & VWRITE) 365 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 366 if (ap->a_accmode & VAPPEND) 367 mode |= NFSACCESS_EXTEND; 368 if (ap->a_accmode & VEXEC) 369 mode |= NFSACCESS_EXECUTE; 370 if (ap->a_accmode & VDELETE) 371 mode |= NFSACCESS_DELETE; 372 } else { 373 if (ap->a_accmode & VWRITE) 374 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 375 if (ap->a_accmode & VAPPEND) 376 mode |= NFSACCESS_EXTEND; 377 if (ap->a_accmode & VEXEC) 378 mode |= NFSACCESS_LOOKUP; 379 if (ap->a_accmode & VDELETE) 380 mode |= NFSACCESS_DELETE; 381 if (ap->a_accmode & VDELETE_CHILD) 382 mode |= NFSACCESS_MODIFY; 383 } 384 /* XXX safety belt, only make blanket request if caching */ 385 if (nfsaccess_cache_timeout > 0) { 386 wmode = NFSACCESS_READ | NFSACCESS_MODIFY | 387 NFSACCESS_EXTEND | NFSACCESS_EXECUTE | 388 NFSACCESS_DELETE | NFSACCESS_LOOKUP; 389 } else { 390 wmode = mode; 391 } 392 393 /* 394 * Does our cached result allow us to give a definite yes to 395 * this request? 396 */ 397 gotahit = 0; 398 mtx_lock(&np->n_mtx); 399 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 400 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) { 401 if (time_second < (np->n_accesscache[i].stamp 402 + nfsaccess_cache_timeout) && 403 (np->n_accesscache[i].mode & mode) == mode) { 404 NFSINCRGLOBAL(nfsstatsv1.accesscache_hits); 405 gotahit = 1; 406 } 407 break; 408 } 409 } 410 mtx_unlock(&np->n_mtx); 411 #ifdef KDTRACE_HOOKS 412 if (gotahit != 0) 413 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp, 414 ap->a_cred->cr_uid, mode); 415 else 416 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp, 417 ap->a_cred->cr_uid, mode); 418 #endif 419 if (gotahit == 0) { 420 /* 421 * Either a no, or a don't know. Go to the wire. 422 */ 423 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses); 424 error = nfs34_access_otw(vp, wmode, ap->a_td, 425 ap->a_cred, &rmode); 426 if (!error && 427 (rmode & mode) != mode) 428 error = EACCES; 429 } 430 return (error); 431 } else { 432 if ((error = nfsspec_access(ap)) != 0) { 433 return (error); 434 } 435 /* 436 * Attempt to prevent a mapped root from accessing a file 437 * which it shouldn't. We try to read a byte from the file 438 * if the user is root and the file is not zero length. 439 * After calling nfsspec_access, we should have the correct 440 * file size cached. 441 */ 442 mtx_lock(&np->n_mtx); 443 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD) 444 && VTONFS(vp)->n_size > 0) { 445 struct iovec aiov; 446 struct uio auio; 447 char buf[1]; 448 449 mtx_unlock(&np->n_mtx); 450 aiov.iov_base = buf; 451 aiov.iov_len = 1; 452 auio.uio_iov = &aiov; 453 auio.uio_iovcnt = 1; 454 auio.uio_offset = 0; 455 auio.uio_resid = 1; 456 auio.uio_segflg = UIO_SYSSPACE; 457 auio.uio_rw = UIO_READ; 458 auio.uio_td = ap->a_td; 459 460 if (vp->v_type == VREG) 461 error = ncl_readrpc(vp, &auio, ap->a_cred); 462 else if (vp->v_type == VDIR) { 463 char* bp; 464 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 465 aiov.iov_base = bp; 466 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 467 error = ncl_readdirrpc(vp, &auio, ap->a_cred, 468 ap->a_td); 469 free(bp, M_TEMP); 470 } else if (vp->v_type == VLNK) 471 error = ncl_readlinkrpc(vp, &auio, ap->a_cred); 472 else 473 error = EACCES; 474 } else 475 mtx_unlock(&np->n_mtx); 476 return (error); 477 } 478 } 479 480 481 /* 482 * nfs open vnode op 483 * Check to see if the type is ok 484 * and that deletion is not in progress. 485 * For paged in text files, you will need to flush the page cache 486 * if consistency is lost. 487 */ 488 /* ARGSUSED */ 489 static int 490 nfs_open(struct vop_open_args *ap) 491 { 492 struct vnode *vp = ap->a_vp; 493 struct nfsnode *np = VTONFS(vp); 494 struct vattr vattr; 495 int error; 496 int fmode = ap->a_mode; 497 struct ucred *cred; 498 499 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) 500 return (EOPNOTSUPP); 501 502 /* 503 * For NFSv4, we need to do the Open Op before cache validation, 504 * so that we conform to RFC3530 Sec. 9.3.1. 505 */ 506 if (NFS_ISV4(vp)) { 507 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td); 508 if (error) { 509 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 510 (gid_t)0); 511 return (error); 512 } 513 } 514 515 /* 516 * Now, if this Open will be doing reading, re-validate/flush the 517 * cache, so that Close/Open coherency is maintained. 518 */ 519 mtx_lock(&np->n_mtx); 520 if (np->n_flag & NMODIFIED) { 521 mtx_unlock(&np->n_mtx); 522 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 523 if (error == EINTR || error == EIO) { 524 if (NFS_ISV4(vp)) 525 (void) nfsrpc_close(vp, 0, ap->a_td); 526 return (error); 527 } 528 mtx_lock(&np->n_mtx); 529 np->n_attrstamp = 0; 530 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 531 if (vp->v_type == VDIR) 532 np->n_direofoffset = 0; 533 mtx_unlock(&np->n_mtx); 534 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 535 if (error) { 536 if (NFS_ISV4(vp)) 537 (void) nfsrpc_close(vp, 0, ap->a_td); 538 return (error); 539 } 540 mtx_lock(&np->n_mtx); 541 np->n_mtime = vattr.va_mtime; 542 if (NFS_ISV4(vp)) 543 np->n_change = vattr.va_filerev; 544 } else { 545 mtx_unlock(&np->n_mtx); 546 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 547 if (error) { 548 if (NFS_ISV4(vp)) 549 (void) nfsrpc_close(vp, 0, ap->a_td); 550 return (error); 551 } 552 mtx_lock(&np->n_mtx); 553 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) || 554 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 555 if (vp->v_type == VDIR) 556 np->n_direofoffset = 0; 557 mtx_unlock(&np->n_mtx); 558 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 559 if (error == EINTR || error == EIO) { 560 if (NFS_ISV4(vp)) 561 (void) nfsrpc_close(vp, 0, ap->a_td); 562 return (error); 563 } 564 mtx_lock(&np->n_mtx); 565 np->n_mtime = vattr.va_mtime; 566 if (NFS_ISV4(vp)) 567 np->n_change = vattr.va_filerev; 568 } 569 } 570 571 /* 572 * If the object has >= 1 O_DIRECT active opens, we disable caching. 573 */ 574 if (newnfs_directio_enable && (fmode & O_DIRECT) && 575 (vp->v_type == VREG)) { 576 if (np->n_directio_opens == 0) { 577 mtx_unlock(&np->n_mtx); 578 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 579 if (error) { 580 if (NFS_ISV4(vp)) 581 (void) nfsrpc_close(vp, 0, ap->a_td); 582 return (error); 583 } 584 mtx_lock(&np->n_mtx); 585 np->n_flag |= NNONCACHE; 586 } 587 np->n_directio_opens++; 588 } 589 590 /* If opened for writing via NFSv4.1 or later, mark that for pNFS. */ 591 if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0) 592 np->n_flag |= NWRITEOPENED; 593 594 /* 595 * If this is an open for writing, capture a reference to the 596 * credentials, so they can be used by ncl_putpages(). Using 597 * these write credentials is preferable to the credentials of 598 * whatever thread happens to be doing the VOP_PUTPAGES() since 599 * the write RPCs are less likely to fail with EACCES. 600 */ 601 if ((fmode & FWRITE) != 0) { 602 cred = np->n_writecred; 603 np->n_writecred = crhold(ap->a_cred); 604 } else 605 cred = NULL; 606 mtx_unlock(&np->n_mtx); 607 608 if (cred != NULL) 609 crfree(cred); 610 vnode_create_vobject(vp, vattr.va_size, ap->a_td); 611 return (0); 612 } 613 614 /* 615 * nfs close vnode op 616 * What an NFS client should do upon close after writing is a debatable issue. 617 * Most NFS clients push delayed writes to the server upon close, basically for 618 * two reasons: 619 * 1 - So that any write errors may be reported back to the client process 620 * doing the close system call. By far the two most likely errors are 621 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 622 * 2 - To put a worst case upper bound on cache inconsistency between 623 * multiple clients for the file. 624 * There is also a consistency problem for Version 2 of the protocol w.r.t. 625 * not being able to tell if other clients are writing a file concurrently, 626 * since there is no way of knowing if the changed modify time in the reply 627 * is only due to the write for this client. 628 * (NFS Version 3 provides weak cache consistency data in the reply that 629 * should be sufficient to detect and handle this case.) 630 * 631 * The current code does the following: 632 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 633 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 634 * or commit them (this satisfies 1 and 2 except for the 635 * case where the server crashes after this close but 636 * before the commit RPC, which is felt to be "good 637 * enough". Changing the last argument to ncl_flush() to 638 * a 1 would force a commit operation, if it is felt a 639 * commit is necessary now. 640 * for NFS Version 4 - flush the dirty buffers and commit them, if 641 * nfscl_mustflush() says this is necessary. 642 * It is necessary if there is no write delegation held, 643 * in order to satisfy open/close coherency. 644 * If the file isn't cached on local stable storage, 645 * it may be necessary in order to detect "out of space" 646 * errors from the server, if the write delegation 647 * issued by the server doesn't allow the file to grow. 648 */ 649 /* ARGSUSED */ 650 static int 651 nfs_close(struct vop_close_args *ap) 652 { 653 struct vnode *vp = ap->a_vp; 654 struct nfsnode *np = VTONFS(vp); 655 struct nfsvattr nfsva; 656 struct ucred *cred; 657 int error = 0, ret, localcred = 0; 658 int fmode = ap->a_fflag; 659 660 if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF)) 661 return (0); 662 /* 663 * During shutdown, a_cred isn't valid, so just use root. 664 */ 665 if (ap->a_cred == NOCRED) { 666 cred = newnfs_getcred(); 667 localcred = 1; 668 } else { 669 cred = ap->a_cred; 670 } 671 if (vp->v_type == VREG) { 672 /* 673 * Examine and clean dirty pages, regardless of NMODIFIED. 674 * This closes a major hole in close-to-open consistency. 675 * We want to push out all dirty pages (and buffers) on 676 * close, regardless of whether they were dirtied by 677 * mmap'ed writes or via write(). 678 */ 679 if (nfs_clean_pages_on_close && vp->v_object) { 680 VM_OBJECT_WLOCK(vp->v_object); 681 vm_object_page_clean(vp->v_object, 0, 0, 0); 682 VM_OBJECT_WUNLOCK(vp->v_object); 683 } 684 mtx_lock(&np->n_mtx); 685 if (np->n_flag & NMODIFIED) { 686 mtx_unlock(&np->n_mtx); 687 if (NFS_ISV3(vp)) { 688 /* 689 * Under NFSv3 we have dirty buffers to dispose of. We 690 * must flush them to the NFS server. We have the option 691 * of waiting all the way through the commit rpc or just 692 * waiting for the initial write. The default is to only 693 * wait through the initial write so the data is in the 694 * server's cache, which is roughly similar to the state 695 * a standard disk subsystem leaves the file in on close(). 696 * 697 * We cannot clear the NMODIFIED bit in np->n_flag due to 698 * potential races with other processes, and certainly 699 * cannot clear it if we don't commit. 700 * These races occur when there is no longer the old 701 * traditional vnode locking implemented for Vnode Ops. 702 */ 703 int cm = newnfs_commit_on_close ? 1 : 0; 704 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm, 0); 705 /* np->n_flag &= ~NMODIFIED; */ 706 } else if (NFS_ISV4(vp)) { 707 if (nfscl_mustflush(vp) != 0) { 708 int cm = newnfs_commit_on_close ? 1 : 0; 709 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, 710 cm, 0); 711 /* 712 * as above w.r.t races when clearing 713 * NMODIFIED. 714 * np->n_flag &= ~NMODIFIED; 715 */ 716 } 717 } else 718 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 719 mtx_lock(&np->n_mtx); 720 } 721 /* 722 * Invalidate the attribute cache in all cases. 723 * An open is going to fetch fresh attrs any way, other procs 724 * on this node that have file open will be forced to do an 725 * otw attr fetch, but this is safe. 726 * --> A user found that their RPC count dropped by 20% when 727 * this was commented out and I can't see any requirement 728 * for it, so I've disabled it when negative lookups are 729 * enabled. (What does this have to do with negative lookup 730 * caching? Well nothing, except it was reported by the 731 * same user that needed negative lookup caching and I wanted 732 * there to be a way to disable it to see if it 733 * is the cause of some caching/coherency issue that might 734 * crop up.) 735 */ 736 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) { 737 np->n_attrstamp = 0; 738 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 739 } 740 if (np->n_flag & NWRITEERR) { 741 np->n_flag &= ~NWRITEERR; 742 error = np->n_error; 743 } 744 mtx_unlock(&np->n_mtx); 745 } 746 747 if (NFS_ISV4(vp)) { 748 /* 749 * Get attributes so "change" is up to date. 750 */ 751 if (error == 0 && nfscl_mustflush(vp) != 0 && 752 vp->v_type == VREG && 753 (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) { 754 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva, 755 NULL); 756 if (!ret) { 757 np->n_change = nfsva.na_filerev; 758 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 759 NULL, 0, 0); 760 } 761 } 762 763 /* 764 * and do the close. 765 */ 766 ret = nfsrpc_close(vp, 0, ap->a_td); 767 if (!error && ret) 768 error = ret; 769 if (error) 770 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 771 (gid_t)0); 772 } 773 if (newnfs_directio_enable) 774 KASSERT((np->n_directio_asyncwr == 0), 775 ("nfs_close: dirty unflushed (%d) directio buffers\n", 776 np->n_directio_asyncwr)); 777 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) { 778 mtx_lock(&np->n_mtx); 779 KASSERT((np->n_directio_opens > 0), 780 ("nfs_close: unexpectedly value (0) of n_directio_opens\n")); 781 np->n_directio_opens--; 782 if (np->n_directio_opens == 0) 783 np->n_flag &= ~NNONCACHE; 784 mtx_unlock(&np->n_mtx); 785 } 786 if (localcred) 787 NFSFREECRED(cred); 788 return (error); 789 } 790 791 /* 792 * nfs getattr call from vfs. 793 */ 794 static int 795 nfs_getattr(struct vop_getattr_args *ap) 796 { 797 struct vnode *vp = ap->a_vp; 798 struct thread *td = curthread; /* XXX */ 799 struct nfsnode *np = VTONFS(vp); 800 int error = 0; 801 struct nfsvattr nfsva; 802 struct vattr *vap = ap->a_vap; 803 struct vattr vattr; 804 805 /* 806 * Update local times for special files. 807 */ 808 mtx_lock(&np->n_mtx); 809 if (np->n_flag & (NACC | NUPD)) 810 np->n_flag |= NCHG; 811 mtx_unlock(&np->n_mtx); 812 /* 813 * First look in the cache. 814 */ 815 if (ncl_getattrcache(vp, &vattr) == 0) { 816 vap->va_type = vattr.va_type; 817 vap->va_mode = vattr.va_mode; 818 vap->va_nlink = vattr.va_nlink; 819 vap->va_uid = vattr.va_uid; 820 vap->va_gid = vattr.va_gid; 821 vap->va_fsid = vattr.va_fsid; 822 vap->va_fileid = vattr.va_fileid; 823 vap->va_size = vattr.va_size; 824 vap->va_blocksize = vattr.va_blocksize; 825 vap->va_atime = vattr.va_atime; 826 vap->va_mtime = vattr.va_mtime; 827 vap->va_ctime = vattr.va_ctime; 828 vap->va_gen = vattr.va_gen; 829 vap->va_flags = vattr.va_flags; 830 vap->va_rdev = vattr.va_rdev; 831 vap->va_bytes = vattr.va_bytes; 832 vap->va_filerev = vattr.va_filerev; 833 /* 834 * Get the local modify time for the case of a write 835 * delegation. 836 */ 837 nfscl_deleggetmodtime(vp, &vap->va_mtime); 838 return (0); 839 } 840 841 if (NFS_ISV34(vp) && nfs_prime_access_cache && 842 nfsaccess_cache_timeout > 0) { 843 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses); 844 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL); 845 if (ncl_getattrcache(vp, ap->a_vap) == 0) { 846 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime); 847 return (0); 848 } 849 } 850 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL); 851 if (!error) 852 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0); 853 if (!error) { 854 /* 855 * Get the local modify time for the case of a write 856 * delegation. 857 */ 858 nfscl_deleggetmodtime(vp, &vap->va_mtime); 859 } else if (NFS_ISV4(vp)) { 860 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 861 } 862 return (error); 863 } 864 865 /* 866 * nfs setattr call. 867 */ 868 static int 869 nfs_setattr(struct vop_setattr_args *ap) 870 { 871 struct vnode *vp = ap->a_vp; 872 struct nfsnode *np = VTONFS(vp); 873 struct thread *td = curthread; /* XXX */ 874 struct vattr *vap = ap->a_vap; 875 int error = 0; 876 u_quad_t tsize; 877 878 #ifndef nolint 879 tsize = (u_quad_t)0; 880 #endif 881 882 /* 883 * Setting of flags and marking of atimes are not supported. 884 */ 885 if (vap->va_flags != VNOVAL) 886 return (EOPNOTSUPP); 887 888 /* 889 * Disallow write attempts if the filesystem is mounted read-only. 890 */ 891 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 892 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 893 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 894 (vp->v_mount->mnt_flag & MNT_RDONLY)) 895 return (EROFS); 896 if (vap->va_size != VNOVAL) { 897 switch (vp->v_type) { 898 case VDIR: 899 return (EISDIR); 900 case VCHR: 901 case VBLK: 902 case VSOCK: 903 case VFIFO: 904 if (vap->va_mtime.tv_sec == VNOVAL && 905 vap->va_atime.tv_sec == VNOVAL && 906 vap->va_mode == (mode_t)VNOVAL && 907 vap->va_uid == (uid_t)VNOVAL && 908 vap->va_gid == (gid_t)VNOVAL) 909 return (0); 910 vap->va_size = VNOVAL; 911 break; 912 default: 913 /* 914 * Disallow write attempts if the filesystem is 915 * mounted read-only. 916 */ 917 if (vp->v_mount->mnt_flag & MNT_RDONLY) 918 return (EROFS); 919 /* 920 * We run vnode_pager_setsize() early (why?), 921 * we must set np->n_size now to avoid vinvalbuf 922 * V_SAVE races that might setsize a lower 923 * value. 924 */ 925 mtx_lock(&np->n_mtx); 926 tsize = np->n_size; 927 mtx_unlock(&np->n_mtx); 928 error = ncl_meta_setsize(vp, ap->a_cred, td, 929 vap->va_size); 930 mtx_lock(&np->n_mtx); 931 if (np->n_flag & NMODIFIED) { 932 tsize = np->n_size; 933 mtx_unlock(&np->n_mtx); 934 if (vap->va_size == 0) 935 error = ncl_vinvalbuf(vp, 0, td, 1); 936 else 937 error = ncl_vinvalbuf(vp, V_SAVE, td, 1); 938 if (error) { 939 vnode_pager_setsize(vp, tsize); 940 return (error); 941 } 942 /* 943 * Call nfscl_delegmodtime() to set the modify time 944 * locally, as required. 945 */ 946 nfscl_delegmodtime(vp); 947 } else 948 mtx_unlock(&np->n_mtx); 949 /* 950 * np->n_size has already been set to vap->va_size 951 * in ncl_meta_setsize(). We must set it again since 952 * nfs_loadattrcache() could be called through 953 * ncl_meta_setsize() and could modify np->n_size. 954 */ 955 mtx_lock(&np->n_mtx); 956 np->n_vattr.na_size = np->n_size = vap->va_size; 957 mtx_unlock(&np->n_mtx); 958 } 959 } else { 960 mtx_lock(&np->n_mtx); 961 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) && 962 (np->n_flag & NMODIFIED) && vp->v_type == VREG) { 963 mtx_unlock(&np->n_mtx); 964 if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 && 965 (error == EINTR || error == EIO)) 966 return (error); 967 } else 968 mtx_unlock(&np->n_mtx); 969 } 970 error = nfs_setattrrpc(vp, vap, ap->a_cred, td); 971 if (error && vap->va_size != VNOVAL) { 972 mtx_lock(&np->n_mtx); 973 np->n_size = np->n_vattr.na_size = tsize; 974 vnode_pager_setsize(vp, tsize); 975 mtx_unlock(&np->n_mtx); 976 } 977 return (error); 978 } 979 980 /* 981 * Do an nfs setattr rpc. 982 */ 983 static int 984 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred, 985 struct thread *td) 986 { 987 struct nfsnode *np = VTONFS(vp); 988 int error, ret, attrflag, i; 989 struct nfsvattr nfsva; 990 991 if (NFS_ISV34(vp)) { 992 mtx_lock(&np->n_mtx); 993 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) 994 np->n_accesscache[i].stamp = 0; 995 np->n_flag |= NDELEGMOD; 996 mtx_unlock(&np->n_mtx); 997 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp); 998 } 999 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag, 1000 NULL); 1001 if (attrflag) { 1002 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1003 if (ret && !error) 1004 error = ret; 1005 } 1006 if (error && NFS_ISV4(vp)) 1007 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid); 1008 return (error); 1009 } 1010 1011 /* 1012 * nfs lookup call, one step at a time... 1013 * First look in cache 1014 * If not found, unlock the directory nfsnode and do the rpc 1015 */ 1016 static int 1017 nfs_lookup(struct vop_lookup_args *ap) 1018 { 1019 struct componentname *cnp = ap->a_cnp; 1020 struct vnode *dvp = ap->a_dvp; 1021 struct vnode **vpp = ap->a_vpp; 1022 struct mount *mp = dvp->v_mount; 1023 int flags = cnp->cn_flags; 1024 struct vnode *newvp; 1025 struct nfsmount *nmp; 1026 struct nfsnode *np, *newnp; 1027 int error = 0, attrflag, dattrflag, ltype, ncticks; 1028 struct thread *td = cnp->cn_thread; 1029 struct nfsfh *nfhp; 1030 struct nfsvattr dnfsva, nfsva; 1031 struct vattr vattr; 1032 struct timespec nctime; 1033 1034 *vpp = NULLVP; 1035 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) && 1036 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 1037 return (EROFS); 1038 if (dvp->v_type != VDIR) 1039 return (ENOTDIR); 1040 nmp = VFSTONFS(mp); 1041 np = VTONFS(dvp); 1042 1043 /* For NFSv4, wait until any remove is done. */ 1044 mtx_lock(&np->n_mtx); 1045 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) { 1046 np->n_flag |= NREMOVEWANT; 1047 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0); 1048 } 1049 mtx_unlock(&np->n_mtx); 1050 1051 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) 1052 return (error); 1053 error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks); 1054 if (error > 0 && error != ENOENT) 1055 return (error); 1056 if (error == -1) { 1057 /* 1058 * Lookups of "." are special and always return the 1059 * current directory. cache_lookup() already handles 1060 * associated locking bookkeeping, etc. 1061 */ 1062 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 1063 /* XXX: Is this really correct? */ 1064 if (cnp->cn_nameiop != LOOKUP && 1065 (flags & ISLASTCN)) 1066 cnp->cn_flags |= SAVENAME; 1067 return (0); 1068 } 1069 1070 /* 1071 * We only accept a positive hit in the cache if the 1072 * change time of the file matches our cached copy. 1073 * Otherwise, we discard the cache entry and fallback 1074 * to doing a lookup RPC. We also only trust cache 1075 * entries for less than nm_nametimeo seconds. 1076 * 1077 * To better handle stale file handles and attributes, 1078 * clear the attribute cache of this node if it is a 1079 * leaf component, part of an open() call, and not 1080 * locally modified before fetching the attributes. 1081 * This should allow stale file handles to be detected 1082 * here where we can fall back to a LOOKUP RPC to 1083 * recover rather than having nfs_open() detect the 1084 * stale file handle and failing open(2) with ESTALE. 1085 */ 1086 newvp = *vpp; 1087 newnp = VTONFS(newvp); 1088 if (!(nmp->nm_flag & NFSMNT_NOCTO) && 1089 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1090 !(newnp->n_flag & NMODIFIED)) { 1091 mtx_lock(&newnp->n_mtx); 1092 newnp->n_attrstamp = 0; 1093 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1094 mtx_unlock(&newnp->n_mtx); 1095 } 1096 if (nfscl_nodeleg(newvp, 0) == 0 || 1097 ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) && 1098 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 && 1099 timespeccmp(&vattr.va_ctime, &nctime, ==))) { 1100 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits); 1101 if (cnp->cn_nameiop != LOOKUP && 1102 (flags & ISLASTCN)) 1103 cnp->cn_flags |= SAVENAME; 1104 return (0); 1105 } 1106 cache_purge(newvp); 1107 if (dvp != newvp) 1108 vput(newvp); 1109 else 1110 vrele(newvp); 1111 *vpp = NULLVP; 1112 } else if (error == ENOENT) { 1113 if (dvp->v_iflag & VI_DOOMED) 1114 return (ENOENT); 1115 /* 1116 * We only accept a negative hit in the cache if the 1117 * modification time of the parent directory matches 1118 * the cached copy in the name cache entry. 1119 * Otherwise, we discard all of the negative cache 1120 * entries for this directory. We also only trust 1121 * negative cache entries for up to nm_negnametimeo 1122 * seconds. 1123 */ 1124 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) && 1125 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 && 1126 timespeccmp(&vattr.va_mtime, &nctime, ==)) { 1127 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits); 1128 return (ENOENT); 1129 } 1130 cache_purge_negative(dvp); 1131 } 1132 1133 error = 0; 1134 newvp = NULLVP; 1135 NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses); 1136 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1137 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1138 NULL); 1139 if (dattrflag) 1140 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1141 if (error) { 1142 if (newvp != NULLVP) { 1143 vput(newvp); 1144 *vpp = NULLVP; 1145 } 1146 1147 if (error != ENOENT) { 1148 if (NFS_ISV4(dvp)) 1149 error = nfscl_maperr(td, error, (uid_t)0, 1150 (gid_t)0); 1151 return (error); 1152 } 1153 1154 /* The requested file was not found. */ 1155 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1156 (flags & ISLASTCN)) { 1157 /* 1158 * XXX: UFS does a full VOP_ACCESS(dvp, 1159 * VWRITE) here instead of just checking 1160 * MNT_RDONLY. 1161 */ 1162 if (mp->mnt_flag & MNT_RDONLY) 1163 return (EROFS); 1164 cnp->cn_flags |= SAVENAME; 1165 return (EJUSTRETURN); 1166 } 1167 1168 if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) { 1169 /* 1170 * Cache the modification time of the parent 1171 * directory from the post-op attributes in 1172 * the name cache entry. The negative cache 1173 * entry will be ignored once the directory 1174 * has changed. Don't bother adding the entry 1175 * if the directory has already changed. 1176 */ 1177 mtx_lock(&np->n_mtx); 1178 if (timespeccmp(&np->n_vattr.na_mtime, 1179 &dnfsva.na_mtime, ==)) { 1180 mtx_unlock(&np->n_mtx); 1181 cache_enter_time(dvp, NULL, cnp, 1182 &dnfsva.na_mtime, NULL); 1183 } else 1184 mtx_unlock(&np->n_mtx); 1185 } 1186 return (ENOENT); 1187 } 1188 1189 /* 1190 * Handle RENAME case... 1191 */ 1192 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 1193 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1194 FREE((caddr_t)nfhp, M_NFSFH); 1195 return (EISDIR); 1196 } 1197 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1198 LK_EXCLUSIVE); 1199 if (error) 1200 return (error); 1201 newvp = NFSTOV(np); 1202 if (attrflag) 1203 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1204 0, 1); 1205 *vpp = newvp; 1206 cnp->cn_flags |= SAVENAME; 1207 return (0); 1208 } 1209 1210 if (flags & ISDOTDOT) { 1211 ltype = NFSVOPISLOCKED(dvp); 1212 error = vfs_busy(mp, MBF_NOWAIT); 1213 if (error != 0) { 1214 vfs_ref(mp); 1215 NFSVOPUNLOCK(dvp, 0); 1216 error = vfs_busy(mp, 0); 1217 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1218 vfs_rel(mp); 1219 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) { 1220 vfs_unbusy(mp); 1221 error = ENOENT; 1222 } 1223 if (error != 0) 1224 return (error); 1225 } 1226 NFSVOPUNLOCK(dvp, 0); 1227 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1228 cnp->cn_lkflags); 1229 if (error == 0) 1230 newvp = NFSTOV(np); 1231 vfs_unbusy(mp); 1232 if (newvp != dvp) 1233 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1234 if (dvp->v_iflag & VI_DOOMED) { 1235 if (error == 0) { 1236 if (newvp == dvp) 1237 vrele(newvp); 1238 else 1239 vput(newvp); 1240 } 1241 error = ENOENT; 1242 } 1243 if (error != 0) 1244 return (error); 1245 if (attrflag) 1246 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1247 0, 1); 1248 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1249 FREE((caddr_t)nfhp, M_NFSFH); 1250 VREF(dvp); 1251 newvp = dvp; 1252 if (attrflag) 1253 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1254 0, 1); 1255 } else { 1256 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1257 cnp->cn_lkflags); 1258 if (error) 1259 return (error); 1260 newvp = NFSTOV(np); 1261 if (attrflag) 1262 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1263 0, 1); 1264 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1265 !(np->n_flag & NMODIFIED)) { 1266 /* 1267 * Flush the attribute cache when opening a 1268 * leaf node to ensure that fresh attributes 1269 * are fetched in nfs_open() since we did not 1270 * fetch attributes from the LOOKUP reply. 1271 */ 1272 mtx_lock(&np->n_mtx); 1273 np->n_attrstamp = 0; 1274 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1275 mtx_unlock(&np->n_mtx); 1276 } 1277 } 1278 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1279 cnp->cn_flags |= SAVENAME; 1280 if ((cnp->cn_flags & MAKEENTRY) && 1281 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) && 1282 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0)) 1283 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 1284 newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime); 1285 *vpp = newvp; 1286 return (0); 1287 } 1288 1289 /* 1290 * nfs read call. 1291 * Just call ncl_bioread() to do the work. 1292 */ 1293 static int 1294 nfs_read(struct vop_read_args *ap) 1295 { 1296 struct vnode *vp = ap->a_vp; 1297 1298 switch (vp->v_type) { 1299 case VREG: 1300 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred)); 1301 case VDIR: 1302 return (EISDIR); 1303 default: 1304 return (EOPNOTSUPP); 1305 } 1306 } 1307 1308 /* 1309 * nfs readlink call 1310 */ 1311 static int 1312 nfs_readlink(struct vop_readlink_args *ap) 1313 { 1314 struct vnode *vp = ap->a_vp; 1315 1316 if (vp->v_type != VLNK) 1317 return (EINVAL); 1318 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred)); 1319 } 1320 1321 /* 1322 * Do a readlink rpc. 1323 * Called by ncl_doio() from below the buffer cache. 1324 */ 1325 int 1326 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1327 { 1328 int error, ret, attrflag; 1329 struct nfsvattr nfsva; 1330 1331 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva, 1332 &attrflag, NULL); 1333 if (attrflag) { 1334 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1335 if (ret && !error) 1336 error = ret; 1337 } 1338 if (error && NFS_ISV4(vp)) 1339 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1340 return (error); 1341 } 1342 1343 /* 1344 * nfs read rpc call 1345 * Ditto above 1346 */ 1347 int 1348 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1349 { 1350 int error, ret, attrflag; 1351 struct nfsvattr nfsva; 1352 struct nfsmount *nmp; 1353 1354 nmp = VFSTONFS(vnode_mount(vp)); 1355 error = EIO; 1356 attrflag = 0; 1357 if (NFSHASPNFS(nmp)) 1358 error = nfscl_doiods(vp, uiop, NULL, NULL, 1359 NFSV4OPEN_ACCESSREAD, cred, uiop->uio_td); 1360 NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error); 1361 if (error != 0) 1362 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, 1363 &attrflag, NULL); 1364 if (attrflag) { 1365 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1366 if (ret && !error) 1367 error = ret; 1368 } 1369 if (error && NFS_ISV4(vp)) 1370 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1371 return (error); 1372 } 1373 1374 /* 1375 * nfs write call 1376 */ 1377 int 1378 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 1379 int *iomode, int *must_commit, int called_from_strategy) 1380 { 1381 struct nfsvattr nfsva; 1382 int error, attrflag, ret; 1383 struct nfsmount *nmp; 1384 1385 nmp = VFSTONFS(vnode_mount(vp)); 1386 error = EIO; 1387 attrflag = 0; 1388 if (NFSHASPNFS(nmp)) 1389 error = nfscl_doiods(vp, uiop, iomode, must_commit, 1390 NFSV4OPEN_ACCESSWRITE, cred, uiop->uio_td); 1391 NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error); 1392 if (error != 0) 1393 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred, 1394 uiop->uio_td, &nfsva, &attrflag, NULL, 1395 called_from_strategy); 1396 if (attrflag) { 1397 if (VTONFS(vp)->n_flag & ND_NFSV4) 1398 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1, 1399 1); 1400 else 1401 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1402 1); 1403 if (ret && !error) 1404 error = ret; 1405 } 1406 if (DOINGASYNC(vp)) 1407 *iomode = NFSWRITE_FILESYNC; 1408 if (error && NFS_ISV4(vp)) 1409 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1410 return (error); 1411 } 1412 1413 /* 1414 * nfs mknod rpc 1415 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1416 * mode set to specify the file type and the size field for rdev. 1417 */ 1418 static int 1419 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1420 struct vattr *vap) 1421 { 1422 struct nfsvattr nfsva, dnfsva; 1423 struct vnode *newvp = NULL; 1424 struct nfsnode *np = NULL, *dnp; 1425 struct nfsfh *nfhp; 1426 struct vattr vattr; 1427 int error = 0, attrflag, dattrflag; 1428 u_int32_t rdev; 1429 1430 if (vap->va_type == VCHR || vap->va_type == VBLK) 1431 rdev = vap->va_rdev; 1432 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1433 rdev = 0xffffffff; 1434 else 1435 return (EOPNOTSUPP); 1436 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1437 return (error); 1438 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap, 1439 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva, 1440 &nfsva, &nfhp, &attrflag, &dattrflag, NULL); 1441 if (!error) { 1442 if (!nfhp) 1443 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1444 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, 1445 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1446 NULL); 1447 if (nfhp) 1448 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1449 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); 1450 } 1451 if (dattrflag) 1452 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1453 if (!error) { 1454 newvp = NFSTOV(np); 1455 if (attrflag != 0) { 1456 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1457 0, 1); 1458 if (error != 0) 1459 vput(newvp); 1460 } 1461 } 1462 if (!error) { 1463 *vpp = newvp; 1464 } else if (NFS_ISV4(dvp)) { 1465 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, 1466 vap->va_gid); 1467 } 1468 dnp = VTONFS(dvp); 1469 mtx_lock(&dnp->n_mtx); 1470 dnp->n_flag |= NMODIFIED; 1471 if (!dattrflag) { 1472 dnp->n_attrstamp = 0; 1473 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1474 } 1475 mtx_unlock(&dnp->n_mtx); 1476 return (error); 1477 } 1478 1479 /* 1480 * nfs mknod vop 1481 * just call nfs_mknodrpc() to do the work. 1482 */ 1483 /* ARGSUSED */ 1484 static int 1485 nfs_mknod(struct vop_mknod_args *ap) 1486 { 1487 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap)); 1488 } 1489 1490 static struct mtx nfs_cverf_mtx; 1491 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex", 1492 MTX_DEF); 1493 1494 static nfsquad_t 1495 nfs_get_cverf(void) 1496 { 1497 static nfsquad_t cverf; 1498 nfsquad_t ret; 1499 static int cverf_initialized = 0; 1500 1501 mtx_lock(&nfs_cverf_mtx); 1502 if (cverf_initialized == 0) { 1503 cverf.lval[0] = arc4random(); 1504 cverf.lval[1] = arc4random(); 1505 cverf_initialized = 1; 1506 } else 1507 cverf.qval++; 1508 ret = cverf; 1509 mtx_unlock(&nfs_cverf_mtx); 1510 1511 return (ret); 1512 } 1513 1514 /* 1515 * nfs file create call 1516 */ 1517 static int 1518 nfs_create(struct vop_create_args *ap) 1519 { 1520 struct vnode *dvp = ap->a_dvp; 1521 struct vattr *vap = ap->a_vap; 1522 struct componentname *cnp = ap->a_cnp; 1523 struct nfsnode *np = NULL, *dnp; 1524 struct vnode *newvp = NULL; 1525 struct nfsmount *nmp; 1526 struct nfsvattr dnfsva, nfsva; 1527 struct nfsfh *nfhp; 1528 nfsquad_t cverf; 1529 int error = 0, attrflag, dattrflag, fmode = 0; 1530 struct vattr vattr; 1531 1532 /* 1533 * Oops, not for me.. 1534 */ 1535 if (vap->va_type == VSOCK) 1536 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1537 1538 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1539 return (error); 1540 if (vap->va_vaflags & VA_EXCLUSIVE) 1541 fmode |= O_EXCL; 1542 dnp = VTONFS(dvp); 1543 nmp = VFSTONFS(vnode_mount(dvp)); 1544 again: 1545 /* For NFSv4, wait until any remove is done. */ 1546 mtx_lock(&dnp->n_mtx); 1547 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) { 1548 dnp->n_flag |= NREMOVEWANT; 1549 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0); 1550 } 1551 mtx_unlock(&dnp->n_mtx); 1552 1553 cverf = nfs_get_cverf(); 1554 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1555 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, 1556 &nfhp, &attrflag, &dattrflag, NULL); 1557 if (!error) { 1558 if (nfhp == NULL) 1559 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1560 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, 1561 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1562 NULL); 1563 if (nfhp != NULL) 1564 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1565 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); 1566 } 1567 if (dattrflag) 1568 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1569 if (!error) { 1570 newvp = NFSTOV(np); 1571 if (attrflag == 0) 1572 error = nfsrpc_getattr(newvp, cnp->cn_cred, 1573 cnp->cn_thread, &nfsva, NULL); 1574 if (error == 0) 1575 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1576 0, 1); 1577 } 1578 if (error) { 1579 if (newvp != NULL) { 1580 vput(newvp); 1581 newvp = NULL; 1582 } 1583 if (NFS_ISV34(dvp) && (fmode & O_EXCL) && 1584 error == NFSERR_NOTSUPP) { 1585 fmode &= ~O_EXCL; 1586 goto again; 1587 } 1588 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) { 1589 if (nfscl_checksattr(vap, &nfsva)) { 1590 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred, 1591 cnp->cn_thread, &nfsva, &attrflag, NULL); 1592 if (error && (vap->va_uid != (uid_t)VNOVAL || 1593 vap->va_gid != (gid_t)VNOVAL)) { 1594 /* try again without setting uid/gid */ 1595 vap->va_uid = (uid_t)VNOVAL; 1596 vap->va_gid = (uid_t)VNOVAL; 1597 error = nfsrpc_setattr(newvp, vap, NULL, 1598 cnp->cn_cred, cnp->cn_thread, &nfsva, 1599 &attrflag, NULL); 1600 } 1601 if (attrflag) 1602 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 1603 NULL, 0, 1); 1604 if (error != 0) 1605 vput(newvp); 1606 } 1607 } 1608 if (!error) { 1609 if ((cnp->cn_flags & MAKEENTRY) && attrflag) 1610 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 1611 NULL); 1612 *ap->a_vpp = newvp; 1613 } else if (NFS_ISV4(dvp)) { 1614 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, 1615 vap->va_gid); 1616 } 1617 mtx_lock(&dnp->n_mtx); 1618 dnp->n_flag |= NMODIFIED; 1619 if (!dattrflag) { 1620 dnp->n_attrstamp = 0; 1621 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1622 } 1623 mtx_unlock(&dnp->n_mtx); 1624 return (error); 1625 } 1626 1627 /* 1628 * nfs file remove call 1629 * To try and make nfs semantics closer to ufs semantics, a file that has 1630 * other processes using the vnode is renamed instead of removed and then 1631 * removed later on the last close. 1632 * - If v_usecount > 1 1633 * If a rename is not already in the works 1634 * call nfs_sillyrename() to set it up 1635 * else 1636 * do the remove rpc 1637 */ 1638 static int 1639 nfs_remove(struct vop_remove_args *ap) 1640 { 1641 struct vnode *vp = ap->a_vp; 1642 struct vnode *dvp = ap->a_dvp; 1643 struct componentname *cnp = ap->a_cnp; 1644 struct nfsnode *np = VTONFS(vp); 1645 int error = 0; 1646 struct vattr vattr; 1647 1648 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name")); 1649 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount")); 1650 if (vp->v_type == VDIR) 1651 error = EPERM; 1652 else if (vrefcnt(vp) == 1 || (np->n_sillyrename && 1653 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 1654 vattr.va_nlink > 1)) { 1655 /* 1656 * Purge the name cache so that the chance of a lookup for 1657 * the name succeeding while the remove is in progress is 1658 * minimized. Without node locking it can still happen, such 1659 * that an I/O op returns ESTALE, but since you get this if 1660 * another host removes the file.. 1661 */ 1662 cache_purge(vp); 1663 /* 1664 * throw away biocache buffers, mainly to avoid 1665 * unnecessary delayed writes later. 1666 */ 1667 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1); 1668 /* Do the rpc */ 1669 if (error != EINTR && error != EIO) 1670 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr, 1671 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread); 1672 /* 1673 * Kludge City: If the first reply to the remove rpc is lost.. 1674 * the reply to the retransmitted request will be ENOENT 1675 * since the file was in fact removed 1676 * Therefore, we cheat and return success. 1677 */ 1678 if (error == ENOENT) 1679 error = 0; 1680 } else if (!np->n_sillyrename) 1681 error = nfs_sillyrename(dvp, vp, cnp); 1682 mtx_lock(&np->n_mtx); 1683 np->n_attrstamp = 0; 1684 mtx_unlock(&np->n_mtx); 1685 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 1686 return (error); 1687 } 1688 1689 /* 1690 * nfs file remove rpc called from nfs_inactive 1691 */ 1692 int 1693 ncl_removeit(struct sillyrename *sp, struct vnode *vp) 1694 { 1695 /* 1696 * Make sure that the directory vnode is still valid. 1697 * XXX we should lock sp->s_dvp here. 1698 */ 1699 if (sp->s_dvp->v_type == VBAD) 1700 return (0); 1701 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen, 1702 sp->s_cred, NULL)); 1703 } 1704 1705 /* 1706 * Nfs remove rpc, called from nfs_remove() and ncl_removeit(). 1707 */ 1708 static int 1709 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 1710 int namelen, struct ucred *cred, struct thread *td) 1711 { 1712 struct nfsvattr dnfsva; 1713 struct nfsnode *dnp = VTONFS(dvp); 1714 int error = 0, dattrflag; 1715 1716 mtx_lock(&dnp->n_mtx); 1717 dnp->n_flag |= NREMOVEINPROG; 1718 mtx_unlock(&dnp->n_mtx); 1719 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva, 1720 &dattrflag, NULL); 1721 mtx_lock(&dnp->n_mtx); 1722 if ((dnp->n_flag & NREMOVEWANT)) { 1723 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG); 1724 mtx_unlock(&dnp->n_mtx); 1725 wakeup((caddr_t)dnp); 1726 } else { 1727 dnp->n_flag &= ~NREMOVEINPROG; 1728 mtx_unlock(&dnp->n_mtx); 1729 } 1730 if (dattrflag) 1731 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1732 mtx_lock(&dnp->n_mtx); 1733 dnp->n_flag |= NMODIFIED; 1734 if (!dattrflag) { 1735 dnp->n_attrstamp = 0; 1736 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1737 } 1738 mtx_unlock(&dnp->n_mtx); 1739 if (error && NFS_ISV4(dvp)) 1740 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1741 return (error); 1742 } 1743 1744 /* 1745 * nfs file rename call 1746 */ 1747 static int 1748 nfs_rename(struct vop_rename_args *ap) 1749 { 1750 struct vnode *fvp = ap->a_fvp; 1751 struct vnode *tvp = ap->a_tvp; 1752 struct vnode *fdvp = ap->a_fdvp; 1753 struct vnode *tdvp = ap->a_tdvp; 1754 struct componentname *tcnp = ap->a_tcnp; 1755 struct componentname *fcnp = ap->a_fcnp; 1756 struct nfsnode *fnp = VTONFS(ap->a_fvp); 1757 struct nfsnode *tdnp = VTONFS(ap->a_tdvp); 1758 struct nfsv4node *newv4 = NULL; 1759 int error; 1760 1761 KASSERT((tcnp->cn_flags & HASBUF) != 0 && 1762 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name")); 1763 /* Check for cross-device rename */ 1764 if ((fvp->v_mount != tdvp->v_mount) || 1765 (tvp && (fvp->v_mount != tvp->v_mount))) { 1766 error = EXDEV; 1767 goto out; 1768 } 1769 1770 if (fvp == tvp) { 1771 printf("nfs_rename: fvp == tvp (can't happen)\n"); 1772 error = 0; 1773 goto out; 1774 } 1775 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0) 1776 goto out; 1777 1778 /* 1779 * We have to flush B_DELWRI data prior to renaming 1780 * the file. If we don't, the delayed-write buffers 1781 * can be flushed out later after the file has gone stale 1782 * under NFSV3. NFSV2 does not have this problem because 1783 * ( as far as I can tell ) it flushes dirty buffers more 1784 * often. 1785 * 1786 * Skip the rename operation if the fsync fails, this can happen 1787 * due to the server's volume being full, when we pushed out data 1788 * that was written back to our cache earlier. Not checking for 1789 * this condition can result in potential (silent) data loss. 1790 */ 1791 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread); 1792 NFSVOPUNLOCK(fvp, 0); 1793 if (!error && tvp) 1794 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread); 1795 if (error) 1796 goto out; 1797 1798 /* 1799 * If the tvp exists and is in use, sillyrename it before doing the 1800 * rename of the new file over it. 1801 * XXX Can't sillyrename a directory. 1802 */ 1803 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename && 1804 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1805 vput(tvp); 1806 tvp = NULL; 1807 } 1808 1809 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1810 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1811 tcnp->cn_thread); 1812 1813 if (error == 0 && NFS_ISV4(tdvp)) { 1814 /* 1815 * For NFSv4, check to see if it is the same name and 1816 * replace the name, if it is different. 1817 */ 1818 MALLOC(newv4, struct nfsv4node *, 1819 sizeof (struct nfsv4node) + 1820 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1, 1821 M_NFSV4NODE, M_WAITOK); 1822 mtx_lock(&tdnp->n_mtx); 1823 mtx_lock(&fnp->n_mtx); 1824 if (fnp->n_v4 != NULL && fvp->v_type == VREG && 1825 (fnp->n_v4->n4_namelen != tcnp->cn_namelen || 1826 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4), 1827 tcnp->cn_namelen) || 1828 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen || 1829 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 1830 tdnp->n_fhp->nfh_len))) { 1831 #ifdef notdef 1832 { char nnn[100]; int nnnl; 1833 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99; 1834 bcopy(tcnp->cn_nameptr, nnn, nnnl); 1835 nnn[nnnl] = '\0'; 1836 printf("ren replace=%s\n",nnn); 1837 } 1838 #endif 1839 FREE((caddr_t)fnp->n_v4, M_NFSV4NODE); 1840 fnp->n_v4 = newv4; 1841 newv4 = NULL; 1842 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len; 1843 fnp->n_v4->n4_namelen = tcnp->cn_namelen; 1844 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 1845 tdnp->n_fhp->nfh_len); 1846 NFSBCOPY(tcnp->cn_nameptr, 1847 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen); 1848 } 1849 mtx_unlock(&tdnp->n_mtx); 1850 mtx_unlock(&fnp->n_mtx); 1851 if (newv4 != NULL) 1852 FREE((caddr_t)newv4, M_NFSV4NODE); 1853 } 1854 1855 if (fvp->v_type == VDIR) { 1856 if (tvp != NULL && tvp->v_type == VDIR) 1857 cache_purge(tdvp); 1858 cache_purge(fdvp); 1859 } 1860 1861 out: 1862 if (tdvp == tvp) 1863 vrele(tdvp); 1864 else 1865 vput(tdvp); 1866 if (tvp) 1867 vput(tvp); 1868 vrele(fdvp); 1869 vrele(fvp); 1870 /* 1871 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1872 */ 1873 if (error == ENOENT) 1874 error = 0; 1875 return (error); 1876 } 1877 1878 /* 1879 * nfs file rename rpc called from nfs_remove() above 1880 */ 1881 static int 1882 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp, 1883 struct sillyrename *sp) 1884 { 1885 1886 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen, 1887 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred, 1888 scnp->cn_thread)); 1889 } 1890 1891 /* 1892 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1893 */ 1894 static int 1895 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr, 1896 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr, 1897 int tnamelen, struct ucred *cred, struct thread *td) 1898 { 1899 struct nfsvattr fnfsva, tnfsva; 1900 struct nfsnode *fdnp = VTONFS(fdvp); 1901 struct nfsnode *tdnp = VTONFS(tdvp); 1902 int error = 0, fattrflag, tattrflag; 1903 1904 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp, 1905 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag, 1906 &tattrflag, NULL, NULL); 1907 mtx_lock(&fdnp->n_mtx); 1908 fdnp->n_flag |= NMODIFIED; 1909 if (fattrflag != 0) { 1910 mtx_unlock(&fdnp->n_mtx); 1911 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1); 1912 } else { 1913 fdnp->n_attrstamp = 0; 1914 mtx_unlock(&fdnp->n_mtx); 1915 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp); 1916 } 1917 mtx_lock(&tdnp->n_mtx); 1918 tdnp->n_flag |= NMODIFIED; 1919 if (tattrflag != 0) { 1920 mtx_unlock(&tdnp->n_mtx); 1921 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1); 1922 } else { 1923 tdnp->n_attrstamp = 0; 1924 mtx_unlock(&tdnp->n_mtx); 1925 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 1926 } 1927 if (error && NFS_ISV4(fdvp)) 1928 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1929 return (error); 1930 } 1931 1932 /* 1933 * nfs hard link create call 1934 */ 1935 static int 1936 nfs_link(struct vop_link_args *ap) 1937 { 1938 struct vnode *vp = ap->a_vp; 1939 struct vnode *tdvp = ap->a_tdvp; 1940 struct componentname *cnp = ap->a_cnp; 1941 struct nfsnode *np, *tdnp; 1942 struct nfsvattr nfsva, dnfsva; 1943 int error = 0, attrflag, dattrflag; 1944 1945 /* 1946 * Push all writes to the server, so that the attribute cache 1947 * doesn't get "out of sync" with the server. 1948 * XXX There should be a better way! 1949 */ 1950 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread); 1951 1952 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 1953 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag, 1954 &dattrflag, NULL); 1955 tdnp = VTONFS(tdvp); 1956 mtx_lock(&tdnp->n_mtx); 1957 tdnp->n_flag |= NMODIFIED; 1958 if (dattrflag != 0) { 1959 mtx_unlock(&tdnp->n_mtx); 1960 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1); 1961 } else { 1962 tdnp->n_attrstamp = 0; 1963 mtx_unlock(&tdnp->n_mtx); 1964 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 1965 } 1966 if (attrflag) 1967 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1968 else { 1969 np = VTONFS(vp); 1970 mtx_lock(&np->n_mtx); 1971 np->n_attrstamp = 0; 1972 mtx_unlock(&np->n_mtx); 1973 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 1974 } 1975 /* 1976 * If negative lookup caching is enabled, I might as well 1977 * add an entry for this node. Not necessary for correctness, 1978 * but if negative caching is enabled, then the system 1979 * must care about lookup caching hit rate, so... 1980 */ 1981 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 && 1982 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 1983 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL); 1984 } 1985 if (error && NFS_ISV4(vp)) 1986 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, 1987 (gid_t)0); 1988 return (error); 1989 } 1990 1991 /* 1992 * nfs symbolic link create call 1993 */ 1994 static int 1995 nfs_symlink(struct vop_symlink_args *ap) 1996 { 1997 struct vnode *dvp = ap->a_dvp; 1998 struct vattr *vap = ap->a_vap; 1999 struct componentname *cnp = ap->a_cnp; 2000 struct nfsvattr nfsva, dnfsva; 2001 struct nfsfh *nfhp; 2002 struct nfsnode *np = NULL, *dnp; 2003 struct vnode *newvp = NULL; 2004 int error = 0, attrflag, dattrflag, ret; 2005 2006 vap->va_type = VLNK; 2007 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2008 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, 2009 &nfsva, &nfhp, &attrflag, &dattrflag, NULL); 2010 if (nfhp) { 2011 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, 2012 &np, NULL, LK_EXCLUSIVE); 2013 if (!ret) 2014 newvp = NFSTOV(np); 2015 else if (!error) 2016 error = ret; 2017 } 2018 if (newvp != NULL) { 2019 if (attrflag) 2020 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 2021 0, 1); 2022 } else if (!error) { 2023 /* 2024 * If we do not have an error and we could not extract the 2025 * newvp from the response due to the request being NFSv2, we 2026 * have to do a lookup in order to obtain a newvp to return. 2027 */ 2028 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2029 cnp->cn_cred, cnp->cn_thread, &np); 2030 if (!error) 2031 newvp = NFSTOV(np); 2032 } 2033 if (error) { 2034 if (newvp) 2035 vput(newvp); 2036 if (NFS_ISV4(dvp)) 2037 error = nfscl_maperr(cnp->cn_thread, error, 2038 vap->va_uid, vap->va_gid); 2039 } else { 2040 *ap->a_vpp = newvp; 2041 } 2042 2043 dnp = VTONFS(dvp); 2044 mtx_lock(&dnp->n_mtx); 2045 dnp->n_flag |= NMODIFIED; 2046 if (dattrflag != 0) { 2047 mtx_unlock(&dnp->n_mtx); 2048 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2049 } else { 2050 dnp->n_attrstamp = 0; 2051 mtx_unlock(&dnp->n_mtx); 2052 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2053 } 2054 /* 2055 * If negative lookup caching is enabled, I might as well 2056 * add an entry for this node. Not necessary for correctness, 2057 * but if negative caching is enabled, then the system 2058 * must care about lookup caching hit rate, so... 2059 */ 2060 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2061 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 2062 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL); 2063 } 2064 return (error); 2065 } 2066 2067 /* 2068 * nfs make dir call 2069 */ 2070 static int 2071 nfs_mkdir(struct vop_mkdir_args *ap) 2072 { 2073 struct vnode *dvp = ap->a_dvp; 2074 struct vattr *vap = ap->a_vap; 2075 struct componentname *cnp = ap->a_cnp; 2076 struct nfsnode *np = NULL, *dnp; 2077 struct vnode *newvp = NULL; 2078 struct vattr vattr; 2079 struct nfsfh *nfhp; 2080 struct nfsvattr nfsva, dnfsva; 2081 int error = 0, attrflag, dattrflag, ret; 2082 2083 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) 2084 return (error); 2085 vap->va_type = VDIR; 2086 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2087 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp, 2088 &attrflag, &dattrflag, NULL); 2089 dnp = VTONFS(dvp); 2090 mtx_lock(&dnp->n_mtx); 2091 dnp->n_flag |= NMODIFIED; 2092 if (dattrflag != 0) { 2093 mtx_unlock(&dnp->n_mtx); 2094 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2095 } else { 2096 dnp->n_attrstamp = 0; 2097 mtx_unlock(&dnp->n_mtx); 2098 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2099 } 2100 if (nfhp) { 2101 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, 2102 &np, NULL, LK_EXCLUSIVE); 2103 if (!ret) { 2104 newvp = NFSTOV(np); 2105 if (attrflag) 2106 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 2107 NULL, 0, 1); 2108 } else if (!error) 2109 error = ret; 2110 } 2111 if (!error && newvp == NULL) { 2112 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2113 cnp->cn_cred, cnp->cn_thread, &np); 2114 if (!error) { 2115 newvp = NFSTOV(np); 2116 if (newvp->v_type != VDIR) 2117 error = EEXIST; 2118 } 2119 } 2120 if (error) { 2121 if (newvp) 2122 vput(newvp); 2123 if (NFS_ISV4(dvp)) 2124 error = nfscl_maperr(cnp->cn_thread, error, 2125 vap->va_uid, vap->va_gid); 2126 } else { 2127 /* 2128 * If negative lookup caching is enabled, I might as well 2129 * add an entry for this node. Not necessary for correctness, 2130 * but if negative caching is enabled, then the system 2131 * must care about lookup caching hit rate, so... 2132 */ 2133 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2134 (cnp->cn_flags & MAKEENTRY) && 2135 attrflag != 0 && dattrflag != 0) 2136 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 2137 &dnfsva.na_ctime); 2138 *ap->a_vpp = newvp; 2139 } 2140 return (error); 2141 } 2142 2143 /* 2144 * nfs remove directory call 2145 */ 2146 static int 2147 nfs_rmdir(struct vop_rmdir_args *ap) 2148 { 2149 struct vnode *vp = ap->a_vp; 2150 struct vnode *dvp = ap->a_dvp; 2151 struct componentname *cnp = ap->a_cnp; 2152 struct nfsnode *dnp; 2153 struct nfsvattr dnfsva; 2154 int error, dattrflag; 2155 2156 if (dvp == vp) 2157 return (EINVAL); 2158 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2159 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL); 2160 dnp = VTONFS(dvp); 2161 mtx_lock(&dnp->n_mtx); 2162 dnp->n_flag |= NMODIFIED; 2163 if (dattrflag != 0) { 2164 mtx_unlock(&dnp->n_mtx); 2165 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2166 } else { 2167 dnp->n_attrstamp = 0; 2168 mtx_unlock(&dnp->n_mtx); 2169 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2170 } 2171 2172 cache_purge(dvp); 2173 cache_purge(vp); 2174 if (error && NFS_ISV4(dvp)) 2175 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, 2176 (gid_t)0); 2177 /* 2178 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2179 */ 2180 if (error == ENOENT) 2181 error = 0; 2182 return (error); 2183 } 2184 2185 /* 2186 * nfs readdir call 2187 */ 2188 static int 2189 nfs_readdir(struct vop_readdir_args *ap) 2190 { 2191 struct vnode *vp = ap->a_vp; 2192 struct nfsnode *np = VTONFS(vp); 2193 struct uio *uio = ap->a_uio; 2194 ssize_t tresid, left; 2195 int error = 0; 2196 struct vattr vattr; 2197 2198 if (ap->a_eofflag != NULL) 2199 *ap->a_eofflag = 0; 2200 if (vp->v_type != VDIR) 2201 return(EPERM); 2202 2203 /* 2204 * First, check for hit on the EOF offset cache 2205 */ 2206 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2207 (np->n_flag & NMODIFIED) == 0) { 2208 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) { 2209 mtx_lock(&np->n_mtx); 2210 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) || 2211 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 2212 mtx_unlock(&np->n_mtx); 2213 NFSINCRGLOBAL(nfsstatsv1.direofcache_hits); 2214 if (ap->a_eofflag != NULL) 2215 *ap->a_eofflag = 1; 2216 return (0); 2217 } else 2218 mtx_unlock(&np->n_mtx); 2219 } 2220 } 2221 2222 /* 2223 * NFS always guarantees that directory entries don't straddle 2224 * DIRBLKSIZ boundaries. As such, we need to limit the size 2225 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial 2226 * directory entry. 2227 */ 2228 left = uio->uio_resid % DIRBLKSIZ; 2229 if (left == uio->uio_resid) 2230 return (EINVAL); 2231 uio->uio_resid -= left; 2232 2233 /* 2234 * Call ncl_bioread() to do the real work. 2235 */ 2236 tresid = uio->uio_resid; 2237 error = ncl_bioread(vp, uio, 0, ap->a_cred); 2238 2239 if (!error && uio->uio_resid == tresid) { 2240 NFSINCRGLOBAL(nfsstatsv1.direofcache_misses); 2241 if (ap->a_eofflag != NULL) 2242 *ap->a_eofflag = 1; 2243 } 2244 2245 /* Add the partial DIRBLKSIZ (left) back in. */ 2246 uio->uio_resid += left; 2247 return (error); 2248 } 2249 2250 /* 2251 * Readdir rpc call. 2252 * Called from below the buffer cache by ncl_doio(). 2253 */ 2254 int 2255 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2256 struct thread *td) 2257 { 2258 struct nfsvattr nfsva; 2259 nfsuint64 *cookiep, cookie; 2260 struct nfsnode *dnp = VTONFS(vp); 2261 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2262 int error = 0, eof, attrflag; 2263 2264 KASSERT(uiop->uio_iovcnt == 1 && 2265 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2266 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2267 ("nfs readdirrpc bad uio")); 2268 2269 /* 2270 * If there is no cookie, assume directory was stale. 2271 */ 2272 ncl_dircookie_lock(dnp); 2273 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2274 if (cookiep) { 2275 cookie = *cookiep; 2276 ncl_dircookie_unlock(dnp); 2277 } else { 2278 ncl_dircookie_unlock(dnp); 2279 return (NFSERR_BAD_COOKIE); 2280 } 2281 2282 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2283 (void)ncl_fsinfo(nmp, vp, cred, td); 2284 2285 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva, 2286 &attrflag, &eof, NULL); 2287 if (attrflag) 2288 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2289 2290 if (!error) { 2291 /* 2292 * We are now either at the end of the directory or have filled 2293 * the block. 2294 */ 2295 if (eof) 2296 dnp->n_direofoffset = uiop->uio_offset; 2297 else { 2298 if (uiop->uio_resid > 0) 2299 printf("EEK! readdirrpc resid > 0\n"); 2300 ncl_dircookie_lock(dnp); 2301 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2302 *cookiep = cookie; 2303 ncl_dircookie_unlock(dnp); 2304 } 2305 } else if (NFS_ISV4(vp)) { 2306 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2307 } 2308 return (error); 2309 } 2310 2311 /* 2312 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc(). 2313 */ 2314 int 2315 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2316 struct thread *td) 2317 { 2318 struct nfsvattr nfsva; 2319 nfsuint64 *cookiep, cookie; 2320 struct nfsnode *dnp = VTONFS(vp); 2321 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2322 int error = 0, attrflag, eof; 2323 2324 KASSERT(uiop->uio_iovcnt == 1 && 2325 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2326 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2327 ("nfs readdirplusrpc bad uio")); 2328 2329 /* 2330 * If there is no cookie, assume directory was stale. 2331 */ 2332 ncl_dircookie_lock(dnp); 2333 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2334 if (cookiep) { 2335 cookie = *cookiep; 2336 ncl_dircookie_unlock(dnp); 2337 } else { 2338 ncl_dircookie_unlock(dnp); 2339 return (NFSERR_BAD_COOKIE); 2340 } 2341 2342 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2343 (void)ncl_fsinfo(nmp, vp, cred, td); 2344 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva, 2345 &attrflag, &eof, NULL); 2346 if (attrflag) 2347 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2348 2349 if (!error) { 2350 /* 2351 * We are now either at end of the directory or have filled the 2352 * the block. 2353 */ 2354 if (eof) 2355 dnp->n_direofoffset = uiop->uio_offset; 2356 else { 2357 if (uiop->uio_resid > 0) 2358 printf("EEK! readdirplusrpc resid > 0\n"); 2359 ncl_dircookie_lock(dnp); 2360 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2361 *cookiep = cookie; 2362 ncl_dircookie_unlock(dnp); 2363 } 2364 } else if (NFS_ISV4(vp)) { 2365 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2366 } 2367 return (error); 2368 } 2369 2370 /* 2371 * Silly rename. To make the NFS filesystem that is stateless look a little 2372 * more like the "ufs" a remove of an active vnode is translated to a rename 2373 * to a funny looking filename that is removed by nfs_inactive on the 2374 * nfsnode. There is the potential for another process on a different client 2375 * to create the same funny name between the nfs_lookitup() fails and the 2376 * nfs_rename() completes, but... 2377 */ 2378 static int 2379 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2380 { 2381 struct sillyrename *sp; 2382 struct nfsnode *np; 2383 int error; 2384 short pid; 2385 unsigned int lticks; 2386 2387 cache_purge(dvp); 2388 np = VTONFS(vp); 2389 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir")); 2390 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2391 M_NEWNFSREQ, M_WAITOK); 2392 sp->s_cred = crhold(cnp->cn_cred); 2393 sp->s_dvp = dvp; 2394 VREF(dvp); 2395 2396 /* 2397 * Fudge together a funny name. 2398 * Changing the format of the funny name to accommodate more 2399 * sillynames per directory. 2400 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is 2401 * CPU ticks since boot. 2402 */ 2403 pid = cnp->cn_thread->td_proc->p_pid; 2404 lticks = (unsigned int)ticks; 2405 for ( ; ; ) { 2406 sp->s_namlen = sprintf(sp->s_name, 2407 ".nfs.%08x.%04x4.4", lticks, 2408 pid); 2409 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2410 cnp->cn_thread, NULL)) 2411 break; 2412 lticks++; 2413 } 2414 error = nfs_renameit(dvp, vp, cnp, sp); 2415 if (error) 2416 goto bad; 2417 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2418 cnp->cn_thread, &np); 2419 np->n_sillyrename = sp; 2420 return (0); 2421 bad: 2422 vrele(sp->s_dvp); 2423 crfree(sp->s_cred); 2424 free((caddr_t)sp, M_NEWNFSREQ); 2425 return (error); 2426 } 2427 2428 /* 2429 * Look up a file name and optionally either update the file handle or 2430 * allocate an nfsnode, depending on the value of npp. 2431 * npp == NULL --> just do the lookup 2432 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2433 * handled too 2434 * *npp != NULL --> update the file handle in the vnode 2435 */ 2436 static int 2437 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred, 2438 struct thread *td, struct nfsnode **npp) 2439 { 2440 struct vnode *newvp = NULL, *vp; 2441 struct nfsnode *np, *dnp = VTONFS(dvp); 2442 struct nfsfh *nfhp, *onfhp; 2443 struct nfsvattr nfsva, dnfsva; 2444 struct componentname cn; 2445 int error = 0, attrflag, dattrflag; 2446 u_int hash; 2447 2448 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva, 2449 &nfhp, &attrflag, &dattrflag, NULL); 2450 if (dattrflag) 2451 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2452 if (npp && !error) { 2453 if (*npp != NULL) { 2454 np = *npp; 2455 vp = NFSTOV(np); 2456 /* 2457 * For NFSv4, check to see if it is the same name and 2458 * replace the name, if it is different. 2459 */ 2460 if (np->n_v4 != NULL && nfsva.na_type == VREG && 2461 (np->n_v4->n4_namelen != len || 2462 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) || 2463 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || 2464 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2465 dnp->n_fhp->nfh_len))) { 2466 #ifdef notdef 2467 { char nnn[100]; int nnnl; 2468 nnnl = (len < 100) ? len : 99; 2469 bcopy(name, nnn, nnnl); 2470 nnn[nnnl] = '\0'; 2471 printf("replace=%s\n",nnn); 2472 } 2473 #endif 2474 FREE((caddr_t)np->n_v4, M_NFSV4NODE); 2475 MALLOC(np->n_v4, struct nfsv4node *, 2476 sizeof (struct nfsv4node) + 2477 dnp->n_fhp->nfh_len + len - 1, 2478 M_NFSV4NODE, M_WAITOK); 2479 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; 2480 np->n_v4->n4_namelen = len; 2481 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2482 dnp->n_fhp->nfh_len); 2483 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len); 2484 } 2485 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, 2486 FNV1_32_INIT); 2487 onfhp = np->n_fhp; 2488 /* 2489 * Rehash node for new file handle. 2490 */ 2491 vfs_hash_rehash(vp, hash); 2492 np->n_fhp = nfhp; 2493 if (onfhp != NULL) 2494 FREE((caddr_t)onfhp, M_NFSFH); 2495 newvp = NFSTOV(np); 2496 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) { 2497 FREE((caddr_t)nfhp, M_NFSFH); 2498 VREF(dvp); 2499 newvp = dvp; 2500 } else { 2501 cn.cn_nameptr = name; 2502 cn.cn_namelen = len; 2503 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td, 2504 &np, NULL, LK_EXCLUSIVE); 2505 if (error) 2506 return (error); 2507 newvp = NFSTOV(np); 2508 } 2509 if (!attrflag && *npp == NULL) { 2510 if (newvp == dvp) 2511 vrele(newvp); 2512 else 2513 vput(newvp); 2514 return (ENOENT); 2515 } 2516 if (attrflag) 2517 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 2518 0, 1); 2519 } 2520 if (npp && *npp == NULL) { 2521 if (error) { 2522 if (newvp) { 2523 if (newvp == dvp) 2524 vrele(newvp); 2525 else 2526 vput(newvp); 2527 } 2528 } else 2529 *npp = np; 2530 } 2531 if (error && NFS_ISV4(dvp)) 2532 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2533 return (error); 2534 } 2535 2536 /* 2537 * Nfs Version 3 and 4 commit rpc 2538 */ 2539 int 2540 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred, 2541 struct thread *td) 2542 { 2543 struct nfsvattr nfsva; 2544 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2545 int error, attrflag; 2546 2547 mtx_lock(&nmp->nm_mtx); 2548 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) { 2549 mtx_unlock(&nmp->nm_mtx); 2550 return (0); 2551 } 2552 mtx_unlock(&nmp->nm_mtx); 2553 error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva, 2554 &attrflag, NULL); 2555 if (attrflag != 0) 2556 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 2557 0, 1); 2558 if (error != 0 && NFS_ISV4(vp)) 2559 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2560 return (error); 2561 } 2562 2563 /* 2564 * Strategy routine. 2565 * For async requests when nfsiod(s) are running, queue the request by 2566 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the 2567 * request. 2568 */ 2569 static int 2570 nfs_strategy(struct vop_strategy_args *ap) 2571 { 2572 struct buf *bp; 2573 struct vnode *vp; 2574 struct ucred *cr; 2575 2576 bp = ap->a_bp; 2577 vp = ap->a_vp; 2578 KASSERT(bp->b_vp == vp, ("missing b_getvp")); 2579 KASSERT(!(bp->b_flags & B_DONE), 2580 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); 2581 BUF_ASSERT_HELD(bp); 2582 2583 if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno) 2584 bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize / 2585 DEV_BSIZE); 2586 if (bp->b_iocmd == BIO_READ) 2587 cr = bp->b_rcred; 2588 else 2589 cr = bp->b_wcred; 2590 2591 /* 2592 * If the op is asynchronous and an i/o daemon is waiting 2593 * queue the request, wake it up and wait for completion 2594 * otherwise just do it ourselves. 2595 */ 2596 if ((bp->b_flags & B_ASYNC) == 0 || 2597 ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread)) 2598 (void) ncl_doio(vp, bp, cr, curthread, 1); 2599 return (0); 2600 } 2601 2602 /* 2603 * fsync vnode op. Just call ncl_flush() with commit == 1. 2604 */ 2605 /* ARGSUSED */ 2606 static int 2607 nfs_fsync(struct vop_fsync_args *ap) 2608 { 2609 2610 if (ap->a_vp->v_type != VREG) { 2611 /* 2612 * For NFS, metadata is changed synchronously on the server, 2613 * so there is nothing to flush. Also, ncl_flush() clears 2614 * the NMODIFIED flag and that shouldn't be done here for 2615 * directories. 2616 */ 2617 return (0); 2618 } 2619 return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1, 0)); 2620 } 2621 2622 /* 2623 * Flush all the blocks associated with a vnode. 2624 * Walk through the buffer pool and push any dirty pages 2625 * associated with the vnode. 2626 * If the called_from_renewthread argument is TRUE, it has been called 2627 * from the NFSv4 renew thread and, as such, cannot block indefinitely 2628 * waiting for a buffer write to complete. 2629 */ 2630 int 2631 ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td, 2632 int commit, int called_from_renewthread) 2633 { 2634 struct nfsnode *np = VTONFS(vp); 2635 struct buf *bp; 2636 int i; 2637 struct buf *nbp; 2638 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2639 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos; 2640 int passone = 1, trycnt = 0; 2641 u_quad_t off, endoff, toff; 2642 struct ucred* wcred = NULL; 2643 struct buf **bvec = NULL; 2644 struct bufobj *bo; 2645 #ifndef NFS_COMMITBVECSIZ 2646 #define NFS_COMMITBVECSIZ 20 2647 #endif 2648 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ]; 2649 int bvecsize = 0, bveccount; 2650 2651 if (called_from_renewthread != 0) 2652 slptimeo = hz; 2653 if (nmp->nm_flag & NFSMNT_INT) 2654 slpflag = PCATCH; 2655 if (!commit) 2656 passone = 0; 2657 bo = &vp->v_bufobj; 2658 /* 2659 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the 2660 * server, but has not been committed to stable storage on the server 2661 * yet. On the first pass, the byte range is worked out and the commit 2662 * rpc is done. On the second pass, ncl_writebp() is called to do the 2663 * job. 2664 */ 2665 again: 2666 off = (u_quad_t)-1; 2667 endoff = 0; 2668 bvecpos = 0; 2669 if (NFS_ISV34(vp) && commit) { 2670 if (bvec != NULL && bvec != bvec_on_stack) 2671 free(bvec, M_TEMP); 2672 /* 2673 * Count up how many buffers waiting for a commit. 2674 */ 2675 bveccount = 0; 2676 BO_LOCK(bo); 2677 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2678 if (!BUF_ISLOCKED(bp) && 2679 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) 2680 == (B_DELWRI | B_NEEDCOMMIT)) 2681 bveccount++; 2682 } 2683 /* 2684 * Allocate space to remember the list of bufs to commit. It is 2685 * important to use M_NOWAIT here to avoid a race with nfs_write. 2686 * If we can't get memory (for whatever reason), we will end up 2687 * committing the buffers one-by-one in the loop below. 2688 */ 2689 if (bveccount > NFS_COMMITBVECSIZ) { 2690 /* 2691 * Release the vnode interlock to avoid a lock 2692 * order reversal. 2693 */ 2694 BO_UNLOCK(bo); 2695 bvec = (struct buf **) 2696 malloc(bveccount * sizeof(struct buf *), 2697 M_TEMP, M_NOWAIT); 2698 BO_LOCK(bo); 2699 if (bvec == NULL) { 2700 bvec = bvec_on_stack; 2701 bvecsize = NFS_COMMITBVECSIZ; 2702 } else 2703 bvecsize = bveccount; 2704 } else { 2705 bvec = bvec_on_stack; 2706 bvecsize = NFS_COMMITBVECSIZ; 2707 } 2708 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2709 if (bvecpos >= bvecsize) 2710 break; 2711 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2712 nbp = TAILQ_NEXT(bp, b_bobufs); 2713 continue; 2714 } 2715 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 2716 (B_DELWRI | B_NEEDCOMMIT)) { 2717 BUF_UNLOCK(bp); 2718 nbp = TAILQ_NEXT(bp, b_bobufs); 2719 continue; 2720 } 2721 BO_UNLOCK(bo); 2722 bremfree(bp); 2723 /* 2724 * Work out if all buffers are using the same cred 2725 * so we can deal with them all with one commit. 2726 * 2727 * NOTE: we are not clearing B_DONE here, so we have 2728 * to do it later on in this routine if we intend to 2729 * initiate I/O on the bp. 2730 * 2731 * Note: to avoid loopback deadlocks, we do not 2732 * assign b_runningbufspace. 2733 */ 2734 if (wcred == NULL) 2735 wcred = bp->b_wcred; 2736 else if (wcred != bp->b_wcred) 2737 wcred = NOCRED; 2738 vfs_busy_pages(bp, 1); 2739 2740 BO_LOCK(bo); 2741 /* 2742 * bp is protected by being locked, but nbp is not 2743 * and vfs_busy_pages() may sleep. We have to 2744 * recalculate nbp. 2745 */ 2746 nbp = TAILQ_NEXT(bp, b_bobufs); 2747 2748 /* 2749 * A list of these buffers is kept so that the 2750 * second loop knows which buffers have actually 2751 * been committed. This is necessary, since there 2752 * may be a race between the commit rpc and new 2753 * uncommitted writes on the file. 2754 */ 2755 bvec[bvecpos++] = bp; 2756 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2757 bp->b_dirtyoff; 2758 if (toff < off) 2759 off = toff; 2760 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); 2761 if (toff > endoff) 2762 endoff = toff; 2763 } 2764 BO_UNLOCK(bo); 2765 } 2766 if (bvecpos > 0) { 2767 /* 2768 * Commit data on the server, as required. 2769 * If all bufs are using the same wcred, then use that with 2770 * one call for all of them, otherwise commit each one 2771 * separately. 2772 */ 2773 if (wcred != NOCRED) 2774 retv = ncl_commit(vp, off, (int)(endoff - off), 2775 wcred, td); 2776 else { 2777 retv = 0; 2778 for (i = 0; i < bvecpos; i++) { 2779 off_t off, size; 2780 bp = bvec[i]; 2781 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2782 bp->b_dirtyoff; 2783 size = (u_quad_t)(bp->b_dirtyend 2784 - bp->b_dirtyoff); 2785 retv = ncl_commit(vp, off, (int)size, 2786 bp->b_wcred, td); 2787 if (retv) break; 2788 } 2789 } 2790 2791 if (retv == NFSERR_STALEWRITEVERF) 2792 ncl_clearcommit(vp->v_mount); 2793 2794 /* 2795 * Now, either mark the blocks I/O done or mark the 2796 * blocks dirty, depending on whether the commit 2797 * succeeded. 2798 */ 2799 for (i = 0; i < bvecpos; i++) { 2800 bp = bvec[i]; 2801 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 2802 if (retv) { 2803 /* 2804 * Error, leave B_DELWRI intact 2805 */ 2806 vfs_unbusy_pages(bp); 2807 brelse(bp); 2808 } else { 2809 /* 2810 * Success, remove B_DELWRI ( bundirty() ). 2811 * 2812 * b_dirtyoff/b_dirtyend seem to be NFS 2813 * specific. We should probably move that 2814 * into bundirty(). XXX 2815 */ 2816 bufobj_wref(bo); 2817 bp->b_flags |= B_ASYNC; 2818 bundirty(bp); 2819 bp->b_flags &= ~B_DONE; 2820 bp->b_ioflags &= ~BIO_ERROR; 2821 bp->b_dirtyoff = bp->b_dirtyend = 0; 2822 bufdone(bp); 2823 } 2824 } 2825 } 2826 2827 /* 2828 * Start/do any write(s) that are required. 2829 */ 2830 loop: 2831 BO_LOCK(bo); 2832 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2833 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2834 if (waitfor != MNT_WAIT || passone) 2835 continue; 2836 2837 error = BUF_TIMELOCK(bp, 2838 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 2839 BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo); 2840 if (error == 0) { 2841 BUF_UNLOCK(bp); 2842 goto loop; 2843 } 2844 if (error == ENOLCK) { 2845 error = 0; 2846 goto loop; 2847 } 2848 if (called_from_renewthread != 0) { 2849 /* 2850 * Return EIO so the flush will be retried 2851 * later. 2852 */ 2853 error = EIO; 2854 goto done; 2855 } 2856 if (newnfs_sigintr(nmp, td)) { 2857 error = EINTR; 2858 goto done; 2859 } 2860 if (slpflag == PCATCH) { 2861 slpflag = 0; 2862 slptimeo = 2 * hz; 2863 } 2864 goto loop; 2865 } 2866 if ((bp->b_flags & B_DELWRI) == 0) 2867 panic("nfs_fsync: not dirty"); 2868 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) { 2869 BUF_UNLOCK(bp); 2870 continue; 2871 } 2872 BO_UNLOCK(bo); 2873 bremfree(bp); 2874 if (passone || !commit) 2875 bp->b_flags |= B_ASYNC; 2876 else 2877 bp->b_flags |= B_ASYNC; 2878 bwrite(bp); 2879 if (newnfs_sigintr(nmp, td)) { 2880 error = EINTR; 2881 goto done; 2882 } 2883 goto loop; 2884 } 2885 if (passone) { 2886 passone = 0; 2887 BO_UNLOCK(bo); 2888 goto again; 2889 } 2890 if (waitfor == MNT_WAIT) { 2891 while (bo->bo_numoutput) { 2892 error = bufobj_wwait(bo, slpflag, slptimeo); 2893 if (error) { 2894 BO_UNLOCK(bo); 2895 if (called_from_renewthread != 0) { 2896 /* 2897 * Return EIO so that the flush will be 2898 * retried later. 2899 */ 2900 error = EIO; 2901 goto done; 2902 } 2903 error = newnfs_sigintr(nmp, td); 2904 if (error) 2905 goto done; 2906 if (slpflag == PCATCH) { 2907 slpflag = 0; 2908 slptimeo = 2 * hz; 2909 } 2910 BO_LOCK(bo); 2911 } 2912 } 2913 if (bo->bo_dirty.bv_cnt != 0 && commit) { 2914 BO_UNLOCK(bo); 2915 goto loop; 2916 } 2917 /* 2918 * Wait for all the async IO requests to drain 2919 */ 2920 BO_UNLOCK(bo); 2921 mtx_lock(&np->n_mtx); 2922 while (np->n_directio_asyncwr > 0) { 2923 np->n_flag |= NFSYNCWAIT; 2924 error = newnfs_msleep(td, &np->n_directio_asyncwr, 2925 &np->n_mtx, slpflag | (PRIBIO + 1), 2926 "nfsfsync", 0); 2927 if (error) { 2928 if (newnfs_sigintr(nmp, td)) { 2929 mtx_unlock(&np->n_mtx); 2930 error = EINTR; 2931 goto done; 2932 } 2933 } 2934 } 2935 mtx_unlock(&np->n_mtx); 2936 } else 2937 BO_UNLOCK(bo); 2938 if (NFSHASPNFS(nmp)) { 2939 nfscl_layoutcommit(vp, td); 2940 /* 2941 * Invalidate the attribute cache, since writes to a DS 2942 * won't update the size attribute. 2943 */ 2944 mtx_lock(&np->n_mtx); 2945 np->n_attrstamp = 0; 2946 } else 2947 mtx_lock(&np->n_mtx); 2948 if (np->n_flag & NWRITEERR) { 2949 error = np->n_error; 2950 np->n_flag &= ~NWRITEERR; 2951 } 2952 if (commit && bo->bo_dirty.bv_cnt == 0 && 2953 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0) 2954 np->n_flag &= ~NMODIFIED; 2955 mtx_unlock(&np->n_mtx); 2956 done: 2957 if (bvec != NULL && bvec != bvec_on_stack) 2958 free(bvec, M_TEMP); 2959 if (error == 0 && commit != 0 && waitfor == MNT_WAIT && 2960 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 || 2961 np->n_directio_asyncwr != 0) && trycnt++ < 5) { 2962 /* try, try again... */ 2963 passone = 1; 2964 wcred = NULL; 2965 bvec = NULL; 2966 bvecsize = 0; 2967 printf("try%d\n", trycnt); 2968 goto again; 2969 } 2970 return (error); 2971 } 2972 2973 /* 2974 * NFS advisory byte-level locks. 2975 */ 2976 static int 2977 nfs_advlock(struct vop_advlock_args *ap) 2978 { 2979 struct vnode *vp = ap->a_vp; 2980 struct ucred *cred; 2981 struct nfsnode *np = VTONFS(ap->a_vp); 2982 struct proc *p = (struct proc *)ap->a_id; 2983 struct thread *td = curthread; /* XXX */ 2984 struct vattr va; 2985 int ret, error = EOPNOTSUPP; 2986 u_quad_t size; 2987 2988 if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) { 2989 if (vp->v_type != VREG) 2990 return (EINVAL); 2991 if ((ap->a_flags & F_POSIX) != 0) 2992 cred = p->p_ucred; 2993 else 2994 cred = td->td_ucred; 2995 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 2996 if (vp->v_iflag & VI_DOOMED) { 2997 NFSVOPUNLOCK(vp, 0); 2998 return (EBADF); 2999 } 3000 3001 /* 3002 * If this is unlocking a write locked region, flush and 3003 * commit them before unlocking. This is required by 3004 * RFC3530 Sec. 9.3.2. 3005 */ 3006 if (ap->a_op == F_UNLCK && 3007 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id, 3008 ap->a_flags)) 3009 (void) ncl_flush(vp, MNT_WAIT, cred, td, 1, 0); 3010 3011 /* 3012 * Loop around doing the lock op, while a blocking lock 3013 * must wait for the lock op to succeed. 3014 */ 3015 do { 3016 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op, 3017 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags); 3018 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3019 ap->a_op == F_SETLK) { 3020 NFSVOPUNLOCK(vp, 0); 3021 error = nfs_catnap(PZERO | PCATCH, ret, 3022 "ncladvl"); 3023 if (error) 3024 return (EINTR); 3025 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 3026 if (vp->v_iflag & VI_DOOMED) { 3027 NFSVOPUNLOCK(vp, 0); 3028 return (EBADF); 3029 } 3030 } 3031 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3032 ap->a_op == F_SETLK); 3033 if (ret == NFSERR_DENIED) { 3034 NFSVOPUNLOCK(vp, 0); 3035 return (EAGAIN); 3036 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) { 3037 NFSVOPUNLOCK(vp, 0); 3038 return (ret); 3039 } else if (ret != 0) { 3040 NFSVOPUNLOCK(vp, 0); 3041 return (EACCES); 3042 } 3043 3044 /* 3045 * Now, if we just got a lock, invalidate data in the buffer 3046 * cache, as required, so that the coherency conforms with 3047 * RFC3530 Sec. 9.3.2. 3048 */ 3049 if (ap->a_op == F_SETLK) { 3050 if ((np->n_flag & NMODIFIED) == 0) { 3051 np->n_attrstamp = 0; 3052 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3053 ret = VOP_GETATTR(vp, &va, cred); 3054 } 3055 if ((np->n_flag & NMODIFIED) || ret || 3056 np->n_change != va.va_filerev) { 3057 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1); 3058 np->n_attrstamp = 0; 3059 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3060 ret = VOP_GETATTR(vp, &va, cred); 3061 if (!ret) { 3062 np->n_mtime = va.va_mtime; 3063 np->n_change = va.va_filerev; 3064 } 3065 } 3066 /* Mark that a file lock has been acquired. */ 3067 mtx_lock(&np->n_mtx); 3068 np->n_flag |= NHASBEENLOCKED; 3069 mtx_unlock(&np->n_mtx); 3070 } 3071 NFSVOPUNLOCK(vp, 0); 3072 return (0); 3073 } else if (!NFS_ISV4(vp)) { 3074 error = NFSVOPLOCK(vp, LK_SHARED); 3075 if (error) 3076 return (error); 3077 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3078 size = VTONFS(vp)->n_size; 3079 NFSVOPUNLOCK(vp, 0); 3080 error = lf_advlock(ap, &(vp->v_lockf), size); 3081 } else { 3082 if (nfs_advlock_p != NULL) 3083 error = nfs_advlock_p(ap); 3084 else { 3085 NFSVOPUNLOCK(vp, 0); 3086 error = ENOLCK; 3087 } 3088 } 3089 if (error == 0 && ap->a_op == F_SETLK) { 3090 error = NFSVOPLOCK(vp, LK_SHARED); 3091 if (error == 0) { 3092 /* Mark that a file lock has been acquired. */ 3093 mtx_lock(&np->n_mtx); 3094 np->n_flag |= NHASBEENLOCKED; 3095 mtx_unlock(&np->n_mtx); 3096 NFSVOPUNLOCK(vp, 0); 3097 } 3098 } 3099 } 3100 return (error); 3101 } 3102 3103 /* 3104 * NFS advisory byte-level locks. 3105 */ 3106 static int 3107 nfs_advlockasync(struct vop_advlockasync_args *ap) 3108 { 3109 struct vnode *vp = ap->a_vp; 3110 u_quad_t size; 3111 int error; 3112 3113 if (NFS_ISV4(vp)) 3114 return (EOPNOTSUPP); 3115 error = NFSVOPLOCK(vp, LK_SHARED); 3116 if (error) 3117 return (error); 3118 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3119 size = VTONFS(vp)->n_size; 3120 NFSVOPUNLOCK(vp, 0); 3121 error = lf_advlockasync(ap, &(vp->v_lockf), size); 3122 } else { 3123 NFSVOPUNLOCK(vp, 0); 3124 error = EOPNOTSUPP; 3125 } 3126 return (error); 3127 } 3128 3129 /* 3130 * Print out the contents of an nfsnode. 3131 */ 3132 static int 3133 nfs_print(struct vop_print_args *ap) 3134 { 3135 struct vnode *vp = ap->a_vp; 3136 struct nfsnode *np = VTONFS(vp); 3137 3138 printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid, 3139 (uintmax_t)np->n_vattr.na_fsid); 3140 if (vp->v_type == VFIFO) 3141 fifo_printinfo(vp); 3142 printf("\n"); 3143 return (0); 3144 } 3145 3146 /* 3147 * This is the "real" nfs::bwrite(struct buf*). 3148 * We set B_CACHE if this is a VMIO buffer. 3149 */ 3150 int 3151 ncl_writebp(struct buf *bp, int force __unused, struct thread *td) 3152 { 3153 int s; 3154 int oldflags = bp->b_flags; 3155 #if 0 3156 int retv = 1; 3157 off_t off; 3158 #endif 3159 3160 BUF_ASSERT_HELD(bp); 3161 3162 if (bp->b_flags & B_INVAL) { 3163 brelse(bp); 3164 return(0); 3165 } 3166 3167 bp->b_flags |= B_CACHE; 3168 3169 /* 3170 * Undirty the bp. We will redirty it later if the I/O fails. 3171 */ 3172 3173 s = splbio(); 3174 bundirty(bp); 3175 bp->b_flags &= ~B_DONE; 3176 bp->b_ioflags &= ~BIO_ERROR; 3177 bp->b_iocmd = BIO_WRITE; 3178 3179 bufobj_wref(bp->b_bufobj); 3180 curthread->td_ru.ru_oublock++; 3181 splx(s); 3182 3183 /* 3184 * Note: to avoid loopback deadlocks, we do not 3185 * assign b_runningbufspace. 3186 */ 3187 vfs_busy_pages(bp, 1); 3188 3189 BUF_KERNPROC(bp); 3190 bp->b_iooffset = dbtob(bp->b_blkno); 3191 bstrategy(bp); 3192 3193 if( (oldflags & B_ASYNC) == 0) { 3194 int rtval = bufwait(bp); 3195 3196 if (oldflags & B_DELWRI) { 3197 s = splbio(); 3198 reassignbuf(bp); 3199 splx(s); 3200 } 3201 brelse(bp); 3202 return (rtval); 3203 } 3204 3205 return (0); 3206 } 3207 3208 /* 3209 * nfs special file access vnode op. 3210 * Essentially just get vattr and then imitate iaccess() since the device is 3211 * local to the client. 3212 */ 3213 static int 3214 nfsspec_access(struct vop_access_args *ap) 3215 { 3216 struct vattr *vap; 3217 struct ucred *cred = ap->a_cred; 3218 struct vnode *vp = ap->a_vp; 3219 accmode_t accmode = ap->a_accmode; 3220 struct vattr vattr; 3221 int error; 3222 3223 /* 3224 * Disallow write attempts on filesystems mounted read-only; 3225 * unless the file is a socket, fifo, or a block or character 3226 * device resident on the filesystem. 3227 */ 3228 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3229 switch (vp->v_type) { 3230 case VREG: 3231 case VDIR: 3232 case VLNK: 3233 return (EROFS); 3234 default: 3235 break; 3236 } 3237 } 3238 vap = &vattr; 3239 error = VOP_GETATTR(vp, vap, cred); 3240 if (error) 3241 goto out; 3242 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid, 3243 accmode, cred, NULL); 3244 out: 3245 return error; 3246 } 3247 3248 /* 3249 * Read wrapper for fifos. 3250 */ 3251 static int 3252 nfsfifo_read(struct vop_read_args *ap) 3253 { 3254 struct nfsnode *np = VTONFS(ap->a_vp); 3255 int error; 3256 3257 /* 3258 * Set access flag. 3259 */ 3260 mtx_lock(&np->n_mtx); 3261 np->n_flag |= NACC; 3262 vfs_timestamp(&np->n_atim); 3263 mtx_unlock(&np->n_mtx); 3264 error = fifo_specops.vop_read(ap); 3265 return error; 3266 } 3267 3268 /* 3269 * Write wrapper for fifos. 3270 */ 3271 static int 3272 nfsfifo_write(struct vop_write_args *ap) 3273 { 3274 struct nfsnode *np = VTONFS(ap->a_vp); 3275 3276 /* 3277 * Set update flag. 3278 */ 3279 mtx_lock(&np->n_mtx); 3280 np->n_flag |= NUPD; 3281 vfs_timestamp(&np->n_mtim); 3282 mtx_unlock(&np->n_mtx); 3283 return(fifo_specops.vop_write(ap)); 3284 } 3285 3286 /* 3287 * Close wrapper for fifos. 3288 * 3289 * Update the times on the nfsnode then do fifo close. 3290 */ 3291 static int 3292 nfsfifo_close(struct vop_close_args *ap) 3293 { 3294 struct vnode *vp = ap->a_vp; 3295 struct nfsnode *np = VTONFS(vp); 3296 struct vattr vattr; 3297 struct timespec ts; 3298 3299 mtx_lock(&np->n_mtx); 3300 if (np->n_flag & (NACC | NUPD)) { 3301 vfs_timestamp(&ts); 3302 if (np->n_flag & NACC) 3303 np->n_atim = ts; 3304 if (np->n_flag & NUPD) 3305 np->n_mtim = ts; 3306 np->n_flag |= NCHG; 3307 if (vrefcnt(vp) == 1 && 3308 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3309 VATTR_NULL(&vattr); 3310 if (np->n_flag & NACC) 3311 vattr.va_atime = np->n_atim; 3312 if (np->n_flag & NUPD) 3313 vattr.va_mtime = np->n_mtim; 3314 mtx_unlock(&np->n_mtx); 3315 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3316 goto out; 3317 } 3318 } 3319 mtx_unlock(&np->n_mtx); 3320 out: 3321 return (fifo_specops.vop_close(ap)); 3322 } 3323 3324 /* 3325 * Just call ncl_writebp() with the force argument set to 1. 3326 * 3327 * NOTE: B_DONE may or may not be set in a_bp on call. 3328 */ 3329 static int 3330 nfs_bwrite(struct buf *bp) 3331 { 3332 3333 return (ncl_writebp(bp, 1, curthread)); 3334 } 3335 3336 struct buf_ops buf_ops_newnfs = { 3337 .bop_name = "buf_ops_nfs", 3338 .bop_write = nfs_bwrite, 3339 .bop_strategy = bufstrategy, 3340 .bop_sync = bufsync, 3341 .bop_bdflush = bufbdflush, 3342 }; 3343 3344 static int 3345 nfs_getacl(struct vop_getacl_args *ap) 3346 { 3347 int error; 3348 3349 if (ap->a_type != ACL_TYPE_NFS4) 3350 return (EOPNOTSUPP); 3351 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3352 NULL); 3353 if (error > NFSERR_STALE) { 3354 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3355 error = EPERM; 3356 } 3357 return (error); 3358 } 3359 3360 static int 3361 nfs_setacl(struct vop_setacl_args *ap) 3362 { 3363 int error; 3364 3365 if (ap->a_type != ACL_TYPE_NFS4) 3366 return (EOPNOTSUPP); 3367 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3368 NULL); 3369 if (error > NFSERR_STALE) { 3370 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3371 error = EPERM; 3372 } 3373 return (error); 3374 } 3375 3376 /* 3377 * Return POSIX pathconf information applicable to nfs filesystems. 3378 */ 3379 static int 3380 nfs_pathconf(struct vop_pathconf_args *ap) 3381 { 3382 struct nfsv3_pathconf pc; 3383 struct nfsvattr nfsva; 3384 struct vnode *vp = ap->a_vp; 3385 struct thread *td = curthread; 3386 int attrflag, error; 3387 3388 if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX || 3389 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED || 3390 ap->a_name == _PC_NO_TRUNC)) || 3391 (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) { 3392 /* 3393 * Since only the above 4 a_names are returned by the NFSv3 3394 * Pathconf RPC, there is no point in doing it for others. 3395 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can 3396 * be used for _PC_NFS4_ACL as well. 3397 */ 3398 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva, 3399 &attrflag, NULL); 3400 if (attrflag != 0) 3401 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 3402 1); 3403 if (error != 0) 3404 return (error); 3405 } else { 3406 /* 3407 * For NFSv2 (or NFSv3 when not one of the above 4 a_names), 3408 * just fake them. 3409 */ 3410 pc.pc_linkmax = LINK_MAX; 3411 pc.pc_namemax = NFS_MAXNAMLEN; 3412 pc.pc_notrunc = 1; 3413 pc.pc_chownrestricted = 1; 3414 pc.pc_caseinsensitive = 0; 3415 pc.pc_casepreserving = 1; 3416 error = 0; 3417 } 3418 switch (ap->a_name) { 3419 case _PC_LINK_MAX: 3420 *ap->a_retval = pc.pc_linkmax; 3421 break; 3422 case _PC_NAME_MAX: 3423 *ap->a_retval = pc.pc_namemax; 3424 break; 3425 case _PC_PATH_MAX: 3426 *ap->a_retval = PATH_MAX; 3427 break; 3428 case _PC_PIPE_BUF: 3429 *ap->a_retval = PIPE_BUF; 3430 break; 3431 case _PC_CHOWN_RESTRICTED: 3432 *ap->a_retval = pc.pc_chownrestricted; 3433 break; 3434 case _PC_NO_TRUNC: 3435 *ap->a_retval = pc.pc_notrunc; 3436 break; 3437 case _PC_ACL_EXTENDED: 3438 *ap->a_retval = 0; 3439 break; 3440 case _PC_ACL_NFS4: 3441 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 && 3442 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL)) 3443 *ap->a_retval = 1; 3444 else 3445 *ap->a_retval = 0; 3446 break; 3447 case _PC_ACL_PATH_MAX: 3448 if (NFS_ISV4(vp)) 3449 *ap->a_retval = ACL_MAX_ENTRIES; 3450 else 3451 *ap->a_retval = 3; 3452 break; 3453 case _PC_MAC_PRESENT: 3454 *ap->a_retval = 0; 3455 break; 3456 case _PC_ASYNC_IO: 3457 /* _PC_ASYNC_IO should have been handled by upper layers. */ 3458 KASSERT(0, ("_PC_ASYNC_IO should not get here")); 3459 error = EINVAL; 3460 break; 3461 case _PC_PRIO_IO: 3462 *ap->a_retval = 0; 3463 break; 3464 case _PC_SYNC_IO: 3465 *ap->a_retval = 0; 3466 break; 3467 case _PC_ALLOC_SIZE_MIN: 3468 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize; 3469 break; 3470 case _PC_FILESIZEBITS: 3471 if (NFS_ISV34(vp)) 3472 *ap->a_retval = 64; 3473 else 3474 *ap->a_retval = 32; 3475 break; 3476 case _PC_REC_INCR_XFER_SIZE: 3477 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 3478 break; 3479 case _PC_REC_MAX_XFER_SIZE: 3480 *ap->a_retval = -1; /* means ``unlimited'' */ 3481 break; 3482 case _PC_REC_MIN_XFER_SIZE: 3483 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 3484 break; 3485 case _PC_REC_XFER_ALIGN: 3486 *ap->a_retval = PAGE_SIZE; 3487 break; 3488 case _PC_SYMLINK_MAX: 3489 *ap->a_retval = NFS_MAXPATHLEN; 3490 break; 3491 3492 default: 3493 error = EINVAL; 3494 break; 3495 } 3496 return (error); 3497 } 3498 3499