1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * from nfs_vnops.c 8.16 (Berkeley) 5/27/95 35 */ 36 37 #include <sys/cdefs.h> 38 __FBSDID("$FreeBSD$"); 39 40 /* 41 * vnode op calls for Sun NFS version 2, 3 and 4 42 */ 43 44 #include "opt_inet.h" 45 46 #include <sys/param.h> 47 #include <sys/kernel.h> 48 #include <sys/systm.h> 49 #include <sys/resourcevar.h> 50 #include <sys/proc.h> 51 #include <sys/mount.h> 52 #include <sys/bio.h> 53 #include <sys/buf.h> 54 #include <sys/jail.h> 55 #include <sys/malloc.h> 56 #include <sys/mbuf.h> 57 #include <sys/namei.h> 58 #include <sys/socket.h> 59 #include <sys/vnode.h> 60 #include <sys/dirent.h> 61 #include <sys/fcntl.h> 62 #include <sys/lockf.h> 63 #include <sys/stat.h> 64 #include <sys/sysctl.h> 65 #include <sys/signalvar.h> 66 67 #include <vm/vm.h> 68 #include <vm/vm_extern.h> 69 #include <vm/vm_object.h> 70 71 #include <fs/nfs/nfsport.h> 72 #include <fs/nfsclient/nfsnode.h> 73 #include <fs/nfsclient/nfsmount.h> 74 #include <fs/nfsclient/nfs.h> 75 #include <fs/nfsclient/nfs_kdtrace.h> 76 77 #include <net/if.h> 78 #include <netinet/in.h> 79 #include <netinet/in_var.h> 80 81 #include <nfs/nfs_lock.h> 82 83 #ifdef KDTRACE_HOOKS 84 #include <sys/dtrace_bsd.h> 85 86 dtrace_nfsclient_accesscache_flush_probe_func_t 87 dtrace_nfscl_accesscache_flush_done_probe; 88 uint32_t nfscl_accesscache_flush_done_id; 89 90 dtrace_nfsclient_accesscache_get_probe_func_t 91 dtrace_nfscl_accesscache_get_hit_probe, 92 dtrace_nfscl_accesscache_get_miss_probe; 93 uint32_t nfscl_accesscache_get_hit_id; 94 uint32_t nfscl_accesscache_get_miss_id; 95 96 dtrace_nfsclient_accesscache_load_probe_func_t 97 dtrace_nfscl_accesscache_load_done_probe; 98 uint32_t nfscl_accesscache_load_done_id; 99 #endif /* !KDTRACE_HOOKS */ 100 101 /* Defs */ 102 #define TRUE 1 103 #define FALSE 0 104 105 extern struct nfsstatsv1 nfsstatsv1; 106 extern int nfsrv_useacl; 107 extern int nfscl_debuglevel; 108 MALLOC_DECLARE(M_NEWNFSREQ); 109 110 static vop_read_t nfsfifo_read; 111 static vop_write_t nfsfifo_write; 112 static vop_close_t nfsfifo_close; 113 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *, 114 struct thread *); 115 static vop_lookup_t nfs_lookup; 116 static vop_create_t nfs_create; 117 static vop_mknod_t nfs_mknod; 118 static vop_open_t nfs_open; 119 static vop_pathconf_t nfs_pathconf; 120 static vop_close_t nfs_close; 121 static vop_access_t nfs_access; 122 static vop_getattr_t nfs_getattr; 123 static vop_setattr_t nfs_setattr; 124 static vop_read_t nfs_read; 125 static vop_fsync_t nfs_fsync; 126 static vop_remove_t nfs_remove; 127 static vop_link_t nfs_link; 128 static vop_rename_t nfs_rename; 129 static vop_mkdir_t nfs_mkdir; 130 static vop_rmdir_t nfs_rmdir; 131 static vop_symlink_t nfs_symlink; 132 static vop_readdir_t nfs_readdir; 133 static vop_strategy_t nfs_strategy; 134 static int nfs_lookitup(struct vnode *, char *, int, 135 struct ucred *, struct thread *, struct nfsnode **); 136 static int nfs_sillyrename(struct vnode *, struct vnode *, 137 struct componentname *); 138 static vop_access_t nfsspec_access; 139 static vop_readlink_t nfs_readlink; 140 static vop_print_t nfs_print; 141 static vop_advlock_t nfs_advlock; 142 static vop_advlockasync_t nfs_advlockasync; 143 static vop_getacl_t nfs_getacl; 144 static vop_setacl_t nfs_setacl; 145 146 /* 147 * Global vfs data structures for nfs 148 */ 149 150 static struct vop_vector newnfs_vnodeops_nosig = { 151 .vop_default = &default_vnodeops, 152 .vop_access = nfs_access, 153 .vop_advlock = nfs_advlock, 154 .vop_advlockasync = nfs_advlockasync, 155 .vop_close = nfs_close, 156 .vop_create = nfs_create, 157 .vop_fsync = nfs_fsync, 158 .vop_getattr = nfs_getattr, 159 .vop_getpages = ncl_getpages, 160 .vop_putpages = ncl_putpages, 161 .vop_inactive = ncl_inactive, 162 .vop_link = nfs_link, 163 .vop_lookup = nfs_lookup, 164 .vop_mkdir = nfs_mkdir, 165 .vop_mknod = nfs_mknod, 166 .vop_open = nfs_open, 167 .vop_pathconf = nfs_pathconf, 168 .vop_print = nfs_print, 169 .vop_read = nfs_read, 170 .vop_readdir = nfs_readdir, 171 .vop_readlink = nfs_readlink, 172 .vop_reclaim = ncl_reclaim, 173 .vop_remove = nfs_remove, 174 .vop_rename = nfs_rename, 175 .vop_rmdir = nfs_rmdir, 176 .vop_setattr = nfs_setattr, 177 .vop_strategy = nfs_strategy, 178 .vop_symlink = nfs_symlink, 179 .vop_write = ncl_write, 180 .vop_getacl = nfs_getacl, 181 .vop_setacl = nfs_setacl, 182 }; 183 184 static int 185 nfs_vnodeops_bypass(struct vop_generic_args *a) 186 { 187 188 return (vop_sigdefer(&newnfs_vnodeops_nosig, a)); 189 } 190 191 struct vop_vector newnfs_vnodeops = { 192 .vop_default = &default_vnodeops, 193 .vop_bypass = nfs_vnodeops_bypass, 194 }; 195 196 static struct vop_vector newnfs_fifoops_nosig = { 197 .vop_default = &fifo_specops, 198 .vop_access = nfsspec_access, 199 .vop_close = nfsfifo_close, 200 .vop_fsync = nfs_fsync, 201 .vop_getattr = nfs_getattr, 202 .vop_inactive = ncl_inactive, 203 .vop_pathconf = nfs_pathconf, 204 .vop_print = nfs_print, 205 .vop_read = nfsfifo_read, 206 .vop_reclaim = ncl_reclaim, 207 .vop_setattr = nfs_setattr, 208 .vop_write = nfsfifo_write, 209 }; 210 211 static int 212 nfs_fifoops_bypass(struct vop_generic_args *a) 213 { 214 215 return (vop_sigdefer(&newnfs_fifoops_nosig, a)); 216 } 217 218 struct vop_vector newnfs_fifoops = { 219 .vop_default = &default_vnodeops, 220 .vop_bypass = nfs_fifoops_bypass, 221 }; 222 223 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, 224 struct componentname *cnp, struct vattr *vap); 225 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 226 int namelen, struct ucred *cred, struct thread *td); 227 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, 228 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp, 229 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td); 230 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp, 231 struct componentname *scnp, struct sillyrename *sp); 232 233 /* 234 * Global variables 235 */ 236 SYSCTL_DECL(_vfs_nfs); 237 238 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO; 239 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 240 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 241 242 static int nfs_prime_access_cache = 0; 243 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW, 244 &nfs_prime_access_cache, 0, 245 "Prime NFS ACCESS cache when fetching attributes"); 246 247 static int newnfs_commit_on_close = 0; 248 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW, 249 &newnfs_commit_on_close, 0, "write+commit on close, else only write"); 250 251 static int nfs_clean_pages_on_close = 1; 252 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW, 253 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close"); 254 255 int newnfs_directio_enable = 0; 256 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW, 257 &newnfs_directio_enable, 0, "Enable NFS directio"); 258 259 int nfs_keep_dirty_on_error; 260 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW, 261 &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned"); 262 263 /* 264 * This sysctl allows other processes to mmap a file that has been opened 265 * O_DIRECT by a process. In general, having processes mmap the file while 266 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow 267 * this by default to prevent DoS attacks - to prevent a malicious user from 268 * opening up files O_DIRECT preventing other users from mmap'ing these 269 * files. "Protected" environments where stricter consistency guarantees are 270 * required can disable this knob. The process that opened the file O_DIRECT 271 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not 272 * meaningful. 273 */ 274 int newnfs_directio_allow_mmap = 1; 275 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW, 276 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens"); 277 278 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \ 279 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \ 280 | NFSACCESS_DELETE | NFSACCESS_LOOKUP) 281 282 /* 283 * SMP Locking Note : 284 * The list of locks after the description of the lock is the ordering 285 * of other locks acquired with the lock held. 286 * np->n_mtx : Protects the fields in the nfsnode. 287 VM Object Lock 288 VI_MTX (acquired indirectly) 289 * nmp->nm_mtx : Protects the fields in the nfsmount. 290 rep->r_mtx 291 * ncl_iod_mutex : Global lock, protects shared nfsiod state. 292 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list. 293 nmp->nm_mtx 294 rep->r_mtx 295 * rep->r_mtx : Protects the fields in an nfsreq. 296 */ 297 298 static int 299 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td, 300 struct ucred *cred, u_int32_t *retmode) 301 { 302 int error = 0, attrflag, i, lrupos; 303 u_int32_t rmode; 304 struct nfsnode *np = VTONFS(vp); 305 struct nfsvattr nfsva; 306 307 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag, 308 &rmode, NULL); 309 if (attrflag) 310 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 311 if (!error) { 312 lrupos = 0; 313 mtx_lock(&np->n_mtx); 314 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 315 if (np->n_accesscache[i].uid == cred->cr_uid) { 316 np->n_accesscache[i].mode = rmode; 317 np->n_accesscache[i].stamp = time_second; 318 break; 319 } 320 if (i > 0 && np->n_accesscache[i].stamp < 321 np->n_accesscache[lrupos].stamp) 322 lrupos = i; 323 } 324 if (i == NFS_ACCESSCACHESIZE) { 325 np->n_accesscache[lrupos].uid = cred->cr_uid; 326 np->n_accesscache[lrupos].mode = rmode; 327 np->n_accesscache[lrupos].stamp = time_second; 328 } 329 mtx_unlock(&np->n_mtx); 330 if (retmode != NULL) 331 *retmode = rmode; 332 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0); 333 } else if (NFS_ISV4(vp)) { 334 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 335 } 336 #ifdef KDTRACE_HOOKS 337 if (error != 0) 338 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0, 339 error); 340 #endif 341 return (error); 342 } 343 344 /* 345 * nfs access vnode op. 346 * For nfs version 2, just return ok. File accesses may fail later. 347 * For nfs version 3, use the access rpc to check accessibility. If file modes 348 * are changed on the server, accesses might still fail later. 349 */ 350 static int 351 nfs_access(struct vop_access_args *ap) 352 { 353 struct vnode *vp = ap->a_vp; 354 int error = 0, i, gotahit; 355 u_int32_t mode, wmode, rmode; 356 int v34 = NFS_ISV34(vp); 357 struct nfsnode *np = VTONFS(vp); 358 359 /* 360 * Disallow write attempts on filesystems mounted read-only; 361 * unless the file is a socket, fifo, or a block or character 362 * device resident on the filesystem. 363 */ 364 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS | 365 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL | 366 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) { 367 switch (vp->v_type) { 368 case VREG: 369 case VDIR: 370 case VLNK: 371 return (EROFS); 372 default: 373 break; 374 } 375 } 376 /* 377 * For nfs v3 or v4, check to see if we have done this recently, and if 378 * so return our cached result instead of making an ACCESS call. 379 * If not, do an access rpc, otherwise you are stuck emulating 380 * ufs_access() locally using the vattr. This may not be correct, 381 * since the server may apply other access criteria such as 382 * client uid-->server uid mapping that we do not know about. 383 */ 384 if (v34) { 385 if (ap->a_accmode & VREAD) 386 mode = NFSACCESS_READ; 387 else 388 mode = 0; 389 if (vp->v_type != VDIR) { 390 if (ap->a_accmode & VWRITE) 391 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 392 if (ap->a_accmode & VAPPEND) 393 mode |= NFSACCESS_EXTEND; 394 if (ap->a_accmode & VEXEC) 395 mode |= NFSACCESS_EXECUTE; 396 if (ap->a_accmode & VDELETE) 397 mode |= NFSACCESS_DELETE; 398 } else { 399 if (ap->a_accmode & VWRITE) 400 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 401 if (ap->a_accmode & VAPPEND) 402 mode |= NFSACCESS_EXTEND; 403 if (ap->a_accmode & VEXEC) 404 mode |= NFSACCESS_LOOKUP; 405 if (ap->a_accmode & VDELETE) 406 mode |= NFSACCESS_DELETE; 407 if (ap->a_accmode & VDELETE_CHILD) 408 mode |= NFSACCESS_MODIFY; 409 } 410 /* XXX safety belt, only make blanket request if caching */ 411 if (nfsaccess_cache_timeout > 0) { 412 wmode = NFSACCESS_READ | NFSACCESS_MODIFY | 413 NFSACCESS_EXTEND | NFSACCESS_EXECUTE | 414 NFSACCESS_DELETE | NFSACCESS_LOOKUP; 415 } else { 416 wmode = mode; 417 } 418 419 /* 420 * Does our cached result allow us to give a definite yes to 421 * this request? 422 */ 423 gotahit = 0; 424 mtx_lock(&np->n_mtx); 425 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 426 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) { 427 if (time_second < (np->n_accesscache[i].stamp 428 + nfsaccess_cache_timeout) && 429 (np->n_accesscache[i].mode & mode) == mode) { 430 NFSINCRGLOBAL(nfsstatsv1.accesscache_hits); 431 gotahit = 1; 432 } 433 break; 434 } 435 } 436 mtx_unlock(&np->n_mtx); 437 #ifdef KDTRACE_HOOKS 438 if (gotahit != 0) 439 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp, 440 ap->a_cred->cr_uid, mode); 441 else 442 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp, 443 ap->a_cred->cr_uid, mode); 444 #endif 445 if (gotahit == 0) { 446 /* 447 * Either a no, or a don't know. Go to the wire. 448 */ 449 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses); 450 error = nfs34_access_otw(vp, wmode, ap->a_td, 451 ap->a_cred, &rmode); 452 if (!error && 453 (rmode & mode) != mode) 454 error = EACCES; 455 } 456 return (error); 457 } else { 458 if ((error = nfsspec_access(ap)) != 0) { 459 return (error); 460 } 461 /* 462 * Attempt to prevent a mapped root from accessing a file 463 * which it shouldn't. We try to read a byte from the file 464 * if the user is root and the file is not zero length. 465 * After calling nfsspec_access, we should have the correct 466 * file size cached. 467 */ 468 mtx_lock(&np->n_mtx); 469 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD) 470 && VTONFS(vp)->n_size > 0) { 471 struct iovec aiov; 472 struct uio auio; 473 char buf[1]; 474 475 mtx_unlock(&np->n_mtx); 476 aiov.iov_base = buf; 477 aiov.iov_len = 1; 478 auio.uio_iov = &aiov; 479 auio.uio_iovcnt = 1; 480 auio.uio_offset = 0; 481 auio.uio_resid = 1; 482 auio.uio_segflg = UIO_SYSSPACE; 483 auio.uio_rw = UIO_READ; 484 auio.uio_td = ap->a_td; 485 486 if (vp->v_type == VREG) 487 error = ncl_readrpc(vp, &auio, ap->a_cred); 488 else if (vp->v_type == VDIR) { 489 char* bp; 490 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 491 aiov.iov_base = bp; 492 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 493 error = ncl_readdirrpc(vp, &auio, ap->a_cred, 494 ap->a_td); 495 free(bp, M_TEMP); 496 } else if (vp->v_type == VLNK) 497 error = ncl_readlinkrpc(vp, &auio, ap->a_cred); 498 else 499 error = EACCES; 500 } else 501 mtx_unlock(&np->n_mtx); 502 return (error); 503 } 504 } 505 506 507 /* 508 * nfs open vnode op 509 * Check to see if the type is ok 510 * and that deletion is not in progress. 511 * For paged in text files, you will need to flush the page cache 512 * if consistency is lost. 513 */ 514 /* ARGSUSED */ 515 static int 516 nfs_open(struct vop_open_args *ap) 517 { 518 struct vnode *vp = ap->a_vp; 519 struct nfsnode *np = VTONFS(vp); 520 struct vattr vattr; 521 int error; 522 int fmode = ap->a_mode; 523 struct ucred *cred; 524 vm_object_t obj; 525 526 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) 527 return (EOPNOTSUPP); 528 529 /* 530 * For NFSv4, we need to do the Open Op before cache validation, 531 * so that we conform to RFC3530 Sec. 9.3.1. 532 */ 533 if (NFS_ISV4(vp)) { 534 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td); 535 if (error) { 536 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 537 (gid_t)0); 538 return (error); 539 } 540 } 541 542 /* 543 * Now, if this Open will be doing reading, re-validate/flush the 544 * cache, so that Close/Open coherency is maintained. 545 */ 546 mtx_lock(&np->n_mtx); 547 if (np->n_flag & NMODIFIED) { 548 mtx_unlock(&np->n_mtx); 549 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 550 if (error == EINTR || error == EIO) { 551 if (NFS_ISV4(vp)) 552 (void) nfsrpc_close(vp, 0, ap->a_td); 553 return (error); 554 } 555 mtx_lock(&np->n_mtx); 556 np->n_attrstamp = 0; 557 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 558 if (vp->v_type == VDIR) 559 np->n_direofoffset = 0; 560 mtx_unlock(&np->n_mtx); 561 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 562 if (error) { 563 if (NFS_ISV4(vp)) 564 (void) nfsrpc_close(vp, 0, ap->a_td); 565 return (error); 566 } 567 mtx_lock(&np->n_mtx); 568 np->n_mtime = vattr.va_mtime; 569 if (NFS_ISV4(vp)) 570 np->n_change = vattr.va_filerev; 571 } else { 572 mtx_unlock(&np->n_mtx); 573 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 574 if (error) { 575 if (NFS_ISV4(vp)) 576 (void) nfsrpc_close(vp, 0, ap->a_td); 577 return (error); 578 } 579 mtx_lock(&np->n_mtx); 580 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) || 581 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 582 if (vp->v_type == VDIR) 583 np->n_direofoffset = 0; 584 mtx_unlock(&np->n_mtx); 585 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 586 if (error == EINTR || error == EIO) { 587 if (NFS_ISV4(vp)) 588 (void) nfsrpc_close(vp, 0, ap->a_td); 589 return (error); 590 } 591 mtx_lock(&np->n_mtx); 592 np->n_mtime = vattr.va_mtime; 593 if (NFS_ISV4(vp)) 594 np->n_change = vattr.va_filerev; 595 } 596 } 597 598 /* 599 * If the object has >= 1 O_DIRECT active opens, we disable caching. 600 */ 601 if (newnfs_directio_enable && (fmode & O_DIRECT) && 602 (vp->v_type == VREG)) { 603 if (np->n_directio_opens == 0) { 604 mtx_unlock(&np->n_mtx); 605 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 606 if (error) { 607 if (NFS_ISV4(vp)) 608 (void) nfsrpc_close(vp, 0, ap->a_td); 609 return (error); 610 } 611 mtx_lock(&np->n_mtx); 612 np->n_flag |= NNONCACHE; 613 } 614 np->n_directio_opens++; 615 } 616 617 /* If opened for writing via NFSv4.1 or later, mark that for pNFS. */ 618 if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0) 619 np->n_flag |= NWRITEOPENED; 620 621 /* 622 * If this is an open for writing, capture a reference to the 623 * credentials, so they can be used by ncl_putpages(). Using 624 * these write credentials is preferable to the credentials of 625 * whatever thread happens to be doing the VOP_PUTPAGES() since 626 * the write RPCs are less likely to fail with EACCES. 627 */ 628 if ((fmode & FWRITE) != 0) { 629 cred = np->n_writecred; 630 np->n_writecred = crhold(ap->a_cred); 631 } else 632 cred = NULL; 633 mtx_unlock(&np->n_mtx); 634 635 if (cred != NULL) 636 crfree(cred); 637 vnode_create_vobject(vp, vattr.va_size, ap->a_td); 638 639 /* 640 * If the text file has been mmap'd, flush any dirty pages to the 641 * buffer cache and then... 642 * Make sure all writes are pushed to the NFS server. If this is not 643 * done, the modify time of the file can change while the text 644 * file is being executed. This will cause the process that is 645 * executing the text file to be terminated. 646 */ 647 if (vp->v_writecount <= -1) { 648 if ((obj = vp->v_object) != NULL && 649 (obj->flags & OBJ_MIGHTBEDIRTY) != 0) { 650 VM_OBJECT_WLOCK(obj); 651 vm_object_page_clean(obj, 0, 0, OBJPC_SYNC); 652 VM_OBJECT_WUNLOCK(obj); 653 } 654 655 /* Now, flush the buffer cache. */ 656 ncl_flush(vp, MNT_WAIT, curthread, 0, 0); 657 658 /* And, finally, make sure that n_mtime is up to date. */ 659 np = VTONFS(vp); 660 mtx_lock(&np->n_mtx); 661 np->n_mtime = np->n_vattr.na_mtime; 662 mtx_unlock(&np->n_mtx); 663 } 664 return (0); 665 } 666 667 /* 668 * nfs close vnode op 669 * What an NFS client should do upon close after writing is a debatable issue. 670 * Most NFS clients push delayed writes to the server upon close, basically for 671 * two reasons: 672 * 1 - So that any write errors may be reported back to the client process 673 * doing the close system call. By far the two most likely errors are 674 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 675 * 2 - To put a worst case upper bound on cache inconsistency between 676 * multiple clients for the file. 677 * There is also a consistency problem for Version 2 of the protocol w.r.t. 678 * not being able to tell if other clients are writing a file concurrently, 679 * since there is no way of knowing if the changed modify time in the reply 680 * is only due to the write for this client. 681 * (NFS Version 3 provides weak cache consistency data in the reply that 682 * should be sufficient to detect and handle this case.) 683 * 684 * The current code does the following: 685 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 686 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 687 * or commit them (this satisfies 1 and 2 except for the 688 * case where the server crashes after this close but 689 * before the commit RPC, which is felt to be "good 690 * enough". Changing the last argument to ncl_flush() to 691 * a 1 would force a commit operation, if it is felt a 692 * commit is necessary now. 693 * for NFS Version 4 - flush the dirty buffers and commit them, if 694 * nfscl_mustflush() says this is necessary. 695 * It is necessary if there is no write delegation held, 696 * in order to satisfy open/close coherency. 697 * If the file isn't cached on local stable storage, 698 * it may be necessary in order to detect "out of space" 699 * errors from the server, if the write delegation 700 * issued by the server doesn't allow the file to grow. 701 */ 702 /* ARGSUSED */ 703 static int 704 nfs_close(struct vop_close_args *ap) 705 { 706 struct vnode *vp = ap->a_vp; 707 struct nfsnode *np = VTONFS(vp); 708 struct nfsvattr nfsva; 709 struct ucred *cred; 710 int error = 0, ret, localcred = 0; 711 int fmode = ap->a_fflag; 712 713 if (NFSCL_FORCEDISM(vp->v_mount)) 714 return (0); 715 /* 716 * During shutdown, a_cred isn't valid, so just use root. 717 */ 718 if (ap->a_cred == NOCRED) { 719 cred = newnfs_getcred(); 720 localcred = 1; 721 } else { 722 cred = ap->a_cred; 723 } 724 if (vp->v_type == VREG) { 725 /* 726 * Examine and clean dirty pages, regardless of NMODIFIED. 727 * This closes a major hole in close-to-open consistency. 728 * We want to push out all dirty pages (and buffers) on 729 * close, regardless of whether they were dirtied by 730 * mmap'ed writes or via write(). 731 */ 732 if (nfs_clean_pages_on_close && vp->v_object) { 733 VM_OBJECT_WLOCK(vp->v_object); 734 vm_object_page_clean(vp->v_object, 0, 0, 0); 735 VM_OBJECT_WUNLOCK(vp->v_object); 736 } 737 mtx_lock(&np->n_mtx); 738 if (np->n_flag & NMODIFIED) { 739 mtx_unlock(&np->n_mtx); 740 if (NFS_ISV3(vp)) { 741 /* 742 * Under NFSv3 we have dirty buffers to dispose of. We 743 * must flush them to the NFS server. We have the option 744 * of waiting all the way through the commit rpc or just 745 * waiting for the initial write. The default is to only 746 * wait through the initial write so the data is in the 747 * server's cache, which is roughly similar to the state 748 * a standard disk subsystem leaves the file in on close(). 749 * 750 * We cannot clear the NMODIFIED bit in np->n_flag due to 751 * potential races with other processes, and certainly 752 * cannot clear it if we don't commit. 753 * These races occur when there is no longer the old 754 * traditional vnode locking implemented for Vnode Ops. 755 */ 756 int cm = newnfs_commit_on_close ? 1 : 0; 757 error = ncl_flush(vp, MNT_WAIT, ap->a_td, cm, 0); 758 /* np->n_flag &= ~NMODIFIED; */ 759 } else if (NFS_ISV4(vp)) { 760 if (nfscl_mustflush(vp) != 0) { 761 int cm = newnfs_commit_on_close ? 1 : 0; 762 error = ncl_flush(vp, MNT_WAIT, ap->a_td, 763 cm, 0); 764 /* 765 * as above w.r.t races when clearing 766 * NMODIFIED. 767 * np->n_flag &= ~NMODIFIED; 768 */ 769 } 770 } else { 771 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 772 } 773 mtx_lock(&np->n_mtx); 774 } 775 /* 776 * Invalidate the attribute cache in all cases. 777 * An open is going to fetch fresh attrs any way, other procs 778 * on this node that have file open will be forced to do an 779 * otw attr fetch, but this is safe. 780 * --> A user found that their RPC count dropped by 20% when 781 * this was commented out and I can't see any requirement 782 * for it, so I've disabled it when negative lookups are 783 * enabled. (What does this have to do with negative lookup 784 * caching? Well nothing, except it was reported by the 785 * same user that needed negative lookup caching and I wanted 786 * there to be a way to disable it to see if it 787 * is the cause of some caching/coherency issue that might 788 * crop up.) 789 */ 790 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) { 791 np->n_attrstamp = 0; 792 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 793 } 794 if (np->n_flag & NWRITEERR) { 795 np->n_flag &= ~NWRITEERR; 796 error = np->n_error; 797 } 798 mtx_unlock(&np->n_mtx); 799 } 800 801 if (NFS_ISV4(vp)) { 802 /* 803 * Get attributes so "change" is up to date. 804 */ 805 if (error == 0 && nfscl_mustflush(vp) != 0 && 806 vp->v_type == VREG && 807 (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) { 808 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva, 809 NULL); 810 if (!ret) { 811 np->n_change = nfsva.na_filerev; 812 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 813 NULL, 0, 0); 814 } 815 } 816 817 /* 818 * and do the close. 819 */ 820 ret = nfsrpc_close(vp, 0, ap->a_td); 821 if (!error && ret) 822 error = ret; 823 if (error) 824 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 825 (gid_t)0); 826 } 827 if (newnfs_directio_enable) 828 KASSERT((np->n_directio_asyncwr == 0), 829 ("nfs_close: dirty unflushed (%d) directio buffers\n", 830 np->n_directio_asyncwr)); 831 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) { 832 mtx_lock(&np->n_mtx); 833 KASSERT((np->n_directio_opens > 0), 834 ("nfs_close: unexpectedly value (0) of n_directio_opens\n")); 835 np->n_directio_opens--; 836 if (np->n_directio_opens == 0) 837 np->n_flag &= ~NNONCACHE; 838 mtx_unlock(&np->n_mtx); 839 } 840 if (localcred) 841 NFSFREECRED(cred); 842 return (error); 843 } 844 845 /* 846 * nfs getattr call from vfs. 847 */ 848 static int 849 nfs_getattr(struct vop_getattr_args *ap) 850 { 851 struct vnode *vp = ap->a_vp; 852 struct thread *td = curthread; /* XXX */ 853 struct nfsnode *np = VTONFS(vp); 854 int error = 0; 855 struct nfsvattr nfsva; 856 struct vattr *vap = ap->a_vap; 857 struct vattr vattr; 858 859 /* 860 * Update local times for special files. 861 */ 862 mtx_lock(&np->n_mtx); 863 if (np->n_flag & (NACC | NUPD)) 864 np->n_flag |= NCHG; 865 mtx_unlock(&np->n_mtx); 866 /* 867 * First look in the cache. 868 */ 869 if (ncl_getattrcache(vp, &vattr) == 0) { 870 vap->va_type = vattr.va_type; 871 vap->va_mode = vattr.va_mode; 872 vap->va_nlink = vattr.va_nlink; 873 vap->va_uid = vattr.va_uid; 874 vap->va_gid = vattr.va_gid; 875 vap->va_fsid = vattr.va_fsid; 876 vap->va_fileid = vattr.va_fileid; 877 vap->va_size = vattr.va_size; 878 vap->va_blocksize = vattr.va_blocksize; 879 vap->va_atime = vattr.va_atime; 880 vap->va_mtime = vattr.va_mtime; 881 vap->va_ctime = vattr.va_ctime; 882 vap->va_gen = vattr.va_gen; 883 vap->va_flags = vattr.va_flags; 884 vap->va_rdev = vattr.va_rdev; 885 vap->va_bytes = vattr.va_bytes; 886 vap->va_filerev = vattr.va_filerev; 887 /* 888 * Get the local modify time for the case of a write 889 * delegation. 890 */ 891 nfscl_deleggetmodtime(vp, &vap->va_mtime); 892 return (0); 893 } 894 895 if (NFS_ISV34(vp) && nfs_prime_access_cache && 896 nfsaccess_cache_timeout > 0) { 897 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses); 898 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL); 899 if (ncl_getattrcache(vp, ap->a_vap) == 0) { 900 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime); 901 return (0); 902 } 903 } 904 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL); 905 if (!error) 906 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0); 907 if (!error) { 908 /* 909 * Get the local modify time for the case of a write 910 * delegation. 911 */ 912 nfscl_deleggetmodtime(vp, &vap->va_mtime); 913 } else if (NFS_ISV4(vp)) { 914 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 915 } 916 return (error); 917 } 918 919 /* 920 * nfs setattr call. 921 */ 922 static int 923 nfs_setattr(struct vop_setattr_args *ap) 924 { 925 struct vnode *vp = ap->a_vp; 926 struct nfsnode *np = VTONFS(vp); 927 struct thread *td = curthread; /* XXX */ 928 struct vattr *vap = ap->a_vap; 929 int error = 0; 930 u_quad_t tsize; 931 932 #ifndef nolint 933 tsize = (u_quad_t)0; 934 #endif 935 936 /* 937 * Setting of flags and marking of atimes are not supported. 938 */ 939 if (vap->va_flags != VNOVAL) 940 return (EOPNOTSUPP); 941 942 /* 943 * Disallow write attempts if the filesystem is mounted read-only. 944 */ 945 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 946 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 947 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 948 (vp->v_mount->mnt_flag & MNT_RDONLY)) 949 return (EROFS); 950 if (vap->va_size != VNOVAL) { 951 switch (vp->v_type) { 952 case VDIR: 953 return (EISDIR); 954 case VCHR: 955 case VBLK: 956 case VSOCK: 957 case VFIFO: 958 if (vap->va_mtime.tv_sec == VNOVAL && 959 vap->va_atime.tv_sec == VNOVAL && 960 vap->va_mode == (mode_t)VNOVAL && 961 vap->va_uid == (uid_t)VNOVAL && 962 vap->va_gid == (gid_t)VNOVAL) 963 return (0); 964 vap->va_size = VNOVAL; 965 break; 966 default: 967 /* 968 * Disallow write attempts if the filesystem is 969 * mounted read-only. 970 */ 971 if (vp->v_mount->mnt_flag & MNT_RDONLY) 972 return (EROFS); 973 /* 974 * We run vnode_pager_setsize() early (why?), 975 * we must set np->n_size now to avoid vinvalbuf 976 * V_SAVE races that might setsize a lower 977 * value. 978 */ 979 mtx_lock(&np->n_mtx); 980 tsize = np->n_size; 981 mtx_unlock(&np->n_mtx); 982 error = ncl_meta_setsize(vp, td, vap->va_size); 983 mtx_lock(&np->n_mtx); 984 if (np->n_flag & NMODIFIED) { 985 tsize = np->n_size; 986 mtx_unlock(&np->n_mtx); 987 error = ncl_vinvalbuf(vp, vap->va_size == 0 ? 988 0 : V_SAVE, td, 1); 989 if (error != 0) { 990 vnode_pager_setsize(vp, tsize); 991 return (error); 992 } 993 /* 994 * Call nfscl_delegmodtime() to set the modify time 995 * locally, as required. 996 */ 997 nfscl_delegmodtime(vp); 998 } else 999 mtx_unlock(&np->n_mtx); 1000 /* 1001 * np->n_size has already been set to vap->va_size 1002 * in ncl_meta_setsize(). We must set it again since 1003 * nfs_loadattrcache() could be called through 1004 * ncl_meta_setsize() and could modify np->n_size. 1005 */ 1006 mtx_lock(&np->n_mtx); 1007 np->n_vattr.na_size = np->n_size = vap->va_size; 1008 mtx_unlock(&np->n_mtx); 1009 } 1010 } else { 1011 mtx_lock(&np->n_mtx); 1012 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) && 1013 (np->n_flag & NMODIFIED) && vp->v_type == VREG) { 1014 mtx_unlock(&np->n_mtx); 1015 error = ncl_vinvalbuf(vp, V_SAVE, td, 1); 1016 if (error == EINTR || error == EIO) 1017 return (error); 1018 } else 1019 mtx_unlock(&np->n_mtx); 1020 } 1021 error = nfs_setattrrpc(vp, vap, ap->a_cred, td); 1022 if (error && vap->va_size != VNOVAL) { 1023 mtx_lock(&np->n_mtx); 1024 np->n_size = np->n_vattr.na_size = tsize; 1025 vnode_pager_setsize(vp, tsize); 1026 mtx_unlock(&np->n_mtx); 1027 } 1028 return (error); 1029 } 1030 1031 /* 1032 * Do an nfs setattr rpc. 1033 */ 1034 static int 1035 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred, 1036 struct thread *td) 1037 { 1038 struct nfsnode *np = VTONFS(vp); 1039 int error, ret, attrflag, i; 1040 struct nfsvattr nfsva; 1041 1042 if (NFS_ISV34(vp)) { 1043 mtx_lock(&np->n_mtx); 1044 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) 1045 np->n_accesscache[i].stamp = 0; 1046 np->n_flag |= NDELEGMOD; 1047 mtx_unlock(&np->n_mtx); 1048 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp); 1049 } 1050 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag, 1051 NULL); 1052 if (attrflag) { 1053 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1054 if (ret && !error) 1055 error = ret; 1056 } 1057 if (error && NFS_ISV4(vp)) 1058 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid); 1059 return (error); 1060 } 1061 1062 /* 1063 * nfs lookup call, one step at a time... 1064 * First look in cache 1065 * If not found, unlock the directory nfsnode and do the rpc 1066 */ 1067 static int 1068 nfs_lookup(struct vop_lookup_args *ap) 1069 { 1070 struct componentname *cnp = ap->a_cnp; 1071 struct vnode *dvp = ap->a_dvp; 1072 struct vnode **vpp = ap->a_vpp; 1073 struct mount *mp = dvp->v_mount; 1074 int flags = cnp->cn_flags; 1075 struct vnode *newvp; 1076 struct nfsmount *nmp; 1077 struct nfsnode *np, *newnp; 1078 int error = 0, attrflag, dattrflag, ltype, ncticks; 1079 struct thread *td = cnp->cn_thread; 1080 struct nfsfh *nfhp; 1081 struct nfsvattr dnfsva, nfsva; 1082 struct vattr vattr; 1083 struct timespec nctime; 1084 1085 *vpp = NULLVP; 1086 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) && 1087 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 1088 return (EROFS); 1089 if (dvp->v_type != VDIR) 1090 return (ENOTDIR); 1091 nmp = VFSTONFS(mp); 1092 np = VTONFS(dvp); 1093 1094 /* For NFSv4, wait until any remove is done. */ 1095 mtx_lock(&np->n_mtx); 1096 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) { 1097 np->n_flag |= NREMOVEWANT; 1098 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0); 1099 } 1100 mtx_unlock(&np->n_mtx); 1101 1102 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) 1103 return (error); 1104 error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks); 1105 if (error > 0 && error != ENOENT) 1106 return (error); 1107 if (error == -1) { 1108 /* 1109 * Lookups of "." are special and always return the 1110 * current directory. cache_lookup() already handles 1111 * associated locking bookkeeping, etc. 1112 */ 1113 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 1114 /* XXX: Is this really correct? */ 1115 if (cnp->cn_nameiop != LOOKUP && 1116 (flags & ISLASTCN)) 1117 cnp->cn_flags |= SAVENAME; 1118 return (0); 1119 } 1120 1121 /* 1122 * We only accept a positive hit in the cache if the 1123 * change time of the file matches our cached copy. 1124 * Otherwise, we discard the cache entry and fallback 1125 * to doing a lookup RPC. We also only trust cache 1126 * entries for less than nm_nametimeo seconds. 1127 * 1128 * To better handle stale file handles and attributes, 1129 * clear the attribute cache of this node if it is a 1130 * leaf component, part of an open() call, and not 1131 * locally modified before fetching the attributes. 1132 * This should allow stale file handles to be detected 1133 * here where we can fall back to a LOOKUP RPC to 1134 * recover rather than having nfs_open() detect the 1135 * stale file handle and failing open(2) with ESTALE. 1136 */ 1137 newvp = *vpp; 1138 newnp = VTONFS(newvp); 1139 if (!(nmp->nm_flag & NFSMNT_NOCTO) && 1140 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1141 !(newnp->n_flag & NMODIFIED)) { 1142 mtx_lock(&newnp->n_mtx); 1143 newnp->n_attrstamp = 0; 1144 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1145 mtx_unlock(&newnp->n_mtx); 1146 } 1147 if (nfscl_nodeleg(newvp, 0) == 0 || 1148 ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) && 1149 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 && 1150 timespeccmp(&vattr.va_ctime, &nctime, ==))) { 1151 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits); 1152 if (cnp->cn_nameiop != LOOKUP && 1153 (flags & ISLASTCN)) 1154 cnp->cn_flags |= SAVENAME; 1155 return (0); 1156 } 1157 cache_purge(newvp); 1158 if (dvp != newvp) 1159 vput(newvp); 1160 else 1161 vrele(newvp); 1162 *vpp = NULLVP; 1163 } else if (error == ENOENT) { 1164 if (dvp->v_iflag & VI_DOOMED) 1165 return (ENOENT); 1166 /* 1167 * We only accept a negative hit in the cache if the 1168 * modification time of the parent directory matches 1169 * the cached copy in the name cache entry. 1170 * Otherwise, we discard all of the negative cache 1171 * entries for this directory. We also only trust 1172 * negative cache entries for up to nm_negnametimeo 1173 * seconds. 1174 */ 1175 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) && 1176 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 && 1177 timespeccmp(&vattr.va_mtime, &nctime, ==)) { 1178 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits); 1179 return (ENOENT); 1180 } 1181 cache_purge_negative(dvp); 1182 } 1183 1184 error = 0; 1185 newvp = NULLVP; 1186 NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses); 1187 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1188 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1189 NULL); 1190 if (dattrflag) 1191 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1192 if (error) { 1193 if (newvp != NULLVP) { 1194 vput(newvp); 1195 *vpp = NULLVP; 1196 } 1197 1198 if (error != ENOENT) { 1199 if (NFS_ISV4(dvp)) 1200 error = nfscl_maperr(td, error, (uid_t)0, 1201 (gid_t)0); 1202 return (error); 1203 } 1204 1205 /* The requested file was not found. */ 1206 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1207 (flags & ISLASTCN)) { 1208 /* 1209 * XXX: UFS does a full VOP_ACCESS(dvp, 1210 * VWRITE) here instead of just checking 1211 * MNT_RDONLY. 1212 */ 1213 if (mp->mnt_flag & MNT_RDONLY) 1214 return (EROFS); 1215 cnp->cn_flags |= SAVENAME; 1216 return (EJUSTRETURN); 1217 } 1218 1219 if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) { 1220 /* 1221 * Cache the modification time of the parent 1222 * directory from the post-op attributes in 1223 * the name cache entry. The negative cache 1224 * entry will be ignored once the directory 1225 * has changed. Don't bother adding the entry 1226 * if the directory has already changed. 1227 */ 1228 mtx_lock(&np->n_mtx); 1229 if (timespeccmp(&np->n_vattr.na_mtime, 1230 &dnfsva.na_mtime, ==)) { 1231 mtx_unlock(&np->n_mtx); 1232 cache_enter_time(dvp, NULL, cnp, 1233 &dnfsva.na_mtime, NULL); 1234 } else 1235 mtx_unlock(&np->n_mtx); 1236 } 1237 return (ENOENT); 1238 } 1239 1240 /* 1241 * Handle RENAME case... 1242 */ 1243 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 1244 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1245 free(nfhp, M_NFSFH); 1246 return (EISDIR); 1247 } 1248 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1249 LK_EXCLUSIVE); 1250 if (error) 1251 return (error); 1252 newvp = NFSTOV(np); 1253 if (attrflag) 1254 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1255 0, 1); 1256 *vpp = newvp; 1257 cnp->cn_flags |= SAVENAME; 1258 return (0); 1259 } 1260 1261 if (flags & ISDOTDOT) { 1262 ltype = NFSVOPISLOCKED(dvp); 1263 error = vfs_busy(mp, MBF_NOWAIT); 1264 if (error != 0) { 1265 vfs_ref(mp); 1266 NFSVOPUNLOCK(dvp, 0); 1267 error = vfs_busy(mp, 0); 1268 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1269 vfs_rel(mp); 1270 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) { 1271 vfs_unbusy(mp); 1272 error = ENOENT; 1273 } 1274 if (error != 0) 1275 return (error); 1276 } 1277 NFSVOPUNLOCK(dvp, 0); 1278 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1279 cnp->cn_lkflags); 1280 if (error == 0) 1281 newvp = NFSTOV(np); 1282 vfs_unbusy(mp); 1283 if (newvp != dvp) 1284 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1285 if (dvp->v_iflag & VI_DOOMED) { 1286 if (error == 0) { 1287 if (newvp == dvp) 1288 vrele(newvp); 1289 else 1290 vput(newvp); 1291 } 1292 error = ENOENT; 1293 } 1294 if (error != 0) 1295 return (error); 1296 if (attrflag) 1297 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1298 0, 1); 1299 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1300 free(nfhp, M_NFSFH); 1301 VREF(dvp); 1302 newvp = dvp; 1303 if (attrflag) 1304 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1305 0, 1); 1306 } else { 1307 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1308 cnp->cn_lkflags); 1309 if (error) 1310 return (error); 1311 newvp = NFSTOV(np); 1312 if (attrflag) 1313 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1314 0, 1); 1315 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1316 !(np->n_flag & NMODIFIED)) { 1317 /* 1318 * Flush the attribute cache when opening a 1319 * leaf node to ensure that fresh attributes 1320 * are fetched in nfs_open() since we did not 1321 * fetch attributes from the LOOKUP reply. 1322 */ 1323 mtx_lock(&np->n_mtx); 1324 np->n_attrstamp = 0; 1325 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1326 mtx_unlock(&np->n_mtx); 1327 } 1328 } 1329 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1330 cnp->cn_flags |= SAVENAME; 1331 if ((cnp->cn_flags & MAKEENTRY) && 1332 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) && 1333 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0)) 1334 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 1335 newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime); 1336 *vpp = newvp; 1337 return (0); 1338 } 1339 1340 /* 1341 * nfs read call. 1342 * Just call ncl_bioread() to do the work. 1343 */ 1344 static int 1345 nfs_read(struct vop_read_args *ap) 1346 { 1347 struct vnode *vp = ap->a_vp; 1348 1349 switch (vp->v_type) { 1350 case VREG: 1351 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred)); 1352 case VDIR: 1353 return (EISDIR); 1354 default: 1355 return (EOPNOTSUPP); 1356 } 1357 } 1358 1359 /* 1360 * nfs readlink call 1361 */ 1362 static int 1363 nfs_readlink(struct vop_readlink_args *ap) 1364 { 1365 struct vnode *vp = ap->a_vp; 1366 1367 if (vp->v_type != VLNK) 1368 return (EINVAL); 1369 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred)); 1370 } 1371 1372 /* 1373 * Do a readlink rpc. 1374 * Called by ncl_doio() from below the buffer cache. 1375 */ 1376 int 1377 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1378 { 1379 int error, ret, attrflag; 1380 struct nfsvattr nfsva; 1381 1382 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva, 1383 &attrflag, NULL); 1384 if (attrflag) { 1385 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1386 if (ret && !error) 1387 error = ret; 1388 } 1389 if (error && NFS_ISV4(vp)) 1390 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1391 return (error); 1392 } 1393 1394 /* 1395 * nfs read rpc call 1396 * Ditto above 1397 */ 1398 int 1399 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1400 { 1401 int error, ret, attrflag; 1402 struct nfsvattr nfsva; 1403 struct nfsmount *nmp; 1404 1405 nmp = VFSTONFS(vnode_mount(vp)); 1406 error = EIO; 1407 attrflag = 0; 1408 if (NFSHASPNFS(nmp)) 1409 error = nfscl_doiods(vp, uiop, NULL, NULL, 1410 NFSV4OPEN_ACCESSREAD, 0, cred, uiop->uio_td); 1411 NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error); 1412 if (error != 0) 1413 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, 1414 &attrflag, NULL); 1415 if (attrflag) { 1416 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1417 if (ret && !error) 1418 error = ret; 1419 } 1420 if (error && NFS_ISV4(vp)) 1421 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1422 return (error); 1423 } 1424 1425 /* 1426 * nfs write call 1427 */ 1428 int 1429 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 1430 int *iomode, int *must_commit, int called_from_strategy) 1431 { 1432 struct nfsvattr nfsva; 1433 int error, attrflag, ret; 1434 struct nfsmount *nmp; 1435 1436 nmp = VFSTONFS(vnode_mount(vp)); 1437 error = EIO; 1438 attrflag = 0; 1439 if (NFSHASPNFS(nmp)) 1440 error = nfscl_doiods(vp, uiop, iomode, must_commit, 1441 NFSV4OPEN_ACCESSWRITE, 0, cred, uiop->uio_td); 1442 NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error); 1443 if (error != 0) 1444 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred, 1445 uiop->uio_td, &nfsva, &attrflag, NULL, 1446 called_from_strategy); 1447 if (attrflag) { 1448 if (VTONFS(vp)->n_flag & ND_NFSV4) 1449 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1, 1450 1); 1451 else 1452 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1453 1); 1454 if (ret && !error) 1455 error = ret; 1456 } 1457 if (DOINGASYNC(vp)) 1458 *iomode = NFSWRITE_FILESYNC; 1459 if (error && NFS_ISV4(vp)) 1460 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1461 return (error); 1462 } 1463 1464 /* 1465 * nfs mknod rpc 1466 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1467 * mode set to specify the file type and the size field for rdev. 1468 */ 1469 static int 1470 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1471 struct vattr *vap) 1472 { 1473 struct nfsvattr nfsva, dnfsva; 1474 struct vnode *newvp = NULL; 1475 struct nfsnode *np = NULL, *dnp; 1476 struct nfsfh *nfhp; 1477 struct vattr vattr; 1478 int error = 0, attrflag, dattrflag; 1479 u_int32_t rdev; 1480 1481 if (vap->va_type == VCHR || vap->va_type == VBLK) 1482 rdev = vap->va_rdev; 1483 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1484 rdev = 0xffffffff; 1485 else 1486 return (EOPNOTSUPP); 1487 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1488 return (error); 1489 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap, 1490 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva, 1491 &nfsva, &nfhp, &attrflag, &dattrflag, NULL); 1492 if (!error) { 1493 if (!nfhp) 1494 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1495 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, 1496 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1497 NULL); 1498 if (nfhp) 1499 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1500 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); 1501 } 1502 if (dattrflag) 1503 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1504 if (!error) { 1505 newvp = NFSTOV(np); 1506 if (attrflag != 0) { 1507 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1508 0, 1); 1509 if (error != 0) 1510 vput(newvp); 1511 } 1512 } 1513 if (!error) { 1514 *vpp = newvp; 1515 } else if (NFS_ISV4(dvp)) { 1516 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, 1517 vap->va_gid); 1518 } 1519 dnp = VTONFS(dvp); 1520 mtx_lock(&dnp->n_mtx); 1521 dnp->n_flag |= NMODIFIED; 1522 if (!dattrflag) { 1523 dnp->n_attrstamp = 0; 1524 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1525 } 1526 mtx_unlock(&dnp->n_mtx); 1527 return (error); 1528 } 1529 1530 /* 1531 * nfs mknod vop 1532 * just call nfs_mknodrpc() to do the work. 1533 */ 1534 /* ARGSUSED */ 1535 static int 1536 nfs_mknod(struct vop_mknod_args *ap) 1537 { 1538 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap)); 1539 } 1540 1541 static struct mtx nfs_cverf_mtx; 1542 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex", 1543 MTX_DEF); 1544 1545 static nfsquad_t 1546 nfs_get_cverf(void) 1547 { 1548 static nfsquad_t cverf; 1549 nfsquad_t ret; 1550 static int cverf_initialized = 0; 1551 1552 mtx_lock(&nfs_cverf_mtx); 1553 if (cverf_initialized == 0) { 1554 cverf.lval[0] = arc4random(); 1555 cverf.lval[1] = arc4random(); 1556 cverf_initialized = 1; 1557 } else 1558 cverf.qval++; 1559 ret = cverf; 1560 mtx_unlock(&nfs_cverf_mtx); 1561 1562 return (ret); 1563 } 1564 1565 /* 1566 * nfs file create call 1567 */ 1568 static int 1569 nfs_create(struct vop_create_args *ap) 1570 { 1571 struct vnode *dvp = ap->a_dvp; 1572 struct vattr *vap = ap->a_vap; 1573 struct componentname *cnp = ap->a_cnp; 1574 struct nfsnode *np = NULL, *dnp; 1575 struct vnode *newvp = NULL; 1576 struct nfsmount *nmp; 1577 struct nfsvattr dnfsva, nfsva; 1578 struct nfsfh *nfhp; 1579 nfsquad_t cverf; 1580 int error = 0, attrflag, dattrflag, fmode = 0; 1581 struct vattr vattr; 1582 1583 /* 1584 * Oops, not for me.. 1585 */ 1586 if (vap->va_type == VSOCK) 1587 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1588 1589 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1590 return (error); 1591 if (vap->va_vaflags & VA_EXCLUSIVE) 1592 fmode |= O_EXCL; 1593 dnp = VTONFS(dvp); 1594 nmp = VFSTONFS(vnode_mount(dvp)); 1595 again: 1596 /* For NFSv4, wait until any remove is done. */ 1597 mtx_lock(&dnp->n_mtx); 1598 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) { 1599 dnp->n_flag |= NREMOVEWANT; 1600 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0); 1601 } 1602 mtx_unlock(&dnp->n_mtx); 1603 1604 cverf = nfs_get_cverf(); 1605 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1606 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, 1607 &nfhp, &attrflag, &dattrflag, NULL); 1608 if (!error) { 1609 if (nfhp == NULL) 1610 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1611 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, 1612 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1613 NULL); 1614 if (nfhp != NULL) 1615 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1616 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); 1617 } 1618 if (dattrflag) 1619 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1620 if (!error) { 1621 newvp = NFSTOV(np); 1622 if (attrflag == 0) 1623 error = nfsrpc_getattr(newvp, cnp->cn_cred, 1624 cnp->cn_thread, &nfsva, NULL); 1625 if (error == 0) 1626 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1627 0, 1); 1628 } 1629 if (error) { 1630 if (newvp != NULL) { 1631 vput(newvp); 1632 newvp = NULL; 1633 } 1634 if (NFS_ISV34(dvp) && (fmode & O_EXCL) && 1635 error == NFSERR_NOTSUPP) { 1636 fmode &= ~O_EXCL; 1637 goto again; 1638 } 1639 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) { 1640 if (nfscl_checksattr(vap, &nfsva)) { 1641 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred, 1642 cnp->cn_thread, &nfsva, &attrflag, NULL); 1643 if (error && (vap->va_uid != (uid_t)VNOVAL || 1644 vap->va_gid != (gid_t)VNOVAL)) { 1645 /* try again without setting uid/gid */ 1646 vap->va_uid = (uid_t)VNOVAL; 1647 vap->va_gid = (uid_t)VNOVAL; 1648 error = nfsrpc_setattr(newvp, vap, NULL, 1649 cnp->cn_cred, cnp->cn_thread, &nfsva, 1650 &attrflag, NULL); 1651 } 1652 if (attrflag) 1653 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 1654 NULL, 0, 1); 1655 if (error != 0) 1656 vput(newvp); 1657 } 1658 } 1659 if (!error) { 1660 if ((cnp->cn_flags & MAKEENTRY) && attrflag) 1661 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 1662 NULL); 1663 *ap->a_vpp = newvp; 1664 } else if (NFS_ISV4(dvp)) { 1665 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, 1666 vap->va_gid); 1667 } 1668 mtx_lock(&dnp->n_mtx); 1669 dnp->n_flag |= NMODIFIED; 1670 if (!dattrflag) { 1671 dnp->n_attrstamp = 0; 1672 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1673 } 1674 mtx_unlock(&dnp->n_mtx); 1675 return (error); 1676 } 1677 1678 /* 1679 * nfs file remove call 1680 * To try and make nfs semantics closer to ufs semantics, a file that has 1681 * other processes using the vnode is renamed instead of removed and then 1682 * removed later on the last close. 1683 * - If v_usecount > 1 1684 * If a rename is not already in the works 1685 * call nfs_sillyrename() to set it up 1686 * else 1687 * do the remove rpc 1688 */ 1689 static int 1690 nfs_remove(struct vop_remove_args *ap) 1691 { 1692 struct vnode *vp = ap->a_vp; 1693 struct vnode *dvp = ap->a_dvp; 1694 struct componentname *cnp = ap->a_cnp; 1695 struct nfsnode *np = VTONFS(vp); 1696 int error = 0; 1697 struct vattr vattr; 1698 1699 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name")); 1700 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount")); 1701 if (vp->v_type == VDIR) 1702 error = EPERM; 1703 else if (vrefcnt(vp) == 1 || (np->n_sillyrename && 1704 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 1705 vattr.va_nlink > 1)) { 1706 /* 1707 * Purge the name cache so that the chance of a lookup for 1708 * the name succeeding while the remove is in progress is 1709 * minimized. Without node locking it can still happen, such 1710 * that an I/O op returns ESTALE, but since you get this if 1711 * another host removes the file.. 1712 */ 1713 cache_purge(vp); 1714 /* 1715 * throw away biocache buffers, mainly to avoid 1716 * unnecessary delayed writes later. 1717 */ 1718 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1); 1719 if (error != EINTR && error != EIO) 1720 /* Do the rpc */ 1721 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr, 1722 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread); 1723 /* 1724 * Kludge City: If the first reply to the remove rpc is lost.. 1725 * the reply to the retransmitted request will be ENOENT 1726 * since the file was in fact removed 1727 * Therefore, we cheat and return success. 1728 */ 1729 if (error == ENOENT) 1730 error = 0; 1731 } else if (!np->n_sillyrename) 1732 error = nfs_sillyrename(dvp, vp, cnp); 1733 mtx_lock(&np->n_mtx); 1734 np->n_attrstamp = 0; 1735 mtx_unlock(&np->n_mtx); 1736 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 1737 return (error); 1738 } 1739 1740 /* 1741 * nfs file remove rpc called from nfs_inactive 1742 */ 1743 int 1744 ncl_removeit(struct sillyrename *sp, struct vnode *vp) 1745 { 1746 /* 1747 * Make sure that the directory vnode is still valid. 1748 * XXX we should lock sp->s_dvp here. 1749 */ 1750 if (sp->s_dvp->v_type == VBAD) 1751 return (0); 1752 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen, 1753 sp->s_cred, NULL)); 1754 } 1755 1756 /* 1757 * Nfs remove rpc, called from nfs_remove() and ncl_removeit(). 1758 */ 1759 static int 1760 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 1761 int namelen, struct ucred *cred, struct thread *td) 1762 { 1763 struct nfsvattr dnfsva; 1764 struct nfsnode *dnp = VTONFS(dvp); 1765 int error = 0, dattrflag; 1766 1767 mtx_lock(&dnp->n_mtx); 1768 dnp->n_flag |= NREMOVEINPROG; 1769 mtx_unlock(&dnp->n_mtx); 1770 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva, 1771 &dattrflag, NULL); 1772 mtx_lock(&dnp->n_mtx); 1773 if ((dnp->n_flag & NREMOVEWANT)) { 1774 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG); 1775 mtx_unlock(&dnp->n_mtx); 1776 wakeup((caddr_t)dnp); 1777 } else { 1778 dnp->n_flag &= ~NREMOVEINPROG; 1779 mtx_unlock(&dnp->n_mtx); 1780 } 1781 if (dattrflag) 1782 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1783 mtx_lock(&dnp->n_mtx); 1784 dnp->n_flag |= NMODIFIED; 1785 if (!dattrflag) { 1786 dnp->n_attrstamp = 0; 1787 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1788 } 1789 mtx_unlock(&dnp->n_mtx); 1790 if (error && NFS_ISV4(dvp)) 1791 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1792 return (error); 1793 } 1794 1795 /* 1796 * nfs file rename call 1797 */ 1798 static int 1799 nfs_rename(struct vop_rename_args *ap) 1800 { 1801 struct vnode *fvp = ap->a_fvp; 1802 struct vnode *tvp = ap->a_tvp; 1803 struct vnode *fdvp = ap->a_fdvp; 1804 struct vnode *tdvp = ap->a_tdvp; 1805 struct componentname *tcnp = ap->a_tcnp; 1806 struct componentname *fcnp = ap->a_fcnp; 1807 struct nfsnode *fnp = VTONFS(ap->a_fvp); 1808 struct nfsnode *tdnp = VTONFS(ap->a_tdvp); 1809 struct nfsv4node *newv4 = NULL; 1810 int error; 1811 1812 KASSERT((tcnp->cn_flags & HASBUF) != 0 && 1813 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name")); 1814 /* Check for cross-device rename */ 1815 if ((fvp->v_mount != tdvp->v_mount) || 1816 (tvp && (fvp->v_mount != tvp->v_mount))) { 1817 error = EXDEV; 1818 goto out; 1819 } 1820 1821 if (fvp == tvp) { 1822 printf("nfs_rename: fvp == tvp (can't happen)\n"); 1823 error = 0; 1824 goto out; 1825 } 1826 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0) 1827 goto out; 1828 1829 /* 1830 * We have to flush B_DELWRI data prior to renaming 1831 * the file. If we don't, the delayed-write buffers 1832 * can be flushed out later after the file has gone stale 1833 * under NFSV3. NFSV2 does not have this problem because 1834 * ( as far as I can tell ) it flushes dirty buffers more 1835 * often. 1836 * 1837 * Skip the rename operation if the fsync fails, this can happen 1838 * due to the server's volume being full, when we pushed out data 1839 * that was written back to our cache earlier. Not checking for 1840 * this condition can result in potential (silent) data loss. 1841 */ 1842 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread); 1843 NFSVOPUNLOCK(fvp, 0); 1844 if (!error && tvp) 1845 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread); 1846 if (error) 1847 goto out; 1848 1849 /* 1850 * If the tvp exists and is in use, sillyrename it before doing the 1851 * rename of the new file over it. 1852 * XXX Can't sillyrename a directory. 1853 */ 1854 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename && 1855 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1856 vput(tvp); 1857 tvp = NULL; 1858 } 1859 1860 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1861 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1862 tcnp->cn_thread); 1863 1864 if (error == 0 && NFS_ISV4(tdvp)) { 1865 /* 1866 * For NFSv4, check to see if it is the same name and 1867 * replace the name, if it is different. 1868 */ 1869 newv4 = malloc( 1870 sizeof (struct nfsv4node) + 1871 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1, 1872 M_NFSV4NODE, M_WAITOK); 1873 mtx_lock(&tdnp->n_mtx); 1874 mtx_lock(&fnp->n_mtx); 1875 if (fnp->n_v4 != NULL && fvp->v_type == VREG && 1876 (fnp->n_v4->n4_namelen != tcnp->cn_namelen || 1877 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4), 1878 tcnp->cn_namelen) || 1879 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen || 1880 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 1881 tdnp->n_fhp->nfh_len))) { 1882 #ifdef notdef 1883 { char nnn[100]; int nnnl; 1884 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99; 1885 bcopy(tcnp->cn_nameptr, nnn, nnnl); 1886 nnn[nnnl] = '\0'; 1887 printf("ren replace=%s\n",nnn); 1888 } 1889 #endif 1890 free(fnp->n_v4, M_NFSV4NODE); 1891 fnp->n_v4 = newv4; 1892 newv4 = NULL; 1893 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len; 1894 fnp->n_v4->n4_namelen = tcnp->cn_namelen; 1895 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 1896 tdnp->n_fhp->nfh_len); 1897 NFSBCOPY(tcnp->cn_nameptr, 1898 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen); 1899 } 1900 mtx_unlock(&tdnp->n_mtx); 1901 mtx_unlock(&fnp->n_mtx); 1902 if (newv4 != NULL) 1903 free(newv4, M_NFSV4NODE); 1904 } 1905 1906 if (fvp->v_type == VDIR) { 1907 if (tvp != NULL && tvp->v_type == VDIR) 1908 cache_purge(tdvp); 1909 cache_purge(fdvp); 1910 } 1911 1912 out: 1913 if (tdvp == tvp) 1914 vrele(tdvp); 1915 else 1916 vput(tdvp); 1917 if (tvp) 1918 vput(tvp); 1919 vrele(fdvp); 1920 vrele(fvp); 1921 /* 1922 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1923 */ 1924 if (error == ENOENT) 1925 error = 0; 1926 return (error); 1927 } 1928 1929 /* 1930 * nfs file rename rpc called from nfs_remove() above 1931 */ 1932 static int 1933 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp, 1934 struct sillyrename *sp) 1935 { 1936 1937 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen, 1938 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred, 1939 scnp->cn_thread)); 1940 } 1941 1942 /* 1943 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1944 */ 1945 static int 1946 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr, 1947 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr, 1948 int tnamelen, struct ucred *cred, struct thread *td) 1949 { 1950 struct nfsvattr fnfsva, tnfsva; 1951 struct nfsnode *fdnp = VTONFS(fdvp); 1952 struct nfsnode *tdnp = VTONFS(tdvp); 1953 int error = 0, fattrflag, tattrflag; 1954 1955 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp, 1956 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag, 1957 &tattrflag, NULL, NULL); 1958 mtx_lock(&fdnp->n_mtx); 1959 fdnp->n_flag |= NMODIFIED; 1960 if (fattrflag != 0) { 1961 mtx_unlock(&fdnp->n_mtx); 1962 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1); 1963 } else { 1964 fdnp->n_attrstamp = 0; 1965 mtx_unlock(&fdnp->n_mtx); 1966 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp); 1967 } 1968 mtx_lock(&tdnp->n_mtx); 1969 tdnp->n_flag |= NMODIFIED; 1970 if (tattrflag != 0) { 1971 mtx_unlock(&tdnp->n_mtx); 1972 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1); 1973 } else { 1974 tdnp->n_attrstamp = 0; 1975 mtx_unlock(&tdnp->n_mtx); 1976 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 1977 } 1978 if (error && NFS_ISV4(fdvp)) 1979 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1980 return (error); 1981 } 1982 1983 /* 1984 * nfs hard link create call 1985 */ 1986 static int 1987 nfs_link(struct vop_link_args *ap) 1988 { 1989 struct vnode *vp = ap->a_vp; 1990 struct vnode *tdvp = ap->a_tdvp; 1991 struct componentname *cnp = ap->a_cnp; 1992 struct nfsnode *np, *tdnp; 1993 struct nfsvattr nfsva, dnfsva; 1994 int error = 0, attrflag, dattrflag; 1995 1996 /* 1997 * Push all writes to the server, so that the attribute cache 1998 * doesn't get "out of sync" with the server. 1999 * XXX There should be a better way! 2000 */ 2001 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread); 2002 2003 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 2004 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag, 2005 &dattrflag, NULL); 2006 tdnp = VTONFS(tdvp); 2007 mtx_lock(&tdnp->n_mtx); 2008 tdnp->n_flag |= NMODIFIED; 2009 if (dattrflag != 0) { 2010 mtx_unlock(&tdnp->n_mtx); 2011 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1); 2012 } else { 2013 tdnp->n_attrstamp = 0; 2014 mtx_unlock(&tdnp->n_mtx); 2015 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 2016 } 2017 if (attrflag) 2018 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2019 else { 2020 np = VTONFS(vp); 2021 mtx_lock(&np->n_mtx); 2022 np->n_attrstamp = 0; 2023 mtx_unlock(&np->n_mtx); 2024 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 2025 } 2026 /* 2027 * If negative lookup caching is enabled, I might as well 2028 * add an entry for this node. Not necessary for correctness, 2029 * but if negative caching is enabled, then the system 2030 * must care about lookup caching hit rate, so... 2031 */ 2032 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 && 2033 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 2034 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL); 2035 } 2036 if (error && NFS_ISV4(vp)) 2037 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, 2038 (gid_t)0); 2039 return (error); 2040 } 2041 2042 /* 2043 * nfs symbolic link create call 2044 */ 2045 static int 2046 nfs_symlink(struct vop_symlink_args *ap) 2047 { 2048 struct vnode *dvp = ap->a_dvp; 2049 struct vattr *vap = ap->a_vap; 2050 struct componentname *cnp = ap->a_cnp; 2051 struct nfsvattr nfsva, dnfsva; 2052 struct nfsfh *nfhp; 2053 struct nfsnode *np = NULL, *dnp; 2054 struct vnode *newvp = NULL; 2055 int error = 0, attrflag, dattrflag, ret; 2056 2057 vap->va_type = VLNK; 2058 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2059 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, 2060 &nfsva, &nfhp, &attrflag, &dattrflag, NULL); 2061 if (nfhp) { 2062 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, 2063 &np, NULL, LK_EXCLUSIVE); 2064 if (!ret) 2065 newvp = NFSTOV(np); 2066 else if (!error) 2067 error = ret; 2068 } 2069 if (newvp != NULL) { 2070 if (attrflag) 2071 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 2072 0, 1); 2073 } else if (!error) { 2074 /* 2075 * If we do not have an error and we could not extract the 2076 * newvp from the response due to the request being NFSv2, we 2077 * have to do a lookup in order to obtain a newvp to return. 2078 */ 2079 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2080 cnp->cn_cred, cnp->cn_thread, &np); 2081 if (!error) 2082 newvp = NFSTOV(np); 2083 } 2084 if (error) { 2085 if (newvp) 2086 vput(newvp); 2087 if (NFS_ISV4(dvp)) 2088 error = nfscl_maperr(cnp->cn_thread, error, 2089 vap->va_uid, vap->va_gid); 2090 } else { 2091 *ap->a_vpp = newvp; 2092 } 2093 2094 dnp = VTONFS(dvp); 2095 mtx_lock(&dnp->n_mtx); 2096 dnp->n_flag |= NMODIFIED; 2097 if (dattrflag != 0) { 2098 mtx_unlock(&dnp->n_mtx); 2099 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2100 } else { 2101 dnp->n_attrstamp = 0; 2102 mtx_unlock(&dnp->n_mtx); 2103 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2104 } 2105 /* 2106 * If negative lookup caching is enabled, I might as well 2107 * add an entry for this node. Not necessary for correctness, 2108 * but if negative caching is enabled, then the system 2109 * must care about lookup caching hit rate, so... 2110 */ 2111 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2112 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 2113 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL); 2114 } 2115 return (error); 2116 } 2117 2118 /* 2119 * nfs make dir call 2120 */ 2121 static int 2122 nfs_mkdir(struct vop_mkdir_args *ap) 2123 { 2124 struct vnode *dvp = ap->a_dvp; 2125 struct vattr *vap = ap->a_vap; 2126 struct componentname *cnp = ap->a_cnp; 2127 struct nfsnode *np = NULL, *dnp; 2128 struct vnode *newvp = NULL; 2129 struct vattr vattr; 2130 struct nfsfh *nfhp; 2131 struct nfsvattr nfsva, dnfsva; 2132 int error = 0, attrflag, dattrflag, ret; 2133 2134 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) 2135 return (error); 2136 vap->va_type = VDIR; 2137 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2138 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp, 2139 &attrflag, &dattrflag, NULL); 2140 dnp = VTONFS(dvp); 2141 mtx_lock(&dnp->n_mtx); 2142 dnp->n_flag |= NMODIFIED; 2143 if (dattrflag != 0) { 2144 mtx_unlock(&dnp->n_mtx); 2145 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2146 } else { 2147 dnp->n_attrstamp = 0; 2148 mtx_unlock(&dnp->n_mtx); 2149 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2150 } 2151 if (nfhp) { 2152 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, 2153 &np, NULL, LK_EXCLUSIVE); 2154 if (!ret) { 2155 newvp = NFSTOV(np); 2156 if (attrflag) 2157 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 2158 NULL, 0, 1); 2159 } else if (!error) 2160 error = ret; 2161 } 2162 if (!error && newvp == NULL) { 2163 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2164 cnp->cn_cred, cnp->cn_thread, &np); 2165 if (!error) { 2166 newvp = NFSTOV(np); 2167 if (newvp->v_type != VDIR) 2168 error = EEXIST; 2169 } 2170 } 2171 if (error) { 2172 if (newvp) 2173 vput(newvp); 2174 if (NFS_ISV4(dvp)) 2175 error = nfscl_maperr(cnp->cn_thread, error, 2176 vap->va_uid, vap->va_gid); 2177 } else { 2178 /* 2179 * If negative lookup caching is enabled, I might as well 2180 * add an entry for this node. Not necessary for correctness, 2181 * but if negative caching is enabled, then the system 2182 * must care about lookup caching hit rate, so... 2183 */ 2184 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2185 (cnp->cn_flags & MAKEENTRY) && 2186 attrflag != 0 && dattrflag != 0) 2187 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 2188 &dnfsva.na_ctime); 2189 *ap->a_vpp = newvp; 2190 } 2191 return (error); 2192 } 2193 2194 /* 2195 * nfs remove directory call 2196 */ 2197 static int 2198 nfs_rmdir(struct vop_rmdir_args *ap) 2199 { 2200 struct vnode *vp = ap->a_vp; 2201 struct vnode *dvp = ap->a_dvp; 2202 struct componentname *cnp = ap->a_cnp; 2203 struct nfsnode *dnp; 2204 struct nfsvattr dnfsva; 2205 int error, dattrflag; 2206 2207 if (dvp == vp) 2208 return (EINVAL); 2209 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2210 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL); 2211 dnp = VTONFS(dvp); 2212 mtx_lock(&dnp->n_mtx); 2213 dnp->n_flag |= NMODIFIED; 2214 if (dattrflag != 0) { 2215 mtx_unlock(&dnp->n_mtx); 2216 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2217 } else { 2218 dnp->n_attrstamp = 0; 2219 mtx_unlock(&dnp->n_mtx); 2220 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2221 } 2222 2223 cache_purge(dvp); 2224 cache_purge(vp); 2225 if (error && NFS_ISV4(dvp)) 2226 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, 2227 (gid_t)0); 2228 /* 2229 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2230 */ 2231 if (error == ENOENT) 2232 error = 0; 2233 return (error); 2234 } 2235 2236 /* 2237 * nfs readdir call 2238 */ 2239 static int 2240 nfs_readdir(struct vop_readdir_args *ap) 2241 { 2242 struct vnode *vp = ap->a_vp; 2243 struct nfsnode *np = VTONFS(vp); 2244 struct uio *uio = ap->a_uio; 2245 ssize_t tresid, left; 2246 int error = 0; 2247 struct vattr vattr; 2248 2249 if (ap->a_eofflag != NULL) 2250 *ap->a_eofflag = 0; 2251 if (vp->v_type != VDIR) 2252 return(EPERM); 2253 2254 /* 2255 * First, check for hit on the EOF offset cache 2256 */ 2257 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2258 (np->n_flag & NMODIFIED) == 0) { 2259 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) { 2260 mtx_lock(&np->n_mtx); 2261 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) || 2262 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 2263 mtx_unlock(&np->n_mtx); 2264 NFSINCRGLOBAL(nfsstatsv1.direofcache_hits); 2265 if (ap->a_eofflag != NULL) 2266 *ap->a_eofflag = 1; 2267 return (0); 2268 } else 2269 mtx_unlock(&np->n_mtx); 2270 } 2271 } 2272 2273 /* 2274 * NFS always guarantees that directory entries don't straddle 2275 * DIRBLKSIZ boundaries. As such, we need to limit the size 2276 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial 2277 * directory entry. 2278 */ 2279 left = uio->uio_resid % DIRBLKSIZ; 2280 if (left == uio->uio_resid) 2281 return (EINVAL); 2282 uio->uio_resid -= left; 2283 2284 /* 2285 * Call ncl_bioread() to do the real work. 2286 */ 2287 tresid = uio->uio_resid; 2288 error = ncl_bioread(vp, uio, 0, ap->a_cred); 2289 2290 if (!error && uio->uio_resid == tresid) { 2291 NFSINCRGLOBAL(nfsstatsv1.direofcache_misses); 2292 if (ap->a_eofflag != NULL) 2293 *ap->a_eofflag = 1; 2294 } 2295 2296 /* Add the partial DIRBLKSIZ (left) back in. */ 2297 uio->uio_resid += left; 2298 return (error); 2299 } 2300 2301 /* 2302 * Readdir rpc call. 2303 * Called from below the buffer cache by ncl_doio(). 2304 */ 2305 int 2306 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2307 struct thread *td) 2308 { 2309 struct nfsvattr nfsva; 2310 nfsuint64 *cookiep, cookie; 2311 struct nfsnode *dnp = VTONFS(vp); 2312 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2313 int error = 0, eof, attrflag; 2314 2315 KASSERT(uiop->uio_iovcnt == 1 && 2316 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2317 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2318 ("nfs readdirrpc bad uio")); 2319 2320 /* 2321 * If there is no cookie, assume directory was stale. 2322 */ 2323 ncl_dircookie_lock(dnp); 2324 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2325 if (cookiep) { 2326 cookie = *cookiep; 2327 ncl_dircookie_unlock(dnp); 2328 } else { 2329 ncl_dircookie_unlock(dnp); 2330 return (NFSERR_BAD_COOKIE); 2331 } 2332 2333 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2334 (void)ncl_fsinfo(nmp, vp, cred, td); 2335 2336 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva, 2337 &attrflag, &eof, NULL); 2338 if (attrflag) 2339 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2340 2341 if (!error) { 2342 /* 2343 * We are now either at the end of the directory or have filled 2344 * the block. 2345 */ 2346 if (eof) 2347 dnp->n_direofoffset = uiop->uio_offset; 2348 else { 2349 if (uiop->uio_resid > 0) 2350 printf("EEK! readdirrpc resid > 0\n"); 2351 ncl_dircookie_lock(dnp); 2352 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2353 *cookiep = cookie; 2354 ncl_dircookie_unlock(dnp); 2355 } 2356 } else if (NFS_ISV4(vp)) { 2357 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2358 } 2359 return (error); 2360 } 2361 2362 /* 2363 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc(). 2364 */ 2365 int 2366 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2367 struct thread *td) 2368 { 2369 struct nfsvattr nfsva; 2370 nfsuint64 *cookiep, cookie; 2371 struct nfsnode *dnp = VTONFS(vp); 2372 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2373 int error = 0, attrflag, eof; 2374 2375 KASSERT(uiop->uio_iovcnt == 1 && 2376 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2377 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2378 ("nfs readdirplusrpc bad uio")); 2379 2380 /* 2381 * If there is no cookie, assume directory was stale. 2382 */ 2383 ncl_dircookie_lock(dnp); 2384 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2385 if (cookiep) { 2386 cookie = *cookiep; 2387 ncl_dircookie_unlock(dnp); 2388 } else { 2389 ncl_dircookie_unlock(dnp); 2390 return (NFSERR_BAD_COOKIE); 2391 } 2392 2393 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2394 (void)ncl_fsinfo(nmp, vp, cred, td); 2395 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva, 2396 &attrflag, &eof, NULL); 2397 if (attrflag) 2398 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2399 2400 if (!error) { 2401 /* 2402 * We are now either at end of the directory or have filled the 2403 * the block. 2404 */ 2405 if (eof) 2406 dnp->n_direofoffset = uiop->uio_offset; 2407 else { 2408 if (uiop->uio_resid > 0) 2409 printf("EEK! readdirplusrpc resid > 0\n"); 2410 ncl_dircookie_lock(dnp); 2411 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2412 *cookiep = cookie; 2413 ncl_dircookie_unlock(dnp); 2414 } 2415 } else if (NFS_ISV4(vp)) { 2416 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2417 } 2418 return (error); 2419 } 2420 2421 /* 2422 * Silly rename. To make the NFS filesystem that is stateless look a little 2423 * more like the "ufs" a remove of an active vnode is translated to a rename 2424 * to a funny looking filename that is removed by nfs_inactive on the 2425 * nfsnode. There is the potential for another process on a different client 2426 * to create the same funny name between the nfs_lookitup() fails and the 2427 * nfs_rename() completes, but... 2428 */ 2429 static int 2430 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2431 { 2432 struct sillyrename *sp; 2433 struct nfsnode *np; 2434 int error; 2435 short pid; 2436 unsigned int lticks; 2437 2438 cache_purge(dvp); 2439 np = VTONFS(vp); 2440 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir")); 2441 sp = malloc(sizeof (struct sillyrename), 2442 M_NEWNFSREQ, M_WAITOK); 2443 sp->s_cred = crhold(cnp->cn_cred); 2444 sp->s_dvp = dvp; 2445 VREF(dvp); 2446 2447 /* 2448 * Fudge together a funny name. 2449 * Changing the format of the funny name to accommodate more 2450 * sillynames per directory. 2451 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is 2452 * CPU ticks since boot. 2453 */ 2454 pid = cnp->cn_thread->td_proc->p_pid; 2455 lticks = (unsigned int)ticks; 2456 for ( ; ; ) { 2457 sp->s_namlen = sprintf(sp->s_name, 2458 ".nfs.%08x.%04x4.4", lticks, 2459 pid); 2460 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2461 cnp->cn_thread, NULL)) 2462 break; 2463 lticks++; 2464 } 2465 error = nfs_renameit(dvp, vp, cnp, sp); 2466 if (error) 2467 goto bad; 2468 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2469 cnp->cn_thread, &np); 2470 np->n_sillyrename = sp; 2471 return (0); 2472 bad: 2473 vrele(sp->s_dvp); 2474 crfree(sp->s_cred); 2475 free(sp, M_NEWNFSREQ); 2476 return (error); 2477 } 2478 2479 /* 2480 * Look up a file name and optionally either update the file handle or 2481 * allocate an nfsnode, depending on the value of npp. 2482 * npp == NULL --> just do the lookup 2483 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2484 * handled too 2485 * *npp != NULL --> update the file handle in the vnode 2486 */ 2487 static int 2488 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred, 2489 struct thread *td, struct nfsnode **npp) 2490 { 2491 struct vnode *newvp = NULL, *vp; 2492 struct nfsnode *np, *dnp = VTONFS(dvp); 2493 struct nfsfh *nfhp, *onfhp; 2494 struct nfsvattr nfsva, dnfsva; 2495 struct componentname cn; 2496 int error = 0, attrflag, dattrflag; 2497 u_int hash; 2498 2499 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva, 2500 &nfhp, &attrflag, &dattrflag, NULL); 2501 if (dattrflag) 2502 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2503 if (npp && !error) { 2504 if (*npp != NULL) { 2505 np = *npp; 2506 vp = NFSTOV(np); 2507 /* 2508 * For NFSv4, check to see if it is the same name and 2509 * replace the name, if it is different. 2510 */ 2511 if (np->n_v4 != NULL && nfsva.na_type == VREG && 2512 (np->n_v4->n4_namelen != len || 2513 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) || 2514 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || 2515 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2516 dnp->n_fhp->nfh_len))) { 2517 #ifdef notdef 2518 { char nnn[100]; int nnnl; 2519 nnnl = (len < 100) ? len : 99; 2520 bcopy(name, nnn, nnnl); 2521 nnn[nnnl] = '\0'; 2522 printf("replace=%s\n",nnn); 2523 } 2524 #endif 2525 free(np->n_v4, M_NFSV4NODE); 2526 np->n_v4 = malloc( 2527 sizeof (struct nfsv4node) + 2528 dnp->n_fhp->nfh_len + len - 1, 2529 M_NFSV4NODE, M_WAITOK); 2530 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; 2531 np->n_v4->n4_namelen = len; 2532 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2533 dnp->n_fhp->nfh_len); 2534 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len); 2535 } 2536 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, 2537 FNV1_32_INIT); 2538 onfhp = np->n_fhp; 2539 /* 2540 * Rehash node for new file handle. 2541 */ 2542 vfs_hash_rehash(vp, hash); 2543 np->n_fhp = nfhp; 2544 if (onfhp != NULL) 2545 free(onfhp, M_NFSFH); 2546 newvp = NFSTOV(np); 2547 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) { 2548 free(nfhp, M_NFSFH); 2549 VREF(dvp); 2550 newvp = dvp; 2551 } else { 2552 cn.cn_nameptr = name; 2553 cn.cn_namelen = len; 2554 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td, 2555 &np, NULL, LK_EXCLUSIVE); 2556 if (error) 2557 return (error); 2558 newvp = NFSTOV(np); 2559 } 2560 if (!attrflag && *npp == NULL) { 2561 if (newvp == dvp) 2562 vrele(newvp); 2563 else 2564 vput(newvp); 2565 return (ENOENT); 2566 } 2567 if (attrflag) 2568 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 2569 0, 1); 2570 } 2571 if (npp && *npp == NULL) { 2572 if (error) { 2573 if (newvp) { 2574 if (newvp == dvp) 2575 vrele(newvp); 2576 else 2577 vput(newvp); 2578 } 2579 } else 2580 *npp = np; 2581 } 2582 if (error && NFS_ISV4(dvp)) 2583 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2584 return (error); 2585 } 2586 2587 /* 2588 * Nfs Version 3 and 4 commit rpc 2589 */ 2590 int 2591 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred, 2592 struct thread *td) 2593 { 2594 struct nfsvattr nfsva; 2595 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2596 struct nfsnode *np; 2597 struct uio uio; 2598 int error, attrflag; 2599 2600 np = VTONFS(vp); 2601 error = EIO; 2602 attrflag = 0; 2603 if (NFSHASPNFS(nmp) && (np->n_flag & NDSCOMMIT) != 0) { 2604 uio.uio_offset = offset; 2605 uio.uio_resid = cnt; 2606 error = nfscl_doiods(vp, &uio, NULL, NULL, 2607 NFSV4OPEN_ACCESSWRITE, 1, cred, td); 2608 if (error != 0) { 2609 mtx_lock(&np->n_mtx); 2610 np->n_flag &= ~NDSCOMMIT; 2611 mtx_unlock(&np->n_mtx); 2612 } 2613 } 2614 if (error != 0) { 2615 mtx_lock(&nmp->nm_mtx); 2616 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) { 2617 mtx_unlock(&nmp->nm_mtx); 2618 return (0); 2619 } 2620 mtx_unlock(&nmp->nm_mtx); 2621 error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva, 2622 &attrflag, NULL); 2623 } 2624 if (attrflag != 0) 2625 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 2626 0, 1); 2627 if (error != 0 && NFS_ISV4(vp)) 2628 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2629 return (error); 2630 } 2631 2632 /* 2633 * Strategy routine. 2634 * For async requests when nfsiod(s) are running, queue the request by 2635 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the 2636 * request. 2637 */ 2638 static int 2639 nfs_strategy(struct vop_strategy_args *ap) 2640 { 2641 struct buf *bp; 2642 struct vnode *vp; 2643 struct ucred *cr; 2644 2645 bp = ap->a_bp; 2646 vp = ap->a_vp; 2647 KASSERT(bp->b_vp == vp, ("missing b_getvp")); 2648 KASSERT(!(bp->b_flags & B_DONE), 2649 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); 2650 BUF_ASSERT_HELD(bp); 2651 2652 if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno) 2653 bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize / 2654 DEV_BSIZE); 2655 if (bp->b_iocmd == BIO_READ) 2656 cr = bp->b_rcred; 2657 else 2658 cr = bp->b_wcred; 2659 2660 /* 2661 * If the op is asynchronous and an i/o daemon is waiting 2662 * queue the request, wake it up and wait for completion 2663 * otherwise just do it ourselves. 2664 */ 2665 if ((bp->b_flags & B_ASYNC) == 0 || 2666 ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread)) 2667 (void) ncl_doio(vp, bp, cr, curthread, 1); 2668 return (0); 2669 } 2670 2671 /* 2672 * fsync vnode op. Just call ncl_flush() with commit == 1. 2673 */ 2674 /* ARGSUSED */ 2675 static int 2676 nfs_fsync(struct vop_fsync_args *ap) 2677 { 2678 2679 if (ap->a_vp->v_type != VREG) { 2680 /* 2681 * For NFS, metadata is changed synchronously on the server, 2682 * so there is nothing to flush. Also, ncl_flush() clears 2683 * the NMODIFIED flag and that shouldn't be done here for 2684 * directories. 2685 */ 2686 return (0); 2687 } 2688 return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0)); 2689 } 2690 2691 /* 2692 * Flush all the blocks associated with a vnode. 2693 * Walk through the buffer pool and push any dirty pages 2694 * associated with the vnode. 2695 * If the called_from_renewthread argument is TRUE, it has been called 2696 * from the NFSv4 renew thread and, as such, cannot block indefinitely 2697 * waiting for a buffer write to complete. 2698 */ 2699 int 2700 ncl_flush(struct vnode *vp, int waitfor, struct thread *td, 2701 int commit, int called_from_renewthread) 2702 { 2703 struct nfsnode *np = VTONFS(vp); 2704 struct buf *bp; 2705 int i; 2706 struct buf *nbp; 2707 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2708 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos; 2709 int passone = 1, trycnt = 0; 2710 u_quad_t off, endoff, toff; 2711 struct ucred* wcred = NULL; 2712 struct buf **bvec = NULL; 2713 struct bufobj *bo; 2714 #ifndef NFS_COMMITBVECSIZ 2715 #define NFS_COMMITBVECSIZ 20 2716 #endif 2717 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ]; 2718 u_int bvecsize = 0, bveccount; 2719 2720 if (called_from_renewthread != 0) 2721 slptimeo = hz; 2722 if (nmp->nm_flag & NFSMNT_INT) 2723 slpflag = PCATCH; 2724 if (!commit) 2725 passone = 0; 2726 bo = &vp->v_bufobj; 2727 /* 2728 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the 2729 * server, but has not been committed to stable storage on the server 2730 * yet. On the first pass, the byte range is worked out and the commit 2731 * rpc is done. On the second pass, ncl_writebp() is called to do the 2732 * job. 2733 */ 2734 again: 2735 off = (u_quad_t)-1; 2736 endoff = 0; 2737 bvecpos = 0; 2738 if (NFS_ISV34(vp) && commit) { 2739 if (bvec != NULL && bvec != bvec_on_stack) 2740 free(bvec, M_TEMP); 2741 /* 2742 * Count up how many buffers waiting for a commit. 2743 */ 2744 bveccount = 0; 2745 BO_LOCK(bo); 2746 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2747 if (!BUF_ISLOCKED(bp) && 2748 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) 2749 == (B_DELWRI | B_NEEDCOMMIT)) 2750 bveccount++; 2751 } 2752 /* 2753 * Allocate space to remember the list of bufs to commit. It is 2754 * important to use M_NOWAIT here to avoid a race with nfs_write. 2755 * If we can't get memory (for whatever reason), we will end up 2756 * committing the buffers one-by-one in the loop below. 2757 */ 2758 if (bveccount > NFS_COMMITBVECSIZ) { 2759 /* 2760 * Release the vnode interlock to avoid a lock 2761 * order reversal. 2762 */ 2763 BO_UNLOCK(bo); 2764 bvec = (struct buf **) 2765 malloc(bveccount * sizeof(struct buf *), 2766 M_TEMP, M_NOWAIT); 2767 BO_LOCK(bo); 2768 if (bvec == NULL) { 2769 bvec = bvec_on_stack; 2770 bvecsize = NFS_COMMITBVECSIZ; 2771 } else 2772 bvecsize = bveccount; 2773 } else { 2774 bvec = bvec_on_stack; 2775 bvecsize = NFS_COMMITBVECSIZ; 2776 } 2777 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2778 if (bvecpos >= bvecsize) 2779 break; 2780 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2781 nbp = TAILQ_NEXT(bp, b_bobufs); 2782 continue; 2783 } 2784 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 2785 (B_DELWRI | B_NEEDCOMMIT)) { 2786 BUF_UNLOCK(bp); 2787 nbp = TAILQ_NEXT(bp, b_bobufs); 2788 continue; 2789 } 2790 BO_UNLOCK(bo); 2791 bremfree(bp); 2792 /* 2793 * Work out if all buffers are using the same cred 2794 * so we can deal with them all with one commit. 2795 * 2796 * NOTE: we are not clearing B_DONE here, so we have 2797 * to do it later on in this routine if we intend to 2798 * initiate I/O on the bp. 2799 * 2800 * Note: to avoid loopback deadlocks, we do not 2801 * assign b_runningbufspace. 2802 */ 2803 if (wcred == NULL) 2804 wcred = bp->b_wcred; 2805 else if (wcred != bp->b_wcred) 2806 wcred = NOCRED; 2807 vfs_busy_pages(bp, 1); 2808 2809 BO_LOCK(bo); 2810 /* 2811 * bp is protected by being locked, but nbp is not 2812 * and vfs_busy_pages() may sleep. We have to 2813 * recalculate nbp. 2814 */ 2815 nbp = TAILQ_NEXT(bp, b_bobufs); 2816 2817 /* 2818 * A list of these buffers is kept so that the 2819 * second loop knows which buffers have actually 2820 * been committed. This is necessary, since there 2821 * may be a race between the commit rpc and new 2822 * uncommitted writes on the file. 2823 */ 2824 bvec[bvecpos++] = bp; 2825 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2826 bp->b_dirtyoff; 2827 if (toff < off) 2828 off = toff; 2829 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); 2830 if (toff > endoff) 2831 endoff = toff; 2832 } 2833 BO_UNLOCK(bo); 2834 } 2835 if (bvecpos > 0) { 2836 /* 2837 * Commit data on the server, as required. 2838 * If all bufs are using the same wcred, then use that with 2839 * one call for all of them, otherwise commit each one 2840 * separately. 2841 */ 2842 if (wcred != NOCRED) 2843 retv = ncl_commit(vp, off, (int)(endoff - off), 2844 wcred, td); 2845 else { 2846 retv = 0; 2847 for (i = 0; i < bvecpos; i++) { 2848 off_t off, size; 2849 bp = bvec[i]; 2850 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2851 bp->b_dirtyoff; 2852 size = (u_quad_t)(bp->b_dirtyend 2853 - bp->b_dirtyoff); 2854 retv = ncl_commit(vp, off, (int)size, 2855 bp->b_wcred, td); 2856 if (retv) break; 2857 } 2858 } 2859 2860 if (retv == NFSERR_STALEWRITEVERF) 2861 ncl_clearcommit(vp->v_mount); 2862 2863 /* 2864 * Now, either mark the blocks I/O done or mark the 2865 * blocks dirty, depending on whether the commit 2866 * succeeded. 2867 */ 2868 for (i = 0; i < bvecpos; i++) { 2869 bp = bvec[i]; 2870 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 2871 if (retv) { 2872 /* 2873 * Error, leave B_DELWRI intact 2874 */ 2875 vfs_unbusy_pages(bp); 2876 brelse(bp); 2877 } else { 2878 /* 2879 * Success, remove B_DELWRI ( bundirty() ). 2880 * 2881 * b_dirtyoff/b_dirtyend seem to be NFS 2882 * specific. We should probably move that 2883 * into bundirty(). XXX 2884 */ 2885 bufobj_wref(bo); 2886 bp->b_flags |= B_ASYNC; 2887 bundirty(bp); 2888 bp->b_flags &= ~B_DONE; 2889 bp->b_ioflags &= ~BIO_ERROR; 2890 bp->b_dirtyoff = bp->b_dirtyend = 0; 2891 bufdone(bp); 2892 } 2893 } 2894 } 2895 2896 /* 2897 * Start/do any write(s) that are required. 2898 */ 2899 loop: 2900 BO_LOCK(bo); 2901 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2902 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2903 if (waitfor != MNT_WAIT || passone) 2904 continue; 2905 2906 error = BUF_TIMELOCK(bp, 2907 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 2908 BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo); 2909 if (error == 0) { 2910 BUF_UNLOCK(bp); 2911 goto loop; 2912 } 2913 if (error == ENOLCK) { 2914 error = 0; 2915 goto loop; 2916 } 2917 if (called_from_renewthread != 0) { 2918 /* 2919 * Return EIO so the flush will be retried 2920 * later. 2921 */ 2922 error = EIO; 2923 goto done; 2924 } 2925 if (newnfs_sigintr(nmp, td)) { 2926 error = EINTR; 2927 goto done; 2928 } 2929 if (slpflag == PCATCH) { 2930 slpflag = 0; 2931 slptimeo = 2 * hz; 2932 } 2933 goto loop; 2934 } 2935 if ((bp->b_flags & B_DELWRI) == 0) 2936 panic("nfs_fsync: not dirty"); 2937 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) { 2938 BUF_UNLOCK(bp); 2939 continue; 2940 } 2941 BO_UNLOCK(bo); 2942 bremfree(bp); 2943 if (passone || !commit) 2944 bp->b_flags |= B_ASYNC; 2945 else 2946 bp->b_flags |= B_ASYNC; 2947 bwrite(bp); 2948 if (newnfs_sigintr(nmp, td)) { 2949 error = EINTR; 2950 goto done; 2951 } 2952 goto loop; 2953 } 2954 if (passone) { 2955 passone = 0; 2956 BO_UNLOCK(bo); 2957 goto again; 2958 } 2959 if (waitfor == MNT_WAIT) { 2960 while (bo->bo_numoutput) { 2961 error = bufobj_wwait(bo, slpflag, slptimeo); 2962 if (error) { 2963 BO_UNLOCK(bo); 2964 if (called_from_renewthread != 0) { 2965 /* 2966 * Return EIO so that the flush will be 2967 * retried later. 2968 */ 2969 error = EIO; 2970 goto done; 2971 } 2972 error = newnfs_sigintr(nmp, td); 2973 if (error) 2974 goto done; 2975 if (slpflag == PCATCH) { 2976 slpflag = 0; 2977 slptimeo = 2 * hz; 2978 } 2979 BO_LOCK(bo); 2980 } 2981 } 2982 if (bo->bo_dirty.bv_cnt != 0 && commit) { 2983 BO_UNLOCK(bo); 2984 goto loop; 2985 } 2986 /* 2987 * Wait for all the async IO requests to drain 2988 */ 2989 BO_UNLOCK(bo); 2990 mtx_lock(&np->n_mtx); 2991 while (np->n_directio_asyncwr > 0) { 2992 np->n_flag |= NFSYNCWAIT; 2993 error = newnfs_msleep(td, &np->n_directio_asyncwr, 2994 &np->n_mtx, slpflag | (PRIBIO + 1), 2995 "nfsfsync", 0); 2996 if (error) { 2997 if (newnfs_sigintr(nmp, td)) { 2998 mtx_unlock(&np->n_mtx); 2999 error = EINTR; 3000 goto done; 3001 } 3002 } 3003 } 3004 mtx_unlock(&np->n_mtx); 3005 } else 3006 BO_UNLOCK(bo); 3007 if (NFSHASPNFS(nmp)) { 3008 nfscl_layoutcommit(vp, td); 3009 /* 3010 * Invalidate the attribute cache, since writes to a DS 3011 * won't update the size attribute. 3012 */ 3013 mtx_lock(&np->n_mtx); 3014 np->n_attrstamp = 0; 3015 } else 3016 mtx_lock(&np->n_mtx); 3017 if (np->n_flag & NWRITEERR) { 3018 error = np->n_error; 3019 np->n_flag &= ~NWRITEERR; 3020 } 3021 if (commit && bo->bo_dirty.bv_cnt == 0 && 3022 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0) 3023 np->n_flag &= ~NMODIFIED; 3024 mtx_unlock(&np->n_mtx); 3025 done: 3026 if (bvec != NULL && bvec != bvec_on_stack) 3027 free(bvec, M_TEMP); 3028 if (error == 0 && commit != 0 && waitfor == MNT_WAIT && 3029 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 || 3030 np->n_directio_asyncwr != 0)) { 3031 if (trycnt++ < 5) { 3032 /* try, try again... */ 3033 passone = 1; 3034 wcred = NULL; 3035 bvec = NULL; 3036 bvecsize = 0; 3037 goto again; 3038 } 3039 vn_printf(vp, "ncl_flush failed"); 3040 error = called_from_renewthread != 0 ? EIO : EBUSY; 3041 } 3042 return (error); 3043 } 3044 3045 /* 3046 * NFS advisory byte-level locks. 3047 */ 3048 static int 3049 nfs_advlock(struct vop_advlock_args *ap) 3050 { 3051 struct vnode *vp = ap->a_vp; 3052 struct ucred *cred; 3053 struct nfsnode *np = VTONFS(ap->a_vp); 3054 struct proc *p = (struct proc *)ap->a_id; 3055 struct thread *td = curthread; /* XXX */ 3056 struct vattr va; 3057 int ret, error; 3058 u_quad_t size; 3059 3060 error = NFSVOPLOCK(vp, LK_SHARED); 3061 if (error != 0) 3062 return (EBADF); 3063 if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) { 3064 if (vp->v_type != VREG) { 3065 error = EINVAL; 3066 goto out; 3067 } 3068 if ((ap->a_flags & F_POSIX) != 0) 3069 cred = p->p_ucred; 3070 else 3071 cred = td->td_ucred; 3072 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 3073 if (vp->v_iflag & VI_DOOMED) { 3074 error = EBADF; 3075 goto out; 3076 } 3077 3078 /* 3079 * If this is unlocking a write locked region, flush and 3080 * commit them before unlocking. This is required by 3081 * RFC3530 Sec. 9.3.2. 3082 */ 3083 if (ap->a_op == F_UNLCK && 3084 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id, 3085 ap->a_flags)) 3086 (void) ncl_flush(vp, MNT_WAIT, td, 1, 0); 3087 3088 /* 3089 * Loop around doing the lock op, while a blocking lock 3090 * must wait for the lock op to succeed. 3091 */ 3092 do { 3093 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op, 3094 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags); 3095 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3096 ap->a_op == F_SETLK) { 3097 NFSVOPUNLOCK(vp, 0); 3098 error = nfs_catnap(PZERO | PCATCH, ret, 3099 "ncladvl"); 3100 if (error) 3101 return (EINTR); 3102 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 3103 if (vp->v_iflag & VI_DOOMED) { 3104 error = EBADF; 3105 goto out; 3106 } 3107 } 3108 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3109 ap->a_op == F_SETLK); 3110 if (ret == NFSERR_DENIED) { 3111 error = EAGAIN; 3112 goto out; 3113 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) { 3114 error = ret; 3115 goto out; 3116 } else if (ret != 0) { 3117 error = EACCES; 3118 goto out; 3119 } 3120 3121 /* 3122 * Now, if we just got a lock, invalidate data in the buffer 3123 * cache, as required, so that the coherency conforms with 3124 * RFC3530 Sec. 9.3.2. 3125 */ 3126 if (ap->a_op == F_SETLK) { 3127 if ((np->n_flag & NMODIFIED) == 0) { 3128 np->n_attrstamp = 0; 3129 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3130 ret = VOP_GETATTR(vp, &va, cred); 3131 } 3132 if ((np->n_flag & NMODIFIED) || ret || 3133 np->n_change != va.va_filerev) { 3134 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1); 3135 np->n_attrstamp = 0; 3136 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3137 ret = VOP_GETATTR(vp, &va, cred); 3138 if (!ret) { 3139 np->n_mtime = va.va_mtime; 3140 np->n_change = va.va_filerev; 3141 } 3142 } 3143 /* Mark that a file lock has been acquired. */ 3144 mtx_lock(&np->n_mtx); 3145 np->n_flag |= NHASBEENLOCKED; 3146 mtx_unlock(&np->n_mtx); 3147 } 3148 } else if (!NFS_ISV4(vp)) { 3149 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3150 size = VTONFS(vp)->n_size; 3151 NFSVOPUNLOCK(vp, 0); 3152 error = lf_advlock(ap, &(vp->v_lockf), size); 3153 } else { 3154 if (nfs_advlock_p != NULL) 3155 error = nfs_advlock_p(ap); 3156 else { 3157 NFSVOPUNLOCK(vp, 0); 3158 error = ENOLCK; 3159 } 3160 } 3161 if (error == 0 && ap->a_op == F_SETLK) { 3162 error = NFSVOPLOCK(vp, LK_SHARED); 3163 if (error == 0) { 3164 /* Mark that a file lock has been acquired. */ 3165 mtx_lock(&np->n_mtx); 3166 np->n_flag |= NHASBEENLOCKED; 3167 mtx_unlock(&np->n_mtx); 3168 NFSVOPUNLOCK(vp, 0); 3169 } 3170 } 3171 return (error); 3172 } else 3173 error = EOPNOTSUPP; 3174 out: 3175 NFSVOPUNLOCK(vp, 0); 3176 return (error); 3177 } 3178 3179 /* 3180 * NFS advisory byte-level locks. 3181 */ 3182 static int 3183 nfs_advlockasync(struct vop_advlockasync_args *ap) 3184 { 3185 struct vnode *vp = ap->a_vp; 3186 u_quad_t size; 3187 int error; 3188 3189 if (NFS_ISV4(vp)) 3190 return (EOPNOTSUPP); 3191 error = NFSVOPLOCK(vp, LK_SHARED); 3192 if (error) 3193 return (error); 3194 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3195 size = VTONFS(vp)->n_size; 3196 NFSVOPUNLOCK(vp, 0); 3197 error = lf_advlockasync(ap, &(vp->v_lockf), size); 3198 } else { 3199 NFSVOPUNLOCK(vp, 0); 3200 error = EOPNOTSUPP; 3201 } 3202 return (error); 3203 } 3204 3205 /* 3206 * Print out the contents of an nfsnode. 3207 */ 3208 static int 3209 nfs_print(struct vop_print_args *ap) 3210 { 3211 struct vnode *vp = ap->a_vp; 3212 struct nfsnode *np = VTONFS(vp); 3213 3214 printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid, 3215 (uintmax_t)np->n_vattr.na_fsid); 3216 if (vp->v_type == VFIFO) 3217 fifo_printinfo(vp); 3218 printf("\n"); 3219 return (0); 3220 } 3221 3222 /* 3223 * This is the "real" nfs::bwrite(struct buf*). 3224 * We set B_CACHE if this is a VMIO buffer. 3225 */ 3226 int 3227 ncl_writebp(struct buf *bp, int force __unused, struct thread *td) 3228 { 3229 int oldflags, rtval; 3230 3231 BUF_ASSERT_HELD(bp); 3232 3233 if (bp->b_flags & B_INVAL) { 3234 brelse(bp); 3235 return (0); 3236 } 3237 3238 oldflags = bp->b_flags; 3239 bp->b_flags |= B_CACHE; 3240 3241 /* 3242 * Undirty the bp. We will redirty it later if the I/O fails. 3243 */ 3244 bundirty(bp); 3245 bp->b_flags &= ~B_DONE; 3246 bp->b_ioflags &= ~BIO_ERROR; 3247 bp->b_iocmd = BIO_WRITE; 3248 3249 bufobj_wref(bp->b_bufobj); 3250 curthread->td_ru.ru_oublock++; 3251 3252 /* 3253 * Note: to avoid loopback deadlocks, we do not 3254 * assign b_runningbufspace. 3255 */ 3256 vfs_busy_pages(bp, 1); 3257 3258 BUF_KERNPROC(bp); 3259 bp->b_iooffset = dbtob(bp->b_blkno); 3260 bstrategy(bp); 3261 3262 if ((oldflags & B_ASYNC) != 0) 3263 return (0); 3264 3265 rtval = bufwait(bp); 3266 if (oldflags & B_DELWRI) 3267 reassignbuf(bp); 3268 brelse(bp); 3269 return (rtval); 3270 } 3271 3272 /* 3273 * nfs special file access vnode op. 3274 * Essentially just get vattr and then imitate iaccess() since the device is 3275 * local to the client. 3276 */ 3277 static int 3278 nfsspec_access(struct vop_access_args *ap) 3279 { 3280 struct vattr *vap; 3281 struct ucred *cred = ap->a_cred; 3282 struct vnode *vp = ap->a_vp; 3283 accmode_t accmode = ap->a_accmode; 3284 struct vattr vattr; 3285 int error; 3286 3287 /* 3288 * Disallow write attempts on filesystems mounted read-only; 3289 * unless the file is a socket, fifo, or a block or character 3290 * device resident on the filesystem. 3291 */ 3292 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3293 switch (vp->v_type) { 3294 case VREG: 3295 case VDIR: 3296 case VLNK: 3297 return (EROFS); 3298 default: 3299 break; 3300 } 3301 } 3302 vap = &vattr; 3303 error = VOP_GETATTR(vp, vap, cred); 3304 if (error) 3305 goto out; 3306 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid, 3307 accmode, cred, NULL); 3308 out: 3309 return error; 3310 } 3311 3312 /* 3313 * Read wrapper for fifos. 3314 */ 3315 static int 3316 nfsfifo_read(struct vop_read_args *ap) 3317 { 3318 struct nfsnode *np = VTONFS(ap->a_vp); 3319 int error; 3320 3321 /* 3322 * Set access flag. 3323 */ 3324 mtx_lock(&np->n_mtx); 3325 np->n_flag |= NACC; 3326 vfs_timestamp(&np->n_atim); 3327 mtx_unlock(&np->n_mtx); 3328 error = fifo_specops.vop_read(ap); 3329 return error; 3330 } 3331 3332 /* 3333 * Write wrapper for fifos. 3334 */ 3335 static int 3336 nfsfifo_write(struct vop_write_args *ap) 3337 { 3338 struct nfsnode *np = VTONFS(ap->a_vp); 3339 3340 /* 3341 * Set update flag. 3342 */ 3343 mtx_lock(&np->n_mtx); 3344 np->n_flag |= NUPD; 3345 vfs_timestamp(&np->n_mtim); 3346 mtx_unlock(&np->n_mtx); 3347 return(fifo_specops.vop_write(ap)); 3348 } 3349 3350 /* 3351 * Close wrapper for fifos. 3352 * 3353 * Update the times on the nfsnode then do fifo close. 3354 */ 3355 static int 3356 nfsfifo_close(struct vop_close_args *ap) 3357 { 3358 struct vnode *vp = ap->a_vp; 3359 struct nfsnode *np = VTONFS(vp); 3360 struct vattr vattr; 3361 struct timespec ts; 3362 3363 mtx_lock(&np->n_mtx); 3364 if (np->n_flag & (NACC | NUPD)) { 3365 vfs_timestamp(&ts); 3366 if (np->n_flag & NACC) 3367 np->n_atim = ts; 3368 if (np->n_flag & NUPD) 3369 np->n_mtim = ts; 3370 np->n_flag |= NCHG; 3371 if (vrefcnt(vp) == 1 && 3372 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3373 VATTR_NULL(&vattr); 3374 if (np->n_flag & NACC) 3375 vattr.va_atime = np->n_atim; 3376 if (np->n_flag & NUPD) 3377 vattr.va_mtime = np->n_mtim; 3378 mtx_unlock(&np->n_mtx); 3379 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3380 goto out; 3381 } 3382 } 3383 mtx_unlock(&np->n_mtx); 3384 out: 3385 return (fifo_specops.vop_close(ap)); 3386 } 3387 3388 /* 3389 * Just call ncl_writebp() with the force argument set to 1. 3390 * 3391 * NOTE: B_DONE may or may not be set in a_bp on call. 3392 */ 3393 static int 3394 nfs_bwrite(struct buf *bp) 3395 { 3396 3397 return (ncl_writebp(bp, 1, curthread)); 3398 } 3399 3400 struct buf_ops buf_ops_newnfs = { 3401 .bop_name = "buf_ops_nfs", 3402 .bop_write = nfs_bwrite, 3403 .bop_strategy = bufstrategy, 3404 .bop_sync = bufsync, 3405 .bop_bdflush = bufbdflush, 3406 }; 3407 3408 static int 3409 nfs_getacl(struct vop_getacl_args *ap) 3410 { 3411 int error; 3412 3413 if (ap->a_type != ACL_TYPE_NFS4) 3414 return (EOPNOTSUPP); 3415 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3416 NULL); 3417 if (error > NFSERR_STALE) { 3418 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3419 error = EPERM; 3420 } 3421 return (error); 3422 } 3423 3424 static int 3425 nfs_setacl(struct vop_setacl_args *ap) 3426 { 3427 int error; 3428 3429 if (ap->a_type != ACL_TYPE_NFS4) 3430 return (EOPNOTSUPP); 3431 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3432 NULL); 3433 if (error > NFSERR_STALE) { 3434 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3435 error = EPERM; 3436 } 3437 return (error); 3438 } 3439 3440 /* 3441 * Return POSIX pathconf information applicable to nfs filesystems. 3442 */ 3443 static int 3444 nfs_pathconf(struct vop_pathconf_args *ap) 3445 { 3446 struct nfsv3_pathconf pc; 3447 struct nfsvattr nfsva; 3448 struct vnode *vp = ap->a_vp; 3449 struct thread *td = curthread; 3450 int attrflag, error; 3451 3452 if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX || 3453 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED || 3454 ap->a_name == _PC_NO_TRUNC)) || 3455 (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) { 3456 /* 3457 * Since only the above 4 a_names are returned by the NFSv3 3458 * Pathconf RPC, there is no point in doing it for others. 3459 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can 3460 * be used for _PC_NFS4_ACL as well. 3461 */ 3462 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva, 3463 &attrflag, NULL); 3464 if (attrflag != 0) 3465 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 3466 1); 3467 if (error != 0) 3468 return (error); 3469 } else { 3470 /* 3471 * For NFSv2 (or NFSv3 when not one of the above 4 a_names), 3472 * just fake them. 3473 */ 3474 pc.pc_linkmax = NFS_LINK_MAX; 3475 pc.pc_namemax = NFS_MAXNAMLEN; 3476 pc.pc_notrunc = 1; 3477 pc.pc_chownrestricted = 1; 3478 pc.pc_caseinsensitive = 0; 3479 pc.pc_casepreserving = 1; 3480 error = 0; 3481 } 3482 switch (ap->a_name) { 3483 case _PC_LINK_MAX: 3484 #ifdef _LP64 3485 *ap->a_retval = pc.pc_linkmax; 3486 #else 3487 *ap->a_retval = MIN(LONG_MAX, pc.pc_linkmax); 3488 #endif 3489 break; 3490 case _PC_NAME_MAX: 3491 *ap->a_retval = pc.pc_namemax; 3492 break; 3493 case _PC_PIPE_BUF: 3494 if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO) 3495 *ap->a_retval = PIPE_BUF; 3496 else 3497 error = EINVAL; 3498 break; 3499 case _PC_CHOWN_RESTRICTED: 3500 *ap->a_retval = pc.pc_chownrestricted; 3501 break; 3502 case _PC_NO_TRUNC: 3503 *ap->a_retval = pc.pc_notrunc; 3504 break; 3505 case _PC_ACL_NFS4: 3506 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 && 3507 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL)) 3508 *ap->a_retval = 1; 3509 else 3510 *ap->a_retval = 0; 3511 break; 3512 case _PC_ACL_PATH_MAX: 3513 if (NFS_ISV4(vp)) 3514 *ap->a_retval = ACL_MAX_ENTRIES; 3515 else 3516 *ap->a_retval = 3; 3517 break; 3518 case _PC_PRIO_IO: 3519 *ap->a_retval = 0; 3520 break; 3521 case _PC_SYNC_IO: 3522 *ap->a_retval = 0; 3523 break; 3524 case _PC_ALLOC_SIZE_MIN: 3525 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize; 3526 break; 3527 case _PC_FILESIZEBITS: 3528 if (NFS_ISV34(vp)) 3529 *ap->a_retval = 64; 3530 else 3531 *ap->a_retval = 32; 3532 break; 3533 case _PC_REC_INCR_XFER_SIZE: 3534 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 3535 break; 3536 case _PC_REC_MAX_XFER_SIZE: 3537 *ap->a_retval = -1; /* means ``unlimited'' */ 3538 break; 3539 case _PC_REC_MIN_XFER_SIZE: 3540 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 3541 break; 3542 case _PC_REC_XFER_ALIGN: 3543 *ap->a_retval = PAGE_SIZE; 3544 break; 3545 case _PC_SYMLINK_MAX: 3546 *ap->a_retval = NFS_MAXPATHLEN; 3547 break; 3548 3549 default: 3550 error = vop_stdpathconf(ap); 3551 break; 3552 } 3553 return (error); 3554 } 3555 3556