1 /*- 2 * Copyright (c) 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * Rick Macklem at The University of Guelph. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 4. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * from nfs_vnops.c 8.16 (Berkeley) 5/27/95 33 */ 34 35 #include <sys/cdefs.h> 36 __FBSDID("$FreeBSD$"); 37 38 /* 39 * vnode op calls for Sun NFS version 2, 3 and 4 40 */ 41 42 #include "opt_kdtrace.h" 43 #include "opt_inet.h" 44 45 #include <sys/param.h> 46 #include <sys/kernel.h> 47 #include <sys/systm.h> 48 #include <sys/resourcevar.h> 49 #include <sys/proc.h> 50 #include <sys/mount.h> 51 #include <sys/bio.h> 52 #include <sys/buf.h> 53 #include <sys/jail.h> 54 #include <sys/malloc.h> 55 #include <sys/mbuf.h> 56 #include <sys/namei.h> 57 #include <sys/socket.h> 58 #include <sys/vnode.h> 59 #include <sys/dirent.h> 60 #include <sys/fcntl.h> 61 #include <sys/lockf.h> 62 #include <sys/stat.h> 63 #include <sys/sysctl.h> 64 #include <sys/signalvar.h> 65 66 #include <vm/vm.h> 67 #include <vm/vm_extern.h> 68 #include <vm/vm_object.h> 69 70 #include <fs/nfs/nfsport.h> 71 #include <fs/nfsclient/nfsnode.h> 72 #include <fs/nfsclient/nfsmount.h> 73 #include <fs/nfsclient/nfs.h> 74 #include <fs/nfsclient/nfs_kdtrace.h> 75 76 #include <net/if.h> 77 #include <netinet/in.h> 78 #include <netinet/in_var.h> 79 80 #include <nfs/nfs_lock.h> 81 82 #ifdef KDTRACE_HOOKS 83 #include <sys/dtrace_bsd.h> 84 85 dtrace_nfsclient_accesscache_flush_probe_func_t 86 dtrace_nfscl_accesscache_flush_done_probe; 87 uint32_t nfscl_accesscache_flush_done_id; 88 89 dtrace_nfsclient_accesscache_get_probe_func_t 90 dtrace_nfscl_accesscache_get_hit_probe, 91 dtrace_nfscl_accesscache_get_miss_probe; 92 uint32_t nfscl_accesscache_get_hit_id; 93 uint32_t nfscl_accesscache_get_miss_id; 94 95 dtrace_nfsclient_accesscache_load_probe_func_t 96 dtrace_nfscl_accesscache_load_done_probe; 97 uint32_t nfscl_accesscache_load_done_id; 98 #endif /* !KDTRACE_HOOKS */ 99 100 /* Defs */ 101 #define TRUE 1 102 #define FALSE 0 103 104 extern struct nfsstats newnfsstats; 105 extern int nfsrv_useacl; 106 MALLOC_DECLARE(M_NEWNFSREQ); 107 108 /* 109 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these 110 * calls are not in getblk() and brelse() so that they would not be necessary 111 * here. 112 */ 113 #ifndef B_VMIO 114 #define vfs_busy_pages(bp, f) 115 #endif 116 117 static vop_read_t nfsfifo_read; 118 static vop_write_t nfsfifo_write; 119 static vop_close_t nfsfifo_close; 120 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *, 121 struct thread *); 122 static vop_lookup_t nfs_lookup; 123 static vop_create_t nfs_create; 124 static vop_mknod_t nfs_mknod; 125 static vop_open_t nfs_open; 126 static vop_pathconf_t nfs_pathconf; 127 static vop_close_t nfs_close; 128 static vop_access_t nfs_access; 129 static vop_getattr_t nfs_getattr; 130 static vop_setattr_t nfs_setattr; 131 static vop_read_t nfs_read; 132 static vop_fsync_t nfs_fsync; 133 static vop_remove_t nfs_remove; 134 static vop_link_t nfs_link; 135 static vop_rename_t nfs_rename; 136 static vop_mkdir_t nfs_mkdir; 137 static vop_rmdir_t nfs_rmdir; 138 static vop_symlink_t nfs_symlink; 139 static vop_readdir_t nfs_readdir; 140 static vop_strategy_t nfs_strategy; 141 static vop_lock1_t nfs_lock1; 142 static int nfs_lookitup(struct vnode *, char *, int, 143 struct ucred *, struct thread *, struct nfsnode **); 144 static int nfs_sillyrename(struct vnode *, struct vnode *, 145 struct componentname *); 146 static vop_access_t nfsspec_access; 147 static vop_readlink_t nfs_readlink; 148 static vop_print_t nfs_print; 149 static vop_advlock_t nfs_advlock; 150 static vop_advlockasync_t nfs_advlockasync; 151 static vop_getacl_t nfs_getacl; 152 static vop_setacl_t nfs_setacl; 153 154 /* 155 * Global vfs data structures for nfs 156 */ 157 struct vop_vector newnfs_vnodeops = { 158 .vop_default = &default_vnodeops, 159 .vop_access = nfs_access, 160 .vop_advlock = nfs_advlock, 161 .vop_advlockasync = nfs_advlockasync, 162 .vop_close = nfs_close, 163 .vop_create = nfs_create, 164 .vop_fsync = nfs_fsync, 165 .vop_getattr = nfs_getattr, 166 .vop_getpages = ncl_getpages, 167 .vop_putpages = ncl_putpages, 168 .vop_inactive = ncl_inactive, 169 .vop_link = nfs_link, 170 .vop_lock1 = nfs_lock1, 171 .vop_lookup = nfs_lookup, 172 .vop_mkdir = nfs_mkdir, 173 .vop_mknod = nfs_mknod, 174 .vop_open = nfs_open, 175 .vop_pathconf = nfs_pathconf, 176 .vop_print = nfs_print, 177 .vop_read = nfs_read, 178 .vop_readdir = nfs_readdir, 179 .vop_readlink = nfs_readlink, 180 .vop_reclaim = ncl_reclaim, 181 .vop_remove = nfs_remove, 182 .vop_rename = nfs_rename, 183 .vop_rmdir = nfs_rmdir, 184 .vop_setattr = nfs_setattr, 185 .vop_strategy = nfs_strategy, 186 .vop_symlink = nfs_symlink, 187 .vop_write = ncl_write, 188 .vop_getacl = nfs_getacl, 189 .vop_setacl = nfs_setacl, 190 }; 191 192 struct vop_vector newnfs_fifoops = { 193 .vop_default = &fifo_specops, 194 .vop_access = nfsspec_access, 195 .vop_close = nfsfifo_close, 196 .vop_fsync = nfs_fsync, 197 .vop_getattr = nfs_getattr, 198 .vop_inactive = ncl_inactive, 199 .vop_print = nfs_print, 200 .vop_read = nfsfifo_read, 201 .vop_reclaim = ncl_reclaim, 202 .vop_setattr = nfs_setattr, 203 .vop_write = nfsfifo_write, 204 }; 205 206 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, 207 struct componentname *cnp, struct vattr *vap); 208 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 209 int namelen, struct ucred *cred, struct thread *td); 210 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, 211 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp, 212 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td); 213 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp, 214 struct componentname *scnp, struct sillyrename *sp); 215 216 /* 217 * Global variables 218 */ 219 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1)) 220 221 SYSCTL_DECL(_vfs_nfs); 222 223 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO; 224 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 225 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 226 227 static int nfs_prime_access_cache = 0; 228 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW, 229 &nfs_prime_access_cache, 0, 230 "Prime NFS ACCESS cache when fetching attributes"); 231 232 static int newnfs_commit_on_close = 0; 233 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW, 234 &newnfs_commit_on_close, 0, "write+commit on close, else only write"); 235 236 static int nfs_clean_pages_on_close = 1; 237 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW, 238 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close"); 239 240 int newnfs_directio_enable = 0; 241 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW, 242 &newnfs_directio_enable, 0, "Enable NFS directio"); 243 244 int nfs_keep_dirty_on_error; 245 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW, 246 &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned"); 247 248 /* 249 * This sysctl allows other processes to mmap a file that has been opened 250 * O_DIRECT by a process. In general, having processes mmap the file while 251 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow 252 * this by default to prevent DoS attacks - to prevent a malicious user from 253 * opening up files O_DIRECT preventing other users from mmap'ing these 254 * files. "Protected" environments where stricter consistency guarantees are 255 * required can disable this knob. The process that opened the file O_DIRECT 256 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not 257 * meaningful. 258 */ 259 int newnfs_directio_allow_mmap = 1; 260 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW, 261 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens"); 262 263 #if 0 264 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD, 265 &newnfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count"); 266 267 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD, 268 &newnfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count"); 269 #endif 270 271 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \ 272 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \ 273 | NFSACCESS_DELETE | NFSACCESS_LOOKUP) 274 275 /* 276 * SMP Locking Note : 277 * The list of locks after the description of the lock is the ordering 278 * of other locks acquired with the lock held. 279 * np->n_mtx : Protects the fields in the nfsnode. 280 VM Object Lock 281 VI_MTX (acquired indirectly) 282 * nmp->nm_mtx : Protects the fields in the nfsmount. 283 rep->r_mtx 284 * ncl_iod_mutex : Global lock, protects shared nfsiod state. 285 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list. 286 nmp->nm_mtx 287 rep->r_mtx 288 * rep->r_mtx : Protects the fields in an nfsreq. 289 */ 290 291 static int 292 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td, 293 struct ucred *cred, u_int32_t *retmode) 294 { 295 int error = 0, attrflag, i, lrupos; 296 u_int32_t rmode; 297 struct nfsnode *np = VTONFS(vp); 298 struct nfsvattr nfsva; 299 300 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag, 301 &rmode, NULL); 302 if (attrflag) 303 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 304 if (!error) { 305 lrupos = 0; 306 mtx_lock(&np->n_mtx); 307 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 308 if (np->n_accesscache[i].uid == cred->cr_uid) { 309 np->n_accesscache[i].mode = rmode; 310 np->n_accesscache[i].stamp = time_second; 311 break; 312 } 313 if (i > 0 && np->n_accesscache[i].stamp < 314 np->n_accesscache[lrupos].stamp) 315 lrupos = i; 316 } 317 if (i == NFS_ACCESSCACHESIZE) { 318 np->n_accesscache[lrupos].uid = cred->cr_uid; 319 np->n_accesscache[lrupos].mode = rmode; 320 np->n_accesscache[lrupos].stamp = time_second; 321 } 322 mtx_unlock(&np->n_mtx); 323 if (retmode != NULL) 324 *retmode = rmode; 325 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0); 326 } else if (NFS_ISV4(vp)) { 327 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 328 } 329 #ifdef KDTRACE_HOOKS 330 if (error != 0) 331 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0, 332 error); 333 #endif 334 return (error); 335 } 336 337 /* 338 * nfs access vnode op. 339 * For nfs version 2, just return ok. File accesses may fail later. 340 * For nfs version 3, use the access rpc to check accessibility. If file modes 341 * are changed on the server, accesses might still fail later. 342 */ 343 static int 344 nfs_access(struct vop_access_args *ap) 345 { 346 struct vnode *vp = ap->a_vp; 347 int error = 0, i, gotahit; 348 u_int32_t mode, wmode, rmode; 349 int v34 = NFS_ISV34(vp); 350 struct nfsnode *np = VTONFS(vp); 351 352 /* 353 * Disallow write attempts on filesystems mounted read-only; 354 * unless the file is a socket, fifo, or a block or character 355 * device resident on the filesystem. 356 */ 357 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS | 358 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL | 359 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) { 360 switch (vp->v_type) { 361 case VREG: 362 case VDIR: 363 case VLNK: 364 return (EROFS); 365 default: 366 break; 367 } 368 } 369 /* 370 * For nfs v3 or v4, check to see if we have done this recently, and if 371 * so return our cached result instead of making an ACCESS call. 372 * If not, do an access rpc, otherwise you are stuck emulating 373 * ufs_access() locally using the vattr. This may not be correct, 374 * since the server may apply other access criteria such as 375 * client uid-->server uid mapping that we do not know about. 376 */ 377 if (v34) { 378 if (ap->a_accmode & VREAD) 379 mode = NFSACCESS_READ; 380 else 381 mode = 0; 382 if (vp->v_type != VDIR) { 383 if (ap->a_accmode & VWRITE) 384 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 385 if (ap->a_accmode & VAPPEND) 386 mode |= NFSACCESS_EXTEND; 387 if (ap->a_accmode & VEXEC) 388 mode |= NFSACCESS_EXECUTE; 389 if (ap->a_accmode & VDELETE) 390 mode |= NFSACCESS_DELETE; 391 } else { 392 if (ap->a_accmode & VWRITE) 393 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 394 if (ap->a_accmode & VAPPEND) 395 mode |= NFSACCESS_EXTEND; 396 if (ap->a_accmode & VEXEC) 397 mode |= NFSACCESS_LOOKUP; 398 if (ap->a_accmode & VDELETE) 399 mode |= NFSACCESS_DELETE; 400 if (ap->a_accmode & VDELETE_CHILD) 401 mode |= NFSACCESS_MODIFY; 402 } 403 /* XXX safety belt, only make blanket request if caching */ 404 if (nfsaccess_cache_timeout > 0) { 405 wmode = NFSACCESS_READ | NFSACCESS_MODIFY | 406 NFSACCESS_EXTEND | NFSACCESS_EXECUTE | 407 NFSACCESS_DELETE | NFSACCESS_LOOKUP; 408 } else { 409 wmode = mode; 410 } 411 412 /* 413 * Does our cached result allow us to give a definite yes to 414 * this request? 415 */ 416 gotahit = 0; 417 mtx_lock(&np->n_mtx); 418 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 419 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) { 420 if (time_second < (np->n_accesscache[i].stamp 421 + nfsaccess_cache_timeout) && 422 (np->n_accesscache[i].mode & mode) == mode) { 423 NFSINCRGLOBAL(newnfsstats.accesscache_hits); 424 gotahit = 1; 425 } 426 break; 427 } 428 } 429 mtx_unlock(&np->n_mtx); 430 #ifdef KDTRACE_HOOKS 431 if (gotahit != 0) 432 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp, 433 ap->a_cred->cr_uid, mode); 434 else 435 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp, 436 ap->a_cred->cr_uid, mode); 437 #endif 438 if (gotahit == 0) { 439 /* 440 * Either a no, or a don't know. Go to the wire. 441 */ 442 NFSINCRGLOBAL(newnfsstats.accesscache_misses); 443 error = nfs34_access_otw(vp, wmode, ap->a_td, 444 ap->a_cred, &rmode); 445 if (!error && 446 (rmode & mode) != mode) 447 error = EACCES; 448 } 449 return (error); 450 } else { 451 if ((error = nfsspec_access(ap)) != 0) { 452 return (error); 453 } 454 /* 455 * Attempt to prevent a mapped root from accessing a file 456 * which it shouldn't. We try to read a byte from the file 457 * if the user is root and the file is not zero length. 458 * After calling nfsspec_access, we should have the correct 459 * file size cached. 460 */ 461 mtx_lock(&np->n_mtx); 462 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD) 463 && VTONFS(vp)->n_size > 0) { 464 struct iovec aiov; 465 struct uio auio; 466 char buf[1]; 467 468 mtx_unlock(&np->n_mtx); 469 aiov.iov_base = buf; 470 aiov.iov_len = 1; 471 auio.uio_iov = &aiov; 472 auio.uio_iovcnt = 1; 473 auio.uio_offset = 0; 474 auio.uio_resid = 1; 475 auio.uio_segflg = UIO_SYSSPACE; 476 auio.uio_rw = UIO_READ; 477 auio.uio_td = ap->a_td; 478 479 if (vp->v_type == VREG) 480 error = ncl_readrpc(vp, &auio, ap->a_cred); 481 else if (vp->v_type == VDIR) { 482 char* bp; 483 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 484 aiov.iov_base = bp; 485 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 486 error = ncl_readdirrpc(vp, &auio, ap->a_cred, 487 ap->a_td); 488 free(bp, M_TEMP); 489 } else if (vp->v_type == VLNK) 490 error = ncl_readlinkrpc(vp, &auio, ap->a_cred); 491 else 492 error = EACCES; 493 } else 494 mtx_unlock(&np->n_mtx); 495 return (error); 496 } 497 } 498 499 500 /* 501 * nfs open vnode op 502 * Check to see if the type is ok 503 * and that deletion is not in progress. 504 * For paged in text files, you will need to flush the page cache 505 * if consistency is lost. 506 */ 507 /* ARGSUSED */ 508 static int 509 nfs_open(struct vop_open_args *ap) 510 { 511 struct vnode *vp = ap->a_vp; 512 struct nfsnode *np = VTONFS(vp); 513 struct vattr vattr; 514 int error; 515 int fmode = ap->a_mode; 516 517 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) 518 return (EOPNOTSUPP); 519 520 /* 521 * For NFSv4, we need to do the Open Op before cache validation, 522 * so that we conform to RFC3530 Sec. 9.3.1. 523 */ 524 if (NFS_ISV4(vp)) { 525 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td); 526 if (error) { 527 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 528 (gid_t)0); 529 return (error); 530 } 531 } 532 533 /* 534 * Now, if this Open will be doing reading, re-validate/flush the 535 * cache, so that Close/Open coherency is maintained. 536 */ 537 mtx_lock(&np->n_mtx); 538 if (np->n_flag & NMODIFIED) { 539 mtx_unlock(&np->n_mtx); 540 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 541 if (error == EINTR || error == EIO) { 542 if (NFS_ISV4(vp)) 543 (void) nfsrpc_close(vp, 0, ap->a_td); 544 return (error); 545 } 546 mtx_lock(&np->n_mtx); 547 np->n_attrstamp = 0; 548 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 549 if (vp->v_type == VDIR) 550 np->n_direofoffset = 0; 551 mtx_unlock(&np->n_mtx); 552 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 553 if (error) { 554 if (NFS_ISV4(vp)) 555 (void) nfsrpc_close(vp, 0, ap->a_td); 556 return (error); 557 } 558 mtx_lock(&np->n_mtx); 559 np->n_mtime = vattr.va_mtime; 560 if (NFS_ISV4(vp)) 561 np->n_change = vattr.va_filerev; 562 } else { 563 mtx_unlock(&np->n_mtx); 564 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 565 if (error) { 566 if (NFS_ISV4(vp)) 567 (void) nfsrpc_close(vp, 0, ap->a_td); 568 return (error); 569 } 570 mtx_lock(&np->n_mtx); 571 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) || 572 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 573 if (vp->v_type == VDIR) 574 np->n_direofoffset = 0; 575 mtx_unlock(&np->n_mtx); 576 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 577 if (error == EINTR || error == EIO) { 578 if (NFS_ISV4(vp)) 579 (void) nfsrpc_close(vp, 0, ap->a_td); 580 return (error); 581 } 582 mtx_lock(&np->n_mtx); 583 np->n_mtime = vattr.va_mtime; 584 if (NFS_ISV4(vp)) 585 np->n_change = vattr.va_filerev; 586 } 587 } 588 589 /* 590 * If the object has >= 1 O_DIRECT active opens, we disable caching. 591 */ 592 if (newnfs_directio_enable && (fmode & O_DIRECT) && 593 (vp->v_type == VREG)) { 594 if (np->n_directio_opens == 0) { 595 mtx_unlock(&np->n_mtx); 596 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 597 if (error) { 598 if (NFS_ISV4(vp)) 599 (void) nfsrpc_close(vp, 0, ap->a_td); 600 return (error); 601 } 602 mtx_lock(&np->n_mtx); 603 np->n_flag |= NNONCACHE; 604 } 605 np->n_directio_opens++; 606 } 607 mtx_unlock(&np->n_mtx); 608 vnode_create_vobject(vp, vattr.va_size, ap->a_td); 609 return (0); 610 } 611 612 /* 613 * nfs close vnode op 614 * What an NFS client should do upon close after writing is a debatable issue. 615 * Most NFS clients push delayed writes to the server upon close, basically for 616 * two reasons: 617 * 1 - So that any write errors may be reported back to the client process 618 * doing the close system call. By far the two most likely errors are 619 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 620 * 2 - To put a worst case upper bound on cache inconsistency between 621 * multiple clients for the file. 622 * There is also a consistency problem for Version 2 of the protocol w.r.t. 623 * not being able to tell if other clients are writing a file concurrently, 624 * since there is no way of knowing if the changed modify time in the reply 625 * is only due to the write for this client. 626 * (NFS Version 3 provides weak cache consistency data in the reply that 627 * should be sufficient to detect and handle this case.) 628 * 629 * The current code does the following: 630 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 631 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 632 * or commit them (this satisfies 1 and 2 except for the 633 * case where the server crashes after this close but 634 * before the commit RPC, which is felt to be "good 635 * enough". Changing the last argument to ncl_flush() to 636 * a 1 would force a commit operation, if it is felt a 637 * commit is necessary now. 638 * for NFS Version 4 - flush the dirty buffers and commit them, if 639 * nfscl_mustflush() says this is necessary. 640 * It is necessary if there is no write delegation held, 641 * in order to satisfy open/close coherency. 642 * If the file isn't cached on local stable storage, 643 * it may be necessary in order to detect "out of space" 644 * errors from the server, if the write delegation 645 * issued by the server doesn't allow the file to grow. 646 */ 647 /* ARGSUSED */ 648 static int 649 nfs_close(struct vop_close_args *ap) 650 { 651 struct vnode *vp = ap->a_vp; 652 struct nfsnode *np = VTONFS(vp); 653 struct nfsvattr nfsva; 654 struct ucred *cred; 655 int error = 0, ret, localcred = 0; 656 int fmode = ap->a_fflag; 657 658 if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF)) 659 return (0); 660 /* 661 * During shutdown, a_cred isn't valid, so just use root. 662 */ 663 if (ap->a_cred == NOCRED) { 664 cred = newnfs_getcred(); 665 localcred = 1; 666 } else { 667 cred = ap->a_cred; 668 } 669 if (vp->v_type == VREG) { 670 /* 671 * Examine and clean dirty pages, regardless of NMODIFIED. 672 * This closes a major hole in close-to-open consistency. 673 * We want to push out all dirty pages (and buffers) on 674 * close, regardless of whether they were dirtied by 675 * mmap'ed writes or via write(). 676 */ 677 if (nfs_clean_pages_on_close && vp->v_object) { 678 VM_OBJECT_LOCK(vp->v_object); 679 vm_object_page_clean(vp->v_object, 0, 0, 0); 680 VM_OBJECT_UNLOCK(vp->v_object); 681 } 682 mtx_lock(&np->n_mtx); 683 if (np->n_flag & NMODIFIED) { 684 mtx_unlock(&np->n_mtx); 685 if (NFS_ISV3(vp)) { 686 /* 687 * Under NFSv3 we have dirty buffers to dispose of. We 688 * must flush them to the NFS server. We have the option 689 * of waiting all the way through the commit rpc or just 690 * waiting for the initial write. The default is to only 691 * wait through the initial write so the data is in the 692 * server's cache, which is roughly similar to the state 693 * a standard disk subsystem leaves the file in on close(). 694 * 695 * We cannot clear the NMODIFIED bit in np->n_flag due to 696 * potential races with other processes, and certainly 697 * cannot clear it if we don't commit. 698 * These races occur when there is no longer the old 699 * traditional vnode locking implemented for Vnode Ops. 700 */ 701 int cm = newnfs_commit_on_close ? 1 : 0; 702 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm, 0); 703 /* np->n_flag &= ~NMODIFIED; */ 704 } else if (NFS_ISV4(vp)) { 705 if (nfscl_mustflush(vp) != 0) { 706 int cm = newnfs_commit_on_close ? 1 : 0; 707 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, 708 cm, 0); 709 /* 710 * as above w.r.t races when clearing 711 * NMODIFIED. 712 * np->n_flag &= ~NMODIFIED; 713 */ 714 } 715 } else 716 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 717 mtx_lock(&np->n_mtx); 718 } 719 /* 720 * Invalidate the attribute cache in all cases. 721 * An open is going to fetch fresh attrs any way, other procs 722 * on this node that have file open will be forced to do an 723 * otw attr fetch, but this is safe. 724 * --> A user found that their RPC count dropped by 20% when 725 * this was commented out and I can't see any requirement 726 * for it, so I've disabled it when negative lookups are 727 * enabled. (What does this have to do with negative lookup 728 * caching? Well nothing, except it was reported by the 729 * same user that needed negative lookup caching and I wanted 730 * there to be a way to disable it to see if it 731 * is the cause of some caching/coherency issue that might 732 * crop up.) 733 */ 734 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) { 735 np->n_attrstamp = 0; 736 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 737 } 738 if (np->n_flag & NWRITEERR) { 739 np->n_flag &= ~NWRITEERR; 740 error = np->n_error; 741 } 742 mtx_unlock(&np->n_mtx); 743 } 744 745 if (NFS_ISV4(vp)) { 746 /* 747 * Get attributes so "change" is up to date. 748 */ 749 if (error == 0 && nfscl_mustflush(vp) != 0) { 750 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva, 751 NULL); 752 if (!ret) { 753 np->n_change = nfsva.na_filerev; 754 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 755 NULL, 0, 0); 756 } 757 } 758 759 /* 760 * and do the close. 761 */ 762 ret = nfsrpc_close(vp, 0, ap->a_td); 763 if (!error && ret) 764 error = ret; 765 if (error) 766 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 767 (gid_t)0); 768 } 769 if (newnfs_directio_enable) 770 KASSERT((np->n_directio_asyncwr == 0), 771 ("nfs_close: dirty unflushed (%d) directio buffers\n", 772 np->n_directio_asyncwr)); 773 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) { 774 mtx_lock(&np->n_mtx); 775 KASSERT((np->n_directio_opens > 0), 776 ("nfs_close: unexpectedly value (0) of n_directio_opens\n")); 777 np->n_directio_opens--; 778 if (np->n_directio_opens == 0) 779 np->n_flag &= ~NNONCACHE; 780 mtx_unlock(&np->n_mtx); 781 } 782 if (localcred) 783 NFSFREECRED(cred); 784 return (error); 785 } 786 787 /* 788 * nfs getattr call from vfs. 789 */ 790 static int 791 nfs_getattr(struct vop_getattr_args *ap) 792 { 793 struct vnode *vp = ap->a_vp; 794 struct thread *td = curthread; /* XXX */ 795 struct nfsnode *np = VTONFS(vp); 796 int error = 0; 797 struct nfsvattr nfsva; 798 struct vattr *vap = ap->a_vap; 799 struct vattr vattr; 800 801 /* 802 * Update local times for special files. 803 */ 804 mtx_lock(&np->n_mtx); 805 if (np->n_flag & (NACC | NUPD)) 806 np->n_flag |= NCHG; 807 mtx_unlock(&np->n_mtx); 808 /* 809 * First look in the cache. 810 */ 811 if (ncl_getattrcache(vp, &vattr) == 0) { 812 vap->va_type = vattr.va_type; 813 vap->va_mode = vattr.va_mode; 814 vap->va_nlink = vattr.va_nlink; 815 vap->va_uid = vattr.va_uid; 816 vap->va_gid = vattr.va_gid; 817 vap->va_fsid = vattr.va_fsid; 818 vap->va_fileid = vattr.va_fileid; 819 vap->va_size = vattr.va_size; 820 vap->va_blocksize = vattr.va_blocksize; 821 vap->va_atime = vattr.va_atime; 822 vap->va_mtime = vattr.va_mtime; 823 vap->va_ctime = vattr.va_ctime; 824 vap->va_gen = vattr.va_gen; 825 vap->va_flags = vattr.va_flags; 826 vap->va_rdev = vattr.va_rdev; 827 vap->va_bytes = vattr.va_bytes; 828 vap->va_filerev = vattr.va_filerev; 829 /* 830 * Get the local modify time for the case of a write 831 * delegation. 832 */ 833 nfscl_deleggetmodtime(vp, &vap->va_mtime); 834 return (0); 835 } 836 837 if (NFS_ISV34(vp) && nfs_prime_access_cache && 838 nfsaccess_cache_timeout > 0) { 839 NFSINCRGLOBAL(newnfsstats.accesscache_misses); 840 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL); 841 if (ncl_getattrcache(vp, ap->a_vap) == 0) { 842 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime); 843 return (0); 844 } 845 } 846 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL); 847 if (!error) 848 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0); 849 if (!error) { 850 /* 851 * Get the local modify time for the case of a write 852 * delegation. 853 */ 854 nfscl_deleggetmodtime(vp, &vap->va_mtime); 855 } else if (NFS_ISV4(vp)) { 856 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 857 } 858 return (error); 859 } 860 861 /* 862 * nfs setattr call. 863 */ 864 static int 865 nfs_setattr(struct vop_setattr_args *ap) 866 { 867 struct vnode *vp = ap->a_vp; 868 struct nfsnode *np = VTONFS(vp); 869 struct thread *td = curthread; /* XXX */ 870 struct vattr *vap = ap->a_vap; 871 int error = 0; 872 u_quad_t tsize; 873 874 #ifndef nolint 875 tsize = (u_quad_t)0; 876 #endif 877 878 /* 879 * Setting of flags and marking of atimes are not supported. 880 */ 881 if (vap->va_flags != VNOVAL) 882 return (EOPNOTSUPP); 883 884 /* 885 * Disallow write attempts if the filesystem is mounted read-only. 886 */ 887 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 888 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 889 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 890 (vp->v_mount->mnt_flag & MNT_RDONLY)) 891 return (EROFS); 892 if (vap->va_size != VNOVAL) { 893 switch (vp->v_type) { 894 case VDIR: 895 return (EISDIR); 896 case VCHR: 897 case VBLK: 898 case VSOCK: 899 case VFIFO: 900 if (vap->va_mtime.tv_sec == VNOVAL && 901 vap->va_atime.tv_sec == VNOVAL && 902 vap->va_mode == (mode_t)VNOVAL && 903 vap->va_uid == (uid_t)VNOVAL && 904 vap->va_gid == (gid_t)VNOVAL) 905 return (0); 906 vap->va_size = VNOVAL; 907 break; 908 default: 909 /* 910 * Disallow write attempts if the filesystem is 911 * mounted read-only. 912 */ 913 if (vp->v_mount->mnt_flag & MNT_RDONLY) 914 return (EROFS); 915 /* 916 * We run vnode_pager_setsize() early (why?), 917 * we must set np->n_size now to avoid vinvalbuf 918 * V_SAVE races that might setsize a lower 919 * value. 920 */ 921 mtx_lock(&np->n_mtx); 922 tsize = np->n_size; 923 mtx_unlock(&np->n_mtx); 924 error = ncl_meta_setsize(vp, ap->a_cred, td, 925 vap->va_size); 926 mtx_lock(&np->n_mtx); 927 if (np->n_flag & NMODIFIED) { 928 tsize = np->n_size; 929 mtx_unlock(&np->n_mtx); 930 if (vap->va_size == 0) 931 error = ncl_vinvalbuf(vp, 0, td, 1); 932 else 933 error = ncl_vinvalbuf(vp, V_SAVE, td, 1); 934 if (error) { 935 vnode_pager_setsize(vp, tsize); 936 return (error); 937 } 938 /* 939 * Call nfscl_delegmodtime() to set the modify time 940 * locally, as required. 941 */ 942 nfscl_delegmodtime(vp); 943 } else 944 mtx_unlock(&np->n_mtx); 945 /* 946 * np->n_size has already been set to vap->va_size 947 * in ncl_meta_setsize(). We must set it again since 948 * nfs_loadattrcache() could be called through 949 * ncl_meta_setsize() and could modify np->n_size. 950 */ 951 mtx_lock(&np->n_mtx); 952 np->n_vattr.na_size = np->n_size = vap->va_size; 953 mtx_unlock(&np->n_mtx); 954 }; 955 } else { 956 mtx_lock(&np->n_mtx); 957 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) && 958 (np->n_flag & NMODIFIED) && vp->v_type == VREG) { 959 mtx_unlock(&np->n_mtx); 960 if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 && 961 (error == EINTR || error == EIO)) 962 return (error); 963 } else 964 mtx_unlock(&np->n_mtx); 965 } 966 error = nfs_setattrrpc(vp, vap, ap->a_cred, td); 967 if (error && vap->va_size != VNOVAL) { 968 mtx_lock(&np->n_mtx); 969 np->n_size = np->n_vattr.na_size = tsize; 970 vnode_pager_setsize(vp, tsize); 971 mtx_unlock(&np->n_mtx); 972 } 973 return (error); 974 } 975 976 /* 977 * Do an nfs setattr rpc. 978 */ 979 static int 980 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred, 981 struct thread *td) 982 { 983 struct nfsnode *np = VTONFS(vp); 984 int error, ret, attrflag, i; 985 struct nfsvattr nfsva; 986 987 if (NFS_ISV34(vp)) { 988 mtx_lock(&np->n_mtx); 989 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) 990 np->n_accesscache[i].stamp = 0; 991 np->n_flag |= NDELEGMOD; 992 mtx_unlock(&np->n_mtx); 993 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp); 994 } 995 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag, 996 NULL); 997 if (attrflag) { 998 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 999 if (ret && !error) 1000 error = ret; 1001 } 1002 if (error && NFS_ISV4(vp)) 1003 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid); 1004 return (error); 1005 } 1006 1007 /* 1008 * nfs lookup call, one step at a time... 1009 * First look in cache 1010 * If not found, unlock the directory nfsnode and do the rpc 1011 */ 1012 static int 1013 nfs_lookup(struct vop_lookup_args *ap) 1014 { 1015 struct componentname *cnp = ap->a_cnp; 1016 struct vnode *dvp = ap->a_dvp; 1017 struct vnode **vpp = ap->a_vpp; 1018 struct mount *mp = dvp->v_mount; 1019 int flags = cnp->cn_flags; 1020 struct vnode *newvp; 1021 struct nfsmount *nmp; 1022 struct nfsnode *np, *newnp; 1023 int error = 0, attrflag, dattrflag, ltype, ncticks; 1024 struct thread *td = cnp->cn_thread; 1025 struct nfsfh *nfhp; 1026 struct nfsvattr dnfsva, nfsva; 1027 struct vattr vattr; 1028 struct timespec nctime; 1029 1030 *vpp = NULLVP; 1031 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) && 1032 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 1033 return (EROFS); 1034 if (dvp->v_type != VDIR) 1035 return (ENOTDIR); 1036 nmp = VFSTONFS(mp); 1037 np = VTONFS(dvp); 1038 1039 /* For NFSv4, wait until any remove is done. */ 1040 mtx_lock(&np->n_mtx); 1041 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) { 1042 np->n_flag |= NREMOVEWANT; 1043 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0); 1044 } 1045 mtx_unlock(&np->n_mtx); 1046 1047 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) 1048 return (error); 1049 error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks); 1050 if (error > 0 && error != ENOENT) 1051 return (error); 1052 if (error == -1) { 1053 /* 1054 * Lookups of "." are special and always return the 1055 * current directory. cache_lookup() already handles 1056 * associated locking bookkeeping, etc. 1057 */ 1058 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 1059 /* XXX: Is this really correct? */ 1060 if (cnp->cn_nameiop != LOOKUP && 1061 (flags & ISLASTCN)) 1062 cnp->cn_flags |= SAVENAME; 1063 return (0); 1064 } 1065 1066 /* 1067 * We only accept a positive hit in the cache if the 1068 * change time of the file matches our cached copy. 1069 * Otherwise, we discard the cache entry and fallback 1070 * to doing a lookup RPC. We also only trust cache 1071 * entries for less than nm_nametimeo seconds. 1072 * 1073 * To better handle stale file handles and attributes, 1074 * clear the attribute cache of this node if it is a 1075 * leaf component, part of an open() call, and not 1076 * locally modified before fetching the attributes. 1077 * This should allow stale file handles to be detected 1078 * here where we can fall back to a LOOKUP RPC to 1079 * recover rather than having nfs_open() detect the 1080 * stale file handle and failing open(2) with ESTALE. 1081 */ 1082 newvp = *vpp; 1083 newnp = VTONFS(newvp); 1084 if (!(nmp->nm_flag & NFSMNT_NOCTO) && 1085 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1086 !(newnp->n_flag & NMODIFIED)) { 1087 mtx_lock(&newnp->n_mtx); 1088 newnp->n_attrstamp = 0; 1089 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1090 mtx_unlock(&newnp->n_mtx); 1091 } 1092 if (nfscl_nodeleg(newvp, 0) == 0 || 1093 ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) && 1094 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 && 1095 timespeccmp(&vattr.va_ctime, &nctime, ==))) { 1096 NFSINCRGLOBAL(newnfsstats.lookupcache_hits); 1097 if (cnp->cn_nameiop != LOOKUP && 1098 (flags & ISLASTCN)) 1099 cnp->cn_flags |= SAVENAME; 1100 return (0); 1101 } 1102 cache_purge(newvp); 1103 if (dvp != newvp) 1104 vput(newvp); 1105 else 1106 vrele(newvp); 1107 *vpp = NULLVP; 1108 } else if (error == ENOENT) { 1109 if (dvp->v_iflag & VI_DOOMED) 1110 return (ENOENT); 1111 /* 1112 * We only accept a negative hit in the cache if the 1113 * modification time of the parent directory matches 1114 * the cached copy in the name cache entry. 1115 * Otherwise, we discard all of the negative cache 1116 * entries for this directory. We also only trust 1117 * negative cache entries for up to nm_negnametimeo 1118 * seconds. 1119 */ 1120 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) && 1121 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 && 1122 timespeccmp(&vattr.va_mtime, &nctime, ==)) { 1123 NFSINCRGLOBAL(newnfsstats.lookupcache_hits); 1124 return (ENOENT); 1125 } 1126 cache_purge_negative(dvp); 1127 } 1128 1129 error = 0; 1130 newvp = NULLVP; 1131 NFSINCRGLOBAL(newnfsstats.lookupcache_misses); 1132 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1133 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1134 NULL); 1135 if (dattrflag) 1136 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1137 if (error) { 1138 if (newvp != NULLVP) { 1139 vput(newvp); 1140 *vpp = NULLVP; 1141 } 1142 1143 if (error != ENOENT) { 1144 if (NFS_ISV4(dvp)) 1145 error = nfscl_maperr(td, error, (uid_t)0, 1146 (gid_t)0); 1147 return (error); 1148 } 1149 1150 /* The requested file was not found. */ 1151 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1152 (flags & ISLASTCN)) { 1153 /* 1154 * XXX: UFS does a full VOP_ACCESS(dvp, 1155 * VWRITE) here instead of just checking 1156 * MNT_RDONLY. 1157 */ 1158 if (mp->mnt_flag & MNT_RDONLY) 1159 return (EROFS); 1160 cnp->cn_flags |= SAVENAME; 1161 return (EJUSTRETURN); 1162 } 1163 1164 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE && 1165 dattrflag) { 1166 /* 1167 * Cache the modification time of the parent 1168 * directory from the post-op attributes in 1169 * the name cache entry. The negative cache 1170 * entry will be ignored once the directory 1171 * has changed. Don't bother adding the entry 1172 * if the directory has already changed. 1173 */ 1174 mtx_lock(&np->n_mtx); 1175 if (timespeccmp(&np->n_vattr.na_mtime, 1176 &dnfsva.na_mtime, ==)) { 1177 mtx_unlock(&np->n_mtx); 1178 cache_enter_time(dvp, NULL, cnp, 1179 &dnfsva.na_mtime, NULL); 1180 } else 1181 mtx_unlock(&np->n_mtx); 1182 } 1183 return (ENOENT); 1184 } 1185 1186 /* 1187 * Handle RENAME case... 1188 */ 1189 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 1190 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1191 FREE((caddr_t)nfhp, M_NFSFH); 1192 return (EISDIR); 1193 } 1194 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1195 LK_EXCLUSIVE); 1196 if (error) 1197 return (error); 1198 newvp = NFSTOV(np); 1199 if (attrflag) 1200 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1201 0, 1); 1202 *vpp = newvp; 1203 cnp->cn_flags |= SAVENAME; 1204 return (0); 1205 } 1206 1207 if (flags & ISDOTDOT) { 1208 ltype = NFSVOPISLOCKED(dvp); 1209 error = vfs_busy(mp, MBF_NOWAIT); 1210 if (error != 0) { 1211 vfs_ref(mp); 1212 NFSVOPUNLOCK(dvp, 0); 1213 error = vfs_busy(mp, 0); 1214 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1215 vfs_rel(mp); 1216 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) { 1217 vfs_unbusy(mp); 1218 error = ENOENT; 1219 } 1220 if (error != 0) 1221 return (error); 1222 } 1223 NFSVOPUNLOCK(dvp, 0); 1224 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1225 cnp->cn_lkflags); 1226 if (error == 0) 1227 newvp = NFSTOV(np); 1228 vfs_unbusy(mp); 1229 if (newvp != dvp) 1230 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1231 if (dvp->v_iflag & VI_DOOMED) { 1232 if (error == 0) { 1233 if (newvp == dvp) 1234 vrele(newvp); 1235 else 1236 vput(newvp); 1237 } 1238 error = ENOENT; 1239 } 1240 if (error != 0) 1241 return (error); 1242 if (attrflag) 1243 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1244 0, 1); 1245 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1246 FREE((caddr_t)nfhp, M_NFSFH); 1247 VREF(dvp); 1248 newvp = dvp; 1249 if (attrflag) 1250 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1251 0, 1); 1252 } else { 1253 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1254 cnp->cn_lkflags); 1255 if (error) 1256 return (error); 1257 newvp = NFSTOV(np); 1258 if (attrflag) 1259 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1260 0, 1); 1261 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1262 !(np->n_flag & NMODIFIED)) { 1263 /* 1264 * Flush the attribute cache when opening a 1265 * leaf node to ensure that fresh attributes 1266 * are fetched in nfs_open() since we did not 1267 * fetch attributes from the LOOKUP reply. 1268 */ 1269 mtx_lock(&np->n_mtx); 1270 np->n_attrstamp = 0; 1271 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1272 mtx_unlock(&np->n_mtx); 1273 } 1274 } 1275 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1276 cnp->cn_flags |= SAVENAME; 1277 if ((cnp->cn_flags & MAKEENTRY) && 1278 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) && 1279 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0)) 1280 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 1281 newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime); 1282 *vpp = newvp; 1283 return (0); 1284 } 1285 1286 /* 1287 * nfs read call. 1288 * Just call ncl_bioread() to do the work. 1289 */ 1290 static int 1291 nfs_read(struct vop_read_args *ap) 1292 { 1293 struct vnode *vp = ap->a_vp; 1294 1295 switch (vp->v_type) { 1296 case VREG: 1297 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred)); 1298 case VDIR: 1299 return (EISDIR); 1300 default: 1301 return (EOPNOTSUPP); 1302 } 1303 } 1304 1305 /* 1306 * nfs readlink call 1307 */ 1308 static int 1309 nfs_readlink(struct vop_readlink_args *ap) 1310 { 1311 struct vnode *vp = ap->a_vp; 1312 1313 if (vp->v_type != VLNK) 1314 return (EINVAL); 1315 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred)); 1316 } 1317 1318 /* 1319 * Do a readlink rpc. 1320 * Called by ncl_doio() from below the buffer cache. 1321 */ 1322 int 1323 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1324 { 1325 int error, ret, attrflag; 1326 struct nfsvattr nfsva; 1327 1328 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva, 1329 &attrflag, NULL); 1330 if (attrflag) { 1331 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1332 if (ret && !error) 1333 error = ret; 1334 } 1335 if (error && NFS_ISV4(vp)) 1336 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1337 return (error); 1338 } 1339 1340 /* 1341 * nfs read rpc call 1342 * Ditto above 1343 */ 1344 int 1345 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1346 { 1347 int error, ret, attrflag; 1348 struct nfsvattr nfsva; 1349 1350 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, &attrflag, 1351 NULL); 1352 if (attrflag) { 1353 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1354 if (ret && !error) 1355 error = ret; 1356 } 1357 if (error && NFS_ISV4(vp)) 1358 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1359 return (error); 1360 } 1361 1362 /* 1363 * nfs write call 1364 */ 1365 int 1366 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 1367 int *iomode, int *must_commit, int called_from_strategy) 1368 { 1369 struct nfsvattr nfsva; 1370 int error = 0, attrflag, ret; 1371 1372 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred, 1373 uiop->uio_td, &nfsva, &attrflag, NULL, called_from_strategy); 1374 if (attrflag) { 1375 if (VTONFS(vp)->n_flag & ND_NFSV4) 1376 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1, 1377 1); 1378 else 1379 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1380 1); 1381 if (ret && !error) 1382 error = ret; 1383 } 1384 if (DOINGASYNC(vp)) 1385 *iomode = NFSWRITE_FILESYNC; 1386 if (error && NFS_ISV4(vp)) 1387 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1388 return (error); 1389 } 1390 1391 /* 1392 * nfs mknod rpc 1393 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1394 * mode set to specify the file type and the size field for rdev. 1395 */ 1396 static int 1397 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1398 struct vattr *vap) 1399 { 1400 struct nfsvattr nfsva, dnfsva; 1401 struct vnode *newvp = NULL; 1402 struct nfsnode *np = NULL, *dnp; 1403 struct nfsfh *nfhp; 1404 struct vattr vattr; 1405 int error = 0, attrflag, dattrflag; 1406 u_int32_t rdev; 1407 1408 if (vap->va_type == VCHR || vap->va_type == VBLK) 1409 rdev = vap->va_rdev; 1410 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1411 rdev = 0xffffffff; 1412 else 1413 return (EOPNOTSUPP); 1414 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1415 return (error); 1416 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap, 1417 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva, 1418 &nfsva, &nfhp, &attrflag, &dattrflag, NULL); 1419 if (!error) { 1420 if (!nfhp) 1421 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1422 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, 1423 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1424 NULL); 1425 if (nfhp) 1426 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1427 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); 1428 } 1429 if (dattrflag) 1430 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1431 if (!error) { 1432 newvp = NFSTOV(np); 1433 if (attrflag != 0) { 1434 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1435 0, 1); 1436 if (error != 0) 1437 vput(newvp); 1438 } 1439 } 1440 if (!error) { 1441 *vpp = newvp; 1442 } else if (NFS_ISV4(dvp)) { 1443 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, 1444 vap->va_gid); 1445 } 1446 dnp = VTONFS(dvp); 1447 mtx_lock(&dnp->n_mtx); 1448 dnp->n_flag |= NMODIFIED; 1449 if (!dattrflag) { 1450 dnp->n_attrstamp = 0; 1451 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1452 } 1453 mtx_unlock(&dnp->n_mtx); 1454 return (error); 1455 } 1456 1457 /* 1458 * nfs mknod vop 1459 * just call nfs_mknodrpc() to do the work. 1460 */ 1461 /* ARGSUSED */ 1462 static int 1463 nfs_mknod(struct vop_mknod_args *ap) 1464 { 1465 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap)); 1466 } 1467 1468 static struct mtx nfs_cverf_mtx; 1469 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex", 1470 MTX_DEF); 1471 1472 static nfsquad_t 1473 nfs_get_cverf(void) 1474 { 1475 static nfsquad_t cverf; 1476 nfsquad_t ret; 1477 static int cverf_initialized = 0; 1478 1479 mtx_lock(&nfs_cverf_mtx); 1480 if (cverf_initialized == 0) { 1481 cverf.lval[0] = arc4random(); 1482 cverf.lval[1] = arc4random(); 1483 cverf_initialized = 1; 1484 } else 1485 cverf.qval++; 1486 ret = cverf; 1487 mtx_unlock(&nfs_cverf_mtx); 1488 1489 return (ret); 1490 } 1491 1492 /* 1493 * nfs file create call 1494 */ 1495 static int 1496 nfs_create(struct vop_create_args *ap) 1497 { 1498 struct vnode *dvp = ap->a_dvp; 1499 struct vattr *vap = ap->a_vap; 1500 struct componentname *cnp = ap->a_cnp; 1501 struct nfsnode *np = NULL, *dnp; 1502 struct vnode *newvp = NULL; 1503 struct nfsmount *nmp; 1504 struct nfsvattr dnfsva, nfsva; 1505 struct nfsfh *nfhp; 1506 nfsquad_t cverf; 1507 int error = 0, attrflag, dattrflag, fmode = 0; 1508 struct vattr vattr; 1509 1510 /* 1511 * Oops, not for me.. 1512 */ 1513 if (vap->va_type == VSOCK) 1514 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1515 1516 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1517 return (error); 1518 if (vap->va_vaflags & VA_EXCLUSIVE) 1519 fmode |= O_EXCL; 1520 dnp = VTONFS(dvp); 1521 nmp = VFSTONFS(vnode_mount(dvp)); 1522 again: 1523 /* For NFSv4, wait until any remove is done. */ 1524 mtx_lock(&dnp->n_mtx); 1525 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) { 1526 dnp->n_flag |= NREMOVEWANT; 1527 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0); 1528 } 1529 mtx_unlock(&dnp->n_mtx); 1530 1531 cverf = nfs_get_cverf(); 1532 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1533 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, 1534 &nfhp, &attrflag, &dattrflag, NULL); 1535 if (!error) { 1536 if (nfhp == NULL) 1537 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1538 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, 1539 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1540 NULL); 1541 if (nfhp != NULL) 1542 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1543 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); 1544 } 1545 if (dattrflag) 1546 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1547 if (!error) { 1548 newvp = NFSTOV(np); 1549 if (attrflag) 1550 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1551 0, 1); 1552 } 1553 if (error) { 1554 if (newvp != NULL) { 1555 vput(newvp); 1556 newvp = NULL; 1557 } 1558 if (NFS_ISV34(dvp) && (fmode & O_EXCL) && 1559 error == NFSERR_NOTSUPP) { 1560 fmode &= ~O_EXCL; 1561 goto again; 1562 } 1563 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) { 1564 if (nfscl_checksattr(vap, &nfsva)) { 1565 /* 1566 * We are normally called with only a partially 1567 * initialized VAP. Since the NFSv3 spec says that 1568 * the server may use the file attributes to 1569 * store the verifier, the spec requires us to do a 1570 * SETATTR RPC. FreeBSD servers store the verifier in 1571 * atime, but we can't really assume that all servers 1572 * will so we ensure that our SETATTR sets both atime 1573 * and mtime. 1574 */ 1575 if (vap->va_mtime.tv_sec == VNOVAL) 1576 vfs_timestamp(&vap->va_mtime); 1577 if (vap->va_atime.tv_sec == VNOVAL) 1578 vap->va_atime = vap->va_mtime; 1579 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred, 1580 cnp->cn_thread, &nfsva, &attrflag, NULL); 1581 if (error && (vap->va_uid != (uid_t)VNOVAL || 1582 vap->va_gid != (gid_t)VNOVAL)) { 1583 /* try again without setting uid/gid */ 1584 vap->va_uid = (uid_t)VNOVAL; 1585 vap->va_gid = (uid_t)VNOVAL; 1586 error = nfsrpc_setattr(newvp, vap, NULL, 1587 cnp->cn_cred, cnp->cn_thread, &nfsva, 1588 &attrflag, NULL); 1589 } 1590 if (attrflag) 1591 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 1592 NULL, 0, 1); 1593 if (error != 0) 1594 vput(newvp); 1595 } 1596 } 1597 if (!error) { 1598 if ((cnp->cn_flags & MAKEENTRY) && attrflag) 1599 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 1600 NULL); 1601 *ap->a_vpp = newvp; 1602 } else if (NFS_ISV4(dvp)) { 1603 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, 1604 vap->va_gid); 1605 } 1606 mtx_lock(&dnp->n_mtx); 1607 dnp->n_flag |= NMODIFIED; 1608 if (!dattrflag) { 1609 dnp->n_attrstamp = 0; 1610 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1611 } 1612 mtx_unlock(&dnp->n_mtx); 1613 return (error); 1614 } 1615 1616 /* 1617 * nfs file remove call 1618 * To try and make nfs semantics closer to ufs semantics, a file that has 1619 * other processes using the vnode is renamed instead of removed and then 1620 * removed later on the last close. 1621 * - If v_usecount > 1 1622 * If a rename is not already in the works 1623 * call nfs_sillyrename() to set it up 1624 * else 1625 * do the remove rpc 1626 */ 1627 static int 1628 nfs_remove(struct vop_remove_args *ap) 1629 { 1630 struct vnode *vp = ap->a_vp; 1631 struct vnode *dvp = ap->a_dvp; 1632 struct componentname *cnp = ap->a_cnp; 1633 struct nfsnode *np = VTONFS(vp); 1634 int error = 0; 1635 struct vattr vattr; 1636 1637 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name")); 1638 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount")); 1639 if (vp->v_type == VDIR) 1640 error = EPERM; 1641 else if (vrefcnt(vp) == 1 || (np->n_sillyrename && 1642 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 1643 vattr.va_nlink > 1)) { 1644 /* 1645 * Purge the name cache so that the chance of a lookup for 1646 * the name succeeding while the remove is in progress is 1647 * minimized. Without node locking it can still happen, such 1648 * that an I/O op returns ESTALE, but since you get this if 1649 * another host removes the file.. 1650 */ 1651 cache_purge(vp); 1652 /* 1653 * throw away biocache buffers, mainly to avoid 1654 * unnecessary delayed writes later. 1655 */ 1656 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1); 1657 /* Do the rpc */ 1658 if (error != EINTR && error != EIO) 1659 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr, 1660 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread); 1661 /* 1662 * Kludge City: If the first reply to the remove rpc is lost.. 1663 * the reply to the retransmitted request will be ENOENT 1664 * since the file was in fact removed 1665 * Therefore, we cheat and return success. 1666 */ 1667 if (error == ENOENT) 1668 error = 0; 1669 } else if (!np->n_sillyrename) 1670 error = nfs_sillyrename(dvp, vp, cnp); 1671 mtx_lock(&np->n_mtx); 1672 np->n_attrstamp = 0; 1673 mtx_unlock(&np->n_mtx); 1674 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 1675 return (error); 1676 } 1677 1678 /* 1679 * nfs file remove rpc called from nfs_inactive 1680 */ 1681 int 1682 ncl_removeit(struct sillyrename *sp, struct vnode *vp) 1683 { 1684 /* 1685 * Make sure that the directory vnode is still valid. 1686 * XXX we should lock sp->s_dvp here. 1687 */ 1688 if (sp->s_dvp->v_type == VBAD) 1689 return (0); 1690 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen, 1691 sp->s_cred, NULL)); 1692 } 1693 1694 /* 1695 * Nfs remove rpc, called from nfs_remove() and ncl_removeit(). 1696 */ 1697 static int 1698 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 1699 int namelen, struct ucred *cred, struct thread *td) 1700 { 1701 struct nfsvattr dnfsva; 1702 struct nfsnode *dnp = VTONFS(dvp); 1703 int error = 0, dattrflag; 1704 1705 mtx_lock(&dnp->n_mtx); 1706 dnp->n_flag |= NREMOVEINPROG; 1707 mtx_unlock(&dnp->n_mtx); 1708 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva, 1709 &dattrflag, NULL); 1710 mtx_lock(&dnp->n_mtx); 1711 if ((dnp->n_flag & NREMOVEWANT)) { 1712 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG); 1713 mtx_unlock(&dnp->n_mtx); 1714 wakeup((caddr_t)dnp); 1715 } else { 1716 dnp->n_flag &= ~NREMOVEINPROG; 1717 mtx_unlock(&dnp->n_mtx); 1718 } 1719 if (dattrflag) 1720 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1721 mtx_lock(&dnp->n_mtx); 1722 dnp->n_flag |= NMODIFIED; 1723 if (!dattrflag) { 1724 dnp->n_attrstamp = 0; 1725 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1726 } 1727 mtx_unlock(&dnp->n_mtx); 1728 if (error && NFS_ISV4(dvp)) 1729 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1730 return (error); 1731 } 1732 1733 /* 1734 * nfs file rename call 1735 */ 1736 static int 1737 nfs_rename(struct vop_rename_args *ap) 1738 { 1739 struct vnode *fvp = ap->a_fvp; 1740 struct vnode *tvp = ap->a_tvp; 1741 struct vnode *fdvp = ap->a_fdvp; 1742 struct vnode *tdvp = ap->a_tdvp; 1743 struct componentname *tcnp = ap->a_tcnp; 1744 struct componentname *fcnp = ap->a_fcnp; 1745 struct nfsnode *fnp = VTONFS(ap->a_fvp); 1746 struct nfsnode *tdnp = VTONFS(ap->a_tdvp); 1747 struct nfsv4node *newv4 = NULL; 1748 int error; 1749 1750 KASSERT((tcnp->cn_flags & HASBUF) != 0 && 1751 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name")); 1752 /* Check for cross-device rename */ 1753 if ((fvp->v_mount != tdvp->v_mount) || 1754 (tvp && (fvp->v_mount != tvp->v_mount))) { 1755 error = EXDEV; 1756 goto out; 1757 } 1758 1759 if (fvp == tvp) { 1760 ncl_printf("nfs_rename: fvp == tvp (can't happen)\n"); 1761 error = 0; 1762 goto out; 1763 } 1764 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0) 1765 goto out; 1766 1767 /* 1768 * We have to flush B_DELWRI data prior to renaming 1769 * the file. If we don't, the delayed-write buffers 1770 * can be flushed out later after the file has gone stale 1771 * under NFSV3. NFSV2 does not have this problem because 1772 * ( as far as I can tell ) it flushes dirty buffers more 1773 * often. 1774 * 1775 * Skip the rename operation if the fsync fails, this can happen 1776 * due to the server's volume being full, when we pushed out data 1777 * that was written back to our cache earlier. Not checking for 1778 * this condition can result in potential (silent) data loss. 1779 */ 1780 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread); 1781 NFSVOPUNLOCK(fvp, 0); 1782 if (!error && tvp) 1783 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread); 1784 if (error) 1785 goto out; 1786 1787 /* 1788 * If the tvp exists and is in use, sillyrename it before doing the 1789 * rename of the new file over it. 1790 * XXX Can't sillyrename a directory. 1791 */ 1792 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename && 1793 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1794 vput(tvp); 1795 tvp = NULL; 1796 } 1797 1798 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1799 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1800 tcnp->cn_thread); 1801 1802 if (error == 0 && NFS_ISV4(tdvp)) { 1803 /* 1804 * For NFSv4, check to see if it is the same name and 1805 * replace the name, if it is different. 1806 */ 1807 MALLOC(newv4, struct nfsv4node *, 1808 sizeof (struct nfsv4node) + 1809 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1, 1810 M_NFSV4NODE, M_WAITOK); 1811 mtx_lock(&tdnp->n_mtx); 1812 mtx_lock(&fnp->n_mtx); 1813 if (fnp->n_v4 != NULL && fvp->v_type == VREG && 1814 (fnp->n_v4->n4_namelen != tcnp->cn_namelen || 1815 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4), 1816 tcnp->cn_namelen) || 1817 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen || 1818 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 1819 tdnp->n_fhp->nfh_len))) { 1820 #ifdef notdef 1821 { char nnn[100]; int nnnl; 1822 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99; 1823 bcopy(tcnp->cn_nameptr, nnn, nnnl); 1824 nnn[nnnl] = '\0'; 1825 printf("ren replace=%s\n",nnn); 1826 } 1827 #endif 1828 FREE((caddr_t)fnp->n_v4, M_NFSV4NODE); 1829 fnp->n_v4 = newv4; 1830 newv4 = NULL; 1831 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len; 1832 fnp->n_v4->n4_namelen = tcnp->cn_namelen; 1833 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 1834 tdnp->n_fhp->nfh_len); 1835 NFSBCOPY(tcnp->cn_nameptr, 1836 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen); 1837 } 1838 mtx_unlock(&tdnp->n_mtx); 1839 mtx_unlock(&fnp->n_mtx); 1840 if (newv4 != NULL) 1841 FREE((caddr_t)newv4, M_NFSV4NODE); 1842 } 1843 1844 if (fvp->v_type == VDIR) { 1845 if (tvp != NULL && tvp->v_type == VDIR) 1846 cache_purge(tdvp); 1847 cache_purge(fdvp); 1848 } 1849 1850 out: 1851 if (tdvp == tvp) 1852 vrele(tdvp); 1853 else 1854 vput(tdvp); 1855 if (tvp) 1856 vput(tvp); 1857 vrele(fdvp); 1858 vrele(fvp); 1859 /* 1860 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1861 */ 1862 if (error == ENOENT) 1863 error = 0; 1864 return (error); 1865 } 1866 1867 /* 1868 * nfs file rename rpc called from nfs_remove() above 1869 */ 1870 static int 1871 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp, 1872 struct sillyrename *sp) 1873 { 1874 1875 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen, 1876 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred, 1877 scnp->cn_thread)); 1878 } 1879 1880 /* 1881 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1882 */ 1883 static int 1884 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr, 1885 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr, 1886 int tnamelen, struct ucred *cred, struct thread *td) 1887 { 1888 struct nfsvattr fnfsva, tnfsva; 1889 struct nfsnode *fdnp = VTONFS(fdvp); 1890 struct nfsnode *tdnp = VTONFS(tdvp); 1891 int error = 0, fattrflag, tattrflag; 1892 1893 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp, 1894 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag, 1895 &tattrflag, NULL, NULL); 1896 mtx_lock(&fdnp->n_mtx); 1897 fdnp->n_flag |= NMODIFIED; 1898 if (fattrflag != 0) { 1899 mtx_unlock(&fdnp->n_mtx); 1900 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1); 1901 } else { 1902 fdnp->n_attrstamp = 0; 1903 mtx_unlock(&fdnp->n_mtx); 1904 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp); 1905 } 1906 mtx_lock(&tdnp->n_mtx); 1907 tdnp->n_flag |= NMODIFIED; 1908 if (tattrflag != 0) { 1909 mtx_unlock(&tdnp->n_mtx); 1910 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1); 1911 } else { 1912 tdnp->n_attrstamp = 0; 1913 mtx_unlock(&tdnp->n_mtx); 1914 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 1915 } 1916 if (error && NFS_ISV4(fdvp)) 1917 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1918 return (error); 1919 } 1920 1921 /* 1922 * nfs hard link create call 1923 */ 1924 static int 1925 nfs_link(struct vop_link_args *ap) 1926 { 1927 struct vnode *vp = ap->a_vp; 1928 struct vnode *tdvp = ap->a_tdvp; 1929 struct componentname *cnp = ap->a_cnp; 1930 struct nfsnode *np, *tdnp; 1931 struct nfsvattr nfsva, dnfsva; 1932 int error = 0, attrflag, dattrflag; 1933 1934 if (vp->v_mount != tdvp->v_mount) { 1935 return (EXDEV); 1936 } 1937 1938 /* 1939 * Push all writes to the server, so that the attribute cache 1940 * doesn't get "out of sync" with the server. 1941 * XXX There should be a better way! 1942 */ 1943 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread); 1944 1945 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 1946 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag, 1947 &dattrflag, NULL); 1948 tdnp = VTONFS(tdvp); 1949 mtx_lock(&tdnp->n_mtx); 1950 tdnp->n_flag |= NMODIFIED; 1951 if (dattrflag != 0) { 1952 mtx_unlock(&tdnp->n_mtx); 1953 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1); 1954 } else { 1955 tdnp->n_attrstamp = 0; 1956 mtx_unlock(&tdnp->n_mtx); 1957 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 1958 } 1959 if (attrflag) 1960 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1961 else { 1962 np = VTONFS(vp); 1963 mtx_lock(&np->n_mtx); 1964 np->n_attrstamp = 0; 1965 mtx_unlock(&np->n_mtx); 1966 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 1967 } 1968 /* 1969 * If negative lookup caching is enabled, I might as well 1970 * add an entry for this node. Not necessary for correctness, 1971 * but if negative caching is enabled, then the system 1972 * must care about lookup caching hit rate, so... 1973 */ 1974 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 && 1975 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 1976 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL); 1977 } 1978 if (error && NFS_ISV4(vp)) 1979 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, 1980 (gid_t)0); 1981 return (error); 1982 } 1983 1984 /* 1985 * nfs symbolic link create call 1986 */ 1987 static int 1988 nfs_symlink(struct vop_symlink_args *ap) 1989 { 1990 struct vnode *dvp = ap->a_dvp; 1991 struct vattr *vap = ap->a_vap; 1992 struct componentname *cnp = ap->a_cnp; 1993 struct nfsvattr nfsva, dnfsva; 1994 struct nfsfh *nfhp; 1995 struct nfsnode *np = NULL, *dnp; 1996 struct vnode *newvp = NULL; 1997 int error = 0, attrflag, dattrflag, ret; 1998 1999 vap->va_type = VLNK; 2000 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2001 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, 2002 &nfsva, &nfhp, &attrflag, &dattrflag, NULL); 2003 if (nfhp) { 2004 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, 2005 &np, NULL, LK_EXCLUSIVE); 2006 if (!ret) 2007 newvp = NFSTOV(np); 2008 else if (!error) 2009 error = ret; 2010 } 2011 if (newvp != NULL) { 2012 if (attrflag) 2013 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 2014 0, 1); 2015 } else if (!error) { 2016 /* 2017 * If we do not have an error and we could not extract the 2018 * newvp from the response due to the request being NFSv2, we 2019 * have to do a lookup in order to obtain a newvp to return. 2020 */ 2021 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2022 cnp->cn_cred, cnp->cn_thread, &np); 2023 if (!error) 2024 newvp = NFSTOV(np); 2025 } 2026 if (error) { 2027 if (newvp) 2028 vput(newvp); 2029 if (NFS_ISV4(dvp)) 2030 error = nfscl_maperr(cnp->cn_thread, error, 2031 vap->va_uid, vap->va_gid); 2032 } else { 2033 *ap->a_vpp = newvp; 2034 } 2035 2036 dnp = VTONFS(dvp); 2037 mtx_lock(&dnp->n_mtx); 2038 dnp->n_flag |= NMODIFIED; 2039 if (dattrflag != 0) { 2040 mtx_unlock(&dnp->n_mtx); 2041 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2042 } else { 2043 dnp->n_attrstamp = 0; 2044 mtx_unlock(&dnp->n_mtx); 2045 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2046 } 2047 /* 2048 * If negative lookup caching is enabled, I might as well 2049 * add an entry for this node. Not necessary for correctness, 2050 * but if negative caching is enabled, then the system 2051 * must care about lookup caching hit rate, so... 2052 */ 2053 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2054 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 2055 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL); 2056 } 2057 return (error); 2058 } 2059 2060 /* 2061 * nfs make dir call 2062 */ 2063 static int 2064 nfs_mkdir(struct vop_mkdir_args *ap) 2065 { 2066 struct vnode *dvp = ap->a_dvp; 2067 struct vattr *vap = ap->a_vap; 2068 struct componentname *cnp = ap->a_cnp; 2069 struct nfsnode *np = NULL, *dnp; 2070 struct vnode *newvp = NULL; 2071 struct vattr vattr; 2072 struct nfsfh *nfhp; 2073 struct nfsvattr nfsva, dnfsva; 2074 int error = 0, attrflag, dattrflag, ret; 2075 2076 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) 2077 return (error); 2078 vap->va_type = VDIR; 2079 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2080 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp, 2081 &attrflag, &dattrflag, NULL); 2082 dnp = VTONFS(dvp); 2083 mtx_lock(&dnp->n_mtx); 2084 dnp->n_flag |= NMODIFIED; 2085 if (dattrflag != 0) { 2086 mtx_unlock(&dnp->n_mtx); 2087 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2088 } else { 2089 dnp->n_attrstamp = 0; 2090 mtx_unlock(&dnp->n_mtx); 2091 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2092 } 2093 if (nfhp) { 2094 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, 2095 &np, NULL, LK_EXCLUSIVE); 2096 if (!ret) { 2097 newvp = NFSTOV(np); 2098 if (attrflag) 2099 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 2100 NULL, 0, 1); 2101 } else if (!error) 2102 error = ret; 2103 } 2104 if (!error && newvp == NULL) { 2105 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2106 cnp->cn_cred, cnp->cn_thread, &np); 2107 if (!error) { 2108 newvp = NFSTOV(np); 2109 if (newvp->v_type != VDIR) 2110 error = EEXIST; 2111 } 2112 } 2113 if (error) { 2114 if (newvp) 2115 vput(newvp); 2116 if (NFS_ISV4(dvp)) 2117 error = nfscl_maperr(cnp->cn_thread, error, 2118 vap->va_uid, vap->va_gid); 2119 } else { 2120 /* 2121 * If negative lookup caching is enabled, I might as well 2122 * add an entry for this node. Not necessary for correctness, 2123 * but if negative caching is enabled, then the system 2124 * must care about lookup caching hit rate, so... 2125 */ 2126 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2127 (cnp->cn_flags & MAKEENTRY) && 2128 attrflag != 0 && dattrflag != 0) 2129 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 2130 &dnfsva.na_ctime); 2131 *ap->a_vpp = newvp; 2132 } 2133 return (error); 2134 } 2135 2136 /* 2137 * nfs remove directory call 2138 */ 2139 static int 2140 nfs_rmdir(struct vop_rmdir_args *ap) 2141 { 2142 struct vnode *vp = ap->a_vp; 2143 struct vnode *dvp = ap->a_dvp; 2144 struct componentname *cnp = ap->a_cnp; 2145 struct nfsnode *dnp; 2146 struct nfsvattr dnfsva; 2147 int error, dattrflag; 2148 2149 if (dvp == vp) 2150 return (EINVAL); 2151 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2152 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL); 2153 dnp = VTONFS(dvp); 2154 mtx_lock(&dnp->n_mtx); 2155 dnp->n_flag |= NMODIFIED; 2156 if (dattrflag != 0) { 2157 mtx_unlock(&dnp->n_mtx); 2158 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2159 } else { 2160 dnp->n_attrstamp = 0; 2161 mtx_unlock(&dnp->n_mtx); 2162 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2163 } 2164 2165 cache_purge(dvp); 2166 cache_purge(vp); 2167 if (error && NFS_ISV4(dvp)) 2168 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, 2169 (gid_t)0); 2170 /* 2171 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2172 */ 2173 if (error == ENOENT) 2174 error = 0; 2175 return (error); 2176 } 2177 2178 /* 2179 * nfs readdir call 2180 */ 2181 static int 2182 nfs_readdir(struct vop_readdir_args *ap) 2183 { 2184 struct vnode *vp = ap->a_vp; 2185 struct nfsnode *np = VTONFS(vp); 2186 struct uio *uio = ap->a_uio; 2187 ssize_t tresid; 2188 int error = 0; 2189 struct vattr vattr; 2190 2191 if (vp->v_type != VDIR) 2192 return(EPERM); 2193 2194 /* 2195 * First, check for hit on the EOF offset cache 2196 */ 2197 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2198 (np->n_flag & NMODIFIED) == 0) { 2199 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) { 2200 mtx_lock(&np->n_mtx); 2201 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) || 2202 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 2203 mtx_unlock(&np->n_mtx); 2204 NFSINCRGLOBAL(newnfsstats.direofcache_hits); 2205 return (0); 2206 } else 2207 mtx_unlock(&np->n_mtx); 2208 } 2209 } 2210 2211 /* 2212 * Call ncl_bioread() to do the real work. 2213 */ 2214 tresid = uio->uio_resid; 2215 error = ncl_bioread(vp, uio, 0, ap->a_cred); 2216 2217 if (!error && uio->uio_resid == tresid) 2218 NFSINCRGLOBAL(newnfsstats.direofcache_misses); 2219 return (error); 2220 } 2221 2222 /* 2223 * Readdir rpc call. 2224 * Called from below the buffer cache by ncl_doio(). 2225 */ 2226 int 2227 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2228 struct thread *td) 2229 { 2230 struct nfsvattr nfsva; 2231 nfsuint64 *cookiep, cookie; 2232 struct nfsnode *dnp = VTONFS(vp); 2233 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2234 int error = 0, eof, attrflag; 2235 2236 KASSERT(uiop->uio_iovcnt == 1 && 2237 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2238 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2239 ("nfs readdirrpc bad uio")); 2240 2241 /* 2242 * If there is no cookie, assume directory was stale. 2243 */ 2244 ncl_dircookie_lock(dnp); 2245 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2246 if (cookiep) { 2247 cookie = *cookiep; 2248 ncl_dircookie_unlock(dnp); 2249 } else { 2250 ncl_dircookie_unlock(dnp); 2251 return (NFSERR_BAD_COOKIE); 2252 } 2253 2254 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2255 (void)ncl_fsinfo(nmp, vp, cred, td); 2256 2257 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva, 2258 &attrflag, &eof, NULL); 2259 if (attrflag) 2260 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2261 2262 if (!error) { 2263 /* 2264 * We are now either at the end of the directory or have filled 2265 * the block. 2266 */ 2267 if (eof) 2268 dnp->n_direofoffset = uiop->uio_offset; 2269 else { 2270 if (uiop->uio_resid > 0) 2271 ncl_printf("EEK! readdirrpc resid > 0\n"); 2272 ncl_dircookie_lock(dnp); 2273 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2274 *cookiep = cookie; 2275 ncl_dircookie_unlock(dnp); 2276 } 2277 } else if (NFS_ISV4(vp)) { 2278 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2279 } 2280 return (error); 2281 } 2282 2283 /* 2284 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc(). 2285 */ 2286 int 2287 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2288 struct thread *td) 2289 { 2290 struct nfsvattr nfsva; 2291 nfsuint64 *cookiep, cookie; 2292 struct nfsnode *dnp = VTONFS(vp); 2293 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2294 int error = 0, attrflag, eof; 2295 2296 KASSERT(uiop->uio_iovcnt == 1 && 2297 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2298 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2299 ("nfs readdirplusrpc bad uio")); 2300 2301 /* 2302 * If there is no cookie, assume directory was stale. 2303 */ 2304 ncl_dircookie_lock(dnp); 2305 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2306 if (cookiep) { 2307 cookie = *cookiep; 2308 ncl_dircookie_unlock(dnp); 2309 } else { 2310 ncl_dircookie_unlock(dnp); 2311 return (NFSERR_BAD_COOKIE); 2312 } 2313 2314 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2315 (void)ncl_fsinfo(nmp, vp, cred, td); 2316 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva, 2317 &attrflag, &eof, NULL); 2318 if (attrflag) 2319 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2320 2321 if (!error) { 2322 /* 2323 * We are now either at end of the directory or have filled the 2324 * the block. 2325 */ 2326 if (eof) 2327 dnp->n_direofoffset = uiop->uio_offset; 2328 else { 2329 if (uiop->uio_resid > 0) 2330 ncl_printf("EEK! readdirplusrpc resid > 0\n"); 2331 ncl_dircookie_lock(dnp); 2332 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2333 *cookiep = cookie; 2334 ncl_dircookie_unlock(dnp); 2335 } 2336 } else if (NFS_ISV4(vp)) { 2337 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2338 } 2339 return (error); 2340 } 2341 2342 /* 2343 * Silly rename. To make the NFS filesystem that is stateless look a little 2344 * more like the "ufs" a remove of an active vnode is translated to a rename 2345 * to a funny looking filename that is removed by nfs_inactive on the 2346 * nfsnode. There is the potential for another process on a different client 2347 * to create the same funny name between the nfs_lookitup() fails and the 2348 * nfs_rename() completes, but... 2349 */ 2350 static int 2351 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2352 { 2353 struct sillyrename *sp; 2354 struct nfsnode *np; 2355 int error; 2356 short pid; 2357 unsigned int lticks; 2358 2359 cache_purge(dvp); 2360 np = VTONFS(vp); 2361 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir")); 2362 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2363 M_NEWNFSREQ, M_WAITOK); 2364 sp->s_cred = crhold(cnp->cn_cred); 2365 sp->s_dvp = dvp; 2366 VREF(dvp); 2367 2368 /* 2369 * Fudge together a funny name. 2370 * Changing the format of the funny name to accomodate more 2371 * sillynames per directory. 2372 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is 2373 * CPU ticks since boot. 2374 */ 2375 pid = cnp->cn_thread->td_proc->p_pid; 2376 lticks = (unsigned int)ticks; 2377 for ( ; ; ) { 2378 sp->s_namlen = sprintf(sp->s_name, 2379 ".nfs.%08x.%04x4.4", lticks, 2380 pid); 2381 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2382 cnp->cn_thread, NULL)) 2383 break; 2384 lticks++; 2385 } 2386 error = nfs_renameit(dvp, vp, cnp, sp); 2387 if (error) 2388 goto bad; 2389 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2390 cnp->cn_thread, &np); 2391 np->n_sillyrename = sp; 2392 return (0); 2393 bad: 2394 vrele(sp->s_dvp); 2395 crfree(sp->s_cred); 2396 free((caddr_t)sp, M_NEWNFSREQ); 2397 return (error); 2398 } 2399 2400 /* 2401 * Look up a file name and optionally either update the file handle or 2402 * allocate an nfsnode, depending on the value of npp. 2403 * npp == NULL --> just do the lookup 2404 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2405 * handled too 2406 * *npp != NULL --> update the file handle in the vnode 2407 */ 2408 static int 2409 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred, 2410 struct thread *td, struct nfsnode **npp) 2411 { 2412 struct vnode *newvp = NULL, *vp; 2413 struct nfsnode *np, *dnp = VTONFS(dvp); 2414 struct nfsfh *nfhp, *onfhp; 2415 struct nfsvattr nfsva, dnfsva; 2416 struct componentname cn; 2417 int error = 0, attrflag, dattrflag; 2418 u_int hash; 2419 2420 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva, 2421 &nfhp, &attrflag, &dattrflag, NULL); 2422 if (dattrflag) 2423 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2424 if (npp && !error) { 2425 if (*npp != NULL) { 2426 np = *npp; 2427 vp = NFSTOV(np); 2428 /* 2429 * For NFSv4, check to see if it is the same name and 2430 * replace the name, if it is different. 2431 */ 2432 if (np->n_v4 != NULL && nfsva.na_type == VREG && 2433 (np->n_v4->n4_namelen != len || 2434 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) || 2435 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || 2436 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2437 dnp->n_fhp->nfh_len))) { 2438 #ifdef notdef 2439 { char nnn[100]; int nnnl; 2440 nnnl = (len < 100) ? len : 99; 2441 bcopy(name, nnn, nnnl); 2442 nnn[nnnl] = '\0'; 2443 printf("replace=%s\n",nnn); 2444 } 2445 #endif 2446 FREE((caddr_t)np->n_v4, M_NFSV4NODE); 2447 MALLOC(np->n_v4, struct nfsv4node *, 2448 sizeof (struct nfsv4node) + 2449 dnp->n_fhp->nfh_len + len - 1, 2450 M_NFSV4NODE, M_WAITOK); 2451 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; 2452 np->n_v4->n4_namelen = len; 2453 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2454 dnp->n_fhp->nfh_len); 2455 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len); 2456 } 2457 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, 2458 FNV1_32_INIT); 2459 onfhp = np->n_fhp; 2460 /* 2461 * Rehash node for new file handle. 2462 */ 2463 vfs_hash_rehash(vp, hash); 2464 np->n_fhp = nfhp; 2465 if (onfhp != NULL) 2466 FREE((caddr_t)onfhp, M_NFSFH); 2467 newvp = NFSTOV(np); 2468 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) { 2469 FREE((caddr_t)nfhp, M_NFSFH); 2470 VREF(dvp); 2471 newvp = dvp; 2472 } else { 2473 cn.cn_nameptr = name; 2474 cn.cn_namelen = len; 2475 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td, 2476 &np, NULL, LK_EXCLUSIVE); 2477 if (error) 2478 return (error); 2479 newvp = NFSTOV(np); 2480 } 2481 if (!attrflag && *npp == NULL) { 2482 if (newvp == dvp) 2483 vrele(newvp); 2484 else 2485 vput(newvp); 2486 return (ENOENT); 2487 } 2488 if (attrflag) 2489 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 2490 0, 1); 2491 } 2492 if (npp && *npp == NULL) { 2493 if (error) { 2494 if (newvp) { 2495 if (newvp == dvp) 2496 vrele(newvp); 2497 else 2498 vput(newvp); 2499 } 2500 } else 2501 *npp = np; 2502 } 2503 if (error && NFS_ISV4(dvp)) 2504 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2505 return (error); 2506 } 2507 2508 /* 2509 * Nfs Version 3 and 4 commit rpc 2510 */ 2511 int 2512 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred, 2513 struct thread *td) 2514 { 2515 struct nfsvattr nfsva; 2516 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2517 int error, attrflag; 2518 u_char verf[NFSX_VERF]; 2519 2520 mtx_lock(&nmp->nm_mtx); 2521 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) { 2522 mtx_unlock(&nmp->nm_mtx); 2523 return (0); 2524 } 2525 mtx_unlock(&nmp->nm_mtx); 2526 error = nfsrpc_commit(vp, offset, cnt, cred, td, verf, &nfsva, 2527 &attrflag, NULL); 2528 if (!error) { 2529 mtx_lock(&nmp->nm_mtx); 2530 if (NFSBCMP((caddr_t)nmp->nm_verf, verf, NFSX_VERF)) { 2531 NFSBCOPY(verf, (caddr_t)nmp->nm_verf, NFSX_VERF); 2532 error = NFSERR_STALEWRITEVERF; 2533 } 2534 mtx_unlock(&nmp->nm_mtx); 2535 if (!error && attrflag) 2536 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 2537 0, 1); 2538 } else if (NFS_ISV4(vp)) { 2539 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2540 } 2541 return (error); 2542 } 2543 2544 /* 2545 * Strategy routine. 2546 * For async requests when nfsiod(s) are running, queue the request by 2547 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the 2548 * request. 2549 */ 2550 static int 2551 nfs_strategy(struct vop_strategy_args *ap) 2552 { 2553 struct buf *bp = ap->a_bp; 2554 struct ucred *cr; 2555 2556 KASSERT(!(bp->b_flags & B_DONE), 2557 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); 2558 BUF_ASSERT_HELD(bp); 2559 2560 if (bp->b_iocmd == BIO_READ) 2561 cr = bp->b_rcred; 2562 else 2563 cr = bp->b_wcred; 2564 2565 /* 2566 * If the op is asynchronous and an i/o daemon is waiting 2567 * queue the request, wake it up and wait for completion 2568 * otherwise just do it ourselves. 2569 */ 2570 if ((bp->b_flags & B_ASYNC) == 0 || 2571 ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread)) 2572 (void) ncl_doio(ap->a_vp, bp, cr, curthread, 1); 2573 return (0); 2574 } 2575 2576 /* 2577 * fsync vnode op. Just call ncl_flush() with commit == 1. 2578 */ 2579 /* ARGSUSED */ 2580 static int 2581 nfs_fsync(struct vop_fsync_args *ap) 2582 { 2583 2584 if (ap->a_vp->v_type != VREG) { 2585 /* 2586 * For NFS, metadata is changed synchronously on the server, 2587 * so there is nothing to flush. Also, ncl_flush() clears 2588 * the NMODIFIED flag and that shouldn't be done here for 2589 * directories. 2590 */ 2591 return (0); 2592 } 2593 return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1, 0)); 2594 } 2595 2596 /* 2597 * Flush all the blocks associated with a vnode. 2598 * Walk through the buffer pool and push any dirty pages 2599 * associated with the vnode. 2600 * If the called_from_renewthread argument is TRUE, it has been called 2601 * from the NFSv4 renew thread and, as such, cannot block indefinitely 2602 * waiting for a buffer write to complete. 2603 */ 2604 int 2605 ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td, 2606 int commit, int called_from_renewthread) 2607 { 2608 struct nfsnode *np = VTONFS(vp); 2609 struct buf *bp; 2610 int i; 2611 struct buf *nbp; 2612 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2613 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos; 2614 int passone = 1, trycnt = 0; 2615 u_quad_t off, endoff, toff; 2616 struct ucred* wcred = NULL; 2617 struct buf **bvec = NULL; 2618 struct bufobj *bo; 2619 #ifndef NFS_COMMITBVECSIZ 2620 #define NFS_COMMITBVECSIZ 20 2621 #endif 2622 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ]; 2623 int bvecsize = 0, bveccount; 2624 2625 if (called_from_renewthread != 0) 2626 slptimeo = hz; 2627 if (nmp->nm_flag & NFSMNT_INT) 2628 slpflag = NFS_PCATCH; 2629 if (!commit) 2630 passone = 0; 2631 bo = &vp->v_bufobj; 2632 /* 2633 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the 2634 * server, but has not been committed to stable storage on the server 2635 * yet. On the first pass, the byte range is worked out and the commit 2636 * rpc is done. On the second pass, ncl_writebp() is called to do the 2637 * job. 2638 */ 2639 again: 2640 off = (u_quad_t)-1; 2641 endoff = 0; 2642 bvecpos = 0; 2643 if (NFS_ISV34(vp) && commit) { 2644 if (bvec != NULL && bvec != bvec_on_stack) 2645 free(bvec, M_TEMP); 2646 /* 2647 * Count up how many buffers waiting for a commit. 2648 */ 2649 bveccount = 0; 2650 BO_LOCK(bo); 2651 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2652 if (!BUF_ISLOCKED(bp) && 2653 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) 2654 == (B_DELWRI | B_NEEDCOMMIT)) 2655 bveccount++; 2656 } 2657 /* 2658 * Allocate space to remember the list of bufs to commit. It is 2659 * important to use M_NOWAIT here to avoid a race with nfs_write. 2660 * If we can't get memory (for whatever reason), we will end up 2661 * committing the buffers one-by-one in the loop below. 2662 */ 2663 if (bveccount > NFS_COMMITBVECSIZ) { 2664 /* 2665 * Release the vnode interlock to avoid a lock 2666 * order reversal. 2667 */ 2668 BO_UNLOCK(bo); 2669 bvec = (struct buf **) 2670 malloc(bveccount * sizeof(struct buf *), 2671 M_TEMP, M_NOWAIT); 2672 BO_LOCK(bo); 2673 if (bvec == NULL) { 2674 bvec = bvec_on_stack; 2675 bvecsize = NFS_COMMITBVECSIZ; 2676 } else 2677 bvecsize = bveccount; 2678 } else { 2679 bvec = bvec_on_stack; 2680 bvecsize = NFS_COMMITBVECSIZ; 2681 } 2682 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2683 if (bvecpos >= bvecsize) 2684 break; 2685 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2686 nbp = TAILQ_NEXT(bp, b_bobufs); 2687 continue; 2688 } 2689 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 2690 (B_DELWRI | B_NEEDCOMMIT)) { 2691 BUF_UNLOCK(bp); 2692 nbp = TAILQ_NEXT(bp, b_bobufs); 2693 continue; 2694 } 2695 BO_UNLOCK(bo); 2696 bremfree(bp); 2697 /* 2698 * Work out if all buffers are using the same cred 2699 * so we can deal with them all with one commit. 2700 * 2701 * NOTE: we are not clearing B_DONE here, so we have 2702 * to do it later on in this routine if we intend to 2703 * initiate I/O on the bp. 2704 * 2705 * Note: to avoid loopback deadlocks, we do not 2706 * assign b_runningbufspace. 2707 */ 2708 if (wcred == NULL) 2709 wcred = bp->b_wcred; 2710 else if (wcred != bp->b_wcred) 2711 wcred = NOCRED; 2712 vfs_busy_pages(bp, 1); 2713 2714 BO_LOCK(bo); 2715 /* 2716 * bp is protected by being locked, but nbp is not 2717 * and vfs_busy_pages() may sleep. We have to 2718 * recalculate nbp. 2719 */ 2720 nbp = TAILQ_NEXT(bp, b_bobufs); 2721 2722 /* 2723 * A list of these buffers is kept so that the 2724 * second loop knows which buffers have actually 2725 * been committed. This is necessary, since there 2726 * may be a race between the commit rpc and new 2727 * uncommitted writes on the file. 2728 */ 2729 bvec[bvecpos++] = bp; 2730 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2731 bp->b_dirtyoff; 2732 if (toff < off) 2733 off = toff; 2734 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); 2735 if (toff > endoff) 2736 endoff = toff; 2737 } 2738 BO_UNLOCK(bo); 2739 } 2740 if (bvecpos > 0) { 2741 /* 2742 * Commit data on the server, as required. 2743 * If all bufs are using the same wcred, then use that with 2744 * one call for all of them, otherwise commit each one 2745 * separately. 2746 */ 2747 if (wcred != NOCRED) 2748 retv = ncl_commit(vp, off, (int)(endoff - off), 2749 wcred, td); 2750 else { 2751 retv = 0; 2752 for (i = 0; i < bvecpos; i++) { 2753 off_t off, size; 2754 bp = bvec[i]; 2755 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2756 bp->b_dirtyoff; 2757 size = (u_quad_t)(bp->b_dirtyend 2758 - bp->b_dirtyoff); 2759 retv = ncl_commit(vp, off, (int)size, 2760 bp->b_wcred, td); 2761 if (retv) break; 2762 } 2763 } 2764 2765 if (retv == NFSERR_STALEWRITEVERF) 2766 ncl_clearcommit(vp->v_mount); 2767 2768 /* 2769 * Now, either mark the blocks I/O done or mark the 2770 * blocks dirty, depending on whether the commit 2771 * succeeded. 2772 */ 2773 for (i = 0; i < bvecpos; i++) { 2774 bp = bvec[i]; 2775 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 2776 if (retv) { 2777 /* 2778 * Error, leave B_DELWRI intact 2779 */ 2780 vfs_unbusy_pages(bp); 2781 brelse(bp); 2782 } else { 2783 /* 2784 * Success, remove B_DELWRI ( bundirty() ). 2785 * 2786 * b_dirtyoff/b_dirtyend seem to be NFS 2787 * specific. We should probably move that 2788 * into bundirty(). XXX 2789 */ 2790 bufobj_wref(bo); 2791 bp->b_flags |= B_ASYNC; 2792 bundirty(bp); 2793 bp->b_flags &= ~B_DONE; 2794 bp->b_ioflags &= ~BIO_ERROR; 2795 bp->b_dirtyoff = bp->b_dirtyend = 0; 2796 bufdone(bp); 2797 } 2798 } 2799 } 2800 2801 /* 2802 * Start/do any write(s) that are required. 2803 */ 2804 loop: 2805 BO_LOCK(bo); 2806 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2807 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2808 if (waitfor != MNT_WAIT || passone) 2809 continue; 2810 2811 error = BUF_TIMELOCK(bp, 2812 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 2813 BO_MTX(bo), "nfsfsync", slpflag, slptimeo); 2814 if (error == 0) { 2815 BUF_UNLOCK(bp); 2816 goto loop; 2817 } 2818 if (error == ENOLCK) { 2819 error = 0; 2820 goto loop; 2821 } 2822 if (called_from_renewthread != 0) { 2823 /* 2824 * Return EIO so the flush will be retried 2825 * later. 2826 */ 2827 error = EIO; 2828 goto done; 2829 } 2830 if (newnfs_sigintr(nmp, td)) { 2831 error = EINTR; 2832 goto done; 2833 } 2834 if (slpflag & PCATCH) { 2835 slpflag = 0; 2836 slptimeo = 2 * hz; 2837 } 2838 goto loop; 2839 } 2840 if ((bp->b_flags & B_DELWRI) == 0) 2841 panic("nfs_fsync: not dirty"); 2842 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) { 2843 BUF_UNLOCK(bp); 2844 continue; 2845 } 2846 BO_UNLOCK(bo); 2847 bremfree(bp); 2848 if (passone || !commit) 2849 bp->b_flags |= B_ASYNC; 2850 else 2851 bp->b_flags |= B_ASYNC; 2852 bwrite(bp); 2853 if (newnfs_sigintr(nmp, td)) { 2854 error = EINTR; 2855 goto done; 2856 } 2857 goto loop; 2858 } 2859 if (passone) { 2860 passone = 0; 2861 BO_UNLOCK(bo); 2862 goto again; 2863 } 2864 if (waitfor == MNT_WAIT) { 2865 while (bo->bo_numoutput) { 2866 error = bufobj_wwait(bo, slpflag, slptimeo); 2867 if (error) { 2868 BO_UNLOCK(bo); 2869 if (called_from_renewthread != 0) { 2870 /* 2871 * Return EIO so that the flush will be 2872 * retried later. 2873 */ 2874 error = EIO; 2875 goto done; 2876 } 2877 error = newnfs_sigintr(nmp, td); 2878 if (error) 2879 goto done; 2880 if (slpflag & PCATCH) { 2881 slpflag = 0; 2882 slptimeo = 2 * hz; 2883 } 2884 BO_LOCK(bo); 2885 } 2886 } 2887 if (bo->bo_dirty.bv_cnt != 0 && commit) { 2888 BO_UNLOCK(bo); 2889 goto loop; 2890 } 2891 /* 2892 * Wait for all the async IO requests to drain 2893 */ 2894 BO_UNLOCK(bo); 2895 mtx_lock(&np->n_mtx); 2896 while (np->n_directio_asyncwr > 0) { 2897 np->n_flag |= NFSYNCWAIT; 2898 error = newnfs_msleep(td, &np->n_directio_asyncwr, 2899 &np->n_mtx, slpflag | (PRIBIO + 1), 2900 "nfsfsync", 0); 2901 if (error) { 2902 if (newnfs_sigintr(nmp, td)) { 2903 mtx_unlock(&np->n_mtx); 2904 error = EINTR; 2905 goto done; 2906 } 2907 } 2908 } 2909 mtx_unlock(&np->n_mtx); 2910 } else 2911 BO_UNLOCK(bo); 2912 mtx_lock(&np->n_mtx); 2913 if (np->n_flag & NWRITEERR) { 2914 error = np->n_error; 2915 np->n_flag &= ~NWRITEERR; 2916 } 2917 if (commit && bo->bo_dirty.bv_cnt == 0 && 2918 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0) 2919 np->n_flag &= ~NMODIFIED; 2920 mtx_unlock(&np->n_mtx); 2921 done: 2922 if (bvec != NULL && bvec != bvec_on_stack) 2923 free(bvec, M_TEMP); 2924 if (error == 0 && commit != 0 && waitfor == MNT_WAIT && 2925 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 || 2926 np->n_directio_asyncwr != 0) && trycnt++ < 5) { 2927 /* try, try again... */ 2928 passone = 1; 2929 wcred = NULL; 2930 bvec = NULL; 2931 bvecsize = 0; 2932 printf("try%d\n", trycnt); 2933 goto again; 2934 } 2935 return (error); 2936 } 2937 2938 /* 2939 * NFS advisory byte-level locks. 2940 */ 2941 static int 2942 nfs_advlock(struct vop_advlock_args *ap) 2943 { 2944 struct vnode *vp = ap->a_vp; 2945 struct ucred *cred; 2946 struct nfsnode *np = VTONFS(ap->a_vp); 2947 struct proc *p = (struct proc *)ap->a_id; 2948 struct thread *td = curthread; /* XXX */ 2949 struct vattr va; 2950 int ret, error = EOPNOTSUPP; 2951 u_quad_t size; 2952 2953 if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) { 2954 if (vp->v_type != VREG) 2955 return (EINVAL); 2956 if ((ap->a_flags & F_POSIX) != 0) 2957 cred = p->p_ucred; 2958 else 2959 cred = td->td_ucred; 2960 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 2961 if (vp->v_iflag & VI_DOOMED) { 2962 NFSVOPUNLOCK(vp, 0); 2963 return (EBADF); 2964 } 2965 2966 /* 2967 * If this is unlocking a write locked region, flush and 2968 * commit them before unlocking. This is required by 2969 * RFC3530 Sec. 9.3.2. 2970 */ 2971 if (ap->a_op == F_UNLCK && 2972 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id, 2973 ap->a_flags)) 2974 (void) ncl_flush(vp, MNT_WAIT, cred, td, 1, 0); 2975 2976 /* 2977 * Loop around doing the lock op, while a blocking lock 2978 * must wait for the lock op to succeed. 2979 */ 2980 do { 2981 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op, 2982 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags); 2983 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 2984 ap->a_op == F_SETLK) { 2985 NFSVOPUNLOCK(vp, 0); 2986 error = nfs_catnap(PZERO | PCATCH, ret, 2987 "ncladvl"); 2988 if (error) 2989 return (EINTR); 2990 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 2991 if (vp->v_iflag & VI_DOOMED) { 2992 NFSVOPUNLOCK(vp, 0); 2993 return (EBADF); 2994 } 2995 } 2996 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 2997 ap->a_op == F_SETLK); 2998 if (ret == NFSERR_DENIED) { 2999 NFSVOPUNLOCK(vp, 0); 3000 return (EAGAIN); 3001 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) { 3002 NFSVOPUNLOCK(vp, 0); 3003 return (ret); 3004 } else if (ret != 0) { 3005 NFSVOPUNLOCK(vp, 0); 3006 return (EACCES); 3007 } 3008 3009 /* 3010 * Now, if we just got a lock, invalidate data in the buffer 3011 * cache, as required, so that the coherency conforms with 3012 * RFC3530 Sec. 9.3.2. 3013 */ 3014 if (ap->a_op == F_SETLK) { 3015 if ((np->n_flag & NMODIFIED) == 0) { 3016 np->n_attrstamp = 0; 3017 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3018 ret = VOP_GETATTR(vp, &va, cred); 3019 } 3020 if ((np->n_flag & NMODIFIED) || ret || 3021 np->n_change != va.va_filerev) { 3022 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1); 3023 np->n_attrstamp = 0; 3024 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3025 ret = VOP_GETATTR(vp, &va, cred); 3026 if (!ret) { 3027 np->n_mtime = va.va_mtime; 3028 np->n_change = va.va_filerev; 3029 } 3030 } 3031 } 3032 NFSVOPUNLOCK(vp, 0); 3033 return (0); 3034 } else if (!NFS_ISV4(vp)) { 3035 error = NFSVOPLOCK(vp, LK_SHARED); 3036 if (error) 3037 return (error); 3038 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3039 size = VTONFS(vp)->n_size; 3040 NFSVOPUNLOCK(vp, 0); 3041 error = lf_advlock(ap, &(vp->v_lockf), size); 3042 } else { 3043 if (nfs_advlock_p != NULL) 3044 error = nfs_advlock_p(ap); 3045 else { 3046 NFSVOPUNLOCK(vp, 0); 3047 error = ENOLCK; 3048 } 3049 } 3050 } 3051 return (error); 3052 } 3053 3054 /* 3055 * NFS advisory byte-level locks. 3056 */ 3057 static int 3058 nfs_advlockasync(struct vop_advlockasync_args *ap) 3059 { 3060 struct vnode *vp = ap->a_vp; 3061 u_quad_t size; 3062 int error; 3063 3064 if (NFS_ISV4(vp)) 3065 return (EOPNOTSUPP); 3066 error = NFSVOPLOCK(vp, LK_SHARED); 3067 if (error) 3068 return (error); 3069 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3070 size = VTONFS(vp)->n_size; 3071 NFSVOPUNLOCK(vp, 0); 3072 error = lf_advlockasync(ap, &(vp->v_lockf), size); 3073 } else { 3074 NFSVOPUNLOCK(vp, 0); 3075 error = EOPNOTSUPP; 3076 } 3077 return (error); 3078 } 3079 3080 /* 3081 * Print out the contents of an nfsnode. 3082 */ 3083 static int 3084 nfs_print(struct vop_print_args *ap) 3085 { 3086 struct vnode *vp = ap->a_vp; 3087 struct nfsnode *np = VTONFS(vp); 3088 3089 ncl_printf("\tfileid %ld fsid 0x%x", 3090 np->n_vattr.na_fileid, np->n_vattr.na_fsid); 3091 if (vp->v_type == VFIFO) 3092 fifo_printinfo(vp); 3093 printf("\n"); 3094 return (0); 3095 } 3096 3097 /* 3098 * This is the "real" nfs::bwrite(struct buf*). 3099 * We set B_CACHE if this is a VMIO buffer. 3100 */ 3101 int 3102 ncl_writebp(struct buf *bp, int force __unused, struct thread *td) 3103 { 3104 int s; 3105 int oldflags = bp->b_flags; 3106 #if 0 3107 int retv = 1; 3108 off_t off; 3109 #endif 3110 3111 BUF_ASSERT_HELD(bp); 3112 3113 if (bp->b_flags & B_INVAL) { 3114 brelse(bp); 3115 return(0); 3116 } 3117 3118 bp->b_flags |= B_CACHE; 3119 3120 /* 3121 * Undirty the bp. We will redirty it later if the I/O fails. 3122 */ 3123 3124 s = splbio(); 3125 bundirty(bp); 3126 bp->b_flags &= ~B_DONE; 3127 bp->b_ioflags &= ~BIO_ERROR; 3128 bp->b_iocmd = BIO_WRITE; 3129 3130 bufobj_wref(bp->b_bufobj); 3131 curthread->td_ru.ru_oublock++; 3132 splx(s); 3133 3134 /* 3135 * Note: to avoid loopback deadlocks, we do not 3136 * assign b_runningbufspace. 3137 */ 3138 vfs_busy_pages(bp, 1); 3139 3140 BUF_KERNPROC(bp); 3141 bp->b_iooffset = dbtob(bp->b_blkno); 3142 bstrategy(bp); 3143 3144 if( (oldflags & B_ASYNC) == 0) { 3145 int rtval = bufwait(bp); 3146 3147 if (oldflags & B_DELWRI) { 3148 s = splbio(); 3149 reassignbuf(bp); 3150 splx(s); 3151 } 3152 brelse(bp); 3153 return (rtval); 3154 } 3155 3156 return (0); 3157 } 3158 3159 /* 3160 * nfs special file access vnode op. 3161 * Essentially just get vattr and then imitate iaccess() since the device is 3162 * local to the client. 3163 */ 3164 static int 3165 nfsspec_access(struct vop_access_args *ap) 3166 { 3167 struct vattr *vap; 3168 struct ucred *cred = ap->a_cred; 3169 struct vnode *vp = ap->a_vp; 3170 accmode_t accmode = ap->a_accmode; 3171 struct vattr vattr; 3172 int error; 3173 3174 /* 3175 * Disallow write attempts on filesystems mounted read-only; 3176 * unless the file is a socket, fifo, or a block or character 3177 * device resident on the filesystem. 3178 */ 3179 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3180 switch (vp->v_type) { 3181 case VREG: 3182 case VDIR: 3183 case VLNK: 3184 return (EROFS); 3185 default: 3186 break; 3187 } 3188 } 3189 vap = &vattr; 3190 error = VOP_GETATTR(vp, vap, cred); 3191 if (error) 3192 goto out; 3193 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid, 3194 accmode, cred, NULL); 3195 out: 3196 return error; 3197 } 3198 3199 /* 3200 * Read wrapper for fifos. 3201 */ 3202 static int 3203 nfsfifo_read(struct vop_read_args *ap) 3204 { 3205 struct nfsnode *np = VTONFS(ap->a_vp); 3206 int error; 3207 3208 /* 3209 * Set access flag. 3210 */ 3211 mtx_lock(&np->n_mtx); 3212 np->n_flag |= NACC; 3213 getnanotime(&np->n_atim); 3214 mtx_unlock(&np->n_mtx); 3215 error = fifo_specops.vop_read(ap); 3216 return error; 3217 } 3218 3219 /* 3220 * Write wrapper for fifos. 3221 */ 3222 static int 3223 nfsfifo_write(struct vop_write_args *ap) 3224 { 3225 struct nfsnode *np = VTONFS(ap->a_vp); 3226 3227 /* 3228 * Set update flag. 3229 */ 3230 mtx_lock(&np->n_mtx); 3231 np->n_flag |= NUPD; 3232 getnanotime(&np->n_mtim); 3233 mtx_unlock(&np->n_mtx); 3234 return(fifo_specops.vop_write(ap)); 3235 } 3236 3237 /* 3238 * Close wrapper for fifos. 3239 * 3240 * Update the times on the nfsnode then do fifo close. 3241 */ 3242 static int 3243 nfsfifo_close(struct vop_close_args *ap) 3244 { 3245 struct vnode *vp = ap->a_vp; 3246 struct nfsnode *np = VTONFS(vp); 3247 struct vattr vattr; 3248 struct timespec ts; 3249 3250 mtx_lock(&np->n_mtx); 3251 if (np->n_flag & (NACC | NUPD)) { 3252 getnanotime(&ts); 3253 if (np->n_flag & NACC) 3254 np->n_atim = ts; 3255 if (np->n_flag & NUPD) 3256 np->n_mtim = ts; 3257 np->n_flag |= NCHG; 3258 if (vrefcnt(vp) == 1 && 3259 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3260 VATTR_NULL(&vattr); 3261 if (np->n_flag & NACC) 3262 vattr.va_atime = np->n_atim; 3263 if (np->n_flag & NUPD) 3264 vattr.va_mtime = np->n_mtim; 3265 mtx_unlock(&np->n_mtx); 3266 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3267 goto out; 3268 } 3269 } 3270 mtx_unlock(&np->n_mtx); 3271 out: 3272 return (fifo_specops.vop_close(ap)); 3273 } 3274 3275 /* 3276 * Just call ncl_writebp() with the force argument set to 1. 3277 * 3278 * NOTE: B_DONE may or may not be set in a_bp on call. 3279 */ 3280 static int 3281 nfs_bwrite(struct buf *bp) 3282 { 3283 3284 return (ncl_writebp(bp, 1, curthread)); 3285 } 3286 3287 struct buf_ops buf_ops_newnfs = { 3288 .bop_name = "buf_ops_nfs", 3289 .bop_write = nfs_bwrite, 3290 .bop_strategy = bufstrategy, 3291 .bop_sync = bufsync, 3292 .bop_bdflush = bufbdflush, 3293 }; 3294 3295 /* 3296 * Cloned from vop_stdlock(), and then the ugly hack added. 3297 */ 3298 static int 3299 nfs_lock1(struct vop_lock1_args *ap) 3300 { 3301 struct vnode *vp = ap->a_vp; 3302 int error = 0; 3303 3304 /* 3305 * Since vfs_hash_get() calls vget() and it will no longer work 3306 * for FreeBSD8 with flags == 0, I can only think of this horrible 3307 * hack to work around it. I call vfs_hash_get() with LK_EXCLOTHER 3308 * and then handle it here. All I want for this case is a v_usecount 3309 * on the vnode to use for recovery, while another thread might 3310 * hold a lock on the vnode. I have the other threads blocked, so 3311 * there isn't any race problem. 3312 */ 3313 if ((ap->a_flags & LK_TYPE_MASK) == LK_EXCLOTHER) { 3314 if ((ap->a_flags & LK_INTERLOCK) == 0) 3315 panic("ncllock1"); 3316 if ((vp->v_iflag & VI_DOOMED)) 3317 error = ENOENT; 3318 VI_UNLOCK(vp); 3319 return (error); 3320 } 3321 return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp), 3322 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file, 3323 ap->a_line)); 3324 } 3325 3326 static int 3327 nfs_getacl(struct vop_getacl_args *ap) 3328 { 3329 int error; 3330 3331 if (ap->a_type != ACL_TYPE_NFS4) 3332 return (EOPNOTSUPP); 3333 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3334 NULL); 3335 if (error > NFSERR_STALE) { 3336 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3337 error = EPERM; 3338 } 3339 return (error); 3340 } 3341 3342 static int 3343 nfs_setacl(struct vop_setacl_args *ap) 3344 { 3345 int error; 3346 3347 if (ap->a_type != ACL_TYPE_NFS4) 3348 return (EOPNOTSUPP); 3349 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3350 NULL); 3351 if (error > NFSERR_STALE) { 3352 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3353 error = EPERM; 3354 } 3355 return (error); 3356 } 3357 3358 /* 3359 * Return POSIX pathconf information applicable to nfs filesystems. 3360 */ 3361 static int 3362 nfs_pathconf(struct vop_pathconf_args *ap) 3363 { 3364 struct nfsv3_pathconf pc; 3365 struct nfsvattr nfsva; 3366 struct vnode *vp = ap->a_vp; 3367 struct thread *td = curthread; 3368 int attrflag, error; 3369 3370 if (NFS_ISV4(vp) || (NFS_ISV3(vp) && (ap->a_name == _PC_LINK_MAX || 3371 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED || 3372 ap->a_name == _PC_NO_TRUNC))) { 3373 /* 3374 * Since only the above 4 a_names are returned by the NFSv3 3375 * Pathconf RPC, there is no point in doing it for others. 3376 */ 3377 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva, 3378 &attrflag, NULL); 3379 if (attrflag != 0) 3380 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 3381 1); 3382 if (error != 0) 3383 return (error); 3384 } else { 3385 /* 3386 * For NFSv2 (or NFSv3 when not one of the above 4 a_names), 3387 * just fake them. 3388 */ 3389 pc.pc_linkmax = LINK_MAX; 3390 pc.pc_namemax = NFS_MAXNAMLEN; 3391 pc.pc_notrunc = 1; 3392 pc.pc_chownrestricted = 1; 3393 pc.pc_caseinsensitive = 0; 3394 pc.pc_casepreserving = 1; 3395 error = 0; 3396 } 3397 switch (ap->a_name) { 3398 case _PC_LINK_MAX: 3399 *ap->a_retval = pc.pc_linkmax; 3400 break; 3401 case _PC_NAME_MAX: 3402 *ap->a_retval = pc.pc_namemax; 3403 break; 3404 case _PC_PATH_MAX: 3405 *ap->a_retval = PATH_MAX; 3406 break; 3407 case _PC_PIPE_BUF: 3408 *ap->a_retval = PIPE_BUF; 3409 break; 3410 case _PC_CHOWN_RESTRICTED: 3411 *ap->a_retval = pc.pc_chownrestricted; 3412 break; 3413 case _PC_NO_TRUNC: 3414 *ap->a_retval = pc.pc_notrunc; 3415 break; 3416 case _PC_ACL_EXTENDED: 3417 *ap->a_retval = 0; 3418 break; 3419 case _PC_ACL_NFS4: 3420 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 && 3421 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL)) 3422 *ap->a_retval = 1; 3423 else 3424 *ap->a_retval = 0; 3425 break; 3426 case _PC_ACL_PATH_MAX: 3427 if (NFS_ISV4(vp)) 3428 *ap->a_retval = ACL_MAX_ENTRIES; 3429 else 3430 *ap->a_retval = 3; 3431 break; 3432 case _PC_MAC_PRESENT: 3433 *ap->a_retval = 0; 3434 break; 3435 case _PC_ASYNC_IO: 3436 /* _PC_ASYNC_IO should have been handled by upper layers. */ 3437 KASSERT(0, ("_PC_ASYNC_IO should not get here")); 3438 error = EINVAL; 3439 break; 3440 case _PC_PRIO_IO: 3441 *ap->a_retval = 0; 3442 break; 3443 case _PC_SYNC_IO: 3444 *ap->a_retval = 0; 3445 break; 3446 case _PC_ALLOC_SIZE_MIN: 3447 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize; 3448 break; 3449 case _PC_FILESIZEBITS: 3450 if (NFS_ISV34(vp)) 3451 *ap->a_retval = 64; 3452 else 3453 *ap->a_retval = 32; 3454 break; 3455 case _PC_REC_INCR_XFER_SIZE: 3456 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 3457 break; 3458 case _PC_REC_MAX_XFER_SIZE: 3459 *ap->a_retval = -1; /* means ``unlimited'' */ 3460 break; 3461 case _PC_REC_MIN_XFER_SIZE: 3462 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 3463 break; 3464 case _PC_REC_XFER_ALIGN: 3465 *ap->a_retval = PAGE_SIZE; 3466 break; 3467 case _PC_SYMLINK_MAX: 3468 *ap->a_retval = NFS_MAXPATHLEN; 3469 break; 3470 3471 default: 3472 error = EINVAL; 3473 break; 3474 } 3475 return (error); 3476 } 3477 3478