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