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