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 vop_lock1_t nfs_lock1; 142 static int nfs_lookitup(struct vnode *, char *, int, 143 struct ucred *, struct thread *, struct nfsnode **); 144 static int nfs_sillyrename(struct vnode *, struct vnode *, 145 struct componentname *); 146 static vop_access_t nfsspec_access; 147 static vop_readlink_t nfs_readlink; 148 static vop_print_t nfs_print; 149 static vop_advlock_t nfs_advlock; 150 static vop_advlockasync_t nfs_advlockasync; 151 static vop_getacl_t nfs_getacl; 152 static vop_setacl_t nfs_setacl; 153 154 /* 155 * Global vfs data structures for nfs 156 */ 157 struct vop_vector newnfs_vnodeops = { 158 .vop_default = &default_vnodeops, 159 .vop_access = nfs_access, 160 .vop_advlock = nfs_advlock, 161 .vop_advlockasync = nfs_advlockasync, 162 .vop_close = nfs_close, 163 .vop_create = nfs_create, 164 .vop_fsync = nfs_fsync, 165 .vop_getattr = nfs_getattr, 166 .vop_getpages = ncl_getpages, 167 .vop_putpages = ncl_putpages, 168 .vop_inactive = ncl_inactive, 169 .vop_link = nfs_link, 170 .vop_lock1 = nfs_lock1, 171 .vop_lookup = nfs_lookup, 172 .vop_mkdir = nfs_mkdir, 173 .vop_mknod = nfs_mknod, 174 .vop_open = nfs_open, 175 .vop_pathconf = nfs_pathconf, 176 .vop_print = nfs_print, 177 .vop_read = nfs_read, 178 .vop_readdir = nfs_readdir, 179 .vop_readlink = nfs_readlink, 180 .vop_reclaim = ncl_reclaim, 181 .vop_remove = nfs_remove, 182 .vop_rename = nfs_rename, 183 .vop_rmdir = nfs_rmdir, 184 .vop_setattr = nfs_setattr, 185 .vop_strategy = nfs_strategy, 186 .vop_symlink = nfs_symlink, 187 .vop_write = ncl_write, 188 .vop_getacl = nfs_getacl, 189 .vop_setacl = nfs_setacl, 190 }; 191 192 struct vop_vector newnfs_fifoops = { 193 .vop_default = &fifo_specops, 194 .vop_access = nfsspec_access, 195 .vop_close = nfsfifo_close, 196 .vop_fsync = nfs_fsync, 197 .vop_getattr = nfs_getattr, 198 .vop_inactive = ncl_inactive, 199 .vop_print = nfs_print, 200 .vop_read = nfsfifo_read, 201 .vop_reclaim = ncl_reclaim, 202 .vop_setattr = nfs_setattr, 203 .vop_write = nfsfifo_write, 204 }; 205 206 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, 207 struct componentname *cnp, struct vattr *vap); 208 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 209 int namelen, struct ucred *cred, struct thread *td); 210 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, 211 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp, 212 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td); 213 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp, 214 struct componentname *scnp, struct sillyrename *sp); 215 216 /* 217 * Global variables 218 */ 219 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1)) 220 221 SYSCTL_DECL(_vfs_nfs); 222 223 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO; 224 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 225 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 226 227 static int nfs_prime_access_cache = 0; 228 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW, 229 &nfs_prime_access_cache, 0, 230 "Prime NFS ACCESS cache when fetching attributes"); 231 232 static int newnfs_commit_on_close = 0; 233 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW, 234 &newnfs_commit_on_close, 0, "write+commit on close, else only write"); 235 236 static int nfs_clean_pages_on_close = 1; 237 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW, 238 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close"); 239 240 int newnfs_directio_enable = 0; 241 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW, 242 &newnfs_directio_enable, 0, "Enable NFS directio"); 243 244 int nfs_keep_dirty_on_error; 245 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW, 246 &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned"); 247 248 /* 249 * This sysctl allows other processes to mmap a file that has been opened 250 * O_DIRECT by a process. In general, having processes mmap the file while 251 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow 252 * this by default to prevent DoS attacks - to prevent a malicious user from 253 * opening up files O_DIRECT preventing other users from mmap'ing these 254 * files. "Protected" environments where stricter consistency guarantees are 255 * required can disable this knob. The process that opened the file O_DIRECT 256 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not 257 * meaningful. 258 */ 259 int newnfs_directio_allow_mmap = 1; 260 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW, 261 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens"); 262 263 #if 0 264 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD, 265 &newnfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count"); 266 267 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD, 268 &newnfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count"); 269 #endif 270 271 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \ 272 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \ 273 | NFSACCESS_DELETE | NFSACCESS_LOOKUP) 274 275 /* 276 * SMP Locking Note : 277 * The list of locks after the description of the lock is the ordering 278 * of other locks acquired with the lock held. 279 * np->n_mtx : Protects the fields in the nfsnode. 280 VM Object Lock 281 VI_MTX (acquired indirectly) 282 * nmp->nm_mtx : Protects the fields in the nfsmount. 283 rep->r_mtx 284 * ncl_iod_mutex : Global lock, protects shared nfsiod state. 285 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list. 286 nmp->nm_mtx 287 rep->r_mtx 288 * rep->r_mtx : Protects the fields in an nfsreq. 289 */ 290 291 static int 292 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td, 293 struct ucred *cred, u_int32_t *retmode) 294 { 295 int error = 0, attrflag, i, lrupos; 296 u_int32_t rmode; 297 struct nfsnode *np = VTONFS(vp); 298 struct nfsvattr nfsva; 299 300 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag, 301 &rmode, NULL); 302 if (attrflag) 303 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 304 if (!error) { 305 lrupos = 0; 306 mtx_lock(&np->n_mtx); 307 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 308 if (np->n_accesscache[i].uid == cred->cr_uid) { 309 np->n_accesscache[i].mode = rmode; 310 np->n_accesscache[i].stamp = time_second; 311 break; 312 } 313 if (i > 0 && np->n_accesscache[i].stamp < 314 np->n_accesscache[lrupos].stamp) 315 lrupos = i; 316 } 317 if (i == NFS_ACCESSCACHESIZE) { 318 np->n_accesscache[lrupos].uid = cred->cr_uid; 319 np->n_accesscache[lrupos].mode = rmode; 320 np->n_accesscache[lrupos].stamp = time_second; 321 } 322 mtx_unlock(&np->n_mtx); 323 if (retmode != NULL) 324 *retmode = rmode; 325 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0); 326 } else if (NFS_ISV4(vp)) { 327 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 328 } 329 #ifdef KDTRACE_HOOKS 330 if (error != 0) 331 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0, 332 error); 333 #endif 334 return (error); 335 } 336 337 /* 338 * nfs access vnode op. 339 * For nfs version 2, just return ok. File accesses may fail later. 340 * For nfs version 3, use the access rpc to check accessibility. If file modes 341 * are changed on the server, accesses might still fail later. 342 */ 343 static int 344 nfs_access(struct vop_access_args *ap) 345 { 346 struct vnode *vp = ap->a_vp; 347 int error = 0, i, gotahit; 348 u_int32_t mode, wmode, rmode; 349 int v34 = NFS_ISV34(vp); 350 struct nfsnode *np = VTONFS(vp); 351 352 /* 353 * Disallow write attempts on filesystems mounted read-only; 354 * unless the file is a socket, fifo, or a block or character 355 * device resident on the filesystem. 356 */ 357 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS | 358 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL | 359 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) { 360 switch (vp->v_type) { 361 case VREG: 362 case VDIR: 363 case VLNK: 364 return (EROFS); 365 default: 366 break; 367 } 368 } 369 /* 370 * For nfs v3 or v4, check to see if we have done this recently, and if 371 * so return our cached result instead of making an ACCESS call. 372 * If not, do an access rpc, otherwise you are stuck emulating 373 * ufs_access() locally using the vattr. This may not be correct, 374 * since the server may apply other access criteria such as 375 * client uid-->server uid mapping that we do not know about. 376 */ 377 if (v34) { 378 if (ap->a_accmode & VREAD) 379 mode = NFSACCESS_READ; 380 else 381 mode = 0; 382 if (vp->v_type != VDIR) { 383 if (ap->a_accmode & VWRITE) 384 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 385 if (ap->a_accmode & VAPPEND) 386 mode |= NFSACCESS_EXTEND; 387 if (ap->a_accmode & VEXEC) 388 mode |= NFSACCESS_EXECUTE; 389 if (ap->a_accmode & VDELETE) 390 mode |= NFSACCESS_DELETE; 391 } else { 392 if (ap->a_accmode & VWRITE) 393 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); 394 if (ap->a_accmode & VAPPEND) 395 mode |= NFSACCESS_EXTEND; 396 if (ap->a_accmode & VEXEC) 397 mode |= NFSACCESS_LOOKUP; 398 if (ap->a_accmode & VDELETE) 399 mode |= NFSACCESS_DELETE; 400 if (ap->a_accmode & VDELETE_CHILD) 401 mode |= NFSACCESS_MODIFY; 402 } 403 /* XXX safety belt, only make blanket request if caching */ 404 if (nfsaccess_cache_timeout > 0) { 405 wmode = NFSACCESS_READ | NFSACCESS_MODIFY | 406 NFSACCESS_EXTEND | NFSACCESS_EXECUTE | 407 NFSACCESS_DELETE | NFSACCESS_LOOKUP; 408 } else { 409 wmode = mode; 410 } 411 412 /* 413 * Does our cached result allow us to give a definite yes to 414 * this request? 415 */ 416 gotahit = 0; 417 mtx_lock(&np->n_mtx); 418 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { 419 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) { 420 if (time_second < (np->n_accesscache[i].stamp 421 + nfsaccess_cache_timeout) && 422 (np->n_accesscache[i].mode & mode) == mode) { 423 NFSINCRGLOBAL(newnfsstats.accesscache_hits); 424 gotahit = 1; 425 } 426 break; 427 } 428 } 429 mtx_unlock(&np->n_mtx); 430 #ifdef KDTRACE_HOOKS 431 if (gotahit != 0) 432 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp, 433 ap->a_cred->cr_uid, mode); 434 else 435 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp, 436 ap->a_cred->cr_uid, mode); 437 #endif 438 if (gotahit == 0) { 439 /* 440 * Either a no, or a don't know. Go to the wire. 441 */ 442 NFSINCRGLOBAL(newnfsstats.accesscache_misses); 443 error = nfs34_access_otw(vp, wmode, ap->a_td, 444 ap->a_cred, &rmode); 445 if (!error && 446 (rmode & mode) != mode) 447 error = EACCES; 448 } 449 return (error); 450 } else { 451 if ((error = nfsspec_access(ap)) != 0) { 452 return (error); 453 } 454 /* 455 * Attempt to prevent a mapped root from accessing a file 456 * which it shouldn't. We try to read a byte from the file 457 * if the user is root and the file is not zero length. 458 * After calling nfsspec_access, we should have the correct 459 * file size cached. 460 */ 461 mtx_lock(&np->n_mtx); 462 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD) 463 && VTONFS(vp)->n_size > 0) { 464 struct iovec aiov; 465 struct uio auio; 466 char buf[1]; 467 468 mtx_unlock(&np->n_mtx); 469 aiov.iov_base = buf; 470 aiov.iov_len = 1; 471 auio.uio_iov = &aiov; 472 auio.uio_iovcnt = 1; 473 auio.uio_offset = 0; 474 auio.uio_resid = 1; 475 auio.uio_segflg = UIO_SYSSPACE; 476 auio.uio_rw = UIO_READ; 477 auio.uio_td = ap->a_td; 478 479 if (vp->v_type == VREG) 480 error = ncl_readrpc(vp, &auio, ap->a_cred); 481 else if (vp->v_type == VDIR) { 482 char* bp; 483 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 484 aiov.iov_base = bp; 485 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 486 error = ncl_readdirrpc(vp, &auio, ap->a_cred, 487 ap->a_td); 488 free(bp, M_TEMP); 489 } else if (vp->v_type == VLNK) 490 error = ncl_readlinkrpc(vp, &auio, ap->a_cred); 491 else 492 error = EACCES; 493 } else 494 mtx_unlock(&np->n_mtx); 495 return (error); 496 } 497 } 498 499 500 /* 501 * nfs open vnode op 502 * Check to see if the type is ok 503 * and that deletion is not in progress. 504 * For paged in text files, you will need to flush the page cache 505 * if consistency is lost. 506 */ 507 /* ARGSUSED */ 508 static int 509 nfs_open(struct vop_open_args *ap) 510 { 511 struct vnode *vp = ap->a_vp; 512 struct nfsnode *np = VTONFS(vp); 513 struct vattr vattr; 514 int error; 515 int fmode = ap->a_mode; 516 struct ucred *cred; 517 518 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) 519 return (EOPNOTSUPP); 520 521 /* 522 * For NFSv4, we need to do the Open Op before cache validation, 523 * so that we conform to RFC3530 Sec. 9.3.1. 524 */ 525 if (NFS_ISV4(vp)) { 526 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td); 527 if (error) { 528 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 529 (gid_t)0); 530 return (error); 531 } 532 } 533 534 /* 535 * Now, if this Open will be doing reading, re-validate/flush the 536 * cache, so that Close/Open coherency is maintained. 537 */ 538 mtx_lock(&np->n_mtx); 539 if (np->n_flag & NMODIFIED) { 540 mtx_unlock(&np->n_mtx); 541 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 542 if (error == EINTR || error == EIO) { 543 if (NFS_ISV4(vp)) 544 (void) nfsrpc_close(vp, 0, ap->a_td); 545 return (error); 546 } 547 mtx_lock(&np->n_mtx); 548 np->n_attrstamp = 0; 549 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 550 if (vp->v_type == VDIR) 551 np->n_direofoffset = 0; 552 mtx_unlock(&np->n_mtx); 553 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 554 if (error) { 555 if (NFS_ISV4(vp)) 556 (void) nfsrpc_close(vp, 0, ap->a_td); 557 return (error); 558 } 559 mtx_lock(&np->n_mtx); 560 np->n_mtime = vattr.va_mtime; 561 if (NFS_ISV4(vp)) 562 np->n_change = vattr.va_filerev; 563 } else { 564 mtx_unlock(&np->n_mtx); 565 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 566 if (error) { 567 if (NFS_ISV4(vp)) 568 (void) nfsrpc_close(vp, 0, ap->a_td); 569 return (error); 570 } 571 mtx_lock(&np->n_mtx); 572 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) || 573 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 574 if (vp->v_type == VDIR) 575 np->n_direofoffset = 0; 576 mtx_unlock(&np->n_mtx); 577 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 578 if (error == EINTR || error == EIO) { 579 if (NFS_ISV4(vp)) 580 (void) nfsrpc_close(vp, 0, ap->a_td); 581 return (error); 582 } 583 mtx_lock(&np->n_mtx); 584 np->n_mtime = vattr.va_mtime; 585 if (NFS_ISV4(vp)) 586 np->n_change = vattr.va_filerev; 587 } 588 } 589 590 /* 591 * If the object has >= 1 O_DIRECT active opens, we disable caching. 592 */ 593 if (newnfs_directio_enable && (fmode & O_DIRECT) && 594 (vp->v_type == VREG)) { 595 if (np->n_directio_opens == 0) { 596 mtx_unlock(&np->n_mtx); 597 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 598 if (error) { 599 if (NFS_ISV4(vp)) 600 (void) nfsrpc_close(vp, 0, ap->a_td); 601 return (error); 602 } 603 mtx_lock(&np->n_mtx); 604 np->n_flag |= NNONCACHE; 605 } 606 np->n_directio_opens++; 607 } 608 609 /* If opened for writing via NFSv4.1 or later, mark that for pNFS. */ 610 if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0) 611 np->n_flag |= NWRITEOPENED; 612 613 /* 614 * If this is an open for writing, capture a reference to the 615 * credentials, so they can be used by ncl_putpages(). Using 616 * these write credentials is preferable to the credentials of 617 * whatever thread happens to be doing the VOP_PUTPAGES() since 618 * the write RPCs are less likely to fail with EACCES. 619 */ 620 if ((fmode & FWRITE) != 0) { 621 cred = np->n_writecred; 622 np->n_writecred = crhold(ap->a_cred); 623 } else 624 cred = NULL; 625 mtx_unlock(&np->n_mtx); 626 627 if (cred != NULL) 628 crfree(cred); 629 vnode_create_vobject(vp, vattr.va_size, ap->a_td); 630 return (0); 631 } 632 633 /* 634 * nfs close vnode op 635 * What an NFS client should do upon close after writing is a debatable issue. 636 * Most NFS clients push delayed writes to the server upon close, basically for 637 * two reasons: 638 * 1 - So that any write errors may be reported back to the client process 639 * doing the close system call. By far the two most likely errors are 640 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 641 * 2 - To put a worst case upper bound on cache inconsistency between 642 * multiple clients for the file. 643 * There is also a consistency problem for Version 2 of the protocol w.r.t. 644 * not being able to tell if other clients are writing a file concurrently, 645 * since there is no way of knowing if the changed modify time in the reply 646 * is only due to the write for this client. 647 * (NFS Version 3 provides weak cache consistency data in the reply that 648 * should be sufficient to detect and handle this case.) 649 * 650 * The current code does the following: 651 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 652 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 653 * or commit them (this satisfies 1 and 2 except for the 654 * case where the server crashes after this close but 655 * before the commit RPC, which is felt to be "good 656 * enough". Changing the last argument to ncl_flush() to 657 * a 1 would force a commit operation, if it is felt a 658 * commit is necessary now. 659 * for NFS Version 4 - flush the dirty buffers and commit them, if 660 * nfscl_mustflush() says this is necessary. 661 * It is necessary if there is no write delegation held, 662 * in order to satisfy open/close coherency. 663 * If the file isn't cached on local stable storage, 664 * it may be necessary in order to detect "out of space" 665 * errors from the server, if the write delegation 666 * issued by the server doesn't allow the file to grow. 667 */ 668 /* ARGSUSED */ 669 static int 670 nfs_close(struct vop_close_args *ap) 671 { 672 struct vnode *vp = ap->a_vp; 673 struct nfsnode *np = VTONFS(vp); 674 struct nfsvattr nfsva; 675 struct ucred *cred; 676 int error = 0, ret, localcred = 0; 677 int fmode = ap->a_fflag; 678 679 if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF)) 680 return (0); 681 /* 682 * During shutdown, a_cred isn't valid, so just use root. 683 */ 684 if (ap->a_cred == NOCRED) { 685 cred = newnfs_getcred(); 686 localcred = 1; 687 } else { 688 cred = ap->a_cred; 689 } 690 if (vp->v_type == VREG) { 691 /* 692 * Examine and clean dirty pages, regardless of NMODIFIED. 693 * This closes a major hole in close-to-open consistency. 694 * We want to push out all dirty pages (and buffers) on 695 * close, regardless of whether they were dirtied by 696 * mmap'ed writes or via write(). 697 */ 698 if (nfs_clean_pages_on_close && vp->v_object) { 699 VM_OBJECT_WLOCK(vp->v_object); 700 vm_object_page_clean(vp->v_object, 0, 0, 0); 701 VM_OBJECT_WUNLOCK(vp->v_object); 702 } 703 mtx_lock(&np->n_mtx); 704 if (np->n_flag & NMODIFIED) { 705 mtx_unlock(&np->n_mtx); 706 if (NFS_ISV3(vp)) { 707 /* 708 * Under NFSv3 we have dirty buffers to dispose of. We 709 * must flush them to the NFS server. We have the option 710 * of waiting all the way through the commit rpc or just 711 * waiting for the initial write. The default is to only 712 * wait through the initial write so the data is in the 713 * server's cache, which is roughly similar to the state 714 * a standard disk subsystem leaves the file in on close(). 715 * 716 * We cannot clear the NMODIFIED bit in np->n_flag due to 717 * potential races with other processes, and certainly 718 * cannot clear it if we don't commit. 719 * These races occur when there is no longer the old 720 * traditional vnode locking implemented for Vnode Ops. 721 */ 722 int cm = newnfs_commit_on_close ? 1 : 0; 723 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm, 0); 724 /* np->n_flag &= ~NMODIFIED; */ 725 } else if (NFS_ISV4(vp)) { 726 if (nfscl_mustflush(vp) != 0) { 727 int cm = newnfs_commit_on_close ? 1 : 0; 728 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, 729 cm, 0); 730 /* 731 * as above w.r.t races when clearing 732 * NMODIFIED. 733 * np->n_flag &= ~NMODIFIED; 734 */ 735 } 736 } else 737 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 738 mtx_lock(&np->n_mtx); 739 } 740 /* 741 * Invalidate the attribute cache in all cases. 742 * An open is going to fetch fresh attrs any way, other procs 743 * on this node that have file open will be forced to do an 744 * otw attr fetch, but this is safe. 745 * --> A user found that their RPC count dropped by 20% when 746 * this was commented out and I can't see any requirement 747 * for it, so I've disabled it when negative lookups are 748 * enabled. (What does this have to do with negative lookup 749 * caching? Well nothing, except it was reported by the 750 * same user that needed negative lookup caching and I wanted 751 * there to be a way to disable it to see if it 752 * is the cause of some caching/coherency issue that might 753 * crop up.) 754 */ 755 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) { 756 np->n_attrstamp = 0; 757 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 758 } 759 if (np->n_flag & NWRITEERR) { 760 np->n_flag &= ~NWRITEERR; 761 error = np->n_error; 762 } 763 mtx_unlock(&np->n_mtx); 764 } 765 766 if (NFS_ISV4(vp)) { 767 /* 768 * Get attributes so "change" is up to date. 769 */ 770 if (error == 0 && nfscl_mustflush(vp) != 0 && 771 vp->v_type == VREG && 772 (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) { 773 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva, 774 NULL); 775 if (!ret) { 776 np->n_change = nfsva.na_filerev; 777 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 778 NULL, 0, 0); 779 } 780 } 781 782 /* 783 * and do the close. 784 */ 785 ret = nfsrpc_close(vp, 0, ap->a_td); 786 if (!error && ret) 787 error = ret; 788 if (error) 789 error = nfscl_maperr(ap->a_td, error, (uid_t)0, 790 (gid_t)0); 791 } 792 if (newnfs_directio_enable) 793 KASSERT((np->n_directio_asyncwr == 0), 794 ("nfs_close: dirty unflushed (%d) directio buffers\n", 795 np->n_directio_asyncwr)); 796 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) { 797 mtx_lock(&np->n_mtx); 798 KASSERT((np->n_directio_opens > 0), 799 ("nfs_close: unexpectedly value (0) of n_directio_opens\n")); 800 np->n_directio_opens--; 801 if (np->n_directio_opens == 0) 802 np->n_flag &= ~NNONCACHE; 803 mtx_unlock(&np->n_mtx); 804 } 805 if (localcred) 806 NFSFREECRED(cred); 807 return (error); 808 } 809 810 /* 811 * nfs getattr call from vfs. 812 */ 813 static int 814 nfs_getattr(struct vop_getattr_args *ap) 815 { 816 struct vnode *vp = ap->a_vp; 817 struct thread *td = curthread; /* XXX */ 818 struct nfsnode *np = VTONFS(vp); 819 int error = 0; 820 struct nfsvattr nfsva; 821 struct vattr *vap = ap->a_vap; 822 struct vattr vattr; 823 824 /* 825 * Update local times for special files. 826 */ 827 mtx_lock(&np->n_mtx); 828 if (np->n_flag & (NACC | NUPD)) 829 np->n_flag |= NCHG; 830 mtx_unlock(&np->n_mtx); 831 /* 832 * First look in the cache. 833 */ 834 if (ncl_getattrcache(vp, &vattr) == 0) { 835 vap->va_type = vattr.va_type; 836 vap->va_mode = vattr.va_mode; 837 vap->va_nlink = vattr.va_nlink; 838 vap->va_uid = vattr.va_uid; 839 vap->va_gid = vattr.va_gid; 840 vap->va_fsid = vattr.va_fsid; 841 vap->va_fileid = vattr.va_fileid; 842 vap->va_size = vattr.va_size; 843 vap->va_blocksize = vattr.va_blocksize; 844 vap->va_atime = vattr.va_atime; 845 vap->va_mtime = vattr.va_mtime; 846 vap->va_ctime = vattr.va_ctime; 847 vap->va_gen = vattr.va_gen; 848 vap->va_flags = vattr.va_flags; 849 vap->va_rdev = vattr.va_rdev; 850 vap->va_bytes = vattr.va_bytes; 851 vap->va_filerev = vattr.va_filerev; 852 /* 853 * Get the local modify time for the case of a write 854 * delegation. 855 */ 856 nfscl_deleggetmodtime(vp, &vap->va_mtime); 857 return (0); 858 } 859 860 if (NFS_ISV34(vp) && nfs_prime_access_cache && 861 nfsaccess_cache_timeout > 0) { 862 NFSINCRGLOBAL(newnfsstats.accesscache_misses); 863 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL); 864 if (ncl_getattrcache(vp, ap->a_vap) == 0) { 865 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime); 866 return (0); 867 } 868 } 869 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL); 870 if (!error) 871 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0); 872 if (!error) { 873 /* 874 * Get the local modify time for the case of a write 875 * delegation. 876 */ 877 nfscl_deleggetmodtime(vp, &vap->va_mtime); 878 } else if (NFS_ISV4(vp)) { 879 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 880 } 881 return (error); 882 } 883 884 /* 885 * nfs setattr call. 886 */ 887 static int 888 nfs_setattr(struct vop_setattr_args *ap) 889 { 890 struct vnode *vp = ap->a_vp; 891 struct nfsnode *np = VTONFS(vp); 892 struct thread *td = curthread; /* XXX */ 893 struct vattr *vap = ap->a_vap; 894 int error = 0; 895 u_quad_t tsize; 896 897 #ifndef nolint 898 tsize = (u_quad_t)0; 899 #endif 900 901 /* 902 * Setting of flags and marking of atimes are not supported. 903 */ 904 if (vap->va_flags != VNOVAL) 905 return (EOPNOTSUPP); 906 907 /* 908 * Disallow write attempts if the filesystem is mounted read-only. 909 */ 910 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 911 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 912 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 913 (vp->v_mount->mnt_flag & MNT_RDONLY)) 914 return (EROFS); 915 if (vap->va_size != VNOVAL) { 916 switch (vp->v_type) { 917 case VDIR: 918 return (EISDIR); 919 case VCHR: 920 case VBLK: 921 case VSOCK: 922 case VFIFO: 923 if (vap->va_mtime.tv_sec == VNOVAL && 924 vap->va_atime.tv_sec == VNOVAL && 925 vap->va_mode == (mode_t)VNOVAL && 926 vap->va_uid == (uid_t)VNOVAL && 927 vap->va_gid == (gid_t)VNOVAL) 928 return (0); 929 vap->va_size = VNOVAL; 930 break; 931 default: 932 /* 933 * Disallow write attempts if the filesystem is 934 * mounted read-only. 935 */ 936 if (vp->v_mount->mnt_flag & MNT_RDONLY) 937 return (EROFS); 938 /* 939 * We run vnode_pager_setsize() early (why?), 940 * we must set np->n_size now to avoid vinvalbuf 941 * V_SAVE races that might setsize a lower 942 * value. 943 */ 944 mtx_lock(&np->n_mtx); 945 tsize = np->n_size; 946 mtx_unlock(&np->n_mtx); 947 error = ncl_meta_setsize(vp, ap->a_cred, td, 948 vap->va_size); 949 mtx_lock(&np->n_mtx); 950 if (np->n_flag & NMODIFIED) { 951 tsize = np->n_size; 952 mtx_unlock(&np->n_mtx); 953 if (vap->va_size == 0) 954 error = ncl_vinvalbuf(vp, 0, td, 1); 955 else 956 error = ncl_vinvalbuf(vp, V_SAVE, td, 1); 957 if (error) { 958 vnode_pager_setsize(vp, tsize); 959 return (error); 960 } 961 /* 962 * Call nfscl_delegmodtime() to set the modify time 963 * locally, as required. 964 */ 965 nfscl_delegmodtime(vp); 966 } else 967 mtx_unlock(&np->n_mtx); 968 /* 969 * np->n_size has already been set to vap->va_size 970 * in ncl_meta_setsize(). We must set it again since 971 * nfs_loadattrcache() could be called through 972 * ncl_meta_setsize() and could modify np->n_size. 973 */ 974 mtx_lock(&np->n_mtx); 975 np->n_vattr.na_size = np->n_size = vap->va_size; 976 mtx_unlock(&np->n_mtx); 977 }; 978 } else { 979 mtx_lock(&np->n_mtx); 980 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) && 981 (np->n_flag & NMODIFIED) && vp->v_type == VREG) { 982 mtx_unlock(&np->n_mtx); 983 if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 && 984 (error == EINTR || error == EIO)) 985 return (error); 986 } else 987 mtx_unlock(&np->n_mtx); 988 } 989 error = nfs_setattrrpc(vp, vap, ap->a_cred, td); 990 if (error && vap->va_size != VNOVAL) { 991 mtx_lock(&np->n_mtx); 992 np->n_size = np->n_vattr.na_size = tsize; 993 vnode_pager_setsize(vp, tsize); 994 mtx_unlock(&np->n_mtx); 995 } 996 return (error); 997 } 998 999 /* 1000 * Do an nfs setattr rpc. 1001 */ 1002 static int 1003 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred, 1004 struct thread *td) 1005 { 1006 struct nfsnode *np = VTONFS(vp); 1007 int error, ret, attrflag, i; 1008 struct nfsvattr nfsva; 1009 1010 if (NFS_ISV34(vp)) { 1011 mtx_lock(&np->n_mtx); 1012 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) 1013 np->n_accesscache[i].stamp = 0; 1014 np->n_flag |= NDELEGMOD; 1015 mtx_unlock(&np->n_mtx); 1016 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp); 1017 } 1018 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag, 1019 NULL); 1020 if (attrflag) { 1021 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1022 if (ret && !error) 1023 error = ret; 1024 } 1025 if (error && NFS_ISV4(vp)) 1026 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid); 1027 return (error); 1028 } 1029 1030 /* 1031 * nfs lookup call, one step at a time... 1032 * First look in cache 1033 * If not found, unlock the directory nfsnode and do the rpc 1034 */ 1035 static int 1036 nfs_lookup(struct vop_lookup_args *ap) 1037 { 1038 struct componentname *cnp = ap->a_cnp; 1039 struct vnode *dvp = ap->a_dvp; 1040 struct vnode **vpp = ap->a_vpp; 1041 struct mount *mp = dvp->v_mount; 1042 int flags = cnp->cn_flags; 1043 struct vnode *newvp; 1044 struct nfsmount *nmp; 1045 struct nfsnode *np, *newnp; 1046 int error = 0, attrflag, dattrflag, ltype, ncticks; 1047 struct thread *td = cnp->cn_thread; 1048 struct nfsfh *nfhp; 1049 struct nfsvattr dnfsva, nfsva; 1050 struct vattr vattr; 1051 struct timespec nctime; 1052 1053 *vpp = NULLVP; 1054 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) && 1055 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 1056 return (EROFS); 1057 if (dvp->v_type != VDIR) 1058 return (ENOTDIR); 1059 nmp = VFSTONFS(mp); 1060 np = VTONFS(dvp); 1061 1062 /* For NFSv4, wait until any remove is done. */ 1063 mtx_lock(&np->n_mtx); 1064 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) { 1065 np->n_flag |= NREMOVEWANT; 1066 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0); 1067 } 1068 mtx_unlock(&np->n_mtx); 1069 1070 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) 1071 return (error); 1072 error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks); 1073 if (error > 0 && error != ENOENT) 1074 return (error); 1075 if (error == -1) { 1076 /* 1077 * Lookups of "." are special and always return the 1078 * current directory. cache_lookup() already handles 1079 * associated locking bookkeeping, etc. 1080 */ 1081 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 1082 /* XXX: Is this really correct? */ 1083 if (cnp->cn_nameiop != LOOKUP && 1084 (flags & ISLASTCN)) 1085 cnp->cn_flags |= SAVENAME; 1086 return (0); 1087 } 1088 1089 /* 1090 * We only accept a positive hit in the cache if the 1091 * change time of the file matches our cached copy. 1092 * Otherwise, we discard the cache entry and fallback 1093 * to doing a lookup RPC. We also only trust cache 1094 * entries for less than nm_nametimeo seconds. 1095 * 1096 * To better handle stale file handles and attributes, 1097 * clear the attribute cache of this node if it is a 1098 * leaf component, part of an open() call, and not 1099 * locally modified before fetching the attributes. 1100 * This should allow stale file handles to be detected 1101 * here where we can fall back to a LOOKUP RPC to 1102 * recover rather than having nfs_open() detect the 1103 * stale file handle and failing open(2) with ESTALE. 1104 */ 1105 newvp = *vpp; 1106 newnp = VTONFS(newvp); 1107 if (!(nmp->nm_flag & NFSMNT_NOCTO) && 1108 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1109 !(newnp->n_flag & NMODIFIED)) { 1110 mtx_lock(&newnp->n_mtx); 1111 newnp->n_attrstamp = 0; 1112 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1113 mtx_unlock(&newnp->n_mtx); 1114 } 1115 if (nfscl_nodeleg(newvp, 0) == 0 || 1116 ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) && 1117 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 && 1118 timespeccmp(&vattr.va_ctime, &nctime, ==))) { 1119 NFSINCRGLOBAL(newnfsstats.lookupcache_hits); 1120 if (cnp->cn_nameiop != LOOKUP && 1121 (flags & ISLASTCN)) 1122 cnp->cn_flags |= SAVENAME; 1123 return (0); 1124 } 1125 cache_purge(newvp); 1126 if (dvp != newvp) 1127 vput(newvp); 1128 else 1129 vrele(newvp); 1130 *vpp = NULLVP; 1131 } else if (error == ENOENT) { 1132 if (dvp->v_iflag & VI_DOOMED) 1133 return (ENOENT); 1134 /* 1135 * We only accept a negative hit in the cache if the 1136 * modification time of the parent directory matches 1137 * the cached copy in the name cache entry. 1138 * Otherwise, we discard all of the negative cache 1139 * entries for this directory. We also only trust 1140 * negative cache entries for up to nm_negnametimeo 1141 * seconds. 1142 */ 1143 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) && 1144 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 && 1145 timespeccmp(&vattr.va_mtime, &nctime, ==)) { 1146 NFSINCRGLOBAL(newnfsstats.lookupcache_hits); 1147 return (ENOENT); 1148 } 1149 cache_purge_negative(dvp); 1150 } 1151 1152 error = 0; 1153 newvp = NULLVP; 1154 NFSINCRGLOBAL(newnfsstats.lookupcache_misses); 1155 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1156 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1157 NULL); 1158 if (dattrflag) 1159 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1160 if (error) { 1161 if (newvp != NULLVP) { 1162 vput(newvp); 1163 *vpp = NULLVP; 1164 } 1165 1166 if (error != ENOENT) { 1167 if (NFS_ISV4(dvp)) 1168 error = nfscl_maperr(td, error, (uid_t)0, 1169 (gid_t)0); 1170 return (error); 1171 } 1172 1173 /* The requested file was not found. */ 1174 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1175 (flags & ISLASTCN)) { 1176 /* 1177 * XXX: UFS does a full VOP_ACCESS(dvp, 1178 * VWRITE) here instead of just checking 1179 * MNT_RDONLY. 1180 */ 1181 if (mp->mnt_flag & MNT_RDONLY) 1182 return (EROFS); 1183 cnp->cn_flags |= SAVENAME; 1184 return (EJUSTRETURN); 1185 } 1186 1187 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE && 1188 dattrflag) { 1189 /* 1190 * Cache the modification time of the parent 1191 * directory from the post-op attributes in 1192 * the name cache entry. The negative cache 1193 * entry will be ignored once the directory 1194 * has changed. Don't bother adding the entry 1195 * if the directory has already changed. 1196 */ 1197 mtx_lock(&np->n_mtx); 1198 if (timespeccmp(&np->n_vattr.na_mtime, 1199 &dnfsva.na_mtime, ==)) { 1200 mtx_unlock(&np->n_mtx); 1201 cache_enter_time(dvp, NULL, cnp, 1202 &dnfsva.na_mtime, NULL); 1203 } else 1204 mtx_unlock(&np->n_mtx); 1205 } 1206 return (ENOENT); 1207 } 1208 1209 /* 1210 * Handle RENAME case... 1211 */ 1212 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 1213 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1214 FREE((caddr_t)nfhp, M_NFSFH); 1215 return (EISDIR); 1216 } 1217 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1218 LK_EXCLUSIVE); 1219 if (error) 1220 return (error); 1221 newvp = NFSTOV(np); 1222 if (attrflag) 1223 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1224 0, 1); 1225 *vpp = newvp; 1226 cnp->cn_flags |= SAVENAME; 1227 return (0); 1228 } 1229 1230 if (flags & ISDOTDOT) { 1231 ltype = NFSVOPISLOCKED(dvp); 1232 error = vfs_busy(mp, MBF_NOWAIT); 1233 if (error != 0) { 1234 vfs_ref(mp); 1235 NFSVOPUNLOCK(dvp, 0); 1236 error = vfs_busy(mp, 0); 1237 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1238 vfs_rel(mp); 1239 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) { 1240 vfs_unbusy(mp); 1241 error = ENOENT; 1242 } 1243 if (error != 0) 1244 return (error); 1245 } 1246 NFSVOPUNLOCK(dvp, 0); 1247 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1248 cnp->cn_lkflags); 1249 if (error == 0) 1250 newvp = NFSTOV(np); 1251 vfs_unbusy(mp); 1252 if (newvp != dvp) 1253 NFSVOPLOCK(dvp, ltype | LK_RETRY); 1254 if (dvp->v_iflag & VI_DOOMED) { 1255 if (error == 0) { 1256 if (newvp == dvp) 1257 vrele(newvp); 1258 else 1259 vput(newvp); 1260 } 1261 error = ENOENT; 1262 } 1263 if (error != 0) 1264 return (error); 1265 if (attrflag) 1266 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1267 0, 1); 1268 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { 1269 FREE((caddr_t)nfhp, M_NFSFH); 1270 VREF(dvp); 1271 newvp = dvp; 1272 if (attrflag) 1273 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1274 0, 1); 1275 } else { 1276 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, 1277 cnp->cn_lkflags); 1278 if (error) 1279 return (error); 1280 newvp = NFSTOV(np); 1281 if (attrflag) 1282 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1283 0, 1); 1284 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && 1285 !(np->n_flag & NMODIFIED)) { 1286 /* 1287 * Flush the attribute cache when opening a 1288 * leaf node to ensure that fresh attributes 1289 * are fetched in nfs_open() since we did not 1290 * fetch attributes from the LOOKUP reply. 1291 */ 1292 mtx_lock(&np->n_mtx); 1293 np->n_attrstamp = 0; 1294 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); 1295 mtx_unlock(&np->n_mtx); 1296 } 1297 } 1298 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1299 cnp->cn_flags |= SAVENAME; 1300 if ((cnp->cn_flags & MAKEENTRY) && 1301 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) && 1302 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0)) 1303 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 1304 newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime); 1305 *vpp = newvp; 1306 return (0); 1307 } 1308 1309 /* 1310 * nfs read call. 1311 * Just call ncl_bioread() to do the work. 1312 */ 1313 static int 1314 nfs_read(struct vop_read_args *ap) 1315 { 1316 struct vnode *vp = ap->a_vp; 1317 1318 switch (vp->v_type) { 1319 case VREG: 1320 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred)); 1321 case VDIR: 1322 return (EISDIR); 1323 default: 1324 return (EOPNOTSUPP); 1325 } 1326 } 1327 1328 /* 1329 * nfs readlink call 1330 */ 1331 static int 1332 nfs_readlink(struct vop_readlink_args *ap) 1333 { 1334 struct vnode *vp = ap->a_vp; 1335 1336 if (vp->v_type != VLNK) 1337 return (EINVAL); 1338 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred)); 1339 } 1340 1341 /* 1342 * Do a readlink rpc. 1343 * Called by ncl_doio() from below the buffer cache. 1344 */ 1345 int 1346 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1347 { 1348 int error, ret, attrflag; 1349 struct nfsvattr nfsva; 1350 1351 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva, 1352 &attrflag, NULL); 1353 if (attrflag) { 1354 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1355 if (ret && !error) 1356 error = ret; 1357 } 1358 if (error && NFS_ISV4(vp)) 1359 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1360 return (error); 1361 } 1362 1363 /* 1364 * nfs read rpc call 1365 * Ditto above 1366 */ 1367 int 1368 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) 1369 { 1370 int error, ret, attrflag; 1371 struct nfsvattr nfsva; 1372 struct nfsmount *nmp; 1373 1374 nmp = VFSTONFS(vnode_mount(vp)); 1375 error = EIO; 1376 attrflag = 0; 1377 if (NFSHASPNFS(nmp)) 1378 error = nfscl_doiods(vp, uiop, NULL, NULL, 1379 NFSV4OPEN_ACCESSREAD, cred, uiop->uio_td); 1380 NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error); 1381 if (error != 0) 1382 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, 1383 &attrflag, NULL); 1384 if (attrflag) { 1385 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 1386 if (ret && !error) 1387 error = ret; 1388 } 1389 if (error && NFS_ISV4(vp)) 1390 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1391 return (error); 1392 } 1393 1394 /* 1395 * nfs write call 1396 */ 1397 int 1398 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 1399 int *iomode, int *must_commit, int called_from_strategy) 1400 { 1401 struct nfsvattr nfsva; 1402 int error, attrflag, ret; 1403 struct nfsmount *nmp; 1404 1405 nmp = VFSTONFS(vnode_mount(vp)); 1406 error = EIO; 1407 attrflag = 0; 1408 if (NFSHASPNFS(nmp)) 1409 error = nfscl_doiods(vp, uiop, iomode, must_commit, 1410 NFSV4OPEN_ACCESSWRITE, cred, uiop->uio_td); 1411 NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error); 1412 if (error != 0) 1413 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred, 1414 uiop->uio_td, &nfsva, &attrflag, NULL, 1415 called_from_strategy); 1416 if (attrflag) { 1417 if (VTONFS(vp)->n_flag & ND_NFSV4) 1418 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1, 1419 1); 1420 else 1421 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1422 1); 1423 if (ret && !error) 1424 error = ret; 1425 } 1426 if (DOINGASYNC(vp)) 1427 *iomode = NFSWRITE_FILESYNC; 1428 if (error && NFS_ISV4(vp)) 1429 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); 1430 return (error); 1431 } 1432 1433 /* 1434 * nfs mknod rpc 1435 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1436 * mode set to specify the file type and the size field for rdev. 1437 */ 1438 static int 1439 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1440 struct vattr *vap) 1441 { 1442 struct nfsvattr nfsva, dnfsva; 1443 struct vnode *newvp = NULL; 1444 struct nfsnode *np = NULL, *dnp; 1445 struct nfsfh *nfhp; 1446 struct vattr vattr; 1447 int error = 0, attrflag, dattrflag; 1448 u_int32_t rdev; 1449 1450 if (vap->va_type == VCHR || vap->va_type == VBLK) 1451 rdev = vap->va_rdev; 1452 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1453 rdev = 0xffffffff; 1454 else 1455 return (EOPNOTSUPP); 1456 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1457 return (error); 1458 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap, 1459 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva, 1460 &nfsva, &nfhp, &attrflag, &dattrflag, NULL); 1461 if (!error) { 1462 if (!nfhp) 1463 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1464 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, 1465 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1466 NULL); 1467 if (nfhp) 1468 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1469 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); 1470 } 1471 if (dattrflag) 1472 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1473 if (!error) { 1474 newvp = NFSTOV(np); 1475 if (attrflag != 0) { 1476 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1477 0, 1); 1478 if (error != 0) 1479 vput(newvp); 1480 } 1481 } 1482 if (!error) { 1483 *vpp = newvp; 1484 } else if (NFS_ISV4(dvp)) { 1485 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, 1486 vap->va_gid); 1487 } 1488 dnp = VTONFS(dvp); 1489 mtx_lock(&dnp->n_mtx); 1490 dnp->n_flag |= NMODIFIED; 1491 if (!dattrflag) { 1492 dnp->n_attrstamp = 0; 1493 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1494 } 1495 mtx_unlock(&dnp->n_mtx); 1496 return (error); 1497 } 1498 1499 /* 1500 * nfs mknod vop 1501 * just call nfs_mknodrpc() to do the work. 1502 */ 1503 /* ARGSUSED */ 1504 static int 1505 nfs_mknod(struct vop_mknod_args *ap) 1506 { 1507 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap)); 1508 } 1509 1510 static struct mtx nfs_cverf_mtx; 1511 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex", 1512 MTX_DEF); 1513 1514 static nfsquad_t 1515 nfs_get_cverf(void) 1516 { 1517 static nfsquad_t cverf; 1518 nfsquad_t ret; 1519 static int cverf_initialized = 0; 1520 1521 mtx_lock(&nfs_cverf_mtx); 1522 if (cverf_initialized == 0) { 1523 cverf.lval[0] = arc4random(); 1524 cverf.lval[1] = arc4random(); 1525 cverf_initialized = 1; 1526 } else 1527 cverf.qval++; 1528 ret = cverf; 1529 mtx_unlock(&nfs_cverf_mtx); 1530 1531 return (ret); 1532 } 1533 1534 /* 1535 * nfs file create call 1536 */ 1537 static int 1538 nfs_create(struct vop_create_args *ap) 1539 { 1540 struct vnode *dvp = ap->a_dvp; 1541 struct vattr *vap = ap->a_vap; 1542 struct componentname *cnp = ap->a_cnp; 1543 struct nfsnode *np = NULL, *dnp; 1544 struct vnode *newvp = NULL; 1545 struct nfsmount *nmp; 1546 struct nfsvattr dnfsva, nfsva; 1547 struct nfsfh *nfhp; 1548 nfsquad_t cverf; 1549 int error = 0, attrflag, dattrflag, fmode = 0; 1550 struct vattr vattr; 1551 1552 /* 1553 * Oops, not for me.. 1554 */ 1555 if (vap->va_type == VSOCK) 1556 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1557 1558 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) 1559 return (error); 1560 if (vap->va_vaflags & VA_EXCLUSIVE) 1561 fmode |= O_EXCL; 1562 dnp = VTONFS(dvp); 1563 nmp = VFSTONFS(vnode_mount(dvp)); 1564 again: 1565 /* For NFSv4, wait until any remove is done. */ 1566 mtx_lock(&dnp->n_mtx); 1567 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) { 1568 dnp->n_flag |= NREMOVEWANT; 1569 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0); 1570 } 1571 mtx_unlock(&dnp->n_mtx); 1572 1573 cverf = nfs_get_cverf(); 1574 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1575 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, 1576 &nfhp, &attrflag, &dattrflag, NULL); 1577 if (!error) { 1578 if (nfhp == NULL) 1579 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, 1580 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, 1581 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 1582 NULL); 1583 if (nfhp != NULL) 1584 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, 1585 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); 1586 } 1587 if (dattrflag) 1588 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1589 if (!error) { 1590 newvp = NFSTOV(np); 1591 if (attrflag == 0) 1592 error = nfsrpc_getattr(newvp, cnp->cn_cred, 1593 cnp->cn_thread, &nfsva, NULL); 1594 if (error == 0) 1595 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 1596 0, 1); 1597 } 1598 if (error) { 1599 if (newvp != NULL) { 1600 vput(newvp); 1601 newvp = NULL; 1602 } 1603 if (NFS_ISV34(dvp) && (fmode & O_EXCL) && 1604 error == NFSERR_NOTSUPP) { 1605 fmode &= ~O_EXCL; 1606 goto again; 1607 } 1608 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) { 1609 if (nfscl_checksattr(vap, &nfsva)) { 1610 /* 1611 * We are normally called with only a partially 1612 * initialized VAP. Since the NFSv3 spec says that 1613 * the server may use the file attributes to 1614 * store the verifier, the spec requires us to do a 1615 * SETATTR RPC. FreeBSD servers store the verifier in 1616 * atime, but we can't really assume that all servers 1617 * will so we ensure that our SETATTR sets both atime 1618 * and mtime. 1619 */ 1620 if (vap->va_mtime.tv_sec == VNOVAL) 1621 vfs_timestamp(&vap->va_mtime); 1622 if (vap->va_atime.tv_sec == VNOVAL) 1623 vap->va_atime = vap->va_mtime; 1624 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred, 1625 cnp->cn_thread, &nfsva, &attrflag, NULL); 1626 if (error && (vap->va_uid != (uid_t)VNOVAL || 1627 vap->va_gid != (gid_t)VNOVAL)) { 1628 /* try again without setting uid/gid */ 1629 vap->va_uid = (uid_t)VNOVAL; 1630 vap->va_gid = (uid_t)VNOVAL; 1631 error = nfsrpc_setattr(newvp, vap, NULL, 1632 cnp->cn_cred, cnp->cn_thread, &nfsva, 1633 &attrflag, NULL); 1634 } 1635 if (attrflag) 1636 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 1637 NULL, 0, 1); 1638 if (error != 0) 1639 vput(newvp); 1640 } 1641 } 1642 if (!error) { 1643 if ((cnp->cn_flags & MAKEENTRY) && attrflag) 1644 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 1645 NULL); 1646 *ap->a_vpp = newvp; 1647 } else if (NFS_ISV4(dvp)) { 1648 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, 1649 vap->va_gid); 1650 } 1651 mtx_lock(&dnp->n_mtx); 1652 dnp->n_flag |= NMODIFIED; 1653 if (!dattrflag) { 1654 dnp->n_attrstamp = 0; 1655 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1656 } 1657 mtx_unlock(&dnp->n_mtx); 1658 return (error); 1659 } 1660 1661 /* 1662 * nfs file remove call 1663 * To try and make nfs semantics closer to ufs semantics, a file that has 1664 * other processes using the vnode is renamed instead of removed and then 1665 * removed later on the last close. 1666 * - If v_usecount > 1 1667 * If a rename is not already in the works 1668 * call nfs_sillyrename() to set it up 1669 * else 1670 * do the remove rpc 1671 */ 1672 static int 1673 nfs_remove(struct vop_remove_args *ap) 1674 { 1675 struct vnode *vp = ap->a_vp; 1676 struct vnode *dvp = ap->a_dvp; 1677 struct componentname *cnp = ap->a_cnp; 1678 struct nfsnode *np = VTONFS(vp); 1679 int error = 0; 1680 struct vattr vattr; 1681 1682 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name")); 1683 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount")); 1684 if (vp->v_type == VDIR) 1685 error = EPERM; 1686 else if (vrefcnt(vp) == 1 || (np->n_sillyrename && 1687 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 1688 vattr.va_nlink > 1)) { 1689 /* 1690 * Purge the name cache so that the chance of a lookup for 1691 * the name succeeding while the remove is in progress is 1692 * minimized. Without node locking it can still happen, such 1693 * that an I/O op returns ESTALE, but since you get this if 1694 * another host removes the file.. 1695 */ 1696 cache_purge(vp); 1697 /* 1698 * throw away biocache buffers, mainly to avoid 1699 * unnecessary delayed writes later. 1700 */ 1701 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1); 1702 /* Do the rpc */ 1703 if (error != EINTR && error != EIO) 1704 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr, 1705 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread); 1706 /* 1707 * Kludge City: If the first reply to the remove rpc is lost.. 1708 * the reply to the retransmitted request will be ENOENT 1709 * since the file was in fact removed 1710 * Therefore, we cheat and return success. 1711 */ 1712 if (error == ENOENT) 1713 error = 0; 1714 } else if (!np->n_sillyrename) 1715 error = nfs_sillyrename(dvp, vp, cnp); 1716 mtx_lock(&np->n_mtx); 1717 np->n_attrstamp = 0; 1718 mtx_unlock(&np->n_mtx); 1719 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 1720 return (error); 1721 } 1722 1723 /* 1724 * nfs file remove rpc called from nfs_inactive 1725 */ 1726 int 1727 ncl_removeit(struct sillyrename *sp, struct vnode *vp) 1728 { 1729 /* 1730 * Make sure that the directory vnode is still valid. 1731 * XXX we should lock sp->s_dvp here. 1732 */ 1733 if (sp->s_dvp->v_type == VBAD) 1734 return (0); 1735 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen, 1736 sp->s_cred, NULL)); 1737 } 1738 1739 /* 1740 * Nfs remove rpc, called from nfs_remove() and ncl_removeit(). 1741 */ 1742 static int 1743 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, 1744 int namelen, struct ucred *cred, struct thread *td) 1745 { 1746 struct nfsvattr dnfsva; 1747 struct nfsnode *dnp = VTONFS(dvp); 1748 int error = 0, dattrflag; 1749 1750 mtx_lock(&dnp->n_mtx); 1751 dnp->n_flag |= NREMOVEINPROG; 1752 mtx_unlock(&dnp->n_mtx); 1753 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva, 1754 &dattrflag, NULL); 1755 mtx_lock(&dnp->n_mtx); 1756 if ((dnp->n_flag & NREMOVEWANT)) { 1757 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG); 1758 mtx_unlock(&dnp->n_mtx); 1759 wakeup((caddr_t)dnp); 1760 } else { 1761 dnp->n_flag &= ~NREMOVEINPROG; 1762 mtx_unlock(&dnp->n_mtx); 1763 } 1764 if (dattrflag) 1765 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 1766 mtx_lock(&dnp->n_mtx); 1767 dnp->n_flag |= NMODIFIED; 1768 if (!dattrflag) { 1769 dnp->n_attrstamp = 0; 1770 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 1771 } 1772 mtx_unlock(&dnp->n_mtx); 1773 if (error && NFS_ISV4(dvp)) 1774 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1775 return (error); 1776 } 1777 1778 /* 1779 * nfs file rename call 1780 */ 1781 static int 1782 nfs_rename(struct vop_rename_args *ap) 1783 { 1784 struct vnode *fvp = ap->a_fvp; 1785 struct vnode *tvp = ap->a_tvp; 1786 struct vnode *fdvp = ap->a_fdvp; 1787 struct vnode *tdvp = ap->a_tdvp; 1788 struct componentname *tcnp = ap->a_tcnp; 1789 struct componentname *fcnp = ap->a_fcnp; 1790 struct nfsnode *fnp = VTONFS(ap->a_fvp); 1791 struct nfsnode *tdnp = VTONFS(ap->a_tdvp); 1792 struct nfsv4node *newv4 = NULL; 1793 int error; 1794 1795 KASSERT((tcnp->cn_flags & HASBUF) != 0 && 1796 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name")); 1797 /* Check for cross-device rename */ 1798 if ((fvp->v_mount != tdvp->v_mount) || 1799 (tvp && (fvp->v_mount != tvp->v_mount))) { 1800 error = EXDEV; 1801 goto out; 1802 } 1803 1804 if (fvp == tvp) { 1805 ncl_printf("nfs_rename: fvp == tvp (can't happen)\n"); 1806 error = 0; 1807 goto out; 1808 } 1809 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0) 1810 goto out; 1811 1812 /* 1813 * We have to flush B_DELWRI data prior to renaming 1814 * the file. If we don't, the delayed-write buffers 1815 * can be flushed out later after the file has gone stale 1816 * under NFSV3. NFSV2 does not have this problem because 1817 * ( as far as I can tell ) it flushes dirty buffers more 1818 * often. 1819 * 1820 * Skip the rename operation if the fsync fails, this can happen 1821 * due to the server's volume being full, when we pushed out data 1822 * that was written back to our cache earlier. Not checking for 1823 * this condition can result in potential (silent) data loss. 1824 */ 1825 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread); 1826 NFSVOPUNLOCK(fvp, 0); 1827 if (!error && tvp) 1828 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread); 1829 if (error) 1830 goto out; 1831 1832 /* 1833 * If the tvp exists and is in use, sillyrename it before doing the 1834 * rename of the new file over it. 1835 * XXX Can't sillyrename a directory. 1836 */ 1837 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename && 1838 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1839 vput(tvp); 1840 tvp = NULL; 1841 } 1842 1843 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1844 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1845 tcnp->cn_thread); 1846 1847 if (error == 0 && NFS_ISV4(tdvp)) { 1848 /* 1849 * For NFSv4, check to see if it is the same name and 1850 * replace the name, if it is different. 1851 */ 1852 MALLOC(newv4, struct nfsv4node *, 1853 sizeof (struct nfsv4node) + 1854 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1, 1855 M_NFSV4NODE, M_WAITOK); 1856 mtx_lock(&tdnp->n_mtx); 1857 mtx_lock(&fnp->n_mtx); 1858 if (fnp->n_v4 != NULL && fvp->v_type == VREG && 1859 (fnp->n_v4->n4_namelen != tcnp->cn_namelen || 1860 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4), 1861 tcnp->cn_namelen) || 1862 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen || 1863 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 1864 tdnp->n_fhp->nfh_len))) { 1865 #ifdef notdef 1866 { char nnn[100]; int nnnl; 1867 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99; 1868 bcopy(tcnp->cn_nameptr, nnn, nnnl); 1869 nnn[nnnl] = '\0'; 1870 printf("ren replace=%s\n",nnn); 1871 } 1872 #endif 1873 FREE((caddr_t)fnp->n_v4, M_NFSV4NODE); 1874 fnp->n_v4 = newv4; 1875 newv4 = NULL; 1876 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len; 1877 fnp->n_v4->n4_namelen = tcnp->cn_namelen; 1878 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, 1879 tdnp->n_fhp->nfh_len); 1880 NFSBCOPY(tcnp->cn_nameptr, 1881 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen); 1882 } 1883 mtx_unlock(&tdnp->n_mtx); 1884 mtx_unlock(&fnp->n_mtx); 1885 if (newv4 != NULL) 1886 FREE((caddr_t)newv4, M_NFSV4NODE); 1887 } 1888 1889 if (fvp->v_type == VDIR) { 1890 if (tvp != NULL && tvp->v_type == VDIR) 1891 cache_purge(tdvp); 1892 cache_purge(fdvp); 1893 } 1894 1895 out: 1896 if (tdvp == tvp) 1897 vrele(tdvp); 1898 else 1899 vput(tdvp); 1900 if (tvp) 1901 vput(tvp); 1902 vrele(fdvp); 1903 vrele(fvp); 1904 /* 1905 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1906 */ 1907 if (error == ENOENT) 1908 error = 0; 1909 return (error); 1910 } 1911 1912 /* 1913 * nfs file rename rpc called from nfs_remove() above 1914 */ 1915 static int 1916 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp, 1917 struct sillyrename *sp) 1918 { 1919 1920 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen, 1921 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred, 1922 scnp->cn_thread)); 1923 } 1924 1925 /* 1926 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1927 */ 1928 static int 1929 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr, 1930 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr, 1931 int tnamelen, struct ucred *cred, struct thread *td) 1932 { 1933 struct nfsvattr fnfsva, tnfsva; 1934 struct nfsnode *fdnp = VTONFS(fdvp); 1935 struct nfsnode *tdnp = VTONFS(tdvp); 1936 int error = 0, fattrflag, tattrflag; 1937 1938 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp, 1939 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag, 1940 &tattrflag, NULL, NULL); 1941 mtx_lock(&fdnp->n_mtx); 1942 fdnp->n_flag |= NMODIFIED; 1943 if (fattrflag != 0) { 1944 mtx_unlock(&fdnp->n_mtx); 1945 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1); 1946 } else { 1947 fdnp->n_attrstamp = 0; 1948 mtx_unlock(&fdnp->n_mtx); 1949 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp); 1950 } 1951 mtx_lock(&tdnp->n_mtx); 1952 tdnp->n_flag |= NMODIFIED; 1953 if (tattrflag != 0) { 1954 mtx_unlock(&tdnp->n_mtx); 1955 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1); 1956 } else { 1957 tdnp->n_attrstamp = 0; 1958 mtx_unlock(&tdnp->n_mtx); 1959 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 1960 } 1961 if (error && NFS_ISV4(fdvp)) 1962 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1963 return (error); 1964 } 1965 1966 /* 1967 * nfs hard link create call 1968 */ 1969 static int 1970 nfs_link(struct vop_link_args *ap) 1971 { 1972 struct vnode *vp = ap->a_vp; 1973 struct vnode *tdvp = ap->a_tdvp; 1974 struct componentname *cnp = ap->a_cnp; 1975 struct nfsnode *np, *tdnp; 1976 struct nfsvattr nfsva, dnfsva; 1977 int error = 0, attrflag, dattrflag; 1978 1979 /* 1980 * Push all writes to the server, so that the attribute cache 1981 * doesn't get "out of sync" with the server. 1982 * XXX There should be a better way! 1983 */ 1984 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread); 1985 1986 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 1987 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag, 1988 &dattrflag, NULL); 1989 tdnp = VTONFS(tdvp); 1990 mtx_lock(&tdnp->n_mtx); 1991 tdnp->n_flag |= NMODIFIED; 1992 if (dattrflag != 0) { 1993 mtx_unlock(&tdnp->n_mtx); 1994 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1); 1995 } else { 1996 tdnp->n_attrstamp = 0; 1997 mtx_unlock(&tdnp->n_mtx); 1998 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 1999 } 2000 if (attrflag) 2001 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2002 else { 2003 np = VTONFS(vp); 2004 mtx_lock(&np->n_mtx); 2005 np->n_attrstamp = 0; 2006 mtx_unlock(&np->n_mtx); 2007 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 2008 } 2009 /* 2010 * If negative lookup caching is enabled, I might as well 2011 * add an entry for this node. Not necessary for correctness, 2012 * but if negative caching is enabled, then the system 2013 * must care about lookup caching hit rate, so... 2014 */ 2015 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 && 2016 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 2017 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL); 2018 } 2019 if (error && NFS_ISV4(vp)) 2020 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, 2021 (gid_t)0); 2022 return (error); 2023 } 2024 2025 /* 2026 * nfs symbolic link create call 2027 */ 2028 static int 2029 nfs_symlink(struct vop_symlink_args *ap) 2030 { 2031 struct vnode *dvp = ap->a_dvp; 2032 struct vattr *vap = ap->a_vap; 2033 struct componentname *cnp = ap->a_cnp; 2034 struct nfsvattr nfsva, dnfsva; 2035 struct nfsfh *nfhp; 2036 struct nfsnode *np = NULL, *dnp; 2037 struct vnode *newvp = NULL; 2038 int error = 0, attrflag, dattrflag, ret; 2039 2040 vap->va_type = VLNK; 2041 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2042 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, 2043 &nfsva, &nfhp, &attrflag, &dattrflag, NULL); 2044 if (nfhp) { 2045 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, 2046 &np, NULL, LK_EXCLUSIVE); 2047 if (!ret) 2048 newvp = NFSTOV(np); 2049 else if (!error) 2050 error = ret; 2051 } 2052 if (newvp != NULL) { 2053 if (attrflag) 2054 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 2055 0, 1); 2056 } else if (!error) { 2057 /* 2058 * If we do not have an error and we could not extract the 2059 * newvp from the response due to the request being NFSv2, we 2060 * have to do a lookup in order to obtain a newvp to return. 2061 */ 2062 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2063 cnp->cn_cred, cnp->cn_thread, &np); 2064 if (!error) 2065 newvp = NFSTOV(np); 2066 } 2067 if (error) { 2068 if (newvp) 2069 vput(newvp); 2070 if (NFS_ISV4(dvp)) 2071 error = nfscl_maperr(cnp->cn_thread, error, 2072 vap->va_uid, vap->va_gid); 2073 } else { 2074 *ap->a_vpp = newvp; 2075 } 2076 2077 dnp = VTONFS(dvp); 2078 mtx_lock(&dnp->n_mtx); 2079 dnp->n_flag |= NMODIFIED; 2080 if (dattrflag != 0) { 2081 mtx_unlock(&dnp->n_mtx); 2082 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2083 } else { 2084 dnp->n_attrstamp = 0; 2085 mtx_unlock(&dnp->n_mtx); 2086 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2087 } 2088 /* 2089 * If negative lookup caching is enabled, I might as well 2090 * add an entry for this node. Not necessary for correctness, 2091 * but if negative caching is enabled, then the system 2092 * must care about lookup caching hit rate, so... 2093 */ 2094 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2095 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 2096 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL); 2097 } 2098 return (error); 2099 } 2100 2101 /* 2102 * nfs make dir call 2103 */ 2104 static int 2105 nfs_mkdir(struct vop_mkdir_args *ap) 2106 { 2107 struct vnode *dvp = ap->a_dvp; 2108 struct vattr *vap = ap->a_vap; 2109 struct componentname *cnp = ap->a_cnp; 2110 struct nfsnode *np = NULL, *dnp; 2111 struct vnode *newvp = NULL; 2112 struct vattr vattr; 2113 struct nfsfh *nfhp; 2114 struct nfsvattr nfsva, dnfsva; 2115 int error = 0, attrflag, dattrflag, ret; 2116 2117 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) 2118 return (error); 2119 vap->va_type = VDIR; 2120 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2121 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp, 2122 &attrflag, &dattrflag, NULL); 2123 dnp = VTONFS(dvp); 2124 mtx_lock(&dnp->n_mtx); 2125 dnp->n_flag |= NMODIFIED; 2126 if (dattrflag != 0) { 2127 mtx_unlock(&dnp->n_mtx); 2128 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2129 } else { 2130 dnp->n_attrstamp = 0; 2131 mtx_unlock(&dnp->n_mtx); 2132 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2133 } 2134 if (nfhp) { 2135 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, 2136 &np, NULL, LK_EXCLUSIVE); 2137 if (!ret) { 2138 newvp = NFSTOV(np); 2139 if (attrflag) 2140 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 2141 NULL, 0, 1); 2142 } else if (!error) 2143 error = ret; 2144 } 2145 if (!error && newvp == NULL) { 2146 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2147 cnp->cn_cred, cnp->cn_thread, &np); 2148 if (!error) { 2149 newvp = NFSTOV(np); 2150 if (newvp->v_type != VDIR) 2151 error = EEXIST; 2152 } 2153 } 2154 if (error) { 2155 if (newvp) 2156 vput(newvp); 2157 if (NFS_ISV4(dvp)) 2158 error = nfscl_maperr(cnp->cn_thread, error, 2159 vap->va_uid, vap->va_gid); 2160 } else { 2161 /* 2162 * If negative lookup caching is enabled, I might as well 2163 * add an entry for this node. Not necessary for correctness, 2164 * but if negative caching is enabled, then the system 2165 * must care about lookup caching hit rate, so... 2166 */ 2167 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2168 (cnp->cn_flags & MAKEENTRY) && 2169 attrflag != 0 && dattrflag != 0) 2170 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 2171 &dnfsva.na_ctime); 2172 *ap->a_vpp = newvp; 2173 } 2174 return (error); 2175 } 2176 2177 /* 2178 * nfs remove directory call 2179 */ 2180 static int 2181 nfs_rmdir(struct vop_rmdir_args *ap) 2182 { 2183 struct vnode *vp = ap->a_vp; 2184 struct vnode *dvp = ap->a_dvp; 2185 struct componentname *cnp = ap->a_cnp; 2186 struct nfsnode *dnp; 2187 struct nfsvattr dnfsva; 2188 int error, dattrflag; 2189 2190 if (dvp == vp) 2191 return (EINVAL); 2192 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2193 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL); 2194 dnp = VTONFS(dvp); 2195 mtx_lock(&dnp->n_mtx); 2196 dnp->n_flag |= NMODIFIED; 2197 if (dattrflag != 0) { 2198 mtx_unlock(&dnp->n_mtx); 2199 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2200 } else { 2201 dnp->n_attrstamp = 0; 2202 mtx_unlock(&dnp->n_mtx); 2203 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2204 } 2205 2206 cache_purge(dvp); 2207 cache_purge(vp); 2208 if (error && NFS_ISV4(dvp)) 2209 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, 2210 (gid_t)0); 2211 /* 2212 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2213 */ 2214 if (error == ENOENT) 2215 error = 0; 2216 return (error); 2217 } 2218 2219 /* 2220 * nfs readdir call 2221 */ 2222 static int 2223 nfs_readdir(struct vop_readdir_args *ap) 2224 { 2225 struct vnode *vp = ap->a_vp; 2226 struct nfsnode *np = VTONFS(vp); 2227 struct uio *uio = ap->a_uio; 2228 ssize_t tresid; 2229 int error = 0; 2230 struct vattr vattr; 2231 2232 if (ap->a_eofflag != NULL) 2233 *ap->a_eofflag = 0; 2234 if (vp->v_type != VDIR) 2235 return(EPERM); 2236 2237 /* 2238 * First, check for hit on the EOF offset cache 2239 */ 2240 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2241 (np->n_flag & NMODIFIED) == 0) { 2242 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) { 2243 mtx_lock(&np->n_mtx); 2244 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) || 2245 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 2246 mtx_unlock(&np->n_mtx); 2247 NFSINCRGLOBAL(newnfsstats.direofcache_hits); 2248 if (ap->a_eofflag != NULL) 2249 *ap->a_eofflag = 1; 2250 return (0); 2251 } else 2252 mtx_unlock(&np->n_mtx); 2253 } 2254 } 2255 2256 /* 2257 * Call ncl_bioread() to do the real work. 2258 */ 2259 tresid = uio->uio_resid; 2260 error = ncl_bioread(vp, uio, 0, ap->a_cred); 2261 2262 if (!error && uio->uio_resid == tresid) { 2263 NFSINCRGLOBAL(newnfsstats.direofcache_misses); 2264 if (ap->a_eofflag != NULL) 2265 *ap->a_eofflag = 1; 2266 } 2267 return (error); 2268 } 2269 2270 /* 2271 * Readdir rpc call. 2272 * Called from below the buffer cache by ncl_doio(). 2273 */ 2274 int 2275 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2276 struct thread *td) 2277 { 2278 struct nfsvattr nfsva; 2279 nfsuint64 *cookiep, cookie; 2280 struct nfsnode *dnp = VTONFS(vp); 2281 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2282 int error = 0, eof, attrflag; 2283 2284 KASSERT(uiop->uio_iovcnt == 1 && 2285 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2286 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2287 ("nfs readdirrpc bad uio")); 2288 2289 /* 2290 * If there is no cookie, assume directory was stale. 2291 */ 2292 ncl_dircookie_lock(dnp); 2293 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2294 if (cookiep) { 2295 cookie = *cookiep; 2296 ncl_dircookie_unlock(dnp); 2297 } else { 2298 ncl_dircookie_unlock(dnp); 2299 return (NFSERR_BAD_COOKIE); 2300 } 2301 2302 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2303 (void)ncl_fsinfo(nmp, vp, cred, td); 2304 2305 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva, 2306 &attrflag, &eof, NULL); 2307 if (attrflag) 2308 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2309 2310 if (!error) { 2311 /* 2312 * We are now either at the end of the directory or have filled 2313 * the block. 2314 */ 2315 if (eof) 2316 dnp->n_direofoffset = uiop->uio_offset; 2317 else { 2318 if (uiop->uio_resid > 0) 2319 ncl_printf("EEK! readdirrpc resid > 0\n"); 2320 ncl_dircookie_lock(dnp); 2321 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2322 *cookiep = cookie; 2323 ncl_dircookie_unlock(dnp); 2324 } 2325 } else if (NFS_ISV4(vp)) { 2326 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2327 } 2328 return (error); 2329 } 2330 2331 /* 2332 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc(). 2333 */ 2334 int 2335 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2336 struct thread *td) 2337 { 2338 struct nfsvattr nfsva; 2339 nfsuint64 *cookiep, cookie; 2340 struct nfsnode *dnp = VTONFS(vp); 2341 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2342 int error = 0, attrflag, eof; 2343 2344 KASSERT(uiop->uio_iovcnt == 1 && 2345 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2346 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2347 ("nfs readdirplusrpc bad uio")); 2348 2349 /* 2350 * If there is no cookie, assume directory was stale. 2351 */ 2352 ncl_dircookie_lock(dnp); 2353 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2354 if (cookiep) { 2355 cookie = *cookiep; 2356 ncl_dircookie_unlock(dnp); 2357 } else { 2358 ncl_dircookie_unlock(dnp); 2359 return (NFSERR_BAD_COOKIE); 2360 } 2361 2362 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2363 (void)ncl_fsinfo(nmp, vp, cred, td); 2364 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva, 2365 &attrflag, &eof, NULL); 2366 if (attrflag) 2367 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2368 2369 if (!error) { 2370 /* 2371 * We are now either at end of the directory or have filled the 2372 * the block. 2373 */ 2374 if (eof) 2375 dnp->n_direofoffset = uiop->uio_offset; 2376 else { 2377 if (uiop->uio_resid > 0) 2378 ncl_printf("EEK! readdirplusrpc resid > 0\n"); 2379 ncl_dircookie_lock(dnp); 2380 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2381 *cookiep = cookie; 2382 ncl_dircookie_unlock(dnp); 2383 } 2384 } else if (NFS_ISV4(vp)) { 2385 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2386 } 2387 return (error); 2388 } 2389 2390 /* 2391 * Silly rename. To make the NFS filesystem that is stateless look a little 2392 * more like the "ufs" a remove of an active vnode is translated to a rename 2393 * to a funny looking filename that is removed by nfs_inactive on the 2394 * nfsnode. There is the potential for another process on a different client 2395 * to create the same funny name between the nfs_lookitup() fails and the 2396 * nfs_rename() completes, but... 2397 */ 2398 static int 2399 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2400 { 2401 struct sillyrename *sp; 2402 struct nfsnode *np; 2403 int error; 2404 short pid; 2405 unsigned int lticks; 2406 2407 cache_purge(dvp); 2408 np = VTONFS(vp); 2409 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir")); 2410 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2411 M_NEWNFSREQ, M_WAITOK); 2412 sp->s_cred = crhold(cnp->cn_cred); 2413 sp->s_dvp = dvp; 2414 VREF(dvp); 2415 2416 /* 2417 * Fudge together a funny name. 2418 * Changing the format of the funny name to accomodate more 2419 * sillynames per directory. 2420 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is 2421 * CPU ticks since boot. 2422 */ 2423 pid = cnp->cn_thread->td_proc->p_pid; 2424 lticks = (unsigned int)ticks; 2425 for ( ; ; ) { 2426 sp->s_namlen = sprintf(sp->s_name, 2427 ".nfs.%08x.%04x4.4", lticks, 2428 pid); 2429 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2430 cnp->cn_thread, NULL)) 2431 break; 2432 lticks++; 2433 } 2434 error = nfs_renameit(dvp, vp, cnp, sp); 2435 if (error) 2436 goto bad; 2437 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2438 cnp->cn_thread, &np); 2439 np->n_sillyrename = sp; 2440 return (0); 2441 bad: 2442 vrele(sp->s_dvp); 2443 crfree(sp->s_cred); 2444 free((caddr_t)sp, M_NEWNFSREQ); 2445 return (error); 2446 } 2447 2448 /* 2449 * Look up a file name and optionally either update the file handle or 2450 * allocate an nfsnode, depending on the value of npp. 2451 * npp == NULL --> just do the lookup 2452 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2453 * handled too 2454 * *npp != NULL --> update the file handle in the vnode 2455 */ 2456 static int 2457 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred, 2458 struct thread *td, struct nfsnode **npp) 2459 { 2460 struct vnode *newvp = NULL, *vp; 2461 struct nfsnode *np, *dnp = VTONFS(dvp); 2462 struct nfsfh *nfhp, *onfhp; 2463 struct nfsvattr nfsva, dnfsva; 2464 struct componentname cn; 2465 int error = 0, attrflag, dattrflag; 2466 u_int hash; 2467 2468 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva, 2469 &nfhp, &attrflag, &dattrflag, NULL); 2470 if (dattrflag) 2471 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2472 if (npp && !error) { 2473 if (*npp != NULL) { 2474 np = *npp; 2475 vp = NFSTOV(np); 2476 /* 2477 * For NFSv4, check to see if it is the same name and 2478 * replace the name, if it is different. 2479 */ 2480 if (np->n_v4 != NULL && nfsva.na_type == VREG && 2481 (np->n_v4->n4_namelen != len || 2482 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) || 2483 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || 2484 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2485 dnp->n_fhp->nfh_len))) { 2486 #ifdef notdef 2487 { char nnn[100]; int nnnl; 2488 nnnl = (len < 100) ? len : 99; 2489 bcopy(name, nnn, nnnl); 2490 nnn[nnnl] = '\0'; 2491 printf("replace=%s\n",nnn); 2492 } 2493 #endif 2494 FREE((caddr_t)np->n_v4, M_NFSV4NODE); 2495 MALLOC(np->n_v4, struct nfsv4node *, 2496 sizeof (struct nfsv4node) + 2497 dnp->n_fhp->nfh_len + len - 1, 2498 M_NFSV4NODE, M_WAITOK); 2499 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; 2500 np->n_v4->n4_namelen = len; 2501 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2502 dnp->n_fhp->nfh_len); 2503 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len); 2504 } 2505 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, 2506 FNV1_32_INIT); 2507 onfhp = np->n_fhp; 2508 /* 2509 * Rehash node for new file handle. 2510 */ 2511 vfs_hash_rehash(vp, hash); 2512 np->n_fhp = nfhp; 2513 if (onfhp != NULL) 2514 FREE((caddr_t)onfhp, M_NFSFH); 2515 newvp = NFSTOV(np); 2516 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) { 2517 FREE((caddr_t)nfhp, M_NFSFH); 2518 VREF(dvp); 2519 newvp = dvp; 2520 } else { 2521 cn.cn_nameptr = name; 2522 cn.cn_namelen = len; 2523 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td, 2524 &np, NULL, LK_EXCLUSIVE); 2525 if (error) 2526 return (error); 2527 newvp = NFSTOV(np); 2528 } 2529 if (!attrflag && *npp == NULL) { 2530 if (newvp == dvp) 2531 vrele(newvp); 2532 else 2533 vput(newvp); 2534 return (ENOENT); 2535 } 2536 if (attrflag) 2537 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 2538 0, 1); 2539 } 2540 if (npp && *npp == NULL) { 2541 if (error) { 2542 if (newvp) { 2543 if (newvp == dvp) 2544 vrele(newvp); 2545 else 2546 vput(newvp); 2547 } 2548 } else 2549 *npp = np; 2550 } 2551 if (error && NFS_ISV4(dvp)) 2552 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2553 return (error); 2554 } 2555 2556 /* 2557 * Nfs Version 3 and 4 commit rpc 2558 */ 2559 int 2560 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred, 2561 struct thread *td) 2562 { 2563 struct nfsvattr nfsva; 2564 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2565 int error, attrflag; 2566 2567 mtx_lock(&nmp->nm_mtx); 2568 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) { 2569 mtx_unlock(&nmp->nm_mtx); 2570 return (0); 2571 } 2572 mtx_unlock(&nmp->nm_mtx); 2573 error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva, 2574 &attrflag, NULL); 2575 if (attrflag != 0) 2576 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 2577 0, 1); 2578 if (error != 0 && NFS_ISV4(vp)) 2579 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2580 return (error); 2581 } 2582 2583 /* 2584 * Strategy routine. 2585 * For async requests when nfsiod(s) are running, queue the request by 2586 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the 2587 * request. 2588 */ 2589 static int 2590 nfs_strategy(struct vop_strategy_args *ap) 2591 { 2592 struct buf *bp = ap->a_bp; 2593 struct ucred *cr; 2594 2595 KASSERT(!(bp->b_flags & B_DONE), 2596 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); 2597 BUF_ASSERT_HELD(bp); 2598 2599 if (bp->b_iocmd == BIO_READ) 2600 cr = bp->b_rcred; 2601 else 2602 cr = bp->b_wcred; 2603 2604 /* 2605 * If the op is asynchronous and an i/o daemon is waiting 2606 * queue the request, wake it up and wait for completion 2607 * otherwise just do it ourselves. 2608 */ 2609 if ((bp->b_flags & B_ASYNC) == 0 || 2610 ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread)) 2611 (void) ncl_doio(ap->a_vp, bp, cr, curthread, 1); 2612 return (0); 2613 } 2614 2615 /* 2616 * fsync vnode op. Just call ncl_flush() with commit == 1. 2617 */ 2618 /* ARGSUSED */ 2619 static int 2620 nfs_fsync(struct vop_fsync_args *ap) 2621 { 2622 2623 if (ap->a_vp->v_type != VREG) { 2624 /* 2625 * For NFS, metadata is changed synchronously on the server, 2626 * so there is nothing to flush. Also, ncl_flush() clears 2627 * the NMODIFIED flag and that shouldn't be done here for 2628 * directories. 2629 */ 2630 return (0); 2631 } 2632 return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1, 0)); 2633 } 2634 2635 /* 2636 * Flush all the blocks associated with a vnode. 2637 * Walk through the buffer pool and push any dirty pages 2638 * associated with the vnode. 2639 * If the called_from_renewthread argument is TRUE, it has been called 2640 * from the NFSv4 renew thread and, as such, cannot block indefinitely 2641 * waiting for a buffer write to complete. 2642 */ 2643 int 2644 ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td, 2645 int commit, int called_from_renewthread) 2646 { 2647 struct nfsnode *np = VTONFS(vp); 2648 struct buf *bp; 2649 int i; 2650 struct buf *nbp; 2651 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2652 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos; 2653 int passone = 1, trycnt = 0; 2654 u_quad_t off, endoff, toff; 2655 struct ucred* wcred = NULL; 2656 struct buf **bvec = NULL; 2657 struct bufobj *bo; 2658 #ifndef NFS_COMMITBVECSIZ 2659 #define NFS_COMMITBVECSIZ 20 2660 #endif 2661 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ]; 2662 int bvecsize = 0, bveccount; 2663 2664 if (called_from_renewthread != 0) 2665 slptimeo = hz; 2666 if (nmp->nm_flag & NFSMNT_INT) 2667 slpflag = PCATCH; 2668 if (!commit) 2669 passone = 0; 2670 bo = &vp->v_bufobj; 2671 /* 2672 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the 2673 * server, but has not been committed to stable storage on the server 2674 * yet. On the first pass, the byte range is worked out and the commit 2675 * rpc is done. On the second pass, ncl_writebp() is called to do the 2676 * job. 2677 */ 2678 again: 2679 off = (u_quad_t)-1; 2680 endoff = 0; 2681 bvecpos = 0; 2682 if (NFS_ISV34(vp) && commit) { 2683 if (bvec != NULL && bvec != bvec_on_stack) 2684 free(bvec, M_TEMP); 2685 /* 2686 * Count up how many buffers waiting for a commit. 2687 */ 2688 bveccount = 0; 2689 BO_LOCK(bo); 2690 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2691 if (!BUF_ISLOCKED(bp) && 2692 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) 2693 == (B_DELWRI | B_NEEDCOMMIT)) 2694 bveccount++; 2695 } 2696 /* 2697 * Allocate space to remember the list of bufs to commit. It is 2698 * important to use M_NOWAIT here to avoid a race with nfs_write. 2699 * If we can't get memory (for whatever reason), we will end up 2700 * committing the buffers one-by-one in the loop below. 2701 */ 2702 if (bveccount > NFS_COMMITBVECSIZ) { 2703 /* 2704 * Release the vnode interlock to avoid a lock 2705 * order reversal. 2706 */ 2707 BO_UNLOCK(bo); 2708 bvec = (struct buf **) 2709 malloc(bveccount * sizeof(struct buf *), 2710 M_TEMP, M_NOWAIT); 2711 BO_LOCK(bo); 2712 if (bvec == NULL) { 2713 bvec = bvec_on_stack; 2714 bvecsize = NFS_COMMITBVECSIZ; 2715 } else 2716 bvecsize = bveccount; 2717 } else { 2718 bvec = bvec_on_stack; 2719 bvecsize = NFS_COMMITBVECSIZ; 2720 } 2721 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2722 if (bvecpos >= bvecsize) 2723 break; 2724 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2725 nbp = TAILQ_NEXT(bp, b_bobufs); 2726 continue; 2727 } 2728 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 2729 (B_DELWRI | B_NEEDCOMMIT)) { 2730 BUF_UNLOCK(bp); 2731 nbp = TAILQ_NEXT(bp, b_bobufs); 2732 continue; 2733 } 2734 BO_UNLOCK(bo); 2735 bremfree(bp); 2736 /* 2737 * Work out if all buffers are using the same cred 2738 * so we can deal with them all with one commit. 2739 * 2740 * NOTE: we are not clearing B_DONE here, so we have 2741 * to do it later on in this routine if we intend to 2742 * initiate I/O on the bp. 2743 * 2744 * Note: to avoid loopback deadlocks, we do not 2745 * assign b_runningbufspace. 2746 */ 2747 if (wcred == NULL) 2748 wcred = bp->b_wcred; 2749 else if (wcred != bp->b_wcred) 2750 wcred = NOCRED; 2751 vfs_busy_pages(bp, 1); 2752 2753 BO_LOCK(bo); 2754 /* 2755 * bp is protected by being locked, but nbp is not 2756 * and vfs_busy_pages() may sleep. We have to 2757 * recalculate nbp. 2758 */ 2759 nbp = TAILQ_NEXT(bp, b_bobufs); 2760 2761 /* 2762 * A list of these buffers is kept so that the 2763 * second loop knows which buffers have actually 2764 * been committed. This is necessary, since there 2765 * may be a race between the commit rpc and new 2766 * uncommitted writes on the file. 2767 */ 2768 bvec[bvecpos++] = bp; 2769 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2770 bp->b_dirtyoff; 2771 if (toff < off) 2772 off = toff; 2773 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); 2774 if (toff > endoff) 2775 endoff = toff; 2776 } 2777 BO_UNLOCK(bo); 2778 } 2779 if (bvecpos > 0) { 2780 /* 2781 * Commit data on the server, as required. 2782 * If all bufs are using the same wcred, then use that with 2783 * one call for all of them, otherwise commit each one 2784 * separately. 2785 */ 2786 if (wcred != NOCRED) 2787 retv = ncl_commit(vp, off, (int)(endoff - off), 2788 wcred, td); 2789 else { 2790 retv = 0; 2791 for (i = 0; i < bvecpos; i++) { 2792 off_t off, size; 2793 bp = bvec[i]; 2794 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2795 bp->b_dirtyoff; 2796 size = (u_quad_t)(bp->b_dirtyend 2797 - bp->b_dirtyoff); 2798 retv = ncl_commit(vp, off, (int)size, 2799 bp->b_wcred, td); 2800 if (retv) break; 2801 } 2802 } 2803 2804 if (retv == NFSERR_STALEWRITEVERF) 2805 ncl_clearcommit(vp->v_mount); 2806 2807 /* 2808 * Now, either mark the blocks I/O done or mark the 2809 * blocks dirty, depending on whether the commit 2810 * succeeded. 2811 */ 2812 for (i = 0; i < bvecpos; i++) { 2813 bp = bvec[i]; 2814 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 2815 if (retv) { 2816 /* 2817 * Error, leave B_DELWRI intact 2818 */ 2819 vfs_unbusy_pages(bp); 2820 brelse(bp); 2821 } else { 2822 /* 2823 * Success, remove B_DELWRI ( bundirty() ). 2824 * 2825 * b_dirtyoff/b_dirtyend seem to be NFS 2826 * specific. We should probably move that 2827 * into bundirty(). XXX 2828 */ 2829 bufobj_wref(bo); 2830 bp->b_flags |= B_ASYNC; 2831 bundirty(bp); 2832 bp->b_flags &= ~B_DONE; 2833 bp->b_ioflags &= ~BIO_ERROR; 2834 bp->b_dirtyoff = bp->b_dirtyend = 0; 2835 bufdone(bp); 2836 } 2837 } 2838 } 2839 2840 /* 2841 * Start/do any write(s) that are required. 2842 */ 2843 loop: 2844 BO_LOCK(bo); 2845 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2846 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2847 if (waitfor != MNT_WAIT || passone) 2848 continue; 2849 2850 error = BUF_TIMELOCK(bp, 2851 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 2852 BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo); 2853 if (error == 0) { 2854 BUF_UNLOCK(bp); 2855 goto loop; 2856 } 2857 if (error == ENOLCK) { 2858 error = 0; 2859 goto loop; 2860 } 2861 if (called_from_renewthread != 0) { 2862 /* 2863 * Return EIO so the flush will be retried 2864 * later. 2865 */ 2866 error = EIO; 2867 goto done; 2868 } 2869 if (newnfs_sigintr(nmp, td)) { 2870 error = EINTR; 2871 goto done; 2872 } 2873 if (slpflag == PCATCH) { 2874 slpflag = 0; 2875 slptimeo = 2 * hz; 2876 } 2877 goto loop; 2878 } 2879 if ((bp->b_flags & B_DELWRI) == 0) 2880 panic("nfs_fsync: not dirty"); 2881 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) { 2882 BUF_UNLOCK(bp); 2883 continue; 2884 } 2885 BO_UNLOCK(bo); 2886 bremfree(bp); 2887 if (passone || !commit) 2888 bp->b_flags |= B_ASYNC; 2889 else 2890 bp->b_flags |= B_ASYNC; 2891 bwrite(bp); 2892 if (newnfs_sigintr(nmp, td)) { 2893 error = EINTR; 2894 goto done; 2895 } 2896 goto loop; 2897 } 2898 if (passone) { 2899 passone = 0; 2900 BO_UNLOCK(bo); 2901 goto again; 2902 } 2903 if (waitfor == MNT_WAIT) { 2904 while (bo->bo_numoutput) { 2905 error = bufobj_wwait(bo, slpflag, slptimeo); 2906 if (error) { 2907 BO_UNLOCK(bo); 2908 if (called_from_renewthread != 0) { 2909 /* 2910 * Return EIO so that the flush will be 2911 * retried later. 2912 */ 2913 error = EIO; 2914 goto done; 2915 } 2916 error = newnfs_sigintr(nmp, td); 2917 if (error) 2918 goto done; 2919 if (slpflag == PCATCH) { 2920 slpflag = 0; 2921 slptimeo = 2 * hz; 2922 } 2923 BO_LOCK(bo); 2924 } 2925 } 2926 if (bo->bo_dirty.bv_cnt != 0 && commit) { 2927 BO_UNLOCK(bo); 2928 goto loop; 2929 } 2930 /* 2931 * Wait for all the async IO requests to drain 2932 */ 2933 BO_UNLOCK(bo); 2934 mtx_lock(&np->n_mtx); 2935 while (np->n_directio_asyncwr > 0) { 2936 np->n_flag |= NFSYNCWAIT; 2937 error = newnfs_msleep(td, &np->n_directio_asyncwr, 2938 &np->n_mtx, slpflag | (PRIBIO + 1), 2939 "nfsfsync", 0); 2940 if (error) { 2941 if (newnfs_sigintr(nmp, td)) { 2942 mtx_unlock(&np->n_mtx); 2943 error = EINTR; 2944 goto done; 2945 } 2946 } 2947 } 2948 mtx_unlock(&np->n_mtx); 2949 } else 2950 BO_UNLOCK(bo); 2951 if (NFSHASPNFS(nmp)) { 2952 nfscl_layoutcommit(vp, td); 2953 /* 2954 * Invalidate the attribute cache, since writes to a DS 2955 * won't update the size attribute. 2956 */ 2957 mtx_lock(&np->n_mtx); 2958 np->n_attrstamp = 0; 2959 } else 2960 mtx_lock(&np->n_mtx); 2961 if (np->n_flag & NWRITEERR) { 2962 error = np->n_error; 2963 np->n_flag &= ~NWRITEERR; 2964 } 2965 if (commit && bo->bo_dirty.bv_cnt == 0 && 2966 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0) 2967 np->n_flag &= ~NMODIFIED; 2968 mtx_unlock(&np->n_mtx); 2969 done: 2970 if (bvec != NULL && bvec != bvec_on_stack) 2971 free(bvec, M_TEMP); 2972 if (error == 0 && commit != 0 && waitfor == MNT_WAIT && 2973 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 || 2974 np->n_directio_asyncwr != 0) && trycnt++ < 5) { 2975 /* try, try again... */ 2976 passone = 1; 2977 wcred = NULL; 2978 bvec = NULL; 2979 bvecsize = 0; 2980 printf("try%d\n", trycnt); 2981 goto again; 2982 } 2983 return (error); 2984 } 2985 2986 /* 2987 * NFS advisory byte-level locks. 2988 */ 2989 static int 2990 nfs_advlock(struct vop_advlock_args *ap) 2991 { 2992 struct vnode *vp = ap->a_vp; 2993 struct ucred *cred; 2994 struct nfsnode *np = VTONFS(ap->a_vp); 2995 struct proc *p = (struct proc *)ap->a_id; 2996 struct thread *td = curthread; /* XXX */ 2997 struct vattr va; 2998 int ret, error = EOPNOTSUPP; 2999 u_quad_t size; 3000 3001 if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) { 3002 if (vp->v_type != VREG) 3003 return (EINVAL); 3004 if ((ap->a_flags & F_POSIX) != 0) 3005 cred = p->p_ucred; 3006 else 3007 cred = td->td_ucred; 3008 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 3009 if (vp->v_iflag & VI_DOOMED) { 3010 NFSVOPUNLOCK(vp, 0); 3011 return (EBADF); 3012 } 3013 3014 /* 3015 * If this is unlocking a write locked region, flush and 3016 * commit them before unlocking. This is required by 3017 * RFC3530 Sec. 9.3.2. 3018 */ 3019 if (ap->a_op == F_UNLCK && 3020 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id, 3021 ap->a_flags)) 3022 (void) ncl_flush(vp, MNT_WAIT, cred, td, 1, 0); 3023 3024 /* 3025 * Loop around doing the lock op, while a blocking lock 3026 * must wait for the lock op to succeed. 3027 */ 3028 do { 3029 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op, 3030 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags); 3031 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3032 ap->a_op == F_SETLK) { 3033 NFSVOPUNLOCK(vp, 0); 3034 error = nfs_catnap(PZERO | PCATCH, ret, 3035 "ncladvl"); 3036 if (error) 3037 return (EINTR); 3038 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 3039 if (vp->v_iflag & VI_DOOMED) { 3040 NFSVOPUNLOCK(vp, 0); 3041 return (EBADF); 3042 } 3043 } 3044 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3045 ap->a_op == F_SETLK); 3046 if (ret == NFSERR_DENIED) { 3047 NFSVOPUNLOCK(vp, 0); 3048 return (EAGAIN); 3049 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) { 3050 NFSVOPUNLOCK(vp, 0); 3051 return (ret); 3052 } else if (ret != 0) { 3053 NFSVOPUNLOCK(vp, 0); 3054 return (EACCES); 3055 } 3056 3057 /* 3058 * Now, if we just got a lock, invalidate data in the buffer 3059 * cache, as required, so that the coherency conforms with 3060 * RFC3530 Sec. 9.3.2. 3061 */ 3062 if (ap->a_op == F_SETLK) { 3063 if ((np->n_flag & NMODIFIED) == 0) { 3064 np->n_attrstamp = 0; 3065 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3066 ret = VOP_GETATTR(vp, &va, cred); 3067 } 3068 if ((np->n_flag & NMODIFIED) || ret || 3069 np->n_change != va.va_filerev) { 3070 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1); 3071 np->n_attrstamp = 0; 3072 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3073 ret = VOP_GETATTR(vp, &va, cred); 3074 if (!ret) { 3075 np->n_mtime = va.va_mtime; 3076 np->n_change = va.va_filerev; 3077 } 3078 } 3079 /* Mark that a file lock has been acquired. */ 3080 mtx_lock(&np->n_mtx); 3081 np->n_flag |= NHASBEENLOCKED; 3082 mtx_unlock(&np->n_mtx); 3083 } 3084 NFSVOPUNLOCK(vp, 0); 3085 return (0); 3086 } else if (!NFS_ISV4(vp)) { 3087 error = NFSVOPLOCK(vp, LK_SHARED); 3088 if (error) 3089 return (error); 3090 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3091 size = VTONFS(vp)->n_size; 3092 NFSVOPUNLOCK(vp, 0); 3093 error = lf_advlock(ap, &(vp->v_lockf), size); 3094 } else { 3095 if (nfs_advlock_p != NULL) 3096 error = nfs_advlock_p(ap); 3097 else { 3098 NFSVOPUNLOCK(vp, 0); 3099 error = ENOLCK; 3100 } 3101 } 3102 if (error == 0 && ap->a_op == F_SETLK) { 3103 /* Mark that a file lock has been acquired. */ 3104 mtx_lock(&np->n_mtx); 3105 np->n_flag |= NHASBEENLOCKED; 3106 mtx_unlock(&np->n_mtx); 3107 } 3108 } 3109 return (error); 3110 } 3111 3112 /* 3113 * NFS advisory byte-level locks. 3114 */ 3115 static int 3116 nfs_advlockasync(struct vop_advlockasync_args *ap) 3117 { 3118 struct vnode *vp = ap->a_vp; 3119 u_quad_t size; 3120 int error; 3121 3122 if (NFS_ISV4(vp)) 3123 return (EOPNOTSUPP); 3124 error = NFSVOPLOCK(vp, LK_SHARED); 3125 if (error) 3126 return (error); 3127 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3128 size = VTONFS(vp)->n_size; 3129 NFSVOPUNLOCK(vp, 0); 3130 error = lf_advlockasync(ap, &(vp->v_lockf), size); 3131 } else { 3132 NFSVOPUNLOCK(vp, 0); 3133 error = EOPNOTSUPP; 3134 } 3135 return (error); 3136 } 3137 3138 /* 3139 * Print out the contents of an nfsnode. 3140 */ 3141 static int 3142 nfs_print(struct vop_print_args *ap) 3143 { 3144 struct vnode *vp = ap->a_vp; 3145 struct nfsnode *np = VTONFS(vp); 3146 3147 ncl_printf("\tfileid %ld fsid 0x%x", 3148 np->n_vattr.na_fileid, np->n_vattr.na_fsid); 3149 if (vp->v_type == VFIFO) 3150 fifo_printinfo(vp); 3151 printf("\n"); 3152 return (0); 3153 } 3154 3155 /* 3156 * This is the "real" nfs::bwrite(struct buf*). 3157 * We set B_CACHE if this is a VMIO buffer. 3158 */ 3159 int 3160 ncl_writebp(struct buf *bp, int force __unused, struct thread *td) 3161 { 3162 int s; 3163 int oldflags = bp->b_flags; 3164 #if 0 3165 int retv = 1; 3166 off_t off; 3167 #endif 3168 3169 BUF_ASSERT_HELD(bp); 3170 3171 if (bp->b_flags & B_INVAL) { 3172 brelse(bp); 3173 return(0); 3174 } 3175 3176 bp->b_flags |= B_CACHE; 3177 3178 /* 3179 * Undirty the bp. We will redirty it later if the I/O fails. 3180 */ 3181 3182 s = splbio(); 3183 bundirty(bp); 3184 bp->b_flags &= ~B_DONE; 3185 bp->b_ioflags &= ~BIO_ERROR; 3186 bp->b_iocmd = BIO_WRITE; 3187 3188 bufobj_wref(bp->b_bufobj); 3189 curthread->td_ru.ru_oublock++; 3190 splx(s); 3191 3192 /* 3193 * Note: to avoid loopback deadlocks, we do not 3194 * assign b_runningbufspace. 3195 */ 3196 vfs_busy_pages(bp, 1); 3197 3198 BUF_KERNPROC(bp); 3199 bp->b_iooffset = dbtob(bp->b_blkno); 3200 bstrategy(bp); 3201 3202 if( (oldflags & B_ASYNC) == 0) { 3203 int rtval = bufwait(bp); 3204 3205 if (oldflags & B_DELWRI) { 3206 s = splbio(); 3207 reassignbuf(bp); 3208 splx(s); 3209 } 3210 brelse(bp); 3211 return (rtval); 3212 } 3213 3214 return (0); 3215 } 3216 3217 /* 3218 * nfs special file access vnode op. 3219 * Essentially just get vattr and then imitate iaccess() since the device is 3220 * local to the client. 3221 */ 3222 static int 3223 nfsspec_access(struct vop_access_args *ap) 3224 { 3225 struct vattr *vap; 3226 struct ucred *cred = ap->a_cred; 3227 struct vnode *vp = ap->a_vp; 3228 accmode_t accmode = ap->a_accmode; 3229 struct vattr vattr; 3230 int error; 3231 3232 /* 3233 * Disallow write attempts on filesystems mounted read-only; 3234 * unless the file is a socket, fifo, or a block or character 3235 * device resident on the filesystem. 3236 */ 3237 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3238 switch (vp->v_type) { 3239 case VREG: 3240 case VDIR: 3241 case VLNK: 3242 return (EROFS); 3243 default: 3244 break; 3245 } 3246 } 3247 vap = &vattr; 3248 error = VOP_GETATTR(vp, vap, cred); 3249 if (error) 3250 goto out; 3251 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid, 3252 accmode, cred, NULL); 3253 out: 3254 return error; 3255 } 3256 3257 /* 3258 * Read wrapper for fifos. 3259 */ 3260 static int 3261 nfsfifo_read(struct vop_read_args *ap) 3262 { 3263 struct nfsnode *np = VTONFS(ap->a_vp); 3264 int error; 3265 3266 /* 3267 * Set access flag. 3268 */ 3269 mtx_lock(&np->n_mtx); 3270 np->n_flag |= NACC; 3271 vfs_timestamp(&np->n_atim); 3272 mtx_unlock(&np->n_mtx); 3273 error = fifo_specops.vop_read(ap); 3274 return error; 3275 } 3276 3277 /* 3278 * Write wrapper for fifos. 3279 */ 3280 static int 3281 nfsfifo_write(struct vop_write_args *ap) 3282 { 3283 struct nfsnode *np = VTONFS(ap->a_vp); 3284 3285 /* 3286 * Set update flag. 3287 */ 3288 mtx_lock(&np->n_mtx); 3289 np->n_flag |= NUPD; 3290 vfs_timestamp(&np->n_mtim); 3291 mtx_unlock(&np->n_mtx); 3292 return(fifo_specops.vop_write(ap)); 3293 } 3294 3295 /* 3296 * Close wrapper for fifos. 3297 * 3298 * Update the times on the nfsnode then do fifo close. 3299 */ 3300 static int 3301 nfsfifo_close(struct vop_close_args *ap) 3302 { 3303 struct vnode *vp = ap->a_vp; 3304 struct nfsnode *np = VTONFS(vp); 3305 struct vattr vattr; 3306 struct timespec ts; 3307 3308 mtx_lock(&np->n_mtx); 3309 if (np->n_flag & (NACC | NUPD)) { 3310 vfs_timestamp(&ts); 3311 if (np->n_flag & NACC) 3312 np->n_atim = ts; 3313 if (np->n_flag & NUPD) 3314 np->n_mtim = ts; 3315 np->n_flag |= NCHG; 3316 if (vrefcnt(vp) == 1 && 3317 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3318 VATTR_NULL(&vattr); 3319 if (np->n_flag & NACC) 3320 vattr.va_atime = np->n_atim; 3321 if (np->n_flag & NUPD) 3322 vattr.va_mtime = np->n_mtim; 3323 mtx_unlock(&np->n_mtx); 3324 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3325 goto out; 3326 } 3327 } 3328 mtx_unlock(&np->n_mtx); 3329 out: 3330 return (fifo_specops.vop_close(ap)); 3331 } 3332 3333 /* 3334 * Just call ncl_writebp() with the force argument set to 1. 3335 * 3336 * NOTE: B_DONE may or may not be set in a_bp on call. 3337 */ 3338 static int 3339 nfs_bwrite(struct buf *bp) 3340 { 3341 3342 return (ncl_writebp(bp, 1, curthread)); 3343 } 3344 3345 struct buf_ops buf_ops_newnfs = { 3346 .bop_name = "buf_ops_nfs", 3347 .bop_write = nfs_bwrite, 3348 .bop_strategy = bufstrategy, 3349 .bop_sync = bufsync, 3350 .bop_bdflush = bufbdflush, 3351 }; 3352 3353 /* 3354 * Cloned from vop_stdlock(), and then the ugly hack added. 3355 */ 3356 static int 3357 nfs_lock1(struct vop_lock1_args *ap) 3358 { 3359 struct vnode *vp = ap->a_vp; 3360 int error = 0; 3361 3362 /* 3363 * Since vfs_hash_get() calls vget() and it will no longer work 3364 * for FreeBSD8 with flags == 0, I can only think of this horrible 3365 * hack to work around it. I call vfs_hash_get() with LK_EXCLOTHER 3366 * and then handle it here. All I want for this case is a v_usecount 3367 * on the vnode to use for recovery, while another thread might 3368 * hold a lock on the vnode. I have the other threads blocked, so 3369 * there isn't any race problem. 3370 */ 3371 if ((ap->a_flags & LK_TYPE_MASK) == LK_EXCLOTHER) { 3372 if ((ap->a_flags & LK_INTERLOCK) == 0) 3373 panic("ncllock1"); 3374 if ((vp->v_iflag & VI_DOOMED)) 3375 error = ENOENT; 3376 VI_UNLOCK(vp); 3377 return (error); 3378 } 3379 return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp), 3380 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file, 3381 ap->a_line)); 3382 } 3383 3384 static int 3385 nfs_getacl(struct vop_getacl_args *ap) 3386 { 3387 int error; 3388 3389 if (ap->a_type != ACL_TYPE_NFS4) 3390 return (EOPNOTSUPP); 3391 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3392 NULL); 3393 if (error > NFSERR_STALE) { 3394 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3395 error = EPERM; 3396 } 3397 return (error); 3398 } 3399 3400 static int 3401 nfs_setacl(struct vop_setacl_args *ap) 3402 { 3403 int error; 3404 3405 if (ap->a_type != ACL_TYPE_NFS4) 3406 return (EOPNOTSUPP); 3407 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3408 NULL); 3409 if (error > NFSERR_STALE) { 3410 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3411 error = EPERM; 3412 } 3413 return (error); 3414 } 3415 3416 /* 3417 * Return POSIX pathconf information applicable to nfs filesystems. 3418 */ 3419 static int 3420 nfs_pathconf(struct vop_pathconf_args *ap) 3421 { 3422 struct nfsv3_pathconf pc; 3423 struct nfsvattr nfsva; 3424 struct vnode *vp = ap->a_vp; 3425 struct thread *td = curthread; 3426 int attrflag, error; 3427 3428 if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX || 3429 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED || 3430 ap->a_name == _PC_NO_TRUNC)) || 3431 (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) { 3432 /* 3433 * Since only the above 4 a_names are returned by the NFSv3 3434 * Pathconf RPC, there is no point in doing it for others. 3435 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can 3436 * be used for _PC_NFS4_ACL as well. 3437 */ 3438 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva, 3439 &attrflag, NULL); 3440 if (attrflag != 0) 3441 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 3442 1); 3443 if (error != 0) 3444 return (error); 3445 } else { 3446 /* 3447 * For NFSv2 (or NFSv3 when not one of the above 4 a_names), 3448 * just fake them. 3449 */ 3450 pc.pc_linkmax = LINK_MAX; 3451 pc.pc_namemax = NFS_MAXNAMLEN; 3452 pc.pc_notrunc = 1; 3453 pc.pc_chownrestricted = 1; 3454 pc.pc_caseinsensitive = 0; 3455 pc.pc_casepreserving = 1; 3456 error = 0; 3457 } 3458 switch (ap->a_name) { 3459 case _PC_LINK_MAX: 3460 *ap->a_retval = pc.pc_linkmax; 3461 break; 3462 case _PC_NAME_MAX: 3463 *ap->a_retval = pc.pc_namemax; 3464 break; 3465 case _PC_PATH_MAX: 3466 *ap->a_retval = PATH_MAX; 3467 break; 3468 case _PC_PIPE_BUF: 3469 *ap->a_retval = PIPE_BUF; 3470 break; 3471 case _PC_CHOWN_RESTRICTED: 3472 *ap->a_retval = pc.pc_chownrestricted; 3473 break; 3474 case _PC_NO_TRUNC: 3475 *ap->a_retval = pc.pc_notrunc; 3476 break; 3477 case _PC_ACL_EXTENDED: 3478 *ap->a_retval = 0; 3479 break; 3480 case _PC_ACL_NFS4: 3481 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 && 3482 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL)) 3483 *ap->a_retval = 1; 3484 else 3485 *ap->a_retval = 0; 3486 break; 3487 case _PC_ACL_PATH_MAX: 3488 if (NFS_ISV4(vp)) 3489 *ap->a_retval = ACL_MAX_ENTRIES; 3490 else 3491 *ap->a_retval = 3; 3492 break; 3493 case _PC_MAC_PRESENT: 3494 *ap->a_retval = 0; 3495 break; 3496 case _PC_ASYNC_IO: 3497 /* _PC_ASYNC_IO should have been handled by upper layers. */ 3498 KASSERT(0, ("_PC_ASYNC_IO should not get here")); 3499 error = EINVAL; 3500 break; 3501 case _PC_PRIO_IO: 3502 *ap->a_retval = 0; 3503 break; 3504 case _PC_SYNC_IO: 3505 *ap->a_retval = 0; 3506 break; 3507 case _PC_ALLOC_SIZE_MIN: 3508 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize; 3509 break; 3510 case _PC_FILESIZEBITS: 3511 if (NFS_ISV34(vp)) 3512 *ap->a_retval = 64; 3513 else 3514 *ap->a_retval = 32; 3515 break; 3516 case _PC_REC_INCR_XFER_SIZE: 3517 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 3518 break; 3519 case _PC_REC_MAX_XFER_SIZE: 3520 *ap->a_retval = -1; /* means ``unlimited'' */ 3521 break; 3522 case _PC_REC_MIN_XFER_SIZE: 3523 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 3524 break; 3525 case _PC_REC_XFER_ALIGN: 3526 *ap->a_retval = PAGE_SIZE; 3527 break; 3528 case _PC_SYMLINK_MAX: 3529 *ap->a_retval = NFS_MAXPATHLEN; 3530 break; 3531 3532 default: 3533 error = EINVAL; 3534 break; 3535 } 3536 return (error); 3537 } 3538 3539