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 if (vp->v_mount != tdvp->v_mount) { 1980 return (EXDEV); 1981 } 1982 1983 /* 1984 * Push all writes to the server, so that the attribute cache 1985 * doesn't get "out of sync" with the server. 1986 * XXX There should be a better way! 1987 */ 1988 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread); 1989 1990 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 1991 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag, 1992 &dattrflag, NULL); 1993 tdnp = VTONFS(tdvp); 1994 mtx_lock(&tdnp->n_mtx); 1995 tdnp->n_flag |= NMODIFIED; 1996 if (dattrflag != 0) { 1997 mtx_unlock(&tdnp->n_mtx); 1998 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1); 1999 } else { 2000 tdnp->n_attrstamp = 0; 2001 mtx_unlock(&tdnp->n_mtx); 2002 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); 2003 } 2004 if (attrflag) 2005 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2006 else { 2007 np = VTONFS(vp); 2008 mtx_lock(&np->n_mtx); 2009 np->n_attrstamp = 0; 2010 mtx_unlock(&np->n_mtx); 2011 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 2012 } 2013 /* 2014 * If negative lookup caching is enabled, I might as well 2015 * add an entry for this node. Not necessary for correctness, 2016 * but if negative caching is enabled, then the system 2017 * must care about lookup caching hit rate, so... 2018 */ 2019 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 && 2020 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 2021 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL); 2022 } 2023 if (error && NFS_ISV4(vp)) 2024 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, 2025 (gid_t)0); 2026 return (error); 2027 } 2028 2029 /* 2030 * nfs symbolic link create call 2031 */ 2032 static int 2033 nfs_symlink(struct vop_symlink_args *ap) 2034 { 2035 struct vnode *dvp = ap->a_dvp; 2036 struct vattr *vap = ap->a_vap; 2037 struct componentname *cnp = ap->a_cnp; 2038 struct nfsvattr nfsva, dnfsva; 2039 struct nfsfh *nfhp; 2040 struct nfsnode *np = NULL, *dnp; 2041 struct vnode *newvp = NULL; 2042 int error = 0, attrflag, dattrflag, ret; 2043 2044 vap->va_type = VLNK; 2045 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2046 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, 2047 &nfsva, &nfhp, &attrflag, &dattrflag, NULL); 2048 if (nfhp) { 2049 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, 2050 &np, NULL, LK_EXCLUSIVE); 2051 if (!ret) 2052 newvp = NFSTOV(np); 2053 else if (!error) 2054 error = ret; 2055 } 2056 if (newvp != NULL) { 2057 if (attrflag) 2058 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 2059 0, 1); 2060 } else if (!error) { 2061 /* 2062 * If we do not have an error and we could not extract the 2063 * newvp from the response due to the request being NFSv2, we 2064 * have to do a lookup in order to obtain a newvp to return. 2065 */ 2066 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2067 cnp->cn_cred, cnp->cn_thread, &np); 2068 if (!error) 2069 newvp = NFSTOV(np); 2070 } 2071 if (error) { 2072 if (newvp) 2073 vput(newvp); 2074 if (NFS_ISV4(dvp)) 2075 error = nfscl_maperr(cnp->cn_thread, error, 2076 vap->va_uid, vap->va_gid); 2077 } else { 2078 *ap->a_vpp = newvp; 2079 } 2080 2081 dnp = VTONFS(dvp); 2082 mtx_lock(&dnp->n_mtx); 2083 dnp->n_flag |= NMODIFIED; 2084 if (dattrflag != 0) { 2085 mtx_unlock(&dnp->n_mtx); 2086 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2087 } else { 2088 dnp->n_attrstamp = 0; 2089 mtx_unlock(&dnp->n_mtx); 2090 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2091 } 2092 /* 2093 * If negative lookup caching is enabled, I might as well 2094 * add an entry for this node. Not necessary for correctness, 2095 * but if negative caching is enabled, then the system 2096 * must care about lookup caching hit rate, so... 2097 */ 2098 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2099 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { 2100 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL); 2101 } 2102 return (error); 2103 } 2104 2105 /* 2106 * nfs make dir call 2107 */ 2108 static int 2109 nfs_mkdir(struct vop_mkdir_args *ap) 2110 { 2111 struct vnode *dvp = ap->a_dvp; 2112 struct vattr *vap = ap->a_vap; 2113 struct componentname *cnp = ap->a_cnp; 2114 struct nfsnode *np = NULL, *dnp; 2115 struct vnode *newvp = NULL; 2116 struct vattr vattr; 2117 struct nfsfh *nfhp; 2118 struct nfsvattr nfsva, dnfsva; 2119 int error = 0, attrflag, dattrflag, ret; 2120 2121 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) 2122 return (error); 2123 vap->va_type = VDIR; 2124 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2125 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp, 2126 &attrflag, &dattrflag, NULL); 2127 dnp = VTONFS(dvp); 2128 mtx_lock(&dnp->n_mtx); 2129 dnp->n_flag |= NMODIFIED; 2130 if (dattrflag != 0) { 2131 mtx_unlock(&dnp->n_mtx); 2132 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2133 } else { 2134 dnp->n_attrstamp = 0; 2135 mtx_unlock(&dnp->n_mtx); 2136 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2137 } 2138 if (nfhp) { 2139 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, 2140 &np, NULL, LK_EXCLUSIVE); 2141 if (!ret) { 2142 newvp = NFSTOV(np); 2143 if (attrflag) 2144 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 2145 NULL, 0, 1); 2146 } else if (!error) 2147 error = ret; 2148 } 2149 if (!error && newvp == NULL) { 2150 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2151 cnp->cn_cred, cnp->cn_thread, &np); 2152 if (!error) { 2153 newvp = NFSTOV(np); 2154 if (newvp->v_type != VDIR) 2155 error = EEXIST; 2156 } 2157 } 2158 if (error) { 2159 if (newvp) 2160 vput(newvp); 2161 if (NFS_ISV4(dvp)) 2162 error = nfscl_maperr(cnp->cn_thread, error, 2163 vap->va_uid, vap->va_gid); 2164 } else { 2165 /* 2166 * If negative lookup caching is enabled, I might as well 2167 * add an entry for this node. Not necessary for correctness, 2168 * but if negative caching is enabled, then the system 2169 * must care about lookup caching hit rate, so... 2170 */ 2171 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && 2172 (cnp->cn_flags & MAKEENTRY) && 2173 attrflag != 0 && dattrflag != 0) 2174 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, 2175 &dnfsva.na_ctime); 2176 *ap->a_vpp = newvp; 2177 } 2178 return (error); 2179 } 2180 2181 /* 2182 * nfs remove directory call 2183 */ 2184 static int 2185 nfs_rmdir(struct vop_rmdir_args *ap) 2186 { 2187 struct vnode *vp = ap->a_vp; 2188 struct vnode *dvp = ap->a_dvp; 2189 struct componentname *cnp = ap->a_cnp; 2190 struct nfsnode *dnp; 2191 struct nfsvattr dnfsva; 2192 int error, dattrflag; 2193 2194 if (dvp == vp) 2195 return (EINVAL); 2196 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2197 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL); 2198 dnp = VTONFS(dvp); 2199 mtx_lock(&dnp->n_mtx); 2200 dnp->n_flag |= NMODIFIED; 2201 if (dattrflag != 0) { 2202 mtx_unlock(&dnp->n_mtx); 2203 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2204 } else { 2205 dnp->n_attrstamp = 0; 2206 mtx_unlock(&dnp->n_mtx); 2207 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); 2208 } 2209 2210 cache_purge(dvp); 2211 cache_purge(vp); 2212 if (error && NFS_ISV4(dvp)) 2213 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, 2214 (gid_t)0); 2215 /* 2216 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2217 */ 2218 if (error == ENOENT) 2219 error = 0; 2220 return (error); 2221 } 2222 2223 /* 2224 * nfs readdir call 2225 */ 2226 static int 2227 nfs_readdir(struct vop_readdir_args *ap) 2228 { 2229 struct vnode *vp = ap->a_vp; 2230 struct nfsnode *np = VTONFS(vp); 2231 struct uio *uio = ap->a_uio; 2232 ssize_t tresid; 2233 int error = 0; 2234 struct vattr vattr; 2235 2236 if (ap->a_eofflag != NULL) 2237 *ap->a_eofflag = 0; 2238 if (vp->v_type != VDIR) 2239 return(EPERM); 2240 2241 /* 2242 * First, check for hit on the EOF offset cache 2243 */ 2244 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2245 (np->n_flag & NMODIFIED) == 0) { 2246 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) { 2247 mtx_lock(&np->n_mtx); 2248 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) || 2249 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { 2250 mtx_unlock(&np->n_mtx); 2251 NFSINCRGLOBAL(newnfsstats.direofcache_hits); 2252 if (ap->a_eofflag != NULL) 2253 *ap->a_eofflag = 1; 2254 return (0); 2255 } else 2256 mtx_unlock(&np->n_mtx); 2257 } 2258 } 2259 2260 /* 2261 * Call ncl_bioread() to do the real work. 2262 */ 2263 tresid = uio->uio_resid; 2264 error = ncl_bioread(vp, uio, 0, ap->a_cred); 2265 2266 if (!error && uio->uio_resid == tresid) { 2267 NFSINCRGLOBAL(newnfsstats.direofcache_misses); 2268 if (ap->a_eofflag != NULL) 2269 *ap->a_eofflag = 1; 2270 } 2271 return (error); 2272 } 2273 2274 /* 2275 * Readdir rpc call. 2276 * Called from below the buffer cache by ncl_doio(). 2277 */ 2278 int 2279 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2280 struct thread *td) 2281 { 2282 struct nfsvattr nfsva; 2283 nfsuint64 *cookiep, cookie; 2284 struct nfsnode *dnp = VTONFS(vp); 2285 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2286 int error = 0, eof, attrflag; 2287 2288 KASSERT(uiop->uio_iovcnt == 1 && 2289 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2290 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2291 ("nfs readdirrpc bad uio")); 2292 2293 /* 2294 * If there is no cookie, assume directory was stale. 2295 */ 2296 ncl_dircookie_lock(dnp); 2297 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2298 if (cookiep) { 2299 cookie = *cookiep; 2300 ncl_dircookie_unlock(dnp); 2301 } else { 2302 ncl_dircookie_unlock(dnp); 2303 return (NFSERR_BAD_COOKIE); 2304 } 2305 2306 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2307 (void)ncl_fsinfo(nmp, vp, cred, td); 2308 2309 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva, 2310 &attrflag, &eof, NULL); 2311 if (attrflag) 2312 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2313 2314 if (!error) { 2315 /* 2316 * We are now either at the end of the directory or have filled 2317 * the block. 2318 */ 2319 if (eof) 2320 dnp->n_direofoffset = uiop->uio_offset; 2321 else { 2322 if (uiop->uio_resid > 0) 2323 ncl_printf("EEK! readdirrpc resid > 0\n"); 2324 ncl_dircookie_lock(dnp); 2325 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2326 *cookiep = cookie; 2327 ncl_dircookie_unlock(dnp); 2328 } 2329 } else if (NFS_ISV4(vp)) { 2330 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2331 } 2332 return (error); 2333 } 2334 2335 /* 2336 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc(). 2337 */ 2338 int 2339 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, 2340 struct thread *td) 2341 { 2342 struct nfsvattr nfsva; 2343 nfsuint64 *cookiep, cookie; 2344 struct nfsnode *dnp = VTONFS(vp); 2345 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2346 int error = 0, attrflag, eof; 2347 2348 KASSERT(uiop->uio_iovcnt == 1 && 2349 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && 2350 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, 2351 ("nfs readdirplusrpc bad uio")); 2352 2353 /* 2354 * If there is no cookie, assume directory was stale. 2355 */ 2356 ncl_dircookie_lock(dnp); 2357 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); 2358 if (cookiep) { 2359 cookie = *cookiep; 2360 ncl_dircookie_unlock(dnp); 2361 } else { 2362 ncl_dircookie_unlock(dnp); 2363 return (NFSERR_BAD_COOKIE); 2364 } 2365 2366 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) 2367 (void)ncl_fsinfo(nmp, vp, cred, td); 2368 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva, 2369 &attrflag, &eof, NULL); 2370 if (attrflag) 2371 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 2372 2373 if (!error) { 2374 /* 2375 * We are now either at end of the directory or have filled the 2376 * the block. 2377 */ 2378 if (eof) 2379 dnp->n_direofoffset = uiop->uio_offset; 2380 else { 2381 if (uiop->uio_resid > 0) 2382 ncl_printf("EEK! readdirplusrpc resid > 0\n"); 2383 ncl_dircookie_lock(dnp); 2384 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); 2385 *cookiep = cookie; 2386 ncl_dircookie_unlock(dnp); 2387 } 2388 } else if (NFS_ISV4(vp)) { 2389 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2390 } 2391 return (error); 2392 } 2393 2394 /* 2395 * Silly rename. To make the NFS filesystem that is stateless look a little 2396 * more like the "ufs" a remove of an active vnode is translated to a rename 2397 * to a funny looking filename that is removed by nfs_inactive on the 2398 * nfsnode. There is the potential for another process on a different client 2399 * to create the same funny name between the nfs_lookitup() fails and the 2400 * nfs_rename() completes, but... 2401 */ 2402 static int 2403 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2404 { 2405 struct sillyrename *sp; 2406 struct nfsnode *np; 2407 int error; 2408 short pid; 2409 unsigned int lticks; 2410 2411 cache_purge(dvp); 2412 np = VTONFS(vp); 2413 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir")); 2414 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2415 M_NEWNFSREQ, M_WAITOK); 2416 sp->s_cred = crhold(cnp->cn_cred); 2417 sp->s_dvp = dvp; 2418 VREF(dvp); 2419 2420 /* 2421 * Fudge together a funny name. 2422 * Changing the format of the funny name to accomodate more 2423 * sillynames per directory. 2424 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is 2425 * CPU ticks since boot. 2426 */ 2427 pid = cnp->cn_thread->td_proc->p_pid; 2428 lticks = (unsigned int)ticks; 2429 for ( ; ; ) { 2430 sp->s_namlen = sprintf(sp->s_name, 2431 ".nfs.%08x.%04x4.4", lticks, 2432 pid); 2433 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2434 cnp->cn_thread, NULL)) 2435 break; 2436 lticks++; 2437 } 2438 error = nfs_renameit(dvp, vp, cnp, sp); 2439 if (error) 2440 goto bad; 2441 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2442 cnp->cn_thread, &np); 2443 np->n_sillyrename = sp; 2444 return (0); 2445 bad: 2446 vrele(sp->s_dvp); 2447 crfree(sp->s_cred); 2448 free((caddr_t)sp, M_NEWNFSREQ); 2449 return (error); 2450 } 2451 2452 /* 2453 * Look up a file name and optionally either update the file handle or 2454 * allocate an nfsnode, depending on the value of npp. 2455 * npp == NULL --> just do the lookup 2456 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2457 * handled too 2458 * *npp != NULL --> update the file handle in the vnode 2459 */ 2460 static int 2461 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred, 2462 struct thread *td, struct nfsnode **npp) 2463 { 2464 struct vnode *newvp = NULL, *vp; 2465 struct nfsnode *np, *dnp = VTONFS(dvp); 2466 struct nfsfh *nfhp, *onfhp; 2467 struct nfsvattr nfsva, dnfsva; 2468 struct componentname cn; 2469 int error = 0, attrflag, dattrflag; 2470 u_int hash; 2471 2472 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva, 2473 &nfhp, &attrflag, &dattrflag, NULL); 2474 if (dattrflag) 2475 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); 2476 if (npp && !error) { 2477 if (*npp != NULL) { 2478 np = *npp; 2479 vp = NFSTOV(np); 2480 /* 2481 * For NFSv4, check to see if it is the same name and 2482 * replace the name, if it is different. 2483 */ 2484 if (np->n_v4 != NULL && nfsva.na_type == VREG && 2485 (np->n_v4->n4_namelen != len || 2486 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) || 2487 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || 2488 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2489 dnp->n_fhp->nfh_len))) { 2490 #ifdef notdef 2491 { char nnn[100]; int nnnl; 2492 nnnl = (len < 100) ? len : 99; 2493 bcopy(name, nnn, nnnl); 2494 nnn[nnnl] = '\0'; 2495 printf("replace=%s\n",nnn); 2496 } 2497 #endif 2498 FREE((caddr_t)np->n_v4, M_NFSV4NODE); 2499 MALLOC(np->n_v4, struct nfsv4node *, 2500 sizeof (struct nfsv4node) + 2501 dnp->n_fhp->nfh_len + len - 1, 2502 M_NFSV4NODE, M_WAITOK); 2503 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; 2504 np->n_v4->n4_namelen = len; 2505 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 2506 dnp->n_fhp->nfh_len); 2507 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len); 2508 } 2509 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, 2510 FNV1_32_INIT); 2511 onfhp = np->n_fhp; 2512 /* 2513 * Rehash node for new file handle. 2514 */ 2515 vfs_hash_rehash(vp, hash); 2516 np->n_fhp = nfhp; 2517 if (onfhp != NULL) 2518 FREE((caddr_t)onfhp, M_NFSFH); 2519 newvp = NFSTOV(np); 2520 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) { 2521 FREE((caddr_t)nfhp, M_NFSFH); 2522 VREF(dvp); 2523 newvp = dvp; 2524 } else { 2525 cn.cn_nameptr = name; 2526 cn.cn_namelen = len; 2527 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td, 2528 &np, NULL, LK_EXCLUSIVE); 2529 if (error) 2530 return (error); 2531 newvp = NFSTOV(np); 2532 } 2533 if (!attrflag && *npp == NULL) { 2534 if (newvp == dvp) 2535 vrele(newvp); 2536 else 2537 vput(newvp); 2538 return (ENOENT); 2539 } 2540 if (attrflag) 2541 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 2542 0, 1); 2543 } 2544 if (npp && *npp == NULL) { 2545 if (error) { 2546 if (newvp) { 2547 if (newvp == dvp) 2548 vrele(newvp); 2549 else 2550 vput(newvp); 2551 } 2552 } else 2553 *npp = np; 2554 } 2555 if (error && NFS_ISV4(dvp)) 2556 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2557 return (error); 2558 } 2559 2560 /* 2561 * Nfs Version 3 and 4 commit rpc 2562 */ 2563 int 2564 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred, 2565 struct thread *td) 2566 { 2567 struct nfsvattr nfsva; 2568 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2569 int error, attrflag; 2570 2571 mtx_lock(&nmp->nm_mtx); 2572 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) { 2573 mtx_unlock(&nmp->nm_mtx); 2574 return (0); 2575 } 2576 mtx_unlock(&nmp->nm_mtx); 2577 error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva, 2578 &attrflag, NULL); 2579 if (attrflag != 0) 2580 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 2581 0, 1); 2582 if (error != 0 && NFS_ISV4(vp)) 2583 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 2584 return (error); 2585 } 2586 2587 /* 2588 * Strategy routine. 2589 * For async requests when nfsiod(s) are running, queue the request by 2590 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the 2591 * request. 2592 */ 2593 static int 2594 nfs_strategy(struct vop_strategy_args *ap) 2595 { 2596 struct buf *bp = ap->a_bp; 2597 struct ucred *cr; 2598 2599 KASSERT(!(bp->b_flags & B_DONE), 2600 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); 2601 BUF_ASSERT_HELD(bp); 2602 2603 if (bp->b_iocmd == BIO_READ) 2604 cr = bp->b_rcred; 2605 else 2606 cr = bp->b_wcred; 2607 2608 /* 2609 * If the op is asynchronous and an i/o daemon is waiting 2610 * queue the request, wake it up and wait for completion 2611 * otherwise just do it ourselves. 2612 */ 2613 if ((bp->b_flags & B_ASYNC) == 0 || 2614 ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread)) 2615 (void) ncl_doio(ap->a_vp, bp, cr, curthread, 1); 2616 return (0); 2617 } 2618 2619 /* 2620 * fsync vnode op. Just call ncl_flush() with commit == 1. 2621 */ 2622 /* ARGSUSED */ 2623 static int 2624 nfs_fsync(struct vop_fsync_args *ap) 2625 { 2626 2627 if (ap->a_vp->v_type != VREG) { 2628 /* 2629 * For NFS, metadata is changed synchronously on the server, 2630 * so there is nothing to flush. Also, ncl_flush() clears 2631 * the NMODIFIED flag and that shouldn't be done here for 2632 * directories. 2633 */ 2634 return (0); 2635 } 2636 return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1, 0)); 2637 } 2638 2639 /* 2640 * Flush all the blocks associated with a vnode. 2641 * Walk through the buffer pool and push any dirty pages 2642 * associated with the vnode. 2643 * If the called_from_renewthread argument is TRUE, it has been called 2644 * from the NFSv4 renew thread and, as such, cannot block indefinitely 2645 * waiting for a buffer write to complete. 2646 */ 2647 int 2648 ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td, 2649 int commit, int called_from_renewthread) 2650 { 2651 struct nfsnode *np = VTONFS(vp); 2652 struct buf *bp; 2653 int i; 2654 struct buf *nbp; 2655 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2656 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos; 2657 int passone = 1, trycnt = 0; 2658 u_quad_t off, endoff, toff; 2659 struct ucred* wcred = NULL; 2660 struct buf **bvec = NULL; 2661 struct bufobj *bo; 2662 #ifndef NFS_COMMITBVECSIZ 2663 #define NFS_COMMITBVECSIZ 20 2664 #endif 2665 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ]; 2666 int bvecsize = 0, bveccount; 2667 2668 if (called_from_renewthread != 0) 2669 slptimeo = hz; 2670 if (nmp->nm_flag & NFSMNT_INT) 2671 slpflag = PCATCH; 2672 if (!commit) 2673 passone = 0; 2674 bo = &vp->v_bufobj; 2675 /* 2676 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the 2677 * server, but has not been committed to stable storage on the server 2678 * yet. On the first pass, the byte range is worked out and the commit 2679 * rpc is done. On the second pass, ncl_writebp() is called to do the 2680 * job. 2681 */ 2682 again: 2683 off = (u_quad_t)-1; 2684 endoff = 0; 2685 bvecpos = 0; 2686 if (NFS_ISV34(vp) && commit) { 2687 if (bvec != NULL && bvec != bvec_on_stack) 2688 free(bvec, M_TEMP); 2689 /* 2690 * Count up how many buffers waiting for a commit. 2691 */ 2692 bveccount = 0; 2693 BO_LOCK(bo); 2694 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2695 if (!BUF_ISLOCKED(bp) && 2696 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) 2697 == (B_DELWRI | B_NEEDCOMMIT)) 2698 bveccount++; 2699 } 2700 /* 2701 * Allocate space to remember the list of bufs to commit. It is 2702 * important to use M_NOWAIT here to avoid a race with nfs_write. 2703 * If we can't get memory (for whatever reason), we will end up 2704 * committing the buffers one-by-one in the loop below. 2705 */ 2706 if (bveccount > NFS_COMMITBVECSIZ) { 2707 /* 2708 * Release the vnode interlock to avoid a lock 2709 * order reversal. 2710 */ 2711 BO_UNLOCK(bo); 2712 bvec = (struct buf **) 2713 malloc(bveccount * sizeof(struct buf *), 2714 M_TEMP, M_NOWAIT); 2715 BO_LOCK(bo); 2716 if (bvec == NULL) { 2717 bvec = bvec_on_stack; 2718 bvecsize = NFS_COMMITBVECSIZ; 2719 } else 2720 bvecsize = bveccount; 2721 } else { 2722 bvec = bvec_on_stack; 2723 bvecsize = NFS_COMMITBVECSIZ; 2724 } 2725 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2726 if (bvecpos >= bvecsize) 2727 break; 2728 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2729 nbp = TAILQ_NEXT(bp, b_bobufs); 2730 continue; 2731 } 2732 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 2733 (B_DELWRI | B_NEEDCOMMIT)) { 2734 BUF_UNLOCK(bp); 2735 nbp = TAILQ_NEXT(bp, b_bobufs); 2736 continue; 2737 } 2738 BO_UNLOCK(bo); 2739 bremfree(bp); 2740 /* 2741 * Work out if all buffers are using the same cred 2742 * so we can deal with them all with one commit. 2743 * 2744 * NOTE: we are not clearing B_DONE here, so we have 2745 * to do it later on in this routine if we intend to 2746 * initiate I/O on the bp. 2747 * 2748 * Note: to avoid loopback deadlocks, we do not 2749 * assign b_runningbufspace. 2750 */ 2751 if (wcred == NULL) 2752 wcred = bp->b_wcred; 2753 else if (wcred != bp->b_wcred) 2754 wcred = NOCRED; 2755 vfs_busy_pages(bp, 1); 2756 2757 BO_LOCK(bo); 2758 /* 2759 * bp is protected by being locked, but nbp is not 2760 * and vfs_busy_pages() may sleep. We have to 2761 * recalculate nbp. 2762 */ 2763 nbp = TAILQ_NEXT(bp, b_bobufs); 2764 2765 /* 2766 * A list of these buffers is kept so that the 2767 * second loop knows which buffers have actually 2768 * been committed. This is necessary, since there 2769 * may be a race between the commit rpc and new 2770 * uncommitted writes on the file. 2771 */ 2772 bvec[bvecpos++] = bp; 2773 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2774 bp->b_dirtyoff; 2775 if (toff < off) 2776 off = toff; 2777 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); 2778 if (toff > endoff) 2779 endoff = toff; 2780 } 2781 BO_UNLOCK(bo); 2782 } 2783 if (bvecpos > 0) { 2784 /* 2785 * Commit data on the server, as required. 2786 * If all bufs are using the same wcred, then use that with 2787 * one call for all of them, otherwise commit each one 2788 * separately. 2789 */ 2790 if (wcred != NOCRED) 2791 retv = ncl_commit(vp, off, (int)(endoff - off), 2792 wcred, td); 2793 else { 2794 retv = 0; 2795 for (i = 0; i < bvecpos; i++) { 2796 off_t off, size; 2797 bp = bvec[i]; 2798 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2799 bp->b_dirtyoff; 2800 size = (u_quad_t)(bp->b_dirtyend 2801 - bp->b_dirtyoff); 2802 retv = ncl_commit(vp, off, (int)size, 2803 bp->b_wcred, td); 2804 if (retv) break; 2805 } 2806 } 2807 2808 if (retv == NFSERR_STALEWRITEVERF) 2809 ncl_clearcommit(vp->v_mount); 2810 2811 /* 2812 * Now, either mark the blocks I/O done or mark the 2813 * blocks dirty, depending on whether the commit 2814 * succeeded. 2815 */ 2816 for (i = 0; i < bvecpos; i++) { 2817 bp = bvec[i]; 2818 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 2819 if (retv) { 2820 /* 2821 * Error, leave B_DELWRI intact 2822 */ 2823 vfs_unbusy_pages(bp); 2824 brelse(bp); 2825 } else { 2826 /* 2827 * Success, remove B_DELWRI ( bundirty() ). 2828 * 2829 * b_dirtyoff/b_dirtyend seem to be NFS 2830 * specific. We should probably move that 2831 * into bundirty(). XXX 2832 */ 2833 bufobj_wref(bo); 2834 bp->b_flags |= B_ASYNC; 2835 bundirty(bp); 2836 bp->b_flags &= ~B_DONE; 2837 bp->b_ioflags &= ~BIO_ERROR; 2838 bp->b_dirtyoff = bp->b_dirtyend = 0; 2839 bufdone(bp); 2840 } 2841 } 2842 } 2843 2844 /* 2845 * Start/do any write(s) that are required. 2846 */ 2847 loop: 2848 BO_LOCK(bo); 2849 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 2850 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { 2851 if (waitfor != MNT_WAIT || passone) 2852 continue; 2853 2854 error = BUF_TIMELOCK(bp, 2855 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 2856 BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo); 2857 if (error == 0) { 2858 BUF_UNLOCK(bp); 2859 goto loop; 2860 } 2861 if (error == ENOLCK) { 2862 error = 0; 2863 goto loop; 2864 } 2865 if (called_from_renewthread != 0) { 2866 /* 2867 * Return EIO so the flush will be retried 2868 * later. 2869 */ 2870 error = EIO; 2871 goto done; 2872 } 2873 if (newnfs_sigintr(nmp, td)) { 2874 error = EINTR; 2875 goto done; 2876 } 2877 if (slpflag == PCATCH) { 2878 slpflag = 0; 2879 slptimeo = 2 * hz; 2880 } 2881 goto loop; 2882 } 2883 if ((bp->b_flags & B_DELWRI) == 0) 2884 panic("nfs_fsync: not dirty"); 2885 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) { 2886 BUF_UNLOCK(bp); 2887 continue; 2888 } 2889 BO_UNLOCK(bo); 2890 bremfree(bp); 2891 if (passone || !commit) 2892 bp->b_flags |= B_ASYNC; 2893 else 2894 bp->b_flags |= B_ASYNC; 2895 bwrite(bp); 2896 if (newnfs_sigintr(nmp, td)) { 2897 error = EINTR; 2898 goto done; 2899 } 2900 goto loop; 2901 } 2902 if (passone) { 2903 passone = 0; 2904 BO_UNLOCK(bo); 2905 goto again; 2906 } 2907 if (waitfor == MNT_WAIT) { 2908 while (bo->bo_numoutput) { 2909 error = bufobj_wwait(bo, slpflag, slptimeo); 2910 if (error) { 2911 BO_UNLOCK(bo); 2912 if (called_from_renewthread != 0) { 2913 /* 2914 * Return EIO so that the flush will be 2915 * retried later. 2916 */ 2917 error = EIO; 2918 goto done; 2919 } 2920 error = newnfs_sigintr(nmp, td); 2921 if (error) 2922 goto done; 2923 if (slpflag == PCATCH) { 2924 slpflag = 0; 2925 slptimeo = 2 * hz; 2926 } 2927 BO_LOCK(bo); 2928 } 2929 } 2930 if (bo->bo_dirty.bv_cnt != 0 && commit) { 2931 BO_UNLOCK(bo); 2932 goto loop; 2933 } 2934 /* 2935 * Wait for all the async IO requests to drain 2936 */ 2937 BO_UNLOCK(bo); 2938 mtx_lock(&np->n_mtx); 2939 while (np->n_directio_asyncwr > 0) { 2940 np->n_flag |= NFSYNCWAIT; 2941 error = newnfs_msleep(td, &np->n_directio_asyncwr, 2942 &np->n_mtx, slpflag | (PRIBIO + 1), 2943 "nfsfsync", 0); 2944 if (error) { 2945 if (newnfs_sigintr(nmp, td)) { 2946 mtx_unlock(&np->n_mtx); 2947 error = EINTR; 2948 goto done; 2949 } 2950 } 2951 } 2952 mtx_unlock(&np->n_mtx); 2953 } else 2954 BO_UNLOCK(bo); 2955 if (NFSHASPNFS(nmp)) { 2956 nfscl_layoutcommit(vp, td); 2957 /* 2958 * Invalidate the attribute cache, since writes to a DS 2959 * won't update the size attribute. 2960 */ 2961 mtx_lock(&np->n_mtx); 2962 np->n_attrstamp = 0; 2963 } else 2964 mtx_lock(&np->n_mtx); 2965 if (np->n_flag & NWRITEERR) { 2966 error = np->n_error; 2967 np->n_flag &= ~NWRITEERR; 2968 } 2969 if (commit && bo->bo_dirty.bv_cnt == 0 && 2970 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0) 2971 np->n_flag &= ~NMODIFIED; 2972 mtx_unlock(&np->n_mtx); 2973 done: 2974 if (bvec != NULL && bvec != bvec_on_stack) 2975 free(bvec, M_TEMP); 2976 if (error == 0 && commit != 0 && waitfor == MNT_WAIT && 2977 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 || 2978 np->n_directio_asyncwr != 0) && trycnt++ < 5) { 2979 /* try, try again... */ 2980 passone = 1; 2981 wcred = NULL; 2982 bvec = NULL; 2983 bvecsize = 0; 2984 printf("try%d\n", trycnt); 2985 goto again; 2986 } 2987 return (error); 2988 } 2989 2990 /* 2991 * NFS advisory byte-level locks. 2992 */ 2993 static int 2994 nfs_advlock(struct vop_advlock_args *ap) 2995 { 2996 struct vnode *vp = ap->a_vp; 2997 struct ucred *cred; 2998 struct nfsnode *np = VTONFS(ap->a_vp); 2999 struct proc *p = (struct proc *)ap->a_id; 3000 struct thread *td = curthread; /* XXX */ 3001 struct vattr va; 3002 int ret, error = EOPNOTSUPP; 3003 u_quad_t size; 3004 3005 if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) { 3006 if (vp->v_type != VREG) 3007 return (EINVAL); 3008 if ((ap->a_flags & F_POSIX) != 0) 3009 cred = p->p_ucred; 3010 else 3011 cred = td->td_ucred; 3012 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 3013 if (vp->v_iflag & VI_DOOMED) { 3014 NFSVOPUNLOCK(vp, 0); 3015 return (EBADF); 3016 } 3017 3018 /* 3019 * If this is unlocking a write locked region, flush and 3020 * commit them before unlocking. This is required by 3021 * RFC3530 Sec. 9.3.2. 3022 */ 3023 if (ap->a_op == F_UNLCK && 3024 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id, 3025 ap->a_flags)) 3026 (void) ncl_flush(vp, MNT_WAIT, cred, td, 1, 0); 3027 3028 /* 3029 * Loop around doing the lock op, while a blocking lock 3030 * must wait for the lock op to succeed. 3031 */ 3032 do { 3033 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op, 3034 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags); 3035 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3036 ap->a_op == F_SETLK) { 3037 NFSVOPUNLOCK(vp, 0); 3038 error = nfs_catnap(PZERO | PCATCH, ret, 3039 "ncladvl"); 3040 if (error) 3041 return (EINTR); 3042 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 3043 if (vp->v_iflag & VI_DOOMED) { 3044 NFSVOPUNLOCK(vp, 0); 3045 return (EBADF); 3046 } 3047 } 3048 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && 3049 ap->a_op == F_SETLK); 3050 if (ret == NFSERR_DENIED) { 3051 NFSVOPUNLOCK(vp, 0); 3052 return (EAGAIN); 3053 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) { 3054 NFSVOPUNLOCK(vp, 0); 3055 return (ret); 3056 } else if (ret != 0) { 3057 NFSVOPUNLOCK(vp, 0); 3058 return (EACCES); 3059 } 3060 3061 /* 3062 * Now, if we just got a lock, invalidate data in the buffer 3063 * cache, as required, so that the coherency conforms with 3064 * RFC3530 Sec. 9.3.2. 3065 */ 3066 if (ap->a_op == F_SETLK) { 3067 if ((np->n_flag & NMODIFIED) == 0) { 3068 np->n_attrstamp = 0; 3069 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3070 ret = VOP_GETATTR(vp, &va, cred); 3071 } 3072 if ((np->n_flag & NMODIFIED) || ret || 3073 np->n_change != va.va_filerev) { 3074 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1); 3075 np->n_attrstamp = 0; 3076 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 3077 ret = VOP_GETATTR(vp, &va, cred); 3078 if (!ret) { 3079 np->n_mtime = va.va_mtime; 3080 np->n_change = va.va_filerev; 3081 } 3082 } 3083 /* Mark that a file lock has been acquired. */ 3084 mtx_lock(&np->n_mtx); 3085 np->n_flag |= NHASBEENLOCKED; 3086 mtx_unlock(&np->n_mtx); 3087 } 3088 NFSVOPUNLOCK(vp, 0); 3089 return (0); 3090 } else if (!NFS_ISV4(vp)) { 3091 error = NFSVOPLOCK(vp, LK_SHARED); 3092 if (error) 3093 return (error); 3094 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3095 size = VTONFS(vp)->n_size; 3096 NFSVOPUNLOCK(vp, 0); 3097 error = lf_advlock(ap, &(vp->v_lockf), size); 3098 } else { 3099 if (nfs_advlock_p != NULL) 3100 error = nfs_advlock_p(ap); 3101 else { 3102 NFSVOPUNLOCK(vp, 0); 3103 error = ENOLCK; 3104 } 3105 } 3106 if (error == 0 && ap->a_op == F_SETLK) { 3107 /* Mark that a file lock has been acquired. */ 3108 mtx_lock(&np->n_mtx); 3109 np->n_flag |= NHASBEENLOCKED; 3110 mtx_unlock(&np->n_mtx); 3111 } 3112 } 3113 return (error); 3114 } 3115 3116 /* 3117 * NFS advisory byte-level locks. 3118 */ 3119 static int 3120 nfs_advlockasync(struct vop_advlockasync_args *ap) 3121 { 3122 struct vnode *vp = ap->a_vp; 3123 u_quad_t size; 3124 int error; 3125 3126 if (NFS_ISV4(vp)) 3127 return (EOPNOTSUPP); 3128 error = NFSVOPLOCK(vp, LK_SHARED); 3129 if (error) 3130 return (error); 3131 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { 3132 size = VTONFS(vp)->n_size; 3133 NFSVOPUNLOCK(vp, 0); 3134 error = lf_advlockasync(ap, &(vp->v_lockf), size); 3135 } else { 3136 NFSVOPUNLOCK(vp, 0); 3137 error = EOPNOTSUPP; 3138 } 3139 return (error); 3140 } 3141 3142 /* 3143 * Print out the contents of an nfsnode. 3144 */ 3145 static int 3146 nfs_print(struct vop_print_args *ap) 3147 { 3148 struct vnode *vp = ap->a_vp; 3149 struct nfsnode *np = VTONFS(vp); 3150 3151 ncl_printf("\tfileid %ld fsid 0x%x", 3152 np->n_vattr.na_fileid, np->n_vattr.na_fsid); 3153 if (vp->v_type == VFIFO) 3154 fifo_printinfo(vp); 3155 printf("\n"); 3156 return (0); 3157 } 3158 3159 /* 3160 * This is the "real" nfs::bwrite(struct buf*). 3161 * We set B_CACHE if this is a VMIO buffer. 3162 */ 3163 int 3164 ncl_writebp(struct buf *bp, int force __unused, struct thread *td) 3165 { 3166 int s; 3167 int oldflags = bp->b_flags; 3168 #if 0 3169 int retv = 1; 3170 off_t off; 3171 #endif 3172 3173 BUF_ASSERT_HELD(bp); 3174 3175 if (bp->b_flags & B_INVAL) { 3176 brelse(bp); 3177 return(0); 3178 } 3179 3180 bp->b_flags |= B_CACHE; 3181 3182 /* 3183 * Undirty the bp. We will redirty it later if the I/O fails. 3184 */ 3185 3186 s = splbio(); 3187 bundirty(bp); 3188 bp->b_flags &= ~B_DONE; 3189 bp->b_ioflags &= ~BIO_ERROR; 3190 bp->b_iocmd = BIO_WRITE; 3191 3192 bufobj_wref(bp->b_bufobj); 3193 curthread->td_ru.ru_oublock++; 3194 splx(s); 3195 3196 /* 3197 * Note: to avoid loopback deadlocks, we do not 3198 * assign b_runningbufspace. 3199 */ 3200 vfs_busy_pages(bp, 1); 3201 3202 BUF_KERNPROC(bp); 3203 bp->b_iooffset = dbtob(bp->b_blkno); 3204 bstrategy(bp); 3205 3206 if( (oldflags & B_ASYNC) == 0) { 3207 int rtval = bufwait(bp); 3208 3209 if (oldflags & B_DELWRI) { 3210 s = splbio(); 3211 reassignbuf(bp); 3212 splx(s); 3213 } 3214 brelse(bp); 3215 return (rtval); 3216 } 3217 3218 return (0); 3219 } 3220 3221 /* 3222 * nfs special file access vnode op. 3223 * Essentially just get vattr and then imitate iaccess() since the device is 3224 * local to the client. 3225 */ 3226 static int 3227 nfsspec_access(struct vop_access_args *ap) 3228 { 3229 struct vattr *vap; 3230 struct ucred *cred = ap->a_cred; 3231 struct vnode *vp = ap->a_vp; 3232 accmode_t accmode = ap->a_accmode; 3233 struct vattr vattr; 3234 int error; 3235 3236 /* 3237 * Disallow write attempts on filesystems mounted read-only; 3238 * unless the file is a socket, fifo, or a block or character 3239 * device resident on the filesystem. 3240 */ 3241 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3242 switch (vp->v_type) { 3243 case VREG: 3244 case VDIR: 3245 case VLNK: 3246 return (EROFS); 3247 default: 3248 break; 3249 } 3250 } 3251 vap = &vattr; 3252 error = VOP_GETATTR(vp, vap, cred); 3253 if (error) 3254 goto out; 3255 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid, 3256 accmode, cred, NULL); 3257 out: 3258 return error; 3259 } 3260 3261 /* 3262 * Read wrapper for fifos. 3263 */ 3264 static int 3265 nfsfifo_read(struct vop_read_args *ap) 3266 { 3267 struct nfsnode *np = VTONFS(ap->a_vp); 3268 int error; 3269 3270 /* 3271 * Set access flag. 3272 */ 3273 mtx_lock(&np->n_mtx); 3274 np->n_flag |= NACC; 3275 vfs_timestamp(&np->n_atim); 3276 mtx_unlock(&np->n_mtx); 3277 error = fifo_specops.vop_read(ap); 3278 return error; 3279 } 3280 3281 /* 3282 * Write wrapper for fifos. 3283 */ 3284 static int 3285 nfsfifo_write(struct vop_write_args *ap) 3286 { 3287 struct nfsnode *np = VTONFS(ap->a_vp); 3288 3289 /* 3290 * Set update flag. 3291 */ 3292 mtx_lock(&np->n_mtx); 3293 np->n_flag |= NUPD; 3294 vfs_timestamp(&np->n_mtim); 3295 mtx_unlock(&np->n_mtx); 3296 return(fifo_specops.vop_write(ap)); 3297 } 3298 3299 /* 3300 * Close wrapper for fifos. 3301 * 3302 * Update the times on the nfsnode then do fifo close. 3303 */ 3304 static int 3305 nfsfifo_close(struct vop_close_args *ap) 3306 { 3307 struct vnode *vp = ap->a_vp; 3308 struct nfsnode *np = VTONFS(vp); 3309 struct vattr vattr; 3310 struct timespec ts; 3311 3312 mtx_lock(&np->n_mtx); 3313 if (np->n_flag & (NACC | NUPD)) { 3314 vfs_timestamp(&ts); 3315 if (np->n_flag & NACC) 3316 np->n_atim = ts; 3317 if (np->n_flag & NUPD) 3318 np->n_mtim = ts; 3319 np->n_flag |= NCHG; 3320 if (vrefcnt(vp) == 1 && 3321 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3322 VATTR_NULL(&vattr); 3323 if (np->n_flag & NACC) 3324 vattr.va_atime = np->n_atim; 3325 if (np->n_flag & NUPD) 3326 vattr.va_mtime = np->n_mtim; 3327 mtx_unlock(&np->n_mtx); 3328 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3329 goto out; 3330 } 3331 } 3332 mtx_unlock(&np->n_mtx); 3333 out: 3334 return (fifo_specops.vop_close(ap)); 3335 } 3336 3337 /* 3338 * Just call ncl_writebp() with the force argument set to 1. 3339 * 3340 * NOTE: B_DONE may or may not be set in a_bp on call. 3341 */ 3342 static int 3343 nfs_bwrite(struct buf *bp) 3344 { 3345 3346 return (ncl_writebp(bp, 1, curthread)); 3347 } 3348 3349 struct buf_ops buf_ops_newnfs = { 3350 .bop_name = "buf_ops_nfs", 3351 .bop_write = nfs_bwrite, 3352 .bop_strategy = bufstrategy, 3353 .bop_sync = bufsync, 3354 .bop_bdflush = bufbdflush, 3355 }; 3356 3357 /* 3358 * Cloned from vop_stdlock(), and then the ugly hack added. 3359 */ 3360 static int 3361 nfs_lock1(struct vop_lock1_args *ap) 3362 { 3363 struct vnode *vp = ap->a_vp; 3364 int error = 0; 3365 3366 /* 3367 * Since vfs_hash_get() calls vget() and it will no longer work 3368 * for FreeBSD8 with flags == 0, I can only think of this horrible 3369 * hack to work around it. I call vfs_hash_get() with LK_EXCLOTHER 3370 * and then handle it here. All I want for this case is a v_usecount 3371 * on the vnode to use for recovery, while another thread might 3372 * hold a lock on the vnode. I have the other threads blocked, so 3373 * there isn't any race problem. 3374 */ 3375 if ((ap->a_flags & LK_TYPE_MASK) == LK_EXCLOTHER) { 3376 if ((ap->a_flags & LK_INTERLOCK) == 0) 3377 panic("ncllock1"); 3378 if ((vp->v_iflag & VI_DOOMED)) 3379 error = ENOENT; 3380 VI_UNLOCK(vp); 3381 return (error); 3382 } 3383 return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp), 3384 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file, 3385 ap->a_line)); 3386 } 3387 3388 static int 3389 nfs_getacl(struct vop_getacl_args *ap) 3390 { 3391 int error; 3392 3393 if (ap->a_type != ACL_TYPE_NFS4) 3394 return (EOPNOTSUPP); 3395 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3396 NULL); 3397 if (error > NFSERR_STALE) { 3398 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3399 error = EPERM; 3400 } 3401 return (error); 3402 } 3403 3404 static int 3405 nfs_setacl(struct vop_setacl_args *ap) 3406 { 3407 int error; 3408 3409 if (ap->a_type != ACL_TYPE_NFS4) 3410 return (EOPNOTSUPP); 3411 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, 3412 NULL); 3413 if (error > NFSERR_STALE) { 3414 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); 3415 error = EPERM; 3416 } 3417 return (error); 3418 } 3419 3420 /* 3421 * Return POSIX pathconf information applicable to nfs filesystems. 3422 */ 3423 static int 3424 nfs_pathconf(struct vop_pathconf_args *ap) 3425 { 3426 struct nfsv3_pathconf pc; 3427 struct nfsvattr nfsva; 3428 struct vnode *vp = ap->a_vp; 3429 struct thread *td = curthread; 3430 int attrflag, error; 3431 3432 if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX || 3433 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED || 3434 ap->a_name == _PC_NO_TRUNC)) || 3435 (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) { 3436 /* 3437 * Since only the above 4 a_names are returned by the NFSv3 3438 * Pathconf RPC, there is no point in doing it for others. 3439 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can 3440 * be used for _PC_NFS4_ACL as well. 3441 */ 3442 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva, 3443 &attrflag, NULL); 3444 if (attrflag != 0) 3445 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 3446 1); 3447 if (error != 0) 3448 return (error); 3449 } else { 3450 /* 3451 * For NFSv2 (or NFSv3 when not one of the above 4 a_names), 3452 * just fake them. 3453 */ 3454 pc.pc_linkmax = LINK_MAX; 3455 pc.pc_namemax = NFS_MAXNAMLEN; 3456 pc.pc_notrunc = 1; 3457 pc.pc_chownrestricted = 1; 3458 pc.pc_caseinsensitive = 0; 3459 pc.pc_casepreserving = 1; 3460 error = 0; 3461 } 3462 switch (ap->a_name) { 3463 case _PC_LINK_MAX: 3464 *ap->a_retval = pc.pc_linkmax; 3465 break; 3466 case _PC_NAME_MAX: 3467 *ap->a_retval = pc.pc_namemax; 3468 break; 3469 case _PC_PATH_MAX: 3470 *ap->a_retval = PATH_MAX; 3471 break; 3472 case _PC_PIPE_BUF: 3473 *ap->a_retval = PIPE_BUF; 3474 break; 3475 case _PC_CHOWN_RESTRICTED: 3476 *ap->a_retval = pc.pc_chownrestricted; 3477 break; 3478 case _PC_NO_TRUNC: 3479 *ap->a_retval = pc.pc_notrunc; 3480 break; 3481 case _PC_ACL_EXTENDED: 3482 *ap->a_retval = 0; 3483 break; 3484 case _PC_ACL_NFS4: 3485 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 && 3486 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL)) 3487 *ap->a_retval = 1; 3488 else 3489 *ap->a_retval = 0; 3490 break; 3491 case _PC_ACL_PATH_MAX: 3492 if (NFS_ISV4(vp)) 3493 *ap->a_retval = ACL_MAX_ENTRIES; 3494 else 3495 *ap->a_retval = 3; 3496 break; 3497 case _PC_MAC_PRESENT: 3498 *ap->a_retval = 0; 3499 break; 3500 case _PC_ASYNC_IO: 3501 /* _PC_ASYNC_IO should have been handled by upper layers. */ 3502 KASSERT(0, ("_PC_ASYNC_IO should not get here")); 3503 error = EINVAL; 3504 break; 3505 case _PC_PRIO_IO: 3506 *ap->a_retval = 0; 3507 break; 3508 case _PC_SYNC_IO: 3509 *ap->a_retval = 0; 3510 break; 3511 case _PC_ALLOC_SIZE_MIN: 3512 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize; 3513 break; 3514 case _PC_FILESIZEBITS: 3515 if (NFS_ISV34(vp)) 3516 *ap->a_retval = 64; 3517 else 3518 *ap->a_retval = 32; 3519 break; 3520 case _PC_REC_INCR_XFER_SIZE: 3521 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 3522 break; 3523 case _PC_REC_MAX_XFER_SIZE: 3524 *ap->a_retval = -1; /* means ``unlimited'' */ 3525 break; 3526 case _PC_REC_MIN_XFER_SIZE: 3527 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; 3528 break; 3529 case _PC_REC_XFER_ALIGN: 3530 *ap->a_retval = PAGE_SIZE; 3531 break; 3532 case _PC_SYMLINK_MAX: 3533 *ap->a_retval = NFS_MAXPATHLEN; 3534 break; 3535 3536 default: 3537 error = EINVAL; 3538 break; 3539 } 3540 return (error); 3541 } 3542 3543