1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright 2007 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 /* 27 * Copyright (c) 1983,1984,1985,1986,1987,1988,1989 AT&T. 28 * All rights reserved. 29 */ 30 31 #pragma ident "%Z%%M% %I% %E% SMI" 32 33 #include <sys/param.h> 34 #include <sys/types.h> 35 #include <sys/systm.h> 36 #include <sys/cred.h> 37 #include <sys/time.h> 38 #include <sys/vnode.h> 39 #include <sys/vfs.h> 40 #include <sys/vfs_opreg.h> 41 #include <sys/file.h> 42 #include <sys/filio.h> 43 #include <sys/uio.h> 44 #include <sys/buf.h> 45 #include <sys/mman.h> 46 #include <sys/pathname.h> 47 #include <sys/dirent.h> 48 #include <sys/debug.h> 49 #include <sys/vmsystm.h> 50 #include <sys/fcntl.h> 51 #include <sys/flock.h> 52 #include <sys/swap.h> 53 #include <sys/errno.h> 54 #include <sys/strsubr.h> 55 #include <sys/sysmacros.h> 56 #include <sys/kmem.h> 57 #include <sys/cmn_err.h> 58 #include <sys/pathconf.h> 59 #include <sys/utsname.h> 60 #include <sys/dnlc.h> 61 #include <sys/acl.h> 62 #include <sys/systeminfo.h> 63 #include <sys/atomic.h> 64 #include <sys/policy.h> 65 #include <sys/sdt.h> 66 67 #include <rpc/types.h> 68 #include <rpc/auth.h> 69 #include <rpc/clnt.h> 70 71 #include <nfs/nfs.h> 72 #include <nfs/nfs_clnt.h> 73 #include <nfs/rnode.h> 74 #include <nfs/nfs_acl.h> 75 #include <nfs/lm.h> 76 77 #include <vm/hat.h> 78 #include <vm/as.h> 79 #include <vm/page.h> 80 #include <vm/pvn.h> 81 #include <vm/seg.h> 82 #include <vm/seg_map.h> 83 #include <vm/seg_kpm.h> 84 #include <vm/seg_vn.h> 85 86 #include <fs/fs_subr.h> 87 88 #include <sys/ddi.h> 89 90 static int nfs3_rdwrlbn(vnode_t *, page_t *, u_offset_t, size_t, int, 91 cred_t *); 92 static int nfs3write(vnode_t *, caddr_t, u_offset_t, int, cred_t *, 93 stable_how *); 94 static int nfs3read(vnode_t *, caddr_t, offset_t, int, size_t *, cred_t *); 95 static int nfs3setattr(vnode_t *, struct vattr *, int, cred_t *); 96 static int nfs3_accessx(void *, int, cred_t *); 97 static int nfs3lookup_dnlc(vnode_t *, char *, vnode_t **, cred_t *); 98 static int nfs3lookup_otw(vnode_t *, char *, vnode_t **, cred_t *, int); 99 static int nfs3create(vnode_t *, char *, struct vattr *, enum vcexcl, 100 int, vnode_t **, cred_t *, int); 101 static int nfs3excl_create_settimes(vnode_t *, struct vattr *, cred_t *); 102 static int nfs3mknod(vnode_t *, char *, struct vattr *, enum vcexcl, 103 int, vnode_t **, cred_t *); 104 static int nfs3rename(vnode_t *, char *, vnode_t *, char *, cred_t *, 105 caller_context_t *); 106 static int do_nfs3readdir(vnode_t *, rddir_cache *, cred_t *); 107 static void nfs3readdir(vnode_t *, rddir_cache *, cred_t *); 108 static void nfs3readdirplus(vnode_t *, rddir_cache *, cred_t *); 109 static int nfs3_bio(struct buf *, stable_how *, cred_t *); 110 static int nfs3_getapage(vnode_t *, u_offset_t, size_t, uint_t *, 111 page_t *[], size_t, struct seg *, caddr_t, 112 enum seg_rw, cred_t *); 113 static void nfs3_readahead(vnode_t *, u_offset_t, caddr_t, struct seg *, 114 cred_t *); 115 static int nfs3_sync_putapage(vnode_t *, page_t *, u_offset_t, size_t, 116 int, cred_t *); 117 static int nfs3_sync_pageio(vnode_t *, page_t *, u_offset_t, size_t, 118 int, cred_t *); 119 static int nfs3_commit(vnode_t *, offset3, count3, cred_t *); 120 static void nfs3_set_mod(vnode_t *); 121 static void nfs3_get_commit(vnode_t *); 122 static void nfs3_get_commit_range(vnode_t *, u_offset_t, size_t); 123 #if 0 /* unused */ 124 #ifdef DEBUG 125 static int nfs3_no_uncommitted_pages(vnode_t *); 126 #endif 127 #endif /* unused */ 128 static int nfs3_putpage_commit(vnode_t *, offset_t, size_t, cred_t *); 129 static int nfs3_commit_vp(vnode_t *, u_offset_t, size_t, cred_t *); 130 static int nfs3_sync_commit(vnode_t *, page_t *, offset3, count3, 131 cred_t *); 132 static void nfs3_async_commit(vnode_t *, page_t *, offset3, count3, 133 cred_t *); 134 static void nfs3_delmap_callback(struct as *, void *, uint_t); 135 136 /* 137 * Error flags used to pass information about certain special errors 138 * which need to be handled specially. 139 */ 140 #define NFS_EOF -98 141 #define NFS_VERF_MISMATCH -97 142 143 /* ALIGN64 aligns the given buffer and adjust buffer size to 64 bit */ 144 #define ALIGN64(x, ptr, sz) \ 145 x = ((uintptr_t)(ptr)) & (sizeof (uint64_t) - 1); \ 146 if (x) { \ 147 x = sizeof (uint64_t) - (x); \ 148 sz -= (x); \ 149 ptr += (x); \ 150 } 151 152 /* 153 * These are the vnode ops routines which implement the vnode interface to 154 * the networked file system. These routines just take their parameters, 155 * make them look networkish by putting the right info into interface structs, 156 * and then calling the appropriate remote routine(s) to do the work. 157 * 158 * Note on directory name lookup cacheing: If we detect a stale fhandle, 159 * we purge the directory cache relative to that vnode. This way, the 160 * user won't get burned by the cache repeatedly. See <nfs/rnode.h> for 161 * more details on rnode locking. 162 */ 163 164 static int nfs3_open(vnode_t **, int, cred_t *, caller_context_t *); 165 static int nfs3_close(vnode_t *, int, int, offset_t, cred_t *, 166 caller_context_t *); 167 static int nfs3_read(vnode_t *, struct uio *, int, cred_t *, 168 caller_context_t *); 169 static int nfs3_write(vnode_t *, struct uio *, int, cred_t *, 170 caller_context_t *); 171 static int nfs3_ioctl(vnode_t *, int, intptr_t, int, cred_t *, int *, 172 caller_context_t *); 173 static int nfs3_getattr(vnode_t *, struct vattr *, int, cred_t *, 174 caller_context_t *); 175 static int nfs3_setattr(vnode_t *, struct vattr *, int, cred_t *, 176 caller_context_t *); 177 static int nfs3_access(vnode_t *, int, int, cred_t *, caller_context_t *); 178 static int nfs3_readlink(vnode_t *, struct uio *, cred_t *, 179 caller_context_t *); 180 static int nfs3_fsync(vnode_t *, int, cred_t *, caller_context_t *); 181 static void nfs3_inactive(vnode_t *, cred_t *, caller_context_t *); 182 static int nfs3_lookup(vnode_t *, char *, vnode_t **, 183 struct pathname *, int, vnode_t *, cred_t *, 184 caller_context_t *, int *, pathname_t *); 185 static int nfs3_create(vnode_t *, char *, struct vattr *, enum vcexcl, 186 int, vnode_t **, cred_t *, int, caller_context_t *, 187 vsecattr_t *); 188 static int nfs3_remove(vnode_t *, char *, cred_t *, caller_context_t *, 189 int); 190 static int nfs3_link(vnode_t *, vnode_t *, char *, cred_t *, 191 caller_context_t *, int); 192 static int nfs3_rename(vnode_t *, char *, vnode_t *, char *, cred_t *, 193 caller_context_t *, int); 194 static int nfs3_mkdir(vnode_t *, char *, struct vattr *, vnode_t **, 195 cred_t *, caller_context_t *, int, vsecattr_t *); 196 static int nfs3_rmdir(vnode_t *, char *, vnode_t *, cred_t *, 197 caller_context_t *, int); 198 static int nfs3_symlink(vnode_t *, char *, struct vattr *, char *, 199 cred_t *, caller_context_t *, int); 200 static int nfs3_readdir(vnode_t *, struct uio *, cred_t *, int *, 201 caller_context_t *, int); 202 static int nfs3_fid(vnode_t *, fid_t *, caller_context_t *); 203 static int nfs3_rwlock(vnode_t *, int, caller_context_t *); 204 static void nfs3_rwunlock(vnode_t *, int, caller_context_t *); 205 static int nfs3_seek(vnode_t *, offset_t, offset_t *, caller_context_t *); 206 static int nfs3_getpage(vnode_t *, offset_t, size_t, uint_t *, 207 page_t *[], size_t, struct seg *, caddr_t, 208 enum seg_rw, cred_t *, caller_context_t *); 209 static int nfs3_putpage(vnode_t *, offset_t, size_t, int, cred_t *, 210 caller_context_t *); 211 static int nfs3_map(vnode_t *, offset_t, struct as *, caddr_t *, size_t, 212 uchar_t, uchar_t, uint_t, cred_t *, caller_context_t *); 213 static int nfs3_addmap(vnode_t *, offset_t, struct as *, caddr_t, size_t, 214 uchar_t, uchar_t, uint_t, cred_t *, caller_context_t *); 215 static int nfs3_frlock(vnode_t *, int, struct flock64 *, int, offset_t, 216 struct flk_callback *, cred_t *, caller_context_t *); 217 static int nfs3_space(vnode_t *, int, struct flock64 *, int, offset_t, 218 cred_t *, caller_context_t *); 219 static int nfs3_realvp(vnode_t *, vnode_t **, caller_context_t *); 220 static int nfs3_delmap(vnode_t *, offset_t, struct as *, caddr_t, size_t, 221 uint_t, uint_t, uint_t, cred_t *, caller_context_t *); 222 static int nfs3_pathconf(vnode_t *, int, ulong_t *, cred_t *, 223 caller_context_t *); 224 static int nfs3_pageio(vnode_t *, page_t *, u_offset_t, size_t, int, 225 cred_t *, caller_context_t *); 226 static void nfs3_dispose(vnode_t *, page_t *, int, int, cred_t *, 227 caller_context_t *); 228 static int nfs3_setsecattr(vnode_t *, vsecattr_t *, int, cred_t *, 229 caller_context_t *); 230 static int nfs3_getsecattr(vnode_t *, vsecattr_t *, int, cred_t *, 231 caller_context_t *); 232 static int nfs3_shrlock(vnode_t *, int, struct shrlock *, int, cred_t *, 233 caller_context_t *); 234 235 struct vnodeops *nfs3_vnodeops; 236 237 const fs_operation_def_t nfs3_vnodeops_template[] = { 238 VOPNAME_OPEN, { .vop_open = nfs3_open }, 239 VOPNAME_CLOSE, { .vop_close = nfs3_close }, 240 VOPNAME_READ, { .vop_read = nfs3_read }, 241 VOPNAME_WRITE, { .vop_write = nfs3_write }, 242 VOPNAME_IOCTL, { .vop_ioctl = nfs3_ioctl }, 243 VOPNAME_GETATTR, { .vop_getattr = nfs3_getattr }, 244 VOPNAME_SETATTR, { .vop_setattr = nfs3_setattr }, 245 VOPNAME_ACCESS, { .vop_access = nfs3_access }, 246 VOPNAME_LOOKUP, { .vop_lookup = nfs3_lookup }, 247 VOPNAME_CREATE, { .vop_create = nfs3_create }, 248 VOPNAME_REMOVE, { .vop_remove = nfs3_remove }, 249 VOPNAME_LINK, { .vop_link = nfs3_link }, 250 VOPNAME_RENAME, { .vop_rename = nfs3_rename }, 251 VOPNAME_MKDIR, { .vop_mkdir = nfs3_mkdir }, 252 VOPNAME_RMDIR, { .vop_rmdir = nfs3_rmdir }, 253 VOPNAME_READDIR, { .vop_readdir = nfs3_readdir }, 254 VOPNAME_SYMLINK, { .vop_symlink = nfs3_symlink }, 255 VOPNAME_READLINK, { .vop_readlink = nfs3_readlink }, 256 VOPNAME_FSYNC, { .vop_fsync = nfs3_fsync }, 257 VOPNAME_INACTIVE, { .vop_inactive = nfs3_inactive }, 258 VOPNAME_FID, { .vop_fid = nfs3_fid }, 259 VOPNAME_RWLOCK, { .vop_rwlock = nfs3_rwlock }, 260 VOPNAME_RWUNLOCK, { .vop_rwunlock = nfs3_rwunlock }, 261 VOPNAME_SEEK, { .vop_seek = nfs3_seek }, 262 VOPNAME_FRLOCK, { .vop_frlock = nfs3_frlock }, 263 VOPNAME_SPACE, { .vop_space = nfs3_space }, 264 VOPNAME_REALVP, { .vop_realvp = nfs3_realvp }, 265 VOPNAME_GETPAGE, { .vop_getpage = nfs3_getpage }, 266 VOPNAME_PUTPAGE, { .vop_putpage = nfs3_putpage }, 267 VOPNAME_MAP, { .vop_map = nfs3_map }, 268 VOPNAME_ADDMAP, { .vop_addmap = nfs3_addmap }, 269 VOPNAME_DELMAP, { .vop_delmap = nfs3_delmap }, 270 /* no separate nfs3_dump */ 271 VOPNAME_DUMP, { .vop_dump = nfs_dump }, 272 VOPNAME_PATHCONF, { .vop_pathconf = nfs3_pathconf }, 273 VOPNAME_PAGEIO, { .vop_pageio = nfs3_pageio }, 274 VOPNAME_DISPOSE, { .vop_dispose = nfs3_dispose }, 275 VOPNAME_SETSECATTR, { .vop_setsecattr = nfs3_setsecattr }, 276 VOPNAME_GETSECATTR, { .vop_getsecattr = nfs3_getsecattr }, 277 VOPNAME_SHRLOCK, { .vop_shrlock = nfs3_shrlock }, 278 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 279 NULL, NULL 280 }; 281 282 /* 283 * XXX: This is referenced in modstubs.s 284 */ 285 struct vnodeops * 286 nfs3_getvnodeops(void) 287 { 288 return (nfs3_vnodeops); 289 } 290 291 /* ARGSUSED */ 292 static int 293 nfs3_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct) 294 { 295 int error; 296 struct vattr va; 297 rnode_t *rp; 298 vnode_t *vp; 299 300 vp = *vpp; 301 if (nfs_zone() != VTOMI(vp)->mi_zone) 302 return (EIO); 303 rp = VTOR(vp); 304 mutex_enter(&rp->r_statelock); 305 if (rp->r_cred == NULL) { 306 crhold(cr); 307 rp->r_cred = cr; 308 } 309 mutex_exit(&rp->r_statelock); 310 311 /* 312 * If there is no cached data or if close-to-open 313 * consistency checking is turned off, we can avoid 314 * the over the wire getattr. Otherwise, if the 315 * file system is mounted readonly, then just verify 316 * the caches are up to date using the normal mechanism. 317 * Else, if the file is not mmap'd, then just mark 318 * the attributes as timed out. They will be refreshed 319 * and the caches validated prior to being used. 320 * Else, the file system is mounted writeable so 321 * force an over the wire GETATTR in order to ensure 322 * that all cached data is valid. 323 */ 324 if (vp->v_count > 1 || 325 ((vn_has_cached_data(vp) || HAVE_RDDIR_CACHE(rp)) && 326 !(VTOMI(vp)->mi_flags & MI_NOCTO))) { 327 if (vn_is_readonly(vp)) 328 error = nfs3_validate_caches(vp, cr); 329 else if (rp->r_mapcnt == 0 && vp->v_count == 1) { 330 PURGE_ATTRCACHE(vp); 331 error = 0; 332 } else { 333 va.va_mask = AT_ALL; 334 error = nfs3_getattr_otw(vp, &va, cr); 335 } 336 } else 337 error = 0; 338 339 return (error); 340 } 341 342 /* ARGSUSED */ 343 static int 344 nfs3_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr, 345 caller_context_t *ct) 346 { 347 rnode_t *rp; 348 int error; 349 struct vattr va; 350 351 /* 352 * zone_enter(2) prevents processes from changing zones with NFS files 353 * open; if we happen to get here from the wrong zone we can't do 354 * anything over the wire. 355 */ 356 if (VTOMI(vp)->mi_zone != nfs_zone()) { 357 /* 358 * We could attempt to clean up locks, except we're sure 359 * that the current process didn't acquire any locks on 360 * the file: any attempt to lock a file belong to another zone 361 * will fail, and one can't lock an NFS file and then change 362 * zones, as that fails too. 363 * 364 * Returning an error here is the sane thing to do. A 365 * subsequent call to VN_RELE() which translates to a 366 * nfs3_inactive() will clean up state: if the zone of the 367 * vnode's origin is still alive and kicking, an async worker 368 * thread will handle the request (from the correct zone), and 369 * everything (minus the commit and final nfs3_getattr_otw() 370 * call) should be OK. If the zone is going away 371 * nfs_async_inactive() will throw away cached pages inline. 372 */ 373 return (EIO); 374 } 375 376 /* 377 * If we are using local locking for this filesystem, then 378 * release all of the SYSV style record locks. Otherwise, 379 * we are doing network locking and we need to release all 380 * of the network locks. All of the locks held by this 381 * process on this file are released no matter what the 382 * incoming reference count is. 383 */ 384 if (VTOMI(vp)->mi_flags & MI_LLOCK) { 385 cleanlocks(vp, ttoproc(curthread)->p_pid, 0); 386 cleanshares(vp, ttoproc(curthread)->p_pid); 387 } else 388 nfs_lockrelease(vp, flag, offset, cr); 389 390 if (count > 1) 391 return (0); 392 393 /* 394 * If the file has been `unlinked', then purge the 395 * DNLC so that this vnode will get reycled quicker 396 * and the .nfs* file on the server will get removed. 397 */ 398 rp = VTOR(vp); 399 if (rp->r_unldvp != NULL) 400 dnlc_purge_vp(vp); 401 402 /* 403 * If the file was open for write and there are pages, 404 * then if the file system was mounted using the "no-close- 405 * to-open" semantics, then start an asynchronous flush 406 * of the all of the pages in the file. 407 * else the file system was not mounted using the "no-close- 408 * to-open" semantics, then do a synchronous flush and 409 * commit of all of the dirty and uncommitted pages. 410 * 411 * The asynchronous flush of the pages in the "nocto" path 412 * mostly just associates a cred pointer with the rnode so 413 * writes which happen later will have a better chance of 414 * working. It also starts the data being written to the 415 * server, but without unnecessarily delaying the application. 416 */ 417 if ((flag & FWRITE) && vn_has_cached_data(vp)) { 418 if (VTOMI(vp)->mi_flags & MI_NOCTO) { 419 error = nfs3_putpage(vp, (offset_t)0, 0, B_ASYNC, 420 cr, ct); 421 if (error == EAGAIN) 422 error = 0; 423 } else 424 error = nfs3_putpage_commit(vp, (offset_t)0, 0, cr); 425 if (!error) { 426 mutex_enter(&rp->r_statelock); 427 error = rp->r_error; 428 rp->r_error = 0; 429 mutex_exit(&rp->r_statelock); 430 } 431 } else { 432 mutex_enter(&rp->r_statelock); 433 error = rp->r_error; 434 rp->r_error = 0; 435 mutex_exit(&rp->r_statelock); 436 } 437 438 /* 439 * If RWRITEATTR is set, then issue an over the wire GETATTR to 440 * refresh the attribute cache with a set of attributes which 441 * weren't returned from a WRITE. This will enable the close- 442 * to-open processing to work. 443 */ 444 if (rp->r_flags & RWRITEATTR) 445 (void) nfs3_getattr_otw(vp, &va, cr); 446 447 return (error); 448 } 449 450 /* ARGSUSED */ 451 static int 452 nfs3_directio_read(vnode_t *vp, struct uio *uiop, cred_t *cr) 453 { 454 mntinfo_t *mi; 455 READ3args args; 456 READ3uiores res; 457 int tsize; 458 offset_t offset; 459 ssize_t count; 460 int error; 461 int douprintf; 462 failinfo_t fi; 463 char *sv_hostname; 464 465 mi = VTOMI(vp); 466 ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 467 sv_hostname = VTOR(vp)->r_server->sv_hostname; 468 469 douprintf = 1; 470 args.file = *VTOFH3(vp); 471 fi.vp = vp; 472 fi.fhp = (caddr_t)&args.file; 473 fi.copyproc = nfs3copyfh; 474 fi.lookupproc = nfs3lookup; 475 fi.xattrdirproc = acl_getxattrdir3; 476 477 res.uiop = uiop; 478 479 offset = uiop->uio_loffset; 480 count = uiop->uio_resid; 481 482 do { 483 if (mi->mi_io_kstats) { 484 mutex_enter(&mi->mi_lock); 485 kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats)); 486 mutex_exit(&mi->mi_lock); 487 } 488 489 do { 490 tsize = MIN(mi->mi_tsize, count); 491 args.offset = (offset3)offset; 492 args.count = (count3)tsize; 493 res.size = (uint_t)tsize; 494 error = rfs3call(mi, NFSPROC3_READ, 495 xdr_READ3args, (caddr_t)&args, 496 xdr_READ3uiores, (caddr_t)&res, cr, 497 &douprintf, &res.status, 0, &fi); 498 } while (error == ENFS_TRYAGAIN); 499 500 if (mi->mi_io_kstats) { 501 mutex_enter(&mi->mi_lock); 502 kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats)); 503 mutex_exit(&mi->mi_lock); 504 } 505 506 if (error) 507 return (error); 508 509 error = geterrno3(res.status); 510 if (error) 511 return (error); 512 513 if (res.count != res.size) { 514 zcmn_err(getzoneid(), CE_WARN, 515 "nfs3_directio_read: server %s returned incorrect amount", 516 sv_hostname); 517 return (EIO); 518 } 519 count -= res.count; 520 offset += res.count; 521 if (mi->mi_io_kstats) { 522 mutex_enter(&mi->mi_lock); 523 KSTAT_IO_PTR(mi->mi_io_kstats)->reads++; 524 KSTAT_IO_PTR(mi->mi_io_kstats)->nread += res.count; 525 mutex_exit(&mi->mi_lock); 526 } 527 lwp_stat_update(LWP_STAT_INBLK, 1); 528 } while (count && !res.eof); 529 530 return (0); 531 } 532 533 /* ARGSUSED */ 534 static int 535 nfs3_read(vnode_t *vp, struct uio *uiop, int ioflag, cred_t *cr, 536 caller_context_t *ct) 537 { 538 rnode_t *rp; 539 u_offset_t off; 540 offset_t diff; 541 int on; 542 size_t n; 543 caddr_t base; 544 uint_t flags; 545 int error = 0; 546 mntinfo_t *mi; 547 548 rp = VTOR(vp); 549 mi = VTOMI(vp); 550 551 ASSERT(nfs_rw_lock_held(&rp->r_rwlock, RW_READER)); 552 553 if (nfs_zone() != mi->mi_zone) 554 return (EIO); 555 556 if (vp->v_type != VREG) 557 return (EISDIR); 558 559 if (uiop->uio_resid == 0) 560 return (0); 561 562 if (uiop->uio_loffset < 0 || uiop->uio_loffset + uiop->uio_resid < 0) 563 return (EINVAL); 564 565 /* 566 * Bypass VM if caching has been disabled (e.g., locking) or if 567 * using client-side direct I/O and the file is not mmap'd and 568 * there are no cached pages. 569 */ 570 if ((vp->v_flag & VNOCACHE) || 571 (((rp->r_flags & RDIRECTIO) || (mi->mi_flags & MI_DIRECTIO)) && 572 rp->r_mapcnt == 0 && !vn_has_cached_data(vp))) { 573 return (nfs3_directio_read(vp, uiop, cr)); 574 } 575 576 do { 577 off = uiop->uio_loffset & MAXBMASK; /* mapping offset */ 578 on = uiop->uio_loffset & MAXBOFFSET; /* Relative offset */ 579 n = MIN(MAXBSIZE - on, uiop->uio_resid); 580 581 error = nfs3_validate_caches(vp, cr); 582 if (error) 583 break; 584 585 mutex_enter(&rp->r_statelock); 586 diff = rp->r_size - uiop->uio_loffset; 587 mutex_exit(&rp->r_statelock); 588 if (diff <= 0) 589 break; 590 if (diff < n) 591 n = (size_t)diff; 592 593 if (vpm_enable) { 594 /* 595 * Copy data. 596 */ 597 error = vpm_data_copy(vp, off + on, n, uiop, 598 1, NULL, 0, S_READ); 599 } else { 600 base = segmap_getmapflt(segkmap, vp, off + on, n, 1, 601 S_READ); 602 603 error = uiomove(base + on, n, UIO_READ, uiop); 604 } 605 606 if (!error) { 607 /* 608 * If read a whole block or read to eof, 609 * won't need this buffer again soon. 610 */ 611 mutex_enter(&rp->r_statelock); 612 if (n + on == MAXBSIZE || 613 uiop->uio_loffset == rp->r_size) 614 flags = SM_DONTNEED; 615 else 616 flags = 0; 617 mutex_exit(&rp->r_statelock); 618 if (vpm_enable) { 619 error = vpm_sync_pages(vp, off, n, flags); 620 } else { 621 error = segmap_release(segkmap, base, flags); 622 } 623 } else { 624 if (vpm_enable) { 625 (void) vpm_sync_pages(vp, off, n, 0); 626 } else { 627 (void) segmap_release(segkmap, base, 0); 628 } 629 } 630 } while (!error && uiop->uio_resid > 0); 631 632 return (error); 633 } 634 635 /* ARGSUSED */ 636 static int 637 nfs3_write(vnode_t *vp, struct uio *uiop, int ioflag, cred_t *cr, 638 caller_context_t *ct) 639 { 640 rlim64_t limit = uiop->uio_llimit; 641 rnode_t *rp; 642 u_offset_t off; 643 caddr_t base; 644 uint_t flags; 645 int remainder; 646 size_t n; 647 int on; 648 int error; 649 int resid; 650 offset_t offset; 651 mntinfo_t *mi; 652 uint_t bsize; 653 654 rp = VTOR(vp); 655 656 if (vp->v_type != VREG) 657 return (EISDIR); 658 659 mi = VTOMI(vp); 660 if (nfs_zone() != mi->mi_zone) 661 return (EIO); 662 if (uiop->uio_resid == 0) 663 return (0); 664 665 if (ioflag & FAPPEND) { 666 struct vattr va; 667 668 /* 669 * Must serialize if appending. 670 */ 671 if (nfs_rw_lock_held(&rp->r_rwlock, RW_READER)) { 672 nfs_rw_exit(&rp->r_rwlock); 673 if (nfs_rw_enter_sig(&rp->r_rwlock, RW_WRITER, 674 INTR(vp))) 675 return (EINTR); 676 } 677 678 va.va_mask = AT_SIZE; 679 error = nfs3getattr(vp, &va, cr); 680 if (error) 681 return (error); 682 uiop->uio_loffset = va.va_size; 683 } 684 685 offset = uiop->uio_loffset + uiop->uio_resid; 686 687 if (uiop->uio_loffset < 0 || offset < 0) 688 return (EINVAL); 689 690 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 691 limit = MAXOFFSET_T; 692 693 /* 694 * Check to make sure that the process will not exceed 695 * its limit on file size. It is okay to write up to 696 * the limit, but not beyond. Thus, the write which 697 * reaches the limit will be short and the next write 698 * will return an error. 699 */ 700 remainder = 0; 701 if (offset > limit) { 702 remainder = offset - limit; 703 uiop->uio_resid = limit - uiop->uio_loffset; 704 if (uiop->uio_resid <= 0) { 705 proc_t *p = ttoproc(curthread); 706 707 uiop->uio_resid += remainder; 708 mutex_enter(&p->p_lock); 709 (void) rctl_action(rctlproc_legacy[RLIMIT_FSIZE], 710 p->p_rctls, p, RCA_UNSAFE_SIGINFO); 711 mutex_exit(&p->p_lock); 712 return (EFBIG); 713 } 714 } 715 716 if (nfs_rw_enter_sig(&rp->r_lkserlock, RW_READER, INTR(vp))) 717 return (EINTR); 718 719 /* 720 * Bypass VM if caching has been disabled (e.g., locking) or if 721 * using client-side direct I/O and the file is not mmap'd and 722 * there are no cached pages. 723 */ 724 if ((vp->v_flag & VNOCACHE) || 725 (((rp->r_flags & RDIRECTIO) || (mi->mi_flags & MI_DIRECTIO)) && 726 rp->r_mapcnt == 0 && !vn_has_cached_data(vp))) { 727 size_t bufsize; 728 int count; 729 u_offset_t org_offset; 730 stable_how stab_comm; 731 732 nfs3_fwrite: 733 if (rp->r_flags & RSTALE) { 734 resid = uiop->uio_resid; 735 offset = uiop->uio_loffset; 736 error = rp->r_error; 737 goto bottom; 738 } 739 bufsize = MIN(uiop->uio_resid, mi->mi_stsize); 740 base = kmem_alloc(bufsize, KM_SLEEP); 741 do { 742 if (ioflag & FDSYNC) 743 stab_comm = DATA_SYNC; 744 else 745 stab_comm = FILE_SYNC; 746 resid = uiop->uio_resid; 747 offset = uiop->uio_loffset; 748 count = MIN(uiop->uio_resid, bufsize); 749 org_offset = uiop->uio_loffset; 750 error = uiomove(base, count, UIO_WRITE, uiop); 751 if (!error) { 752 error = nfs3write(vp, base, org_offset, 753 count, cr, &stab_comm); 754 } 755 } while (!error && uiop->uio_resid > 0); 756 kmem_free(base, bufsize); 757 goto bottom; 758 } 759 760 761 bsize = vp->v_vfsp->vfs_bsize; 762 763 do { 764 off = uiop->uio_loffset & MAXBMASK; /* mapping offset */ 765 on = uiop->uio_loffset & MAXBOFFSET; /* Relative offset */ 766 n = MIN(MAXBSIZE - on, uiop->uio_resid); 767 768 resid = uiop->uio_resid; 769 offset = uiop->uio_loffset; 770 771 if (rp->r_flags & RSTALE) { 772 error = rp->r_error; 773 break; 774 } 775 776 /* 777 * Don't create dirty pages faster than they 778 * can be cleaned so that the system doesn't 779 * get imbalanced. If the async queue is 780 * maxed out, then wait for it to drain before 781 * creating more dirty pages. Also, wait for 782 * any threads doing pagewalks in the vop_getattr 783 * entry points so that they don't block for 784 * long periods. 785 */ 786 mutex_enter(&rp->r_statelock); 787 while ((mi->mi_max_threads != 0 && 788 rp->r_awcount > 2 * mi->mi_max_threads) || 789 rp->r_gcount > 0) 790 cv_wait(&rp->r_cv, &rp->r_statelock); 791 mutex_exit(&rp->r_statelock); 792 793 if (vpm_enable) { 794 /* 795 * It will use kpm mappings, so no need to 796 * pass an address. 797 */ 798 error = writerp(rp, NULL, n, uiop, 0); 799 } else { 800 if (segmap_kpm) { 801 int pon = uiop->uio_loffset & PAGEOFFSET; 802 size_t pn = MIN(PAGESIZE - pon, 803 uiop->uio_resid); 804 int pagecreate; 805 806 mutex_enter(&rp->r_statelock); 807 pagecreate = (pon == 0) && (pn == PAGESIZE || 808 uiop->uio_loffset + pn >= rp->r_size); 809 mutex_exit(&rp->r_statelock); 810 811 base = segmap_getmapflt(segkmap, vp, off + on, 812 pn, !pagecreate, S_WRITE); 813 814 error = writerp(rp, base + pon, n, uiop, 815 pagecreate); 816 817 } else { 818 base = segmap_getmapflt(segkmap, vp, off + on, 819 n, 0, S_READ); 820 error = writerp(rp, base + on, n, uiop, 0); 821 } 822 } 823 824 if (!error) { 825 if (mi->mi_flags & MI_NOAC) 826 flags = SM_WRITE; 827 else if ((uiop->uio_loffset % bsize) == 0 || 828 IS_SWAPVP(vp)) { 829 /* 830 * Have written a whole block. 831 * Start an asynchronous write 832 * and mark the buffer to 833 * indicate that it won't be 834 * needed again soon. 835 */ 836 flags = SM_WRITE | SM_ASYNC | SM_DONTNEED; 837 } else 838 flags = 0; 839 if ((ioflag & (FSYNC|FDSYNC)) || 840 (rp->r_flags & ROUTOFSPACE)) { 841 flags &= ~SM_ASYNC; 842 flags |= SM_WRITE; 843 } 844 if (vpm_enable) { 845 error = vpm_sync_pages(vp, off, n, flags); 846 } else { 847 error = segmap_release(segkmap, base, flags); 848 } 849 } else { 850 if (vpm_enable) { 851 (void) vpm_sync_pages(vp, off, n, 0); 852 } else { 853 (void) segmap_release(segkmap, base, 0); 854 } 855 /* 856 * In the event that we got an access error while 857 * faulting in a page for a write-only file just 858 * force a write. 859 */ 860 if (error == EACCES) 861 goto nfs3_fwrite; 862 } 863 } while (!error && uiop->uio_resid > 0); 864 865 bottom: 866 if (error) { 867 uiop->uio_resid = resid + remainder; 868 uiop->uio_loffset = offset; 869 } else 870 uiop->uio_resid += remainder; 871 872 nfs_rw_exit(&rp->r_lkserlock); 873 874 return (error); 875 } 876 877 /* 878 * Flags are composed of {B_ASYNC, B_INVAL, B_FREE, B_DONTNEED} 879 */ 880 static int 881 nfs3_rdwrlbn(vnode_t *vp, page_t *pp, u_offset_t off, size_t len, 882 int flags, cred_t *cr) 883 { 884 struct buf *bp; 885 int error; 886 page_t *savepp; 887 uchar_t fsdata; 888 stable_how stab_comm; 889 890 ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 891 bp = pageio_setup(pp, len, vp, flags); 892 ASSERT(bp != NULL); 893 894 /* 895 * pageio_setup should have set b_addr to 0. This 896 * is correct since we want to do I/O on a page 897 * boundary. bp_mapin will use this addr to calculate 898 * an offset, and then set b_addr to the kernel virtual 899 * address it allocated for us. 900 */ 901 ASSERT(bp->b_un.b_addr == 0); 902 903 bp->b_edev = 0; 904 bp->b_dev = 0; 905 bp->b_lblkno = lbtodb(off); 906 bp->b_file = vp; 907 bp->b_offset = (offset_t)off; 908 bp_mapin(bp); 909 910 /* 911 * Calculate the desired level of stability to write data 912 * on the server and then mark all of the pages to reflect 913 * this. 914 */ 915 if ((flags & (B_WRITE|B_ASYNC)) == (B_WRITE|B_ASYNC) && 916 freemem > desfree) { 917 stab_comm = UNSTABLE; 918 fsdata = C_DELAYCOMMIT; 919 } else { 920 stab_comm = FILE_SYNC; 921 fsdata = C_NOCOMMIT; 922 } 923 924 savepp = pp; 925 do { 926 pp->p_fsdata = fsdata; 927 } while ((pp = pp->p_next) != savepp); 928 929 error = nfs3_bio(bp, &stab_comm, cr); 930 931 bp_mapout(bp); 932 pageio_done(bp); 933 934 /* 935 * If the server wrote pages in a more stable fashion than 936 * was requested, then clear all of the marks in the pages 937 * indicating that COMMIT operations were required. 938 */ 939 if (stab_comm != UNSTABLE && fsdata == C_DELAYCOMMIT) { 940 do { 941 pp->p_fsdata = C_NOCOMMIT; 942 } while ((pp = pp->p_next) != savepp); 943 } 944 945 return (error); 946 } 947 948 /* 949 * Write to file. Writes to remote server in largest size 950 * chunks that the server can handle. Write is synchronous. 951 */ 952 static int 953 nfs3write(vnode_t *vp, caddr_t base, u_offset_t offset, int count, cred_t *cr, 954 stable_how *stab_comm) 955 { 956 mntinfo_t *mi; 957 WRITE3args args; 958 WRITE3res res; 959 int error; 960 int tsize; 961 rnode_t *rp; 962 int douprintf; 963 964 rp = VTOR(vp); 965 mi = VTOMI(vp); 966 967 ASSERT(nfs_zone() == mi->mi_zone); 968 969 args.file = *VTOFH3(vp); 970 args.stable = *stab_comm; 971 972 *stab_comm = FILE_SYNC; 973 974 douprintf = 1; 975 976 do { 977 if ((vp->v_flag & VNOCACHE) || 978 (rp->r_flags & RDIRECTIO) || 979 (mi->mi_flags & MI_DIRECTIO)) 980 tsize = MIN(mi->mi_stsize, count); 981 else 982 tsize = MIN(mi->mi_curwrite, count); 983 args.offset = (offset3)offset; 984 args.count = (count3)tsize; 985 args.data.data_len = (uint_t)tsize; 986 args.data.data_val = base; 987 988 if (mi->mi_io_kstats) { 989 mutex_enter(&mi->mi_lock); 990 kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats)); 991 mutex_exit(&mi->mi_lock); 992 } 993 args.mblk = NULL; 994 do { 995 error = rfs3call(mi, NFSPROC3_WRITE, 996 xdr_WRITE3args, (caddr_t)&args, 997 xdr_WRITE3res, (caddr_t)&res, cr, 998 &douprintf, &res.status, 0, NULL); 999 } while (error == ENFS_TRYAGAIN); 1000 if (mi->mi_io_kstats) { 1001 mutex_enter(&mi->mi_lock); 1002 kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats)); 1003 mutex_exit(&mi->mi_lock); 1004 } 1005 1006 if (error) 1007 return (error); 1008 error = geterrno3(res.status); 1009 if (!error) { 1010 if (res.resok.count > args.count) { 1011 zcmn_err(getzoneid(), CE_WARN, 1012 "nfs3write: server %s wrote %u, " 1013 "requested was %u", 1014 rp->r_server->sv_hostname, 1015 res.resok.count, args.count); 1016 return (EIO); 1017 } 1018 if (res.resok.committed == UNSTABLE) { 1019 *stab_comm = UNSTABLE; 1020 if (args.stable == DATA_SYNC || 1021 args.stable == FILE_SYNC) { 1022 zcmn_err(getzoneid(), CE_WARN, 1023 "nfs3write: server %s did not commit to stable storage", 1024 rp->r_server->sv_hostname); 1025 return (EIO); 1026 } 1027 } 1028 tsize = (int)res.resok.count; 1029 count -= tsize; 1030 base += tsize; 1031 offset += tsize; 1032 if (mi->mi_io_kstats) { 1033 mutex_enter(&mi->mi_lock); 1034 KSTAT_IO_PTR(mi->mi_io_kstats)->writes++; 1035 KSTAT_IO_PTR(mi->mi_io_kstats)->nwritten += 1036 tsize; 1037 mutex_exit(&mi->mi_lock); 1038 } 1039 lwp_stat_update(LWP_STAT_OUBLK, 1); 1040 mutex_enter(&rp->r_statelock); 1041 if (rp->r_flags & RHAVEVERF) { 1042 if (rp->r_verf != res.resok.verf) { 1043 nfs3_set_mod(vp); 1044 rp->r_verf = res.resok.verf; 1045 /* 1046 * If the data was written UNSTABLE, 1047 * then might as well stop because 1048 * the whole block will have to get 1049 * rewritten anyway. 1050 */ 1051 if (*stab_comm == UNSTABLE) { 1052 mutex_exit(&rp->r_statelock); 1053 break; 1054 } 1055 } 1056 } else { 1057 rp->r_verf = res.resok.verf; 1058 rp->r_flags |= RHAVEVERF; 1059 } 1060 /* 1061 * Mark the attribute cache as timed out and 1062 * set RWRITEATTR to indicate that the file 1063 * was modified with a WRITE operation and 1064 * that the attributes can not be trusted. 1065 */ 1066 PURGE_ATTRCACHE_LOCKED(rp); 1067 rp->r_flags |= RWRITEATTR; 1068 mutex_exit(&rp->r_statelock); 1069 } 1070 } while (!error && count); 1071 1072 return (error); 1073 } 1074 1075 /* 1076 * Read from a file. Reads data in largest chunks our interface can handle. 1077 */ 1078 static int 1079 nfs3read(vnode_t *vp, caddr_t base, offset_t offset, int count, 1080 size_t *residp, cred_t *cr) 1081 { 1082 mntinfo_t *mi; 1083 READ3args args; 1084 READ3vres res; 1085 int tsize; 1086 int error; 1087 int douprintf; 1088 failinfo_t fi; 1089 rnode_t *rp; 1090 struct vattr va; 1091 hrtime_t t; 1092 1093 rp = VTOR(vp); 1094 mi = VTOMI(vp); 1095 ASSERT(nfs_zone() == mi->mi_zone); 1096 douprintf = 1; 1097 1098 args.file = *VTOFH3(vp); 1099 fi.vp = vp; 1100 fi.fhp = (caddr_t)&args.file; 1101 fi.copyproc = nfs3copyfh; 1102 fi.lookupproc = nfs3lookup; 1103 fi.xattrdirproc = acl_getxattrdir3; 1104 1105 res.pov.fres.vp = vp; 1106 res.pov.fres.vap = &va; 1107 1108 *residp = count; 1109 do { 1110 if (mi->mi_io_kstats) { 1111 mutex_enter(&mi->mi_lock); 1112 kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats)); 1113 mutex_exit(&mi->mi_lock); 1114 } 1115 1116 do { 1117 if ((vp->v_flag & VNOCACHE) || 1118 (rp->r_flags & RDIRECTIO) || 1119 (mi->mi_flags & MI_DIRECTIO)) 1120 tsize = MIN(mi->mi_tsize, count); 1121 else 1122 tsize = MIN(mi->mi_curread, count); 1123 res.data.data_val = base; 1124 res.data.data_len = tsize; 1125 args.offset = (offset3)offset; 1126 args.count = (count3)tsize; 1127 t = gethrtime(); 1128 error = rfs3call(mi, NFSPROC3_READ, 1129 xdr_READ3args, (caddr_t)&args, 1130 xdr_READ3vres, (caddr_t)&res, cr, 1131 &douprintf, &res.status, 0, &fi); 1132 } while (error == ENFS_TRYAGAIN); 1133 1134 if (mi->mi_io_kstats) { 1135 mutex_enter(&mi->mi_lock); 1136 kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats)); 1137 mutex_exit(&mi->mi_lock); 1138 } 1139 1140 if (error) 1141 return (error); 1142 1143 error = geterrno3(res.status); 1144 if (error) 1145 return (error); 1146 1147 if (res.count != res.data.data_len) { 1148 zcmn_err(getzoneid(), CE_WARN, 1149 "nfs3read: server %s returned incorrect amount", 1150 rp->r_server->sv_hostname); 1151 return (EIO); 1152 } 1153 1154 count -= res.count; 1155 *residp = count; 1156 base += res.count; 1157 offset += res.count; 1158 if (mi->mi_io_kstats) { 1159 mutex_enter(&mi->mi_lock); 1160 KSTAT_IO_PTR(mi->mi_io_kstats)->reads++; 1161 KSTAT_IO_PTR(mi->mi_io_kstats)->nread += res.count; 1162 mutex_exit(&mi->mi_lock); 1163 } 1164 lwp_stat_update(LWP_STAT_INBLK, 1); 1165 } while (count && !res.eof); 1166 1167 if (res.pov.attributes) { 1168 mutex_enter(&rp->r_statelock); 1169 if (!CACHE_VALID(rp, va.va_mtime, va.va_size)) { 1170 mutex_exit(&rp->r_statelock); 1171 PURGE_ATTRCACHE(vp); 1172 } else { 1173 if (rp->r_mtime <= t) 1174 nfs_attrcache_va(vp, &va); 1175 mutex_exit(&rp->r_statelock); 1176 } 1177 } 1178 1179 return (0); 1180 } 1181 1182 /* ARGSUSED */ 1183 static int 1184 nfs3_ioctl(vnode_t *vp, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp, 1185 caller_context_t *ct) 1186 { 1187 1188 if (nfs_zone() != VTOMI(vp)->mi_zone) 1189 return (EIO); 1190 switch (cmd) { 1191 case _FIODIRECTIO: 1192 return (nfs_directio(vp, (int)arg, cr)); 1193 default: 1194 return (ENOTTY); 1195 } 1196 } 1197 1198 /* ARGSUSED */ 1199 static int 1200 nfs3_getattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr, 1201 caller_context_t *ct) 1202 { 1203 int error; 1204 rnode_t *rp; 1205 1206 if (nfs_zone() != VTOMI(vp)->mi_zone) 1207 return (EIO); 1208 /* 1209 * If it has been specified that the return value will 1210 * just be used as a hint, and we are only being asked 1211 * for size, fsid or rdevid, then return the client's 1212 * notion of these values without checking to make sure 1213 * that the attribute cache is up to date. 1214 * The whole point is to avoid an over the wire GETATTR 1215 * call. 1216 */ 1217 rp = VTOR(vp); 1218 if (flags & ATTR_HINT) { 1219 if (vap->va_mask == 1220 (vap->va_mask & (AT_SIZE | AT_FSID | AT_RDEV))) { 1221 mutex_enter(&rp->r_statelock); 1222 if (vap->va_mask | AT_SIZE) 1223 vap->va_size = rp->r_size; 1224 if (vap->va_mask | AT_FSID) 1225 vap->va_fsid = rp->r_attr.va_fsid; 1226 if (vap->va_mask | AT_RDEV) 1227 vap->va_rdev = rp->r_attr.va_rdev; 1228 mutex_exit(&rp->r_statelock); 1229 return (0); 1230 } 1231 } 1232 1233 /* 1234 * Only need to flush pages if asking for the mtime 1235 * and if there any dirty pages or any outstanding 1236 * asynchronous (write) requests for this file. 1237 */ 1238 if (vap->va_mask & AT_MTIME) { 1239 if (vn_has_cached_data(vp) && 1240 ((rp->r_flags & RDIRTY) || rp->r_awcount > 0)) { 1241 mutex_enter(&rp->r_statelock); 1242 rp->r_gcount++; 1243 mutex_exit(&rp->r_statelock); 1244 error = nfs3_putpage(vp, (offset_t)0, 0, 0, cr, ct); 1245 mutex_enter(&rp->r_statelock); 1246 if (error && (error == ENOSPC || error == EDQUOT)) { 1247 if (!rp->r_error) 1248 rp->r_error = error; 1249 } 1250 if (--rp->r_gcount == 0) 1251 cv_broadcast(&rp->r_cv); 1252 mutex_exit(&rp->r_statelock); 1253 } 1254 } 1255 1256 return (nfs3getattr(vp, vap, cr)); 1257 } 1258 1259 /*ARGSUSED4*/ 1260 static int 1261 nfs3_setattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr, 1262 caller_context_t *ct) 1263 { 1264 int error; 1265 struct vattr va; 1266 1267 if (vap->va_mask & AT_NOSET) 1268 return (EINVAL); 1269 if (nfs_zone() != VTOMI(vp)->mi_zone) 1270 return (EIO); 1271 1272 va.va_mask = AT_UID | AT_MODE; 1273 error = nfs3getattr(vp, &va, cr); 1274 if (error) 1275 return (error); 1276 1277 error = secpolicy_vnode_setattr(cr, vp, vap, &va, flags, nfs3_accessx, 1278 vp); 1279 if (error) 1280 return (error); 1281 1282 return (nfs3setattr(vp, vap, flags, cr)); 1283 } 1284 1285 static int 1286 nfs3setattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr) 1287 { 1288 int error; 1289 uint_t mask; 1290 SETATTR3args args; 1291 SETATTR3res res; 1292 int douprintf; 1293 rnode_t *rp; 1294 struct vattr va; 1295 mode_t omode; 1296 vsecattr_t *vsp; 1297 hrtime_t t; 1298 1299 ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 1300 mask = vap->va_mask; 1301 1302 rp = VTOR(vp); 1303 1304 /* 1305 * Only need to flush pages if there are any pages and 1306 * if the file is marked as dirty in some fashion. The 1307 * file must be flushed so that we can accurately 1308 * determine the size of the file and the cached data 1309 * after the SETATTR returns. A file is considered to 1310 * be dirty if it is either marked with RDIRTY, has 1311 * outstanding i/o's active, or is mmap'd. In this 1312 * last case, we can't tell whether there are dirty 1313 * pages, so we flush just to be sure. 1314 */ 1315 if (vn_has_cached_data(vp) && 1316 ((rp->r_flags & RDIRTY) || 1317 rp->r_count > 0 || 1318 rp->r_mapcnt > 0)) { 1319 ASSERT(vp->v_type != VCHR); 1320 error = nfs3_putpage(vp, (offset_t)0, 0, 0, cr, NULL); 1321 if (error && (error == ENOSPC || error == EDQUOT)) { 1322 mutex_enter(&rp->r_statelock); 1323 if (!rp->r_error) 1324 rp->r_error = error; 1325 mutex_exit(&rp->r_statelock); 1326 } 1327 } 1328 1329 args.object = *RTOFH3(rp); 1330 /* 1331 * If the intent is for the server to set the times, 1332 * there is no point in have the mask indicating set mtime or 1333 * atime, because the vap values may be junk, and so result 1334 * in an overflow error. Remove these flags from the vap mask 1335 * before calling in this case, and restore them afterwards. 1336 */ 1337 if ((mask & (AT_ATIME | AT_MTIME)) && !(flags & ATTR_UTIME)) { 1338 /* Use server times, so don't set the args time fields */ 1339 vap->va_mask &= ~(AT_ATIME | AT_MTIME); 1340 error = vattr_to_sattr3(vap, &args.new_attributes); 1341 vap->va_mask |= (mask & (AT_ATIME | AT_MTIME)); 1342 if (mask & AT_ATIME) { 1343 args.new_attributes.atime.set_it = SET_TO_SERVER_TIME; 1344 } 1345 if (mask & AT_MTIME) { 1346 args.new_attributes.mtime.set_it = SET_TO_SERVER_TIME; 1347 } 1348 } else { 1349 /* Either do not set times or use the client specified times */ 1350 error = vattr_to_sattr3(vap, &args.new_attributes); 1351 } 1352 1353 if (error) { 1354 /* req time field(s) overflow - return immediately */ 1355 return (error); 1356 } 1357 1358 va.va_mask = AT_MODE | AT_CTIME; 1359 error = nfs3getattr(vp, &va, cr); 1360 if (error) 1361 return (error); 1362 omode = va.va_mode; 1363 1364 tryagain: 1365 if (mask & AT_SIZE) { 1366 args.guard.check = TRUE; 1367 args.guard.obj_ctime.seconds = va.va_ctime.tv_sec; 1368 args.guard.obj_ctime.nseconds = va.va_ctime.tv_nsec; 1369 } else 1370 args.guard.check = FALSE; 1371 1372 douprintf = 1; 1373 1374 t = gethrtime(); 1375 1376 error = rfs3call(VTOMI(vp), NFSPROC3_SETATTR, 1377 xdr_SETATTR3args, (caddr_t)&args, 1378 xdr_SETATTR3res, (caddr_t)&res, cr, 1379 &douprintf, &res.status, 0, NULL); 1380 1381 /* 1382 * Purge the access cache and ACL cache if changing either the 1383 * owner of the file, the group owner, or the mode. These may 1384 * change the access permissions of the file, so purge old 1385 * information and start over again. 1386 */ 1387 if (mask & (AT_UID | AT_GID | AT_MODE)) { 1388 (void) nfs_access_purge_rp(rp); 1389 if (rp->r_secattr != NULL) { 1390 mutex_enter(&rp->r_statelock); 1391 vsp = rp->r_secattr; 1392 rp->r_secattr = NULL; 1393 mutex_exit(&rp->r_statelock); 1394 if (vsp != NULL) 1395 nfs_acl_free(vsp); 1396 } 1397 } 1398 1399 if (error) { 1400 PURGE_ATTRCACHE(vp); 1401 return (error); 1402 } 1403 1404 error = geterrno3(res.status); 1405 if (!error) { 1406 /* 1407 * If changing the size of the file, invalidate 1408 * any local cached data which is no longer part 1409 * of the file. We also possibly invalidate the 1410 * last page in the file. We could use 1411 * pvn_vpzero(), but this would mark the page as 1412 * modified and require it to be written back to 1413 * the server for no particularly good reason. 1414 * This way, if we access it, then we bring it 1415 * back in. A read should be cheaper than a 1416 * write. 1417 */ 1418 if (mask & AT_SIZE) { 1419 nfs_invalidate_pages(vp, 1420 (vap->va_size & PAGEMASK), cr); 1421 } 1422 nfs3_cache_wcc_data(vp, &res.resok.obj_wcc, t, cr); 1423 /* 1424 * Some servers will change the mode to clear the setuid 1425 * and setgid bits when changing the uid or gid. The 1426 * client needs to compensate appropriately. 1427 */ 1428 if (mask & (AT_UID | AT_GID)) { 1429 int terror; 1430 1431 va.va_mask = AT_MODE; 1432 terror = nfs3getattr(vp, &va, cr); 1433 if (!terror && 1434 (((mask & AT_MODE) && va.va_mode != vap->va_mode) || 1435 (!(mask & AT_MODE) && va.va_mode != omode))) { 1436 va.va_mask = AT_MODE; 1437 if (mask & AT_MODE) 1438 va.va_mode = vap->va_mode; 1439 else 1440 va.va_mode = omode; 1441 (void) nfs3setattr(vp, &va, 0, cr); 1442 } 1443 } 1444 } else { 1445 nfs3_cache_wcc_data(vp, &res.resfail.obj_wcc, t, cr); 1446 /* 1447 * If we got back a "not synchronized" error, then 1448 * we need to retry with a new guard value. The 1449 * guard value used is the change time. If the 1450 * server returned post_op_attr, then we can just 1451 * retry because we have the latest attributes. 1452 * Otherwise, we issue a GETATTR to get the latest 1453 * attributes and then retry. If we couldn't get 1454 * the attributes this way either, then we give 1455 * up because we can't complete the operation as 1456 * required. 1457 */ 1458 if (res.status == NFS3ERR_NOT_SYNC) { 1459 va.va_mask = AT_CTIME; 1460 if (nfs3getattr(vp, &va, cr) == 0) 1461 goto tryagain; 1462 } 1463 PURGE_STALE_FH(error, vp, cr); 1464 } 1465 1466 return (error); 1467 } 1468 1469 static int 1470 nfs3_accessx(void *vp, int mode, cred_t *cr) 1471 { 1472 ASSERT(nfs_zone() == VTOMI((vnode_t *)vp)->mi_zone); 1473 return (nfs3_access(vp, mode, 0, cr, NULL)); 1474 } 1475 1476 /* ARGSUSED */ 1477 static int 1478 nfs3_access(vnode_t *vp, int mode, int flags, cred_t *cr, caller_context_t *ct) 1479 { 1480 int error; 1481 ACCESS3args args; 1482 ACCESS3res res; 1483 int douprintf; 1484 uint32 acc; 1485 rnode_t *rp; 1486 cred_t *cred, *ncr, *ncrfree = NULL; 1487 failinfo_t fi; 1488 nfs_access_type_t cacc; 1489 hrtime_t t; 1490 1491 acc = 0; 1492 if (nfs_zone() != VTOMI(vp)->mi_zone) 1493 return (EIO); 1494 if (mode & VREAD) 1495 acc |= ACCESS3_READ; 1496 if (mode & VWRITE) { 1497 if (vn_is_readonly(vp) && !IS_DEVVP(vp)) 1498 return (EROFS); 1499 if (vp->v_type == VDIR) 1500 acc |= ACCESS3_DELETE; 1501 acc |= ACCESS3_MODIFY | ACCESS3_EXTEND; 1502 } 1503 if (mode & VEXEC) { 1504 if (vp->v_type == VDIR) 1505 acc |= ACCESS3_LOOKUP; 1506 else 1507 acc |= ACCESS3_EXECUTE; 1508 } 1509 1510 rp = VTOR(vp); 1511 args.object = *VTOFH3(vp); 1512 if (vp->v_type == VDIR) { 1513 args.access = ACCESS3_READ | ACCESS3_DELETE | ACCESS3_MODIFY | 1514 ACCESS3_EXTEND | ACCESS3_LOOKUP; 1515 } else { 1516 args.access = ACCESS3_READ | ACCESS3_MODIFY | ACCESS3_EXTEND | 1517 ACCESS3_EXECUTE; 1518 } 1519 fi.vp = vp; 1520 fi.fhp = (caddr_t)&args.object; 1521 fi.copyproc = nfs3copyfh; 1522 fi.lookupproc = nfs3lookup; 1523 fi.xattrdirproc = acl_getxattrdir3; 1524 1525 cred = cr; 1526 /* 1527 * ncr and ncrfree both initially 1528 * point to the memory area returned 1529 * by crnetadjust(); 1530 * ncrfree not NULL when exiting means 1531 * that we need to release it 1532 */ 1533 ncr = crnetadjust(cred); 1534 ncrfree = ncr; 1535 tryagain: 1536 if (rp->r_acache != NULL) { 1537 cacc = nfs_access_check(rp, acc, cred); 1538 if (cacc == NFS_ACCESS_ALLOWED) { 1539 if (ncrfree != NULL) 1540 crfree(ncrfree); 1541 return (0); 1542 } 1543 if (cacc == NFS_ACCESS_DENIED) { 1544 /* 1545 * If the cred can be adjusted, try again 1546 * with the new cred. 1547 */ 1548 if (ncr != NULL) { 1549 cred = ncr; 1550 ncr = NULL; 1551 goto tryagain; 1552 } 1553 if (ncrfree != NULL) 1554 crfree(ncrfree); 1555 return (EACCES); 1556 } 1557 } 1558 1559 douprintf = 1; 1560 1561 t = gethrtime(); 1562 1563 error = rfs3call(VTOMI(vp), NFSPROC3_ACCESS, 1564 xdr_ACCESS3args, (caddr_t)&args, 1565 xdr_ACCESS3res, (caddr_t)&res, cred, 1566 &douprintf, &res.status, 0, &fi); 1567 1568 if (error) { 1569 if (ncrfree != NULL) 1570 crfree(ncrfree); 1571 return (error); 1572 } 1573 1574 error = geterrno3(res.status); 1575 if (!error) { 1576 nfs3_cache_post_op_attr(vp, &res.resok.obj_attributes, t, cr); 1577 nfs_access_cache(rp, args.access, res.resok.access, cred); 1578 /* 1579 * we just cached results with cred; if cred is the 1580 * adjusted credentials from crnetadjust, we do not want 1581 * to release them before exiting: hence setting ncrfree 1582 * to NULL 1583 */ 1584 if (cred != cr) 1585 ncrfree = NULL; 1586 if ((acc & res.resok.access) != acc) { 1587 /* 1588 * If the cred can be adjusted, try again 1589 * with the new cred. 1590 */ 1591 if (ncr != NULL) { 1592 cred = ncr; 1593 ncr = NULL; 1594 goto tryagain; 1595 } 1596 error = EACCES; 1597 } 1598 } else { 1599 nfs3_cache_post_op_attr(vp, &res.resfail.obj_attributes, t, cr); 1600 PURGE_STALE_FH(error, vp, cr); 1601 } 1602 1603 if (ncrfree != NULL) 1604 crfree(ncrfree); 1605 1606 return (error); 1607 } 1608 1609 static int nfs3_do_symlink_cache = 1; 1610 1611 /* ARGSUSED */ 1612 static int 1613 nfs3_readlink(vnode_t *vp, struct uio *uiop, cred_t *cr, caller_context_t *ct) 1614 { 1615 int error; 1616 READLINK3args args; 1617 READLINK3res res; 1618 nfspath3 resdata_backup; 1619 rnode_t *rp; 1620 int douprintf; 1621 int len; 1622 failinfo_t fi; 1623 hrtime_t t; 1624 1625 /* 1626 * Can't readlink anything other than a symbolic link. 1627 */ 1628 if (vp->v_type != VLNK) 1629 return (EINVAL); 1630 if (nfs_zone() != VTOMI(vp)->mi_zone) 1631 return (EIO); 1632 1633 rp = VTOR(vp); 1634 if (nfs3_do_symlink_cache && rp->r_symlink.contents != NULL) { 1635 error = nfs3_validate_caches(vp, cr); 1636 if (error) 1637 return (error); 1638 mutex_enter(&rp->r_statelock); 1639 if (rp->r_symlink.contents != NULL) { 1640 error = uiomove(rp->r_symlink.contents, 1641 rp->r_symlink.len, UIO_READ, uiop); 1642 mutex_exit(&rp->r_statelock); 1643 return (error); 1644 } 1645 mutex_exit(&rp->r_statelock); 1646 } 1647 1648 args.symlink = *VTOFH3(vp); 1649 fi.vp = vp; 1650 fi.fhp = (caddr_t)&args.symlink; 1651 fi.copyproc = nfs3copyfh; 1652 fi.lookupproc = nfs3lookup; 1653 fi.xattrdirproc = acl_getxattrdir3; 1654 1655 res.resok.data = kmem_alloc(MAXPATHLEN, KM_SLEEP); 1656 1657 resdata_backup = res.resok.data; 1658 1659 douprintf = 1; 1660 1661 t = gethrtime(); 1662 1663 error = rfs3call(VTOMI(vp), NFSPROC3_READLINK, 1664 xdr_nfs_fh3, (caddr_t)&args, 1665 xdr_READLINK3res, (caddr_t)&res, cr, 1666 &douprintf, &res.status, 0, &fi); 1667 1668 if (res.resok.data == nfs3nametoolong) 1669 error = EINVAL; 1670 1671 if (error) { 1672 kmem_free(resdata_backup, MAXPATHLEN); 1673 return (error); 1674 } 1675 1676 error = geterrno3(res.status); 1677 if (!error) { 1678 nfs3_cache_post_op_attr(vp, &res.resok.symlink_attributes, t, 1679 cr); 1680 len = strlen(res.resok.data); 1681 error = uiomove(res.resok.data, len, UIO_READ, uiop); 1682 if (nfs3_do_symlink_cache && rp->r_symlink.contents == NULL) { 1683 mutex_enter(&rp->r_statelock); 1684 if (rp->r_symlink.contents == NULL) { 1685 rp->r_symlink.contents = res.resok.data; 1686 rp->r_symlink.len = len; 1687 rp->r_symlink.size = MAXPATHLEN; 1688 mutex_exit(&rp->r_statelock); 1689 } else { 1690 mutex_exit(&rp->r_statelock); 1691 1692 kmem_free((void *)res.resok.data, MAXPATHLEN); 1693 } 1694 } else { 1695 kmem_free((void *)res.resok.data, MAXPATHLEN); 1696 } 1697 } else { 1698 nfs3_cache_post_op_attr(vp, 1699 &res.resfail.symlink_attributes, t, cr); 1700 PURGE_STALE_FH(error, vp, cr); 1701 1702 kmem_free((void *)res.resok.data, MAXPATHLEN); 1703 1704 } 1705 1706 /* 1707 * The over the wire error for attempting to readlink something 1708 * other than a symbolic link is ENXIO. However, we need to 1709 * return EINVAL instead of ENXIO, so we map it here. 1710 */ 1711 return (error == ENXIO ? EINVAL : error); 1712 } 1713 1714 /* 1715 * Flush local dirty pages to stable storage on the server. 1716 * 1717 * If FNODSYNC is specified, then there is nothing to do because 1718 * metadata changes are not cached on the client before being 1719 * sent to the server. 1720 */ 1721 /* ARGSUSED */ 1722 static int 1723 nfs3_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct) 1724 { 1725 int error; 1726 1727 if ((syncflag & FNODSYNC) || IS_SWAPVP(vp)) 1728 return (0); 1729 if (nfs_zone() != VTOMI(vp)->mi_zone) 1730 return (EIO); 1731 1732 error = nfs3_putpage_commit(vp, (offset_t)0, 0, cr); 1733 if (!error) 1734 error = VTOR(vp)->r_error; 1735 return (error); 1736 } 1737 1738 /* 1739 * Weirdness: if the file was removed or the target of a rename 1740 * operation while it was open, it got renamed instead. Here we 1741 * remove the renamed file. 1742 */ 1743 /* ARGSUSED */ 1744 static void 1745 nfs3_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) 1746 { 1747 rnode_t *rp; 1748 1749 ASSERT(vp != DNLC_NO_VNODE); 1750 1751 /* 1752 * If this is coming from the wrong zone, we let someone in the right 1753 * zone take care of it asynchronously. We can get here due to 1754 * VN_RELE() being called from pageout() or fsflush(). This call may 1755 * potentially turn into an expensive no-op if, for instance, v_count 1756 * gets incremented in the meantime, but it's still correct. 1757 */ 1758 if (nfs_zone() != VTOMI(vp)->mi_zone) { 1759 nfs_async_inactive(vp, cr, nfs3_inactive); 1760 return; 1761 } 1762 1763 rp = VTOR(vp); 1764 redo: 1765 if (rp->r_unldvp != NULL) { 1766 /* 1767 * Save the vnode pointer for the directory where the 1768 * unlinked-open file got renamed, then set it to NULL 1769 * to prevent another thread from getting here before 1770 * we're done with the remove. While we have the 1771 * statelock, make local copies of the pertinent rnode 1772 * fields. If we weren't to do this in an atomic way, the 1773 * the unl* fields could become inconsistent with respect 1774 * to each other due to a race condition between this 1775 * code and nfs_remove(). See bug report 1034328. 1776 */ 1777 mutex_enter(&rp->r_statelock); 1778 if (rp->r_unldvp != NULL) { 1779 vnode_t *unldvp; 1780 char *unlname; 1781 cred_t *unlcred; 1782 REMOVE3args args; 1783 REMOVE3res res; 1784 int douprintf; 1785 int error; 1786 hrtime_t t; 1787 1788 unldvp = rp->r_unldvp; 1789 rp->r_unldvp = NULL; 1790 unlname = rp->r_unlname; 1791 rp->r_unlname = NULL; 1792 unlcred = rp->r_unlcred; 1793 rp->r_unlcred = NULL; 1794 mutex_exit(&rp->r_statelock); 1795 1796 /* 1797 * If there are any dirty pages left, then flush 1798 * them. This is unfortunate because they just 1799 * may get thrown away during the remove operation, 1800 * but we have to do this for correctness. 1801 */ 1802 if (vn_has_cached_data(vp) && 1803 ((rp->r_flags & RDIRTY) || rp->r_count > 0)) { 1804 ASSERT(vp->v_type != VCHR); 1805 error = nfs3_putpage(vp, (offset_t)0, 0, 0, 1806 cr, ct); 1807 if (error) { 1808 mutex_enter(&rp->r_statelock); 1809 if (!rp->r_error) 1810 rp->r_error = error; 1811 mutex_exit(&rp->r_statelock); 1812 } 1813 } 1814 1815 /* 1816 * Do the remove operation on the renamed file 1817 */ 1818 setdiropargs3(&args.object, unlname, unldvp); 1819 1820 douprintf = 1; 1821 1822 t = gethrtime(); 1823 1824 error = rfs3call(VTOMI(unldvp), NFSPROC3_REMOVE, 1825 xdr_diropargs3, (caddr_t)&args, 1826 xdr_REMOVE3res, (caddr_t)&res, unlcred, 1827 &douprintf, &res.status, 0, NULL); 1828 1829 if (error) { 1830 PURGE_ATTRCACHE(unldvp); 1831 } else { 1832 error = geterrno3(res.status); 1833 if (!error) { 1834 nfs3_cache_wcc_data(unldvp, 1835 &res.resok.dir_wcc, t, cr); 1836 if (HAVE_RDDIR_CACHE(VTOR(unldvp))) 1837 nfs_purge_rddir_cache(unldvp); 1838 } else { 1839 nfs3_cache_wcc_data(unldvp, 1840 &res.resfail.dir_wcc, t, cr); 1841 PURGE_STALE_FH(error, unldvp, cr); 1842 } 1843 } 1844 1845 /* 1846 * Release stuff held for the remove 1847 */ 1848 VN_RELE(unldvp); 1849 kmem_free(unlname, MAXNAMELEN); 1850 crfree(unlcred); 1851 goto redo; 1852 } 1853 mutex_exit(&rp->r_statelock); 1854 } 1855 1856 rp_addfree(rp, cr); 1857 } 1858 1859 /* 1860 * Remote file system operations having to do with directory manipulation. 1861 */ 1862 1863 /* ARGSUSED */ 1864 static int 1865 nfs3_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp, 1866 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct, 1867 int *direntflags, pathname_t *realpnp) 1868 { 1869 int error; 1870 vnode_t *vp; 1871 vnode_t *avp = NULL; 1872 rnode_t *drp; 1873 1874 if (nfs_zone() != VTOMI(dvp)->mi_zone) 1875 return (EPERM); 1876 1877 drp = VTOR(dvp); 1878 1879 /* 1880 * Are we looking up extended attributes? If so, "dvp" is 1881 * the file or directory for which we want attributes, and 1882 * we need a lookup of the hidden attribute directory 1883 * before we lookup the rest of the path. 1884 */ 1885 if (flags & LOOKUP_XATTR) { 1886 bool_t cflag = ((flags & CREATE_XATTR_DIR) != 0); 1887 mntinfo_t *mi; 1888 1889 mi = VTOMI(dvp); 1890 if (!(mi->mi_flags & MI_EXTATTR)) 1891 return (EINVAL); 1892 1893 if (nfs_rw_enter_sig(&drp->r_rwlock, RW_READER, INTR(dvp))) 1894 return (EINTR); 1895 1896 (void) nfs3lookup_dnlc(dvp, XATTR_DIR_NAME, &avp, cr); 1897 if (avp == NULL) 1898 error = acl_getxattrdir3(dvp, &avp, cflag, cr, 0); 1899 else 1900 error = 0; 1901 1902 nfs_rw_exit(&drp->r_rwlock); 1903 1904 if (error) { 1905 if (mi->mi_flags & MI_EXTATTR) 1906 return (error); 1907 return (EINVAL); 1908 } 1909 dvp = avp; 1910 drp = VTOR(dvp); 1911 } 1912 1913 if (nfs_rw_enter_sig(&drp->r_rwlock, RW_READER, INTR(dvp))) { 1914 error = EINTR; 1915 goto out; 1916 } 1917 1918 error = nfs3lookup(dvp, nm, vpp, pnp, flags, rdir, cr, 0); 1919 1920 nfs_rw_exit(&drp->r_rwlock); 1921 1922 /* 1923 * If vnode is a device, create special vnode. 1924 */ 1925 if (!error && IS_DEVVP(*vpp)) { 1926 vp = *vpp; 1927 *vpp = specvp(vp, vp->v_rdev, vp->v_type, cr); 1928 VN_RELE(vp); 1929 } 1930 1931 out: 1932 if (avp != NULL) 1933 VN_RELE(avp); 1934 1935 return (error); 1936 } 1937 1938 static int nfs3_lookup_neg_cache = 1; 1939 1940 #ifdef DEBUG 1941 static int nfs3_lookup_dnlc_hits = 0; 1942 static int nfs3_lookup_dnlc_misses = 0; 1943 static int nfs3_lookup_dnlc_neg_hits = 0; 1944 static int nfs3_lookup_dnlc_disappears = 0; 1945 static int nfs3_lookup_dnlc_lookups = 0; 1946 #endif 1947 1948 /* ARGSUSED */ 1949 int 1950 nfs3lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp, 1951 int flags, vnode_t *rdir, cred_t *cr, int rfscall_flags) 1952 { 1953 int error; 1954 rnode_t *drp; 1955 1956 ASSERT(nfs_zone() == VTOMI(dvp)->mi_zone); 1957 /* 1958 * If lookup is for "", just return dvp. Don't need 1959 * to send it over the wire, look it up in the dnlc, 1960 * or perform any access checks. 1961 */ 1962 if (*nm == '\0') { 1963 VN_HOLD(dvp); 1964 *vpp = dvp; 1965 return (0); 1966 } 1967 1968 /* 1969 * Can't do lookups in non-directories. 1970 */ 1971 if (dvp->v_type != VDIR) 1972 return (ENOTDIR); 1973 1974 /* 1975 * If we're called with RFSCALL_SOFT, it's important that 1976 * the only rfscall is one we make directly; if we permit 1977 * an access call because we're looking up "." or validating 1978 * a dnlc hit, we'll deadlock because that rfscall will not 1979 * have the RFSCALL_SOFT set. 1980 */ 1981 if (rfscall_flags & RFSCALL_SOFT) 1982 goto callit; 1983 1984 /* 1985 * If lookup is for ".", just return dvp. Don't need 1986 * to send it over the wire or look it up in the dnlc, 1987 * just need to check access. 1988 */ 1989 if (strcmp(nm, ".") == 0) { 1990 error = nfs3_access(dvp, VEXEC, 0, cr, NULL); 1991 if (error) 1992 return (error); 1993 VN_HOLD(dvp); 1994 *vpp = dvp; 1995 return (0); 1996 } 1997 1998 drp = VTOR(dvp); 1999 if (!(drp->r_flags & RLOOKUP)) { 2000 mutex_enter(&drp->r_statelock); 2001 drp->r_flags |= RLOOKUP; 2002 mutex_exit(&drp->r_statelock); 2003 } 2004 2005 /* 2006 * Lookup this name in the DNLC. If there was a valid entry, 2007 * then return the results of the lookup. 2008 */ 2009 error = nfs3lookup_dnlc(dvp, nm, vpp, cr); 2010 if (error || *vpp != NULL) 2011 return (error); 2012 2013 callit: 2014 error = nfs3lookup_otw(dvp, nm, vpp, cr, rfscall_flags); 2015 2016 return (error); 2017 } 2018 2019 static int 2020 nfs3lookup_dnlc(vnode_t *dvp, char *nm, vnode_t **vpp, cred_t *cr) 2021 { 2022 int error; 2023 vnode_t *vp; 2024 2025 ASSERT(*nm != '\0'); 2026 ASSERT(nfs_zone() == VTOMI(dvp)->mi_zone); 2027 /* 2028 * Lookup this name in the DNLC. If successful, then validate 2029 * the caches and then recheck the DNLC. The DNLC is rechecked 2030 * just in case this entry got invalidated during the call 2031 * to nfs3_validate_caches. 2032 * 2033 * An assumption is being made that it is safe to say that a 2034 * file exists which may not on the server. Any operations to 2035 * the server will fail with ESTALE. 2036 */ 2037 #ifdef DEBUG 2038 nfs3_lookup_dnlc_lookups++; 2039 #endif 2040 vp = dnlc_lookup(dvp, nm); 2041 if (vp != NULL) { 2042 VN_RELE(vp); 2043 if (vp == DNLC_NO_VNODE && !vn_is_readonly(dvp)) { 2044 PURGE_ATTRCACHE(dvp); 2045 } 2046 error = nfs3_validate_caches(dvp, cr); 2047 if (error) 2048 return (error); 2049 vp = dnlc_lookup(dvp, nm); 2050 if (vp != NULL) { 2051 error = nfs3_access(dvp, VEXEC, 0, cr, NULL); 2052 if (error) { 2053 VN_RELE(vp); 2054 return (error); 2055 } 2056 if (vp == DNLC_NO_VNODE) { 2057 VN_RELE(vp); 2058 #ifdef DEBUG 2059 nfs3_lookup_dnlc_neg_hits++; 2060 #endif 2061 return (ENOENT); 2062 } 2063 *vpp = vp; 2064 #ifdef DEBUG 2065 nfs3_lookup_dnlc_hits++; 2066 #endif 2067 return (0); 2068 } 2069 #ifdef DEBUG 2070 nfs3_lookup_dnlc_disappears++; 2071 #endif 2072 } 2073 #ifdef DEBUG 2074 else 2075 nfs3_lookup_dnlc_misses++; 2076 #endif 2077 2078 *vpp = NULL; 2079 2080 return (0); 2081 } 2082 2083 static int 2084 nfs3lookup_otw(vnode_t *dvp, char *nm, vnode_t **vpp, cred_t *cr, 2085 int rfscall_flags) 2086 { 2087 int error; 2088 LOOKUP3args args; 2089 LOOKUP3vres res; 2090 int douprintf; 2091 struct vattr vattr; 2092 struct vattr dvattr; 2093 vnode_t *vp; 2094 failinfo_t fi; 2095 hrtime_t t; 2096 2097 ASSERT(*nm != '\0'); 2098 ASSERT(dvp->v_type == VDIR); 2099 ASSERT(nfs_zone() == VTOMI(dvp)->mi_zone); 2100 2101 setdiropargs3(&args.what, nm, dvp); 2102 2103 fi.vp = dvp; 2104 fi.fhp = (caddr_t)&args.what.dir; 2105 fi.copyproc = nfs3copyfh; 2106 fi.lookupproc = nfs3lookup; 2107 fi.xattrdirproc = acl_getxattrdir3; 2108 res.obj_attributes.fres.vp = dvp; 2109 res.obj_attributes.fres.vap = &vattr; 2110 res.dir_attributes.fres.vp = dvp; 2111 res.dir_attributes.fres.vap = &dvattr; 2112 2113 douprintf = 1; 2114 2115 t = gethrtime(); 2116 2117 error = rfs3call(VTOMI(dvp), NFSPROC3_LOOKUP, 2118 xdr_diropargs3, (caddr_t)&args, 2119 xdr_LOOKUP3vres, (caddr_t)&res, cr, 2120 &douprintf, &res.status, rfscall_flags, &fi); 2121 2122 if (error) 2123 return (error); 2124 2125 nfs3_cache_post_op_vattr(dvp, &res.dir_attributes, t, cr); 2126 2127 error = geterrno3(res.status); 2128 if (error) { 2129 PURGE_STALE_FH(error, dvp, cr); 2130 if (error == ENOENT && nfs3_lookup_neg_cache) 2131 dnlc_enter(dvp, nm, DNLC_NO_VNODE); 2132 return (error); 2133 } 2134 2135 if (res.obj_attributes.attributes) { 2136 vp = makenfs3node_va(&res.object, res.obj_attributes.fres.vap, 2137 dvp->v_vfsp, t, cr, VTOR(dvp)->r_path, nm); 2138 } else { 2139 vp = makenfs3node_va(&res.object, NULL, 2140 dvp->v_vfsp, t, cr, VTOR(dvp)->r_path, nm); 2141 if (vp->v_type == VNON) { 2142 vattr.va_mask = AT_TYPE; 2143 error = nfs3getattr(vp, &vattr, cr); 2144 if (error) { 2145 VN_RELE(vp); 2146 return (error); 2147 } 2148 vp->v_type = vattr.va_type; 2149 } 2150 } 2151 2152 if (!(rfscall_flags & RFSCALL_SOFT)) 2153 dnlc_update(dvp, nm, vp); 2154 2155 *vpp = vp; 2156 2157 return (error); 2158 } 2159 2160 #ifdef DEBUG 2161 static int nfs3_create_misses = 0; 2162 #endif 2163 2164 /* ARGSUSED */ 2165 static int 2166 nfs3_create(vnode_t *dvp, char *nm, struct vattr *va, enum vcexcl exclusive, 2167 int mode, vnode_t **vpp, cred_t *cr, int lfaware, caller_context_t *ct, 2168 vsecattr_t *vsecp) 2169 { 2170 int error; 2171 vnode_t *vp; 2172 rnode_t *rp; 2173 struct vattr vattr; 2174 rnode_t *drp; 2175 vnode_t *tempvp; 2176 2177 drp = VTOR(dvp); 2178 if (nfs_zone() != VTOMI(dvp)->mi_zone) 2179 return (EPERM); 2180 if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp))) 2181 return (EINTR); 2182 2183 top: 2184 /* 2185 * We make a copy of the attributes because the caller does not 2186 * expect us to change what va points to. 2187 */ 2188 vattr = *va; 2189 2190 /* 2191 * If the pathname is "", just use dvp. Don't need 2192 * to send it over the wire, look it up in the dnlc, 2193 * or perform any access checks. 2194 */ 2195 if (*nm == '\0') { 2196 error = 0; 2197 VN_HOLD(dvp); 2198 vp = dvp; 2199 /* 2200 * If the pathname is ".", just use dvp. Don't need 2201 * to send it over the wire or look it up in the dnlc, 2202 * just need to check access. 2203 */ 2204 } else if (strcmp(nm, ".") == 0) { 2205 error = nfs3_access(dvp, VEXEC, 0, cr, ct); 2206 if (error) { 2207 nfs_rw_exit(&drp->r_rwlock); 2208 return (error); 2209 } 2210 VN_HOLD(dvp); 2211 vp = dvp; 2212 /* 2213 * We need to go over the wire, just to be sure whether the 2214 * file exists or not. Using the DNLC can be dangerous in 2215 * this case when making a decision regarding existence. 2216 */ 2217 } else { 2218 error = nfs3lookup_otw(dvp, nm, &vp, cr, 0); 2219 } 2220 if (!error) { 2221 if (exclusive == EXCL) 2222 error = EEXIST; 2223 else if (vp->v_type == VDIR && (mode & VWRITE)) 2224 error = EISDIR; 2225 else { 2226 /* 2227 * If vnode is a device, create special vnode. 2228 */ 2229 if (IS_DEVVP(vp)) { 2230 tempvp = vp; 2231 vp = specvp(vp, vp->v_rdev, vp->v_type, cr); 2232 VN_RELE(tempvp); 2233 } 2234 if (!(error = VOP_ACCESS(vp, mode, 0, cr, ct))) { 2235 if ((vattr.va_mask & AT_SIZE) && 2236 vp->v_type == VREG) { 2237 rp = VTOR(vp); 2238 /* 2239 * Check here for large file handled 2240 * by LF-unaware process (as 2241 * ufs_create() does) 2242 */ 2243 if (!(lfaware & FOFFMAX)) { 2244 mutex_enter(&rp->r_statelock); 2245 if (rp->r_size > MAXOFF32_T) 2246 error = EOVERFLOW; 2247 mutex_exit(&rp->r_statelock); 2248 } 2249 if (!error) { 2250 vattr.va_mask = AT_SIZE; 2251 error = nfs3setattr(vp, 2252 &vattr, 0, cr); 2253 } 2254 } 2255 } 2256 } 2257 nfs_rw_exit(&drp->r_rwlock); 2258 if (error) { 2259 VN_RELE(vp); 2260 } else { 2261 /* 2262 * existing file got truncated, notify. 2263 */ 2264 vnevent_create(vp, ct); 2265 *vpp = vp; 2266 } 2267 return (error); 2268 } 2269 2270 dnlc_remove(dvp, nm); 2271 2272 /* 2273 * Decide what the group-id of the created file should be. 2274 * Set it in attribute list as advisory... 2275 */ 2276 error = setdirgid(dvp, &vattr.va_gid, cr); 2277 if (error) { 2278 nfs_rw_exit(&drp->r_rwlock); 2279 return (error); 2280 } 2281 vattr.va_mask |= AT_GID; 2282 2283 ASSERT(vattr.va_mask & AT_TYPE); 2284 if (vattr.va_type == VREG) { 2285 ASSERT(vattr.va_mask & AT_MODE); 2286 if (MANDMODE(vattr.va_mode)) { 2287 nfs_rw_exit(&drp->r_rwlock); 2288 return (EACCES); 2289 } 2290 error = nfs3create(dvp, nm, &vattr, exclusive, mode, vpp, cr, 2291 lfaware); 2292 /* 2293 * If this is not an exclusive create, then the CREATE 2294 * request will be made with the GUARDED mode set. This 2295 * means that the server will return EEXIST if the file 2296 * exists. The file could exist because of a retransmitted 2297 * request. In this case, we recover by starting over and 2298 * checking to see whether the file exists. This second 2299 * time through it should and a CREATE request will not be 2300 * sent. 2301 * 2302 * This handles the problem of a dangling CREATE request 2303 * which contains attributes which indicate that the file 2304 * should be truncated. This retransmitted request could 2305 * possibly truncate valid data in the file if not caught 2306 * by the duplicate request mechanism on the server or if 2307 * not caught by other means. The scenario is: 2308 * 2309 * Client transmits CREATE request with size = 0 2310 * Client times out, retransmits request. 2311 * Response to the first request arrives from the server 2312 * and the client proceeds on. 2313 * Client writes data to the file. 2314 * The server now processes retransmitted CREATE request 2315 * and truncates file. 2316 * 2317 * The use of the GUARDED CREATE request prevents this from 2318 * happening because the retransmitted CREATE would fail 2319 * with EEXIST and would not truncate the file. 2320 */ 2321 if (error == EEXIST && exclusive == NONEXCL) { 2322 #ifdef DEBUG 2323 nfs3_create_misses++; 2324 #endif 2325 goto top; 2326 } 2327 nfs_rw_exit(&drp->r_rwlock); 2328 return (error); 2329 } 2330 error = nfs3mknod(dvp, nm, &vattr, exclusive, mode, vpp, cr); 2331 nfs_rw_exit(&drp->r_rwlock); 2332 return (error); 2333 } 2334 2335 /* ARGSUSED */ 2336 static int 2337 nfs3create(vnode_t *dvp, char *nm, struct vattr *va, enum vcexcl exclusive, 2338 int mode, vnode_t **vpp, cred_t *cr, int lfaware) 2339 { 2340 int error; 2341 CREATE3args args; 2342 CREATE3res res; 2343 int douprintf; 2344 vnode_t *vp; 2345 struct vattr vattr; 2346 nfstime3 *verfp; 2347 rnode_t *rp; 2348 timestruc_t now; 2349 hrtime_t t; 2350 2351 ASSERT(nfs_zone() == VTOMI(dvp)->mi_zone); 2352 setdiropargs3(&args.where, nm, dvp); 2353 if (exclusive == EXCL) { 2354 args.how.mode = EXCLUSIVE; 2355 /* 2356 * Construct the create verifier. This verifier needs 2357 * to be unique between different clients. It also needs 2358 * to vary for each exclusive create request generated 2359 * from the client to the server. 2360 * 2361 * The first attempt is made to use the hostid and a 2362 * unique number on the client. If the hostid has not 2363 * been set, the high resolution time that the exclusive 2364 * create request is being made is used. This will work 2365 * unless two different clients, both with the hostid 2366 * not set, attempt an exclusive create request on the 2367 * same file, at exactly the same clock time. The 2368 * chances of this happening seem small enough to be 2369 * reasonable. 2370 */ 2371 verfp = (nfstime3 *)&args.how.createhow3_u.verf; 2372 verfp->seconds = nfs_atoi(hw_serial); 2373 if (verfp->seconds != 0) 2374 verfp->nseconds = newnum(); 2375 else { 2376 gethrestime(&now); 2377 verfp->seconds = now.tv_sec; 2378 verfp->nseconds = now.tv_nsec; 2379 } 2380 /* 2381 * Since the server will use this value for the mtime, 2382 * make sure that it can't overflow. Zero out the MSB. 2383 * The actual value does not matter here, only its uniqeness. 2384 */ 2385 verfp->seconds %= INT32_MAX; 2386 } else { 2387 /* 2388 * Issue the non-exclusive create in guarded mode. This 2389 * may result in some false EEXIST responses for 2390 * retransmitted requests, but these will be handled at 2391 * a higher level. By using GUARDED, duplicate requests 2392 * to do file truncation and possible access problems 2393 * can be avoided. 2394 */ 2395 args.how.mode = GUARDED; 2396 error = vattr_to_sattr3(va, 2397 &args.how.createhow3_u.obj_attributes); 2398 if (error) { 2399 /* req time field(s) overflow - return immediately */ 2400 return (error); 2401 } 2402 } 2403 2404 douprintf = 1; 2405 2406 t = gethrtime(); 2407 2408 error = rfs3call(VTOMI(dvp), NFSPROC3_CREATE, 2409 xdr_CREATE3args, (caddr_t)&args, 2410 xdr_CREATE3res, (caddr_t)&res, cr, 2411 &douprintf, &res.status, 0, NULL); 2412 2413 if (error) { 2414 PURGE_ATTRCACHE(dvp); 2415 return (error); 2416 } 2417 2418 error = geterrno3(res.status); 2419 if (!error) { 2420 nfs3_cache_wcc_data(dvp, &res.resok.dir_wcc, t, cr); 2421 if (HAVE_RDDIR_CACHE(VTOR(dvp))) 2422 nfs_purge_rddir_cache(dvp); 2423 2424 /* 2425 * On exclusive create the times need to be explicitly 2426 * set to clear any potential verifier that may be stored 2427 * in one of these fields (see comment below). This 2428 * is done here to cover the case where no post op attrs 2429 * were returned or a 'invalid' time was returned in 2430 * the attributes. 2431 */ 2432 if (exclusive == EXCL) 2433 va->va_mask |= (AT_MTIME | AT_ATIME); 2434 2435 if (!res.resok.obj.handle_follows) { 2436 error = nfs3lookup(dvp, nm, &vp, NULL, 0, NULL, cr, 0); 2437 if (error) 2438 return (error); 2439 } else { 2440 if (res.resok.obj_attributes.attributes) { 2441 vp = makenfs3node(&res.resok.obj.handle, 2442 &res.resok.obj_attributes.attr, 2443 dvp->v_vfsp, t, cr, NULL, NULL); 2444 } else { 2445 vp = makenfs3node(&res.resok.obj.handle, NULL, 2446 dvp->v_vfsp, t, cr, NULL, NULL); 2447 2448 /* 2449 * On an exclusive create, it is possible 2450 * that attributes were returned but those 2451 * postop attributes failed to decode 2452 * properly. If this is the case, 2453 * then most likely the atime or mtime 2454 * were invalid for our client; this 2455 * is caused by the server storing the 2456 * create verifier in one of the time 2457 * fields(most likely mtime). 2458 * So... we are going to setattr just the 2459 * atime/mtime to clear things up. 2460 */ 2461 if (exclusive == EXCL) { 2462 if (error = 2463 nfs3excl_create_settimes(vp, 2464 va, cr)) { 2465 /* 2466 * Setting the times failed. 2467 * Remove the file and return 2468 * the error. 2469 */ 2470 VN_RELE(vp); 2471 (void) nfs3_remove(dvp, 2472 nm, cr, NULL, 0); 2473 return (error); 2474 } 2475 } 2476 2477 /* 2478 * This handles the non-exclusive case 2479 * and the exclusive case where no post op 2480 * attrs were returned. 2481 */ 2482 if (vp->v_type == VNON) { 2483 vattr.va_mask = AT_TYPE; 2484 error = nfs3getattr(vp, &vattr, cr); 2485 if (error) { 2486 VN_RELE(vp); 2487 return (error); 2488 } 2489 vp->v_type = vattr.va_type; 2490 } 2491 } 2492 dnlc_update(dvp, nm, vp); 2493 } 2494 2495 rp = VTOR(vp); 2496 2497 /* 2498 * Check here for large file handled by 2499 * LF-unaware process (as ufs_create() does) 2500 */ 2501 if ((va->va_mask & AT_SIZE) && vp->v_type == VREG && 2502 !(lfaware & FOFFMAX)) { 2503 mutex_enter(&rp->r_statelock); 2504 if (rp->r_size > MAXOFF32_T) { 2505 mutex_exit(&rp->r_statelock); 2506 VN_RELE(vp); 2507 return (EOVERFLOW); 2508 } 2509 mutex_exit(&rp->r_statelock); 2510 } 2511 2512 if (exclusive == EXCL && 2513 (va->va_mask & ~(AT_GID | AT_SIZE))) { 2514 /* 2515 * If doing an exclusive create, then generate 2516 * a SETATTR to set the initial attributes. 2517 * Try to set the mtime and the atime to the 2518 * server's current time. It is somewhat 2519 * expected that these fields will be used to 2520 * store the exclusive create cookie. If not, 2521 * server implementors will need to know that 2522 * a SETATTR will follow an exclusive create 2523 * and the cookie should be destroyed if 2524 * appropriate. This work may have been done 2525 * earlier in this function if post op attrs 2526 * were not available. 2527 * 2528 * The AT_GID and AT_SIZE bits are turned off 2529 * so that the SETATTR request will not attempt 2530 * to process these. The gid will be set 2531 * separately if appropriate. The size is turned 2532 * off because it is assumed that a new file will 2533 * be created empty and if the file wasn't empty, 2534 * then the exclusive create will have failed 2535 * because the file must have existed already. 2536 * Therefore, no truncate operation is needed. 2537 */ 2538 va->va_mask &= ~(AT_GID | AT_SIZE); 2539 error = nfs3setattr(vp, va, 0, cr); 2540 if (error) { 2541 /* 2542 * Couldn't correct the attributes of 2543 * the newly created file and the 2544 * attributes are wrong. Remove the 2545 * file and return an error to the 2546 * application. 2547 */ 2548 VN_RELE(vp); 2549 (void) nfs3_remove(dvp, nm, cr, NULL, 0); 2550 return (error); 2551 } 2552 } 2553 2554 if (va->va_gid != rp->r_attr.va_gid) { 2555 /* 2556 * If the gid on the file isn't right, then 2557 * generate a SETATTR to attempt to change 2558 * it. This may or may not work, depending 2559 * upon the server's semantics for allowing 2560 * file ownership changes. 2561 */ 2562 va->va_mask = AT_GID; 2563 (void) nfs3setattr(vp, va, 0, cr); 2564 } 2565 2566 /* 2567 * If vnode is a device create special vnode 2568 */ 2569 if (IS_DEVVP(vp)) { 2570 *vpp = specvp(vp, vp->v_rdev, vp->v_type, cr); 2571 VN_RELE(vp); 2572 } else 2573 *vpp = vp; 2574 } else { 2575 nfs3_cache_wcc_data(dvp, &res.resfail.dir_wcc, t, cr); 2576 PURGE_STALE_FH(error, dvp, cr); 2577 } 2578 2579 return (error); 2580 } 2581 2582 /* 2583 * Special setattr function to take care of rest of atime/mtime 2584 * after successful exclusive create. This function exists to avoid 2585 * handling attributes from the server; exclusive the atime/mtime fields 2586 * may be 'invalid' in client's view and therefore can not be trusted. 2587 */ 2588 static int 2589 nfs3excl_create_settimes(vnode_t *vp, struct vattr *vap, cred_t *cr) 2590 { 2591 int error; 2592 uint_t mask; 2593 SETATTR3args args; 2594 SETATTR3res res; 2595 int douprintf; 2596 rnode_t *rp; 2597 hrtime_t t; 2598 2599 ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 2600 /* save the caller's mask so that it can be reset later */ 2601 mask = vap->va_mask; 2602 2603 rp = VTOR(vp); 2604 2605 args.object = *RTOFH3(rp); 2606 args.guard.check = FALSE; 2607 2608 /* Use the mask to initialize the arguments */ 2609 vap->va_mask = 0; 2610 error = vattr_to_sattr3(vap, &args.new_attributes); 2611 2612 /* We want to set just atime/mtime on this request */ 2613 args.new_attributes.atime.set_it = SET_TO_SERVER_TIME; 2614 args.new_attributes.mtime.set_it = SET_TO_SERVER_TIME; 2615 2616 douprintf = 1; 2617 2618 t = gethrtime(); 2619 2620 error = rfs3call(VTOMI(vp), NFSPROC3_SETATTR, 2621 xdr_SETATTR3args, (caddr_t)&args, 2622 xdr_SETATTR3res, (caddr_t)&res, cr, 2623 &douprintf, &res.status, 0, NULL); 2624 2625 if (error) { 2626 vap->va_mask = mask; 2627 return (error); 2628 } 2629 2630 error = geterrno3(res.status); 2631 if (!error) { 2632 /* 2633 * It is important to pick up the attributes. 2634 * Since this is the exclusive create path, the 2635 * attributes on the initial create were ignored 2636 * and we need these to have the correct info. 2637 */ 2638 nfs3_cache_wcc_data(vp, &res.resok.obj_wcc, t, cr); 2639 /* 2640 * No need to do the atime/mtime work again so clear 2641 * the bits. 2642 */ 2643 mask &= ~(AT_ATIME | AT_MTIME); 2644 } else { 2645 nfs3_cache_wcc_data(vp, &res.resfail.obj_wcc, t, cr); 2646 } 2647 2648 vap->va_mask = mask; 2649 2650 return (error); 2651 } 2652 2653 /* ARGSUSED */ 2654 static int 2655 nfs3mknod(vnode_t *dvp, char *nm, struct vattr *va, enum vcexcl exclusive, 2656 int mode, vnode_t **vpp, cred_t *cr) 2657 { 2658 int error; 2659 MKNOD3args args; 2660 MKNOD3res res; 2661 int douprintf; 2662 vnode_t *vp; 2663 struct vattr vattr; 2664 hrtime_t t; 2665 2666 ASSERT(nfs_zone() == VTOMI(dvp)->mi_zone); 2667 switch (va->va_type) { 2668 case VCHR: 2669 case VBLK: 2670 setdiropargs3(&args.where, nm, dvp); 2671 args.what.type = (va->va_type == VCHR) ? NF3CHR : NF3BLK; 2672 error = vattr_to_sattr3(va, 2673 &args.what.mknoddata3_u.device.dev_attributes); 2674 if (error) { 2675 /* req time field(s) overflow - return immediately */ 2676 return (error); 2677 } 2678 args.what.mknoddata3_u.device.spec.specdata1 = 2679 getmajor(va->va_rdev); 2680 args.what.mknoddata3_u.device.spec.specdata2 = 2681 getminor(va->va_rdev); 2682 break; 2683 2684 case VFIFO: 2685 case VSOCK: 2686 setdiropargs3(&args.where, nm, dvp); 2687 args.what.type = (va->va_type == VFIFO) ? NF3FIFO : NF3SOCK; 2688 error = vattr_to_sattr3(va, 2689 &args.what.mknoddata3_u.pipe_attributes); 2690 if (error) { 2691 /* req time field(s) overflow - return immediately */ 2692 return (error); 2693 } 2694 break; 2695 2696 default: 2697 return (EINVAL); 2698 } 2699 2700 douprintf = 1; 2701 2702 t = gethrtime(); 2703 2704 error = rfs3call(VTOMI(dvp), NFSPROC3_MKNOD, 2705 xdr_MKNOD3args, (caddr_t)&args, 2706 xdr_MKNOD3res, (caddr_t)&res, cr, 2707 &douprintf, &res.status, 0, NULL); 2708 2709 if (error) { 2710 PURGE_ATTRCACHE(dvp); 2711 return (error); 2712 } 2713 2714 error = geterrno3(res.status); 2715 if (!error) { 2716 nfs3_cache_wcc_data(dvp, &res.resok.dir_wcc, t, cr); 2717 if (HAVE_RDDIR_CACHE(VTOR(dvp))) 2718 nfs_purge_rddir_cache(dvp); 2719 2720 if (!res.resok.obj.handle_follows) { 2721 error = nfs3lookup(dvp, nm, &vp, NULL, 0, NULL, cr, 0); 2722 if (error) 2723 return (error); 2724 } else { 2725 if (res.resok.obj_attributes.attributes) { 2726 vp = makenfs3node(&res.resok.obj.handle, 2727 &res.resok.obj_attributes.attr, 2728 dvp->v_vfsp, t, cr, NULL, NULL); 2729 } else { 2730 vp = makenfs3node(&res.resok.obj.handle, NULL, 2731 dvp->v_vfsp, t, cr, NULL, NULL); 2732 if (vp->v_type == VNON) { 2733 vattr.va_mask = AT_TYPE; 2734 error = nfs3getattr(vp, &vattr, cr); 2735 if (error) { 2736 VN_RELE(vp); 2737 return (error); 2738 } 2739 vp->v_type = vattr.va_type; 2740 } 2741 2742 } 2743 dnlc_update(dvp, nm, vp); 2744 } 2745 2746 if (va->va_gid != VTOR(vp)->r_attr.va_gid) { 2747 va->va_mask = AT_GID; 2748 (void) nfs3setattr(vp, va, 0, cr); 2749 } 2750 2751 /* 2752 * If vnode is a device create special vnode 2753 */ 2754 if (IS_DEVVP(vp)) { 2755 *vpp = specvp(vp, vp->v_rdev, vp->v_type, cr); 2756 VN_RELE(vp); 2757 } else 2758 *vpp = vp; 2759 } else { 2760 nfs3_cache_wcc_data(dvp, &res.resfail.dir_wcc, t, cr); 2761 PURGE_STALE_FH(error, dvp, cr); 2762 } 2763 return (error); 2764 } 2765 2766 /* 2767 * Weirdness: if the vnode to be removed is open 2768 * we rename it instead of removing it and nfs_inactive 2769 * will remove the new name. 2770 */ 2771 /* ARGSUSED */ 2772 static int 2773 nfs3_remove(vnode_t *dvp, char *nm, cred_t *cr, caller_context_t *ct, int flags) 2774 { 2775 int error; 2776 REMOVE3args args; 2777 REMOVE3res res; 2778 vnode_t *vp; 2779 char *tmpname; 2780 int douprintf; 2781 rnode_t *rp; 2782 rnode_t *drp; 2783 hrtime_t t; 2784 2785 if (nfs_zone() != VTOMI(dvp)->mi_zone) 2786 return (EPERM); 2787 drp = VTOR(dvp); 2788 if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp))) 2789 return (EINTR); 2790 2791 error = nfs3lookup(dvp, nm, &vp, NULL, 0, NULL, cr, 0); 2792 if (error) { 2793 nfs_rw_exit(&drp->r_rwlock); 2794 return (error); 2795 } 2796 2797 if (vp->v_type == VDIR && secpolicy_fs_linkdir(cr, dvp->v_vfsp)) { 2798 VN_RELE(vp); 2799 nfs_rw_exit(&drp->r_rwlock); 2800 return (EPERM); 2801 } 2802 2803 /* 2804 * First just remove the entry from the name cache, as it 2805 * is most likely the only entry for this vp. 2806 */ 2807 dnlc_remove(dvp, nm); 2808 2809 /* 2810 * If the file has a v_count > 1 then there may be more than one 2811 * entry in the name cache due multiple links or an open file, 2812 * but we don't have the real reference count so flush all 2813 * possible entries. 2814 */ 2815 if (vp->v_count > 1) 2816 dnlc_purge_vp(vp); 2817 2818 /* 2819 * Now we have the real reference count on the vnode 2820 */ 2821 rp = VTOR(vp); 2822 mutex_enter(&rp->r_statelock); 2823 if (vp->v_count > 1 && 2824 (rp->r_unldvp == NULL || strcmp(nm, rp->r_unlname) == 0)) { 2825 mutex_exit(&rp->r_statelock); 2826 tmpname = newname(); 2827 error = nfs3rename(dvp, nm, dvp, tmpname, cr, ct); 2828 if (error) 2829 kmem_free(tmpname, MAXNAMELEN); 2830 else { 2831 mutex_enter(&rp->r_statelock); 2832 if (rp->r_unldvp == NULL) { 2833 VN_HOLD(dvp); 2834 rp->r_unldvp = dvp; 2835 if (rp->r_unlcred != NULL) 2836 crfree(rp->r_unlcred); 2837 crhold(cr); 2838 rp->r_unlcred = cr; 2839 rp->r_unlname = tmpname; 2840 } else { 2841 kmem_free(rp->r_unlname, MAXNAMELEN); 2842 rp->r_unlname = tmpname; 2843 } 2844 mutex_exit(&rp->r_statelock); 2845 } 2846 } else { 2847 mutex_exit(&rp->r_statelock); 2848 /* 2849 * We need to flush any dirty pages which happen to 2850 * be hanging around before removing the file. This 2851 * shouldn't happen very often and mostly on file 2852 * systems mounted "nocto". 2853 */ 2854 if (vn_has_cached_data(vp) && 2855 ((rp->r_flags & RDIRTY) || rp->r_count > 0)) { 2856 error = nfs3_putpage(vp, (offset_t)0, 0, 0, cr, ct); 2857 if (error && (error == ENOSPC || error == EDQUOT)) { 2858 mutex_enter(&rp->r_statelock); 2859 if (!rp->r_error) 2860 rp->r_error = error; 2861 mutex_exit(&rp->r_statelock); 2862 } 2863 } 2864 2865 setdiropargs3(&args.object, nm, dvp); 2866 2867 douprintf = 1; 2868 2869 t = gethrtime(); 2870 2871 error = rfs3call(VTOMI(dvp), NFSPROC3_REMOVE, 2872 xdr_diropargs3, (caddr_t)&args, 2873 xdr_REMOVE3res, (caddr_t)&res, cr, 2874 &douprintf, &res.status, 0, NULL); 2875 2876 /* 2877 * The xattr dir may be gone after last attr is removed, 2878 * so flush it from dnlc. 2879 */ 2880 if (dvp->v_flag & V_XATTRDIR) 2881 dnlc_purge_vp(dvp); 2882 2883 PURGE_ATTRCACHE(vp); 2884 2885 if (error) { 2886 PURGE_ATTRCACHE(dvp); 2887 } else { 2888 error = geterrno3(res.status); 2889 if (!error) { 2890 nfs3_cache_wcc_data(dvp, &res.resok.dir_wcc, t, 2891 cr); 2892 if (HAVE_RDDIR_CACHE(drp)) 2893 nfs_purge_rddir_cache(dvp); 2894 } else { 2895 nfs3_cache_wcc_data(dvp, &res.resfail.dir_wcc, 2896 t, cr); 2897 PURGE_STALE_FH(error, dvp, cr); 2898 } 2899 } 2900 } 2901 2902 if (error == 0) { 2903 vnevent_remove(vp, dvp, nm, ct); 2904 } 2905 VN_RELE(vp); 2906 2907 nfs_rw_exit(&drp->r_rwlock); 2908 2909 return (error); 2910 } 2911 2912 /* ARGSUSED */ 2913 static int 2914 nfs3_link(vnode_t *tdvp, vnode_t *svp, char *tnm, cred_t *cr, 2915 caller_context_t *ct, int flags) 2916 { 2917 int error; 2918 LINK3args args; 2919 LINK3res res; 2920 vnode_t *realvp; 2921 int douprintf; 2922 mntinfo_t *mi; 2923 rnode_t *tdrp; 2924 hrtime_t t; 2925 2926 if (nfs_zone() != VTOMI(tdvp)->mi_zone) 2927 return (EPERM); 2928 if (VOP_REALVP(svp, &realvp, ct) == 0) 2929 svp = realvp; 2930 2931 mi = VTOMI(svp); 2932 2933 if (!(mi->mi_flags & MI_LINK)) 2934 return (EOPNOTSUPP); 2935 2936 args.file = *VTOFH3(svp); 2937 setdiropargs3(&args.link, tnm, tdvp); 2938 2939 tdrp = VTOR(tdvp); 2940 if (nfs_rw_enter_sig(&tdrp->r_rwlock, RW_WRITER, INTR(tdvp))) 2941 return (EINTR); 2942 2943 dnlc_remove(tdvp, tnm); 2944 2945 douprintf = 1; 2946 2947 t = gethrtime(); 2948 2949 error = rfs3call(mi, NFSPROC3_LINK, 2950 xdr_LINK3args, (caddr_t)&args, 2951 xdr_LINK3res, (caddr_t)&res, cr, 2952 &douprintf, &res.status, 0, NULL); 2953 2954 if (error) { 2955 PURGE_ATTRCACHE(tdvp); 2956 PURGE_ATTRCACHE(svp); 2957 nfs_rw_exit(&tdrp->r_rwlock); 2958 return (error); 2959 } 2960 2961 error = geterrno3(res.status); 2962 2963 if (!error) { 2964 nfs3_cache_post_op_attr(svp, &res.resok.file_attributes, t, cr); 2965 nfs3_cache_wcc_data(tdvp, &res.resok.linkdir_wcc, t, cr); 2966 if (HAVE_RDDIR_CACHE(tdrp)) 2967 nfs_purge_rddir_cache(tdvp); 2968 dnlc_update(tdvp, tnm, svp); 2969 } else { 2970 nfs3_cache_post_op_attr(svp, &res.resfail.file_attributes, t, 2971 cr); 2972 nfs3_cache_wcc_data(tdvp, &res.resfail.linkdir_wcc, t, cr); 2973 if (error == EOPNOTSUPP) { 2974 mutex_enter(&mi->mi_lock); 2975 mi->mi_flags &= ~MI_LINK; 2976 mutex_exit(&mi->mi_lock); 2977 } 2978 } 2979 2980 nfs_rw_exit(&tdrp->r_rwlock); 2981 2982 if (!error) { 2983 /* 2984 * Notify the source file of this link operation. 2985 */ 2986 vnevent_link(svp, ct); 2987 } 2988 return (error); 2989 } 2990 2991 /* ARGSUSED */ 2992 static int 2993 nfs3_rename(vnode_t *odvp, char *onm, vnode_t *ndvp, char *nnm, cred_t *cr, 2994 caller_context_t *ct, int flags) 2995 { 2996 vnode_t *realvp; 2997 2998 if (nfs_zone() != VTOMI(odvp)->mi_zone) 2999 return (EPERM); 3000 if (VOP_REALVP(ndvp, &realvp, ct) == 0) 3001 ndvp = realvp; 3002 3003 return (nfs3rename(odvp, onm, ndvp, nnm, cr, ct)); 3004 } 3005 3006 /* 3007 * nfs3rename does the real work of renaming in NFS Version 3. 3008 */ 3009 static int 3010 nfs3rename(vnode_t *odvp, char *onm, vnode_t *ndvp, char *nnm, cred_t *cr, 3011 caller_context_t *ct) 3012 { 3013 int error; 3014 RENAME3args args; 3015 RENAME3res res; 3016 int douprintf; 3017 vnode_t *nvp = NULL; 3018 vnode_t *ovp = NULL; 3019 char *tmpname; 3020 rnode_t *rp; 3021 rnode_t *odrp; 3022 rnode_t *ndrp; 3023 hrtime_t t; 3024 3025 ASSERT(nfs_zone() == VTOMI(odvp)->mi_zone); 3026 3027 if (strcmp(onm, ".") == 0 || strcmp(onm, "..") == 0 || 3028 strcmp(nnm, ".") == 0 || strcmp(nnm, "..") == 0) 3029 return (EINVAL); 3030 3031 odrp = VTOR(odvp); 3032 ndrp = VTOR(ndvp); 3033 if ((intptr_t)odrp < (intptr_t)ndrp) { 3034 if (nfs_rw_enter_sig(&odrp->r_rwlock, RW_WRITER, INTR(odvp))) 3035 return (EINTR); 3036 if (nfs_rw_enter_sig(&ndrp->r_rwlock, RW_WRITER, INTR(ndvp))) { 3037 nfs_rw_exit(&odrp->r_rwlock); 3038 return (EINTR); 3039 } 3040 } else { 3041 if (nfs_rw_enter_sig(&ndrp->r_rwlock, RW_WRITER, INTR(ndvp))) 3042 return (EINTR); 3043 if (nfs_rw_enter_sig(&odrp->r_rwlock, RW_WRITER, INTR(odvp))) { 3044 nfs_rw_exit(&ndrp->r_rwlock); 3045 return (EINTR); 3046 } 3047 } 3048 3049 /* 3050 * Lookup the target file. If it exists, it needs to be 3051 * checked to see whether it is a mount point and whether 3052 * it is active (open). 3053 */ 3054 error = nfs3lookup(ndvp, nnm, &nvp, NULL, 0, NULL, cr, 0); 3055 if (!error) { 3056 /* 3057 * If this file has been mounted on, then just 3058 * return busy because renaming to it would remove 3059 * the mounted file system from the name space. 3060 */ 3061 if (vn_mountedvfs(nvp) != NULL) { 3062 VN_RELE(nvp); 3063 nfs_rw_exit(&odrp->r_rwlock); 3064 nfs_rw_exit(&ndrp->r_rwlock); 3065 return (EBUSY); 3066 } 3067 3068 /* 3069 * Purge the name cache of all references to this vnode 3070 * so that we can check the reference count to infer 3071 * whether it is active or not. 3072 */ 3073 /* 3074 * First just remove the entry from the name cache, as it 3075 * is most likely the only entry for this vp. 3076 */ 3077 dnlc_remove(ndvp, nnm); 3078 /* 3079 * If the file has a v_count > 1 then there may be more 3080 * than one entry in the name cache due multiple links 3081 * or an open file, but we don't have the real reference 3082 * count so flush all possible entries. 3083 */ 3084 if (nvp->v_count > 1) 3085 dnlc_purge_vp(nvp); 3086 3087 /* 3088 * If the vnode is active and is not a directory, 3089 * arrange to rename it to a 3090 * temporary file so that it will continue to be 3091 * accessible. This implements the "unlink-open-file" 3092 * semantics for the target of a rename operation. 3093 * Before doing this though, make sure that the 3094 * source and target files are not already the same. 3095 */ 3096 if (nvp->v_count > 1 && nvp->v_type != VDIR) { 3097 /* 3098 * Lookup the source name. 3099 */ 3100 error = nfs3lookup(odvp, onm, &ovp, NULL, 0, NULL, 3101 cr, 0); 3102 3103 /* 3104 * The source name *should* already exist. 3105 */ 3106 if (error) { 3107 VN_RELE(nvp); 3108 nfs_rw_exit(&odrp->r_rwlock); 3109 nfs_rw_exit(&ndrp->r_rwlock); 3110 return (error); 3111 } 3112 3113 /* 3114 * Compare the two vnodes. If they are the same, 3115 * just release all held vnodes and return success. 3116 */ 3117 if (ovp == nvp) { 3118 VN_RELE(ovp); 3119 VN_RELE(nvp); 3120 nfs_rw_exit(&odrp->r_rwlock); 3121 nfs_rw_exit(&ndrp->r_rwlock); 3122 return (0); 3123 } 3124 3125 /* 3126 * Can't mix and match directories and non- 3127 * directories in rename operations. We already 3128 * know that the target is not a directory. If 3129 * the source is a directory, return an error. 3130 */ 3131 if (ovp->v_type == VDIR) { 3132 VN_RELE(ovp); 3133 VN_RELE(nvp); 3134 nfs_rw_exit(&odrp->r_rwlock); 3135 nfs_rw_exit(&ndrp->r_rwlock); 3136 return (ENOTDIR); 3137 } 3138 3139 /* 3140 * The target file exists, is not the same as 3141 * the source file, and is active. Link it 3142 * to a temporary filename to avoid having 3143 * the server removing the file completely. 3144 */ 3145 tmpname = newname(); 3146 error = nfs3_link(ndvp, nvp, tmpname, cr, NULL, 0); 3147 if (error == EOPNOTSUPP) { 3148 error = nfs3_rename(ndvp, nnm, ndvp, tmpname, 3149 cr, NULL, 0); 3150 } 3151 if (error) { 3152 kmem_free(tmpname, MAXNAMELEN); 3153 VN_RELE(ovp); 3154 VN_RELE(nvp); 3155 nfs_rw_exit(&odrp->r_rwlock); 3156 nfs_rw_exit(&ndrp->r_rwlock); 3157 return (error); 3158 } 3159 rp = VTOR(nvp); 3160 mutex_enter(&rp->r_statelock); 3161 if (rp->r_unldvp == NULL) { 3162 VN_HOLD(ndvp); 3163 rp->r_unldvp = ndvp; 3164 if (rp->r_unlcred != NULL) 3165 crfree(rp->r_unlcred); 3166 crhold(cr); 3167 rp->r_unlcred = cr; 3168 rp->r_unlname = tmpname; 3169 } else { 3170 kmem_free(rp->r_unlname, MAXNAMELEN); 3171 rp->r_unlname = tmpname; 3172 } 3173 mutex_exit(&rp->r_statelock); 3174 } 3175 } 3176 3177 if (ovp == NULL) { 3178 /* 3179 * When renaming directories to be a subdirectory of a 3180 * different parent, the dnlc entry for ".." will no 3181 * longer be valid, so it must be removed. 3182 * 3183 * We do a lookup here to determine whether we are renaming 3184 * a directory and we need to check if we are renaming 3185 * an unlinked file. This might have already been done 3186 * in previous code, so we check ovp == NULL to avoid 3187 * doing it twice. 3188 */ 3189 3190 error = nfs3lookup(odvp, onm, &ovp, NULL, 0, NULL, cr, 0); 3191 /* 3192 * The source name *should* already exist. 3193 */ 3194 if (error) { 3195 nfs_rw_exit(&odrp->r_rwlock); 3196 nfs_rw_exit(&ndrp->r_rwlock); 3197 if (nvp) { 3198 VN_RELE(nvp); 3199 } 3200 return (error); 3201 } 3202 ASSERT(ovp != NULL); 3203 } 3204 3205 dnlc_remove(odvp, onm); 3206 dnlc_remove(ndvp, nnm); 3207 3208 setdiropargs3(&args.from, onm, odvp); 3209 setdiropargs3(&args.to, nnm, ndvp); 3210 3211 douprintf = 1; 3212 3213 t = gethrtime(); 3214 3215 error = rfs3call(VTOMI(odvp), NFSPROC3_RENAME, 3216 xdr_RENAME3args, (caddr_t)&args, 3217 xdr_RENAME3res, (caddr_t)&res, cr, 3218 &douprintf, &res.status, 0, NULL); 3219 3220 if (error) { 3221 PURGE_ATTRCACHE(odvp); 3222 PURGE_ATTRCACHE(ndvp); 3223 VN_RELE(ovp); 3224 nfs_rw_exit(&odrp->r_rwlock); 3225 nfs_rw_exit(&ndrp->r_rwlock); 3226 if (nvp) { 3227 VN_RELE(nvp); 3228 } 3229 return (error); 3230 } 3231 3232 error = geterrno3(res.status); 3233 3234 if (!error) { 3235 nfs3_cache_wcc_data(odvp, &res.resok.fromdir_wcc, t, cr); 3236 if (HAVE_RDDIR_CACHE(odrp)) 3237 nfs_purge_rddir_cache(odvp); 3238 if (ndvp != odvp) { 3239 nfs3_cache_wcc_data(ndvp, &res.resok.todir_wcc, t, cr); 3240 if (HAVE_RDDIR_CACHE(ndrp)) 3241 nfs_purge_rddir_cache(ndvp); 3242 } 3243 /* 3244 * when renaming directories to be a subdirectory of a 3245 * different parent, the dnlc entry for ".." will no 3246 * longer be valid, so it must be removed 3247 */ 3248 rp = VTOR(ovp); 3249 if (ndvp != odvp) { 3250 if (ovp->v_type == VDIR) { 3251 dnlc_remove(ovp, ".."); 3252 if (HAVE_RDDIR_CACHE(rp)) 3253 nfs_purge_rddir_cache(ovp); 3254 } 3255 } 3256 3257 /* 3258 * If we are renaming the unlinked file, update the 3259 * r_unldvp and r_unlname as needed. 3260 */ 3261 mutex_enter(&rp->r_statelock); 3262 if (rp->r_unldvp != NULL) { 3263 if (strcmp(rp->r_unlname, onm) == 0) { 3264 (void) strncpy(rp->r_unlname, nnm, MAXNAMELEN); 3265 rp->r_unlname[MAXNAMELEN - 1] = '\0'; 3266 3267 if (ndvp != rp->r_unldvp) { 3268 VN_RELE(rp->r_unldvp); 3269 rp->r_unldvp = ndvp; 3270 VN_HOLD(ndvp); 3271 } 3272 } 3273 } 3274 mutex_exit(&rp->r_statelock); 3275 } else { 3276 nfs3_cache_wcc_data(odvp, &res.resfail.fromdir_wcc, t, cr); 3277 if (ndvp != odvp) { 3278 nfs3_cache_wcc_data(ndvp, &res.resfail.todir_wcc, t, 3279 cr); 3280 } 3281 /* 3282 * System V defines rename to return EEXIST, not 3283 * ENOTEMPTY if the target directory is not empty. 3284 * Over the wire, the error is NFSERR_ENOTEMPTY 3285 * which geterrno maps to ENOTEMPTY. 3286 */ 3287 if (error == ENOTEMPTY) 3288 error = EEXIST; 3289 } 3290 3291 if (error == 0) { 3292 if (nvp) 3293 vnevent_rename_dest(nvp, ndvp, nnm, ct); 3294 3295 if (odvp != ndvp) 3296 vnevent_rename_dest_dir(ndvp, ct); 3297 ASSERT(ovp != NULL); 3298 vnevent_rename_src(ovp, odvp, onm, ct); 3299 } 3300 3301 if (nvp) { 3302 VN_RELE(nvp); 3303 } 3304 VN_RELE(ovp); 3305 3306 nfs_rw_exit(&odrp->r_rwlock); 3307 nfs_rw_exit(&ndrp->r_rwlock); 3308 3309 return (error); 3310 } 3311 3312 /* ARGSUSED */ 3313 static int 3314 nfs3_mkdir(vnode_t *dvp, char *nm, struct vattr *va, vnode_t **vpp, cred_t *cr, 3315 caller_context_t *ct, int flags, vsecattr_t *vsecp) 3316 { 3317 int error; 3318 MKDIR3args args; 3319 MKDIR3res res; 3320 int douprintf; 3321 struct vattr vattr; 3322 vnode_t *vp; 3323 rnode_t *drp; 3324 hrtime_t t; 3325 3326 if (nfs_zone() != VTOMI(dvp)->mi_zone) 3327 return (EPERM); 3328 setdiropargs3(&args.where, nm, dvp); 3329 3330 /* 3331 * Decide what the group-id and set-gid bit of the created directory 3332 * should be. May have to do a setattr to get the gid right. 3333 */ 3334 error = setdirgid(dvp, &va->va_gid, cr); 3335 if (error) 3336 return (error); 3337 error = setdirmode(dvp, &va->va_mode, cr); 3338 if (error) 3339 return (error); 3340 va->va_mask |= AT_MODE|AT_GID; 3341 3342 error = vattr_to_sattr3(va, &args.attributes); 3343 if (error) { 3344 /* req time field(s) overflow - return immediately */ 3345 return (error); 3346 } 3347 3348 drp = VTOR(dvp); 3349 if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp))) 3350 return (EINTR); 3351 3352 dnlc_remove(dvp, nm); 3353 3354 douprintf = 1; 3355 3356 t = gethrtime(); 3357 3358 error = rfs3call(VTOMI(dvp), NFSPROC3_MKDIR, 3359 xdr_MKDIR3args, (caddr_t)&args, 3360 xdr_MKDIR3res, (caddr_t)&res, cr, 3361 &douprintf, &res.status, 0, NULL); 3362 3363 if (error) { 3364 PURGE_ATTRCACHE(dvp); 3365 nfs_rw_exit(&drp->r_rwlock); 3366 return (error); 3367 } 3368 3369 error = geterrno3(res.status); 3370 if (!error) { 3371 nfs3_cache_wcc_data(dvp, &res.resok.dir_wcc, t, cr); 3372 if (HAVE_RDDIR_CACHE(drp)) 3373 nfs_purge_rddir_cache(dvp); 3374 3375 if (!res.resok.obj.handle_follows) { 3376 error = nfs3lookup(dvp, nm, &vp, NULL, 0, NULL, cr, 0); 3377 if (error) { 3378 nfs_rw_exit(&drp->r_rwlock); 3379 return (error); 3380 } 3381 } else { 3382 if (res.resok.obj_attributes.attributes) { 3383 vp = makenfs3node(&res.resok.obj.handle, 3384 &res.resok.obj_attributes.attr, 3385 dvp->v_vfsp, t, cr, NULL, NULL); 3386 } else { 3387 vp = makenfs3node(&res.resok.obj.handle, NULL, 3388 dvp->v_vfsp, t, cr, NULL, NULL); 3389 if (vp->v_type == VNON) { 3390 vattr.va_mask = AT_TYPE; 3391 error = nfs3getattr(vp, &vattr, cr); 3392 if (error) { 3393 VN_RELE(vp); 3394 nfs_rw_exit(&drp->r_rwlock); 3395 return (error); 3396 } 3397 vp->v_type = vattr.va_type; 3398 } 3399 } 3400 dnlc_update(dvp, nm, vp); 3401 } 3402 if (va->va_gid != VTOR(vp)->r_attr.va_gid) { 3403 va->va_mask = AT_GID; 3404 (void) nfs3setattr(vp, va, 0, cr); 3405 } 3406 *vpp = vp; 3407 } else { 3408 nfs3_cache_wcc_data(dvp, &res.resfail.dir_wcc, t, cr); 3409 PURGE_STALE_FH(error, dvp, cr); 3410 } 3411 3412 nfs_rw_exit(&drp->r_rwlock); 3413 3414 return (error); 3415 } 3416 3417 /* ARGSUSED */ 3418 static int 3419 nfs3_rmdir(vnode_t *dvp, char *nm, vnode_t *cdir, cred_t *cr, 3420 caller_context_t *ct, int flags) 3421 { 3422 int error; 3423 RMDIR3args args; 3424 RMDIR3res res; 3425 vnode_t *vp; 3426 int douprintf; 3427 rnode_t *drp; 3428 hrtime_t t; 3429 3430 if (nfs_zone() != VTOMI(dvp)->mi_zone) 3431 return (EPERM); 3432 drp = VTOR(dvp); 3433 if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp))) 3434 return (EINTR); 3435 3436 /* 3437 * Attempt to prevent a rmdir(".") from succeeding. 3438 */ 3439 error = nfs3lookup(dvp, nm, &vp, NULL, 0, NULL, cr, 0); 3440 if (error) { 3441 nfs_rw_exit(&drp->r_rwlock); 3442 return (error); 3443 } 3444 3445 if (vp == cdir) { 3446 VN_RELE(vp); 3447 nfs_rw_exit(&drp->r_rwlock); 3448 return (EINVAL); 3449 } 3450 3451 setdiropargs3(&args.object, nm, dvp); 3452 3453 /* 3454 * First just remove the entry from the name cache, as it 3455 * is most likely an entry for this vp. 3456 */ 3457 dnlc_remove(dvp, nm); 3458 3459 /* 3460 * If there vnode reference count is greater than one, then 3461 * there may be additional references in the DNLC which will 3462 * need to be purged. First, trying removing the entry for 3463 * the parent directory and see if that removes the additional 3464 * reference(s). If that doesn't do it, then use dnlc_purge_vp 3465 * to completely remove any references to the directory which 3466 * might still exist in the DNLC. 3467 */ 3468 if (vp->v_count > 1) { 3469 dnlc_remove(vp, ".."); 3470 if (vp->v_count > 1) 3471 dnlc_purge_vp(vp); 3472 } 3473 3474 douprintf = 1; 3475 3476 t = gethrtime(); 3477 3478 error = rfs3call(VTOMI(dvp), NFSPROC3_RMDIR, 3479 xdr_diropargs3, (caddr_t)&args, 3480 xdr_RMDIR3res, (caddr_t)&res, cr, 3481 &douprintf, &res.status, 0, NULL); 3482 3483 PURGE_ATTRCACHE(vp); 3484 3485 if (error) { 3486 PURGE_ATTRCACHE(dvp); 3487 VN_RELE(vp); 3488 nfs_rw_exit(&drp->r_rwlock); 3489 return (error); 3490 } 3491 3492 error = geterrno3(res.status); 3493 if (!error) { 3494 nfs3_cache_wcc_data(dvp, &res.resok.dir_wcc, t, cr); 3495 if (HAVE_RDDIR_CACHE(drp)) 3496 nfs_purge_rddir_cache(dvp); 3497 if (HAVE_RDDIR_CACHE(VTOR(vp))) 3498 nfs_purge_rddir_cache(vp); 3499 } else { 3500 nfs3_cache_wcc_data(dvp, &res.resfail.dir_wcc, t, cr); 3501 PURGE_STALE_FH(error, dvp, cr); 3502 /* 3503 * System V defines rmdir to return EEXIST, not 3504 * ENOTEMPTY if the directory is not empty. Over 3505 * the wire, the error is NFSERR_ENOTEMPTY which 3506 * geterrno maps to ENOTEMPTY. 3507 */ 3508 if (error == ENOTEMPTY) 3509 error = EEXIST; 3510 } 3511 3512 if (error == 0) { 3513 vnevent_rmdir(vp, dvp, nm, ct); 3514 } 3515 VN_RELE(vp); 3516 3517 nfs_rw_exit(&drp->r_rwlock); 3518 3519 return (error); 3520 } 3521 3522 /* ARGSUSED */ 3523 static int 3524 nfs3_symlink(vnode_t *dvp, char *lnm, struct vattr *tva, char *tnm, cred_t *cr, 3525 caller_context_t *ct, int flags) 3526 { 3527 int error; 3528 SYMLINK3args args; 3529 SYMLINK3res res; 3530 int douprintf; 3531 mntinfo_t *mi; 3532 vnode_t *vp; 3533 rnode_t *rp; 3534 char *contents; 3535 rnode_t *drp; 3536 hrtime_t t; 3537 3538 mi = VTOMI(dvp); 3539 3540 if (nfs_zone() != mi->mi_zone) 3541 return (EPERM); 3542 if (!(mi->mi_flags & MI_SYMLINK)) 3543 return (EOPNOTSUPP); 3544 3545 setdiropargs3(&args.where, lnm, dvp); 3546 error = vattr_to_sattr3(tva, &args.symlink.symlink_attributes); 3547 if (error) { 3548 /* req time field(s) overflow - return immediately */ 3549 return (error); 3550 } 3551 args.symlink.symlink_data = tnm; 3552 3553 drp = VTOR(dvp); 3554 if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp))) 3555 return (EINTR); 3556 3557 dnlc_remove(dvp, lnm); 3558 3559 douprintf = 1; 3560 3561 t = gethrtime(); 3562 3563 error = rfs3call(mi, NFSPROC3_SYMLINK, 3564 xdr_SYMLINK3args, (caddr_t)&args, 3565 xdr_SYMLINK3res, (caddr_t)&res, cr, 3566 &douprintf, &res.status, 0, NULL); 3567 3568 if (error) { 3569 PURGE_ATTRCACHE(dvp); 3570 nfs_rw_exit(&drp->r_rwlock); 3571 return (error); 3572 } 3573 3574 error = geterrno3(res.status); 3575 if (!error) { 3576 nfs3_cache_wcc_data(dvp, &res.resok.dir_wcc, t, cr); 3577 if (HAVE_RDDIR_CACHE(drp)) 3578 nfs_purge_rddir_cache(dvp); 3579 3580 if (res.resok.obj.handle_follows) { 3581 if (res.resok.obj_attributes.attributes) { 3582 vp = makenfs3node(&res.resok.obj.handle, 3583 &res.resok.obj_attributes.attr, 3584 dvp->v_vfsp, t, cr, NULL, NULL); 3585 } else { 3586 vp = makenfs3node(&res.resok.obj.handle, NULL, 3587 dvp->v_vfsp, t, cr, NULL, NULL); 3588 vp->v_type = VLNK; 3589 vp->v_rdev = 0; 3590 } 3591 dnlc_update(dvp, lnm, vp); 3592 rp = VTOR(vp); 3593 if (nfs3_do_symlink_cache && 3594 rp->r_symlink.contents == NULL) { 3595 3596 contents = kmem_alloc(MAXPATHLEN, 3597 KM_NOSLEEP); 3598 3599 if (contents != NULL) { 3600 mutex_enter(&rp->r_statelock); 3601 if (rp->r_symlink.contents == NULL) { 3602 rp->r_symlink.len = strlen(tnm); 3603 bcopy(tnm, contents, 3604 rp->r_symlink.len); 3605 rp->r_symlink.contents = 3606 contents; 3607 rp->r_symlink.size = MAXPATHLEN; 3608 mutex_exit(&rp->r_statelock); 3609 } else { 3610 mutex_exit(&rp->r_statelock); 3611 kmem_free((void *)contents, 3612 MAXPATHLEN); 3613 } 3614 } 3615 } 3616 VN_RELE(vp); 3617 } 3618 } else { 3619 nfs3_cache_wcc_data(dvp, &res.resfail.dir_wcc, t, cr); 3620 PURGE_STALE_FH(error, dvp, cr); 3621 if (error == EOPNOTSUPP) { 3622 mutex_enter(&mi->mi_lock); 3623 mi->mi_flags &= ~MI_SYMLINK; 3624 mutex_exit(&mi->mi_lock); 3625 } 3626 } 3627 3628 nfs_rw_exit(&drp->r_rwlock); 3629 3630 return (error); 3631 } 3632 3633 #ifdef DEBUG 3634 static int nfs3_readdir_cache_hits = 0; 3635 static int nfs3_readdir_cache_shorts = 0; 3636 static int nfs3_readdir_cache_waits = 0; 3637 static int nfs3_readdir_cache_misses = 0; 3638 static int nfs3_readdir_readahead = 0; 3639 #endif 3640 3641 static int nfs3_shrinkreaddir = 0; 3642 3643 /* 3644 * Read directory entries. 3645 * There are some weird things to look out for here. The uio_loffset 3646 * field is either 0 or it is the offset returned from a previous 3647 * readdir. It is an opaque value used by the server to find the 3648 * correct directory block to read. The count field is the number 3649 * of blocks to read on the server. This is advisory only, the server 3650 * may return only one block's worth of entries. Entries may be compressed 3651 * on the server. 3652 */ 3653 /* ARGSUSED */ 3654 static int 3655 nfs3_readdir(vnode_t *vp, struct uio *uiop, cred_t *cr, int *eofp, 3656 caller_context_t *ct, int flags) 3657 { 3658 int error; 3659 size_t count; 3660 rnode_t *rp; 3661 rddir_cache *rdc; 3662 rddir_cache *nrdc; 3663 rddir_cache *rrdc; 3664 #ifdef DEBUG 3665 int missed; 3666 #endif 3667 int doreadahead; 3668 rddir_cache srdc; 3669 avl_index_t where; 3670 3671 if (nfs_zone() != VTOMI(vp)->mi_zone) 3672 return (EIO); 3673 rp = VTOR(vp); 3674 3675 ASSERT(nfs_rw_lock_held(&rp->r_rwlock, RW_READER)); 3676 3677 /* 3678 * Make sure that the directory cache is valid. 3679 */ 3680 if (HAVE_RDDIR_CACHE(rp)) { 3681 if (nfs_disable_rddir_cache) { 3682 /* 3683 * Setting nfs_disable_rddir_cache in /etc/system 3684 * allows interoperability with servers that do not 3685 * properly update the attributes of directories. 3686 * Any cached information gets purged before an 3687 * access is made to it. 3688 */ 3689 nfs_purge_rddir_cache(vp); 3690 } else { 3691 error = nfs3_validate_caches(vp, cr); 3692 if (error) 3693 return (error); 3694 } 3695 } 3696 3697 /* 3698 * It is possible that some servers may not be able to correctly 3699 * handle a large READDIR or READDIRPLUS request due to bugs in 3700 * their implementation. In order to continue to interoperate 3701 * with them, this workaround is provided to limit the maximum 3702 * size of a READDIRPLUS request to 1024. In any case, the request 3703 * size is limited to MAXBSIZE. 3704 */ 3705 count = MIN(uiop->uio_iov->iov_len, 3706 nfs3_shrinkreaddir ? 1024 : MAXBSIZE); 3707 3708 nrdc = NULL; 3709 #ifdef DEBUG 3710 missed = 0; 3711 #endif 3712 top: 3713 /* 3714 * Short circuit last readdir which always returns 0 bytes. 3715 * This can be done after the directory has been read through 3716 * completely at least once. This will set r_direof which 3717 * can be used to find the value of the last cookie. 3718 */ 3719 mutex_enter(&rp->r_statelock); 3720 if (rp->r_direof != NULL && 3721 uiop->uio_loffset == rp->r_direof->nfs3_ncookie) { 3722 mutex_exit(&rp->r_statelock); 3723 #ifdef DEBUG 3724 nfs3_readdir_cache_shorts++; 3725 #endif 3726 if (eofp) 3727 *eofp = 1; 3728 if (nrdc != NULL) 3729 rddir_cache_rele(nrdc); 3730 return (0); 3731 } 3732 /* 3733 * Look for a cache entry. Cache entries are identified 3734 * by the NFS cookie value and the byte count requested. 3735 */ 3736 srdc.nfs3_cookie = uiop->uio_loffset; 3737 srdc.buflen = count; 3738 rdc = avl_find(&rp->r_dir, &srdc, &where); 3739 if (rdc != NULL) { 3740 rddir_cache_hold(rdc); 3741 /* 3742 * If the cache entry is in the process of being 3743 * filled in, wait until this completes. The 3744 * RDDIRWAIT bit is set to indicate that someone 3745 * is waiting and then the thread currently 3746 * filling the entry is done, it should do a 3747 * cv_broadcast to wakeup all of the threads 3748 * waiting for it to finish. 3749 */ 3750 if (rdc->flags & RDDIR) { 3751 nfs_rw_exit(&rp->r_rwlock); 3752 rdc->flags |= RDDIRWAIT; 3753 #ifdef DEBUG 3754 nfs3_readdir_cache_waits++; 3755 #endif 3756 if (!cv_wait_sig(&rdc->cv, &rp->r_statelock)) { 3757 /* 3758 * We got interrupted, probably 3759 * the user typed ^C or an alarm 3760 * fired. We free the new entry 3761 * if we allocated one. 3762 */ 3763 mutex_exit(&rp->r_statelock); 3764 (void) nfs_rw_enter_sig(&rp->r_rwlock, 3765 RW_READER, FALSE); 3766 rddir_cache_rele(rdc); 3767 if (nrdc != NULL) 3768 rddir_cache_rele(nrdc); 3769 return (EINTR); 3770 } 3771 mutex_exit(&rp->r_statelock); 3772 (void) nfs_rw_enter_sig(&rp->r_rwlock, 3773 RW_READER, FALSE); 3774 rddir_cache_rele(rdc); 3775 goto top; 3776 } 3777 /* 3778 * Check to see if a readdir is required to 3779 * fill the entry. If so, mark this entry 3780 * as being filled, remove our reference, 3781 * and branch to the code to fill the entry. 3782 */ 3783 if (rdc->flags & RDDIRREQ) { 3784 rdc->flags &= ~RDDIRREQ; 3785 rdc->flags |= RDDIR; 3786 if (nrdc != NULL) 3787 rddir_cache_rele(nrdc); 3788 nrdc = rdc; 3789 mutex_exit(&rp->r_statelock); 3790 goto bottom; 3791 } 3792 #ifdef DEBUG 3793 if (!missed) 3794 nfs3_readdir_cache_hits++; 3795 #endif 3796 /* 3797 * If an error occurred while attempting 3798 * to fill the cache entry, just return it. 3799 */ 3800 if (rdc->error) { 3801 error = rdc->error; 3802 mutex_exit(&rp->r_statelock); 3803 rddir_cache_rele(rdc); 3804 if (nrdc != NULL) 3805 rddir_cache_rele(nrdc); 3806 return (error); 3807 } 3808 3809 /* 3810 * The cache entry is complete and good, 3811 * copyout the dirent structs to the calling 3812 * thread. 3813 */ 3814 error = uiomove(rdc->entries, rdc->entlen, UIO_READ, uiop); 3815 3816 /* 3817 * If no error occurred during the copyout, 3818 * update the offset in the uio struct to 3819 * contain the value of the next cookie 3820 * and set the eof value appropriately. 3821 */ 3822 if (!error) { 3823 uiop->uio_loffset = rdc->nfs3_ncookie; 3824 if (eofp) 3825 *eofp = rdc->eof; 3826 } 3827 3828 /* 3829 * Decide whether to do readahead. 3830 * 3831 * Don't if have already read to the end of 3832 * directory. There is nothing more to read. 3833 * 3834 * Don't if the application is not doing 3835 * lookups in the directory. The readahead 3836 * is only effective if the application can 3837 * be doing work while an async thread is 3838 * handling the over the wire request. 3839 */ 3840 if (rdc->eof) { 3841 rp->r_direof = rdc; 3842 doreadahead = FALSE; 3843 } else if (!(rp->r_flags & RLOOKUP)) 3844 doreadahead = FALSE; 3845 else 3846 doreadahead = TRUE; 3847 3848 if (!doreadahead) { 3849 mutex_exit(&rp->r_statelock); 3850 rddir_cache_rele(rdc); 3851 if (nrdc != NULL) 3852 rddir_cache_rele(nrdc); 3853 return (error); 3854 } 3855 3856 /* 3857 * Check to see whether we found an entry 3858 * for the readahead. If so, we don't need 3859 * to do anything further, so free the new 3860 * entry if one was allocated. Otherwise, 3861 * allocate a new entry, add it to the cache, 3862 * and then initiate an asynchronous readdir 3863 * operation to fill it. 3864 */ 3865 srdc.nfs3_cookie = rdc->nfs3_ncookie; 3866 srdc.buflen = count; 3867 rrdc = avl_find(&rp->r_dir, &srdc, &where); 3868 if (rrdc != NULL) { 3869 if (nrdc != NULL) 3870 rddir_cache_rele(nrdc); 3871 } else { 3872 if (nrdc != NULL) 3873 rrdc = nrdc; 3874 else { 3875 rrdc = rddir_cache_alloc(KM_NOSLEEP); 3876 } 3877 if (rrdc != NULL) { 3878 rrdc->nfs3_cookie = rdc->nfs3_ncookie; 3879 rrdc->buflen = count; 3880 avl_insert(&rp->r_dir, rrdc, where); 3881 rddir_cache_hold(rrdc); 3882 mutex_exit(&rp->r_statelock); 3883 rddir_cache_rele(rdc); 3884 #ifdef DEBUG 3885 nfs3_readdir_readahead++; 3886 #endif 3887 nfs_async_readdir(vp, rrdc, cr, do_nfs3readdir); 3888 return (error); 3889 } 3890 } 3891 3892 mutex_exit(&rp->r_statelock); 3893 rddir_cache_rele(rdc); 3894 return (error); 3895 } 3896 3897 /* 3898 * Didn't find an entry in the cache. Construct a new empty 3899 * entry and link it into the cache. Other processes attempting 3900 * to access this entry will need to wait until it is filled in. 3901 * 3902 * Since kmem_alloc may block, another pass through the cache 3903 * will need to be taken to make sure that another process 3904 * hasn't already added an entry to the cache for this request. 3905 */ 3906 if (nrdc == NULL) { 3907 mutex_exit(&rp->r_statelock); 3908 nrdc = rddir_cache_alloc(KM_SLEEP); 3909 nrdc->nfs3_cookie = uiop->uio_loffset; 3910 nrdc->buflen = count; 3911 goto top; 3912 } 3913 3914 /* 3915 * Add this entry to the cache. 3916 */ 3917 avl_insert(&rp->r_dir, nrdc, where); 3918 rddir_cache_hold(nrdc); 3919 mutex_exit(&rp->r_statelock); 3920 3921 bottom: 3922 #ifdef DEBUG 3923 missed = 1; 3924 nfs3_readdir_cache_misses++; 3925 #endif 3926 /* 3927 * Do the readdir. This routine decides whether to use 3928 * READDIR or READDIRPLUS. 3929 */ 3930 error = do_nfs3readdir(vp, nrdc, cr); 3931 3932 /* 3933 * If this operation failed, just return the error which occurred. 3934 */ 3935 if (error != 0) 3936 return (error); 3937 3938 /* 3939 * Since the RPC operation will have taken sometime and blocked 3940 * this process, another pass through the cache will need to be 3941 * taken to find the correct cache entry. It is possible that 3942 * the correct cache entry will not be there (although one was 3943 * added) because the directory changed during the RPC operation 3944 * and the readdir cache was flushed. In this case, just start 3945 * over. It is hoped that this will not happen too often... :-) 3946 */ 3947 nrdc = NULL; 3948 goto top; 3949 /* NOTREACHED */ 3950 } 3951 3952 static int 3953 do_nfs3readdir(vnode_t *vp, rddir_cache *rdc, cred_t *cr) 3954 { 3955 int error; 3956 rnode_t *rp; 3957 mntinfo_t *mi; 3958 3959 rp = VTOR(vp); 3960 mi = VTOMI(vp); 3961 ASSERT(nfs_zone() == mi->mi_zone); 3962 /* 3963 * Issue the proper request. 3964 * 3965 * If the server does not support READDIRPLUS, then use READDIR. 3966 * 3967 * Otherwise -- 3968 * Issue a READDIRPLUS if reading to fill an empty cache or if 3969 * an application has performed a lookup in the directory which 3970 * required an over the wire lookup. The use of READDIRPLUS 3971 * will help to (re)populate the DNLC. 3972 */ 3973 if (!(mi->mi_flags & MI_READDIRONLY) && 3974 (rp->r_flags & (RLOOKUP | RREADDIRPLUS))) { 3975 if (rp->r_flags & RREADDIRPLUS) { 3976 mutex_enter(&rp->r_statelock); 3977 rp->r_flags &= ~RREADDIRPLUS; 3978 mutex_exit(&rp->r_statelock); 3979 } 3980 nfs3readdirplus(vp, rdc, cr); 3981 if (rdc->error == EOPNOTSUPP) 3982 nfs3readdir(vp, rdc, cr); 3983 } else 3984 nfs3readdir(vp, rdc, cr); 3985 3986 mutex_enter(&rp->r_statelock); 3987 rdc->flags &= ~RDDIR; 3988 if (rdc->flags & RDDIRWAIT) { 3989 rdc->flags &= ~RDDIRWAIT; 3990 cv_broadcast(&rdc->cv); 3991 } 3992 error = rdc->error; 3993 if (error) 3994 rdc->flags |= RDDIRREQ; 3995 mutex_exit(&rp->r_statelock); 3996 3997 rddir_cache_rele(rdc); 3998 3999 return (error); 4000 } 4001 4002 static void 4003 nfs3readdir(vnode_t *vp, rddir_cache *rdc, cred_t *cr) 4004 { 4005 int error; 4006 READDIR3args args; 4007 READDIR3vres res; 4008 vattr_t dva; 4009 rnode_t *rp; 4010 int douprintf; 4011 failinfo_t fi, *fip = NULL; 4012 mntinfo_t *mi; 4013 hrtime_t t; 4014 4015 rp = VTOR(vp); 4016 mi = VTOMI(vp); 4017 ASSERT(nfs_zone() == mi->mi_zone); 4018 4019 args.dir = *RTOFH3(rp); 4020 args.cookie = (cookie3)rdc->nfs3_cookie; 4021 args.cookieverf = rp->r_cookieverf; 4022 args.count = rdc->buflen; 4023 4024 /* 4025 * NFS client failover support 4026 * suppress failover unless we have a zero cookie 4027 */ 4028 if (args.cookie == (cookie3) 0) { 4029 fi.vp = vp; 4030 fi.fhp = (caddr_t)&args.dir; 4031 fi.copyproc = nfs3copyfh; 4032 fi.lookupproc = nfs3lookup; 4033 fi.xattrdirproc = acl_getxattrdir3; 4034 fip = &fi; 4035 } 4036 4037 #ifdef DEBUG 4038 rdc->entries = rddir_cache_buf_alloc(rdc->buflen, KM_SLEEP); 4039 #else 4040 rdc->entries = kmem_alloc(rdc->buflen, KM_SLEEP); 4041 #endif 4042 4043 res.entries = (dirent64_t *)rdc->entries; 4044 res.entries_size = rdc->buflen; 4045 res.dir_attributes.fres.vap = &dva; 4046 res.dir_attributes.fres.vp = vp; 4047 res.loff = rdc->nfs3_cookie; 4048 4049 douprintf = 1; 4050 4051 if (mi->mi_io_kstats) { 4052 mutex_enter(&mi->mi_lock); 4053 kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats)); 4054 mutex_exit(&mi->mi_lock); 4055 } 4056 4057 t = gethrtime(); 4058 4059 error = rfs3call(VTOMI(vp), NFSPROC3_READDIR, 4060 xdr_READDIR3args, (caddr_t)&args, 4061 xdr_READDIR3vres, (caddr_t)&res, cr, 4062 &douprintf, &res.status, 0, fip); 4063 4064 if (mi->mi_io_kstats) { 4065 mutex_enter(&mi->mi_lock); 4066 kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats)); 4067 mutex_exit(&mi->mi_lock); 4068 } 4069 4070 if (error) 4071 goto err; 4072 4073 nfs3_cache_post_op_vattr(vp, &res.dir_attributes, t, cr); 4074 4075 error = geterrno3(res.status); 4076 if (error) { 4077 PURGE_STALE_FH(error, vp, cr); 4078 goto err; 4079 } 4080 4081 if (mi->mi_io_kstats) { 4082 mutex_enter(&mi->mi_lock); 4083 KSTAT_IO_PTR(mi->mi_io_kstats)->reads++; 4084 KSTAT_IO_PTR(mi->mi_io_kstats)->nread += res.size; 4085 mutex_exit(&mi->mi_lock); 4086 } 4087 4088 rdc->nfs3_ncookie = res.loff; 4089 rp->r_cookieverf = res.cookieverf; 4090 rdc->eof = res.eof ? 1 : 0; 4091 rdc->entlen = res.size; 4092 ASSERT(rdc->entlen <= rdc->buflen); 4093 rdc->error = 0; 4094 return; 4095 4096 err: 4097 kmem_free(rdc->entries, rdc->buflen); 4098 rdc->entries = NULL; 4099 rdc->error = error; 4100 } 4101 4102 /* 4103 * Read directory entries. 4104 * There are some weird things to look out for here. The uio_loffset 4105 * field is either 0 or it is the offset returned from a previous 4106 * readdir. It is an opaque value used by the server to find the 4107 * correct directory block to read. The count field is the number 4108 * of blocks to read on the server. This is advisory only, the server 4109 * may return only one block's worth of entries. Entries may be compressed 4110 * on the server. 4111 */ 4112 static void 4113 nfs3readdirplus(vnode_t *vp, rddir_cache *rdc, cred_t *cr) 4114 { 4115 int error; 4116 READDIRPLUS3args args; 4117 READDIRPLUS3vres res; 4118 vattr_t dva; 4119 rnode_t *rp; 4120 mntinfo_t *mi; 4121 int douprintf; 4122 failinfo_t fi, *fip = NULL; 4123 4124 rp = VTOR(vp); 4125 mi = VTOMI(vp); 4126 ASSERT(nfs_zone() == mi->mi_zone); 4127 4128 args.dir = *RTOFH3(rp); 4129 args.cookie = (cookie3)rdc->nfs3_cookie; 4130 args.cookieverf = rp->r_cookieverf; 4131 args.dircount = rdc->buflen; 4132 args.maxcount = mi->mi_tsize; 4133 4134 /* 4135 * NFS client failover support 4136 * suppress failover unless we have a zero cookie 4137 */ 4138 if (args.cookie == (cookie3)0) { 4139 fi.vp = vp; 4140 fi.fhp = (caddr_t)&args.dir; 4141 fi.copyproc = nfs3copyfh; 4142 fi.lookupproc = nfs3lookup; 4143 fi.xattrdirproc = acl_getxattrdir3; 4144 fip = &fi; 4145 } 4146 4147 #ifdef DEBUG 4148 rdc->entries = rddir_cache_buf_alloc(rdc->buflen, KM_SLEEP); 4149 #else 4150 rdc->entries = kmem_alloc(rdc->buflen, KM_SLEEP); 4151 #endif 4152 4153 res.entries = (dirent64_t *)rdc->entries; 4154 res.entries_size = rdc->buflen; 4155 res.dir_attributes.fres.vap = &dva; 4156 res.dir_attributes.fres.vp = vp; 4157 res.loff = rdc->nfs3_cookie; 4158 res.credentials = cr; 4159 4160 douprintf = 1; 4161 4162 if (mi->mi_io_kstats) { 4163 mutex_enter(&mi->mi_lock); 4164 kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats)); 4165 mutex_exit(&mi->mi_lock); 4166 } 4167 4168 res.time = gethrtime(); 4169 4170 error = rfs3call(mi, NFSPROC3_READDIRPLUS, 4171 xdr_READDIRPLUS3args, (caddr_t)&args, 4172 xdr_READDIRPLUS3vres, (caddr_t)&res, cr, 4173 &douprintf, &res.status, 0, fip); 4174 4175 if (mi->mi_io_kstats) { 4176 mutex_enter(&mi->mi_lock); 4177 kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats)); 4178 mutex_exit(&mi->mi_lock); 4179 } 4180 4181 if (error) { 4182 goto err; 4183 } 4184 4185 nfs3_cache_post_op_vattr(vp, &res.dir_attributes, res.time, cr); 4186 4187 error = geterrno3(res.status); 4188 if (error) { 4189 PURGE_STALE_FH(error, vp, cr); 4190 if (error == EOPNOTSUPP) { 4191 mutex_enter(&mi->mi_lock); 4192 mi->mi_flags |= MI_READDIRONLY; 4193 mutex_exit(&mi->mi_lock); 4194 } 4195 goto err; 4196 } 4197 4198 if (mi->mi_io_kstats) { 4199 mutex_enter(&mi->mi_lock); 4200 KSTAT_IO_PTR(mi->mi_io_kstats)->reads++; 4201 KSTAT_IO_PTR(mi->mi_io_kstats)->nread += res.size; 4202 mutex_exit(&mi->mi_lock); 4203 } 4204 4205 rdc->nfs3_ncookie = res.loff; 4206 rp->r_cookieverf = res.cookieverf; 4207 rdc->eof = res.eof ? 1 : 0; 4208 rdc->entlen = res.size; 4209 ASSERT(rdc->entlen <= rdc->buflen); 4210 rdc->error = 0; 4211 4212 return; 4213 4214 err: 4215 kmem_free(rdc->entries, rdc->buflen); 4216 rdc->entries = NULL; 4217 rdc->error = error; 4218 } 4219 4220 #ifdef DEBUG 4221 static int nfs3_bio_do_stop = 0; 4222 #endif 4223 4224 static int 4225 nfs3_bio(struct buf *bp, stable_how *stab_comm, cred_t *cr) 4226 { 4227 rnode_t *rp = VTOR(bp->b_vp); 4228 int count; 4229 int error; 4230 cred_t *cred; 4231 offset_t offset; 4232 4233 ASSERT(nfs_zone() == VTOMI(bp->b_vp)->mi_zone); 4234 offset = ldbtob(bp->b_lblkno); 4235 4236 DTRACE_IO1(start, struct buf *, bp); 4237 4238 if (bp->b_flags & B_READ) { 4239 mutex_enter(&rp->r_statelock); 4240 if (rp->r_cred != NULL) { 4241 cred = rp->r_cred; 4242 crhold(cred); 4243 } else { 4244 rp->r_cred = cr; 4245 crhold(cr); 4246 cred = cr; 4247 crhold(cred); 4248 } 4249 mutex_exit(&rp->r_statelock); 4250 read_again: 4251 error = bp->b_error = nfs3read(bp->b_vp, bp->b_un.b_addr, 4252 offset, bp->b_bcount, &bp->b_resid, cred); 4253 crfree(cred); 4254 if (!error) { 4255 if (bp->b_resid) { 4256 /* 4257 * Didn't get it all because we hit EOF, 4258 * zero all the memory beyond the EOF. 4259 */ 4260 /* bzero(rdaddr + */ 4261 bzero(bp->b_un.b_addr + 4262 bp->b_bcount - bp->b_resid, bp->b_resid); 4263 } 4264 mutex_enter(&rp->r_statelock); 4265 if (bp->b_resid == bp->b_bcount && 4266 offset >= rp->r_size) { 4267 /* 4268 * We didn't read anything at all as we are 4269 * past EOF. Return an error indicator back 4270 * but don't destroy the pages (yet). 4271 */ 4272 error = NFS_EOF; 4273 } 4274 mutex_exit(&rp->r_statelock); 4275 } else if (error == EACCES) { 4276 mutex_enter(&rp->r_statelock); 4277 if (cred != cr) { 4278 if (rp->r_cred != NULL) 4279 crfree(rp->r_cred); 4280 rp->r_cred = cr; 4281 crhold(cr); 4282 cred = cr; 4283 crhold(cred); 4284 mutex_exit(&rp->r_statelock); 4285 goto read_again; 4286 } 4287 mutex_exit(&rp->r_statelock); 4288 } 4289 } else { 4290 if (!(rp->r_flags & RSTALE)) { 4291 mutex_enter(&rp->r_statelock); 4292 if (rp->r_cred != NULL) { 4293 cred = rp->r_cred; 4294 crhold(cred); 4295 } else { 4296 rp->r_cred = cr; 4297 crhold(cr); 4298 cred = cr; 4299 crhold(cred); 4300 } 4301 mutex_exit(&rp->r_statelock); 4302 write_again: 4303 mutex_enter(&rp->r_statelock); 4304 count = MIN(bp->b_bcount, rp->r_size - offset); 4305 mutex_exit(&rp->r_statelock); 4306 if (count < 0) 4307 cmn_err(CE_PANIC, "nfs3_bio: write count < 0"); 4308 #ifdef DEBUG 4309 if (count == 0) { 4310 zcmn_err(getzoneid(), CE_WARN, 4311 "nfs3_bio: zero length write at %lld", 4312 offset); 4313 nfs_printfhandle(&rp->r_fh); 4314 if (nfs3_bio_do_stop) 4315 debug_enter("nfs3_bio"); 4316 } 4317 #endif 4318 error = nfs3write(bp->b_vp, bp->b_un.b_addr, offset, 4319 count, cred, stab_comm); 4320 if (error == EACCES) { 4321 mutex_enter(&rp->r_statelock); 4322 if (cred != cr) { 4323 if (rp->r_cred != NULL) 4324 crfree(rp->r_cred); 4325 rp->r_cred = cr; 4326 crhold(cr); 4327 crfree(cred); 4328 cred = cr; 4329 crhold(cred); 4330 mutex_exit(&rp->r_statelock); 4331 goto write_again; 4332 } 4333 mutex_exit(&rp->r_statelock); 4334 } 4335 bp->b_error = error; 4336 if (error && error != EINTR) { 4337 /* 4338 * Don't print EDQUOT errors on the console. 4339 * Don't print asynchronous EACCES errors. 4340 * Don't print EFBIG errors. 4341 * Print all other write errors. 4342 */ 4343 if (error != EDQUOT && error != EFBIG && 4344 (error != EACCES || 4345 !(bp->b_flags & B_ASYNC))) 4346 nfs_write_error(bp->b_vp, error, cred); 4347 /* 4348 * Update r_error and r_flags as appropriate. 4349 * If the error was ESTALE, then mark the 4350 * rnode as not being writeable and save 4351 * the error status. Otherwise, save any 4352 * errors which occur from asynchronous 4353 * page invalidations. Any errors occurring 4354 * from other operations should be saved 4355 * by the caller. 4356 */ 4357 mutex_enter(&rp->r_statelock); 4358 if (error == ESTALE) { 4359 rp->r_flags |= RSTALE; 4360 if (!rp->r_error) 4361 rp->r_error = error; 4362 } else if (!rp->r_error && 4363 (bp->b_flags & 4364 (B_INVAL|B_FORCE|B_ASYNC)) == 4365 (B_INVAL|B_FORCE|B_ASYNC)) { 4366 rp->r_error = error; 4367 } 4368 mutex_exit(&rp->r_statelock); 4369 } 4370 crfree(cred); 4371 } else 4372 error = rp->r_error; 4373 } 4374 4375 if (error != 0 && error != NFS_EOF) 4376 bp->b_flags |= B_ERROR; 4377 4378 DTRACE_IO1(done, struct buf *, bp); 4379 4380 return (error); 4381 } 4382 4383 /* ARGSUSED */ 4384 static int 4385 nfs3_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) 4386 { 4387 rnode_t *rp; 4388 4389 if (nfs_zone() != VTOMI(vp)->mi_zone) 4390 return (EIO); 4391 rp = VTOR(vp); 4392 4393 if (fidp->fid_len < (ushort_t)rp->r_fh.fh_len) { 4394 fidp->fid_len = rp->r_fh.fh_len; 4395 return (ENOSPC); 4396 } 4397 fidp->fid_len = rp->r_fh.fh_len; 4398 bcopy(rp->r_fh.fh_buf, fidp->fid_data, fidp->fid_len); 4399 return (0); 4400 } 4401 4402 /* ARGSUSED2 */ 4403 static int 4404 nfs3_rwlock(vnode_t *vp, int write_lock, caller_context_t *ctp) 4405 { 4406 rnode_t *rp = VTOR(vp); 4407 4408 if (!write_lock) { 4409 (void) nfs_rw_enter_sig(&rp->r_rwlock, RW_READER, FALSE); 4410 return (V_WRITELOCK_FALSE); 4411 } 4412 4413 if ((rp->r_flags & RDIRECTIO) || (VTOMI(vp)->mi_flags & MI_DIRECTIO)) { 4414 (void) nfs_rw_enter_sig(&rp->r_rwlock, RW_READER, FALSE); 4415 if (rp->r_mapcnt == 0 && !vn_has_cached_data(vp)) 4416 return (V_WRITELOCK_FALSE); 4417 nfs_rw_exit(&rp->r_rwlock); 4418 } 4419 4420 (void) nfs_rw_enter_sig(&rp->r_rwlock, RW_WRITER, FALSE); 4421 return (V_WRITELOCK_TRUE); 4422 } 4423 4424 /* ARGSUSED */ 4425 static void 4426 nfs3_rwunlock(vnode_t *vp, int write_lock, caller_context_t *ctp) 4427 { 4428 rnode_t *rp = VTOR(vp); 4429 4430 nfs_rw_exit(&rp->r_rwlock); 4431 } 4432 4433 /* ARGSUSED */ 4434 static int 4435 nfs3_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, caller_context_t *ct) 4436 { 4437 4438 /* 4439 * Because we stuff the readdir cookie into the offset field 4440 * someone may attempt to do an lseek with the cookie which 4441 * we want to succeed. 4442 */ 4443 if (vp->v_type == VDIR) 4444 return (0); 4445 if (*noffp < 0) 4446 return (EINVAL); 4447 return (0); 4448 } 4449 4450 /* 4451 * number of nfs3_bsize blocks to read ahead. 4452 */ 4453 static int nfs3_nra = 4; 4454 4455 #ifdef DEBUG 4456 static int nfs3_lostpage = 0; /* number of times we lost original page */ 4457 #endif 4458 4459 /* 4460 * Return all the pages from [off..off+len) in file 4461 */ 4462 /* ARGSUSED */ 4463 static int 4464 nfs3_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 4465 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 4466 enum seg_rw rw, cred_t *cr, caller_context_t *ct) 4467 { 4468 rnode_t *rp; 4469 int error; 4470 mntinfo_t *mi; 4471 4472 if (vp->v_flag & VNOMAP) 4473 return (ENOSYS); 4474 4475 if (nfs_zone() != VTOMI(vp)->mi_zone) 4476 return (EIO); 4477 if (protp != NULL) 4478 *protp = PROT_ALL; 4479 4480 /* 4481 * Now valididate that the caches are up to date. 4482 */ 4483 error = nfs3_validate_caches(vp, cr); 4484 if (error) 4485 return (error); 4486 4487 rp = VTOR(vp); 4488 mi = VTOMI(vp); 4489 retry: 4490 mutex_enter(&rp->r_statelock); 4491 4492 /* 4493 * Don't create dirty pages faster than they 4494 * can be cleaned so that the system doesn't 4495 * get imbalanced. If the async queue is 4496 * maxed out, then wait for it to drain before 4497 * creating more dirty pages. Also, wait for 4498 * any threads doing pagewalks in the vop_getattr 4499 * entry points so that they don't block for 4500 * long periods. 4501 */ 4502 if (rw == S_CREATE) { 4503 while ((mi->mi_max_threads != 0 && 4504 rp->r_awcount > 2 * mi->mi_max_threads) || 4505 rp->r_gcount > 0) 4506 cv_wait(&rp->r_cv, &rp->r_statelock); 4507 } 4508 4509 /* 4510 * If we are getting called as a side effect of an nfs_write() 4511 * operation the local file size might not be extended yet. 4512 * In this case we want to be able to return pages of zeroes. 4513 */ 4514 if (off + len > rp->r_size + PAGEOFFSET && seg != segkmap) { 4515 mutex_exit(&rp->r_statelock); 4516 return (EFAULT); /* beyond EOF */ 4517 } 4518 4519 mutex_exit(&rp->r_statelock); 4520 4521 if (len <= PAGESIZE) { 4522 error = nfs3_getapage(vp, off, len, protp, pl, plsz, 4523 seg, addr, rw, cr); 4524 } else { 4525 error = pvn_getpages(nfs3_getapage, vp, off, len, protp, 4526 pl, plsz, seg, addr, rw, cr); 4527 } 4528 4529 switch (error) { 4530 case NFS_EOF: 4531 nfs_purge_caches(vp, NFS_NOPURGE_DNLC, cr); 4532 goto retry; 4533 case ESTALE: 4534 PURGE_STALE_FH(error, vp, cr); 4535 } 4536 4537 return (error); 4538 } 4539 4540 /* 4541 * Called from pvn_getpages or nfs3_getpage to get a particular page. 4542 */ 4543 /* ARGSUSED */ 4544 static int 4545 nfs3_getapage(vnode_t *vp, u_offset_t off, size_t len, uint_t *protp, 4546 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 4547 enum seg_rw rw, cred_t *cr) 4548 { 4549 rnode_t *rp; 4550 uint_t bsize; 4551 struct buf *bp; 4552 page_t *pp; 4553 u_offset_t lbn; 4554 u_offset_t io_off; 4555 u_offset_t blkoff; 4556 u_offset_t rablkoff; 4557 size_t io_len; 4558 uint_t blksize; 4559 int error; 4560 int readahead; 4561 int readahead_issued = 0; 4562 int ra_window; /* readahead window */ 4563 page_t *pagefound; 4564 page_t *savepp; 4565 4566 if (nfs_zone() != VTOMI(vp)->mi_zone) 4567 return (EIO); 4568 rp = VTOR(vp); 4569 bsize = MAX(vp->v_vfsp->vfs_bsize, PAGESIZE); 4570 4571 reread: 4572 bp = NULL; 4573 pp = NULL; 4574 pagefound = NULL; 4575 4576 if (pl != NULL) 4577 pl[0] = NULL; 4578 4579 error = 0; 4580 lbn = off / bsize; 4581 blkoff = lbn * bsize; 4582 4583 /* 4584 * Queueing up the readahead before doing the synchronous read 4585 * results in a significant increase in read throughput because 4586 * of the increased parallelism between the async threads and 4587 * the process context. 4588 */ 4589 if ((off & ((vp->v_vfsp->vfs_bsize) - 1)) == 0 && 4590 rw != S_CREATE && 4591 !(vp->v_flag & VNOCACHE)) { 4592 mutex_enter(&rp->r_statelock); 4593 4594 /* 4595 * Calculate the number of readaheads to do. 4596 * a) No readaheads at offset = 0. 4597 * b) Do maximum(nfs3_nra) readaheads when the readahead 4598 * window is closed. 4599 * c) Do readaheads between 1 to (nfs3_nra - 1) depending 4600 * upon how far the readahead window is open or close. 4601 * d) No readaheads if rp->r_nextr is not within the scope 4602 * of the readahead window (random i/o). 4603 */ 4604 4605 if (off == 0) 4606 readahead = 0; 4607 else if (blkoff == rp->r_nextr) 4608 readahead = nfs3_nra; 4609 else if (rp->r_nextr > blkoff && 4610 ((ra_window = (rp->r_nextr - blkoff) / bsize) 4611 <= (nfs3_nra - 1))) 4612 readahead = nfs3_nra - ra_window; 4613 else 4614 readahead = 0; 4615 4616 rablkoff = rp->r_nextr; 4617 while (readahead > 0 && rablkoff + bsize < rp->r_size) { 4618 mutex_exit(&rp->r_statelock); 4619 if (nfs_async_readahead(vp, rablkoff + bsize, 4620 addr + (rablkoff + bsize - off), seg, cr, 4621 nfs3_readahead) < 0) { 4622 mutex_enter(&rp->r_statelock); 4623 break; 4624 } 4625 readahead--; 4626 rablkoff += bsize; 4627 /* 4628 * Indicate that we did a readahead so 4629 * readahead offset is not updated 4630 * by the synchronous read below. 4631 */ 4632 readahead_issued = 1; 4633 mutex_enter(&rp->r_statelock); 4634 /* 4635 * set readahead offset to 4636 * offset of last async readahead 4637 * request. 4638 */ 4639 rp->r_nextr = rablkoff; 4640 } 4641 mutex_exit(&rp->r_statelock); 4642 } 4643 4644 again: 4645 if ((pagefound = page_exists(vp, off)) == NULL) { 4646 if (pl == NULL) { 4647 (void) nfs_async_readahead(vp, blkoff, addr, seg, cr, 4648 nfs3_readahead); 4649 } else if (rw == S_CREATE) { 4650 /* 4651 * Block for this page is not allocated, or the offset 4652 * is beyond the current allocation size, or we're 4653 * allocating a swap slot and the page was not found, 4654 * so allocate it and return a zero page. 4655 */ 4656 if ((pp = page_create_va(vp, off, 4657 PAGESIZE, PG_WAIT, seg, addr)) == NULL) 4658 cmn_err(CE_PANIC, "nfs3_getapage: page_create"); 4659 io_len = PAGESIZE; 4660 mutex_enter(&rp->r_statelock); 4661 rp->r_nextr = off + PAGESIZE; 4662 mutex_exit(&rp->r_statelock); 4663 } else { 4664 /* 4665 * Need to go to server to get a BLOCK, exception to 4666 * that being while reading at offset = 0 or doing 4667 * random i/o, in that case read only a PAGE. 4668 */ 4669 mutex_enter(&rp->r_statelock); 4670 if (blkoff < rp->r_size && 4671 blkoff + bsize >= rp->r_size) { 4672 /* 4673 * If only a block or less is left in 4674 * the file, read all that is remaining. 4675 */ 4676 if (rp->r_size <= off) { 4677 /* 4678 * Trying to access beyond EOF, 4679 * set up to get at least one page. 4680 */ 4681 blksize = off + PAGESIZE - blkoff; 4682 } else 4683 blksize = rp->r_size - blkoff; 4684 } else if ((off == 0) || 4685 (off != rp->r_nextr && !readahead_issued)) { 4686 blksize = PAGESIZE; 4687 blkoff = off; /* block = page here */ 4688 } else 4689 blksize = bsize; 4690 mutex_exit(&rp->r_statelock); 4691 4692 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 4693 &io_len, blkoff, blksize, 0); 4694 4695 /* 4696 * Some other thread has entered the page, 4697 * so just use it. 4698 */ 4699 if (pp == NULL) 4700 goto again; 4701 4702 /* 4703 * Now round the request size up to page boundaries. 4704 * This ensures that the entire page will be 4705 * initialized to zeroes if EOF is encountered. 4706 */ 4707 io_len = ptob(btopr(io_len)); 4708 4709 bp = pageio_setup(pp, io_len, vp, B_READ); 4710 ASSERT(bp != NULL); 4711 4712 /* 4713 * pageio_setup should have set b_addr to 0. This 4714 * is correct since we want to do I/O on a page 4715 * boundary. bp_mapin will use this addr to calculate 4716 * an offset, and then set b_addr to the kernel virtual 4717 * address it allocated for us. 4718 */ 4719 ASSERT(bp->b_un.b_addr == 0); 4720 4721 bp->b_edev = 0; 4722 bp->b_dev = 0; 4723 bp->b_lblkno = lbtodb(io_off); 4724 bp->b_file = vp; 4725 bp->b_offset = (offset_t)off; 4726 bp_mapin(bp); 4727 4728 /* 4729 * If doing a write beyond what we believe is EOF, 4730 * don't bother trying to read the pages from the 4731 * server, we'll just zero the pages here. We 4732 * don't check that the rw flag is S_WRITE here 4733 * because some implementations may attempt a 4734 * read access to the buffer before copying data. 4735 */ 4736 mutex_enter(&rp->r_statelock); 4737 if (io_off >= rp->r_size && seg == segkmap) { 4738 mutex_exit(&rp->r_statelock); 4739 bzero(bp->b_un.b_addr, io_len); 4740 } else { 4741 mutex_exit(&rp->r_statelock); 4742 error = nfs3_bio(bp, NULL, cr); 4743 } 4744 4745 /* 4746 * Unmap the buffer before freeing it. 4747 */ 4748 bp_mapout(bp); 4749 pageio_done(bp); 4750 4751 savepp = pp; 4752 do { 4753 pp->p_fsdata = C_NOCOMMIT; 4754 } while ((pp = pp->p_next) != savepp); 4755 4756 if (error == NFS_EOF) { 4757 /* 4758 * If doing a write system call just return 4759 * zeroed pages, else user tried to get pages 4760 * beyond EOF, return error. We don't check 4761 * that the rw flag is S_WRITE here because 4762 * some implementations may attempt a read 4763 * access to the buffer before copying data. 4764 */ 4765 if (seg == segkmap) 4766 error = 0; 4767 else 4768 error = EFAULT; 4769 } 4770 4771 if (!readahead_issued && !error) { 4772 mutex_enter(&rp->r_statelock); 4773 rp->r_nextr = io_off + io_len; 4774 mutex_exit(&rp->r_statelock); 4775 } 4776 } 4777 } 4778 4779 out: 4780 if (pl == NULL) 4781 return (error); 4782 4783 if (error) { 4784 if (pp != NULL) 4785 pvn_read_done(pp, B_ERROR); 4786 return (error); 4787 } 4788 4789 if (pagefound) { 4790 se_t se = (rw == S_CREATE ? SE_EXCL : SE_SHARED); 4791 4792 /* 4793 * Page exists in the cache, acquire the appropriate lock. 4794 * If this fails, start all over again. 4795 */ 4796 if ((pp = page_lookup(vp, off, se)) == NULL) { 4797 #ifdef DEBUG 4798 nfs3_lostpage++; 4799 #endif 4800 goto reread; 4801 } 4802 pl[0] = pp; 4803 pl[1] = NULL; 4804 return (0); 4805 } 4806 4807 if (pp != NULL) 4808 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 4809 4810 return (error); 4811 } 4812 4813 static void 4814 nfs3_readahead(vnode_t *vp, u_offset_t blkoff, caddr_t addr, struct seg *seg, 4815 cred_t *cr) 4816 { 4817 int error; 4818 page_t *pp; 4819 u_offset_t io_off; 4820 size_t io_len; 4821 struct buf *bp; 4822 uint_t bsize, blksize; 4823 rnode_t *rp = VTOR(vp); 4824 page_t *savepp; 4825 4826 ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 4827 bsize = MAX(vp->v_vfsp->vfs_bsize, PAGESIZE); 4828 4829 mutex_enter(&rp->r_statelock); 4830 if (blkoff < rp->r_size && blkoff + bsize > rp->r_size) { 4831 /* 4832 * If less than a block left in file read less 4833 * than a block. 4834 */ 4835 blksize = rp->r_size - blkoff; 4836 } else 4837 blksize = bsize; 4838 mutex_exit(&rp->r_statelock); 4839 4840 pp = pvn_read_kluster(vp, blkoff, segkmap, addr, 4841 &io_off, &io_len, blkoff, blksize, 1); 4842 /* 4843 * The isra flag passed to the kluster function is 1, we may have 4844 * gotten a return value of NULL for a variety of reasons (# of free 4845 * pages < minfree, someone entered the page on the vnode etc). In all 4846 * cases, we want to punt on the readahead. 4847 */ 4848 if (pp == NULL) 4849 return; 4850 4851 /* 4852 * Now round the request size up to page boundaries. 4853 * This ensures that the entire page will be 4854 * initialized to zeroes if EOF is encountered. 4855 */ 4856 io_len = ptob(btopr(io_len)); 4857 4858 bp = pageio_setup(pp, io_len, vp, B_READ); 4859 ASSERT(bp != NULL); 4860 4861 /* 4862 * pageio_setup should have set b_addr to 0. This is correct since 4863 * we want to do I/O on a page boundary. bp_mapin() will use this addr 4864 * to calculate an offset, and then set b_addr to the kernel virtual 4865 * address it allocated for us. 4866 */ 4867 ASSERT(bp->b_un.b_addr == 0); 4868 4869 bp->b_edev = 0; 4870 bp->b_dev = 0; 4871 bp->b_lblkno = lbtodb(io_off); 4872 bp->b_file = vp; 4873 bp->b_offset = (offset_t)blkoff; 4874 bp_mapin(bp); 4875 4876 /* 4877 * If doing a write beyond what we believe is EOF, don't bother trying 4878 * to read the pages from the server, we'll just zero the pages here. 4879 * We don't check that the rw flag is S_WRITE here because some 4880 * implementations may attempt a read access to the buffer before 4881 * copying data. 4882 */ 4883 mutex_enter(&rp->r_statelock); 4884 if (io_off >= rp->r_size && seg == segkmap) { 4885 mutex_exit(&rp->r_statelock); 4886 bzero(bp->b_un.b_addr, io_len); 4887 error = 0; 4888 } else { 4889 mutex_exit(&rp->r_statelock); 4890 error = nfs3_bio(bp, NULL, cr); 4891 if (error == NFS_EOF) 4892 error = 0; 4893 } 4894 4895 /* 4896 * Unmap the buffer before freeing it. 4897 */ 4898 bp_mapout(bp); 4899 pageio_done(bp); 4900 4901 savepp = pp; 4902 do { 4903 pp->p_fsdata = C_NOCOMMIT; 4904 } while ((pp = pp->p_next) != savepp); 4905 4906 pvn_read_done(pp, error ? B_READ | B_ERROR : B_READ); 4907 4908 /* 4909 * In case of error set readahead offset 4910 * to the lowest offset. 4911 * pvn_read_done() calls VN_DISPOSE to destroy the pages 4912 */ 4913 if (error && rp->r_nextr > io_off) { 4914 mutex_enter(&rp->r_statelock); 4915 if (rp->r_nextr > io_off) 4916 rp->r_nextr = io_off; 4917 mutex_exit(&rp->r_statelock); 4918 } 4919 } 4920 4921 /* 4922 * Flags are composed of {B_INVAL, B_FREE, B_DONTNEED, B_FORCE} 4923 * If len == 0, do from off to EOF. 4924 * 4925 * The normal cases should be len == 0 && off == 0 (entire vp list), 4926 * len == MAXBSIZE (from segmap_release actions), and len == PAGESIZE 4927 * (from pageout). 4928 */ 4929 /* ARGSUSED */ 4930 static int 4931 nfs3_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr, 4932 caller_context_t *ct) 4933 { 4934 int error; 4935 rnode_t *rp; 4936 4937 ASSERT(cr != NULL); 4938 4939 /* 4940 * XXX - Why should this check be made here? 4941 */ 4942 if (vp->v_flag & VNOMAP) 4943 return (ENOSYS); 4944 if (len == 0 && !(flags & B_INVAL) && vn_is_readonly(vp)) 4945 return (0); 4946 if (!(flags & B_ASYNC) && nfs_zone() != VTOMI(vp)->mi_zone) 4947 return (EIO); 4948 4949 rp = VTOR(vp); 4950 mutex_enter(&rp->r_statelock); 4951 rp->r_count++; 4952 mutex_exit(&rp->r_statelock); 4953 error = nfs_putpages(vp, off, len, flags, cr); 4954 mutex_enter(&rp->r_statelock); 4955 rp->r_count--; 4956 cv_broadcast(&rp->r_cv); 4957 mutex_exit(&rp->r_statelock); 4958 4959 return (error); 4960 } 4961 4962 /* 4963 * Write out a single page, possibly klustering adjacent dirty pages. 4964 */ 4965 int 4966 nfs3_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp, 4967 int flags, cred_t *cr) 4968 { 4969 u_offset_t io_off; 4970 u_offset_t lbn_off; 4971 u_offset_t lbn; 4972 size_t io_len; 4973 uint_t bsize; 4974 int error; 4975 rnode_t *rp; 4976 4977 ASSERT(!vn_is_readonly(vp)); 4978 ASSERT(pp != NULL); 4979 ASSERT(cr != NULL); 4980 ASSERT((flags & B_ASYNC) || nfs_zone() == VTOMI(vp)->mi_zone); 4981 4982 rp = VTOR(vp); 4983 ASSERT(rp->r_count > 0); 4984 4985 bsize = MAX(vp->v_vfsp->vfs_bsize, PAGESIZE); 4986 lbn = pp->p_offset / bsize; 4987 lbn_off = lbn * bsize; 4988 4989 /* 4990 * Find a kluster that fits in one block, or in 4991 * one page if pages are bigger than blocks. If 4992 * there is less file space allocated than a whole 4993 * page, we'll shorten the i/o request below. 4994 */ 4995 pp = pvn_write_kluster(vp, pp, &io_off, &io_len, lbn_off, 4996 roundup(bsize, PAGESIZE), flags); 4997 4998 /* 4999 * pvn_write_kluster shouldn't have returned a page with offset 5000 * behind the original page we were given. Verify that. 5001 */ 5002 ASSERT((pp->p_offset / bsize) >= lbn); 5003 5004 /* 5005 * Now pp will have the list of kept dirty pages marked for 5006 * write back. It will also handle invalidation and freeing 5007 * of pages that are not dirty. Check for page length rounding 5008 * problems. 5009 */ 5010 if (io_off + io_len > lbn_off + bsize) { 5011 ASSERT((io_off + io_len) - (lbn_off + bsize) < PAGESIZE); 5012 io_len = lbn_off + bsize - io_off; 5013 } 5014 /* 5015 * The RMODINPROGRESS flag makes sure that nfs(3)_bio() sees a 5016 * consistent value of r_size. RMODINPROGRESS is set in writerp(). 5017 * When RMODINPROGRESS is set it indicates that a uiomove() is in 5018 * progress and the r_size has not been made consistent with the 5019 * new size of the file. When the uiomove() completes the r_size is 5020 * updated and the RMODINPROGRESS flag is cleared. 5021 * 5022 * The RMODINPROGRESS flag makes sure that nfs(3)_bio() sees a 5023 * consistent value of r_size. Without this handshaking, it is 5024 * possible that nfs(3)_bio() picks up the old value of r_size 5025 * before the uiomove() in writerp() completes. This will result 5026 * in the write through nfs(3)_bio() being dropped. 5027 * 5028 * More precisely, there is a window between the time the uiomove() 5029 * completes and the time the r_size is updated. If a VOP_PUTPAGE() 5030 * operation intervenes in this window, the page will be picked up, 5031 * because it is dirty (it will be unlocked, unless it was 5032 * pagecreate'd). When the page is picked up as dirty, the dirty 5033 * bit is reset (pvn_getdirty()). In nfs(3)write(), r_size is 5034 * checked. This will still be the old size. Therefore the page will 5035 * not be written out. When segmap_release() calls VOP_PUTPAGE(), 5036 * the page will be found to be clean and the write will be dropped. 5037 */ 5038 if (rp->r_flags & RMODINPROGRESS) { 5039 mutex_enter(&rp->r_statelock); 5040 if ((rp->r_flags & RMODINPROGRESS) && 5041 rp->r_modaddr + MAXBSIZE > io_off && 5042 rp->r_modaddr < io_off + io_len) { 5043 page_t *plist; 5044 /* 5045 * A write is in progress for this region of the file. 5046 * If we did not detect RMODINPROGRESS here then this 5047 * path through nfs_putapage() would eventually go to 5048 * nfs(3)_bio() and may not write out all of the data 5049 * in the pages. We end up losing data. So we decide 5050 * to set the modified bit on each page in the page 5051 * list and mark the rnode with RDIRTY. This write 5052 * will be restarted at some later time. 5053 */ 5054 plist = pp; 5055 while (plist != NULL) { 5056 pp = plist; 5057 page_sub(&plist, pp); 5058 hat_setmod(pp); 5059 page_io_unlock(pp); 5060 page_unlock(pp); 5061 } 5062 rp->r_flags |= RDIRTY; 5063 mutex_exit(&rp->r_statelock); 5064 if (offp) 5065 *offp = io_off; 5066 if (lenp) 5067 *lenp = io_len; 5068 return (0); 5069 } 5070 mutex_exit(&rp->r_statelock); 5071 } 5072 5073 if (flags & B_ASYNC) { 5074 error = nfs_async_putapage(vp, pp, io_off, io_len, flags, cr, 5075 nfs3_sync_putapage); 5076 } else 5077 error = nfs3_sync_putapage(vp, pp, io_off, io_len, flags, cr); 5078 5079 if (offp) 5080 *offp = io_off; 5081 if (lenp) 5082 *lenp = io_len; 5083 return (error); 5084 } 5085 5086 static int 5087 nfs3_sync_putapage(vnode_t *vp, page_t *pp, u_offset_t io_off, size_t io_len, 5088 int flags, cred_t *cr) 5089 { 5090 int error; 5091 rnode_t *rp; 5092 5093 ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 5094 5095 flags |= B_WRITE; 5096 5097 error = nfs3_rdwrlbn(vp, pp, io_off, io_len, flags, cr); 5098 5099 rp = VTOR(vp); 5100 5101 if ((error == ENOSPC || error == EDQUOT || error == EFBIG || 5102 error == EACCES) && 5103 (flags & (B_INVAL|B_FORCE)) != (B_INVAL|B_FORCE)) { 5104 if (!(rp->r_flags & ROUTOFSPACE)) { 5105 mutex_enter(&rp->r_statelock); 5106 rp->r_flags |= ROUTOFSPACE; 5107 mutex_exit(&rp->r_statelock); 5108 } 5109 flags |= B_ERROR; 5110 pvn_write_done(pp, flags); 5111 /* 5112 * If this was not an async thread, then try again to 5113 * write out the pages, but this time, also destroy 5114 * them whether or not the write is successful. This 5115 * will prevent memory from filling up with these 5116 * pages and destroying them is the only alternative 5117 * if they can't be written out. 5118 * 5119 * Don't do this if this is an async thread because 5120 * when the pages are unlocked in pvn_write_done, 5121 * some other thread could have come along, locked 5122 * them, and queued for an async thread. It would be 5123 * possible for all of the async threads to be tied 5124 * up waiting to lock the pages again and they would 5125 * all already be locked and waiting for an async 5126 * thread to handle them. Deadlock. 5127 */ 5128 if (!(flags & B_ASYNC)) { 5129 error = nfs3_putpage(vp, io_off, io_len, 5130 B_INVAL | B_FORCE, cr, NULL); 5131 } 5132 } else { 5133 if (error) 5134 flags |= B_ERROR; 5135 else if (rp->r_flags & ROUTOFSPACE) { 5136 mutex_enter(&rp->r_statelock); 5137 rp->r_flags &= ~ROUTOFSPACE; 5138 mutex_exit(&rp->r_statelock); 5139 } 5140 pvn_write_done(pp, flags); 5141 if (freemem < desfree) 5142 (void) nfs3_commit_vp(vp, (u_offset_t)0, 0, cr); 5143 } 5144 5145 return (error); 5146 } 5147 5148 /* ARGSUSED */ 5149 static int 5150 nfs3_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 5151 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, 5152 cred_t *cr, caller_context_t *ct) 5153 { 5154 struct segvn_crargs vn_a; 5155 int error; 5156 rnode_t *rp; 5157 struct vattr va; 5158 5159 if (nfs_zone() != VTOMI(vp)->mi_zone) 5160 return (EIO); 5161 5162 if (vp->v_flag & VNOMAP) 5163 return (ENOSYS); 5164 5165 if (off < 0 || off + len < 0) 5166 return (ENXIO); 5167 5168 if (vp->v_type != VREG) 5169 return (ENODEV); 5170 5171 /* 5172 * If there is cached data and if close-to-open consistency 5173 * checking is not turned off and if the file system is not 5174 * mounted readonly, then force an over the wire getattr. 5175 * Otherwise, just invoke nfs3getattr to get a copy of the 5176 * attributes. The attribute cache will be used unless it 5177 * is timed out and if it is, then an over the wire getattr 5178 * will be issued. 5179 */ 5180 va.va_mask = AT_ALL; 5181 if (vn_has_cached_data(vp) && 5182 !(VTOMI(vp)->mi_flags & MI_NOCTO) && !vn_is_readonly(vp)) 5183 error = nfs3_getattr_otw(vp, &va, cr); 5184 else 5185 error = nfs3getattr(vp, &va, cr); 5186 if (error) 5187 return (error); 5188 5189 /* 5190 * Check to see if the vnode is currently marked as not cachable. 5191 * This means portions of the file are locked (through VOP_FRLOCK). 5192 * In this case the map request must be refused. We use 5193 * rp->r_lkserlock to avoid a race with concurrent lock requests. 5194 */ 5195 rp = VTOR(vp); 5196 if (nfs_rw_enter_sig(&rp->r_lkserlock, RW_READER, INTR(vp))) 5197 return (EINTR); 5198 5199 if (vp->v_flag & VNOCACHE) { 5200 error = EAGAIN; 5201 goto done; 5202 } 5203 5204 /* 5205 * Don't allow concurrent locks and mapping if mandatory locking is 5206 * enabled. 5207 */ 5208 if ((flk_has_remote_locks(vp) || lm_has_sleep(vp)) && 5209 MANDLOCK(vp, va.va_mode)) { 5210 error = EAGAIN; 5211 goto done; 5212 } 5213 5214 as_rangelock(as); 5215 if (!(flags & MAP_FIXED)) { 5216 map_addr(addrp, len, off, 1, flags); 5217 if (*addrp == NULL) { 5218 as_rangeunlock(as); 5219 error = ENOMEM; 5220 goto done; 5221 } 5222 } else { 5223 /* 5224 * User specified address - blow away any previous mappings 5225 */ 5226 (void) as_unmap(as, *addrp, len); 5227 } 5228 5229 vn_a.vp = vp; 5230 vn_a.offset = off; 5231 vn_a.type = (flags & MAP_TYPE); 5232 vn_a.prot = (uchar_t)prot; 5233 vn_a.maxprot = (uchar_t)maxprot; 5234 vn_a.flags = (flags & ~MAP_TYPE); 5235 vn_a.cred = cr; 5236 vn_a.amp = NULL; 5237 vn_a.szc = 0; 5238 vn_a.lgrp_mem_policy_flags = 0; 5239 5240 error = as_map(as, *addrp, len, segvn_create, &vn_a); 5241 as_rangeunlock(as); 5242 5243 done: 5244 nfs_rw_exit(&rp->r_lkserlock); 5245 return (error); 5246 } 5247 5248 /* ARGSUSED */ 5249 static int 5250 nfs3_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 5251 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, 5252 cred_t *cr, caller_context_t *ct) 5253 { 5254 rnode_t *rp; 5255 5256 if (vp->v_flag & VNOMAP) 5257 return (ENOSYS); 5258 if (nfs_zone() != VTOMI(vp)->mi_zone) 5259 return (EIO); 5260 5261 /* 5262 * Need to hold rwlock while incrementing the mapcnt so that 5263 * mmap'ing can be serialized with writes so that the caching 5264 * can be handled correctly. 5265 */ 5266 rp = VTOR(vp); 5267 if (nfs_rw_enter_sig(&rp->r_rwlock, RW_WRITER, INTR(vp))) 5268 return (EINTR); 5269 atomic_add_long((ulong_t *)&rp->r_mapcnt, btopr(len)); 5270 nfs_rw_exit(&rp->r_rwlock); 5271 5272 return (0); 5273 } 5274 5275 /* ARGSUSED */ 5276 static int 5277 nfs3_frlock(vnode_t *vp, int cmd, struct flock64 *bfp, int flag, 5278 offset_t offset, struct flk_callback *flk_cbp, cred_t *cr, 5279 caller_context_t *ct) 5280 { 5281 netobj lm_fh3; 5282 int rc; 5283 u_offset_t start, end; 5284 rnode_t *rp; 5285 int error = 0, intr = INTR(vp); 5286 5287 if (nfs_zone() != VTOMI(vp)->mi_zone) 5288 return (EIO); 5289 /* check for valid cmd parameter */ 5290 if (cmd != F_GETLK && cmd != F_SETLK && cmd != F_SETLKW) 5291 return (EINVAL); 5292 5293 /* Verify l_type. */ 5294 switch (bfp->l_type) { 5295 case F_RDLCK: 5296 if (cmd != F_GETLK && !(flag & FREAD)) 5297 return (EBADF); 5298 break; 5299 case F_WRLCK: 5300 if (cmd != F_GETLK && !(flag & FWRITE)) 5301 return (EBADF); 5302 break; 5303 case F_UNLCK: 5304 intr = 0; 5305 break; 5306 5307 default: 5308 return (EINVAL); 5309 } 5310 5311 /* check the validity of the lock range */ 5312 if (rc = flk_convert_lock_data(vp, bfp, &start, &end, offset)) 5313 return (rc); 5314 if (rc = flk_check_lock_data(start, end, MAXEND)) 5315 return (rc); 5316 5317 /* 5318 * If the filesystem is mounted using local locking, pass the 5319 * request off to the local locking code. 5320 */ 5321 if (VTOMI(vp)->mi_flags & MI_LLOCK) { 5322 if (cmd == F_SETLK || cmd == F_SETLKW) { 5323 /* 5324 * For complete safety, we should be holding 5325 * r_lkserlock. However, we can't call 5326 * lm_safelock and then fs_frlock while 5327 * holding r_lkserlock, so just invoke 5328 * lm_safelock and expect that this will 5329 * catch enough of the cases. 5330 */ 5331 if (!lm_safelock(vp, bfp, cr)) 5332 return (EAGAIN); 5333 } 5334 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct)); 5335 } 5336 5337 rp = VTOR(vp); 5338 5339 /* 5340 * Check whether the given lock request can proceed, given the 5341 * current file mappings. 5342 */ 5343 if (nfs_rw_enter_sig(&rp->r_lkserlock, RW_WRITER, intr)) 5344 return (EINTR); 5345 if (cmd == F_SETLK || cmd == F_SETLKW) { 5346 if (!lm_safelock(vp, bfp, cr)) { 5347 rc = EAGAIN; 5348 goto done; 5349 } 5350 } 5351 5352 /* 5353 * Flush the cache after waiting for async I/O to finish. For new 5354 * locks, this is so that the process gets the latest bits from the 5355 * server. For unlocks, this is so that other clients see the 5356 * latest bits once the file has been unlocked. If currently dirty 5357 * pages can't be flushed, then don't allow a lock to be set. But 5358 * allow unlocks to succeed, to avoid having orphan locks on the 5359 * server. 5360 */ 5361 if (cmd != F_GETLK) { 5362 mutex_enter(&rp->r_statelock); 5363 while (rp->r_count > 0) { 5364 if (intr) { 5365 klwp_t *lwp = ttolwp(curthread); 5366 5367 if (lwp != NULL) 5368 lwp->lwp_nostop++; 5369 if (cv_wait_sig(&rp->r_cv, &rp->r_statelock) == 0) { 5370 if (lwp != NULL) 5371 lwp->lwp_nostop--; 5372 rc = EINTR; 5373 break; 5374 } 5375 if (lwp != NULL) 5376 lwp->lwp_nostop--; 5377 } else 5378 cv_wait(&rp->r_cv, &rp->r_statelock); 5379 } 5380 mutex_exit(&rp->r_statelock); 5381 if (rc != 0) 5382 goto done; 5383 error = nfs3_putpage(vp, (offset_t)0, 0, B_INVAL, cr, ct); 5384 if (error) { 5385 if (error == ENOSPC || error == EDQUOT) { 5386 mutex_enter(&rp->r_statelock); 5387 if (!rp->r_error) 5388 rp->r_error = error; 5389 mutex_exit(&rp->r_statelock); 5390 } 5391 if (bfp->l_type != F_UNLCK) { 5392 rc = ENOLCK; 5393 goto done; 5394 } 5395 } 5396 } 5397 5398 lm_fh3.n_len = VTOFH3(vp)->fh3_length; 5399 lm_fh3.n_bytes = (char *)&(VTOFH3(vp)->fh3_u.data); 5400 5401 /* 5402 * Call the lock manager to do the real work of contacting 5403 * the server and obtaining the lock. 5404 */ 5405 rc = lm4_frlock(vp, cmd, bfp, flag, offset, cr, &lm_fh3, flk_cbp); 5406 5407 if (rc == 0) 5408 nfs_lockcompletion(vp, cmd); 5409 5410 done: 5411 nfs_rw_exit(&rp->r_lkserlock); 5412 return (rc); 5413 } 5414 5415 /* 5416 * Free storage space associated with the specified vnode. The portion 5417 * to be freed is specified by bfp->l_start and bfp->l_len (already 5418 * normalized to a "whence" of 0). 5419 * 5420 * This is an experimental facility whose continued existence is not 5421 * guaranteed. Currently, we only support the special case 5422 * of l_len == 0, meaning free to end of file. 5423 */ 5424 /* ARGSUSED */ 5425 static int 5426 nfs3_space(vnode_t *vp, int cmd, struct flock64 *bfp, int flag, 5427 offset_t offset, cred_t *cr, caller_context_t *ct) 5428 { 5429 int error; 5430 5431 ASSERT(vp->v_type == VREG); 5432 if (cmd != F_FREESP) 5433 return (EINVAL); 5434 if (nfs_zone() != VTOMI(vp)->mi_zone) 5435 return (EIO); 5436 5437 error = convoff(vp, bfp, 0, offset); 5438 if (!error) { 5439 ASSERT(bfp->l_start >= 0); 5440 if (bfp->l_len == 0) { 5441 struct vattr va; 5442 5443 /* 5444 * ftruncate should not change the ctime and 5445 * mtime if we truncate the file to its 5446 * previous size. 5447 */ 5448 va.va_mask = AT_SIZE; 5449 error = nfs3getattr(vp, &va, cr); 5450 if (error || va.va_size == bfp->l_start) 5451 return (error); 5452 va.va_mask = AT_SIZE; 5453 va.va_size = bfp->l_start; 5454 error = nfs3setattr(vp, &va, 0, cr); 5455 } else 5456 error = EINVAL; 5457 } 5458 5459 return (error); 5460 } 5461 5462 /* ARGSUSED */ 5463 static int 5464 nfs3_realvp(vnode_t *vp, vnode_t **vpp, caller_context_t *ct) 5465 { 5466 5467 return (EINVAL); 5468 } 5469 5470 /* 5471 * Setup and add an address space callback to do the work of the delmap call. 5472 * The callback will (and must be) deleted in the actual callback function. 5473 * 5474 * This is done in order to take care of the problem that we have with holding 5475 * the address space's a_lock for a long period of time (e.g. if the NFS server 5476 * is down). Callbacks will be executed in the address space code while the 5477 * a_lock is not held. Holding the address space's a_lock causes things such 5478 * as ps and fork to hang because they are trying to acquire this lock as well. 5479 */ 5480 /* ARGSUSED */ 5481 static int 5482 nfs3_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 5483 size_t len, uint_t prot, uint_t maxprot, uint_t flags, 5484 cred_t *cr, caller_context_t *ct) 5485 { 5486 int caller_found; 5487 int error; 5488 rnode_t *rp; 5489 nfs_delmap_args_t *dmapp; 5490 nfs_delmapcall_t *delmap_call; 5491 5492 if (vp->v_flag & VNOMAP) 5493 return (ENOSYS); 5494 /* 5495 * A process may not change zones if it has NFS pages mmap'ed 5496 * in, so we can't legitimately get here from the wrong zone. 5497 */ 5498 ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 5499 5500 rp = VTOR(vp); 5501 5502 /* 5503 * The way that the address space of this process deletes its mapping 5504 * of this file is via the following call chains: 5505 * - as_free()->SEGOP_UNMAP()/segvn_unmap()->VOP_DELMAP()/nfs3_delmap() 5506 * - as_unmap()->SEGOP_UNMAP()/segvn_unmap()->VOP_DELMAP()/nfs3_delmap() 5507 * 5508 * With the use of address space callbacks we are allowed to drop the 5509 * address space lock, a_lock, while executing the NFS operations that 5510 * need to go over the wire. Returning EAGAIN to the caller of this 5511 * function is what drives the execution of the callback that we add 5512 * below. The callback will be executed by the address space code 5513 * after dropping the a_lock. When the callback is finished, since 5514 * we dropped the a_lock, it must be re-acquired and segvn_unmap() 5515 * is called again on the same segment to finish the rest of the work 5516 * that needs to happen during unmapping. 5517 * 5518 * This action of calling back into the segment driver causes 5519 * nfs3_delmap() to get called again, but since the callback was 5520 * already executed at this point, it already did the work and there 5521 * is nothing left for us to do. 5522 * 5523 * To Summarize: 5524 * - The first time nfs3_delmap is called by the current thread is when 5525 * we add the caller associated with this delmap to the delmap caller 5526 * list, add the callback, and return EAGAIN. 5527 * - The second time in this call chain when nfs3_delmap is called we 5528 * will find this caller in the delmap caller list and realize there 5529 * is no more work to do thus removing this caller from the list and 5530 * returning the error that was set in the callback execution. 5531 */ 5532 caller_found = nfs_find_and_delete_delmapcall(rp, &error); 5533 if (caller_found) { 5534 /* 5535 * 'error' is from the actual delmap operations. To avoid 5536 * hangs, we need to handle the return of EAGAIN differently 5537 * since this is what drives the callback execution. 5538 * In this case, we don't want to return EAGAIN and do the 5539 * callback execution because there are none to execute. 5540 */ 5541 if (error == EAGAIN) 5542 return (0); 5543 else 5544 return (error); 5545 } 5546 5547 /* current caller was not in the list */ 5548 delmap_call = nfs_init_delmapcall(); 5549 5550 mutex_enter(&rp->r_statelock); 5551 list_insert_tail(&rp->r_indelmap, delmap_call); 5552 mutex_exit(&rp->r_statelock); 5553 5554 dmapp = kmem_alloc(sizeof (nfs_delmap_args_t), KM_SLEEP); 5555 5556 dmapp->vp = vp; 5557 dmapp->off = off; 5558 dmapp->addr = addr; 5559 dmapp->len = len; 5560 dmapp->prot = prot; 5561 dmapp->maxprot = maxprot; 5562 dmapp->flags = flags; 5563 dmapp->cr = cr; 5564 dmapp->caller = delmap_call; 5565 5566 error = as_add_callback(as, nfs3_delmap_callback, dmapp, 5567 AS_UNMAP_EVENT, addr, len, KM_SLEEP); 5568 5569 return (error ? error : EAGAIN); 5570 } 5571 5572 /* 5573 * Remove some pages from an mmap'd vnode. Just update the 5574 * count of pages. If doing close-to-open, then flush and 5575 * commit all of the pages associated with this file. 5576 * Otherwise, start an asynchronous page flush to write out 5577 * any dirty pages. This will also associate a credential 5578 * with the rnode which can be used to write the pages. 5579 */ 5580 /* ARGSUSED */ 5581 static void 5582 nfs3_delmap_callback(struct as *as, void *arg, uint_t event) 5583 { 5584 int error; 5585 rnode_t *rp; 5586 mntinfo_t *mi; 5587 nfs_delmap_args_t *dmapp = (nfs_delmap_args_t *)arg; 5588 5589 rp = VTOR(dmapp->vp); 5590 mi = VTOMI(dmapp->vp); 5591 5592 atomic_add_long((ulong_t *)&rp->r_mapcnt, -btopr(dmapp->len)); 5593 ASSERT(rp->r_mapcnt >= 0); 5594 5595 /* 5596 * Initiate a page flush and potential commit if there are 5597 * pages, the file system was not mounted readonly, the segment 5598 * was mapped shared, and the pages themselves were writeable. 5599 */ 5600 if (vn_has_cached_data(dmapp->vp) && !vn_is_readonly(dmapp->vp) && 5601 dmapp->flags == MAP_SHARED && (dmapp->maxprot & PROT_WRITE)) { 5602 mutex_enter(&rp->r_statelock); 5603 rp->r_flags |= RDIRTY; 5604 mutex_exit(&rp->r_statelock); 5605 /* 5606 * If this is a cross-zone access a sync putpage won't work, so 5607 * the best we can do is try an async putpage. That seems 5608 * better than something more draconian such as discarding the 5609 * dirty pages. 5610 */ 5611 if ((mi->mi_flags & MI_NOCTO) || 5612 nfs_zone() != mi->mi_zone) 5613 error = nfs3_putpage(dmapp->vp, dmapp->off, dmapp->len, 5614 B_ASYNC, dmapp->cr, NULL); 5615 else 5616 error = nfs3_putpage_commit(dmapp->vp, dmapp->off, 5617 dmapp->len, dmapp->cr); 5618 if (!error) { 5619 mutex_enter(&rp->r_statelock); 5620 error = rp->r_error; 5621 rp->r_error = 0; 5622 mutex_exit(&rp->r_statelock); 5623 } 5624 } else 5625 error = 0; 5626 5627 if ((rp->r_flags & RDIRECTIO) || (mi->mi_flags & MI_DIRECTIO)) 5628 (void) nfs3_putpage(dmapp->vp, dmapp->off, dmapp->len, 5629 B_INVAL, dmapp->cr, NULL); 5630 5631 dmapp->caller->error = error; 5632 (void) as_delete_callback(as, arg); 5633 kmem_free(dmapp, sizeof (nfs_delmap_args_t)); 5634 } 5635 5636 static int nfs3_pathconf_disable_cache = 0; 5637 5638 #ifdef DEBUG 5639 static int nfs3_pathconf_cache_hits = 0; 5640 static int nfs3_pathconf_cache_misses = 0; 5641 #endif 5642 5643 /* ARGSUSED */ 5644 static int 5645 nfs3_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 5646 caller_context_t *ct) 5647 { 5648 int error; 5649 PATHCONF3args args; 5650 PATHCONF3res res; 5651 int douprintf; 5652 failinfo_t fi; 5653 rnode_t *rp; 5654 hrtime_t t; 5655 5656 if (nfs_zone() != VTOMI(vp)->mi_zone) 5657 return (EIO); 5658 /* 5659 * Large file spec - need to base answer on info stored 5660 * on original FSINFO response. 5661 */ 5662 if (cmd == _PC_FILESIZEBITS) { 5663 unsigned long long ll; 5664 long l = 1; 5665 5666 ll = VTOMI(vp)->mi_maxfilesize; 5667 5668 if (ll == 0) { 5669 *valp = 0; 5670 return (0); 5671 } 5672 5673 if (ll & 0xffffffff00000000) { 5674 l += 32; ll >>= 32; 5675 } 5676 if (ll & 0xffff0000) { 5677 l += 16; ll >>= 16; 5678 } 5679 if (ll & 0xff00) { 5680 l += 8; ll >>= 8; 5681 } 5682 if (ll & 0xf0) { 5683 l += 4; ll >>= 4; 5684 } 5685 if (ll & 0xc) { 5686 l += 2; ll >>= 2; 5687 } 5688 if (ll & 0x2) 5689 l += 2; 5690 else if (ll & 0x1) 5691 l += 1; 5692 *valp = l; 5693 return (0); 5694 } 5695 5696 if (cmd == _PC_ACL_ENABLED) { 5697 *valp = _ACL_ACLENT_ENABLED; 5698 return (0); 5699 } 5700 5701 if (cmd == _PC_XATTR_EXISTS) { 5702 error = 0; 5703 *valp = 0; 5704 if (vp->v_vfsp->vfs_flag & VFS_XATTR) { 5705 vnode_t *avp; 5706 rnode_t *rp; 5707 int error = 0; 5708 mntinfo_t *mi = VTOMI(vp); 5709 5710 if (!(mi->mi_flags & MI_EXTATTR)) 5711 return (0); 5712 5713 rp = VTOR(vp); 5714 if (nfs_rw_enter_sig(&rp->r_rwlock, RW_READER, 5715 INTR(vp))) 5716 return (EINTR); 5717 5718 error = nfs3lookup_dnlc(vp, XATTR_DIR_NAME, &avp, cr); 5719 if (error || avp == NULL) 5720 error = acl_getxattrdir3(vp, &avp, 0, cr, 0); 5721 5722 nfs_rw_exit(&rp->r_rwlock); 5723 5724 if (error == 0 && avp != NULL) { 5725 VN_RELE(avp); 5726 *valp = 1; 5727 } else if (error == ENOENT) 5728 error = 0; 5729 } 5730 return (error); 5731 } 5732 5733 rp = VTOR(vp); 5734 if (rp->r_pathconf != NULL) { 5735 mutex_enter(&rp->r_statelock); 5736 if (rp->r_pathconf != NULL && nfs3_pathconf_disable_cache) { 5737 kmem_free(rp->r_pathconf, sizeof (*rp->r_pathconf)); 5738 rp->r_pathconf = NULL; 5739 } 5740 if (rp->r_pathconf != NULL) { 5741 error = 0; 5742 switch (cmd) { 5743 case _PC_LINK_MAX: 5744 *valp = rp->r_pathconf->link_max; 5745 break; 5746 case _PC_NAME_MAX: 5747 *valp = rp->r_pathconf->name_max; 5748 break; 5749 case _PC_PATH_MAX: 5750 case _PC_SYMLINK_MAX: 5751 *valp = MAXPATHLEN; 5752 break; 5753 case _PC_CHOWN_RESTRICTED: 5754 *valp = rp->r_pathconf->chown_restricted; 5755 break; 5756 case _PC_NO_TRUNC: 5757 *valp = rp->r_pathconf->no_trunc; 5758 break; 5759 default: 5760 error = EINVAL; 5761 break; 5762 } 5763 mutex_exit(&rp->r_statelock); 5764 #ifdef DEBUG 5765 nfs3_pathconf_cache_hits++; 5766 #endif 5767 return (error); 5768 } 5769 mutex_exit(&rp->r_statelock); 5770 } 5771 #ifdef DEBUG 5772 nfs3_pathconf_cache_misses++; 5773 #endif 5774 5775 args.object = *VTOFH3(vp); 5776 fi.vp = vp; 5777 fi.fhp = (caddr_t)&args.object; 5778 fi.copyproc = nfs3copyfh; 5779 fi.lookupproc = nfs3lookup; 5780 fi.xattrdirproc = acl_getxattrdir3; 5781 5782 douprintf = 1; 5783 5784 t = gethrtime(); 5785 5786 error = rfs3call(VTOMI(vp), NFSPROC3_PATHCONF, 5787 xdr_nfs_fh3, (caddr_t)&args, 5788 xdr_PATHCONF3res, (caddr_t)&res, cr, 5789 &douprintf, &res.status, 0, &fi); 5790 5791 if (error) 5792 return (error); 5793 5794 error = geterrno3(res.status); 5795 5796 if (!error) { 5797 nfs3_cache_post_op_attr(vp, &res.resok.obj_attributes, t, cr); 5798 if (!nfs3_pathconf_disable_cache) { 5799 mutex_enter(&rp->r_statelock); 5800 if (rp->r_pathconf == NULL) { 5801 rp->r_pathconf = kmem_alloc( 5802 sizeof (*rp->r_pathconf), KM_NOSLEEP); 5803 if (rp->r_pathconf != NULL) 5804 *rp->r_pathconf = res.resok.info; 5805 } 5806 mutex_exit(&rp->r_statelock); 5807 } 5808 switch (cmd) { 5809 case _PC_LINK_MAX: 5810 *valp = res.resok.info.link_max; 5811 break; 5812 case _PC_NAME_MAX: 5813 *valp = res.resok.info.name_max; 5814 break; 5815 case _PC_PATH_MAX: 5816 case _PC_SYMLINK_MAX: 5817 *valp = MAXPATHLEN; 5818 break; 5819 case _PC_CHOWN_RESTRICTED: 5820 *valp = res.resok.info.chown_restricted; 5821 break; 5822 case _PC_NO_TRUNC: 5823 *valp = res.resok.info.no_trunc; 5824 break; 5825 default: 5826 return (EINVAL); 5827 } 5828 } else { 5829 nfs3_cache_post_op_attr(vp, &res.resfail.obj_attributes, t, cr); 5830 PURGE_STALE_FH(error, vp, cr); 5831 } 5832 5833 return (error); 5834 } 5835 5836 /* 5837 * Called by async thread to do synchronous pageio. Do the i/o, wait 5838 * for it to complete, and cleanup the page list when done. 5839 */ 5840 static int 5841 nfs3_sync_pageio(vnode_t *vp, page_t *pp, u_offset_t io_off, size_t io_len, 5842 int flags, cred_t *cr) 5843 { 5844 int error; 5845 5846 ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 5847 error = nfs3_rdwrlbn(vp, pp, io_off, io_len, flags, cr); 5848 if (flags & B_READ) 5849 pvn_read_done(pp, (error ? B_ERROR : 0) | flags); 5850 else 5851 pvn_write_done(pp, (error ? B_ERROR : 0) | flags); 5852 return (error); 5853 } 5854 5855 /* ARGSUSED */ 5856 static int 5857 nfs3_pageio(vnode_t *vp, page_t *pp, u_offset_t io_off, size_t io_len, 5858 int flags, cred_t *cr, caller_context_t *ct) 5859 { 5860 int error; 5861 rnode_t *rp; 5862 5863 if (pp == NULL) 5864 return (EINVAL); 5865 if (!(flags & B_ASYNC) && nfs_zone() != VTOMI(vp)->mi_zone) 5866 return (EIO); 5867 5868 rp = VTOR(vp); 5869 mutex_enter(&rp->r_statelock); 5870 rp->r_count++; 5871 mutex_exit(&rp->r_statelock); 5872 5873 if (flags & B_ASYNC) { 5874 error = nfs_async_pageio(vp, pp, io_off, io_len, flags, cr, 5875 nfs3_sync_pageio); 5876 } else 5877 error = nfs3_rdwrlbn(vp, pp, io_off, io_len, flags, cr); 5878 mutex_enter(&rp->r_statelock); 5879 rp->r_count--; 5880 cv_broadcast(&rp->r_cv); 5881 mutex_exit(&rp->r_statelock); 5882 return (error); 5883 } 5884 5885 /* ARGSUSED */ 5886 static void 5887 nfs3_dispose(vnode_t *vp, page_t *pp, int fl, int dn, cred_t *cr, 5888 caller_context_t *ct) 5889 { 5890 int error; 5891 rnode_t *rp; 5892 page_t *plist; 5893 page_t *pptr; 5894 offset3 offset; 5895 count3 len; 5896 k_sigset_t smask; 5897 5898 /* 5899 * We should get called with fl equal to either B_FREE or 5900 * B_INVAL. Any other value is illegal. 5901 * 5902 * The page that we are either supposed to free or destroy 5903 * should be exclusive locked and its io lock should not 5904 * be held. 5905 */ 5906 ASSERT(fl == B_FREE || fl == B_INVAL); 5907 ASSERT((PAGE_EXCL(pp) && !page_iolock_assert(pp)) || panicstr); 5908 rp = VTOR(vp); 5909 5910 /* 5911 * If the page doesn't need to be committed or we shouldn't 5912 * even bother attempting to commit it, then just make sure 5913 * that the p_fsdata byte is clear and then either free or 5914 * destroy the page as appropriate. 5915 */ 5916 if (pp->p_fsdata == C_NOCOMMIT || (rp->r_flags & RSTALE)) { 5917 pp->p_fsdata = C_NOCOMMIT; 5918 if (fl == B_FREE) 5919 page_free(pp, dn); 5920 else 5921 page_destroy(pp, dn); 5922 return; 5923 } 5924 5925 /* 5926 * If there is a page invalidation operation going on, then 5927 * if this is one of the pages being destroyed, then just 5928 * clear the p_fsdata byte and then either free or destroy 5929 * the page as appropriate. 5930 */ 5931 mutex_enter(&rp->r_statelock); 5932 if ((rp->r_flags & RTRUNCATE) && pp->p_offset >= rp->r_truncaddr) { 5933 mutex_exit(&rp->r_statelock); 5934 pp->p_fsdata = C_NOCOMMIT; 5935 if (fl == B_FREE) 5936 page_free(pp, dn); 5937 else 5938 page_destroy(pp, dn); 5939 return; 5940 } 5941 5942 /* 5943 * If we are freeing this page and someone else is already 5944 * waiting to do a commit, then just unlock the page and 5945 * return. That other thread will take care of commiting 5946 * this page. The page can be freed sometime after the 5947 * commit has finished. Otherwise, if the page is marked 5948 * as delay commit, then we may be getting called from 5949 * pvn_write_done, one page at a time. This could result 5950 * in one commit per page, so we end up doing lots of small 5951 * commits instead of fewer larger commits. This is bad, 5952 * we want do as few commits as possible. 5953 */ 5954 if (fl == B_FREE) { 5955 if (rp->r_flags & RCOMMITWAIT) { 5956 page_unlock(pp); 5957 mutex_exit(&rp->r_statelock); 5958 return; 5959 } 5960 if (pp->p_fsdata == C_DELAYCOMMIT) { 5961 pp->p_fsdata = C_COMMIT; 5962 page_unlock(pp); 5963 mutex_exit(&rp->r_statelock); 5964 return; 5965 } 5966 } 5967 5968 /* 5969 * Check to see if there is a signal which would prevent an 5970 * attempt to commit the pages from being successful. If so, 5971 * then don't bother with all of the work to gather pages and 5972 * generate the unsuccessful RPC. Just return from here and 5973 * let the page be committed at some later time. 5974 */ 5975 sigintr(&smask, VTOMI(vp)->mi_flags & MI_INT); 5976 if (ttolwp(curthread) != NULL && ISSIG(curthread, JUSTLOOKING)) { 5977 sigunintr(&smask); 5978 page_unlock(pp); 5979 mutex_exit(&rp->r_statelock); 5980 return; 5981 } 5982 sigunintr(&smask); 5983 5984 /* 5985 * We are starting to need to commit pages, so let's try 5986 * to commit as many as possible at once to reduce the 5987 * overhead. 5988 * 5989 * Set the `commit inprogress' state bit. We must 5990 * first wait until any current one finishes. Then 5991 * we initialize the c_pages list with this page. 5992 */ 5993 while (rp->r_flags & RCOMMIT) { 5994 rp->r_flags |= RCOMMITWAIT; 5995 cv_wait(&rp->r_commit.c_cv, &rp->r_statelock); 5996 rp->r_flags &= ~RCOMMITWAIT; 5997 } 5998 rp->r_flags |= RCOMMIT; 5999 mutex_exit(&rp->r_statelock); 6000 ASSERT(rp->r_commit.c_pages == NULL); 6001 rp->r_commit.c_pages = pp; 6002 rp->r_commit.c_commbase = (offset3)pp->p_offset; 6003 rp->r_commit.c_commlen = PAGESIZE; 6004 6005 /* 6006 * Gather together all other pages which can be committed. 6007 * They will all be chained off r_commit.c_pages. 6008 */ 6009 nfs3_get_commit(vp); 6010 6011 /* 6012 * Clear the `commit inprogress' status and disconnect 6013 * the list of pages to be committed from the rnode. 6014 * At this same time, we also save the starting offset 6015 * and length of data to be committed on the server. 6016 */ 6017 plist = rp->r_commit.c_pages; 6018 rp->r_commit.c_pages = NULL; 6019 offset = rp->r_commit.c_commbase; 6020 len = rp->r_commit.c_commlen; 6021 mutex_enter(&rp->r_statelock); 6022 rp->r_flags &= ~RCOMMIT; 6023 cv_broadcast(&rp->r_commit.c_cv); 6024 mutex_exit(&rp->r_statelock); 6025 6026 if (curproc == proc_pageout || curproc == proc_fsflush || 6027 nfs_zone() != VTOMI(vp)->mi_zone) { 6028 nfs_async_commit(vp, plist, offset, len, cr, nfs3_async_commit); 6029 return; 6030 } 6031 6032 /* 6033 * Actually generate the COMMIT3 over the wire operation. 6034 */ 6035 error = nfs3_commit(vp, offset, len, cr); 6036 6037 /* 6038 * If we got an error during the commit, just unlock all 6039 * of the pages. The pages will get retransmitted to the 6040 * server during a putpage operation. 6041 */ 6042 if (error) { 6043 while (plist != NULL) { 6044 pptr = plist; 6045 page_sub(&plist, pptr); 6046 page_unlock(pptr); 6047 } 6048 return; 6049 } 6050 6051 /* 6052 * We've tried as hard as we can to commit the data to stable 6053 * storage on the server. We release the rest of the pages 6054 * and clear the commit required state. They will be put 6055 * onto the tail of the cachelist if they are nolonger 6056 * mapped. 6057 */ 6058 while (plist != pp) { 6059 pptr = plist; 6060 page_sub(&plist, pptr); 6061 pptr->p_fsdata = C_NOCOMMIT; 6062 (void) page_release(pptr, 1); 6063 } 6064 6065 /* 6066 * It is possible that nfs3_commit didn't return error but 6067 * some other thread has modified the page we are going 6068 * to free/destroy. 6069 * In this case we need to rewrite the page. Do an explicit check 6070 * before attempting to free/destroy the page. If modified, needs to 6071 * be rewritten so unlock the page and return. 6072 */ 6073 if (hat_ismod(pp)) { 6074 pp->p_fsdata = C_NOCOMMIT; 6075 page_unlock(pp); 6076 return; 6077 } 6078 6079 /* 6080 * Now, as appropriate, either free or destroy the page 6081 * that we were called with. 6082 */ 6083 pp->p_fsdata = C_NOCOMMIT; 6084 if (fl == B_FREE) 6085 page_free(pp, dn); 6086 else 6087 page_destroy(pp, dn); 6088 } 6089 6090 static int 6091 nfs3_commit(vnode_t *vp, offset3 offset, count3 count, cred_t *cr) 6092 { 6093 int error; 6094 rnode_t *rp; 6095 COMMIT3args args; 6096 COMMIT3res res; 6097 int douprintf; 6098 cred_t *cred; 6099 6100 rp = VTOR(vp); 6101 ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 6102 6103 mutex_enter(&rp->r_statelock); 6104 if (rp->r_cred != NULL) { 6105 cred = rp->r_cred; 6106 crhold(cred); 6107 } else { 6108 rp->r_cred = cr; 6109 crhold(cr); 6110 cred = cr; 6111 crhold(cred); 6112 } 6113 mutex_exit(&rp->r_statelock); 6114 6115 args.file = *VTOFH3(vp); 6116 args.offset = offset; 6117 args.count = count; 6118 6119 doitagain: 6120 douprintf = 1; 6121 error = rfs3call(VTOMI(vp), NFSPROC3_COMMIT, 6122 xdr_COMMIT3args, (caddr_t)&args, 6123 xdr_COMMIT3res, (caddr_t)&res, cred, 6124 &douprintf, &res.status, 0, NULL); 6125 6126 crfree(cred); 6127 6128 if (error) 6129 return (error); 6130 6131 error = geterrno3(res.status); 6132 if (!error) { 6133 ASSERT(rp->r_flags & RHAVEVERF); 6134 mutex_enter(&rp->r_statelock); 6135 if (rp->r_verf == res.resok.verf) { 6136 mutex_exit(&rp->r_statelock); 6137 return (0); 6138 } 6139 nfs3_set_mod(vp); 6140 rp->r_verf = res.resok.verf; 6141 mutex_exit(&rp->r_statelock); 6142 error = NFS_VERF_MISMATCH; 6143 } else { 6144 if (error == EACCES) { 6145 mutex_enter(&rp->r_statelock); 6146 if (cred != cr) { 6147 if (rp->r_cred != NULL) 6148 crfree(rp->r_cred); 6149 rp->r_cred = cr; 6150 crhold(cr); 6151 cred = cr; 6152 crhold(cred); 6153 mutex_exit(&rp->r_statelock); 6154 goto doitagain; 6155 } 6156 mutex_exit(&rp->r_statelock); 6157 } 6158 /* 6159 * Can't do a PURGE_STALE_FH here because this 6160 * can cause a deadlock. nfs3_commit can 6161 * be called from nfs3_dispose which can be called 6162 * indirectly via pvn_vplist_dirty. PURGE_STALE_FH 6163 * can call back to pvn_vplist_dirty. 6164 */ 6165 if (error == ESTALE) { 6166 mutex_enter(&rp->r_statelock); 6167 rp->r_flags |= RSTALE; 6168 if (!rp->r_error) 6169 rp->r_error = error; 6170 mutex_exit(&rp->r_statelock); 6171 PURGE_ATTRCACHE(vp); 6172 } else { 6173 mutex_enter(&rp->r_statelock); 6174 if (!rp->r_error) 6175 rp->r_error = error; 6176 mutex_exit(&rp->r_statelock); 6177 } 6178 } 6179 6180 return (error); 6181 } 6182 6183 static void 6184 nfs3_set_mod(vnode_t *vp) 6185 { 6186 page_t *pp; 6187 kmutex_t *vphm; 6188 6189 ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 6190 vphm = page_vnode_mutex(vp); 6191 mutex_enter(vphm); 6192 if ((pp = vp->v_pages) != NULL) { 6193 do { 6194 if (pp->p_fsdata != C_NOCOMMIT) { 6195 hat_setmod(pp); 6196 pp->p_fsdata = C_NOCOMMIT; 6197 } 6198 } while ((pp = pp->p_vpnext) != vp->v_pages); 6199 } 6200 mutex_exit(vphm); 6201 } 6202 6203 6204 /* 6205 * This routine is used to gather together a page list of the pages 6206 * which are to be committed on the server. This routine must not 6207 * be called if the calling thread holds any locked pages. 6208 * 6209 * The calling thread must have set RCOMMIT. This bit is used to 6210 * serialize access to the commit structure in the rnode. As long 6211 * as the thread has set RCOMMIT, then it can manipulate the commit 6212 * structure without requiring any other locks. 6213 */ 6214 static void 6215 nfs3_get_commit(vnode_t *vp) 6216 { 6217 rnode_t *rp; 6218 page_t *pp; 6219 kmutex_t *vphm; 6220 6221 rp = VTOR(vp); 6222 6223 ASSERT(rp->r_flags & RCOMMIT); 6224 6225 vphm = page_vnode_mutex(vp); 6226 mutex_enter(vphm); 6227 6228 /* 6229 * If there are no pages associated with this vnode, then 6230 * just return. 6231 */ 6232 if ((pp = vp->v_pages) == NULL) { 6233 mutex_exit(vphm); 6234 return; 6235 } 6236 6237 /* 6238 * Step through all of the pages associated with this vnode 6239 * looking for pages which need to be committed. 6240 */ 6241 do { 6242 /* 6243 * If this page does not need to be committed or is 6244 * modified, then just skip it. 6245 */ 6246 if (pp->p_fsdata == C_NOCOMMIT || hat_ismod(pp)) 6247 continue; 6248 6249 /* 6250 * Attempt to lock the page. If we can't, then 6251 * someone else is messing with it and we will 6252 * just skip it. 6253 */ 6254 if (!page_trylock(pp, SE_EXCL)) 6255 continue; 6256 6257 /* 6258 * If this page does not need to be committed or is 6259 * modified, then just skip it. Recheck now that 6260 * the page is locked. 6261 */ 6262 if (pp->p_fsdata == C_NOCOMMIT || hat_ismod(pp)) { 6263 page_unlock(pp); 6264 continue; 6265 } 6266 6267 if (PP_ISFREE(pp)) { 6268 cmn_err(CE_PANIC, "nfs3_get_commit: %p is free", 6269 (void *)pp); 6270 } 6271 6272 /* 6273 * The page needs to be committed and we locked it. 6274 * Update the base and length parameters and add it 6275 * to r_pages. 6276 */ 6277 if (rp->r_commit.c_pages == NULL) { 6278 rp->r_commit.c_commbase = (offset3)pp->p_offset; 6279 rp->r_commit.c_commlen = PAGESIZE; 6280 } else if (pp->p_offset < rp->r_commit.c_commbase) { 6281 rp->r_commit.c_commlen = rp->r_commit.c_commbase - 6282 (offset3)pp->p_offset + rp->r_commit.c_commlen; 6283 rp->r_commit.c_commbase = (offset3)pp->p_offset; 6284 } else if ((rp->r_commit.c_commbase + rp->r_commit.c_commlen) 6285 <= pp->p_offset) { 6286 rp->r_commit.c_commlen = (offset3)pp->p_offset - 6287 rp->r_commit.c_commbase + PAGESIZE; 6288 } 6289 page_add(&rp->r_commit.c_pages, pp); 6290 } while ((pp = pp->p_vpnext) != vp->v_pages); 6291 6292 mutex_exit(vphm); 6293 } 6294 6295 /* 6296 * This routine is used to gather together a page list of the pages 6297 * which are to be committed on the server. This routine must not 6298 * be called if the calling thread holds any locked pages. 6299 * 6300 * The calling thread must have set RCOMMIT. This bit is used to 6301 * serialize access to the commit structure in the rnode. As long 6302 * as the thread has set RCOMMIT, then it can manipulate the commit 6303 * structure without requiring any other locks. 6304 */ 6305 static void 6306 nfs3_get_commit_range(vnode_t *vp, u_offset_t soff, size_t len) 6307 { 6308 6309 rnode_t *rp; 6310 page_t *pp; 6311 u_offset_t end; 6312 u_offset_t off; 6313 6314 ASSERT(len != 0); 6315 6316 rp = VTOR(vp); 6317 6318 ASSERT(rp->r_flags & RCOMMIT); 6319 ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 6320 6321 /* 6322 * If there are no pages associated with this vnode, then 6323 * just return. 6324 */ 6325 if ((pp = vp->v_pages) == NULL) 6326 return; 6327 6328 /* 6329 * Calculate the ending offset. 6330 */ 6331 end = soff + len; 6332 6333 for (off = soff; off < end; off += PAGESIZE) { 6334 /* 6335 * Lookup each page by vp, offset. 6336 */ 6337 if ((pp = page_lookup_nowait(vp, off, SE_EXCL)) == NULL) 6338 continue; 6339 6340 /* 6341 * If this page does not need to be committed or is 6342 * modified, then just skip it. 6343 */ 6344 if (pp->p_fsdata == C_NOCOMMIT || hat_ismod(pp)) { 6345 page_unlock(pp); 6346 continue; 6347 } 6348 6349 ASSERT(PP_ISFREE(pp) == 0); 6350 6351 /* 6352 * The page needs to be committed and we locked it. 6353 * Update the base and length parameters and add it 6354 * to r_pages. 6355 */ 6356 if (rp->r_commit.c_pages == NULL) { 6357 rp->r_commit.c_commbase = (offset3)pp->p_offset; 6358 rp->r_commit.c_commlen = PAGESIZE; 6359 } else { 6360 rp->r_commit.c_commlen = (offset3)pp->p_offset - 6361 rp->r_commit.c_commbase + PAGESIZE; 6362 } 6363 page_add(&rp->r_commit.c_pages, pp); 6364 } 6365 } 6366 6367 #if 0 /* unused */ 6368 #ifdef DEBUG 6369 static int 6370 nfs3_no_uncommitted_pages(vnode_t *vp) 6371 { 6372 page_t *pp; 6373 kmutex_t *vphm; 6374 6375 vphm = page_vnode_mutex(vp); 6376 mutex_enter(vphm); 6377 if ((pp = vp->v_pages) != NULL) { 6378 do { 6379 if (pp->p_fsdata != C_NOCOMMIT) { 6380 mutex_exit(vphm); 6381 return (0); 6382 } 6383 } while ((pp = pp->p_vpnext) != vp->v_pages); 6384 } 6385 mutex_exit(vphm); 6386 6387 return (1); 6388 } 6389 #endif 6390 #endif 6391 6392 static int 6393 nfs3_putpage_commit(vnode_t *vp, offset_t poff, size_t plen, cred_t *cr) 6394 { 6395 int error; 6396 writeverf3 write_verf; 6397 rnode_t *rp = VTOR(vp); 6398 6399 ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 6400 /* 6401 * Flush the data portion of the file and then commit any 6402 * portions which need to be committed. This may need to 6403 * be done twice if the server has changed state since 6404 * data was last written. The data will need to be 6405 * rewritten to the server and then a new commit done. 6406 * 6407 * In fact, this may need to be done several times if the 6408 * server is having problems and crashing while we are 6409 * attempting to do this. 6410 */ 6411 6412 top: 6413 /* 6414 * Do a flush based on the poff and plen arguments. This 6415 * will asynchronously write out any modified pages in the 6416 * range specified by (poff, plen). This starts all of the 6417 * i/o operations which will be waited for in the next 6418 * call to nfs3_putpage 6419 */ 6420 6421 mutex_enter(&rp->r_statelock); 6422 write_verf = rp->r_verf; 6423 mutex_exit(&rp->r_statelock); 6424 6425 error = nfs3_putpage(vp, poff, plen, B_ASYNC, cr, NULL); 6426 if (error == EAGAIN) 6427 error = 0; 6428 6429 /* 6430 * Do a flush based on the poff and plen arguments. This 6431 * will synchronously write out any modified pages in the 6432 * range specified by (poff, plen) and wait until all of 6433 * the asynchronous i/o's in that range are done as well. 6434 */ 6435 if (!error) 6436 error = nfs3_putpage(vp, poff, plen, 0, cr, NULL); 6437 6438 if (error) 6439 return (error); 6440 6441 mutex_enter(&rp->r_statelock); 6442 if (rp->r_verf != write_verf) { 6443 mutex_exit(&rp->r_statelock); 6444 goto top; 6445 } 6446 mutex_exit(&rp->r_statelock); 6447 6448 /* 6449 * Now commit any pages which might need to be committed. 6450 * If the error, NFS_VERF_MISMATCH, is returned, then 6451 * start over with the flush operation. 6452 */ 6453 6454 error = nfs3_commit_vp(vp, poff, plen, cr); 6455 6456 if (error == NFS_VERF_MISMATCH) 6457 goto top; 6458 6459 return (error); 6460 } 6461 6462 static int 6463 nfs3_commit_vp(vnode_t *vp, u_offset_t poff, size_t plen, cred_t *cr) 6464 { 6465 rnode_t *rp; 6466 page_t *plist; 6467 offset3 offset; 6468 count3 len; 6469 6470 6471 rp = VTOR(vp); 6472 6473 if (nfs_zone() != VTOMI(vp)->mi_zone) 6474 return (EIO); 6475 /* 6476 * Set the `commit inprogress' state bit. We must 6477 * first wait until any current one finishes. 6478 */ 6479 mutex_enter(&rp->r_statelock); 6480 while (rp->r_flags & RCOMMIT) { 6481 rp->r_flags |= RCOMMITWAIT; 6482 cv_wait(&rp->r_commit.c_cv, &rp->r_statelock); 6483 rp->r_flags &= ~RCOMMITWAIT; 6484 } 6485 rp->r_flags |= RCOMMIT; 6486 mutex_exit(&rp->r_statelock); 6487 6488 /* 6489 * Gather together all of the pages which need to be 6490 * committed. 6491 */ 6492 if (plen == 0) 6493 nfs3_get_commit(vp); 6494 else 6495 nfs3_get_commit_range(vp, poff, plen); 6496 6497 /* 6498 * Clear the `commit inprogress' bit and disconnect the 6499 * page list which was gathered together in nfs3_get_commit. 6500 */ 6501 plist = rp->r_commit.c_pages; 6502 rp->r_commit.c_pages = NULL; 6503 offset = rp->r_commit.c_commbase; 6504 len = rp->r_commit.c_commlen; 6505 mutex_enter(&rp->r_statelock); 6506 rp->r_flags &= ~RCOMMIT; 6507 cv_broadcast(&rp->r_commit.c_cv); 6508 mutex_exit(&rp->r_statelock); 6509 6510 /* 6511 * If any pages need to be committed, commit them and 6512 * then unlock them so that they can be freed some 6513 * time later. 6514 */ 6515 if (plist != NULL) { 6516 /* 6517 * No error occurred during the flush portion 6518 * of this operation, so now attempt to commit 6519 * the data to stable storage on the server. 6520 * 6521 * This will unlock all of the pages on the list. 6522 */ 6523 return (nfs3_sync_commit(vp, plist, offset, len, cr)); 6524 } 6525 return (0); 6526 } 6527 6528 static int 6529 nfs3_sync_commit(vnode_t *vp, page_t *plist, offset3 offset, count3 count, 6530 cred_t *cr) 6531 { 6532 int error; 6533 page_t *pp; 6534 6535 ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 6536 error = nfs3_commit(vp, offset, count, cr); 6537 6538 /* 6539 * If we got an error, then just unlock all of the pages 6540 * on the list. 6541 */ 6542 if (error) { 6543 while (plist != NULL) { 6544 pp = plist; 6545 page_sub(&plist, pp); 6546 page_unlock(pp); 6547 } 6548 return (error); 6549 } 6550 /* 6551 * We've tried as hard as we can to commit the data to stable 6552 * storage on the server. We just unlock the pages and clear 6553 * the commit required state. They will get freed later. 6554 */ 6555 while (plist != NULL) { 6556 pp = plist; 6557 page_sub(&plist, pp); 6558 pp->p_fsdata = C_NOCOMMIT; 6559 page_unlock(pp); 6560 } 6561 6562 return (error); 6563 } 6564 6565 static void 6566 nfs3_async_commit(vnode_t *vp, page_t *plist, offset3 offset, count3 count, 6567 cred_t *cr) 6568 { 6569 ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 6570 (void) nfs3_sync_commit(vp, plist, offset, count, cr); 6571 } 6572 6573 /* ARGSUSED */ 6574 static int 6575 nfs3_setsecattr(vnode_t *vp, vsecattr_t *vsecattr, int flag, cred_t *cr, 6576 caller_context_t *ct) 6577 { 6578 int error; 6579 mntinfo_t *mi; 6580 6581 mi = VTOMI(vp); 6582 6583 if (nfs_zone() != mi->mi_zone) 6584 return (EIO); 6585 6586 if (mi->mi_flags & MI_ACL) { 6587 error = acl_setacl3(vp, vsecattr, flag, cr); 6588 if (mi->mi_flags & MI_ACL) 6589 return (error); 6590 } 6591 6592 return (ENOSYS); 6593 } 6594 6595 /* ARGSUSED */ 6596 static int 6597 nfs3_getsecattr(vnode_t *vp, vsecattr_t *vsecattr, int flag, cred_t *cr, 6598 caller_context_t *ct) 6599 { 6600 int error; 6601 mntinfo_t *mi; 6602 6603 mi = VTOMI(vp); 6604 6605 if (nfs_zone() != mi->mi_zone) 6606 return (EIO); 6607 6608 if (mi->mi_flags & MI_ACL) { 6609 error = acl_getacl3(vp, vsecattr, flag, cr); 6610 if (mi->mi_flags & MI_ACL) 6611 return (error); 6612 } 6613 6614 return (fs_fab_acl(vp, vsecattr, flag, cr, ct)); 6615 } 6616 6617 /* ARGSUSED */ 6618 static int 6619 nfs3_shrlock(vnode_t *vp, int cmd, struct shrlock *shr, int flag, cred_t *cr, 6620 caller_context_t *ct) 6621 { 6622 int error; 6623 struct shrlock nshr; 6624 struct nfs_owner nfs_owner; 6625 netobj lm_fh3; 6626 6627 if (nfs_zone() != VTOMI(vp)->mi_zone) 6628 return (EIO); 6629 6630 /* 6631 * check for valid cmd parameter 6632 */ 6633 if (cmd != F_SHARE && cmd != F_UNSHARE && cmd != F_HASREMOTELOCKS) 6634 return (EINVAL); 6635 6636 /* 6637 * Check access permissions 6638 */ 6639 if (cmd == F_SHARE && 6640 (((shr->s_access & F_RDACC) && !(flag & FREAD)) || 6641 ((shr->s_access & F_WRACC) && !(flag & FWRITE)))) 6642 return (EBADF); 6643 6644 /* 6645 * If the filesystem is mounted using local locking, pass the 6646 * request off to the local share code. 6647 */ 6648 if (VTOMI(vp)->mi_flags & MI_LLOCK) 6649 return (fs_shrlock(vp, cmd, shr, flag, cr, ct)); 6650 6651 switch (cmd) { 6652 case F_SHARE: 6653 case F_UNSHARE: 6654 lm_fh3.n_len = VTOFH3(vp)->fh3_length; 6655 lm_fh3.n_bytes = (char *)&(VTOFH3(vp)->fh3_u.data); 6656 6657 /* 6658 * If passed an owner that is too large to fit in an 6659 * nfs_owner it is likely a recursive call from the 6660 * lock manager client and pass it straight through. If 6661 * it is not a nfs_owner then simply return an error. 6662 */ 6663 if (shr->s_own_len > sizeof (nfs_owner.lowner)) { 6664 if (((struct nfs_owner *)shr->s_owner)->magic != 6665 NFS_OWNER_MAGIC) 6666 return (EINVAL); 6667 6668 if (error = lm4_shrlock(vp, cmd, shr, flag, &lm_fh3)) { 6669 error = set_errno(error); 6670 } 6671 return (error); 6672 } 6673 /* 6674 * Remote share reservations owner is a combination of 6675 * a magic number, hostname, and the local owner 6676 */ 6677 bzero(&nfs_owner, sizeof (nfs_owner)); 6678 nfs_owner.magic = NFS_OWNER_MAGIC; 6679 (void) strncpy(nfs_owner.hname, uts_nodename(), 6680 sizeof (nfs_owner.hname)); 6681 bcopy(shr->s_owner, nfs_owner.lowner, shr->s_own_len); 6682 nshr.s_access = shr->s_access; 6683 nshr.s_deny = shr->s_deny; 6684 nshr.s_sysid = 0; 6685 nshr.s_pid = ttoproc(curthread)->p_pid; 6686 nshr.s_own_len = sizeof (nfs_owner); 6687 nshr.s_owner = (caddr_t)&nfs_owner; 6688 6689 if (error = lm4_shrlock(vp, cmd, &nshr, flag, &lm_fh3)) { 6690 error = set_errno(error); 6691 } 6692 6693 break; 6694 6695 case F_HASREMOTELOCKS: 6696 /* 6697 * NFS client can't store remote locks itself 6698 */ 6699 shr->s_access = 0; 6700 error = 0; 6701 break; 6702 6703 default: 6704 error = EINVAL; 6705 break; 6706 } 6707 6708 return (error); 6709 } 6710