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 2009 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 #include <sys/param.h> 32 #include <sys/types.h> 33 #include <sys/systm.h> 34 #include <sys/cred.h> 35 #include <sys/time.h> 36 #include <sys/vnode.h> 37 #include <sys/vfs.h> 38 #include <sys/vfs_opreg.h> 39 #include <sys/file.h> 40 #include <sys/filio.h> 41 #include <sys/uio.h> 42 #include <sys/buf.h> 43 #include <sys/mman.h> 44 #include <sys/pathname.h> 45 #include <sys/dirent.h> 46 #include <sys/debug.h> 47 #include <sys/vmsystm.h> 48 #include <sys/fcntl.h> 49 #include <sys/flock.h> 50 #include <sys/swap.h> 51 #include <sys/errno.h> 52 #include <sys/strsubr.h> 53 #include <sys/sysmacros.h> 54 #include <sys/kmem.h> 55 #include <sys/cmn_err.h> 56 #include <sys/pathconf.h> 57 #include <sys/utsname.h> 58 #include <sys/dnlc.h> 59 #include <sys/acl.h> 60 #include <sys/systeminfo.h> 61 #include <sys/atomic.h> 62 #include <sys/policy.h> 63 #include <sys/sdt.h> 64 #include <sys/zone.h> 65 66 #include <rpc/types.h> 67 #include <rpc/auth.h> 68 #include <rpc/clnt.h> 69 #include <rpc/rpc_rdma.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 res.wlist = NULL; 480 481 offset = uiop->uio_loffset; 482 count = uiop->uio_resid; 483 484 do { 485 if (mi->mi_io_kstats) { 486 mutex_enter(&mi->mi_lock); 487 kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats)); 488 mutex_exit(&mi->mi_lock); 489 } 490 491 do { 492 tsize = MIN(mi->mi_tsize, count); 493 args.offset = (offset3)offset; 494 args.count = (count3)tsize; 495 res.size = (uint_t)tsize; 496 args.res_uiop = uiop; 497 args.res_data_val_alt = NULL; 498 499 error = rfs3call(mi, NFSPROC3_READ, 500 xdr_READ3args, (caddr_t)&args, 501 xdr_READ3uiores, (caddr_t)&res, cr, 502 &douprintf, &res.status, 0, &fi); 503 } while (error == ENFS_TRYAGAIN); 504 505 if (mi->mi_io_kstats) { 506 mutex_enter(&mi->mi_lock); 507 kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats)); 508 mutex_exit(&mi->mi_lock); 509 } 510 511 if (error) 512 return (error); 513 514 error = geterrno3(res.status); 515 if (error) 516 return (error); 517 518 if (res.count != res.size) { 519 zcmn_err(getzoneid(), CE_WARN, 520 "nfs3_directio_read: server %s returned incorrect amount", 521 sv_hostname); 522 return (EIO); 523 } 524 count -= res.count; 525 offset += res.count; 526 if (mi->mi_io_kstats) { 527 mutex_enter(&mi->mi_lock); 528 KSTAT_IO_PTR(mi->mi_io_kstats)->reads++; 529 KSTAT_IO_PTR(mi->mi_io_kstats)->nread += res.count; 530 mutex_exit(&mi->mi_lock); 531 } 532 lwp_stat_update(LWP_STAT_INBLK, 1); 533 } while (count && !res.eof); 534 535 return (0); 536 } 537 538 /* ARGSUSED */ 539 static int 540 nfs3_read(vnode_t *vp, struct uio *uiop, int ioflag, cred_t *cr, 541 caller_context_t *ct) 542 { 543 rnode_t *rp; 544 u_offset_t off; 545 offset_t diff; 546 int on; 547 size_t n; 548 caddr_t base; 549 uint_t flags; 550 int error = 0; 551 mntinfo_t *mi; 552 553 rp = VTOR(vp); 554 mi = VTOMI(vp); 555 556 ASSERT(nfs_rw_lock_held(&rp->r_rwlock, RW_READER)); 557 558 if (nfs_zone() != mi->mi_zone) 559 return (EIO); 560 561 if (vp->v_type != VREG) 562 return (EISDIR); 563 564 if (uiop->uio_resid == 0) 565 return (0); 566 567 if (uiop->uio_loffset < 0 || uiop->uio_loffset + uiop->uio_resid < 0) 568 return (EINVAL); 569 570 /* 571 * Bypass VM if caching has been disabled (e.g., locking) or if 572 * using client-side direct I/O and the file is not mmap'd and 573 * there are no cached pages. 574 */ 575 if ((vp->v_flag & VNOCACHE) || 576 (((rp->r_flags & RDIRECTIO) || (mi->mi_flags & MI_DIRECTIO)) && 577 rp->r_mapcnt == 0 && rp->r_inmap == 0 && 578 !vn_has_cached_data(vp))) { 579 return (nfs3_directio_read(vp, uiop, cr)); 580 } 581 582 do { 583 off = uiop->uio_loffset & MAXBMASK; /* mapping offset */ 584 on = uiop->uio_loffset & MAXBOFFSET; /* Relative offset */ 585 n = MIN(MAXBSIZE - on, uiop->uio_resid); 586 587 error = nfs3_validate_caches(vp, cr); 588 if (error) 589 break; 590 591 mutex_enter(&rp->r_statelock); 592 while (rp->r_flags & RINCACHEPURGE) { 593 if (!cv_wait_sig(&rp->r_cv, &rp->r_statelock)) { 594 mutex_exit(&rp->r_statelock); 595 return (EINTR); 596 } 597 } 598 diff = rp->r_size - uiop->uio_loffset; 599 mutex_exit(&rp->r_statelock); 600 if (diff <= 0) 601 break; 602 if (diff < n) 603 n = (size_t)diff; 604 605 if (vpm_enable) { 606 /* 607 * Copy data. 608 */ 609 error = vpm_data_copy(vp, off + on, n, uiop, 610 1, NULL, 0, S_READ); 611 } else { 612 base = segmap_getmapflt(segkmap, vp, off + on, n, 1, 613 S_READ); 614 615 error = uiomove(base + on, n, UIO_READ, uiop); 616 } 617 618 if (!error) { 619 /* 620 * If read a whole block or read to eof, 621 * won't need this buffer again soon. 622 */ 623 mutex_enter(&rp->r_statelock); 624 if (n + on == MAXBSIZE || 625 uiop->uio_loffset == rp->r_size) 626 flags = SM_DONTNEED; 627 else 628 flags = 0; 629 mutex_exit(&rp->r_statelock); 630 if (vpm_enable) { 631 error = vpm_sync_pages(vp, off, n, flags); 632 } else { 633 error = segmap_release(segkmap, base, flags); 634 } 635 } else { 636 if (vpm_enable) { 637 (void) vpm_sync_pages(vp, off, n, 0); 638 } else { 639 (void) segmap_release(segkmap, base, 0); 640 } 641 } 642 } while (!error && uiop->uio_resid > 0); 643 644 return (error); 645 } 646 647 /* ARGSUSED */ 648 static int 649 nfs3_write(vnode_t *vp, struct uio *uiop, int ioflag, cred_t *cr, 650 caller_context_t *ct) 651 { 652 rlim64_t limit = uiop->uio_llimit; 653 rnode_t *rp; 654 u_offset_t off; 655 caddr_t base; 656 uint_t flags; 657 int remainder; 658 size_t n; 659 int on; 660 int error; 661 int resid; 662 offset_t offset; 663 mntinfo_t *mi; 664 uint_t bsize; 665 666 rp = VTOR(vp); 667 668 if (vp->v_type != VREG) 669 return (EISDIR); 670 671 mi = VTOMI(vp); 672 if (nfs_zone() != mi->mi_zone) 673 return (EIO); 674 if (uiop->uio_resid == 0) 675 return (0); 676 677 if (ioflag & FAPPEND) { 678 struct vattr va; 679 680 /* 681 * Must serialize if appending. 682 */ 683 if (nfs_rw_lock_held(&rp->r_rwlock, RW_READER)) { 684 nfs_rw_exit(&rp->r_rwlock); 685 if (nfs_rw_enter_sig(&rp->r_rwlock, RW_WRITER, 686 INTR(vp))) 687 return (EINTR); 688 } 689 690 va.va_mask = AT_SIZE; 691 error = nfs3getattr(vp, &va, cr); 692 if (error) 693 return (error); 694 uiop->uio_loffset = va.va_size; 695 } 696 697 offset = uiop->uio_loffset + uiop->uio_resid; 698 699 if (uiop->uio_loffset < 0 || offset < 0) 700 return (EINVAL); 701 702 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 703 limit = MAXOFFSET_T; 704 705 /* 706 * Check to make sure that the process will not exceed 707 * its limit on file size. It is okay to write up to 708 * the limit, but not beyond. Thus, the write which 709 * reaches the limit will be short and the next write 710 * will return an error. 711 */ 712 remainder = 0; 713 if (offset > limit) { 714 remainder = offset - limit; 715 uiop->uio_resid = limit - uiop->uio_loffset; 716 if (uiop->uio_resid <= 0) { 717 proc_t *p = ttoproc(curthread); 718 719 uiop->uio_resid += remainder; 720 mutex_enter(&p->p_lock); 721 (void) rctl_action(rctlproc_legacy[RLIMIT_FSIZE], 722 p->p_rctls, p, RCA_UNSAFE_SIGINFO); 723 mutex_exit(&p->p_lock); 724 return (EFBIG); 725 } 726 } 727 728 if (nfs_rw_enter_sig(&rp->r_lkserlock, RW_READER, INTR(vp))) 729 return (EINTR); 730 731 /* 732 * Bypass VM if caching has been disabled (e.g., locking) or if 733 * using client-side direct I/O and the file is not mmap'd and 734 * there are no cached pages. 735 */ 736 if ((vp->v_flag & VNOCACHE) || 737 (((rp->r_flags & RDIRECTIO) || (mi->mi_flags & MI_DIRECTIO)) && 738 rp->r_mapcnt == 0 && rp->r_inmap == 0 && 739 !vn_has_cached_data(vp))) { 740 size_t bufsize; 741 int count; 742 u_offset_t org_offset; 743 stable_how stab_comm; 744 745 nfs3_fwrite: 746 if (rp->r_flags & RSTALE) { 747 resid = uiop->uio_resid; 748 offset = uiop->uio_loffset; 749 error = rp->r_error; 750 /* 751 * A close may have cleared r_error, if so, 752 * propagate ESTALE error return properly 753 */ 754 if (error == 0) 755 error = ESTALE; 756 goto bottom; 757 } 758 bufsize = MIN(uiop->uio_resid, mi->mi_stsize); 759 base = kmem_alloc(bufsize, KM_SLEEP); 760 do { 761 if (ioflag & FDSYNC) 762 stab_comm = DATA_SYNC; 763 else 764 stab_comm = FILE_SYNC; 765 resid = uiop->uio_resid; 766 offset = uiop->uio_loffset; 767 count = MIN(uiop->uio_resid, bufsize); 768 org_offset = uiop->uio_loffset; 769 error = uiomove(base, count, UIO_WRITE, uiop); 770 if (!error) { 771 error = nfs3write(vp, base, org_offset, 772 count, cr, &stab_comm); 773 } 774 } while (!error && uiop->uio_resid > 0); 775 kmem_free(base, bufsize); 776 goto bottom; 777 } 778 779 780 bsize = vp->v_vfsp->vfs_bsize; 781 782 do { 783 off = uiop->uio_loffset & MAXBMASK; /* mapping offset */ 784 on = uiop->uio_loffset & MAXBOFFSET; /* Relative offset */ 785 n = MIN(MAXBSIZE - on, uiop->uio_resid); 786 787 resid = uiop->uio_resid; 788 offset = uiop->uio_loffset; 789 790 if (rp->r_flags & RSTALE) { 791 error = rp->r_error; 792 /* 793 * A close may have cleared r_error, if so, 794 * propagate ESTALE error return properly 795 */ 796 if (error == 0) 797 error = ESTALE; 798 break; 799 } 800 801 /* 802 * Don't create dirty pages faster than they 803 * can be cleaned so that the system doesn't 804 * get imbalanced. If the async queue is 805 * maxed out, then wait for it to drain before 806 * creating more dirty pages. Also, wait for 807 * any threads doing pagewalks in the vop_getattr 808 * entry points so that they don't block for 809 * long periods. 810 */ 811 mutex_enter(&rp->r_statelock); 812 while ((mi->mi_max_threads != 0 && 813 rp->r_awcount > 2 * mi->mi_max_threads) || 814 rp->r_gcount > 0) { 815 if (INTR(vp)) { 816 klwp_t *lwp = ttolwp(curthread); 817 818 if (lwp != NULL) 819 lwp->lwp_nostop++; 820 if (!cv_wait_sig(&rp->r_cv, &rp->r_statelock)) { 821 mutex_exit(&rp->r_statelock); 822 if (lwp != NULL) 823 lwp->lwp_nostop--; 824 error = EINTR; 825 goto bottom; 826 } 827 if (lwp != NULL) 828 lwp->lwp_nostop--; 829 } else 830 cv_wait(&rp->r_cv, &rp->r_statelock); 831 } 832 mutex_exit(&rp->r_statelock); 833 834 /* 835 * Touch the page and fault it in if it is not in core 836 * before segmap_getmapflt or vpm_data_copy can lock it. 837 * This is to avoid the deadlock if the buffer is mapped 838 * to the same file through mmap which we want to write. 839 */ 840 uio_prefaultpages((long)n, uiop); 841 842 if (vpm_enable) { 843 /* 844 * It will use kpm mappings, so no need to 845 * pass an address. 846 */ 847 error = writerp(rp, NULL, n, uiop, 0); 848 } else { 849 if (segmap_kpm) { 850 int pon = uiop->uio_loffset & PAGEOFFSET; 851 size_t pn = MIN(PAGESIZE - pon, 852 uiop->uio_resid); 853 int pagecreate; 854 855 mutex_enter(&rp->r_statelock); 856 pagecreate = (pon == 0) && (pn == PAGESIZE || 857 uiop->uio_loffset + pn >= rp->r_size); 858 mutex_exit(&rp->r_statelock); 859 860 base = segmap_getmapflt(segkmap, vp, off + on, 861 pn, !pagecreate, S_WRITE); 862 863 error = writerp(rp, base + pon, n, uiop, 864 pagecreate); 865 866 } else { 867 base = segmap_getmapflt(segkmap, vp, off + on, 868 n, 0, S_READ); 869 error = writerp(rp, base + on, n, uiop, 0); 870 } 871 } 872 873 if (!error) { 874 if (mi->mi_flags & MI_NOAC) 875 flags = SM_WRITE; 876 else if ((uiop->uio_loffset % bsize) == 0 || 877 IS_SWAPVP(vp)) { 878 /* 879 * Have written a whole block. 880 * Start an asynchronous write 881 * and mark the buffer to 882 * indicate that it won't be 883 * needed again soon. 884 */ 885 flags = SM_WRITE | SM_ASYNC | SM_DONTNEED; 886 } else 887 flags = 0; 888 if ((ioflag & (FSYNC|FDSYNC)) || 889 (rp->r_flags & ROUTOFSPACE)) { 890 flags &= ~SM_ASYNC; 891 flags |= SM_WRITE; 892 } 893 if (vpm_enable) { 894 error = vpm_sync_pages(vp, off, n, flags); 895 } else { 896 error = segmap_release(segkmap, base, flags); 897 } 898 } else { 899 if (vpm_enable) { 900 (void) vpm_sync_pages(vp, off, n, 0); 901 } else { 902 (void) segmap_release(segkmap, base, 0); 903 } 904 /* 905 * In the event that we got an access error while 906 * faulting in a page for a write-only file just 907 * force a write. 908 */ 909 if (error == EACCES) 910 goto nfs3_fwrite; 911 } 912 } while (!error && uiop->uio_resid > 0); 913 914 bottom: 915 if (error) { 916 uiop->uio_resid = resid + remainder; 917 uiop->uio_loffset = offset; 918 } else 919 uiop->uio_resid += remainder; 920 921 nfs_rw_exit(&rp->r_lkserlock); 922 923 return (error); 924 } 925 926 /* 927 * Flags are composed of {B_ASYNC, B_INVAL, B_FREE, B_DONTNEED} 928 */ 929 static int 930 nfs3_rdwrlbn(vnode_t *vp, page_t *pp, u_offset_t off, size_t len, 931 int flags, cred_t *cr) 932 { 933 struct buf *bp; 934 int error; 935 page_t *savepp; 936 uchar_t fsdata; 937 stable_how stab_comm; 938 939 ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 940 bp = pageio_setup(pp, len, vp, flags); 941 ASSERT(bp != NULL); 942 943 /* 944 * pageio_setup should have set b_addr to 0. This 945 * is correct since we want to do I/O on a page 946 * boundary. bp_mapin will use this addr to calculate 947 * an offset, and then set b_addr to the kernel virtual 948 * address it allocated for us. 949 */ 950 ASSERT(bp->b_un.b_addr == 0); 951 952 bp->b_edev = 0; 953 bp->b_dev = 0; 954 bp->b_lblkno = lbtodb(off); 955 bp->b_file = vp; 956 bp->b_offset = (offset_t)off; 957 bp_mapin(bp); 958 959 /* 960 * Calculate the desired level of stability to write data 961 * on the server and then mark all of the pages to reflect 962 * this. 963 */ 964 if ((flags & (B_WRITE|B_ASYNC)) == (B_WRITE|B_ASYNC) && 965 freemem > desfree) { 966 stab_comm = UNSTABLE; 967 fsdata = C_DELAYCOMMIT; 968 } else { 969 stab_comm = FILE_SYNC; 970 fsdata = C_NOCOMMIT; 971 } 972 973 savepp = pp; 974 do { 975 pp->p_fsdata = fsdata; 976 } while ((pp = pp->p_next) != savepp); 977 978 error = nfs3_bio(bp, &stab_comm, cr); 979 980 bp_mapout(bp); 981 pageio_done(bp); 982 983 /* 984 * If the server wrote pages in a more stable fashion than 985 * was requested, then clear all of the marks in the pages 986 * indicating that COMMIT operations were required. 987 */ 988 if (stab_comm != UNSTABLE && fsdata == C_DELAYCOMMIT) { 989 do { 990 pp->p_fsdata = C_NOCOMMIT; 991 } while ((pp = pp->p_next) != savepp); 992 } 993 994 return (error); 995 } 996 997 /* 998 * Write to file. Writes to remote server in largest size 999 * chunks that the server can handle. Write is synchronous. 1000 */ 1001 static int 1002 nfs3write(vnode_t *vp, caddr_t base, u_offset_t offset, int count, cred_t *cr, 1003 stable_how *stab_comm) 1004 { 1005 mntinfo_t *mi; 1006 WRITE3args args; 1007 WRITE3res res; 1008 int error; 1009 int tsize; 1010 rnode_t *rp; 1011 int douprintf; 1012 1013 rp = VTOR(vp); 1014 mi = VTOMI(vp); 1015 1016 ASSERT(nfs_zone() == mi->mi_zone); 1017 1018 args.file = *VTOFH3(vp); 1019 args.stable = *stab_comm; 1020 1021 *stab_comm = FILE_SYNC; 1022 1023 douprintf = 1; 1024 1025 do { 1026 if ((vp->v_flag & VNOCACHE) || 1027 (rp->r_flags & RDIRECTIO) || 1028 (mi->mi_flags & MI_DIRECTIO)) 1029 tsize = MIN(mi->mi_stsize, count); 1030 else 1031 tsize = MIN(mi->mi_curwrite, count); 1032 args.offset = (offset3)offset; 1033 args.count = (count3)tsize; 1034 args.data.data_len = (uint_t)tsize; 1035 args.data.data_val = base; 1036 1037 if (mi->mi_io_kstats) { 1038 mutex_enter(&mi->mi_lock); 1039 kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats)); 1040 mutex_exit(&mi->mi_lock); 1041 } 1042 args.mblk = NULL; 1043 do { 1044 error = rfs3call(mi, NFSPROC3_WRITE, 1045 xdr_WRITE3args, (caddr_t)&args, 1046 xdr_WRITE3res, (caddr_t)&res, cr, 1047 &douprintf, &res.status, 0, NULL); 1048 } while (error == ENFS_TRYAGAIN); 1049 if (mi->mi_io_kstats) { 1050 mutex_enter(&mi->mi_lock); 1051 kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats)); 1052 mutex_exit(&mi->mi_lock); 1053 } 1054 1055 if (error) 1056 return (error); 1057 error = geterrno3(res.status); 1058 if (!error) { 1059 if (res.resok.count > args.count) { 1060 zcmn_err(getzoneid(), CE_WARN, 1061 "nfs3write: server %s wrote %u, " 1062 "requested was %u", 1063 rp->r_server->sv_hostname, 1064 res.resok.count, args.count); 1065 return (EIO); 1066 } 1067 if (res.resok.committed == UNSTABLE) { 1068 *stab_comm = UNSTABLE; 1069 if (args.stable == DATA_SYNC || 1070 args.stable == FILE_SYNC) { 1071 zcmn_err(getzoneid(), CE_WARN, 1072 "nfs3write: server %s did not commit to stable storage", 1073 rp->r_server->sv_hostname); 1074 return (EIO); 1075 } 1076 } 1077 tsize = (int)res.resok.count; 1078 count -= tsize; 1079 base += tsize; 1080 offset += tsize; 1081 if (mi->mi_io_kstats) { 1082 mutex_enter(&mi->mi_lock); 1083 KSTAT_IO_PTR(mi->mi_io_kstats)->writes++; 1084 KSTAT_IO_PTR(mi->mi_io_kstats)->nwritten += 1085 tsize; 1086 mutex_exit(&mi->mi_lock); 1087 } 1088 lwp_stat_update(LWP_STAT_OUBLK, 1); 1089 mutex_enter(&rp->r_statelock); 1090 if (rp->r_flags & RHAVEVERF) { 1091 if (rp->r_verf != res.resok.verf) { 1092 nfs3_set_mod(vp); 1093 rp->r_verf = res.resok.verf; 1094 /* 1095 * If the data was written UNSTABLE, 1096 * then might as well stop because 1097 * the whole block will have to get 1098 * rewritten anyway. 1099 */ 1100 if (*stab_comm == UNSTABLE) { 1101 mutex_exit(&rp->r_statelock); 1102 break; 1103 } 1104 } 1105 } else { 1106 rp->r_verf = res.resok.verf; 1107 rp->r_flags |= RHAVEVERF; 1108 } 1109 /* 1110 * Mark the attribute cache as timed out and 1111 * set RWRITEATTR to indicate that the file 1112 * was modified with a WRITE operation and 1113 * that the attributes can not be trusted. 1114 */ 1115 PURGE_ATTRCACHE_LOCKED(rp); 1116 rp->r_flags |= RWRITEATTR; 1117 mutex_exit(&rp->r_statelock); 1118 } 1119 } while (!error && count); 1120 1121 return (error); 1122 } 1123 1124 /* 1125 * Read from a file. Reads data in largest chunks our interface can handle. 1126 */ 1127 static int 1128 nfs3read(vnode_t *vp, caddr_t base, offset_t offset, int count, 1129 size_t *residp, cred_t *cr) 1130 { 1131 mntinfo_t *mi; 1132 READ3args args; 1133 READ3vres res; 1134 int tsize; 1135 int error; 1136 int douprintf; 1137 failinfo_t fi; 1138 rnode_t *rp; 1139 struct vattr va; 1140 hrtime_t t; 1141 1142 rp = VTOR(vp); 1143 mi = VTOMI(vp); 1144 ASSERT(nfs_zone() == mi->mi_zone); 1145 douprintf = 1; 1146 1147 args.file = *VTOFH3(vp); 1148 fi.vp = vp; 1149 fi.fhp = (caddr_t)&args.file; 1150 fi.copyproc = nfs3copyfh; 1151 fi.lookupproc = nfs3lookup; 1152 fi.xattrdirproc = acl_getxattrdir3; 1153 1154 res.pov.fres.vp = vp; 1155 res.pov.fres.vap = &va; 1156 1157 res.wlist = NULL; 1158 *residp = count; 1159 do { 1160 if (mi->mi_io_kstats) { 1161 mutex_enter(&mi->mi_lock); 1162 kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats)); 1163 mutex_exit(&mi->mi_lock); 1164 } 1165 1166 do { 1167 if ((vp->v_flag & VNOCACHE) || 1168 (rp->r_flags & RDIRECTIO) || 1169 (mi->mi_flags & MI_DIRECTIO)) 1170 tsize = MIN(mi->mi_tsize, count); 1171 else 1172 tsize = MIN(mi->mi_curread, count); 1173 res.data.data_val = base; 1174 res.data.data_len = tsize; 1175 args.offset = (offset3)offset; 1176 args.count = (count3)tsize; 1177 args.res_uiop = NULL; 1178 args.res_data_val_alt = base; 1179 1180 t = gethrtime(); 1181 error = rfs3call(mi, NFSPROC3_READ, 1182 xdr_READ3args, (caddr_t)&args, 1183 xdr_READ3vres, (caddr_t)&res, cr, 1184 &douprintf, &res.status, 0, &fi); 1185 } while (error == ENFS_TRYAGAIN); 1186 1187 if (mi->mi_io_kstats) { 1188 mutex_enter(&mi->mi_lock); 1189 kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats)); 1190 mutex_exit(&mi->mi_lock); 1191 } 1192 1193 if (error) 1194 return (error); 1195 1196 error = geterrno3(res.status); 1197 if (error) 1198 return (error); 1199 1200 if (res.count != res.data.data_len) { 1201 zcmn_err(getzoneid(), CE_WARN, 1202 "nfs3read: server %s returned incorrect amount", 1203 rp->r_server->sv_hostname); 1204 return (EIO); 1205 } 1206 1207 count -= res.count; 1208 *residp = count; 1209 base += res.count; 1210 offset += res.count; 1211 if (mi->mi_io_kstats) { 1212 mutex_enter(&mi->mi_lock); 1213 KSTAT_IO_PTR(mi->mi_io_kstats)->reads++; 1214 KSTAT_IO_PTR(mi->mi_io_kstats)->nread += res.count; 1215 mutex_exit(&mi->mi_lock); 1216 } 1217 lwp_stat_update(LWP_STAT_INBLK, 1); 1218 } while (count && !res.eof); 1219 1220 if (res.pov.attributes) { 1221 mutex_enter(&rp->r_statelock); 1222 if (!CACHE_VALID(rp, va.va_mtime, va.va_size)) { 1223 mutex_exit(&rp->r_statelock); 1224 PURGE_ATTRCACHE(vp); 1225 } else { 1226 if (rp->r_mtime <= t) 1227 nfs_attrcache_va(vp, &va); 1228 mutex_exit(&rp->r_statelock); 1229 } 1230 } 1231 1232 return (0); 1233 } 1234 1235 /* ARGSUSED */ 1236 static int 1237 nfs3_ioctl(vnode_t *vp, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp, 1238 caller_context_t *ct) 1239 { 1240 1241 if (nfs_zone() != VTOMI(vp)->mi_zone) 1242 return (EIO); 1243 switch (cmd) { 1244 case _FIODIRECTIO: 1245 return (nfs_directio(vp, (int)arg, cr)); 1246 default: 1247 return (ENOTTY); 1248 } 1249 } 1250 1251 /* ARGSUSED */ 1252 static int 1253 nfs3_getattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr, 1254 caller_context_t *ct) 1255 { 1256 int error; 1257 rnode_t *rp; 1258 1259 if (nfs_zone() != VTOMI(vp)->mi_zone) 1260 return (EIO); 1261 /* 1262 * If it has been specified that the return value will 1263 * just be used as a hint, and we are only being asked 1264 * for size, fsid or rdevid, then return the client's 1265 * notion of these values without checking to make sure 1266 * that the attribute cache is up to date. 1267 * The whole point is to avoid an over the wire GETATTR 1268 * call. 1269 */ 1270 rp = VTOR(vp); 1271 if (flags & ATTR_HINT) { 1272 if (vap->va_mask == 1273 (vap->va_mask & (AT_SIZE | AT_FSID | AT_RDEV))) { 1274 mutex_enter(&rp->r_statelock); 1275 if (vap->va_mask | AT_SIZE) 1276 vap->va_size = rp->r_size; 1277 if (vap->va_mask | AT_FSID) 1278 vap->va_fsid = rp->r_attr.va_fsid; 1279 if (vap->va_mask | AT_RDEV) 1280 vap->va_rdev = rp->r_attr.va_rdev; 1281 mutex_exit(&rp->r_statelock); 1282 return (0); 1283 } 1284 } 1285 1286 /* 1287 * Only need to flush pages if asking for the mtime 1288 * and if there any dirty pages or any outstanding 1289 * asynchronous (write) requests for this file. 1290 */ 1291 if (vap->va_mask & AT_MTIME) { 1292 if (vn_has_cached_data(vp) && 1293 ((rp->r_flags & RDIRTY) || rp->r_awcount > 0)) { 1294 mutex_enter(&rp->r_statelock); 1295 rp->r_gcount++; 1296 mutex_exit(&rp->r_statelock); 1297 error = nfs3_putpage(vp, (offset_t)0, 0, 0, cr, ct); 1298 mutex_enter(&rp->r_statelock); 1299 if (error && (error == ENOSPC || error == EDQUOT)) { 1300 if (!rp->r_error) 1301 rp->r_error = error; 1302 } 1303 if (--rp->r_gcount == 0) 1304 cv_broadcast(&rp->r_cv); 1305 mutex_exit(&rp->r_statelock); 1306 } 1307 } 1308 1309 return (nfs3getattr(vp, vap, cr)); 1310 } 1311 1312 /*ARGSUSED4*/ 1313 static int 1314 nfs3_setattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr, 1315 caller_context_t *ct) 1316 { 1317 int error; 1318 struct vattr va; 1319 1320 if (vap->va_mask & AT_NOSET) 1321 return (EINVAL); 1322 if (nfs_zone() != VTOMI(vp)->mi_zone) 1323 return (EIO); 1324 1325 va.va_mask = AT_UID | AT_MODE; 1326 error = nfs3getattr(vp, &va, cr); 1327 if (error) 1328 return (error); 1329 1330 error = secpolicy_vnode_setattr(cr, vp, vap, &va, flags, nfs3_accessx, 1331 vp); 1332 if (error) 1333 return (error); 1334 1335 return (nfs3setattr(vp, vap, flags, cr)); 1336 } 1337 1338 static int 1339 nfs3setattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr) 1340 { 1341 int error; 1342 uint_t mask; 1343 SETATTR3args args; 1344 SETATTR3res res; 1345 int douprintf; 1346 rnode_t *rp; 1347 struct vattr va; 1348 mode_t omode; 1349 vsecattr_t *vsp; 1350 hrtime_t t; 1351 1352 ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 1353 mask = vap->va_mask; 1354 1355 rp = VTOR(vp); 1356 1357 /* 1358 * Only need to flush pages if there are any pages and 1359 * if the file is marked as dirty in some fashion. The 1360 * file must be flushed so that we can accurately 1361 * determine the size of the file and the cached data 1362 * after the SETATTR returns. A file is considered to 1363 * be dirty if it is either marked with RDIRTY, has 1364 * outstanding i/o's active, or is mmap'd. In this 1365 * last case, we can't tell whether there are dirty 1366 * pages, so we flush just to be sure. 1367 */ 1368 if (vn_has_cached_data(vp) && 1369 ((rp->r_flags & RDIRTY) || 1370 rp->r_count > 0 || 1371 rp->r_mapcnt > 0)) { 1372 ASSERT(vp->v_type != VCHR); 1373 error = nfs3_putpage(vp, (offset_t)0, 0, 0, cr, NULL); 1374 if (error && (error == ENOSPC || error == EDQUOT)) { 1375 mutex_enter(&rp->r_statelock); 1376 if (!rp->r_error) 1377 rp->r_error = error; 1378 mutex_exit(&rp->r_statelock); 1379 } 1380 } 1381 1382 args.object = *RTOFH3(rp); 1383 /* 1384 * If the intent is for the server to set the times, 1385 * there is no point in have the mask indicating set mtime or 1386 * atime, because the vap values may be junk, and so result 1387 * in an overflow error. Remove these flags from the vap mask 1388 * before calling in this case, and restore them afterwards. 1389 */ 1390 if ((mask & (AT_ATIME | AT_MTIME)) && !(flags & ATTR_UTIME)) { 1391 /* Use server times, so don't set the args time fields */ 1392 vap->va_mask &= ~(AT_ATIME | AT_MTIME); 1393 error = vattr_to_sattr3(vap, &args.new_attributes); 1394 vap->va_mask |= (mask & (AT_ATIME | AT_MTIME)); 1395 if (mask & AT_ATIME) { 1396 args.new_attributes.atime.set_it = SET_TO_SERVER_TIME; 1397 } 1398 if (mask & AT_MTIME) { 1399 args.new_attributes.mtime.set_it = SET_TO_SERVER_TIME; 1400 } 1401 } else { 1402 /* Either do not set times or use the client specified times */ 1403 error = vattr_to_sattr3(vap, &args.new_attributes); 1404 } 1405 1406 if (error) { 1407 /* req time field(s) overflow - return immediately */ 1408 return (error); 1409 } 1410 1411 va.va_mask = AT_MODE | AT_CTIME; 1412 error = nfs3getattr(vp, &va, cr); 1413 if (error) 1414 return (error); 1415 omode = va.va_mode; 1416 1417 tryagain: 1418 if (mask & AT_SIZE) { 1419 args.guard.check = TRUE; 1420 args.guard.obj_ctime.seconds = va.va_ctime.tv_sec; 1421 args.guard.obj_ctime.nseconds = va.va_ctime.tv_nsec; 1422 } else 1423 args.guard.check = FALSE; 1424 1425 douprintf = 1; 1426 1427 t = gethrtime(); 1428 1429 error = rfs3call(VTOMI(vp), NFSPROC3_SETATTR, 1430 xdr_SETATTR3args, (caddr_t)&args, 1431 xdr_SETATTR3res, (caddr_t)&res, cr, 1432 &douprintf, &res.status, 0, NULL); 1433 1434 /* 1435 * Purge the access cache and ACL cache if changing either the 1436 * owner of the file, the group owner, or the mode. These may 1437 * change the access permissions of the file, so purge old 1438 * information and start over again. 1439 */ 1440 if (mask & (AT_UID | AT_GID | AT_MODE)) { 1441 (void) nfs_access_purge_rp(rp); 1442 if (rp->r_secattr != NULL) { 1443 mutex_enter(&rp->r_statelock); 1444 vsp = rp->r_secattr; 1445 rp->r_secattr = NULL; 1446 mutex_exit(&rp->r_statelock); 1447 if (vsp != NULL) 1448 nfs_acl_free(vsp); 1449 } 1450 } 1451 1452 if (error) { 1453 PURGE_ATTRCACHE(vp); 1454 return (error); 1455 } 1456 1457 error = geterrno3(res.status); 1458 if (!error) { 1459 /* 1460 * If changing the size of the file, invalidate 1461 * any local cached data which is no longer part 1462 * of the file. We also possibly invalidate the 1463 * last page in the file. We could use 1464 * pvn_vpzero(), but this would mark the page as 1465 * modified and require it to be written back to 1466 * the server for no particularly good reason. 1467 * This way, if we access it, then we bring it 1468 * back in. A read should be cheaper than a 1469 * write. 1470 */ 1471 if (mask & AT_SIZE) { 1472 nfs_invalidate_pages(vp, 1473 (vap->va_size & PAGEMASK), cr); 1474 } 1475 nfs3_cache_wcc_data(vp, &res.resok.obj_wcc, t, cr); 1476 /* 1477 * Some servers will change the mode to clear the setuid 1478 * and setgid bits when changing the uid or gid. The 1479 * client needs to compensate appropriately. 1480 */ 1481 if (mask & (AT_UID | AT_GID)) { 1482 int terror; 1483 1484 va.va_mask = AT_MODE; 1485 terror = nfs3getattr(vp, &va, cr); 1486 if (!terror && 1487 (((mask & AT_MODE) && va.va_mode != vap->va_mode) || 1488 (!(mask & AT_MODE) && va.va_mode != omode))) { 1489 va.va_mask = AT_MODE; 1490 if (mask & AT_MODE) 1491 va.va_mode = vap->va_mode; 1492 else 1493 va.va_mode = omode; 1494 (void) nfs3setattr(vp, &va, 0, cr); 1495 } 1496 } 1497 } else { 1498 nfs3_cache_wcc_data(vp, &res.resfail.obj_wcc, t, cr); 1499 /* 1500 * If we got back a "not synchronized" error, then 1501 * we need to retry with a new guard value. The 1502 * guard value used is the change time. If the 1503 * server returned post_op_attr, then we can just 1504 * retry because we have the latest attributes. 1505 * Otherwise, we issue a GETATTR to get the latest 1506 * attributes and then retry. If we couldn't get 1507 * the attributes this way either, then we give 1508 * up because we can't complete the operation as 1509 * required. 1510 */ 1511 if (res.status == NFS3ERR_NOT_SYNC) { 1512 va.va_mask = AT_CTIME; 1513 if (nfs3getattr(vp, &va, cr) == 0) 1514 goto tryagain; 1515 } 1516 PURGE_STALE_FH(error, vp, cr); 1517 } 1518 1519 return (error); 1520 } 1521 1522 static int 1523 nfs3_accessx(void *vp, int mode, cred_t *cr) 1524 { 1525 ASSERT(nfs_zone() == VTOMI((vnode_t *)vp)->mi_zone); 1526 return (nfs3_access(vp, mode, 0, cr, NULL)); 1527 } 1528 1529 /* ARGSUSED */ 1530 static int 1531 nfs3_access(vnode_t *vp, int mode, int flags, cred_t *cr, caller_context_t *ct) 1532 { 1533 int error; 1534 ACCESS3args args; 1535 ACCESS3res res; 1536 int douprintf; 1537 uint32 acc; 1538 rnode_t *rp; 1539 cred_t *cred, *ncr, *ncrfree = NULL; 1540 failinfo_t fi; 1541 nfs_access_type_t cacc; 1542 hrtime_t t; 1543 1544 acc = 0; 1545 if (nfs_zone() != VTOMI(vp)->mi_zone) 1546 return (EIO); 1547 if (mode & VREAD) 1548 acc |= ACCESS3_READ; 1549 if (mode & VWRITE) { 1550 if (vn_is_readonly(vp) && !IS_DEVVP(vp)) 1551 return (EROFS); 1552 if (vp->v_type == VDIR) 1553 acc |= ACCESS3_DELETE; 1554 acc |= ACCESS3_MODIFY | ACCESS3_EXTEND; 1555 } 1556 if (mode & VEXEC) { 1557 if (vp->v_type == VDIR) 1558 acc |= ACCESS3_LOOKUP; 1559 else 1560 acc |= ACCESS3_EXECUTE; 1561 } 1562 1563 rp = VTOR(vp); 1564 args.object = *VTOFH3(vp); 1565 if (vp->v_type == VDIR) { 1566 args.access = ACCESS3_READ | ACCESS3_DELETE | ACCESS3_MODIFY | 1567 ACCESS3_EXTEND | ACCESS3_LOOKUP; 1568 } else { 1569 args.access = ACCESS3_READ | ACCESS3_MODIFY | ACCESS3_EXTEND | 1570 ACCESS3_EXECUTE; 1571 } 1572 fi.vp = vp; 1573 fi.fhp = (caddr_t)&args.object; 1574 fi.copyproc = nfs3copyfh; 1575 fi.lookupproc = nfs3lookup; 1576 fi.xattrdirproc = acl_getxattrdir3; 1577 1578 cred = cr; 1579 /* 1580 * ncr and ncrfree both initially 1581 * point to the memory area returned 1582 * by crnetadjust(); 1583 * ncrfree not NULL when exiting means 1584 * that we need to release it 1585 */ 1586 ncr = crnetadjust(cred); 1587 ncrfree = ncr; 1588 tryagain: 1589 if (rp->r_acache != NULL) { 1590 cacc = nfs_access_check(rp, acc, cred); 1591 if (cacc == NFS_ACCESS_ALLOWED) { 1592 if (ncrfree != NULL) 1593 crfree(ncrfree); 1594 return (0); 1595 } 1596 if (cacc == NFS_ACCESS_DENIED) { 1597 /* 1598 * If the cred can be adjusted, try again 1599 * with the new cred. 1600 */ 1601 if (ncr != NULL) { 1602 cred = ncr; 1603 ncr = NULL; 1604 goto tryagain; 1605 } 1606 if (ncrfree != NULL) 1607 crfree(ncrfree); 1608 return (EACCES); 1609 } 1610 } 1611 1612 douprintf = 1; 1613 1614 t = gethrtime(); 1615 1616 error = rfs3call(VTOMI(vp), NFSPROC3_ACCESS, 1617 xdr_ACCESS3args, (caddr_t)&args, 1618 xdr_ACCESS3res, (caddr_t)&res, cred, 1619 &douprintf, &res.status, 0, &fi); 1620 1621 if (error) { 1622 if (ncrfree != NULL) 1623 crfree(ncrfree); 1624 return (error); 1625 } 1626 1627 error = geterrno3(res.status); 1628 if (!error) { 1629 nfs3_cache_post_op_attr(vp, &res.resok.obj_attributes, t, cr); 1630 nfs_access_cache(rp, args.access, res.resok.access, cred); 1631 /* 1632 * we just cached results with cred; if cred is the 1633 * adjusted credentials from crnetadjust, we do not want 1634 * to release them before exiting: hence setting ncrfree 1635 * to NULL 1636 */ 1637 if (cred != cr) 1638 ncrfree = NULL; 1639 if ((acc & res.resok.access) != acc) { 1640 /* 1641 * If the cred can be adjusted, try again 1642 * with the new cred. 1643 */ 1644 if (ncr != NULL) { 1645 cred = ncr; 1646 ncr = NULL; 1647 goto tryagain; 1648 } 1649 error = EACCES; 1650 } 1651 } else { 1652 nfs3_cache_post_op_attr(vp, &res.resfail.obj_attributes, t, cr); 1653 PURGE_STALE_FH(error, vp, cr); 1654 } 1655 1656 if (ncrfree != NULL) 1657 crfree(ncrfree); 1658 1659 return (error); 1660 } 1661 1662 static int nfs3_do_symlink_cache = 1; 1663 1664 /* ARGSUSED */ 1665 static int 1666 nfs3_readlink(vnode_t *vp, struct uio *uiop, cred_t *cr, caller_context_t *ct) 1667 { 1668 int error; 1669 READLINK3args args; 1670 READLINK3res res; 1671 nfspath3 resdata_backup; 1672 rnode_t *rp; 1673 int douprintf; 1674 int len; 1675 failinfo_t fi; 1676 hrtime_t t; 1677 1678 /* 1679 * Can't readlink anything other than a symbolic link. 1680 */ 1681 if (vp->v_type != VLNK) 1682 return (EINVAL); 1683 if (nfs_zone() != VTOMI(vp)->mi_zone) 1684 return (EIO); 1685 1686 rp = VTOR(vp); 1687 if (nfs3_do_symlink_cache && rp->r_symlink.contents != NULL) { 1688 error = nfs3_validate_caches(vp, cr); 1689 if (error) 1690 return (error); 1691 mutex_enter(&rp->r_statelock); 1692 if (rp->r_symlink.contents != NULL) { 1693 error = uiomove(rp->r_symlink.contents, 1694 rp->r_symlink.len, UIO_READ, uiop); 1695 mutex_exit(&rp->r_statelock); 1696 return (error); 1697 } 1698 mutex_exit(&rp->r_statelock); 1699 } 1700 1701 args.symlink = *VTOFH3(vp); 1702 fi.vp = vp; 1703 fi.fhp = (caddr_t)&args.symlink; 1704 fi.copyproc = nfs3copyfh; 1705 fi.lookupproc = nfs3lookup; 1706 fi.xattrdirproc = acl_getxattrdir3; 1707 1708 res.resok.data = kmem_alloc(MAXPATHLEN, KM_SLEEP); 1709 1710 resdata_backup = res.resok.data; 1711 1712 douprintf = 1; 1713 1714 t = gethrtime(); 1715 1716 error = rfs3call(VTOMI(vp), NFSPROC3_READLINK, 1717 xdr_READLINK3args, (caddr_t)&args, 1718 xdr_READLINK3res, (caddr_t)&res, cr, 1719 &douprintf, &res.status, 0, &fi); 1720 1721 if (res.resok.data == nfs3nametoolong) 1722 error = EINVAL; 1723 1724 if (error) { 1725 kmem_free(resdata_backup, MAXPATHLEN); 1726 return (error); 1727 } 1728 1729 error = geterrno3(res.status); 1730 if (!error) { 1731 nfs3_cache_post_op_attr(vp, &res.resok.symlink_attributes, t, 1732 cr); 1733 len = strlen(res.resok.data); 1734 error = uiomove(res.resok.data, len, UIO_READ, uiop); 1735 if (nfs3_do_symlink_cache && rp->r_symlink.contents == NULL) { 1736 mutex_enter(&rp->r_statelock); 1737 if (rp->r_symlink.contents == NULL) { 1738 rp->r_symlink.contents = res.resok.data; 1739 rp->r_symlink.len = len; 1740 rp->r_symlink.size = MAXPATHLEN; 1741 mutex_exit(&rp->r_statelock); 1742 } else { 1743 mutex_exit(&rp->r_statelock); 1744 1745 kmem_free((void *)res.resok.data, MAXPATHLEN); 1746 } 1747 } else { 1748 kmem_free((void *)res.resok.data, MAXPATHLEN); 1749 } 1750 } else { 1751 nfs3_cache_post_op_attr(vp, 1752 &res.resfail.symlink_attributes, t, cr); 1753 PURGE_STALE_FH(error, vp, cr); 1754 1755 kmem_free((void *)res.resok.data, MAXPATHLEN); 1756 1757 } 1758 1759 /* 1760 * The over the wire error for attempting to readlink something 1761 * other than a symbolic link is ENXIO. However, we need to 1762 * return EINVAL instead of ENXIO, so we map it here. 1763 */ 1764 return (error == ENXIO ? EINVAL : error); 1765 } 1766 1767 /* 1768 * Flush local dirty pages to stable storage on the server. 1769 * 1770 * If FNODSYNC is specified, then there is nothing to do because 1771 * metadata changes are not cached on the client before being 1772 * sent to the server. 1773 */ 1774 /* ARGSUSED */ 1775 static int 1776 nfs3_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct) 1777 { 1778 int error; 1779 1780 if ((syncflag & FNODSYNC) || IS_SWAPVP(vp)) 1781 return (0); 1782 if (nfs_zone() != VTOMI(vp)->mi_zone) 1783 return (EIO); 1784 1785 error = nfs3_putpage_commit(vp, (offset_t)0, 0, cr); 1786 if (!error) 1787 error = VTOR(vp)->r_error; 1788 return (error); 1789 } 1790 1791 /* 1792 * Weirdness: if the file was removed or the target of a rename 1793 * operation while it was open, it got renamed instead. Here we 1794 * remove the renamed file. 1795 */ 1796 /* ARGSUSED */ 1797 static void 1798 nfs3_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) 1799 { 1800 rnode_t *rp; 1801 1802 ASSERT(vp != DNLC_NO_VNODE); 1803 1804 /* 1805 * If this is coming from the wrong zone, we let someone in the right 1806 * zone take care of it asynchronously. We can get here due to 1807 * VN_RELE() being called from pageout() or fsflush(). This call may 1808 * potentially turn into an expensive no-op if, for instance, v_count 1809 * gets incremented in the meantime, but it's still correct. 1810 */ 1811 if (nfs_zone() != VTOMI(vp)->mi_zone) { 1812 nfs_async_inactive(vp, cr, nfs3_inactive); 1813 return; 1814 } 1815 1816 rp = VTOR(vp); 1817 redo: 1818 if (rp->r_unldvp != NULL) { 1819 /* 1820 * Save the vnode pointer for the directory where the 1821 * unlinked-open file got renamed, then set it to NULL 1822 * to prevent another thread from getting here before 1823 * we're done with the remove. While we have the 1824 * statelock, make local copies of the pertinent rnode 1825 * fields. If we weren't to do this in an atomic way, the 1826 * the unl* fields could become inconsistent with respect 1827 * to each other due to a race condition between this 1828 * code and nfs_remove(). See bug report 1034328. 1829 */ 1830 mutex_enter(&rp->r_statelock); 1831 if (rp->r_unldvp != NULL) { 1832 vnode_t *unldvp; 1833 char *unlname; 1834 cred_t *unlcred; 1835 REMOVE3args args; 1836 REMOVE3res res; 1837 int douprintf; 1838 int error; 1839 hrtime_t t; 1840 1841 unldvp = rp->r_unldvp; 1842 rp->r_unldvp = NULL; 1843 unlname = rp->r_unlname; 1844 rp->r_unlname = NULL; 1845 unlcred = rp->r_unlcred; 1846 rp->r_unlcred = NULL; 1847 mutex_exit(&rp->r_statelock); 1848 1849 /* 1850 * If there are any dirty pages left, then flush 1851 * them. This is unfortunate because they just 1852 * may get thrown away during the remove operation, 1853 * but we have to do this for correctness. 1854 */ 1855 if (vn_has_cached_data(vp) && 1856 ((rp->r_flags & RDIRTY) || rp->r_count > 0)) { 1857 ASSERT(vp->v_type != VCHR); 1858 error = nfs3_putpage(vp, (offset_t)0, 0, 0, 1859 cr, ct); 1860 if (error) { 1861 mutex_enter(&rp->r_statelock); 1862 if (!rp->r_error) 1863 rp->r_error = error; 1864 mutex_exit(&rp->r_statelock); 1865 } 1866 } 1867 1868 /* 1869 * Do the remove operation on the renamed file 1870 */ 1871 setdiropargs3(&args.object, unlname, unldvp); 1872 1873 douprintf = 1; 1874 1875 t = gethrtime(); 1876 1877 error = rfs3call(VTOMI(unldvp), NFSPROC3_REMOVE, 1878 xdr_diropargs3, (caddr_t)&args, 1879 xdr_REMOVE3res, (caddr_t)&res, unlcred, 1880 &douprintf, &res.status, 0, NULL); 1881 1882 if (error) { 1883 PURGE_ATTRCACHE(unldvp); 1884 } else { 1885 error = geterrno3(res.status); 1886 if (!error) { 1887 nfs3_cache_wcc_data(unldvp, 1888 &res.resok.dir_wcc, t, cr); 1889 if (HAVE_RDDIR_CACHE(VTOR(unldvp))) 1890 nfs_purge_rddir_cache(unldvp); 1891 } else { 1892 nfs3_cache_wcc_data(unldvp, 1893 &res.resfail.dir_wcc, t, cr); 1894 PURGE_STALE_FH(error, unldvp, cr); 1895 } 1896 } 1897 1898 /* 1899 * Release stuff held for the remove 1900 */ 1901 VN_RELE(unldvp); 1902 kmem_free(unlname, MAXNAMELEN); 1903 crfree(unlcred); 1904 goto redo; 1905 } 1906 mutex_exit(&rp->r_statelock); 1907 } 1908 1909 rp_addfree(rp, cr); 1910 } 1911 1912 /* 1913 * Remote file system operations having to do with directory manipulation. 1914 */ 1915 1916 /* ARGSUSED */ 1917 static int 1918 nfs3_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp, 1919 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct, 1920 int *direntflags, pathname_t *realpnp) 1921 { 1922 int error; 1923 vnode_t *vp; 1924 vnode_t *avp = NULL; 1925 rnode_t *drp; 1926 1927 if (nfs_zone() != VTOMI(dvp)->mi_zone) 1928 return (EPERM); 1929 1930 drp = VTOR(dvp); 1931 1932 /* 1933 * Are we looking up extended attributes? If so, "dvp" is 1934 * the file or directory for which we want attributes, and 1935 * we need a lookup of the hidden attribute directory 1936 * before we lookup the rest of the path. 1937 */ 1938 if (flags & LOOKUP_XATTR) { 1939 bool_t cflag = ((flags & CREATE_XATTR_DIR) != 0); 1940 mntinfo_t *mi; 1941 1942 mi = VTOMI(dvp); 1943 if (!(mi->mi_flags & MI_EXTATTR)) 1944 return (EINVAL); 1945 1946 if (nfs_rw_enter_sig(&drp->r_rwlock, RW_READER, INTR(dvp))) 1947 return (EINTR); 1948 1949 (void) nfs3lookup_dnlc(dvp, XATTR_DIR_NAME, &avp, cr); 1950 if (avp == NULL) 1951 error = acl_getxattrdir3(dvp, &avp, cflag, cr, 0); 1952 else 1953 error = 0; 1954 1955 nfs_rw_exit(&drp->r_rwlock); 1956 1957 if (error) { 1958 if (mi->mi_flags & MI_EXTATTR) 1959 return (error); 1960 return (EINVAL); 1961 } 1962 dvp = avp; 1963 drp = VTOR(dvp); 1964 } 1965 1966 if (nfs_rw_enter_sig(&drp->r_rwlock, RW_READER, INTR(dvp))) { 1967 error = EINTR; 1968 goto out; 1969 } 1970 1971 error = nfs3lookup(dvp, nm, vpp, pnp, flags, rdir, cr, 0); 1972 1973 nfs_rw_exit(&drp->r_rwlock); 1974 1975 /* 1976 * If vnode is a device, create special vnode. 1977 */ 1978 if (!error && IS_DEVVP(*vpp)) { 1979 vp = *vpp; 1980 *vpp = specvp(vp, vp->v_rdev, vp->v_type, cr); 1981 VN_RELE(vp); 1982 } 1983 1984 out: 1985 if (avp != NULL) 1986 VN_RELE(avp); 1987 1988 return (error); 1989 } 1990 1991 static int nfs3_lookup_neg_cache = 1; 1992 1993 #ifdef DEBUG 1994 static int nfs3_lookup_dnlc_hits = 0; 1995 static int nfs3_lookup_dnlc_misses = 0; 1996 static int nfs3_lookup_dnlc_neg_hits = 0; 1997 static int nfs3_lookup_dnlc_disappears = 0; 1998 static int nfs3_lookup_dnlc_lookups = 0; 1999 #endif 2000 2001 /* ARGSUSED */ 2002 int 2003 nfs3lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp, 2004 int flags, vnode_t *rdir, cred_t *cr, int rfscall_flags) 2005 { 2006 int error; 2007 rnode_t *drp; 2008 2009 ASSERT(nfs_zone() == VTOMI(dvp)->mi_zone); 2010 /* 2011 * If lookup is for "", just return dvp. Don't need 2012 * to send it over the wire, look it up in the dnlc, 2013 * or perform any access checks. 2014 */ 2015 if (*nm == '\0') { 2016 VN_HOLD(dvp); 2017 *vpp = dvp; 2018 return (0); 2019 } 2020 2021 /* 2022 * Can't do lookups in non-directories. 2023 */ 2024 if (dvp->v_type != VDIR) 2025 return (ENOTDIR); 2026 2027 /* 2028 * If we're called with RFSCALL_SOFT, it's important that 2029 * the only rfscall is one we make directly; if we permit 2030 * an access call because we're looking up "." or validating 2031 * a dnlc hit, we'll deadlock because that rfscall will not 2032 * have the RFSCALL_SOFT set. 2033 */ 2034 if (rfscall_flags & RFSCALL_SOFT) 2035 goto callit; 2036 2037 /* 2038 * If lookup is for ".", just return dvp. Don't need 2039 * to send it over the wire or look it up in the dnlc, 2040 * just need to check access. 2041 */ 2042 if (strcmp(nm, ".") == 0) { 2043 error = nfs3_access(dvp, VEXEC, 0, cr, NULL); 2044 if (error) 2045 return (error); 2046 VN_HOLD(dvp); 2047 *vpp = dvp; 2048 return (0); 2049 } 2050 2051 drp = VTOR(dvp); 2052 if (!(drp->r_flags & RLOOKUP)) { 2053 mutex_enter(&drp->r_statelock); 2054 drp->r_flags |= RLOOKUP; 2055 mutex_exit(&drp->r_statelock); 2056 } 2057 2058 /* 2059 * Lookup this name in the DNLC. If there was a valid entry, 2060 * then return the results of the lookup. 2061 */ 2062 error = nfs3lookup_dnlc(dvp, nm, vpp, cr); 2063 if (error || *vpp != NULL) 2064 return (error); 2065 2066 callit: 2067 error = nfs3lookup_otw(dvp, nm, vpp, cr, rfscall_flags); 2068 2069 return (error); 2070 } 2071 2072 static int 2073 nfs3lookup_dnlc(vnode_t *dvp, char *nm, vnode_t **vpp, cred_t *cr) 2074 { 2075 int error; 2076 vnode_t *vp; 2077 2078 ASSERT(*nm != '\0'); 2079 ASSERT(nfs_zone() == VTOMI(dvp)->mi_zone); 2080 /* 2081 * Lookup this name in the DNLC. If successful, then validate 2082 * the caches and then recheck the DNLC. The DNLC is rechecked 2083 * just in case this entry got invalidated during the call 2084 * to nfs3_validate_caches. 2085 * 2086 * An assumption is being made that it is safe to say that a 2087 * file exists which may not on the server. Any operations to 2088 * the server will fail with ESTALE. 2089 */ 2090 #ifdef DEBUG 2091 nfs3_lookup_dnlc_lookups++; 2092 #endif 2093 vp = dnlc_lookup(dvp, nm); 2094 if (vp != NULL) { 2095 VN_RELE(vp); 2096 if (vp == DNLC_NO_VNODE && !vn_is_readonly(dvp)) { 2097 PURGE_ATTRCACHE(dvp); 2098 } 2099 error = nfs3_validate_caches(dvp, cr); 2100 if (error) 2101 return (error); 2102 vp = dnlc_lookup(dvp, nm); 2103 if (vp != NULL) { 2104 error = nfs3_access(dvp, VEXEC, 0, cr, NULL); 2105 if (error) { 2106 VN_RELE(vp); 2107 return (error); 2108 } 2109 if (vp == DNLC_NO_VNODE) { 2110 VN_RELE(vp); 2111 #ifdef DEBUG 2112 nfs3_lookup_dnlc_neg_hits++; 2113 #endif 2114 return (ENOENT); 2115 } 2116 *vpp = vp; 2117 #ifdef DEBUG 2118 nfs3_lookup_dnlc_hits++; 2119 #endif 2120 return (0); 2121 } 2122 #ifdef DEBUG 2123 nfs3_lookup_dnlc_disappears++; 2124 #endif 2125 } 2126 #ifdef DEBUG 2127 else 2128 nfs3_lookup_dnlc_misses++; 2129 #endif 2130 2131 *vpp = NULL; 2132 2133 return (0); 2134 } 2135 2136 static int 2137 nfs3lookup_otw(vnode_t *dvp, char *nm, vnode_t **vpp, cred_t *cr, 2138 int rfscall_flags) 2139 { 2140 int error; 2141 LOOKUP3args args; 2142 LOOKUP3vres res; 2143 int douprintf; 2144 struct vattr vattr; 2145 struct vattr dvattr; 2146 vnode_t *vp; 2147 failinfo_t fi; 2148 hrtime_t t; 2149 2150 ASSERT(*nm != '\0'); 2151 ASSERT(dvp->v_type == VDIR); 2152 ASSERT(nfs_zone() == VTOMI(dvp)->mi_zone); 2153 2154 setdiropargs3(&args.what, nm, dvp); 2155 2156 fi.vp = dvp; 2157 fi.fhp = (caddr_t)&args.what.dir; 2158 fi.copyproc = nfs3copyfh; 2159 fi.lookupproc = nfs3lookup; 2160 fi.xattrdirproc = acl_getxattrdir3; 2161 res.obj_attributes.fres.vp = dvp; 2162 res.obj_attributes.fres.vap = &vattr; 2163 res.dir_attributes.fres.vp = dvp; 2164 res.dir_attributes.fres.vap = &dvattr; 2165 2166 douprintf = 1; 2167 2168 t = gethrtime(); 2169 2170 error = rfs3call(VTOMI(dvp), NFSPROC3_LOOKUP, 2171 xdr_diropargs3, (caddr_t)&args, 2172 xdr_LOOKUP3vres, (caddr_t)&res, cr, 2173 &douprintf, &res.status, rfscall_flags, &fi); 2174 2175 if (error) 2176 return (error); 2177 2178 nfs3_cache_post_op_vattr(dvp, &res.dir_attributes, t, cr); 2179 2180 error = geterrno3(res.status); 2181 if (error) { 2182 PURGE_STALE_FH(error, dvp, cr); 2183 if (error == ENOENT && nfs3_lookup_neg_cache) 2184 dnlc_enter(dvp, nm, DNLC_NO_VNODE); 2185 return (error); 2186 } 2187 2188 if (res.obj_attributes.attributes) { 2189 vp = makenfs3node_va(&res.object, res.obj_attributes.fres.vap, 2190 dvp->v_vfsp, t, cr, VTOR(dvp)->r_path, nm); 2191 } else { 2192 vp = makenfs3node_va(&res.object, NULL, 2193 dvp->v_vfsp, t, cr, VTOR(dvp)->r_path, nm); 2194 if (vp->v_type == VNON) { 2195 vattr.va_mask = AT_TYPE; 2196 error = nfs3getattr(vp, &vattr, cr); 2197 if (error) { 2198 VN_RELE(vp); 2199 return (error); 2200 } 2201 vp->v_type = vattr.va_type; 2202 } 2203 } 2204 2205 if (!(rfscall_flags & RFSCALL_SOFT)) 2206 dnlc_update(dvp, nm, vp); 2207 2208 *vpp = vp; 2209 2210 return (error); 2211 } 2212 2213 #ifdef DEBUG 2214 static int nfs3_create_misses = 0; 2215 #endif 2216 2217 /* ARGSUSED */ 2218 static int 2219 nfs3_create(vnode_t *dvp, char *nm, struct vattr *va, enum vcexcl exclusive, 2220 int mode, vnode_t **vpp, cred_t *cr, int lfaware, caller_context_t *ct, 2221 vsecattr_t *vsecp) 2222 { 2223 int error; 2224 vnode_t *vp; 2225 rnode_t *rp; 2226 struct vattr vattr; 2227 rnode_t *drp; 2228 vnode_t *tempvp; 2229 2230 drp = VTOR(dvp); 2231 if (nfs_zone() != VTOMI(dvp)->mi_zone) 2232 return (EPERM); 2233 if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp))) 2234 return (EINTR); 2235 2236 top: 2237 /* 2238 * We make a copy of the attributes because the caller does not 2239 * expect us to change what va points to. 2240 */ 2241 vattr = *va; 2242 2243 /* 2244 * If the pathname is "", just use dvp. Don't need 2245 * to send it over the wire, look it up in the dnlc, 2246 * or perform any access checks. 2247 */ 2248 if (*nm == '\0') { 2249 error = 0; 2250 VN_HOLD(dvp); 2251 vp = dvp; 2252 /* 2253 * If the pathname is ".", just use dvp. Don't need 2254 * to send it over the wire or look it up in the dnlc, 2255 * just need to check access. 2256 */ 2257 } else if (strcmp(nm, ".") == 0) { 2258 error = nfs3_access(dvp, VEXEC, 0, cr, ct); 2259 if (error) { 2260 nfs_rw_exit(&drp->r_rwlock); 2261 return (error); 2262 } 2263 VN_HOLD(dvp); 2264 vp = dvp; 2265 /* 2266 * We need to go over the wire, just to be sure whether the 2267 * file exists or not. Using the DNLC can be dangerous in 2268 * this case when making a decision regarding existence. 2269 */ 2270 } else { 2271 error = nfs3lookup_otw(dvp, nm, &vp, cr, 0); 2272 } 2273 if (!error) { 2274 if (exclusive == EXCL) 2275 error = EEXIST; 2276 else if (vp->v_type == VDIR && (mode & VWRITE)) 2277 error = EISDIR; 2278 else { 2279 /* 2280 * If vnode is a device, create special vnode. 2281 */ 2282 if (IS_DEVVP(vp)) { 2283 tempvp = vp; 2284 vp = specvp(vp, vp->v_rdev, vp->v_type, cr); 2285 VN_RELE(tempvp); 2286 } 2287 if (!(error = VOP_ACCESS(vp, mode, 0, cr, ct))) { 2288 if ((vattr.va_mask & AT_SIZE) && 2289 vp->v_type == VREG) { 2290 rp = VTOR(vp); 2291 /* 2292 * Check here for large file handled 2293 * by LF-unaware process (as 2294 * ufs_create() does) 2295 */ 2296 if (!(lfaware & FOFFMAX)) { 2297 mutex_enter(&rp->r_statelock); 2298 if (rp->r_size > MAXOFF32_T) 2299 error = EOVERFLOW; 2300 mutex_exit(&rp->r_statelock); 2301 } 2302 if (!error) { 2303 vattr.va_mask = AT_SIZE; 2304 error = nfs3setattr(vp, 2305 &vattr, 0, cr); 2306 } 2307 } 2308 } 2309 } 2310 nfs_rw_exit(&drp->r_rwlock); 2311 if (error) { 2312 VN_RELE(vp); 2313 } else { 2314 /* 2315 * existing file got truncated, notify. 2316 */ 2317 vnevent_create(vp, ct); 2318 *vpp = vp; 2319 } 2320 return (error); 2321 } 2322 2323 dnlc_remove(dvp, nm); 2324 2325 /* 2326 * Decide what the group-id of the created file should be. 2327 * Set it in attribute list as advisory... 2328 */ 2329 error = setdirgid(dvp, &vattr.va_gid, cr); 2330 if (error) { 2331 nfs_rw_exit(&drp->r_rwlock); 2332 return (error); 2333 } 2334 vattr.va_mask |= AT_GID; 2335 2336 ASSERT(vattr.va_mask & AT_TYPE); 2337 if (vattr.va_type == VREG) { 2338 ASSERT(vattr.va_mask & AT_MODE); 2339 if (MANDMODE(vattr.va_mode)) { 2340 nfs_rw_exit(&drp->r_rwlock); 2341 return (EACCES); 2342 } 2343 error = nfs3create(dvp, nm, &vattr, exclusive, mode, vpp, cr, 2344 lfaware); 2345 /* 2346 * If this is not an exclusive create, then the CREATE 2347 * request will be made with the GUARDED mode set. This 2348 * means that the server will return EEXIST if the file 2349 * exists. The file could exist because of a retransmitted 2350 * request. In this case, we recover by starting over and 2351 * checking to see whether the file exists. This second 2352 * time through it should and a CREATE request will not be 2353 * sent. 2354 * 2355 * This handles the problem of a dangling CREATE request 2356 * which contains attributes which indicate that the file 2357 * should be truncated. This retransmitted request could 2358 * possibly truncate valid data in the file if not caught 2359 * by the duplicate request mechanism on the server or if 2360 * not caught by other means. The scenario is: 2361 * 2362 * Client transmits CREATE request with size = 0 2363 * Client times out, retransmits request. 2364 * Response to the first request arrives from the server 2365 * and the client proceeds on. 2366 * Client writes data to the file. 2367 * The server now processes retransmitted CREATE request 2368 * and truncates file. 2369 * 2370 * The use of the GUARDED CREATE request prevents this from 2371 * happening because the retransmitted CREATE would fail 2372 * with EEXIST and would not truncate the file. 2373 */ 2374 if (error == EEXIST && exclusive == NONEXCL) { 2375 #ifdef DEBUG 2376 nfs3_create_misses++; 2377 #endif 2378 goto top; 2379 } 2380 nfs_rw_exit(&drp->r_rwlock); 2381 return (error); 2382 } 2383 error = nfs3mknod(dvp, nm, &vattr, exclusive, mode, vpp, cr); 2384 nfs_rw_exit(&drp->r_rwlock); 2385 return (error); 2386 } 2387 2388 /* ARGSUSED */ 2389 static int 2390 nfs3create(vnode_t *dvp, char *nm, struct vattr *va, enum vcexcl exclusive, 2391 int mode, vnode_t **vpp, cred_t *cr, int lfaware) 2392 { 2393 int error; 2394 CREATE3args args; 2395 CREATE3res res; 2396 int douprintf; 2397 vnode_t *vp; 2398 struct vattr vattr; 2399 nfstime3 *verfp; 2400 rnode_t *rp; 2401 timestruc_t now; 2402 hrtime_t t; 2403 2404 ASSERT(nfs_zone() == VTOMI(dvp)->mi_zone); 2405 setdiropargs3(&args.where, nm, dvp); 2406 if (exclusive == EXCL) { 2407 args.how.mode = EXCLUSIVE; 2408 /* 2409 * Construct the create verifier. This verifier needs 2410 * to be unique between different clients. It also needs 2411 * to vary for each exclusive create request generated 2412 * from the client to the server. 2413 * 2414 * The first attempt is made to use the hostid and a 2415 * unique number on the client. If the hostid has not 2416 * been set, the high resolution time that the exclusive 2417 * create request is being made is used. This will work 2418 * unless two different clients, both with the hostid 2419 * not set, attempt an exclusive create request on the 2420 * same file, at exactly the same clock time. The 2421 * chances of this happening seem small enough to be 2422 * reasonable. 2423 */ 2424 verfp = (nfstime3 *)&args.how.createhow3_u.verf; 2425 verfp->seconds = zone_get_hostid(NULL); 2426 if (verfp->seconds != 0) 2427 verfp->nseconds = newnum(); 2428 else { 2429 gethrestime(&now); 2430 verfp->seconds = now.tv_sec; 2431 verfp->nseconds = now.tv_nsec; 2432 } 2433 /* 2434 * Since the server will use this value for the mtime, 2435 * make sure that it can't overflow. Zero out the MSB. 2436 * The actual value does not matter here, only its uniqeness. 2437 */ 2438 verfp->seconds %= INT32_MAX; 2439 } else { 2440 /* 2441 * Issue the non-exclusive create in guarded mode. This 2442 * may result in some false EEXIST responses for 2443 * retransmitted requests, but these will be handled at 2444 * a higher level. By using GUARDED, duplicate requests 2445 * to do file truncation and possible access problems 2446 * can be avoided. 2447 */ 2448 args.how.mode = GUARDED; 2449 error = vattr_to_sattr3(va, 2450 &args.how.createhow3_u.obj_attributes); 2451 if (error) { 2452 /* req time field(s) overflow - return immediately */ 2453 return (error); 2454 } 2455 } 2456 2457 douprintf = 1; 2458 2459 t = gethrtime(); 2460 2461 error = rfs3call(VTOMI(dvp), NFSPROC3_CREATE, 2462 xdr_CREATE3args, (caddr_t)&args, 2463 xdr_CREATE3res, (caddr_t)&res, cr, 2464 &douprintf, &res.status, 0, NULL); 2465 2466 if (error) { 2467 PURGE_ATTRCACHE(dvp); 2468 return (error); 2469 } 2470 2471 error = geterrno3(res.status); 2472 if (!error) { 2473 nfs3_cache_wcc_data(dvp, &res.resok.dir_wcc, t, cr); 2474 if (HAVE_RDDIR_CACHE(VTOR(dvp))) 2475 nfs_purge_rddir_cache(dvp); 2476 2477 /* 2478 * On exclusive create the times need to be explicitly 2479 * set to clear any potential verifier that may be stored 2480 * in one of these fields (see comment below). This 2481 * is done here to cover the case where no post op attrs 2482 * were returned or a 'invalid' time was returned in 2483 * the attributes. 2484 */ 2485 if (exclusive == EXCL) 2486 va->va_mask |= (AT_MTIME | AT_ATIME); 2487 2488 if (!res.resok.obj.handle_follows) { 2489 error = nfs3lookup(dvp, nm, &vp, NULL, 0, NULL, cr, 0); 2490 if (error) 2491 return (error); 2492 } else { 2493 if (res.resok.obj_attributes.attributes) { 2494 vp = makenfs3node(&res.resok.obj.handle, 2495 &res.resok.obj_attributes.attr, 2496 dvp->v_vfsp, t, cr, NULL, NULL); 2497 } else { 2498 vp = makenfs3node(&res.resok.obj.handle, NULL, 2499 dvp->v_vfsp, t, cr, NULL, NULL); 2500 2501 /* 2502 * On an exclusive create, it is possible 2503 * that attributes were returned but those 2504 * postop attributes failed to decode 2505 * properly. If this is the case, 2506 * then most likely the atime or mtime 2507 * were invalid for our client; this 2508 * is caused by the server storing the 2509 * create verifier in one of the time 2510 * fields(most likely mtime). 2511 * So... we are going to setattr just the 2512 * atime/mtime to clear things up. 2513 */ 2514 if (exclusive == EXCL) { 2515 if (error = 2516 nfs3excl_create_settimes(vp, 2517 va, cr)) { 2518 /* 2519 * Setting the times failed. 2520 * Remove the file and return 2521 * the error. 2522 */ 2523 VN_RELE(vp); 2524 (void) nfs3_remove(dvp, 2525 nm, cr, NULL, 0); 2526 return (error); 2527 } 2528 } 2529 2530 /* 2531 * This handles the non-exclusive case 2532 * and the exclusive case where no post op 2533 * attrs were returned. 2534 */ 2535 if (vp->v_type == VNON) { 2536 vattr.va_mask = AT_TYPE; 2537 error = nfs3getattr(vp, &vattr, cr); 2538 if (error) { 2539 VN_RELE(vp); 2540 return (error); 2541 } 2542 vp->v_type = vattr.va_type; 2543 } 2544 } 2545 dnlc_update(dvp, nm, vp); 2546 } 2547 2548 rp = VTOR(vp); 2549 2550 /* 2551 * Check here for large file handled by 2552 * LF-unaware process (as ufs_create() does) 2553 */ 2554 if ((va->va_mask & AT_SIZE) && vp->v_type == VREG && 2555 !(lfaware & FOFFMAX)) { 2556 mutex_enter(&rp->r_statelock); 2557 if (rp->r_size > MAXOFF32_T) { 2558 mutex_exit(&rp->r_statelock); 2559 VN_RELE(vp); 2560 return (EOVERFLOW); 2561 } 2562 mutex_exit(&rp->r_statelock); 2563 } 2564 2565 if (exclusive == EXCL && 2566 (va->va_mask & ~(AT_GID | AT_SIZE))) { 2567 /* 2568 * If doing an exclusive create, then generate 2569 * a SETATTR to set the initial attributes. 2570 * Try to set the mtime and the atime to the 2571 * server's current time. It is somewhat 2572 * expected that these fields will be used to 2573 * store the exclusive create cookie. If not, 2574 * server implementors will need to know that 2575 * a SETATTR will follow an exclusive create 2576 * and the cookie should be destroyed if 2577 * appropriate. This work may have been done 2578 * earlier in this function if post op attrs 2579 * were not available. 2580 * 2581 * The AT_GID and AT_SIZE bits are turned off 2582 * so that the SETATTR request will not attempt 2583 * to process these. The gid will be set 2584 * separately if appropriate. The size is turned 2585 * off because it is assumed that a new file will 2586 * be created empty and if the file wasn't empty, 2587 * then the exclusive create will have failed 2588 * because the file must have existed already. 2589 * Therefore, no truncate operation is needed. 2590 */ 2591 va->va_mask &= ~(AT_GID | AT_SIZE); 2592 error = nfs3setattr(vp, va, 0, cr); 2593 if (error) { 2594 /* 2595 * Couldn't correct the attributes of 2596 * the newly created file and the 2597 * attributes are wrong. Remove the 2598 * file and return an error to the 2599 * application. 2600 */ 2601 VN_RELE(vp); 2602 (void) nfs3_remove(dvp, nm, cr, NULL, 0); 2603 return (error); 2604 } 2605 } 2606 2607 if (va->va_gid != rp->r_attr.va_gid) { 2608 /* 2609 * If the gid on the file isn't right, then 2610 * generate a SETATTR to attempt to change 2611 * it. This may or may not work, depending 2612 * upon the server's semantics for allowing 2613 * file ownership changes. 2614 */ 2615 va->va_mask = AT_GID; 2616 (void) nfs3setattr(vp, va, 0, cr); 2617 } 2618 2619 /* 2620 * If vnode is a device create special vnode 2621 */ 2622 if (IS_DEVVP(vp)) { 2623 *vpp = specvp(vp, vp->v_rdev, vp->v_type, cr); 2624 VN_RELE(vp); 2625 } else 2626 *vpp = vp; 2627 } else { 2628 nfs3_cache_wcc_data(dvp, &res.resfail.dir_wcc, t, cr); 2629 PURGE_STALE_FH(error, dvp, cr); 2630 } 2631 2632 return (error); 2633 } 2634 2635 /* 2636 * Special setattr function to take care of rest of atime/mtime 2637 * after successful exclusive create. This function exists to avoid 2638 * handling attributes from the server; exclusive the atime/mtime fields 2639 * may be 'invalid' in client's view and therefore can not be trusted. 2640 */ 2641 static int 2642 nfs3excl_create_settimes(vnode_t *vp, struct vattr *vap, cred_t *cr) 2643 { 2644 int error; 2645 uint_t mask; 2646 SETATTR3args args; 2647 SETATTR3res res; 2648 int douprintf; 2649 rnode_t *rp; 2650 hrtime_t t; 2651 2652 ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 2653 /* save the caller's mask so that it can be reset later */ 2654 mask = vap->va_mask; 2655 2656 rp = VTOR(vp); 2657 2658 args.object = *RTOFH3(rp); 2659 args.guard.check = FALSE; 2660 2661 /* Use the mask to initialize the arguments */ 2662 vap->va_mask = 0; 2663 error = vattr_to_sattr3(vap, &args.new_attributes); 2664 2665 /* We want to set just atime/mtime on this request */ 2666 args.new_attributes.atime.set_it = SET_TO_SERVER_TIME; 2667 args.new_attributes.mtime.set_it = SET_TO_SERVER_TIME; 2668 2669 douprintf = 1; 2670 2671 t = gethrtime(); 2672 2673 error = rfs3call(VTOMI(vp), NFSPROC3_SETATTR, 2674 xdr_SETATTR3args, (caddr_t)&args, 2675 xdr_SETATTR3res, (caddr_t)&res, cr, 2676 &douprintf, &res.status, 0, NULL); 2677 2678 if (error) { 2679 vap->va_mask = mask; 2680 return (error); 2681 } 2682 2683 error = geterrno3(res.status); 2684 if (!error) { 2685 /* 2686 * It is important to pick up the attributes. 2687 * Since this is the exclusive create path, the 2688 * attributes on the initial create were ignored 2689 * and we need these to have the correct info. 2690 */ 2691 nfs3_cache_wcc_data(vp, &res.resok.obj_wcc, t, cr); 2692 /* 2693 * No need to do the atime/mtime work again so clear 2694 * the bits. 2695 */ 2696 mask &= ~(AT_ATIME | AT_MTIME); 2697 } else { 2698 nfs3_cache_wcc_data(vp, &res.resfail.obj_wcc, t, cr); 2699 } 2700 2701 vap->va_mask = mask; 2702 2703 return (error); 2704 } 2705 2706 /* ARGSUSED */ 2707 static int 2708 nfs3mknod(vnode_t *dvp, char *nm, struct vattr *va, enum vcexcl exclusive, 2709 int mode, vnode_t **vpp, cred_t *cr) 2710 { 2711 int error; 2712 MKNOD3args args; 2713 MKNOD3res res; 2714 int douprintf; 2715 vnode_t *vp; 2716 struct vattr vattr; 2717 hrtime_t t; 2718 2719 ASSERT(nfs_zone() == VTOMI(dvp)->mi_zone); 2720 switch (va->va_type) { 2721 case VCHR: 2722 case VBLK: 2723 setdiropargs3(&args.where, nm, dvp); 2724 args.what.type = (va->va_type == VCHR) ? NF3CHR : NF3BLK; 2725 error = vattr_to_sattr3(va, 2726 &args.what.mknoddata3_u.device.dev_attributes); 2727 if (error) { 2728 /* req time field(s) overflow - return immediately */ 2729 return (error); 2730 } 2731 args.what.mknoddata3_u.device.spec.specdata1 = 2732 getmajor(va->va_rdev); 2733 args.what.mknoddata3_u.device.spec.specdata2 = 2734 getminor(va->va_rdev); 2735 break; 2736 2737 case VFIFO: 2738 case VSOCK: 2739 setdiropargs3(&args.where, nm, dvp); 2740 args.what.type = (va->va_type == VFIFO) ? NF3FIFO : NF3SOCK; 2741 error = vattr_to_sattr3(va, 2742 &args.what.mknoddata3_u.pipe_attributes); 2743 if (error) { 2744 /* req time field(s) overflow - return immediately */ 2745 return (error); 2746 } 2747 break; 2748 2749 default: 2750 return (EINVAL); 2751 } 2752 2753 douprintf = 1; 2754 2755 t = gethrtime(); 2756 2757 error = rfs3call(VTOMI(dvp), NFSPROC3_MKNOD, 2758 xdr_MKNOD3args, (caddr_t)&args, 2759 xdr_MKNOD3res, (caddr_t)&res, cr, 2760 &douprintf, &res.status, 0, NULL); 2761 2762 if (error) { 2763 PURGE_ATTRCACHE(dvp); 2764 return (error); 2765 } 2766 2767 error = geterrno3(res.status); 2768 if (!error) { 2769 nfs3_cache_wcc_data(dvp, &res.resok.dir_wcc, t, cr); 2770 if (HAVE_RDDIR_CACHE(VTOR(dvp))) 2771 nfs_purge_rddir_cache(dvp); 2772 2773 if (!res.resok.obj.handle_follows) { 2774 error = nfs3lookup(dvp, nm, &vp, NULL, 0, NULL, cr, 0); 2775 if (error) 2776 return (error); 2777 } else { 2778 if (res.resok.obj_attributes.attributes) { 2779 vp = makenfs3node(&res.resok.obj.handle, 2780 &res.resok.obj_attributes.attr, 2781 dvp->v_vfsp, t, cr, NULL, NULL); 2782 } else { 2783 vp = makenfs3node(&res.resok.obj.handle, NULL, 2784 dvp->v_vfsp, t, cr, NULL, NULL); 2785 if (vp->v_type == VNON) { 2786 vattr.va_mask = AT_TYPE; 2787 error = nfs3getattr(vp, &vattr, cr); 2788 if (error) { 2789 VN_RELE(vp); 2790 return (error); 2791 } 2792 vp->v_type = vattr.va_type; 2793 } 2794 2795 } 2796 dnlc_update(dvp, nm, vp); 2797 } 2798 2799 if (va->va_gid != VTOR(vp)->r_attr.va_gid) { 2800 va->va_mask = AT_GID; 2801 (void) nfs3setattr(vp, va, 0, cr); 2802 } 2803 2804 /* 2805 * If vnode is a device create special vnode 2806 */ 2807 if (IS_DEVVP(vp)) { 2808 *vpp = specvp(vp, vp->v_rdev, vp->v_type, cr); 2809 VN_RELE(vp); 2810 } else 2811 *vpp = vp; 2812 } else { 2813 nfs3_cache_wcc_data(dvp, &res.resfail.dir_wcc, t, cr); 2814 PURGE_STALE_FH(error, dvp, cr); 2815 } 2816 return (error); 2817 } 2818 2819 /* 2820 * Weirdness: if the vnode to be removed is open 2821 * we rename it instead of removing it and nfs_inactive 2822 * will remove the new name. 2823 */ 2824 /* ARGSUSED */ 2825 static int 2826 nfs3_remove(vnode_t *dvp, char *nm, cred_t *cr, caller_context_t *ct, int flags) 2827 { 2828 int error; 2829 REMOVE3args args; 2830 REMOVE3res res; 2831 vnode_t *vp; 2832 char *tmpname; 2833 int douprintf; 2834 rnode_t *rp; 2835 rnode_t *drp; 2836 hrtime_t t; 2837 2838 if (nfs_zone() != VTOMI(dvp)->mi_zone) 2839 return (EPERM); 2840 drp = VTOR(dvp); 2841 if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp))) 2842 return (EINTR); 2843 2844 error = nfs3lookup(dvp, nm, &vp, NULL, 0, NULL, cr, 0); 2845 if (error) { 2846 nfs_rw_exit(&drp->r_rwlock); 2847 return (error); 2848 } 2849 2850 if (vp->v_type == VDIR && secpolicy_fs_linkdir(cr, dvp->v_vfsp)) { 2851 VN_RELE(vp); 2852 nfs_rw_exit(&drp->r_rwlock); 2853 return (EPERM); 2854 } 2855 2856 /* 2857 * First just remove the entry from the name cache, as it 2858 * is most likely the only entry for this vp. 2859 */ 2860 dnlc_remove(dvp, nm); 2861 2862 /* 2863 * If the file has a v_count > 1 then there may be more than one 2864 * entry in the name cache due multiple links or an open file, 2865 * but we don't have the real reference count so flush all 2866 * possible entries. 2867 */ 2868 if (vp->v_count > 1) 2869 dnlc_purge_vp(vp); 2870 2871 /* 2872 * Now we have the real reference count on the vnode 2873 */ 2874 rp = VTOR(vp); 2875 mutex_enter(&rp->r_statelock); 2876 if (vp->v_count > 1 && 2877 (rp->r_unldvp == NULL || strcmp(nm, rp->r_unlname) == 0)) { 2878 mutex_exit(&rp->r_statelock); 2879 tmpname = newname(); 2880 error = nfs3rename(dvp, nm, dvp, tmpname, cr, ct); 2881 if (error) 2882 kmem_free(tmpname, MAXNAMELEN); 2883 else { 2884 mutex_enter(&rp->r_statelock); 2885 if (rp->r_unldvp == NULL) { 2886 VN_HOLD(dvp); 2887 rp->r_unldvp = dvp; 2888 if (rp->r_unlcred != NULL) 2889 crfree(rp->r_unlcred); 2890 crhold(cr); 2891 rp->r_unlcred = cr; 2892 rp->r_unlname = tmpname; 2893 } else { 2894 kmem_free(rp->r_unlname, MAXNAMELEN); 2895 rp->r_unlname = tmpname; 2896 } 2897 mutex_exit(&rp->r_statelock); 2898 } 2899 } else { 2900 mutex_exit(&rp->r_statelock); 2901 /* 2902 * We need to flush any dirty pages which happen to 2903 * be hanging around before removing the file. This 2904 * shouldn't happen very often and mostly on file 2905 * systems mounted "nocto". 2906 */ 2907 if (vn_has_cached_data(vp) && 2908 ((rp->r_flags & RDIRTY) || rp->r_count > 0)) { 2909 error = nfs3_putpage(vp, (offset_t)0, 0, 0, cr, ct); 2910 if (error && (error == ENOSPC || error == EDQUOT)) { 2911 mutex_enter(&rp->r_statelock); 2912 if (!rp->r_error) 2913 rp->r_error = error; 2914 mutex_exit(&rp->r_statelock); 2915 } 2916 } 2917 2918 setdiropargs3(&args.object, nm, dvp); 2919 2920 douprintf = 1; 2921 2922 t = gethrtime(); 2923 2924 error = rfs3call(VTOMI(dvp), NFSPROC3_REMOVE, 2925 xdr_diropargs3, (caddr_t)&args, 2926 xdr_REMOVE3res, (caddr_t)&res, cr, 2927 &douprintf, &res.status, 0, NULL); 2928 2929 /* 2930 * The xattr dir may be gone after last attr is removed, 2931 * so flush it from dnlc. 2932 */ 2933 if (dvp->v_flag & V_XATTRDIR) 2934 dnlc_purge_vp(dvp); 2935 2936 PURGE_ATTRCACHE(vp); 2937 2938 if (error) { 2939 PURGE_ATTRCACHE(dvp); 2940 } else { 2941 error = geterrno3(res.status); 2942 if (!error) { 2943 nfs3_cache_wcc_data(dvp, &res.resok.dir_wcc, t, 2944 cr); 2945 if (HAVE_RDDIR_CACHE(drp)) 2946 nfs_purge_rddir_cache(dvp); 2947 } else { 2948 nfs3_cache_wcc_data(dvp, &res.resfail.dir_wcc, 2949 t, cr); 2950 PURGE_STALE_FH(error, dvp, cr); 2951 } 2952 } 2953 } 2954 2955 if (error == 0) { 2956 vnevent_remove(vp, dvp, nm, ct); 2957 } 2958 VN_RELE(vp); 2959 2960 nfs_rw_exit(&drp->r_rwlock); 2961 2962 return (error); 2963 } 2964 2965 /* ARGSUSED */ 2966 static int 2967 nfs3_link(vnode_t *tdvp, vnode_t *svp, char *tnm, cred_t *cr, 2968 caller_context_t *ct, int flags) 2969 { 2970 int error; 2971 LINK3args args; 2972 LINK3res res; 2973 vnode_t *realvp; 2974 int douprintf; 2975 mntinfo_t *mi; 2976 rnode_t *tdrp; 2977 hrtime_t t; 2978 2979 if (nfs_zone() != VTOMI(tdvp)->mi_zone) 2980 return (EPERM); 2981 if (VOP_REALVP(svp, &realvp, ct) == 0) 2982 svp = realvp; 2983 2984 mi = VTOMI(svp); 2985 2986 if (!(mi->mi_flags & MI_LINK)) 2987 return (EOPNOTSUPP); 2988 2989 args.file = *VTOFH3(svp); 2990 setdiropargs3(&args.link, tnm, tdvp); 2991 2992 tdrp = VTOR(tdvp); 2993 if (nfs_rw_enter_sig(&tdrp->r_rwlock, RW_WRITER, INTR(tdvp))) 2994 return (EINTR); 2995 2996 dnlc_remove(tdvp, tnm); 2997 2998 douprintf = 1; 2999 3000 t = gethrtime(); 3001 3002 error = rfs3call(mi, NFSPROC3_LINK, 3003 xdr_LINK3args, (caddr_t)&args, 3004 xdr_LINK3res, (caddr_t)&res, cr, 3005 &douprintf, &res.status, 0, NULL); 3006 3007 if (error) { 3008 PURGE_ATTRCACHE(tdvp); 3009 PURGE_ATTRCACHE(svp); 3010 nfs_rw_exit(&tdrp->r_rwlock); 3011 return (error); 3012 } 3013 3014 error = geterrno3(res.status); 3015 3016 if (!error) { 3017 nfs3_cache_post_op_attr(svp, &res.resok.file_attributes, t, cr); 3018 nfs3_cache_wcc_data(tdvp, &res.resok.linkdir_wcc, t, cr); 3019 if (HAVE_RDDIR_CACHE(tdrp)) 3020 nfs_purge_rddir_cache(tdvp); 3021 dnlc_update(tdvp, tnm, svp); 3022 } else { 3023 nfs3_cache_post_op_attr(svp, &res.resfail.file_attributes, t, 3024 cr); 3025 nfs3_cache_wcc_data(tdvp, &res.resfail.linkdir_wcc, t, cr); 3026 if (error == EOPNOTSUPP) { 3027 mutex_enter(&mi->mi_lock); 3028 mi->mi_flags &= ~MI_LINK; 3029 mutex_exit(&mi->mi_lock); 3030 } 3031 } 3032 3033 nfs_rw_exit(&tdrp->r_rwlock); 3034 3035 if (!error) { 3036 /* 3037 * Notify the source file of this link operation. 3038 */ 3039 vnevent_link(svp, ct); 3040 } 3041 return (error); 3042 } 3043 3044 /* ARGSUSED */ 3045 static int 3046 nfs3_rename(vnode_t *odvp, char *onm, vnode_t *ndvp, char *nnm, cred_t *cr, 3047 caller_context_t *ct, int flags) 3048 { 3049 vnode_t *realvp; 3050 3051 if (nfs_zone() != VTOMI(odvp)->mi_zone) 3052 return (EPERM); 3053 if (VOP_REALVP(ndvp, &realvp, ct) == 0) 3054 ndvp = realvp; 3055 3056 return (nfs3rename(odvp, onm, ndvp, nnm, cr, ct)); 3057 } 3058 3059 /* 3060 * nfs3rename does the real work of renaming in NFS Version 3. 3061 */ 3062 static int 3063 nfs3rename(vnode_t *odvp, char *onm, vnode_t *ndvp, char *nnm, cred_t *cr, 3064 caller_context_t *ct) 3065 { 3066 int error; 3067 RENAME3args args; 3068 RENAME3res res; 3069 int douprintf; 3070 vnode_t *nvp = NULL; 3071 vnode_t *ovp = NULL; 3072 char *tmpname; 3073 rnode_t *rp; 3074 rnode_t *odrp; 3075 rnode_t *ndrp; 3076 hrtime_t t; 3077 3078 ASSERT(nfs_zone() == VTOMI(odvp)->mi_zone); 3079 3080 if (strcmp(onm, ".") == 0 || strcmp(onm, "..") == 0 || 3081 strcmp(nnm, ".") == 0 || strcmp(nnm, "..") == 0) 3082 return (EINVAL); 3083 3084 odrp = VTOR(odvp); 3085 ndrp = VTOR(ndvp); 3086 if ((intptr_t)odrp < (intptr_t)ndrp) { 3087 if (nfs_rw_enter_sig(&odrp->r_rwlock, RW_WRITER, INTR(odvp))) 3088 return (EINTR); 3089 if (nfs_rw_enter_sig(&ndrp->r_rwlock, RW_WRITER, INTR(ndvp))) { 3090 nfs_rw_exit(&odrp->r_rwlock); 3091 return (EINTR); 3092 } 3093 } else { 3094 if (nfs_rw_enter_sig(&ndrp->r_rwlock, RW_WRITER, INTR(ndvp))) 3095 return (EINTR); 3096 if (nfs_rw_enter_sig(&odrp->r_rwlock, RW_WRITER, INTR(odvp))) { 3097 nfs_rw_exit(&ndrp->r_rwlock); 3098 return (EINTR); 3099 } 3100 } 3101 3102 /* 3103 * Lookup the target file. If it exists, it needs to be 3104 * checked to see whether it is a mount point and whether 3105 * it is active (open). 3106 */ 3107 error = nfs3lookup(ndvp, nnm, &nvp, NULL, 0, NULL, cr, 0); 3108 if (!error) { 3109 /* 3110 * If this file has been mounted on, then just 3111 * return busy because renaming to it would remove 3112 * the mounted file system from the name space. 3113 */ 3114 if (vn_mountedvfs(nvp) != NULL) { 3115 VN_RELE(nvp); 3116 nfs_rw_exit(&odrp->r_rwlock); 3117 nfs_rw_exit(&ndrp->r_rwlock); 3118 return (EBUSY); 3119 } 3120 3121 /* 3122 * Purge the name cache of all references to this vnode 3123 * so that we can check the reference count to infer 3124 * whether it is active or not. 3125 */ 3126 /* 3127 * First just remove the entry from the name cache, as it 3128 * is most likely the only entry for this vp. 3129 */ 3130 dnlc_remove(ndvp, nnm); 3131 /* 3132 * If the file has a v_count > 1 then there may be more 3133 * than one entry in the name cache due multiple links 3134 * or an open file, but we don't have the real reference 3135 * count so flush all possible entries. 3136 */ 3137 if (nvp->v_count > 1) 3138 dnlc_purge_vp(nvp); 3139 3140 /* 3141 * If the vnode is active and is not a directory, 3142 * arrange to rename it to a 3143 * temporary file so that it will continue to be 3144 * accessible. This implements the "unlink-open-file" 3145 * semantics for the target of a rename operation. 3146 * Before doing this though, make sure that the 3147 * source and target files are not already the same. 3148 */ 3149 if (nvp->v_count > 1 && nvp->v_type != VDIR) { 3150 /* 3151 * Lookup the source name. 3152 */ 3153 error = nfs3lookup(odvp, onm, &ovp, NULL, 0, NULL, 3154 cr, 0); 3155 3156 /* 3157 * The source name *should* already exist. 3158 */ 3159 if (error) { 3160 VN_RELE(nvp); 3161 nfs_rw_exit(&odrp->r_rwlock); 3162 nfs_rw_exit(&ndrp->r_rwlock); 3163 return (error); 3164 } 3165 3166 /* 3167 * Compare the two vnodes. If they are the same, 3168 * just release all held vnodes and return success. 3169 */ 3170 if (ovp == nvp) { 3171 VN_RELE(ovp); 3172 VN_RELE(nvp); 3173 nfs_rw_exit(&odrp->r_rwlock); 3174 nfs_rw_exit(&ndrp->r_rwlock); 3175 return (0); 3176 } 3177 3178 /* 3179 * Can't mix and match directories and non- 3180 * directories in rename operations. We already 3181 * know that the target is not a directory. If 3182 * the source is a directory, return an error. 3183 */ 3184 if (ovp->v_type == VDIR) { 3185 VN_RELE(ovp); 3186 VN_RELE(nvp); 3187 nfs_rw_exit(&odrp->r_rwlock); 3188 nfs_rw_exit(&ndrp->r_rwlock); 3189 return (ENOTDIR); 3190 } 3191 3192 /* 3193 * The target file exists, is not the same as 3194 * the source file, and is active. Link it 3195 * to a temporary filename to avoid having 3196 * the server removing the file completely. 3197 */ 3198 tmpname = newname(); 3199 error = nfs3_link(ndvp, nvp, tmpname, cr, NULL, 0); 3200 if (error == EOPNOTSUPP) { 3201 error = nfs3_rename(ndvp, nnm, ndvp, tmpname, 3202 cr, NULL, 0); 3203 } 3204 if (error) { 3205 kmem_free(tmpname, MAXNAMELEN); 3206 VN_RELE(ovp); 3207 VN_RELE(nvp); 3208 nfs_rw_exit(&odrp->r_rwlock); 3209 nfs_rw_exit(&ndrp->r_rwlock); 3210 return (error); 3211 } 3212 rp = VTOR(nvp); 3213 mutex_enter(&rp->r_statelock); 3214 if (rp->r_unldvp == NULL) { 3215 VN_HOLD(ndvp); 3216 rp->r_unldvp = ndvp; 3217 if (rp->r_unlcred != NULL) 3218 crfree(rp->r_unlcred); 3219 crhold(cr); 3220 rp->r_unlcred = cr; 3221 rp->r_unlname = tmpname; 3222 } else { 3223 kmem_free(rp->r_unlname, MAXNAMELEN); 3224 rp->r_unlname = tmpname; 3225 } 3226 mutex_exit(&rp->r_statelock); 3227 } 3228 } 3229 3230 if (ovp == NULL) { 3231 /* 3232 * When renaming directories to be a subdirectory of a 3233 * different parent, the dnlc entry for ".." will no 3234 * longer be valid, so it must be removed. 3235 * 3236 * We do a lookup here to determine whether we are renaming 3237 * a directory and we need to check if we are renaming 3238 * an unlinked file. This might have already been done 3239 * in previous code, so we check ovp == NULL to avoid 3240 * doing it twice. 3241 */ 3242 3243 error = nfs3lookup(odvp, onm, &ovp, NULL, 0, NULL, cr, 0); 3244 /* 3245 * The source name *should* already exist. 3246 */ 3247 if (error) { 3248 nfs_rw_exit(&odrp->r_rwlock); 3249 nfs_rw_exit(&ndrp->r_rwlock); 3250 if (nvp) { 3251 VN_RELE(nvp); 3252 } 3253 return (error); 3254 } 3255 ASSERT(ovp != NULL); 3256 } 3257 3258 dnlc_remove(odvp, onm); 3259 dnlc_remove(ndvp, nnm); 3260 3261 setdiropargs3(&args.from, onm, odvp); 3262 setdiropargs3(&args.to, nnm, ndvp); 3263 3264 douprintf = 1; 3265 3266 t = gethrtime(); 3267 3268 error = rfs3call(VTOMI(odvp), NFSPROC3_RENAME, 3269 xdr_RENAME3args, (caddr_t)&args, 3270 xdr_RENAME3res, (caddr_t)&res, cr, 3271 &douprintf, &res.status, 0, NULL); 3272 3273 if (error) { 3274 PURGE_ATTRCACHE(odvp); 3275 PURGE_ATTRCACHE(ndvp); 3276 VN_RELE(ovp); 3277 nfs_rw_exit(&odrp->r_rwlock); 3278 nfs_rw_exit(&ndrp->r_rwlock); 3279 if (nvp) { 3280 VN_RELE(nvp); 3281 } 3282 return (error); 3283 } 3284 3285 error = geterrno3(res.status); 3286 3287 if (!error) { 3288 nfs3_cache_wcc_data(odvp, &res.resok.fromdir_wcc, t, cr); 3289 if (HAVE_RDDIR_CACHE(odrp)) 3290 nfs_purge_rddir_cache(odvp); 3291 if (ndvp != odvp) { 3292 nfs3_cache_wcc_data(ndvp, &res.resok.todir_wcc, t, cr); 3293 if (HAVE_RDDIR_CACHE(ndrp)) 3294 nfs_purge_rddir_cache(ndvp); 3295 } 3296 /* 3297 * when renaming directories to be a subdirectory of a 3298 * different parent, the dnlc entry for ".." will no 3299 * longer be valid, so it must be removed 3300 */ 3301 rp = VTOR(ovp); 3302 if (ndvp != odvp) { 3303 if (ovp->v_type == VDIR) { 3304 dnlc_remove(ovp, ".."); 3305 if (HAVE_RDDIR_CACHE(rp)) 3306 nfs_purge_rddir_cache(ovp); 3307 } 3308 } 3309 3310 /* 3311 * If we are renaming the unlinked file, update the 3312 * r_unldvp and r_unlname as needed. 3313 */ 3314 mutex_enter(&rp->r_statelock); 3315 if (rp->r_unldvp != NULL) { 3316 if (strcmp(rp->r_unlname, onm) == 0) { 3317 (void) strncpy(rp->r_unlname, nnm, MAXNAMELEN); 3318 rp->r_unlname[MAXNAMELEN - 1] = '\0'; 3319 3320 if (ndvp != rp->r_unldvp) { 3321 VN_RELE(rp->r_unldvp); 3322 rp->r_unldvp = ndvp; 3323 VN_HOLD(ndvp); 3324 } 3325 } 3326 } 3327 mutex_exit(&rp->r_statelock); 3328 } else { 3329 nfs3_cache_wcc_data(odvp, &res.resfail.fromdir_wcc, t, cr); 3330 if (ndvp != odvp) { 3331 nfs3_cache_wcc_data(ndvp, &res.resfail.todir_wcc, t, 3332 cr); 3333 } 3334 /* 3335 * System V defines rename to return EEXIST, not 3336 * ENOTEMPTY if the target directory is not empty. 3337 * Over the wire, the error is NFSERR_ENOTEMPTY 3338 * which geterrno maps to ENOTEMPTY. 3339 */ 3340 if (error == ENOTEMPTY) 3341 error = EEXIST; 3342 } 3343 3344 if (error == 0) { 3345 if (nvp) 3346 vnevent_rename_dest(nvp, ndvp, nnm, ct); 3347 3348 if (odvp != ndvp) 3349 vnevent_rename_dest_dir(ndvp, ct); 3350 ASSERT(ovp != NULL); 3351 vnevent_rename_src(ovp, odvp, onm, ct); 3352 } 3353 3354 if (nvp) { 3355 VN_RELE(nvp); 3356 } 3357 VN_RELE(ovp); 3358 3359 nfs_rw_exit(&odrp->r_rwlock); 3360 nfs_rw_exit(&ndrp->r_rwlock); 3361 3362 return (error); 3363 } 3364 3365 /* ARGSUSED */ 3366 static int 3367 nfs3_mkdir(vnode_t *dvp, char *nm, struct vattr *va, vnode_t **vpp, cred_t *cr, 3368 caller_context_t *ct, int flags, vsecattr_t *vsecp) 3369 { 3370 int error; 3371 MKDIR3args args; 3372 MKDIR3res res; 3373 int douprintf; 3374 struct vattr vattr; 3375 vnode_t *vp; 3376 rnode_t *drp; 3377 hrtime_t t; 3378 3379 if (nfs_zone() != VTOMI(dvp)->mi_zone) 3380 return (EPERM); 3381 setdiropargs3(&args.where, nm, dvp); 3382 3383 /* 3384 * Decide what the group-id and set-gid bit of the created directory 3385 * should be. May have to do a setattr to get the gid right. 3386 */ 3387 error = setdirgid(dvp, &va->va_gid, cr); 3388 if (error) 3389 return (error); 3390 error = setdirmode(dvp, &va->va_mode, cr); 3391 if (error) 3392 return (error); 3393 va->va_mask |= AT_MODE|AT_GID; 3394 3395 error = vattr_to_sattr3(va, &args.attributes); 3396 if (error) { 3397 /* req time field(s) overflow - return immediately */ 3398 return (error); 3399 } 3400 3401 drp = VTOR(dvp); 3402 if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp))) 3403 return (EINTR); 3404 3405 dnlc_remove(dvp, nm); 3406 3407 douprintf = 1; 3408 3409 t = gethrtime(); 3410 3411 error = rfs3call(VTOMI(dvp), NFSPROC3_MKDIR, 3412 xdr_MKDIR3args, (caddr_t)&args, 3413 xdr_MKDIR3res, (caddr_t)&res, cr, 3414 &douprintf, &res.status, 0, NULL); 3415 3416 if (error) { 3417 PURGE_ATTRCACHE(dvp); 3418 nfs_rw_exit(&drp->r_rwlock); 3419 return (error); 3420 } 3421 3422 error = geterrno3(res.status); 3423 if (!error) { 3424 nfs3_cache_wcc_data(dvp, &res.resok.dir_wcc, t, cr); 3425 if (HAVE_RDDIR_CACHE(drp)) 3426 nfs_purge_rddir_cache(dvp); 3427 3428 if (!res.resok.obj.handle_follows) { 3429 error = nfs3lookup(dvp, nm, &vp, NULL, 0, NULL, cr, 0); 3430 if (error) { 3431 nfs_rw_exit(&drp->r_rwlock); 3432 return (error); 3433 } 3434 } else { 3435 if (res.resok.obj_attributes.attributes) { 3436 vp = makenfs3node(&res.resok.obj.handle, 3437 &res.resok.obj_attributes.attr, 3438 dvp->v_vfsp, t, cr, NULL, NULL); 3439 } else { 3440 vp = makenfs3node(&res.resok.obj.handle, NULL, 3441 dvp->v_vfsp, t, cr, NULL, NULL); 3442 if (vp->v_type == VNON) { 3443 vattr.va_mask = AT_TYPE; 3444 error = nfs3getattr(vp, &vattr, cr); 3445 if (error) { 3446 VN_RELE(vp); 3447 nfs_rw_exit(&drp->r_rwlock); 3448 return (error); 3449 } 3450 vp->v_type = vattr.va_type; 3451 } 3452 } 3453 dnlc_update(dvp, nm, vp); 3454 } 3455 if (va->va_gid != VTOR(vp)->r_attr.va_gid) { 3456 va->va_mask = AT_GID; 3457 (void) nfs3setattr(vp, va, 0, cr); 3458 } 3459 *vpp = vp; 3460 } else { 3461 nfs3_cache_wcc_data(dvp, &res.resfail.dir_wcc, t, cr); 3462 PURGE_STALE_FH(error, dvp, cr); 3463 } 3464 3465 nfs_rw_exit(&drp->r_rwlock); 3466 3467 return (error); 3468 } 3469 3470 /* ARGSUSED */ 3471 static int 3472 nfs3_rmdir(vnode_t *dvp, char *nm, vnode_t *cdir, cred_t *cr, 3473 caller_context_t *ct, int flags) 3474 { 3475 int error; 3476 RMDIR3args args; 3477 RMDIR3res res; 3478 vnode_t *vp; 3479 int douprintf; 3480 rnode_t *drp; 3481 hrtime_t t; 3482 3483 if (nfs_zone() != VTOMI(dvp)->mi_zone) 3484 return (EPERM); 3485 drp = VTOR(dvp); 3486 if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp))) 3487 return (EINTR); 3488 3489 /* 3490 * Attempt to prevent a rmdir(".") from succeeding. 3491 */ 3492 error = nfs3lookup(dvp, nm, &vp, NULL, 0, NULL, cr, 0); 3493 if (error) { 3494 nfs_rw_exit(&drp->r_rwlock); 3495 return (error); 3496 } 3497 3498 if (vp == cdir) { 3499 VN_RELE(vp); 3500 nfs_rw_exit(&drp->r_rwlock); 3501 return (EINVAL); 3502 } 3503 3504 setdiropargs3(&args.object, nm, dvp); 3505 3506 /* 3507 * First just remove the entry from the name cache, as it 3508 * is most likely an entry for this vp. 3509 */ 3510 dnlc_remove(dvp, nm); 3511 3512 /* 3513 * If there vnode reference count is greater than one, then 3514 * there may be additional references in the DNLC which will 3515 * need to be purged. First, trying removing the entry for 3516 * the parent directory and see if that removes the additional 3517 * reference(s). If that doesn't do it, then use dnlc_purge_vp 3518 * to completely remove any references to the directory which 3519 * might still exist in the DNLC. 3520 */ 3521 if (vp->v_count > 1) { 3522 dnlc_remove(vp, ".."); 3523 if (vp->v_count > 1) 3524 dnlc_purge_vp(vp); 3525 } 3526 3527 douprintf = 1; 3528 3529 t = gethrtime(); 3530 3531 error = rfs3call(VTOMI(dvp), NFSPROC3_RMDIR, 3532 xdr_diropargs3, (caddr_t)&args, 3533 xdr_RMDIR3res, (caddr_t)&res, cr, 3534 &douprintf, &res.status, 0, NULL); 3535 3536 PURGE_ATTRCACHE(vp); 3537 3538 if (error) { 3539 PURGE_ATTRCACHE(dvp); 3540 VN_RELE(vp); 3541 nfs_rw_exit(&drp->r_rwlock); 3542 return (error); 3543 } 3544 3545 error = geterrno3(res.status); 3546 if (!error) { 3547 nfs3_cache_wcc_data(dvp, &res.resok.dir_wcc, t, cr); 3548 if (HAVE_RDDIR_CACHE(drp)) 3549 nfs_purge_rddir_cache(dvp); 3550 if (HAVE_RDDIR_CACHE(VTOR(vp))) 3551 nfs_purge_rddir_cache(vp); 3552 } else { 3553 nfs3_cache_wcc_data(dvp, &res.resfail.dir_wcc, t, cr); 3554 PURGE_STALE_FH(error, dvp, cr); 3555 /* 3556 * System V defines rmdir to return EEXIST, not 3557 * ENOTEMPTY if the directory is not empty. Over 3558 * the wire, the error is NFSERR_ENOTEMPTY which 3559 * geterrno maps to ENOTEMPTY. 3560 */ 3561 if (error == ENOTEMPTY) 3562 error = EEXIST; 3563 } 3564 3565 if (error == 0) { 3566 vnevent_rmdir(vp, dvp, nm, ct); 3567 } 3568 VN_RELE(vp); 3569 3570 nfs_rw_exit(&drp->r_rwlock); 3571 3572 return (error); 3573 } 3574 3575 /* ARGSUSED */ 3576 static int 3577 nfs3_symlink(vnode_t *dvp, char *lnm, struct vattr *tva, char *tnm, cred_t *cr, 3578 caller_context_t *ct, int flags) 3579 { 3580 int error; 3581 SYMLINK3args args; 3582 SYMLINK3res res; 3583 int douprintf; 3584 mntinfo_t *mi; 3585 vnode_t *vp; 3586 rnode_t *rp; 3587 char *contents; 3588 rnode_t *drp; 3589 hrtime_t t; 3590 3591 mi = VTOMI(dvp); 3592 3593 if (nfs_zone() != mi->mi_zone) 3594 return (EPERM); 3595 if (!(mi->mi_flags & MI_SYMLINK)) 3596 return (EOPNOTSUPP); 3597 3598 setdiropargs3(&args.where, lnm, dvp); 3599 error = vattr_to_sattr3(tva, &args.symlink.symlink_attributes); 3600 if (error) { 3601 /* req time field(s) overflow - return immediately */ 3602 return (error); 3603 } 3604 args.symlink.symlink_data = tnm; 3605 3606 drp = VTOR(dvp); 3607 if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp))) 3608 return (EINTR); 3609 3610 dnlc_remove(dvp, lnm); 3611 3612 douprintf = 1; 3613 3614 t = gethrtime(); 3615 3616 error = rfs3call(mi, NFSPROC3_SYMLINK, 3617 xdr_SYMLINK3args, (caddr_t)&args, 3618 xdr_SYMLINK3res, (caddr_t)&res, cr, 3619 &douprintf, &res.status, 0, NULL); 3620 3621 if (error) { 3622 PURGE_ATTRCACHE(dvp); 3623 nfs_rw_exit(&drp->r_rwlock); 3624 return (error); 3625 } 3626 3627 error = geterrno3(res.status); 3628 if (!error) { 3629 nfs3_cache_wcc_data(dvp, &res.resok.dir_wcc, t, cr); 3630 if (HAVE_RDDIR_CACHE(drp)) 3631 nfs_purge_rddir_cache(dvp); 3632 3633 if (res.resok.obj.handle_follows) { 3634 if (res.resok.obj_attributes.attributes) { 3635 vp = makenfs3node(&res.resok.obj.handle, 3636 &res.resok.obj_attributes.attr, 3637 dvp->v_vfsp, t, cr, NULL, NULL); 3638 } else { 3639 vp = makenfs3node(&res.resok.obj.handle, NULL, 3640 dvp->v_vfsp, t, cr, NULL, NULL); 3641 vp->v_type = VLNK; 3642 vp->v_rdev = 0; 3643 } 3644 dnlc_update(dvp, lnm, vp); 3645 rp = VTOR(vp); 3646 if (nfs3_do_symlink_cache && 3647 rp->r_symlink.contents == NULL) { 3648 3649 contents = kmem_alloc(MAXPATHLEN, 3650 KM_NOSLEEP); 3651 3652 if (contents != NULL) { 3653 mutex_enter(&rp->r_statelock); 3654 if (rp->r_symlink.contents == NULL) { 3655 rp->r_symlink.len = strlen(tnm); 3656 bcopy(tnm, contents, 3657 rp->r_symlink.len); 3658 rp->r_symlink.contents = 3659 contents; 3660 rp->r_symlink.size = MAXPATHLEN; 3661 mutex_exit(&rp->r_statelock); 3662 } else { 3663 mutex_exit(&rp->r_statelock); 3664 kmem_free((void *)contents, 3665 MAXPATHLEN); 3666 } 3667 } 3668 } 3669 VN_RELE(vp); 3670 } 3671 } else { 3672 nfs3_cache_wcc_data(dvp, &res.resfail.dir_wcc, t, cr); 3673 PURGE_STALE_FH(error, dvp, cr); 3674 if (error == EOPNOTSUPP) { 3675 mutex_enter(&mi->mi_lock); 3676 mi->mi_flags &= ~MI_SYMLINK; 3677 mutex_exit(&mi->mi_lock); 3678 } 3679 } 3680 3681 nfs_rw_exit(&drp->r_rwlock); 3682 3683 return (error); 3684 } 3685 3686 #ifdef DEBUG 3687 static int nfs3_readdir_cache_hits = 0; 3688 static int nfs3_readdir_cache_shorts = 0; 3689 static int nfs3_readdir_cache_waits = 0; 3690 static int nfs3_readdir_cache_misses = 0; 3691 static int nfs3_readdir_readahead = 0; 3692 #endif 3693 3694 static int nfs3_shrinkreaddir = 0; 3695 3696 /* 3697 * Read directory entries. 3698 * There are some weird things to look out for here. The uio_loffset 3699 * field is either 0 or it is the offset returned from a previous 3700 * readdir. It is an opaque value used by the server to find the 3701 * correct directory block to read. The count field is the number 3702 * of blocks to read on the server. This is advisory only, the server 3703 * may return only one block's worth of entries. Entries may be compressed 3704 * on the server. 3705 */ 3706 /* ARGSUSED */ 3707 static int 3708 nfs3_readdir(vnode_t *vp, struct uio *uiop, cred_t *cr, int *eofp, 3709 caller_context_t *ct, int flags) 3710 { 3711 int error; 3712 size_t count; 3713 rnode_t *rp; 3714 rddir_cache *rdc; 3715 rddir_cache *nrdc; 3716 rddir_cache *rrdc; 3717 #ifdef DEBUG 3718 int missed; 3719 #endif 3720 int doreadahead; 3721 rddir_cache srdc; 3722 avl_index_t where; 3723 3724 if (nfs_zone() != VTOMI(vp)->mi_zone) 3725 return (EIO); 3726 rp = VTOR(vp); 3727 3728 ASSERT(nfs_rw_lock_held(&rp->r_rwlock, RW_READER)); 3729 3730 /* 3731 * Make sure that the directory cache is valid. 3732 */ 3733 if (HAVE_RDDIR_CACHE(rp)) { 3734 if (nfs_disable_rddir_cache) { 3735 /* 3736 * Setting nfs_disable_rddir_cache in /etc/system 3737 * allows interoperability with servers that do not 3738 * properly update the attributes of directories. 3739 * Any cached information gets purged before an 3740 * access is made to it. 3741 */ 3742 nfs_purge_rddir_cache(vp); 3743 } else { 3744 error = nfs3_validate_caches(vp, cr); 3745 if (error) 3746 return (error); 3747 } 3748 } 3749 3750 /* 3751 * It is possible that some servers may not be able to correctly 3752 * handle a large READDIR or READDIRPLUS request due to bugs in 3753 * their implementation. In order to continue to interoperate 3754 * with them, this workaround is provided to limit the maximum 3755 * size of a READDIRPLUS request to 1024. In any case, the request 3756 * size is limited to MAXBSIZE. 3757 */ 3758 count = MIN(uiop->uio_iov->iov_len, 3759 nfs3_shrinkreaddir ? 1024 : MAXBSIZE); 3760 3761 nrdc = NULL; 3762 #ifdef DEBUG 3763 missed = 0; 3764 #endif 3765 top: 3766 /* 3767 * Short circuit last readdir which always returns 0 bytes. 3768 * This can be done after the directory has been read through 3769 * completely at least once. This will set r_direof which 3770 * can be used to find the value of the last cookie. 3771 */ 3772 mutex_enter(&rp->r_statelock); 3773 if (rp->r_direof != NULL && 3774 uiop->uio_loffset == rp->r_direof->nfs3_ncookie) { 3775 mutex_exit(&rp->r_statelock); 3776 #ifdef DEBUG 3777 nfs3_readdir_cache_shorts++; 3778 #endif 3779 if (eofp) 3780 *eofp = 1; 3781 if (nrdc != NULL) 3782 rddir_cache_rele(nrdc); 3783 return (0); 3784 } 3785 /* 3786 * Look for a cache entry. Cache entries are identified 3787 * by the NFS cookie value and the byte count requested. 3788 */ 3789 srdc.nfs3_cookie = uiop->uio_loffset; 3790 srdc.buflen = count; 3791 rdc = avl_find(&rp->r_dir, &srdc, &where); 3792 if (rdc != NULL) { 3793 rddir_cache_hold(rdc); 3794 /* 3795 * If the cache entry is in the process of being 3796 * filled in, wait until this completes. The 3797 * RDDIRWAIT bit is set to indicate that someone 3798 * is waiting and then the thread currently 3799 * filling the entry is done, it should do a 3800 * cv_broadcast to wakeup all of the threads 3801 * waiting for it to finish. 3802 */ 3803 if (rdc->flags & RDDIR) { 3804 nfs_rw_exit(&rp->r_rwlock); 3805 rdc->flags |= RDDIRWAIT; 3806 #ifdef DEBUG 3807 nfs3_readdir_cache_waits++; 3808 #endif 3809 if (!cv_wait_sig(&rdc->cv, &rp->r_statelock)) { 3810 /* 3811 * We got interrupted, probably 3812 * the user typed ^C or an alarm 3813 * fired. We free the new entry 3814 * if we allocated one. 3815 */ 3816 mutex_exit(&rp->r_statelock); 3817 (void) nfs_rw_enter_sig(&rp->r_rwlock, 3818 RW_READER, FALSE); 3819 rddir_cache_rele(rdc); 3820 if (nrdc != NULL) 3821 rddir_cache_rele(nrdc); 3822 return (EINTR); 3823 } 3824 mutex_exit(&rp->r_statelock); 3825 (void) nfs_rw_enter_sig(&rp->r_rwlock, 3826 RW_READER, FALSE); 3827 rddir_cache_rele(rdc); 3828 goto top; 3829 } 3830 /* 3831 * Check to see if a readdir is required to 3832 * fill the entry. If so, mark this entry 3833 * as being filled, remove our reference, 3834 * and branch to the code to fill the entry. 3835 */ 3836 if (rdc->flags & RDDIRREQ) { 3837 rdc->flags &= ~RDDIRREQ; 3838 rdc->flags |= RDDIR; 3839 if (nrdc != NULL) 3840 rddir_cache_rele(nrdc); 3841 nrdc = rdc; 3842 mutex_exit(&rp->r_statelock); 3843 goto bottom; 3844 } 3845 #ifdef DEBUG 3846 if (!missed) 3847 nfs3_readdir_cache_hits++; 3848 #endif 3849 /* 3850 * If an error occurred while attempting 3851 * to fill the cache entry, just return it. 3852 */ 3853 if (rdc->error) { 3854 error = rdc->error; 3855 mutex_exit(&rp->r_statelock); 3856 rddir_cache_rele(rdc); 3857 if (nrdc != NULL) 3858 rddir_cache_rele(nrdc); 3859 return (error); 3860 } 3861 3862 /* 3863 * The cache entry is complete and good, 3864 * copyout the dirent structs to the calling 3865 * thread. 3866 */ 3867 error = uiomove(rdc->entries, rdc->entlen, UIO_READ, uiop); 3868 3869 /* 3870 * If no error occurred during the copyout, 3871 * update the offset in the uio struct to 3872 * contain the value of the next cookie 3873 * and set the eof value appropriately. 3874 */ 3875 if (!error) { 3876 uiop->uio_loffset = rdc->nfs3_ncookie; 3877 if (eofp) 3878 *eofp = rdc->eof; 3879 } 3880 3881 /* 3882 * Decide whether to do readahead. 3883 * 3884 * Don't if have already read to the end of 3885 * directory. There is nothing more to read. 3886 * 3887 * Don't if the application is not doing 3888 * lookups in the directory. The readahead 3889 * is only effective if the application can 3890 * be doing work while an async thread is 3891 * handling the over the wire request. 3892 */ 3893 if (rdc->eof) { 3894 rp->r_direof = rdc; 3895 doreadahead = FALSE; 3896 } else if (!(rp->r_flags & RLOOKUP)) 3897 doreadahead = FALSE; 3898 else 3899 doreadahead = TRUE; 3900 3901 if (!doreadahead) { 3902 mutex_exit(&rp->r_statelock); 3903 rddir_cache_rele(rdc); 3904 if (nrdc != NULL) 3905 rddir_cache_rele(nrdc); 3906 return (error); 3907 } 3908 3909 /* 3910 * Check to see whether we found an entry 3911 * for the readahead. If so, we don't need 3912 * to do anything further, so free the new 3913 * entry if one was allocated. Otherwise, 3914 * allocate a new entry, add it to the cache, 3915 * and then initiate an asynchronous readdir 3916 * operation to fill it. 3917 */ 3918 srdc.nfs3_cookie = rdc->nfs3_ncookie; 3919 srdc.buflen = count; 3920 rrdc = avl_find(&rp->r_dir, &srdc, &where); 3921 if (rrdc != NULL) { 3922 if (nrdc != NULL) 3923 rddir_cache_rele(nrdc); 3924 } else { 3925 if (nrdc != NULL) 3926 rrdc = nrdc; 3927 else { 3928 rrdc = rddir_cache_alloc(KM_NOSLEEP); 3929 } 3930 if (rrdc != NULL) { 3931 rrdc->nfs3_cookie = rdc->nfs3_ncookie; 3932 rrdc->buflen = count; 3933 avl_insert(&rp->r_dir, rrdc, where); 3934 rddir_cache_hold(rrdc); 3935 mutex_exit(&rp->r_statelock); 3936 rddir_cache_rele(rdc); 3937 #ifdef DEBUG 3938 nfs3_readdir_readahead++; 3939 #endif 3940 nfs_async_readdir(vp, rrdc, cr, do_nfs3readdir); 3941 return (error); 3942 } 3943 } 3944 3945 mutex_exit(&rp->r_statelock); 3946 rddir_cache_rele(rdc); 3947 return (error); 3948 } 3949 3950 /* 3951 * Didn't find an entry in the cache. Construct a new empty 3952 * entry and link it into the cache. Other processes attempting 3953 * to access this entry will need to wait until it is filled in. 3954 * 3955 * Since kmem_alloc may block, another pass through the cache 3956 * will need to be taken to make sure that another process 3957 * hasn't already added an entry to the cache for this request. 3958 */ 3959 if (nrdc == NULL) { 3960 mutex_exit(&rp->r_statelock); 3961 nrdc = rddir_cache_alloc(KM_SLEEP); 3962 nrdc->nfs3_cookie = uiop->uio_loffset; 3963 nrdc->buflen = count; 3964 goto top; 3965 } 3966 3967 /* 3968 * Add this entry to the cache. 3969 */ 3970 avl_insert(&rp->r_dir, nrdc, where); 3971 rddir_cache_hold(nrdc); 3972 mutex_exit(&rp->r_statelock); 3973 3974 bottom: 3975 #ifdef DEBUG 3976 missed = 1; 3977 nfs3_readdir_cache_misses++; 3978 #endif 3979 /* 3980 * Do the readdir. This routine decides whether to use 3981 * READDIR or READDIRPLUS. 3982 */ 3983 error = do_nfs3readdir(vp, nrdc, cr); 3984 3985 /* 3986 * If this operation failed, just return the error which occurred. 3987 */ 3988 if (error != 0) 3989 return (error); 3990 3991 /* 3992 * Since the RPC operation will have taken sometime and blocked 3993 * this process, another pass through the cache will need to be 3994 * taken to find the correct cache entry. It is possible that 3995 * the correct cache entry will not be there (although one was 3996 * added) because the directory changed during the RPC operation 3997 * and the readdir cache was flushed. In this case, just start 3998 * over. It is hoped that this will not happen too often... :-) 3999 */ 4000 nrdc = NULL; 4001 goto top; 4002 /* NOTREACHED */ 4003 } 4004 4005 static int 4006 do_nfs3readdir(vnode_t *vp, rddir_cache *rdc, cred_t *cr) 4007 { 4008 int error; 4009 rnode_t *rp; 4010 mntinfo_t *mi; 4011 4012 rp = VTOR(vp); 4013 mi = VTOMI(vp); 4014 ASSERT(nfs_zone() == mi->mi_zone); 4015 /* 4016 * Issue the proper request. 4017 * 4018 * If the server does not support READDIRPLUS, then use READDIR. 4019 * 4020 * Otherwise -- 4021 * Issue a READDIRPLUS if reading to fill an empty cache or if 4022 * an application has performed a lookup in the directory which 4023 * required an over the wire lookup. The use of READDIRPLUS 4024 * will help to (re)populate the DNLC. 4025 */ 4026 if (!(mi->mi_flags & MI_READDIRONLY) && 4027 (rp->r_flags & (RLOOKUP | RREADDIRPLUS))) { 4028 if (rp->r_flags & RREADDIRPLUS) { 4029 mutex_enter(&rp->r_statelock); 4030 rp->r_flags &= ~RREADDIRPLUS; 4031 mutex_exit(&rp->r_statelock); 4032 } 4033 nfs3readdirplus(vp, rdc, cr); 4034 if (rdc->error == EOPNOTSUPP) 4035 nfs3readdir(vp, rdc, cr); 4036 } else 4037 nfs3readdir(vp, rdc, cr); 4038 4039 mutex_enter(&rp->r_statelock); 4040 rdc->flags &= ~RDDIR; 4041 if (rdc->flags & RDDIRWAIT) { 4042 rdc->flags &= ~RDDIRWAIT; 4043 cv_broadcast(&rdc->cv); 4044 } 4045 error = rdc->error; 4046 if (error) 4047 rdc->flags |= RDDIRREQ; 4048 mutex_exit(&rp->r_statelock); 4049 4050 rddir_cache_rele(rdc); 4051 4052 return (error); 4053 } 4054 4055 static void 4056 nfs3readdir(vnode_t *vp, rddir_cache *rdc, cred_t *cr) 4057 { 4058 int error; 4059 READDIR3args args; 4060 READDIR3vres res; 4061 vattr_t dva; 4062 rnode_t *rp; 4063 int douprintf; 4064 failinfo_t fi, *fip = NULL; 4065 mntinfo_t *mi; 4066 hrtime_t t; 4067 4068 rp = VTOR(vp); 4069 mi = VTOMI(vp); 4070 ASSERT(nfs_zone() == mi->mi_zone); 4071 4072 args.dir = *RTOFH3(rp); 4073 args.cookie = (cookie3)rdc->nfs3_cookie; 4074 args.cookieverf = rp->r_cookieverf; 4075 args.count = rdc->buflen; 4076 4077 /* 4078 * NFS client failover support 4079 * suppress failover unless we have a zero cookie 4080 */ 4081 if (args.cookie == (cookie3) 0) { 4082 fi.vp = vp; 4083 fi.fhp = (caddr_t)&args.dir; 4084 fi.copyproc = nfs3copyfh; 4085 fi.lookupproc = nfs3lookup; 4086 fi.xattrdirproc = acl_getxattrdir3; 4087 fip = &fi; 4088 } 4089 4090 #ifdef DEBUG 4091 rdc->entries = rddir_cache_buf_alloc(rdc->buflen, KM_SLEEP); 4092 #else 4093 rdc->entries = kmem_alloc(rdc->buflen, KM_SLEEP); 4094 #endif 4095 4096 res.entries = (dirent64_t *)rdc->entries; 4097 res.entries_size = rdc->buflen; 4098 res.dir_attributes.fres.vap = &dva; 4099 res.dir_attributes.fres.vp = vp; 4100 res.loff = rdc->nfs3_cookie; 4101 4102 douprintf = 1; 4103 4104 if (mi->mi_io_kstats) { 4105 mutex_enter(&mi->mi_lock); 4106 kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats)); 4107 mutex_exit(&mi->mi_lock); 4108 } 4109 4110 t = gethrtime(); 4111 4112 error = rfs3call(VTOMI(vp), NFSPROC3_READDIR, 4113 xdr_READDIR3args, (caddr_t)&args, 4114 xdr_READDIR3vres, (caddr_t)&res, cr, 4115 &douprintf, &res.status, 0, fip); 4116 4117 if (mi->mi_io_kstats) { 4118 mutex_enter(&mi->mi_lock); 4119 kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats)); 4120 mutex_exit(&mi->mi_lock); 4121 } 4122 4123 if (error) 4124 goto err; 4125 4126 nfs3_cache_post_op_vattr(vp, &res.dir_attributes, t, cr); 4127 4128 error = geterrno3(res.status); 4129 if (error) { 4130 PURGE_STALE_FH(error, vp, cr); 4131 goto err; 4132 } 4133 4134 if (mi->mi_io_kstats) { 4135 mutex_enter(&mi->mi_lock); 4136 KSTAT_IO_PTR(mi->mi_io_kstats)->reads++; 4137 KSTAT_IO_PTR(mi->mi_io_kstats)->nread += res.size; 4138 mutex_exit(&mi->mi_lock); 4139 } 4140 4141 rdc->nfs3_ncookie = res.loff; 4142 rp->r_cookieverf = res.cookieverf; 4143 rdc->eof = res.eof ? 1 : 0; 4144 rdc->entlen = res.size; 4145 ASSERT(rdc->entlen <= rdc->buflen); 4146 rdc->error = 0; 4147 return; 4148 4149 err: 4150 kmem_free(rdc->entries, rdc->buflen); 4151 rdc->entries = NULL; 4152 rdc->error = error; 4153 } 4154 4155 /* 4156 * Read directory entries. 4157 * There are some weird things to look out for here. The uio_loffset 4158 * field is either 0 or it is the offset returned from a previous 4159 * readdir. It is an opaque value used by the server to find the 4160 * correct directory block to read. The count field is the number 4161 * of blocks to read on the server. This is advisory only, the server 4162 * may return only one block's worth of entries. Entries may be compressed 4163 * on the server. 4164 */ 4165 static void 4166 nfs3readdirplus(vnode_t *vp, rddir_cache *rdc, cred_t *cr) 4167 { 4168 int error; 4169 READDIRPLUS3args args; 4170 READDIRPLUS3vres res; 4171 vattr_t dva; 4172 rnode_t *rp; 4173 mntinfo_t *mi; 4174 int douprintf; 4175 failinfo_t fi, *fip = NULL; 4176 4177 rp = VTOR(vp); 4178 mi = VTOMI(vp); 4179 ASSERT(nfs_zone() == mi->mi_zone); 4180 4181 args.dir = *RTOFH3(rp); 4182 args.cookie = (cookie3)rdc->nfs3_cookie; 4183 args.cookieverf = rp->r_cookieverf; 4184 args.dircount = rdc->buflen; 4185 args.maxcount = mi->mi_tsize; 4186 4187 /* 4188 * NFS client failover support 4189 * suppress failover unless we have a zero cookie 4190 */ 4191 if (args.cookie == (cookie3)0) { 4192 fi.vp = vp; 4193 fi.fhp = (caddr_t)&args.dir; 4194 fi.copyproc = nfs3copyfh; 4195 fi.lookupproc = nfs3lookup; 4196 fi.xattrdirproc = acl_getxattrdir3; 4197 fip = &fi; 4198 } 4199 4200 #ifdef DEBUG 4201 rdc->entries = rddir_cache_buf_alloc(rdc->buflen, KM_SLEEP); 4202 #else 4203 rdc->entries = kmem_alloc(rdc->buflen, KM_SLEEP); 4204 #endif 4205 4206 res.entries = (dirent64_t *)rdc->entries; 4207 res.entries_size = rdc->buflen; 4208 res.dir_attributes.fres.vap = &dva; 4209 res.dir_attributes.fres.vp = vp; 4210 res.loff = rdc->nfs3_cookie; 4211 res.credentials = cr; 4212 4213 douprintf = 1; 4214 4215 if (mi->mi_io_kstats) { 4216 mutex_enter(&mi->mi_lock); 4217 kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats)); 4218 mutex_exit(&mi->mi_lock); 4219 } 4220 4221 res.time = gethrtime(); 4222 4223 error = rfs3call(mi, NFSPROC3_READDIRPLUS, 4224 xdr_READDIRPLUS3args, (caddr_t)&args, 4225 xdr_READDIRPLUS3vres, (caddr_t)&res, cr, 4226 &douprintf, &res.status, 0, fip); 4227 4228 if (mi->mi_io_kstats) { 4229 mutex_enter(&mi->mi_lock); 4230 kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats)); 4231 mutex_exit(&mi->mi_lock); 4232 } 4233 4234 if (error) { 4235 goto err; 4236 } 4237 4238 nfs3_cache_post_op_vattr(vp, &res.dir_attributes, res.time, cr); 4239 4240 error = geterrno3(res.status); 4241 if (error) { 4242 PURGE_STALE_FH(error, vp, cr); 4243 if (error == EOPNOTSUPP) { 4244 mutex_enter(&mi->mi_lock); 4245 mi->mi_flags |= MI_READDIRONLY; 4246 mutex_exit(&mi->mi_lock); 4247 } 4248 goto err; 4249 } 4250 4251 if (mi->mi_io_kstats) { 4252 mutex_enter(&mi->mi_lock); 4253 KSTAT_IO_PTR(mi->mi_io_kstats)->reads++; 4254 KSTAT_IO_PTR(mi->mi_io_kstats)->nread += res.size; 4255 mutex_exit(&mi->mi_lock); 4256 } 4257 4258 rdc->nfs3_ncookie = res.loff; 4259 rp->r_cookieverf = res.cookieverf; 4260 rdc->eof = res.eof ? 1 : 0; 4261 rdc->entlen = res.size; 4262 ASSERT(rdc->entlen <= rdc->buflen); 4263 rdc->error = 0; 4264 4265 return; 4266 4267 err: 4268 kmem_free(rdc->entries, rdc->buflen); 4269 rdc->entries = NULL; 4270 rdc->error = error; 4271 } 4272 4273 #ifdef DEBUG 4274 static int nfs3_bio_do_stop = 0; 4275 #endif 4276 4277 static int 4278 nfs3_bio(struct buf *bp, stable_how *stab_comm, cred_t *cr) 4279 { 4280 rnode_t *rp = VTOR(bp->b_vp); 4281 int count; 4282 int error; 4283 cred_t *cred; 4284 offset_t offset; 4285 4286 ASSERT(nfs_zone() == VTOMI(bp->b_vp)->mi_zone); 4287 offset = ldbtob(bp->b_lblkno); 4288 4289 DTRACE_IO1(start, struct buf *, bp); 4290 4291 if (bp->b_flags & B_READ) { 4292 mutex_enter(&rp->r_statelock); 4293 if (rp->r_cred != NULL) { 4294 cred = rp->r_cred; 4295 crhold(cred); 4296 } else { 4297 rp->r_cred = cr; 4298 crhold(cr); 4299 cred = cr; 4300 crhold(cred); 4301 } 4302 mutex_exit(&rp->r_statelock); 4303 read_again: 4304 error = bp->b_error = nfs3read(bp->b_vp, bp->b_un.b_addr, 4305 offset, bp->b_bcount, &bp->b_resid, cred); 4306 crfree(cred); 4307 if (!error) { 4308 if (bp->b_resid) { 4309 /* 4310 * Didn't get it all because we hit EOF, 4311 * zero all the memory beyond the EOF. 4312 */ 4313 /* bzero(rdaddr + */ 4314 bzero(bp->b_un.b_addr + 4315 bp->b_bcount - bp->b_resid, bp->b_resid); 4316 } 4317 mutex_enter(&rp->r_statelock); 4318 if (bp->b_resid == bp->b_bcount && 4319 offset >= rp->r_size) { 4320 /* 4321 * We didn't read anything at all as we are 4322 * past EOF. Return an error indicator back 4323 * but don't destroy the pages (yet). 4324 */ 4325 error = NFS_EOF; 4326 } 4327 mutex_exit(&rp->r_statelock); 4328 } else if (error == EACCES) { 4329 mutex_enter(&rp->r_statelock); 4330 if (cred != cr) { 4331 if (rp->r_cred != NULL) 4332 crfree(rp->r_cred); 4333 rp->r_cred = cr; 4334 crhold(cr); 4335 cred = cr; 4336 crhold(cred); 4337 mutex_exit(&rp->r_statelock); 4338 goto read_again; 4339 } 4340 mutex_exit(&rp->r_statelock); 4341 } 4342 } else { 4343 if (!(rp->r_flags & RSTALE)) { 4344 mutex_enter(&rp->r_statelock); 4345 if (rp->r_cred != NULL) { 4346 cred = rp->r_cred; 4347 crhold(cred); 4348 } else { 4349 rp->r_cred = cr; 4350 crhold(cr); 4351 cred = cr; 4352 crhold(cred); 4353 } 4354 mutex_exit(&rp->r_statelock); 4355 write_again: 4356 mutex_enter(&rp->r_statelock); 4357 count = MIN(bp->b_bcount, rp->r_size - offset); 4358 mutex_exit(&rp->r_statelock); 4359 if (count < 0) 4360 cmn_err(CE_PANIC, "nfs3_bio: write count < 0"); 4361 #ifdef DEBUG 4362 if (count == 0) { 4363 zcmn_err(getzoneid(), CE_WARN, 4364 "nfs3_bio: zero length write at %lld", 4365 offset); 4366 nfs_printfhandle(&rp->r_fh); 4367 if (nfs3_bio_do_stop) 4368 debug_enter("nfs3_bio"); 4369 } 4370 #endif 4371 error = nfs3write(bp->b_vp, bp->b_un.b_addr, offset, 4372 count, cred, stab_comm); 4373 if (error == EACCES) { 4374 mutex_enter(&rp->r_statelock); 4375 if (cred != cr) { 4376 if (rp->r_cred != NULL) 4377 crfree(rp->r_cred); 4378 rp->r_cred = cr; 4379 crhold(cr); 4380 crfree(cred); 4381 cred = cr; 4382 crhold(cred); 4383 mutex_exit(&rp->r_statelock); 4384 goto write_again; 4385 } 4386 mutex_exit(&rp->r_statelock); 4387 } 4388 bp->b_error = error; 4389 if (error && error != EINTR) { 4390 /* 4391 * Don't print EDQUOT errors on the console. 4392 * Don't print asynchronous EACCES errors. 4393 * Don't print EFBIG errors. 4394 * Print all other write errors. 4395 */ 4396 if (error != EDQUOT && error != EFBIG && 4397 (error != EACCES || 4398 !(bp->b_flags & B_ASYNC))) 4399 nfs_write_error(bp->b_vp, error, cred); 4400 /* 4401 * Update r_error and r_flags as appropriate. 4402 * If the error was ESTALE, then mark the 4403 * rnode as not being writeable and save 4404 * the error status. Otherwise, save any 4405 * errors which occur from asynchronous 4406 * page invalidations. Any errors occurring 4407 * from other operations should be saved 4408 * by the caller. 4409 */ 4410 mutex_enter(&rp->r_statelock); 4411 if (error == ESTALE) { 4412 rp->r_flags |= RSTALE; 4413 if (!rp->r_error) 4414 rp->r_error = error; 4415 } else if (!rp->r_error && 4416 (bp->b_flags & 4417 (B_INVAL|B_FORCE|B_ASYNC)) == 4418 (B_INVAL|B_FORCE|B_ASYNC)) { 4419 rp->r_error = error; 4420 } 4421 mutex_exit(&rp->r_statelock); 4422 } 4423 crfree(cred); 4424 } else { 4425 error = rp->r_error; 4426 /* 4427 * A close may have cleared r_error, if so, 4428 * propagate ESTALE error return properly 4429 */ 4430 if (error == 0) 4431 error = ESTALE; 4432 } 4433 } 4434 4435 if (error != 0 && error != NFS_EOF) 4436 bp->b_flags |= B_ERROR; 4437 4438 DTRACE_IO1(done, struct buf *, bp); 4439 4440 return (error); 4441 } 4442 4443 /* ARGSUSED */ 4444 static int 4445 nfs3_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) 4446 { 4447 rnode_t *rp; 4448 4449 if (nfs_zone() != VTOMI(vp)->mi_zone) 4450 return (EIO); 4451 rp = VTOR(vp); 4452 4453 if (fidp->fid_len < (ushort_t)rp->r_fh.fh_len) { 4454 fidp->fid_len = rp->r_fh.fh_len; 4455 return (ENOSPC); 4456 } 4457 fidp->fid_len = rp->r_fh.fh_len; 4458 bcopy(rp->r_fh.fh_buf, fidp->fid_data, fidp->fid_len); 4459 return (0); 4460 } 4461 4462 /* ARGSUSED2 */ 4463 static int 4464 nfs3_rwlock(vnode_t *vp, int write_lock, caller_context_t *ctp) 4465 { 4466 rnode_t *rp = VTOR(vp); 4467 4468 if (!write_lock) { 4469 (void) nfs_rw_enter_sig(&rp->r_rwlock, RW_READER, FALSE); 4470 return (V_WRITELOCK_FALSE); 4471 } 4472 4473 if ((rp->r_flags & RDIRECTIO) || (VTOMI(vp)->mi_flags & MI_DIRECTIO)) { 4474 (void) nfs_rw_enter_sig(&rp->r_rwlock, RW_READER, FALSE); 4475 if (rp->r_mapcnt == 0 && !vn_has_cached_data(vp)) 4476 return (V_WRITELOCK_FALSE); 4477 nfs_rw_exit(&rp->r_rwlock); 4478 } 4479 4480 (void) nfs_rw_enter_sig(&rp->r_rwlock, RW_WRITER, FALSE); 4481 return (V_WRITELOCK_TRUE); 4482 } 4483 4484 /* ARGSUSED */ 4485 static void 4486 nfs3_rwunlock(vnode_t *vp, int write_lock, caller_context_t *ctp) 4487 { 4488 rnode_t *rp = VTOR(vp); 4489 4490 nfs_rw_exit(&rp->r_rwlock); 4491 } 4492 4493 /* ARGSUSED */ 4494 static int 4495 nfs3_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, caller_context_t *ct) 4496 { 4497 4498 /* 4499 * Because we stuff the readdir cookie into the offset field 4500 * someone may attempt to do an lseek with the cookie which 4501 * we want to succeed. 4502 */ 4503 if (vp->v_type == VDIR) 4504 return (0); 4505 if (*noffp < 0) 4506 return (EINVAL); 4507 return (0); 4508 } 4509 4510 /* 4511 * number of nfs3_bsize blocks to read ahead. 4512 */ 4513 static int nfs3_nra = 4; 4514 4515 #ifdef DEBUG 4516 static int nfs3_lostpage = 0; /* number of times we lost original page */ 4517 #endif 4518 4519 /* 4520 * Return all the pages from [off..off+len) in file 4521 */ 4522 /* ARGSUSED */ 4523 static int 4524 nfs3_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 4525 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 4526 enum seg_rw rw, cred_t *cr, caller_context_t *ct) 4527 { 4528 rnode_t *rp; 4529 int error; 4530 mntinfo_t *mi; 4531 4532 if (vp->v_flag & VNOMAP) 4533 return (ENOSYS); 4534 4535 if (nfs_zone() != VTOMI(vp)->mi_zone) 4536 return (EIO); 4537 if (protp != NULL) 4538 *protp = PROT_ALL; 4539 4540 /* 4541 * Now valididate that the caches are up to date. 4542 */ 4543 error = nfs3_validate_caches(vp, cr); 4544 if (error) 4545 return (error); 4546 4547 rp = VTOR(vp); 4548 mi = VTOMI(vp); 4549 retry: 4550 mutex_enter(&rp->r_statelock); 4551 4552 /* 4553 * Don't create dirty pages faster than they 4554 * can be cleaned so that the system doesn't 4555 * get imbalanced. If the async queue is 4556 * maxed out, then wait for it to drain before 4557 * creating more dirty pages. Also, wait for 4558 * any threads doing pagewalks in the vop_getattr 4559 * entry points so that they don't block for 4560 * long periods. 4561 */ 4562 if (rw == S_CREATE) { 4563 while ((mi->mi_max_threads != 0 && 4564 rp->r_awcount > 2 * mi->mi_max_threads) || 4565 rp->r_gcount > 0) 4566 cv_wait(&rp->r_cv, &rp->r_statelock); 4567 } 4568 4569 /* 4570 * If we are getting called as a side effect of an nfs_write() 4571 * operation the local file size might not be extended yet. 4572 * In this case we want to be able to return pages of zeroes. 4573 */ 4574 if (off + len > rp->r_size + PAGEOFFSET && seg != segkmap) { 4575 mutex_exit(&rp->r_statelock); 4576 return (EFAULT); /* beyond EOF */ 4577 } 4578 4579 mutex_exit(&rp->r_statelock); 4580 4581 if (len <= PAGESIZE) { 4582 error = nfs3_getapage(vp, off, len, protp, pl, plsz, 4583 seg, addr, rw, cr); 4584 } else { 4585 error = pvn_getpages(nfs3_getapage, vp, off, len, protp, 4586 pl, plsz, seg, addr, rw, cr); 4587 } 4588 4589 switch (error) { 4590 case NFS_EOF: 4591 nfs_purge_caches(vp, NFS_NOPURGE_DNLC, cr); 4592 goto retry; 4593 case ESTALE: 4594 PURGE_STALE_FH(error, vp, cr); 4595 } 4596 4597 return (error); 4598 } 4599 4600 /* 4601 * Called from pvn_getpages or nfs3_getpage to get a particular page. 4602 */ 4603 /* ARGSUSED */ 4604 static int 4605 nfs3_getapage(vnode_t *vp, u_offset_t off, size_t len, uint_t *protp, 4606 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 4607 enum seg_rw rw, cred_t *cr) 4608 { 4609 rnode_t *rp; 4610 uint_t bsize; 4611 struct buf *bp; 4612 page_t *pp; 4613 u_offset_t lbn; 4614 u_offset_t io_off; 4615 u_offset_t blkoff; 4616 u_offset_t rablkoff; 4617 size_t io_len; 4618 uint_t blksize; 4619 int error; 4620 int readahead; 4621 int readahead_issued = 0; 4622 int ra_window; /* readahead window */ 4623 page_t *pagefound; 4624 page_t *savepp; 4625 4626 if (nfs_zone() != VTOMI(vp)->mi_zone) 4627 return (EIO); 4628 rp = VTOR(vp); 4629 bsize = MAX(vp->v_vfsp->vfs_bsize, PAGESIZE); 4630 4631 reread: 4632 bp = NULL; 4633 pp = NULL; 4634 pagefound = NULL; 4635 4636 if (pl != NULL) 4637 pl[0] = NULL; 4638 4639 error = 0; 4640 lbn = off / bsize; 4641 blkoff = lbn * bsize; 4642 4643 /* 4644 * Queueing up the readahead before doing the synchronous read 4645 * results in a significant increase in read throughput because 4646 * of the increased parallelism between the async threads and 4647 * the process context. 4648 */ 4649 if ((off & ((vp->v_vfsp->vfs_bsize) - 1)) == 0 && 4650 rw != S_CREATE && 4651 !(vp->v_flag & VNOCACHE)) { 4652 mutex_enter(&rp->r_statelock); 4653 4654 /* 4655 * Calculate the number of readaheads to do. 4656 * a) No readaheads at offset = 0. 4657 * b) Do maximum(nfs3_nra) readaheads when the readahead 4658 * window is closed. 4659 * c) Do readaheads between 1 to (nfs3_nra - 1) depending 4660 * upon how far the readahead window is open or close. 4661 * d) No readaheads if rp->r_nextr is not within the scope 4662 * of the readahead window (random i/o). 4663 */ 4664 4665 if (off == 0) 4666 readahead = 0; 4667 else if (blkoff == rp->r_nextr) 4668 readahead = nfs3_nra; 4669 else if (rp->r_nextr > blkoff && 4670 ((ra_window = (rp->r_nextr - blkoff) / bsize) 4671 <= (nfs3_nra - 1))) 4672 readahead = nfs3_nra - ra_window; 4673 else 4674 readahead = 0; 4675 4676 rablkoff = rp->r_nextr; 4677 while (readahead > 0 && rablkoff + bsize < rp->r_size) { 4678 mutex_exit(&rp->r_statelock); 4679 if (nfs_async_readahead(vp, rablkoff + bsize, 4680 addr + (rablkoff + bsize - off), seg, cr, 4681 nfs3_readahead) < 0) { 4682 mutex_enter(&rp->r_statelock); 4683 break; 4684 } 4685 readahead--; 4686 rablkoff += bsize; 4687 /* 4688 * Indicate that we did a readahead so 4689 * readahead offset is not updated 4690 * by the synchronous read below. 4691 */ 4692 readahead_issued = 1; 4693 mutex_enter(&rp->r_statelock); 4694 /* 4695 * set readahead offset to 4696 * offset of last async readahead 4697 * request. 4698 */ 4699 rp->r_nextr = rablkoff; 4700 } 4701 mutex_exit(&rp->r_statelock); 4702 } 4703 4704 again: 4705 if ((pagefound = page_exists(vp, off)) == NULL) { 4706 if (pl == NULL) { 4707 (void) nfs_async_readahead(vp, blkoff, addr, seg, cr, 4708 nfs3_readahead); 4709 } else if (rw == S_CREATE) { 4710 /* 4711 * Block for this page is not allocated, or the offset 4712 * is beyond the current allocation size, or we're 4713 * allocating a swap slot and the page was not found, 4714 * so allocate it and return a zero page. 4715 */ 4716 if ((pp = page_create_va(vp, off, 4717 PAGESIZE, PG_WAIT, seg, addr)) == NULL) 4718 cmn_err(CE_PANIC, "nfs3_getapage: page_create"); 4719 io_len = PAGESIZE; 4720 mutex_enter(&rp->r_statelock); 4721 rp->r_nextr = off + PAGESIZE; 4722 mutex_exit(&rp->r_statelock); 4723 } else { 4724 /* 4725 * Need to go to server to get a BLOCK, exception to 4726 * that being while reading at offset = 0 or doing 4727 * random i/o, in that case read only a PAGE. 4728 */ 4729 mutex_enter(&rp->r_statelock); 4730 if (blkoff < rp->r_size && 4731 blkoff + bsize >= rp->r_size) { 4732 /* 4733 * If only a block or less is left in 4734 * the file, read all that is remaining. 4735 */ 4736 if (rp->r_size <= off) { 4737 /* 4738 * Trying to access beyond EOF, 4739 * set up to get at least one page. 4740 */ 4741 blksize = off + PAGESIZE - blkoff; 4742 } else 4743 blksize = rp->r_size - blkoff; 4744 } else if ((off == 0) || 4745 (off != rp->r_nextr && !readahead_issued)) { 4746 blksize = PAGESIZE; 4747 blkoff = off; /* block = page here */ 4748 } else 4749 blksize = bsize; 4750 mutex_exit(&rp->r_statelock); 4751 4752 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 4753 &io_len, blkoff, blksize, 0); 4754 4755 /* 4756 * Some other thread has entered the page, 4757 * so just use it. 4758 */ 4759 if (pp == NULL) 4760 goto again; 4761 4762 /* 4763 * Now round the request size up to page boundaries. 4764 * This ensures that the entire page will be 4765 * initialized to zeroes if EOF is encountered. 4766 */ 4767 io_len = ptob(btopr(io_len)); 4768 4769 bp = pageio_setup(pp, io_len, vp, B_READ); 4770 ASSERT(bp != NULL); 4771 4772 /* 4773 * pageio_setup should have set b_addr to 0. This 4774 * is correct since we want to do I/O on a page 4775 * boundary. bp_mapin will use this addr to calculate 4776 * an offset, and then set b_addr to the kernel virtual 4777 * address it allocated for us. 4778 */ 4779 ASSERT(bp->b_un.b_addr == 0); 4780 4781 bp->b_edev = 0; 4782 bp->b_dev = 0; 4783 bp->b_lblkno = lbtodb(io_off); 4784 bp->b_file = vp; 4785 bp->b_offset = (offset_t)off; 4786 bp_mapin(bp); 4787 4788 /* 4789 * If doing a write beyond what we believe is EOF, 4790 * don't bother trying to read the pages from the 4791 * server, we'll just zero the pages here. We 4792 * don't check that the rw flag is S_WRITE here 4793 * because some implementations may attempt a 4794 * read access to the buffer before copying data. 4795 */ 4796 mutex_enter(&rp->r_statelock); 4797 if (io_off >= rp->r_size && seg == segkmap) { 4798 mutex_exit(&rp->r_statelock); 4799 bzero(bp->b_un.b_addr, io_len); 4800 } else { 4801 mutex_exit(&rp->r_statelock); 4802 error = nfs3_bio(bp, NULL, cr); 4803 } 4804 4805 /* 4806 * Unmap the buffer before freeing it. 4807 */ 4808 bp_mapout(bp); 4809 pageio_done(bp); 4810 4811 savepp = pp; 4812 do { 4813 pp->p_fsdata = C_NOCOMMIT; 4814 } while ((pp = pp->p_next) != savepp); 4815 4816 if (error == NFS_EOF) { 4817 /* 4818 * If doing a write system call just return 4819 * zeroed pages, else user tried to get pages 4820 * beyond EOF, return error. We don't check 4821 * that the rw flag is S_WRITE here because 4822 * some implementations may attempt a read 4823 * access to the buffer before copying data. 4824 */ 4825 if (seg == segkmap) 4826 error = 0; 4827 else 4828 error = EFAULT; 4829 } 4830 4831 if (!readahead_issued && !error) { 4832 mutex_enter(&rp->r_statelock); 4833 rp->r_nextr = io_off + io_len; 4834 mutex_exit(&rp->r_statelock); 4835 } 4836 } 4837 } 4838 4839 out: 4840 if (pl == NULL) 4841 return (error); 4842 4843 if (error) { 4844 if (pp != NULL) 4845 pvn_read_done(pp, B_ERROR); 4846 return (error); 4847 } 4848 4849 if (pagefound) { 4850 se_t se = (rw == S_CREATE ? SE_EXCL : SE_SHARED); 4851 4852 /* 4853 * Page exists in the cache, acquire the appropriate lock. 4854 * If this fails, start all over again. 4855 */ 4856 if ((pp = page_lookup(vp, off, se)) == NULL) { 4857 #ifdef DEBUG 4858 nfs3_lostpage++; 4859 #endif 4860 goto reread; 4861 } 4862 pl[0] = pp; 4863 pl[1] = NULL; 4864 return (0); 4865 } 4866 4867 if (pp != NULL) 4868 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 4869 4870 return (error); 4871 } 4872 4873 static void 4874 nfs3_readahead(vnode_t *vp, u_offset_t blkoff, caddr_t addr, struct seg *seg, 4875 cred_t *cr) 4876 { 4877 int error; 4878 page_t *pp; 4879 u_offset_t io_off; 4880 size_t io_len; 4881 struct buf *bp; 4882 uint_t bsize, blksize; 4883 rnode_t *rp = VTOR(vp); 4884 page_t *savepp; 4885 4886 ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 4887 bsize = MAX(vp->v_vfsp->vfs_bsize, PAGESIZE); 4888 4889 mutex_enter(&rp->r_statelock); 4890 if (blkoff < rp->r_size && blkoff + bsize > rp->r_size) { 4891 /* 4892 * If less than a block left in file read less 4893 * than a block. 4894 */ 4895 blksize = rp->r_size - blkoff; 4896 } else 4897 blksize = bsize; 4898 mutex_exit(&rp->r_statelock); 4899 4900 pp = pvn_read_kluster(vp, blkoff, segkmap, addr, 4901 &io_off, &io_len, blkoff, blksize, 1); 4902 /* 4903 * The isra flag passed to the kluster function is 1, we may have 4904 * gotten a return value of NULL for a variety of reasons (# of free 4905 * pages < minfree, someone entered the page on the vnode etc). In all 4906 * cases, we want to punt on the readahead. 4907 */ 4908 if (pp == NULL) 4909 return; 4910 4911 /* 4912 * Now round the request size up to page boundaries. 4913 * This ensures that the entire page will be 4914 * initialized to zeroes if EOF is encountered. 4915 */ 4916 io_len = ptob(btopr(io_len)); 4917 4918 bp = pageio_setup(pp, io_len, vp, B_READ); 4919 ASSERT(bp != NULL); 4920 4921 /* 4922 * pageio_setup should have set b_addr to 0. This is correct since 4923 * we want to do I/O on a page boundary. bp_mapin() will use this addr 4924 * to calculate an offset, and then set b_addr to the kernel virtual 4925 * address it allocated for us. 4926 */ 4927 ASSERT(bp->b_un.b_addr == 0); 4928 4929 bp->b_edev = 0; 4930 bp->b_dev = 0; 4931 bp->b_lblkno = lbtodb(io_off); 4932 bp->b_file = vp; 4933 bp->b_offset = (offset_t)blkoff; 4934 bp_mapin(bp); 4935 4936 /* 4937 * If doing a write beyond what we believe is EOF, don't bother trying 4938 * to read the pages from the server, we'll just zero the pages here. 4939 * We don't check that the rw flag is S_WRITE here because some 4940 * implementations may attempt a read access to the buffer before 4941 * copying data. 4942 */ 4943 mutex_enter(&rp->r_statelock); 4944 if (io_off >= rp->r_size && seg == segkmap) { 4945 mutex_exit(&rp->r_statelock); 4946 bzero(bp->b_un.b_addr, io_len); 4947 error = 0; 4948 } else { 4949 mutex_exit(&rp->r_statelock); 4950 error = nfs3_bio(bp, NULL, cr); 4951 if (error == NFS_EOF) 4952 error = 0; 4953 } 4954 4955 /* 4956 * Unmap the buffer before freeing it. 4957 */ 4958 bp_mapout(bp); 4959 pageio_done(bp); 4960 4961 savepp = pp; 4962 do { 4963 pp->p_fsdata = C_NOCOMMIT; 4964 } while ((pp = pp->p_next) != savepp); 4965 4966 pvn_read_done(pp, error ? B_READ | B_ERROR : B_READ); 4967 4968 /* 4969 * In case of error set readahead offset 4970 * to the lowest offset. 4971 * pvn_read_done() calls VN_DISPOSE to destroy the pages 4972 */ 4973 if (error && rp->r_nextr > io_off) { 4974 mutex_enter(&rp->r_statelock); 4975 if (rp->r_nextr > io_off) 4976 rp->r_nextr = io_off; 4977 mutex_exit(&rp->r_statelock); 4978 } 4979 } 4980 4981 /* 4982 * Flags are composed of {B_INVAL, B_FREE, B_DONTNEED, B_FORCE} 4983 * If len == 0, do from off to EOF. 4984 * 4985 * The normal cases should be len == 0 && off == 0 (entire vp list), 4986 * len == MAXBSIZE (from segmap_release actions), and len == PAGESIZE 4987 * (from pageout). 4988 */ 4989 /* ARGSUSED */ 4990 static int 4991 nfs3_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr, 4992 caller_context_t *ct) 4993 { 4994 int error; 4995 rnode_t *rp; 4996 4997 ASSERT(cr != NULL); 4998 4999 /* 5000 * XXX - Why should this check be made here? 5001 */ 5002 if (vp->v_flag & VNOMAP) 5003 return (ENOSYS); 5004 if (len == 0 && !(flags & B_INVAL) && vn_is_readonly(vp)) 5005 return (0); 5006 if (!(flags & B_ASYNC) && nfs_zone() != VTOMI(vp)->mi_zone) 5007 return (EIO); 5008 5009 rp = VTOR(vp); 5010 mutex_enter(&rp->r_statelock); 5011 rp->r_count++; 5012 mutex_exit(&rp->r_statelock); 5013 error = nfs_putpages(vp, off, len, flags, cr); 5014 mutex_enter(&rp->r_statelock); 5015 rp->r_count--; 5016 cv_broadcast(&rp->r_cv); 5017 mutex_exit(&rp->r_statelock); 5018 5019 return (error); 5020 } 5021 5022 /* 5023 * Write out a single page, possibly klustering adjacent dirty pages. 5024 */ 5025 int 5026 nfs3_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp, 5027 int flags, cred_t *cr) 5028 { 5029 u_offset_t io_off; 5030 u_offset_t lbn_off; 5031 u_offset_t lbn; 5032 size_t io_len; 5033 uint_t bsize; 5034 int error; 5035 rnode_t *rp; 5036 5037 ASSERT(!vn_is_readonly(vp)); 5038 ASSERT(pp != NULL); 5039 ASSERT(cr != NULL); 5040 ASSERT((flags & B_ASYNC) || nfs_zone() == VTOMI(vp)->mi_zone); 5041 5042 rp = VTOR(vp); 5043 ASSERT(rp->r_count > 0); 5044 5045 bsize = MAX(vp->v_vfsp->vfs_bsize, PAGESIZE); 5046 lbn = pp->p_offset / bsize; 5047 lbn_off = lbn * bsize; 5048 5049 /* 5050 * Find a kluster that fits in one block, or in 5051 * one page if pages are bigger than blocks. If 5052 * there is less file space allocated than a whole 5053 * page, we'll shorten the i/o request below. 5054 */ 5055 pp = pvn_write_kluster(vp, pp, &io_off, &io_len, lbn_off, 5056 roundup(bsize, PAGESIZE), flags); 5057 5058 /* 5059 * pvn_write_kluster shouldn't have returned a page with offset 5060 * behind the original page we were given. Verify that. 5061 */ 5062 ASSERT((pp->p_offset / bsize) >= lbn); 5063 5064 /* 5065 * Now pp will have the list of kept dirty pages marked for 5066 * write back. It will also handle invalidation and freeing 5067 * of pages that are not dirty. Check for page length rounding 5068 * problems. 5069 */ 5070 if (io_off + io_len > lbn_off + bsize) { 5071 ASSERT((io_off + io_len) - (lbn_off + bsize) < PAGESIZE); 5072 io_len = lbn_off + bsize - io_off; 5073 } 5074 /* 5075 * The RMODINPROGRESS flag makes sure that nfs(3)_bio() sees a 5076 * consistent value of r_size. RMODINPROGRESS is set in writerp(). 5077 * When RMODINPROGRESS is set it indicates that a uiomove() is in 5078 * progress and the r_size has not been made consistent with the 5079 * new size of the file. When the uiomove() completes the r_size is 5080 * updated and the RMODINPROGRESS flag is cleared. 5081 * 5082 * The RMODINPROGRESS flag makes sure that nfs(3)_bio() sees a 5083 * consistent value of r_size. Without this handshaking, it is 5084 * possible that nfs(3)_bio() picks up the old value of r_size 5085 * before the uiomove() in writerp() completes. This will result 5086 * in the write through nfs(3)_bio() being dropped. 5087 * 5088 * More precisely, there is a window between the time the uiomove() 5089 * completes and the time the r_size is updated. If a VOP_PUTPAGE() 5090 * operation intervenes in this window, the page will be picked up, 5091 * because it is dirty (it will be unlocked, unless it was 5092 * pagecreate'd). When the page is picked up as dirty, the dirty 5093 * bit is reset (pvn_getdirty()). In nfs(3)write(), r_size is 5094 * checked. This will still be the old size. Therefore the page will 5095 * not be written out. When segmap_release() calls VOP_PUTPAGE(), 5096 * the page will be found to be clean and the write will be dropped. 5097 */ 5098 if (rp->r_flags & RMODINPROGRESS) { 5099 mutex_enter(&rp->r_statelock); 5100 if ((rp->r_flags & RMODINPROGRESS) && 5101 rp->r_modaddr + MAXBSIZE > io_off && 5102 rp->r_modaddr < io_off + io_len) { 5103 page_t *plist; 5104 /* 5105 * A write is in progress for this region of the file. 5106 * If we did not detect RMODINPROGRESS here then this 5107 * path through nfs_putapage() would eventually go to 5108 * nfs(3)_bio() and may not write out all of the data 5109 * in the pages. We end up losing data. So we decide 5110 * to set the modified bit on each page in the page 5111 * list and mark the rnode with RDIRTY. This write 5112 * will be restarted at some later time. 5113 */ 5114 plist = pp; 5115 while (plist != NULL) { 5116 pp = plist; 5117 page_sub(&plist, pp); 5118 hat_setmod(pp); 5119 page_io_unlock(pp); 5120 page_unlock(pp); 5121 } 5122 rp->r_flags |= RDIRTY; 5123 mutex_exit(&rp->r_statelock); 5124 if (offp) 5125 *offp = io_off; 5126 if (lenp) 5127 *lenp = io_len; 5128 return (0); 5129 } 5130 mutex_exit(&rp->r_statelock); 5131 } 5132 5133 if (flags & B_ASYNC) { 5134 error = nfs_async_putapage(vp, pp, io_off, io_len, flags, cr, 5135 nfs3_sync_putapage); 5136 } else 5137 error = nfs3_sync_putapage(vp, pp, io_off, io_len, flags, cr); 5138 5139 if (offp) 5140 *offp = io_off; 5141 if (lenp) 5142 *lenp = io_len; 5143 return (error); 5144 } 5145 5146 static int 5147 nfs3_sync_putapage(vnode_t *vp, page_t *pp, u_offset_t io_off, size_t io_len, 5148 int flags, cred_t *cr) 5149 { 5150 int error; 5151 rnode_t *rp; 5152 5153 ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 5154 5155 flags |= B_WRITE; 5156 5157 error = nfs3_rdwrlbn(vp, pp, io_off, io_len, flags, cr); 5158 5159 rp = VTOR(vp); 5160 5161 if ((error == ENOSPC || error == EDQUOT || error == EFBIG || 5162 error == EACCES) && 5163 (flags & (B_INVAL|B_FORCE)) != (B_INVAL|B_FORCE)) { 5164 if (!(rp->r_flags & ROUTOFSPACE)) { 5165 mutex_enter(&rp->r_statelock); 5166 rp->r_flags |= ROUTOFSPACE; 5167 mutex_exit(&rp->r_statelock); 5168 } 5169 flags |= B_ERROR; 5170 pvn_write_done(pp, flags); 5171 /* 5172 * If this was not an async thread, then try again to 5173 * write out the pages, but this time, also destroy 5174 * them whether or not the write is successful. This 5175 * will prevent memory from filling up with these 5176 * pages and destroying them is the only alternative 5177 * if they can't be written out. 5178 * 5179 * Don't do this if this is an async thread because 5180 * when the pages are unlocked in pvn_write_done, 5181 * some other thread could have come along, locked 5182 * them, and queued for an async thread. It would be 5183 * possible for all of the async threads to be tied 5184 * up waiting to lock the pages again and they would 5185 * all already be locked and waiting for an async 5186 * thread to handle them. Deadlock. 5187 */ 5188 if (!(flags & B_ASYNC)) { 5189 error = nfs3_putpage(vp, io_off, io_len, 5190 B_INVAL | B_FORCE, cr, NULL); 5191 } 5192 } else { 5193 if (error) 5194 flags |= B_ERROR; 5195 else if (rp->r_flags & ROUTOFSPACE) { 5196 mutex_enter(&rp->r_statelock); 5197 rp->r_flags &= ~ROUTOFSPACE; 5198 mutex_exit(&rp->r_statelock); 5199 } 5200 pvn_write_done(pp, flags); 5201 if (freemem < desfree) 5202 (void) nfs3_commit_vp(vp, (u_offset_t)0, 0, cr); 5203 } 5204 5205 return (error); 5206 } 5207 5208 /* ARGSUSED */ 5209 static int 5210 nfs3_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 5211 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, 5212 cred_t *cr, caller_context_t *ct) 5213 { 5214 struct segvn_crargs vn_a; 5215 int error; 5216 rnode_t *rp; 5217 struct vattr va; 5218 5219 if (nfs_zone() != VTOMI(vp)->mi_zone) 5220 return (EIO); 5221 5222 if (vp->v_flag & VNOMAP) 5223 return (ENOSYS); 5224 5225 if (off < 0 || off + len < 0) 5226 return (ENXIO); 5227 5228 if (vp->v_type != VREG) 5229 return (ENODEV); 5230 5231 /* 5232 * If there is cached data and if close-to-open consistency 5233 * checking is not turned off and if the file system is not 5234 * mounted readonly, then force an over the wire getattr. 5235 * Otherwise, just invoke nfs3getattr to get a copy of the 5236 * attributes. The attribute cache will be used unless it 5237 * is timed out and if it is, then an over the wire getattr 5238 * will be issued. 5239 */ 5240 va.va_mask = AT_ALL; 5241 if (vn_has_cached_data(vp) && 5242 !(VTOMI(vp)->mi_flags & MI_NOCTO) && !vn_is_readonly(vp)) 5243 error = nfs3_getattr_otw(vp, &va, cr); 5244 else 5245 error = nfs3getattr(vp, &va, cr); 5246 if (error) 5247 return (error); 5248 5249 /* 5250 * Check to see if the vnode is currently marked as not cachable. 5251 * This means portions of the file are locked (through VOP_FRLOCK). 5252 * In this case the map request must be refused. We use 5253 * rp->r_lkserlock to avoid a race with concurrent lock requests. 5254 */ 5255 rp = VTOR(vp); 5256 5257 /* 5258 * Atomically increment r_inmap after acquiring r_rwlock. The 5259 * idea here is to acquire r_rwlock to block read/write and 5260 * not to protect r_inmap. r_inmap will inform nfs3_read/write() 5261 * that we are in nfs3_map(). Now, r_rwlock is acquired in order 5262 * and we can prevent the deadlock that would have occurred 5263 * when nfs3_addmap() would have acquired it out of order. 5264 * 5265 * Since we are not protecting r_inmap by any lock, we do not 5266 * hold any lock when we decrement it. We atomically decrement 5267 * r_inmap after we release r_lkserlock. 5268 */ 5269 5270 if (nfs_rw_enter_sig(&rp->r_rwlock, RW_WRITER, INTR(vp))) 5271 return (EINTR); 5272 atomic_add_int(&rp->r_inmap, 1); 5273 nfs_rw_exit(&rp->r_rwlock); 5274 5275 if (nfs_rw_enter_sig(&rp->r_lkserlock, RW_READER, INTR(vp))) { 5276 atomic_add_int(&rp->r_inmap, -1); 5277 return (EINTR); 5278 } 5279 5280 if (vp->v_flag & VNOCACHE) { 5281 error = EAGAIN; 5282 goto done; 5283 } 5284 5285 /* 5286 * Don't allow concurrent locks and mapping if mandatory locking is 5287 * enabled. 5288 */ 5289 if ((flk_has_remote_locks(vp) || lm_has_sleep(vp)) && 5290 MANDLOCK(vp, va.va_mode)) { 5291 error = EAGAIN; 5292 goto done; 5293 } 5294 5295 as_rangelock(as); 5296 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags); 5297 if (error != 0) { 5298 as_rangeunlock(as); 5299 goto done; 5300 } 5301 5302 vn_a.vp = vp; 5303 vn_a.offset = off; 5304 vn_a.type = (flags & MAP_TYPE); 5305 vn_a.prot = (uchar_t)prot; 5306 vn_a.maxprot = (uchar_t)maxprot; 5307 vn_a.flags = (flags & ~MAP_TYPE); 5308 vn_a.cred = cr; 5309 vn_a.amp = NULL; 5310 vn_a.szc = 0; 5311 vn_a.lgrp_mem_policy_flags = 0; 5312 5313 error = as_map(as, *addrp, len, segvn_create, &vn_a); 5314 as_rangeunlock(as); 5315 5316 done: 5317 nfs_rw_exit(&rp->r_lkserlock); 5318 atomic_add_int(&rp->r_inmap, -1); 5319 return (error); 5320 } 5321 5322 /* ARGSUSED */ 5323 static int 5324 nfs3_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 5325 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, 5326 cred_t *cr, caller_context_t *ct) 5327 { 5328 rnode_t *rp; 5329 5330 if (vp->v_flag & VNOMAP) 5331 return (ENOSYS); 5332 if (nfs_zone() != VTOMI(vp)->mi_zone) 5333 return (EIO); 5334 5335 rp = VTOR(vp); 5336 atomic_add_long((ulong_t *)&rp->r_mapcnt, btopr(len)); 5337 5338 return (0); 5339 } 5340 5341 /* ARGSUSED */ 5342 static int 5343 nfs3_frlock(vnode_t *vp, int cmd, struct flock64 *bfp, int flag, 5344 offset_t offset, struct flk_callback *flk_cbp, cred_t *cr, 5345 caller_context_t *ct) 5346 { 5347 netobj lm_fh3; 5348 int rc; 5349 u_offset_t start, end; 5350 rnode_t *rp; 5351 int error = 0, intr = INTR(vp); 5352 5353 if (nfs_zone() != VTOMI(vp)->mi_zone) 5354 return (EIO); 5355 /* check for valid cmd parameter */ 5356 if (cmd != F_GETLK && cmd != F_SETLK && cmd != F_SETLKW) 5357 return (EINVAL); 5358 5359 /* Verify l_type. */ 5360 switch (bfp->l_type) { 5361 case F_RDLCK: 5362 if (cmd != F_GETLK && !(flag & FREAD)) 5363 return (EBADF); 5364 break; 5365 case F_WRLCK: 5366 if (cmd != F_GETLK && !(flag & FWRITE)) 5367 return (EBADF); 5368 break; 5369 case F_UNLCK: 5370 intr = 0; 5371 break; 5372 5373 default: 5374 return (EINVAL); 5375 } 5376 5377 /* check the validity of the lock range */ 5378 if (rc = flk_convert_lock_data(vp, bfp, &start, &end, offset)) 5379 return (rc); 5380 if (rc = flk_check_lock_data(start, end, MAXEND)) 5381 return (rc); 5382 5383 /* 5384 * If the filesystem is mounted using local locking, pass the 5385 * request off to the local locking code. 5386 */ 5387 if (VTOMI(vp)->mi_flags & MI_LLOCK) { 5388 if (cmd == F_SETLK || cmd == F_SETLKW) { 5389 /* 5390 * For complete safety, we should be holding 5391 * r_lkserlock. However, we can't call 5392 * lm_safelock and then fs_frlock while 5393 * holding r_lkserlock, so just invoke 5394 * lm_safelock and expect that this will 5395 * catch enough of the cases. 5396 */ 5397 if (!lm_safelock(vp, bfp, cr)) 5398 return (EAGAIN); 5399 } 5400 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct)); 5401 } 5402 5403 rp = VTOR(vp); 5404 5405 /* 5406 * Check whether the given lock request can proceed, given the 5407 * current file mappings. 5408 */ 5409 if (nfs_rw_enter_sig(&rp->r_lkserlock, RW_WRITER, intr)) 5410 return (EINTR); 5411 if (cmd == F_SETLK || cmd == F_SETLKW) { 5412 if (!lm_safelock(vp, bfp, cr)) { 5413 rc = EAGAIN; 5414 goto done; 5415 } 5416 } 5417 5418 /* 5419 * Flush the cache after waiting for async I/O to finish. For new 5420 * locks, this is so that the process gets the latest bits from the 5421 * server. For unlocks, this is so that other clients see the 5422 * latest bits once the file has been unlocked. If currently dirty 5423 * pages can't be flushed, then don't allow a lock to be set. But 5424 * allow unlocks to succeed, to avoid having orphan locks on the 5425 * server. 5426 */ 5427 if (cmd != F_GETLK) { 5428 mutex_enter(&rp->r_statelock); 5429 while (rp->r_count > 0) { 5430 if (intr) { 5431 klwp_t *lwp = ttolwp(curthread); 5432 5433 if (lwp != NULL) 5434 lwp->lwp_nostop++; 5435 if (cv_wait_sig(&rp->r_cv, 5436 &rp->r_statelock) == 0) { 5437 if (lwp != NULL) 5438 lwp->lwp_nostop--; 5439 rc = EINTR; 5440 break; 5441 } 5442 if (lwp != NULL) 5443 lwp->lwp_nostop--; 5444 } else 5445 cv_wait(&rp->r_cv, &rp->r_statelock); 5446 } 5447 mutex_exit(&rp->r_statelock); 5448 if (rc != 0) 5449 goto done; 5450 error = nfs3_putpage(vp, (offset_t)0, 0, B_INVAL, cr, ct); 5451 if (error) { 5452 if (error == ENOSPC || error == EDQUOT) { 5453 mutex_enter(&rp->r_statelock); 5454 if (!rp->r_error) 5455 rp->r_error = error; 5456 mutex_exit(&rp->r_statelock); 5457 } 5458 if (bfp->l_type != F_UNLCK) { 5459 rc = ENOLCK; 5460 goto done; 5461 } 5462 } 5463 } 5464 5465 lm_fh3.n_len = VTOFH3(vp)->fh3_length; 5466 lm_fh3.n_bytes = (char *)&(VTOFH3(vp)->fh3_u.data); 5467 5468 /* 5469 * Call the lock manager to do the real work of contacting 5470 * the server and obtaining the lock. 5471 */ 5472 rc = lm4_frlock(vp, cmd, bfp, flag, offset, cr, &lm_fh3, flk_cbp); 5473 5474 if (rc == 0) 5475 nfs_lockcompletion(vp, cmd); 5476 5477 done: 5478 nfs_rw_exit(&rp->r_lkserlock); 5479 return (rc); 5480 } 5481 5482 /* 5483 * Free storage space associated with the specified vnode. The portion 5484 * to be freed is specified by bfp->l_start and bfp->l_len (already 5485 * normalized to a "whence" of 0). 5486 * 5487 * This is an experimental facility whose continued existence is not 5488 * guaranteed. Currently, we only support the special case 5489 * of l_len == 0, meaning free to end of file. 5490 */ 5491 /* ARGSUSED */ 5492 static int 5493 nfs3_space(vnode_t *vp, int cmd, struct flock64 *bfp, int flag, 5494 offset_t offset, cred_t *cr, caller_context_t *ct) 5495 { 5496 int error; 5497 5498 ASSERT(vp->v_type == VREG); 5499 if (cmd != F_FREESP) 5500 return (EINVAL); 5501 if (nfs_zone() != VTOMI(vp)->mi_zone) 5502 return (EIO); 5503 5504 error = convoff(vp, bfp, 0, offset); 5505 if (!error) { 5506 ASSERT(bfp->l_start >= 0); 5507 if (bfp->l_len == 0) { 5508 struct vattr va; 5509 5510 /* 5511 * ftruncate should not change the ctime and 5512 * mtime if we truncate the file to its 5513 * previous size. 5514 */ 5515 va.va_mask = AT_SIZE; 5516 error = nfs3getattr(vp, &va, cr); 5517 if (error || va.va_size == bfp->l_start) 5518 return (error); 5519 va.va_mask = AT_SIZE; 5520 va.va_size = bfp->l_start; 5521 error = nfs3setattr(vp, &va, 0, cr); 5522 } else 5523 error = EINVAL; 5524 } 5525 5526 return (error); 5527 } 5528 5529 /* ARGSUSED */ 5530 static int 5531 nfs3_realvp(vnode_t *vp, vnode_t **vpp, caller_context_t *ct) 5532 { 5533 5534 return (EINVAL); 5535 } 5536 5537 /* 5538 * Setup and add an address space callback to do the work of the delmap call. 5539 * The callback will (and must be) deleted in the actual callback function. 5540 * 5541 * This is done in order to take care of the problem that we have with holding 5542 * the address space's a_lock for a long period of time (e.g. if the NFS server 5543 * is down). Callbacks will be executed in the address space code while the 5544 * a_lock is not held. Holding the address space's a_lock causes things such 5545 * as ps and fork to hang because they are trying to acquire this lock as well. 5546 */ 5547 /* ARGSUSED */ 5548 static int 5549 nfs3_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 5550 size_t len, uint_t prot, uint_t maxprot, uint_t flags, 5551 cred_t *cr, caller_context_t *ct) 5552 { 5553 int caller_found; 5554 int error; 5555 rnode_t *rp; 5556 nfs_delmap_args_t *dmapp; 5557 nfs_delmapcall_t *delmap_call; 5558 5559 if (vp->v_flag & VNOMAP) 5560 return (ENOSYS); 5561 /* 5562 * A process may not change zones if it has NFS pages mmap'ed 5563 * in, so we can't legitimately get here from the wrong zone. 5564 */ 5565 ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 5566 5567 rp = VTOR(vp); 5568 5569 /* 5570 * The way that the address space of this process deletes its mapping 5571 * of this file is via the following call chains: 5572 * - as_free()->SEGOP_UNMAP()/segvn_unmap()->VOP_DELMAP()/nfs3_delmap() 5573 * - as_unmap()->SEGOP_UNMAP()/segvn_unmap()->VOP_DELMAP()/nfs3_delmap() 5574 * 5575 * With the use of address space callbacks we are allowed to drop the 5576 * address space lock, a_lock, while executing the NFS operations that 5577 * need to go over the wire. Returning EAGAIN to the caller of this 5578 * function is what drives the execution of the callback that we add 5579 * below. The callback will be executed by the address space code 5580 * after dropping the a_lock. When the callback is finished, since 5581 * we dropped the a_lock, it must be re-acquired and segvn_unmap() 5582 * is called again on the same segment to finish the rest of the work 5583 * that needs to happen during unmapping. 5584 * 5585 * This action of calling back into the segment driver causes 5586 * nfs3_delmap() to get called again, but since the callback was 5587 * already executed at this point, it already did the work and there 5588 * is nothing left for us to do. 5589 * 5590 * To Summarize: 5591 * - The first time nfs3_delmap is called by the current thread is when 5592 * we add the caller associated with this delmap to the delmap caller 5593 * list, add the callback, and return EAGAIN. 5594 * - The second time in this call chain when nfs3_delmap is called we 5595 * will find this caller in the delmap caller list and realize there 5596 * is no more work to do thus removing this caller from the list and 5597 * returning the error that was set in the callback execution. 5598 */ 5599 caller_found = nfs_find_and_delete_delmapcall(rp, &error); 5600 if (caller_found) { 5601 /* 5602 * 'error' is from the actual delmap operations. To avoid 5603 * hangs, we need to handle the return of EAGAIN differently 5604 * since this is what drives the callback execution. 5605 * In this case, we don't want to return EAGAIN and do the 5606 * callback execution because there are none to execute. 5607 */ 5608 if (error == EAGAIN) 5609 return (0); 5610 else 5611 return (error); 5612 } 5613 5614 /* current caller was not in the list */ 5615 delmap_call = nfs_init_delmapcall(); 5616 5617 mutex_enter(&rp->r_statelock); 5618 list_insert_tail(&rp->r_indelmap, delmap_call); 5619 mutex_exit(&rp->r_statelock); 5620 5621 dmapp = kmem_alloc(sizeof (nfs_delmap_args_t), KM_SLEEP); 5622 5623 dmapp->vp = vp; 5624 dmapp->off = off; 5625 dmapp->addr = addr; 5626 dmapp->len = len; 5627 dmapp->prot = prot; 5628 dmapp->maxprot = maxprot; 5629 dmapp->flags = flags; 5630 dmapp->cr = cr; 5631 dmapp->caller = delmap_call; 5632 5633 error = as_add_callback(as, nfs3_delmap_callback, dmapp, 5634 AS_UNMAP_EVENT, addr, len, KM_SLEEP); 5635 5636 return (error ? error : EAGAIN); 5637 } 5638 5639 /* 5640 * Remove some pages from an mmap'd vnode. Just update the 5641 * count of pages. If doing close-to-open, then flush and 5642 * commit all of the pages associated with this file. 5643 * Otherwise, start an asynchronous page flush to write out 5644 * any dirty pages. This will also associate a credential 5645 * with the rnode which can be used to write the pages. 5646 */ 5647 /* ARGSUSED */ 5648 static void 5649 nfs3_delmap_callback(struct as *as, void *arg, uint_t event) 5650 { 5651 int error; 5652 rnode_t *rp; 5653 mntinfo_t *mi; 5654 nfs_delmap_args_t *dmapp = (nfs_delmap_args_t *)arg; 5655 5656 rp = VTOR(dmapp->vp); 5657 mi = VTOMI(dmapp->vp); 5658 5659 atomic_add_long((ulong_t *)&rp->r_mapcnt, -btopr(dmapp->len)); 5660 ASSERT(rp->r_mapcnt >= 0); 5661 5662 /* 5663 * Initiate a page flush and potential commit if there are 5664 * pages, the file system was not mounted readonly, the segment 5665 * was mapped shared, and the pages themselves were writeable. 5666 */ 5667 if (vn_has_cached_data(dmapp->vp) && !vn_is_readonly(dmapp->vp) && 5668 dmapp->flags == MAP_SHARED && (dmapp->maxprot & PROT_WRITE)) { 5669 mutex_enter(&rp->r_statelock); 5670 rp->r_flags |= RDIRTY; 5671 mutex_exit(&rp->r_statelock); 5672 /* 5673 * If this is a cross-zone access a sync putpage won't work, so 5674 * the best we can do is try an async putpage. That seems 5675 * better than something more draconian such as discarding the 5676 * dirty pages. 5677 */ 5678 if ((mi->mi_flags & MI_NOCTO) || 5679 nfs_zone() != mi->mi_zone) 5680 error = nfs3_putpage(dmapp->vp, dmapp->off, dmapp->len, 5681 B_ASYNC, dmapp->cr, NULL); 5682 else 5683 error = nfs3_putpage_commit(dmapp->vp, dmapp->off, 5684 dmapp->len, dmapp->cr); 5685 if (!error) { 5686 mutex_enter(&rp->r_statelock); 5687 error = rp->r_error; 5688 rp->r_error = 0; 5689 mutex_exit(&rp->r_statelock); 5690 } 5691 } else 5692 error = 0; 5693 5694 if ((rp->r_flags & RDIRECTIO) || (mi->mi_flags & MI_DIRECTIO)) 5695 (void) nfs3_putpage(dmapp->vp, dmapp->off, dmapp->len, 5696 B_INVAL, dmapp->cr, NULL); 5697 5698 dmapp->caller->error = error; 5699 (void) as_delete_callback(as, arg); 5700 kmem_free(dmapp, sizeof (nfs_delmap_args_t)); 5701 } 5702 5703 static int nfs3_pathconf_disable_cache = 0; 5704 5705 #ifdef DEBUG 5706 static int nfs3_pathconf_cache_hits = 0; 5707 static int nfs3_pathconf_cache_misses = 0; 5708 #endif 5709 5710 /* ARGSUSED */ 5711 static int 5712 nfs3_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 5713 caller_context_t *ct) 5714 { 5715 int error; 5716 PATHCONF3args args; 5717 PATHCONF3res res; 5718 int douprintf; 5719 failinfo_t fi; 5720 rnode_t *rp; 5721 hrtime_t t; 5722 5723 if (nfs_zone() != VTOMI(vp)->mi_zone) 5724 return (EIO); 5725 /* 5726 * Large file spec - need to base answer on info stored 5727 * on original FSINFO response. 5728 */ 5729 if (cmd == _PC_FILESIZEBITS) { 5730 unsigned long long ll; 5731 long l = 1; 5732 5733 ll = VTOMI(vp)->mi_maxfilesize; 5734 5735 if (ll == 0) { 5736 *valp = 0; 5737 return (0); 5738 } 5739 5740 if (ll & 0xffffffff00000000) { 5741 l += 32; ll >>= 32; 5742 } 5743 if (ll & 0xffff0000) { 5744 l += 16; ll >>= 16; 5745 } 5746 if (ll & 0xff00) { 5747 l += 8; ll >>= 8; 5748 } 5749 if (ll & 0xf0) { 5750 l += 4; ll >>= 4; 5751 } 5752 if (ll & 0xc) { 5753 l += 2; ll >>= 2; 5754 } 5755 if (ll & 0x2) 5756 l += 2; 5757 else if (ll & 0x1) 5758 l += 1; 5759 *valp = l; 5760 return (0); 5761 } 5762 5763 if (cmd == _PC_ACL_ENABLED) { 5764 *valp = _ACL_ACLENT_ENABLED; 5765 return (0); 5766 } 5767 5768 if (cmd == _PC_XATTR_EXISTS) { 5769 error = 0; 5770 *valp = 0; 5771 if (vp->v_vfsp->vfs_flag & VFS_XATTR) { 5772 vnode_t *avp; 5773 rnode_t *rp; 5774 int error = 0; 5775 mntinfo_t *mi = VTOMI(vp); 5776 5777 if (!(mi->mi_flags & MI_EXTATTR)) 5778 return (0); 5779 5780 rp = VTOR(vp); 5781 if (nfs_rw_enter_sig(&rp->r_rwlock, RW_READER, 5782 INTR(vp))) 5783 return (EINTR); 5784 5785 error = nfs3lookup_dnlc(vp, XATTR_DIR_NAME, &avp, cr); 5786 if (error || avp == NULL) 5787 error = acl_getxattrdir3(vp, &avp, 0, cr, 0); 5788 5789 nfs_rw_exit(&rp->r_rwlock); 5790 5791 if (error == 0 && avp != NULL) { 5792 error = do_xattr_exists_check(avp, valp, cr); 5793 VN_RELE(avp); 5794 } else if (error == ENOENT) { 5795 error = 0; 5796 *valp = 0; 5797 } 5798 } 5799 return (error); 5800 } 5801 5802 rp = VTOR(vp); 5803 if (rp->r_pathconf != NULL) { 5804 mutex_enter(&rp->r_statelock); 5805 if (rp->r_pathconf != NULL && nfs3_pathconf_disable_cache) { 5806 kmem_free(rp->r_pathconf, sizeof (*rp->r_pathconf)); 5807 rp->r_pathconf = NULL; 5808 } 5809 if (rp->r_pathconf != NULL) { 5810 error = 0; 5811 switch (cmd) { 5812 case _PC_LINK_MAX: 5813 *valp = rp->r_pathconf->link_max; 5814 break; 5815 case _PC_NAME_MAX: 5816 *valp = rp->r_pathconf->name_max; 5817 break; 5818 case _PC_PATH_MAX: 5819 case _PC_SYMLINK_MAX: 5820 *valp = MAXPATHLEN; 5821 break; 5822 case _PC_CHOWN_RESTRICTED: 5823 *valp = rp->r_pathconf->chown_restricted; 5824 break; 5825 case _PC_NO_TRUNC: 5826 *valp = rp->r_pathconf->no_trunc; 5827 break; 5828 default: 5829 error = EINVAL; 5830 break; 5831 } 5832 mutex_exit(&rp->r_statelock); 5833 #ifdef DEBUG 5834 nfs3_pathconf_cache_hits++; 5835 #endif 5836 return (error); 5837 } 5838 mutex_exit(&rp->r_statelock); 5839 } 5840 #ifdef DEBUG 5841 nfs3_pathconf_cache_misses++; 5842 #endif 5843 5844 args.object = *VTOFH3(vp); 5845 fi.vp = vp; 5846 fi.fhp = (caddr_t)&args.object; 5847 fi.copyproc = nfs3copyfh; 5848 fi.lookupproc = nfs3lookup; 5849 fi.xattrdirproc = acl_getxattrdir3; 5850 5851 douprintf = 1; 5852 5853 t = gethrtime(); 5854 5855 error = rfs3call(VTOMI(vp), NFSPROC3_PATHCONF, 5856 xdr_nfs_fh3, (caddr_t)&args, 5857 xdr_PATHCONF3res, (caddr_t)&res, cr, 5858 &douprintf, &res.status, 0, &fi); 5859 5860 if (error) 5861 return (error); 5862 5863 error = geterrno3(res.status); 5864 5865 if (!error) { 5866 nfs3_cache_post_op_attr(vp, &res.resok.obj_attributes, t, cr); 5867 if (!nfs3_pathconf_disable_cache) { 5868 mutex_enter(&rp->r_statelock); 5869 if (rp->r_pathconf == NULL) { 5870 rp->r_pathconf = kmem_alloc( 5871 sizeof (*rp->r_pathconf), KM_NOSLEEP); 5872 if (rp->r_pathconf != NULL) 5873 *rp->r_pathconf = res.resok.info; 5874 } 5875 mutex_exit(&rp->r_statelock); 5876 } 5877 switch (cmd) { 5878 case _PC_LINK_MAX: 5879 *valp = res.resok.info.link_max; 5880 break; 5881 case _PC_NAME_MAX: 5882 *valp = res.resok.info.name_max; 5883 break; 5884 case _PC_PATH_MAX: 5885 case _PC_SYMLINK_MAX: 5886 *valp = MAXPATHLEN; 5887 break; 5888 case _PC_CHOWN_RESTRICTED: 5889 *valp = res.resok.info.chown_restricted; 5890 break; 5891 case _PC_NO_TRUNC: 5892 *valp = res.resok.info.no_trunc; 5893 break; 5894 default: 5895 return (EINVAL); 5896 } 5897 } else { 5898 nfs3_cache_post_op_attr(vp, &res.resfail.obj_attributes, t, cr); 5899 PURGE_STALE_FH(error, vp, cr); 5900 } 5901 5902 return (error); 5903 } 5904 5905 /* 5906 * Called by async thread to do synchronous pageio. Do the i/o, wait 5907 * for it to complete, and cleanup the page list when done. 5908 */ 5909 static int 5910 nfs3_sync_pageio(vnode_t *vp, page_t *pp, u_offset_t io_off, size_t io_len, 5911 int flags, cred_t *cr) 5912 { 5913 int error; 5914 5915 ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 5916 error = nfs3_rdwrlbn(vp, pp, io_off, io_len, flags, cr); 5917 if (flags & B_READ) 5918 pvn_read_done(pp, (error ? B_ERROR : 0) | flags); 5919 else 5920 pvn_write_done(pp, (error ? B_ERROR : 0) | flags); 5921 return (error); 5922 } 5923 5924 /* ARGSUSED */ 5925 static int 5926 nfs3_pageio(vnode_t *vp, page_t *pp, u_offset_t io_off, size_t io_len, 5927 int flags, cred_t *cr, caller_context_t *ct) 5928 { 5929 int error; 5930 rnode_t *rp; 5931 5932 if (pp == NULL) 5933 return (EINVAL); 5934 if (!(flags & B_ASYNC) && nfs_zone() != VTOMI(vp)->mi_zone) 5935 return (EIO); 5936 5937 rp = VTOR(vp); 5938 mutex_enter(&rp->r_statelock); 5939 rp->r_count++; 5940 mutex_exit(&rp->r_statelock); 5941 5942 if (flags & B_ASYNC) { 5943 error = nfs_async_pageio(vp, pp, io_off, io_len, flags, cr, 5944 nfs3_sync_pageio); 5945 } else 5946 error = nfs3_rdwrlbn(vp, pp, io_off, io_len, flags, cr); 5947 mutex_enter(&rp->r_statelock); 5948 rp->r_count--; 5949 cv_broadcast(&rp->r_cv); 5950 mutex_exit(&rp->r_statelock); 5951 return (error); 5952 } 5953 5954 /* ARGSUSED */ 5955 static void 5956 nfs3_dispose(vnode_t *vp, page_t *pp, int fl, int dn, cred_t *cr, 5957 caller_context_t *ct) 5958 { 5959 int error; 5960 rnode_t *rp; 5961 page_t *plist; 5962 page_t *pptr; 5963 offset3 offset; 5964 count3 len; 5965 k_sigset_t smask; 5966 5967 /* 5968 * We should get called with fl equal to either B_FREE or 5969 * B_INVAL. Any other value is illegal. 5970 * 5971 * The page that we are either supposed to free or destroy 5972 * should be exclusive locked and its io lock should not 5973 * be held. 5974 */ 5975 ASSERT(fl == B_FREE || fl == B_INVAL); 5976 ASSERT((PAGE_EXCL(pp) && !page_iolock_assert(pp)) || panicstr); 5977 rp = VTOR(vp); 5978 5979 /* 5980 * If the page doesn't need to be committed or we shouldn't 5981 * even bother attempting to commit it, then just make sure 5982 * that the p_fsdata byte is clear and then either free or 5983 * destroy the page as appropriate. 5984 */ 5985 if (pp->p_fsdata == C_NOCOMMIT || (rp->r_flags & RSTALE)) { 5986 pp->p_fsdata = C_NOCOMMIT; 5987 if (fl == B_FREE) 5988 page_free(pp, dn); 5989 else 5990 page_destroy(pp, dn); 5991 return; 5992 } 5993 5994 /* 5995 * If there is a page invalidation operation going on, then 5996 * if this is one of the pages being destroyed, then just 5997 * clear the p_fsdata byte and then either free or destroy 5998 * the page as appropriate. 5999 */ 6000 mutex_enter(&rp->r_statelock); 6001 if ((rp->r_flags & RTRUNCATE) && pp->p_offset >= rp->r_truncaddr) { 6002 mutex_exit(&rp->r_statelock); 6003 pp->p_fsdata = C_NOCOMMIT; 6004 if (fl == B_FREE) 6005 page_free(pp, dn); 6006 else 6007 page_destroy(pp, dn); 6008 return; 6009 } 6010 6011 /* 6012 * If we are freeing this page and someone else is already 6013 * waiting to do a commit, then just unlock the page and 6014 * return. That other thread will take care of commiting 6015 * this page. The page can be freed sometime after the 6016 * commit has finished. Otherwise, if the page is marked 6017 * as delay commit, then we may be getting called from 6018 * pvn_write_done, one page at a time. This could result 6019 * in one commit per page, so we end up doing lots of small 6020 * commits instead of fewer larger commits. This is bad, 6021 * we want do as few commits as possible. 6022 */ 6023 if (fl == B_FREE) { 6024 if (rp->r_flags & RCOMMITWAIT) { 6025 page_unlock(pp); 6026 mutex_exit(&rp->r_statelock); 6027 return; 6028 } 6029 if (pp->p_fsdata == C_DELAYCOMMIT) { 6030 pp->p_fsdata = C_COMMIT; 6031 page_unlock(pp); 6032 mutex_exit(&rp->r_statelock); 6033 return; 6034 } 6035 } 6036 6037 /* 6038 * Check to see if there is a signal which would prevent an 6039 * attempt to commit the pages from being successful. If so, 6040 * then don't bother with all of the work to gather pages and 6041 * generate the unsuccessful RPC. Just return from here and 6042 * let the page be committed at some later time. 6043 */ 6044 sigintr(&smask, VTOMI(vp)->mi_flags & MI_INT); 6045 if (ttolwp(curthread) != NULL && ISSIG(curthread, JUSTLOOKING)) { 6046 sigunintr(&smask); 6047 page_unlock(pp); 6048 mutex_exit(&rp->r_statelock); 6049 return; 6050 } 6051 sigunintr(&smask); 6052 6053 /* 6054 * We are starting to need to commit pages, so let's try 6055 * to commit as many as possible at once to reduce the 6056 * overhead. 6057 * 6058 * Set the `commit inprogress' state bit. We must 6059 * first wait until any current one finishes. Then 6060 * we initialize the c_pages list with this page. 6061 */ 6062 while (rp->r_flags & RCOMMIT) { 6063 rp->r_flags |= RCOMMITWAIT; 6064 cv_wait(&rp->r_commit.c_cv, &rp->r_statelock); 6065 rp->r_flags &= ~RCOMMITWAIT; 6066 } 6067 rp->r_flags |= RCOMMIT; 6068 mutex_exit(&rp->r_statelock); 6069 ASSERT(rp->r_commit.c_pages == NULL); 6070 rp->r_commit.c_pages = pp; 6071 rp->r_commit.c_commbase = (offset3)pp->p_offset; 6072 rp->r_commit.c_commlen = PAGESIZE; 6073 6074 /* 6075 * Gather together all other pages which can be committed. 6076 * They will all be chained off r_commit.c_pages. 6077 */ 6078 nfs3_get_commit(vp); 6079 6080 /* 6081 * Clear the `commit inprogress' status and disconnect 6082 * the list of pages to be committed from the rnode. 6083 * At this same time, we also save the starting offset 6084 * and length of data to be committed on the server. 6085 */ 6086 plist = rp->r_commit.c_pages; 6087 rp->r_commit.c_pages = NULL; 6088 offset = rp->r_commit.c_commbase; 6089 len = rp->r_commit.c_commlen; 6090 mutex_enter(&rp->r_statelock); 6091 rp->r_flags &= ~RCOMMIT; 6092 cv_broadcast(&rp->r_commit.c_cv); 6093 mutex_exit(&rp->r_statelock); 6094 6095 if (curproc == proc_pageout || curproc == proc_fsflush || 6096 nfs_zone() != VTOMI(vp)->mi_zone) { 6097 nfs_async_commit(vp, plist, offset, len, cr, nfs3_async_commit); 6098 return; 6099 } 6100 6101 /* 6102 * Actually generate the COMMIT3 over the wire operation. 6103 */ 6104 error = nfs3_commit(vp, offset, len, cr); 6105 6106 /* 6107 * If we got an error during the commit, just unlock all 6108 * of the pages. The pages will get retransmitted to the 6109 * server during a putpage operation. 6110 */ 6111 if (error) { 6112 while (plist != NULL) { 6113 pptr = plist; 6114 page_sub(&plist, pptr); 6115 page_unlock(pptr); 6116 } 6117 return; 6118 } 6119 6120 /* 6121 * We've tried as hard as we can to commit the data to stable 6122 * storage on the server. We release the rest of the pages 6123 * and clear the commit required state. They will be put 6124 * onto the tail of the cachelist if they are nolonger 6125 * mapped. 6126 */ 6127 while (plist != pp) { 6128 pptr = plist; 6129 page_sub(&plist, pptr); 6130 pptr->p_fsdata = C_NOCOMMIT; 6131 (void) page_release(pptr, 1); 6132 } 6133 6134 /* 6135 * It is possible that nfs3_commit didn't return error but 6136 * some other thread has modified the page we are going 6137 * to free/destroy. 6138 * In this case we need to rewrite the page. Do an explicit check 6139 * before attempting to free/destroy the page. If modified, needs to 6140 * be rewritten so unlock the page and return. 6141 */ 6142 if (hat_ismod(pp)) { 6143 pp->p_fsdata = C_NOCOMMIT; 6144 page_unlock(pp); 6145 return; 6146 } 6147 6148 /* 6149 * Now, as appropriate, either free or destroy the page 6150 * that we were called with. 6151 */ 6152 pp->p_fsdata = C_NOCOMMIT; 6153 if (fl == B_FREE) 6154 page_free(pp, dn); 6155 else 6156 page_destroy(pp, dn); 6157 } 6158 6159 static int 6160 nfs3_commit(vnode_t *vp, offset3 offset, count3 count, cred_t *cr) 6161 { 6162 int error; 6163 rnode_t *rp; 6164 COMMIT3args args; 6165 COMMIT3res res; 6166 int douprintf; 6167 cred_t *cred; 6168 6169 rp = VTOR(vp); 6170 ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 6171 6172 mutex_enter(&rp->r_statelock); 6173 if (rp->r_cred != NULL) { 6174 cred = rp->r_cred; 6175 crhold(cred); 6176 } else { 6177 rp->r_cred = cr; 6178 crhold(cr); 6179 cred = cr; 6180 crhold(cred); 6181 } 6182 mutex_exit(&rp->r_statelock); 6183 6184 args.file = *VTOFH3(vp); 6185 args.offset = offset; 6186 args.count = count; 6187 6188 doitagain: 6189 douprintf = 1; 6190 error = rfs3call(VTOMI(vp), NFSPROC3_COMMIT, 6191 xdr_COMMIT3args, (caddr_t)&args, 6192 xdr_COMMIT3res, (caddr_t)&res, cred, 6193 &douprintf, &res.status, 0, NULL); 6194 6195 crfree(cred); 6196 6197 if (error) 6198 return (error); 6199 6200 error = geterrno3(res.status); 6201 if (!error) { 6202 ASSERT(rp->r_flags & RHAVEVERF); 6203 mutex_enter(&rp->r_statelock); 6204 if (rp->r_verf == res.resok.verf) { 6205 mutex_exit(&rp->r_statelock); 6206 return (0); 6207 } 6208 nfs3_set_mod(vp); 6209 rp->r_verf = res.resok.verf; 6210 mutex_exit(&rp->r_statelock); 6211 error = NFS_VERF_MISMATCH; 6212 } else { 6213 if (error == EACCES) { 6214 mutex_enter(&rp->r_statelock); 6215 if (cred != cr) { 6216 if (rp->r_cred != NULL) 6217 crfree(rp->r_cred); 6218 rp->r_cred = cr; 6219 crhold(cr); 6220 cred = cr; 6221 crhold(cred); 6222 mutex_exit(&rp->r_statelock); 6223 goto doitagain; 6224 } 6225 mutex_exit(&rp->r_statelock); 6226 } 6227 /* 6228 * Can't do a PURGE_STALE_FH here because this 6229 * can cause a deadlock. nfs3_commit can 6230 * be called from nfs3_dispose which can be called 6231 * indirectly via pvn_vplist_dirty. PURGE_STALE_FH 6232 * can call back to pvn_vplist_dirty. 6233 */ 6234 if (error == ESTALE) { 6235 mutex_enter(&rp->r_statelock); 6236 rp->r_flags |= RSTALE; 6237 if (!rp->r_error) 6238 rp->r_error = error; 6239 mutex_exit(&rp->r_statelock); 6240 PURGE_ATTRCACHE(vp); 6241 } else { 6242 mutex_enter(&rp->r_statelock); 6243 if (!rp->r_error) 6244 rp->r_error = error; 6245 mutex_exit(&rp->r_statelock); 6246 } 6247 } 6248 6249 return (error); 6250 } 6251 6252 static void 6253 nfs3_set_mod(vnode_t *vp) 6254 { 6255 page_t *pp; 6256 kmutex_t *vphm; 6257 6258 ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 6259 vphm = page_vnode_mutex(vp); 6260 mutex_enter(vphm); 6261 if ((pp = vp->v_pages) != NULL) { 6262 do { 6263 if (pp->p_fsdata != C_NOCOMMIT) { 6264 hat_setmod(pp); 6265 pp->p_fsdata = C_NOCOMMIT; 6266 } 6267 } while ((pp = pp->p_vpnext) != vp->v_pages); 6268 } 6269 mutex_exit(vphm); 6270 } 6271 6272 6273 /* 6274 * This routine is used to gather together a page list of the pages 6275 * which are to be committed on the server. This routine must not 6276 * be called if the calling thread holds any locked pages. 6277 * 6278 * The calling thread must have set RCOMMIT. This bit is used to 6279 * serialize access to the commit structure in the rnode. As long 6280 * as the thread has set RCOMMIT, then it can manipulate the commit 6281 * structure without requiring any other locks. 6282 */ 6283 static void 6284 nfs3_get_commit(vnode_t *vp) 6285 { 6286 rnode_t *rp; 6287 page_t *pp; 6288 kmutex_t *vphm; 6289 6290 rp = VTOR(vp); 6291 6292 ASSERT(rp->r_flags & RCOMMIT); 6293 6294 vphm = page_vnode_mutex(vp); 6295 mutex_enter(vphm); 6296 6297 /* 6298 * If there are no pages associated with this vnode, then 6299 * just return. 6300 */ 6301 if ((pp = vp->v_pages) == NULL) { 6302 mutex_exit(vphm); 6303 return; 6304 } 6305 6306 /* 6307 * Step through all of the pages associated with this vnode 6308 * looking for pages which need to be committed. 6309 */ 6310 do { 6311 /* 6312 * If this page does not need to be committed or is 6313 * modified, then just skip it. 6314 */ 6315 if (pp->p_fsdata == C_NOCOMMIT || hat_ismod(pp)) 6316 continue; 6317 6318 /* 6319 * Attempt to lock the page. If we can't, then 6320 * someone else is messing with it and we will 6321 * just skip it. 6322 */ 6323 if (!page_trylock(pp, SE_EXCL)) 6324 continue; 6325 6326 /* 6327 * If this page does not need to be committed or is 6328 * modified, then just skip it. Recheck now that 6329 * the page is locked. 6330 */ 6331 if (pp->p_fsdata == C_NOCOMMIT || hat_ismod(pp)) { 6332 page_unlock(pp); 6333 continue; 6334 } 6335 6336 if (PP_ISFREE(pp)) { 6337 cmn_err(CE_PANIC, "nfs3_get_commit: %p is free", 6338 (void *)pp); 6339 } 6340 6341 /* 6342 * The page needs to be committed and we locked it. 6343 * Update the base and length parameters and add it 6344 * to r_pages. 6345 */ 6346 if (rp->r_commit.c_pages == NULL) { 6347 rp->r_commit.c_commbase = (offset3)pp->p_offset; 6348 rp->r_commit.c_commlen = PAGESIZE; 6349 } else if (pp->p_offset < rp->r_commit.c_commbase) { 6350 rp->r_commit.c_commlen = rp->r_commit.c_commbase - 6351 (offset3)pp->p_offset + rp->r_commit.c_commlen; 6352 rp->r_commit.c_commbase = (offset3)pp->p_offset; 6353 } else if ((rp->r_commit.c_commbase + rp->r_commit.c_commlen) 6354 <= pp->p_offset) { 6355 rp->r_commit.c_commlen = (offset3)pp->p_offset - 6356 rp->r_commit.c_commbase + PAGESIZE; 6357 } 6358 page_add(&rp->r_commit.c_pages, pp); 6359 } while ((pp = pp->p_vpnext) != vp->v_pages); 6360 6361 mutex_exit(vphm); 6362 } 6363 6364 /* 6365 * This routine is used to gather together a page list of the pages 6366 * which are to be committed on the server. This routine must not 6367 * be called if the calling thread holds any locked pages. 6368 * 6369 * The calling thread must have set RCOMMIT. This bit is used to 6370 * serialize access to the commit structure in the rnode. As long 6371 * as the thread has set RCOMMIT, then it can manipulate the commit 6372 * structure without requiring any other locks. 6373 */ 6374 static void 6375 nfs3_get_commit_range(vnode_t *vp, u_offset_t soff, size_t len) 6376 { 6377 6378 rnode_t *rp; 6379 page_t *pp; 6380 u_offset_t end; 6381 u_offset_t off; 6382 6383 ASSERT(len != 0); 6384 6385 rp = VTOR(vp); 6386 6387 ASSERT(rp->r_flags & RCOMMIT); 6388 ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 6389 6390 /* 6391 * If there are no pages associated with this vnode, then 6392 * just return. 6393 */ 6394 if ((pp = vp->v_pages) == NULL) 6395 return; 6396 6397 /* 6398 * Calculate the ending offset. 6399 */ 6400 end = soff + len; 6401 6402 for (off = soff; off < end; off += PAGESIZE) { 6403 /* 6404 * Lookup each page by vp, offset. 6405 */ 6406 if ((pp = page_lookup_nowait(vp, off, SE_EXCL)) == NULL) 6407 continue; 6408 6409 /* 6410 * If this page does not need to be committed or is 6411 * modified, then just skip it. 6412 */ 6413 if (pp->p_fsdata == C_NOCOMMIT || hat_ismod(pp)) { 6414 page_unlock(pp); 6415 continue; 6416 } 6417 6418 ASSERT(PP_ISFREE(pp) == 0); 6419 6420 /* 6421 * The page needs to be committed and we locked it. 6422 * Update the base and length parameters and add it 6423 * to r_pages. 6424 */ 6425 if (rp->r_commit.c_pages == NULL) { 6426 rp->r_commit.c_commbase = (offset3)pp->p_offset; 6427 rp->r_commit.c_commlen = PAGESIZE; 6428 } else { 6429 rp->r_commit.c_commlen = (offset3)pp->p_offset - 6430 rp->r_commit.c_commbase + PAGESIZE; 6431 } 6432 page_add(&rp->r_commit.c_pages, pp); 6433 } 6434 } 6435 6436 #if 0 /* unused */ 6437 #ifdef DEBUG 6438 static int 6439 nfs3_no_uncommitted_pages(vnode_t *vp) 6440 { 6441 page_t *pp; 6442 kmutex_t *vphm; 6443 6444 vphm = page_vnode_mutex(vp); 6445 mutex_enter(vphm); 6446 if ((pp = vp->v_pages) != NULL) { 6447 do { 6448 if (pp->p_fsdata != C_NOCOMMIT) { 6449 mutex_exit(vphm); 6450 return (0); 6451 } 6452 } while ((pp = pp->p_vpnext) != vp->v_pages); 6453 } 6454 mutex_exit(vphm); 6455 6456 return (1); 6457 } 6458 #endif 6459 #endif 6460 6461 static int 6462 nfs3_putpage_commit(vnode_t *vp, offset_t poff, size_t plen, cred_t *cr) 6463 { 6464 int error; 6465 writeverf3 write_verf; 6466 rnode_t *rp = VTOR(vp); 6467 6468 ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 6469 /* 6470 * Flush the data portion of the file and then commit any 6471 * portions which need to be committed. This may need to 6472 * be done twice if the server has changed state since 6473 * data was last written. The data will need to be 6474 * rewritten to the server and then a new commit done. 6475 * 6476 * In fact, this may need to be done several times if the 6477 * server is having problems and crashing while we are 6478 * attempting to do this. 6479 */ 6480 6481 top: 6482 /* 6483 * Do a flush based on the poff and plen arguments. This 6484 * will asynchronously write out any modified pages in the 6485 * range specified by (poff, plen). This starts all of the 6486 * i/o operations which will be waited for in the next 6487 * call to nfs3_putpage 6488 */ 6489 6490 mutex_enter(&rp->r_statelock); 6491 write_verf = rp->r_verf; 6492 mutex_exit(&rp->r_statelock); 6493 6494 error = nfs3_putpage(vp, poff, plen, B_ASYNC, cr, NULL); 6495 if (error == EAGAIN) 6496 error = 0; 6497 6498 /* 6499 * Do a flush based on the poff and plen arguments. This 6500 * will synchronously write out any modified pages in the 6501 * range specified by (poff, plen) and wait until all of 6502 * the asynchronous i/o's in that range are done as well. 6503 */ 6504 if (!error) 6505 error = nfs3_putpage(vp, poff, plen, 0, cr, NULL); 6506 6507 if (error) 6508 return (error); 6509 6510 mutex_enter(&rp->r_statelock); 6511 if (rp->r_verf != write_verf) { 6512 mutex_exit(&rp->r_statelock); 6513 goto top; 6514 } 6515 mutex_exit(&rp->r_statelock); 6516 6517 /* 6518 * Now commit any pages which might need to be committed. 6519 * If the error, NFS_VERF_MISMATCH, is returned, then 6520 * start over with the flush operation. 6521 */ 6522 6523 error = nfs3_commit_vp(vp, poff, plen, cr); 6524 6525 if (error == NFS_VERF_MISMATCH) 6526 goto top; 6527 6528 return (error); 6529 } 6530 6531 static int 6532 nfs3_commit_vp(vnode_t *vp, u_offset_t poff, size_t plen, cred_t *cr) 6533 { 6534 rnode_t *rp; 6535 page_t *plist; 6536 offset3 offset; 6537 count3 len; 6538 6539 6540 rp = VTOR(vp); 6541 6542 if (nfs_zone() != VTOMI(vp)->mi_zone) 6543 return (EIO); 6544 /* 6545 * Set the `commit inprogress' state bit. We must 6546 * first wait until any current one finishes. 6547 */ 6548 mutex_enter(&rp->r_statelock); 6549 while (rp->r_flags & RCOMMIT) { 6550 rp->r_flags |= RCOMMITWAIT; 6551 cv_wait(&rp->r_commit.c_cv, &rp->r_statelock); 6552 rp->r_flags &= ~RCOMMITWAIT; 6553 } 6554 rp->r_flags |= RCOMMIT; 6555 mutex_exit(&rp->r_statelock); 6556 6557 /* 6558 * Gather together all of the pages which need to be 6559 * committed. 6560 */ 6561 if (plen == 0) 6562 nfs3_get_commit(vp); 6563 else 6564 nfs3_get_commit_range(vp, poff, plen); 6565 6566 /* 6567 * Clear the `commit inprogress' bit and disconnect the 6568 * page list which was gathered together in nfs3_get_commit. 6569 */ 6570 plist = rp->r_commit.c_pages; 6571 rp->r_commit.c_pages = NULL; 6572 offset = rp->r_commit.c_commbase; 6573 len = rp->r_commit.c_commlen; 6574 mutex_enter(&rp->r_statelock); 6575 rp->r_flags &= ~RCOMMIT; 6576 cv_broadcast(&rp->r_commit.c_cv); 6577 mutex_exit(&rp->r_statelock); 6578 6579 /* 6580 * If any pages need to be committed, commit them and 6581 * then unlock them so that they can be freed some 6582 * time later. 6583 */ 6584 if (plist != NULL) { 6585 /* 6586 * No error occurred during the flush portion 6587 * of this operation, so now attempt to commit 6588 * the data to stable storage on the server. 6589 * 6590 * This will unlock all of the pages on the list. 6591 */ 6592 return (nfs3_sync_commit(vp, plist, offset, len, cr)); 6593 } 6594 return (0); 6595 } 6596 6597 static int 6598 nfs3_sync_commit(vnode_t *vp, page_t *plist, offset3 offset, count3 count, 6599 cred_t *cr) 6600 { 6601 int error; 6602 page_t *pp; 6603 6604 ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 6605 error = nfs3_commit(vp, offset, count, cr); 6606 6607 /* 6608 * If we got an error, then just unlock all of the pages 6609 * on the list. 6610 */ 6611 if (error) { 6612 while (plist != NULL) { 6613 pp = plist; 6614 page_sub(&plist, pp); 6615 page_unlock(pp); 6616 } 6617 return (error); 6618 } 6619 /* 6620 * We've tried as hard as we can to commit the data to stable 6621 * storage on the server. We just unlock the pages and clear 6622 * the commit required state. They will get freed later. 6623 */ 6624 while (plist != NULL) { 6625 pp = plist; 6626 page_sub(&plist, pp); 6627 pp->p_fsdata = C_NOCOMMIT; 6628 page_unlock(pp); 6629 } 6630 6631 return (error); 6632 } 6633 6634 static void 6635 nfs3_async_commit(vnode_t *vp, page_t *plist, offset3 offset, count3 count, 6636 cred_t *cr) 6637 { 6638 ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 6639 (void) nfs3_sync_commit(vp, plist, offset, count, cr); 6640 } 6641 6642 /* ARGSUSED */ 6643 static int 6644 nfs3_setsecattr(vnode_t *vp, vsecattr_t *vsecattr, int flag, cred_t *cr, 6645 caller_context_t *ct) 6646 { 6647 int error; 6648 mntinfo_t *mi; 6649 6650 mi = VTOMI(vp); 6651 6652 if (nfs_zone() != mi->mi_zone) 6653 return (EIO); 6654 6655 if (mi->mi_flags & MI_ACL) { 6656 error = acl_setacl3(vp, vsecattr, flag, cr); 6657 if (mi->mi_flags & MI_ACL) 6658 return (error); 6659 } 6660 6661 return (ENOSYS); 6662 } 6663 6664 /* ARGSUSED */ 6665 static int 6666 nfs3_getsecattr(vnode_t *vp, vsecattr_t *vsecattr, int flag, cred_t *cr, 6667 caller_context_t *ct) 6668 { 6669 int error; 6670 mntinfo_t *mi; 6671 6672 mi = VTOMI(vp); 6673 6674 if (nfs_zone() != mi->mi_zone) 6675 return (EIO); 6676 6677 if (mi->mi_flags & MI_ACL) { 6678 error = acl_getacl3(vp, vsecattr, flag, cr); 6679 if (mi->mi_flags & MI_ACL) 6680 return (error); 6681 } 6682 6683 return (fs_fab_acl(vp, vsecattr, flag, cr, ct)); 6684 } 6685 6686 /* ARGSUSED */ 6687 static int 6688 nfs3_shrlock(vnode_t *vp, int cmd, struct shrlock *shr, int flag, cred_t *cr, 6689 caller_context_t *ct) 6690 { 6691 int error; 6692 struct shrlock nshr; 6693 struct nfs_owner nfs_owner; 6694 netobj lm_fh3; 6695 6696 if (nfs_zone() != VTOMI(vp)->mi_zone) 6697 return (EIO); 6698 6699 /* 6700 * check for valid cmd parameter 6701 */ 6702 if (cmd != F_SHARE && cmd != F_UNSHARE && cmd != F_HASREMOTELOCKS) 6703 return (EINVAL); 6704 6705 /* 6706 * Check access permissions 6707 */ 6708 if (cmd == F_SHARE && 6709 (((shr->s_access & F_RDACC) && !(flag & FREAD)) || 6710 ((shr->s_access & F_WRACC) && !(flag & FWRITE)))) 6711 return (EBADF); 6712 6713 /* 6714 * If the filesystem is mounted using local locking, pass the 6715 * request off to the local share code. 6716 */ 6717 if (VTOMI(vp)->mi_flags & MI_LLOCK) 6718 return (fs_shrlock(vp, cmd, shr, flag, cr, ct)); 6719 6720 switch (cmd) { 6721 case F_SHARE: 6722 case F_UNSHARE: 6723 lm_fh3.n_len = VTOFH3(vp)->fh3_length; 6724 lm_fh3.n_bytes = (char *)&(VTOFH3(vp)->fh3_u.data); 6725 6726 /* 6727 * If passed an owner that is too large to fit in an 6728 * nfs_owner it is likely a recursive call from the 6729 * lock manager client and pass it straight through. If 6730 * it is not a nfs_owner then simply return an error. 6731 */ 6732 if (shr->s_own_len > sizeof (nfs_owner.lowner)) { 6733 if (((struct nfs_owner *)shr->s_owner)->magic != 6734 NFS_OWNER_MAGIC) 6735 return (EINVAL); 6736 6737 if (error = lm4_shrlock(vp, cmd, shr, flag, &lm_fh3)) { 6738 error = set_errno(error); 6739 } 6740 return (error); 6741 } 6742 /* 6743 * Remote share reservations owner is a combination of 6744 * a magic number, hostname, and the local owner 6745 */ 6746 bzero(&nfs_owner, sizeof (nfs_owner)); 6747 nfs_owner.magic = NFS_OWNER_MAGIC; 6748 (void) strncpy(nfs_owner.hname, uts_nodename(), 6749 sizeof (nfs_owner.hname)); 6750 bcopy(shr->s_owner, nfs_owner.lowner, shr->s_own_len); 6751 nshr.s_access = shr->s_access; 6752 nshr.s_deny = shr->s_deny; 6753 nshr.s_sysid = 0; 6754 nshr.s_pid = ttoproc(curthread)->p_pid; 6755 nshr.s_own_len = sizeof (nfs_owner); 6756 nshr.s_owner = (caddr_t)&nfs_owner; 6757 6758 if (error = lm4_shrlock(vp, cmd, &nshr, flag, &lm_fh3)) { 6759 error = set_errno(error); 6760 } 6761 6762 break; 6763 6764 case F_HASREMOTELOCKS: 6765 /* 6766 * NFS client can't store remote locks itself 6767 */ 6768 shr->s_access = 0; 6769 error = 0; 6770 break; 6771 6772 default: 6773 error = EINVAL; 6774 break; 6775 } 6776 6777 return (error); 6778 } 6779
Indexes created Wed Nov 25 23:03:41 UTC 2009
Terms of Use |
Privacy |
Trademarks |
Copyright Policy |
Site Guidelines |
Help
Your use of this web site or any of its content or software indicates your agreement to be bound by these Terms of Use.
Copyright © 1995-2008 Sun Microsystems, Inc.