1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 */ 35 36 #include <sys/capsicum.h> 37 #include <sys/extattr.h> 38 39 /* 40 * Functions that perform the vfs operations required by the routines in 41 * nfsd_serv.c. It is hoped that this change will make the server more 42 * portable. 43 */ 44 45 #include <fs/nfs/nfsport.h> 46 #include <security/mac/mac_framework.h> 47 #include <sys/callout.h> 48 #include <sys/filio.h> 49 #include <sys/hash.h> 50 #include <sys/osd.h> 51 #include <sys/sysctl.h> 52 #include <nlm/nlm_prot.h> 53 #include <nlm/nlm.h> 54 #include <vm/vm_param.h> 55 #include <vm/vnode_pager.h> 56 57 FEATURE(nfsd, "NFSv4 server"); 58 59 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1; 60 extern int nfsrv_useacl; 61 extern int newnfs_numnfsd; 62 extern int nfsrv_sessionhashsize; 63 extern struct nfslayouthash *nfslayouthash; 64 extern int nfsrv_layouthashsize; 65 extern struct mtx nfsrv_dslock_mtx; 66 extern int nfs_pnfsiothreads; 67 extern volatile int nfsrv_devidcnt; 68 extern int nfsrv_maxpnfsmirror; 69 extern uint32_t nfs_srvmaxio; 70 extern int nfs_bufpackets; 71 extern u_long sb_max_adj; 72 extern struct nfsv4lock nfsv4rootfs_lock; 73 74 NFSD_VNET_DECLARE(int, nfsrv_numnfsd); 75 NFSD_VNET_DECLARE(struct nfsrv_stablefirst, nfsrv_stablefirst); 76 NFSD_VNET_DECLARE(SVCPOOL *, nfsrvd_pool); 77 NFSD_VNET_DECLARE(struct nfsclienthashhead *, nfsclienthash); 78 NFSD_VNET_DECLARE(struct nfslockhashhead *, nfslockhash); 79 NFSD_VNET_DECLARE(struct nfssessionhash *, nfssessionhash); 80 NFSD_VNET_DECLARE(struct nfsv4lock, nfsd_suspend_lock); 81 NFSD_VNET_DECLARE(struct nfsstatsv1 *, nfsstatsv1_p); 82 83 NFSDLOCKMUTEX; 84 NFSSTATESPINLOCK; 85 struct mtx nfsrc_udpmtx; 86 struct mtx nfs_v4root_mutex; 87 struct mtx nfsrv_dontlistlock_mtx; 88 struct mtx nfsrv_recalllock_mtx; 89 struct nfsrvfh nfs_pubfh; 90 int nfs_pubfhset = 0; 91 int nfsd_debuglevel = 0; 92 static pid_t nfsd_master_pid = (pid_t)-1; 93 static char nfsd_master_comm[MAXCOMLEN + 1]; 94 static struct timeval nfsd_master_start; 95 static uint32_t nfsv4_sysid = 0; 96 static fhandle_t zerofh; 97 98 NFSD_VNET_DEFINE(struct proc *, nfsd_master_proc) = NULL; 99 NFSD_VNET_DEFINE(struct nfsrvhashhead *, nfsrvudphashtbl); 100 NFSD_VNET_DEFINE(struct nfsrchash_bucket *, nfsrchash_table); 101 NFSD_VNET_DEFINE(struct nfsrchash_bucket *, nfsrcahash_table); 102 NFSD_VNET_DEFINE(struct nfsrvfh, nfs_rootfh); 103 NFSD_VNET_DEFINE(int, nfs_rootfhset) = 0; 104 NFSD_VNET_DEFINE(struct callout, nfsd_callout); 105 NFSD_VNET_DEFINE_STATIC(struct mount *, nfsv4root_mnt); 106 NFSD_VNET_DEFINE_STATIC(struct vfsoptlist, nfsv4root_opt); 107 NFSD_VNET_DEFINE_STATIC(struct vfsoptlist, nfsv4root_newopt); 108 NFSD_VNET_DEFINE_STATIC(bool, nfsrv_suspend_nfsd) = false; 109 NFSD_VNET_DEFINE_STATIC(bool, nfsrv_mntinited) = false; 110 111 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *, 112 struct ucred *); 113 static void nfsvno_updateds(struct vnode *, struct ucred *, struct thread *); 114 115 int nfsrv_enable_crossmntpt = 1; 116 static int nfs_commit_blks; 117 static int nfs_commit_miss; 118 extern int nfsrv_issuedelegs; 119 extern int nfsrv_dolocallocks; 120 extern struct nfsdevicehead nfsrv_devidhead; 121 122 /* Map d_type to vnode type. */ 123 static uint8_t dtype_to_vnode[DT_WHT + 1] = { VNON, VFIFO, VCHR, VNON, VDIR, 124 VNON, VBLK, VNON, VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON }; 125 126 static int nfsrv_createiovec(int, struct mbuf **, struct mbuf **, 127 struct iovec **); 128 static int nfsrv_createiovec_extpgs(int, int, struct mbuf **, 129 struct mbuf **, struct iovec **); 130 static int nfsrv_createiovecw(int, struct mbuf *, char *, struct iovec **, 131 int *); 132 static void nfs_dtypetovtype(struct nfsvattr *, struct vnode *, uint8_t); 133 static void nfsrv_pnfscreate(struct vnode *, struct vattr *, struct ucred *, 134 NFSPROC_T *); 135 static void nfsrv_pnfsremovesetup(struct vnode *, NFSPROC_T *, struct vnode **, 136 int *, char *, fhandle_t *); 137 static void nfsrv_pnfsremove(struct vnode **, int, char *, fhandle_t *, 138 NFSPROC_T *); 139 static int nfsrv_proxyds(struct vnode *, off_t, int, struct ucred *, 140 struct thread *, int, struct mbuf **, char *, struct mbuf **, 141 struct nfsvattr *, struct acl *, off_t *, int, bool *); 142 static int nfsrv_setextattr(struct vnode *, struct nfsvattr *, NFSPROC_T *); 143 static int nfsrv_readdsrpc(fhandle_t *, off_t, int, struct ucred *, 144 NFSPROC_T *, struct nfsmount *, struct mbuf **, struct mbuf **); 145 static int nfsrv_writedsrpc(fhandle_t *, off_t, int, struct ucred *, 146 NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct mbuf **, 147 char *, int *); 148 static int nfsrv_allocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *, 149 NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *); 150 static int nfsrv_deallocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *, 151 NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *); 152 static int nfsrv_setacldsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, 153 struct vnode *, struct nfsmount **, int, struct acl *, int *); 154 static int nfsrv_setattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, 155 struct vnode *, struct nfsmount **, int, struct nfsvattr *, int *); 156 static int nfsrv_getattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, 157 struct vnode *, struct nfsmount *, struct nfsvattr *); 158 static int nfsrv_seekdsrpc(fhandle_t *, off_t *, int, bool *, struct ucred *, 159 NFSPROC_T *, struct nfsmount *); 160 static int nfsrv_putfhname(fhandle_t *, char *); 161 static int nfsrv_pnfslookupds(struct vnode *, struct vnode *, 162 struct pnfsdsfile *, struct vnode **, NFSPROC_T *); 163 static void nfsrv_pnfssetfh(struct vnode *, struct pnfsdsfile *, char *, char *, 164 struct vnode *, NFSPROC_T *); 165 static int nfsrv_dsremove(struct vnode *, char *, struct ucred *, NFSPROC_T *); 166 static int nfsrv_dssetacl(struct vnode *, struct acl *, struct ucred *, 167 NFSPROC_T *); 168 static int nfsrv_pnfsstatfs(struct statfs *, struct mount *); 169 170 int nfs_pnfsio(task_fn_t *, void *); 171 172 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 173 "NFS server"); 174 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW, 175 &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points"); 176 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks, 177 0, ""); 178 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss, 179 0, ""); 180 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW, 181 &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations"); 182 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel, 183 0, "Debug level for NFS server"); 184 NFSD_VNET_DECLARE(int, nfsd_enable_stringtouid); 185 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, 186 CTLFLAG_NFSD_VNET | CTLFLAG_RW, &NFSD_VNET_NAME(nfsd_enable_stringtouid), 187 0, "Enable nfsd to accept numeric owner_names"); 188 static int nfsrv_pnfsgetdsattr = 1; 189 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsgetdsattr, CTLFLAG_RW, 190 &nfsrv_pnfsgetdsattr, 0, "When set getattr gets DS attributes via RPC"); 191 static bool nfsrv_recalldeleg = false; 192 SYSCTL_BOOL(_vfs_nfsd, OID_AUTO, recalldeleg, CTLFLAG_RW, 193 &nfsrv_recalldeleg, 0, 194 "When set remove/rename recalls delegations for same client"); 195 196 /* 197 * nfsrv_dsdirsize can only be increased and only when the nfsd threads are 198 * not running. 199 * The dsN subdirectories for the increased values must have been created 200 * on all DS servers before this increase is done. 201 */ 202 u_int nfsrv_dsdirsize = 20; 203 static int 204 sysctl_dsdirsize(SYSCTL_HANDLER_ARGS) 205 { 206 int error, newdsdirsize; 207 208 newdsdirsize = nfsrv_dsdirsize; 209 error = sysctl_handle_int(oidp, &newdsdirsize, 0, req); 210 if (error != 0 || req->newptr == NULL) 211 return (error); 212 if (newdsdirsize <= nfsrv_dsdirsize || newdsdirsize > 10000 || 213 newnfs_numnfsd != 0) 214 return (EINVAL); 215 nfsrv_dsdirsize = newdsdirsize; 216 return (0); 217 } 218 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, dsdirsize, 219 CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(nfsrv_dsdirsize), 220 sysctl_dsdirsize, "IU", "Number of dsN subdirs on the DS servers"); 221 222 /* 223 * nfs_srvmaxio can only be increased and only when the nfsd threads are 224 * not running. The setting must be a power of 2, with the current limit of 225 * 1Mbyte. 226 */ 227 static int 228 sysctl_srvmaxio(SYSCTL_HANDLER_ARGS) 229 { 230 int error; 231 u_int newsrvmaxio; 232 uint64_t tval; 233 234 newsrvmaxio = nfs_srvmaxio; 235 error = sysctl_handle_int(oidp, &newsrvmaxio, 0, req); 236 if (error != 0 || req->newptr == NULL) 237 return (error); 238 if (newsrvmaxio == nfs_srvmaxio) 239 return (0); 240 if (newsrvmaxio < nfs_srvmaxio) { 241 printf("nfsd: vfs.nfsd.srvmaxio can only be increased\n"); 242 return (EINVAL); 243 } 244 if (newsrvmaxio > 1048576) { 245 printf("nfsd: vfs.nfsd.srvmaxio cannot be > 1Mbyte\n"); 246 return (EINVAL); 247 } 248 if ((newsrvmaxio & (newsrvmaxio - 1)) != 0) { 249 printf("nfsd: vfs.nfsd.srvmaxio must be a power of 2\n"); 250 return (EINVAL); 251 } 252 253 /* 254 * Check that kern.ipc.maxsockbuf is large enough for 255 * newsrviomax, given the setting of vfs.nfs.bufpackets. 256 */ 257 if ((newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets > 258 sb_max_adj) { 259 /* 260 * Suggest vfs.nfs.bufpackets * maximum RPC message for 261 * sb_max_adj. 262 */ 263 tval = (newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets; 264 265 /* 266 * Convert suggested sb_max_adj value to a suggested 267 * sb_max value, which is what is set via kern.ipc.maxsockbuf. 268 * Perform the inverse calculation of (from uipc_sockbuf.c): 269 * sb_max_adj = (u_quad_t)sb_max * MCLBYTES / 270 * (MSIZE + MCLBYTES); 271 * XXX If the calculation of sb_max_adj from sb_max changes, 272 * this calculation must be changed as well. 273 */ 274 tval *= (MSIZE + MCLBYTES); /* Brackets for readability. */ 275 tval += MCLBYTES - 1; /* Round up divide. */ 276 tval /= MCLBYTES; 277 printf("nfsd: set kern.ipc.maxsockbuf to a minimum of " 278 "%ju to support %ubyte NFS I/O\n", (uintmax_t)tval, 279 newsrvmaxio); 280 return (EINVAL); 281 } 282 283 NFSD_LOCK(); 284 if (newnfs_numnfsd != 0) { 285 NFSD_UNLOCK(); 286 printf("nfsd: cannot set vfs.nfsd.srvmaxio when nfsd " 287 "threads are running\n"); 288 return (EINVAL); 289 } 290 291 292 nfs_srvmaxio = newsrvmaxio; 293 NFSD_UNLOCK(); 294 return (0); 295 } 296 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, srvmaxio, 297 CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0, 298 sysctl_srvmaxio, "IU", "Maximum I/O size in bytes"); 299 300 static int 301 sysctl_dolocallocks(SYSCTL_HANDLER_ARGS) 302 { 303 int error, igotlock, newdolocallocks; 304 305 newdolocallocks = nfsrv_dolocallocks; 306 error = sysctl_handle_int(oidp, &newdolocallocks, 0, req); 307 if (error != 0 || req->newptr == NULL) 308 return (error); 309 if (newdolocallocks == nfsrv_dolocallocks) 310 return (0); 311 if (jailed(curthread->td_ucred)) 312 return (EINVAL); 313 314 NFSLOCKV4ROOTMUTEX(); 315 do { 316 igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, 317 NFSV4ROOTLOCKMUTEXPTR, NULL); 318 } while (!igotlock); 319 NFSUNLOCKV4ROOTMUTEX(); 320 321 nfsrv_dolocallocks = newdolocallocks; 322 323 NFSLOCKV4ROOTMUTEX(); 324 nfsv4_unlock(&nfsv4rootfs_lock, 0); 325 NFSUNLOCKV4ROOTMUTEX(); 326 return (0); 327 } 328 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, enable_locallocks, 329 CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0, 330 sysctl_dolocallocks, "IU", "Enable nfsd to acquire local locks on files"); 331 332 #define MAX_REORDERED_RPC 16 333 #define NUM_HEURISTIC 1031 334 #define NHUSE_INIT 64 335 #define NHUSE_INC 16 336 #define NHUSE_MAX 2048 337 338 static struct nfsheur { 339 struct vnode *nh_vp; /* vp to match (unreferenced pointer) */ 340 off_t nh_nextoff; /* next offset for sequential detection */ 341 int nh_use; /* use count for selection */ 342 int nh_seqcount; /* heuristic */ 343 } nfsheur[NUM_HEURISTIC]; 344 345 /* 346 * Heuristic to detect sequential operation. 347 */ 348 static struct nfsheur * 349 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp) 350 { 351 struct nfsheur *nh; 352 int hi, try; 353 354 /* Locate best candidate. */ 355 try = 32; 356 hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC; 357 nh = &nfsheur[hi]; 358 while (try--) { 359 if (nfsheur[hi].nh_vp == vp) { 360 nh = &nfsheur[hi]; 361 break; 362 } 363 if (nfsheur[hi].nh_use > 0) 364 --nfsheur[hi].nh_use; 365 hi = (hi + 1) % NUM_HEURISTIC; 366 if (nfsheur[hi].nh_use < nh->nh_use) 367 nh = &nfsheur[hi]; 368 } 369 370 /* Initialize hint if this is a new file. */ 371 if (nh->nh_vp != vp) { 372 nh->nh_vp = vp; 373 nh->nh_nextoff = uio->uio_offset; 374 nh->nh_use = NHUSE_INIT; 375 if (uio->uio_offset == 0) 376 nh->nh_seqcount = 4; 377 else 378 nh->nh_seqcount = 1; 379 } 380 381 /* Calculate heuristic. */ 382 if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) || 383 uio->uio_offset == nh->nh_nextoff) { 384 /* See comments in vfs_vnops.c:sequential_heuristic(). */ 385 nh->nh_seqcount += howmany(uio->uio_resid, 16384); 386 if (nh->nh_seqcount > IO_SEQMAX) 387 nh->nh_seqcount = IO_SEQMAX; 388 } else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC * 389 imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) { 390 /* Probably a reordered RPC, leave seqcount alone. */ 391 } else if (nh->nh_seqcount > 1) { 392 nh->nh_seqcount /= 2; 393 } else { 394 nh->nh_seqcount = 0; 395 } 396 nh->nh_use += NHUSE_INC; 397 if (nh->nh_use > NHUSE_MAX) 398 nh->nh_use = NHUSE_MAX; 399 return (nh); 400 } 401 402 /* 403 * Get attributes into nfsvattr structure. 404 */ 405 int 406 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, 407 struct nfsrv_descript *nd, struct thread *p, int vpislocked, 408 nfsattrbit_t *attrbitp) 409 { 410 int error, gotattr, lockedit = 0; 411 struct nfsvattr na; 412 413 if (vpislocked == 0) { 414 /* 415 * When vpislocked == 0, the vnode is either exclusively 416 * locked by this thread or not locked by this thread. 417 * As such, shared lock it, if not exclusively locked. 418 */ 419 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) { 420 lockedit = 1; 421 NFSVOPLOCK(vp, LK_SHARED | LK_RETRY); 422 } 423 } 424 425 /* 426 * Acquire the Change, Size, TimeAccess, TimeModify and SpaceUsed 427 * attributes, as required. 428 * This needs to be done for regular files if: 429 * - non-NFSv4 RPCs or 430 * - when attrbitp == NULL or 431 * - an NFSv4 RPC with any of the above attributes in attrbitp. 432 * A return of 0 for nfsrv_proxyds() indicates that it has acquired 433 * these attributes. nfsrv_proxyds() will return an error if the 434 * server is not a pNFS one. 435 */ 436 gotattr = 0; 437 if (vp->v_type == VREG && nfsrv_devidcnt > 0 && (attrbitp == NULL || 438 (nd->nd_flag & ND_NFSV4) == 0 || 439 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_CHANGE) || 440 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE) || 441 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS) || 442 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEMODIFY) || 443 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEUSED))) { 444 error = nfsrv_proxyds(vp, 0, 0, nd->nd_cred, p, 445 NFSPROC_GETATTR, NULL, NULL, NULL, &na, NULL, NULL, 0, 446 NULL); 447 if (error == 0) 448 gotattr = 1; 449 } 450 451 nvap->na_bsdflags = 0; 452 nvap->na_flags = 0; 453 error = VOP_GETATTR(vp, &nvap->na_vattr, nd->nd_cred); 454 if (lockedit != 0) 455 NFSVOPUNLOCK(vp); 456 457 /* 458 * If we got the Change, Size and Modify Time from the DS, 459 * replace them. 460 */ 461 if (gotattr != 0) { 462 nvap->na_atime = na.na_atime; 463 nvap->na_mtime = na.na_mtime; 464 nvap->na_filerev = na.na_filerev; 465 nvap->na_size = na.na_size; 466 nvap->na_bytes = na.na_bytes; 467 } 468 NFSD_DEBUG(4, "nfsvno_getattr: gotattr=%d err=%d chg=%ju\n", gotattr, 469 error, (uintmax_t)na.na_filerev); 470 471 NFSEXITCODE(error); 472 return (error); 473 } 474 475 /* 476 * Get a file handle for a vnode. 477 */ 478 int 479 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p) 480 { 481 int error; 482 483 NFSBZERO((caddr_t)fhp, sizeof(fhandle_t)); 484 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid; 485 error = VOP_VPTOFH(vp, &fhp->fh_fid); 486 487 NFSEXITCODE(error); 488 return (error); 489 } 490 491 /* 492 * Perform access checking for vnodes obtained from file handles that would 493 * refer to files already opened by a Unix client. You cannot just use 494 * vn_writechk() and VOP_ACCESSX() for two reasons. 495 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write 496 * case. 497 * 2 - The owner is to be given access irrespective of mode bits for some 498 * operations, so that processes that chmod after opening a file don't 499 * break. 500 */ 501 int 502 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred, 503 struct nfsexstuff *exp, struct thread *p, int override, int vpislocked, 504 u_int32_t *supportedtypep) 505 { 506 struct vattr vattr; 507 int error = 0, getret = 0; 508 509 if (vpislocked == 0) { 510 if (NFSVOPLOCK(vp, LK_SHARED) != 0) { 511 error = EPERM; 512 goto out; 513 } 514 } 515 if (accmode & VWRITE) { 516 /* Just vn_writechk() changed to check rdonly */ 517 /* 518 * Disallow write attempts on read-only file systems; 519 * unless the file is a socket or a block or character 520 * device resident on the file system. 521 */ 522 if (NFSVNO_EXRDONLY(exp) || 523 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 524 switch (vp->v_type) { 525 case VREG: 526 case VDIR: 527 case VLNK: 528 error = EROFS; 529 default: 530 break; 531 } 532 } 533 /* 534 * If there's shared text associated with 535 * the inode, try to free it up once. If 536 * we fail, we can't allow writing. 537 */ 538 if (VOP_IS_TEXT(vp) && error == 0) 539 error = ETXTBSY; 540 } 541 if (error != 0) { 542 if (vpislocked == 0) 543 NFSVOPUNLOCK(vp); 544 goto out; 545 } 546 547 /* 548 * Should the override still be applied when ACLs are enabled? 549 */ 550 error = VOP_ACCESSX(vp, accmode, cred, p); 551 if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) { 552 /* 553 * Try again with VEXPLICIT_DENY, to see if the test for 554 * deletion is supported. 555 */ 556 error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p); 557 if (error == 0) { 558 if (vp->v_type == VDIR) { 559 accmode &= ~(VDELETE | VDELETE_CHILD); 560 accmode |= VWRITE; 561 error = VOP_ACCESSX(vp, accmode, cred, p); 562 } else if (supportedtypep != NULL) { 563 *supportedtypep &= ~NFSACCESS_DELETE; 564 } 565 } 566 } 567 568 /* 569 * Allow certain operations for the owner (reads and writes 570 * on files that are already open). 571 */ 572 if (override != NFSACCCHK_NOOVERRIDE && 573 (error == EPERM || error == EACCES)) { 574 if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT)) 575 error = 0; 576 else if (override & NFSACCCHK_ALLOWOWNER) { 577 getret = VOP_GETATTR(vp, &vattr, cred); 578 if (getret == 0 && cred->cr_uid == vattr.va_uid) 579 error = 0; 580 } 581 } 582 if (vpislocked == 0) 583 NFSVOPUNLOCK(vp); 584 585 out: 586 NFSEXITCODE(error); 587 return (error); 588 } 589 590 /* 591 * Set attribute(s) vnop. 592 */ 593 int 594 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred, 595 struct thread *p, struct nfsexstuff *exp) 596 { 597 u_quad_t savsize = 0; 598 int error, savedit; 599 time_t savbtime; 600 601 /* 602 * If this is an exported file system and a pNFS service is running, 603 * don't VOP_SETATTR() of size for the MDS file system. 604 */ 605 savedit = 0; 606 error = 0; 607 if (vp->v_type == VREG && (vp->v_mount->mnt_flag & MNT_EXPORTED) != 0 && 608 nfsrv_devidcnt != 0 && nvap->na_vattr.va_size != VNOVAL && 609 nvap->na_vattr.va_size > 0) { 610 savsize = nvap->na_vattr.va_size; 611 nvap->na_vattr.va_size = VNOVAL; 612 if (nvap->na_vattr.va_uid != (uid_t)VNOVAL || 613 nvap->na_vattr.va_gid != (gid_t)VNOVAL || 614 nvap->na_vattr.va_mode != (mode_t)VNOVAL || 615 nvap->na_vattr.va_atime.tv_sec != VNOVAL || 616 nvap->na_vattr.va_mtime.tv_sec != VNOVAL) 617 savedit = 1; 618 else 619 savedit = 2; 620 } 621 if (savedit != 2) 622 error = VOP_SETATTR(vp, &nvap->na_vattr, cred); 623 if (savedit != 0) 624 nvap->na_vattr.va_size = savsize; 625 if (error == 0 && (nvap->na_vattr.va_uid != (uid_t)VNOVAL || 626 nvap->na_vattr.va_gid != (gid_t)VNOVAL || 627 nvap->na_vattr.va_size != VNOVAL || 628 nvap->na_vattr.va_mode != (mode_t)VNOVAL || 629 nvap->na_vattr.va_atime.tv_sec != VNOVAL || 630 nvap->na_vattr.va_mtime.tv_sec != VNOVAL)) { 631 /* Never modify birthtime on a DS file. */ 632 savbtime = nvap->na_vattr.va_birthtime.tv_sec; 633 nvap->na_vattr.va_birthtime.tv_sec = VNOVAL; 634 /* For a pNFS server, set the attributes on the DS file. */ 635 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETATTR, 636 NULL, NULL, NULL, nvap, NULL, NULL, 0, NULL); 637 nvap->na_vattr.va_birthtime.tv_sec = savbtime; 638 if (error == ENOENT) 639 error = 0; 640 } 641 NFSEXITCODE(error); 642 return (error); 643 } 644 645 /* 646 * Set up nameidata for a lookup() call and do it. 647 */ 648 int 649 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp, 650 struct vnode *dp, int islocked, struct nfsexstuff *exp, 651 struct vnode **retdirp) 652 { 653 struct componentname *cnp = &ndp->ni_cnd; 654 int i; 655 struct iovec aiov; 656 struct uio auio; 657 int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen; 658 int error = 0; 659 char *cp; 660 661 *retdirp = NULL; 662 cnp->cn_nameptr = cnp->cn_pnbuf; 663 ndp->ni_lcf = 0; 664 /* 665 * Extract and set starting directory. 666 */ 667 if (dp->v_type != VDIR) { 668 if (islocked) 669 vput(dp); 670 else 671 vrele(dp); 672 nfsvno_relpathbuf(ndp); 673 error = ENOTDIR; 674 goto out1; 675 } 676 if (islocked) 677 NFSVOPUNLOCK(dp); 678 VREF(dp); 679 *retdirp = dp; 680 if (NFSVNO_EXRDONLY(exp)) 681 cnp->cn_flags |= RDONLY; 682 ndp->ni_segflg = UIO_SYSSPACE; 683 684 if (nd->nd_flag & ND_PUBLOOKUP) { 685 ndp->ni_loopcnt = 0; 686 if (cnp->cn_pnbuf[0] == '/') { 687 vrele(dp); 688 /* 689 * Check for degenerate pathnames here, since lookup() 690 * panics on them. 691 */ 692 for (i = 1; i < ndp->ni_pathlen; i++) 693 if (cnp->cn_pnbuf[i] != '/') 694 break; 695 if (i == ndp->ni_pathlen) { 696 error = NFSERR_ACCES; 697 goto out; 698 } 699 dp = rootvnode; 700 VREF(dp); 701 } 702 } else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) || 703 (nd->nd_flag & ND_NFSV4) == 0) { 704 /* 705 * Only cross mount points for NFSv4 when doing a 706 * mount while traversing the file system above 707 * the mount point, unless nfsrv_enable_crossmntpt is set. 708 */ 709 cnp->cn_flags |= NOCROSSMOUNT; 710 } 711 712 /* 713 * Initialize for scan, set ni_startdir and bump ref on dp again 714 * because lookup() will dereference ni_startdir. 715 */ 716 717 ndp->ni_startdir = dp; 718 ndp->ni_rootdir = rootvnode; 719 ndp->ni_topdir = NULL; 720 721 if (!lockleaf) 722 cnp->cn_flags |= LOCKLEAF; 723 for (;;) { 724 cnp->cn_nameptr = cnp->cn_pnbuf; 725 /* 726 * Call lookup() to do the real work. If an error occurs, 727 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and 728 * we do not have to dereference anything before returning. 729 * In either case ni_startdir will be dereferenced and NULLed 730 * out. 731 */ 732 error = vfs_lookup(ndp); 733 if (error) 734 break; 735 736 /* 737 * Check for encountering a symbolic link. Trivial 738 * termination occurs if no symlink encountered. 739 */ 740 if ((cnp->cn_flags & ISSYMLINK) == 0) { 741 if (ndp->ni_vp && !lockleaf) 742 NFSVOPUNLOCK(ndp->ni_vp); 743 break; 744 } 745 746 /* 747 * Validate symlink 748 */ 749 if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1) 750 NFSVOPUNLOCK(ndp->ni_dvp); 751 if (!(nd->nd_flag & ND_PUBLOOKUP)) { 752 error = EINVAL; 753 goto badlink2; 754 } 755 756 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) { 757 error = ELOOP; 758 goto badlink2; 759 } 760 if (ndp->ni_pathlen > 1) 761 cp = uma_zalloc(namei_zone, M_WAITOK); 762 else 763 cp = cnp->cn_pnbuf; 764 aiov.iov_base = cp; 765 aiov.iov_len = MAXPATHLEN; 766 auio.uio_iov = &aiov; 767 auio.uio_iovcnt = 1; 768 auio.uio_offset = 0; 769 auio.uio_rw = UIO_READ; 770 auio.uio_segflg = UIO_SYSSPACE; 771 auio.uio_td = NULL; 772 auio.uio_resid = MAXPATHLEN; 773 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred); 774 if (error) { 775 badlink1: 776 if (ndp->ni_pathlen > 1) 777 uma_zfree(namei_zone, cp); 778 badlink2: 779 vrele(ndp->ni_dvp); 780 vput(ndp->ni_vp); 781 break; 782 } 783 linklen = MAXPATHLEN - auio.uio_resid; 784 if (linklen == 0) { 785 error = ENOENT; 786 goto badlink1; 787 } 788 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) { 789 error = ENAMETOOLONG; 790 goto badlink1; 791 } 792 793 /* 794 * Adjust or replace path 795 */ 796 if (ndp->ni_pathlen > 1) { 797 NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen); 798 uma_zfree(namei_zone, cnp->cn_pnbuf); 799 cnp->cn_pnbuf = cp; 800 } else 801 cnp->cn_pnbuf[linklen] = '\0'; 802 ndp->ni_pathlen += linklen; 803 804 /* 805 * Cleanup refs for next loop and check if root directory 806 * should replace current directory. Normally ni_dvp 807 * becomes the new base directory and is cleaned up when 808 * we loop. Explicitly null pointers after invalidation 809 * to clarify operation. 810 */ 811 vput(ndp->ni_vp); 812 ndp->ni_vp = NULL; 813 814 if (cnp->cn_pnbuf[0] == '/') { 815 vrele(ndp->ni_dvp); 816 ndp->ni_dvp = ndp->ni_rootdir; 817 VREF(ndp->ni_dvp); 818 } 819 ndp->ni_startdir = ndp->ni_dvp; 820 ndp->ni_dvp = NULL; 821 } 822 if (!lockleaf) 823 cnp->cn_flags &= ~LOCKLEAF; 824 825 out: 826 if (error) { 827 nfsvno_relpathbuf(ndp); 828 ndp->ni_vp = NULL; 829 ndp->ni_dvp = NULL; 830 ndp->ni_startdir = NULL; 831 } else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) { 832 ndp->ni_dvp = NULL; 833 } 834 835 out1: 836 NFSEXITCODE2(error, nd); 837 return (error); 838 } 839 840 /* 841 * Set up a pathname buffer and return a pointer to it and, optionally 842 * set a hash pointer. 843 */ 844 void 845 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp) 846 { 847 struct componentname *cnp = &ndp->ni_cnd; 848 849 cnp->cn_flags |= (NOMACCHECK); 850 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK); 851 if (hashpp != NULL) 852 *hashpp = NULL; 853 *bufpp = cnp->cn_pnbuf; 854 } 855 856 /* 857 * Release the above path buffer, if not released by nfsvno_namei(). 858 */ 859 void 860 nfsvno_relpathbuf(struct nameidata *ndp) 861 { 862 863 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf); 864 ndp->ni_cnd.cn_pnbuf = NULL; 865 } 866 867 /* 868 * Readlink vnode op into an mbuf list. 869 */ 870 int 871 nfsvno_readlink(struct vnode *vp, struct ucred *cred, int maxextsiz, 872 struct thread *p, struct mbuf **mpp, struct mbuf **mpendp, int *lenp) 873 { 874 struct iovec *iv; 875 struct uio io, *uiop = &io; 876 struct mbuf *mp, *mp3; 877 int len, tlen, error = 0; 878 879 len = NFS_MAXPATHLEN; 880 if (maxextsiz > 0) 881 uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz, 882 &mp3, &mp, &iv); 883 else 884 uiop->uio_iovcnt = nfsrv_createiovec(len, &mp3, &mp, &iv); 885 uiop->uio_iov = iv; 886 uiop->uio_offset = 0; 887 uiop->uio_resid = len; 888 uiop->uio_rw = UIO_READ; 889 uiop->uio_segflg = UIO_SYSSPACE; 890 uiop->uio_td = NULL; 891 error = VOP_READLINK(vp, uiop, cred); 892 free(iv, M_TEMP); 893 if (error) { 894 m_freem(mp3); 895 *lenp = 0; 896 goto out; 897 } 898 if (uiop->uio_resid > 0) { 899 len -= uiop->uio_resid; 900 tlen = NFSM_RNDUP(len); 901 if (tlen == 0) { 902 m_freem(mp3); 903 mp3 = mp = NULL; 904 } else if (tlen != NFS_MAXPATHLEN || tlen != len) 905 mp = nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, 906 tlen - len); 907 } 908 *lenp = len; 909 *mpp = mp3; 910 *mpendp = mp; 911 912 out: 913 NFSEXITCODE(error); 914 return (error); 915 } 916 917 /* 918 * Create an mbuf chain and an associated iovec that can be used to Read 919 * or Getextattr of data. 920 * Upon success, return pointers to the first and last mbufs in the chain 921 * plus the malloc'd iovec and its iovlen. 922 */ 923 static int 924 nfsrv_createiovec(int len, struct mbuf **mpp, struct mbuf **mpendp, 925 struct iovec **ivp) 926 { 927 struct mbuf *m, *m2 = NULL, *m3; 928 struct iovec *iv; 929 int i, left, siz; 930 931 left = len; 932 m3 = NULL; 933 /* 934 * Generate the mbuf list with the uio_iov ref. to it. 935 */ 936 i = 0; 937 while (left > 0) { 938 NFSMGET(m); 939 MCLGET(m, M_WAITOK); 940 m->m_len = 0; 941 siz = min(M_TRAILINGSPACE(m), left); 942 left -= siz; 943 i++; 944 if (m3) 945 m2->m_next = m; 946 else 947 m3 = m; 948 m2 = m; 949 } 950 *ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK); 951 m = m3; 952 left = len; 953 i = 0; 954 while (left > 0) { 955 if (m == NULL) 956 panic("nfsrv_createiovec iov"); 957 siz = min(M_TRAILINGSPACE(m), left); 958 if (siz > 0) { 959 iv->iov_base = mtod(m, caddr_t) + m->m_len; 960 iv->iov_len = siz; 961 m->m_len += siz; 962 left -= siz; 963 iv++; 964 i++; 965 } 966 m = m->m_next; 967 } 968 *mpp = m3; 969 *mpendp = m2; 970 return (i); 971 } 972 973 /* 974 * Create an mbuf chain and an associated iovec that can be used to Read 975 * or Getextattr of data. 976 * Upon success, return pointers to the first and last mbufs in the chain 977 * plus the malloc'd iovec and its iovlen. 978 * Same as above, but creates ext_pgs mbuf(s). 979 */ 980 static int 981 nfsrv_createiovec_extpgs(int len, int maxextsiz, struct mbuf **mpp, 982 struct mbuf **mpendp, struct iovec **ivp) 983 { 984 struct mbuf *m, *m2 = NULL, *m3; 985 struct iovec *iv; 986 int i, left, pgno, siz; 987 988 left = len; 989 m3 = NULL; 990 /* 991 * Generate the mbuf list with the uio_iov ref. to it. 992 */ 993 i = 0; 994 while (left > 0) { 995 siz = min(left, maxextsiz); 996 m = mb_alloc_ext_plus_pages(siz, M_WAITOK); 997 left -= siz; 998 i += m->m_epg_npgs; 999 if (m3 != NULL) 1000 m2->m_next = m; 1001 else 1002 m3 = m; 1003 m2 = m; 1004 } 1005 *ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK); 1006 m = m3; 1007 left = len; 1008 i = 0; 1009 pgno = 0; 1010 while (left > 0) { 1011 if (m == NULL) 1012 panic("nfsvno_createiovec_extpgs iov"); 1013 siz = min(PAGE_SIZE, left); 1014 if (siz > 0) { 1015 iv->iov_base = (void *)PHYS_TO_DMAP(m->m_epg_pa[pgno]); 1016 iv->iov_len = siz; 1017 m->m_len += siz; 1018 if (pgno == m->m_epg_npgs - 1) 1019 m->m_epg_last_len = siz; 1020 left -= siz; 1021 iv++; 1022 i++; 1023 pgno++; 1024 } 1025 if (pgno == m->m_epg_npgs && left > 0) { 1026 m = m->m_next; 1027 if (m == NULL) 1028 panic("nfsvno_createiovec_extpgs iov"); 1029 pgno = 0; 1030 } 1031 } 1032 *mpp = m3; 1033 *mpendp = m2; 1034 return (i); 1035 } 1036 1037 /* 1038 * Read vnode op call into mbuf list. 1039 */ 1040 int 1041 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred, 1042 int maxextsiz, struct thread *p, struct mbuf **mpp, 1043 struct mbuf **mpendp) 1044 { 1045 struct mbuf *m; 1046 struct iovec *iv; 1047 int error = 0, len, tlen, ioflag = 0; 1048 struct mbuf *m3; 1049 struct uio io, *uiop = &io; 1050 struct nfsheur *nh; 1051 1052 /* 1053 * Attempt to read from a DS file. A return of ENOENT implies 1054 * there is no DS file to read. 1055 */ 1056 error = nfsrv_proxyds(vp, off, cnt, cred, p, NFSPROC_READDS, mpp, 1057 NULL, mpendp, NULL, NULL, NULL, 0, NULL); 1058 if (error != ENOENT) 1059 return (error); 1060 1061 len = NFSM_RNDUP(cnt); 1062 if (maxextsiz > 0) 1063 uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz, 1064 &m3, &m, &iv); 1065 else 1066 uiop->uio_iovcnt = nfsrv_createiovec(len, &m3, &m, &iv); 1067 uiop->uio_iov = iv; 1068 uiop->uio_offset = off; 1069 uiop->uio_resid = len; 1070 uiop->uio_rw = UIO_READ; 1071 uiop->uio_segflg = UIO_SYSSPACE; 1072 uiop->uio_td = NULL; 1073 nh = nfsrv_sequential_heuristic(uiop, vp); 1074 ioflag |= nh->nh_seqcount << IO_SEQSHIFT; 1075 /* XXX KDM make this more systematic? */ 1076 NFSD_VNET(nfsstatsv1_p)->srvbytes[NFSV4OP_READ] += uiop->uio_resid; 1077 error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred); 1078 free(iv, M_TEMP); 1079 if (error) { 1080 m_freem(m3); 1081 *mpp = NULL; 1082 goto out; 1083 } 1084 nh->nh_nextoff = uiop->uio_offset; 1085 tlen = len - uiop->uio_resid; 1086 cnt = cnt < tlen ? cnt : tlen; 1087 tlen = NFSM_RNDUP(cnt); 1088 if (tlen == 0) { 1089 m_freem(m3); 1090 m3 = m = NULL; 1091 } else if (len != tlen || tlen != cnt) 1092 m = nfsrv_adj(m3, len - tlen, tlen - cnt); 1093 *mpp = m3; 1094 *mpendp = m; 1095 1096 out: 1097 NFSEXITCODE(error); 1098 return (error); 1099 } 1100 1101 /* 1102 * Create the iovec for the mbuf chain passed in as an argument. 1103 * The "cp" argument is where the data starts within the first mbuf in 1104 * the chain. It returns the iovec and the iovcnt. 1105 */ 1106 static int 1107 nfsrv_createiovecw(int retlen, struct mbuf *m, char *cp, struct iovec **ivpp, 1108 int *iovcntp) 1109 { 1110 struct mbuf *mp; 1111 struct iovec *ivp; 1112 int cnt, i, len; 1113 1114 /* 1115 * Loop through the mbuf chain, counting how many mbufs are a 1116 * part of this write operation, so the iovec size is known. 1117 */ 1118 cnt = 0; 1119 len = retlen; 1120 mp = m; 1121 i = mtod(mp, caddr_t) + mp->m_len - cp; 1122 while (len > 0) { 1123 if (i > 0) { 1124 len -= i; 1125 cnt++; 1126 } 1127 mp = mp->m_next; 1128 if (!mp) { 1129 if (len > 0) 1130 return (EBADRPC); 1131 } else 1132 i = mp->m_len; 1133 } 1134 1135 /* Now, create the iovec. */ 1136 mp = m; 1137 *ivpp = ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, 1138 M_WAITOK); 1139 *iovcntp = cnt; 1140 i = mtod(mp, caddr_t) + mp->m_len - cp; 1141 len = retlen; 1142 while (len > 0) { 1143 if (mp == NULL) 1144 panic("nfsrv_createiovecw"); 1145 if (i > 0) { 1146 i = min(i, len); 1147 ivp->iov_base = cp; 1148 ivp->iov_len = i; 1149 ivp++; 1150 len -= i; 1151 } 1152 mp = mp->m_next; 1153 if (mp) { 1154 i = mp->m_len; 1155 cp = mtod(mp, caddr_t); 1156 } 1157 } 1158 return (0); 1159 } 1160 1161 /* 1162 * Write vnode op from an mbuf list. 1163 */ 1164 int 1165 nfsvno_write(struct vnode *vp, off_t off, int retlen, int *stable, 1166 struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p) 1167 { 1168 struct iovec *iv; 1169 int cnt, ioflags, error; 1170 struct uio io, *uiop = &io; 1171 struct nfsheur *nh; 1172 1173 /* 1174 * Attempt to write to a DS file. A return of ENOENT implies 1175 * there is no DS file to write. 1176 */ 1177 error = nfsrv_proxyds(vp, off, retlen, cred, p, NFSPROC_WRITEDS, 1178 &mp, cp, NULL, NULL, NULL, NULL, 0, NULL); 1179 if (error != ENOENT) { 1180 *stable = NFSWRITE_FILESYNC; 1181 return (error); 1182 } 1183 1184 if (*stable == NFSWRITE_UNSTABLE) 1185 ioflags = IO_NODELOCKED; 1186 else 1187 ioflags = (IO_SYNC | IO_NODELOCKED); 1188 error = nfsrv_createiovecw(retlen, mp, cp, &iv, &cnt); 1189 if (error != 0) 1190 return (error); 1191 uiop->uio_iov = iv; 1192 uiop->uio_iovcnt = cnt; 1193 uiop->uio_resid = retlen; 1194 uiop->uio_rw = UIO_WRITE; 1195 uiop->uio_segflg = UIO_SYSSPACE; 1196 NFSUIOPROC(uiop, p); 1197 uiop->uio_offset = off; 1198 nh = nfsrv_sequential_heuristic(uiop, vp); 1199 ioflags |= nh->nh_seqcount << IO_SEQSHIFT; 1200 /* XXX KDM make this more systematic? */ 1201 NFSD_VNET(nfsstatsv1_p)->srvbytes[NFSV4OP_WRITE] += uiop->uio_resid; 1202 error = VOP_WRITE(vp, uiop, ioflags, cred); 1203 if (error == 0) 1204 nh->nh_nextoff = uiop->uio_offset; 1205 free(iv, M_TEMP); 1206 1207 NFSEXITCODE(error); 1208 return (error); 1209 } 1210 1211 /* 1212 * Common code for creating a regular file (plus special files for V2). 1213 */ 1214 int 1215 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp, 1216 struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp, 1217 int32_t *cverf, NFSDEV_T rdev, struct nfsexstuff *exp) 1218 { 1219 u_quad_t tempsize; 1220 int error; 1221 struct thread *p = curthread; 1222 1223 error = nd->nd_repstat; 1224 if (!error && ndp->ni_vp == NULL) { 1225 if (nvap->na_type == VREG || nvap->na_type == VSOCK) { 1226 error = VOP_CREATE(ndp->ni_dvp, 1227 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); 1228 /* For a pNFS server, create the data file on a DS. */ 1229 if (error == 0 && nvap->na_type == VREG) { 1230 /* 1231 * Create a data file on a DS for a pNFS server. 1232 * This function just returns if not 1233 * running a pNFS DS or the creation fails. 1234 */ 1235 nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr, 1236 nd->nd_cred, p); 1237 } 1238 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : 1239 NULL, false); 1240 nfsvno_relpathbuf(ndp); 1241 if (!error) { 1242 if (*exclusive_flagp) { 1243 *exclusive_flagp = 0; 1244 NFSVNO_ATTRINIT(nvap); 1245 nvap->na_atime.tv_sec = cverf[0]; 1246 nvap->na_atime.tv_nsec = cverf[1]; 1247 error = VOP_SETATTR(ndp->ni_vp, 1248 &nvap->na_vattr, nd->nd_cred); 1249 if (error != 0) { 1250 vput(ndp->ni_vp); 1251 ndp->ni_vp = NULL; 1252 error = NFSERR_NOTSUPP; 1253 } 1254 } 1255 } 1256 /* 1257 * NFS V2 Only. nfsrvd_mknod() does this for V3. 1258 * (This implies, just get out on an error.) 1259 */ 1260 } else if (nvap->na_type == VCHR || nvap->na_type == VBLK || 1261 nvap->na_type == VFIFO) { 1262 if (nvap->na_type == VCHR && rdev == 0xffffffff) 1263 nvap->na_type = VFIFO; 1264 if (nvap->na_type != VFIFO && 1265 (error = priv_check_cred(nd->nd_cred, PRIV_VFS_MKNOD_DEV))) { 1266 nfsvno_relpathbuf(ndp); 1267 vput(ndp->ni_dvp); 1268 goto out; 1269 } 1270 nvap->na_rdev = rdev; 1271 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp, 1272 &ndp->ni_cnd, &nvap->na_vattr); 1273 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : 1274 NULL, false); 1275 nfsvno_relpathbuf(ndp); 1276 if (error) 1277 goto out; 1278 } else { 1279 nfsvno_relpathbuf(ndp); 1280 vput(ndp->ni_dvp); 1281 error = ENXIO; 1282 goto out; 1283 } 1284 *vpp = ndp->ni_vp; 1285 } else { 1286 /* 1287 * Handle cases where error is already set and/or 1288 * the file exists. 1289 * 1 - clean up the lookup 1290 * 2 - iff !error and na_size set, truncate it 1291 */ 1292 nfsvno_relpathbuf(ndp); 1293 *vpp = ndp->ni_vp; 1294 if (ndp->ni_dvp == *vpp) 1295 vrele(ndp->ni_dvp); 1296 else 1297 vput(ndp->ni_dvp); 1298 if (!error && nvap->na_size != VNOVAL) { 1299 error = nfsvno_accchk(*vpp, VWRITE, 1300 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, 1301 NFSACCCHK_VPISLOCKED, NULL); 1302 if (!error) { 1303 tempsize = nvap->na_size; 1304 NFSVNO_ATTRINIT(nvap); 1305 nvap->na_size = tempsize; 1306 error = nfsvno_setattr(*vpp, nvap, 1307 nd->nd_cred, p, exp); 1308 } 1309 } 1310 if (error) 1311 vput(*vpp); 1312 } 1313 1314 out: 1315 NFSEXITCODE(error); 1316 return (error); 1317 } 1318 1319 /* 1320 * Do a mknod vnode op. 1321 */ 1322 int 1323 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred, 1324 struct thread *p) 1325 { 1326 int error = 0; 1327 __enum_uint8(vtype) vtyp; 1328 1329 vtyp = nvap->na_type; 1330 /* 1331 * Iff doesn't exist, create it. 1332 */ 1333 if (ndp->ni_vp) { 1334 nfsvno_relpathbuf(ndp); 1335 vput(ndp->ni_dvp); 1336 vrele(ndp->ni_vp); 1337 error = EEXIST; 1338 goto out; 1339 } 1340 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) { 1341 nfsvno_relpathbuf(ndp); 1342 vput(ndp->ni_dvp); 1343 error = NFSERR_BADTYPE; 1344 goto out; 1345 } 1346 if (vtyp == VSOCK) { 1347 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, 1348 &ndp->ni_cnd, &nvap->na_vattr); 1349 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, 1350 false); 1351 nfsvno_relpathbuf(ndp); 1352 } else { 1353 if (nvap->na_type != VFIFO && 1354 (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV))) { 1355 nfsvno_relpathbuf(ndp); 1356 vput(ndp->ni_dvp); 1357 goto out; 1358 } 1359 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp, 1360 &ndp->ni_cnd, &nvap->na_vattr); 1361 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, 1362 false); 1363 nfsvno_relpathbuf(ndp); 1364 /* 1365 * Since VOP_MKNOD returns the ni_vp, I can't 1366 * see any reason to do the lookup. 1367 */ 1368 } 1369 1370 out: 1371 NFSEXITCODE(error); 1372 return (error); 1373 } 1374 1375 /* 1376 * Mkdir vnode op. 1377 */ 1378 int 1379 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid, 1380 struct ucred *cred, struct thread *p, struct nfsexstuff *exp) 1381 { 1382 int error = 0; 1383 1384 if (ndp->ni_vp != NULL) { 1385 if (ndp->ni_dvp == ndp->ni_vp) 1386 vrele(ndp->ni_dvp); 1387 else 1388 vput(ndp->ni_dvp); 1389 vrele(ndp->ni_vp); 1390 nfsvno_relpathbuf(ndp); 1391 error = EEXIST; 1392 goto out; 1393 } 1394 error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, 1395 &nvap->na_vattr); 1396 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, false); 1397 nfsvno_relpathbuf(ndp); 1398 1399 out: 1400 NFSEXITCODE(error); 1401 return (error); 1402 } 1403 1404 /* 1405 * symlink vnode op. 1406 */ 1407 int 1408 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp, 1409 int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p, 1410 struct nfsexstuff *exp) 1411 { 1412 int error = 0; 1413 1414 if (ndp->ni_vp) { 1415 nfsvno_relpathbuf(ndp); 1416 if (ndp->ni_dvp == ndp->ni_vp) 1417 vrele(ndp->ni_dvp); 1418 else 1419 vput(ndp->ni_dvp); 1420 vrele(ndp->ni_vp); 1421 error = EEXIST; 1422 goto out; 1423 } 1424 1425 error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, 1426 &nvap->na_vattr, pathcp); 1427 /* 1428 * Although FreeBSD still had the lookup code in 1429 * it for 7/current, there doesn't seem to be any 1430 * point, since VOP_SYMLINK() returns the ni_vp. 1431 * Just vput it for v2. 1432 */ 1433 VOP_VPUT_PAIR(ndp->ni_dvp, &ndp->ni_vp, !not_v2 && error == 0); 1434 nfsvno_relpathbuf(ndp); 1435 1436 out: 1437 NFSEXITCODE(error); 1438 return (error); 1439 } 1440 1441 /* 1442 * Parse symbolic link arguments. 1443 * This function has an ugly side effect. It will malloc() an area for 1444 * the symlink and set iov_base to point to it, only if it succeeds. 1445 * So, if it returns with uiop->uio_iov->iov_base != NULL, that must 1446 * be FREE'd later. 1447 */ 1448 int 1449 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap, 1450 struct thread *p, char **pathcpp, int *lenp) 1451 { 1452 u_int32_t *tl; 1453 char *pathcp = NULL; 1454 int error = 0, len; 1455 struct nfsv2_sattr *sp; 1456 1457 *pathcpp = NULL; 1458 *lenp = 0; 1459 if ((nd->nd_flag & ND_NFSV3) && 1460 (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p))) 1461 goto nfsmout; 1462 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 1463 len = fxdr_unsigned(int, *tl); 1464 if (len > NFS_MAXPATHLEN || len <= 0) { 1465 error = EBADRPC; 1466 goto nfsmout; 1467 } 1468 pathcp = malloc(len + 1, M_TEMP, M_WAITOK); 1469 error = nfsrv_mtostr(nd, pathcp, len); 1470 if (error) 1471 goto nfsmout; 1472 if (nd->nd_flag & ND_NFSV2) { 1473 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1474 nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode); 1475 } 1476 *pathcpp = pathcp; 1477 *lenp = len; 1478 NFSEXITCODE2(0, nd); 1479 return (0); 1480 nfsmout: 1481 if (pathcp) 1482 free(pathcp, M_TEMP); 1483 NFSEXITCODE2(error, nd); 1484 return (error); 1485 } 1486 1487 /* 1488 * Remove a non-directory object. 1489 */ 1490 int 1491 nfsvno_removesub(struct nameidata *ndp, bool is_v4, struct nfsrv_descript *nd, 1492 struct thread *p, struct nfsexstuff *exp) 1493 { 1494 struct vnode *vp, *dsdvp[NFSDEV_MAXMIRRORS], *newvp; 1495 struct mount *mp; 1496 int error = 0, mirrorcnt, ret; 1497 char fname[PNFS_FILENAME_LEN + 1]; 1498 fhandle_t fh; 1499 1500 vp = ndp->ni_vp; 1501 dsdvp[0] = NULL; 1502 if (vp->v_type == VDIR) { 1503 error = NFSERR_ISDIR; 1504 } else if (is_v4) { 1505 if (nfsrv_recalldeleg || (nd->nd_flag & ND_NFSV41) == 0) 1506 error = nfsrv_checkremove(vp, 1, NULL, 1507 (nfsquad_t)((u_quad_t)0), p); 1508 else 1509 error = nfsrv_checkremove(vp, 1, NULL, nd->nd_clientid, 1510 p); 1511 } 1512 if (error == 0) 1513 nfsrv_pnfsremovesetup(vp, p, dsdvp, &mirrorcnt, fname, &fh); 1514 if (!error) 1515 error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd); 1516 if (error == 0 && dsdvp[0] != NULL) 1517 nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p); 1518 if (is_v4 && (nd->nd_flag & ND_NFSV41) != 0 && error == 0) 1519 error = nfsvno_getfh(vp, &fh, p); 1520 if (ndp->ni_dvp == vp) 1521 vrele(ndp->ni_dvp); 1522 else 1523 vput(ndp->ni_dvp); 1524 vput(vp); 1525 1526 /* Use ret to determine if the file still exists. */ 1527 if (is_v4 && (nd->nd_flag & ND_NFSV41) != 0 && error == 0) { 1528 mp = vfs_busyfs(&fh.fh_fsid); 1529 if (mp != NULL) { 1530 /* Find out if the file still exists. */ 1531 ret = VFS_FHTOVP(mp, &fh.fh_fid, LK_SHARED, &newvp); 1532 if (ret == 0) 1533 vput(newvp); 1534 else 1535 ret = ESTALE; 1536 vfs_unbusy(mp); 1537 } else { 1538 ret = ESTALE; 1539 } 1540 if (ret == ESTALE) { 1541 /* Get rid of any delegation. */ 1542 nfsrv_removedeleg(&fh, nd, p); 1543 } 1544 } 1545 1546 nfsvno_relpathbuf(ndp); 1547 NFSEXITCODE(error); 1548 return (error); 1549 } 1550 1551 /* 1552 * Remove a directory. 1553 */ 1554 int 1555 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred, 1556 struct thread *p, struct nfsexstuff *exp) 1557 { 1558 struct vnode *vp; 1559 int error = 0; 1560 1561 vp = ndp->ni_vp; 1562 if (vp->v_type != VDIR) { 1563 error = ENOTDIR; 1564 goto out; 1565 } 1566 /* 1567 * No rmdir "." please. 1568 */ 1569 if (ndp->ni_dvp == vp) { 1570 error = EINVAL; 1571 goto out; 1572 } 1573 /* 1574 * The root of a mounted filesystem cannot be deleted. 1575 */ 1576 if (vp->v_vflag & VV_ROOT) 1577 error = EBUSY; 1578 out: 1579 if (!error) 1580 error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd); 1581 if (ndp->ni_dvp == vp) 1582 vrele(ndp->ni_dvp); 1583 else 1584 vput(ndp->ni_dvp); 1585 vput(vp); 1586 nfsvno_relpathbuf(ndp); 1587 NFSEXITCODE(error); 1588 return (error); 1589 } 1590 1591 /* 1592 * Rename vnode op. 1593 */ 1594 int 1595 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp, 1596 struct nfsrv_descript *nd, struct thread *p) 1597 { 1598 struct vnode *fvp, *tvp, *tdvp, *dsdvp[NFSDEV_MAXMIRRORS], *newvp; 1599 struct mount *mp; 1600 int error = 0, mirrorcnt, ret; 1601 char fname[PNFS_FILENAME_LEN + 1]; 1602 fhandle_t fh, fh2; 1603 1604 dsdvp[0] = NULL; 1605 fvp = fromndp->ni_vp; 1606 if (nd->nd_repstat != 0) { 1607 vrele(fromndp->ni_dvp); 1608 vrele(fvp); 1609 error = nd->nd_repstat; 1610 goto out1; 1611 } 1612 tdvp = tondp->ni_dvp; 1613 tvp = tondp->ni_vp; 1614 if (tvp != NULL) { 1615 if (fvp->v_type == VDIR && tvp->v_type != VDIR) { 1616 error = (nd->nd_flag & ND_NFSV2) ? EISDIR : EEXIST; 1617 goto out; 1618 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { 1619 error = (nd->nd_flag & ND_NFSV2) ? ENOTDIR : EEXIST; 1620 goto out; 1621 } 1622 if (tvp->v_type == VDIR && tvp->v_mountedhere) { 1623 error = (nd->nd_flag & ND_NFSV2) ? ENOTEMPTY : EXDEV; 1624 goto out; 1625 } 1626 1627 /* 1628 * A rename to '.' or '..' results in a prematurely 1629 * unlocked vnode on FreeBSD5, so I'm just going to fail that 1630 * here. 1631 */ 1632 if ((tondp->ni_cnd.cn_namelen == 1 && 1633 tondp->ni_cnd.cn_nameptr[0] == '.') || 1634 (tondp->ni_cnd.cn_namelen == 2 && 1635 tondp->ni_cnd.cn_nameptr[0] == '.' && 1636 tondp->ni_cnd.cn_nameptr[1] == '.')) { 1637 error = EINVAL; 1638 goto out; 1639 } 1640 } 1641 if (fvp->v_type == VDIR && fvp->v_mountedhere) { 1642 error = (nd->nd_flag & ND_NFSV2) ? ENOTEMPTY : EXDEV; 1643 goto out; 1644 } 1645 if (fvp->v_mount != tdvp->v_mount) { 1646 error = (nd->nd_flag & ND_NFSV2) ? ENOTEMPTY : EXDEV; 1647 goto out; 1648 } 1649 if (fvp == tdvp) { 1650 error = (nd->nd_flag & ND_NFSV2) ? ENOTEMPTY : EINVAL; 1651 goto out; 1652 } 1653 if (fvp == tvp) { 1654 /* 1655 * If source and destination are the same, there is nothing to 1656 * do. Set error to -1 to indicate this. 1657 */ 1658 error = -1; 1659 goto out; 1660 } 1661 if (nd->nd_flag & ND_NFSV4) { 1662 if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) { 1663 if (nfsrv_recalldeleg || (nd->nd_flag & ND_NFSV41) == 0) 1664 error = nfsrv_checkremove(fvp, 0, NULL, 1665 (nfsquad_t)((u_quad_t)0), p); 1666 else 1667 error = nfsrv_checkremove(fvp, 0, NULL, 1668 nd->nd_clientid, p); 1669 NFSVOPUNLOCK(fvp); 1670 } else 1671 error = EPERM; 1672 if (tvp && !error) { 1673 if (nfsrv_recalldeleg || (nd->nd_flag & ND_NFSV41) == 0) 1674 error = nfsrv_checkremove(tvp, 1, NULL, 1675 (nfsquad_t)((u_quad_t)0), p); 1676 else 1677 error = nfsrv_checkremove(tvp, 1, NULL, 1678 nd->nd_clientid, p); 1679 } 1680 } else { 1681 /* 1682 * For NFSv2 and NFSv3, try to get rid of the delegation, so 1683 * that the NFSv4 client won't be confused by the rename. 1684 * Since nfsd_recalldelegation() can only be called on an 1685 * unlocked vnode at this point and fvp is the file that will 1686 * still exist after the rename, just do fvp. 1687 */ 1688 nfsd_recalldelegation(fvp, p); 1689 } 1690 if (error == 0 && tvp != NULL) { 1691 if ((nd->nd_flag & ND_NFSV41) != 0) 1692 error = nfsvno_getfh(tvp, &fh2, p); 1693 if (error == 0) 1694 nfsrv_pnfsremovesetup(tvp, p, dsdvp, &mirrorcnt, fname, 1695 &fh); 1696 NFSD_DEBUG(4, "nfsvno_rename: pnfsremovesetup" 1697 " dsdvp=%p\n", dsdvp[0]); 1698 } 1699 out: 1700 if (!error) { 1701 error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp, 1702 &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp, 1703 &tondp->ni_cnd); 1704 } else { 1705 if (tdvp == tvp) 1706 vrele(tdvp); 1707 else 1708 vput(tdvp); 1709 if (tvp) 1710 vput(tvp); 1711 vrele(fromndp->ni_dvp); 1712 vrele(fvp); 1713 if (error == -1) 1714 error = 0; 1715 } 1716 1717 /* 1718 * If dsdvp[0] != NULL, it was set up by nfsrv_pnfsremovesetup() and 1719 * if the rename succeeded, the DS file for the tvp needs to be 1720 * removed. 1721 */ 1722 if (error == 0 && dsdvp[0] != NULL) { 1723 nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p); 1724 NFSD_DEBUG(4, "nfsvno_rename: pnfsremove\n"); 1725 } 1726 1727 /* Use ret to determine if the file still exists. */ 1728 if ((nd->nd_flag & ND_NFSV41) != 0 && error == 0) { 1729 mp = vfs_busyfs(&fh2.fh_fsid); 1730 if (mp != NULL) { 1731 /* Find out if the file still exists. */ 1732 ret = VFS_FHTOVP(mp, &fh2.fh_fid, LK_SHARED, &newvp); 1733 if (ret == 0) 1734 vput(newvp); 1735 else 1736 ret = ESTALE; 1737 vfs_unbusy(mp); 1738 } else { 1739 ret = ESTALE; 1740 } 1741 if (ret == ESTALE) { 1742 /* Get rid of any delegation. */ 1743 nfsrv_removedeleg(&fh2, nd, p); 1744 } 1745 } 1746 1747 nfsvno_relpathbuf(tondp); 1748 out1: 1749 nfsvno_relpathbuf(fromndp); 1750 NFSEXITCODE(error); 1751 return (error); 1752 } 1753 1754 /* 1755 * Link vnode op. 1756 */ 1757 int 1758 nfsvno_link(struct nameidata *ndp, struct vnode *vp, nfsquad_t clientid, 1759 struct ucred *cred, struct thread *p, struct nfsexstuff *exp) 1760 { 1761 struct vnode *xp; 1762 int error = 0; 1763 1764 xp = ndp->ni_vp; 1765 if (xp != NULL) { 1766 error = EEXIST; 1767 } else { 1768 xp = ndp->ni_dvp; 1769 if (vp->v_mount != xp->v_mount) 1770 error = EXDEV; 1771 } 1772 if (!error) { 1773 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 1774 if (!VN_IS_DOOMED(vp)) { 1775 error = nfsrv_checkremove(vp, 0, NULL, clientid, p); 1776 if (error == 0) 1777 error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd); 1778 } else 1779 error = EPERM; 1780 if (ndp->ni_dvp == vp) { 1781 vrele(ndp->ni_dvp); 1782 NFSVOPUNLOCK(vp); 1783 } else { 1784 vref(vp); 1785 VOP_VPUT_PAIR(ndp->ni_dvp, &vp, true); 1786 } 1787 } else { 1788 if (ndp->ni_dvp == ndp->ni_vp) 1789 vrele(ndp->ni_dvp); 1790 else 1791 vput(ndp->ni_dvp); 1792 if (ndp->ni_vp) 1793 vrele(ndp->ni_vp); 1794 } 1795 nfsvno_relpathbuf(ndp); 1796 NFSEXITCODE(error); 1797 return (error); 1798 } 1799 1800 /* 1801 * Do the fsync() appropriate for the commit. 1802 */ 1803 int 1804 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred, 1805 struct thread *td) 1806 { 1807 int error = 0; 1808 1809 /* 1810 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of 1811 * file is done. At this time VOP_FSYNC does not accept offset and 1812 * byte count parameters so call VOP_FSYNC the whole file for now. 1813 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3. 1814 * File systems that do not use the buffer cache (as indicated 1815 * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC(). 1816 */ 1817 if (cnt == 0 || cnt > MAX_COMMIT_COUNT || 1818 (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) { 1819 /* 1820 * Give up and do the whole thing 1821 */ 1822 vnode_pager_clean_sync(vp); 1823 error = VOP_FSYNC(vp, MNT_WAIT, td); 1824 } else { 1825 /* 1826 * Locate and synchronously write any buffers that fall 1827 * into the requested range. Note: we are assuming that 1828 * f_iosize is a power of 2. 1829 */ 1830 int iosize = vp->v_mount->mnt_stat.f_iosize; 1831 int iomask = iosize - 1; 1832 struct bufobj *bo; 1833 daddr_t lblkno; 1834 1835 /* 1836 * Align to iosize boundary, super-align to page boundary. 1837 */ 1838 if (off & iomask) { 1839 cnt += off & iomask; 1840 off &= ~(u_quad_t)iomask; 1841 } 1842 if (off & PAGE_MASK) { 1843 cnt += off & PAGE_MASK; 1844 off &= ~(u_quad_t)PAGE_MASK; 1845 } 1846 lblkno = off / iosize; 1847 1848 if (vp->v_object && vm_object_mightbedirty(vp->v_object)) { 1849 VM_OBJECT_WLOCK(vp->v_object); 1850 vm_object_page_clean(vp->v_object, off, off + cnt, 1851 OBJPC_SYNC); 1852 VM_OBJECT_WUNLOCK(vp->v_object); 1853 } 1854 1855 bo = &vp->v_bufobj; 1856 BO_LOCK(bo); 1857 while (cnt > 0) { 1858 struct buf *bp; 1859 1860 /* 1861 * If we have a buffer and it is marked B_DELWRI we 1862 * have to lock and write it. Otherwise the prior 1863 * write is assumed to have already been committed. 1864 * 1865 * gbincore() can return invalid buffers now so we 1866 * have to check that bit as well (though B_DELWRI 1867 * should not be set if B_INVAL is set there could be 1868 * a race here since we haven't locked the buffer). 1869 */ 1870 if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) { 1871 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | 1872 LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) { 1873 BO_LOCK(bo); 1874 continue; /* retry */ 1875 } 1876 if ((bp->b_flags & (B_DELWRI|B_INVAL)) == 1877 B_DELWRI) { 1878 bremfree(bp); 1879 bp->b_flags &= ~B_ASYNC; 1880 bwrite(bp); 1881 ++nfs_commit_miss; 1882 } else 1883 BUF_UNLOCK(bp); 1884 BO_LOCK(bo); 1885 } 1886 ++nfs_commit_blks; 1887 if (cnt < iosize) 1888 break; 1889 cnt -= iosize; 1890 ++lblkno; 1891 } 1892 BO_UNLOCK(bo); 1893 } 1894 NFSEXITCODE(error); 1895 return (error); 1896 } 1897 1898 /* 1899 * Statfs vnode op. 1900 */ 1901 int 1902 nfsvno_statfs(struct vnode *vp, struct statfs *sf) 1903 { 1904 struct statfs *tsf; 1905 int error; 1906 1907 tsf = NULL; 1908 if (nfsrv_devidcnt > 0) { 1909 /* For a pNFS service, get the DS numbers. */ 1910 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK | M_ZERO); 1911 error = nfsrv_pnfsstatfs(tsf, vp->v_mount); 1912 if (error != 0) { 1913 free(tsf, M_TEMP); 1914 tsf = NULL; 1915 } 1916 } 1917 error = VFS_STATFS(vp->v_mount, sf); 1918 if (error == 0) { 1919 if (tsf != NULL) { 1920 sf->f_blocks = tsf->f_blocks; 1921 sf->f_bavail = tsf->f_bavail; 1922 sf->f_bfree = tsf->f_bfree; 1923 sf->f_bsize = tsf->f_bsize; 1924 } 1925 /* 1926 * Since NFS handles these values as unsigned on the 1927 * wire, there is no way to represent negative values, 1928 * so set them to 0. Without this, they will appear 1929 * to be very large positive values for clients like 1930 * Solaris10. 1931 */ 1932 if (sf->f_bavail < 0) 1933 sf->f_bavail = 0; 1934 if (sf->f_ffree < 0) 1935 sf->f_ffree = 0; 1936 } 1937 free(tsf, M_TEMP); 1938 NFSEXITCODE(error); 1939 return (error); 1940 } 1941 1942 /* 1943 * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but 1944 * must handle nfsrv_opencheck() calls after any other access checks. 1945 */ 1946 void 1947 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp, 1948 nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp, 1949 int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create, 1950 NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, bool done_namei, 1951 struct nfsexstuff *exp, struct vnode **vpp) 1952 { 1953 struct vnode *vp = NULL; 1954 u_quad_t tempsize; 1955 struct nfsexstuff nes; 1956 struct thread *p = curthread; 1957 uint32_t oldrepstat; 1958 1959 if (ndp->ni_vp == NULL) { 1960 /* 1961 * If nfsrv_opencheck() sets nd_repstat, done_namei needs to be 1962 * set true, since cleanup after nfsvno_namei() is needed. 1963 */ 1964 oldrepstat = nd->nd_repstat; 1965 nd->nd_repstat = nfsrv_opencheck(clientid, 1966 stateidp, stp, NULL, nd, p, nd->nd_repstat); 1967 if (nd->nd_repstat != 0 && oldrepstat == 0) 1968 done_namei = true; 1969 } 1970 if (!nd->nd_repstat) { 1971 if (ndp->ni_vp == NULL) { 1972 nd->nd_repstat = VOP_CREATE(ndp->ni_dvp, 1973 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); 1974 /* For a pNFS server, create the data file on a DS. */ 1975 if (nd->nd_repstat == 0) { 1976 /* 1977 * Create a data file on a DS for a pNFS server. 1978 * This function just returns if not 1979 * running a pNFS DS or the creation fails. 1980 */ 1981 nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr, 1982 cred, p); 1983 } 1984 VOP_VPUT_PAIR(ndp->ni_dvp, nd->nd_repstat == 0 ? 1985 &ndp->ni_vp : NULL, false); 1986 nfsvno_relpathbuf(ndp); 1987 if (!nd->nd_repstat) { 1988 if (*exclusive_flagp) { 1989 *exclusive_flagp = 0; 1990 NFSVNO_ATTRINIT(nvap); 1991 nvap->na_atime.tv_sec = cverf[0]; 1992 nvap->na_atime.tv_nsec = cverf[1]; 1993 nd->nd_repstat = VOP_SETATTR(ndp->ni_vp, 1994 &nvap->na_vattr, cred); 1995 if (nd->nd_repstat != 0) { 1996 vput(ndp->ni_vp); 1997 ndp->ni_vp = NULL; 1998 nd->nd_repstat = NFSERR_NOTSUPP; 1999 } else 2000 NFSSETBIT_ATTRBIT(attrbitp, 2001 NFSATTRBIT_TIMEACCESS); 2002 } else { 2003 nfsrv_fixattr(nd, ndp->ni_vp, nvap, 2004 aclp, p, attrbitp, exp); 2005 } 2006 } 2007 vp = ndp->ni_vp; 2008 } else { 2009 nfsvno_relpathbuf(ndp); 2010 vp = ndp->ni_vp; 2011 if (create == NFSV4OPEN_CREATE) { 2012 if (ndp->ni_dvp == vp) 2013 vrele(ndp->ni_dvp); 2014 else 2015 vput(ndp->ni_dvp); 2016 } 2017 if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) { 2018 if (ndp->ni_cnd.cn_flags & RDONLY) 2019 NFSVNO_SETEXRDONLY(&nes); 2020 else 2021 NFSVNO_EXINIT(&nes); 2022 nd->nd_repstat = nfsvno_accchk(vp, 2023 VWRITE, cred, &nes, p, 2024 NFSACCCHK_NOOVERRIDE, 2025 NFSACCCHK_VPISLOCKED, NULL); 2026 nd->nd_repstat = nfsrv_opencheck(clientid, 2027 stateidp, stp, vp, nd, p, nd->nd_repstat); 2028 if (!nd->nd_repstat) { 2029 tempsize = nvap->na_size; 2030 NFSVNO_ATTRINIT(nvap); 2031 nvap->na_size = tempsize; 2032 nd->nd_repstat = nfsvno_setattr(vp, 2033 nvap, cred, p, exp); 2034 } 2035 } else if (vp->v_type == VREG) { 2036 nd->nd_repstat = nfsrv_opencheck(clientid, 2037 stateidp, stp, vp, nd, p, nd->nd_repstat); 2038 } 2039 } 2040 } else if (done_namei) { 2041 KASSERT(create == NFSV4OPEN_CREATE, 2042 ("nfsvno_open: not create")); 2043 /* 2044 * done_namei is set when nfsvno_namei() has completed 2045 * successfully, but a subsequent error was set in 2046 * nd_repstat. As such, cleanup of the nfsvno_namei() 2047 * results is required. 2048 */ 2049 nfsvno_relpathbuf(ndp); 2050 if (ndp->ni_dvp == ndp->ni_vp) 2051 vrele(ndp->ni_dvp); 2052 else 2053 vput(ndp->ni_dvp); 2054 if (ndp->ni_vp) 2055 vput(ndp->ni_vp); 2056 } 2057 *vpp = vp; 2058 2059 NFSEXITCODE2(0, nd); 2060 } 2061 2062 /* 2063 * Updates the file rev and sets the mtime and ctime 2064 * to the current clock time, returning the va_filerev and va_Xtime 2065 * values. 2066 * Return ESTALE to indicate the vnode is VIRF_DOOMED. 2067 */ 2068 int 2069 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap, 2070 struct nfsrv_descript *nd, struct thread *p) 2071 { 2072 struct vattr va; 2073 2074 VATTR_NULL(&va); 2075 vfs_timestamp(&va.va_mtime); 2076 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) { 2077 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 2078 if (VN_IS_DOOMED(vp)) 2079 return (ESTALE); 2080 } 2081 (void) VOP_SETATTR(vp, &va, nd->nd_cred); 2082 (void) nfsvno_getattr(vp, nvap, nd, p, 1, NULL); 2083 return (0); 2084 } 2085 2086 /* 2087 * Glue routine to nfsv4_fillattr(). 2088 */ 2089 int 2090 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp, 2091 struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp, 2092 struct ucred *cred, struct thread *p, int isdgram, int reterr, 2093 int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno) 2094 { 2095 struct statfs *sf; 2096 int error; 2097 2098 sf = NULL; 2099 if (nfsrv_devidcnt > 0 && 2100 (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEAVAIL) || 2101 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEFREE) || 2102 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACETOTAL))) { 2103 sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK | M_ZERO); 2104 error = nfsrv_pnfsstatfs(sf, mp); 2105 if (error != 0) { 2106 free(sf, M_TEMP); 2107 sf = NULL; 2108 } 2109 } 2110 error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror, 2111 attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root, 2112 mounted_on_fileno, sf); 2113 free(sf, M_TEMP); 2114 NFSEXITCODE2(0, nd); 2115 return (error); 2116 } 2117 2118 /* 2119 * Convert a dirent d_type to a vnode type. 2120 */ 2121 static void nfs_dtypetovtype(struct nfsvattr *nvap, struct vnode *vp, 2122 uint8_t dtype) 2123 { 2124 2125 if ((vn_irflag_read(vp) & VIRF_NAMEDDIR) != 0) { 2126 nvap->na_type = VREG; 2127 nvap->na_bsdflags |= SFBSD_NAMEDATTR; 2128 } else if (dtype <= DT_WHT) { 2129 nvap->na_type = dtype_to_vnode[dtype]; 2130 } else { 2131 nvap->na_type = VNON; 2132 } 2133 } 2134 2135 /* Since the Readdir vnode ops vary, put the entire functions in here. */ 2136 /* 2137 * nfs readdir service 2138 * - mallocs what it thinks is enough to read 2139 * count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR 2140 * - calls VOP_READDIR() 2141 * - loops around building the reply 2142 * if the output generated exceeds count break out of loop 2143 * The NFSM_CLGET macro is used here so that the reply will be packed 2144 * tightly in mbuf clusters. 2145 * - it trims out records with d_fileno == 0 2146 * this doesn't matter for Unix clients, but they might confuse clients 2147 * for other os'. 2148 * - it trims out records with d_type == DT_WHT 2149 * these cannot be seen through NFS (unless we extend the protocol) 2150 * The alternate call nfsrvd_readdirplus() does lookups as well. 2151 * PS: The NFS protocol spec. does not clarify what the "count" byte 2152 * argument is a count of.. just name strings and file id's or the 2153 * entire reply rpc or ... 2154 * I tried just file name and id sizes and it confused the Sun client, 2155 * so I am using the full rpc size now. The "paranoia.." comment refers 2156 * to including the status longwords that are not a part of the dir. 2157 * "entry" structures, but are in the rpc. 2158 */ 2159 int 2160 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram, 2161 struct vnode *vp, struct nfsexstuff *exp) 2162 { 2163 struct dirent *dp; 2164 u_int32_t *tl; 2165 int dirlen; 2166 char *cpos, *cend, *rbuf; 2167 struct nfsvattr at; 2168 int nlen, error = 0, getret = 1; 2169 int siz, cnt, fullsiz, eofflag, ncookies; 2170 u_int64_t off, toff, verf __unused; 2171 uint64_t *cookies = NULL, *cookiep; 2172 struct uio io; 2173 struct iovec iv; 2174 int is_ufs; 2175 struct thread *p = curthread; 2176 2177 if (nd->nd_repstat) { 2178 nfsrv_postopattr(nd, getret, &at); 2179 goto out; 2180 } 2181 if (nd->nd_flag & ND_NFSV2) { 2182 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2183 off = fxdr_unsigned(u_quad_t, *tl++); 2184 } else { 2185 NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2186 off = fxdr_hyper(tl); 2187 tl += 2; 2188 verf = fxdr_hyper(tl); 2189 tl += 2; 2190 } 2191 toff = off; 2192 cnt = fxdr_unsigned(int, *tl); 2193 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0) 2194 cnt = NFS_SRVMAXDATA(nd); 2195 siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1)); 2196 fullsiz = siz; 2197 if (nd->nd_flag & ND_NFSV3) { 2198 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, 2199 NULL); 2200 #if 0 2201 /* 2202 * va_filerev is not sufficient as a cookie verifier, 2203 * since it is not supposed to change when entries are 2204 * removed/added unless that offset cookies returned to 2205 * the client are no longer valid. 2206 */ 2207 if (!nd->nd_repstat && toff && verf != at.na_filerev) 2208 nd->nd_repstat = NFSERR_BAD_COOKIE; 2209 #endif 2210 } 2211 if (!nd->nd_repstat && vp->v_type != VDIR) 2212 nd->nd_repstat = NFSERR_NOTDIR; 2213 if (nd->nd_repstat == 0 && cnt == 0) { 2214 if (nd->nd_flag & ND_NFSV2) 2215 /* NFSv2 does not have NFSERR_TOOSMALL */ 2216 nd->nd_repstat = EPERM; 2217 else 2218 nd->nd_repstat = NFSERR_TOOSMALL; 2219 } 2220 if (!nd->nd_repstat) 2221 nd->nd_repstat = nfsvno_accchk(vp, VEXEC, 2222 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, 2223 NFSACCCHK_VPISLOCKED, NULL); 2224 if (nd->nd_repstat) { 2225 vput(vp); 2226 if (nd->nd_flag & ND_NFSV3) 2227 nfsrv_postopattr(nd, getret, &at); 2228 goto out; 2229 } 2230 is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0; 2231 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2232 again: 2233 eofflag = 0; 2234 if (cookies) { 2235 free(cookies, M_TEMP); 2236 cookies = NULL; 2237 } 2238 2239 iv.iov_base = rbuf; 2240 iv.iov_len = siz; 2241 io.uio_iov = &iv; 2242 io.uio_iovcnt = 1; 2243 io.uio_offset = (off_t)off; 2244 io.uio_resid = siz; 2245 io.uio_segflg = UIO_SYSSPACE; 2246 io.uio_rw = UIO_READ; 2247 io.uio_td = NULL; 2248 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies, 2249 &cookies); 2250 off = (u_int64_t)io.uio_offset; 2251 if (io.uio_resid) 2252 siz -= io.uio_resid; 2253 2254 if (!cookies && !nd->nd_repstat) 2255 nd->nd_repstat = NFSERR_PERM; 2256 if (nd->nd_flag & ND_NFSV3) { 2257 getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); 2258 if (!nd->nd_repstat) 2259 nd->nd_repstat = getret; 2260 } 2261 2262 /* 2263 * Handles the failed cases. nd->nd_repstat == 0 past here. 2264 */ 2265 if (nd->nd_repstat) { 2266 vput(vp); 2267 free(rbuf, M_TEMP); 2268 if (cookies) 2269 free(cookies, M_TEMP); 2270 if (nd->nd_flag & ND_NFSV3) 2271 nfsrv_postopattr(nd, getret, &at); 2272 goto out; 2273 } 2274 /* 2275 * If nothing read, return eof 2276 * rpc reply 2277 */ 2278 if (siz == 0) { 2279 vput(vp); 2280 if (nd->nd_flag & ND_NFSV2) { 2281 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2282 } else { 2283 nfsrv_postopattr(nd, getret, &at); 2284 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2285 txdr_hyper(at.na_filerev, tl); 2286 tl += 2; 2287 } 2288 *tl++ = newnfs_false; 2289 *tl = newnfs_true; 2290 free(rbuf, M_TEMP); 2291 free(cookies, M_TEMP); 2292 goto out; 2293 } 2294 2295 /* 2296 * Check for degenerate cases of nothing useful read. 2297 * If so go try again 2298 */ 2299 cpos = rbuf; 2300 cend = rbuf + siz; 2301 dp = (struct dirent *)cpos; 2302 cookiep = cookies; 2303 2304 /* 2305 * For some reason FreeBSD's ufs_readdir() chooses to back the 2306 * directory offset up to a block boundary, so it is necessary to 2307 * skip over the records that precede the requested offset. This 2308 * requires the assumption that file offset cookies monotonically 2309 * increase. 2310 */ 2311 while (cpos < cend && ncookies > 0 && 2312 (dp->d_fileno == 0 || dp->d_type == DT_WHT || 2313 (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) { 2314 cpos += dp->d_reclen; 2315 dp = (struct dirent *)cpos; 2316 cookiep++; 2317 ncookies--; 2318 } 2319 if (cpos >= cend || ncookies == 0) { 2320 siz = fullsiz; 2321 toff = off; 2322 goto again; 2323 } 2324 vput(vp); 2325 2326 /* 2327 * If cnt > MCLBYTES and the reply will not be saved, use 2328 * ext_pgs mbufs for TLS. 2329 * For NFSv4.0, we do not know for sure if the reply will 2330 * be saved, so do not use ext_pgs mbufs for NFSv4.0. 2331 */ 2332 if (cnt > MCLBYTES && siz > MCLBYTES && 2333 (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS && 2334 (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4) 2335 nd->nd_flag |= ND_EXTPG; 2336 2337 /* 2338 * dirlen is the size of the reply, including all XDR and must 2339 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate 2340 * if the XDR should be included in "count", but to be safe, we do. 2341 * (Include the two booleans at the end of the reply in dirlen now.) 2342 */ 2343 if (nd->nd_flag & ND_NFSV3) { 2344 nfsrv_postopattr(nd, getret, &at); 2345 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2346 txdr_hyper(at.na_filerev, tl); 2347 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED; 2348 } else { 2349 dirlen = 2 * NFSX_UNSIGNED; 2350 } 2351 2352 /* Loop through the records and build reply */ 2353 while (cpos < cend && ncookies > 0) { 2354 nlen = dp->d_namlen; 2355 if (dp->d_fileno != 0 && dp->d_type != DT_WHT && 2356 nlen <= NFS_MAXNAMLEN) { 2357 if (nd->nd_flag & ND_NFSV3) 2358 dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen)); 2359 else 2360 dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen)); 2361 if (dirlen > cnt) { 2362 eofflag = 0; 2363 break; 2364 } 2365 2366 /* 2367 * Build the directory record xdr from 2368 * the dirent entry. 2369 */ 2370 if (nd->nd_flag & ND_NFSV3) { 2371 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2372 *tl++ = newnfs_true; 2373 txdr_hyper(dp->d_fileno, tl); 2374 } else { 2375 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2376 *tl++ = newnfs_true; 2377 *tl = txdr_unsigned(dp->d_fileno); 2378 } 2379 (void) nfsm_strtom(nd, dp->d_name, nlen); 2380 if (nd->nd_flag & ND_NFSV3) { 2381 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2382 txdr_hyper(*cookiep, tl); 2383 } else { 2384 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 2385 *tl = txdr_unsigned(*cookiep); 2386 } 2387 } 2388 cpos += dp->d_reclen; 2389 dp = (struct dirent *)cpos; 2390 cookiep++; 2391 ncookies--; 2392 } 2393 if (cpos < cend) 2394 eofflag = 0; 2395 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2396 *tl++ = newnfs_false; 2397 if (eofflag) 2398 *tl = newnfs_true; 2399 else 2400 *tl = newnfs_false; 2401 free(rbuf, M_TEMP); 2402 free(cookies, M_TEMP); 2403 2404 out: 2405 NFSEXITCODE2(0, nd); 2406 return (0); 2407 nfsmout: 2408 vput(vp); 2409 NFSEXITCODE2(error, nd); 2410 return (error); 2411 } 2412 2413 /* 2414 * Readdirplus for V3 and Readdir for V4. 2415 */ 2416 int 2417 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram, 2418 struct vnode *vp, struct nfsexstuff *exp) 2419 { 2420 struct dirent *dp; 2421 u_int32_t *tl; 2422 int dirlen; 2423 char *cpos, *cend, *rbuf; 2424 struct vnode *nvp; 2425 fhandle_t nfh; 2426 struct nfsvattr nva, at, *nvap = &nva; 2427 struct mbuf *mb0, *mb1; 2428 struct nfsreferral *refp; 2429 int nlen, r, error = 0, getret = 1, usevget = 1; 2430 int siz, cnt, fullsiz, eofflag, ncookies, entrycnt; 2431 caddr_t bpos0, bpos1; 2432 u_int64_t off, toff, verf __unused; 2433 uint64_t *cookies = NULL, *cookiep; 2434 nfsattrbit_t attrbits, rderrbits, savbits, refbits; 2435 struct uio io; 2436 struct iovec iv; 2437 struct componentname cn; 2438 int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls; 2439 struct mount *mp, *new_mp; 2440 uint64_t mounted_on_fileno; 2441 struct thread *p = curthread; 2442 int bextpg0, bextpg1, bextpgsiz0, bextpgsiz1; 2443 2444 if (nd->nd_repstat) { 2445 nfsrv_postopattr(nd, getret, &at); 2446 goto out; 2447 } 2448 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2449 off = fxdr_hyper(tl); 2450 toff = off; 2451 tl += 2; 2452 verf = fxdr_hyper(tl); 2453 tl += 2; 2454 siz = fxdr_unsigned(int, *tl++); 2455 cnt = fxdr_unsigned(int, *tl); 2456 2457 /* 2458 * Use the server's maximum data transfer size as the upper bound 2459 * on reply datalen. 2460 */ 2461 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0) 2462 cnt = NFS_SRVMAXDATA(nd); 2463 2464 /* 2465 * siz is a "hint" of how much directory information (name, fileid, 2466 * cookie) should be in the reply. At least one client "hints" 0, 2467 * so I set it to cnt for that case. I also round it up to the 2468 * next multiple of DIRBLKSIZ. 2469 * Since the size of a Readdirplus directory entry reply will always 2470 * be greater than a directory entry returned by VOP_READDIR(), it 2471 * does not make sense to read more than NFS_SRVMAXDATA() via 2472 * VOP_READDIR(). 2473 */ 2474 if (siz <= 0) 2475 siz = cnt; 2476 else if (siz > NFS_SRVMAXDATA(nd)) 2477 siz = NFS_SRVMAXDATA(nd); 2478 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1)); 2479 2480 if (nd->nd_flag & ND_NFSV4) { 2481 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); 2482 if (error) 2483 goto nfsmout; 2484 NFSSET_ATTRBIT(&savbits, &attrbits); 2485 NFSSET_ATTRBIT(&refbits, &attrbits); 2486 NFSCLRNOTFILLABLE_ATTRBIT(&attrbits, nd); 2487 NFSZERO_ATTRBIT(&rderrbits); 2488 NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR); 2489 /* 2490 * If these 4 bits are the only attributes requested by the 2491 * client, they can be satisfied without acquiring the vnode 2492 * for the file object unless it is a directory. 2493 * This will be indicated by savbits being all 0s. 2494 */ 2495 NFSCLRBIT_ATTRBIT(&savbits, NFSATTRBIT_TYPE); 2496 NFSCLRBIT_ATTRBIT(&savbits, NFSATTRBIT_FILEID); 2497 NFSCLRBIT_ATTRBIT(&savbits, NFSATTRBIT_MOUNTEDONFILEID); 2498 NFSCLRBIT_ATTRBIT(&savbits, NFSATTRBIT_RDATTRERROR); 2499 } else { 2500 NFSZERO_ATTRBIT(&attrbits); 2501 } 2502 fullsiz = siz; 2503 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); 2504 #if 0 2505 if (!nd->nd_repstat) { 2506 if (off && verf != at.na_filerev) { 2507 /* 2508 * va_filerev is not sufficient as a cookie verifier, 2509 * since it is not supposed to change when entries are 2510 * removed/added unless that offset cookies returned to 2511 * the client are no longer valid. 2512 */ 2513 if (nd->nd_flag & ND_NFSV4) { 2514 nd->nd_repstat = NFSERR_NOTSAME; 2515 } else { 2516 nd->nd_repstat = NFSERR_BAD_COOKIE; 2517 } 2518 } 2519 } 2520 #endif 2521 if (!nd->nd_repstat && vp->v_type != VDIR) 2522 nd->nd_repstat = NFSERR_NOTDIR; 2523 if (!nd->nd_repstat && cnt == 0) 2524 nd->nd_repstat = NFSERR_TOOSMALL; 2525 if (!nd->nd_repstat) 2526 nd->nd_repstat = nfsvno_accchk(vp, VEXEC, 2527 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, 2528 NFSACCCHK_VPISLOCKED, NULL); 2529 if (nd->nd_repstat) { 2530 vput(vp); 2531 if (nd->nd_flag & ND_NFSV3) 2532 nfsrv_postopattr(nd, getret, &at); 2533 goto out; 2534 } 2535 is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0; 2536 is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0; 2537 2538 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2539 again: 2540 eofflag = 0; 2541 if (cookies) { 2542 free(cookies, M_TEMP); 2543 cookies = NULL; 2544 } 2545 2546 iv.iov_base = rbuf; 2547 iv.iov_len = siz; 2548 io.uio_iov = &iv; 2549 io.uio_iovcnt = 1; 2550 io.uio_offset = (off_t)off; 2551 io.uio_resid = siz; 2552 io.uio_segflg = UIO_SYSSPACE; 2553 io.uio_rw = UIO_READ; 2554 io.uio_td = NULL; 2555 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies, 2556 &cookies); 2557 off = (u_int64_t)io.uio_offset; 2558 if (io.uio_resid) 2559 siz -= io.uio_resid; 2560 2561 getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); 2562 2563 if (!cookies && !nd->nd_repstat) 2564 nd->nd_repstat = NFSERR_PERM; 2565 if (!nd->nd_repstat) 2566 nd->nd_repstat = getret; 2567 if (nd->nd_repstat) { 2568 vput(vp); 2569 if (cookies) 2570 free(cookies, M_TEMP); 2571 free(rbuf, M_TEMP); 2572 if (nd->nd_flag & ND_NFSV3) 2573 nfsrv_postopattr(nd, getret, &at); 2574 goto out; 2575 } 2576 /* 2577 * If nothing read, return eof 2578 * rpc reply 2579 */ 2580 if (siz == 0) { 2581 vput(vp); 2582 if (nd->nd_flag & ND_NFSV3) 2583 nfsrv_postopattr(nd, getret, &at); 2584 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2585 txdr_hyper(at.na_filerev, tl); 2586 tl += 2; 2587 *tl++ = newnfs_false; 2588 *tl = newnfs_true; 2589 free(cookies, M_TEMP); 2590 free(rbuf, M_TEMP); 2591 goto out; 2592 } 2593 2594 /* 2595 * Check for degenerate cases of nothing useful read. 2596 * If so go try again 2597 */ 2598 cpos = rbuf; 2599 cend = rbuf + siz; 2600 dp = (struct dirent *)cpos; 2601 cookiep = cookies; 2602 2603 /* 2604 * For some reason FreeBSD's ufs_readdir() chooses to back the 2605 * directory offset up to a block boundary, so it is necessary to 2606 * skip over the records that precede the requested offset. This 2607 * requires the assumption that file offset cookies monotonically 2608 * increase. 2609 */ 2610 while (cpos < cend && ncookies > 0 && 2611 (dp->d_fileno == 0 || dp->d_type == DT_WHT || 2612 (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) || 2613 ((nd->nd_flag & ND_NFSV4) && 2614 ((dp->d_namlen == 1 && dp->d_name[0] == '.') || 2615 (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) { 2616 cpos += dp->d_reclen; 2617 dp = (struct dirent *)cpos; 2618 cookiep++; 2619 ncookies--; 2620 } 2621 if (cpos >= cend || ncookies == 0) { 2622 siz = fullsiz; 2623 toff = off; 2624 goto again; 2625 } 2626 2627 /* 2628 * Busy the file system so that the mount point won't go away 2629 * and, as such, VFS_VGET() can be used safely. 2630 */ 2631 mp = vp->v_mount; 2632 vfs_ref(mp); 2633 NFSVOPUNLOCK(vp); 2634 nd->nd_repstat = vfs_busy(mp, 0); 2635 vfs_rel(mp); 2636 if (nd->nd_repstat != 0) { 2637 vrele(vp); 2638 free(cookies, M_TEMP); 2639 free(rbuf, M_TEMP); 2640 if (nd->nd_flag & ND_NFSV3) 2641 nfsrv_postopattr(nd, getret, &at); 2642 goto out; 2643 } 2644 2645 /* 2646 * Check to see if entries in this directory can be safely acquired 2647 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required. 2648 * ZFS snapshot directories need VOP_LOOKUP(), so that any 2649 * automount of the snapshot directory that is required will 2650 * be done. 2651 * This needs to be done here for NFSv4, since NFSv4 never does 2652 * a VFS_VGET() for "." or "..". 2653 */ 2654 if (is_zfs == 1) { 2655 r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp); 2656 if (r == EOPNOTSUPP) { 2657 usevget = 0; 2658 cn.cn_nameiop = LOOKUP; 2659 cn.cn_lkflags = LK_SHARED | LK_RETRY; 2660 cn.cn_cred = nd->nd_cred; 2661 } else if (r == 0) 2662 vput(nvp); 2663 } 2664 2665 /* 2666 * If the reply is likely to exceed MCLBYTES and the reply will 2667 * not be saved, use ext_pgs mbufs for TLS. 2668 * It is difficult to predict how large each entry will be and 2669 * how many entries have been read, so just assume the directory 2670 * entries grow by a factor of 4 when attributes are included. 2671 * For NFSv4.0, we do not know for sure if the reply will 2672 * be saved, so do not use ext_pgs mbufs for NFSv4.0. 2673 */ 2674 if (cnt > MCLBYTES && siz > MCLBYTES / 4 && 2675 (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS && 2676 (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4) 2677 nd->nd_flag |= ND_EXTPG; 2678 2679 /* 2680 * Save this position, in case there is an error before one entry 2681 * is created. 2682 */ 2683 mb0 = nd->nd_mb; 2684 bpos0 = nd->nd_bpos; 2685 bextpg0 = nd->nd_bextpg; 2686 bextpgsiz0 = nd->nd_bextpgsiz; 2687 2688 /* 2689 * Fill in the first part of the reply. 2690 * dirlen is the reply length in bytes and cannot exceed cnt. 2691 * (Include the two booleans at the end of the reply in dirlen now, 2692 * so we recognize when we have exceeded cnt.) 2693 */ 2694 if (nd->nd_flag & ND_NFSV3) { 2695 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED; 2696 nfsrv_postopattr(nd, getret, &at); 2697 } else { 2698 dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED; 2699 } 2700 NFSM_BUILD(tl, u_int32_t *, NFSX_VERF); 2701 txdr_hyper(at.na_filerev, tl); 2702 2703 /* 2704 * Save this position, in case there is an empty reply needed. 2705 */ 2706 mb1 = nd->nd_mb; 2707 bpos1 = nd->nd_bpos; 2708 bextpg1 = nd->nd_bextpg; 2709 bextpgsiz1 = nd->nd_bextpgsiz; 2710 2711 /* Loop through the records and build reply */ 2712 entrycnt = 0; 2713 while (cpos < cend && ncookies > 0 && dirlen < cnt) { 2714 nlen = dp->d_namlen; 2715 if (dp->d_fileno != 0 && dp->d_type != DT_WHT && 2716 nlen <= NFS_MAXNAMLEN && 2717 ((nd->nd_flag & ND_NFSV3) || nlen > 2 || 2718 (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.')) 2719 || (nlen == 1 && dp->d_name[0] != '.'))) { 2720 /* 2721 * Save the current position in the reply, in case 2722 * this entry exceeds cnt. 2723 */ 2724 mb1 = nd->nd_mb; 2725 bpos1 = nd->nd_bpos; 2726 bextpg1 = nd->nd_bextpg; 2727 bextpgsiz1 = nd->nd_bextpgsiz; 2728 2729 /* 2730 * For readdir_and_lookup get the vnode using 2731 * the file number. 2732 */ 2733 nvp = NULL; 2734 refp = NULL; 2735 r = 0; 2736 at_root = 0; 2737 needs_unbusy = 0; 2738 new_mp = mp; 2739 mounted_on_fileno = (uint64_t)dp->d_fileno; 2740 if ((nd->nd_flag & ND_NFSV3) || 2741 NFSNONZERO_ATTRBIT(&savbits) || 2742 dp->d_type == DT_UNKNOWN || 2743 (dp->d_type == DT_DIR && 2744 nfsrv_enable_crossmntpt != 0)) { 2745 if (nd->nd_flag & ND_NFSV4) 2746 refp = nfsv4root_getreferral(NULL, 2747 vp, dp->d_fileno); 2748 if (refp == NULL) { 2749 if (usevget) 2750 r = VFS_VGET(mp, dp->d_fileno, 2751 LK_SHARED, &nvp); 2752 else 2753 r = EOPNOTSUPP; 2754 if (r == 0 && (vn_irflag_read(vp) & 2755 VIRF_NAMEDDIR) != 0) 2756 vn_irflag_set_cond(nvp, 2757 VIRF_NAMEDATTR); 2758 if (r == EOPNOTSUPP) { 2759 if (usevget) { 2760 usevget = 0; 2761 cn.cn_nameiop = LOOKUP; 2762 cn.cn_lkflags = 2763 LK_SHARED | 2764 LK_RETRY; 2765 cn.cn_cred = 2766 nd->nd_cred; 2767 } 2768 cn.cn_nameptr = dp->d_name; 2769 cn.cn_namelen = nlen; 2770 cn.cn_flags = ISLASTCN | 2771 NOFOLLOW | LOCKLEAF; 2772 if ((vn_irflag_read(vp) & 2773 VIRF_NAMEDDIR) != 0) 2774 cn.cn_flags |= 2775 OPENNAMED; 2776 if (nlen == 2 && 2777 dp->d_name[0] == '.' && 2778 dp->d_name[1] == '.') 2779 cn.cn_flags |= 2780 ISDOTDOT; 2781 if (NFSVOPLOCK(vp, LK_SHARED) 2782 != 0) { 2783 nd->nd_repstat = EPERM; 2784 break; 2785 } 2786 if ((vp->v_vflag & VV_ROOT) != 0 2787 && (cn.cn_flags & ISDOTDOT) 2788 != 0) { 2789 vref(vp); 2790 nvp = vp; 2791 r = 0; 2792 } else { 2793 r = VOP_LOOKUP(vp, &nvp, 2794 &cn); 2795 if (vp != nvp) 2796 NFSVOPUNLOCK(vp); 2797 } 2798 } 2799 2800 /* 2801 * For NFSv4, check to see if nvp is 2802 * a mount point and get the mount 2803 * point vnode, as required. 2804 */ 2805 if (r == 0 && 2806 nfsrv_enable_crossmntpt != 0 && 2807 (nd->nd_flag & ND_NFSV4) != 0 && 2808 nvp->v_type == VDIR && 2809 nvp->v_mountedhere != NULL) { 2810 new_mp = nvp->v_mountedhere; 2811 r = vfs_busy(new_mp, 0); 2812 vput(nvp); 2813 nvp = NULL; 2814 if (r == 0) { 2815 r = VFS_ROOT(new_mp, 2816 LK_SHARED, &nvp); 2817 needs_unbusy = 1; 2818 if (r == 0) 2819 at_root = 1; 2820 } 2821 } 2822 } 2823 2824 /* 2825 * If we failed to look up the entry, then it 2826 * has become invalid, most likely removed. 2827 */ 2828 if (r != 0) { 2829 if (needs_unbusy) 2830 vfs_unbusy(new_mp); 2831 goto invalid; 2832 } 2833 KASSERT(refp != NULL || nvp != NULL, 2834 ("%s: undetected lookup error", __func__)); 2835 2836 if (refp == NULL && 2837 ((nd->nd_flag & ND_NFSV3) || 2838 NFSNONZERO_ATTRBIT(&attrbits))) { 2839 r = nfsvno_getfh(nvp, &nfh, p); 2840 if (!r) 2841 r = nfsvno_getattr(nvp, nvap, nd, p, 2842 1, &attrbits); 2843 if (r == 0 && is_zfs == 1 && 2844 nfsrv_enable_crossmntpt != 0 && 2845 (nd->nd_flag & ND_NFSV4) != 0 && 2846 nvp->v_type == VDIR && 2847 vp->v_mount != nvp->v_mount) { 2848 /* 2849 * For a ZFS snapshot, there is a 2850 * pseudo mount that does not set 2851 * v_mountedhere, so it needs to 2852 * be detected via a different 2853 * mount structure. 2854 */ 2855 at_root = 1; 2856 if (new_mp == mp) 2857 new_mp = nvp->v_mount; 2858 } 2859 } 2860 2861 /* 2862 * If we failed to get attributes of the entry, 2863 * then just skip it for NFSv3 (the traditional 2864 * behavior in the old NFS server). 2865 * For NFSv4 the behavior is controlled by 2866 * RDATTRERROR: we either ignore the error or 2867 * fail the request. 2868 * The exception is EOPNOTSUPP, which can be 2869 * returned by nfsvno_getfh() for certain 2870 * file systems, such as devfs. This indicates 2871 * that the file system cannot be exported, 2872 * so just skip over the entry. 2873 * Note that RDATTRERROR is never set for NFSv3. 2874 */ 2875 if (r != 0) { 2876 if (!NFSISSET_ATTRBIT(&attrbits, 2877 NFSATTRBIT_RDATTRERROR) || 2878 r == EOPNOTSUPP) { 2879 vput(nvp); 2880 if (needs_unbusy != 0) 2881 vfs_unbusy(new_mp); 2882 if ((nd->nd_flag & ND_NFSV3) || 2883 r == EOPNOTSUPP) 2884 goto invalid; 2885 nd->nd_repstat = r; 2886 break; 2887 } 2888 } 2889 } else if (NFSNONZERO_ATTRBIT(&attrbits)) { 2890 /* Only need Type and/or Fileid. */ 2891 VATTR_NULL(&nvap->na_vattr); 2892 nvap->na_fileid = dp->d_fileno; 2893 nfs_dtypetovtype(nvap, vp, dp->d_type); 2894 } 2895 2896 /* 2897 * Build the directory record xdr 2898 */ 2899 if (nd->nd_flag & ND_NFSV3) { 2900 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2901 *tl++ = newnfs_true; 2902 txdr_hyper(dp->d_fileno, tl); 2903 dirlen += nfsm_strtom(nd, dp->d_name, nlen); 2904 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2905 txdr_hyper(*cookiep, tl); 2906 nfsrv_postopattr(nd, 0, nvap); 2907 dirlen += nfsm_fhtom(NULL, nd, (u_int8_t *)&nfh, 2908 0, 1); 2909 dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR); 2910 if (nvp != NULL) 2911 vput(nvp); 2912 } else { 2913 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2914 *tl++ = newnfs_true; 2915 txdr_hyper(*cookiep, tl); 2916 dirlen += nfsm_strtom(nd, dp->d_name, nlen); 2917 if (nvp != NULL) { 2918 supports_nfsv4acls = 2919 nfs_supportsnfsv4acls(nvp); 2920 NFSVOPUNLOCK(nvp); 2921 } else 2922 supports_nfsv4acls = 0; 2923 if (refp != NULL) { 2924 dirlen += nfsrv_putreferralattr(nd, 2925 &refbits, refp, 0, 2926 &nd->nd_repstat); 2927 if (nd->nd_repstat) { 2928 if (nvp != NULL) 2929 vrele(nvp); 2930 if (needs_unbusy != 0) 2931 vfs_unbusy(new_mp); 2932 break; 2933 } 2934 } else if (r) { 2935 dirlen += nfsvno_fillattr(nd, new_mp, 2936 nvp, nvap, &nfh, r, &rderrbits, 2937 nd->nd_cred, p, isdgram, 0, 2938 supports_nfsv4acls, at_root, 2939 mounted_on_fileno); 2940 } else { 2941 dirlen += nfsvno_fillattr(nd, new_mp, 2942 nvp, nvap, &nfh, r, &attrbits, 2943 nd->nd_cred, p, isdgram, 0, 2944 supports_nfsv4acls, at_root, 2945 mounted_on_fileno); 2946 } 2947 if (nvp != NULL) 2948 vrele(nvp); 2949 dirlen += (3 * NFSX_UNSIGNED); 2950 } 2951 if (needs_unbusy != 0) 2952 vfs_unbusy(new_mp); 2953 if (dirlen <= cnt) 2954 entrycnt++; 2955 } 2956 invalid: 2957 cpos += dp->d_reclen; 2958 dp = (struct dirent *)cpos; 2959 cookiep++; 2960 ncookies--; 2961 } 2962 vrele(vp); 2963 vfs_unbusy(mp); 2964 2965 /* 2966 * If dirlen > cnt, we must strip off the last entry. If that 2967 * results in an empty reply, report NFSERR_TOOSMALL. 2968 */ 2969 if (dirlen > cnt || nd->nd_repstat) { 2970 if (!nd->nd_repstat && entrycnt == 0) 2971 nd->nd_repstat = NFSERR_TOOSMALL; 2972 if (nd->nd_repstat) { 2973 nfsm_trimtrailing(nd, mb0, bpos0, bextpg0, bextpgsiz0); 2974 if (nd->nd_flag & ND_NFSV3) 2975 nfsrv_postopattr(nd, getret, &at); 2976 } else 2977 nfsm_trimtrailing(nd, mb1, bpos1, bextpg1, bextpgsiz1); 2978 eofflag = 0; 2979 } else if (cpos < cend) 2980 eofflag = 0; 2981 if (!nd->nd_repstat) { 2982 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2983 *tl++ = newnfs_false; 2984 if (eofflag) 2985 *tl = newnfs_true; 2986 else 2987 *tl = newnfs_false; 2988 } 2989 free(cookies, M_TEMP); 2990 free(rbuf, M_TEMP); 2991 2992 out: 2993 NFSEXITCODE2(0, nd); 2994 return (0); 2995 nfsmout: 2996 vput(vp); 2997 NFSEXITCODE2(error, nd); 2998 return (error); 2999 } 3000 3001 /* 3002 * Get the settable attributes out of the mbuf list. 3003 * (Return 0 or EBADRPC) 3004 */ 3005 int 3006 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap, 3007 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p) 3008 { 3009 u_int32_t *tl; 3010 struct nfsv2_sattr *sp; 3011 int error = 0, toclient = 0; 3012 3013 switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) { 3014 case ND_NFSV2: 3015 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 3016 /* 3017 * Some old clients didn't fill in the high order 16bits. 3018 * --> check the low order 2 bytes for 0xffff 3019 */ 3020 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff) 3021 nvap->na_mode = nfstov_mode(sp->sa_mode); 3022 if (sp->sa_uid != newnfs_xdrneg1) 3023 nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid); 3024 if (sp->sa_gid != newnfs_xdrneg1) 3025 nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid); 3026 if (sp->sa_size != newnfs_xdrneg1) 3027 nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size); 3028 if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) { 3029 #ifdef notyet 3030 fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime); 3031 #else 3032 nvap->na_atime.tv_sec = 3033 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec); 3034 nvap->na_atime.tv_nsec = 0; 3035 #endif 3036 } 3037 if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1) 3038 fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime); 3039 break; 3040 case ND_NFSV3: 3041 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3042 if (*tl == newnfs_true) { 3043 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3044 nvap->na_mode = nfstov_mode(*tl); 3045 } 3046 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3047 if (*tl == newnfs_true) { 3048 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3049 nvap->na_uid = fxdr_unsigned(uid_t, *tl); 3050 } 3051 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3052 if (*tl == newnfs_true) { 3053 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3054 nvap->na_gid = fxdr_unsigned(gid_t, *tl); 3055 } 3056 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3057 if (*tl == newnfs_true) { 3058 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 3059 nvap->na_size = fxdr_hyper(tl); 3060 } 3061 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3062 switch (fxdr_unsigned(int, *tl)) { 3063 case NFSV3SATTRTIME_TOCLIENT: 3064 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 3065 fxdr_nfsv3time(tl, &nvap->na_atime); 3066 toclient = 1; 3067 break; 3068 case NFSV3SATTRTIME_TOSERVER: 3069 vfs_timestamp(&nvap->na_atime); 3070 nvap->na_vaflags |= VA_UTIMES_NULL; 3071 break; 3072 } 3073 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3074 switch (fxdr_unsigned(int, *tl)) { 3075 case NFSV3SATTRTIME_TOCLIENT: 3076 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 3077 fxdr_nfsv3time(tl, &nvap->na_mtime); 3078 nvap->na_vaflags &= ~VA_UTIMES_NULL; 3079 break; 3080 case NFSV3SATTRTIME_TOSERVER: 3081 vfs_timestamp(&nvap->na_mtime); 3082 if (!toclient) 3083 nvap->na_vaflags |= VA_UTIMES_NULL; 3084 break; 3085 } 3086 break; 3087 case ND_NFSV4: 3088 error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p); 3089 } 3090 nfsmout: 3091 NFSEXITCODE2(error, nd); 3092 return (error); 3093 } 3094 3095 /* 3096 * Handle the setable attributes for V4. 3097 * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise. 3098 */ 3099 int 3100 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap, 3101 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p) 3102 { 3103 u_int32_t *tl; 3104 int attrsum = 0; 3105 int i, j; 3106 int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0; 3107 int moderet, toclient = 0; 3108 u_char *cp, namestr[NFSV4_SMALLSTR + 1]; 3109 uid_t uid; 3110 gid_t gid; 3111 u_short mode, mask; /* Same type as va_mode. */ 3112 struct vattr va; 3113 3114 error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup); 3115 if (error) 3116 goto nfsmout; 3117 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3118 attrsize = fxdr_unsigned(int, *tl); 3119 3120 /* 3121 * Loop around getting the setable attributes. If an unsupported 3122 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return. 3123 * Once nd_repstat != 0, do not set the attribute value, but keep 3124 * parsing the attribute(s). 3125 */ 3126 if (retnotsup) { 3127 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3128 bitpos = NFSATTRBIT_MAX; 3129 } else { 3130 bitpos = 0; 3131 if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_HIDDEN) || 3132 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SYSTEM)) 3133 nvap->na_flags = 0; 3134 } 3135 moderet = 0; 3136 for (; bitpos < NFSATTRBIT_MAX; bitpos++) { 3137 if (attrsum > attrsize) { 3138 error = NFSERR_BADXDR; 3139 goto nfsmout; 3140 } 3141 if (NFSISSET_ATTRBIT(attrbitp, bitpos)) 3142 switch (bitpos) { 3143 case NFSATTRBIT_SIZE: 3144 NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); 3145 if (!nd->nd_repstat) { 3146 if (vp != NULL && vp->v_type != VREG) 3147 nd->nd_repstat = (vp->v_type == VDIR) ? 3148 NFSERR_ISDIR : NFSERR_INVAL; 3149 else 3150 nvap->na_size = fxdr_hyper(tl); 3151 } 3152 attrsum += NFSX_HYPER; 3153 break; 3154 case NFSATTRBIT_ACL: 3155 error = nfsrv_dissectacl(nd, aclp, true, &aceerr, 3156 &aclsize, p); 3157 if (error) 3158 goto nfsmout; 3159 if (aceerr && !nd->nd_repstat) 3160 nd->nd_repstat = aceerr; 3161 attrsum += aclsize; 3162 break; 3163 case NFSATTRBIT_ARCHIVE: 3164 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3165 if (!nd->nd_repstat) 3166 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3167 attrsum += NFSX_UNSIGNED; 3168 break; 3169 case NFSATTRBIT_HIDDEN: 3170 NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); 3171 if (nd->nd_repstat == 0) { 3172 if (*tl == newnfs_true) 3173 nvap->na_flags |= UF_HIDDEN; 3174 } 3175 attrsum += NFSX_UNSIGNED; 3176 break; 3177 case NFSATTRBIT_MIMETYPE: 3178 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3179 i = fxdr_unsigned(int, *tl); 3180 error = nfsm_advance(nd, NFSM_RNDUP(i), -1); 3181 if (error) 3182 goto nfsmout; 3183 if (!nd->nd_repstat) 3184 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3185 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i)); 3186 break; 3187 case NFSATTRBIT_MODE: 3188 moderet = NFSERR_INVAL; /* Can't do MODESETMASKED. */ 3189 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3190 if (!nd->nd_repstat) 3191 nvap->na_mode = nfstov_mode(*tl); 3192 attrsum += NFSX_UNSIGNED; 3193 break; 3194 case NFSATTRBIT_OWNER: 3195 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3196 j = fxdr_unsigned(int, *tl); 3197 if (j < 0) { 3198 error = NFSERR_BADXDR; 3199 goto nfsmout; 3200 } 3201 if (j > NFSV4_SMALLSTR) 3202 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK); 3203 else 3204 cp = namestr; 3205 error = nfsrv_mtostr(nd, cp, j); 3206 if (error) { 3207 if (j > NFSV4_SMALLSTR) 3208 free(cp, M_NFSSTRING); 3209 goto nfsmout; 3210 } 3211 if (!nd->nd_repstat) { 3212 nd->nd_repstat = nfsv4_strtouid(nd, cp, j, 3213 &uid); 3214 if (!nd->nd_repstat) 3215 nvap->na_uid = uid; 3216 } 3217 if (j > NFSV4_SMALLSTR) 3218 free(cp, M_NFSSTRING); 3219 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); 3220 break; 3221 case NFSATTRBIT_OWNERGROUP: 3222 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3223 j = fxdr_unsigned(int, *tl); 3224 if (j < 0) { 3225 error = NFSERR_BADXDR; 3226 goto nfsmout; 3227 } 3228 if (j > NFSV4_SMALLSTR) 3229 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK); 3230 else 3231 cp = namestr; 3232 error = nfsrv_mtostr(nd, cp, j); 3233 if (error) { 3234 if (j > NFSV4_SMALLSTR) 3235 free(cp, M_NFSSTRING); 3236 goto nfsmout; 3237 } 3238 if (!nd->nd_repstat) { 3239 nd->nd_repstat = nfsv4_strtogid(nd, cp, j, 3240 &gid); 3241 if (!nd->nd_repstat) 3242 nvap->na_gid = gid; 3243 } 3244 if (j > NFSV4_SMALLSTR) 3245 free(cp, M_NFSSTRING); 3246 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); 3247 break; 3248 case NFSATTRBIT_SYSTEM: 3249 NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); 3250 if (nd->nd_repstat == 0) { 3251 if (*tl == newnfs_true) 3252 nvap->na_flags |= UF_SYSTEM; 3253 } 3254 attrsum += NFSX_UNSIGNED; 3255 break; 3256 case NFSATTRBIT_TIMEACCESSSET: 3257 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3258 attrsum += NFSX_UNSIGNED; 3259 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) { 3260 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3261 if (!nd->nd_repstat) 3262 fxdr_nfsv4time(tl, &nvap->na_atime); 3263 toclient = 1; 3264 attrsum += NFSX_V4TIME; 3265 } else if (!nd->nd_repstat) { 3266 vfs_timestamp(&nvap->na_atime); 3267 nvap->na_vaflags |= VA_UTIMES_NULL; 3268 } 3269 break; 3270 case NFSATTRBIT_TIMEBACKUP: 3271 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3272 if (!nd->nd_repstat) 3273 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3274 attrsum += NFSX_V4TIME; 3275 break; 3276 case NFSATTRBIT_TIMECREATE: 3277 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3278 if (!nd->nd_repstat) 3279 fxdr_nfsv4time(tl, &nvap->na_btime); 3280 attrsum += NFSX_V4TIME; 3281 break; 3282 case NFSATTRBIT_TIMEMODIFYSET: 3283 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3284 attrsum += NFSX_UNSIGNED; 3285 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) { 3286 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3287 if (!nd->nd_repstat) 3288 fxdr_nfsv4time(tl, &nvap->na_mtime); 3289 nvap->na_vaflags &= ~VA_UTIMES_NULL; 3290 attrsum += NFSX_V4TIME; 3291 } else if (!nd->nd_repstat) { 3292 vfs_timestamp(&nvap->na_mtime); 3293 if (!toclient) 3294 nvap->na_vaflags |= VA_UTIMES_NULL; 3295 } 3296 break; 3297 case NFSATTRBIT_MODESETMASKED: 3298 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 3299 mode = fxdr_unsigned(u_short, *tl++); 3300 mask = fxdr_unsigned(u_short, *tl); 3301 /* 3302 * vp == NULL implies an Open/Create operation. 3303 * This attribute can only be used for Setattr and 3304 * only for NFSv4.1 or higher. 3305 * If moderet != 0, a mode attribute has also been 3306 * specified and this attribute cannot be done in the 3307 * same Setattr operation. 3308 */ 3309 if (!nd->nd_repstat) { 3310 if ((nd->nd_flag & ND_NFSV41) == 0) 3311 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3312 else if ((mode & ~07777) != 0 || 3313 (mask & ~07777) != 0 || vp == NULL) 3314 nd->nd_repstat = NFSERR_INVAL; 3315 else if (moderet == 0) 3316 moderet = VOP_GETATTR(vp, &va, 3317 nd->nd_cred); 3318 if (moderet == 0) 3319 nvap->na_mode = (mode & mask) | 3320 (va.va_mode & ~mask); 3321 else 3322 nd->nd_repstat = moderet; 3323 } 3324 attrsum += 2 * NFSX_UNSIGNED; 3325 break; 3326 case NFSATTRBIT_MODEUMASK: 3327 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 3328 mode = fxdr_unsigned(u_short, *tl++); 3329 mask = fxdr_unsigned(u_short, *tl); 3330 /* 3331 * If moderet != 0, mode has already been done. 3332 * If vp != NULL, this is not a file object creation. 3333 */ 3334 if (!nd->nd_repstat) { 3335 if ((nd->nd_flag & ND_NFSV42) == 0) 3336 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3337 else if ((mask & ~0777) != 0 || vp != NULL || 3338 moderet != 0) 3339 nd->nd_repstat = NFSERR_INVAL; 3340 else 3341 nvap->na_mode = (mode & ~mask); 3342 } 3343 attrsum += 2 * NFSX_UNSIGNED; 3344 break; 3345 default: 3346 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3347 /* 3348 * set bitpos so we drop out of the loop. 3349 */ 3350 bitpos = NFSATTRBIT_MAX; 3351 break; 3352 } 3353 } 3354 3355 /* 3356 * some clients pad the attrlist, so we need to skip over the 3357 * padding. This also skips over unparsed non-supported attributes. 3358 */ 3359 if (attrsum > attrsize) { 3360 error = NFSERR_BADXDR; 3361 } else { 3362 attrsize = NFSM_RNDUP(attrsize); 3363 if (attrsum < attrsize) 3364 error = nfsm_advance(nd, attrsize - attrsum, -1); 3365 } 3366 nfsmout: 3367 NFSEXITCODE2(error, nd); 3368 return (error); 3369 } 3370 3371 /* 3372 * Check/setup export credentials. 3373 */ 3374 int 3375 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp, 3376 struct ucred *credanon, bool testsec) 3377 { 3378 int error; 3379 3380 /* 3381 * Check/setup credentials. 3382 */ 3383 if (nd->nd_flag & ND_GSS) 3384 exp->nes_exflag &= ~MNT_EXPORTANON; 3385 3386 /* 3387 * Check to see if the operation is allowed for this security flavor. 3388 */ 3389 error = 0; 3390 if (testsec) { 3391 error = nfsvno_testexp(nd, exp); 3392 if (error != 0) 3393 goto out; 3394 } 3395 3396 /* 3397 * Check to see if the file system is exported V4 only. 3398 */ 3399 if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) { 3400 error = NFSERR_PROGNOTV4; 3401 goto out; 3402 } 3403 3404 /* 3405 * Now, map the user credentials. 3406 * (Note that ND_AUTHNONE will only be set for an NFSv3 3407 * Fsinfo RPC. If set for anything else, this code might need 3408 * to change.) 3409 */ 3410 if (NFSVNO_EXPORTED(exp)) { 3411 if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) || 3412 NFSVNO_EXPORTANON(exp) || 3413 (nd->nd_flag & ND_AUTHNONE) != 0) { 3414 nd->nd_cred->cr_uid = credanon->cr_uid; 3415 /* 3416 * 'credanon' is already a 'struct ucred' that was built 3417 * internally with calls to crsetgroups_fallback(), so 3418 * we don't need a fallback here. 3419 */ 3420 crsetgroups(nd->nd_cred, credanon->cr_ngroups, 3421 credanon->cr_groups); 3422 } else if ((nd->nd_flag & ND_GSS) == 0) { 3423 /* 3424 * If using AUTH_SYS, call nfsrv_getgrpscred() to see 3425 * if there is a replacement credential with a group 3426 * list set up by "nfsuserd -manage-gids". 3427 * If there is no replacement, nfsrv_getgrpscred() 3428 * simply returns its argument. 3429 */ 3430 nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred); 3431 } 3432 } 3433 3434 out: 3435 NFSEXITCODE2(error, nd); 3436 return (error); 3437 } 3438 3439 /* 3440 * Check exports. 3441 */ 3442 int 3443 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp, 3444 struct ucred **credp) 3445 { 3446 int error; 3447 3448 error = 0; 3449 *credp = NULL; 3450 MNT_ILOCK(mp); 3451 if (mp->mnt_exjail == NULL || 3452 mp->mnt_exjail->cr_prison != curthread->td_ucred->cr_prison) 3453 error = EACCES; 3454 MNT_IUNLOCK(mp); 3455 if (error == 0) 3456 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp, 3457 &exp->nes_numsecflavor, exp->nes_secflavors); 3458 if (error) { 3459 if (NFSD_VNET(nfs_rootfhset)) { 3460 exp->nes_exflag = 0; 3461 exp->nes_numsecflavor = 0; 3462 error = 0; 3463 } 3464 } else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor > 3465 MAXSECFLAVORS) { 3466 printf("nfsvno_checkexp: numsecflavors out of range\n"); 3467 exp->nes_numsecflavor = 0; 3468 error = EACCES; 3469 } 3470 NFSEXITCODE(error); 3471 return (error); 3472 } 3473 3474 /* 3475 * Get a vnode for a file handle and export stuff. 3476 */ 3477 int 3478 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam, 3479 int lktype, struct vnode **vpp, struct nfsexstuff *exp, 3480 struct ucred **credp) 3481 { 3482 int error; 3483 3484 *credp = NULL; 3485 exp->nes_numsecflavor = 0; 3486 error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp); 3487 if (error != 0) 3488 /* Make sure the server replies ESTALE to the client. */ 3489 error = ESTALE; 3490 if (nam && !error) { 3491 MNT_ILOCK(mp); 3492 if (mp->mnt_exjail == NULL || 3493 mp->mnt_exjail->cr_prison != curthread->td_ucred->cr_prison) 3494 error = EACCES; 3495 MNT_IUNLOCK(mp); 3496 if (error == 0) 3497 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp, 3498 &exp->nes_numsecflavor, exp->nes_secflavors); 3499 if (error) { 3500 if (NFSD_VNET(nfs_rootfhset)) { 3501 exp->nes_exflag = 0; 3502 exp->nes_numsecflavor = 0; 3503 error = 0; 3504 } else { 3505 vput(*vpp); 3506 } 3507 } else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor > 3508 MAXSECFLAVORS) { 3509 printf("nfsvno_fhtovp: numsecflavors out of range\n"); 3510 exp->nes_numsecflavor = 0; 3511 error = EACCES; 3512 vput(*vpp); 3513 } 3514 } 3515 NFSEXITCODE(error); 3516 return (error); 3517 } 3518 3519 /* 3520 * nfsd_fhtovp() - convert a fh to a vnode ptr 3521 * - look up fsid in mount list (if not found ret error) 3522 * - get vp and export rights by calling nfsvno_fhtovp() 3523 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon 3524 * for AUTH_SYS 3525 * - if mpp != NULL, return the mount point so that it can 3526 * be used for vn_finished_write() by the caller 3527 */ 3528 void 3529 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype, 3530 struct vnode **vpp, struct nfsexstuff *exp, 3531 struct mount **mpp, int startwrite, int nextop) 3532 { 3533 struct mount *mp, *mpw; 3534 struct ucred *credanon; 3535 fhandle_t *fhp; 3536 int error; 3537 3538 if (mpp != NULL) 3539 *mpp = NULL; 3540 *vpp = NULL; 3541 fhp = (fhandle_t *)nfp->nfsrvfh_data; 3542 mp = vfs_busyfs(&fhp->fh_fsid); 3543 if (mp == NULL) { 3544 nd->nd_repstat = ESTALE; 3545 goto out; 3546 } 3547 3548 if (startwrite) { 3549 mpw = mp; 3550 error = vn_start_write(NULL, &mpw, V_WAIT); 3551 if (error != 0) { 3552 mpw = NULL; 3553 vfs_unbusy(mp); 3554 nd->nd_repstat = ESTALE; 3555 goto out; 3556 } 3557 if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp))) 3558 lktype = LK_EXCLUSIVE; 3559 } else 3560 mpw = NULL; 3561 3562 nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp, 3563 &credanon); 3564 vfs_unbusy(mp); 3565 3566 if (nd->nd_repstat == 0 && 3567 nfp->nfsrvfh_len >= NFSX_MYFH + NFSX_V4NAMEDDIRFH && 3568 nfp->nfsrvfh_len <= NFSX_MYFH + NFSX_V4NAMEDATTRFH) { 3569 if (nfp->nfsrvfh_len == NFSX_MYFH + NFSX_V4NAMEDDIRFH) 3570 vn_irflag_set_cond(*vpp, VIRF_NAMEDDIR); 3571 else 3572 vn_irflag_set_cond(*vpp, VIRF_NAMEDATTR); 3573 } 3574 3575 /* 3576 * For NFSv4 without a pseudo root fs, unexported file handles 3577 * can be returned, so that Lookup works everywhere. 3578 */ 3579 if (!nd->nd_repstat && exp->nes_exflag == 0 && 3580 !(nd->nd_flag & ND_NFSV4)) { 3581 vput(*vpp); 3582 *vpp = NULL; 3583 nd->nd_repstat = EACCES; 3584 } 3585 3586 /* 3587 * Personally, I've never seen any point in requiring a 3588 * reserved port#, since only in the rare case where the 3589 * clients are all boxes with secure system privileges, 3590 * does it provide any enhanced security, but... some people 3591 * believe it to be useful and keep putting this code back in. 3592 * (There is also some "security checker" out there that 3593 * complains if the nfs server doesn't enforce this.) 3594 * However, note the following: 3595 * RFC3530 (NFSv4) specifies that a reserved port# not be 3596 * required. 3597 * RFC2623 recommends that, if a reserved port# is checked for, 3598 * that there be a way to turn that off--> ifdef'd. 3599 */ 3600 #ifdef NFS_REQRSVPORT 3601 if (!nd->nd_repstat) { 3602 struct sockaddr_in *saddr; 3603 struct sockaddr_in6 *saddr6; 3604 3605 saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *); 3606 saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *); 3607 if (!(nd->nd_flag & ND_NFSV4) && 3608 ((saddr->sin_family == AF_INET && 3609 ntohs(saddr->sin_port) >= IPPORT_RESERVED) || 3610 (saddr6->sin6_family == AF_INET6 && 3611 ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) { 3612 vput(*vpp); 3613 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 3614 } 3615 } 3616 #endif /* NFS_REQRSVPORT */ 3617 3618 /* 3619 * Check/setup credentials. 3620 */ 3621 if (!nd->nd_repstat) { 3622 nd->nd_saveduid = nd->nd_cred->cr_uid; 3623 nd->nd_repstat = nfsd_excred(nd, exp, credanon, 3624 nfsrv_checkwrongsec(nd, nextop, (*vpp)->v_type)); 3625 if (nd->nd_repstat) 3626 vput(*vpp); 3627 } 3628 if (credanon != NULL) 3629 crfree(credanon); 3630 if (nd->nd_repstat) { 3631 vn_finished_write(mpw); 3632 *vpp = NULL; 3633 } else if (mpp != NULL) { 3634 *mpp = mpw; 3635 } 3636 3637 out: 3638 NFSEXITCODE2(0, nd); 3639 } 3640 3641 /* 3642 * glue for fp. 3643 */ 3644 static int 3645 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp) 3646 { 3647 struct filedesc *fdp; 3648 struct file *fp; 3649 int error = 0; 3650 3651 fdp = p->td_proc->p_fd; 3652 if (fd < 0 || fd >= fdp->fd_nfiles || 3653 (fp = fdp->fd_ofiles[fd].fde_file) == NULL) { 3654 error = EBADF; 3655 goto out; 3656 } 3657 *fpp = fp; 3658 3659 out: 3660 NFSEXITCODE(error); 3661 return (error); 3662 } 3663 3664 /* 3665 * Called from nfssvc() to update the exports list. Just call 3666 * vfs_export(). This has to be done, since the v4 root fake fs isn't 3667 * in the mount list. 3668 */ 3669 int 3670 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p) 3671 { 3672 struct nfsex_args *nfsexargp = (struct nfsex_args *)argp; 3673 int error = 0; 3674 struct nameidata nd; 3675 fhandle_t fh; 3676 3677 error = vfs_export(NFSD_VNET(nfsv4root_mnt), &nfsexargp->export, false); 3678 if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0) 3679 NFSD_VNET(nfs_rootfhset) = 0; 3680 else if (error == 0) { 3681 if (nfsexargp->fspec == NULL) { 3682 error = EPERM; 3683 goto out; 3684 } 3685 /* 3686 * If fspec != NULL, this is the v4root path. 3687 */ 3688 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, nfsexargp->fspec); 3689 if ((error = namei(&nd)) != 0) 3690 goto out; 3691 error = nfsvno_getfh(nd.ni_vp, &fh, p); 3692 vrele(nd.ni_vp); 3693 if (!error) { 3694 NFSD_VNET(nfs_rootfh).nfsrvfh_len = NFSX_MYFH; 3695 NFSBCOPY((caddr_t)&fh, 3696 NFSD_VNET(nfs_rootfh).nfsrvfh_data, 3697 sizeof (fhandle_t)); 3698 NFSD_VNET(nfs_rootfhset) = 1; 3699 } 3700 } 3701 3702 out: 3703 NFSEXITCODE(error); 3704 return (error); 3705 } 3706 3707 /* 3708 * This function needs to test to see if the system is near its limit 3709 * for memory allocation via malloc() or mget() and return True iff 3710 * either of these resources are near their limit. 3711 * XXX (For now, this is just a stub.) 3712 */ 3713 int nfsrv_testmalloclimit = 0; 3714 int 3715 nfsrv_mallocmget_limit(void) 3716 { 3717 static int printmesg = 0; 3718 static int testval = 1; 3719 3720 if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) { 3721 if ((printmesg++ % 100) == 0) 3722 printf("nfsd: malloc/mget near limit\n"); 3723 return (1); 3724 } 3725 return (0); 3726 } 3727 3728 /* 3729 * BSD specific initialization of a mount point. 3730 */ 3731 void 3732 nfsd_mntinit(void) 3733 { 3734 3735 NFSD_LOCK(); 3736 if (NFSD_VNET(nfsrv_mntinited)) { 3737 NFSD_UNLOCK(); 3738 return; 3739 } 3740 NFSD_VNET(nfsrv_mntinited) = true; 3741 nfsrvd_init(0); 3742 NFSD_UNLOCK(); 3743 3744 NFSD_VNET(nfsv4root_mnt) = malloc(sizeof(struct mount), M_TEMP, 3745 M_WAITOK | M_ZERO); 3746 NFSD_VNET(nfsv4root_mnt)->mnt_flag = (MNT_RDONLY | MNT_EXPORTED); 3747 mtx_init(&NFSD_VNET(nfsv4root_mnt)->mnt_mtx, "nfs4mnt", NULL, MTX_DEF); 3748 lockinit(&NFSD_VNET(nfsv4root_mnt)->mnt_explock, PVFS, "explock", 0, 0); 3749 TAILQ_INIT(&NFSD_VNET(nfsv4root_mnt)->mnt_nvnodelist); 3750 TAILQ_INIT(&NFSD_VNET(nfsv4root_mnt)->mnt_lazyvnodelist); 3751 NFSD_VNET(nfsv4root_mnt)->mnt_export = NULL; 3752 TAILQ_INIT(&NFSD_VNET(nfsv4root_opt)); 3753 TAILQ_INIT(&NFSD_VNET(nfsv4root_newopt)); 3754 NFSD_VNET(nfsv4root_mnt)->mnt_opt = &NFSD_VNET(nfsv4root_opt); 3755 NFSD_VNET(nfsv4root_mnt)->mnt_optnew = &NFSD_VNET(nfsv4root_newopt); 3756 NFSD_VNET(nfsv4root_mnt)->mnt_nvnodelistsize = 0; 3757 NFSD_VNET(nfsv4root_mnt)->mnt_lazyvnodelistsize = 0; 3758 callout_init(&NFSD_VNET(nfsd_callout), 1); 3759 3760 nfsrvd_initcache(); 3761 nfsd_init(); 3762 } 3763 3764 static void 3765 nfsd_timer(void *arg) 3766 { 3767 struct vnet *vnetp; 3768 3769 vnetp = (struct vnet *)arg; 3770 NFSD_CURVNET_SET_QUIET(vnetp); 3771 nfsrv_servertimer(vnetp); 3772 callout_reset_sbt(&NFSD_VNET(nfsd_callout), SBT_1S, SBT_1S, nfsd_timer, 3773 arg, 0); 3774 NFSD_CURVNET_RESTORE(); 3775 } 3776 3777 /* 3778 * Get a vnode for a file handle, without checking exports, etc. 3779 */ 3780 struct vnode * 3781 nfsvno_getvp(fhandle_t *fhp) 3782 { 3783 struct mount *mp; 3784 struct vnode *vp; 3785 int error; 3786 3787 mp = vfs_busyfs(&fhp->fh_fsid); 3788 if (mp == NULL) 3789 return (NULL); 3790 error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp); 3791 vfs_unbusy(mp); 3792 if (error) 3793 return (NULL); 3794 return (vp); 3795 } 3796 3797 /* 3798 * Do a local VOP_ADVLOCK(). 3799 */ 3800 int 3801 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first, 3802 u_int64_t end, struct thread *td) 3803 { 3804 int error = 0; 3805 struct flock fl; 3806 u_int64_t tlen; 3807 3808 if (nfsrv_dolocallocks == 0) 3809 goto out; 3810 ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked"); 3811 3812 fl.l_whence = SEEK_SET; 3813 fl.l_type = ftype; 3814 fl.l_start = (off_t)first; 3815 if (end == NFS64BITSSET) { 3816 fl.l_len = 0; 3817 } else { 3818 tlen = end - first; 3819 fl.l_len = (off_t)tlen; 3820 } 3821 /* 3822 * For FreeBSD8, the l_pid and l_sysid must be set to the same 3823 * values for all calls, so that all locks will be held by the 3824 * nfsd server. (The nfsd server handles conflicts between the 3825 * various clients.) 3826 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024 3827 * bytes, so it can't be put in l_sysid. 3828 */ 3829 if (nfsv4_sysid == 0) 3830 nfsv4_sysid = nlm_acquire_next_sysid(); 3831 fl.l_pid = (pid_t)0; 3832 fl.l_sysid = (int)nfsv4_sysid; 3833 3834 if (ftype == F_UNLCK) 3835 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl, 3836 (F_POSIX | F_REMOTE)); 3837 else 3838 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl, 3839 (F_POSIX | F_REMOTE)); 3840 3841 out: 3842 NFSEXITCODE(error); 3843 return (error); 3844 } 3845 3846 /* 3847 * Check the nfsv4 root exports. 3848 */ 3849 int 3850 nfsvno_v4rootexport(struct nfsrv_descript *nd) 3851 { 3852 struct ucred *credanon; 3853 int error = 0, numsecflavor, secflavors[MAXSECFLAVORS], i; 3854 uint64_t exflags; 3855 3856 error = vfs_stdcheckexp(NFSD_VNET(nfsv4root_mnt), nd->nd_nam, &exflags, 3857 &credanon, &numsecflavor, secflavors); 3858 if (error) { 3859 error = NFSERR_PROGUNAVAIL; 3860 goto out; 3861 } 3862 if (credanon != NULL) 3863 crfree(credanon); 3864 for (i = 0; i < numsecflavor; i++) { 3865 if (secflavors[i] == AUTH_SYS) 3866 nd->nd_flag |= ND_EXAUTHSYS; 3867 else if (secflavors[i] == RPCSEC_GSS_KRB5) 3868 nd->nd_flag |= ND_EXGSS; 3869 else if (secflavors[i] == RPCSEC_GSS_KRB5I) 3870 nd->nd_flag |= ND_EXGSSINTEGRITY; 3871 else if (secflavors[i] == RPCSEC_GSS_KRB5P) 3872 nd->nd_flag |= ND_EXGSSPRIVACY; 3873 } 3874 3875 /* And set ND_EXxx flags for TLS. */ 3876 if ((exflags & MNT_EXTLS) != 0) { 3877 nd->nd_flag |= ND_EXTLS; 3878 if ((exflags & MNT_EXTLSCERT) != 0) 3879 nd->nd_flag |= ND_EXTLSCERT; 3880 if ((exflags & MNT_EXTLSCERTUSER) != 0) 3881 nd->nd_flag |= ND_EXTLSCERTUSER; 3882 } 3883 3884 out: 3885 NFSEXITCODE(error); 3886 return (error); 3887 } 3888 3889 /* 3890 * Nfs server pseudo system call for the nfsd's 3891 */ 3892 /* 3893 * MPSAFE 3894 */ 3895 static int 3896 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap) 3897 { 3898 struct file *fp; 3899 struct nfsd_addsock_args sockarg; 3900 struct nfsd_nfsd_args nfsdarg; 3901 struct nfsd_nfsd_oargs onfsdarg; 3902 struct nfsd_pnfsd_args pnfsdarg; 3903 struct vnode *vp, *nvp, *curdvp; 3904 struct pnfsdsfile *pf; 3905 struct nfsdevice *ds, *fds; 3906 cap_rights_t rights; 3907 int buflen, error, ret; 3908 char *buf, *cp, *cp2, *cp3; 3909 char fname[PNFS_FILENAME_LEN + 1]; 3910 3911 NFSD_CURVNET_SET(NFSD_TD_TO_VNET(td)); 3912 if (uap->flag & NFSSVC_NFSDADDSOCK) { 3913 error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg)); 3914 if (error) 3915 goto out; 3916 /* 3917 * Since we don't know what rights might be required, 3918 * pretend that we need them all. It is better to be too 3919 * careful than too reckless. 3920 */ 3921 error = fget(td, sockarg.sock, 3922 cap_rights_init_one(&rights, CAP_SOCK_SERVER), &fp); 3923 if (error != 0) 3924 goto out; 3925 if (fp->f_type != DTYPE_SOCKET) { 3926 fdrop(fp, td); 3927 error = EPERM; 3928 goto out; 3929 } 3930 error = nfsrvd_addsock(fp); 3931 fdrop(fp, td); 3932 } else if (uap->flag & NFSSVC_NFSDNFSD) { 3933 if (uap->argp == NULL) { 3934 error = EINVAL; 3935 goto out; 3936 } 3937 if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) { 3938 error = copyin(uap->argp, &onfsdarg, sizeof(onfsdarg)); 3939 if (error == 0) { 3940 nfsdarg.principal = onfsdarg.principal; 3941 nfsdarg.minthreads = onfsdarg.minthreads; 3942 nfsdarg.maxthreads = onfsdarg.maxthreads; 3943 nfsdarg.version = 1; 3944 nfsdarg.addr = NULL; 3945 nfsdarg.addrlen = 0; 3946 nfsdarg.dnshost = NULL; 3947 nfsdarg.dnshostlen = 0; 3948 nfsdarg.dspath = NULL; 3949 nfsdarg.dspathlen = 0; 3950 nfsdarg.mdspath = NULL; 3951 nfsdarg.mdspathlen = 0; 3952 nfsdarg.mirrorcnt = 1; 3953 } 3954 } else 3955 error = copyin(uap->argp, &nfsdarg, sizeof(nfsdarg)); 3956 if (error) 3957 goto out; 3958 if (nfsdarg.addrlen > 0 && nfsdarg.addrlen < 10000 && 3959 nfsdarg.dnshostlen > 0 && nfsdarg.dnshostlen < 10000 && 3960 nfsdarg.dspathlen > 0 && nfsdarg.dspathlen < 10000 && 3961 nfsdarg.mdspathlen > 0 && nfsdarg.mdspathlen < 10000 && 3962 nfsdarg.mirrorcnt >= 1 && 3963 nfsdarg.mirrorcnt <= NFSDEV_MAXMIRRORS && 3964 nfsdarg.addr != NULL && nfsdarg.dnshost != NULL && 3965 nfsdarg.dspath != NULL && nfsdarg.mdspath != NULL) { 3966 NFSD_DEBUG(1, "addrlen=%d dspathlen=%d dnslen=%d" 3967 " mdspathlen=%d mirrorcnt=%d\n", nfsdarg.addrlen, 3968 nfsdarg.dspathlen, nfsdarg.dnshostlen, 3969 nfsdarg.mdspathlen, nfsdarg.mirrorcnt); 3970 cp = malloc(nfsdarg.addrlen + 1, M_TEMP, M_WAITOK); 3971 error = copyin(nfsdarg.addr, cp, nfsdarg.addrlen); 3972 if (error != 0) { 3973 free(cp, M_TEMP); 3974 goto out; 3975 } 3976 cp[nfsdarg.addrlen] = '\0'; /* Ensure nul term. */ 3977 nfsdarg.addr = cp; 3978 cp = malloc(nfsdarg.dnshostlen + 1, M_TEMP, M_WAITOK); 3979 error = copyin(nfsdarg.dnshost, cp, nfsdarg.dnshostlen); 3980 if (error != 0) { 3981 free(nfsdarg.addr, M_TEMP); 3982 free(cp, M_TEMP); 3983 goto out; 3984 } 3985 cp[nfsdarg.dnshostlen] = '\0'; /* Ensure nul term. */ 3986 nfsdarg.dnshost = cp; 3987 cp = malloc(nfsdarg.dspathlen + 1, M_TEMP, M_WAITOK); 3988 error = copyin(nfsdarg.dspath, cp, nfsdarg.dspathlen); 3989 if (error != 0) { 3990 free(nfsdarg.addr, M_TEMP); 3991 free(nfsdarg.dnshost, M_TEMP); 3992 free(cp, M_TEMP); 3993 goto out; 3994 } 3995 cp[nfsdarg.dspathlen] = '\0'; /* Ensure nul term. */ 3996 nfsdarg.dspath = cp; 3997 cp = malloc(nfsdarg.mdspathlen + 1, M_TEMP, M_WAITOK); 3998 error = copyin(nfsdarg.mdspath, cp, nfsdarg.mdspathlen); 3999 if (error != 0) { 4000 free(nfsdarg.addr, M_TEMP); 4001 free(nfsdarg.dnshost, M_TEMP); 4002 free(nfsdarg.dspath, M_TEMP); 4003 free(cp, M_TEMP); 4004 goto out; 4005 } 4006 cp[nfsdarg.mdspathlen] = '\0'; /* Ensure nul term. */ 4007 nfsdarg.mdspath = cp; 4008 } else { 4009 nfsdarg.addr = NULL; 4010 nfsdarg.addrlen = 0; 4011 nfsdarg.dnshost = NULL; 4012 nfsdarg.dnshostlen = 0; 4013 nfsdarg.dspath = NULL; 4014 nfsdarg.dspathlen = 0; 4015 nfsdarg.mdspath = NULL; 4016 nfsdarg.mdspathlen = 0; 4017 nfsdarg.mirrorcnt = 1; 4018 } 4019 nfsd_timer(NFSD_TD_TO_VNET(td)); 4020 error = nfsrvd_nfsd(td, &nfsdarg); 4021 callout_drain(&NFSD_VNET(nfsd_callout)); 4022 free(nfsdarg.addr, M_TEMP); 4023 free(nfsdarg.dnshost, M_TEMP); 4024 free(nfsdarg.dspath, M_TEMP); 4025 free(nfsdarg.mdspath, M_TEMP); 4026 } else if (uap->flag & NFSSVC_PNFSDS) { 4027 error = copyin(uap->argp, &pnfsdarg, sizeof(pnfsdarg)); 4028 if (error == 0 && (pnfsdarg.op == PNFSDOP_DELDSSERVER || 4029 pnfsdarg.op == PNFSDOP_FORCEDELDS)) { 4030 cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 4031 error = copyinstr(pnfsdarg.dspath, cp, PATH_MAX + 1, 4032 NULL); 4033 if (error == 0) 4034 error = nfsrv_deldsserver(pnfsdarg.op, cp, td); 4035 free(cp, M_TEMP); 4036 } else if (error == 0 && pnfsdarg.op == PNFSDOP_COPYMR) { 4037 cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 4038 buflen = sizeof(*pf) * NFSDEV_MAXMIRRORS; 4039 buf = malloc(buflen, M_TEMP, M_WAITOK); 4040 error = copyinstr(pnfsdarg.mdspath, cp, PATH_MAX + 1, 4041 NULL); 4042 NFSD_DEBUG(4, "pnfsdcopymr cp mdspath=%d\n", error); 4043 if (error == 0 && pnfsdarg.dspath != NULL) { 4044 cp2 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 4045 error = copyinstr(pnfsdarg.dspath, cp2, 4046 PATH_MAX + 1, NULL); 4047 NFSD_DEBUG(4, "pnfsdcopymr cp dspath=%d\n", 4048 error); 4049 } else 4050 cp2 = NULL; 4051 if (error == 0 && pnfsdarg.curdspath != NULL) { 4052 cp3 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 4053 error = copyinstr(pnfsdarg.curdspath, cp3, 4054 PATH_MAX + 1, NULL); 4055 NFSD_DEBUG(4, "pnfsdcopymr cp curdspath=%d\n", 4056 error); 4057 } else 4058 cp3 = NULL; 4059 curdvp = NULL; 4060 fds = NULL; 4061 if (error == 0) 4062 error = nfsrv_mdscopymr(cp, cp2, cp3, buf, 4063 &buflen, fname, td, &vp, &nvp, &pf, &ds, 4064 &fds); 4065 NFSD_DEBUG(4, "nfsrv_mdscopymr=%d\n", error); 4066 if (error == 0) { 4067 if (pf->dsf_dir >= nfsrv_dsdirsize) { 4068 printf("copymr: dsdir out of range\n"); 4069 pf->dsf_dir = 0; 4070 } 4071 NFSD_DEBUG(4, "copymr: buflen=%d\n", buflen); 4072 error = nfsrv_copymr(vp, nvp, 4073 ds->nfsdev_dsdir[pf->dsf_dir], ds, pf, 4074 (struct pnfsdsfile *)buf, 4075 buflen / sizeof(*pf), td->td_ucred, td); 4076 vput(vp); 4077 vput(nvp); 4078 if (fds != NULL && error == 0) { 4079 curdvp = fds->nfsdev_dsdir[pf->dsf_dir]; 4080 ret = vn_lock(curdvp, LK_EXCLUSIVE); 4081 if (ret == 0) { 4082 nfsrv_dsremove(curdvp, fname, 4083 td->td_ucred, td); 4084 NFSVOPUNLOCK(curdvp); 4085 } 4086 } 4087 NFSD_DEBUG(4, "nfsrv_copymr=%d\n", error); 4088 } 4089 free(cp, M_TEMP); 4090 free(cp2, M_TEMP); 4091 free(cp3, M_TEMP); 4092 free(buf, M_TEMP); 4093 } 4094 } else { 4095 error = nfssvc_srvcall(td, uap, td->td_ucred); 4096 } 4097 4098 out: 4099 NFSD_CURVNET_RESTORE(); 4100 NFSEXITCODE(error); 4101 return (error); 4102 } 4103 4104 static int 4105 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred) 4106 { 4107 struct nfsex_args export; 4108 struct nfsex_oldargs oexp; 4109 struct file *fp = NULL; 4110 int stablefd, i, len; 4111 struct nfsd_clid adminrevoke; 4112 struct nfsd_dumplist dumplist; 4113 struct nfsd_dumpclients *dumpclients; 4114 struct nfsd_dumplocklist dumplocklist; 4115 struct nfsd_dumplocks *dumplocks; 4116 struct nameidata nd; 4117 vnode_t vp; 4118 int error = EINVAL, igotlock; 4119 struct proc *procp; 4120 gid_t *grps; 4121 4122 if (uap->flag & NFSSVC_PUBLICFH) { 4123 NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data, 4124 sizeof (fhandle_t)); 4125 error = copyin(uap->argp, 4126 &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t)); 4127 if (!error) 4128 nfs_pubfhset = 1; 4129 } else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) == 4130 (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) { 4131 error = copyin(uap->argp,(caddr_t)&export, 4132 sizeof (struct nfsex_args)); 4133 if (!error) { 4134 grps = NULL; 4135 if (export.export.ex_ngroups > NGROUPS_MAX || 4136 export.export.ex_ngroups < 0) 4137 error = EINVAL; 4138 else if (export.export.ex_ngroups > 0) { 4139 grps = malloc(export.export.ex_ngroups * 4140 sizeof(gid_t), M_TEMP, M_WAITOK); 4141 error = copyin(export.export.ex_groups, grps, 4142 export.export.ex_ngroups * sizeof(gid_t)); 4143 export.export.ex_groups = grps; 4144 } else 4145 export.export.ex_groups = NULL; 4146 if (!error) 4147 error = nfsrv_v4rootexport(&export, cred, p); 4148 free(grps, M_TEMP); 4149 } 4150 } else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) == 4151 NFSSVC_V4ROOTEXPORT) { 4152 error = copyin(uap->argp,(caddr_t)&oexp, 4153 sizeof (struct nfsex_oldargs)); 4154 if (!error) { 4155 memset(&export.export, 0, sizeof(export.export)); 4156 export.export.ex_flags = (uint64_t)oexp.export.ex_flags; 4157 export.export.ex_root = oexp.export.ex_root; 4158 export.export.ex_uid = oexp.export.ex_anon.cr_uid; 4159 export.export.ex_ngroups = 4160 oexp.export.ex_anon.cr_ngroups; 4161 export.export.ex_groups = NULL; 4162 if (export.export.ex_ngroups > XU_NGROUPS || 4163 export.export.ex_ngroups < 0) 4164 error = EINVAL; 4165 else if (export.export.ex_ngroups > 0) { 4166 export.export.ex_groups = malloc( 4167 export.export.ex_ngroups * sizeof(gid_t), 4168 M_TEMP, M_WAITOK); 4169 for (i = 0; i < export.export.ex_ngroups; i++) 4170 export.export.ex_groups[i] = 4171 oexp.export.ex_anon.cr_groups[i]; 4172 } 4173 export.export.ex_addr = oexp.export.ex_addr; 4174 export.export.ex_addrlen = oexp.export.ex_addrlen; 4175 export.export.ex_mask = oexp.export.ex_mask; 4176 export.export.ex_masklen = oexp.export.ex_masklen; 4177 export.export.ex_indexfile = oexp.export.ex_indexfile; 4178 export.export.ex_numsecflavors = 4179 oexp.export.ex_numsecflavors; 4180 if (export.export.ex_numsecflavors >= MAXSECFLAVORS || 4181 export.export.ex_numsecflavors < 0) 4182 error = EINVAL; 4183 else { 4184 for (i = 0; i < export.export.ex_numsecflavors; 4185 i++) 4186 export.export.ex_secflavors[i] = 4187 oexp.export.ex_secflavors[i]; 4188 } 4189 export.fspec = oexp.fspec; 4190 if (error == 0) 4191 error = nfsrv_v4rootexport(&export, cred, p); 4192 free(export.export.ex_groups, M_TEMP); 4193 } 4194 } else if (uap->flag & NFSSVC_NOPUBLICFH) { 4195 nfs_pubfhset = 0; 4196 error = 0; 4197 } else if (uap->flag & NFSSVC_STABLERESTART) { 4198 error = copyin(uap->argp, (caddr_t)&stablefd, 4199 sizeof (int)); 4200 if (!error) 4201 error = fp_getfvp(p, stablefd, &fp, &vp); 4202 if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE)) 4203 error = EBADF; 4204 if (!error && NFSD_VNET(nfsrv_numnfsd) != 0) 4205 error = ENXIO; 4206 if (!error) { 4207 NFSD_VNET(nfsrv_stablefirst).nsf_fp = fp; 4208 nfsrv_setupstable(p); 4209 } 4210 } else if (uap->flag & NFSSVC_ADMINREVOKE) { 4211 error = copyin(uap->argp, (caddr_t)&adminrevoke, 4212 sizeof (struct nfsd_clid)); 4213 if (!error) 4214 error = nfsrv_adminrevoke(&adminrevoke, p); 4215 } else if (uap->flag & NFSSVC_DUMPCLIENTS) { 4216 error = copyin(uap->argp, (caddr_t)&dumplist, 4217 sizeof (struct nfsd_dumplist)); 4218 if (!error && (dumplist.ndl_size < 1 || 4219 dumplist.ndl_size > NFSRV_MAXDUMPLIST)) 4220 error = EPERM; 4221 if (!error) { 4222 len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size; 4223 dumpclients = malloc(len, M_TEMP, M_WAITOK | M_ZERO); 4224 nfsrv_dumpclients(dumpclients, dumplist.ndl_size); 4225 error = copyout(dumpclients, dumplist.ndl_list, len); 4226 free(dumpclients, M_TEMP); 4227 } 4228 } else if (uap->flag & NFSSVC_DUMPLOCKS) { 4229 error = copyin(uap->argp, (caddr_t)&dumplocklist, 4230 sizeof (struct nfsd_dumplocklist)); 4231 if (!error && (dumplocklist.ndllck_size < 1 || 4232 dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST)) 4233 error = EPERM; 4234 if (!error) 4235 error = nfsrv_lookupfilename(&nd, 4236 dumplocklist.ndllck_fname, p); 4237 if (!error) { 4238 len = sizeof (struct nfsd_dumplocks) * 4239 dumplocklist.ndllck_size; 4240 dumplocks = malloc(len, M_TEMP, M_WAITOK | M_ZERO); 4241 nfsrv_dumplocks(nd.ni_vp, dumplocks, 4242 dumplocklist.ndllck_size, p); 4243 vput(nd.ni_vp); 4244 error = copyout(dumplocks, dumplocklist.ndllck_list, 4245 len); 4246 free(dumplocks, M_TEMP); 4247 } 4248 } else if (uap->flag & NFSSVC_BACKUPSTABLE) { 4249 procp = p->td_proc; 4250 PROC_LOCK(procp); 4251 nfsd_master_pid = procp->p_pid; 4252 bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1); 4253 nfsd_master_start = procp->p_stats->p_start; 4254 NFSD_VNET(nfsd_master_proc) = procp; 4255 PROC_UNLOCK(procp); 4256 } else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) { 4257 NFSLOCKV4ROOTMUTEX(); 4258 if (!NFSD_VNET(nfsrv_suspend_nfsd)) { 4259 /* Lock out all nfsd threads */ 4260 do { 4261 igotlock = nfsv4_lock( 4262 &NFSD_VNET(nfsd_suspend_lock), 1, NULL, 4263 NFSV4ROOTLOCKMUTEXPTR, NULL); 4264 } while (igotlock == 0 && 4265 !NFSD_VNET(nfsrv_suspend_nfsd)); 4266 NFSD_VNET(nfsrv_suspend_nfsd) = true; 4267 } 4268 NFSUNLOCKV4ROOTMUTEX(); 4269 error = 0; 4270 } else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) { 4271 NFSLOCKV4ROOTMUTEX(); 4272 if (NFSD_VNET(nfsrv_suspend_nfsd)) { 4273 nfsv4_unlock(&NFSD_VNET(nfsd_suspend_lock), 0); 4274 NFSD_VNET(nfsrv_suspend_nfsd) = false; 4275 } 4276 NFSUNLOCKV4ROOTMUTEX(); 4277 error = 0; 4278 } 4279 4280 NFSEXITCODE(error); 4281 return (error); 4282 } 4283 4284 /* 4285 * Check exports. 4286 * Returns 0 if ok, 1 otherwise. 4287 */ 4288 int 4289 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp) 4290 { 4291 int i; 4292 4293 if ((NFSVNO_EXTLS(exp) && (nd->nd_flag & ND_TLS) == 0) || 4294 (NFSVNO_EXTLSCERT(exp) && 4295 (nd->nd_flag & ND_TLSCERT) == 0) || 4296 (NFSVNO_EXTLSCERTUSER(exp) && 4297 (nd->nd_flag & ND_TLSCERTUSER) == 0)) { 4298 if ((nd->nd_flag & ND_NFSV4) != 0) 4299 return (NFSERR_WRONGSEC); 4300 #ifdef notnow 4301 /* There is currently no auth_stat for this. */ 4302 else if ((nd->nd_flag & ND_TLS) == 0) 4303 return (NFSERR_AUTHERR | AUTH_NEEDS_TLS); 4304 else 4305 return (NFSERR_AUTHERR | AUTH_NEEDS_TLS_MUTUAL_HOST); 4306 #endif 4307 else 4308 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 4309 } 4310 4311 /* 4312 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to use 4313 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS. 4314 */ 4315 if ((nd->nd_flag & ND_NFSV3) != 0 && nd->nd_procnum == NFSPROC_FSINFO) 4316 return (0); 4317 4318 /* 4319 * This seems odd, but allow the case where the security flavor 4320 * list is empty. This happens when NFSv4 is traversing non-exported 4321 * file systems. Exported file systems should always have a non-empty 4322 * security flavor list. 4323 */ 4324 if (exp->nes_numsecflavor == 0) 4325 return (0); 4326 4327 for (i = 0; i < exp->nes_numsecflavor; i++) { 4328 /* 4329 * The tests for privacy and integrity must be first, 4330 * since ND_GSS is set for everything but AUTH_SYS. 4331 */ 4332 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P && 4333 (nd->nd_flag & ND_GSSPRIVACY)) 4334 return (0); 4335 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I && 4336 (nd->nd_flag & ND_GSSINTEGRITY)) 4337 return (0); 4338 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 && 4339 (nd->nd_flag & ND_GSS)) 4340 return (0); 4341 if (exp->nes_secflavors[i] == AUTH_SYS && 4342 (nd->nd_flag & ND_GSS) == 0) 4343 return (0); 4344 } 4345 if ((nd->nd_flag & ND_NFSV4) != 0) 4346 return (NFSERR_WRONGSEC); 4347 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 4348 } 4349 4350 /* 4351 * Calculate a hash value for the fid in a file handle. 4352 */ 4353 uint32_t 4354 nfsrv_hashfh(fhandle_t *fhp) 4355 { 4356 uint32_t hashval; 4357 4358 hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0); 4359 return (hashval); 4360 } 4361 4362 /* 4363 * Calculate a hash value for the sessionid. 4364 */ 4365 uint32_t 4366 nfsrv_hashsessionid(uint8_t *sessionid) 4367 { 4368 uint32_t hashval; 4369 4370 hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0); 4371 return (hashval); 4372 } 4373 4374 /* 4375 * Signal the userland master nfsd to backup the stable restart file. 4376 */ 4377 void 4378 nfsrv_backupstable(void) 4379 { 4380 struct proc *procp; 4381 4382 if (NFSD_VNET(nfsd_master_proc) != NULL) { 4383 procp = pfind(nfsd_master_pid); 4384 /* Try to make sure it is the correct process. */ 4385 if (procp == NFSD_VNET(nfsd_master_proc) && 4386 procp->p_stats->p_start.tv_sec == 4387 nfsd_master_start.tv_sec && 4388 procp->p_stats->p_start.tv_usec == 4389 nfsd_master_start.tv_usec && 4390 strcmp(procp->p_comm, nfsd_master_comm) == 0) 4391 kern_psignal(procp, SIGUSR2); 4392 else 4393 NFSD_VNET(nfsd_master_proc) = NULL; 4394 4395 if (procp != NULL) 4396 PROC_UNLOCK(procp); 4397 } 4398 } 4399 4400 /* 4401 * Create a DS data file for nfsrv_pnfscreate(). Called for each mirror. 4402 * The arguments are in a structure, so that they can be passed through 4403 * taskqueue for a kernel process to execute this function. 4404 */ 4405 struct nfsrvdscreate { 4406 int done; 4407 int inprog; 4408 struct task tsk; 4409 struct ucred *tcred; 4410 struct vnode *dvp; 4411 NFSPROC_T *p; 4412 struct pnfsdsfile *pf; 4413 int err; 4414 fhandle_t fh; 4415 struct vattr va; 4416 struct vattr createva; 4417 }; 4418 4419 int 4420 nfsrv_dscreate(struct vnode *dvp, struct vattr *vap, struct vattr *nvap, 4421 fhandle_t *fhp, struct pnfsdsfile *pf, struct pnfsdsattr *dsa, 4422 char *fnamep, struct ucred *tcred, NFSPROC_T *p, struct vnode **nvpp) 4423 { 4424 struct vnode *nvp; 4425 struct nameidata named; 4426 struct vattr va; 4427 char *bufp; 4428 u_long *hashp; 4429 struct nfsnode *np; 4430 struct nfsmount *nmp; 4431 int error; 4432 4433 NFSNAMEICNDSET(&named.ni_cnd, tcred, CREATE, 4434 LOCKPARENT | LOCKLEAF | NOCACHE); 4435 nfsvno_setpathbuf(&named, &bufp, &hashp); 4436 named.ni_cnd.cn_lkflags = LK_EXCLUSIVE; 4437 named.ni_cnd.cn_nameptr = bufp; 4438 if (fnamep != NULL) { 4439 strlcpy(bufp, fnamep, PNFS_FILENAME_LEN + 1); 4440 named.ni_cnd.cn_namelen = strlen(bufp); 4441 } else 4442 named.ni_cnd.cn_namelen = nfsrv_putfhname(fhp, bufp); 4443 NFSD_DEBUG(4, "nfsrv_dscreate: dvp=%p fname=%s\n", dvp, bufp); 4444 4445 /* Create the date file in the DS mount. */ 4446 error = NFSVOPLOCK(dvp, LK_EXCLUSIVE); 4447 if (error == 0) { 4448 error = VOP_CREATE(dvp, &nvp, &named.ni_cnd, vap); 4449 vref(dvp); 4450 VOP_VPUT_PAIR(dvp, error == 0 ? &nvp : NULL, false); 4451 if (error == 0) { 4452 /* Set the ownership of the file. */ 4453 error = VOP_SETATTR(nvp, nvap, tcred); 4454 NFSD_DEBUG(4, "nfsrv_dscreate:" 4455 " setattr-uid=%d\n", error); 4456 if (error != 0) 4457 vput(nvp); 4458 } 4459 if (error != 0) 4460 printf("pNFS: pnfscreate failed=%d\n", error); 4461 } else 4462 printf("pNFS: pnfscreate vnlock=%d\n", error); 4463 if (error == 0) { 4464 np = VTONFS(nvp); 4465 nmp = VFSTONFS(nvp->v_mount); 4466 if (strcmp(nvp->v_mount->mnt_vfc->vfc_name, "nfs") 4467 != 0 || nmp->nm_nam->sa_len > sizeof( 4468 struct sockaddr_in6) || 4469 np->n_fhp->nfh_len != NFSX_MYFH) { 4470 printf("Bad DS file: fstype=%s salen=%d" 4471 " fhlen=%d\n", 4472 nvp->v_mount->mnt_vfc->vfc_name, 4473 nmp->nm_nam->sa_len, np->n_fhp->nfh_len); 4474 error = ENOENT; 4475 } 4476 4477 /* Set extattrs for the DS on the MDS file. */ 4478 if (error == 0) { 4479 if (dsa != NULL) { 4480 error = VOP_GETATTR(nvp, &va, tcred); 4481 if (error == 0) { 4482 dsa->dsa_filerev = va.va_filerev; 4483 dsa->dsa_size = va.va_size; 4484 dsa->dsa_atime = va.va_atime; 4485 dsa->dsa_mtime = va.va_mtime; 4486 dsa->dsa_bytes = va.va_bytes; 4487 } 4488 } 4489 if (error == 0) { 4490 NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, 4491 NFSX_MYFH); 4492 NFSBCOPY(nmp->nm_nam, &pf->dsf_sin, 4493 nmp->nm_nam->sa_len); 4494 NFSBCOPY(named.ni_cnd.cn_nameptr, 4495 pf->dsf_filename, 4496 sizeof(pf->dsf_filename)); 4497 } 4498 } else 4499 printf("pNFS: pnfscreate can't get DS" 4500 " attr=%d\n", error); 4501 if (nvpp != NULL && error == 0) 4502 *nvpp = nvp; 4503 else 4504 vput(nvp); 4505 } 4506 nfsvno_relpathbuf(&named); 4507 return (error); 4508 } 4509 4510 /* 4511 * Start up the thread that will execute nfsrv_dscreate(). 4512 */ 4513 static void 4514 start_dscreate(void *arg, int pending) 4515 { 4516 struct nfsrvdscreate *dsc; 4517 4518 dsc = (struct nfsrvdscreate *)arg; 4519 dsc->err = nfsrv_dscreate(dsc->dvp, &dsc->createva, &dsc->va, &dsc->fh, 4520 dsc->pf, NULL, NULL, dsc->tcred, dsc->p, NULL); 4521 dsc->done = 1; 4522 NFSD_DEBUG(4, "start_dscreate: err=%d\n", dsc->err); 4523 } 4524 4525 /* 4526 * Create a pNFS data file on the Data Server(s). 4527 */ 4528 static void 4529 nfsrv_pnfscreate(struct vnode *vp, struct vattr *vap, struct ucred *cred, 4530 NFSPROC_T *p) 4531 { 4532 struct nfsrvdscreate *dsc, *tdsc = NULL; 4533 struct nfsdevice *ds, *tds, *fds; 4534 struct mount *mp; 4535 struct pnfsdsfile *pf, *tpf; 4536 struct pnfsdsattr dsattr; 4537 struct vattr va; 4538 struct vnode *dvp[NFSDEV_MAXMIRRORS]; 4539 struct nfsmount *nmp; 4540 fhandle_t fh; 4541 uid_t vauid; 4542 gid_t vagid; 4543 u_short vamode; 4544 struct ucred *tcred; 4545 int dsdir[NFSDEV_MAXMIRRORS], error, i, mirrorcnt, ret; 4546 int failpos, timo; 4547 4548 /* Get a DS server directory in a round-robin order. */ 4549 mirrorcnt = 1; 4550 mp = vp->v_mount; 4551 ds = fds = NULL; 4552 NFSDDSLOCK(); 4553 /* 4554 * Search for the first entry that handles this MDS fs, but use the 4555 * first entry for all MDS fs's otherwise. 4556 */ 4557 TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) { 4558 if (tds->nfsdev_nmp != NULL) { 4559 if (tds->nfsdev_mdsisset == 0 && ds == NULL) 4560 ds = tds; 4561 else if (tds->nfsdev_mdsisset != 0 && fsidcmp( 4562 &mp->mnt_stat.f_fsid, &tds->nfsdev_mdsfsid) == 0) { 4563 ds = fds = tds; 4564 break; 4565 } 4566 } 4567 } 4568 if (ds == NULL) { 4569 NFSDDSUNLOCK(); 4570 NFSD_DEBUG(4, "nfsrv_pnfscreate: no srv\n"); 4571 return; 4572 } 4573 i = dsdir[0] = ds->nfsdev_nextdir; 4574 ds->nfsdev_nextdir = (ds->nfsdev_nextdir + 1) % nfsrv_dsdirsize; 4575 dvp[0] = ds->nfsdev_dsdir[i]; 4576 tds = TAILQ_NEXT(ds, nfsdev_list); 4577 if (nfsrv_maxpnfsmirror > 1 && tds != NULL) { 4578 TAILQ_FOREACH_FROM(tds, &nfsrv_devidhead, nfsdev_list) { 4579 if (tds->nfsdev_nmp != NULL && 4580 ((tds->nfsdev_mdsisset == 0 && fds == NULL) || 4581 (tds->nfsdev_mdsisset != 0 && fds != NULL && 4582 fsidcmp(&mp->mnt_stat.f_fsid, 4583 &tds->nfsdev_mdsfsid) == 0))) { 4584 dsdir[mirrorcnt] = i; 4585 dvp[mirrorcnt] = tds->nfsdev_dsdir[i]; 4586 mirrorcnt++; 4587 if (mirrorcnt >= nfsrv_maxpnfsmirror) 4588 break; 4589 } 4590 } 4591 } 4592 /* Put at end of list to implement round-robin usage. */ 4593 TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list); 4594 TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list); 4595 NFSDDSUNLOCK(); 4596 dsc = NULL; 4597 if (mirrorcnt > 1) 4598 tdsc = dsc = malloc(sizeof(*dsc) * (mirrorcnt - 1), M_TEMP, 4599 M_WAITOK | M_ZERO); 4600 tpf = pf = malloc(sizeof(*pf) * nfsrv_maxpnfsmirror, M_TEMP, M_WAITOK | 4601 M_ZERO); 4602 4603 error = nfsvno_getfh(vp, &fh, p); 4604 if (error == 0) 4605 error = VOP_GETATTR(vp, &va, cred); 4606 if (error == 0) { 4607 /* Set the attributes for "vp" to Setattr the DS vp. */ 4608 vauid = va.va_uid; 4609 vagid = va.va_gid; 4610 vamode = va.va_mode; 4611 VATTR_NULL(&va); 4612 va.va_uid = vauid; 4613 va.va_gid = vagid; 4614 va.va_mode = vamode; 4615 va.va_size = 0; 4616 } else 4617 printf("pNFS: pnfscreate getfh+attr=%d\n", error); 4618 4619 NFSD_DEBUG(4, "nfsrv_pnfscreate: cruid=%d crgid=%d\n", cred->cr_uid, 4620 cred->cr_gid); 4621 /* Make data file name based on FH. */ 4622 tcred = newnfs_getcred(); 4623 4624 /* 4625 * Create the file on each DS mirror, using kernel process(es) for the 4626 * additional mirrors. 4627 */ 4628 failpos = -1; 4629 for (i = 0; i < mirrorcnt - 1 && error == 0; i++, tpf++, tdsc++) { 4630 tpf->dsf_dir = dsdir[i]; 4631 tdsc->tcred = tcred; 4632 tdsc->p = p; 4633 tdsc->pf = tpf; 4634 tdsc->createva = *vap; 4635 NFSBCOPY(&fh, &tdsc->fh, sizeof(fh)); 4636 tdsc->va = va; 4637 tdsc->dvp = dvp[i]; 4638 tdsc->done = 0; 4639 tdsc->inprog = 0; 4640 tdsc->err = 0; 4641 ret = EIO; 4642 if (nfs_pnfsiothreads != 0) { 4643 ret = nfs_pnfsio(start_dscreate, tdsc); 4644 NFSD_DEBUG(4, "nfsrv_pnfscreate: nfs_pnfsio=%d\n", ret); 4645 } 4646 if (ret != 0) { 4647 ret = nfsrv_dscreate(dvp[i], vap, &va, &fh, tpf, NULL, 4648 NULL, tcred, p, NULL); 4649 if (ret != 0) { 4650 KASSERT(error == 0, ("nfsrv_dscreate err=%d", 4651 error)); 4652 if (failpos == -1 && nfsds_failerr(ret)) 4653 failpos = i; 4654 else 4655 error = ret; 4656 } 4657 } 4658 } 4659 if (error == 0) { 4660 tpf->dsf_dir = dsdir[mirrorcnt - 1]; 4661 error = nfsrv_dscreate(dvp[mirrorcnt - 1], vap, &va, &fh, tpf, 4662 &dsattr, NULL, tcred, p, NULL); 4663 if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(error)) { 4664 failpos = mirrorcnt - 1; 4665 error = 0; 4666 } 4667 } 4668 timo = hz / 50; /* Wait for 20msec. */ 4669 if (timo < 1) 4670 timo = 1; 4671 /* Wait for kernel task(s) to complete. */ 4672 for (tdsc = dsc, i = 0; i < mirrorcnt - 1; i++, tdsc++) { 4673 while (tdsc->inprog != 0 && tdsc->done == 0) 4674 tsleep(&tdsc->tsk, PVFS, "srvdcr", timo); 4675 if (tdsc->err != 0) { 4676 if (failpos == -1 && nfsds_failerr(tdsc->err)) 4677 failpos = i; 4678 else if (error == 0) 4679 error = tdsc->err; 4680 } 4681 } 4682 4683 /* 4684 * If failpos has been set, that mirror has failed, so it needs 4685 * to be disabled. 4686 */ 4687 if (failpos >= 0) { 4688 nmp = VFSTONFS(dvp[failpos]->v_mount); 4689 NFSLOCKMNT(nmp); 4690 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | 4691 NFSMNTP_CANCELRPCS)) == 0) { 4692 nmp->nm_privflag |= NFSMNTP_CANCELRPCS; 4693 NFSUNLOCKMNT(nmp); 4694 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p); 4695 NFSD_DEBUG(4, "dscreatfail fail=%d ds=%p\n", failpos, 4696 ds); 4697 if (ds != NULL) 4698 nfsrv_killrpcs(nmp); 4699 NFSLOCKMNT(nmp); 4700 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 4701 wakeup(nmp); 4702 } 4703 NFSUNLOCKMNT(nmp); 4704 } 4705 4706 NFSFREECRED(tcred); 4707 if (error == 0) { 4708 ASSERT_VOP_ELOCKED(vp, "nfsrv_pnfscreate vp"); 4709 4710 NFSD_DEBUG(4, "nfsrv_pnfscreate: mirrorcnt=%d maxmirror=%d\n", 4711 mirrorcnt, nfsrv_maxpnfsmirror); 4712 /* 4713 * For all mirrors that couldn't be created, fill in the 4714 * *pf structure, but with an IP address == 0.0.0.0. 4715 */ 4716 tpf = pf + mirrorcnt; 4717 for (i = mirrorcnt; i < nfsrv_maxpnfsmirror; i++, tpf++) { 4718 *tpf = *pf; 4719 tpf->dsf_sin.sin_family = AF_INET; 4720 tpf->dsf_sin.sin_len = sizeof(struct sockaddr_in); 4721 tpf->dsf_sin.sin_addr.s_addr = 0; 4722 tpf->dsf_sin.sin_port = 0; 4723 } 4724 4725 error = vn_extattr_set(vp, IO_NODELOCKED, 4726 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", 4727 sizeof(*pf) * nfsrv_maxpnfsmirror, (char *)pf, p); 4728 if (error == 0) 4729 error = vn_extattr_set(vp, IO_NODELOCKED, 4730 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", 4731 sizeof(dsattr), (char *)&dsattr, p); 4732 if (error != 0) 4733 printf("pNFS: pnfscreate setextattr=%d\n", 4734 error); 4735 } else 4736 printf("pNFS: pnfscreate=%d\n", error); 4737 free(pf, M_TEMP); 4738 free(dsc, M_TEMP); 4739 } 4740 4741 /* 4742 * Get the information needed to remove the pNFS Data Server file from the 4743 * Metadata file. Upon success, ddvp is set non-NULL to the locked 4744 * DS directory vnode. The caller must unlock *ddvp when done with it. 4745 */ 4746 static void 4747 nfsrv_pnfsremovesetup(struct vnode *vp, NFSPROC_T *p, struct vnode **dvpp, 4748 int *mirrorcntp, char *fname, fhandle_t *fhp) 4749 { 4750 struct vattr va; 4751 struct ucred *tcred; 4752 char *buf; 4753 int buflen, error; 4754 4755 dvpp[0] = NULL; 4756 /* If not an exported regular file or not a pNFS server, just return. */ 4757 if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 || 4758 nfsrv_devidcnt == 0) 4759 return; 4760 4761 /* Check to see if this is the last hard link. */ 4762 tcred = newnfs_getcred(); 4763 error = VOP_GETATTR(vp, &va, tcred); 4764 NFSFREECRED(tcred); 4765 if (error != 0) { 4766 printf("pNFS: nfsrv_pnfsremovesetup getattr=%d\n", error); 4767 return; 4768 } 4769 if (va.va_nlink > 1) 4770 return; 4771 4772 error = nfsvno_getfh(vp, fhp, p); 4773 if (error != 0) { 4774 printf("pNFS: nfsrv_pnfsremovesetup getfh=%d\n", error); 4775 return; 4776 } 4777 4778 buflen = 1024; 4779 buf = malloc(buflen, M_TEMP, M_WAITOK); 4780 /* Get the directory vnode for the DS mount and the file handle. */ 4781 error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, dvpp, 4782 NULL, NULL, fname, NULL, NULL, NULL, NULL, NULL); 4783 free(buf, M_TEMP); 4784 if (error != 0) 4785 printf("pNFS: nfsrv_pnfsremovesetup getsockmnt=%d\n", error); 4786 } 4787 4788 /* 4789 * Remove a DS data file for nfsrv_pnfsremove(). Called for each mirror. 4790 * The arguments are in a structure, so that they can be passed through 4791 * taskqueue for a kernel process to execute this function. 4792 */ 4793 struct nfsrvdsremove { 4794 int done; 4795 int inprog; 4796 struct task tsk; 4797 struct ucred *tcred; 4798 struct vnode *dvp; 4799 NFSPROC_T *p; 4800 int err; 4801 char fname[PNFS_FILENAME_LEN + 1]; 4802 }; 4803 4804 static int 4805 nfsrv_dsremove(struct vnode *dvp, char *fname, struct ucred *tcred, 4806 NFSPROC_T *p) 4807 { 4808 struct nameidata named; 4809 struct vnode *nvp; 4810 char *bufp; 4811 u_long *hashp; 4812 int error; 4813 4814 error = NFSVOPLOCK(dvp, LK_EXCLUSIVE); 4815 if (error != 0) 4816 return (error); 4817 named.ni_cnd.cn_nameiop = DELETE; 4818 named.ni_cnd.cn_lkflags = LK_EXCLUSIVE | LK_RETRY; 4819 named.ni_cnd.cn_cred = tcred; 4820 named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF; 4821 nfsvno_setpathbuf(&named, &bufp, &hashp); 4822 named.ni_cnd.cn_nameptr = bufp; 4823 named.ni_cnd.cn_namelen = strlen(fname); 4824 strlcpy(bufp, fname, NAME_MAX); 4825 NFSD_DEBUG(4, "nfsrv_pnfsremove: filename=%s\n", bufp); 4826 error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd); 4827 NFSD_DEBUG(4, "nfsrv_pnfsremove: aft LOOKUP=%d\n", error); 4828 if (error == 0) { 4829 error = VOP_REMOVE(dvp, nvp, &named.ni_cnd); 4830 vput(nvp); 4831 } 4832 NFSVOPUNLOCK(dvp); 4833 nfsvno_relpathbuf(&named); 4834 if (error != 0) 4835 printf("pNFS: nfsrv_pnfsremove failed=%d\n", error); 4836 return (error); 4837 } 4838 4839 /* 4840 * Start up the thread that will execute nfsrv_dsremove(). 4841 */ 4842 static void 4843 start_dsremove(void *arg, int pending) 4844 { 4845 struct nfsrvdsremove *dsrm; 4846 4847 dsrm = (struct nfsrvdsremove *)arg; 4848 dsrm->err = nfsrv_dsremove(dsrm->dvp, dsrm->fname, dsrm->tcred, 4849 dsrm->p); 4850 dsrm->done = 1; 4851 NFSD_DEBUG(4, "start_dsremove: err=%d\n", dsrm->err); 4852 } 4853 4854 /* 4855 * Remove a pNFS data file from a Data Server. 4856 * nfsrv_pnfsremovesetup() must have been called before the MDS file was 4857 * removed to set up the dvp and fill in the FH. 4858 */ 4859 static void 4860 nfsrv_pnfsremove(struct vnode **dvp, int mirrorcnt, char *fname, fhandle_t *fhp, 4861 NFSPROC_T *p) 4862 { 4863 struct ucred *tcred; 4864 struct nfsrvdsremove *dsrm, *tdsrm; 4865 struct nfsdevice *ds; 4866 struct nfsmount *nmp; 4867 int failpos, i, ret, timo; 4868 4869 tcred = newnfs_getcred(); 4870 dsrm = NULL; 4871 if (mirrorcnt > 1) 4872 dsrm = malloc(sizeof(*dsrm) * mirrorcnt - 1, M_TEMP, M_WAITOK); 4873 /* 4874 * Remove the file on each DS mirror, using kernel process(es) for the 4875 * additional mirrors. 4876 */ 4877 failpos = -1; 4878 for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) { 4879 tdsrm->tcred = tcred; 4880 tdsrm->p = p; 4881 tdsrm->dvp = dvp[i]; 4882 strlcpy(tdsrm->fname, fname, PNFS_FILENAME_LEN + 1); 4883 tdsrm->inprog = 0; 4884 tdsrm->done = 0; 4885 tdsrm->err = 0; 4886 ret = EIO; 4887 if (nfs_pnfsiothreads != 0) { 4888 ret = nfs_pnfsio(start_dsremove, tdsrm); 4889 NFSD_DEBUG(4, "nfsrv_pnfsremove: nfs_pnfsio=%d\n", ret); 4890 } 4891 if (ret != 0) { 4892 ret = nfsrv_dsremove(dvp[i], fname, tcred, p); 4893 if (failpos == -1 && nfsds_failerr(ret)) 4894 failpos = i; 4895 } 4896 } 4897 ret = nfsrv_dsremove(dvp[mirrorcnt - 1], fname, tcred, p); 4898 if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(ret)) 4899 failpos = mirrorcnt - 1; 4900 timo = hz / 50; /* Wait for 20msec. */ 4901 if (timo < 1) 4902 timo = 1; 4903 /* Wait for kernel task(s) to complete. */ 4904 for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) { 4905 while (tdsrm->inprog != 0 && tdsrm->done == 0) 4906 tsleep(&tdsrm->tsk, PVFS, "srvdsrm", timo); 4907 if (failpos == -1 && nfsds_failerr(tdsrm->err)) 4908 failpos = i; 4909 } 4910 4911 /* 4912 * If failpos has been set, that mirror has failed, so it needs 4913 * to be disabled. 4914 */ 4915 if (failpos >= 0) { 4916 nmp = VFSTONFS(dvp[failpos]->v_mount); 4917 NFSLOCKMNT(nmp); 4918 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | 4919 NFSMNTP_CANCELRPCS)) == 0) { 4920 nmp->nm_privflag |= NFSMNTP_CANCELRPCS; 4921 NFSUNLOCKMNT(nmp); 4922 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p); 4923 NFSD_DEBUG(4, "dsremovefail fail=%d ds=%p\n", failpos, 4924 ds); 4925 if (ds != NULL) 4926 nfsrv_killrpcs(nmp); 4927 NFSLOCKMNT(nmp); 4928 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 4929 wakeup(nmp); 4930 } 4931 NFSUNLOCKMNT(nmp); 4932 } 4933 4934 /* Get rid all layouts for the file. */ 4935 nfsrv_freefilelayouts(fhp); 4936 4937 NFSFREECRED(tcred); 4938 free(dsrm, M_TEMP); 4939 } 4940 4941 /* 4942 * Generate a file name based on the file handle and put it in *bufp. 4943 * Return the number of bytes generated. 4944 */ 4945 static int 4946 nfsrv_putfhname(fhandle_t *fhp, char *bufp) 4947 { 4948 int i; 4949 uint8_t *cp; 4950 const uint8_t *hexdigits = "0123456789abcdef"; 4951 4952 cp = (uint8_t *)fhp; 4953 for (i = 0; i < sizeof(*fhp); i++) { 4954 bufp[2 * i] = hexdigits[(*cp >> 4) & 0xf]; 4955 bufp[2 * i + 1] = hexdigits[*cp++ & 0xf]; 4956 } 4957 bufp[2 * i] = '\0'; 4958 return (2 * i); 4959 } 4960 4961 /* 4962 * Update the Metadata file's attributes from the DS file when a Read/Write 4963 * layout is returned. 4964 * Basically just call nfsrv_proxyds() with procedure == NFSPROC_LAYOUTRETURN 4965 * so that it does a nfsrv_getattrdsrpc() and nfsrv_setextattr() on the DS file. 4966 */ 4967 int 4968 nfsrv_updatemdsattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p) 4969 { 4970 struct ucred *tcred; 4971 int error; 4972 4973 /* Do this as root so that it won't fail with EACCES. */ 4974 tcred = newnfs_getcred(); 4975 error = nfsrv_proxyds(vp, 0, 0, tcred, p, NFSPROC_LAYOUTRETURN, 4976 NULL, NULL, NULL, nap, NULL, NULL, 0, NULL); 4977 NFSFREECRED(tcred); 4978 return (error); 4979 } 4980 4981 /* 4982 * Set the NFSv4 ACL on the DS file to the same ACL as the MDS file. 4983 */ 4984 static int 4985 nfsrv_dssetacl(struct vnode *vp, struct acl *aclp, struct ucred *cred, 4986 NFSPROC_T *p) 4987 { 4988 int error; 4989 4990 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETACL, 4991 NULL, NULL, NULL, NULL, aclp, NULL, 0, NULL); 4992 return (error); 4993 } 4994 4995 static int 4996 nfsrv_proxyds(struct vnode *vp, off_t off, int cnt, struct ucred *cred, 4997 struct thread *p, int ioproc, struct mbuf **mpp, char *cp, 4998 struct mbuf **mpp2, struct nfsvattr *nap, struct acl *aclp, 4999 off_t *offp, int content, bool *eofp) 5000 { 5001 struct nfsmount *nmp[NFSDEV_MAXMIRRORS], *failnmp; 5002 fhandle_t fh[NFSDEV_MAXMIRRORS]; 5003 struct vnode *dvp[NFSDEV_MAXMIRRORS]; 5004 struct nfsdevice *ds; 5005 struct pnfsdsattr dsattr; 5006 struct opnfsdsattr odsattr; 5007 char *buf; 5008 int buflen, error, failpos, i, mirrorcnt, origmircnt, trycnt; 5009 5010 NFSD_DEBUG(4, "in nfsrv_proxyds\n"); 5011 /* 5012 * If not a regular file, not exported or not a pNFS server, 5013 * just return ENOENT. 5014 */ 5015 if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 || 5016 nfsrv_devidcnt == 0) 5017 return (ENOENT); 5018 5019 buflen = 1024; 5020 buf = malloc(buflen, M_TEMP, M_WAITOK); 5021 error = 0; 5022 5023 /* 5024 * For Getattr, get the Change attribute (va_filerev) and size (va_size) 5025 * from the MetaData file's extended attribute. 5026 */ 5027 if (ioproc == NFSPROC_GETATTR) { 5028 error = vn_extattr_get(vp, IO_NODELOCKED, 5029 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", &buflen, buf, 5030 p); 5031 if (error == 0) { 5032 if (buflen == sizeof(odsattr)) { 5033 NFSBCOPY(buf, &odsattr, buflen); 5034 nap->na_filerev = odsattr.dsa_filerev; 5035 nap->na_size = odsattr.dsa_size; 5036 nap->na_atime = odsattr.dsa_atime; 5037 nap->na_mtime = odsattr.dsa_mtime; 5038 /* 5039 * Fake na_bytes by rounding up na_size. 5040 * Since we don't know the block size, just 5041 * use BLKDEV_IOSIZE. 5042 */ 5043 nap->na_bytes = (odsattr.dsa_size + 5044 BLKDEV_IOSIZE - 1) & ~(BLKDEV_IOSIZE - 1); 5045 } else if (buflen == sizeof(dsattr)) { 5046 NFSBCOPY(buf, &dsattr, buflen); 5047 nap->na_filerev = dsattr.dsa_filerev; 5048 nap->na_size = dsattr.dsa_size; 5049 nap->na_atime = dsattr.dsa_atime; 5050 nap->na_mtime = dsattr.dsa_mtime; 5051 nap->na_bytes = dsattr.dsa_bytes; 5052 } else 5053 error = ENXIO; 5054 } 5055 if (error == 0) { 5056 /* 5057 * If nfsrv_pnfsgetdsattr is 0 or nfsrv_checkdsattr() 5058 * returns 0, just return now. nfsrv_checkdsattr() 5059 * returns 0 if there is no Read/Write layout 5060 * plus either an Open/Write_access or Write 5061 * delegation issued to a client for the file. 5062 */ 5063 if (nfsrv_pnfsgetdsattr == 0 || 5064 nfsrv_checkdsattr(vp, p) == 0) { 5065 free(buf, M_TEMP); 5066 return (error); 5067 } 5068 } 5069 5070 /* 5071 * Clear ENOATTR so the code below will attempt to do a 5072 * nfsrv_getattrdsrpc() to get the attributes and (re)create 5073 * the extended attribute. 5074 */ 5075 if (error == ENOATTR) 5076 error = 0; 5077 } 5078 5079 origmircnt = -1; 5080 trycnt = 0; 5081 tryagain: 5082 if (error == 0) { 5083 buflen = 1024; 5084 if (ioproc == NFSPROC_READDS && NFSVOPISLOCKED(vp) == 5085 LK_EXCLUSIVE) 5086 printf("nfsrv_proxyds: Readds vp exclusively locked\n"); 5087 error = nfsrv_dsgetsockmnt(vp, LK_SHARED, buf, &buflen, 5088 &mirrorcnt, p, dvp, fh, NULL, NULL, NULL, NULL, NULL, 5089 NULL, NULL); 5090 if (error == 0) { 5091 for (i = 0; i < mirrorcnt; i++) 5092 nmp[i] = VFSTONFS(dvp[i]->v_mount); 5093 } else 5094 printf("pNFS: proxy getextattr sockaddr=%d\n", error); 5095 } else 5096 printf("pNFS: nfsrv_dsgetsockmnt=%d\n", error); 5097 if (error == 0) { 5098 failpos = -1; 5099 if (origmircnt == -1) 5100 origmircnt = mirrorcnt; 5101 /* 5102 * If failpos is set to a mirror#, then that mirror has 5103 * failed and will be disabled. For Read, Getattr and Seek, the 5104 * function only tries one mirror, so if that mirror has 5105 * failed, it will need to be retried. As such, increment 5106 * tryitagain for these cases. 5107 * For Write, Setattr and Setacl, the function tries all 5108 * mirrors and will not return an error for the case where 5109 * one mirror has failed. For these cases, the functioning 5110 * mirror(s) will have been modified, so a retry isn't 5111 * necessary. These functions will set failpos for the 5112 * failed mirror#. 5113 */ 5114 if (ioproc == NFSPROC_READDS) { 5115 error = nfsrv_readdsrpc(fh, off, cnt, cred, p, nmp[0], 5116 mpp, mpp2); 5117 if (nfsds_failerr(error) && mirrorcnt > 1) { 5118 /* 5119 * Setting failpos will cause the mirror 5120 * to be disabled and then a retry of this 5121 * read is required. 5122 */ 5123 failpos = 0; 5124 error = 0; 5125 trycnt++; 5126 } 5127 } else if (ioproc == NFSPROC_WRITEDS) 5128 error = nfsrv_writedsrpc(fh, off, cnt, cred, p, vp, 5129 &nmp[0], mirrorcnt, mpp, cp, &failpos); 5130 else if (ioproc == NFSPROC_SETATTR) 5131 error = nfsrv_setattrdsrpc(fh, cred, p, vp, &nmp[0], 5132 mirrorcnt, nap, &failpos); 5133 else if (ioproc == NFSPROC_SETACL) 5134 error = nfsrv_setacldsrpc(fh, cred, p, vp, &nmp[0], 5135 mirrorcnt, aclp, &failpos); 5136 else if (ioproc == NFSPROC_SEEKDS) { 5137 error = nfsrv_seekdsrpc(fh, offp, content, eofp, cred, 5138 p, nmp[0]); 5139 if (nfsds_failerr(error) && mirrorcnt > 1) { 5140 /* 5141 * Setting failpos will cause the mirror 5142 * to be disabled and then a retry of this 5143 * read is required. 5144 */ 5145 failpos = 0; 5146 error = 0; 5147 trycnt++; 5148 } 5149 } else if (ioproc == NFSPROC_ALLOCATE) 5150 error = nfsrv_allocatedsrpc(fh, off, *offp, cred, p, vp, 5151 &nmp[0], mirrorcnt, &failpos); 5152 else if (ioproc == NFSPROC_DEALLOCATE) 5153 error = nfsrv_deallocatedsrpc(fh, off, *offp, cred, p, 5154 vp, &nmp[0], mirrorcnt, &failpos); 5155 else { 5156 error = nfsrv_getattrdsrpc(&fh[mirrorcnt - 1], cred, p, 5157 vp, nmp[mirrorcnt - 1], nap); 5158 if (nfsds_failerr(error) && mirrorcnt > 1) { 5159 /* 5160 * Setting failpos will cause the mirror 5161 * to be disabled and then a retry of this 5162 * getattr is required. 5163 */ 5164 failpos = mirrorcnt - 1; 5165 error = 0; 5166 trycnt++; 5167 } 5168 } 5169 ds = NULL; 5170 if (failpos >= 0) { 5171 failnmp = nmp[failpos]; 5172 NFSLOCKMNT(failnmp); 5173 if ((failnmp->nm_privflag & (NFSMNTP_FORCEDISM | 5174 NFSMNTP_CANCELRPCS)) == 0) { 5175 failnmp->nm_privflag |= NFSMNTP_CANCELRPCS; 5176 NFSUNLOCKMNT(failnmp); 5177 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, 5178 failnmp, p); 5179 NFSD_DEBUG(4, "dsldsnmp fail=%d ds=%p\n", 5180 failpos, ds); 5181 if (ds != NULL) 5182 nfsrv_killrpcs(failnmp); 5183 NFSLOCKMNT(failnmp); 5184 failnmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 5185 wakeup(failnmp); 5186 } 5187 NFSUNLOCKMNT(failnmp); 5188 } 5189 for (i = 0; i < mirrorcnt; i++) 5190 NFSVOPUNLOCK(dvp[i]); 5191 NFSD_DEBUG(4, "nfsrv_proxyds: aft RPC=%d trya=%d\n", error, 5192 trycnt); 5193 /* Try the Read/Getattr again if a mirror was deleted. */ 5194 if (ds != NULL && trycnt > 0 && trycnt < origmircnt) 5195 goto tryagain; 5196 } else { 5197 /* Return ENOENT for any Extended Attribute error. */ 5198 error = ENOENT; 5199 } 5200 free(buf, M_TEMP); 5201 NFSD_DEBUG(4, "nfsrv_proxyds: error=%d\n", error); 5202 return (error); 5203 } 5204 5205 /* 5206 * Get the DS mount point, fh and directory from the "pnfsd.dsfile" extended 5207 * attribute. 5208 * newnmpp - If it points to a non-NULL nmp, that is the destination and needs 5209 * to be checked. If it points to a NULL nmp, then it returns 5210 * a suitable destination. 5211 * curnmp - If non-NULL, it is the source mount for the copy. 5212 */ 5213 int 5214 nfsrv_dsgetsockmnt(struct vnode *vp, int lktype, char *buf, int *buflenp, 5215 int *mirrorcntp, NFSPROC_T *p, struct vnode **dvpp, fhandle_t *fhp, 5216 char *devid, char *fnamep, struct vnode **nvpp, struct nfsmount **newnmpp, 5217 struct nfsmount *curnmp, int *ippos, int *dsdirp) 5218 { 5219 struct vnode *dvp, *nvp = NULL, **tdvpp; 5220 struct mount *mp; 5221 struct nfsmount *nmp, *newnmp; 5222 struct sockaddr *sad; 5223 struct sockaddr_in *sin; 5224 struct nfsdevice *ds, *tds, *fndds; 5225 struct pnfsdsfile *pf; 5226 uint32_t dsdir; 5227 int error, fhiszero, fnd, gotone, i, mirrorcnt; 5228 5229 ASSERT_VOP_LOCKED(vp, "nfsrv_dsgetsockmnt vp"); 5230 *mirrorcntp = 1; 5231 tdvpp = dvpp; 5232 if (nvpp != NULL) 5233 *nvpp = NULL; 5234 if (dvpp != NULL) 5235 *dvpp = NULL; 5236 if (ippos != NULL) 5237 *ippos = -1; 5238 if (newnmpp != NULL) 5239 newnmp = *newnmpp; 5240 else 5241 newnmp = NULL; 5242 mp = vp->v_mount; 5243 error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, 5244 "pnfsd.dsfile", buflenp, buf, p); 5245 mirrorcnt = *buflenp / sizeof(*pf); 5246 if (error == 0 && (mirrorcnt < 1 || mirrorcnt > NFSDEV_MAXMIRRORS || 5247 *buflenp != sizeof(*pf) * mirrorcnt)) 5248 error = ENOATTR; 5249 5250 pf = (struct pnfsdsfile *)buf; 5251 /* If curnmp != NULL, check for a match in the mirror list. */ 5252 if (curnmp != NULL && error == 0) { 5253 fnd = 0; 5254 for (i = 0; i < mirrorcnt; i++, pf++) { 5255 sad = (struct sockaddr *)&pf->dsf_sin; 5256 if (nfsaddr2_match(sad, curnmp->nm_nam)) { 5257 if (ippos != NULL) 5258 *ippos = i; 5259 fnd = 1; 5260 break; 5261 } 5262 } 5263 if (fnd == 0) 5264 error = ENXIO; 5265 } 5266 5267 gotone = 0; 5268 pf = (struct pnfsdsfile *)buf; 5269 NFSD_DEBUG(4, "nfsrv_dsgetsockmnt: mirrorcnt=%d err=%d\n", mirrorcnt, 5270 error); 5271 for (i = 0; i < mirrorcnt && error == 0; i++, pf++) { 5272 fhiszero = 0; 5273 sad = (struct sockaddr *)&pf->dsf_sin; 5274 sin = &pf->dsf_sin; 5275 dsdir = pf->dsf_dir; 5276 if (dsdir >= nfsrv_dsdirsize) { 5277 printf("nfsrv_dsgetsockmnt: dsdir=%d\n", dsdir); 5278 error = ENOATTR; 5279 } else if (nvpp != NULL && newnmp != NULL && 5280 nfsaddr2_match(sad, newnmp->nm_nam)) 5281 error = EEXIST; 5282 if (error == 0) { 5283 if (ippos != NULL && curnmp == NULL && 5284 sad->sa_family == AF_INET && 5285 sin->sin_addr.s_addr == 0) 5286 *ippos = i; 5287 if (NFSBCMP(&zerofh, &pf->dsf_fh, sizeof(zerofh)) == 0) 5288 fhiszero = 1; 5289 /* Use the socket address to find the mount point. */ 5290 fndds = NULL; 5291 NFSDDSLOCK(); 5292 /* Find a match for the IP address. */ 5293 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 5294 if (ds->nfsdev_nmp != NULL) { 5295 dvp = ds->nfsdev_dvp; 5296 nmp = VFSTONFS(dvp->v_mount); 5297 if (nmp != ds->nfsdev_nmp) 5298 printf("different2 nmp %p %p\n", 5299 nmp, ds->nfsdev_nmp); 5300 if (nfsaddr2_match(sad, nmp->nm_nam)) { 5301 fndds = ds; 5302 break; 5303 } 5304 } 5305 } 5306 if (fndds != NULL && newnmpp != NULL && 5307 newnmp == NULL) { 5308 /* Search for a place to make a mirror copy. */ 5309 TAILQ_FOREACH(tds, &nfsrv_devidhead, 5310 nfsdev_list) { 5311 if (tds->nfsdev_nmp != NULL && 5312 fndds != tds && 5313 ((tds->nfsdev_mdsisset == 0 && 5314 fndds->nfsdev_mdsisset == 0) || 5315 (tds->nfsdev_mdsisset != 0 && 5316 fndds->nfsdev_mdsisset != 0 && 5317 fsidcmp(&tds->nfsdev_mdsfsid, 5318 &mp->mnt_stat.f_fsid) == 0))) { 5319 *newnmpp = tds->nfsdev_nmp; 5320 break; 5321 } 5322 } 5323 if (tds != NULL) { 5324 /* 5325 * Move this entry to the end of the 5326 * list, so it won't be selected as 5327 * easily the next time. 5328 */ 5329 TAILQ_REMOVE(&nfsrv_devidhead, tds, 5330 nfsdev_list); 5331 TAILQ_INSERT_TAIL(&nfsrv_devidhead, tds, 5332 nfsdev_list); 5333 } 5334 } 5335 NFSDDSUNLOCK(); 5336 if (fndds != NULL) { 5337 dvp = fndds->nfsdev_dsdir[dsdir]; 5338 if (lktype != 0 || fhiszero != 0 || 5339 (nvpp != NULL && *nvpp == NULL)) { 5340 if (fhiszero != 0) 5341 error = vn_lock(dvp, 5342 LK_EXCLUSIVE); 5343 else if (lktype != 0) 5344 error = vn_lock(dvp, lktype); 5345 else 5346 error = vn_lock(dvp, LK_SHARED); 5347 /* 5348 * If the file handle is all 0's, try to 5349 * do a Lookup against the DS to acquire 5350 * it. 5351 * If dvpp == NULL or the Lookup fails, 5352 * unlock dvp after the call. 5353 */ 5354 if (error == 0 && (fhiszero != 0 || 5355 (nvpp != NULL && *nvpp == NULL))) { 5356 error = nfsrv_pnfslookupds(vp, 5357 dvp, pf, &nvp, p); 5358 if (error == 0) { 5359 if (fhiszero != 0) 5360 nfsrv_pnfssetfh( 5361 vp, pf, 5362 devid, 5363 fnamep, 5364 nvp, p); 5365 if (nvpp != NULL && 5366 *nvpp == NULL) { 5367 *nvpp = nvp; 5368 *dsdirp = dsdir; 5369 } else 5370 vput(nvp); 5371 } 5372 if (error != 0 || lktype == 0) 5373 NFSVOPUNLOCK(dvp); 5374 } 5375 } 5376 if (error == 0) { 5377 gotone++; 5378 NFSD_DEBUG(4, "gotone=%d\n", gotone); 5379 if (devid != NULL) { 5380 NFSBCOPY(fndds->nfsdev_deviceid, 5381 devid, NFSX_V4DEVICEID); 5382 devid += NFSX_V4DEVICEID; 5383 } 5384 if (dvpp != NULL) 5385 *tdvpp++ = dvp; 5386 if (fhp != NULL) 5387 NFSBCOPY(&pf->dsf_fh, fhp++, 5388 NFSX_MYFH); 5389 if (fnamep != NULL && gotone == 1) 5390 strlcpy(fnamep, 5391 pf->dsf_filename, 5392 sizeof(pf->dsf_filename)); 5393 } else 5394 NFSD_DEBUG(4, "nfsrv_dsgetsockmnt " 5395 "err=%d\n", error); 5396 } 5397 } 5398 } 5399 if (error == 0 && gotone == 0) 5400 error = ENOENT; 5401 5402 NFSD_DEBUG(4, "eo nfsrv_dsgetsockmnt: gotone=%d err=%d\n", gotone, 5403 error); 5404 if (error == 0) 5405 *mirrorcntp = gotone; 5406 else { 5407 if (gotone > 0 && dvpp != NULL) { 5408 /* 5409 * If the error didn't occur on the first one and 5410 * dvpp != NULL, the one(s) prior to the failure will 5411 * have locked dvp's that need to be unlocked. 5412 */ 5413 for (i = 0; i < gotone; i++) { 5414 NFSVOPUNLOCK(*dvpp); 5415 *dvpp++ = NULL; 5416 } 5417 } 5418 /* 5419 * If it found the vnode to be copied from before a failure, 5420 * it needs to be vput()'d. 5421 */ 5422 if (nvpp != NULL && *nvpp != NULL) { 5423 vput(*nvpp); 5424 *nvpp = NULL; 5425 } 5426 } 5427 return (error); 5428 } 5429 5430 /* 5431 * Set the extended attribute for the Change attribute. 5432 */ 5433 static int 5434 nfsrv_setextattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p) 5435 { 5436 struct pnfsdsattr dsattr; 5437 int error; 5438 5439 ASSERT_VOP_ELOCKED(vp, "nfsrv_setextattr vp"); 5440 dsattr.dsa_filerev = nap->na_filerev; 5441 dsattr.dsa_size = nap->na_size; 5442 dsattr.dsa_atime = nap->na_atime; 5443 dsattr.dsa_mtime = nap->na_mtime; 5444 dsattr.dsa_bytes = nap->na_bytes; 5445 error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, 5446 "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p); 5447 if (error != 0) 5448 printf("pNFS: setextattr=%d\n", error); 5449 return (error); 5450 } 5451 5452 static int 5453 nfsrv_readdsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred, 5454 NFSPROC_T *p, struct nfsmount *nmp, struct mbuf **mpp, struct mbuf **mpendp) 5455 { 5456 uint32_t *tl; 5457 struct nfsrv_descript *nd; 5458 nfsv4stateid_t st; 5459 struct mbuf *m, *m2; 5460 int error = 0, retlen, tlen, trimlen; 5461 5462 NFSD_DEBUG(4, "in nfsrv_readdsrpc\n"); 5463 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5464 *mpp = NULL; 5465 /* 5466 * Use a stateid where other is an alternating 01010 pattern and 5467 * seqid is 0xffffffff. This value is not defined as special by 5468 * the RFC and is used by the FreeBSD NFS server to indicate an 5469 * MDS->DS proxy operation. 5470 */ 5471 st.other[0] = 0x55555555; 5472 st.other[1] = 0x55555555; 5473 st.other[2] = 0x55555555; 5474 st.seqid = 0xffffffff; 5475 nfscl_reqstart(nd, NFSPROC_READDS, nmp, (u_int8_t *)fhp, sizeof(*fhp), 5476 NULL, NULL, 0, 0, cred); 5477 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5478 NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3); 5479 txdr_hyper(off, tl); 5480 *(tl + 2) = txdr_unsigned(len); 5481 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 5482 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5483 if (error != 0) { 5484 free(nd, M_TEMP); 5485 return (error); 5486 } 5487 if (nd->nd_repstat == 0) { 5488 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 5489 NFSM_STRSIZ(retlen, len); 5490 if (retlen > 0) { 5491 /* Trim off the pre-data XDR from the mbuf chain. */ 5492 m = nd->nd_mrep; 5493 while (m != NULL && m != nd->nd_md) { 5494 if (m->m_next == nd->nd_md) { 5495 m->m_next = NULL; 5496 m_freem(nd->nd_mrep); 5497 nd->nd_mrep = m = nd->nd_md; 5498 } else 5499 m = m->m_next; 5500 } 5501 if (m == NULL) { 5502 printf("nfsrv_readdsrpc: busted mbuf list\n"); 5503 error = ENOENT; 5504 goto nfsmout; 5505 } 5506 5507 /* 5508 * Now, adjust first mbuf so that any XDR before the 5509 * read data is skipped over. 5510 */ 5511 trimlen = nd->nd_dpos - mtod(m, char *); 5512 if (trimlen > 0) { 5513 m->m_len -= trimlen; 5514 NFSM_DATAP(m, trimlen); 5515 } 5516 5517 /* 5518 * Truncate the mbuf chain at retlen bytes of data, 5519 * plus XDR padding that brings the length up to a 5520 * multiple of 4. 5521 */ 5522 tlen = NFSM_RNDUP(retlen); 5523 do { 5524 if (m->m_len >= tlen) { 5525 m->m_len = tlen; 5526 tlen = 0; 5527 m2 = m->m_next; 5528 m->m_next = NULL; 5529 m_freem(m2); 5530 break; 5531 } 5532 tlen -= m->m_len; 5533 m = m->m_next; 5534 } while (m != NULL); 5535 if (tlen > 0) { 5536 printf("nfsrv_readdsrpc: busted mbuf list\n"); 5537 error = ENOENT; 5538 goto nfsmout; 5539 } 5540 *mpp = nd->nd_mrep; 5541 *mpendp = m; 5542 nd->nd_mrep = NULL; 5543 } 5544 } else 5545 error = nd->nd_repstat; 5546 nfsmout: 5547 /* If nd->nd_mrep is already NULL, this is a no-op. */ 5548 m_freem(nd->nd_mrep); 5549 free(nd, M_TEMP); 5550 NFSD_DEBUG(4, "nfsrv_readdsrpc error=%d\n", error); 5551 return (error); 5552 } 5553 5554 /* 5555 * Do a write RPC on a DS data file, using this structure for the arguments, 5556 * so that this function can be executed by a separate kernel process. 5557 */ 5558 struct nfsrvwritedsdorpc { 5559 int done; 5560 int inprog; 5561 struct task tsk; 5562 fhandle_t fh; 5563 off_t off; 5564 int len; 5565 struct nfsmount *nmp; 5566 struct ucred *cred; 5567 NFSPROC_T *p; 5568 struct mbuf *m; 5569 int err; 5570 }; 5571 5572 static int 5573 nfsrv_writedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, int len, 5574 struct nfsvattr *nap, struct mbuf *m, struct ucred *cred, NFSPROC_T *p) 5575 { 5576 uint32_t *tl; 5577 struct nfsrv_descript *nd; 5578 nfsattrbit_t attrbits; 5579 nfsv4stateid_t st; 5580 int commit, error, retlen; 5581 5582 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5583 nfscl_reqstart(nd, NFSPROC_WRITE, nmp, (u_int8_t *)fhp, 5584 sizeof(fhandle_t), NULL, NULL, 0, 0, cred); 5585 5586 /* 5587 * Use a stateid where other is an alternating 01010 pattern and 5588 * seqid is 0xffffffff. This value is not defined as special by 5589 * the RFC and is used by the FreeBSD NFS server to indicate an 5590 * MDS->DS proxy operation. 5591 */ 5592 st.other[0] = 0x55555555; 5593 st.other[1] = 0x55555555; 5594 st.other[2] = 0x55555555; 5595 st.seqid = 0xffffffff; 5596 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5597 NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED); 5598 txdr_hyper(off, tl); 5599 tl += 2; 5600 /* 5601 * Do all writes FileSync, since the server doesn't hold onto dirty 5602 * buffers. Since clients should be accessing the DS servers directly 5603 * using the pNFS layouts, this just needs to work correctly as a 5604 * fallback. 5605 */ 5606 *tl++ = txdr_unsigned(NFSWRITE_FILESYNC); 5607 *tl = txdr_unsigned(len); 5608 NFSD_DEBUG(4, "nfsrv_writedsdorpc: len=%d\n", len); 5609 5610 /* Put data in mbuf chain. */ 5611 nd->nd_mb->m_next = m; 5612 5613 /* Set nd_mb and nd_bpos to end of data. */ 5614 while (m->m_next != NULL) 5615 m = m->m_next; 5616 nd->nd_mb = m; 5617 nfsm_set(nd, m->m_len); 5618 NFSD_DEBUG(4, "nfsrv_writedsdorpc: lastmb len=%d\n", m->m_len); 5619 5620 /* Do a Getattr for the attributes that change upon writing. */ 5621 NFSZERO_ATTRBIT(&attrbits); 5622 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 5623 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 5624 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 5625 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 5626 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 5627 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 5628 *tl = txdr_unsigned(NFSV4OP_GETATTR); 5629 (void) nfsrv_putattrbit(nd, &attrbits); 5630 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, 5631 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5632 if (error != 0) { 5633 free(nd, M_TEMP); 5634 return (error); 5635 } 5636 NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft writerpc=%d\n", nd->nd_repstat); 5637 /* Get rid of weak cache consistency data for now. */ 5638 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == 5639 (ND_NFSV4 | ND_V4WCCATTR)) { 5640 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5641 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL); 5642 NFSD_DEBUG(4, "nfsrv_writedsdorpc: wcc attr=%d\n", error); 5643 if (error != 0) 5644 goto nfsmout; 5645 /* 5646 * Get rid of Op# and status for next op. 5647 */ 5648 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5649 if (*++tl != 0) 5650 nd->nd_flag |= ND_NOMOREDATA; 5651 } 5652 if (nd->nd_repstat == 0) { 5653 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF); 5654 retlen = fxdr_unsigned(int, *tl++); 5655 commit = fxdr_unsigned(int, *tl); 5656 if (commit != NFSWRITE_FILESYNC) 5657 error = NFSERR_IO; 5658 NFSD_DEBUG(4, "nfsrv_writedsdorpc:retlen=%d commit=%d err=%d\n", 5659 retlen, commit, error); 5660 } else 5661 error = nd->nd_repstat; 5662 /* We have no use for the Write Verifier since we use FileSync. */ 5663 5664 /* 5665 * Get the Change, Size, Access Time and Modify Time attributes and set 5666 * on the Metadata file, so its attributes will be what the file's 5667 * would be if it had been written. 5668 */ 5669 if (error == 0) { 5670 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5671 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5672 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL); 5673 } 5674 NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft loadattr=%d\n", error); 5675 nfsmout: 5676 m_freem(nd->nd_mrep); 5677 free(nd, M_TEMP); 5678 NFSD_DEBUG(4, "nfsrv_writedsdorpc error=%d\n", error); 5679 return (error); 5680 } 5681 5682 /* 5683 * Start up the thread that will execute nfsrv_writedsdorpc(). 5684 */ 5685 static void 5686 start_writedsdorpc(void *arg, int pending) 5687 { 5688 struct nfsrvwritedsdorpc *drpc; 5689 5690 drpc = (struct nfsrvwritedsdorpc *)arg; 5691 drpc->err = nfsrv_writedsdorpc(drpc->nmp, &drpc->fh, drpc->off, 5692 drpc->len, NULL, drpc->m, drpc->cred, drpc->p); 5693 drpc->done = 1; 5694 NFSD_DEBUG(4, "start_writedsdorpc: err=%d\n", drpc->err); 5695 } 5696 5697 static int 5698 nfsrv_writedsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred, 5699 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 5700 struct mbuf **mpp, char *cp, int *failposp) 5701 { 5702 struct nfsrvwritedsdorpc *drpc, *tdrpc = NULL; 5703 struct nfsvattr na; 5704 struct mbuf *m; 5705 int error, i, offs, ret, timo; 5706 5707 NFSD_DEBUG(4, "in nfsrv_writedsrpc\n"); 5708 KASSERT(*mpp != NULL, ("nfsrv_writedsrpc: NULL mbuf chain")); 5709 drpc = NULL; 5710 if (mirrorcnt > 1) 5711 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 5712 M_WAITOK); 5713 5714 /* Calculate offset in mbuf chain that data starts. */ 5715 offs = cp - mtod(*mpp, char *); 5716 NFSD_DEBUG(4, "nfsrv_writedsrpc: mcopy offs=%d len=%d\n", offs, len); 5717 5718 /* 5719 * Do the write RPC for every DS, using a separate kernel process 5720 * for every DS except the last one. 5721 */ 5722 error = 0; 5723 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5724 tdrpc->done = 0; 5725 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 5726 tdrpc->off = off; 5727 tdrpc->len = len; 5728 tdrpc->nmp = *nmpp; 5729 tdrpc->cred = cred; 5730 tdrpc->p = p; 5731 tdrpc->inprog = 0; 5732 tdrpc->err = 0; 5733 tdrpc->m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK); 5734 ret = EIO; 5735 if (nfs_pnfsiothreads != 0) { 5736 ret = nfs_pnfsio(start_writedsdorpc, tdrpc); 5737 NFSD_DEBUG(4, "nfsrv_writedsrpc: nfs_pnfsio=%d\n", 5738 ret); 5739 } 5740 if (ret != 0) { 5741 ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, NULL, 5742 tdrpc->m, cred, p); 5743 if (nfsds_failerr(ret) && *failposp == -1) 5744 *failposp = i; 5745 else if (error == 0 && ret != 0) 5746 error = ret; 5747 } 5748 nmpp++; 5749 fhp++; 5750 } 5751 m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK); 5752 ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, &na, m, cred, p); 5753 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 5754 *failposp = mirrorcnt - 1; 5755 else if (error == 0 && ret != 0) 5756 error = ret; 5757 if (error == 0) 5758 error = nfsrv_setextattr(vp, &na, p); 5759 NFSD_DEBUG(4, "nfsrv_writedsrpc: aft setextat=%d\n", error); 5760 tdrpc = drpc; 5761 timo = hz / 50; /* Wait for 20msec. */ 5762 if (timo < 1) 5763 timo = 1; 5764 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5765 /* Wait for RPCs on separate threads to complete. */ 5766 while (tdrpc->inprog != 0 && tdrpc->done == 0) 5767 tsleep(&tdrpc->tsk, PVFS, "srvwrds", timo); 5768 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 5769 *failposp = i; 5770 else if (error == 0 && tdrpc->err != 0) 5771 error = tdrpc->err; 5772 } 5773 free(drpc, M_TEMP); 5774 return (error); 5775 } 5776 5777 /* 5778 * Do a allocate RPC on a DS data file, using this structure for the arguments, 5779 * so that this function can be executed by a separate kernel process. 5780 */ 5781 struct nfsrvallocatedsdorpc { 5782 int done; 5783 int inprog; 5784 struct task tsk; 5785 fhandle_t fh; 5786 off_t off; 5787 off_t len; 5788 struct nfsmount *nmp; 5789 struct ucred *cred; 5790 NFSPROC_T *p; 5791 int err; 5792 }; 5793 5794 static int 5795 nfsrv_allocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, 5796 off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p) 5797 { 5798 uint32_t *tl; 5799 struct nfsrv_descript *nd; 5800 nfsattrbit_t attrbits; 5801 nfsv4stateid_t st; 5802 int error; 5803 5804 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5805 nfscl_reqstart(nd, NFSPROC_ALLOCATE, nmp, (u_int8_t *)fhp, 5806 sizeof(fhandle_t), NULL, NULL, 0, 0, cred); 5807 5808 /* 5809 * Use a stateid where other is an alternating 01010 pattern and 5810 * seqid is 0xffffffff. This value is not defined as special by 5811 * the RFC and is used by the FreeBSD NFS server to indicate an 5812 * MDS->DS proxy operation. 5813 */ 5814 st.other[0] = 0x55555555; 5815 st.other[1] = 0x55555555; 5816 st.other[2] = 0x55555555; 5817 st.seqid = 0xffffffff; 5818 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5819 NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED); 5820 txdr_hyper(off, tl); tl += 2; 5821 txdr_hyper(len, tl); tl += 2; 5822 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: len=%jd\n", (intmax_t)len); 5823 5824 *tl = txdr_unsigned(NFSV4OP_GETATTR); 5825 NFSGETATTR_ATTRBIT(&attrbits); 5826 nfsrv_putattrbit(nd, &attrbits); 5827 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, 5828 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5829 if (error != 0) { 5830 free(nd, M_TEMP); 5831 return (error); 5832 } 5833 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft allocaterpc=%d\n", 5834 nd->nd_repstat); 5835 if (nd->nd_repstat == 0) { 5836 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5837 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5838 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL); 5839 } else 5840 error = nd->nd_repstat; 5841 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft loadattr=%d\n", error); 5842 nfsmout: 5843 m_freem(nd->nd_mrep); 5844 free(nd, M_TEMP); 5845 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc error=%d\n", error); 5846 return (error); 5847 } 5848 5849 /* 5850 * Start up the thread that will execute nfsrv_allocatedsdorpc(). 5851 */ 5852 static void 5853 start_allocatedsdorpc(void *arg, int pending) 5854 { 5855 struct nfsrvallocatedsdorpc *drpc; 5856 5857 drpc = (struct nfsrvallocatedsdorpc *)arg; 5858 drpc->err = nfsrv_allocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off, 5859 drpc->len, NULL, drpc->cred, drpc->p); 5860 drpc->done = 1; 5861 NFSD_DEBUG(4, "start_allocatedsdorpc: err=%d\n", drpc->err); 5862 } 5863 5864 static int 5865 nfsrv_allocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred, 5866 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 5867 int *failposp) 5868 { 5869 struct nfsrvallocatedsdorpc *drpc, *tdrpc = NULL; 5870 struct nfsvattr na; 5871 int error, i, ret, timo; 5872 5873 NFSD_DEBUG(4, "in nfsrv_allocatedsrpc\n"); 5874 drpc = NULL; 5875 if (mirrorcnt > 1) 5876 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 5877 M_WAITOK); 5878 5879 /* 5880 * Do the allocate RPC for every DS, using a separate kernel process 5881 * for every DS except the last one. 5882 */ 5883 error = 0; 5884 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5885 tdrpc->done = 0; 5886 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 5887 tdrpc->off = off; 5888 tdrpc->len = len; 5889 tdrpc->nmp = *nmpp; 5890 tdrpc->cred = cred; 5891 tdrpc->p = p; 5892 tdrpc->inprog = 0; 5893 tdrpc->err = 0; 5894 ret = EIO; 5895 if (nfs_pnfsiothreads != 0) { 5896 ret = nfs_pnfsio(start_allocatedsdorpc, tdrpc); 5897 NFSD_DEBUG(4, "nfsrv_allocatedsrpc: nfs_pnfsio=%d\n", 5898 ret); 5899 } 5900 if (ret != 0) { 5901 ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, NULL, 5902 cred, p); 5903 if (nfsds_failerr(ret) && *failposp == -1) 5904 *failposp = i; 5905 else if (error == 0 && ret != 0) 5906 error = ret; 5907 } 5908 nmpp++; 5909 fhp++; 5910 } 5911 ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p); 5912 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 5913 *failposp = mirrorcnt - 1; 5914 else if (error == 0 && ret != 0) 5915 error = ret; 5916 if (error == 0) 5917 error = nfsrv_setextattr(vp, &na, p); 5918 NFSD_DEBUG(4, "nfsrv_allocatedsrpc: aft setextat=%d\n", error); 5919 tdrpc = drpc; 5920 timo = hz / 50; /* Wait for 20msec. */ 5921 if (timo < 1) 5922 timo = 1; 5923 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5924 /* Wait for RPCs on separate threads to complete. */ 5925 while (tdrpc->inprog != 0 && tdrpc->done == 0) 5926 tsleep(&tdrpc->tsk, PVFS, "srvalds", timo); 5927 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 5928 *failposp = i; 5929 else if (error == 0 && tdrpc->err != 0) 5930 error = tdrpc->err; 5931 } 5932 free(drpc, M_TEMP); 5933 return (error); 5934 } 5935 5936 /* 5937 * Do a deallocate RPC on a DS data file, using this structure for the 5938 * arguments, so that this function can be executed by a separate kernel 5939 * process. 5940 */ 5941 struct nfsrvdeallocatedsdorpc { 5942 int done; 5943 int inprog; 5944 struct task tsk; 5945 fhandle_t fh; 5946 off_t off; 5947 off_t len; 5948 struct nfsmount *nmp; 5949 struct ucred *cred; 5950 NFSPROC_T *p; 5951 int err; 5952 }; 5953 5954 static int 5955 nfsrv_deallocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, 5956 off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p) 5957 { 5958 uint32_t *tl; 5959 struct nfsrv_descript *nd; 5960 nfsattrbit_t attrbits; 5961 nfsv4stateid_t st; 5962 int error; 5963 5964 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5965 nfscl_reqstart(nd, NFSPROC_DEALLOCATE, nmp, (u_int8_t *)fhp, 5966 sizeof(fhandle_t), NULL, NULL, 0, 0, cred); 5967 5968 /* 5969 * Use a stateid where other is an alternating 01010 pattern and 5970 * seqid is 0xffffffff. This value is not defined as special by 5971 * the RFC and is used by the FreeBSD NFS server to indicate an 5972 * MDS->DS proxy operation. 5973 */ 5974 st.other[0] = 0x55555555; 5975 st.other[1] = 0x55555555; 5976 st.other[2] = 0x55555555; 5977 st.seqid = 0xffffffff; 5978 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5979 NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED); 5980 txdr_hyper(off, tl); tl += 2; 5981 txdr_hyper(len, tl); tl += 2; 5982 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: len=%jd\n", (intmax_t)len); 5983 5984 /* Do a Getattr for the attributes that change upon writing. */ 5985 NFSZERO_ATTRBIT(&attrbits); 5986 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 5987 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 5988 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 5989 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 5990 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 5991 *tl = txdr_unsigned(NFSV4OP_GETATTR); 5992 nfsrv_putattrbit(nd, &attrbits); 5993 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, 5994 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5995 if (error != 0) { 5996 free(nd, M_TEMP); 5997 return (error); 5998 } 5999 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft deallocaterpc=%d\n", 6000 nd->nd_repstat); 6001 /* Get rid of weak cache consistency data for now. */ 6002 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == 6003 (ND_NFSV4 | ND_V4WCCATTR)) { 6004 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 6005 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL); 6006 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: wcc attr=%d\n", error); 6007 if (error != 0) 6008 goto nfsmout; 6009 /* 6010 * Get rid of Op# and status for next op. 6011 */ 6012 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 6013 if (*++tl != 0) 6014 nd->nd_flag |= ND_NOMOREDATA; 6015 } 6016 if (nd->nd_repstat == 0) { 6017 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 6018 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 6019 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL); 6020 } else 6021 error = nd->nd_repstat; 6022 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft loadattr=%d\n", error); 6023 nfsmout: 6024 m_freem(nd->nd_mrep); 6025 free(nd, M_TEMP); 6026 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc error=%d\n", error); 6027 return (error); 6028 } 6029 6030 /* 6031 * Start up the thread that will execute nfsrv_deallocatedsdorpc(). 6032 */ 6033 static void 6034 start_deallocatedsdorpc(void *arg, int pending) 6035 { 6036 struct nfsrvdeallocatedsdorpc *drpc; 6037 6038 drpc = (struct nfsrvdeallocatedsdorpc *)arg; 6039 drpc->err = nfsrv_deallocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off, 6040 drpc->len, NULL, drpc->cred, drpc->p); 6041 drpc->done = 1; 6042 NFSD_DEBUG(4, "start_deallocatedsdorpc: err=%d\n", drpc->err); 6043 } 6044 6045 static int 6046 nfsrv_deallocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred, 6047 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 6048 int *failposp) 6049 { 6050 struct nfsrvdeallocatedsdorpc *drpc, *tdrpc = NULL; 6051 struct nfsvattr na; 6052 int error, i, ret, timo; 6053 6054 NFSD_DEBUG(4, "in nfsrv_deallocatedsrpc\n"); 6055 drpc = NULL; 6056 if (mirrorcnt > 1) 6057 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 6058 M_WAITOK); 6059 6060 /* 6061 * Do the deallocate RPC for every DS, using a separate kernel process 6062 * for every DS except the last one. 6063 */ 6064 error = 0; 6065 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6066 tdrpc->done = 0; 6067 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 6068 tdrpc->off = off; 6069 tdrpc->len = len; 6070 tdrpc->nmp = *nmpp; 6071 tdrpc->cred = cred; 6072 tdrpc->p = p; 6073 tdrpc->inprog = 0; 6074 tdrpc->err = 0; 6075 ret = EIO; 6076 if (nfs_pnfsiothreads != 0) { 6077 ret = nfs_pnfsio(start_deallocatedsdorpc, tdrpc); 6078 NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: nfs_pnfsio=%d\n", 6079 ret); 6080 } 6081 if (ret != 0) { 6082 ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len, 6083 NULL, cred, p); 6084 if (nfsds_failerr(ret) && *failposp == -1) 6085 *failposp = i; 6086 else if (error == 0 && ret != 0) 6087 error = ret; 6088 } 6089 nmpp++; 6090 fhp++; 6091 } 6092 ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p); 6093 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 6094 *failposp = mirrorcnt - 1; 6095 else if (error == 0 && ret != 0) 6096 error = ret; 6097 if (error == 0) 6098 error = nfsrv_setextattr(vp, &na, p); 6099 NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: aft setextat=%d\n", error); 6100 tdrpc = drpc; 6101 timo = hz / 50; /* Wait for 20msec. */ 6102 if (timo < 1) 6103 timo = 1; 6104 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6105 /* Wait for RPCs on separate threads to complete. */ 6106 while (tdrpc->inprog != 0 && tdrpc->done == 0) 6107 tsleep(&tdrpc->tsk, PVFS, "srvalds", timo); 6108 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 6109 *failposp = i; 6110 else if (error == 0 && tdrpc->err != 0) 6111 error = tdrpc->err; 6112 } 6113 free(drpc, M_TEMP); 6114 return (error); 6115 } 6116 6117 static int 6118 nfsrv_setattrdsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 6119 struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap, 6120 struct nfsvattr *dsnap) 6121 { 6122 uint32_t *tl; 6123 struct nfsrv_descript *nd; 6124 nfsv4stateid_t st; 6125 nfsattrbit_t attrbits; 6126 int error; 6127 6128 NFSD_DEBUG(4, "in nfsrv_setattrdsdorpc\n"); 6129 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 6130 /* 6131 * Use a stateid where other is an alternating 01010 pattern and 6132 * seqid is 0xffffffff. This value is not defined as special by 6133 * the RFC and is used by the FreeBSD NFS server to indicate an 6134 * MDS->DS proxy operation. 6135 */ 6136 st.other[0] = 0x55555555; 6137 st.other[1] = 0x55555555; 6138 st.other[2] = 0x55555555; 6139 st.seqid = 0xffffffff; 6140 nfscl_reqstart(nd, NFSPROC_SETATTR, nmp, (u_int8_t *)fhp, sizeof(*fhp), 6141 NULL, NULL, 0, 0, cred); 6142 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 6143 nfscl_fillsattr(nd, &nap->na_vattr, vp, NFSSATTR_FULL, 0); 6144 6145 /* Do a Getattr for the attributes that change due to writing. */ 6146 NFSZERO_ATTRBIT(&attrbits); 6147 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 6148 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 6149 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 6150 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 6151 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 6152 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 6153 *tl = txdr_unsigned(NFSV4OP_GETATTR); 6154 (void) nfsrv_putattrbit(nd, &attrbits); 6155 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 6156 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 6157 if (error != 0) { 6158 free(nd, M_TEMP); 6159 return (error); 6160 } 6161 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattrrpc=%d\n", 6162 nd->nd_repstat); 6163 /* Get rid of weak cache consistency data for now. */ 6164 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == 6165 (ND_NFSV4 | ND_V4WCCATTR)) { 6166 error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL, 6167 NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL); 6168 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: wcc attr=%d\n", error); 6169 if (error != 0) 6170 goto nfsmout; 6171 /* 6172 * Get rid of Op# and status for next op. 6173 */ 6174 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 6175 if (*++tl != 0) 6176 nd->nd_flag |= ND_NOMOREDATA; 6177 } 6178 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); 6179 if (error != 0) 6180 goto nfsmout; 6181 if (nd->nd_repstat != 0) 6182 error = nd->nd_repstat; 6183 /* 6184 * Get the Change, Size, Access Time and Modify Time attributes and set 6185 * on the Metadata file, so its attributes will be what the file's 6186 * would be if it had been written. 6187 */ 6188 if (error == 0) { 6189 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 6190 error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL, 6191 NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 6192 NULL); 6193 } 6194 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattr loadattr=%d\n", error); 6195 nfsmout: 6196 m_freem(nd->nd_mrep); 6197 free(nd, M_TEMP); 6198 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc error=%d\n", error); 6199 return (error); 6200 } 6201 6202 struct nfsrvsetattrdsdorpc { 6203 int done; 6204 int inprog; 6205 struct task tsk; 6206 fhandle_t fh; 6207 struct nfsmount *nmp; 6208 struct vnode *vp; 6209 struct ucred *cred; 6210 NFSPROC_T *p; 6211 struct nfsvattr na; 6212 struct nfsvattr dsna; 6213 int err; 6214 }; 6215 6216 /* 6217 * Start up the thread that will execute nfsrv_setattrdsdorpc(). 6218 */ 6219 static void 6220 start_setattrdsdorpc(void *arg, int pending) 6221 { 6222 struct nfsrvsetattrdsdorpc *drpc; 6223 6224 drpc = (struct nfsrvsetattrdsdorpc *)arg; 6225 drpc->err = nfsrv_setattrdsdorpc(&drpc->fh, drpc->cred, drpc->p, 6226 drpc->vp, drpc->nmp, &drpc->na, &drpc->dsna); 6227 drpc->done = 1; 6228 } 6229 6230 static int 6231 nfsrv_setattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 6232 struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 6233 struct nfsvattr *nap, int *failposp) 6234 { 6235 struct nfsrvsetattrdsdorpc *drpc, *tdrpc = NULL; 6236 struct nfsvattr na; 6237 int error, i, ret, timo; 6238 6239 NFSD_DEBUG(4, "in nfsrv_setattrdsrpc\n"); 6240 drpc = NULL; 6241 if (mirrorcnt > 1) 6242 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 6243 M_WAITOK); 6244 6245 /* 6246 * Do the setattr RPC for every DS, using a separate kernel process 6247 * for every DS except the last one. 6248 */ 6249 error = 0; 6250 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6251 tdrpc->done = 0; 6252 tdrpc->inprog = 0; 6253 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 6254 tdrpc->nmp = *nmpp; 6255 tdrpc->vp = vp; 6256 tdrpc->cred = cred; 6257 tdrpc->p = p; 6258 tdrpc->na = *nap; 6259 tdrpc->err = 0; 6260 ret = EIO; 6261 if (nfs_pnfsiothreads != 0) { 6262 ret = nfs_pnfsio(start_setattrdsdorpc, tdrpc); 6263 NFSD_DEBUG(4, "nfsrv_setattrdsrpc: nfs_pnfsio=%d\n", 6264 ret); 6265 } 6266 if (ret != 0) { 6267 ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, 6268 &na); 6269 if (nfsds_failerr(ret) && *failposp == -1) 6270 *failposp = i; 6271 else if (error == 0 && ret != 0) 6272 error = ret; 6273 } 6274 nmpp++; 6275 fhp++; 6276 } 6277 ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na); 6278 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 6279 *failposp = mirrorcnt - 1; 6280 else if (error == 0 && ret != 0) 6281 error = ret; 6282 if (error == 0) 6283 error = nfsrv_setextattr(vp, &na, p); 6284 NFSD_DEBUG(4, "nfsrv_setattrdsrpc: aft setextat=%d\n", error); 6285 tdrpc = drpc; 6286 timo = hz / 50; /* Wait for 20msec. */ 6287 if (timo < 1) 6288 timo = 1; 6289 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6290 /* Wait for RPCs on separate threads to complete. */ 6291 while (tdrpc->inprog != 0 && tdrpc->done == 0) 6292 tsleep(&tdrpc->tsk, PVFS, "srvsads", timo); 6293 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 6294 *failposp = i; 6295 else if (error == 0 && tdrpc->err != 0) 6296 error = tdrpc->err; 6297 } 6298 free(drpc, M_TEMP); 6299 return (error); 6300 } 6301 6302 /* 6303 * Do a Setattr of an NFSv4 ACL on the DS file. 6304 */ 6305 static int 6306 nfsrv_setacldsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 6307 struct vnode *vp, struct nfsmount *nmp, struct acl *aclp) 6308 { 6309 struct nfsrv_descript *nd; 6310 nfsv4stateid_t st; 6311 nfsattrbit_t attrbits; 6312 int error; 6313 6314 NFSD_DEBUG(4, "in nfsrv_setacldsdorpc\n"); 6315 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 6316 /* 6317 * Use a stateid where other is an alternating 01010 pattern and 6318 * seqid is 0xffffffff. This value is not defined as special by 6319 * the RFC and is used by the FreeBSD NFS server to indicate an 6320 * MDS->DS proxy operation. 6321 */ 6322 st.other[0] = 0x55555555; 6323 st.other[1] = 0x55555555; 6324 st.other[2] = 0x55555555; 6325 st.seqid = 0xffffffff; 6326 nfscl_reqstart(nd, NFSPROC_SETACL, nmp, (u_int8_t *)fhp, sizeof(*fhp), 6327 NULL, NULL, 0, 0, cred); 6328 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 6329 NFSZERO_ATTRBIT(&attrbits); 6330 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_ACL); 6331 /* 6332 * The "vp" argument to nfsv4_fillattr() is only used for vnode_type(), 6333 * so passing in the metadata "vp" will be ok, since it is of 6334 * the same type (VREG). 6335 */ 6336 nfsv4_fillattr(nd, NULL, vp, aclp, NULL, NULL, 0, &attrbits, NULL, 6337 NULL, 0, 0, 0, 0, 0, NULL); 6338 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 6339 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 6340 if (error != 0) { 6341 free(nd, M_TEMP); 6342 return (error); 6343 } 6344 NFSD_DEBUG(4, "nfsrv_setacldsdorpc: aft setaclrpc=%d\n", 6345 nd->nd_repstat); 6346 error = nd->nd_repstat; 6347 m_freem(nd->nd_mrep); 6348 free(nd, M_TEMP); 6349 return (error); 6350 } 6351 6352 struct nfsrvsetacldsdorpc { 6353 int done; 6354 int inprog; 6355 struct task tsk; 6356 fhandle_t fh; 6357 struct nfsmount *nmp; 6358 struct vnode *vp; 6359 struct ucred *cred; 6360 NFSPROC_T *p; 6361 struct acl *aclp; 6362 int err; 6363 }; 6364 6365 /* 6366 * Start up the thread that will execute nfsrv_setacldsdorpc(). 6367 */ 6368 static void 6369 start_setacldsdorpc(void *arg, int pending) 6370 { 6371 struct nfsrvsetacldsdorpc *drpc; 6372 6373 drpc = (struct nfsrvsetacldsdorpc *)arg; 6374 drpc->err = nfsrv_setacldsdorpc(&drpc->fh, drpc->cred, drpc->p, 6375 drpc->vp, drpc->nmp, drpc->aclp); 6376 drpc->done = 1; 6377 } 6378 6379 static int 6380 nfsrv_setacldsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 6381 struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct acl *aclp, 6382 int *failposp) 6383 { 6384 struct nfsrvsetacldsdorpc *drpc, *tdrpc = NULL; 6385 int error, i, ret, timo; 6386 6387 NFSD_DEBUG(4, "in nfsrv_setacldsrpc\n"); 6388 drpc = NULL; 6389 if (mirrorcnt > 1) 6390 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 6391 M_WAITOK); 6392 6393 /* 6394 * Do the setattr RPC for every DS, using a separate kernel process 6395 * for every DS except the last one. 6396 */ 6397 error = 0; 6398 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6399 tdrpc->done = 0; 6400 tdrpc->inprog = 0; 6401 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 6402 tdrpc->nmp = *nmpp; 6403 tdrpc->vp = vp; 6404 tdrpc->cred = cred; 6405 tdrpc->p = p; 6406 tdrpc->aclp = aclp; 6407 tdrpc->err = 0; 6408 ret = EIO; 6409 if (nfs_pnfsiothreads != 0) { 6410 ret = nfs_pnfsio(start_setacldsdorpc, tdrpc); 6411 NFSD_DEBUG(4, "nfsrv_setacldsrpc: nfs_pnfsio=%d\n", 6412 ret); 6413 } 6414 if (ret != 0) { 6415 ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, 6416 aclp); 6417 if (nfsds_failerr(ret) && *failposp == -1) 6418 *failposp = i; 6419 else if (error == 0 && ret != 0) 6420 error = ret; 6421 } 6422 nmpp++; 6423 fhp++; 6424 } 6425 ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp); 6426 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 6427 *failposp = mirrorcnt - 1; 6428 else if (error == 0 && ret != 0) 6429 error = ret; 6430 NFSD_DEBUG(4, "nfsrv_setacldsrpc: aft setextat=%d\n", error); 6431 tdrpc = drpc; 6432 timo = hz / 50; /* Wait for 20msec. */ 6433 if (timo < 1) 6434 timo = 1; 6435 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6436 /* Wait for RPCs on separate threads to complete. */ 6437 while (tdrpc->inprog != 0 && tdrpc->done == 0) 6438 tsleep(&tdrpc->tsk, PVFS, "srvacds", timo); 6439 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 6440 *failposp = i; 6441 else if (error == 0 && tdrpc->err != 0) 6442 error = tdrpc->err; 6443 } 6444 free(drpc, M_TEMP); 6445 return (error); 6446 } 6447 6448 /* 6449 * Getattr call to the DS for the attributes that change due to writing. 6450 */ 6451 static int 6452 nfsrv_getattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 6453 struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap) 6454 { 6455 struct nfsrv_descript *nd; 6456 int error; 6457 nfsattrbit_t attrbits; 6458 6459 NFSD_DEBUG(4, "in nfsrv_getattrdsrpc\n"); 6460 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 6461 nfscl_reqstart(nd, NFSPROC_GETATTR, nmp, (u_int8_t *)fhp, 6462 sizeof(fhandle_t), NULL, NULL, 0, 0, cred); 6463 NFSZERO_ATTRBIT(&attrbits); 6464 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 6465 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 6466 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 6467 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 6468 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 6469 (void) nfsrv_putattrbit(nd, &attrbits); 6470 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 6471 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 6472 if (error != 0) { 6473 free(nd, M_TEMP); 6474 return (error); 6475 } 6476 NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft getattrrpc=%d\n", 6477 nd->nd_repstat); 6478 if (nd->nd_repstat == 0) { 6479 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, 6480 NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, 6481 NULL, NULL, NULL); 6482 /* 6483 * We can only save the updated values in the extended 6484 * attribute if the vp is exclusively locked. 6485 * This should happen when any of the following operations 6486 * occur on the vnode: 6487 * Close, Delegreturn, LayoutCommit, LayoutReturn 6488 * As such, the updated extended attribute should get saved 6489 * before nfsrv_checkdsattr() returns 0 and allows the cached 6490 * attributes to be returned without calling this function. 6491 */ 6492 if (error == 0 && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) { 6493 error = nfsrv_setextattr(vp, nap, p); 6494 NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft setextat=%d\n", 6495 error); 6496 } 6497 } else 6498 error = nd->nd_repstat; 6499 m_freem(nd->nd_mrep); 6500 free(nd, M_TEMP); 6501 NFSD_DEBUG(4, "nfsrv_getattrdsrpc error=%d\n", error); 6502 return (error); 6503 } 6504 6505 /* 6506 * Seek call to a DS. 6507 */ 6508 static int 6509 nfsrv_seekdsrpc(fhandle_t *fhp, off_t *offp, int content, bool *eofp, 6510 struct ucred *cred, NFSPROC_T *p, struct nfsmount *nmp) 6511 { 6512 uint32_t *tl; 6513 struct nfsrv_descript *nd; 6514 nfsv4stateid_t st; 6515 int error; 6516 6517 NFSD_DEBUG(4, "in nfsrv_seekdsrpc\n"); 6518 /* 6519 * Use a stateid where other is an alternating 01010 pattern and 6520 * seqid is 0xffffffff. This value is not defined as special by 6521 * the RFC and is used by the FreeBSD NFS server to indicate an 6522 * MDS->DS proxy operation. 6523 */ 6524 st.other[0] = 0x55555555; 6525 st.other[1] = 0x55555555; 6526 st.other[2] = 0x55555555; 6527 st.seqid = 0xffffffff; 6528 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 6529 nfscl_reqstart(nd, NFSPROC_SEEKDS, nmp, (u_int8_t *)fhp, 6530 sizeof(fhandle_t), NULL, NULL, 0, 0, cred); 6531 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 6532 NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + NFSX_UNSIGNED); 6533 txdr_hyper(*offp, tl); tl += 2; 6534 *tl = txdr_unsigned(content); 6535 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 6536 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 6537 if (error != 0) { 6538 free(nd, M_TEMP); 6539 return (error); 6540 } 6541 NFSD_DEBUG(4, "nfsrv_seekdsrpc: aft seekrpc=%d\n", nd->nd_repstat); 6542 if (nd->nd_repstat == 0) { 6543 NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED + NFSX_HYPER); 6544 if (*tl++ == newnfs_true) 6545 *eofp = true; 6546 else 6547 *eofp = false; 6548 *offp = fxdr_hyper(tl); 6549 } else 6550 error = nd->nd_repstat; 6551 nfsmout: 6552 m_freem(nd->nd_mrep); 6553 free(nd, M_TEMP); 6554 NFSD_DEBUG(4, "nfsrv_seekdsrpc error=%d\n", error); 6555 return (error); 6556 } 6557 6558 /* 6559 * Get the device id and file handle for a DS file. 6560 */ 6561 int 6562 nfsrv_dsgetdevandfh(struct vnode *vp, NFSPROC_T *p, int *mirrorcntp, 6563 fhandle_t *fhp, char *devid) 6564 { 6565 int buflen, error; 6566 char *buf; 6567 6568 buflen = 1024; 6569 buf = malloc(buflen, M_TEMP, M_WAITOK); 6570 error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, NULL, 6571 fhp, devid, NULL, NULL, NULL, NULL, NULL, NULL); 6572 free(buf, M_TEMP); 6573 return (error); 6574 } 6575 6576 /* 6577 * Do a Lookup against the DS for the filename. 6578 */ 6579 static int 6580 nfsrv_pnfslookupds(struct vnode *vp, struct vnode *dvp, struct pnfsdsfile *pf, 6581 struct vnode **nvpp, NFSPROC_T *p) 6582 { 6583 struct nameidata named; 6584 struct ucred *tcred; 6585 char *bufp; 6586 u_long *hashp; 6587 struct vnode *nvp; 6588 int error; 6589 6590 tcred = newnfs_getcred(); 6591 named.ni_cnd.cn_nameiop = LOOKUP; 6592 named.ni_cnd.cn_lkflags = LK_SHARED | LK_RETRY; 6593 named.ni_cnd.cn_cred = tcred; 6594 named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF; 6595 nfsvno_setpathbuf(&named, &bufp, &hashp); 6596 named.ni_cnd.cn_nameptr = bufp; 6597 named.ni_cnd.cn_namelen = strlen(pf->dsf_filename); 6598 strlcpy(bufp, pf->dsf_filename, NAME_MAX); 6599 NFSD_DEBUG(4, "nfsrv_pnfslookupds: filename=%s\n", bufp); 6600 error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd); 6601 NFSD_DEBUG(4, "nfsrv_pnfslookupds: aft LOOKUP=%d\n", error); 6602 NFSFREECRED(tcred); 6603 nfsvno_relpathbuf(&named); 6604 if (error == 0) 6605 *nvpp = nvp; 6606 NFSD_DEBUG(4, "eo nfsrv_pnfslookupds=%d\n", error); 6607 return (error); 6608 } 6609 6610 /* 6611 * Set the file handle to the correct one. 6612 */ 6613 static void 6614 nfsrv_pnfssetfh(struct vnode *vp, struct pnfsdsfile *pf, char *devid, 6615 char *fnamep, struct vnode *nvp, NFSPROC_T *p) 6616 { 6617 struct nfsnode *np; 6618 int ret = 0; 6619 6620 np = VTONFS(nvp); 6621 NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH); 6622 /* 6623 * We can only do a vn_set_extattr() if the vnode is exclusively 6624 * locked and vn_start_write() has been done. If devid != NULL or 6625 * fnamep != NULL or the vnode is shared locked, vn_start_write() 6626 * may not have been done. 6627 * If not done now, it will be done on a future call. 6628 */ 6629 if (devid == NULL && fnamep == NULL && NFSVOPISLOCKED(vp) == 6630 LK_EXCLUSIVE) 6631 ret = vn_extattr_set(vp, IO_NODELOCKED, 6632 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf), 6633 (char *)pf, p); 6634 NFSD_DEBUG(4, "eo nfsrv_pnfssetfh=%d\n", ret); 6635 } 6636 6637 /* 6638 * Cause RPCs waiting on "nmp" to fail. This is called for a DS mount point 6639 * when the DS has failed. 6640 */ 6641 void 6642 nfsrv_killrpcs(struct nfsmount *nmp) 6643 { 6644 6645 /* 6646 * Call newnfs_nmcancelreqs() to cause 6647 * any RPCs in progress on the mount point to 6648 * fail. 6649 * This will cause any process waiting for an 6650 * RPC to complete while holding a vnode lock 6651 * on the mounted-on vnode (such as "df" or 6652 * a non-forced "umount") to fail. 6653 * This will unlock the mounted-on vnode so 6654 * a forced dismount can succeed. 6655 * The NFSMNTP_CANCELRPCS flag should be set when this function is 6656 * called. 6657 */ 6658 newnfs_nmcancelreqs(nmp); 6659 } 6660 6661 /* 6662 * Sum up the statfs info for each of the DSs, so that the client will 6663 * receive the total for all DSs. 6664 */ 6665 static int 6666 nfsrv_pnfsstatfs(struct statfs *sf, struct mount *mp) 6667 { 6668 struct statfs *tsf; 6669 struct nfsdevice *ds; 6670 struct vnode **dvpp, **tdvpp, *dvp; 6671 uint64_t tot; 6672 int cnt, error = 0, i; 6673 6674 if (nfsrv_devidcnt <= 0) 6675 return (ENXIO); 6676 dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK); 6677 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK); 6678 6679 /* Get an array of the dvps for the DSs. */ 6680 tdvpp = dvpp; 6681 i = 0; 6682 NFSDDSLOCK(); 6683 /* First, search for matches for same file system. */ 6684 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 6685 if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 && 6686 fsidcmp(&ds->nfsdev_mdsfsid, &mp->mnt_stat.f_fsid) == 0) { 6687 if (++i > nfsrv_devidcnt) 6688 break; 6689 *tdvpp++ = ds->nfsdev_dvp; 6690 } 6691 } 6692 /* 6693 * If no matches for same file system, total all servers not assigned 6694 * to a file system. 6695 */ 6696 if (i == 0) { 6697 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 6698 if (ds->nfsdev_nmp != NULL && 6699 ds->nfsdev_mdsisset == 0) { 6700 if (++i > nfsrv_devidcnt) 6701 break; 6702 *tdvpp++ = ds->nfsdev_dvp; 6703 } 6704 } 6705 } 6706 NFSDDSUNLOCK(); 6707 cnt = i; 6708 6709 /* Do a VFS_STATFS() for each of the DSs and sum them up. */ 6710 tdvpp = dvpp; 6711 for (i = 0; i < cnt && error == 0; i++) { 6712 dvp = *tdvpp++; 6713 error = VFS_STATFS(dvp->v_mount, tsf); 6714 if (error == 0) { 6715 if (sf->f_bsize == 0) { 6716 if (tsf->f_bsize > 0) 6717 sf->f_bsize = tsf->f_bsize; 6718 else 6719 sf->f_bsize = 8192; 6720 } 6721 if (tsf->f_blocks > 0) { 6722 if (sf->f_bsize != tsf->f_bsize) { 6723 tot = tsf->f_blocks * tsf->f_bsize; 6724 sf->f_blocks += (tot / sf->f_bsize); 6725 } else 6726 sf->f_blocks += tsf->f_blocks; 6727 } 6728 if (tsf->f_bfree > 0) { 6729 if (sf->f_bsize != tsf->f_bsize) { 6730 tot = tsf->f_bfree * tsf->f_bsize; 6731 sf->f_bfree += (tot / sf->f_bsize); 6732 } else 6733 sf->f_bfree += tsf->f_bfree; 6734 } 6735 if (tsf->f_bavail > 0) { 6736 if (sf->f_bsize != tsf->f_bsize) { 6737 tot = tsf->f_bavail * tsf->f_bsize; 6738 sf->f_bavail += (tot / sf->f_bsize); 6739 } else 6740 sf->f_bavail += tsf->f_bavail; 6741 } 6742 } 6743 } 6744 free(tsf, M_TEMP); 6745 free(dvpp, M_TEMP); 6746 return (error); 6747 } 6748 6749 /* 6750 * Set an NFSv4 acl. 6751 */ 6752 int 6753 nfsrv_setacl(struct vnode *vp, NFSACL_T *aclp, struct ucred *cred, NFSPROC_T *p) 6754 { 6755 int error; 6756 6757 if (nfsrv_useacl == 0 || nfs_supportsnfsv4acls(vp) == 0) { 6758 error = NFSERR_ATTRNOTSUPP; 6759 goto out; 6760 } 6761 /* 6762 * With NFSv4 ACLs, chmod(2) may need to add additional entries. 6763 * Make sure it has enough room for that - splitting every entry 6764 * into two and appending "canonical six" entries at the end. 6765 * Cribbed out of kern/vfs_acl.c - Rick M. 6766 */ 6767 if (aclp->acl_cnt > (ACL_MAX_ENTRIES - 6) / 2) { 6768 error = NFSERR_ATTRNOTSUPP; 6769 goto out; 6770 } 6771 error = VOP_SETACL(vp, ACL_TYPE_NFS4, aclp, cred, p); 6772 if (error == 0) { 6773 error = nfsrv_dssetacl(vp, aclp, cred, p); 6774 if (error == ENOENT) 6775 error = 0; 6776 } 6777 6778 out: 6779 NFSEXITCODE(error); 6780 return (error); 6781 } 6782 6783 /* 6784 * Seek vnode op call (actually it is a VOP_IOCTL()). 6785 * This function is called with the vnode locked, but unlocks and vrele()s 6786 * the vp before returning. 6787 */ 6788 int 6789 nfsvno_seek(struct nfsrv_descript *nd, struct vnode *vp, u_long cmd, 6790 off_t *offp, int content, bool *eofp, struct ucred *cred, NFSPROC_T *p) 6791 { 6792 struct nfsvattr at; 6793 int error, ret; 6794 6795 ASSERT_VOP_LOCKED(vp, "nfsvno_seek vp"); 6796 /* 6797 * Attempt to seek on a DS file. A return of ENOENT implies 6798 * there is no DS file to seek on. 6799 */ 6800 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SEEKDS, NULL, 6801 NULL, NULL, NULL, NULL, offp, content, eofp); 6802 if (error != ENOENT) { 6803 vput(vp); 6804 return (error); 6805 } 6806 6807 /* 6808 * Do the VOP_IOCTL() call. For the case where *offp == file_size, 6809 * VOP_IOCTL() will return ENXIO. However, the correct reply for 6810 * NFSv4.2 is *eofp == true and error == 0 for this case. 6811 */ 6812 NFSVOPUNLOCK(vp); 6813 error = VOP_IOCTL(vp, cmd, offp, 0, cred, p); 6814 *eofp = false; 6815 if (error == ENXIO || (error == 0 && cmd == FIOSEEKHOLE)) { 6816 /* Handle the cases where we might be at EOF. */ 6817 ret = nfsvno_getattr(vp, &at, nd, p, 0, NULL); 6818 if (ret == 0 && *offp == at.na_size) { 6819 *eofp = true; 6820 error = 0; 6821 } 6822 if (ret != 0 && error == 0) 6823 error = ret; 6824 } 6825 vrele(vp); 6826 NFSEXITCODE(error); 6827 return (error); 6828 } 6829 6830 /* 6831 * Allocate vnode op call. 6832 */ 6833 int 6834 nfsvno_allocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred, 6835 NFSPROC_T *p) 6836 { 6837 int error; 6838 off_t olen; 6839 6840 ASSERT_VOP_ELOCKED(vp, "nfsvno_allocate vp"); 6841 /* 6842 * Attempt to allocate on a DS file. A return of ENOENT implies 6843 * there is no DS file to allocate on. 6844 */ 6845 error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_ALLOCATE, NULL, 6846 NULL, NULL, NULL, NULL, &len, 0, NULL); 6847 if (error != ENOENT) 6848 return (error); 6849 6850 /* 6851 * Do the actual VOP_ALLOCATE(), looping so long as 6852 * progress is being made, to achieve completion. 6853 */ 6854 do { 6855 olen = len; 6856 error = VOP_ALLOCATE(vp, &off, &len, IO_SYNC, cred); 6857 if (error == 0 && len > 0 && olen > len) 6858 maybe_yield(); 6859 } while (error == 0 && len > 0 && olen > len); 6860 if (error == 0 && len > 0) 6861 error = NFSERR_IO; 6862 NFSEXITCODE(error); 6863 return (error); 6864 } 6865 6866 /* 6867 * Deallocate vnode op call. 6868 */ 6869 int 6870 nfsvno_deallocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred, 6871 NFSPROC_T *p) 6872 { 6873 int error; 6874 off_t olen; 6875 6876 ASSERT_VOP_ELOCKED(vp, "nfsvno_deallocate vp"); 6877 /* 6878 * Attempt to deallocate on a DS file. A return of ENOENT implies 6879 * there is no DS file to deallocate on. 6880 */ 6881 error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_DEALLOCATE, NULL, 6882 NULL, NULL, NULL, NULL, &len, 0, NULL); 6883 if (error != ENOENT) 6884 return (error); 6885 6886 /* 6887 * Do the actual VOP_DEALLOCATE(), looping so long as 6888 * progress is being made, to achieve completion. 6889 */ 6890 do { 6891 olen = len; 6892 error = VOP_DEALLOCATE(vp, &off, &len, 0, IO_SYNC, cred); 6893 if (error == 0 && len > 0 && olen > len) 6894 maybe_yield(); 6895 } while (error == 0 && len > 0 && olen > len); 6896 if (error == 0 && len > 0) 6897 error = NFSERR_IO; 6898 NFSEXITCODE(error); 6899 return (error); 6900 } 6901 6902 /* 6903 * Get Extended Atribute vnode op into an mbuf list. 6904 */ 6905 int 6906 nfsvno_getxattr(struct vnode *vp, char *name, uint32_t maxresp, 6907 struct ucred *cred, uint64_t flag, int maxextsiz, struct thread *p, 6908 struct mbuf **mpp, struct mbuf **mpendp, int *lenp) 6909 { 6910 struct iovec *iv; 6911 struct uio io, *uiop = &io; 6912 struct mbuf *m, *m2; 6913 int alen, error, len, tlen; 6914 size_t siz; 6915 6916 /* First, find out the size of the extended attribute. */ 6917 error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL, 6918 &siz, cred, p); 6919 if (error != 0) 6920 return (NFSERR_NOXATTR); 6921 if (siz > maxresp - NFS_MAXXDR) 6922 return (NFSERR_XATTR2BIG); 6923 len = siz; 6924 tlen = NFSM_RNDUP(len); 6925 if (tlen > 0) { 6926 /* 6927 * If cnt > MCLBYTES and the reply will not be saved, use 6928 * ext_pgs mbufs for TLS. 6929 * For NFSv4.0, we do not know for sure if the reply will 6930 * be saved, so do not use ext_pgs mbufs for NFSv4.0. 6931 * Always use ext_pgs mbufs if ND_EXTPG is set. 6932 */ 6933 if ((flag & ND_EXTPG) != 0 || (tlen > MCLBYTES && 6934 (flag & (ND_TLS | ND_SAVEREPLY)) == ND_TLS && 6935 (flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)) 6936 uiop->uio_iovcnt = nfsrv_createiovec_extpgs(tlen, 6937 maxextsiz, &m, &m2, &iv); 6938 else 6939 uiop->uio_iovcnt = nfsrv_createiovec(tlen, &m, &m2, 6940 &iv); 6941 uiop->uio_iov = iv; 6942 } else { 6943 uiop->uio_iovcnt = 0; 6944 uiop->uio_iov = iv = NULL; 6945 m = m2 = NULL; 6946 } 6947 uiop->uio_offset = 0; 6948 uiop->uio_resid = tlen; 6949 uiop->uio_rw = UIO_READ; 6950 uiop->uio_segflg = UIO_SYSSPACE; 6951 uiop->uio_td = p; 6952 #ifdef MAC 6953 error = mac_vnode_check_getextattr(cred, vp, EXTATTR_NAMESPACE_USER, 6954 name); 6955 if (error != 0) 6956 goto out; 6957 #endif 6958 6959 if (tlen > 0) 6960 error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop, 6961 NULL, cred, p); 6962 if (error != 0) 6963 goto out; 6964 if (uiop->uio_resid > 0) { 6965 alen = tlen; 6966 len = tlen - uiop->uio_resid; 6967 tlen = NFSM_RNDUP(len); 6968 if (alen != tlen) 6969 printf("nfsvno_getxattr: weird size read\n"); 6970 if (tlen == 0) { 6971 m_freem(m); 6972 m = m2 = NULL; 6973 } else if (alen != tlen || tlen != len) 6974 m2 = nfsrv_adj(m, alen - tlen, tlen - len); 6975 } 6976 *lenp = len; 6977 *mpp = m; 6978 *mpendp = m2; 6979 6980 out: 6981 if (error != 0) { 6982 if (m != NULL) 6983 m_freem(m); 6984 *lenp = 0; 6985 } 6986 free(iv, M_TEMP); 6987 NFSEXITCODE(error); 6988 return (error); 6989 } 6990 6991 /* 6992 * Set Extended attribute vnode op from an mbuf list. 6993 */ 6994 int 6995 nfsvno_setxattr(struct vnode *vp, char *name, int len, struct mbuf *m, 6996 char *cp, struct ucred *cred, struct thread *p) 6997 { 6998 struct iovec *iv; 6999 struct uio uio, *uiop = &uio; 7000 int cnt, error; 7001 7002 error = 0; 7003 #ifdef MAC 7004 error = mac_vnode_check_setextattr(cred, vp, EXTATTR_NAMESPACE_USER, 7005 name); 7006 #endif 7007 if (error != 0) 7008 goto out; 7009 7010 uiop->uio_rw = UIO_WRITE; 7011 uiop->uio_segflg = UIO_SYSSPACE; 7012 uiop->uio_td = p; 7013 uiop->uio_offset = 0; 7014 uiop->uio_resid = len; 7015 if (len > 0) { 7016 error = nfsrv_createiovecw(len, m, cp, &iv, &cnt); 7017 uiop->uio_iov = iv; 7018 uiop->uio_iovcnt = cnt; 7019 } else { 7020 uiop->uio_iov = iv = NULL; 7021 uiop->uio_iovcnt = 0; 7022 } 7023 if (error == 0) { 7024 error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop, 7025 cred, p); 7026 if (error == 0) { 7027 if (vp->v_type == VREG && nfsrv_devidcnt != 0) 7028 nfsvno_updateds(vp, cred, p); 7029 error = VOP_FSYNC(vp, MNT_WAIT, p); 7030 } 7031 free(iv, M_TEMP); 7032 } 7033 7034 out: 7035 NFSEXITCODE(error); 7036 return (error); 7037 } 7038 7039 /* 7040 * For a pNFS server, the DS file's ctime and 7041 * va_filerev (TimeMetadata and Change) needs to 7042 * be updated. This is a hack, but works by 7043 * flipping the S_ISGID bit in va_mode and then 7044 * flipping it back. 7045 * It does result in two MDS->DS RPCs, but creating 7046 * a custom RPC just to do this seems overkill, since 7047 * Setxattr/Rmxattr will not be done that frequently. 7048 * If it fails part way through, that is not too 7049 * serious, since the DS file is never executed. 7050 */ 7051 static void 7052 nfsvno_updateds(struct vnode *vp, struct ucred *cred, NFSPROC_T *p) 7053 { 7054 struct nfsvattr nva; 7055 int ret; 7056 u_short tmode; 7057 7058 ret = VOP_GETATTR(vp, &nva.na_vattr, cred); 7059 if (ret == 0) { 7060 tmode = nva.na_mode; 7061 NFSVNO_ATTRINIT(&nva); 7062 tmode ^= S_ISGID; 7063 NFSVNO_SETATTRVAL(&nva, mode, tmode); 7064 ret = nfsrv_proxyds(vp, 0, 0, cred, p, 7065 NFSPROC_SETATTR, NULL, NULL, NULL, &nva, 7066 NULL, NULL, 0, NULL); 7067 if (ret == 0) { 7068 tmode ^= S_ISGID; 7069 NFSVNO_SETATTRVAL(&nva, mode, tmode); 7070 ret = nfsrv_proxyds(vp, 0, 0, cred, p, 7071 NFSPROC_SETATTR, NULL, NULL, NULL, 7072 &nva, NULL, NULL, 0, NULL); 7073 } 7074 } 7075 } 7076 7077 /* 7078 * Remove Extended attribute vnode op. 7079 */ 7080 int 7081 nfsvno_rmxattr(struct nfsrv_descript *nd, struct vnode *vp, char *name, 7082 struct ucred *cred, struct thread *p) 7083 { 7084 int error; 7085 7086 /* 7087 * Get rid of any delegations. I am not sure why this is required, 7088 * but RFC-8276 says so. 7089 */ 7090 error = nfsrv_checkremove(vp, 0, nd, nd->nd_clientid, p); 7091 if (error != 0) 7092 goto out; 7093 #ifdef MAC 7094 error = mac_vnode_check_deleteextattr(cred, vp, EXTATTR_NAMESPACE_USER, 7095 name); 7096 if (error != 0) 7097 goto out; 7098 #endif 7099 7100 error = VOP_DELETEEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, cred, p); 7101 if (error == EOPNOTSUPP) 7102 error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL, 7103 cred, p); 7104 if (error == 0) { 7105 if (vp->v_type == VREG && nfsrv_devidcnt != 0) 7106 nfsvno_updateds(vp, cred, p); 7107 error = VOP_FSYNC(vp, MNT_WAIT, p); 7108 } 7109 out: 7110 NFSEXITCODE(error); 7111 return (error); 7112 } 7113 7114 /* 7115 * List Extended Atribute vnode op into an mbuf list. 7116 */ 7117 int 7118 nfsvno_listxattr(struct vnode *vp, uint64_t cookie, struct ucred *cred, 7119 struct thread *p, u_char **bufp, uint32_t *lenp, bool *eofp) 7120 { 7121 struct iovec iv; 7122 struct uio io; 7123 int error; 7124 size_t siz; 7125 7126 *bufp = NULL; 7127 /* First, find out the size of the extended attribute. */ 7128 error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, NULL, &siz, cred, 7129 p); 7130 if (error != 0) 7131 return (NFSERR_NOXATTR); 7132 if (siz <= cookie) { 7133 *lenp = 0; 7134 *eofp = true; 7135 goto out; 7136 } 7137 if (siz > cookie + *lenp) { 7138 siz = cookie + *lenp; 7139 *eofp = false; 7140 } else 7141 *eofp = true; 7142 /* Just choose a sanity limit of 10Mbytes for malloc(M_TEMP). */ 7143 if (siz > 10 * 1024 * 1024) { 7144 error = NFSERR_XATTR2BIG; 7145 goto out; 7146 } 7147 *bufp = malloc(siz, M_TEMP, M_WAITOK); 7148 iv.iov_base = *bufp; 7149 iv.iov_len = siz; 7150 io.uio_iovcnt = 1; 7151 io.uio_iov = &iv; 7152 io.uio_offset = 0; 7153 io.uio_resid = siz; 7154 io.uio_rw = UIO_READ; 7155 io.uio_segflg = UIO_SYSSPACE; 7156 io.uio_td = p; 7157 #ifdef MAC 7158 error = mac_vnode_check_listextattr(cred, vp, EXTATTR_NAMESPACE_USER); 7159 if (error != 0) 7160 goto out; 7161 #endif 7162 7163 error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, &io, NULL, cred, 7164 p); 7165 if (error != 0) 7166 goto out; 7167 if (io.uio_resid > 0) 7168 siz -= io.uio_resid; 7169 *lenp = siz; 7170 7171 out: 7172 if (error != 0) { 7173 free(*bufp, M_TEMP); 7174 *bufp = NULL; 7175 } 7176 NFSEXITCODE(error); 7177 return (error); 7178 } 7179 7180 /* 7181 * Trim trailing data off the mbuf list being built. 7182 */ 7183 void 7184 nfsm_trimtrailing(struct nfsrv_descript *nd, struct mbuf *mb, char *bpos, 7185 int bextpg, int bextpgsiz) 7186 { 7187 vm_page_t pg; 7188 int fullpgsiz, i; 7189 7190 if (mb->m_next != NULL) { 7191 m_freem(mb->m_next); 7192 mb->m_next = NULL; 7193 } 7194 if ((mb->m_flags & M_EXTPG) != 0) { 7195 KASSERT(bextpg >= 0 && bextpg < mb->m_epg_npgs, 7196 ("nfsm_trimtrailing: bextpg out of range")); 7197 KASSERT(bpos == (char *)(void *) 7198 PHYS_TO_DMAP(mb->m_epg_pa[bextpg]) + PAGE_SIZE - bextpgsiz, 7199 ("nfsm_trimtrailing: bextpgsiz bad!")); 7200 7201 /* First, get rid of any pages after this position. */ 7202 for (i = mb->m_epg_npgs - 1; i > bextpg; i--) { 7203 pg = PHYS_TO_VM_PAGE(mb->m_epg_pa[i]); 7204 vm_page_unwire_noq(pg); 7205 vm_page_free(pg); 7206 } 7207 mb->m_epg_npgs = bextpg + 1; 7208 if (bextpg == 0) 7209 fullpgsiz = PAGE_SIZE - mb->m_epg_1st_off; 7210 else 7211 fullpgsiz = PAGE_SIZE; 7212 mb->m_epg_last_len = fullpgsiz - bextpgsiz; 7213 mb->m_len = m_epg_pagelen(mb, 0, mb->m_epg_1st_off); 7214 for (i = 1; i < mb->m_epg_npgs; i++) 7215 mb->m_len += m_epg_pagelen(mb, i, 0); 7216 nd->nd_bextpgsiz = bextpgsiz; 7217 nd->nd_bextpg = bextpg; 7218 } else 7219 mb->m_len = bpos - mtod(mb, char *); 7220 nd->nd_mb = mb; 7221 nd->nd_bpos = bpos; 7222 } 7223 7224 7225 /* 7226 * Check to see if a put file handle operation should test for 7227 * NFSERR_WRONGSEC, although NFSv3 actually returns NFSERR_AUTHERR. 7228 * When Open is the next operation, NFSERR_WRONGSEC cannot be 7229 * replied for the Open cases that use a component. This can 7230 * be identified by the fact that the file handle's type is VDIR. 7231 */ 7232 bool 7233 nfsrv_checkwrongsec(struct nfsrv_descript *nd, int nextop, __enum_uint8(vtype) vtyp) 7234 { 7235 7236 if ((nd->nd_flag & ND_NFSV4) == 0) 7237 return (true); 7238 7239 if ((nd->nd_flag & ND_LASTOP) != 0) 7240 return (false); 7241 7242 if (nextop == NFSV4OP_PUTROOTFH || nextop == NFSV4OP_PUTFH || 7243 nextop == NFSV4OP_PUTPUBFH || nextop == NFSV4OP_RESTOREFH || 7244 nextop == NFSV4OP_LOOKUP || nextop == NFSV4OP_LOOKUPP || 7245 nextop == NFSV4OP_SECINFO || nextop == NFSV4OP_SECINFONONAME) 7246 return (false); 7247 if (nextop == NFSV4OP_OPEN && vtyp == VDIR) 7248 return (false); 7249 return (true); 7250 } 7251 7252 /* 7253 * Check DSs marked no space. 7254 */ 7255 void 7256 nfsrv_checknospc(void) 7257 { 7258 struct statfs *tsf; 7259 struct nfsdevice *ds; 7260 struct vnode **dvpp, **tdvpp, *dvp; 7261 char *devid, *tdevid; 7262 int cnt, error = 0, i; 7263 7264 if (nfsrv_devidcnt <= 0) 7265 return; 7266 dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK); 7267 devid = malloc(nfsrv_devidcnt * NFSX_V4DEVICEID, M_TEMP, M_WAITOK); 7268 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK); 7269 7270 /* Get an array of the dvps for the DSs. */ 7271 tdvpp = dvpp; 7272 tdevid = devid; 7273 i = 0; 7274 NFSDDSLOCK(); 7275 /* First, search for matches for same file system. */ 7276 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 7277 if (ds->nfsdev_nmp != NULL && ds->nfsdev_nospc) { 7278 if (++i > nfsrv_devidcnt) 7279 break; 7280 *tdvpp++ = ds->nfsdev_dvp; 7281 NFSBCOPY(ds->nfsdev_deviceid, tdevid, NFSX_V4DEVICEID); 7282 tdevid += NFSX_V4DEVICEID; 7283 } 7284 } 7285 NFSDDSUNLOCK(); 7286 7287 /* Do a VFS_STATFS() for each of the DSs and clear no space. */ 7288 cnt = i; 7289 tdvpp = dvpp; 7290 tdevid = devid; 7291 for (i = 0; i < cnt && error == 0; i++) { 7292 dvp = *tdvpp++; 7293 error = VFS_STATFS(dvp->v_mount, tsf); 7294 if (error == 0 && tsf->f_bavail > 0) { 7295 NFSD_DEBUG(1, "nfsrv_checknospc: reset nospc\n"); 7296 nfsrv_marknospc(tdevid, false); 7297 } 7298 tdevid += NFSX_V4DEVICEID; 7299 } 7300 free(tsf, M_TEMP); 7301 free(dvpp, M_TEMP); 7302 free(devid, M_TEMP); 7303 } 7304 7305 /* 7306 * Initialize everything that needs to be initialized for a vnet. 7307 */ 7308 static void 7309 nfsrv_vnetinit(const void *unused __unused) 7310 { 7311 7312 nfsd_mntinit(); 7313 } 7314 VNET_SYSINIT(nfsrv_vnetinit, SI_SUB_VNET_DONE, SI_ORDER_ANY, 7315 nfsrv_vnetinit, NULL); 7316 7317 /* 7318 * Clean up everything that is in a vnet and needs to be 7319 * done when the jail is destroyed or the module unloaded. 7320 */ 7321 static void 7322 nfsrv_cleanup(const void *unused __unused) 7323 { 7324 int i; 7325 7326 NFSD_LOCK(); 7327 if (!NFSD_VNET(nfsrv_mntinited)) { 7328 NFSD_UNLOCK(); 7329 return; 7330 } 7331 NFSD_VNET(nfsrv_mntinited) = false; 7332 NFSD_UNLOCK(); 7333 7334 /* Clean out all NFSv4 state. */ 7335 nfsrv_throwawayallstate(curthread); 7336 7337 /* Clean the NFS server reply cache */ 7338 nfsrvd_cleancache(); 7339 7340 /* Clean out v4root exports. */ 7341 if (NFSD_VNET(nfsv4root_mnt)->mnt_export != NULL) { 7342 vfs_free_addrlist(NFSD_VNET(nfsv4root_mnt)->mnt_export); 7343 free(NFSD_VNET(nfsv4root_mnt)->mnt_export, M_MOUNT); 7344 NFSD_VNET(nfsv4root_mnt)->mnt_export = NULL; 7345 } 7346 7347 /* Free up the krpc server pool. */ 7348 if (NFSD_VNET(nfsrvd_pool) != NULL) 7349 svcpool_destroy(NFSD_VNET(nfsrvd_pool)); 7350 7351 /* and get rid of the locks */ 7352 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { 7353 mtx_destroy(&NFSD_VNET(nfsrchash_table)[i].mtx); 7354 mtx_destroy(&NFSD_VNET(nfsrcahash_table)[i].mtx); 7355 } 7356 mtx_destroy(&NFSD_VNET(nfsv4root_mnt)->mnt_mtx); 7357 for (i = 0; i < nfsrv_sessionhashsize; i++) 7358 mtx_destroy(&NFSD_VNET(nfssessionhash)[i].mtx); 7359 lockdestroy(&NFSD_VNET(nfsv4root_mnt)->mnt_explock); 7360 free(NFSD_VNET(nfsrvudphashtbl), M_NFSRVCACHE); 7361 free(NFSD_VNET(nfsrchash_table), M_NFSRVCACHE); 7362 free(NFSD_VNET(nfsrcahash_table), M_NFSRVCACHE); 7363 free(NFSD_VNET(nfsclienthash), M_NFSDCLIENT); 7364 free(NFSD_VNET(nfslockhash), M_NFSDLOCKFILE); 7365 free(NFSD_VNET(nfssessionhash), M_NFSDSESSION); 7366 free(NFSD_VNET(nfsv4root_mnt), M_TEMP); 7367 NFSD_VNET(nfsv4root_mnt) = NULL; 7368 } 7369 VNET_SYSUNINIT(nfsrv_cleanup, SI_SUB_VNET_DONE, SI_ORDER_ANY, 7370 nfsrv_cleanup, NULL); 7371 7372 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *); 7373 7374 /* 7375 * Called once to initialize data structures... 7376 */ 7377 static int 7378 nfsd_modevent(module_t mod, int type, void *data) 7379 { 7380 int error = 0, i; 7381 static int loaded = 0; 7382 7383 switch (type) { 7384 case MOD_LOAD: 7385 if (loaded) 7386 goto out; 7387 newnfs_portinit(); 7388 mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF); 7389 mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF); 7390 mtx_init(&nfsrv_dontlistlock_mtx, "nfs4dnl", NULL, MTX_DEF); 7391 mtx_init(&nfsrv_recalllock_mtx, "nfs4rec", NULL, MTX_DEF); 7392 #ifdef VV_DISABLEDELEG 7393 vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation; 7394 vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation; 7395 #endif 7396 nfsd_call_nfsd = nfssvc_nfsd; 7397 loaded = 1; 7398 break; 7399 7400 case MOD_UNLOAD: 7401 if (newnfs_numnfsd != 0) { 7402 error = EBUSY; 7403 break; 7404 } 7405 7406 #ifdef VV_DISABLEDELEG 7407 vn_deleg_ops.vndeleg_recall = NULL; 7408 vn_deleg_ops.vndeleg_disable = NULL; 7409 #endif 7410 nfsd_call_nfsd = NULL; 7411 mtx_destroy(&nfsrc_udpmtx); 7412 mtx_destroy(&nfs_v4root_mutex); 7413 mtx_destroy(&nfsrv_dontlistlock_mtx); 7414 mtx_destroy(&nfsrv_recalllock_mtx); 7415 if (nfslayouthash != NULL) { 7416 for (i = 0; i < nfsrv_layouthashsize; i++) 7417 mtx_destroy(&nfslayouthash[i].mtx); 7418 free(nfslayouthash, M_NFSDSESSION); 7419 } 7420 loaded = 0; 7421 break; 7422 default: 7423 error = EOPNOTSUPP; 7424 break; 7425 } 7426 7427 out: 7428 NFSEXITCODE(error); 7429 return (error); 7430 } 7431 static moduledata_t nfsd_mod = { 7432 "nfsd", 7433 nfsd_modevent, 7434 NULL, 7435 }; 7436 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY); 7437 7438 /* So that loader and kldload(2) can find us, wherever we are.. */ 7439 MODULE_VERSION(nfsd, 1); 7440 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1); 7441 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1); 7442 MODULE_DEPEND(nfsd, krpc, 1, 1, 1); 7443 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1); 7444