nfs_clport.c revision 1.1.1.1.12.1
1 /* $NetBSD: nfs_clport.c,v 1.1.1.1.12.1 2016/12/05 10:55:25 skrll Exp $ */ 2 /*- 3 * Copyright (c) 1989, 1993 4 * The Regents of the University of California. All rights reserved. 5 * 6 * This code is derived from software contributed to Berkeley by 7 * Rick Macklem at The University of Guelph. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 */ 34 35 #include <sys/cdefs.h> 36 /* __FBSDID("FreeBSD: head/sys/fs/nfsclient/nfs_clport.c 299413 2016-05-11 06:35:46Z kib "); */ 37 __RCSID("$NetBSD: nfs_clport.c,v 1.1.1.1.12.1 2016/12/05 10:55:25 skrll Exp $"); 38 39 #include "opt_inet.h" 40 #include "opt_inet6.h" 41 42 #include <sys/capsicum.h> 43 44 /* 45 * generally, I don't like #includes inside .h files, but it seems to 46 * be the easiest way to handle the port. 47 */ 48 #include <sys/fail.h> 49 #include <sys/hash.h> 50 #include <sys/sysctl.h> 51 #include <fs/nfs/nfsport.h> 52 #include <netinet/in_fib.h> 53 #include <netinet/if_ether.h> 54 #include <netinet6/ip6_var.h> 55 #include <net/if_types.h> 56 57 #include <fs/nfsclient/nfs_kdtrace.h> 58 59 #ifdef KDTRACE_HOOKS 60 dtrace_nfsclient_attrcache_flush_probe_func_t 61 dtrace_nfscl_attrcache_flush_done_probe; 62 uint32_t nfscl_attrcache_flush_done_id; 63 64 dtrace_nfsclient_attrcache_get_hit_probe_func_t 65 dtrace_nfscl_attrcache_get_hit_probe; 66 uint32_t nfscl_attrcache_get_hit_id; 67 68 dtrace_nfsclient_attrcache_get_miss_probe_func_t 69 dtrace_nfscl_attrcache_get_miss_probe; 70 uint32_t nfscl_attrcache_get_miss_id; 71 72 dtrace_nfsclient_attrcache_load_probe_func_t 73 dtrace_nfscl_attrcache_load_done_probe; 74 uint32_t nfscl_attrcache_load_done_id; 75 #endif /* !KDTRACE_HOOKS */ 76 77 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1; 78 extern struct vop_vector newnfs_vnodeops; 79 extern struct vop_vector newnfs_fifoops; 80 extern uma_zone_t newnfsnode_zone; 81 extern struct buf_ops buf_ops_newnfs; 82 extern int ncl_pbuf_freecnt; 83 extern short nfsv4_cbport; 84 extern int nfscl_enablecallb; 85 extern int nfs_numnfscbd; 86 extern int nfscl_inited; 87 struct mtx nfs_clstate_mutex; 88 struct mtx ncl_iod_mutex; 89 NFSDLOCKMUTEX; 90 91 extern void (*ncl_call_invalcaches)(struct vnode *); 92 93 SYSCTL_DECL(_vfs_nfs); 94 static int ncl_fileid_maxwarnings = 10; 95 SYSCTL_INT(_vfs_nfs, OID_AUTO, fileid_maxwarnings, CTLFLAG_RWTUN, 96 &ncl_fileid_maxwarnings, 0, 97 "Limit fileid corruption warnings; 0 is off; -1 is unlimited"); 98 static volatile int ncl_fileid_nwarnings; 99 100 static void nfscl_warn_fileid(struct nfsmount *, struct nfsvattr *, 101 struct nfsvattr *); 102 103 /* 104 * Comparison function for vfs_hash functions. 105 */ 106 int 107 newnfs_vncmpf(struct vnode *vp, void *arg) 108 { 109 struct nfsfh *nfhp = (struct nfsfh *)arg; 110 struct nfsnode *np = VTONFS(vp); 111 112 if (np->n_fhp->nfh_len != nfhp->nfh_len || 113 NFSBCMP(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len)) 114 return (1); 115 return (0); 116 } 117 118 /* 119 * Look up a vnode/nfsnode by file handle. 120 * Callers must check for mount points!! 121 * In all cases, a pointer to a 122 * nfsnode structure is returned. 123 * This variant takes a "struct nfsfh *" as second argument and uses 124 * that structure up, either by hanging off the nfsnode or FREEing it. 125 */ 126 int 127 nfscl_nget(struct mount *mntp, struct vnode *dvp, struct nfsfh *nfhp, 128 struct componentname *cnp, struct thread *td, struct nfsnode **npp, 129 void *stuff, int lkflags) 130 { 131 struct nfsnode *np, *dnp; 132 struct vnode *vp, *nvp; 133 struct nfsv4node *newd, *oldd; 134 int error; 135 u_int hash; 136 struct nfsmount *nmp; 137 138 nmp = VFSTONFS(mntp); 139 dnp = VTONFS(dvp); 140 *npp = NULL; 141 142 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT); 143 144 error = vfs_hash_get(mntp, hash, lkflags, 145 td, &nvp, newnfs_vncmpf, nfhp); 146 if (error == 0 && nvp != NULL) { 147 /* 148 * I believe there is a slight chance that vgonel() could 149 * get called on this vnode between when NFSVOPLOCK() drops 150 * the VI_LOCK() and vget() acquires it again, so that it 151 * hasn't yet had v_usecount incremented. If this were to 152 * happen, the VI_DOOMED flag would be set, so check for 153 * that here. Since we now have the v_usecount incremented, 154 * we should be ok until we vrele() it, if the VI_DOOMED 155 * flag isn't set now. 156 */ 157 VI_LOCK(nvp); 158 if ((nvp->v_iflag & VI_DOOMED)) { 159 VI_UNLOCK(nvp); 160 vrele(nvp); 161 error = ENOENT; 162 } else { 163 VI_UNLOCK(nvp); 164 } 165 } 166 if (error) { 167 FREE((caddr_t)nfhp, M_NFSFH); 168 return (error); 169 } 170 if (nvp != NULL) { 171 np = VTONFS(nvp); 172 /* 173 * For NFSv4, check to see if it is the same name and 174 * replace the name, if it is different. 175 */ 176 oldd = newd = NULL; 177 if ((nmp->nm_flag & NFSMNT_NFSV4) && np->n_v4 != NULL && 178 nvp->v_type == VREG && 179 (np->n_v4->n4_namelen != cnp->cn_namelen || 180 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4), 181 cnp->cn_namelen) || 182 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || 183 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 184 dnp->n_fhp->nfh_len))) { 185 MALLOC(newd, struct nfsv4node *, 186 sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len + 187 + cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK); 188 NFSLOCKNODE(np); 189 if (newd != NULL && np->n_v4 != NULL && nvp->v_type == VREG 190 && (np->n_v4->n4_namelen != cnp->cn_namelen || 191 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4), 192 cnp->cn_namelen) || 193 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || 194 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 195 dnp->n_fhp->nfh_len))) { 196 oldd = np->n_v4; 197 np->n_v4 = newd; 198 newd = NULL; 199 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; 200 np->n_v4->n4_namelen = cnp->cn_namelen; 201 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 202 dnp->n_fhp->nfh_len); 203 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4), 204 cnp->cn_namelen); 205 } 206 NFSUNLOCKNODE(np); 207 } 208 if (newd != NULL) 209 FREE((caddr_t)newd, M_NFSV4NODE); 210 if (oldd != NULL) 211 FREE((caddr_t)oldd, M_NFSV4NODE); 212 *npp = np; 213 FREE((caddr_t)nfhp, M_NFSFH); 214 return (0); 215 } 216 np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO); 217 218 error = getnewvnode(nfs_vnode_tag, mntp, &newnfs_vnodeops, &nvp); 219 if (error) { 220 uma_zfree(newnfsnode_zone, np); 221 FREE((caddr_t)nfhp, M_NFSFH); 222 return (error); 223 } 224 vp = nvp; 225 KASSERT(vp->v_bufobj.bo_bsize != 0, ("nfscl_nget: bo_bsize == 0")); 226 vp->v_bufobj.bo_ops = &buf_ops_newnfs; 227 vp->v_data = np; 228 np->n_vnode = vp; 229 /* 230 * Initialize the mutex even if the vnode is going to be a loser. 231 * This simplifies the logic in reclaim, which can then unconditionally 232 * destroy the mutex (in the case of the loser, or if hash_insert 233 * happened to return an error no special casing is needed). 234 */ 235 mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK); 236 237 /* 238 * Are we getting the root? If so, make sure the vnode flags 239 * are correct 240 */ 241 if ((nfhp->nfh_len == nmp->nm_fhsize) && 242 !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len)) { 243 if (vp->v_type == VNON) 244 vp->v_type = VDIR; 245 vp->v_vflag |= VV_ROOT; 246 } 247 248 np->n_fhp = nfhp; 249 /* 250 * For NFSv4, we have to attach the directory file handle and 251 * file name, so that Open Ops can be done later. 252 */ 253 if (nmp->nm_flag & NFSMNT_NFSV4) { 254 MALLOC(np->n_v4, struct nfsv4node *, sizeof (struct nfsv4node) 255 + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE, 256 M_WAITOK); 257 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; 258 np->n_v4->n4_namelen = cnp->cn_namelen; 259 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 260 dnp->n_fhp->nfh_len); 261 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4), 262 cnp->cn_namelen); 263 } else { 264 np->n_v4 = NULL; 265 } 266 267 /* 268 * NFS supports recursive and shared locking. 269 */ 270 lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL); 271 VN_LOCK_AREC(vp); 272 VN_LOCK_ASHARE(vp); 273 error = insmntque(vp, mntp); 274 if (error != 0) { 275 *npp = NULL; 276 mtx_destroy(&np->n_mtx); 277 FREE((caddr_t)nfhp, M_NFSFH); 278 if (np->n_v4 != NULL) 279 FREE((caddr_t)np->n_v4, M_NFSV4NODE); 280 uma_zfree(newnfsnode_zone, np); 281 return (error); 282 } 283 error = vfs_hash_insert(vp, hash, lkflags, 284 td, &nvp, newnfs_vncmpf, nfhp); 285 if (error) 286 return (error); 287 if (nvp != NULL) { 288 *npp = VTONFS(nvp); 289 /* vfs_hash_insert() vput()'s the losing vnode */ 290 return (0); 291 } 292 *npp = np; 293 294 return (0); 295 } 296 297 /* 298 * Another variant of nfs_nget(). This one is only used by reopen. It 299 * takes almost the same args as nfs_nget(), but only succeeds if an entry 300 * exists in the cache. (Since files should already be "open" with a 301 * vnode ref cnt on the node when reopen calls this, it should always 302 * succeed.) 303 * Also, don't get a vnode lock, since it may already be locked by some 304 * other process that is handling it. This is ok, since all other threads 305 * on the client are blocked by the nfsc_lock being exclusively held by the 306 * caller of this function. 307 */ 308 int 309 nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize, 310 struct thread *td, struct nfsnode **npp) 311 { 312 struct vnode *nvp; 313 u_int hash; 314 struct nfsfh *nfhp; 315 int error; 316 317 *npp = NULL; 318 /* For forced dismounts, just return error. */ 319 if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF)) 320 return (EINTR); 321 MALLOC(nfhp, struct nfsfh *, sizeof (struct nfsfh) + fhsize, 322 M_NFSFH, M_WAITOK); 323 bcopy(fhp, &nfhp->nfh_fh[0], fhsize); 324 nfhp->nfh_len = fhsize; 325 326 hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT); 327 328 /* 329 * First, try to get the vnode locked, but don't block for the lock. 330 */ 331 error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp, 332 newnfs_vncmpf, nfhp); 333 if (error == 0 && nvp != NULL) { 334 NFSVOPUNLOCK(nvp, 0); 335 } else if (error == EBUSY) { 336 /* 337 * It is safe so long as a vflush() with 338 * FORCECLOSE has not been done. Since the Renew thread is 339 * stopped and the MNTK_UNMOUNTF flag is set before doing 340 * a vflush() with FORCECLOSE, we should be ok here. 341 */ 342 if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF)) 343 error = EINTR; 344 else { 345 vfs_hash_ref(mntp, hash, td, &nvp, newnfs_vncmpf, nfhp); 346 if (nvp == NULL) { 347 error = ENOENT; 348 } else if ((nvp->v_iflag & VI_DOOMED) != 0) { 349 error = ENOENT; 350 vrele(nvp); 351 } else { 352 error = 0; 353 } 354 } 355 } 356 FREE(nfhp, M_NFSFH); 357 if (error) 358 return (error); 359 if (nvp != NULL) { 360 *npp = VTONFS(nvp); 361 return (0); 362 } 363 return (EINVAL); 364 } 365 366 static void 367 nfscl_warn_fileid(struct nfsmount *nmp, struct nfsvattr *oldnap, 368 struct nfsvattr *newnap) 369 { 370 int off; 371 372 if (ncl_fileid_maxwarnings >= 0 && 373 ncl_fileid_nwarnings >= ncl_fileid_maxwarnings) 374 return; 375 off = 0; 376 if (ncl_fileid_maxwarnings >= 0) { 377 if (++ncl_fileid_nwarnings >= ncl_fileid_maxwarnings) 378 off = 1; 379 } 380 381 printf("newnfs: server '%s' error: fileid changed. " 382 "fsid %jx:%jx: expected fileid %#jx, got %#jx. " 383 "(BROKEN NFS SERVER OR MIDDLEWARE)\n", 384 nmp->nm_com.nmcom_hostname, 385 (uintmax_t)nmp->nm_fsid[0], 386 (uintmax_t)nmp->nm_fsid[1], 387 (uintmax_t)oldnap->na_fileid, 388 (uintmax_t)newnap->na_fileid); 389 390 if (off) 391 printf("newnfs: Logged %d times about fileid corruption; " 392 "going quiet to avoid spamming logs excessively. (Limit " 393 "is: %d).\n", ncl_fileid_nwarnings, 394 ncl_fileid_maxwarnings); 395 } 396 397 /* 398 * Load the attribute cache (that lives in the nfsnode entry) with 399 * the attributes of the second argument and 400 * Iff vaper not NULL 401 * copy the attributes to *vaper 402 * Similar to nfs_loadattrcache(), except the attributes are passed in 403 * instead of being parsed out of the mbuf list. 404 */ 405 int 406 nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper, 407 void *stuff, int writeattr, int dontshrink) 408 { 409 struct vnode *vp = *vpp; 410 struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper; 411 struct nfsnode *np; 412 struct nfsmount *nmp; 413 struct timespec mtime_save; 414 u_quad_t nsize; 415 int setnsize, error, force_fid_err; 416 417 error = 0; 418 setnsize = 0; 419 nsize = 0; 420 421 /* 422 * If v_type == VNON it is a new node, so fill in the v_type, 423 * n_mtime fields. Check to see if it represents a special 424 * device, and if so, check for a possible alias. Once the 425 * correct vnode has been obtained, fill in the rest of the 426 * information. 427 */ 428 np = VTONFS(vp); 429 NFSLOCKNODE(np); 430 if (vp->v_type != nvap->va_type) { 431 vp->v_type = nvap->va_type; 432 if (vp->v_type == VFIFO) 433 vp->v_op = &newnfs_fifoops; 434 np->n_mtime = nvap->va_mtime; 435 } 436 nmp = VFSTONFS(vp->v_mount); 437 vap = &np->n_vattr.na_vattr; 438 mtime_save = vap->va_mtime; 439 if (writeattr) { 440 np->n_vattr.na_filerev = nap->na_filerev; 441 np->n_vattr.na_size = nap->na_size; 442 np->n_vattr.na_mtime = nap->na_mtime; 443 np->n_vattr.na_ctime = nap->na_ctime; 444 np->n_vattr.na_fsid = nap->na_fsid; 445 np->n_vattr.na_mode = nap->na_mode; 446 } else { 447 force_fid_err = 0; 448 KFAIL_POINT_ERROR(DEBUG_FP, nfscl_force_fileid_warning, 449 force_fid_err); 450 /* 451 * BROKEN NFS SERVER OR MIDDLEWARE 452 * 453 * Certain NFS servers (certain old proprietary filers ca. 454 * 2006) or broken middleboxes (e.g. WAN accelerator products) 455 * will respond to GETATTR requests with results for a 456 * different fileid. 457 * 458 * The WAN accelerator we've observed not only serves stale 459 * cache results for a given file, it also occasionally serves 460 * results for wholly different files. This causes surprising 461 * problems; for example the cached size attribute of a file 462 * may truncate down and then back up, resulting in zero 463 * regions in file contents read by applications. We observed 464 * this reliably with Clang and .c files during parallel build. 465 * A pcap revealed packet fragmentation and GETATTR RPC 466 * responses with wholly wrong fileids. 467 */ 468 if ((np->n_vattr.na_fileid != 0 && 469 np->n_vattr.na_fileid != nap->na_fileid) || 470 force_fid_err) { 471 nfscl_warn_fileid(nmp, &np->n_vattr, nap); 472 error = EIDRM; 473 goto out; 474 } 475 NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr, 476 sizeof (struct nfsvattr)); 477 } 478 479 /* 480 * For NFSv4, if the node's fsid is not equal to the mount point's 481 * fsid, return the low order 32bits of the node's fsid. This 482 * allows getcwd(3) to work. There is a chance that the fsid might 483 * be the same as a local fs, but since this is in an NFS mount 484 * point, I don't think that will cause any problems? 485 */ 486 if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) && 487 (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] || 488 nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) { 489 /* 490 * va_fsid needs to be set to some value derived from 491 * np->n_vattr.na_filesid that is not equal 492 * vp->v_mount->mnt_stat.f_fsid[0], so that it changes 493 * from the value used for the top level server volume 494 * in the mounted subtree. 495 */ 496 if (vp->v_mount->mnt_stat.f_fsid.val[0] != 497 (uint32_t)np->n_vattr.na_filesid[0]) 498 vap->va_fsid = (uint32_t)np->n_vattr.na_filesid[0]; 499 else 500 vap->va_fsid = (uint32_t)hash32_buf( 501 np->n_vattr.na_filesid, 2 * sizeof(uint64_t), 0); 502 } else 503 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 504 np->n_attrstamp = time_second; 505 if (vap->va_size != np->n_size) { 506 if (vap->va_type == VREG) { 507 if (dontshrink && vap->va_size < np->n_size) { 508 /* 509 * We've been told not to shrink the file; 510 * zero np->n_attrstamp to indicate that 511 * the attributes are stale. 512 */ 513 vap->va_size = np->n_size; 514 np->n_attrstamp = 0; 515 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 516 vnode_pager_setsize(vp, np->n_size); 517 } else if (np->n_flag & NMODIFIED) { 518 /* 519 * We've modified the file: Use the larger 520 * of our size, and the server's size. 521 */ 522 if (vap->va_size < np->n_size) { 523 vap->va_size = np->n_size; 524 } else { 525 np->n_size = vap->va_size; 526 np->n_flag |= NSIZECHANGED; 527 } 528 vnode_pager_setsize(vp, np->n_size); 529 } else if (vap->va_size < np->n_size) { 530 /* 531 * When shrinking the size, the call to 532 * vnode_pager_setsize() cannot be done 533 * with the mutex held, so delay it until 534 * after the mtx_unlock call. 535 */ 536 nsize = np->n_size = vap->va_size; 537 np->n_flag |= NSIZECHANGED; 538 setnsize = 1; 539 } else { 540 np->n_size = vap->va_size; 541 np->n_flag |= NSIZECHANGED; 542 vnode_pager_setsize(vp, np->n_size); 543 } 544 } else { 545 np->n_size = vap->va_size; 546 } 547 } 548 /* 549 * The following checks are added to prevent a race between (say) 550 * a READDIR+ and a WRITE. 551 * READDIR+, WRITE requests sent out. 552 * READDIR+ resp, WRITE resp received on client. 553 * However, the WRITE resp was handled before the READDIR+ resp 554 * causing the post op attrs from the write to be loaded first 555 * and the attrs from the READDIR+ to be loaded later. If this 556 * happens, we have stale attrs loaded into the attrcache. 557 * We detect this by for the mtime moving back. We invalidate the 558 * attrcache when this happens. 559 */ 560 if (timespeccmp(&mtime_save, &vap->va_mtime, >)) { 561 /* Size changed or mtime went backwards */ 562 np->n_attrstamp = 0; 563 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 564 } 565 if (vaper != NULL) { 566 NFSBCOPY((caddr_t)vap, (caddr_t)vaper, sizeof(*vap)); 567 if (np->n_flag & NCHG) { 568 if (np->n_flag & NACC) 569 vaper->va_atime = np->n_atim; 570 if (np->n_flag & NUPD) 571 vaper->va_mtime = np->n_mtim; 572 } 573 } 574 575 out: 576 #ifdef KDTRACE_HOOKS 577 if (np->n_attrstamp != 0) 578 KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, error); 579 #endif 580 NFSUNLOCKNODE(np); 581 if (setnsize) 582 vnode_pager_setsize(vp, nsize); 583 return (error); 584 } 585 586 /* 587 * Fill in the client id name. For these bytes: 588 * 1 - they must be unique 589 * 2 - they should be persistent across client reboots 590 * 1 is more critical than 2 591 * Use the mount point's unique id plus either the uuid or, if that 592 * isn't set, random junk. 593 */ 594 void 595 nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen) 596 { 597 int uuidlen; 598 599 /* 600 * First, put in the 64bit mount point identifier. 601 */ 602 if (idlen >= sizeof (u_int64_t)) { 603 NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t)); 604 cp += sizeof (u_int64_t); 605 idlen -= sizeof (u_int64_t); 606 } 607 608 /* 609 * If uuid is non-zero length, use it. 610 */ 611 uuidlen = strlen(uuid); 612 if (uuidlen > 0 && idlen >= uuidlen) { 613 NFSBCOPY(uuid, cp, uuidlen); 614 cp += uuidlen; 615 idlen -= uuidlen; 616 } 617 618 /* 619 * This only normally happens if the uuid isn't set. 620 */ 621 while (idlen > 0) { 622 *cp++ = (u_int8_t)(arc4random() % 256); 623 idlen--; 624 } 625 } 626 627 /* 628 * Fill in a lock owner name. For now, pid + the process's creation time. 629 */ 630 void 631 nfscl_filllockowner(void *id, u_int8_t *cp, int flags) 632 { 633 union { 634 u_int32_t lval; 635 u_int8_t cval[4]; 636 } tl; 637 struct proc *p; 638 639 if (id == NULL) { 640 printf("NULL id\n"); 641 bzero(cp, NFSV4CL_LOCKNAMELEN); 642 return; 643 } 644 if ((flags & F_POSIX) != 0) { 645 p = (struct proc *)id; 646 tl.lval = p->p_pid; 647 *cp++ = tl.cval[0]; 648 *cp++ = tl.cval[1]; 649 *cp++ = tl.cval[2]; 650 *cp++ = tl.cval[3]; 651 tl.lval = p->p_stats->p_start.tv_sec; 652 *cp++ = tl.cval[0]; 653 *cp++ = tl.cval[1]; 654 *cp++ = tl.cval[2]; 655 *cp++ = tl.cval[3]; 656 tl.lval = p->p_stats->p_start.tv_usec; 657 *cp++ = tl.cval[0]; 658 *cp++ = tl.cval[1]; 659 *cp++ = tl.cval[2]; 660 *cp = tl.cval[3]; 661 } else if ((flags & F_FLOCK) != 0) { 662 bcopy(&id, cp, sizeof(id)); 663 bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id)); 664 } else { 665 printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n"); 666 bzero(cp, NFSV4CL_LOCKNAMELEN); 667 } 668 } 669 670 /* 671 * Find the parent process for the thread passed in as an argument. 672 * If none exists, return NULL, otherwise return a thread for the parent. 673 * (Can be any of the threads, since it is only used for td->td_proc.) 674 */ 675 NFSPROC_T * 676 nfscl_getparent(struct thread *td) 677 { 678 struct proc *p; 679 struct thread *ptd; 680 681 if (td == NULL) 682 return (NULL); 683 p = td->td_proc; 684 if (p->p_pid == 0) 685 return (NULL); 686 p = p->p_pptr; 687 if (p == NULL) 688 return (NULL); 689 ptd = TAILQ_FIRST(&p->p_threads); 690 return (ptd); 691 } 692 693 /* 694 * Start up the renew kernel thread. 695 */ 696 static void 697 start_nfscl(void *arg) 698 { 699 struct nfsclclient *clp; 700 struct thread *td; 701 702 clp = (struct nfsclclient *)arg; 703 td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads); 704 nfscl_renewthread(clp, td); 705 kproc_exit(0); 706 } 707 708 void 709 nfscl_start_renewthread(struct nfsclclient *clp) 710 { 711 712 kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0, 713 "nfscl"); 714 } 715 716 /* 717 * Handle wcc_data. 718 * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr 719 * as the first Op after PutFH. 720 * (For NFSv4, the postop attributes are after the Op, so they can't be 721 * parsed here. A separate call to nfscl_postop_attr() is required.) 722 */ 723 int 724 nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp, 725 struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff) 726 { 727 u_int32_t *tl; 728 struct nfsnode *np = VTONFS(vp); 729 struct nfsvattr nfsva; 730 int error = 0; 731 732 if (wccflagp != NULL) 733 *wccflagp = 0; 734 if (nd->nd_flag & ND_NFSV3) { 735 *flagp = 0; 736 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 737 if (*tl == newnfs_true) { 738 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 739 if (wccflagp != NULL) { 740 mtx_lock(&np->n_mtx); 741 *wccflagp = (np->n_mtime.tv_sec == 742 fxdr_unsigned(u_int32_t, *(tl + 2)) && 743 np->n_mtime.tv_nsec == 744 fxdr_unsigned(u_int32_t, *(tl + 3))); 745 mtx_unlock(&np->n_mtx); 746 } 747 } 748 error = nfscl_postop_attr(nd, nap, flagp, stuff); 749 } else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) 750 == (ND_NFSV4 | ND_V4WCCATTR)) { 751 error = nfsv4_loadattr(nd, NULL, &nfsva, NULL, 752 NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, 753 NULL, NULL, NULL, NULL, NULL); 754 if (error) 755 return (error); 756 /* 757 * Get rid of Op# and status for next op. 758 */ 759 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 760 if (*++tl) 761 nd->nd_flag |= ND_NOMOREDATA; 762 if (wccflagp != NULL && 763 nfsva.na_vattr.va_mtime.tv_sec != 0) { 764 mtx_lock(&np->n_mtx); 765 *wccflagp = (np->n_mtime.tv_sec == 766 nfsva.na_vattr.va_mtime.tv_sec && 767 np->n_mtime.tv_nsec == 768 nfsva.na_vattr.va_mtime.tv_sec); 769 mtx_unlock(&np->n_mtx); 770 } 771 } 772 nfsmout: 773 return (error); 774 } 775 776 /* 777 * Get postop attributes. 778 */ 779 int 780 nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp, 781 void *stuff) 782 { 783 u_int32_t *tl; 784 int error = 0; 785 786 *retp = 0; 787 if (nd->nd_flag & ND_NOMOREDATA) 788 return (error); 789 if (nd->nd_flag & ND_NFSV3) { 790 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 791 *retp = fxdr_unsigned(int, *tl); 792 } else if (nd->nd_flag & ND_NFSV4) { 793 /* 794 * For NFSv4, the postop attr are at the end, so no point 795 * in looking if nd_repstat != 0. 796 */ 797 if (!nd->nd_repstat) { 798 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 799 if (*(tl + 1)) 800 /* should never happen since nd_repstat != 0 */ 801 nd->nd_flag |= ND_NOMOREDATA; 802 else 803 *retp = 1; 804 } 805 } else if (!nd->nd_repstat) { 806 /* For NFSv2, the attributes are here iff nd_repstat == 0 */ 807 *retp = 1; 808 } 809 if (*retp) { 810 error = nfsm_loadattr(nd, nap); 811 if (error) 812 *retp = 0; 813 } 814 nfsmout: 815 return (error); 816 } 817 818 /* 819 * Fill in the setable attributes. The full argument indicates whether 820 * to fill in them all or just mode and time. 821 */ 822 void 823 nfscl_fillsattr(struct nfsrv_descript *nd, struct vattr *vap, 824 struct vnode *vp, int flags, u_int32_t rdev) 825 { 826 u_int32_t *tl; 827 struct nfsv2_sattr *sp; 828 nfsattrbit_t attrbits; 829 830 switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) { 831 case ND_NFSV2: 832 NFSM_BUILD(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 833 if (vap->va_mode == (mode_t)VNOVAL) 834 sp->sa_mode = newnfs_xdrneg1; 835 else 836 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 837 if (vap->va_uid == (uid_t)VNOVAL) 838 sp->sa_uid = newnfs_xdrneg1; 839 else 840 sp->sa_uid = txdr_unsigned(vap->va_uid); 841 if (vap->va_gid == (gid_t)VNOVAL) 842 sp->sa_gid = newnfs_xdrneg1; 843 else 844 sp->sa_gid = txdr_unsigned(vap->va_gid); 845 if (flags & NFSSATTR_SIZE0) 846 sp->sa_size = 0; 847 else if (flags & NFSSATTR_SIZENEG1) 848 sp->sa_size = newnfs_xdrneg1; 849 else if (flags & NFSSATTR_SIZERDEV) 850 sp->sa_size = txdr_unsigned(rdev); 851 else 852 sp->sa_size = txdr_unsigned(vap->va_size); 853 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 854 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 855 break; 856 case ND_NFSV3: 857 if (vap->va_mode != (mode_t)VNOVAL) { 858 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 859 *tl++ = newnfs_true; 860 *tl = txdr_unsigned(vap->va_mode); 861 } else { 862 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 863 *tl = newnfs_false; 864 } 865 if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL) { 866 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 867 *tl++ = newnfs_true; 868 *tl = txdr_unsigned(vap->va_uid); 869 } else { 870 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 871 *tl = newnfs_false; 872 } 873 if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL) { 874 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 875 *tl++ = newnfs_true; 876 *tl = txdr_unsigned(vap->va_gid); 877 } else { 878 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 879 *tl = newnfs_false; 880 } 881 if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL) { 882 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 883 *tl++ = newnfs_true; 884 txdr_hyper(vap->va_size, tl); 885 } else { 886 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 887 *tl = newnfs_false; 888 } 889 if (vap->va_atime.tv_sec != VNOVAL) { 890 if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) { 891 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 892 *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT); 893 txdr_nfsv3time(&vap->va_atime, tl); 894 } else { 895 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 896 *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER); 897 } 898 } else { 899 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 900 *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE); 901 } 902 if (vap->va_mtime.tv_sec != VNOVAL) { 903 if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) { 904 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 905 *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT); 906 txdr_nfsv3time(&vap->va_mtime, tl); 907 } else { 908 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 909 *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER); 910 } 911 } else { 912 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 913 *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE); 914 } 915 break; 916 case ND_NFSV4: 917 NFSZERO_ATTRBIT(&attrbits); 918 if (vap->va_mode != (mode_t)VNOVAL) 919 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_MODE); 920 if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL) 921 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNER); 922 if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL) 923 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNERGROUP); 924 if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL) 925 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 926 if (vap->va_atime.tv_sec != VNOVAL) 927 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESSSET); 928 if (vap->va_mtime.tv_sec != VNOVAL) 929 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFYSET); 930 (void) nfsv4_fillattr(nd, vp->v_mount, vp, NULL, vap, NULL, 0, 931 &attrbits, NULL, NULL, 0, 0, 0, 0, (uint64_t)0); 932 break; 933 } 934 } 935 936 /* 937 * nfscl_request() - mostly a wrapper for newnfs_request(). 938 */ 939 int 940 nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p, 941 struct ucred *cred, void *stuff) 942 { 943 int ret, vers; 944 struct nfsmount *nmp; 945 946 nmp = VFSTONFS(vp->v_mount); 947 if (nd->nd_flag & ND_NFSV4) 948 vers = NFS_VER4; 949 else if (nd->nd_flag & ND_NFSV3) 950 vers = NFS_VER3; 951 else 952 vers = NFS_VER2; 953 ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred, 954 NFS_PROG, vers, NULL, 1, NULL, NULL); 955 return (ret); 956 } 957 958 /* 959 * fill in this bsden's variant of statfs using nfsstatfs. 960 */ 961 void 962 nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs) 963 { 964 struct statfs *sbp = (struct statfs *)statfs; 965 966 if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) { 967 sbp->f_bsize = NFS_FABLKSIZE; 968 sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE; 969 sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE; 970 /* 971 * Although sf_abytes is uint64_t and f_bavail is int64_t, 972 * the value after dividing by NFS_FABLKSIZE is small 973 * enough that it will fit in 63bits, so it is ok to 974 * assign it to f_bavail without fear that it will become 975 * negative. 976 */ 977 sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE; 978 sbp->f_files = sfp->sf_tfiles; 979 /* Since f_ffree is int64_t, clip it to 63bits. */ 980 if (sfp->sf_ffiles > INT64_MAX) 981 sbp->f_ffree = INT64_MAX; 982 else 983 sbp->f_ffree = sfp->sf_ffiles; 984 } else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) { 985 /* 986 * The type casts to (int32_t) ensure that this code is 987 * compatible with the old NFS client, in that it will 988 * propagate bit31 to the high order bits. This may or may 989 * not be correct for NFSv2, but since it is a legacy 990 * environment, I'd rather retain backwards compatibility. 991 */ 992 sbp->f_bsize = (int32_t)sfp->sf_bsize; 993 sbp->f_blocks = (int32_t)sfp->sf_blocks; 994 sbp->f_bfree = (int32_t)sfp->sf_bfree; 995 sbp->f_bavail = (int32_t)sfp->sf_bavail; 996 sbp->f_files = 0; 997 sbp->f_ffree = 0; 998 } 999 } 1000 1001 /* 1002 * Use the fsinfo stuff to update the mount point. 1003 */ 1004 void 1005 nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp) 1006 { 1007 1008 if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) && 1009 fsp->fs_wtpref >= NFS_FABLKSIZE) 1010 nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) & 1011 ~(NFS_FABLKSIZE - 1); 1012 if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) { 1013 nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1); 1014 if (nmp->nm_wsize == 0) 1015 nmp->nm_wsize = fsp->fs_wtmax; 1016 } 1017 if (nmp->nm_wsize < NFS_FABLKSIZE) 1018 nmp->nm_wsize = NFS_FABLKSIZE; 1019 if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) && 1020 fsp->fs_rtpref >= NFS_FABLKSIZE) 1021 nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) & 1022 ~(NFS_FABLKSIZE - 1); 1023 if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) { 1024 nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1); 1025 if (nmp->nm_rsize == 0) 1026 nmp->nm_rsize = fsp->fs_rtmax; 1027 } 1028 if (nmp->nm_rsize < NFS_FABLKSIZE) 1029 nmp->nm_rsize = NFS_FABLKSIZE; 1030 if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize) 1031 && fsp->fs_dtpref >= NFS_DIRBLKSIZ) 1032 nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) & 1033 ~(NFS_DIRBLKSIZ - 1); 1034 if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) { 1035 nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1); 1036 if (nmp->nm_readdirsize == 0) 1037 nmp->nm_readdirsize = fsp->fs_rtmax; 1038 } 1039 if (nmp->nm_readdirsize < NFS_DIRBLKSIZ) 1040 nmp->nm_readdirsize = NFS_DIRBLKSIZ; 1041 if (fsp->fs_maxfilesize > 0 && 1042 fsp->fs_maxfilesize < nmp->nm_maxfilesize) 1043 nmp->nm_maxfilesize = fsp->fs_maxfilesize; 1044 nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp); 1045 nmp->nm_state |= NFSSTA_GOTFSINFO; 1046 } 1047 1048 /* 1049 * Lookups source address which should be used to communicate with 1050 * @nmp and stores it inside @pdst. 1051 * 1052 * Returns 0 on success. 1053 */ 1054 u_int8_t * 1055 nfscl_getmyip(struct nfsmount *nmp, struct in6_addr *paddr, int *isinet6p) 1056 { 1057 #if defined(INET6) || defined(INET) 1058 int error, fibnum; 1059 1060 fibnum = curthread->td_proc->p_fibnum; 1061 #endif 1062 #ifdef INET 1063 if (nmp->nm_nam->sa_family == AF_INET) { 1064 struct sockaddr_in *sin; 1065 struct nhop4_extended nh_ext; 1066 1067 sin = (struct sockaddr_in *)nmp->nm_nam; 1068 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred)); 1069 error = fib4_lookup_nh_ext(fibnum, sin->sin_addr, 0, 0, 1070 &nh_ext); 1071 CURVNET_RESTORE(); 1072 if (error != 0) 1073 return (NULL); 1074 1075 if ((ntohl(nh_ext.nh_src.s_addr) >> IN_CLASSA_NSHIFT) == 1076 IN_LOOPBACKNET) { 1077 /* Ignore loopback addresses */ 1078 return (NULL); 1079 } 1080 1081 *isinet6p = 0; 1082 *((struct in_addr *)paddr) = nh_ext.nh_src; 1083 1084 return (u_int8_t *)paddr; 1085 } 1086 #endif 1087 #ifdef INET6 1088 if (nmp->nm_nam->sa_family == AF_INET6) { 1089 struct sockaddr_in6 *sin6; 1090 1091 sin6 = (struct sockaddr_in6 *)nmp->nm_nam; 1092 1093 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred)); 1094 error = in6_selectsrc_addr(fibnum, &sin6->sin6_addr, 1095 sin6->sin6_scope_id, NULL, paddr, NULL); 1096 CURVNET_RESTORE(); 1097 if (error != 0) 1098 return (NULL); 1099 1100 if (IN6_IS_ADDR_LOOPBACK(paddr)) 1101 return (NULL); 1102 1103 /* Scope is embedded in */ 1104 *isinet6p = 1; 1105 1106 return (u_int8_t *)paddr; 1107 } 1108 #endif 1109 return (NULL); 1110 } 1111 1112 /* 1113 * Copy NFS uid, gids from the cred structure. 1114 */ 1115 void 1116 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr) 1117 { 1118 int i; 1119 1120 KASSERT(cr->cr_ngroups >= 0, 1121 ("newnfs_copyincred: negative cr_ngroups")); 1122 nfscr->nfsc_uid = cr->cr_uid; 1123 nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1); 1124 for (i = 0; i < nfscr->nfsc_ngroups; i++) 1125 nfscr->nfsc_groups[i] = cr->cr_groups[i]; 1126 } 1127 1128 1129 /* 1130 * Do any client specific initialization. 1131 */ 1132 void 1133 nfscl_init(void) 1134 { 1135 static int inited = 0; 1136 1137 if (inited) 1138 return; 1139 inited = 1; 1140 nfscl_inited = 1; 1141 ncl_pbuf_freecnt = nswbuf / 2 + 1; 1142 } 1143 1144 /* 1145 * Check each of the attributes to be set, to ensure they aren't already 1146 * the correct value. Disable setting ones already correct. 1147 */ 1148 int 1149 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap) 1150 { 1151 1152 if (vap->va_mode != (mode_t)VNOVAL) { 1153 if (vap->va_mode == nvap->na_mode) 1154 vap->va_mode = (mode_t)VNOVAL; 1155 } 1156 if (vap->va_uid != (uid_t)VNOVAL) { 1157 if (vap->va_uid == nvap->na_uid) 1158 vap->va_uid = (uid_t)VNOVAL; 1159 } 1160 if (vap->va_gid != (gid_t)VNOVAL) { 1161 if (vap->va_gid == nvap->na_gid) 1162 vap->va_gid = (gid_t)VNOVAL; 1163 } 1164 if (vap->va_size != VNOVAL) { 1165 if (vap->va_size == nvap->na_size) 1166 vap->va_size = VNOVAL; 1167 } 1168 1169 /* 1170 * We are normally called with only a partially initialized 1171 * VAP. Since the NFSv3 spec says that server may use the 1172 * file attributes to store the verifier, the spec requires 1173 * us to do a SETATTR RPC. FreeBSD servers store the verifier 1174 * in atime, but we can't really assume that all servers will 1175 * so we ensure that our SETATTR sets both atime and mtime. 1176 * Set the VA_UTIMES_NULL flag for this case, so that 1177 * the server's time will be used. This is needed to 1178 * work around a bug in some Solaris servers, where 1179 * setting the time TOCLIENT causes the Setattr RPC 1180 * to return NFS_OK, but not set va_mode. 1181 */ 1182 if (vap->va_mtime.tv_sec == VNOVAL) { 1183 vfs_timestamp(&vap->va_mtime); 1184 vap->va_vaflags |= VA_UTIMES_NULL; 1185 } 1186 if (vap->va_atime.tv_sec == VNOVAL) 1187 vap->va_atime = vap->va_mtime; 1188 return (1); 1189 } 1190 1191 /* 1192 * Map nfsv4 errors to errno.h errors. 1193 * The uid and gid arguments are only used for NFSERR_BADOWNER and that 1194 * error should only be returned for the Open, Create and Setattr Ops. 1195 * As such, most calls can just pass in 0 for those arguments. 1196 */ 1197 APPLESTATIC int 1198 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid) 1199 { 1200 struct proc *p; 1201 1202 if (error < 10000) 1203 return (error); 1204 if (td != NULL) 1205 p = td->td_proc; 1206 else 1207 p = NULL; 1208 switch (error) { 1209 case NFSERR_BADOWNER: 1210 tprintf(p, LOG_INFO, 1211 "No name and/or group mapping for uid,gid:(%d,%d)\n", 1212 uid, gid); 1213 return (EPERM); 1214 case NFSERR_BADNAME: 1215 case NFSERR_BADCHAR: 1216 printf("nfsv4 char/name not handled by server\n"); 1217 return (ENOENT); 1218 case NFSERR_STALECLIENTID: 1219 case NFSERR_STALESTATEID: 1220 case NFSERR_EXPIRED: 1221 case NFSERR_BADSTATEID: 1222 case NFSERR_BADSESSION: 1223 printf("nfsv4 recover err returned %d\n", error); 1224 return (EIO); 1225 case NFSERR_BADHANDLE: 1226 case NFSERR_SERVERFAULT: 1227 case NFSERR_BADTYPE: 1228 case NFSERR_FHEXPIRED: 1229 case NFSERR_RESOURCE: 1230 case NFSERR_MOVED: 1231 case NFSERR_NOFILEHANDLE: 1232 case NFSERR_MINORVERMISMATCH: 1233 case NFSERR_OLDSTATEID: 1234 case NFSERR_BADSEQID: 1235 case NFSERR_LEASEMOVED: 1236 case NFSERR_RECLAIMBAD: 1237 case NFSERR_BADXDR: 1238 case NFSERR_OPILLEGAL: 1239 printf("nfsv4 client/server protocol prob err=%d\n", 1240 error); 1241 return (EIO); 1242 default: 1243 tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error); 1244 return (EIO); 1245 }; 1246 } 1247 1248 /* 1249 * Check to see if the process for this owner exists. Return 1 if it doesn't 1250 * and 0 otherwise. 1251 */ 1252 int 1253 nfscl_procdoesntexist(u_int8_t *own) 1254 { 1255 union { 1256 u_int32_t lval; 1257 u_int8_t cval[4]; 1258 } tl; 1259 struct proc *p; 1260 pid_t pid; 1261 int ret = 0; 1262 1263 tl.cval[0] = *own++; 1264 tl.cval[1] = *own++; 1265 tl.cval[2] = *own++; 1266 tl.cval[3] = *own++; 1267 pid = tl.lval; 1268 p = pfind_locked(pid); 1269 if (p == NULL) 1270 return (1); 1271 if (p->p_stats == NULL) { 1272 PROC_UNLOCK(p); 1273 return (0); 1274 } 1275 tl.cval[0] = *own++; 1276 tl.cval[1] = *own++; 1277 tl.cval[2] = *own++; 1278 tl.cval[3] = *own++; 1279 if (tl.lval != p->p_stats->p_start.tv_sec) { 1280 ret = 1; 1281 } else { 1282 tl.cval[0] = *own++; 1283 tl.cval[1] = *own++; 1284 tl.cval[2] = *own++; 1285 tl.cval[3] = *own; 1286 if (tl.lval != p->p_stats->p_start.tv_usec) 1287 ret = 1; 1288 } 1289 PROC_UNLOCK(p); 1290 return (ret); 1291 } 1292 1293 /* 1294 * - nfs pseudo system call for the client 1295 */ 1296 /* 1297 * MPSAFE 1298 */ 1299 static int 1300 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap) 1301 { 1302 struct file *fp; 1303 struct nfscbd_args nfscbdarg; 1304 struct nfsd_nfscbd_args nfscbdarg2; 1305 struct nameidata nd; 1306 struct nfscl_dumpmntopts dumpmntopts; 1307 cap_rights_t rights; 1308 char *buf; 1309 int error; 1310 1311 if (uap->flag & NFSSVC_CBADDSOCK) { 1312 error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg)); 1313 if (error) 1314 return (error); 1315 /* 1316 * Since we don't know what rights might be required, 1317 * pretend that we need them all. It is better to be too 1318 * careful than too reckless. 1319 */ 1320 error = fget(td, nfscbdarg.sock, 1321 cap_rights_init(&rights, CAP_SOCK_CLIENT), &fp); 1322 if (error) 1323 return (error); 1324 if (fp->f_type != DTYPE_SOCKET) { 1325 fdrop(fp, td); 1326 return (EPERM); 1327 } 1328 error = nfscbd_addsock(fp); 1329 fdrop(fp, td); 1330 if (!error && nfscl_enablecallb == 0) { 1331 nfsv4_cbport = nfscbdarg.port; 1332 nfscl_enablecallb = 1; 1333 } 1334 } else if (uap->flag & NFSSVC_NFSCBD) { 1335 if (uap->argp == NULL) 1336 return (EINVAL); 1337 error = copyin(uap->argp, (caddr_t)&nfscbdarg2, 1338 sizeof(nfscbdarg2)); 1339 if (error) 1340 return (error); 1341 error = nfscbd_nfsd(td, &nfscbdarg2); 1342 } else if (uap->flag & NFSSVC_DUMPMNTOPTS) { 1343 error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts)); 1344 if (error == 0 && (dumpmntopts.ndmnt_blen < 256 || 1345 dumpmntopts.ndmnt_blen > 1024)) 1346 error = EINVAL; 1347 if (error == 0) 1348 error = nfsrv_lookupfilename(&nd, 1349 dumpmntopts.ndmnt_fname, td); 1350 if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, 1351 "nfs") != 0) { 1352 vput(nd.ni_vp); 1353 error = EINVAL; 1354 } 1355 if (error == 0) { 1356 buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK); 1357 nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf, 1358 dumpmntopts.ndmnt_blen); 1359 vput(nd.ni_vp); 1360 error = copyout(buf, dumpmntopts.ndmnt_buf, 1361 dumpmntopts.ndmnt_blen); 1362 free(buf, M_TEMP); 1363 } 1364 } else { 1365 error = EINVAL; 1366 } 1367 return (error); 1368 } 1369 1370 extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *); 1371 1372 /* 1373 * Called once to initialize data structures... 1374 */ 1375 static int 1376 nfscl_modevent(module_t mod, int type, void *data) 1377 { 1378 int error = 0; 1379 static int loaded = 0; 1380 1381 switch (type) { 1382 case MOD_LOAD: 1383 if (loaded) 1384 return (0); 1385 newnfs_portinit(); 1386 mtx_init(&nfs_clstate_mutex, "nfs_clstate_mutex", NULL, 1387 MTX_DEF); 1388 mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF); 1389 nfscl_init(); 1390 NFSD_LOCK(); 1391 nfsrvd_cbinit(0); 1392 NFSD_UNLOCK(); 1393 ncl_call_invalcaches = ncl_invalcaches; 1394 nfsd_call_nfscl = nfssvc_nfscl; 1395 loaded = 1; 1396 break; 1397 1398 case MOD_UNLOAD: 1399 if (nfs_numnfscbd != 0) { 1400 error = EBUSY; 1401 break; 1402 } 1403 1404 /* 1405 * XXX: Unloading of nfscl module is unsupported. 1406 */ 1407 #if 0 1408 ncl_call_invalcaches = NULL; 1409 nfsd_call_nfscl = NULL; 1410 /* and get rid of the mutexes */ 1411 mtx_destroy(&nfs_clstate_mutex); 1412 mtx_destroy(&ncl_iod_mutex); 1413 loaded = 0; 1414 break; 1415 #else 1416 /* FALLTHROUGH */ 1417 #endif 1418 default: 1419 error = EOPNOTSUPP; 1420 break; 1421 } 1422 return error; 1423 } 1424 static moduledata_t nfscl_mod = { 1425 "nfscl", 1426 nfscl_modevent, 1427 NULL, 1428 }; 1429 DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST); 1430 1431 /* So that loader and kldload(2) can find us, wherever we are.. */ 1432 MODULE_VERSION(nfscl, 1); 1433 MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1); 1434 MODULE_DEPEND(nfscl, krpc, 1, 1, 1); 1435 MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1); 1436 MODULE_DEPEND(nfscl, nfslock, 1, 1, 1); 1437 1438
Indexes created Sun Oct 12 09:09:55 GMT 2025