nfs_clport.c revision 330897
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 * 4. 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/cdefs.h> 37__FBSDID("$FreeBSD: stable/11/sys/fs/nfsclient/nfs_clport.c 330897 2018-03-14 03:19:51Z eadler $"); 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 60dtrace_nfsclient_attrcache_flush_probe_func_t 61 dtrace_nfscl_attrcache_flush_done_probe; 62uint32_t nfscl_attrcache_flush_done_id; 63 64dtrace_nfsclient_attrcache_get_hit_probe_func_t 65 dtrace_nfscl_attrcache_get_hit_probe; 66uint32_t nfscl_attrcache_get_hit_id; 67 68dtrace_nfsclient_attrcache_get_miss_probe_func_t 69 dtrace_nfscl_attrcache_get_miss_probe; 70uint32_t nfscl_attrcache_get_miss_id; 71 72dtrace_nfsclient_attrcache_load_probe_func_t 73 dtrace_nfscl_attrcache_load_done_probe; 74uint32_t nfscl_attrcache_load_done_id; 75#endif /* !KDTRACE_HOOKS */ 76 77extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1; 78extern struct vop_vector newnfs_vnodeops; 79extern struct vop_vector newnfs_fifoops; 80extern uma_zone_t newnfsnode_zone; 81extern struct buf_ops buf_ops_newnfs; 82extern int ncl_pbuf_freecnt; 83extern short nfsv4_cbport; 84extern int nfscl_enablecallb; 85extern int nfs_numnfscbd; 86extern int nfscl_inited; 87struct mtx ncl_iod_mutex; 88NFSDLOCKMUTEX; 89 90extern void (*ncl_call_invalcaches)(struct vnode *); 91 92SYSCTL_DECL(_vfs_nfs); 93static int ncl_fileid_maxwarnings = 10; 94SYSCTL_INT(_vfs_nfs, OID_AUTO, fileid_maxwarnings, CTLFLAG_RWTUN, 95 &ncl_fileid_maxwarnings, 0, 96 "Limit fileid corruption warnings; 0 is off; -1 is unlimited"); 97static volatile int ncl_fileid_nwarnings; 98 99static void nfscl_warn_fileid(struct nfsmount *, struct nfsvattr *, 100 struct nfsvattr *); 101 102/* 103 * Comparison function for vfs_hash functions. 104 */ 105int 106newnfs_vncmpf(struct vnode *vp, void *arg) 107{ 108 struct nfsfh *nfhp = (struct nfsfh *)arg; 109 struct nfsnode *np = VTONFS(vp); 110 111 if (np->n_fhp->nfh_len != nfhp->nfh_len || 112 NFSBCMP(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len)) 113 return (1); 114 return (0); 115} 116 117/* 118 * Look up a vnode/nfsnode by file handle. 119 * Callers must check for mount points!! 120 * In all cases, a pointer to a 121 * nfsnode structure is returned. 122 * This variant takes a "struct nfsfh *" as second argument and uses 123 * that structure up, either by hanging off the nfsnode or FREEing it. 124 */ 125int 126nfscl_nget(struct mount *mntp, struct vnode *dvp, struct nfsfh *nfhp, 127 struct componentname *cnp, struct thread *td, struct nfsnode **npp, 128 void *stuff, int lkflags) 129{ 130 struct nfsnode *np, *dnp; 131 struct vnode *vp, *nvp; 132 struct nfsv4node *newd, *oldd; 133 int error; 134 u_int hash; 135 struct nfsmount *nmp; 136 137 nmp = VFSTONFS(mntp); 138 dnp = VTONFS(dvp); 139 *npp = NULL; 140 141 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT); 142 143 error = vfs_hash_get(mntp, hash, lkflags, 144 td, &nvp, newnfs_vncmpf, nfhp); 145 if (error == 0 && nvp != NULL) { 146 /* 147 * I believe there is a slight chance that vgonel() could 148 * get called on this vnode between when NFSVOPLOCK() drops 149 * the VI_LOCK() and vget() acquires it again, so that it 150 * hasn't yet had v_usecount incremented. If this were to 151 * happen, the VI_DOOMED flag would be set, so check for 152 * that here. Since we now have the v_usecount incremented, 153 * we should be ok until we vrele() it, if the VI_DOOMED 154 * flag isn't set now. 155 */ 156 VI_LOCK(nvp); 157 if ((nvp->v_iflag & VI_DOOMED)) { 158 VI_UNLOCK(nvp); 159 vrele(nvp); 160 error = ENOENT; 161 } else { 162 VI_UNLOCK(nvp); 163 } 164 } 165 if (error) { 166 FREE((caddr_t)nfhp, M_NFSFH); 167 return (error); 168 } 169 if (nvp != NULL) { 170 np = VTONFS(nvp); 171 /* 172 * For NFSv4, check to see if it is the same name and 173 * replace the name, if it is different. 174 */ 175 oldd = newd = NULL; 176 if ((nmp->nm_flag & NFSMNT_NFSV4) && np->n_v4 != NULL && 177 nvp->v_type == VREG && 178 (np->n_v4->n4_namelen != cnp->cn_namelen || 179 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4), 180 cnp->cn_namelen) || 181 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || 182 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 183 dnp->n_fhp->nfh_len))) { 184 MALLOC(newd, struct nfsv4node *, 185 sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len + 186 + cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK); 187 NFSLOCKNODE(np); 188 if (newd != NULL && np->n_v4 != NULL && nvp->v_type == VREG 189 && (np->n_v4->n4_namelen != cnp->cn_namelen || 190 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4), 191 cnp->cn_namelen) || 192 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || 193 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 194 dnp->n_fhp->nfh_len))) { 195 oldd = np->n_v4; 196 np->n_v4 = newd; 197 newd = NULL; 198 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; 199 np->n_v4->n4_namelen = cnp->cn_namelen; 200 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 201 dnp->n_fhp->nfh_len); 202 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4), 203 cnp->cn_namelen); 204 } 205 NFSUNLOCKNODE(np); 206 } 207 if (newd != NULL) 208 FREE((caddr_t)newd, M_NFSV4NODE); 209 if (oldd != NULL) 210 FREE((caddr_t)oldd, M_NFSV4NODE); 211 *npp = np; 212 FREE((caddr_t)nfhp, M_NFSFH); 213 return (0); 214 } 215 np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO); 216 217 error = getnewvnode(nfs_vnode_tag, mntp, &newnfs_vnodeops, &nvp); 218 if (error) { 219 uma_zfree(newnfsnode_zone, np); 220 FREE((caddr_t)nfhp, M_NFSFH); 221 return (error); 222 } 223 vp = nvp; 224 KASSERT(vp->v_bufobj.bo_bsize != 0, ("nfscl_nget: bo_bsize == 0")); 225 vp->v_bufobj.bo_ops = &buf_ops_newnfs; 226 vp->v_data = np; 227 np->n_vnode = vp; 228 /* 229 * Initialize the mutex even if the vnode is going to be a loser. 230 * This simplifies the logic in reclaim, which can then unconditionally 231 * destroy the mutex (in the case of the loser, or if hash_insert 232 * happened to return an error no special casing is needed). 233 */ 234 mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK); 235 lockinit(&np->n_excl, PVFS, "nfsupg", VLKTIMEOUT, LK_NOSHARE | 236 LK_CANRECURSE); 237 238 /* 239 * Are we getting the root? If so, make sure the vnode flags 240 * are correct 241 */ 242 if ((nfhp->nfh_len == nmp->nm_fhsize) && 243 !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len)) { 244 if (vp->v_type == VNON) 245 vp->v_type = VDIR; 246 vp->v_vflag |= VV_ROOT; 247 } 248 249 np->n_fhp = nfhp; 250 /* 251 * For NFSv4, we have to attach the directory file handle and 252 * file name, so that Open Ops can be done later. 253 */ 254 if (nmp->nm_flag & NFSMNT_NFSV4) { 255 MALLOC(np->n_v4, struct nfsv4node *, sizeof (struct nfsv4node) 256 + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE, 257 M_WAITOK); 258 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; 259 np->n_v4->n4_namelen = cnp->cn_namelen; 260 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 261 dnp->n_fhp->nfh_len); 262 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4), 263 cnp->cn_namelen); 264 } else { 265 np->n_v4 = NULL; 266 } 267 268 /* 269 * NFS supports recursive and shared locking. 270 */ 271 lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL); 272 VN_LOCK_AREC(vp); 273 VN_LOCK_ASHARE(vp); 274 error = insmntque(vp, mntp); 275 if (error != 0) { 276 *npp = NULL; 277 mtx_destroy(&np->n_mtx); 278 lockdestroy(&np->n_excl); 279 FREE((caddr_t)nfhp, M_NFSFH); 280 if (np->n_v4 != NULL) 281 FREE((caddr_t)np->n_v4, M_NFSV4NODE); 282 uma_zfree(newnfsnode_zone, np); 283 return (error); 284 } 285 error = vfs_hash_insert(vp, hash, lkflags, 286 td, &nvp, newnfs_vncmpf, nfhp); 287 if (error) 288 return (error); 289 if (nvp != NULL) { 290 *npp = VTONFS(nvp); 291 /* vfs_hash_insert() vput()'s the losing vnode */ 292 return (0); 293 } 294 *npp = np; 295 296 return (0); 297} 298 299/* 300 * Another variant of nfs_nget(). This one is only used by reopen. It 301 * takes almost the same args as nfs_nget(), but only succeeds if an entry 302 * exists in the cache. (Since files should already be "open" with a 303 * vnode ref cnt on the node when reopen calls this, it should always 304 * succeed.) 305 * Also, don't get a vnode lock, since it may already be locked by some 306 * other process that is handling it. This is ok, since all other threads 307 * on the client are blocked by the nfsc_lock being exclusively held by the 308 * caller of this function. 309 */ 310int 311nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize, 312 struct thread *td, struct nfsnode **npp) 313{ 314 struct vnode *nvp; 315 u_int hash; 316 struct nfsfh *nfhp; 317 int error; 318 319 *npp = NULL; 320 /* For forced dismounts, just return error. */ 321 if (NFSCL_FORCEDISM(mntp)) 322 return (EINTR); 323 MALLOC(nfhp, struct nfsfh *, sizeof (struct nfsfh) + fhsize, 324 M_NFSFH, M_WAITOK); 325 bcopy(fhp, &nfhp->nfh_fh[0], fhsize); 326 nfhp->nfh_len = fhsize; 327 328 hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT); 329 330 /* 331 * First, try to get the vnode locked, but don't block for the lock. 332 */ 333 error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp, 334 newnfs_vncmpf, nfhp); 335 if (error == 0 && nvp != NULL) { 336 NFSVOPUNLOCK(nvp, 0); 337 } else if (error == EBUSY) { 338 /* 339 * It is safe so long as a vflush() with 340 * FORCECLOSE has not been done. Since the Renew thread is 341 * stopped and the MNTK_UNMOUNTF flag is set before doing 342 * a vflush() with FORCECLOSE, we should be ok here. 343 */ 344 if (NFSCL_FORCEDISM(mntp)) 345 error = EINTR; 346 else { 347 vfs_hash_ref(mntp, hash, td, &nvp, newnfs_vncmpf, nfhp); 348 if (nvp == NULL) { 349 error = ENOENT; 350 } else if ((nvp->v_iflag & VI_DOOMED) != 0) { 351 error = ENOENT; 352 vrele(nvp); 353 } else { 354 error = 0; 355 } 356 } 357 } 358 FREE(nfhp, M_NFSFH); 359 if (error) 360 return (error); 361 if (nvp != NULL) { 362 *npp = VTONFS(nvp); 363 return (0); 364 } 365 return (EINVAL); 366} 367 368static void 369nfscl_warn_fileid(struct nfsmount *nmp, struct nfsvattr *oldnap, 370 struct nfsvattr *newnap) 371{ 372 int off; 373 374 if (ncl_fileid_maxwarnings >= 0 && 375 ncl_fileid_nwarnings >= ncl_fileid_maxwarnings) 376 return; 377 off = 0; 378 if (ncl_fileid_maxwarnings >= 0) { 379 if (++ncl_fileid_nwarnings >= ncl_fileid_maxwarnings) 380 off = 1; 381 } 382 383 printf("newnfs: server '%s' error: fileid changed. " 384 "fsid %jx:%jx: expected fileid %#jx, got %#jx. " 385 "(BROKEN NFS SERVER OR MIDDLEWARE)\n", 386 nmp->nm_com.nmcom_hostname, 387 (uintmax_t)nmp->nm_fsid[0], 388 (uintmax_t)nmp->nm_fsid[1], 389 (uintmax_t)oldnap->na_fileid, 390 (uintmax_t)newnap->na_fileid); 391 392 if (off) 393 printf("newnfs: Logged %d times about fileid corruption; " 394 "going quiet to avoid spamming logs excessively. (Limit " 395 "is: %d).\n", ncl_fileid_nwarnings, 396 ncl_fileid_maxwarnings); 397} 398 399/* 400 * Load the attribute cache (that lives in the nfsnode entry) with 401 * the attributes of the second argument and 402 * Iff vaper not NULL 403 * copy the attributes to *vaper 404 * Similar to nfs_loadattrcache(), except the attributes are passed in 405 * instead of being parsed out of the mbuf list. 406 */ 407int 408nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper, 409 void *stuff, int writeattr, int dontshrink) 410{ 411 struct vnode *vp = *vpp; 412 struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper; 413 struct nfsnode *np; 414 struct nfsmount *nmp; 415 struct timespec mtime_save; 416 u_quad_t nsize; 417 int setnsize, error, force_fid_err; 418 419 error = 0; 420 setnsize = 0; 421 nsize = 0; 422 423 /* 424 * If v_type == VNON it is a new node, so fill in the v_type, 425 * n_mtime fields. Check to see if it represents a special 426 * device, and if so, check for a possible alias. Once the 427 * correct vnode has been obtained, fill in the rest of the 428 * information. 429 */ 430 np = VTONFS(vp); 431 NFSLOCKNODE(np); 432 if (vp->v_type != nvap->va_type) { 433 vp->v_type = nvap->va_type; 434 if (vp->v_type == VFIFO) 435 vp->v_op = &newnfs_fifoops; 436 np->n_mtime = nvap->va_mtime; 437 } 438 nmp = VFSTONFS(vp->v_mount); 439 vap = &np->n_vattr.na_vattr; 440 mtime_save = vap->va_mtime; 441 if (writeattr) { 442 np->n_vattr.na_filerev = nap->na_filerev; 443 np->n_vattr.na_size = nap->na_size; 444 np->n_vattr.na_mtime = nap->na_mtime; 445 np->n_vattr.na_ctime = nap->na_ctime; 446 np->n_vattr.na_fsid = nap->na_fsid; 447 np->n_vattr.na_mode = nap->na_mode; 448 } else { 449 force_fid_err = 0; 450 KFAIL_POINT_ERROR(DEBUG_FP, nfscl_force_fileid_warning, 451 force_fid_err); 452 /* 453 * BROKEN NFS SERVER OR MIDDLEWARE 454 * 455 * Certain NFS servers (certain old proprietary filers ca. 456 * 2006) or broken middleboxes (e.g. WAN accelerator products) 457 * will respond to GETATTR requests with results for a 458 * different fileid. 459 * 460 * The WAN accelerator we've observed not only serves stale 461 * cache results for a given file, it also occasionally serves 462 * results for wholly different files. This causes surprising 463 * problems; for example the cached size attribute of a file 464 * may truncate down and then back up, resulting in zero 465 * regions in file contents read by applications. We observed 466 * this reliably with Clang and .c files during parallel build. 467 * A pcap revealed packet fragmentation and GETATTR RPC 468 * responses with wholly wrong fileids. 469 */ 470 if ((np->n_vattr.na_fileid != 0 && 471 np->n_vattr.na_fileid != nap->na_fileid) || 472 force_fid_err) { 473 nfscl_warn_fileid(nmp, &np->n_vattr, nap); 474 error = EIDRM; 475 goto out; 476 } 477 NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr, 478 sizeof (struct nfsvattr)); 479 } 480 481 /* 482 * For NFSv4, if the node's fsid is not equal to the mount point's 483 * fsid, return the low order 32bits of the node's fsid. This 484 * allows getcwd(3) to work. There is a chance that the fsid might 485 * be the same as a local fs, but since this is in an NFS mount 486 * point, I don't think that will cause any problems? 487 */ 488 if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) && 489 (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] || 490 nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) { 491 /* 492 * va_fsid needs to be set to some value derived from 493 * np->n_vattr.na_filesid that is not equal 494 * vp->v_mount->mnt_stat.f_fsid[0], so that it changes 495 * from the value used for the top level server volume 496 * in the mounted subtree. 497 */ 498 if (vp->v_mount->mnt_stat.f_fsid.val[0] != 499 (uint32_t)np->n_vattr.na_filesid[0]) 500 vap->va_fsid = (uint32_t)np->n_vattr.na_filesid[0]; 501 else 502 vap->va_fsid = (uint32_t)hash32_buf( 503 np->n_vattr.na_filesid, 2 * sizeof(uint64_t), 0); 504 } else 505 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 506 np->n_attrstamp = time_second; 507 if (vap->va_size != np->n_size) { 508 if (vap->va_type == VREG) { 509 if (dontshrink && vap->va_size < np->n_size) { 510 /* 511 * We've been told not to shrink the file; 512 * zero np->n_attrstamp to indicate that 513 * the attributes are stale. 514 */ 515 vap->va_size = np->n_size; 516 np->n_attrstamp = 0; 517 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 518 vnode_pager_setsize(vp, np->n_size); 519 } else if (np->n_flag & NMODIFIED) { 520 /* 521 * We've modified the file: Use the larger 522 * of our size, and the server's size. 523 */ 524 if (vap->va_size < np->n_size) { 525 vap->va_size = np->n_size; 526 } else { 527 np->n_size = vap->va_size; 528 np->n_flag |= NSIZECHANGED; 529 } 530 vnode_pager_setsize(vp, np->n_size); 531 } else if (vap->va_size < np->n_size) { 532 /* 533 * When shrinking the size, the call to 534 * vnode_pager_setsize() cannot be done 535 * with the mutex held, so delay it until 536 * after the mtx_unlock call. 537 */ 538 nsize = np->n_size = vap->va_size; 539 np->n_flag |= NSIZECHANGED; 540 setnsize = 1; 541 } else { 542 np->n_size = vap->va_size; 543 np->n_flag |= NSIZECHANGED; 544 vnode_pager_setsize(vp, np->n_size); 545 } 546 } else { 547 np->n_size = vap->va_size; 548 } 549 } 550 /* 551 * The following checks are added to prevent a race between (say) 552 * a READDIR+ and a WRITE. 553 * READDIR+, WRITE requests sent out. 554 * READDIR+ resp, WRITE resp received on client. 555 * However, the WRITE resp was handled before the READDIR+ resp 556 * causing the post op attrs from the write to be loaded first 557 * and the attrs from the READDIR+ to be loaded later. If this 558 * happens, we have stale attrs loaded into the attrcache. 559 * We detect this by for the mtime moving back. We invalidate the 560 * attrcache when this happens. 561 */ 562 if (timespeccmp(&mtime_save, &vap->va_mtime, >)) { 563 /* Size changed or mtime went backwards */ 564 np->n_attrstamp = 0; 565 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 566 } 567 if (vaper != NULL) { 568 NFSBCOPY((caddr_t)vap, (caddr_t)vaper, sizeof(*vap)); 569 if (np->n_flag & NCHG) { 570 if (np->n_flag & NACC) 571 vaper->va_atime = np->n_atim; 572 if (np->n_flag & NUPD) 573 vaper->va_mtime = np->n_mtim; 574 } 575 } 576 577out: 578#ifdef KDTRACE_HOOKS 579 if (np->n_attrstamp != 0) 580 KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, error); 581#endif 582 NFSUNLOCKNODE(np); 583 if (setnsize) 584 vnode_pager_setsize(vp, nsize); 585 return (error); 586} 587 588/* 589 * Fill in the client id name. For these bytes: 590 * 1 - they must be unique 591 * 2 - they should be persistent across client reboots 592 * 1 is more critical than 2 593 * Use the mount point's unique id plus either the uuid or, if that 594 * isn't set, random junk. 595 */ 596void 597nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen) 598{ 599 int uuidlen; 600 601 /* 602 * First, put in the 64bit mount point identifier. 603 */ 604 if (idlen >= sizeof (u_int64_t)) { 605 NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t)); 606 cp += sizeof (u_int64_t); 607 idlen -= sizeof (u_int64_t); 608 } 609 610 /* 611 * If uuid is non-zero length, use it. 612 */ 613 uuidlen = strlen(uuid); 614 if (uuidlen > 0 && idlen >= uuidlen) { 615 NFSBCOPY(uuid, cp, uuidlen); 616 cp += uuidlen; 617 idlen -= uuidlen; 618 } 619 620 /* 621 * This only normally happens if the uuid isn't set. 622 */ 623 while (idlen > 0) { 624 *cp++ = (u_int8_t)(arc4random() % 256); 625 idlen--; 626 } 627} 628 629/* 630 * Fill in a lock owner name. For now, pid + the process's creation time. 631 */ 632void 633nfscl_filllockowner(void *id, u_int8_t *cp, int flags) 634{ 635 union { 636 u_int32_t lval; 637 u_int8_t cval[4]; 638 } tl; 639 struct proc *p; 640 641 if (id == NULL) { 642 /* Return the single open_owner of all 0 bytes. */ 643 bzero(cp, NFSV4CL_LOCKNAMELEN); 644 return; 645 } 646 if ((flags & F_POSIX) != 0) { 647 p = (struct proc *)id; 648 tl.lval = p->p_pid; 649 *cp++ = tl.cval[0]; 650 *cp++ = tl.cval[1]; 651 *cp++ = tl.cval[2]; 652 *cp++ = tl.cval[3]; 653 tl.lval = p->p_stats->p_start.tv_sec; 654 *cp++ = tl.cval[0]; 655 *cp++ = tl.cval[1]; 656 *cp++ = tl.cval[2]; 657 *cp++ = tl.cval[3]; 658 tl.lval = p->p_stats->p_start.tv_usec; 659 *cp++ = tl.cval[0]; 660 *cp++ = tl.cval[1]; 661 *cp++ = tl.cval[2]; 662 *cp = tl.cval[3]; 663 } else if ((flags & F_FLOCK) != 0) { 664 bcopy(&id, cp, sizeof(id)); 665 bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id)); 666 } else { 667 printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n"); 668 bzero(cp, NFSV4CL_LOCKNAMELEN); 669 } 670} 671 672/* 673 * Find the parent process for the thread passed in as an argument. 674 * If none exists, return NULL, otherwise return a thread for the parent. 675 * (Can be any of the threads, since it is only used for td->td_proc.) 676 */ 677NFSPROC_T * 678nfscl_getparent(struct thread *td) 679{ 680 struct proc *p; 681 struct thread *ptd; 682 683 if (td == NULL) 684 return (NULL); 685 p = td->td_proc; 686 if (p->p_pid == 0) 687 return (NULL); 688 p = p->p_pptr; 689 if (p == NULL) 690 return (NULL); 691 ptd = TAILQ_FIRST(&p->p_threads); 692 return (ptd); 693} 694 695/* 696 * Start up the renew kernel thread. 697 */ 698static void 699start_nfscl(void *arg) 700{ 701 struct nfsclclient *clp; 702 struct thread *td; 703 704 clp = (struct nfsclclient *)arg; 705 td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads); 706 nfscl_renewthread(clp, td); 707 kproc_exit(0); 708} 709 710void 711nfscl_start_renewthread(struct nfsclclient *clp) 712{ 713 714 kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0, 715 "nfscl"); 716} 717 718/* 719 * Handle wcc_data. 720 * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr 721 * as the first Op after PutFH. 722 * (For NFSv4, the postop attributes are after the Op, so they can't be 723 * parsed here. A separate call to nfscl_postop_attr() is required.) 724 */ 725int 726nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp, 727 struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff) 728{ 729 u_int32_t *tl; 730 struct nfsnode *np = VTONFS(vp); 731 struct nfsvattr nfsva; 732 int error = 0; 733 734 if (wccflagp != NULL) 735 *wccflagp = 0; 736 if (nd->nd_flag & ND_NFSV3) { 737 *flagp = 0; 738 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 739 if (*tl == newnfs_true) { 740 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 741 if (wccflagp != NULL) { 742 mtx_lock(&np->n_mtx); 743 *wccflagp = (np->n_mtime.tv_sec == 744 fxdr_unsigned(u_int32_t, *(tl + 2)) && 745 np->n_mtime.tv_nsec == 746 fxdr_unsigned(u_int32_t, *(tl + 3))); 747 mtx_unlock(&np->n_mtx); 748 } 749 } 750 error = nfscl_postop_attr(nd, nap, flagp, stuff); 751 if (wccflagp != NULL && *flagp == 0) 752 *wccflagp = 0; 753 } else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) 754 == (ND_NFSV4 | ND_V4WCCATTR)) { 755 error = nfsv4_loadattr(nd, NULL, &nfsva, NULL, 756 NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, 757 NULL, NULL, NULL, NULL, NULL); 758 if (error) 759 return (error); 760 /* 761 * Get rid of Op# and status for next op. 762 */ 763 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 764 if (*++tl) 765 nd->nd_flag |= ND_NOMOREDATA; 766 if (wccflagp != NULL && 767 nfsva.na_vattr.va_mtime.tv_sec != 0) { 768 mtx_lock(&np->n_mtx); 769 *wccflagp = (np->n_mtime.tv_sec == 770 nfsva.na_vattr.va_mtime.tv_sec && 771 np->n_mtime.tv_nsec == 772 nfsva.na_vattr.va_mtime.tv_sec); 773 mtx_unlock(&np->n_mtx); 774 } 775 } 776nfsmout: 777 return (error); 778} 779 780/* 781 * Get postop attributes. 782 */ 783int 784nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp, 785 void *stuff) 786{ 787 u_int32_t *tl; 788 int error = 0; 789 790 *retp = 0; 791 if (nd->nd_flag & ND_NOMOREDATA) 792 return (error); 793 if (nd->nd_flag & ND_NFSV3) { 794 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 795 *retp = fxdr_unsigned(int, *tl); 796 } else if (nd->nd_flag & ND_NFSV4) { 797 /* 798 * For NFSv4, the postop attr are at the end, so no point 799 * in looking if nd_repstat != 0. 800 */ 801 if (!nd->nd_repstat) { 802 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 803 if (*(tl + 1)) 804 /* should never happen since nd_repstat != 0 */ 805 nd->nd_flag |= ND_NOMOREDATA; 806 else 807 *retp = 1; 808 } 809 } else if (!nd->nd_repstat) { 810 /* For NFSv2, the attributes are here iff nd_repstat == 0 */ 811 *retp = 1; 812 } 813 if (*retp) { 814 error = nfsm_loadattr(nd, nap); 815 if (error) 816 *retp = 0; 817 } 818nfsmout: 819 return (error); 820} 821 822/* 823 * Fill in the setable attributes. The full argument indicates whether 824 * to fill in them all or just mode and time. 825 */ 826void 827nfscl_fillsattr(struct nfsrv_descript *nd, struct vattr *vap, 828 struct vnode *vp, int flags, u_int32_t rdev) 829{ 830 u_int32_t *tl; 831 struct nfsv2_sattr *sp; 832 nfsattrbit_t attrbits; 833 834 switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) { 835 case ND_NFSV2: 836 NFSM_BUILD(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 837 if (vap->va_mode == (mode_t)VNOVAL) 838 sp->sa_mode = newnfs_xdrneg1; 839 else 840 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 841 if (vap->va_uid == (uid_t)VNOVAL) 842 sp->sa_uid = newnfs_xdrneg1; 843 else 844 sp->sa_uid = txdr_unsigned(vap->va_uid); 845 if (vap->va_gid == (gid_t)VNOVAL) 846 sp->sa_gid = newnfs_xdrneg1; 847 else 848 sp->sa_gid = txdr_unsigned(vap->va_gid); 849 if (flags & NFSSATTR_SIZE0) 850 sp->sa_size = 0; 851 else if (flags & NFSSATTR_SIZENEG1) 852 sp->sa_size = newnfs_xdrneg1; 853 else if (flags & NFSSATTR_SIZERDEV) 854 sp->sa_size = txdr_unsigned(rdev); 855 else 856 sp->sa_size = txdr_unsigned(vap->va_size); 857 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 858 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 859 break; 860 case ND_NFSV3: 861 if (vap->va_mode != (mode_t)VNOVAL) { 862 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 863 *tl++ = newnfs_true; 864 *tl = txdr_unsigned(vap->va_mode); 865 } else { 866 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 867 *tl = newnfs_false; 868 } 869 if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL) { 870 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 871 *tl++ = newnfs_true; 872 *tl = txdr_unsigned(vap->va_uid); 873 } else { 874 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 875 *tl = newnfs_false; 876 } 877 if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL) { 878 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 879 *tl++ = newnfs_true; 880 *tl = txdr_unsigned(vap->va_gid); 881 } else { 882 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 883 *tl = newnfs_false; 884 } 885 if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL) { 886 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 887 *tl++ = newnfs_true; 888 txdr_hyper(vap->va_size, tl); 889 } else { 890 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 891 *tl = newnfs_false; 892 } 893 if (vap->va_atime.tv_sec != VNOVAL) { 894 if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) { 895 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 896 *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT); 897 txdr_nfsv3time(&vap->va_atime, tl); 898 } else { 899 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 900 *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER); 901 } 902 } else { 903 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 904 *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE); 905 } 906 if (vap->va_mtime.tv_sec != VNOVAL) { 907 if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) { 908 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 909 *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT); 910 txdr_nfsv3time(&vap->va_mtime, tl); 911 } else { 912 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 913 *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER); 914 } 915 } else { 916 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 917 *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE); 918 } 919 break; 920 case ND_NFSV4: 921 NFSZERO_ATTRBIT(&attrbits); 922 if (vap->va_mode != (mode_t)VNOVAL) 923 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_MODE); 924 if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL) 925 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNER); 926 if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL) 927 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNERGROUP); 928 if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL) 929 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 930 if (vap->va_atime.tv_sec != VNOVAL) 931 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESSSET); 932 if (vap->va_mtime.tv_sec != VNOVAL) 933 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFYSET); 934 (void) nfsv4_fillattr(nd, vp->v_mount, vp, NULL, vap, NULL, 0, 935 &attrbits, NULL, NULL, 0, 0, 0, 0, (uint64_t)0); 936 break; 937 } 938} 939 940/* 941 * nfscl_request() - mostly a wrapper for newnfs_request(). 942 */ 943int 944nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p, 945 struct ucred *cred, void *stuff) 946{ 947 int ret, vers; 948 struct nfsmount *nmp; 949 950 nmp = VFSTONFS(vp->v_mount); 951 if (nd->nd_flag & ND_NFSV4) 952 vers = NFS_VER4; 953 else if (nd->nd_flag & ND_NFSV3) 954 vers = NFS_VER3; 955 else 956 vers = NFS_VER2; 957 ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred, 958 NFS_PROG, vers, NULL, 1, NULL, NULL); 959 return (ret); 960} 961 962/* 963 * fill in this bsden's variant of statfs using nfsstatfs. 964 */ 965void 966nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs) 967{ 968 struct statfs *sbp = (struct statfs *)statfs; 969 970 if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) { 971 sbp->f_bsize = NFS_FABLKSIZE; 972 sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE; 973 sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE; 974 /* 975 * Although sf_abytes is uint64_t and f_bavail is int64_t, 976 * the value after dividing by NFS_FABLKSIZE is small 977 * enough that it will fit in 63bits, so it is ok to 978 * assign it to f_bavail without fear that it will become 979 * negative. 980 */ 981 sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE; 982 sbp->f_files = sfp->sf_tfiles; 983 /* Since f_ffree is int64_t, clip it to 63bits. */ 984 if (sfp->sf_ffiles > INT64_MAX) 985 sbp->f_ffree = INT64_MAX; 986 else 987 sbp->f_ffree = sfp->sf_ffiles; 988 } else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) { 989 /* 990 * The type casts to (int32_t) ensure that this code is 991 * compatible with the old NFS client, in that it will 992 * propagate bit31 to the high order bits. This may or may 993 * not be correct for NFSv2, but since it is a legacy 994 * environment, I'd rather retain backwards compatibility. 995 */ 996 sbp->f_bsize = (int32_t)sfp->sf_bsize; 997 sbp->f_blocks = (int32_t)sfp->sf_blocks; 998 sbp->f_bfree = (int32_t)sfp->sf_bfree; 999 sbp->f_bavail = (int32_t)sfp->sf_bavail; 1000 sbp->f_files = 0; 1001 sbp->f_ffree = 0; 1002 } 1003} 1004 1005/* 1006 * Use the fsinfo stuff to update the mount point. 1007 */ 1008void 1009nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp) 1010{ 1011 1012 if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) && 1013 fsp->fs_wtpref >= NFS_FABLKSIZE) 1014 nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) & 1015 ~(NFS_FABLKSIZE - 1); 1016 if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) { 1017 nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1); 1018 if (nmp->nm_wsize == 0) 1019 nmp->nm_wsize = fsp->fs_wtmax; 1020 } 1021 if (nmp->nm_wsize < NFS_FABLKSIZE) 1022 nmp->nm_wsize = NFS_FABLKSIZE; 1023 if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) && 1024 fsp->fs_rtpref >= NFS_FABLKSIZE) 1025 nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) & 1026 ~(NFS_FABLKSIZE - 1); 1027 if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) { 1028 nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1); 1029 if (nmp->nm_rsize == 0) 1030 nmp->nm_rsize = fsp->fs_rtmax; 1031 } 1032 if (nmp->nm_rsize < NFS_FABLKSIZE) 1033 nmp->nm_rsize = NFS_FABLKSIZE; 1034 if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize) 1035 && fsp->fs_dtpref >= NFS_DIRBLKSIZ) 1036 nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) & 1037 ~(NFS_DIRBLKSIZ - 1); 1038 if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) { 1039 nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1); 1040 if (nmp->nm_readdirsize == 0) 1041 nmp->nm_readdirsize = fsp->fs_rtmax; 1042 } 1043 if (nmp->nm_readdirsize < NFS_DIRBLKSIZ) 1044 nmp->nm_readdirsize = NFS_DIRBLKSIZ; 1045 if (fsp->fs_maxfilesize > 0 && 1046 fsp->fs_maxfilesize < nmp->nm_maxfilesize) 1047 nmp->nm_maxfilesize = fsp->fs_maxfilesize; 1048 nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp); 1049 nmp->nm_state |= NFSSTA_GOTFSINFO; 1050} 1051 1052/* 1053 * Lookups source address which should be used to communicate with 1054 * @nmp and stores it inside @pdst. 1055 * 1056 * Returns 0 on success. 1057 */ 1058u_int8_t * 1059nfscl_getmyip(struct nfsmount *nmp, struct in6_addr *paddr, int *isinet6p) 1060{ 1061#if defined(INET6) || defined(INET) 1062 int error, fibnum; 1063 1064 fibnum = curthread->td_proc->p_fibnum; 1065#endif 1066#ifdef INET 1067 if (nmp->nm_nam->sa_family == AF_INET) { 1068 struct sockaddr_in *sin; 1069 struct nhop4_extended nh_ext; 1070 1071 sin = (struct sockaddr_in *)nmp->nm_nam; 1072 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred)); 1073 error = fib4_lookup_nh_ext(fibnum, sin->sin_addr, 0, 0, 1074 &nh_ext); 1075 CURVNET_RESTORE(); 1076 if (error != 0) 1077 return (NULL); 1078 1079 if ((ntohl(nh_ext.nh_src.s_addr) >> IN_CLASSA_NSHIFT) == 1080 IN_LOOPBACKNET) { 1081 /* Ignore loopback addresses */ 1082 return (NULL); 1083 } 1084 1085 *isinet6p = 0; 1086 *((struct in_addr *)paddr) = nh_ext.nh_src; 1087 1088 return (u_int8_t *)paddr; 1089 } 1090#endif 1091#ifdef INET6 1092 if (nmp->nm_nam->sa_family == AF_INET6) { 1093 struct sockaddr_in6 *sin6; 1094 1095 sin6 = (struct sockaddr_in6 *)nmp->nm_nam; 1096 1097 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred)); 1098 error = in6_selectsrc_addr(fibnum, &sin6->sin6_addr, 1099 sin6->sin6_scope_id, NULL, paddr, NULL); 1100 CURVNET_RESTORE(); 1101 if (error != 0) 1102 return (NULL); 1103 1104 if (IN6_IS_ADDR_LOOPBACK(paddr)) 1105 return (NULL); 1106 1107 /* Scope is embedded in */ 1108 *isinet6p = 1; 1109 1110 return (u_int8_t *)paddr; 1111 } 1112#endif 1113 return (NULL); 1114} 1115 1116/* 1117 * Copy NFS uid, gids from the cred structure. 1118 */ 1119void 1120newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr) 1121{ 1122 int i; 1123 1124 KASSERT(cr->cr_ngroups >= 0, 1125 ("newnfs_copyincred: negative cr_ngroups")); 1126 nfscr->nfsc_uid = cr->cr_uid; 1127 nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1); 1128 for (i = 0; i < nfscr->nfsc_ngroups; i++) 1129 nfscr->nfsc_groups[i] = cr->cr_groups[i]; 1130} 1131 1132 1133/* 1134 * Do any client specific initialization. 1135 */ 1136void 1137nfscl_init(void) 1138{ 1139 static int inited = 0; 1140 1141 if (inited) 1142 return; 1143 inited = 1; 1144 nfscl_inited = 1; 1145 ncl_pbuf_freecnt = nswbuf / 2 + 1; 1146} 1147 1148/* 1149 * Check each of the attributes to be set, to ensure they aren't already 1150 * the correct value. Disable setting ones already correct. 1151 */ 1152int 1153nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap) 1154{ 1155 1156 if (vap->va_mode != (mode_t)VNOVAL) { 1157 if (vap->va_mode == nvap->na_mode) 1158 vap->va_mode = (mode_t)VNOVAL; 1159 } 1160 if (vap->va_uid != (uid_t)VNOVAL) { 1161 if (vap->va_uid == nvap->na_uid) 1162 vap->va_uid = (uid_t)VNOVAL; 1163 } 1164 if (vap->va_gid != (gid_t)VNOVAL) { 1165 if (vap->va_gid == nvap->na_gid) 1166 vap->va_gid = (gid_t)VNOVAL; 1167 } 1168 if (vap->va_size != VNOVAL) { 1169 if (vap->va_size == nvap->na_size) 1170 vap->va_size = VNOVAL; 1171 } 1172 1173 /* 1174 * We are normally called with only a partially initialized 1175 * VAP. Since the NFSv3 spec says that server may use the 1176 * file attributes to store the verifier, the spec requires 1177 * us to do a SETATTR RPC. FreeBSD servers store the verifier 1178 * in atime, but we can't really assume that all servers will 1179 * so we ensure that our SETATTR sets both atime and mtime. 1180 * Set the VA_UTIMES_NULL flag for this case, so that 1181 * the server's time will be used. This is needed to 1182 * work around a bug in some Solaris servers, where 1183 * setting the time TOCLIENT causes the Setattr RPC 1184 * to return NFS_OK, but not set va_mode. 1185 */ 1186 if (vap->va_mtime.tv_sec == VNOVAL) { 1187 vfs_timestamp(&vap->va_mtime); 1188 vap->va_vaflags |= VA_UTIMES_NULL; 1189 } 1190 if (vap->va_atime.tv_sec == VNOVAL) 1191 vap->va_atime = vap->va_mtime; 1192 return (1); 1193} 1194 1195/* 1196 * Map nfsv4 errors to errno.h errors. 1197 * The uid and gid arguments are only used for NFSERR_BADOWNER and that 1198 * error should only be returned for the Open, Create and Setattr Ops. 1199 * As such, most calls can just pass in 0 for those arguments. 1200 */ 1201APPLESTATIC int 1202nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid) 1203{ 1204 struct proc *p; 1205 1206 if (error < 10000 || error >= NFSERR_STALEWRITEVERF) 1207 return (error); 1208 if (td != NULL) 1209 p = td->td_proc; 1210 else 1211 p = NULL; 1212 switch (error) { 1213 case NFSERR_BADOWNER: 1214 tprintf(p, LOG_INFO, 1215 "No name and/or group mapping for uid,gid:(%d,%d)\n", 1216 uid, gid); 1217 return (EPERM); 1218 case NFSERR_BADNAME: 1219 case NFSERR_BADCHAR: 1220 printf("nfsv4 char/name not handled by server\n"); 1221 return (ENOENT); 1222 case NFSERR_STALECLIENTID: 1223 case NFSERR_STALESTATEID: 1224 case NFSERR_EXPIRED: 1225 case NFSERR_BADSTATEID: 1226 case NFSERR_BADSESSION: 1227 printf("nfsv4 recover err returned %d\n", error); 1228 return (EIO); 1229 case NFSERR_BADHANDLE: 1230 case NFSERR_SERVERFAULT: 1231 case NFSERR_BADTYPE: 1232 case NFSERR_FHEXPIRED: 1233 case NFSERR_RESOURCE: 1234 case NFSERR_MOVED: 1235 case NFSERR_NOFILEHANDLE: 1236 case NFSERR_MINORVERMISMATCH: 1237 case NFSERR_OLDSTATEID: 1238 case NFSERR_BADSEQID: 1239 case NFSERR_LEASEMOVED: 1240 case NFSERR_RECLAIMBAD: 1241 case NFSERR_BADXDR: 1242 case NFSERR_OPILLEGAL: 1243 printf("nfsv4 client/server protocol prob err=%d\n", 1244 error); 1245 return (EIO); 1246 default: 1247 tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error); 1248 return (EIO); 1249 }; 1250} 1251 1252/* 1253 * Check to see if the process for this owner exists. Return 1 if it doesn't 1254 * and 0 otherwise. 1255 */ 1256int 1257nfscl_procdoesntexist(u_int8_t *own) 1258{ 1259 union { 1260 u_int32_t lval; 1261 u_int8_t cval[4]; 1262 } tl; 1263 struct proc *p; 1264 pid_t pid; 1265 int i, ret = 0; 1266 1267 /* For the single open_owner of all 0 bytes, just return 0. */ 1268 for (i = 0; i < NFSV4CL_LOCKNAMELEN; i++) 1269 if (own[i] != 0) 1270 break; 1271 if (i == NFSV4CL_LOCKNAMELEN) 1272 return (0); 1273 1274 tl.cval[0] = *own++; 1275 tl.cval[1] = *own++; 1276 tl.cval[2] = *own++; 1277 tl.cval[3] = *own++; 1278 pid = tl.lval; 1279 p = pfind_locked(pid); 1280 if (p == NULL) 1281 return (1); 1282 if (p->p_stats == NULL) { 1283 PROC_UNLOCK(p); 1284 return (0); 1285 } 1286 tl.cval[0] = *own++; 1287 tl.cval[1] = *own++; 1288 tl.cval[2] = *own++; 1289 tl.cval[3] = *own++; 1290 if (tl.lval != p->p_stats->p_start.tv_sec) { 1291 ret = 1; 1292 } else { 1293 tl.cval[0] = *own++; 1294 tl.cval[1] = *own++; 1295 tl.cval[2] = *own++; 1296 tl.cval[3] = *own; 1297 if (tl.lval != p->p_stats->p_start.tv_usec) 1298 ret = 1; 1299 } 1300 PROC_UNLOCK(p); 1301 return (ret); 1302} 1303 1304/* 1305 * - nfs pseudo system call for the client 1306 */ 1307/* 1308 * MPSAFE 1309 */ 1310static int 1311nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap) 1312{ 1313 struct file *fp; 1314 struct nfscbd_args nfscbdarg; 1315 struct nfsd_nfscbd_args nfscbdarg2; 1316 struct nameidata nd; 1317 struct nfscl_dumpmntopts dumpmntopts; 1318 cap_rights_t rights; 1319 char *buf; 1320 int error; 1321 struct mount *mp; 1322 struct nfsmount *nmp; 1323 1324 if (uap->flag & NFSSVC_CBADDSOCK) { 1325 error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg)); 1326 if (error) 1327 return (error); 1328 /* 1329 * Since we don't know what rights might be required, 1330 * pretend that we need them all. It is better to be too 1331 * careful than too reckless. 1332 */ 1333 error = fget(td, nfscbdarg.sock, 1334 cap_rights_init(&rights, CAP_SOCK_CLIENT), &fp); 1335 if (error) 1336 return (error); 1337 if (fp->f_type != DTYPE_SOCKET) { 1338 fdrop(fp, td); 1339 return (EPERM); 1340 } 1341 error = nfscbd_addsock(fp); 1342 fdrop(fp, td); 1343 if (!error && nfscl_enablecallb == 0) { 1344 nfsv4_cbport = nfscbdarg.port; 1345 nfscl_enablecallb = 1; 1346 } 1347 } else if (uap->flag & NFSSVC_NFSCBD) { 1348 if (uap->argp == NULL) 1349 return (EINVAL); 1350 error = copyin(uap->argp, (caddr_t)&nfscbdarg2, 1351 sizeof(nfscbdarg2)); 1352 if (error) 1353 return (error); 1354 error = nfscbd_nfsd(td, &nfscbdarg2); 1355 } else if (uap->flag & NFSSVC_DUMPMNTOPTS) { 1356 error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts)); 1357 if (error == 0 && (dumpmntopts.ndmnt_blen < 256 || 1358 dumpmntopts.ndmnt_blen > 1024)) 1359 error = EINVAL; 1360 if (error == 0) 1361 error = nfsrv_lookupfilename(&nd, 1362 dumpmntopts.ndmnt_fname, td); 1363 if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, 1364 "nfs") != 0) { 1365 vput(nd.ni_vp); 1366 error = EINVAL; 1367 } 1368 if (error == 0) { 1369 buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK); 1370 nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf, 1371 dumpmntopts.ndmnt_blen); 1372 vput(nd.ni_vp); 1373 error = copyout(buf, dumpmntopts.ndmnt_buf, 1374 dumpmntopts.ndmnt_blen); 1375 free(buf, M_TEMP); 1376 } 1377 } else if (uap->flag & NFSSVC_FORCEDISM) { 1378 buf = malloc(MNAMELEN + 1, M_TEMP, M_WAITOK); 1379 error = copyinstr(uap->argp, buf, MNAMELEN + 1, NULL); 1380 if (error == 0) { 1381 nmp = NULL; 1382 mtx_lock(&mountlist_mtx); 1383 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 1384 if (strcmp(mp->mnt_stat.f_mntonname, buf) == 1385 0 && strcmp(mp->mnt_stat.f_fstypename, 1386 "nfs") == 0 && mp->mnt_data != NULL) { 1387 nmp = VFSTONFS(mp); 1388 mtx_lock(&nmp->nm_mtx); 1389 if ((nmp->nm_privflag & 1390 NFSMNTP_FORCEDISM) == 0) { 1391 nmp->nm_privflag |= 1392 (NFSMNTP_FORCEDISM | 1393 NFSMNTP_CANCELRPCS); 1394 mtx_unlock(&nmp->nm_mtx); 1395 } else { 1396 nmp = NULL; 1397 mtx_unlock(&nmp->nm_mtx); 1398 } 1399 break; 1400 } 1401 } 1402 mtx_unlock(&mountlist_mtx); 1403 1404 if (nmp != NULL) { 1405 /* 1406 * Call newnfs_nmcancelreqs() to cause 1407 * any RPCs in progress on the mount point to 1408 * fail. 1409 * This will cause any process waiting for an 1410 * RPC to complete while holding a vnode lock 1411 * on the mounted-on vnode (such as "df" or 1412 * a non-forced "umount") to fail. 1413 * This will unlock the mounted-on vnode so 1414 * a forced dismount can succeed. 1415 * Then clear NFSMNTP_CANCELRPCS and wakeup(), 1416 * so that nfs_unmount() can complete. 1417 */ 1418 newnfs_nmcancelreqs(nmp); 1419 mtx_lock(&nmp->nm_mtx); 1420 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 1421 wakeup(nmp); 1422 mtx_unlock(&nmp->nm_mtx); 1423 } else 1424 error = EINVAL; 1425 } 1426 free(buf, M_TEMP); 1427 } else { 1428 error = EINVAL; 1429 } 1430 return (error); 1431} 1432 1433extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *); 1434 1435/* 1436 * Called once to initialize data structures... 1437 */ 1438static int 1439nfscl_modevent(module_t mod, int type, void *data) 1440{ 1441 int error = 0; 1442 static int loaded = 0; 1443 1444 switch (type) { 1445 case MOD_LOAD: 1446 if (loaded) 1447 return (0); 1448 newnfs_portinit(); 1449 mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF); 1450 nfscl_init(); 1451 NFSD_LOCK(); 1452 nfsrvd_cbinit(0); 1453 NFSD_UNLOCK(); 1454 ncl_call_invalcaches = ncl_invalcaches; 1455 nfsd_call_nfscl = nfssvc_nfscl; 1456 loaded = 1; 1457 break; 1458 1459 case MOD_UNLOAD: 1460 if (nfs_numnfscbd != 0) { 1461 error = EBUSY; 1462 break; 1463 } 1464 1465 /* 1466 * XXX: Unloading of nfscl module is unsupported. 1467 */ 1468#if 0 1469 ncl_call_invalcaches = NULL; 1470 nfsd_call_nfscl = NULL; 1471 /* and get rid of the mutexes */ 1472 mtx_destroy(&ncl_iod_mutex); 1473 loaded = 0; 1474 break; 1475#else 1476 /* FALLTHROUGH */ 1477#endif 1478 default: 1479 error = EOPNOTSUPP; 1480 break; 1481 } 1482 return error; 1483} 1484static moduledata_t nfscl_mod = { 1485 "nfscl", 1486 nfscl_modevent, 1487 NULL, 1488}; 1489DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST); 1490 1491/* So that loader and kldload(2) can find us, wherever we are.. */ 1492MODULE_VERSION(nfscl, 1); 1493MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1); 1494MODULE_DEPEND(nfscl, krpc, 1, 1, 1); 1495MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1); 1496MODULE_DEPEND(nfscl, nfslock, 1, 1, 1); 1497 1498