nfs_subs.c revision 247502
1/*- 2 * Copyright (c) 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * Rick Macklem at The University of Guelph. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 4. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * @(#)nfs_subs.c 8.8 (Berkeley) 5/22/95 33 */ 34 35#include <sys/cdefs.h> 36__FBSDID("$FreeBSD: stable/9/sys/nfsclient/nfs_subs.c 247502 2013-02-28 21:57:38Z jhb $"); 37 38/* 39 * These functions support the macros and help fiddle mbuf chains for 40 * the nfs op functions. They do things like create the rpc header and 41 * copy data between mbuf chains and uio lists. 42 */ 43 44#include "opt_kdtrace.h" 45 46#include <sys/param.h> 47#include <sys/systm.h> 48#include <sys/kernel.h> 49#include <sys/bio.h> 50#include <sys/buf.h> 51#include <sys/proc.h> 52#include <sys/mount.h> 53#include <sys/vnode.h> 54#include <sys/namei.h> 55#include <sys/mbuf.h> 56#include <sys/socket.h> 57#include <sys/stat.h> 58#include <sys/malloc.h> 59#include <sys/sysent.h> 60#include <sys/syscall.h> 61#include <sys/sysproto.h> 62#include <sys/taskqueue.h> 63 64#include <vm/vm.h> 65#include <vm/vm_object.h> 66#include <vm/vm_extern.h> 67#include <vm/uma.h> 68 69#include <nfs/nfsproto.h> 70#include <nfsclient/nfs.h> 71#include <nfsclient/nfsnode.h> 72#include <nfs/nfs_kdtrace.h> 73#include <nfs/xdr_subs.h> 74#include <nfsclient/nfsm_subs.h> 75#include <nfsclient/nfsmount.h> 76 77#include <netinet/in.h> 78 79/* 80 * Note that stdarg.h and the ANSI style va_start macro is used for both 81 * ANSI and traditional C compilers. 82 */ 83#include <machine/stdarg.h> 84 85#ifdef KDTRACE_HOOKS 86dtrace_nfsclient_attrcache_flush_probe_func_t 87 dtrace_nfsclient_attrcache_flush_done_probe; 88uint32_t nfsclient_attrcache_flush_done_id; 89 90dtrace_nfsclient_attrcache_get_hit_probe_func_t 91 dtrace_nfsclient_attrcache_get_hit_probe; 92uint32_t nfsclient_attrcache_get_hit_id; 93 94dtrace_nfsclient_attrcache_get_miss_probe_func_t 95 dtrace_nfsclient_attrcache_get_miss_probe; 96uint32_t nfsclient_attrcache_get_miss_id; 97 98dtrace_nfsclient_attrcache_load_probe_func_t 99 dtrace_nfsclient_attrcache_load_done_probe; 100uint32_t nfsclient_attrcache_load_done_id; 101#endif /* !KDTRACE_HOOKS */ 102 103/* 104 * Data items converted to xdr at startup, since they are constant 105 * This is kinda hokey, but may save a little time doing byte swaps 106 */ 107u_int32_t nfs_xdrneg1; 108u_int32_t nfs_true, nfs_false; 109 110/* And other global data */ 111static u_int32_t nfs_xid = 0; 112static enum vtype nv2tov_type[8]= { 113 VNON, VREG, VDIR, VBLK, VCHR, VLNK, VNON, VNON 114}; 115 116int nfs_ticks; 117int nfs_pbuf_freecnt = -1; /* start out unlimited */ 118 119struct nfs_bufq nfs_bufq; 120static struct mtx nfs_xid_mtx; 121struct task nfs_nfsiodnew_task; 122 123/* 124 * and the reverse mapping from generic to Version 2 procedure numbers 125 */ 126int nfsv2_procid[NFS_NPROCS] = { 127 NFSV2PROC_NULL, 128 NFSV2PROC_GETATTR, 129 NFSV2PROC_SETATTR, 130 NFSV2PROC_LOOKUP, 131 NFSV2PROC_NOOP, 132 NFSV2PROC_READLINK, 133 NFSV2PROC_READ, 134 NFSV2PROC_WRITE, 135 NFSV2PROC_CREATE, 136 NFSV2PROC_MKDIR, 137 NFSV2PROC_SYMLINK, 138 NFSV2PROC_CREATE, 139 NFSV2PROC_REMOVE, 140 NFSV2PROC_RMDIR, 141 NFSV2PROC_RENAME, 142 NFSV2PROC_LINK, 143 NFSV2PROC_READDIR, 144 NFSV2PROC_NOOP, 145 NFSV2PROC_STATFS, 146 NFSV2PROC_NOOP, 147 NFSV2PROC_NOOP, 148 NFSV2PROC_NOOP, 149 NFSV2PROC_NOOP, 150}; 151 152LIST_HEAD(nfsnodehashhead, nfsnode); 153 154u_int32_t 155nfs_xid_gen(void) 156{ 157 uint32_t xid; 158 159 mtx_lock(&nfs_xid_mtx); 160 161 /* Get a pretty random xid to start with */ 162 if (!nfs_xid) 163 nfs_xid = random(); 164 /* 165 * Skip zero xid if it should ever happen. 166 */ 167 if (++nfs_xid == 0) 168 nfs_xid++; 169 xid = nfs_xid; 170 mtx_unlock(&nfs_xid_mtx); 171 return xid; 172} 173 174/* 175 * Create the header for an rpc request packet 176 * The hsiz is the size of the rest of the nfs request header. 177 * (just used to decide if a cluster is a good idea) 178 */ 179struct mbuf * 180nfsm_reqhead(struct vnode *vp, u_long procid, int hsiz) 181{ 182 struct mbuf *mb; 183 184 MGET(mb, M_WAIT, MT_DATA); 185 if (hsiz >= MINCLSIZE) 186 MCLGET(mb, M_WAIT); 187 mb->m_len = 0; 188 return (mb); 189} 190 191/* 192 * copies a uio scatter/gather list to an mbuf chain. 193 * NOTE: can ony handle iovcnt == 1 194 */ 195int 196nfsm_uiotombuf(struct uio *uiop, struct mbuf **mq, int siz, caddr_t *bpos) 197{ 198 char *uiocp; 199 struct mbuf *mp, *mp2; 200 int xfer, left, mlen; 201 int uiosiz, clflg, rem; 202 char *cp; 203 204 KASSERT(uiop->uio_iovcnt == 1, ("nfsm_uiotombuf: iovcnt != 1")); 205 206 if (siz > MLEN) /* or should it >= MCLBYTES ?? */ 207 clflg = 1; 208 else 209 clflg = 0; 210 rem = nfsm_rndup(siz)-siz; 211 mp = mp2 = *mq; 212 while (siz > 0) { 213 left = uiop->uio_iov->iov_len; 214 uiocp = uiop->uio_iov->iov_base; 215 if (left > siz) 216 left = siz; 217 uiosiz = left; 218 while (left > 0) { 219 mlen = M_TRAILINGSPACE(mp); 220 if (mlen == 0) { 221 MGET(mp, M_WAIT, MT_DATA); 222 if (clflg) 223 MCLGET(mp, M_WAIT); 224 mp->m_len = 0; 225 mp2->m_next = mp; 226 mp2 = mp; 227 mlen = M_TRAILINGSPACE(mp); 228 } 229 xfer = (left > mlen) ? mlen : left; 230#ifdef notdef 231 /* Not Yet.. */ 232 if (uiop->uio_iov->iov_op != NULL) 233 (*(uiop->uio_iov->iov_op)) 234 (uiocp, mtod(mp, caddr_t)+mp->m_len, xfer); 235 else 236#endif 237 if (uiop->uio_segflg == UIO_SYSSPACE) 238 bcopy(uiocp, mtod(mp, caddr_t)+mp->m_len, xfer); 239 else 240 copyin(uiocp, mtod(mp, caddr_t)+mp->m_len, xfer); 241 mp->m_len += xfer; 242 left -= xfer; 243 uiocp += xfer; 244 uiop->uio_offset += xfer; 245 uiop->uio_resid -= xfer; 246 } 247 uiop->uio_iov->iov_base = 248 (char *)uiop->uio_iov->iov_base + uiosiz; 249 uiop->uio_iov->iov_len -= uiosiz; 250 siz -= uiosiz; 251 } 252 if (rem > 0) { 253 if (rem > M_TRAILINGSPACE(mp)) { 254 MGET(mp, M_WAIT, MT_DATA); 255 mp->m_len = 0; 256 mp2->m_next = mp; 257 } 258 cp = mtod(mp, caddr_t)+mp->m_len; 259 for (left = 0; left < rem; left++) 260 *cp++ = '\0'; 261 mp->m_len += rem; 262 *bpos = cp; 263 } else 264 *bpos = mtod(mp, caddr_t)+mp->m_len; 265 *mq = mp; 266 return (0); 267} 268 269/* 270 * Copy a string into mbufs for the hard cases... 271 */ 272int 273nfsm_strtmbuf(struct mbuf **mb, char **bpos, const char *cp, long siz) 274{ 275 struct mbuf *m1 = NULL, *m2; 276 long left, xfer, len, tlen; 277 u_int32_t *tl; 278 int putsize; 279 280 putsize = 1; 281 m2 = *mb; 282 left = M_TRAILINGSPACE(m2); 283 if (left > 0) { 284 tl = ((u_int32_t *)(*bpos)); 285 *tl++ = txdr_unsigned(siz); 286 putsize = 0; 287 left -= NFSX_UNSIGNED; 288 m2->m_len += NFSX_UNSIGNED; 289 if (left > 0) { 290 bcopy(cp, (caddr_t) tl, left); 291 siz -= left; 292 cp += left; 293 m2->m_len += left; 294 left = 0; 295 } 296 } 297 /* Loop around adding mbufs */ 298 while (siz > 0) { 299 MGET(m1, M_WAIT, MT_DATA); 300 if (siz > MLEN) 301 MCLGET(m1, M_WAIT); 302 m1->m_len = NFSMSIZ(m1); 303 m2->m_next = m1; 304 m2 = m1; 305 tl = mtod(m1, u_int32_t *); 306 tlen = 0; 307 if (putsize) { 308 *tl++ = txdr_unsigned(siz); 309 m1->m_len -= NFSX_UNSIGNED; 310 tlen = NFSX_UNSIGNED; 311 putsize = 0; 312 } 313 if (siz < m1->m_len) { 314 len = nfsm_rndup(siz); 315 xfer = siz; 316 if (xfer < len) 317 *(tl+(xfer>>2)) = 0; 318 } else { 319 xfer = len = m1->m_len; 320 } 321 bcopy(cp, (caddr_t) tl, xfer); 322 m1->m_len = len+tlen; 323 siz -= xfer; 324 cp += xfer; 325 } 326 *mb = m1; 327 *bpos = mtod(m1, caddr_t)+m1->m_len; 328 return (0); 329} 330 331/* 332 * Called once to initialize data structures... 333 */ 334int 335nfs_init(struct vfsconf *vfsp) 336{ 337 int i; 338 339 nfsmount_zone = uma_zcreate("NFSMOUNT", sizeof(struct nfsmount), 340 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 341 nfs_true = txdr_unsigned(TRUE); 342 nfs_false = txdr_unsigned(FALSE); 343 nfs_xdrneg1 = txdr_unsigned(-1); 344 nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000; 345 if (nfs_ticks < 1) 346 nfs_ticks = 1; 347 /* Ensure async daemons disabled */ 348 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) { 349 nfs_iodwant[i] = NFSIOD_NOT_AVAILABLE; 350 nfs_iodmount[i] = NULL; 351 } 352 nfs_nhinit(); /* Init the nfsnode table */ 353 354 /* 355 * Initialize reply list and start timer 356 */ 357 mtx_init(&nfs_iod_mtx, "NFS iod lock", NULL, MTX_DEF); 358 mtx_init(&nfs_xid_mtx, "NFS xid lock", NULL, MTX_DEF); 359 TASK_INIT(&nfs_nfsiodnew_task, 0, nfs_nfsiodnew_tq, NULL); 360 361 nfs_pbuf_freecnt = nswbuf / 2 + 1; 362 363 return (0); 364} 365 366int 367nfs_uninit(struct vfsconf *vfsp) 368{ 369 int i; 370 371 /* 372 * Tell all nfsiod processes to exit. Clear nfs_iodmax, and wakeup 373 * any sleeping nfsiods so they check nfs_iodmax and exit. 374 * Drain nfsiodnew task before we wait for them to finish. 375 */ 376 mtx_lock(&nfs_iod_mtx); 377 nfs_iodmax = 0; 378 mtx_unlock(&nfs_iod_mtx); 379 taskqueue_drain(taskqueue_thread, &nfs_nfsiodnew_task); 380 mtx_lock(&nfs_iod_mtx); 381 for (i = 0; i < nfs_numasync; i++) 382 if (nfs_iodwant[i] == NFSIOD_AVAILABLE) 383 wakeup(&nfs_iodwant[i]); 384 /* The last nfsiod to exit will wake us up when nfs_numasync hits 0 */ 385 while (nfs_numasync) 386 msleep(&nfs_numasync, &nfs_iod_mtx, PWAIT, "ioddie", 0); 387 mtx_unlock(&nfs_iod_mtx); 388 nfs_nhuninit(); 389 uma_zdestroy(nfsmount_zone); 390 return (0); 391} 392 393void 394nfs_dircookie_lock(struct nfsnode *np) 395{ 396 mtx_lock(&np->n_mtx); 397 while (np->n_flag & NDIRCOOKIELK) 398 (void) msleep(&np->n_flag, &np->n_mtx, PZERO, "nfsdirlk", 0); 399 np->n_flag |= NDIRCOOKIELK; 400 mtx_unlock(&np->n_mtx); 401} 402 403void 404nfs_dircookie_unlock(struct nfsnode *np) 405{ 406 mtx_lock(&np->n_mtx); 407 np->n_flag &= ~NDIRCOOKIELK; 408 wakeup(&np->n_flag); 409 mtx_unlock(&np->n_mtx); 410} 411 412int 413nfs_upgrade_vnlock(struct vnode *vp) 414{ 415 int old_lock; 416 417 ASSERT_VOP_LOCKED(vp, "nfs_upgrade_vnlock"); 418 old_lock = VOP_ISLOCKED(vp); 419 if (old_lock != LK_EXCLUSIVE) { 420 KASSERT(old_lock == LK_SHARED, 421 ("nfs_upgrade_vnlock: wrong old_lock %d", old_lock)); 422 /* Upgrade to exclusive lock, this might block */ 423 vn_lock(vp, LK_UPGRADE | LK_RETRY); 424 } 425 return (old_lock); 426} 427 428void 429nfs_downgrade_vnlock(struct vnode *vp, int old_lock) 430{ 431 if (old_lock != LK_EXCLUSIVE) { 432 KASSERT(old_lock == LK_SHARED, ("wrong old_lock %d", old_lock)); 433 /* Downgrade from exclusive lock. */ 434 vn_lock(vp, LK_DOWNGRADE | LK_RETRY); 435 } 436} 437 438void 439nfs_printf(const char *fmt, ...) 440{ 441 va_list ap; 442 443 mtx_lock(&Giant); 444 va_start(ap, fmt); 445 vprintf(fmt, ap); 446 va_end(ap); 447 mtx_unlock(&Giant); 448} 449 450/* 451 * Attribute cache routines. 452 * nfs_loadattrcache() - loads or updates the cache contents from attributes 453 * that are on the mbuf list 454 * nfs_getattrcache() - returns valid attributes if found in cache, returns 455 * error otherwise 456 */ 457 458/* 459 * Load the attribute cache (that lives in the nfsnode entry) with 460 * the values on the mbuf list and 461 * Iff vap not NULL 462 * copy the attributes to *vaper 463 */ 464int 465nfs_loadattrcache(struct vnode **vpp, struct mbuf **mdp, caddr_t *dposp, 466 struct vattr *vaper, int dontshrink) 467{ 468 struct vnode *vp = *vpp; 469 struct vattr *vap; 470 struct nfs_fattr *fp; 471 struct nfsnode *np = NULL; 472 int32_t t1; 473 caddr_t cp2; 474 int rdev; 475 struct mbuf *md; 476 enum vtype vtyp; 477 u_short vmode; 478 struct timespec mtime, mtime_save; 479 int v3 = NFS_ISV3(vp); 480 int error = 0; 481 482 md = *mdp; 483 t1 = (mtod(md, caddr_t) + md->m_len) - *dposp; 484 cp2 = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, M_WAIT); 485 if (cp2 == NULL) { 486 error = EBADRPC; 487 goto out; 488 } 489 fp = (struct nfs_fattr *)cp2; 490 if (v3) { 491 vtyp = nfsv3tov_type(fp->fa_type); 492 vmode = fxdr_unsigned(u_short, fp->fa_mode); 493 rdev = makedev(fxdr_unsigned(int, fp->fa3_rdev.specdata1), 494 fxdr_unsigned(int, fp->fa3_rdev.specdata2)); 495 fxdr_nfsv3time(&fp->fa3_mtime, &mtime); 496 } else { 497 vtyp = nfsv2tov_type(fp->fa_type); 498 vmode = fxdr_unsigned(u_short, fp->fa_mode); 499 /* 500 * XXX 501 * 502 * The duplicate information returned in fa_type and fa_mode 503 * is an ambiguity in the NFS version 2 protocol. 504 * 505 * VREG should be taken literally as a regular file. If a 506 * server intents to return some type information differently 507 * in the upper bits of the mode field (e.g. for sockets, or 508 * FIFOs), NFSv2 mandates fa_type to be VNON. Anyway, we 509 * leave the examination of the mode bits even in the VREG 510 * case to avoid breakage for bogus servers, but we make sure 511 * that there are actually type bits set in the upper part of 512 * fa_mode (and failing that, trust the va_type field). 513 * 514 * NFSv3 cleared the issue, and requires fa_mode to not 515 * contain any type information (while also introduing sockets 516 * and FIFOs for fa_type). 517 */ 518 if (vtyp == VNON || (vtyp == VREG && (vmode & S_IFMT) != 0)) 519 vtyp = IFTOVT(vmode); 520 rdev = fxdr_unsigned(int32_t, fp->fa2_rdev); 521 fxdr_nfsv2time(&fp->fa2_mtime, &mtime); 522 523 /* 524 * Really ugly NFSv2 kludge. 525 */ 526 if (vtyp == VCHR && rdev == 0xffffffff) 527 vtyp = VFIFO; 528 } 529 530 /* 531 * If v_type == VNON it is a new node, so fill in the v_type, 532 * n_mtime fields. Check to see if it represents a special 533 * device, and if so, check for a possible alias. Once the 534 * correct vnode has been obtained, fill in the rest of the 535 * information. 536 */ 537 np = VTONFS(vp); 538 mtx_lock(&np->n_mtx); 539 if (vp->v_type != vtyp) { 540 vp->v_type = vtyp; 541 if (vp->v_type == VFIFO) 542 vp->v_op = &nfs_fifoops; 543 np->n_mtime = mtime; 544 } 545 vap = &np->n_vattr; 546 vap->va_type = vtyp; 547 vap->va_mode = (vmode & 07777); 548 vap->va_rdev = rdev; 549 mtime_save = vap->va_mtime; 550 vap->va_mtime = mtime; 551 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 552 if (v3) { 553 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink); 554 vap->va_uid = fxdr_unsigned(uid_t, fp->fa_uid); 555 vap->va_gid = fxdr_unsigned(gid_t, fp->fa_gid); 556 vap->va_size = fxdr_hyper(&fp->fa3_size); 557 vap->va_blocksize = NFS_FABLKSIZE; 558 vap->va_bytes = fxdr_hyper(&fp->fa3_used); 559 vap->va_fileid = fxdr_unsigned(int32_t, 560 fp->fa3_fileid.nfsuquad[1]); 561 fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime); 562 fxdr_nfsv3time(&fp->fa3_ctime, &vap->va_ctime); 563 vap->va_flags = 0; 564 vap->va_filerev = 0; 565 } else { 566 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink); 567 vap->va_uid = fxdr_unsigned(uid_t, fp->fa_uid); 568 vap->va_gid = fxdr_unsigned(gid_t, fp->fa_gid); 569 vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size); 570 vap->va_blocksize = fxdr_unsigned(int32_t, fp->fa2_blocksize); 571 vap->va_bytes = (u_quad_t)fxdr_unsigned(int32_t, fp->fa2_blocks) 572 * NFS_FABLKSIZE; 573 vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid); 574 fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime); 575 vap->va_flags = 0; 576 vap->va_ctime.tv_sec = fxdr_unsigned(u_int32_t, 577 fp->fa2_ctime.nfsv2_sec); 578 vap->va_ctime.tv_nsec = 0; 579 vap->va_gen = fxdr_unsigned(u_int32_t, fp->fa2_ctime.nfsv2_usec); 580 vap->va_filerev = 0; 581 } 582 np->n_attrstamp = time_second; 583 if (vap->va_size != np->n_size) { 584 if (vap->va_type == VREG) { 585 if (dontshrink && vap->va_size < np->n_size) { 586 /* 587 * We've been told not to shrink the file; 588 * zero np->n_attrstamp to indicate that 589 * the attributes are stale. 590 */ 591 vap->va_size = np->n_size; 592 np->n_attrstamp = 0; 593 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 594 } else if (np->n_flag & NMODIFIED) { 595 /* 596 * We've modified the file: Use the larger 597 * of our size, and the server's size. 598 */ 599 if (vap->va_size < np->n_size) { 600 vap->va_size = np->n_size; 601 } else { 602 np->n_size = vap->va_size; 603 np->n_flag |= NSIZECHANGED; 604 } 605 } else { 606 np->n_size = vap->va_size; 607 np->n_flag |= NSIZECHANGED; 608 } 609 vnode_pager_setsize(vp, np->n_size); 610 } else { 611 np->n_size = vap->va_size; 612 } 613 } 614 /* 615 * The following checks are added to prevent a race between (say) 616 * a READDIR+ and a WRITE. 617 * READDIR+, WRITE requests sent out. 618 * READDIR+ resp, WRITE resp received on client. 619 * However, the WRITE resp was handled before the READDIR+ resp 620 * causing the post op attrs from the write to be loaded first 621 * and the attrs from the READDIR+ to be loaded later. If this 622 * happens, we have stale attrs loaded into the attrcache. 623 * We detect this by for the mtime moving back. We invalidate the 624 * attrcache when this happens. 625 */ 626 if (timespeccmp(&mtime_save, &vap->va_mtime, >)) { 627 /* Size changed or mtime went backwards */ 628 np->n_attrstamp = 0; 629 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 630 } 631 if (vaper != NULL) { 632 bcopy((caddr_t)vap, (caddr_t)vaper, sizeof(*vap)); 633 if (np->n_flag & NCHG) { 634 if (np->n_flag & NACC) 635 vaper->va_atime = np->n_atim; 636 if (np->n_flag & NUPD) 637 vaper->va_mtime = np->n_mtim; 638 } 639 } 640 641#ifdef KDTRACE_HOOKS 642 if (np->n_attrstamp != 0) 643 KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, &np->n_vattr, 0); 644#endif 645 mtx_unlock(&np->n_mtx); 646out: 647#ifdef KDTRACE_HOOKS 648 if (error) 649 KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, NULL, error); 650#endif 651 return (error); 652} 653 654#ifdef NFS_ACDEBUG 655#include <sys/sysctl.h> 656SYSCTL_DECL(_vfs_oldnfs); 657static int nfs_acdebug; 658SYSCTL_INT(_vfs_oldnfs, OID_AUTO, acdebug, CTLFLAG_RW, &nfs_acdebug, 0, 659 "Toggle acdebug (attribute cache debug) flag"); 660#endif 661 662/* 663 * Check the time stamp 664 * If the cache is valid, copy contents to *vap and return 0 665 * otherwise return an error 666 */ 667int 668nfs_getattrcache(struct vnode *vp, struct vattr *vaper) 669{ 670 struct nfsnode *np; 671 struct vattr *vap; 672 struct nfsmount *nmp; 673 int timeo; 674 675 np = VTONFS(vp); 676 vap = &np->n_vattr; 677 nmp = VFSTONFS(vp->v_mount); 678#ifdef NFS_ACDEBUG 679 mtx_lock(&Giant); /* nfs_printf() */ 680#endif 681 mtx_lock(&np->n_mtx); 682 /* XXX n_mtime doesn't seem to be updated on a miss-and-reload */ 683 timeo = (time_second - np->n_mtime.tv_sec) / 10; 684 685#ifdef NFS_ACDEBUG 686 if (nfs_acdebug>1) 687 nfs_printf("nfs_getattrcache: initial timeo = %d\n", timeo); 688#endif 689 690 if (vap->va_type == VDIR) { 691 if ((np->n_flag & NMODIFIED) || timeo < nmp->nm_acdirmin) 692 timeo = nmp->nm_acdirmin; 693 else if (timeo > nmp->nm_acdirmax) 694 timeo = nmp->nm_acdirmax; 695 } else { 696 if ((np->n_flag & NMODIFIED) || timeo < nmp->nm_acregmin) 697 timeo = nmp->nm_acregmin; 698 else if (timeo > nmp->nm_acregmax) 699 timeo = nmp->nm_acregmax; 700 } 701 702#ifdef NFS_ACDEBUG 703 if (nfs_acdebug > 2) 704 nfs_printf("acregmin %d; acregmax %d; acdirmin %d; acdirmax %d\n", 705 nmp->nm_acregmin, nmp->nm_acregmax, 706 nmp->nm_acdirmin, nmp->nm_acdirmax); 707 708 if (nfs_acdebug) 709 nfs_printf("nfs_getattrcache: age = %d; final timeo = %d\n", 710 (time_second - np->n_attrstamp), timeo); 711#endif 712 713 if ((time_second - np->n_attrstamp) >= timeo) { 714 nfsstats.attrcache_misses++; 715 mtx_unlock(&np->n_mtx); 716#ifdef NFS_ACDEBUG 717 mtx_unlock(&Giant); /* nfs_printf() */ 718#endif 719 KDTRACE_NFS_ATTRCACHE_GET_MISS(vp); 720 return (ENOENT); 721 } 722 nfsstats.attrcache_hits++; 723 if (vap->va_size != np->n_size) { 724 if (vap->va_type == VREG) { 725 if (np->n_flag & NMODIFIED) { 726 if (vap->va_size < np->n_size) 727 vap->va_size = np->n_size; 728 else 729 np->n_size = vap->va_size; 730 } else { 731 np->n_size = vap->va_size; 732 } 733 vnode_pager_setsize(vp, np->n_size); 734 } else { 735 np->n_size = vap->va_size; 736 } 737 } 738 bcopy((caddr_t)vap, (caddr_t)vaper, sizeof(struct vattr)); 739 if (np->n_flag & NCHG) { 740 if (np->n_flag & NACC) 741 vaper->va_atime = np->n_atim; 742 if (np->n_flag & NUPD) 743 vaper->va_mtime = np->n_mtim; 744 } 745 mtx_unlock(&np->n_mtx); 746#ifdef NFS_ACDEBUG 747 mtx_unlock(&Giant); /* nfs_printf() */ 748#endif 749 KDTRACE_NFS_ATTRCACHE_GET_HIT(vp, vap); 750 return (0); 751} 752 753/* 754 * Purge all cached information about an NFS vnode including name 755 * cache entries, the attribute cache, and the access cache. This is 756 * called when an NFS request for a node fails with a stale 757 * filehandle. 758 */ 759void 760nfs_purgecache(struct vnode *vp) 761{ 762 struct nfsnode *np; 763 int i; 764 765 np = VTONFS(vp); 766 cache_purge(vp); 767 mtx_lock(&np->n_mtx); 768 np->n_attrstamp = 0; 769 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 770 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) 771 np->n_accesscache[i].stamp = 0; 772 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp); 773 mtx_unlock(&np->n_mtx); 774} 775 776static nfsuint64 nfs_nullcookie = { { 0, 0 } }; 777/* 778 * This function finds the directory cookie that corresponds to the 779 * logical byte offset given. 780 */ 781nfsuint64 * 782nfs_getcookie(struct nfsnode *np, off_t off, int add) 783{ 784 struct nfsdmap *dp, *dp2; 785 int pos; 786 nfsuint64 *retval = NULL; 787 788 pos = (uoff_t)off / NFS_DIRBLKSIZ; 789 if (pos == 0 || off < 0) { 790 KASSERT(!add, ("nfs getcookie add at <= 0")); 791 return (&nfs_nullcookie); 792 } 793 pos--; 794 dp = LIST_FIRST(&np->n_cookies); 795 if (!dp) { 796 if (add) { 797 dp = malloc(sizeof (struct nfsdmap), 798 M_NFSDIROFF, M_WAITOK); 799 dp->ndm_eocookie = 0; 800 LIST_INSERT_HEAD(&np->n_cookies, dp, ndm_list); 801 } else 802 goto out; 803 } 804 while (pos >= NFSNUMCOOKIES) { 805 pos -= NFSNUMCOOKIES; 806 if (LIST_NEXT(dp, ndm_list)) { 807 if (!add && dp->ndm_eocookie < NFSNUMCOOKIES && 808 pos >= dp->ndm_eocookie) 809 goto out; 810 dp = LIST_NEXT(dp, ndm_list); 811 } else if (add) { 812 dp2 = malloc(sizeof (struct nfsdmap), 813 M_NFSDIROFF, M_WAITOK); 814 dp2->ndm_eocookie = 0; 815 LIST_INSERT_AFTER(dp, dp2, ndm_list); 816 dp = dp2; 817 } else 818 goto out; 819 } 820 if (pos >= dp->ndm_eocookie) { 821 if (add) 822 dp->ndm_eocookie = pos + 1; 823 else 824 goto out; 825 } 826 retval = &dp->ndm_cookies[pos]; 827out: 828 return (retval); 829} 830 831/* 832 * Invalidate cached directory information, except for the actual directory 833 * blocks (which are invalidated separately). 834 * Done mainly to avoid the use of stale offset cookies. 835 */ 836void 837nfs_invaldir(struct vnode *vp) 838{ 839 struct nfsnode *np = VTONFS(vp); 840 841 KASSERT(vp->v_type == VDIR, ("nfs: invaldir not dir")); 842 nfs_dircookie_lock(np); 843 np->n_direofoffset = 0; 844 np->n_cookieverf.nfsuquad[0] = 0; 845 np->n_cookieverf.nfsuquad[1] = 0; 846 if (LIST_FIRST(&np->n_cookies)) 847 LIST_FIRST(&np->n_cookies)->ndm_eocookie = 0; 848 nfs_dircookie_unlock(np); 849} 850 851/* 852 * The write verifier has changed (probably due to a server reboot), so all 853 * B_NEEDCOMMIT blocks will have to be written again. Since they are on the 854 * dirty block list as B_DELWRI, all this takes is clearing the B_NEEDCOMMIT 855 * and B_CLUSTEROK flags. Once done the new write verifier can be set for the 856 * mount point. 857 * 858 * B_CLUSTEROK must be cleared along with B_NEEDCOMMIT because stage 1 data 859 * writes are not clusterable. 860 */ 861void 862nfs_clearcommit(struct mount *mp) 863{ 864 struct vnode *vp, *nvp; 865 struct buf *bp, *nbp; 866 struct bufobj *bo; 867 868 MNT_VNODE_FOREACH_ALL(vp, mp, nvp) { 869 bo = &vp->v_bufobj; 870 vholdl(vp); 871 VI_UNLOCK(vp); 872 BO_LOCK(bo); 873 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 874 if (!BUF_ISLOCKED(bp) && 875 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) 876 == (B_DELWRI | B_NEEDCOMMIT)) 877 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 878 } 879 BO_UNLOCK(bo); 880 vdrop(vp); 881 } 882} 883 884/* 885 * Helper functions for former macros. Some of these should be 886 * moved to their callers. 887 */ 888 889int 890nfsm_mtofh_xx(struct vnode *d, struct vnode **v, int v3, int *f, 891 struct mbuf **md, caddr_t *dpos) 892{ 893 struct nfsnode *ttnp; 894 struct vnode *ttvp; 895 nfsfh_t *ttfhp; 896 u_int32_t *tl; 897 int ttfhsize; 898 int t1; 899 900 if (v3) { 901 tl = nfsm_dissect_xx(NFSX_UNSIGNED, md, dpos); 902 if (tl == NULL) 903 return EBADRPC; 904 *f = fxdr_unsigned(int, *tl); 905 } else 906 *f = 1; 907 if (*f) { 908 t1 = nfsm_getfh_xx(&ttfhp, &ttfhsize, (v3), md, dpos); 909 if (t1 != 0) 910 return t1; 911 t1 = nfs_nget(d->v_mount, ttfhp, ttfhsize, &ttnp, LK_EXCLUSIVE); 912 if (t1 != 0) 913 return t1; 914 *v = NFSTOV(ttnp); 915 } 916 if (v3) { 917 tl = nfsm_dissect_xx(NFSX_UNSIGNED, md, dpos); 918 if (tl == NULL) 919 return EBADRPC; 920 if (*f) 921 *f = fxdr_unsigned(int, *tl); 922 else if (fxdr_unsigned(int, *tl)) 923 nfsm_adv_xx(NFSX_V3FATTR, md, dpos); 924 } 925 if (*f) { 926 ttvp = *v; 927 t1 = nfs_loadattrcache(&ttvp, md, dpos, NULL, 0); 928 if (t1) 929 return t1; 930 *v = ttvp; 931 } 932 return 0; 933} 934 935int 936nfsm_getfh_xx(nfsfh_t **f, int *s, int v3, struct mbuf **md, caddr_t *dpos) 937{ 938 u_int32_t *tl; 939 940 if (v3) { 941 tl = nfsm_dissect_xx(NFSX_UNSIGNED, md, dpos); 942 if (tl == NULL) 943 return EBADRPC; 944 *s = fxdr_unsigned(int, *tl); 945 if (*s <= 0 || *s > NFSX_V3FHMAX) 946 return EBADRPC; 947 } else 948 *s = NFSX_V2FH; 949 *f = nfsm_dissect_xx(nfsm_rndup(*s), md, dpos); 950 if (*f == NULL) 951 return EBADRPC; 952 else 953 return 0; 954} 955 956 957int 958nfsm_loadattr_xx(struct vnode **v, struct vattr *va, struct mbuf **md, 959 caddr_t *dpos) 960{ 961 int t1; 962 963 struct vnode *ttvp = *v; 964 t1 = nfs_loadattrcache(&ttvp, md, dpos, va, 0); 965 if (t1 != 0) 966 return t1; 967 *v = ttvp; 968 return 0; 969} 970 971int 972nfsm_postop_attr_xx(struct vnode **v, int *f, struct vattr *va, 973 struct mbuf **md, caddr_t *dpos) 974{ 975 u_int32_t *tl; 976 int t1; 977 978 struct vnode *ttvp = *v; 979 tl = nfsm_dissect_xx(NFSX_UNSIGNED, md, dpos); 980 if (tl == NULL) 981 return EBADRPC; 982 *f = fxdr_unsigned(int, *tl); 983 if (*f != 0) { 984 t1 = nfs_loadattrcache(&ttvp, md, dpos, va, 1); 985 if (t1 != 0) { 986 *f = 0; 987 return t1; 988 } 989 *v = ttvp; 990 } 991 return 0; 992} 993 994int 995nfsm_wcc_data_xx(struct vnode **v, int *f, struct mbuf **md, caddr_t *dpos) 996{ 997 u_int32_t *tl; 998 int ttattrf, ttretf = 0; 999 int t1; 1000 1001 tl = nfsm_dissect_xx(NFSX_UNSIGNED, md, dpos); 1002 if (tl == NULL) 1003 return EBADRPC; 1004 if (*tl == nfs_true) { 1005 tl = nfsm_dissect_xx(6 * NFSX_UNSIGNED, md, dpos); 1006 if (tl == NULL) 1007 return EBADRPC; 1008 mtx_lock(&(VTONFS(*v))->n_mtx); 1009 if (*f) 1010 ttretf = (VTONFS(*v)->n_mtime.tv_sec == fxdr_unsigned(u_int32_t, *(tl + 2)) && 1011 VTONFS(*v)->n_mtime.tv_nsec == fxdr_unsigned(u_int32_t, *(tl + 3))); 1012 mtx_unlock(&(VTONFS(*v))->n_mtx); 1013 } 1014 t1 = nfsm_postop_attr_xx(v, &ttattrf, NULL, md, dpos); 1015 if (t1) 1016 return t1; 1017 if (*f) 1018 *f = ttretf; 1019 else 1020 *f = ttattrf; 1021 return 0; 1022} 1023 1024int 1025nfsm_strtom_xx(const char *a, int s, int m, struct mbuf **mb, caddr_t *bpos) 1026{ 1027 u_int32_t *tl; 1028 int t1; 1029 1030 if (s > m) 1031 return ENAMETOOLONG; 1032 t1 = nfsm_rndup(s) + NFSX_UNSIGNED; 1033 if (t1 <= M_TRAILINGSPACE(*mb)) { 1034 tl = nfsm_build_xx(t1, mb, bpos); 1035 *tl++ = txdr_unsigned(s); 1036 *(tl + ((t1 >> 2) - 2)) = 0; 1037 bcopy(a, tl, s); 1038 } else { 1039 t1 = nfsm_strtmbuf(mb, bpos, a, s); 1040 if (t1 != 0) 1041 return t1; 1042 } 1043 return 0; 1044} 1045 1046int 1047nfsm_fhtom_xx(struct vnode *v, int v3, struct mbuf **mb, caddr_t *bpos) 1048{ 1049 u_int32_t *tl; 1050 int t1; 1051 caddr_t cp; 1052 1053 if (v3) { 1054 t1 = nfsm_rndup(VTONFS(v)->n_fhsize) + NFSX_UNSIGNED; 1055 if (t1 < M_TRAILINGSPACE(*mb)) { 1056 tl = nfsm_build_xx(t1, mb, bpos); 1057 *tl++ = txdr_unsigned(VTONFS(v)->n_fhsize); 1058 *(tl + ((t1 >> 2) - 2)) = 0; 1059 bcopy(VTONFS(v)->n_fhp, tl, VTONFS(v)->n_fhsize); 1060 } else { 1061 t1 = nfsm_strtmbuf(mb, bpos, 1062 (const char *)VTONFS(v)->n_fhp, 1063 VTONFS(v)->n_fhsize); 1064 if (t1 != 0) 1065 return t1; 1066 } 1067 } else { 1068 cp = nfsm_build_xx(NFSX_V2FH, mb, bpos); 1069 bcopy(VTONFS(v)->n_fhp, cp, NFSX_V2FH); 1070 } 1071 return 0; 1072} 1073 1074void 1075nfsm_v3attrbuild_xx(struct vattr *va, int full, struct mbuf **mb, 1076 caddr_t *bpos) 1077{ 1078 u_int32_t *tl; 1079 1080 if (va->va_mode != (mode_t)VNOVAL) { 1081 tl = nfsm_build_xx(2 * NFSX_UNSIGNED, mb, bpos); 1082 *tl++ = nfs_true; 1083 *tl = txdr_unsigned(va->va_mode); 1084 } else { 1085 tl = nfsm_build_xx(NFSX_UNSIGNED, mb, bpos); 1086 *tl = nfs_false; 1087 } 1088 if (full && va->va_uid != (uid_t)VNOVAL) { 1089 tl = nfsm_build_xx(2 * NFSX_UNSIGNED, mb, bpos); 1090 *tl++ = nfs_true; 1091 *tl = txdr_unsigned(va->va_uid); 1092 } else { 1093 tl = nfsm_build_xx(NFSX_UNSIGNED, mb, bpos); 1094 *tl = nfs_false; 1095 } 1096 if (full && va->va_gid != (gid_t)VNOVAL) { 1097 tl = nfsm_build_xx(2 * NFSX_UNSIGNED, mb, bpos); 1098 *tl++ = nfs_true; 1099 *tl = txdr_unsigned(va->va_gid); 1100 } else { 1101 tl = nfsm_build_xx(NFSX_UNSIGNED, mb, bpos); 1102 *tl = nfs_false; 1103 } 1104 if (full && va->va_size != VNOVAL) { 1105 tl = nfsm_build_xx(3 * NFSX_UNSIGNED, mb, bpos); 1106 *tl++ = nfs_true; 1107 txdr_hyper(va->va_size, tl); 1108 } else { 1109 tl = nfsm_build_xx(NFSX_UNSIGNED, mb, bpos); 1110 *tl = nfs_false; 1111 } 1112 if (va->va_atime.tv_sec != VNOVAL) { 1113 if ((va->va_vaflags & VA_UTIMES_NULL) == 0) { 1114 tl = nfsm_build_xx(3 * NFSX_UNSIGNED, mb, bpos); 1115 *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT); 1116 txdr_nfsv3time(&va->va_atime, tl); 1117 } else { 1118 tl = nfsm_build_xx(NFSX_UNSIGNED, mb, bpos); 1119 *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER); 1120 } 1121 } else { 1122 tl = nfsm_build_xx(NFSX_UNSIGNED, mb, bpos); 1123 *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE); 1124 } 1125 if (va->va_mtime.tv_sec != VNOVAL) { 1126 if ((va->va_vaflags & VA_UTIMES_NULL) == 0) { 1127 tl = nfsm_build_xx(3 * NFSX_UNSIGNED, mb, bpos); 1128 *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT); 1129 txdr_nfsv3time(&va->va_mtime, tl); 1130 } else { 1131 tl = nfsm_build_xx(NFSX_UNSIGNED, mb, bpos); 1132 *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER); 1133 } 1134 } else { 1135 tl = nfsm_build_xx(NFSX_UNSIGNED, mb, bpos); 1136 *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE); 1137 } 1138} 1139