nfs_bio.c revision 1.63.2.11
1 /* $NetBSD: nfs_bio.c,v 1.63.2.11 2002/04/01 07:49:05 nathanw Exp $ */ 2 3 /* 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)nfs_bio.c 8.9 (Berkeley) 3/30/95 39 */ 40 41 #include <sys/cdefs.h> 42 __KERNEL_RCSID(0, "$NetBSD: nfs_bio.c,v 1.63.2.11 2002/04/01 07:49:05 nathanw Exp $"); 43 44 #include "opt_nfs.h" 45 #include "opt_ddb.h" 46 47 #include <sys/param.h> 48 #include <sys/systm.h> 49 #include <sys/resourcevar.h> 50 #include <sys/signalvar.h> 51 #include <sys/lwp.h> 52 #include <sys/proc.h> 53 #include <sys/buf.h> 54 #include <sys/vnode.h> 55 #include <sys/mount.h> 56 #include <sys/kernel.h> 57 #include <sys/namei.h> 58 #include <sys/dirent.h> 59 #include <sys/malloc.h> 60 61 #include <uvm/uvm_extern.h> 62 #include <uvm/uvm.h> 63 64 #include <nfs/rpcv2.h> 65 #include <nfs/nfsproto.h> 66 #include <nfs/nfs.h> 67 #include <nfs/nfsmount.h> 68 #include <nfs/nqnfs.h> 69 #include <nfs/nfsnode.h> 70 #include <nfs/nfs_var.h> 71 72 extern int nfs_numasync; 73 extern int nfs_commitsize; 74 extern struct nfsstats nfsstats; 75 76 /* 77 * Vnode op for read using bio 78 * Any similarity to readip() is purely coincidental 79 */ 80 int 81 nfs_bioread(vp, uio, ioflag, cred, cflag) 82 struct vnode *vp; 83 struct uio *uio; 84 int ioflag, cflag; 85 struct ucred *cred; 86 { 87 struct nfsnode *np = VTONFS(vp); 88 struct buf *bp = NULL, *rabp; 89 struct vattr vattr; 90 struct proc *p; 91 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 92 struct nfsdircache *ndp = NULL, *nndp = NULL; 93 caddr_t baddr, ep, edp; 94 int got_buf = 0, error = 0, n = 0, on = 0, en, enn; 95 int enough = 0; 96 struct dirent *dp, *pdp; 97 off_t curoff = 0; 98 99 #ifdef DIAGNOSTIC 100 if (uio->uio_rw != UIO_READ) 101 panic("nfs_read mode"); 102 #endif 103 if (uio->uio_resid == 0) 104 return (0); 105 if (vp->v_type != VDIR && uio->uio_offset < 0) 106 return (EINVAL); 107 p = uio->uio_procp; 108 #ifndef NFS_V2_ONLY 109 if ((nmp->nm_flag & NFSMNT_NFSV3) && 110 !(nmp->nm_iflag & NFSMNT_GOTFSINFO)) 111 (void)nfs_fsinfo(nmp, vp, cred, p); 112 #endif 113 if (vp->v_type != VDIR && 114 (uio->uio_offset + uio->uio_resid) > nmp->nm_maxfilesize) 115 return (EFBIG); 116 117 /* 118 * For nfs, cache consistency can only be maintained approximately. 119 * Although RFC1094 does not specify the criteria, the following is 120 * believed to be compatible with the reference port. 121 * For nqnfs, full cache consistency is maintained within the loop. 122 * For nfs: 123 * If the file's modify time on the server has changed since the 124 * last read rpc or you have written to the file, 125 * you may have lost data cache consistency with the 126 * server, so flush all of the file's data out of the cache. 127 * Then force a getattr rpc to ensure that you have up to date 128 * attributes. 129 * NB: This implies that cache data can be read when up to 130 * NFS_ATTRTIMEO seconds out of date. If you find that you need current 131 * attributes this could be forced by setting n_attrstamp to 0 before 132 * the VOP_GETATTR() call. 133 */ 134 135 if ((nmp->nm_flag & NFSMNT_NQNFS) == 0 && vp->v_type != VLNK) { 136 if (np->n_flag & NMODIFIED) { 137 if (vp->v_type != VREG) { 138 if (vp->v_type != VDIR) 139 panic("nfs: bioread, not dir"); 140 nfs_invaldircache(vp, 0); 141 np->n_direofoffset = 0; 142 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 143 if (error) 144 return (error); 145 } 146 np->n_attrstamp = 0; 147 error = VOP_GETATTR(vp, &vattr, cred, p); 148 if (error) 149 return (error); 150 np->n_mtime = vattr.va_mtime.tv_sec; 151 } else { 152 error = VOP_GETATTR(vp, &vattr, cred, p); 153 if (error) 154 return (error); 155 if (np->n_mtime != vattr.va_mtime.tv_sec) { 156 if (vp->v_type == VDIR) { 157 nfs_invaldircache(vp, 0); 158 np->n_direofoffset = 0; 159 } 160 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 161 if (error) 162 return (error); 163 np->n_mtime = vattr.va_mtime.tv_sec; 164 } 165 } 166 } 167 168 /* 169 * update the cached read creds for this node. 170 */ 171 172 if (np->n_rcred) { 173 crfree(np->n_rcred); 174 } 175 np->n_rcred = cred; 176 crhold(cred); 177 178 do { 179 #ifndef NFS_V2_ONLY 180 /* 181 * Get a valid lease. If cached data is stale, flush it. 182 */ 183 if (nmp->nm_flag & NFSMNT_NQNFS) { 184 if (NQNFS_CKINVALID(vp, np, ND_READ)) { 185 do { 186 error = nqnfs_getlease(vp, ND_READ, cred, p); 187 } while (error == NQNFS_EXPIRED); 188 if (error) 189 return (error); 190 if (np->n_lrev != np->n_brev || 191 (np->n_flag & NQNFSNONCACHE) || 192 ((np->n_flag & NMODIFIED) && vp->v_type == VDIR)) { 193 if (vp->v_type == VDIR) { 194 nfs_invaldircache(vp, 0); 195 np->n_direofoffset = 0; 196 } 197 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 198 if (error) 199 return (error); 200 np->n_brev = np->n_lrev; 201 } 202 } else if (vp->v_type == VDIR && (np->n_flag & NMODIFIED)) { 203 nfs_invaldircache(vp, 0); 204 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 205 np->n_direofoffset = 0; 206 if (error) 207 return (error); 208 } 209 } 210 #endif 211 /* 212 * Don't cache symlinks. 213 */ 214 if (np->n_flag & NQNFSNONCACHE 215 || ((vp->v_flag & VROOT) && vp->v_type == VLNK)) { 216 switch (vp->v_type) { 217 case VREG: 218 return (nfs_readrpc(vp, uio)); 219 case VLNK: 220 return (nfs_readlinkrpc(vp, uio, cred)); 221 case VDIR: 222 break; 223 default: 224 printf(" NQNFSNONCACHE: type %x unexpected\n", 225 vp->v_type); 226 }; 227 } 228 baddr = (caddr_t)0; 229 switch (vp->v_type) { 230 case VREG: 231 nfsstats.biocache_reads++; 232 233 error = 0; 234 if (uio->uio_offset >= np->n_size) { 235 break; 236 } 237 while (uio->uio_resid > 0) { 238 void *win; 239 vsize_t bytelen = MIN(np->n_size - uio->uio_offset, 240 uio->uio_resid); 241 242 if (bytelen == 0) 243 break; 244 win = ubc_alloc(&vp->v_uobj, uio->uio_offset, 245 &bytelen, UBC_READ); 246 error = uiomove(win, bytelen, uio); 247 ubc_release(win, 0); 248 if (error) { 249 break; 250 } 251 } 252 n = 0; 253 break; 254 255 case VLNK: 256 nfsstats.biocache_readlinks++; 257 bp = nfs_getcacheblk(vp, (daddr_t)0, NFS_MAXPATHLEN, p); 258 if (!bp) 259 return (EINTR); 260 if ((bp->b_flags & B_DONE) == 0) { 261 bp->b_flags |= B_READ; 262 error = nfs_doio(bp, p); 263 if (error) { 264 brelse(bp); 265 return (error); 266 } 267 } 268 n = MIN(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid); 269 got_buf = 1; 270 on = 0; 271 break; 272 case VDIR: 273 diragain: 274 nfsstats.biocache_readdirs++; 275 ndp = nfs_searchdircache(vp, uio->uio_offset, 276 (nmp->nm_flag & NFSMNT_XLATECOOKIE), 0); 277 if (!ndp) { 278 /* 279 * We've been handed a cookie that is not 280 * in the cache. If we're not translating 281 * 32 <-> 64, it may be a value that was 282 * flushed out of the cache because it grew 283 * too big. Let the server judge if it's 284 * valid or not. In the translation case, 285 * we have no way of validating this value, 286 * so punt. 287 */ 288 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) 289 return (EINVAL); 290 ndp = nfs_enterdircache(vp, uio->uio_offset, 291 uio->uio_offset, 0, 0); 292 } 293 294 if (uio->uio_offset != 0 && 295 ndp->dc_cookie == np->n_direofoffset) { 296 nfsstats.direofcache_hits++; 297 return (0); 298 } 299 300 bp = nfs_getcacheblk(vp, ndp->dc_blkno, NFS_DIRBLKSIZ, p); 301 if (!bp) 302 return (EINTR); 303 if ((bp->b_flags & B_DONE) == 0) { 304 bp->b_flags |= B_READ; 305 bp->b_dcookie = ndp->dc_blkcookie; 306 error = nfs_doio(bp, p); 307 if (error) { 308 /* 309 * Yuck! The directory has been modified on the 310 * server. Punt and let the userland code 311 * deal with it. 312 */ 313 brelse(bp); 314 if (error == NFSERR_BAD_COOKIE) { 315 nfs_invaldircache(vp, 0); 316 nfs_vinvalbuf(vp, 0, cred, p, 1); 317 error = EINVAL; 318 } 319 return (error); 320 } 321 } 322 323 /* 324 * Just return if we hit EOF right away with this 325 * block. Always check here, because direofoffset 326 * may have been set by an nfsiod since the last 327 * check. 328 */ 329 if (np->n_direofoffset != 0 && 330 ndp->dc_blkcookie == np->n_direofoffset) { 331 brelse(bp); 332 return (0); 333 } 334 335 /* 336 * Find the entry we were looking for in the block. 337 */ 338 339 en = ndp->dc_entry; 340 341 pdp = dp = (struct dirent *)bp->b_data; 342 edp = bp->b_data + bp->b_bcount - bp->b_resid; 343 enn = 0; 344 while (enn < en && (caddr_t)dp < edp) { 345 pdp = dp; 346 dp = (struct dirent *)((caddr_t)dp + dp->d_reclen); 347 enn++; 348 } 349 350 /* 351 * If the entry number was bigger than the number of 352 * entries in the block, or the cookie of the previous 353 * entry doesn't match, the directory cache is 354 * stale. Flush it and try again (i.e. go to 355 * the server). 356 */ 357 if ((caddr_t)dp >= edp || (caddr_t)dp + dp->d_reclen > edp || 358 (en > 0 && NFS_GETCOOKIE(pdp) != ndp->dc_cookie)) { 359 #ifdef DEBUG 360 printf("invalid cache: %p %p %p off %lx %lx\n", 361 pdp, dp, edp, 362 (unsigned long)uio->uio_offset, 363 (unsigned long)NFS_GETCOOKIE(pdp)); 364 #endif 365 brelse(bp); 366 nfs_invaldircache(vp, 0); 367 nfs_vinvalbuf(vp, 0, cred, p, 0); 368 goto diragain; 369 } 370 371 on = (caddr_t)dp - bp->b_data; 372 373 /* 374 * Cache all entries that may be exported to the 375 * user, as they may be thrown back at us. The 376 * NFSBIO_CACHECOOKIES flag indicates that all 377 * entries are being 'exported', so cache them all. 378 */ 379 380 if (en == 0 && pdp == dp) { 381 dp = (struct dirent *) 382 ((caddr_t)dp + dp->d_reclen); 383 enn++; 384 } 385 386 if (uio->uio_resid < (bp->b_bcount - bp->b_resid - on)) { 387 n = uio->uio_resid; 388 enough = 1; 389 } else 390 n = bp->b_bcount - bp->b_resid - on; 391 392 ep = bp->b_data + on + n; 393 394 /* 395 * Find last complete entry to copy, caching entries 396 * (if requested) as we go. 397 */ 398 399 while ((caddr_t)dp < ep && (caddr_t)dp + dp->d_reclen <= ep) { 400 if (cflag & NFSBIO_CACHECOOKIES) { 401 nndp = nfs_enterdircache(vp, NFS_GETCOOKIE(pdp), 402 ndp->dc_blkcookie, enn, bp->b_lblkno); 403 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) { 404 NFS_STASHCOOKIE32(pdp, 405 nndp->dc_cookie32); 406 } 407 } 408 pdp = dp; 409 dp = (struct dirent *)((caddr_t)dp + dp->d_reclen); 410 enn++; 411 } 412 413 /* 414 * If the last requested entry was not the last in the 415 * buffer (happens if NFS_DIRFRAGSIZ < NFS_DIRBLKSIZ), 416 * cache the cookie of the last requested one, and 417 * set of the offset to it. 418 */ 419 420 if ((on + n) < bp->b_bcount - bp->b_resid) { 421 curoff = NFS_GETCOOKIE(pdp); 422 nndp = nfs_enterdircache(vp, curoff, ndp->dc_blkcookie, 423 enn, bp->b_lblkno); 424 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) { 425 NFS_STASHCOOKIE32(pdp, nndp->dc_cookie32); 426 curoff = nndp->dc_cookie32; 427 } 428 } else 429 curoff = bp->b_dcookie; 430 431 /* 432 * Always cache the entry for the next block, 433 * so that readaheads can use it. 434 */ 435 nndp = nfs_enterdircache(vp, bp->b_dcookie, bp->b_dcookie, 0,0); 436 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) { 437 if (curoff == bp->b_dcookie) { 438 NFS_STASHCOOKIE32(pdp, nndp->dc_cookie32); 439 curoff = nndp->dc_cookie32; 440 } 441 } 442 443 n = ((caddr_t)pdp + pdp->d_reclen) - (bp->b_data + on); 444 445 /* 446 * If not eof and read aheads are enabled, start one. 447 * (You need the current block first, so that you have the 448 * directory offset cookie of the next block.) 449 */ 450 if (nfs_numasync > 0 && nmp->nm_readahead > 0 && 451 np->n_direofoffset == 0 && !(np->n_flag & NQNFSNONCACHE)) { 452 rabp = nfs_getcacheblk(vp, nndp->dc_blkno, 453 NFS_DIRBLKSIZ, p); 454 if (rabp) { 455 if ((rabp->b_flags & (B_DONE | B_DELWRI)) == 0) { 456 rabp->b_dcookie = nndp->dc_cookie; 457 rabp->b_flags |= (B_READ | B_ASYNC); 458 if (nfs_asyncio(rabp)) { 459 rabp->b_flags |= B_INVAL; 460 brelse(rabp); 461 } 462 } else 463 brelse(rabp); 464 } 465 } 466 got_buf = 1; 467 break; 468 default: 469 printf(" nfsbioread: type %x unexpected\n",vp->v_type); 470 break; 471 } 472 473 if (n > 0) { 474 if (!baddr) 475 baddr = bp->b_data; 476 error = uiomove(baddr + on, (int)n, uio); 477 } 478 switch (vp->v_type) { 479 case VREG: 480 break; 481 case VLNK: 482 n = 0; 483 break; 484 case VDIR: 485 if (np->n_flag & NQNFSNONCACHE) 486 bp->b_flags |= B_INVAL; 487 uio->uio_offset = curoff; 488 if (enough) 489 n = 0; 490 break; 491 default: 492 printf(" nfsbioread: type %x unexpected\n",vp->v_type); 493 } 494 if (got_buf) 495 brelse(bp); 496 } while (error == 0 && uio->uio_resid > 0 && n > 0); 497 return (error); 498 } 499 500 /* 501 * Vnode op for write using bio 502 */ 503 int 504 nfs_write(v) 505 void *v; 506 { 507 struct vop_write_args /* { 508 struct vnode *a_vp; 509 struct uio *a_uio; 510 int a_ioflag; 511 struct ucred *a_cred; 512 } */ *ap = v; 513 struct uio *uio = ap->a_uio; 514 struct proc *p = uio->uio_procp; 515 struct vnode *vp = ap->a_vp; 516 struct nfsnode *np = VTONFS(vp); 517 struct ucred *cred = ap->a_cred; 518 int ioflag = ap->a_ioflag; 519 struct vattr vattr; 520 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 521 void *win; 522 voff_t oldoff, origoff; 523 vsize_t bytelen; 524 int error = 0, iomode, must_commit; 525 526 #ifdef DIAGNOSTIC 527 if (uio->uio_rw != UIO_WRITE) 528 panic("nfs_write mode"); 529 if (uio->uio_segflg == UIO_USERSPACE && 530 uio->uio_procp != curproc->l_proc) 531 panic("nfs_write proc"); 532 #endif 533 if (vp->v_type != VREG) 534 return (EIO); 535 if (np->n_flag & NWRITEERR) { 536 np->n_flag &= ~NWRITEERR; 537 return (np->n_error); 538 } 539 #ifndef NFS_V2_ONLY 540 if ((nmp->nm_flag & NFSMNT_NFSV3) && 541 !(nmp->nm_iflag & NFSMNT_GOTFSINFO)) 542 (void)nfs_fsinfo(nmp, vp, cred, p); 543 #endif 544 if (ioflag & (IO_APPEND | IO_SYNC)) { 545 if (np->n_flag & NMODIFIED) { 546 np->n_attrstamp = 0; 547 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 548 if (error) 549 return (error); 550 } 551 if (ioflag & IO_APPEND) { 552 np->n_attrstamp = 0; 553 error = VOP_GETATTR(vp, &vattr, cred, p); 554 if (error) 555 return (error); 556 uio->uio_offset = np->n_size; 557 } 558 } 559 if (uio->uio_offset < 0) 560 return (EINVAL); 561 if ((uio->uio_offset + uio->uio_resid) > nmp->nm_maxfilesize) 562 return (EFBIG); 563 if (uio->uio_resid == 0) 564 return (0); 565 /* 566 * Maybe this should be above the vnode op call, but so long as 567 * file servers have no limits, i don't think it matters 568 */ 569 if (p && uio->uio_offset + uio->uio_resid > 570 p->p_rlimit[RLIMIT_FSIZE].rlim_cur) { 571 psignal(p, SIGXFSZ); 572 return (EFBIG); 573 } 574 575 /* 576 * update the cached write creds for this node. 577 */ 578 579 if (np->n_wcred) { 580 crfree(np->n_wcred); 581 } 582 np->n_wcred = cred; 583 crhold(cred); 584 585 if ((np->n_flag & NQNFSNONCACHE) && uio->uio_iovcnt == 1) { 586 iomode = NFSV3WRITE_FILESYNC; 587 error = nfs_writerpc(vp, uio, &iomode, &must_commit); 588 if (must_commit) 589 nfs_clearcommit(vp->v_mount); 590 return (error); 591 } 592 593 origoff = uio->uio_offset; 594 do { 595 oldoff = uio->uio_offset; 596 bytelen = uio->uio_resid; 597 598 #ifndef NFS_V2_ONLY 599 /* 600 * Check for a valid write lease. 601 */ 602 if ((nmp->nm_flag & NFSMNT_NQNFS) && 603 NQNFS_CKINVALID(vp, np, ND_WRITE)) { 604 do { 605 error = nqnfs_getlease(vp, ND_WRITE, cred, p); 606 } while (error == NQNFS_EXPIRED); 607 if (error) 608 return (error); 609 if (np->n_lrev != np->n_brev || 610 (np->n_flag & NQNFSNONCACHE)) { 611 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 612 if (error) 613 return (error); 614 np->n_brev = np->n_lrev; 615 } 616 } 617 #endif 618 nfsstats.biocache_writes++; 619 620 np->n_flag |= NMODIFIED; 621 if (np->n_size < uio->uio_offset + bytelen) { 622 np->n_size = uio->uio_offset + bytelen; 623 } 624 if ((uio->uio_offset & PAGE_MASK) == 0 && 625 ((uio->uio_offset + bytelen) & PAGE_MASK) == 0) { 626 win = ubc_alloc(&vp->v_uobj, uio->uio_offset, &bytelen, 627 UBC_WRITE | UBC_FAULTBUSY); 628 } else { 629 win = ubc_alloc(&vp->v_uobj, uio->uio_offset, &bytelen, 630 UBC_WRITE); 631 } 632 error = uiomove(win, bytelen, uio); 633 ubc_release(win, 0); 634 if (error) { 635 break; 636 } 637 638 /* 639 * update UVM's notion of the size now that we've 640 * copied the data into the vnode's pages. 641 */ 642 643 if (vp->v_size < uio->uio_offset) { 644 uvm_vnp_setsize(vp, uio->uio_offset); 645 } 646 647 if ((oldoff & ~(nmp->nm_wsize - 1)) != 648 (uio->uio_offset & ~(nmp->nm_wsize - 1))) { 649 simple_lock(&vp->v_interlock); 650 error = VOP_PUTPAGES(vp, 651 trunc_page(oldoff & ~(nmp->nm_wsize - 1)), 652 round_page((uio->uio_offset + nmp->nm_wsize - 1) & 653 ~(nmp->nm_wsize - 1)), PGO_CLEANIT); 654 } 655 } while (uio->uio_resid > 0); 656 if ((np->n_flag & NQNFSNONCACHE) || (ioflag & IO_SYNC)) { 657 simple_lock(&vp->v_interlock); 658 error = VOP_PUTPAGES(vp, 659 trunc_page(origoff & ~(nmp->nm_wsize - 1)), 660 round_page((uio->uio_offset + nmp->nm_wsize - 1) & 661 ~(nmp->nm_wsize - 1)), 662 PGO_CLEANIT | PGO_SYNCIO); 663 } 664 return error; 665 } 666 667 /* 668 * Get an nfs cache block. 669 * Allocate a new one if the block isn't currently in the cache 670 * and return the block marked busy. If the calling process is 671 * interrupted by a signal for an interruptible mount point, return 672 * NULL. 673 */ 674 struct buf * 675 nfs_getcacheblk(vp, bn, size, p) 676 struct vnode *vp; 677 daddr_t bn; 678 int size; 679 struct proc *p; 680 { 681 struct buf *bp; 682 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 683 684 if (nmp->nm_flag & NFSMNT_INT) { 685 bp = getblk(vp, bn, size, PCATCH, 0); 686 while (bp == NULL) { 687 if (nfs_sigintr(nmp, NULL, p)) 688 return (NULL); 689 bp = getblk(vp, bn, size, 0, 2 * hz); 690 } 691 } else 692 bp = getblk(vp, bn, size, 0, 0); 693 return (bp); 694 } 695 696 /* 697 * Flush and invalidate all dirty buffers. If another process is already 698 * doing the flush, just wait for completion. 699 */ 700 int 701 nfs_vinvalbuf(vp, flags, cred, p, intrflg) 702 struct vnode *vp; 703 int flags; 704 struct ucred *cred; 705 struct proc *p; 706 int intrflg; 707 { 708 struct nfsnode *np = VTONFS(vp); 709 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 710 int error = 0, slpflag, slptimeo; 711 712 if ((nmp->nm_flag & NFSMNT_INT) == 0) 713 intrflg = 0; 714 if (intrflg) { 715 slpflag = PCATCH; 716 slptimeo = 2 * hz; 717 } else { 718 slpflag = 0; 719 slptimeo = 0; 720 } 721 /* 722 * First wait for any other process doing a flush to complete. 723 */ 724 while (np->n_flag & NFLUSHINPROG) { 725 np->n_flag |= NFLUSHWANT; 726 error = tsleep((caddr_t)&np->n_flag, PRIBIO + 2, "nfsvinval", 727 slptimeo); 728 if (error && intrflg && nfs_sigintr(nmp, NULL, p)) 729 return (EINTR); 730 } 731 732 /* 733 * Now, flush as required. 734 */ 735 np->n_flag |= NFLUSHINPROG; 736 error = vinvalbuf(vp, flags, cred, p, slpflag, 0); 737 while (error) { 738 if (intrflg && nfs_sigintr(nmp, NULL, p)) { 739 np->n_flag &= ~NFLUSHINPROG; 740 if (np->n_flag & NFLUSHWANT) { 741 np->n_flag &= ~NFLUSHWANT; 742 wakeup((caddr_t)&np->n_flag); 743 } 744 return (EINTR); 745 } 746 error = vinvalbuf(vp, flags, cred, p, 0, slptimeo); 747 } 748 np->n_flag &= ~(NMODIFIED | NFLUSHINPROG); 749 if (np->n_flag & NFLUSHWANT) { 750 np->n_flag &= ~NFLUSHWANT; 751 wakeup((caddr_t)&np->n_flag); 752 } 753 return (0); 754 } 755 756 /* 757 * Initiate asynchronous I/O. Return an error if no nfsiods are available. 758 * This is mainly to avoid queueing async I/O requests when the nfsiods 759 * are all hung on a dead server. 760 */ 761 762 int 763 nfs_asyncio(bp) 764 struct buf *bp; 765 { 766 int i; 767 struct nfsmount *nmp; 768 int gotiod, slpflag = 0, slptimeo = 0, error; 769 770 if (nfs_numasync == 0) 771 return (EIO); 772 773 nmp = VFSTONFS(bp->b_vp->v_mount); 774 again: 775 if (nmp->nm_flag & NFSMNT_INT) 776 slpflag = PCATCH; 777 gotiod = FALSE; 778 779 /* 780 * Find a free iod to process this request. 781 */ 782 783 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 784 if (nfs_iodwant[i]) { 785 /* 786 * Found one, so wake it up and tell it which 787 * mount to process. 788 */ 789 nfs_iodwant[i] = NULL; 790 nfs_iodmount[i] = nmp; 791 nmp->nm_bufqiods++; 792 wakeup((caddr_t)&nfs_iodwant[i]); 793 gotiod = TRUE; 794 break; 795 } 796 797 /* 798 * If none are free, we may already have an iod working on this mount 799 * point. If so, it will process our request. 800 */ 801 802 if (!gotiod && nmp->nm_bufqiods > 0) 803 gotiod = TRUE; 804 805 /* 806 * If we have an iod which can process the request, then queue 807 * the buffer. 808 */ 809 810 if (gotiod) { 811 812 /* 813 * Ensure that the queue never grows too large. 814 */ 815 816 while (nmp->nm_bufqlen >= 2*nfs_numasync) { 817 nmp->nm_bufqwant = TRUE; 818 error = tsleep(&nmp->nm_bufq, slpflag | PRIBIO, 819 "nfsaio", slptimeo); 820 if (error) { 821 if (nfs_sigintr(nmp, NULL, curproc->l_proc)) 822 return (EINTR); 823 if (slpflag == PCATCH) { 824 slpflag = 0; 825 slptimeo = 2 * hz; 826 } 827 } 828 829 /* 830 * We might have lost our iod while sleeping, 831 * so check and loop if nescessary. 832 */ 833 834 if (nmp->nm_bufqiods == 0) 835 goto again; 836 } 837 TAILQ_INSERT_TAIL(&nmp->nm_bufq, bp, b_freelist); 838 nmp->nm_bufqlen++; 839 return (0); 840 } 841 842 /* 843 * All the iods are busy on other mounts, so return EIO to 844 * force the caller to process the i/o synchronously. 845 */ 846 847 return (EIO); 848 } 849 850 /* 851 * Do an I/O operation to/from a cache block. This may be called 852 * synchronously or from an nfsiod. 853 */ 854 int 855 nfs_doio(bp, p) 856 struct buf *bp; 857 struct proc *p; 858 { 859 struct uio *uiop; 860 struct vnode *vp; 861 struct nfsnode *np; 862 struct nfsmount *nmp; 863 int error = 0, diff, len, iomode, must_commit = 0; 864 int pushedrange; 865 struct uio uio; 866 struct iovec io; 867 off_t off, cnt; 868 struct uvm_object *uobj; 869 UVMHIST_FUNC("nfs_doio"); UVMHIST_CALLED(ubchist); 870 871 vp = bp->b_vp; 872 uobj = &vp->v_uobj; 873 np = VTONFS(vp); 874 nmp = VFSTONFS(vp->v_mount); 875 uiop = &uio; 876 uiop->uio_iov = &io; 877 uiop->uio_iovcnt = 1; 878 uiop->uio_segflg = UIO_SYSSPACE; 879 uiop->uio_procp = p; 880 881 /* 882 * Historically, paging was done with physio, but no more... 883 */ 884 if (bp->b_flags & B_PHYS) { 885 /* 886 * ...though reading /dev/drum still gets us here. 887 */ 888 io.iov_len = uiop->uio_resid = bp->b_bcount; 889 /* mapping was done by vmapbuf() */ 890 io.iov_base = bp->b_data; 891 uiop->uio_offset = ((off_t)bp->b_blkno) << DEV_BSHIFT; 892 if (bp->b_flags & B_READ) { 893 uiop->uio_rw = UIO_READ; 894 nfsstats.read_physios++; 895 error = nfs_readrpc(vp, uiop); 896 } else { 897 iomode = NFSV3WRITE_DATASYNC; 898 uiop->uio_rw = UIO_WRITE; 899 nfsstats.write_physios++; 900 error = nfs_writerpc(vp, uiop, &iomode, &must_commit); 901 } 902 if (error) { 903 bp->b_flags |= B_ERROR; 904 bp->b_error = error; 905 } 906 } else if (bp->b_flags & B_READ) { 907 io.iov_len = uiop->uio_resid = bp->b_bcount; 908 io.iov_base = bp->b_data; 909 uiop->uio_rw = UIO_READ; 910 switch (vp->v_type) { 911 case VREG: 912 uiop->uio_offset = ((off_t)bp->b_blkno) << DEV_BSHIFT; 913 nfsstats.read_bios++; 914 error = nfs_readrpc(vp, uiop); 915 if (!error && uiop->uio_resid) { 916 917 /* 918 * If len > 0, there is a hole in the file and 919 * no writes after the hole have been pushed to 920 * the server yet. 921 * Just zero fill the rest of the valid area. 922 */ 923 924 diff = bp->b_bcount - uiop->uio_resid; 925 len = np->n_size - ((((off_t)bp->b_blkno) << DEV_BSHIFT) 926 + diff); 927 if (len > 0) { 928 len = MIN(len, uiop->uio_resid); 929 memset((char *)bp->b_data + diff, 0, len); 930 } 931 } 932 if (p && (vp->v_flag & VTEXT) && 933 (((nmp->nm_flag & NFSMNT_NQNFS) && 934 NQNFS_CKINVALID(vp, np, ND_READ) && 935 np->n_lrev != np->n_brev) || 936 (!(nmp->nm_flag & NFSMNT_NQNFS) && 937 np->n_mtime != np->n_vattr->va_mtime.tv_sec))) { 938 uprintf("Process killed due to " 939 "text file modification\n"); 940 psignal(p, SIGKILL); 941 #if 0 /* XXX NJWLWP */ 942 p->p_holdcnt++; 943 #endif 944 } 945 break; 946 case VLNK: 947 uiop->uio_offset = (off_t)0; 948 nfsstats.readlink_bios++; 949 error = nfs_readlinkrpc(vp, uiop, curproc->l_proc->p_ucred); 950 break; 951 case VDIR: 952 nfsstats.readdir_bios++; 953 uiop->uio_offset = bp->b_dcookie; 954 if (nmp->nm_flag & NFSMNT_RDIRPLUS) { 955 error = nfs_readdirplusrpc(vp, uiop, curproc->l_proc->p_ucred); 956 if (error == NFSERR_NOTSUPP) 957 nmp->nm_flag &= ~NFSMNT_RDIRPLUS; 958 } 959 if ((nmp->nm_flag & NFSMNT_RDIRPLUS) == 0) 960 error = nfs_readdirrpc(vp, uiop, curproc->l_proc->p_ucred); 961 if (!error) { 962 bp->b_dcookie = uiop->uio_offset; 963 } 964 break; 965 default: 966 printf("nfs_doio: type %x unexpected\n",vp->v_type); 967 break; 968 } 969 if (error) { 970 bp->b_flags |= B_ERROR; 971 bp->b_error = error; 972 } 973 } else { 974 int i, npages = bp->b_bufsize >> PAGE_SHIFT; 975 struct vm_page *pgs[npages]; 976 boolean_t needcommit = TRUE; 977 978 if ((bp->b_flags & B_ASYNC) != 0 && NFS_ISV3(vp)) { 979 iomode = NFSV3WRITE_UNSTABLE; 980 } else { 981 iomode = NFSV3WRITE_FILESYNC; 982 } 983 984 for (i = 0; i < npages; i++) { 985 pgs[i] = uvm_pageratop((vaddr_t)bp->b_data + 986 (i << PAGE_SHIFT)); 987 if ((pgs[i]->flags & PG_NEEDCOMMIT) == 0) { 988 needcommit = FALSE; 989 } 990 } 991 if (!needcommit && iomode == NFSV3WRITE_UNSTABLE) { 992 for (i = 0; i < npages; i++) { 993 pgs[i]->flags |= PG_NEEDCOMMIT | PG_RDONLY; 994 pmap_page_protect(pgs[i], VM_PROT_READ); 995 } 996 } 997 998 uiop->uio_offset = (((off_t)bp->b_blkno) << DEV_BSHIFT); 999 off = uiop->uio_offset; 1000 cnt = bp->b_bcount; 1001 1002 /* 1003 * Send the data to the server if necessary, 1004 * otherwise just send a commit rpc. 1005 */ 1006 1007 if (needcommit) { 1008 1009 /* 1010 * If the buffer is in the range that we already committed, 1011 * there's nothing to do. 1012 * 1013 * If it's in the range that we need to commit, push the 1014 * whole range at once, otherwise only push the buffer. 1015 * In both these cases, acquire the commit lock to avoid 1016 * other processes modifying the range. 1017 */ 1018 1019 lockmgr(&np->n_commitlock, LK_EXCLUSIVE, NULL); 1020 if (!nfs_in_committed_range(vp, off, bp->b_bcount)) { 1021 if (nfs_in_tobecommitted_range(vp, off, bp->b_bcount)) { 1022 pushedrange = 1; 1023 off = np->n_pushlo; 1024 cnt = np->n_pushhi - np->n_pushlo; 1025 } else { 1026 pushedrange = 0; 1027 } 1028 error = nfs_commit(vp, off, cnt, curproc->l_proc); 1029 if (error == 0) { 1030 if (pushedrange) { 1031 nfs_merge_commit_ranges(vp); 1032 } else { 1033 nfs_add_committed_range(vp, off, cnt); 1034 } 1035 } 1036 } 1037 lockmgr(&np->n_commitlock, LK_RELEASE, NULL); 1038 if (!error) { 1039 bp->b_resid = 0; 1040 simple_lock(&uobj->vmobjlock); 1041 for (i = 0; i < npages; i++) { 1042 pgs[i]->flags &= ~(PG_NEEDCOMMIT | PG_RDONLY); 1043 } 1044 simple_unlock(&uobj->vmobjlock); 1045 biodone(bp); 1046 return (0); 1047 } else if (error == NFSERR_STALEWRITEVERF) { 1048 nfs_clearcommit(bp->b_vp->v_mount); 1049 } 1050 } 1051 io.iov_base = bp->b_data; 1052 io.iov_len = uiop->uio_resid = bp->b_bcount; 1053 uiop->uio_rw = UIO_WRITE; 1054 nfsstats.write_bios++; 1055 error = nfs_writerpc(vp, uiop, &iomode, &must_commit); 1056 if (!error && iomode == NFSV3WRITE_UNSTABLE) { 1057 lockmgr(&np->n_commitlock, LK_EXCLUSIVE, NULL); 1058 nfs_add_tobecommitted_range(vp, off, cnt); 1059 simple_lock(&uobj->vmobjlock); 1060 for (i = 0; i < npages; i++) { 1061 pgs[i]->flags &= ~PG_CLEAN; 1062 } 1063 simple_unlock(&uobj->vmobjlock); 1064 if (np->n_pushhi - np->n_pushlo > nfs_commitsize) { 1065 off = np->n_pushlo; 1066 cnt = nfs_commitsize >> 1; 1067 error = nfs_commit(vp, off, cnt, curproc->l_proc); 1068 if (!error) { 1069 nfs_add_committed_range(vp, off, cnt); 1070 nfs_del_tobecommitted_range(vp, off, cnt); 1071 } 1072 } 1073 lockmgr(&np->n_commitlock, LK_RELEASE, NULL); 1074 } else if (!error && needcommit) { 1075 lockmgr(&np->n_commitlock, LK_EXCLUSIVE, NULL); 1076 nfs_del_committed_range(vp, off, cnt); 1077 lockmgr(&np->n_commitlock, LK_RELEASE, NULL); 1078 simple_lock(&uobj->vmobjlock); 1079 for (i = 0; i < npages; i++) { 1080 pgs[i]->flags &= ~(PG_NEEDCOMMIT | PG_RDONLY); 1081 } 1082 simple_unlock(&uobj->vmobjlock); 1083 } 1084 } 1085 bp->b_resid = uiop->uio_resid; 1086 if (must_commit || (error == NFSERR_STALEWRITEVERF)) { 1087 nfs_clearcommit(vp->v_mount); 1088 } 1089 biodone(bp); 1090 return (error); 1091 } 1092 1093 /* 1094 * Vnode op for VM getpages. 1095 */ 1096 1097 int 1098 nfs_getpages(v) 1099 void *v; 1100 { 1101 struct vop_getpages_args /* { 1102 struct vnode *a_vp; 1103 voff_t a_offset; 1104 struct vm_page **a_m; 1105 int *a_count; 1106 int a_centeridx; 1107 vm_prot_t a_access_type; 1108 int a_advice; 1109 int a_flags; 1110 } */ *ap = v; 1111 1112 struct vnode *vp = ap->a_vp; 1113 struct uvm_object *uobj = &vp->v_uobj; 1114 struct nfsnode *np = VTONFS(vp); 1115 struct vm_page *pg, **pgs; 1116 off_t origoffset, len; 1117 int i, error, npages; 1118 boolean_t v3 = NFS_ISV3(vp); 1119 boolean_t write = (ap->a_access_type & VM_PROT_WRITE) != 0; 1120 boolean_t locked = (ap->a_flags & PGO_LOCKED) != 0; 1121 1122 /* 1123 * update the cached read creds for this node. 1124 */ 1125 1126 if (np->n_rcred) { 1127 crfree(np->n_rcred); 1128 } 1129 np->n_rcred = curproc->l_proc->p_ucred; 1130 crhold(np->n_rcred); 1131 1132 /* 1133 * call the genfs code to get the pages. 1134 */ 1135 1136 npages = *ap->a_count; 1137 error = genfs_getpages(v); 1138 if (error) { 1139 return error; 1140 } 1141 1142 /* 1143 * for read faults where the nfs node is not yet marked NMODIFIED, 1144 * set PG_RDONLY on the pages so that we come back here if someone 1145 * tries to modify later via the mapping that will be entered for 1146 * this fault. 1147 */ 1148 1149 pgs = ap->a_m; 1150 if (!write && (np->n_flag & NMODIFIED) == 0 && pgs != NULL) { 1151 if (!locked) { 1152 simple_lock(&uobj->vmobjlock); 1153 } 1154 for (i = 0; i < npages; i++) { 1155 pg = pgs[i]; 1156 if (pg == NULL || pg == PGO_DONTCARE) { 1157 continue; 1158 } 1159 pg->flags |= PG_RDONLY; 1160 } 1161 if (!locked) { 1162 simple_unlock(&uobj->vmobjlock); 1163 } 1164 } 1165 if (!write) { 1166 return error; 1167 } 1168 1169 /* 1170 * this is a write fault, update the commit info. 1171 */ 1172 1173 origoffset = ap->a_offset; 1174 len = npages << PAGE_SHIFT; 1175 1176 np->n_flag |= NMODIFIED; 1177 if (v3) { 1178 lockmgr(&np->n_commitlock, LK_EXCLUSIVE, NULL); 1179 nfs_del_committed_range(vp, origoffset, len); 1180 nfs_del_tobecommitted_range(vp, origoffset, len); 1181 } 1182 if (!locked) { 1183 simple_lock(&uobj->vmobjlock); 1184 } 1185 for (i = 0; i < npages; i++) { 1186 pg = pgs[i]; 1187 if (pg == NULL || pg == PGO_DONTCARE) { 1188 continue; 1189 } 1190 pg->flags &= ~(PG_NEEDCOMMIT | PG_RDONLY); 1191 } 1192 if (!locked) { 1193 simple_unlock(&uobj->vmobjlock); 1194 } 1195 if (v3) { 1196 lockmgr(&np->n_commitlock, LK_RELEASE, NULL); 1197 } 1198 return 0; 1199 } 1200
Indexes created Mon Oct 13 08:09:50 GMT 2025