nfs_bio.c revision 1.37.2.1
1 /* $NetBSD: nfs_bio.c,v 1.37.2.1 1997/10/23 01:42:53 mellon 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 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/resourcevar.h> 45 #include <sys/signalvar.h> 46 #include <sys/proc.h> 47 #include <sys/buf.h> 48 #include <sys/vnode.h> 49 #include <sys/trace.h> 50 #include <sys/mount.h> 51 #include <sys/kernel.h> 52 #include <sys/namei.h> 53 #include <sys/dirent.h> 54 55 #include <vm/vm.h> 56 57 #include <nfs/rpcv2.h> 58 #include <nfs/nfsproto.h> 59 #include <nfs/nfs.h> 60 #include <nfs/nfsmount.h> 61 #include <nfs/nqnfs.h> 62 #include <nfs/nfsnode.h> 63 #include <nfs/nfs_var.h> 64 65 extern int nfs_numasync; 66 extern struct nfsstats nfsstats; 67 68 /* 69 * Vnode op for read using bio 70 * Any similarity to readip() is purely coincidental 71 */ 72 int 73 nfs_bioread(vp, uio, ioflag, cred, cflag) 74 register struct vnode *vp; 75 register struct uio *uio; 76 int ioflag, cflag; 77 struct ucred *cred; 78 { 79 register struct nfsnode *np = VTONFS(vp); 80 register int biosize, diff; 81 struct buf *bp = NULL, *rabp; 82 struct vattr vattr; 83 struct proc *p; 84 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 85 struct nfsdircache *ndp = NULL, *nndp = NULL; 86 daddr_t lbn, bn, rabn; 87 caddr_t baddr, ep, edp; 88 int got_buf = 0, nra, error = 0, n = 0, on = 0, not_readin, en, enn; 89 int enough = 0; 90 struct dirent *dp, *pdp; 91 off_t curoff = 0; 92 93 #ifdef DIAGNOSTIC 94 if (uio->uio_rw != UIO_READ) 95 panic("nfs_read mode"); 96 #endif 97 if (uio->uio_resid == 0) 98 return (0); 99 if (vp->v_type != VDIR && uio->uio_offset < 0) 100 return (EINVAL); 101 p = uio->uio_procp; 102 if ((nmp->nm_flag & NFSMNT_NFSV3) && 103 !(nmp->nm_iflag & NFSMNT_GOTFSINFO)) 104 (void)nfs_fsinfo(nmp, vp, cred, p); 105 if (vp->v_type != VDIR && 106 (uio->uio_offset + uio->uio_resid) > nmp->nm_maxfilesize) 107 return (EFBIG); 108 biosize = nmp->nm_rsize; 109 /* 110 * For nfs, cache consistency can only be maintained approximately. 111 * Although RFC1094 does not specify the criteria, the following is 112 * believed to be compatible with the reference port. 113 * For nqnfs, full cache consistency is maintained within the loop. 114 * For nfs: 115 * If the file's modify time on the server has changed since the 116 * last read rpc or you have written to the file, 117 * you may have lost data cache consistency with the 118 * server, so flush all of the file's data out of the cache. 119 * Then force a getattr rpc to ensure that you have up to date 120 * attributes. 121 * NB: This implies that cache data can be read when up to 122 * NFS_ATTRTIMEO seconds out of date. If you find that you need current 123 * attributes this could be forced by setting n_attrstamp to 0 before 124 * the VOP_GETATTR() call. 125 */ 126 if ((nmp->nm_flag & NFSMNT_NQNFS) == 0 && vp->v_type != VLNK) { 127 if (np->n_flag & NMODIFIED) { 128 if (vp->v_type != VREG) { 129 if (vp->v_type != VDIR) 130 panic("nfs: bioread, not dir"); 131 nfs_invaldircache(vp, 0); 132 np->n_direofoffset = 0; 133 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 134 if (error) 135 return (error); 136 } 137 np->n_attrstamp = 0; 138 error = VOP_GETATTR(vp, &vattr, cred, p); 139 if (error) 140 return (error); 141 np->n_mtime = vattr.va_mtime.tv_sec; 142 } else { 143 error = VOP_GETATTR(vp, &vattr, cred, p); 144 if (error) 145 return (error); 146 if (np->n_mtime != vattr.va_mtime.tv_sec) { 147 if (vp->v_type == VDIR) { 148 nfs_invaldircache(vp, 0); 149 np->n_direofoffset = 0; 150 } 151 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 152 if (error) 153 return (error); 154 np->n_mtime = vattr.va_mtime.tv_sec; 155 } 156 } 157 } 158 do { 159 160 /* 161 * Get a valid lease. If cached data is stale, flush it. 162 */ 163 if (nmp->nm_flag & NFSMNT_NQNFS) { 164 if (NQNFS_CKINVALID(vp, np, ND_READ)) { 165 do { 166 error = nqnfs_getlease(vp, ND_READ, cred, p); 167 } while (error == NQNFS_EXPIRED); 168 if (error) 169 return (error); 170 if (np->n_lrev != np->n_brev || 171 (np->n_flag & NQNFSNONCACHE) || 172 ((np->n_flag & NMODIFIED) && vp->v_type == VDIR)) { 173 if (vp->v_type == VDIR) { 174 nfs_invaldircache(vp, 0); 175 np->n_direofoffset = 0; 176 } 177 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 178 if (error) 179 return (error); 180 np->n_brev = np->n_lrev; 181 } 182 } else if (vp->v_type == VDIR && (np->n_flag & NMODIFIED)) { 183 nfs_invaldircache(vp, 0); 184 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 185 np->n_direofoffset = 0; 186 if (error) 187 return (error); 188 } 189 } 190 /* 191 * Don't cache symlinks. 192 */ 193 if (np->n_flag & NQNFSNONCACHE 194 || ((vp->v_flag & VROOT) && vp->v_type == VLNK)) { 195 switch (vp->v_type) { 196 case VREG: 197 return (nfs_readrpc(vp, uio, cred)); 198 case VLNK: 199 return (nfs_readlinkrpc(vp, uio, cred)); 200 case VDIR: 201 break; 202 default: 203 printf(" NQNFSNONCACHE: type %x unexpected\n", 204 vp->v_type); 205 }; 206 } 207 baddr = (caddr_t)0; 208 switch (vp->v_type) { 209 case VREG: 210 nfsstats.biocache_reads++; 211 lbn = uio->uio_offset / biosize; 212 on = uio->uio_offset & (biosize - 1); 213 bn = lbn * (biosize / DEV_BSIZE); 214 not_readin = 1; 215 216 /* 217 * Start the read ahead(s), as required. 218 */ 219 if (nfs_numasync > 0 && nmp->nm_readahead > 0 && 220 lbn - 1 == vp->v_lastr) { 221 for (nra = 0; nra < nmp->nm_readahead && 222 (lbn + 1 + nra) * biosize < np->n_size; nra++) { 223 rabn = (lbn + 1 + nra) * (biosize / DEV_BSIZE); 224 if (!incore(vp, rabn)) { 225 rabp = nfs_getcacheblk(vp, rabn, biosize, p); 226 if (!rabp) 227 return (EINTR); 228 if ((rabp->b_flags & (B_DELWRI | B_DONE)) == 0) { 229 rabp->b_flags |= (B_READ | B_ASYNC); 230 if (nfs_asyncio(rabp, cred)) { 231 rabp->b_flags |= B_INVAL; 232 brelse(rabp); 233 } 234 } else 235 brelse(rabp); 236 } 237 } 238 } 239 240 /* 241 * If the block is in the cache and has the required data 242 * in a valid region, just copy it out. 243 * Otherwise, get the block and write back/read in, 244 * as required. 245 */ 246 if ((bp = incore(vp, bn)) && 247 (bp->b_flags & (B_BUSY | B_WRITEINPROG)) == 248 (B_BUSY | B_WRITEINPROG)) 249 got_buf = 0; 250 else { 251 again: 252 bp = nfs_getcacheblk(vp, bn, biosize, p); 253 if (!bp) 254 return (EINTR); 255 got_buf = 1; 256 if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0) { 257 bp->b_flags |= B_READ; 258 not_readin = 0; 259 error = nfs_doio(bp, cred, p); 260 if (error) { 261 brelse(bp); 262 return (error); 263 } 264 } 265 } 266 n = min((unsigned)(biosize - on), uio->uio_resid); 267 diff = np->n_size - uio->uio_offset; 268 if (diff < n) 269 n = diff; 270 if (not_readin && n > 0) { 271 if (on < bp->b_validoff || (on + n) > bp->b_validend) { 272 if (!got_buf) { 273 bp = nfs_getcacheblk(vp, bn, biosize, p); 274 if (!bp) 275 return (EINTR); 276 got_buf = 1; 277 } 278 bp->b_flags |= B_INVAFTERWRITE; 279 if (bp->b_dirtyend > 0) { 280 if ((bp->b_flags & B_DELWRI) == 0) 281 panic("nfsbioread"); 282 if (VOP_BWRITE(bp) == EINTR) 283 return (EINTR); 284 } else 285 brelse(bp); 286 goto again; 287 } 288 } 289 vp->v_lastr = lbn; 290 diff = (on >= bp->b_validend) ? 0 : (bp->b_validend - on); 291 if (diff < n) 292 n = diff; 293 break; 294 case VLNK: 295 nfsstats.biocache_readlinks++; 296 bp = nfs_getcacheblk(vp, (daddr_t)0, NFS_MAXPATHLEN, p); 297 if (!bp) 298 return (EINTR); 299 if ((bp->b_flags & B_DONE) == 0) { 300 bp->b_flags |= B_READ; 301 error = nfs_doio(bp, cred, p); 302 if (error) { 303 brelse(bp); 304 return (error); 305 } 306 } 307 n = min(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid); 308 got_buf = 1; 309 on = 0; 310 break; 311 case VDIR: 312 diragain: 313 nfsstats.biocache_readdirs++; 314 ndp = nfs_searchdircache(vp, uio->uio_offset, 315 (nmp->nm_flag & NFSMNT_XLATECOOKIE), 0); 316 if (!ndp) { 317 /* 318 * We've been handed a cookie that is not 319 * in the cache. If we're not translating 320 * 32 <-> 64, it may be a value that was 321 * flushed out of the cache because it grew 322 * too big. Let the server judge if it's 323 * valid or not. In the translation case, 324 * we have no way of validating this value, 325 * so punt. 326 */ 327 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) 328 return (EINVAL); 329 ndp = nfs_enterdircache(vp, uio->uio_offset, 330 uio->uio_offset, 0, 0); 331 } 332 333 if (uio->uio_offset != 0 && 334 ndp->dc_cookie == np->n_direofoffset) { 335 nfsstats.direofcache_hits++; 336 return (0); 337 } 338 339 bp = nfs_getcacheblk(vp, ndp->dc_blkno, NFS_DIRBLKSIZ, p); 340 if (!bp) 341 return (EINTR); 342 if ((bp->b_flags & B_DONE) == 0) { 343 bp->b_flags |= B_READ; 344 bp->b_dcookie = ndp->dc_blkcookie; 345 error = nfs_doio(bp, cred, p); 346 if (error) { 347 /* 348 * Yuck! The directory has been modified on the 349 * server. Punt and let the userland code 350 * deal with it. 351 */ 352 brelse(bp); 353 if (error == NFSERR_BAD_COOKIE) { 354 nfs_invaldircache(vp, 0); 355 nfs_vinvalbuf(vp, 0, cred, p, 1); 356 error = EINVAL; 357 } 358 return (error); 359 } 360 /* 361 * Just return if we hit EOF right away with this 362 * block. 363 */ 364 if (ndp->dc_blkcookie == np->n_direofoffset) { 365 brelse(bp); 366 return (0); 367 } 368 } 369 370 /* 371 * Find the entry we were looking for in the block. 372 */ 373 374 en = ndp->dc_entry; 375 376 pdp = dp = (struct dirent *)bp->b_data; 377 edp = bp->b_data + bp->b_validend; 378 enn = 0; 379 while (enn < en && (caddr_t)dp < edp) { 380 pdp = dp; 381 dp = (struct dirent *)((caddr_t)dp + dp->d_reclen); 382 enn++; 383 } 384 385 /* 386 * If the entry number was bigger than the number of 387 * entries in the block, or the cookie of the previous 388 * entry doesn't match, the directory cache is 389 * stale. Flush it and try again (i.e. go to 390 * the server). 391 */ 392 if ((caddr_t)dp >= edp || (caddr_t)dp + dp->d_reclen > edp || 393 (en > 0 && NFS_GETCOOKIE(pdp) != ndp->dc_cookie)) { 394 #ifdef DEBUG 395 printf("invalid cache: %p %p %p off %lx %lx\n", 396 pdp, dp, edp, 397 (unsigned long)uio->uio_offset, 398 (unsigned long)NFS_GETCOOKIE(pdp)); 399 #endif 400 brelse(bp); 401 nfs_invaldircache(vp, 0); 402 nfs_vinvalbuf(vp, 0, cred, p, 0); 403 goto diragain; 404 } 405 406 on = (caddr_t)dp - bp->b_data; 407 408 /* 409 * Cache all entries that may be exported to the 410 * user, as they may be thrown back at us. The 411 * NFSBIO_CACHECOOKIES flag indicates that all 412 * entries are being 'exported', so cache them all. 413 */ 414 415 if (en == 0 && pdp == dp) { 416 dp = (struct dirent *) 417 ((caddr_t)dp + dp->d_reclen); 418 enn++; 419 } 420 421 if (uio->uio_resid < (bp->b_validend - on)) { 422 n = uio->uio_resid; 423 enough = 1; 424 } else 425 n = bp->b_validend - on; 426 427 ep = bp->b_data + on + n; 428 429 /* 430 * Find last complete entry to copy, caching entries 431 * (if requested) as we go. 432 */ 433 434 while ((caddr_t)dp < ep && (caddr_t)dp + dp->d_reclen <= ep) { 435 if (cflag & NFSBIO_CACHECOOKIES) { 436 nndp = nfs_enterdircache(vp, NFS_GETCOOKIE(pdp), 437 ndp->dc_blkcookie, enn, bp->b_lblkno); 438 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) { 439 NFS_STASHCOOKIE32(pdp, 440 nndp->dc_cookie32); 441 } 442 } 443 pdp = dp; 444 dp = (struct dirent *)((caddr_t)dp + dp->d_reclen); 445 enn++; 446 } 447 448 /* 449 * If the last requested entry was not the last in the 450 * buffer (happens if NFS_DIRFRAGSIZ < NFS_DIRBLKSIZ), 451 * cache the cookie of the last requested one, and 452 * set of the offset to it. 453 */ 454 455 if ((on + n) < bp->b_validend) { 456 curoff = NFS_GETCOOKIE(pdp); 457 nndp = nfs_enterdircache(vp, curoff, ndp->dc_blkcookie, 458 enn, bp->b_lblkno); 459 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) { 460 NFS_STASHCOOKIE32(pdp, nndp->dc_cookie32); 461 curoff = nndp->dc_cookie32; 462 } 463 } else 464 curoff = bp->b_dcookie; 465 466 /* 467 * Always cache the entry for the next block, 468 * so that readaheads can use it. 469 */ 470 nndp = nfs_enterdircache(vp, bp->b_dcookie, bp->b_dcookie, 0,0); 471 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) { 472 if (curoff == bp->b_dcookie) { 473 NFS_STASHCOOKIE32(pdp, nndp->dc_cookie32); 474 curoff = nndp->dc_cookie32; 475 } 476 } 477 478 n = ((caddr_t)pdp + pdp->d_reclen) - (bp->b_data + on); 479 480 /* 481 * If not eof and read aheads are enabled, start one. 482 * (You need the current block first, so that you have the 483 * directory offset cookie of the next block.) 484 */ 485 if (nfs_numasync > 0 && nmp->nm_readahead > 0 && 486 np->n_direofoffset == 0 && !(np->n_flag & NQNFSNONCACHE)) { 487 rabp = nfs_getcacheblk(vp, nndp->dc_blkno, 488 NFS_DIRBLKSIZ, p); 489 if (rabp) { 490 if ((rabp->b_flags & (B_DONE | B_DELWRI)) == 0) { 491 rabp->b_dcookie = nndp->dc_cookie; 492 rabp->b_flags |= (B_READ | B_ASYNC); 493 if (nfs_asyncio(rabp, cred)) { 494 rabp->b_flags |= B_INVAL; 495 brelse(rabp); 496 } 497 } else 498 brelse(rabp); 499 } 500 } 501 got_buf = 1; 502 break; 503 default: 504 printf(" nfsbioread: type %x unexpected\n",vp->v_type); 505 break; 506 }; 507 508 if (n > 0) { 509 if (!baddr) 510 baddr = bp->b_data; 511 error = uiomove(baddr + on, (int)n, uio); 512 } 513 switch (vp->v_type) { 514 case VREG: 515 break; 516 case VLNK: 517 n = 0; 518 break; 519 case VDIR: 520 if (np->n_flag & NQNFSNONCACHE) 521 bp->b_flags |= B_INVAL; 522 uio->uio_offset = curoff; 523 if (enough) 524 n = 0; 525 break; 526 default: 527 printf(" nfsbioread: type %x unexpected\n",vp->v_type); 528 } 529 if (got_buf) 530 brelse(bp); 531 } while (error == 0 && uio->uio_resid > 0 && n > 0); 532 return (error); 533 } 534 535 /* 536 * Vnode op for write using bio 537 */ 538 int 539 nfs_write(v) 540 void *v; 541 { 542 struct vop_write_args /* { 543 struct vnode *a_vp; 544 struct uio *a_uio; 545 int a_ioflag; 546 struct ucred *a_cred; 547 } */ *ap = v; 548 register int biosize; 549 register struct uio *uio = ap->a_uio; 550 struct proc *p = uio->uio_procp; 551 register struct vnode *vp = ap->a_vp; 552 struct nfsnode *np = VTONFS(vp); 553 register struct ucred *cred = ap->a_cred; 554 int ioflag = ap->a_ioflag; 555 struct buf *bp; 556 struct vattr vattr; 557 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 558 daddr_t lbn, bn; 559 int n, on, error = 0, iomode, must_commit; 560 561 #ifdef DIAGNOSTIC 562 if (uio->uio_rw != UIO_WRITE) 563 panic("nfs_write mode"); 564 if (uio->uio_segflg == UIO_USERSPACE && uio->uio_procp != curproc) 565 panic("nfs_write proc"); 566 #endif 567 if (vp->v_type != VREG) 568 return (EIO); 569 if (np->n_flag & NWRITEERR) { 570 np->n_flag &= ~NWRITEERR; 571 return (np->n_error); 572 } 573 if ((nmp->nm_flag & NFSMNT_NFSV3) && 574 !(nmp->nm_iflag & NFSMNT_GOTFSINFO)) 575 (void)nfs_fsinfo(nmp, vp, cred, p); 576 if (ioflag & (IO_APPEND | IO_SYNC)) { 577 if (np->n_flag & NMODIFIED) { 578 np->n_attrstamp = 0; 579 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 580 if (error) 581 return (error); 582 } 583 if (ioflag & IO_APPEND) { 584 np->n_attrstamp = 0; 585 error = VOP_GETATTR(vp, &vattr, cred, p); 586 if (error) 587 return (error); 588 uio->uio_offset = np->n_size; 589 } 590 } 591 if (uio->uio_offset < 0) 592 return (EINVAL); 593 if ((uio->uio_offset + uio->uio_resid) > nmp->nm_maxfilesize) 594 return (EFBIG); 595 if (uio->uio_resid == 0) 596 return (0); 597 /* 598 * Maybe this should be above the vnode op call, but so long as 599 * file servers have no limits, i don't think it matters 600 */ 601 if (p && uio->uio_offset + uio->uio_resid > 602 p->p_rlimit[RLIMIT_FSIZE].rlim_cur) { 603 psignal(p, SIGXFSZ); 604 return (EFBIG); 605 } 606 /* 607 * I use nm_rsize, not nm_wsize so that all buffer cache blocks 608 * will be the same size within a filesystem. nfs_writerpc will 609 * still use nm_wsize when sizing the rpc's. 610 */ 611 biosize = nmp->nm_rsize; 612 do { 613 614 /* 615 * XXX make sure we aren't cached in the VM page cache 616 */ 617 (void)vnode_pager_uncache(vp); 618 619 /* 620 * Check for a valid write lease. 621 */ 622 if ((nmp->nm_flag & NFSMNT_NQNFS) && 623 NQNFS_CKINVALID(vp, np, ND_WRITE)) { 624 do { 625 error = nqnfs_getlease(vp, ND_WRITE, cred, p); 626 } while (error == NQNFS_EXPIRED); 627 if (error) 628 return (error); 629 if (np->n_lrev != np->n_brev || 630 (np->n_flag & NQNFSNONCACHE)) { 631 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 632 if (error) 633 return (error); 634 np->n_brev = np->n_lrev; 635 } 636 } 637 if ((np->n_flag & NQNFSNONCACHE) && uio->uio_iovcnt == 1) { 638 iomode = NFSV3WRITE_FILESYNC; 639 error = nfs_writerpc(vp, uio, cred, &iomode, &must_commit); 640 if (must_commit) 641 nfs_clearcommit(vp->v_mount); 642 return (error); 643 } 644 nfsstats.biocache_writes++; 645 lbn = uio->uio_offset / biosize; 646 on = uio->uio_offset & (biosize-1); 647 n = min((unsigned)(biosize - on), uio->uio_resid); 648 bn = lbn * (biosize / DEV_BSIZE); 649 again: 650 bp = nfs_getcacheblk(vp, bn, biosize, p); 651 if (!bp) 652 return (EINTR); 653 if (bp->b_wcred == NOCRED) { 654 crhold(cred); 655 bp->b_wcred = cred; 656 } 657 np->n_flag |= NMODIFIED; 658 if (uio->uio_offset + n > np->n_size) { 659 np->n_size = uio->uio_offset + n; 660 vnode_pager_setsize(vp, np->n_size); 661 } 662 663 /* 664 * If the new write will leave a contiguous dirty 665 * area, just update the b_dirtyoff and b_dirtyend, 666 * otherwise force a write rpc of the old dirty area. 667 */ 668 if (bp->b_dirtyend > 0 && 669 (on > bp->b_dirtyend || (on + n) < bp->b_dirtyoff)) { 670 bp->b_proc = p; 671 if (VOP_BWRITE(bp) == EINTR) 672 return (EINTR); 673 goto again; 674 } 675 676 /* 677 * Check for valid write lease and get one as required. 678 * In case getblk() and/or bwrite() delayed us. 679 */ 680 if ((nmp->nm_flag & NFSMNT_NQNFS) && 681 NQNFS_CKINVALID(vp, np, ND_WRITE)) { 682 do { 683 error = nqnfs_getlease(vp, ND_WRITE, cred, p); 684 } while (error == NQNFS_EXPIRED); 685 if (error) { 686 brelse(bp); 687 return (error); 688 } 689 if (np->n_lrev != np->n_brev || 690 (np->n_flag & NQNFSNONCACHE)) { 691 brelse(bp); 692 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 693 if (error) 694 return (error); 695 np->n_brev = np->n_lrev; 696 goto again; 697 } 698 } 699 error = uiomove((char *)bp->b_data + on, n, uio); 700 if (error) { 701 bp->b_flags |= B_ERROR; 702 brelse(bp); 703 return (error); 704 } 705 if (bp->b_dirtyend > 0) { 706 bp->b_dirtyoff = min(on, bp->b_dirtyoff); 707 bp->b_dirtyend = max((on + n), bp->b_dirtyend); 708 } else { 709 bp->b_dirtyoff = on; 710 bp->b_dirtyend = on + n; 711 } 712 if (bp->b_validend == 0 || bp->b_validend < bp->b_dirtyoff || 713 bp->b_validoff > bp->b_dirtyend) { 714 bp->b_validoff = bp->b_dirtyoff; 715 bp->b_validend = bp->b_dirtyend; 716 } else { 717 bp->b_validoff = min(bp->b_validoff, bp->b_dirtyoff); 718 bp->b_validend = max(bp->b_validend, bp->b_dirtyend); 719 } 720 721 /* 722 * Since this block is being modified, it must be written 723 * again and not just committed. 724 */ 725 bp->b_flags &= ~B_NEEDCOMMIT; 726 727 /* 728 * If the lease is non-cachable or IO_SYNC do bwrite(). 729 */ 730 if ((np->n_flag & NQNFSNONCACHE) || (ioflag & IO_SYNC)) { 731 bp->b_proc = p; 732 error = VOP_BWRITE(bp); 733 if (error) 734 return (error); 735 if (np->n_flag & NQNFSNONCACHE) { 736 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 737 if (error) 738 return (error); 739 } 740 } else if ((n + on) == biosize && 741 (nmp->nm_flag & NFSMNT_NQNFS) == 0) { 742 bp->b_proc = (struct proc *)0; 743 bp->b_flags |= B_ASYNC; 744 (void)nfs_writebp(bp, 0); 745 } else { 746 bdwrite(bp); 747 } 748 } while (uio->uio_resid > 0 && n > 0); 749 return (0); 750 } 751 752 /* 753 * Get an nfs cache block. 754 * Allocate a new one if the block isn't currently in the cache 755 * and return the block marked busy. If the calling process is 756 * interrupted by a signal for an interruptible mount point, return 757 * NULL. 758 */ 759 struct buf * 760 nfs_getcacheblk(vp, bn, size, p) 761 struct vnode *vp; 762 daddr_t bn; 763 int size; 764 struct proc *p; 765 { 766 register struct buf *bp; 767 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 768 769 if (nmp->nm_flag & NFSMNT_INT) { 770 bp = getblk(vp, bn, size, PCATCH, 0); 771 while (bp == (struct buf *)0) { 772 if (nfs_sigintr(nmp, (struct nfsreq *)0, p)) 773 return ((struct buf *)0); 774 bp = getblk(vp, bn, size, 0, 2 * hz); 775 } 776 } else 777 bp = getblk(vp, bn, size, 0, 0); 778 return (bp); 779 } 780 781 /* 782 * Flush and invalidate all dirty buffers. If another process is already 783 * doing the flush, just wait for completion. 784 */ 785 int 786 nfs_vinvalbuf(vp, flags, cred, p, intrflg) 787 struct vnode *vp; 788 int flags; 789 struct ucred *cred; 790 struct proc *p; 791 int intrflg; 792 { 793 register struct nfsnode *np = VTONFS(vp); 794 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 795 int error = 0, slpflag, slptimeo; 796 797 if ((nmp->nm_flag & NFSMNT_INT) == 0) 798 intrflg = 0; 799 if (intrflg) { 800 slpflag = PCATCH; 801 slptimeo = 2 * hz; 802 } else { 803 slpflag = 0; 804 slptimeo = 0; 805 } 806 /* 807 * First wait for any other process doing a flush to complete. 808 */ 809 while (np->n_flag & NFLUSHINPROG) { 810 np->n_flag |= NFLUSHWANT; 811 error = tsleep((caddr_t)&np->n_flag, PRIBIO + 2, "nfsvinval", 812 slptimeo); 813 if (error && intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) 814 return (EINTR); 815 } 816 817 /* 818 * Now, flush as required. 819 */ 820 np->n_flag |= NFLUSHINPROG; 821 error = vinvalbuf(vp, flags, cred, p, slpflag, 0); 822 while (error) { 823 if (intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) { 824 np->n_flag &= ~NFLUSHINPROG; 825 if (np->n_flag & NFLUSHWANT) { 826 np->n_flag &= ~NFLUSHWANT; 827 wakeup((caddr_t)&np->n_flag); 828 } 829 return (EINTR); 830 } 831 error = vinvalbuf(vp, flags, cred, p, 0, slptimeo); 832 } 833 np->n_flag &= ~(NMODIFIED | NFLUSHINPROG); 834 if (np->n_flag & NFLUSHWANT) { 835 np->n_flag &= ~NFLUSHWANT; 836 wakeup((caddr_t)&np->n_flag); 837 } 838 return (0); 839 } 840 841 /* 842 * Initiate asynchronous I/O. Return an error if no nfsiods are available. 843 * This is mainly to avoid queueing async I/O requests when the nfsiods 844 * are all hung on a dead server. 845 */ 846 int 847 nfs_asyncio(bp, cred) 848 register struct buf *bp; 849 struct ucred *cred; 850 { 851 register int i; 852 register struct nfsmount *nmp; 853 int gotiod, slpflag = 0, slptimeo = 0, error; 854 855 if (nfs_numasync == 0) 856 return (EIO); 857 858 859 nmp = VFSTONFS(bp->b_vp->v_mount); 860 again: 861 if (nmp->nm_flag & NFSMNT_INT) 862 slpflag = PCATCH; 863 gotiod = FALSE; 864 865 /* 866 * Find a free iod to process this request. 867 */ 868 869 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 870 if (nfs_iodwant[i]) { 871 /* 872 * Found one, so wake it up and tell it which 873 * mount to process. 874 */ 875 nfs_iodwant[i] = (struct proc *)0; 876 nfs_iodmount[i] = nmp; 877 nmp->nm_bufqiods++; 878 wakeup((caddr_t)&nfs_iodwant[i]); 879 gotiod = TRUE; 880 break; 881 } 882 /* 883 * If none are free, we may already have an iod working on this mount 884 * point. If so, it will process our request. 885 */ 886 if (!gotiod && nmp->nm_bufqiods > 0) 887 gotiod = TRUE; 888 889 /* 890 * If we have an iod which can process the request, then queue 891 * the buffer. 892 */ 893 if (gotiod) { 894 /* 895 * Ensure that the queue never grows too large. 896 */ 897 while (nmp->nm_bufqlen >= 2*nfs_numasync) { 898 nmp->nm_bufqwant = TRUE; 899 error = tsleep(&nmp->nm_bufq, slpflag | PRIBIO, 900 "nfsaio", slptimeo); 901 if (error) { 902 if (nfs_sigintr(nmp, NULL, bp->b_proc)) 903 return (EINTR); 904 if (slpflag == PCATCH) { 905 slpflag = 0; 906 slptimeo = 2 * hz; 907 } 908 } 909 /* 910 * We might have lost our iod while sleeping, 911 * so check and loop if nescessary. 912 */ 913 if (nmp->nm_bufqiods == 0) 914 goto again; 915 } 916 917 if (bp->b_flags & B_READ) { 918 if (bp->b_rcred == NOCRED && cred != NOCRED) { 919 crhold(cred); 920 bp->b_rcred = cred; 921 } 922 } else { 923 bp->b_flags |= B_WRITEINPROG; 924 if (bp->b_wcred == NOCRED && cred != NOCRED) { 925 crhold(cred); 926 bp->b_wcred = cred; 927 } 928 } 929 930 TAILQ_INSERT_TAIL(&nmp->nm_bufq, bp, b_freelist); 931 nmp->nm_bufqlen++; 932 return (0); 933 } 934 935 /* 936 * All the iods are busy on other mounts, so return EIO to 937 * force the caller to process the i/o synchronously. 938 */ 939 return (EIO); 940 } 941 942 /* 943 * Do an I/O operation to/from a cache block. This may be called 944 * synchronously or from an nfsiod. 945 */ 946 int 947 nfs_doio(bp, cr, p) 948 register struct buf *bp; 949 struct ucred *cr; 950 struct proc *p; 951 { 952 register struct uio *uiop; 953 register struct vnode *vp; 954 struct nfsnode *np; 955 struct nfsmount *nmp; 956 int error = 0, diff, len, iomode, must_commit = 0; 957 struct uio uio; 958 struct iovec io; 959 960 vp = bp->b_vp; 961 np = VTONFS(vp); 962 nmp = VFSTONFS(vp->v_mount); 963 uiop = &uio; 964 uiop->uio_iov = &io; 965 uiop->uio_iovcnt = 1; 966 uiop->uio_segflg = UIO_SYSSPACE; 967 uiop->uio_procp = p; 968 969 /* 970 * Historically, paging was done with physio, but no more... 971 */ 972 if (bp->b_flags & B_PHYS) { 973 /* 974 * ...though reading /dev/drum still gets us here. 975 */ 976 io.iov_len = uiop->uio_resid = bp->b_bcount; 977 /* mapping was done by vmapbuf() */ 978 io.iov_base = bp->b_data; 979 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; 980 if (bp->b_flags & B_READ) { 981 uiop->uio_rw = UIO_READ; 982 nfsstats.read_physios++; 983 error = nfs_readrpc(vp, uiop, cr); 984 } else { 985 iomode = NFSV3WRITE_DATASYNC; 986 uiop->uio_rw = UIO_WRITE; 987 nfsstats.write_physios++; 988 error = nfs_writerpc(vp, uiop, cr, &iomode, &must_commit); 989 } 990 if (error) { 991 bp->b_flags |= B_ERROR; 992 bp->b_error = error; 993 } 994 } else if (bp->b_flags & B_READ) { 995 io.iov_len = uiop->uio_resid = bp->b_bcount; 996 io.iov_base = bp->b_data; 997 uiop->uio_rw = UIO_READ; 998 switch (vp->v_type) { 999 case VREG: 1000 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; 1001 nfsstats.read_bios++; 1002 error = nfs_readrpc(vp, uiop, cr); 1003 if (!error) { 1004 bp->b_validoff = 0; 1005 if (uiop->uio_resid) { 1006 /* 1007 * If len > 0, there is a hole in the file and 1008 * no writes after the hole have been pushed to 1009 * the server yet. 1010 * Just zero fill the rest of the valid area. 1011 */ 1012 diff = bp->b_bcount - uiop->uio_resid; 1013 len = np->n_size - (((u_quad_t)bp->b_blkno) * DEV_BSIZE 1014 + diff); 1015 if (len > 0) { 1016 len = min(len, uiop->uio_resid); 1017 bzero((char *)bp->b_data + diff, len); 1018 bp->b_validend = diff + len; 1019 } else 1020 bp->b_validend = diff; 1021 } else 1022 bp->b_validend = bp->b_bcount; 1023 } 1024 if (p && (vp->v_flag & VTEXT) && 1025 (((nmp->nm_flag & NFSMNT_NQNFS) && 1026 NQNFS_CKINVALID(vp, np, ND_READ) && 1027 np->n_lrev != np->n_brev) || 1028 (!(nmp->nm_flag & NFSMNT_NQNFS) && 1029 np->n_mtime != np->n_vattr->va_mtime.tv_sec))) { 1030 uprintf("Process killed due to text file modification\n"); 1031 psignal(p, SIGKILL); 1032 p->p_holdcnt++; 1033 } 1034 break; 1035 case VLNK: 1036 uiop->uio_offset = (off_t)0; 1037 nfsstats.readlink_bios++; 1038 error = nfs_readlinkrpc(vp, uiop, cr); 1039 break; 1040 case VDIR: 1041 nfsstats.readdir_bios++; 1042 uiop->uio_offset = bp->b_dcookie; 1043 if (nmp->nm_flag & NFSMNT_RDIRPLUS) { 1044 error = nfs_readdirplusrpc(vp, uiop, cr); 1045 if (error == NFSERR_NOTSUPP) 1046 nmp->nm_flag &= ~NFSMNT_RDIRPLUS; 1047 } 1048 if ((nmp->nm_flag & NFSMNT_RDIRPLUS) == 0) 1049 error = nfs_readdirrpc(vp, uiop, cr); 1050 if (!error) { 1051 bp->b_dcookie = uiop->uio_offset; 1052 bp->b_validoff = 0; 1053 bp->b_validend = bp->b_bcount - uiop->uio_resid; 1054 } 1055 break; 1056 default: 1057 printf("nfs_doio: type %x unexpected\n",vp->v_type); 1058 break; 1059 }; 1060 if (error) { 1061 bp->b_flags |= B_ERROR; 1062 bp->b_error = error; 1063 } 1064 } else { 1065 io.iov_len = uiop->uio_resid = bp->b_dirtyend 1066 - bp->b_dirtyoff; 1067 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE 1068 + bp->b_dirtyoff; 1069 io.iov_base = (char *)bp->b_data + bp->b_dirtyoff; 1070 uiop->uio_rw = UIO_WRITE; 1071 nfsstats.write_bios++; 1072 if ((bp->b_flags & (B_ASYNC | B_NEEDCOMMIT | B_NOCACHE)) == B_ASYNC) 1073 iomode = NFSV3WRITE_UNSTABLE; 1074 else 1075 iomode = NFSV3WRITE_FILESYNC; 1076 bp->b_flags |= B_WRITEINPROG; 1077 #ifdef fvdl_debug 1078 printf("nfs_doio(%x): bp %x doff %d dend %d\n", 1079 vp, bp, bp->b_dirtyoff, bp->b_dirtyend); 1080 #endif 1081 error = nfs_writerpc(vp, uiop, cr, &iomode, &must_commit); 1082 if (!error && iomode == NFSV3WRITE_UNSTABLE) 1083 bp->b_flags |= B_NEEDCOMMIT; 1084 else 1085 bp->b_flags &= ~B_NEEDCOMMIT; 1086 bp->b_flags &= ~B_WRITEINPROG; 1087 1088 /* 1089 * For an interrupted write, the buffer is still valid and the 1090 * write hasn't been pushed to the server yet, so we can't set 1091 * B_ERROR and report the interruption by setting B_EINTR. For 1092 * the B_ASYNC case, B_EINTR is not relevant, so the rpc attempt 1093 * is essentially a noop. 1094 * For the case of a V3 write rpc not being committed to stable 1095 * storage, the block is still dirty and requires either a commit 1096 * rpc or another write rpc with iomode == NFSV3WRITE_FILESYNC 1097 * before the block is reused. This is indicated by setting the 1098 * B_DELWRI and B_NEEDCOMMIT flags. 1099 */ 1100 if (error == EINTR || (!error && (bp->b_flags & B_NEEDCOMMIT))) { 1101 bp->b_flags |= B_DELWRI; 1102 1103 /* 1104 * Since for the B_ASYNC case, nfs_bwrite() has reassigned the 1105 * buffer to the clean list, we have to reassign it back to the 1106 * dirty one. Ugh. 1107 */ 1108 if (bp->b_flags & B_ASYNC) 1109 reassignbuf(bp, vp); 1110 else if (error) 1111 bp->b_flags |= B_EINTR; 1112 } else { 1113 if (error) { 1114 bp->b_flags |= B_ERROR; 1115 bp->b_error = np->n_error = error; 1116 np->n_flag |= NWRITEERR; 1117 } 1118 bp->b_dirtyoff = bp->b_dirtyend = 0; 1119 } 1120 } 1121 bp->b_resid = uiop->uio_resid; 1122 if (must_commit) 1123 nfs_clearcommit(vp->v_mount); 1124 biodone(bp); 1125 return (error); 1126 } 1127
Indexes created Mon Oct 13 21:09:55 GMT 2025