nfs_bio.c revision 1.35
1 /* $NetBSD: nfs_bio.c,v 1.35 1997/10/19 01:46:20 fvdl 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 for (nra = 0; nra < nmp->nm_readahead && 221 (lbn + 1 + nra) * biosize < np->n_size; nra++) { 222 rabn = (lbn + 1 + nra) * (biosize / DEV_BSIZE); 223 if (!incore(vp, rabn)) { 224 rabp = nfs_getcacheblk(vp, rabn, biosize, p); 225 if (!rabp) 226 return (EINTR); 227 if ((rabp->b_flags & (B_DELWRI | B_DONE)) == 0) { 228 rabp->b_flags |= (B_READ | B_ASYNC); 229 if (nfs_asyncio(rabp, cred)) { 230 rabp->b_flags |= B_INVAL; 231 brelse(rabp); 232 } 233 } else 234 brelse(rabp); 235 } 236 } 237 } 238 239 /* 240 * If the block is in the cache and has the required data 241 * in a valid region, just copy it out. 242 * Otherwise, get the block and write back/read in, 243 * as required. 244 */ 245 if ((bp = incore(vp, bn)) && 246 (bp->b_flags & (B_BUSY | B_WRITEINPROG)) == 247 (B_BUSY | B_WRITEINPROG)) 248 got_buf = 0; 249 else { 250 again: 251 bp = nfs_getcacheblk(vp, bn, biosize, p); 252 if (!bp) 253 return (EINTR); 254 got_buf = 1; 255 if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0) { 256 bp->b_flags |= B_READ; 257 not_readin = 0; 258 error = nfs_doio(bp, cred, p); 259 if (error) { 260 brelse(bp); 261 return (error); 262 } 263 } 264 } 265 n = min((unsigned)(biosize - on), uio->uio_resid); 266 diff = np->n_size - uio->uio_offset; 267 if (diff < n) 268 n = diff; 269 if (not_readin && n > 0) { 270 if (on < bp->b_validoff || (on + n) > bp->b_validend) { 271 if (!got_buf) { 272 bp = nfs_getcacheblk(vp, bn, biosize, p); 273 if (!bp) 274 return (EINTR); 275 got_buf = 1; 276 } 277 bp->b_flags |= B_INVAFTERWRITE; 278 if (bp->b_dirtyend > 0) { 279 if ((bp->b_flags & B_DELWRI) == 0) 280 panic("nfsbioread"); 281 if (VOP_BWRITE(bp) == EINTR) 282 return (EINTR); 283 } else 284 brelse(bp); 285 goto again; 286 } 287 } 288 vp->v_lastr = lbn; 289 diff = (on >= bp->b_validend) ? 0 : (bp->b_validend - on); 290 if (diff < n) 291 n = diff; 292 break; 293 case VLNK: 294 nfsstats.biocache_readlinks++; 295 bp = nfs_getcacheblk(vp, (daddr_t)0, NFS_MAXPATHLEN, p); 296 if (!bp) 297 return (EINTR); 298 if ((bp->b_flags & B_DONE) == 0) { 299 bp->b_flags |= B_READ; 300 error = nfs_doio(bp, cred, p); 301 if (error) { 302 brelse(bp); 303 return (error); 304 } 305 } 306 n = min(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid); 307 got_buf = 1; 308 on = 0; 309 break; 310 case VDIR: 311 diragain: 312 nfsstats.biocache_readdirs++; 313 ndp = nfs_searchdircache(vp, uio->uio_offset, 314 (nmp->nm_flag & NFSMNT_XLATECOOKIE), 0); 315 if (!ndp) { 316 /* 317 * We've been handed a cookie that is not 318 * in the cache. If we're not translating 319 * 32 <-> 64, it may be a value that was 320 * flushed out of the cache because it grew 321 * too big. Let the server judge if it's 322 * valid or not. In the translation case, 323 * we have no way of validating this value, 324 * so punt. 325 */ 326 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) 327 return (EINVAL); 328 ndp = nfs_enterdircache(vp, uio->uio_offset, 329 uio->uio_offset, 0, 0); 330 } 331 332 if (uio->uio_offset != 0 && 333 ndp->dc_cookie == np->n_direofoffset) { 334 nfsstats.direofcache_hits++; 335 return (0); 336 } 337 338 bp = nfs_getcacheblk(vp, ndp->dc_blkno, NFS_DIRBLKSIZ, p); 339 if (!bp) 340 return (EINTR); 341 if ((bp->b_flags & B_DONE) == 0) { 342 bp->b_flags |= B_READ; 343 bp->b_dcookie = ndp->dc_blkcookie; 344 error = nfs_doio(bp, cred, p); 345 if (error) { 346 /* 347 * Yuck! The directory has been modified on the 348 * server. Punt and let the userland code 349 * deal with it. 350 */ 351 brelse(bp); 352 if (error == NFSERR_BAD_COOKIE) { 353 nfs_invaldircache(vp, 0); 354 nfs_vinvalbuf(vp, 0, cred, p, 1); 355 error = EINVAL; 356 } 357 return (error); 358 } 359 } 360 361 /* 362 * Find the entry we were looking for in the block. 363 */ 364 365 en = ndp->dc_entry; 366 367 pdp = dp = (struct dirent *)bp->b_data; 368 edp = bp->b_data + bp->b_validend; 369 enn = 0; 370 while (enn < en && (caddr_t)dp < edp) { 371 pdp = dp; 372 dp = (struct dirent *)((caddr_t)dp + dp->d_reclen); 373 enn++; 374 } 375 376 /* 377 * If the entry number was bigger than the number of 378 * entries in the block, or the cookie of the previous 379 * entry doesn't match, the directory cache is 380 * stale. Flush it and try again (i.e. go to 381 * the server). 382 */ 383 if ((caddr_t)dp >= edp || (caddr_t)dp + dp->d_reclen > edp || 384 (en > 0 && NFS_GETCOOKIE(pdp) != ndp->dc_cookie)) { 385 #ifdef DEBUG 386 printf("invalid cache: %p %p %p len %u off %lx %lx\n", 387 pdp, dp, edp, dp->d_reclen, 388 (unsigned long)uio->uio_offset, 389 (unsigned long)NFS_GETCOOKIE(pdp)); 390 #endif 391 brelse(bp); 392 nfs_invaldircache(vp, 0); 393 nfs_vinvalbuf(vp, 0, cred, p, 0); 394 goto diragain; 395 } 396 397 on = (caddr_t)dp - bp->b_data; 398 399 /* 400 * Cache all entries that may be exported to the 401 * user, as they may be thrown back at us. The 402 * NFSBIO_CACHECOOKIES flag indicates that all 403 * entries are being 'exported', so cache them all. 404 */ 405 406 if (en == 0 && pdp == dp) { 407 dp = (struct dirent *) 408 ((caddr_t)dp + dp->d_reclen); 409 enn++; 410 } 411 412 if (uio->uio_resid < (bp->b_validend - on)) { 413 n = uio->uio_resid; 414 enough = 1; 415 } else 416 n = bp->b_validend - on; 417 418 ep = bp->b_data + on + n; 419 420 /* 421 * Find last complete entry to copy, caching entries 422 * (if requested) as we go. 423 */ 424 425 while ((caddr_t)dp < ep && (caddr_t)dp + dp->d_reclen <= ep) { 426 if (cflag & NFSBIO_CACHECOOKIES) { 427 nndp = nfs_enterdircache(vp, NFS_GETCOOKIE(pdp), 428 ndp->dc_blkcookie, enn, bp->b_lblkno); 429 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) { 430 NFS_STASHCOOKIE32(pdp, 431 nndp->dc_cookie32); 432 } 433 } 434 pdp = dp; 435 dp = (struct dirent *)((caddr_t)dp + dp->d_reclen); 436 enn++; 437 } 438 439 /* 440 * If the last requested entry was not the last in the 441 * buffer (happens if NFS_DIRFRAGSIZ < NFS_DIRBLKSIZ), 442 * cache the cookie of the last requested one, and 443 * set of the offset to it. 444 */ 445 446 if ((on + n) < bp->b_validend) { 447 curoff = NFS_GETCOOKIE(pdp); 448 nndp = nfs_enterdircache(vp, curoff, ndp->dc_blkcookie, 449 enn, bp->b_lblkno); 450 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) { 451 NFS_STASHCOOKIE32(pdp, nndp->dc_cookie32); 452 curoff = nndp->dc_cookie32; 453 } 454 } else 455 curoff = bp->b_dcookie; 456 457 /* 458 * Always cache the entry for the next block, 459 * so that readaheads can use it. 460 */ 461 nndp = nfs_enterdircache(vp, bp->b_dcookie, bp->b_dcookie, 0,0); 462 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) { 463 if (curoff == bp->b_dcookie) { 464 NFS_STASHCOOKIE32(pdp, nndp->dc_cookie32); 465 curoff = nndp->dc_cookie32; 466 } 467 } 468 469 n = ((caddr_t)pdp + pdp->d_reclen) - (bp->b_data + on); 470 471 /* 472 * If not eof and read aheads are enabled, start one. 473 * (You need the current block first, so that you have the 474 * directory offset cookie of the next block.) 475 */ 476 if (nfs_numasync > 0 && nmp->nm_readahead > 0 && 477 np->n_direofoffset == 0 && !(np->n_flag & NQNFSNONCACHE)) { 478 rabp = nfs_getcacheblk(vp, nndp->dc_blkno, 479 NFS_DIRBLKSIZ, p); 480 if (rabp) { 481 if ((rabp->b_flags & (B_DONE | B_DELWRI)) == 0) { 482 rabp->b_dcookie = nndp->dc_cookie; 483 rabp->b_flags |= (B_READ | B_ASYNC); 484 if (nfs_asyncio(rabp, cred)) { 485 rabp->b_flags |= B_INVAL; 486 brelse(rabp); 487 } 488 } else 489 brelse(rabp); 490 } 491 } 492 got_buf = 1; 493 break; 494 default: 495 printf(" nfsbioread: type %x unexpected\n",vp->v_type); 496 break; 497 }; 498 499 if (n > 0) { 500 if (!baddr) 501 baddr = bp->b_data; 502 error = uiomove(baddr + on, (int)n, uio); 503 } 504 switch (vp->v_type) { 505 case VREG: 506 break; 507 case VLNK: 508 n = 0; 509 break; 510 case VDIR: 511 if (np->n_flag & NQNFSNONCACHE) 512 bp->b_flags |= B_INVAL; 513 uio->uio_offset = curoff; 514 if (enough) 515 n = 0; 516 break; 517 default: 518 printf(" nfsbioread: type %x unexpected\n",vp->v_type); 519 } 520 if (got_buf) 521 brelse(bp); 522 } while (error == 0 && uio->uio_resid > 0 && n > 0); 523 return (error); 524 } 525 526 /* 527 * Vnode op for write using bio 528 */ 529 int 530 nfs_write(v) 531 void *v; 532 { 533 struct vop_write_args /* { 534 struct vnode *a_vp; 535 struct uio *a_uio; 536 int a_ioflag; 537 struct ucred *a_cred; 538 } */ *ap = v; 539 register int biosize; 540 register struct uio *uio = ap->a_uio; 541 struct proc *p = uio->uio_procp; 542 register struct vnode *vp = ap->a_vp; 543 struct nfsnode *np = VTONFS(vp); 544 register struct ucred *cred = ap->a_cred; 545 int ioflag = ap->a_ioflag; 546 struct buf *bp; 547 struct vattr vattr; 548 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 549 daddr_t lbn, bn; 550 int n, on, error = 0, iomode, must_commit; 551 552 #ifdef DIAGNOSTIC 553 if (uio->uio_rw != UIO_WRITE) 554 panic("nfs_write mode"); 555 if (uio->uio_segflg == UIO_USERSPACE && uio->uio_procp != curproc) 556 panic("nfs_write proc"); 557 #endif 558 if (vp->v_type != VREG) 559 return (EIO); 560 if (np->n_flag & NWRITEERR) { 561 np->n_flag &= ~NWRITEERR; 562 return (np->n_error); 563 } 564 if ((nmp->nm_flag & NFSMNT_NFSV3) && 565 !(nmp->nm_iflag & NFSMNT_GOTFSINFO)) 566 (void)nfs_fsinfo(nmp, vp, cred, p); 567 if (ioflag & (IO_APPEND | IO_SYNC)) { 568 if (np->n_flag & NMODIFIED) { 569 np->n_attrstamp = 0; 570 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 571 if (error) 572 return (error); 573 } 574 if (ioflag & IO_APPEND) { 575 np->n_attrstamp = 0; 576 error = VOP_GETATTR(vp, &vattr, cred, p); 577 if (error) 578 return (error); 579 uio->uio_offset = np->n_size; 580 } 581 } 582 if (uio->uio_offset < 0) 583 return (EINVAL); 584 if ((uio->uio_offset + uio->uio_resid) > nmp->nm_maxfilesize) 585 return (EFBIG); 586 if (uio->uio_resid == 0) 587 return (0); 588 /* 589 * Maybe this should be above the vnode op call, but so long as 590 * file servers have no limits, i don't think it matters 591 */ 592 if (p && uio->uio_offset + uio->uio_resid > 593 p->p_rlimit[RLIMIT_FSIZE].rlim_cur) { 594 psignal(p, SIGXFSZ); 595 return (EFBIG); 596 } 597 /* 598 * I use nm_rsize, not nm_wsize so that all buffer cache blocks 599 * will be the same size within a filesystem. nfs_writerpc will 600 * still use nm_wsize when sizing the rpc's. 601 */ 602 biosize = nmp->nm_rsize; 603 do { 604 605 /* 606 * XXX make sure we aren't cached in the VM page cache 607 */ 608 (void)vnode_pager_uncache(vp); 609 610 /* 611 * Check for a valid write lease. 612 */ 613 if ((nmp->nm_flag & NFSMNT_NQNFS) && 614 NQNFS_CKINVALID(vp, np, ND_WRITE)) { 615 do { 616 error = nqnfs_getlease(vp, ND_WRITE, cred, p); 617 } while (error == NQNFS_EXPIRED); 618 if (error) 619 return (error); 620 if (np->n_lrev != np->n_brev || 621 (np->n_flag & NQNFSNONCACHE)) { 622 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 623 if (error) 624 return (error); 625 np->n_brev = np->n_lrev; 626 } 627 } 628 if ((np->n_flag & NQNFSNONCACHE) && uio->uio_iovcnt == 1) { 629 iomode = NFSV3WRITE_FILESYNC; 630 error = nfs_writerpc(vp, uio, cred, &iomode, &must_commit); 631 if (must_commit) 632 nfs_clearcommit(vp->v_mount); 633 return (error); 634 } 635 nfsstats.biocache_writes++; 636 lbn = uio->uio_offset / biosize; 637 on = uio->uio_offset & (biosize-1); 638 n = min((unsigned)(biosize - on), uio->uio_resid); 639 bn = lbn * (biosize / DEV_BSIZE); 640 again: 641 bp = nfs_getcacheblk(vp, bn, biosize, p); 642 if (!bp) 643 return (EINTR); 644 if (bp->b_wcred == NOCRED) { 645 crhold(cred); 646 bp->b_wcred = cred; 647 } 648 np->n_flag |= NMODIFIED; 649 if (uio->uio_offset + n > np->n_size) { 650 np->n_size = uio->uio_offset + n; 651 vnode_pager_setsize(vp, np->n_size); 652 } 653 654 /* 655 * If the new write will leave a contiguous dirty 656 * area, just update the b_dirtyoff and b_dirtyend, 657 * otherwise force a write rpc of the old dirty area. 658 */ 659 if (bp->b_dirtyend > 0 && 660 (on > bp->b_dirtyend || (on + n) < bp->b_dirtyoff)) { 661 bp->b_proc = p; 662 if (VOP_BWRITE(bp) == EINTR) 663 return (EINTR); 664 goto again; 665 } 666 667 /* 668 * Check for valid write lease and get one as required. 669 * In case getblk() and/or bwrite() delayed us. 670 */ 671 if ((nmp->nm_flag & NFSMNT_NQNFS) && 672 NQNFS_CKINVALID(vp, np, ND_WRITE)) { 673 do { 674 error = nqnfs_getlease(vp, ND_WRITE, cred, p); 675 } while (error == NQNFS_EXPIRED); 676 if (error) { 677 brelse(bp); 678 return (error); 679 } 680 if (np->n_lrev != np->n_brev || 681 (np->n_flag & NQNFSNONCACHE)) { 682 brelse(bp); 683 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 684 if (error) 685 return (error); 686 np->n_brev = np->n_lrev; 687 goto again; 688 } 689 } 690 error = uiomove((char *)bp->b_data + on, n, uio); 691 if (error) { 692 bp->b_flags |= B_ERROR; 693 brelse(bp); 694 return (error); 695 } 696 if (bp->b_dirtyend > 0) { 697 bp->b_dirtyoff = min(on, bp->b_dirtyoff); 698 bp->b_dirtyend = max((on + n), bp->b_dirtyend); 699 } else { 700 bp->b_dirtyoff = on; 701 bp->b_dirtyend = on + n; 702 } 703 if (bp->b_validend == 0 || bp->b_validend < bp->b_dirtyoff || 704 bp->b_validoff > bp->b_dirtyend) { 705 bp->b_validoff = bp->b_dirtyoff; 706 bp->b_validend = bp->b_dirtyend; 707 } else { 708 bp->b_validoff = min(bp->b_validoff, bp->b_dirtyoff); 709 bp->b_validend = max(bp->b_validend, bp->b_dirtyend); 710 } 711 712 /* 713 * Since this block is being modified, it must be written 714 * again and not just committed. 715 */ 716 bp->b_flags &= ~B_NEEDCOMMIT; 717 718 /* 719 * If the lease is non-cachable or IO_SYNC do bwrite(). 720 */ 721 if ((np->n_flag & NQNFSNONCACHE) || (ioflag & IO_SYNC)) { 722 bp->b_proc = p; 723 error = VOP_BWRITE(bp); 724 if (error) 725 return (error); 726 if (np->n_flag & NQNFSNONCACHE) { 727 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 728 if (error) 729 return (error); 730 } 731 } else if ((n + on) == biosize && 732 (nmp->nm_flag & NFSMNT_NQNFS) == 0) { 733 bp->b_proc = (struct proc *)0; 734 bp->b_flags |= B_ASYNC; 735 (void)nfs_writebp(bp, 0); 736 } else { 737 bdwrite(bp); 738 } 739 } while (uio->uio_resid > 0 && n > 0); 740 return (0); 741 } 742 743 /* 744 * Get an nfs cache block. 745 * Allocate a new one if the block isn't currently in the cache 746 * and return the block marked busy. If the calling process is 747 * interrupted by a signal for an interruptible mount point, return 748 * NULL. 749 */ 750 struct buf * 751 nfs_getcacheblk(vp, bn, size, p) 752 struct vnode *vp; 753 daddr_t bn; 754 int size; 755 struct proc *p; 756 { 757 register struct buf *bp; 758 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 759 760 if (nmp->nm_flag & NFSMNT_INT) { 761 bp = getblk(vp, bn, size, PCATCH, 0); 762 while (bp == (struct buf *)0) { 763 if (nfs_sigintr(nmp, (struct nfsreq *)0, p)) 764 return ((struct buf *)0); 765 bp = getblk(vp, bn, size, 0, 2 * hz); 766 } 767 } else 768 bp = getblk(vp, bn, size, 0, 0); 769 return (bp); 770 } 771 772 /* 773 * Flush and invalidate all dirty buffers. If another process is already 774 * doing the flush, just wait for completion. 775 */ 776 int 777 nfs_vinvalbuf(vp, flags, cred, p, intrflg) 778 struct vnode *vp; 779 int flags; 780 struct ucred *cred; 781 struct proc *p; 782 int intrflg; 783 { 784 register struct nfsnode *np = VTONFS(vp); 785 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 786 int error = 0, slpflag, slptimeo; 787 788 if ((nmp->nm_flag & NFSMNT_INT) == 0) 789 intrflg = 0; 790 if (intrflg) { 791 slpflag = PCATCH; 792 slptimeo = 2 * hz; 793 } else { 794 slpflag = 0; 795 slptimeo = 0; 796 } 797 /* 798 * First wait for any other process doing a flush to complete. 799 */ 800 while (np->n_flag & NFLUSHINPROG) { 801 np->n_flag |= NFLUSHWANT; 802 error = tsleep((caddr_t)&np->n_flag, PRIBIO + 2, "nfsvinval", 803 slptimeo); 804 if (error && intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) 805 return (EINTR); 806 } 807 808 /* 809 * Now, flush as required. 810 */ 811 np->n_flag |= NFLUSHINPROG; 812 error = vinvalbuf(vp, flags, cred, p, slpflag, 0); 813 while (error) { 814 if (intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) { 815 np->n_flag &= ~NFLUSHINPROG; 816 if (np->n_flag & NFLUSHWANT) { 817 np->n_flag &= ~NFLUSHWANT; 818 wakeup((caddr_t)&np->n_flag); 819 } 820 return (EINTR); 821 } 822 error = vinvalbuf(vp, flags, cred, p, 0, slptimeo); 823 } 824 np->n_flag &= ~(NMODIFIED | NFLUSHINPROG); 825 if (np->n_flag & NFLUSHWANT) { 826 np->n_flag &= ~NFLUSHWANT; 827 wakeup((caddr_t)&np->n_flag); 828 } 829 return (0); 830 } 831 832 /* 833 * Initiate asynchronous I/O. Return an error if no nfsiods are available. 834 * This is mainly to avoid queueing async I/O requests when the nfsiods 835 * are all hung on a dead server. 836 */ 837 int 838 nfs_asyncio(bp, cred) 839 register struct buf *bp; 840 struct ucred *cred; 841 { 842 register int i; 843 register struct nfsmount *nmp; 844 int gotiod, slpflag = 0, slptimeo = 0, error; 845 846 if (nfs_numasync == 0) 847 return (EIO); 848 849 850 nmp = VFSTONFS(bp->b_vp->v_mount); 851 again: 852 if (nmp->nm_flag & NFSMNT_INT) 853 slpflag = PCATCH; 854 gotiod = FALSE; 855 856 /* 857 * Find a free iod to process this request. 858 */ 859 860 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 861 if (nfs_iodwant[i]) { 862 /* 863 * Found one, so wake it up and tell it which 864 * mount to process. 865 */ 866 nfs_iodwant[i] = (struct proc *)0; 867 nfs_iodmount[i] = nmp; 868 nmp->nm_bufqiods++; 869 wakeup((caddr_t)&nfs_iodwant[i]); 870 gotiod = TRUE; 871 break; 872 } 873 /* 874 * If none are free, we may already have an iod working on this mount 875 * point. If so, it will process our request. 876 */ 877 if (!gotiod && nmp->nm_bufqiods > 0) 878 gotiod = TRUE; 879 880 /* 881 * If we have an iod which can process the request, then queue 882 * the buffer. 883 */ 884 if (gotiod) { 885 /* 886 * Ensure that the queue never grows too large. 887 */ 888 while (nmp->nm_bufqlen >= 2*nfs_numasync) { 889 nmp->nm_bufqwant = TRUE; 890 error = tsleep(&nmp->nm_bufq, slpflag | PRIBIO, 891 "nfsaio", slptimeo); 892 if (error) { 893 if (nfs_sigintr(nmp, NULL, bp->b_proc)) 894 return (EINTR); 895 if (slpflag == PCATCH) { 896 slpflag = 0; 897 slptimeo = 2 * hz; 898 } 899 } 900 /* 901 * We might have lost our iod while sleeping, 902 * so check and loop if nescessary. 903 */ 904 if (nmp->nm_bufqiods == 0) 905 goto again; 906 } 907 908 if (bp->b_flags & B_READ) { 909 if (bp->b_rcred == NOCRED && cred != NOCRED) { 910 crhold(cred); 911 bp->b_rcred = cred; 912 } 913 } else { 914 bp->b_flags |= B_WRITEINPROG; 915 if (bp->b_wcred == NOCRED && cred != NOCRED) { 916 crhold(cred); 917 bp->b_wcred = cred; 918 } 919 } 920 921 TAILQ_INSERT_TAIL(&nmp->nm_bufq, bp, b_freelist); 922 nmp->nm_bufqlen++; 923 return (0); 924 } 925 926 /* 927 * All the iods are busy on other mounts, so return EIO to 928 * force the caller to process the i/o synchronously. 929 */ 930 return (EIO); 931 } 932 933 /* 934 * Do an I/O operation to/from a cache block. This may be called 935 * synchronously or from an nfsiod. 936 */ 937 int 938 nfs_doio(bp, cr, p) 939 register struct buf *bp; 940 struct ucred *cr; 941 struct proc *p; 942 { 943 register struct uio *uiop; 944 register struct vnode *vp; 945 struct nfsnode *np; 946 struct nfsmount *nmp; 947 int error = 0, diff, len, iomode, must_commit = 0; 948 struct uio uio; 949 struct iovec io; 950 951 vp = bp->b_vp; 952 np = VTONFS(vp); 953 nmp = VFSTONFS(vp->v_mount); 954 uiop = &uio; 955 uiop->uio_iov = &io; 956 uiop->uio_iovcnt = 1; 957 uiop->uio_segflg = UIO_SYSSPACE; 958 uiop->uio_procp = p; 959 960 /* 961 * Historically, paging was done with physio, but no more... 962 */ 963 if (bp->b_flags & B_PHYS) { 964 /* 965 * ...though reading /dev/drum still gets us here. 966 */ 967 io.iov_len = uiop->uio_resid = bp->b_bcount; 968 /* mapping was done by vmapbuf() */ 969 io.iov_base = bp->b_data; 970 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; 971 if (bp->b_flags & B_READ) { 972 uiop->uio_rw = UIO_READ; 973 nfsstats.read_physios++; 974 error = nfs_readrpc(vp, uiop, cr); 975 } else { 976 iomode = NFSV3WRITE_DATASYNC; 977 uiop->uio_rw = UIO_WRITE; 978 nfsstats.write_physios++; 979 error = nfs_writerpc(vp, uiop, cr, &iomode, &must_commit); 980 } 981 if (error) { 982 bp->b_flags |= B_ERROR; 983 bp->b_error = error; 984 } 985 } else if (bp->b_flags & B_READ) { 986 io.iov_len = uiop->uio_resid = bp->b_bcount; 987 io.iov_base = bp->b_data; 988 uiop->uio_rw = UIO_READ; 989 switch (vp->v_type) { 990 case VREG: 991 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; 992 nfsstats.read_bios++; 993 error = nfs_readrpc(vp, uiop, cr); 994 if (!error) { 995 bp->b_validoff = 0; 996 if (uiop->uio_resid) { 997 /* 998 * If len > 0, there is a hole in the file and 999 * no writes after the hole have been pushed to 1000 * the server yet. 1001 * Just zero fill the rest of the valid area. 1002 */ 1003 diff = bp->b_bcount - uiop->uio_resid; 1004 len = np->n_size - (((u_quad_t)bp->b_blkno) * DEV_BSIZE 1005 + diff); 1006 if (len > 0) { 1007 len = min(len, uiop->uio_resid); 1008 bzero((char *)bp->b_data + diff, len); 1009 bp->b_validend = diff + len; 1010 } else 1011 bp->b_validend = diff; 1012 } else 1013 bp->b_validend = bp->b_bcount; 1014 } 1015 if (p && (vp->v_flag & VTEXT) && 1016 (((nmp->nm_flag & NFSMNT_NQNFS) && 1017 NQNFS_CKINVALID(vp, np, ND_READ) && 1018 np->n_lrev != np->n_brev) || 1019 (!(nmp->nm_flag & NFSMNT_NQNFS) && 1020 np->n_mtime != np->n_vattr->va_mtime.tv_sec))) { 1021 uprintf("Process killed due to text file modification\n"); 1022 psignal(p, SIGKILL); 1023 p->p_holdcnt++; 1024 } 1025 break; 1026 case VLNK: 1027 uiop->uio_offset = (off_t)0; 1028 nfsstats.readlink_bios++; 1029 error = nfs_readlinkrpc(vp, uiop, cr); 1030 break; 1031 case VDIR: 1032 nfsstats.readdir_bios++; 1033 uiop->uio_offset = bp->b_dcookie; 1034 if (nmp->nm_flag & NFSMNT_RDIRPLUS) { 1035 error = nfs_readdirplusrpc(vp, uiop, cr); 1036 if (error == NFSERR_NOTSUPP) 1037 nmp->nm_flag &= ~NFSMNT_RDIRPLUS; 1038 } 1039 if ((nmp->nm_flag & NFSMNT_RDIRPLUS) == 0) 1040 error = nfs_readdirrpc(vp, uiop, cr); 1041 if (!error) { 1042 bp->b_dcookie = uiop->uio_offset; 1043 bp->b_validoff = 0; 1044 bp->b_validend = bp->b_bcount - uiop->uio_resid; 1045 } 1046 break; 1047 default: 1048 printf("nfs_doio: type %x unexpected\n",vp->v_type); 1049 break; 1050 }; 1051 if (error) { 1052 bp->b_flags |= B_ERROR; 1053 bp->b_error = error; 1054 } 1055 } else { 1056 io.iov_len = uiop->uio_resid = bp->b_dirtyend 1057 - bp->b_dirtyoff; 1058 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE 1059 + bp->b_dirtyoff; 1060 io.iov_base = (char *)bp->b_data + bp->b_dirtyoff; 1061 uiop->uio_rw = UIO_WRITE; 1062 nfsstats.write_bios++; 1063 if ((bp->b_flags & (B_ASYNC | B_NEEDCOMMIT | B_NOCACHE)) == B_ASYNC) 1064 iomode = NFSV3WRITE_UNSTABLE; 1065 else 1066 iomode = NFSV3WRITE_FILESYNC; 1067 bp->b_flags |= B_WRITEINPROG; 1068 #ifdef fvdl_debug 1069 printf("nfs_doio(%x): bp %x doff %d dend %d\n", 1070 vp, bp, bp->b_dirtyoff, bp->b_dirtyend); 1071 #endif 1072 error = nfs_writerpc(vp, uiop, cr, &iomode, &must_commit); 1073 if (!error && iomode == NFSV3WRITE_UNSTABLE) 1074 bp->b_flags |= B_NEEDCOMMIT; 1075 else 1076 bp->b_flags &= ~B_NEEDCOMMIT; 1077 bp->b_flags &= ~B_WRITEINPROG; 1078 1079 /* 1080 * For an interrupted write, the buffer is still valid and the 1081 * write hasn't been pushed to the server yet, so we can't set 1082 * B_ERROR and report the interruption by setting B_EINTR. For 1083 * the B_ASYNC case, B_EINTR is not relevant, so the rpc attempt 1084 * is essentially a noop. 1085 * For the case of a V3 write rpc not being committed to stable 1086 * storage, the block is still dirty and requires either a commit 1087 * rpc or another write rpc with iomode == NFSV3WRITE_FILESYNC 1088 * before the block is reused. This is indicated by setting the 1089 * B_DELWRI and B_NEEDCOMMIT flags. 1090 */ 1091 if (error == EINTR || (!error && (bp->b_flags & B_NEEDCOMMIT))) { 1092 bp->b_flags |= B_DELWRI; 1093 1094 /* 1095 * Since for the B_ASYNC case, nfs_bwrite() has reassigned the 1096 * buffer to the clean list, we have to reassign it back to the 1097 * dirty one. Ugh. 1098 */ 1099 if (bp->b_flags & B_ASYNC) 1100 reassignbuf(bp, vp); 1101 else if (error) 1102 bp->b_flags |= B_EINTR; 1103 } else { 1104 if (error) { 1105 bp->b_flags |= B_ERROR; 1106 bp->b_error = np->n_error = error; 1107 np->n_flag |= NWRITEERR; 1108 } 1109 bp->b_dirtyoff = bp->b_dirtyend = 0; 1110 } 1111 } 1112 bp->b_resid = uiop->uio_resid; 1113 if (must_commit) 1114 nfs_clearcommit(vp->v_mount); 1115 biodone(bp); 1116 return (error); 1117 } 1118
Indexes created Mon Oct 20 08:09:54 GMT 2025