nfs_bio.c revision 1.34
1 /* $NetBSD: nfs_bio.c,v 1.34 1997/10/10 01:53:18 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; 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); 132 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 133 if (error) 134 return (error); 135 } 136 np->n_attrstamp = 0; 137 error = VOP_GETATTR(vp, &vattr, cred, p); 138 if (error) 139 return (error); 140 np->n_mtime = vattr.va_mtime.tv_sec; 141 } else { 142 error = VOP_GETATTR(vp, &vattr, cred, p); 143 if (error) 144 return (error); 145 if (np->n_mtime != vattr.va_mtime.tv_sec) { 146 if (vp->v_type == VDIR) 147 nfs_invaldircache(vp); 148 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 149 if (error) 150 return (error); 151 np->n_mtime = vattr.va_mtime.tv_sec; 152 } 153 } 154 } 155 do { 156 157 /* 158 * Get a valid lease. If cached data is stale, flush it. 159 */ 160 if (nmp->nm_flag & NFSMNT_NQNFS) { 161 if (NQNFS_CKINVALID(vp, np, ND_READ)) { 162 do { 163 error = nqnfs_getlease(vp, ND_READ, cred, p); 164 } while (error == NQNFS_EXPIRED); 165 if (error) 166 return (error); 167 if (np->n_lrev != np->n_brev || 168 (np->n_flag & NQNFSNONCACHE) || 169 ((np->n_flag & NMODIFIED) && vp->v_type == VDIR)) { 170 if (vp->v_type == VDIR) 171 nfs_invaldircache(vp); 172 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 173 if (error) 174 return (error); 175 np->n_brev = np->n_lrev; 176 } 177 } else if (vp->v_type == VDIR && (np->n_flag & NMODIFIED)) { 178 nfs_invaldircache(vp); 179 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 180 if (error) 181 return (error); 182 } 183 } 184 /* 185 * Don't cache symlinks. 186 */ 187 if (np->n_flag & NQNFSNONCACHE 188 || ((vp->v_flag & VROOT) && vp->v_type == VLNK)) { 189 switch (vp->v_type) { 190 case VREG: 191 return (nfs_readrpc(vp, uio, cred)); 192 case VLNK: 193 return (nfs_readlinkrpc(vp, uio, cred)); 194 case VDIR: 195 break; 196 default: 197 printf(" NQNFSNONCACHE: type %x unexpected\n", 198 vp->v_type); 199 }; 200 } 201 baddr = (caddr_t)0; 202 switch (vp->v_type) { 203 case VREG: 204 nfsstats.biocache_reads++; 205 lbn = uio->uio_offset / biosize; 206 on = uio->uio_offset & (biosize - 1); 207 bn = lbn * (biosize / DEV_BSIZE); 208 not_readin = 1; 209 210 /* 211 * Start the read ahead(s), as required. 212 */ 213 if (nfs_numasync > 0 && nmp->nm_readahead > 0) { 214 for (nra = 0; nra < nmp->nm_readahead && 215 (lbn + 1 + nra) * biosize < np->n_size; nra++) { 216 rabn = (lbn + 1 + nra) * (biosize / DEV_BSIZE); 217 if (!incore(vp, rabn)) { 218 rabp = nfs_getcacheblk(vp, rabn, biosize, p); 219 if (!rabp) 220 return (EINTR); 221 if ((rabp->b_flags & (B_DELWRI | B_DONE)) == 0) { 222 rabp->b_flags |= (B_READ | B_ASYNC); 223 if (nfs_asyncio(rabp, cred)) { 224 rabp->b_flags |= B_INVAL; 225 brelse(rabp); 226 } 227 } else 228 brelse(rabp); 229 } 230 } 231 } 232 233 /* 234 * If the block is in the cache and has the required data 235 * in a valid region, just copy it out. 236 * Otherwise, get the block and write back/read in, 237 * as required. 238 */ 239 if ((bp = incore(vp, bn)) && 240 (bp->b_flags & (B_BUSY | B_WRITEINPROG)) == 241 (B_BUSY | B_WRITEINPROG)) 242 got_buf = 0; 243 else { 244 again: 245 bp = nfs_getcacheblk(vp, bn, biosize, p); 246 if (!bp) 247 return (EINTR); 248 got_buf = 1; 249 if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0) { 250 bp->b_flags |= B_READ; 251 not_readin = 0; 252 error = nfs_doio(bp, cred, p); 253 if (error) { 254 brelse(bp); 255 return (error); 256 } 257 } 258 } 259 n = min((unsigned)(biosize - on), uio->uio_resid); 260 diff = np->n_size - uio->uio_offset; 261 if (diff < n) 262 n = diff; 263 if (not_readin && n > 0) { 264 if (on < bp->b_validoff || (on + n) > bp->b_validend) { 265 if (!got_buf) { 266 bp = nfs_getcacheblk(vp, bn, biosize, p); 267 if (!bp) 268 return (EINTR); 269 got_buf = 1; 270 } 271 bp->b_flags |= B_INVAFTERWRITE; 272 if (bp->b_dirtyend > 0) { 273 if ((bp->b_flags & B_DELWRI) == 0) 274 panic("nfsbioread"); 275 if (VOP_BWRITE(bp) == EINTR) 276 return (EINTR); 277 } else 278 brelse(bp); 279 goto again; 280 } 281 } 282 vp->v_lastr = lbn; 283 diff = (on >= bp->b_validend) ? 0 : (bp->b_validend - on); 284 if (diff < n) 285 n = diff; 286 break; 287 case VLNK: 288 nfsstats.biocache_readlinks++; 289 bp = nfs_getcacheblk(vp, (daddr_t)0, NFS_MAXPATHLEN, p); 290 if (!bp) 291 return (EINTR); 292 if ((bp->b_flags & B_DONE) == 0) { 293 bp->b_flags |= B_READ; 294 error = nfs_doio(bp, cred, p); 295 if (error) { 296 brelse(bp); 297 return (error); 298 } 299 } 300 n = min(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid); 301 got_buf = 1; 302 on = 0; 303 break; 304 case VDIR: 305 diragain: 306 nfsstats.biocache_readdirs++; 307 if (uio->uio_offset != 0 && 308 uio->uio_offset == np->n_direofoffset) 309 return (0); 310 ndp = nfs_lookdircache(vp, uio->uio_offset, 0, 0, 1); 311 #ifdef DIAGNOSTIC 312 if (!ndp) 313 panic("nfs_bioread: bad dir cache"); 314 #endif 315 bp = nfs_getcacheblk(vp, ndp->dc_blkno, NFS_DIRBLKSIZ, p); 316 if (!bp) 317 return (EINTR); 318 if ((bp->b_flags & B_DONE) == 0) { 319 bp->b_flags |= B_READ; 320 bp->b_dcookie = ndp->dc_cookie; 321 error = nfs_doio(bp, cred, p); 322 if (error) { 323 /* 324 * Yuck! The directory has been modified on the 325 * server. Punt and let the userland code 326 * deal with it. 327 */ 328 brelse(bp); 329 if (error == NFSERR_BAD_COOKIE) { 330 nfs_invaldircache(vp); 331 nfs_vinvalbuf(vp, 0, cred, p, 1); 332 error = EINVAL; 333 } 334 return (error); 335 } 336 } 337 338 /* 339 * Find the entry we were looking for in the block. 340 */ 341 342 en = ndp->dc_entry; 343 344 pdp = dp = (struct dirent *)bp->b_data; 345 edp = bp->b_data + bp->b_validend; 346 enn = 0; 347 while (enn < en && (caddr_t)dp < edp) { 348 pdp = dp; 349 dp = (struct dirent *)((caddr_t)dp + dp->d_reclen); 350 enn++; 351 } 352 353 /* 354 * If the entry number was bigger than the number of 355 * entries in the block, or the cookie of the previous 356 * entry doesn't match, the directory cache is 357 * stale. Flush it and try again (i.e. go to 358 * the server). 359 */ 360 if ((caddr_t)dp >= edp || (caddr_t)dp + dp->d_reclen > edp || 361 (en > 0 && NFS_GETCOOKIE(pdp) != uio->uio_offset)) { 362 #ifdef DEBUG 363 printf("invalid cache: %p %p %p len %u off %lx %lx\n", 364 pdp, dp, edp, dp->d_reclen, 365 (unsigned long)uio->uio_offset, 366 (unsigned long)NFS_GETCOOKIE(pdp)); 367 #endif 368 brelse(bp); 369 nfs_invaldircache(vp); 370 nfs_vinvalbuf(vp, 0, cred, p, 0); 371 goto diragain; 372 } 373 374 on = (caddr_t)dp - bp->b_data; 375 376 /* 377 * Cache all entries that may be exported to the 378 * user, as they may be thrown back at us. The 379 * NFSBIO_CACHECOOKIES flag indicates that all 380 * entries are being 'exported', so cache them all. 381 */ 382 383 if (en == 0 && pdp == dp) { 384 dp = (struct dirent *) 385 ((caddr_t)dp + dp->d_reclen); 386 enn++; 387 } 388 389 if (uio->uio_resid < (bp->b_validend - on)) { 390 n = uio->uio_resid; 391 enough = 1; 392 } else 393 n = bp->b_validend - on; 394 395 ep = bp->b_data + on + n; 396 397 /* 398 * Find last complete entry to copy, caching entries 399 * (if requested) as we go. 400 */ 401 402 while ((caddr_t)dp < ep && (caddr_t)dp + dp->d_reclen <= ep) { 403 if (cflag & NFSBIO_CACHECOOKIES) 404 nfs_lookdircache(vp, NFS_GETCOOKIE(pdp), enn, 405 bp->b_lblkno, 1); 406 pdp = dp; 407 dp = (struct dirent *)((caddr_t)dp + dp->d_reclen); 408 enn++; 409 } 410 411 /* 412 * If the last requested entry was not the last in the 413 * buffer (happens if NFS_DIRFRAGSIZ < NFS_DIRBLKSIZ), 414 * cache the cookie of the last requested one, and 415 * set of the offset to it. 416 */ 417 418 if ((on + n) < bp->b_validend) { 419 curoff = NFS_GETCOOKIE(pdp); 420 nfs_lookdircache(vp, curoff, enn, bp->b_lblkno, 1); 421 } else 422 curoff = bp->b_dcookie; 423 424 n = ((caddr_t)pdp + pdp->d_reclen) - (bp->b_data + on); 425 426 /* 427 * If not eof and read aheads are enabled, start one. 428 * (You need the current block first, so that you have the 429 * directory offset cookie of the next block.) 430 */ 431 if (nfs_numasync > 0 && nmp->nm_readahead > 0 && 432 np->n_direofoffset == 0 && !(np->n_flag & NQNFSNONCACHE)) { 433 ndp = nfs_lookdircache(vp, bp->b_dcookie, 0, 0, 1); 434 rabp = nfs_getcacheblk(vp, ndp->dc_blkno, 435 NFS_DIRBLKSIZ, p); 436 if (rabp) { 437 if ((rabp->b_flags & (B_DONE | B_DELWRI)) == 0) { 438 rabp->b_dcookie = ndp->dc_cookie; 439 rabp->b_flags |= (B_READ | B_ASYNC); 440 if (nfs_asyncio(rabp, cred)) { 441 rabp->b_flags |= B_INVAL; 442 brelse(rabp); 443 } 444 } else 445 brelse(rabp); 446 } 447 } 448 got_buf = 1; 449 break; 450 default: 451 printf(" nfsbioread: type %x unexpected\n",vp->v_type); 452 break; 453 }; 454 455 if (n > 0) { 456 if (!baddr) 457 baddr = bp->b_data; 458 error = uiomove(baddr + on, (int)n, uio); 459 } 460 switch (vp->v_type) { 461 case VREG: 462 break; 463 case VLNK: 464 n = 0; 465 break; 466 case VDIR: 467 if (np->n_flag & NQNFSNONCACHE) 468 bp->b_flags |= B_INVAL; 469 uio->uio_offset = curoff; 470 if (enough) 471 n = 0; 472 break; 473 default: 474 printf(" nfsbioread: type %x unexpected\n",vp->v_type); 475 } 476 if (got_buf) 477 brelse(bp); 478 } while (error == 0 && uio->uio_resid > 0 && n > 0); 479 return (error); 480 } 481 482 /* 483 * Vnode op for write using bio 484 */ 485 int 486 nfs_write(v) 487 void *v; 488 { 489 struct vop_write_args /* { 490 struct vnode *a_vp; 491 struct uio *a_uio; 492 int a_ioflag; 493 struct ucred *a_cred; 494 } */ *ap = v; 495 register int biosize; 496 register struct uio *uio = ap->a_uio; 497 struct proc *p = uio->uio_procp; 498 register struct vnode *vp = ap->a_vp; 499 struct nfsnode *np = VTONFS(vp); 500 register struct ucred *cred = ap->a_cred; 501 int ioflag = ap->a_ioflag; 502 struct buf *bp; 503 struct vattr vattr; 504 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 505 daddr_t lbn, bn; 506 int n, on, error = 0, iomode, must_commit; 507 508 #ifdef DIAGNOSTIC 509 if (uio->uio_rw != UIO_WRITE) 510 panic("nfs_write mode"); 511 if (uio->uio_segflg == UIO_USERSPACE && uio->uio_procp != curproc) 512 panic("nfs_write proc"); 513 #endif 514 if (vp->v_type != VREG) 515 return (EIO); 516 if (np->n_flag & NWRITEERR) { 517 np->n_flag &= ~NWRITEERR; 518 return (np->n_error); 519 } 520 if ((nmp->nm_flag & NFSMNT_NFSV3) && 521 !(nmp->nm_iflag & NFSMNT_GOTFSINFO)) 522 (void)nfs_fsinfo(nmp, vp, cred, p); 523 if (ioflag & (IO_APPEND | IO_SYNC)) { 524 if (np->n_flag & NMODIFIED) { 525 np->n_attrstamp = 0; 526 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 527 if (error) 528 return (error); 529 } 530 if (ioflag & IO_APPEND) { 531 np->n_attrstamp = 0; 532 error = VOP_GETATTR(vp, &vattr, cred, p); 533 if (error) 534 return (error); 535 uio->uio_offset = np->n_size; 536 } 537 } 538 if (uio->uio_offset < 0) 539 return (EINVAL); 540 if ((uio->uio_offset + uio->uio_resid) > nmp->nm_maxfilesize) 541 return (EFBIG); 542 if (uio->uio_resid == 0) 543 return (0); 544 /* 545 * Maybe this should be above the vnode op call, but so long as 546 * file servers have no limits, i don't think it matters 547 */ 548 if (p && uio->uio_offset + uio->uio_resid > 549 p->p_rlimit[RLIMIT_FSIZE].rlim_cur) { 550 psignal(p, SIGXFSZ); 551 return (EFBIG); 552 } 553 /* 554 * I use nm_rsize, not nm_wsize so that all buffer cache blocks 555 * will be the same size within a filesystem. nfs_writerpc will 556 * still use nm_wsize when sizing the rpc's. 557 */ 558 biosize = nmp->nm_rsize; 559 do { 560 561 /* 562 * XXX make sure we aren't cached in the VM page cache 563 */ 564 (void)vnode_pager_uncache(vp); 565 566 /* 567 * Check for a valid write lease. 568 */ 569 if ((nmp->nm_flag & NFSMNT_NQNFS) && 570 NQNFS_CKINVALID(vp, np, ND_WRITE)) { 571 do { 572 error = nqnfs_getlease(vp, ND_WRITE, cred, p); 573 } while (error == NQNFS_EXPIRED); 574 if (error) 575 return (error); 576 if (np->n_lrev != np->n_brev || 577 (np->n_flag & NQNFSNONCACHE)) { 578 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 579 if (error) 580 return (error); 581 np->n_brev = np->n_lrev; 582 } 583 } 584 if ((np->n_flag & NQNFSNONCACHE) && uio->uio_iovcnt == 1) { 585 iomode = NFSV3WRITE_FILESYNC; 586 error = nfs_writerpc(vp, uio, cred, &iomode, &must_commit); 587 if (must_commit) 588 nfs_clearcommit(vp->v_mount); 589 return (error); 590 } 591 nfsstats.biocache_writes++; 592 lbn = uio->uio_offset / biosize; 593 on = uio->uio_offset & (biosize-1); 594 n = min((unsigned)(biosize - on), uio->uio_resid); 595 bn = lbn * (biosize / DEV_BSIZE); 596 again: 597 bp = nfs_getcacheblk(vp, bn, biosize, p); 598 if (!bp) 599 return (EINTR); 600 if (bp->b_wcred == NOCRED) { 601 crhold(cred); 602 bp->b_wcred = cred; 603 } 604 np->n_flag |= NMODIFIED; 605 if (uio->uio_offset + n > np->n_size) { 606 np->n_size = uio->uio_offset + n; 607 vnode_pager_setsize(vp, np->n_size); 608 } 609 610 /* 611 * If the new write will leave a contiguous dirty 612 * area, just update the b_dirtyoff and b_dirtyend, 613 * otherwise force a write rpc of the old dirty area. 614 */ 615 if (bp->b_dirtyend > 0 && 616 (on > bp->b_dirtyend || (on + n) < bp->b_dirtyoff)) { 617 bp->b_proc = p; 618 if (VOP_BWRITE(bp) == EINTR) 619 return (EINTR); 620 goto again; 621 } 622 623 /* 624 * Check for valid write lease and get one as required. 625 * In case getblk() and/or bwrite() delayed us. 626 */ 627 if ((nmp->nm_flag & NFSMNT_NQNFS) && 628 NQNFS_CKINVALID(vp, np, ND_WRITE)) { 629 do { 630 error = nqnfs_getlease(vp, ND_WRITE, cred, p); 631 } while (error == NQNFS_EXPIRED); 632 if (error) { 633 brelse(bp); 634 return (error); 635 } 636 if (np->n_lrev != np->n_brev || 637 (np->n_flag & NQNFSNONCACHE)) { 638 brelse(bp); 639 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 640 if (error) 641 return (error); 642 np->n_brev = np->n_lrev; 643 goto again; 644 } 645 } 646 error = uiomove((char *)bp->b_data + on, n, uio); 647 if (error) { 648 bp->b_flags |= B_ERROR; 649 brelse(bp); 650 return (error); 651 } 652 if (bp->b_dirtyend > 0) { 653 bp->b_dirtyoff = min(on, bp->b_dirtyoff); 654 bp->b_dirtyend = max((on + n), bp->b_dirtyend); 655 } else { 656 bp->b_dirtyoff = on; 657 bp->b_dirtyend = on + n; 658 } 659 if (bp->b_validend == 0 || bp->b_validend < bp->b_dirtyoff || 660 bp->b_validoff > bp->b_dirtyend) { 661 bp->b_validoff = bp->b_dirtyoff; 662 bp->b_validend = bp->b_dirtyend; 663 } else { 664 bp->b_validoff = min(bp->b_validoff, bp->b_dirtyoff); 665 bp->b_validend = max(bp->b_validend, bp->b_dirtyend); 666 } 667 668 /* 669 * Since this block is being modified, it must be written 670 * again and not just committed. 671 */ 672 bp->b_flags &= ~B_NEEDCOMMIT; 673 674 /* 675 * If the lease is non-cachable or IO_SYNC do bwrite(). 676 */ 677 if ((np->n_flag & NQNFSNONCACHE) || (ioflag & IO_SYNC)) { 678 bp->b_proc = p; 679 error = VOP_BWRITE(bp); 680 if (error) 681 return (error); 682 if (np->n_flag & NQNFSNONCACHE) { 683 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 684 if (error) 685 return (error); 686 } 687 } else if ((n + on) == biosize && 688 (nmp->nm_flag & NFSMNT_NQNFS) == 0) { 689 bp->b_proc = (struct proc *)0; 690 bp->b_flags |= B_ASYNC; 691 (void)nfs_writebp(bp, 0); 692 } else { 693 bdwrite(bp); 694 } 695 } while (uio->uio_resid > 0 && n > 0); 696 return (0); 697 } 698 699 /* 700 * Get an nfs cache block. 701 * Allocate a new one if the block isn't currently in the cache 702 * and return the block marked busy. If the calling process is 703 * interrupted by a signal for an interruptible mount point, return 704 * NULL. 705 */ 706 struct buf * 707 nfs_getcacheblk(vp, bn, size, p) 708 struct vnode *vp; 709 daddr_t bn; 710 int size; 711 struct proc *p; 712 { 713 register struct buf *bp; 714 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 715 716 if (nmp->nm_flag & NFSMNT_INT) { 717 bp = getblk(vp, bn, size, PCATCH, 0); 718 while (bp == (struct buf *)0) { 719 if (nfs_sigintr(nmp, (struct nfsreq *)0, p)) 720 return ((struct buf *)0); 721 bp = getblk(vp, bn, size, 0, 2 * hz); 722 } 723 } else 724 bp = getblk(vp, bn, size, 0, 0); 725 return (bp); 726 } 727 728 /* 729 * Flush and invalidate all dirty buffers. If another process is already 730 * doing the flush, just wait for completion. 731 */ 732 int 733 nfs_vinvalbuf(vp, flags, cred, p, intrflg) 734 struct vnode *vp; 735 int flags; 736 struct ucred *cred; 737 struct proc *p; 738 int intrflg; 739 { 740 register struct nfsnode *np = VTONFS(vp); 741 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 742 int error = 0, slpflag, slptimeo; 743 744 if ((nmp->nm_flag & NFSMNT_INT) == 0) 745 intrflg = 0; 746 if (intrflg) { 747 slpflag = PCATCH; 748 slptimeo = 2 * hz; 749 } else { 750 slpflag = 0; 751 slptimeo = 0; 752 } 753 /* 754 * First wait for any other process doing a flush to complete. 755 */ 756 while (np->n_flag & NFLUSHINPROG) { 757 np->n_flag |= NFLUSHWANT; 758 error = tsleep((caddr_t)&np->n_flag, PRIBIO + 2, "nfsvinval", 759 slptimeo); 760 if (error && intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) 761 return (EINTR); 762 } 763 764 /* 765 * Now, flush as required. 766 */ 767 np->n_flag |= NFLUSHINPROG; 768 error = vinvalbuf(vp, flags, cred, p, slpflag, 0); 769 while (error) { 770 if (intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) { 771 np->n_flag &= ~NFLUSHINPROG; 772 if (np->n_flag & NFLUSHWANT) { 773 np->n_flag &= ~NFLUSHWANT; 774 wakeup((caddr_t)&np->n_flag); 775 } 776 return (EINTR); 777 } 778 error = vinvalbuf(vp, flags, cred, p, 0, slptimeo); 779 } 780 np->n_flag &= ~(NMODIFIED | NFLUSHINPROG); 781 if (np->n_flag & NFLUSHWANT) { 782 np->n_flag &= ~NFLUSHWANT; 783 wakeup((caddr_t)&np->n_flag); 784 } 785 return (0); 786 } 787 788 /* 789 * Initiate asynchronous I/O. Return an error if no nfsiods are available. 790 * This is mainly to avoid queueing async I/O requests when the nfsiods 791 * are all hung on a dead server. 792 */ 793 int 794 nfs_asyncio(bp, cred) 795 register struct buf *bp; 796 struct ucred *cred; 797 { 798 register int i; 799 register struct nfsmount *nmp; 800 int gotiod, slpflag = 0, slptimeo = 0, error; 801 802 if (nfs_numasync == 0) 803 return (EIO); 804 805 806 nmp = VFSTONFS(bp->b_vp->v_mount); 807 again: 808 if (nmp->nm_flag & NFSMNT_INT) 809 slpflag = PCATCH; 810 gotiod = FALSE; 811 812 /* 813 * Find a free iod to process this request. 814 */ 815 816 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 817 if (nfs_iodwant[i]) { 818 /* 819 * Found one, so wake it up and tell it which 820 * mount to process. 821 */ 822 nfs_iodwant[i] = (struct proc *)0; 823 nfs_iodmount[i] = nmp; 824 nmp->nm_bufqiods++; 825 wakeup((caddr_t)&nfs_iodwant[i]); 826 gotiod = TRUE; 827 break; 828 } 829 /* 830 * If none are free, we may already have an iod working on this mount 831 * point. If so, it will process our request. 832 */ 833 if (!gotiod && nmp->nm_bufqiods > 0) 834 gotiod = TRUE; 835 836 /* 837 * If we have an iod which can process the request, then queue 838 * the buffer. 839 */ 840 if (gotiod) { 841 /* 842 * Ensure that the queue never grows too large. 843 */ 844 while (nmp->nm_bufqlen >= 2*nfs_numasync) { 845 nmp->nm_bufqwant = TRUE; 846 error = tsleep(&nmp->nm_bufq, slpflag | PRIBIO, 847 "nfsaio", slptimeo); 848 if (error) { 849 if (nfs_sigintr(nmp, NULL, bp->b_proc)) 850 return (EINTR); 851 if (slpflag == PCATCH) { 852 slpflag = 0; 853 slptimeo = 2 * hz; 854 } 855 } 856 /* 857 * We might have lost our iod while sleeping, 858 * so check and loop if nescessary. 859 */ 860 if (nmp->nm_bufqiods == 0) 861 goto again; 862 } 863 864 if (bp->b_flags & B_READ) { 865 if (bp->b_rcred == NOCRED && cred != NOCRED) { 866 crhold(cred); 867 bp->b_rcred = cred; 868 } 869 } else { 870 bp->b_flags |= B_WRITEINPROG; 871 if (bp->b_wcred == NOCRED && cred != NOCRED) { 872 crhold(cred); 873 bp->b_wcred = cred; 874 } 875 } 876 877 TAILQ_INSERT_TAIL(&nmp->nm_bufq, bp, b_freelist); 878 nmp->nm_bufqlen++; 879 return (0); 880 } 881 882 /* 883 * All the iods are busy on other mounts, so return EIO to 884 * force the caller to process the i/o synchronously. 885 */ 886 return (EIO); 887 } 888 889 /* 890 * Do an I/O operation to/from a cache block. This may be called 891 * synchronously or from an nfsiod. 892 */ 893 int 894 nfs_doio(bp, cr, p) 895 register struct buf *bp; 896 struct ucred *cr; 897 struct proc *p; 898 { 899 register struct uio *uiop; 900 register struct vnode *vp; 901 struct nfsnode *np; 902 struct nfsmount *nmp; 903 int error = 0, diff, len, iomode, must_commit = 0; 904 struct uio uio; 905 struct iovec io; 906 907 vp = bp->b_vp; 908 np = VTONFS(vp); 909 nmp = VFSTONFS(vp->v_mount); 910 uiop = &uio; 911 uiop->uio_iov = &io; 912 uiop->uio_iovcnt = 1; 913 uiop->uio_segflg = UIO_SYSSPACE; 914 uiop->uio_procp = p; 915 916 /* 917 * Historically, paging was done with physio, but no more... 918 */ 919 if (bp->b_flags & B_PHYS) { 920 /* 921 * ...though reading /dev/drum still gets us here. 922 */ 923 io.iov_len = uiop->uio_resid = bp->b_bcount; 924 /* mapping was done by vmapbuf() */ 925 io.iov_base = bp->b_data; 926 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; 927 if (bp->b_flags & B_READ) { 928 uiop->uio_rw = UIO_READ; 929 nfsstats.read_physios++; 930 error = nfs_readrpc(vp, uiop, cr); 931 } else { 932 iomode = NFSV3WRITE_DATASYNC; 933 uiop->uio_rw = UIO_WRITE; 934 nfsstats.write_physios++; 935 error = nfs_writerpc(vp, uiop, cr, &iomode, &must_commit); 936 } 937 if (error) { 938 bp->b_flags |= B_ERROR; 939 bp->b_error = error; 940 } 941 } else if (bp->b_flags & B_READ) { 942 io.iov_len = uiop->uio_resid = bp->b_bcount; 943 io.iov_base = bp->b_data; 944 uiop->uio_rw = UIO_READ; 945 switch (vp->v_type) { 946 case VREG: 947 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; 948 nfsstats.read_bios++; 949 error = nfs_readrpc(vp, uiop, cr); 950 if (!error) { 951 bp->b_validoff = 0; 952 if (uiop->uio_resid) { 953 /* 954 * If len > 0, there is a hole in the file and 955 * no writes after the hole have been pushed to 956 * the server yet. 957 * Just zero fill the rest of the valid area. 958 */ 959 diff = bp->b_bcount - uiop->uio_resid; 960 len = np->n_size - (((u_quad_t)bp->b_blkno) * DEV_BSIZE 961 + diff); 962 if (len > 0) { 963 len = min(len, uiop->uio_resid); 964 bzero((char *)bp->b_data + diff, len); 965 bp->b_validend = diff + len; 966 } else 967 bp->b_validend = diff; 968 } else 969 bp->b_validend = bp->b_bcount; 970 } 971 if (p && (vp->v_flag & VTEXT) && 972 (((nmp->nm_flag & NFSMNT_NQNFS) && 973 NQNFS_CKINVALID(vp, np, ND_READ) && 974 np->n_lrev != np->n_brev) || 975 (!(nmp->nm_flag & NFSMNT_NQNFS) && 976 np->n_mtime != np->n_vattr.va_mtime.tv_sec))) { 977 uprintf("Process killed due to text file modification\n"); 978 psignal(p, SIGKILL); 979 p->p_holdcnt++; 980 } 981 break; 982 case VLNK: 983 uiop->uio_offset = (off_t)0; 984 nfsstats.readlink_bios++; 985 error = nfs_readlinkrpc(vp, uiop, cr); 986 break; 987 case VDIR: 988 nfsstats.readdir_bios++; 989 uiop->uio_offset = bp->b_dcookie; 990 if (nmp->nm_flag & NFSMNT_RDIRPLUS) { 991 error = nfs_readdirplusrpc(vp, uiop, cr); 992 if (error == NFSERR_NOTSUPP) 993 nmp->nm_flag &= ~NFSMNT_RDIRPLUS; 994 } 995 if ((nmp->nm_flag & NFSMNT_RDIRPLUS) == 0) 996 error = nfs_readdirrpc(vp, uiop, cr); 997 if (!error) { 998 bp->b_dcookie = uiop->uio_offset; 999 bp->b_validoff = 0; 1000 bp->b_validend = bp->b_bcount - uiop->uio_resid; 1001 } 1002 break; 1003 default: 1004 printf("nfs_doio: type %x unexpected\n",vp->v_type); 1005 break; 1006 }; 1007 if (error) { 1008 bp->b_flags |= B_ERROR; 1009 bp->b_error = error; 1010 } 1011 } else { 1012 io.iov_len = uiop->uio_resid = bp->b_dirtyend 1013 - bp->b_dirtyoff; 1014 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE 1015 + bp->b_dirtyoff; 1016 io.iov_base = (char *)bp->b_data + bp->b_dirtyoff; 1017 uiop->uio_rw = UIO_WRITE; 1018 nfsstats.write_bios++; 1019 if ((bp->b_flags & (B_ASYNC | B_NEEDCOMMIT | B_NOCACHE)) == B_ASYNC) 1020 iomode = NFSV3WRITE_UNSTABLE; 1021 else 1022 iomode = NFSV3WRITE_FILESYNC; 1023 bp->b_flags |= B_WRITEINPROG; 1024 #ifdef fvdl_debug 1025 printf("nfs_doio(%x): bp %x doff %d dend %d\n", 1026 vp, bp, bp->b_dirtyoff, bp->b_dirtyend); 1027 #endif 1028 error = nfs_writerpc(vp, uiop, cr, &iomode, &must_commit); 1029 if (!error && iomode == NFSV3WRITE_UNSTABLE) 1030 bp->b_flags |= B_NEEDCOMMIT; 1031 else 1032 bp->b_flags &= ~B_NEEDCOMMIT; 1033 bp->b_flags &= ~B_WRITEINPROG; 1034 1035 /* 1036 * For an interrupted write, the buffer is still valid and the 1037 * write hasn't been pushed to the server yet, so we can't set 1038 * B_ERROR and report the interruption by setting B_EINTR. For 1039 * the B_ASYNC case, B_EINTR is not relevant, so the rpc attempt 1040 * is essentially a noop. 1041 * For the case of a V3 write rpc not being committed to stable 1042 * storage, the block is still dirty and requires either a commit 1043 * rpc or another write rpc with iomode == NFSV3WRITE_FILESYNC 1044 * before the block is reused. This is indicated by setting the 1045 * B_DELWRI and B_NEEDCOMMIT flags. 1046 */ 1047 if (error == EINTR || (!error && (bp->b_flags & B_NEEDCOMMIT))) { 1048 bp->b_flags |= B_DELWRI; 1049 1050 /* 1051 * Since for the B_ASYNC case, nfs_bwrite() has reassigned the 1052 * buffer to the clean list, we have to reassign it back to the 1053 * dirty one. Ugh. 1054 */ 1055 if (bp->b_flags & B_ASYNC) 1056 reassignbuf(bp, vp); 1057 else if (error) 1058 bp->b_flags |= B_EINTR; 1059 } else { 1060 if (error) { 1061 bp->b_flags |= B_ERROR; 1062 bp->b_error = np->n_error = error; 1063 np->n_flag |= NWRITEERR; 1064 } 1065 bp->b_dirtyoff = bp->b_dirtyend = 0; 1066 } 1067 } 1068 bp->b_resid = uiop->uio_resid; 1069 if (must_commit) 1070 nfs_clearcommit(vp->v_mount); 1071 biodone(bp); 1072 return (error); 1073 } 1074
Indexes created Mon Oct 13 08:09:50 GMT 2025