nfs_bio.c revision 1.33
1 /* $NetBSD: nfs_bio.c,v 1.33 1997/07/17 23:54:27 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 54 #include <vm/vm.h> 55 56 #include <nfs/rpcv2.h> 57 #include <nfs/nfsproto.h> 58 #include <nfs/nfs.h> 59 #include <nfs/nfsmount.h> 60 #include <nfs/nqnfs.h> 61 #include <nfs/nfsnode.h> 62 #include <nfs/nfs_var.h> 63 64 extern int nfs_numasync; 65 extern struct nfsstats nfsstats; 66 67 /* 68 * Vnode op for read using bio 69 * Any similarity to readip() is purely coincidental 70 */ 71 int 72 nfs_bioread(vp, uio, ioflag, cred) 73 register struct vnode *vp; 74 register struct uio *uio; 75 int ioflag; 76 struct ucred *cred; 77 { 78 register struct nfsnode *np = VTONFS(vp); 79 register int biosize, diff, i; 80 struct buf *bp = NULL, *rabp; 81 struct vattr vattr; 82 struct proc *p; 83 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 84 daddr_t lbn, bn, rabn; 85 caddr_t baddr; 86 int got_buf = 0, nra, error = 0, n = 0, on = 0, not_readin; 87 88 #ifdef DIAGNOSTIC 89 if (uio->uio_rw != UIO_READ) 90 panic("nfs_read mode"); 91 #endif 92 if (uio->uio_resid == 0) 93 return (0); 94 if (uio->uio_offset < 0) 95 return (EINVAL); 96 p = uio->uio_procp; 97 if ((nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_GOTFSINFO)) == NFSMNT_NFSV3) 98 (void)nfs_fsinfo(nmp, vp, cred, p); 99 if ((uio->uio_offset + uio->uio_resid) > nmp->nm_maxfilesize) 100 return (EFBIG); 101 biosize = nmp->nm_rsize; 102 /* 103 * For nfs, cache consistency can only be maintained approximately. 104 * Although RFC1094 does not specify the criteria, the following is 105 * believed to be compatible with the reference port. 106 * For nqnfs, full cache consistency is maintained within the loop. 107 * For nfs: 108 * If the file's modify time on the server has changed since the 109 * last read rpc or you have written to the file, 110 * you may have lost data cache consistency with the 111 * server, so flush all of the file's data out of the cache. 112 * Then force a getattr rpc to ensure that you have up to date 113 * attributes. 114 * NB: This implies that cache data can be read when up to 115 * NFS_ATTRTIMEO seconds out of date. If you find that you need current 116 * attributes this could be forced by setting n_attrstamp to 0 before 117 * the VOP_GETATTR() call. 118 */ 119 if ((nmp->nm_flag & NFSMNT_NQNFS) == 0 && vp->v_type != VLNK) { 120 if (np->n_flag & NMODIFIED) { 121 if (vp->v_type != VREG) { 122 if (vp->v_type != VDIR) 123 panic("nfs: bioread, not dir"); 124 nfs_invaldir(vp); 125 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 126 if (error) 127 return (error); 128 } 129 np->n_attrstamp = 0; 130 error = VOP_GETATTR(vp, &vattr, cred, p); 131 if (error) 132 return (error); 133 np->n_mtime = vattr.va_mtime.tv_sec; 134 } else { 135 error = VOP_GETATTR(vp, &vattr, cred, p); 136 if (error) 137 return (error); 138 if (np->n_mtime != vattr.va_mtime.tv_sec) { 139 if (vp->v_type == VDIR) 140 nfs_invaldir(vp); 141 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 142 if (error) 143 return (error); 144 np->n_mtime = vattr.va_mtime.tv_sec; 145 } 146 } 147 } 148 do { 149 150 /* 151 * Get a valid lease. If cached data is stale, flush it. 152 */ 153 if (nmp->nm_flag & NFSMNT_NQNFS) { 154 if (NQNFS_CKINVALID(vp, np, ND_READ)) { 155 do { 156 error = nqnfs_getlease(vp, ND_READ, cred, p); 157 } while (error == NQNFS_EXPIRED); 158 if (error) 159 return (error); 160 if (np->n_lrev != np->n_brev || 161 (np->n_flag & NQNFSNONCACHE) || 162 ((np->n_flag & NMODIFIED) && vp->v_type == VDIR)) { 163 if (vp->v_type == VDIR) 164 nfs_invaldir(vp); 165 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 166 if (error) 167 return (error); 168 np->n_brev = np->n_lrev; 169 } 170 } else if (vp->v_type == VDIR && (np->n_flag & NMODIFIED)) { 171 nfs_invaldir(vp); 172 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 173 if (error) 174 return (error); 175 } 176 } 177 /* 178 * Don't cache symlinks. 179 */ 180 if (np->n_flag & NQNFSNONCACHE 181 || ((vp->v_flag & VROOT) && vp->v_type == VLNK)) { 182 switch (vp->v_type) { 183 case VREG: 184 return (nfs_readrpc(vp, uio, cred)); 185 case VLNK: 186 return (nfs_readlinkrpc(vp, uio, cred)); 187 case VDIR: 188 break; 189 default: 190 printf(" NQNFSNONCACHE: type %x unexpected\n", 191 vp->v_type); 192 }; 193 } 194 baddr = (caddr_t)0; 195 switch (vp->v_type) { 196 case VREG: 197 nfsstats.biocache_reads++; 198 lbn = uio->uio_offset / biosize; 199 on = uio->uio_offset & (biosize - 1); 200 bn = lbn * (biosize / DEV_BSIZE); 201 not_readin = 1; 202 203 /* 204 * Start the read ahead(s), as required. 205 */ 206 if (nfs_numasync > 0 && nmp->nm_readahead > 0) { 207 for (nra = 0; nra < nmp->nm_readahead && 208 (lbn + 1 + nra) * biosize < np->n_size; nra++) { 209 rabn = (lbn + 1 + nra) * (biosize / DEV_BSIZE); 210 if (!incore(vp, rabn)) { 211 rabp = nfs_getcacheblk(vp, rabn, biosize, p); 212 if (!rabp) 213 return (EINTR); 214 if ((rabp->b_flags & (B_DELWRI | B_DONE)) == 0) { 215 rabp->b_flags |= (B_READ | B_ASYNC); 216 if (nfs_asyncio(rabp, cred)) { 217 rabp->b_flags |= B_INVAL; 218 brelse(rabp); 219 } 220 } else 221 brelse(rabp); 222 } 223 } 224 } 225 226 /* 227 * If the block is in the cache and has the required data 228 * in a valid region, just copy it out. 229 * Otherwise, get the block and write back/read in, 230 * as required. 231 */ 232 if ((bp = incore(vp, bn)) && 233 (bp->b_flags & (B_BUSY | B_WRITEINPROG)) == 234 (B_BUSY | B_WRITEINPROG)) 235 got_buf = 0; 236 else { 237 again: 238 bp = nfs_getcacheblk(vp, bn, biosize, p); 239 if (!bp) 240 return (EINTR); 241 got_buf = 1; 242 if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0) { 243 bp->b_flags |= B_READ; 244 not_readin = 0; 245 error = nfs_doio(bp, cred, p); 246 if (error) { 247 brelse(bp); 248 return (error); 249 } 250 } 251 } 252 n = min((unsigned)(biosize - on), uio->uio_resid); 253 diff = np->n_size - uio->uio_offset; 254 if (diff < n) 255 n = diff; 256 if (not_readin && n > 0) { 257 if (on < bp->b_validoff || (on + n) > bp->b_validend) { 258 if (!got_buf) { 259 bp = nfs_getcacheblk(vp, bn, biosize, p); 260 if (!bp) 261 return (EINTR); 262 got_buf = 1; 263 } 264 bp->b_flags |= B_INVAFTERWRITE; 265 if (bp->b_dirtyend > 0) { 266 if ((bp->b_flags & B_DELWRI) == 0) 267 panic("nfsbioread"); 268 if (VOP_BWRITE(bp) == EINTR) 269 return (EINTR); 270 } else 271 brelse(bp); 272 goto again; 273 } 274 } 275 vp->v_lastr = lbn; 276 diff = (on >= bp->b_validend) ? 0 : (bp->b_validend - on); 277 if (diff < n) 278 n = diff; 279 break; 280 case VLNK: 281 nfsstats.biocache_readlinks++; 282 bp = nfs_getcacheblk(vp, (daddr_t)0, NFS_MAXPATHLEN, p); 283 if (!bp) 284 return (EINTR); 285 if ((bp->b_flags & B_DONE) == 0) { 286 bp->b_flags |= B_READ; 287 error = nfs_doio(bp, cred, p); 288 if (error) { 289 brelse(bp); 290 return (error); 291 } 292 } 293 n = min(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid); 294 got_buf = 1; 295 on = 0; 296 break; 297 case VDIR: 298 if (uio->uio_resid < NFS_READDIRBLKSIZ) 299 return (0); 300 nfsstats.biocache_readdirs++; 301 lbn = uio->uio_offset / NFS_DIRBLKSIZ; 302 on = uio->uio_offset & (NFS_DIRBLKSIZ - 1); 303 bp = nfs_getcacheblk(vp, lbn, NFS_DIRBLKSIZ, p); 304 if (!bp) 305 return (EINTR); 306 if ((bp->b_flags & B_DONE) == 0) { 307 bp->b_flags |= B_READ; 308 error = nfs_doio(bp, cred, p); 309 if (error) { 310 brelse(bp); 311 while (error == NFSERR_BAD_COOKIE) { 312 nfs_invaldir(vp); 313 error = nfs_vinvalbuf(vp, 0, cred, p, 1); 314 /* 315 * Yuck! The directory has been modified on the 316 * server. The only way to get the block is by 317 * reading from the beginning to get all the 318 * offset cookies. 319 */ 320 for (i = 0; i <= lbn && !error; i++) { 321 bp = nfs_getcacheblk(vp, i, NFS_DIRBLKSIZ, p); 322 if (!bp) 323 return (EINTR); 324 if ((bp->b_flags & B_DONE) == 0) { 325 bp->b_flags |= B_READ; 326 error = nfs_doio(bp, cred, p); 327 if (error) 328 brelse(bp); 329 } 330 } 331 } 332 if (error) 333 return (error); 334 } 335 } 336 337 /* 338 * If not eof and read aheads are enabled, start one. 339 * (You need the current block first, so that you have the 340 * directory offset cookie of the next block.) 341 */ 342 if (nfs_numasync > 0 && nmp->nm_readahead > 0 && 343 (np->n_direofoffset == 0 || 344 (lbn + 1) * NFS_DIRBLKSIZ < np->n_direofoffset) && 345 !(np->n_flag & NQNFSNONCACHE) && 346 !incore(vp, lbn + 1)) { 347 rabp = nfs_getcacheblk(vp, lbn + 1, NFS_DIRBLKSIZ, p); 348 if (rabp) { 349 if ((rabp->b_flags & (B_DONE | B_DELWRI)) == 0) { 350 rabp->b_flags |= (B_READ | B_ASYNC); 351 if (nfs_asyncio(rabp, cred)) { 352 rabp->b_flags |= B_INVAL; 353 brelse(rabp); 354 } 355 } else 356 brelse(rabp); 357 } 358 } 359 n = min(uio->uio_resid, NFS_DIRBLKSIZ - bp->b_resid - on); 360 got_buf = 1; 361 break; 362 default: 363 printf(" nfsbioread: type %x unexpected\n",vp->v_type); 364 break; 365 }; 366 367 if (n > 0) { 368 if (!baddr) 369 baddr = bp->b_data; 370 error = uiomove(baddr + on, (int)n, uio); 371 } 372 switch (vp->v_type) { 373 case VREG: 374 break; 375 case VLNK: 376 n = 0; 377 break; 378 case VDIR: 379 if (np->n_flag & NQNFSNONCACHE) 380 bp->b_flags |= B_INVAL; 381 break; 382 default: 383 printf(" nfsbioread: type %x unexpected\n",vp->v_type); 384 } 385 if (got_buf) 386 brelse(bp); 387 } while (error == 0 && uio->uio_resid > 0 && n > 0); 388 return (error); 389 } 390 391 /* 392 * Vnode op for write using bio 393 */ 394 int 395 nfs_write(v) 396 void *v; 397 { 398 struct vop_write_args /* { 399 struct vnode *a_vp; 400 struct uio *a_uio; 401 int a_ioflag; 402 struct ucred *a_cred; 403 } */ *ap = v; 404 register int biosize; 405 register struct uio *uio = ap->a_uio; 406 struct proc *p = uio->uio_procp; 407 register struct vnode *vp = ap->a_vp; 408 struct nfsnode *np = VTONFS(vp); 409 register struct ucred *cred = ap->a_cred; 410 int ioflag = ap->a_ioflag; 411 struct buf *bp; 412 struct vattr vattr; 413 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 414 daddr_t lbn, bn; 415 int n, on, error = 0, iomode, must_commit; 416 417 #ifdef DIAGNOSTIC 418 if (uio->uio_rw != UIO_WRITE) 419 panic("nfs_write mode"); 420 if (uio->uio_segflg == UIO_USERSPACE && uio->uio_procp != curproc) 421 panic("nfs_write proc"); 422 #endif 423 if (vp->v_type != VREG) 424 return (EIO); 425 if (np->n_flag & NWRITEERR) { 426 np->n_flag &= ~NWRITEERR; 427 return (np->n_error); 428 } 429 if ((nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_GOTFSINFO)) == NFSMNT_NFSV3) 430 (void)nfs_fsinfo(nmp, vp, cred, p); 431 if (ioflag & (IO_APPEND | IO_SYNC)) { 432 if (np->n_flag & NMODIFIED) { 433 np->n_attrstamp = 0; 434 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 435 if (error) 436 return (error); 437 } 438 if (ioflag & IO_APPEND) { 439 np->n_attrstamp = 0; 440 error = VOP_GETATTR(vp, &vattr, cred, p); 441 if (error) 442 return (error); 443 uio->uio_offset = np->n_size; 444 } 445 } 446 if (uio->uio_offset < 0) 447 return (EINVAL); 448 if ((uio->uio_offset + uio->uio_resid) > nmp->nm_maxfilesize) 449 return (EFBIG); 450 if (uio->uio_resid == 0) 451 return (0); 452 /* 453 * Maybe this should be above the vnode op call, but so long as 454 * file servers have no limits, i don't think it matters 455 */ 456 if (p && uio->uio_offset + uio->uio_resid > 457 p->p_rlimit[RLIMIT_FSIZE].rlim_cur) { 458 psignal(p, SIGXFSZ); 459 return (EFBIG); 460 } 461 /* 462 * I use nm_rsize, not nm_wsize so that all buffer cache blocks 463 * will be the same size within a filesystem. nfs_writerpc will 464 * still use nm_wsize when sizing the rpc's. 465 */ 466 biosize = nmp->nm_rsize; 467 do { 468 469 /* 470 * XXX make sure we aren't cached in the VM page cache 471 */ 472 (void)vnode_pager_uncache(vp); 473 474 /* 475 * Check for a valid write lease. 476 */ 477 if ((nmp->nm_flag & NFSMNT_NQNFS) && 478 NQNFS_CKINVALID(vp, np, ND_WRITE)) { 479 do { 480 error = nqnfs_getlease(vp, ND_WRITE, cred, p); 481 } while (error == NQNFS_EXPIRED); 482 if (error) 483 return (error); 484 if (np->n_lrev != np->n_brev || 485 (np->n_flag & NQNFSNONCACHE)) { 486 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 487 if (error) 488 return (error); 489 np->n_brev = np->n_lrev; 490 } 491 } 492 if ((np->n_flag & NQNFSNONCACHE) && uio->uio_iovcnt == 1) { 493 iomode = NFSV3WRITE_FILESYNC; 494 error = nfs_writerpc(vp, uio, cred, &iomode, &must_commit); 495 if (must_commit) 496 nfs_clearcommit(vp->v_mount); 497 return (error); 498 } 499 nfsstats.biocache_writes++; 500 lbn = uio->uio_offset / biosize; 501 on = uio->uio_offset & (biosize-1); 502 n = min((unsigned)(biosize - on), uio->uio_resid); 503 bn = lbn * (biosize / DEV_BSIZE); 504 again: 505 bp = nfs_getcacheblk(vp, bn, biosize, p); 506 if (!bp) 507 return (EINTR); 508 if (bp->b_wcred == NOCRED) { 509 crhold(cred); 510 bp->b_wcred = cred; 511 } 512 np->n_flag |= NMODIFIED; 513 if (uio->uio_offset + n > np->n_size) { 514 np->n_size = uio->uio_offset + n; 515 vnode_pager_setsize(vp, np->n_size); 516 } 517 518 /* 519 * If the new write will leave a contiguous dirty 520 * area, just update the b_dirtyoff and b_dirtyend, 521 * otherwise force a write rpc of the old dirty area. 522 */ 523 if (bp->b_dirtyend > 0 && 524 (on > bp->b_dirtyend || (on + n) < bp->b_dirtyoff)) { 525 bp->b_proc = p; 526 if (VOP_BWRITE(bp) == EINTR) 527 return (EINTR); 528 goto again; 529 } 530 531 /* 532 * Check for valid write lease and get one as required. 533 * In case getblk() and/or bwrite() delayed us. 534 */ 535 if ((nmp->nm_flag & NFSMNT_NQNFS) && 536 NQNFS_CKINVALID(vp, np, ND_WRITE)) { 537 do { 538 error = nqnfs_getlease(vp, ND_WRITE, cred, p); 539 } while (error == NQNFS_EXPIRED); 540 if (error) { 541 brelse(bp); 542 return (error); 543 } 544 if (np->n_lrev != np->n_brev || 545 (np->n_flag & NQNFSNONCACHE)) { 546 brelse(bp); 547 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 548 if (error) 549 return (error); 550 np->n_brev = np->n_lrev; 551 goto again; 552 } 553 } 554 error = uiomove((char *)bp->b_data + on, n, uio); 555 if (error) { 556 bp->b_flags |= B_ERROR; 557 brelse(bp); 558 return (error); 559 } 560 if (bp->b_dirtyend > 0) { 561 bp->b_dirtyoff = min(on, bp->b_dirtyoff); 562 bp->b_dirtyend = max((on + n), bp->b_dirtyend); 563 } else { 564 bp->b_dirtyoff = on; 565 bp->b_dirtyend = on + n; 566 } 567 if (bp->b_validend == 0 || bp->b_validend < bp->b_dirtyoff || 568 bp->b_validoff > bp->b_dirtyend) { 569 bp->b_validoff = bp->b_dirtyoff; 570 bp->b_validend = bp->b_dirtyend; 571 } else { 572 bp->b_validoff = min(bp->b_validoff, bp->b_dirtyoff); 573 bp->b_validend = max(bp->b_validend, bp->b_dirtyend); 574 } 575 576 /* 577 * Since this block is being modified, it must be written 578 * again and not just committed. 579 */ 580 bp->b_flags &= ~B_NEEDCOMMIT; 581 582 /* 583 * If the lease is non-cachable or IO_SYNC do bwrite(). 584 */ 585 if ((np->n_flag & NQNFSNONCACHE) || (ioflag & IO_SYNC)) { 586 bp->b_proc = p; 587 error = VOP_BWRITE(bp); 588 if (error) 589 return (error); 590 if (np->n_flag & NQNFSNONCACHE) { 591 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 592 if (error) 593 return (error); 594 } 595 } else if ((n + on) == biosize && 596 (nmp->nm_flag & NFSMNT_NQNFS) == 0) { 597 bp->b_proc = (struct proc *)0; 598 bp->b_flags |= B_ASYNC; 599 (void)nfs_writebp(bp, 0); 600 } else { 601 bdwrite(bp); 602 } 603 } while (uio->uio_resid > 0 && n > 0); 604 return (0); 605 } 606 607 /* 608 * Get an nfs cache block. 609 * Allocate a new one if the block isn't currently in the cache 610 * and return the block marked busy. If the calling process is 611 * interrupted by a signal for an interruptible mount point, return 612 * NULL. 613 */ 614 struct buf * 615 nfs_getcacheblk(vp, bn, size, p) 616 struct vnode *vp; 617 daddr_t bn; 618 int size; 619 struct proc *p; 620 { 621 register struct buf *bp; 622 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 623 624 if (nmp->nm_flag & NFSMNT_INT) { 625 bp = getblk(vp, bn, size, PCATCH, 0); 626 while (bp == (struct buf *)0) { 627 if (nfs_sigintr(nmp, (struct nfsreq *)0, p)) 628 return ((struct buf *)0); 629 bp = getblk(vp, bn, size, 0, 2 * hz); 630 } 631 } else 632 bp = getblk(vp, bn, size, 0, 0); 633 return (bp); 634 } 635 636 /* 637 * Flush and invalidate all dirty buffers. If another process is already 638 * doing the flush, just wait for completion. 639 */ 640 int 641 nfs_vinvalbuf(vp, flags, cred, p, intrflg) 642 struct vnode *vp; 643 int flags; 644 struct ucred *cred; 645 struct proc *p; 646 int intrflg; 647 { 648 register struct nfsnode *np = VTONFS(vp); 649 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 650 int error = 0, slpflag, slptimeo; 651 652 if ((nmp->nm_flag & NFSMNT_INT) == 0) 653 intrflg = 0; 654 if (intrflg) { 655 slpflag = PCATCH; 656 slptimeo = 2 * hz; 657 } else { 658 slpflag = 0; 659 slptimeo = 0; 660 } 661 /* 662 * First wait for any other process doing a flush to complete. 663 */ 664 while (np->n_flag & NFLUSHINPROG) { 665 np->n_flag |= NFLUSHWANT; 666 error = tsleep((caddr_t)&np->n_flag, PRIBIO + 2, "nfsvinval", 667 slptimeo); 668 if (error && intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) 669 return (EINTR); 670 } 671 672 /* 673 * Now, flush as required. 674 */ 675 np->n_flag |= NFLUSHINPROG; 676 error = vinvalbuf(vp, flags, cred, p, slpflag, 0); 677 while (error) { 678 if (intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) { 679 np->n_flag &= ~NFLUSHINPROG; 680 if (np->n_flag & NFLUSHWANT) { 681 np->n_flag &= ~NFLUSHWANT; 682 wakeup((caddr_t)&np->n_flag); 683 } 684 return (EINTR); 685 } 686 error = vinvalbuf(vp, flags, cred, p, 0, slptimeo); 687 } 688 np->n_flag &= ~(NMODIFIED | NFLUSHINPROG); 689 if (np->n_flag & NFLUSHWANT) { 690 np->n_flag &= ~NFLUSHWANT; 691 wakeup((caddr_t)&np->n_flag); 692 } 693 return (0); 694 } 695 696 /* 697 * Initiate asynchronous I/O. Return an error if no nfsiods are available. 698 * This is mainly to avoid queueing async I/O requests when the nfsiods 699 * are all hung on a dead server. 700 */ 701 int 702 nfs_asyncio(bp, cred) 703 register struct buf *bp; 704 struct ucred *cred; 705 { 706 register int i; 707 register struct nfsmount *nmp; 708 int gotiod, slpflag = 0, slptimeo = 0, error; 709 710 if (nfs_numasync == 0) 711 return (EIO); 712 713 714 nmp = VFSTONFS(bp->b_vp->v_mount); 715 again: 716 if (nmp->nm_flag & NFSMNT_INT) 717 slpflag = PCATCH; 718 gotiod = FALSE; 719 720 /* 721 * Find a free iod to process this request. 722 */ 723 724 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 725 if (nfs_iodwant[i]) { 726 /* 727 * Found one, so wake it up and tell it which 728 * mount to process. 729 */ 730 nfs_iodwant[i] = (struct proc *)0; 731 nfs_iodmount[i] = nmp; 732 nmp->nm_bufqiods++; 733 wakeup((caddr_t)&nfs_iodwant[i]); 734 gotiod = TRUE; 735 break; 736 } 737 /* 738 * If none are free, we may already have an iod working on this mount 739 * point. If so, it will process our request. 740 */ 741 if (!gotiod && nmp->nm_bufqiods > 0) 742 gotiod = TRUE; 743 744 /* 745 * If we have an iod which can process the request, then queue 746 * the buffer. 747 */ 748 if (gotiod) { 749 /* 750 * Ensure that the queue never grows too large. 751 */ 752 while (nmp->nm_bufqlen >= 2*nfs_numasync) { 753 nmp->nm_bufqwant = TRUE; 754 error = tsleep(&nmp->nm_bufq, slpflag | PRIBIO, 755 "nfsaio", slptimeo); 756 if (error) { 757 if (nfs_sigintr(nmp, NULL, bp->b_proc)) 758 return (EINTR); 759 if (slpflag == PCATCH) { 760 slpflag = 0; 761 slptimeo = 2 * hz; 762 } 763 } 764 /* 765 * We might have lost our iod while sleeping, 766 * so check and loop if nescessary. 767 */ 768 if (nmp->nm_bufqiods == 0) 769 goto again; 770 } 771 772 if (bp->b_flags & B_READ) { 773 if (bp->b_rcred == NOCRED && cred != NOCRED) { 774 crhold(cred); 775 bp->b_rcred = cred; 776 } 777 } else { 778 bp->b_flags |= B_WRITEINPROG; 779 if (bp->b_wcred == NOCRED && cred != NOCRED) { 780 crhold(cred); 781 bp->b_wcred = cred; 782 } 783 } 784 785 TAILQ_INSERT_TAIL(&nmp->nm_bufq, bp, b_freelist); 786 nmp->nm_bufqlen++; 787 return (0); 788 } 789 790 /* 791 * All the iods are busy on other mounts, so return EIO to 792 * force the caller to process the i/o synchronously. 793 */ 794 return (EIO); 795 } 796 797 /* 798 * Do an I/O operation to/from a cache block. This may be called 799 * synchronously or from an nfsiod. 800 */ 801 int 802 nfs_doio(bp, cr, p) 803 register struct buf *bp; 804 struct ucred *cr; 805 struct proc *p; 806 { 807 register struct uio *uiop; 808 register struct vnode *vp; 809 struct nfsnode *np; 810 struct nfsmount *nmp; 811 int error = 0, diff, len, iomode, must_commit = 0; 812 struct uio uio; 813 struct iovec io; 814 815 vp = bp->b_vp; 816 np = VTONFS(vp); 817 nmp = VFSTONFS(vp->v_mount); 818 uiop = &uio; 819 uiop->uio_iov = &io; 820 uiop->uio_iovcnt = 1; 821 uiop->uio_segflg = UIO_SYSSPACE; 822 uiop->uio_procp = p; 823 824 /* 825 * Historically, paging was done with physio, but no more... 826 */ 827 if (bp->b_flags & B_PHYS) { 828 /* 829 * ...though reading /dev/drum still gets us here. 830 */ 831 io.iov_len = uiop->uio_resid = bp->b_bcount; 832 /* mapping was done by vmapbuf() */ 833 io.iov_base = bp->b_data; 834 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; 835 if (bp->b_flags & B_READ) { 836 uiop->uio_rw = UIO_READ; 837 nfsstats.read_physios++; 838 error = nfs_readrpc(vp, uiop, cr); 839 } else { 840 iomode = NFSV3WRITE_DATASYNC; 841 uiop->uio_rw = UIO_WRITE; 842 nfsstats.write_physios++; 843 error = nfs_writerpc(vp, uiop, cr, &iomode, &must_commit); 844 } 845 if (error) { 846 bp->b_flags |= B_ERROR; 847 bp->b_error = error; 848 } 849 } else if (bp->b_flags & B_READ) { 850 io.iov_len = uiop->uio_resid = bp->b_bcount; 851 io.iov_base = bp->b_data; 852 uiop->uio_rw = UIO_READ; 853 switch (vp->v_type) { 854 case VREG: 855 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; 856 nfsstats.read_bios++; 857 error = nfs_readrpc(vp, uiop, cr); 858 if (!error) { 859 bp->b_validoff = 0; 860 if (uiop->uio_resid) { 861 /* 862 * If len > 0, there is a hole in the file and 863 * no writes after the hole have been pushed to 864 * the server yet. 865 * Just zero fill the rest of the valid area. 866 */ 867 diff = bp->b_bcount - uiop->uio_resid; 868 len = np->n_size - (((u_quad_t)bp->b_blkno) * DEV_BSIZE 869 + diff); 870 if (len > 0) { 871 len = min(len, uiop->uio_resid); 872 bzero((char *)bp->b_data + diff, len); 873 bp->b_validend = diff + len; 874 } else 875 bp->b_validend = diff; 876 } else 877 bp->b_validend = bp->b_bcount; 878 } 879 if (p && (vp->v_flag & VTEXT) && 880 (((nmp->nm_flag & NFSMNT_NQNFS) && 881 NQNFS_CKINVALID(vp, np, ND_READ) && 882 np->n_lrev != np->n_brev) || 883 (!(nmp->nm_flag & NFSMNT_NQNFS) && 884 np->n_mtime != np->n_vattr.va_mtime.tv_sec))) { 885 uprintf("Process killed due to text file modification\n"); 886 psignal(p, SIGKILL); 887 p->p_holdcnt++; 888 } 889 break; 890 case VLNK: 891 uiop->uio_offset = (off_t)0; 892 nfsstats.readlink_bios++; 893 error = nfs_readlinkrpc(vp, uiop, cr); 894 break; 895 case VDIR: 896 nfsstats.readdir_bios++; 897 uiop->uio_offset = ((u_quad_t)bp->b_lblkno) * NFS_DIRBLKSIZ; 898 if (nmp->nm_flag & NFSMNT_RDIRPLUS) { 899 error = nfs_readdirplusrpc(vp, uiop, cr); 900 if (error == NFSERR_NOTSUPP) 901 nmp->nm_flag &= ~NFSMNT_RDIRPLUS; 902 } 903 if ((nmp->nm_flag & NFSMNT_RDIRPLUS) == 0) 904 error = nfs_readdirrpc(vp, uiop, cr); 905 break; 906 default: 907 printf("nfs_doio: type %x unexpected\n",vp->v_type); 908 break; 909 }; 910 if (error) { 911 bp->b_flags |= B_ERROR; 912 bp->b_error = error; 913 } 914 } else { 915 io.iov_len = uiop->uio_resid = bp->b_dirtyend 916 - bp->b_dirtyoff; 917 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE 918 + bp->b_dirtyoff; 919 io.iov_base = (char *)bp->b_data + bp->b_dirtyoff; 920 uiop->uio_rw = UIO_WRITE; 921 nfsstats.write_bios++; 922 if ((bp->b_flags & (B_ASYNC | B_NEEDCOMMIT | B_NOCACHE)) == B_ASYNC) 923 iomode = NFSV3WRITE_UNSTABLE; 924 else 925 iomode = NFSV3WRITE_FILESYNC; 926 bp->b_flags |= B_WRITEINPROG; 927 #ifdef fvdl_debug 928 printf("nfs_doio(%x): bp %x doff %d dend %d\n", 929 vp, bp, bp->b_dirtyoff, bp->b_dirtyend); 930 #endif 931 error = nfs_writerpc(vp, uiop, cr, &iomode, &must_commit); 932 if (!error && iomode == NFSV3WRITE_UNSTABLE) 933 bp->b_flags |= B_NEEDCOMMIT; 934 else 935 bp->b_flags &= ~B_NEEDCOMMIT; 936 bp->b_flags &= ~B_WRITEINPROG; 937 938 /* 939 * For an interrupted write, the buffer is still valid and the 940 * write hasn't been pushed to the server yet, so we can't set 941 * B_ERROR and report the interruption by setting B_EINTR. For 942 * the B_ASYNC case, B_EINTR is not relevant, so the rpc attempt 943 * is essentially a noop. 944 * For the case of a V3 write rpc not being committed to stable 945 * storage, the block is still dirty and requires either a commit 946 * rpc or another write rpc with iomode == NFSV3WRITE_FILESYNC 947 * before the block is reused. This is indicated by setting the 948 * B_DELWRI and B_NEEDCOMMIT flags. 949 */ 950 if (error == EINTR || (!error && (bp->b_flags & B_NEEDCOMMIT))) { 951 bp->b_flags |= B_DELWRI; 952 953 /* 954 * Since for the B_ASYNC case, nfs_bwrite() has reassigned the 955 * buffer to the clean list, we have to reassign it back to the 956 * dirty one. Ugh. 957 */ 958 if (bp->b_flags & B_ASYNC) 959 reassignbuf(bp, vp); 960 else if (error) 961 bp->b_flags |= B_EINTR; 962 } else { 963 if (error) { 964 bp->b_flags |= B_ERROR; 965 bp->b_error = np->n_error = error; 966 np->n_flag |= NWRITEERR; 967 } 968 bp->b_dirtyoff = bp->b_dirtyend = 0; 969 } 970 } 971 bp->b_resid = uiop->uio_resid; 972 if (must_commit) 973 nfs_clearcommit(vp->v_mount); 974 biodone(bp); 975 return (error); 976 } 977
Indexes created Tue Oct 14 11:09:46 GMT 2025