nfs_bio.c revision 1.23
1 /* $NetBSD: nfs_bio.c,v 1.23 1996/02/09 21:48:19 christos 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 8.5 (Berkeley) 1/4/94 39 */ 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/resourcevar.h> 44 #include <sys/proc.h> 45 #include <sys/buf.h> 46 #include <sys/vnode.h> 47 #include <sys/trace.h> 48 #include <sys/mount.h> 49 #include <sys/kernel.h> 50 #include <sys/namei.h> 51 #include <sys/signalvar.h> 52 53 #include <vm/vm.h> 54 55 #include <nfs/nfsnode.h> 56 #include <nfs/rpcv2.h> 57 #include <nfs/nfsv2.h> 58 #include <nfs/nfs.h> 59 #include <nfs/nfsmount.h> 60 #include <nfs/nqnfs.h> 61 #include <nfs/nfs_var.h> 62 63 extern struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON]; 64 extern int nfs_numasync; 65 66 /* 67 * Vnode op for read using bio 68 * Any similarity to readip() is purely coincidental 69 */ 70 int 71 nfs_bioread(vp, uio, ioflag, cred) 72 register struct vnode *vp; 73 register struct uio *uio; 74 int ioflag; 75 struct ucred *cred; 76 { 77 register struct nfsnode *np = VTONFS(vp); 78 register int biosize, diff; 79 struct buf *bp = NULL, *rabp; 80 struct vattr vattr; 81 struct proc *p; 82 struct nfsmount *nmp; 83 daddr_t lbn, bn, rabn; 84 caddr_t baddr; 85 int got_buf = 0, nra, error = 0, n = 0, on = 0, not_readin; 86 87 #ifdef lint 88 ioflag = ioflag; 89 #endif /* lint */ 90 #ifdef DIAGNOSTIC 91 if (uio->uio_rw != UIO_READ) 92 panic("nfs_read mode"); 93 #endif 94 if (uio->uio_resid == 0) 95 return (0); 96 if (uio->uio_offset < 0 && vp->v_type != VDIR) 97 return (EINVAL); 98 nmp = VFSTONFS(vp->v_mount); 99 biosize = nmp->nm_rsize; 100 p = uio->uio_procp; 101 /* 102 * For nfs, cache consistency can only be maintained approximately. 103 * Although RFC1094 does not specify the criteria, the following is 104 * believed to be compatible with the reference port. 105 * For nqnfs, full cache consistency is maintained within the loop. 106 * For nfs: 107 * If the file's modify time on the server has changed since the 108 * last read rpc or you have written to the file, 109 * you may have lost data cache consistency with the 110 * server, so flush all of the file's data out of the cache. 111 * Then force a getattr rpc to ensure that you have up to date 112 * attributes. 113 * The mount flag NFSMNT_MYWRITE says "Assume that my writes are 114 * the ones changing the modify time. 115 * NB: This implies that cache data can be read when up to 116 * NFS_ATTRTIMEO seconds out of date. If you find that you need current 117 * attributes this could be forced by setting n_attrstamp to 0 before 118 * the VOP_GETATTR() call. 119 */ 120 if ((nmp->nm_flag & NFSMNT_NQNFS) == 0 && vp->v_type != VLNK) { 121 if (np->n_flag & NMODIFIED) { 122 if ((nmp->nm_flag & NFSMNT_MYWRITE) == 0 || 123 vp->v_type != VREG) { 124 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 125 if (error) 126 return (error); 127 } 128 np->n_attrstamp = 0; 129 np->n_direofoffset = 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 if ((error = VOP_GETATTR(vp, &vattr, cred, p)) != 0) 136 return (error); 137 if (np->n_mtime != vattr.va_mtime.tv_sec) { 138 np->n_direofoffset = 0; 139 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 140 if (error) 141 return (error); 142 np->n_mtime = vattr.va_mtime.tv_sec; 143 } 144 } 145 } 146 do { 147 148 /* 149 * Get a valid lease. If cached data is stale, flush it. 150 */ 151 if (nmp->nm_flag & NFSMNT_NQNFS) { 152 if (NQNFS_CKINVALID(vp, np, NQL_READ)) { 153 do { 154 error = nqnfs_getlease(vp, NQL_READ, cred, p); 155 } while (error == NQNFS_EXPIRED); 156 if (error) 157 return (error); 158 if (np->n_lrev != np->n_brev || 159 (np->n_flag & NQNFSNONCACHE) || 160 ((np->n_flag & NMODIFIED) && vp->v_type == VDIR)) { 161 if (vp->v_type == VDIR) { 162 np->n_direofoffset = 0; 163 cache_purge(vp); 164 } 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 np->n_direofoffset = 0; 172 cache_purge(vp); 173 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 174 if (error) 175 return (error); 176 } 177 } 178 if (np->n_flag & NQNFSNONCACHE) { 179 switch (vp->v_type) { 180 case VREG: 181 error = nfs_readrpc(vp, uio, cred); 182 break; 183 case VLNK: 184 error = nfs_readlinkrpc(vp, uio, cred); 185 break; 186 case VDIR: 187 error = nfs_readdirrpc(vp, uio, cred); 188 break; 189 case VCHR: 190 case VSOCK: 191 case VFIFO: 192 case VBAD: 193 case VNON: 194 case VBLK: 195 break; 196 }; 197 return (error); 198 } 199 baddr = (caddr_t)0; 200 switch (vp->v_type) { 201 case VREG: 202 nfsstats.biocache_reads++; 203 lbn = uio->uio_offset / biosize; 204 on = uio->uio_offset & (biosize-1); 205 bn = lbn * (biosize / DEV_BSIZE); 206 not_readin = 1; 207 208 /* 209 * Start the read ahead(s), as required. 210 */ 211 if (nfs_numasync > 0 && nmp->nm_readahead > 0 && 212 lbn == vp->v_lastr + 1) { 213 for (nra = 0; nra < nmp->nm_readahead && 214 (lbn + 1 + nra) * biosize < np->n_size; nra++) { 215 rabn = (lbn + 1 + nra) * (biosize / DEV_BSIZE); 216 if (!incore(vp, rabn)) { 217 rabp = nfs_getcacheblk(vp, rabn, biosize, p); 218 if (!rabp) 219 return (EINTR); 220 if ((rabp->b_flags & (B_DELWRI | B_DONE)) == 0) { 221 rabp->b_flags |= (B_READ | B_ASYNC); 222 if (nfs_asyncio(rabp, cred)) { 223 rabp->b_flags |= B_INVAL; 224 brelse(rabp); 225 } 226 } else 227 brelse(rabp); 228 } 229 } 230 } 231 232 /* 233 * If the block is in the cache and has the required data 234 * in a valid region, just copy it out. 235 * Otherwise, get the block and write back/read in, 236 * as required. 237 */ 238 if ((bp = incore(vp, bn)) && 239 (bp->b_flags & (B_BUSY | B_WRITEINPROG)) == 240 (B_BUSY | B_WRITEINPROG)) 241 got_buf = 0; 242 else { 243 again: 244 bp = nfs_getcacheblk(vp, bn, biosize, p); 245 if (!bp) 246 return (EINTR); 247 got_buf = 1; 248 if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0) { 249 bp->b_flags |= B_READ; 250 not_readin = 0; 251 error = nfs_doio(bp, cred, p); 252 if (error) { 253 brelse(bp); 254 return (error); 255 } 256 } 257 } 258 n = min((unsigned)(biosize - on), uio->uio_resid); 259 diff = np->n_size - uio->uio_offset; 260 if (diff < n) 261 n = diff; 262 if (not_readin && n > 0) { 263 if (on < bp->b_validoff || (on + n) > bp->b_validend) { 264 if (!got_buf) { 265 bp = nfs_getcacheblk(vp, bn, biosize, p); 266 if (!bp) 267 return (EINTR); 268 got_buf = 1; 269 } 270 bp->b_flags |= B_INVAL; 271 if (bp->b_dirtyend > 0) { 272 if ((bp->b_flags & B_DELWRI) == 0) 273 panic("nfsbioread"); 274 if (VOP_BWRITE(bp) == EINTR) 275 return (EINTR); 276 } else 277 brelse(bp); 278 goto again; 279 } 280 } 281 vp->v_lastr = lbn; 282 diff = (on >= bp->b_validend) ? 0 : (bp->b_validend - on); 283 if (diff < n) 284 n = diff; 285 break; 286 case VLNK: 287 nfsstats.biocache_readlinks++; 288 bp = nfs_getcacheblk(vp, (daddr_t)0, NFS_MAXPATHLEN, p); 289 if (!bp) 290 return (EINTR); 291 if ((bp->b_flags & B_DONE) == 0) { 292 bp->b_flags |= B_READ; 293 if ((error = nfs_doio(bp, cred, p)) != 0) { 294 brelse(bp); 295 return (error); 296 } 297 } 298 n = min(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid); 299 got_buf = 1; 300 on = 0; 301 break; 302 case VDIR: 303 if (uio->uio_resid < NFS_DIRBLKSIZ) 304 return (0); 305 nfsstats.biocache_readdirs++; 306 bn = (daddr_t)uio->uio_offset; 307 bp = nfs_getcacheblk(vp, bn, NFS_DIRBLKSIZ, p); 308 if (!bp) 309 return (EINTR); 310 if ((bp->b_flags & B_DONE) == 0) { 311 bp->b_flags |= B_READ; 312 if ((error = nfs_doio(bp, cred, p)) != 0) { 313 brelse(bp); 314 return (error); 315 } 316 } 317 318 /* 319 * If not eof and read aheads are enabled, start one. 320 * (You need the current block first, so that you have the 321 * directory offset cookie of the next block. 322 */ 323 rabn = bp->b_blkno; 324 if (nfs_numasync > 0 && nmp->nm_readahead > 0 && 325 rabn != 0 && rabn != np->n_direofoffset && 326 !incore(vp, rabn)) { 327 rabp = nfs_getcacheblk(vp, rabn, NFS_DIRBLKSIZ, p); 328 if (rabp) { 329 if ((rabp->b_flags & (B_DONE | B_DELWRI)) == 0) { 330 rabp->b_flags |= (B_READ | B_ASYNC); 331 if (nfs_asyncio(rabp, cred)) { 332 rabp->b_flags |= B_INVAL; 333 brelse(rabp); 334 } 335 } else 336 brelse(rabp); 337 } 338 } 339 on = 0; 340 n = min(uio->uio_resid, NFS_DIRBLKSIZ - bp->b_resid); 341 got_buf = 1; 342 break; 343 case VBAD: 344 case VSOCK: 345 case VCHR: 346 case VBLK: 347 case VNON: 348 case VFIFO: 349 break; 350 }; 351 352 if (n > 0) { 353 if (!baddr) 354 baddr = bp->b_data; 355 error = uiomove(baddr + on, (int)n, uio); 356 } 357 switch (vp->v_type) { 358 case VLNK: 359 n = 0; 360 break; 361 case VDIR: 362 uio->uio_offset = bp->b_blkno; 363 break; 364 case VREG: 365 case VBAD: 366 case VFIFO: 367 case VSOCK: 368 case VCHR: 369 case VBLK: 370 case VNON: 371 break; 372 }; 373 if (got_buf) 374 brelse(bp); 375 } while (error == 0 && uio->uio_resid > 0 && n > 0); 376 return (error); 377 } 378 379 /* 380 * Vnode op for write using bio 381 */ 382 int 383 nfs_write(v) 384 void *v; 385 { 386 struct vop_write_args /* { 387 struct vnode a_vp; 388 struct uio *a_uio; 389 int a_ioflag; 390 struct ucred *a_cred; 391 } */ *ap = v; 392 register int biosize; 393 register struct uio *uio = ap->a_uio; 394 struct proc *p = uio->uio_procp; 395 register struct vnode *vp = ap->a_vp; 396 struct nfsnode *np = VTONFS(vp); 397 register struct ucred *cred = ap->a_cred; 398 int ioflag = ap->a_ioflag; 399 struct buf *bp; 400 struct vattr vattr; 401 struct nfsmount *nmp; 402 daddr_t lbn, bn; 403 int n, on, error = 0; 404 405 #ifdef DIAGNOSTIC 406 if (uio->uio_rw != UIO_WRITE) 407 panic("nfs_write mode"); 408 if (uio->uio_segflg == UIO_USERSPACE && uio->uio_procp != curproc) 409 panic("nfs_write proc"); 410 #endif 411 if (vp->v_type != VREG) 412 return (EIO); 413 if (np->n_flag & NWRITEERR) { 414 np->n_flag &= ~NWRITEERR; 415 return (np->n_error); 416 } 417 if (ioflag & (IO_APPEND | IO_SYNC)) { 418 if (np->n_flag & NMODIFIED) { 419 np->n_attrstamp = 0; 420 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 421 if (error) 422 return (error); 423 } 424 if (ioflag & IO_APPEND) { 425 np->n_attrstamp = 0; 426 error = VOP_GETATTR(vp, &vattr, cred, p); 427 if (error) 428 return (error); 429 uio->uio_offset = np->n_size; 430 } 431 } 432 nmp = VFSTONFS(vp->v_mount); 433 if (uio->uio_offset < 0) 434 return (EINVAL); 435 if (uio->uio_resid == 0) 436 return (0); 437 /* 438 * Maybe this should be above the vnode op call, but so long as 439 * file servers have no limits, i don't think it matters 440 */ 441 if (p && uio->uio_offset + uio->uio_resid > 442 p->p_rlimit[RLIMIT_FSIZE].rlim_cur) { 443 psignal(p, SIGXFSZ); 444 return (EFBIG); 445 } 446 /* 447 * I use nm_rsize, not nm_wsize so that all buffer cache blocks 448 * will be the same size within a filesystem. nfs_writerpc will 449 * still use nm_wsize when sizing the rpc's. 450 */ 451 biosize = nmp->nm_rsize; 452 do { 453 454 /* 455 * XXX make sure we aren't cached in the VM page cache 456 */ 457 (void)vnode_pager_uncache(vp); 458 459 /* 460 * Check for a valid write lease. 461 * If non-cachable, just do the rpc 462 */ 463 if ((nmp->nm_flag & NFSMNT_NQNFS) && 464 NQNFS_CKINVALID(vp, np, NQL_WRITE)) { 465 do { 466 error = nqnfs_getlease(vp, NQL_WRITE, cred, p); 467 } while (error == NQNFS_EXPIRED); 468 if (error) 469 return (error); 470 if (np->n_lrev != np->n_brev || 471 (np->n_flag & NQNFSNONCACHE)) { 472 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 473 if (error) 474 return (error); 475 np->n_brev = np->n_lrev; 476 } 477 } 478 if (np->n_flag & NQNFSNONCACHE) 479 return (nfs_writerpc(vp, uio, cred, ioflag)); 480 nfsstats.biocache_writes++; 481 lbn = uio->uio_offset / biosize; 482 on = uio->uio_offset & (biosize-1); 483 n = min((unsigned)(biosize - on), uio->uio_resid); 484 bn = lbn * (biosize / DEV_BSIZE); 485 again: 486 bp = nfs_getcacheblk(vp, bn, biosize, p); 487 if (!bp) 488 return (EINTR); 489 if (bp->b_wcred == NOCRED) { 490 crhold(cred); 491 bp->b_wcred = cred; 492 } 493 np->n_flag |= NMODIFIED; 494 if (uio->uio_offset + n > np->n_size) { 495 np->n_size = uio->uio_offset + n; 496 vnode_pager_setsize(vp, (u_long)np->n_size); 497 } 498 499 /* 500 * If the new write will leave a contiguous dirty 501 * area, just update the b_dirtyoff and b_dirtyend, 502 * otherwise force a write rpc of the old dirty area. 503 */ 504 if (bp->b_dirtyend > 0 && 505 (on > bp->b_dirtyend || (on + n) < bp->b_dirtyoff)) { 506 bp->b_proc = p; 507 if (VOP_BWRITE(bp) == EINTR) 508 return (EINTR); 509 goto again; 510 } 511 512 /* 513 * Check for valid write lease and get one as required. 514 * In case getblk() and/or bwrite() delayed us. 515 */ 516 if ((nmp->nm_flag & NFSMNT_NQNFS) && 517 NQNFS_CKINVALID(vp, np, NQL_WRITE)) { 518 do { 519 error = nqnfs_getlease(vp, NQL_WRITE, cred, p); 520 } while (error == NQNFS_EXPIRED); 521 if (error) { 522 brelse(bp); 523 return (error); 524 } 525 if (np->n_lrev != np->n_brev || 526 (np->n_flag & NQNFSNONCACHE)) { 527 brelse(bp); 528 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 529 if (error) 530 return (error); 531 np->n_brev = np->n_lrev; 532 goto again; 533 } 534 } 535 error = uiomove((char *)bp->b_data + on, n, uio); 536 if (error) { 537 bp->b_flags |= B_ERROR; 538 brelse(bp); 539 return (error); 540 } 541 if (bp->b_dirtyend > 0) { 542 bp->b_dirtyoff = min(on, bp->b_dirtyoff); 543 bp->b_dirtyend = max((on + n), bp->b_dirtyend); 544 } else { 545 bp->b_dirtyoff = on; 546 bp->b_dirtyend = on + n; 547 } 548 #ifndef notdef 549 if (bp->b_validend == 0 || bp->b_validend < bp->b_dirtyoff || 550 bp->b_validoff > bp->b_dirtyend) { 551 bp->b_validoff = bp->b_dirtyoff; 552 bp->b_validend = bp->b_dirtyend; 553 } else { 554 bp->b_validoff = min(bp->b_validoff, bp->b_dirtyoff); 555 bp->b_validend = max(bp->b_validend, bp->b_dirtyend); 556 } 557 #else 558 bp->b_validoff = bp->b_dirtyoff; 559 bp->b_validend = bp->b_dirtyend; 560 #endif 561 if (ioflag & IO_APPEND) 562 bp->b_flags |= B_APPENDWRITE; 563 564 /* 565 * If the lease is non-cachable or IO_SYNC do bwrite(). 566 */ 567 if ((np->n_flag & NQNFSNONCACHE) || (ioflag & IO_SYNC)) { 568 bp->b_proc = p; 569 if ((error = VOP_BWRITE(bp)) != 0) 570 return (error); 571 } else if ((n + on) == biosize && 572 (nmp->nm_flag & NFSMNT_NQNFS) == 0) { 573 bp->b_proc = (struct proc *)0; 574 bawrite(bp); 575 } else 576 bdwrite(bp); 577 } while (uio->uio_resid > 0 && n > 0); 578 return (0); 579 } 580 581 /* 582 * Get an nfs cache block. 583 * Allocate a new one if the block isn't currently in the cache 584 * and return the block marked busy. If the calling process is 585 * interrupted by a signal for an interruptible mount point, return 586 * NULL. 587 */ 588 struct buf * 589 nfs_getcacheblk(vp, bn, size, p) 590 struct vnode *vp; 591 daddr_t bn; 592 int size; 593 struct proc *p; 594 { 595 register struct buf *bp; 596 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 597 598 if (nmp->nm_flag & NFSMNT_INT) { 599 bp = getblk(vp, bn, size, PCATCH, 0); 600 while (bp == (struct buf *)0) { 601 if (nfs_sigintr(nmp, (struct nfsreq *)0, p)) 602 return ((struct buf *)0); 603 bp = getblk(vp, bn, size, 0, 2 * hz); 604 } 605 } else 606 bp = getblk(vp, bn, size, 0, 0); 607 return (bp); 608 } 609 610 /* 611 * Flush and invalidate all dirty buffers. If another process is already 612 * doing the flush, just wait for completion. 613 */ 614 int 615 nfs_vinvalbuf(vp, flags, cred, p, intrflg) 616 struct vnode *vp; 617 int flags; 618 struct ucred *cred; 619 struct proc *p; 620 int intrflg; 621 { 622 register struct nfsnode *np = VTONFS(vp); 623 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 624 int error = 0, slpflag, slptimeo; 625 626 if ((nmp->nm_flag & NFSMNT_INT) == 0) 627 intrflg = 0; 628 if (intrflg) { 629 slpflag = PCATCH; 630 slptimeo = 2 * hz; 631 } else { 632 slpflag = 0; 633 slptimeo = 0; 634 } 635 /* 636 * First wait for any other process doing a flush to complete. 637 */ 638 while (np->n_flag & NFLUSHINPROG) { 639 np->n_flag |= NFLUSHWANT; 640 error = tsleep((caddr_t)&np->n_flag, PRIBIO + 2, "nfsvinval", 641 slptimeo); 642 if (error && intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) 643 return (EINTR); 644 } 645 646 /* 647 * Now, flush as required. 648 */ 649 np->n_flag |= NFLUSHINPROG; 650 error = vinvalbuf(vp, flags, cred, p, slpflag, 0); 651 while (error) { 652 if (intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) { 653 np->n_flag &= ~NFLUSHINPROG; 654 if (np->n_flag & NFLUSHWANT) { 655 np->n_flag &= ~NFLUSHWANT; 656 wakeup((caddr_t)&np->n_flag); 657 } 658 return (EINTR); 659 } 660 error = vinvalbuf(vp, flags, cred, p, 0, slptimeo); 661 } 662 np->n_flag &= ~(NMODIFIED | NFLUSHINPROG); 663 if (np->n_flag & NFLUSHWANT) { 664 np->n_flag &= ~NFLUSHWANT; 665 wakeup((caddr_t)&np->n_flag); 666 } 667 return (0); 668 } 669 670 /* 671 * Initiate asynchronous I/O. Return an error if no nfsiods are available. 672 * This is mainly to avoid queueing async I/O requests when the nfsiods 673 * are all hung on a dead server. 674 */ 675 int 676 nfs_asyncio(bp, cred) 677 register struct buf *bp; 678 struct ucred *cred; 679 { 680 register int i; 681 682 if (nfs_numasync == 0) 683 return (EIO); 684 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 685 if (nfs_iodwant[i]) { 686 if (bp->b_flags & B_READ) { 687 if (bp->b_rcred == NOCRED && cred != NOCRED) { 688 crhold(cred); 689 bp->b_rcred = cred; 690 } 691 } else { 692 if (bp->b_wcred == NOCRED && cred != NOCRED) { 693 crhold(cred); 694 bp->b_wcred = cred; 695 } 696 } 697 698 TAILQ_INSERT_TAIL(&nfs_bufq, bp, b_freelist); 699 nfs_iodwant[i] = (struct proc *)0; 700 wakeup((caddr_t)&nfs_iodwant[i]); 701 return (0); 702 } 703 return (EIO); 704 } 705 706 /* 707 * Do an I/O operation to/from a cache block. This may be called 708 * synchronously or from an nfsiod. 709 */ 710 int 711 nfs_doio(bp, cr, p) 712 register struct buf *bp; 713 struct ucred *cr; 714 struct proc *p; 715 { 716 register struct uio *uiop; 717 register struct vnode *vp; 718 struct nfsnode *np; 719 struct nfsmount *nmp; 720 int error = 0, diff, len; 721 struct uio uio; 722 struct iovec io; 723 724 vp = bp->b_vp; 725 np = VTONFS(vp); 726 nmp = VFSTONFS(vp->v_mount); 727 uiop = &uio; 728 uiop->uio_iov = &io; 729 uiop->uio_iovcnt = 1; 730 uiop->uio_segflg = UIO_SYSSPACE; 731 uiop->uio_procp = p; 732 733 /* 734 * Historically, paging was done with physio, but no more... 735 */ 736 if (bp->b_flags & B_PHYS) { 737 /* 738 * ...though reading /dev/drum still gets us here. 739 */ 740 io.iov_len = uiop->uio_resid = bp->b_bcount; 741 /* mapping was done by vmapbuf() */ 742 io.iov_base = bp->b_data; 743 uiop->uio_offset = bp->b_blkno * DEV_BSIZE; 744 if (bp->b_flags & B_READ) { 745 uiop->uio_rw = UIO_READ; 746 nfsstats.read_physios++; 747 error = nfs_readrpc(vp, uiop, cr); 748 } else { 749 uiop->uio_rw = UIO_WRITE; 750 nfsstats.write_physios++; 751 error = nfs_writerpc(vp, uiop, cr, 0); 752 } 753 if (error) { 754 bp->b_flags |= B_ERROR; 755 bp->b_error = error; 756 } 757 } else if (bp->b_flags & B_READ) { 758 io.iov_len = uiop->uio_resid = bp->b_bcount; 759 io.iov_base = bp->b_data; 760 uiop->uio_rw = UIO_READ; 761 switch (vp->v_type) { 762 case VREG: 763 uiop->uio_offset = bp->b_blkno * DEV_BSIZE; 764 nfsstats.read_bios++; 765 error = nfs_readrpc(vp, uiop, cr); 766 if (!error) { 767 bp->b_validoff = 0; 768 if (uiop->uio_resid) { 769 /* 770 * If len > 0, there is a hole in the file and 771 * no writes after the hole have been pushed to 772 * the server yet. 773 * Just zero fill the rest of the valid area. 774 */ 775 diff = bp->b_bcount - uiop->uio_resid; 776 len = np->n_size - (bp->b_blkno * DEV_BSIZE 777 + diff); 778 if (len > 0) { 779 len = min(len, uiop->uio_resid); 780 bzero((char *)bp->b_data + diff, len); 781 bp->b_validend = diff + len; 782 } else 783 bp->b_validend = diff; 784 } else 785 bp->b_validend = bp->b_bcount; 786 } 787 if (p && (vp->v_flag & VTEXT) && 788 (((nmp->nm_flag & NFSMNT_NQNFS) && 789 NQNFS_CKINVALID(vp, np, NQL_READ) && 790 np->n_lrev != np->n_brev) || 791 (!(nmp->nm_flag & NFSMNT_NQNFS) && 792 np->n_mtime != np->n_vattr.va_mtime.tv_sec))) { 793 uprintf("Process killed due to text file modification\n"); 794 psignal(p, SIGKILL); 795 p->p_holdcnt++; 796 } 797 break; 798 case VLNK: 799 uiop->uio_offset = 0; 800 nfsstats.readlink_bios++; 801 error = nfs_readlinkrpc(vp, uiop, cr); 802 break; 803 case VDIR: 804 uiop->uio_offset = bp->b_lblkno; 805 nfsstats.readdir_bios++; 806 if (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) 807 error = nfs_readdirlookrpc(vp, uiop, cr); 808 else 809 error = nfs_readdirrpc(vp, uiop, cr); 810 /* 811 * Save offset cookie in b_blkno. 812 */ 813 bp->b_blkno = uiop->uio_offset; 814 break; 815 case VNON: 816 case VBLK: 817 case VCHR: 818 case VFIFO: 819 case VBAD: 820 case VSOCK: 821 }; 822 if (error) { 823 bp->b_flags |= B_ERROR; 824 bp->b_error = error; 825 } 826 } else { 827 io.iov_len = uiop->uio_resid = bp->b_dirtyend 828 - bp->b_dirtyoff; 829 uiop->uio_offset = (bp->b_blkno * DEV_BSIZE) 830 + bp->b_dirtyoff; 831 io.iov_base = (char *)bp->b_data + bp->b_dirtyoff; 832 uiop->uio_rw = UIO_WRITE; 833 nfsstats.write_bios++; 834 if (bp->b_flags & B_APPENDWRITE) 835 error = nfs_writerpc(vp, uiop, cr, IO_APPEND); 836 else 837 error = nfs_writerpc(vp, uiop, cr, 0); 838 bp->b_flags &= ~(B_WRITEINPROG | B_APPENDWRITE); 839 840 /* 841 * For an interrupted write, the buffer is still valid and the 842 * write hasn't been pushed to the server yet, so we can't set 843 * B_ERROR and report the interruption by setting B_EINTR. For 844 * the B_ASYNC case, B_EINTR is not relevant, so the rpc attempt 845 * is essentially a noop. 846 */ 847 if (error == EINTR) { 848 bp->b_flags &= ~B_INVAL; 849 bp->b_flags |= B_DELWRI; 850 851 /* 852 * Since for the B_ASYNC case, nfs_bwrite() has reassigned the 853 * buffer to the clean list, we have to reassign it back to the 854 * dirty one. Ugh. 855 */ 856 if (bp->b_flags & B_ASYNC) 857 reassignbuf(bp, vp); 858 else 859 bp->b_flags |= B_EINTR; 860 } else { 861 if (error) { 862 bp->b_flags |= B_ERROR; 863 bp->b_error = np->n_error = error; 864 np->n_flag |= NWRITEERR; 865 } 866 bp->b_dirtyoff = bp->b_dirtyend = 0; 867 } 868 } 869 bp->b_resid = uiop->uio_resid; 870 biodone(bp); 871 return (error); 872 } 873
Indexes created Wed Oct 15 07:09:58 GMT 2025