nfs_bio.c revision 13612
1/* 2 * Copyright (c) 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * Rick Macklem at The University of Guelph. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by the University of 19 * California, Berkeley and its contributors. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)nfs_bio.c 8.5 (Berkeley) 1/4/94 37 * $Id: nfs_bio.c,v 1.21 1995/12/17 21:12:13 phk Exp $ 38 */ 39 40#include <sys/param.h> 41#include <sys/systm.h> 42#include <sys/resourcevar.h> 43#include <sys/signalvar.h> 44#include <sys/proc.h> 45#include <sys/buf.h> 46#include <sys/vnode.h> 47#include <sys/mount.h> 48#include <sys/kernel.h> 49 50#include <vm/vm.h> 51#include <vm/vm_param.h> 52#include <vm/vm_extern.h> 53 54#include <nfs/rpcv2.h> 55#include <nfs/nfsproto.h> 56#include <nfs/nfs.h> 57#include <nfs/nfsmount.h> 58#include <nfs/nqnfs.h> 59#include <nfs/nfsnode.h> 60 61static struct buf *nfs_getcacheblk __P((struct vnode *vp, daddr_t bn, int size, 62 struct proc *p)); 63 64extern struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON]; 65extern int nfs_numasync; 66extern struct nfsstats nfsstats; 67 68/* 69 * Ifdefs for FreeBSD-current's merged VM/buffer cache. It is unfortunate 70 * that this isn't done inside getblk() and brelse() so these calls 71 * wouldn't need to be here. 72 */ 73#ifdef B_VMIO 74#define vnode_pager_uncache(vp) 75#else 76#define vfs_busy_pages(bp, f) 77#define vfs_unbusy_pages(bp) 78#define vfs_dirty_pages(bp) 79#endif 80 81/* 82 * Vnode op for read using bio 83 * Any similarity to readip() is purely coincidental 84 */ 85int 86nfs_bioread(vp, uio, ioflag, cred) 87 register struct vnode *vp; 88 register struct uio *uio; 89 int ioflag; 90 struct ucred *cred; 91{ 92 register struct nfsnode *np = VTONFS(vp); 93 register int biosize, diff, i; 94 struct buf *bp = 0, *rabp; 95 struct vattr vattr; 96 struct proc *p; 97 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 98 daddr_t lbn, rabn; 99 int bufsize; 100 int nra, error = 0, n = 0, on = 0, not_readin; 101 102#ifdef DIAGNOSTIC 103 if (uio->uio_rw != UIO_READ) 104 panic("nfs_read mode"); 105#endif 106 if (uio->uio_resid == 0) 107 return (0); 108 if (uio->uio_offset < 0) 109 return (EINVAL); 110 p = uio->uio_procp; 111 if ((nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_GOTFSINFO)) == NFSMNT_NFSV3) 112 (void)nfs_fsinfo(nmp, vp, cred, p); 113 biosize = vp->v_mount->mnt_stat.f_iosize; 114 /* 115 * For nfs, cache consistency can only be maintained approximately. 116 * Although RFC1094 does not specify the criteria, the following is 117 * believed to be compatible with the reference port. 118 * For nqnfs, full cache consistency is maintained within the loop. 119 * For nfs: 120 * If the file's modify time on the server has changed since the 121 * last read rpc or you have written to the file, 122 * you may have lost data cache consistency with the 123 * server, so flush all of the file's data out of the cache. 124 * Then force a getattr rpc to ensure that you have up to date 125 * attributes. 126 * NB: This implies that cache data can be read when up to 127 * NFS_ATTRTIMEO seconds out of date. If you find that you need current 128 * attributes this could be forced by setting n_attrstamp to 0 before 129 * the VOP_GETATTR() call. 130 */ 131 if ((nmp->nm_flag & NFSMNT_NQNFS) == 0) { 132 if (np->n_flag & NMODIFIED) { 133 if (vp->v_type != VREG) { 134 if (vp->v_type != VDIR) 135 panic("nfs: bioread, not dir"); 136 nfs_invaldir(vp); 137 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 138 if (error) 139 return (error); 140 } 141 np->n_attrstamp = 0; 142 error = VOP_GETATTR(vp, &vattr, cred, p); 143 if (error) 144 return (error); 145 np->n_mtime = vattr.va_mtime.ts_sec; 146 } else { 147 error = VOP_GETATTR(vp, &vattr, cred, p); 148 if (error) 149 return (error); 150 if (np->n_mtime != vattr.va_mtime.ts_sec) { 151 if (vp->v_type == VDIR) 152 nfs_invaldir(vp); 153 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 154 if (error) 155 return (error); 156 np->n_mtime = vattr.va_mtime.ts_sec; 157 } 158 } 159 } 160 do { 161 162 /* 163 * Get a valid lease. If cached data is stale, flush it. 164 */ 165 if (nmp->nm_flag & NFSMNT_NQNFS) { 166 if (NQNFS_CKINVALID(vp, np, ND_READ)) { 167 do { 168 error = nqnfs_getlease(vp, ND_READ, cred, p); 169 } while (error == NQNFS_EXPIRED); 170 if (error) 171 return (error); 172 if (np->n_lrev != np->n_brev || 173 (np->n_flag & NQNFSNONCACHE) || 174 ((np->n_flag & NMODIFIED) && vp->v_type == VDIR)) { 175 if (vp->v_type == VDIR) 176 nfs_invaldir(vp); 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_invaldir(vp); 184 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 185 if (error) 186 return (error); 187 } 188 } 189 if (np->n_flag & NQNFSNONCACHE) { 190 switch (vp->v_type) { 191 case VREG: 192 return (nfs_readrpc(vp, uio, cred)); 193 case VLNK: 194 return (nfs_readlinkrpc(vp, uio, cred)); 195 case VDIR: 196 break; 197 default: 198 printf(" NQNFSNONCACHE: type %x unexpected\n", 199 vp->v_type); 200 }; 201 } 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 not_readin = 1; 208 209 /* 210 * Start the read ahead(s), as required. 211 */ 212 if (nfs_numasync > 0 && nmp->nm_readahead > 0) { 213 for (nra = 0; nra < nmp->nm_readahead && 214 (off_t)(lbn + 1 + nra) * biosize < np->n_size; nra++) { 215 rabn = lbn + 1 + nra; 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_CACHE|B_DELWRI)) == 0) { 221 rabp->b_flags |= (B_READ | B_ASYNC); 222 vfs_busy_pages(rabp, 0); 223 if (nfs_asyncio(rabp, cred)) { 224 rabp->b_flags |= B_INVAL|B_ERROR; 225 vfs_unbusy_pages(rabp); 226 brelse(rabp); 227 } 228 } else { 229 brelse(rabp); 230 } 231 } 232 } 233 } 234 235 /* 236 * If the block is in the cache and has the required data 237 * in a valid region, just copy it out. 238 * Otherwise, get the block and write back/read in, 239 * as required. 240 */ 241again: 242 bufsize = biosize; 243 if ((off_t)(lbn + 1) * biosize > np->n_size && 244 (off_t)(lbn + 1) * biosize - np->n_size < biosize) { 245 bufsize = np->n_size - lbn * biosize; 246 bufsize = (bufsize + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1); 247 } 248 bp = nfs_getcacheblk(vp, lbn, bufsize, p); 249 if (!bp) 250 return (EINTR); 251 if ((bp->b_flags & B_CACHE) == 0) { 252 bp->b_flags |= B_READ; 253 not_readin = 0; 254 vfs_busy_pages(bp, 0); 255 error = nfs_doio(bp, cred, p); 256 if (error) { 257 brelse(bp); 258 return (error); 259 } 260 } 261 if (bufsize > on) { 262 n = min((unsigned)(bufsize - on), uio->uio_resid); 263 } else { 264 n = 0; 265 } 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 bp->b_flags |= B_NOCACHE; 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_CACHE) == 0) { 293 bp->b_flags |= B_READ; 294 vfs_busy_pages(bp, 0); 295 error = nfs_doio(bp, cred, p); 296 if (error) { 297 bp->b_flags |= B_ERROR; 298 brelse(bp); 299 return (error); 300 } 301 } 302 n = min(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid); 303 on = 0; 304 break; 305 case VDIR: 306 nfsstats.biocache_readdirs++; 307 lbn = uio->uio_offset / NFS_DIRBLKSIZ; 308 on = uio->uio_offset & (NFS_DIRBLKSIZ - 1); 309 bp = nfs_getcacheblk(vp, lbn, NFS_DIRBLKSIZ, p); 310 if (!bp) 311 return (EINTR); 312 if ((bp->b_flags & B_CACHE) == 0) { 313 bp->b_flags |= B_READ; 314 vfs_busy_pages(bp, 0); 315 error = nfs_doio(bp, cred, p); 316 if (error) { 317 brelse(bp); 318 while (error == NFSERR_BAD_COOKIE) { 319 nfs_invaldir(vp); 320 error = nfs_vinvalbuf(vp, 0, cred, p, 1); 321 /* 322 * Yuck! The directory has been modified on the 323 * server. The only way to get the block is by 324 * reading from the beginning to get all the 325 * offset cookies. 326 */ 327 for (i = 0; i <= lbn && !error; i++) { 328 bp = nfs_getcacheblk(vp, i, NFS_DIRBLKSIZ, p); 329 if (!bp) 330 return (EINTR); 331 if ((bp->b_flags & B_DONE) == 0) { 332 bp->b_flags |= B_READ; 333 vfs_busy_pages(bp, 0); 334 error = nfs_doio(bp, cred, p); 335 if (error) 336 brelse(bp); 337 } 338 } 339 } 340 if (error) 341 return (error); 342 } 343 } 344 345 /* 346 * If not eof and read aheads are enabled, start one. 347 * (You need the current block first, so that you have the 348 * directory offset cookie of the next block.) 349 */ 350 if (nfs_numasync > 0 && nmp->nm_readahead > 0 && 351 (np->n_direofoffset == 0 || 352 (lbn + 1) * NFS_DIRBLKSIZ < np->n_direofoffset) && 353 !(np->n_flag & NQNFSNONCACHE) && 354 !incore(vp, lbn + 1)) { 355 rabp = nfs_getcacheblk(vp, lbn + 1, NFS_DIRBLKSIZ, p); 356 if (rabp) { 357 if ((rabp->b_flags & (B_CACHE|B_DELWRI)) == 0) { 358 rabp->b_flags |= (B_READ | B_ASYNC); 359 vfs_busy_pages(rabp, 0); 360 if (nfs_asyncio(rabp, cred)) { 361 rabp->b_flags |= B_INVAL|B_ERROR; 362 vfs_unbusy_pages(rabp); 363 brelse(rabp); 364 } 365 } else { 366 brelse(rabp); 367 } 368 } 369 } 370 n = min(uio->uio_resid, NFS_DIRBLKSIZ - bp->b_resid - on); 371 break; 372 default: 373 printf(" nfs_bioread: type %x unexpected\n",vp->v_type); 374 break; 375 }; 376 377 if (n > 0) { 378 error = uiomove(bp->b_data + on, (int)n, uio); 379 } 380 switch (vp->v_type) { 381 case VREG: 382 break; 383 case VLNK: 384 n = 0; 385 break; 386 case VDIR: 387 if (np->n_flag & NQNFSNONCACHE) 388 bp->b_flags |= B_INVAL; 389 break; 390 default: 391 printf(" nfs_bioread: type %x unexpected\n",vp->v_type); 392 } 393 brelse(bp); 394 } while (error == 0 && uio->uio_resid > 0 && n > 0); 395 return (error); 396} 397 398/* 399 * Vnode op for write using bio 400 */ 401int 402nfs_write(ap) 403 struct vop_write_args /* { 404 struct vnode *a_vp; 405 struct uio *a_uio; 406 int a_ioflag; 407 struct ucred *a_cred; 408 } */ *ap; 409{ 410 register int biosize; 411 register struct uio *uio = ap->a_uio; 412 struct proc *p = uio->uio_procp; 413 register struct vnode *vp = ap->a_vp; 414 struct nfsnode *np = VTONFS(vp); 415 register struct ucred *cred = ap->a_cred; 416 int ioflag = ap->a_ioflag; 417 struct buf *bp; 418 struct vattr vattr; 419 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 420 daddr_t lbn; 421 int bufsize; 422 int n, on, error = 0, iomode, must_commit; 423 424#ifdef DIAGNOSTIC 425 if (uio->uio_rw != UIO_WRITE) 426 panic("nfs_write mode"); 427 if (uio->uio_segflg == UIO_USERSPACE && uio->uio_procp != curproc) 428 panic("nfs_write proc"); 429#endif 430 if (vp->v_type != VREG) 431 return (EIO); 432 if (np->n_flag & NWRITEERR) { 433 np->n_flag &= ~NWRITEERR; 434 return (np->n_error); 435 } 436 if ((nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_GOTFSINFO)) == NFSMNT_NFSV3) 437 (void)nfs_fsinfo(nmp, vp, cred, p); 438 if (ioflag & (IO_APPEND | IO_SYNC)) { 439 if (np->n_flag & NMODIFIED) { 440 np->n_attrstamp = 0; 441 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 442 if (error) 443 return (error); 444 } 445 if (ioflag & IO_APPEND) { 446 np->n_attrstamp = 0; 447 error = VOP_GETATTR(vp, &vattr, cred, p); 448 if (error) 449 return (error); 450 uio->uio_offset = np->n_size; 451 } 452 } 453 if (uio->uio_offset < 0) 454 return (EINVAL); 455 if (uio->uio_resid == 0) 456 return (0); 457 /* 458 * Maybe this should be above the vnode op call, but so long as 459 * file servers have no limits, i don't think it matters 460 */ 461 if (p && uio->uio_offset + uio->uio_resid > 462 p->p_rlimit[RLIMIT_FSIZE].rlim_cur) { 463 psignal(p, SIGXFSZ); 464 return (EFBIG); 465 } 466 /* 467 * I use nm_rsize, not nm_wsize so that all buffer cache blocks 468 * will be the same size within a filesystem. nfs_writerpc will 469 * still use nm_wsize when sizing the rpc's. 470 */ 471 biosize = vp->v_mount->mnt_stat.f_iosize; 472 do { 473 474 /* 475 * XXX make sure we aren't cached in the VM page cache 476 */ 477 /* 478 * Check for a valid write lease. 479 */ 480 if ((nmp->nm_flag & NFSMNT_NQNFS) && 481 NQNFS_CKINVALID(vp, np, ND_WRITE)) { 482 do { 483 error = nqnfs_getlease(vp, ND_WRITE, cred, p); 484 } while (error == NQNFS_EXPIRED); 485 if (error) 486 return (error); 487 if (np->n_lrev != np->n_brev || 488 (np->n_flag & NQNFSNONCACHE)) { 489 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 490 if (error) 491 return (error); 492 np->n_brev = np->n_lrev; 493 } 494 } 495 if ((np->n_flag & NQNFSNONCACHE) && uio->uio_iovcnt == 1) { 496 iomode = NFSV3WRITE_FILESYNC; 497 error = nfs_writerpc(vp, uio, cred, &iomode, &must_commit); 498 if (must_commit) 499 nfs_clearcommit(vp->v_mount); 500 return (error); 501 } 502 nfsstats.biocache_writes++; 503 lbn = uio->uio_offset / biosize; 504 on = uio->uio_offset & (biosize-1); 505 n = min((unsigned)(biosize - on), uio->uio_resid); 506again: 507 if (uio->uio_offset + n > np->n_size) { 508 np->n_size = uio->uio_offset + n; 509 vnode_pager_setsize(vp, (u_long)np->n_size); 510 } 511 bufsize = biosize; 512 if ((lbn + 1) * biosize > np->n_size) { 513 bufsize = np->n_size - lbn * biosize; 514 bufsize = (bufsize + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1); 515 } 516 bp = nfs_getcacheblk(vp, lbn, bufsize, p); 517 if (!bp) 518 return (EINTR); 519 if (bp->b_wcred == NOCRED) { 520 crhold(cred); 521 bp->b_wcred = cred; 522 } 523 np->n_flag |= NMODIFIED; 524 525 if ((bp->b_blkno * DEV_BSIZE) + bp->b_dirtyend > np->n_size) { 526 bp->b_dirtyend = np->n_size - (bp->b_blkno * DEV_BSIZE); 527 } 528 529 /* 530 * If the new write will leave a contiguous dirty 531 * area, just update the b_dirtyoff and b_dirtyend, 532 * otherwise force a write rpc of the old dirty area. 533 */ 534 if (bp->b_dirtyend > 0 && 535 (on > bp->b_dirtyend || (on + n) < bp->b_dirtyoff)) { 536 bp->b_proc = p; 537 if (VOP_BWRITE(bp) == EINTR) 538 return (EINTR); 539 goto again; 540 } 541 542 /* 543 * Check for valid write lease and get one as required. 544 * In case getblk() and/or bwrite() delayed us. 545 */ 546 if ((nmp->nm_flag & NFSMNT_NQNFS) && 547 NQNFS_CKINVALID(vp, np, ND_WRITE)) { 548 do { 549 error = nqnfs_getlease(vp, ND_WRITE, cred, p); 550 } while (error == NQNFS_EXPIRED); 551 if (error) { 552 brelse(bp); 553 return (error); 554 } 555 if (np->n_lrev != np->n_brev || 556 (np->n_flag & NQNFSNONCACHE)) { 557 brelse(bp); 558 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 559 if (error) 560 return (error); 561 np->n_brev = np->n_lrev; 562 goto again; 563 } 564 } 565 error = uiomove((char *)bp->b_data + on, n, uio); 566 if (error) { 567 bp->b_flags |= B_ERROR; 568 brelse(bp); 569 return (error); 570 } 571 if (bp->b_dirtyend > 0) { 572 bp->b_dirtyoff = min(on, bp->b_dirtyoff); 573 bp->b_dirtyend = max((on + n), bp->b_dirtyend); 574 } else { 575 bp->b_dirtyoff = on; 576 bp->b_dirtyend = on + n; 577 } 578 if (bp->b_validend == 0 || bp->b_validend < bp->b_dirtyoff || 579 bp->b_validoff > bp->b_dirtyend) { 580 bp->b_validoff = bp->b_dirtyoff; 581 bp->b_validend = bp->b_dirtyend; 582 } else { 583 bp->b_validoff = min(bp->b_validoff, bp->b_dirtyoff); 584 bp->b_validend = max(bp->b_validend, bp->b_dirtyend); 585 } 586 /* 587 * If the lease is non-cachable or IO_SYNC do bwrite(). 588 */ 589 if ((np->n_flag & NQNFSNONCACHE) || (ioflag & IO_SYNC)) { 590 bp->b_proc = p; 591 error = VOP_BWRITE(bp); 592 if (error) 593 return (error); 594 if (np->n_flag & NQNFSNONCACHE) { 595 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 596 if (error) 597 return (error); 598 } 599 } else if ((n + on) == biosize && 600 (nmp->nm_flag & NFSMNT_NQNFS) == 0) { 601 bp->b_proc = (struct proc *)0; 602 bp->b_flags |= B_ASYNC; 603 (void)nfs_writebp(bp, 0); 604 } else 605 bdwrite(bp); 606 } while (uio->uio_resid > 0 && n > 0); 607 return (0); 608} 609 610/* 611 * Get an nfs cache block. 612 * Allocate a new one if the block isn't currently in the cache 613 * and return the block marked busy. If the calling process is 614 * interrupted by a signal for an interruptible mount point, return 615 * NULL. 616 */ 617static struct buf * 618nfs_getcacheblk(vp, bn, size, p) 619 struct vnode *vp; 620 daddr_t bn; 621 int size; 622 struct proc *p; 623{ 624 register struct buf *bp; 625 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 626 int biosize = vp->v_mount->mnt_stat.f_iosize; 627 628 if (nmp->nm_flag & NFSMNT_INT) { 629 bp = getblk(vp, bn, size, PCATCH, 0); 630 while (bp == (struct buf *)0) { 631 if (nfs_sigintr(nmp, (struct nfsreq *)0, p)) 632 return ((struct buf *)0); 633 bp = getblk(vp, bn, size, 0, 2 * hz); 634 } 635 } else 636 bp = getblk(vp, bn, size, 0, 0); 637 638 if( vp->v_type == VREG) 639 bp->b_blkno = (bn * biosize) / DEV_BSIZE; 640 641 return (bp); 642} 643 644/* 645 * Flush and invalidate all dirty buffers. If another process is already 646 * doing the flush, just wait for completion. 647 */ 648int 649nfs_vinvalbuf(vp, flags, cred, p, intrflg) 650 struct vnode *vp; 651 int flags; 652 struct ucred *cred; 653 struct proc *p; 654 int intrflg; 655{ 656 register struct nfsnode *np = VTONFS(vp); 657 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 658 int error = 0, slpflag, slptimeo; 659 660 if ((nmp->nm_flag & NFSMNT_INT) == 0) 661 intrflg = 0; 662 if (intrflg) { 663 slpflag = PCATCH; 664 slptimeo = 2 * hz; 665 } else { 666 slpflag = 0; 667 slptimeo = 0; 668 } 669 /* 670 * First wait for any other process doing a flush to complete. 671 */ 672 while (np->n_flag & NFLUSHINPROG) { 673 np->n_flag |= NFLUSHWANT; 674 error = tsleep((caddr_t)&np->n_flag, PRIBIO + 2, "nfsvinval", 675 slptimeo); 676 if (error && intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) 677 return (EINTR); 678 } 679 680 /* 681 * Now, flush as required. 682 */ 683 np->n_flag |= NFLUSHINPROG; 684 error = vinvalbuf(vp, flags, cred, p, slpflag, 0); 685 while (error) { 686 if (intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) { 687 np->n_flag &= ~NFLUSHINPROG; 688 if (np->n_flag & NFLUSHWANT) { 689 np->n_flag &= ~NFLUSHWANT; 690 wakeup((caddr_t)&np->n_flag); 691 } 692 return (EINTR); 693 } 694 error = vinvalbuf(vp, flags, cred, p, 0, slptimeo); 695 } 696 np->n_flag &= ~(NMODIFIED | NFLUSHINPROG); 697 if (np->n_flag & NFLUSHWANT) { 698 np->n_flag &= ~NFLUSHWANT; 699 wakeup((caddr_t)&np->n_flag); 700 } 701 return (0); 702} 703 704/* 705 * Initiate asynchronous I/O. Return an error if no nfsiods are available. 706 * This is mainly to avoid queueing async I/O requests when the nfsiods 707 * are all hung on a dead server. 708 */ 709int 710nfs_asyncio(bp, cred) 711 register struct buf *bp; 712 struct ucred *cred; 713{ 714 register int i; 715 716 if (nfs_numasync == 0) 717 return (EIO); 718 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 719 if (nfs_iodwant[i]) { 720 if (bp->b_flags & B_READ) { 721 if (bp->b_rcred == NOCRED && cred != NOCRED) { 722 crhold(cred); 723 bp->b_rcred = cred; 724 } 725 } else { 726 bp->b_flags |= B_WRITEINPROG; 727 if (bp->b_wcred == NOCRED && cred != NOCRED) { 728 crhold(cred); 729 bp->b_wcred = cred; 730 } 731 } 732 733 TAILQ_INSERT_TAIL(&nfs_bufq, bp, b_freelist); 734 nfs_iodwant[i] = (struct proc *)0; 735 wakeup((caddr_t)&nfs_iodwant[i]); 736 return (0); 737 } 738 739 /* 740 * If it is a read or a write already marked B_WRITEINPROG or B_NOCACHE 741 * return EIO so the process will call nfs_doio() and do it 742 * synchronously. 743 */ 744 if (bp->b_flags & (B_READ | B_WRITEINPROG | B_NOCACHE)) 745 return (EIO); 746 747 /* 748 * Just turn the async write into a delayed write, instead of 749 * doing in synchronously. Hopefully, at least one of the nfsiods 750 * is currently doing a write for this file and will pick up the 751 * delayed writes before going back to sleep. 752 */ 753 bp->b_flags |= B_DELWRI; 754 reassignbuf(bp, bp->b_vp); 755 biodone(bp); 756 return (0); 757} 758 759/* 760 * Do an I/O operation to/from a cache block. This may be called 761 * synchronously or from an nfsiod. 762 */ 763int 764nfs_doio(bp, cr, p) 765 register struct buf *bp; 766 struct ucred *cr; 767 struct proc *p; 768{ 769 register struct uio *uiop; 770 register struct vnode *vp; 771 struct nfsnode *np; 772 struct nfsmount *nmp; 773 int error = 0, diff, len, iomode, must_commit = 0; 774 struct uio uio; 775 struct iovec io; 776 777 vp = bp->b_vp; 778 np = VTONFS(vp); 779 nmp = VFSTONFS(vp->v_mount); 780 uiop = &uio; 781 uiop->uio_iov = &io; 782 uiop->uio_iovcnt = 1; 783 uiop->uio_segflg = UIO_SYSSPACE; 784 uiop->uio_procp = p; 785 786 /* 787 * Historically, paging was done with physio, but no more. 788 */ 789 if (bp->b_flags & B_PHYS) { 790 /* 791 * ...though reading /dev/drum still gets us here. 792 */ 793 io.iov_len = uiop->uio_resid = bp->b_bcount; 794 /* mapping was done by vmapbuf() */ 795 io.iov_base = bp->b_data; 796 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; 797 if (bp->b_flags & B_READ) { 798 uiop->uio_rw = UIO_READ; 799 nfsstats.read_physios++; 800 error = nfs_readrpc(vp, uiop, cr); 801 } else { 802 int com; 803 804 iomode = NFSV3WRITE_DATASYNC; 805 uiop->uio_rw = UIO_WRITE; 806 nfsstats.write_physios++; 807 error = nfs_writerpc(vp, uiop, cr, &iomode, &com); 808 } 809 if (error) { 810 bp->b_flags |= B_ERROR; 811 bp->b_error = error; 812 } 813 } else if (bp->b_flags & B_READ) { 814 io.iov_len = uiop->uio_resid = bp->b_bcount; 815 io.iov_base = bp->b_data; 816 uiop->uio_rw = UIO_READ; 817 switch (vp->v_type) { 818 case VREG: 819 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; 820 nfsstats.read_bios++; 821 error = nfs_readrpc(vp, uiop, cr); 822 if (!error) { 823 bp->b_validoff = 0; 824 if (uiop->uio_resid) { 825 /* 826 * If len > 0, there is a hole in the file and 827 * no writes after the hole have been pushed to 828 * the server yet. 829 * Just zero fill the rest of the valid area. 830 */ 831 diff = bp->b_bcount - uiop->uio_resid; 832 len = np->n_size - (((u_quad_t)bp->b_blkno) * DEV_BSIZE 833 + diff); 834 if (len > 0) { 835 len = min(len, uiop->uio_resid); 836 bzero((char *)bp->b_data + diff, len); 837 bp->b_validend = diff + len; 838 } else 839 bp->b_validend = diff; 840 } else 841 bp->b_validend = bp->b_bcount; 842 } 843 if (p && (vp->v_flag & VTEXT) && 844 (((nmp->nm_flag & NFSMNT_NQNFS) && 845 NQNFS_CKINVALID(vp, np, ND_READ) && 846 np->n_lrev != np->n_brev) || 847 (!(nmp->nm_flag & NFSMNT_NQNFS) && 848 np->n_mtime != np->n_vattr.va_mtime.ts_sec))) { 849 uprintf("Process killed due to text file modification\n"); 850 psignal(p, SIGKILL); 851#ifdef __NetBSD__ 852 p->p_holdcnt++; 853#else 854 p->p_flag |= P_NOSWAP; 855#endif 856 } 857 break; 858 case VLNK: 859 uiop->uio_offset = (off_t)0; 860 nfsstats.readlink_bios++; 861 error = nfs_readlinkrpc(vp, uiop, cr); 862 break; 863 case VDIR: 864 nfsstats.readdir_bios++; 865 uiop->uio_offset = ((u_quad_t)bp->b_lblkno) * NFS_DIRBLKSIZ; 866 if (nmp->nm_flag & NFSMNT_RDIRPLUS) { 867 error = nfs_readdirplusrpc(vp, uiop, cr); 868 if (error == NFSERR_NOTSUPP) 869 nmp->nm_flag &= ~NFSMNT_RDIRPLUS; 870 } 871 if ((nmp->nm_flag & NFSMNT_RDIRPLUS) == 0) 872 error = nfs_readdirrpc(vp, uiop, cr); 873 break; 874 default: 875 printf("nfs_doio: type %x unexpected\n",vp->v_type); 876 break; 877 }; 878 if (error) { 879 bp->b_flags |= B_ERROR; 880 bp->b_error = error; 881 } 882 } else { 883 if (((bp->b_blkno * DEV_BSIZE) + bp->b_dirtyend) > np->n_size) 884 bp->b_dirtyend = np->n_size - (bp->b_blkno * DEV_BSIZE); 885 886 if (bp->b_dirtyend > bp->b_dirtyoff) { 887 io.iov_len = uiop->uio_resid = bp->b_dirtyend 888 - bp->b_dirtyoff; 889 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE 890 + bp->b_dirtyoff; 891 io.iov_base = (char *)bp->b_data + bp->b_dirtyoff; 892 uiop->uio_rw = UIO_WRITE; 893 nfsstats.write_bios++; 894 if ((bp->b_flags & (B_ASYNC | B_NEEDCOMMIT | B_NOCACHE)) == B_ASYNC) 895 iomode = NFSV3WRITE_UNSTABLE; 896 else 897 iomode = NFSV3WRITE_FILESYNC; 898 bp->b_flags |= B_WRITEINPROG; 899 error = nfs_writerpc(vp, uiop, cr, &iomode, &must_commit); 900 if (!error && iomode == NFSV3WRITE_UNSTABLE) 901 bp->b_flags |= B_NEEDCOMMIT; 902 else 903 bp->b_flags &= ~B_NEEDCOMMIT; 904 bp->b_flags &= ~B_WRITEINPROG; 905 906 /* 907 * For an interrupted write, the buffer is still valid 908 * and the write hasn't been pushed to the server yet, 909 * so we can't set B_ERROR and report the interruption 910 * by setting B_EINTR. For the B_ASYNC case, B_EINTR 911 * is not relevant, so the rpc attempt is essentially 912 * a noop. For the case of a V3 write rpc not being 913 * committed to stable storage, the block is still 914 * dirty and requires either a commit rpc or another 915 * write rpc with iomode == NFSV3WRITE_FILESYNC before 916 * the block is reused. This is indicated by setting 917 * the B_DELWRI and B_NEEDCOMMIT flags. 918 */ 919 if (error == EINTR 920 || (!error && (bp->b_flags & B_NEEDCOMMIT))) { 921 bp->b_flags &= ~(B_INVAL|B_NOCACHE); 922 bp->b_flags |= B_DELWRI; 923 924 /* 925 * Since for the B_ASYNC case, nfs_bwrite() has reassigned the 926 * buffer to the clean list, we have to reassign it back to the 927 * dirty one. Ugh. 928 */ 929 if (bp->b_flags & B_ASYNC) 930 reassignbuf(bp, vp); 931 else 932 bp->b_flags |= B_EINTR; 933 } else { 934 if (error) { 935 bp->b_flags |= B_ERROR; 936 bp->b_error = np->n_error = error; 937 np->n_flag |= NWRITEERR; 938 } 939 bp->b_dirtyoff = bp->b_dirtyend = 0; 940 } 941 } else { 942 bp->b_resid = 0; 943 biodone(bp); 944 return (0); 945 } 946 } 947 bp->b_resid = uiop->uio_resid; 948 if (must_commit) 949 nfs_clearcommit(vp->v_mount); 950 biodone(bp); 951 return (error); 952} 953