nfs_bio.c revision 18866
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 1.25 1996/09/19 18:20:54 nate 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#include <sys/sysctl.h> 50 51#include <vm/vm.h> 52#include <vm/vm_param.h> 53#include <vm/vm_extern.h> 54 55#include <nfs/rpcv2.h> 56#include <nfs/nfsproto.h> 57#include <nfs/nfs.h> 58#include <nfs/nfsmount.h> 59#include <nfs/nqnfs.h> 60#include <nfs/nfsnode.h> 61 62static struct buf *nfs_getcacheblk __P((struct vnode *vp, daddr_t bn, int size, 63 struct proc *p)); 64 65extern struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON]; 66extern int nfs_numasync; 67extern struct nfsstats nfsstats; 68 69int nfs_dwrite = 1; 70SYSCTL_INT(_vfs_nfs, OID_AUTO, dwrite, CTLFLAG_RW, &nfs_dwrite, 0, ""); 71 72/* 73 * Ifdefs for FreeBSD-current's merged VM/buffer cache. It is unfortunate 74 * that this isn't done inside getblk() and brelse() so these calls 75 * wouldn't need to be here. 76 */ 77#ifdef B_VMIO 78#define vnode_pager_uncache(vp) 79#else 80#define vfs_busy_pages(bp, f) 81#define vfs_unbusy_pages(bp) 82#define vfs_dirty_pages(bp) 83#endif 84 85/* 86 * Vnode op for read using bio 87 * Any similarity to readip() is purely coincidental 88 */ 89int 90nfs_bioread(vp, uio, ioflag, cred) 91 register struct vnode *vp; 92 register struct uio *uio; 93 int ioflag; 94 struct ucred *cred; 95{ 96 register struct nfsnode *np = VTONFS(vp); 97 register int biosize, diff, i; 98 struct buf *bp = 0, *rabp; 99 struct vattr vattr; 100 struct proc *p; 101 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 102 daddr_t lbn, rabn; 103 int bufsize; 104 int nra, error = 0, n = 0, on = 0, not_readin; 105 106#ifdef DIAGNOSTIC 107 if (uio->uio_rw != UIO_READ) 108 panic("nfs_read mode"); 109#endif 110 if (uio->uio_resid == 0) 111 return (0); 112 if (uio->uio_offset < 0) 113 return (EINVAL); 114 p = uio->uio_procp; 115 if ((nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_GOTFSINFO)) == NFSMNT_NFSV3) 116 (void)nfs_fsinfo(nmp, vp, cred, p); 117 biosize = vp->v_mount->mnt_stat.f_iosize; 118 /* 119 * For nfs, cache consistency can only be maintained approximately. 120 * Although RFC1094 does not specify the criteria, the following is 121 * believed to be compatible with the reference port. 122 * For nqnfs, full cache consistency is maintained within the loop. 123 * For nfs: 124 * If the file's modify time on the server has changed since the 125 * last read rpc or you have written to the file, 126 * you may have lost data cache consistency with the 127 * server, so flush all of the file's data out of the cache. 128 * Then force a getattr rpc to ensure that you have up to date 129 * attributes. 130 * NB: This implies that cache data can be read when up to 131 * NFS_ATTRTIMEO seconds out of date. If you find that you need current 132 * attributes this could be forced by setting n_attrstamp to 0 before 133 * the VOP_GETATTR() call. 134 */ 135 if ((nmp->nm_flag & NFSMNT_NQNFS) == 0) { 136 if (np->n_flag & NMODIFIED) { 137 if (vp->v_type != VREG) { 138 if (vp->v_type != VDIR) 139 panic("nfs: bioread, not dir"); 140 nfs_invaldir(vp); 141 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 142 if (error) 143 return (error); 144 } 145 np->n_attrstamp = 0; 146 error = VOP_GETATTR(vp, &vattr, cred, p); 147 if (error) 148 return (error); 149 np->n_mtime = vattr.va_mtime.tv_sec; 150 } else { 151 error = VOP_GETATTR(vp, &vattr, cred, p); 152 if (error) 153 return (error); 154 if (np->n_mtime != vattr.va_mtime.tv_sec) { 155 if (vp->v_type == VDIR) 156 nfs_invaldir(vp); 157 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 158 if (error) 159 return (error); 160 np->n_mtime = vattr.va_mtime.tv_sec; 161 } 162 } 163 } 164 do { 165 166 /* 167 * Get a valid lease. If cached data is stale, flush it. 168 */ 169 if (nmp->nm_flag & NFSMNT_NQNFS) { 170 if (NQNFS_CKINVALID(vp, np, ND_READ)) { 171 do { 172 error = nqnfs_getlease(vp, ND_READ, cred, p); 173 } while (error == NQNFS_EXPIRED); 174 if (error) 175 return (error); 176 if (np->n_lrev != np->n_brev || 177 (np->n_flag & NQNFSNONCACHE) || 178 ((np->n_flag & NMODIFIED) && vp->v_type == VDIR)) { 179 if (vp->v_type == VDIR) 180 nfs_invaldir(vp); 181 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 182 if (error) 183 return (error); 184 np->n_brev = np->n_lrev; 185 } 186 } else if (vp->v_type == VDIR && (np->n_flag & NMODIFIED)) { 187 nfs_invaldir(vp); 188 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 189 if (error) 190 return (error); 191 } 192 } 193 if (np->n_flag & NQNFSNONCACHE) { 194 switch (vp->v_type) { 195 case VREG: 196 return (nfs_readrpc(vp, uio, cred)); 197 case VLNK: 198 return (nfs_readlinkrpc(vp, uio, cred)); 199 case VDIR: 200 break; 201 default: 202 printf(" NQNFSNONCACHE: type %x unexpected\n", 203 vp->v_type); 204 }; 205 } 206 switch (vp->v_type) { 207 case VREG: 208 nfsstats.biocache_reads++; 209 lbn = uio->uio_offset / biosize; 210 on = uio->uio_offset & (biosize - 1); 211 not_readin = 1; 212 213 /* 214 * Start the read ahead(s), as required. 215 */ 216 if (nfs_numasync > 0 && nmp->nm_readahead > 0) { 217 for (nra = 0; nra < nmp->nm_readahead && 218 (off_t)(lbn + 1 + nra) * biosize < np->n_size; nra++) { 219 rabn = lbn + 1 + nra; 220 if (!incore(vp, rabn)) { 221 rabp = nfs_getcacheblk(vp, rabn, biosize, p); 222 if (!rabp) 223 return (EINTR); 224 if ((rabp->b_flags & (B_CACHE|B_DELWRI)) == 0) { 225 rabp->b_flags |= (B_READ | B_ASYNC); 226 vfs_busy_pages(rabp, 0); 227 if (nfs_asyncio(rabp, cred)) { 228 rabp->b_flags |= B_INVAL|B_ERROR; 229 vfs_unbusy_pages(rabp); 230 brelse(rabp); 231 } 232 } else { 233 brelse(rabp); 234 } 235 } 236 } 237 } 238 239 /* 240 * If the block is in the cache and has the required data 241 * in a valid region, just copy it out. 242 * Otherwise, get the block and write back/read in, 243 * as required. 244 */ 245again: 246 bufsize = biosize; 247 if ((off_t)(lbn + 1) * biosize > np->n_size && 248 (off_t)(lbn + 1) * biosize - np->n_size < biosize) { 249 bufsize = np->n_size - lbn * biosize; 250 bufsize = (bufsize + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1); 251 } 252 bp = nfs_getcacheblk(vp, lbn, bufsize, p); 253 if (!bp) 254 return (EINTR); 255 if ((bp->b_flags & B_CACHE) == 0) { 256 bp->b_flags |= B_READ; 257 bp->b_flags &= ~(B_DONE | B_ERROR | B_INVAL); 258 not_readin = 0; 259 vfs_busy_pages(bp, 0); 260 error = nfs_doio(bp, cred, p); 261 if (error) { 262 brelse(bp); 263 return (error); 264 } 265 } 266 if (bufsize > on) { 267 n = min((unsigned)(bufsize - on), uio->uio_resid); 268 } else { 269 n = 0; 270 } 271 diff = np->n_size - uio->uio_offset; 272 if (diff < n) 273 n = diff; 274 if (not_readin && n > 0) { 275 if (on < bp->b_validoff || (on + n) > bp->b_validend) { 276 bp->b_flags |= B_NOCACHE; 277 if (bp->b_dirtyend > 0) { 278 if ((bp->b_flags & B_DELWRI) == 0) 279 panic("nfsbioread"); 280 if (VOP_BWRITE(bp) == EINTR) 281 return (EINTR); 282 } else 283 brelse(bp); 284 goto again; 285 } 286 } 287 vp->v_lastr = lbn; 288 diff = (on >= bp->b_validend) ? 0 : (bp->b_validend - on); 289 if (diff < n) 290 n = diff; 291 break; 292 case VLNK: 293 nfsstats.biocache_readlinks++; 294 bp = nfs_getcacheblk(vp, (daddr_t)0, NFS_MAXPATHLEN, p); 295 if (!bp) 296 return (EINTR); 297 if ((bp->b_flags & B_CACHE) == 0) { 298 bp->b_flags |= B_READ; 299 vfs_busy_pages(bp, 0); 300 error = nfs_doio(bp, cred, p); 301 if (error) { 302 bp->b_flags |= B_ERROR; 303 brelse(bp); 304 return (error); 305 } 306 } 307 n = min(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid); 308 on = 0; 309 break; 310 case VDIR: 311 nfsstats.biocache_readdirs++; 312 lbn = uio->uio_offset / NFS_DIRBLKSIZ; 313 on = uio->uio_offset & (NFS_DIRBLKSIZ - 1); 314 bp = nfs_getcacheblk(vp, lbn, NFS_DIRBLKSIZ, p); 315 if (!bp) 316 return (EINTR); 317 if ((bp->b_flags & B_CACHE) == 0) { 318 bp->b_flags |= B_READ; 319 vfs_busy_pages(bp, 0); 320 error = nfs_doio(bp, cred, p); 321 if (error) { 322 brelse(bp); 323 while (error == NFSERR_BAD_COOKIE) { 324 nfs_invaldir(vp); 325 error = nfs_vinvalbuf(vp, 0, cred, p, 1); 326 /* 327 * Yuck! The directory has been modified on the 328 * server. The only way to get the block is by 329 * reading from the beginning to get all the 330 * offset cookies. 331 */ 332 for (i = 0; i <= lbn && !error; i++) { 333 bp = nfs_getcacheblk(vp, i, NFS_DIRBLKSIZ, p); 334 if (!bp) 335 return (EINTR); 336 if ((bp->b_flags & B_DONE) == 0) { 337 bp->b_flags |= B_READ; 338 vfs_busy_pages(bp, 0); 339 error = nfs_doio(bp, cred, p); 340 if (error) 341 brelse(bp); 342 } 343 } 344 } 345 if (error) 346 return (error); 347 } 348 } 349 350 /* 351 * If not eof and read aheads are enabled, start one. 352 * (You need the current block first, so that you have the 353 * directory offset cookie of the next block.) 354 */ 355 if (nfs_numasync > 0 && nmp->nm_readahead > 0 && 356 (np->n_direofoffset == 0 || 357 (lbn + 1) * NFS_DIRBLKSIZ < np->n_direofoffset) && 358 !(np->n_flag & NQNFSNONCACHE) && 359 !incore(vp, lbn + 1)) { 360 rabp = nfs_getcacheblk(vp, lbn + 1, NFS_DIRBLKSIZ, p); 361 if (rabp) { 362 if ((rabp->b_flags & (B_CACHE|B_DELWRI)) == 0) { 363 rabp->b_flags |= (B_READ | B_ASYNC); 364 vfs_busy_pages(rabp, 0); 365 if (nfs_asyncio(rabp, cred)) { 366 rabp->b_flags |= B_INVAL|B_ERROR; 367 vfs_unbusy_pages(rabp); 368 brelse(rabp); 369 } 370 } else { 371 brelse(rabp); 372 } 373 } 374 } 375 n = min(uio->uio_resid, NFS_DIRBLKSIZ - bp->b_resid - on); 376 break; 377 default: 378 printf(" nfs_bioread: type %x unexpected\n",vp->v_type); 379 break; 380 }; 381 382 if (n > 0) { 383 error = uiomove(bp->b_data + on, (int)n, uio); 384 } 385 switch (vp->v_type) { 386 case VREG: 387 break; 388 case VLNK: 389 n = 0; 390 break; 391 case VDIR: 392 if (np->n_flag & NQNFSNONCACHE) 393 bp->b_flags |= B_INVAL; 394 break; 395 default: 396 printf(" nfs_bioread: type %x unexpected\n",vp->v_type); 397 } 398 brelse(bp); 399 } while (error == 0 && uio->uio_resid > 0 && n > 0); 400 return (error); 401} 402 403/* 404 * Vnode op for write using bio 405 */ 406int 407nfs_write(ap) 408 struct vop_write_args /* { 409 struct vnode *a_vp; 410 struct uio *a_uio; 411 int a_ioflag; 412 struct ucred *a_cred; 413 } */ *ap; 414{ 415 register int biosize; 416 register struct uio *uio = ap->a_uio; 417 struct proc *p = uio->uio_procp; 418 register struct vnode *vp = ap->a_vp; 419 struct nfsnode *np = VTONFS(vp); 420 register struct ucred *cred = ap->a_cred; 421 int ioflag = ap->a_ioflag; 422 struct buf *bp; 423 struct vattr vattr; 424 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 425 daddr_t lbn; 426 int bufsize; 427 int n, on, error = 0, iomode, must_commit; 428 429#ifdef DIAGNOSTIC 430 if (uio->uio_rw != UIO_WRITE) 431 panic("nfs_write mode"); 432 if (uio->uio_segflg == UIO_USERSPACE && uio->uio_procp != curproc) 433 panic("nfs_write proc"); 434#endif 435 if (vp->v_type != VREG) 436 return (EIO); 437 if (np->n_flag & NWRITEERR) { 438 np->n_flag &= ~NWRITEERR; 439 return (np->n_error); 440 } 441 if ((nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_GOTFSINFO)) == NFSMNT_NFSV3) 442 (void)nfs_fsinfo(nmp, vp, cred, p); 443 if (ioflag & (IO_APPEND | IO_SYNC)) { 444 if (np->n_flag & NMODIFIED) { 445 np->n_attrstamp = 0; 446 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 447 if (error) 448 return (error); 449 } 450 if (ioflag & IO_APPEND) { 451 np->n_attrstamp = 0; 452 error = VOP_GETATTR(vp, &vattr, cred, p); 453 if (error) 454 return (error); 455 uio->uio_offset = np->n_size; 456 } 457 } 458 if (uio->uio_offset < 0) 459 return (EINVAL); 460 if (uio->uio_resid == 0) 461 return (0); 462 /* 463 * Maybe this should be above the vnode op call, but so long as 464 * file servers have no limits, i don't think it matters 465 */ 466 if (p && uio->uio_offset + uio->uio_resid > 467 p->p_rlimit[RLIMIT_FSIZE].rlim_cur) { 468 psignal(p, SIGXFSZ); 469 return (EFBIG); 470 } 471 /* 472 * I use nm_rsize, not nm_wsize so that all buffer cache blocks 473 * will be the same size within a filesystem. nfs_writerpc will 474 * still use nm_wsize when sizing the rpc's. 475 */ 476 biosize = vp->v_mount->mnt_stat.f_iosize; 477 do { 478 479 /* 480 * XXX make sure we aren't cached in the VM page cache 481 */ 482 /* 483 * Check for a valid write lease. 484 */ 485 if ((nmp->nm_flag & NFSMNT_NQNFS) && 486 NQNFS_CKINVALID(vp, np, ND_WRITE)) { 487 do { 488 error = nqnfs_getlease(vp, ND_WRITE, cred, p); 489 } while (error == NQNFS_EXPIRED); 490 if (error) 491 return (error); 492 if (np->n_lrev != np->n_brev || 493 (np->n_flag & NQNFSNONCACHE)) { 494 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 495 if (error) 496 return (error); 497 np->n_brev = np->n_lrev; 498 } 499 } 500 if ((np->n_flag & NQNFSNONCACHE) && uio->uio_iovcnt == 1) { 501 iomode = NFSV3WRITE_FILESYNC; 502 error = nfs_writerpc(vp, uio, cred, &iomode, &must_commit); 503 if (must_commit) 504 nfs_clearcommit(vp->v_mount); 505 return (error); 506 } 507 nfsstats.biocache_writes++; 508 lbn = uio->uio_offset / biosize; 509 on = uio->uio_offset & (biosize-1); 510 n = min((unsigned)(biosize - on), uio->uio_resid); 511again: 512 if (uio->uio_offset + n > np->n_size) { 513 np->n_size = uio->uio_offset + n; 514 vnode_pager_setsize(vp, (u_long)np->n_size); 515 } 516 bufsize = biosize; 517 if ((lbn + 1) * biosize > np->n_size) { 518 bufsize = np->n_size - lbn * biosize; 519 bufsize = (bufsize + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1); 520 } 521 bp = nfs_getcacheblk(vp, lbn, bufsize, p); 522 if (!bp) 523 return (EINTR); 524 if (bp->b_wcred == NOCRED) { 525 crhold(cred); 526 bp->b_wcred = cred; 527 } 528 np->n_flag |= NMODIFIED; 529 530 if ((bp->b_blkno * DEV_BSIZE) + bp->b_dirtyend > np->n_size) { 531 bp->b_dirtyend = np->n_size - (bp->b_blkno * DEV_BSIZE); 532 } 533 534 /* 535 * If the new write will leave a contiguous dirty 536 * area, just update the b_dirtyoff and b_dirtyend, 537 * otherwise force a write rpc of the old dirty area. 538 */ 539 if (bp->b_dirtyend > 0 && 540 (on > bp->b_dirtyend || (on + n) < bp->b_dirtyoff)) { 541 bp->b_proc = p; 542 if (VOP_BWRITE(bp) == EINTR) 543 return (EINTR); 544 goto again; 545 } 546 547 /* 548 * Check for valid write lease and get one as required. 549 * In case getblk() and/or bwrite() delayed us. 550 */ 551 if ((nmp->nm_flag & NFSMNT_NQNFS) && 552 NQNFS_CKINVALID(vp, np, ND_WRITE)) { 553 do { 554 error = nqnfs_getlease(vp, ND_WRITE, cred, p); 555 } while (error == NQNFS_EXPIRED); 556 if (error) { 557 brelse(bp); 558 return (error); 559 } 560 if (np->n_lrev != np->n_brev || 561 (np->n_flag & NQNFSNONCACHE)) { 562 brelse(bp); 563 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 564 if (error) 565 return (error); 566 np->n_brev = np->n_lrev; 567 goto again; 568 } 569 } 570 error = uiomove((char *)bp->b_data + on, n, uio); 571 if (error) { 572 bp->b_flags |= B_ERROR; 573 brelse(bp); 574 return (error); 575 } 576 if (bp->b_dirtyend > 0) { 577 bp->b_dirtyoff = min(on, bp->b_dirtyoff); 578 bp->b_dirtyend = max((on + n), bp->b_dirtyend); 579 } else { 580 bp->b_dirtyoff = on; 581 bp->b_dirtyend = on + n; 582 } 583 if (bp->b_validend == 0 || bp->b_validend < bp->b_dirtyoff || 584 bp->b_validoff > bp->b_dirtyend) { 585 bp->b_validoff = bp->b_dirtyoff; 586 bp->b_validend = bp->b_dirtyend; 587 } else { 588 bp->b_validoff = min(bp->b_validoff, bp->b_dirtyoff); 589 bp->b_validend = max(bp->b_validend, bp->b_dirtyend); 590 } 591 592 /* 593 * Since this block is being modified, it must be written 594 * again and not just committed. 595 */ 596 bp->b_flags &= ~B_NEEDCOMMIT; 597 598 /* 599 * If the lease is non-cachable or IO_SYNC do bwrite(). 600 */ 601 if ((np->n_flag & NQNFSNONCACHE) || (ioflag & IO_SYNC)) { 602 bp->b_proc = p; 603 error = VOP_BWRITE(bp); 604 if (error) 605 return (error); 606 if (np->n_flag & NQNFSNONCACHE) { 607 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 608 if (error) 609 return (error); 610 } 611 } else if ((n + on) == biosize && 612 (nmp->nm_flag & NFSMNT_NQNFS) == 0) { 613 bp->b_proc = (struct proc *)0; 614 bp->b_flags |= B_ASYNC; 615 (void)nfs_writebp(bp, 0); 616 } else 617 bdwrite(bp); 618 } while (uio->uio_resid > 0 && n > 0); 619 return (0); 620} 621 622/* 623 * Get an nfs cache block. 624 * Allocate a new one if the block isn't currently in the cache 625 * and return the block marked busy. If the calling process is 626 * interrupted by a signal for an interruptible mount point, return 627 * NULL. 628 */ 629static struct buf * 630nfs_getcacheblk(vp, bn, size, p) 631 struct vnode *vp; 632 daddr_t bn; 633 int size; 634 struct proc *p; 635{ 636 register struct buf *bp; 637 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 638 int biosize = vp->v_mount->mnt_stat.f_iosize; 639 640 if (nmp->nm_flag & NFSMNT_INT) { 641 bp = getblk(vp, bn, size, PCATCH, 0); 642 while (bp == (struct buf *)0) { 643 if (nfs_sigintr(nmp, (struct nfsreq *)0, p)) 644 return ((struct buf *)0); 645 bp = getblk(vp, bn, size, 0, 2 * hz); 646 } 647 } else 648 bp = getblk(vp, bn, size, 0, 0); 649 650 if( vp->v_type == VREG) 651 bp->b_blkno = (bn * biosize) / DEV_BSIZE; 652 653 return (bp); 654} 655 656/* 657 * Flush and invalidate all dirty buffers. If another process is already 658 * doing the flush, just wait for completion. 659 */ 660int 661nfs_vinvalbuf(vp, flags, cred, p, intrflg) 662 struct vnode *vp; 663 int flags; 664 struct ucred *cred; 665 struct proc *p; 666 int intrflg; 667{ 668 register struct nfsnode *np = VTONFS(vp); 669 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 670 int error = 0, slpflag, slptimeo; 671 672 if ((nmp->nm_flag & NFSMNT_INT) == 0) 673 intrflg = 0; 674 if (intrflg) { 675 slpflag = PCATCH; 676 slptimeo = 2 * hz; 677 } else { 678 slpflag = 0; 679 slptimeo = 0; 680 } 681 /* 682 * First wait for any other process doing a flush to complete. 683 */ 684 while (np->n_flag & NFLUSHINPROG) { 685 np->n_flag |= NFLUSHWANT; 686 error = tsleep((caddr_t)&np->n_flag, PRIBIO + 2, "nfsvinval", 687 slptimeo); 688 if (error && intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) 689 return (EINTR); 690 } 691 692 /* 693 * Now, flush as required. 694 */ 695 np->n_flag |= NFLUSHINPROG; 696 error = vinvalbuf(vp, flags, cred, p, slpflag, 0); 697 while (error) { 698 if (intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) { 699 np->n_flag &= ~NFLUSHINPROG; 700 if (np->n_flag & NFLUSHWANT) { 701 np->n_flag &= ~NFLUSHWANT; 702 wakeup((caddr_t)&np->n_flag); 703 } 704 return (EINTR); 705 } 706 error = vinvalbuf(vp, flags, cred, p, 0, slptimeo); 707 } 708 np->n_flag &= ~(NMODIFIED | NFLUSHINPROG); 709 if (np->n_flag & NFLUSHWANT) { 710 np->n_flag &= ~NFLUSHWANT; 711 wakeup((caddr_t)&np->n_flag); 712 } 713 return (0); 714} 715 716/* 717 * Initiate asynchronous I/O. Return an error if no nfsiods are available. 718 * This is mainly to avoid queueing async I/O requests when the nfsiods 719 * are all hung on a dead server. 720 */ 721int 722nfs_asyncio(bp, cred) 723 register struct buf *bp; 724 struct ucred *cred; 725{ 726 register int i; 727 728 if (nfs_numasync == 0) 729 return (EIO); 730 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 731 if (nfs_iodwant[i]) { 732 if (bp->b_flags & B_READ) { 733 if (bp->b_rcred == NOCRED && cred != NOCRED) { 734 crhold(cred); 735 bp->b_rcred = cred; 736 } 737 } else { 738 bp->b_flags |= B_WRITEINPROG; 739 if (bp->b_wcred == NOCRED && cred != NOCRED) { 740 crhold(cred); 741 bp->b_wcred = cred; 742 } 743 } 744 745 TAILQ_INSERT_TAIL(&nfs_bufq, bp, b_freelist); 746 nfs_iodwant[i] = (struct proc *)0; 747 wakeup((caddr_t)&nfs_iodwant[i]); 748 return (0); 749 } 750 751 /* 752 * If it is a read or a write already marked B_WRITEINPROG or B_NOCACHE 753 * return EIO so the process will call nfs_doio() and do it 754 * synchronously. 755 */ 756 if (bp->b_flags & (B_READ | B_WRITEINPROG | B_NOCACHE)) 757 return (EIO); 758 759 /* 760 * Allow the administrator to override the choice of using a delayed 761 * write since it is a pessimization for some servers, notably some 762 * Solaris servers. 763 */ 764 if (!nfs_dwrite) 765 return (EIO); 766 767 /* 768 * Just turn the async write into a delayed write, instead of 769 * doing in synchronously. Hopefully, at least one of the nfsiods 770 * is currently doing a write for this file and will pick up the 771 * delayed writes before going back to sleep. 772 */ 773 bp->b_flags |= B_DELWRI; 774 reassignbuf(bp, bp->b_vp); 775 biodone(bp); 776 return (0); 777} 778 779/* 780 * Do an I/O operation to/from a cache block. This may be called 781 * synchronously or from an nfsiod. 782 */ 783int 784nfs_doio(bp, cr, p) 785 register struct buf *bp; 786 struct ucred *cr; 787 struct proc *p; 788{ 789 register struct uio *uiop; 790 register struct vnode *vp; 791 struct nfsnode *np; 792 struct nfsmount *nmp; 793 int error = 0, diff, len, iomode, must_commit = 0; 794 struct uio uio; 795 struct iovec io; 796 797 vp = bp->b_vp; 798 np = VTONFS(vp); 799 nmp = VFSTONFS(vp->v_mount); 800 uiop = &uio; 801 uiop->uio_iov = &io; 802 uiop->uio_iovcnt = 1; 803 uiop->uio_segflg = UIO_SYSSPACE; 804 uiop->uio_procp = p; 805 806 /* 807 * Historically, paging was done with physio, but no more. 808 */ 809 if (bp->b_flags & B_PHYS) { 810 /* 811 * ...though reading /dev/drum still gets us here. 812 */ 813 io.iov_len = uiop->uio_resid = bp->b_bcount; 814 /* mapping was done by vmapbuf() */ 815 io.iov_base = bp->b_data; 816 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; 817 if (bp->b_flags & B_READ) { 818 uiop->uio_rw = UIO_READ; 819 nfsstats.read_physios++; 820 error = nfs_readrpc(vp, uiop, cr); 821 } else { 822 int com; 823 824 iomode = NFSV3WRITE_DATASYNC; 825 uiop->uio_rw = UIO_WRITE; 826 nfsstats.write_physios++; 827 error = nfs_writerpc(vp, uiop, cr, &iomode, &com); 828 } 829 if (error) { 830 bp->b_flags |= B_ERROR; 831 bp->b_error = error; 832 } 833 } else if (bp->b_flags & B_READ) { 834 io.iov_len = uiop->uio_resid = bp->b_bcount; 835 io.iov_base = bp->b_data; 836 uiop->uio_rw = UIO_READ; 837 switch (vp->v_type) { 838 case VREG: 839 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; 840 nfsstats.read_bios++; 841 error = nfs_readrpc(vp, uiop, cr); 842 if (!error) { 843 bp->b_validoff = 0; 844 if (uiop->uio_resid) { 845 /* 846 * If len > 0, there is a hole in the file and 847 * no writes after the hole have been pushed to 848 * the server yet. 849 * Just zero fill the rest of the valid area. 850 */ 851 diff = bp->b_bcount - uiop->uio_resid; 852 len = np->n_size - (((u_quad_t)bp->b_blkno) * DEV_BSIZE 853 + diff); 854 if (len > 0) { 855 len = min(len, uiop->uio_resid); 856 bzero((char *)bp->b_data + diff, len); 857 bp->b_validend = diff + len; 858 } else 859 bp->b_validend = diff; 860 } else 861 bp->b_validend = bp->b_bcount; 862 } 863 if (p && (vp->v_flag & VTEXT) && 864 (((nmp->nm_flag & NFSMNT_NQNFS) && 865 NQNFS_CKINVALID(vp, np, ND_READ) && 866 np->n_lrev != np->n_brev) || 867 (!(nmp->nm_flag & NFSMNT_NQNFS) && 868 np->n_mtime != np->n_vattr.va_mtime.tv_sec))) { 869 uprintf("Process killed due to text file modification\n"); 870 psignal(p, SIGKILL); 871#ifdef __NetBSD__ 872 p->p_holdcnt++; 873#else 874 p->p_flag |= P_NOSWAP; 875#endif 876 } 877 break; 878 case VLNK: 879 uiop->uio_offset = (off_t)0; 880 nfsstats.readlink_bios++; 881 error = nfs_readlinkrpc(vp, uiop, cr); 882 break; 883 case VDIR: 884 nfsstats.readdir_bios++; 885 uiop->uio_offset = ((u_quad_t)bp->b_lblkno) * NFS_DIRBLKSIZ; 886 if (nmp->nm_flag & NFSMNT_RDIRPLUS) { 887 error = nfs_readdirplusrpc(vp, uiop, cr); 888 if (error == NFSERR_NOTSUPP) 889 nmp->nm_flag &= ~NFSMNT_RDIRPLUS; 890 } 891 if ((nmp->nm_flag & NFSMNT_RDIRPLUS) == 0) 892 error = nfs_readdirrpc(vp, uiop, cr); 893 break; 894 default: 895 printf("nfs_doio: type %x unexpected\n",vp->v_type); 896 break; 897 }; 898 if (error) { 899 bp->b_flags |= B_ERROR; 900 bp->b_error = error; 901 } 902 } else { 903 if (((bp->b_blkno * DEV_BSIZE) + bp->b_dirtyend) > np->n_size) 904 bp->b_dirtyend = np->n_size - (bp->b_blkno * DEV_BSIZE); 905 906 if (bp->b_dirtyend > bp->b_dirtyoff) { 907 io.iov_len = uiop->uio_resid = bp->b_dirtyend 908 - bp->b_dirtyoff; 909 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE 910 + bp->b_dirtyoff; 911 io.iov_base = (char *)bp->b_data + bp->b_dirtyoff; 912 uiop->uio_rw = UIO_WRITE; 913 nfsstats.write_bios++; 914 if ((bp->b_flags & (B_ASYNC | B_NEEDCOMMIT | B_NOCACHE)) == B_ASYNC) 915 iomode = NFSV3WRITE_UNSTABLE; 916 else 917 iomode = NFSV3WRITE_FILESYNC; 918 bp->b_flags |= B_WRITEINPROG; 919 error = nfs_writerpc(vp, uiop, cr, &iomode, &must_commit); 920 if (!error && iomode == NFSV3WRITE_UNSTABLE) 921 bp->b_flags |= B_NEEDCOMMIT; 922 else 923 bp->b_flags &= ~B_NEEDCOMMIT; 924 bp->b_flags &= ~B_WRITEINPROG; 925 926 /* 927 * For an interrupted write, the buffer is still valid 928 * and the write hasn't been pushed to the server yet, 929 * so we can't set B_ERROR and report the interruption 930 * by setting B_EINTR. For the B_ASYNC case, B_EINTR 931 * is not relevant, so the rpc attempt is essentially 932 * a noop. For the case of a V3 write rpc not being 933 * committed to stable storage, the block is still 934 * dirty and requires either a commit rpc or another 935 * write rpc with iomode == NFSV3WRITE_FILESYNC before 936 * the block is reused. This is indicated by setting 937 * the B_DELWRI and B_NEEDCOMMIT flags. 938 */ 939 if (error == EINTR 940 || (!error && (bp->b_flags & B_NEEDCOMMIT))) { 941 bp->b_flags &= ~(B_INVAL|B_NOCACHE); 942 bp->b_flags |= B_DELWRI; 943 944 /* 945 * Since for the B_ASYNC case, nfs_bwrite() has reassigned the 946 * buffer to the clean list, we have to reassign it back to the 947 * dirty one. Ugh. 948 */ 949 if (bp->b_flags & B_ASYNC) 950 reassignbuf(bp, vp); 951 else 952 bp->b_flags |= B_EINTR; 953 } else { 954 if (error) { 955 bp->b_flags |= B_ERROR; 956 bp->b_error = np->n_error = error; 957 np->n_flag |= NWRITEERR; 958 } 959 bp->b_dirtyoff = bp->b_dirtyend = 0; 960 } 961 } else { 962 bp->b_resid = 0; 963 biodone(bp); 964 return (0); 965 } 966 } 967 bp->b_resid = uiop->uio_resid; 968 if (must_commit) 969 nfs_clearcommit(vp->v_mount); 970 biodone(bp); 971 return (error); 972} 973