nfs_bio.c revision 22521
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 8.9 (Berkeley) 3/30/95 37 * $FreeBSD: head/sys/nfsclient/nfs_bio.c 22521 1997-02-10 02:22:35Z dyson $ 38 */ 39 40 41#include <sys/param.h> 42#include <sys/systm.h> 43#include <sys/resourcevar.h> 44#include <sys/signalvar.h> 45#include <sys/proc.h> 46#include <sys/buf.h> 47#include <sys/vnode.h> 48#include <sys/mount.h> 49#include <sys/kernel.h> 50#include <sys/sysctl.h> 51 52#include <vm/vm.h> 53#include <vm/vm_param.h> 54#include <vm/vm_extern.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 63static struct buf *nfs_getcacheblk __P((struct vnode *vp, daddr_t bn, int size, 64 struct proc *p)); 65 66extern int nfs_numasync; 67extern struct nfsstats nfsstats; 68 69/* 70 * Vnode op for read using bio 71 * Any similarity to readip() is purely coincidental 72 */ 73int 74nfs_bioread(vp, uio, ioflag, cred) 75 register struct vnode *vp; 76 register struct uio *uio; 77 int ioflag; 78 struct ucred *cred; 79{ 80 register struct nfsnode *np = VTONFS(vp); 81 register int biosize, diff, i; 82 struct buf *bp = 0, *rabp; 83 struct vattr vattr; 84 struct proc *p; 85 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 86 daddr_t lbn, rabn; 87 int bufsize; 88 int nra, error = 0, n = 0, on = 0, not_readin; 89 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) 97 return (EINVAL); 98 p = uio->uio_procp; 99 if ((nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_GOTFSINFO)) == NFSMNT_NFSV3) 100 (void)nfs_fsinfo(nmp, vp, cred, p); 101 biosize = vp->v_mount->mnt_stat.f_iosize; 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) { 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 if (np->n_flag & NQNFSNONCACHE) { 178 switch (vp->v_type) { 179 case VREG: 180 return (nfs_readrpc(vp, uio, cred)); 181 case VLNK: 182 return (nfs_readlinkrpc(vp, uio, cred)); 183 case VDIR: 184 break; 185 default: 186 printf(" NQNFSNONCACHE: type %x unexpected\n", 187 vp->v_type); 188 }; 189 } 190 switch (vp->v_type) { 191 case VREG: 192 nfsstats.biocache_reads++; 193 lbn = uio->uio_offset / biosize; 194 on = uio->uio_offset & (biosize - 1); 195 not_readin = 1; 196 197 /* 198 * Start the read ahead(s), as required. 199 */ 200 if (nfs_numasync > 0 && nmp->nm_readahead > 0) { 201 for (nra = 0; nra < nmp->nm_readahead && 202 (off_t)(lbn + 1 + nra) * biosize < np->n_size; nra++) { 203 rabn = lbn + 1 + nra; 204 if (!incore(vp, rabn)) { 205 rabp = nfs_getcacheblk(vp, rabn, biosize, p); 206 if (!rabp) 207 return (EINTR); 208 if ((rabp->b_flags & (B_CACHE|B_DELWRI)) == 0) { 209 rabp->b_flags |= (B_READ | B_ASYNC); 210 vfs_busy_pages(rabp, 0); 211 if (nfs_asyncio(rabp, cred)) { 212 rabp->b_flags |= B_INVAL|B_ERROR; 213 vfs_unbusy_pages(rabp); 214 brelse(rabp); 215 } 216 } else 217 brelse(rabp); 218 } 219 } 220 } 221 222 /* 223 * If the block is in the cache and has the required data 224 * in a valid region, just copy it out. 225 * Otherwise, get the block and write back/read in, 226 * as required. 227 */ 228again: 229 bufsize = biosize; 230 if ((off_t)(lbn + 1) * biosize > np->n_size && 231 (off_t)(lbn + 1) * biosize - np->n_size < biosize) { 232 bufsize = np->n_size - lbn * biosize; 233 bufsize = (bufsize + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1); 234 } 235 bp = nfs_getcacheblk(vp, lbn, bufsize, p); 236 if (!bp) 237 return (EINTR); 238 if ((bp->b_flags & B_CACHE) == 0) { 239 bp->b_flags |= B_READ; 240 bp->b_flags &= ~(B_DONE | B_ERROR | B_INVAL); 241 not_readin = 0; 242 vfs_busy_pages(bp, 0); 243 error = nfs_doio(bp, cred, p); 244 if (error) { 245 brelse(bp); 246 return (error); 247 } 248 } 249 if (bufsize > on) { 250 n = min((unsigned)(bufsize - on), uio->uio_resid); 251 } else { 252 n = 0; 253 } 254 diff = np->n_size - uio->uio_offset; 255 if (diff < n) 256 n = diff; 257 if (not_readin && n > 0) { 258 if (on < bp->b_validoff || (on + n) > bp->b_validend) { 259 bp->b_flags |= B_NOCACHE; 260 bp->b_flags |= B_INVAFTERWRITE; 261 if (bp->b_dirtyend > 0) { 262 if ((bp->b_flags & B_DELWRI) == 0) 263 panic("nfsbioread"); 264 if (VOP_BWRITE(bp) == EINTR) 265 return (EINTR); 266 } else 267 brelse(bp); 268 goto again; 269 } 270 } 271 vp->v_lastr = lbn; 272 diff = (on >= bp->b_validend) ? 0 : (bp->b_validend - on); 273 if (diff < n) 274 n = diff; 275 break; 276 case VLNK: 277 nfsstats.biocache_readlinks++; 278 bp = nfs_getcacheblk(vp, (daddr_t)0, NFS_MAXPATHLEN, p); 279 if (!bp) 280 return (EINTR); 281 if ((bp->b_flags & B_CACHE) == 0) { 282 bp->b_flags |= B_READ; 283 vfs_busy_pages(bp, 0); 284 error = nfs_doio(bp, cred, p); 285 if (error) { 286 bp->b_flags |= B_ERROR; 287 brelse(bp); 288 return (error); 289 } 290 } 291 n = min(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid); 292 on = 0; 293 break; 294 case VDIR: 295 nfsstats.biocache_readdirs++; 296 lbn = uio->uio_offset / NFS_DIRBLKSIZ; 297 on = uio->uio_offset & (NFS_DIRBLKSIZ - 1); 298 bp = nfs_getcacheblk(vp, lbn, NFS_DIRBLKSIZ, p); 299 if (!bp) 300 return (EINTR); 301 if ((bp->b_flags & B_CACHE) == 0) { 302 bp->b_flags |= B_READ; 303 vfs_busy_pages(bp, 0); 304 error = nfs_doio(bp, cred, p); 305 if (error) { 306 vfs_unbusy_pages(bp); 307 brelse(bp); 308 while (error == NFSERR_BAD_COOKIE) { 309 nfs_invaldir(vp); 310 error = nfs_vinvalbuf(vp, 0, cred, p, 1); 311 /* 312 * Yuck! The directory has been modified on the 313 * server. The only way to get the block is by 314 * reading from the beginning to get all the 315 * offset cookies. 316 */ 317 for (i = 0; i <= lbn && !error; i++) { 318 bp = nfs_getcacheblk(vp, i, NFS_DIRBLKSIZ, p); 319 if (!bp) 320 return (EINTR); 321 if ((bp->b_flags & B_DONE) == 0) { 322 bp->b_flags |= B_READ; 323 vfs_busy_pages(bp, 0); 324 error = nfs_doio(bp, cred, p); 325 if (error) { 326 vfs_unbusy_pages(bp); 327 brelse(bp); 328 } else if (i < lbn) 329 brelse(bp); 330 } 331 } 332 } 333 if (error) 334 return (error); 335 } 336 } 337 338 /* 339 * If not eof and read aheads are enabled, start one. 340 * (You need the current block first, so that you have the 341 * directory offset cookie of the next block.) 342 */ 343 if (nfs_numasync > 0 && nmp->nm_readahead > 0 && 344 (np->n_direofoffset == 0 || 345 (lbn + 1) * NFS_DIRBLKSIZ < np->n_direofoffset) && 346 !(np->n_flag & NQNFSNONCACHE) && 347 !incore(vp, lbn + 1)) { 348 rabp = nfs_getcacheblk(vp, lbn + 1, NFS_DIRBLKSIZ, p); 349 if (rabp) { 350 if ((rabp->b_flags & (B_CACHE|B_DELWRI)) == 0) { 351 rabp->b_flags |= (B_READ | B_ASYNC); 352 vfs_busy_pages(rabp, 0); 353 if (nfs_asyncio(rabp, cred)) { 354 rabp->b_flags |= B_INVAL|B_ERROR; 355 vfs_unbusy_pages(rabp); 356 brelse(rabp); 357 } 358 } else { 359 brelse(rabp); 360 } 361 } 362 } 363 n = min(uio->uio_resid, NFS_DIRBLKSIZ - bp->b_resid - on); 364 break; 365 default: 366 printf(" nfs_bioread: type %x unexpected\n",vp->v_type); 367 break; 368 }; 369 370 if (n > 0) { 371 error = uiomove(bp->b_data + on, (int)n, uio); 372 } 373 switch (vp->v_type) { 374 case VREG: 375 break; 376 case VLNK: 377 n = 0; 378 break; 379 case VDIR: 380 if (np->n_flag & NQNFSNONCACHE) 381 bp->b_flags |= B_INVAL; 382 break; 383 default: 384 printf(" nfs_bioread: type %x unexpected\n",vp->v_type); 385 } 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 */ 394int 395nfs_write(ap) 396 struct vop_write_args /* { 397 struct vnode *a_vp; 398 struct uio *a_uio; 399 int a_ioflag; 400 struct ucred *a_cred; 401 } */ *ap; 402{ 403 register int biosize; 404 register struct uio *uio = ap->a_uio; 405 struct proc *p = uio->uio_procp; 406 register struct vnode *vp = ap->a_vp; 407 struct nfsnode *np = VTONFS(vp); 408 register struct ucred *cred = ap->a_cred; 409 int ioflag = ap->a_ioflag; 410 struct buf *bp; 411 struct vattr vattr; 412 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 413 daddr_t lbn; 414 int bufsize; 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_resid == 0) 449 return (0); 450 /* 451 * Maybe this should be above the vnode op call, but so long as 452 * file servers have no limits, i don't think it matters 453 */ 454 if (p && uio->uio_offset + uio->uio_resid > 455 p->p_rlimit[RLIMIT_FSIZE].rlim_cur) { 456 psignal(p, SIGXFSZ); 457 return (EFBIG); 458 } 459 /* 460 * I use nm_rsize, not nm_wsize so that all buffer cache blocks 461 * will be the same size within a filesystem. nfs_writerpc will 462 * still use nm_wsize when sizing the rpc's. 463 */ 464 biosize = vp->v_mount->mnt_stat.f_iosize; 465 do { 466 /* 467 * Check for a valid write lease. 468 */ 469 if ((nmp->nm_flag & NFSMNT_NQNFS) && 470 NQNFS_CKINVALID(vp, np, ND_WRITE)) { 471 do { 472 error = nqnfs_getlease(vp, ND_WRITE, cred, p); 473 } while (error == NQNFS_EXPIRED); 474 if (error) 475 return (error); 476 if (np->n_lrev != np->n_brev || 477 (np->n_flag & NQNFSNONCACHE)) { 478 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 479 if (error) 480 return (error); 481 np->n_brev = np->n_lrev; 482 } 483 } 484 if ((np->n_flag & NQNFSNONCACHE) && uio->uio_iovcnt == 1) { 485 iomode = NFSV3WRITE_FILESYNC; 486 error = nfs_writerpc(vp, uio, cred, &iomode, &must_commit); 487 if (must_commit) 488 nfs_clearcommit(vp->v_mount); 489 return (error); 490 } 491 nfsstats.biocache_writes++; 492 lbn = uio->uio_offset / biosize; 493 on = uio->uio_offset & (biosize-1); 494 n = min((unsigned)(biosize - on), uio->uio_resid); 495again: 496 if (uio->uio_offset + n > np->n_size) { 497 np->n_size = uio->uio_offset + n; 498 vnode_pager_setsize(vp, (u_long)np->n_size); 499 } 500 bufsize = biosize; 501 if ((lbn + 1) * biosize > np->n_size) { 502 bufsize = np->n_size - lbn * biosize; 503 bufsize = (bufsize + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1); 504 } 505 bp = nfs_getcacheblk(vp, lbn, bufsize, 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; 513 514 if ((bp->b_blkno * DEV_BSIZE) + bp->b_dirtyend > np->n_size) { 515 bp->b_dirtyend = np->n_size - (bp->b_blkno * DEV_BSIZE); 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 } while (uio->uio_resid > 0 && n > 0); 603 return (0); 604} 605 606/* 607 * Get an nfs cache block. 608 * Allocate a new one if the block isn't currently in the cache 609 * and return the block marked busy. If the calling process is 610 * interrupted by a signal for an interruptible mount point, return 611 * NULL. 612 */ 613static struct buf * 614nfs_getcacheblk(vp, bn, size, p) 615 struct vnode *vp; 616 daddr_t bn; 617 int size; 618 struct proc *p; 619{ 620 register struct buf *bp; 621 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 622 int biosize = vp->v_mount->mnt_stat.f_iosize; 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); 633 634 if( vp->v_type == VREG) 635 bp->b_blkno = (bn * biosize) / DEV_BSIZE; 636 637 return (bp); 638} 639 640/* 641 * Flush and invalidate all dirty buffers. If another process is already 642 * doing the flush, just wait for completion. 643 */ 644int 645nfs_vinvalbuf(vp, flags, cred, p, intrflg) 646 struct vnode *vp; 647 int flags; 648 struct ucred *cred; 649 struct proc *p; 650 int intrflg; 651{ 652 register struct nfsnode *np = VTONFS(vp); 653 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 654 int error = 0, slpflag, slptimeo; 655 656 if ((nmp->nm_flag & NFSMNT_INT) == 0) 657 intrflg = 0; 658 if (intrflg) { 659 slpflag = PCATCH; 660 slptimeo = 2 * hz; 661 } else { 662 slpflag = 0; 663 slptimeo = 0; 664 } 665 /* 666 * First wait for any other process doing a flush to complete. 667 */ 668 while (np->n_flag & NFLUSHINPROG) { 669 np->n_flag |= NFLUSHWANT; 670 error = tsleep((caddr_t)&np->n_flag, PRIBIO + 2, "nfsvinval", 671 slptimeo); 672 if (error && intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) 673 return (EINTR); 674 } 675 676 /* 677 * Now, flush as required. 678 */ 679 np->n_flag |= NFLUSHINPROG; 680 error = vinvalbuf(vp, flags, cred, p, slpflag, 0); 681 while (error) { 682 if (intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) { 683 np->n_flag &= ~NFLUSHINPROG; 684 if (np->n_flag & NFLUSHWANT) { 685 np->n_flag &= ~NFLUSHWANT; 686 wakeup((caddr_t)&np->n_flag); 687 } 688 return (EINTR); 689 } 690 error = vinvalbuf(vp, flags, cred, p, 0, slptimeo); 691 } 692 np->n_flag &= ~(NMODIFIED | NFLUSHINPROG); 693 if (np->n_flag & NFLUSHWANT) { 694 np->n_flag &= ~NFLUSHWANT; 695 wakeup((caddr_t)&np->n_flag); 696 } 697 return (0); 698} 699 700/* 701 * Initiate asynchronous I/O. Return an error if no nfsiods are available. 702 * This is mainly to avoid queueing async I/O requests when the nfsiods 703 * are all hung on a dead server. 704 */ 705int 706nfs_asyncio(bp, cred) 707 register struct buf *bp; 708 struct ucred *cred; 709{ 710 struct nfsmount *nmp; 711 int i; 712 int gotiod; 713 int slpflag = 0; 714 int slptimeo = 0; 715 int error; 716 717 if (nfs_numasync == 0) 718 return (EIO); 719 720 nmp = VFSTONFS(bp->b_vp->v_mount); 721again: 722 if (nmp->nm_flag & NFSMNT_INT) 723 slpflag = PCATCH; 724 gotiod = FALSE; 725 726 /* 727 * Find a free iod to process this request. 728 */ 729 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 730 if (nfs_iodwant[i]) { 731 /* 732 * Found one, so wake it up and tell it which 733 * mount to process. 734 */ 735 NFS_DPF(ASYNCIO, 736 ("nfs_asyncio: waking iod %d for mount %p\n", 737 i, nmp)); 738 nfs_iodwant[i] = (struct proc *)0; 739 nfs_iodmount[i] = nmp; 740 nmp->nm_bufqiods++; 741 wakeup((caddr_t)&nfs_iodwant[i]); 742 gotiod = TRUE; 743 } 744 745 /* 746 * If none are free, we may already have an iod working on this mount 747 * point. If so, it will process our request. 748 */ 749 if (!gotiod) { 750 if (nmp->nm_bufqiods > 0) { 751 NFS_DPF(ASYNCIO, 752 ("nfs_asyncio: %d iods are already processing mount %p\n", 753 nmp->nm_bufqiods, nmp)); 754 gotiod = TRUE; 755 } 756 } 757 758 /* 759 * If we have an iod which can process the request, then queue 760 * the buffer. 761 */ 762 if (gotiod) { 763 /* 764 * Ensure that the queue never grows too large. 765 */ 766 while (nmp->nm_bufqlen >= 2*nfs_numasync) { 767 NFS_DPF(ASYNCIO, 768 ("nfs_asyncio: waiting for mount %p queue to drain\n", nmp)); 769 nmp->nm_bufqwant = TRUE; 770 error = tsleep(&nmp->nm_bufq, slpflag | PRIBIO, 771 "nfsaio", slptimeo); 772 if (error) { 773 if (nfs_sigintr(nmp, NULL, bp->b_proc)) 774 return (EINTR); 775 if (slpflag == PCATCH) { 776 slpflag = 0; 777 slptimeo = 2 * hz; 778 } 779 } 780 /* 781 * We might have lost our iod while sleeping, 782 * so check and loop if nescessary. 783 */ 784 if (nmp->nm_bufqiods == 0) { 785 NFS_DPF(ASYNCIO, 786 ("nfs_asyncio: no iods after mount %p queue was drained, looping\n", nmp)); 787 goto again; 788 } 789 } 790 791 if (bp->b_flags & B_READ) { 792 if (bp->b_rcred == NOCRED && cred != NOCRED) { 793 crhold(cred); 794 bp->b_rcred = cred; 795 } 796 } else { 797 bp->b_flags |= B_WRITEINPROG; 798 if (bp->b_wcred == NOCRED && cred != NOCRED) { 799 crhold(cred); 800 bp->b_wcred = cred; 801 } 802 } 803 804 TAILQ_INSERT_TAIL(&nmp->nm_bufq, bp, b_freelist); 805 nmp->nm_bufqlen++; 806 return (0); 807 } 808 809 /* 810 * All the iods are busy on other mounts, so return EIO to 811 * force the caller to process the i/o synchronously. 812 */ 813 NFS_DPF(ASYNCIO, ("nfs_asyncio: no iods available, i/o is synchronous\n")); 814 return (EIO); 815} 816 817/* 818 * Do an I/O operation to/from a cache block. This may be called 819 * synchronously or from an nfsiod. 820 */ 821int 822nfs_doio(bp, cr, p) 823 register struct buf *bp; 824 struct ucred *cr; 825 struct proc *p; 826{ 827 register struct uio *uiop; 828 register struct vnode *vp; 829 struct nfsnode *np; 830 struct nfsmount *nmp; 831 int error = 0, diff, len, iomode, must_commit = 0; 832 struct uio uio; 833 struct iovec io; 834 835 vp = bp->b_vp; 836 np = VTONFS(vp); 837 nmp = VFSTONFS(vp->v_mount); 838 uiop = &uio; 839 uiop->uio_iov = &io; 840 uiop->uio_iovcnt = 1; 841 uiop->uio_segflg = UIO_SYSSPACE; 842 uiop->uio_procp = p; 843 844 /* 845 * Historically, paging was done with physio, but no more. 846 */ 847 if (bp->b_flags & B_PHYS) { 848 /* 849 * ...though reading /dev/drum still gets us here. 850 */ 851 io.iov_len = uiop->uio_resid = bp->b_bcount; 852 /* mapping was done by vmapbuf() */ 853 io.iov_base = bp->b_data; 854 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; 855 if (bp->b_flags & B_READ) { 856 uiop->uio_rw = UIO_READ; 857 nfsstats.read_physios++; 858 error = nfs_readrpc(vp, uiop, cr); 859 } else { 860 int com; 861 862 iomode = NFSV3WRITE_DATASYNC; 863 uiop->uio_rw = UIO_WRITE; 864 nfsstats.write_physios++; 865 error = nfs_writerpc(vp, uiop, cr, &iomode, &com); 866 } 867 if (error) { 868 bp->b_flags |= B_ERROR; 869 bp->b_error = error; 870 } 871 } else if (bp->b_flags & B_READ) { 872 io.iov_len = uiop->uio_resid = bp->b_bcount; 873 io.iov_base = bp->b_data; 874 uiop->uio_rw = UIO_READ; 875 switch (vp->v_type) { 876 case VREG: 877 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; 878 nfsstats.read_bios++; 879 error = nfs_readrpc(vp, uiop, cr); 880 if (!error) { 881 bp->b_validoff = 0; 882 if (uiop->uio_resid) { 883 /* 884 * If len > 0, there is a hole in the file and 885 * no writes after the hole have been pushed to 886 * the server yet. 887 * Just zero fill the rest of the valid area. 888 */ 889 diff = bp->b_bcount - uiop->uio_resid; 890 len = np->n_size - (((u_quad_t)bp->b_blkno) * DEV_BSIZE 891 + diff); 892 if (len > 0) { 893 len = min(len, uiop->uio_resid); 894 bzero((char *)bp->b_data + diff, len); 895 bp->b_validend = diff + len; 896 } else 897 bp->b_validend = diff; 898 } else 899 bp->b_validend = bp->b_bcount; 900 } 901 if (p && (vp->v_flag & VTEXT) && 902 (((nmp->nm_flag & NFSMNT_NQNFS) && 903 NQNFS_CKINVALID(vp, np, ND_READ) && 904 np->n_lrev != np->n_brev) || 905 (!(nmp->nm_flag & NFSMNT_NQNFS) && 906 np->n_mtime != np->n_vattr.va_mtime.tv_sec))) { 907 uprintf("Process killed due to text file modification\n"); 908 psignal(p, SIGKILL); 909#ifdef __NetBSD__ 910 p->p_holdcnt++; 911#else 912 p->p_flag |= P_NOSWAP; 913#endif 914 } 915 break; 916 case VLNK: 917 uiop->uio_offset = (off_t)0; 918 nfsstats.readlink_bios++; 919 error = nfs_readlinkrpc(vp, uiop, cr); 920 break; 921 case VDIR: 922 nfsstats.readdir_bios++; 923 uiop->uio_offset = ((u_quad_t)bp->b_lblkno) * NFS_DIRBLKSIZ; 924 if (nmp->nm_flag & NFSMNT_RDIRPLUS) { 925 error = nfs_readdirplusrpc(vp, uiop, cr); 926 if (error == NFSERR_NOTSUPP) 927 nmp->nm_flag &= ~NFSMNT_RDIRPLUS; 928 } 929 if ((nmp->nm_flag & NFSMNT_RDIRPLUS) == 0) 930 error = nfs_readdirrpc(vp, uiop, cr); 931 break; 932 default: 933 printf("nfs_doio: type %x unexpected\n",vp->v_type); 934 break; 935 }; 936 if (error) { 937 bp->b_flags |= B_ERROR; 938 bp->b_error = error; 939 } 940 } else { 941 if (((bp->b_blkno * DEV_BSIZE) + bp->b_dirtyend) > np->n_size) 942 bp->b_dirtyend = np->n_size - (bp->b_blkno * DEV_BSIZE); 943 944 if (bp->b_dirtyend > bp->b_dirtyoff) { 945 io.iov_len = uiop->uio_resid = bp->b_dirtyend 946 - bp->b_dirtyoff; 947 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE 948 + bp->b_dirtyoff; 949 io.iov_base = (char *)bp->b_data + bp->b_dirtyoff; 950 uiop->uio_rw = UIO_WRITE; 951 nfsstats.write_bios++; 952 if ((bp->b_flags & (B_ASYNC | B_NEEDCOMMIT | B_NOCACHE)) == B_ASYNC) 953 iomode = NFSV3WRITE_UNSTABLE; 954 else 955 iomode = NFSV3WRITE_FILESYNC; 956 bp->b_flags |= B_WRITEINPROG; 957 error = nfs_writerpc(vp, uiop, cr, &iomode, &must_commit); 958 if (!error && iomode == NFSV3WRITE_UNSTABLE) 959 bp->b_flags |= B_NEEDCOMMIT; 960 else 961 bp->b_flags &= ~B_NEEDCOMMIT; 962 bp->b_flags &= ~B_WRITEINPROG; 963 964 /* 965 * For an interrupted write, the buffer is still valid 966 * and the write hasn't been pushed to the server yet, 967 * so we can't set B_ERROR and report the interruption 968 * by setting B_EINTR. For the B_ASYNC case, B_EINTR 969 * is not relevant, so the rpc attempt is essentially 970 * a noop. For the case of a V3 write rpc not being 971 * committed to stable storage, the block is still 972 * dirty and requires either a commit rpc or another 973 * write rpc with iomode == NFSV3WRITE_FILESYNC before 974 * the block is reused. This is indicated by setting 975 * the B_DELWRI and B_NEEDCOMMIT flags. 976 */ 977 if (error == EINTR 978 || (!error && (bp->b_flags & B_NEEDCOMMIT))) { 979 bp->b_flags &= ~(B_INVAL|B_NOCACHE); 980 bp->b_flags |= B_DELWRI; 981 982 /* 983 * Since for the B_ASYNC case, nfs_bwrite() has reassigned the 984 * buffer to the clean list, we have to reassign it back to the 985 * dirty one. Ugh. 986 */ 987 if (bp->b_flags & B_ASYNC) 988 reassignbuf(bp, vp); 989 else 990 bp->b_flags |= B_EINTR; 991 } else { 992 if (error) { 993 bp->b_flags |= B_ERROR; 994 bp->b_error = np->n_error = error; 995 np->n_flag |= NWRITEERR; 996 } 997 bp->b_dirtyoff = bp->b_dirtyend = 0; 998 } 999 } else { 1000 bp->b_resid = 0; 1001 biodone(bp); 1002 return (0); 1003 } 1004 } 1005 bp->b_resid = uiop->uio_resid; 1006 if (must_commit) 1007 nfs_clearcommit(vp->v_mount); 1008 biodone(bp); 1009 return (error); 1010} 1011