nfs_bio.c revision 21673
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 * $FreeBSD: head/sys/nfsclient/nfs_bio.c 21673 1997-01-14 07:20:47Z jkh $ 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 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.tv_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.tv_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.tv_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 bp->b_flags &= ~(B_DONE | B_ERROR | B_INVAL); 254 not_readin = 0; 255 vfs_busy_pages(bp, 0); 256 error = nfs_doio(bp, cred, p); 257 if (error) { 258 brelse(bp); 259 return (error); 260 } 261 } 262 if (bufsize > on) { 263 n = min((unsigned)(bufsize - on), uio->uio_resid); 264 } else { 265 n = 0; 266 } 267 diff = np->n_size - uio->uio_offset; 268 if (diff < n) 269 n = diff; 270 if (not_readin && n > 0) { 271 if (on < bp->b_validoff || (on + n) > bp->b_validend) { 272 bp->b_flags |= B_NOCACHE; 273 if (bp->b_dirtyend > 0) { 274 if ((bp->b_flags & B_DELWRI) == 0) 275 panic("nfsbioread"); 276 if (VOP_BWRITE(bp) == EINTR) 277 return (EINTR); 278 } else 279 brelse(bp); 280 goto again; 281 } 282 } 283 vp->v_lastr = lbn; 284 diff = (on >= bp->b_validend) ? 0 : (bp->b_validend - on); 285 if (diff < n) 286 n = diff; 287 break; 288 case VLNK: 289 nfsstats.biocache_readlinks++; 290 bp = nfs_getcacheblk(vp, (daddr_t)0, NFS_MAXPATHLEN, p); 291 if (!bp) 292 return (EINTR); 293 if ((bp->b_flags & B_CACHE) == 0) { 294 bp->b_flags |= B_READ; 295 vfs_busy_pages(bp, 0); 296 error = nfs_doio(bp, cred, p); 297 if (error) { 298 bp->b_flags |= B_ERROR; 299 brelse(bp); 300 return (error); 301 } 302 } 303 n = min(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid); 304 on = 0; 305 break; 306 case VDIR: 307 nfsstats.biocache_readdirs++; 308 lbn = uio->uio_offset / NFS_DIRBLKSIZ; 309 on = uio->uio_offset & (NFS_DIRBLKSIZ - 1); 310 bp = nfs_getcacheblk(vp, lbn, NFS_DIRBLKSIZ, p); 311 if (!bp) 312 return (EINTR); 313 if ((bp->b_flags & B_CACHE) == 0) { 314 bp->b_flags |= B_READ; 315 vfs_busy_pages(bp, 0); 316 error = nfs_doio(bp, cred, p); 317 if (error) { 318 vfs_unbusy_pages(bp); 319 brelse(bp); 320 while (error == NFSERR_BAD_COOKIE) { 321 nfs_invaldir(vp); 322 error = nfs_vinvalbuf(vp, 0, cred, p, 1); 323 /* 324 * Yuck! The directory has been modified on the 325 * server. The only way to get the block is by 326 * reading from the beginning to get all the 327 * offset cookies. 328 */ 329 for (i = 0; i <= lbn && !error; i++) { 330 bp = nfs_getcacheblk(vp, i, NFS_DIRBLKSIZ, p); 331 if (!bp) 332 return (EINTR); 333 if ((bp->b_flags & B_DONE) == 0) { 334 bp->b_flags |= B_READ; 335 vfs_busy_pages(bp, 0); 336 error = nfs_doio(bp, cred, p); 337 if (error) { 338 vfs_unbusy_pages(bp); 339 brelse(bp); 340 } else if (i < lbn) 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 struct nfsmount *nmp; 727 int i; 728 int gotiod; 729 int slpflag = 0; 730 int slptimeo = 0; 731 int error; 732 733 if (nfs_numasync == 0) 734 return (EIO); 735 736 nmp = VFSTONFS(bp->b_vp->v_mount); 737again: 738 if (nmp->nm_flag & NFSMNT_INT) 739 slpflag = PCATCH; 740 gotiod = FALSE; 741 742 /* 743 * Find a free iod to process this request. 744 */ 745 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 746 if (nfs_iodwant[i]) { 747 /* 748 * Found one, so wake it up and tell it which 749 * mount to process. 750 */ 751 NFS_DPF(ASYNCIO, 752 ("nfs_asyncio: waking iod %d for mount %p\n", 753 i, nmp)); 754 nfs_iodwant[i] = (struct proc *)0; 755 nfs_iodmount[i] = nmp; 756 nmp->nm_bufqiods++; 757 wakeup((caddr_t)&nfs_iodwant[i]); 758 gotiod = TRUE; 759 } 760 761 /* 762 * If none are free, we may already have an iod working on this mount 763 * point. If so, it will process our request. 764 */ 765 if (!gotiod) { 766 if (nmp->nm_bufqiods > 0) { 767 NFS_DPF(ASYNCIO, 768 ("nfs_asyncio: %d iods are already processing mount %p\n", 769 nmp->nm_bufqiods, nmp)); 770 gotiod = TRUE; 771 } 772 } 773 774 /* 775 * If we have an iod which can process the request, then queue 776 * the buffer. 777 */ 778 if (gotiod) { 779 /* 780 * Ensure that the queue never grows too large. 781 */ 782 while (nmp->nm_bufqlen >= 2*nfs_numasync) { 783 NFS_DPF(ASYNCIO, 784 ("nfs_asyncio: waiting for mount %p queue to drain\n", nmp)); 785 nmp->nm_bufqwant = TRUE; 786 error = tsleep(&nmp->nm_bufq, slpflag | PRIBIO, 787 "nfsaio", slptimeo); 788 if (error) { 789 if (nfs_sigintr(nmp, NULL, bp->b_proc)) 790 return (EINTR); 791 if (slpflag == PCATCH) { 792 slpflag = 0; 793 slptimeo = 2 * hz; 794 } 795 } 796 /* 797 * We might have lost our iod while sleeping, 798 * so check and loop if nescessary. 799 */ 800 if (nmp->nm_bufqiods == 0) { 801 NFS_DPF(ASYNCIO, 802 ("nfs_asyncio: no iods after mount %p queue was drained, looping\n", nmp)); 803 goto again; 804 } 805 } 806 807 if (bp->b_flags & B_READ) { 808 if (bp->b_rcred == NOCRED && cred != NOCRED) { 809 crhold(cred); 810 bp->b_rcred = cred; 811 } 812 } else { 813 bp->b_flags |= B_WRITEINPROG; 814 if (bp->b_wcred == NOCRED && cred != NOCRED) { 815 crhold(cred); 816 bp->b_wcred = cred; 817 } 818 } 819 820 TAILQ_INSERT_TAIL(&nmp->nm_bufq, bp, b_freelist); 821 nmp->nm_bufqlen++; 822 return (0); 823 } 824 825 /* 826 * All the iods are busy on other mounts, so return EIO to 827 * force the caller to process the i/o synchronously. 828 */ 829 NFS_DPF(ASYNCIO, ("nfs_asyncio: no iods available, i/o is synchronous\n")); 830 return (EIO); 831} 832 833/* 834 * Do an I/O operation to/from a cache block. This may be called 835 * synchronously or from an nfsiod. 836 */ 837int 838nfs_doio(bp, cr, p) 839 register struct buf *bp; 840 struct ucred *cr; 841 struct proc *p; 842{ 843 register struct uio *uiop; 844 register struct vnode *vp; 845 struct nfsnode *np; 846 struct nfsmount *nmp; 847 int error = 0, diff, len, iomode, must_commit = 0; 848 struct uio uio; 849 struct iovec io; 850 851 vp = bp->b_vp; 852 np = VTONFS(vp); 853 nmp = VFSTONFS(vp->v_mount); 854 uiop = &uio; 855 uiop->uio_iov = &io; 856 uiop->uio_iovcnt = 1; 857 uiop->uio_segflg = UIO_SYSSPACE; 858 uiop->uio_procp = p; 859 860 /* 861 * Historically, paging was done with physio, but no more. 862 */ 863 if (bp->b_flags & B_PHYS) { 864 /* 865 * ...though reading /dev/drum still gets us here. 866 */ 867 io.iov_len = uiop->uio_resid = bp->b_bcount; 868 /* mapping was done by vmapbuf() */ 869 io.iov_base = bp->b_data; 870 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; 871 if (bp->b_flags & B_READ) { 872 uiop->uio_rw = UIO_READ; 873 nfsstats.read_physios++; 874 error = nfs_readrpc(vp, uiop, cr); 875 } else { 876 int com; 877 878 iomode = NFSV3WRITE_DATASYNC; 879 uiop->uio_rw = UIO_WRITE; 880 nfsstats.write_physios++; 881 error = nfs_writerpc(vp, uiop, cr, &iomode, &com); 882 } 883 if (error) { 884 bp->b_flags |= B_ERROR; 885 bp->b_error = error; 886 } 887 } else if (bp->b_flags & B_READ) { 888 io.iov_len = uiop->uio_resid = bp->b_bcount; 889 io.iov_base = bp->b_data; 890 uiop->uio_rw = UIO_READ; 891 switch (vp->v_type) { 892 case VREG: 893 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; 894 nfsstats.read_bios++; 895 error = nfs_readrpc(vp, uiop, cr); 896 if (!error) { 897 bp->b_validoff = 0; 898 if (uiop->uio_resid) { 899 /* 900 * If len > 0, there is a hole in the file and 901 * no writes after the hole have been pushed to 902 * the server yet. 903 * Just zero fill the rest of the valid area. 904 */ 905 diff = bp->b_bcount - uiop->uio_resid; 906 len = np->n_size - (((u_quad_t)bp->b_blkno) * DEV_BSIZE 907 + diff); 908 if (len > 0) { 909 len = min(len, uiop->uio_resid); 910 bzero((char *)bp->b_data + diff, len); 911 bp->b_validend = diff + len; 912 } else 913 bp->b_validend = diff; 914 } else 915 bp->b_validend = bp->b_bcount; 916 } 917 if (p && (vp->v_flag & VTEXT) && 918 (((nmp->nm_flag & NFSMNT_NQNFS) && 919 NQNFS_CKINVALID(vp, np, ND_READ) && 920 np->n_lrev != np->n_brev) || 921 (!(nmp->nm_flag & NFSMNT_NQNFS) && 922 np->n_mtime != np->n_vattr.va_mtime.tv_sec))) { 923 uprintf("Process killed due to text file modification\n"); 924 psignal(p, SIGKILL); 925#ifdef __NetBSD__ 926 p->p_holdcnt++; 927#else 928 p->p_flag |= P_NOSWAP; 929#endif 930 } 931 break; 932 case VLNK: 933 uiop->uio_offset = (off_t)0; 934 nfsstats.readlink_bios++; 935 error = nfs_readlinkrpc(vp, uiop, cr); 936 break; 937 case VDIR: 938 nfsstats.readdir_bios++; 939 uiop->uio_offset = ((u_quad_t)bp->b_lblkno) * NFS_DIRBLKSIZ; 940 if (nmp->nm_flag & NFSMNT_RDIRPLUS) { 941 error = nfs_readdirplusrpc(vp, uiop, cr); 942 if (error == NFSERR_NOTSUPP) 943 nmp->nm_flag &= ~NFSMNT_RDIRPLUS; 944 } 945 if ((nmp->nm_flag & NFSMNT_RDIRPLUS) == 0) 946 error = nfs_readdirrpc(vp, uiop, cr); 947 break; 948 default: 949 printf("nfs_doio: type %x unexpected\n",vp->v_type); 950 break; 951 }; 952 if (error) { 953 bp->b_flags |= B_ERROR; 954 bp->b_error = error; 955 } 956 } else { 957 if (((bp->b_blkno * DEV_BSIZE) + bp->b_dirtyend) > np->n_size) 958 bp->b_dirtyend = np->n_size - (bp->b_blkno * DEV_BSIZE); 959 960 if (bp->b_dirtyend > bp->b_dirtyoff) { 961 io.iov_len = uiop->uio_resid = bp->b_dirtyend 962 - bp->b_dirtyoff; 963 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE 964 + bp->b_dirtyoff; 965 io.iov_base = (char *)bp->b_data + bp->b_dirtyoff; 966 uiop->uio_rw = UIO_WRITE; 967 nfsstats.write_bios++; 968 if ((bp->b_flags & (B_ASYNC | B_NEEDCOMMIT | B_NOCACHE)) == B_ASYNC) 969 iomode = NFSV3WRITE_UNSTABLE; 970 else 971 iomode = NFSV3WRITE_FILESYNC; 972 bp->b_flags |= B_WRITEINPROG; 973 error = nfs_writerpc(vp, uiop, cr, &iomode, &must_commit); 974 if (!error && iomode == NFSV3WRITE_UNSTABLE) 975 bp->b_flags |= B_NEEDCOMMIT; 976 else 977 bp->b_flags &= ~B_NEEDCOMMIT; 978 bp->b_flags &= ~B_WRITEINPROG; 979 980 /* 981 * For an interrupted write, the buffer is still valid 982 * and the write hasn't been pushed to the server yet, 983 * so we can't set B_ERROR and report the interruption 984 * by setting B_EINTR. For the B_ASYNC case, B_EINTR 985 * is not relevant, so the rpc attempt is essentially 986 * a noop. For the case of a V3 write rpc not being 987 * committed to stable storage, the block is still 988 * dirty and requires either a commit rpc or another 989 * write rpc with iomode == NFSV3WRITE_FILESYNC before 990 * the block is reused. This is indicated by setting 991 * the B_DELWRI and B_NEEDCOMMIT flags. 992 */ 993 if (error == EINTR 994 || (!error && (bp->b_flags & B_NEEDCOMMIT))) { 995 bp->b_flags &= ~(B_INVAL|B_NOCACHE); 996 bp->b_flags |= B_DELWRI; 997 998 /* 999 * Since for the B_ASYNC case, nfs_bwrite() has reassigned the 1000 * buffer to the clean list, we have to reassign it back to the 1001 * dirty one. Ugh. 1002 */ 1003 if (bp->b_flags & B_ASYNC) 1004 reassignbuf(bp, vp); 1005 else 1006 bp->b_flags |= B_EINTR; 1007 } else { 1008 if (error) { 1009 bp->b_flags |= B_ERROR; 1010 bp->b_error = np->n_error = error; 1011 np->n_flag |= NWRITEERR; 1012 } 1013 bp->b_dirtyoff = bp->b_dirtyend = 0; 1014 } 1015 } else { 1016 bp->b_resid = 0; 1017 biodone(bp); 1018 return (0); 1019 } 1020 } 1021 bp->b_resid = uiop->uio_resid; 1022 if (must_commit) 1023 nfs_clearcommit(vp->v_mount); 1024 biodone(bp); 1025 return (error); 1026} 1027