nfs_bio.c revision 31617
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 * $Id: nfs_bio.c,v 1.44 1997/09/10 19:52:25 phk Exp $ 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 51#include <vm/vm.h> 52#include <vm/vm_extern.h> 53#include <vm/vm_prot.h> 54#include <vm/vm_page.h> 55#include <vm/vm_object.h> 56#include <vm/vm_pager.h> 57#include <vm/vnode_pager.h> 58 59#include <nfs/rpcv2.h> 60#include <nfs/nfsproto.h> 61#include <nfs/nfs.h> 62#include <nfs/nfsmount.h> 63#include <nfs/nqnfs.h> 64#include <nfs/nfsnode.h> 65 66static struct buf *nfs_getcacheblk __P((struct vnode *vp, daddr_t bn, int size, 67 struct proc *p)); 68 69extern int nfs_numasync; 70extern struct nfsstats nfsstats; 71 72/* 73 * Vnode op for VM getpages. 74 */ 75int 76nfs_getpages(ap) 77 struct vop_getpages_args *ap; 78{ 79 int i, bsize; 80 vm_object_t obj; 81 int pcount; 82 struct uio auio; 83 struct iovec aiov; 84 int error; 85 vm_page_t m; 86 87 if (!(ap->a_vp->v_flag & VVMIO)) { 88 printf("nfs_getpages: called with non-VMIO vnode??\n"); 89 return EOPNOTSUPP; 90 } 91 92 pcount = round_page(ap->a_count) / PAGE_SIZE; 93 94 obj = ap->a_m[ap->a_reqpage]->object; 95 bsize = ap->a_vp->v_mount->mnt_stat.f_iosize; 96 97 for (i = 0; i < pcount; i++) { 98 if (i != ap->a_reqpage) { 99 vnode_pager_freepage(ap->a_m[i]); 100 } 101 } 102 m = ap->a_m[ap->a_reqpage]; 103 104 m->busy++; 105 m->flags &= ~PG_BUSY; 106 107 auio.uio_iov = &aiov; 108 auio.uio_iovcnt = 1; 109 aiov.iov_base = 0; 110 aiov.iov_len = PAGE_SIZE; 111 auio.uio_resid = PAGE_SIZE; 112 auio.uio_offset = IDX_TO_OFF(m->pindex); 113 auio.uio_segflg = UIO_NOCOPY; 114 auio.uio_rw = UIO_READ; 115 auio.uio_procp = curproc; 116 error = nfs_bioread(ap->a_vp, &auio, IO_NODELOCKED, curproc->p_ucred, 1); 117 118 m->flags |= PG_BUSY; 119 m->busy--; 120 121 if (error && (auio.uio_resid == PAGE_SIZE)) 122 return VM_PAGER_ERROR; 123 return 0; 124} 125 126/* 127 * Vnode op for read using bio 128 * Any similarity to readip() is purely coincidental 129 */ 130int 131nfs_bioread(vp, uio, ioflag, cred, getpages) 132 register struct vnode *vp; 133 register struct uio *uio; 134 int ioflag; 135 struct ucred *cred; 136 int getpages; 137{ 138 register struct nfsnode *np = VTONFS(vp); 139 register int biosize, diff, i; 140 struct buf *bp = 0, *rabp; 141 struct vattr vattr; 142 struct proc *p; 143 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 144 daddr_t lbn, rabn; 145 int bufsize; 146 int nra, error = 0, n = 0, on = 0, not_readin; 147 148#ifdef DIAGNOSTIC 149 if (uio->uio_rw != UIO_READ) 150 panic("nfs_read mode"); 151#endif 152 if (uio->uio_resid == 0) 153 return (0); 154 if (uio->uio_offset < 0) 155 return (EINVAL); 156 p = uio->uio_procp; 157 if ((nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_GOTFSINFO)) == NFSMNT_NFSV3) 158 (void)nfs_fsinfo(nmp, vp, cred, p); 159 biosize = vp->v_mount->mnt_stat.f_iosize; 160 /* 161 * For nfs, cache consistency can only be maintained approximately. 162 * Although RFC1094 does not specify the criteria, the following is 163 * believed to be compatible with the reference port. 164 * For nqnfs, full cache consistency is maintained within the loop. 165 * For nfs: 166 * If the file's modify time on the server has changed since the 167 * last read rpc or you have written to the file, 168 * you may have lost data cache consistency with the 169 * server, so flush all of the file's data out of the cache. 170 * Then force a getattr rpc to ensure that you have up to date 171 * attributes. 172 * NB: This implies that cache data can be read when up to 173 * NFS_ATTRTIMEO seconds out of date. If you find that you need current 174 * attributes this could be forced by setting n_attrstamp to 0 before 175 * the VOP_GETATTR() call. 176 */ 177 if ((nmp->nm_flag & NFSMNT_NQNFS) == 0) { 178 if (np->n_flag & NMODIFIED) { 179 if (vp->v_type != VREG) { 180 if (vp->v_type != VDIR) 181 panic("nfs: bioread, not dir"); 182 nfs_invaldir(vp); 183 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 184 if (error) 185 return (error); 186 } 187 np->n_attrstamp = 0; 188 error = VOP_GETATTR(vp, &vattr, cred, p); 189 if (error) 190 return (error); 191 np->n_mtime = vattr.va_mtime.tv_sec; 192 } else { 193 error = VOP_GETATTR(vp, &vattr, cred, p); 194 if (error) 195 return (error); 196 if (np->n_mtime != vattr.va_mtime.tv_sec) { 197 if (vp->v_type == VDIR) 198 nfs_invaldir(vp); 199 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 200 if (error) 201 return (error); 202 np->n_mtime = vattr.va_mtime.tv_sec; 203 } 204 } 205 } 206 do { 207 208 /* 209 * Get a valid lease. If cached data is stale, flush it. 210 */ 211 if (nmp->nm_flag & NFSMNT_NQNFS) { 212 if (NQNFS_CKINVALID(vp, np, ND_READ)) { 213 do { 214 error = nqnfs_getlease(vp, ND_READ, cred, p); 215 } while (error == NQNFS_EXPIRED); 216 if (error) 217 return (error); 218 if (np->n_lrev != np->n_brev || 219 (np->n_flag & NQNFSNONCACHE) || 220 ((np->n_flag & NMODIFIED) && vp->v_type == VDIR)) { 221 if (vp->v_type == VDIR) 222 nfs_invaldir(vp); 223 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 224 if (error) 225 return (error); 226 np->n_brev = np->n_lrev; 227 } 228 } else if (vp->v_type == VDIR && (np->n_flag & NMODIFIED)) { 229 nfs_invaldir(vp); 230 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 231 if (error) 232 return (error); 233 } 234 } 235 if (np->n_flag & NQNFSNONCACHE) { 236 switch (vp->v_type) { 237 case VREG: 238 return (nfs_readrpc(vp, uio, cred)); 239 case VLNK: 240 return (nfs_readlinkrpc(vp, uio, cred)); 241 case VDIR: 242 break; 243 default: 244 printf(" NQNFSNONCACHE: type %x unexpected\n", 245 vp->v_type); 246 }; 247 } 248 switch (vp->v_type) { 249 case VREG: 250 nfsstats.biocache_reads++; 251 lbn = uio->uio_offset / biosize; 252 on = uio->uio_offset & (biosize - 1); 253 not_readin = 1; 254 255 /* 256 * Start the read ahead(s), as required. 257 */ 258 if (nfs_numasync > 0 && nmp->nm_readahead > 0) { 259 for (nra = 0; nra < nmp->nm_readahead && 260 (off_t)(lbn + 1 + nra) * biosize < np->n_size; nra++) { 261 rabn = lbn + 1 + nra; 262 if (!incore(vp, rabn)) { 263 rabp = nfs_getcacheblk(vp, rabn, biosize, p); 264 if (!rabp) 265 return (EINTR); 266 if ((rabp->b_flags & (B_CACHE|B_DELWRI)) == 0) { 267 rabp->b_flags |= (B_READ | B_ASYNC); 268 vfs_busy_pages(rabp, 0); 269 if (nfs_asyncio(rabp, cred)) { 270 rabp->b_flags |= B_INVAL|B_ERROR; 271 vfs_unbusy_pages(rabp); 272 brelse(rabp); 273 } 274 } else 275 brelse(rabp); 276 } 277 } 278 } 279 280 /* 281 * If the block is in the cache and has the required data 282 * in a valid region, just copy it out. 283 * Otherwise, get the block and write back/read in, 284 * as required. 285 */ 286again: 287 bufsize = biosize; 288 if ((off_t)(lbn + 1) * biosize > np->n_size && 289 (off_t)(lbn + 1) * biosize - np->n_size < biosize) { 290 bufsize = np->n_size - lbn * biosize; 291 bufsize = (bufsize + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1); 292 } 293 bp = nfs_getcacheblk(vp, lbn, bufsize, p); 294 if (!bp) 295 return (EINTR); 296 /* 297 * If we are being called from nfs_getpages, we must 298 * make sure the buffer is a vmio buffer. The vp will 299 * already be setup for vmio but there may be some old 300 * non-vmio buffers attached to it. 301 */ 302 if (getpages && !(bp->b_flags & B_VMIO)) { 303#ifdef DIAGNOSTIC 304 printf("nfs_bioread: non vmio buf found, discarding\n"); 305#endif 306 bp->b_flags |= B_NOCACHE; 307 bp->b_flags |= B_INVAFTERWRITE; 308 if (bp->b_dirtyend > 0) { 309 if ((bp->b_flags & B_DELWRI) == 0) 310 panic("nfsbioread"); 311 if (VOP_BWRITE(bp) == EINTR) 312 return (EINTR); 313 } else 314 brelse(bp); 315 goto again; 316 } 317 if ((bp->b_flags & B_CACHE) == 0) { 318 bp->b_flags |= B_READ; 319 bp->b_flags &= ~(B_DONE | B_ERROR | B_INVAL); 320 not_readin = 0; 321 vfs_busy_pages(bp, 0); 322 error = nfs_doio(bp, cred, p); 323 if (error) { 324 brelse(bp); 325 return (error); 326 } 327 } 328 if (bufsize > on) { 329 n = min((unsigned)(bufsize - on), uio->uio_resid); 330 } else { 331 n = 0; 332 } 333 diff = np->n_size - uio->uio_offset; 334 if (diff < n) 335 n = diff; 336 if (not_readin && n > 0) { 337 if (on < bp->b_validoff || (on + n) > bp->b_validend) { 338 bp->b_flags |= B_NOCACHE; 339 bp->b_flags |= B_INVAFTERWRITE; 340 if (bp->b_dirtyend > 0) { 341 if ((bp->b_flags & B_DELWRI) == 0) 342 panic("nfsbioread"); 343 if (VOP_BWRITE(bp) == EINTR) 344 return (EINTR); 345 } else 346 brelse(bp); 347 goto again; 348 } 349 } 350 vp->v_lastr = lbn; 351 diff = (on >= bp->b_validend) ? 0 : (bp->b_validend - on); 352 if (diff < n) 353 n = diff; 354 break; 355 case VLNK: 356 nfsstats.biocache_readlinks++; 357 bp = nfs_getcacheblk(vp, (daddr_t)0, NFS_MAXPATHLEN, p); 358 if (!bp) 359 return (EINTR); 360 if ((bp->b_flags & B_CACHE) == 0) { 361 bp->b_flags |= B_READ; 362 vfs_busy_pages(bp, 0); 363 error = nfs_doio(bp, cred, p); 364 if (error) { 365 bp->b_flags |= B_ERROR; 366 brelse(bp); 367 return (error); 368 } 369 } 370 n = min(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid); 371 on = 0; 372 break; 373 case VDIR: 374 nfsstats.biocache_readdirs++; 375 if (np->n_direofoffset 376 && uio->uio_offset >= np->n_direofoffset) { 377 return (0); 378 } 379 lbn = uio->uio_offset / NFS_DIRBLKSIZ; 380 on = uio->uio_offset & (NFS_DIRBLKSIZ - 1); 381 bp = nfs_getcacheblk(vp, lbn, NFS_DIRBLKSIZ, p); 382 if (!bp) 383 return (EINTR); 384 if ((bp->b_flags & B_CACHE) == 0) { 385 bp->b_flags |= B_READ; 386 vfs_busy_pages(bp, 0); 387 error = nfs_doio(bp, cred, p); 388 if (error) { 389 vfs_unbusy_pages(bp); 390 brelse(bp); 391 while (error == NFSERR_BAD_COOKIE) { 392 nfs_invaldir(vp); 393 error = nfs_vinvalbuf(vp, 0, cred, p, 1); 394 /* 395 * Yuck! The directory has been modified on the 396 * server. The only way to get the block is by 397 * reading from the beginning to get all the 398 * offset cookies. 399 */ 400 for (i = 0; i <= lbn && !error; i++) { 401 if (np->n_direofoffset 402 && (i * NFS_DIRBLKSIZ) >= np->n_direofoffset) 403 return (0); 404 bp = nfs_getcacheblk(vp, i, NFS_DIRBLKSIZ, p); 405 if (!bp) 406 return (EINTR); 407 if ((bp->b_flags & B_DONE) == 0) { 408 bp->b_flags |= B_READ; 409 vfs_busy_pages(bp, 0); 410 error = nfs_doio(bp, cred, p); 411 if (error) { 412 vfs_unbusy_pages(bp); 413 brelse(bp); 414 } else if (i < lbn) 415 brelse(bp); 416 } 417 } 418 } 419 if (error) 420 return (error); 421 } 422 } 423 424 /* 425 * If not eof and read aheads are enabled, start one. 426 * (You need the current block first, so that you have the 427 * directory offset cookie of the next block.) 428 */ 429 if (nfs_numasync > 0 && nmp->nm_readahead > 0 && 430 (np->n_direofoffset == 0 || 431 (lbn + 1) * NFS_DIRBLKSIZ < np->n_direofoffset) && 432 !(np->n_flag & NQNFSNONCACHE) && 433 !incore(vp, lbn + 1)) { 434 rabp = nfs_getcacheblk(vp, lbn + 1, NFS_DIRBLKSIZ, p); 435 if (rabp) { 436 if ((rabp->b_flags & (B_CACHE|B_DELWRI)) == 0) { 437 rabp->b_flags |= (B_READ | B_ASYNC); 438 vfs_busy_pages(rabp, 0); 439 if (nfs_asyncio(rabp, cred)) { 440 rabp->b_flags |= B_INVAL|B_ERROR; 441 vfs_unbusy_pages(rabp); 442 brelse(rabp); 443 } 444 } else { 445 brelse(rabp); 446 } 447 } 448 } 449 /* 450 * Make sure we use a signed variant of min() since 451 * the second term may be negative. 452 */ 453 n = lmin(uio->uio_resid, NFS_DIRBLKSIZ - bp->b_resid - on); 454 break; 455 default: 456 printf(" nfs_bioread: type %x unexpected\n",vp->v_type); 457 break; 458 }; 459 460 if (n > 0) { 461 error = uiomove(bp->b_data + on, (int)n, uio); 462 } 463 switch (vp->v_type) { 464 case VREG: 465 break; 466 case VLNK: 467 n = 0; 468 break; 469 case VDIR: 470 if (np->n_flag & NQNFSNONCACHE) 471 bp->b_flags |= B_INVAL; 472 break; 473 default: 474 printf(" nfs_bioread: type %x unexpected\n",vp->v_type); 475 } 476 brelse(bp); 477 } while (error == 0 && uio->uio_resid > 0 && n > 0); 478 return (error); 479} 480 481/* 482 * Vnode op for write using bio 483 */ 484int 485nfs_write(ap) 486 struct vop_write_args /* { 487 struct vnode *a_vp; 488 struct uio *a_uio; 489 int a_ioflag; 490 struct ucred *a_cred; 491 } */ *ap; 492{ 493 register int biosize; 494 register struct uio *uio = ap->a_uio; 495 struct proc *p = uio->uio_procp; 496 register struct vnode *vp = ap->a_vp; 497 struct nfsnode *np = VTONFS(vp); 498 register struct ucred *cred = ap->a_cred; 499 int ioflag = ap->a_ioflag; 500 struct buf *bp; 501 struct vattr vattr; 502 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 503 daddr_t lbn; 504 int bufsize; 505 int n, on, error = 0, iomode, must_commit; 506 507#ifdef DIAGNOSTIC 508 if (uio->uio_rw != UIO_WRITE) 509 panic("nfs_write mode"); 510 if (uio->uio_segflg == UIO_USERSPACE && uio->uio_procp != curproc) 511 panic("nfs_write proc"); 512#endif 513 if (vp->v_type != VREG) 514 return (EIO); 515 if (np->n_flag & NWRITEERR) { 516 np->n_flag &= ~NWRITEERR; 517 return (np->n_error); 518 } 519 if ((nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_GOTFSINFO)) == NFSMNT_NFSV3) 520 (void)nfs_fsinfo(nmp, vp, cred, p); 521 if (ioflag & (IO_APPEND | IO_SYNC)) { 522 if (np->n_flag & NMODIFIED) { 523 np->n_attrstamp = 0; 524 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 525 if (error) 526 return (error); 527 } 528 if (ioflag & IO_APPEND) { 529 np->n_attrstamp = 0; 530 error = VOP_GETATTR(vp, &vattr, cred, p); 531 if (error) 532 return (error); 533 uio->uio_offset = np->n_size; 534 } 535 } 536 if (uio->uio_offset < 0) 537 return (EINVAL); 538 if (uio->uio_resid == 0) 539 return (0); 540 /* 541 * Maybe this should be above the vnode op call, but so long as 542 * file servers have no limits, i don't think it matters 543 */ 544 if (p && uio->uio_offset + uio->uio_resid > 545 p->p_rlimit[RLIMIT_FSIZE].rlim_cur) { 546 psignal(p, SIGXFSZ); 547 return (EFBIG); 548 } 549 /* 550 * I use nm_rsize, not nm_wsize so that all buffer cache blocks 551 * will be the same size within a filesystem. nfs_writerpc will 552 * still use nm_wsize when sizing the rpc's. 553 */ 554 biosize = vp->v_mount->mnt_stat.f_iosize; 555 do { 556 /* 557 * Check for a valid write lease. 558 */ 559 if ((nmp->nm_flag & NFSMNT_NQNFS) && 560 NQNFS_CKINVALID(vp, np, ND_WRITE)) { 561 do { 562 error = nqnfs_getlease(vp, ND_WRITE, cred, p); 563 } while (error == NQNFS_EXPIRED); 564 if (error) 565 return (error); 566 if (np->n_lrev != np->n_brev || 567 (np->n_flag & NQNFSNONCACHE)) { 568 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 569 if (error) 570 return (error); 571 np->n_brev = np->n_lrev; 572 } 573 } 574 if ((np->n_flag & NQNFSNONCACHE) && uio->uio_iovcnt == 1) { 575 iomode = NFSV3WRITE_FILESYNC; 576 error = nfs_writerpc(vp, uio, cred, &iomode, &must_commit); 577 if (must_commit) 578 nfs_clearcommit(vp->v_mount); 579 return (error); 580 } 581 nfsstats.biocache_writes++; 582 lbn = uio->uio_offset / biosize; 583 on = uio->uio_offset & (biosize-1); 584 n = min((unsigned)(biosize - on), uio->uio_resid); 585again: 586 if (uio->uio_offset + n > np->n_size) { 587 np->n_size = uio->uio_offset + n; 588 np->n_flag |= NMODIFIED; 589 vnode_pager_setsize(vp, (u_long)np->n_size); 590 } 591 bufsize = biosize; 592 if ((lbn + 1) * biosize > np->n_size) { 593 bufsize = np->n_size - lbn * biosize; 594 bufsize = (bufsize + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1); 595 } 596 bp = nfs_getcacheblk(vp, lbn, bufsize, p); 597 if (!bp) 598 return (EINTR); 599 if (bp->b_wcred == NOCRED) { 600 crhold(cred); 601 bp->b_wcred = cred; 602 } 603 np->n_flag |= NMODIFIED; 604 605 if ((bp->b_blkno * DEV_BSIZE) + bp->b_dirtyend > np->n_size) { 606 bp->b_dirtyend = np->n_size - (bp->b_blkno * DEV_BSIZE); 607 } 608 609 /* 610 * If the new write will leave a contiguous dirty 611 * area, just update the b_dirtyoff and b_dirtyend, 612 * otherwise force a write rpc of the old dirty area. 613 */ 614 if (bp->b_dirtyend > 0 && 615 (on > bp->b_dirtyend || (on + n) < bp->b_dirtyoff)) { 616 bp->b_proc = p; 617 if (VOP_BWRITE(bp) == EINTR) 618 return (EINTR); 619 goto again; 620 } 621 622 /* 623 * Check for valid write lease and get one as required. 624 * In case getblk() and/or bwrite() delayed us. 625 */ 626 if ((nmp->nm_flag & NFSMNT_NQNFS) && 627 NQNFS_CKINVALID(vp, np, ND_WRITE)) { 628 do { 629 error = nqnfs_getlease(vp, ND_WRITE, cred, p); 630 } while (error == NQNFS_EXPIRED); 631 if (error) { 632 brelse(bp); 633 return (error); 634 } 635 if (np->n_lrev != np->n_brev || 636 (np->n_flag & NQNFSNONCACHE)) { 637 brelse(bp); 638 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 639 if (error) 640 return (error); 641 np->n_brev = np->n_lrev; 642 goto again; 643 } 644 } 645 error = uiomove((char *)bp->b_data + on, n, uio); 646 if (error) { 647 bp->b_flags |= B_ERROR; 648 brelse(bp); 649 return (error); 650 } 651 if (bp->b_dirtyend > 0) { 652 bp->b_dirtyoff = min(on, bp->b_dirtyoff); 653 bp->b_dirtyend = max((on + n), bp->b_dirtyend); 654 } else { 655 bp->b_dirtyoff = on; 656 bp->b_dirtyend = on + n; 657 } 658 if (bp->b_validend == 0 || bp->b_validend < bp->b_dirtyoff || 659 bp->b_validoff > bp->b_dirtyend) { 660 bp->b_validoff = bp->b_dirtyoff; 661 bp->b_validend = bp->b_dirtyend; 662 } else { 663 bp->b_validoff = min(bp->b_validoff, bp->b_dirtyoff); 664 bp->b_validend = max(bp->b_validend, bp->b_dirtyend); 665 } 666 667 /* 668 * Since this block is being modified, it must be written 669 * again and not just committed. 670 */ 671 bp->b_flags &= ~B_NEEDCOMMIT; 672 673 /* 674 * If the lease is non-cachable or IO_SYNC do bwrite(). 675 */ 676 if ((np->n_flag & NQNFSNONCACHE) || (ioflag & IO_SYNC)) { 677 bp->b_proc = p; 678 error = VOP_BWRITE(bp); 679 if (error) 680 return (error); 681 if (np->n_flag & NQNFSNONCACHE) { 682 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 683 if (error) 684 return (error); 685 } 686 } else if ((n + on) == biosize && 687 (nmp->nm_flag & NFSMNT_NQNFS) == 0) { 688 bp->b_proc = (struct proc *)0; 689 bp->b_flags |= B_ASYNC; 690 (void)nfs_writebp(bp, 0); 691 } else 692 bdwrite(bp); 693 } while (uio->uio_resid > 0 && n > 0); 694 return (0); 695} 696 697/* 698 * Get an nfs cache block. 699 * Allocate a new one if the block isn't currently in the cache 700 * and return the block marked busy. If the calling process is 701 * interrupted by a signal for an interruptible mount point, return 702 * NULL. 703 */ 704static struct buf * 705nfs_getcacheblk(vp, bn, size, p) 706 struct vnode *vp; 707 daddr_t bn; 708 int size; 709 struct proc *p; 710{ 711 register struct buf *bp; 712 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 713 int biosize = vp->v_mount->mnt_stat.f_iosize; 714 715 if (nmp->nm_flag & NFSMNT_INT) { 716 bp = getblk(vp, bn, size, PCATCH, 0); 717 while (bp == (struct buf *)0) { 718 if (nfs_sigintr(nmp, (struct nfsreq *)0, p)) 719 return ((struct buf *)0); 720 bp = getblk(vp, bn, size, 0, 2 * hz); 721 } 722 } else 723 bp = getblk(vp, bn, size, 0, 0); 724 725 if( vp->v_type == VREG) 726 bp->b_blkno = (bn * biosize) / DEV_BSIZE; 727 728 return (bp); 729} 730 731/* 732 * Flush and invalidate all dirty buffers. If another process is already 733 * doing the flush, just wait for completion. 734 */ 735int 736nfs_vinvalbuf(vp, flags, cred, p, intrflg) 737 struct vnode *vp; 738 int flags; 739 struct ucred *cred; 740 struct proc *p; 741 int intrflg; 742{ 743 register struct nfsnode *np = VTONFS(vp); 744 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 745 int error = 0, slpflag, slptimeo; 746 747 if ((nmp->nm_flag & NFSMNT_INT) == 0) 748 intrflg = 0; 749 if (intrflg) { 750 slpflag = PCATCH; 751 slptimeo = 2 * hz; 752 } else { 753 slpflag = 0; 754 slptimeo = 0; 755 } 756 /* 757 * First wait for any other process doing a flush to complete. 758 */ 759 while (np->n_flag & NFLUSHINPROG) { 760 np->n_flag |= NFLUSHWANT; 761 error = tsleep((caddr_t)&np->n_flag, PRIBIO + 2, "nfsvinval", 762 slptimeo); 763 if (error && intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) 764 return (EINTR); 765 } 766 767 /* 768 * Now, flush as required. 769 */ 770 np->n_flag |= NFLUSHINPROG; 771 error = vinvalbuf(vp, flags, cred, p, slpflag, 0); 772 while (error) { 773 if (intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) { 774 np->n_flag &= ~NFLUSHINPROG; 775 if (np->n_flag & NFLUSHWANT) { 776 np->n_flag &= ~NFLUSHWANT; 777 wakeup((caddr_t)&np->n_flag); 778 } 779 return (EINTR); 780 } 781 error = vinvalbuf(vp, flags, cred, p, 0, slptimeo); 782 } 783 np->n_flag &= ~(NMODIFIED | NFLUSHINPROG); 784 if (np->n_flag & NFLUSHWANT) { 785 np->n_flag &= ~NFLUSHWANT; 786 wakeup((caddr_t)&np->n_flag); 787 } 788 return (0); 789} 790 791/* 792 * Initiate asynchronous I/O. Return an error if no nfsiods are available. 793 * This is mainly to avoid queueing async I/O requests when the nfsiods 794 * are all hung on a dead server. 795 */ 796int 797nfs_asyncio(bp, cred) 798 register struct buf *bp; 799 struct ucred *cred; 800{ 801 struct nfsmount *nmp; 802 int i; 803 int gotiod; 804 int slpflag = 0; 805 int slptimeo = 0; 806 int error; 807 808 if (nfs_numasync == 0) 809 return (EIO); 810 811 nmp = VFSTONFS(bp->b_vp->v_mount); 812again: 813 if (nmp->nm_flag & NFSMNT_INT) 814 slpflag = PCATCH; 815 gotiod = FALSE; 816 817 /* 818 * Find a free iod to process this request. 819 */ 820 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 821 if (nfs_iodwant[i]) { 822 /* 823 * Found one, so wake it up and tell it which 824 * mount to process. 825 */ 826 NFS_DPF(ASYNCIO, 827 ("nfs_asyncio: waking iod %d for mount %p\n", 828 i, nmp)); 829 nfs_iodwant[i] = (struct proc *)0; 830 nfs_iodmount[i] = nmp; 831 nmp->nm_bufqiods++; 832 wakeup((caddr_t)&nfs_iodwant[i]); 833 gotiod = TRUE; 834 break; 835 } 836 837 /* 838 * If none are free, we may already have an iod working on this mount 839 * point. If so, it will process our request. 840 */ 841 if (!gotiod) { 842 if (nmp->nm_bufqiods > 0) { 843 NFS_DPF(ASYNCIO, 844 ("nfs_asyncio: %d iods are already processing mount %p\n", 845 nmp->nm_bufqiods, nmp)); 846 gotiod = TRUE; 847 } 848 } 849 850 /* 851 * If we have an iod which can process the request, then queue 852 * the buffer. 853 */ 854 if (gotiod) { 855 /* 856 * Ensure that the queue never grows too large. 857 */ 858 while (nmp->nm_bufqlen >= 2*nfs_numasync) { 859 NFS_DPF(ASYNCIO, 860 ("nfs_asyncio: waiting for mount %p queue to drain\n", nmp)); 861 nmp->nm_bufqwant = TRUE; 862 error = tsleep(&nmp->nm_bufq, slpflag | PRIBIO, 863 "nfsaio", slptimeo); 864 if (error) { 865 if (nfs_sigintr(nmp, NULL, bp->b_proc)) 866 return (EINTR); 867 if (slpflag == PCATCH) { 868 slpflag = 0; 869 slptimeo = 2 * hz; 870 } 871 } 872 /* 873 * We might have lost our iod while sleeping, 874 * so check and loop if nescessary. 875 */ 876 if (nmp->nm_bufqiods == 0) { 877 NFS_DPF(ASYNCIO, 878 ("nfs_asyncio: no iods after mount %p queue was drained, looping\n", nmp)); 879 goto again; 880 } 881 } 882 883 if (bp->b_flags & B_READ) { 884 if (bp->b_rcred == NOCRED && cred != NOCRED) { 885 crhold(cred); 886 bp->b_rcred = cred; 887 } 888 } else { 889 bp->b_flags |= B_WRITEINPROG; 890 if (bp->b_wcred == NOCRED && cred != NOCRED) { 891 crhold(cred); 892 bp->b_wcred = cred; 893 } 894 } 895 896 TAILQ_INSERT_TAIL(&nmp->nm_bufq, bp, b_freelist); 897 nmp->nm_bufqlen++; 898 return (0); 899 } 900 901 /* 902 * All the iods are busy on other mounts, so return EIO to 903 * force the caller to process the i/o synchronously. 904 */ 905 NFS_DPF(ASYNCIO, ("nfs_asyncio: no iods available, i/o is synchronous\n")); 906 return (EIO); 907} 908 909/* 910 * Do an I/O operation to/from a cache block. This may be called 911 * synchronously or from an nfsiod. 912 */ 913int 914nfs_doio(bp, cr, p) 915 register struct buf *bp; 916 struct ucred *cr; 917 struct proc *p; 918{ 919 register struct uio *uiop; 920 register struct vnode *vp; 921 struct nfsnode *np; 922 struct nfsmount *nmp; 923 int error = 0, diff, len, iomode, must_commit = 0; 924 struct uio uio; 925 struct iovec io; 926 927 vp = bp->b_vp; 928 np = VTONFS(vp); 929 nmp = VFSTONFS(vp->v_mount); 930 uiop = &uio; 931 uiop->uio_iov = &io; 932 uiop->uio_iovcnt = 1; 933 uiop->uio_segflg = UIO_SYSSPACE; 934 uiop->uio_procp = p; 935 936 /* 937 * Historically, paging was done with physio, but no more. 938 */ 939 if (bp->b_flags & B_PHYS) { 940 /* 941 * ...though reading /dev/drum still gets us here. 942 */ 943 io.iov_len = uiop->uio_resid = bp->b_bcount; 944 /* mapping was done by vmapbuf() */ 945 io.iov_base = bp->b_data; 946 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; 947 if (bp->b_flags & B_READ) { 948 uiop->uio_rw = UIO_READ; 949 nfsstats.read_physios++; 950 error = nfs_readrpc(vp, uiop, cr); 951 } else { 952 int com; 953 954 iomode = NFSV3WRITE_DATASYNC; 955 uiop->uio_rw = UIO_WRITE; 956 nfsstats.write_physios++; 957 error = nfs_writerpc(vp, uiop, cr, &iomode, &com); 958 } 959 if (error) { 960 bp->b_flags |= B_ERROR; 961 bp->b_error = error; 962 } 963 } else if (bp->b_flags & B_READ) { 964 io.iov_len = uiop->uio_resid = bp->b_bcount; 965 io.iov_base = bp->b_data; 966 uiop->uio_rw = UIO_READ; 967 switch (vp->v_type) { 968 case VREG: 969 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; 970 nfsstats.read_bios++; 971 error = nfs_readrpc(vp, uiop, cr); 972 if (!error) { 973 bp->b_validoff = 0; 974 if (uiop->uio_resid) { 975 /* 976 * If len > 0, there is a hole in the file and 977 * no writes after the hole have been pushed to 978 * the server yet. 979 * Just zero fill the rest of the valid area. 980 */ 981 diff = bp->b_bcount - uiop->uio_resid; 982 len = np->n_size - (((u_quad_t)bp->b_blkno) * DEV_BSIZE 983 + diff); 984 if (len > 0) { 985 len = min(len, uiop->uio_resid); 986 bzero((char *)bp->b_data + diff, len); 987 bp->b_validend = diff + len; 988 } else 989 bp->b_validend = diff; 990 } else 991 bp->b_validend = bp->b_bcount; 992 } 993 if (p && (vp->v_flag & VTEXT) && 994 (((nmp->nm_flag & NFSMNT_NQNFS) && 995 NQNFS_CKINVALID(vp, np, ND_READ) && 996 np->n_lrev != np->n_brev) || 997 (!(nmp->nm_flag & NFSMNT_NQNFS) && 998 np->n_mtime != np->n_vattr.va_mtime.tv_sec))) { 999 uprintf("Process killed due to text file modification\n"); 1000 psignal(p, SIGKILL); 1001 p->p_flag |= P_NOSWAP; 1002 } 1003 break; 1004 case VLNK: 1005 uiop->uio_offset = (off_t)0; 1006 nfsstats.readlink_bios++; 1007 error = nfs_readlinkrpc(vp, uiop, cr); 1008 break; 1009 case VDIR: 1010 nfsstats.readdir_bios++; 1011 uiop->uio_offset = ((u_quad_t)bp->b_lblkno) * NFS_DIRBLKSIZ; 1012 if (nmp->nm_flag & NFSMNT_RDIRPLUS) { 1013 error = nfs_readdirplusrpc(vp, uiop, cr); 1014 if (error == NFSERR_NOTSUPP) 1015 nmp->nm_flag &= ~NFSMNT_RDIRPLUS; 1016 } 1017 if ((nmp->nm_flag & NFSMNT_RDIRPLUS) == 0) 1018 error = nfs_readdirrpc(vp, uiop, cr); 1019 break; 1020 default: 1021 printf("nfs_doio: type %x unexpected\n",vp->v_type); 1022 break; 1023 }; 1024 if (error) { 1025 bp->b_flags |= B_ERROR; 1026 bp->b_error = error; 1027 } 1028 } else { 1029 if (((bp->b_blkno * DEV_BSIZE) + bp->b_dirtyend) > np->n_size) 1030 bp->b_dirtyend = np->n_size - (bp->b_blkno * DEV_BSIZE); 1031 1032 if (bp->b_dirtyend > bp->b_dirtyoff) { 1033 io.iov_len = uiop->uio_resid = bp->b_dirtyend 1034 - bp->b_dirtyoff; 1035 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE 1036 + bp->b_dirtyoff; 1037 io.iov_base = (char *)bp->b_data + bp->b_dirtyoff; 1038 uiop->uio_rw = UIO_WRITE; 1039 nfsstats.write_bios++; 1040 if ((bp->b_flags & (B_ASYNC | B_NEEDCOMMIT | B_NOCACHE | B_CLUSTER)) == B_ASYNC) 1041 iomode = NFSV3WRITE_UNSTABLE; 1042 else 1043 iomode = NFSV3WRITE_FILESYNC; 1044 bp->b_flags |= B_WRITEINPROG; 1045 error = nfs_writerpc(vp, uiop, cr, &iomode, &must_commit); 1046 if (!error && iomode == NFSV3WRITE_UNSTABLE) { 1047 bp->b_flags |= B_NEEDCOMMIT; 1048 if (bp->b_dirtyoff == 0 1049 && bp->b_dirtyend == bp->b_bufsize) 1050 bp->b_flags |= B_CLUSTEROK; 1051 } else 1052 bp->b_flags &= ~B_NEEDCOMMIT; 1053 bp->b_flags &= ~B_WRITEINPROG; 1054 1055 /* 1056 * For an interrupted write, the buffer is still valid 1057 * and the write hasn't been pushed to the server yet, 1058 * so we can't set B_ERROR and report the interruption 1059 * by setting B_EINTR. For the B_ASYNC case, B_EINTR 1060 * is not relevant, so the rpc attempt is essentially 1061 * a noop. For the case of a V3 write rpc not being 1062 * committed to stable storage, the block is still 1063 * dirty and requires either a commit rpc or another 1064 * write rpc with iomode == NFSV3WRITE_FILESYNC before 1065 * the block is reused. This is indicated by setting 1066 * the B_DELWRI and B_NEEDCOMMIT flags. 1067 */ 1068 if (error == EINTR 1069 || (!error && (bp->b_flags & B_NEEDCOMMIT))) { 1070 bp->b_flags &= ~(B_INVAL|B_NOCACHE); 1071 ++numdirtybuffers; 1072 bp->b_flags |= B_DELWRI; 1073 1074 /* 1075 * Since for the B_ASYNC case, nfs_bwrite() has reassigned the 1076 * buffer to the clean list, we have to reassign it back to the 1077 * dirty one. Ugh. 1078 */ 1079 if (bp->b_flags & B_ASYNC) 1080 reassignbuf(bp, vp); 1081 else 1082 bp->b_flags |= B_EINTR; 1083 } else { 1084 if (error) { 1085 bp->b_flags |= B_ERROR; 1086 bp->b_error = np->n_error = error; 1087 np->n_flag |= NWRITEERR; 1088 } 1089 bp->b_dirtyoff = bp->b_dirtyend = 0; 1090 } 1091 } else { 1092 bp->b_resid = 0; 1093 biodone(bp); 1094 return (0); 1095 } 1096 } 1097 bp->b_resid = uiop->uio_resid; 1098 if (must_commit) 1099 nfs_clearcommit(vp->v_mount); 1100 biodone(bp); 1101 return (error); 1102} 1103