1 /* $OpenBSD: nfs_bio.c,v 1.68 2009/10/19 22:24:18 jsg Exp $ */ 2 /* $NetBSD: nfs_bio.c,v 1.25.4.2 1996/07/08 20:47:04 jtc Exp $ */ 3 4 /* 5 * Copyright (c) 1989, 1993 6 * The Regents of the University of California. All rights reserved. 7 * 8 * This code is derived from software contributed to Berkeley by 9 * Rick Macklem at The University of Guelph. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. Neither the name of the University nor the names of its contributors 20 * may be used to endorse or promote products derived from this software 21 * without specific prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 * 35 * @(#)nfs_bio.c 8.9 (Berkeley) 3/30/95 36 */ 37 38 #include <sys/param.h> 39 #include <sys/systm.h> 40 #include <sys/resourcevar.h> 41 #include <sys/signalvar.h> 42 #include <sys/proc.h> 43 #include <sys/buf.h> 44 #include <sys/vnode.h> 45 #include <sys/mount.h> 46 #include <sys/kernel.h> 47 #include <sys/namei.h> 48 #include <sys/queue.h> 49 #include <sys/time.h> 50 51 #include <uvm/uvm_extern.h> 52 53 #include <nfs/rpcv2.h> 54 #include <nfs/nfsproto.h> 55 #include <nfs/nfs.h> 56 #include <nfs/nfsmount.h> 57 #include <nfs/nfsnode.h> 58 #include <nfs/nfs_var.h> 59 60 extern int nfs_numasync; 61 extern struct nfsstats nfsstats; 62 struct nfs_bufqhead nfs_bufq; 63 uint32_t nfs_bufqmax, nfs_bufqlen; 64 65 /* 66 * Vnode op for read using bio 67 * Any similarity to readip() is purely coincidental 68 */ 69 int 70 nfs_bioread(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *cred) 71 { 72 struct nfsnode *np = VTONFS(vp); 73 int biosize, diff; 74 struct buf *bp = NULL, *rabp; 75 struct vattr vattr; 76 struct proc *p; 77 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 78 daddr64_t lbn, bn, rabn; 79 caddr_t baddr; 80 int got_buf = 0, nra, error = 0, n = 0, on = 0, not_readin; 81 off_t offdiff; 82 83 #ifdef DIAGNOSTIC 84 if (uio->uio_rw != UIO_READ) 85 panic("nfs_read mode"); 86 #endif 87 if (uio->uio_resid == 0) 88 return (0); 89 if (uio->uio_offset < 0) 90 return (EINVAL); 91 p = uio->uio_procp; 92 if ((nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_GOTFSINFO)) == NFSMNT_NFSV3) 93 (void)nfs_fsinfo(nmp, vp, cred, p); 94 biosize = nmp->nm_rsize; 95 /* 96 * For nfs, cache consistency can only be maintained approximately. 97 * Although RFC1094 does not specify the criteria, the following is 98 * believed to be compatible with the reference port. 99 * For nfs: 100 * If the file's modify time on the server has changed since the 101 * last read rpc or you have written to the file, 102 * you may have lost data cache consistency with the 103 * server, so flush all of the file's data out of the cache. 104 * Then force a getattr rpc to ensure that you have up to date 105 * attributes. 106 */ 107 if (np->n_flag & NMODIFIED) { 108 NFS_INVALIDATE_ATTRCACHE(np); 109 error = VOP_GETATTR(vp, &vattr, cred, p); 110 if (error) 111 return (error); 112 np->n_mtime = vattr.va_mtime; 113 } else { 114 error = VOP_GETATTR(vp, &vattr, cred, p); 115 if (error) 116 return (error); 117 if (timespeccmp(&np->n_mtime, &vattr.va_mtime, !=)) { 118 error = nfs_vinvalbuf(vp, V_SAVE, cred, p); 119 if (error) 120 return (error); 121 np->n_mtime = vattr.va_mtime; 122 } 123 } 124 125 /* 126 * update the cache read creds for this vnode 127 */ 128 if (np->n_rcred) 129 crfree(np->n_rcred); 130 np->n_rcred = cred; 131 crhold(cred); 132 133 do { 134 if ((vp->v_flag & VROOT) && vp->v_type == VLNK) { 135 return (nfs_readlinkrpc(vp, uio, cred)); 136 } 137 baddr = (caddr_t)0; 138 switch (vp->v_type) { 139 case VREG: 140 nfsstats.biocache_reads++; 141 lbn = uio->uio_offset / biosize; 142 on = uio->uio_offset & (biosize - 1); 143 bn = lbn * (biosize / DEV_BSIZE); 144 not_readin = 1; 145 146 /* 147 * Start the read ahead(s), as required. 148 */ 149 if (nfs_numasync > 0 && nmp->nm_readahead > 0) { 150 for (nra = 0; nra < nmp->nm_readahead && 151 (lbn + 1 + nra) * biosize < np->n_size; nra++) { 152 rabn = (lbn + 1 + nra) * (biosize / DEV_BSIZE); 153 if (!incore(vp, rabn)) { 154 rabp = nfs_getcacheblk(vp, rabn, biosize, p); 155 if (!rabp) 156 return (EINTR); 157 if ((rabp->b_flags & (B_DELWRI | B_DONE)) == 0) { 158 rabp->b_flags |= (B_READ | B_ASYNC); 159 if (nfs_asyncio(rabp)) { 160 rabp->b_flags |= B_INVAL; 161 brelse(rabp); 162 } 163 } else 164 brelse(rabp); 165 } 166 } 167 } 168 169 again: 170 bp = nfs_getcacheblk(vp, bn, biosize, p); 171 if (!bp) 172 return (EINTR); 173 got_buf = 1; 174 if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0) { 175 bp->b_flags |= B_READ; 176 not_readin = 0; 177 error = nfs_doio(bp, p); 178 if (error) { 179 brelse(bp); 180 return (error); 181 } 182 } 183 n = min((unsigned)(biosize - on), uio->uio_resid); 184 offdiff = np->n_size - uio->uio_offset; 185 if (offdiff < (off_t)n) 186 n = (int)offdiff; 187 if (not_readin && n > 0) { 188 if (on < bp->b_validoff || (on + n) > bp->b_validend) { 189 bp->b_flags |= B_INVAFTERWRITE; 190 if (bp->b_dirtyend > 0) { 191 if ((bp->b_flags & B_DELWRI) == 0) 192 panic("nfsbioread"); 193 if (VOP_BWRITE(bp) == EINTR) 194 return (EINTR); 195 } else 196 brelse(bp); 197 goto again; 198 } 199 } 200 diff = (on >= bp->b_validend) ? 0 : (bp->b_validend - on); 201 if (diff < n) 202 n = diff; 203 break; 204 case VLNK: 205 nfsstats.biocache_readlinks++; 206 bp = nfs_getcacheblk(vp, 0, NFS_MAXPATHLEN, p); 207 if (!bp) 208 return (EINTR); 209 if ((bp->b_flags & B_DONE) == 0) { 210 bp->b_flags |= B_READ; 211 error = nfs_doio(bp, p); 212 if (error) { 213 brelse(bp); 214 return (error); 215 } 216 } 217 n = min(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid); 218 got_buf = 1; 219 on = 0; 220 break; 221 default: 222 printf(" nfsbioread: type %x unexpected\n",vp->v_type); 223 break; 224 } 225 226 if (n > 0) { 227 if (!baddr) 228 baddr = bp->b_data; 229 error = uiomove(baddr + on, (int)n, uio); 230 } 231 switch (vp->v_type) { 232 case VREG: 233 break; 234 case VLNK: 235 n = 0; 236 break; 237 default: 238 printf(" nfsbioread: type %x unexpected\n",vp->v_type); 239 } 240 if (got_buf) 241 brelse(bp); 242 } while (error == 0 && uio->uio_resid > 0 && n > 0); 243 return (error); 244 } 245 246 /* 247 * Vnode op for write using bio 248 */ 249 int 250 nfs_write(void *v) 251 { 252 struct vop_write_args *ap = v; 253 int biosize; 254 struct uio *uio = ap->a_uio; 255 struct proc *p = uio->uio_procp; 256 struct vnode *vp = ap->a_vp; 257 struct nfsnode *np = VTONFS(vp); 258 struct ucred *cred = ap->a_cred; 259 int ioflag = ap->a_ioflag; 260 struct buf *bp; 261 struct vattr vattr; 262 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 263 daddr64_t lbn, bn; 264 int n, on, error = 0, extended = 0, wrotedta = 0, truncated = 0; 265 266 #ifdef DIAGNOSTIC 267 if (uio->uio_rw != UIO_WRITE) 268 panic("nfs_write mode"); 269 if (uio->uio_segflg == UIO_USERSPACE && uio->uio_procp != curproc) 270 panic("nfs_write proc"); 271 #endif 272 if (vp->v_type != VREG) 273 return (EIO); 274 if (np->n_flag & NWRITEERR) { 275 np->n_flag &= ~NWRITEERR; 276 return (np->n_error); 277 } 278 if ((nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_GOTFSINFO)) == NFSMNT_NFSV3) 279 (void)nfs_fsinfo(nmp, vp, cred, p); 280 if (ioflag & (IO_APPEND | IO_SYNC)) { 281 if (np->n_flag & NMODIFIED) { 282 NFS_INVALIDATE_ATTRCACHE(np); 283 error = nfs_vinvalbuf(vp, V_SAVE, cred, p); 284 if (error) 285 return (error); 286 } 287 if (ioflag & IO_APPEND) { 288 NFS_INVALIDATE_ATTRCACHE(np); 289 error = VOP_GETATTR(vp, &vattr, cred, p); 290 if (error) 291 return (error); 292 uio->uio_offset = np->n_size; 293 } 294 } 295 if (uio->uio_offset < 0) 296 return (EINVAL); 297 if (uio->uio_resid == 0) 298 return (0); 299 /* 300 * Maybe this should be above the vnode op call, but so long as 301 * file servers have no limits, i don't think it matters 302 */ 303 if (p && uio->uio_offset + uio->uio_resid > 304 p->p_rlimit[RLIMIT_FSIZE].rlim_cur) { 305 psignal(p, SIGXFSZ); 306 return (EFBIG); 307 } 308 309 /* 310 * update the cache write creds for this node. 311 */ 312 if (np->n_wcred) 313 crfree(np->n_wcred); 314 np->n_wcred = cred; 315 crhold(cred); 316 317 /* 318 * I use nm_rsize, not nm_wsize so that all buffer cache blocks 319 * will be the same size within a filesystem. nfs_writerpc will 320 * still use nm_wsize when sizing the rpc's. 321 */ 322 biosize = nmp->nm_rsize; 323 do { 324 325 /* 326 * XXX make sure we aren't cached in the VM page cache 327 */ 328 uvm_vnp_uncache(vp); 329 330 nfsstats.biocache_writes++; 331 lbn = uio->uio_offset / biosize; 332 on = uio->uio_offset & (biosize-1); 333 n = min((unsigned)(biosize - on), uio->uio_resid); 334 bn = lbn * (biosize / DEV_BSIZE); 335 again: 336 bp = nfs_getcacheblk(vp, bn, biosize, p); 337 if (!bp) 338 return (EINTR); 339 np->n_flag |= NMODIFIED; 340 if (uio->uio_offset + n > np->n_size) { 341 np->n_size = uio->uio_offset + n; 342 uvm_vnp_setsize(vp, (u_long)np->n_size); 343 extended = 1; 344 } else if (uio->uio_offset + n < np->n_size) 345 truncated = 1; 346 347 /* 348 * If the new write will leave a contiguous dirty 349 * area, just update the b_dirtyoff and b_dirtyend, 350 * otherwise force a write rpc of the old dirty area. 351 */ 352 if (bp->b_dirtyend > 0 && 353 (on > bp->b_dirtyend || (on + n) < bp->b_dirtyoff)) { 354 bp->b_proc = p; 355 if (VOP_BWRITE(bp) == EINTR) 356 return (EINTR); 357 goto again; 358 } 359 360 error = uiomove((char *)bp->b_data + on, n, uio); 361 if (error) { 362 bp->b_flags |= B_ERROR; 363 brelse(bp); 364 return (error); 365 } 366 if (bp->b_dirtyend > 0) { 367 bp->b_dirtyoff = min(on, bp->b_dirtyoff); 368 bp->b_dirtyend = max((on + n), bp->b_dirtyend); 369 } else { 370 bp->b_dirtyoff = on; 371 bp->b_dirtyend = on + n; 372 } 373 if (bp->b_validend == 0 || bp->b_validend < bp->b_dirtyoff || 374 bp->b_validoff > bp->b_dirtyend) { 375 bp->b_validoff = bp->b_dirtyoff; 376 bp->b_validend = bp->b_dirtyend; 377 } else { 378 bp->b_validoff = min(bp->b_validoff, bp->b_dirtyoff); 379 bp->b_validend = max(bp->b_validend, bp->b_dirtyend); 380 } 381 382 wrotedta = 1; 383 384 /* 385 * Since this block is being modified, it must be written 386 * again and not just committed. 387 */ 388 389 if (NFS_ISV3(vp)) { 390 rw_enter_write(&np->n_commitlock); 391 if (bp->b_flags & B_NEEDCOMMIT) { 392 bp->b_flags &= ~B_NEEDCOMMIT; 393 nfs_del_tobecommitted_range(vp, bp); 394 } 395 nfs_del_committed_range(vp, bp); 396 rw_exit_write(&np->n_commitlock); 397 } else 398 bp->b_flags &= ~B_NEEDCOMMIT; 399 400 if (ioflag & IO_SYNC) { 401 bp->b_proc = p; 402 error = VOP_BWRITE(bp); 403 if (error) 404 return (error); 405 } else if ((n + on) == biosize) { 406 bp->b_proc = NULL; 407 bp->b_flags |= B_ASYNC; 408 (void)nfs_writebp(bp, 0); 409 } else { 410 bdwrite(bp); 411 } 412 } while (uio->uio_resid > 0 && n > 0); 413 414 if (wrotedta) 415 VN_KNOTE(vp, NOTE_WRITE | (extended ? NOTE_EXTEND : 0) | 416 (truncated ? NOTE_TRUNCATE : 0)); 417 418 return (0); 419 } 420 421 /* 422 * Get an nfs cache block. 423 * Allocate a new one if the block isn't currently in the cache 424 * and return the block marked busy. If the calling process is 425 * interrupted by a signal for an interruptible mount point, return 426 * NULL. 427 */ 428 struct buf * 429 nfs_getcacheblk(struct vnode *vp, daddr64_t bn, int size, struct proc *p) 430 { 431 struct buf *bp; 432 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 433 434 if (nmp->nm_flag & NFSMNT_INT) { 435 bp = getblk(vp, bn, size, PCATCH, 0); 436 while (bp == NULL) { 437 if (nfs_sigintr(nmp, NULL, p)) 438 return (NULL); 439 bp = getblk(vp, bn, size, 0, 2 * hz); 440 } 441 } else 442 bp = getblk(vp, bn, size, 0, 0); 443 return (bp); 444 } 445 446 /* 447 * Flush and invalidate all dirty buffers. If another process is already 448 * doing the flush, just wait for completion. 449 */ 450 int 451 nfs_vinvalbuf(struct vnode *vp, int flags, struct ucred *cred, struct proc *p) 452 { 453 struct nfsmount *nmp= VFSTONFS(vp->v_mount); 454 struct nfsnode *np = VTONFS(vp); 455 int error, sintr, stimeo; 456 457 error = sintr = stimeo = 0; 458 459 if (ISSET(nmp->nm_flag, NFSMNT_INT)) { 460 sintr = PCATCH; 461 stimeo = 2 * hz; 462 } 463 464 /* First wait for any other process doing a flush to complete. */ 465 while (np->n_flag & NFLUSHINPROG) { 466 np->n_flag |= NFLUSHWANT; 467 error = tsleep(&np->n_flag, PRIBIO|sintr, "nfsvinval", stimeo); 468 if (error && sintr && nfs_sigintr(nmp, NULL, p)) 469 return (EINTR); 470 } 471 472 /* Now, flush as required. */ 473 np->n_flag |= NFLUSHINPROG; 474 error = vinvalbuf(vp, flags, cred, p, sintr, 0); 475 while (error) { 476 if (sintr && nfs_sigintr(nmp, NULL, p)) { 477 np->n_flag &= ~NFLUSHINPROG; 478 if (np->n_flag & NFLUSHWANT) { 479 np->n_flag &= ~NFLUSHWANT; 480 wakeup(&np->n_flag); 481 } 482 return (EINTR); 483 } 484 error = vinvalbuf(vp, flags, cred, p, 0, stimeo); 485 } 486 np->n_flag &= ~(NMODIFIED | NFLUSHINPROG); 487 if (np->n_flag & NFLUSHWANT) { 488 np->n_flag &= ~NFLUSHWANT; 489 wakeup(&np->n_flag); 490 } 491 return (0); 492 } 493 494 /* 495 * Initiate asynchronous I/O. Return an error if no nfsiods are available. 496 * This is mainly to avoid queueing async I/O requests when the nfsiods 497 * are all hung on a dead server. 498 */ 499 int 500 nfs_asyncio(struct buf *bp) 501 { 502 if (nfs_numasync == 0) 503 goto out; 504 505 if (nfs_bufqlen > nfs_bufqmax) 506 goto out; /* too many bufs in use, force sync */ 507 508 if ((bp->b_flags & B_READ) == 0) { 509 bp->b_flags |= B_WRITEINPROG; 510 } 511 512 TAILQ_INSERT_TAIL(&nfs_bufq, bp, b_freelist); 513 nfs_bufqlen++; 514 515 wakeup_one(&nfs_bufq); 516 return (0); 517 518 out: 519 nfsstats.forcedsync++; 520 return (EIO); 521 } 522 523 /* 524 * Do an I/O operation to/from a cache block. This may be called 525 * synchronously or from an nfsiod. 526 */ 527 int 528 nfs_doio(struct buf *bp, struct proc *p) 529 { 530 struct uio *uiop; 531 struct vnode *vp; 532 struct nfsnode *np; 533 struct nfsmount *nmp; 534 int s, error = 0, diff, len, iomode, must_commit = 0; 535 struct uio uio; 536 struct iovec io; 537 538 vp = bp->b_vp; 539 np = VTONFS(vp); 540 nmp = VFSTONFS(vp->v_mount); 541 uiop = &uio; 542 uiop->uio_iov = &io; 543 uiop->uio_iovcnt = 1; 544 uiop->uio_segflg = UIO_SYSSPACE; 545 uiop->uio_procp = p; 546 547 /* 548 * Historically, paging was done with physio, but no more. 549 */ 550 if (bp->b_flags & B_PHYS) { 551 io.iov_len = uiop->uio_resid = bp->b_bcount; 552 /* mapping was done by vmapbuf() */ 553 io.iov_base = bp->b_data; 554 uiop->uio_offset = ((off_t)bp->b_blkno) << DEV_BSHIFT; 555 if (bp->b_flags & B_READ) { 556 uiop->uio_rw = UIO_READ; 557 nfsstats.read_physios++; 558 error = nfs_readrpc(vp, uiop); 559 } else { 560 iomode = NFSV3WRITE_DATASYNC; 561 uiop->uio_rw = UIO_WRITE; 562 nfsstats.write_physios++; 563 error = nfs_writerpc(vp, uiop, &iomode, &must_commit); 564 } 565 if (error) { 566 bp->b_flags |= B_ERROR; 567 bp->b_error = error; 568 } 569 } else if (bp->b_flags & B_READ) { 570 io.iov_len = uiop->uio_resid = bp->b_bcount; 571 io.iov_base = bp->b_data; 572 uiop->uio_rw = UIO_READ; 573 switch (vp->v_type) { 574 case VREG: 575 uiop->uio_offset = ((off_t)bp->b_blkno) << DEV_BSHIFT; 576 nfsstats.read_bios++; 577 bcstats.pendingreads++; 578 bcstats.numreads++; 579 error = nfs_readrpc(vp, uiop); 580 if (!error) { 581 bp->b_validoff = 0; 582 if (uiop->uio_resid) { 583 /* 584 * If len > 0, there is a hole in the file and 585 * no writes after the hole have been pushed to 586 * the server yet. 587 * Just zero fill the rest of the valid area. 588 */ 589 diff = bp->b_bcount - uiop->uio_resid; 590 len = np->n_size - ((((off_t)bp->b_blkno) << DEV_BSHIFT) 591 + diff); 592 if (len > 0) { 593 len = min(len, uiop->uio_resid); 594 bzero((char *)bp->b_data + diff, len); 595 bp->b_validend = diff + len; 596 } else 597 bp->b_validend = diff; 598 } else 599 bp->b_validend = bp->b_bcount; 600 } 601 if (p && (vp->v_flag & VTEXT) && 602 (timespeccmp(&np->n_mtime, &np->n_vattr.va_mtime, !=))) { 603 uprintf("Process killed due to text file modification\n"); 604 psignal(p, SIGKILL); 605 } 606 break; 607 case VLNK: 608 uiop->uio_offset = (off_t)0; 609 nfsstats.readlink_bios++; 610 bcstats.pendingreads++; 611 bcstats.numreads++; 612 error = nfs_readlinkrpc(vp, uiop, curproc->p_ucred); 613 break; 614 default: 615 printf("nfs_doio: type %x unexpected\n", vp->v_type); 616 break; 617 }; 618 if (error) { 619 bp->b_flags |= B_ERROR; 620 bp->b_error = error; 621 } 622 } else { 623 io.iov_len = uiop->uio_resid = bp->b_dirtyend 624 - bp->b_dirtyoff; 625 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE 626 + bp->b_dirtyoff; 627 io.iov_base = (char *)bp->b_data + bp->b_dirtyoff; 628 uiop->uio_rw = UIO_WRITE; 629 nfsstats.write_bios++; 630 bcstats.pendingwrites++; 631 bcstats.numwrites++; 632 if ((bp->b_flags & (B_ASYNC | B_NEEDCOMMIT | B_NOCACHE)) == B_ASYNC) 633 iomode = NFSV3WRITE_UNSTABLE; 634 else 635 iomode = NFSV3WRITE_FILESYNC; 636 bp->b_flags |= B_WRITEINPROG; 637 error = nfs_writerpc(vp, uiop, &iomode, &must_commit); 638 639 rw_enter_write(&np->n_commitlock); 640 if (!error && iomode == NFSV3WRITE_UNSTABLE) { 641 bp->b_flags |= B_NEEDCOMMIT; 642 nfs_add_tobecommitted_range(vp, bp); 643 } else { 644 bp->b_flags &= ~B_NEEDCOMMIT; 645 nfs_del_committed_range(vp, bp); 646 } 647 rw_exit_write(&np->n_commitlock); 648 649 bp->b_flags &= ~B_WRITEINPROG; 650 651 /* 652 * For an interrupted write, the buffer is still valid and the 653 * write hasn't been pushed to the server yet, so we can't set 654 * B_ERROR and report the interruption by setting B_EINTR. For 655 * the B_ASYNC case, B_EINTR is not relevant, so the rpc attempt 656 * is essentially a noop. 657 * For the case of a V3 write rpc not being committed to stable 658 * storage, the block is still dirty and requires either a commit 659 * rpc or another write rpc with iomode == NFSV3WRITE_FILESYNC 660 * before the block is reused. This is indicated by setting the 661 * B_DELWRI and B_NEEDCOMMIT flags. 662 */ 663 if (error == EINTR || (!error && (bp->b_flags & B_NEEDCOMMIT))) { 664 s = splbio(); 665 buf_dirty(bp); 666 splx(s); 667 668 if (!(bp->b_flags & B_ASYNC) && error) 669 bp->b_flags |= B_EINTR; 670 } else { 671 if (error) { 672 bp->b_flags |= B_ERROR; 673 bp->b_error = np->n_error = error; 674 np->n_flag |= NWRITEERR; 675 } 676 bp->b_dirtyoff = bp->b_dirtyend = 0; 677 } 678 } 679 bp->b_resid = uiop->uio_resid; 680 if (must_commit) 681 nfs_clearcommit(vp->v_mount); 682 s = splbio(); 683 biodone(bp); 684 splx(s); 685 return (error); 686 } 687