1 /* $OpenBSD: nfs_bio.c,v 1.73 2012/07/11 12:39:20 guenther 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, 1)) { 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 panic("nfsbioread: type %x unexpected", 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 } 231 232 if (vp->v_type == VLNK) 233 n = 0; 234 235 if (got_buf) 236 brelse(bp); 237 } while (error == 0 && uio->uio_resid > 0 && n > 0); 238 return (error); 239 } 240 241 /* 242 * Vnode op for write using bio 243 */ 244 int 245 nfs_write(void *v) 246 { 247 struct vop_write_args *ap = v; 248 int biosize; 249 struct uio *uio = ap->a_uio; 250 struct proc *p = uio->uio_procp; 251 struct vnode *vp = ap->a_vp; 252 struct nfsnode *np = VTONFS(vp); 253 struct ucred *cred = ap->a_cred; 254 int ioflag = ap->a_ioflag; 255 struct buf *bp; 256 struct vattr vattr; 257 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 258 daddr64_t lbn, bn; 259 int n, on, error = 0, extended = 0, wrotedta = 0, truncated = 0; 260 int overrun; 261 262 #ifdef DIAGNOSTIC 263 if (uio->uio_rw != UIO_WRITE) 264 panic("nfs_write mode"); 265 if (uio->uio_segflg == UIO_USERSPACE && uio->uio_procp != curproc) 266 panic("nfs_write proc"); 267 #endif 268 if (vp->v_type != VREG) 269 return (EIO); 270 if (np->n_flag & NWRITEERR) { 271 np->n_flag &= ~NWRITEERR; 272 return (np->n_error); 273 } 274 if ((nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_GOTFSINFO)) == NFSMNT_NFSV3) 275 (void)nfs_fsinfo(nmp, vp, cred, p); 276 if (ioflag & (IO_APPEND | IO_SYNC)) { 277 if (np->n_flag & NMODIFIED) { 278 NFS_INVALIDATE_ATTRCACHE(np); 279 error = nfs_vinvalbuf(vp, V_SAVE, cred, p); 280 if (error) 281 return (error); 282 } 283 if (ioflag & IO_APPEND) { 284 NFS_INVALIDATE_ATTRCACHE(np); 285 error = VOP_GETATTR(vp, &vattr, cred, p); 286 if (error) 287 return (error); 288 uio->uio_offset = np->n_size; 289 } 290 } 291 if (uio->uio_offset < 0) 292 return (EINVAL); 293 if (uio->uio_resid == 0) 294 return (0); 295 296 /* do the filesize rlimit check */ 297 if ((error = vn_fsizechk(vp, uio, ioflag, &overrun))) 298 return (error); 299 300 /* 301 * update the cache write creds for this node. 302 */ 303 if (np->n_wcred) 304 crfree(np->n_wcred); 305 np->n_wcred = cred; 306 crhold(cred); 307 308 /* 309 * I use nm_rsize, not nm_wsize so that all buffer cache blocks 310 * will be the same size within a filesystem. nfs_writerpc will 311 * still use nm_wsize when sizing the rpc's. 312 */ 313 biosize = nmp->nm_rsize; 314 do { 315 316 /* 317 * XXX make sure we aren't cached in the VM page cache 318 */ 319 uvm_vnp_uncache(vp); 320 321 nfsstats.biocache_writes++; 322 lbn = uio->uio_offset / biosize; 323 on = uio->uio_offset & (biosize-1); 324 n = min((unsigned)(biosize - on), uio->uio_resid); 325 bn = lbn * (biosize / DEV_BSIZE); 326 again: 327 bp = nfs_getcacheblk(vp, bn, biosize, p); 328 if (!bp) { 329 error = EINTR; 330 goto out; 331 } 332 np->n_flag |= NMODIFIED; 333 if (uio->uio_offset + n > np->n_size) { 334 np->n_size = uio->uio_offset + n; 335 uvm_vnp_setsize(vp, (u_long)np->n_size); 336 extended = 1; 337 } else if (uio->uio_offset + n < np->n_size) 338 truncated = 1; 339 340 /* 341 * If the new write will leave a contiguous dirty 342 * area, just update the b_dirtyoff and b_dirtyend, 343 * otherwise force a write rpc of the old dirty area. 344 */ 345 if (bp->b_dirtyend > 0 && 346 (on > bp->b_dirtyend || (on + n) < bp->b_dirtyoff)) { 347 bp->b_proc = p; 348 if (VOP_BWRITE(bp) == EINTR) { 349 error = EINTR; 350 goto out; 351 } 352 goto again; 353 } 354 355 error = uiomove((char *)bp->b_data + on, n, uio); 356 if (error) { 357 bp->b_flags |= B_ERROR; 358 brelse(bp); 359 goto out; 360 } 361 if (bp->b_dirtyend > 0) { 362 bp->b_dirtyoff = min(on, bp->b_dirtyoff); 363 bp->b_dirtyend = max((on + n), bp->b_dirtyend); 364 } else { 365 bp->b_dirtyoff = on; 366 bp->b_dirtyend = on + n; 367 } 368 if (bp->b_validend == 0 || bp->b_validend < bp->b_dirtyoff || 369 bp->b_validoff > bp->b_dirtyend) { 370 bp->b_validoff = bp->b_dirtyoff; 371 bp->b_validend = bp->b_dirtyend; 372 } else { 373 bp->b_validoff = min(bp->b_validoff, bp->b_dirtyoff); 374 bp->b_validend = max(bp->b_validend, bp->b_dirtyend); 375 } 376 377 wrotedta = 1; 378 379 /* 380 * Since this block is being modified, it must be written 381 * again and not just committed. 382 */ 383 384 if (NFS_ISV3(vp)) { 385 rw_enter_write(&np->n_commitlock); 386 if (bp->b_flags & B_NEEDCOMMIT) { 387 bp->b_flags &= ~B_NEEDCOMMIT; 388 nfs_del_tobecommitted_range(vp, bp); 389 } 390 nfs_del_committed_range(vp, bp); 391 rw_exit_write(&np->n_commitlock); 392 } else 393 bp->b_flags &= ~B_NEEDCOMMIT; 394 395 if (ioflag & IO_SYNC) { 396 bp->b_proc = p; 397 error = VOP_BWRITE(bp); 398 if (error) 399 goto out; 400 } else if ((n + on) == biosize) { 401 bp->b_proc = NULL; 402 bp->b_flags |= B_ASYNC; 403 (void)nfs_writebp(bp, 0); 404 } else { 405 bdwrite(bp); 406 } 407 } while (uio->uio_resid > 0 && n > 0); 408 409 /*out: XXX belongs here??? */ 410 if (wrotedta) 411 VN_KNOTE(vp, NOTE_WRITE | (extended ? NOTE_EXTEND : 0) | 412 (truncated ? NOTE_TRUNCATE : 0)); 413 414 out: 415 /* correct the result for writes clamped by vn_fsizechk() */ 416 uio->uio_resid += overrun; 417 418 return (error); 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, int readahead) 501 { 502 if (nfs_numasync == 0) 503 goto out; 504 505 while (nfs_bufqlen > nfs_bufqmax) 506 if (readahead) 507 goto out; 508 else 509 tsleep(&nfs_bufqlen, PRIBIO, "nfs_bufq", 0); 510 511 if ((bp->b_flags & B_READ) == 0) { 512 bp->b_flags |= B_WRITEINPROG; 513 } 514 515 TAILQ_INSERT_TAIL(&nfs_bufq, bp, b_freelist); 516 nfs_bufqlen++; 517 518 wakeup_one(&nfs_bufq); 519 return (0); 520 521 out: 522 nfsstats.forcedsync++; 523 return (EIO); 524 } 525 526 /* 527 * Do an I/O operation to/from a cache block. This may be called 528 * synchronously or from an nfsiod. 529 */ 530 int 531 nfs_doio(struct buf *bp, struct proc *p) 532 { 533 struct uio *uiop; 534 struct vnode *vp; 535 struct nfsnode *np; 536 struct nfsmount *nmp; 537 int s, error = 0, diff, len, iomode, must_commit = 0; 538 struct uio uio; 539 struct iovec io; 540 541 vp = bp->b_vp; 542 np = VTONFS(vp); 543 nmp = VFSTONFS(vp->v_mount); 544 uiop = &uio; 545 uiop->uio_iov = &io; 546 uiop->uio_iovcnt = 1; 547 uiop->uio_segflg = UIO_SYSSPACE; 548 uiop->uio_procp = p; 549 550 /* 551 * Historically, paging was done with physio, but no more. 552 */ 553 if (bp->b_flags & B_PHYS) { 554 io.iov_len = uiop->uio_resid = bp->b_bcount; 555 /* mapping was done by vmapbuf() */ 556 io.iov_base = bp->b_data; 557 uiop->uio_offset = ((off_t)bp->b_blkno) << DEV_BSHIFT; 558 if (bp->b_flags & B_READ) { 559 uiop->uio_rw = UIO_READ; 560 nfsstats.read_physios++; 561 error = nfs_readrpc(vp, uiop); 562 } else { 563 iomode = NFSV3WRITE_DATASYNC; 564 uiop->uio_rw = UIO_WRITE; 565 nfsstats.write_physios++; 566 error = nfs_writerpc(vp, uiop, &iomode, &must_commit); 567 } 568 if (error) { 569 bp->b_flags |= B_ERROR; 570 bp->b_error = error; 571 } 572 } else if (bp->b_flags & B_READ) { 573 io.iov_len = uiop->uio_resid = bp->b_bcount; 574 io.iov_base = bp->b_data; 575 uiop->uio_rw = UIO_READ; 576 switch (vp->v_type) { 577 case VREG: 578 uiop->uio_offset = ((off_t)bp->b_blkno) << DEV_BSHIFT; 579 nfsstats.read_bios++; 580 bcstats.pendingreads++; 581 bcstats.numreads++; 582 error = nfs_readrpc(vp, uiop); 583 if (!error) { 584 bp->b_validoff = 0; 585 if (uiop->uio_resid) { 586 /* 587 * If len > 0, there is a hole in the file and 588 * no writes after the hole have been pushed to 589 * the server yet. 590 * Just zero fill the rest of the valid area. 591 */ 592 diff = bp->b_bcount - uiop->uio_resid; 593 len = np->n_size - ((((off_t)bp->b_blkno) << DEV_BSHIFT) 594 + diff); 595 if (len > 0) { 596 len = min(len, uiop->uio_resid); 597 bzero((char *)bp->b_data + diff, len); 598 bp->b_validend = diff + len; 599 } else 600 bp->b_validend = diff; 601 } else 602 bp->b_validend = bp->b_bcount; 603 } 604 if (p && (vp->v_flag & VTEXT) && 605 (timespeccmp(&np->n_mtime, &np->n_vattr.va_mtime, !=))) { 606 uprintf("Process killed due to text file modification\n"); 607 psignal(p, SIGKILL); 608 } 609 break; 610 case VLNK: 611 uiop->uio_offset = (off_t)0; 612 nfsstats.readlink_bios++; 613 bcstats.pendingreads++; 614 bcstats.numreads++; 615 error = nfs_readlinkrpc(vp, uiop, curproc->p_ucred); 616 break; 617 default: 618 panic("nfs_doio: type %x unexpected", vp->v_type); 619 break; 620 }; 621 if (error) { 622 bp->b_flags |= B_ERROR; 623 bp->b_error = error; 624 } 625 } else { 626 io.iov_len = uiop->uio_resid = bp->b_dirtyend 627 - bp->b_dirtyoff; 628 uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE 629 + bp->b_dirtyoff; 630 io.iov_base = (char *)bp->b_data + bp->b_dirtyoff; 631 uiop->uio_rw = UIO_WRITE; 632 nfsstats.write_bios++; 633 bcstats.pendingwrites++; 634 bcstats.numwrites++; 635 if ((bp->b_flags & (B_ASYNC | B_NEEDCOMMIT | B_NOCACHE)) == B_ASYNC) 636 iomode = NFSV3WRITE_UNSTABLE; 637 else 638 iomode = NFSV3WRITE_FILESYNC; 639 bp->b_flags |= B_WRITEINPROG; 640 error = nfs_writerpc(vp, uiop, &iomode, &must_commit); 641 642 rw_enter_write(&np->n_commitlock); 643 if (!error && iomode == NFSV3WRITE_UNSTABLE) { 644 bp->b_flags |= B_NEEDCOMMIT; 645 nfs_add_tobecommitted_range(vp, bp); 646 } else { 647 bp->b_flags &= ~B_NEEDCOMMIT; 648 nfs_del_committed_range(vp, bp); 649 } 650 rw_exit_write(&np->n_commitlock); 651 652 bp->b_flags &= ~B_WRITEINPROG; 653 654 /* 655 * For an interrupted write, the buffer is still valid and the 656 * write hasn't been pushed to the server yet, so we can't set 657 * B_ERROR and report the interruption by setting B_EINTR. For 658 * the B_ASYNC case, B_EINTR is not relevant, so the rpc attempt 659 * is essentially a noop. 660 * For the case of a V3 write rpc not being committed to stable 661 * storage, the block is still dirty and requires either a commit 662 * rpc or another write rpc with iomode == NFSV3WRITE_FILESYNC 663 * before the block is reused. This is indicated by setting the 664 * B_DELWRI and B_NEEDCOMMIT flags. 665 */ 666 if (error == EINTR || (!error && (bp->b_flags & B_NEEDCOMMIT))) { 667 s = splbio(); 668 buf_dirty(bp); 669 splx(s); 670 671 if (!(bp->b_flags & B_ASYNC) && error) 672 bp->b_flags |= B_EINTR; 673 } else { 674 if (error) { 675 bp->b_flags |= B_ERROR; 676 bp->b_error = np->n_error = error; 677 np->n_flag |= NWRITEERR; 678 } 679 bp->b_dirtyoff = bp->b_dirtyend = 0; 680 } 681 } 682 bp->b_resid = uiop->uio_resid; 683 if (must_commit) 684 nfs_clearcommit(vp->v_mount); 685 s = splbio(); 686 biodone(bp); 687 splx(s); 688 return (error); 689 } 690