nfs_bio.c revision 1.18
1 /* $NetBSD: nfs_bio.c,v 1.18 1995/01/10 06:50:03 mycroft Exp $ */ 2 3 /* 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)nfs_bio.c 8.5 (Berkeley) 1/4/94 39 */ 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/resourcevar.h> 44 #include <sys/proc.h> 45 #include <sys/buf.h> 46 #include <sys/vnode.h> 47 #include <sys/trace.h> 48 #include <sys/mount.h> 49 #include <sys/kernel.h> 50 51 #include <vm/vm.h> 52 53 #include <nfs/nfsnode.h> 54 #include <nfs/rpcv2.h> 55 #include <nfs/nfsv2.h> 56 #include <nfs/nfs.h> 57 #include <nfs/nfsmount.h> 58 #include <nfs/nqnfs.h> 59 60 struct buf *incore(), *nfs_getcacheblk(); 61 extern struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON]; 62 extern int nfs_numasync; 63 64 /* 65 * Vnode op for read using bio 66 * Any similarity to readip() is purely coincidental 67 */ 68 nfs_bioread(vp, uio, ioflag, cred) 69 register struct vnode *vp; 70 register struct uio *uio; 71 int ioflag; 72 struct ucred *cred; 73 { 74 register struct nfsnode *np = VTONFS(vp); 75 register int biosize, diff; 76 struct buf *bp, *rabp; 77 struct vattr vattr; 78 struct proc *p; 79 struct nfsmount *nmp; 80 daddr_t lbn, bn, rabn; 81 caddr_t baddr; 82 int got_buf, nra, error = 0, n, on, not_readin; 83 84 #ifdef lint 85 ioflag = ioflag; 86 #endif /* lint */ 87 #ifdef DIAGNOSTIC 88 if (uio->uio_rw != UIO_READ) 89 panic("nfs_read mode"); 90 #endif 91 if (uio->uio_resid == 0) 92 return (0); 93 if (uio->uio_offset < 0 && vp->v_type != VDIR) 94 return (EINVAL); 95 nmp = VFSTONFS(vp->v_mount); 96 biosize = nmp->nm_rsize; 97 p = uio->uio_procp; 98 /* 99 * For nfs, cache consistency can only be maintained approximately. 100 * Although RFC1094 does not specify the criteria, the following is 101 * believed to be compatible with the reference port. 102 * For nqnfs, full cache consistency is maintained within the loop. 103 * For nfs: 104 * If the file's modify time on the server has changed since the 105 * last read rpc or you have written to the file, 106 * you may have lost data cache consistency with the 107 * server, so flush all of the file's data out of the cache. 108 * Then force a getattr rpc to ensure that you have up to date 109 * attributes. 110 * The mount flag NFSMNT_MYWRITE says "Assume that my writes are 111 * the ones changing the modify time. 112 * NB: This implies that cache data can be read when up to 113 * NFS_ATTRTIMEO seconds out of date. If you find that you need current 114 * attributes this could be forced by setting n_attrstamp to 0 before 115 * the VOP_GETATTR() call. 116 */ 117 if ((nmp->nm_flag & NFSMNT_NQNFS) == 0 && vp->v_type != VLNK) { 118 if (np->n_flag & NMODIFIED) { 119 if ((nmp->nm_flag & NFSMNT_MYWRITE) == 0 || 120 vp->v_type != VREG) { 121 if (error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1)) 122 return (error); 123 } 124 np->n_attrstamp = 0; 125 np->n_direofoffset = 0; 126 if (error = VOP_GETATTR(vp, &vattr, cred, p)) 127 return (error); 128 np->n_mtime = vattr.va_mtime.ts_sec; 129 } else { 130 if (error = VOP_GETATTR(vp, &vattr, cred, p)) 131 return (error); 132 if (np->n_mtime != vattr.va_mtime.ts_sec) { 133 np->n_direofoffset = 0; 134 if (error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1)) 135 return (error); 136 np->n_mtime = vattr.va_mtime.ts_sec; 137 } 138 } 139 } 140 do { 141 142 /* 143 * Get a valid lease. If cached data is stale, flush it. 144 */ 145 if (nmp->nm_flag & NFSMNT_NQNFS) { 146 if (NQNFS_CKINVALID(vp, np, NQL_READ)) { 147 do { 148 error = nqnfs_getlease(vp, NQL_READ, cred, p); 149 } while (error == NQNFS_EXPIRED); 150 if (error) 151 return (error); 152 if (np->n_lrev != np->n_brev || 153 (np->n_flag & NQNFSNONCACHE) || 154 ((np->n_flag & NMODIFIED) && vp->v_type == VDIR)) { 155 if (vp->v_type == VDIR) { 156 np->n_direofoffset = 0; 157 cache_purge(vp); 158 } 159 if (error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1)) 160 return (error); 161 np->n_brev = np->n_lrev; 162 } 163 } else if (vp->v_type == VDIR && (np->n_flag & NMODIFIED)) { 164 np->n_direofoffset = 0; 165 cache_purge(vp); 166 if (error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1)) 167 return (error); 168 } 169 } 170 if (np->n_flag & NQNFSNONCACHE) { 171 switch (vp->v_type) { 172 case VREG: 173 error = nfs_readrpc(vp, uio, cred); 174 break; 175 case VLNK: 176 error = nfs_readlinkrpc(vp, uio, cred); 177 break; 178 case VDIR: 179 error = nfs_readdirrpc(vp, uio, cred); 180 break; 181 }; 182 return (error); 183 } 184 baddr = (caddr_t)0; 185 switch (vp->v_type) { 186 case VREG: 187 nfsstats.biocache_reads++; 188 lbn = uio->uio_offset / biosize; 189 on = uio->uio_offset & (biosize-1); 190 bn = lbn * (biosize / DEV_BSIZE); 191 not_readin = 1; 192 193 /* 194 * Start the read ahead(s), as required. 195 */ 196 if (nfs_numasync > 0 && nmp->nm_readahead > 0 && 197 lbn == vp->v_lastr + 1) { 198 for (nra = 0; nra < nmp->nm_readahead && 199 (lbn + 1 + nra) * biosize < np->n_size; nra++) { 200 rabn = (lbn + 1 + nra) * (biosize / DEV_BSIZE); 201 if (!incore(vp, rabn)) { 202 rabp = nfs_getcacheblk(vp, rabn, biosize, p); 203 if (!rabp) 204 return (EINTR); 205 if ((rabp->b_flags & (B_DELWRI | B_DONE)) == 0) { 206 rabp->b_flags |= (B_READ | B_ASYNC); 207 if (nfs_asyncio(rabp, cred)) { 208 rabp->b_flags |= B_INVAL; 209 brelse(rabp); 210 } 211 } 212 } 213 } 214 } 215 216 /* 217 * If the block is in the cache and has the required data 218 * in a valid region, just copy it out. 219 * Otherwise, get the block and write back/read in, 220 * as required. 221 */ 222 if ((bp = incore(vp, bn)) && 223 (bp->b_flags & (B_BUSY | B_WRITEINPROG)) == 224 (B_BUSY | B_WRITEINPROG)) 225 got_buf = 0; 226 else { 227 again: 228 bp = nfs_getcacheblk(vp, bn, biosize, p); 229 if (!bp) 230 return (EINTR); 231 got_buf = 1; 232 if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0) { 233 bp->b_flags |= B_READ; 234 not_readin = 0; 235 if (error = nfs_doio(bp, cred, p)) { 236 brelse(bp); 237 return (error); 238 } 239 } 240 } 241 n = min((unsigned)(biosize - on), uio->uio_resid); 242 diff = np->n_size - uio->uio_offset; 243 if (diff < n) 244 n = diff; 245 if (not_readin && n > 0) { 246 if (on < bp->b_validoff || (on + n) > bp->b_validend) { 247 if (!got_buf) { 248 bp = nfs_getcacheblk(vp, bn, biosize, p); 249 if (!bp) 250 return (EINTR); 251 got_buf = 1; 252 } 253 bp->b_flags |= B_INVAL; 254 if (bp->b_dirtyend > 0) { 255 if ((bp->b_flags & B_DELWRI) == 0) 256 panic("nfsbioread"); 257 if (VOP_BWRITE(bp) == EINTR) 258 return (EINTR); 259 } else 260 brelse(bp); 261 goto again; 262 } 263 } 264 vp->v_lastr = lbn; 265 diff = (on >= bp->b_validend) ? 0 : (bp->b_validend - on); 266 if (diff < n) 267 n = diff; 268 break; 269 case VLNK: 270 nfsstats.biocache_readlinks++; 271 bp = nfs_getcacheblk(vp, (daddr_t)0, NFS_MAXPATHLEN, p); 272 if (!bp) 273 return (EINTR); 274 if ((bp->b_flags & B_DONE) == 0) { 275 bp->b_flags |= B_READ; 276 if (error = nfs_doio(bp, cred, p)) { 277 brelse(bp); 278 return (error); 279 } 280 } 281 n = min(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid); 282 got_buf = 1; 283 on = 0; 284 break; 285 case VDIR: 286 if (uio->uio_resid < NFS_DIRBLKSIZ) 287 return (0); 288 nfsstats.biocache_readdirs++; 289 bn = (daddr_t)uio->uio_offset; 290 bp = nfs_getcacheblk(vp, bn, NFS_DIRBLKSIZ, p); 291 if (!bp) 292 return (EINTR); 293 if ((bp->b_flags & B_DONE) == 0) { 294 bp->b_flags |= B_READ; 295 if (error = nfs_doio(bp, cred, p)) { 296 brelse(bp); 297 return (error); 298 } 299 } 300 301 /* 302 * If not eof and read aheads are enabled, start one. 303 * (You need the current block first, so that you have the 304 * directory offset cookie of the next block. 305 */ 306 rabn = bp->b_blkno; 307 if (nfs_numasync > 0 && nmp->nm_readahead > 0 && 308 rabn != 0 && rabn != np->n_direofoffset && 309 !incore(vp, rabn)) { 310 rabp = nfs_getcacheblk(vp, rabn, NFS_DIRBLKSIZ, p); 311 if (rabp) { 312 if ((rabp->b_flags & (B_DONE | B_DELWRI)) == 0) { 313 rabp->b_flags |= (B_READ | B_ASYNC); 314 if (nfs_asyncio(rabp, cred)) { 315 rabp->b_flags |= B_INVAL; 316 brelse(rabp); 317 } 318 } 319 } 320 } 321 on = 0; 322 n = min(uio->uio_resid, NFS_DIRBLKSIZ - bp->b_resid); 323 got_buf = 1; 324 break; 325 }; 326 327 if (n > 0) { 328 if (!baddr) 329 baddr = bp->b_data; 330 error = uiomove(baddr + on, (int)n, uio); 331 } 332 switch (vp->v_type) { 333 case VREG: 334 if (n + on == biosize || uio->uio_offset == np->n_size) 335 bp->b_flags |= B_AGE; 336 break; 337 case VLNK: 338 n = 0; 339 break; 340 case VDIR: 341 uio->uio_offset = bp->b_blkno; 342 break; 343 }; 344 if (got_buf) 345 brelse(bp); 346 } while (error == 0 && uio->uio_resid > 0 && n > 0); 347 return (error); 348 } 349 350 /* 351 * Vnode op for write using bio 352 */ 353 nfs_write(ap) 354 struct vop_write_args /* { 355 struct vnode *a_vp; 356 struct uio *a_uio; 357 int a_ioflag; 358 struct ucred *a_cred; 359 } */ *ap; 360 { 361 register int biosize; 362 register struct uio *uio = ap->a_uio; 363 struct proc *p = uio->uio_procp; 364 register struct vnode *vp = ap->a_vp; 365 struct nfsnode *np = VTONFS(vp); 366 register struct ucred *cred = ap->a_cred; 367 int ioflag = ap->a_ioflag; 368 struct buf *bp; 369 struct vattr vattr; 370 struct nfsmount *nmp; 371 daddr_t lbn, bn; 372 int n, on, error = 0; 373 374 #ifdef DIAGNOSTIC 375 if (uio->uio_rw != UIO_WRITE) 376 panic("nfs_write mode"); 377 if (uio->uio_segflg == UIO_USERSPACE && uio->uio_procp != curproc) 378 panic("nfs_write proc"); 379 #endif 380 if (vp->v_type != VREG) 381 return (EIO); 382 if (np->n_flag & NWRITEERR) { 383 np->n_flag &= ~NWRITEERR; 384 return (np->n_error); 385 } 386 if (ioflag & (IO_APPEND | IO_SYNC)) { 387 if (np->n_flag & NMODIFIED) { 388 np->n_attrstamp = 0; 389 if (error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1)) 390 return (error); 391 } 392 if (ioflag & IO_APPEND) { 393 np->n_attrstamp = 0; 394 if (error = VOP_GETATTR(vp, &vattr, cred, p)) 395 return (error); 396 uio->uio_offset = np->n_size; 397 } 398 } 399 nmp = VFSTONFS(vp->v_mount); 400 if (uio->uio_offset < 0) 401 return (EINVAL); 402 if (uio->uio_resid == 0) 403 return (0); 404 /* 405 * Maybe this should be above the vnode op call, but so long as 406 * file servers have no limits, i don't think it matters 407 */ 408 if (p && uio->uio_offset + uio->uio_resid > 409 p->p_rlimit[RLIMIT_FSIZE].rlim_cur) { 410 psignal(p, SIGXFSZ); 411 return (EFBIG); 412 } 413 /* 414 * I use nm_rsize, not nm_wsize so that all buffer cache blocks 415 * will be the same size within a filesystem. nfs_writerpc will 416 * still use nm_wsize when sizing the rpc's. 417 */ 418 biosize = nmp->nm_rsize; 419 do { 420 421 /* 422 * XXX make sure we aren't cached in the VM page cache 423 */ 424 (void)vnode_pager_uncache(vp); 425 426 /* 427 * Check for a valid write lease. 428 * If non-cachable, just do the rpc 429 */ 430 if ((nmp->nm_flag & NFSMNT_NQNFS) && 431 NQNFS_CKINVALID(vp, np, NQL_WRITE)) { 432 do { 433 error = nqnfs_getlease(vp, NQL_WRITE, cred, p); 434 } while (error == NQNFS_EXPIRED); 435 if (error) 436 return (error); 437 if (np->n_lrev != np->n_brev || 438 (np->n_flag & NQNFSNONCACHE)) { 439 if (error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1)) 440 return (error); 441 np->n_brev = np->n_lrev; 442 } 443 } 444 if (np->n_flag & NQNFSNONCACHE) 445 return (nfs_writerpc(vp, uio, cred, ioflag)); 446 nfsstats.biocache_writes++; 447 lbn = uio->uio_offset / biosize; 448 on = uio->uio_offset & (biosize-1); 449 n = min((unsigned)(biosize - on), uio->uio_resid); 450 bn = lbn * (biosize / DEV_BSIZE); 451 again: 452 bp = nfs_getcacheblk(vp, bn, biosize, p); 453 if (!bp) 454 return (EINTR); 455 if (bp->b_wcred == NOCRED) { 456 crhold(cred); 457 bp->b_wcred = cred; 458 } 459 np->n_flag |= NMODIFIED; 460 if (uio->uio_offset + n > np->n_size) { 461 np->n_size = uio->uio_offset + n; 462 vnode_pager_setsize(vp, (u_long)np->n_size); 463 } 464 465 /* 466 * If the new write will leave a contiguous dirty 467 * area, just update the b_dirtyoff and b_dirtyend, 468 * otherwise force a write rpc of the old dirty area. 469 */ 470 if (bp->b_dirtyend > 0 && 471 (on > bp->b_dirtyend || (on + n) < bp->b_dirtyoff)) { 472 bp->b_proc = p; 473 if (VOP_BWRITE(bp) == EINTR) 474 return (EINTR); 475 goto again; 476 } 477 478 /* 479 * Check for valid write lease and get one as required. 480 * In case getblk() and/or bwrite() delayed us. 481 */ 482 if ((nmp->nm_flag & NFSMNT_NQNFS) && 483 NQNFS_CKINVALID(vp, np, NQL_WRITE)) { 484 do { 485 error = nqnfs_getlease(vp, NQL_WRITE, cred, p); 486 } while (error == NQNFS_EXPIRED); 487 if (error) { 488 brelse(bp); 489 return (error); 490 } 491 if (np->n_lrev != np->n_brev || 492 (np->n_flag & NQNFSNONCACHE)) { 493 brelse(bp); 494 if (error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1)) 495 return (error); 496 np->n_brev = np->n_lrev; 497 goto again; 498 } 499 } 500 if (error = uiomove((char *)bp->b_data + on, n, uio)) { 501 bp->b_flags |= B_ERROR; 502 brelse(bp); 503 return (error); 504 } 505 if (bp->b_dirtyend > 0) { 506 bp->b_dirtyoff = min(on, bp->b_dirtyoff); 507 bp->b_dirtyend = max((on + n), bp->b_dirtyend); 508 } else { 509 bp->b_dirtyoff = on; 510 bp->b_dirtyend = on + n; 511 } 512 #ifndef notdef 513 if (bp->b_validend == 0 || bp->b_validend < bp->b_dirtyoff || 514 bp->b_validoff > bp->b_dirtyend) { 515 bp->b_validoff = bp->b_dirtyoff; 516 bp->b_validend = bp->b_dirtyend; 517 } else { 518 bp->b_validoff = min(bp->b_validoff, bp->b_dirtyoff); 519 bp->b_validend = max(bp->b_validend, bp->b_dirtyend); 520 } 521 #else 522 bp->b_validoff = bp->b_dirtyoff; 523 bp->b_validend = bp->b_dirtyend; 524 #endif 525 if (ioflag & IO_APPEND) 526 bp->b_flags |= B_APPENDWRITE; 527 528 /* 529 * If the lease is non-cachable or IO_SYNC do bwrite(). 530 */ 531 if ((np->n_flag & NQNFSNONCACHE) || (ioflag & IO_SYNC)) { 532 bp->b_proc = p; 533 if (error = VOP_BWRITE(bp)) 534 return (error); 535 } else if ((n + on) == biosize && 536 (nmp->nm_flag & NFSMNT_NQNFS) == 0) { 537 bp->b_proc = (struct proc *)0; 538 bawrite(bp); 539 } else 540 bdwrite(bp); 541 } while (uio->uio_resid > 0 && n > 0); 542 return (0); 543 } 544 545 /* 546 * Get an nfs cache block. 547 * Allocate a new one if the block isn't currently in the cache 548 * and return the block marked busy. If the calling process is 549 * interrupted by a signal for an interruptible mount point, return 550 * NULL. 551 */ 552 struct buf * 553 nfs_getcacheblk(vp, bn, size, p) 554 struct vnode *vp; 555 daddr_t bn; 556 int size; 557 struct proc *p; 558 { 559 register struct buf *bp; 560 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 561 562 if (nmp->nm_flag & NFSMNT_INT) { 563 bp = getblk(vp, bn, size, PCATCH, 0); 564 while (bp == (struct buf *)0) { 565 if (nfs_sigintr(nmp, (struct nfsreq *)0, p)) 566 return ((struct buf *)0); 567 bp = getblk(vp, bn, size, 0, 2 * hz); 568 } 569 } else 570 bp = getblk(vp, bn, size, 0, 0); 571 return (bp); 572 } 573 574 /* 575 * Flush and invalidate all dirty buffers. If another process is already 576 * doing the flush, just wait for completion. 577 */ 578 nfs_vinvalbuf(vp, flags, cred, p, intrflg) 579 struct vnode *vp; 580 int flags; 581 struct ucred *cred; 582 struct proc *p; 583 int intrflg; 584 { 585 register struct nfsnode *np = VTONFS(vp); 586 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 587 int error = 0, slpflag, slptimeo; 588 589 if ((nmp->nm_flag & NFSMNT_INT) == 0) 590 intrflg = 0; 591 if (intrflg) { 592 slpflag = PCATCH; 593 slptimeo = 2 * hz; 594 } else { 595 slpflag = 0; 596 slptimeo = 0; 597 } 598 /* 599 * First wait for any other process doing a flush to complete. 600 */ 601 while (np->n_flag & NFLUSHINPROG) { 602 np->n_flag |= NFLUSHWANT; 603 error = tsleep((caddr_t)&np->n_flag, PRIBIO + 2, "nfsvinval", 604 slptimeo); 605 if (error && intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) 606 return (EINTR); 607 } 608 609 /* 610 * Now, flush as required. 611 */ 612 np->n_flag |= NFLUSHINPROG; 613 error = vinvalbuf(vp, flags, cred, p, slpflag, 0); 614 while (error) { 615 if (intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) { 616 np->n_flag &= ~NFLUSHINPROG; 617 if (np->n_flag & NFLUSHWANT) { 618 np->n_flag &= ~NFLUSHWANT; 619 wakeup((caddr_t)&np->n_flag); 620 } 621 return (EINTR); 622 } 623 error = vinvalbuf(vp, flags, cred, p, 0, slptimeo); 624 } 625 np->n_flag &= ~(NMODIFIED | NFLUSHINPROG); 626 if (np->n_flag & NFLUSHWANT) { 627 np->n_flag &= ~NFLUSHWANT; 628 wakeup((caddr_t)&np->n_flag); 629 } 630 return (0); 631 } 632 633 /* 634 * Initiate asynchronous I/O. Return an error if no nfsiods are available. 635 * This is mainly to avoid queueing async I/O requests when the nfsiods 636 * are all hung on a dead server. 637 */ 638 nfs_asyncio(bp, cred) 639 register struct buf *bp; 640 struct ucred *cred; 641 { 642 register int i; 643 644 if (nfs_numasync == 0) 645 return (EIO); 646 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 647 if (nfs_iodwant[i]) { 648 if (bp->b_flags & B_READ) { 649 if (bp->b_rcred == NOCRED && cred != NOCRED) { 650 crhold(cred); 651 bp->b_rcred = cred; 652 } 653 } else { 654 if (bp->b_wcred == NOCRED && cred != NOCRED) { 655 crhold(cred); 656 bp->b_wcred = cred; 657 } 658 } 659 660 TAILQ_INSERT_TAIL(&nfs_bufq, bp, b_freelist); 661 nfs_iodwant[i] = (struct proc *)0; 662 wakeup((caddr_t)&nfs_iodwant[i]); 663 return (0); 664 } 665 return (EIO); 666 } 667 668 /* 669 * Do an I/O operation to/from a cache block. This may be called 670 * synchronously or from an nfsiod. 671 */ 672 int 673 nfs_doio(bp, cr, p) 674 register struct buf *bp; 675 struct cred *cr; 676 struct proc *p; 677 { 678 register struct uio *uiop; 679 register struct vnode *vp; 680 struct nfsnode *np; 681 struct nfsmount *nmp; 682 int error, diff, len; 683 struct uio uio; 684 struct iovec io; 685 686 vp = bp->b_vp; 687 np = VTONFS(vp); 688 nmp = VFSTONFS(vp->v_mount); 689 uiop = &uio; 690 uiop->uio_iov = &io; 691 uiop->uio_iovcnt = 1; 692 uiop->uio_segflg = UIO_SYSSPACE; 693 uiop->uio_procp = p; 694 695 /* 696 * Historically, paging was done with physio, but no more... 697 */ 698 if (bp->b_flags & B_PHYS) { 699 /* 700 * ...though reading /dev/drum still gets us here. 701 */ 702 io.iov_len = uiop->uio_resid = bp->b_bcount; 703 /* mapping was done by vmapbuf() */ 704 io.iov_base = bp->b_data; 705 uiop->uio_offset = bp->b_blkno * DEV_BSIZE; 706 if (bp->b_flags & B_READ) { 707 uiop->uio_rw = UIO_READ; 708 nfsstats.read_physios++; 709 error = nfs_readrpc(vp, uiop, cr); 710 } else { 711 uiop->uio_rw = UIO_WRITE; 712 nfsstats.write_physios++; 713 error = nfs_writerpc(vp, uiop, cr); 714 } 715 if (error) { 716 bp->b_flags |= B_ERROR; 717 bp->b_error = error; 718 } 719 } else if (bp->b_flags & B_READ) { 720 io.iov_len = uiop->uio_resid = bp->b_bcount; 721 io.iov_base = bp->b_data; 722 uiop->uio_rw = UIO_READ; 723 switch (vp->v_type) { 724 case VREG: 725 uiop->uio_offset = bp->b_blkno * DEV_BSIZE; 726 nfsstats.read_bios++; 727 error = nfs_readrpc(vp, uiop, cr); 728 if (!error) { 729 bp->b_validoff = 0; 730 if (uiop->uio_resid) { 731 /* 732 * If len > 0, there is a hole in the file and 733 * no writes after the hole have been pushed to 734 * the server yet. 735 * Just zero fill the rest of the valid area. 736 */ 737 diff = bp->b_bcount - uiop->uio_resid; 738 len = np->n_size - (bp->b_blkno * DEV_BSIZE 739 + diff); 740 if (len > 0) { 741 len = min(len, uiop->uio_resid); 742 bzero((char *)bp->b_data + diff, len); 743 bp->b_validend = diff + len; 744 } else 745 bp->b_validend = diff; 746 } else 747 bp->b_validend = bp->b_bcount; 748 } 749 if (p && (vp->v_flag & VTEXT) && 750 (((nmp->nm_flag & NFSMNT_NQNFS) && 751 NQNFS_CKINVALID(vp, np, NQL_READ) && 752 np->n_lrev != np->n_brev) || 753 (!(nmp->nm_flag & NFSMNT_NQNFS) && 754 np->n_mtime != np->n_vattr.va_mtime.ts_sec))) { 755 uprintf("Process killed due to text file modification\n"); 756 psignal(p, SIGKILL); 757 p->p_holdcnt++; 758 } 759 break; 760 case VLNK: 761 uiop->uio_offset = 0; 762 nfsstats.readlink_bios++; 763 error = nfs_readlinkrpc(vp, uiop, cr); 764 break; 765 case VDIR: 766 uiop->uio_offset = bp->b_lblkno; 767 nfsstats.readdir_bios++; 768 if (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) 769 error = nfs_readdirlookrpc(vp, uiop, cr); 770 else 771 error = nfs_readdirrpc(vp, uiop, cr); 772 /* 773 * Save offset cookie in b_blkno. 774 */ 775 bp->b_blkno = uiop->uio_offset; 776 break; 777 }; 778 if (error) { 779 bp->b_flags |= B_ERROR; 780 bp->b_error = error; 781 } 782 } else { 783 io.iov_len = uiop->uio_resid = bp->b_dirtyend 784 - bp->b_dirtyoff; 785 uiop->uio_offset = (bp->b_blkno * DEV_BSIZE) 786 + bp->b_dirtyoff; 787 io.iov_base = (char *)bp->b_data + bp->b_dirtyoff; 788 uiop->uio_rw = UIO_WRITE; 789 nfsstats.write_bios++; 790 if (bp->b_flags & B_APPENDWRITE) 791 error = nfs_writerpc(vp, uiop, cr, IO_APPEND); 792 else 793 error = nfs_writerpc(vp, uiop, cr, 0); 794 bp->b_flags &= ~(B_WRITEINPROG | B_APPENDWRITE); 795 796 /* 797 * For an interrupted write, the buffer is still valid and the 798 * write hasn't been pushed to the server yet, so we can't set 799 * B_ERROR and report the interruption by setting B_EINTR. For 800 * the B_ASYNC case, B_EINTR is not relevant, so the rpc attempt 801 * is essentially a noop. 802 */ 803 if (error == EINTR) { 804 bp->b_flags &= ~B_INVAL; 805 bp->b_flags |= B_DELWRI; 806 807 /* 808 * Since for the B_ASYNC case, nfs_bwrite() has reassigned the 809 * buffer to the clean list, we have to reassign it back to the 810 * dirty one. Ugh. 811 */ 812 if (bp->b_flags & B_ASYNC) 813 reassignbuf(bp, vp); 814 else 815 bp->b_flags |= B_EINTR; 816 } else { 817 if (error) { 818 bp->b_flags |= B_ERROR; 819 bp->b_error = np->n_error = error; 820 np->n_flag |= NWRITEERR; 821 } 822 bp->b_dirtyoff = bp->b_dirtyend = 0; 823 } 824 } 825 bp->b_resid = uiop->uio_resid; 826 biodone(bp); 827 return (error); 828 } 829
Indexes created Wed Oct 15 16:09:53 GMT 2025