nfs_bio.c revision 1.13
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 * from: @(#)nfs_bio.c 8.5 (Berkeley) 1/4/94 37 * $Id: nfs_bio.c,v 1.13 1994/06/15 19:59:52 mycroft Exp $ 38 */ 39 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/resourcevar.h> 43 #include <sys/proc.h> 44 #include <sys/buf.h> 45 #include <sys/vnode.h> 46 #include <sys/trace.h> 47 #include <sys/mount.h> 48 #include <sys/kernel.h> 49 50 #include <vm/vm.h> 51 52 #include <nfs/nfsnode.h> 53 #include <nfs/rpcv2.h> 54 #include <nfs/nfsv2.h> 55 #include <nfs/nfs.h> 56 #include <nfs/nfsmount.h> 57 #include <nfs/nqnfs.h> 58 59 struct buf *incore(), *nfs_getcacheblk(); 60 extern struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON]; 61 extern int nfs_numasync; 62 63 /* 64 * Vnode op for read using bio 65 * Any similarity to readip() is purely coincidental 66 */ 67 nfs_bioread(vp, uio, ioflag, cred) 68 register struct vnode *vp; 69 register struct uio *uio; 70 int ioflag; 71 struct ucred *cred; 72 { 73 register struct nfsnode *np = VTONFS(vp); 74 register int biosize, diff; 75 struct buf *bp, *rabp; 76 struct vattr vattr; 77 struct proc *p; 78 struct nfsmount *nmp; 79 daddr_t lbn, bn, rabn; 80 caddr_t baddr; 81 int got_buf, nra, error = 0, n, on, not_readin; 82 83 #ifdef lint 84 ioflag = ioflag; 85 #endif /* lint */ 86 #ifdef DIAGNOSTIC 87 if (uio->uio_rw != UIO_READ) 88 panic("nfs_read mode"); 89 #endif 90 if (uio->uio_resid == 0) 91 return (0); 92 if (uio->uio_offset < 0 && vp->v_type != VDIR) 93 return (EINVAL); 94 nmp = VFSTONFS(vp->v_mount); 95 biosize = nmp->nm_rsize; 96 p = uio->uio_procp; 97 /* 98 * For nfs, cache consistency can only be maintained approximately. 99 * Although RFC1094 does not specify the criteria, the following is 100 * believed to be compatible with the reference port. 101 * For nqnfs, full cache consistency is maintained within the loop. 102 * For nfs: 103 * If the file's modify time on the server has changed since the 104 * last read rpc or you have written to the file, 105 * you may have lost data cache consistency with the 106 * server, so flush all of the file's data out of the cache. 107 * Then force a getattr rpc to ensure that you have up to date 108 * attributes. 109 * The mount flag NFSMNT_MYWRITE says "Assume that my writes are 110 * the ones changing the modify time. 111 * NB: This implies that cache data can be read when up to 112 * NFS_ATTRTIMEO seconds out of date. If you find that you need current 113 * attributes this could be forced by setting n_attrstamp to 0 before 114 * the VOP_GETATTR() call. 115 */ 116 if ((nmp->nm_flag & NFSMNT_NQNFS) == 0 && vp->v_type != VLNK) { 117 if (np->n_flag & NMODIFIED) { 118 if ((nmp->nm_flag & NFSMNT_MYWRITE) == 0 || 119 vp->v_type != VREG) { 120 if (error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1)) 121 return (error); 122 } 123 np->n_attrstamp = 0; 124 np->n_direofoffset = 0; 125 if (error = VOP_GETATTR(vp, &vattr, cred, p)) 126 return (error); 127 np->n_mtime = vattr.va_mtime.ts_sec; 128 } else { 129 if (error = VOP_GETATTR(vp, &vattr, cred, p)) 130 return (error); 131 if (np->n_mtime != vattr.va_mtime.ts_sec) { 132 np->n_direofoffset = 0; 133 if (error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1)) 134 return (error); 135 np->n_mtime = vattr.va_mtime.ts_sec; 136 } 137 } 138 } 139 do { 140 141 /* 142 * Get a valid lease. If cached data is stale, flush it. 143 */ 144 if (nmp->nm_flag & NFSMNT_NQNFS) { 145 if (NQNFS_CKINVALID(vp, np, NQL_READ)) { 146 do { 147 error = nqnfs_getlease(vp, NQL_READ, cred, p); 148 } while (error == NQNFS_EXPIRED); 149 if (error) 150 return (error); 151 if (np->n_lrev != np->n_brev || 152 (np->n_flag & NQNFSNONCACHE) || 153 ((np->n_flag & NMODIFIED) && vp->v_type == VDIR)) { 154 if (vp->v_type == VDIR) { 155 np->n_direofoffset = 0; 156 cache_purge(vp); 157 } 158 if (error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1)) 159 return (error); 160 np->n_brev = np->n_lrev; 161 } 162 } else if (vp->v_type == VDIR && (np->n_flag & NMODIFIED)) { 163 np->n_direofoffset = 0; 164 cache_purge(vp); 165 if (error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1)) 166 return (error); 167 } 168 } 169 if (np->n_flag & NQNFSNONCACHE) { 170 switch (vp->v_type) { 171 case VREG: 172 error = nfs_readrpc(vp, uio, cred); 173 break; 174 case VLNK: 175 error = nfs_readlinkrpc(vp, uio, cred); 176 break; 177 case VDIR: 178 error = nfs_readdirrpc(vp, uio, cred); 179 break; 180 }; 181 return (error); 182 } 183 baddr = (caddr_t)0; 184 switch (vp->v_type) { 185 case VREG: 186 nfsstats.biocache_reads++; 187 lbn = uio->uio_offset / biosize; 188 on = uio->uio_offset & (biosize-1); 189 bn = lbn * (biosize / DEV_BSIZE); 190 not_readin = 1; 191 192 /* 193 * Start the read ahead(s), as required. 194 */ 195 if (nfs_numasync > 0 && nmp->nm_readahead > 0 && 196 lbn == vp->v_lastr + 1) { 197 for (nra = 0; nra < nmp->nm_readahead && 198 (lbn + 1 + nra) * biosize < np->n_size; nra++) { 199 rabn = (lbn + 1 + nra) * (biosize / DEV_BSIZE); 200 if (!incore(vp, rabn)) { 201 rabp = nfs_getcacheblk(vp, rabn, biosize, p); 202 if (!rabp) 203 return (EINTR); 204 if ((rabp->b_flags & (B_DELWRI | B_DONE)) == 0) { 205 rabp->b_flags |= (B_READ | B_ASYNC); 206 if (nfs_asyncio(rabp, cred)) { 207 rabp->b_flags |= B_INVAL; 208 brelse(rabp); 209 } 210 } 211 } 212 } 213 } 214 215 /* 216 * If the block is in the cache and has the required data 217 * in a valid region, just copy it out. 218 * Otherwise, get the block and write back/read in, 219 * as required. 220 */ 221 if ((bp = incore(vp, bn)) && 222 (bp->b_flags & (B_BUSY | B_WRITEINPROG)) == 223 (B_BUSY | B_WRITEINPROG)) 224 got_buf = 0; 225 else { 226 again: 227 bp = nfs_getcacheblk(vp, bn, biosize, p); 228 if (!bp) 229 return (EINTR); 230 got_buf = 1; 231 if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0) { 232 bp->b_flags |= B_READ; 233 not_readin = 0; 234 if (error = nfs_doio(bp, cred, p)) { 235 brelse(bp); 236 return (error); 237 } 238 } 239 } 240 n = min((unsigned)(biosize - on), uio->uio_resid); 241 diff = np->n_size - uio->uio_offset; 242 if (diff < n) 243 n = diff; 244 if (not_readin && n > 0) { 245 if (on < bp->b_validoff || (on + n) > bp->b_validend) { 246 if (!got_buf) { 247 bp = nfs_getcacheblk(vp, bn, biosize, p); 248 if (!bp) 249 return (EINTR); 250 got_buf = 1; 251 } 252 bp->b_flags |= B_INVAL; 253 if (bp->b_dirtyend > 0) { 254 if ((bp->b_flags & B_DELWRI) == 0) 255 panic("nfsbioread"); 256 if (VOP_BWRITE(bp) == EINTR) 257 return (EINTR); 258 } else 259 brelse(bp); 260 goto again; 261 } 262 } 263 vp->v_lastr = lbn; 264 diff = (on >= bp->b_validend) ? 0 : (bp->b_validend - on); 265 if (diff < n) 266 n = diff; 267 break; 268 case VLNK: 269 nfsstats.biocache_readlinks++; 270 bp = nfs_getcacheblk(vp, (daddr_t)0, NFS_MAXPATHLEN, p); 271 if (!bp) 272 return (EINTR); 273 if ((bp->b_flags & B_DONE) == 0) { 274 bp->b_flags |= B_READ; 275 if (error = nfs_doio(bp, cred, p)) { 276 brelse(bp); 277 return (error); 278 } 279 } 280 n = min(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid); 281 got_buf = 1; 282 on = 0; 283 break; 284 case VDIR: 285 nfsstats.biocache_readdirs++; 286 bn = (daddr_t)uio->uio_offset; 287 bp = nfs_getcacheblk(vp, bn, NFS_DIRBLKSIZ, p); 288 if (!bp) 289 return (EINTR); 290 if ((bp->b_flags & B_DONE) == 0) { 291 bp->b_flags |= B_READ; 292 if (error = nfs_doio(bp, cred, p)) { 293 brelse(bp); 294 return (error); 295 } 296 } 297 298 /* 299 * If not eof and read aheads are enabled, start one. 300 * (You need the current block first, so that you have the 301 * directory offset cookie of the next block. 302 */ 303 rabn = bp->b_blkno; 304 if (nfs_numasync > 0 && nmp->nm_readahead > 0 && 305 rabn != 0 && rabn != np->n_direofoffset && 306 !incore(vp, rabn)) { 307 rabp = nfs_getcacheblk(vp, rabn, NFS_DIRBLKSIZ, p); 308 if (rabp) { 309 if ((rabp->b_flags & (B_DONE | B_DELWRI)) == 0) { 310 rabp->b_flags |= (B_READ | B_ASYNC); 311 if (nfs_asyncio(rabp, cred)) { 312 rabp->b_flags |= B_INVAL; 313 brelse(rabp); 314 } 315 } 316 } 317 } 318 on = 0; 319 n = min(uio->uio_resid, NFS_DIRBLKSIZ - bp->b_resid); 320 got_buf = 1; 321 break; 322 }; 323 324 if (n > 0) { 325 if (!baddr) 326 baddr = bp->b_data; 327 error = uiomove(baddr + on, (int)n, uio); 328 } 329 switch (vp->v_type) { 330 case VREG: 331 if (n + on == biosize || uio->uio_offset == np->n_size) 332 bp->b_flags |= B_AGE; 333 break; 334 case VLNK: 335 n = 0; 336 break; 337 case VDIR: 338 uio->uio_offset = bp->b_blkno; 339 break; 340 }; 341 if (got_buf) 342 brelse(bp); 343 } while (error == 0 && uio->uio_resid > 0 && n > 0); 344 return (error); 345 } 346 347 /* 348 * Vnode op for write using bio 349 */ 350 nfs_write(ap) 351 struct vop_write_args /* { 352 struct vnode *a_vp; 353 struct uio *a_uio; 354 int a_ioflag; 355 struct ucred *a_cred; 356 } */ *ap; 357 { 358 register int biosize; 359 register struct uio *uio = ap->a_uio; 360 struct proc *p = uio->uio_procp; 361 register struct vnode *vp = ap->a_vp; 362 struct nfsnode *np = VTONFS(vp); 363 register struct ucred *cred = ap->a_cred; 364 int ioflag = ap->a_ioflag; 365 struct buf *bp; 366 struct vattr vattr; 367 struct nfsmount *nmp; 368 daddr_t lbn, bn; 369 int n, on, error = 0; 370 371 #ifdef DIAGNOSTIC 372 if (uio->uio_rw != UIO_WRITE) 373 panic("nfs_write mode"); 374 if (uio->uio_segflg == UIO_USERSPACE && uio->uio_procp != curproc) 375 panic("nfs_write proc"); 376 #endif 377 if (vp->v_type != VREG) 378 return (EIO); 379 if (np->n_flag & NWRITEERR) { 380 np->n_flag &= ~NWRITEERR; 381 return (np->n_error); 382 } 383 if (ioflag & (IO_APPEND | IO_SYNC)) { 384 if (np->n_flag & NMODIFIED) { 385 np->n_attrstamp = 0; 386 if (error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1)) 387 return (error); 388 } 389 if (ioflag & IO_APPEND) { 390 np->n_attrstamp = 0; 391 if (error = VOP_GETATTR(vp, &vattr, cred, p)) 392 return (error); 393 uio->uio_offset = np->n_size; 394 } 395 } 396 nmp = VFSTONFS(vp->v_mount); 397 if (uio->uio_offset < 0) 398 return (EINVAL); 399 if (uio->uio_resid == 0) 400 return (0); 401 /* 402 * Maybe this should be above the vnode op call, but so long as 403 * file servers have no limits, i don't think it matters 404 */ 405 if (p && uio->uio_offset + uio->uio_resid > 406 p->p_rlimit[RLIMIT_FSIZE].rlim_cur) { 407 psignal(p, SIGXFSZ); 408 return (EFBIG); 409 } 410 /* 411 * I use nm_rsize, not nm_wsize so that all buffer cache blocks 412 * will be the same size within a filesystem. nfs_writerpc will 413 * still use nm_wsize when sizing the rpc's. 414 */ 415 biosize = nmp->nm_rsize; 416 do { 417 418 /* 419 * Check for a valid write lease. 420 * If non-cachable, just do the rpc 421 */ 422 if ((nmp->nm_flag & NFSMNT_NQNFS) && 423 NQNFS_CKINVALID(vp, np, NQL_WRITE)) { 424 do { 425 error = nqnfs_getlease(vp, NQL_WRITE, cred, p); 426 } while (error == NQNFS_EXPIRED); 427 if (error) 428 return (error); 429 if (np->n_lrev != np->n_brev || 430 (np->n_flag & NQNFSNONCACHE)) { 431 if (error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1)) 432 return (error); 433 np->n_brev = np->n_lrev; 434 } 435 } 436 if (np->n_flag & NQNFSNONCACHE) 437 return (nfs_writerpc(vp, uio, cred, ioflag)); 438 nfsstats.biocache_writes++; 439 lbn = uio->uio_offset / biosize; 440 on = uio->uio_offset & (biosize-1); 441 n = min((unsigned)(biosize - on), uio->uio_resid); 442 bn = lbn * (biosize / DEV_BSIZE); 443 again: 444 bp = nfs_getcacheblk(vp, bn, biosize, p); 445 if (!bp) 446 return (EINTR); 447 if (bp->b_wcred == NOCRED) { 448 crhold(cred); 449 bp->b_wcred = cred; 450 } 451 np->n_flag |= NMODIFIED; 452 if (uio->uio_offset + n > np->n_size) { 453 np->n_size = uio->uio_offset + n; 454 vnode_pager_setsize(vp, (u_long)np->n_size); 455 } 456 457 /* 458 * If the new write will leave a contiguous dirty 459 * area, just update the b_dirtyoff and b_dirtyend, 460 * otherwise force a write rpc of the old dirty area. 461 */ 462 if (bp->b_dirtyend > 0 && 463 (on > bp->b_dirtyend || (on + n) < bp->b_dirtyoff)) { 464 bp->b_proc = p; 465 if (VOP_BWRITE(bp) == EINTR) 466 return (EINTR); 467 goto again; 468 } 469 470 /* 471 * Check for valid write lease and get one as required. 472 * In case getblk() and/or bwrite() delayed us. 473 */ 474 if ((nmp->nm_flag & NFSMNT_NQNFS) && 475 NQNFS_CKINVALID(vp, np, NQL_WRITE)) { 476 do { 477 error = nqnfs_getlease(vp, NQL_WRITE, cred, p); 478 } while (error == NQNFS_EXPIRED); 479 if (error) { 480 brelse(bp); 481 return (error); 482 } 483 if (np->n_lrev != np->n_brev || 484 (np->n_flag & NQNFSNONCACHE)) { 485 brelse(bp); 486 if (error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1)) 487 return (error); 488 np->n_brev = np->n_lrev; 489 goto again; 490 } 491 } 492 if (error = uiomove((char *)bp->b_data + on, n, uio)) { 493 bp->b_flags |= B_ERROR; 494 brelse(bp); 495 return (error); 496 } 497 if (bp->b_dirtyend > 0) { 498 bp->b_dirtyoff = min(on, bp->b_dirtyoff); 499 bp->b_dirtyend = max((on + n), bp->b_dirtyend); 500 } else { 501 bp->b_dirtyoff = on; 502 bp->b_dirtyend = on + n; 503 } 504 #ifndef notdef 505 if (bp->b_validend == 0 || bp->b_validend < bp->b_dirtyoff || 506 bp->b_validoff > bp->b_dirtyend) { 507 bp->b_validoff = bp->b_dirtyoff; 508 bp->b_validend = bp->b_dirtyend; 509 } else { 510 bp->b_validoff = min(bp->b_validoff, bp->b_dirtyoff); 511 bp->b_validend = max(bp->b_validend, bp->b_dirtyend); 512 } 513 #else 514 bp->b_validoff = bp->b_dirtyoff; 515 bp->b_validend = bp->b_dirtyend; 516 #endif 517 if (ioflag & IO_APPEND) 518 bp->b_flags |= B_APPENDWRITE; 519 520 /* 521 * If the lease is non-cachable or IO_SYNC do bwrite(). 522 */ 523 if ((np->n_flag & NQNFSNONCACHE) || (ioflag & IO_SYNC)) { 524 bp->b_proc = p; 525 if (error = VOP_BWRITE(bp)) 526 return (error); 527 } else if ((n + on) == biosize && 528 (nmp->nm_flag & NFSMNT_NQNFS) == 0) { 529 bp->b_proc = (struct proc *)0; 530 bawrite(bp); 531 } else 532 bdwrite(bp); 533 } while (uio->uio_resid > 0 && n > 0); 534 return (0); 535 } 536 537 /* 538 * Get an nfs cache block. 539 * Allocate a new one if the block isn't currently in the cache 540 * and return the block marked busy. If the calling process is 541 * interrupted by a signal for an interruptible mount point, return 542 * NULL. 543 */ 544 struct buf * 545 nfs_getcacheblk(vp, bn, size, p) 546 struct vnode *vp; 547 daddr_t bn; 548 int size; 549 struct proc *p; 550 { 551 register struct buf *bp; 552 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 553 554 if (nmp->nm_flag & NFSMNT_INT) { 555 bp = getblk(vp, bn, size, PCATCH, 0); 556 while (bp == (struct buf *)0) { 557 if (nfs_sigintr(nmp, (struct nfsreq *)0, p)) 558 return ((struct buf *)0); 559 bp = getblk(vp, bn, size, 0, 2 * hz); 560 } 561 } else 562 bp = getblk(vp, bn, size, 0, 0); 563 return (bp); 564 } 565 566 /* 567 * Flush and invalidate all dirty buffers. If another process is already 568 * doing the flush, just wait for completion. 569 */ 570 nfs_vinvalbuf(vp, flags, cred, p, intrflg) 571 struct vnode *vp; 572 int flags; 573 struct ucred *cred; 574 struct proc *p; 575 int intrflg; 576 { 577 register struct nfsnode *np = VTONFS(vp); 578 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 579 int error = 0, slpflag, slptimeo; 580 581 if ((nmp->nm_flag & NFSMNT_INT) == 0) 582 intrflg = 0; 583 if (intrflg) { 584 slpflag = PCATCH; 585 slptimeo = 2 * hz; 586 } else { 587 slpflag = 0; 588 slptimeo = 0; 589 } 590 /* 591 * First wait for any other process doing a flush to complete. 592 */ 593 while (np->n_flag & NFLUSHINPROG) { 594 np->n_flag |= NFLUSHWANT; 595 error = tsleep((caddr_t)&np->n_flag, PRIBIO + 2, "nfsvinval", 596 slptimeo); 597 if (error && intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) 598 return (EINTR); 599 } 600 601 /* 602 * Now, flush as required. 603 */ 604 np->n_flag |= NFLUSHINPROG; 605 error = vinvalbuf(vp, flags, cred, p, slpflag, 0); 606 while (error) { 607 if (intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) { 608 np->n_flag &= ~NFLUSHINPROG; 609 if (np->n_flag & NFLUSHWANT) { 610 np->n_flag &= ~NFLUSHWANT; 611 wakeup((caddr_t)&np->n_flag); 612 } 613 return (EINTR); 614 } 615 error = vinvalbuf(vp, flags, cred, p, 0, slptimeo); 616 } 617 np->n_flag &= ~(NMODIFIED | NFLUSHINPROG); 618 if (np->n_flag & NFLUSHWANT) { 619 np->n_flag &= ~NFLUSHWANT; 620 wakeup((caddr_t)&np->n_flag); 621 } 622 return (0); 623 } 624 625 /* 626 * Initiate asynchronous I/O. Return an error if no nfsiods are available. 627 * This is mainly to avoid queueing async I/O requests when the nfsiods 628 * are all hung on a dead server. 629 */ 630 nfs_asyncio(bp, cred) 631 register struct buf *bp; 632 struct ucred *cred; 633 { 634 register int i; 635 636 if (nfs_numasync == 0) 637 return (EIO); 638 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 639 if (nfs_iodwant[i]) { 640 if (bp->b_flags & B_READ) { 641 if (bp->b_rcred == NOCRED && cred != NOCRED) { 642 crhold(cred); 643 bp->b_rcred = cred; 644 } 645 } else { 646 if (bp->b_wcred == NOCRED && cred != NOCRED) { 647 crhold(cred); 648 bp->b_wcred = cred; 649 } 650 } 651 652 TAILQ_INSERT_TAIL(&nfs_bufq, bp, b_freelist); 653 nfs_iodwant[i] = (struct proc *)0; 654 wakeup((caddr_t)&nfs_iodwant[i]); 655 return (0); 656 } 657 return (EIO); 658 } 659 660 /* 661 * Do an I/O operation to/from a cache block. This may be called 662 * synchronously or from an nfsiod. 663 */ 664 int 665 nfs_doio(bp, cr, p) 666 register struct buf *bp; 667 struct cred *cr; 668 struct proc *p; 669 { 670 register struct uio *uiop; 671 register struct vnode *vp; 672 struct nfsnode *np; 673 struct nfsmount *nmp; 674 int error, diff, len; 675 struct uio uio; 676 struct iovec io; 677 678 vp = bp->b_vp; 679 np = VTONFS(vp); 680 nmp = VFSTONFS(vp->v_mount); 681 uiop = &uio; 682 uiop->uio_iov = &io; 683 uiop->uio_iovcnt = 1; 684 uiop->uio_segflg = UIO_SYSSPACE; 685 uiop->uio_procp = p; 686 687 /* 688 * Historically, paging was done with physio, but no more. 689 */ 690 if (bp->b_flags & B_PHYS) 691 panic("doio phys"); 692 if (bp->b_flags & B_READ) { 693 io.iov_len = uiop->uio_resid = bp->b_bcount; 694 io.iov_base = bp->b_data; 695 uiop->uio_rw = UIO_READ; 696 switch (vp->v_type) { 697 case VREG: 698 uiop->uio_offset = bp->b_blkno * DEV_BSIZE; 699 nfsstats.read_bios++; 700 error = nfs_readrpc(vp, uiop, cr); 701 if (!error) { 702 bp->b_validoff = 0; 703 if (uiop->uio_resid) { 704 /* 705 * If len > 0, there is a hole in the file and 706 * no writes after the hole have been pushed to 707 * the server yet. 708 * Just zero fill the rest of the valid area. 709 */ 710 diff = bp->b_bcount - uiop->uio_resid; 711 len = np->n_size - (bp->b_blkno * DEV_BSIZE 712 + diff); 713 if (len > 0) { 714 len = min(len, uiop->uio_resid); 715 bzero((char *)bp->b_data + diff, len); 716 bp->b_validend = diff + len; 717 } else 718 bp->b_validend = diff; 719 } else 720 bp->b_validend = bp->b_bcount; 721 } 722 if (p && (vp->v_flag & VTEXT) && 723 (((nmp->nm_flag & NFSMNT_NQNFS) && 724 np->n_lrev != np->n_brev) || 725 (!(nmp->nm_flag & NFSMNT_NQNFS) && 726 np->n_mtime != np->n_vattr.va_mtime.ts_sec))) { 727 uprintf("Process killed due to text file modification\n"); 728 psignal(p, SIGKILL); 729 p->p_holdcnt++; 730 } 731 break; 732 case VLNK: 733 uiop->uio_offset = 0; 734 nfsstats.readlink_bios++; 735 error = nfs_readlinkrpc(vp, uiop, cr); 736 break; 737 case VDIR: 738 uiop->uio_offset = bp->b_lblkno; 739 nfsstats.readdir_bios++; 740 if (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) 741 error = nfs_readdirlookrpc(vp, uiop, cr); 742 else 743 error = nfs_readdirrpc(vp, uiop, cr); 744 /* 745 * Save offset cookie in b_blkno. 746 */ 747 bp->b_blkno = uiop->uio_offset; 748 break; 749 }; 750 if (error) { 751 bp->b_flags |= B_ERROR; 752 bp->b_error = error; 753 } 754 } else { 755 io.iov_len = uiop->uio_resid = bp->b_dirtyend 756 - bp->b_dirtyoff; 757 uiop->uio_offset = (bp->b_blkno * DEV_BSIZE) 758 + bp->b_dirtyoff; 759 io.iov_base = (char *)bp->b_data + bp->b_dirtyoff; 760 uiop->uio_rw = UIO_WRITE; 761 nfsstats.write_bios++; 762 if (bp->b_flags & B_APPENDWRITE) 763 error = nfs_writerpc(vp, uiop, cr, IO_APPEND); 764 else 765 error = nfs_writerpc(vp, uiop, cr, 0); 766 bp->b_flags &= ~(B_WRITEINPROG | B_APPENDWRITE); 767 768 /* 769 * For an interrupted write, the buffer is still valid and the 770 * write hasn't been pushed to the server yet, so we can't set 771 * B_ERROR and report the interruption by setting B_EINTR. For 772 * the B_ASYNC case, B_EINTR is not relevant, so the rpc attempt 773 * is essentially a noop. 774 */ 775 if (error == EINTR) { 776 bp->b_flags &= ~B_INVAL; 777 bp->b_flags |= B_DELWRI; 778 779 /* 780 * Since for the B_ASYNC case, nfs_bwrite() has reassigned the 781 * buffer to the clean list, we have to reassign it back to the 782 * dirty one. Ugh. 783 */ 784 if (bp->b_flags & B_ASYNC) 785 reassignbuf(bp, vp); 786 else 787 bp->b_flags |= B_EINTR; 788 } else { 789 if (error) { 790 bp->b_flags |= B_ERROR; 791 bp->b_error = np->n_error = error; 792 np->n_flag |= NWRITEERR; 793 } 794 bp->b_dirtyoff = bp->b_dirtyend = 0; 795 } 796 } 797 bp->b_resid = uiop->uio_resid; 798 biodone(bp); 799 return (error); 800 } 801
Indexes created Mon Oct 13 00:09:54 GMT 2025