nfs_bio.c revision 1.63.2.14
1 /* $NetBSD: nfs_bio.c,v 1.63.2.14 2002/06/24 22:12:03 nathanw 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 8.9 (Berkeley) 3/30/95 39 */ 40 41 #include <sys/cdefs.h> 42 __KERNEL_RCSID(0, "$NetBSD: nfs_bio.c,v 1.63.2.14 2002/06/24 22:12:03 nathanw Exp $"); 43 44 #include "opt_nfs.h" 45 #include "opt_ddb.h" 46 47 #include <sys/param.h> 48 #include <sys/systm.h> 49 #include <sys/resourcevar.h> 50 #include <sys/signalvar.h> 51 #include <sys/lwp.h> 52 #include <sys/proc.h> 53 #include <sys/buf.h> 54 #include <sys/vnode.h> 55 #include <sys/mount.h> 56 #include <sys/kernel.h> 57 #include <sys/namei.h> 58 #include <sys/dirent.h> 59 #include <sys/malloc.h> 60 61 #include <uvm/uvm_extern.h> 62 #include <uvm/uvm.h> 63 64 #include <nfs/rpcv2.h> 65 #include <nfs/nfsproto.h> 66 #include <nfs/nfs.h> 67 #include <nfs/nfsmount.h> 68 #include <nfs/nqnfs.h> 69 #include <nfs/nfsnode.h> 70 #include <nfs/nfs_var.h> 71 72 extern int nfs_numasync; 73 extern int nfs_commitsize; 74 extern struct nfsstats nfsstats; 75 76 /* 77 * Vnode op for read using bio 78 * Any similarity to readip() is purely coincidental 79 */ 80 int 81 nfs_bioread(vp, uio, ioflag, cred, cflag) 82 struct vnode *vp; 83 struct uio *uio; 84 int ioflag, cflag; 85 struct ucred *cred; 86 { 87 struct nfsnode *np = VTONFS(vp); 88 struct buf *bp = NULL, *rabp; 89 struct vattr vattr; 90 struct proc *p; 91 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 92 struct nfsdircache *ndp = NULL, *nndp = NULL; 93 caddr_t baddr, ep, edp; 94 int got_buf = 0, error = 0, n = 0, on = 0, en, enn; 95 int enough = 0; 96 struct dirent *dp, *pdp; 97 off_t curoff = 0; 98 99 #ifdef DIAGNOSTIC 100 if (uio->uio_rw != UIO_READ) 101 panic("nfs_read mode"); 102 #endif 103 if (uio->uio_resid == 0) 104 return (0); 105 if (vp->v_type != VDIR && uio->uio_offset < 0) 106 return (EINVAL); 107 p = uio->uio_procp; 108 #ifndef NFS_V2_ONLY 109 if ((nmp->nm_flag & NFSMNT_NFSV3) && 110 !(nmp->nm_iflag & NFSMNT_GOTFSINFO)) 111 (void)nfs_fsinfo(nmp, vp, cred, p); 112 #endif 113 if (vp->v_type != VDIR && 114 (uio->uio_offset + uio->uio_resid) > nmp->nm_maxfilesize) 115 return (EFBIG); 116 117 /* 118 * For nfs, cache consistency can only be maintained approximately. 119 * Although RFC1094 does not specify the criteria, the following is 120 * believed to be compatible with the reference port. 121 * For nqnfs, full cache consistency is maintained within the loop. 122 * For nfs: 123 * If the file's modify time on the server has changed since the 124 * last read rpc or you have written to the file, 125 * you may have lost data cache consistency with the 126 * server, so flush all of the file's data out of the cache. 127 * Then force a getattr rpc to ensure that you have up to date 128 * attributes. 129 * NB: This implies that cache data can be read when up to 130 * NFS_ATTRTIMEO seconds out of date. If you find that you need current 131 * attributes this could be forced by setting n_attrstamp to 0 before 132 * the VOP_GETATTR() call. 133 */ 134 135 if ((nmp->nm_flag & NFSMNT_NQNFS) == 0 && vp->v_type != VLNK) { 136 if (np->n_flag & NMODIFIED) { 137 if (vp->v_type != VREG) { 138 if (vp->v_type != VDIR) 139 panic("nfs: bioread, not dir"); 140 nfs_invaldircache(vp, 0); 141 np->n_direofoffset = 0; 142 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 143 if (error) 144 return (error); 145 } 146 np->n_attrstamp = 0; 147 error = VOP_GETATTR(vp, &vattr, cred, p); 148 if (error) 149 return (error); 150 np->n_mtime = vattr.va_mtime.tv_sec; 151 } else { 152 error = VOP_GETATTR(vp, &vattr, cred, p); 153 if (error) 154 return (error); 155 if (np->n_mtime != vattr.va_mtime.tv_sec) { 156 if (vp->v_type == VDIR) { 157 nfs_invaldircache(vp, 0); 158 np->n_direofoffset = 0; 159 } 160 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 161 if (error) 162 return (error); 163 np->n_mtime = vattr.va_mtime.tv_sec; 164 } 165 } 166 } 167 168 /* 169 * update the cached read creds for this node. 170 */ 171 172 if (np->n_rcred) { 173 crfree(np->n_rcred); 174 } 175 np->n_rcred = cred; 176 crhold(cred); 177 178 do { 179 #ifndef NFS_V2_ONLY 180 /* 181 * Get a valid lease. If cached data is stale, flush it. 182 */ 183 if (nmp->nm_flag & NFSMNT_NQNFS) { 184 if (NQNFS_CKINVALID(vp, np, ND_READ)) { 185 do { 186 error = nqnfs_getlease(vp, ND_READ, cred, p); 187 } while (error == NQNFS_EXPIRED); 188 if (error) 189 return (error); 190 if (np->n_lrev != np->n_brev || 191 (np->n_flag & NQNFSNONCACHE) || 192 ((np->n_flag & NMODIFIED) && vp->v_type == VDIR)) { 193 if (vp->v_type == VDIR) { 194 nfs_invaldircache(vp, 0); 195 np->n_direofoffset = 0; 196 } 197 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 198 if (error) 199 return (error); 200 np->n_brev = np->n_lrev; 201 } 202 } else if (vp->v_type == VDIR && (np->n_flag & NMODIFIED)) { 203 nfs_invaldircache(vp, 0); 204 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 205 np->n_direofoffset = 0; 206 if (error) 207 return (error); 208 } 209 } 210 #endif 211 /* 212 * Don't cache symlinks. 213 */ 214 if (np->n_flag & NQNFSNONCACHE 215 || ((vp->v_flag & VROOT) && vp->v_type == VLNK)) { 216 switch (vp->v_type) { 217 case VREG: 218 return (nfs_readrpc(vp, uio)); 219 case VLNK: 220 return (nfs_readlinkrpc(vp, uio, cred)); 221 case VDIR: 222 break; 223 default: 224 printf(" NQNFSNONCACHE: type %x unexpected\n", 225 vp->v_type); 226 }; 227 } 228 baddr = (caddr_t)0; 229 switch (vp->v_type) { 230 case VREG: 231 nfsstats.biocache_reads++; 232 233 error = 0; 234 if (uio->uio_offset >= np->n_size) { 235 break; 236 } 237 while (uio->uio_resid > 0) { 238 void *win; 239 vsize_t bytelen = MIN(np->n_size - uio->uio_offset, 240 uio->uio_resid); 241 242 if (bytelen == 0) 243 break; 244 win = ubc_alloc(&vp->v_uobj, uio->uio_offset, 245 &bytelen, UBC_READ); 246 error = uiomove(win, bytelen, uio); 247 ubc_release(win, 0); 248 if (error) { 249 break; 250 } 251 } 252 n = 0; 253 break; 254 255 case VLNK: 256 nfsstats.biocache_readlinks++; 257 bp = nfs_getcacheblk(vp, (daddr_t)0, NFS_MAXPATHLEN, p); 258 if (!bp) 259 return (EINTR); 260 if ((bp->b_flags & B_DONE) == 0) { 261 bp->b_flags |= B_READ; 262 error = nfs_doio(bp, p); 263 if (error) { 264 brelse(bp); 265 return (error); 266 } 267 } 268 n = MIN(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid); 269 got_buf = 1; 270 on = 0; 271 break; 272 case VDIR: 273 diragain: 274 nfsstats.biocache_readdirs++; 275 ndp = nfs_searchdircache(vp, uio->uio_offset, 276 (nmp->nm_flag & NFSMNT_XLATECOOKIE), 0); 277 if (!ndp) { 278 /* 279 * We've been handed a cookie that is not 280 * in the cache. If we're not translating 281 * 32 <-> 64, it may be a value that was 282 * flushed out of the cache because it grew 283 * too big. Let the server judge if it's 284 * valid or not. In the translation case, 285 * we have no way of validating this value, 286 * so punt. 287 */ 288 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) 289 return (EINVAL); 290 ndp = nfs_enterdircache(vp, uio->uio_offset, 291 uio->uio_offset, 0, 0); 292 } 293 294 if (uio->uio_offset != 0 && 295 ndp->dc_cookie == np->n_direofoffset) { 296 nfsstats.direofcache_hits++; 297 return (0); 298 } 299 300 bp = nfs_getcacheblk(vp, ndp->dc_blkno, NFS_DIRBLKSIZ, p); 301 if (!bp) 302 return (EINTR); 303 if ((bp->b_flags & B_DONE) == 0) { 304 bp->b_flags |= B_READ; 305 bp->b_dcookie = ndp->dc_blkcookie; 306 error = nfs_doio(bp, p); 307 if (error) { 308 /* 309 * Yuck! The directory has been modified on the 310 * server. Punt and let the userland code 311 * deal with it. 312 */ 313 brelse(bp); 314 if (error == NFSERR_BAD_COOKIE) { 315 nfs_invaldircache(vp, 0); 316 nfs_vinvalbuf(vp, 0, cred, p, 1); 317 error = EINVAL; 318 } 319 return (error); 320 } 321 } 322 323 /* 324 * Just return if we hit EOF right away with this 325 * block. Always check here, because direofoffset 326 * may have been set by an nfsiod since the last 327 * check. 328 */ 329 if (np->n_direofoffset != 0 && 330 ndp->dc_blkcookie == np->n_direofoffset) { 331 brelse(bp); 332 return (0); 333 } 334 335 /* 336 * Find the entry we were looking for in the block. 337 */ 338 339 en = ndp->dc_entry; 340 341 pdp = dp = (struct dirent *)bp->b_data; 342 edp = bp->b_data + bp->b_bcount - bp->b_resid; 343 enn = 0; 344 while (enn < en && (caddr_t)dp < edp) { 345 pdp = dp; 346 dp = (struct dirent *)((caddr_t)dp + dp->d_reclen); 347 enn++; 348 } 349 350 /* 351 * If the entry number was bigger than the number of 352 * entries in the block, or the cookie of the previous 353 * entry doesn't match, the directory cache is 354 * stale. Flush it and try again (i.e. go to 355 * the server). 356 */ 357 if ((caddr_t)dp >= edp || (caddr_t)dp + dp->d_reclen > edp || 358 (en > 0 && NFS_GETCOOKIE(pdp) != ndp->dc_cookie)) { 359 #ifdef DEBUG 360 printf("invalid cache: %p %p %p off %lx %lx\n", 361 pdp, dp, edp, 362 (unsigned long)uio->uio_offset, 363 (unsigned long)NFS_GETCOOKIE(pdp)); 364 #endif 365 brelse(bp); 366 nfs_invaldircache(vp, 0); 367 nfs_vinvalbuf(vp, 0, cred, p, 0); 368 goto diragain; 369 } 370 371 on = (caddr_t)dp - bp->b_data; 372 373 /* 374 * Cache all entries that may be exported to the 375 * user, as they may be thrown back at us. The 376 * NFSBIO_CACHECOOKIES flag indicates that all 377 * entries are being 'exported', so cache them all. 378 */ 379 380 if (en == 0 && pdp == dp) { 381 dp = (struct dirent *) 382 ((caddr_t)dp + dp->d_reclen); 383 enn++; 384 } 385 386 if (uio->uio_resid < (bp->b_bcount - bp->b_resid - on)) { 387 n = uio->uio_resid; 388 enough = 1; 389 } else 390 n = bp->b_bcount - bp->b_resid - on; 391 392 ep = bp->b_data + on + n; 393 394 /* 395 * Find last complete entry to copy, caching entries 396 * (if requested) as we go. 397 */ 398 399 while ((caddr_t)dp < ep && (caddr_t)dp + dp->d_reclen <= ep) { 400 if (cflag & NFSBIO_CACHECOOKIES) { 401 nndp = nfs_enterdircache(vp, NFS_GETCOOKIE(pdp), 402 ndp->dc_blkcookie, enn, bp->b_lblkno); 403 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) { 404 NFS_STASHCOOKIE32(pdp, 405 nndp->dc_cookie32); 406 } 407 } 408 pdp = dp; 409 dp = (struct dirent *)((caddr_t)dp + dp->d_reclen); 410 enn++; 411 } 412 413 /* 414 * If the last requested entry was not the last in the 415 * buffer (happens if NFS_DIRFRAGSIZ < NFS_DIRBLKSIZ), 416 * cache the cookie of the last requested one, and 417 * set of the offset to it. 418 */ 419 420 if ((on + n) < bp->b_bcount - bp->b_resid) { 421 curoff = NFS_GETCOOKIE(pdp); 422 nndp = nfs_enterdircache(vp, curoff, ndp->dc_blkcookie, 423 enn, bp->b_lblkno); 424 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) { 425 NFS_STASHCOOKIE32(pdp, nndp->dc_cookie32); 426 curoff = nndp->dc_cookie32; 427 } 428 } else 429 curoff = bp->b_dcookie; 430 431 /* 432 * Always cache the entry for the next block, 433 * so that readaheads can use it. 434 */ 435 nndp = nfs_enterdircache(vp, bp->b_dcookie, bp->b_dcookie, 0,0); 436 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) { 437 if (curoff == bp->b_dcookie) { 438 NFS_STASHCOOKIE32(pdp, nndp->dc_cookie32); 439 curoff = nndp->dc_cookie32; 440 } 441 } 442 443 n = ((caddr_t)pdp + pdp->d_reclen) - (bp->b_data + on); 444 445 /* 446 * If not eof and read aheads are enabled, start one. 447 * (You need the current block first, so that you have the 448 * directory offset cookie of the next block.) 449 */ 450 if (nfs_numasync > 0 && nmp->nm_readahead > 0 && 451 np->n_direofoffset == 0 && !(np->n_flag & NQNFSNONCACHE)) { 452 rabp = nfs_getcacheblk(vp, nndp->dc_blkno, 453 NFS_DIRBLKSIZ, p); 454 if (rabp) { 455 if ((rabp->b_flags & (B_DONE | B_DELWRI)) == 0) { 456 rabp->b_dcookie = nndp->dc_cookie; 457 rabp->b_flags |= (B_READ | B_ASYNC); 458 if (nfs_asyncio(rabp)) { 459 rabp->b_flags |= B_INVAL; 460 brelse(rabp); 461 } 462 } else 463 brelse(rabp); 464 } 465 } 466 got_buf = 1; 467 break; 468 default: 469 printf(" nfsbioread: type %x unexpected\n",vp->v_type); 470 break; 471 } 472 473 if (n > 0) { 474 if (!baddr) 475 baddr = bp->b_data; 476 error = uiomove(baddr + on, (int)n, uio); 477 } 478 switch (vp->v_type) { 479 case VREG: 480 break; 481 case VLNK: 482 n = 0; 483 break; 484 case VDIR: 485 if (np->n_flag & NQNFSNONCACHE) 486 bp->b_flags |= B_INVAL; 487 uio->uio_offset = curoff; 488 if (enough) 489 n = 0; 490 break; 491 default: 492 printf(" nfsbioread: type %x unexpected\n",vp->v_type); 493 } 494 if (got_buf) 495 brelse(bp); 496 } while (error == 0 && uio->uio_resid > 0 && n > 0); 497 return (error); 498 } 499 500 /* 501 * Vnode op for write using bio 502 */ 503 int 504 nfs_write(v) 505 void *v; 506 { 507 struct vop_write_args /* { 508 struct vnode *a_vp; 509 struct uio *a_uio; 510 int a_ioflag; 511 struct ucred *a_cred; 512 } */ *ap = v; 513 struct uio *uio = ap->a_uio; 514 struct proc *p = uio->uio_procp; 515 struct vnode *vp = ap->a_vp; 516 struct nfsnode *np = VTONFS(vp); 517 struct ucred *cred = ap->a_cred; 518 int ioflag = ap->a_ioflag; 519 struct vattr vattr; 520 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 521 void *win; 522 voff_t oldoff, origoff; 523 vsize_t bytelen; 524 int error = 0, iomode, must_commit; 525 526 #ifdef DIAGNOSTIC 527 if (uio->uio_rw != UIO_WRITE) 528 panic("nfs_write mode"); 529 if (uio->uio_segflg == UIO_USERSPACE && 530 uio->uio_procp != curproc) 531 panic("nfs_write proc"); 532 #endif 533 if (vp->v_type != VREG) 534 return (EIO); 535 if (np->n_flag & NWRITEERR) { 536 np->n_flag &= ~NWRITEERR; 537 return (np->n_error); 538 } 539 #ifndef NFS_V2_ONLY 540 if ((nmp->nm_flag & NFSMNT_NFSV3) && 541 !(nmp->nm_iflag & NFSMNT_GOTFSINFO)) 542 (void)nfs_fsinfo(nmp, vp, cred, p); 543 #endif 544 if (ioflag & (IO_APPEND | IO_SYNC)) { 545 if (np->n_flag & NMODIFIED) { 546 np->n_attrstamp = 0; 547 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 548 if (error) 549 return (error); 550 } 551 if (ioflag & IO_APPEND) { 552 np->n_attrstamp = 0; 553 error = VOP_GETATTR(vp, &vattr, cred, p); 554 if (error) 555 return (error); 556 uio->uio_offset = np->n_size; 557 } 558 } 559 if (uio->uio_offset < 0) 560 return (EINVAL); 561 if ((uio->uio_offset + uio->uio_resid) > nmp->nm_maxfilesize) 562 return (EFBIG); 563 if (uio->uio_resid == 0) 564 return (0); 565 /* 566 * Maybe this should be above the vnode op call, but so long as 567 * file servers have no limits, i don't think it matters 568 */ 569 if (p && uio->uio_offset + uio->uio_resid > 570 p->p_rlimit[RLIMIT_FSIZE].rlim_cur) { 571 psignal(p, SIGXFSZ); 572 return (EFBIG); 573 } 574 575 /* 576 * update the cached write creds for this node. 577 */ 578 579 if (np->n_wcred) { 580 crfree(np->n_wcred); 581 } 582 np->n_wcred = cred; 583 crhold(cred); 584 585 if ((np->n_flag & NQNFSNONCACHE) && uio->uio_iovcnt == 1) { 586 iomode = NFSV3WRITE_FILESYNC; 587 error = nfs_writerpc(vp, uio, &iomode, &must_commit); 588 if (must_commit) 589 nfs_clearcommit(vp->v_mount); 590 return (error); 591 } 592 593 origoff = uio->uio_offset; 594 do { 595 oldoff = uio->uio_offset; 596 bytelen = uio->uio_resid; 597 598 #ifndef NFS_V2_ONLY 599 /* 600 * Check for a valid write lease. 601 */ 602 if ((nmp->nm_flag & NFSMNT_NQNFS) && 603 NQNFS_CKINVALID(vp, np, ND_WRITE)) { 604 do { 605 error = nqnfs_getlease(vp, ND_WRITE, cred, p); 606 } while (error == NQNFS_EXPIRED); 607 if (error) 608 return (error); 609 if (np->n_lrev != np->n_brev || 610 (np->n_flag & NQNFSNONCACHE)) { 611 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 612 if (error) 613 return (error); 614 np->n_brev = np->n_lrev; 615 } 616 } 617 #endif 618 nfsstats.biocache_writes++; 619 620 np->n_flag |= NMODIFIED; 621 if (np->n_size < uio->uio_offset + bytelen) { 622 np->n_size = uio->uio_offset + bytelen; 623 } 624 if ((uio->uio_offset & PAGE_MASK) == 0 && 625 (bytelen & PAGE_MASK) == 0 && 626 uio->uio_offset >= vp->v_size) { 627 win = ubc_alloc(&vp->v_uobj, uio->uio_offset, &bytelen, 628 UBC_WRITE | UBC_FAULTBUSY); 629 } else { 630 win = ubc_alloc(&vp->v_uobj, uio->uio_offset, &bytelen, 631 UBC_WRITE); 632 } 633 error = uiomove(win, bytelen, uio); 634 ubc_release(win, 0); 635 if (error) { 636 break; 637 } 638 639 /* 640 * update UVM's notion of the size now that we've 641 * copied the data into the vnode's pages. 642 */ 643 644 if (vp->v_size < uio->uio_offset) { 645 uvm_vnp_setsize(vp, uio->uio_offset); 646 } 647 648 if ((oldoff & ~(nmp->nm_wsize - 1)) != 649 (uio->uio_offset & ~(nmp->nm_wsize - 1))) { 650 simple_lock(&vp->v_interlock); 651 error = VOP_PUTPAGES(vp, 652 trunc_page(oldoff & ~(nmp->nm_wsize - 1)), 653 round_page((uio->uio_offset + nmp->nm_wsize - 1) & 654 ~(nmp->nm_wsize - 1)), PGO_CLEANIT); 655 } 656 } while (uio->uio_resid > 0); 657 if ((np->n_flag & NQNFSNONCACHE) || (ioflag & IO_SYNC)) { 658 simple_lock(&vp->v_interlock); 659 error = VOP_PUTPAGES(vp, 660 trunc_page(origoff & ~(nmp->nm_wsize - 1)), 661 round_page((uio->uio_offset + nmp->nm_wsize - 1) & 662 ~(nmp->nm_wsize - 1)), 663 PGO_CLEANIT | PGO_SYNCIO); 664 } 665 return error; 666 } 667 668 /* 669 * Get an nfs cache block. 670 * Allocate a new one if the block isn't currently in the cache 671 * and return the block marked busy. If the calling process is 672 * interrupted by a signal for an interruptible mount point, return 673 * NULL. 674 */ 675 struct buf * 676 nfs_getcacheblk(vp, bn, size, p) 677 struct vnode *vp; 678 daddr_t bn; 679 int size; 680 struct proc *p; 681 { 682 struct buf *bp; 683 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 684 685 if (nmp->nm_flag & NFSMNT_INT) { 686 bp = getblk(vp, bn, size, PCATCH, 0); 687 while (bp == NULL) { 688 if (nfs_sigintr(nmp, NULL, p)) 689 return (NULL); 690 bp = getblk(vp, bn, size, 0, 2 * hz); 691 } 692 } else 693 bp = getblk(vp, bn, size, 0, 0); 694 return (bp); 695 } 696 697 /* 698 * Flush and invalidate all dirty buffers. If another process is already 699 * doing the flush, just wait for completion. 700 */ 701 int 702 nfs_vinvalbuf(vp, flags, cred, p, intrflg) 703 struct vnode *vp; 704 int flags; 705 struct ucred *cred; 706 struct proc *p; 707 int intrflg; 708 { 709 struct nfsnode *np = VTONFS(vp); 710 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 711 int error = 0, slpflag, slptimeo; 712 713 if ((nmp->nm_flag & NFSMNT_INT) == 0) 714 intrflg = 0; 715 if (intrflg) { 716 slpflag = PCATCH; 717 slptimeo = 2 * hz; 718 } else { 719 slpflag = 0; 720 slptimeo = 0; 721 } 722 /* 723 * First wait for any other process doing a flush to complete. 724 */ 725 while (np->n_flag & NFLUSHINPROG) { 726 np->n_flag |= NFLUSHWANT; 727 error = tsleep((caddr_t)&np->n_flag, PRIBIO + 2, "nfsvinval", 728 slptimeo); 729 if (error && intrflg && nfs_sigintr(nmp, NULL, p)) 730 return (EINTR); 731 } 732 733 /* 734 * Now, flush as required. 735 */ 736 np->n_flag |= NFLUSHINPROG; 737 error = vinvalbuf(vp, flags, cred, p, slpflag, 0); 738 while (error) { 739 if (intrflg && nfs_sigintr(nmp, NULL, p)) { 740 np->n_flag &= ~NFLUSHINPROG; 741 if (np->n_flag & NFLUSHWANT) { 742 np->n_flag &= ~NFLUSHWANT; 743 wakeup((caddr_t)&np->n_flag); 744 } 745 return (EINTR); 746 } 747 error = vinvalbuf(vp, flags, cred, p, 0, slptimeo); 748 } 749 np->n_flag &= ~(NMODIFIED | NFLUSHINPROG); 750 if (np->n_flag & NFLUSHWANT) { 751 np->n_flag &= ~NFLUSHWANT; 752 wakeup((caddr_t)&np->n_flag); 753 } 754 return (0); 755 } 756 757 /* 758 * Initiate asynchronous I/O. Return an error if no nfsiods are available. 759 * This is mainly to avoid queueing async I/O requests when the nfsiods 760 * are all hung on a dead server. 761 */ 762 763 int 764 nfs_asyncio(bp) 765 struct buf *bp; 766 { 767 int i; 768 struct nfsmount *nmp; 769 int gotiod, slpflag = 0, slptimeo = 0, error; 770 771 if (nfs_numasync == 0) 772 return (EIO); 773 774 nmp = VFSTONFS(bp->b_vp->v_mount); 775 again: 776 if (nmp->nm_flag & NFSMNT_INT) 777 slpflag = PCATCH; 778 gotiod = FALSE; 779 780 /* 781 * Find a free iod to process this request. 782 */ 783 784 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 785 if (nfs_iodwant[i]) { 786 /* 787 * Found one, so wake it up and tell it which 788 * mount to process. 789 */ 790 nfs_iodwant[i] = NULL; 791 nfs_iodmount[i] = nmp; 792 nmp->nm_bufqiods++; 793 wakeup((caddr_t)&nfs_iodwant[i]); 794 gotiod = TRUE; 795 break; 796 } 797 798 /* 799 * If none are free, we may already have an iod working on this mount 800 * point. If so, it will process our request. 801 */ 802 803 if (!gotiod && nmp->nm_bufqiods > 0) 804 gotiod = TRUE; 805 806 /* 807 * If we have an iod which can process the request, then queue 808 * the buffer. 809 */ 810 811 if (gotiod) { 812 813 /* 814 * Ensure that the queue never grows too large. 815 */ 816 817 while (nmp->nm_bufqlen >= 2*nfs_numasync) { 818 nmp->nm_bufqwant = TRUE; 819 error = tsleep(&nmp->nm_bufq, slpflag | PRIBIO, 820 "nfsaio", slptimeo); 821 if (error) { 822 if (nfs_sigintr(nmp, NULL, curproc)) 823 return (EINTR); 824 if (slpflag == PCATCH) { 825 slpflag = 0; 826 slptimeo = 2 * hz; 827 } 828 } 829 830 /* 831 * We might have lost our iod while sleeping, 832 * so check and loop if nescessary. 833 */ 834 835 if (nmp->nm_bufqiods == 0) 836 goto again; 837 } 838 TAILQ_INSERT_TAIL(&nmp->nm_bufq, bp, b_freelist); 839 nmp->nm_bufqlen++; 840 return (0); 841 } 842 843 /* 844 * All the iods are busy on other mounts, so return EIO to 845 * force the caller to process the i/o synchronously. 846 */ 847 848 return (EIO); 849 } 850 851 /* 852 * Do an I/O operation to/from a cache block. This may be called 853 * synchronously or from an nfsiod. 854 */ 855 int 856 nfs_doio(bp, p) 857 struct buf *bp; 858 struct proc *p; 859 { 860 struct uio *uiop; 861 struct vnode *vp; 862 struct nfsnode *np; 863 struct nfsmount *nmp; 864 int error = 0, diff, len, iomode, must_commit = 0; 865 int pushedrange; 866 struct uio uio; 867 struct iovec io; 868 off_t off, cnt; 869 struct uvm_object *uobj; 870 UVMHIST_FUNC("nfs_doio"); UVMHIST_CALLED(ubchist); 871 872 vp = bp->b_vp; 873 uobj = &vp->v_uobj; 874 np = VTONFS(vp); 875 nmp = VFSTONFS(vp->v_mount); 876 uiop = &uio; 877 uiop->uio_iov = &io; 878 uiop->uio_iovcnt = 1; 879 uiop->uio_segflg = UIO_SYSSPACE; 880 uiop->uio_procp = p; 881 882 /* 883 * Historically, paging was done with physio, but no more... 884 */ 885 if (bp->b_flags & B_PHYS) { 886 /* 887 * ...though reading /dev/drum still gets us here. 888 */ 889 io.iov_len = uiop->uio_resid = bp->b_bcount; 890 /* mapping was done by vmapbuf() */ 891 io.iov_base = bp->b_data; 892 uiop->uio_offset = ((off_t)bp->b_blkno) << DEV_BSHIFT; 893 if (bp->b_flags & B_READ) { 894 uiop->uio_rw = UIO_READ; 895 nfsstats.read_physios++; 896 error = nfs_readrpc(vp, uiop); 897 } else { 898 iomode = NFSV3WRITE_DATASYNC; 899 uiop->uio_rw = UIO_WRITE; 900 nfsstats.write_physios++; 901 error = nfs_writerpc(vp, uiop, &iomode, &must_commit); 902 } 903 if (error) { 904 bp->b_flags |= B_ERROR; 905 bp->b_error = error; 906 } 907 } else if (bp->b_flags & B_READ) { 908 io.iov_len = uiop->uio_resid = bp->b_bcount; 909 io.iov_base = bp->b_data; 910 uiop->uio_rw = UIO_READ; 911 switch (vp->v_type) { 912 case VREG: 913 uiop->uio_offset = ((off_t)bp->b_blkno) << DEV_BSHIFT; 914 nfsstats.read_bios++; 915 error = nfs_readrpc(vp, uiop); 916 if (!error && uiop->uio_resid) { 917 918 /* 919 * If len > 0, there is a hole in the file and 920 * no writes after the hole have been pushed to 921 * the server yet. 922 * Just zero fill the rest of the valid area. 923 */ 924 925 diff = bp->b_bcount - uiop->uio_resid; 926 len = np->n_size - ((((off_t)bp->b_blkno) << DEV_BSHIFT) 927 + diff); 928 if (len > 0) { 929 len = MIN(len, uiop->uio_resid); 930 memset((char *)bp->b_data + diff, 0, len); 931 } 932 } 933 if (p && (vp->v_flag & VTEXT) && 934 (((nmp->nm_flag & NFSMNT_NQNFS) && 935 NQNFS_CKINVALID(vp, np, ND_READ) && 936 np->n_lrev != np->n_brev) || 937 (!(nmp->nm_flag & NFSMNT_NQNFS) && 938 np->n_mtime != np->n_vattr->va_mtime.tv_sec))) { 939 uprintf("Process killed due to " 940 "text file modification\n"); 941 psignal(p, SIGKILL); 942 #if 0 /* XXX NJWLWP */ 943 p->p_holdcnt++; 944 #endif 945 } 946 break; 947 case VLNK: 948 uiop->uio_offset = (off_t)0; 949 nfsstats.readlink_bios++; 950 error = nfs_readlinkrpc(vp, uiop, curproc->p_ucred); 951 break; 952 case VDIR: 953 nfsstats.readdir_bios++; 954 uiop->uio_offset = bp->b_dcookie; 955 if (nmp->nm_flag & NFSMNT_RDIRPLUS) { 956 error = nfs_readdirplusrpc(vp, uiop, curproc->p_ucred); 957 if (error == NFSERR_NOTSUPP) 958 nmp->nm_flag &= ~NFSMNT_RDIRPLUS; 959 } 960 if ((nmp->nm_flag & NFSMNT_RDIRPLUS) == 0) 961 error = nfs_readdirrpc(vp, uiop, curproc->p_ucred); 962 if (!error) { 963 bp->b_dcookie = uiop->uio_offset; 964 } 965 break; 966 default: 967 printf("nfs_doio: type %x unexpected\n",vp->v_type); 968 break; 969 } 970 if (error) { 971 bp->b_flags |= B_ERROR; 972 bp->b_error = error; 973 } 974 } else { 975 int i, npages = bp->b_bufsize >> PAGE_SHIFT; 976 struct vm_page *pgs[npages]; 977 boolean_t needcommit = TRUE; 978 979 if ((bp->b_flags & B_ASYNC) != 0 && NFS_ISV3(vp)) { 980 iomode = NFSV3WRITE_UNSTABLE; 981 } else { 982 iomode = NFSV3WRITE_FILESYNC; 983 } 984 985 for (i = 0; i < npages; i++) { 986 pgs[i] = uvm_pageratop((vaddr_t)bp->b_data + 987 (i << PAGE_SHIFT)); 988 if ((pgs[i]->flags & PG_NEEDCOMMIT) == 0) { 989 needcommit = FALSE; 990 } 991 } 992 if (!needcommit && iomode == NFSV3WRITE_UNSTABLE) { 993 for (i = 0; i < npages; i++) { 994 pgs[i]->flags |= PG_NEEDCOMMIT | PG_RDONLY; 995 pmap_page_protect(pgs[i], VM_PROT_READ); 996 } 997 } 998 999 uiop->uio_offset = (((off_t)bp->b_blkno) << DEV_BSHIFT); 1000 off = uiop->uio_offset; 1001 cnt = bp->b_bcount; 1002 1003 /* 1004 * Send the data to the server if necessary, 1005 * otherwise just send a commit rpc. 1006 */ 1007 1008 if (needcommit) { 1009 1010 /* 1011 * If the buffer is in the range that we already committed, 1012 * there's nothing to do. 1013 * 1014 * If it's in the range that we need to commit, push the 1015 * whole range at once, otherwise only push the buffer. 1016 * In both these cases, acquire the commit lock to avoid 1017 * other processes modifying the range. 1018 */ 1019 1020 lockmgr(&np->n_commitlock, LK_EXCLUSIVE, NULL); 1021 if (!nfs_in_committed_range(vp, off, bp->b_bcount)) { 1022 if (nfs_in_tobecommitted_range(vp, off, bp->b_bcount)) { 1023 pushedrange = 1; 1024 off = np->n_pushlo; 1025 cnt = np->n_pushhi - np->n_pushlo; 1026 } else { 1027 pushedrange = 0; 1028 } 1029 error = nfs_commit(vp, off, cnt, curproc); 1030 if (error == 0) { 1031 if (pushedrange) { 1032 nfs_merge_commit_ranges(vp); 1033 } else { 1034 nfs_add_committed_range(vp, off, cnt); 1035 } 1036 } 1037 } 1038 lockmgr(&np->n_commitlock, LK_RELEASE, NULL); 1039 if (!error) { 1040 bp->b_resid = 0; 1041 simple_lock(&uobj->vmobjlock); 1042 for (i = 0; i < npages; i++) { 1043 pgs[i]->flags &= ~(PG_NEEDCOMMIT | PG_RDONLY); 1044 } 1045 simple_unlock(&uobj->vmobjlock); 1046 biodone(bp); 1047 return (0); 1048 } else if (error == NFSERR_STALEWRITEVERF) { 1049 nfs_clearcommit(bp->b_vp->v_mount); 1050 } 1051 } 1052 io.iov_base = bp->b_data; 1053 io.iov_len = uiop->uio_resid = bp->b_bcount; 1054 uiop->uio_rw = UIO_WRITE; 1055 nfsstats.write_bios++; 1056 error = nfs_writerpc(vp, uiop, &iomode, &must_commit); 1057 if (!error && iomode == NFSV3WRITE_UNSTABLE) { 1058 lockmgr(&np->n_commitlock, LK_EXCLUSIVE, NULL); 1059 nfs_add_tobecommitted_range(vp, off, cnt); 1060 simple_lock(&uobj->vmobjlock); 1061 for (i = 0; i < npages; i++) { 1062 pgs[i]->flags &= ~PG_CLEAN; 1063 } 1064 simple_unlock(&uobj->vmobjlock); 1065 if (np->n_pushhi - np->n_pushlo > nfs_commitsize) { 1066 off = np->n_pushlo; 1067 cnt = nfs_commitsize >> 1; 1068 error = nfs_commit(vp, off, cnt, curproc); 1069 if (!error) { 1070 nfs_add_committed_range(vp, off, cnt); 1071 nfs_del_tobecommitted_range(vp, off, cnt); 1072 } 1073 } 1074 lockmgr(&np->n_commitlock, LK_RELEASE, NULL); 1075 } else if (!error && needcommit) { 1076 lockmgr(&np->n_commitlock, LK_EXCLUSIVE, NULL); 1077 nfs_del_committed_range(vp, off, cnt); 1078 lockmgr(&np->n_commitlock, LK_RELEASE, NULL); 1079 simple_lock(&uobj->vmobjlock); 1080 for (i = 0; i < npages; i++) { 1081 pgs[i]->flags &= ~(PG_NEEDCOMMIT | PG_RDONLY); 1082 } 1083 simple_unlock(&uobj->vmobjlock); 1084 } 1085 } 1086 bp->b_resid = uiop->uio_resid; 1087 if (must_commit || (error == NFSERR_STALEWRITEVERF)) { 1088 nfs_clearcommit(vp->v_mount); 1089 } 1090 biodone(bp); 1091 return (error); 1092 } 1093 1094 /* 1095 * Vnode op for VM getpages. 1096 */ 1097 1098 int 1099 nfs_getpages(v) 1100 void *v; 1101 { 1102 struct vop_getpages_args /* { 1103 struct vnode *a_vp; 1104 voff_t a_offset; 1105 struct vm_page **a_m; 1106 int *a_count; 1107 int a_centeridx; 1108 vm_prot_t a_access_type; 1109 int a_advice; 1110 int a_flags; 1111 } */ *ap = v; 1112 1113 struct vnode *vp = ap->a_vp; 1114 struct uvm_object *uobj = &vp->v_uobj; 1115 struct nfsnode *np = VTONFS(vp); 1116 const int npages = *ap->a_count; 1117 struct vm_page *pg, **pgs, *opgs[npages]; 1118 off_t origoffset, len; 1119 int i, error; 1120 boolean_t v3 = NFS_ISV3(vp); 1121 boolean_t write = (ap->a_access_type & VM_PROT_WRITE) != 0; 1122 boolean_t locked = (ap->a_flags & PGO_LOCKED) != 0; 1123 1124 /* 1125 * update the cached read creds for this node. 1126 */ 1127 1128 if (np->n_rcred) { 1129 crfree(np->n_rcred); 1130 } 1131 np->n_rcred = curproc->p_ucred; 1132 crhold(np->n_rcred); 1133 1134 /* 1135 * call the genfs code to get the pages. `pgs' may be NULL 1136 * when doing read-ahead. 1137 */ 1138 1139 pgs = ap->a_m; 1140 if (write && locked && v3) { 1141 KASSERT(pgs != NULL); 1142 #ifdef DEBUG 1143 1144 /* 1145 * If PGO_LOCKED is set, real pages shouldn't exists 1146 * in the array. 1147 */ 1148 1149 for (i = 0; i < npages; i++) 1150 KDASSERT(pgs[i] == NULL || pgs[i] == PGO_DONTCARE); 1151 #endif 1152 memcpy(opgs, pgs, npages * sizeof(struct vm_pages *)); 1153 } 1154 error = genfs_getpages(v); 1155 if (error) { 1156 return (error); 1157 } 1158 1159 /* 1160 * for read faults where the nfs node is not yet marked NMODIFIED, 1161 * set PG_RDONLY on the pages so that we come back here if someone 1162 * tries to modify later via the mapping that will be entered for 1163 * this fault. 1164 */ 1165 1166 if (!write && (np->n_flag & NMODIFIED) == 0 && pgs != NULL) { 1167 if (!locked) { 1168 simple_lock(&uobj->vmobjlock); 1169 } 1170 for (i = 0; i < npages; i++) { 1171 pg = pgs[i]; 1172 if (pg == NULL || pg == PGO_DONTCARE) { 1173 continue; 1174 } 1175 pg->flags |= PG_RDONLY; 1176 } 1177 if (!locked) { 1178 simple_unlock(&uobj->vmobjlock); 1179 } 1180 } 1181 if (!write) { 1182 return (0); 1183 } 1184 1185 /* 1186 * this is a write fault, update the commit info. 1187 */ 1188 1189 origoffset = ap->a_offset; 1190 len = npages << PAGE_SHIFT; 1191 1192 if (v3) { 1193 error = lockmgr(&np->n_commitlock, 1194 LK_EXCLUSIVE | (locked ? LK_NOWAIT : 0), NULL); 1195 if (error) { 1196 KASSERT(locked != 0); 1197 1198 /* 1199 * Since PGO_LOCKED is set, we need to unbusy 1200 * all pages fetched by genfs_getpages() above, 1201 * tell the caller that there are no pages 1202 * available and put back original pgs array. 1203 */ 1204 1205 uvm_lock_pageq(); 1206 uvm_page_unbusy(pgs, npages); 1207 uvm_unlock_pageq(); 1208 *ap->a_count = 0; 1209 memcpy(pgs, opgs, 1210 npages * sizeof(struct vm_pages *)); 1211 return (error); 1212 } 1213 nfs_del_committed_range(vp, origoffset, len); 1214 nfs_del_tobecommitted_range(vp, origoffset, len); 1215 } 1216 np->n_flag |= NMODIFIED; 1217 if (!locked) { 1218 simple_lock(&uobj->vmobjlock); 1219 } 1220 for (i = 0; i < npages; i++) { 1221 pg = pgs[i]; 1222 if (pg == NULL || pg == PGO_DONTCARE) { 1223 continue; 1224 } 1225 pg->flags &= ~(PG_NEEDCOMMIT | PG_RDONLY); 1226 } 1227 if (!locked) { 1228 simple_unlock(&uobj->vmobjlock); 1229 } 1230 if (v3) { 1231 lockmgr(&np->n_commitlock, LK_RELEASE, NULL); 1232 } 1233 return (0); 1234 } 1235
Indexes created Sun Oct 12 20:09:57 GMT 2025