nfs_bio.c revision 1.45.8.2
1 /* $NetBSD: nfs_bio.c,v 1.45.8.2 2000/12/08 09:19:20 bouyer 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 "opt_nfs.h" 42 #include "opt_ddb.h" 43 44 #include <sys/param.h> 45 #include <sys/systm.h> 46 #include <sys/resourcevar.h> 47 #include <sys/signalvar.h> 48 #include <sys/proc.h> 49 #include <sys/buf.h> 50 #include <sys/vnode.h> 51 #include <sys/trace.h> 52 #include <sys/mount.h> 53 #include <sys/kernel.h> 54 #include <sys/namei.h> 55 #include <sys/dirent.h> 56 #include <sys/malloc.h> 57 58 #include <uvm/uvm_extern.h> 59 #include <uvm/uvm.h> 60 61 #include <nfs/rpcv2.h> 62 #include <nfs/nfsproto.h> 63 #include <nfs/nfs.h> 64 #include <nfs/nfsmount.h> 65 #include <nfs/nqnfs.h> 66 #include <nfs/nfsnode.h> 67 #include <nfs/nfs_var.h> 68 69 extern int nfs_numasync; 70 extern struct nfsstats nfsstats; 71 72 /* 73 * Vnode op for read using bio 74 * Any similarity to readip() is purely coincidental 75 */ 76 int 77 nfs_bioread(vp, uio, ioflag, cred, cflag) 78 struct vnode *vp; 79 struct uio *uio; 80 int ioflag, cflag; 81 struct ucred *cred; 82 { 83 struct nfsnode *np = VTONFS(vp); 84 int biosize; 85 struct buf *bp = NULL, *rabp; 86 struct vattr vattr; 87 struct proc *p; 88 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 89 struct nfsdircache *ndp = NULL, *nndp = NULL; 90 caddr_t baddr, ep, edp; 91 int got_buf = 0, error = 0, n = 0, on = 0, en, enn; 92 int enough = 0; 93 struct dirent *dp, *pdp; 94 off_t curoff = 0; 95 96 #ifdef DIAGNOSTIC 97 if (uio->uio_rw != UIO_READ) 98 panic("nfs_read mode"); 99 #endif 100 if (uio->uio_resid == 0) 101 return (0); 102 if (vp->v_type != VDIR && uio->uio_offset < 0) 103 return (EINVAL); 104 p = uio->uio_procp; 105 #ifndef NFS_V2_ONLY 106 if ((nmp->nm_flag & NFSMNT_NFSV3) && 107 !(nmp->nm_iflag & NFSMNT_GOTFSINFO)) 108 (void)nfs_fsinfo(nmp, vp, cred, p); 109 #endif 110 if (vp->v_type != VDIR && 111 (uio->uio_offset + uio->uio_resid) > nmp->nm_maxfilesize) 112 return (EFBIG); 113 biosize = nmp->nm_rsize; 114 115 /* 116 * For nfs, cache consistency can only be maintained approximately. 117 * Although RFC1094 does not specify the criteria, the following is 118 * believed to be compatible with the reference port. 119 * For nqnfs, full cache consistency is maintained within the loop. 120 * For nfs: 121 * If the file's modify time on the server has changed since the 122 * last read rpc or you have written to the file, 123 * you may have lost data cache consistency with the 124 * server, so flush all of the file's data out of the cache. 125 * Then force a getattr rpc to ensure that you have up to date 126 * attributes. 127 * NB: This implies that cache data can be read when up to 128 * NFS_ATTRTIMEO seconds out of date. If you find that you need current 129 * attributes this could be forced by setting n_attrstamp to 0 before 130 * the VOP_GETATTR() call. 131 */ 132 133 if ((nmp->nm_flag & NFSMNT_NQNFS) == 0 && vp->v_type != VLNK) { 134 if (np->n_flag & NMODIFIED) { 135 if (vp->v_type != VREG) { 136 if (vp->v_type != VDIR) 137 panic("nfs: bioread, not dir"); 138 nfs_invaldircache(vp, 0); 139 np->n_direofoffset = 0; 140 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 141 if (error) 142 return (error); 143 } 144 np->n_attrstamp = 0; 145 error = VOP_GETATTR(vp, &vattr, cred, p); 146 if (error) 147 return (error); 148 np->n_mtime = vattr.va_mtime.tv_sec; 149 } else { 150 error = VOP_GETATTR(vp, &vattr, cred, p); 151 if (error) 152 return (error); 153 if (np->n_mtime != vattr.va_mtime.tv_sec) { 154 if (vp->v_type == VDIR) { 155 nfs_invaldircache(vp, 0); 156 np->n_direofoffset = 0; 157 } 158 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 159 if (error) 160 return (error); 161 np->n_mtime = vattr.va_mtime.tv_sec; 162 } 163 } 164 } 165 166 /* 167 * update the cached read creds for this node. 168 */ 169 170 if (np->n_rcred) { 171 crfree(np->n_rcred); 172 } 173 np->n_rcred = cred; 174 crhold(cred); 175 176 do { 177 #ifndef NFS_V2_ONLY 178 /* 179 * Get a valid lease. If cached data is stale, flush it. 180 */ 181 if (nmp->nm_flag & NFSMNT_NQNFS) { 182 if (NQNFS_CKINVALID(vp, np, ND_READ)) { 183 do { 184 error = nqnfs_getlease(vp, ND_READ, cred, p); 185 } while (error == NQNFS_EXPIRED); 186 if (error) 187 return (error); 188 if (np->n_lrev != np->n_brev || 189 (np->n_flag & NQNFSNONCACHE) || 190 ((np->n_flag & NMODIFIED) && vp->v_type == VDIR)) { 191 if (vp->v_type == VDIR) { 192 nfs_invaldircache(vp, 0); 193 np->n_direofoffset = 0; 194 } 195 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 196 if (error) 197 return (error); 198 np->n_brev = np->n_lrev; 199 } 200 } else if (vp->v_type == VDIR && (np->n_flag & NMODIFIED)) { 201 nfs_invaldircache(vp, 0); 202 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 203 np->n_direofoffset = 0; 204 if (error) 205 return (error); 206 } 207 } 208 #endif 209 /* 210 * Don't cache symlinks. 211 */ 212 if (np->n_flag & NQNFSNONCACHE 213 || ((vp->v_flag & VROOT) && vp->v_type == VLNK)) { 214 switch (vp->v_type) { 215 case VREG: 216 return (nfs_readrpc(vp, uio)); 217 case VLNK: 218 return (nfs_readlinkrpc(vp, uio, cred)); 219 case VDIR: 220 break; 221 default: 222 printf(" NQNFSNONCACHE: type %x unexpected\n", 223 vp->v_type); 224 }; 225 } 226 baddr = (caddr_t)0; 227 switch (vp->v_type) { 228 case VREG: 229 nfsstats.biocache_reads++; 230 231 error = 0; 232 while (uio->uio_resid > 0) { 233 void *win; 234 vsize_t bytelen = min(np->n_size - uio->uio_offset, 235 uio->uio_resid); 236 237 if (bytelen == 0) 238 break; 239 win = ubc_alloc(&vp->v_uvm.u_obj, uio->uio_offset, 240 &bytelen, UBC_READ); 241 error = uiomove(win, bytelen, uio); 242 ubc_release(win, 0); 243 if (error) { 244 break; 245 } 246 } 247 n = 0; 248 break; 249 250 case VLNK: 251 nfsstats.biocache_readlinks++; 252 bp = nfs_getcacheblk(vp, (daddr_t)0, NFS_MAXPATHLEN, p); 253 if (!bp) 254 return (EINTR); 255 if ((bp->b_flags & B_DONE) == 0) { 256 bp->b_flags |= B_READ; 257 error = nfs_doio(bp, p); 258 if (error) { 259 brelse(bp); 260 return (error); 261 } 262 } 263 n = min(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid); 264 got_buf = 1; 265 on = 0; 266 break; 267 case VDIR: 268 diragain: 269 nfsstats.biocache_readdirs++; 270 ndp = nfs_searchdircache(vp, uio->uio_offset, 271 (nmp->nm_flag & NFSMNT_XLATECOOKIE), 0); 272 if (!ndp) { 273 /* 274 * We've been handed a cookie that is not 275 * in the cache. If we're not translating 276 * 32 <-> 64, it may be a value that was 277 * flushed out of the cache because it grew 278 * too big. Let the server judge if it's 279 * valid or not. In the translation case, 280 * we have no way of validating this value, 281 * so punt. 282 */ 283 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) 284 return (EINVAL); 285 ndp = nfs_enterdircache(vp, uio->uio_offset, 286 uio->uio_offset, 0, 0); 287 } 288 289 if (uio->uio_offset != 0 && 290 ndp->dc_cookie == np->n_direofoffset) { 291 nfsstats.direofcache_hits++; 292 return (0); 293 } 294 295 bp = nfs_getcacheblk(vp, ndp->dc_blkno, NFS_DIRBLKSIZ, p); 296 if (!bp) 297 return (EINTR); 298 if ((bp->b_flags & B_DONE) == 0) { 299 bp->b_flags |= B_READ; 300 bp->b_dcookie = ndp->dc_blkcookie; 301 error = nfs_doio(bp, p); 302 if (error) { 303 /* 304 * Yuck! The directory has been modified on the 305 * server. Punt and let the userland code 306 * deal with it. 307 */ 308 brelse(bp); 309 if (error == NFSERR_BAD_COOKIE) { 310 nfs_invaldircache(vp, 0); 311 nfs_vinvalbuf(vp, 0, cred, p, 1); 312 error = EINVAL; 313 } 314 return (error); 315 } 316 } 317 318 /* 319 * Just return if we hit EOF right away with this 320 * block. Always check here, because direofoffset 321 * may have been set by an nfsiod since the last 322 * check. 323 */ 324 if (np->n_direofoffset != 0 && 325 ndp->dc_blkcookie == np->n_direofoffset) { 326 brelse(bp); 327 return (0); 328 } 329 330 /* 331 * Find the entry we were looking for in the block. 332 */ 333 334 en = ndp->dc_entry; 335 336 pdp = dp = (struct dirent *)bp->b_data; 337 edp = bp->b_data + bp->b_bcount; 338 enn = 0; 339 while (enn < en && (caddr_t)dp < edp) { 340 pdp = dp; 341 dp = (struct dirent *)((caddr_t)dp + dp->d_reclen); 342 enn++; 343 } 344 345 /* 346 * If the entry number was bigger than the number of 347 * entries in the block, or the cookie of the previous 348 * entry doesn't match, the directory cache is 349 * stale. Flush it and try again (i.e. go to 350 * the server). 351 */ 352 if ((caddr_t)dp >= edp || (caddr_t)dp + dp->d_reclen > edp || 353 (en > 0 && NFS_GETCOOKIE(pdp) != ndp->dc_cookie)) { 354 #ifdef DEBUG 355 printf("invalid cache: %p %p %p off %lx %lx\n", 356 pdp, dp, edp, 357 (unsigned long)uio->uio_offset, 358 (unsigned long)NFS_GETCOOKIE(pdp)); 359 #endif 360 brelse(bp); 361 nfs_invaldircache(vp, 0); 362 nfs_vinvalbuf(vp, 0, cred, p, 0); 363 goto diragain; 364 } 365 366 on = (caddr_t)dp - bp->b_data; 367 368 /* 369 * Cache all entries that may be exported to the 370 * user, as they may be thrown back at us. The 371 * NFSBIO_CACHECOOKIES flag indicates that all 372 * entries are being 'exported', so cache them all. 373 */ 374 375 if (en == 0 && pdp == dp) { 376 dp = (struct dirent *) 377 ((caddr_t)dp + dp->d_reclen); 378 enn++; 379 } 380 381 if (uio->uio_resid < (bp->b_bcount - on)) { 382 n = uio->uio_resid; 383 enough = 1; 384 } else 385 n = bp->b_bcount - on; 386 387 ep = bp->b_data + on + n; 388 389 /* 390 * Find last complete entry to copy, caching entries 391 * (if requested) as we go. 392 */ 393 394 while ((caddr_t)dp < ep && (caddr_t)dp + dp->d_reclen <= ep) { 395 if (cflag & NFSBIO_CACHECOOKIES) { 396 nndp = nfs_enterdircache(vp, NFS_GETCOOKIE(pdp), 397 ndp->dc_blkcookie, enn, bp->b_lblkno); 398 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) { 399 NFS_STASHCOOKIE32(pdp, 400 nndp->dc_cookie32); 401 } 402 } 403 pdp = dp; 404 dp = (struct dirent *)((caddr_t)dp + dp->d_reclen); 405 enn++; 406 } 407 408 /* 409 * If the last requested entry was not the last in the 410 * buffer (happens if NFS_DIRFRAGSIZ < NFS_DIRBLKSIZ), 411 * cache the cookie of the last requested one, and 412 * set of the offset to it. 413 */ 414 415 if ((on + n) < bp->b_bcount) { 416 curoff = NFS_GETCOOKIE(pdp); 417 nndp = nfs_enterdircache(vp, curoff, ndp->dc_blkcookie, 418 enn, bp->b_lblkno); 419 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) { 420 NFS_STASHCOOKIE32(pdp, nndp->dc_cookie32); 421 curoff = nndp->dc_cookie32; 422 } 423 } else 424 curoff = bp->b_dcookie; 425 426 /* 427 * Always cache the entry for the next block, 428 * so that readaheads can use it. 429 */ 430 nndp = nfs_enterdircache(vp, bp->b_dcookie, bp->b_dcookie, 0,0); 431 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) { 432 if (curoff == bp->b_dcookie) { 433 NFS_STASHCOOKIE32(pdp, nndp->dc_cookie32); 434 curoff = nndp->dc_cookie32; 435 } 436 } 437 438 n = ((caddr_t)pdp + pdp->d_reclen) - (bp->b_data + on); 439 440 /* 441 * If not eof and read aheads are enabled, start one. 442 * (You need the current block first, so that you have the 443 * directory offset cookie of the next block.) 444 */ 445 if (nfs_numasync > 0 && nmp->nm_readahead > 0 && 446 np->n_direofoffset == 0 && !(np->n_flag & NQNFSNONCACHE)) { 447 rabp = nfs_getcacheblk(vp, nndp->dc_blkno, 448 NFS_DIRBLKSIZ, p); 449 if (rabp) { 450 if ((rabp->b_flags & (B_DONE | B_DELWRI)) == 0) { 451 rabp->b_dcookie = nndp->dc_cookie; 452 rabp->b_flags |= (B_READ | B_ASYNC); 453 if (nfs_asyncio(rabp)) { 454 rabp->b_flags |= B_INVAL; 455 brelse(rabp); 456 } 457 } else 458 brelse(rabp); 459 } 460 } 461 got_buf = 1; 462 break; 463 default: 464 printf(" nfsbioread: type %x unexpected\n",vp->v_type); 465 break; 466 } 467 468 if (n > 0) { 469 if (!baddr) 470 baddr = bp->b_data; 471 error = uiomove(baddr + on, (int)n, uio); 472 } 473 switch (vp->v_type) { 474 case VREG: 475 break; 476 case VLNK: 477 n = 0; 478 break; 479 case VDIR: 480 if (np->n_flag & NQNFSNONCACHE) 481 bp->b_flags |= B_INVAL; 482 uio->uio_offset = curoff; 483 if (enough) 484 n = 0; 485 break; 486 default: 487 printf(" nfsbioread: type %x unexpected\n",vp->v_type); 488 } 489 if (got_buf) 490 brelse(bp); 491 } while (error == 0 && uio->uio_resid > 0 && n > 0); 492 return (error); 493 } 494 495 /* 496 * Vnode op for write using bio 497 */ 498 int 499 nfs_write(v) 500 void *v; 501 { 502 struct vop_write_args /* { 503 struct vnode *a_vp; 504 struct uio *a_uio; 505 int a_ioflag; 506 struct ucred *a_cred; 507 } */ *ap = v; 508 struct uio *uio = ap->a_uio; 509 struct proc *p = uio->uio_procp; 510 struct vnode *vp = ap->a_vp; 511 struct nfsnode *np = VTONFS(vp); 512 struct ucred *cred = ap->a_cred; 513 int ioflag = ap->a_ioflag; 514 struct vattr vattr; 515 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 516 int error = 0, iomode, must_commit; 517 int rv; 518 519 #ifdef DIAGNOSTIC 520 if (uio->uio_rw != UIO_WRITE) 521 panic("nfs_write mode"); 522 if (uio->uio_segflg == UIO_USERSPACE && uio->uio_procp != curproc) 523 panic("nfs_write proc"); 524 #endif 525 if (vp->v_type != VREG) 526 return (EIO); 527 if (np->n_flag & NWRITEERR) { 528 np->n_flag &= ~NWRITEERR; 529 return (np->n_error); 530 } 531 #ifndef NFS_V2_ONLY 532 if ((nmp->nm_flag & NFSMNT_NFSV3) && 533 !(nmp->nm_iflag & NFSMNT_GOTFSINFO)) 534 (void)nfs_fsinfo(nmp, vp, cred, p); 535 #endif 536 if (ioflag & (IO_APPEND | IO_SYNC)) { 537 if (np->n_flag & NMODIFIED) { 538 np->n_attrstamp = 0; 539 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 540 if (error) 541 return (error); 542 } 543 if (ioflag & IO_APPEND) { 544 np->n_attrstamp = 0; 545 error = VOP_GETATTR(vp, &vattr, cred, p); 546 if (error) 547 return (error); 548 uio->uio_offset = np->n_size; 549 } 550 } 551 if (uio->uio_offset < 0) 552 return (EINVAL); 553 if ((uio->uio_offset + uio->uio_resid) > nmp->nm_maxfilesize) 554 return (EFBIG); 555 if (uio->uio_resid == 0) 556 return (0); 557 /* 558 * Maybe this should be above the vnode op call, but so long as 559 * file servers have no limits, i don't think it matters 560 */ 561 if (p && uio->uio_offset + uio->uio_resid > 562 p->p_rlimit[RLIMIT_FSIZE].rlim_cur) { 563 psignal(p, SIGXFSZ); 564 return (EFBIG); 565 } 566 567 /* 568 * update the cached write creds for this node. 569 */ 570 571 if (np->n_wcred) { 572 crfree(np->n_wcred); 573 } 574 np->n_wcred = cred; 575 crhold(cred); 576 577 if ((np->n_flag & NQNFSNONCACHE) && uio->uio_iovcnt == 1) { 578 iomode = NFSV3WRITE_FILESYNC; 579 error = nfs_writerpc(vp, uio, &iomode, &must_commit); 580 if (must_commit) 581 nfs_clearcommit(vp->v_mount); 582 return (error); 583 } 584 585 do { 586 void *win; 587 voff_t oldoff = uio->uio_offset; 588 vsize_t bytelen = uio->uio_resid; 589 590 #ifndef NFS_V2_ONLY 591 /* 592 * Check for a valid write lease. 593 */ 594 if ((nmp->nm_flag & NFSMNT_NQNFS) && 595 NQNFS_CKINVALID(vp, np, ND_WRITE)) { 596 do { 597 error = nqnfs_getlease(vp, ND_WRITE, cred, p); 598 } while (error == NQNFS_EXPIRED); 599 if (error) 600 return (error); 601 if (np->n_lrev != np->n_brev || 602 (np->n_flag & NQNFSNONCACHE)) { 603 error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 604 if (error) 605 return (error); 606 np->n_brev = np->n_lrev; 607 } 608 } 609 #endif 610 nfsstats.biocache_writes++; 611 612 np->n_flag |= NMODIFIED; 613 if (np->n_size < uio->uio_offset + bytelen) { 614 np->n_size = uio->uio_offset + bytelen; 615 uvm_vnp_setsize(vp, np->n_size); 616 } 617 win = ubc_alloc(&vp->v_uvm.u_obj, uio->uio_offset, &bytelen, 618 UBC_WRITE); 619 error = uiomove(win, bytelen, uio); 620 if (error) { 621 memset((void *)trunc_page((vaddr_t)win), 0, 622 round_page((vaddr_t)win + bytelen) - 623 trunc_page((vaddr_t)win)); 624 } 625 ubc_release(win, 0); 626 rv = 1; 627 if ((np->n_flag & NQNFSNONCACHE) || (ioflag & IO_SYNC)) { 628 simple_lock(&vp->v_uvm.u_obj.vmobjlock); 629 rv = vp->v_uvm.u_obj.pgops->pgo_flush( 630 &vp->v_uvm.u_obj, 631 oldoff & ~(nmp->nm_wsize - 1), 632 uio->uio_offset & ~(nmp->nm_wsize - 1), 633 PGO_CLEANIT|PGO_SYNCIO); 634 simple_unlock(&vp->v_uvm.u_obj.vmobjlock); 635 } else if ((oldoff & ~(nmp->nm_wsize - 1)) != 636 (uio->uio_offset & ~(nmp->nm_wsize - 1))) { 637 simple_lock(&vp->v_uvm.u_obj.vmobjlock); 638 rv = vp->v_uvm.u_obj.pgops->pgo_flush( 639 &vp->v_uvm.u_obj, 640 oldoff & ~(nmp->nm_wsize - 1), 641 uio->uio_offset & ~(nmp->nm_wsize - 1), 642 PGO_CLEANIT|PGO_WEAK); 643 simple_unlock(&vp->v_uvm.u_obj.vmobjlock); 644 } 645 if (!rv) { 646 error = EIO; 647 break; 648 } 649 } while (uio->uio_resid > 0); 650 return error; 651 } 652 653 /* 654 * Get an nfs cache block. 655 * Allocate a new one if the block isn't currently in the cache 656 * and return the block marked busy. If the calling process is 657 * interrupted by a signal for an interruptible mount point, return 658 * NULL. 659 */ 660 struct buf * 661 nfs_getcacheblk(vp, bn, size, p) 662 struct vnode *vp; 663 daddr_t bn; 664 int size; 665 struct proc *p; 666 { 667 struct buf *bp; 668 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 669 670 if (nmp->nm_flag & NFSMNT_INT) { 671 bp = getblk(vp, bn, size, PCATCH, 0); 672 while (bp == NULL) { 673 if (nfs_sigintr(nmp, NULL, p)) 674 return (NULL); 675 bp = getblk(vp, bn, size, 0, 2 * hz); 676 } 677 } else 678 bp = getblk(vp, bn, size, 0, 0); 679 return (bp); 680 } 681 682 /* 683 * Flush and invalidate all dirty buffers. If another process is already 684 * doing the flush, just wait for completion. 685 */ 686 int 687 nfs_vinvalbuf(vp, flags, cred, p, intrflg) 688 struct vnode *vp; 689 int flags; 690 struct ucred *cred; 691 struct proc *p; 692 int intrflg; 693 { 694 struct nfsnode *np = VTONFS(vp); 695 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 696 int error = 0, slpflag, slptimeo; 697 698 if ((nmp->nm_flag & NFSMNT_INT) == 0) 699 intrflg = 0; 700 if (intrflg) { 701 slpflag = PCATCH; 702 slptimeo = 2 * hz; 703 } else { 704 slpflag = 0; 705 slptimeo = 0; 706 } 707 /* 708 * First wait for any other process doing a flush to complete. 709 */ 710 while (np->n_flag & NFLUSHINPROG) { 711 np->n_flag |= NFLUSHWANT; 712 error = tsleep((caddr_t)&np->n_flag, PRIBIO + 2, "nfsvinval", 713 slptimeo); 714 if (error && intrflg && nfs_sigintr(nmp, NULL, p)) 715 return (EINTR); 716 } 717 718 /* 719 * Now, flush as required. 720 */ 721 np->n_flag |= NFLUSHINPROG; 722 error = vinvalbuf(vp, flags, cred, p, slpflag, 0); 723 while (error) { 724 if (intrflg && nfs_sigintr(nmp, NULL, p)) { 725 np->n_flag &= ~NFLUSHINPROG; 726 if (np->n_flag & NFLUSHWANT) { 727 np->n_flag &= ~NFLUSHWANT; 728 wakeup((caddr_t)&np->n_flag); 729 } 730 return (EINTR); 731 } 732 error = vinvalbuf(vp, flags, cred, p, 0, slptimeo); 733 } 734 np->n_flag &= ~(NMODIFIED | NFLUSHINPROG); 735 if (np->n_flag & NFLUSHWANT) { 736 np->n_flag &= ~NFLUSHWANT; 737 wakeup((caddr_t)&np->n_flag); 738 } 739 return (0); 740 } 741 742 /* 743 * Initiate asynchronous I/O. Return an error if no nfsiods are available. 744 * This is mainly to avoid queueing async I/O requests when the nfsiods 745 * are all hung on a dead server. 746 */ 747 int 748 nfs_asyncio(bp) 749 struct buf *bp; 750 { 751 int i; 752 struct nfsmount *nmp; 753 int gotiod, slpflag = 0, slptimeo = 0, error; 754 755 if (nfs_numasync == 0) 756 return (EIO); 757 758 759 nmp = VFSTONFS(bp->b_vp->v_mount); 760 again: 761 if (nmp->nm_flag & NFSMNT_INT) 762 slpflag = PCATCH; 763 gotiod = FALSE; 764 765 /* 766 * Find a free iod to process this request. 767 */ 768 769 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 770 if (nfs_iodwant[i]) { 771 /* 772 * Found one, so wake it up and tell it which 773 * mount to process. 774 */ 775 nfs_iodwant[i] = NULL; 776 nfs_iodmount[i] = nmp; 777 nmp->nm_bufqiods++; 778 wakeup((caddr_t)&nfs_iodwant[i]); 779 gotiod = TRUE; 780 break; 781 } 782 /* 783 * If none are free, we may already have an iod working on this mount 784 * point. If so, it will process our request. 785 */ 786 if (!gotiod && nmp->nm_bufqiods > 0) 787 gotiod = TRUE; 788 789 /* 790 * If we have an iod which can process the request, then queue 791 * the buffer. 792 */ 793 if (gotiod) { 794 /* 795 * Ensure that the queue never grows too large. 796 */ 797 while (nmp->nm_bufqlen >= 2*nfs_numasync) { 798 nmp->nm_bufqwant = TRUE; 799 error = tsleep(&nmp->nm_bufq, slpflag | PRIBIO, 800 "nfsaio", slptimeo); 801 if (error) { 802 if (nfs_sigintr(nmp, NULL, bp->b_proc)) 803 return (EINTR); 804 if (slpflag == PCATCH) { 805 slpflag = 0; 806 slptimeo = 2 * hz; 807 } 808 } 809 /* 810 * We might have lost our iod while sleeping, 811 * so check and loop if nescessary. 812 */ 813 if (nmp->nm_bufqiods == 0) 814 goto again; 815 } 816 TAILQ_INSERT_TAIL(&nmp->nm_bufq, bp, b_freelist); 817 nmp->nm_bufqlen++; 818 return (0); 819 } 820 821 /* 822 * All the iods are busy on other mounts, so return EIO to 823 * force the caller to process the i/o synchronously. 824 */ 825 return (EIO); 826 } 827 828 /* 829 * Do an I/O operation to/from a cache block. This may be called 830 * synchronously or from an nfsiod. 831 */ 832 int 833 nfs_doio(bp, p) 834 struct buf *bp; 835 struct proc *p; 836 { 837 struct uio *uiop; 838 struct vnode *vp; 839 struct nfsnode *np; 840 struct nfsmount *nmp; 841 int error = 0, diff, len, iomode, must_commit = 0; 842 struct uio uio; 843 struct iovec io; 844 845 vp = bp->b_vp; 846 np = VTONFS(vp); 847 nmp = VFSTONFS(vp->v_mount); 848 uiop = &uio; 849 uiop->uio_iov = &io; 850 uiop->uio_iovcnt = 1; 851 uiop->uio_segflg = UIO_SYSSPACE; 852 uiop->uio_procp = p; 853 854 /* 855 * Historically, paging was done with physio, but no more... 856 */ 857 if (bp->b_flags & B_PHYS) { 858 /* 859 * ...though reading /dev/drum still gets us here. 860 */ 861 io.iov_len = uiop->uio_resid = bp->b_bcount; 862 /* mapping was done by vmapbuf() */ 863 io.iov_base = bp->b_data; 864 uiop->uio_offset = ((off_t)bp->b_blkno) << DEV_BSHIFT; 865 if (bp->b_flags & B_READ) { 866 uiop->uio_rw = UIO_READ; 867 nfsstats.read_physios++; 868 error = nfs_readrpc(vp, uiop); 869 } else { 870 iomode = NFSV3WRITE_DATASYNC; 871 uiop->uio_rw = UIO_WRITE; 872 nfsstats.write_physios++; 873 error = nfs_writerpc(vp, uiop, &iomode, &must_commit); 874 } 875 if (error) { 876 bp->b_flags |= B_ERROR; 877 bp->b_error = error; 878 } 879 } else if (bp->b_flags & B_READ) { 880 io.iov_len = uiop->uio_resid = bp->b_bcount; 881 io.iov_base = bp->b_data; 882 uiop->uio_rw = UIO_READ; 883 switch (vp->v_type) { 884 case VREG: 885 uiop->uio_offset = ((off_t)bp->b_blkno) << DEV_BSHIFT; 886 nfsstats.read_bios++; 887 error = nfs_readrpc(vp, uiop); 888 if (!error && uiop->uio_resid) { 889 890 /* 891 * If len > 0, there is a hole in the file and 892 * no writes after the hole have been pushed to 893 * the server yet. 894 * Just zero fill the rest of the valid area. 895 */ 896 897 diff = bp->b_bcount - uiop->uio_resid; 898 len = np->n_size - ((((off_t)bp->b_blkno) << DEV_BSHIFT) 899 + diff); 900 if (len > 0) { 901 len = min(len, uiop->uio_resid); 902 memset((char *)bp->b_data + diff, 0, len); 903 } 904 } 905 if (p && (vp->v_flag & VTEXT) && 906 (((nmp->nm_flag & NFSMNT_NQNFS) && 907 NQNFS_CKINVALID(vp, np, ND_READ) && 908 np->n_lrev != np->n_brev) || 909 (!(nmp->nm_flag & NFSMNT_NQNFS) && 910 np->n_mtime != np->n_vattr->va_mtime.tv_sec))) { 911 uprintf("Process killed due to " 912 "text file modification\n"); 913 psignal(p, SIGKILL); 914 p->p_holdcnt++; 915 } 916 break; 917 case VLNK: 918 uiop->uio_offset = (off_t)0; 919 nfsstats.readlink_bios++; 920 error = nfs_readlinkrpc(vp, uiop, curproc->p_ucred); 921 break; 922 case VDIR: 923 nfsstats.readdir_bios++; 924 uiop->uio_offset = bp->b_dcookie; 925 if (nmp->nm_flag & NFSMNT_RDIRPLUS) { 926 error = nfs_readdirplusrpc(vp, uiop, curproc->p_ucred); 927 if (error == NFSERR_NOTSUPP) 928 nmp->nm_flag &= ~NFSMNT_RDIRPLUS; 929 } 930 if ((nmp->nm_flag & NFSMNT_RDIRPLUS) == 0) 931 error = nfs_readdirrpc(vp, uiop, curproc->p_ucred); 932 if (!error) { 933 bp->b_dcookie = uiop->uio_offset; 934 } 935 break; 936 default: 937 printf("nfs_doio: type %x unexpected\n",vp->v_type); 938 break; 939 } 940 if (error) { 941 bp->b_flags |= B_ERROR; 942 bp->b_error = error; 943 } 944 } else { 945 /* 946 * If B_NEEDCOMMIT is set, a commit rpc may do the trick. If not 947 * an actual write will have to be scheduled. 948 */ 949 950 io.iov_base = bp->b_data; 951 io.iov_len = uiop->uio_resid = bp->b_bcount; 952 uiop->uio_offset = (((off_t)bp->b_blkno) << DEV_BSHIFT); 953 uiop->uio_rw = UIO_WRITE; 954 nfsstats.write_bios++; 955 iomode = NFSV3WRITE_UNSTABLE; 956 error = nfs_writerpc(vp, uiop, &iomode, &must_commit); 957 } 958 bp->b_resid = uiop->uio_resid; 959 if (must_commit) 960 nfs_clearcommit(vp->v_mount); 961 biodone(bp); 962 return (error); 963 } 964 965 /* 966 * Vnode op for VM getpages. 967 */ 968 int 969 nfs_getpages(v) 970 void *v; 971 { 972 struct vop_getpages_args /* { 973 struct vnode *a_vp; 974 voff_t a_offset; 975 vm_page_t *a_m; 976 int *a_count; 977 int a_centeridx; 978 vm_prot_t a_access_type; 979 int a_advice; 980 int a_flags; 981 } */ *ap = v; 982 983 off_t eof, offset, origoffset, startoffset, endoffset; 984 int s, i, error, npages, orignpages, npgs, ridx, pidx, pcount; 985 vaddr_t kva; 986 struct buf *bp, *mbp; 987 struct vnode *vp = ap->a_vp; 988 struct nfsnode *np = VTONFS(vp); 989 struct uvm_object *uobj = &vp->v_uvm.u_obj; 990 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 991 size_t bytes, iobytes, tailbytes, totalbytes, skipbytes; 992 int flags = ap->a_flags; 993 int bsize; 994 struct vm_page *pgs[16]; /* XXXUBC 16 */ 995 boolean_t v3 = NFS_ISV3(vp); 996 boolean_t async = (flags & PGO_SYNCIO) == 0; 997 boolean_t write = (ap->a_access_type & VM_PROT_WRITE) != 0; 998 999 UVMHIST_FUNC("nfs_getpages"); UVMHIST_CALLED(ubchist); 1000 UVMHIST_LOG(ubchist, "vp %p off 0x%x count %d", vp, (int)ap->a_offset, 1001 *ap->a_count,0); 1002 1003 #ifdef DIAGNOSTIC 1004 if (ap->a_centeridx < 0 || ap->a_centeridx >= *ap->a_count) { 1005 panic("nfs_getpages: centeridx %d out of range", 1006 ap->a_centeridx); 1007 } 1008 #endif 1009 1010 error = 0; 1011 origoffset = ap->a_offset; 1012 eof = vp->v_uvm.u_size; 1013 if (origoffset >= eof) { 1014 if ((flags & PGO_LOCKED) == 0) { 1015 simple_unlock(&uobj->vmobjlock); 1016 } 1017 UVMHIST_LOG(ubchist, "off 0x%x past EOF 0x%x", 1018 (int)origoffset, (int)eof,0,0); 1019 return EINVAL; 1020 } 1021 1022 if (flags & PGO_LOCKED) { 1023 uvn_findpages(uobj, origoffset, ap->a_count, ap->a_m, 1024 UFP_NOWAIT|UFP_NOALLOC); 1025 return 0; 1026 } 1027 1028 /* vnode is VOP_LOCKed, uobj is locked */ 1029 1030 bsize = nmp->nm_rsize; 1031 orignpages = min(*ap->a_count, 1032 round_page(eof - origoffset) >> PAGE_SHIFT); 1033 npages = orignpages; 1034 startoffset = origoffset & ~(bsize - 1); 1035 endoffset = round_page((origoffset + (npages << PAGE_SHIFT) 1036 + bsize - 1) & ~(bsize - 1)); 1037 endoffset = min(endoffset, round_page(eof)); 1038 ridx = (origoffset - startoffset) >> PAGE_SHIFT; 1039 1040 if (!async && !write) { 1041 int rapages = max(PAGE_SIZE, nmp->nm_rsize) >> PAGE_SHIFT; 1042 1043 (void) VOP_GETPAGES(vp, endoffset, NULL, &rapages, 0, 1044 VM_PROT_READ, 0, 0); 1045 simple_lock(&uobj->vmobjlock); 1046 } 1047 1048 UVMHIST_LOG(ubchist, "npages %d offset 0x%x", npages, 1049 (int)origoffset, 0,0); 1050 memset(pgs, 0, sizeof(pgs)); 1051 uvn_findpages(uobj, origoffset, &npages, &pgs[ridx], UFP_ALL); 1052 1053 if (flags & PGO_OVERWRITE) { 1054 UVMHIST_LOG(ubchist, "PGO_OVERWRITE",0,0,0,0); 1055 1056 /* XXXUBC for now, zero the page if we allocated it */ 1057 for (i = 0; i < npages; i++) { 1058 struct vm_page *pg = pgs[ridx + i]; 1059 1060 if (pg->flags & PG_FAKE) { 1061 uvm_pagezero(pg); 1062 pg->flags &= ~(PG_FAKE); 1063 } 1064 } 1065 goto out; 1066 } 1067 1068 /* 1069 * if the pages are already resident, just return them. 1070 */ 1071 1072 for (i = 0; i < npages; i++) { 1073 struct vm_page *pg = pgs[ridx + i]; 1074 1075 if ((pg->flags & PG_FAKE) != 0 || 1076 ((ap->a_access_type & VM_PROT_WRITE) && 1077 (pg->flags & PG_RDONLY))) { 1078 break; 1079 } 1080 } 1081 if (i == npages) { 1082 UVMHIST_LOG(ubchist, "returning cached pages", 0,0,0,0); 1083 goto out; 1084 } 1085 1086 /* 1087 * the page wasn't resident and we're not overwriting, 1088 * so we're going to have to do some i/o. 1089 * find any additional pages needed to cover the expanded range. 1090 */ 1091 1092 if (startoffset != origoffset || 1093 startoffset + (npages << PAGE_SHIFT) != endoffset) { 1094 UVMHIST_LOG(ubchist, "reset npages start 0x%x end 0x%x", 1095 (int)startoffset, (int)endoffset, 0,0); 1096 npages = (endoffset - startoffset) >> PAGE_SHIFT; 1097 KASSERT(npages != 0); 1098 npgs = npages; 1099 uvn_findpages(uobj, startoffset, &npgs, pgs, UFP_ALL); 1100 } 1101 simple_unlock(&uobj->vmobjlock); 1102 1103 /* 1104 * update the cached read creds for this node. 1105 */ 1106 1107 if (np->n_rcred) { 1108 crfree(np->n_rcred); 1109 } 1110 np->n_rcred = curproc->p_ucred; 1111 crhold(np->n_rcred); 1112 1113 /* 1114 * read the desired page(s). 1115 */ 1116 1117 totalbytes = npages << PAGE_SHIFT; 1118 bytes = min(totalbytes, vp->v_uvm.u_size - startoffset); 1119 tailbytes = totalbytes - bytes; 1120 skipbytes = 0; 1121 1122 kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WAITOK | 1123 UVMPAGER_MAPIN_READ); 1124 1125 s = splbio(); 1126 mbp = pool_get(&bufpool, PR_WAITOK); 1127 splx(s); 1128 mbp->b_bufsize = totalbytes; 1129 mbp->b_data = (void *)kva; 1130 mbp->b_resid = mbp->b_bcount = bytes; 1131 mbp->b_flags = B_BUSY|B_READ| (async ? B_CALL|B_ASYNC : 0); 1132 mbp->b_iodone = uvm_aio_biodone; 1133 mbp->b_vp = vp; 1134 LIST_INIT(&mbp->b_dep); 1135 1136 /* 1137 * if EOF is in the middle of the last page, zero the part past EOF. 1138 */ 1139 1140 if (tailbytes > 0) { 1141 memset((char *)kva + bytes, 0, tailbytes); 1142 } 1143 1144 /* 1145 * now loop over the pages, reading as needed. 1146 */ 1147 1148 bp = NULL; 1149 for (offset = startoffset; 1150 bytes > 0; 1151 offset += iobytes, bytes -= iobytes) { 1152 1153 /* 1154 * skip pages which don't need to be read. 1155 */ 1156 1157 pidx = (offset - startoffset) >> PAGE_SHIFT; 1158 UVMHIST_LOG(ubchist, "pidx %d offset 0x%x startoffset 0x%x", 1159 pidx, (int)offset, (int)startoffset,0); 1160 while ((pgs[pidx]->flags & PG_FAKE) == 0) { 1161 size_t b; 1162 1163 #ifdef DEBUG 1164 if (offset & (PAGE_SIZE - 1)) { 1165 panic("nfs_getpages: skipping from middle " 1166 "of page"); 1167 } 1168 #endif 1169 1170 b = min(PAGE_SIZE, bytes); 1171 offset += b; 1172 bytes -= b; 1173 skipbytes += b; 1174 pidx++; 1175 UVMHIST_LOG(ubchist, "skipping, new offset 0x%x", 1176 (int)offset, 0,0,0); 1177 if (bytes == 0) { 1178 goto loopdone; 1179 } 1180 } 1181 1182 /* 1183 * see how many pages can be read with this i/o. 1184 * reduce the i/o size if necessary. 1185 */ 1186 1187 iobytes = bytes; 1188 if (offset + iobytes > round_page(offset)) { 1189 pcount = 1; 1190 while (pidx + pcount < npages && 1191 pgs[pidx + pcount]->flags & PG_FAKE) { 1192 pcount++; 1193 } 1194 iobytes = min(iobytes, (pcount << PAGE_SHIFT) - 1195 (offset - trunc_page(offset))); 1196 } 1197 iobytes = min(iobytes, nmp->nm_rsize); 1198 1199 /* 1200 * allocate a sub-buf for this piece of the i/o 1201 * (or just use mbp if there's only 1 piece), 1202 * and start it going. 1203 */ 1204 1205 if (offset == startoffset && iobytes == bytes) { 1206 bp = mbp; 1207 } else { 1208 s = splbio(); 1209 bp = pool_get(&bufpool, PR_WAITOK); 1210 splx(s); 1211 bp->b_data = (char *)kva + offset - startoffset; 1212 bp->b_resid = bp->b_bcount = iobytes; 1213 bp->b_flags = B_BUSY|B_READ|B_CALL|B_ASYNC; 1214 bp->b_iodone = uvm_aio_biodone1; 1215 bp->b_vp = vp; 1216 LIST_INIT(&bp->b_dep); 1217 } 1218 bp->b_private = mbp; 1219 bp->b_lblkno = bp->b_blkno = offset >> DEV_BSHIFT; 1220 1221 UVMHIST_LOG(ubchist, "bp %p offset 0x%x bcount 0x%x blkno 0x%x", 1222 bp, offset, iobytes, bp->b_blkno); 1223 1224 VOP_STRATEGY(bp); 1225 } 1226 1227 loopdone: 1228 if (skipbytes) { 1229 s = splbio(); 1230 mbp->b_resid -= skipbytes; 1231 if (mbp->b_resid == 0) { 1232 biodone(mbp); 1233 } 1234 splx(s); 1235 } 1236 if (async) { 1237 UVMHIST_LOG(ubchist, "returning PEND",0,0,0,0); 1238 return EINPROGRESS; 1239 } 1240 if (bp != NULL) { 1241 error = biowait(mbp); 1242 } 1243 s = splbio(); 1244 pool_put(&bufpool, mbp); 1245 splx(s); 1246 uvm_pagermapout(kva, npages); 1247 1248 if (write && v3) { 1249 lockmgr(&np->n_commitlock, LK_EXCLUSIVE, NULL); 1250 nfs_del_committed_range(vp, origoffset, npages); 1251 nfs_del_tobecommitted_range(vp, origoffset, npages); 1252 for (i = 0; i < npages; i++) { 1253 if (pgs[i] == NULL) { 1254 continue; 1255 } 1256 pgs[i]->flags &= ~(PG_NEEDCOMMIT|PG_RDONLY); 1257 } 1258 lockmgr(&np->n_commitlock, LK_RELEASE, NULL); 1259 } 1260 1261 simple_lock(&uobj->vmobjlock); 1262 1263 out: 1264 uvm_lock_pageq(); 1265 if (error) { 1266 for (i = 0; i < npages; i++) { 1267 if (pgs[i] == NULL) { 1268 continue; 1269 } 1270 UVMHIST_LOG(ubchist, "examining pg %p flags 0x%x", 1271 pgs[i], pgs[i]->flags, 0,0); 1272 if ((pgs[i]->flags & PG_FAKE) == 0) { 1273 continue; 1274 } 1275 if (pgs[i]->flags & PG_WANTED) { 1276 wakeup(pgs[i]); 1277 } 1278 uvm_pagefree(pgs[i]); 1279 } 1280 goto done; 1281 } 1282 1283 UVMHIST_LOG(ubchist, "ridx %d count %d", ridx, npages, 0,0); 1284 for (i = 0; i < npages; i++) { 1285 if (pgs[i] == NULL) { 1286 continue; 1287 } 1288 UVMHIST_LOG(ubchist, "examining pg %p flags 0x%x", 1289 pgs[i], pgs[i]->flags, 0,0); 1290 if (pgs[i]->flags & PG_FAKE) { 1291 UVMHIST_LOG(ubchist, "unfaking pg %p offset 0x%x", 1292 pgs[i], (int)pgs[i]->offset,0,0); 1293 pgs[i]->flags &= ~(PG_FAKE); 1294 pmap_clear_modify(pgs[i]); 1295 pmap_clear_reference(pgs[i]); 1296 } 1297 if (i < ridx || i >= ridx + orignpages || async) { 1298 UVMHIST_LOG(ubchist, "unbusy pg %p offset 0x%x", 1299 pgs[i], (int)pgs[i]->offset,0,0); 1300 KASSERT((pgs[i]->flags & PG_RELEASED) == 0); 1301 if (pgs[i]->flags & PG_WANTED) { 1302 wakeup(pgs[i]); 1303 } 1304 if (pgs[i]->wire_count == 0) { 1305 uvm_pageactivate(pgs[i]); 1306 } 1307 pgs[i]->flags &= ~(PG_WANTED|PG_BUSY); 1308 UVM_PAGE_OWN(pgs[i], NULL); 1309 } 1310 } 1311 1312 done: 1313 uvm_unlock_pageq(); 1314 simple_unlock(&uobj->vmobjlock); 1315 if (ap->a_m != NULL) { 1316 memcpy(ap->a_m, &pgs[ridx], 1317 *ap->a_count * sizeof(struct vm_page *)); 1318 } 1319 1320 UVMHIST_LOG(ubchist, "done -> %d", error, 0,0,0); 1321 return error; 1322 } 1323 1324 /* 1325 * Vnode op for VM putpages. 1326 */ 1327 int 1328 nfs_putpages(v) 1329 void *v; 1330 { 1331 struct vop_putpages_args /* { 1332 struct vnode *a_vp; 1333 struct vm_page **a_m; 1334 int a_count; 1335 int a_flags; 1336 int *a_rtvals; 1337 } */ *ap = v; 1338 1339 struct vnode *vp = ap->a_vp; 1340 struct nfsnode *np = VTONFS(vp); 1341 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1342 struct buf *bp, *mbp; 1343 struct vm_page **pgs = ap->a_m; 1344 int flags = ap->a_flags; 1345 int npages = ap->a_count; 1346 int s, error = 0, i; 1347 size_t bytes, iobytes, skipbytes; 1348 vaddr_t kva; 1349 off_t offset, origoffset, commitoff; 1350 uint32_t commitbytes; 1351 boolean_t v3 = NFS_ISV3(vp); 1352 boolean_t async = (flags & PGO_SYNCIO) == 0; 1353 boolean_t weak = (flags & PGO_WEAK) && v3; 1354 UVMHIST_FUNC("nfs_putpages"); UVMHIST_CALLED(ubchist); 1355 1356 UVMHIST_LOG(ubchist, "vp %p pgp %p count %d", 1357 vp, ap->a_m, ap->a_count,0); 1358 1359 simple_unlock(&vp->v_uvm.u_obj.vmobjlock); 1360 1361 origoffset = pgs[0]->offset; 1362 bytes = min(ap->a_count << PAGE_SHIFT, vp->v_uvm.u_size - origoffset); 1363 skipbytes = 0; 1364 1365 /* 1366 * if the range has been committed already, mark the pages thus. 1367 * if the range just needs to be committed, we're done 1368 * if it's a weak putpage, otherwise commit the range. 1369 */ 1370 1371 if (v3) { 1372 lockmgr(&np->n_commitlock, LK_EXCLUSIVE, NULL); 1373 if (nfs_in_committed_range(vp, origoffset, bytes)) { 1374 goto committed; 1375 } 1376 if (nfs_in_tobecommitted_range(vp, origoffset, bytes)) { 1377 if (weak) { 1378 lockmgr(&np->n_commitlock, LK_RELEASE, NULL); 1379 return 0; 1380 } else { 1381 commitoff = np->n_pushlo; 1382 commitbytes = (uint32_t)(np->n_pushhi - 1383 np->n_pushlo); 1384 goto commit; 1385 } 1386 } 1387 lockmgr(&np->n_commitlock, LK_RELEASE, NULL); 1388 } 1389 1390 /* 1391 * otherwise write or commit all the pages. 1392 */ 1393 1394 kva = uvm_pagermapin(pgs, ap->a_count, UVMPAGER_MAPIN_WAITOK| 1395 UVMPAGER_MAPIN_WRITE); 1396 1397 s = splbio(); 1398 vp->v_numoutput += 2; 1399 mbp = pool_get(&bufpool, PR_WAITOK); 1400 UVMHIST_LOG(ubchist, "vp %p mbp %p num now %d bytes 0x%x", 1401 vp, mbp, vp->v_numoutput, bytes); 1402 splx(s); 1403 mbp->b_bufsize = npages << PAGE_SHIFT; 1404 mbp->b_data = (void *)kva; 1405 mbp->b_resid = mbp->b_bcount = bytes; 1406 mbp->b_flags = B_BUSY|B_WRITE|B_AGE | 1407 (async ? B_CALL|B_ASYNC : 0) | 1408 (curproc == uvm.pagedaemon_proc ? B_PDAEMON : 0); 1409 mbp->b_iodone = uvm_aio_aiodone; 1410 mbp->b_vp = vp; 1411 LIST_INIT(&mbp->b_dep); 1412 1413 for (offset = origoffset; 1414 bytes > 0; 1415 offset += iobytes, bytes -= iobytes) { 1416 iobytes = min(nmp->nm_wsize, bytes); 1417 1418 /* 1419 * skip writing any pages which only need a commit. 1420 */ 1421 1422 if ((pgs[(offset - origoffset) >> PAGE_SHIFT]->flags & 1423 PG_NEEDCOMMIT) != 0) { 1424 iobytes = PAGE_SIZE; 1425 skipbytes += min(iobytes, vp->v_uvm.u_size - offset); 1426 continue; 1427 } 1428 1429 /* if it's really one i/o, don't make a second buf */ 1430 if (offset == origoffset && iobytes == bytes) { 1431 bp = mbp; 1432 } else { 1433 s = splbio(); 1434 vp->v_numoutput++; 1435 bp = pool_get(&bufpool, PR_WAITOK); 1436 UVMHIST_LOG(ubchist, "vp %p bp %p num now %d", 1437 vp, bp, vp->v_numoutput, 0); 1438 splx(s); 1439 bp->b_data = (char *)kva + (offset - origoffset); 1440 bp->b_resid = bp->b_bcount = iobytes; 1441 bp->b_flags = B_BUSY|B_WRITE|B_CALL|B_ASYNC; 1442 bp->b_iodone = uvm_aio_biodone1; 1443 bp->b_vp = vp; 1444 LIST_INIT(&bp->b_dep); 1445 } 1446 bp->b_private = mbp; 1447 bp->b_lblkno = bp->b_blkno = (daddr_t)(offset >> DEV_BSHIFT); 1448 UVMHIST_LOG(ubchist, "bp %p numout %d", 1449 bp, vp->v_numoutput,0,0); 1450 VOP_STRATEGY(bp); 1451 } 1452 if (skipbytes) { 1453 UVMHIST_LOG(ubchist, "skipbytes %d", bytes, 0,0,0); 1454 s = splbio(); 1455 mbp->b_resid -= skipbytes; 1456 if (mbp->b_resid == 0) { 1457 biodone(mbp); 1458 } 1459 splx(s); 1460 } 1461 if (async) { 1462 return EINPROGRESS; 1463 } 1464 error = biowait(mbp); 1465 1466 s = splbio(); 1467 vwakeup(mbp); 1468 pool_put(&bufpool, mbp); 1469 splx(s); 1470 1471 uvm_pagermapout(kva, ap->a_count); 1472 if (error || !v3) { 1473 UVMHIST_LOG(ubchist, "returning error %d", error, 0,0,0); 1474 return error; 1475 } 1476 1477 /* 1478 * for a weak put, mark the range as "to be committed" 1479 * and mark the pages read-only so that we will be notified 1480 * to remove the pages from the "to be committed" range 1481 * if they are made dirty again. 1482 * for a strong put, commit the pages and remove them from the 1483 * "to be committed" range. also, mark them as writable 1484 * and not cleanable with just a commit. 1485 */ 1486 1487 lockmgr(&np->n_commitlock, LK_EXCLUSIVE, NULL); 1488 if (weak) { 1489 nfs_add_tobecommitted_range(vp, origoffset, 1490 npages << PAGE_SHIFT); 1491 for (i = 0; i < npages; i++) { 1492 pgs[i]->flags |= PG_NEEDCOMMIT|PG_RDONLY; 1493 } 1494 } else { 1495 commitoff = origoffset; 1496 commitbytes = npages << PAGE_SHIFT; 1497 commit: 1498 error = nfs_commit(vp, commitoff, commitbytes, curproc); 1499 nfs_del_tobecommitted_range(vp, commitoff, commitbytes); 1500 committed: 1501 for (i = 0; i < npages; i++) { 1502 pgs[i]->flags &= ~(PG_NEEDCOMMIT|PG_RDONLY); 1503 } 1504 } 1505 lockmgr(&np->n_commitlock, LK_RELEASE, NULL); 1506 return error; 1507 } 1508
Indexes created Tue Oct 14 15:09:51 GMT 2025