ulfs_readwrite.c revision 1.7.8.2
1 /* $NetBSD: ulfs_readwrite.c,v 1.7.8.2 2015/06/06 14:40:30 skrll Exp $ */ 2 /* from NetBSD: ufs_readwrite.c,v 1.105 2013/01/22 09:39:18 dholland Exp */ 3 4 /*- 5 * Copyright (c) 1993 6 * The Regents of the University of California. All rights reserved. 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. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * @(#)ufs_readwrite.c 8.11 (Berkeley) 5/8/95 33 */ 34 35 #include <sys/cdefs.h> 36 __KERNEL_RCSID(1, "$NetBSD: ulfs_readwrite.c,v 1.7.8.2 2015/06/06 14:40:30 skrll Exp $"); 37 38 #ifdef LFS_READWRITE 39 #define FS struct lfs 40 #define I_FS i_lfs 41 #define READ lfs_read 42 #define READ_S "lfs_read" 43 #define WRITE lfs_write 44 #define WRITE_S "lfs_write" 45 #define BUFRD lfs_bufrd 46 #define BUFWR lfs_bufwr 47 #define fs_bsize lfs_bsize 48 #define fs_bmask lfs_bmask 49 #else 50 #define FS struct fs 51 #define I_FS i_fs 52 #define READ ffs_read 53 #define READ_S "ffs_read" 54 #define WRITE ffs_write 55 #define WRITE_S "ffs_write" 56 #define BUFRD ffs_bufrd 57 #define BUFWR ffs_bufwr 58 #endif 59 60 static int ulfs_post_read_update(struct vnode *, int, int); 61 static int ulfs_post_write_update(struct vnode *, struct uio *, int, 62 kauth_cred_t, off_t, int, int, int); 63 64 /* 65 * Vnode op for reading. 66 */ 67 /* ARGSUSED */ 68 int 69 READ(void *v) 70 { 71 struct vop_read_args /* { 72 struct vnode *a_vp; 73 struct uio *a_uio; 74 int a_ioflag; 75 kauth_cred_t a_cred; 76 } */ *ap = v; 77 struct vnode *vp; 78 struct inode *ip; 79 struct uio *uio; 80 FS *fs; 81 vsize_t bytelen; 82 int error, ioflag, advice; 83 84 vp = ap->a_vp; 85 ip = VTOI(vp); 86 fs = ip->I_FS; 87 uio = ap->a_uio; 88 ioflag = ap->a_ioflag; 89 error = 0; 90 91 KASSERT(uio->uio_rw == UIO_READ); 92 KASSERT(vp->v_type == VREG || vp->v_type == VDIR); 93 94 /* XXX Eliminate me by refusing directory reads from userland. */ 95 if (vp->v_type == VDIR) 96 return BUFRD(vp, uio, ioflag, ap->a_cred); 97 #ifdef LFS_READWRITE 98 /* XXX Eliminate me by using ufs_bufio in lfs. */ 99 if (vp->v_type == VREG && ip->i_number == LFS_IFILE_INUM) 100 return BUFRD(vp, uio, ioflag, ap->a_cred); 101 #endif 102 if ((u_int64_t)uio->uio_offset > fs->um_maxfilesize) 103 return (EFBIG); 104 if (uio->uio_resid == 0) 105 return (0); 106 107 #ifndef LFS_READWRITE 108 if ((ip->i_flags & (SF_SNAPSHOT | SF_SNAPINVAL)) == SF_SNAPSHOT) 109 return ffs_snapshot_read(vp, uio, ioflag); 110 #endif /* !LFS_READWRITE */ 111 112 fstrans_start(vp->v_mount, FSTRANS_SHARED); 113 114 if (uio->uio_offset >= ip->i_size) 115 goto out; 116 117 KASSERT(vp->v_type == VREG); 118 advice = IO_ADV_DECODE(ap->a_ioflag); 119 while (uio->uio_resid > 0) { 120 if (ioflag & IO_DIRECT) { 121 genfs_directio(vp, uio, ioflag); 122 } 123 bytelen = MIN(ip->i_size - uio->uio_offset, uio->uio_resid); 124 if (bytelen == 0) 125 break; 126 error = ubc_uiomove(&vp->v_uobj, uio, bytelen, advice, 127 UBC_READ | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp)); 128 if (error) 129 break; 130 } 131 132 out: 133 error = ulfs_post_read_update(vp, ap->a_ioflag, error); 134 fstrans_done(vp->v_mount); 135 return (error); 136 } 137 138 /* 139 * UFS op for reading via the buffer cache 140 */ 141 int 142 BUFRD(struct vnode *vp, struct uio *uio, int ioflag, kauth_cred_t cred) 143 { 144 struct inode *ip; 145 FS *fs; 146 struct buf *bp; 147 daddr_t lbn, nextlbn; 148 off_t bytesinfile; 149 long size, xfersize, blkoffset; 150 int error; 151 152 KASSERT(VOP_ISLOCKED(vp)); 153 KASSERT(vp->v_type == VDIR || vp->v_type == VLNK || 154 vp->v_type == VREG); 155 KASSERT(uio->uio_rw == UIO_READ); 156 157 ip = VTOI(vp); 158 fs = ip->I_FS; 159 error = 0; 160 161 KASSERT(vp->v_type != VLNK || ip->i_size < fs->um_maxsymlinklen); 162 KASSERT(vp->v_type != VLNK || fs->um_maxsymlinklen != 0 || 163 DIP(ip, blocks) == 0); 164 KASSERT(vp->v_type != VREG || vp == fs->lfs_ivnode); 165 KASSERT(vp->v_type != VREG || ip->i_number == LFS_IFILE_INUM); 166 167 if (uio->uio_offset > fs->um_maxfilesize) 168 return EFBIG; 169 if (uio->uio_resid == 0) 170 return 0; 171 172 #ifndef LFS_READWRITE 173 KASSERT(!ISSET(ip->i_flags, (SF_SNAPSHOT | SF_SNAPINVAL))); 174 #endif 175 176 fstrans_start(vp->v_mount, FSTRANS_SHARED); 177 178 if (uio->uio_offset >= ip->i_size) 179 goto out; 180 181 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) { 182 bytesinfile = ip->i_size - uio->uio_offset; 183 if (bytesinfile <= 0) 184 break; 185 lbn = lfs_lblkno(fs, uio->uio_offset); 186 nextlbn = lbn + 1; 187 size = lfs_blksize(fs, ip, lbn); 188 blkoffset = lfs_blkoff(fs, uio->uio_offset); 189 xfersize = MIN(MIN(fs->fs_bsize - blkoffset, uio->uio_resid), 190 bytesinfile); 191 192 if (lfs_lblktosize(fs, nextlbn) >= ip->i_size) 193 error = bread(vp, lbn, size, 0, &bp); 194 else { 195 int nextsize = lfs_blksize(fs, ip, nextlbn); 196 error = breadn(vp, lbn, 197 size, &nextlbn, &nextsize, 1, 0, &bp); 198 } 199 if (error) 200 break; 201 202 /* 203 * We should only get non-zero b_resid when an I/O error 204 * has occurred, which should cause us to break above. 205 * However, if the short read did not cause an error, 206 * then we want to ensure that we do not uiomove bad 207 * or uninitialized data. 208 */ 209 size -= bp->b_resid; 210 if (size < xfersize) { 211 if (size == 0) 212 break; 213 xfersize = size; 214 } 215 error = uiomove((char *)bp->b_data + blkoffset, xfersize, uio); 216 if (error) 217 break; 218 brelse(bp, 0); 219 } 220 if (bp != NULL) 221 brelse(bp, 0); 222 223 out: 224 error = ulfs_post_read_update(vp, ioflag, error); 225 fstrans_done(vp->v_mount); 226 return (error); 227 } 228 229 static int 230 ulfs_post_read_update(struct vnode *vp, int ioflag, int oerror) 231 { 232 struct inode *ip = VTOI(vp); 233 int error = oerror; 234 235 if (!(vp->v_mount->mnt_flag & MNT_NOATIME)) { 236 ip->i_flag |= IN_ACCESS; 237 if ((ioflag & IO_SYNC) == IO_SYNC) { 238 error = lfs_update(vp, NULL, NULL, UPDATE_WAIT); 239 } 240 } 241 242 /* Read error overrides any inode update error. */ 243 if (oerror) 244 error = oerror; 245 return error; 246 } 247 248 /* 249 * Vnode op for writing. 250 */ 251 int 252 WRITE(void *v) 253 { 254 struct vop_write_args /* { 255 struct vnode *a_vp; 256 struct uio *a_uio; 257 int a_ioflag; 258 kauth_cred_t a_cred; 259 } */ *ap = v; 260 struct vnode *vp; 261 struct uio *uio; 262 struct inode *ip; 263 FS *fs; 264 kauth_cred_t cred; 265 off_t osize, origoff, oldoff, preallocoff, endallocoff, nsize; 266 int blkoffset, error, flags, ioflag, resid; 267 int aflag; 268 int extended=0; 269 vsize_t bytelen; 270 bool async; 271 272 cred = ap->a_cred; 273 ioflag = ap->a_ioflag; 274 uio = ap->a_uio; 275 vp = ap->a_vp; 276 ip = VTOI(vp); 277 278 KASSERT(vp->v_size == ip->i_size); 279 KASSERT(uio->uio_rw == UIO_WRITE); 280 KASSERT(vp->v_type == VREG); 281 282 if (ioflag & IO_APPEND) 283 uio->uio_offset = ip->i_size; 284 if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size) 285 return (EPERM); 286 287 fs = ip->I_FS; 288 if (uio->uio_offset < 0 || 289 (u_int64_t)uio->uio_offset + uio->uio_resid > fs->um_maxfilesize) 290 return (EFBIG); 291 #ifdef LFS_READWRITE 292 /* Disallow writes to the Ifile, even if noschg flag is removed */ 293 /* XXX can this go away when the Ifile is no longer in the namespace? */ 294 if (vp == fs->lfs_ivnode) 295 return (EPERM); 296 #endif 297 if (uio->uio_resid == 0) 298 return (0); 299 300 fstrans_start(vp->v_mount, FSTRANS_SHARED); 301 302 flags = ioflag & IO_SYNC ? B_SYNC : 0; 303 async = vp->v_mount->mnt_flag & MNT_ASYNC; 304 origoff = uio->uio_offset; 305 resid = uio->uio_resid; 306 osize = ip->i_size; 307 error = 0; 308 309 KASSERT(vp->v_type == VREG); 310 311 #ifdef LFS_READWRITE 312 async = true; 313 lfs_availwait(fs, lfs_btofsb(fs, uio->uio_resid)); 314 lfs_check(vp, LFS_UNUSED_LBN, 0); 315 #endif /* !LFS_READWRITE */ 316 317 preallocoff = round_page(lfs_blkroundup(fs, MAX(osize, uio->uio_offset))); 318 aflag = ioflag & IO_SYNC ? B_SYNC : 0; 319 nsize = MAX(osize, uio->uio_offset + uio->uio_resid); 320 endallocoff = nsize - lfs_blkoff(fs, nsize); 321 322 /* 323 * if we're increasing the file size, deal with expanding 324 * the fragment if there is one. 325 */ 326 327 if (nsize > osize && lfs_lblkno(fs, osize) < ULFS_NDADDR && 328 lfs_lblkno(fs, osize) != lfs_lblkno(fs, nsize) && 329 lfs_blkroundup(fs, osize) != osize) { 330 off_t eob; 331 332 eob = lfs_blkroundup(fs, osize); 333 uvm_vnp_setwritesize(vp, eob); 334 error = ulfs_balloc_range(vp, osize, eob - osize, cred, aflag); 335 if (error) 336 goto out; 337 if (flags & B_SYNC) { 338 mutex_enter(vp->v_interlock); 339 VOP_PUTPAGES(vp, trunc_page(osize & fs->fs_bmask), 340 round_page(eob), 341 PGO_CLEANIT | PGO_SYNCIO); 342 } 343 } 344 345 while (uio->uio_resid > 0) { 346 int ubc_flags = UBC_WRITE; 347 bool overwrite; /* if we're overwrite a whole block */ 348 off_t newoff; 349 350 if (ioflag & IO_DIRECT) { 351 genfs_directio(vp, uio, ioflag); 352 } 353 354 oldoff = uio->uio_offset; 355 blkoffset = lfs_blkoff(fs, uio->uio_offset); 356 bytelen = MIN(fs->fs_bsize - blkoffset, uio->uio_resid); 357 if (bytelen == 0) { 358 break; 359 } 360 361 /* 362 * if we're filling in a hole, allocate the blocks now and 363 * initialize the pages first. if we're extending the file, 364 * we can safely allocate blocks without initializing pages 365 * since the new blocks will be inaccessible until the write 366 * is complete. 367 */ 368 overwrite = uio->uio_offset >= preallocoff && 369 uio->uio_offset < endallocoff; 370 if (!overwrite && (vp->v_vflag & VV_MAPPED) == 0 && 371 lfs_blkoff(fs, uio->uio_offset) == 0 && 372 (uio->uio_offset & PAGE_MASK) == 0) { 373 vsize_t len; 374 375 len = trunc_page(bytelen); 376 len -= lfs_blkoff(fs, len); 377 if (len > 0) { 378 overwrite = true; 379 bytelen = len; 380 } 381 } 382 383 newoff = oldoff + bytelen; 384 if (vp->v_size < newoff) { 385 uvm_vnp_setwritesize(vp, newoff); 386 } 387 388 if (!overwrite) { 389 error = ulfs_balloc_range(vp, uio->uio_offset, bytelen, 390 cred, aflag); 391 if (error) 392 break; 393 } else { 394 genfs_node_wrlock(vp); 395 error = GOP_ALLOC(vp, uio->uio_offset, bytelen, 396 aflag, cred); 397 genfs_node_unlock(vp); 398 if (error) 399 break; 400 ubc_flags |= UBC_FAULTBUSY; 401 } 402 403 /* 404 * copy the data. 405 */ 406 407 error = ubc_uiomove(&vp->v_uobj, uio, bytelen, 408 IO_ADV_DECODE(ioflag), ubc_flags | UBC_UNMAP_FLAG(vp)); 409 410 /* 411 * update UVM's notion of the size now that we've 412 * copied the data into the vnode's pages. 413 * 414 * we should update the size even when uiomove failed. 415 */ 416 417 if (vp->v_size < newoff) { 418 uvm_vnp_setsize(vp, newoff); 419 extended = 1; 420 } 421 422 if (error) 423 break; 424 425 /* 426 * flush what we just wrote if necessary. 427 * XXXUBC simplistic async flushing. 428 */ 429 430 #ifndef LFS_READWRITE 431 if (!async && oldoff >> 16 != uio->uio_offset >> 16) { 432 mutex_enter(vp->v_interlock); 433 error = VOP_PUTPAGES(vp, (oldoff >> 16) << 16, 434 (uio->uio_offset >> 16) << 16, 435 PGO_CLEANIT | PGO_LAZY); 436 if (error) 437 break; 438 } 439 #else 440 __USE(async); 441 #endif 442 } 443 if (error == 0 && ioflag & IO_SYNC) { 444 mutex_enter(vp->v_interlock); 445 error = VOP_PUTPAGES(vp, trunc_page(origoff & fs->fs_bmask), 446 round_page(lfs_blkroundup(fs, uio->uio_offset)), 447 PGO_CLEANIT | PGO_SYNCIO); 448 } 449 450 out: 451 error = ulfs_post_write_update(vp, uio, ioflag, cred, osize, resid, 452 extended, error); 453 fstrans_done(vp->v_mount); 454 455 return (error); 456 } 457 458 /* 459 * UFS op for writing via the buffer cache 460 */ 461 int 462 BUFWR(struct vnode *vp, struct uio *uio, int ioflag, kauth_cred_t cred) 463 { 464 struct inode *ip; 465 FS *fs; 466 int flags; 467 struct buf *bp; 468 off_t osize; 469 int resid, xfersize, size, blkoffset; 470 daddr_t lbn; 471 int extended=0; 472 int error; 473 #ifdef LFS_READWRITE 474 bool need_unreserve = false; 475 #endif 476 477 KASSERT(ISSET(ioflag, IO_NODELOCKED)); 478 KASSERT(VOP_ISLOCKED(vp) == LK_EXCLUSIVE); 479 KASSERT(vp->v_type == VDIR || vp->v_type == VLNK); 480 KASSERT(vp->v_type != VDIR || ISSET(ioflag, IO_SYNC)); 481 KASSERT(uio->uio_rw == UIO_WRITE); 482 483 ip = VTOI(vp); 484 fs = ip->I_FS; 485 486 KASSERT(vp->v_size == ip->i_size); 487 488 if (uio->uio_offset < 0 || 489 uio->uio_resid > fs->um_maxfilesize || 490 uio->uio_offset > (fs->um_maxfilesize - uio->uio_resid)) 491 return EFBIG; 492 #ifdef LFS_READWRITE 493 KASSERT(vp != fs->lfs_ivnode); 494 #endif 495 if (uio->uio_resid == 0) 496 return 0; 497 498 fstrans_start(vp->v_mount, FSTRANS_SHARED); 499 500 flags = ioflag & IO_SYNC ? B_SYNC : 0; 501 resid = uio->uio_resid; 502 osize = ip->i_size; 503 error = 0; 504 505 KASSERT(vp->v_type != VREG); 506 507 #ifdef LFS_READWRITE 508 lfs_availwait(fs, lfs_btofsb(fs, uio->uio_resid)); 509 lfs_check(vp, LFS_UNUSED_LBN, 0); 510 #endif /* !LFS_READWRITE */ 511 512 /* XXX Should never have pages cached here. */ 513 KASSERT(vp->v_uobj.uo_npages == 0); 514 while (uio->uio_resid > 0) { 515 lbn = lfs_lblkno(fs, uio->uio_offset); 516 blkoffset = lfs_blkoff(fs, uio->uio_offset); 517 xfersize = MIN(fs->fs_bsize - blkoffset, uio->uio_resid); 518 if (fs->fs_bsize > xfersize) 519 flags |= B_CLRBUF; 520 else 521 flags &= ~B_CLRBUF; 522 523 #ifdef LFS_READWRITE 524 error = lfs_reserve(fs, vp, NULL, 525 lfs_btofsb(fs, (ULFS_NIADDR + 1) << fs->lfs_bshift)); 526 if (error) 527 break; 528 need_unreserve = true; 529 #endif 530 error = lfs_balloc(vp, uio->uio_offset, xfersize, cred, flags, 531 &bp); 532 533 if (error) 534 break; 535 if (uio->uio_offset + xfersize > ip->i_size) { 536 ip->i_size = uio->uio_offset + xfersize; 537 DIP_ASSIGN(ip, size, ip->i_size); 538 uvm_vnp_setsize(vp, ip->i_size); 539 extended = 1; 540 } 541 size = lfs_blksize(fs, ip, lbn) - bp->b_resid; 542 if (xfersize > size) 543 xfersize = size; 544 545 error = uiomove((char *)bp->b_data + blkoffset, xfersize, uio); 546 547 /* 548 * if we didn't clear the block and the uiomove failed, 549 * the buf will now contain part of some other file, 550 * so we need to invalidate it. 551 */ 552 if (error && (flags & B_CLRBUF) == 0) { 553 brelse(bp, BC_INVAL); 554 break; 555 } 556 #ifdef LFS_READWRITE 557 (void)VOP_BWRITE(bp->b_vp, bp); 558 lfs_reserve(fs, vp, NULL, 559 -lfs_btofsb(fs, (ULFS_NIADDR + 1) << fs->lfs_bshift)); 560 need_unreserve = false; 561 #else 562 if (ioflag & IO_SYNC) 563 (void)bwrite(bp); 564 else if (xfersize + blkoffset == fs->fs_bsize) 565 bawrite(bp); 566 else 567 bdwrite(bp); 568 #endif 569 if (error || xfersize == 0) 570 break; 571 } 572 #ifdef LFS_READWRITE 573 if (need_unreserve) { 574 lfs_reserve(fs, vp, NULL, 575 -lfs_btofsb(fs, (ULFS_NIADDR + 1) << fs->lfs_bshift)); 576 } 577 #endif 578 579 error = ulfs_post_write_update(vp, uio, ioflag, cred, osize, resid, 580 extended, error); 581 fstrans_done(vp->v_mount); 582 583 return (error); 584 } 585 586 static int 587 ulfs_post_write_update(struct vnode *vp, struct uio *uio, int ioflag, 588 kauth_cred_t cred, off_t osize, int resid, int extended, int oerror) 589 { 590 struct inode *ip = VTOI(vp); 591 int error = oerror; 592 593 /* Trigger ctime and mtime updates, and atime if MNT_RELATIME. */ 594 ip->i_flag |= IN_CHANGE | IN_UPDATE; 595 if (vp->v_mount->mnt_flag & MNT_RELATIME) 596 ip->i_flag |= IN_ACCESS; 597 598 /* 599 * If we successfully wrote any data and we are not the superuser, 600 * we clear the setuid and setgid bits as a precaution against 601 * tampering. 602 */ 603 if (resid > uio->uio_resid && cred) { 604 if (ip->i_mode & ISUID) { 605 if (kauth_authorize_vnode(cred, 606 KAUTH_VNODE_RETAIN_SUID, vp, NULL, EPERM) != 0) { 607 ip->i_mode &= ~ISUID; 608 DIP_ASSIGN(ip, mode, ip->i_mode); 609 } 610 } 611 612 if (ip->i_mode & ISGID) { 613 if (kauth_authorize_vnode(cred, 614 KAUTH_VNODE_RETAIN_SGID, vp, NULL, EPERM) != 0) { 615 ip->i_mode &= ~ISGID; 616 DIP_ASSIGN(ip, mode, ip->i_mode); 617 } 618 } 619 } 620 621 /* If we successfully wrote anything, notify kevent listeners. */ 622 if (resid > uio->uio_resid) 623 VN_KNOTE(vp, NOTE_WRITE | (extended ? NOTE_EXTEND : 0)); 624 625 /* 626 * Update the size on disk: truncate back to original size on 627 * error, or reflect the new size on success. 628 */ 629 if (error) { 630 (void) lfs_truncate(vp, osize, ioflag & IO_SYNC, cred); 631 uio->uio_offset -= resid - uio->uio_resid; 632 uio->uio_resid = resid; 633 } else if (resid > uio->uio_resid && (ioflag & IO_SYNC) == IO_SYNC) { 634 error = lfs_update(vp, NULL, NULL, UPDATE_WAIT); 635 } else { 636 /* nothing */ 637 } 638 639 /* Make sure the vnode uvm size matches the inode file size. */ 640 KASSERT(vp->v_size == ip->i_size); 641 642 /* Write error overrides any inode update error. */ 643 if (oerror) 644 error = oerror; 645 return error; 646 } 647
Indexes created Fri Oct 03 02:09:57 GMT 2025