1 /* $NetBSD: ufs_readwrite.c,v 1.131 2026/07/22 14:44:49 hannken Exp $ */ 2 3 /*- 4 * Copyright (c) 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * @(#)ufs_readwrite.c 8.11 (Berkeley) 5/8/95 32 */ 33 34 #include <sys/cdefs.h> 35 __KERNEL_RCSID(1, "$NetBSD: ufs_readwrite.c,v 1.131 2026/07/22 14:44:49 hannken Exp $"); 36 37 #include <sys/bitops.h> 38 #include <sys/sdt.h> 39 40 #define FS struct fs 41 #define I_FS i_fs 42 #define READ ffs_read 43 #define READ_S "ffs_read" 44 #define WRITE ffs_write 45 #define WRITE_S "ffs_write" 46 #define BUFRD ffs_bufrd 47 #define BUFWR ffs_bufwr 48 #define ufs_blkoff ffs_blkoff 49 #define ufs_blksize ffs_blksize 50 #define ufs_lblkno ffs_lblkno 51 #define ufs_lblktosize ffs_lblktosize 52 #define ufs_blkroundup ffs_blkroundup 53 54 static int ufs_post_read_update(struct vnode *, int, int); 55 static int ufs_post_write_update(struct vnode *, struct uio *, int, 56 kauth_cred_t, off_t, int, int); 57 58 /* 59 * Vnode op for reading. 60 */ 61 /* ARGSUSED */ 62 int 63 READ(void *v) 64 { 65 struct vop_read_args /* { 66 struct vnode *a_vp; 67 struct uio *a_uio; 68 int a_ioflag; 69 kauth_cred_t a_cred; 70 } */ *ap = v; 71 struct vnode *vp; 72 struct inode *ip; 73 struct uio *uio; 74 struct ufsmount *ump; 75 vsize_t bytelen; 76 int error, ioflag, advice; 77 78 vp = ap->a_vp; 79 ip = VTOI(vp); 80 ump = ip->i_ump; 81 uio = ap->a_uio; 82 ioflag = ap->a_ioflag; 83 error = 0; 84 85 KASSERT(uio->uio_rw == UIO_READ); 86 KASSERT(vp->v_type == VREG || vp->v_type == VDIR); 87 88 /* XXX Eliminate me by refusing directory reads from userland. */ 89 if (vp->v_type == VDIR) 90 return BUFRD(vp, uio, ioflag, ap->a_cred); 91 if ((u_int64_t)uio->uio_offset > ump->um_maxfilesize) 92 return SET_ERROR(EFBIG); 93 if (uio->uio_resid == 0) 94 return 0; 95 96 if ((ip->i_flags & (SF_SNAPSHOT | SF_SNAPINVAL)) == SF_SNAPSHOT) 97 #ifndef FFS_NO_SNAPSHOT 98 return ffs_snapshot_read(vp, uio, ioflag); 99 #else 100 panic("%s: unexpected snapshot", __func__); 101 #endif 102 103 if (uio->uio_offset >= ip->i_size) 104 goto out; 105 106 KASSERT(vp->v_type == VREG); 107 advice = IO_ADV_DECODE(ap->a_ioflag); 108 while (uio->uio_resid > 0) { 109 if (ioflag & IO_DIRECT) { 110 genfs_directio(vp, uio, ioflag); 111 } 112 bytelen = MIN(ip->i_size - uio->uio_offset, uio->uio_resid); 113 if (bytelen == 0) 114 break; 115 error = ubc_uiomove(&vp->v_uobj, uio, bytelen, advice, 116 UBC_READ | UBC_PARTIALOK | UBC_VNODE_FLAGS(vp)); 117 if (error) 118 break; 119 } 120 121 out: 122 error = ufs_post_read_update(vp, ap->a_ioflag, error); 123 return error; 124 } 125 126 /* 127 * UFS op for reading via the buffer cache 128 */ 129 int 130 BUFRD(struct vnode *vp, struct uio *uio, int ioflag, kauth_cred_t cred) 131 { 132 struct inode *ip; 133 struct ufsmount *ump; 134 FS *fs; 135 struct buf *bp; 136 daddr_t lbn, nextlbn; 137 off_t bytesinfile; 138 long size, xfersize, blkoffset; 139 int error; 140 141 KASSERT(VOP_ISLOCKED(vp)); 142 KASSERT(vp->v_type == VDIR || vp->v_type == VLNK); 143 KASSERT(uio->uio_rw == UIO_READ); 144 145 ip = VTOI(vp); 146 ump = ip->i_ump; 147 fs = ip->I_FS; 148 error = 0; 149 150 KASSERT(vp->v_type != VLNK || ip->i_size >= ump->um_maxsymlinklen); 151 KASSERT(vp->v_type != VLNK || ump->um_maxsymlinklen != 0 || 152 DIP(ip, blocks) != 0); 153 154 if (uio->uio_offset > ump->um_maxfilesize) 155 return SET_ERROR(EFBIG); 156 if (uio->uio_resid == 0) 157 return 0; 158 159 KASSERT(!ISSET(ip->i_flags, (SF_SNAPSHOT | SF_SNAPINVAL))); 160 161 if (uio->uio_offset >= ip->i_size) 162 goto out; 163 164 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) { 165 bytesinfile = ip->i_size - uio->uio_offset; 166 if (bytesinfile <= 0) 167 break; 168 lbn = ufs_lblkno(fs, uio->uio_offset); 169 nextlbn = lbn + 1; 170 size = ufs_blksize(fs, ip, lbn); 171 blkoffset = ufs_blkoff(fs, uio->uio_offset); 172 xfersize = MIN(MIN(fs->fs_bsize - blkoffset, uio->uio_resid), 173 bytesinfile); 174 175 if (ufs_lblktosize(fs, nextlbn) >= ip->i_size) 176 error = bread(vp, lbn, size, 0, &bp); 177 else { 178 int nextsize = ufs_blksize(fs, ip, nextlbn); 179 error = breadn(vp, lbn, 180 size, &nextlbn, &nextsize, 1, 0, &bp); 181 } 182 if (error) 183 break; 184 185 /* 186 * We should only get non-zero b_resid when an I/O error 187 * has occurred, which should cause us to break above. 188 * However, if the short read did not cause an error, 189 * then we want to ensure that we do not uiomove bad 190 * or uninitialized data. 191 */ 192 size -= bp->b_resid; 193 if (size < xfersize) { 194 if (size == 0) 195 break; 196 xfersize = size; 197 } 198 error = uiomove((char *)bp->b_data + blkoffset, xfersize, uio); 199 if (error) 200 break; 201 brelse(bp, 0); 202 } 203 if (bp != NULL) 204 brelse(bp, 0); 205 206 out: 207 error = ufs_post_read_update(vp, ioflag, error); 208 return error; 209 } 210 211 static int 212 ufs_post_read_update(struct vnode *vp, int ioflag, int oerror) 213 { 214 struct inode *ip = VTOI(vp); 215 int error = oerror; 216 217 if (!(vp->v_mount->mnt_flag & MNT_NOATIME)) { 218 ip->i_flag |= IN_ACCESS; 219 if ((ioflag & IO_SYNC) == IO_SYNC) { 220 error = UFS_WAPBL_BEGIN(vp->v_mount); 221 if (error) 222 goto out; 223 error = UFS_UPDATE(vp, NULL, NULL, UPDATE_WAIT); 224 UFS_WAPBL_END(vp->v_mount); 225 } 226 } 227 228 out: 229 /* Read error overrides any inode update error. */ 230 if (oerror) 231 error = oerror; 232 return error; 233 } 234 235 /* 236 * Vnode op for writing. 237 */ 238 int 239 WRITE(void *v) 240 { 241 struct vop_write_args /* { 242 struct vnode *a_vp; 243 struct uio *a_uio; 244 int a_ioflag; 245 kauth_cred_t a_cred; 246 } */ *ap = v; 247 struct vnode *vp; 248 struct uio *uio; 249 struct inode *ip; 250 FS *fs; 251 kauth_cred_t cred; 252 off_t osize, origoff, oldoff, preallocoff, endallocoff, nsize; 253 int blkoffset, error, flags, ioflag, resid; 254 int aflag; 255 vsize_t bytelen; 256 bool async; 257 struct ufsmount *ump; 258 const unsigned int fshift = ilog2(MAXPHYS); 259 260 cred = ap->a_cred; 261 ioflag = ap->a_ioflag; 262 uio = ap->a_uio; 263 vp = ap->a_vp; 264 ip = VTOI(vp); 265 ump = ip->i_ump; 266 267 KASSERT(vp->v_size == ip->i_size); 268 KASSERT(uio->uio_rw == UIO_WRITE); 269 KASSERT(vp->v_type == VREG); 270 KASSERT(!ISSET(ioflag, IO_JOURNALLOCKED)); 271 UFS_WAPBL_JUNLOCK_ASSERT(vp->v_mount); 272 273 if (ioflag & IO_APPEND) 274 uio->uio_offset = ip->i_size; 275 if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size) 276 return SET_ERROR(EPERM); 277 278 fs = ip->I_FS; 279 if (uio->uio_offset < 0 || 280 (u_int64_t)uio->uio_offset + uio->uio_resid > ump->um_maxfilesize) 281 return SET_ERROR(EFBIG); 282 if (uio->uio_resid == 0) 283 return 0; 284 285 flags = ioflag & IO_SYNC ? B_SYNC : 0; 286 async = vp->v_mount->mnt_flag & MNT_ASYNC; 287 origoff = uio->uio_offset; 288 resid = uio->uio_resid; 289 osize = ip->i_size; 290 error = 0; 291 292 KASSERT(vp->v_type == VREG); 293 294 /* 295 * XXX The entire write operation must occur in a single WAPBL 296 * transaction because it may allocate disk blocks, if 297 * appending or filling holes, which is allowed to happen only 298 * if the write fully succeeds. 299 * 300 * If ubc_uiomove fails in the middle with EFAULT, we can clean 301 * up at the end with UFS_TRUNCATE. But if the power fails in 302 * the middle, there would be nobody to deallocate the blocks, 303 * without an fsck to globally analyze the file system. 304 * 305 * If the increasingly inaccurately named WAPBL were augmented 306 * with rollback records for block allocations, then we could 307 * split this into multiple transactions and commit the 308 * allocations in the last one. 309 * 310 * But WAPBL doesn't have that notion now, so we'll have to 311 * live with gigantic transactions and WAPBL tentacles in 312 * genfs_getpages/putpages to cope with the possibility that 313 * the transaction may or may not be locked on entry to the 314 * page cache. 315 * 316 * And even if we added that notion to WAPBL, it wouldn't help 317 * us get rid of the tentacles in genfs_getpages/putpages 318 * because we'd have to interoperate with old implementations 319 * that assume they can replay the log without fsck. 320 */ 321 error = UFS_WAPBL_BEGIN(vp->v_mount); 322 if (error) { 323 return error; 324 } 325 326 327 preallocoff = round_page(ufs_blkroundup(fs, MAX(osize, uio->uio_offset))); 328 aflag = ioflag & IO_SYNC ? B_SYNC : 0; 329 nsize = MAX(osize, uio->uio_offset + uio->uio_resid); 330 endallocoff = nsize - ufs_blkoff(fs, nsize); 331 332 /* 333 * if we're increasing the file size, deal with expanding 334 * the fragment if there is one. 335 */ 336 337 if (nsize > osize && ufs_lblkno(fs, osize) < UFS_NDADDR && 338 ufs_lblkno(fs, osize) != ufs_lblkno(fs, nsize) && 339 ufs_blkroundup(fs, osize) != osize) { 340 off_t eob; 341 342 eob = ufs_blkroundup(fs, osize); 343 uvm_vnp_setwritesize(vp, eob); 344 error = ufs_balloc_range(vp, osize, eob - osize, cred, aflag); 345 if (error) 346 goto out; 347 if (flags & B_SYNC) { 348 rw_enter(vp->v_uobj.vmobjlock, RW_WRITER); 349 VOP_PUTPAGES(vp, trunc_page(osize & fs->fs_bmask), 350 round_page(eob), 351 PGO_CLEANIT | PGO_SYNCIO | PGO_JOURNALLOCKED); 352 } 353 } 354 355 while (uio->uio_resid > 0) { 356 int ubc_flags = UBC_WRITE; 357 bool overwrite; /* if we're overwrite a whole block */ 358 off_t newoff; 359 360 if (ioflag & IO_DIRECT) { 361 genfs_directio(vp, uio, ioflag | IO_JOURNALLOCKED); 362 } 363 364 oldoff = uio->uio_offset; 365 blkoffset = ufs_blkoff(fs, uio->uio_offset); 366 bytelen = MIN(fs->fs_bsize - blkoffset, uio->uio_resid); 367 if (bytelen == 0) { 368 break; 369 } 370 371 /* 372 * if we're filling in a hole, allocate the blocks now and 373 * initialize the pages first. if we're extending the file, 374 * we can safely allocate blocks without initializing pages 375 * since the new blocks will be inaccessible until the write 376 * is complete. 377 */ 378 overwrite = uio->uio_offset >= preallocoff && 379 uio->uio_offset < endallocoff; 380 if (!overwrite && (vp->v_vflag & VV_MAPPED) == 0 && 381 ufs_blkoff(fs, uio->uio_offset) == 0 && 382 (uio->uio_offset & PAGE_MASK) == 0) { 383 vsize_t len; 384 385 len = trunc_page(bytelen); 386 len -= ufs_blkoff(fs, len); 387 if (len > 0) { 388 overwrite = true; 389 bytelen = len; 390 } 391 } 392 393 newoff = oldoff + bytelen; 394 if (vp->v_size < newoff) { 395 uvm_vnp_setwritesize(vp, newoff); 396 } 397 398 if (!overwrite) { 399 error = ufs_balloc_range(vp, uio->uio_offset, bytelen, 400 cred, aflag); 401 if (error) 402 break; 403 } else { 404 genfs_node_wrlock(vp); 405 error = GOP_ALLOC(vp, uio->uio_offset, bytelen, 406 aflag, cred); 407 genfs_node_unlock(vp); 408 if (error) 409 break; 410 ubc_flags |= UBC_FAULTBUSY; 411 } 412 413 /* 414 * copy the data. 415 */ 416 417 error = ubc_uiomove(&vp->v_uobj, uio, bytelen, 418 IO_ADV_DECODE(ioflag), ubc_flags | UBC_VNODE_FLAGS(vp)); 419 420 /* 421 * update UVM's notion of the size now that we've 422 * copied the data into the vnode's pages. 423 * 424 * we should update the size even when uiomove failed. 425 */ 426 427 if (vp->v_size < newoff) { 428 uvm_vnp_setsize(vp, newoff); 429 } 430 431 if (error) 432 break; 433 434 /* 435 * flush what we just wrote if necessary. 436 * XXXUBC simplistic async flushing. 437 */ 438 439 if (!async && oldoff >> fshift != uio->uio_offset >> fshift) { 440 rw_enter(vp->v_uobj.vmobjlock, RW_WRITER); 441 error = VOP_PUTPAGES(vp, (oldoff >> fshift) << fshift, 442 (uio->uio_offset >> fshift) << fshift, 443 PGO_CLEANIT | PGO_JOURNALLOCKED | PGO_LAZY); 444 if (error) 445 break; 446 } 447 } 448 if (error == 0 && ioflag & IO_SYNC) { 449 rw_enter(vp->v_uobj.vmobjlock, RW_WRITER); 450 error = VOP_PUTPAGES(vp, trunc_page(origoff & fs->fs_bmask), 451 round_page(ufs_blkroundup(fs, uio->uio_offset)), 452 PGO_CLEANIT | PGO_SYNCIO | PGO_JOURNALLOCKED); 453 } 454 455 out: 456 error = ufs_post_write_update(vp, uio, ioflag, cred, osize, resid, 457 error); 458 UFS_WAPBL_END(vp->v_mount); 459 460 return error; 461 } 462 463 /* 464 * UFS op for writing via the buffer cache 465 */ 466 int 467 BUFWR(struct vnode *vp, struct uio *uio, int ioflag, kauth_cred_t cred) 468 { 469 struct inode *ip; 470 struct ufsmount *ump; 471 FS *fs; 472 int flags; 473 struct buf *bp; 474 off_t osize; 475 int resid, xfersize, size, blkoffset; 476 daddr_t lbn; 477 int error; 478 479 KASSERT(ISSET(ioflag, IO_NODELOCKED)); 480 KASSERT(VOP_ISLOCKED(vp) == LK_EXCLUSIVE); 481 KASSERT(vp->v_type == VDIR || vp->v_type == VLNK); 482 KASSERT(vp->v_type != VDIR || ISSET(ioflag, IO_SYNC)); 483 KASSERT(uio->uio_rw == UIO_WRITE); 484 KASSERT(ISSET(ioflag, IO_JOURNALLOCKED)); 485 UFS_WAPBL_JLOCK_ASSERT(vp->v_mount); 486 487 ip = VTOI(vp); 488 ump = ip->i_ump; 489 fs = ip->I_FS; 490 491 KASSERT(vp->v_size == ip->i_size); 492 493 if (uio->uio_offset < 0 || 494 uio->uio_resid > ump->um_maxfilesize || 495 uio->uio_offset > (ump->um_maxfilesize - uio->uio_resid)) 496 return SET_ERROR(EFBIG); 497 if (uio->uio_resid == 0) 498 return 0; 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); 506 507 508 /* XXX Should never have pages cached here. */ 509 KASSERT(vp->v_uobj.uo_npages == 0); 510 while (uio->uio_resid > 0) { 511 lbn = ufs_lblkno(fs, uio->uio_offset); 512 blkoffset = ufs_blkoff(fs, uio->uio_offset); 513 xfersize = MIN(fs->fs_bsize - blkoffset, uio->uio_resid); 514 if (fs->fs_bsize > xfersize) 515 flags |= B_CLRBUF; 516 else 517 flags &= ~B_CLRBUF; 518 519 error = UFS_BALLOC(vp, uio->uio_offset, xfersize, cred, flags, 520 &bp); 521 522 if (error) 523 break; 524 if (uio->uio_offset + xfersize > ip->i_size) { 525 ip->i_size = uio->uio_offset + xfersize; 526 DIP_ASSIGN(ip, size, ip->i_size); 527 uvm_vnp_setsize(vp, ip->i_size); 528 } 529 size = ufs_blksize(fs, ip, lbn) - bp->b_resid; 530 if (xfersize > size) 531 xfersize = size; 532 533 error = uiomove((char *)bp->b_data + blkoffset, xfersize, uio); 534 535 /* 536 * if we didn't clear the block and the uiomove failed, 537 * the buf will now contain part of some other file, 538 * so we need to invalidate it. 539 */ 540 if (error && (flags & B_CLRBUF) == 0) { 541 brelse(bp, BC_INVAL); 542 break; 543 } 544 if (ioflag & IO_SYNC) 545 (void)bwrite(bp); 546 else if (xfersize + blkoffset == fs->fs_bsize) 547 bawrite(bp); 548 else 549 bdwrite(bp); 550 if (error || xfersize == 0) 551 break; 552 } 553 554 error = ufs_post_write_update(vp, uio, ioflag, cred, osize, resid, 555 error); 556 557 return error; 558 } 559 560 static int 561 ufs_post_write_update(struct vnode *vp, struct uio *uio, int ioflag, 562 kauth_cred_t cred, off_t osize, int resid, int oerror) 563 { 564 struct inode *ip = VTOI(vp); 565 int error = oerror; 566 567 /* Trigger ctime and mtime updates, and atime if MNT_RELATIME. */ 568 ip->i_flag |= IN_CHANGE | IN_UPDATE; 569 if (vp->v_mount->mnt_flag & MNT_RELATIME) 570 ip->i_flag |= IN_ACCESS; 571 572 /* 573 * If we successfully wrote any data and we are not the superuser, 574 * we clear the setuid and setgid bits as a precaution against 575 * tampering. 576 */ 577 if (resid > uio->uio_resid && cred) { 578 if (ip->i_mode & ISUID) { 579 if (kauth_authorize_vnode(cred, 580 KAUTH_VNODE_RETAIN_SUID, vp, NULL, EPERM) != 0) { 581 ip->i_mode &= ~ISUID; 582 DIP_ASSIGN(ip, mode, ip->i_mode); 583 } 584 } 585 586 if (ip->i_mode & ISGID) { 587 if (kauth_authorize_vnode(cred, 588 KAUTH_VNODE_RETAIN_SGID, vp, NULL, EPERM) != 0) { 589 ip->i_mode &= ~ISGID; 590 DIP_ASSIGN(ip, mode, ip->i_mode); 591 } 592 } 593 } 594 595 /* 596 * Update the size on disk: truncate back to original size on 597 * error, or reflect the new size on success. 598 */ 599 if (error) { 600 (void) UFS_TRUNCATE(vp, osize, ioflag & IO_SYNC, cred); 601 uio->uio_offset -= resid - uio->uio_resid; 602 uio->uio_resid = resid; 603 } else if (resid > uio->uio_resid && (ioflag & IO_SYNC) == IO_SYNC) 604 error = UFS_UPDATE(vp, NULL, NULL, UPDATE_WAIT); 605 else 606 UFS_WAPBL_UPDATE(vp, NULL, NULL, 0); 607 608 /* Make sure the vnode uvm size matches the inode file size. */ 609 KASSERT(vp->v_size == ip->i_size); 610 611 /* Write error overrides any inode update error. */ 612 if (oerror) 613 error = oerror; 614 return error; 615 } 616
Indexes created Sun Oct 25 00:26:45 UTC 2026