1 /*- 2 * Copyright (c) 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)ufs_readwrite.c 8.11 (Berkeley) 5/8/95 30 * $FreeBSD: src/sys/ufs/ufs/ufs_readwrite.c,v 1.65.2.14 2003/04/04 22:21:29 tegge Exp $ 31 */ 32 33 #define BLKSIZE(a, b, c) blksize(a, b, c) 34 #define FS struct fs 35 #define I_FS i_fs 36 37 #include <vm/vm.h> 38 #include <vm/vm_object.h> 39 #include <vm/vm_pager.h> 40 #include <vm/vm_map.h> 41 #include <vm/vnode_pager.h> 42 #include <sys/event.h> 43 #include <sys/vmmeter.h> 44 #include <sys/sysctl.h> 45 #include <vm/vm_page2.h> 46 47 #include "opt_directio.h" 48 49 #define VN_KNOTE(vp, b) \ 50 KNOTE((struct klist *)&vp->v_pollinfo.vpi_kqinfo.ki_note, (b)) 51 52 #ifdef DIRECTIO 53 extern int ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone); 54 #endif 55 56 SYSCTL_DECL(_vfs_ffs); 57 58 /* 59 * Vnode op for reading. 60 * 61 * ffs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 62 * struct ucred *a_cred) 63 */ 64 /* ARGSUSED */ 65 int 66 ffs_read(struct vop_read_args *ap) 67 { 68 struct vnode *vp; 69 struct inode *ip; 70 struct uio *uio; 71 FS *fs; 72 struct buf *bp; 73 off_t bytesinfile; 74 int xfersize, blkoffset; 75 int error, orig_resid; 76 int seqcount; 77 int ioflag; 78 79 vp = ap->a_vp; 80 seqcount = ap->a_ioflag >> 16; 81 ip = VTOI(vp); 82 uio = ap->a_uio; 83 ioflag = ap->a_ioflag; 84 #ifdef DIRECTIO 85 if ((ioflag & IO_DIRECT) != 0) { 86 int workdone; 87 88 error = ffs_rawread(vp, uio, &workdone); 89 if (error || workdone) 90 return error; 91 } 92 #endif 93 94 #ifdef DIAGNOSTIC 95 if (uio->uio_rw != UIO_READ) 96 panic("ffs_read: mode"); 97 98 if (vp->v_type == VLNK) { 99 if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen) 100 panic("ffs_read: short symlink"); 101 } else if (vp->v_type != VREG && vp->v_type != VDIR) 102 panic("ffs_read: type %d", vp->v_type); 103 #endif 104 fs = ip->I_FS; 105 if ((uint64_t)uio->uio_offset > fs->fs_maxfilesize) 106 return (EFBIG); 107 108 orig_resid = uio->uio_resid; 109 if (orig_resid <= 0) 110 return (0); 111 112 bytesinfile = ip->i_size - uio->uio_offset; 113 if (bytesinfile <= 0) { 114 if ((vp->v_mount->mnt_flag & MNT_NOATIME) == 0) 115 ip->i_flag |= IN_ACCESS; 116 return 0; 117 } 118 119 /* 120 * Ok so we couldn't do it all in one vm trick... 121 * so cycle around trying smaller bites.. 122 */ 123 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) { 124 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0) 125 break; 126 127 error = ffs_blkatoff_ra(vp, uio->uio_offset, NULL, 128 &bp, seqcount); 129 if (error) 130 break; 131 132 /* 133 * If IO_DIRECT then set B_DIRECT for the buffer. This 134 * will cause us to attempt to release the buffer later on 135 * and will cause the buffer cache to attempt to free the 136 * underlying pages. 137 */ 138 if (ioflag & IO_DIRECT) 139 bp->b_flags |= B_DIRECT; 140 141 /* 142 * We should only get non-zero b_resid when an I/O error 143 * has occurred, which should cause us to break above. 144 * However, if the short read did not cause an error, 145 * then we want to ensure that we do not uiomove bad 146 * or uninitialized data. 147 * 148 * XXX b_resid is only valid when an actual I/O has occured 149 * and may be incorrect if the buffer is B_CACHE or if the 150 * last op on the buffer was a failed write. This KASSERT 151 * is a precursor to removing it from the UFS code. 152 */ 153 KASSERT(bp->b_resid == 0, ("bp->b_resid != 0")); 154 155 /* 156 * Calculate how much data we can copy 157 */ 158 blkoffset = blkoff(fs, uio->uio_offset); 159 xfersize = bp->b_bufsize - blkoffset; 160 if (xfersize > uio->uio_resid) 161 xfersize = uio->uio_resid; 162 if (xfersize > bytesinfile) 163 xfersize = bytesinfile; 164 if (xfersize <= 0) { 165 panic("ufs_readwrite: impossible xfersize: %d", 166 xfersize); 167 } 168 169 /* 170 * otherwise use the general form 171 */ 172 error = uiomovebp(bp, bp->b_data + blkoffset, xfersize, uio); 173 174 if (error) 175 break; 176 177 if ((ioflag & (IO_VMIO|IO_DIRECT)) && 178 (LIST_FIRST(&bp->b_dep) == NULL)) { 179 /* 180 * If there are no dependencies, and it's VMIO, 181 * then we don't need the buf, mark it available 182 * for freeing. The VM has the data. 183 */ 184 bp->b_flags |= B_RELBUF; 185 brelse(bp); 186 } else { 187 /* 188 * Otherwise let whoever 189 * made the request take care of 190 * freeing it. We just queue 191 * it onto another list. 192 */ 193 bqrelse(bp); 194 } 195 } 196 197 /* 198 * This can only happen in the case of an error 199 * because the loop above resets bp to NULL on each iteration 200 * and on normal completion has not set a new value into it. 201 * so it must have come from a 'break' statement 202 */ 203 if (bp != NULL) { 204 if ((ioflag & (IO_VMIO|IO_DIRECT)) && 205 (LIST_FIRST(&bp->b_dep) == NULL)) { 206 bp->b_flags |= B_RELBUF; 207 brelse(bp); 208 } else { 209 bqrelse(bp); 210 } 211 } 212 213 if ((error == 0 || uio->uio_resid != orig_resid) && 214 (vp->v_mount->mnt_flag & MNT_NOATIME) == 0) 215 ip->i_flag |= IN_ACCESS; 216 return (error); 217 } 218 219 /* 220 * Vnode op for writing. 221 * 222 * ffs_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 223 * struct ucred *a_cred) 224 */ 225 int 226 ffs_write(struct vop_write_args *ap) 227 { 228 struct vnode *vp; 229 struct uio *uio; 230 struct inode *ip; 231 FS *fs; 232 struct buf *bp; 233 ufs_daddr_t lbn; 234 off_t osize; 235 off_t nsize; 236 int seqcount; 237 int blkoffset, error, extended, flags, ioflag, resid, size, xfersize; 238 struct thread *td; 239 240 extended = 0; 241 seqcount = ap->a_ioflag >> 16; 242 ioflag = ap->a_ioflag; 243 uio = ap->a_uio; 244 vp = ap->a_vp; 245 ip = VTOI(vp); 246 247 #ifdef DIAGNOSTIC 248 if (uio->uio_rw != UIO_WRITE) 249 panic("ffs_write: mode"); 250 #endif 251 252 switch (vp->v_type) { 253 case VREG: 254 if (ioflag & IO_APPEND) 255 uio->uio_offset = ip->i_size; 256 if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size) 257 return (EPERM); 258 /* FALLTHROUGH */ 259 case VLNK: 260 break; 261 case VDIR: 262 panic("ffs_write: dir write"); 263 break; 264 default: 265 panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type, 266 (int)uio->uio_offset, 267 (int)uio->uio_resid 268 ); 269 } 270 271 fs = ip->I_FS; 272 if (uio->uio_offset < 0 || 273 (uint64_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize) { 274 return (EFBIG); 275 } 276 /* 277 * Maybe this should be above the vnode op call, but so long as 278 * file servers have no limits, I don't think it matters. 279 */ 280 td = uio->uio_td; 281 if (vp->v_type == VREG && td && td->td_proc && 282 uio->uio_offset + uio->uio_resid > 283 td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) { 284 lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ); 285 return (EFBIG); 286 } 287 288 resid = uio->uio_resid; 289 osize = ip->i_size; 290 291 /* 292 * NOTE! These B_ flags are actually balloc-only flags, not buffer 293 * flags. They are similar to the BA_ flags in fbsd. 294 */ 295 if (seqcount > B_SEQMAX) 296 flags = B_SEQMAX << B_SEQSHIFT; 297 else 298 flags = seqcount << B_SEQSHIFT; 299 if ((ioflag & IO_SYNC) && !DOINGASYNC(vp)) 300 flags |= B_SYNC; 301 302 for (error = 0; uio->uio_resid > 0;) { 303 lbn = lblkno(fs, uio->uio_offset); 304 blkoffset = blkoff(fs, uio->uio_offset); 305 xfersize = fs->fs_bsize - blkoffset; 306 if (uio->uio_resid < xfersize) 307 xfersize = uio->uio_resid; 308 309 if (uio->uio_offset + xfersize > ip->i_size) { 310 nsize = uio->uio_offset + xfersize; 311 nvnode_pager_setsize(vp, nsize, 312 blkoffresize(fs, nsize), blkoff(fs, nsize)); 313 } 314 315 #if 0 316 /* 317 * If doing a dummy write to flush the buffer for a 318 * putpages we must perform a read-before-write to 319 * fill in any missing spots and clear any invalid 320 * areas. Otherwise a multi-page buffer may not properly 321 * flush. 322 * 323 * We must clear any invalid areas 324 */ 325 if (uio->uio_segflg == UIO_NOCOPY) { 326 error = ffs_blkatoff(vp, uio->uio_offset, NULL, &bp); 327 if (error) 328 break; 329 bqrelse(bp); 330 } 331 #endif 332 333 /* 334 * We must clear invalid areas. 335 */ 336 if (xfersize < fs->fs_bsize || uio->uio_segflg == UIO_NOCOPY) 337 flags |= B_CLRBUF; 338 else 339 flags &= ~B_CLRBUF; 340 /* XXX is uio->uio_offset the right thing here? */ 341 error = VOP_BALLOC(vp, uio->uio_offset, xfersize, 342 ap->a_cred, flags, &bp); 343 if (error != 0) 344 break; 345 /* 346 * If the buffer is not valid and we did not clear garbage 347 * out above, we have to do so here even though the write 348 * covers the entire buffer in order to avoid a mmap()/write 349 * race where another process may see the garbage prior to 350 * the uiomove() for a write replacing it. 351 */ 352 if ((bp->b_flags & B_CACHE) == 0 && (flags & B_CLRBUF) == 0) 353 vfs_bio_clrbuf(bp); 354 if (ioflag & IO_DIRECT) 355 bp->b_flags |= B_DIRECT; 356 if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL)) 357 bp->b_flags |= B_NOCACHE; 358 359 if (uio->uio_offset + xfersize > ip->i_size) { 360 ip->i_size = uio->uio_offset + xfersize; 361 extended = 1; 362 } 363 364 size = BLKSIZE(fs, ip, lbn) - bp->b_resid; 365 if (size < xfersize) 366 xfersize = size; 367 368 error = uiomovebp(bp, bp->b_data + blkoffset, xfersize, uio); 369 if ((ioflag & (IO_VMIO|IO_DIRECT)) && 370 (LIST_FIRST(&bp->b_dep) == NULL)) { 371 bp->b_flags |= B_RELBUF; 372 } 373 374 /* 375 * If IO_SYNC each buffer is written synchronously. Otherwise 376 * if we have a severe page deficiency write the buffer 377 * asynchronously. Otherwise try to cluster, and if that 378 * doesn't do it then either do an async write (if O_DIRECT), 379 * or a delayed write (if not). 380 */ 381 382 if (ioflag & IO_SYNC) { 383 (void)bwrite(bp); 384 } else if (vm_page_count_severe() || 385 buf_dirty_count_severe() || 386 (ioflag & IO_ASYNC)) { 387 bp->b_flags |= B_CLUSTEROK; 388 bawrite(bp); 389 } else if (xfersize + blkoffset == fs->fs_bsize) { 390 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) { 391 bp->b_flags |= B_CLUSTEROK; 392 cluster_write(bp, (off_t)ip->i_size, fs->fs_bsize, seqcount); 393 } else { 394 bawrite(bp); 395 } 396 } else if (ioflag & IO_DIRECT) { 397 bp->b_flags |= B_CLUSTEROK; 398 bawrite(bp); 399 } else { 400 bp->b_flags |= B_CLUSTEROK; 401 bdwrite(bp); 402 } 403 if (error || xfersize == 0) 404 break; 405 ip->i_flag |= IN_CHANGE | IN_UPDATE; 406 } 407 /* 408 * If we successfully wrote any data, and we are not the superuser 409 * we clear the setuid and setgid bits as a precaution against 410 * tampering. 411 */ 412 if (resid > uio->uio_resid && ap->a_cred && ap->a_cred->cr_uid != 0) 413 ip->i_mode &= ~(ISUID | ISGID); 414 if (resid > uio->uio_resid) 415 VN_KNOTE(vp, NOTE_WRITE | (extended ? NOTE_EXTEND : 0)); 416 if (error) { 417 if (ioflag & IO_UNIT) { 418 (void)ffs_truncate(vp, osize, ioflag & IO_SYNC, 419 ap->a_cred); 420 uio->uio_offset -= resid - uio->uio_resid; 421 uio->uio_resid = resid; 422 } 423 } else if (resid > uio->uio_resid && (ioflag & IO_SYNC)) { 424 error = ffs_update(vp, 1); 425 } 426 427 return (error); 428 } 429 430