| 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. All advertising materials mentioning features or use of this software |
| 14 | * must display the following acknowledgement: |
| 15 | * This product includes software developed by the University of |
| 16 | * California, Berkeley and its contributors. |
| 17 | * 4. Neither the name of the University nor the names of its contributors |
| 18 | * may be used to endorse or promote products derived from this software |
| 19 | * without specific prior written permission. |
| 20 | * |
| 21 | * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND |
| 22 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 23 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 24 | * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE |
| 25 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
| 26 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
| 27 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
| 28 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
| 29 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
| 30 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 31 | * SUCH DAMAGE. |
| 32 | * |
| 33 | * @(#)ufs_readwrite.c 8.11 (Berkeley) 5/8/95 |
| 34 | * $FreeBSD: src/sys/ufs/ufs/ufs_readwrite.c,v 1.65.2.14 2003/04/04 22:21:29 tegge Exp $ |
| 35 | * $DragonFly: src/sys/vfs/ufs/ufs_readwrite.c,v 1.26 2008/06/19 23:27:39 dillon Exp $ |
| 36 | */ |
| 37 | |
| 38 | #define BLKSIZE(a, b, c) blksize(a, b, c) |
| 39 | #define FS struct fs |
| 40 | #define I_FS i_fs |
| 41 | |
| 42 | #include <vm/vm.h> |
| 43 | #include <vm/vm_object.h> |
| 44 | #include <vm/vm_pager.h> |
| 45 | #include <vm/vm_map.h> |
| 46 | #include <vm/vnode_pager.h> |
| 47 | #include <sys/event.h> |
| 48 | #include <sys/vmmeter.h> |
| 49 | #include <sys/sysctl.h> |
| 50 | #include <vm/vm_page2.h> |
| 51 | |
| 52 | #include "opt_directio.h" |
| 53 | |
| 54 | #define VN_KNOTE(vp, b) \ |
| 55 | KNOTE((struct klist *)&vp->v_pollinfo.vpi_kqinfo.ki_note, (b)) |
| 56 | |
| 57 | #ifdef DIRECTIO |
| 58 | extern int ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone); |
| 59 | #endif |
| 60 | |
| 61 | SYSCTL_DECL(_vfs_ffs); |
| 62 | |
| 63 | /* |
| 64 | * Vnode op for reading. |
| 65 | * |
| 66 | * ffs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, |
| 67 | * struct ucred *a_cred) |
| 68 | */ |
| 69 | /* ARGSUSED */ |
| 70 | int |
| 71 | ffs_read(struct vop_read_args *ap) |
| 72 | { |
| 73 | struct vnode *vp; |
| 74 | struct inode *ip; |
| 75 | struct uio *uio; |
| 76 | FS *fs; |
| 77 | struct buf *bp; |
| 78 | off_t bytesinfile; |
| 79 | int xfersize, blkoffset; |
| 80 | int error, orig_resid; |
| 81 | u_short mode; |
| 82 | int seqcount; |
| 83 | int ioflag; |
| 84 | |
| 85 | vp = ap->a_vp; |
| 86 | seqcount = ap->a_ioflag >> 16; |
| 87 | ip = VTOI(vp); |
| 88 | mode = ip->i_mode; |
| 89 | uio = ap->a_uio; |
| 90 | ioflag = ap->a_ioflag; |
| 91 | #ifdef DIRECTIO |
| 92 | if ((ioflag & IO_DIRECT) != 0) { |
| 93 | int workdone; |
| 94 | |
| 95 | error = ffs_rawread(vp, uio, &workdone); |
| 96 | if (error || workdone) |
| 97 | return error; |
| 98 | } |
| 99 | #endif |
| 100 | |
| 101 | #ifdef DIAGNOSTIC |
| 102 | if (uio->uio_rw != UIO_READ) |
| 103 | panic("ffs_read: mode"); |
| 104 | |
| 105 | if (vp->v_type == VLNK) { |
| 106 | if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen) |
| 107 | panic("ffs_read: short symlink"); |
| 108 | } else if (vp->v_type != VREG && vp->v_type != VDIR) |
| 109 | panic("ffs_read: type %d", vp->v_type); |
| 110 | #endif |
| 111 | fs = ip->I_FS; |
| 112 | if ((uint64_t)uio->uio_offset > fs->fs_maxfilesize) |
| 113 | return (EFBIG); |
| 114 | |
| 115 | orig_resid = uio->uio_resid; |
| 116 | if (orig_resid <= 0) |
| 117 | return (0); |
| 118 | |
| 119 | bytesinfile = ip->i_size - uio->uio_offset; |
| 120 | if (bytesinfile <= 0) { |
| 121 | if ((vp->v_mount->mnt_flag & MNT_NOATIME) == 0) |
| 122 | ip->i_flag |= IN_ACCESS; |
| 123 | return 0; |
| 124 | } |
| 125 | |
| 126 | /* |
| 127 | * Ok so we couldn't do it all in one vm trick... |
| 128 | * so cycle around trying smaller bites.. |
| 129 | */ |
| 130 | for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) { |
| 131 | if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0) |
| 132 | break; |
| 133 | |
| 134 | error = ffs_blkatoff_ra(vp, uio->uio_offset, NULL, |
| 135 | &bp, seqcount); |
| 136 | if (error) |
| 137 | break; |
| 138 | |
| 139 | /* |
| 140 | * If IO_DIRECT then set B_DIRECT for the buffer. This |
| 141 | * will cause us to attempt to release the buffer later on |
| 142 | * and will cause the buffer cache to attempt to free the |
| 143 | * underlying pages. |
| 144 | */ |
| 145 | if (ioflag & IO_DIRECT) |
| 146 | bp->b_flags |= B_DIRECT; |
| 147 | |
| 148 | /* |
| 149 | * We should only get non-zero b_resid when an I/O error |
| 150 | * has occurred, which should cause us to break above. |
| 151 | * However, if the short read did not cause an error, |
| 152 | * then we want to ensure that we do not uiomove bad |
| 153 | * or uninitialized data. |
| 154 | * |
| 155 | * XXX b_resid is only valid when an actual I/O has occured |
| 156 | * and may be incorrect if the buffer is B_CACHE or if the |
| 157 | * last op on the buffer was a failed write. This KASSERT |
| 158 | * is a precursor to removing it from the UFS code. |
| 159 | */ |
| 160 | KASSERT(bp->b_resid == 0, ("bp->b_resid != 0")); |
| 161 | |
| 162 | /* |
| 163 | * Calculate how much data we can copy |
| 164 | */ |
| 165 | blkoffset = blkoff(fs, uio->uio_offset); |
| 166 | xfersize = bp->b_bufsize - blkoffset; |
| 167 | if (xfersize > uio->uio_resid) |
| 168 | xfersize = uio->uio_resid; |
| 169 | if (xfersize > bytesinfile) |
| 170 | xfersize = bytesinfile; |
| 171 | if (xfersize <= 0) { |
| 172 | panic("ufs_readwrite: impossible xfersize: %d", |
| 173 | xfersize); |
| 174 | } |
| 175 | |
| 176 | /* |
| 177 | * otherwise use the general form |
| 178 | */ |
| 179 | error = uiomovebp(bp, bp->b_data + blkoffset, xfersize, uio); |
| 180 | |
| 181 | if (error) |
| 182 | break; |
| 183 | |
| 184 | if ((ioflag & (IO_VMIO|IO_DIRECT)) && |
| 185 | (LIST_FIRST(&bp->b_dep) == NULL)) { |
| 186 | /* |
| 187 | * If there are no dependencies, and it's VMIO, |
| 188 | * then we don't need the buf, mark it available |
| 189 | * for freeing. The VM has the data. |
| 190 | */ |
| 191 | bp->b_flags |= B_RELBUF; |
| 192 | brelse(bp); |
| 193 | } else { |
| 194 | /* |
| 195 | * Otherwise let whoever |
| 196 | * made the request take care of |
| 197 | * freeing it. We just queue |
| 198 | * it onto another list. |
| 199 | */ |
| 200 | bqrelse(bp); |
| 201 | } |
| 202 | } |
| 203 | |
| 204 | /* |
| 205 | * This can only happen in the case of an error |
| 206 | * because the loop above resets bp to NULL on each iteration |
| 207 | * and on normal completion has not set a new value into it. |
| 208 | * so it must have come from a 'break' statement |
| 209 | */ |
| 210 | if (bp != NULL) { |
| 211 | if ((ioflag & (IO_VMIO|IO_DIRECT)) && |
| 212 | (LIST_FIRST(&bp->b_dep) == NULL)) { |
| 213 | bp->b_flags |= B_RELBUF; |
| 214 | brelse(bp); |
| 215 | } else { |
| 216 | bqrelse(bp); |
| 217 | } |
| 218 | } |
| 219 | |
| 220 | if ((error == 0 || uio->uio_resid != orig_resid) && |
| 221 | (vp->v_mount->mnt_flag & MNT_NOATIME) == 0) |
| 222 | ip->i_flag |= IN_ACCESS; |
| 223 | return (error); |
| 224 | } |
| 225 | |
| 226 | /* |
| 227 | * Vnode op for writing. |
| 228 | * |
| 229 | * ffs_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, |
| 230 | * struct ucred *a_cred) |
| 231 | */ |
| 232 | int |
| 233 | ffs_write(struct vop_write_args *ap) |
| 234 | { |
| 235 | struct vnode *vp; |
| 236 | struct uio *uio; |
| 237 | struct inode *ip; |
| 238 | FS *fs; |
| 239 | struct buf *bp; |
| 240 | ufs_daddr_t lbn; |
| 241 | off_t osize; |
| 242 | off_t nsize; |
| 243 | int seqcount; |
| 244 | int blkoffset, error, extended, flags, ioflag, resid, size, xfersize; |
| 245 | struct thread *td; |
| 246 | |
| 247 | extended = 0; |
| 248 | seqcount = ap->a_ioflag >> 16; |
| 249 | ioflag = ap->a_ioflag; |
| 250 | uio = ap->a_uio; |
| 251 | vp = ap->a_vp; |
| 252 | ip = VTOI(vp); |
| 253 | |
| 254 | #ifdef DIAGNOSTIC |
| 255 | if (uio->uio_rw != UIO_WRITE) |
| 256 | panic("ffs_write: mode"); |
| 257 | #endif |
| 258 | |
| 259 | switch (vp->v_type) { |
| 260 | case VREG: |
| 261 | if (ioflag & IO_APPEND) |
| 262 | uio->uio_offset = ip->i_size; |
| 263 | if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size) |
| 264 | return (EPERM); |
| 265 | /* FALLTHROUGH */ |
| 266 | case VLNK: |
| 267 | break; |
| 268 | case VDIR: |
| 269 | panic("ffs_write: dir write"); |
| 270 | break; |
| 271 | default: |
| 272 | panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type, |
| 273 | (int)uio->uio_offset, |
| 274 | (int)uio->uio_resid |
| 275 | ); |
| 276 | } |
| 277 | |
| 278 | fs = ip->I_FS; |
| 279 | if (uio->uio_offset < 0 || |
| 280 | (uint64_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize) { |
| 281 | return (EFBIG); |
| 282 | } |
| 283 | /* |
| 284 | * Maybe this should be above the vnode op call, but so long as |
| 285 | * file servers have no limits, I don't think it matters. |
| 286 | */ |
| 287 | td = uio->uio_td; |
| 288 | if (vp->v_type == VREG && td && td->td_proc && |
| 289 | uio->uio_offset + uio->uio_resid > |
| 290 | td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) { |
| 291 | lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ); |
| 292 | return (EFBIG); |
| 293 | } |
| 294 | |
| 295 | resid = uio->uio_resid; |
| 296 | osize = ip->i_size; |
| 297 | |
| 298 | /* |
| 299 | * NOTE! These B_ flags are actually balloc-only flags, not buffer |
| 300 | * flags. They are similar to the BA_ flags in fbsd. |
| 301 | */ |
| 302 | if (seqcount > B_SEQMAX) |
| 303 | flags = B_SEQMAX << B_SEQSHIFT; |
| 304 | else |
| 305 | flags = seqcount << B_SEQSHIFT; |
| 306 | if ((ioflag & IO_SYNC) && !DOINGASYNC(vp)) |
| 307 | flags |= B_SYNC; |
| 308 | |
| 309 | for (error = 0; uio->uio_resid > 0;) { |
| 310 | lbn = lblkno(fs, uio->uio_offset); |
| 311 | blkoffset = blkoff(fs, uio->uio_offset); |
| 312 | xfersize = fs->fs_bsize - blkoffset; |
| 313 | if (uio->uio_resid < xfersize) |
| 314 | xfersize = uio->uio_resid; |
| 315 | |
| 316 | if (uio->uio_offset + xfersize > ip->i_size) { |
| 317 | nsize = uio->uio_offset + xfersize; |
| 318 | nvnode_pager_setsize(vp, nsize, |
| 319 | blkoffresize(fs, nsize), blkoff(fs, nsize)); |
| 320 | } |
| 321 | |
| 322 | #if 0 |
| 323 | /* |
| 324 | * If doing a dummy write to flush the buffer for a |
| 325 | * putpages we must perform a read-before-write to |
| 326 | * fill in any missing spots and clear any invalid |
| 327 | * areas. Otherwise a multi-page buffer may not properly |
| 328 | * flush. |
| 329 | * |
| 330 | * We must clear any invalid areas |
| 331 | */ |
| 332 | if (uio->uio_segflg == UIO_NOCOPY) { |
| 333 | error = ffs_blkatoff(vp, uio->uio_offset, NULL, &bp); |
| 334 | if (error) |
| 335 | break; |
| 336 | bqrelse(bp); |
| 337 | } |
| 338 | #endif |
| 339 | |
| 340 | /* |
| 341 | * We must clear invalid areas. |
| 342 | */ |
| 343 | if (xfersize < fs->fs_bsize || uio->uio_segflg == UIO_NOCOPY) |
| 344 | flags |= B_CLRBUF; |
| 345 | else |
| 346 | flags &= ~B_CLRBUF; |
| 347 | /* XXX is uio->uio_offset the right thing here? */ |
| 348 | error = VOP_BALLOC(vp, uio->uio_offset, xfersize, |
| 349 | ap->a_cred, flags, &bp); |
| 350 | if (error != 0) |
| 351 | break; |
| 352 | /* |
| 353 | * If the buffer is not valid and we did not clear garbage |
| 354 | * out above, we have to do so here even though the write |
| 355 | * covers the entire buffer in order to avoid a mmap()/write |
| 356 | * race where another process may see the garbage prior to |
| 357 | * the uiomove() for a write replacing it. |
| 358 | */ |
| 359 | if ((bp->b_flags & B_CACHE) == 0 && (flags & B_CLRBUF) == 0) |
| 360 | vfs_bio_clrbuf(bp); |
| 361 | if (ioflag & IO_DIRECT) |
| 362 | bp->b_flags |= B_DIRECT; |
| 363 | if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL)) |
| 364 | bp->b_flags |= B_NOCACHE; |
| 365 | |
| 366 | if (uio->uio_offset + xfersize > ip->i_size) { |
| 367 | ip->i_size = uio->uio_offset + xfersize; |
| 368 | extended = 1; |
| 369 | } |
| 370 | |
| 371 | size = BLKSIZE(fs, ip, lbn) - bp->b_resid; |
| 372 | if (size < xfersize) |
| 373 | xfersize = size; |
| 374 | |
| 375 | error = uiomovebp(bp, bp->b_data + blkoffset, xfersize, uio); |
| 376 | if ((ioflag & (IO_VMIO|IO_DIRECT)) && |
| 377 | (LIST_FIRST(&bp->b_dep) == NULL)) { |
| 378 | bp->b_flags |= B_RELBUF; |
| 379 | } |
| 380 | |
| 381 | /* |
| 382 | * If IO_SYNC each buffer is written synchronously. Otherwise |
| 383 | * if we have a severe page deficiency write the buffer |
| 384 | * asynchronously. Otherwise try to cluster, and if that |
| 385 | * doesn't do it then either do an async write (if O_DIRECT), |
| 386 | * or a delayed write (if not). |
| 387 | */ |
| 388 | |
| 389 | if (ioflag & IO_SYNC) { |
| 390 | (void)bwrite(bp); |
| 391 | } else if (vm_page_count_severe() || |
| 392 | buf_dirty_count_severe() || |
| 393 | (ioflag & IO_ASYNC)) { |
| 394 | bp->b_flags |= B_CLUSTEROK; |
| 395 | bawrite(bp); |
| 396 | } else if (xfersize + blkoffset == fs->fs_bsize) { |
| 397 | if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) { |
| 398 | bp->b_flags |= B_CLUSTEROK; |
| 399 | cluster_write(bp, (off_t)ip->i_size, fs->fs_bsize, seqcount); |
| 400 | } else { |
| 401 | bawrite(bp); |
| 402 | } |
| 403 | } else if (ioflag & IO_DIRECT) { |
| 404 | bp->b_flags |= B_CLUSTEROK; |
| 405 | bawrite(bp); |
| 406 | } else { |
| 407 | bp->b_flags |= B_CLUSTEROK; |
| 408 | bdwrite(bp); |
| 409 | } |
| 410 | if (error || xfersize == 0) |
| 411 | break; |
| 412 | ip->i_flag |= IN_CHANGE | IN_UPDATE; |
| 413 | } |
| 414 | /* |
| 415 | * If we successfully wrote any data, and we are not the superuser |
| 416 | * we clear the setuid and setgid bits as a precaution against |
| 417 | * tampering. |
| 418 | */ |
| 419 | if (resid > uio->uio_resid && ap->a_cred && ap->a_cred->cr_uid != 0) |
| 420 | ip->i_mode &= ~(ISUID | ISGID); |
| 421 | if (resid > uio->uio_resid) |
| 422 | VN_KNOTE(vp, NOTE_WRITE | (extended ? NOTE_EXTEND : 0)); |
| 423 | if (error) { |
| 424 | if (ioflag & IO_UNIT) { |
| 425 | (void)ffs_truncate(vp, osize, ioflag & IO_SYNC, |
| 426 | ap->a_cred); |
| 427 | uio->uio_offset -= resid - uio->uio_resid; |
| 428 | uio->uio_resid = resid; |
| 429 | } |
| 430 | } else if (resid > uio->uio_resid && (ioflag & IO_SYNC)) { |
| 431 | error = ffs_update(vp, 1); |
| 432 | } |
| 433 | |
| 434 | return (error); |
| 435 | } |
| 436 | |