1 /*- 2 * Copyright (c) 2000-2003 Tor Egge 3 * 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 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 * 26 * $FreeBSD$ 27 */ 28 29 #include <sys/param.h> 30 #include <sys/systm.h> 31 #include <sys/fcntl.h> 32 #include <sys/file.h> 33 #include <sys/stat.h> 34 #include <sys/proc.h> 35 #include <sys/limits.h> 36 #include <sys/mount.h> 37 #include <sys/namei.h> 38 #include <sys/vnode.h> 39 #include <sys/conf.h> 40 #include <sys/filio.h> 41 #include <sys/ttycom.h> 42 #include <sys/bio.h> 43 #include <sys/buf.h> 44 #include <ufs/ufs/quota.h> 45 #include <ufs/ufs/inode.h> 46 #include <ufs/ffs/fs.h> 47 48 #include <vm/vm.h> 49 #include <vm/vm_extern.h> 50 #include <vm/vm_object.h> 51 #include <sys/kernel.h> 52 #include <sys/sysctl.h> 53 54 static int ffs_rawread_readahead(struct vnode *vp, 55 caddr_t udata, 56 off_t offset, 57 size_t len, 58 struct thread *td, 59 struct buf *bp, 60 caddr_t sa); 61 static int ffs_rawread_main(struct vnode *vp, 62 struct uio *uio); 63 64 static int ffs_rawread_sync(struct vnode *vp, struct thread *td); 65 66 int ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone); 67 68 void ffs_rawread_setup(void); 69 70 static void ffs_rawreadwakeup(struct buf *bp); 71 72 73 SYSCTL_DECL(_vfs_ffs); 74 75 static int ffsrawbufcnt = 4; 76 SYSCTL_INT(_vfs_ffs, OID_AUTO, ffsrawbufcnt, CTLFLAG_RD, &ffsrawbufcnt, 0, 77 "Buffers available for raw reads"); 78 79 static int allowrawread = 1; 80 SYSCTL_INT(_vfs_ffs, OID_AUTO, allowrawread, CTLFLAG_RW, &allowrawread, 0, 81 "Flag to enable raw reads"); 82 83 static int rawreadahead = 1; 84 SYSCTL_INT(_vfs_ffs, OID_AUTO, rawreadahead, CTLFLAG_RW, &rawreadahead, 0, 85 "Flag to enable readahead for long raw reads"); 86 87 88 void 89 ffs_rawread_setup(void) 90 { 91 ffsrawbufcnt = (nswbuf > 100 ) ? (nswbuf - (nswbuf >> 4)) : nswbuf - 8; 92 } 93 94 95 static int 96 ffs_rawread_sync(struct vnode *vp, struct thread *td) 97 { 98 int spl; 99 int error; 100 int upgraded; 101 102 GIANT_REQUIRED; 103 /* Check for dirty mmap, pending writes and dirty buffers */ 104 spl = splbio(); 105 VI_LOCK(vp); 106 if (vp->v_numoutput > 0 || 107 !TAILQ_EMPTY(&vp->v_dirtyblkhd) || 108 (vp->v_iflag & VI_OBJDIRTY) != 0) { 109 splx(spl); 110 VI_UNLOCK(vp); 111 112 if (VOP_ISLOCKED(vp, td) != LK_EXCLUSIVE) { 113 upgraded = 1; 114 /* Upgrade to exclusive lock, this might block */ 115 VOP_LOCK(vp, LK_UPGRADE | LK_NOPAUSE, td); 116 } else 117 upgraded = 0; 118 119 120 /* Attempt to msync mmap() regions to clean dirty mmap */ 121 VI_LOCK(vp); 122 if ((vp->v_iflag & VI_OBJDIRTY) != 0) { 123 struct vm_object *obj; 124 VI_UNLOCK(vp); 125 if (VOP_GETVOBJECT(vp, &obj) == 0) 126 vm_object_page_clean(obj, 0, 0, OBJPC_SYNC); 127 VI_LOCK(vp); 128 } 129 130 /* Wait for pending writes to complete */ 131 spl = splbio(); 132 while (vp->v_numoutput) { 133 vp->v_iflag |= VI_BWAIT; 134 error = msleep((caddr_t)&vp->v_numoutput, 135 VI_MTX(vp), 136 PRIBIO + 1, 137 "rawrdfls", 0); 138 if (error != 0) { 139 splx(spl); 140 VI_UNLOCK(vp); 141 if (upgraded != 0) 142 VOP_LOCK(vp, LK_DOWNGRADE, td); 143 return (error); 144 } 145 } 146 /* Flush dirty buffers */ 147 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) { 148 splx(spl); 149 VI_UNLOCK(vp); 150 if ((error = VOP_FSYNC(vp, NOCRED, MNT_WAIT, td)) != 0) { 151 if (upgraded != 0) 152 VOP_LOCK(vp, LK_DOWNGRADE, td); 153 return (error); 154 } 155 VI_LOCK(vp); 156 spl = splbio(); 157 if (vp->v_numoutput > 0 || 158 !TAILQ_EMPTY(&vp->v_dirtyblkhd)) 159 panic("ffs_rawread_sync: dirty bufs"); 160 } 161 splx(spl); 162 VI_UNLOCK(vp); 163 if (upgraded != 0) 164 VOP_LOCK(vp, LK_DOWNGRADE, td); 165 } else { 166 splx(spl); 167 VI_UNLOCK(vp); 168 } 169 return 0; 170 } 171 172 173 static int 174 ffs_rawread_readahead(struct vnode *vp, 175 caddr_t udata, 176 off_t offset, 177 size_t len, 178 struct thread *td, 179 struct buf *bp, 180 caddr_t sa) 181 { 182 int error; 183 u_int iolen; 184 off_t blockno; 185 int blockoff; 186 int bsize; 187 struct vnode *dp; 188 int bforwards; 189 190 GIANT_REQUIRED; 191 bsize = vp->v_mount->mnt_stat.f_iosize; 192 193 iolen = ((vm_offset_t) udata) & PAGE_MASK; 194 bp->b_bcount = len; 195 if (bp->b_bcount + iolen > bp->b_kvasize) { 196 bp->b_bcount = bp->b_kvasize; 197 if (iolen != 0) 198 bp->b_bcount -= PAGE_SIZE; 199 } 200 bp->b_flags = B_PHYS; 201 bp->b_iocmd = BIO_READ; 202 bp->b_iodone = ffs_rawreadwakeup; 203 bp->b_data = udata; 204 bp->b_saveaddr = sa; 205 bp->b_offset = offset; 206 blockno = bp->b_offset / bsize; 207 blockoff = (bp->b_offset % bsize) / DEV_BSIZE; 208 if ((daddr_t) blockno != blockno) { 209 return EINVAL; /* blockno overflow */ 210 } 211 212 bp->b_lblkno = bp->b_blkno = blockno; 213 214 error = VOP_BMAP(vp, bp->b_lblkno, &dp, &bp->b_blkno, &bforwards, 215 NULL); 216 if (error != 0) { 217 return error; 218 } 219 if (bp->b_blkno == -1) { 220 221 /* Fill holes with NULs to preserve semantics */ 222 223 if (bp->b_bcount + blockoff * DEV_BSIZE > bsize) 224 bp->b_bcount = bsize - blockoff * DEV_BSIZE; 225 bp->b_bufsize = bp->b_bcount; 226 227 if (vmapbuf(bp) < 0) 228 return EFAULT; 229 230 if (ticks - PCPU_GET(switchticks) >= hogticks) 231 uio_yield(); 232 bzero(bp->b_data, bp->b_bufsize); 233 234 /* Mark operation completed (similar to bufdone()) */ 235 236 bp->b_resid = 0; 237 bp->b_flags |= B_DONE; 238 return 0; 239 } 240 241 if (bp->b_bcount + blockoff * DEV_BSIZE > bsize * (1 + bforwards)) 242 bp->b_bcount = bsize * (1 + bforwards) - blockoff * DEV_BSIZE; 243 bp->b_bufsize = bp->b_bcount; 244 bp->b_blkno += blockoff; 245 bp->b_dev = dp->v_rdev; 246 247 if (vmapbuf(bp) < 0) 248 return EFAULT; 249 250 if (dp->v_type == VCHR) 251 (void) VOP_SPECSTRATEGY(dp, bp); 252 else 253 (void) VOP_STRATEGY(dp, bp); 254 return 0; 255 } 256 257 258 static int 259 ffs_rawread_main(struct vnode *vp, 260 struct uio *uio) 261 { 262 int error, nerror; 263 struct buf *bp, *nbp, *tbp; 264 caddr_t sa, nsa, tsa; 265 u_int iolen; 266 int spl; 267 caddr_t udata; 268 long resid; 269 off_t offset; 270 struct thread *td; 271 272 GIANT_REQUIRED; 273 td = uio->uio_td ? uio->uio_td : curthread; 274 udata = uio->uio_iov->iov_base; 275 resid = uio->uio_resid; 276 offset = uio->uio_offset; 277 278 /* 279 * keep the process from being swapped 280 */ 281 PHOLD(td->td_proc); 282 283 error = 0; 284 nerror = 0; 285 286 bp = NULL; 287 nbp = NULL; 288 sa = NULL; 289 nsa = NULL; 290 291 while (resid > 0) { 292 293 if (bp == NULL) { /* Setup first read */ 294 /* XXX: Leave some bufs for swap */ 295 bp = getpbuf(&ffsrawbufcnt); 296 sa = bp->b_data; 297 bp->b_vp = vp; 298 error = ffs_rawread_readahead(vp, udata, offset, 299 resid, td, bp, sa); 300 if (error != 0) 301 break; 302 303 if (resid > bp->b_bufsize) { /* Setup fist readahead */ 304 /* XXX: Leave bufs for swap */ 305 if (rawreadahead != 0) 306 nbp = trypbuf(&ffsrawbufcnt); 307 else 308 nbp = NULL; 309 if (nbp != NULL) { 310 nsa = nbp->b_data; 311 nbp->b_vp = vp; 312 313 nerror = ffs_rawread_readahead(vp, 314 udata + 315 bp->b_bufsize, 316 offset + 317 bp->b_bufsize, 318 resid - 319 bp->b_bufsize, 320 td, 321 nbp, 322 nsa); 323 if (nerror) { 324 relpbuf(nbp, &ffsrawbufcnt); 325 nbp = NULL; 326 } 327 } 328 } 329 } 330 331 spl = splbio(); 332 bwait(bp, PRIBIO, "rawrd"); 333 splx(spl); 334 335 vunmapbuf(bp); 336 337 iolen = bp->b_bcount - bp->b_resid; 338 if (iolen == 0 && (bp->b_ioflags & BIO_ERROR) == 0) { 339 nerror = 0; /* Ignore possible beyond EOF error */ 340 break; /* EOF */ 341 } 342 343 if ((bp->b_ioflags & BIO_ERROR) != 0) { 344 error = bp->b_error; 345 break; 346 } 347 resid -= iolen; 348 udata += iolen; 349 offset += iolen; 350 if (iolen < bp->b_bufsize) { 351 /* Incomplete read. Try to read remaining part */ 352 error = ffs_rawread_readahead(vp, 353 udata, 354 offset, 355 bp->b_bufsize - iolen, 356 td, 357 bp, 358 sa); 359 if (error != 0) 360 break; 361 } else if (nbp != NULL) { /* Complete read with readahead */ 362 363 tbp = bp; 364 bp = nbp; 365 nbp = tbp; 366 367 tsa = sa; 368 sa = nsa; 369 nsa = tsa; 370 371 if (resid <= bp->b_bufsize) { /* No more readaheads */ 372 relpbuf(nbp, &ffsrawbufcnt); 373 nbp = NULL; 374 } else { /* Setup next readahead */ 375 nerror = ffs_rawread_readahead(vp, 376 udata + 377 bp->b_bufsize, 378 offset + 379 bp->b_bufsize, 380 resid - 381 bp->b_bufsize, 382 td, 383 nbp, 384 nsa); 385 if (nerror != 0) { 386 relpbuf(nbp, &ffsrawbufcnt); 387 nbp = NULL; 388 } 389 } 390 } else if (nerror != 0) {/* Deferred Readahead error */ 391 break; 392 } else if (resid > 0) { /* More to read, no readahead */ 393 error = ffs_rawread_readahead(vp, udata, offset, 394 resid, td, bp, sa); 395 if (error != 0) 396 break; 397 } 398 } 399 400 if (bp != NULL) 401 relpbuf(bp, &ffsrawbufcnt); 402 if (nbp != NULL) { /* Run down readahead buffer */ 403 spl = splbio(); 404 bwait(nbp, PRIBIO, "rawrd"); 405 splx(spl); 406 vunmapbuf(nbp); 407 relpbuf(nbp, &ffsrawbufcnt); 408 } 409 410 if (error == 0) 411 error = nerror; 412 PRELE(td->td_proc); 413 uio->uio_iov->iov_base = udata; 414 uio->uio_resid = resid; 415 uio->uio_offset = offset; 416 return error; 417 } 418 419 420 int 421 ffs_rawread(struct vnode *vp, 422 struct uio *uio, 423 int *workdone) 424 { 425 if (allowrawread != 0 && 426 uio->uio_iovcnt == 1 && 427 uio->uio_segflg == UIO_USERSPACE && 428 uio->uio_resid == uio->uio_iov->iov_len && 429 (((uio->uio_td != NULL) ? uio->uio_td : curthread)->td_flags & 430 TDF_DEADLKTREAT) == 0) { 431 int secsize; /* Media sector size */ 432 off_t filebytes; /* Bytes left of file */ 433 int blockbytes; /* Bytes left of file in full blocks */ 434 int partialbytes; /* Bytes in last partial block */ 435 int skipbytes; /* Bytes not to read in ffs_rawread */ 436 struct inode *ip; 437 int error; 438 439 440 /* Only handle sector aligned reads */ 441 ip = VTOI(vp); 442 secsize = ip->i_devvp->v_rdev->si_bsize_phys; 443 if ((uio->uio_offset & (secsize - 1)) == 0 && 444 (uio->uio_resid & (secsize - 1)) == 0) { 445 446 /* Sync dirty pages and buffers if needed */ 447 error = ffs_rawread_sync(vp, 448 (uio->uio_td != NULL) ? 449 uio->uio_td : curthread); 450 if (error != 0) 451 return error; 452 453 /* Check for end of file */ 454 if (ip->i_size > uio->uio_offset) { 455 filebytes = ip->i_size - uio->uio_offset; 456 457 /* No special eof handling needed ? */ 458 if (uio->uio_resid <= filebytes) { 459 *workdone = 1; 460 return ffs_rawread_main(vp, uio); 461 } 462 463 partialbytes = ((unsigned int) ip->i_size) % 464 ip->i_fs->fs_bsize; 465 blockbytes = (int) filebytes - partialbytes; 466 if (blockbytes > 0) { 467 skipbytes = uio->uio_resid - 468 blockbytes; 469 uio->uio_resid = blockbytes; 470 error = ffs_rawread_main(vp, uio); 471 uio->uio_resid += skipbytes; 472 if (error != 0) 473 return error; 474 /* Read remaining part using buffer */ 475 } 476 } 477 } 478 } 479 *workdone = 0; 480 return 0; 481 } 482 483 484 static void 485 ffs_rawreadwakeup(struct buf *bp) 486 { 487 bdone(bp); 488 } 489