sys/ufs/ffs/ffs_rawread.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
   3  *
   4  * Copyright (c) 2000-2003 Tor Egge
   5  * 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  *
  16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  26  * SUCH DAMAGE.
  27  */
  28
  29 #include <sys/cdefs.h>
  30 __FBSDID("$FreeBSD$");
  31
  32 #include <sys/param.h>
  33 #include <sys/systm.h>
  34 #include <sys/fcntl.h>
  35 #include <sys/file.h>
  36 #include <sys/stat.h>
  37 #include <sys/proc.h>
  38 #include <sys/limits.h>
  39 #include <sys/mount.h>
  40 #include <sys/namei.h>
  41 #include <sys/vnode.h>
  42 #include <sys/conf.h>
  43 #include <sys/filio.h>
  44 #include <sys/ttycom.h>
  45 #include <sys/bio.h>
  46 #include <sys/buf.h>
  47 #include <sys/rwlock.h>
  48 #include <ufs/ufs/extattr.h>
  49 #include <ufs/ufs/quota.h>
  50 #include <ufs/ufs/inode.h>
  51 #include <ufs/ufs/ufsmount.h>
  52 #include <ufs/ufs/ufs_extern.h>
  53 #include <ufs/ffs/fs.h>
  54 #include <ufs/ffs/ffs_extern.h>
  55
  56 #include <vm/vm.h>
  57 #include <vm/vm_extern.h>
  58 #include <vm/vm_object.h>
  59 #include <sys/kernel.h>
  60 #include <sys/sysctl.h>
  61
  62 static int ffs_rawread_readahead(struct vnode *vp,
  63                                  caddr_t udata,
  64                                  off_t offset,
  65                                  size_t len,
  66                                  struct thread *td,
  67                                  struct buf *bp);
  68 static int ffs_rawread_main(struct vnode *vp,
  69                             struct uio *uio);
  70
  71 static int ffs_rawread_sync(struct vnode *vp);
  72
  73 int ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
  74
  75 SYSCTL_DECL(_vfs_ffs);
  76
  77 static int ffsrawbufcnt = 4;
  78 SYSCTL_INT(_vfs_ffs, OID_AUTO, ffsrawbufcnt, CTLFLAG_RD, &ffsrawbufcnt, 0,
  79            "Buffers available for raw reads");
  80
  81 static int allowrawread = 1;
  82 SYSCTL_INT(_vfs_ffs, OID_AUTO, allowrawread, CTLFLAG_RW, &allowrawread, 0,
  83            "Flag to enable raw reads");
  84
  85 static int rawreadahead = 1;
  86 SYSCTL_INT(_vfs_ffs, OID_AUTO, rawreadahead, CTLFLAG_RW, &rawreadahead, 0,
  87            "Flag to enable readahead for long raw reads");
  88
  89 static void
  90 ffs_rawread_setup(void *arg __unused)
  91 {
  92
  93         ffsrawbufcnt = (nswbuf > 100 ) ? (nswbuf - (nswbuf >> 4)) : nswbuf - 8;
  94 }
  95 SYSINIT(ffs_raw, SI_SUB_VM_CONF, SI_ORDER_ANY, ffs_rawread_setup, NULL);
  96
  97 static int
  98 ffs_rawread_sync(struct vnode *vp)
  99 {
 100         int error;
 101         int upgraded;
 102         struct bufobj *bo;
 103         struct mount *mp;
 104         vm_object_t obj;
 105
 106         /* Check for dirty mmap, pending writes and dirty buffers */
 107         bo = &vp->v_bufobj;
 108         BO_LOCK(bo);
 109         VI_LOCK(vp);
 110         if (bo->bo_numoutput > 0 ||
 111             bo->bo_dirty.bv_cnt > 0 ||
 112             ((obj = vp->v_object) != NULL &&
 113              (obj->flags & OBJ_MIGHTBEDIRTY) != 0)) {
 114                 VI_UNLOCK(vp);
 115                 BO_UNLOCK(bo);
 116
 117                 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
 118                         if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
 119                                 upgraded = 1;
 120                         else
 121                                 upgraded = 0;
 122                         VOP_UNLOCK(vp, 0);
 123                         (void) vn_start_write(vp, &mp, V_WAIT);
 124                         VOP_LOCK(vp, LK_EXCLUSIVE);
 125                 } else if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
 126                         upgraded = 1;
 127                         /* Upgrade to exclusive lock, this might block */
 128                         VOP_LOCK(vp, LK_UPGRADE);
 129                 } else
 130                         upgraded = 0;
 131
 132
 133                 VI_LOCK(vp);
 134                 /* Check if vnode was reclaimed while unlocked. */
 135                 if ((vp->v_iflag & VI_DOOMED) != 0) {
 136                         VI_UNLOCK(vp);
 137                         if (upgraded != 0)
 138                                 VOP_LOCK(vp, LK_DOWNGRADE);
 139                         vn_finished_write(mp);
 140                         return (EIO);
 141                 }
 142                 /* Attempt to msync mmap() regions to clean dirty mmap */
 143                 if ((obj = vp->v_object) != NULL &&
 144                     (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
 145                         VI_UNLOCK(vp);
 146                         VM_OBJECT_WLOCK(obj);
 147                         vm_object_page_clean(obj, 0, 0, OBJPC_SYNC);
 148                         VM_OBJECT_WUNLOCK(obj);
 149                 } else
 150                         VI_UNLOCK(vp);
 151
 152                 /* Wait for pending writes to complete */
 153                 BO_LOCK(bo);
 154                 error = bufobj_wwait(&vp->v_bufobj, 0, 0);
 155                 if (error != 0) {
 156                         /* XXX: can't happen with a zero timeout ??? */
 157                         BO_UNLOCK(bo);
 158                         if (upgraded != 0)
 159                                 VOP_LOCK(vp, LK_DOWNGRADE);
 160                         vn_finished_write(mp);
 161                         return (error);
 162                 }
 163                 /* Flush dirty buffers */
 164                 if (bo->bo_dirty.bv_cnt > 0) {
 165                         BO_UNLOCK(bo);
 166                         if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0) {
 167                                 if (upgraded != 0)
 168                                         VOP_LOCK(vp, LK_DOWNGRADE);
 169                                 vn_finished_write(mp);
 170                                 return (error);
 171                         }
 172                         BO_LOCK(bo);
 173                         if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
 174                                 panic("ffs_rawread_sync: dirty bufs");
 175                 }
 176                 BO_UNLOCK(bo);
 177                 if (upgraded != 0)
 178                         VOP_LOCK(vp, LK_DOWNGRADE);
 179                 vn_finished_write(mp);
 180         } else {
 181                 VI_UNLOCK(vp);
 182                 BO_UNLOCK(bo);
 183         }
 184         return 0;
 185 }
 186
 187
 188 static int
 189 ffs_rawread_readahead(struct vnode *vp,
 190                       caddr_t udata,
 191                       off_t offset,
 192                       size_t len,
 193                       struct thread *td,
 194                       struct buf *bp)
 195 {
 196         int error;
 197         u_int iolen;
 198         off_t blockno;
 199         int blockoff;
 200         int bsize;
 201         struct vnode *dp;
 202         int bforwards;
 203         struct inode *ip;
 204         ufs2_daddr_t blkno;
 205
 206         bsize = vp->v_mount->mnt_stat.f_iosize;
 207
 208         ip = VTOI(vp);
 209         dp = ITODEVVP(ip);
 210
 211         iolen = ((vm_offset_t) udata) & PAGE_MASK;
 212         bp->b_bcount = len;
 213         if (bp->b_bcount + iolen > bp->b_kvasize) {
 214                 bp->b_bcount = bp->b_kvasize;
 215                 if (iolen != 0)
 216                         bp->b_bcount -= PAGE_SIZE;
 217         }
 218         bp->b_flags = 0;        /* XXX necessary ? */
 219         bp->b_iocmd = BIO_READ;
 220         bp->b_iodone = bdone;
 221         bp->b_data = udata;
 222         blockno = offset / bsize;
 223         blockoff = (offset % bsize) / DEV_BSIZE;
 224         if ((daddr_t) blockno != blockno) {
 225                 return EINVAL; /* blockno overflow */
 226         }
 227
 228         bp->b_lblkno = bp->b_blkno = blockno;
 229
 230         error = ufs_bmaparray(vp, bp->b_lblkno, &blkno, NULL, &bforwards, NULL);
 231         if (error != 0)
 232                 return error;
 233         if (blkno == -1) {
 234
 235                 /* Fill holes with NULs to preserve semantics */
 236
 237                 if (bp->b_bcount + blockoff * DEV_BSIZE > bsize)
 238                         bp->b_bcount = bsize - blockoff * DEV_BSIZE;
 239                 bp->b_bufsize = bp->b_bcount;
 240
 241                 if (vmapbuf(bp, 1) < 0)
 242                         return EFAULT;
 243
 244                 maybe_yield();
 245                 bzero(bp->b_data, bp->b_bufsize);
 246
 247                 /* Mark operation completed (similar to bufdone()) */
 248
 249                 bp->b_resid = 0;
 250                 bp->b_flags |= B_DONE;
 251                 return 0;
 252         }
 253         bp->b_blkno = blkno + blockoff;
 254         bp->b_offset = bp->b_iooffset = (blkno + blockoff) * DEV_BSIZE;
 255
 256         if (bp->b_bcount + blockoff * DEV_BSIZE > bsize * (1 + bforwards))
 257                 bp->b_bcount = bsize * (1 + bforwards) - blockoff * DEV_BSIZE;
 258         bp->b_bufsize = bp->b_bcount;
 259
 260         if (vmapbuf(bp, 1) < 0)
 261                 return EFAULT;
 262
 263         BO_STRATEGY(&dp->v_bufobj, bp);
 264         return 0;
 265 }
 266
 267
 268 static int
 269 ffs_rawread_main(struct vnode *vp,
 270                  struct uio *uio)
 271 {
 272         int error, nerror;
 273         struct buf *bp, *nbp, *tbp;
 274         u_int iolen;
 275         caddr_t udata;
 276         long resid;
 277         off_t offset;
 278         struct thread *td;
 279
 280         td = uio->uio_td ? uio->uio_td : curthread;
 281         udata = uio->uio_iov->iov_base;
 282         resid = uio->uio_resid;
 283         offset = uio->uio_offset;
 284
 285         /*
 286          * keep the process from being swapped
 287          */
 288         PHOLD(td->td_proc);
 289
 290         error = 0;
 291         nerror = 0;
 292
 293         bp = NULL;
 294         nbp = NULL;
 295
 296         while (resid > 0) {
 297
 298                 if (bp == NULL) { /* Setup first read */
 299                         /* XXX: Leave some bufs for swap */
 300                         bp = getpbuf(&ffsrawbufcnt);
 301                         pbgetvp(vp, bp);
 302                         error = ffs_rawread_readahead(vp, udata, offset,
 303                                                      resid, td, bp);
 304                         if (error != 0)
 305                                 break;
 306
 307                         if (resid > bp->b_bufsize) { /* Setup fist readahead */
 308                                 /* XXX: Leave bufs for swap */
 309                                 if (rawreadahead != 0)
 310                                         nbp = trypbuf(&ffsrawbufcnt);
 311                                 else
 312                                         nbp = NULL;
 313                                 if (nbp != NULL) {
 314                                         pbgetvp(vp, nbp);
 315
 316                                         nerror = ffs_rawread_readahead(vp,
 317                                                                        udata +
 318                                                                        bp->b_bufsize,
 319                                                                        offset +
 320                                                                        bp->b_bufsize,
 321                                                                        resid -
 322                                                                        bp->b_bufsize,
 323                                                                        td,
 324                                                                        nbp);
 325                                         if (nerror) {
 326                                                 pbrelvp(nbp);
 327                                                 relpbuf(nbp, &ffsrawbufcnt);
 328                                                 nbp = NULL;
 329                                         }
 330                                 }
 331                         }
 332                 }
 333
 334                 bwait(bp, PRIBIO, "rawrd");
 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                         if (error != 0)
 359                                 break;
 360                 } else if (nbp != NULL) { /* Complete read with readahead */
 361
 362                         tbp = bp;
 363                         bp = nbp;
 364                         nbp = tbp;
 365
 366                         if (resid <= bp->b_bufsize) { /* No more readaheads */
 367                                 pbrelvp(nbp);
 368                                 relpbuf(nbp, &ffsrawbufcnt);
 369                                 nbp = NULL;
 370                         } else { /* Setup next readahead */
 371                                 nerror = ffs_rawread_readahead(vp,
 372                                                                udata +
 373                                                                bp->b_bufsize,
 374                                                                offset +
 375                                                                bp->b_bufsize,
 376                                                                resid -
 377                                                                bp->b_bufsize,
 378                                                                td,
 379                                                                nbp);
 380                                 if (nerror != 0) {
 381                                         pbrelvp(nbp);
 382                                         relpbuf(nbp, &ffsrawbufcnt);
 383                                         nbp = NULL;
 384                                 }
 385                         }
 386                 } else if (nerror != 0) {/* Deferred Readahead error */
 387                         break;
 388                 }  else if (resid > 0) { /* More to read, no readahead */
 389                         error = ffs_rawread_readahead(vp, udata, offset,
 390                                                       resid, td, bp);
 391                         if (error != 0)
 392                                 break;
 393                 }
 394         }
 395
 396         if (bp != NULL) {
 397                 pbrelvp(bp);
 398                 relpbuf(bp, &ffsrawbufcnt);
 399         }
 400         if (nbp != NULL) {                      /* Run down readahead buffer */
 401                 bwait(nbp, PRIBIO, "rawrd");
 402                 vunmapbuf(nbp);
 403                 pbrelvp(nbp);
 404                 relpbuf(nbp, &ffsrawbufcnt);
 405         }
 406
 407         if (error == 0)
 408                 error = nerror;
 409         PRELE(td->td_proc);
 410         uio->uio_iov->iov_base = udata;
 411         uio->uio_resid = resid;
 412         uio->uio_offset = offset;
 413         return error;
 414 }
 415
 416
 417 int
 418 ffs_rawread(struct vnode *vp,
 419             struct uio *uio,
 420             int *workdone)
 421 {
 422         if (allowrawread != 0 &&
 423             uio->uio_iovcnt == 1 &&
 424             uio->uio_segflg == UIO_USERSPACE &&
 425             uio->uio_resid == uio->uio_iov->iov_len &&
 426             (((uio->uio_td != NULL) ? uio->uio_td : curthread)->td_pflags &
 427              TDP_DEADLKTREAT) == 0) {
 428                 int secsize;            /* Media sector size */
 429                 off_t filebytes;        /* Bytes left of file */
 430                 int blockbytes;         /* Bytes left of file in full blocks */
 431                 int partialbytes;       /* Bytes in last partial block */
 432                 int skipbytes;          /* Bytes not to read in ffs_rawread */
 433                 struct inode *ip;
 434                 int error;
 435
 436
 437                 /* Only handle sector aligned reads */
 438                 ip = VTOI(vp);
 439                 secsize = ITODEVVP(ip)->v_bufobj.bo_bsize;
 440                 if ((uio->uio_offset & (secsize - 1)) == 0 &&
 441                     (uio->uio_resid & (secsize - 1)) == 0) {
 442
 443                         /* Sync dirty pages and buffers if needed */
 444                         error = ffs_rawread_sync(vp);
 445                         if (error != 0)
 446                                 return error;
 447
 448                         /* Check for end of file */
 449                         if (ip->i_size > uio->uio_offset) {
 450                                 filebytes = ip->i_size - uio->uio_offset;
 451
 452                                 /* No special eof handling needed ? */
 453                                 if (uio->uio_resid <= filebytes) {
 454                                         *workdone = 1;
 455                                         return ffs_rawread_main(vp, uio);
 456                                 }
 457
 458                                 partialbytes = ((unsigned int) ip->i_size) %
 459                                     ITOFS(ip)->fs_bsize;
 460                                 blockbytes = (int) filebytes - partialbytes;
 461                                 if (blockbytes > 0) {
 462                                         skipbytes = uio->uio_resid -
 463                                                 blockbytes;
 464                                         uio->uio_resid = blockbytes;
 465                                         error = ffs_rawread_main(vp, uio);
 466                                         uio->uio_resid += skipbytes;
 467                                         if (error != 0)
 468                                                 return error;
 469                                         /* Read remaining part using buffer */
 470                                 }
 471                         }
 472                 }
 473         }
 474         *workdone = 0;
 475         return 0;
 476 }