sys/ufs/ffs/ffs_rawread.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-2-Clause
   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 #include <sys/param.h>
  31 #include <sys/systm.h>
  32 #include <sys/fcntl.h>
  33 #include <sys/file.h>
  34 #include <sys/stat.h>
  35 #include <sys/proc.h>
  36 #include <sys/limits.h>
  37 #include <sys/mount.h>
  38 #include <sys/namei.h>
  39 #include <sys/vnode.h>
  40 #include <sys/conf.h>
  41 #include <sys/filio.h>
  42 #include <sys/ttycom.h>
  43 #include <sys/bio.h>
  44 #include <sys/buf.h>
  45 #include <sys/rwlock.h>
  46 #include <ufs/ufs/extattr.h>
  47 #include <ufs/ufs/quota.h>
  48 #include <ufs/ufs/inode.h>
  49 #include <ufs/ufs/ufsmount.h>
  50 #include <ufs/ufs/ufs_extern.h>
  51 #include <ufs/ffs/fs.h>
  52 #include <ufs/ffs/ffs_extern.h>
  53
  54 #include <vm/vm.h>
  55 #include <vm/vm_extern.h>
  56 #include <vm/vm_object.h>
  57 #include <sys/kernel.h>
  58 #include <sys/sysctl.h>
  59
  60 static int ffs_rawread_readahead(struct vnode *vp,
  61                                  caddr_t udata,
  62                                  off_t offset,
  63                                  size_t len,
  64                                  struct thread *td,
  65                                  struct buf *bp);
  66 static int ffs_rawread_main(struct vnode *vp,
  67                             struct uio *uio);
  68
  69 static int ffs_rawread_sync(struct vnode *vp);
  70
  71 int ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
  72
  73 SYSCTL_DECL(_vfs_ffs);
  74
  75 static uma_zone_t ffsraw_pbuf_zone;
  76
  77 static int allowrawread = 1;
  78 SYSCTL_INT(_vfs_ffs, OID_AUTO, allowrawread, CTLFLAG_RW, &allowrawread, 0,
  79            "Flag to enable raw reads");
  80
  81 static int rawreadahead = 1;
  82 SYSCTL_INT(_vfs_ffs, OID_AUTO, rawreadahead, CTLFLAG_RW, &rawreadahead, 0,
  83            "Flag to enable readahead for long raw reads");
  84
  85 static void
  86 ffs_rawread_setup(void *arg __unused)
  87 {
  88
  89         ffsraw_pbuf_zone = pbuf_zsecond_create("ffsrawpbuf",
  90             (nswbuf > 100 ) ?  (nswbuf - (nswbuf >> 4)) : nswbuf - 8);
  91 }
  92 SYSINIT(ffs_raw, SI_SUB_VM_CONF, SI_ORDER_ANY, ffs_rawread_setup, NULL);
  93
  94 static int
  95 ffs_rawread_sync(struct vnode *vp)
  96 {
  97         int error;
  98         int upgraded;
  99         struct bufobj *bo;
 100         struct mount *mp;
 101         vm_object_t obj;
 102
 103         /* Check for dirty mmap, pending writes and dirty buffers */
 104         bo = &vp->v_bufobj;
 105         BO_LOCK(bo);
 106         VI_LOCK(vp);
 107         if (bo->bo_numoutput > 0 ||
 108             bo->bo_dirty.bv_cnt > 0 ||
 109             ((obj = vp->v_object) != NULL &&
 110              vm_object_mightbedirty(obj))) {
 111                 VI_UNLOCK(vp);
 112                 BO_UNLOCK(bo);
 113
 114                 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
 115                         if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
 116                                 upgraded = 1;
 117                         else
 118                                 upgraded = 0;
 119                         VOP_UNLOCK(vp);
 120                         (void) vn_start_write(vp, &mp, V_WAIT);
 121                         VOP_LOCK(vp, LK_EXCLUSIVE);
 122                 } else if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
 123                         upgraded = 1;
 124                         /* Upgrade to exclusive lock, this might block */
 125                         VOP_LOCK(vp, LK_UPGRADE);
 126                 } else
 127                         upgraded = 0;
 128
 129
 130                 VI_LOCK(vp);
 131                 /* Check if vnode was reclaimed while unlocked. */
 132                 if (VN_IS_DOOMED(vp)) {
 133                         VI_UNLOCK(vp);
 134                         if (upgraded != 0)
 135                                 VOP_LOCK(vp, LK_DOWNGRADE);
 136                         vn_finished_write(mp);
 137                         return (EIO);
 138                 }
 139                 /* Attempt to msync mmap() regions to clean dirty mmap */
 140                 if ((obj = vp->v_object) != NULL &&
 141                     vm_object_mightbedirty(obj)) {
 142                         VI_UNLOCK(vp);
 143                         VM_OBJECT_WLOCK(obj);
 144                         vm_object_page_clean(obj, 0, 0, OBJPC_SYNC);
 145                         VM_OBJECT_WUNLOCK(obj);
 146                 } else
 147                         VI_UNLOCK(vp);
 148
 149                 /* Wait for pending writes to complete */
 150                 BO_LOCK(bo);
 151                 error = bufobj_wwait(&vp->v_bufobj, 0, 0);
 152                 if (error != 0) {
 153                         /* XXX: can't happen with a zero timeout ??? */
 154                         BO_UNLOCK(bo);
 155                         if (upgraded != 0)
 156                                 VOP_LOCK(vp, LK_DOWNGRADE);
 157                         vn_finished_write(mp);
 158                         return (error);
 159                 }
 160                 /* Flush dirty buffers */
 161                 if (bo->bo_dirty.bv_cnt > 0) {
 162                         BO_UNLOCK(bo);
 163                         if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0) {
 164                                 if (upgraded != 0)
 165                                         VOP_LOCK(vp, LK_DOWNGRADE);
 166                                 vn_finished_write(mp);
 167                                 return (error);
 168                         }
 169                         BO_LOCK(bo);
 170                         if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
 171                                 panic("ffs_rawread_sync: dirty bufs");
 172                 }
 173                 BO_UNLOCK(bo);
 174                 if (upgraded != 0)
 175                         VOP_LOCK(vp, LK_DOWNGRADE);
 176                 vn_finished_write(mp);
 177         } else {
 178                 VI_UNLOCK(vp);
 179                 BO_UNLOCK(bo);
 180         }
 181         return 0;
 182 }
 183
 184 static int
 185 ffs_rawread_readahead(struct vnode *vp,
 186                       caddr_t udata,
 187                       off_t offset,
 188                       size_t len,
 189                       struct thread *td,
 190                       struct buf *bp)
 191 {
 192         int error;
 193         uint64_t iolen;
 194         off_t blockno;
 195         int blockoff;
 196         int bsize;
 197         struct vnode *dp;
 198         int bforwards;
 199         struct inode *ip;
 200         ufs2_daddr_t blkno;
 201
 202         bsize = vp->v_mount->mnt_stat.f_iosize;
 203
 204         ip = VTOI(vp);
 205         dp = ITODEVVP(ip);
 206
 207         iolen = ((vm_offset_t) udata) & PAGE_MASK;
 208         bp->b_bcount = len;
 209         if (bp->b_bcount + iolen > bp->b_kvasize) {
 210                 bp->b_bcount = bp->b_kvasize;
 211                 if (iolen != 0)
 212                         bp->b_bcount -= PAGE_SIZE;
 213         }
 214         bp->b_flags = 0;        /* XXX necessary ? */
 215         bp->b_iocmd = BIO_READ;
 216         bp->b_iodone = bdone;
 217         blockno = offset / bsize;
 218         blockoff = (offset % bsize) / DEV_BSIZE;
 219         if ((daddr_t) blockno != blockno) {
 220                 return EINVAL; /* blockno overflow */
 221         }
 222
 223         bp->b_lblkno = bp->b_blkno = blockno;
 224
 225         error = ufs_bmaparray(vp, bp->b_lblkno, &blkno, NULL, &bforwards, NULL);
 226         if (error != 0)
 227                 return error;
 228         if (blkno == -1) {
 229                 /* Fill holes with NULs to preserve semantics */
 230
 231                 if (bp->b_bcount + blockoff * DEV_BSIZE > bsize)
 232                         bp->b_bcount = bsize - blockoff * DEV_BSIZE;
 233
 234                 if (vmapbuf(bp, udata, bp->b_bcount, 1) < 0)
 235                         return EFAULT;
 236
 237                 maybe_yield();
 238                 bzero(bp->b_data, bp->b_bufsize);
 239
 240                 /* Mark operation completed (similar to bufdone()) */
 241
 242                 bp->b_resid = 0;
 243                 bp->b_flags |= B_DONE;
 244                 return 0;
 245         }
 246         bp->b_blkno = blkno + blockoff;
 247         bp->b_offset = bp->b_iooffset = (blkno + blockoff) * DEV_BSIZE;
 248
 249         if (bp->b_bcount + blockoff * DEV_BSIZE > bsize * (1 + bforwards))
 250                 bp->b_bcount = bsize * (1 + bforwards) - blockoff * DEV_BSIZE;
 251
 252         if (vmapbuf(bp, udata, bp->b_bcount, 1) < 0)
 253                 return EFAULT;
 254
 255         BO_STRATEGY(&dp->v_bufobj, bp);
 256         return 0;
 257 }
 258
 259 static int
 260 ffs_rawread_main(struct vnode *vp,
 261                  struct uio *uio)
 262 {
 263         int error, nerror;
 264         struct buf *bp, *nbp, *tbp;
 265         uint64_t iolen;
 266         caddr_t udata;
 267         long resid;
 268         off_t offset;
 269         struct thread *td;
 270
 271         td = uio->uio_td ? uio->uio_td : curthread;
 272         udata = uio->uio_iov->iov_base;
 273         resid = uio->uio_resid;
 274         offset = uio->uio_offset;
 275
 276         /*
 277          * keep the process from being swapped
 278          */
 279         PHOLD(td->td_proc);
 280
 281         error = 0;
 282         nerror = 0;
 283
 284         bp = NULL;
 285         nbp = NULL;
 286
 287         while (resid > 0) {
 288
 289                 if (bp == NULL) { /* Setup first read */
 290                         bp = uma_zalloc(ffsraw_pbuf_zone, M_WAITOK);
 291                         pbgetvp(vp, bp);
 292                         error = ffs_rawread_readahead(vp, udata, offset,
 293                                                      resid, td, bp);
 294                         if (error != 0)
 295                                 break;
 296
 297                         if (resid > bp->b_bufsize) { /* Setup fist readahead */
 298                                 if (rawreadahead != 0)
 299                                         nbp = uma_zalloc(ffsraw_pbuf_zone,
 300                                             M_NOWAIT);
 301                                 else
 302                                         nbp = NULL;
 303                                 if (nbp != NULL) {
 304                                         pbgetvp(vp, nbp);
 305
 306                                         nerror = ffs_rawread_readahead(vp,
 307                                                                        udata +
 308                                                                        bp->b_bufsize,
 309                                                                        offset +
 310                                                                        bp->b_bufsize,
 311                                                                        resid -
 312                                                                        bp->b_bufsize,
 313                                                                        td,
 314                                                                        nbp);
 315                                         if (nerror) {
 316                                                 pbrelvp(nbp);
 317                                                 uma_zfree(ffsraw_pbuf_zone,
 318                                                     nbp);
 319                                                 nbp = NULL;
 320                                         }
 321                                 }
 322                         }
 323                 }
 324
 325                 bwait(bp, PRIBIO, "rawrd");
 326                 vunmapbuf(bp);
 327
 328                 iolen = bp->b_bcount - bp->b_resid;
 329                 if (iolen == 0 && (bp->b_ioflags & BIO_ERROR) == 0) {
 330                         nerror = 0;     /* Ignore possible beyond EOF error */
 331                         break; /* EOF */
 332                 }
 333
 334                 if ((bp->b_ioflags & BIO_ERROR) != 0) {
 335                         error = bp->b_error;
 336                         break;
 337                 }
 338                 resid -= iolen;
 339                 udata += iolen;
 340                 offset += iolen;
 341                 if (iolen < bp->b_bufsize) {
 342                         /* Incomplete read.  Try to read remaining part */
 343                         error = ffs_rawread_readahead(vp,
 344                                                       udata,
 345                                                       offset,
 346                                                       bp->b_bufsize - iolen,
 347                                                       td,
 348                                                       bp);
 349                         if (error != 0)
 350                                 break;
 351                 } else if (nbp != NULL) { /* Complete read with readahead */
 352
 353                         tbp = bp;
 354                         bp = nbp;
 355                         nbp = tbp;
 356
 357                         if (resid <= bp->b_bufsize) { /* No more readaheads */
 358                                 pbrelvp(nbp);
 359                                 uma_zfree(ffsraw_pbuf_zone, nbp);
 360                                 nbp = NULL;
 361                         } else { /* Setup next readahead */
 362                                 nerror = ffs_rawread_readahead(vp,
 363                                                                udata +
 364                                                                bp->b_bufsize,
 365                                                                offset +
 366                                                                bp->b_bufsize,
 367                                                                resid -
 368                                                                bp->b_bufsize,
 369                                                                td,
 370                                                                nbp);
 371                                 if (nerror != 0) {
 372                                         pbrelvp(nbp);
 373                                         uma_zfree(ffsraw_pbuf_zone, nbp);
 374                                         nbp = NULL;
 375                                 }
 376                         }
 377                 } else if (nerror != 0) {/* Deferred Readahead error */
 378                         break;
 379                 }  else if (resid > 0) { /* More to read, no readahead */
 380                         error = ffs_rawread_readahead(vp, udata, offset,
 381                                                       resid, td, bp);
 382                         if (error != 0)
 383                                 break;
 384                 }
 385         }
 386
 387         if (bp != NULL) {
 388                 pbrelvp(bp);
 389                 uma_zfree(ffsraw_pbuf_zone, bp);
 390         }
 391         if (nbp != NULL) {                      /* Run down readahead buffer */
 392                 bwait(nbp, PRIBIO, "rawrd");
 393                 vunmapbuf(nbp);
 394                 pbrelvp(nbp);
 395                 uma_zfree(ffsraw_pbuf_zone, nbp);
 396         }
 397
 398         if (error == 0)
 399                 error = nerror;
 400         PRELE(td->td_proc);
 401         uio->uio_iov->iov_base = udata;
 402         uio->uio_resid = resid;
 403         uio->uio_offset = offset;
 404         return error;
 405 }
 406
 407 int
 408 ffs_rawread(struct vnode *vp,
 409             struct uio *uio,
 410             int *workdone)
 411 {
 412         if (allowrawread != 0 &&
 413             uio->uio_iovcnt == 1 &&
 414             uio->uio_segflg == UIO_USERSPACE &&
 415             uio->uio_resid == uio->uio_iov->iov_len &&
 416             (((uio->uio_td != NULL) ? uio->uio_td : curthread)->td_pflags &
 417              TDP_DEADLKTREAT) == 0) {
 418                 int secsize;            /* Media sector size */
 419                 off_t filebytes;        /* Bytes left of file */
 420                 int blockbytes;         /* Bytes left of file in full blocks */
 421                 int partialbytes;       /* Bytes in last partial block */
 422                 int skipbytes;          /* Bytes not to read in ffs_rawread */
 423                 struct inode *ip;
 424                 int error;
 425
 426
 427                 /* Only handle sector aligned reads */
 428                 ip = VTOI(vp);
 429                 secsize = ITODEVVP(ip)->v_bufobj.bo_bsize;
 430                 if ((uio->uio_offset & (secsize - 1)) == 0 &&
 431                     (uio->uio_resid & (secsize - 1)) == 0) {
 432
 433                         /* Sync dirty pages and buffers if needed */
 434                         error = ffs_rawread_sync(vp);
 435                         if (error != 0)
 436                                 return error;
 437
 438                         /* Check for end of file */
 439                         if (ip->i_size > uio->uio_offset) {
 440                                 filebytes = ip->i_size - uio->uio_offset;
 441
 442                                 /* No special eof handling needed ? */
 443                                 if (uio->uio_resid <= filebytes) {
 444                                         *workdone = 1;
 445                                         return ffs_rawread_main(vp, uio);
 446                                 }
 447
 448                                 partialbytes = ((unsigned int) ip->i_size) %
 449                                     ITOFS(ip)->fs_bsize;
 450                                 blockbytes = (int) filebytes - partialbytes;
 451                                 if (blockbytes > 0) {
 452                                         skipbytes = uio->uio_resid -
 453                                                 blockbytes;
 454                                         uio->uio_resid = blockbytes;
 455                                         error = ffs_rawread_main(vp, uio);
 456                                         uio->uio_resid += skipbytes;
 457                                         if (error != 0)
 458                                                 return error;
 459                                         /* Read remaining part using buffer */
 460                                 }
 461                         }
 462                 }
 463         }
 464         *workdone = 0;
 465         return 0;
 466 }