linux/fs/xfs/xfs_inode.c
<<
>>
Prefs
   1/*
   2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   3 * All Rights Reserved.
   4 *
   5 * This program is free software; you can redistribute it and/or
   6 * modify it under the terms of the GNU General Public License as
   7 * published by the Free Software Foundation.
   8 *
   9 * This program is distributed in the hope that it would be useful,
  10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12 * GNU General Public License for more details.
  13 *
  14 * You should have received a copy of the GNU General Public License
  15 * along with this program; if not, write the Free Software Foundation,
  16 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17 */
  18#include <linux/log2.h>
  19
  20#include "xfs.h"
  21#include "xfs_fs.h"
  22#include "xfs_types.h"
  23#include "xfs_log.h"
  24#include "xfs_inum.h"
  25#include "xfs_trans.h"
  26#include "xfs_trans_priv.h"
  27#include "xfs_sb.h"
  28#include "xfs_ag.h"
  29#include "xfs_mount.h"
  30#include "xfs_bmap_btree.h"
  31#include "xfs_alloc_btree.h"
  32#include "xfs_ialloc_btree.h"
  33#include "xfs_attr_sf.h"
  34#include "xfs_dinode.h"
  35#include "xfs_inode.h"
  36#include "xfs_buf_item.h"
  37#include "xfs_inode_item.h"
  38#include "xfs_btree.h"
  39#include "xfs_alloc.h"
  40#include "xfs_ialloc.h"
  41#include "xfs_bmap.h"
  42#include "xfs_error.h"
  43#include "xfs_utils.h"
  44#include "xfs_quota.h"
  45#include "xfs_filestream.h"
  46#include "xfs_vnodeops.h"
  47#include "xfs_trace.h"
  48
  49kmem_zone_t *xfs_ifork_zone;
  50kmem_zone_t *xfs_inode_zone;
  51
  52/*
  53 * Used in xfs_itruncate_extents().  This is the maximum number of extents
  54 * freed from a file in a single transaction.
  55 */
  56#define XFS_ITRUNC_MAX_EXTENTS  2
  57
  58STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
  59STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
  60STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
  61STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
  62
  63/*
  64 * helper function to extract extent size hint from inode
  65 */
  66xfs_extlen_t
  67xfs_get_extsz_hint(
  68        struct xfs_inode        *ip)
  69{
  70        if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
  71                return ip->i_d.di_extsize;
  72        if (XFS_IS_REALTIME_INODE(ip))
  73                return ip->i_mount->m_sb.sb_rextsize;
  74        return 0;
  75}
  76
  77#ifdef DEBUG
  78/*
  79 * Make sure that the extents in the given memory buffer
  80 * are valid.
  81 */
  82STATIC void
  83xfs_validate_extents(
  84        xfs_ifork_t             *ifp,
  85        int                     nrecs,
  86        xfs_exntfmt_t           fmt)
  87{
  88        xfs_bmbt_irec_t         irec;
  89        xfs_bmbt_rec_host_t     rec;
  90        int                     i;
  91
  92        for (i = 0; i < nrecs; i++) {
  93                xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
  94                rec.l0 = get_unaligned(&ep->l0);
  95                rec.l1 = get_unaligned(&ep->l1);
  96                xfs_bmbt_get_all(&rec, &irec);
  97                if (fmt == XFS_EXTFMT_NOSTATE)
  98                        ASSERT(irec.br_state == XFS_EXT_NORM);
  99        }
 100}
 101#else /* DEBUG */
 102#define xfs_validate_extents(ifp, nrecs, fmt)
 103#endif /* DEBUG */
 104
 105/*
 106 * Check that none of the inode's in the buffer have a next
 107 * unlinked field of 0.
 108 */
 109#if defined(DEBUG)
 110void
 111xfs_inobp_check(
 112        xfs_mount_t     *mp,
 113        xfs_buf_t       *bp)
 114{
 115        int             i;
 116        int             j;
 117        xfs_dinode_t    *dip;
 118
 119        j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
 120
 121        for (i = 0; i < j; i++) {
 122                dip = (xfs_dinode_t *)xfs_buf_offset(bp,
 123                                        i * mp->m_sb.sb_inodesize);
 124                if (!dip->di_next_unlinked)  {
 125                        xfs_alert(mp,
 126        "Detected bogus zero next_unlinked field in incore inode buffer 0x%p.",
 127                                bp);
 128                        ASSERT(dip->di_next_unlinked);
 129                }
 130        }
 131}
 132#endif
 133
 134/*
 135 * This routine is called to map an inode to the buffer containing the on-disk
 136 * version of the inode.  It returns a pointer to the buffer containing the
 137 * on-disk inode in the bpp parameter, and in the dipp parameter it returns a
 138 * pointer to the on-disk inode within that buffer.
 139 *
 140 * If a non-zero error is returned, then the contents of bpp and dipp are
 141 * undefined.
 142 */
 143int
 144xfs_imap_to_bp(
 145        struct xfs_mount        *mp,
 146        struct xfs_trans        *tp,
 147        struct xfs_imap         *imap,
 148        struct xfs_dinode       **dipp,
 149        struct xfs_buf          **bpp,
 150        uint                    buf_flags,
 151        uint                    iget_flags)
 152{
 153        struct xfs_buf          *bp;
 154        int                     error;
 155        int                     i;
 156        int                     ni;
 157
 158        buf_flags |= XBF_UNMAPPED;
 159        error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
 160                                   (int)imap->im_len, buf_flags, &bp);
 161        if (error) {
 162                if (error != EAGAIN) {
 163                        xfs_warn(mp,
 164                                "%s: xfs_trans_read_buf() returned error %d.",
 165                                __func__, error);
 166                } else {
 167                        ASSERT(buf_flags & XBF_TRYLOCK);
 168                }
 169                return error;
 170        }
 171
 172        /*
 173         * Validate the magic number and version of every inode in the buffer
 174         * (if DEBUG kernel) or the first inode in the buffer, otherwise.
 175         */
 176#ifdef DEBUG
 177        ni = BBTOB(imap->im_len) >> mp->m_sb.sb_inodelog;
 178#else   /* usual case */
 179        ni = 1;
 180#endif
 181
 182        for (i = 0; i < ni; i++) {
 183                int             di_ok;
 184                xfs_dinode_t    *dip;
 185
 186                dip = (xfs_dinode_t *)xfs_buf_offset(bp,
 187                                        (i << mp->m_sb.sb_inodelog));
 188                di_ok = dip->di_magic == cpu_to_be16(XFS_DINODE_MAGIC) &&
 189                            XFS_DINODE_GOOD_VERSION(dip->di_version);
 190                if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
 191                                                XFS_ERRTAG_ITOBP_INOTOBP,
 192                                                XFS_RANDOM_ITOBP_INOTOBP))) {
 193                        if (iget_flags & XFS_IGET_UNTRUSTED) {
 194                                xfs_trans_brelse(tp, bp);
 195                                return XFS_ERROR(EINVAL);
 196                        }
 197                        XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_HIGH,
 198                                             mp, dip);
 199#ifdef DEBUG
 200                        xfs_emerg(mp,
 201                                "bad inode magic/vsn daddr %lld #%d (magic=%x)",
 202                                (unsigned long long)imap->im_blkno, i,
 203                                be16_to_cpu(dip->di_magic));
 204                        ASSERT(0);
 205#endif
 206                        xfs_trans_brelse(tp, bp);
 207                        return XFS_ERROR(EFSCORRUPTED);
 208                }
 209        }
 210
 211        xfs_inobp_check(mp, bp);
 212
 213        *bpp = bp;
 214        *dipp = (struct xfs_dinode *)xfs_buf_offset(bp, imap->im_boffset);
 215        return 0;
 216}
 217
 218/*
 219 * Move inode type and inode format specific information from the
 220 * on-disk inode to the in-core inode.  For fifos, devs, and sockets
 221 * this means set if_rdev to the proper value.  For files, directories,
 222 * and symlinks this means to bring in the in-line data or extent
 223 * pointers.  For a file in B-tree format, only the root is immediately
 224 * brought in-core.  The rest will be in-lined in if_extents when it
 225 * is first referenced (see xfs_iread_extents()).
 226 */
 227STATIC int
 228xfs_iformat(
 229        xfs_inode_t             *ip,
 230        xfs_dinode_t            *dip)
 231{
 232        xfs_attr_shortform_t    *atp;
 233        int                     size;
 234        int                     error = 0;
 235        xfs_fsize_t             di_size;
 236
 237        if (unlikely(be32_to_cpu(dip->di_nextents) +
 238                     be16_to_cpu(dip->di_anextents) >
 239                     be64_to_cpu(dip->di_nblocks))) {
 240                xfs_warn(ip->i_mount,
 241                        "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
 242                        (unsigned long long)ip->i_ino,
 243                        (int)(be32_to_cpu(dip->di_nextents) +
 244                              be16_to_cpu(dip->di_anextents)),
 245                        (unsigned long long)
 246                                be64_to_cpu(dip->di_nblocks));
 247                XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
 248                                     ip->i_mount, dip);
 249                return XFS_ERROR(EFSCORRUPTED);
 250        }
 251
 252        if (unlikely(dip->di_forkoff > ip->i_mount->m_sb.sb_inodesize)) {
 253                xfs_warn(ip->i_mount, "corrupt dinode %Lu, forkoff = 0x%x.",
 254                        (unsigned long long)ip->i_ino,
 255                        dip->di_forkoff);
 256                XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
 257                                     ip->i_mount, dip);
 258                return XFS_ERROR(EFSCORRUPTED);
 259        }
 260
 261        if (unlikely((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) &&
 262                     !ip->i_mount->m_rtdev_targp)) {
 263                xfs_warn(ip->i_mount,
 264                        "corrupt dinode %Lu, has realtime flag set.",
 265                        ip->i_ino);
 266                XFS_CORRUPTION_ERROR("xfs_iformat(realtime)",
 267                                     XFS_ERRLEVEL_LOW, ip->i_mount, dip);
 268                return XFS_ERROR(EFSCORRUPTED);
 269        }
 270
 271        switch (ip->i_d.di_mode & S_IFMT) {
 272        case S_IFIFO:
 273        case S_IFCHR:
 274        case S_IFBLK:
 275        case S_IFSOCK:
 276                if (unlikely(dip->di_format != XFS_DINODE_FMT_DEV)) {
 277                        XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
 278                                              ip->i_mount, dip);
 279                        return XFS_ERROR(EFSCORRUPTED);
 280                }
 281                ip->i_d.di_size = 0;
 282                ip->i_df.if_u2.if_rdev = xfs_dinode_get_rdev(dip);
 283                break;
 284
 285        case S_IFREG:
 286        case S_IFLNK:
 287        case S_IFDIR:
 288                switch (dip->di_format) {
 289                case XFS_DINODE_FMT_LOCAL:
 290                        /*
 291                         * no local regular files yet
 292                         */
 293                        if (unlikely(S_ISREG(be16_to_cpu(dip->di_mode)))) {
 294                                xfs_warn(ip->i_mount,
 295                        "corrupt inode %Lu (local format for regular file).",
 296                                        (unsigned long long) ip->i_ino);
 297                                XFS_CORRUPTION_ERROR("xfs_iformat(4)",
 298                                                     XFS_ERRLEVEL_LOW,
 299                                                     ip->i_mount, dip);
 300                                return XFS_ERROR(EFSCORRUPTED);
 301                        }
 302
 303                        di_size = be64_to_cpu(dip->di_size);
 304                        if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
 305                                xfs_warn(ip->i_mount,
 306                        "corrupt inode %Lu (bad size %Ld for local inode).",
 307                                        (unsigned long long) ip->i_ino,
 308                                        (long long) di_size);
 309                                XFS_CORRUPTION_ERROR("xfs_iformat(5)",
 310                                                     XFS_ERRLEVEL_LOW,
 311                                                     ip->i_mount, dip);
 312                                return XFS_ERROR(EFSCORRUPTED);
 313                        }
 314
 315                        size = (int)di_size;
 316                        error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
 317                        break;
 318                case XFS_DINODE_FMT_EXTENTS:
 319                        error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
 320                        break;
 321                case XFS_DINODE_FMT_BTREE:
 322                        error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
 323                        break;
 324                default:
 325                        XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
 326                                         ip->i_mount);
 327                        return XFS_ERROR(EFSCORRUPTED);
 328                }
 329                break;
 330
 331        default:
 332                XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
 333                return XFS_ERROR(EFSCORRUPTED);
 334        }
 335        if (error) {
 336                return error;
 337        }
 338        if (!XFS_DFORK_Q(dip))
 339                return 0;
 340
 341        ASSERT(ip->i_afp == NULL);
 342        ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP | KM_NOFS);
 343
 344        switch (dip->di_aformat) {
 345        case XFS_DINODE_FMT_LOCAL:
 346                atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
 347                size = be16_to_cpu(atp->hdr.totsize);
 348
 349                if (unlikely(size < sizeof(struct xfs_attr_sf_hdr))) {
 350                        xfs_warn(ip->i_mount,
 351                                "corrupt inode %Lu (bad attr fork size %Ld).",
 352                                (unsigned long long) ip->i_ino,
 353                                (long long) size);
 354                        XFS_CORRUPTION_ERROR("xfs_iformat(8)",
 355                                             XFS_ERRLEVEL_LOW,
 356                                             ip->i_mount, dip);
 357                        return XFS_ERROR(EFSCORRUPTED);
 358                }
 359
 360                error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
 361                break;
 362        case XFS_DINODE_FMT_EXTENTS:
 363                error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
 364                break;
 365        case XFS_DINODE_FMT_BTREE:
 366                error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
 367                break;
 368        default:
 369                error = XFS_ERROR(EFSCORRUPTED);
 370                break;
 371        }
 372        if (error) {
 373                kmem_zone_free(xfs_ifork_zone, ip->i_afp);
 374                ip->i_afp = NULL;
 375                xfs_idestroy_fork(ip, XFS_DATA_FORK);
 376        }
 377        return error;
 378}
 379
 380/*
 381 * The file is in-lined in the on-disk inode.
 382 * If it fits into if_inline_data, then copy
 383 * it there, otherwise allocate a buffer for it
 384 * and copy the data there.  Either way, set
 385 * if_data to point at the data.
 386 * If we allocate a buffer for the data, make
 387 * sure that its size is a multiple of 4 and
 388 * record the real size in i_real_bytes.
 389 */
 390STATIC int
 391xfs_iformat_local(
 392        xfs_inode_t     *ip,
 393        xfs_dinode_t    *dip,
 394        int             whichfork,
 395        int             size)
 396{
 397        xfs_ifork_t     *ifp;
 398        int             real_size;
 399
 400        /*
 401         * If the size is unreasonable, then something
 402         * is wrong and we just bail out rather than crash in
 403         * kmem_alloc() or memcpy() below.
 404         */
 405        if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
 406                xfs_warn(ip->i_mount,
 407        "corrupt inode %Lu (bad size %d for local fork, size = %d).",
 408                        (unsigned long long) ip->i_ino, size,
 409                        XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
 410                XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
 411                                     ip->i_mount, dip);
 412                return XFS_ERROR(EFSCORRUPTED);
 413        }
 414        ifp = XFS_IFORK_PTR(ip, whichfork);
 415        real_size = 0;
 416        if (size == 0)
 417                ifp->if_u1.if_data = NULL;
 418        else if (size <= sizeof(ifp->if_u2.if_inline_data))
 419                ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
 420        else {
 421                real_size = roundup(size, 4);
 422                ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP | KM_NOFS);
 423        }
 424        ifp->if_bytes = size;
 425        ifp->if_real_bytes = real_size;
 426        if (size)
 427                memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
 428        ifp->if_flags &= ~XFS_IFEXTENTS;
 429        ifp->if_flags |= XFS_IFINLINE;
 430        return 0;
 431}
 432
 433/*
 434 * The file consists of a set of extents all
 435 * of which fit into the on-disk inode.
 436 * If there are few enough extents to fit into
 437 * the if_inline_ext, then copy them there.
 438 * Otherwise allocate a buffer for them and copy
 439 * them into it.  Either way, set if_extents
 440 * to point at the extents.
 441 */
 442STATIC int
 443xfs_iformat_extents(
 444        xfs_inode_t     *ip,
 445        xfs_dinode_t    *dip,
 446        int             whichfork)
 447{
 448        xfs_bmbt_rec_t  *dp;
 449        xfs_ifork_t     *ifp;
 450        int             nex;
 451        int             size;
 452        int             i;
 453
 454        ifp = XFS_IFORK_PTR(ip, whichfork);
 455        nex = XFS_DFORK_NEXTENTS(dip, whichfork);
 456        size = nex * (uint)sizeof(xfs_bmbt_rec_t);
 457
 458        /*
 459         * If the number of extents is unreasonable, then something
 460         * is wrong and we just bail out rather than crash in
 461         * kmem_alloc() or memcpy() below.
 462         */
 463        if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
 464                xfs_warn(ip->i_mount, "corrupt inode %Lu ((a)extents = %d).",
 465                        (unsigned long long) ip->i_ino, nex);
 466                XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
 467                                     ip->i_mount, dip);
 468                return XFS_ERROR(EFSCORRUPTED);
 469        }
 470
 471        ifp->if_real_bytes = 0;
 472        if (nex == 0)
 473                ifp->if_u1.if_extents = NULL;
 474        else if (nex <= XFS_INLINE_EXTS)
 475                ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
 476        else
 477                xfs_iext_add(ifp, 0, nex);
 478
 479        ifp->if_bytes = size;
 480        if (size) {
 481                dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
 482                xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip));
 483                for (i = 0; i < nex; i++, dp++) {
 484                        xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
 485                        ep->l0 = get_unaligned_be64(&dp->l0);
 486                        ep->l1 = get_unaligned_be64(&dp->l1);
 487                }
 488                XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork);
 489                if (whichfork != XFS_DATA_FORK ||
 490                        XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
 491                                if (unlikely(xfs_check_nostate_extents(
 492                                    ifp, 0, nex))) {
 493                                        XFS_ERROR_REPORT("xfs_iformat_extents(2)",
 494                                                         XFS_ERRLEVEL_LOW,
 495                                                         ip->i_mount);
 496                                        return XFS_ERROR(EFSCORRUPTED);
 497                                }
 498        }
 499        ifp->if_flags |= XFS_IFEXTENTS;
 500        return 0;
 501}
 502
 503/*
 504 * The file has too many extents to fit into
 505 * the inode, so they are in B-tree format.
 506 * Allocate a buffer for the root of the B-tree
 507 * and copy the root into it.  The i_extents
 508 * field will remain NULL until all of the
 509 * extents are read in (when they are needed).
 510 */
 511STATIC int
 512xfs_iformat_btree(
 513        xfs_inode_t             *ip,
 514        xfs_dinode_t            *dip,
 515        int                     whichfork)
 516{
 517        xfs_bmdr_block_t        *dfp;
 518        xfs_ifork_t             *ifp;
 519        /* REFERENCED */
 520        int                     nrecs;
 521        int                     size;
 522
 523        ifp = XFS_IFORK_PTR(ip, whichfork);
 524        dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
 525        size = XFS_BMAP_BROOT_SPACE(dfp);
 526        nrecs = be16_to_cpu(dfp->bb_numrecs);
 527
 528        /*
 529         * blow out if -- fork has less extents than can fit in
 530         * fork (fork shouldn't be a btree format), root btree
 531         * block has more records than can fit into the fork,
 532         * or the number of extents is greater than the number of
 533         * blocks.
 534         */
 535        if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <=
 536                        XFS_IFORK_MAXEXT(ip, whichfork) ||
 537                     XFS_BMDR_SPACE_CALC(nrecs) >
 538                        XFS_DFORK_SIZE(dip, ip->i_mount, whichfork) ||
 539                     XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
 540                xfs_warn(ip->i_mount, "corrupt inode %Lu (btree).",
 541                        (unsigned long long) ip->i_ino);
 542                XFS_CORRUPTION_ERROR("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
 543                                 ip->i_mount, dip);
 544                return XFS_ERROR(EFSCORRUPTED);
 545        }
 546
 547        ifp->if_broot_bytes = size;
 548        ifp->if_broot = kmem_alloc(size, KM_SLEEP | KM_NOFS);
 549        ASSERT(ifp->if_broot != NULL);
 550        /*
 551         * Copy and convert from the on-disk structure
 552         * to the in-memory structure.
 553         */
 554        xfs_bmdr_to_bmbt(ip->i_mount, dfp,
 555                         XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
 556                         ifp->if_broot, size);
 557        ifp->if_flags &= ~XFS_IFEXTENTS;
 558        ifp->if_flags |= XFS_IFBROOT;
 559
 560        return 0;
 561}
 562
 563STATIC void
 564xfs_dinode_from_disk(
 565        xfs_icdinode_t          *to,
 566        xfs_dinode_t            *from)
 567{
 568        to->di_magic = be16_to_cpu(from->di_magic);
 569        to->di_mode = be16_to_cpu(from->di_mode);
 570        to->di_version = from ->di_version;
 571        to->di_format = from->di_format;
 572        to->di_onlink = be16_to_cpu(from->di_onlink);
 573        to->di_uid = be32_to_cpu(from->di_uid);
 574        to->di_gid = be32_to_cpu(from->di_gid);
 575        to->di_nlink = be32_to_cpu(from->di_nlink);
 576        to->di_projid_lo = be16_to_cpu(from->di_projid_lo);
 577        to->di_projid_hi = be16_to_cpu(from->di_projid_hi);
 578        memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
 579        to->di_flushiter = be16_to_cpu(from->di_flushiter);
 580        to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec);
 581        to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec);
 582        to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec);
 583        to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec);
 584        to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec);
 585        to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec);
 586        to->di_size = be64_to_cpu(from->di_size);
 587        to->di_nblocks = be64_to_cpu(from->di_nblocks);
 588        to->di_extsize = be32_to_cpu(from->di_extsize);
 589        to->di_nextents = be32_to_cpu(from->di_nextents);
 590        to->di_anextents = be16_to_cpu(from->di_anextents);
 591        to->di_forkoff = from->di_forkoff;
 592        to->di_aformat  = from->di_aformat;
 593        to->di_dmevmask = be32_to_cpu(from->di_dmevmask);
 594        to->di_dmstate  = be16_to_cpu(from->di_dmstate);
 595        to->di_flags    = be16_to_cpu(from->di_flags);
 596        to->di_gen      = be32_to_cpu(from->di_gen);
 597}
 598
 599void
 600xfs_dinode_to_disk(
 601        xfs_dinode_t            *to,
 602        xfs_icdinode_t          *from)
 603{
 604        to->di_magic = cpu_to_be16(from->di_magic);
 605        to->di_mode = cpu_to_be16(from->di_mode);
 606        to->di_version = from ->di_version;
 607        to->di_format = from->di_format;
 608        to->di_onlink = cpu_to_be16(from->di_onlink);
 609        to->di_uid = cpu_to_be32(from->di_uid);
 610        to->di_gid = cpu_to_be32(from->di_gid);
 611        to->di_nlink = cpu_to_be32(from->di_nlink);
 612        to->di_projid_lo = cpu_to_be16(from->di_projid_lo);
 613        to->di_projid_hi = cpu_to_be16(from->di_projid_hi);
 614        memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
 615        to->di_flushiter = cpu_to_be16(from->di_flushiter);
 616        to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
 617        to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
 618        to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
 619        to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
 620        to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
 621        to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
 622        to->di_size = cpu_to_be64(from->di_size);
 623        to->di_nblocks = cpu_to_be64(from->di_nblocks);
 624        to->di_extsize = cpu_to_be32(from->di_extsize);
 625        to->di_nextents = cpu_to_be32(from->di_nextents);
 626        to->di_anextents = cpu_to_be16(from->di_anextents);
 627        to->di_forkoff = from->di_forkoff;
 628        to->di_aformat = from->di_aformat;
 629        to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
 630        to->di_dmstate = cpu_to_be16(from->di_dmstate);
 631        to->di_flags = cpu_to_be16(from->di_flags);
 632        to->di_gen = cpu_to_be32(from->di_gen);
 633}
 634
 635STATIC uint
 636_xfs_dic2xflags(
 637        __uint16_t              di_flags)
 638{
 639        uint                    flags = 0;
 640
 641        if (di_flags & XFS_DIFLAG_ANY) {
 642                if (di_flags & XFS_DIFLAG_REALTIME)
 643                        flags |= XFS_XFLAG_REALTIME;
 644                if (di_flags & XFS_DIFLAG_PREALLOC)
 645                        flags |= XFS_XFLAG_PREALLOC;
 646                if (di_flags & XFS_DIFLAG_IMMUTABLE)
 647                        flags |= XFS_XFLAG_IMMUTABLE;
 648                if (di_flags & XFS_DIFLAG_APPEND)
 649                        flags |= XFS_XFLAG_APPEND;
 650                if (di_flags & XFS_DIFLAG_SYNC)
 651                        flags |= XFS_XFLAG_SYNC;
 652                if (di_flags & XFS_DIFLAG_NOATIME)
 653                        flags |= XFS_XFLAG_NOATIME;
 654                if (di_flags & XFS_DIFLAG_NODUMP)
 655                        flags |= XFS_XFLAG_NODUMP;
 656                if (di_flags & XFS_DIFLAG_RTINHERIT)
 657                        flags |= XFS_XFLAG_RTINHERIT;
 658                if (di_flags & XFS_DIFLAG_PROJINHERIT)
 659                        flags |= XFS_XFLAG_PROJINHERIT;
 660                if (di_flags & XFS_DIFLAG_NOSYMLINKS)
 661                        flags |= XFS_XFLAG_NOSYMLINKS;
 662                if (di_flags & XFS_DIFLAG_EXTSIZE)
 663                        flags |= XFS_XFLAG_EXTSIZE;
 664                if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
 665                        flags |= XFS_XFLAG_EXTSZINHERIT;
 666                if (di_flags & XFS_DIFLAG_NODEFRAG)
 667                        flags |= XFS_XFLAG_NODEFRAG;
 668                if (di_flags & XFS_DIFLAG_FILESTREAM)
 669                        flags |= XFS_XFLAG_FILESTREAM;
 670        }
 671
 672        return flags;
 673}
 674
 675uint
 676xfs_ip2xflags(
 677        xfs_inode_t             *ip)
 678{
 679        xfs_icdinode_t          *dic = &ip->i_d;
 680
 681        return _xfs_dic2xflags(dic->di_flags) |
 682                                (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
 683}
 684
 685uint
 686xfs_dic2xflags(
 687        xfs_dinode_t            *dip)
 688{
 689        return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
 690                                (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
 691}
 692
 693/*
 694 * Read the disk inode attributes into the in-core inode structure.
 695 */
 696int
 697xfs_iread(
 698        xfs_mount_t     *mp,
 699        xfs_trans_t     *tp,
 700        xfs_inode_t     *ip,
 701        uint            iget_flags)
 702{
 703        xfs_buf_t       *bp;
 704        xfs_dinode_t    *dip;
 705        int             error;
 706
 707        /*
 708         * Fill in the location information in the in-core inode.
 709         */
 710        error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, iget_flags);
 711        if (error)
 712                return error;
 713
 714        /*
 715         * Get pointers to the on-disk inode and the buffer containing it.
 716         */
 717        error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &bp, 0, iget_flags);
 718        if (error)
 719                return error;
 720
 721        /*
 722         * If we got something that isn't an inode it means someone
 723         * (nfs or dmi) has a stale handle.
 724         */
 725        if (dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC)) {
 726#ifdef DEBUG
 727                xfs_alert(mp,
 728                        "%s: dip->di_magic (0x%x) != XFS_DINODE_MAGIC (0x%x)",
 729                        __func__, be16_to_cpu(dip->di_magic), XFS_DINODE_MAGIC);
 730#endif /* DEBUG */
 731                error = XFS_ERROR(EINVAL);
 732                goto out_brelse;
 733        }
 734
 735        /*
 736         * If the on-disk inode is already linked to a directory
 737         * entry, copy all of the inode into the in-core inode.
 738         * xfs_iformat() handles copying in the inode format
 739         * specific information.
 740         * Otherwise, just get the truly permanent information.
 741         */
 742        if (dip->di_mode) {
 743                xfs_dinode_from_disk(&ip->i_d, dip);
 744                error = xfs_iformat(ip, dip);
 745                if (error)  {
 746#ifdef DEBUG
 747                        xfs_alert(mp, "%s: xfs_iformat() returned error %d",
 748                                __func__, error);
 749#endif /* DEBUG */
 750                        goto out_brelse;
 751                }
 752        } else {
 753                ip->i_d.di_magic = be16_to_cpu(dip->di_magic);
 754                ip->i_d.di_version = dip->di_version;
 755                ip->i_d.di_gen = be32_to_cpu(dip->di_gen);
 756                ip->i_d.di_flushiter = be16_to_cpu(dip->di_flushiter);
 757                /*
 758                 * Make sure to pull in the mode here as well in
 759                 * case the inode is released without being used.
 760                 * This ensures that xfs_inactive() will see that
 761                 * the inode is already free and not try to mess
 762                 * with the uninitialized part of it.
 763                 */
 764                ip->i_d.di_mode = 0;
 765        }
 766
 767        /*
 768         * The inode format changed when we moved the link count and
 769         * made it 32 bits long.  If this is an old format inode,
 770         * convert it in memory to look like a new one.  If it gets
 771         * flushed to disk we will convert back before flushing or
 772         * logging it.  We zero out the new projid field and the old link
 773         * count field.  We'll handle clearing the pad field (the remains
 774         * of the old uuid field) when we actually convert the inode to
 775         * the new format. We don't change the version number so that we
 776         * can distinguish this from a real new format inode.
 777         */
 778        if (ip->i_d.di_version == 1) {
 779                ip->i_d.di_nlink = ip->i_d.di_onlink;
 780                ip->i_d.di_onlink = 0;
 781                xfs_set_projid(ip, 0);
 782        }
 783
 784        ip->i_delayed_blks = 0;
 785
 786        /*
 787         * Mark the buffer containing the inode as something to keep
 788         * around for a while.  This helps to keep recently accessed
 789         * meta-data in-core longer.
 790         */
 791        xfs_buf_set_ref(bp, XFS_INO_REF);
 792
 793        /*
 794         * Use xfs_trans_brelse() to release the buffer containing the
 795         * on-disk inode, because it was acquired with xfs_trans_read_buf()
 796         * in xfs_imap_to_bp() above.  If tp is NULL, this is just a normal
 797         * brelse().  If we're within a transaction, then xfs_trans_brelse()
 798         * will only release the buffer if it is not dirty within the
 799         * transaction.  It will be OK to release the buffer in this case,
 800         * because inodes on disk are never destroyed and we will be
 801         * locking the new in-core inode before putting it in the hash
 802         * table where other processes can find it.  Thus we don't have
 803         * to worry about the inode being changed just because we released
 804         * the buffer.
 805         */
 806 out_brelse:
 807        xfs_trans_brelse(tp, bp);
 808        return error;
 809}
 810
 811/*
 812 * Read in extents from a btree-format inode.
 813 * Allocate and fill in if_extents.  Real work is done in xfs_bmap.c.
 814 */
 815int
 816xfs_iread_extents(
 817        xfs_trans_t     *tp,
 818        xfs_inode_t     *ip,
 819        int             whichfork)
 820{
 821        int             error;
 822        xfs_ifork_t     *ifp;
 823        xfs_extnum_t    nextents;
 824
 825        if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
 826                XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
 827                                 ip->i_mount);
 828                return XFS_ERROR(EFSCORRUPTED);
 829        }
 830        nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
 831        ifp = XFS_IFORK_PTR(ip, whichfork);
 832
 833        /*
 834         * We know that the size is valid (it's checked in iformat_btree)
 835         */
 836        ifp->if_bytes = ifp->if_real_bytes = 0;
 837        ifp->if_flags |= XFS_IFEXTENTS;
 838        xfs_iext_add(ifp, 0, nextents);
 839        error = xfs_bmap_read_extents(tp, ip, whichfork);
 840        if (error) {
 841                xfs_iext_destroy(ifp);
 842                ifp->if_flags &= ~XFS_IFEXTENTS;
 843                return error;
 844        }
 845        xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
 846        return 0;
 847}
 848
 849/*
 850 * Allocate an inode on disk and return a copy of its in-core version.
 851 * The in-core inode is locked exclusively.  Set mode, nlink, and rdev
 852 * appropriately within the inode.  The uid and gid for the inode are
 853 * set according to the contents of the given cred structure.
 854 *
 855 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
 856 * has a free inode available, call xfs_iget()
 857 * to obtain the in-core version of the allocated inode.  Finally,
 858 * fill in the inode and log its initial contents.  In this case,
 859 * ialloc_context would be set to NULL and call_again set to false.
 860 *
 861 * If xfs_dialloc() does not have an available inode,
 862 * it will replenish its supply by doing an allocation. Since we can
 863 * only do one allocation within a transaction without deadlocks, we
 864 * must commit the current transaction before returning the inode itself.
 865 * In this case, therefore, we will set call_again to true and return.
 866 * The caller should then commit the current transaction, start a new
 867 * transaction, and call xfs_ialloc() again to actually get the inode.
 868 *
 869 * To ensure that some other process does not grab the inode that
 870 * was allocated during the first call to xfs_ialloc(), this routine
 871 * also returns the [locked] bp pointing to the head of the freelist
 872 * as ialloc_context.  The caller should hold this buffer across
 873 * the commit and pass it back into this routine on the second call.
 874 *
 875 * If we are allocating quota inodes, we do not have a parent inode
 876 * to attach to or associate with (i.e. pip == NULL) because they
 877 * are not linked into the directory structure - they are attached
 878 * directly to the superblock - and so have no parent.
 879 */
 880int
 881xfs_ialloc(
 882        xfs_trans_t     *tp,
 883        xfs_inode_t     *pip,
 884        umode_t         mode,
 885        xfs_nlink_t     nlink,
 886        xfs_dev_t       rdev,
 887        prid_t          prid,
 888        int             okalloc,
 889        xfs_buf_t       **ialloc_context,
 890        xfs_inode_t     **ipp)
 891{
 892        xfs_ino_t       ino;
 893        xfs_inode_t     *ip;
 894        uint            flags;
 895        int             error;
 896        timespec_t      tv;
 897        int             filestreams = 0;
 898
 899        /*
 900         * Call the space management code to pick
 901         * the on-disk inode to be allocated.
 902         */
 903        error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
 904                            ialloc_context, &ino);
 905        if (error)
 906                return error;
 907        if (*ialloc_context || ino == NULLFSINO) {
 908                *ipp = NULL;
 909                return 0;
 910        }
 911        ASSERT(*ialloc_context == NULL);
 912
 913        /*
 914         * Get the in-core inode with the lock held exclusively.
 915         * This is because we're setting fields here we need
 916         * to prevent others from looking at until we're done.
 917         */
 918        error = xfs_iget(tp->t_mountp, tp, ino, XFS_IGET_CREATE,
 919                         XFS_ILOCK_EXCL, &ip);
 920        if (error)
 921                return error;
 922        ASSERT(ip != NULL);
 923
 924        ip->i_d.di_mode = mode;
 925        ip->i_d.di_onlink = 0;
 926        ip->i_d.di_nlink = nlink;
 927        ASSERT(ip->i_d.di_nlink == nlink);
 928        ip->i_d.di_uid = current_fsuid();
 929        ip->i_d.di_gid = current_fsgid();
 930        xfs_set_projid(ip, prid);
 931        memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
 932
 933        /*
 934         * If the superblock version is up to where we support new format
 935         * inodes and this is currently an old format inode, then change
 936         * the inode version number now.  This way we only do the conversion
 937         * here rather than here and in the flush/logging code.
 938         */
 939        if (xfs_sb_version_hasnlink(&tp->t_mountp->m_sb) &&
 940            ip->i_d.di_version == 1) {
 941                ip->i_d.di_version = 2;
 942                /*
 943                 * We've already zeroed the old link count, the projid field,
 944                 * and the pad field.
 945                 */
 946        }
 947
 948        /*
 949         * Project ids won't be stored on disk if we are using a version 1 inode.
 950         */
 951        if ((prid != 0) && (ip->i_d.di_version == 1))
 952                xfs_bump_ino_vers2(tp, ip);
 953
 954        if (pip && XFS_INHERIT_GID(pip)) {
 955                ip->i_d.di_gid = pip->i_d.di_gid;
 956                if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) {
 957                        ip->i_d.di_mode |= S_ISGID;
 958                }
 959        }
 960
 961        /*
 962         * If the group ID of the new file does not match the effective group
 963         * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
 964         * (and only if the irix_sgid_inherit compatibility variable is set).
 965         */
 966        if ((irix_sgid_inherit) &&
 967            (ip->i_d.di_mode & S_ISGID) &&
 968            (!in_group_p((gid_t)ip->i_d.di_gid))) {
 969                ip->i_d.di_mode &= ~S_ISGID;
 970        }
 971
 972        ip->i_d.di_size = 0;
 973        ip->i_d.di_nextents = 0;
 974        ASSERT(ip->i_d.di_nblocks == 0);
 975
 976        nanotime(&tv);
 977        ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
 978        ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
 979        ip->i_d.di_atime = ip->i_d.di_mtime;
 980        ip->i_d.di_ctime = ip->i_d.di_mtime;
 981
 982        /*
 983         * di_gen will have been taken care of in xfs_iread.
 984         */
 985        ip->i_d.di_extsize = 0;
 986        ip->i_d.di_dmevmask = 0;
 987        ip->i_d.di_dmstate = 0;
 988        ip->i_d.di_flags = 0;
 989        flags = XFS_ILOG_CORE;
 990        switch (mode & S_IFMT) {
 991        case S_IFIFO:
 992        case S_IFCHR:
 993        case S_IFBLK:
 994        case S_IFSOCK:
 995                ip->i_d.di_format = XFS_DINODE_FMT_DEV;
 996                ip->i_df.if_u2.if_rdev = rdev;
 997                ip->i_df.if_flags = 0;
 998                flags |= XFS_ILOG_DEV;
 999                break;
1000        case S_IFREG:
1001                /*
1002                 * we can't set up filestreams until after the VFS inode
1003                 * is set up properly.
1004                 */
1005                if (pip && xfs_inode_is_filestream(pip))
1006                        filestreams = 1;
1007                /* fall through */
1008        case S_IFDIR:
1009                if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1010                        uint    di_flags = 0;
1011
1012                        if (S_ISDIR(mode)) {
1013                                if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1014                                        di_flags |= XFS_DIFLAG_RTINHERIT;
1015                                if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1016                                        di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1017                                        ip->i_d.di_extsize = pip->i_d.di_extsize;
1018                                }
1019                        } else if (S_ISREG(mode)) {
1020                                if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1021                                        di_flags |= XFS_DIFLAG_REALTIME;
1022                                if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1023                                        di_flags |= XFS_DIFLAG_EXTSIZE;
1024                                        ip->i_d.di_extsize = pip->i_d.di_extsize;
1025                                }
1026                        }
1027                        if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1028                            xfs_inherit_noatime)
1029                                di_flags |= XFS_DIFLAG_NOATIME;
1030                        if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1031                            xfs_inherit_nodump)
1032                                di_flags |= XFS_DIFLAG_NODUMP;
1033                        if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1034                            xfs_inherit_sync)
1035                                di_flags |= XFS_DIFLAG_SYNC;
1036                        if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1037                            xfs_inherit_nosymlinks)
1038                                di_flags |= XFS_DIFLAG_NOSYMLINKS;
1039                        if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1040                                di_flags |= XFS_DIFLAG_PROJINHERIT;
1041                        if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1042                            xfs_inherit_nodefrag)
1043                                di_flags |= XFS_DIFLAG_NODEFRAG;
1044                        if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1045                                di_flags |= XFS_DIFLAG_FILESTREAM;
1046                        ip->i_d.di_flags |= di_flags;
1047                }
1048                /* FALLTHROUGH */
1049        case S_IFLNK:
1050                ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1051                ip->i_df.if_flags = XFS_IFEXTENTS;
1052                ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1053                ip->i_df.if_u1.if_extents = NULL;
1054                break;
1055        default:
1056                ASSERT(0);
1057        }
1058        /*
1059         * Attribute fork settings for new inode.
1060         */
1061        ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1062        ip->i_d.di_anextents = 0;
1063
1064        /*
1065         * Log the new values stuffed into the inode.
1066         */
1067        xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1068        xfs_trans_log_inode(tp, ip, flags);
1069
1070        /* now that we have an i_mode we can setup inode ops and unlock */
1071        xfs_setup_inode(ip);
1072
1073        /* now we have set up the vfs inode we can associate the filestream */
1074        if (filestreams) {
1075                error = xfs_filestream_associate(pip, ip);
1076                if (error < 0)
1077                        return -error;
1078                if (!error)
1079                        xfs_iflags_set(ip, XFS_IFILESTREAM);
1080        }
1081
1082        *ipp = ip;
1083        return 0;
1084}
1085
1086/*
1087 * Free up the underlying blocks past new_size.  The new size must be smaller
1088 * than the current size.  This routine can be used both for the attribute and
1089 * data fork, and does not modify the inode size, which is left to the caller.
1090 *
1091 * The transaction passed to this routine must have made a permanent log
1092 * reservation of at least XFS_ITRUNCATE_LOG_RES.  This routine may commit the
1093 * given transaction and start new ones, so make sure everything involved in
1094 * the transaction is tidy before calling here.  Some transaction will be
1095 * returned to the caller to be committed.  The incoming transaction must
1096 * already include the inode, and both inode locks must be held exclusively.
1097 * The inode must also be "held" within the transaction.  On return the inode
1098 * will be "held" within the returned transaction.  This routine does NOT
1099 * require any disk space to be reserved for it within the transaction.
1100 *
1101 * If we get an error, we must return with the inode locked and linked into the
1102 * current transaction. This keeps things simple for the higher level code,
1103 * because it always knows that the inode is locked and held in the transaction
1104 * that returns to it whether errors occur or not.  We don't mark the inode
1105 * dirty on error so that transactions can be easily aborted if possible.
1106 */
1107int
1108xfs_itruncate_extents(
1109        struct xfs_trans        **tpp,
1110        struct xfs_inode        *ip,
1111        int                     whichfork,
1112        xfs_fsize_t             new_size)
1113{
1114        struct xfs_mount        *mp = ip->i_mount;
1115        struct xfs_trans        *tp = *tpp;
1116        struct xfs_trans        *ntp;
1117        xfs_bmap_free_t         free_list;
1118        xfs_fsblock_t           first_block;
1119        xfs_fileoff_t           first_unmap_block;
1120        xfs_fileoff_t           last_block;
1121        xfs_filblks_t           unmap_len;
1122        int                     committed;
1123        int                     error = 0;
1124        int                     done = 0;
1125
1126        ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1127        ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1128               xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1129        ASSERT(new_size <= XFS_ISIZE(ip));
1130        ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1131        ASSERT(ip->i_itemp != NULL);
1132        ASSERT(ip->i_itemp->ili_lock_flags == 0);
1133        ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1134
1135        trace_xfs_itruncate_extents_start(ip, new_size);
1136
1137        /*
1138         * Since it is possible for space to become allocated beyond
1139         * the end of the file (in a crash where the space is allocated
1140         * but the inode size is not yet updated), simply remove any
1141         * blocks which show up between the new EOF and the maximum
1142         * possible file size.  If the first block to be removed is
1143         * beyond the maximum file size (ie it is the same as last_block),
1144         * then there is nothing to do.
1145         */
1146        first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1147        last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1148        if (first_unmap_block == last_block)
1149                return 0;
1150
1151        ASSERT(first_unmap_block < last_block);
1152        unmap_len = last_block - first_unmap_block + 1;
1153        while (!done) {
1154                xfs_bmap_init(&free_list, &first_block);
1155                error = xfs_bunmapi(tp, ip,
1156                                    first_unmap_block, unmap_len,
1157                                    xfs_bmapi_aflag(whichfork),
1158                                    XFS_ITRUNC_MAX_EXTENTS,
1159                                    &first_block, &free_list,
1160                                    &done);
1161                if (error)
1162                        goto out_bmap_cancel;
1163
1164                /*
1165                 * Duplicate the transaction that has the permanent
1166                 * reservation and commit the old transaction.
1167                 */
1168                error = xfs_bmap_finish(&tp, &free_list, &committed);
1169                if (committed)
1170                        xfs_trans_ijoin(tp, ip, 0);
1171                if (error)
1172                        goto out_bmap_cancel;
1173
1174                if (committed) {
1175                        /*
1176                         * Mark the inode dirty so it will be logged and
1177                         * moved forward in the log as part of every commit.
1178                         */
1179                        xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1180                }
1181
1182                ntp = xfs_trans_dup(tp);
1183                error = xfs_trans_commit(tp, 0);
1184                tp = ntp;
1185
1186                xfs_trans_ijoin(tp, ip, 0);
1187
1188                if (error)
1189                        goto out;
1190
1191                /*
1192                 * Transaction commit worked ok so we can drop the extra ticket
1193                 * reference that we gained in xfs_trans_dup()
1194                 */
1195                xfs_log_ticket_put(tp->t_ticket);
1196                error = xfs_trans_reserve(tp, 0,
1197                                        XFS_ITRUNCATE_LOG_RES(mp), 0,
1198                                        XFS_TRANS_PERM_LOG_RES,
1199                                        XFS_ITRUNCATE_LOG_COUNT);
1200                if (error)
1201                        goto out;
1202        }
1203
1204        /*
1205         * Always re-log the inode so that our permanent transaction can keep
1206         * on rolling it forward in the log.
1207         */
1208        xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1209
1210        trace_xfs_itruncate_extents_end(ip, new_size);
1211
1212out:
1213        *tpp = tp;
1214        return error;
1215out_bmap_cancel:
1216        /*
1217         * If the bunmapi call encounters an error, return to the caller where
1218         * the transaction can be properly aborted.  We just need to make sure
1219         * we're not holding any resources that we were not when we came in.
1220         */
1221        xfs_bmap_cancel(&free_list);
1222        goto out;
1223}
1224
1225/*
1226 * This is called when the inode's link count goes to 0.
1227 * We place the on-disk inode on a list in the AGI.  It
1228 * will be pulled from this list when the inode is freed.
1229 */
1230int
1231xfs_iunlink(
1232        xfs_trans_t     *tp,
1233        xfs_inode_t     *ip)
1234{
1235        xfs_mount_t     *mp;
1236        xfs_agi_t       *agi;
1237        xfs_dinode_t    *dip;
1238        xfs_buf_t       *agibp;
1239        xfs_buf_t       *ibp;
1240        xfs_agino_t     agino;
1241        short           bucket_index;
1242        int             offset;
1243        int             error;
1244
1245        ASSERT(ip->i_d.di_nlink == 0);
1246        ASSERT(ip->i_d.di_mode != 0);
1247
1248        mp = tp->t_mountp;
1249
1250        /*
1251         * Get the agi buffer first.  It ensures lock ordering
1252         * on the list.
1253         */
1254        error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1255        if (error)
1256                return error;
1257        agi = XFS_BUF_TO_AGI(agibp);
1258
1259        /*
1260         * Get the index into the agi hash table for the
1261         * list this inode will go on.
1262         */
1263        agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1264        ASSERT(agino != 0);
1265        bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1266        ASSERT(agi->agi_unlinked[bucket_index]);
1267        ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1268
1269        if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
1270                /*
1271                 * There is already another inode in the bucket we need
1272                 * to add ourselves to.  Add us at the front of the list.
1273                 * Here we put the head pointer into our next pointer,
1274                 * and then we fall through to point the head at us.
1275                 */
1276                error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1277                                       0, 0);
1278                if (error)
1279                        return error;
1280
1281                ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
1282                dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1283                offset = ip->i_imap.im_boffset +
1284                        offsetof(xfs_dinode_t, di_next_unlinked);
1285                xfs_trans_inode_buf(tp, ibp);
1286                xfs_trans_log_buf(tp, ibp, offset,
1287                                  (offset + sizeof(xfs_agino_t) - 1));
1288                xfs_inobp_check(mp, ibp);
1289        }
1290
1291        /*
1292         * Point the bucket head pointer at the inode being inserted.
1293         */
1294        ASSERT(agino != 0);
1295        agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1296        offset = offsetof(xfs_agi_t, agi_unlinked) +
1297                (sizeof(xfs_agino_t) * bucket_index);
1298        xfs_trans_log_buf(tp, agibp, offset,
1299                          (offset + sizeof(xfs_agino_t) - 1));
1300        return 0;
1301}
1302
1303/*
1304 * Pull the on-disk inode from the AGI unlinked list.
1305 */
1306STATIC int
1307xfs_iunlink_remove(
1308        xfs_trans_t     *tp,
1309        xfs_inode_t     *ip)
1310{
1311        xfs_ino_t       next_ino;
1312        xfs_mount_t     *mp;
1313        xfs_agi_t       *agi;
1314        xfs_dinode_t    *dip;
1315        xfs_buf_t       *agibp;
1316        xfs_buf_t       *ibp;
1317        xfs_agnumber_t  agno;
1318        xfs_agino_t     agino;
1319        xfs_agino_t     next_agino;
1320        xfs_buf_t       *last_ibp;
1321        xfs_dinode_t    *last_dip = NULL;
1322        short           bucket_index;
1323        int             offset, last_offset = 0;
1324        int             error;
1325
1326        mp = tp->t_mountp;
1327        agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1328
1329        /*
1330         * Get the agi buffer first.  It ensures lock ordering
1331         * on the list.
1332         */
1333        error = xfs_read_agi(mp, tp, agno, &agibp);
1334        if (error)
1335                return error;
1336
1337        agi = XFS_BUF_TO_AGI(agibp);
1338
1339        /*
1340         * Get the index into the agi hash table for the
1341         * list this inode will go on.
1342         */
1343        agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1344        ASSERT(agino != 0);
1345        bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1346        ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
1347        ASSERT(agi->agi_unlinked[bucket_index]);
1348
1349        if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
1350                /*
1351                 * We're at the head of the list.  Get the inode's on-disk
1352                 * buffer to see if there is anyone after us on the list.
1353                 * Only modify our next pointer if it is not already NULLAGINO.
1354                 * This saves us the overhead of dealing with the buffer when
1355                 * there is no need to change it.
1356                 */
1357                error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1358                                       0, 0);
1359                if (error) {
1360                        xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
1361                                __func__, error);
1362                        return error;
1363                }
1364                next_agino = be32_to_cpu(dip->di_next_unlinked);
1365                ASSERT(next_agino != 0);
1366                if (next_agino != NULLAGINO) {
1367                        dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1368                        offset = ip->i_imap.im_boffset +
1369                                offsetof(xfs_dinode_t, di_next_unlinked);
1370                        xfs_trans_inode_buf(tp, ibp);
1371                        xfs_trans_log_buf(tp, ibp, offset,
1372                                          (offset + sizeof(xfs_agino_t) - 1));
1373                        xfs_inobp_check(mp, ibp);
1374                } else {
1375                        xfs_trans_brelse(tp, ibp);
1376                }
1377                /*
1378                 * Point the bucket head pointer at the next inode.
1379                 */
1380                ASSERT(next_agino != 0);
1381                ASSERT(next_agino != agino);
1382                agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
1383                offset = offsetof(xfs_agi_t, agi_unlinked) +
1384                        (sizeof(xfs_agino_t) * bucket_index);
1385                xfs_trans_log_buf(tp, agibp, offset,
1386                                  (offset + sizeof(xfs_agino_t) - 1));
1387        } else {
1388                /*
1389                 * We need to search the list for the inode being freed.
1390                 */
1391                next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1392                last_ibp = NULL;
1393                while (next_agino != agino) {
1394                        struct xfs_imap imap;
1395
1396                        if (last_ibp)
1397                                xfs_trans_brelse(tp, last_ibp);
1398
1399                        imap.im_blkno = 0;
1400                        next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
1401
1402                        error = xfs_imap(mp, tp, next_ino, &imap, 0);
1403                        if (error) {
1404                                xfs_warn(mp,
1405        "%s: xfs_imap returned error %d.",
1406                                         __func__, error);
1407                                return error;
1408                        }
1409
1410                        error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
1411                                               &last_ibp, 0, 0);
1412                        if (error) {
1413                                xfs_warn(mp,
1414        "%s: xfs_imap_to_bp returned error %d.",
1415                                        __func__, error);
1416                                return error;
1417                        }
1418
1419                        last_offset = imap.im_boffset;
1420                        next_agino = be32_to_cpu(last_dip->di_next_unlinked);
1421                        ASSERT(next_agino != NULLAGINO);
1422                        ASSERT(next_agino != 0);
1423                }
1424
1425                /*
1426                 * Now last_ibp points to the buffer previous to us on the
1427                 * unlinked list.  Pull us from the list.
1428                 */
1429                error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1430                                       0, 0);
1431                if (error) {
1432                        xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
1433                                __func__, error);
1434                        return error;
1435                }
1436                next_agino = be32_to_cpu(dip->di_next_unlinked);
1437                ASSERT(next_agino != 0);
1438                ASSERT(next_agino != agino);
1439                if (next_agino != NULLAGINO) {
1440                        dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1441                        offset = ip->i_imap.im_boffset +
1442                                offsetof(xfs_dinode_t, di_next_unlinked);
1443                        xfs_trans_inode_buf(tp, ibp);
1444                        xfs_trans_log_buf(tp, ibp, offset,
1445                                          (offset + sizeof(xfs_agino_t) - 1));
1446                        xfs_inobp_check(mp, ibp);
1447                } else {
1448                        xfs_trans_brelse(tp, ibp);
1449                }
1450                /*
1451                 * Point the previous inode on the list to the next inode.
1452                 */
1453                last_dip->di_next_unlinked = cpu_to_be32(next_agino);
1454                ASSERT(next_agino != 0);
1455                offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
1456                xfs_trans_inode_buf(tp, last_ibp);
1457                xfs_trans_log_buf(tp, last_ibp, offset,
1458                                  (offset + sizeof(xfs_agino_t) - 1));
1459                xfs_inobp_check(mp, last_ibp);
1460        }
1461        return 0;
1462}
1463
1464/*
1465 * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1466 * inodes that are in memory - they all must be marked stale and attached to
1467 * the cluster buffer.
1468 */
1469STATIC int
1470xfs_ifree_cluster(
1471        xfs_inode_t     *free_ip,
1472        xfs_trans_t     *tp,
1473        xfs_ino_t       inum)
1474{
1475        xfs_mount_t             *mp = free_ip->i_mount;
1476        int                     blks_per_cluster;
1477        int                     nbufs;
1478        int                     ninodes;
1479        int                     i, j;
1480        xfs_daddr_t             blkno;
1481        xfs_buf_t               *bp;
1482        xfs_inode_t             *ip;
1483        xfs_inode_log_item_t    *iip;
1484        xfs_log_item_t          *lip;
1485        struct xfs_perag        *pag;
1486
1487        pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
1488        if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
1489                blks_per_cluster = 1;
1490                ninodes = mp->m_sb.sb_inopblock;
1491                nbufs = XFS_IALLOC_BLOCKS(mp);
1492        } else {
1493                blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
1494                                        mp->m_sb.sb_blocksize;
1495                ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
1496                nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
1497        }
1498
1499        for (j = 0; j < nbufs; j++, inum += ninodes) {
1500                blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
1501                                         XFS_INO_TO_AGBNO(mp, inum));
1502
1503                /*
1504                 * We obtain and lock the backing buffer first in the process
1505                 * here, as we have to ensure that any dirty inode that we
1506                 * can't get the flush lock on is attached to the buffer.
1507                 * If we scan the in-memory inodes first, then buffer IO can
1508                 * complete before we get a lock on it, and hence we may fail
1509                 * to mark all the active inodes on the buffer stale.
1510                 */
1511                bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
1512                                        mp->m_bsize * blks_per_cluster, 0);
1513
1514                if (!bp)
1515                        return ENOMEM;
1516                /*
1517                 * Walk the inodes already attached to the buffer and mark them
1518                 * stale. These will all have the flush locks held, so an
1519                 * in-memory inode walk can't lock them. By marking them all
1520                 * stale first, we will not attempt to lock them in the loop
1521                 * below as the XFS_ISTALE flag will be set.
1522                 */
1523                lip = bp->b_fspriv;
1524                while (lip) {
1525                        if (lip->li_type == XFS_LI_INODE) {
1526                                iip = (xfs_inode_log_item_t *)lip;
1527                                ASSERT(iip->ili_logged == 1);
1528                                lip->li_cb = xfs_istale_done;
1529                                xfs_trans_ail_copy_lsn(mp->m_ail,
1530                                                        &iip->ili_flush_lsn,
1531                                                        &iip->ili_item.li_lsn);
1532                                xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
1533                        }
1534                        lip = lip->li_bio_list;
1535                }
1536
1537
1538                /*
1539                 * For each inode in memory attempt to add it to the inode
1540                 * buffer and set it up for being staled on buffer IO
1541                 * completion.  This is safe as we've locked out tail pushing
1542                 * and flushing by locking the buffer.
1543                 *
1544                 * We have already marked every inode that was part of a
1545                 * transaction stale above, which means there is no point in
1546                 * even trying to lock them.
1547                 */
1548                for (i = 0; i < ninodes; i++) {
1549retry:
1550                        rcu_read_lock();
1551                        ip = radix_tree_lookup(&pag->pag_ici_root,
1552                                        XFS_INO_TO_AGINO(mp, (inum + i)));
1553
1554                        /* Inode not in memory, nothing to do */
1555                        if (!ip) {
1556                                rcu_read_unlock();
1557                                continue;
1558                        }
1559
1560                        /*
1561                         * because this is an RCU protected lookup, we could
1562                         * find a recently freed or even reallocated inode
1563                         * during the lookup. We need to check under the
1564                         * i_flags_lock for a valid inode here. Skip it if it
1565                         * is not valid, the wrong inode or stale.
1566                         */
1567                        spin_lock(&ip->i_flags_lock);
1568                        if (ip->i_ino != inum + i ||
1569                            __xfs_iflags_test(ip, XFS_ISTALE)) {
1570                                spin_unlock(&ip->i_flags_lock);
1571                                rcu_read_unlock();
1572                                continue;
1573                        }
1574                        spin_unlock(&ip->i_flags_lock);
1575
1576                        /*
1577                         * Don't try to lock/unlock the current inode, but we
1578                         * _cannot_ skip the other inodes that we did not find
1579                         * in the list attached to the buffer and are not
1580                         * already marked stale. If we can't lock it, back off
1581                         * and retry.
1582                         */
1583                        if (ip != free_ip &&
1584                            !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
1585                                rcu_read_unlock();
1586                                delay(1);
1587                                goto retry;
1588                        }
1589                        rcu_read_unlock();
1590
1591                        xfs_iflock(ip);
1592                        xfs_iflags_set(ip, XFS_ISTALE);
1593
1594                        /*
1595                         * we don't need to attach clean inodes or those only
1596                         * with unlogged changes (which we throw away, anyway).
1597                         */
1598                        iip = ip->i_itemp;
1599                        if (!iip || xfs_inode_clean(ip)) {
1600                                ASSERT(ip != free_ip);
1601                                xfs_ifunlock(ip);
1602                                xfs_iunlock(ip, XFS_ILOCK_EXCL);
1603                                continue;
1604                        }
1605
1606                        iip->ili_last_fields = iip->ili_fields;
1607                        iip->ili_fields = 0;
1608                        iip->ili_logged = 1;
1609                        xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
1610                                                &iip->ili_item.li_lsn);
1611
1612                        xfs_buf_attach_iodone(bp, xfs_istale_done,
1613                                                  &iip->ili_item);
1614
1615                        if (ip != free_ip)
1616                                xfs_iunlock(ip, XFS_ILOCK_EXCL);
1617                }
1618
1619                xfs_trans_stale_inode_buf(tp, bp);
1620                xfs_trans_binval(tp, bp);
1621        }
1622
1623        xfs_perag_put(pag);
1624        return 0;
1625}
1626
1627/*
1628 * This is called to return an inode to the inode free list.
1629 * The inode should already be truncated to 0 length and have
1630 * no pages associated with it.  This routine also assumes that
1631 * the inode is already a part of the transaction.
1632 *
1633 * The on-disk copy of the inode will have been added to the list
1634 * of unlinked inodes in the AGI. We need to remove the inode from
1635 * that list atomically with respect to freeing it here.
1636 */
1637int
1638xfs_ifree(
1639        xfs_trans_t     *tp,
1640        xfs_inode_t     *ip,
1641        xfs_bmap_free_t *flist)
1642{
1643        int                     error;
1644        int                     delete;
1645        xfs_ino_t               first_ino;
1646        xfs_dinode_t            *dip;
1647        xfs_buf_t               *ibp;
1648
1649        ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1650        ASSERT(ip->i_d.di_nlink == 0);
1651        ASSERT(ip->i_d.di_nextents == 0);
1652        ASSERT(ip->i_d.di_anextents == 0);
1653        ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode));
1654        ASSERT(ip->i_d.di_nblocks == 0);
1655
1656        /*
1657         * Pull the on-disk inode from the AGI unlinked list.
1658         */
1659        error = xfs_iunlink_remove(tp, ip);
1660        if (error != 0) {
1661                return error;
1662        }
1663
1664        error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
1665        if (error != 0) {
1666                return error;
1667        }
1668        ip->i_d.di_mode = 0;            /* mark incore inode as free */
1669        ip->i_d.di_flags = 0;
1670        ip->i_d.di_dmevmask = 0;
1671        ip->i_d.di_forkoff = 0;         /* mark the attr fork not in use */
1672        ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1673        ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1674        /*
1675         * Bump the generation count so no one will be confused
1676         * by reincarnations of this inode.
1677         */
1678        ip->i_d.di_gen++;
1679
1680        xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1681
1682        error = xfs_imap_to_bp(ip->i_mount, tp, &ip->i_imap, &dip, &ibp,
1683                               0, 0);
1684        if (error)
1685                return error;
1686
1687        /*
1688        * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
1689        * from picking up this inode when it is reclaimed (its incore state
1690        * initialzed but not flushed to disk yet). The in-core di_mode is
1691        * already cleared  and a corresponding transaction logged.
1692        * The hack here just synchronizes the in-core to on-disk
1693        * di_mode value in advance before the actual inode sync to disk.
1694        * This is OK because the inode is already unlinked and would never
1695        * change its di_mode again for this inode generation.
1696        * This is a temporary hack that would require a proper fix
1697        * in the future.
1698        */
1699        dip->di_mode = 0;
1700
1701        if (delete) {
1702                error = xfs_ifree_cluster(ip, tp, first_ino);
1703        }
1704
1705        return error;
1706}
1707
1708/*
1709 * Reallocate the space for if_broot based on the number of records
1710 * being added or deleted as indicated in rec_diff.  Move the records
1711 * and pointers in if_broot to fit the new size.  When shrinking this
1712 * will eliminate holes between the records and pointers created by
1713 * the caller.  When growing this will create holes to be filled in
1714 * by the caller.
1715 *
1716 * The caller must not request to add more records than would fit in
1717 * the on-disk inode root.  If the if_broot is currently NULL, then
1718 * if we adding records one will be allocated.  The caller must also
1719 * not request that the number of records go below zero, although
1720 * it can go to zero.
1721 *
1722 * ip -- the inode whose if_broot area is changing
1723 * ext_diff -- the change in the number of records, positive or negative,
1724 *       requested for the if_broot array.
1725 */
1726void
1727xfs_iroot_realloc(
1728        xfs_inode_t             *ip,
1729        int                     rec_diff,
1730        int                     whichfork)
1731{
1732        struct xfs_mount        *mp = ip->i_mount;
1733        int                     cur_max;
1734        xfs_ifork_t             *ifp;
1735        struct xfs_btree_block  *new_broot;
1736        int                     new_max;
1737        size_t                  new_size;
1738        char                    *np;
1739        char                    *op;
1740
1741        /*
1742         * Handle the degenerate case quietly.
1743         */
1744        if (rec_diff == 0) {
1745                return;
1746        }
1747
1748        ifp = XFS_IFORK_PTR(ip, whichfork);
1749        if (rec_diff > 0) {
1750                /*
1751                 * If there wasn't any memory allocated before, just
1752                 * allocate it now and get out.
1753                 */
1754                if (ifp->if_broot_bytes == 0) {
1755                        new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
1756                        ifp->if_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
1757                        ifp->if_broot_bytes = (int)new_size;
1758                        return;
1759                }
1760
1761                /*
1762                 * If there is already an existing if_broot, then we need
1763                 * to realloc() it and shift the pointers to their new
1764                 * location.  The records don't change location because
1765                 * they are kept butted up against the btree block header.
1766                 */
1767                cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
1768                new_max = cur_max + rec_diff;
1769                new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
1770                ifp->if_broot = kmem_realloc(ifp->if_broot, new_size,
1771                                (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
1772                                KM_SLEEP | KM_NOFS);
1773                op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
1774                                                     ifp->if_broot_bytes);
1775                np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
1776                                                     (int)new_size);
1777                ifp->if_broot_bytes = (int)new_size;
1778                ASSERT(ifp->if_broot_bytes <=
1779                        XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
1780                memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
1781                return;
1782        }
1783
1784        /*
1785         * rec_diff is less than 0.  In this case, we are shrinking the
1786         * if_broot buffer.  It must already exist.  If we go to zero
1787         * records, just get rid of the root and clear the status bit.
1788         */
1789        ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
1790        cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
1791        new_max = cur_max + rec_diff;
1792        ASSERT(new_max >= 0);
1793        if (new_max > 0)
1794                new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
1795        else
1796                new_size = 0;
1797        if (new_size > 0) {
1798                new_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
1799                /*
1800                 * First copy over the btree block header.
1801                 */
1802                memcpy(new_broot, ifp->if_broot, XFS_BTREE_LBLOCK_LEN);
1803        } else {
1804                new_broot = NULL;
1805                ifp->if_flags &= ~XFS_IFBROOT;
1806        }
1807
1808        /*
1809         * Only copy the records and pointers if there are any.
1810         */
1811        if (new_max > 0) {
1812                /*
1813                 * First copy the records.
1814                 */
1815                op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
1816                np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
1817                memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
1818
1819                /*
1820                 * Then copy the pointers.
1821                 */
1822                op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
1823                                                     ifp->if_broot_bytes);
1824                np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
1825                                                     (int)new_size);
1826                memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
1827        }
1828        kmem_free(ifp->if_broot);
1829        ifp->if_broot = new_broot;
1830        ifp->if_broot_bytes = (int)new_size;
1831        ASSERT(ifp->if_broot_bytes <=
1832                XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
1833        return;
1834}
1835
1836
1837/*
1838 * This is called when the amount of space needed for if_data
1839 * is increased or decreased.  The change in size is indicated by
1840 * the number of bytes that need to be added or deleted in the
1841 * byte_diff parameter.
1842 *
1843 * If the amount of space needed has decreased below the size of the
1844 * inline buffer, then switch to using the inline buffer.  Otherwise,
1845 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
1846 * to what is needed.
1847 *
1848 * ip -- the inode whose if_data area is changing
1849 * byte_diff -- the change in the number of bytes, positive or negative,
1850 *       requested for the if_data array.
1851 */
1852void
1853xfs_idata_realloc(
1854        xfs_inode_t     *ip,
1855        int             byte_diff,
1856        int             whichfork)
1857{
1858        xfs_ifork_t     *ifp;
1859        int             new_size;
1860        int             real_size;
1861
1862        if (byte_diff == 0) {
1863                return;
1864        }
1865
1866        ifp = XFS_IFORK_PTR(ip, whichfork);
1867        new_size = (int)ifp->if_bytes + byte_diff;
1868        ASSERT(new_size >= 0);
1869
1870        if (new_size == 0) {
1871                if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
1872                        kmem_free(ifp->if_u1.if_data);
1873                }
1874                ifp->if_u1.if_data = NULL;
1875                real_size = 0;
1876        } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
1877                /*
1878                 * If the valid extents/data can fit in if_inline_ext/data,
1879                 * copy them from the malloc'd vector and free it.
1880                 */
1881                if (ifp->if_u1.if_data == NULL) {
1882                        ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
1883                } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
1884                        ASSERT(ifp->if_real_bytes != 0);
1885                        memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
1886                              new_size);
1887                        kmem_free(ifp->if_u1.if_data);
1888                        ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
1889                }
1890                real_size = 0;
1891        } else {
1892                /*
1893                 * Stuck with malloc/realloc.
1894                 * For inline data, the underlying buffer must be
1895                 * a multiple of 4 bytes in size so that it can be
1896                 * logged and stay on word boundaries.  We enforce
1897                 * that here.
1898                 */
1899                real_size = roundup(new_size, 4);
1900                if (ifp->if_u1.if_data == NULL) {
1901                        ASSERT(ifp->if_real_bytes == 0);
1902                        ifp->if_u1.if_data = kmem_alloc(real_size,
1903                                                        KM_SLEEP | KM_NOFS);
1904                } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
1905                        /*
1906                         * Only do the realloc if the underlying size
1907                         * is really changing.
1908                         */
1909                        if (ifp->if_real_bytes != real_size) {
1910                                ifp->if_u1.if_data =
1911                                        kmem_realloc(ifp->if_u1.if_data,
1912                                                        real_size,
1913                                                        ifp->if_real_bytes,
1914                                                        KM_SLEEP | KM_NOFS);
1915                        }
1916                } else {
1917                        ASSERT(ifp->if_real_bytes == 0);
1918                        ifp->if_u1.if_data = kmem_alloc(real_size,
1919                                                        KM_SLEEP | KM_NOFS);
1920                        memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
1921                                ifp->if_bytes);
1922                }
1923        }
1924        ifp->if_real_bytes = real_size;
1925        ifp->if_bytes = new_size;
1926        ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
1927}
1928
1929void
1930xfs_idestroy_fork(
1931        xfs_inode_t     *ip,
1932        int             whichfork)
1933{
1934        xfs_ifork_t     *ifp;
1935
1936        ifp = XFS_IFORK_PTR(ip, whichfork);
1937        if (ifp->if_broot != NULL) {
1938                kmem_free(ifp->if_broot);
1939                ifp->if_broot = NULL;
1940        }
1941
1942        /*
1943         * If the format is local, then we can't have an extents
1944         * array so just look for an inline data array.  If we're
1945         * not local then we may or may not have an extents list,
1946         * so check and free it up if we do.
1947         */
1948        if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1949                if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
1950                    (ifp->if_u1.if_data != NULL)) {
1951                        ASSERT(ifp->if_real_bytes != 0);
1952                        kmem_free(ifp->if_u1.if_data);
1953                        ifp->if_u1.if_data = NULL;
1954                        ifp->if_real_bytes = 0;
1955                }
1956        } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
1957                   ((ifp->if_flags & XFS_IFEXTIREC) ||
1958                    ((ifp->if_u1.if_extents != NULL) &&
1959                     (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
1960                ASSERT(ifp->if_real_bytes != 0);
1961                xfs_iext_destroy(ifp);
1962        }
1963        ASSERT(ifp->if_u1.if_extents == NULL ||
1964               ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
1965        ASSERT(ifp->if_real_bytes == 0);
1966        if (whichfork == XFS_ATTR_FORK) {
1967                kmem_zone_free(xfs_ifork_zone, ip->i_afp);
1968                ip->i_afp = NULL;
1969        }
1970}
1971
1972/*
1973 * This is called to unpin an inode.  The caller must have the inode locked
1974 * in at least shared mode so that the buffer cannot be subsequently pinned
1975 * once someone is waiting for it to be unpinned.
1976 */
1977static void
1978xfs_iunpin(
1979        struct xfs_inode        *ip)
1980{
1981        ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
1982
1983        trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
1984
1985        /* Give the log a push to start the unpinning I/O */
1986        xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
1987
1988}
1989
1990static void
1991__xfs_iunpin_wait(
1992        struct xfs_inode        *ip)
1993{
1994        wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
1995        DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
1996
1997        xfs_iunpin(ip);
1998
1999        do {
2000                prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2001                if (xfs_ipincount(ip))
2002                        io_schedule();
2003        } while (xfs_ipincount(ip));
2004        finish_wait(wq, &wait.wait);
2005}
2006
2007void
2008xfs_iunpin_wait(
2009        struct xfs_inode        *ip)
2010{
2011        if (xfs_ipincount(ip))
2012                __xfs_iunpin_wait(ip);
2013}
2014
2015/*
2016 * xfs_iextents_copy()
2017 *
2018 * This is called to copy the REAL extents (as opposed to the delayed
2019 * allocation extents) from the inode into the given buffer.  It
2020 * returns the number of bytes copied into the buffer.
2021 *
2022 * If there are no delayed allocation extents, then we can just
2023 * memcpy() the extents into the buffer.  Otherwise, we need to
2024 * examine each extent in turn and skip those which are delayed.
2025 */
2026int
2027xfs_iextents_copy(
2028        xfs_inode_t             *ip,
2029        xfs_bmbt_rec_t          *dp,
2030        int                     whichfork)
2031{
2032        int                     copied;
2033        int                     i;
2034        xfs_ifork_t             *ifp;
2035        int                     nrecs;
2036        xfs_fsblock_t           start_block;
2037
2038        ifp = XFS_IFORK_PTR(ip, whichfork);
2039        ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2040        ASSERT(ifp->if_bytes > 0);
2041
2042        nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2043        XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
2044        ASSERT(nrecs > 0);
2045
2046        /*
2047         * There are some delayed allocation extents in the
2048         * inode, so copy the extents one at a time and skip
2049         * the delayed ones.  There must be at least one
2050         * non-delayed extent.
2051         */
2052        copied = 0;
2053        for (i = 0; i < nrecs; i++) {
2054                xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
2055                start_block = xfs_bmbt_get_startblock(ep);
2056                if (isnullstartblock(start_block)) {
2057                        /*
2058                         * It's a delayed allocation extent, so skip it.
2059                         */
2060                        continue;
2061                }
2062
2063                /* Translate to on disk format */
2064                put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
2065                put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
2066                dp++;
2067                copied++;
2068        }
2069        ASSERT(copied != 0);
2070        xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
2071
2072        return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2073}
2074
2075/*
2076 * Each of the following cases stores data into the same region
2077 * of the on-disk inode, so only one of them can be valid at
2078 * any given time. While it is possible to have conflicting formats
2079 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2080 * in EXTENTS format, this can only happen when the fork has
2081 * changed formats after being modified but before being flushed.
2082 * In these cases, the format always takes precedence, because the
2083 * format indicates the current state of the fork.
2084 */
2085/*ARGSUSED*/
2086STATIC void
2087xfs_iflush_fork(
2088        xfs_inode_t             *ip,
2089        xfs_dinode_t            *dip,
2090        xfs_inode_log_item_t    *iip,
2091        int                     whichfork,
2092        xfs_buf_t               *bp)
2093{
2094        char                    *cp;
2095        xfs_ifork_t             *ifp;
2096        xfs_mount_t             *mp;
2097#ifdef XFS_TRANS_DEBUG
2098        int                     first;
2099#endif
2100        static const short      brootflag[2] =
2101                { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2102        static const short      dataflag[2] =
2103                { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2104        static const short      extflag[2] =
2105                { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2106
2107        if (!iip)
2108                return;
2109        ifp = XFS_IFORK_PTR(ip, whichfork);
2110        /*
2111         * This can happen if we gave up in iformat in an error path,
2112         * for the attribute fork.
2113         */
2114        if (!ifp) {
2115                ASSERT(whichfork == XFS_ATTR_FORK);
2116                return;
2117        }
2118        cp = XFS_DFORK_PTR(dip, whichfork);
2119        mp = ip->i_mount;
2120        switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2121        case XFS_DINODE_FMT_LOCAL:
2122                if ((iip->ili_fields & dataflag[whichfork]) &&
2123                    (ifp->if_bytes > 0)) {
2124                        ASSERT(ifp->if_u1.if_data != NULL);
2125                        ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2126                        memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2127                }
2128                break;
2129
2130        case XFS_DINODE_FMT_EXTENTS:
2131                ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2132                       !(iip->ili_fields & extflag[whichfork]));
2133                if ((iip->ili_fields & extflag[whichfork]) &&
2134                    (ifp->if_bytes > 0)) {
2135                        ASSERT(xfs_iext_get_ext(ifp, 0));
2136                        ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2137                        (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2138                                whichfork);
2139                }
2140                break;
2141
2142        case XFS_DINODE_FMT_BTREE:
2143                if ((iip->ili_fields & brootflag[whichfork]) &&
2144                    (ifp->if_broot_bytes > 0)) {
2145                        ASSERT(ifp->if_broot != NULL);
2146                        ASSERT(ifp->if_broot_bytes <=
2147                               (XFS_IFORK_SIZE(ip, whichfork) +
2148                                XFS_BROOT_SIZE_ADJ));
2149                        xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
2150                                (xfs_bmdr_block_t *)cp,
2151                                XFS_DFORK_SIZE(dip, mp, whichfork));
2152                }
2153                break;
2154
2155        case XFS_DINODE_FMT_DEV:
2156                if (iip->ili_fields & XFS_ILOG_DEV) {
2157                        ASSERT(whichfork == XFS_DATA_FORK);
2158                        xfs_dinode_put_rdev(dip, ip->i_df.if_u2.if_rdev);
2159                }
2160                break;
2161
2162        case XFS_DINODE_FMT_UUID:
2163                if (iip->ili_fields & XFS_ILOG_UUID) {
2164                        ASSERT(whichfork == XFS_DATA_FORK);
2165                        memcpy(XFS_DFORK_DPTR(dip),
2166                               &ip->i_df.if_u2.if_uuid,
2167                               sizeof(uuid_t));
2168                }
2169                break;
2170
2171        default:
2172                ASSERT(0);
2173                break;
2174        }
2175}
2176
2177STATIC int
2178xfs_iflush_cluster(
2179        xfs_inode_t     *ip,
2180        xfs_buf_t       *bp)
2181{
2182        xfs_mount_t             *mp = ip->i_mount;
2183        struct xfs_perag        *pag;
2184        unsigned long           first_index, mask;
2185        unsigned long           inodes_per_cluster;
2186        int                     ilist_size;
2187        xfs_inode_t             **ilist;
2188        xfs_inode_t             *iq;
2189        int                     nr_found;
2190        int                     clcount = 0;
2191        int                     bufwasdelwri;
2192        int                     i;
2193
2194        pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2195
2196        inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2197        ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
2198        ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
2199        if (!ilist)
2200                goto out_put;
2201
2202        mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
2203        first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
2204        rcu_read_lock();
2205        /* really need a gang lookup range call here */
2206        nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
2207                                        first_index, inodes_per_cluster);
2208        if (nr_found == 0)
2209                goto out_free;
2210
2211        for (i = 0; i < nr_found; i++) {
2212                iq = ilist[i];
2213                if (iq == ip)
2214                        continue;
2215
2216                /*
2217                 * because this is an RCU protected lookup, we could find a
2218                 * recently freed or even reallocated inode during the lookup.
2219                 * We need to check under the i_flags_lock for a valid inode
2220                 * here. Skip it if it is not valid or the wrong inode.
2221                 */
2222                spin_lock(&ip->i_flags_lock);
2223                if (!ip->i_ino ||
2224                    (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
2225                        spin_unlock(&ip->i_flags_lock);
2226                        continue;
2227                }
2228                spin_unlock(&ip->i_flags_lock);
2229
2230                /*
2231                 * Do an un-protected check to see if the inode is dirty and
2232                 * is a candidate for flushing.  These checks will be repeated
2233                 * later after the appropriate locks are acquired.
2234                 */
2235                if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
2236                        continue;
2237
2238                /*
2239                 * Try to get locks.  If any are unavailable or it is pinned,
2240                 * then this inode cannot be flushed and is skipped.
2241                 */
2242
2243                if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
2244                        continue;
2245                if (!xfs_iflock_nowait(iq)) {
2246                        xfs_iunlock(iq, XFS_ILOCK_SHARED);
2247                        continue;
2248                }
2249                if (xfs_ipincount(iq)) {
2250                        xfs_ifunlock(iq);
2251                        xfs_iunlock(iq, XFS_ILOCK_SHARED);
2252                        continue;
2253                }
2254
2255                /*
2256                 * arriving here means that this inode can be flushed.  First
2257                 * re-check that it's dirty before flushing.
2258                 */
2259                if (!xfs_inode_clean(iq)) {
2260                        int     error;
2261                        error = xfs_iflush_int(iq, bp);
2262                        if (error) {
2263                                xfs_iunlock(iq, XFS_ILOCK_SHARED);
2264                                goto cluster_corrupt_out;
2265                        }
2266                        clcount++;
2267                } else {
2268                        xfs_ifunlock(iq);
2269                }
2270                xfs_iunlock(iq, XFS_ILOCK_SHARED);
2271        }
2272
2273        if (clcount) {
2274                XFS_STATS_INC(xs_icluster_flushcnt);
2275                XFS_STATS_ADD(xs_icluster_flushinode, clcount);
2276        }
2277
2278out_free:
2279        rcu_read_unlock();
2280        kmem_free(ilist);
2281out_put:
2282        xfs_perag_put(pag);
2283        return 0;
2284
2285
2286cluster_corrupt_out:
2287        /*
2288         * Corruption detected in the clustering loop.  Invalidate the
2289         * inode buffer and shut down the filesystem.
2290         */
2291        rcu_read_unlock();
2292        /*
2293         * Clean up the buffer.  If it was delwri, just release it --
2294         * brelse can handle it with no problems.  If not, shut down the
2295         * filesystem before releasing the buffer.
2296         */
2297        bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
2298        if (bufwasdelwri)
2299                xfs_buf_relse(bp);
2300
2301        xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2302
2303        if (!bufwasdelwri) {
2304                /*
2305                 * Just like incore_relse: if we have b_iodone functions,
2306                 * mark the buffer as an error and call them.  Otherwise
2307                 * mark it as stale and brelse.
2308                 */
2309                if (bp->b_iodone) {
2310                        XFS_BUF_UNDONE(bp);
2311                        xfs_buf_stale(bp);
2312                        xfs_buf_ioerror(bp, EIO);
2313                        xfs_buf_ioend(bp, 0);
2314                } else {
2315                        xfs_buf_stale(bp);
2316                        xfs_buf_relse(bp);
2317                }
2318        }
2319
2320        /*
2321         * Unlocks the flush lock
2322         */
2323        xfs_iflush_abort(iq, false);
2324        kmem_free(ilist);
2325        xfs_perag_put(pag);
2326        return XFS_ERROR(EFSCORRUPTED);
2327}
2328
2329/*
2330 * Flush dirty inode metadata into the backing buffer.
2331 *
2332 * The caller must have the inode lock and the inode flush lock held.  The
2333 * inode lock will still be held upon return to the caller, and the inode
2334 * flush lock will be released after the inode has reached the disk.
2335 *
2336 * The caller must write out the buffer returned in *bpp and release it.
2337 */
2338int
2339xfs_iflush(
2340        struct xfs_inode        *ip,
2341        struct xfs_buf          **bpp)
2342{
2343        struct xfs_mount        *mp = ip->i_mount;
2344        struct xfs_buf          *bp;
2345        struct xfs_dinode       *dip;
2346        int                     error;
2347
2348        XFS_STATS_INC(xs_iflush_count);
2349
2350        ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2351        ASSERT(xfs_isiflocked(ip));
2352        ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2353               ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
2354
2355        *bpp = NULL;
2356
2357        xfs_iunpin_wait(ip);
2358
2359        /*
2360         * For stale inodes we cannot rely on the backing buffer remaining
2361         * stale in cache for the remaining life of the stale inode and so
2362         * xfs_imap_to_bp() below may give us a buffer that no longer contains
2363         * inodes below. We have to check this after ensuring the inode is
2364         * unpinned so that it is safe to reclaim the stale inode after the
2365         * flush call.
2366         */
2367        if (xfs_iflags_test(ip, XFS_ISTALE)) {
2368                xfs_ifunlock(ip);
2369                return 0;
2370        }
2371
2372        /*
2373         * This may have been unpinned because the filesystem is shutting
2374         * down forcibly. If that's the case we must not write this inode
2375         * to disk, because the log record didn't make it to disk.
2376         *
2377         * We also have to remove the log item from the AIL in this case,
2378         * as we wait for an empty AIL as part of the unmount process.
2379         */
2380        if (XFS_FORCED_SHUTDOWN(mp)) {
2381                error = XFS_ERROR(EIO);
2382                goto abort_out;
2383        }
2384
2385        /*
2386         * Get the buffer containing the on-disk inode.
2387         */
2388        error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
2389                               0);
2390        if (error || !bp) {
2391                xfs_ifunlock(ip);
2392                return error;
2393        }
2394
2395        /*
2396         * First flush out the inode that xfs_iflush was called with.
2397         */
2398        error = xfs_iflush_int(ip, bp);
2399        if (error)
2400                goto corrupt_out;
2401
2402        /*
2403         * If the buffer is pinned then push on the log now so we won't
2404         * get stuck waiting in the write for too long.
2405         */
2406        if (xfs_buf_ispinned(bp))
2407                xfs_log_force(mp, 0);
2408
2409        /*
2410         * inode clustering:
2411         * see if other inodes can be gathered into this write
2412         */
2413        error = xfs_iflush_cluster(ip, bp);
2414        if (error)
2415                goto cluster_corrupt_out;
2416
2417        *bpp = bp;
2418        return 0;
2419
2420corrupt_out:
2421        xfs_buf_relse(bp);
2422        xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2423cluster_corrupt_out:
2424        error = XFS_ERROR(EFSCORRUPTED);
2425abort_out:
2426        /*
2427         * Unlocks the flush lock
2428         */
2429        xfs_iflush_abort(ip, false);
2430        return error;
2431}
2432
2433
2434STATIC int
2435xfs_iflush_int(
2436        xfs_inode_t             *ip,
2437        xfs_buf_t               *bp)
2438{
2439        xfs_inode_log_item_t    *iip;
2440        xfs_dinode_t            *dip;
2441        xfs_mount_t             *mp;
2442#ifdef XFS_TRANS_DEBUG
2443        int                     first;
2444#endif
2445
2446        ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2447        ASSERT(xfs_isiflocked(ip));
2448        ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2449               ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
2450
2451        iip = ip->i_itemp;
2452        mp = ip->i_mount;
2453
2454        /* set *dip = inode's place in the buffer */
2455        dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
2456
2457        if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
2458                               mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
2459                xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2460                        "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
2461                        __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
2462                goto corrupt_out;
2463        }
2464        if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
2465                                mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
2466                xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2467                        "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
2468                        __func__, ip->i_ino, ip, ip->i_d.di_magic);
2469                goto corrupt_out;
2470        }
2471        if (S_ISREG(ip->i_d.di_mode)) {
2472                if (XFS_TEST_ERROR(
2473                    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2474                    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
2475                    mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
2476                        xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2477                                "%s: Bad regular inode %Lu, ptr 0x%p",
2478                                __func__, ip->i_ino, ip);
2479                        goto corrupt_out;
2480                }
2481        } else if (S_ISDIR(ip->i_d.di_mode)) {
2482                if (XFS_TEST_ERROR(
2483                    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2484                    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
2485                    (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
2486                    mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
2487                        xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2488                                "%s: Bad directory inode %Lu, ptr 0x%p",
2489                                __func__, ip->i_ino, ip);
2490                        goto corrupt_out;
2491                }
2492        }
2493        if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
2494                                ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
2495                                XFS_RANDOM_IFLUSH_5)) {
2496                xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2497                        "%s: detected corrupt incore inode %Lu, "
2498                        "total extents = %d, nblocks = %Ld, ptr 0x%p",
2499                        __func__, ip->i_ino,
2500                        ip->i_d.di_nextents + ip->i_d.di_anextents,
2501                        ip->i_d.di_nblocks, ip);
2502                goto corrupt_out;
2503        }
2504        if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
2505                                mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
2506                xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2507                        "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
2508                        __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
2509                goto corrupt_out;
2510        }
2511        /*
2512         * bump the flush iteration count, used to detect flushes which
2513         * postdate a log record during recovery.
2514         */
2515
2516        ip->i_d.di_flushiter++;
2517
2518        /*
2519         * Copy the dirty parts of the inode into the on-disk
2520         * inode.  We always copy out the core of the inode,
2521         * because if the inode is dirty at all the core must
2522         * be.
2523         */
2524        xfs_dinode_to_disk(dip, &ip->i_d);
2525
2526        /* Wrap, we never let the log put out DI_MAX_FLUSH */
2527        if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
2528                ip->i_d.di_flushiter = 0;
2529
2530        /*
2531         * If this is really an old format inode and the superblock version
2532         * has not been updated to support only new format inodes, then
2533         * convert back to the old inode format.  If the superblock version
2534         * has been updated, then make the conversion permanent.
2535         */
2536        ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
2537        if (ip->i_d.di_version == 1) {
2538                if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
2539                        /*
2540                         * Convert it back.
2541                         */
2542                        ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
2543                        dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink);
2544                } else {
2545                        /*
2546                         * The superblock version has already been bumped,
2547                         * so just make the conversion to the new inode
2548                         * format permanent.
2549                         */
2550                        ip->i_d.di_version = 2;
2551                        dip->di_version = 2;
2552                        ip->i_d.di_onlink = 0;
2553                        dip->di_onlink = 0;
2554                        memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
2555                        memset(&(dip->di_pad[0]), 0,
2556                              sizeof(dip->di_pad));
2557                        ASSERT(xfs_get_projid(ip) == 0);
2558                }
2559        }
2560
2561        xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
2562        if (XFS_IFORK_Q(ip))
2563                xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
2564        xfs_inobp_check(mp, bp);
2565
2566        /*
2567         * We've recorded everything logged in the inode, so we'd like to clear
2568         * the ili_fields bits so we don't log and flush things unnecessarily.
2569         * However, we can't stop logging all this information until the data
2570         * we've copied into the disk buffer is written to disk.  If we did we
2571         * might overwrite the copy of the inode in the log with all the data
2572         * after re-logging only part of it, and in the face of a crash we
2573         * wouldn't have all the data we need to recover.
2574         *
2575         * What we do is move the bits to the ili_last_fields field.  When
2576         * logging the inode, these bits are moved back to the ili_fields field.
2577         * In the xfs_iflush_done() routine we clear ili_last_fields, since we
2578         * know that the information those bits represent is permanently on
2579         * disk.  As long as the flush completes before the inode is logged
2580         * again, then both ili_fields and ili_last_fields will be cleared.
2581         *
2582         * We can play with the ili_fields bits here, because the inode lock
2583         * must be held exclusively in order to set bits there and the flush
2584         * lock protects the ili_last_fields bits.  Set ili_logged so the flush
2585         * done routine can tell whether or not to look in the AIL.  Also, store
2586         * the current LSN of the inode so that we can tell whether the item has
2587         * moved in the AIL from xfs_iflush_done().  In order to read the lsn we
2588         * need the AIL lock, because it is a 64 bit value that cannot be read
2589         * atomically.
2590         */
2591        if (iip != NULL && iip->ili_fields != 0) {
2592                iip->ili_last_fields = iip->ili_fields;
2593                iip->ili_fields = 0;
2594                iip->ili_logged = 1;
2595
2596                xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2597                                        &iip->ili_item.li_lsn);
2598
2599                /*
2600                 * Attach the function xfs_iflush_done to the inode's
2601                 * buffer.  This will remove the inode from the AIL
2602                 * and unlock the inode's flush lock when the inode is
2603                 * completely written to disk.
2604                 */
2605                xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
2606
2607                ASSERT(bp->b_fspriv != NULL);
2608                ASSERT(bp->b_iodone != NULL);
2609        } else {
2610                /*
2611                 * We're flushing an inode which is not in the AIL and has
2612                 * not been logged.  For this case we can immediately drop
2613                 * the inode flush lock because we can avoid the whole
2614                 * AIL state thing.  It's OK to drop the flush lock now,
2615                 * because we've already locked the buffer and to do anything
2616                 * you really need both.
2617                 */
2618                if (iip != NULL) {
2619                        ASSERT(iip->ili_logged == 0);
2620                        ASSERT(iip->ili_last_fields == 0);
2621                        ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
2622                }
2623                xfs_ifunlock(ip);
2624        }
2625
2626        return 0;
2627
2628corrupt_out:
2629        return XFS_ERROR(EFSCORRUPTED);
2630}
2631
2632/*
2633 * Return a pointer to the extent record at file index idx.
2634 */
2635xfs_bmbt_rec_host_t *
2636xfs_iext_get_ext(
2637        xfs_ifork_t     *ifp,           /* inode fork pointer */
2638        xfs_extnum_t    idx)            /* index of target extent */
2639{
2640        ASSERT(idx >= 0);
2641        ASSERT(idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t));
2642
2643        if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
2644                return ifp->if_u1.if_ext_irec->er_extbuf;
2645        } else if (ifp->if_flags & XFS_IFEXTIREC) {
2646                xfs_ext_irec_t  *erp;           /* irec pointer */
2647                int             erp_idx = 0;    /* irec index */
2648                xfs_extnum_t    page_idx = idx; /* ext index in target list */
2649
2650                erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
2651                return &erp->er_extbuf[page_idx];
2652        } else if (ifp->if_bytes) {
2653                return &ifp->if_u1.if_extents[idx];
2654        } else {
2655                return NULL;
2656        }
2657}
2658
2659/*
2660 * Insert new item(s) into the extent records for incore inode
2661 * fork 'ifp'.  'count' new items are inserted at index 'idx'.
2662 */
2663void
2664xfs_iext_insert(
2665        xfs_inode_t     *ip,            /* incore inode pointer */
2666        xfs_extnum_t    idx,            /* starting index of new items */
2667        xfs_extnum_t    count,          /* number of inserted items */
2668        xfs_bmbt_irec_t *new,           /* items to insert */
2669        int             state)          /* type of extent conversion */
2670{
2671        xfs_ifork_t     *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
2672        xfs_extnum_t    i;              /* extent record index */
2673
2674        trace_xfs_iext_insert(ip, idx, new, state, _RET_IP_);
2675
2676        ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2677        xfs_iext_add(ifp, idx, count);
2678        for (i = idx; i < idx + count; i++, new++)
2679                xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
2680}
2681
2682/*
2683 * This is called when the amount of space required for incore file
2684 * extents needs to be increased. The ext_diff parameter stores the
2685 * number of new extents being added and the idx parameter contains
2686 * the extent index where the new extents will be added. If the new
2687 * extents are being appended, then we just need to (re)allocate and
2688 * initialize the space. Otherwise, if the new extents are being
2689 * inserted into the middle of the existing entries, a bit more work
2690 * is required to make room for the new extents to be inserted. The
2691 * caller is responsible for filling in the new extent entries upon
2692 * return.
2693 */
2694void
2695xfs_iext_add(
2696        xfs_ifork_t     *ifp,           /* inode fork pointer */
2697        xfs_extnum_t    idx,            /* index to begin adding exts */
2698        int             ext_diff)       /* number of extents to add */
2699{
2700        int             byte_diff;      /* new bytes being added */
2701        int             new_size;       /* size of extents after adding */
2702        xfs_extnum_t    nextents;       /* number of extents in file */
2703
2704        nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2705        ASSERT((idx >= 0) && (idx <= nextents));
2706        byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
2707        new_size = ifp->if_bytes + byte_diff;
2708        /*
2709         * If the new number of extents (nextents + ext_diff)
2710         * fits inside the inode, then continue to use the inline
2711         * extent buffer.
2712         */
2713        if (nextents + ext_diff <= XFS_INLINE_EXTS) {
2714                if (idx < nextents) {
2715                        memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
2716                                &ifp->if_u2.if_inline_ext[idx],
2717                                (nextents - idx) * sizeof(xfs_bmbt_rec_t));
2718                        memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
2719                }
2720                ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
2721                ifp->if_real_bytes = 0;
2722        }
2723        /*
2724         * Otherwise use a linear (direct) extent list.
2725         * If the extents are currently inside the inode,
2726         * xfs_iext_realloc_direct will switch us from
2727         * inline to direct extent allocation mode.
2728         */
2729        else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
2730                xfs_iext_realloc_direct(ifp, new_size);
2731                if (idx < nextents) {
2732                        memmove(&ifp->if_u1.if_extents[idx + ext_diff],
2733                                &ifp->if_u1.if_extents[idx],
2734                                (nextents - idx) * sizeof(xfs_bmbt_rec_t));
2735                        memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
2736                }
2737        }
2738        /* Indirection array */
2739        else {
2740                xfs_ext_irec_t  *erp;
2741                int             erp_idx = 0;
2742                int             page_idx = idx;
2743
2744                ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
2745                if (ifp->if_flags & XFS_IFEXTIREC) {
2746                        erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
2747                } else {
2748                        xfs_iext_irec_init(ifp);
2749                        ASSERT(ifp->if_flags & XFS_IFEXTIREC);
2750                        erp = ifp->if_u1.if_ext_irec;
2751                }
2752                /* Extents fit in target extent page */
2753                if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
2754                        if (page_idx < erp->er_extcount) {
2755                                memmove(&erp->er_extbuf[page_idx + ext_diff],
2756                                        &erp->er_extbuf[page_idx],
2757                                        (erp->er_extcount - page_idx) *
2758                                        sizeof(xfs_bmbt_rec_t));
2759                                memset(&erp->er_extbuf[page_idx], 0, byte_diff);
2760                        }
2761                        erp->er_extcount += ext_diff;
2762                        xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
2763                }
2764                /* Insert a new extent page */
2765                else if (erp) {
2766                        xfs_iext_add_indirect_multi(ifp,
2767                                erp_idx, page_idx, ext_diff);
2768                }
2769                /*
2770                 * If extent(s) are being appended to the last page in
2771                 * the indirection array and the new extent(s) don't fit
2772                 * in the page, then erp is NULL and erp_idx is set to
2773                 * the next index needed in the indirection array.
2774                 */
2775                else {
2776                        int     count = ext_diff;
2777
2778                        while (count) {
2779                                erp = xfs_iext_irec_new(ifp, erp_idx);
2780                                erp->er_extcount = count;
2781                                count -= MIN(count, (int)XFS_LINEAR_EXTS);
2782                                if (count) {
2783                                        erp_idx++;
2784                                }
2785                        }
2786                }
2787        }
2788        ifp->if_bytes = new_size;
2789}
2790
2791/*
2792 * This is called when incore extents are being added to the indirection
2793 * array and the new extents do not fit in the target extent list. The
2794 * erp_idx parameter contains the irec index for the target extent list
2795 * in the indirection array, and the idx parameter contains the extent
2796 * index within the list. The number of extents being added is stored
2797 * in the count parameter.
2798 *
2799 *    |-------|   |-------|
2800 *    |       |   |       |    idx - number of extents before idx
2801 *    |  idx  |   | count |
2802 *    |       |   |       |    count - number of extents being inserted at idx
2803 *    |-------|   |-------|
2804 *    | count |   | nex2  |    nex2 - number of extents after idx + count
2805 *    |-------|   |-------|
2806 */
2807void
2808xfs_iext_add_indirect_multi(
2809        xfs_ifork_t     *ifp,                   /* inode fork pointer */
2810        int             erp_idx,                /* target extent irec index */
2811        xfs_extnum_t    idx,                    /* index within target list */
2812        int             count)                  /* new extents being added */
2813{
2814        int             byte_diff;              /* new bytes being added */
2815        xfs_ext_irec_t  *erp;                   /* pointer to irec entry */
2816        xfs_extnum_t    ext_diff;               /* number of extents to add */
2817        xfs_extnum_t    ext_cnt;                /* new extents still needed */
2818        xfs_extnum_t    nex2;                   /* extents after idx + count */
2819        xfs_bmbt_rec_t  *nex2_ep = NULL;        /* temp list for nex2 extents */
2820        int             nlists;                 /* number of irec's (lists) */
2821
2822        ASSERT(ifp->if_flags & XFS_IFEXTIREC);
2823        erp = &ifp->if_u1.if_ext_irec[erp_idx];
2824        nex2 = erp->er_extcount - idx;
2825        nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
2826
2827        /*
2828         * Save second part of target extent list
2829         * (all extents past */
2830        if (nex2) {
2831                byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
2832                nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS);
2833                memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
2834                erp->er_extcount -= nex2;
2835                xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
2836                memset(&erp->er_extbuf[idx], 0, byte_diff);
2837        }
2838
2839        /*
2840         * Add the new extents to the end of the target
2841         * list, then allocate new irec record(s) and
2842         * extent buffer(s) as needed to store the rest
2843         * of the new extents.
2844         */
2845        ext_cnt = count;
2846        ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
2847        if (ext_diff) {
2848                erp->er_extcount += ext_diff;
2849                xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
2850                ext_cnt -= ext_diff;
2851        }
2852        while (ext_cnt) {
2853                erp_idx++;
2854                erp = xfs_iext_irec_new(ifp, erp_idx);
2855                ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
2856                erp->er_extcount = ext_diff;
2857                xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
2858                ext_cnt -= ext_diff;
2859        }
2860
2861        /* Add nex2 extents back to indirection array */
2862        if (nex2) {
2863                xfs_extnum_t    ext_avail;
2864                int             i;
2865
2866                byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
2867                ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
2868                i = 0;
2869                /*
2870                 * If nex2 extents fit in the current page, append
2871                 * nex2_ep after the new extents.
2872                 */
2873                if (nex2 <= ext_avail) {
2874                        i = erp->er_extcount;
2875                }
2876                /*
2877                 * Otherwise, check if space is available in the
2878                 * next page.
2879                 */
2880                else if ((erp_idx < nlists - 1) &&
2881                         (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
2882                          ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
2883                        erp_idx++;
2884                        erp++;
2885                        /* Create a hole for nex2 extents */
2886                        memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
2887                                erp->er_extcount * sizeof(xfs_bmbt_rec_t));
2888                }
2889                /*
2890                 * Final choice, create a new extent page for
2891                 * nex2 extents.
2892                 */
2893                else {
2894                        erp_idx++;
2895                        erp = xfs_iext_irec_new(ifp, erp_idx);
2896                }
2897                memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
2898                kmem_free(nex2_ep);
2899                erp->er_extcount += nex2;
2900                xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
2901        }
2902}
2903
2904/*
2905 * This is called when the amount of space required for incore file
2906 * extents needs to be decreased. The ext_diff parameter stores the
2907 * number of extents to be removed and the idx parameter contains
2908 * the extent index where the extents will be removed from.
2909 *
2910 * If the amount of space needed has decreased below the linear
2911 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
2912 * extent array.  Otherwise, use kmem_realloc() to adjust the
2913 * size to what is needed.
2914 */
2915void
2916xfs_iext_remove(
2917        xfs_inode_t     *ip,            /* incore inode pointer */
2918        xfs_extnum_t    idx,            /* index to begin removing exts */
2919        int             ext_diff,       /* number of extents to remove */
2920        int             state)          /* type of extent conversion */
2921{
2922        xfs_ifork_t     *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
2923        xfs_extnum_t    nextents;       /* number of extents in file */
2924        int             new_size;       /* size of extents after removal */
2925
2926        trace_xfs_iext_remove(ip, idx, state, _RET_IP_);
2927
2928        ASSERT(ext_diff > 0);
2929        nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2930        new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
2931
2932        if (new_size == 0) {
2933                xfs_iext_destroy(ifp);
2934        } else if (ifp->if_flags & XFS_IFEXTIREC) {
2935                xfs_iext_remove_indirect(ifp, idx, ext_diff);
2936        } else if (ifp->if_real_bytes) {
2937                xfs_iext_remove_direct(ifp, idx, ext_diff);
2938        } else {
2939                xfs_iext_remove_inline(ifp, idx, ext_diff);
2940        }
2941        ifp->if_bytes = new_size;
2942}
2943
2944/*
2945 * This removes ext_diff extents from the inline buffer, beginning
2946 * at extent index idx.
2947 */
2948void
2949xfs_iext_remove_inline(
2950        xfs_ifork_t     *ifp,           /* inode fork pointer */
2951        xfs_extnum_t    idx,            /* index to begin removing exts */
2952        int             ext_diff)       /* number of extents to remove */
2953{
2954        int             nextents;       /* number of extents in file */
2955
2956        ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
2957        ASSERT(idx < XFS_INLINE_EXTS);
2958        nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2959        ASSERT(((nextents - ext_diff) > 0) &&
2960                (nextents - ext_diff) < XFS_INLINE_EXTS);
2961
2962        if (idx + ext_diff < nextents) {
2963                memmove(&ifp->if_u2.if_inline_ext[idx],
2964                        &ifp->if_u2.if_inline_ext[idx + ext_diff],
2965                        (nextents - (idx + ext_diff)) *
2966                         sizeof(xfs_bmbt_rec_t));
2967                memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
2968                        0, ext_diff * sizeof(xfs_bmbt_rec_t));
2969        } else {
2970                memset(&ifp->if_u2.if_inline_ext[idx], 0,
2971                        ext_diff * sizeof(xfs_bmbt_rec_t));
2972        }
2973}
2974
2975/*
2976 * This removes ext_diff extents from a linear (direct) extent list,
2977 * beginning at extent index idx. If the extents are being removed
2978 * from the end of the list (ie. truncate) then we just need to re-
2979 * allocate the list to remove the extra space. Otherwise, if the
2980 * extents are being removed from the middle of the existing extent
2981 * entries, then we first need to move the extent records beginning
2982 * at idx + ext_diff up in the list to overwrite the records being
2983 * removed, then remove the extra space via kmem_realloc.
2984 */
2985void
2986xfs_iext_remove_direct(
2987        xfs_ifork_t     *ifp,           /* inode fork pointer */
2988        xfs_extnum_t    idx,            /* index to begin removing exts */
2989        int             ext_diff)       /* number of extents to remove */
2990{
2991        xfs_extnum_t    nextents;       /* number of extents in file */
2992        int             new_size;       /* size of extents after removal */
2993
2994        ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
2995        new_size = ifp->if_bytes -
2996                (ext_diff * sizeof(xfs_bmbt_rec_t));
2997        nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2998
2999        if (new_size == 0) {
3000                xfs_iext_destroy(ifp);
3001                return;
3002        }
3003        /* Move extents up in the list (if needed) */
3004        if (idx + ext_diff < nextents) {
3005                memmove(&ifp->if_u1.if_extents[idx],
3006                        &ifp->if_u1.if_extents[idx + ext_diff],
3007                        (nextents - (idx + ext_diff)) *
3008                         sizeof(xfs_bmbt_rec_t));
3009        }
3010        memset(&ifp->if_u1.if_extents[nextents - ext_diff],
3011                0, ext_diff * sizeof(xfs_bmbt_rec_t));
3012        /*
3013         * Reallocate the direct extent list. If the extents
3014         * will fit inside the inode then xfs_iext_realloc_direct
3015         * will switch from direct to inline extent allocation
3016         * mode for us.
3017         */
3018        xfs_iext_realloc_direct(ifp, new_size);
3019        ifp->if_bytes = new_size;
3020}
3021
3022/*
3023 * This is called when incore extents are being removed from the
3024 * indirection array and the extents being removed span multiple extent
3025 * buffers. The idx parameter contains the file extent index where we
3026 * want to begin removing extents, and the count parameter contains
3027 * how many extents need to be removed.
3028 *
3029 *    |-------|   |-------|
3030 *    | nex1  |   |       |    nex1 - number of extents before idx
3031 *    |-------|   | count |
3032 *    |       |   |       |    count - number of extents being removed at idx
3033 *    | count |   |-------|
3034 *    |       |   | nex2  |    nex2 - number of extents after idx + count
3035 *    |-------|   |-------|
3036 */
3037void
3038xfs_iext_remove_indirect(
3039        xfs_ifork_t     *ifp,           /* inode fork pointer */
3040        xfs_extnum_t    idx,            /* index to begin removing extents */
3041        int             count)          /* number of extents to remove */
3042{
3043        xfs_ext_irec_t  *erp;           /* indirection array pointer */
3044        int             erp_idx = 0;    /* indirection array index */
3045        xfs_extnum_t    ext_cnt;        /* extents left to remove */
3046        xfs_extnum_t    ext_diff;       /* extents to remove in current list */
3047        xfs_extnum_t    nex1;           /* number of extents before idx */
3048        xfs_extnum_t    nex2;           /* extents after idx + count */
3049        int             page_idx = idx; /* index in target extent list */
3050
3051        ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3052        erp = xfs_iext_idx_to_irec(ifp,  &page_idx, &erp_idx, 0);
3053        ASSERT(erp != NULL);
3054        nex1 = page_idx;
3055        ext_cnt = count;
3056        while (ext_cnt) {
3057                nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
3058                ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
3059                /*
3060                 * Check for deletion of entire list;
3061                 * xfs_iext_irec_remove() updates extent offsets.
3062                 */
3063                if (ext_diff == erp->er_extcount) {
3064                        xfs_iext_irec_remove(ifp, erp_idx);
3065                        ext_cnt -= ext_diff;
3066                        nex1 = 0;
3067                        if (ext_cnt) {
3068                                ASSERT(erp_idx < ifp->if_real_bytes /
3069                                        XFS_IEXT_BUFSZ);
3070                                erp = &ifp->if_u1.if_ext_irec[erp_idx];
3071                                nex1 = 0;
3072                                continue;
3073                        } else {
3074                                break;
3075                        }
3076                }
3077                /* Move extents up (if needed) */
3078                if (nex2) {
3079                        memmove(&erp->er_extbuf[nex1],
3080                                &erp->er_extbuf[nex1 + ext_diff],
3081                                nex2 * sizeof(xfs_bmbt_rec_t));
3082                }
3083                /* Zero out rest of page */
3084                memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
3085                        ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
3086                /* Update remaining counters */
3087                erp->er_extcount -= ext_diff;
3088                xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
3089                ext_cnt -= ext_diff;
3090                nex1 = 0;
3091                erp_idx++;
3092                erp++;
3093        }
3094        ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
3095        xfs_iext_irec_compact(ifp);
3096}
3097
3098/*
3099 * Create, destroy, or resize a linear (direct) block of extents.
3100 */
3101void
3102xfs_iext_realloc_direct(
3103        xfs_ifork_t     *ifp,           /* inode fork pointer */
3104        int             new_size)       /* new size of extents */
3105{
3106        int             rnew_size;      /* real new size of extents */
3107
3108        rnew_size = new_size;
3109
3110        ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
3111                ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
3112                 (new_size != ifp->if_real_bytes)));
3113
3114        /* Free extent records */
3115        if (new_size == 0) {
3116                xfs_iext_destroy(ifp);
3117        }
3118        /* Resize direct extent list and zero any new bytes */
3119        else if (ifp->if_real_bytes) {
3120                /* Check if extents will fit inside the inode */
3121                if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
3122                        xfs_iext_direct_to_inline(ifp, new_size /
3123                                (uint)sizeof(xfs_bmbt_rec_t));
3124                        ifp->if_bytes = new_size;
3125                        return;
3126                }
3127                if (!is_power_of_2(new_size)){
3128                        rnew_size = roundup_pow_of_two(new_size);
3129                }
3130                if (rnew_size != ifp->if_real_bytes) {
3131                        ifp->if_u1.if_extents =
3132                                kmem_realloc(ifp->if_u1.if_extents,
3133                                                rnew_size,
3134                                                ifp->if_real_bytes, KM_NOFS);
3135                }
3136                if (rnew_size > ifp->if_real_bytes) {
3137                        memset(&ifp->if_u1.if_extents[ifp->if_bytes /
3138                                (uint)sizeof(xfs_bmbt_rec_t)], 0,
3139                                rnew_size - ifp->if_real_bytes);
3140                }
3141        }
3142        /*
3143         * Switch from the inline extent buffer to a direct
3144         * extent list. Be sure to include the inline extent
3145         * bytes in new_size.
3146         */
3147        else {
3148                new_size += ifp->if_bytes;
3149                if (!is_power_of_2(new_size)) {
3150                        rnew_size = roundup_pow_of_two(new_size);
3151                }
3152                xfs_iext_inline_to_direct(ifp, rnew_size);
3153        }
3154        ifp->if_real_bytes = rnew_size;
3155        ifp->if_bytes = new_size;
3156}
3157
3158/*
3159 * Switch from linear (direct) extent records to inline buffer.
3160 */
3161void
3162xfs_iext_direct_to_inline(
3163        xfs_ifork_t     *ifp,           /* inode fork pointer */
3164        xfs_extnum_t    nextents)       /* number of extents in file */
3165{
3166        ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3167        ASSERT(nextents <= XFS_INLINE_EXTS);
3168        /*
3169         * The inline buffer was zeroed when we switched
3170         * from inline to direct extent allocation mode,
3171         * so we don't need to clear it here.
3172         */
3173        memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
3174                nextents * sizeof(xfs_bmbt_rec_t));
3175        kmem_free(ifp->if_u1.if_extents);
3176        ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3177        ifp->if_real_bytes = 0;
3178}
3179
3180/*
3181 * Switch from inline buffer to linear (direct) extent records.
3182 * new_size should already be rounded up to the next power of 2
3183 * by the caller (when appropriate), so use new_size as it is.
3184 * However, since new_size may be rounded up, we can't update
3185 * if_bytes here. It is the caller's responsibility to update
3186 * if_bytes upon return.
3187 */
3188void
3189xfs_iext_inline_to_direct(
3190        xfs_ifork_t     *ifp,           /* inode fork pointer */
3191        int             new_size)       /* number of extents in file */
3192{
3193        ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS);
3194        memset(ifp->if_u1.if_extents, 0, new_size);
3195        if (ifp->if_bytes) {
3196                memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
3197                        ifp->if_bytes);
3198                memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3199                        sizeof(xfs_bmbt_rec_t));
3200        }
3201        ifp->if_real_bytes = new_size;
3202}
3203
3204/*
3205 * Resize an extent indirection array to new_size bytes.
3206 */
3207STATIC void
3208xfs_iext_realloc_indirect(
3209        xfs_ifork_t     *ifp,           /* inode fork pointer */
3210        int             new_size)       /* new indirection array size */
3211{
3212        int             nlists;         /* number of irec's (ex lists) */
3213        int             size;           /* current indirection array size */
3214
3215        ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3216        nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3217        size = nlists * sizeof(xfs_ext_irec_t);
3218        ASSERT(ifp->if_real_bytes);
3219        ASSERT((new_size >= 0) && (new_size != size));
3220        if (new_size == 0) {
3221                xfs_iext_destroy(ifp);
3222        } else {
3223                ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
3224                        kmem_realloc(ifp->if_u1.if_ext_irec,
3225                                new_size, size, KM_NOFS);
3226        }
3227}
3228
3229/*
3230 * Switch from indirection array to linear (direct) extent allocations.
3231 */
3232STATIC void
3233xfs_iext_indirect_to_direct(
3234         xfs_ifork_t    *ifp)           /* inode fork pointer */
3235{
3236        xfs_bmbt_rec_host_t *ep;        /* extent record pointer */
3237        xfs_extnum_t    nextents;       /* number of extents in file */
3238        int             size;           /* size of file extents */
3239
3240        ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3241        nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3242        ASSERT(nextents <= XFS_LINEAR_EXTS);
3243        size = nextents * sizeof(xfs_bmbt_rec_t);
3244
3245        xfs_iext_irec_compact_pages(ifp);
3246        ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
3247
3248        ep = ifp->if_u1.if_ext_irec->er_extbuf;
3249        kmem_free(ifp->if_u1.if_ext_irec);
3250        ifp->if_flags &= ~XFS_IFEXTIREC;
3251        ifp->if_u1.if_extents = ep;
3252        ifp->if_bytes = size;
3253        if (nextents < XFS_LINEAR_EXTS) {
3254                xfs_iext_realloc_direct(ifp, size);
3255        }
3256}
3257
3258/*
3259 * Free incore file extents.
3260 */
3261void
3262xfs_iext_destroy(
3263        xfs_ifork_t     *ifp)           /* inode fork pointer */
3264{
3265        if (ifp->if_flags & XFS_IFEXTIREC) {
3266                int     erp_idx;
3267                int     nlists;
3268
3269                nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3270                for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
3271                        xfs_iext_irec_remove(ifp, erp_idx);
3272                }
3273                ifp->if_flags &= ~XFS_IFEXTIREC;
3274        } else if (ifp->if_real_bytes) {
3275                kmem_free(ifp->if_u1.if_extents);
3276        } else if (ifp->if_bytes) {
3277                memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3278                        sizeof(xfs_bmbt_rec_t));
3279        }
3280        ifp->if_u1.if_extents = NULL;
3281        ifp->if_real_bytes = 0;
3282        ifp->if_bytes = 0;
3283}
3284
3285/*
3286 * Return a pointer to the extent record for file system block bno.
3287 */
3288xfs_bmbt_rec_host_t *                   /* pointer to found extent record */
3289xfs_iext_bno_to_ext(
3290        xfs_ifork_t     *ifp,           /* inode fork pointer */
3291        xfs_fileoff_t   bno,            /* block number to search for */
3292        xfs_extnum_t    *idxp)          /* index of target extent */
3293{
3294        xfs_bmbt_rec_host_t *base;      /* pointer to first extent */
3295        xfs_filblks_t   blockcount = 0; /* number of blocks in extent */
3296        xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */
3297        xfs_ext_irec_t  *erp = NULL;    /* indirection array pointer */
3298        int             high;           /* upper boundary in search */
3299        xfs_extnum_t    idx = 0;        /* index of target extent */
3300        int             low;            /* lower boundary in search */
3301        xfs_extnum_t    nextents;       /* number of file extents */
3302        xfs_fileoff_t   startoff = 0;   /* start offset of extent */
3303
3304        nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3305        if (nextents == 0) {
3306                *idxp = 0;
3307                return NULL;
3308        }
3309        low = 0;
3310        if (ifp->if_flags & XFS_IFEXTIREC) {
3311                /* Find target extent list */
3312                int     erp_idx = 0;
3313                erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
3314                base = erp->er_extbuf;
3315                high = erp->er_extcount - 1;
3316        } else {
3317                base = ifp->if_u1.if_extents;
3318                high = nextents - 1;
3319        }
3320        /* Binary search extent records */
3321        while (low <= high) {
3322                idx = (low + high) >> 1;
3323                ep = base + idx;
3324                startoff = xfs_bmbt_get_startoff(ep);
3325                blockcount = xfs_bmbt_get_blockcount(ep);
3326                if (bno < startoff) {
3327                        high = idx - 1;
3328                } else if (bno >= startoff + blockcount) {
3329                        low = idx + 1;
3330                } else {
3331                        /* Convert back to file-based extent index */
3332                        if (ifp->if_flags & XFS_IFEXTIREC) {
3333                                idx += erp->er_extoff;
3334                        }
3335                        *idxp = idx;
3336                        return ep;
3337                }
3338        }
3339        /* Convert back to file-based extent index */
3340        if (ifp->if_flags & XFS_IFEXTIREC) {
3341                idx += erp->er_extoff;
3342        }
3343        if (bno >= startoff + blockcount) {
3344                if (++idx == nextents) {
3345                        ep = NULL;
3346                } else {
3347                        ep = xfs_iext_get_ext(ifp, idx);
3348                }
3349        }
3350        *idxp = idx;
3351        return ep;
3352}
3353
3354/*
3355 * Return a pointer to the indirection array entry containing the
3356 * extent record for filesystem block bno. Store the index of the
3357 * target irec in *erp_idxp.
3358 */
3359xfs_ext_irec_t *                        /* pointer to found extent record */
3360xfs_iext_bno_to_irec(
3361        xfs_ifork_t     *ifp,           /* inode fork pointer */
3362        xfs_fileoff_t   bno,            /* block number to search for */
3363        int             *erp_idxp)      /* irec index of target ext list */
3364{
3365        xfs_ext_irec_t  *erp = NULL;    /* indirection array pointer */
3366        xfs_ext_irec_t  *erp_next;      /* next indirection array entry */
3367        int             erp_idx;        /* indirection array index */
3368        int             nlists;         /* number of extent irec's (lists) */
3369        int             high;           /* binary search upper limit */
3370        int             low;            /* binary search lower limit */
3371
3372        ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3373        nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3374        erp_idx = 0;
3375        low = 0;
3376        high = nlists - 1;
3377        while (low <= high) {
3378                erp_idx = (low + high) >> 1;
3379                erp = &ifp->if_u1.if_ext_irec[erp_idx];
3380                erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
3381                if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
3382                        high = erp_idx - 1;
3383                } else if (erp_next && bno >=
3384                           xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
3385                        low = erp_idx + 1;
3386                } else {
3387                        break;
3388                }
3389        }
3390        *erp_idxp = erp_idx;
3391        return erp;
3392}
3393
3394/*
3395 * Return a pointer to the indirection array entry containing the
3396 * extent record at file extent index *idxp. Store the index of the
3397 * target irec in *erp_idxp and store the page index of the target
3398 * extent record in *idxp.
3399 */
3400xfs_ext_irec_t *
3401xfs_iext_idx_to_irec(
3402        xfs_ifork_t     *ifp,           /* inode fork pointer */
3403        xfs_extnum_t    *idxp,          /* extent index (file -> page) */
3404        int             *erp_idxp,      /* pointer to target irec */
3405        int             realloc)        /* new bytes were just added */
3406{
3407        xfs_ext_irec_t  *prev;          /* pointer to previous irec */
3408        xfs_ext_irec_t  *erp = NULL;    /* pointer to current irec */
3409        int             erp_idx;        /* indirection array index */
3410        int             nlists;         /* number of irec's (ex lists) */
3411        int             high;           /* binary search upper limit */
3412        int             low;            /* binary search lower limit */
3413        xfs_extnum_t    page_idx = *idxp; /* extent index in target list */
3414
3415        ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3416        ASSERT(page_idx >= 0);
3417        ASSERT(page_idx <= ifp->if_bytes / sizeof(xfs_bmbt_rec_t));
3418        ASSERT(page_idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t) || realloc);
3419
3420        nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3421        erp_idx = 0;
3422        low = 0;
3423        high = nlists - 1;
3424
3425        /* Binary search extent irec's */
3426        while (low <= high) {
3427                erp_idx = (low + high) >> 1;
3428                erp = &ifp->if_u1.if_ext_irec[erp_idx];
3429                prev = erp_idx > 0 ? erp - 1 : NULL;
3430                if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
3431                     realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
3432                        high = erp_idx - 1;
3433                } else if (page_idx > erp->er_extoff + erp->er_extcount ||
3434                           (page_idx == erp->er_extoff + erp->er_extcount &&
3435                            !realloc)) {
3436                        low = erp_idx + 1;
3437                } else if (page_idx == erp->er_extoff + erp->er_extcount &&
3438                           erp->er_extcount == XFS_LINEAR_EXTS) {
3439                        ASSERT(realloc);
3440                        page_idx = 0;
3441                        erp_idx++;
3442                        erp = erp_idx < nlists ? erp + 1 : NULL;
3443                        break;
3444                } else {
3445                        page_idx -= erp->er_extoff;
3446                        break;
3447                }
3448        }
3449        *idxp = page_idx;
3450        *erp_idxp = erp_idx;
3451        return(erp);
3452}
3453
3454/*
3455 * Allocate and initialize an indirection array once the space needed
3456 * for incore extents increases above XFS_IEXT_BUFSZ.
3457 */
3458void
3459xfs_iext_irec_init(
3460        xfs_ifork_t     *ifp)           /* inode fork pointer */
3461{
3462        xfs_ext_irec_t  *erp;           /* indirection array pointer */
3463        xfs_extnum_t    nextents;       /* number of extents in file */
3464
3465        ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3466        nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3467        ASSERT(nextents <= XFS_LINEAR_EXTS);
3468
3469        erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS);
3470
3471        if (nextents == 0) {
3472                ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
3473        } else if (!ifp->if_real_bytes) {
3474                xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
3475        } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
3476                xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
3477        }
3478        erp->er_extbuf = ifp->if_u1.if_extents;
3479        erp->er_extcount = nextents;
3480        erp->er_extoff = 0;
3481
3482        ifp->if_flags |= XFS_IFEXTIREC;
3483        ifp->if_real_bytes = XFS_IEXT_BUFSZ;
3484        ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
3485        ifp->if_u1.if_ext_irec = erp;
3486
3487        return;
3488}
3489
3490/*
3491 * Allocate and initialize a new entry in the indirection array.
3492 */
3493xfs_ext_irec_t *
3494xfs_iext_irec_new(
3495        xfs_ifork_t     *ifp,           /* inode fork pointer */
3496        int             erp_idx)        /* index for new irec */
3497{
3498        xfs_ext_irec_t  *erp;           /* indirection array pointer */
3499        int             i;              /* loop counter */
3500        int             nlists;         /* number of irec's (ex lists) */
3501
3502        ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3503        nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3504
3505        /* Resize indirection array */
3506        xfs_iext_realloc_indirect(ifp, ++nlists *
3507                                  sizeof(xfs_ext_irec_t));
3508        /*
3509         * Move records down in the array so the
3510         * new page can use erp_idx.
3511         */
3512        erp = ifp->if_u1.if_ext_irec;
3513        for (i = nlists - 1; i > erp_idx; i--) {
3514                memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
3515        }
3516        ASSERT(i == erp_idx);
3517
3518        /* Initialize new extent record */
3519        erp = ifp->if_u1.if_ext_irec;
3520        erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
3521        ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
3522        memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
3523        erp[erp_idx].er_extcount = 0;
3524        erp[erp_idx].er_extoff = erp_idx > 0 ?
3525                erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
3526        return (&erp[erp_idx]);
3527}
3528
3529/*
3530 * Remove a record from the indirection array.
3531 */
3532void
3533xfs_iext_irec_remove(
3534        xfs_ifork_t     *ifp,           /* inode fork pointer */
3535        int             erp_idx)        /* irec index to remove */
3536{
3537        xfs_ext_irec_t  *erp;           /* indirection array pointer */
3538        int             i;              /* loop counter */
3539        int             nlists;         /* number of irec's (ex lists) */
3540
3541        ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3542        nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3543        erp = &ifp->if_u1.if_ext_irec[erp_idx];
3544        if (erp->er_extbuf) {
3545                xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
3546                        -erp->er_extcount);
3547                kmem_free(erp->er_extbuf);
3548        }
3549        /* Compact extent records */
3550        erp = ifp->if_u1.if_ext_irec;
3551        for (i = erp_idx; i < nlists - 1; i++) {
3552                memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
3553        }
3554        /*
3555         * Manually free the last extent record from the indirection
3556         * array.  A call to xfs_iext_realloc_indirect() with a size
3557         * of zero would result in a call to xfs_iext_destroy() which
3558         * would in turn call this function again, creating a nasty
3559         * infinite loop.
3560         */
3561        if (--nlists) {
3562                xfs_iext_realloc_indirect(ifp,
3563                        nlists * sizeof(xfs_ext_irec_t));
3564        } else {
3565                kmem_free(ifp->if_u1.if_ext_irec);
3566        }
3567        ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
3568}
3569
3570/*
3571 * This is called to clean up large amounts of unused memory allocated
3572 * by the indirection array.  Before compacting anything though, verify
3573 * that the indirection array is still needed and switch back to the
3574 * linear extent list (or even the inline buffer) if possible.  The
3575 * compaction policy is as follows:
3576 *
3577 *    Full Compaction: Extents fit into a single page (or inline buffer)
3578 * Partial Compaction: Extents occupy less than 50% of allocated space
3579 *      No Compaction: Extents occupy at least 50% of allocated space
3580 */
3581void
3582xfs_iext_irec_compact(
3583        xfs_ifork_t     *ifp)           /* inode fork pointer */
3584{
3585        xfs_extnum_t    nextents;       /* number of extents in file */
3586        int             nlists;         /* number of irec's (ex lists) */
3587
3588        ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3589        nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3590        nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3591
3592        if (nextents == 0) {
3593                xfs_iext_destroy(ifp);
3594        } else if (nextents <= XFS_INLINE_EXTS) {
3595                xfs_iext_indirect_to_direct(ifp);
3596                xfs_iext_direct_to_inline(ifp, nextents);
3597        } else if (nextents <= XFS_LINEAR_EXTS) {
3598                xfs_iext_indirect_to_direct(ifp);
3599        } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
3600                xfs_iext_irec_compact_pages(ifp);
3601        }
3602}
3603
3604/*
3605 * Combine extents from neighboring extent pages.
3606 */
3607void
3608xfs_iext_irec_compact_pages(
3609        xfs_ifork_t     *ifp)           /* inode fork pointer */
3610{
3611        xfs_ext_irec_t  *erp, *erp_next;/* pointers to irec entries */
3612        int             erp_idx = 0;    /* indirection array index */
3613        int             nlists;         /* number of irec's (ex lists) */
3614
3615        ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3616        nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3617        while (erp_idx < nlists - 1) {
3618                erp = &ifp->if_u1.if_ext_irec[erp_idx];
3619                erp_next = erp + 1;
3620                if (erp_next->er_extcount <=
3621                    (XFS_LINEAR_EXTS - erp->er_extcount)) {
3622                        memcpy(&erp->er_extbuf[erp->er_extcount],
3623                                erp_next->er_extbuf, erp_next->er_extcount *
3624                                sizeof(xfs_bmbt_rec_t));
3625                        erp->er_extcount += erp_next->er_extcount;
3626                        /*
3627                         * Free page before removing extent record
3628                         * so er_extoffs don't get modified in
3629                         * xfs_iext_irec_remove.
3630                         */
3631                        kmem_free(erp_next->er_extbuf);
3632                        erp_next->er_extbuf = NULL;
3633                        xfs_iext_irec_remove(ifp, erp_idx + 1);
3634                        nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3635                } else {
3636                        erp_idx++;
3637                }
3638        }
3639}
3640
3641/*
3642 * This is called to update the er_extoff field in the indirection
3643 * array when extents have been added or removed from one of the
3644 * extent lists. erp_idx contains the irec index to begin updating
3645 * at and ext_diff contains the number of extents that were added
3646 * or removed.
3647 */
3648void
3649xfs_iext_irec_update_extoffs(
3650        xfs_ifork_t     *ifp,           /* inode fork pointer */
3651        int             erp_idx,        /* irec index to update */
3652        int             ext_diff)       /* number of new extents */
3653{
3654        int             i;              /* loop counter */
3655        int             nlists;         /* number of irec's (ex lists */
3656
3657        ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3658        nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3659        for (i = erp_idx; i < nlists; i++) {
3660                ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;
3661        }
3662}
3663
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.