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,
1513                                        XBF_UNMAPPED);
1514
1515                if (!bp)
1516                        return ENOMEM;
1517                /*
1518                 * Walk the inodes already attached to the buffer and mark them
1519                 * stale. These will all have the flush locks held, so an
1520                 * in-memory inode walk can't lock them. By marking them all
1521                 * stale first, we will not attempt to lock them in the loop
1522                 * below as the XFS_ISTALE flag will be set.
1523                 */
1524                lip = bp->b_fspriv;
1525                while (lip) {
1526                        if (lip->li_type == XFS_LI_INODE) {
1527                                iip = (xfs_inode_log_item_t *)lip;
1528                                ASSERT(iip->ili_logged == 1);
1529                                lip->li_cb = xfs_istale_done;
1530                                xfs_trans_ail_copy_lsn(mp->m_ail,
1531                                                        &iip->ili_flush_lsn,
1532                                                        &iip->ili_item.li_lsn);
1533                                xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
1534                        }
1535                        lip = lip->li_bio_list;
1536                }
1537
1538
1539                /*
1540                 * For each inode in memory attempt to add it to the inode
1541                 * buffer and set it up for being staled on buffer IO
1542                 * completion.  This is safe as we've locked out tail pushing
1543                 * and flushing by locking the buffer.
1544                 *
1545                 * We have already marked every inode that was part of a
1546                 * transaction stale above, which means there is no point in
1547                 * even trying to lock them.
1548                 */
1549                for (i = 0; i < ninodes; i++) {
1550retry:
1551                        rcu_read_lock();
1552                        ip = radix_tree_lookup(&pag->pag_ici_root,
1553                                        XFS_INO_TO_AGINO(mp, (inum + i)));
1554
1555                        /* Inode not in memory, nothing to do */
1556                        if (!ip) {
1557                                rcu_read_unlock();
1558                                continue;
1559                        }
1560
1561                        /*
1562                         * because this is an RCU protected lookup, we could
1563                         * find a recently freed or even reallocated inode
1564                         * during the lookup. We need to check under the
1565                         * i_flags_lock for a valid inode here. Skip it if it
1566                         * is not valid, the wrong inode or stale.
1567                         */
1568                        spin_lock(&ip->i_flags_lock);
1569                        if (ip->i_ino != inum + i ||
1570                            __xfs_iflags_test(ip, XFS_ISTALE)) {
1571                                spin_unlock(&ip->i_flags_lock);
1572                                rcu_read_unlock();
1573                                continue;
1574                        }
1575                        spin_unlock(&ip->i_flags_lock);
1576
1577                        /*
1578                         * Don't try to lock/unlock the current inode, but we
1579                         * _cannot_ skip the other inodes that we did not find
1580                         * in the list attached to the buffer and are not
1581                         * already marked stale. If we can't lock it, back off
1582                         * and retry.
1583                         */
1584                        if (ip != free_ip &&
1585                            !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
1586                                rcu_read_unlock();
1587                                delay(1);
1588                                goto retry;
1589                        }
1590                        rcu_read_unlock();
1591
1592                        xfs_iflock(ip);
1593                        xfs_iflags_set(ip, XFS_ISTALE);
1594
1595                        /*
1596                         * we don't need to attach clean inodes or those only
1597                         * with unlogged changes (which we throw away, anyway).
1598                         */
1599                        iip = ip->i_itemp;
1600                        if (!iip || xfs_inode_clean(ip)) {
1601                                ASSERT(ip != free_ip);
1602                                xfs_ifunlock(ip);
1603                                xfs_iunlock(ip, XFS_ILOCK_EXCL);
1604                                continue;
1605                        }
1606
1607                        iip->ili_last_fields = iip->ili_fields;
1608                        iip->ili_fields = 0;
1609                        iip->ili_logged = 1;
1610                        xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
1611                                                &iip->ili_item.li_lsn);
1612
1613                        xfs_buf_attach_iodone(bp, xfs_istale_done,
1614                                                  &iip->ili_item);
1615
1616                        if (ip != free_ip)
1617                                xfs_iunlock(ip, XFS_ILOCK_EXCL);
1618                }
1619
1620                xfs_trans_stale_inode_buf(tp, bp);
1621                xfs_trans_binval(tp, bp);
1622        }
1623
1624        xfs_perag_put(pag);
1625        return 0;
1626}
1627
1628/*
1629 * This is called to return an inode to the inode free list.
1630 * The inode should already be truncated to 0 length and have
1631 * no pages associated with it.  This routine also assumes that
1632 * the inode is already a part of the transaction.
1633 *
1634 * The on-disk copy of the inode will have been added to the list
1635 * of unlinked inodes in the AGI. We need to remove the inode from
1636 * that list atomically with respect to freeing it here.
1637 */
1638int
1639xfs_ifree(
1640        xfs_trans_t     *tp,
1641        xfs_inode_t     *ip,
1642        xfs_bmap_free_t *flist)
1643{
1644        int                     error;
1645        int                     delete;
1646        xfs_ino_t               first_ino;
1647        xfs_dinode_t            *dip;
1648        xfs_buf_t               *ibp;
1649
1650        ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1651        ASSERT(ip->i_d.di_nlink == 0);
1652        ASSERT(ip->i_d.di_nextents == 0);
1653        ASSERT(ip->i_d.di_anextents == 0);
1654        ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode));
1655        ASSERT(ip->i_d.di_nblocks == 0);
1656
1657        /*
1658         * Pull the on-disk inode from the AGI unlinked list.
1659         */
1660        error = xfs_iunlink_remove(tp, ip);
1661        if (error != 0) {
1662                return error;
1663        }
1664
1665        error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
1666        if (error != 0) {
1667                return error;
1668        }
1669        ip->i_d.di_mode = 0;            /* mark incore inode as free */
1670        ip->i_d.di_flags = 0;
1671        ip->i_d.di_dmevmask = 0;
1672        ip->i_d.di_forkoff = 0;         /* mark the attr fork not in use */
1673        ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1674        ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1675        /*
1676         * Bump the generation count so no one will be confused
1677         * by reincarnations of this inode.
1678         */
1679        ip->i_d.di_gen++;
1680
1681        xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1682
1683        error = xfs_imap_to_bp(ip->i_mount, tp, &ip->i_imap, &dip, &ibp,
1684                               0, 0);
1685        if (error)
1686                return error;
1687
1688        /*
1689        * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
1690        * from picking up this inode when it is reclaimed (its incore state
1691        * initialzed but not flushed to disk yet). The in-core di_mode is
1692        * already cleared  and a corresponding transaction logged.
1693        * The hack here just synchronizes the in-core to on-disk
1694        * di_mode value in advance before the actual inode sync to disk.
1695        * This is OK because the inode is already unlinked and would never
1696        * change its di_mode again for this inode generation.
1697        * This is a temporary hack that would require a proper fix
1698        * in the future.
1699        */
1700        dip->di_mode = 0;
1701
1702        if (delete) {
1703                error = xfs_ifree_cluster(ip, tp, first_ino);
1704        }
1705
1706        return error;
1707}
1708
1709/*
1710 * Reallocate the space for if_broot based on the number of records
1711 * being added or deleted as indicated in rec_diff.  Move the records
1712 * and pointers in if_broot to fit the new size.  When shrinking this
1713 * will eliminate holes between the records and pointers created by
1714 * the caller.  When growing this will create holes to be filled in
1715 * by the caller.
1716 *
1717 * The caller must not request to add more records than would fit in
1718 * the on-disk inode root.  If the if_broot is currently NULL, then
1719 * if we adding records one will be allocated.  The caller must also
1720 * not request that the number of records go below zero, although
1721 * it can go to zero.
1722 *
1723 * ip -- the inode whose if_broot area is changing
1724 * ext_diff -- the change in the number of records, positive or negative,
1725 *       requested for the if_broot array.
1726 */
1727void
1728xfs_iroot_realloc(
1729        xfs_inode_t             *ip,
1730        int                     rec_diff,
1731        int                     whichfork)
1732{
1733        struct xfs_mount        *mp = ip->i_mount;
1734        int                     cur_max;
1735        xfs_ifork_t             *ifp;
1736        struct xfs_btree_block  *new_broot;
1737        int                     new_max;
1738        size_t                  new_size;
1739        char                    *np;
1740        char                    *op;
1741
1742        /*
1743         * Handle the degenerate case quietly.
1744         */
1745        if (rec_diff == 0) {
1746                return;
1747        }
1748
1749        ifp = XFS_IFORK_PTR(ip, whichfork);
1750        if (rec_diff > 0) {
1751                /*
1752                 * If there wasn't any memory allocated before, just
1753                 * allocate it now and get out.
1754                 */
1755                if (ifp->if_broot_bytes == 0) {
1756                        new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
1757                        ifp->if_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
1758                        ifp->if_broot_bytes = (int)new_size;
1759                        return;
1760                }
1761
1762                /*
1763                 * If there is already an existing if_broot, then we need
1764                 * to realloc() it and shift the pointers to their new
1765                 * location.  The records don't change location because
1766                 * they are kept butted up against the btree block header.
1767                 */
1768                cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
1769                new_max = cur_max + rec_diff;
1770                new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
1771                ifp->if_broot = kmem_realloc(ifp->if_broot, new_size,
1772                                (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
1773                                KM_SLEEP | KM_NOFS);
1774                op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
1775                                                     ifp->if_broot_bytes);
1776                np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
1777                                                     (int)new_size);
1778                ifp->if_broot_bytes = (int)new_size;
1779                ASSERT(ifp->if_broot_bytes <=
1780                        XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
1781                memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
1782                return;
1783        }
1784
1785        /*
1786         * rec_diff is less than 0.  In this case, we are shrinking the
1787         * if_broot buffer.  It must already exist.  If we go to zero
1788         * records, just get rid of the root and clear the status bit.
1789         */
1790        ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
1791        cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
1792        new_max = cur_max + rec_diff;
1793        ASSERT(new_max >= 0);
1794        if (new_max > 0)
1795                new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
1796        else
1797                new_size = 0;
1798        if (new_size > 0) {
1799                new_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
1800                /*
1801                 * First copy over the btree block header.
1802                 */
1803                memcpy(new_broot, ifp->if_broot, XFS_BTREE_LBLOCK_LEN);
1804        } else {
1805                new_broot = NULL;
1806                ifp->if_flags &= ~XFS_IFBROOT;
1807        }
1808
1809        /*
1810         * Only copy the records and pointers if there are any.
1811         */
1812        if (new_max > 0) {
1813                /*
1814                 * First copy the records.
1815                 */
1816                op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
1817                np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
1818                memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
1819
1820                /*
1821                 * Then copy the pointers.
1822                 */
1823                op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
1824                                                     ifp->if_broot_bytes);
1825                np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
1826                                                     (int)new_size);
1827                memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
1828        }
1829        kmem_free(ifp->if_broot);
1830        ifp->if_broot = new_broot;
1831        ifp->if_broot_bytes = (int)new_size;
1832        ASSERT(ifp->if_broot_bytes <=
1833                XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
1834        return;
1835}
1836
1837
1838/*
1839 * This is called when the amount of space needed for if_data
1840 * is increased or decreased.  The change in size is indicated by
1841 * the number of bytes that need to be added or deleted in the
1842 * byte_diff parameter.
1843 *
1844 * If the amount of space needed has decreased below the size of the
1845 * inline buffer, then switch to using the inline buffer.  Otherwise,
1846 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
1847 * to what is needed.
1848 *
1849 * ip -- the inode whose if_data area is changing
1850 * byte_diff -- the change in the number of bytes, positive or negative,
1851 *       requested for the if_data array.
1852 */
1853void
1854xfs_idata_realloc(
1855        xfs_inode_t     *ip,
1856        int             byte_diff,
1857        int             whichfork)
1858{
1859        xfs_ifork_t     *ifp;
1860        int             new_size;
1861        int             real_size;
1862
1863        if (byte_diff == 0) {
1864                return;
1865        }
1866
1867        ifp = XFS_IFORK_PTR(ip, whichfork);
1868        new_size = (int)ifp->if_bytes + byte_diff;
1869        ASSERT(new_size >= 0);
1870
1871        if (new_size == 0) {
1872                if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
1873                        kmem_free(ifp->if_u1.if_data);
1874                }
1875                ifp->if_u1.if_data = NULL;
1876                real_size = 0;
1877        } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
1878                /*
1879                 * If the valid extents/data can fit in if_inline_ext/data,
1880                 * copy them from the malloc'd vector and free it.
1881                 */
1882                if (ifp->if_u1.if_data == NULL) {
1883                        ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
1884                } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
1885                        ASSERT(ifp->if_real_bytes != 0);
1886                        memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
1887                              new_size);
1888                        kmem_free(ifp->if_u1.if_data);
1889                        ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
1890                }
1891                real_size = 0;
1892        } else {
1893                /*
1894                 * Stuck with malloc/realloc.
1895                 * For inline data, the underlying buffer must be
1896                 * a multiple of 4 bytes in size so that it can be
1897                 * logged and stay on word boundaries.  We enforce
1898                 * that here.
1899                 */
1900                real_size = roundup(new_size, 4);
1901                if (ifp->if_u1.if_data == NULL) {
1902                        ASSERT(ifp->if_real_bytes == 0);
1903                        ifp->if_u1.if_data = kmem_alloc(real_size,
1904                                                        KM_SLEEP | KM_NOFS);
1905                } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
1906                        /*
1907                         * Only do the realloc if the underlying size
1908                         * is really changing.
1909                         */
1910                        if (ifp->if_real_bytes != real_size) {
1911                                ifp->if_u1.if_data =
1912                                        kmem_realloc(ifp->if_u1.if_data,
1913                                                        real_size,
1914                                                        ifp->if_real_bytes,
1915                                                        KM_SLEEP | KM_NOFS);
1916                        }
1917                } else {
1918                        ASSERT(ifp->if_real_bytes == 0);
1919                        ifp->if_u1.if_data = kmem_alloc(real_size,
1920                                                        KM_SLEEP | KM_NOFS);
1921                        memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
1922                                ifp->if_bytes);
1923                }
1924        }
1925        ifp->if_real_bytes = real_size;
1926        ifp->if_bytes = new_size;
1927        ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
1928}
1929
1930void
1931xfs_idestroy_fork(
1932        xfs_inode_t     *ip,
1933        int             whichfork)
1934{
1935        xfs_ifork_t     *ifp;
1936
1937        ifp = XFS_IFORK_PTR(ip, whichfork);
1938        if (ifp->if_broot != NULL) {
1939                kmem_free(ifp->if_broot);
1940                ifp->if_broot = NULL;
1941        }
1942
1943        /*
1944         * If the format is local, then we can't have an extents
1945         * array so just look for an inline data array.  If we're
1946         * not local then we may or may not have an extents list,
1947         * so check and free it up if we do.
1948         */
1949        if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1950                if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
1951                    (ifp->if_u1.if_data != NULL)) {
1952                        ASSERT(ifp->if_real_bytes != 0);
1953                        kmem_free(ifp->if_u1.if_data);
1954                        ifp->if_u1.if_data = NULL;
1955                        ifp->if_real_bytes = 0;
1956                }
1957        } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
1958                   ((ifp->if_flags & XFS_IFEXTIREC) ||
1959                    ((ifp->if_u1.if_extents != NULL) &&
1960                     (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
1961                ASSERT(ifp->if_real_bytes != 0);
1962                xfs_iext_destroy(ifp);
1963        }
1964        ASSERT(ifp->if_u1.if_extents == NULL ||
1965               ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
1966        ASSERT(ifp->if_real_bytes == 0);
1967        if (whichfork == XFS_ATTR_FORK) {
1968                kmem_zone_free(xfs_ifork_zone, ip->i_afp);
1969                ip->i_afp = NULL;
1970        }
1971}
1972
1973/*
1974 * This is called to unpin an inode.  The caller must have the inode locked
1975 * in at least shared mode so that the buffer cannot be subsequently pinned
1976 * once someone is waiting for it to be unpinned.
1977 */
1978static void
1979xfs_iunpin(
1980        struct xfs_inode        *ip)
1981{
1982        ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
1983
1984        trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
1985
1986        /* Give the log a push to start the unpinning I/O */
1987        xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
1988
1989}
1990
1991static void
1992__xfs_iunpin_wait(
1993        struct xfs_inode        *ip)
1994{
1995        wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
1996        DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
1997
1998        xfs_iunpin(ip);
1999
2000        do {
2001                prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2002                if (xfs_ipincount(ip))
2003                        io_schedule();
2004        } while (xfs_ipincount(ip));
2005        finish_wait(wq, &wait.wait);
2006}
2007
2008void
2009xfs_iunpin_wait(
2010        struct xfs_inode        *ip)
2011{
2012        if (xfs_ipincount(ip))
2013                __xfs_iunpin_wait(ip);
2014}
2015
2016/*
2017 * xfs_iextents_copy()
2018 *
2019 * This is called to copy the REAL extents (as opposed to the delayed
2020 * allocation extents) from the inode into the given buffer.  It
2021 * returns the number of bytes copied into the buffer.
2022 *
2023 * If there are no delayed allocation extents, then we can just
2024 * memcpy() the extents into the buffer.  Otherwise, we need to
2025 * examine each extent in turn and skip those which are delayed.
2026 */
2027int
2028xfs_iextents_copy(
2029        xfs_inode_t             *ip,
2030        xfs_bmbt_rec_t          *dp,
2031        int                     whichfork)
2032{
2033        int                     copied;
2034        int                     i;
2035        xfs_ifork_t             *ifp;
2036        int                     nrecs;
2037        xfs_fsblock_t           start_block;
2038
2039        ifp = XFS_IFORK_PTR(ip, whichfork);
2040        ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2041        ASSERT(ifp->if_bytes > 0);
2042
2043        nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2044        XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
2045        ASSERT(nrecs > 0);
2046
2047        /*
2048         * There are some delayed allocation extents in the
2049         * inode, so copy the extents one at a time and skip
2050         * the delayed ones.  There must be at least one
2051         * non-delayed extent.
2052         */
2053        copied = 0;
2054        for (i = 0; i < nrecs; i++) {
2055                xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
2056                start_block = xfs_bmbt_get_startblock(ep);
2057                if (isnullstartblock(start_block)) {
2058                        /*
2059                         * It's a delayed allocation extent, so skip it.
2060                         */
2061                        continue;
2062                }
2063
2064                /* Translate to on disk format */
2065                put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
2066                put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
2067                dp++;
2068                copied++;
2069        }
2070        ASSERT(copied != 0);
2071        xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
2072
2073        return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2074}
2075
2076/*
2077 * Each of the following cases stores data into the same region
2078 * of the on-disk inode, so only one of them can be valid at
2079 * any given time. While it is possible to have conflicting formats
2080 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2081 * in EXTENTS format, this can only happen when the fork has
2082 * changed formats after being modified but before being flushed.
2083 * In these cases, the format always takes precedence, because the
2084 * format indicates the current state of the fork.
2085 */
2086/*ARGSUSED*/
2087STATIC void
2088xfs_iflush_fork(
2089        xfs_inode_t             *ip,
2090        xfs_dinode_t            *dip,
2091        xfs_inode_log_item_t    *iip,
2092        int                     whichfork,
2093        xfs_buf_t               *bp)
2094{
2095        char                    *cp;
2096        xfs_ifork_t             *ifp;
2097        xfs_mount_t             *mp;
2098#ifdef XFS_TRANS_DEBUG
2099        int                     first;
2100#endif
2101        static const short      brootflag[2] =
2102                { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2103        static const short      dataflag[2] =
2104                { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2105        static const short      extflag[2] =
2106                { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2107
2108        if (!iip)
2109                return;
2110        ifp = XFS_IFORK_PTR(ip, whichfork);
2111        /*
2112         * This can happen if we gave up in iformat in an error path,
2113         * for the attribute fork.
2114         */
2115        if (!ifp) {
2116                ASSERT(whichfork == XFS_ATTR_FORK);
2117                return;
2118        }
2119        cp = XFS_DFORK_PTR(dip, whichfork);
2120        mp = ip->i_mount;
2121        switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2122        case XFS_DINODE_FMT_LOCAL:
2123                if ((iip->ili_fields & dataflag[whichfork]) &&
2124                    (ifp->if_bytes > 0)) {
2125                        ASSERT(ifp->if_u1.if_data != NULL);
2126                        ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2127                        memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2128                }
2129                break;
2130
2131        case XFS_DINODE_FMT_EXTENTS:
2132                ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2133                       !(iip->ili_fields & extflag[whichfork]));
2134                if ((iip->ili_fields & extflag[whichfork]) &&
2135                    (ifp->if_bytes > 0)) {
2136                        ASSERT(xfs_iext_get_ext(ifp, 0));
2137                        ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2138                        (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2139                                whichfork);
2140                }
2141                break;
2142
2143        case XFS_DINODE_FMT_BTREE:
2144                if ((iip->ili_fields & brootflag[whichfork]) &&
2145                    (ifp->if_broot_bytes > 0)) {
2146                        ASSERT(ifp->if_broot != NULL);
2147                        ASSERT(ifp->if_broot_bytes <=
2148                               (XFS_IFORK_SIZE(ip, whichfork) +
2149                                XFS_BROOT_SIZE_ADJ));
2150                        xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
2151                                (xfs_bmdr_block_t *)cp,
2152                                XFS_DFORK_SIZE(dip, mp, whichfork));
2153                }
2154                break;
2155
2156        case XFS_DINODE_FMT_DEV:
2157                if (iip->ili_fields & XFS_ILOG_DEV) {
2158                        ASSERT(whichfork == XFS_DATA_FORK);
2159                        xfs_dinode_put_rdev(dip, ip->i_df.if_u2.if_rdev);
2160                }
2161                break;
2162
2163        case XFS_DINODE_FMT_UUID:
2164                if (iip->ili_fields & XFS_ILOG_UUID) {
2165                        ASSERT(whichfork == XFS_DATA_FORK);
2166                        memcpy(XFS_DFORK_DPTR(dip),
2167                               &ip->i_df.if_u2.if_uuid,
2168                               sizeof(uuid_t));
2169                }
2170                break;
2171
2172        default:
2173                ASSERT(0);
2174                break;
2175        }
2176}
2177
2178STATIC int
2179xfs_iflush_cluster(
2180        xfs_inode_t     *ip,
2181        xfs_buf_t       *bp)
2182{
2183        xfs_mount_t             *mp = ip->i_mount;
2184        struct xfs_perag        *pag;
2185        unsigned long           first_index, mask;
2186        unsigned long           inodes_per_cluster;
2187        int                     ilist_size;
2188        xfs_inode_t             **ilist;
2189        xfs_inode_t             *iq;
2190        int                     nr_found;
2191        int                     clcount = 0;
2192        int                     bufwasdelwri;
2193        int                     i;
2194
2195        pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2196
2197        inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2198        ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
2199        ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
2200        if (!ilist)
2201                goto out_put;
2202
2203        mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
2204        first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
2205        rcu_read_lock();
2206        /* really need a gang lookup range call here */
2207        nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
2208                                        first_index, inodes_per_cluster);
2209        if (nr_found == 0)
2210                goto out_free;
2211
2212        for (i = 0; i < nr_found; i++) {
2213                iq = ilist[i];
2214                if (iq == ip)
2215                        continue;
2216
2217                /*
2218                 * because this is an RCU protected lookup, we could find a
2219                 * recently freed or even reallocated inode during the lookup.
2220                 * We need to check under the i_flags_lock for a valid inode
2221                 * here. Skip it if it is not valid or the wrong inode.
2222                 */
2223                spin_lock(&ip->i_flags_lock);
2224                if (!ip->i_ino ||
2225                    (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
2226                        spin_unlock(&ip->i_flags_lock);
2227                        continue;
2228                }
2229                spin_unlock(&ip->i_flags_lock);
2230
2231                /*
2232                 * Do an un-protected check to see if the inode is dirty and
2233                 * is a candidate for flushing.  These checks will be repeated
2234                 * later after the appropriate locks are acquired.
2235                 */
2236                if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
2237                        continue;
2238
2239                /*
2240                 * Try to get locks.  If any are unavailable or it is pinned,
2241                 * then this inode cannot be flushed and is skipped.
2242                 */
2243
2244                if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
2245                        continue;
2246                if (!xfs_iflock_nowait(iq)) {
2247                        xfs_iunlock(iq, XFS_ILOCK_SHARED);
2248                        continue;
2249                }
2250                if (xfs_ipincount(iq)) {
2251                        xfs_ifunlock(iq);
2252                        xfs_iunlock(iq, XFS_ILOCK_SHARED);
2253                        continue;
2254                }
2255
2256                /*
2257                 * arriving here means that this inode can be flushed.  First
2258                 * re-check that it's dirty before flushing.
2259                 */
2260                if (!xfs_inode_clean(iq)) {
2261                        int     error;
2262                        error = xfs_iflush_int(iq, bp);
2263                        if (error) {
2264                                xfs_iunlock(iq, XFS_ILOCK_SHARED);
2265                                goto cluster_corrupt_out;
2266                        }
2267                        clcount++;
2268                } else {
2269                        xfs_ifunlock(iq);
2270                }
2271                xfs_iunlock(iq, XFS_ILOCK_SHARED);
2272        }
2273
2274        if (clcount) {
2275                XFS_STATS_INC(xs_icluster_flushcnt);
2276                XFS_STATS_ADD(xs_icluster_flushinode, clcount);
2277        }
2278
2279out_free:
2280        rcu_read_unlock();
2281        kmem_free(ilist);
2282out_put:
2283        xfs_perag_put(pag);
2284        return 0;
2285
2286
2287cluster_corrupt_out:
2288        /*
2289         * Corruption detected in the clustering loop.  Invalidate the
2290         * inode buffer and shut down the filesystem.
2291         */
2292        rcu_read_unlock();
2293        /*
2294         * Clean up the buffer.  If it was delwri, just release it --
2295         * brelse can handle it with no problems.  If not, shut down the
2296         * filesystem before releasing the buffer.
2297         */
2298        bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
2299        if (bufwasdelwri)
2300                xfs_buf_relse(bp);
2301
2302        xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2303
2304        if (!bufwasdelwri) {
2305                /*
2306                 * Just like incore_relse: if we have b_iodone functions,
2307                 * mark the buffer as an error and call them.  Otherwise
2308                 * mark it as stale and brelse.
2309                 */
2310                if (bp->b_iodone) {
2311                        XFS_BUF_UNDONE(bp);
2312                        xfs_buf_stale(bp);
2313                        xfs_buf_ioerror(bp, EIO);
2314                        xfs_buf_ioend(bp, 0);
2315                } else {
2316                        xfs_buf_stale(bp);
2317                        xfs_buf_relse(bp);
2318                }
2319        }
2320
2321        /*
2322         * Unlocks the flush lock
2323         */
2324        xfs_iflush_abort(iq, false);
2325        kmem_free(ilist);
2326        xfs_perag_put(pag);
2327        return XFS_ERROR(EFSCORRUPTED);
2328}
2329
2330/*
2331 * Flush dirty inode metadata into the backing buffer.
2332 *
2333 * The caller must have the inode lock and the inode flush lock held.  The
2334 * inode lock will still be held upon return to the caller, and the inode
2335 * flush lock will be released after the inode has reached the disk.
2336 *
2337 * The caller must write out the buffer returned in *bpp and release it.
2338 */
2339int
2340xfs_iflush(
2341        struct xfs_inode        *ip,
2342        struct xfs_buf          **bpp)
2343{
2344        struct xfs_mount        *mp = ip->i_mount;
2345        struct xfs_buf          *bp;
2346        struct xfs_dinode       *dip;
2347        int                     error;
2348
2349        XFS_STATS_INC(xs_iflush_count);
2350
2351        ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2352        ASSERT(xfs_isiflocked(ip));
2353        ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2354               ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
2355
2356        *bpp = NULL;
2357
2358        xfs_iunpin_wait(ip);
2359
2360        /*
2361         * For stale inodes we cannot rely on the backing buffer remaining
2362         * stale in cache for the remaining life of the stale inode and so
2363         * xfs_imap_to_bp() below may give us a buffer that no longer contains
2364         * inodes below. We have to check this after ensuring the inode is
2365         * unpinned so that it is safe to reclaim the stale inode after the
2366         * flush call.
2367         */
2368        if (xfs_iflags_test(ip, XFS_ISTALE)) {
2369                xfs_ifunlock(ip);
2370                return 0;
2371        }
2372
2373        /*
2374         * This may have been unpinned because the filesystem is shutting
2375         * down forcibly. If that's the case we must not write this inode
2376         * to disk, because the log record didn't make it to disk.
2377         *
2378         * We also have to remove the log item from the AIL in this case,
2379         * as we wait for an empty AIL as part of the unmount process.
2380         */
2381        if (XFS_FORCED_SHUTDOWN(mp)) {
2382                error = XFS_ERROR(EIO);
2383                goto abort_out;
2384        }
2385
2386        /*
2387         * Get the buffer containing the on-disk inode.
2388         */
2389        error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
2390                               0);
2391        if (error || !bp) {
2392                xfs_ifunlock(ip);
2393                return error;
2394        }
2395
2396        /*
2397         * First flush out the inode that xfs_iflush was called with.
2398         */
2399        error = xfs_iflush_int(ip, bp);
2400        if (error)
2401                goto corrupt_out;
2402
2403        /*
2404         * If the buffer is pinned then push on the log now so we won't
2405         * get stuck waiting in the write for too long.
2406         */
2407        if (xfs_buf_ispinned(bp))
2408                xfs_log_force(mp, 0);
2409
2410        /*
2411         * inode clustering:
2412         * see if other inodes can be gathered into this write
2413         */
2414        error = xfs_iflush_cluster(ip, bp);
2415        if (error)
2416                goto cluster_corrupt_out;
2417
2418        *bpp = bp;
2419        return 0;
2420
2421corrupt_out:
2422        xfs_buf_relse(bp);
2423        xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2424cluster_corrupt_out:
2425        error = XFS_ERROR(EFSCORRUPTED);
2426abort_out:
2427        /*
2428         * Unlocks the flush lock
2429         */
2430        xfs_iflush_abort(ip, false);
2431        return error;
2432}
2433
2434
2435STATIC int
2436xfs_iflush_int(
2437        xfs_inode_t             *ip,
2438        xfs_buf_t               *bp)
2439{
2440        xfs_inode_log_item_t    *iip;
2441        xfs_dinode_t            *dip;
2442        xfs_mount_t             *mp;
2443#ifdef XFS_TRANS_DEBUG
2444        int                     first;
2445#endif
2446
2447        ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2448        ASSERT(xfs_isiflocked(ip));
2449        ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2450               ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
2451
2452        iip = ip->i_itemp;
2453        mp = ip->i_mount;
2454
2455        /* set *dip = inode's place in the buffer */
2456        dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
2457
2458        if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
2459                               mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
2460                xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2461                        "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
2462                        __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
2463                goto corrupt_out;
2464        }
2465        if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
2466                                mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
2467                xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2468                        "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
2469                        __func__, ip->i_ino, ip, ip->i_d.di_magic);
2470                goto corrupt_out;
2471        }
2472        if (S_ISREG(ip->i_d.di_mode)) {
2473                if (XFS_TEST_ERROR(
2474                    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2475                    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
2476                    mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
2477                        xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2478                                "%s: Bad regular inode %Lu, ptr 0x%p",
2479                                __func__, ip->i_ino, ip);
2480                        goto corrupt_out;
2481                }
2482        } else if (S_ISDIR(ip->i_d.di_mode)) {
2483                if (XFS_TEST_ERROR(
2484                    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2485                    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
2486                    (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
2487                    mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
2488                        xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2489                                "%s: Bad directory inode %Lu, ptr 0x%p",
2490                                __func__, ip->i_ino, ip);
2491                        goto corrupt_out;
2492                }
2493        }
2494        if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
2495                                ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
2496                                XFS_RANDOM_IFLUSH_5)) {
2497                xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2498                        "%s: detected corrupt incore inode %Lu, "
2499                        "total extents = %d, nblocks = %Ld, ptr 0x%p",
2500                        __func__, ip->i_ino,
2501                        ip->i_d.di_nextents + ip->i_d.di_anextents,
2502                        ip->i_d.di_nblocks, ip);
2503                goto corrupt_out;
2504        }
2505        if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
2506                                mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
2507                xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2508                        "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
2509                        __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
2510                goto corrupt_out;
2511        }
2512        /*
2513         * bump the flush iteration count, used to detect flushes which
2514         * postdate a log record during recovery.
2515         */
2516
2517        ip->i_d.di_flushiter++;
2518
2519        /*
2520         * Copy the dirty parts of the inode into the on-disk
2521         * inode.  We always copy out the core of the inode,
2522         * because if the inode is dirty at all the core must
2523         * be.
2524         */
2525        xfs_dinode_to_disk(dip, &ip->i_d);
2526
2527        /* Wrap, we never let the log put out DI_MAX_FLUSH */
2528        if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
2529                ip->i_d.di_flushiter = 0;
2530
2531        /*
2532         * If this is really an old format inode and the superblock version
2533         * has not been updated to support only new format inodes, then
2534         * convert back to the old inode format.  If the superblock version
2535         * has been updated, then make the conversion permanent.
2536         */
2537        ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
2538        if (ip->i_d.di_version == 1) {
2539                if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
2540                        /*
2541                         * Convert it back.
2542                         */
2543                        ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
2544                        dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink);
2545                } else {
2546                        /*
2547                         * The superblock version has already been bumped,
2548                         * so just make the conversion to the new inode
2549                         * format permanent.
2550                         */
2551                        ip->i_d.di_version = 2;
2552                        dip->di_version = 2;
2553                        ip->i_d.di_onlink = 0;
2554                        dip->di_onlink = 0;
2555                        memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
2556                        memset(&(dip->di_pad[0]), 0,
2557                              sizeof(dip->di_pad));
2558                        ASSERT(xfs_get_projid(ip) == 0);
2559                }
2560        }
2561
2562        xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
2563        if (XFS_IFORK_Q(ip))
2564                xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
2565        xfs_inobp_check(mp, bp);
2566
2567        /*
2568         * We've recorded everything logged in the inode, so we'd like to clear
2569         * the ili_fields bits so we don't log and flush things unnecessarily.
2570         * However, we can't stop logging all this information until the data
2571         * we've copied into the disk buffer is written to disk.  If we did we
2572         * might overwrite the copy of the inode in the log with all the data
2573         * after re-logging only part of it, and in the face of a crash we
2574         * wouldn't have all the data we need to recover.
2575         *
2576         * What we do is move the bits to the ili_last_fields field.  When
2577         * logging the inode, these bits are moved back to the ili_fields field.
2578         * In the xfs_iflush_done() routine we clear ili_last_fields, since we
2579         * know that the information those bits represent is permanently on
2580         * disk.  As long as the flush completes before the inode is logged
2581         * again, then both ili_fields and ili_last_fields will be cleared.
2582         *
2583         * We can play with the ili_fields bits here, because the inode lock
2584         * must be held exclusively in order to set bits there and the flush
2585         * lock protects the ili_last_fields bits.  Set ili_logged so the flush
2586         * done routine can tell whether or not to look in the AIL.  Also, store
2587         * the current LSN of the inode so that we can tell whether the item has
2588         * moved in the AIL from xfs_iflush_done().  In order to read the lsn we
2589         * need the AIL lock, because it is a 64 bit value that cannot be read
2590         * atomically.
2591         */
2592        if (iip != NULL && iip->ili_fields != 0) {
2593                iip->ili_last_fields = iip->ili_fields;
2594                iip->ili_fields = 0;
2595                iip->ili_logged = 1;
2596
2597                xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2598                                        &iip->ili_item.li_lsn);
2599
2600                /*
2601                 * Attach the function xfs_iflush_done to the inode's
2602                 * buffer.  This will remove the inode from the AIL
2603                 * and unlock the inode's flush lock when the inode is
2604                 * completely written to disk.
2605                 */
2606                xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
2607
2608                ASSERT(bp->b_fspriv != NULL);
2609                ASSERT(bp->b_iodone != NULL);
2610        } else {
2611                /*
2612                 * We're flushing an inode which is not in the AIL and has
2613                 * not been logged.  For this case we can immediately drop
2614                 * the inode flush lock because we can avoid the whole
2615                 * AIL state thing.  It's OK to drop the flush lock now,
2616                 * because we've already locked the buffer and to do anything
2617                 * you really need both.
2618                 */
2619                if (iip != NULL) {
2620                        ASSERT(iip->ili_logged == 0);
2621                        ASSERT(iip->ili_last_fields == 0);
2622                        ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
2623                }
2624                xfs_ifunlock(ip);
2625        }
2626
2627        return 0;
2628
2629corrupt_out:
2630        return XFS_ERROR(EFSCORRUPTED);
2631}
2632
2633/*
2634 * Return a pointer to the extent record at file index idx.
2635 */
2636xfs_bmbt_rec_host_t *
2637xfs_iext_get_ext(
2638        xfs_ifork_t     *ifp,           /* inode fork pointer */
2639        xfs_extnum_t    idx)            /* index of target extent */
2640{
2641        ASSERT(idx >= 0);
2642        ASSERT(idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t));
2643
2644        if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
2645                return ifp->if_u1.if_ext_irec->er_extbuf;
2646        } else if (ifp->if_flags & XFS_IFEXTIREC) {
2647                xfs_ext_irec_t  *erp;           /* irec pointer */
2648                int             erp_idx = 0;    /* irec index */
2649                xfs_extnum_t    page_idx = idx; /* ext index in target list */
2650
2651                erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
2652                return &erp->er_extbuf[page_idx];
2653        } else if (ifp->if_bytes) {
2654                return &ifp->if_u1.if_extents[idx];
2655        } else {
2656                return NULL;
2657        }
2658}
2659
2660/*
2661 * Insert new item(s) into the extent records for incore inode
2662 * fork 'ifp'.  'count' new items are inserted at index 'idx'.
2663 */
2664void
2665xfs_iext_insert(
2666        xfs_inode_t     *ip,            /* incore inode pointer */
2667        xfs_extnum_t    idx,            /* starting index of new items */
2668        xfs_extnum_t    count,          /* number of inserted items */
2669        xfs_bmbt_irec_t *new,           /* items to insert */
2670        int             state)          /* type of extent conversion */
2671{
2672        xfs_ifork_t     *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
2673        xfs_extnum_t    i;              /* extent record index */
2674
2675        trace_xfs_iext_insert(ip, idx, new, state, _RET_IP_);
2676
2677        ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2678        xfs_iext_add(ifp, idx, count);
2679        for (i = idx; i < idx + count; i++, new++)
2680                xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
2681}
2682
2683/*
2684 * This is called when the amount of space required for incore file
2685 * extents needs to be increased. The ext_diff parameter stores the
2686 * number of new extents being added and the idx parameter contains
2687 * the extent index where the new extents will be added. If the new
2688 * extents are being appended, then we just need to (re)allocate and
2689 * initialize the space. Otherwise, if the new extents are being
2690 * inserted into the middle of the existing entries, a bit more work
2691 * is required to make room for the new extents to be inserted. The
2692 * caller is responsible for filling in the new extent entries upon
2693 * return.
2694 */
2695void
2696xfs_iext_add(
2697        xfs_ifork_t     *ifp,           /* inode fork pointer */
2698        xfs_extnum_t    idx,            /* index to begin adding exts */
2699        int             ext_diff)       /* number of extents to add */
2700{
2701        int             byte_diff;      /* new bytes being added */
2702        int             new_size;       /* size of extents after adding */
2703        xfs_extnum_t    nextents;       /* number of extents in file */
2704
2705        nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2706        ASSERT((idx >= 0) && (idx <= nextents));
2707        byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
2708        new_size = ifp->if_bytes + byte_diff;
2709        /*
2710         * If the new number of extents (nextents + ext_diff)
2711         * fits inside the inode, then continue to use the inline
2712         * extent buffer.
2713         */
2714        if (nextents + ext_diff <= XFS_INLINE_EXTS) {
2715                if (idx < nextents) {
2716                        memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
2717                                &ifp->if_u2.if_inline_ext[idx],
2718                                (nextents - idx) * sizeof(xfs_bmbt_rec_t));
2719                        memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
2720                }
2721                ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
2722                ifp->if_real_bytes = 0;
2723        }
2724        /*
2725         * Otherwise use a linear (direct) extent list.
2726         * If the extents are currently inside the inode,
2727         * xfs_iext_realloc_direct will switch us from
2728         * inline to direct extent allocation mode.
2729         */
2730        else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
2731                xfs_iext_realloc_direct(ifp, new_size);
2732                if (idx < nextents) {
2733                        memmove(&ifp->if_u1.if_extents[idx + ext_diff],
2734                                &ifp->if_u1.if_extents[idx],
2735                                (nextents - idx) * sizeof(xfs_bmbt_rec_t));
2736                        memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
2737                }
2738        }
2739        /* Indirection array */
2740        else {
2741                xfs_ext_irec_t  *erp;
2742                int             erp_idx = 0;
2743                int             page_idx = idx;
2744
2745                ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
2746                if (ifp->if_flags & XFS_IFEXTIREC) {
2747                        erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
2748                } else {
2749                        xfs_iext_irec_init(ifp);
2750                        ASSERT(ifp->if_flags & XFS_IFEXTIREC);
2751                        erp = ifp->if_u1.if_ext_irec;
2752                }
2753                /* Extents fit in target extent page */
2754                if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
2755                        if (page_idx < erp->er_extcount) {
2756                                memmove(&erp->er_extbuf[page_idx + ext_diff],
2757                                        &erp->er_extbuf[page_idx],
2758                                        (erp->er_extcount - page_idx) *
2759                                        sizeof(xfs_bmbt_rec_t));
2760                                memset(&erp->er_extbuf[page_idx], 0, byte_diff);
2761                        }
2762                        erp->er_extcount += ext_diff;
2763                        xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
2764                }
2765                /* Insert a new extent page */
2766                else if (erp) {
2767                        xfs_iext_add_indirect_multi(ifp,
2768                                erp_idx, page_idx, ext_diff);
2769                }
2770                /*
2771                 * If extent(s) are being appended to the last page in
2772                 * the indirection array and the new extent(s) don't fit
2773                 * in the page, then erp is NULL and erp_idx is set to
2774                 * the next index needed in the indirection array.
2775                 */
2776                else {
2777                        int     count = ext_diff;
2778
2779                        while (count) {
2780                                erp = xfs_iext_irec_new(ifp, erp_idx);
2781                                erp->er_extcount = count;
2782                                count -= MIN(count, (int)XFS_LINEAR_EXTS);
2783                                if (count) {
2784                                        erp_idx++;
2785                                }
2786                        }
2787                }
2788        }
2789        ifp->if_bytes = new_size;
2790}
2791
2792/*
2793 * This is called when incore extents are being added to the indirection
2794 * array and the new extents do not fit in the target extent list. The
2795 * erp_idx parameter contains the irec index for the target extent list
2796 * in the indirection array, and the idx parameter contains the extent
2797 * index within the list. The number of extents being added is stored
2798 * in the count parameter.
2799 *
2800 *    |-------|   |-------|
2801 *    |       |   |       |    idx - number of extents before idx
2802 *    |  idx  |   | count |
2803 *    |       |   |       |    count - number of extents being inserted at idx
2804 *    |-------|   |-------|
2805 *    | count |   | nex2  |    nex2 - number of extents after idx + count
2806 *    |-------|   |-------|
2807 */
2808void
2809xfs_iext_add_indirect_multi(
2810        xfs_ifork_t     *ifp,                   /* inode fork pointer */
2811        int             erp_idx,                /* target extent irec index */
2812        xfs_extnum_t    idx,                    /* index within target list */
2813        int             count)                  /* new extents being added */
2814{
2815        int             byte_diff;              /* new bytes being added */
2816        xfs_ext_irec_t  *erp;                   /* pointer to irec entry */
2817        xfs_extnum_t    ext_diff;               /* number of extents to add */
2818        xfs_extnum_t    ext_cnt;                /* new extents still needed */
2819        xfs_extnum_t    nex2;                   /* extents after idx + count */
2820        xfs_bmbt_rec_t  *nex2_ep = NULL;        /* temp list for nex2 extents */
2821        int             nlists;                 /* number of irec's (lists) */
2822
2823        ASSERT(ifp->if_flags & XFS_IFEXTIREC);
2824        erp = &ifp->if_u1.if_ext_irec[erp_idx];
2825        nex2 = erp->er_extcount - idx;
2826        nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
2827
2828        /*
2829         * Save second part of target extent list
2830         * (all extents past */
2831        if (nex2) {
2832                byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
2833                nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS);
2834                memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
2835                erp->er_extcount -= nex2;
2836                xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
2837                memset(&erp->er_extbuf[idx], 0, byte_diff);
2838        }
2839
2840        /*
2841         * Add the new extents to the end of the target
2842         * list, then allocate new irec record(s) and
2843         * extent buffer(s) as needed to store the rest
2844         * of the new extents.
2845         */
2846        ext_cnt = count;
2847        ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
2848        if (ext_diff) {
2849                erp->er_extcount += ext_diff;
2850                xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
2851                ext_cnt -= ext_diff;
2852        }
2853        while (ext_cnt) {
2854                erp_idx++;
2855                erp = xfs_iext_irec_new(ifp, erp_idx);
2856                ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
2857                erp->er_extcount = ext_diff;
2858                xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
2859                ext_cnt -= ext_diff;
2860        }
2861
2862        /* Add nex2 extents back to indirection array */
2863        if (nex2) {
2864                xfs_extnum_t    ext_avail;
2865                int             i;
2866
2867                byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
2868                ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
2869                i = 0;
2870                /*
2871                 * If nex2 extents fit in the current page, append
2872                 * nex2_ep after the new extents.
2873                 */
2874                if (nex2 <= ext_avail) {
2875                        i = erp->er_extcount;
2876                }
2877                /*
2878                 * Otherwise, check if space is available in the
2879                 * next page.
2880                 */
2881                else if ((erp_idx < nlists - 1) &&
2882                         (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
2883                          ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
2884                        erp_idx++;
2885                        erp++;
2886                        /* Create a hole for nex2 extents */
2887                        memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
2888                                erp->er_extcount * sizeof(xfs_bmbt_rec_t));
2889                }
2890                /*
2891                 * Final choice, create a new extent page for
2892                 * nex2 extents.
2893                 */
2894                else {
2895                        erp_idx++;
2896                        erp = xfs_iext_irec_new(ifp, erp_idx);
2897                }
2898                memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
2899                kmem_free(nex2_ep);
2900                erp->er_extcount += nex2;
2901                xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
2902        }
2903}
2904
2905/*
2906 * This is called when the amount of space required for incore file
2907 * extents needs to be decreased. The ext_diff parameter stores the
2908 * number of extents to be removed and the idx parameter contains
2909 * the extent index where the extents will be removed from.
2910 *
2911 * If the amount of space needed has decreased below the linear
2912 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
2913 * extent array.  Otherwise, use kmem_realloc() to adjust the
2914 * size to what is needed.
2915 */
2916void
2917xfs_iext_remove(
2918        xfs_inode_t     *ip,            /* incore inode pointer */
2919        xfs_extnum_t    idx,            /* index to begin removing exts */
2920        int             ext_diff,       /* number of extents to remove */
2921        int             state)          /* type of extent conversion */
2922{
2923        xfs_ifork_t     *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
2924        xfs_extnum_t    nextents;       /* number of extents in file */
2925        int             new_size;       /* size of extents after removal */
2926
2927        trace_xfs_iext_remove(ip, idx, state, _RET_IP_);
2928
2929        ASSERT(ext_diff > 0);
2930        nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2931        new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
2932
2933        if (new_size == 0) {
2934                xfs_iext_destroy(ifp);
2935        } else if (ifp->if_flags & XFS_IFEXTIREC) {
2936                xfs_iext_remove_indirect(ifp, idx, ext_diff);
2937        } else if (ifp->if_real_bytes) {
2938                xfs_iext_remove_direct(ifp, idx, ext_diff);
2939        } else {
2940                xfs_iext_remove_inline(ifp, idx, ext_diff);
2941        }
2942        ifp->if_bytes = new_size;
2943}
2944
2945/*
2946 * This removes ext_diff extents from the inline buffer, beginning
2947 * at extent index idx.
2948 */
2949void
2950xfs_iext_remove_inline(
2951        xfs_ifork_t     *ifp,           /* inode fork pointer */
2952        xfs_extnum_t    idx,            /* index to begin removing exts */
2953        int             ext_diff)       /* number of extents to remove */
2954{
2955        int             nextents;       /* number of extents in file */
2956
2957        ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
2958        ASSERT(idx < XFS_INLINE_EXTS);
2959        nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2960        ASSERT(((nextents - ext_diff) > 0) &&
2961                (nextents - ext_diff) < XFS_INLINE_EXTS);
2962
2963        if (idx + ext_diff < nextents) {
2964                memmove(&ifp->if_u2.if_inline_ext[idx],
2965                        &ifp->if_u2.if_inline_ext[idx + ext_diff],
2966                        (nextents - (idx + ext_diff)) *
2967                         sizeof(xfs_bmbt_rec_t));
2968                memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
2969                        0, ext_diff * sizeof(xfs_bmbt_rec_t));
2970        } else {
2971                memset(&ifp->if_u2.if_inline_ext[idx], 0,
2972                        ext_diff * sizeof(xfs_bmbt_rec_t));
2973        }
2974}
2975
2976/*
2977 * This removes ext_diff extents from a linear (direct) extent list,
2978 * beginning at extent index idx. If the extents are being removed
2979 * from the end of the list (ie. truncate) then we just need to re-
2980 * allocate the list to remove the extra space. Otherwise, if the
2981 * extents are being removed from the middle of the existing extent
2982 * entries, then we first need to move the extent records beginning
2983 * at idx + ext_diff up in the list to overwrite the records being
2984 * removed, then remove the extra space via kmem_realloc.
2985 */
2986void
2987xfs_iext_remove_direct(
2988        xfs_ifork_t     *ifp,           /* inode fork pointer */
2989        xfs_extnum_t    idx,            /* index to begin removing exts */
2990        int             ext_diff)       /* number of extents to remove */
2991{
2992        xfs_extnum_t    nextents;       /* number of extents in file */
2993        int             new_size;       /* size of extents after removal */
2994
2995        ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
2996        new_size = ifp->if_bytes -
2997                (ext_diff * sizeof(xfs_bmbt_rec_t));
2998        nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2999
3000        if (new_size == 0) {
3001                xfs_iext_destroy(ifp);
3002                return;
3003        }
3004        /* Move extents up in the list (if needed) */
3005        if (idx + ext_diff < nextents) {
3006                memmove(&ifp->if_u1.if_extents[idx],
3007                        &ifp->if_u1.if_extents[idx + ext_diff],
3008                        (nextents - (idx + ext_diff)) *
3009                         sizeof(xfs_bmbt_rec_t));
3010        }
3011        memset(&ifp->if_u1.if_extents[nextents - ext_diff],
3012                0, ext_diff * sizeof(xfs_bmbt_rec_t));
3013        /*
3014         * Reallocate the direct extent list. If the extents
3015         * will fit inside the inode then xfs_iext_realloc_direct
3016         * will switch from direct to inline extent allocation
3017         * mode for us.
3018         */
3019        xfs_iext_realloc_direct(ifp, new_size);
3020        ifp->if_bytes = new_size;
3021}
3022
3023/*
3024 * This is called when incore extents are being removed from the
3025 * indirection array and the extents being removed span multiple extent
3026 * buffers. The idx parameter contains the file extent index where we
3027 * want to begin removing extents, and the count parameter contains
3028 * how many extents need to be removed.
3029 *
3030 *    |-------|   |-------|
3031 *    | nex1  |   |       |    nex1 - number of extents before idx
3032 *    |-------|   | count |
3033 *    |       |   |       |    count - number of extents being removed at idx
3034 *    | count |   |-------|
3035 *    |       |   | nex2  |    nex2 - number of extents after idx + count
3036 *    |-------|   |-------|
3037 */
3038void
3039xfs_iext_remove_indirect(
3040        xfs_ifork_t     *ifp,           /* inode fork pointer */
3041        xfs_extnum_t    idx,            /* index to begin removing extents */
3042        int             count)          /* number of extents to remove */
3043{
3044        xfs_ext_irec_t  *erp;           /* indirection array pointer */
3045        int             erp_idx = 0;    /* indirection array index */
3046        xfs_extnum_t    ext_cnt;        /* extents left to remove */
3047        xfs_extnum_t    ext_diff;       /* extents to remove in current list */
3048        xfs_extnum_t    nex1;           /* number of extents before idx */
3049        xfs_extnum_t    nex2;           /* extents after idx + count */
3050        int             page_idx = idx; /* index in target extent list */
3051
3052        ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3053        erp = xfs_iext_idx_to_irec(ifp,  &page_idx, &erp_idx, 0);
3054        ASSERT(erp != NULL);
3055        nex1 = page_idx;
3056        ext_cnt = count;
3057        while (ext_cnt) {
3058                nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
3059                ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
3060                /*
3061                 * Check for deletion of entire list;
3062                 * xfs_iext_irec_remove() updates extent offsets.
3063                 */
3064                if (ext_diff == erp->er_extcount) {
3065                        xfs_iext_irec_remove(ifp, erp_idx);
3066                        ext_cnt -= ext_diff;
3067                        nex1 = 0;
3068                        if (ext_cnt) {
3069                                ASSERT(erp_idx < ifp->if_real_bytes /
3070                                        XFS_IEXT_BUFSZ);
3071                                erp = &ifp->if_u1.if_ext_irec[erp_idx];
3072                                nex1 = 0;
3073                                continue;
3074                        } else {
3075                                break;
3076                        }
3077                }
3078                /* Move extents up (if needed) */
3079                if (nex2) {
3080                        memmove(&erp->er_extbuf[nex1],
3081                                &erp->er_extbuf[nex1 + ext_diff],
3082                                nex2 * sizeof(xfs_bmbt_rec_t));
3083                }
3084                /* Zero out rest of page */
3085                memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
3086                        ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
3087                /* Update remaining counters */
3088                erp->er_extcount -= ext_diff;
3089                xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
3090                ext_cnt -= ext_diff;
3091                nex1 = 0;
3092                erp_idx++;
3093                erp++;
3094        }
3095        ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
3096        xfs_iext_irec_compact(ifp);
3097}
3098
3099/*
3100 * Create, destroy, or resize a linear (direct) block of extents.
3101 */
3102void
3103xfs_iext_realloc_direct(
3104        xfs_ifork_t     *ifp,           /* inode fork pointer */
3105        int             new_size)       /* new size of extents */
3106{
3107        int             rnew_size;      /* real new size of extents */
3108
3109        rnew_size = new_size;
3110
3111        ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
3112                ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
3113                 (new_size != ifp->if_real_bytes)));
3114
3115        /* Free extent records */
3116        if (new_size == 0) {
3117                xfs_iext_destroy(ifp);
3118        }
3119        /* Resize direct extent list and zero any new bytes */
3120        else if (ifp->if_real_bytes) {
3121                /* Check if extents will fit inside the inode */
3122                if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
3123                        xfs_iext_direct_to_inline(ifp, new_size /
3124                                (uint)sizeof(xfs_bmbt_rec_t));
3125                        ifp->if_bytes = new_size;
3126                        return;
3127                }
3128                if (!is_power_of_2(new_size)){
3129                        rnew_size = roundup_pow_of_two(new_size);
3130                }
3131                if (rnew_size != ifp->if_real_bytes) {
3132                        ifp->if_u1.if_extents =
3133                                kmem_realloc(ifp->if_u1.if_extents,
3134                                                rnew_size,
3135                                                ifp->if_real_bytes, KM_NOFS);
3136                }
3137                if (rnew_size > ifp->if_real_bytes) {
3138                        memset(&ifp->if_u1.if_extents[ifp->if_bytes /
3139                                (uint)sizeof(xfs_bmbt_rec_t)], 0,
3140                                rnew_size - ifp->if_real_bytes);
3141                }
3142        }
3143        /*
3144         * Switch from the inline extent buffer to a direct
3145         * extent list. Be sure to include the inline extent
3146         * bytes in new_size.
3147         */
3148        else {
3149                new_size += ifp->if_bytes;
3150                if (!is_power_of_2(new_size)) {
3151                        rnew_size = roundup_pow_of_two(new_size);
3152                }
3153                xfs_iext_inline_to_direct(ifp, rnew_size);
3154        }
3155        ifp->if_real_bytes = rnew_size;
3156        ifp->if_bytes = new_size;
3157}
3158
3159/*
3160 * Switch from linear (direct) extent records to inline buffer.
3161 */
3162void
3163xfs_iext_direct_to_inline(
3164        xfs_ifork_t     *ifp,           /* inode fork pointer */
3165        xfs_extnum_t    nextents)       /* number of extents in file */
3166{
3167        ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3168        ASSERT(nextents <= XFS_INLINE_EXTS);
3169        /*
3170         * The inline buffer was zeroed when we switched
3171         * from inline to direct extent allocation mode,
3172         * so we don't need to clear it here.
3173         */
3174        memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
3175                nextents * sizeof(xfs_bmbt_rec_t));
3176        kmem_free(ifp->if_u1.if_extents);
3177        ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3178        ifp->if_real_bytes = 0;
3179}
3180
3181/*
3182 * Switch from inline buffer to linear (direct) extent records.
3183 * new_size should already be rounded up to the next power of 2
3184 * by the caller (when appropriate), so use new_size as it is.
3185 * However, since new_size may be rounded up, we can't update
3186 * if_bytes here. It is the caller's responsibility to update
3187 * if_bytes upon return.
3188 */
3189void
3190xfs_iext_inline_to_direct(
3191        xfs_ifork_t     *ifp,           /* inode fork pointer */
3192        int             new_size)       /* number of extents in file */
3193{
3194        ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS);
3195        memset(ifp->if_u1.if_extents, 0, new_size);
3196        if (ifp->if_bytes) {
3197                memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
3198                        ifp->if_bytes);
3199                memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3200                        sizeof(xfs_bmbt_rec_t));
3201        }
3202        ifp->if_real_bytes = new_size;
3203}
3204
3205/*
3206 * Resize an extent indirection array to new_size bytes.
3207 */
3208STATIC void
3209xfs_iext_realloc_indirect(
3210        xfs_ifork_t     *ifp,           /* inode fork pointer */
3211        int             new_size)       /* new indirection array size */
3212{
3213        int             nlists;         /* number of irec's (ex lists) */
3214        int             size;           /* current indirection array size */
3215
3216        ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3217        nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3218        size = nlists * sizeof(xfs_ext_irec_t);
3219        ASSERT(ifp->if_real_bytes);
3220        ASSERT((new_size >= 0) && (new_size != size));
3221        if (new_size == 0) {
3222                xfs_iext_destroy(ifp);
3223        } else {
3224                ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
3225                        kmem_realloc(ifp->if_u1.if_ext_irec,
3226                                new_size, size, KM_NOFS);
3227        }
3228}
3229
3230/*
3231 * Switch from indirection array to linear (direct) extent allocations.
3232 */
3233STATIC void
3234xfs_iext_indirect_to_direct(
3235         xfs_ifork_t    *ifp)           /* inode fork pointer */
3236{
3237        xfs_bmbt_rec_host_t *ep;        /* extent record pointer */
3238        xfs_extnum_t    nextents;       /* number of extents in file */
3239        int             size;           /* size of file extents */
3240
3241        ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3242        nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3243        ASSERT(nextents <= XFS_LINEAR_EXTS);
3244        size = nextents * sizeof(xfs_bmbt_rec_t);
3245
3246        xfs_iext_irec_compact_pages(ifp);
3247        ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
3248
3249        ep = ifp->if_u1.if_ext_irec->er_extbuf;
3250        kmem_free(ifp->if_u1.if_ext_irec);
3251        ifp->if_flags &= ~XFS_IFEXTIREC;
3252        ifp->if_u1.if_extents = ep;
3253        ifp->if_bytes = size;
3254        if (nextents < XFS_LINEAR_EXTS) {
3255                xfs_iext_realloc_direct(ifp, size);
3256        }
3257}
3258
3259/*
3260 * Free incore file extents.
3261 */
3262void
3263xfs_iext_destroy(
3264        xfs_ifork_t     *ifp)           /* inode fork pointer */
3265{
3266        if (ifp->if_flags & XFS_IFEXTIREC) {
3267                int     erp_idx;
3268                int     nlists;
3269
3270                nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3271                for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
3272                        xfs_iext_irec_remove(ifp, erp_idx);
3273                }
3274                ifp->if_flags &= ~XFS_IFEXTIREC;
3275        } else if (ifp->if_real_bytes) {
3276                kmem_free(ifp->if_u1.if_extents);
3277        } else if (ifp->if_bytes) {
3278                memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3279                        sizeof(xfs_bmbt_rec_t));
3280        }
3281        ifp->if_u1.if_extents = NULL;
3282        ifp->if_real_bytes = 0;
3283        ifp->if_bytes = 0;
3284}
3285
3286/*
3287 * Return a pointer to the extent record for file system block bno.
3288 */
3289xfs_bmbt_rec_host_t *                   /* pointer to found extent record */
3290xfs_iext_bno_to_ext(
3291        xfs_ifork_t     *ifp,           /* inode fork pointer */
3292        xfs_fileoff_t   bno,            /* block number to search for */
3293        xfs_extnum_t    *idxp)          /* index of target extent */
3294{
3295        xfs_bmbt_rec_host_t *base;      /* pointer to first extent */
3296        xfs_filblks_t   blockcount = 0; /* number of blocks in extent */
3297        xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */
3298        xfs_ext_irec_t  *erp = NULL;    /* indirection array pointer */
3299        int             high;           /* upper boundary in search */
3300        xfs_extnum_t    idx = 0;        /* index of target extent */
3301        int             low;            /* lower boundary in search */
3302        xfs_extnum_t    nextents;       /* number of file extents */
3303        xfs_fileoff_t   startoff = 0;   /* start offset of extent */
3304
3305        nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3306        if (nextents == 0) {
3307                *idxp = 0;
3308                return NULL;
3309        }
3310        low = 0;
3311        if (ifp->if_flags & XFS_IFEXTIREC) {
3312                /* Find target extent list */
3313                int     erp_idx = 0;
3314                erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
3315                base = erp->er_extbuf;
3316                high = erp->er_extcount - 1;
3317        } else {
3318                base = ifp->if_u1.if_extents;
3319                high = nextents - 1;
3320        }
3321        /* Binary search extent records */
3322        while (low <= high) {
3323                idx = (low + high) >> 1;
3324                ep = base + idx;
3325                startoff = xfs_bmbt_get_startoff(ep);
3326                blockcount = xfs_bmbt_get_blockcount(ep);
3327                if (bno < startoff) {
3328                        high = idx - 1;
3329                } else if (bno >= startoff + blockcount) {
3330                        low = idx + 1;
3331                } else {
3332                        /* Convert back to file-based extent index */
3333                        if (ifp->if_flags & XFS_IFEXTIREC) {
3334                                idx += erp->er_extoff;
3335                        }
3336                        *idxp = idx;
3337                        return ep;
3338                }
3339        }
3340        /* Convert back to file-based extent index */
3341        if (ifp->if_flags & XFS_IFEXTIREC) {
3342                idx += erp->er_extoff;
3343        }
3344        if (bno >= startoff + blockcount) {
3345                if (++idx == nextents) {
3346                        ep = NULL;
3347                } else {
3348                        ep = xfs_iext_get_ext(ifp, idx);
3349                }
3350        }
3351        *idxp = idx;
3352        return ep;
3353}
3354
3355/*
3356 * Return a pointer to the indirection array entry containing the
3357 * extent record for filesystem block bno. Store the index of the
3358 * target irec in *erp_idxp.
3359 */
3360xfs_ext_irec_t *                        /* pointer to found extent record */
3361xfs_iext_bno_to_irec(
3362        xfs_ifork_t     *ifp,           /* inode fork pointer */
3363        xfs_fileoff_t   bno,            /* block number to search for */
3364        int             *erp_idxp)      /* irec index of target ext list */
3365{
3366        xfs_ext_irec_t  *erp = NULL;    /* indirection array pointer */
3367        xfs_ext_irec_t  *erp_next;      /* next indirection array entry */
3368        int             erp_idx;        /* indirection array index */
3369        int             nlists;         /* number of extent irec's (lists) */
3370        int             high;           /* binary search upper limit */
3371        int             low;            /* binary search lower limit */
3372
3373        ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3374        nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3375        erp_idx = 0;
3376        low = 0;
3377        high = nlists - 1;
3378        while (low <= high) {
3379                erp_idx = (low + high) >> 1;
3380                erp = &ifp->if_u1.if_ext_irec[erp_idx];
3381                erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
3382                if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
3383                        high = erp_idx - 1;
3384                } else if (erp_next && bno >=
3385                           xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
3386                        low = erp_idx + 1;
3387                } else {
3388                        break;
3389                }
3390        }
3391        *erp_idxp = erp_idx;
3392        return erp;
3393}
3394
3395/*
3396 * Return a pointer to the indirection array entry containing the
3397 * extent record at file extent index *idxp. Store the index of the
3398 * target irec in *erp_idxp and store the page index of the target
3399 * extent record in *idxp.
3400 */
3401xfs_ext_irec_t *
3402xfs_iext_idx_to_irec(
3403        xfs_ifork_t     *ifp,           /* inode fork pointer */
3404        xfs_extnum_t    *idxp,          /* extent index (file -> page) */
3405        int             *erp_idxp,      /* pointer to target irec */
3406        int             realloc)        /* new bytes were just added */
3407{
3408        xfs_ext_irec_t  *prev;          /* pointer to previous irec */
3409        xfs_ext_irec_t  *erp = NULL;    /* pointer to current irec */
3410        int             erp_idx;        /* indirection array index */
3411        int             nlists;         /* number of irec's (ex lists) */
3412        int             high;           /* binary search upper limit */
3413        int             low;            /* binary search lower limit */
3414        xfs_extnum_t    page_idx = *idxp; /* extent index in target list */
3415
3416        ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3417        ASSERT(page_idx >= 0);
3418        ASSERT(page_idx <= ifp->if_bytes / sizeof(xfs_bmbt_rec_t));
3419        ASSERT(page_idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t) || realloc);
3420
3421        nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3422        erp_idx = 0;
3423        low = 0;
3424        high = nlists - 1;
3425
3426        /* Binary search extent irec's */
3427        while (low <= high) {
3428                erp_idx = (low + high) >> 1;
3429                erp = &ifp->if_u1.if_ext_irec[erp_idx];
3430                prev = erp_idx > 0 ? erp - 1 : NULL;
3431                if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
3432                     realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
3433                        high = erp_idx - 1;
3434                } else if (page_idx > erp->er_extoff + erp->er_extcount ||
3435                           (page_idx == erp->er_extoff + erp->er_extcount &&
3436                            !realloc)) {
3437                        low = erp_idx + 1;
3438                } else if (page_idx == erp->er_extoff + erp->er_extcount &&
3439                           erp->er_extcount == XFS_LINEAR_EXTS) {
3440                        ASSERT(realloc);
3441                        page_idx = 0;
3442                        erp_idx++;
3443                        erp = erp_idx < nlists ? erp + 1 : NULL;
3444                        break;
3445                } else {
3446                        page_idx -= erp->er_extoff;
3447                        break;
3448                }
3449        }
3450        *idxp = page_idx;
3451        *erp_idxp = erp_idx;
3452        return(erp);
3453}
3454
3455/*
3456 * Allocate and initialize an indirection array once the space needed
3457 * for incore extents increases above XFS_IEXT_BUFSZ.
3458 */
3459void
3460xfs_iext_irec_init(
3461        xfs_ifork_t     *ifp)           /* inode fork pointer */
3462{
3463        xfs_ext_irec_t  *erp;           /* indirection array pointer */
3464        xfs_extnum_t    nextents;       /* number of extents in file */
3465
3466        ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3467        nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3468        ASSERT(nextents <= XFS_LINEAR_EXTS);
3469
3470        erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS);
3471
3472        if (nextents == 0) {
3473                ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
3474        } else if (!ifp->if_real_bytes) {
3475                xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
3476        } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
3477                xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
3478        }
3479        erp->er_extbuf = ifp->if_u1.if_extents;
3480        erp->er_extcount = nextents;
3481        erp->er_extoff = 0;
3482
3483        ifp->if_flags |= XFS_IFEXTIREC;
3484        ifp->if_real_bytes = XFS_IEXT_BUFSZ;
3485        ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
3486        ifp->if_u1.if_ext_irec = erp;
3487
3488        return;
3489}
3490
3491/*
3492 * Allocate and initialize a new entry in the indirection array.
3493 */
3494xfs_ext_irec_t *
3495xfs_iext_irec_new(
3496        xfs_ifork_t     *ifp,           /* inode fork pointer */
3497        int             erp_idx)        /* index for new irec */
3498{
3499        xfs_ext_irec_t  *erp;           /* indirection array pointer */
3500        int             i;              /* loop counter */
3501        int             nlists;         /* number of irec's (ex lists) */
3502
3503        ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3504        nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3505
3506        /* Resize indirection array */
3507        xfs_iext_realloc_indirect(ifp, ++nlists *
3508                                  sizeof(xfs_ext_irec_t));
3509        /*
3510         * Move records down in the array so the
3511         * new page can use erp_idx.
3512         */
3513        erp = ifp->if_u1.if_ext_irec;
3514        for (i = nlists - 1; i > erp_idx; i--) {
3515                memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
3516        }
3517        ASSERT(i == erp_idx);
3518
3519        /* Initialize new extent record */
3520        erp = ifp->if_u1.if_ext_irec;
3521        erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
3522        ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
3523        memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
3524        erp[erp_idx].er_extcount = 0;
3525        erp[erp_idx].er_extoff = erp_idx > 0 ?
3526                erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
3527        return (&erp[erp_idx]);
3528}
3529
3530/*
3531 * Remove a record from the indirection array.
3532 */
3533void
3534xfs_iext_irec_remove(
3535        xfs_ifork_t     *ifp,           /* inode fork pointer */
3536        int             erp_idx)        /* irec index to remove */
3537{
3538        xfs_ext_irec_t  *erp;           /* indirection array pointer */
3539        int             i;              /* loop counter */
3540        int             nlists;         /* number of irec's (ex lists) */
3541
3542        ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3543        nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3544        erp = &ifp->if_u1.if_ext_irec[erp_idx];
3545        if (erp->er_extbuf) {
3546                xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
3547                        -erp->er_extcount);
3548                kmem_free(erp->er_extbuf);
3549        }
3550        /* Compact extent records */
3551        erp = ifp->if_u1.if_ext_irec;
3552        for (i = erp_idx; i < nlists - 1; i++) {
3553                memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
3554        }
3555        /*
3556         * Manually free the last extent record from the indirection
3557         * array.  A call to xfs_iext_realloc_indirect() with a size
3558         * of zero would result in a call to xfs_iext_destroy() which
3559         * would in turn call this function again, creating a nasty
3560         * infinite loop.
3561         */
3562        if (--nlists) {
3563                xfs_iext_realloc_indirect(ifp,
3564                        nlists * sizeof(xfs_ext_irec_t));
3565        } else {
3566                kmem_free(ifp->if_u1.if_ext_irec);
3567        }
3568        ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
3569}
3570
3571/*
3572 * This is called to clean up large amounts of unused memory allocated
3573 * by the indirection array.  Before compacting anything though, verify
3574 * that the indirection array is still needed and switch back to the
3575 * linear extent list (or even the inline buffer) if possible.  The
3576 * compaction policy is as follows:
3577 *
3578 *    Full Compaction: Extents fit into a single page (or inline buffer)
3579 * Partial Compaction: Extents occupy less than 50% of allocated space
3580 *      No Compaction: Extents occupy at least 50% of allocated space
3581 */
3582void
3583xfs_iext_irec_compact(
3584        xfs_ifork_t     *ifp)           /* inode fork pointer */
3585{
3586        xfs_extnum_t    nextents;       /* number of extents in file */
3587        int             nlists;         /* number of irec's (ex lists) */
3588
3589        ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3590        nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3591        nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3592
3593        if (nextents == 0) {
3594                xfs_iext_destroy(ifp);
3595        } else if (nextents <= XFS_INLINE_EXTS) {
3596                xfs_iext_indirect_to_direct(ifp);
3597                xfs_iext_direct_to_inline(ifp, nextents);
3598        } else if (nextents <= XFS_LINEAR_EXTS) {
3599                xfs_iext_indirect_to_direct(ifp);
3600        } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
3601                xfs_iext_irec_compact_pages(ifp);
3602        }
3603}
3604
3605/*
3606 * Combine extents from neighboring extent pages.
3607 */
3608void
3609xfs_iext_irec_compact_pages(
3610        xfs_ifork_t     *ifp)           /* inode fork pointer */
3611{
3612        xfs_ext_irec_t  *erp, *erp_next;/* pointers to irec entries */
3613        int             erp_idx = 0;    /* indirection array index */
3614        int             nlists;         /* number of irec's (ex lists) */
3615
3616        ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3617        nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3618        while (erp_idx < nlists - 1) {
3619                erp = &ifp->if_u1.if_ext_irec[erp_idx];
3620                erp_next = erp + 1;
3621                if (erp_next->er_extcount <=
3622                    (XFS_LINEAR_EXTS - erp->er_extcount)) {
3623                        memcpy(&erp->er_extbuf[erp->er_extcount],
3624                                erp_next->er_extbuf, erp_next->er_extcount *
3625                                sizeof(xfs_bmbt_rec_t));
3626                        erp->er_extcount += erp_next->er_extcount;
3627                        /*
3628                         * Free page before removing extent record
3629                         * so er_extoffs don't get modified in
3630                         * xfs_iext_irec_remove.
3631                         */
3632                        kmem_free(erp_next->er_extbuf);
3633                        erp_next->er_extbuf = NULL;
3634                        xfs_iext_irec_remove(ifp, erp_idx + 1);
3635                        nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3636                } else {
3637                        erp_idx++;
3638                }
3639        }
3640}
3641
3642/*
3643 * This is called to update the er_extoff field in the indirection
3644 * array when extents have been added or removed from one of the
3645 * extent lists. erp_idx contains the irec index to begin updating
3646 * at and ext_diff contains the number of extents that were added
3647 * or removed.
3648 */
3649void
3650xfs_iext_irec_update_extoffs(
3651        xfs_ifork_t     *ifp,           /* inode fork pointer */
3652        int             erp_idx,        /* irec index to update */
3653        int             ext_diff)       /* number of new extents */
3654{
3655        int             i;              /* loop counter */
3656        int             nlists;         /* number of irec's (ex lists */
3657
3658        ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3659        nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3660        for (i = erp_idx; i < nlists; i++) {
3661                ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;
3662        }
3663}
3664
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.