linux/fs/xfs/xfs_trans.c
<<
>>
Prefs
   1/*
   2 * Copyright (c) 2000-2003,2005 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 "xfs.h"
  19#include "xfs_fs.h"
  20#include "xfs_types.h"
  21#include "xfs_bit.h"
  22#include "xfs_log.h"
  23#include "xfs_inum.h"
  24#include "xfs_trans.h"
  25#include "xfs_sb.h"
  26#include "xfs_ag.h"
  27#include "xfs_dir2.h"
  28#include "xfs_dmapi.h"
  29#include "xfs_mount.h"
  30#include "xfs_error.h"
  31#include "xfs_da_btree.h"
  32#include "xfs_bmap_btree.h"
  33#include "xfs_alloc_btree.h"
  34#include "xfs_ialloc_btree.h"
  35#include "xfs_dir2_sf.h"
  36#include "xfs_attr_sf.h"
  37#include "xfs_dinode.h"
  38#include "xfs_inode.h"
  39#include "xfs_btree.h"
  40#include "xfs_ialloc.h"
  41#include "xfs_alloc.h"
  42#include "xfs_bmap.h"
  43#include "xfs_quota.h"
  44#include "xfs_trans_priv.h"
  45#include "xfs_trans_space.h"
  46#include "xfs_inode_item.h"
  47
  48
  49STATIC void     xfs_trans_apply_sb_deltas(xfs_trans_t *);
  50STATIC uint     xfs_trans_count_vecs(xfs_trans_t *);
  51STATIC void     xfs_trans_fill_vecs(xfs_trans_t *, xfs_log_iovec_t *);
  52STATIC void     xfs_trans_uncommit(xfs_trans_t *, uint);
  53STATIC void     xfs_trans_committed(xfs_trans_t *, int);
  54STATIC void     xfs_trans_chunk_committed(xfs_log_item_chunk_t *, xfs_lsn_t, int);
  55STATIC void     xfs_trans_free(xfs_trans_t *);
  56
  57kmem_zone_t     *xfs_trans_zone;
  58
  59
  60/*
  61 * Reservation functions here avoid a huge stack in xfs_trans_init
  62 * due to register overflow from temporaries in the calculations.
  63 */
  64
  65STATIC uint
  66xfs_calc_write_reservation(xfs_mount_t *mp)
  67{
  68        return XFS_CALC_WRITE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
  69}
  70
  71STATIC uint
  72xfs_calc_itruncate_reservation(xfs_mount_t *mp)
  73{
  74        return XFS_CALC_ITRUNCATE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
  75}
  76
  77STATIC uint
  78xfs_calc_rename_reservation(xfs_mount_t *mp)
  79{
  80        return XFS_CALC_RENAME_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
  81}
  82
  83STATIC uint
  84xfs_calc_link_reservation(xfs_mount_t *mp)
  85{
  86        return XFS_CALC_LINK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
  87}
  88
  89STATIC uint
  90xfs_calc_remove_reservation(xfs_mount_t *mp)
  91{
  92        return XFS_CALC_REMOVE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
  93}
  94
  95STATIC uint
  96xfs_calc_symlink_reservation(xfs_mount_t *mp)
  97{
  98        return XFS_CALC_SYMLINK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
  99}
 100
 101STATIC uint
 102xfs_calc_create_reservation(xfs_mount_t *mp)
 103{
 104        return XFS_CALC_CREATE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
 105}
 106
 107STATIC uint
 108xfs_calc_mkdir_reservation(xfs_mount_t *mp)
 109{
 110        return XFS_CALC_MKDIR_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
 111}
 112
 113STATIC uint
 114xfs_calc_ifree_reservation(xfs_mount_t *mp)
 115{
 116        return XFS_CALC_IFREE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
 117}
 118
 119STATIC uint
 120xfs_calc_ichange_reservation(xfs_mount_t *mp)
 121{
 122        return XFS_CALC_ICHANGE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
 123}
 124
 125STATIC uint
 126xfs_calc_growdata_reservation(xfs_mount_t *mp)
 127{
 128        return XFS_CALC_GROWDATA_LOG_RES(mp);
 129}
 130
 131STATIC uint
 132xfs_calc_growrtalloc_reservation(xfs_mount_t *mp)
 133{
 134        return XFS_CALC_GROWRTALLOC_LOG_RES(mp);
 135}
 136
 137STATIC uint
 138xfs_calc_growrtzero_reservation(xfs_mount_t *mp)
 139{
 140        return XFS_CALC_GROWRTZERO_LOG_RES(mp);
 141}
 142
 143STATIC uint
 144xfs_calc_growrtfree_reservation(xfs_mount_t *mp)
 145{
 146        return XFS_CALC_GROWRTFREE_LOG_RES(mp);
 147}
 148
 149STATIC uint
 150xfs_calc_swrite_reservation(xfs_mount_t *mp)
 151{
 152        return XFS_CALC_SWRITE_LOG_RES(mp);
 153}
 154
 155STATIC uint
 156xfs_calc_writeid_reservation(xfs_mount_t *mp)
 157{
 158        return XFS_CALC_WRITEID_LOG_RES(mp);
 159}
 160
 161STATIC uint
 162xfs_calc_addafork_reservation(xfs_mount_t *mp)
 163{
 164        return XFS_CALC_ADDAFORK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
 165}
 166
 167STATIC uint
 168xfs_calc_attrinval_reservation(xfs_mount_t *mp)
 169{
 170        return XFS_CALC_ATTRINVAL_LOG_RES(mp);
 171}
 172
 173STATIC uint
 174xfs_calc_attrset_reservation(xfs_mount_t *mp)
 175{
 176        return XFS_CALC_ATTRSET_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
 177}
 178
 179STATIC uint
 180xfs_calc_attrrm_reservation(xfs_mount_t *mp)
 181{
 182        return XFS_CALC_ATTRRM_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
 183}
 184
 185STATIC uint
 186xfs_calc_clear_agi_bucket_reservation(xfs_mount_t *mp)
 187{
 188        return XFS_CALC_CLEAR_AGI_BUCKET_LOG_RES(mp);
 189}
 190
 191/*
 192 * Initialize the precomputed transaction reservation values
 193 * in the mount structure.
 194 */
 195void
 196xfs_trans_init(
 197        xfs_mount_t     *mp)
 198{
 199        xfs_trans_reservations_t        *resp;
 200
 201        resp = &(mp->m_reservations);
 202        resp->tr_write = xfs_calc_write_reservation(mp);
 203        resp->tr_itruncate = xfs_calc_itruncate_reservation(mp);
 204        resp->tr_rename = xfs_calc_rename_reservation(mp);
 205        resp->tr_link = xfs_calc_link_reservation(mp);
 206        resp->tr_remove = xfs_calc_remove_reservation(mp);
 207        resp->tr_symlink = xfs_calc_symlink_reservation(mp);
 208        resp->tr_create = xfs_calc_create_reservation(mp);
 209        resp->tr_mkdir = xfs_calc_mkdir_reservation(mp);
 210        resp->tr_ifree = xfs_calc_ifree_reservation(mp);
 211        resp->tr_ichange = xfs_calc_ichange_reservation(mp);
 212        resp->tr_growdata = xfs_calc_growdata_reservation(mp);
 213        resp->tr_swrite = xfs_calc_swrite_reservation(mp);
 214        resp->tr_writeid = xfs_calc_writeid_reservation(mp);
 215        resp->tr_addafork = xfs_calc_addafork_reservation(mp);
 216        resp->tr_attrinval = xfs_calc_attrinval_reservation(mp);
 217        resp->tr_attrset = xfs_calc_attrset_reservation(mp);
 218        resp->tr_attrrm = xfs_calc_attrrm_reservation(mp);
 219        resp->tr_clearagi = xfs_calc_clear_agi_bucket_reservation(mp);
 220        resp->tr_growrtalloc = xfs_calc_growrtalloc_reservation(mp);
 221        resp->tr_growrtzero = xfs_calc_growrtzero_reservation(mp);
 222        resp->tr_growrtfree = xfs_calc_growrtfree_reservation(mp);
 223}
 224
 225/*
 226 * This routine is called to allocate a transaction structure.
 227 * The type parameter indicates the type of the transaction.  These
 228 * are enumerated in xfs_trans.h.
 229 *
 230 * Dynamically allocate the transaction structure from the transaction
 231 * zone, initialize it, and return it to the caller.
 232 */
 233xfs_trans_t *
 234xfs_trans_alloc(
 235        xfs_mount_t     *mp,
 236        uint            type)
 237{
 238        xfs_wait_for_freeze(mp, SB_FREEZE_TRANS);
 239        return _xfs_trans_alloc(mp, type);
 240}
 241
 242xfs_trans_t *
 243_xfs_trans_alloc(
 244        xfs_mount_t     *mp,
 245        uint            type)
 246{
 247        xfs_trans_t     *tp;
 248
 249        atomic_inc(&mp->m_active_trans);
 250
 251        tp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP);
 252        tp->t_magic = XFS_TRANS_MAGIC;
 253        tp->t_type = type;
 254        tp->t_mountp = mp;
 255        tp->t_items_free = XFS_LIC_NUM_SLOTS;
 256        tp->t_busy_free = XFS_LBC_NUM_SLOTS;
 257        xfs_lic_init(&(tp->t_items));
 258        XFS_LBC_INIT(&(tp->t_busy));
 259        return tp;
 260}
 261
 262/*
 263 * This is called to create a new transaction which will share the
 264 * permanent log reservation of the given transaction.  The remaining
 265 * unused block and rt extent reservations are also inherited.  This
 266 * implies that the original transaction is no longer allowed to allocate
 267 * blocks.  Locks and log items, however, are no inherited.  They must
 268 * be added to the new transaction explicitly.
 269 */
 270xfs_trans_t *
 271xfs_trans_dup(
 272        xfs_trans_t     *tp)
 273{
 274        xfs_trans_t     *ntp;
 275
 276        ntp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP);
 277
 278        /*
 279         * Initialize the new transaction structure.
 280         */
 281        ntp->t_magic = XFS_TRANS_MAGIC;
 282        ntp->t_type = tp->t_type;
 283        ntp->t_mountp = tp->t_mountp;
 284        ntp->t_items_free = XFS_LIC_NUM_SLOTS;
 285        ntp->t_busy_free = XFS_LBC_NUM_SLOTS;
 286        xfs_lic_init(&(ntp->t_items));
 287        XFS_LBC_INIT(&(ntp->t_busy));
 288
 289        ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
 290        ASSERT(tp->t_ticket != NULL);
 291
 292        ntp->t_flags = XFS_TRANS_PERM_LOG_RES | (tp->t_flags & XFS_TRANS_RESERVE);
 293        ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket);
 294        ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used;
 295        tp->t_blk_res = tp->t_blk_res_used;
 296        ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
 297        tp->t_rtx_res = tp->t_rtx_res_used;
 298        ntp->t_pflags = tp->t_pflags;
 299
 300        xfs_trans_dup_dqinfo(tp, ntp);
 301
 302        atomic_inc(&tp->t_mountp->m_active_trans);
 303        return ntp;
 304}
 305
 306/*
 307 * This is called to reserve free disk blocks and log space for the
 308 * given transaction.  This must be done before allocating any resources
 309 * within the transaction.
 310 *
 311 * This will return ENOSPC if there are not enough blocks available.
 312 * It will sleep waiting for available log space.
 313 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
 314 * is used by long running transactions.  If any one of the reservations
 315 * fails then they will all be backed out.
 316 *
 317 * This does not do quota reservations. That typically is done by the
 318 * caller afterwards.
 319 */
 320int
 321xfs_trans_reserve(
 322        xfs_trans_t     *tp,
 323        uint            blocks,
 324        uint            logspace,
 325        uint            rtextents,
 326        uint            flags,
 327        uint            logcount)
 328{
 329        int             log_flags;
 330        int             error = 0;
 331        int             rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
 332
 333        /* Mark this thread as being in a transaction */
 334        current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
 335
 336        /*
 337         * Attempt to reserve the needed disk blocks by decrementing
 338         * the number needed from the number available.  This will
 339         * fail if the count would go below zero.
 340         */
 341        if (blocks > 0) {
 342                error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FDBLOCKS,
 343                                          -((int64_t)blocks), rsvd);
 344                if (error != 0) {
 345                        current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
 346                        return (XFS_ERROR(ENOSPC));
 347                }
 348                tp->t_blk_res += blocks;
 349        }
 350
 351        /*
 352         * Reserve the log space needed for this transaction.
 353         */
 354        if (logspace > 0) {
 355                ASSERT((tp->t_log_res == 0) || (tp->t_log_res == logspace));
 356                ASSERT((tp->t_log_count == 0) ||
 357                        (tp->t_log_count == logcount));
 358                if (flags & XFS_TRANS_PERM_LOG_RES) {
 359                        log_flags = XFS_LOG_PERM_RESERV;
 360                        tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
 361                } else {
 362                        ASSERT(tp->t_ticket == NULL);
 363                        ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
 364                        log_flags = 0;
 365                }
 366
 367                error = xfs_log_reserve(tp->t_mountp, logspace, logcount,
 368                                        &tp->t_ticket,
 369                                        XFS_TRANSACTION, log_flags, tp->t_type);
 370                if (error) {
 371                        goto undo_blocks;
 372                }
 373                tp->t_log_res = logspace;
 374                tp->t_log_count = logcount;
 375        }
 376
 377        /*
 378         * Attempt to reserve the needed realtime extents by decrementing
 379         * the number needed from the number available.  This will
 380         * fail if the count would go below zero.
 381         */
 382        if (rtextents > 0) {
 383                error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FREXTENTS,
 384                                          -((int64_t)rtextents), rsvd);
 385                if (error) {
 386                        error = XFS_ERROR(ENOSPC);
 387                        goto undo_log;
 388                }
 389                tp->t_rtx_res += rtextents;
 390        }
 391
 392        return 0;
 393
 394        /*
 395         * Error cases jump to one of these labels to undo any
 396         * reservations which have already been performed.
 397         */
 398undo_log:
 399        if (logspace > 0) {
 400                if (flags & XFS_TRANS_PERM_LOG_RES) {
 401                        log_flags = XFS_LOG_REL_PERM_RESERV;
 402                } else {
 403                        log_flags = 0;
 404                }
 405                xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, log_flags);
 406                tp->t_ticket = NULL;
 407                tp->t_log_res = 0;
 408                tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
 409        }
 410
 411undo_blocks:
 412        if (blocks > 0) {
 413                (void) xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FDBLOCKS,
 414                                         (int64_t)blocks, rsvd);
 415                tp->t_blk_res = 0;
 416        }
 417
 418        current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
 419
 420        return error;
 421}
 422
 423
 424/*
 425 * Record the indicated change to the given field for application
 426 * to the file system's superblock when the transaction commits.
 427 * For now, just store the change in the transaction structure.
 428 *
 429 * Mark the transaction structure to indicate that the superblock
 430 * needs to be updated before committing.
 431 *
 432 * Because we may not be keeping track of allocated/free inodes and
 433 * used filesystem blocks in the superblock, we do not mark the
 434 * superblock dirty in this transaction if we modify these fields.
 435 * We still need to update the transaction deltas so that they get
 436 * applied to the incore superblock, but we don't want them to
 437 * cause the superblock to get locked and logged if these are the
 438 * only fields in the superblock that the transaction modifies.
 439 */
 440void
 441xfs_trans_mod_sb(
 442        xfs_trans_t     *tp,
 443        uint            field,
 444        int64_t         delta)
 445{
 446        uint32_t        flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY);
 447        xfs_mount_t     *mp = tp->t_mountp;
 448
 449        switch (field) {
 450        case XFS_TRANS_SB_ICOUNT:
 451                tp->t_icount_delta += delta;
 452                if (xfs_sb_version_haslazysbcount(&mp->m_sb))
 453                        flags &= ~XFS_TRANS_SB_DIRTY;
 454                break;
 455        case XFS_TRANS_SB_IFREE:
 456                tp->t_ifree_delta += delta;
 457                if (xfs_sb_version_haslazysbcount(&mp->m_sb))
 458                        flags &= ~XFS_TRANS_SB_DIRTY;
 459                break;
 460        case XFS_TRANS_SB_FDBLOCKS:
 461                /*
 462                 * Track the number of blocks allocated in the
 463                 * transaction.  Make sure it does not exceed the
 464                 * number reserved.
 465                 */
 466                if (delta < 0) {
 467                        tp->t_blk_res_used += (uint)-delta;
 468                        ASSERT(tp->t_blk_res_used <= tp->t_blk_res);
 469                }
 470                tp->t_fdblocks_delta += delta;
 471                if (xfs_sb_version_haslazysbcount(&mp->m_sb))
 472                        flags &= ~XFS_TRANS_SB_DIRTY;
 473                break;
 474        case XFS_TRANS_SB_RES_FDBLOCKS:
 475                /*
 476                 * The allocation has already been applied to the
 477                 * in-core superblock's counter.  This should only
 478                 * be applied to the on-disk superblock.
 479                 */
 480                ASSERT(delta < 0);
 481                tp->t_res_fdblocks_delta += delta;
 482                if (xfs_sb_version_haslazysbcount(&mp->m_sb))
 483                        flags &= ~XFS_TRANS_SB_DIRTY;
 484                break;
 485        case XFS_TRANS_SB_FREXTENTS:
 486                /*
 487                 * Track the number of blocks allocated in the
 488                 * transaction.  Make sure it does not exceed the
 489                 * number reserved.
 490                 */
 491                if (delta < 0) {
 492                        tp->t_rtx_res_used += (uint)-delta;
 493                        ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
 494                }
 495                tp->t_frextents_delta += delta;
 496                break;
 497        case XFS_TRANS_SB_RES_FREXTENTS:
 498                /*
 499                 * The allocation has already been applied to the
 500                 * in-core superblock's counter.  This should only
 501                 * be applied to the on-disk superblock.
 502                 */
 503                ASSERT(delta < 0);
 504                tp->t_res_frextents_delta += delta;
 505                break;
 506        case XFS_TRANS_SB_DBLOCKS:
 507                ASSERT(delta > 0);
 508                tp->t_dblocks_delta += delta;
 509                break;
 510        case XFS_TRANS_SB_AGCOUNT:
 511                ASSERT(delta > 0);
 512                tp->t_agcount_delta += delta;
 513                break;
 514        case XFS_TRANS_SB_IMAXPCT:
 515                tp->t_imaxpct_delta += delta;
 516                break;
 517        case XFS_TRANS_SB_REXTSIZE:
 518                tp->t_rextsize_delta += delta;
 519                break;
 520        case XFS_TRANS_SB_RBMBLOCKS:
 521                tp->t_rbmblocks_delta += delta;
 522                break;
 523        case XFS_TRANS_SB_RBLOCKS:
 524                tp->t_rblocks_delta += delta;
 525                break;
 526        case XFS_TRANS_SB_REXTENTS:
 527                tp->t_rextents_delta += delta;
 528                break;
 529        case XFS_TRANS_SB_REXTSLOG:
 530                tp->t_rextslog_delta += delta;
 531                break;
 532        default:
 533                ASSERT(0);
 534                return;
 535        }
 536
 537        tp->t_flags |= flags;
 538}
 539
 540/*
 541 * xfs_trans_apply_sb_deltas() is called from the commit code
 542 * to bring the superblock buffer into the current transaction
 543 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
 544 *
 545 * For now we just look at each field allowed to change and change
 546 * it if necessary.
 547 */
 548STATIC void
 549xfs_trans_apply_sb_deltas(
 550        xfs_trans_t     *tp)
 551{
 552        xfs_dsb_t       *sbp;
 553        xfs_buf_t       *bp;
 554        int             whole = 0;
 555
 556        bp = xfs_trans_getsb(tp, tp->t_mountp, 0);
 557        sbp = XFS_BUF_TO_SBP(bp);
 558
 559        /*
 560         * Check that superblock mods match the mods made to AGF counters.
 561         */
 562        ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) ==
 563               (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta +
 564                tp->t_ag_btree_delta));
 565
 566        /*
 567         * Only update the superblock counters if we are logging them
 568         */
 569        if (!xfs_sb_version_haslazysbcount(&(tp->t_mountp->m_sb))) {
 570                if (tp->t_icount_delta)
 571                        be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta);
 572                if (tp->t_ifree_delta)
 573                        be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta);
 574                if (tp->t_fdblocks_delta)
 575                        be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta);
 576                if (tp->t_res_fdblocks_delta)
 577                        be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta);
 578        }
 579
 580        if (tp->t_frextents_delta)
 581                be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta);
 582        if (tp->t_res_frextents_delta)
 583                be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta);
 584
 585        if (tp->t_dblocks_delta) {
 586                be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta);
 587                whole = 1;
 588        }
 589        if (tp->t_agcount_delta) {
 590                be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta);
 591                whole = 1;
 592        }
 593        if (tp->t_imaxpct_delta) {
 594                sbp->sb_imax_pct += tp->t_imaxpct_delta;
 595                whole = 1;
 596        }
 597        if (tp->t_rextsize_delta) {
 598                be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta);
 599                whole = 1;
 600        }
 601        if (tp->t_rbmblocks_delta) {
 602                be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta);
 603                whole = 1;
 604        }
 605        if (tp->t_rblocks_delta) {
 606                be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta);
 607                whole = 1;
 608        }
 609        if (tp->t_rextents_delta) {
 610                be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta);
 611                whole = 1;
 612        }
 613        if (tp->t_rextslog_delta) {
 614                sbp->sb_rextslog += tp->t_rextslog_delta;
 615                whole = 1;
 616        }
 617
 618        if (whole)
 619                /*
 620                 * Log the whole thing, the fields are noncontiguous.
 621                 */
 622                xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_dsb_t) - 1);
 623        else
 624                /*
 625                 * Since all the modifiable fields are contiguous, we
 626                 * can get away with this.
 627                 */
 628                xfs_trans_log_buf(tp, bp, offsetof(xfs_dsb_t, sb_icount),
 629                                  offsetof(xfs_dsb_t, sb_frextents) +
 630                                  sizeof(sbp->sb_frextents) - 1);
 631}
 632
 633/*
 634 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations
 635 * and apply superblock counter changes to the in-core superblock.  The
 636 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
 637 * applied to the in-core superblock.  The idea is that that has already been
 638 * done.
 639 *
 640 * This is done efficiently with a single call to xfs_mod_incore_sb_batch().
 641 * However, we have to ensure that we only modify each superblock field only
 642 * once because the application of the delta values may not be atomic. That can
 643 * lead to ENOSPC races occurring if we have two separate modifcations of the
 644 * free space counter to put back the entire reservation and then take away
 645 * what we used.
 646 *
 647 * If we are not logging superblock counters, then the inode allocated/free and
 648 * used block counts are not updated in the on disk superblock. In this case,
 649 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
 650 * still need to update the incore superblock with the changes.
 651 */
 652STATIC void
 653xfs_trans_unreserve_and_mod_sb(
 654        xfs_trans_t     *tp)
 655{
 656        xfs_mod_sb_t    msb[14];        /* If you add cases, add entries */
 657        xfs_mod_sb_t    *msbp;
 658        xfs_mount_t     *mp = tp->t_mountp;
 659        /* REFERENCED */
 660        int             error;
 661        int             rsvd;
 662        int64_t         blkdelta = 0;
 663        int64_t         rtxdelta = 0;
 664
 665        msbp = msb;
 666        rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
 667
 668        /* calculate free blocks delta */
 669        if (tp->t_blk_res > 0)
 670                blkdelta = tp->t_blk_res;
 671
 672        if ((tp->t_fdblocks_delta != 0) &&
 673            (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
 674             (tp->t_flags & XFS_TRANS_SB_DIRTY)))
 675                blkdelta += tp->t_fdblocks_delta;
 676
 677        if (blkdelta != 0) {
 678                msbp->msb_field = XFS_SBS_FDBLOCKS;
 679                msbp->msb_delta = blkdelta;
 680                msbp++;
 681        }
 682
 683        /* calculate free realtime extents delta */
 684        if (tp->t_rtx_res > 0)
 685                rtxdelta = tp->t_rtx_res;
 686
 687        if ((tp->t_frextents_delta != 0) &&
 688            (tp->t_flags & XFS_TRANS_SB_DIRTY))
 689                rtxdelta += tp->t_frextents_delta;
 690
 691        if (rtxdelta != 0) {
 692                msbp->msb_field = XFS_SBS_FREXTENTS;
 693                msbp->msb_delta = rtxdelta;
 694                msbp++;
 695        }
 696
 697        /* apply remaining deltas */
 698
 699        if (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
 700             (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
 701                if (tp->t_icount_delta != 0) {
 702                        msbp->msb_field = XFS_SBS_ICOUNT;
 703                        msbp->msb_delta = tp->t_icount_delta;
 704                        msbp++;
 705                }
 706                if (tp->t_ifree_delta != 0) {
 707                        msbp->msb_field = XFS_SBS_IFREE;
 708                        msbp->msb_delta = tp->t_ifree_delta;
 709                        msbp++;
 710                }
 711        }
 712
 713        if (tp->t_flags & XFS_TRANS_SB_DIRTY) {
 714                if (tp->t_dblocks_delta != 0) {
 715                        msbp->msb_field = XFS_SBS_DBLOCKS;
 716                        msbp->msb_delta = tp->t_dblocks_delta;
 717                        msbp++;
 718                }
 719                if (tp->t_agcount_delta != 0) {
 720                        msbp->msb_field = XFS_SBS_AGCOUNT;
 721                        msbp->msb_delta = tp->t_agcount_delta;
 722                        msbp++;
 723                }
 724                if (tp->t_imaxpct_delta != 0) {
 725                        msbp->msb_field = XFS_SBS_IMAX_PCT;
 726                        msbp->msb_delta = tp->t_imaxpct_delta;
 727                        msbp++;
 728                }
 729                if (tp->t_rextsize_delta != 0) {
 730                        msbp->msb_field = XFS_SBS_REXTSIZE;
 731                        msbp->msb_delta = tp->t_rextsize_delta;
 732                        msbp++;
 733                }
 734                if (tp->t_rbmblocks_delta != 0) {
 735                        msbp->msb_field = XFS_SBS_RBMBLOCKS;
 736                        msbp->msb_delta = tp->t_rbmblocks_delta;
 737                        msbp++;
 738                }
 739                if (tp->t_rblocks_delta != 0) {
 740                        msbp->msb_field = XFS_SBS_RBLOCKS;
 741                        msbp->msb_delta = tp->t_rblocks_delta;
 742                        msbp++;
 743                }
 744                if (tp->t_rextents_delta != 0) {
 745                        msbp->msb_field = XFS_SBS_REXTENTS;
 746                        msbp->msb_delta = tp->t_rextents_delta;
 747                        msbp++;
 748                }
 749                if (tp->t_rextslog_delta != 0) {
 750                        msbp->msb_field = XFS_SBS_REXTSLOG;
 751                        msbp->msb_delta = tp->t_rextslog_delta;
 752                        msbp++;
 753                }
 754        }
 755
 756        /*
 757         * If we need to change anything, do it.
 758         */
 759        if (msbp > msb) {
 760                error = xfs_mod_incore_sb_batch(tp->t_mountp, msb,
 761                        (uint)(msbp - msb), rsvd);
 762                ASSERT(error == 0);
 763        }
 764}
 765
 766
 767/*
 768 * xfs_trans_commit
 769 *
 770 * Commit the given transaction to the log a/synchronously.
 771 *
 772 * XFS disk error handling mechanism is not based on a typical
 773 * transaction abort mechanism. Logically after the filesystem
 774 * gets marked 'SHUTDOWN', we can't let any new transactions
 775 * be durable - ie. committed to disk - because some metadata might
 776 * be inconsistent. In such cases, this returns an error, and the
 777 * caller may assume that all locked objects joined to the transaction
 778 * have already been unlocked as if the commit had succeeded.
 779 * Do not reference the transaction structure after this call.
 780 */
 781 /*ARGSUSED*/
 782int
 783_xfs_trans_commit(
 784        xfs_trans_t     *tp,
 785        uint            flags,
 786        int             *log_flushed)
 787{
 788        xfs_log_iovec_t         *log_vector;
 789        int                     nvec;
 790        xfs_mount_t             *mp;
 791        xfs_lsn_t               commit_lsn;
 792        /* REFERENCED */
 793        int                     error;
 794        int                     log_flags;
 795        int                     sync;
 796#define XFS_TRANS_LOGVEC_COUNT  16
 797        xfs_log_iovec_t         log_vector_fast[XFS_TRANS_LOGVEC_COUNT];
 798        void                    *commit_iclog;
 799        int                     shutdown;
 800
 801        commit_lsn = -1;
 802
 803        /*
 804         * Determine whether this commit is releasing a permanent
 805         * log reservation or not.
 806         */
 807        if (flags & XFS_TRANS_RELEASE_LOG_RES) {
 808                ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
 809                log_flags = XFS_LOG_REL_PERM_RESERV;
 810        } else {
 811                log_flags = 0;
 812        }
 813        mp = tp->t_mountp;
 814
 815        /*
 816         * If there is nothing to be logged by the transaction,
 817         * then unlock all of the items associated with the
 818         * transaction and free the transaction structure.
 819         * Also make sure to return any reserved blocks to
 820         * the free pool.
 821         */
 822shut_us_down:
 823        shutdown = XFS_FORCED_SHUTDOWN(mp) ? EIO : 0;
 824        if (!(tp->t_flags & XFS_TRANS_DIRTY) || shutdown) {
 825                xfs_trans_unreserve_and_mod_sb(tp);
 826                /*
 827                 * It is indeed possible for the transaction to be
 828                 * not dirty but the dqinfo portion to be. All that
 829                 * means is that we have some (non-persistent) quota
 830                 * reservations that need to be unreserved.
 831                 */
 832                xfs_trans_unreserve_and_mod_dquots(tp);
 833                if (tp->t_ticket) {
 834                        commit_lsn = xfs_log_done(mp, tp->t_ticket,
 835                                                        NULL, log_flags);
 836                        if (commit_lsn == -1 && !shutdown)
 837                                shutdown = XFS_ERROR(EIO);
 838                }
 839                current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
 840                xfs_trans_free_items(tp, shutdown? XFS_TRANS_ABORT : 0);
 841                xfs_trans_free_busy(tp);
 842                xfs_trans_free(tp);
 843                XFS_STATS_INC(xs_trans_empty);
 844                return (shutdown);
 845        }
 846        ASSERT(tp->t_ticket != NULL);
 847
 848        /*
 849         * If we need to update the superblock, then do it now.
 850         */
 851        if (tp->t_flags & XFS_TRANS_SB_DIRTY)
 852                xfs_trans_apply_sb_deltas(tp);
 853        xfs_trans_apply_dquot_deltas(tp);
 854
 855        /*
 856         * Ask each log item how many log_vector entries it will
 857         * need so we can figure out how many to allocate.
 858         * Try to avoid the kmem_alloc() call in the common case
 859         * by using a vector from the stack when it fits.
 860         */
 861        nvec = xfs_trans_count_vecs(tp);
 862        if (nvec == 0) {
 863                xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
 864                goto shut_us_down;
 865        } else if (nvec <= XFS_TRANS_LOGVEC_COUNT) {
 866                log_vector = log_vector_fast;
 867        } else {
 868                log_vector = (xfs_log_iovec_t *)kmem_alloc(nvec *
 869                                                   sizeof(xfs_log_iovec_t),
 870                                                   KM_SLEEP);
 871        }
 872
 873        /*
 874         * Fill in the log_vector and pin the logged items, and
 875         * then write the transaction to the log.
 876         */
 877        xfs_trans_fill_vecs(tp, log_vector);
 878
 879        error = xfs_log_write(mp, log_vector, nvec, tp->t_ticket, &(tp->t_lsn));
 880
 881        /*
 882         * The transaction is committed incore here, and can go out to disk
 883         * at any time after this call.  However, all the items associated
 884         * with the transaction are still locked and pinned in memory.
 885         */
 886        commit_lsn = xfs_log_done(mp, tp->t_ticket, &commit_iclog, log_flags);
 887
 888        tp->t_commit_lsn = commit_lsn;
 889        if (nvec > XFS_TRANS_LOGVEC_COUNT) {
 890                kmem_free(log_vector);
 891        }
 892
 893        /*
 894         * If we got a log write error. Unpin the logitems that we
 895         * had pinned, clean up, free trans structure, and return error.
 896         */
 897        if (error || commit_lsn == -1) {
 898                current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
 899                xfs_trans_uncommit(tp, flags|XFS_TRANS_ABORT);
 900                return XFS_ERROR(EIO);
 901        }
 902
 903        /*
 904         * Once the transaction has committed, unused
 905         * reservations need to be released and changes to
 906         * the superblock need to be reflected in the in-core
 907         * version.  Do that now.
 908         */
 909        xfs_trans_unreserve_and_mod_sb(tp);
 910
 911        sync = tp->t_flags & XFS_TRANS_SYNC;
 912
 913        /*
 914         * Tell the LM to call the transaction completion routine
 915         * when the log write with LSN commit_lsn completes (e.g.
 916         * when the transaction commit really hits the on-disk log).
 917         * After this call we cannot reference tp, because the call
 918         * can happen at any time and the call will free the transaction
 919         * structure pointed to by tp.  The only case where we call
 920         * the completion routine (xfs_trans_committed) directly is
 921         * if the log is turned off on a debug kernel or we're
 922         * running in simulation mode (the log is explicitly turned
 923         * off).
 924         */
 925        tp->t_logcb.cb_func = (void(*)(void*, int))xfs_trans_committed;
 926        tp->t_logcb.cb_arg = tp;
 927
 928        /*
 929         * We need to pass the iclog buffer which was used for the
 930         * transaction commit record into this function, and attach
 931         * the callback to it. The callback must be attached before
 932         * the items are unlocked to avoid racing with other threads
 933         * waiting for an item to unlock.
 934         */
 935        shutdown = xfs_log_notify(mp, commit_iclog, &(tp->t_logcb));
 936
 937        /*
 938         * Mark this thread as no longer being in a transaction
 939         */
 940        current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
 941
 942        /*
 943         * Once all the items of the transaction have been copied
 944         * to the in core log and the callback is attached, the
 945         * items can be unlocked.
 946         *
 947         * This will free descriptors pointing to items which were
 948         * not logged since there is nothing more to do with them.
 949         * For items which were logged, we will keep pointers to them
 950         * so they can be unpinned after the transaction commits to disk.
 951         * This will also stamp each modified meta-data item with
 952         * the commit lsn of this transaction for dependency tracking
 953         * purposes.
 954         */
 955        xfs_trans_unlock_items(tp, commit_lsn);
 956
 957        /*
 958         * If we detected a log error earlier, finish committing
 959         * the transaction now (unpin log items, etc).
 960         *
 961         * Order is critical here, to avoid using the transaction
 962         * pointer after its been freed (by xfs_trans_committed
 963         * either here now, or as a callback).  We cannot do this
 964         * step inside xfs_log_notify as was done earlier because
 965         * of this issue.
 966         */
 967        if (shutdown)
 968                xfs_trans_committed(tp, XFS_LI_ABORTED);
 969
 970        /*
 971         * Now that the xfs_trans_committed callback has been attached,
 972         * and the items are released we can finally allow the iclog to
 973         * go to disk.
 974         */
 975        error = xfs_log_release_iclog(mp, commit_iclog);
 976
 977        /*
 978         * If the transaction needs to be synchronous, then force the
 979         * log out now and wait for it.
 980         */
 981        if (sync) {
 982                if (!error) {
 983                        error = _xfs_log_force(mp, commit_lsn,
 984                                      XFS_LOG_FORCE | XFS_LOG_SYNC,
 985                                      log_flushed);
 986                }
 987                XFS_STATS_INC(xs_trans_sync);
 988        } else {
 989                XFS_STATS_INC(xs_trans_async);
 990        }
 991
 992        return (error);
 993}
 994
 995
 996/*
 997 * Total up the number of log iovecs needed to commit this
 998 * transaction.  The transaction itself needs one for the
 999 * transaction header.  Ask each dirty item in turn how many
1000 * it needs to get the total.
1001 */
1002STATIC uint
1003xfs_trans_count_vecs(
1004        xfs_trans_t     *tp)
1005{
1006        int                     nvecs;
1007        xfs_log_item_desc_t     *lidp;
1008
1009        nvecs = 1;
1010        lidp = xfs_trans_first_item(tp);
1011        ASSERT(lidp != NULL);
1012
1013        /* In the non-debug case we need to start bailing out if we
1014         * didn't find a log_item here, return zero and let trans_commit
1015         * deal with it.
1016         */
1017        if (lidp == NULL)
1018                return 0;
1019
1020        while (lidp != NULL) {
1021                /*
1022                 * Skip items which aren't dirty in this transaction.
1023                 */
1024                if (!(lidp->lid_flags & XFS_LID_DIRTY)) {
1025                        lidp = xfs_trans_next_item(tp, lidp);
1026                        continue;
1027                }
1028                lidp->lid_size = IOP_SIZE(lidp->lid_item);
1029                nvecs += lidp->lid_size;
1030                lidp = xfs_trans_next_item(tp, lidp);
1031        }
1032
1033        return nvecs;
1034}
1035
1036/*
1037 * Called from the trans_commit code when we notice that
1038 * the filesystem is in the middle of a forced shutdown.
1039 */
1040STATIC void
1041xfs_trans_uncommit(
1042        xfs_trans_t     *tp,
1043        uint            flags)
1044{
1045        xfs_log_item_desc_t     *lidp;
1046
1047        for (lidp = xfs_trans_first_item(tp);
1048             lidp != NULL;
1049             lidp = xfs_trans_next_item(tp, lidp)) {
1050                /*
1051                 * Unpin all but those that aren't dirty.
1052                 */
1053                if (lidp->lid_flags & XFS_LID_DIRTY)
1054                        IOP_UNPIN_REMOVE(lidp->lid_item, tp);
1055        }
1056
1057        xfs_trans_unreserve_and_mod_sb(tp);
1058        xfs_trans_unreserve_and_mod_dquots(tp);
1059
1060        xfs_trans_free_items(tp, flags);
1061        xfs_trans_free_busy(tp);
1062        xfs_trans_free(tp);
1063}
1064
1065/*
1066 * Fill in the vector with pointers to data to be logged
1067 * by this transaction.  The transaction header takes
1068 * the first vector, and then each dirty item takes the
1069 * number of vectors it indicated it needed in xfs_trans_count_vecs().
1070 *
1071 * As each item fills in the entries it needs, also pin the item
1072 * so that it cannot be flushed out until the log write completes.
1073 */
1074STATIC void
1075xfs_trans_fill_vecs(
1076        xfs_trans_t             *tp,
1077        xfs_log_iovec_t         *log_vector)
1078{
1079        xfs_log_item_desc_t     *lidp;
1080        xfs_log_iovec_t         *vecp;
1081        uint                    nitems;
1082
1083        /*
1084         * Skip over the entry for the transaction header, we'll
1085         * fill that in at the end.
1086         */
1087        vecp = log_vector + 1;          /* pointer arithmetic */
1088
1089        nitems = 0;
1090        lidp = xfs_trans_first_item(tp);
1091        ASSERT(lidp != NULL);
1092        while (lidp != NULL) {
1093                /*
1094                 * Skip items which aren't dirty in this transaction.
1095                 */
1096                if (!(lidp->lid_flags & XFS_LID_DIRTY)) {
1097                        lidp = xfs_trans_next_item(tp, lidp);
1098                        continue;
1099                }
1100                /*
1101                 * The item may be marked dirty but not log anything.
1102                 * This can be used to get called when a transaction
1103                 * is committed.
1104                 */
1105                if (lidp->lid_size) {
1106                        nitems++;
1107                }
1108                IOP_FORMAT(lidp->lid_item, vecp);
1109                vecp += lidp->lid_size;         /* pointer arithmetic */
1110                IOP_PIN(lidp->lid_item);
1111                lidp = xfs_trans_next_item(tp, lidp);
1112        }
1113
1114        /*
1115         * Now that we've counted the number of items in this
1116         * transaction, fill in the transaction header.
1117         */
1118        tp->t_header.th_magic = XFS_TRANS_HEADER_MAGIC;
1119        tp->t_header.th_type = tp->t_type;
1120        tp->t_header.th_num_items = nitems;
1121        log_vector->i_addr = (xfs_caddr_t)&tp->t_header;
1122        log_vector->i_len = sizeof(xfs_trans_header_t);
1123        XLOG_VEC_SET_TYPE(log_vector, XLOG_REG_TYPE_TRANSHDR);
1124}
1125
1126
1127/*
1128 * Unlock all of the transaction's items and free the transaction.
1129 * The transaction must not have modified any of its items, because
1130 * there is no way to restore them to their previous state.
1131 *
1132 * If the transaction has made a log reservation, make sure to release
1133 * it as well.
1134 */
1135void
1136xfs_trans_cancel(
1137        xfs_trans_t             *tp,
1138        int                     flags)
1139{
1140        int                     log_flags;
1141#ifdef DEBUG
1142        xfs_log_item_chunk_t    *licp;
1143        xfs_log_item_desc_t     *lidp;
1144        xfs_log_item_t          *lip;
1145        int                     i;
1146#endif
1147        xfs_mount_t             *mp = tp->t_mountp;
1148
1149        /*
1150         * See if the caller is being too lazy to figure out if
1151         * the transaction really needs an abort.
1152         */
1153        if ((flags & XFS_TRANS_ABORT) && !(tp->t_flags & XFS_TRANS_DIRTY))
1154                flags &= ~XFS_TRANS_ABORT;
1155        /*
1156         * See if the caller is relying on us to shut down the
1157         * filesystem.  This happens in paths where we detect
1158         * corruption and decide to give up.
1159         */
1160        if ((tp->t_flags & XFS_TRANS_DIRTY) && !XFS_FORCED_SHUTDOWN(mp)) {
1161                XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
1162                xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1163        }
1164#ifdef DEBUG
1165        if (!(flags & XFS_TRANS_ABORT)) {
1166                licp = &(tp->t_items);
1167                while (licp != NULL) {
1168                        lidp = licp->lic_descs;
1169                        for (i = 0; i < licp->lic_unused; i++, lidp++) {
1170                                if (xfs_lic_isfree(licp, i)) {
1171                                        continue;
1172                                }
1173
1174                                lip = lidp->lid_item;
1175                                if (!XFS_FORCED_SHUTDOWN(mp))
1176                                        ASSERT(!(lip->li_type == XFS_LI_EFD));
1177                        }
1178                        licp = licp->lic_next;
1179                }
1180        }
1181#endif
1182        xfs_trans_unreserve_and_mod_sb(tp);
1183        xfs_trans_unreserve_and_mod_dquots(tp);
1184
1185        if (tp->t_ticket) {
1186                if (flags & XFS_TRANS_RELEASE_LOG_RES) {
1187                        ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1188                        log_flags = XFS_LOG_REL_PERM_RESERV;
1189                } else {
1190                        log_flags = 0;
1191                }
1192                xfs_log_done(mp, tp->t_ticket, NULL, log_flags);
1193        }
1194
1195        /* mark this thread as no longer being in a transaction */
1196        current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
1197
1198        xfs_trans_free_items(tp, flags);
1199        xfs_trans_free_busy(tp);
1200        xfs_trans_free(tp);
1201}
1202
1203
1204/*
1205 * Free the transaction structure.  If there is more clean up
1206 * to do when the structure is freed, add it here.
1207 */
1208STATIC void
1209xfs_trans_free(
1210        xfs_trans_t     *tp)
1211{
1212        atomic_dec(&tp->t_mountp->m_active_trans);
1213        xfs_trans_free_dqinfo(tp);
1214        kmem_zone_free(xfs_trans_zone, tp);
1215}
1216
1217/*
1218 * Roll from one trans in the sequence of PERMANENT transactions to
1219 * the next: permanent transactions are only flushed out when
1220 * committed with XFS_TRANS_RELEASE_LOG_RES, but we still want as soon
1221 * as possible to let chunks of it go to the log. So we commit the
1222 * chunk we've been working on and get a new transaction to continue.
1223 */
1224int
1225xfs_trans_roll(
1226        struct xfs_trans        **tpp,
1227        struct xfs_inode        *dp)
1228{
1229        struct xfs_trans        *trans;
1230        unsigned int            logres, count;
1231        int                     error;
1232
1233        /*
1234         * Ensure that the inode is always logged.
1235         */
1236        trans = *tpp;
1237        xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
1238
1239        /*
1240         * Copy the critical parameters from one trans to the next.
1241         */
1242        logres = trans->t_log_res;
1243        count = trans->t_log_count;
1244        *tpp = xfs_trans_dup(trans);
1245
1246        /*
1247         * Commit the current transaction.
1248         * If this commit failed, then it'd just unlock those items that
1249         * are not marked ihold. That also means that a filesystem shutdown
1250         * is in progress. The caller takes the responsibility to cancel
1251         * the duplicate transaction that gets returned.
1252         */
1253        error = xfs_trans_commit(trans, 0);
1254        if (error)
1255                return (error);
1256
1257        trans = *tpp;
1258
1259        /*
1260         * transaction commit worked ok so we can drop the extra ticket
1261         * reference that we gained in xfs_trans_dup()
1262         */
1263        xfs_log_ticket_put(trans->t_ticket);
1264
1265
1266        /*
1267         * Reserve space in the log for th next transaction.
1268         * This also pushes items in the "AIL", the list of logged items,
1269         * out to disk if they are taking up space at the tail of the log
1270         * that we want to use.  This requires that either nothing be locked
1271         * across this call, or that anything that is locked be logged in
1272         * the prior and the next transactions.
1273         */
1274        error = xfs_trans_reserve(trans, 0, logres, 0,
1275                                  XFS_TRANS_PERM_LOG_RES, count);
1276        /*
1277         *  Ensure that the inode is in the new transaction and locked.
1278         */
1279        if (error)
1280                return error;
1281
1282        xfs_trans_ijoin(trans, dp, XFS_ILOCK_EXCL);
1283        xfs_trans_ihold(trans, dp);
1284        return 0;
1285}
1286
1287/*
1288 * THIS SHOULD BE REWRITTEN TO USE xfs_trans_next_item().
1289 *
1290 * This is typically called by the LM when a transaction has been fully
1291 * committed to disk.  It needs to unpin the items which have
1292 * been logged by the transaction and update their positions
1293 * in the AIL if necessary.
1294 * This also gets called when the transactions didn't get written out
1295 * because of an I/O error. Abortflag & XFS_LI_ABORTED is set then.
1296 *
1297 * Call xfs_trans_chunk_committed() to process the items in
1298 * each chunk.
1299 */
1300STATIC void
1301xfs_trans_committed(
1302        xfs_trans_t     *tp,
1303        int             abortflag)
1304{
1305        xfs_log_item_chunk_t    *licp;
1306        xfs_log_item_chunk_t    *next_licp;
1307        xfs_log_busy_chunk_t    *lbcp;
1308        xfs_log_busy_slot_t     *lbsp;
1309        int                     i;
1310
1311        /*
1312         * Call the transaction's completion callback if there
1313         * is one.
1314         */
1315        if (tp->t_callback != NULL) {
1316                tp->t_callback(tp, tp->t_callarg);
1317        }
1318
1319        /*
1320         * Special case the chunk embedded in the transaction.
1321         */
1322        licp = &(tp->t_items);
1323        if (!(xfs_lic_are_all_free(licp))) {
1324                xfs_trans_chunk_committed(licp, tp->t_lsn, abortflag);
1325        }
1326
1327        /*
1328         * Process the items in each chunk in turn.
1329         */
1330        licp = licp->lic_next;
1331        while (licp != NULL) {
1332                ASSERT(!xfs_lic_are_all_free(licp));
1333                xfs_trans_chunk_committed(licp, tp->t_lsn, abortflag);
1334                next_licp = licp->lic_next;
1335                kmem_free(licp);
1336                licp = next_licp;
1337        }
1338
1339        /*
1340         * Clear all the per-AG busy list items listed in this transaction
1341         */
1342        lbcp = &tp->t_busy;
1343        while (lbcp != NULL) {
1344                for (i = 0, lbsp = lbcp->lbc_busy; i < lbcp->lbc_unused; i++, lbsp++) {
1345                        if (!XFS_LBC_ISFREE(lbcp, i)) {
1346                                xfs_alloc_clear_busy(tp, lbsp->lbc_ag,
1347                                                     lbsp->lbc_idx);
1348                        }
1349                }
1350                lbcp = lbcp->lbc_next;
1351        }
1352        xfs_trans_free_busy(tp);
1353
1354        /*
1355         * That's it for the transaction structure.  Free it.
1356         */
1357        xfs_trans_free(tp);
1358}
1359
1360/*
1361 * This is called to perform the commit processing for each
1362 * item described by the given chunk.
1363 *
1364 * The commit processing consists of unlocking items which were
1365 * held locked with the SYNC_UNLOCK attribute, calling the committed
1366 * routine of each logged item, updating the item's position in the AIL
1367 * if necessary, and unpinning each item.  If the committed routine
1368 * returns -1, then do nothing further with the item because it
1369 * may have been freed.
1370 *
1371 * Since items are unlocked when they are copied to the incore
1372 * log, it is possible for two transactions to be completing
1373 * and manipulating the same item simultaneously.  The AIL lock
1374 * will protect the lsn field of each item.  The value of this
1375 * field can never go backwards.
1376 *
1377 * We unpin the items after repositioning them in the AIL, because
1378 * otherwise they could be immediately flushed and we'd have to race
1379 * with the flusher trying to pull the item from the AIL as we add it.
1380 */
1381STATIC void
1382xfs_trans_chunk_committed(
1383        xfs_log_item_chunk_t    *licp,
1384        xfs_lsn_t               lsn,
1385        int                     aborted)
1386{
1387        xfs_log_item_desc_t     *lidp;
1388        xfs_log_item_t          *lip;
1389        xfs_lsn_t               item_lsn;
1390        int                     i;
1391
1392        lidp = licp->lic_descs;
1393        for (i = 0; i < licp->lic_unused; i++, lidp++) {
1394                struct xfs_ail          *ailp;
1395
1396                if (xfs_lic_isfree(licp, i)) {
1397                        continue;
1398                }
1399
1400                lip = lidp->lid_item;
1401                if (aborted)
1402                        lip->li_flags |= XFS_LI_ABORTED;
1403
1404                /*
1405                 * Send in the ABORTED flag to the COMMITTED routine
1406                 * so that it knows whether the transaction was aborted
1407                 * or not.
1408                 */
1409                item_lsn = IOP_COMMITTED(lip, lsn);
1410
1411                /*
1412                 * If the committed routine returns -1, make
1413                 * no more references to the item.
1414                 */
1415                if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0) {
1416                        continue;
1417                }
1418
1419                /*
1420                 * If the returned lsn is greater than what it
1421                 * contained before, update the location of the
1422                 * item in the AIL.  If it is not, then do nothing.
1423                 * Items can never move backwards in the AIL.
1424                 *
1425                 * While the new lsn should usually be greater, it
1426                 * is possible that a later transaction completing
1427                 * simultaneously with an earlier one using the
1428                 * same item could complete first with a higher lsn.
1429                 * This would cause the earlier transaction to fail
1430                 * the test below.
1431                 */
1432                ailp = lip->li_ailp;
1433                spin_lock(&ailp->xa_lock);
1434                if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0) {
1435                        /*
1436                         * This will set the item's lsn to item_lsn
1437                         * and update the position of the item in
1438                         * the AIL.
1439                         *
1440                         * xfs_trans_ail_update() drops the AIL lock.
1441                         */
1442                        xfs_trans_ail_update(ailp, lip, item_lsn);
1443                } else {
1444                        spin_unlock(&ailp->xa_lock);
1445                }
1446
1447                /*
1448                 * Now that we've repositioned the item in the AIL,
1449                 * unpin it so it can be flushed. Pass information
1450                 * about buffer stale state down from the log item
1451                 * flags, if anyone else stales the buffer we do not
1452                 * want to pay any attention to it.
1453                 */
1454                IOP_UNPIN(lip, lidp->lid_flags & XFS_LID_BUF_STALE);
1455        }
1456}
1457