linux/fs/xfs/xfs_log_cil.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (c) 2010 Red Hat, Inc. All Rights Reserved.
   4 */
   5
   6#include "xfs.h"
   7#include "xfs_fs.h"
   8#include "xfs_format.h"
   9#include "xfs_log_format.h"
  10#include "xfs_shared.h"
  11#include "xfs_trans_resv.h"
  12#include "xfs_mount.h"
  13#include "xfs_extent_busy.h"
  14#include "xfs_trans.h"
  15#include "xfs_trans_priv.h"
  16#include "xfs_log.h"
  17#include "xfs_log_priv.h"
  18#include "xfs_trace.h"
  19
  20struct workqueue_struct *xfs_discard_wq;
  21
  22/*
  23 * Allocate a new ticket. Failing to get a new ticket makes it really hard to
  24 * recover, so we don't allow failure here. Also, we allocate in a context that
  25 * we don't want to be issuing transactions from, so we need to tell the
  26 * allocation code this as well.
  27 *
  28 * We don't reserve any space for the ticket - we are going to steal whatever
  29 * space we require from transactions as they commit. To ensure we reserve all
  30 * the space required, we need to set the current reservation of the ticket to
  31 * zero so that we know to steal the initial transaction overhead from the
  32 * first transaction commit.
  33 */
  34static struct xlog_ticket *
  35xlog_cil_ticket_alloc(
  36        struct xlog     *log)
  37{
  38        struct xlog_ticket *tic;
  39
  40        tic = xlog_ticket_alloc(log, 0, 1, XFS_TRANSACTION, 0);
  41
  42        /*
  43         * set the current reservation to zero so we know to steal the basic
  44         * transaction overhead reservation from the first transaction commit.
  45         */
  46        tic->t_curr_res = 0;
  47        return tic;
  48}
  49
  50/*
  51 * After the first stage of log recovery is done, we know where the head and
  52 * tail of the log are. We need this log initialisation done before we can
  53 * initialise the first CIL checkpoint context.
  54 *
  55 * Here we allocate a log ticket to track space usage during a CIL push.  This
  56 * ticket is passed to xlog_write() directly so that we don't slowly leak log
  57 * space by failing to account for space used by log headers and additional
  58 * region headers for split regions.
  59 */
  60void
  61xlog_cil_init_post_recovery(
  62        struct xlog     *log)
  63{
  64        log->l_cilp->xc_ctx->ticket = xlog_cil_ticket_alloc(log);
  65        log->l_cilp->xc_ctx->sequence = 1;
  66}
  67
  68static inline int
  69xlog_cil_iovec_space(
  70        uint    niovecs)
  71{
  72        return round_up((sizeof(struct xfs_log_vec) +
  73                                        niovecs * sizeof(struct xfs_log_iovec)),
  74                        sizeof(uint64_t));
  75}
  76
  77/*
  78 * Allocate or pin log vector buffers for CIL insertion.
  79 *
  80 * The CIL currently uses disposable buffers for copying a snapshot of the
  81 * modified items into the log during a push. The biggest problem with this is
  82 * the requirement to allocate the disposable buffer during the commit if:
  83 *      a) does not exist; or
  84 *      b) it is too small
  85 *
  86 * If we do this allocation within xlog_cil_insert_format_items(), it is done
  87 * under the xc_ctx_lock, which means that a CIL push cannot occur during
  88 * the memory allocation. This means that we have a potential deadlock situation
  89 * under low memory conditions when we have lots of dirty metadata pinned in
  90 * the CIL and we need a CIL commit to occur to free memory.
  91 *
  92 * To avoid this, we need to move the memory allocation outside the
  93 * xc_ctx_lock, but because the log vector buffers are disposable, that opens
  94 * up a TOCTOU race condition w.r.t. the CIL committing and removing the log
  95 * vector buffers between the check and the formatting of the item into the
  96 * log vector buffer within the xc_ctx_lock.
  97 *
  98 * Because the log vector buffer needs to be unchanged during the CIL push
  99 * process, we cannot share the buffer between the transaction commit (which
 100 * modifies the buffer) and the CIL push context that is writing the changes
 101 * into the log. This means skipping preallocation of buffer space is
 102 * unreliable, but we most definitely do not want to be allocating and freeing
 103 * buffers unnecessarily during commits when overwrites can be done safely.
 104 *
 105 * The simplest solution to this problem is to allocate a shadow buffer when a
 106 * log item is committed for the second time, and then to only use this buffer
 107 * if necessary. The buffer can remain attached to the log item until such time
 108 * it is needed, and this is the buffer that is reallocated to match the size of
 109 * the incoming modification. Then during the formatting of the item we can swap
 110 * the active buffer with the new one if we can't reuse the existing buffer. We
 111 * don't free the old buffer as it may be reused on the next modification if
 112 * it's size is right, otherwise we'll free and reallocate it at that point.
 113 *
 114 * This function builds a vector for the changes in each log item in the
 115 * transaction. It then works out the length of the buffer needed for each log
 116 * item, allocates them and attaches the vector to the log item in preparation
 117 * for the formatting step which occurs under the xc_ctx_lock.
 118 *
 119 * While this means the memory footprint goes up, it avoids the repeated
 120 * alloc/free pattern that repeated modifications of an item would otherwise
 121 * cause, and hence minimises the CPU overhead of such behaviour.
 122 */
 123static void
 124xlog_cil_alloc_shadow_bufs(
 125        struct xlog             *log,
 126        struct xfs_trans        *tp)
 127{
 128        struct xfs_log_item     *lip;
 129
 130        list_for_each_entry(lip, &tp->t_items, li_trans) {
 131                struct xfs_log_vec *lv;
 132                int     niovecs = 0;
 133                int     nbytes = 0;
 134                int     buf_size;
 135                bool    ordered = false;
 136
 137                /* Skip items which aren't dirty in this transaction. */
 138                if (!test_bit(XFS_LI_DIRTY, &lip->li_flags))
 139                        continue;
 140
 141                /* get number of vecs and size of data to be stored */
 142                lip->li_ops->iop_size(lip, &niovecs, &nbytes);
 143
 144                /*
 145                 * Ordered items need to be tracked but we do not wish to write
 146                 * them. We need a logvec to track the object, but we do not
 147                 * need an iovec or buffer to be allocated for copying data.
 148                 */
 149                if (niovecs == XFS_LOG_VEC_ORDERED) {
 150                        ordered = true;
 151                        niovecs = 0;
 152                        nbytes = 0;
 153                }
 154
 155                /*
 156                 * We 64-bit align the length of each iovec so that the start
 157                 * of the next one is naturally aligned.  We'll need to
 158                 * account for that slack space here. Then round nbytes up
 159                 * to 64-bit alignment so that the initial buffer alignment is
 160                 * easy to calculate and verify.
 161                 */
 162                nbytes += niovecs * sizeof(uint64_t);
 163                nbytes = round_up(nbytes, sizeof(uint64_t));
 164
 165                /*
 166                 * The data buffer needs to start 64-bit aligned, so round up
 167                 * that space to ensure we can align it appropriately and not
 168                 * overrun the buffer.
 169                 */
 170                buf_size = nbytes + xlog_cil_iovec_space(niovecs);
 171
 172                /*
 173                 * if we have no shadow buffer, or it is too small, we need to
 174                 * reallocate it.
 175                 */
 176                if (!lip->li_lv_shadow ||
 177                    buf_size > lip->li_lv_shadow->lv_size) {
 178
 179                        /*
 180                         * We free and allocate here as a realloc would copy
 181                         * unnecessary data. We don't use kmem_zalloc() for the
 182                         * same reason - we don't need to zero the data area in
 183                         * the buffer, only the log vector header and the iovec
 184                         * storage.
 185                         */
 186                        kmem_free(lip->li_lv_shadow);
 187
 188                        lv = kmem_alloc_large(buf_size, KM_NOFS);
 189                        memset(lv, 0, xlog_cil_iovec_space(niovecs));
 190
 191                        lv->lv_item = lip;
 192                        lv->lv_size = buf_size;
 193                        if (ordered)
 194                                lv->lv_buf_len = XFS_LOG_VEC_ORDERED;
 195                        else
 196                                lv->lv_iovecp = (struct xfs_log_iovec *)&lv[1];
 197                        lip->li_lv_shadow = lv;
 198                } else {
 199                        /* same or smaller, optimise common overwrite case */
 200                        lv = lip->li_lv_shadow;
 201                        if (ordered)
 202                                lv->lv_buf_len = XFS_LOG_VEC_ORDERED;
 203                        else
 204                                lv->lv_buf_len = 0;
 205                        lv->lv_bytes = 0;
 206                        lv->lv_next = NULL;
 207                }
 208
 209                /* Ensure the lv is set up according to ->iop_size */
 210                lv->lv_niovecs = niovecs;
 211
 212                /* The allocated data region lies beyond the iovec region */
 213                lv->lv_buf = (char *)lv + xlog_cil_iovec_space(niovecs);
 214        }
 215
 216}
 217
 218/*
 219 * Prepare the log item for insertion into the CIL. Calculate the difference in
 220 * log space and vectors it will consume, and if it is a new item pin it as
 221 * well.
 222 */
 223STATIC void
 224xfs_cil_prepare_item(
 225        struct xlog             *log,
 226        struct xfs_log_vec      *lv,
 227        struct xfs_log_vec      *old_lv,
 228        int                     *diff_len,
 229        int                     *diff_iovecs)
 230{
 231        /* Account for the new LV being passed in */
 232        if (lv->lv_buf_len != XFS_LOG_VEC_ORDERED) {
 233                *diff_len += lv->lv_bytes;
 234                *diff_iovecs += lv->lv_niovecs;
 235        }
 236
 237        /*
 238         * If there is no old LV, this is the first time we've seen the item in
 239         * this CIL context and so we need to pin it. If we are replacing the
 240         * old_lv, then remove the space it accounts for and make it the shadow
 241         * buffer for later freeing. In both cases we are now switching to the
 242         * shadow buffer, so update the pointer to it appropriately.
 243         */
 244        if (!old_lv) {
 245                if (lv->lv_item->li_ops->iop_pin)
 246                        lv->lv_item->li_ops->iop_pin(lv->lv_item);
 247                lv->lv_item->li_lv_shadow = NULL;
 248        } else if (old_lv != lv) {
 249                ASSERT(lv->lv_buf_len != XFS_LOG_VEC_ORDERED);
 250
 251                *diff_len -= old_lv->lv_bytes;
 252                *diff_iovecs -= old_lv->lv_niovecs;
 253                lv->lv_item->li_lv_shadow = old_lv;
 254        }
 255
 256        /* attach new log vector to log item */
 257        lv->lv_item->li_lv = lv;
 258
 259        /*
 260         * If this is the first time the item is being committed to the
 261         * CIL, store the sequence number on the log item so we can
 262         * tell in future commits whether this is the first checkpoint
 263         * the item is being committed into.
 264         */
 265        if (!lv->lv_item->li_seq)
 266                lv->lv_item->li_seq = log->l_cilp->xc_ctx->sequence;
 267}
 268
 269/*
 270 * Format log item into a flat buffers
 271 *
 272 * For delayed logging, we need to hold a formatted buffer containing all the
 273 * changes on the log item. This enables us to relog the item in memory and
 274 * write it out asynchronously without needing to relock the object that was
 275 * modified at the time it gets written into the iclog.
 276 *
 277 * This function takes the prepared log vectors attached to each log item, and
 278 * formats the changes into the log vector buffer. The buffer it uses is
 279 * dependent on the current state of the vector in the CIL - the shadow lv is
 280 * guaranteed to be large enough for the current modification, but we will only
 281 * use that if we can't reuse the existing lv. If we can't reuse the existing
 282 * lv, then simple swap it out for the shadow lv. We don't free it - that is
 283 * done lazily either by th enext modification or the freeing of the log item.
 284 *
 285 * We don't set up region headers during this process; we simply copy the
 286 * regions into the flat buffer. We can do this because we still have to do a
 287 * formatting step to write the regions into the iclog buffer.  Writing the
 288 * ophdrs during the iclog write means that we can support splitting large
 289 * regions across iclog boundares without needing a change in the format of the
 290 * item/region encapsulation.
 291 *
 292 * Hence what we need to do now is change the rewrite the vector array to point
 293 * to the copied region inside the buffer we just allocated. This allows us to
 294 * format the regions into the iclog as though they are being formatted
 295 * directly out of the objects themselves.
 296 */
 297static void
 298xlog_cil_insert_format_items(
 299        struct xlog             *log,
 300        struct xfs_trans        *tp,
 301        int                     *diff_len,
 302        int                     *diff_iovecs)
 303{
 304        struct xfs_log_item     *lip;
 305
 306
 307        /* Bail out if we didn't find a log item.  */
 308        if (list_empty(&tp->t_items)) {
 309                ASSERT(0);
 310                return;
 311        }
 312
 313        list_for_each_entry(lip, &tp->t_items, li_trans) {
 314                struct xfs_log_vec *lv;
 315                struct xfs_log_vec *old_lv = NULL;
 316                struct xfs_log_vec *shadow;
 317                bool    ordered = false;
 318
 319                /* Skip items which aren't dirty in this transaction. */
 320                if (!test_bit(XFS_LI_DIRTY, &lip->li_flags))
 321                        continue;
 322
 323                /*
 324                 * The formatting size information is already attached to
 325                 * the shadow lv on the log item.
 326                 */
 327                shadow = lip->li_lv_shadow;
 328                if (shadow->lv_buf_len == XFS_LOG_VEC_ORDERED)
 329                        ordered = true;
 330
 331                /* Skip items that do not have any vectors for writing */
 332                if (!shadow->lv_niovecs && !ordered)
 333                        continue;
 334
 335                /* compare to existing item size */
 336                old_lv = lip->li_lv;
 337                if (lip->li_lv && shadow->lv_size <= lip->li_lv->lv_size) {
 338                        /* same or smaller, optimise common overwrite case */
 339                        lv = lip->li_lv;
 340                        lv->lv_next = NULL;
 341
 342                        if (ordered)
 343                                goto insert;
 344
 345                        /*
 346                         * set the item up as though it is a new insertion so
 347                         * that the space reservation accounting is correct.
 348                         */
 349                        *diff_iovecs -= lv->lv_niovecs;
 350                        *diff_len -= lv->lv_bytes;
 351
 352                        /* Ensure the lv is set up according to ->iop_size */
 353                        lv->lv_niovecs = shadow->lv_niovecs;
 354
 355                        /* reset the lv buffer information for new formatting */
 356                        lv->lv_buf_len = 0;
 357                        lv->lv_bytes = 0;
 358                        lv->lv_buf = (char *)lv +
 359                                        xlog_cil_iovec_space(lv->lv_niovecs);
 360                } else {
 361                        /* switch to shadow buffer! */
 362                        lv = shadow;
 363                        lv->lv_item = lip;
 364                        if (ordered) {
 365                                /* track as an ordered logvec */
 366                                ASSERT(lip->li_lv == NULL);
 367                                goto insert;
 368                        }
 369                }
 370
 371                ASSERT(IS_ALIGNED((unsigned long)lv->lv_buf, sizeof(uint64_t)));
 372                lip->li_ops->iop_format(lip, lv);
 373insert:
 374                xfs_cil_prepare_item(log, lv, old_lv, diff_len, diff_iovecs);
 375        }
 376}
 377
 378/*
 379 * Insert the log items into the CIL and calculate the difference in space
 380 * consumed by the item. Add the space to the checkpoint ticket and calculate
 381 * if the change requires additional log metadata. If it does, take that space
 382 * as well. Remove the amount of space we added to the checkpoint ticket from
 383 * the current transaction ticket so that the accounting works out correctly.
 384 */
 385static void
 386xlog_cil_insert_items(
 387        struct xlog             *log,
 388        struct xfs_trans        *tp)
 389{
 390        struct xfs_cil          *cil = log->l_cilp;
 391        struct xfs_cil_ctx      *ctx = cil->xc_ctx;
 392        struct xfs_log_item     *lip;
 393        int                     len = 0;
 394        int                     diff_iovecs = 0;
 395        int                     iclog_space;
 396        int                     iovhdr_res = 0, split_res = 0, ctx_res = 0;
 397
 398        ASSERT(tp);
 399
 400        /*
 401         * We can do this safely because the context can't checkpoint until we
 402         * are done so it doesn't matter exactly how we update the CIL.
 403         */
 404        xlog_cil_insert_format_items(log, tp, &len, &diff_iovecs);
 405
 406        spin_lock(&cil->xc_cil_lock);
 407
 408        /* account for space used by new iovec headers  */
 409        iovhdr_res = diff_iovecs * sizeof(xlog_op_header_t);
 410        len += iovhdr_res;
 411        ctx->nvecs += diff_iovecs;
 412
 413        /* attach the transaction to the CIL if it has any busy extents */
 414        if (!list_empty(&tp->t_busy))
 415                list_splice_init(&tp->t_busy, &ctx->busy_extents);
 416
 417        /*
 418         * Now transfer enough transaction reservation to the context ticket
 419         * for the checkpoint. The context ticket is special - the unit
 420         * reservation has to grow as well as the current reservation as we
 421         * steal from tickets so we can correctly determine the space used
 422         * during the transaction commit.
 423         */
 424        if (ctx->ticket->t_curr_res == 0) {
 425                ctx_res = ctx->ticket->t_unit_res;
 426                ctx->ticket->t_curr_res = ctx_res;
 427                tp->t_ticket->t_curr_res -= ctx_res;
 428        }
 429
 430        /* do we need space for more log record headers? */
 431        iclog_space = log->l_iclog_size - log->l_iclog_hsize;
 432        if (len > 0 && (ctx->space_used / iclog_space !=
 433                                (ctx->space_used + len) / iclog_space)) {
 434                split_res = (len + iclog_space - 1) / iclog_space;
 435                /* need to take into account split region headers, too */
 436                split_res *= log->l_iclog_hsize + sizeof(struct xlog_op_header);
 437                ctx->ticket->t_unit_res += split_res;
 438                ctx->ticket->t_curr_res += split_res;
 439                tp->t_ticket->t_curr_res -= split_res;
 440                ASSERT(tp->t_ticket->t_curr_res >= len);
 441        }
 442        tp->t_ticket->t_curr_res -= len;
 443        ctx->space_used += len;
 444
 445        /*
 446         * If we've overrun the reservation, dump the tx details before we move
 447         * the log items. Shutdown is imminent...
 448         */
 449        if (WARN_ON(tp->t_ticket->t_curr_res < 0)) {
 450                xfs_warn(log->l_mp, "Transaction log reservation overrun:");
 451                xfs_warn(log->l_mp,
 452                         "  log items: %d bytes (iov hdrs: %d bytes)",
 453                         len, iovhdr_res);
 454                xfs_warn(log->l_mp, "  split region headers: %d bytes",
 455                         split_res);
 456                xfs_warn(log->l_mp, "  ctx ticket: %d bytes", ctx_res);
 457                xlog_print_trans(tp);
 458        }
 459
 460        /*
 461         * Now (re-)position everything modified at the tail of the CIL.
 462         * We do this here so we only need to take the CIL lock once during
 463         * the transaction commit.
 464         */
 465        list_for_each_entry(lip, &tp->t_items, li_trans) {
 466
 467                /* Skip items which aren't dirty in this transaction. */
 468                if (!test_bit(XFS_LI_DIRTY, &lip->li_flags))
 469                        continue;
 470
 471                /*
 472                 * Only move the item if it isn't already at the tail. This is
 473                 * to prevent a transient list_empty() state when reinserting
 474                 * an item that is already the only item in the CIL.
 475                 */
 476                if (!list_is_last(&lip->li_cil, &cil->xc_cil))
 477                        list_move_tail(&lip->li_cil, &cil->xc_cil);
 478        }
 479
 480        spin_unlock(&cil->xc_cil_lock);
 481
 482        if (tp->t_ticket->t_curr_res < 0)
 483                xfs_force_shutdown(log->l_mp, SHUTDOWN_LOG_IO_ERROR);
 484}
 485
 486static void
 487xlog_cil_free_logvec(
 488        struct xfs_log_vec      *log_vector)
 489{
 490        struct xfs_log_vec      *lv;
 491
 492        for (lv = log_vector; lv; ) {
 493                struct xfs_log_vec *next = lv->lv_next;
 494                kmem_free(lv);
 495                lv = next;
 496        }
 497}
 498
 499static void
 500xlog_discard_endio_work(
 501        struct work_struct      *work)
 502{
 503        struct xfs_cil_ctx      *ctx =
 504                container_of(work, struct xfs_cil_ctx, discard_endio_work);
 505        struct xfs_mount        *mp = ctx->cil->xc_log->l_mp;
 506
 507        xfs_extent_busy_clear(mp, &ctx->busy_extents, false);
 508        kmem_free(ctx);
 509}
 510
 511/*
 512 * Queue up the actual completion to a thread to avoid IRQ-safe locking for
 513 * pagb_lock.  Note that we need a unbounded workqueue, otherwise we might
 514 * get the execution delayed up to 30 seconds for weird reasons.
 515 */
 516static void
 517xlog_discard_endio(
 518        struct bio              *bio)
 519{
 520        struct xfs_cil_ctx      *ctx = bio->bi_private;
 521
 522        INIT_WORK(&ctx->discard_endio_work, xlog_discard_endio_work);
 523        queue_work(xfs_discard_wq, &ctx->discard_endio_work);
 524        bio_put(bio);
 525}
 526
 527static void
 528xlog_discard_busy_extents(
 529        struct xfs_mount        *mp,
 530        struct xfs_cil_ctx      *ctx)
 531{
 532        struct list_head        *list = &ctx->busy_extents;
 533        struct xfs_extent_busy  *busyp;
 534        struct bio              *bio = NULL;
 535        struct blk_plug         plug;
 536        int                     error = 0;
 537
 538        ASSERT(mp->m_flags & XFS_MOUNT_DISCARD);
 539
 540        blk_start_plug(&plug);
 541        list_for_each_entry(busyp, list, list) {
 542                trace_xfs_discard_extent(mp, busyp->agno, busyp->bno,
 543                                         busyp->length);
 544
 545                error = __blkdev_issue_discard(mp->m_ddev_targp->bt_bdev,
 546                                XFS_AGB_TO_DADDR(mp, busyp->agno, busyp->bno),
 547                                XFS_FSB_TO_BB(mp, busyp->length),
 548                                GFP_NOFS, 0, &bio);
 549                if (error && error != -EOPNOTSUPP) {
 550                        xfs_info(mp,
 551         "discard failed for extent [0x%llx,%u], error %d",
 552                                 (unsigned long long)busyp->bno,
 553                                 busyp->length,
 554                                 error);
 555                        break;
 556                }
 557        }
 558
 559        if (bio) {
 560                bio->bi_private = ctx;
 561                bio->bi_end_io = xlog_discard_endio;
 562                submit_bio(bio);
 563        } else {
 564                xlog_discard_endio_work(&ctx->discard_endio_work);
 565        }
 566        blk_finish_plug(&plug);
 567}
 568
 569/*
 570 * Mark all items committed and clear busy extents. We free the log vector
 571 * chains in a separate pass so that we unpin the log items as quickly as
 572 * possible.
 573 */
 574static void
 575xlog_cil_committed(
 576        struct xfs_cil_ctx      *ctx)
 577{
 578        struct xfs_mount        *mp = ctx->cil->xc_log->l_mp;
 579        bool                    abort = XLOG_FORCED_SHUTDOWN(ctx->cil->xc_log);
 580
 581        /*
 582         * If the I/O failed, we're aborting the commit and already shutdown.
 583         * Wake any commit waiters before aborting the log items so we don't
 584         * block async log pushers on callbacks. Async log pushers explicitly do
 585         * not wait on log force completion because they may be holding locks
 586         * required to unpin items.
 587         */
 588        if (abort) {
 589                spin_lock(&ctx->cil->xc_push_lock);
 590                wake_up_all(&ctx->cil->xc_commit_wait);
 591                spin_unlock(&ctx->cil->xc_push_lock);
 592        }
 593
 594        xfs_trans_committed_bulk(ctx->cil->xc_log->l_ailp, ctx->lv_chain,
 595                                        ctx->start_lsn, abort);
 596
 597        xfs_extent_busy_sort(&ctx->busy_extents);
 598        xfs_extent_busy_clear(mp, &ctx->busy_extents,
 599                             (mp->m_flags & XFS_MOUNT_DISCARD) && !abort);
 600
 601        spin_lock(&ctx->cil->xc_push_lock);
 602        list_del(&ctx->committing);
 603        spin_unlock(&ctx->cil->xc_push_lock);
 604
 605        xlog_cil_free_logvec(ctx->lv_chain);
 606
 607        if (!list_empty(&ctx->busy_extents))
 608                xlog_discard_busy_extents(mp, ctx);
 609        else
 610                kmem_free(ctx);
 611}
 612
 613void
 614xlog_cil_process_committed(
 615        struct list_head        *list)
 616{
 617        struct xfs_cil_ctx      *ctx;
 618
 619        while ((ctx = list_first_entry_or_null(list,
 620                        struct xfs_cil_ctx, iclog_entry))) {
 621                list_del(&ctx->iclog_entry);
 622                xlog_cil_committed(ctx);
 623        }
 624}
 625
 626/*
 627 * Push the Committed Item List to the log.
 628 *
 629 * If the current sequence is the same as xc_push_seq we need to do a flush. If
 630 * xc_push_seq is less than the current sequence, then it has already been
 631 * flushed and we don't need to do anything - the caller will wait for it to
 632 * complete if necessary.
 633 *
 634 * xc_push_seq is checked unlocked against the sequence number for a match.
 635 * Hence we can allow log forces to run racily and not issue pushes for the
 636 * same sequence twice.  If we get a race between multiple pushes for the same
 637 * sequence they will block on the first one and then abort, hence avoiding
 638 * needless pushes.
 639 */
 640static void
 641xlog_cil_push_work(
 642        struct work_struct      *work)
 643{
 644        struct xfs_cil          *cil =
 645                container_of(work, struct xfs_cil, xc_push_work);
 646        struct xlog             *log = cil->xc_log;
 647        struct xfs_log_vec      *lv;
 648        struct xfs_cil_ctx      *ctx;
 649        struct xfs_cil_ctx      *new_ctx;
 650        struct xlog_in_core     *commit_iclog;
 651        struct xlog_ticket      *tic;
 652        int                     num_iovecs;
 653        int                     error = 0;
 654        struct xfs_trans_header thdr;
 655        struct xfs_log_iovec    lhdr;
 656        struct xfs_log_vec      lvhdr = { NULL };
 657        xfs_lsn_t               preflush_tail_lsn;
 658        xfs_lsn_t               commit_lsn;
 659        xfs_csn_t               push_seq;
 660        struct bio              bio;
 661        DECLARE_COMPLETION_ONSTACK(bdev_flush);
 662
 663        new_ctx = kmem_zalloc(sizeof(*new_ctx), KM_NOFS);
 664        new_ctx->ticket = xlog_cil_ticket_alloc(log);
 665
 666        down_write(&cil->xc_ctx_lock);
 667        ctx = cil->xc_ctx;
 668
 669        spin_lock(&cil->xc_push_lock);
 670        push_seq = cil->xc_push_seq;
 671        ASSERT(push_seq <= ctx->sequence);
 672
 673        /*
 674         * As we are about to switch to a new, empty CIL context, we no longer
 675         * need to throttle tasks on CIL space overruns. Wake any waiters that
 676         * the hard push throttle may have caught so they can start committing
 677         * to the new context. The ctx->xc_push_lock provides the serialisation
 678         * necessary for safely using the lockless waitqueue_active() check in
 679         * this context.
 680         */
 681        if (waitqueue_active(&cil->xc_push_wait))
 682                wake_up_all(&cil->xc_push_wait);
 683
 684        /*
 685         * Check if we've anything to push. If there is nothing, then we don't
 686         * move on to a new sequence number and so we have to be able to push
 687         * this sequence again later.
 688         */
 689        if (list_empty(&cil->xc_cil)) {
 690                cil->xc_push_seq = 0;
 691                spin_unlock(&cil->xc_push_lock);
 692                goto out_skip;
 693        }
 694
 695
 696        /* check for a previously pushed sequence */
 697        if (push_seq < cil->xc_ctx->sequence) {
 698                spin_unlock(&cil->xc_push_lock);
 699                goto out_skip;
 700        }
 701
 702        /*
 703         * We are now going to push this context, so add it to the committing
 704         * list before we do anything else. This ensures that anyone waiting on
 705         * this push can easily detect the difference between a "push in
 706         * progress" and "CIL is empty, nothing to do".
 707         *
 708         * IOWs, a wait loop can now check for:
 709         *      the current sequence not being found on the committing list;
 710         *      an empty CIL; and
 711         *      an unchanged sequence number
 712         * to detect a push that had nothing to do and therefore does not need
 713         * waiting on. If the CIL is not empty, we get put on the committing
 714         * list before emptying the CIL and bumping the sequence number. Hence
 715         * an empty CIL and an unchanged sequence number means we jumped out
 716         * above after doing nothing.
 717         *
 718         * Hence the waiter will either find the commit sequence on the
 719         * committing list or the sequence number will be unchanged and the CIL
 720         * still dirty. In that latter case, the push has not yet started, and
 721         * so the waiter will have to continue trying to check the CIL
 722         * committing list until it is found. In extreme cases of delay, the
 723         * sequence may fully commit between the attempts the wait makes to wait
 724         * on the commit sequence.
 725         */
 726        list_add(&ctx->committing, &cil->xc_committing);
 727        spin_unlock(&cil->xc_push_lock);
 728
 729        /*
 730         * The CIL is stable at this point - nothing new will be added to it
 731         * because we hold the flush lock exclusively. Hence we can now issue
 732         * a cache flush to ensure all the completed metadata in the journal we
 733         * are about to overwrite is on stable storage.
 734         *
 735         * Because we are issuing this cache flush before we've written the
 736         * tail lsn to the iclog, we can have metadata IO completions move the
 737         * tail forwards between the completion of this flush and the iclog
 738         * being written. In this case, we need to re-issue the cache flush
 739         * before the iclog write. To detect whether the log tail moves, sample
 740         * the tail LSN *before* we issue the flush.
 741         */
 742        preflush_tail_lsn = atomic64_read(&log->l_tail_lsn);
 743        xfs_flush_bdev_async(&bio, log->l_mp->m_ddev_targp->bt_bdev,
 744                                &bdev_flush);
 745
 746        /*
 747         * Pull all the log vectors off the items in the CIL, and remove the
 748         * items from the CIL. We don't need the CIL lock here because it's only
 749         * needed on the transaction commit side which is currently locked out
 750         * by the flush lock.
 751         */
 752        lv = NULL;
 753        num_iovecs = 0;
 754        while (!list_empty(&cil->xc_cil)) {
 755                struct xfs_log_item     *item;
 756
 757                item = list_first_entry(&cil->xc_cil,
 758                                        struct xfs_log_item, li_cil);
 759                list_del_init(&item->li_cil);
 760                if (!ctx->lv_chain)
 761                        ctx->lv_chain = item->li_lv;
 762                else
 763                        lv->lv_next = item->li_lv;
 764                lv = item->li_lv;
 765                item->li_lv = NULL;
 766                num_iovecs += lv->lv_niovecs;
 767        }
 768
 769        /*
 770         * initialise the new context and attach it to the CIL. Then attach
 771         * the current context to the CIL committing list so it can be found
 772         * during log forces to extract the commit lsn of the sequence that
 773         * needs to be forced.
 774         */
 775        INIT_LIST_HEAD(&new_ctx->committing);
 776        INIT_LIST_HEAD(&new_ctx->busy_extents);
 777        new_ctx->sequence = ctx->sequence + 1;
 778        new_ctx->cil = cil;
 779        cil->xc_ctx = new_ctx;
 780
 781        /*
 782         * The switch is now done, so we can drop the context lock and move out
 783         * of a shared context. We can't just go straight to the commit record,
 784         * though - we need to synchronise with previous and future commits so
 785         * that the commit records are correctly ordered in the log to ensure
 786         * that we process items during log IO completion in the correct order.
 787         *
 788         * For example, if we get an EFI in one checkpoint and the EFD in the
 789         * next (e.g. due to log forces), we do not want the checkpoint with
 790         * the EFD to be committed before the checkpoint with the EFI.  Hence
 791         * we must strictly order the commit records of the checkpoints so
 792         * that: a) the checkpoint callbacks are attached to the iclogs in the
 793         * correct order; and b) the checkpoints are replayed in correct order
 794         * in log recovery.
 795         *
 796         * Hence we need to add this context to the committing context list so
 797         * that higher sequences will wait for us to write out a commit record
 798         * before they do.
 799         *
 800         * xfs_log_force_seq requires us to mirror the new sequence into the cil
 801         * structure atomically with the addition of this sequence to the
 802         * committing list. This also ensures that we can do unlocked checks
 803         * against the current sequence in log forces without risking
 804         * deferencing a freed context pointer.
 805         */
 806        spin_lock(&cil->xc_push_lock);
 807        cil->xc_current_sequence = new_ctx->sequence;
 808        spin_unlock(&cil->xc_push_lock);
 809        up_write(&cil->xc_ctx_lock);
 810
 811        /*
 812         * Build a checkpoint transaction header and write it to the log to
 813         * begin the transaction. We need to account for the space used by the
 814         * transaction header here as it is not accounted for in xlog_write().
 815         *
 816         * The LSN we need to pass to the log items on transaction commit is
 817         * the LSN reported by the first log vector write. If we use the commit
 818         * record lsn then we can move the tail beyond the grant write head.
 819         */
 820        tic = ctx->ticket;
 821        thdr.th_magic = XFS_TRANS_HEADER_MAGIC;
 822        thdr.th_type = XFS_TRANS_CHECKPOINT;
 823        thdr.th_tid = tic->t_tid;
 824        thdr.th_num_items = num_iovecs;
 825        lhdr.i_addr = &thdr;
 826        lhdr.i_len = sizeof(xfs_trans_header_t);
 827        lhdr.i_type = XLOG_REG_TYPE_TRANSHDR;
 828        tic->t_curr_res -= lhdr.i_len + sizeof(xlog_op_header_t);
 829
 830        lvhdr.lv_niovecs = 1;
 831        lvhdr.lv_iovecp = &lhdr;
 832        lvhdr.lv_next = ctx->lv_chain;
 833
 834        /*
 835         * Before we format and submit the first iclog, we have to ensure that
 836         * the metadata writeback ordering cache flush is complete.
 837         */
 838        wait_for_completion(&bdev_flush);
 839
 840        error = xlog_write(log, &lvhdr, tic, &ctx->start_lsn, NULL,
 841                                XLOG_START_TRANS);
 842        if (error)
 843                goto out_abort_free_ticket;
 844
 845        /*
 846         * now that we've written the checkpoint into the log, strictly
 847         * order the commit records so replay will get them in the right order.
 848         */
 849restart:
 850        spin_lock(&cil->xc_push_lock);
 851        list_for_each_entry(new_ctx, &cil->xc_committing, committing) {
 852                /*
 853                 * Avoid getting stuck in this loop because we were woken by the
 854                 * shutdown, but then went back to sleep once already in the
 855                 * shutdown state.
 856                 */
 857                if (XLOG_FORCED_SHUTDOWN(log)) {
 858                        spin_unlock(&cil->xc_push_lock);
 859                        goto out_abort_free_ticket;
 860                }
 861
 862                /*
 863                 * Higher sequences will wait for this one so skip them.
 864                 * Don't wait for our own sequence, either.
 865                 */
 866                if (new_ctx->sequence >= ctx->sequence)
 867                        continue;
 868                if (!new_ctx->commit_lsn) {
 869                        /*
 870                         * It is still being pushed! Wait for the push to
 871                         * complete, then start again from the beginning.
 872                         */
 873                        xlog_wait(&cil->xc_commit_wait, &cil->xc_push_lock);
 874                        goto restart;
 875                }
 876        }
 877        spin_unlock(&cil->xc_push_lock);
 878
 879        error = xlog_commit_record(log, tic, &commit_iclog, &commit_lsn);
 880        if (error)
 881                goto out_abort_free_ticket;
 882
 883        xfs_log_ticket_ungrant(log, tic);
 884
 885        /*
 886         * Once we attach the ctx to the iclog, a shutdown can process the
 887         * iclog, run the callbacks and free the ctx. The only thing preventing
 888         * this potential UAF situation here is that we are holding the
 889         * icloglock. Hence we cannot access the ctx once we have attached the
 890         * callbacks and dropped the icloglock.
 891         */
 892        spin_lock(&log->l_icloglock);
 893        if (commit_iclog->ic_state == XLOG_STATE_IOERROR) {
 894                spin_unlock(&log->l_icloglock);
 895                goto out_abort;
 896        }
 897        ASSERT_ALWAYS(commit_iclog->ic_state == XLOG_STATE_ACTIVE ||
 898                      commit_iclog->ic_state == XLOG_STATE_WANT_SYNC);
 899        list_add_tail(&ctx->iclog_entry, &commit_iclog->ic_callbacks);
 900
 901        /*
 902         * now the checkpoint commit is complete and we've attached the
 903         * callbacks to the iclog we can assign the commit LSN to the context
 904         * and wake up anyone who is waiting for the commit to complete.
 905         */
 906        spin_lock(&cil->xc_push_lock);
 907        ctx->commit_lsn = commit_lsn;
 908        wake_up_all(&cil->xc_commit_wait);
 909        spin_unlock(&cil->xc_push_lock);
 910
 911        /*
 912         * If the checkpoint spans multiple iclogs, wait for all previous iclogs
 913         * to complete before we submit the commit_iclog. We can't use state
 914         * checks for this - ACTIVE can be either a past completed iclog or a
 915         * future iclog being filled, while WANT_SYNC through SYNC_DONE can be a
 916         * past or future iclog awaiting IO or ordered IO completion to be run.
 917         * In the latter case, if it's a future iclog and we wait on it, the we
 918         * will hang because it won't get processed through to ic_force_wait
 919         * wakeup until this commit_iclog is written to disk.  Hence we use the
 920         * iclog header lsn and compare it to the commit lsn to determine if we
 921         * need to wait on iclogs or not.
 922         *
 923         * NOTE: It is not safe to reference the ctx after this check as we drop
 924         * the icloglock if we have to wait for completion of other iclogs.
 925         */
 926        if (ctx->start_lsn != commit_lsn) {
 927                xfs_lsn_t       plsn;
 928
 929                plsn = be64_to_cpu(commit_iclog->ic_prev->ic_header.h_lsn);
 930                if (plsn && XFS_LSN_CMP(plsn, commit_lsn) < 0) {
 931                        /*
 932                         * Waiting on ic_force_wait orders the completion of
 933                         * iclogs older than ic_prev. Hence we only need to wait
 934                         * on the most recent older iclog here.
 935                         */
 936                        xlog_wait_on_iclog(commit_iclog->ic_prev);
 937                        spin_lock(&log->l_icloglock);
 938                }
 939
 940                /*
 941                 * We need to issue a pre-flush so that the ordering for this
 942                 * checkpoint is correctly preserved down to stable storage.
 943                 */
 944                commit_iclog->ic_flags |= XLOG_ICL_NEED_FLUSH;
 945        }
 946
 947        /*
 948         * The commit iclog must be written to stable storage to guarantee
 949         * journal IO vs metadata writeback IO is correctly ordered on stable
 950         * storage.
 951         */
 952        commit_iclog->ic_flags |= XLOG_ICL_NEED_FUA;
 953        xlog_state_release_iclog(log, commit_iclog, preflush_tail_lsn);
 954        spin_unlock(&log->l_icloglock);
 955        return;
 956
 957out_skip:
 958        up_write(&cil->xc_ctx_lock);
 959        xfs_log_ticket_put(new_ctx->ticket);
 960        kmem_free(new_ctx);
 961        return;
 962
 963out_abort_free_ticket:
 964        xfs_log_ticket_ungrant(log, tic);
 965out_abort:
 966        ASSERT(XLOG_FORCED_SHUTDOWN(log));
 967        xlog_cil_committed(ctx);
 968}
 969
 970/*
 971 * We need to push CIL every so often so we don't cache more than we can fit in
 972 * the log. The limit really is that a checkpoint can't be more than half the
 973 * log (the current checkpoint is not allowed to overwrite the previous
 974 * checkpoint), but commit latency and memory usage limit this to a smaller
 975 * size.
 976 */
 977static void
 978xlog_cil_push_background(
 979        struct xlog     *log) __releases(cil->xc_ctx_lock)
 980{
 981        struct xfs_cil  *cil = log->l_cilp;
 982
 983        /*
 984         * The cil won't be empty because we are called while holding the
 985         * context lock so whatever we added to the CIL will still be there
 986         */
 987        ASSERT(!list_empty(&cil->xc_cil));
 988
 989        /*
 990         * Don't do a background push if we haven't used up all the
 991         * space available yet.
 992         */
 993        if (cil->xc_ctx->space_used < XLOG_CIL_SPACE_LIMIT(log)) {
 994                up_read(&cil->xc_ctx_lock);
 995                return;
 996        }
 997
 998        spin_lock(&cil->xc_push_lock);
 999        if (cil->xc_push_seq < cil->xc_current_sequence) {
1000                cil->xc_push_seq = cil->xc_current_sequence;
1001                queue_work(log->l_mp->m_cil_workqueue, &cil->xc_push_work);
1002        }
1003
1004        /*
1005         * Drop the context lock now, we can't hold that if we need to sleep
1006         * because we are over the blocking threshold. The push_lock is still
1007         * held, so blocking threshold sleep/wakeup is still correctly
1008         * serialised here.
1009         */
1010        up_read(&cil->xc_ctx_lock);
1011
1012        /*
1013         * If we are well over the space limit, throttle the work that is being
1014         * done until the push work on this context has begun. Enforce the hard
1015         * throttle on all transaction commits once it has been activated, even
1016         * if the committing transactions have resulted in the space usage
1017         * dipping back down under the hard limit.
1018         *
1019         * The ctx->xc_push_lock provides the serialisation necessary for safely
1020         * using the lockless waitqueue_active() check in this context.
1021         */
1022        if (cil->xc_ctx->space_used >= XLOG_CIL_BLOCKING_SPACE_LIMIT(log) ||
1023            waitqueue_active(&cil->xc_push_wait)) {
1024                trace_xfs_log_cil_wait(log, cil->xc_ctx->ticket);
1025                ASSERT(cil->xc_ctx->space_used < log->l_logsize);
1026                xlog_wait(&cil->xc_push_wait, &cil->xc_push_lock);
1027                return;
1028        }
1029
1030        spin_unlock(&cil->xc_push_lock);
1031
1032}
1033
1034/*
1035 * xlog_cil_push_now() is used to trigger an immediate CIL push to the sequence
1036 * number that is passed. When it returns, the work will be queued for
1037 * @push_seq, but it won't be completed. The caller is expected to do any
1038 * waiting for push_seq to complete if it is required.
1039 */
1040static void
1041xlog_cil_push_now(
1042        struct xlog     *log,
1043        xfs_lsn_t       push_seq)
1044{
1045        struct xfs_cil  *cil = log->l_cilp;
1046
1047        if (!cil)
1048                return;
1049
1050        ASSERT(push_seq && push_seq <= cil->xc_current_sequence);
1051
1052        /* start on any pending background push to minimise wait time on it */
1053        flush_work(&cil->xc_push_work);
1054
1055        /*
1056         * If the CIL is empty or we've already pushed the sequence then
1057         * there's no work we need to do.
1058         */
1059        spin_lock(&cil->xc_push_lock);
1060        if (list_empty(&cil->xc_cil) || push_seq <= cil->xc_push_seq) {
1061                spin_unlock(&cil->xc_push_lock);
1062                return;
1063        }
1064
1065        cil->xc_push_seq = push_seq;
1066        queue_work(log->l_mp->m_cil_workqueue, &cil->xc_push_work);
1067        spin_unlock(&cil->xc_push_lock);
1068}
1069
1070bool
1071xlog_cil_empty(
1072        struct xlog     *log)
1073{
1074        struct xfs_cil  *cil = log->l_cilp;
1075        bool            empty = false;
1076
1077        spin_lock(&cil->xc_push_lock);
1078        if (list_empty(&cil->xc_cil))
1079                empty = true;
1080        spin_unlock(&cil->xc_push_lock);
1081        return empty;
1082}
1083
1084/*
1085 * Commit a transaction with the given vector to the Committed Item List.
1086 *
1087 * To do this, we need to format the item, pin it in memory if required and
1088 * account for the space used by the transaction. Once we have done that we
1089 * need to release the unused reservation for the transaction, attach the
1090 * transaction to the checkpoint context so we carry the busy extents through
1091 * to checkpoint completion, and then unlock all the items in the transaction.
1092 *
1093 * Called with the context lock already held in read mode to lock out
1094 * background commit, returns without it held once background commits are
1095 * allowed again.
1096 */
1097void
1098xlog_cil_commit(
1099        struct xlog             *log,
1100        struct xfs_trans        *tp,
1101        xfs_csn_t               *commit_seq,
1102        bool                    regrant)
1103{
1104        struct xfs_cil          *cil = log->l_cilp;
1105        struct xfs_log_item     *lip, *next;
1106
1107        /*
1108         * Do all necessary memory allocation before we lock the CIL.
1109         * This ensures the allocation does not deadlock with a CIL
1110         * push in memory reclaim (e.g. from kswapd).
1111         */
1112        xlog_cil_alloc_shadow_bufs(log, tp);
1113
1114        /* lock out background commit */
1115        down_read(&cil->xc_ctx_lock);
1116
1117        xlog_cil_insert_items(log, tp);
1118
1119        if (regrant && !XLOG_FORCED_SHUTDOWN(log))
1120                xfs_log_ticket_regrant(log, tp->t_ticket);
1121        else
1122                xfs_log_ticket_ungrant(log, tp->t_ticket);
1123        tp->t_ticket = NULL;
1124        xfs_trans_unreserve_and_mod_sb(tp);
1125
1126        /*
1127         * Once all the items of the transaction have been copied to the CIL,
1128         * the items can be unlocked and possibly freed.
1129         *
1130         * This needs to be done before we drop the CIL context lock because we
1131         * have to update state in the log items and unlock them before they go
1132         * to disk. If we don't, then the CIL checkpoint can race with us and
1133         * we can run checkpoint completion before we've updated and unlocked
1134         * the log items. This affects (at least) processing of stale buffers,
1135         * inodes and EFIs.
1136         */
1137        trace_xfs_trans_commit_items(tp, _RET_IP_);
1138        list_for_each_entry_safe(lip, next, &tp->t_items, li_trans) {
1139                xfs_trans_del_item(lip);
1140                if (lip->li_ops->iop_committing)
1141                        lip->li_ops->iop_committing(lip, cil->xc_ctx->sequence);
1142        }
1143        if (commit_seq)
1144                *commit_seq = cil->xc_ctx->sequence;
1145
1146        /* xlog_cil_push_background() releases cil->xc_ctx_lock */
1147        xlog_cil_push_background(log);
1148}
1149
1150/*
1151 * Conditionally push the CIL based on the sequence passed in.
1152 *
1153 * We only need to push if we haven't already pushed the sequence
1154 * number given. Hence the only time we will trigger a push here is
1155 * if the push sequence is the same as the current context.
1156 *
1157 * We return the current commit lsn to allow the callers to determine if a
1158 * iclog flush is necessary following this call.
1159 */
1160xfs_lsn_t
1161xlog_cil_force_seq(
1162        struct xlog     *log,
1163        xfs_csn_t       sequence)
1164{
1165        struct xfs_cil          *cil = log->l_cilp;
1166        struct xfs_cil_ctx      *ctx;
1167        xfs_lsn_t               commit_lsn = NULLCOMMITLSN;
1168
1169        ASSERT(sequence <= cil->xc_current_sequence);
1170
1171        /*
1172         * check to see if we need to force out the current context.
1173         * xlog_cil_push() handles racing pushes for the same sequence,
1174         * so no need to deal with it here.
1175         */
1176restart:
1177        xlog_cil_push_now(log, sequence);
1178
1179        /*
1180         * See if we can find a previous sequence still committing.
1181         * We need to wait for all previous sequence commits to complete
1182         * before allowing the force of push_seq to go ahead. Hence block
1183         * on commits for those as well.
1184         */
1185        spin_lock(&cil->xc_push_lock);
1186        list_for_each_entry(ctx, &cil->xc_committing, committing) {
1187                /*
1188                 * Avoid getting stuck in this loop because we were woken by the
1189                 * shutdown, but then went back to sleep once already in the
1190                 * shutdown state.
1191                 */
1192                if (XLOG_FORCED_SHUTDOWN(log))
1193                        goto out_shutdown;
1194                if (ctx->sequence > sequence)
1195                        continue;
1196                if (!ctx->commit_lsn) {
1197                        /*
1198                         * It is still being pushed! Wait for the push to
1199                         * complete, then start again from the beginning.
1200                         */
1201                        xlog_wait(&cil->xc_commit_wait, &cil->xc_push_lock);
1202                        goto restart;
1203                }
1204                if (ctx->sequence != sequence)
1205                        continue;
1206                /* found it! */
1207                commit_lsn = ctx->commit_lsn;
1208        }
1209
1210        /*
1211         * The call to xlog_cil_push_now() executes the push in the background.
1212         * Hence by the time we have got here it our sequence may not have been
1213         * pushed yet. This is true if the current sequence still matches the
1214         * push sequence after the above wait loop and the CIL still contains
1215         * dirty objects. This is guaranteed by the push code first adding the
1216         * context to the committing list before emptying the CIL.
1217         *
1218         * Hence if we don't find the context in the committing list and the
1219         * current sequence number is unchanged then the CIL contents are
1220         * significant.  If the CIL is empty, if means there was nothing to push
1221         * and that means there is nothing to wait for. If the CIL is not empty,
1222         * it means we haven't yet started the push, because if it had started
1223         * we would have found the context on the committing list.
1224         */
1225        if (sequence == cil->xc_current_sequence &&
1226            !list_empty(&cil->xc_cil)) {
1227                spin_unlock(&cil->xc_push_lock);
1228                goto restart;
1229        }
1230
1231        spin_unlock(&cil->xc_push_lock);
1232        return commit_lsn;
1233
1234        /*
1235         * We detected a shutdown in progress. We need to trigger the log force
1236         * to pass through it's iclog state machine error handling, even though
1237         * we are already in a shutdown state. Hence we can't return
1238         * NULLCOMMITLSN here as that has special meaning to log forces (i.e.
1239         * LSN is already stable), so we return a zero LSN instead.
1240         */
1241out_shutdown:
1242        spin_unlock(&cil->xc_push_lock);
1243        return 0;
1244}
1245
1246/*
1247 * Check if the current log item was first committed in this sequence.
1248 * We can't rely on just the log item being in the CIL, we have to check
1249 * the recorded commit sequence number.
1250 *
1251 * Note: for this to be used in a non-racy manner, it has to be called with
1252 * CIL flushing locked out. As a result, it should only be used during the
1253 * transaction commit process when deciding what to format into the item.
1254 */
1255bool
1256xfs_log_item_in_current_chkpt(
1257        struct xfs_log_item *lip)
1258{
1259        struct xfs_cil_ctx *ctx = lip->li_mountp->m_log->l_cilp->xc_ctx;
1260
1261        if (list_empty(&lip->li_cil))
1262                return false;
1263
1264        /*
1265         * li_seq is written on the first commit of a log item to record the
1266         * first checkpoint it is written to. Hence if it is different to the
1267         * current sequence, we're in a new checkpoint.
1268         */
1269        return lip->li_seq == ctx->sequence;
1270}
1271
1272/*
1273 * Perform initial CIL structure initialisation.
1274 */
1275int
1276xlog_cil_init(
1277        struct xlog     *log)
1278{
1279        struct xfs_cil  *cil;
1280        struct xfs_cil_ctx *ctx;
1281
1282        cil = kmem_zalloc(sizeof(*cil), KM_MAYFAIL);
1283        if (!cil)
1284                return -ENOMEM;
1285
1286        ctx = kmem_zalloc(sizeof(*ctx), KM_MAYFAIL);
1287        if (!ctx) {
1288                kmem_free(cil);
1289                return -ENOMEM;
1290        }
1291
1292        INIT_WORK(&cil->xc_push_work, xlog_cil_push_work);
1293        INIT_LIST_HEAD(&cil->xc_cil);
1294        INIT_LIST_HEAD(&cil->xc_committing);
1295        spin_lock_init(&cil->xc_cil_lock);
1296        spin_lock_init(&cil->xc_push_lock);
1297        init_waitqueue_head(&cil->xc_push_wait);
1298        init_rwsem(&cil->xc_ctx_lock);
1299        init_waitqueue_head(&cil->xc_commit_wait);
1300
1301        INIT_LIST_HEAD(&ctx->committing);
1302        INIT_LIST_HEAD(&ctx->busy_extents);
1303        ctx->sequence = 1;
1304        ctx->cil = cil;
1305        cil->xc_ctx = ctx;
1306        cil->xc_current_sequence = ctx->sequence;
1307
1308        cil->xc_log = log;
1309        log->l_cilp = cil;
1310        return 0;
1311}
1312
1313void
1314xlog_cil_destroy(
1315        struct xlog     *log)
1316{
1317        if (log->l_cilp->xc_ctx) {
1318                if (log->l_cilp->xc_ctx->ticket)
1319                        xfs_log_ticket_put(log->l_cilp->xc_ctx->ticket);
1320                kmem_free(log->l_cilp->xc_ctx);
1321        }
1322
1323        ASSERT(list_empty(&log->l_cilp->xc_cil));
1324        kmem_free(log->l_cilp);
1325}
1326
1327
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.