2 * Copyright (C) 2016 Oracle. All Rights Reserved.
4 * Author: Darrick J. Wong <darrick.wong@oracle.com>
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version 2
9 * of the License, or (at your option) any later version.
11 * This program is distributed in the hope that it would be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA.
22 #include "xfs_format.h"
23 #include "xfs_log_format.h"
24 #include "xfs_trans_resv.h"
26 #include "xfs_mount.h"
27 #include "xfs_defer.h"
28 #include "xfs_trans.h"
29 #include "xfs_trans_priv.h"
30 #include "xfs_buf_item.h"
31 #include "xfs_refcount_item.h"
33 #include "xfs_refcount.h"
36 kmem_zone_t *xfs_cui_zone;
37 kmem_zone_t *xfs_cud_zone;
39 static inline struct xfs_cui_log_item *CUI_ITEM(struct xfs_log_item *lip)
41 return container_of(lip, struct xfs_cui_log_item, cui_item);
46 struct xfs_cui_log_item *cuip)
48 if (cuip->cui_format.cui_nextents > XFS_CUI_MAX_FAST_EXTENTS)
51 kmem_zone_free(xfs_cui_zone, cuip);
56 struct xfs_log_item *lip,
60 struct xfs_cui_log_item *cuip = CUI_ITEM(lip);
63 *nbytes += xfs_cui_log_format_sizeof(cuip->cui_format.cui_nextents);
67 * This is called to fill in the vector of log iovecs for the
68 * given cui log item. We use only 1 iovec, and we point that
69 * at the cui_log_format structure embedded in the cui item.
70 * It is at this point that we assert that all of the extent
71 * slots in the cui item have been filled.
75 struct xfs_log_item *lip,
76 struct xfs_log_vec *lv)
78 struct xfs_cui_log_item *cuip = CUI_ITEM(lip);
79 struct xfs_log_iovec *vecp = NULL;
81 ASSERT(atomic_read(&cuip->cui_next_extent) ==
82 cuip->cui_format.cui_nextents);
84 cuip->cui_format.cui_type = XFS_LI_CUI;
85 cuip->cui_format.cui_size = 1;
87 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_CUI_FORMAT, &cuip->cui_format,
88 xfs_cui_log_format_sizeof(cuip->cui_format.cui_nextents));
92 * Pinning has no meaning for an cui item, so just return.
96 struct xfs_log_item *lip)
101 * The unpin operation is the last place an CUI is manipulated in the log. It is
102 * either inserted in the AIL or aborted in the event of a log I/O error. In
103 * either case, the CUI transaction has been successfully committed to make it
104 * this far. Therefore, we expect whoever committed the CUI to either construct
105 * and commit the CUD or drop the CUD's reference in the event of error. Simply
106 * drop the log's CUI reference now that the log is done with it.
110 struct xfs_log_item *lip,
113 struct xfs_cui_log_item *cuip = CUI_ITEM(lip);
115 xfs_cui_release(cuip);
119 * CUI items have no locking or pushing. However, since CUIs are pulled from
120 * the AIL when their corresponding CUDs are committed to disk, their situation
121 * is very similar to being pinned. Return XFS_ITEM_PINNED so that the caller
122 * will eventually flush the log. This should help in getting the CUI out of
127 struct xfs_log_item *lip,
128 struct list_head *buffer_list)
130 return XFS_ITEM_PINNED;
134 * The CUI has been either committed or aborted if the transaction has been
135 * cancelled. If the transaction was cancelled, an CUD isn't going to be
136 * constructed and thus we free the CUI here directly.
140 struct xfs_log_item *lip)
142 if (lip->li_flags & XFS_LI_ABORTED)
143 xfs_cui_item_free(CUI_ITEM(lip));
147 * The CUI is logged only once and cannot be moved in the log, so simply return
148 * the lsn at which it's been logged.
151 xfs_cui_item_committed(
152 struct xfs_log_item *lip,
159 * The CUI dependency tracking op doesn't do squat. It can't because
160 * it doesn't know where the free extent is coming from. The dependency
161 * tracking has to be handled by the "enclosing" metadata object. For
162 * example, for inodes, the inode is locked throughout the extent freeing
163 * so the dependency should be recorded there.
166 xfs_cui_item_committing(
167 struct xfs_log_item *lip,
173 * This is the ops vector shared by all cui log items.
175 static const struct xfs_item_ops xfs_cui_item_ops = {
176 .iop_size = xfs_cui_item_size,
177 .iop_format = xfs_cui_item_format,
178 .iop_pin = xfs_cui_item_pin,
179 .iop_unpin = xfs_cui_item_unpin,
180 .iop_unlock = xfs_cui_item_unlock,
181 .iop_committed = xfs_cui_item_committed,
182 .iop_push = xfs_cui_item_push,
183 .iop_committing = xfs_cui_item_committing,
187 * Allocate and initialize an cui item with the given number of extents.
189 struct xfs_cui_log_item *
191 struct xfs_mount *mp,
195 struct xfs_cui_log_item *cuip;
197 ASSERT(nextents > 0);
198 if (nextents > XFS_CUI_MAX_FAST_EXTENTS)
199 cuip = kmem_zalloc(xfs_cui_log_item_sizeof(nextents),
202 cuip = kmem_zone_zalloc(xfs_cui_zone, KM_SLEEP);
204 xfs_log_item_init(mp, &cuip->cui_item, XFS_LI_CUI, &xfs_cui_item_ops);
205 cuip->cui_format.cui_nextents = nextents;
206 cuip->cui_format.cui_id = (uintptr_t)(void *)cuip;
207 atomic_set(&cuip->cui_next_extent, 0);
208 atomic_set(&cuip->cui_refcount, 2);
214 * Freeing the CUI requires that we remove it from the AIL if it has already
215 * been placed there. However, the CUI may not yet have been placed in the AIL
216 * when called by xfs_cui_release() from CUD processing due to the ordering of
217 * committed vs unpin operations in bulk insert operations. Hence the reference
218 * count to ensure only the last caller frees the CUI.
222 struct xfs_cui_log_item *cuip)
224 ASSERT(atomic_read(&cuip->cui_refcount) > 0);
225 if (atomic_dec_and_test(&cuip->cui_refcount)) {
226 xfs_trans_ail_remove(&cuip->cui_item, SHUTDOWN_LOG_IO_ERROR);
227 xfs_cui_item_free(cuip);
231 static inline struct xfs_cud_log_item *CUD_ITEM(struct xfs_log_item *lip)
233 return container_of(lip, struct xfs_cud_log_item, cud_item);
238 struct xfs_log_item *lip,
243 *nbytes += sizeof(struct xfs_cud_log_format);
247 * This is called to fill in the vector of log iovecs for the
248 * given cud log item. We use only 1 iovec, and we point that
249 * at the cud_log_format structure embedded in the cud item.
250 * It is at this point that we assert that all of the extent
251 * slots in the cud item have been filled.
255 struct xfs_log_item *lip,
256 struct xfs_log_vec *lv)
258 struct xfs_cud_log_item *cudp = CUD_ITEM(lip);
259 struct xfs_log_iovec *vecp = NULL;
261 cudp->cud_format.cud_type = XFS_LI_CUD;
262 cudp->cud_format.cud_size = 1;
264 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_CUD_FORMAT, &cudp->cud_format,
265 sizeof(struct xfs_cud_log_format));
269 * Pinning has no meaning for an cud item, so just return.
273 struct xfs_log_item *lip)
278 * Since pinning has no meaning for an cud item, unpinning does
283 struct xfs_log_item *lip,
289 * There isn't much you can do to push on an cud item. It is simply stuck
290 * waiting for the log to be flushed to disk.
294 struct xfs_log_item *lip,
295 struct list_head *buffer_list)
297 return XFS_ITEM_PINNED;
301 * The CUD is either committed or aborted if the transaction is cancelled. If
302 * the transaction is cancelled, drop our reference to the CUI and free the
307 struct xfs_log_item *lip)
309 struct xfs_cud_log_item *cudp = CUD_ITEM(lip);
311 if (lip->li_flags & XFS_LI_ABORTED) {
312 xfs_cui_release(cudp->cud_cuip);
313 kmem_zone_free(xfs_cud_zone, cudp);
318 * When the cud item is committed to disk, all we need to do is delete our
319 * reference to our partner cui item and then free ourselves. Since we're
320 * freeing ourselves we must return -1 to keep the transaction code from
321 * further referencing this item.
324 xfs_cud_item_committed(
325 struct xfs_log_item *lip,
328 struct xfs_cud_log_item *cudp = CUD_ITEM(lip);
331 * Drop the CUI reference regardless of whether the CUD has been
332 * aborted. Once the CUD transaction is constructed, it is the sole
333 * responsibility of the CUD to release the CUI (even if the CUI is
334 * aborted due to log I/O error).
336 xfs_cui_release(cudp->cud_cuip);
337 kmem_zone_free(xfs_cud_zone, cudp);
339 return (xfs_lsn_t)-1;
343 * The CUD dependency tracking op doesn't do squat. It can't because
344 * it doesn't know where the free extent is coming from. The dependency
345 * tracking has to be handled by the "enclosing" metadata object. For
346 * example, for inodes, the inode is locked throughout the extent freeing
347 * so the dependency should be recorded there.
350 xfs_cud_item_committing(
351 struct xfs_log_item *lip,
357 * This is the ops vector shared by all cud log items.
359 static const struct xfs_item_ops xfs_cud_item_ops = {
360 .iop_size = xfs_cud_item_size,
361 .iop_format = xfs_cud_item_format,
362 .iop_pin = xfs_cud_item_pin,
363 .iop_unpin = xfs_cud_item_unpin,
364 .iop_unlock = xfs_cud_item_unlock,
365 .iop_committed = xfs_cud_item_committed,
366 .iop_push = xfs_cud_item_push,
367 .iop_committing = xfs_cud_item_committing,
371 * Allocate and initialize an cud item with the given number of extents.
373 struct xfs_cud_log_item *
375 struct xfs_mount *mp,
376 struct xfs_cui_log_item *cuip)
379 struct xfs_cud_log_item *cudp;
381 cudp = kmem_zone_zalloc(xfs_cud_zone, KM_SLEEP);
382 xfs_log_item_init(mp, &cudp->cud_item, XFS_LI_CUD, &xfs_cud_item_ops);
383 cudp->cud_cuip = cuip;
384 cudp->cud_format.cud_cui_id = cuip->cui_format.cui_id;
390 * Process a refcount update intent item that was recovered from the log.
391 * We need to update the refcountbt.
395 struct xfs_mount *mp,
396 struct xfs_cui_log_item *cuip)
400 unsigned int refc_type;
401 struct xfs_phys_extent *refc;
402 xfs_fsblock_t startblock_fsb;
404 struct xfs_cud_log_item *cudp;
405 struct xfs_trans *tp;
406 struct xfs_btree_cur *rcur = NULL;
407 enum xfs_refcount_intent_type type;
408 xfs_fsblock_t firstfsb;
409 xfs_fsblock_t new_fsb;
410 xfs_extlen_t new_len;
411 struct xfs_bmbt_irec irec;
412 struct xfs_defer_ops dfops;
413 bool requeue_only = false;
415 ASSERT(!test_bit(XFS_CUI_RECOVERED, &cuip->cui_flags));
418 * First check the validity of the extents described by the
419 * CUI. If any are bad, then assume that all are bad and
422 for (i = 0; i < cuip->cui_format.cui_nextents; i++) {
423 refc = &cuip->cui_format.cui_extents[i];
424 startblock_fsb = XFS_BB_TO_FSB(mp,
425 XFS_FSB_TO_DADDR(mp, refc->pe_startblock));
426 switch (refc->pe_flags & XFS_REFCOUNT_EXTENT_TYPE_MASK) {
427 case XFS_REFCOUNT_INCREASE:
428 case XFS_REFCOUNT_DECREASE:
429 case XFS_REFCOUNT_ALLOC_COW:
430 case XFS_REFCOUNT_FREE_COW:
437 if (!op_ok || startblock_fsb == 0 ||
439 startblock_fsb >= mp->m_sb.sb_dblocks ||
440 refc->pe_len >= mp->m_sb.sb_agblocks ||
441 (refc->pe_flags & ~XFS_REFCOUNT_EXTENT_FLAGS)) {
443 * This will pull the CUI from the AIL and
444 * free the memory associated with it.
446 set_bit(XFS_CUI_RECOVERED, &cuip->cui_flags);
447 xfs_cui_release(cuip);
453 * Under normal operation, refcount updates are deferred, so we
454 * wouldn't be adding them directly to a transaction. All
455 * refcount updates manage reservation usage internally and
456 * dynamically by deferring work that won't fit in the
457 * transaction. Normally, any work that needs to be deferred
458 * gets attached to the same defer_ops that scheduled the
459 * refcount update. However, we're in log recovery here, so we
460 * we create our own defer_ops and use that to finish up any
461 * work that doesn't fit.
463 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
466 cudp = xfs_trans_get_cud(tp, cuip);
468 xfs_defer_init(&dfops, &firstfsb);
469 for (i = 0; i < cuip->cui_format.cui_nextents; i++) {
470 refc = &cuip->cui_format.cui_extents[i];
471 refc_type = refc->pe_flags & XFS_REFCOUNT_EXTENT_TYPE_MASK;
473 case XFS_REFCOUNT_INCREASE:
474 case XFS_REFCOUNT_DECREASE:
475 case XFS_REFCOUNT_ALLOC_COW:
476 case XFS_REFCOUNT_FREE_COW:
480 error = -EFSCORRUPTED;
484 new_fsb = refc->pe_startblock;
485 new_len = refc->pe_len;
487 error = xfs_trans_log_finish_refcount_update(tp, cudp,
488 &dfops, type, refc->pe_startblock, refc->pe_len,
489 &new_fsb, &new_len, &rcur);
493 /* Requeue what we didn't finish. */
495 irec.br_startblock = new_fsb;
496 irec.br_blockcount = new_len;
498 case XFS_REFCOUNT_INCREASE:
499 error = xfs_refcount_increase_extent(
500 tp->t_mountp, &dfops, &irec);
502 case XFS_REFCOUNT_DECREASE:
503 error = xfs_refcount_decrease_extent(
504 tp->t_mountp, &dfops, &irec);
506 case XFS_REFCOUNT_ALLOC_COW:
507 error = xfs_refcount_alloc_cow_extent(
508 tp->t_mountp, &dfops,
512 case XFS_REFCOUNT_FREE_COW:
513 error = xfs_refcount_free_cow_extent(
514 tp->t_mountp, &dfops,
527 xfs_refcount_finish_one_cleanup(tp, rcur, error);
528 error = xfs_defer_finish(&tp, &dfops, NULL);
531 set_bit(XFS_CUI_RECOVERED, &cuip->cui_flags);
532 error = xfs_trans_commit(tp);
536 xfs_refcount_finish_one_cleanup(tp, rcur, error);
538 xfs_defer_cancel(&dfops);
539 xfs_trans_cancel(tp);