1/*
  2 * Copyright (C) 2016 Oracle.  All Rights Reserved.
  3 *
  4 * Author: Darrick J. Wong <darrick.wong@oracle.com>
  5 *
  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.
 10 *
 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.
 15 *
 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.
 19 */
 20#include "xfs.h"
 21#include "xfs_fs.h"
 22#include "xfs_format.h"
 23#include "xfs_log_format.h"
 24#include "xfs_trans_resv.h"
 25#include "xfs_bit.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"
 32#include "xfs_log.h"
 33#include "xfs_refcount.h"
 34
 35
 36kmem_zone_t	*xfs_cui_zone;
 37kmem_zone_t	*xfs_cud_zone;
 38
 39static inline struct xfs_cui_log_item *CUI_ITEM(struct xfs_log_item *lip)
 40{
 41	return container_of(lip, struct xfs_cui_log_item, cui_item);
 42}
 43
 44void
 45xfs_cui_item_free(
 46	struct xfs_cui_log_item	*cuip)
 47{
 48	if (cuip->cui_format.cui_nextents > XFS_CUI_MAX_FAST_EXTENTS)
 49		kmem_free(cuip);
 50	else
 51		kmem_zone_free(xfs_cui_zone, cuip);
 52}
 53
 54STATIC void
 55xfs_cui_item_size(
 56	struct xfs_log_item	*lip,
 57	int			*nvecs,
 58	int			*nbytes)
 59{
 60	struct xfs_cui_log_item	*cuip = CUI_ITEM(lip);
 61
 62	*nvecs += 1;
 63	*nbytes += xfs_cui_log_format_sizeof(cuip->cui_format.cui_nextents);
 64}
 65
 66/*
 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.
 72 */
 73STATIC void
 74xfs_cui_item_format(
 75	struct xfs_log_item	*lip,
 76	struct xfs_log_vec	*lv)
 77{
 78	struct xfs_cui_log_item	*cuip = CUI_ITEM(lip);
 79	struct xfs_log_iovec	*vecp = NULL;
 80
 81	ASSERT(atomic_read(&cuip->cui_next_extent) ==
 82			cuip->cui_format.cui_nextents);
 83
 84	cuip->cui_format.cui_type = XFS_LI_CUI;
 85	cuip->cui_format.cui_size = 1;
 86
 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));
 89}
 90
 91/*
 92 * Pinning has no meaning for an cui item, so just return.
 93 */
 94STATIC void
 95xfs_cui_item_pin(
 96	struct xfs_log_item	*lip)
 97{
 98}
 99
100/*
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.
107 */
108STATIC void
109xfs_cui_item_unpin(
110	struct xfs_log_item	*lip,
111	int			remove)
112{
113	struct xfs_cui_log_item	*cuip = CUI_ITEM(lip);
114
115	xfs_cui_release(cuip);
116}
117
118/*
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
123 * the AIL.
124 */
125STATIC uint
126xfs_cui_item_push(
127	struct xfs_log_item	*lip,
128	struct list_head	*buffer_list)
129{
130	return XFS_ITEM_PINNED;
131}
132
133/*
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.
137 */
138STATIC void
139xfs_cui_item_unlock(
140	struct xfs_log_item	*lip)
141{
142	if (lip->li_flags & XFS_LI_ABORTED)
143		xfs_cui_item_free(CUI_ITEM(lip));
144}
145
146/*
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.
149 */
150STATIC xfs_lsn_t
151xfs_cui_item_committed(
152	struct xfs_log_item	*lip,
153	xfs_lsn_t		lsn)
154{
155	return lsn;
156}
157
158/*
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.
164 */
165STATIC void
166xfs_cui_item_committing(
167	struct xfs_log_item	*lip,
168	xfs_lsn_t		lsn)
169{
170}
171
172/*
173 * This is the ops vector shared by all cui log items.
174 */
175static 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,
184};
185
186/*
187 * Allocate and initialize an cui item with the given number of extents.
188 */
189struct xfs_cui_log_item *
190xfs_cui_init(
191	struct xfs_mount		*mp,
192	uint				nextents)
193
194{
195	struct xfs_cui_log_item		*cuip;
196
197	ASSERT(nextents > 0);
198	if (nextents > XFS_CUI_MAX_FAST_EXTENTS)
199		cuip = kmem_zalloc(xfs_cui_log_item_sizeof(nextents),
200				KM_SLEEP);
201	else
202		cuip = kmem_zone_zalloc(xfs_cui_zone, KM_SLEEP);
203
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);
209
210	return cuip;
211}
212
213/*
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.
219 */
220void
221xfs_cui_release(
222	struct xfs_cui_log_item	*cuip)
223{
224	if (atomic_dec_and_test(&cuip->cui_refcount)) {
225		xfs_trans_ail_remove(&cuip->cui_item, SHUTDOWN_LOG_IO_ERROR);
226		xfs_cui_item_free(cuip);
227	}
228}
229
230static inline struct xfs_cud_log_item *CUD_ITEM(struct xfs_log_item *lip)
231{
232	return container_of(lip, struct xfs_cud_log_item, cud_item);
233}
234
235STATIC void
236xfs_cud_item_size(
237	struct xfs_log_item	*lip,
238	int			*nvecs,
239	int			*nbytes)
240{
241	*nvecs += 1;
242	*nbytes += sizeof(struct xfs_cud_log_format);
243}
244
245/*
246 * This is called to fill in the vector of log iovecs for the
247 * given cud log item. We use only 1 iovec, and we point that
248 * at the cud_log_format structure embedded in the cud item.
249 * It is at this point that we assert that all of the extent
250 * slots in the cud item have been filled.
251 */
252STATIC void
253xfs_cud_item_format(
254	struct xfs_log_item	*lip,
255	struct xfs_log_vec	*lv)
256{
257	struct xfs_cud_log_item	*cudp = CUD_ITEM(lip);
258	struct xfs_log_iovec	*vecp = NULL;
259
260	cudp->cud_format.cud_type = XFS_LI_CUD;
261	cudp->cud_format.cud_size = 1;
262
263	xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_CUD_FORMAT, &cudp->cud_format,
264			sizeof(struct xfs_cud_log_format));
265}
266
267/*
268 * Pinning has no meaning for an cud item, so just return.
269 */
270STATIC void
271xfs_cud_item_pin(
272	struct xfs_log_item	*lip)
273{
274}
275
276/*
277 * Since pinning has no meaning for an cud item, unpinning does
278 * not either.
279 */
280STATIC void
281xfs_cud_item_unpin(
282	struct xfs_log_item	*lip,
283	int			remove)
284{
285}
286
287/*
288 * There isn't much you can do to push on an cud item.  It is simply stuck
289 * waiting for the log to be flushed to disk.
290 */
291STATIC uint
292xfs_cud_item_push(
293	struct xfs_log_item	*lip,
294	struct list_head	*buffer_list)
295{
296	return XFS_ITEM_PINNED;
297}
298
299/*
300 * The CUD is either committed or aborted if the transaction is cancelled. If
301 * the transaction is cancelled, drop our reference to the CUI and free the
302 * CUD.
303 */
304STATIC void
305xfs_cud_item_unlock(
306	struct xfs_log_item	*lip)
307{
308	struct xfs_cud_log_item	*cudp = CUD_ITEM(lip);
309
310	if (lip->li_flags & XFS_LI_ABORTED) {
311		xfs_cui_release(cudp->cud_cuip);
312		kmem_zone_free(xfs_cud_zone, cudp);
313	}
314}
315
316/*
317 * When the cud item is committed to disk, all we need to do is delete our
318 * reference to our partner cui item and then free ourselves. Since we're
319 * freeing ourselves we must return -1 to keep the transaction code from
320 * further referencing this item.
321 */
322STATIC xfs_lsn_t
323xfs_cud_item_committed(
324	struct xfs_log_item	*lip,
325	xfs_lsn_t		lsn)
326{
327	struct xfs_cud_log_item	*cudp = CUD_ITEM(lip);
328
329	/*
330	 * Drop the CUI reference regardless of whether the CUD has been
331	 * aborted. Once the CUD transaction is constructed, it is the sole
332	 * responsibility of the CUD to release the CUI (even if the CUI is
333	 * aborted due to log I/O error).
334	 */
335	xfs_cui_release(cudp->cud_cuip);
336	kmem_zone_free(xfs_cud_zone, cudp);
337
338	return (xfs_lsn_t)-1;
339}
340
341/*
342 * The CUD dependency tracking op doesn't do squat.  It can't because
343 * it doesn't know where the free extent is coming from.  The dependency
344 * tracking has to be handled by the "enclosing" metadata object.  For
345 * example, for inodes, the inode is locked throughout the extent freeing
346 * so the dependency should be recorded there.
347 */
348STATIC void
349xfs_cud_item_committing(
350	struct xfs_log_item	*lip,
351	xfs_lsn_t		lsn)
352{
353}
354
355/*
356 * This is the ops vector shared by all cud log items.
357 */
358static const struct xfs_item_ops xfs_cud_item_ops = {
359	.iop_size	= xfs_cud_item_size,
360	.iop_format	= xfs_cud_item_format,
361	.iop_pin	= xfs_cud_item_pin,
362	.iop_unpin	= xfs_cud_item_unpin,
363	.iop_unlock	= xfs_cud_item_unlock,
364	.iop_committed	= xfs_cud_item_committed,
365	.iop_push	= xfs_cud_item_push,
366	.iop_committing = xfs_cud_item_committing,
367};
368
369/*
370 * Allocate and initialize an cud item with the given number of extents.
371 */
372struct xfs_cud_log_item *
373xfs_cud_init(
374	struct xfs_mount		*mp,
375	struct xfs_cui_log_item		*cuip)
376
377{
378	struct xfs_cud_log_item	*cudp;
379
380	cudp = kmem_zone_zalloc(xfs_cud_zone, KM_SLEEP);
381	xfs_log_item_init(mp, &cudp->cud_item, XFS_LI_CUD, &xfs_cud_item_ops);
382	cudp->cud_cuip = cuip;
383	cudp->cud_format.cud_cui_id = cuip->cui_format.cui_id;
384
385	return cudp;
386}
387
388/*
389 * Process a refcount update intent item that was recovered from the log.
390 * We need to update the refcountbt.
391 */
392int
393xfs_cui_recover(
394	struct xfs_mount		*mp,
395	struct xfs_cui_log_item		*cuip)
396{
397	int				i;
398	int				error = 0;
399	unsigned int			refc_type;
400	struct xfs_phys_extent		*refc;
401	xfs_fsblock_t			startblock_fsb;
402	bool				op_ok;
403	struct xfs_cud_log_item		*cudp;
404	struct xfs_trans		*tp;
405	struct xfs_btree_cur		*rcur = NULL;
406	enum xfs_refcount_intent_type	type;
407	xfs_fsblock_t			firstfsb;
408	xfs_fsblock_t			new_fsb;
409	xfs_extlen_t			new_len;
410	struct xfs_bmbt_irec		irec;
411	struct xfs_defer_ops		dfops;
412	bool				requeue_only = false;
413
414	ASSERT(!test_bit(XFS_CUI_RECOVERED, &cuip->cui_flags));
415
416	/*
417	 * First check the validity of the extents described by the
418	 * CUI.  If any are bad, then assume that all are bad and
419	 * just toss the CUI.
420	 */
421	for (i = 0; i < cuip->cui_format.cui_nextents; i++) {
422		refc = &cuip->cui_format.cui_extents[i];
423		startblock_fsb = XFS_BB_TO_FSB(mp,
424				   XFS_FSB_TO_DADDR(mp, refc->pe_startblock));
425		switch (refc->pe_flags & XFS_REFCOUNT_EXTENT_TYPE_MASK) {
426		case XFS_REFCOUNT_INCREASE:
427		case XFS_REFCOUNT_DECREASE:
428		case XFS_REFCOUNT_ALLOC_COW:
429		case XFS_REFCOUNT_FREE_COW:
430			op_ok = true;
431			break;
432		default:
433			op_ok = false;
434			break;
435		}
436		if (!op_ok || startblock_fsb == 0 ||
437		    refc->pe_len == 0 ||
438		    startblock_fsb >= mp->m_sb.sb_dblocks ||
439		    refc->pe_len >= mp->m_sb.sb_agblocks ||
440		    (refc->pe_flags & ~XFS_REFCOUNT_EXTENT_FLAGS)) {
441			/*
442			 * This will pull the CUI from the AIL and
443			 * free the memory associated with it.
444			 */
445			set_bit(XFS_CUI_RECOVERED, &cuip->cui_flags);
446			xfs_cui_release(cuip);
447			return -EIO;
448		}
449	}
450
451	/*
452	 * Under normal operation, refcount updates are deferred, so we
453	 * wouldn't be adding them directly to a transaction.  All
454	 * refcount updates manage reservation usage internally and
455	 * dynamically by deferring work that won't fit in the
456	 * transaction.  Normally, any work that needs to be deferred
457	 * gets attached to the same defer_ops that scheduled the
458	 * refcount update.  However, we're in log recovery here, so we
459	 * we create our own defer_ops and use that to finish up any
460	 * work that doesn't fit.
461	 */
462	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
463	if (error)
464		return error;
465	cudp = xfs_trans_get_cud(tp, cuip);
466
467	xfs_defer_init(&dfops, &firstfsb);
468	for (i = 0; i < cuip->cui_format.cui_nextents; i++) {
469		refc = &cuip->cui_format.cui_extents[i];
470		refc_type = refc->pe_flags & XFS_REFCOUNT_EXTENT_TYPE_MASK;
471		switch (refc_type) {
472		case XFS_REFCOUNT_INCREASE:
473		case XFS_REFCOUNT_DECREASE:
474		case XFS_REFCOUNT_ALLOC_COW:
475		case XFS_REFCOUNT_FREE_COW:
476			type = refc_type;
477			break;
478		default:
479			error = -EFSCORRUPTED;
480			goto abort_error;
481		}
482		if (requeue_only) {
483			new_fsb = refc->pe_startblock;
484			new_len = refc->pe_len;
485		} else
486			error = xfs_trans_log_finish_refcount_update(tp, cudp,
487				&dfops, type, refc->pe_startblock, refc->pe_len,
488				&new_fsb, &new_len, &rcur);
489		if (error)
490			goto abort_error;
491
492		/* Requeue what we didn't finish. */
493		if (new_len > 0) {
494			irec.br_startblock = new_fsb;
495			irec.br_blockcount = new_len;
496			switch (type) {
497			case XFS_REFCOUNT_INCREASE:
498				error = xfs_refcount_increase_extent(
499						tp->t_mountp, &dfops, &irec);
500				break;
501			case XFS_REFCOUNT_DECREASE:
502				error = xfs_refcount_decrease_extent(
503						tp->t_mountp, &dfops, &irec);
504				break;
505			case XFS_REFCOUNT_ALLOC_COW:
506				error = xfs_refcount_alloc_cow_extent(
507						tp->t_mountp, &dfops,
508						irec.br_startblock,
509						irec.br_blockcount);
510				break;
511			case XFS_REFCOUNT_FREE_COW:
512				error = xfs_refcount_free_cow_extent(
513						tp->t_mountp, &dfops,
514						irec.br_startblock,
515						irec.br_blockcount);
516				break;
517			default:
518				ASSERT(0);
519			}
520			if (error)
521				goto abort_error;
522			requeue_only = true;
523		}
524	}
525
526	xfs_refcount_finish_one_cleanup(tp, rcur, error);
527	error = xfs_defer_finish(&tp, &dfops, NULL);
528	if (error)
529		goto abort_defer;
530	set_bit(XFS_CUI_RECOVERED, &cuip->cui_flags);
531	error = xfs_trans_commit(tp);
532	return error;
533
534abort_error:
535	xfs_refcount_finish_one_cleanup(tp, rcur, error);
536abort_defer:
537	xfs_defer_cancel(&dfops);
538	xfs_trans_cancel(tp);
539	return error;
540}