1// SPDX-License-Identifier: GPL-2.0+
  2/*
  3 * Copyright (C) 2016 Oracle.  All Rights Reserved.
  4 * Author: Darrick J. Wong <darrick.wong@oracle.com>
  5 */
  6#include "xfs.h"
  7#include "xfs_fs.h"
  8#include "xfs_format.h"
  9#include "xfs_log_format.h"
 10#include "xfs_trans_resv.h"
 11#include "xfs_bit.h"
 12#include "xfs_shared.h"
 13#include "xfs_mount.h"
 14#include "xfs_defer.h"
 15#include "xfs_inode.h"
 16#include "xfs_trans.h"
 17#include "xfs_trans_priv.h"
 18#include "xfs_bmap_item.h"
 19#include "xfs_log.h"
 20#include "xfs_bmap.h"
 21#include "xfs_icache.h"
 22#include "xfs_bmap_btree.h"
 23#include "xfs_trans_space.h"
 24
 25
 26kmem_zone_t	*xfs_bui_zone;
 27kmem_zone_t	*xfs_bud_zone;
 28
 29static inline struct xfs_bui_log_item *BUI_ITEM(struct xfs_log_item *lip)
 30{
 31	return container_of(lip, struct xfs_bui_log_item, bui_item);
 32}
 33
 34void
 35xfs_bui_item_free(
 36	struct xfs_bui_log_item	*buip)
 37{
 38	kmem_zone_free(xfs_bui_zone, buip);
 39}
 40
 41/*
 42 * Freeing the BUI requires that we remove it from the AIL if it has already
 43 * been placed there. However, the BUI may not yet have been placed in the AIL
 44 * when called by xfs_bui_release() from BUD processing due to the ordering of
 45 * committed vs unpin operations in bulk insert operations. Hence the reference
 46 * count to ensure only the last caller frees the BUI.
 47 */
 48void
 49xfs_bui_release(
 50	struct xfs_bui_log_item	*buip)
 51{
 52	ASSERT(atomic_read(&buip->bui_refcount) > 0);
 53	if (atomic_dec_and_test(&buip->bui_refcount)) {
 54		xfs_trans_ail_remove(&buip->bui_item, SHUTDOWN_LOG_IO_ERROR);
 55		xfs_bui_item_free(buip);
 56	}
 57}
 58
 59
 60STATIC void
 61xfs_bui_item_size(
 62	struct xfs_log_item	*lip,
 63	int			*nvecs,
 64	int			*nbytes)
 65{
 66	struct xfs_bui_log_item	*buip = BUI_ITEM(lip);
 67
 68	*nvecs += 1;
 69	*nbytes += xfs_bui_log_format_sizeof(buip->bui_format.bui_nextents);
 70}
 71
 72/*
 73 * This is called to fill in the vector of log iovecs for the
 74 * given bui log item. We use only 1 iovec, and we point that
 75 * at the bui_log_format structure embedded in the bui item.
 76 * It is at this point that we assert that all of the extent
 77 * slots in the bui item have been filled.
 78 */
 79STATIC void
 80xfs_bui_item_format(
 81	struct xfs_log_item	*lip,
 82	struct xfs_log_vec	*lv)
 83{
 84	struct xfs_bui_log_item	*buip = BUI_ITEM(lip);
 85	struct xfs_log_iovec	*vecp = NULL;
 86
 87	ASSERT(atomic_read(&buip->bui_next_extent) ==
 88			buip->bui_format.bui_nextents);
 89
 90	buip->bui_format.bui_type = XFS_LI_BUI;
 91	buip->bui_format.bui_size = 1;
 92
 93	xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_BUI_FORMAT, &buip->bui_format,
 94			xfs_bui_log_format_sizeof(buip->bui_format.bui_nextents));
 95}
 96
 97/*
 98 * The unpin operation is the last place an BUI is manipulated in the log. It is
 99 * either inserted in the AIL or aborted in the event of a log I/O error. In
100 * either case, the BUI transaction has been successfully committed to make it
101 * this far. Therefore, we expect whoever committed the BUI to either construct
102 * and commit the BUD or drop the BUD's reference in the event of error. Simply
103 * drop the log's BUI reference now that the log is done with it.
104 */
105STATIC void
106xfs_bui_item_unpin(
107	struct xfs_log_item	*lip,
108	int			remove)
109{
110	struct xfs_bui_log_item	*buip = BUI_ITEM(lip);
111
112	xfs_bui_release(buip);
113}
114
115/*
116 * The BUI has been either committed or aborted if the transaction has been
117 * cancelled. If the transaction was cancelled, an BUD isn't going to be
118 * constructed and thus we free the BUI here directly.
119 */
120STATIC void
121xfs_bui_item_release(
122	struct xfs_log_item	*lip)
123{
124	xfs_bui_release(BUI_ITEM(lip));
125}
126
127static const struct xfs_item_ops xfs_bui_item_ops = {
128	.iop_size	= xfs_bui_item_size,
129	.iop_format	= xfs_bui_item_format,
130	.iop_unpin	= xfs_bui_item_unpin,
131	.iop_release	= xfs_bui_item_release,
132};
133
134/*
135 * Allocate and initialize an bui item with the given number of extents.
136 */
137struct xfs_bui_log_item *
138xfs_bui_init(
139	struct xfs_mount		*mp)
140
141{
142	struct xfs_bui_log_item		*buip;
143
144	buip = kmem_zone_zalloc(xfs_bui_zone, 0);
145
146	xfs_log_item_init(mp, &buip->bui_item, XFS_LI_BUI, &xfs_bui_item_ops);
147	buip->bui_format.bui_nextents = XFS_BUI_MAX_FAST_EXTENTS;
148	buip->bui_format.bui_id = (uintptr_t)(void *)buip;
149	atomic_set(&buip->bui_next_extent, 0);
150	atomic_set(&buip->bui_refcount, 2);
151
152	return buip;
153}
154
155static inline struct xfs_bud_log_item *BUD_ITEM(struct xfs_log_item *lip)
156{
157	return container_of(lip, struct xfs_bud_log_item, bud_item);
158}
159
160STATIC void
161xfs_bud_item_size(
162	struct xfs_log_item	*lip,
163	int			*nvecs,
164	int			*nbytes)
165{
166	*nvecs += 1;
167	*nbytes += sizeof(struct xfs_bud_log_format);
168}
169
170/*
171 * This is called to fill in the vector of log iovecs for the
172 * given bud log item. We use only 1 iovec, and we point that
173 * at the bud_log_format structure embedded in the bud item.
174 * It is at this point that we assert that all of the extent
175 * slots in the bud item have been filled.
176 */
177STATIC void
178xfs_bud_item_format(
179	struct xfs_log_item	*lip,
180	struct xfs_log_vec	*lv)
181{
182	struct xfs_bud_log_item	*budp = BUD_ITEM(lip);
183	struct xfs_log_iovec	*vecp = NULL;
184
185	budp->bud_format.bud_type = XFS_LI_BUD;
186	budp->bud_format.bud_size = 1;
187
188	xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_BUD_FORMAT, &budp->bud_format,
189			sizeof(struct xfs_bud_log_format));
190}
191
192/*
193 * The BUD is either committed or aborted if the transaction is cancelled. If
194 * the transaction is cancelled, drop our reference to the BUI and free the
195 * BUD.
196 */
197STATIC void
198xfs_bud_item_release(
199	struct xfs_log_item	*lip)
200{
201	struct xfs_bud_log_item	*budp = BUD_ITEM(lip);
202
203	xfs_bui_release(budp->bud_buip);
204	kmem_zone_free(xfs_bud_zone, budp);
205}
206
207static const struct xfs_item_ops xfs_bud_item_ops = {
208	.flags		= XFS_ITEM_RELEASE_WHEN_COMMITTED,
209	.iop_size	= xfs_bud_item_size,
210	.iop_format	= xfs_bud_item_format,
211	.iop_release	= xfs_bud_item_release,
212};
213
214static struct xfs_bud_log_item *
215xfs_trans_get_bud(
216	struct xfs_trans		*tp,
217	struct xfs_bui_log_item		*buip)
218{
219	struct xfs_bud_log_item		*budp;
220
221	budp = kmem_zone_zalloc(xfs_bud_zone, 0);
222	xfs_log_item_init(tp->t_mountp, &budp->bud_item, XFS_LI_BUD,
223			  &xfs_bud_item_ops);
224	budp->bud_buip = buip;
225	budp->bud_format.bud_bui_id = buip->bui_format.bui_id;
226
227	xfs_trans_add_item(tp, &budp->bud_item);
228	return budp;
229}
230
231/*
232 * Finish an bmap update and log it to the BUD. Note that the
233 * transaction is marked dirty regardless of whether the bmap update
234 * succeeds or fails to support the BUI/BUD lifecycle rules.
235 */
236static int
237xfs_trans_log_finish_bmap_update(
238	struct xfs_trans		*tp,
239	struct xfs_bud_log_item		*budp,
240	enum xfs_bmap_intent_type	type,
241	struct xfs_inode		*ip,
242	int				whichfork,
243	xfs_fileoff_t			startoff,
244	xfs_fsblock_t			startblock,
245	xfs_filblks_t			*blockcount,
246	xfs_exntst_t			state)
247{
248	int				error;
249
250	error = xfs_bmap_finish_one(tp, ip, type, whichfork, startoff,
251			startblock, blockcount, state);
252
253	/*
254	 * Mark the transaction dirty, even on error. This ensures the
255	 * transaction is aborted, which:
256	 *
257	 * 1.) releases the BUI and frees the BUD
258	 * 2.) shuts down the filesystem
259	 */
260	tp->t_flags |= XFS_TRANS_DIRTY;
261	set_bit(XFS_LI_DIRTY, &budp->bud_item.li_flags);
262
263	return error;
264}
265
266/* Sort bmap intents by inode. */
267static int
268xfs_bmap_update_diff_items(
269	void				*priv,
270	struct list_head		*a,
271	struct list_head		*b)
272{
273	struct xfs_bmap_intent		*ba;
274	struct xfs_bmap_intent		*bb;
275
276	ba = container_of(a, struct xfs_bmap_intent, bi_list);
277	bb = container_of(b, struct xfs_bmap_intent, bi_list);
278	return ba->bi_owner->i_ino - bb->bi_owner->i_ino;
279}
280
281/* Get an BUI. */
282STATIC void *
283xfs_bmap_update_create_intent(
284	struct xfs_trans		*tp,
285	unsigned int			count)
286{
287	struct xfs_bui_log_item		*buip;
288
289	ASSERT(count == XFS_BUI_MAX_FAST_EXTENTS);
290	ASSERT(tp != NULL);
291
292	buip = xfs_bui_init(tp->t_mountp);
293	ASSERT(buip != NULL);
294
295	/*
296	 * Get a log_item_desc to point at the new item.
297	 */
298	xfs_trans_add_item(tp, &buip->bui_item);
299	return buip;
300}
301
302/* Set the map extent flags for this mapping. */
303static void
304xfs_trans_set_bmap_flags(
305	struct xfs_map_extent		*bmap,
306	enum xfs_bmap_intent_type	type,
307	int				whichfork,
308	xfs_exntst_t			state)
309{
310	bmap->me_flags = 0;
311	switch (type) {
312	case XFS_BMAP_MAP:
313	case XFS_BMAP_UNMAP:
314		bmap->me_flags = type;
315		break;
316	default:
317		ASSERT(0);
318	}
319	if (state == XFS_EXT_UNWRITTEN)
320		bmap->me_flags |= XFS_BMAP_EXTENT_UNWRITTEN;
321	if (whichfork == XFS_ATTR_FORK)
322		bmap->me_flags |= XFS_BMAP_EXTENT_ATTR_FORK;
323}
324
325/* Log bmap updates in the intent item. */
326STATIC void
327xfs_bmap_update_log_item(
328	struct xfs_trans		*tp,
329	void				*intent,
330	struct list_head		*item)
331{
332	struct xfs_bui_log_item		*buip = intent;
333	struct xfs_bmap_intent		*bmap;
334	uint				next_extent;
335	struct xfs_map_extent		*map;
336
337	bmap = container_of(item, struct xfs_bmap_intent, bi_list);
338
339	tp->t_flags |= XFS_TRANS_DIRTY;
340	set_bit(XFS_LI_DIRTY, &buip->bui_item.li_flags);
341
342	/*
343	 * atomic_inc_return gives us the value after the increment;
344	 * we want to use it as an array index so we need to subtract 1 from
345	 * it.
346	 */
347	next_extent = atomic_inc_return(&buip->bui_next_extent) - 1;
348	ASSERT(next_extent < buip->bui_format.bui_nextents);
349	map = &buip->bui_format.bui_extents[next_extent];
350	map->me_owner = bmap->bi_owner->i_ino;
351	map->me_startblock = bmap->bi_bmap.br_startblock;
352	map->me_startoff = bmap->bi_bmap.br_startoff;
353	map->me_len = bmap->bi_bmap.br_blockcount;
354	xfs_trans_set_bmap_flags(map, bmap->bi_type, bmap->bi_whichfork,
355			bmap->bi_bmap.br_state);
356}
357
358/* Get an BUD so we can process all the deferred rmap updates. */
359STATIC void *
360xfs_bmap_update_create_done(
361	struct xfs_trans		*tp,
362	void				*intent,
363	unsigned int			count)
364{
365	return xfs_trans_get_bud(tp, intent);
366}
367
368/* Process a deferred rmap update. */
369STATIC int
370xfs_bmap_update_finish_item(
371	struct xfs_trans		*tp,
372	struct list_head		*item,
373	void				*done_item,
374	void				**state)
375{
376	struct xfs_bmap_intent		*bmap;
377	xfs_filblks_t			count;
378	int				error;
379
380	bmap = container_of(item, struct xfs_bmap_intent, bi_list);
381	count = bmap->bi_bmap.br_blockcount;
382	error = xfs_trans_log_finish_bmap_update(tp, done_item,
383			bmap->bi_type,
384			bmap->bi_owner, bmap->bi_whichfork,
385			bmap->bi_bmap.br_startoff,
386			bmap->bi_bmap.br_startblock,
387			&count,
388			bmap->bi_bmap.br_state);
389	if (!error && count > 0) {
390		ASSERT(bmap->bi_type == XFS_BMAP_UNMAP);
391		bmap->bi_bmap.br_blockcount = count;
392		return -EAGAIN;
393	}
394	kmem_free(bmap);
395	return error;
396}
397
398/* Abort all pending BUIs. */
399STATIC void
400xfs_bmap_update_abort_intent(
401	void				*intent)
402{
403	xfs_bui_release(intent);
404}
405
406/* Cancel a deferred rmap update. */
407STATIC void
408xfs_bmap_update_cancel_item(
409	struct list_head		*item)
410{
411	struct xfs_bmap_intent		*bmap;
412
413	bmap = container_of(item, struct xfs_bmap_intent, bi_list);
414	kmem_free(bmap);
415}
416
417const struct xfs_defer_op_type xfs_bmap_update_defer_type = {
418	.max_items	= XFS_BUI_MAX_FAST_EXTENTS,
419	.diff_items	= xfs_bmap_update_diff_items,
420	.create_intent	= xfs_bmap_update_create_intent,
421	.abort_intent	= xfs_bmap_update_abort_intent,
422	.log_item	= xfs_bmap_update_log_item,
423	.create_done	= xfs_bmap_update_create_done,
424	.finish_item	= xfs_bmap_update_finish_item,
425	.cancel_item	= xfs_bmap_update_cancel_item,
426};
427
428/*
429 * Process a bmap update intent item that was recovered from the log.
430 * We need to update some inode's bmbt.
431 */
432int
433xfs_bui_recover(
434	struct xfs_trans		*parent_tp,
435	struct xfs_bui_log_item		*buip)
436{
437	int				error = 0;
438	unsigned int			bui_type;
439	struct xfs_map_extent		*bmap;
440	xfs_fsblock_t			startblock_fsb;
441	xfs_fsblock_t			inode_fsb;
442	xfs_filblks_t			count;
443	bool				op_ok;
444	struct xfs_bud_log_item		*budp;
445	enum xfs_bmap_intent_type	type;
446	int				whichfork;
447	xfs_exntst_t			state;
448	struct xfs_trans		*tp;
449	struct xfs_inode		*ip = NULL;
450	struct xfs_bmbt_irec		irec;
451	struct xfs_mount		*mp = parent_tp->t_mountp;
452
453	ASSERT(!test_bit(XFS_BUI_RECOVERED, &buip->bui_flags));
454
455	/* Only one mapping operation per BUI... */
456	if (buip->bui_format.bui_nextents != XFS_BUI_MAX_FAST_EXTENTS) {
457		set_bit(XFS_BUI_RECOVERED, &buip->bui_flags);
458		xfs_bui_release(buip);
459		return -EIO;
460	}
461
462	/*
463	 * First check the validity of the extent described by the
464	 * BUI.  If anything is bad, then toss the BUI.
465	 */
466	bmap = &buip->bui_format.bui_extents[0];
467	startblock_fsb = XFS_BB_TO_FSB(mp,
468			   XFS_FSB_TO_DADDR(mp, bmap->me_startblock));
469	inode_fsb = XFS_BB_TO_FSB(mp, XFS_FSB_TO_DADDR(mp,
470			XFS_INO_TO_FSB(mp, bmap->me_owner)));
471	switch (bmap->me_flags & XFS_BMAP_EXTENT_TYPE_MASK) {
472	case XFS_BMAP_MAP:
473	case XFS_BMAP_UNMAP:
474		op_ok = true;
475		break;
476	default:
477		op_ok = false;
478		break;
479	}
480	if (!op_ok || startblock_fsb == 0 ||
481	    bmap->me_len == 0 ||
482	    inode_fsb == 0 ||
483	    startblock_fsb >= mp->m_sb.sb_dblocks ||
484	    bmap->me_len >= mp->m_sb.sb_agblocks ||
485	    inode_fsb >= mp->m_sb.sb_dblocks ||
486	    (bmap->me_flags & ~XFS_BMAP_EXTENT_FLAGS)) {
487		/*
488		 * This will pull the BUI from the AIL and
489		 * free the memory associated with it.
490		 */
491		set_bit(XFS_BUI_RECOVERED, &buip->bui_flags);
492		xfs_bui_release(buip);
493		return -EIO;
494	}
495
496	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate,
497			XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK), 0, 0, &tp);
498	if (error)
499		return error;
500	/*
501	 * Recovery stashes all deferred ops during intent processing and
502	 * finishes them on completion. Transfer current dfops state to this
503	 * transaction and transfer the result back before we return.
504	 */
505	xfs_defer_move(tp, parent_tp);
506	budp = xfs_trans_get_bud(tp, buip);
507
508	/* Grab the inode. */
509	error = xfs_iget(mp, tp, bmap->me_owner, 0, XFS_ILOCK_EXCL, &ip);
510	if (error)
511		goto err_inode;
512
513	if (VFS_I(ip)->i_nlink == 0)
514		xfs_iflags_set(ip, XFS_IRECOVERY);
515
516	/* Process deferred bmap item. */
517	state = (bmap->me_flags & XFS_BMAP_EXTENT_UNWRITTEN) ?
518			XFS_EXT_UNWRITTEN : XFS_EXT_NORM;
519	whichfork = (bmap->me_flags & XFS_BMAP_EXTENT_ATTR_FORK) ?
520			XFS_ATTR_FORK : XFS_DATA_FORK;
521	bui_type = bmap->me_flags & XFS_BMAP_EXTENT_TYPE_MASK;
522	switch (bui_type) {
523	case XFS_BMAP_MAP:
524	case XFS_BMAP_UNMAP:
525		type = bui_type;
526		break;
527	default:
528		error = -EFSCORRUPTED;
529		goto err_inode;
530	}
531	xfs_trans_ijoin(tp, ip, 0);
532
533	count = bmap->me_len;
534	error = xfs_trans_log_finish_bmap_update(tp, budp, type, ip, whichfork,
535			bmap->me_startoff, bmap->me_startblock, &count, state);
536	if (error)
537		goto err_inode;
538
539	if (count > 0) {
540		ASSERT(type == XFS_BMAP_UNMAP);
541		irec.br_startblock = bmap->me_startblock;
542		irec.br_blockcount = count;
543		irec.br_startoff = bmap->me_startoff;
544		irec.br_state = state;
545		xfs_bmap_unmap_extent(tp, ip, &irec);
546	}
547
548	set_bit(XFS_BUI_RECOVERED, &buip->bui_flags);
549	xfs_defer_move(parent_tp, tp);
550	error = xfs_trans_commit(tp);
551	xfs_iunlock(ip, XFS_ILOCK_EXCL);
552	xfs_irele(ip);
553
554	return error;
555
556err_inode:
557	xfs_defer_move(parent_tp, tp);
558	xfs_trans_cancel(tp);
559	if (ip) {
560		xfs_iunlock(ip, XFS_ILOCK_EXCL);
561		xfs_irele(ip);
562	}
563	return error;
564}