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 #include "xfs_error.h"
25 #include "xfs_log_priv.h"
26 #include "xfs_log_recover.h"
27 
28 struct kmem_cache	*xfs_bui_cache;
29 struct kmem_cache	*xfs_bud_cache;
30 
31 static const struct xfs_item_ops xfs_bui_item_ops;
32 
BUI_ITEM(struct xfs_log_item * lip)33 static inline struct xfs_bui_log_item *BUI_ITEM(struct xfs_log_item *lip)
34 {
35 	return container_of(lip, struct xfs_bui_log_item, bui_item);
36 }
37 
38 STATIC void
xfs_bui_item_free(struct xfs_bui_log_item * buip)39 xfs_bui_item_free(
40 	struct xfs_bui_log_item	*buip)
41 {
42 	kmem_free(buip->bui_item.li_lv_shadow);
43 	kmem_cache_free(xfs_bui_cache, buip);
44 }
45 
46 /*
47  * Freeing the BUI requires that we remove it from the AIL if it has already
48  * been placed there. However, the BUI may not yet have been placed in the AIL
49  * when called by xfs_bui_release() from BUD processing due to the ordering of
50  * committed vs unpin operations in bulk insert operations. Hence the reference
51  * count to ensure only the last caller frees the BUI.
52  */
53 STATIC void
xfs_bui_release(struct xfs_bui_log_item * buip)54 xfs_bui_release(
55 	struct xfs_bui_log_item	*buip)
56 {
57 	ASSERT(atomic_read(&buip->bui_refcount) > 0);
58 	if (!atomic_dec_and_test(&buip->bui_refcount))
59 		return;
60 
61 	xfs_trans_ail_delete(&buip->bui_item, 0);
62 	xfs_bui_item_free(buip);
63 }
64 
65 
66 STATIC void
xfs_bui_item_size(struct xfs_log_item * lip,int * nvecs,int * nbytes)67 xfs_bui_item_size(
68 	struct xfs_log_item	*lip,
69 	int			*nvecs,
70 	int			*nbytes)
71 {
72 	struct xfs_bui_log_item	*buip = BUI_ITEM(lip);
73 
74 	*nvecs += 1;
75 	*nbytes += xfs_bui_log_format_sizeof(buip->bui_format.bui_nextents);
76 }
77 
78 /*
79  * This is called to fill in the vector of log iovecs for the
80  * given bui log item. We use only 1 iovec, and we point that
81  * at the bui_log_format structure embedded in the bui item.
82  * It is at this point that we assert that all of the extent
83  * slots in the bui item have been filled.
84  */
85 STATIC void
xfs_bui_item_format(struct xfs_log_item * lip,struct xfs_log_vec * lv)86 xfs_bui_item_format(
87 	struct xfs_log_item	*lip,
88 	struct xfs_log_vec	*lv)
89 {
90 	struct xfs_bui_log_item	*buip = BUI_ITEM(lip);
91 	struct xfs_log_iovec	*vecp = NULL;
92 
93 	ASSERT(atomic_read(&buip->bui_next_extent) ==
94 			buip->bui_format.bui_nextents);
95 
96 	buip->bui_format.bui_type = XFS_LI_BUI;
97 	buip->bui_format.bui_size = 1;
98 
99 	xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_BUI_FORMAT, &buip->bui_format,
100 			xfs_bui_log_format_sizeof(buip->bui_format.bui_nextents));
101 }
102 
103 /*
104  * The unpin operation is the last place an BUI is manipulated in the log. It is
105  * either inserted in the AIL or aborted in the event of a log I/O error. In
106  * either case, the BUI transaction has been successfully committed to make it
107  * this far. Therefore, we expect whoever committed the BUI to either construct
108  * and commit the BUD or drop the BUD's reference in the event of error. Simply
109  * drop the log's BUI reference now that the log is done with it.
110  */
111 STATIC void
xfs_bui_item_unpin(struct xfs_log_item * lip,int remove)112 xfs_bui_item_unpin(
113 	struct xfs_log_item	*lip,
114 	int			remove)
115 {
116 	struct xfs_bui_log_item	*buip = BUI_ITEM(lip);
117 
118 	xfs_bui_release(buip);
119 }
120 
121 /*
122  * The BUI has been either committed or aborted if the transaction has been
123  * cancelled. If the transaction was cancelled, an BUD isn't going to be
124  * constructed and thus we free the BUI here directly.
125  */
126 STATIC void
xfs_bui_item_release(struct xfs_log_item * lip)127 xfs_bui_item_release(
128 	struct xfs_log_item	*lip)
129 {
130 	xfs_bui_release(BUI_ITEM(lip));
131 }
132 
133 /*
134  * Allocate and initialize an bui item with the given number of extents.
135  */
136 STATIC struct xfs_bui_log_item *
xfs_bui_init(struct xfs_mount * mp)137 xfs_bui_init(
138 	struct xfs_mount		*mp)
139 
140 {
141 	struct xfs_bui_log_item		*buip;
142 
143 	buip = kmem_cache_zalloc(xfs_bui_cache, GFP_KERNEL | __GFP_NOFAIL);
144 
145 	xfs_log_item_init(mp, &buip->bui_item, XFS_LI_BUI, &xfs_bui_item_ops);
146 	buip->bui_format.bui_nextents = XFS_BUI_MAX_FAST_EXTENTS;
147 	buip->bui_format.bui_id = (uintptr_t)(void *)buip;
148 	atomic_set(&buip->bui_next_extent, 0);
149 	atomic_set(&buip->bui_refcount, 2);
150 
151 	return buip;
152 }
153 
BUD_ITEM(struct xfs_log_item * lip)154 static inline struct xfs_bud_log_item *BUD_ITEM(struct xfs_log_item *lip)
155 {
156 	return container_of(lip, struct xfs_bud_log_item, bud_item);
157 }
158 
159 STATIC void
xfs_bud_item_size(struct xfs_log_item * lip,int * nvecs,int * nbytes)160 xfs_bud_item_size(
161 	struct xfs_log_item	*lip,
162 	int			*nvecs,
163 	int			*nbytes)
164 {
165 	*nvecs += 1;
166 	*nbytes += sizeof(struct xfs_bud_log_format);
167 }
168 
169 /*
170  * This is called to fill in the vector of log iovecs for the
171  * given bud log item. We use only 1 iovec, and we point that
172  * at the bud_log_format structure embedded in the bud item.
173  * It is at this point that we assert that all of the extent
174  * slots in the bud item have been filled.
175  */
176 STATIC void
xfs_bud_item_format(struct xfs_log_item * lip,struct xfs_log_vec * lv)177 xfs_bud_item_format(
178 	struct xfs_log_item	*lip,
179 	struct xfs_log_vec	*lv)
180 {
181 	struct xfs_bud_log_item	*budp = BUD_ITEM(lip);
182 	struct xfs_log_iovec	*vecp = NULL;
183 
184 	budp->bud_format.bud_type = XFS_LI_BUD;
185 	budp->bud_format.bud_size = 1;
186 
187 	xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_BUD_FORMAT, &budp->bud_format,
188 			sizeof(struct xfs_bud_log_format));
189 }
190 
191 /*
192  * The BUD is either committed or aborted if the transaction is cancelled. If
193  * the transaction is cancelled, drop our reference to the BUI and free the
194  * BUD.
195  */
196 STATIC void
xfs_bud_item_release(struct xfs_log_item * lip)197 xfs_bud_item_release(
198 	struct xfs_log_item	*lip)
199 {
200 	struct xfs_bud_log_item	*budp = BUD_ITEM(lip);
201 
202 	xfs_bui_release(budp->bud_buip);
203 	kmem_free(budp->bud_item.li_lv_shadow);
204 	kmem_cache_free(xfs_bud_cache, budp);
205 }
206 
207 static struct xfs_log_item *
xfs_bud_item_intent(struct xfs_log_item * lip)208 xfs_bud_item_intent(
209 	struct xfs_log_item	*lip)
210 {
211 	return &BUD_ITEM(lip)->bud_buip->bui_item;
212 }
213 
214 static const struct xfs_item_ops xfs_bud_item_ops = {
215 	.flags		= XFS_ITEM_RELEASE_WHEN_COMMITTED |
216 			  XFS_ITEM_INTENT_DONE,
217 	.iop_size	= xfs_bud_item_size,
218 	.iop_format	= xfs_bud_item_format,
219 	.iop_release	= xfs_bud_item_release,
220 	.iop_intent	= xfs_bud_item_intent,
221 };
222 
223 static struct xfs_bud_log_item *
xfs_trans_get_bud(struct xfs_trans * tp,struct xfs_bui_log_item * buip)224 xfs_trans_get_bud(
225 	struct xfs_trans		*tp,
226 	struct xfs_bui_log_item		*buip)
227 {
228 	struct xfs_bud_log_item		*budp;
229 
230 	budp = kmem_cache_zalloc(xfs_bud_cache, GFP_KERNEL | __GFP_NOFAIL);
231 	xfs_log_item_init(tp->t_mountp, &budp->bud_item, XFS_LI_BUD,
232 			  &xfs_bud_item_ops);
233 	budp->bud_buip = buip;
234 	budp->bud_format.bud_bui_id = buip->bui_format.bui_id;
235 
236 	xfs_trans_add_item(tp, &budp->bud_item);
237 	return budp;
238 }
239 
240 /*
241  * Finish an bmap update and log it to the BUD. Note that the
242  * transaction is marked dirty regardless of whether the bmap update
243  * succeeds or fails to support the BUI/BUD lifecycle rules.
244  */
245 static int
xfs_trans_log_finish_bmap_update(struct xfs_trans * tp,struct xfs_bud_log_item * budp,enum xfs_bmap_intent_type type,struct xfs_inode * ip,int whichfork,xfs_fileoff_t startoff,xfs_fsblock_t startblock,xfs_filblks_t * blockcount,xfs_exntst_t state)246 xfs_trans_log_finish_bmap_update(
247 	struct xfs_trans		*tp,
248 	struct xfs_bud_log_item		*budp,
249 	enum xfs_bmap_intent_type	type,
250 	struct xfs_inode		*ip,
251 	int				whichfork,
252 	xfs_fileoff_t			startoff,
253 	xfs_fsblock_t			startblock,
254 	xfs_filblks_t			*blockcount,
255 	xfs_exntst_t			state)
256 {
257 	int				error;
258 
259 	error = xfs_bmap_finish_one(tp, ip, type, whichfork, startoff,
260 			startblock, blockcount, state);
261 
262 	/*
263 	 * Mark the transaction dirty, even on error. This ensures the
264 	 * transaction is aborted, which:
265 	 *
266 	 * 1.) releases the BUI and frees the BUD
267 	 * 2.) shuts down the filesystem
268 	 */
269 	tp->t_flags |= XFS_TRANS_DIRTY | XFS_TRANS_HAS_INTENT_DONE;
270 	set_bit(XFS_LI_DIRTY, &budp->bud_item.li_flags);
271 
272 	return error;
273 }
274 
275 /* Sort bmap intents by inode. */
276 static int
xfs_bmap_update_diff_items(void * priv,const struct list_head * a,const struct list_head * b)277 xfs_bmap_update_diff_items(
278 	void				*priv,
279 	const struct list_head		*a,
280 	const struct list_head		*b)
281 {
282 	struct xfs_bmap_intent		*ba;
283 	struct xfs_bmap_intent		*bb;
284 
285 	ba = container_of(a, struct xfs_bmap_intent, bi_list);
286 	bb = container_of(b, struct xfs_bmap_intent, bi_list);
287 	return ba->bi_owner->i_ino - bb->bi_owner->i_ino;
288 }
289 
290 /* Set the map extent flags for this mapping. */
291 static void
xfs_trans_set_bmap_flags(struct xfs_map_extent * bmap,enum xfs_bmap_intent_type type,int whichfork,xfs_exntst_t state)292 xfs_trans_set_bmap_flags(
293 	struct xfs_map_extent		*bmap,
294 	enum xfs_bmap_intent_type	type,
295 	int				whichfork,
296 	xfs_exntst_t			state)
297 {
298 	bmap->me_flags = 0;
299 	switch (type) {
300 	case XFS_BMAP_MAP:
301 	case XFS_BMAP_UNMAP:
302 		bmap->me_flags = type;
303 		break;
304 	default:
305 		ASSERT(0);
306 	}
307 	if (state == XFS_EXT_UNWRITTEN)
308 		bmap->me_flags |= XFS_BMAP_EXTENT_UNWRITTEN;
309 	if (whichfork == XFS_ATTR_FORK)
310 		bmap->me_flags |= XFS_BMAP_EXTENT_ATTR_FORK;
311 }
312 
313 /* Log bmap updates in the intent item. */
314 STATIC void
xfs_bmap_update_log_item(struct xfs_trans * tp,struct xfs_bui_log_item * buip,struct xfs_bmap_intent * bmap)315 xfs_bmap_update_log_item(
316 	struct xfs_trans		*tp,
317 	struct xfs_bui_log_item		*buip,
318 	struct xfs_bmap_intent		*bmap)
319 {
320 	uint				next_extent;
321 	struct xfs_map_extent		*map;
322 
323 	tp->t_flags |= XFS_TRANS_DIRTY;
324 	set_bit(XFS_LI_DIRTY, &buip->bui_item.li_flags);
325 
326 	/*
327 	 * atomic_inc_return gives us the value after the increment;
328 	 * we want to use it as an array index so we need to subtract 1 from
329 	 * it.
330 	 */
331 	next_extent = atomic_inc_return(&buip->bui_next_extent) - 1;
332 	ASSERT(next_extent < buip->bui_format.bui_nextents);
333 	map = &buip->bui_format.bui_extents[next_extent];
334 	map->me_owner = bmap->bi_owner->i_ino;
335 	map->me_startblock = bmap->bi_bmap.br_startblock;
336 	map->me_startoff = bmap->bi_bmap.br_startoff;
337 	map->me_len = bmap->bi_bmap.br_blockcount;
338 	xfs_trans_set_bmap_flags(map, bmap->bi_type, bmap->bi_whichfork,
339 			bmap->bi_bmap.br_state);
340 }
341 
342 static struct xfs_log_item *
xfs_bmap_update_create_intent(struct xfs_trans * tp,struct list_head * items,unsigned int count,bool sort)343 xfs_bmap_update_create_intent(
344 	struct xfs_trans		*tp,
345 	struct list_head		*items,
346 	unsigned int			count,
347 	bool				sort)
348 {
349 	struct xfs_mount		*mp = tp->t_mountp;
350 	struct xfs_bui_log_item		*buip = xfs_bui_init(mp);
351 	struct xfs_bmap_intent		*bmap;
352 
353 	ASSERT(count == XFS_BUI_MAX_FAST_EXTENTS);
354 
355 	xfs_trans_add_item(tp, &buip->bui_item);
356 	if (sort)
357 		list_sort(mp, items, xfs_bmap_update_diff_items);
358 	list_for_each_entry(bmap, items, bi_list)
359 		xfs_bmap_update_log_item(tp, buip, bmap);
360 	return &buip->bui_item;
361 }
362 
363 /* Get an BUD so we can process all the deferred rmap updates. */
364 static struct xfs_log_item *
xfs_bmap_update_create_done(struct xfs_trans * tp,struct xfs_log_item * intent,unsigned int count)365 xfs_bmap_update_create_done(
366 	struct xfs_trans		*tp,
367 	struct xfs_log_item		*intent,
368 	unsigned int			count)
369 {
370 	return &xfs_trans_get_bud(tp, BUI_ITEM(intent))->bud_item;
371 }
372 
373 /* Process a deferred rmap update. */
374 STATIC int
xfs_bmap_update_finish_item(struct xfs_trans * tp,struct xfs_log_item * done,struct list_head * item,struct xfs_btree_cur ** state)375 xfs_bmap_update_finish_item(
376 	struct xfs_trans		*tp,
377 	struct xfs_log_item		*done,
378 	struct list_head		*item,
379 	struct xfs_btree_cur		**state)
380 {
381 	struct xfs_bmap_intent		*bmap;
382 	xfs_filblks_t			count;
383 	int				error;
384 
385 	bmap = container_of(item, struct xfs_bmap_intent, bi_list);
386 	count = bmap->bi_bmap.br_blockcount;
387 	error = xfs_trans_log_finish_bmap_update(tp, BUD_ITEM(done),
388 			bmap->bi_type,
389 			bmap->bi_owner, bmap->bi_whichfork,
390 			bmap->bi_bmap.br_startoff,
391 			bmap->bi_bmap.br_startblock,
392 			&count,
393 			bmap->bi_bmap.br_state);
394 	if (!error && count > 0) {
395 		ASSERT(bmap->bi_type == XFS_BMAP_UNMAP);
396 		bmap->bi_bmap.br_blockcount = count;
397 		return -EAGAIN;
398 	}
399 	kmem_cache_free(xfs_bmap_intent_cache, bmap);
400 	return error;
401 }
402 
403 /* Abort all pending BUIs. */
404 STATIC void
xfs_bmap_update_abort_intent(struct xfs_log_item * intent)405 xfs_bmap_update_abort_intent(
406 	struct xfs_log_item		*intent)
407 {
408 	xfs_bui_release(BUI_ITEM(intent));
409 }
410 
411 /* Cancel a deferred rmap update. */
412 STATIC void
xfs_bmap_update_cancel_item(struct list_head * item)413 xfs_bmap_update_cancel_item(
414 	struct list_head		*item)
415 {
416 	struct xfs_bmap_intent		*bmap;
417 
418 	bmap = container_of(item, struct xfs_bmap_intent, bi_list);
419 	kmem_cache_free(xfs_bmap_intent_cache, bmap);
420 }
421 
422 const struct xfs_defer_op_type xfs_bmap_update_defer_type = {
423 	.max_items	= XFS_BUI_MAX_FAST_EXTENTS,
424 	.create_intent	= xfs_bmap_update_create_intent,
425 	.abort_intent	= xfs_bmap_update_abort_intent,
426 	.create_done	= xfs_bmap_update_create_done,
427 	.finish_item	= xfs_bmap_update_finish_item,
428 	.cancel_item	= xfs_bmap_update_cancel_item,
429 };
430 
431 /* Is this recovered BUI ok? */
432 static inline bool
xfs_bui_validate(struct xfs_mount * mp,struct xfs_bui_log_item * buip)433 xfs_bui_validate(
434 	struct xfs_mount		*mp,
435 	struct xfs_bui_log_item		*buip)
436 {
437 	struct xfs_map_extent		*bmap;
438 
439 	/* Only one mapping operation per BUI... */
440 	if (buip->bui_format.bui_nextents != XFS_BUI_MAX_FAST_EXTENTS)
441 		return false;
442 
443 	bmap = &buip->bui_format.bui_extents[0];
444 
445 	if (bmap->me_flags & ~XFS_BMAP_EXTENT_FLAGS)
446 		return false;
447 
448 	switch (bmap->me_flags & XFS_BMAP_EXTENT_TYPE_MASK) {
449 	case XFS_BMAP_MAP:
450 	case XFS_BMAP_UNMAP:
451 		break;
452 	default:
453 		return false;
454 	}
455 
456 	if (!xfs_verify_ino(mp, bmap->me_owner))
457 		return false;
458 
459 	if (!xfs_verify_fileext(mp, bmap->me_startoff, bmap->me_len))
460 		return false;
461 
462 	return xfs_verify_fsbext(mp, bmap->me_startblock, bmap->me_len);
463 }
464 
465 /*
466  * Process a bmap update intent item that was recovered from the log.
467  * We need to update some inode's bmbt.
468  */
469 STATIC int
xfs_bui_item_recover(struct xfs_log_item * lip,struct list_head * capture_list)470 xfs_bui_item_recover(
471 	struct xfs_log_item		*lip,
472 	struct list_head		*capture_list)
473 {
474 	struct xfs_bmbt_irec		irec;
475 	struct xfs_bui_log_item		*buip = BUI_ITEM(lip);
476 	struct xfs_trans		*tp;
477 	struct xfs_inode		*ip = NULL;
478 	struct xfs_mount		*mp = lip->li_log->l_mp;
479 	struct xfs_map_extent		*bmap;
480 	struct xfs_bud_log_item		*budp;
481 	xfs_filblks_t			count;
482 	xfs_exntst_t			state;
483 	unsigned int			bui_type;
484 	int				whichfork;
485 	int				iext_delta;
486 	int				error = 0;
487 
488 	if (!xfs_bui_validate(mp, buip)) {
489 		XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
490 				&buip->bui_format, sizeof(buip->bui_format));
491 		return -EFSCORRUPTED;
492 	}
493 
494 	bmap = &buip->bui_format.bui_extents[0];
495 	state = (bmap->me_flags & XFS_BMAP_EXTENT_UNWRITTEN) ?
496 			XFS_EXT_UNWRITTEN : XFS_EXT_NORM;
497 	whichfork = (bmap->me_flags & XFS_BMAP_EXTENT_ATTR_FORK) ?
498 			XFS_ATTR_FORK : XFS_DATA_FORK;
499 	bui_type = bmap->me_flags & XFS_BMAP_EXTENT_TYPE_MASK;
500 
501 	error = xlog_recover_iget(mp, bmap->me_owner, &ip);
502 	if (error)
503 		return error;
504 
505 	/* Allocate transaction and do the work. */
506 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate,
507 			XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK), 0, 0, &tp);
508 	if (error)
509 		goto err_rele;
510 
511 	budp = xfs_trans_get_bud(tp, buip);
512 	xfs_ilock(ip, XFS_ILOCK_EXCL);
513 	xfs_trans_ijoin(tp, ip, 0);
514 
515 	if (bui_type == XFS_BMAP_MAP)
516 		iext_delta = XFS_IEXT_ADD_NOSPLIT_CNT;
517 	else
518 		iext_delta = XFS_IEXT_PUNCH_HOLE_CNT;
519 
520 	error = xfs_iext_count_may_overflow(ip, whichfork, iext_delta);
521 	if (error == -EFBIG)
522 		error = xfs_iext_count_upgrade(tp, ip, iext_delta);
523 	if (error)
524 		goto err_cancel;
525 
526 	count = bmap->me_len;
527 	error = xfs_trans_log_finish_bmap_update(tp, budp, bui_type, ip,
528 			whichfork, bmap->me_startoff, bmap->me_startblock,
529 			&count, state);
530 	if (error == -EFSCORRUPTED)
531 		XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bmap,
532 				sizeof(*bmap));
533 	if (error)
534 		goto err_cancel;
535 
536 	if (count > 0) {
537 		ASSERT(bui_type == XFS_BMAP_UNMAP);
538 		irec.br_startblock = bmap->me_startblock;
539 		irec.br_blockcount = count;
540 		irec.br_startoff = bmap->me_startoff;
541 		irec.br_state = state;
542 		xfs_bmap_unmap_extent(tp, ip, &irec);
543 	}
544 
545 	/*
546 	 * Commit transaction, which frees the transaction and saves the inode
547 	 * for later replay activities.
548 	 */
549 	error = xfs_defer_ops_capture_and_commit(tp, capture_list);
550 	if (error)
551 		goto err_unlock;
552 
553 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
554 	xfs_irele(ip);
555 	return 0;
556 
557 err_cancel:
558 	xfs_trans_cancel(tp);
559 err_unlock:
560 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
561 err_rele:
562 	xfs_irele(ip);
563 	return error;
564 }
565 
566 STATIC bool
xfs_bui_item_match(struct xfs_log_item * lip,uint64_t intent_id)567 xfs_bui_item_match(
568 	struct xfs_log_item	*lip,
569 	uint64_t		intent_id)
570 {
571 	return BUI_ITEM(lip)->bui_format.bui_id == intent_id;
572 }
573 
574 /* Relog an intent item to push the log tail forward. */
575 static struct xfs_log_item *
xfs_bui_item_relog(struct xfs_log_item * intent,struct xfs_trans * tp)576 xfs_bui_item_relog(
577 	struct xfs_log_item		*intent,
578 	struct xfs_trans		*tp)
579 {
580 	struct xfs_bud_log_item		*budp;
581 	struct xfs_bui_log_item		*buip;
582 	struct xfs_map_extent		*extp;
583 	unsigned int			count;
584 
585 	count = BUI_ITEM(intent)->bui_format.bui_nextents;
586 	extp = BUI_ITEM(intent)->bui_format.bui_extents;
587 
588 	tp->t_flags |= XFS_TRANS_DIRTY;
589 	budp = xfs_trans_get_bud(tp, BUI_ITEM(intent));
590 	set_bit(XFS_LI_DIRTY, &budp->bud_item.li_flags);
591 
592 	buip = xfs_bui_init(tp->t_mountp);
593 	memcpy(buip->bui_format.bui_extents, extp, count * sizeof(*extp));
594 	atomic_set(&buip->bui_next_extent, count);
595 	xfs_trans_add_item(tp, &buip->bui_item);
596 	set_bit(XFS_LI_DIRTY, &buip->bui_item.li_flags);
597 	return &buip->bui_item;
598 }
599 
600 static const struct xfs_item_ops xfs_bui_item_ops = {
601 	.flags		= XFS_ITEM_INTENT,
602 	.iop_size	= xfs_bui_item_size,
603 	.iop_format	= xfs_bui_item_format,
604 	.iop_unpin	= xfs_bui_item_unpin,
605 	.iop_release	= xfs_bui_item_release,
606 	.iop_recover	= xfs_bui_item_recover,
607 	.iop_match	= xfs_bui_item_match,
608 	.iop_relog	= xfs_bui_item_relog,
609 };
610 
611 static inline void
xfs_bui_copy_format(struct xfs_bui_log_format * dst,const struct xfs_bui_log_format * src)612 xfs_bui_copy_format(
613 	struct xfs_bui_log_format	*dst,
614 	const struct xfs_bui_log_format	*src)
615 {
616 	unsigned int			i;
617 
618 	memcpy(dst, src, offsetof(struct xfs_bui_log_format, bui_extents));
619 
620 	for (i = 0; i < src->bui_nextents; i++)
621 		memcpy(&dst->bui_extents[i], &src->bui_extents[i],
622 				sizeof(struct xfs_map_extent));
623 }
624 
625 /*
626  * This routine is called to create an in-core extent bmap update
627  * item from the bui format structure which was logged on disk.
628  * It allocates an in-core bui, copies the extents from the format
629  * structure into it, and adds the bui to the AIL with the given
630  * LSN.
631  */
632 STATIC int
xlog_recover_bui_commit_pass2(struct xlog * log,struct list_head * buffer_list,struct xlog_recover_item * item,xfs_lsn_t lsn)633 xlog_recover_bui_commit_pass2(
634 	struct xlog			*log,
635 	struct list_head		*buffer_list,
636 	struct xlog_recover_item	*item,
637 	xfs_lsn_t			lsn)
638 {
639 	struct xfs_mount		*mp = log->l_mp;
640 	struct xfs_bui_log_item		*buip;
641 	struct xfs_bui_log_format	*bui_formatp;
642 	size_t				len;
643 
644 	bui_formatp = item->ri_buf[0].i_addr;
645 
646 	if (item->ri_buf[0].i_len < xfs_bui_log_format_sizeof(0)) {
647 		XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
648 				item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
649 		return -EFSCORRUPTED;
650 	}
651 
652 	if (bui_formatp->bui_nextents != XFS_BUI_MAX_FAST_EXTENTS) {
653 		XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
654 				item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
655 		return -EFSCORRUPTED;
656 	}
657 
658 	len = xfs_bui_log_format_sizeof(bui_formatp->bui_nextents);
659 	if (item->ri_buf[0].i_len != len) {
660 		XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
661 				item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
662 		return -EFSCORRUPTED;
663 	}
664 
665 	buip = xfs_bui_init(mp);
666 	xfs_bui_copy_format(&buip->bui_format, bui_formatp);
667 	atomic_set(&buip->bui_next_extent, bui_formatp->bui_nextents);
668 	/*
669 	 * Insert the intent into the AIL directly and drop one reference so
670 	 * that finishing or canceling the work will drop the other.
671 	 */
672 	xfs_trans_ail_insert(log->l_ailp, &buip->bui_item, lsn);
673 	xfs_bui_release(buip);
674 	return 0;
675 }
676 
677 const struct xlog_recover_item_ops xlog_bui_item_ops = {
678 	.item_type		= XFS_LI_BUI,
679 	.commit_pass2		= xlog_recover_bui_commit_pass2,
680 };
681 
682 /*
683  * This routine is called when an BUD format structure is found in a committed
684  * transaction in the log. Its purpose is to cancel the corresponding BUI if it
685  * was still in the log. To do this it searches the AIL for the BUI with an id
686  * equal to that in the BUD format structure. If we find it we drop the BUD
687  * reference, which removes the BUI from the AIL and frees it.
688  */
689 STATIC int
xlog_recover_bud_commit_pass2(struct xlog * log,struct list_head * buffer_list,struct xlog_recover_item * item,xfs_lsn_t lsn)690 xlog_recover_bud_commit_pass2(
691 	struct xlog			*log,
692 	struct list_head		*buffer_list,
693 	struct xlog_recover_item	*item,
694 	xfs_lsn_t			lsn)
695 {
696 	struct xfs_bud_log_format	*bud_formatp;
697 
698 	bud_formatp = item->ri_buf[0].i_addr;
699 	if (item->ri_buf[0].i_len != sizeof(struct xfs_bud_log_format)) {
700 		XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, log->l_mp,
701 				item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
702 		return -EFSCORRUPTED;
703 	}
704 
705 	xlog_recover_release_intent(log, XFS_LI_BUI, bud_formatp->bud_bui_id);
706 	return 0;
707 }
708 
709 const struct xlog_recover_item_ops xlog_bud_item_ops = {
710 	.item_type		= XFS_LI_BUD,
711 	.commit_pass2		= xlog_recover_bud_commit_pass2,
712 };
713