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 #include "xfs_ag.h"
28 
29 struct kmem_cache	*xfs_bui_cache;
30 struct kmem_cache	*xfs_bud_cache;
31 
32 static const struct xfs_item_ops xfs_bui_item_ops;
33 
BUI_ITEM(struct xfs_log_item * lip)34 static inline struct xfs_bui_log_item *BUI_ITEM(struct xfs_log_item *lip)
35 {
36 	return container_of(lip, struct xfs_bui_log_item, bui_item);
37 }
38 
39 STATIC void
xfs_bui_item_free(struct xfs_bui_log_item * buip)40 xfs_bui_item_free(
41 	struct xfs_bui_log_item	*buip)
42 {
43 	kmem_free(buip->bui_item.li_lv_shadow);
44 	kmem_cache_free(xfs_bui_cache, buip);
45 }
46 
47 /*
48  * Freeing the BUI requires that we remove it from the AIL if it has already
49  * been placed there. However, the BUI may not yet have been placed in the AIL
50  * when called by xfs_bui_release() from BUD processing due to the ordering of
51  * committed vs unpin operations in bulk insert operations. Hence the reference
52  * count to ensure only the last caller frees the BUI.
53  */
54 STATIC void
xfs_bui_release(struct xfs_bui_log_item * buip)55 xfs_bui_release(
56 	struct xfs_bui_log_item	*buip)
57 {
58 	ASSERT(atomic_read(&buip->bui_refcount) > 0);
59 	if (!atomic_dec_and_test(&buip->bui_refcount))
60 		return;
61 
62 	xfs_trans_ail_delete(&buip->bui_item, 0);
63 	xfs_bui_item_free(buip);
64 }
65 
66 
67 STATIC void
xfs_bui_item_size(struct xfs_log_item * lip,int * nvecs,int * nbytes)68 xfs_bui_item_size(
69 	struct xfs_log_item	*lip,
70 	int			*nvecs,
71 	int			*nbytes)
72 {
73 	struct xfs_bui_log_item	*buip = BUI_ITEM(lip);
74 
75 	*nvecs += 1;
76 	*nbytes += xfs_bui_log_format_sizeof(buip->bui_format.bui_nextents);
77 }
78 
79 /*
80  * This is called to fill in the vector of log iovecs for the
81  * given bui log item. We use only 1 iovec, and we point that
82  * at the bui_log_format structure embedded in the bui item.
83  * It is at this point that we assert that all of the extent
84  * slots in the bui item have been filled.
85  */
86 STATIC void
xfs_bui_item_format(struct xfs_log_item * lip,struct xfs_log_vec * lv)87 xfs_bui_item_format(
88 	struct xfs_log_item	*lip,
89 	struct xfs_log_vec	*lv)
90 {
91 	struct xfs_bui_log_item	*buip = BUI_ITEM(lip);
92 	struct xfs_log_iovec	*vecp = NULL;
93 
94 	ASSERT(atomic_read(&buip->bui_next_extent) ==
95 			buip->bui_format.bui_nextents);
96 
97 	buip->bui_format.bui_type = XFS_LI_BUI;
98 	buip->bui_format.bui_size = 1;
99 
100 	xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_BUI_FORMAT, &buip->bui_format,
101 			xfs_bui_log_format_sizeof(buip->bui_format.bui_nextents));
102 }
103 
104 /*
105  * The unpin operation is the last place an BUI is manipulated in the log. It is
106  * either inserted in the AIL or aborted in the event of a log I/O error. In
107  * either case, the BUI transaction has been successfully committed to make it
108  * this far. Therefore, we expect whoever committed the BUI to either construct
109  * and commit the BUD or drop the BUD's reference in the event of error. Simply
110  * drop the log's BUI reference now that the log is done with it.
111  */
112 STATIC void
xfs_bui_item_unpin(struct xfs_log_item * lip,int remove)113 xfs_bui_item_unpin(
114 	struct xfs_log_item	*lip,
115 	int			remove)
116 {
117 	struct xfs_bui_log_item	*buip = BUI_ITEM(lip);
118 
119 	xfs_bui_release(buip);
120 }
121 
122 /*
123  * The BUI has been either committed or aborted if the transaction has been
124  * cancelled. If the transaction was cancelled, an BUD isn't going to be
125  * constructed and thus we free the BUI here directly.
126  */
127 STATIC void
xfs_bui_item_release(struct xfs_log_item * lip)128 xfs_bui_item_release(
129 	struct xfs_log_item	*lip)
130 {
131 	xfs_bui_release(BUI_ITEM(lip));
132 }
133 
134 /*
135  * Allocate and initialize an bui item with the given number of extents.
136  */
137 STATIC struct xfs_bui_log_item *
xfs_bui_init(struct xfs_mount * mp)138 xfs_bui_init(
139 	struct xfs_mount		*mp)
140 
141 {
142 	struct xfs_bui_log_item		*buip;
143 
144 	buip = kmem_cache_zalloc(xfs_bui_cache, GFP_KERNEL | __GFP_NOFAIL);
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 
BUD_ITEM(struct xfs_log_item * lip)155 static 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 
160 STATIC void
xfs_bud_item_size(struct xfs_log_item * lip,int * nvecs,int * nbytes)161 xfs_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  */
177 STATIC void
xfs_bud_item_format(struct xfs_log_item * lip,struct xfs_log_vec * lv)178 xfs_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  */
197 STATIC void
xfs_bud_item_release(struct xfs_log_item * lip)198 xfs_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_free(budp->bud_item.li_lv_shadow);
205 	kmem_cache_free(xfs_bud_cache, budp);
206 }
207 
208 static struct xfs_log_item *
xfs_bud_item_intent(struct xfs_log_item * lip)209 xfs_bud_item_intent(
210 	struct xfs_log_item	*lip)
211 {
212 	return &BUD_ITEM(lip)->bud_buip->bui_item;
213 }
214 
215 static const struct xfs_item_ops xfs_bud_item_ops = {
216 	.flags		= XFS_ITEM_RELEASE_WHEN_COMMITTED |
217 			  XFS_ITEM_INTENT_DONE,
218 	.iop_size	= xfs_bud_item_size,
219 	.iop_format	= xfs_bud_item_format,
220 	.iop_release	= xfs_bud_item_release,
221 	.iop_intent	= xfs_bud_item_intent,
222 };
223 
224 static struct xfs_bud_log_item *
xfs_trans_get_bud(struct xfs_trans * tp,struct xfs_bui_log_item * buip)225 xfs_trans_get_bud(
226 	struct xfs_trans		*tp,
227 	struct xfs_bui_log_item		*buip)
228 {
229 	struct xfs_bud_log_item		*budp;
230 
231 	budp = kmem_cache_zalloc(xfs_bud_cache, GFP_KERNEL | __GFP_NOFAIL);
232 	xfs_log_item_init(tp->t_mountp, &budp->bud_item, XFS_LI_BUD,
233 			  &xfs_bud_item_ops);
234 	budp->bud_buip = buip;
235 	budp->bud_format.bud_bui_id = buip->bui_format.bui_id;
236 
237 	xfs_trans_add_item(tp, &budp->bud_item);
238 	return budp;
239 }
240 
241 /*
242  * Finish an bmap update and log it to the BUD. Note that the
243  * transaction is marked dirty regardless of whether the bmap update
244  * succeeds or fails to support the BUI/BUD lifecycle rules.
245  */
246 static int
xfs_trans_log_finish_bmap_update(struct xfs_trans * tp,struct xfs_bud_log_item * budp,struct xfs_bmap_intent * bi)247 xfs_trans_log_finish_bmap_update(
248 	struct xfs_trans		*tp,
249 	struct xfs_bud_log_item		*budp,
250 	struct xfs_bmap_intent		*bi)
251 {
252 	int				error;
253 
254 	error = xfs_bmap_finish_one(tp, bi);
255 
256 	/*
257 	 * Mark the transaction dirty, even on error. This ensures the
258 	 * transaction is aborted, which:
259 	 *
260 	 * 1.) releases the BUI and frees the BUD
261 	 * 2.) shuts down the filesystem
262 	 */
263 	tp->t_flags |= XFS_TRANS_DIRTY | XFS_TRANS_HAS_INTENT_DONE;
264 	set_bit(XFS_LI_DIRTY, &budp->bud_item.li_flags);
265 
266 	return error;
267 }
268 
269 /* Sort bmap intents by inode. */
270 static int
xfs_bmap_update_diff_items(void * priv,const struct list_head * a,const struct list_head * b)271 xfs_bmap_update_diff_items(
272 	void				*priv,
273 	const struct list_head		*a,
274 	const struct list_head		*b)
275 {
276 	struct xfs_bmap_intent		*ba;
277 	struct xfs_bmap_intent		*bb;
278 
279 	ba = container_of(a, struct xfs_bmap_intent, bi_list);
280 	bb = container_of(b, struct xfs_bmap_intent, bi_list);
281 	return ba->bi_owner->i_ino - bb->bi_owner->i_ino;
282 }
283 
284 /* Set the map extent flags for this mapping. */
285 static void
xfs_trans_set_bmap_flags(struct xfs_map_extent * map,enum xfs_bmap_intent_type type,int whichfork,xfs_exntst_t state)286 xfs_trans_set_bmap_flags(
287 	struct xfs_map_extent		*map,
288 	enum xfs_bmap_intent_type	type,
289 	int				whichfork,
290 	xfs_exntst_t			state)
291 {
292 	map->me_flags = 0;
293 	switch (type) {
294 	case XFS_BMAP_MAP:
295 	case XFS_BMAP_UNMAP:
296 		map->me_flags = type;
297 		break;
298 	default:
299 		ASSERT(0);
300 	}
301 	if (state == XFS_EXT_UNWRITTEN)
302 		map->me_flags |= XFS_BMAP_EXTENT_UNWRITTEN;
303 	if (whichfork == XFS_ATTR_FORK)
304 		map->me_flags |= XFS_BMAP_EXTENT_ATTR_FORK;
305 }
306 
307 /* Log bmap updates in the intent item. */
308 STATIC void
xfs_bmap_update_log_item(struct xfs_trans * tp,struct xfs_bui_log_item * buip,struct xfs_bmap_intent * bi)309 xfs_bmap_update_log_item(
310 	struct xfs_trans		*tp,
311 	struct xfs_bui_log_item		*buip,
312 	struct xfs_bmap_intent		*bi)
313 {
314 	uint				next_extent;
315 	struct xfs_map_extent		*map;
316 
317 	tp->t_flags |= XFS_TRANS_DIRTY;
318 	set_bit(XFS_LI_DIRTY, &buip->bui_item.li_flags);
319 
320 	/*
321 	 * atomic_inc_return gives us the value after the increment;
322 	 * we want to use it as an array index so we need to subtract 1 from
323 	 * it.
324 	 */
325 	next_extent = atomic_inc_return(&buip->bui_next_extent) - 1;
326 	ASSERT(next_extent < buip->bui_format.bui_nextents);
327 	map = &buip->bui_format.bui_extents[next_extent];
328 	map->me_owner = bi->bi_owner->i_ino;
329 	map->me_startblock = bi->bi_bmap.br_startblock;
330 	map->me_startoff = bi->bi_bmap.br_startoff;
331 	map->me_len = bi->bi_bmap.br_blockcount;
332 	xfs_trans_set_bmap_flags(map, bi->bi_type, bi->bi_whichfork,
333 			bi->bi_bmap.br_state);
334 }
335 
336 static struct xfs_log_item *
xfs_bmap_update_create_intent(struct xfs_trans * tp,struct list_head * items,unsigned int count,bool sort)337 xfs_bmap_update_create_intent(
338 	struct xfs_trans		*tp,
339 	struct list_head		*items,
340 	unsigned int			count,
341 	bool				sort)
342 {
343 	struct xfs_mount		*mp = tp->t_mountp;
344 	struct xfs_bui_log_item		*buip = xfs_bui_init(mp);
345 	struct xfs_bmap_intent		*bi;
346 
347 	ASSERT(count == XFS_BUI_MAX_FAST_EXTENTS);
348 
349 	xfs_trans_add_item(tp, &buip->bui_item);
350 	if (sort)
351 		list_sort(mp, items, xfs_bmap_update_diff_items);
352 	list_for_each_entry(bi, items, bi_list)
353 		xfs_bmap_update_log_item(tp, buip, bi);
354 	return &buip->bui_item;
355 }
356 
357 /* Get an BUD so we can process all the deferred rmap updates. */
358 static struct xfs_log_item *
xfs_bmap_update_create_done(struct xfs_trans * tp,struct xfs_log_item * intent,unsigned int count)359 xfs_bmap_update_create_done(
360 	struct xfs_trans		*tp,
361 	struct xfs_log_item		*intent,
362 	unsigned int			count)
363 {
364 	return &xfs_trans_get_bud(tp, BUI_ITEM(intent))->bud_item;
365 }
366 
367 /* Take a passive ref to the AG containing the space we're mapping. */
368 void
xfs_bmap_update_get_group(struct xfs_mount * mp,struct xfs_bmap_intent * bi)369 xfs_bmap_update_get_group(
370 	struct xfs_mount	*mp,
371 	struct xfs_bmap_intent	*bi)
372 {
373 	xfs_agnumber_t		agno;
374 
375 	agno = XFS_FSB_TO_AGNO(mp, bi->bi_bmap.br_startblock);
376 
377 	/*
378 	 * Bump the intent count on behalf of the deferred rmap and refcount
379 	 * intent items that that we can queue when we finish this bmap work.
380 	 * This new intent item will bump the intent count before the bmap
381 	 * intent drops the intent count, ensuring that the intent count
382 	 * remains nonzero across the transaction roll.
383 	 */
384 	bi->bi_pag = xfs_perag_intent_get(mp, agno);
385 }
386 
387 /* Release a passive AG ref after finishing mapping work. */
388 static inline void
xfs_bmap_update_put_group(struct xfs_bmap_intent * bi)389 xfs_bmap_update_put_group(
390 	struct xfs_bmap_intent	*bi)
391 {
392 	xfs_perag_intent_put(bi->bi_pag);
393 }
394 
395 /* Process a deferred rmap update. */
396 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)397 xfs_bmap_update_finish_item(
398 	struct xfs_trans		*tp,
399 	struct xfs_log_item		*done,
400 	struct list_head		*item,
401 	struct xfs_btree_cur		**state)
402 {
403 	struct xfs_bmap_intent		*bi;
404 	int				error;
405 
406 	bi = container_of(item, struct xfs_bmap_intent, bi_list);
407 
408 	error = xfs_trans_log_finish_bmap_update(tp, BUD_ITEM(done), bi);
409 	if (!error && bi->bi_bmap.br_blockcount > 0) {
410 		ASSERT(bi->bi_type == XFS_BMAP_UNMAP);
411 		return -EAGAIN;
412 	}
413 
414 	xfs_bmap_update_put_group(bi);
415 	kmem_cache_free(xfs_bmap_intent_cache, bi);
416 	return error;
417 }
418 
419 /* Abort all pending BUIs. */
420 STATIC void
xfs_bmap_update_abort_intent(struct xfs_log_item * intent)421 xfs_bmap_update_abort_intent(
422 	struct xfs_log_item		*intent)
423 {
424 	xfs_bui_release(BUI_ITEM(intent));
425 }
426 
427 /* Cancel a deferred bmap update. */
428 STATIC void
xfs_bmap_update_cancel_item(struct list_head * item)429 xfs_bmap_update_cancel_item(
430 	struct list_head		*item)
431 {
432 	struct xfs_bmap_intent		*bi;
433 
434 	bi = container_of(item, struct xfs_bmap_intent, bi_list);
435 
436 	xfs_bmap_update_put_group(bi);
437 	kmem_cache_free(xfs_bmap_intent_cache, bi);
438 }
439 
440 const struct xfs_defer_op_type xfs_bmap_update_defer_type = {
441 	.max_items	= XFS_BUI_MAX_FAST_EXTENTS,
442 	.create_intent	= xfs_bmap_update_create_intent,
443 	.abort_intent	= xfs_bmap_update_abort_intent,
444 	.create_done	= xfs_bmap_update_create_done,
445 	.finish_item	= xfs_bmap_update_finish_item,
446 	.cancel_item	= xfs_bmap_update_cancel_item,
447 };
448 
449 /* Is this recovered BUI ok? */
450 static inline bool
xfs_bui_validate(struct xfs_mount * mp,struct xfs_bui_log_item * buip)451 xfs_bui_validate(
452 	struct xfs_mount		*mp,
453 	struct xfs_bui_log_item		*buip)
454 {
455 	struct xfs_map_extent		*map;
456 
457 	/* Only one mapping operation per BUI... */
458 	if (buip->bui_format.bui_nextents != XFS_BUI_MAX_FAST_EXTENTS)
459 		return false;
460 
461 	map = &buip->bui_format.bui_extents[0];
462 
463 	if (map->me_flags & ~XFS_BMAP_EXTENT_FLAGS)
464 		return false;
465 
466 	switch (map->me_flags & XFS_BMAP_EXTENT_TYPE_MASK) {
467 	case XFS_BMAP_MAP:
468 	case XFS_BMAP_UNMAP:
469 		break;
470 	default:
471 		return false;
472 	}
473 
474 	if (!xfs_verify_ino(mp, map->me_owner))
475 		return false;
476 
477 	if (!xfs_verify_fileext(mp, map->me_startoff, map->me_len))
478 		return false;
479 
480 	return xfs_verify_fsbext(mp, map->me_startblock, map->me_len);
481 }
482 
483 /*
484  * Process a bmap update intent item that was recovered from the log.
485  * We need to update some inode's bmbt.
486  */
487 STATIC int
xfs_bui_item_recover(struct xfs_log_item * lip,struct list_head * capture_list)488 xfs_bui_item_recover(
489 	struct xfs_log_item		*lip,
490 	struct list_head		*capture_list)
491 {
492 	struct xfs_bmap_intent		fake = { };
493 	struct xfs_trans_res		resv;
494 	struct xfs_bui_log_item		*buip = BUI_ITEM(lip);
495 	struct xfs_trans		*tp;
496 	struct xfs_inode		*ip = NULL;
497 	struct xfs_mount		*mp = lip->li_log->l_mp;
498 	struct xfs_map_extent		*map;
499 	struct xfs_bud_log_item		*budp;
500 	int				iext_delta;
501 	int				error = 0;
502 
503 	if (!xfs_bui_validate(mp, buip)) {
504 		XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
505 				&buip->bui_format, sizeof(buip->bui_format));
506 		return -EFSCORRUPTED;
507 	}
508 
509 	map = &buip->bui_format.bui_extents[0];
510 	fake.bi_whichfork = (map->me_flags & XFS_BMAP_EXTENT_ATTR_FORK) ?
511 			XFS_ATTR_FORK : XFS_DATA_FORK;
512 	fake.bi_type = map->me_flags & XFS_BMAP_EXTENT_TYPE_MASK;
513 
514 	error = xlog_recover_iget(mp, map->me_owner, &ip);
515 	if (error)
516 		return error;
517 
518 	/* Allocate transaction and do the work. */
519 	resv = xlog_recover_resv(&M_RES(mp)->tr_itruncate);
520 	error = xfs_trans_alloc(mp, &resv,
521 			XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK), 0, 0, &tp);
522 	if (error)
523 		goto err_rele;
524 
525 	budp = xfs_trans_get_bud(tp, buip);
526 	xfs_ilock(ip, XFS_ILOCK_EXCL);
527 	xfs_trans_ijoin(tp, ip, 0);
528 
529 	if (fake.bi_type == XFS_BMAP_MAP)
530 		iext_delta = XFS_IEXT_ADD_NOSPLIT_CNT;
531 	else
532 		iext_delta = XFS_IEXT_PUNCH_HOLE_CNT;
533 
534 	error = xfs_iext_count_may_overflow(ip, fake.bi_whichfork, iext_delta);
535 	if (error == -EFBIG)
536 		error = xfs_iext_count_upgrade(tp, ip, iext_delta);
537 	if (error)
538 		goto err_cancel;
539 
540 	fake.bi_owner = ip;
541 	fake.bi_bmap.br_startblock = map->me_startblock;
542 	fake.bi_bmap.br_startoff = map->me_startoff;
543 	fake.bi_bmap.br_blockcount = map->me_len;
544 	fake.bi_bmap.br_state = (map->me_flags & XFS_BMAP_EXTENT_UNWRITTEN) ?
545 			XFS_EXT_UNWRITTEN : XFS_EXT_NORM;
546 
547 	xfs_bmap_update_get_group(mp, &fake);
548 	error = xfs_trans_log_finish_bmap_update(tp, budp, &fake);
549 	if (error == -EFSCORRUPTED)
550 		XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, map,
551 				sizeof(*map));
552 	xfs_bmap_update_put_group(&fake);
553 	if (error)
554 		goto err_cancel;
555 
556 	if (fake.bi_bmap.br_blockcount > 0) {
557 		ASSERT(fake.bi_type == XFS_BMAP_UNMAP);
558 		xfs_bmap_unmap_extent(tp, ip, &fake.bi_bmap);
559 	}
560 
561 	/*
562 	 * Commit transaction, which frees the transaction and saves the inode
563 	 * for later replay activities.
564 	 */
565 	error = xfs_defer_ops_capture_and_commit(tp, capture_list);
566 	if (error)
567 		goto err_unlock;
568 
569 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
570 	xfs_irele(ip);
571 	return 0;
572 
573 err_cancel:
574 	xfs_trans_cancel(tp);
575 err_unlock:
576 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
577 err_rele:
578 	xfs_irele(ip);
579 	return error;
580 }
581 
582 STATIC bool
xfs_bui_item_match(struct xfs_log_item * lip,uint64_t intent_id)583 xfs_bui_item_match(
584 	struct xfs_log_item	*lip,
585 	uint64_t		intent_id)
586 {
587 	return BUI_ITEM(lip)->bui_format.bui_id == intent_id;
588 }
589 
590 /* Relog an intent item to push the log tail forward. */
591 static struct xfs_log_item *
xfs_bui_item_relog(struct xfs_log_item * intent,struct xfs_trans * tp)592 xfs_bui_item_relog(
593 	struct xfs_log_item		*intent,
594 	struct xfs_trans		*tp)
595 {
596 	struct xfs_bud_log_item		*budp;
597 	struct xfs_bui_log_item		*buip;
598 	struct xfs_map_extent		*map;
599 	unsigned int			count;
600 
601 	count = BUI_ITEM(intent)->bui_format.bui_nextents;
602 	map = BUI_ITEM(intent)->bui_format.bui_extents;
603 
604 	tp->t_flags |= XFS_TRANS_DIRTY;
605 	budp = xfs_trans_get_bud(tp, BUI_ITEM(intent));
606 	set_bit(XFS_LI_DIRTY, &budp->bud_item.li_flags);
607 
608 	buip = xfs_bui_init(tp->t_mountp);
609 	memcpy(buip->bui_format.bui_extents, map, count * sizeof(*map));
610 	atomic_set(&buip->bui_next_extent, count);
611 	xfs_trans_add_item(tp, &buip->bui_item);
612 	set_bit(XFS_LI_DIRTY, &buip->bui_item.li_flags);
613 	return &buip->bui_item;
614 }
615 
616 static const struct xfs_item_ops xfs_bui_item_ops = {
617 	.flags		= XFS_ITEM_INTENT,
618 	.iop_size	= xfs_bui_item_size,
619 	.iop_format	= xfs_bui_item_format,
620 	.iop_unpin	= xfs_bui_item_unpin,
621 	.iop_release	= xfs_bui_item_release,
622 	.iop_recover	= xfs_bui_item_recover,
623 	.iop_match	= xfs_bui_item_match,
624 	.iop_relog	= xfs_bui_item_relog,
625 };
626 
627 static inline void
xfs_bui_copy_format(struct xfs_bui_log_format * dst,const struct xfs_bui_log_format * src)628 xfs_bui_copy_format(
629 	struct xfs_bui_log_format	*dst,
630 	const struct xfs_bui_log_format	*src)
631 {
632 	unsigned int			i;
633 
634 	memcpy(dst, src, offsetof(struct xfs_bui_log_format, bui_extents));
635 
636 	for (i = 0; i < src->bui_nextents; i++)
637 		memcpy(&dst->bui_extents[i], &src->bui_extents[i],
638 				sizeof(struct xfs_map_extent));
639 }
640 
641 /*
642  * This routine is called to create an in-core extent bmap update
643  * item from the bui format structure which was logged on disk.
644  * It allocates an in-core bui, copies the extents from the format
645  * structure into it, and adds the bui to the AIL with the given
646  * LSN.
647  */
648 STATIC int
xlog_recover_bui_commit_pass2(struct xlog * log,struct list_head * buffer_list,struct xlog_recover_item * item,xfs_lsn_t lsn)649 xlog_recover_bui_commit_pass2(
650 	struct xlog			*log,
651 	struct list_head		*buffer_list,
652 	struct xlog_recover_item	*item,
653 	xfs_lsn_t			lsn)
654 {
655 	struct xfs_mount		*mp = log->l_mp;
656 	struct xfs_bui_log_item		*buip;
657 	struct xfs_bui_log_format	*bui_formatp;
658 	size_t				len;
659 
660 	bui_formatp = item->ri_buf[0].i_addr;
661 
662 	if (item->ri_buf[0].i_len < xfs_bui_log_format_sizeof(0)) {
663 		XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
664 				item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
665 		return -EFSCORRUPTED;
666 	}
667 
668 	if (bui_formatp->bui_nextents != XFS_BUI_MAX_FAST_EXTENTS) {
669 		XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
670 				item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
671 		return -EFSCORRUPTED;
672 	}
673 
674 	len = xfs_bui_log_format_sizeof(bui_formatp->bui_nextents);
675 	if (item->ri_buf[0].i_len != len) {
676 		XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
677 				item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
678 		return -EFSCORRUPTED;
679 	}
680 
681 	buip = xfs_bui_init(mp);
682 	xfs_bui_copy_format(&buip->bui_format, bui_formatp);
683 	atomic_set(&buip->bui_next_extent, bui_formatp->bui_nextents);
684 	/*
685 	 * Insert the intent into the AIL directly and drop one reference so
686 	 * that finishing or canceling the work will drop the other.
687 	 */
688 	xfs_trans_ail_insert(log->l_ailp, &buip->bui_item, lsn);
689 	xfs_bui_release(buip);
690 	return 0;
691 }
692 
693 const struct xlog_recover_item_ops xlog_bui_item_ops = {
694 	.item_type		= XFS_LI_BUI,
695 	.commit_pass2		= xlog_recover_bui_commit_pass2,
696 };
697 
698 /*
699  * This routine is called when an BUD format structure is found in a committed
700  * transaction in the log. Its purpose is to cancel the corresponding BUI if it
701  * was still in the log. To do this it searches the AIL for the BUI with an id
702  * equal to that in the BUD format structure. If we find it we drop the BUD
703  * reference, which removes the BUI from the AIL and frees it.
704  */
705 STATIC int
xlog_recover_bud_commit_pass2(struct xlog * log,struct list_head * buffer_list,struct xlog_recover_item * item,xfs_lsn_t lsn)706 xlog_recover_bud_commit_pass2(
707 	struct xlog			*log,
708 	struct list_head		*buffer_list,
709 	struct xlog_recover_item	*item,
710 	xfs_lsn_t			lsn)
711 {
712 	struct xfs_bud_log_format	*bud_formatp;
713 
714 	bud_formatp = item->ri_buf[0].i_addr;
715 	if (item->ri_buf[0].i_len != sizeof(struct xfs_bud_log_format)) {
716 		XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, log->l_mp,
717 				item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
718 		return -EFSCORRUPTED;
719 	}
720 
721 	xlog_recover_release_intent(log, XFS_LI_BUI, bud_formatp->bud_bui_id);
722 	return 0;
723 }
724 
725 const struct xlog_recover_item_ops xlog_bud_item_ops = {
726 	.item_type		= XFS_LI_BUD,
727 	.commit_pass2		= xlog_recover_bud_commit_pass2,
728 };
729