Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2001,2005 Silicon Graphics, Inc.
4 * All Rights Reserved.
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_ag.h"
15#include "xfs_defer.h"
16#include "xfs_trans.h"
17#include "xfs_trans_priv.h"
18#include "xfs_extfree_item.h"
19#include "xfs_log.h"
20#include "xfs_btree.h"
21#include "xfs_rmap.h"
22#include "xfs_alloc.h"
23#include "xfs_bmap.h"
24#include "xfs_trace.h"
25#include "xfs_error.h"
26#include "xfs_log_priv.h"
27#include "xfs_log_recover.h"
28#include "xfs_rtalloc.h"
29#include "xfs_inode.h"
30#include "xfs_rtbitmap.h"
31#include "xfs_rtgroup.h"
32
33struct kmem_cache *xfs_efi_cache;
34struct kmem_cache *xfs_efd_cache;
35
36static const struct xfs_item_ops xfs_efi_item_ops;
37
38static inline struct xfs_efi_log_item *EFI_ITEM(struct xfs_log_item *lip)
39{
40 return container_of(lip, struct xfs_efi_log_item, efi_item);
41}
42
43STATIC void
44xfs_efi_item_free(
45 struct xfs_efi_log_item *efip)
46{
47 kvfree(efip->efi_item.li_lv_shadow);
48 if (efip->efi_format.efi_nextents > XFS_EFI_MAX_FAST_EXTENTS)
49 kfree(efip);
50 else
51 kmem_cache_free(xfs_efi_cache, efip);
52}
53
54/*
55 * Freeing the efi requires that we remove it from the AIL if it has already
56 * been placed there. However, the EFI may not yet have been placed in the AIL
57 * when called by xfs_efi_release() from EFD processing due to the ordering of
58 * committed vs unpin operations in bulk insert operations. Hence the reference
59 * count to ensure only the last caller frees the EFI.
60 */
61STATIC void
62xfs_efi_release(
63 struct xfs_efi_log_item *efip)
64{
65 ASSERT(atomic_read(&efip->efi_refcount) > 0);
66 if (!atomic_dec_and_test(&efip->efi_refcount))
67 return;
68
69 xfs_trans_ail_delete(&efip->efi_item, 0);
70 xfs_efi_item_free(efip);
71}
72
73STATIC void
74xfs_efi_item_size(
75 struct xfs_log_item *lip,
76 int *nvecs,
77 int *nbytes)
78{
79 struct xfs_efi_log_item *efip = EFI_ITEM(lip);
80
81 *nvecs += 1;
82 *nbytes += xfs_efi_log_format_sizeof(efip->efi_format.efi_nextents);
83}
84
85/*
86 * This is called to fill in the vector of log iovecs for the
87 * given efi log item. We use only 1 iovec, and we point that
88 * at the efi_log_format structure embedded in the efi item.
89 * It is at this point that we assert that all of the extent
90 * slots in the efi item have been filled.
91 */
92STATIC void
93xfs_efi_item_format(
94 struct xfs_log_item *lip,
95 struct xfs_log_vec *lv)
96{
97 struct xfs_efi_log_item *efip = EFI_ITEM(lip);
98 struct xfs_log_iovec *vecp = NULL;
99
100 ASSERT(atomic_read(&efip->efi_next_extent) ==
101 efip->efi_format.efi_nextents);
102 ASSERT(lip->li_type == XFS_LI_EFI || lip->li_type == XFS_LI_EFI_RT);
103
104 efip->efi_format.efi_type = lip->li_type;
105 efip->efi_format.efi_size = 1;
106
107 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_EFI_FORMAT, &efip->efi_format,
108 xfs_efi_log_format_sizeof(efip->efi_format.efi_nextents));
109}
110
111/*
112 * The unpin operation is the last place an EFI is manipulated in the log. It is
113 * either inserted in the AIL or aborted in the event of a log I/O error. In
114 * either case, the EFI transaction has been successfully committed to make it
115 * this far. Therefore, we expect whoever committed the EFI to either construct
116 * and commit the EFD or drop the EFD's reference in the event of error. Simply
117 * drop the log's EFI reference now that the log is done with it.
118 */
119STATIC void
120xfs_efi_item_unpin(
121 struct xfs_log_item *lip,
122 int remove)
123{
124 struct xfs_efi_log_item *efip = EFI_ITEM(lip);
125 xfs_efi_release(efip);
126}
127
128/*
129 * The EFI has been either committed or aborted if the transaction has been
130 * cancelled. If the transaction was cancelled, an EFD isn't going to be
131 * constructed and thus we free the EFI here directly.
132 */
133STATIC void
134xfs_efi_item_release(
135 struct xfs_log_item *lip)
136{
137 xfs_efi_release(EFI_ITEM(lip));
138}
139
140/*
141 * Allocate and initialize an efi item with the given number of extents.
142 */
143STATIC struct xfs_efi_log_item *
144xfs_efi_init(
145 struct xfs_mount *mp,
146 unsigned short item_type,
147 uint nextents)
148{
149 struct xfs_efi_log_item *efip;
150
151 ASSERT(item_type == XFS_LI_EFI || item_type == XFS_LI_EFI_RT);
152 ASSERT(nextents > 0);
153
154 if (nextents > XFS_EFI_MAX_FAST_EXTENTS) {
155 efip = kzalloc(xfs_efi_log_item_sizeof(nextents),
156 GFP_KERNEL | __GFP_NOFAIL);
157 } else {
158 efip = kmem_cache_zalloc(xfs_efi_cache,
159 GFP_KERNEL | __GFP_NOFAIL);
160 }
161
162 xfs_log_item_init(mp, &efip->efi_item, item_type, &xfs_efi_item_ops);
163 efip->efi_format.efi_nextents = nextents;
164 efip->efi_format.efi_id = (uintptr_t)(void *)efip;
165 atomic_set(&efip->efi_next_extent, 0);
166 atomic_set(&efip->efi_refcount, 2);
167
168 return efip;
169}
170
171/*
172 * Copy an EFI format buffer from the given buf, and into the destination
173 * EFI format structure.
174 * The given buffer can be in 32 bit or 64 bit form (which has different padding),
175 * one of which will be the native format for this kernel.
176 * It will handle the conversion of formats if necessary.
177 */
178STATIC int
179xfs_efi_copy_format(xfs_log_iovec_t *buf, xfs_efi_log_format_t *dst_efi_fmt)
180{
181 xfs_efi_log_format_t *src_efi_fmt = buf->i_addr;
182 uint i;
183 uint len = xfs_efi_log_format_sizeof(src_efi_fmt->efi_nextents);
184 uint len32 = xfs_efi_log_format32_sizeof(src_efi_fmt->efi_nextents);
185 uint len64 = xfs_efi_log_format64_sizeof(src_efi_fmt->efi_nextents);
186
187 if (buf->i_len == len) {
188 memcpy(dst_efi_fmt, src_efi_fmt,
189 offsetof(struct xfs_efi_log_format, efi_extents));
190 for (i = 0; i < src_efi_fmt->efi_nextents; i++)
191 memcpy(&dst_efi_fmt->efi_extents[i],
192 &src_efi_fmt->efi_extents[i],
193 sizeof(struct xfs_extent));
194 return 0;
195 } else if (buf->i_len == len32) {
196 xfs_efi_log_format_32_t *src_efi_fmt_32 = buf->i_addr;
197
198 dst_efi_fmt->efi_type = src_efi_fmt_32->efi_type;
199 dst_efi_fmt->efi_size = src_efi_fmt_32->efi_size;
200 dst_efi_fmt->efi_nextents = src_efi_fmt_32->efi_nextents;
201 dst_efi_fmt->efi_id = src_efi_fmt_32->efi_id;
202 for (i = 0; i < dst_efi_fmt->efi_nextents; i++) {
203 dst_efi_fmt->efi_extents[i].ext_start =
204 src_efi_fmt_32->efi_extents[i].ext_start;
205 dst_efi_fmt->efi_extents[i].ext_len =
206 src_efi_fmt_32->efi_extents[i].ext_len;
207 }
208 return 0;
209 } else if (buf->i_len == len64) {
210 xfs_efi_log_format_64_t *src_efi_fmt_64 = buf->i_addr;
211
212 dst_efi_fmt->efi_type = src_efi_fmt_64->efi_type;
213 dst_efi_fmt->efi_size = src_efi_fmt_64->efi_size;
214 dst_efi_fmt->efi_nextents = src_efi_fmt_64->efi_nextents;
215 dst_efi_fmt->efi_id = src_efi_fmt_64->efi_id;
216 for (i = 0; i < dst_efi_fmt->efi_nextents; i++) {
217 dst_efi_fmt->efi_extents[i].ext_start =
218 src_efi_fmt_64->efi_extents[i].ext_start;
219 dst_efi_fmt->efi_extents[i].ext_len =
220 src_efi_fmt_64->efi_extents[i].ext_len;
221 }
222 return 0;
223 }
224 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, NULL, buf->i_addr,
225 buf->i_len);
226 return -EFSCORRUPTED;
227}
228
229static inline struct xfs_efd_log_item *EFD_ITEM(struct xfs_log_item *lip)
230{
231 return container_of(lip, struct xfs_efd_log_item, efd_item);
232}
233
234STATIC void
235xfs_efd_item_free(struct xfs_efd_log_item *efdp)
236{
237 kvfree(efdp->efd_item.li_lv_shadow);
238 if (efdp->efd_format.efd_nextents > XFS_EFD_MAX_FAST_EXTENTS)
239 kfree(efdp);
240 else
241 kmem_cache_free(xfs_efd_cache, efdp);
242}
243
244STATIC void
245xfs_efd_item_size(
246 struct xfs_log_item *lip,
247 int *nvecs,
248 int *nbytes)
249{
250 struct xfs_efd_log_item *efdp = EFD_ITEM(lip);
251
252 *nvecs += 1;
253 *nbytes += xfs_efd_log_format_sizeof(efdp->efd_format.efd_nextents);
254}
255
256/*
257 * This is called to fill in the vector of log iovecs for the
258 * given efd log item. We use only 1 iovec, and we point that
259 * at the efd_log_format structure embedded in the efd item.
260 * It is at this point that we assert that all of the extent
261 * slots in the efd item have been filled.
262 */
263STATIC void
264xfs_efd_item_format(
265 struct xfs_log_item *lip,
266 struct xfs_log_vec *lv)
267{
268 struct xfs_efd_log_item *efdp = EFD_ITEM(lip);
269 struct xfs_log_iovec *vecp = NULL;
270
271 ASSERT(efdp->efd_next_extent == efdp->efd_format.efd_nextents);
272 ASSERT(lip->li_type == XFS_LI_EFD || lip->li_type == XFS_LI_EFD_RT);
273
274 efdp->efd_format.efd_type = lip->li_type;
275 efdp->efd_format.efd_size = 1;
276
277 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_EFD_FORMAT, &efdp->efd_format,
278 xfs_efd_log_format_sizeof(efdp->efd_format.efd_nextents));
279}
280
281/*
282 * The EFD is either committed or aborted if the transaction is cancelled. If
283 * the transaction is cancelled, drop our reference to the EFI and free the EFD.
284 */
285STATIC void
286xfs_efd_item_release(
287 struct xfs_log_item *lip)
288{
289 struct xfs_efd_log_item *efdp = EFD_ITEM(lip);
290
291 xfs_efi_release(efdp->efd_efip);
292 xfs_efd_item_free(efdp);
293}
294
295static struct xfs_log_item *
296xfs_efd_item_intent(
297 struct xfs_log_item *lip)
298{
299 return &EFD_ITEM(lip)->efd_efip->efi_item;
300}
301
302static const struct xfs_item_ops xfs_efd_item_ops = {
303 .flags = XFS_ITEM_RELEASE_WHEN_COMMITTED |
304 XFS_ITEM_INTENT_DONE,
305 .iop_size = xfs_efd_item_size,
306 .iop_format = xfs_efd_item_format,
307 .iop_release = xfs_efd_item_release,
308 .iop_intent = xfs_efd_item_intent,
309};
310
311static inline struct xfs_extent_free_item *xefi_entry(const struct list_head *e)
312{
313 return list_entry(e, struct xfs_extent_free_item, xefi_list);
314}
315
316static inline bool
317xfs_efi_item_isrt(const struct xfs_log_item *lip)
318{
319 ASSERT(lip->li_type == XFS_LI_EFI || lip->li_type == XFS_LI_EFI_RT);
320
321 return lip->li_type == XFS_LI_EFI_RT;
322}
323
324/*
325 * Fill the EFD with all extents from the EFI when we need to roll the
326 * transaction and continue with a new EFI.
327 *
328 * This simply copies all the extents in the EFI to the EFD rather than make
329 * assumptions about which extents in the EFI have already been processed. We
330 * currently keep the xefi list in the same order as the EFI extent list, but
331 * that may not always be the case. Copying everything avoids leaving a landmine
332 * were we fail to cancel all the extents in an EFI if the xefi list is
333 * processed in a different order to the extents in the EFI.
334 */
335static void
336xfs_efd_from_efi(
337 struct xfs_efd_log_item *efdp)
338{
339 struct xfs_efi_log_item *efip = efdp->efd_efip;
340 uint i;
341
342 ASSERT(efip->efi_format.efi_nextents > 0);
343 ASSERT(efdp->efd_next_extent < efip->efi_format.efi_nextents);
344
345 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
346 efdp->efd_format.efd_extents[i] =
347 efip->efi_format.efi_extents[i];
348 }
349 efdp->efd_next_extent = efip->efi_format.efi_nextents;
350}
351
352static void
353xfs_efd_add_extent(
354 struct xfs_efd_log_item *efdp,
355 struct xfs_extent_free_item *xefi)
356{
357 struct xfs_extent *extp;
358
359 ASSERT(efdp->efd_next_extent < efdp->efd_format.efd_nextents);
360
361 extp = &efdp->efd_format.efd_extents[efdp->efd_next_extent];
362 extp->ext_start = xefi->xefi_startblock;
363 extp->ext_len = xefi->xefi_blockcount;
364
365 efdp->efd_next_extent++;
366}
367
368/* Sort bmap items by AG. */
369static int
370xfs_extent_free_diff_items(
371 void *priv,
372 const struct list_head *a,
373 const struct list_head *b)
374{
375 struct xfs_extent_free_item *ra = xefi_entry(a);
376 struct xfs_extent_free_item *rb = xefi_entry(b);
377
378 return ra->xefi_group->xg_gno - rb->xefi_group->xg_gno;
379}
380
381/* Log a free extent to the intent item. */
382STATIC void
383xfs_extent_free_log_item(
384 struct xfs_trans *tp,
385 struct xfs_efi_log_item *efip,
386 struct xfs_extent_free_item *xefi)
387{
388 uint next_extent;
389 struct xfs_extent *extp;
390
391 /*
392 * atomic_inc_return gives us the value after the increment;
393 * we want to use it as an array index so we need to subtract 1 from
394 * it.
395 */
396 next_extent = atomic_inc_return(&efip->efi_next_extent) - 1;
397 ASSERT(next_extent < efip->efi_format.efi_nextents);
398 extp = &efip->efi_format.efi_extents[next_extent];
399 extp->ext_start = xefi->xefi_startblock;
400 extp->ext_len = xefi->xefi_blockcount;
401}
402
403static struct xfs_log_item *
404__xfs_extent_free_create_intent(
405 struct xfs_trans *tp,
406 struct list_head *items,
407 unsigned int count,
408 bool sort,
409 unsigned short item_type)
410{
411 struct xfs_mount *mp = tp->t_mountp;
412 struct xfs_efi_log_item *efip;
413 struct xfs_extent_free_item *xefi;
414
415 ASSERT(count > 0);
416
417 efip = xfs_efi_init(mp, item_type, count);
418 if (sort)
419 list_sort(mp, items, xfs_extent_free_diff_items);
420 list_for_each_entry(xefi, items, xefi_list)
421 xfs_extent_free_log_item(tp, efip, xefi);
422 return &efip->efi_item;
423}
424
425static struct xfs_log_item *
426xfs_extent_free_create_intent(
427 struct xfs_trans *tp,
428 struct list_head *items,
429 unsigned int count,
430 bool sort)
431{
432 return __xfs_extent_free_create_intent(tp, items, count, sort,
433 XFS_LI_EFI);
434}
435
436static inline unsigned short
437xfs_efd_type_from_efi(const struct xfs_efi_log_item *efip)
438{
439 return xfs_efi_item_isrt(&efip->efi_item) ? XFS_LI_EFD_RT : XFS_LI_EFD;
440}
441
442/* Get an EFD so we can process all the free extents. */
443static struct xfs_log_item *
444xfs_extent_free_create_done(
445 struct xfs_trans *tp,
446 struct xfs_log_item *intent,
447 unsigned int count)
448{
449 struct xfs_efi_log_item *efip = EFI_ITEM(intent);
450 struct xfs_efd_log_item *efdp;
451
452 ASSERT(count > 0);
453
454 if (count > XFS_EFD_MAX_FAST_EXTENTS) {
455 efdp = kzalloc(xfs_efd_log_item_sizeof(count),
456 GFP_KERNEL | __GFP_NOFAIL);
457 } else {
458 efdp = kmem_cache_zalloc(xfs_efd_cache,
459 GFP_KERNEL | __GFP_NOFAIL);
460 }
461
462 xfs_log_item_init(tp->t_mountp, &efdp->efd_item,
463 xfs_efd_type_from_efi(efip), &xfs_efd_item_ops);
464 efdp->efd_efip = efip;
465 efdp->efd_format.efd_nextents = count;
466 efdp->efd_format.efd_efi_id = efip->efi_format.efi_id;
467
468 return &efdp->efd_item;
469}
470
471static inline const struct xfs_defer_op_type *
472xefi_ops(
473 struct xfs_extent_free_item *xefi)
474{
475 if (xfs_efi_is_realtime(xefi))
476 return &xfs_rtextent_free_defer_type;
477 if (xefi->xefi_agresv == XFS_AG_RESV_AGFL)
478 return &xfs_agfl_free_defer_type;
479 return &xfs_extent_free_defer_type;
480}
481
482/* Add this deferred EFI to the transaction. */
483void
484xfs_extent_free_defer_add(
485 struct xfs_trans *tp,
486 struct xfs_extent_free_item *xefi,
487 struct xfs_defer_pending **dfpp)
488{
489 struct xfs_mount *mp = tp->t_mountp;
490
491 xefi->xefi_group = xfs_group_intent_get(mp, xefi->xefi_startblock,
492 xfs_efi_is_realtime(xefi) ? XG_TYPE_RTG : XG_TYPE_AG);
493
494 trace_xfs_extent_free_defer(mp, xefi);
495 *dfpp = xfs_defer_add(tp, &xefi->xefi_list, xefi_ops(xefi));
496}
497
498/* Cancel a free extent. */
499STATIC void
500xfs_extent_free_cancel_item(
501 struct list_head *item)
502{
503 struct xfs_extent_free_item *xefi = xefi_entry(item);
504
505 xfs_group_intent_put(xefi->xefi_group);
506 kmem_cache_free(xfs_extfree_item_cache, xefi);
507}
508
509/* Process a free extent. */
510STATIC int
511xfs_extent_free_finish_item(
512 struct xfs_trans *tp,
513 struct xfs_log_item *done,
514 struct list_head *item,
515 struct xfs_btree_cur **state)
516{
517 struct xfs_owner_info oinfo = { };
518 struct xfs_extent_free_item *xefi = xefi_entry(item);
519 struct xfs_efd_log_item *efdp = EFD_ITEM(done);
520 struct xfs_mount *mp = tp->t_mountp;
521 xfs_agblock_t agbno;
522 int error = 0;
523
524 agbno = XFS_FSB_TO_AGBNO(mp, xefi->xefi_startblock);
525
526 oinfo.oi_owner = xefi->xefi_owner;
527 if (xefi->xefi_flags & XFS_EFI_ATTR_FORK)
528 oinfo.oi_flags |= XFS_OWNER_INFO_ATTR_FORK;
529 if (xefi->xefi_flags & XFS_EFI_BMBT_BLOCK)
530 oinfo.oi_flags |= XFS_OWNER_INFO_BMBT_BLOCK;
531
532 trace_xfs_extent_free_deferred(mp, xefi);
533
534 /*
535 * If we need a new transaction to make progress, the caller will log a
536 * new EFI with the current contents. It will also log an EFD to cancel
537 * the existing EFI, and so we need to copy all the unprocessed extents
538 * in this EFI to the EFD so this works correctly.
539 */
540 if (!(xefi->xefi_flags & XFS_EFI_CANCELLED))
541 error = __xfs_free_extent(tp, to_perag(xefi->xefi_group), agbno,
542 xefi->xefi_blockcount, &oinfo, xefi->xefi_agresv,
543 xefi->xefi_flags & XFS_EFI_SKIP_DISCARD);
544 if (error == -EAGAIN) {
545 xfs_efd_from_efi(efdp);
546 return error;
547 }
548
549 xfs_efd_add_extent(efdp, xefi);
550 xfs_extent_free_cancel_item(item);
551 return error;
552}
553
554/* Abort all pending EFIs. */
555STATIC void
556xfs_extent_free_abort_intent(
557 struct xfs_log_item *intent)
558{
559 xfs_efi_release(EFI_ITEM(intent));
560}
561
562/*
563 * AGFL blocks are accounted differently in the reserve pools and are not
564 * inserted into the busy extent list.
565 */
566STATIC int
567xfs_agfl_free_finish_item(
568 struct xfs_trans *tp,
569 struct xfs_log_item *done,
570 struct list_head *item,
571 struct xfs_btree_cur **state)
572{
573 struct xfs_owner_info oinfo = { };
574 struct xfs_mount *mp = tp->t_mountp;
575 struct xfs_efd_log_item *efdp = EFD_ITEM(done);
576 struct xfs_extent_free_item *xefi = xefi_entry(item);
577 struct xfs_buf *agbp;
578 int error;
579 xfs_agblock_t agbno;
580
581 ASSERT(xefi->xefi_blockcount == 1);
582 agbno = XFS_FSB_TO_AGBNO(mp, xefi->xefi_startblock);
583 oinfo.oi_owner = xefi->xefi_owner;
584
585 trace_xfs_agfl_free_deferred(mp, xefi);
586
587 error = xfs_alloc_read_agf(to_perag(xefi->xefi_group), tp, 0, &agbp);
588 if (!error)
589 error = xfs_free_ag_extent(tp, agbp, agbno, 1, &oinfo,
590 XFS_AG_RESV_AGFL);
591
592 xfs_efd_add_extent(efdp, xefi);
593 xfs_extent_free_cancel_item(&xefi->xefi_list);
594 return error;
595}
596
597/* Is this recovered EFI ok? */
598static inline bool
599xfs_efi_validate_ext(
600 struct xfs_mount *mp,
601 bool isrt,
602 struct xfs_extent *extp)
603{
604 if (isrt)
605 return xfs_verify_rtbext(mp, extp->ext_start, extp->ext_len);
606
607 return xfs_verify_fsbext(mp, extp->ext_start, extp->ext_len);
608}
609
610static inline void
611xfs_efi_recover_work(
612 struct xfs_mount *mp,
613 struct xfs_defer_pending *dfp,
614 bool isrt,
615 struct xfs_extent *extp)
616{
617 struct xfs_extent_free_item *xefi;
618
619 xefi = kmem_cache_zalloc(xfs_extfree_item_cache,
620 GFP_KERNEL | __GFP_NOFAIL);
621 xefi->xefi_startblock = extp->ext_start;
622 xefi->xefi_blockcount = extp->ext_len;
623 xefi->xefi_agresv = XFS_AG_RESV_NONE;
624 xefi->xefi_owner = XFS_RMAP_OWN_UNKNOWN;
625 xefi->xefi_group = xfs_group_intent_get(mp, extp->ext_start,
626 isrt ? XG_TYPE_RTG : XG_TYPE_AG);
627 if (isrt)
628 xefi->xefi_flags |= XFS_EFI_REALTIME;
629
630 xfs_defer_add_item(dfp, &xefi->xefi_list);
631}
632
633/*
634 * Process an extent free intent item that was recovered from
635 * the log. We need to free the extents that it describes.
636 */
637STATIC int
638xfs_extent_free_recover_work(
639 struct xfs_defer_pending *dfp,
640 struct list_head *capture_list)
641{
642 struct xfs_trans_res resv;
643 struct xfs_log_item *lip = dfp->dfp_intent;
644 struct xfs_efi_log_item *efip = EFI_ITEM(lip);
645 struct xfs_mount *mp = lip->li_log->l_mp;
646 struct xfs_trans *tp;
647 int i;
648 int error = 0;
649 bool isrt = xfs_efi_item_isrt(lip);
650
651 /*
652 * First check the validity of the extents described by the EFI. If
653 * any are bad, then assume that all are bad and just toss the EFI.
654 * Mixing RT and non-RT extents in the same EFI item is not allowed.
655 */
656 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
657 if (!xfs_efi_validate_ext(mp, isrt,
658 &efip->efi_format.efi_extents[i])) {
659 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
660 &efip->efi_format,
661 sizeof(efip->efi_format));
662 return -EFSCORRUPTED;
663 }
664
665 xfs_efi_recover_work(mp, dfp, isrt,
666 &efip->efi_format.efi_extents[i]);
667 }
668
669 resv = xlog_recover_resv(&M_RES(mp)->tr_itruncate);
670 error = xfs_trans_alloc(mp, &resv, 0, 0, 0, &tp);
671 if (error)
672 return error;
673
674 error = xlog_recover_finish_intent(tp, dfp);
675 if (error == -EFSCORRUPTED)
676 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
677 &efip->efi_format,
678 sizeof(efip->efi_format));
679 if (error)
680 goto abort_error;
681
682 return xfs_defer_ops_capture_and_commit(tp, capture_list);
683
684abort_error:
685 xfs_trans_cancel(tp);
686 return error;
687}
688
689/* Relog an intent item to push the log tail forward. */
690static struct xfs_log_item *
691xfs_extent_free_relog_intent(
692 struct xfs_trans *tp,
693 struct xfs_log_item *intent,
694 struct xfs_log_item *done_item)
695{
696 struct xfs_efd_log_item *efdp = EFD_ITEM(done_item);
697 struct xfs_efi_log_item *efip;
698 struct xfs_extent *extp;
699 unsigned int count;
700
701 count = EFI_ITEM(intent)->efi_format.efi_nextents;
702 extp = EFI_ITEM(intent)->efi_format.efi_extents;
703
704 ASSERT(intent->li_type == XFS_LI_EFI || intent->li_type == XFS_LI_EFI_RT);
705
706 efdp->efd_next_extent = count;
707 memcpy(efdp->efd_format.efd_extents, extp, count * sizeof(*extp));
708
709 efip = xfs_efi_init(tp->t_mountp, intent->li_type, count);
710 memcpy(efip->efi_format.efi_extents, extp, count * sizeof(*extp));
711 atomic_set(&efip->efi_next_extent, count);
712
713 return &efip->efi_item;
714}
715
716const struct xfs_defer_op_type xfs_extent_free_defer_type = {
717 .name = "extent_free",
718 .max_items = XFS_EFI_MAX_FAST_EXTENTS,
719 .create_intent = xfs_extent_free_create_intent,
720 .abort_intent = xfs_extent_free_abort_intent,
721 .create_done = xfs_extent_free_create_done,
722 .finish_item = xfs_extent_free_finish_item,
723 .cancel_item = xfs_extent_free_cancel_item,
724 .recover_work = xfs_extent_free_recover_work,
725 .relog_intent = xfs_extent_free_relog_intent,
726};
727
728/* sub-type with special handling for AGFL deferred frees */
729const struct xfs_defer_op_type xfs_agfl_free_defer_type = {
730 .name = "agfl_free",
731 .max_items = XFS_EFI_MAX_FAST_EXTENTS,
732 .create_intent = xfs_extent_free_create_intent,
733 .abort_intent = xfs_extent_free_abort_intent,
734 .create_done = xfs_extent_free_create_done,
735 .finish_item = xfs_agfl_free_finish_item,
736 .cancel_item = xfs_extent_free_cancel_item,
737 .recover_work = xfs_extent_free_recover_work,
738 .relog_intent = xfs_extent_free_relog_intent,
739};
740
741#ifdef CONFIG_XFS_RT
742/* Create a realtime extent freeing */
743static struct xfs_log_item *
744xfs_rtextent_free_create_intent(
745 struct xfs_trans *tp,
746 struct list_head *items,
747 unsigned int count,
748 bool sort)
749{
750 return __xfs_extent_free_create_intent(tp, items, count, sort,
751 XFS_LI_EFI_RT);
752}
753
754/* Process a free realtime extent. */
755STATIC int
756xfs_rtextent_free_finish_item(
757 struct xfs_trans *tp,
758 struct xfs_log_item *done,
759 struct list_head *item,
760 struct xfs_btree_cur **state)
761{
762 struct xfs_mount *mp = tp->t_mountp;
763 struct xfs_extent_free_item *xefi = xefi_entry(item);
764 struct xfs_efd_log_item *efdp = EFD_ITEM(done);
765 struct xfs_rtgroup **rtgp = (struct xfs_rtgroup **)state;
766 int error = 0;
767
768 trace_xfs_extent_free_deferred(mp, xefi);
769
770 if (!(xefi->xefi_flags & XFS_EFI_CANCELLED)) {
771 if (*rtgp != to_rtg(xefi->xefi_group)) {
772 *rtgp = to_rtg(xefi->xefi_group);
773 xfs_rtgroup_lock(*rtgp, XFS_RTGLOCK_BITMAP);
774 xfs_rtgroup_trans_join(tp, *rtgp,
775 XFS_RTGLOCK_BITMAP);
776 }
777 error = xfs_rtfree_blocks(tp, *rtgp,
778 xefi->xefi_startblock, xefi->xefi_blockcount);
779 }
780 if (error == -EAGAIN) {
781 xfs_efd_from_efi(efdp);
782 return error;
783 }
784
785 xfs_efd_add_extent(efdp, xefi);
786 xfs_extent_free_cancel_item(item);
787 return error;
788}
789
790const struct xfs_defer_op_type xfs_rtextent_free_defer_type = {
791 .name = "rtextent_free",
792 .max_items = XFS_EFI_MAX_FAST_EXTENTS,
793 .create_intent = xfs_rtextent_free_create_intent,
794 .abort_intent = xfs_extent_free_abort_intent,
795 .create_done = xfs_extent_free_create_done,
796 .finish_item = xfs_rtextent_free_finish_item,
797 .cancel_item = xfs_extent_free_cancel_item,
798 .recover_work = xfs_extent_free_recover_work,
799 .relog_intent = xfs_extent_free_relog_intent,
800};
801#else
802const struct xfs_defer_op_type xfs_rtextent_free_defer_type = {
803 .name = "rtextent_free",
804};
805#endif /* CONFIG_XFS_RT */
806
807STATIC bool
808xfs_efi_item_match(
809 struct xfs_log_item *lip,
810 uint64_t intent_id)
811{
812 return EFI_ITEM(lip)->efi_format.efi_id == intent_id;
813}
814
815static const struct xfs_item_ops xfs_efi_item_ops = {
816 .flags = XFS_ITEM_INTENT,
817 .iop_size = xfs_efi_item_size,
818 .iop_format = xfs_efi_item_format,
819 .iop_unpin = xfs_efi_item_unpin,
820 .iop_release = xfs_efi_item_release,
821 .iop_match = xfs_efi_item_match,
822};
823
824/*
825 * This routine is called to create an in-core extent free intent
826 * item from the efi format structure which was logged on disk.
827 * It allocates an in-core efi, copies the extents from the format
828 * structure into it, and adds the efi to the AIL with the given
829 * LSN.
830 */
831STATIC int
832xlog_recover_efi_commit_pass2(
833 struct xlog *log,
834 struct list_head *buffer_list,
835 struct xlog_recover_item *item,
836 xfs_lsn_t lsn)
837{
838 struct xfs_mount *mp = log->l_mp;
839 struct xfs_efi_log_item *efip;
840 struct xfs_efi_log_format *efi_formatp;
841 int error;
842
843 efi_formatp = item->ri_buf[0].i_addr;
844
845 if (item->ri_buf[0].i_len < xfs_efi_log_format_sizeof(0)) {
846 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
847 item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
848 return -EFSCORRUPTED;
849 }
850
851 efip = xfs_efi_init(mp, ITEM_TYPE(item), efi_formatp->efi_nextents);
852 error = xfs_efi_copy_format(&item->ri_buf[0], &efip->efi_format);
853 if (error) {
854 xfs_efi_item_free(efip);
855 return error;
856 }
857 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
858
859 xlog_recover_intent_item(log, &efip->efi_item, lsn,
860 &xfs_extent_free_defer_type);
861 return 0;
862}
863
864const struct xlog_recover_item_ops xlog_efi_item_ops = {
865 .item_type = XFS_LI_EFI,
866 .commit_pass2 = xlog_recover_efi_commit_pass2,
867};
868
869#ifdef CONFIG_XFS_RT
870STATIC int
871xlog_recover_rtefi_commit_pass2(
872 struct xlog *log,
873 struct list_head *buffer_list,
874 struct xlog_recover_item *item,
875 xfs_lsn_t lsn)
876{
877 struct xfs_mount *mp = log->l_mp;
878 struct xfs_efi_log_item *efip;
879 struct xfs_efi_log_format *efi_formatp;
880 int error;
881
882 efi_formatp = item->ri_buf[0].i_addr;
883
884 if (item->ri_buf[0].i_len < xfs_efi_log_format_sizeof(0)) {
885 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
886 item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
887 return -EFSCORRUPTED;
888 }
889
890 efip = xfs_efi_init(mp, ITEM_TYPE(item), efi_formatp->efi_nextents);
891 error = xfs_efi_copy_format(&item->ri_buf[0], &efip->efi_format);
892 if (error) {
893 xfs_efi_item_free(efip);
894 return error;
895 }
896 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
897
898 xlog_recover_intent_item(log, &efip->efi_item, lsn,
899 &xfs_rtextent_free_defer_type);
900 return 0;
901}
902#else
903STATIC int
904xlog_recover_rtefi_commit_pass2(
905 struct xlog *log,
906 struct list_head *buffer_list,
907 struct xlog_recover_item *item,
908 xfs_lsn_t lsn)
909{
910 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, log->l_mp,
911 item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
912 return -EFSCORRUPTED;
913}
914#endif
915
916const struct xlog_recover_item_ops xlog_rtefi_item_ops = {
917 .item_type = XFS_LI_EFI_RT,
918 .commit_pass2 = xlog_recover_rtefi_commit_pass2,
919};
920
921/*
922 * This routine is called when an EFD format structure is found in a committed
923 * transaction in the log. Its purpose is to cancel the corresponding EFI if it
924 * was still in the log. To do this it searches the AIL for the EFI with an id
925 * equal to that in the EFD format structure. If we find it we drop the EFD
926 * reference, which removes the EFI from the AIL and frees it.
927 */
928STATIC int
929xlog_recover_efd_commit_pass2(
930 struct xlog *log,
931 struct list_head *buffer_list,
932 struct xlog_recover_item *item,
933 xfs_lsn_t lsn)
934{
935 struct xfs_efd_log_format *efd_formatp;
936 int buflen = item->ri_buf[0].i_len;
937
938 efd_formatp = item->ri_buf[0].i_addr;
939
940 if (buflen < sizeof(struct xfs_efd_log_format)) {
941 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, log->l_mp,
942 efd_formatp, buflen);
943 return -EFSCORRUPTED;
944 }
945
946 if (item->ri_buf[0].i_len != xfs_efd_log_format32_sizeof(
947 efd_formatp->efd_nextents) &&
948 item->ri_buf[0].i_len != xfs_efd_log_format64_sizeof(
949 efd_formatp->efd_nextents)) {
950 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, log->l_mp,
951 efd_formatp, buflen);
952 return -EFSCORRUPTED;
953 }
954
955 xlog_recover_release_intent(log, XFS_LI_EFI, efd_formatp->efd_efi_id);
956 return 0;
957}
958
959const struct xlog_recover_item_ops xlog_efd_item_ops = {
960 .item_type = XFS_LI_EFD,
961 .commit_pass2 = xlog_recover_efd_commit_pass2,
962};
963
964#ifdef CONFIG_XFS_RT
965STATIC int
966xlog_recover_rtefd_commit_pass2(
967 struct xlog *log,
968 struct list_head *buffer_list,
969 struct xlog_recover_item *item,
970 xfs_lsn_t lsn)
971{
972 struct xfs_efd_log_format *efd_formatp;
973 int buflen = item->ri_buf[0].i_len;
974
975 efd_formatp = item->ri_buf[0].i_addr;
976
977 if (buflen < sizeof(struct xfs_efd_log_format)) {
978 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, log->l_mp,
979 efd_formatp, buflen);
980 return -EFSCORRUPTED;
981 }
982
983 if (item->ri_buf[0].i_len != xfs_efd_log_format32_sizeof(
984 efd_formatp->efd_nextents) &&
985 item->ri_buf[0].i_len != xfs_efd_log_format64_sizeof(
986 efd_formatp->efd_nextents)) {
987 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, log->l_mp,
988 efd_formatp, buflen);
989 return -EFSCORRUPTED;
990 }
991
992 xlog_recover_release_intent(log, XFS_LI_EFI_RT,
993 efd_formatp->efd_efi_id);
994 return 0;
995}
996#else
997# define xlog_recover_rtefd_commit_pass2 xlog_recover_rtefi_commit_pass2
998#endif
999
1000const struct xlog_recover_item_ops xlog_rtefd_item_ops = {
1001 .item_type = XFS_LI_EFD_RT,
1002 .commit_pass2 = xlog_recover_rtefd_commit_pass2,
1003};
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2001,2005 Silicon Graphics, Inc.
4 * All Rights Reserved.
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_ag.h"
15#include "xfs_defer.h"
16#include "xfs_trans.h"
17#include "xfs_trans_priv.h"
18#include "xfs_extfree_item.h"
19#include "xfs_log.h"
20#include "xfs_btree.h"
21#include "xfs_rmap.h"
22#include "xfs_alloc.h"
23#include "xfs_bmap.h"
24#include "xfs_trace.h"
25#include "xfs_error.h"
26#include "xfs_log_priv.h"
27#include "xfs_log_recover.h"
28
29struct kmem_cache *xfs_efi_cache;
30struct kmem_cache *xfs_efd_cache;
31
32static const struct xfs_item_ops xfs_efi_item_ops;
33
34static inline struct xfs_efi_log_item *EFI_ITEM(struct xfs_log_item *lip)
35{
36 return container_of(lip, struct xfs_efi_log_item, efi_item);
37}
38
39STATIC void
40xfs_efi_item_free(
41 struct xfs_efi_log_item *efip)
42{
43 kvfree(efip->efi_item.li_lv_shadow);
44 if (efip->efi_format.efi_nextents > XFS_EFI_MAX_FAST_EXTENTS)
45 kfree(efip);
46 else
47 kmem_cache_free(xfs_efi_cache, efip);
48}
49
50/*
51 * Freeing the efi requires that we remove it from the AIL if it has already
52 * been placed there. However, the EFI may not yet have been placed in the AIL
53 * when called by xfs_efi_release() from EFD processing due to the ordering of
54 * committed vs unpin operations in bulk insert operations. Hence the reference
55 * count to ensure only the last caller frees the EFI.
56 */
57STATIC void
58xfs_efi_release(
59 struct xfs_efi_log_item *efip)
60{
61 ASSERT(atomic_read(&efip->efi_refcount) > 0);
62 if (!atomic_dec_and_test(&efip->efi_refcount))
63 return;
64
65 xfs_trans_ail_delete(&efip->efi_item, 0);
66 xfs_efi_item_free(efip);
67}
68
69STATIC void
70xfs_efi_item_size(
71 struct xfs_log_item *lip,
72 int *nvecs,
73 int *nbytes)
74{
75 struct xfs_efi_log_item *efip = EFI_ITEM(lip);
76
77 *nvecs += 1;
78 *nbytes += xfs_efi_log_format_sizeof(efip->efi_format.efi_nextents);
79}
80
81/*
82 * This is called to fill in the vector of log iovecs for the
83 * given efi log item. We use only 1 iovec, and we point that
84 * at the efi_log_format structure embedded in the efi item.
85 * It is at this point that we assert that all of the extent
86 * slots in the efi item have been filled.
87 */
88STATIC void
89xfs_efi_item_format(
90 struct xfs_log_item *lip,
91 struct xfs_log_vec *lv)
92{
93 struct xfs_efi_log_item *efip = EFI_ITEM(lip);
94 struct xfs_log_iovec *vecp = NULL;
95
96 ASSERT(atomic_read(&efip->efi_next_extent) ==
97 efip->efi_format.efi_nextents);
98
99 efip->efi_format.efi_type = XFS_LI_EFI;
100 efip->efi_format.efi_size = 1;
101
102 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_EFI_FORMAT,
103 &efip->efi_format,
104 xfs_efi_log_format_sizeof(efip->efi_format.efi_nextents));
105}
106
107
108/*
109 * The unpin operation is the last place an EFI is manipulated in the log. It is
110 * either inserted in the AIL or aborted in the event of a log I/O error. In
111 * either case, the EFI transaction has been successfully committed to make it
112 * this far. Therefore, we expect whoever committed the EFI to either construct
113 * and commit the EFD or drop the EFD's reference in the event of error. Simply
114 * drop the log's EFI reference now that the log is done with it.
115 */
116STATIC void
117xfs_efi_item_unpin(
118 struct xfs_log_item *lip,
119 int remove)
120{
121 struct xfs_efi_log_item *efip = EFI_ITEM(lip);
122 xfs_efi_release(efip);
123}
124
125/*
126 * The EFI has been either committed or aborted if the transaction has been
127 * cancelled. If the transaction was cancelled, an EFD isn't going to be
128 * constructed and thus we free the EFI here directly.
129 */
130STATIC void
131xfs_efi_item_release(
132 struct xfs_log_item *lip)
133{
134 xfs_efi_release(EFI_ITEM(lip));
135}
136
137/*
138 * Allocate and initialize an efi item with the given number of extents.
139 */
140STATIC struct xfs_efi_log_item *
141xfs_efi_init(
142 struct xfs_mount *mp,
143 uint nextents)
144
145{
146 struct xfs_efi_log_item *efip;
147
148 ASSERT(nextents > 0);
149 if (nextents > XFS_EFI_MAX_FAST_EXTENTS) {
150 efip = kzalloc(xfs_efi_log_item_sizeof(nextents),
151 GFP_KERNEL | __GFP_NOFAIL);
152 } else {
153 efip = kmem_cache_zalloc(xfs_efi_cache,
154 GFP_KERNEL | __GFP_NOFAIL);
155 }
156
157 xfs_log_item_init(mp, &efip->efi_item, XFS_LI_EFI, &xfs_efi_item_ops);
158 efip->efi_format.efi_nextents = nextents;
159 efip->efi_format.efi_id = (uintptr_t)(void *)efip;
160 atomic_set(&efip->efi_next_extent, 0);
161 atomic_set(&efip->efi_refcount, 2);
162
163 return efip;
164}
165
166/*
167 * Copy an EFI format buffer from the given buf, and into the destination
168 * EFI format structure.
169 * The given buffer can be in 32 bit or 64 bit form (which has different padding),
170 * one of which will be the native format for this kernel.
171 * It will handle the conversion of formats if necessary.
172 */
173STATIC int
174xfs_efi_copy_format(xfs_log_iovec_t *buf, xfs_efi_log_format_t *dst_efi_fmt)
175{
176 xfs_efi_log_format_t *src_efi_fmt = buf->i_addr;
177 uint i;
178 uint len = xfs_efi_log_format_sizeof(src_efi_fmt->efi_nextents);
179 uint len32 = xfs_efi_log_format32_sizeof(src_efi_fmt->efi_nextents);
180 uint len64 = xfs_efi_log_format64_sizeof(src_efi_fmt->efi_nextents);
181
182 if (buf->i_len == len) {
183 memcpy(dst_efi_fmt, src_efi_fmt,
184 offsetof(struct xfs_efi_log_format, efi_extents));
185 for (i = 0; i < src_efi_fmt->efi_nextents; i++)
186 memcpy(&dst_efi_fmt->efi_extents[i],
187 &src_efi_fmt->efi_extents[i],
188 sizeof(struct xfs_extent));
189 return 0;
190 } else if (buf->i_len == len32) {
191 xfs_efi_log_format_32_t *src_efi_fmt_32 = buf->i_addr;
192
193 dst_efi_fmt->efi_type = src_efi_fmt_32->efi_type;
194 dst_efi_fmt->efi_size = src_efi_fmt_32->efi_size;
195 dst_efi_fmt->efi_nextents = src_efi_fmt_32->efi_nextents;
196 dst_efi_fmt->efi_id = src_efi_fmt_32->efi_id;
197 for (i = 0; i < dst_efi_fmt->efi_nextents; i++) {
198 dst_efi_fmt->efi_extents[i].ext_start =
199 src_efi_fmt_32->efi_extents[i].ext_start;
200 dst_efi_fmt->efi_extents[i].ext_len =
201 src_efi_fmt_32->efi_extents[i].ext_len;
202 }
203 return 0;
204 } else if (buf->i_len == len64) {
205 xfs_efi_log_format_64_t *src_efi_fmt_64 = buf->i_addr;
206
207 dst_efi_fmt->efi_type = src_efi_fmt_64->efi_type;
208 dst_efi_fmt->efi_size = src_efi_fmt_64->efi_size;
209 dst_efi_fmt->efi_nextents = src_efi_fmt_64->efi_nextents;
210 dst_efi_fmt->efi_id = src_efi_fmt_64->efi_id;
211 for (i = 0; i < dst_efi_fmt->efi_nextents; i++) {
212 dst_efi_fmt->efi_extents[i].ext_start =
213 src_efi_fmt_64->efi_extents[i].ext_start;
214 dst_efi_fmt->efi_extents[i].ext_len =
215 src_efi_fmt_64->efi_extents[i].ext_len;
216 }
217 return 0;
218 }
219 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, NULL, buf->i_addr,
220 buf->i_len);
221 return -EFSCORRUPTED;
222}
223
224static inline struct xfs_efd_log_item *EFD_ITEM(struct xfs_log_item *lip)
225{
226 return container_of(lip, struct xfs_efd_log_item, efd_item);
227}
228
229STATIC void
230xfs_efd_item_free(struct xfs_efd_log_item *efdp)
231{
232 kvfree(efdp->efd_item.li_lv_shadow);
233 if (efdp->efd_format.efd_nextents > XFS_EFD_MAX_FAST_EXTENTS)
234 kfree(efdp);
235 else
236 kmem_cache_free(xfs_efd_cache, efdp);
237}
238
239STATIC void
240xfs_efd_item_size(
241 struct xfs_log_item *lip,
242 int *nvecs,
243 int *nbytes)
244{
245 struct xfs_efd_log_item *efdp = EFD_ITEM(lip);
246
247 *nvecs += 1;
248 *nbytes += xfs_efd_log_format_sizeof(efdp->efd_format.efd_nextents);
249}
250
251/*
252 * This is called to fill in the vector of log iovecs for the
253 * given efd log item. We use only 1 iovec, and we point that
254 * at the efd_log_format structure embedded in the efd item.
255 * It is at this point that we assert that all of the extent
256 * slots in the efd item have been filled.
257 */
258STATIC void
259xfs_efd_item_format(
260 struct xfs_log_item *lip,
261 struct xfs_log_vec *lv)
262{
263 struct xfs_efd_log_item *efdp = EFD_ITEM(lip);
264 struct xfs_log_iovec *vecp = NULL;
265
266 ASSERT(efdp->efd_next_extent == efdp->efd_format.efd_nextents);
267
268 efdp->efd_format.efd_type = XFS_LI_EFD;
269 efdp->efd_format.efd_size = 1;
270
271 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_EFD_FORMAT,
272 &efdp->efd_format,
273 xfs_efd_log_format_sizeof(efdp->efd_format.efd_nextents));
274}
275
276/*
277 * The EFD is either committed or aborted if the transaction is cancelled. If
278 * the transaction is cancelled, drop our reference to the EFI and free the EFD.
279 */
280STATIC void
281xfs_efd_item_release(
282 struct xfs_log_item *lip)
283{
284 struct xfs_efd_log_item *efdp = EFD_ITEM(lip);
285
286 xfs_efi_release(efdp->efd_efip);
287 xfs_efd_item_free(efdp);
288}
289
290static struct xfs_log_item *
291xfs_efd_item_intent(
292 struct xfs_log_item *lip)
293{
294 return &EFD_ITEM(lip)->efd_efip->efi_item;
295}
296
297static const struct xfs_item_ops xfs_efd_item_ops = {
298 .flags = XFS_ITEM_RELEASE_WHEN_COMMITTED |
299 XFS_ITEM_INTENT_DONE,
300 .iop_size = xfs_efd_item_size,
301 .iop_format = xfs_efd_item_format,
302 .iop_release = xfs_efd_item_release,
303 .iop_intent = xfs_efd_item_intent,
304};
305
306/*
307 * Fill the EFD with all extents from the EFI when we need to roll the
308 * transaction and continue with a new EFI.
309 *
310 * This simply copies all the extents in the EFI to the EFD rather than make
311 * assumptions about which extents in the EFI have already been processed. We
312 * currently keep the xefi list in the same order as the EFI extent list, but
313 * that may not always be the case. Copying everything avoids leaving a landmine
314 * were we fail to cancel all the extents in an EFI if the xefi list is
315 * processed in a different order to the extents in the EFI.
316 */
317static void
318xfs_efd_from_efi(
319 struct xfs_efd_log_item *efdp)
320{
321 struct xfs_efi_log_item *efip = efdp->efd_efip;
322 uint i;
323
324 ASSERT(efip->efi_format.efi_nextents > 0);
325 ASSERT(efdp->efd_next_extent < efip->efi_format.efi_nextents);
326
327 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
328 efdp->efd_format.efd_extents[i] =
329 efip->efi_format.efi_extents[i];
330 }
331 efdp->efd_next_extent = efip->efi_format.efi_nextents;
332}
333
334/* Sort bmap items by AG. */
335static int
336xfs_extent_free_diff_items(
337 void *priv,
338 const struct list_head *a,
339 const struct list_head *b)
340{
341 struct xfs_extent_free_item *ra;
342 struct xfs_extent_free_item *rb;
343
344 ra = container_of(a, struct xfs_extent_free_item, xefi_list);
345 rb = container_of(b, struct xfs_extent_free_item, xefi_list);
346
347 return ra->xefi_pag->pag_agno - rb->xefi_pag->pag_agno;
348}
349
350/* Log a free extent to the intent item. */
351STATIC void
352xfs_extent_free_log_item(
353 struct xfs_trans *tp,
354 struct xfs_efi_log_item *efip,
355 struct xfs_extent_free_item *xefi)
356{
357 uint next_extent;
358 struct xfs_extent *extp;
359
360 /*
361 * atomic_inc_return gives us the value after the increment;
362 * we want to use it as an array index so we need to subtract 1 from
363 * it.
364 */
365 next_extent = atomic_inc_return(&efip->efi_next_extent) - 1;
366 ASSERT(next_extent < efip->efi_format.efi_nextents);
367 extp = &efip->efi_format.efi_extents[next_extent];
368 extp->ext_start = xefi->xefi_startblock;
369 extp->ext_len = xefi->xefi_blockcount;
370}
371
372static struct xfs_log_item *
373xfs_extent_free_create_intent(
374 struct xfs_trans *tp,
375 struct list_head *items,
376 unsigned int count,
377 bool sort)
378{
379 struct xfs_mount *mp = tp->t_mountp;
380 struct xfs_efi_log_item *efip = xfs_efi_init(mp, count);
381 struct xfs_extent_free_item *xefi;
382
383 ASSERT(count > 0);
384
385 if (sort)
386 list_sort(mp, items, xfs_extent_free_diff_items);
387 list_for_each_entry(xefi, items, xefi_list)
388 xfs_extent_free_log_item(tp, efip, xefi);
389 return &efip->efi_item;
390}
391
392/* Get an EFD so we can process all the free extents. */
393static struct xfs_log_item *
394xfs_extent_free_create_done(
395 struct xfs_trans *tp,
396 struct xfs_log_item *intent,
397 unsigned int count)
398{
399 struct xfs_efi_log_item *efip = EFI_ITEM(intent);
400 struct xfs_efd_log_item *efdp;
401
402 ASSERT(count > 0);
403
404 if (count > XFS_EFD_MAX_FAST_EXTENTS) {
405 efdp = kzalloc(xfs_efd_log_item_sizeof(count),
406 GFP_KERNEL | __GFP_NOFAIL);
407 } else {
408 efdp = kmem_cache_zalloc(xfs_efd_cache,
409 GFP_KERNEL | __GFP_NOFAIL);
410 }
411
412 xfs_log_item_init(tp->t_mountp, &efdp->efd_item, XFS_LI_EFD,
413 &xfs_efd_item_ops);
414 efdp->efd_efip = efip;
415 efdp->efd_format.efd_nextents = count;
416 efdp->efd_format.efd_efi_id = efip->efi_format.efi_id;
417
418 return &efdp->efd_item;
419}
420
421/* Take a passive ref to the AG containing the space we're freeing. */
422void
423xfs_extent_free_get_group(
424 struct xfs_mount *mp,
425 struct xfs_extent_free_item *xefi)
426{
427 xfs_agnumber_t agno;
428
429 agno = XFS_FSB_TO_AGNO(mp, xefi->xefi_startblock);
430 xefi->xefi_pag = xfs_perag_intent_get(mp, agno);
431}
432
433/* Release a passive AG ref after some freeing work. */
434static inline void
435xfs_extent_free_put_group(
436 struct xfs_extent_free_item *xefi)
437{
438 xfs_perag_intent_put(xefi->xefi_pag);
439}
440
441/* Process a free extent. */
442STATIC int
443xfs_extent_free_finish_item(
444 struct xfs_trans *tp,
445 struct xfs_log_item *done,
446 struct list_head *item,
447 struct xfs_btree_cur **state)
448{
449 struct xfs_owner_info oinfo = { };
450 struct xfs_extent_free_item *xefi;
451 struct xfs_efd_log_item *efdp = EFD_ITEM(done);
452 struct xfs_mount *mp = tp->t_mountp;
453 struct xfs_extent *extp;
454 uint next_extent;
455 xfs_agblock_t agbno;
456 int error = 0;
457
458 xefi = container_of(item, struct xfs_extent_free_item, xefi_list);
459 agbno = XFS_FSB_TO_AGBNO(mp, xefi->xefi_startblock);
460
461 oinfo.oi_owner = xefi->xefi_owner;
462 if (xefi->xefi_flags & XFS_EFI_ATTR_FORK)
463 oinfo.oi_flags |= XFS_OWNER_INFO_ATTR_FORK;
464 if (xefi->xefi_flags & XFS_EFI_BMBT_BLOCK)
465 oinfo.oi_flags |= XFS_OWNER_INFO_BMBT_BLOCK;
466
467 trace_xfs_bmap_free_deferred(tp->t_mountp, xefi->xefi_pag->pag_agno, 0,
468 agbno, xefi->xefi_blockcount);
469
470 /*
471 * If we need a new transaction to make progress, the caller will log a
472 * new EFI with the current contents. It will also log an EFD to cancel
473 * the existing EFI, and so we need to copy all the unprocessed extents
474 * in this EFI to the EFD so this works correctly.
475 */
476 if (!(xefi->xefi_flags & XFS_EFI_CANCELLED))
477 error = __xfs_free_extent(tp, xefi->xefi_pag, agbno,
478 xefi->xefi_blockcount, &oinfo, xefi->xefi_agresv,
479 xefi->xefi_flags & XFS_EFI_SKIP_DISCARD);
480 if (error == -EAGAIN) {
481 xfs_efd_from_efi(efdp);
482 return error;
483 }
484
485 /* Add the work we finished to the EFD, even though nobody uses that */
486 next_extent = efdp->efd_next_extent;
487 ASSERT(next_extent < efdp->efd_format.efd_nextents);
488 extp = &(efdp->efd_format.efd_extents[next_extent]);
489 extp->ext_start = xefi->xefi_startblock;
490 extp->ext_len = xefi->xefi_blockcount;
491 efdp->efd_next_extent++;
492
493 xfs_extent_free_put_group(xefi);
494 kmem_cache_free(xfs_extfree_item_cache, xefi);
495 return error;
496}
497
498/* Abort all pending EFIs. */
499STATIC void
500xfs_extent_free_abort_intent(
501 struct xfs_log_item *intent)
502{
503 xfs_efi_release(EFI_ITEM(intent));
504}
505
506/* Cancel a free extent. */
507STATIC void
508xfs_extent_free_cancel_item(
509 struct list_head *item)
510{
511 struct xfs_extent_free_item *xefi;
512
513 xefi = container_of(item, struct xfs_extent_free_item, xefi_list);
514
515 xfs_extent_free_put_group(xefi);
516 kmem_cache_free(xfs_extfree_item_cache, xefi);
517}
518
519/*
520 * AGFL blocks are accounted differently in the reserve pools and are not
521 * inserted into the busy extent list.
522 */
523STATIC int
524xfs_agfl_free_finish_item(
525 struct xfs_trans *tp,
526 struct xfs_log_item *done,
527 struct list_head *item,
528 struct xfs_btree_cur **state)
529{
530 struct xfs_owner_info oinfo = { };
531 struct xfs_mount *mp = tp->t_mountp;
532 struct xfs_efd_log_item *efdp = EFD_ITEM(done);
533 struct xfs_extent_free_item *xefi;
534 struct xfs_extent *extp;
535 struct xfs_buf *agbp;
536 int error;
537 xfs_agblock_t agbno;
538 uint next_extent;
539
540 xefi = container_of(item, struct xfs_extent_free_item, xefi_list);
541 ASSERT(xefi->xefi_blockcount == 1);
542 agbno = XFS_FSB_TO_AGBNO(mp, xefi->xefi_startblock);
543 oinfo.oi_owner = xefi->xefi_owner;
544
545 trace_xfs_agfl_free_deferred(mp, xefi->xefi_pag->pag_agno, 0, agbno,
546 xefi->xefi_blockcount);
547
548 error = xfs_alloc_read_agf(xefi->xefi_pag, tp, 0, &agbp);
549 if (!error)
550 error = xfs_free_agfl_block(tp, xefi->xefi_pag->pag_agno,
551 agbno, agbp, &oinfo);
552
553 next_extent = efdp->efd_next_extent;
554 ASSERT(next_extent < efdp->efd_format.efd_nextents);
555 extp = &(efdp->efd_format.efd_extents[next_extent]);
556 extp->ext_start = xefi->xefi_startblock;
557 extp->ext_len = xefi->xefi_blockcount;
558 efdp->efd_next_extent++;
559
560 xfs_extent_free_put_group(xefi);
561 kmem_cache_free(xfs_extfree_item_cache, xefi);
562 return error;
563}
564
565/* Is this recovered EFI ok? */
566static inline bool
567xfs_efi_validate_ext(
568 struct xfs_mount *mp,
569 struct xfs_extent *extp)
570{
571 return xfs_verify_fsbext(mp, extp->ext_start, extp->ext_len);
572}
573
574static inline void
575xfs_efi_recover_work(
576 struct xfs_mount *mp,
577 struct xfs_defer_pending *dfp,
578 struct xfs_extent *extp)
579{
580 struct xfs_extent_free_item *xefi;
581
582 xefi = kmem_cache_zalloc(xfs_extfree_item_cache,
583 GFP_KERNEL | __GFP_NOFAIL);
584 xefi->xefi_startblock = extp->ext_start;
585 xefi->xefi_blockcount = extp->ext_len;
586 xefi->xefi_agresv = XFS_AG_RESV_NONE;
587 xefi->xefi_owner = XFS_RMAP_OWN_UNKNOWN;
588 xfs_extent_free_get_group(mp, xefi);
589
590 xfs_defer_add_item(dfp, &xefi->xefi_list);
591}
592
593/*
594 * Process an extent free intent item that was recovered from
595 * the log. We need to free the extents that it describes.
596 */
597STATIC int
598xfs_extent_free_recover_work(
599 struct xfs_defer_pending *dfp,
600 struct list_head *capture_list)
601{
602 struct xfs_trans_res resv;
603 struct xfs_log_item *lip = dfp->dfp_intent;
604 struct xfs_efi_log_item *efip = EFI_ITEM(lip);
605 struct xfs_mount *mp = lip->li_log->l_mp;
606 struct xfs_trans *tp;
607 int i;
608 int error = 0;
609
610 /*
611 * First check the validity of the extents described by the
612 * EFI. If any are bad, then assume that all are bad and
613 * just toss the EFI.
614 */
615 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
616 if (!xfs_efi_validate_ext(mp,
617 &efip->efi_format.efi_extents[i])) {
618 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
619 &efip->efi_format,
620 sizeof(efip->efi_format));
621 return -EFSCORRUPTED;
622 }
623
624 xfs_efi_recover_work(mp, dfp, &efip->efi_format.efi_extents[i]);
625 }
626
627 resv = xlog_recover_resv(&M_RES(mp)->tr_itruncate);
628 error = xfs_trans_alloc(mp, &resv, 0, 0, 0, &tp);
629 if (error)
630 return error;
631
632 error = xlog_recover_finish_intent(tp, dfp);
633 if (error == -EFSCORRUPTED)
634 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
635 &efip->efi_format,
636 sizeof(efip->efi_format));
637 if (error)
638 goto abort_error;
639
640 return xfs_defer_ops_capture_and_commit(tp, capture_list);
641
642abort_error:
643 xfs_trans_cancel(tp);
644 return error;
645}
646
647/* Relog an intent item to push the log tail forward. */
648static struct xfs_log_item *
649xfs_extent_free_relog_intent(
650 struct xfs_trans *tp,
651 struct xfs_log_item *intent,
652 struct xfs_log_item *done_item)
653{
654 struct xfs_efd_log_item *efdp = EFD_ITEM(done_item);
655 struct xfs_efi_log_item *efip;
656 struct xfs_extent *extp;
657 unsigned int count;
658
659 count = EFI_ITEM(intent)->efi_format.efi_nextents;
660 extp = EFI_ITEM(intent)->efi_format.efi_extents;
661
662 efdp->efd_next_extent = count;
663 memcpy(efdp->efd_format.efd_extents, extp, count * sizeof(*extp));
664
665 efip = xfs_efi_init(tp->t_mountp, count);
666 memcpy(efip->efi_format.efi_extents, extp, count * sizeof(*extp));
667 atomic_set(&efip->efi_next_extent, count);
668
669 return &efip->efi_item;
670}
671
672const struct xfs_defer_op_type xfs_extent_free_defer_type = {
673 .name = "extent_free",
674 .max_items = XFS_EFI_MAX_FAST_EXTENTS,
675 .create_intent = xfs_extent_free_create_intent,
676 .abort_intent = xfs_extent_free_abort_intent,
677 .create_done = xfs_extent_free_create_done,
678 .finish_item = xfs_extent_free_finish_item,
679 .cancel_item = xfs_extent_free_cancel_item,
680 .recover_work = xfs_extent_free_recover_work,
681 .relog_intent = xfs_extent_free_relog_intent,
682};
683
684/* sub-type with special handling for AGFL deferred frees */
685const struct xfs_defer_op_type xfs_agfl_free_defer_type = {
686 .name = "agfl_free",
687 .max_items = XFS_EFI_MAX_FAST_EXTENTS,
688 .create_intent = xfs_extent_free_create_intent,
689 .abort_intent = xfs_extent_free_abort_intent,
690 .create_done = xfs_extent_free_create_done,
691 .finish_item = xfs_agfl_free_finish_item,
692 .cancel_item = xfs_extent_free_cancel_item,
693 .recover_work = xfs_extent_free_recover_work,
694 .relog_intent = xfs_extent_free_relog_intent,
695};
696
697STATIC bool
698xfs_efi_item_match(
699 struct xfs_log_item *lip,
700 uint64_t intent_id)
701{
702 return EFI_ITEM(lip)->efi_format.efi_id == intent_id;
703}
704
705static const struct xfs_item_ops xfs_efi_item_ops = {
706 .flags = XFS_ITEM_INTENT,
707 .iop_size = xfs_efi_item_size,
708 .iop_format = xfs_efi_item_format,
709 .iop_unpin = xfs_efi_item_unpin,
710 .iop_release = xfs_efi_item_release,
711 .iop_match = xfs_efi_item_match,
712};
713
714/*
715 * This routine is called to create an in-core extent free intent
716 * item from the efi format structure which was logged on disk.
717 * It allocates an in-core efi, copies the extents from the format
718 * structure into it, and adds the efi to the AIL with the given
719 * LSN.
720 */
721STATIC int
722xlog_recover_efi_commit_pass2(
723 struct xlog *log,
724 struct list_head *buffer_list,
725 struct xlog_recover_item *item,
726 xfs_lsn_t lsn)
727{
728 struct xfs_mount *mp = log->l_mp;
729 struct xfs_efi_log_item *efip;
730 struct xfs_efi_log_format *efi_formatp;
731 int error;
732
733 efi_formatp = item->ri_buf[0].i_addr;
734
735 if (item->ri_buf[0].i_len < xfs_efi_log_format_sizeof(0)) {
736 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
737 item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
738 return -EFSCORRUPTED;
739 }
740
741 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
742 error = xfs_efi_copy_format(&item->ri_buf[0], &efip->efi_format);
743 if (error) {
744 xfs_efi_item_free(efip);
745 return error;
746 }
747 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
748
749 xlog_recover_intent_item(log, &efip->efi_item, lsn,
750 &xfs_extent_free_defer_type);
751 return 0;
752}
753
754const struct xlog_recover_item_ops xlog_efi_item_ops = {
755 .item_type = XFS_LI_EFI,
756 .commit_pass2 = xlog_recover_efi_commit_pass2,
757};
758
759/*
760 * This routine is called when an EFD format structure is found in a committed
761 * transaction in the log. Its purpose is to cancel the corresponding EFI if it
762 * was still in the log. To do this it searches the AIL for the EFI with an id
763 * equal to that in the EFD format structure. If we find it we drop the EFD
764 * reference, which removes the EFI from the AIL and frees it.
765 */
766STATIC int
767xlog_recover_efd_commit_pass2(
768 struct xlog *log,
769 struct list_head *buffer_list,
770 struct xlog_recover_item *item,
771 xfs_lsn_t lsn)
772{
773 struct xfs_efd_log_format *efd_formatp;
774 int buflen = item->ri_buf[0].i_len;
775
776 efd_formatp = item->ri_buf[0].i_addr;
777
778 if (buflen < sizeof(struct xfs_efd_log_format)) {
779 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, log->l_mp,
780 efd_formatp, buflen);
781 return -EFSCORRUPTED;
782 }
783
784 if (item->ri_buf[0].i_len != xfs_efd_log_format32_sizeof(
785 efd_formatp->efd_nextents) &&
786 item->ri_buf[0].i_len != xfs_efd_log_format64_sizeof(
787 efd_formatp->efd_nextents)) {
788 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, log->l_mp,
789 efd_formatp, buflen);
790 return -EFSCORRUPTED;
791 }
792
793 xlog_recover_release_intent(log, XFS_LI_EFI, efd_formatp->efd_efi_id);
794 return 0;
795}
796
797const struct xlog_recover_item_ops xlog_efd_item_ops = {
798 .item_type = XFS_LI_EFD,
799 .commit_pass2 = xlog_recover_efd_commit_pass2,
800};