1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2001,2005 Silicon Graphics, Inc.
8 #include "xfs_format.h"
9 #include "xfs_log_format.h"
10 #include "xfs_trans_resv.h"
12 #include "xfs_shared.h"
13 #include "xfs_mount.h"
14 #include "xfs_defer.h"
15 #include "xfs_trans.h"
16 #include "xfs_trans_priv.h"
17 #include "xfs_extfree_item.h"
19 #include "xfs_btree.h"
21 #include "xfs_alloc.h"
23 #include "xfs_trace.h"
24 #include "xfs_error.h"
25 #include "xfs_log_priv.h"
26 #include "xfs_log_recover.h"
28 kmem_zone_t *xfs_efi_zone;
29 kmem_zone_t *xfs_efd_zone;
31 static const struct xfs_item_ops xfs_efi_item_ops;
33 static inline struct xfs_efi_log_item *EFI_ITEM(struct xfs_log_item *lip)
35 return container_of(lip, struct xfs_efi_log_item, efi_item);
40 struct xfs_efi_log_item *efip)
42 kmem_free(efip->efi_item.li_lv_shadow);
43 if (efip->efi_format.efi_nextents > XFS_EFI_MAX_FAST_EXTENTS)
46 kmem_cache_free(xfs_efi_zone, efip);
50 * Freeing the efi requires that we remove it from the AIL if it has already
51 * been placed there. However, the EFI may not yet have been placed in the AIL
52 * when called by xfs_efi_release() from EFD processing due to the ordering of
53 * committed vs unpin operations in bulk insert operations. Hence the reference
54 * count to ensure only the last caller frees the EFI.
58 struct xfs_efi_log_item *efip)
60 ASSERT(atomic_read(&efip->efi_refcount) > 0);
61 if (atomic_dec_and_test(&efip->efi_refcount)) {
62 xfs_trans_ail_delete(&efip->efi_item, SHUTDOWN_LOG_IO_ERROR);
63 xfs_efi_item_free(efip);
68 * This returns the number of iovecs needed to log the given efi item.
69 * We only need 1 iovec for an efi item. It just logs the efi_log_format
74 struct xfs_efi_log_item *efip)
76 return sizeof(struct xfs_efi_log_format) +
77 (efip->efi_format.efi_nextents - 1) * sizeof(xfs_extent_t);
82 struct xfs_log_item *lip,
87 *nbytes += xfs_efi_item_sizeof(EFI_ITEM(lip));
91 * This is called to fill in the vector of log iovecs for the
92 * given efi log item. We use only 1 iovec, and we point that
93 * at the efi_log_format structure embedded in the efi item.
94 * It is at this point that we assert that all of the extent
95 * slots in the efi item have been filled.
99 struct xfs_log_item *lip,
100 struct xfs_log_vec *lv)
102 struct xfs_efi_log_item *efip = EFI_ITEM(lip);
103 struct xfs_log_iovec *vecp = NULL;
105 ASSERT(atomic_read(&efip->efi_next_extent) ==
106 efip->efi_format.efi_nextents);
108 efip->efi_format.efi_type = XFS_LI_EFI;
109 efip->efi_format.efi_size = 1;
111 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_EFI_FORMAT,
113 xfs_efi_item_sizeof(efip));
118 * The unpin operation is the last place an EFI is manipulated in the log. It is
119 * either inserted in the AIL or aborted in the event of a log I/O error. In
120 * either case, the EFI transaction has been successfully committed to make it
121 * this far. Therefore, we expect whoever committed the EFI to either construct
122 * and commit the EFD or drop the EFD's reference in the event of error. Simply
123 * drop the log's EFI reference now that the log is done with it.
127 struct xfs_log_item *lip,
130 struct xfs_efi_log_item *efip = EFI_ITEM(lip);
131 xfs_efi_release(efip);
135 * The EFI has been either committed or aborted if the transaction has been
136 * cancelled. If the transaction was cancelled, an EFD isn't going to be
137 * constructed and thus we free the EFI here directly.
140 xfs_efi_item_release(
141 struct xfs_log_item *lip)
143 xfs_efi_release(EFI_ITEM(lip));
147 * Allocate and initialize an efi item with the given number of extents.
149 STATIC struct xfs_efi_log_item *
151 struct xfs_mount *mp,
155 struct xfs_efi_log_item *efip;
158 ASSERT(nextents > 0);
159 if (nextents > XFS_EFI_MAX_FAST_EXTENTS) {
160 size = (uint)(sizeof(struct xfs_efi_log_item) +
161 ((nextents - 1) * sizeof(xfs_extent_t)));
162 efip = kmem_zalloc(size, 0);
164 efip = kmem_zone_zalloc(xfs_efi_zone, 0);
167 xfs_log_item_init(mp, &efip->efi_item, XFS_LI_EFI, &xfs_efi_item_ops);
168 efip->efi_format.efi_nextents = nextents;
169 efip->efi_format.efi_id = (uintptr_t)(void *)efip;
170 atomic_set(&efip->efi_next_extent, 0);
171 atomic_set(&efip->efi_refcount, 2);
177 * Copy an EFI format buffer from the given buf, and into the destination
178 * EFI format structure.
179 * The given buffer can be in 32 bit or 64 bit form (which has different padding),
180 * one of which will be the native format for this kernel.
181 * It will handle the conversion of formats if necessary.
184 xfs_efi_copy_format(xfs_log_iovec_t *buf, xfs_efi_log_format_t *dst_efi_fmt)
186 xfs_efi_log_format_t *src_efi_fmt = buf->i_addr;
188 uint len = sizeof(xfs_efi_log_format_t) +
189 (src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_t);
190 uint len32 = sizeof(xfs_efi_log_format_32_t) +
191 (src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_32_t);
192 uint len64 = sizeof(xfs_efi_log_format_64_t) +
193 (src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_64_t);
195 if (buf->i_len == len) {
196 memcpy((char *)dst_efi_fmt, (char*)src_efi_fmt, len);
198 } else if (buf->i_len == len32) {
199 xfs_efi_log_format_32_t *src_efi_fmt_32 = buf->i_addr;
201 dst_efi_fmt->efi_type = src_efi_fmt_32->efi_type;
202 dst_efi_fmt->efi_size = src_efi_fmt_32->efi_size;
203 dst_efi_fmt->efi_nextents = src_efi_fmt_32->efi_nextents;
204 dst_efi_fmt->efi_id = src_efi_fmt_32->efi_id;
205 for (i = 0; i < dst_efi_fmt->efi_nextents; i++) {
206 dst_efi_fmt->efi_extents[i].ext_start =
207 src_efi_fmt_32->efi_extents[i].ext_start;
208 dst_efi_fmt->efi_extents[i].ext_len =
209 src_efi_fmt_32->efi_extents[i].ext_len;
212 } else if (buf->i_len == len64) {
213 xfs_efi_log_format_64_t *src_efi_fmt_64 = buf->i_addr;
215 dst_efi_fmt->efi_type = src_efi_fmt_64->efi_type;
216 dst_efi_fmt->efi_size = src_efi_fmt_64->efi_size;
217 dst_efi_fmt->efi_nextents = src_efi_fmt_64->efi_nextents;
218 dst_efi_fmt->efi_id = src_efi_fmt_64->efi_id;
219 for (i = 0; i < dst_efi_fmt->efi_nextents; i++) {
220 dst_efi_fmt->efi_extents[i].ext_start =
221 src_efi_fmt_64->efi_extents[i].ext_start;
222 dst_efi_fmt->efi_extents[i].ext_len =
223 src_efi_fmt_64->efi_extents[i].ext_len;
227 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, NULL);
228 return -EFSCORRUPTED;
231 static inline struct xfs_efd_log_item *EFD_ITEM(struct xfs_log_item *lip)
233 return container_of(lip, struct xfs_efd_log_item, efd_item);
237 xfs_efd_item_free(struct xfs_efd_log_item *efdp)
239 kmem_free(efdp->efd_item.li_lv_shadow);
240 if (efdp->efd_format.efd_nextents > XFS_EFD_MAX_FAST_EXTENTS)
243 kmem_cache_free(xfs_efd_zone, efdp);
247 * This returns the number of iovecs needed to log the given efd item.
248 * We only need 1 iovec for an efd item. It just logs the efd_log_format
253 struct xfs_efd_log_item *efdp)
255 return sizeof(xfs_efd_log_format_t) +
256 (efdp->efd_format.efd_nextents - 1) * sizeof(xfs_extent_t);
261 struct xfs_log_item *lip,
266 *nbytes += xfs_efd_item_sizeof(EFD_ITEM(lip));
270 * This is called to fill in the vector of log iovecs for the
271 * given efd log item. We use only 1 iovec, and we point that
272 * at the efd_log_format structure embedded in the efd item.
273 * It is at this point that we assert that all of the extent
274 * slots in the efd item have been filled.
278 struct xfs_log_item *lip,
279 struct xfs_log_vec *lv)
281 struct xfs_efd_log_item *efdp = EFD_ITEM(lip);
282 struct xfs_log_iovec *vecp = NULL;
284 ASSERT(efdp->efd_next_extent == efdp->efd_format.efd_nextents);
286 efdp->efd_format.efd_type = XFS_LI_EFD;
287 efdp->efd_format.efd_size = 1;
289 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_EFD_FORMAT,
291 xfs_efd_item_sizeof(efdp));
295 * The EFD is either committed or aborted if the transaction is cancelled. If
296 * the transaction is cancelled, drop our reference to the EFI and free the EFD.
299 xfs_efd_item_release(
300 struct xfs_log_item *lip)
302 struct xfs_efd_log_item *efdp = EFD_ITEM(lip);
304 xfs_efi_release(efdp->efd_efip);
305 xfs_efd_item_free(efdp);
308 static const struct xfs_item_ops xfs_efd_item_ops = {
309 .flags = XFS_ITEM_RELEASE_WHEN_COMMITTED,
310 .iop_size = xfs_efd_item_size,
311 .iop_format = xfs_efd_item_format,
312 .iop_release = xfs_efd_item_release,
316 * Allocate an "extent free done" log item that will hold nextents worth of
317 * extents. The caller must use all nextents extents, because we are not
318 * flexible about this at all.
320 static struct xfs_efd_log_item *
322 struct xfs_trans *tp,
323 struct xfs_efi_log_item *efip,
324 unsigned int nextents)
326 struct xfs_efd_log_item *efdp;
328 ASSERT(nextents > 0);
330 if (nextents > XFS_EFD_MAX_FAST_EXTENTS) {
331 efdp = kmem_zalloc(sizeof(struct xfs_efd_log_item) +
332 (nextents - 1) * sizeof(struct xfs_extent),
335 efdp = kmem_zone_zalloc(xfs_efd_zone, 0);
338 xfs_log_item_init(tp->t_mountp, &efdp->efd_item, XFS_LI_EFD,
340 efdp->efd_efip = efip;
341 efdp->efd_format.efd_nextents = nextents;
342 efdp->efd_format.efd_efi_id = efip->efi_format.efi_id;
344 xfs_trans_add_item(tp, &efdp->efd_item);
349 * Free an extent and log it to the EFD. Note that the transaction is marked
350 * dirty regardless of whether the extent free succeeds or fails to support the
351 * EFI/EFD lifecycle rules.
354 xfs_trans_free_extent(
355 struct xfs_trans *tp,
356 struct xfs_efd_log_item *efdp,
357 xfs_fsblock_t start_block,
358 xfs_extlen_t ext_len,
359 const struct xfs_owner_info *oinfo,
362 struct xfs_mount *mp = tp->t_mountp;
363 struct xfs_extent *extp;
365 xfs_agnumber_t agno = XFS_FSB_TO_AGNO(mp, start_block);
366 xfs_agblock_t agbno = XFS_FSB_TO_AGBNO(mp,
370 trace_xfs_bmap_free_deferred(tp->t_mountp, agno, 0, agbno, ext_len);
372 error = __xfs_free_extent(tp, start_block, ext_len,
373 oinfo, XFS_AG_RESV_NONE, skip_discard);
375 * Mark the transaction dirty, even on error. This ensures the
376 * transaction is aborted, which:
378 * 1.) releases the EFI and frees the EFD
379 * 2.) shuts down the filesystem
381 tp->t_flags |= XFS_TRANS_DIRTY;
382 set_bit(XFS_LI_DIRTY, &efdp->efd_item.li_flags);
384 next_extent = efdp->efd_next_extent;
385 ASSERT(next_extent < efdp->efd_format.efd_nextents);
386 extp = &(efdp->efd_format.efd_extents[next_extent]);
387 extp->ext_start = start_block;
388 extp->ext_len = ext_len;
389 efdp->efd_next_extent++;
394 /* Sort bmap items by AG. */
396 xfs_extent_free_diff_items(
401 struct xfs_mount *mp = priv;
402 struct xfs_extent_free_item *ra;
403 struct xfs_extent_free_item *rb;
405 ra = container_of(a, struct xfs_extent_free_item, xefi_list);
406 rb = container_of(b, struct xfs_extent_free_item, xefi_list);
407 return XFS_FSB_TO_AGNO(mp, ra->xefi_startblock) -
408 XFS_FSB_TO_AGNO(mp, rb->xefi_startblock);
411 /* Log a free extent to the intent item. */
413 xfs_extent_free_log_item(
414 struct xfs_trans *tp,
415 struct xfs_efi_log_item *efip,
416 struct xfs_extent_free_item *free)
419 struct xfs_extent *extp;
421 tp->t_flags |= XFS_TRANS_DIRTY;
422 set_bit(XFS_LI_DIRTY, &efip->efi_item.li_flags);
425 * atomic_inc_return gives us the value after the increment;
426 * we want to use it as an array index so we need to subtract 1 from
429 next_extent = atomic_inc_return(&efip->efi_next_extent) - 1;
430 ASSERT(next_extent < efip->efi_format.efi_nextents);
431 extp = &efip->efi_format.efi_extents[next_extent];
432 extp->ext_start = free->xefi_startblock;
433 extp->ext_len = free->xefi_blockcount;
436 static struct xfs_log_item *
437 xfs_extent_free_create_intent(
438 struct xfs_trans *tp,
439 struct list_head *items,
443 struct xfs_mount *mp = tp->t_mountp;
444 struct xfs_efi_log_item *efip = xfs_efi_init(mp, count);
445 struct xfs_extent_free_item *free;
449 xfs_trans_add_item(tp, &efip->efi_item);
451 list_sort(mp, items, xfs_extent_free_diff_items);
452 list_for_each_entry(free, items, xefi_list)
453 xfs_extent_free_log_item(tp, efip, free);
454 return &efip->efi_item;
457 /* Get an EFD so we can process all the free extents. */
458 static struct xfs_log_item *
459 xfs_extent_free_create_done(
460 struct xfs_trans *tp,
461 struct xfs_log_item *intent,
464 return &xfs_trans_get_efd(tp, EFI_ITEM(intent), count)->efd_item;
467 /* Process a free extent. */
469 xfs_extent_free_finish_item(
470 struct xfs_trans *tp,
471 struct xfs_log_item *done,
472 struct list_head *item,
473 struct xfs_btree_cur **state)
475 struct xfs_extent_free_item *free;
478 free = container_of(item, struct xfs_extent_free_item, xefi_list);
479 error = xfs_trans_free_extent(tp, EFD_ITEM(done),
480 free->xefi_startblock,
481 free->xefi_blockcount,
482 &free->xefi_oinfo, free->xefi_skip_discard);
487 /* Abort all pending EFIs. */
489 xfs_extent_free_abort_intent(
490 struct xfs_log_item *intent)
492 xfs_efi_release(EFI_ITEM(intent));
495 /* Cancel a free extent. */
497 xfs_extent_free_cancel_item(
498 struct list_head *item)
500 struct xfs_extent_free_item *free;
502 free = container_of(item, struct xfs_extent_free_item, xefi_list);
506 const struct xfs_defer_op_type xfs_extent_free_defer_type = {
507 .max_items = XFS_EFI_MAX_FAST_EXTENTS,
508 .create_intent = xfs_extent_free_create_intent,
509 .abort_intent = xfs_extent_free_abort_intent,
510 .create_done = xfs_extent_free_create_done,
511 .finish_item = xfs_extent_free_finish_item,
512 .cancel_item = xfs_extent_free_cancel_item,
516 * AGFL blocks are accounted differently in the reserve pools and are not
517 * inserted into the busy extent list.
520 xfs_agfl_free_finish_item(
521 struct xfs_trans *tp,
522 struct xfs_log_item *done,
523 struct list_head *item,
524 struct xfs_btree_cur **state)
526 struct xfs_mount *mp = tp->t_mountp;
527 struct xfs_efd_log_item *efdp = EFD_ITEM(done);
528 struct xfs_extent_free_item *free;
529 struct xfs_extent *extp;
530 struct xfs_buf *agbp;
536 free = container_of(item, struct xfs_extent_free_item, xefi_list);
537 ASSERT(free->xefi_blockcount == 1);
538 agno = XFS_FSB_TO_AGNO(mp, free->xefi_startblock);
539 agbno = XFS_FSB_TO_AGBNO(mp, free->xefi_startblock);
541 trace_xfs_agfl_free_deferred(mp, agno, 0, agbno, free->xefi_blockcount);
543 error = xfs_alloc_read_agf(mp, tp, agno, 0, &agbp);
545 error = xfs_free_agfl_block(tp, agno, agbno, agbp,
549 * Mark the transaction dirty, even on error. This ensures the
550 * transaction is aborted, which:
552 * 1.) releases the EFI and frees the EFD
553 * 2.) shuts down the filesystem
555 tp->t_flags |= XFS_TRANS_DIRTY;
556 set_bit(XFS_LI_DIRTY, &efdp->efd_item.li_flags);
558 next_extent = efdp->efd_next_extent;
559 ASSERT(next_extent < efdp->efd_format.efd_nextents);
560 extp = &(efdp->efd_format.efd_extents[next_extent]);
561 extp->ext_start = free->xefi_startblock;
562 extp->ext_len = free->xefi_blockcount;
563 efdp->efd_next_extent++;
569 /* sub-type with special handling for AGFL deferred frees */
570 const struct xfs_defer_op_type xfs_agfl_free_defer_type = {
571 .max_items = XFS_EFI_MAX_FAST_EXTENTS,
572 .create_intent = xfs_extent_free_create_intent,
573 .abort_intent = xfs_extent_free_abort_intent,
574 .create_done = xfs_extent_free_create_done,
575 .finish_item = xfs_agfl_free_finish_item,
576 .cancel_item = xfs_extent_free_cancel_item,
580 * Process an extent free intent item that was recovered from
581 * the log. We need to free the extents that it describes.
584 xfs_efi_item_recover(
585 struct xfs_log_item *lip,
586 struct xfs_trans *parent_tp)
588 struct xfs_efi_log_item *efip = EFI_ITEM(lip);
589 struct xfs_mount *mp = parent_tp->t_mountp;
590 struct xfs_efd_log_item *efdp;
591 struct xfs_trans *tp;
592 struct xfs_extent *extp;
593 xfs_fsblock_t startblock_fsb;
598 * First check the validity of the extents described by the
599 * EFI. If any are bad, then assume that all are bad and
602 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
603 extp = &efip->efi_format.efi_extents[i];
604 startblock_fsb = XFS_BB_TO_FSB(mp,
605 XFS_FSB_TO_DADDR(mp, extp->ext_start));
606 if (startblock_fsb == 0 ||
607 extp->ext_len == 0 ||
608 startblock_fsb >= mp->m_sb.sb_dblocks ||
609 extp->ext_len >= mp->m_sb.sb_agblocks) {
611 * This will pull the EFI from the AIL and
612 * free the memory associated with it.
614 xfs_efi_release(efip);
615 return -EFSCORRUPTED;
619 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
622 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
624 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
625 extp = &efip->efi_format.efi_extents[i];
626 error = xfs_trans_free_extent(tp, efdp, extp->ext_start,
628 &XFS_RMAP_OINFO_ANY_OWNER, false);
634 error = xfs_trans_commit(tp);
638 xfs_trans_cancel(tp);
644 struct xfs_log_item *lip,
647 return EFI_ITEM(lip)->efi_format.efi_id == intent_id;
650 static const struct xfs_item_ops xfs_efi_item_ops = {
651 .iop_size = xfs_efi_item_size,
652 .iop_format = xfs_efi_item_format,
653 .iop_unpin = xfs_efi_item_unpin,
654 .iop_release = xfs_efi_item_release,
655 .iop_recover = xfs_efi_item_recover,
656 .iop_match = xfs_efi_item_match,
660 * This routine is called to create an in-core extent free intent
661 * item from the efi format structure which was logged on disk.
662 * It allocates an in-core efi, copies the extents from the format
663 * structure into it, and adds the efi to the AIL with the given
667 xlog_recover_efi_commit_pass2(
669 struct list_head *buffer_list,
670 struct xlog_recover_item *item,
673 struct xfs_mount *mp = log->l_mp;
674 struct xfs_efi_log_item *efip;
675 struct xfs_efi_log_format *efi_formatp;
678 efi_formatp = item->ri_buf[0].i_addr;
680 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
681 error = xfs_efi_copy_format(&item->ri_buf[0], &efip->efi_format);
683 xfs_efi_item_free(efip);
686 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
688 * Insert the intent into the AIL directly and drop one reference so
689 * that finishing or canceling the work will drop the other.
691 xfs_trans_ail_insert(log->l_ailp, &efip->efi_item, lsn);
692 xfs_efi_release(efip);
696 const struct xlog_recover_item_ops xlog_efi_item_ops = {
697 .item_type = XFS_LI_EFI,
698 .commit_pass2 = xlog_recover_efi_commit_pass2,
702 * This routine is called when an EFD format structure is found in a committed
703 * transaction in the log. Its purpose is to cancel the corresponding EFI if it
704 * was still in the log. To do this it searches the AIL for the EFI with an id
705 * equal to that in the EFD format structure. If we find it we drop the EFD
706 * reference, which removes the EFI from the AIL and frees it.
709 xlog_recover_efd_commit_pass2(
711 struct list_head *buffer_list,
712 struct xlog_recover_item *item,
715 struct xfs_efd_log_format *efd_formatp;
717 efd_formatp = item->ri_buf[0].i_addr;
718 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
719 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
720 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
721 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
723 xlog_recover_release_intent(log, XFS_LI_EFI, efd_formatp->efd_efi_id);
727 const struct xlog_recover_item_ops xlog_efd_item_ops = {
728 .item_type = XFS_LI_EFD,
729 .commit_pass2 = xlog_recover_efd_commit_pass2,