1 // SPDX-License-Identifier: GPL-2.0-or-later
3  * Copyright (C) 2022-2023 Oracle.  All Rights Reserved.
58  * if the rmapbt says there is an owner of this block other than @oinfo, then
59  * the block is crosslinked.  Remove the reverse mapping and continue.
61  * If there is one rmap record, we can free the block, which removes the
62  * reverse mapping but doesn't add the block to the free space.  Our repair
63  * strategy is to hope the other metadata objects crosslinked on this block
67  * If there are no rmap records at all, we also free the block.  If the btree
70  * had to have been an rmap entry for the block to have ended up on @bitmap,
74  * the btree we're trying to rebuild and the block is indeed owned by another
75  * data structure with the same rmap owner, then the block will be in sublist
77  * with only the same rmap owner but the block is not owned by something with
78  * the same rmap owner, the block will be freed.
107 /* Put a block back on the AGFL. */
122 	 * Since we're "freeing" a lost block onto the AGFL, we have to  in xreap_put_freelist()
123 	 * create an rmap for the block prior to merging it or else other  in xreap_put_freelist()
126 	error = xfs_rmap_alloc(sc->tp, sc->sa.agf_bp, sc->sa.pag, agbno, 1,  in xreap_put_freelist()
131 	/* Put the block on the AGFL. */  in xreap_put_freelist()
132 	error = xfs_alloc_read_agfl(sc->sa.pag, sc->tp, &agfl_bp);  in xreap_put_freelist()
136 	error = xfs_alloc_put_freelist(sc->sa.pag, sc->tp, sc->sa.agf_bp,  in xreap_put_freelist()
140 	xfs_extent_busy_insert(sc->tp, sc->sa.pag, agbno, 1,  in xreap_put_freelist()
149 	if (rs->force_roll)  in xreap_dirty()
151 	if (rs->deferred)  in xreap_dirty()
153 	if (rs->invalidated)  in xreap_dirty()
155 	if (rs->total_deferred)  in xreap_dirty()
170 	if (rs->force_roll)  in xreap_want_roll()
172 	if (rs->deferred > XREP_MAX_ITRUNCATE_EFIS)  in xreap_want_roll()
174 	if (rs->invalidated > XREAP_MAX_BINVAL)  in xreap_want_roll()
181 	rs->total_deferred += rs->deferred;  in xreap_reset()
182 	rs->deferred = 0;  in xreap_reset()
183 	rs->invalidated = 0;  in xreap_reset()
184 	rs->force_roll = false;  in xreap_reset()
199 	if (rs->force_roll)  in xreap_want_defer_finish()
201 	if (rs->total_deferred > XREAP_MAX_DEFER_CHAIN)  in xreap_want_defer_finish()
208 	rs->total_deferred = 0;  in xreap_defer_finish_reset()
209 	rs->deferred = 0;  in xreap_defer_finish_reset()
210 	rs->invalidated = 0;  in xreap_defer_finish_reset()
211 	rs->force_roll = false;  in xreap_defer_finish_reset()
240 	scan->__sector_count += scan->daddr_step;  in xrep_bufscan_advance()
241 	while (scan->__sector_count <= scan->max_sectors) {  in xrep_bufscan_advance()
245 		error = xfs_buf_incore(mp->m_ddev_targp, scan->daddr,  in xrep_bufscan_advance()
246 				scan->__sector_count, XBF_LIVESCAN, &bp);  in xrep_bufscan_advance()
250 		scan->__sector_count += scan->daddr_step;  in xrep_bufscan_advance()
263 	struct xfs_scrub	*sc = rs->sc;  in xreap_agextent_binval()
264 	struct xfs_perag	*pag = sc->sa.pag;  in xreap_agextent_binval()
265 	struct xfs_mount	*mp = sc->mp;  in xreap_agextent_binval()
266 	xfs_agnumber_t		agno = sc->sa.pag->pag_agno;  in xreap_agextent_binval()
271 	 * Avoid invalidating AG headers and post-EOFS blocks because we never  in xreap_agextent_binval()
275 	    !xfs_verify_agbno(pag, agbno_next - 1))  in xreap_agextent_binval()
289 							agbno_next - bno),  in xreap_agextent_binval()
295 			xfs_trans_bjoin(sc->tp, bp);  in xreap_agextent_binval()
296 			xfs_trans_binval(sc->tp, bp);  in xreap_agextent_binval()
297 			rs->invalidated++;  in xreap_agextent_binval()
304 			if (rs->invalidated > XREAP_MAX_BINVAL) {  in xreap_agextent_binval()
305 				*aglenp -= agbno_next - bno;  in xreap_agextent_binval()
314 	trace_xreap_agextent_binval(sc->sa.pag, agbno, *aglenp);  in xreap_agextent_binval()
319  * call.  Cross-linked blocks should have their reverse mappings removed, but
320  * single-owner extents can be freed.  AGFL blocks can only be put back one at
331 	struct xfs_scrub	*sc = rs->sc;  in xreap_agextent_select()
339 	 * block of this extent.  If so, the block is crosslinked.  in xreap_agextent_select()
341 	cur = xfs_rmapbt_init_cursor(sc->mp, sc->tp, sc->sa.agf_bp,  in xreap_agextent_select()
342 			sc->sa.pag);  in xreap_agextent_select()
343 	error = xfs_rmap_has_other_keys(cur, agbno, 1, rs->oinfo,  in xreap_agextent_select()
349 	if (rs->resv == XFS_AG_RESV_AGFL)  in xreap_agextent_select()
359 		error = xfs_rmap_has_other_keys(cur, bno, 1, rs->oinfo,  in xreap_agextent_select()
373 	trace_xreap_agextent_select(sc->sa.pag, agbno, len, *crosslinked);  in xreap_agextent_select()
390 	struct xfs_scrub	*sc = rs->sc;  in xreap_agextent_iter()
394 	fsbno = XFS_AGB_TO_FSB(sc->mp, sc->sa.pag->pag_agno, agbno);  in xreap_agextent_iter()
397 	 * If there are other rmappings, this block is cross linked and must  in xreap_agextent_iter()
399 	 * we were the only owner of the block, so free the extent, which will  in xreap_agextent_iter()
402 	 * XXX: XFS doesn't support detecting the case where a single block  in xreap_agextent_iter()
403 	 * metadata structure is crosslinked with a multi-block structure  in xreap_agextent_iter()
410 		trace_xreap_dispose_unmap_extent(sc->sa.pag, agbno, *aglenp);  in xreap_agextent_iter()
412 		rs->force_roll = true;  in xreap_agextent_iter()
414 		if (rs->oinfo == &XFS_RMAP_OINFO_COW) {  in xreap_agextent_iter()
420 			xfs_refcount_free_cow_extent(sc->tp, fsbno, *aglenp);  in xreap_agextent_iter()
424 		return xfs_rmap_free(sc->tp, sc->sa.agf_bp, sc->sa.pag, agbno,  in xreap_agextent_iter()
425 				*aglenp, rs->oinfo);  in xreap_agextent_iter()
428 	trace_xreap_dispose_free_extent(sc->sa.pag, agbno, *aglenp);  in xreap_agextent_iter()
449 	if (rs->oinfo == &XFS_RMAP_OINFO_COW) {  in xreap_agextent_iter()
450 		ASSERT(rs->resv == XFS_AG_RESV_NONE);  in xreap_agextent_iter()
452 		xfs_refcount_free_cow_extent(sc->tp, fsbno, *aglenp);  in xreap_agextent_iter()
453 		error = xfs_free_extent_later(sc->tp, fsbno, *aglenp, NULL,  in xreap_agextent_iter()
454 				rs->resv, XFS_FREE_EXTENT_SKIP_DISCARD);  in xreap_agextent_iter()
458 		rs->force_roll = true;  in xreap_agextent_iter()
463 	if (rs->resv == XFS_AG_RESV_AGFL) {  in xreap_agextent_iter()
469 		rs->force_roll = true;  in xreap_agextent_iter()
479 	error = xfs_free_extent_later(sc->tp, fsbno, *aglenp, rs->oinfo,  in xreap_agextent_iter()
480 			rs->resv, XFS_FREE_EXTENT_SKIP_DISCARD);  in xreap_agextent_iter()
484 	rs->deferred++;  in xreap_agextent_iter()
485 	if (rs->deferred % 2 == 0)  in xreap_agextent_iter()
486 		xfs_defer_add_barrier(sc->tp);  in xreap_agextent_iter()
491  * Break an AG metadata extent into sub-extents by fate (crosslinked, not
492  * crosslinked), and dispose of each sub-extent separately.
501 	struct xfs_scrub	*sc = rs->sc;  in xreap_agmeta_extent()
506 	ASSERT(sc->ip == NULL);  in xreap_agmeta_extent()
539 /* Dispose of every block of every AG metadata extent in the bitmap. */
554 	ASSERT(xfs_has_rmapbt(sc->mp));  in xrep_reap_agblocks()
555 	ASSERT(sc->ip == NULL);  in xrep_reap_agblocks()
568  * Break a file metadata extent into sub-extents by fate (crosslinked, not
569  * crosslinked), and dispose of each sub-extent separately.  The extent must
579 	struct xfs_scrub	*sc = rs->sc;  in xreap_fsmeta_extent()
580 	xfs_agnumber_t		agno = XFS_FSB_TO_AGNO(sc->mp, fsbno);  in xreap_fsmeta_extent()
581 	xfs_agblock_t		agbno = XFS_FSB_TO_AGBNO(sc->mp, fsbno);  in xreap_fsmeta_extent()
586 	ASSERT(sc->ip != NULL);  in xreap_fsmeta_extent()
587 	ASSERT(!sc->sa.pag);  in xreap_fsmeta_extent()
593 	sc->sa.pag = xfs_perag_get(sc->mp, agno);  in xreap_fsmeta_extent()
594 	if (!sc->sa.pag)  in xreap_fsmeta_extent()
595 		return -EFSCORRUPTED;  in xreap_fsmeta_extent()
597 	error = xfs_alloc_read_agf(sc->sa.pag, sc->tp, 0, &sc->sa.agf_bp);  in xreap_fsmeta_extent()
629 			xfs_trans_bhold(sc->tp, sc->sa.agf_bp);  in xreap_fsmeta_extent()
630 			error = xfs_trans_roll_inode(&sc->tp, sc->ip);  in xreap_fsmeta_extent()
631 			xfs_trans_bjoin(sc->tp, sc->sa.agf_bp);  in xreap_fsmeta_extent()
641 	xfs_trans_brelse(sc->tp, sc->sa.agf_bp);  in xreap_fsmeta_extent()
642 	sc->sa.agf_bp = NULL;  in xreap_fsmeta_extent()
644 	xfs_perag_put(sc->sa.pag);  in xreap_fsmeta_extent()
645 	sc->sa.pag = NULL;  in xreap_fsmeta_extent()
650  * Dispose of every block of every fs metadata extent in the bitmap.
666 	ASSERT(xfs_has_rmapbt(sc->mp));  in xrep_reap_fsblocks()
667 	ASSERT(sc->ip != NULL);  in xrep_reap_fsblocks()
684  * This first step determines the longest subset of the passed-in imap
691 	struct xfs_inode	*ip,  in xreap_bmapi_select()  argument
704 	agbno = XFS_FSB_TO_AGBNO(sc->mp, imap->br_startblock);  in xreap_bmapi_select()
705 	agbno_next = agbno + imap->br_blockcount;  in xreap_bmapi_select()
707 	cur = xfs_rmapbt_init_cursor(sc->mp, sc->tp, sc->sa.agf_bp,  in xreap_bmapi_select()
708 			sc->sa.pag);  in xreap_bmapi_select()
710 	xfs_rmap_ino_owner(&oinfo, ip->i_ino, whichfork, imap->br_startoff);  in xreap_bmapi_select()
732 	imap->br_blockcount = len;  in xreap_bmapi_select()
733 	trace_xreap_bmapi_select(sc->sa.pag, agbno, len, *crosslinked);  in xreap_bmapi_select()
752 	for (i = 0; i < bp->b_map_count; i++) {  in xreap_buf_loggable()
756 		chunks = DIV_ROUND_UP(BBTOB(bp->b_maps[i].bm_len),  in xreap_buf_loggable()
773 	struct xfs_inode	*ip,  in xreap_bmapi_binval()  argument
777 	struct xfs_mount	*mp = sc->mp;  in xreap_bmapi_binval()
778 	struct xfs_perag	*pag = sc->sa.pag;  in xreap_bmapi_binval()
783 	xfs_agnumber_t		agno = sc->sa.pag->pag_agno;  in xreap_bmapi_binval()
791 	 * Avoid invalidating AG headers and post-EOFS blocks because we never  in xreap_bmapi_binval()
794 	agbno = bno = XFS_FSB_TO_AGBNO(sc->mp, imap->br_startblock);  in xreap_bmapi_binval()
795 	agbno_next = agbno + imap->br_blockcount;  in xreap_bmapi_binval()
797 	    !xfs_verify_agbno(pag, agbno_next - 1))  in xreap_bmapi_binval()
802 	 * means that for each block in the mapping, there could exist an  in xreap_bmapi_binval()
803 	 * xfs_buf indexed by that block with any length up to the maximum  in xreap_bmapi_binval()
808 	off = imap->br_startoff + imap->br_blockcount;  in xreap_bmapi_binval()
814 		error = xfs_bmapi_read(ip, off, max_off - off, &hmap,  in xreap_bmapi_binval()
820 			return -EFSCORRUPTED;  in xreap_bmapi_binval()
828 	scan_blocks = off - imap->br_startoff;  in xreap_bmapi_binval()
849 				xfs_trans_bjoin(sc->tp, bp);  in xreap_bmapi_binval()
850 				xfs_trans_binval(sc->tp, bp);  in xreap_bmapi_binval()
863 				imap->br_blockcount = agbno_next - bno;  in xreap_bmapi_binval()
869 		scan_blocks--;  in xreap_bmapi_binval()
873 	trace_xreap_bmapi_binval(sc->sa.pag, agbno, imap->br_blockcount);  in xreap_bmapi_binval()
879  * The number of blocks disposed of is returned in @imap->br_blockcount.
884 	struct xfs_inode		*ip,  in xrep_reap_bmapi_iter()  argument
893 		 * If there are other rmappings, this block is cross linked and  in xrep_reap_bmapi_iter()
897 		 * anybody else who thinks they own the block, even though that  in xrep_reap_bmapi_iter()
900 		trace_xreap_dispose_unmap_extent(sc->sa.pag,  in xrep_reap_bmapi_iter()
901 				XFS_FSB_TO_AGBNO(sc->mp, imap->br_startblock),  in xrep_reap_bmapi_iter()
902 				imap->br_blockcount);  in xrep_reap_bmapi_iter()
909 		xfs_bmap_unmap_extent(sc->tp, ip, whichfork, imap);  in xrep_reap_bmapi_iter()
910 		xfs_trans_mod_dquot_byino(sc->tp, ip, XFS_TRANS_DQ_BCOUNT,  in xrep_reap_bmapi_iter()
911 				-(int64_t)imap->br_blockcount);  in xrep_reap_bmapi_iter()
912 		xfs_rmap_unmap_extent(sc->tp, ip, whichfork, imap);  in xrep_reap_bmapi_iter()
917 	 * If the block is not crosslinked, we can invalidate all the incore  in xrep_reap_bmapi_iter()
920 	 * by a block starting before the first block of the extent but overlap  in xrep_reap_bmapi_iter()
923 	trace_xreap_dispose_free_extent(sc->sa.pag,  in xrep_reap_bmapi_iter()
924 			XFS_FSB_TO_AGBNO(sc->mp, imap->br_startblock),  in xrep_reap_bmapi_iter()
925 			imap->br_blockcount);  in xrep_reap_bmapi_iter()
933 	error = xreap_bmapi_binval(sc, ip, whichfork, imap);  in xrep_reap_bmapi_iter()
934 	if (error || imap->br_blockcount == 0)  in xrep_reap_bmapi_iter()
942 	xfs_bmap_unmap_extent(sc->tp, ip, whichfork, imap);  in xrep_reap_bmapi_iter()
943 	xfs_trans_mod_dquot_byino(sc->tp, ip, XFS_TRANS_DQ_BCOUNT,  in xrep_reap_bmapi_iter()
944 			-(int64_t)imap->br_blockcount);  in xrep_reap_bmapi_iter()
945 	return xfs_free_extent_later(sc->tp, imap->br_startblock,  in xrep_reap_bmapi_iter()
946 			imap->br_blockcount, NULL, XFS_AG_RESV_NONE,  in xrep_reap_bmapi_iter()
957 	struct xfs_inode		*ip,  in xreap_ifork_extent()  argument
965 	ASSERT(sc->sa.pag == NULL);  in xreap_ifork_extent()
967 	trace_xreap_ifork_extent(sc, ip, whichfork, imap);  in xreap_ifork_extent()
969 	agno = XFS_FSB_TO_AGNO(sc->mp, imap->br_startblock);  in xreap_ifork_extent()
970 	sc->sa.pag = xfs_perag_get(sc->mp, agno);  in xreap_ifork_extent()
971 	if (!sc->sa.pag)  in xreap_ifork_extent()
972 		return -EFSCORRUPTED;  in xreap_ifork_extent()
974 	error = xfs_alloc_read_agf(sc->sa.pag, sc->tp, 0, &sc->sa.agf_bp);  in xreap_ifork_extent()
982 	error = xreap_bmapi_select(sc, ip, whichfork, imap, &crosslinked);  in xreap_ifork_extent()
986 	error = xrep_reap_bmapi_iter(sc, ip, whichfork, imap, crosslinked);  in xreap_ifork_extent()
991 	xfs_trans_brelse(sc->tp, sc->sa.agf_bp);  in xreap_ifork_extent()
992 	sc->sa.agf_bp = NULL;  in xreap_ifork_extent()
994 	xfs_perag_put(sc->sa.pag);  in xreap_ifork_extent()
995 	sc->sa.pag = NULL;  in xreap_ifork_extent()
1000  * Dispose of each block mapped to the given fork of the given file.  Callers
1001  * must hold ILOCK_EXCL, and ip can only be sc->ip or sc->tempip.  The fork
1007 	struct xfs_inode	*ip,  in xrep_reap_ifork()  argument
1014 	ASSERT(xfs_has_rmapbt(sc->mp));  in xrep_reap_ifork()
1015 	ASSERT(ip == sc->ip || ip == sc->tempip);  in xrep_reap_ifork()
1016 	ASSERT(whichfork == XFS_ATTR_FORK || !XFS_IS_REALTIME_INODE(ip));  in xrep_reap_ifork()
1023 		error = xfs_bmapi_read(ip, off, XFS_MAX_FILEOFF - off, &imap,  in xrep_reap_ifork()
1029 			return -EFSCORRUPTED;  in xrep_reap_ifork()
1037 			error = xreap_ifork_extent(sc, ip, whichfork, &imap);  in xrep_reap_ifork()
1041 			error = xfs_defer_finish(&sc->tp);  in xrep_reap_ifork()