init_waitqueue_head(&pag->pag_active_wq);
pag->pagb_count = 0;
pag->pagb_tree = RB_ROOT;
+ xfs_hooks_init(&pag->pag_rmap_update_hooks);
#endif /* __KERNEL__ */
error = xfs_buf_cache_init(&pag->pag_bcache);
* inconsistencies.
*/
struct xfs_defer_drain pag_intents_drain;
+
+ /* Hook to feed rmapbt updates to an active online repair. */
+ struct xfs_hooks pag_rmap_update_hooks;
#endif /* __KERNEL__ */
};
return error;
}
+#ifdef CONFIG_XFS_LIVE_HOOKS
+/*
+ * Use a static key here to reduce the overhead of rmapbt live updates. If
+ * the compiler supports jump labels, the static branch will be replaced by a
+ * nop sled when there are no hook users. Online fsck is currently the only
+ * caller, so this is a reasonable tradeoff.
+ *
+ * Note: Patching the kernel code requires taking the cpu hotplug lock. Other
+ * parts of the kernel allocate memory with that lock held, which means that
+ * XFS callers cannot hold any locks that might be used by memory reclaim or
+ * writeback when calling the static_branch_{inc,dec} functions.
+ */
+DEFINE_STATIC_XFS_HOOK_SWITCH(xfs_rmap_hooks_switch);
+
+void
+xfs_rmap_hook_disable(void)
+{
+ xfs_hooks_switch_off(&xfs_rmap_hooks_switch);
+}
+
+void
+xfs_rmap_hook_enable(void)
+{
+ xfs_hooks_switch_on(&xfs_rmap_hooks_switch);
+}
+
+/* Call downstream hooks for a reverse mapping update. */
+static inline void
+xfs_rmap_update_hook(
+ struct xfs_trans *tp,
+ struct xfs_perag *pag,
+ enum xfs_rmap_intent_type op,
+ xfs_agblock_t startblock,
+ xfs_extlen_t blockcount,
+ bool unwritten,
+ const struct xfs_owner_info *oinfo)
+{
+ if (xfs_hooks_switched_on(&xfs_rmap_hooks_switch)) {
+ struct xfs_rmap_update_params p = {
+ .startblock = startblock,
+ .blockcount = blockcount,
+ .unwritten = unwritten,
+ .oinfo = *oinfo, /* struct copy */
+ };
+
+ if (pag)
+ xfs_hooks_call(&pag->pag_rmap_update_hooks, op, &p);
+ }
+}
+
+/* Call the specified function during a reverse mapping update. */
+int
+xfs_rmap_hook_add(
+ struct xfs_perag *pag,
+ struct xfs_rmap_hook *hook)
+{
+ return xfs_hooks_add(&pag->pag_rmap_update_hooks, &hook->rmap_hook);
+}
+
+/* Stop calling the specified function during a reverse mapping update. */
+void
+xfs_rmap_hook_del(
+ struct xfs_perag *pag,
+ struct xfs_rmap_hook *hook)
+{
+ xfs_hooks_del(&pag->pag_rmap_update_hooks, &hook->rmap_hook);
+}
+
+/* Configure rmap update hook functions. */
+void
+xfs_rmap_hook_setup(
+ struct xfs_rmap_hook *hook,
+ notifier_fn_t mod_fn)
+{
+ xfs_hook_setup(&hook->rmap_hook, mod_fn);
+}
+#else
+# define xfs_rmap_update_hook(t, p, o, s, b, u, oi) do { } while (0)
+#endif /* CONFIG_XFS_LIVE_HOOKS */
+
/*
* Remove a reference to an extent in the rmap btree.
*/
return 0;
cur = xfs_rmapbt_init_cursor(mp, tp, agbp, pag);
-
+ xfs_rmap_update_hook(tp, pag, XFS_RMAP_UNMAP, bno, len, false, oinfo);
error = xfs_rmap_unmap(cur, bno, len, false, oinfo);
xfs_btree_del_cursor(cur, error);
return 0;
cur = xfs_rmapbt_init_cursor(mp, tp, agbp, pag);
+ xfs_rmap_update_hook(tp, pag, XFS_RMAP_MAP, bno, len, false, oinfo);
error = xfs_rmap_map(cur, bno, len, false, oinfo);
xfs_btree_del_cursor(cur, error);
xfs_trans_brelse(tp, agbp);
}
+/* Commit an rmap operation into the ondisk tree. */
+int
+__xfs_rmap_finish_intent(
+ struct xfs_btree_cur *rcur,
+ enum xfs_rmap_intent_type op,
+ xfs_agblock_t bno,
+ xfs_extlen_t len,
+ const struct xfs_owner_info *oinfo,
+ bool unwritten)
+{
+ switch (op) {
+ case XFS_RMAP_ALLOC:
+ case XFS_RMAP_MAP:
+ return xfs_rmap_map(rcur, bno, len, unwritten, oinfo);
+ case XFS_RMAP_MAP_SHARED:
+ return xfs_rmap_map_shared(rcur, bno, len, unwritten, oinfo);
+ case XFS_RMAP_FREE:
+ case XFS_RMAP_UNMAP:
+ return xfs_rmap_unmap(rcur, bno, len, unwritten, oinfo);
+ case XFS_RMAP_UNMAP_SHARED:
+ return xfs_rmap_unmap_shared(rcur, bno, len, unwritten, oinfo);
+ case XFS_RMAP_CONVERT:
+ return xfs_rmap_convert(rcur, bno, len, !unwritten, oinfo);
+ case XFS_RMAP_CONVERT_SHARED:
+ return xfs_rmap_convert_shared(rcur, bno, len, !unwritten,
+ oinfo);
+ default:
+ ASSERT(0);
+ return -EFSCORRUPTED;
+ }
+}
+
/*
* Process one of the deferred rmap operations. We pass back the
* btree cursor to maintain our lock on the rmapbt between calls.
unwritten = ri->ri_bmap.br_state == XFS_EXT_UNWRITTEN;
bno = XFS_FSB_TO_AGBNO(rcur->bc_mp, ri->ri_bmap.br_startblock);
- switch (ri->ri_type) {
- case XFS_RMAP_ALLOC:
- case XFS_RMAP_MAP:
- error = xfs_rmap_map(rcur, bno, ri->ri_bmap.br_blockcount,
- unwritten, &oinfo);
- break;
- case XFS_RMAP_MAP_SHARED:
- error = xfs_rmap_map_shared(rcur, bno,
- ri->ri_bmap.br_blockcount, unwritten, &oinfo);
- break;
- case XFS_RMAP_FREE:
- case XFS_RMAP_UNMAP:
- error = xfs_rmap_unmap(rcur, bno, ri->ri_bmap.br_blockcount,
- unwritten, &oinfo);
- break;
- case XFS_RMAP_UNMAP_SHARED:
- error = xfs_rmap_unmap_shared(rcur, bno,
- ri->ri_bmap.br_blockcount, unwritten, &oinfo);
- break;
- case XFS_RMAP_CONVERT:
- error = xfs_rmap_convert(rcur, bno, ri->ri_bmap.br_blockcount,
- !unwritten, &oinfo);
- break;
- case XFS_RMAP_CONVERT_SHARED:
- error = xfs_rmap_convert_shared(rcur, bno,
- ri->ri_bmap.br_blockcount, !unwritten, &oinfo);
- break;
- default:
- ASSERT(0);
- error = -EFSCORRUPTED;
- }
+ error = __xfs_rmap_finish_intent(rcur, ri->ri_type, bno,
+ ri->ri_bmap.br_blockcount, &oinfo, unwritten);
+ if (error)
+ return error;
- return error;
+ xfs_rmap_update_hook(tp, ri->ri_pag, ri->ri_type, bno,
+ ri->ri_bmap.br_blockcount, unwritten, &oinfo);
+ return 0;
}
/*
struct xfs_btree_cur *rcur, int error);
int xfs_rmap_finish_one(struct xfs_trans *tp, struct xfs_rmap_intent *ri,
struct xfs_btree_cur **pcur);
+int __xfs_rmap_finish_intent(struct xfs_btree_cur *rcur,
+ enum xfs_rmap_intent_type op, xfs_agblock_t bno,
+ xfs_extlen_t len, const struct xfs_owner_info *oinfo,
+ bool unwritten);
int xfs_rmap_lookup_le_range(struct xfs_btree_cur *cur, xfs_agblock_t bno,
uint64_t owner, uint64_t offset, unsigned int flags,
int __init xfs_rmap_intent_init_cache(void);
void xfs_rmap_intent_destroy_cache(void);
+/*
+ * Parameters for tracking reverse mapping changes. The hook function arg
+ * parameter is enum xfs_rmap_intent_type, and the rest is below.
+ */
+struct xfs_rmap_update_params {
+ xfs_agblock_t startblock;
+ xfs_extlen_t blockcount;
+ struct xfs_owner_info oinfo;
+ bool unwritten;
+};
+
+#ifdef CONFIG_XFS_LIVE_HOOKS
+
+struct xfs_rmap_hook {
+ struct xfs_hook rmap_hook;
+};
+
+void xfs_rmap_hook_disable(void);
+void xfs_rmap_hook_enable(void);
+
+int xfs_rmap_hook_add(struct xfs_perag *pag, struct xfs_rmap_hook *hook);
+void xfs_rmap_hook_del(struct xfs_perag *pag, struct xfs_rmap_hook *hook);
+void xfs_rmap_hook_setup(struct xfs_rmap_hook *hook, notifier_fn_t mod_fn);
+#endif
+
#endif /* __XFS_RMAP_H__ */
if (scrub_fsgates & XCHK_FSGATES_DIRENTS)
xfs_dir_hook_enable();
+ if (scrub_fsgates & XCHK_FSGATES_RMAP)
+ xfs_rmap_hook_enable();
+
sc->flags |= scrub_fsgates;
}
return xmbuf_alloc(sc->mp, descr, &sc->xmbtp);
}
+
+/*
+ * Create a dummy transaction for use in a live update hook function. This
+ * function MUST NOT be called from regular repair code because the current
+ * process' transaction is saved via the cookie.
+ */
+int
+xrep_trans_alloc_hook_dummy(
+ struct xfs_mount *mp,
+ void **cookiep,
+ struct xfs_trans **tpp)
+{
+ int error;
+
+ *cookiep = current->journal_info;
+ current->journal_info = NULL;
+
+ error = xfs_trans_alloc_empty(mp, tpp);
+ if (!error)
+ return 0;
+
+ current->journal_info = *cookiep;
+ *cookiep = NULL;
+ return error;
+}
+
+/* Cancel a dummy transaction used by a live update hook function. */
+void
+xrep_trans_cancel_hook_dummy(
+ void **cookiep,
+ struct xfs_trans *tp)
+{
+ xfs_trans_cancel(tp);
+ current->journal_info = *cookiep;
+ *cookiep = NULL;
+}
int xrep_reinit_pagf(struct xfs_scrub *sc);
int xrep_reinit_pagi(struct xfs_scrub *sc);
+int xrep_trans_alloc_hook_dummy(struct xfs_mount *mp, void **cookiep,
+ struct xfs_trans **tpp);
+void xrep_trans_cancel_hook_dummy(void **cookiep, struct xfs_trans *tp);
+
#else
#define xrep_ino_dqattach(sc) (0)
/* new rmapbt information */
struct xrep_newbt new_btree;
+ /* lock for the xfbtree and xfile */
+ struct mutex lock;
+
/* rmap records generated from primary metadata */
struct xfbtree rmap_btree;
/* in-memory btree cursor for the xfs_btree_bload iteration */
struct xfs_btree_cur *mcur;
+ /* Hooks into rmap update code. */
+ struct xfs_rmap_hook rhook;
+
/* inode scan cursor */
struct xchk_iscan iscan;
char *descr;
int error;
+ xchk_fsgates_enable(sc, XCHK_FSGATES_RMAP);
+
descr = xchk_xfile_ag_descr(sc, "reverse mapping records");
error = xrep_setup_xfbtree(sc, descr);
kfree(descr);
if (xchk_should_terminate(sc, &error))
return error;
+ if (xchk_iscan_aborted(&rr->iscan))
+ return -EFSCORRUPTED;
+
trace_xrep_rmap_found(sc->mp, sc->sa.pag->pag_agno, &rmap);
+ mutex_lock(&rr->lock);
mcur = xfs_rmapbt_mem_cursor(sc->sa.pag, sc->tp, &rr->rmap_btree);
error = xfs_rmap_map_raw(mcur, &rmap);
xfs_btree_del_cursor(mcur, error);
if (error)
goto out_cancel;
- return xfbtree_trans_commit(&rr->rmap_btree, sc->tp);
+ error = xfbtree_trans_commit(&rr->rmap_btree, sc->tp);
+ if (error)
+ goto out_abort;
+
+ mutex_unlock(&rr->lock);
+ return 0;
out_cancel:
xfbtree_trans_cancel(&rr->rmap_btree, sc->tp);
+out_abort:
+ xchk_iscan_abort(&rr->iscan);
+ mutex_unlock(&rr->lock);
return error;
}
if (error)
return error;
+ /*
+ * If a hook failed to update the in-memory btree, we lack the data to
+ * continue the repair.
+ */
+ if (xchk_iscan_aborted(&rr->iscan))
+ return -EFSCORRUPTED;
+
/*
* Now that we have everything locked again, we need to count the
* number of rmap records stashed in the btree. This should reflect
return error;
}
+static inline bool
+xrep_rmapbt_want_live_update(
+ struct xchk_iscan *iscan,
+ const struct xfs_owner_info *oi)
+{
+ if (xchk_iscan_aborted(iscan))
+ return false;
+
+ /*
+ * Before unlocking the AG header to perform the inode scan, we
+ * recorded reverse mappings for all AG metadata except for the OWN_AG
+ * metadata. IOWs, the in-memory btree knows about the AG headers, the
+ * two inode btrees, the CoW staging extents, and the refcount btrees.
+ * For these types of metadata, we need to record the live updates in
+ * the in-memory rmap btree.
+ *
+ * However, we do not scan the free space btrees or the AGFL until we
+ * have re-locked the AGF and are ready to reserve space for the new
+ * rmap btree, so we do not want live updates for OWN_AG metadata.
+ */
+ if (XFS_RMAP_NON_INODE_OWNER(oi->oi_owner))
+ return oi->oi_owner != XFS_RMAP_OWN_AG;
+
+ /* Ignore updates to files that the scanner hasn't visited yet. */
+ return xchk_iscan_want_live_update(iscan, oi->oi_owner);
+}
+
+/*
+ * Apply a rmapbt update from the regular filesystem into our shadow btree.
+ * We're running from the thread that owns the AGF buffer and is generating
+ * the update, so we must be careful about which parts of the struct xrep_rmap
+ * that we change.
+ */
+static int
+xrep_rmapbt_live_update(
+ struct notifier_block *nb,
+ unsigned long action,
+ void *data)
+{
+ struct xfs_rmap_update_params *p = data;
+ struct xrep_rmap *rr;
+ struct xfs_mount *mp;
+ struct xfs_btree_cur *mcur;
+ struct xfs_trans *tp;
+ void *txcookie;
+ int error;
+
+ rr = container_of(nb, struct xrep_rmap, rhook.rmap_hook.nb);
+ mp = rr->sc->mp;
+
+ if (!xrep_rmapbt_want_live_update(&rr->iscan, &p->oinfo))
+ goto out_unlock;
+
+ trace_xrep_rmap_live_update(mp, rr->sc->sa.pag->pag_agno, action, p);
+
+ error = xrep_trans_alloc_hook_dummy(mp, &txcookie, &tp);
+ if (error)
+ goto out_abort;
+
+ mutex_lock(&rr->lock);
+ mcur = xfs_rmapbt_mem_cursor(rr->sc->sa.pag, tp, &rr->rmap_btree);
+ error = __xfs_rmap_finish_intent(mcur, action, p->startblock,
+ p->blockcount, &p->oinfo, p->unwritten);
+ xfs_btree_del_cursor(mcur, error);
+ if (error)
+ goto out_cancel;
+
+ error = xfbtree_trans_commit(&rr->rmap_btree, tp);
+ if (error)
+ goto out_cancel;
+
+ xrep_trans_cancel_hook_dummy(&txcookie, tp);
+ mutex_unlock(&rr->lock);
+ return NOTIFY_DONE;
+
+out_cancel:
+ xfbtree_trans_cancel(&rr->rmap_btree, tp);
+ xrep_trans_cancel_hook_dummy(&txcookie, tp);
+out_abort:
+ mutex_unlock(&rr->lock);
+ xchk_iscan_abort(&rr->iscan);
+out_unlock:
+ return NOTIFY_DONE;
+}
+
/* Set up the filesystem scan components. */
STATIC int
xrep_rmap_setup_scan(
struct xfs_scrub *sc = rr->sc;
int error;
+ mutex_init(&rr->lock);
+
/* Set up in-memory rmap btree */
error = xfs_rmapbt_mem_init(sc->mp, &rr->rmap_btree, sc->xmbtp,
sc->sa.pag->pag_agno);
if (error)
- return error;
+ goto out_mutex;
/* Retry iget every tenth of a second for up to 30 seconds. */
xchk_iscan_start(sc, 30000, 100, &rr->iscan);
+
+ /*
+ * Hook into live rmap operations so that we can update our in-memory
+ * btree to reflect live changes on the filesystem. Since we drop the
+ * AGF buffer to scan all the inodes, we need this piece to avoid
+ * installing a stale btree.
+ */
+ ASSERT(sc->flags & XCHK_FSGATES_RMAP);
+ xfs_rmap_hook_setup(&rr->rhook, xrep_rmapbt_live_update);
+ error = xfs_rmap_hook_add(sc->sa.pag, &rr->rhook);
+ if (error)
+ goto out_iscan;
return 0;
+
+out_iscan:
+ xchk_iscan_teardown(&rr->iscan);
+ xfbtree_destroy(&rr->rmap_btree);
+out_mutex:
+ mutex_destroy(&rr->lock);
+ return error;
}
/* Tear down scan components. */
xrep_rmap_teardown(
struct xrep_rmap *rr)
{
+ struct xfs_scrub *sc = rr->sc;
+
+ xchk_iscan_abort(&rr->iscan);
+ xfs_rmap_hook_del(sc->sa.pag, &rr->rhook);
xchk_iscan_teardown(&rr->iscan);
xfbtree_destroy(&rr->rmap_btree);
+ mutex_destroy(&rr->lock);
}
/* Repair the rmap btree for some AG. */
struct xrep_rmap *rr = sc->buf;
int error;
- /* Functionality is not yet complete. */
- return xrep_notsupported(sc);
-
error = xrep_rmap_setup_scan(rr);
if (error)
return error;
#include "xfs_qm.h"
#include "xfs_scrub.h"
#include "xfs_buf_mem.h"
+#include "xfs_rmap.h"
#include "scrub/scrub.h"
#include "scrub/common.h"
#include "scrub/trace.h"
if (sc->flags & XCHK_FSGATES_DIRENTS)
xfs_dir_hook_disable();
+ if (sc->flags & XCHK_FSGATES_RMAP)
+ xfs_rmap_hook_disable();
+
sc->flags &= ~XCHK_FSGATES_ALL;
}
#define XCHK_NEED_DRAIN (1U << 3) /* scrub needs to drain defer ops */
#define XCHK_FSGATES_QUOTA (1U << 4) /* quota live update enabled */
#define XCHK_FSGATES_DIRENTS (1U << 5) /* directory live update enabled */
+#define XCHK_FSGATES_RMAP (1U << 6) /* rmapbt live update enabled */
#define XREP_RESET_PERAG_RESV (1U << 30) /* must reset AG space reservation */
#define XREP_ALREADY_FIXED (1U << 31) /* checking our repair work */
*/
#define XCHK_FSGATES_ALL (XCHK_FSGATES_DRAIN | \
XCHK_FSGATES_QUOTA | \
- XCHK_FSGATES_DIRENTS)
+ XCHK_FSGATES_DIRENTS | \
+ XCHK_FSGATES_RMAP)
/* Metadata scrubbers */
int xchk_tester(struct xfs_scrub *sc);
#include "xfs_quota_defs.h"
#include "xfs_da_format.h"
#include "xfs_dir2.h"
+#include "xfs_rmap.h"
#include "scrub/scrub.h"
#include "scrub/xfile.h"
#include "scrub/xfarray.h"
struct xchk_iscan;
struct xchk_nlink;
struct xchk_fscounters;
+struct xfs_rmap_update_params;
/*
* ftrace's __print_symbolic requires that all enum values be wrapped in the
{ XCHK_NEED_DRAIN, "need_drain" }, \
{ XCHK_FSGATES_QUOTA, "fsgates_quota" }, \
{ XCHK_FSGATES_DIRENTS, "fsgates_dirents" }, \
+ { XCHK_FSGATES_RMAP, "fsgates_rmap" }, \
{ XREP_RESET_PERAG_RESV, "reset_perag_resv" }, \
{ XREP_ALREADY_FIXED, "already_fixed" }
+TRACE_DEFINE_ENUM(XFS_RMAP_MAP);
+TRACE_DEFINE_ENUM(XFS_RMAP_MAP_SHARED);
+TRACE_DEFINE_ENUM(XFS_RMAP_UNMAP);
+TRACE_DEFINE_ENUM(XFS_RMAP_UNMAP_SHARED);
+TRACE_DEFINE_ENUM(XFS_RMAP_CONVERT);
+TRACE_DEFINE_ENUM(XFS_RMAP_CONVERT_SHARED);
+TRACE_DEFINE_ENUM(XFS_RMAP_ALLOC);
+TRACE_DEFINE_ENUM(XFS_RMAP_FREE);
+
DECLARE_EVENT_CLASS(xchk_class,
TP_PROTO(struct xfs_inode *ip, struct xfs_scrub_metadata *sm,
int error),
DEFINE_SCRUB_NLINKS_DIFF_EVENT(xrep_nlinks_update_inode);
DEFINE_SCRUB_NLINKS_DIFF_EVENT(xrep_nlinks_unfixable_inode);
+TRACE_EVENT(xrep_rmap_live_update,
+ TP_PROTO(struct xfs_mount *mp, xfs_agnumber_t agno, unsigned int op,
+ const struct xfs_rmap_update_params *p),
+ TP_ARGS(mp, agno, op, p),
+ TP_STRUCT__entry(
+ __field(dev_t, dev)
+ __field(xfs_agnumber_t, agno)
+ __field(unsigned int, op)
+ __field(xfs_agblock_t, agbno)
+ __field(xfs_extlen_t, len)
+ __field(uint64_t, owner)
+ __field(uint64_t, offset)
+ __field(unsigned int, flags)
+ ),
+ TP_fast_assign(
+ __entry->dev = mp->m_super->s_dev;
+ __entry->agno = agno;
+ __entry->op = op;
+ __entry->agbno = p->startblock;
+ __entry->len = p->blockcount;
+ xfs_owner_info_unpack(&p->oinfo, &__entry->owner,
+ &__entry->offset, &__entry->flags);
+ if (p->unwritten)
+ __entry->flags |= XFS_RMAP_UNWRITTEN;
+ ),
+ TP_printk("dev %d:%d agno 0x%x op %d agbno 0x%x fsbcount 0x%x owner 0x%llx fileoff 0x%llx flags 0x%x",
+ MAJOR(__entry->dev), MINOR(__entry->dev),
+ __entry->agno,
+ __entry->op,
+ __entry->agbno,
+ __entry->len,
+ __entry->owner,
+ __entry->offset,
+ __entry->flags)
+);
+
#endif /* IS_ENABLED(CONFIG_XFS_ONLINE_REPAIR) */
#endif /* _TRACE_XFS_SCRUB_TRACE_H */
projects / linux.git / commitdiff