]> git.sesse.net Git - bcachefs-tools-debian/blobdiff - libbcachefs/journal.h
Update bcachefs sources to 4b5917839c bcachefs: Fix a null ptr deref in check_xattr()
[bcachefs-tools-debian] / libbcachefs / journal.h
index c39cbbf1bccd068c65f5206cf429d70981e1995b..008a2e25a4fac93df681233b679c9fc6eaca95d0 100644 (file)
@@ -29,8 +29,8 @@
  *
  * Synchronous updates are specified by passing a closure (@flush_cl) to
  * bch2_btree_insert() or bch_btree_insert_node(), which then pass that parameter
- * down to the journalling code. That closure will will wait on the journal
- * write to complete (via closure_wait()).
+ * down to the journalling code. That closure will wait on the journal write to
+ * complete (via closure_wait()).
  *
  * If the index update wasn't synchronous, the journal entry will be
  * written out after 10 ms have elapsed, by default (the delay_ms field
@@ -141,7 +141,10 @@ static inline u64 journal_cur_seq(struct journal *j)
        return j->pin.back - 1;
 }
 
-void bch2_journal_set_has_inum(struct journal *, u64, u64);
+static inline u64 journal_last_unwritten_seq(struct journal *j)
+{
+       return j->seq_ondisk + 1;
+}
 
 static inline int journal_state_count(union journal_res_state s, int idx)
 {
@@ -196,9 +199,9 @@ journal_res_entry(struct journal *j, struct journal_res *res)
        return vstruct_idx(j->buf[res->idx].data, res->offset);
 }
 
-static inline unsigned journal_entry_set(struct jset_entry *entry, unsigned type,
+static inline unsigned journal_entry_init(struct jset_entry *entry, unsigned type,
                                          enum btree_id id, unsigned level,
-                                         const void *data, unsigned u64s)
+                                         unsigned u64s)
 {
        entry->u64s     = cpu_to_le16(u64s);
        entry->btree_id = id;
@@ -207,32 +210,33 @@ static inline unsigned journal_entry_set(struct jset_entry *entry, unsigned type
        entry->pad[0]   = 0;
        entry->pad[1]   = 0;
        entry->pad[2]   = 0;
-       memcpy_u64s_small(entry->_data, data, u64s);
-
        return jset_u64s(u64s);
 }
 
-static inline void bch2_journal_add_entry(struct journal *j, struct journal_res *res,
-                                         unsigned type, enum btree_id id,
-                                         unsigned level,
+static inline unsigned journal_entry_set(struct jset_entry *entry, unsigned type,
+                                         enum btree_id id, unsigned level,
                                          const void *data, unsigned u64s)
 {
-       unsigned actual = journal_entry_set(journal_res_entry(j, res),
-                              type, id, level, data, u64s);
+       unsigned ret = journal_entry_init(entry, type, id, level, u64s);
+
+       memcpy_u64s_small(entry->_data, data, u64s);
+       return ret;
+}
+
+static inline struct jset_entry *
+bch2_journal_add_entry(struct journal *j, struct journal_res *res,
+                        unsigned type, enum btree_id id,
+                        unsigned level, unsigned u64s)
+{
+       struct jset_entry *entry = journal_res_entry(j, res);
+       unsigned actual = journal_entry_init(entry, type, id, level, u64s);
 
        EBUG_ON(!res->ref);
        EBUG_ON(actual > res->u64s);
 
        res->offset     += actual;
        res->u64s       -= actual;
-}
-
-static inline void bch2_journal_add_keys(struct journal *j, struct journal_res *res,
-                                       enum btree_id id, unsigned level,
-                                       const struct bkey_i *k)
-{
-       bch2_journal_add_entry(j, res, BCH_JSET_ENTRY_btree_keys,
-                              id, level, k, k->k.u64s);
+       return entry;
 }
 
 static inline bool journal_entry_empty(struct jset *j)
@@ -261,9 +265,6 @@ static inline void bch2_journal_buf_put(struct journal *j, unsigned idx)
                                    .buf3_count = idx == 3,
                                    }).v, &j->reservations.counter);
 
-       EBUG_ON(((s.idx - idx) & 3) >
-               ((s.idx - s.unwritten_idx) & 3));
-
        if (!journal_state_count(s, idx) && idx == s.unwritten_idx)
                __bch2_journal_buf_put(j);
 }
@@ -283,7 +284,7 @@ static inline void bch2_journal_res_put(struct journal *j,
        while (res->u64s)
                bch2_journal_add_entry(j, res,
                                       BCH_JSET_ENTRY_btree_keys,
-                                      0, 0, NULL, 0);
+                                      0, 0, 0);
 
        bch2_journal_buf_put(j, res->idx);
 
@@ -293,9 +294,14 @@ static inline void bch2_journal_res_put(struct journal *j,
 int bch2_journal_res_get_slowpath(struct journal *, struct journal_res *,
                                  unsigned);
 
-#define JOURNAL_RES_GET_NONBLOCK       (1 << 0)
-#define JOURNAL_RES_GET_CHECK          (1 << 1)
-#define JOURNAL_RES_GET_RESERVED       (1 << 2)
+/* First bits for BCH_WATERMARK: */
+enum journal_res_flags {
+       __JOURNAL_RES_GET_NONBLOCK      = BCH_WATERMARK_BITS,
+       __JOURNAL_RES_GET_CHECK,
+};
+
+#define JOURNAL_RES_GET_NONBLOCK       (1 << __JOURNAL_RES_GET_NONBLOCK)
+#define JOURNAL_RES_GET_CHECK          (1 << __JOURNAL_RES_GET_CHECK)
 
 static inline int journal_res_get_fast(struct journal *j,
                                       struct journal_res *res,
@@ -316,8 +322,7 @@ static inline int journal_res_get_fast(struct journal *j,
 
                EBUG_ON(!journal_state_count(new, new.idx));
 
-               if (!(flags & JOURNAL_RES_GET_RESERVED) &&
-                   !test_bit(JOURNAL_MAY_GET_UNRESERVED, &j->flags))
+               if ((flags & BCH_WATERMARK_MASK) < j->watermark)
                        return 0;
 
                new.cur_entry_offset += res->u64s;
@@ -370,23 +375,27 @@ out:
 
 /* journal_preres: */
 
-static inline bool journal_check_may_get_unreserved(struct journal *j)
+static inline void journal_set_watermark(struct journal *j)
 {
        union journal_preres_state s = READ_ONCE(j->prereserved);
-       bool ret = s.reserved < s.remaining &&
-               fifo_free(&j->pin) > 8;
-
-       lockdep_assert_held(&j->lock);
-
-       if (ret != test_bit(JOURNAL_MAY_GET_UNRESERVED, &j->flags)) {
-               if (ret) {
-                       set_bit(JOURNAL_MAY_GET_UNRESERVED, &j->flags);
-                       journal_wake(j);
-               } else {
-                       clear_bit(JOURNAL_MAY_GET_UNRESERVED, &j->flags);
-               }
-       }
-       return ret;
+       unsigned watermark = BCH_WATERMARK_stripe;
+
+       if (fifo_free(&j->pin) < j->pin.size / 4)
+               watermark = max_t(unsigned, watermark, BCH_WATERMARK_copygc);
+       if (fifo_free(&j->pin) < j->pin.size / 8)
+               watermark = max_t(unsigned, watermark, BCH_WATERMARK_reclaim);
+
+       if (s.reserved > s.remaining)
+               watermark = max_t(unsigned, watermark, BCH_WATERMARK_copygc);
+       if (!s.remaining)
+               watermark = max_t(unsigned, watermark, BCH_WATERMARK_reclaim);
+
+       if (watermark == j->watermark)
+               return;
+
+       swap(watermark, j->watermark);
+       if (watermark > j->watermark)
+               journal_wake(j);
 }
 
 static inline void bch2_journal_preres_put(struct journal *j,
@@ -406,12 +415,8 @@ static inline void bch2_journal_preres_put(struct journal *j,
                closure_wake_up(&j->preres_wait);
        }
 
-       if (s.reserved <= s.remaining &&
-           !test_bit(JOURNAL_MAY_GET_UNRESERVED, &j->flags)) {
-               spin_lock(&j->lock);
-               journal_check_may_get_unreserved(j);
-               spin_unlock(&j->lock);
-       }
+       if (s.reserved <= s.remaining && j->watermark)
+               journal_set_watermark(j);
 }
 
 int __bch2_journal_preres_get(struct journal *,
@@ -426,14 +431,14 @@ static inline int bch2_journal_preres_get_fast(struct journal *j,
        int d = new_u64s - res->u64s;
        union journal_preres_state old, new;
        u64 v = atomic64_read(&j->prereserved.counter);
+       enum bch_watermark watermark = flags & BCH_WATERMARK_MASK;
        int ret;
 
        do {
                old.v = new.v = v;
                ret = 0;
 
-               if ((flags & JOURNAL_RES_GET_RESERVED) ||
-                   test_bit(JOURNAL_NOCHANGES, &j->flags) ||
+               if (watermark == BCH_WATERMARK_reclaim ||
                    new.reserved + d < new.remaining) {
                        new.reserved += d;
                        ret = 1;
@@ -461,7 +466,7 @@ static inline int bch2_journal_preres_get(struct journal *j,
                return 0;
 
        if (flags & JOURNAL_RES_GET_NONBLOCK)
-               return -EAGAIN;
+               return -BCH_ERR_journal_preres_get_blocked;
 
        return __bch2_journal_preres_get(j, res, new_u64s, flags);
 }
@@ -477,6 +482,7 @@ void bch2_journal_flush_async(struct journal *, struct closure *);
 
 int bch2_journal_flush_seq(struct journal *, u64);
 int bch2_journal_flush(struct journal *);
+bool bch2_journal_noflush_seq(struct journal *, u64);
 int bch2_journal_meta(struct journal *);
 
 void bch2_journal_halt(struct journal *);
@@ -501,6 +507,7 @@ void bch2_journal_block(struct journal *);
 void __bch2_journal_debug_to_text(struct printbuf *, struct journal *);
 void bch2_journal_debug_to_text(struct printbuf *, struct journal *);
 void bch2_journal_pins_to_text(struct printbuf *, struct journal *);
+bool bch2_journal_seq_pins_to_text(struct printbuf *, struct journal *, u64 *);
 
 int bch2_set_nr_journal_buckets(struct bch_fs *, struct bch_dev *,
                                unsigned nr);
@@ -509,7 +516,7 @@ int bch2_dev_journal_alloc(struct bch_dev *);
 void bch2_dev_journal_stop(struct journal *, struct bch_dev *);
 
 void bch2_fs_journal_stop(struct journal *);
-int bch2_fs_journal_start(struct journal *, u64, struct list_head *);
+int bch2_fs_journal_start(struct journal *, u64);
 
 void bch2_dev_journal_exit(struct bch_dev *);
 int bch2_dev_journal_init(struct bch_dev *, struct bch_sb *);