bcachefs: Be more conservation about journal pre-reservations

- Try to always keep 1/8th of the journal free, on top of
   pre-reservations
 - Move the check for whether the journal is stuck to
   bch2_journal_space_available, and make it only fire when there aren't
   any journal writes in flight (that might free up space by updating
   last_seq)

Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

fs/bcachefs/journal.c

@@ -442,20 +442,6 @@ static int __journal_res_get(struct journal *j, struct journal_res *res,
 	if (!ret)
 		goto retry;
 
-	if (WARN_ONCE(ret == cur_entry_journal_full &&
-		      !can_discard &&
-		      (flags & JOURNAL_RES_GET_RESERVED),
-		      "JOURNAL_RES_GET_RESERVED set but journal full")) {
-		char *buf;
-
-		buf = kmalloc(4096, GFP_NOFS);
-		if (buf) {
-			bch2_journal_debug_to_text(&_PBUF(buf, 4096), j);
-			pr_err("\n%s", buf);
-			kfree(buf);
-		}
-	}
-
 	/*
 	 * Journal is full - can't rely on reclaim from work item due to
 	 * freezing:
@@ -1139,7 +1125,7 @@ int bch2_fs_journal_init(struct journal *j)
 
 /* debug: */
 
-void bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
+void __bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
 {
 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
 	union journal_res_state s;
@@ -1147,7 +1133,6 @@ void bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
 	unsigned i;
 
 	rcu_read_lock();
-	spin_lock(&j->lock);
 	s = READ_ONCE(j->reservations);
 
 	pr_buf(out,
@@ -1247,10 +1232,16 @@ void bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
 		       ja->cur_idx,		ja->bucket_seq[ja->cur_idx]);
 	}
 
-	spin_unlock(&j->lock);
 	rcu_read_unlock();
 }
 
+void bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
+{
+	spin_lock(&j->lock);
+	__bch2_journal_debug_to_text(out, j);
+	spin_unlock(&j->lock);
+}
+
 void bch2_journal_pins_to_text(struct printbuf *out, struct journal *j)
 {
 	struct journal_entry_pin_list *pin_list;

fs/bcachefs/journal.h

@@ -386,7 +386,7 @@ static inline int bch2_journal_res_get(struct journal *j, struct journal_res *re
 static inline bool journal_check_may_get_unreserved(struct journal *j)
 {
 	union journal_preres_state s = READ_ONCE(j->prereserved);
-	bool ret = s.reserved <= s.remaining &&
+	bool ret = s.reserved < s.remaining &&
 		fifo_free(&j->pin) > 8;
 
 	lockdep_assert_held(&j->lock);
@@ -510,6 +510,7 @@ static inline void bch2_journal_set_replay_done(struct journal *j)
 void bch2_journal_unblock(struct journal *);
 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 *);
 

fs/bcachefs/journal_io.c

@@ -1098,7 +1098,6 @@ static void journal_write_done(struct closure *cl)
 	if (!w->noflush) {
 		j->flushed_seq_ondisk = seq;
 		j->last_seq_ondisk = last_seq;
-		bch2_journal_space_available(j);
 	}
 
 	/*
@@ -1122,6 +1121,8 @@ static void journal_write_done(struct closure *cl)
 	} while ((v = atomic64_cmpxchg(&j->reservations.counter,
 				       old.v, new.v)) != old.v);
 
+	bch2_journal_space_available(j);
+
 	closure_wake_up(&w->wait);
 	journal_wake(j);
 

fs/bcachefs/journal_reclaim.c

@@ -2,6 +2,7 @@
 
 #include "bcachefs.h"
 #include "btree_key_cache.h"
+#include "error.h"
 #include "journal.h"
 #include "journal_io.h"
 #include "journal_reclaim.h"
@@ -159,7 +160,7 @@ void bch2_journal_space_available(struct journal *j)
 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
 	struct bch_dev *ca;
 	unsigned clean, clean_ondisk, total;
-	unsigned overhead, u64s_remaining = 0;
+	s64 u64s_remaining = 0;
 	unsigned max_entry_size	 = min(j->buf[0].buf_size >> 9,
 				       j->buf[1].buf_size >> 9);
 	unsigned i, nr_online = 0, nr_devs_want;
@@ -208,22 +209,37 @@ void bch2_journal_space_available(struct journal *j)
 	clean		= j->space[journal_space_clean].total;
 	total		= j->space[journal_space_total].total;
 
-	if (!j->space[journal_space_discarded].next_entry)
+	if (!clean_ondisk &&
+	    j->reservations.idx ==
+	    j->reservations.unwritten_idx) {
+		char *buf = kmalloc(4096, GFP_ATOMIC);
+
+		bch_err(c, "journal stuck");
+		if (buf) {
+			__bch2_journal_debug_to_text(&_PBUF(buf, 4096), j);
+			pr_err("\n%s", buf);
+			kfree(buf);
+		}
+
+		bch2_fatal_error(c);
+		ret = cur_entry_journal_stuck;
+	} else if (!j->space[journal_space_discarded].next_entry)
 		ret = cur_entry_journal_full;
 	else if (!fifo_free(&j->pin))
 		ret = cur_entry_journal_pin_full;
 
-	if ((clean - clean_ondisk <= total / 8) &&
+	if ((j->space[journal_space_clean_ondisk].next_entry <
+	     j->space[journal_space_clean_ondisk].total) &&
+	    (clean - clean_ondisk <= total / 8) &&
 	    (clean_ondisk * 2 > clean ))
 		set_bit(JOURNAL_MAY_SKIP_FLUSH, &j->flags);
 	else
 		clear_bit(JOURNAL_MAY_SKIP_FLUSH, &j->flags);
 
-	overhead = DIV_ROUND_UP(clean, max_entry_size) *
-		journal_entry_overhead(j);
-	u64s_remaining = clean << 6;
-	u64s_remaining = max_t(int, 0, u64s_remaining - overhead);
-	u64s_remaining /= 4;
+	u64s_remaining  = (u64) clean << 6;
+	u64s_remaining -= (u64) total << 3;
+	u64s_remaining = max(0LL, u64s_remaining);
+	u64s_remaining /= 2;
 out:
 	j->cur_entry_sectors	= !ret ? j->space[journal_space_discarded].next_entry : 0;
 	j->cur_entry_error	= ret;
@@ -572,6 +588,9 @@ static int __bch2_journal_reclaim(struct journal *j, bool direct)
 		    c->btree_cache.used  * 3)
 			min_nr = 1;
 
+		if (fifo_free(&j->pin) <= 32)
+			min_nr = 1;
+
 		min_nr = max(min_nr, bch2_nr_btree_keys_need_flush(c));
 
 		trace_journal_reclaim_start(c,

fs/bcachefs/journal_types.h

@@ -172,6 +172,7 @@ struct journal {
 		cur_entry_blocked,
 		cur_entry_journal_full,
 		cur_entry_journal_pin_full,
+		cur_entry_journal_stuck,
 		cur_entry_insufficient_devices,
 	}			cur_entry_error;