Update bcachefs sources to 10ab39f2fa bcachefs: Improvements to the journal read...

[bcachefs-tools-debian] / libbcachefs / io.c

diff --git a/libbcachefs/io.c b/libbcachefs/io.c
index 8c441050b02a6af8c4ed40bc4b29528ded51be9e..5c9c3cf54edd1c70c80c857626428c428ffbed22 100644 (file)
--- a/libbcachefs/io.c
+++ b/libbcachefs/io.c
@@ -31,9 +31,17 @@
 
 #include <linux/blkdev.h>
 #include <linux/random.h>
+#include <linux/sched/mm.h>
 
 #include <trace/events/bcachefs.h>
 
+const char *bch2_blk_status_to_str(blk_status_t status)
+{
+       if (status == BLK_STS_REMOVED)
+               return "device removed";
+       return blk_status_to_str(status);
+}
+
 static bool bch2_target_congested(struct bch_fs *c, u16 target)
 {
        const struct bch_devs_mask *devs;
@@ -46,7 +54,9 @@ static bool bch2_target_congested(struct bch_fs *c, u16 target)
                return false;
 
        rcu_read_lock();
-       devs = bch2_target_to_mask(c, target);
+       devs = bch2_target_to_mask(c, target) ?:
+               &c->rw_devs[BCH_DATA_user];
+
        for_each_set_bit(d, devs->d, BCH_SB_MEMBERS_MAX) {
                ca = rcu_dereference(c->devs[d]);
                if (!ca)
@@ -463,7 +473,8 @@ void bch2_submit_wbio_replicas(struct bch_write_bio *wbio, struct bch_fs *c,
 
                n->c                    = c;
                n->dev                  = ptr->dev;
-               n->have_ioref           = bch2_dev_get_ioref(ca, WRITE);
+               n->have_ioref           = bch2_dev_get_ioref(ca,
+                                       type == BCH_DATA_btree ? READ : WRITE);
                n->submit_time          = local_clock();
                n->bio.bi_iter.bi_sector = ptr->offset;
 
@@ -611,7 +622,8 @@ static void bch2_write_endio(struct bio *bio)
        struct bch_fs *c                = wbio->c;
        struct bch_dev *ca              = bch_dev_bkey_exists(c, wbio->dev);
 
-       if (bch2_dev_io_err_on(bio->bi_status, ca, "data write"))
+       if (bch2_dev_io_err_on(bio->bi_status, ca, "data write: %s",
+                              bch2_blk_status_to_str(bio->bi_status)))
                set_bit(wbio->dev, op->failed.d);
 
        if (wbio->have_ioref) {
@@ -1053,7 +1065,10 @@ static void __bch2_write(struct closure *cl)
        struct write_point *wp;
        struct bio *bio;
        bool skip_put = true;
+       unsigned nofs_flags;
        int ret;
+
+       nofs_flags = memalloc_nofs_save();
 again:
        memset(&op->failed, 0, sizeof(op->failed));
 
@@ -1079,6 +1094,11 @@ again:
                        goto err;
                }
 
+               /*
+                * The copygc thread is now global, which means it's no longer
+                * freeing up space on specific disks, which means that
+                * allocations for specific disks may hang arbitrarily long:
+                */
                wp = bch2_alloc_sectors_start(c,
                        op->target,
                        op->opts.erasure_code,
@@ -1088,7 +1108,8 @@ again:
                        op->nr_replicas_required,
                        op->alloc_reserve,
                        op->flags,
-                       (op->flags & BCH_WRITE_ALLOC_NOWAIT) ? NULL : cl);
+                       (op->flags & (BCH_WRITE_ALLOC_NOWAIT|
+                                     BCH_WRITE_ONLY_SPECIFIED_DEVS)) ? NULL : cl);
                EBUG_ON(!wp);
 
                if (unlikely(IS_ERR(wp))) {
@@ -1100,6 +1121,16 @@ again:
                        goto flush_io;
                }
 
+               /*
+                * It's possible for the allocator to fail, put us on the
+                * freelist waitlist, and then succeed in one of various retry
+                * paths: if that happens, we need to disable the skip_put
+                * optimization because otherwise there won't necessarily be a
+                * barrier before we free the bch_write_op:
+                */
+               if (atomic_read(&cl->remaining) & CLOSURE_WAITING)
+                       skip_put = false;
+
                bch2_open_bucket_get(c, wp, &op->open_buckets);
                ret = bch2_write_extent(op, wp, &bio);
                bch2_alloc_sectors_done(c, wp);
@@ -1129,19 +1160,21 @@ again:
                key_to_write = (void *) (op->insert_keys.keys_p +
                                         key_to_write_offset);
 
-               bch2_submit_wbio_replicas(to_wbio(bio), c, BCH_DATA_USER,
+               bch2_submit_wbio_replicas(to_wbio(bio), c, BCH_DATA_user,
                                          key_to_write);
        } while (ret);
 
        if (!skip_put)
                continue_at(cl, bch2_write_index, index_update_wq(op));
+out:
+       memalloc_nofs_restore(nofs_flags);
        return;
 err:
        op->error = ret;
        op->flags |= BCH_WRITE_DONE;
 
        continue_at(cl, bch2_write_index, index_update_wq(op));
-       return;
+       goto out;
 flush_io:
        /*
         * If the write can't all be submitted at once, we generally want to
@@ -1152,7 +1185,7 @@ flush_io:
         */
        if (current->flags & PF_WQ_WORKER) {
                continue_at(cl, bch2_write_index, index_update_wq(op));
-               return;
+               goto out;
        }
 
        closure_sync(cl);
@@ -1163,7 +1196,7 @@ flush_io:
                if (op->error) {
                        op->flags |= BCH_WRITE_DONE;
                        continue_at_nobarrier(cl, bch2_write_done, NULL);
-                       return;
+                       goto out;
                }
        }
 
@@ -1921,7 +1954,8 @@ static void bch2_read_endio(struct bio *bio)
        if (!rbio->split)
                rbio->bio.bi_end_io = rbio->end_io;
 
-       if (bch2_dev_io_err_on(bio->bi_status, ca, "data read")) {
+       if (bch2_dev_io_err_on(bio->bi_status, ca, "data read; %s",
+                              bch2_blk_status_to_str(bio->bi_status))) {
                bch2_rbio_error(rbio, READ_RETRY_AVOID, bio->bi_status);
                return;
        }
@@ -2174,7 +2208,7 @@ get_bio:
                        goto out;
                }
 
-               this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_USER],
+               this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_user],
                             bio_sectors(&rbio->bio));
                bio_set_dev(&rbio->bio, ca->disk_sb.bdev);