X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Freplicas.c;h=5b591c59bc3eadaf86ffd83360609950c9eaad5a;hb=21ae8a4b715acd326e6404ce6409ae329566eb64;hp=8e6cccd3938373bd2cb5d531931d71c04cff398a;hpb=f87850496ea0ad0e5e8f8feb1c454662d0973530;p=bcachefs-tools-debian diff --git a/libbcachefs/replicas.c b/libbcachefs/replicas.c index 8e6cccd..5b591c5 100644 --- a/libbcachefs/replicas.c +++ b/libbcachefs/replicas.c @@ -36,30 +36,47 @@ static void bch2_cpu_replicas_sort(struct bch_replicas_cpu *r) eytzinger0_sort(r->entries, r->nr, r->entry_size, memcmp, NULL); } +static void bch2_replicas_entry_v0_to_text(struct printbuf *out, + struct bch_replicas_entry_v0 *e) +{ + unsigned i; + + if (e->data_type < BCH_DATA_NR) + prt_printf(out, "%s", bch2_data_types[e->data_type]); + else + prt_printf(out, "(invalid data type %u)", e->data_type); + + prt_printf(out, ": %u [", e->nr_devs); + for (i = 0; i < e->nr_devs; i++) + prt_printf(out, i ? " %u" : "%u", e->devs[i]); + prt_printf(out, "]"); +} + void bch2_replicas_entry_to_text(struct printbuf *out, struct bch_replicas_entry *e) { unsigned i; - pr_buf(out, "%s: %u/%u [", - bch2_data_types[e->data_type], - e->nr_required, - e->nr_devs); + if (e->data_type < BCH_DATA_NR) + prt_printf(out, "%s", bch2_data_types[e->data_type]); + else + prt_printf(out, "(invalid data type %u)", e->data_type); + prt_printf(out, ": %u/%u [", e->nr_required, e->nr_devs); for (i = 0; i < e->nr_devs; i++) - pr_buf(out, i ? " %u" : "%u", e->devs[i]); - pr_buf(out, "]"); + prt_printf(out, i ? " %u" : "%u", e->devs[i]); + prt_printf(out, "]"); } void bch2_cpu_replicas_to_text(struct printbuf *out, - struct bch_replicas_cpu *r) + struct bch_replicas_cpu *r) { struct bch_replicas_entry *e; bool first = true; for_each_cpu_replicas_entry(r, e) { if (!first) - pr_buf(out, " "); + prt_printf(out, " "); first = false; bch2_replicas_entry_to_text(out, e); @@ -255,7 +272,7 @@ static void __replicas_table_update_pcpu(struct bch_fs_usage __percpu *dst_p, { unsigned src_nr = sizeof(struct bch_fs_usage) / sizeof(u64) + src_r->nr; struct bch_fs_usage *dst, *src = (void *) - bch2_acc_percpu_u64s((void *) src_p, src_nr); + bch2_acc_percpu_u64s((u64 __percpu *) src_p, src_nr); preempt_disable(); dst = this_cpu_ptr(dst_p); @@ -319,7 +336,7 @@ out: return ret; err: bch_err(c, "error updating replicas table: memory allocation failure"); - ret = -ENOMEM; + ret = -BCH_ERR_ENOMEM_replicas_table; goto out; } @@ -366,14 +383,18 @@ static int bch2_mark_replicas_slowpath(struct bch_fs *c, if (c->replicas_gc.entries && !__replicas_has_entry(&c->replicas_gc, new_entry)) { new_gc = cpu_replicas_add_entry(&c->replicas_gc, new_entry); - if (!new_gc.entries) + if (!new_gc.entries) { + ret = -BCH_ERR_ENOMEM_cpu_replicas; goto err; + } } if (!__replicas_has_entry(&c->replicas, new_entry)) { new_r = cpu_replicas_add_entry(&c->replicas, new_entry); - if (!new_r.entries) + if (!new_r.entries) { + ret = -BCH_ERR_ENOMEM_cpu_replicas; goto err; + } ret = bch2_cpu_replicas_to_sb_replicas(c, &new_r); if (ret) @@ -408,78 +429,18 @@ out: return ret; err: - bch_err(c, "error adding replicas entry: memory allocation failure"); - ret = -ENOMEM; + bch_err(c, "error adding replicas entry: %s", bch2_err_str(ret)); goto out; } -static int __bch2_mark_replicas(struct bch_fs *c, - struct bch_replicas_entry *r, - bool check) -{ - return likely(bch2_replicas_marked(c, r)) ? 0 - : check ? -1 - : bch2_mark_replicas_slowpath(c, r); -} - int bch2_mark_replicas(struct bch_fs *c, struct bch_replicas_entry *r) { - return __bch2_mark_replicas(c, r, false); -} - -static int __bch2_mark_bkey_replicas(struct bch_fs *c, struct bkey_s_c k, - bool check) -{ - struct bch_replicas_padded search; - struct bch_devs_list cached = bch2_bkey_cached_devs(k); - unsigned i; - int ret; - - for (i = 0; i < cached.nr; i++) { - bch2_replicas_entry_cached(&search.e, cached.devs[i]); - - ret = __bch2_mark_replicas(c, &search.e, check); - if (ret) - return ret; - } - - bch2_bkey_to_replicas(&search.e, k); - - ret = __bch2_mark_replicas(c, &search.e, check); - if (ret) - return ret; - - if (search.e.data_type == BCH_DATA_parity) { - search.e.data_type = BCH_DATA_cached; - ret = __bch2_mark_replicas(c, &search.e, check); - if (ret) - return ret; - - search.e.data_type = BCH_DATA_user; - ret = __bch2_mark_replicas(c, &search.e, check); - if (ret) - return ret; - } - - return 0; + return likely(bch2_replicas_marked(c, r)) + ? 0 : bch2_mark_replicas_slowpath(c, r); } /* replicas delta list: */ -bool bch2_replicas_delta_list_marked(struct bch_fs *c, - struct replicas_delta_list *r) -{ - struct replicas_delta *d = r->d; - struct replicas_delta *top = (void *) r->d + r->used; - - percpu_rwsem_assert_held(&c->mark_lock); - - for (d = r->d; d != top; d = replicas_delta_next(d)) - if (bch2_replicas_entry_idx(c, &d->r) < 0) - return false; - return true; -} - int bch2_replicas_delta_list_mark(struct bch_fs *c, struct replicas_delta_list *r) { @@ -492,19 +453,6 @@ int bch2_replicas_delta_list_mark(struct bch_fs *c, return ret; } -/* bkey replicas: */ - -bool bch2_bkey_replicas_marked(struct bch_fs *c, - struct bkey_s_c k) -{ - return __bch2_mark_bkey_replicas(c, k, true) == 0; -} - -int bch2_mark_bkey_replicas(struct bch_fs *c, struct bkey_s_c k) -{ - return __bch2_mark_bkey_replicas(c, k, false); -} - /* * Old replicas_gc mechanism: only used for journal replicas entries now, should * die at some point: @@ -512,40 +460,17 @@ int bch2_mark_bkey_replicas(struct bch_fs *c, struct bkey_s_c k) int bch2_replicas_gc_end(struct bch_fs *c, int ret) { - unsigned i; - lockdep_assert_held(&c->replicas_gc_lock); + if (ret) + goto err; + mutex_lock(&c->sb_lock); percpu_down_write(&c->mark_lock); - /* - * this is kind of crappy; the replicas gc mechanism needs to be ripped - * out - */ - - for (i = 0; i < c->replicas.nr; i++) { - struct bch_replicas_entry *e = - cpu_replicas_entry(&c->replicas, i); - struct bch_replicas_cpu n; - - if (!__replicas_has_entry(&c->replicas_gc, e) && - bch2_fs_usage_read_one(c, &c->usage_base->replicas[i])) { - n = cpu_replicas_add_entry(&c->replicas_gc, e); - if (!n.entries) { - ret = -ENOSPC; - goto err; - } - - swap(n, c->replicas_gc); - kfree(n.entries); - } - } - - if (bch2_cpu_replicas_to_sb_replicas(c, &c->replicas_gc)) { - ret = -ENOSPC; + ret = bch2_cpu_replicas_to_sb_replicas(c, &c->replicas_gc); + if (ret) goto err; - } ret = replicas_table_update(c, &c->replicas_gc); err: @@ -589,7 +514,7 @@ int bch2_replicas_gc_start(struct bch_fs *c, unsigned typemask) if (!c->replicas_gc.entries) { mutex_unlock(&c->sb_lock); bch_err(c, "error allocating c->replicas_gc"); - return -ENOMEM; + return -BCH_ERR_ENOMEM_replicas_gc; } for_each_cpu_replicas_entry(&c->replicas, e) @@ -603,8 +528,14 @@ int bch2_replicas_gc_start(struct bch_fs *c, unsigned typemask) return 0; } -/* New much simpler mechanism for clearing out unneeded replicas entries: */ - +/* + * New much simpler mechanism for clearing out unneeded replicas entries - drop + * replicas entries that have 0 sectors used. + * + * However, we don't track sector counts for journal usage, so this doesn't drop + * any BCH_DATA_journal entries; the old bch2_replicas_gc_(start|end) mechanism + * is retained for that. + */ int bch2_replicas_gc2(struct bch_fs *c) { struct bch_replicas_cpu new = { 0 }; @@ -618,7 +549,7 @@ retry: new.entries = kcalloc(nr, new.entry_size, GFP_KERNEL); if (!new.entries) { bch_err(c, "error allocating c->replicas_gc"); - return -ENOMEM; + return -BCH_ERR_ENOMEM_replicas_gc; } mutex_lock(&c->sb_lock); @@ -648,10 +579,9 @@ retry: bch2_cpu_replicas_sort(&new); - if (bch2_cpu_replicas_to_sb_replicas(c, &new)) { - ret = -ENOSPC; + ret = bch2_cpu_replicas_to_sb_replicas(c, &new); + if (ret) goto err; - } ret = replicas_table_update(c, &new); err: @@ -678,7 +608,7 @@ int bch2_replicas_set_usage(struct bch_fs *c, n = cpu_replicas_add_entry(&c->replicas, r); if (!n.entries) - return -ENOMEM; + return -BCH_ERR_ENOMEM_cpu_replicas; ret = replicas_table_update(c, &n); if (ret) @@ -712,7 +642,7 @@ __bch2_sb_replicas_to_cpu_replicas(struct bch_sb_field_replicas *sb_r, cpu_r->entries = kcalloc(nr, entry_size, GFP_KERNEL); if (!cpu_r->entries) - return -ENOMEM; + return -BCH_ERR_ENOMEM_cpu_replicas; cpu_r->nr = nr; cpu_r->entry_size = entry_size; @@ -744,7 +674,7 @@ __bch2_sb_replicas_v0_to_cpu_replicas(struct bch_sb_field_replicas_v0 *sb_r, cpu_r->entries = kcalloc(nr, entry_size, GFP_KERNEL); if (!cpu_r->entries) - return -ENOMEM; + return -BCH_ERR_ENOMEM_cpu_replicas; cpu_r->nr = nr; cpu_r->entry_size = entry_size; @@ -774,9 +704,8 @@ int bch2_sb_replicas_to_cpu_replicas(struct bch_fs *c) ret = __bch2_sb_replicas_to_cpu_replicas(sb_v1, &new_r); else if ((sb_v0 = bch2_sb_get_replicas_v0(c->disk_sb.sb))) ret = __bch2_sb_replicas_v0_to_cpu_replicas(sb_v0, &new_r); - if (ret) - return -ENOMEM; + return ret; bch2_cpu_replicas_sort(&new_r); @@ -806,7 +735,7 @@ static int bch2_cpu_replicas_to_sb_replicas_v0(struct bch_fs *c, sb_r = bch2_sb_resize_replicas_v0(&c->disk_sb, DIV_ROUND_UP(bytes, sizeof(u64))); if (!sb_r) - return -ENOSPC; + return -BCH_ERR_ENOSPC_sb_replicas; bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas); sb_r = bch2_sb_get_replicas_v0(c->disk_sb.sb); @@ -851,7 +780,7 @@ static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *c, sb_r = bch2_sb_resize_replicas(&c->disk_sb, DIV_ROUND_UP(bytes, sizeof(u64))); if (!sb_r) - return -ENOSPC; + return -BCH_ERR_ENOSPC_sb_replicas; bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas_v0); sb_r = bch2_sb_get_replicas(c->disk_sb.sb); @@ -872,67 +801,79 @@ static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *c, return 0; } -static const char *check_dup_replicas_entries(struct bch_replicas_cpu *cpu_r) +static int bch2_cpu_replicas_validate(struct bch_replicas_cpu *cpu_r, + struct bch_sb *sb, + struct printbuf *err) { - unsigned i; + struct bch_sb_field_members *mi = bch2_sb_get_members(sb); + unsigned i, j; sort_cmp_size(cpu_r->entries, cpu_r->nr, cpu_r->entry_size, memcmp, NULL); - for (i = 0; i + 1 < cpu_r->nr; i++) { - struct bch_replicas_entry *l = + for (i = 0; i < cpu_r->nr; i++) { + struct bch_replicas_entry *e = cpu_replicas_entry(cpu_r, i); - struct bch_replicas_entry *r = - cpu_replicas_entry(cpu_r, i + 1); - BUG_ON(memcmp(l, r, cpu_r->entry_size) > 0); + if (e->data_type >= BCH_DATA_NR) { + prt_printf(err, "invalid data type in entry "); + bch2_replicas_entry_to_text(err, e); + return -BCH_ERR_invalid_sb_replicas; + } - if (!memcmp(l, r, cpu_r->entry_size)) - return "duplicate replicas entry"; - } + if (!e->nr_devs) { + prt_printf(err, "no devices in entry "); + bch2_replicas_entry_to_text(err, e); + return -BCH_ERR_invalid_sb_replicas; + } - return NULL; -} + if (e->nr_required > 1 && + e->nr_required >= e->nr_devs) { + prt_printf(err, "bad nr_required in entry "); + bch2_replicas_entry_to_text(err, e); + return -BCH_ERR_invalid_sb_replicas; + } -static const char *bch2_sb_validate_replicas(struct bch_sb *sb, struct bch_sb_field *f) -{ - struct bch_sb_field_replicas *sb_r = field_to_type(f, replicas); - struct bch_sb_field_members *mi = bch2_sb_get_members(sb); - struct bch_replicas_cpu cpu_r = { .entries = NULL }; - struct bch_replicas_entry *e; - const char *err; - unsigned i; + for (j = 0; j < e->nr_devs; j++) + if (!bch2_dev_exists(sb, mi, e->devs[j])) { + prt_printf(err, "invalid device %u in entry ", e->devs[j]); + bch2_replicas_entry_to_text(err, e); + return -BCH_ERR_invalid_sb_replicas; + } - for_each_replicas_entry(sb_r, e) { - err = "invalid replicas entry: invalid data type"; - if (e->data_type >= BCH_DATA_NR) - goto err; + if (i + 1 < cpu_r->nr) { + struct bch_replicas_entry *n = + cpu_replicas_entry(cpu_r, i + 1); - err = "invalid replicas entry: no devices"; - if (!e->nr_devs) - goto err; + BUG_ON(memcmp(e, n, cpu_r->entry_size) > 0); - err = "invalid replicas entry: bad nr_required"; - if (e->nr_required > 1 && - e->nr_required >= e->nr_devs) - goto err; - - err = "invalid replicas entry: invalid device"; - for (i = 0; i < e->nr_devs; i++) - if (!bch2_dev_exists(sb, mi, e->devs[i])) - goto err; + if (!memcmp(e, n, cpu_r->entry_size)) { + prt_printf(err, "duplicate replicas entry "); + bch2_replicas_entry_to_text(err, e); + return -BCH_ERR_invalid_sb_replicas; + } + } } - err = "cannot allocate memory"; - if (__bch2_sb_replicas_to_cpu_replicas(sb_r, &cpu_r)) - goto err; + return 0; +} - err = check_dup_replicas_entries(&cpu_r); -err: +static int bch2_sb_replicas_validate(struct bch_sb *sb, struct bch_sb_field *f, + struct printbuf *err) +{ + struct bch_sb_field_replicas *sb_r = field_to_type(f, replicas); + struct bch_replicas_cpu cpu_r; + int ret; + + ret = __bch2_sb_replicas_to_cpu_replicas(sb_r, &cpu_r); + if (ret) + return ret; + + ret = bch2_cpu_replicas_validate(&cpu_r, sb, err); kfree(cpu_r.entries); - return err; + return ret; } static void bch2_sb_replicas_to_text(struct printbuf *out, @@ -945,54 +886,56 @@ static void bch2_sb_replicas_to_text(struct printbuf *out, for_each_replicas_entry(r, e) { if (!first) - pr_buf(out, " "); + prt_printf(out, " "); first = false; bch2_replicas_entry_to_text(out, e); } + prt_newline(out); } const struct bch_sb_field_ops bch_sb_field_ops_replicas = { - .validate = bch2_sb_validate_replicas, + .validate = bch2_sb_replicas_validate, .to_text = bch2_sb_replicas_to_text, }; -static const char *bch2_sb_validate_replicas_v0(struct bch_sb *sb, struct bch_sb_field *f) +static int bch2_sb_replicas_v0_validate(struct bch_sb *sb, struct bch_sb_field *f, + struct printbuf *err) { struct bch_sb_field_replicas_v0 *sb_r = field_to_type(f, replicas_v0); - struct bch_sb_field_members *mi = bch2_sb_get_members(sb); - struct bch_replicas_cpu cpu_r = { .entries = NULL }; - struct bch_replicas_entry_v0 *e; - const char *err; - unsigned i; + struct bch_replicas_cpu cpu_r; + int ret; - for_each_replicas_entry_v0(sb_r, e) { - err = "invalid replicas entry: invalid data type"; - if (e->data_type >= BCH_DATA_NR) - goto err; + ret = __bch2_sb_replicas_v0_to_cpu_replicas(sb_r, &cpu_r); + if (ret) + return ret; - err = "invalid replicas entry: no devices"; - if (!e->nr_devs) - goto err; + ret = bch2_cpu_replicas_validate(&cpu_r, sb, err); + kfree(cpu_r.entries); + return ret; +} - err = "invalid replicas entry: invalid device"; - for (i = 0; i < e->nr_devs; i++) - if (!bch2_dev_exists(sb, mi, e->devs[i])) - goto err; - } +static void bch2_sb_replicas_v0_to_text(struct printbuf *out, + struct bch_sb *sb, + struct bch_sb_field *f) +{ + struct bch_sb_field_replicas_v0 *sb_r = field_to_type(f, replicas_v0); + struct bch_replicas_entry_v0 *e; + bool first = true; - err = "cannot allocate memory"; - if (__bch2_sb_replicas_v0_to_cpu_replicas(sb_r, &cpu_r)) - goto err; + for_each_replicas_entry(sb_r, e) { + if (!first) + prt_printf(out, " "); + first = false; - err = check_dup_replicas_entries(&cpu_r); -err: - kfree(cpu_r.entries); - return err; + bch2_replicas_entry_v0_to_text(out, e); + } + prt_newline(out); } const struct bch_sb_field_ops bch_sb_field_ops_replicas_v0 = { - .validate = bch2_sb_validate_replicas_v0, + .validate = bch2_sb_replicas_v0_validate, + .to_text = bch2_sb_replicas_v0_to_text, }; /* Query replicas: */ @@ -1008,6 +951,9 @@ bool bch2_have_enough_devs(struct bch_fs *c, struct bch_devs_mask devs, unsigned i, nr_online = 0, nr_failed = 0, dflags = 0; bool metadata = e->data_type < BCH_DATA_user; + if (e->data_type == BCH_DATA_cached) + continue; + for (i = 0; i < e->nr_devs; i++) { struct bch_dev *ca = bch_dev_bkey_exists(c, e->devs[i]); @@ -1030,11 +976,12 @@ bool bch2_have_enough_devs(struct bch_fs *c, struct bch_devs_mask devs, if (dflags & ~flags) { if (print) { - char buf[100]; + struct printbuf buf = PRINTBUF; - bch2_replicas_entry_to_text(&PBUF(buf), e); + bch2_replicas_entry_to_text(&buf, e); bch_err(c, "insufficient devices online (%u) for replicas entry %s", - nr_online, buf); + nr_online, buf.buf); + printbuf_exit(&buf); } ret = false; break; @@ -1046,19 +993,42 @@ bool bch2_have_enough_devs(struct bch_fs *c, struct bch_devs_mask devs, return ret; } -unsigned bch2_dev_has_data(struct bch_fs *c, struct bch_dev *ca) +unsigned bch2_sb_dev_has_data(struct bch_sb *sb, unsigned dev) { - struct bch_replicas_entry *e; - unsigned i, ret = 0; + struct bch_sb_field_replicas *replicas; + struct bch_sb_field_replicas_v0 *replicas_v0; + unsigned i, data_has = 0; + + replicas = bch2_sb_get_replicas(sb); + replicas_v0 = bch2_sb_get_replicas_v0(sb); + + if (replicas) { + struct bch_replicas_entry *r; + + for_each_replicas_entry(replicas, r) + for (i = 0; i < r->nr_devs; i++) + if (r->devs[i] == dev) + data_has |= 1 << r->data_type; + } else if (replicas_v0) { + struct bch_replicas_entry_v0 *r; + + for_each_replicas_entry_v0(replicas_v0, r) + for (i = 0; i < r->nr_devs; i++) + if (r->devs[i] == dev) + data_has |= 1 << r->data_type; + } - percpu_down_read(&c->mark_lock); - for_each_cpu_replicas_entry(&c->replicas, e) - for (i = 0; i < e->nr_devs; i++) - if (e->devs[i] == ca->dev_idx) - ret |= 1 << e->data_type; + return data_has; +} - percpu_up_read(&c->mark_lock); +unsigned bch2_dev_has_data(struct bch_fs *c, struct bch_dev *ca) +{ + unsigned ret; + + mutex_lock(&c->sb_lock); + ret = bch2_sb_dev_has_data(c->disk_sb.sb, ca->dev_idx); + mutex_unlock(&c->sb_lock); return ret; }