* Code for manipulating bucket marks for garbage collection.
*
* Copyright 2014 Datera, Inc.
- *
- * Bucket states:
- * - free bucket: mark == 0
- * The bucket contains no data and will not be read
- *
- * - allocator bucket: owned_by_allocator == 1
- * The bucket is on a free list, or it is an open bucket
- *
- * - cached bucket: owned_by_allocator == 0 &&
- * dirty_sectors == 0 &&
- * cached_sectors > 0
- * The bucket contains data but may be safely discarded as there are
- * enough replicas of the data on other cache devices, or it has been
- * written back to the backing device
- *
- * - dirty bucket: owned_by_allocator == 0 &&
- * dirty_sectors > 0
- * The bucket contains data that we must not discard (either only copy,
- * or one of the 'main copies' for data requiring multiple replicas)
- *
- * - metadata bucket: owned_by_allocator == 0 && is_metadata == 1
- * This is a btree node, journal or gen/prio bucket
- *
- * Lifecycle:
- *
- * bucket invalidated => bucket on freelist => open bucket =>
- * [dirty bucket =>] cached bucket => bucket invalidated => ...
- *
- * Note that cache promotion can skip the dirty bucket step, as data
- * is copied from a deeper tier to a shallower tier, onto a cached
- * bucket.
- * Note also that a cached bucket can spontaneously become dirty --
- * see below.
- *
- * Only a traversal of the key space can determine whether a bucket is
- * truly dirty or cached.
- *
- * Transitions:
- *
- * - free => allocator: bucket was invalidated
- * - cached => allocator: bucket was invalidated
- *
- * - allocator => dirty: open bucket was filled up
- * - allocator => cached: open bucket was filled up
- * - allocator => metadata: metadata was allocated
- *
- * - dirty => cached: dirty sectors were copied to a deeper tier
- * - dirty => free: dirty sectors were overwritten or moved (copy gc)
- * - cached => free: cached sectors were overwritten
- *
- * - metadata => free: metadata was freed
- *
- * Oddities:
- * - cached => dirty: a device was removed so formerly replicated data
- * is no longer sufficiently replicated
- * - free => cached: cannot happen
- * - free => dirty: cannot happen
- * - free => metadata: cannot happen
*/
#include "bcachefs.h"
#include "buckets.h"
#include "ec.h"
#include "error.h"
+#include "inode.h"
#include "movinggc.h"
+#include "recovery.h"
+#include "reflink.h"
#include "replicas.h"
+#include "subvolume.h"
#include <linux/preempt.h>
#include <trace/events/bcachefs.h>
+static inline void fs_usage_data_type_to_base(struct bch_fs_usage *fs_usage,
+ enum bch_data_type data_type,
+ s64 sectors)
+{
+ switch (data_type) {
+ case BCH_DATA_btree:
+ fs_usage->btree += sectors;
+ break;
+ case BCH_DATA_user:
+ case BCH_DATA_parity:
+ fs_usage->data += sectors;
+ break;
+ case BCH_DATA_cached:
+ fs_usage->cached += sectors;
+ break;
+ default:
+ break;
+ }
+}
+
/*
* Clear journal_seq_valid for buckets for which it's not needed, to prevent
* wraparound:
void bch2_bucket_seq_cleanup(struct bch_fs *c)
{
u64 journal_seq = atomic64_read(&c->journal.seq);
- u16 last_seq_ondisk = c->journal.last_seq_ondisk;
+ u16 last_seq_ondisk = c->journal.flushed_seq_ondisk;
struct bch_dev *ca;
struct bucket_array *buckets;
struct bucket *g;
void bch2_fs_usage_initialize(struct bch_fs *c)
{
struct bch_fs_usage *usage;
+ struct bch_dev *ca;
unsigned i;
percpu_down_write(&c->mark_lock);
usage = c->usage_base;
- bch2_fs_usage_acc_to_base(c, 0);
- bch2_fs_usage_acc_to_base(c, 1);
+ for (i = 0; i < ARRAY_SIZE(c->usage); i++)
+ bch2_fs_usage_acc_to_base(c, i);
for (i = 0; i < BCH_REPLICAS_MAX; i++)
usage->reserved += usage->persistent_reserved[i];
struct bch_replicas_entry *e =
cpu_replicas_entry(&c->replicas, i);
- switch (e->data_type) {
- case BCH_DATA_BTREE:
- usage->btree += usage->replicas[i];
- break;
- case BCH_DATA_USER:
- usage->data += usage->replicas[i];
- break;
- case BCH_DATA_CACHED:
- usage->cached += usage->replicas[i];
- break;
- }
+ fs_usage_data_type_to_base(usage, e->data_type, usage->replicas[i]);
}
- percpu_up_write(&c->mark_lock);
-}
+ for_each_member_device(ca, c, i) {
+ struct bch_dev_usage dev = bch2_dev_usage_read(ca);
-void bch2_fs_usage_scratch_put(struct bch_fs *c, struct bch_fs_usage *fs_usage)
-{
- if (fs_usage == c->usage_scratch)
- mutex_unlock(&c->usage_scratch_lock);
- else
- kfree(fs_usage);
+ usage->hidden += (dev.d[BCH_DATA_sb].buckets +
+ dev.d[BCH_DATA_journal].buckets) *
+ ca->mi.bucket_size;
+ }
+
+ percpu_up_write(&c->mark_lock);
}
-struct bch_fs_usage *bch2_fs_usage_scratch_get(struct bch_fs *c)
+static inline struct bch_dev_usage *dev_usage_ptr(struct bch_dev *ca,
+ unsigned journal_seq,
+ bool gc)
{
- struct bch_fs_usage *ret;
- unsigned bytes = fs_usage_u64s(c) * sizeof(u64);
-
- ret = kzalloc(bytes, GFP_NOWAIT);
- if (ret)
- return ret;
-
- if (mutex_trylock(&c->usage_scratch_lock))
- goto out_pool;
-
- ret = kzalloc(bytes, GFP_NOFS);
- if (ret)
- return ret;
+ BUG_ON(!gc && !journal_seq);
- mutex_lock(&c->usage_scratch_lock);
-out_pool:
- ret = c->usage_scratch;
- memset(ret, 0, bytes);
- return ret;
+ return this_cpu_ptr(gc
+ ? ca->usage_gc
+ : ca->usage[journal_seq & JOURNAL_BUF_MASK]);
}
-struct bch_dev_usage bch2_dev_usage_read(struct bch_fs *c, struct bch_dev *ca)
+struct bch_dev_usage bch2_dev_usage_read(struct bch_dev *ca)
{
+ struct bch_fs *c = ca->fs;
struct bch_dev_usage ret;
+ unsigned seq, i, u64s = dev_usage_u64s();
- memset(&ret, 0, sizeof(ret));
- acc_u64s_percpu((u64 *) &ret,
- (u64 __percpu *) ca->usage[0],
- sizeof(ret) / sizeof(u64));
+ do {
+ seq = read_seqcount_begin(&c->usage_lock);
+ memcpy(&ret, ca->usage_base, u64s * sizeof(u64));
+ for (i = 0; i < ARRAY_SIZE(ca->usage); i++)
+ acc_u64s_percpu((u64 *) &ret, (u64 __percpu *) ca->usage[i], u64s);
+ } while (read_seqcount_retry(&c->usage_lock, seq));
return ret;
}
unsigned journal_seq,
bool gc)
{
+ percpu_rwsem_assert_held(&c->mark_lock);
+ BUG_ON(!gc && !journal_seq);
+
return this_cpu_ptr(gc
? c->usage_gc
- : c->usage[journal_seq & 1]);
+ : c->usage[journal_seq & JOURNAL_BUF_MASK]);
}
u64 bch2_fs_usage_read_one(struct bch_fs *c, u64 *v)
{
ssize_t offset = v - (u64 *) c->usage_base;
- unsigned seq;
+ unsigned i, seq;
u64 ret;
BUG_ON(offset < 0 || offset >= fs_usage_u64s(c));
do {
seq = read_seqcount_begin(&c->usage_lock);
- ret = *v +
- percpu_u64_get((u64 __percpu *) c->usage[0] + offset) +
- percpu_u64_get((u64 __percpu *) c->usage[1] + offset);
+ ret = *v;
+
+ for (i = 0; i < ARRAY_SIZE(c->usage); i++)
+ ret += percpu_u64_get((u64 __percpu *) c->usage[i] + offset);
} while (read_seqcount_retry(&c->usage_lock, seq));
return ret;
}
-struct bch_fs_usage *bch2_fs_usage_read(struct bch_fs *c)
+struct bch_fs_usage_online *bch2_fs_usage_read(struct bch_fs *c)
{
- struct bch_fs_usage *ret;
- unsigned seq, v, u64s = fs_usage_u64s(c);
-retry:
- ret = kmalloc(u64s * sizeof(u64), GFP_NOFS);
- if (unlikely(!ret))
- return NULL;
+ struct bch_fs_usage_online *ret;
+ unsigned seq, i, u64s;
percpu_down_read(&c->mark_lock);
- v = fs_usage_u64s(c);
- if (unlikely(u64s != v)) {
- u64s = v;
+ ret = kmalloc(sizeof(struct bch_fs_usage_online) +
+ sizeof(u64) * c->replicas.nr, GFP_NOFS);
+ if (unlikely(!ret)) {
percpu_up_read(&c->mark_lock);
- kfree(ret);
- goto retry;
+ return NULL;
}
+ ret->online_reserved = percpu_u64_get(c->online_reserved);
+
+ u64s = fs_usage_u64s(c);
do {
seq = read_seqcount_begin(&c->usage_lock);
- memcpy(ret, c->usage_base, u64s * sizeof(u64));
- acc_u64s_percpu((u64 *) ret, (u64 __percpu *) c->usage[0], u64s);
- acc_u64s_percpu((u64 *) ret, (u64 __percpu *) c->usage[1], u64s);
+ memcpy(&ret->u, c->usage_base, u64s * sizeof(u64));
+ for (i = 0; i < ARRAY_SIZE(c->usage); i++)
+ acc_u64s_percpu((u64 *) &ret->u, (u64 __percpu *) c->usage[i], u64s);
} while (read_seqcount_retry(&c->usage_lock, seq));
return ret;
void bch2_fs_usage_acc_to_base(struct bch_fs *c, unsigned idx)
{
- unsigned u64s = fs_usage_u64s(c);
+ struct bch_dev *ca;
+ unsigned i, u64s = fs_usage_u64s(c);
- BUG_ON(idx >= 2);
+ BUG_ON(idx >= ARRAY_SIZE(c->usage));
+ preempt_disable();
write_seqcount_begin(&c->usage_lock);
acc_u64s_percpu((u64 *) c->usage_base,
(u64 __percpu *) c->usage[idx], u64s);
percpu_memset(c->usage[idx], 0, u64s * sizeof(u64));
+ rcu_read_lock();
+ for_each_member_device_rcu(ca, c, i, NULL) {
+ u64s = dev_usage_u64s();
+
+ acc_u64s_percpu((u64 *) ca->usage_base,
+ (u64 __percpu *) ca->usage[idx], u64s);
+ percpu_memset(ca->usage[idx], 0, u64s * sizeof(u64));
+ }
+ rcu_read_unlock();
+
write_seqcount_end(&c->usage_lock);
+ preempt_enable();
}
void bch2_fs_usage_to_text(struct printbuf *out,
struct bch_fs *c,
- struct bch_fs_usage *fs_usage)
+ struct bch_fs_usage_online *fs_usage)
{
unsigned i;
pr_buf(out, "capacity:\t\t\t%llu\n", c->capacity);
pr_buf(out, "hidden:\t\t\t\t%llu\n",
- fs_usage->hidden);
+ fs_usage->u.hidden);
pr_buf(out, "data:\t\t\t\t%llu\n",
- fs_usage->data);
+ fs_usage->u.data);
pr_buf(out, "cached:\t\t\t\t%llu\n",
- fs_usage->cached);
+ fs_usage->u.cached);
pr_buf(out, "reserved:\t\t\t%llu\n",
- fs_usage->reserved);
+ fs_usage->u.reserved);
pr_buf(out, "nr_inodes:\t\t\t%llu\n",
- fs_usage->nr_inodes);
+ fs_usage->u.nr_inodes);
pr_buf(out, "online reserved:\t\t%llu\n",
fs_usage->online_reserved);
for (i = 0;
- i < ARRAY_SIZE(fs_usage->persistent_reserved);
+ i < ARRAY_SIZE(fs_usage->u.persistent_reserved);
i++) {
pr_buf(out, "%u replicas:\n", i + 1);
pr_buf(out, "\treserved:\t\t%llu\n",
- fs_usage->persistent_reserved[i]);
+ fs_usage->u.persistent_reserved[i]);
}
for (i = 0; i < c->replicas.nr; i++) {
pr_buf(out, "\t");
bch2_replicas_entry_to_text(out, e);
- pr_buf(out, ":\t%llu\n", fs_usage->replicas[i]);
+ pr_buf(out, ":\t%llu\n", fs_usage->u.replicas[i]);
}
}
-#define RESERVE_FACTOR 6
-
static u64 reserve_factor(u64 r)
{
return r + (round_up(r, (1 << RESERVE_FACTOR)) >> RESERVE_FACTOR);
}
-static u64 avail_factor(u64 r)
+u64 bch2_fs_sectors_used(struct bch_fs *c, struct bch_fs_usage_online *fs_usage)
{
- return (r << RESERVE_FACTOR) / ((1 << RESERVE_FACTOR) + 1);
-}
-
-u64 bch2_fs_sectors_used(struct bch_fs *c, struct bch_fs_usage *fs_usage)
-{
- return min(fs_usage->hidden +
- fs_usage->btree +
- fs_usage->data +
- reserve_factor(fs_usage->reserved +
+ return min(fs_usage->u.hidden +
+ fs_usage->u.btree +
+ fs_usage->u.data +
+ reserve_factor(fs_usage->u.reserved +
fs_usage->online_reserved),
c->capacity);
}
data = bch2_fs_usage_read_one(c, &c->usage_base->data) +
bch2_fs_usage_read_one(c, &c->usage_base->btree);
reserved = bch2_fs_usage_read_one(c, &c->usage_base->reserved) +
- bch2_fs_usage_read_one(c, &c->usage_base->online_reserved);
+ percpu_u64_get(c->online_reserved);
ret.used = min(ret.capacity, data + reserve_factor(reserved));
ret.free = ret.capacity - ret.used;
return !is_available_bucket(m);
}
-static inline int is_fragmented_bucket(struct bucket_mark m,
- struct bch_dev *ca)
-{
- if (!m.owned_by_allocator &&
- m.data_type == BCH_DATA_USER &&
- bucket_sectors_used(m))
- return max_t(int, 0, (int) ca->mi.bucket_size -
- bucket_sectors_used(m));
- return 0;
-}
-
-static inline enum bch_data_type bucket_type(struct bucket_mark m)
+static inline int bucket_sectors_fragmented(struct bch_dev *ca,
+ struct bucket_mark m)
{
- return m.cached_sectors && !m.dirty_sectors
- ? BCH_DATA_CACHED
- : m.data_type;
+ return bucket_sectors_used(m)
+ ? max(0, (int) ca->mi.bucket_size - (int) bucket_sectors_used(m))
+ : 0;
}
-static bool bucket_became_unavailable(struct bucket_mark old,
- struct bucket_mark new)
+static inline int is_stripe_data_bucket(struct bucket_mark m)
{
- return is_available_bucket(old) &&
- !is_available_bucket(new);
+ return m.stripe && m.data_type != BCH_DATA_parity;
}
-int bch2_fs_usage_apply(struct bch_fs *c,
- struct bch_fs_usage *fs_usage,
- struct disk_reservation *disk_res,
- unsigned journal_seq)
+static inline enum bch_data_type bucket_type(struct bucket_mark m)
{
- s64 added = fs_usage->data + fs_usage->reserved;
- s64 should_not_have_added;
- int ret = 0;
-
- percpu_rwsem_assert_held(&c->mark_lock);
-
- /*
- * Not allowed to reduce sectors_available except by getting a
- * reservation:
- */
- should_not_have_added = added - (s64) (disk_res ? disk_res->sectors : 0);
- if (WARN_ONCE(should_not_have_added > 0,
- "disk usage increased by %lli without a reservation",
- should_not_have_added)) {
- atomic64_sub(should_not_have_added, &c->sectors_available);
- added -= should_not_have_added;
- ret = -1;
- }
-
- if (added > 0) {
- disk_res->sectors -= added;
- fs_usage->online_reserved -= added;
- }
-
- preempt_disable();
- acc_u64s((u64 *) fs_usage_ptr(c, journal_seq, false),
- (u64 *) fs_usage, fs_usage_u64s(c));
- preempt_enable();
-
- return ret;
+ return m.cached_sectors && !m.dirty_sectors
+ ? BCH_DATA_cached
+ : m.data_type;
}
static inline void account_bucket(struct bch_fs_usage *fs_usage,
enum bch_data_type type,
int nr, s64 size)
{
- if (type == BCH_DATA_SB || type == BCH_DATA_JOURNAL)
+ if (type == BCH_DATA_sb || type == BCH_DATA_journal)
fs_usage->hidden += size;
- dev_usage->buckets[type] += nr;
+ dev_usage->d[type].buckets += nr;
}
static void bch2_dev_usage_update(struct bch_fs *c, struct bch_dev *ca,
- struct bch_fs_usage *fs_usage,
struct bucket_mark old, struct bucket_mark new,
- bool gc)
+ u64 journal_seq, bool gc)
{
- struct bch_dev_usage *dev_usage;
-
- percpu_rwsem_assert_held(&c->mark_lock);
-
- bch2_fs_inconsistent_on(old.data_type && new.data_type &&
- old.data_type != new.data_type, c,
- "different types of data in same bucket: %s, %s",
- bch2_data_types[old.data_type],
- bch2_data_types[new.data_type]);
+ struct bch_fs_usage *fs_usage;
+ struct bch_dev_usage *u;
preempt_disable();
- dev_usage = this_cpu_ptr(ca->usage[gc]);
+ fs_usage = fs_usage_ptr(c, journal_seq, gc);
+ u = dev_usage_ptr(ca, journal_seq, gc);
if (bucket_type(old))
- account_bucket(fs_usage, dev_usage, bucket_type(old),
+ account_bucket(fs_usage, u, bucket_type(old),
-1, -ca->mi.bucket_size);
if (bucket_type(new))
- account_bucket(fs_usage, dev_usage, bucket_type(new),
+ account_bucket(fs_usage, u, bucket_type(new),
1, ca->mi.bucket_size);
- dev_usage->buckets_alloc +=
- (int) new.owned_by_allocator - (int) old.owned_by_allocator;
- dev_usage->buckets_ec +=
- (int) new.stripe - (int) old.stripe;
- dev_usage->buckets_unavailable +=
+ u->buckets_ec += (int) new.stripe - (int) old.stripe;
+ u->buckets_unavailable +=
is_unavailable_bucket(new) - is_unavailable_bucket(old);
- dev_usage->sectors[old.data_type] -= old.dirty_sectors;
- dev_usage->sectors[new.data_type] += new.dirty_sectors;
- dev_usage->sectors[BCH_DATA_CACHED] +=
+ u->d[old.data_type].sectors -= old.dirty_sectors;
+ u->d[new.data_type].sectors += new.dirty_sectors;
+ u->d[BCH_DATA_cached].sectors +=
(int) new.cached_sectors - (int) old.cached_sectors;
- dev_usage->sectors_fragmented +=
- is_fragmented_bucket(new, ca) - is_fragmented_bucket(old, ca);
+
+ u->d[old.data_type].fragmented -= bucket_sectors_fragmented(ca, old);
+ u->d[new.data_type].fragmented += bucket_sectors_fragmented(ca, new);
+
preempt_enable();
if (!is_available_bucket(old) && is_available_bucket(new))
bch2_wake_allocator(ca);
}
-void bch2_dev_usage_from_buckets(struct bch_fs *c)
+static inline int __update_replicas(struct bch_fs *c,
+ struct bch_fs_usage *fs_usage,
+ struct bch_replicas_entry *r,
+ s64 sectors)
{
- struct bch_dev *ca;
- struct bucket_mark old = { .v.counter = 0 };
- struct bucket_array *buckets;
- struct bucket *g;
- unsigned i;
- int cpu;
-
- c->usage_base->hidden = 0;
-
- for_each_member_device(ca, c, i) {
- for_each_possible_cpu(cpu)
- memset(per_cpu_ptr(ca->usage[0], cpu), 0,
- sizeof(*ca->usage[0]));
+ int idx = bch2_replicas_entry_idx(c, r);
- buckets = bucket_array(ca);
+ if (idx < 0)
+ return -1;
- for_each_bucket(g, buckets)
- bch2_dev_usage_update(c, ca, c->usage_base,
- old, g->mark, false);
- }
+ fs_usage_data_type_to_base(fs_usage, r->data_type, sectors);
+ fs_usage->replicas[idx] += sectors;
+ return 0;
}
-#define bucket_data_cmpxchg(c, ca, fs_usage, g, new, expr) \
-({ \
- struct bucket_mark _old = bucket_cmpxchg(g, new, expr); \
- \
- bch2_dev_usage_update(c, ca, fs_usage, _old, new, gc); \
- _old; \
-})
-
-static inline void update_replicas(struct bch_fs *c,
- struct bch_fs_usage *fs_usage,
- struct bch_replicas_entry *r,
- s64 sectors)
+static inline int update_replicas(struct bch_fs *c, struct bkey_s_c k,
+ struct bch_replicas_entry *r, s64 sectors,
+ unsigned journal_seq, bool gc)
{
- int idx = bch2_replicas_entry_idx(c, r);
+ struct bch_fs_usage __percpu *fs_usage;
+ int idx, ret = 0;
+ char buf[200];
- BUG_ON(idx < 0);
- BUG_ON(!sectors);
+ percpu_down_read(&c->mark_lock);
- switch (r->data_type) {
- case BCH_DATA_BTREE:
- fs_usage->btree += sectors;
- break;
- case BCH_DATA_USER:
- fs_usage->data += sectors;
- break;
- case BCH_DATA_CACHED:
- fs_usage->cached += sectors;
- break;
+ idx = bch2_replicas_entry_idx(c, r);
+ if (idx < 0 &&
+ (test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags) ||
+ fsck_err(c, "no replicas entry\n"
+ " while marking %s",
+ (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf)))) {
+ percpu_up_read(&c->mark_lock);
+ ret = bch2_mark_replicas(c, r);
+ if (ret)
+ return ret;
+
+ percpu_down_read(&c->mark_lock);
+ idx = bch2_replicas_entry_idx(c, r);
+ }
+ if (idx < 0) {
+ ret = -1;
+ goto err;
}
+
+ preempt_disable();
+ fs_usage = fs_usage_ptr(c, journal_seq, gc);
+ fs_usage_data_type_to_base(fs_usage, r->data_type, sectors);
fs_usage->replicas[idx] += sectors;
+ preempt_enable();
+err:
+fsck_err:
+ percpu_up_read(&c->mark_lock);
+ return ret;
}
-static inline void update_cached_sectors(struct bch_fs *c,
- struct bch_fs_usage *fs_usage,
- unsigned dev, s64 sectors)
+static inline int update_cached_sectors(struct bch_fs *c,
+ struct bkey_s_c k,
+ unsigned dev, s64 sectors,
+ unsigned journal_seq, bool gc)
{
struct bch_replicas_padded r;
bch2_replicas_entry_cached(&r.e, dev);
- update_replicas(c, fs_usage, &r.e, sectors);
+ return update_replicas(c, k, &r.e, sectors, journal_seq, gc);
}
static struct replicas_delta_list *
{
struct replicas_delta_list *d = trans->fs_usage_deltas;
unsigned new_size = d ? (d->size + more) * 2 : 128;
+ unsigned alloc_size = sizeof(*d) + new_size;
+
+ WARN_ON_ONCE(alloc_size > REPLICAS_DELTA_LIST_MAX);
if (!d || d->used + more > d->size) {
- d = krealloc(d, sizeof(*d) + new_size, GFP_NOIO|__GFP_ZERO);
- BUG_ON(!d);
+ d = krealloc(d, alloc_size, GFP_NOIO|__GFP_ZERO);
+
+ BUG_ON(!d && alloc_size > REPLICAS_DELTA_LIST_MAX);
+
+ if (!d) {
+ d = mempool_alloc(&trans->c->replicas_delta_pool, GFP_NOIO);
+ memset(d, 0, REPLICAS_DELTA_LIST_MAX);
+
+ if (trans->fs_usage_deltas)
+ memcpy(d, trans->fs_usage_deltas,
+ trans->fs_usage_deltas->size + sizeof(*d));
+
+ new_size = REPLICAS_DELTA_LIST_MAX - sizeof(*d);
+ kfree(trans->fs_usage_deltas);
+ }
d->size = new_size;
trans->fs_usage_deltas = d;
{
struct replicas_delta_list *d;
struct replicas_delta *n;
- unsigned b = replicas_entry_bytes(r) + 8;
+ unsigned b;
+
+ if (!sectors)
+ return;
+ b = replicas_entry_bytes(r) + 8;
d = replicas_deltas_realloc(trans, b);
n = (void *) d->d + d->used;
n->delta = sectors;
memcpy(&n->r, r, replicas_entry_bytes(r));
+ bch2_replicas_entry_sort(&n->r);
d->used += b;
}
update_replicas_list(trans, &r.e, sectors);
}
-void bch2_replicas_delta_list_apply(struct bch_fs *c,
- struct bch_fs_usage *fs_usage,
- struct replicas_delta_list *r)
-{
- struct replicas_delta *d = r->d;
- struct replicas_delta *top = (void *) r->d + r->used;
-
- acc_u64s((u64 *) fs_usage,
- (u64 *) &r->fs_usage, sizeof(*fs_usage) / sizeof(u64));
-
- while (d != top) {
- BUG_ON((void *) d > (void *) top);
-
- update_replicas(c, fs_usage, &d->r, d->delta);
-
- d = (void *) d + replicas_entry_bytes(&d->r) + 8;
- }
-}
-
-#define do_mark_fn(fn, c, pos, flags, ...) \
-({ \
- int gc, ret = 0; \
- \
- percpu_rwsem_assert_held(&c->mark_lock); \
- \
- for (gc = 0; gc < 2 && !ret; gc++) \
- if (!gc == !(flags & BCH_BUCKET_MARK_GC) || \
- (gc && gc_visited(c, pos))) \
- ret = fn(c, __VA_ARGS__, gc); \
- ret; \
-})
-
-static int __bch2_invalidate_bucket(struct bch_fs *c, struct bch_dev *ca,
- size_t b, struct bucket_mark *ret,
- bool gc)
-{
- struct bch_fs_usage *fs_usage = fs_usage_ptr(c, 0, gc);
- struct bucket *g = __bucket(ca, b, gc);
- struct bucket_mark old, new;
-
- old = bucket_data_cmpxchg(c, ca, fs_usage, g, new, ({
- BUG_ON(!is_available_bucket(new));
-
- new.owned_by_allocator = true;
- new.dirty = true;
- new.data_type = 0;
- new.cached_sectors = 0;
- new.dirty_sectors = 0;
- new.gen++;
- }));
-
- if (old.cached_sectors)
- update_cached_sectors(c, fs_usage, ca->dev_idx,
- -((s64) old.cached_sectors));
-
- if (!gc)
- *ret = old;
- return 0;
-}
-
-void bch2_invalidate_bucket(struct bch_fs *c, struct bch_dev *ca,
- size_t b, struct bucket_mark *old)
-{
- do_mark_fn(__bch2_invalidate_bucket, c, gc_phase(GC_PHASE_START), 0,
- ca, b, old);
-
- if (!old->owned_by_allocator && old->cached_sectors)
- trace_invalidate(ca, bucket_to_sector(ca, b),
- old->cached_sectors);
-}
-
-static int __bch2_mark_alloc_bucket(struct bch_fs *c, struct bch_dev *ca,
- size_t b, bool owned_by_allocator,
- bool gc)
+void bch2_mark_alloc_bucket(struct bch_fs *c, struct bch_dev *ca,
+ size_t b, bool owned_by_allocator)
{
- struct bch_fs_usage *fs_usage = fs_usage_ptr(c, 0, gc);
- struct bucket *g = __bucket(ca, b, gc);
+ struct bucket *g = bucket(ca, b);
struct bucket_mark old, new;
- old = bucket_data_cmpxchg(c, ca, fs_usage, g, new, ({
+ old = bucket_cmpxchg(g, new, ({
new.owned_by_allocator = owned_by_allocator;
}));
- BUG_ON(!gc &&
- !owned_by_allocator && !old.owned_by_allocator);
-
- return 0;
-}
-
-void bch2_mark_alloc_bucket(struct bch_fs *c, struct bch_dev *ca,
- size_t b, bool owned_by_allocator,
- struct gc_pos pos, unsigned flags)
-{
- preempt_disable();
-
- do_mark_fn(__bch2_mark_alloc_bucket, c, pos, flags,
- ca, b, owned_by_allocator);
-
- preempt_enable();
+ BUG_ON(owned_by_allocator == old.owned_by_allocator);
}
-static int bch2_mark_alloc(struct bch_fs *c, struct bkey_s_c k,
- struct bch_fs_usage *fs_usage,
- u64 journal_seq, unsigned flags)
+static int bch2_mark_alloc(struct btree_trans *trans,
+ struct bkey_s_c old, struct bkey_s_c new,
+ unsigned flags)
{
- bool gc = flags & BCH_BUCKET_MARK_GC;
- struct bkey_alloc_unpacked u;
+ bool gc = flags & BTREE_TRIGGER_GC;
+ u64 journal_seq = trans->journal_res.seq;
+ struct bch_fs *c = trans->c;
+ struct bkey_alloc_unpacked old_u = bch2_alloc_unpack(old);
+ struct bkey_alloc_unpacked new_u = bch2_alloc_unpack(new);
struct bch_dev *ca;
struct bucket *g;
- struct bucket_mark old, m;
+ struct bucket_mark old_m, m;
+ int ret = 0;
/*
* alloc btree is read in by bch2_alloc_read, not gc:
*/
- if ((flags & BCH_BUCKET_MARK_GC) &&
- !(flags & BCH_BUCKET_MARK_BUCKET_INVALIDATE))
+ if ((flags & BTREE_TRIGGER_GC) &&
+ !(flags & BTREE_TRIGGER_BUCKET_INVALIDATE))
return 0;
- ca = bch_dev_bkey_exists(c, k.k->p.inode);
+ if ((flags & BTREE_TRIGGER_INSERT) &&
+ !old_u.data_type != !new_u.data_type &&
+ new.k->type == KEY_TYPE_alloc_v3) {
+ struct bch_alloc_v3 *v = (struct bch_alloc_v3 *) new.v;
+
+ BUG_ON(!journal_seq);
- if (k.k->p.offset >= ca->mi.nbuckets)
+ /*
+ * If the btree updates referring to a bucket weren't flushed
+ * before the bucket became empty again, then the we don't have
+ * to wait on a journal flush before we can reuse the bucket:
+ */
+ v->journal_seq = !new_u.data_type &&
+ bch2_journal_noflush_seq(&c->journal, journal_seq)
+ ? 0 : cpu_to_le64(journal_seq);
+ }
+
+ ca = bch_dev_bkey_exists(c, new.k->p.inode);
+
+ if (new.k->p.offset >= ca->mi.nbuckets)
return 0;
- g = __bucket(ca, k.k->p.offset, gc);
- u = bch2_alloc_unpack(k);
+ percpu_down_read(&c->mark_lock);
+ if (!gc && new_u.gen != old_u.gen)
+ *bucket_gen(ca, new.k->p.offset) = new_u.gen;
+
+ g = __bucket(ca, new.k->p.offset, gc);
- old = bucket_cmpxchg(g, m, ({
- m.gen = u.gen;
- m.data_type = u.data_type;
- m.dirty_sectors = u.dirty_sectors;
- m.cached_sectors = u.cached_sectors;
+ old_m = bucket_cmpxchg(g, m, ({
+ m.gen = new_u.gen;
+ m.data_type = new_u.data_type;
+ m.dirty_sectors = new_u.dirty_sectors;
+ m.cached_sectors = new_u.cached_sectors;
+ m.stripe = new_u.stripe != 0;
if (journal_seq) {
m.journal_seq_valid = 1;
}
}));
- if (!(flags & BCH_BUCKET_MARK_ALLOC_READ))
- bch2_dev_usage_update(c, ca, fs_usage, old, m, gc);
+ bch2_dev_usage_update(c, ca, old_m, m, journal_seq, gc);
- g->io_time[READ] = u.read_time;
- g->io_time[WRITE] = u.write_time;
- g->oldest_gen = u.oldest_gen;
+ g->io_time[READ] = new_u.read_time;
+ g->io_time[WRITE] = new_u.write_time;
+ g->oldest_gen = new_u.oldest_gen;
g->gen_valid = 1;
+ g->stripe = new_u.stripe;
+ g->stripe_redundancy = new_u.stripe_redundancy;
+ percpu_up_read(&c->mark_lock);
/*
* need to know if we're getting called from the invalidate path or
* not:
*/
- if ((flags & BCH_BUCKET_MARK_BUCKET_INVALIDATE) &&
- old.cached_sectors) {
- update_cached_sectors(c, fs_usage, ca->dev_idx,
- -old.cached_sectors);
- trace_invalidate(ca, bucket_to_sector(ca, k.k->p.offset),
- old.cached_sectors);
+ if ((flags & BTREE_TRIGGER_BUCKET_INVALIDATE) &&
+ old_m.cached_sectors) {
+ ret = update_cached_sectors(c, new, ca->dev_idx,
+ -old_m.cached_sectors,
+ journal_seq, gc);
+ if (ret) {
+ bch2_fs_fatal_error(c, "bch2_mark_alloc(): no replicas entry while updating cached sectors");
+ return ret;
+ }
+
+ trace_invalidate(ca, bucket_to_sector(ca, new.k->p.offset),
+ old_m.cached_sectors);
}
return 0;
overflow; \
})
-static int __bch2_mark_metadata_bucket(struct bch_fs *c, struct bch_dev *ca,
- size_t b, enum bch_data_type type,
- unsigned sectors, bool gc)
+void bch2_mark_metadata_bucket(struct bch_fs *c, struct bch_dev *ca,
+ size_t b, enum bch_data_type data_type,
+ unsigned sectors, struct gc_pos pos,
+ unsigned flags)
{
- struct bucket *g = __bucket(ca, b, gc);
+ struct bucket *g;
struct bucket_mark old, new;
bool overflow;
- BUG_ON(type != BCH_DATA_SB &&
- type != BCH_DATA_JOURNAL);
+ BUG_ON(!(flags & BTREE_TRIGGER_GC));
+ BUG_ON(data_type != BCH_DATA_sb &&
+ data_type != BCH_DATA_journal);
+
+ /*
+ * Backup superblock might be past the end of our normal usable space:
+ */
+ if (b >= ca->mi.nbuckets)
+ return;
+ percpu_down_read(&c->mark_lock);
+ g = gc_bucket(ca, b);
old = bucket_cmpxchg(g, new, ({
- new.dirty = true;
- new.data_type = type;
+ new.data_type = data_type;
overflow = checked_add(new.dirty_sectors, sectors);
}));
+ bch2_fs_inconsistent_on(old.data_type &&
+ old.data_type != data_type, c,
+ "different types of data in same bucket: %s, %s",
+ bch2_data_types[old.data_type],
+ bch2_data_types[data_type]);
+
bch2_fs_inconsistent_on(overflow, c,
- "bucket sector count overflow: %u + %u > U16_MAX",
+ "bucket %u:%zu gen %u data type %s sector count overflow: %u + %u > U16_MAX",
+ ca->dev_idx, b, new.gen,
+ bch2_data_types[old.data_type ?: data_type],
old.dirty_sectors, sectors);
- if (c)
- bch2_dev_usage_update(c, ca, fs_usage_ptr(c, 0, gc),
- old, new, gc);
-
- return 0;
+ bch2_dev_usage_update(c, ca, old, new, 0, true);
+ percpu_up_read(&c->mark_lock);
}
-void bch2_mark_metadata_bucket(struct bch_fs *c, struct bch_dev *ca,
- size_t b, enum bch_data_type type,
- unsigned sectors, struct gc_pos pos,
- unsigned flags)
+static s64 ptr_disk_sectors(s64 sectors, struct extent_ptr_decoded p)
{
- BUG_ON(type != BCH_DATA_SB &&
- type != BCH_DATA_JOURNAL);
-
- preempt_disable();
-
- if (likely(c)) {
- do_mark_fn(__bch2_mark_metadata_bucket, c, pos, flags,
- ca, b, type, sectors);
- } else {
- __bch2_mark_metadata_bucket(c, ca, b, type, sectors, 0);
- }
+ EBUG_ON(sectors < 0);
- preempt_enable();
+ return p.crc.compression_type &&
+ p.crc.compression_type != BCH_COMPRESSION_TYPE_incompressible
+ ? DIV_ROUND_UP_ULL(sectors * p.crc.compressed_size,
+ p.crc.uncompressed_size)
+ : sectors;
}
-static s64 ptr_disk_sectors_delta(struct extent_ptr_decoded p,
- unsigned offset, s64 delta,
- unsigned flags)
+static int check_bucket_ref(struct bch_fs *c,
+ struct bkey_s_c k,
+ const struct bch_extent_ptr *ptr,
+ s64 sectors, enum bch_data_type ptr_data_type,
+ u8 bucket_gen, u8 bucket_data_type,
+ u16 dirty_sectors, u16 cached_sectors)
{
- if (flags & BCH_BUCKET_MARK_OVERWRITE_SPLIT) {
- BUG_ON(offset + -delta > p.crc.live_size);
+ size_t bucket_nr = PTR_BUCKET_NR(bch_dev_bkey_exists(c, ptr->dev), ptr);
+ u16 bucket_sectors = !ptr->cached
+ ? dirty_sectors
+ : cached_sectors;
+ char buf[200];
- return -((s64) ptr_disk_sectors(p)) +
- __ptr_disk_sectors(p, offset) +
- __ptr_disk_sectors(p, p.crc.live_size -
- offset + delta);
- } else if (flags & BCH_BUCKET_MARK_OVERWRITE) {
- BUG_ON(offset + -delta > p.crc.live_size);
+ if (gen_after(ptr->gen, bucket_gen)) {
+ bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
+ "bucket %u:%zu gen %u data type %s: ptr gen %u newer than bucket gen\n"
+ "while marking %s",
+ ptr->dev, bucket_nr, bucket_gen,
+ bch2_data_types[bucket_data_type ?: ptr_data_type],
+ ptr->gen,
+ (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf));
+ return -EIO;
+ }
- return -((s64) ptr_disk_sectors(p)) +
- __ptr_disk_sectors(p, p.crc.live_size +
- delta);
- } else {
- return ptr_disk_sectors(p);
+ if (gen_cmp(bucket_gen, ptr->gen) > BUCKET_GC_GEN_MAX) {
+ bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
+ "bucket %u:%zu gen %u data type %s: ptr gen %u too stale\n"
+ "while marking %s",
+ ptr->dev, bucket_nr, bucket_gen,
+ bch2_data_types[bucket_data_type ?: ptr_data_type],
+ ptr->gen,
+ (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf));
+ return -EIO;
}
-}
-static void bucket_set_stripe(struct bch_fs *c,
- const struct bch_stripe *v,
- struct bch_fs_usage *fs_usage,
- u64 journal_seq,
- unsigned flags)
-{
- bool enabled = !(flags & BCH_BUCKET_MARK_OVERWRITE);
- bool gc = flags & BCH_BUCKET_MARK_GC;
- unsigned i;
+ if (bucket_gen != ptr->gen && !ptr->cached) {
+ bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
+ "bucket %u:%zu gen %u data type %s: stale dirty ptr (gen %u)\n"
+ "while marking %s",
+ ptr->dev, bucket_nr, bucket_gen,
+ bch2_data_types[bucket_data_type ?: ptr_data_type],
+ ptr->gen,
+ (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf));
+ return -EIO;
+ }
- for (i = 0; i < v->nr_blocks; i++) {
- const struct bch_extent_ptr *ptr = v->ptrs + i;
- struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
- struct bucket *g = PTR_BUCKET(ca, ptr, gc);
- struct bucket_mark new, old;
-
- old = bucket_data_cmpxchg(c, ca, fs_usage, g, new, ({
- new.dirty = true;
- new.stripe = enabled;
- if (journal_seq) {
- new.journal_seq_valid = 1;
- new.journal_seq = journal_seq;
- }
- }));
+ if (bucket_gen != ptr->gen)
+ return 1;
+
+ if (bucket_data_type && ptr_data_type &&
+ bucket_data_type != ptr_data_type) {
+ bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
+ "bucket %u:%zu gen %u different types of data in same bucket: %s, %s\n"
+ "while marking %s",
+ ptr->dev, bucket_nr, bucket_gen,
+ bch2_data_types[bucket_data_type],
+ bch2_data_types[ptr_data_type],
+ (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf));
+ return -EIO;
+ }
- /*
- * XXX write repair code for these, flag stripe as possibly bad
- */
- if (old.gen != ptr->gen)
- bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
- "stripe with stale pointer");
-#if 0
- /*
- * We'd like to check for these, but these checks don't work
- * yet:
- */
- if (old.stripe && enabled)
- bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
- "multiple stripes using same bucket");
-
- if (!old.stripe && !enabled)
- bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
- "deleting stripe but bucket not marked as stripe bucket");
-#endif
+ if ((unsigned) (bucket_sectors + sectors) > U16_MAX) {
+ bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
+ "bucket %u:%zu gen %u data type %s sector count overflow: %u + %lli > U16_MAX\n"
+ "while marking %s",
+ ptr->dev, bucket_nr, bucket_gen,
+ bch2_data_types[bucket_data_type ?: ptr_data_type],
+ bucket_sectors, sectors,
+ (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf));
+ return -EIO;
}
+
+ return 0;
}
-static bool bch2_mark_pointer(struct bch_fs *c,
- struct extent_ptr_decoded p,
- s64 sectors, enum bch_data_type data_type,
- struct bch_fs_usage *fs_usage,
+static int mark_stripe_bucket(struct btree_trans *trans,
+ struct bkey_s_c k,
+ unsigned ptr_idx,
u64 journal_seq, unsigned flags)
{
- bool gc = flags & BCH_BUCKET_MARK_GC;
- struct bucket_mark old, new;
- struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev);
- struct bucket *g = PTR_BUCKET(ca, &p.ptr, gc);
- bool overflow;
- u64 v;
+ struct bch_fs *c = trans->c;
+ const struct bch_stripe *s = bkey_s_c_to_stripe(k).v;
+ unsigned nr_data = s->nr_blocks - s->nr_redundant;
+ bool parity = ptr_idx >= nr_data;
+ enum bch_data_type data_type = parity ? BCH_DATA_parity : 0;
+ s64 sectors = parity ? le16_to_cpu(s->sectors) : 0;
+ const struct bch_extent_ptr *ptr = s->ptrs + ptr_idx;
+ struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
+ struct bucket *g;
+ struct bucket_mark new, old;
+ char buf[200];
+ int ret = 0;
- v = atomic64_read(&g->_mark.v);
- do {
- new.v.counter = old.v.counter = v;
+ BUG_ON(!(flags & BTREE_TRIGGER_GC));
- new.dirty = true;
+ /* * XXX doesn't handle deletion */
- /*
- * Check this after reading bucket mark to guard against
- * the allocator invalidating a bucket after we've already
- * checked the gen
- */
- if (gen_after(new.gen, p.ptr.gen)) {
- /* XXX write repair code for this */
- if (!p.ptr.cached &&
- test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags))
- bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
- "stale dirty pointer");
- return true;
+ percpu_down_read(&c->mark_lock);
+ g = PTR_GC_BUCKET(ca, ptr);
+
+ if (g->mark.dirty_sectors ||
+ (g->stripe && g->stripe != k.k->p.offset)) {
+ bch2_fs_inconsistent(c,
+ "bucket %u:%zu gen %u: multiple stripes using same bucket\n%s",
+ ptr->dev, PTR_BUCKET_NR(ca, ptr), g->mark.gen,
+ (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf));
+ ret = -EINVAL;
+ goto err;
+ }
+
+ old = bucket_cmpxchg(g, new, ({
+ ret = check_bucket_ref(c, k, ptr, sectors, data_type,
+ new.gen, new.data_type,
+ new.dirty_sectors, new.cached_sectors);
+ if (ret)
+ goto err;
+
+ new.dirty_sectors += sectors;
+ if (data_type)
+ new.data_type = data_type;
+
+ if (journal_seq) {
+ new.journal_seq_valid = 1;
+ new.journal_seq = journal_seq;
}
- if (!p.ptr.cached)
- overflow = checked_add(new.dirty_sectors, sectors);
- else
- overflow = checked_add(new.cached_sectors, sectors);
+ new.stripe = true;
+ }));
- if (!new.dirty_sectors &&
- !new.cached_sectors) {
- new.data_type = 0;
+ g->stripe = k.k->p.offset;
+ g->stripe_redundancy = s->nr_redundant;
- if (journal_seq) {
- new.journal_seq_valid = 1;
- new.journal_seq = journal_seq;
- }
- } else {
- new.data_type = data_type;
+ bch2_dev_usage_update(c, ca, old, new, journal_seq, true);
+err:
+ percpu_up_read(&c->mark_lock);
+
+ return 0;
+}
+
+static int __mark_pointer(struct btree_trans *trans,
+ struct bkey_s_c k,
+ const struct bch_extent_ptr *ptr,
+ s64 sectors, enum bch_data_type ptr_data_type,
+ u8 bucket_gen, u8 *bucket_data_type,
+ u16 *dirty_sectors, u16 *cached_sectors)
+{
+ u16 *dst_sectors = !ptr->cached
+ ? dirty_sectors
+ : cached_sectors;
+ int ret = check_bucket_ref(trans->c, k, ptr, sectors, ptr_data_type,
+ bucket_gen, *bucket_data_type,
+ *dirty_sectors, *cached_sectors);
+
+ if (ret)
+ return ret;
+
+ *dst_sectors += sectors;
+ *bucket_data_type = *dirty_sectors || *cached_sectors
+ ? ptr_data_type : 0;
+ return 0;
+}
+
+static int bch2_mark_pointer(struct btree_trans *trans,
+ struct bkey_s_c k,
+ struct extent_ptr_decoded p,
+ s64 sectors, enum bch_data_type data_type,
+ unsigned flags)
+{
+ u64 journal_seq = trans->journal_res.seq;
+ struct bch_fs *c = trans->c;
+ struct bucket_mark old, new;
+ struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev);
+ struct bucket *g;
+ u8 bucket_data_type;
+ u64 v;
+ int ret = 0;
+
+ BUG_ON(!(flags & BTREE_TRIGGER_GC));
+
+ percpu_down_read(&c->mark_lock);
+ g = PTR_GC_BUCKET(ca, &p.ptr);
+
+ v = atomic64_read(&g->_mark.v);
+ do {
+ new.v.counter = old.v.counter = v;
+ bucket_data_type = new.data_type;
+
+ ret = __mark_pointer(trans, k, &p.ptr, sectors,
+ data_type, new.gen,
+ &bucket_data_type,
+ &new.dirty_sectors,
+ &new.cached_sectors);
+ if (ret)
+ goto err;
+
+ new.data_type = bucket_data_type;
+
+ if (journal_seq) {
+ new.journal_seq_valid = 1;
+ new.journal_seq = journal_seq;
}
- if (flags & BCH_BUCKET_MARK_NOATOMIC) {
+ if (flags & BTREE_TRIGGER_NOATOMIC) {
g->_mark = new;
break;
}
old.v.counter,
new.v.counter)) != old.v.counter);
- bch2_fs_inconsistent_on(overflow, c,
- "bucket sector count overflow: %u + %lli > U16_MAX",
- !p.ptr.cached
- ? old.dirty_sectors
- : old.cached_sectors, sectors);
-
- bch2_dev_usage_update(c, ca, fs_usage, old, new, gc);
-
- BUG_ON(!gc && bucket_became_unavailable(old, new));
+ bch2_dev_usage_update(c, ca, old, new, journal_seq, true);
+err:
+ percpu_up_read(&c->mark_lock);
- return false;
+ return ret;
}
-static int bch2_mark_stripe_ptr(struct bch_fs *c,
+static int bch2_mark_stripe_ptr(struct btree_trans *trans,
+ struct bkey_s_c k,
struct bch_extent_stripe_ptr p,
enum bch_data_type data_type,
- struct bch_fs_usage *fs_usage,
- s64 sectors, unsigned flags)
+ s64 sectors,
+ unsigned flags)
{
- bool gc = flags & BCH_BUCKET_MARK_GC;
- struct stripe *m;
- unsigned old, new, nr_data;
- int blocks_nonempty_delta;
- s64 parity_sectors;
+ struct bch_fs *c = trans->c;
+ struct bch_replicas_padded r;
+ struct gc_stripe *m;
- BUG_ON(!sectors);
+ BUG_ON(!(flags & BTREE_TRIGGER_GC));
- m = genradix_ptr(&c->stripes[gc], p.idx);
+ m = genradix_ptr_alloc(&c->gc_stripes, p.idx, GFP_KERNEL);
+ if (!m) {
+ bch_err(c, "error allocating memory for gc_stripes, idx %llu",
+ (u64) p.idx);
+ return -ENOMEM;
+ }
spin_lock(&c->ec_stripes_heap_lock);
spin_unlock(&c->ec_stripes_heap_lock);
bch_err_ratelimited(c, "pointer to nonexistent stripe %llu",
(u64) p.idx);
+ bch2_inconsistent_error(c);
return -EIO;
}
- BUG_ON(m->r.e.data_type != data_type);
-
- nr_data = m->nr_blocks - m->nr_redundant;
-
- parity_sectors = DIV_ROUND_UP(abs(sectors) * m->nr_redundant, nr_data);
-
- if (sectors < 0)
- parity_sectors = -parity_sectors;
- sectors += parity_sectors;
-
- old = m->block_sectors[p.block];
m->block_sectors[p.block] += sectors;
- new = m->block_sectors[p.block];
-
- blocks_nonempty_delta = (int) !!new - (int) !!old;
- if (blocks_nonempty_delta) {
- m->blocks_nonempty += blocks_nonempty_delta;
-
- if (!gc)
- bch2_stripes_heap_update(c, m, p.idx);
- }
-
- m->dirty = true;
+ r = m->r;
spin_unlock(&c->ec_stripes_heap_lock);
- update_replicas(c, fs_usage, &m->r.e, sectors);
+ r.e.data_type = data_type;
+ update_replicas(c, k, &r.e, sectors, trans->journal_res.seq, true);
return 0;
}
-static int bch2_mark_extent(struct bch_fs *c, struct bkey_s_c k,
- unsigned offset, s64 sectors,
- enum bch_data_type data_type,
- struct bch_fs_usage *fs_usage,
- unsigned journal_seq, unsigned flags)
+static int bch2_mark_extent(struct btree_trans *trans,
+ struct bkey_s_c old, struct bkey_s_c new,
+ unsigned flags)
{
+ u64 journal_seq = trans->journal_res.seq;
+ struct bch_fs *c = trans->c;
+ struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE ? old: new;
struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
const union bch_extent_entry *entry;
struct extent_ptr_decoded p;
struct bch_replicas_padded r;
+ enum bch_data_type data_type = bkey_is_btree_ptr(k.k)
+ ? BCH_DATA_btree
+ : BCH_DATA_user;
+ s64 sectors = bkey_is_btree_ptr(k.k)
+ ? btree_sectors(c)
+ : k.k->size;
s64 dirty_sectors = 0;
- unsigned i;
+ bool stale;
int ret;
+ BUG_ON(!(flags & BTREE_TRIGGER_GC));
+
r.e.data_type = data_type;
r.e.nr_devs = 0;
r.e.nr_required = 1;
- BUG_ON(!sectors);
-
bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
- s64 disk_sectors = data_type == BCH_DATA_BTREE
- ? sectors
- : ptr_disk_sectors_delta(p, offset, sectors, flags);
- bool stale = bch2_mark_pointer(c, p, disk_sectors, data_type,
- fs_usage, journal_seq, flags);
+ s64 disk_sectors = ptr_disk_sectors(sectors, p);
+
+ if (flags & BTREE_TRIGGER_OVERWRITE)
+ disk_sectors = -disk_sectors;
+
+ ret = bch2_mark_pointer(trans, k, p, disk_sectors,
+ data_type, flags);
+ if (ret < 0)
+ return ret;
+
+ stale = ret > 0;
if (p.ptr.cached) {
- if (disk_sectors && !stale)
- update_cached_sectors(c, fs_usage, p.ptr.dev,
- disk_sectors);
- } else if (!p.ec_nr) {
+ if (!stale) {
+ ret = update_cached_sectors(c, k, p.ptr.dev,
+ disk_sectors, journal_seq, true);
+ if (ret) {
+ bch2_fs_fatal_error(c, "bch2_mark_extent(): no replicas entry while updating cached sectors");
+ return ret;
+ }
+ }
+ } else if (!p.has_ec) {
dirty_sectors += disk_sectors;
r.e.devs[r.e.nr_devs++] = p.ptr.dev;
} else {
- for (i = 0; i < p.ec_nr; i++) {
- ret = bch2_mark_stripe_ptr(c, p.ec[i],
- data_type, fs_usage,
- disk_sectors, flags);
- if (ret)
- return ret;
- }
-
+ ret = bch2_mark_stripe_ptr(trans, k, p.ec, data_type,
+ disk_sectors, flags);
+ if (ret)
+ return ret;
+
+ /*
+ * There may be other dirty pointers in this extent, but
+ * if so they're not required for mounting if we have an
+ * erasure coded pointer in this extent:
+ */
r.e.nr_required = 0;
}
}
- if (dirty_sectors)
- update_replicas(c, fs_usage, &r.e, dirty_sectors);
+ if (r.e.nr_devs) {
+ ret = update_replicas(c, k, &r.e, dirty_sectors, journal_seq, true);
+ if (ret) {
+ char buf[200];
+
+ bch2_bkey_val_to_text(&PBUF(buf), c, k);
+ bch2_fs_fatal_error(c, "no replicas entry for %s", buf);
+ return ret;
+ }
+ }
return 0;
}
-static int bch2_mark_stripe(struct bch_fs *c, struct bkey_s_c k,
- struct bch_fs_usage *fs_usage,
- u64 journal_seq, unsigned flags)
+static int bch2_mark_stripe(struct btree_trans *trans,
+ struct bkey_s_c old, struct bkey_s_c new,
+ unsigned flags)
{
- bool gc = flags & BCH_BUCKET_MARK_GC;
- struct bkey_s_c_stripe s = bkey_s_c_to_stripe(k);
- size_t idx = s.k->p.offset;
- struct stripe *m = genradix_ptr(&c->stripes[gc], idx);
+ bool gc = flags & BTREE_TRIGGER_GC;
+ u64 journal_seq = trans->journal_res.seq;
+ struct bch_fs *c = trans->c;
+ u64 idx = new.k->p.offset;
+ const struct bch_stripe *old_s = old.k->type == KEY_TYPE_stripe
+ ? bkey_s_c_to_stripe(old).v : NULL;
+ const struct bch_stripe *new_s = new.k->type == KEY_TYPE_stripe
+ ? bkey_s_c_to_stripe(new).v : NULL;
unsigned i;
+ int ret;
- spin_lock(&c->ec_stripes_heap_lock);
+ BUG_ON(gc && old_s);
- if (!m || ((flags & BCH_BUCKET_MARK_OVERWRITE) && !m->alive)) {
- spin_unlock(&c->ec_stripes_heap_lock);
- bch_err_ratelimited(c, "error marking nonexistent stripe %zu",
- idx);
- return -1;
- }
+ if (!gc) {
+ struct stripe *m = genradix_ptr(&c->stripes, idx);
+
+ if (!m || (old_s && !m->alive)) {
+ char buf1[200], buf2[200];
+
+ bch2_bkey_val_to_text(&PBUF(buf1), c, old);
+ bch2_bkey_val_to_text(&PBUF(buf2), c, new);
+ bch_err_ratelimited(c, "error marking nonexistent stripe %llu while marking\n"
+ "old %s\n"
+ "new %s", idx, buf1, buf2);
+ bch2_inconsistent_error(c);
+ return -1;
+ }
+
+ if (!new_s) {
+ spin_lock(&c->ec_stripes_heap_lock);
+ bch2_stripes_heap_del(c, m, idx);
+ spin_unlock(&c->ec_stripes_heap_lock);
- if (!(flags & BCH_BUCKET_MARK_OVERWRITE)) {
- m->sectors = le16_to_cpu(s.v->sectors);
- m->algorithm = s.v->algorithm;
- m->nr_blocks = s.v->nr_blocks;
- m->nr_redundant = s.v->nr_redundant;
+ memset(m, 0, sizeof(*m));
+ } else {
+ m->alive = true;
+ m->sectors = le16_to_cpu(new_s->sectors);
+ m->algorithm = new_s->algorithm;
+ m->nr_blocks = new_s->nr_blocks;
+ m->nr_redundant = new_s->nr_redundant;
+ m->blocks_nonempty = 0;
+
+ for (i = 0; i < new_s->nr_blocks; i++)
+ m->blocks_nonempty += !!stripe_blockcount_get(new_s, i);
- bch2_bkey_to_replicas(&m->r.e, k);
+ spin_lock(&c->ec_stripes_heap_lock);
+ bch2_stripes_heap_update(c, m, idx);
+ spin_unlock(&c->ec_stripes_heap_lock);
+ }
+ } else {
+ struct gc_stripe *m =
+ genradix_ptr_alloc(&c->gc_stripes, idx, GFP_KERNEL);
+ if (!m) {
+ bch_err(c, "error allocating memory for gc_stripes, idx %llu",
+ idx);
+ return -ENOMEM;
+ }
/*
- * XXX: account for stripes somehow here
+ * This will be wrong when we bring back runtime gc: we should
+ * be unmarking the old key and then marking the new key
*/
-#if 0
- update_replicas(c, fs_usage, &m->r.e, stripe_sectors);
-#endif
-
- /* gc recalculates these fields: */
- if (!(flags & BCH_BUCKET_MARK_GC)) {
- for (i = 0; i < s.v->nr_blocks; i++) {
- m->block_sectors[i] =
- stripe_blockcount_get(s.v, i);
- m->blocks_nonempty += !!m->block_sectors[i];
- }
+ m->alive = true;
+ m->sectors = le16_to_cpu(new_s->sectors);
+ m->nr_blocks = new_s->nr_blocks;
+ m->nr_redundant = new_s->nr_redundant;
+
+ for (i = 0; i < new_s->nr_blocks; i++)
+ m->ptrs[i] = new_s->ptrs[i];
+
+ bch2_bkey_to_replicas(&m->r.e, new);
+
+ /*
+ * gc recalculates this field from stripe ptr
+ * references:
+ */
+ memset(m->block_sectors, 0, sizeof(m->block_sectors));
+
+ for (i = 0; i < new_s->nr_blocks; i++) {
+ ret = mark_stripe_bucket(trans, new, i, journal_seq, flags);
+ if (ret)
+ return ret;
}
- if (!gc)
- bch2_stripes_heap_update(c, m, idx);
- m->alive = true;
- } else {
- if (!gc)
- bch2_stripes_heap_del(c, m, idx);
- memset(m, 0, sizeof(*m));
- }
+ ret = update_replicas(c, new, &m->r.e,
+ ((s64) m->sectors * m->nr_redundant),
+ journal_seq, gc);
+ if (ret) {
+ char buf[200];
- spin_unlock(&c->ec_stripes_heap_lock);
+ bch2_bkey_val_to_text(&PBUF(buf), c, new);
+ bch2_fs_fatal_error(c, "no replicas entry for %s", buf);
+ return ret;
+ }
+ }
- bucket_set_stripe(c, s.v, fs_usage, 0, flags);
return 0;
}
-int bch2_mark_key_locked(struct bch_fs *c,
- struct bkey_s_c k,
- unsigned offset, s64 sectors,
- struct bch_fs_usage *fs_usage,
- u64 journal_seq, unsigned flags)
+static int bch2_mark_inode(struct btree_trans *trans,
+ struct bkey_s_c old, struct bkey_s_c new,
+ unsigned flags)
{
- int ret = 0;
+ struct bch_fs *c = trans->c;
+ struct bch_fs_usage __percpu *fs_usage;
+ u64 journal_seq = trans->journal_res.seq;
- preempt_disable();
+ if (flags & BTREE_TRIGGER_INSERT) {
+ struct bch_inode_v2 *v = (struct bch_inode_v2 *) new.v;
- if (!fs_usage || (flags & BCH_BUCKET_MARK_GC))
- fs_usage = fs_usage_ptr(c, journal_seq,
- flags & BCH_BUCKET_MARK_GC);
+ BUG_ON(!journal_seq);
+ BUG_ON(new.k->type != KEY_TYPE_inode_v2);
- switch (k.k->type) {
- case KEY_TYPE_alloc:
- ret = bch2_mark_alloc(c, k, fs_usage, journal_seq, flags);
- break;
- case KEY_TYPE_btree_ptr:
- sectors = !(flags & BCH_BUCKET_MARK_OVERWRITE)
- ? c->opts.btree_node_size
- : -c->opts.btree_node_size;
+ v->bi_journal_seq = cpu_to_le64(journal_seq);
+ }
- ret = bch2_mark_extent(c, k, offset, sectors, BCH_DATA_BTREE,
- fs_usage, journal_seq, flags);
- break;
- case KEY_TYPE_extent:
- case KEY_TYPE_reflink_v:
- ret = bch2_mark_extent(c, k, offset, sectors, BCH_DATA_USER,
- fs_usage, journal_seq, flags);
- break;
- case KEY_TYPE_stripe:
- ret = bch2_mark_stripe(c, k, fs_usage, journal_seq, flags);
- break;
- case KEY_TYPE_inode:
- if (!(flags & BCH_BUCKET_MARK_OVERWRITE))
- fs_usage->nr_inodes++;
- else
- fs_usage->nr_inodes--;
- break;
- case KEY_TYPE_reservation: {
- unsigned replicas = bkey_s_c_to_reservation(k).v->nr_replicas;
+ if (flags & BTREE_TRIGGER_GC) {
+ percpu_down_read(&c->mark_lock);
+ preempt_disable();
- sectors *= replicas;
- replicas = clamp_t(unsigned, replicas, 1,
- ARRAY_SIZE(fs_usage->persistent_reserved));
+ fs_usage = fs_usage_ptr(c, journal_seq, flags & BTREE_TRIGGER_GC);
+ fs_usage->nr_inodes += bkey_is_inode(new.k);
+ fs_usage->nr_inodes -= bkey_is_inode(old.k);
- fs_usage->reserved += sectors;
- fs_usage->persistent_reserved[replicas - 1] += sectors;
- break;
- }
+ preempt_enable();
+ percpu_up_read(&c->mark_lock);
}
+ return 0;
+}
+
+static int bch2_mark_reservation(struct btree_trans *trans,
+ struct bkey_s_c old, struct bkey_s_c new,
+ unsigned flags)
+{
+ struct bch_fs *c = trans->c;
+ struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE ? old: new;
+ struct bch_fs_usage __percpu *fs_usage;
+ unsigned replicas = bkey_s_c_to_reservation(k).v->nr_replicas;
+ s64 sectors = (s64) k.k->size;
+
+ BUG_ON(!(flags & BTREE_TRIGGER_GC));
+
+ if (flags & BTREE_TRIGGER_OVERWRITE)
+ sectors = -sectors;
+ sectors *= replicas;
+
+ percpu_down_read(&c->mark_lock);
+ preempt_disable();
+
+ fs_usage = fs_usage_ptr(c, trans->journal_res.seq, flags & BTREE_TRIGGER_GC);
+ replicas = clamp_t(unsigned, replicas, 1,
+ ARRAY_SIZE(fs_usage->persistent_reserved));
+
+ fs_usage->reserved += sectors;
+ fs_usage->persistent_reserved[replicas - 1] += sectors;
preempt_enable();
+ percpu_up_read(&c->mark_lock);
- return ret;
+ return 0;
}
-int bch2_mark_key(struct bch_fs *c, struct bkey_s_c k,
- unsigned offset, s64 sectors,
- struct bch_fs_usage *fs_usage,
- u64 journal_seq, unsigned flags)
+static s64 __bch2_mark_reflink_p(struct bch_fs *c, struct bkey_s_c_reflink_p p,
+ u64 *idx, unsigned flags, size_t r_idx)
{
- int ret;
+ struct reflink_gc *r;
+ int add = !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1;
+ s64 ret = 0;
- percpu_down_read(&c->mark_lock);
- ret = bch2_mark_key_locked(c, k, offset, sectors,
- fs_usage, journal_seq, flags);
- percpu_up_read(&c->mark_lock);
+ if (r_idx >= c->reflink_gc_nr)
+ goto not_found;
+ r = genradix_ptr(&c->reflink_gc_table, r_idx);
+ if (*idx < r->offset - r->size)
+ goto not_found;
+
+ BUG_ON((s64) r->refcount + add < 0);
+
+ r->refcount += add;
+ *idx = r->offset;
+ return 0;
+not_found:
+ *idx = U64_MAX;
+ ret = -EIO;
+
+ /*
+ * XXX: we're replacing the entire reflink pointer with an error
+ * key, we should just be replacing the part that was missing:
+ */
+ if (fsck_err(c, "%llu:%llu len %u points to nonexistent indirect extent %llu",
+ p.k->p.inode, p.k->p.offset, p.k->size, *idx)) {
+ struct bkey_i_error new;
+
+ bkey_init(&new.k);
+ new.k.type = KEY_TYPE_error;
+ new.k.p = p.k->p;
+ new.k.size = p.k->size;
+ ret = bch2_journal_key_insert(c, BTREE_ID_extents, 0, &new.k_i);
+ }
+fsck_err:
return ret;
}
-inline int bch2_mark_overwrite(struct btree_trans *trans,
- struct btree_iter *iter,
- struct bkey_s_c old,
- struct bkey_i *new,
- struct bch_fs_usage *fs_usage,
+static int bch2_mark_reflink_p(struct btree_trans *trans,
+ struct bkey_s_c old, struct bkey_s_c new,
unsigned flags)
{
- struct bch_fs *c = trans->c;
- struct btree *b = iter->l[0].b;
- unsigned offset = 0;
- s64 sectors = 0;
+ struct bch_fs *c = trans->c;
+ struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE ? old: new;
+ struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);
+ struct reflink_gc *ref;
+ size_t l, r, m;
+ u64 idx = le64_to_cpu(p.v->idx);
+ u64 end = le64_to_cpu(p.v->idx) + p.k->size;
+ int ret = 0;
- flags |= BCH_BUCKET_MARK_OVERWRITE;
+ BUG_ON(!(flags & BTREE_TRIGGER_GC));
- if (btree_node_is_extents(b)
- ? bkey_cmp(new->k.p, bkey_start_pos(old.k)) <= 0
- : bkey_cmp(new->k.p, old.k->p))
- return 0;
+ if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix) {
+ idx -= le32_to_cpu(p.v->front_pad);
+ end += le32_to_cpu(p.v->back_pad);
+ }
- if (btree_node_is_extents(b)) {
- switch (bch2_extent_overlap(&new->k, old.k)) {
- case BCH_EXTENT_OVERLAP_ALL:
- offset = 0;
- sectors = -((s64) old.k->size);
- break;
- case BCH_EXTENT_OVERLAP_BACK:
- offset = bkey_start_offset(&new->k) -
- bkey_start_offset(old.k);
- sectors = bkey_start_offset(&new->k) -
- old.k->p.offset;
- break;
- case BCH_EXTENT_OVERLAP_FRONT:
- offset = 0;
- sectors = bkey_start_offset(old.k) -
- new->k.p.offset;
- break;
- case BCH_EXTENT_OVERLAP_MIDDLE:
- offset = bkey_start_offset(&new->k) -
- bkey_start_offset(old.k);
- sectors = -((s64) new->k.size);
- flags |= BCH_BUCKET_MARK_OVERWRITE_SPLIT;
- break;
- }
+ l = 0;
+ r = c->reflink_gc_nr;
+ while (l < r) {
+ m = l + (r - l) / 2;
- BUG_ON(sectors >= 0);
+ ref = genradix_ptr(&c->reflink_gc_table, m);
+ if (ref->offset <= idx)
+ l = m + 1;
+ else
+ r = m;
}
- return bch2_mark_key_locked(c, old, offset, sectors, fs_usage,
- trans->journal_res.seq, flags) ?: 1;
+ while (idx < end && !ret)
+ ret = __bch2_mark_reflink_p(c, p, &idx, flags, l++);
+
+ return ret;
}
-int bch2_mark_update(struct btree_trans *trans,
- struct btree_insert_entry *insert,
- struct bch_fs_usage *fs_usage,
- unsigned flags)
+int bch2_mark_key(struct btree_trans *trans,
+ struct bkey_s_c old,
+ struct bkey_s_c new,
+ unsigned flags)
{
- struct bch_fs *c = trans->c;
- struct btree_iter *iter = insert->iter;
- struct btree *b = iter->l[0].b;
- struct btree_node_iter node_iter = iter->l[0].iter;
- struct bkey_packed *_k;
- int ret = 0;
+ struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE ? old: new;
- if (!btree_node_type_needs_gc(iter->btree_id))
+ switch (k.k->type) {
+ case KEY_TYPE_alloc:
+ case KEY_TYPE_alloc_v2:
+ case KEY_TYPE_alloc_v3:
+ return bch2_mark_alloc(trans, old, new, flags);
+ case KEY_TYPE_btree_ptr:
+ case KEY_TYPE_btree_ptr_v2:
+ case KEY_TYPE_extent:
+ case KEY_TYPE_reflink_v:
+ return bch2_mark_extent(trans, old, new, flags);
+ case KEY_TYPE_stripe:
+ return bch2_mark_stripe(trans, old, new, flags);
+ case KEY_TYPE_inode:
+ case KEY_TYPE_inode_v2:
+ return bch2_mark_inode(trans, old, new, flags);
+ case KEY_TYPE_reservation:
+ return bch2_mark_reservation(trans, old, new, flags);
+ case KEY_TYPE_reflink_p:
+ return bch2_mark_reflink_p(trans, old, new, flags);
+ case KEY_TYPE_snapshot:
+ return bch2_mark_snapshot(trans, old, new, flags);
+ default:
return 0;
+ }
+}
- if (!(trans->flags & BTREE_INSERT_NOMARK_INSERT))
- bch2_mark_key_locked(c, bkey_i_to_s_c(insert->k),
- 0, insert->k->k.size,
- fs_usage, trans->journal_res.seq,
- BCH_BUCKET_MARK_INSERT|flags);
+int bch2_mark_update(struct btree_trans *trans, struct btree_path *path,
+ struct bkey_i *new, unsigned flags)
+{
+ struct bkey _deleted = KEY(0, 0, 0);
+ struct bkey_s_c deleted = (struct bkey_s_c) { &_deleted, NULL };
+ struct bkey_s_c old;
+ struct bkey unpacked;
+ int ret;
- if (unlikely(trans->flags & BTREE_INSERT_NOMARK_OVERWRITES))
- return 0;
+ _deleted.p = path->pos;
- /*
- * For non extents, we only mark the new key, not the key being
- * overwritten - unless we're actually deleting:
- */
- if ((iter->btree_id == BTREE_ID_ALLOC ||
- iter->btree_id == BTREE_ID_EC) &&
- !bkey_deleted(&insert->k->k))
+ if (unlikely(flags & BTREE_TRIGGER_NORUN))
return 0;
- while ((_k = bch2_btree_node_iter_peek_filter(&node_iter, b,
- KEY_TYPE_discard))) {
- struct bkey unpacked;
- struct bkey_s_c k = bkey_disassemble(b, _k, &unpacked);
+ if (!btree_node_type_needs_gc(path->btree_id))
+ return 0;
- ret = bch2_mark_overwrite(trans, iter, k, insert->k,
- fs_usage, flags);
- if (ret <= 0)
- break;
+ old = bch2_btree_path_peek_slot(path, &unpacked);
- bch2_btree_node_iter_advance(&node_iter, b);
+ if (old.k->type == new->k.type &&
+ ((1U << old.k->type) & BTREE_TRIGGER_WANTS_OLD_AND_NEW)) {
+ ret = bch2_mark_key(trans, old, bkey_i_to_s_c(new),
+ BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE|flags);
+ } else {
+ ret = bch2_mark_key(trans, deleted, bkey_i_to_s_c(new),
+ BTREE_TRIGGER_INSERT|flags) ?:
+ bch2_mark_key(trans, old, deleted,
+ BTREE_TRIGGER_OVERWRITE|flags);
}
return ret;
}
-void bch2_trans_fs_usage_apply(struct btree_trans *trans,
- struct bch_fs_usage *fs_usage)
+static noinline __cold
+void fs_usage_apply_warn(struct btree_trans *trans,
+ unsigned disk_res_sectors,
+ s64 should_not_have_added)
{
struct bch_fs *c = trans->c;
struct btree_insert_entry *i;
- static int warned_disk_usage = 0;
- u64 disk_res_sectors = trans->disk_res ? trans->disk_res->sectors : 0;
char buf[200];
- if (!bch2_fs_usage_apply(c, fs_usage, trans->disk_res,
- trans->journal_res.seq) ||
- warned_disk_usage ||
- xchg(&warned_disk_usage, 1))
- return;
-
- bch_err(c, "disk usage increased more than %llu sectors reserved",
- disk_res_sectors);
-
- trans_for_each_update_iter(trans, i) {
- struct btree_iter *iter = i->iter;
- struct btree *b = iter->l[0].b;
- struct btree_node_iter node_iter = iter->l[0].iter;
- struct bkey_packed *_k;
+ bch_err(c, "disk usage increased %lli more than %u sectors reserved",
+ should_not_have_added, disk_res_sectors);
+ trans_for_each_update(trans, i) {
pr_err("while inserting");
bch2_bkey_val_to_text(&PBUF(buf), c, bkey_i_to_s_c(i->k));
pr_err("%s", buf);
pr_err("overlapping with");
- node_iter = iter->l[0].iter;
- while ((_k = bch2_btree_node_iter_peek_filter(&node_iter, b,
- KEY_TYPE_discard))) {
- struct bkey unpacked;
- struct bkey_s_c k;
-
- k = bkey_disassemble(b, _k, &unpacked);
-
- if (btree_node_is_extents(b)
- ? bkey_cmp(i->k->k.p, bkey_start_pos(k.k)) <= 0
- : bkey_cmp(i->k->k.p, k.k->p))
- break;
+ if (!i->cached) {
+ struct bkey u;
+ struct bkey_s_c k = bch2_btree_path_peek_slot(i->path, &u);
bch2_bkey_val_to_text(&PBUF(buf), c, k);
pr_err("%s", buf);
+ } else {
+ struct bkey_cached *ck = (void *) i->path->l[0].b;
- bch2_btree_node_iter_advance(&node_iter, b);
+ if (ck->valid) {
+ bch2_bkey_val_to_text(&PBUF(buf), c, bkey_i_to_s_c(ck->k));
+ pr_err("%s", buf);
+ }
}
}
+ __WARN();
}
-/* trans_mark: */
-
-static int trans_get_key(struct btree_trans *trans,
- enum btree_id btree_id, struct bpos pos,
- struct btree_iter **iter,
- struct bkey_s_c *k)
+int bch2_trans_fs_usage_apply(struct btree_trans *trans,
+ struct replicas_delta_list *deltas)
{
- struct btree_insert_entry *i;
- int ret;
+ struct bch_fs *c = trans->c;
+ static int warned_disk_usage = 0;
+ bool warn = false;
+ unsigned disk_res_sectors = trans->disk_res ? trans->disk_res->sectors : 0;
+ struct replicas_delta *d = deltas->d, *d2;
+ struct replicas_delta *top = (void *) deltas->d + deltas->used;
+ struct bch_fs_usage *dst;
+ s64 added = 0, should_not_have_added;
+ unsigned i;
- for (i = trans->updates;
- i < trans->updates + trans->nr_updates;
- i++)
- if (!i->deferred &&
- i->iter->btree_id == btree_id &&
- (btree_node_type_is_extents(btree_id)
- ? bkey_cmp(pos, bkey_start_pos(&i->k->k)) >= 0 &&
- bkey_cmp(pos, i->k->k.p) < 0
- : !bkey_cmp(pos, i->iter->pos))) {
- *iter = i->iter;
- *k = bkey_i_to_s_c(i->k);
- return 1;
+ percpu_down_read(&c->mark_lock);
+ preempt_disable();
+ dst = fs_usage_ptr(c, trans->journal_res.seq, false);
+
+ for (d = deltas->d; d != top; d = replicas_delta_next(d)) {
+ switch (d->r.data_type) {
+ case BCH_DATA_btree:
+ case BCH_DATA_user:
+ case BCH_DATA_parity:
+ added += d->delta;
}
- *iter = __bch2_trans_get_iter(trans, btree_id, pos,
- BTREE_ITER_SLOTS|BTREE_ITER_INTENT, 0);
- if (IS_ERR(*iter))
- return PTR_ERR(*iter);
+ if (__update_replicas(c, dst, &d->r, d->delta))
+ goto need_mark;
+ }
- bch2_trans_iter_free_on_commit(trans, *iter);
+ dst->nr_inodes += deltas->nr_inodes;
- *k = bch2_btree_iter_peek_slot(*iter);
- ret = bkey_err(*k);
- if (ret)
- bch2_trans_iter_put(trans, *iter);
- return ret;
-}
+ for (i = 0; i < BCH_REPLICAS_MAX; i++) {
+ added += deltas->persistent_reserved[i];
+ dst->reserved += deltas->persistent_reserved[i];
+ dst->persistent_reserved[i] += deltas->persistent_reserved[i];
+ }
-static void *trans_update_key(struct btree_trans *trans,
- struct btree_iter *iter,
- unsigned u64s)
-{
- struct bkey_i *new_k;
- unsigned i;
+ /*
+ * Not allowed to reduce sectors_available except by getting a
+ * reservation:
+ */
+ should_not_have_added = added - (s64) disk_res_sectors;
+ if (unlikely(should_not_have_added > 0)) {
+ u64 old, new, v = atomic64_read(&c->sectors_available);
- new_k = bch2_trans_kmalloc(trans, u64s * sizeof(u64));
- if (IS_ERR(new_k))
- return new_k;
+ do {
+ old = v;
+ new = max_t(s64, 0, old - should_not_have_added);
+ } while ((v = atomic64_cmpxchg(&c->sectors_available,
+ old, new)) != old);
- bkey_init(&new_k->k);
- new_k->k.p = iter->pos;
+ added -= should_not_have_added;
+ warn = true;
+ }
- for (i = 0; i < trans->nr_updates; i++)
- if (!trans->updates[i].deferred &&
- trans->updates[i].iter == iter) {
- trans->updates[i].k = new_k;
- return new_k;
- }
+ if (added > 0) {
+ trans->disk_res->sectors -= added;
+ this_cpu_sub(*c->online_reserved, added);
+ }
- bch2_trans_update(trans, ((struct btree_insert_entry) {
- .iter = iter,
- .k = new_k,
- .triggered = true,
- }));
+ preempt_enable();
+ percpu_up_read(&c->mark_lock);
+
+ if (unlikely(warn) && !xchg(&warned_disk_usage, 1))
+ fs_usage_apply_warn(trans, disk_res_sectors, should_not_have_added);
+ return 0;
+need_mark:
+ /* revert changes: */
+ for (d2 = deltas->d; d2 != d; d2 = replicas_delta_next(d2))
+ BUG_ON(__update_replicas(c, dst, &d2->r, -d2->delta));
- return new_k;
+ preempt_enable();
+ percpu_up_read(&c->mark_lock);
+ return -1;
}
-static int bch2_trans_mark_pointer(struct btree_trans *trans,
- struct extent_ptr_decoded p,
- s64 sectors, enum bch_data_type data_type)
+/* trans_mark: */
+
+static int bch2_trans_start_alloc_update(struct btree_trans *trans, struct btree_iter *iter,
+ const struct bch_extent_ptr *ptr,
+ struct bkey_alloc_unpacked *u)
{
struct bch_fs *c = trans->c;
- struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev);
- struct btree_iter *iter;
+ struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
struct bkey_s_c k;
- struct bucket *g;
- struct bucket_mark m;
- struct bkey_alloc_unpacked u;
- struct bkey_i_alloc *a;
- bool overflow;
int ret;
- ret = trans_get_key(trans, BTREE_ID_ALLOC,
- POS(p.ptr.dev, PTR_BUCKET_NR(ca, &p.ptr)),
- &iter, &k);
- if (ret < 0)
+ bch2_trans_iter_init(trans, iter, BTREE_ID_alloc,
+ POS(ptr->dev, PTR_BUCKET_NR(ca, ptr)),
+ BTREE_ITER_WITH_UPDATES|
+ BTREE_ITER_CACHED|
+ BTREE_ITER_INTENT);
+ k = bch2_btree_iter_peek_slot(iter);
+ ret = bkey_err(k);
+ if (ret) {
+ bch2_trans_iter_exit(trans, iter);
return ret;
-
- if (!ret) {
- percpu_down_read(&c->mark_lock);
- g = bucket(ca, iter->pos.offset);
- m = READ_ONCE(g->mark);
- u = alloc_mem_to_key(g, m);
- percpu_up_read(&c->mark_lock);
- } else {
- u = bch2_alloc_unpack(k);
}
- if (gen_after(u.gen, p.ptr.gen)) {
- ret = 1;
- goto out;
- }
-
- if (!p.ptr.cached)
- overflow = checked_add(u.dirty_sectors, sectors);
- else
- overflow = checked_add(u.cached_sectors, sectors);
+ *u = bch2_alloc_unpack(k);
+ return 0;
+}
- u.data_type = u.dirty_sectors || u.cached_sectors
- ? data_type : 0;
+static int bch2_trans_mark_pointer(struct btree_trans *trans,
+ struct bkey_s_c k, struct extent_ptr_decoded p,
+ s64 sectors, enum bch_data_type data_type)
+{
+ struct btree_iter iter;
+ struct bkey_alloc_unpacked u;
+ int ret;
- bch2_fs_inconsistent_on(overflow, c,
- "bucket sector count overflow: %u + %lli > U16_MAX",
- !p.ptr.cached
- ? u.dirty_sectors
- : u.cached_sectors, sectors);
+ ret = bch2_trans_start_alloc_update(trans, &iter, &p.ptr, &u);
+ if (ret)
+ return ret;
- a = trans_update_key(trans, iter, BKEY_ALLOC_U64s_MAX);
- ret = PTR_ERR_OR_ZERO(a);
+ ret = __mark_pointer(trans, k, &p.ptr, sectors, data_type,
+ u.gen, &u.data_type,
+ &u.dirty_sectors, &u.cached_sectors);
if (ret)
goto out;
- bkey_alloc_init(&a->k_i);
- a->k.p = iter->pos;
- bch2_alloc_pack(a, u);
+ ret = bch2_alloc_write(trans, &iter, &u, 0);
+ if (ret)
+ goto out;
out:
- bch2_trans_iter_put(trans, iter);
+ bch2_trans_iter_exit(trans, &iter);
return ret;
}
static int bch2_trans_mark_stripe_ptr(struct btree_trans *trans,
- struct bch_extent_stripe_ptr p,
+ struct extent_ptr_decoded p,
s64 sectors, enum bch_data_type data_type)
{
struct bch_fs *c = trans->c;
- struct bch_replicas_padded r;
- struct btree_iter *iter;
- struct bkey_i *new_k;
+ struct btree_iter iter;
struct bkey_s_c k;
- struct bkey_s_stripe s;
- unsigned nr_data;
- s64 parity_sectors;
+ struct bkey_i_stripe *s;
+ struct bch_replicas_padded r;
int ret = 0;
- ret = trans_get_key(trans, BTREE_ID_EC, POS(0, p.idx), &iter, &k);
- if (ret < 0)
- return ret;
+ bch2_trans_iter_init(trans, &iter, BTREE_ID_stripes, POS(0, p.ec.idx),
+ BTREE_ITER_INTENT|
+ BTREE_ITER_WITH_UPDATES);
+ k = bch2_btree_iter_peek_slot(&iter);
+ ret = bkey_err(k);
+ if (ret)
+ goto err;
if (k.k->type != KEY_TYPE_stripe) {
bch2_fs_inconsistent(c,
"pointer to nonexistent stripe %llu",
- (u64) p.idx);
+ (u64) p.ec.idx);
+ bch2_inconsistent_error(c);
ret = -EIO;
- goto out;
+ goto err;
}
- new_k = trans_update_key(trans, iter, k.k->u64s);
- ret = PTR_ERR_OR_ZERO(new_k);
- if (ret)
- goto out;
-
- bkey_reassemble(new_k, k);
- s = bkey_i_to_s_stripe(new_k);
-
- nr_data = s.v->nr_blocks - s.v->nr_redundant;
-
- parity_sectors = DIV_ROUND_UP(abs(sectors) * s.v->nr_redundant, nr_data);
+ if (!bch2_ptr_matches_stripe(bkey_s_c_to_stripe(k).v, p)) {
+ bch2_fs_inconsistent(c,
+ "stripe pointer doesn't match stripe %llu",
+ (u64) p.ec.idx);
+ ret = -EIO;
+ goto err;
+ }
- if (sectors < 0)
- parity_sectors = -parity_sectors;
+ s = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
+ ret = PTR_ERR_OR_ZERO(s);
+ if (ret)
+ goto err;
- stripe_blockcount_set(s.v, p.block,
- stripe_blockcount_get(s.v, p.block) +
- sectors + parity_sectors);
+ bkey_reassemble(&s->k_i, k);
+ stripe_blockcount_set(&s->v, p.ec.block,
+ stripe_blockcount_get(&s->v, p.ec.block) +
+ sectors);
- bch2_bkey_to_replicas(&r.e, s.s_c);
+ ret = bch2_trans_update(trans, &iter, &s->k_i, 0);
+ if (ret)
+ goto err;
+ bch2_bkey_to_replicas(&r.e, bkey_i_to_s_c(&s->k_i));
+ r.e.data_type = data_type;
update_replicas_list(trans, &r.e, sectors);
-out:
- bch2_trans_iter_put(trans, iter);
+err:
+ bch2_trans_iter_exit(trans, &iter);
return ret;
}
static int bch2_trans_mark_extent(struct btree_trans *trans,
- struct bkey_s_c k, unsigned offset,
- s64 sectors, unsigned flags,
- enum bch_data_type data_type)
+ struct bkey_s_c k, unsigned flags)
{
+ struct bch_fs *c = trans->c;
struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
const union bch_extent_entry *entry;
struct extent_ptr_decoded p;
struct bch_replicas_padded r;
+ enum bch_data_type data_type = bkey_is_btree_ptr(k.k)
+ ? BCH_DATA_btree
+ : BCH_DATA_user;
+ s64 sectors = bkey_is_btree_ptr(k.k)
+ ? btree_sectors(c)
+ : k.k->size;
s64 dirty_sectors = 0;
bool stale;
- unsigned i;
int ret;
r.e.data_type = data_type;
r.e.nr_devs = 0;
r.e.nr_required = 1;
- BUG_ON(!sectors);
-
bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
- s64 disk_sectors = data_type == BCH_DATA_BTREE
- ? sectors
- : ptr_disk_sectors_delta(p, offset, sectors, flags);
+ s64 disk_sectors = ptr_disk_sectors(sectors, p);
- /*
- * can happen due to rounding with compressed extents:
- */
- if (!disk_sectors)
- continue;
+ if (flags & BTREE_TRIGGER_OVERWRITE)
+ disk_sectors = -disk_sectors;
- ret = bch2_trans_mark_pointer(trans, p, disk_sectors,
- data_type);
+ ret = bch2_trans_mark_pointer(trans, k, p,
+ disk_sectors, data_type);
if (ret < 0)
return ret;
stale = ret > 0;
if (p.ptr.cached) {
- if (disk_sectors && !stale)
+ if (!stale)
update_cached_sectors_list(trans, p.ptr.dev,
disk_sectors);
- } else if (!p.ec_nr) {
+ } else if (!p.has_ec) {
dirty_sectors += disk_sectors;
r.e.devs[r.e.nr_devs++] = p.ptr.dev;
} else {
- for (i = 0; i < p.ec_nr; i++) {
- ret = bch2_trans_mark_stripe_ptr(trans, p.ec[i],
- disk_sectors, data_type);
- if (ret)
- return ret;
- }
+ ret = bch2_trans_mark_stripe_ptr(trans, p,
+ disk_sectors, data_type);
+ if (ret)
+ return ret;
r.e.nr_required = 0;
}
}
- if (dirty_sectors)
+ if (r.e.nr_devs)
update_replicas_list(trans, &r.e, dirty_sectors);
return 0;
}
+static int bch2_trans_mark_stripe_bucket(struct btree_trans *trans,
+ struct bkey_s_c_stripe s,
+ unsigned idx, bool deleting)
+{
+ struct bch_fs *c = trans->c;
+ const struct bch_extent_ptr *ptr = &s.v->ptrs[idx];
+ struct btree_iter iter;
+ struct bkey_alloc_unpacked u;
+ enum bch_data_type data_type = idx >= s.v->nr_blocks - s.v->nr_redundant
+ ? BCH_DATA_parity : 0;
+ s64 sectors = data_type ? le16_to_cpu(s.v->sectors) : 0;
+ int ret = 0;
+
+ if (deleting)
+ sectors = -sectors;
+
+ ret = bch2_trans_start_alloc_update(trans, &iter, ptr, &u);
+ if (ret)
+ return ret;
+
+ ret = check_bucket_ref(c, s.s_c, ptr, sectors, data_type,
+ u.gen, u.data_type,
+ u.dirty_sectors, u.cached_sectors);
+ if (ret)
+ goto err;
+
+ if (!deleting) {
+ if (bch2_fs_inconsistent_on(u.stripe ||
+ u.stripe_redundancy, c,
+ "bucket %llu:%llu gen %u data type %s dirty_sectors %u: multiple stripes using same bucket (%u, %llu)",
+ iter.pos.inode, iter.pos.offset, u.gen,
+ bch2_data_types[u.data_type],
+ u.dirty_sectors,
+ u.stripe, s.k->p.offset)) {
+ ret = -EIO;
+ goto err;
+ }
+
+ if (bch2_fs_inconsistent_on(data_type && u.dirty_sectors, c,
+ "bucket %llu:%llu gen %u data type %s dirty_sectors %u: data already in stripe bucket %llu",
+ iter.pos.inode, iter.pos.offset, u.gen,
+ bch2_data_types[u.data_type],
+ u.dirty_sectors,
+ s.k->p.offset)) {
+ ret = -EIO;
+ goto err;
+ }
+
+ u.stripe = s.k->p.offset;
+ u.stripe_redundancy = s.v->nr_redundant;
+ } else {
+ if (bch2_fs_inconsistent_on(u.stripe != s.k->p.offset ||
+ u.stripe_redundancy != s.v->nr_redundant, c,
+ "bucket %llu:%llu gen %u: not marked as stripe when deleting stripe %llu (got %u)",
+ iter.pos.inode, iter.pos.offset, u.gen,
+ s.k->p.offset, u.stripe)) {
+ ret = -EIO;
+ goto err;
+ }
+
+ u.stripe = 0;
+ u.stripe_redundancy = 0;
+ }
+
+ u.dirty_sectors += sectors;
+ if (data_type)
+ u.data_type = !deleting ? data_type : 0;
+
+ ret = bch2_alloc_write(trans, &iter, &u, 0);
+ if (ret)
+ goto err;
+err:
+ bch2_trans_iter_exit(trans, &iter);
+ return ret;
+}
+
+static int bch2_trans_mark_stripe(struct btree_trans *trans,
+ struct bkey_s_c old, struct bkey_s_c new,
+ unsigned flags)
+{
+ struct bkey_s_c_stripe old_s = { .k = NULL };
+ struct bkey_s_c_stripe new_s = { .k = NULL };
+ struct bch_replicas_padded r;
+ unsigned i, nr_blocks;
+ int ret = 0;
+
+ if (old.k->type == KEY_TYPE_stripe)
+ old_s = bkey_s_c_to_stripe(old);
+ if (new.k->type == KEY_TYPE_stripe)
+ new_s = bkey_s_c_to_stripe(new);
+
+ /*
+ * If the pointers aren't changing, we don't need to do anything:
+ */
+ if (new_s.k && old_s.k &&
+ new_s.v->nr_blocks == old_s.v->nr_blocks &&
+ new_s.v->nr_redundant == old_s.v->nr_redundant &&
+ !memcmp(old_s.v->ptrs, new_s.v->ptrs,
+ new_s.v->nr_blocks * sizeof(struct bch_extent_ptr)))
+ return 0;
+
+ BUG_ON(new_s.k && old_s.k &&
+ (new_s.v->nr_blocks != old_s.v->nr_blocks ||
+ new_s.v->nr_redundant != old_s.v->nr_redundant));
+
+ nr_blocks = new_s.k ? new_s.v->nr_blocks : old_s.v->nr_blocks;
+
+ if (new_s.k) {
+ s64 sectors = le16_to_cpu(new_s.v->sectors);
+
+ bch2_bkey_to_replicas(&r.e, new);
+ update_replicas_list(trans, &r.e, sectors * new_s.v->nr_redundant);
+ }
+
+ if (old_s.k) {
+ s64 sectors = -((s64) le16_to_cpu(old_s.v->sectors));
+
+ bch2_bkey_to_replicas(&r.e, old);
+ update_replicas_list(trans, &r.e, sectors * old_s.v->nr_redundant);
+ }
+
+ for (i = 0; i < nr_blocks; i++) {
+ if (new_s.k && old_s.k &&
+ !memcmp(&new_s.v->ptrs[i],
+ &old_s.v->ptrs[i],
+ sizeof(new_s.v->ptrs[i])))
+ continue;
+
+ if (new_s.k) {
+ ret = bch2_trans_mark_stripe_bucket(trans, new_s, i, false);
+ if (ret)
+ break;
+ }
+
+ if (old_s.k) {
+ ret = bch2_trans_mark_stripe_bucket(trans, old_s, i, true);
+ if (ret)
+ break;
+ }
+ }
+
+ return ret;
+}
+
+static int bch2_trans_mark_inode(struct btree_trans *trans,
+ struct bkey_s_c old,
+ struct bkey_s_c new,
+ unsigned flags)
+{
+ int nr = bkey_is_inode(new.k) - bkey_is_inode(old.k);
+
+ if (nr) {
+ struct replicas_delta_list *d =
+ replicas_deltas_realloc(trans, 0);
+ d->nr_inodes += nr;
+ }
+
+ return 0;
+}
+
+static int bch2_trans_mark_reservation(struct btree_trans *trans,
+ struct bkey_s_c k, unsigned flags)
+{
+ unsigned replicas = bkey_s_c_to_reservation(k).v->nr_replicas;
+ s64 sectors = (s64) k.k->size;
+ struct replicas_delta_list *d;
+
+ if (flags & BTREE_TRIGGER_OVERWRITE)
+ sectors = -sectors;
+ sectors *= replicas;
+
+ d = replicas_deltas_realloc(trans, 0);
+
+ replicas = clamp_t(unsigned, replicas, 1,
+ ARRAY_SIZE(d->persistent_reserved));
+
+ d->persistent_reserved[replicas - 1] += sectors;
+ return 0;
+}
+
static int __bch2_trans_mark_reflink_p(struct btree_trans *trans,
struct bkey_s_c_reflink_p p,
- u64 idx, unsigned sectors,
- unsigned flags)
+ u64 *idx, unsigned flags)
{
struct bch_fs *c = trans->c;
- struct btree_iter *iter;
- struct bkey_i *new_k;
+ struct btree_iter iter;
struct bkey_s_c k;
- struct bkey_i_reflink_v *r_v;
- s64 ret;
+ struct bkey_i *n;
+ __le64 *refcount;
+ int add = !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1;
+ char buf[200];
+ int ret;
- ret = trans_get_key(trans, BTREE_ID_REFLINK,
- POS(0, idx), &iter, &k);
- if (ret < 0)
- return ret;
+ bch2_trans_iter_init(trans, &iter, BTREE_ID_reflink, POS(0, *idx),
+ BTREE_ITER_INTENT|
+ BTREE_ITER_WITH_UPDATES);
+ k = bch2_btree_iter_peek_slot(&iter);
+ ret = bkey_err(k);
+ if (ret)
+ goto err;
- if (k.k->type != KEY_TYPE_reflink_v) {
+ n = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
+ ret = PTR_ERR_OR_ZERO(n);
+ if (ret)
+ goto err;
+
+ bkey_reassemble(n, k);
+
+ refcount = bkey_refcount(n);
+ if (!refcount) {
+ bch2_bkey_val_to_text(&PBUF(buf), c, p.s_c);
bch2_fs_inconsistent(c,
- "%llu:%llu len %u points to nonexistent indirect extent %llu",
- p.k->p.inode, p.k->p.offset, p.k->size, idx);
+ "nonexistent indirect extent at %llu while marking\n %s",
+ *idx, buf);
ret = -EIO;
goto err;
}
- if ((flags & BCH_BUCKET_MARK_OVERWRITE) &&
- (bkey_start_offset(k.k) < idx ||
- k.k->p.offset > idx + sectors))
- goto out;
+ if (!*refcount && (flags & BTREE_TRIGGER_OVERWRITE)) {
+ bch2_bkey_val_to_text(&PBUF(buf), c, p.s_c);
+ bch2_fs_inconsistent(c,
+ "indirect extent refcount underflow at %llu while marking\n %s",
+ *idx, buf);
+ ret = -EIO;
+ goto err;
+ }
- bch2_btree_iter_set_pos(iter, bkey_start_pos(k.k));
- BUG_ON(iter->uptodate > BTREE_ITER_NEED_PEEK);
+ if (flags & BTREE_TRIGGER_INSERT) {
+ struct bch_reflink_p *v = (struct bch_reflink_p *) p.v;
+ u64 pad;
- new_k = trans_update_key(trans, iter, k.k->u64s);
- ret = PTR_ERR_OR_ZERO(new_k);
- if (ret)
- goto err;
+ pad = max_t(s64, le32_to_cpu(v->front_pad),
+ le64_to_cpu(v->idx) - bkey_start_offset(k.k));
+ BUG_ON(pad > U32_MAX);
+ v->front_pad = cpu_to_le32(pad);
- bkey_reassemble(new_k, k);
- r_v = bkey_i_to_reflink_v(new_k);
+ pad = max_t(s64, le32_to_cpu(v->back_pad),
+ k.k->p.offset - p.k->size - le64_to_cpu(v->idx));
+ BUG_ON(pad > U32_MAX);
+ v->back_pad = cpu_to_le32(pad);
+ }
- le64_add_cpu(&r_v->v.refcount,
- !(flags & BCH_BUCKET_MARK_OVERWRITE) ? 1 : -1);
+ le64_add_cpu(refcount, add);
- if (!r_v->v.refcount) {
- r_v->k.type = KEY_TYPE_deleted;
- set_bkey_val_u64s(&r_v->k, 0);
+ if (!*refcount) {
+ n->k.type = KEY_TYPE_deleted;
+ set_bkey_val_u64s(&n->k, 0);
}
-out:
- ret = k.k->p.offset - idx;
+
+ bch2_btree_iter_set_pos_to_extent_start(&iter);
+ ret = bch2_trans_update(trans, &iter, n, 0);
+ if (ret)
+ goto err;
+
+ *idx = k.k->p.offset;
err:
- bch2_trans_iter_put(trans, iter);
+ bch2_trans_iter_exit(trans, &iter);
return ret;
}
static int bch2_trans_mark_reflink_p(struct btree_trans *trans,
- struct bkey_s_c_reflink_p p, unsigned offset,
- s64 sectors, unsigned flags)
+ struct bkey_s_c k, unsigned flags)
{
- u64 idx = le64_to_cpu(p.v->idx) + offset;
- s64 ret = 0;
-
- sectors = abs(sectors);
- BUG_ON(offset + sectors > p.k->size);
+ struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);
+ u64 idx, end_idx;
+ int ret = 0;
- while (sectors) {
- ret = __bch2_trans_mark_reflink_p(trans, p, idx, sectors, flags);
- if (ret < 0)
- break;
+ if (flags & BTREE_TRIGGER_INSERT) {
+ struct bch_reflink_p *v = (struct bch_reflink_p *) p.v;
- idx += ret;
- sectors = max_t(s64, 0LL, sectors - ret);
- ret = 0;
+ v->front_pad = v->back_pad = 0;
}
+ idx = le64_to_cpu(p.v->idx) - le32_to_cpu(p.v->front_pad);
+ end_idx = le64_to_cpu(p.v->idx) + p.k->size +
+ le32_to_cpu(p.v->back_pad);
+
+ while (idx < end_idx && !ret)
+ ret = __bch2_trans_mark_reflink_p(trans, p, &idx, flags);
+
return ret;
}
-int bch2_trans_mark_key(struct btree_trans *trans, struct bkey_s_c k,
- unsigned offset, s64 sectors, unsigned flags)
+int bch2_trans_mark_key(struct btree_trans *trans, struct bkey_s_c old,
+ struct bkey_s_c new, unsigned flags)
{
- struct replicas_delta_list *d;
- struct bch_fs *c = trans->c;
+ struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE ? old: new;
switch (k.k->type) {
case KEY_TYPE_btree_ptr:
- sectors = !(flags & BCH_BUCKET_MARK_OVERWRITE)
- ? c->opts.btree_node_size
- : -c->opts.btree_node_size;
-
- return bch2_trans_mark_extent(trans, k, offset, sectors,
- flags, BCH_DATA_BTREE);
+ case KEY_TYPE_btree_ptr_v2:
case KEY_TYPE_extent:
case KEY_TYPE_reflink_v:
- return bch2_trans_mark_extent(trans, k, offset, sectors,
- flags, BCH_DATA_USER);
+ return bch2_trans_mark_extent(trans, k, flags);
+ case KEY_TYPE_stripe:
+ return bch2_trans_mark_stripe(trans, old, new, flags);
case KEY_TYPE_inode:
- d = replicas_deltas_realloc(trans, 0);
-
- if (!(flags & BCH_BUCKET_MARK_OVERWRITE))
- d->fs_usage.nr_inodes++;
- else
- d->fs_usage.nr_inodes--;
- return 0;
- case KEY_TYPE_reservation: {
- unsigned replicas = bkey_s_c_to_reservation(k).v->nr_replicas;
-
- d = replicas_deltas_realloc(trans, 0);
-
- sectors *= replicas;
- replicas = clamp_t(unsigned, replicas, 1,
- ARRAY_SIZE(d->fs_usage.persistent_reserved));
-
- d->fs_usage.reserved += sectors;
- d->fs_usage.persistent_reserved[replicas - 1] += sectors;
- return 0;
- }
+ case KEY_TYPE_inode_v2:
+ return bch2_trans_mark_inode(trans, old, new, flags);
+ case KEY_TYPE_reservation:
+ return bch2_trans_mark_reservation(trans, k, flags);
case KEY_TYPE_reflink_p:
- return bch2_trans_mark_reflink_p(trans,
- bkey_s_c_to_reflink_p(k),
- offset, sectors, flags);
+ return bch2_trans_mark_reflink_p(trans, k, flags);
default:
return 0;
}
}
-int bch2_trans_mark_update(struct btree_trans *trans,
- struct btree_iter *iter,
- struct bkey_i *insert)
+static int __bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
+ struct bch_dev *ca, size_t b,
+ enum bch_data_type type,
+ unsigned sectors)
{
- struct btree *b = iter->l[0].b;
- struct btree_node_iter node_iter = iter->l[0].iter;
- struct bkey_packed *_k;
- int ret;
+ struct bch_fs *c = trans->c;
+ struct btree_iter iter;
+ struct bkey_alloc_unpacked u;
+ struct bch_extent_ptr ptr = {
+ .dev = ca->dev_idx,
+ .offset = bucket_to_sector(ca, b),
+ };
+ int ret = 0;
- if (!btree_node_type_needs_gc(iter->btree_id))
+ /*
+ * Backup superblock might be past the end of our normal usable space:
+ */
+ if (b >= ca->mi.nbuckets)
return 0;
- ret = bch2_trans_mark_key(trans, bkey_i_to_s_c(insert),
- 0, insert->k.size, BCH_BUCKET_MARK_INSERT);
+ ret = bch2_trans_start_alloc_update(trans, &iter, &ptr, &u);
if (ret)
return ret;
- while ((_k = bch2_btree_node_iter_peek_filter(&node_iter, b,
- KEY_TYPE_discard))) {
- struct bkey unpacked;
- struct bkey_s_c k;
- unsigned offset = 0;
- s64 sectors = 0;
- unsigned flags = BCH_BUCKET_MARK_OVERWRITE;
+ if (u.data_type && u.data_type != type) {
+ bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
+ "bucket %llu:%llu gen %u different types of data in same bucket: %s, %s\n"
+ "while marking %s",
+ iter.pos.inode, iter.pos.offset, u.gen,
+ bch2_data_types[u.data_type],
+ bch2_data_types[type],
+ bch2_data_types[type]);
+ ret = -EIO;
+ goto out;
+ }
- k = bkey_disassemble(b, _k, &unpacked);
+ u.data_type = type;
+ u.dirty_sectors = sectors;
- if (btree_node_is_extents(b)
- ? bkey_cmp(insert->k.p, bkey_start_pos(k.k)) <= 0
- : bkey_cmp(insert->k.p, k.k->p))
- break;
+ ret = bch2_alloc_write(trans, &iter, &u, 0);
+ if (ret)
+ goto out;
+out:
+ bch2_trans_iter_exit(trans, &iter);
+ return ret;
+}
- if (btree_node_is_extents(b)) {
- switch (bch2_extent_overlap(&insert->k, k.k)) {
- case BCH_EXTENT_OVERLAP_ALL:
- offset = 0;
- sectors = -((s64) k.k->size);
- break;
- case BCH_EXTENT_OVERLAP_BACK:
- offset = bkey_start_offset(&insert->k) -
- bkey_start_offset(k.k);
- sectors = bkey_start_offset(&insert->k) -
- k.k->p.offset;
- break;
- case BCH_EXTENT_OVERLAP_FRONT:
- offset = 0;
- sectors = bkey_start_offset(k.k) -
- insert->k.p.offset;
- break;
- case BCH_EXTENT_OVERLAP_MIDDLE:
- offset = bkey_start_offset(&insert->k) -
- bkey_start_offset(k.k);
- sectors = -((s64) insert->k.size);
- flags |= BCH_BUCKET_MARK_OVERWRITE_SPLIT;
- break;
- }
+int bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
+ struct bch_dev *ca, size_t b,
+ enum bch_data_type type,
+ unsigned sectors)
+{
+ return __bch2_trans_do(trans, NULL, NULL, 0,
+ __bch2_trans_mark_metadata_bucket(trans, ca, b, type, sectors));
+}
+
+static int bch2_trans_mark_metadata_sectors(struct btree_trans *trans,
+ struct bch_dev *ca,
+ u64 start, u64 end,
+ enum bch_data_type type,
+ u64 *bucket, unsigned *bucket_sectors)
+{
+ do {
+ u64 b = sector_to_bucket(ca, start);
+ unsigned sectors =
+ min_t(u64, bucket_to_sector(ca, b + 1), end) - start;
+
+ if (b != *bucket && *bucket_sectors) {
+ int ret = bch2_trans_mark_metadata_bucket(trans, ca, *bucket,
+ type, *bucket_sectors);
+ if (ret)
+ return ret;
+
+ *bucket_sectors = 0;
+ }
+
+ *bucket = b;
+ *bucket_sectors += sectors;
+ start += sectors;
+ } while (start < end);
+
+ return 0;
+}
+
+static int __bch2_trans_mark_dev_sb(struct btree_trans *trans,
+ struct bch_dev *ca)
+{
+ struct bch_sb_layout *layout = &ca->disk_sb.sb->layout;
+ u64 bucket = 0;
+ unsigned i, bucket_sectors = 0;
+ int ret;
+
+ for (i = 0; i < layout->nr_superblocks; i++) {
+ u64 offset = le64_to_cpu(layout->sb_offset[i]);
- BUG_ON(sectors >= 0);
+ if (offset == BCH_SB_SECTOR) {
+ ret = bch2_trans_mark_metadata_sectors(trans, ca,
+ 0, BCH_SB_SECTOR,
+ BCH_DATA_sb, &bucket, &bucket_sectors);
+ if (ret)
+ return ret;
}
- ret = bch2_trans_mark_key(trans, k, offset, sectors, flags);
+ ret = bch2_trans_mark_metadata_sectors(trans, ca, offset,
+ offset + (1 << layout->sb_max_size_bits),
+ BCH_DATA_sb, &bucket, &bucket_sectors);
+ if (ret)
+ return ret;
+ }
+
+ if (bucket_sectors) {
+ ret = bch2_trans_mark_metadata_bucket(trans, ca,
+ bucket, BCH_DATA_sb, bucket_sectors);
if (ret)
return ret;
+ }
- bch2_btree_node_iter_advance(&node_iter, b);
+ for (i = 0; i < ca->journal.nr; i++) {
+ ret = bch2_trans_mark_metadata_bucket(trans, ca,
+ ca->journal.buckets[i],
+ BCH_DATA_journal, ca->mi.bucket_size);
+ if (ret)
+ return ret;
}
return 0;
}
-/* Disk reservations: */
-
-static u64 bch2_recalc_sectors_available(struct bch_fs *c)
+int bch2_trans_mark_dev_sb(struct bch_fs *c, struct bch_dev *ca)
{
- percpu_u64_set(&c->pcpu->sectors_available, 0);
-
- return avail_factor(__bch2_fs_usage_read_short(c).free);
+ return bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
+ __bch2_trans_mark_dev_sb(&trans, ca));
}
-void __bch2_disk_reservation_put(struct bch_fs *c, struct disk_reservation *res)
-{
- percpu_down_read(&c->mark_lock);
- this_cpu_sub(c->usage[0]->online_reserved,
- res->sectors);
- percpu_up_read(&c->mark_lock);
-
- res->sectors = 0;
-}
+/* Disk reservations: */
#define SECTORS_CACHE 1024
int bch2_disk_reservation_add(struct bch_fs *c, struct disk_reservation *res,
- unsigned sectors, int flags)
+ u64 sectors, int flags)
{
struct bch_fs_pcpu *pcpu;
u64 old, v, get;
if (get < sectors) {
preempt_enable();
- percpu_up_read(&c->mark_lock);
goto recalculate;
}
} while ((v = atomic64_cmpxchg(&c->sectors_available,
out:
pcpu->sectors_available -= sectors;
- this_cpu_add(c->usage[0]->online_reserved, sectors);
+ this_cpu_add(*c->online_reserved, sectors);
res->sectors += sectors;
preempt_enable();
return 0;
recalculate:
- percpu_down_write(&c->mark_lock);
+ mutex_lock(&c->sectors_available_lock);
- sectors_available = bch2_recalc_sectors_available(c);
+ percpu_u64_set(&c->pcpu->sectors_available, 0);
+ sectors_available = avail_factor(__bch2_fs_usage_read_short(c).free);
if (sectors <= sectors_available ||
(flags & BCH_DISK_RESERVATION_NOFAIL)) {
atomic64_set(&c->sectors_available,
max_t(s64, 0, sectors_available - sectors));
- this_cpu_add(c->usage[0]->online_reserved, sectors);
+ this_cpu_add(*c->online_reserved, sectors);
res->sectors += sectors;
ret = 0;
} else {
ret = -ENOSPC;
}
- percpu_up_write(&c->mark_lock);
+ mutex_unlock(&c->sectors_available_lock);
+ percpu_up_read(&c->mark_lock);
return ret;
}
buckets->nbuckets * sizeof(struct bucket));
}
+static void bucket_gens_free_rcu(struct rcu_head *rcu)
+{
+ struct bucket_gens *buckets =
+ container_of(rcu, struct bucket_gens, rcu);
+
+ kvpfree(buckets, sizeof(struct bucket_array) + buckets->nbuckets);
+}
+
int bch2_dev_buckets_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
{
struct bucket_array *buckets = NULL, *old_buckets = NULL;
+ struct bucket_gens *bucket_gens = NULL, *old_bucket_gens = NULL;
unsigned long *buckets_nouse = NULL;
- unsigned long *buckets_written = NULL;
alloc_fifo free[RESERVE_NR];
alloc_fifo free_inc;
alloc_heap alloc_heap;
- copygc_heap copygc_heap;
size_t btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
- ca->mi.bucket_size / c->opts.btree_node_size);
+ ca->mi.bucket_size / btree_sectors(c));
/* XXX: these should be tunable */
size_t reserve_none = max_t(size_t, 1, nbuckets >> 9);
- size_t copygc_reserve = max_t(size_t, 2, nbuckets >> 7);
+ size_t copygc_reserve = max_t(size_t, 2, nbuckets >> 6);
size_t free_inc_nr = max(max_t(size_t, 1, nbuckets >> 12),
btree_reserve * 2);
- bool resize = ca->buckets[0] != NULL,
- start_copygc = ca->copygc_thread != NULL;
+ bool resize = ca->buckets[0] != NULL;
int ret = -ENOMEM;
unsigned i;
memset(&free, 0, sizeof(free));
memset(&free_inc, 0, sizeof(free_inc));
memset(&alloc_heap, 0, sizeof(alloc_heap));
- memset(©gc_heap, 0, sizeof(copygc_heap));
if (!(buckets = kvpmalloc(sizeof(struct bucket_array) +
nbuckets * sizeof(struct bucket),
GFP_KERNEL|__GFP_ZERO)) ||
- !(buckets_nouse = kvpmalloc(BITS_TO_LONGS(nbuckets) *
- sizeof(unsigned long),
+ !(bucket_gens = kvpmalloc(sizeof(struct bucket_gens) + nbuckets,
GFP_KERNEL|__GFP_ZERO)) ||
- !(buckets_written = kvpmalloc(BITS_TO_LONGS(nbuckets) *
+ !(buckets_nouse = kvpmalloc(BITS_TO_LONGS(nbuckets) *
sizeof(unsigned long),
GFP_KERNEL|__GFP_ZERO)) ||
- !init_fifo(&free[RESERVE_BTREE], btree_reserve, GFP_KERNEL) ||
!init_fifo(&free[RESERVE_MOVINGGC],
copygc_reserve, GFP_KERNEL) ||
!init_fifo(&free[RESERVE_NONE], reserve_none, GFP_KERNEL) ||
!init_fifo(&free_inc, free_inc_nr, GFP_KERNEL) ||
- !init_heap(&alloc_heap, ALLOC_SCAN_BATCH(ca) << 1, GFP_KERNEL) ||
- !init_heap(©gc_heap, copygc_reserve, GFP_KERNEL))
+ !init_heap(&alloc_heap, ALLOC_SCAN_BATCH(ca) << 1, GFP_KERNEL))
goto err;
buckets->first_bucket = ca->mi.first_bucket;
buckets->nbuckets = nbuckets;
+ bucket_gens->first_bucket = ca->mi.first_bucket;
+ bucket_gens->nbuckets = nbuckets;
- bch2_copygc_stop(ca);
+ bch2_copygc_stop(c);
if (resize) {
down_write(&c->gc_lock);
}
old_buckets = bucket_array(ca);
+ old_bucket_gens = rcu_dereference_protected(ca->bucket_gens, 1);
if (resize) {
size_t n = min(buckets->nbuckets, old_buckets->nbuckets);
memcpy(buckets->b,
old_buckets->b,
n * sizeof(struct bucket));
+ memcpy(bucket_gens->b,
+ old_bucket_gens->b,
+ n);
memcpy(buckets_nouse,
ca->buckets_nouse,
BITS_TO_LONGS(n) * sizeof(unsigned long));
- memcpy(buckets_written,
- ca->buckets_written,
- BITS_TO_LONGS(n) * sizeof(unsigned long));
}
rcu_assign_pointer(ca->buckets[0], buckets);
- buckets = old_buckets;
+ rcu_assign_pointer(ca->bucket_gens, bucket_gens);
+ buckets = old_buckets;
+ bucket_gens = old_bucket_gens;
swap(ca->buckets_nouse, buckets_nouse);
- swap(ca->buckets_written, buckets_written);
- if (resize)
+ if (resize) {
percpu_up_write(&c->mark_lock);
+ up_write(&c->gc_lock);
+ }
spin_lock(&c->freelist_lock);
for (i = 0; i < RESERVE_NR; i++) {
/* with gc lock held, alloc_heap can't be in use: */
swap(ca->alloc_heap, alloc_heap);
- /* and we shut down copygc: */
- swap(ca->copygc_heap, copygc_heap);
-
nbuckets = ca->mi.nbuckets;
- if (resize) {
+ if (resize)
up_write(&ca->bucket_lock);
- up_write(&c->gc_lock);
- }
-
- if (start_copygc &&
- bch2_copygc_start(c, ca))
- bch_err(ca, "error restarting copygc thread");
ret = 0;
err:
- free_heap(©gc_heap);
free_heap(&alloc_heap);
free_fifo(&free_inc);
for (i = 0; i < RESERVE_NR; i++)
free_fifo(&free[i]);
kvpfree(buckets_nouse,
BITS_TO_LONGS(nbuckets) * sizeof(unsigned long));
- kvpfree(buckets_written,
- BITS_TO_LONGS(nbuckets) * sizeof(unsigned long));
+ if (bucket_gens)
+ call_rcu(&old_buckets->rcu, bucket_gens_free_rcu);
if (buckets)
call_rcu(&old_buckets->rcu, buckets_free_rcu);
{
unsigned i;
- free_heap(&ca->copygc_heap);
free_heap(&ca->alloc_heap);
free_fifo(&ca->free_inc);
for (i = 0; i < RESERVE_NR; i++)
free_fifo(&ca->free[i]);
- kvpfree(ca->buckets_written,
- BITS_TO_LONGS(ca->mi.nbuckets) * sizeof(unsigned long));
kvpfree(ca->buckets_nouse,
BITS_TO_LONGS(ca->mi.nbuckets) * sizeof(unsigned long));
kvpfree(rcu_dereference_protected(ca->buckets[0], 1),
sizeof(struct bucket_array) +
ca->mi.nbuckets * sizeof(struct bucket));
- free_percpu(ca->usage[0]);
+ for (i = 0; i < ARRAY_SIZE(ca->usage); i++)
+ free_percpu(ca->usage[i]);
+ kfree(ca->usage_base);
}
int bch2_dev_buckets_alloc(struct bch_fs *c, struct bch_dev *ca)
{
- if (!(ca->usage[0] = alloc_percpu(struct bch_dev_usage)))
+ unsigned i;
+
+ ca->usage_base = kzalloc(sizeof(struct bch_dev_usage), GFP_KERNEL);
+ if (!ca->usage_base)
return -ENOMEM;
+ for (i = 0; i < ARRAY_SIZE(ca->usage); i++) {
+ ca->usage[i] = alloc_percpu(struct bch_dev_usage);
+ if (!ca->usage[i])
+ return -ENOMEM;
+ }
+
return bch2_dev_buckets_resize(c, ca, ca->mi.nbuckets);;
}