* 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>
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;
unsigned journal_seq,
bool gc)
{
+ BUG_ON(!gc && !journal_seq);
+
return this_cpu_ptr(gc
? ca->usage_gc
: ca->usage[journal_seq & JOURNAL_BUF_MASK]);
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 & JOURNAL_BUF_MASK]);
percpu_down_read(&c->mark_lock);
ret = kmalloc(sizeof(struct bch_fs_usage_online) +
- sizeof(u64) + c->replicas.nr, GFP_NOFS);
+ sizeof(u64) * c->replicas.nr, GFP_NOFS);
if (unlikely(!ret)) {
percpu_up_read(&c->mark_lock);
return NULL;
}
}
-#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)
-{
- return div_u64(r << RESERVE_FACTOR, (1 << RESERVE_FACTOR) + 1);
-}
-
u64 bch2_fs_sectors_used(struct bch_fs *c, struct bch_fs_usage_online *fs_usage)
{
return min(fs_usage->u.hidden +
: m.data_type;
}
-static bool bucket_became_unavailable(struct bucket_mark old,
- struct bucket_mark new)
-{
- return is_available_bucket(old) &&
- !is_available_bucket(new);
-}
-
static inline void account_bucket(struct bch_fs_usage *fs_usage,
struct bch_dev_usage *dev_usage,
enum bch_data_type type,
}
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,
u64 journal_seq, bool gc)
{
+ struct bch_fs_usage *fs_usage;
struct bch_dev_usage *u;
- percpu_rwsem_assert_held(&c->mark_lock);
-
preempt_disable();
- if (!fs_usage)
- fs_usage = fs_usage_ptr(c, journal_seq, gc);
+ fs_usage = fs_usage_ptr(c, journal_seq, gc);
u = dev_usage_ptr(ca, journal_seq, gc);
if (bucket_type(old))
bch2_wake_allocator(ca);
}
-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 bch_fs_usage *fs_usage,
+ struct bch_replicas_entry *r,
+ s64 sectors)
{
int idx = bch2_replicas_entry_idx(c, r);
- BUG_ON(idx < 0);
+ if (idx < 0)
+ return -1;
+
+ fs_usage_data_type_to_base(fs_usage, r->data_type, sectors);
+ fs_usage->replicas[idx] += sectors;
+ return 0;
+}
+
+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)
+{
+ struct bch_fs_usage __percpu *fs_usage;
+ int idx, ret = 0;
+ char buf[200];
+
+ percpu_down_read(&c->mark_lock);
+
+ 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;
update_replicas_list(trans, &r.e, sectors);
}
-#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 & BTREE_TRIGGER_GC) || \
- (gc && gc_visited(c, pos))) \
- ret = fn(c, __VA_ARGS__, gc); \
- ret; \
-})
-
-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 bucket *g = __bucket(ca, b, gc);
+ struct bucket *g = bucket(ca, b);
struct bucket_mark old, 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,
+static int bch2_mark_alloc(struct btree_trans *trans,
struct bkey_s_c old, struct bkey_s_c new,
- struct bch_fs_usage *fs_usage,
- u64 journal_seq, unsigned flags)
+ unsigned flags)
{
bool gc = flags & BTREE_TRIGGER_GC;
- struct bkey_alloc_unpacked u;
+ 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, m;
-
- /* We don't do anything for deletions - do we?: */
- if (new.k->type != KEY_TYPE_alloc &&
- new.k->type != KEY_TYPE_alloc_v2)
- return 0;
+ int ret = 0;
/*
* alloc btree is read in by bch2_alloc_read, not gc:
!(flags & BTREE_TRIGGER_BUCKET_INVALIDATE))
return 0;
+ 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 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;
+ 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);
- u = bch2_alloc_unpack(new);
old_m = 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;
- m.stripe = u.stripe != 0;
+ 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;
}
}));
- bch2_dev_usage_update(c, ca, fs_usage, old_m, m, journal_seq, 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 = u.stripe;
- g->stripe_redundancy = u.stripe_redundancy;
+ 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
if ((flags & BTREE_TRIGGER_BUCKET_INVALIDATE) &&
old_m.cached_sectors) {
- update_cached_sectors(c, fs_usage, ca->dev_idx,
- -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);
}
overflow; \
})
-static int __bch2_mark_metadata_bucket(struct bch_fs *c, struct bch_dev *ca,
- size_t b, enum bch_data_type data_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(!(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.data_type = data_type;
overflow = checked_add(new.dirty_sectors, sectors);
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, 0, gc);
-
- return 0;
-}
-
-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)
-{
- 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);
- }
-
- preempt_enable();
-}
-
-static s64 disk_sectors_scaled(unsigned n, unsigned d, unsigned sectors)
-{
- return DIV_ROUND_UP(sectors * n, d);
+ bch2_dev_usage_update(c, ca, old, new, 0, true);
+ percpu_up_read(&c->mark_lock);
}
-static s64 __ptr_disk_sectors_delta(unsigned old_size,
- unsigned offset, s64 delta,
- unsigned flags,
- unsigned n, unsigned d)
+static s64 ptr_disk_sectors(s64 sectors, struct extent_ptr_decoded p)
{
- BUG_ON(!n || !d);
-
- if (flags & BTREE_TRIGGER_OVERWRITE_SPLIT) {
- BUG_ON(offset + -delta > old_size);
-
- return -disk_sectors_scaled(n, d, old_size) +
- disk_sectors_scaled(n, d, offset) +
- disk_sectors_scaled(n, d, old_size - offset + delta);
- } else if (flags & BTREE_TRIGGER_OVERWRITE) {
- BUG_ON(offset + -delta > old_size);
+ EBUG_ON(sectors < 0);
- return -disk_sectors_scaled(n, d, old_size) +
- disk_sectors_scaled(n, d, old_size + delta);
- } else {
- return disk_sectors_scaled(n, d, delta);
- }
-}
-
-static s64 ptr_disk_sectors_delta(struct extent_ptr_decoded p,
- unsigned offset, s64 delta,
- unsigned flags)
-{
- return __ptr_disk_sectors_delta(p.crc.live_size,
- offset, delta, flags,
- p.crc.compressed_size,
- p.crc.uncompressed_size);
+ 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 int check_bucket_ref(struct bch_fs *c, struct bkey_s_c k,
+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,
return 0;
}
-static int mark_stripe_bucket(struct bch_fs *c, struct bkey_s_c k,
- unsigned ptr_idx,
- struct bch_fs_usage *fs_usage,
- u64 journal_seq, unsigned flags)
+static int mark_stripe_bucket(struct btree_trans *trans,
+ struct bkey_s_c k,
+ unsigned ptr_idx,
+ u64 journal_seq, unsigned flags)
{
+ 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;
- bool gc = flags & BTREE_TRIGGER_GC;
struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
- struct bucket *g = PTR_BUCKET(ca, ptr, gc);
+ struct bucket *g;
struct bucket_mark new, old;
char buf[200];
- int ret;
+ int ret = 0;
+
+ BUG_ON(!(flags & BTREE_TRIGGER_GC));
- if (g->stripe && g->stripe != k.k->p.offset) {
+ /* * XXX doesn't handle deletion */
+
+ 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));
- return -EINVAL;
+ ret = -EINVAL;
+ goto err;
}
old = bucket_cmpxchg(g, new, ({
- ret = check_bucket_ref(c, k, ptr, 0, 0, new.gen, new.data_type,
+ ret = check_bucket_ref(c, k, ptr, sectors, data_type,
+ new.gen, new.data_type,
new.dirty_sectors, new.cached_sectors);
if (ret)
- return ret;
+ goto err;
- if (parity) {
- new.data_type = BCH_DATA_parity;
- new.dirty_sectors = le16_to_cpu(s->sectors);
- }
+ 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;
}
+
+ new.stripe = true;
}));
g->stripe = k.k->p.offset;
g->stripe_redundancy = s->nr_redundant;
- bch2_dev_usage_update(c, ca, fs_usage, old, new, journal_seq, gc);
+ 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 bch_fs *c, struct bkey_s_c k,
+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 *dst_sectors = !ptr->cached
? dirty_sectors
: cached_sectors;
- int ret = check_bucket_ref(c, k, ptr, sectors, ptr_data_type,
+ int ret = check_bucket_ref(trans->c, k, ptr, sectors, ptr_data_type,
bucket_gen, *bucket_data_type,
*dirty_sectors, *cached_sectors);
return 0;
}
-static int bch2_mark_pointer(struct bch_fs *c, struct bkey_s_c k,
+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,
- struct bch_fs_usage *fs_usage,
- u64 journal_seq, unsigned flags)
+ unsigned flags)
{
- bool gc = flags & BTREE_TRIGGER_GC;
+ 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 = PTR_BUCKET(ca, &p.ptr, gc);
+ struct bucket *g;
u8 bucket_data_type;
u64 v;
- int ret;
+ 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(c, k, &p.ptr, sectors, data_type, new.gen,
+ ret = __mark_pointer(trans, k, &p.ptr, sectors,
+ data_type, new.gen,
&bucket_data_type,
&new.dirty_sectors,
&new.cached_sectors);
if (ret)
- return ret;
+ goto err;
new.data_type = bucket_data_type;
old.v.counter,
new.v.counter)) != old.v.counter);
- bch2_dev_usage_update(c, ca, fs_usage, old, new, journal_seq, 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 0;
+ 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 & BTREE_TRIGGER_GC;
+ struct bch_fs *c = trans->c;
struct bch_replicas_padded r;
- struct stripe *m;
- unsigned i, blocks_nonempty = 0;
+ struct gc_stripe *m;
- m = genradix_ptr(&c->stripes[gc], p.idx);
+ BUG_ON(!(flags & BTREE_TRIGGER_GC));
+
+ 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;
}
m->block_sectors[p.block] += sectors;
r = m->r;
-
- for (i = 0; i < m->nr_blocks; i++)
- blocks_nonempty += m->block_sectors[i] != 0;
-
- if (m->blocks_nonempty != blocks_nonempty) {
- m->blocks_nonempty = blocks_nonempty;
- if (!gc)
- bch2_stripes_heap_update(c, m, p.idx);
- }
-
spin_unlock(&c->ec_stripes_heap_lock);
r.e.data_type = data_type;
- update_replicas(c, fs_usage, &r.e, sectors);
+ update_replicas(c, k, &r.e, sectors, trans->journal_res.seq, true);
return 0;
}
-static int bch2_mark_extent(struct bch_fs *c,
+static int bch2_mark_extent(struct btree_trans *trans,
struct bkey_s_c old, struct bkey_s_c new,
- unsigned offset, s64 sectors,
- enum bch_data_type data_type,
- struct bch_fs_usage *fs_usage,
- unsigned journal_seq, unsigned flags)
+ unsigned flags)
{
- struct bkey_s_c k = flags & BTREE_TRIGGER_INSERT ? new : old;
+ 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;
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);
+ s64 disk_sectors = ptr_disk_sectors(sectors, p);
- ret = bch2_mark_pointer(c, k, p, disk_sectors, data_type,
- fs_usage, journal_seq, flags);
+ 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 (!stale)
- update_cached_sectors(c, fs_usage, p.ptr.dev,
- disk_sectors);
+ 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 {
- ret = bch2_mark_stripe_ptr(c, p.ec, data_type,
- fs_usage, disk_sectors, flags);
+ ret = bch2_mark_stripe_ptr(trans, k, p.ec, data_type,
+ disk_sectors, flags);
if (ret)
return ret;
}
}
- if (r.e.nr_devs)
- 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,
+static int bch2_mark_stripe(struct btree_trans *trans,
struct bkey_s_c old, struct bkey_s_c new,
- struct bch_fs_usage *fs_usage,
- u64 journal_seq, unsigned flags)
+ unsigned flags)
{
bool gc = flags & BTREE_TRIGGER_GC;
- size_t idx = new.k->p.offset;
+ 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;
- struct stripe *m = genradix_ptr(&c->stripes[gc], idx);
unsigned i;
int ret;
BUG_ON(gc && old_s);
- if (!m || (old_s && !m->alive)) {
- bch_err_ratelimited(c, "error marking nonexistent stripe %zu",
- idx);
- return -1;
- }
+ if (!gc) {
+ struct stripe *m = genradix_ptr(&c->stripes, idx);
- 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 (!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);
+
+ 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);
- memset(m, 0, sizeof(*m));
+ 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;
+ }
+ /*
+ * This will be wrong when we bring back runtime gc: we should
+ * be unmarking the old key and then marking the new key
+ */
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->block_sectors[i] =
- stripe_blockcount_get(new_s, i);
- m->blocks_nonempty += !!m->block_sectors[i];
+ for (i = 0; i < new_s->nr_blocks; i++)
m->ptrs[i] = new_s->ptrs[i];
- }
bch2_bkey_to_replicas(&m->r.e, new);
- if (!gc) {
- spin_lock(&c->ec_stripes_heap_lock);
- bch2_stripes_heap_update(c, m, idx);
- spin_unlock(&c->ec_stripes_heap_lock);
- }
- }
-
- if (gc) {
/*
* gc recalculates this field from stripe ptr
* references:
*/
memset(m->block_sectors, 0, sizeof(m->block_sectors));
- m->blocks_nonempty = 0;
for (i = 0; i < new_s->nr_blocks; i++) {
- ret = mark_stripe_bucket(c, new, i, fs_usage,
- journal_seq, flags);
+ ret = mark_stripe_bucket(trans, new, i, journal_seq, flags);
if (ret)
return ret;
}
- update_replicas(c, fs_usage, &m->r.e,
- ((s64) m->sectors * m->nr_redundant));
+ ret = update_replicas(c, new, &m->r.e,
+ ((s64) m->sectors * m->nr_redundant),
+ journal_seq, gc);
+ if (ret) {
+ char buf[200];
+
+ bch2_bkey_val_to_text(&PBUF(buf), c, new);
+ bch2_fs_fatal_error(c, "no replicas entry for %s", buf);
+ return ret;
+ }
}
return 0;
}
-static int bch2_mark_key_locked(struct bch_fs *c,
- struct bkey_s_c old,
- struct bkey_s_c new,
- 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)
{
- struct bkey_s_c k = flags & BTREE_TRIGGER_INSERT ? new : old;
- int ret = 0;
-
- BUG_ON(!(flags & (BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE)));
+ 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 & BTREE_TRIGGER_GC))
- fs_usage = fs_usage_ptr(c, journal_seq,
- flags & BTREE_TRIGGER_GC);
+ BUG_ON(!journal_seq);
+ BUG_ON(new.k->type != KEY_TYPE_inode_v2);
- switch (k.k->type) {
- case KEY_TYPE_alloc:
- case KEY_TYPE_alloc_v2:
- ret = bch2_mark_alloc(c, old, new, fs_usage, journal_seq, flags);
- break;
- case KEY_TYPE_btree_ptr:
- case KEY_TYPE_btree_ptr_v2:
- sectors = !(flags & BTREE_TRIGGER_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, old, new, 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, old, new, offset, sectors,
- BCH_DATA_user, fs_usage, journal_seq, flags);
- break;
- case KEY_TYPE_stripe:
- ret = bch2_mark_stripe(c, old, new, fs_usage, journal_seq, flags);
- break;
- case KEY_TYPE_inode:
- fs_usage->nr_inodes += new.k->type == KEY_TYPE_inode;
- fs_usage->nr_inodes -= old.k->type == KEY_TYPE_inode;
- 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);
}
-
- preempt_enable();
-
- return ret;
+ return 0;
}
-int bch2_mark_key(struct bch_fs *c, struct bkey_s_c new,
- unsigned offset, s64 sectors,
- struct bch_fs_usage *fs_usage,
- u64 journal_seq, unsigned flags)
+static int bch2_mark_reservation(struct btree_trans *trans,
+ struct bkey_s_c old, struct bkey_s_c new,
+ unsigned flags)
{
- struct bkey deleted;
- struct bkey_s_c old = (struct bkey_s_c) { &deleted, NULL };
- int ret;
+ 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;
- bkey_init(&deleted);
+ BUG_ON(!(flags & BTREE_TRIGGER_GC));
+
+ if (flags & BTREE_TRIGGER_OVERWRITE)
+ sectors = -sectors;
+ sectors *= replicas;
percpu_down_read(&c->mark_lock);
- ret = bch2_mark_key_locked(c, old, new, offset, sectors,
- fs_usage, journal_seq,
- BTREE_TRIGGER_INSERT|flags);
+ 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_update(struct btree_trans *trans,
- struct btree_iter *iter,
- struct bkey_i *new,
- struct bch_fs_usage *fs_usage,
- 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)
{
- struct bch_fs *c = trans->c;
- struct bkey_s_c old;
- struct bkey unpacked;
- int ret = 0;
+ struct reflink_gc *r;
+ int add = !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1;
+ s64 ret = 0;
- if (unlikely(flags & BTREE_TRIGGER_NORUN))
- return 0;
+ if (r_idx >= c->reflink_gc_nr)
+ goto not_found;
- if (!btree_node_type_needs_gc(iter->btree_id))
- return 0;
+ r = genradix_ptr(&c->reflink_gc_table, r_idx);
+ if (*idx < r->offset - r->size)
+ goto not_found;
- bkey_init(&unpacked);
- old = (struct bkey_s_c) { &unpacked, NULL };
+ BUG_ON((s64) r->refcount + add < 0);
- if (!btree_node_type_is_extents(iter->btree_id)) {
- /* iterators should be uptodate, shouldn't get errors here: */
- if (btree_iter_type(iter) != BTREE_ITER_CACHED) {
- old = bch2_btree_iter_peek_slot(iter);
- BUG_ON(bkey_err(old));
- } else {
- struct bkey_cached *ck = (void *) iter->l[0].b;
+ r->refcount += add;
+ *idx = r->offset;
+ return 0;
+not_found:
+ *idx = U64_MAX;
+ ret = -EIO;
- if (ck->valid)
- old = bkey_i_to_s_c(ck->k);
- }
+ /*
+ * 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;
+}
- if (old.k->type == new->k.type) {
- bch2_mark_key_locked(c, old, bkey_i_to_s_c(new), 0, 0,
- fs_usage, trans->journal_res.seq,
- BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE|flags);
+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 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;
- } else {
- bch2_mark_key_locked(c, old, bkey_i_to_s_c(new), 0, 0,
- fs_usage, trans->journal_res.seq,
- BTREE_TRIGGER_INSERT|flags);
- bch2_mark_key_locked(c, old, bkey_i_to_s_c(new), 0, 0,
- fs_usage, trans->journal_res.seq,
- BTREE_TRIGGER_OVERWRITE|flags);
- }
- } else {
- struct btree_iter *copy;
+ BUG_ON(!(flags & BTREE_TRIGGER_GC));
- BUG_ON(btree_iter_type(iter) == BTREE_ITER_CACHED);
- bch2_mark_key_locked(c, old, bkey_i_to_s_c(new),
- 0, new->k.size,
- fs_usage, trans->journal_res.seq,
- BTREE_TRIGGER_INSERT|flags);
+ 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);
+ }
- copy = bch2_trans_copy_iter(trans, iter);
+ l = 0;
+ r = c->reflink_gc_nr;
+ while (l < r) {
+ m = l + (r - l) / 2;
- for_each_btree_key_continue(copy, 0, old, ret) {
- unsigned offset = 0;
- s64 sectors = -((s64) old.k->size);
+ ref = genradix_ptr(&c->reflink_gc_table, m);
+ if (ref->offset <= idx)
+ l = m + 1;
+ else
+ r = m;
+ }
- flags |= BTREE_TRIGGER_OVERWRITE;
+ while (idx < end && !ret)
+ ret = __bch2_mark_reflink_p(c, p, &idx, flags, l++);
- if (bkey_cmp(new->k.p, bkey_start_pos(old.k)) <= 0)
- break;
+ return ret;
+}
- 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 |= BTREE_TRIGGER_OVERWRITE_SPLIT;
- break;
- }
+int bch2_mark_key(struct btree_trans *trans,
+ struct bkey_s_c old,
+ struct bkey_s_c new,
+ unsigned flags)
+{
+ struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE ? old: new;
- BUG_ON(sectors >= 0);
+ 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;
+ }
+}
- ret = bch2_mark_key_locked(c, old, bkey_i_to_s_c(new),
- offset, sectors, fs_usage,
- trans->journal_res.seq, flags) ?: 1;
- if (ret <= 0)
- break;
- }
- bch2_trans_iter_put(trans, copy);
+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;
+
+ _deleted.p = path->pos;
+
+ if (unlikely(flags & BTREE_TRIGGER_NORUN))
+ return 0;
+
+ if (!btree_node_type_needs_gc(path->btree_id))
+ return 0;
+
+ old = bch2_btree_path_peek_slot(path, &unpacked);
+
+ 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;
static noinline __cold
void fs_usage_apply_warn(struct btree_trans *trans,
- unsigned disk_res_sectors)
+ unsigned disk_res_sectors,
+ s64 should_not_have_added)
{
struct bch_fs *c = trans->c;
struct btree_insert_entry *i;
char buf[200];
- bch_err(c, "disk usage increased more than %u sectors reserved",
- disk_res_sectors);
+ 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");
pr_err("%s", buf);
pr_err("overlapping with");
- if (btree_iter_type(i->iter) != BTREE_ITER_CACHED) {
- struct btree_iter *copy = bch2_trans_copy_iter(trans, i->iter);
- struct bkey_s_c k;
- int ret;
+ if (!i->cached) {
+ struct bkey u;
+ struct bkey_s_c k = bch2_btree_path_peek_slot(i->path, &u);
- for_each_btree_key_continue(copy, 0, k, ret) {
- if (btree_node_type_is_extents(i->iter->btree_id)
- ? bkey_cmp(i->k->k.p, bkey_start_pos(k.k)) <= 0
- : bkey_cmp(i->k->k.p, k.k->p))
- break;
-
- bch2_bkey_val_to_text(&PBUF(buf), c, k);
- pr_err("%s", buf);
- }
- bch2_trans_iter_put(trans, copy);
+ bch2_bkey_val_to_text(&PBUF(buf), c, k);
+ pr_err("%s", buf);
} else {
- struct bkey_cached *ck = (void *) i->iter->l[0].b;
+ struct bkey_cached *ck = (void *) i->path->l[0].b;
if (ck->valid) {
bch2_bkey_val_to_text(&PBUF(buf), c, bkey_i_to_s_c(ck->k));
}
}
}
+ __WARN();
}
-void bch2_trans_fs_usage_apply(struct btree_trans *trans,
- struct replicas_delta_list *deltas)
+int bch2_trans_fs_usage_apply(struct btree_trans *trans,
+ struct replicas_delta_list *deltas)
{
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;
+ 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;
- percpu_rwsem_assert_held(&c->mark_lock);
-
+ percpu_down_read(&c->mark_lock);
preempt_disable();
dst = fs_usage_ptr(c, trans->journal_res.seq, false);
added += d->delta;
}
- update_replicas(c, dst, &d->r, d->delta);
+ if (__update_replicas(c, dst, &d->r, d->delta))
+ goto need_mark;
}
dst->nr_inodes += deltas->nr_inodes;
*/
should_not_have_added = added - (s64) disk_res_sectors;
if (unlikely(should_not_have_added > 0)) {
- atomic64_sub(should_not_have_added, &c->sectors_available);
+ u64 old, new, v = atomic64_read(&c->sectors_available);
+
+ do {
+ old = v;
+ new = max_t(s64, 0, old - should_not_have_added);
+ } while ((v = atomic64_cmpxchg(&c->sectors_available,
+ old, new)) != old);
+
added -= should_not_have_added;
warn = 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);
-}
-
-/* trans_mark: */
-
-static struct btree_iter *trans_get_update(struct btree_trans *trans,
- enum btree_id btree_id, struct bpos pos,
- struct bkey_s_c *k)
-{
- struct btree_insert_entry *i;
-
- trans_for_each_update(trans, i)
- if (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))) {
- *k = bkey_i_to_s_c(i->k);
-
- /* ugly hack.. */
- BUG_ON(btree_iter_live(trans, i->iter));
- trans->iters_live |= 1ULL << i->iter->idx;
- return i->iter;
- }
+ 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 NULL;
+ preempt_enable();
+ percpu_up_read(&c->mark_lock);
+ return -1;
}
-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)
-{
- unsigned flags = btree_id != BTREE_ID_alloc
- ? BTREE_ITER_SLOTS
- : BTREE_ITER_CACHED;
- int ret;
-
- *iter = trans_get_update(trans, btree_id, pos, k);
- if (*iter)
- return 1;
-
- *iter = bch2_trans_get_iter(trans, btree_id, pos,
- flags|BTREE_ITER_INTENT);
- *k = __bch2_btree_iter_peek(*iter, flags);
- ret = bkey_err(*k);
- if (ret)
- bch2_trans_iter_put(trans, *iter);
- return ret;
-}
+/* trans_mark: */
-static struct bkey_alloc_buf *
-bch2_trans_start_alloc_update(struct btree_trans *trans, struct btree_iter **_iter,
+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, ptr->dev);
- struct bpos pos = POS(ptr->dev, PTR_BUCKET_NR(ca, ptr));
- struct bucket *g;
- struct btree_iter *iter;
struct bkey_s_c k;
- struct bkey_alloc_buf *a;
int ret;
- a = bch2_trans_kmalloc(trans, sizeof(struct bkey_alloc_buf));
- if (IS_ERR(a))
- return a;
-
- iter = trans_get_update(trans, BTREE_ID_alloc, pos, &k);
- if (iter) {
- *u = bch2_alloc_unpack(k);
- } else {
- iter = bch2_trans_get_iter(trans, BTREE_ID_alloc, pos,
- BTREE_ITER_CACHED|
- BTREE_ITER_CACHED_NOFILL|
- BTREE_ITER_INTENT);
- ret = bch2_btree_iter_traverse(iter);
- if (ret) {
- bch2_trans_iter_put(trans, iter);
- return ERR_PTR(ret);
- }
-
- percpu_down_read(&c->mark_lock);
- g = bucket(ca, pos.offset);
- *u = alloc_mem_to_key(iter, g, READ_ONCE(g->mark));
- percpu_up_read(&c->mark_lock);
+ 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;
}
- *_iter = iter;
- return a;
+ *u = bch2_alloc_unpack(k);
+ return 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 bch_fs *c = trans->c;
- struct btree_iter *iter;
+ struct btree_iter iter;
struct bkey_alloc_unpacked u;
- struct bkey_alloc_buf *a;
int ret;
- a = bch2_trans_start_alloc_update(trans, &iter, &p.ptr, &u);
- if (IS_ERR(a))
- return PTR_ERR(a);
+ ret = bch2_trans_start_alloc_update(trans, &iter, &p.ptr, &u);
+ if (ret)
+ return ret;
- ret = __mark_pointer(c, k, &p.ptr, sectors, data_type, u.gen, &u.data_type,
+ 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;
- bch2_alloc_pack(c, a, u);
- bch2_trans_update(trans, iter, &a->k, 0);
+ 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;
}
s64 sectors, enum bch_data_type data_type)
{
struct bch_fs *c = trans->c;
- struct btree_iter *iter;
+ struct btree_iter iter;
struct bkey_s_c k;
struct bkey_i_stripe *s;
struct bch_replicas_padded r;
int ret = 0;
- ret = trans_get_key(trans, BTREE_ID_stripes, POS(0, p.ec.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.ec.idx);
+ bch2_inconsistent_error(c);
ret = -EIO;
- goto out;
+ goto err;
}
if (!bch2_ptr_matches_stripe(bkey_s_c_to_stripe(k).v, p)) {
"stripe pointer doesn't match stripe %llu",
(u64) p.ec.idx);
ret = -EIO;
- goto out;
+ goto err;
}
s = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
ret = PTR_ERR_OR_ZERO(s);
if (ret)
- goto out;
+ goto err;
bkey_reassemble(&s->k_i, k);
stripe_blockcount_set(&s->v, p.ec.block,
stripe_blockcount_get(&s->v, p.ec.block) +
sectors);
- bch2_trans_update(trans, iter, &s->k_i, 0);
+
+ 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;
int ret;
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);
- ret = bch2_trans_mark_pointer(trans, k, p, disk_sectors,
- data_type);
+ if (flags & BTREE_TRIGGER_OVERWRITE)
+ disk_sectors = -disk_sectors;
+
+ ret = bch2_trans_mark_pointer(trans, k, p,
+ disk_sectors, data_type);
if (ret < 0)
return ret;
return 0;
}
-static int bch2_trans_mark_stripe_alloc_ref(struct btree_trans *trans,
- struct bkey_s_c_stripe s,
- unsigned idx, bool deleting)
+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 bkey_alloc_buf *a;
- struct btree_iter *iter;
+ struct btree_iter iter;
struct bkey_alloc_unpacked u;
- bool parity = idx >= s.v->nr_blocks - s.v->nr_redundant;
+ 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;
- a = bch2_trans_start_alloc_update(trans, &iter, ptr, &u);
- if (IS_ERR(a))
- return PTR_ERR(a);
+ if (deleting)
+ sectors = -sectors;
- if (parity) {
- s64 sectors = le16_to_cpu(s.v->sectors);
-
- if (deleting)
- sectors = -sectors;
+ ret = bch2_trans_start_alloc_update(trans, &iter, ptr, &u);
+ if (ret)
+ return ret;
- u.dirty_sectors += sectors;
- u.data_type = u.dirty_sectors
- ? BCH_DATA_parity
- : 0;
- }
+ 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 != s.k->p.offset, c,
- "bucket %llu:%llu gen %u: multiple stripes using same bucket (%u, %llu)",
- iter->pos.inode, iter->pos.offset, u.gen,
+ 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;
}
- bch2_alloc_pack(c, a, u);
- bch2_trans_update(trans, iter, &a->k, 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_put(trans, iter);
+ bch2_trans_iter_exit(trans, &iter);
return ret;
}
struct bkey_s_c old, struct bkey_s_c new,
unsigned flags)
{
- struct bkey_s_c_stripe old_s = { NULL };
- struct bkey_s_c_stripe new_s = { NULL };
+ 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;
+ unsigned i, nr_blocks;
int ret = 0;
if (old.k->type == KEY_TYPE_stripe)
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);
-
- for (i = 0; i < new_s.v->nr_blocks; i++) {
- ret = bch2_trans_mark_stripe_alloc_ref(trans, new_s,
- i, false);
- if (ret)
- return ret;
- }
}
if (old_s.k) {
bch2_bkey_to_replicas(&r.e, old);
update_replicas_list(trans, &r.e, sectors * old_s.v->nr_redundant);
+ }
- for (i = 0; i < old_s.v->nr_blocks; i++) {
- ret = bch2_trans_mark_stripe_alloc_ref(trans, old_s,
- i, true);
+ 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)
- return ret;
+ break;
+ }
+
+ if (old_s.k) {
+ ret = bch2_trans_mark_stripe_bucket(trans, old_s, i, true);
+ if (ret)
+ break;
}
}
return ret;
}
-static __le64 *bkey_refcount(struct bkey_i *k)
+static int bch2_trans_mark_inode(struct btree_trans *trans,
+ struct bkey_s_c old,
+ struct bkey_s_c new,
+ unsigned flags)
{
- switch (k->k.type) {
- case KEY_TYPE_reflink_v:
- return &bkey_i_to_reflink_v(k)->v.refcount;
- case KEY_TYPE_indirect_inline_data:
- return &bkey_i_to_indirect_inline_data(k)->v.refcount;
- default:
- return NULL;
+ 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 btree_iter iter;
struct bkey_s_c k;
struct bkey_i *n;
__le64 *refcount;
- s64 ret;
-
- ret = trans_get_key(trans, BTREE_ID_reflink,
- POS(0, idx), &iter, &k);
- if (ret < 0)
- return ret;
-
- if ((flags & BTREE_TRIGGER_OVERWRITE) &&
- (bkey_start_offset(k.k) < idx ||
- k.k->p.offset > idx + sectors))
- goto out;
+ int add = !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1;
+ char buf[200];
+ int ret;
- sectors = k.k->p.offset - idx;
+ 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;
n = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
ret = PTR_ERR_OR_ZERO(n);
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;
}
- le64_add_cpu(refcount, !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1);
+ 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;
+ }
+
+ if (flags & BTREE_TRIGGER_INSERT) {
+ struct bch_reflink_p *v = (struct bch_reflink_p *) p.v;
+ u64 pad;
+
+ 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);
+
+ 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(refcount, add);
if (!*refcount) {
n->k.type = KEY_TYPE_deleted;
set_bkey_val_u64s(&n->k, 0);
}
- bch2_btree_iter_set_pos(iter, bkey_start_pos(k.k));
- bch2_trans_update(trans, iter, n, 0);
-out:
- ret = sectors;
+ 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 old,
- struct bkey_s_c new,
- 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 bch_fs *c = trans->c;
- struct bkey_s_c k = flags & BTREE_TRIGGER_INSERT ? new : old;
- struct replicas_delta_list *d;
-
- BUG_ON(!(flags & (BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE)));
+ struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE ? old: new;
switch (k.k->type) {
case KEY_TYPE_btree_ptr:
case KEY_TYPE_btree_ptr_v2:
- sectors = !(flags & BTREE_TRIGGER_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_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: {
- int nr = (new.k->type == KEY_TYPE_inode) -
- (old.k->type == KEY_TYPE_inode);
-
- if (nr) {
- d = replicas_deltas_realloc(trans, 0);
- d->nr_inodes += nr;
- }
-
- 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->persistent_reserved));
-
- d->persistent_reserved[replicas - 1] += sectors;
- return 0;
- }
+ case KEY_TYPE_inode:
+ 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 *new,
- unsigned flags)
-{
- struct bkey_s_c old;
- int ret;
-
- if (unlikely(flags & BTREE_TRIGGER_NORUN))
- return 0;
-
- if (!btree_node_type_needs_gc(iter->btree_id))
- return 0;
-
- if (!btree_node_type_is_extents(iter->btree_id)) {
- /* iterators should be uptodate, shouldn't get errors here: */
- if (btree_iter_type(iter) != BTREE_ITER_CACHED) {
- old = bch2_btree_iter_peek_slot(iter);
- BUG_ON(bkey_err(old));
- } else {
- struct bkey_cached *ck = (void *) iter->l[0].b;
-
- BUG_ON(!ck->valid);
- old = bkey_i_to_s_c(ck->k);
- }
-
- if (old.k->type == new->k.type) {
- ret = bch2_trans_mark_key(trans, old, bkey_i_to_s_c(new), 0, 0,
- BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE|flags);
- } else {
- ret = bch2_trans_mark_key(trans, old, bkey_i_to_s_c(new), 0, 0,
- BTREE_TRIGGER_INSERT|flags) ?:
- bch2_trans_mark_key(trans, old, bkey_i_to_s_c(new), 0, 0,
- BTREE_TRIGGER_OVERWRITE|flags);
- }
- } else {
- struct btree_iter *copy;
- struct bkey _old;
-
- EBUG_ON(btree_iter_type(iter) == BTREE_ITER_CACHED);
-
- bkey_init(&_old);
- old = (struct bkey_s_c) { &_old, NULL };
-
- ret = bch2_trans_mark_key(trans, old, bkey_i_to_s_c(new),
- 0, new->k.size,
- BTREE_TRIGGER_INSERT);
- if (ret)
- return ret;
-
- copy = bch2_trans_copy_iter(trans, iter);
-
- for_each_btree_key_continue(copy, 0, old, ret) {
- unsigned offset = 0;
- s64 sectors = -((s64) old.k->size);
-
- flags |= BTREE_TRIGGER_OVERWRITE;
-
- if (bkey_cmp(new->k.p, bkey_start_pos(old.k)) <= 0)
- break;
-
- 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 |= BTREE_TRIGGER_OVERWRITE_SPLIT;
- break;
- }
-
- BUG_ON(sectors >= 0);
-
- ret = bch2_trans_mark_key(trans, old, bkey_i_to_s_c(new),
- offset, sectors, flags);
- if (ret)
- break;
- }
- bch2_trans_iter_put(trans, copy);
- }
-
- return ret;
-}
-
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 bch_fs *c = trans->c;
- struct btree_iter *iter;
+ struct btree_iter iter;
struct bkey_alloc_unpacked u;
- struct bkey_alloc_buf *a;
struct bch_extent_ptr ptr = {
.dev = ca->dev_idx,
.offset = bucket_to_sector(ca, b),
};
int ret = 0;
- a = bch2_trans_start_alloc_update(trans, &iter, &ptr, &u);
- if (IS_ERR(a))
- return PTR_ERR(a);
+ /*
+ * Backup superblock might be past the end of our normal usable space:
+ */
+ if (b >= ca->mi.nbuckets)
+ return 0;
+
+ ret = bch2_trans_start_alloc_update(trans, &iter, &ptr, &u);
+ if (ret)
+ return ret;
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,
+ iter.pos.inode, iter.pos.offset, u.gen,
bch2_data_types[u.data_type],
bch2_data_types[type],
bch2_data_types[type]);
goto out;
}
- if ((unsigned) (u.dirty_sectors + sectors) > ca->mi.bucket_size) {
- bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
- "bucket %llu:%llu gen %u data type %s sector count overflow: %u + %u > %u\n"
- "while marking %s",
- iter->pos.inode, iter->pos.offset, u.gen,
- bch2_data_types[u.data_type ?: type],
- u.dirty_sectors, sectors, ca->mi.bucket_size,
- bch2_data_types[type]);
- ret = -EIO;
- goto out;
- }
-
- if (u.data_type == type &&
- u.dirty_sectors == sectors)
- goto out;
-
u.data_type = type;
u.dirty_sectors = sectors;
- bch2_alloc_pack(c, a, u);
- bch2_trans_update(trans, iter, &a->k, 0);
+ 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;
}
int bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
- struct disk_reservation *res,
struct bch_dev *ca, size_t b,
enum bch_data_type type,
unsigned sectors)
{
- return __bch2_trans_do(trans, res, NULL, 0,
- __bch2_trans_mark_metadata_bucket(trans, ca, b, BCH_DATA_journal,
- ca->mi.bucket_size));
-
+ 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 disk_reservation *res,
struct bch_dev *ca,
u64 start, u64 end,
enum bch_data_type type,
u64 *bucket, unsigned *bucket_sectors)
{
- int ret;
-
do {
u64 b = sector_to_bucket(ca, start);
unsigned sectors =
min_t(u64, bucket_to_sector(ca, b + 1), end) - start;
- if (b != *bucket) {
- if (*bucket_sectors) {
- ret = bch2_trans_mark_metadata_bucket(trans, res, ca,
- *bucket, type, *bucket_sectors);
- if (ret)
- return ret;
- }
+ if (b != *bucket && *bucket_sectors) {
+ int ret = bch2_trans_mark_metadata_bucket(trans, ca, *bucket,
+ type, *bucket_sectors);
+ if (ret)
+ return ret;
- *bucket = b;
- *bucket_sectors = 0;
+ *bucket_sectors = 0;
}
+ *bucket = b;
*bucket_sectors += sectors;
start += sectors;
- } while (!ret && start < end);
+ } while (start < end);
return 0;
}
static int __bch2_trans_mark_dev_sb(struct btree_trans *trans,
- struct disk_reservation *res,
- struct bch_dev *ca)
+ struct bch_dev *ca)
{
struct bch_sb_layout *layout = &ca->disk_sb.sb->layout;
u64 bucket = 0;
u64 offset = le64_to_cpu(layout->sb_offset[i]);
if (offset == BCH_SB_SECTOR) {
- ret = bch2_trans_mark_metadata_sectors(trans, res, ca,
+ 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_metadata_sectors(trans, res, ca, offset,
+ ret = bch2_trans_mark_metadata_sectors(trans, ca, offset,
offset + (1 << layout->sb_max_size_bits),
BCH_DATA_sb, &bucket, &bucket_sectors);
if (ret)
}
if (bucket_sectors) {
- ret = bch2_trans_mark_metadata_bucket(trans, res, ca,
+ ret = bch2_trans_mark_metadata_bucket(trans, ca,
bucket, BCH_DATA_sb, bucket_sectors);
if (ret)
return ret;
}
for (i = 0; i < ca->journal.nr; i++) {
- ret = bch2_trans_mark_metadata_bucket(trans, res, ca,
+ ret = bch2_trans_mark_metadata_bucket(trans, ca,
ca->journal.buckets[i],
BCH_DATA_journal, ca->mi.bucket_size);
if (ret)
return 0;
}
-int bch2_trans_mark_dev_sb(struct bch_fs *c,
- struct disk_reservation *res,
- struct bch_dev *ca)
+int bch2_trans_mark_dev_sb(struct bch_fs *c, struct bch_dev *ca)
{
- return bch2_trans_do(c, res, NULL, 0,
- __bch2_trans_mark_dev_sb(&trans, res, ca));
+ return bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
+ __bch2_trans_mark_dev_sb(&trans, ca));
}
/* Disk reservations: */
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;
alloc_fifo free[RESERVE_NR];
alloc_fifo free_inc;
alloc_heap alloc_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 >> 6);
if (!(buckets = kvpmalloc(sizeof(struct bucket_array) +
nbuckets * sizeof(struct bucket),
GFP_KERNEL|__GFP_ZERO)) ||
+ !(bucket_gens = kvpmalloc(sizeof(struct bucket_gens) + nbuckets,
+ GFP_KERNEL|__GFP_ZERO)) ||
!(buckets_nouse = kvpmalloc(BITS_TO_LONGS(nbuckets) *
sizeof(unsigned long),
GFP_KERNEL|__GFP_ZERO)) ||
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(c);
}
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));
}
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);
free_fifo(&free[i]);
kvpfree(buckets_nouse,
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);