mirrors
/
linux-stable
mirror of https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git
1
0
Fork 0
Code Issues Releases Wiki Activity
linux-stable/fs/bcachefs/btree_write_buffer.c

671 lines
18 KiB
C
Raw Permalink Normal View History

bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
// SPDX-License-Identifier: GPL-2.0
#include "bcachefs.h"
#include "btree_locking.h"
#include "btree_update.h"
#include "btree_update_interior.h"
#include "btree_write_buffer.h"
#include "error.h"
#include "journal.h"
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
#include "journal_io.h"
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
#include "journal_reclaim.h"
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
#include <linux/prefetch.h>
bcachefs: Fix journal pins in btree write buffer btree write buffer flush has two phases - in natural key order, which is more efficient but may fail - then in journal order The journal order flush was assuming that keys were still correctly ordered by journal sequence number - but due to coalescing by the previous phase, we need an additional sort. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2024-03-22 08:01:27 +00:00
#include <linux/sort.h>
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
bcachefs: Journal pins must always have a flush_fn flush_fn is how we identify journal pins in debugfs - this is a debugging aid. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-09 18:19:00 +00:00
static int bch2_btree_write_buffer_journal_flush(struct journal *,
struct journal_entry_pin *, u64);
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
static int bch2_journal_keys_to_write_buffer(struct bch_fs *, struct journal_buf *);
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
bcachefs: wb_key_cmp -> wb_key_ref_cmp Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-12-27 23:23:34 +00:00
static inline bool __wb_key_ref_cmp(const struct wb_key_ref *l, const struct wb_key_ref *r)
bcachefs: Inline btree write buffer sort The sort in the btree write buffer flush path is a very hot path, and it's particularly performance sensitive since it's single threaded and can block every other thread on a multithreaded write workload. It's well worth doing a sort with inlined cmp and swap functions. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-04 04:06:56 +00:00
{
return (cmp_int(l->hi, r->hi) ?:
cmp_int(l->mi, r->mi) ?:
cmp_int(l->lo, r->lo)) >= 0;
}
bcachefs: wb_key_cmp -> wb_key_ref_cmp Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-12-27 23:23:34 +00:00
static inline bool wb_key_ref_cmp(const struct wb_key_ref *l, const struct wb_key_ref *r)
bcachefs: Inline btree write buffer sort The sort in the btree write buffer flush path is a very hot path, and it's particularly performance sensitive since it's single threaded and can block every other thread on a multithreaded write workload. It's well worth doing a sort with inlined cmp and swap functions. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-04 04:06:56 +00:00
{
#ifdef CONFIG_X86_64
int cmp;
asm("mov (%[l]), %%rax;"
"sub (%[r]), %%rax;"
"mov 8(%[l]), %%rax;"
"sbb 8(%[r]), %%rax;"
"mov 16(%[l]), %%rax;"
"sbb 16(%[r]), %%rax;"
: "=@ccae" (cmp)
: [l] "r" (l), [r] "r" (r)
: "rax", "cc");
bcachefs: wb_key_cmp -> wb_key_ref_cmp Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-12-27 23:23:34 +00:00
EBUG_ON(cmp != __wb_key_ref_cmp(l, r));
bcachefs: Inline btree write buffer sort The sort in the btree write buffer flush path is a very hot path, and it's particularly performance sensitive since it's single threaded and can block every other thread on a multithreaded write workload. It's well worth doing a sort with inlined cmp and swap functions. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-04 04:06:56 +00:00
return cmp;
#else
bcachefs: wb_key_cmp -> wb_key_ref_cmp Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-12-27 23:23:34 +00:00
return __wb_key_ref_cmp(l, r);
bcachefs: Inline btree write buffer sort The sort in the btree write buffer flush path is a very hot path, and it's particularly performance sensitive since it's single threaded and can block every other thread on a multithreaded write workload. It's well worth doing a sort with inlined cmp and swap functions. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-04 04:06:56 +00:00
#endif
}
bcachefs: Fix journal pins in btree write buffer btree write buffer flush has two phases - in natural key order, which is more efficient but may fail - then in journal order The journal order flush was assuming that keys were still correctly ordered by journal sequence number - but due to coalescing by the previous phase, we need an additional sort. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2024-03-22 08:01:27 +00:00
static int wb_key_seq_cmp(const void *_l, const void *_r)
{
const struct btree_write_buffered_key *l = _l;
const struct btree_write_buffered_key *r = _r;
return cmp_int(l->journal_seq, r->journal_seq);
}
bcachefs: Inline btree write buffer sort The sort in the btree write buffer flush path is a very hot path, and it's particularly performance sensitive since it's single threaded and can block every other thread on a multithreaded write workload. It's well worth doing a sort with inlined cmp and swap functions. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-04 04:06:56 +00:00
/* Compare excluding idx, the low 24 bits: */
static inline bool wb_key_eq(const void *_l, const void *_r)
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
{
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
const struct wb_key_ref *l = _l;
const struct wb_key_ref *r = _r;
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
bcachefs: Inline btree write buffer sort The sort in the btree write buffer flush path is a very hot path, and it's particularly performance sensitive since it's single threaded and can block every other thread on a multithreaded write workload. It's well worth doing a sort with inlined cmp and swap functions. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-04 04:06:56 +00:00
return !((l->hi ^ r->hi)|
(l->mi ^ r->mi)|
((l->lo >> 24) ^ (r->lo >> 24)));
}
static noinline void wb_sort(struct wb_key_ref *base, size_t num)
{
size_t n = num, a = num / 2;
if (!a) /* num < 2 || size == 0 */
return;
for (;;) {
size_t b, c, d;
if (a) /* Building heap: sift down --a */
--a;
else if (--n) /* Sorting: Extract root to --n */
swap(base[0], base[n]);
else /* Sort complete */
break;
/*
* Sift element at "a" down into heap. This is the
* "bottom-up" variant, which significantly reduces
* calls to cmp_func(): we find the sift-down path all
* the way to the leaves (one compare per level), then
* backtrack to find where to insert the target element.
*
* Because elements tend to sift down close to the leaves,
* this uses fewer compares than doing two per level
* on the way down. (A bit more than half as many on
* average, 3/4 worst-case.)
*/
for (b = a; c = 2*b + 1, (d = c + 1) < n;)
bcachefs: wb_key_cmp -> wb_key_ref_cmp Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-12-27 23:23:34 +00:00
b = wb_key_ref_cmp(base + c, base + d) ? c : d;
bcachefs: Inline btree write buffer sort The sort in the btree write buffer flush path is a very hot path, and it's particularly performance sensitive since it's single threaded and can block every other thread on a multithreaded write workload. It's well worth doing a sort with inlined cmp and swap functions. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-04 04:06:56 +00:00
if (d == n) /* Special case last leaf with no sibling */
b = c;
/* Now backtrack from "b" to the correct location for "a" */
bcachefs: wb_key_cmp -> wb_key_ref_cmp Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-12-27 23:23:34 +00:00
while (b != a && wb_key_ref_cmp(base + a, base + b))
bcachefs: Inline btree write buffer sort The sort in the btree write buffer flush path is a very hot path, and it's particularly performance sensitive since it's single threaded and can block every other thread on a multithreaded write workload. It's well worth doing a sort with inlined cmp and swap functions. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-04 04:06:56 +00:00
b = (b - 1) / 2;
c = b; /* Where "a" belongs */
while (b != a) { /* Shift it into place */
b = (b - 1) / 2;
swap(base[b], base[c]);
}
}
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
}
bcachefs: wb_flush_one_slowpath() A bit of refactoring for better inlining in the main btree write buffer flush path. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-27 02:58:11 +00:00
static noinline int wb_flush_one_slowpath(struct btree_trans *trans,
struct btree_iter *iter,
struct btree_write_buffered_key *wb)
{
bcachefs: btree_iter -> btree_path_idx_t Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-12-04 05:39:38 +00:00
struct btree_path *path = btree_iter_path(trans, iter);
bch2_btree_node_unlock_write(trans, path, path->l[0].b);
bcachefs: wb_flush_one_slowpath() A bit of refactoring for better inlining in the main btree write buffer flush path. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-27 02:58:11 +00:00
trans->journal_res.seq = wb->journal_seq;
return bch2_trans_update(trans, iter, &wb->k,
bcachefs: iter/update/trigger/str_hash flag cleanup Combine iter/update/trigger/str_hash flags into a single enum, and x-macroize them for a to_text() function later. These flags are all for a specific iter/key/update context, so it makes sense to group them together - iter/update/trigger flags were already given distinct bits, this cleans up and unifies that handling. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2024-04-07 22:05:34 +00:00
BTREE_UPDATE_internal_snapshot_node) ?:
bcachefs: wb_flush_one_slowpath() A bit of refactoring for better inlining in the main btree write buffer flush path. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-27 02:58:11 +00:00
bch2_trans_commit(trans, NULL, NULL,
BCH_TRANS_COMMIT_no_enospc|
BCH_TRANS_COMMIT_no_check_rw|
BCH_TRANS_COMMIT_no_journal_res|
BCH_TRANS_COMMIT_journal_reclaim);
}
static inline int wb_flush_one(struct btree_trans *trans, struct btree_iter *iter,
struct btree_write_buffered_key *wb,
bool *write_locked, size_t *fast)
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
{
struct btree_path *path;
int ret;
bcachefs: wb_flush_one_slowpath() A bit of refactoring for better inlining in the main btree write buffer flush path. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-27 02:58:11 +00:00
EBUG_ON(!wb->journal_seq);
bcachefs: Prep work for variable size btree node buffers bcachefs btree nodes are big - typically 256k - and btree roots are pinned in memory. As we're now up to 18 btrees, we now have significant memory overhead in mostly empty btree roots. And in the future we're going to start enforcing that certain btree node boundaries exist, to solve lock contention issues - analagous to XFS's AGIs. Thus, we need to start allocating smaller btree node buffers when we can. This patch changes code that refers to the filesystem constant c->opts.btree_node_size to refer to the btree node buffer size - btree_buf_bytes() - where appropriate. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2024-01-16 18:29:59 +00:00
EBUG_ON(!trans->c->btree_write_buffer.flushing.pin.seq);
EBUG_ON(trans->c->btree_write_buffer.flushing.pin.seq > wb->journal_seq);
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
ret = bch2_btree_iter_traverse(iter);
if (ret)
return ret;
bcachefs: Avoiding dropping/retaking write locks in bch2_btree_write_buffer_flush_one() Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-07 15:42:53 +00:00
/*
* We can't clone a path that has write locks: unshare it now, before
* set_pos and traverse():
*/
bcachefs: btree_iter -> btree_path_idx_t Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-12-04 05:39:38 +00:00
if (btree_iter_path(trans, iter)->ref > 1)
iter->path = __bch2_btree_path_make_mut(trans, iter->path, true, _THIS_IP_);
bcachefs: Avoiding dropping/retaking write locks in bch2_btree_write_buffer_flush_one() Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-07 15:42:53 +00:00
bcachefs: btree_iter -> btree_path_idx_t Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-12-04 05:39:38 +00:00
path = btree_iter_path(trans, iter);
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
if (!*write_locked) {
ret = bch2_btree_node_lock_write(trans, path, &path->l[0].b->c);
if (ret)
return ret;
bch2_btree_node_prep_for_write(trans, path, path->l[0].b);
*write_locked = true;
}
bcachefs: Prep work for variable size btree node buffers bcachefs btree nodes are big - typically 256k - and btree roots are pinned in memory. As we're now up to 18 btrees, we now have significant memory overhead in mostly empty btree roots. And in the future we're going to start enforcing that certain btree node boundaries exist, to solve lock contention issues - analagous to XFS's AGIs. Thus, we need to start allocating smaller btree node buffers when we can. This patch changes code that refers to the filesystem constant c->opts.btree_node_size to refer to the btree node buffer size - btree_buf_bytes() - where appropriate. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2024-01-16 18:29:59 +00:00
if (unlikely(!bch2_btree_node_insert_fits(path->l[0].b, wb->k.k.u64s))) {
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
*write_locked = false;
bcachefs: wb_flush_one_slowpath() A bit of refactoring for better inlining in the main btree write buffer flush path. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-27 02:58:11 +00:00
return wb_flush_one_slowpath(trans, iter, wb);
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
}
bch2_btree_insert_key_leaf(trans, path, &wb->k, wb->journal_seq);
(*fast)++;
return 0;
}
bcachefs: use prejournaled key updates for write buffer flushes The write buffer mechanism journals keys twice in certain situations. A key is always journaled on write buffer insertion, and is potentially journaled again if a write buffer flush falls into either of the slow btree insert paths. This has shown to cause journal recovery ordering problems in the event of an untimely crash. For example, consider if a key is inserted into index 0 of a write buffer, the active write buffer switches to index 1, the key is deleted in index 1, and then index 0 is flushed. If the original key is rejournaled in the btree update from the index 0 flush, the (now deleted) key is journaled in a seq buffer ahead of the latest version of key (which was journaled when the key was deleted in index 1). If the fs crashes while this is still observable in the log, recovery sees the key from the btree update after the delete key from the write buffer insert, which is the incorrect order. This problem is occasionally reproduced by generic/388 and generally manifests as one or more backpointer entry inconsistencies. To avoid this problem, never rejournal write buffered key updates to the associated btree. Instead, use prejournaled key updates to pass the journal seq of the write buffer insert down to the btree insert, which updates the btree leaf pin to reflect the seq of the key. Note that tracking the seq is required instead of just using NOJOURNAL here because otherwise we lose protection of the write buffer pin when the buffer is flushed, which means the key can fall off the tail of the on-disk journal before the btree leaf is flushed and lead to similar recovery inconsistencies. Signed-off-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-07-19 12:53:06 +00:00
/*
* Update a btree with a write buffered key using the journal seq of the
* original write buffer insert.
*
* It is not safe to rejournal the key once it has been inserted into the write
* buffer because that may break recovery ordering. For example, the key may
* have already been modified in the active write buffer in a seq that comes
* before the current transaction. If we were to journal this key again and
* crash, recovery would process updates in the wrong order.
*/
static int
btree_write_buffered_insert(struct btree_trans *trans,
struct btree_write_buffered_key *wb)
{
struct btree_iter iter;
int ret;
bch2_trans_iter_init(trans, &iter, wb->btree, bkey_start_pos(&wb->k.k),
bcachefs: iter/update/trigger/str_hash flag cleanup Combine iter/update/trigger/str_hash flags into a single enum, and x-macroize them for a to_text() function later. These flags are all for a specific iter/key/update context, so it makes sense to group them together - iter/update/trigger flags were already given distinct bits, this cleans up and unifies that handling. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2024-04-07 22:05:34 +00:00
BTREE_ITER_cached|BTREE_ITER_intent);
bcachefs: use prejournaled key updates for write buffer flushes The write buffer mechanism journals keys twice in certain situations. A key is always journaled on write buffer insertion, and is potentially journaled again if a write buffer flush falls into either of the slow btree insert paths. This has shown to cause journal recovery ordering problems in the event of an untimely crash. For example, consider if a key is inserted into index 0 of a write buffer, the active write buffer switches to index 1, the key is deleted in index 1, and then index 0 is flushed. If the original key is rejournaled in the btree update from the index 0 flush, the (now deleted) key is journaled in a seq buffer ahead of the latest version of key (which was journaled when the key was deleted in index 1). If the fs crashes while this is still observable in the log, recovery sees the key from the btree update after the delete key from the write buffer insert, which is the incorrect order. This problem is occasionally reproduced by generic/388 and generally manifests as one or more backpointer entry inconsistencies. To avoid this problem, never rejournal write buffered key updates to the associated btree. Instead, use prejournaled key updates to pass the journal seq of the write buffer insert down to the btree insert, which updates the btree leaf pin to reflect the seq of the key. Note that tracking the seq is required instead of just using NOJOURNAL here because otherwise we lose protection of the write buffer pin when the buffer is flushed, which means the key can fall off the tail of the on-disk journal before the btree leaf is flushed and lead to similar recovery inconsistencies. Signed-off-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-07-19 12:53:06 +00:00
bcachefs: Kill BTREE_UPDATE_PREJOURNAL With the previous patch that reworks BTREE_INSERT_JOURNAL_REPLAY, we can now switch the btree write buffer to use it for flushing. This has the advantage that transaction commits don't need to take a journal reservation at all. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-09 03:04:29 +00:00
trans->journal_res.seq = wb->journal_seq;
bcachefs: use prejournaled key updates for write buffer flushes The write buffer mechanism journals keys twice in certain situations. A key is always journaled on write buffer insertion, and is potentially journaled again if a write buffer flush falls into either of the slow btree insert paths. This has shown to cause journal recovery ordering problems in the event of an untimely crash. For example, consider if a key is inserted into index 0 of a write buffer, the active write buffer switches to index 1, the key is deleted in index 1, and then index 0 is flushed. If the original key is rejournaled in the btree update from the index 0 flush, the (now deleted) key is journaled in a seq buffer ahead of the latest version of key (which was journaled when the key was deleted in index 1). If the fs crashes while this is still observable in the log, recovery sees the key from the btree update after the delete key from the write buffer insert, which is the incorrect order. This problem is occasionally reproduced by generic/388 and generally manifests as one or more backpointer entry inconsistencies. To avoid this problem, never rejournal write buffered key updates to the associated btree. Instead, use prejournaled key updates to pass the journal seq of the write buffer insert down to the btree insert, which updates the btree leaf pin to reflect the seq of the key. Note that tracking the seq is required instead of just using NOJOURNAL here because otherwise we lose protection of the write buffer pin when the buffer is flushed, which means the key can fall off the tail of the on-disk journal before the btree leaf is flushed and lead to similar recovery inconsistencies. Signed-off-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-07-19 12:53:06 +00:00
ret = bch2_btree_iter_traverse(&iter) ?:
bcachefs: Kill BTREE_UPDATE_PREJOURNAL With the previous patch that reworks BTREE_INSERT_JOURNAL_REPLAY, we can now switch the btree write buffer to use it for flushing. This has the advantage that transaction commits don't need to take a journal reservation at all. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-09 03:04:29 +00:00
bch2_trans_update(trans, &iter, &wb->k,
bcachefs: iter/update/trigger/str_hash flag cleanup Combine iter/update/trigger/str_hash flags into a single enum, and x-macroize them for a to_text() function later. These flags are all for a specific iter/key/update context, so it makes sense to group them together - iter/update/trigger flags were already given distinct bits, this cleans up and unifies that handling. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2024-04-07 22:05:34 +00:00
BTREE_UPDATE_internal_snapshot_node);
bcachefs: use prejournaled key updates for write buffer flushes The write buffer mechanism journals keys twice in certain situations. A key is always journaled on write buffer insertion, and is potentially journaled again if a write buffer flush falls into either of the slow btree insert paths. This has shown to cause journal recovery ordering problems in the event of an untimely crash. For example, consider if a key is inserted into index 0 of a write buffer, the active write buffer switches to index 1, the key is deleted in index 1, and then index 0 is flushed. If the original key is rejournaled in the btree update from the index 0 flush, the (now deleted) key is journaled in a seq buffer ahead of the latest version of key (which was journaled when the key was deleted in index 1). If the fs crashes while this is still observable in the log, recovery sees the key from the btree update after the delete key from the write buffer insert, which is the incorrect order. This problem is occasionally reproduced by generic/388 and generally manifests as one or more backpointer entry inconsistencies. To avoid this problem, never rejournal write buffered key updates to the associated btree. Instead, use prejournaled key updates to pass the journal seq of the write buffer insert down to the btree insert, which updates the btree leaf pin to reflect the seq of the key. Note that tracking the seq is required instead of just using NOJOURNAL here because otherwise we lose protection of the write buffer pin when the buffer is flushed, which means the key can fall off the tail of the on-disk journal before the btree leaf is flushed and lead to similar recovery inconsistencies. Signed-off-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-07-19 12:53:06 +00:00
bch2_trans_iter_exit(trans, &iter);
return ret;
}
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
static void move_keys_from_inc_to_flushing(struct btree_write_buffer *wb)
{
struct bch_fs *c = container_of(wb, struct bch_fs, btree_write_buffer);
struct journal *j = &c->journal;
if (!wb->inc.keys.nr)
return;
bch2_journal_pin_add(j, wb->inc.keys.data[0].journal_seq, &wb->flushing.pin,
bch2_btree_write_buffer_journal_flush);
darray_resize(&wb->flushing.keys, min_t(size_t, 1U << 20, wb->flushing.keys.nr + wb->inc.keys.nr));
darray_resize(&wb->sorted, wb->flushing.keys.size);
if (!wb->flushing.keys.nr && wb->sorted.size >= wb->inc.keys.nr) {
swap(wb->flushing.keys, wb->inc.keys);
goto out;
}
size_t nr = min(darray_room(wb->flushing.keys),
wb->sorted.size - wb->flushing.keys.nr);
nr = min(nr, wb->inc.keys.nr);
memcpy(&darray_top(wb->flushing.keys),
wb->inc.keys.data,
sizeof(wb->inc.keys.data[0]) * nr);
memmove(wb->inc.keys.data,
wb->inc.keys.data + nr,
sizeof(wb->inc.keys.data[0]) * (wb->inc.keys.nr - nr));
wb->flushing.keys.nr += nr;
wb->inc.keys.nr -= nr;
out:
if (!wb->inc.keys.nr)
bch2_journal_pin_drop(j, &wb->inc.pin);
else
bch2_journal_pin_update(j, wb->inc.keys.data[0].journal_seq, &wb->inc.pin,
bch2_btree_write_buffer_journal_flush);
if (j->watermark) {
spin_lock(&j->lock);
bch2_journal_set_watermark(j);
spin_unlock(&j->lock);
}
BUG_ON(wb->sorted.size < wb->flushing.keys.nr);
}
static int bch2_btree_write_buffer_flush_locked(struct btree_trans *trans)
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
{
struct bch_fs *c = trans->c;
struct journal *j = &c->journal;
struct btree_write_buffer *wb = &c->btree_write_buffer;
struct btree_iter iter = { NULL };
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
size_t skipped = 0, fast = 0, slowpath = 0;
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
bool write_locked = false;
int ret = 0;
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
bch2_trans_unlock(trans);
bch2_trans_begin(trans);
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
mutex_lock(&wb->inc.lock);
move_keys_from_inc_to_flushing(wb);
mutex_unlock(&wb->inc.lock);
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
for (size_t i = 0; i < wb->flushing.keys.nr; i++) {
wb->sorted.data[i].idx = i;
wb->sorted.data[i].btree = wb->flushing.keys.data[i].btree;
bcachefs: Inline btree write buffer sort The sort in the btree write buffer flush path is a very hot path, and it's particularly performance sensitive since it's single threaded and can block every other thread on a multithreaded write workload. It's well worth doing a sort with inlined cmp and swap functions. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-04 04:06:56 +00:00
memcpy(&wb->sorted.data[i].pos, &wb->flushing.keys.data[i].k.k.p, sizeof(struct bpos));
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
}
wb->sorted.nr = wb->flushing.keys.nr;
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
/*
* We first sort so that we can detect and skip redundant updates, and
* then we attempt to flush in sorted btree order, as this is most
* efficient.
*
* However, since we're not flushing in the order they appear in the
* journal we won't be able to drop our journal pin until everything is
bcachefs: more aggressive fast path write buffer key flushing The btree write buffer flush code is prone to causing journal deadlock due to inefficient use and release of reservation space. Reservation is not pre-reserved for write buffered keys (as is done for key cache keys, for example), because the write buffer flush side uses a fast path that attempts insertion without need for any reservation at all. The write buffer flush attempts to deal with this by inserting keys using the BTREE_INSERT_JOURNAL_RECLAIM flag to return an error on journal reservations that require blocking. Upon first error, it falls back to a slow path that inserts in journal order and supports moving the associated journal pin forward. The problem is that under pathological conditions (i.e. smaller log, larger write buffer and journal reservation pressure), we've seen instances where the fast path fails fairly quickly without having completed many insertions, and then the slow path is unable to push the journal pin forward enough to free up the space it needs to completely flush the buffer. This problem is occasionally reproduced by fstest generic/333. To avoid this problem, update the fast path algorithm to skip key inserts that fail due to inability to acquire needed journal reservation without immediately breaking out of the loop. Instead, insert as many keys as possible, zap the sequence numbers to mark them as processed, and then fall back to the slow path to process the remaining set in journal order. This reduces the amount of journal reservation that might be required to flush the entire buffer and increases the odds that the slow path is able to move the journal pin forward and free up space as keys are processed. Signed-off-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-03-17 12:54:01 +00:00
* flushed - which means this could deadlock the journal if we weren't
bcachefs: Rename BTREE_INSERT flags BTREE_INSERT flags are actually transaction commit flags - rename them for clarity. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-11 21:31:50 +00:00
* passing BCH_TRANS_COMMIT_journal_reclaim. This causes the update to fail
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
* if it would block taking a journal reservation.
*
bcachefs: more aggressive fast path write buffer key flushing The btree write buffer flush code is prone to causing journal deadlock due to inefficient use and release of reservation space. Reservation is not pre-reserved for write buffered keys (as is done for key cache keys, for example), because the write buffer flush side uses a fast path that attempts insertion without need for any reservation at all. The write buffer flush attempts to deal with this by inserting keys using the BTREE_INSERT_JOURNAL_RECLAIM flag to return an error on journal reservations that require blocking. Upon first error, it falls back to a slow path that inserts in journal order and supports moving the associated journal pin forward. The problem is that under pathological conditions (i.e. smaller log, larger write buffer and journal reservation pressure), we've seen instances where the fast path fails fairly quickly without having completed many insertions, and then the slow path is unable to push the journal pin forward enough to free up the space it needs to completely flush the buffer. This problem is occasionally reproduced by fstest generic/333. To avoid this problem, update the fast path algorithm to skip key inserts that fail due to inability to acquire needed journal reservation without immediately breaking out of the loop. Instead, insert as many keys as possible, zap the sequence numbers to mark them as processed, and then fall back to the slow path to process the remaining set in journal order. This reduces the amount of journal reservation that might be required to flush the entire buffer and increases the odds that the slow path is able to move the journal pin forward and free up space as keys are processed. Signed-off-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-03-17 12:54:01 +00:00
* If that happens, simply skip the key so we can optimistically insert
* as many keys as possible in the fast path.
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
*/
bcachefs: Inline btree write buffer sort The sort in the btree write buffer flush path is a very hot path, and it's particularly performance sensitive since it's single threaded and can block every other thread on a multithreaded write workload. It's well worth doing a sort with inlined cmp and swap functions. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-04 04:06:56 +00:00
wb_sort(wb->sorted.data, wb->sorted.nr);
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
darray_for_each(wb->sorted, i) {
struct btree_write_buffered_key *k = &wb->flushing.keys.data[i->idx];
for (struct wb_key_ref *n = i + 1; n < min(i + 4, &darray_top(wb->sorted)); n++)
prefetch(&wb->flushing.keys.data[n->idx]);
BUG_ON(!k->journal_seq);
if (i + 1 < &darray_top(wb->sorted) &&
bcachefs: Inline btree write buffer sort The sort in the btree write buffer flush path is a very hot path, and it's particularly performance sensitive since it's single threaded and can block every other thread on a multithreaded write workload. It's well worth doing a sort with inlined cmp and swap functions. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-04 04:06:56 +00:00
wb_key_eq(i, i + 1)) {
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
struct btree_write_buffered_key *n = &wb->flushing.keys.data[i[1].idx];
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
skipped++;
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
n->journal_seq = min_t(u64, n->journal_seq, k->journal_seq);
k->journal_seq = 0;
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
continue;
}
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
if (write_locked) {
bcachefs: btree_iter -> btree_path_idx_t Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-12-04 05:39:38 +00:00
struct btree_path *path = btree_iter_path(trans, &iter);
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
if (path->btree_id != i->btree ||
bpos_gt(k->k.k.p, path->l[0].b->key.k.p)) {
bch2_btree_node_unlock_write(trans, path, path->l[0].b);
write_locked = false;
bcachefs: Fix btree node merging on write buffer btrees The btree write buffer flush fastpath that avoids the main transaction commit path had the unfortunate side effect of not doing btree node merging. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-12-27 03:42:34 +00:00
ret = lockrestart_do(trans,
bch2_btree_iter_traverse(&iter) ?:
bch2_foreground_maybe_merge(trans, iter.path, 0,
BCH_WATERMARK_reclaim|
BCH_TRANS_COMMIT_journal_reclaim|
BCH_TRANS_COMMIT_no_check_rw|
BCH_TRANS_COMMIT_no_enospc));
if (ret)
goto err;
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
}
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
}
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
if (!iter.path || iter.btree_id != k->btree) {
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
bch2_trans_iter_exit(trans, &iter);
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
bch2_trans_iter_init(trans, &iter, k->btree, k->k.k.p,
bcachefs: iter/update/trigger/str_hash flag cleanup Combine iter/update/trigger/str_hash flags into a single enum, and x-macroize them for a to_text() function later. These flags are all for a specific iter/key/update context, so it makes sense to group them together - iter/update/trigger flags were already given distinct bits, this cleans up and unifies that handling. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2024-04-07 22:05:34 +00:00
BTREE_ITER_intent|BTREE_ITER_all_snapshots);
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
}
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
bch2_btree_iter_set_pos(&iter, k->k.k.p);
bcachefs: btree_iter -> btree_path_idx_t Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-12-04 05:39:38 +00:00
btree_iter_path(trans, &iter)->preserve = false;
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
do {
bcachefs: more write buffer refactoring prep work for big rewrite - no functional changes in this patch. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-27 03:06:48 +00:00
if (race_fault()) {
ret = -BCH_ERR_journal_reclaim_would_deadlock;
break;
}
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
ret = wb_flush_one(trans, &iter, k, &write_locked, &fast);
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
if (!write_locked)
bch2_trans_begin(trans);
} while (bch2_err_matches(ret, BCH_ERR_transaction_restart));
bcachefs: wb_flush_one_slowpath() A bit of refactoring for better inlining in the main btree write buffer flush path. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-27 02:58:11 +00:00
if (!ret) {
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
k->journal_seq = 0;
bcachefs: wb_flush_one_slowpath() A bit of refactoring for better inlining in the main btree write buffer flush path. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-27 02:58:11 +00:00
} else if (ret == -BCH_ERR_journal_reclaim_would_deadlock) {
bcachefs: more aggressive fast path write buffer key flushing The btree write buffer flush code is prone to causing journal deadlock due to inefficient use and release of reservation space. Reservation is not pre-reserved for write buffered keys (as is done for key cache keys, for example), because the write buffer flush side uses a fast path that attempts insertion without need for any reservation at all. The write buffer flush attempts to deal with this by inserting keys using the BTREE_INSERT_JOURNAL_RECLAIM flag to return an error on journal reservations that require blocking. Upon first error, it falls back to a slow path that inserts in journal order and supports moving the associated journal pin forward. The problem is that under pathological conditions (i.e. smaller log, larger write buffer and journal reservation pressure), we've seen instances where the fast path fails fairly quickly without having completed many insertions, and then the slow path is unable to push the journal pin forward enough to free up the space it needs to completely flush the buffer. This problem is occasionally reproduced by fstest generic/333. To avoid this problem, update the fast path algorithm to skip key inserts that fail due to inability to acquire needed journal reservation without immediately breaking out of the loop. Instead, insert as many keys as possible, zap the sequence numbers to mark them as processed, and then fall back to the slow path to process the remaining set in journal order. This reduces the amount of journal reservation that might be required to flush the entire buffer and increases the odds that the slow path is able to move the journal pin forward and free up space as keys are processed. Signed-off-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-03-17 12:54:01 +00:00
slowpath++;
bcachefs: wb_flush_one_slowpath() A bit of refactoring for better inlining in the main btree write buffer flush path. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-27 02:58:11 +00:00
ret = 0;
} else
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
break;
}
bcachefs: btree_iter -> btree_path_idx_t Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-12-04 05:39:38 +00:00
if (write_locked) {
struct btree_path *path = btree_iter_path(trans, &iter);
bch2_btree_node_unlock_write(trans, path, path->l[0].b);
}
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
bch2_trans_iter_exit(trans, &iter);
bcachefs: more write buffer refactoring prep work for big rewrite - no functional changes in this patch. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-27 03:06:48 +00:00
if (ret)
goto err;
if (slowpath) {
/*
* Flush in the order they were present in the journal, so that
* we can release journal pins:
* The fastpath zapped the seq of keys that were successfully flushed so
* we can skip those here.
*/
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
trace_and_count(c, write_buffer_flush_slowpath, trans, slowpath, wb->flushing.keys.nr);
bcachefs: more write buffer refactoring prep work for big rewrite - no functional changes in this patch. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-27 03:06:48 +00:00
bcachefs: Fix journal pins in btree write buffer btree write buffer flush has two phases - in natural key order, which is more efficient but may fail - then in journal order The journal order flush was assuming that keys were still correctly ordered by journal sequence number - but due to coalescing by the previous phase, we need an additional sort. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2024-03-22 08:01:27 +00:00
sort(wb->flushing.keys.data,
wb->flushing.keys.nr,
sizeof(wb->flushing.keys.data[0]),
wb_key_seq_cmp, NULL);
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
darray_for_each(wb->flushing.keys, i) {
bcachefs: more write buffer refactoring prep work for big rewrite - no functional changes in this patch. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-27 03:06:48 +00:00
if (!i->journal_seq)
continue;
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
bch2_journal_pin_update(j, i->journal_seq, &wb->flushing.pin,
bch2_btree_write_buffer_journal_flush);
bch2_trans_begin(trans);
bcachefs: more write buffer refactoring prep work for big rewrite - no functional changes in this patch. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-27 03:06:48 +00:00
ret = commit_do(trans, NULL, NULL,
BCH_WATERMARK_reclaim|
bcachefs: Fix btree node merging on write buffer btrees The btree write buffer flush fastpath that avoids the main transaction commit path had the unfortunate side effect of not doing btree node merging. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-12-27 03:42:34 +00:00
BCH_TRANS_COMMIT_journal_reclaim|
bcachefs: more write buffer refactoring prep work for big rewrite - no functional changes in this patch. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-27 03:06:48 +00:00
BCH_TRANS_COMMIT_no_check_rw|
BCH_TRANS_COMMIT_no_enospc|
bcachefs: Fix btree node merging on write buffer btrees The btree write buffer flush fastpath that avoids the main transaction commit path had the unfortunate side effect of not doing btree node merging. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-12-27 03:42:34 +00:00
BCH_TRANS_COMMIT_no_journal_res ,
bcachefs: more write buffer refactoring prep work for big rewrite - no functional changes in this patch. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-27 03:06:48 +00:00
btree_write_buffered_insert(trans, i));
if (ret)
goto err;
}
}
err:
bcachefs: Improve bch2_fatal_error() error messages should always include __func__ Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2024-03-18 01:51:19 +00:00
bch2_fs_fatal_err_on(ret, c, "%s", bch2_err_str(ret));
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
trace_write_buffer_flush(trans, wb->flushing.keys.nr, skipped, fast, 0);
bch2_journal_pin_drop(j, &wb->flushing.pin);
wb->flushing.keys.nr = 0;
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
return ret;
}
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
static int fetch_wb_keys_from_journal(struct bch_fs *c, u64 seq)
{
struct journal *j = &c->journal;
struct journal_buf *buf;
int ret = 0;
while (!ret && (buf = bch2_next_write_buffer_flush_journal_buf(j, seq))) {
ret = bch2_journal_keys_to_write_buffer(c, buf);
mutex_unlock(&j->buf_lock);
}
return ret;
}
static int btree_write_buffer_flush_seq(struct btree_trans *trans, u64 seq)
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
{
bcachefs: Improve btree write buffer tracepoints - add a tracepoint for write_buffer_flush_sync; this is expensive - fix the write_buffer_flush_slowpath tracepoint Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-03 02:31:16 +00:00
struct bch_fs *c = trans->c;
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
struct btree_write_buffer *wb = &c->btree_write_buffer;
int ret = 0, fetch_from_journal_err;
bcachefs: Improve btree write buffer tracepoints - add a tracepoint for write_buffer_flush_sync; this is expensive - fix the write_buffer_flush_slowpath tracepoint Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-03 02:31:16 +00:00
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
do {
bch2_trans_unlock(trans);
bcachefs: Clean up btree write buffer write ref handling __bch2_btree_write_buffer_flush() now assumes a write ref is already held (as called by the transaction commit path); and the wrappers bch2_write_buffer_flush() and flush_sync() take an explicit write ref. This means internally the write buffer code can always use BTREE_INSERT_NOCHECK_RW, instead of in the previous code passing flags around and hoping the NOCHECK_RW flag was always carried around correctly. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 23:37:15 +00:00
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
fetch_from_journal_err = fetch_wb_keys_from_journal(c, seq);
/*
* On memory allocation failure, bch2_btree_write_buffer_flush_locked()
* is not guaranteed to empty wb->inc:
*/
mutex_lock(&wb->flushing.lock);
ret = bch2_btree_write_buffer_flush_locked(trans);
mutex_unlock(&wb->flushing.lock);
} while (!ret &&
(fetch_from_journal_err ||
(wb->inc.pin.seq && wb->inc.pin.seq <= seq) ||
(wb->flushing.pin.seq && wb->flushing.pin.seq <= seq)));
bcachefs: Improve btree write buffer tracepoints - add a tracepoint for write_buffer_flush_sync; this is expensive - fix the write_buffer_flush_slowpath tracepoint Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-03 02:31:16 +00:00
bcachefs: Clean up btree write buffer write ref handling __bch2_btree_write_buffer_flush() now assumes a write ref is already held (as called by the transaction commit path); and the wrappers bch2_write_buffer_flush() and flush_sync() take an explicit write ref. This means internally the write buffer code can always use BTREE_INSERT_NOCHECK_RW, instead of in the previous code passing flags around and hoping the NOCHECK_RW flag was always carried around correctly. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 23:37:15 +00:00
return ret;
}
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
static int bch2_btree_write_buffer_journal_flush(struct journal *j,
struct journal_entry_pin *_pin, u64 seq)
{
struct bch_fs *c = container_of(j, struct bch_fs, journal);
return bch2_trans_run(c, btree_write_buffer_flush_seq(trans, seq));
}
int bch2_btree_write_buffer_flush_sync(struct btree_trans *trans)
{
struct bch_fs *c = trans->c;
trace_and_count(c, write_buffer_flush_sync, trans, _RET_IP_);
return btree_write_buffer_flush_seq(trans, journal_cur_seq(&c->journal));
}
bcachefs: Clean up btree write buffer write ref handling __bch2_btree_write_buffer_flush() now assumes a write ref is already held (as called by the transaction commit path); and the wrappers bch2_write_buffer_flush() and flush_sync() take an explicit write ref. This means internally the write buffer code can always use BTREE_INSERT_NOCHECK_RW, instead of in the previous code passing flags around and hoping the NOCHECK_RW flag was always carried around correctly. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 23:37:15 +00:00
int bch2_btree_write_buffer_flush_nocheck_rw(struct btree_trans *trans)
{
bcachefs: bch2_btree_write_buffer_flush_locked() Minor refactoring - improved naming, and move the responsibility for flush_lock to the caller instead of having it be shared. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-03 00:32:19 +00:00
struct bch_fs *c = trans->c;
struct btree_write_buffer *wb = &c->btree_write_buffer;
int ret = 0;
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
if (mutex_trylock(&wb->flushing.lock)) {
bcachefs: bch2_btree_write_buffer_flush_locked() Minor refactoring - improved naming, and move the responsibility for flush_lock to the caller instead of having it be shared. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-03 00:32:19 +00:00
ret = bch2_btree_write_buffer_flush_locked(trans);
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
mutex_unlock(&wb->flushing.lock);
bcachefs: bch2_btree_write_buffer_flush_locked() Minor refactoring - improved naming, and move the responsibility for flush_lock to the caller instead of having it be shared. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-03 00:32:19 +00:00
}
return ret;
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
}
bcachefs: bch2_btree_write_buffer_flush() -> bch2_btree_write_buffer_tryflush() More accurate naming. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-03 00:36:00 +00:00
int bch2_btree_write_buffer_tryflush(struct btree_trans *trans)
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
{
bcachefs: Clean up btree write buffer write ref handling __bch2_btree_write_buffer_flush() now assumes a write ref is already held (as called by the transaction commit path); and the wrappers bch2_write_buffer_flush() and flush_sync() take an explicit write ref. This means internally the write buffer code can always use BTREE_INSERT_NOCHECK_RW, instead of in the previous code passing flags around and hoping the NOCHECK_RW flag was always carried around correctly. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 23:37:15 +00:00
struct bch_fs *c = trans->c;
if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_btree_write_buffer))
return -BCH_ERR_erofs_no_writes;
int ret = bch2_btree_write_buffer_flush_nocheck_rw(trans);
bch2_write_ref_put(c, BCH_WRITE_REF_btree_write_buffer);
return ret;
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
}
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
static void bch2_btree_write_buffer_flush_work(struct work_struct *work)
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
{
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
struct bch_fs *c = container_of(work, struct bch_fs, btree_write_buffer.flush_work);
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
struct btree_write_buffer *wb = &c->btree_write_buffer;
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
int ret;
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
mutex_lock(&wb->flushing.lock);
do {
ret = bch2_trans_run(c, bch2_btree_write_buffer_flush_locked(trans));
} while (!ret && bch2_btree_write_buffer_should_flush(c));
mutex_unlock(&wb->flushing.lock);
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
bch2_write_ref_put(c, BCH_WRITE_REF_btree_write_buffer);
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
}
bcachefs: __bch2_journal_key_to_wb -> bch2_journal_key_to_wb_slowpath Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-12-28 01:26:30 +00:00
int bch2_journal_key_to_wb_slowpath(struct bch_fs *c,
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
struct journal_keys_to_wb *dst,
enum btree_id btree, struct bkey_i *k)
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
{
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
struct btree_write_buffer *wb = &c->btree_write_buffer;
int ret;
retry:
ret = darray_make_room_gfp(&dst->wb->keys, 1, GFP_KERNEL);
if (!ret && dst->wb == &wb->flushing)
ret = darray_resize(&wb->sorted, wb->flushing.keys.size);
if (unlikely(ret)) {
if (dst->wb == &c->btree_write_buffer.flushing) {
mutex_unlock(&dst->wb->lock);
dst->wb = &c->btree_write_buffer.inc;
bch2_journal_pin_add(&c->journal, dst->seq, &dst->wb->pin,
bch2_btree_write_buffer_journal_flush);
goto retry;
}
return ret;
}
dst->room = darray_room(dst->wb->keys);
if (dst->wb == &wb->flushing)
dst->room = min(dst->room, wb->sorted.size - wb->flushing.keys.nr);
BUG_ON(!dst->room);
BUG_ON(!dst->seq);
struct btree_write_buffered_key *wb_k = &darray_top(dst->wb->keys);
wb_k->journal_seq = dst->seq;
wb_k->btree = btree;
bkey_copy(&wb_k->k, k);
dst->wb->keys.nr++;
dst->room--;
return 0;
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
}
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
void bch2_journal_keys_to_write_buffer_start(struct bch_fs *c, struct journal_keys_to_wb *dst, u64 seq)
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
{
struct btree_write_buffer *wb = &c->btree_write_buffer;
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
if (mutex_trylock(&wb->flushing.lock)) {
mutex_lock(&wb->inc.lock);
move_keys_from_inc_to_flushing(wb);
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
/*
* Attempt to skip wb->inc, and add keys directly to
* wb->flushing, saving us a copy later:
*/
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
if (!wb->inc.keys.nr) {
dst->wb = &wb->flushing;
} else {
mutex_unlock(&wb->flushing.lock);
dst->wb = &wb->inc;
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
}
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
} else {
mutex_lock(&wb->inc.lock);
dst->wb = &wb->inc;
}
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
dst->room = darray_room(dst->wb->keys);
if (dst->wb == &wb->flushing)
dst->room = min(dst->room, wb->sorted.size - wb->flushing.keys.nr);
dst->seq = seq;
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
bch2_journal_pin_add(&c->journal, seq, &dst->wb->pin,
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
bch2_btree_write_buffer_journal_flush);
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
}
void bch2_journal_keys_to_write_buffer_end(struct bch_fs *c, struct journal_keys_to_wb *dst)
{
struct btree_write_buffer *wb = &c->btree_write_buffer;
if (!dst->wb->keys.nr)
bch2_journal_pin_drop(&c->journal, &dst->wb->pin);
if (bch2_btree_write_buffer_should_flush(c) &&
__bch2_write_ref_tryget(c, BCH_WRITE_REF_btree_write_buffer) &&
!queue_work(system_unbound_wq, &c->btree_write_buffer.flush_work))
bch2_write_ref_put(c, BCH_WRITE_REF_btree_write_buffer);
if (dst->wb == &wb->flushing)
mutex_unlock(&wb->flushing.lock);
mutex_unlock(&wb->inc.lock);
}
static int bch2_journal_keys_to_write_buffer(struct bch_fs *c, struct journal_buf *buf)
{
struct journal_keys_to_wb dst;
int ret = 0;
bch2_journal_keys_to_write_buffer_start(c, &dst, le64_to_cpu(buf->data->seq));
for_each_jset_entry_type(entry, buf->data, BCH_JSET_ENTRY_write_buffer_keys) {
jset_entry_for_each_key(entry, k) {
ret = bch2_journal_key_to_wb(c, &dst, entry->btree_id, k);
if (ret)
goto out;
}
entry->type = BCH_JSET_ENTRY_btree_keys;
}
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
bcachefs: Fix journal_buf bitfield accesses All jounal_buf bitfield updates must happen under the journal lock - perhaps we should just switch these to atomic bit flags. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2024-02-17 22:54:39 +00:00
spin_lock(&c->journal.lock);
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
buf->need_flush_to_write_buffer = false;
bcachefs: Fix journal_buf bitfield accesses All jounal_buf bitfield updates must happen under the journal lock - perhaps we should just switch these to atomic bit flags. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2024-02-17 22:54:39 +00:00
spin_unlock(&c->journal.lock);
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
out:
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
bch2_journal_keys_to_write_buffer_end(c, &dst);
return ret;
}
static int wb_keys_resize(struct btree_write_buffer_keys *wb, size_t new_size)
{
if (wb->keys.size >= new_size)
return 0;
if (!mutex_trylock(&wb->lock))
return -EINTR;
int ret = darray_resize(&wb->keys, new_size);
mutex_unlock(&wb->lock);
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
return ret;
}
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
int bch2_btree_write_buffer_resize(struct bch_fs *c, size_t new_size)
{
struct btree_write_buffer *wb = &c->btree_write_buffer;
return wb_keys_resize(&wb->flushing, new_size) ?:
wb_keys_resize(&wb->inc, new_size);
}
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
void bch2_fs_btree_write_buffer_exit(struct bch_fs *c)
{
struct btree_write_buffer *wb = &c->btree_write_buffer;
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
BUG_ON((wb->inc.keys.nr || wb->flushing.keys.nr) &&
!bch2_journal_error(&c->journal));
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
darray_exit(&wb->sorted);
darray_exit(&wb->flushing.keys);
darray_exit(&wb->inc.keys);
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
}
int bch2_fs_btree_write_buffer_init(struct bch_fs *c)
{
struct btree_write_buffer *wb = &c->btree_write_buffer;
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
mutex_init(&wb->inc.lock);
mutex_init(&wb->flushing.lock);
INIT_WORK(&wb->flush_work, bch2_btree_write_buffer_flush_work);
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
/* Will be resized by journal as needed: */
unsigned initial_size = 1 << 16;
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
bcachefs: btree write buffer now slurps keys from journal Previosuly, the transaction commit path would have to add keys to the btree write buffer as a separate operation, requiring additional global synchronization. This patch introduces a new journal entry type, which indicates that the keys need to be copied into the btree write buffer prior to being written out. We switch the journal entry type back to JSET_ENTRY_btree_keys prior to write, so this is not an on disk format change. Flushing the btree write buffer may require pulling keys out of journal entries yet to be written, and quiescing outstanding journal reservations; we previously added journal->buf_lock for synchronization with the journal write path. We also can't put strict bounds on the number of keys in the journal destined for the write buffer, which means we might overflow the size of the preallocated buffer and have to reallocate - this introduces a potentially fatal memory allocation failure. This is something we'll have to watch for, if it becomes an issue in practice we can do additional mitigation. The transaction commit path no longer has to explicitly check if the write buffer is full and wait on flushing; this is another performance optimization. Instead, when the btree write buffer is close to full we change the journal watermark, so that only reservations for journal reclaim are allowed. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-11-02 22:57:19 +00:00
return darray_make_room(&wb->inc.keys, initial_size) ?:
darray_make_room(&wb->flushing.keys, initial_size) ?:
darray_make_room(&wb->sorted, initial_size);
bcachefs: Btree write buffer This adds a new method of doing btree updates - a straight write buffer, implemented as a flat fixed size array. This is only useful when we don't need to read from the btree in order to do the update, and when reading is infrequent - perfect for the LRU btree. This will make LRU btree updates fast enough that we'll be able to use it for persistently indexing buckets by fragmentation, which will be a massive boost to copygc performance. Changes: - A new btree_insert_type enum, for btree_insert_entries. Specifies btree, btree key cache, or btree write buffer. - bch2_trans_update_buffered(): updates via the btree write buffer don't need a btree path, so we need a new update path. - Transaction commit path changes: The update to the btree write buffer both mutates global, and can fail if there isn't currently room. Therefore we do all write buffer updates in the transaction all at once, and also if it fails we have to revert filesystem usage counter changes. If there isn't room we flush the write buffer in the transaction commit error path and retry. - A new persistent option, for specifying the number of entries in the write buffer. Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-01-04 05:00:50 +00:00
}