From d6d539c9a7ad0655e5ad46b5e869f1b20bce8953 Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Sat, 17 Aug 2024 19:51:13 -0400 Subject: bcachefs: Reallocate table when we're increasing size Fixes: c2f6e16a6771 ("bcachefs: Increase size of cuckoo hash table on too many rehashes") Signed-off-by: Kent Overstreet --- fs/bcachefs/buckets_waiting_for_journal.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/fs/bcachefs/buckets_waiting_for_journal.c b/fs/bcachefs/buckets_waiting_for_journal.c index f70eb2127d32..f9fb150eda70 100644 --- a/fs/bcachefs/buckets_waiting_for_journal.c +++ b/fs/bcachefs/buckets_waiting_for_journal.c @@ -107,7 +107,7 @@ int bch2_set_bucket_needs_journal_commit(struct buckets_waiting_for_journal *b, nr_elements += t->d[i].journal_seq > flushed_seq; new_bits = ilog2(roundup_pow_of_two(nr_elements * 3)); - +realloc: n = kvmalloc(sizeof(*n) + (sizeof(n->d[0]) << new_bits), GFP_KERNEL); if (!n) { ret = -BCH_ERR_ENOMEM_buckets_waiting_for_journal_set; @@ -118,6 +118,8 @@ retry_rehash: if (nr_rehashes_this_size == 3) { new_bits++; nr_rehashes_this_size = 0; + kvfree(n); + goto realloc; } nr_rehashes++; -- cgit v1.2.3 From d9f49c3106e404776afcf6c5682357f4fe088beb Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Sat, 17 Aug 2024 16:41:39 -0400 Subject: bcachefs: fix field-spanning write warning attempts to retrofit memory safety onto C are increasingly annoying ------------[ cut here ]------------ memcpy: detected field-spanning write (size 4) of single field "&k.replicas" at fs/bcachefs/replicas.c:454 (size 3) WARNING: CPU: 5 PID: 6525 at fs/bcachefs/replicas.c:454 bch2_replicas_gc2+0x2cb/0x400 [bcachefs] bch2_replicas_gc2+0x2cb/0x400: bch2_replicas_gc2 at /home/ojab/src/bcachefs/fs/bcachefs/replicas.c:454 (discriminator 3) Modules linked in: dm_mod tun nf_conntrack_netlink nfnetlink xt_addrtype br_netfilter overlay msr sctp bcachefs lz4hc_compress lz4_compress libcrc32c xor raid6_pq lz4_decompress pps_ldisc pps_core wireguard libchacha20poly1305 chacha_x86_64 poly1305_x86_64 ip6_udp_tunnel udp_tunnel curve25519_x86_64 libcurve25519_generic libchacha sit tunnel4 ip_tunnel af_packet bridge stp llc ip6table_nat ip6table_filter ip6_tables xt_MASQUERADE xt_conntrack iptable_nat nf_nat nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 iptable_filter ip_tables x_tables tcp_bbr sch_fq_codel efivarfs nls_iso8859_1 nls_cp437 vfat fat cdc_mbim cdc_wdm cdc_ncm cdc_ether usbnet r8152 input_leds joydev mii amdgpu mousedev hid_generic usbhid hid ath10k_pci amd_atl edac_mce_amd ath10k_core kvm_amd ath kvm mac80211 bfq crc32_pclmul crc32c_intel polyval_clmulni polyval_generic sha512_ssse3 sha256_ssse3 sha1_ssse3 snd_hda_codec_generic snd_hda_codec_hdmi snd_hda_intel snd_intel_dspcfg i2c_algo_bit drm_exec snd_hda_codec r8169 drm_suballoc_helper aesni_intel gf128mul crypto_simd amdxcp realtek mfd_core tpm_crb drm_buddy snd_hwdep mdio_devres libarc4 cryptd tpm_tis wmi_bmof cfg80211 evdev libphy snd_hda_core tpm_tis_core gpu_sched rapl xhci_pci xhci_hcd snd_pcm drm_display_helper snd_timer tpm sp5100_tco rfkill efi_pstore mpt3sas drm_ttm_helper ahci usbcore libaescfb ccp snd ttm 8250 libahci watchdog soundcore raid_class sha1_generic acpi_cpufreq k10temp 8250_base usb_common scsi_transport_sas i2c_piix4 hwmon video serial_mctrl_gpio serial_base ecdh_generic wmi rtc_cmos backlight ecc gpio_amdpt rng_core gpio_generic button CPU: 5 UID: 0 PID: 6525 Comm: bcachefs Tainted: G W 6.11.0-rc1-ojab-00058-g224bc118aec9 #6 6d5debde398d2a84851f42ab300dae32c2992027 Tainted: [W]=WARN RIP: 0010:bch2_replicas_gc2+0x2cb/0x400 [bcachefs] Code: c7 c2 60 91 d1 c1 48 89 c6 48 c7 c7 98 91 d1 c1 4c 89 14 24 44 89 5c 24 08 48 89 44 24 20 c6 05 fa 68 04 00 01 e8 05 a3 40 e4 <0f> 0b 4c 8b 14 24 44 8b 5c 24 08 48 8b 44 24 20 e9 55 fe ff ff 8b RSP: 0018:ffffb434c9263d60 EFLAGS: 00010246 RAX: 0000000000000000 RBX: ffff9a8efa79cc00 RCX: 0000000000000000 RDX: 0000000000000000 RSI: 0000000000000000 RDI: 0000000000000000 RBP: ffffb434c9263de0 R08: 0000000000000000 R09: 0000000000000000 R10: 0000000000000000 R11: 0000000000000000 R12: 0000000000000005 R13: ffff9a8efa73c300 R14: ffff9a8d9e880000 R15: ffff9a8d9e8806f8 FS: 0000000000000000(0000) GS:ffff9a9410c80000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 0000565423373090 CR3: 0000000164e30000 CR4: 00000000003506f0 Call Trace: ? __warn+0x97/0x150 ? bch2_replicas_gc2+0x2cb/0x400 [bcachefs 9803eca5e131ef28f26250ede34072d5b50d98b3] bch2_replicas_gc2+0x2cb/0x400: bch2_replicas_gc2 at /home/ojab/src/bcachefs/fs/bcachefs/replicas.c:454 (discriminator 3) ? report_bug+0x196/0x1c0 ? handle_bug+0x3c/0x70 ? exc_invalid_op+0x17/0x80 ? __wake_up_klogd.part.0+0x4c/0x80 ? asm_exc_invalid_op+0x16/0x20 ? bch2_replicas_gc2+0x2cb/0x400 [bcachefs 9803eca5e131ef28f26250ede34072d5b50d98b3] bch2_replicas_gc2+0x2cb/0x400: bch2_replicas_gc2 at /home/ojab/src/bcachefs/fs/bcachefs/replicas.c:454 (discriminator 3) ? bch2_dev_usage_read+0xa0/0xa0 [bcachefs 9803eca5e131ef28f26250ede34072d5b50d98b3] bch2_dev_usage_read+0xa0/0xa0: discard_in_flight_remove at /home/ojab/src/bcachefs/fs/bcachefs/alloc_background.c:1712 Signed-off-by: Kent Overstreet --- fs/bcachefs/replicas.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/fs/bcachefs/replicas.c b/fs/bcachefs/replicas.c index 1223b710755d..12b1d28b7eb4 100644 --- a/fs/bcachefs/replicas.c +++ b/fs/bcachefs/replicas.c @@ -451,7 +451,8 @@ retry: .type = BCH_DISK_ACCOUNTING_replicas, }; - memcpy(&k.replicas, e, replicas_entry_bytes(e)); + unsafe_memcpy(&k.replicas, e, replicas_entry_bytes(e), + "embedded variable length struct"); struct bpos p = disk_accounting_pos_to_bpos(&k); -- cgit v1.2.3 From 47cdc7b14417a40af6a5d5909f1d28a5a23fc11d Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Sat, 17 Aug 2024 17:38:43 -0400 Subject: bcachefs: Fix incorrect gfp flags fixes: 00488 WARNING: CPU: 9 PID: 194 at mm/page_alloc.c:4410 __alloc_pages_noprof+0x1818/0x1888 00488 Modules linked in: 00488 CPU: 9 UID: 0 PID: 194 Comm: kworker/u66:1 Not tainted 6.11.0-rc1-ktest-g18fa10d6495f #2931 00488 Hardware name: linux,dummy-virt (DT) 00488 Workqueue: writeback wb_workfn (flush-bcachefs-2) 00488 pstate: 20001005 (nzCv daif -PAN -UAO -TCO -DIT +SSBS BTYPE=--) 00488 pc : __alloc_pages_noprof+0x1818/0x1888 00488 lr : __alloc_pages_noprof+0x5f4/0x1888 00488 sp : ffffff80ccd8ed00 00488 x29: ffffff80ccd8ed00 x28: 0000000000000000 x27: dfffffc000000000 00488 x26: 0000000000000010 x25: 0000000000000002 x24: 0000000000000000 00488 x23: 0000000000000000 x22: 1ffffff0199b1dbe x21: ffffff80cc680900 00488 x20: 0000000000000000 x19: ffffff80ccd8eed0 x18: 0000000000000000 00488 x17: ffffff80cc58a010 x16: dfffffc000000000 x15: 1ffffff00474e518 00488 x14: 1ffffff00474e518 x13: 1ffffff00474e518 x12: ffffffb8104701b9 00488 x11: 1ffffff8104701b8 x10: ffffffb8104701b8 x9 : ffffffc08043cde8 00488 x8 : 00000047efb8fe48 x7 : ffffff80ccd8ee20 x6 : 0000000000048000 00488 x5 : 1ffffff810470138 x4 : 0000000000000050 x3 : 1ffffff0199b1d94 00488 x2 : ffffffb0199b1d94 x1 : 0000000000000001 x0 : ffffffc082387448 00488 Call trace: 00488 __alloc_pages_noprof+0x1818/0x1888 00488 new_slab+0x284/0x2f0 00488 ___slab_alloc+0x208/0x8e0 00488 __kmalloc_noprof+0x328/0x340 00488 __bch2_writepage+0x106c/0x1830 00488 write_cache_pages+0xa0/0xe8 due to __GFP_NOFAIL without allowing reclaim Signed-off-by: Kent Overstreet --- fs/bcachefs/fs-io-buffered.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/fs/bcachefs/fs-io-buffered.c b/fs/bcachefs/fs-io-buffered.c index cc33d763f722..184d03851676 100644 --- a/fs/bcachefs/fs-io-buffered.c +++ b/fs/bcachefs/fs-io-buffered.c @@ -534,7 +534,7 @@ do_io: if (f_sectors > w->tmp_sectors) { kfree(w->tmp); - w->tmp = kcalloc(f_sectors, sizeof(struct bch_folio_sector), __GFP_NOFAIL); + w->tmp = kcalloc(f_sectors, sizeof(struct bch_folio_sector), GFP_NOFS|__GFP_NOFAIL); w->tmp_sectors = f_sectors; } -- cgit v1.2.3 From 2102bdac67b55bf2d1df4ff757bced74e94a5f74 Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Sat, 17 Aug 2024 21:03:07 -0400 Subject: bcachefs: Extra debug for data move path We don't have sufficient information to debug: https://github.com/koverstreet/bcachefs/issues/726 - print out durability of extent ptrs, when non default - print the number of replicas we need in data_update_to_text() Signed-off-by: Kent Overstreet --- fs/bcachefs/data_update.c | 3 +++ fs/bcachefs/extents.c | 2 ++ 2 files changed, 5 insertions(+) diff --git a/fs/bcachefs/data_update.c b/fs/bcachefs/data_update.c index 6a854c918496..1ca628e93e87 100644 --- a/fs/bcachefs/data_update.c +++ b/fs/bcachefs/data_update.c @@ -475,6 +475,9 @@ void bch2_data_update_opts_to_text(struct printbuf *out, struct bch_fs *c, bch2_compression_opt_to_text(out, background_compression(*io_opts)); prt_newline(out); + prt_str(out, "opts.replicas:\t"); + prt_u64(out, io_opts->data_replicas); + prt_str(out, "extra replicas:\t"); prt_u64(out, data_opts->extra_replicas); } diff --git a/fs/bcachefs/extents.c b/fs/bcachefs/extents.c index 4419ad3e454e..9406f82fc255 100644 --- a/fs/bcachefs/extents.c +++ b/fs/bcachefs/extents.c @@ -1017,6 +1017,8 @@ void bch2_extent_ptr_to_text(struct printbuf *out, struct bch_fs *c, const struc prt_printf(out, "ptr: %u:%llu:%u gen %u", ptr->dev, b, offset, ptr->gen); + if (ca->mi.durability != 1) + prt_printf(out, " d=%u", ca->mi.durability); if (ptr->cached) prt_str(out, " cached"); if (ptr->unwritten) -- cgit v1.2.3 From 3f53d050416e88122d53aabbadb1fede998004da Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Sun, 18 Aug 2024 12:22:23 -0400 Subject: bcachefs: bch2_data_update_init() cleanup Factor out some helpers - this function has gotten much too big. Signed-off-by: Kent Overstreet --- fs/bcachefs/data_update.c | 179 ++++++++++++++++++++++++++-------------------- 1 file changed, 101 insertions(+), 78 deletions(-) diff --git a/fs/bcachefs/data_update.c b/fs/bcachefs/data_update.c index 1ca628e93e87..5f49f4953b19 100644 --- a/fs/bcachefs/data_update.c +++ b/fs/bcachefs/data_update.c @@ -20,6 +20,76 @@ #include "subvolume.h" #include "trace.h" +static void bkey_put_dev_refs(struct bch_fs *c, struct bkey_s_c k) +{ + struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k); + + bkey_for_each_ptr(ptrs, ptr) + bch2_dev_put(bch2_dev_have_ref(c, ptr->dev)); +} + +static bool bkey_get_dev_refs(struct bch_fs *c, struct bkey_s_c k) +{ + struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k); + + bkey_for_each_ptr(ptrs, ptr) { + if (!bch2_dev_tryget(c, ptr->dev)) { + bkey_for_each_ptr(ptrs, ptr2) { + if (ptr2 == ptr) + break; + bch2_dev_put(bch2_dev_have_ref(c, ptr2->dev)); + } + return false; + } + } + return true; +} + +static void bkey_nocow_unlock(struct bch_fs *c, struct bkey_s_c k) +{ + struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k); + + bkey_for_each_ptr(ptrs, ptr) { + struct bch_dev *ca = bch2_dev_have_ref(c, ptr->dev); + struct bpos bucket = PTR_BUCKET_POS(ca, ptr); + + bch2_bucket_nocow_unlock(&c->nocow_locks, bucket, 0); + } +} + +static bool bkey_nocow_lock(struct bch_fs *c, struct moving_context *ctxt, struct bkey_s_c k) +{ + struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k); + + bkey_for_each_ptr(ptrs, ptr) { + struct bch_dev *ca = bch2_dev_have_ref(c, ptr->dev); + struct bpos bucket = PTR_BUCKET_POS(ca, ptr); + + if (ctxt) { + bool locked; + + move_ctxt_wait_event(ctxt, + (locked = bch2_bucket_nocow_trylock(&c->nocow_locks, bucket, 0)) || + list_empty(&ctxt->ios)); + + if (!locked) + bch2_bucket_nocow_lock(&c->nocow_locks, bucket, 0); + } else { + if (!bch2_bucket_nocow_trylock(&c->nocow_locks, bucket, 0)) { + bkey_for_each_ptr(ptrs, ptr2) { + if (ptr2 == ptr) + break; + + bucket = PTR_BUCKET_POS(ca, ptr2); + bch2_bucket_nocow_unlock(&c->nocow_locks, bucket, 0); + } + return false; + } + } + } + return true; +} + static void trace_move_extent_finish2(struct bch_fs *c, struct bkey_s_c k) { if (trace_move_extent_finish_enabled()) { @@ -355,17 +425,11 @@ void bch2_data_update_read_done(struct data_update *m, void bch2_data_update_exit(struct data_update *update) { struct bch_fs *c = update->op.c; - struct bkey_ptrs_c ptrs = - bch2_bkey_ptrs_c(bkey_i_to_s_c(update->k.k)); - - bkey_for_each_ptr(ptrs, ptr) { - struct bch_dev *ca = bch2_dev_have_ref(c, ptr->dev); - if (c->opts.nocow_enabled) - bch2_bucket_nocow_unlock(&c->nocow_locks, - PTR_BUCKET_POS(ca, ptr), 0); - bch2_dev_put(ca); - } + struct bkey_s_c k = bkey_i_to_s_c(update->k.k); + if (c->opts.nocow_enabled) + bkey_nocow_unlock(c, k); + bkey_put_dev_refs(c, k); bch2_bkey_buf_exit(&update->k, c); bch2_disk_reservation_put(c, &update->op.res); bch2_bio_free_pages_pool(c, &update->op.wbio.bio); @@ -546,7 +610,6 @@ int bch2_data_update_init(struct btree_trans *trans, const union bch_extent_entry *entry; struct extent_ptr_decoded p; unsigned i, reserve_sectors = k.k->size * data_opts.extra_replicas; - unsigned ptrs_locked = 0; int ret = 0; /* @@ -557,6 +620,15 @@ int bch2_data_update_init(struct btree_trans *trans, if (unlikely(k.k->p.snapshot && !bch2_snapshot_equiv(c, k.k->p.snapshot))) return -BCH_ERR_data_update_done; + if (!bkey_get_dev_refs(c, k)) + return -BCH_ERR_data_update_done; + + if (c->opts.nocow_enabled && + !bkey_nocow_lock(c, ctxt, k)) { + bkey_put_dev_refs(c, k); + return -BCH_ERR_nocow_lock_blocked; + } + bch2_bkey_buf_init(&m->k); bch2_bkey_buf_reassemble(&m->k, c, k); m->btree_id = btree_id; @@ -578,40 +650,24 @@ int bch2_data_update_init(struct btree_trans *trans, m->op.compression_opt = background_compression(io_opts); m->op.watermark = m->data_opts.btree_insert_flags & BCH_WATERMARK_MASK; - bkey_for_each_ptr(ptrs, ptr) { - if (!bch2_dev_tryget(c, ptr->dev)) { - bkey_for_each_ptr(ptrs, ptr2) { - if (ptr2 == ptr) - break; - bch2_dev_put(bch2_dev_have_ref(c, ptr2->dev)); - } - return -BCH_ERR_data_update_done; - } - } - unsigned durability_have = 0, durability_removing = 0; i = 0; bkey_for_each_ptr_decode(k.k, ptrs, p, entry) { - struct bch_dev *ca = bch2_dev_have_ref(c, p.ptr.dev); - struct bpos bucket = PTR_BUCKET_POS(ca, &p.ptr); - bool locked; - - rcu_read_lock(); - if (((1U << i) & m->data_opts.rewrite_ptrs)) { - BUG_ON(p.ptr.cached); - - if (crc_is_compressed(p.crc)) - reserve_sectors += k.k->size; - - m->op.nr_replicas += bch2_extent_ptr_desired_durability(c, &p); - durability_removing += bch2_extent_ptr_desired_durability(c, &p); - } else if (!p.ptr.cached && - !((1U << i) & m->data_opts.kill_ptrs)) { - bch2_dev_list_add_dev(&m->op.devs_have, p.ptr.dev); - durability_have += bch2_extent_ptr_durability(c, &p); + if (!p.ptr.cached) { + rcu_read_lock(); + if (BIT(i) & m->data_opts.rewrite_ptrs) { + if (crc_is_compressed(p.crc)) + reserve_sectors += k.k->size; + + m->op.nr_replicas += bch2_extent_ptr_desired_durability(c, &p); + durability_removing += bch2_extent_ptr_desired_durability(c, &p); + } else if (!(BIT(i) & m->data_opts.kill_ptrs)) { + bch2_dev_list_add_dev(&m->op.devs_have, p.ptr.dev); + durability_have += bch2_extent_ptr_durability(c, &p); + } + rcu_read_unlock(); } - rcu_read_unlock(); /* * op->csum_type is normally initialized from the fs/file's @@ -626,24 +682,6 @@ int bch2_data_update_init(struct btree_trans *trans, if (p.crc.compression_type == BCH_COMPRESSION_TYPE_incompressible) m->op.incompressible = true; - if (c->opts.nocow_enabled) { - if (ctxt) { - move_ctxt_wait_event(ctxt, - (locked = bch2_bucket_nocow_trylock(&c->nocow_locks, - bucket, 0)) || - list_empty(&ctxt->ios)); - - if (!locked) - bch2_bucket_nocow_lock(&c->nocow_locks, bucket, 0); - } else { - if (!bch2_bucket_nocow_trylock(&c->nocow_locks, bucket, 0)) { - ret = -BCH_ERR_nocow_lock_blocked; - goto err; - } - } - ptrs_locked |= (1U << i); - } - i++; } @@ -664,7 +702,7 @@ int bch2_data_update_init(struct btree_trans *trans, /* if iter == NULL, it's just a promote */ if (iter) ret = bch2_extent_drop_ptrs(trans, iter, k, m->data_opts); - goto done; + goto out; } m->op.nr_replicas = min(durability_removing, durability_required) + @@ -684,8 +722,7 @@ int bch2_data_update_init(struct btree_trans *trans, bch2_data_update_to_text(&buf, m); WARN(1, "trying to move an extent, but nr_replicas=0\n%s", buf.buf); printbuf_exit(&buf); - ret = -BCH_ERR_data_update_done; - goto done; + goto out; } m->op.nr_replicas_required = m->op.nr_replicas; @@ -696,30 +733,16 @@ int bch2_data_update_init(struct btree_trans *trans, ? 0 : BCH_DISK_RESERVATION_NOFAIL); if (ret) - goto err; + goto out; } if (bkey_extent_is_unwritten(k)) { bch2_update_unwritten_extent(trans, m); - goto done; + goto out; } return 0; -err: - i = 0; - bkey_for_each_ptr_decode(k.k, ptrs, p, entry) { - struct bch_dev *ca = bch2_dev_have_ref(c, p.ptr.dev); - struct bpos bucket = PTR_BUCKET_POS(ca, &p.ptr); - if ((1U << i) & ptrs_locked) - bch2_bucket_nocow_unlock(&c->nocow_locks, bucket, 0); - bch2_dev_put(ca); - i++; - } - - bch2_bkey_buf_exit(&m->k, c); - bch2_bio_free_pages_pool(c, &m->op.wbio.bio); - return ret; -done: +out: bch2_data_update_exit(m); return ret ?: -BCH_ERR_data_update_done; } -- cgit v1.2.3 From 8cc0e50614520c6c609c6ae32a65d0591b7865a1 Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Sun, 18 Aug 2024 13:13:39 -0400 Subject: bcachefs: Fix "trying to move an extent, but nr_replicas=0" data_update_init() does a bunch of complicated stuff to decide how many replicas to add, since we only want to increase an extent's durability on an explicit rereplicate, but extent pointers may be on devices with different durability settings. There was a corner case when evacuating a device that had been set to durability=0 after data had been written to it, and extents on that device had already been rereplicated - then evacuate only needs to drop pointers on that device, not move them. So the assert for !m->op.nr_replicas was spurious; this was a perfectly legitimate case that needed to be handled. Signed-off-by: Kent Overstreet --- fs/bcachefs/data_update.c | 29 ++++++++++++----------------- 1 file changed, 12 insertions(+), 17 deletions(-) diff --git a/fs/bcachefs/data_update.c b/fs/bcachefs/data_update.c index 5f49f4953b19..65176d51b502 100644 --- a/fs/bcachefs/data_update.c +++ b/fs/bcachefs/data_update.c @@ -695,16 +695,6 @@ int bch2_data_update_init(struct btree_trans *trans, * Increasing replication is an explicit operation triggered by * rereplicate, currently, so that users don't get an unexpected -ENOSPC */ - if (!(m->data_opts.write_flags & BCH_WRITE_CACHED) && - !durability_required) { - m->data_opts.kill_ptrs |= m->data_opts.rewrite_ptrs; - m->data_opts.rewrite_ptrs = 0; - /* if iter == NULL, it's just a promote */ - if (iter) - ret = bch2_extent_drop_ptrs(trans, iter, k, m->data_opts); - goto out; - } - m->op.nr_replicas = min(durability_removing, durability_required) + m->data_opts.extra_replicas; @@ -716,17 +706,22 @@ int bch2_data_update_init(struct btree_trans *trans, if (!(durability_have + durability_removing)) m->op.nr_replicas = max((unsigned) m->op.nr_replicas, 1); - if (!m->op.nr_replicas) { - struct printbuf buf = PRINTBUF; + m->op.nr_replicas_required = m->op.nr_replicas; - bch2_data_update_to_text(&buf, m); - WARN(1, "trying to move an extent, but nr_replicas=0\n%s", buf.buf); - printbuf_exit(&buf); + /* + * It might turn out that we don't need any new replicas, if the + * replicas or durability settings have been changed since the extent + * was written: + */ + if (!m->op.nr_replicas) { + m->data_opts.kill_ptrs |= m->data_opts.rewrite_ptrs; + m->data_opts.rewrite_ptrs = 0; + /* if iter == NULL, it's just a promote */ + if (iter) + ret = bch2_extent_drop_ptrs(trans, iter, k, m->data_opts); goto out; } - m->op.nr_replicas_required = m->op.nr_replicas; - if (reserve_sectors) { ret = bch2_disk_reservation_add(c, &m->op.res, reserve_sectors, m->data_opts.extra_replicas -- cgit v1.2.3 From 548e7f51679bf0ec3cdc2027d780c5d06a2a7ac6 Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Sun, 18 Aug 2024 13:24:26 -0400 Subject: bcachefs: setting bcachefs_effective.* xattrs is a noop bcachefs_effective.* xattrs show the options inherited from parent directories (as well as explicitly set); this namespace is not for setting bcachefs options. Change the .set() handler to a noop so that if e.g. rsync is copying xattrs it'll do the right thing, and only copy xattrs in the bcachefs.* namespace. We don't want to return an error, because that will cause rsync to bail out or get spammy. Signed-off-by: Kent Overstreet --- fs/bcachefs/xattr.c | 12 +++++++++++- 1 file changed, 11 insertions(+), 1 deletion(-) diff --git a/fs/bcachefs/xattr.c b/fs/bcachefs/xattr.c index f2b4c17a0307..331f944d73dc 100644 --- a/fs/bcachefs/xattr.c +++ b/fs/bcachefs/xattr.c @@ -612,10 +612,20 @@ static int bch2_xattr_bcachefs_get_effective( name, buffer, size, true); } +/* Noop - xattrs in the bcachefs_effective namespace are inherited */ +static int bch2_xattr_bcachefs_set_effective(const struct xattr_handler *handler, + struct mnt_idmap *idmap, + struct dentry *dentry, struct inode *vinode, + const char *name, const void *value, + size_t size, int flags) +{ + return 0; +} + static const struct xattr_handler bch_xattr_bcachefs_effective_handler = { .prefix = "bcachefs_effective.", .get = bch2_xattr_bcachefs_get_effective, - .set = bch2_xattr_bcachefs_set, + .set = bch2_xattr_bcachefs_set_effective, }; #endif /* NO_BCACHEFS_FS */ -- cgit v1.2.3 From 49203a6b9d12bfd1a223a67847a631a78f1cd782 Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Sun, 18 Aug 2024 15:08:12 -0400 Subject: bcachefs: Fix failure to relock in btree_node_get() discovered by new trans->locked asserts Signed-off-by: Kent Overstreet --- fs/bcachefs/btree_cache.c | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/fs/bcachefs/btree_cache.c b/fs/bcachefs/btree_cache.c index f5d85b50b6f2..cc778d7e769e 100644 --- a/fs/bcachefs/btree_cache.c +++ b/fs/bcachefs/btree_cache.c @@ -974,6 +974,10 @@ retry: bch2_btree_node_wait_on_read(b); + ret = bch2_trans_relock(trans); + if (ret) + return ERR_PTR(ret); + /* * should_be_locked is not set on this path yet, so we need to * relock it specifically: -- cgit v1.2.3 From e150a7e89c4727176d07f5a0a8966fc2af05821c Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Sun, 18 Aug 2024 20:18:34 -0400 Subject: bcachefs: Fix bch2_trigger_alloc assert On testing on an old mangled filesystem, we missed a case. Fixes: bd864bc2d907 ("bcachefs: Fix bch2_trigger_alloc when upgrading from old versions") Signed-off-by: Kent Overstreet --- fs/bcachefs/alloc_background.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/fs/bcachefs/alloc_background.c b/fs/bcachefs/alloc_background.c index fd3a2522bc3e..488d0710f7b9 100644 --- a/fs/bcachefs/alloc_background.c +++ b/fs/bcachefs/alloc_background.c @@ -829,7 +829,7 @@ int bch2_trigger_alloc(struct btree_trans *trans, if (likely(new.k->type == KEY_TYPE_alloc_v4)) { new_a = bkey_s_to_alloc_v4(new).v; } else { - BUG_ON(!(flags & BTREE_TRIGGER_gc)); + BUG_ON(!(flags & (BTREE_TRIGGER_gc|BTREE_TRIGGER_check_repair))); struct bkey_i_alloc_v4 *new_ka = bch2_alloc_to_v4_mut_inlined(trans, new.s_c); ret = PTR_ERR_OR_ZERO(new_ka); -- cgit v1.2.3 From c2a503f3e98e191d86738f5438a3a2b69575c830 Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Sun, 18 Aug 2024 20:38:49 -0400 Subject: bcachefs: Fix bch2_bucket_gens_init() Comparing the wrong bpos - this was missed because normally bucket_gens_init() runs on brand new filesystems, but this bug caused it to overwrite bucket_gens keys with 0s when upgrading ancient filesystems. Signed-off-by: Kent Overstreet --- fs/bcachefs/alloc_background.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/fs/bcachefs/alloc_background.c b/fs/bcachefs/alloc_background.c index 488d0710f7b9..ac933142aeda 100644 --- a/fs/bcachefs/alloc_background.c +++ b/fs/bcachefs/alloc_background.c @@ -556,7 +556,7 @@ int bch2_bucket_gens_init(struct bch_fs *c) struct bpos pos = alloc_gens_pos(iter.pos, &offset); int ret2 = 0; - if (have_bucket_gens_key && bkey_cmp(iter.pos, pos)) { + if (have_bucket_gens_key && !bkey_eq(g.k.p, pos)) { ret2 = bch2_btree_insert_trans(trans, BTREE_ID_bucket_gens, &g.k_i, 0) ?: bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc); if (ret2) -- cgit v1.2.3 From b8db1bd8020d5fecb3bf46cd8b954a657c20ba14 Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Mon, 19 Aug 2024 16:13:16 -0400 Subject: bcachefs: fix time_stats_to_text() Fixes: 7423330e30ab ("bcachefs: prt_printf() now respects \r\n\t") Signed-off-by: Kent Overstreet --- fs/bcachefs/util.c | 1 - 1 file changed, 1 deletion(-) diff --git a/fs/bcachefs/util.c b/fs/bcachefs/util.c index 138320eaa2ad..1b8554460af4 100644 --- a/fs/bcachefs/util.c +++ b/fs/bcachefs/util.c @@ -416,7 +416,6 @@ void bch2_time_stats_to_text(struct printbuf *out, struct bch2_time_stats *stats printbuf_tabstop_push(out, TABSTOP_SIZE + 2); prt_printf(out, "\tsince mount\r\trecent\r\n"); - prt_printf(out, "recent"); printbuf_tabstops_reset(out); printbuf_tabstop_push(out, out->indent + 20); -- cgit v1.2.3 From cecc328240609df17395dfd0ea03cc813d8be36d Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Tue, 20 Aug 2024 11:25:39 -0400 Subject: bcachefs: fix missing bch2_err_str() Signed-off-by: Kent Overstreet --- fs/bcachefs/fsck.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/fs/bcachefs/fsck.c b/fs/bcachefs/fsck.c index 9138944c5ae6..267c2336b115 100644 --- a/fs/bcachefs/fsck.c +++ b/fs/bcachefs/fsck.c @@ -2469,8 +2469,7 @@ static int check_path(struct btree_trans *trans, pathbuf *p, struct bkey_s_c ino : bch2_inode_unpack(inode_k, &inode); if (ret) { /* Should have been caught in dirents pass */ - if (!bch2_err_matches(ret, BCH_ERR_transaction_restart)) - bch_err(c, "error looking up parent directory: %i", ret); + bch_err_msg(c, ret, "error looking up parent directory"); break; } -- cgit v1.2.3 From 1dceae4cc12aa6389d9a8706f0d2a94d1679e79d Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Tue, 20 Aug 2024 12:10:33 -0400 Subject: bcachefs: unlock_long() before resort in journal replay Fix another SRCU splat - this one pretty harmless. Signed-off-by: Kent Overstreet --- fs/bcachefs/recovery.c | 1 + 1 file changed, 1 insertion(+) diff --git a/fs/bcachefs/recovery.c b/fs/bcachefs/recovery.c index d89eb43c5ce9..11368dfa96b2 100644 --- a/fs/bcachefs/recovery.c +++ b/fs/bcachefs/recovery.c @@ -322,6 +322,7 @@ int bch2_journal_replay(struct bch_fs *c) } } + bch2_trans_unlock_long(trans); /* * Now, replay any remaining keys in the order in which they appear in * the journal, unpinning those journal entries as we go: -- cgit v1.2.3 From 3c5d0b72a8e8c19c960e8fefb7463067e58b6bc4 Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Mon, 19 Aug 2024 15:22:55 -0400 Subject: bcachefs: fix failure to relock in bch2_btree_node_mem_alloc() We weren't always so strict about trans->locked state - but now we are, and new assertions are shaking some bugs out. Signed-off-by: Kent Overstreet --- fs/bcachefs/btree_cache.c | 17 ++++++++++++++ fs/bcachefs/btree_cache.h | 2 ++ fs/bcachefs/btree_update_interior.c | 46 ++++++++++++++++++++----------------- 3 files changed, 44 insertions(+), 21 deletions(-) diff --git a/fs/bcachefs/btree_cache.c b/fs/bcachefs/btree_cache.c index cc778d7e769e..063725ecb2b3 100644 --- a/fs/bcachefs/btree_cache.c +++ b/fs/bcachefs/btree_cache.c @@ -159,6 +159,16 @@ struct btree *__bch2_btree_node_mem_alloc(struct bch_fs *c) return b; } +void bch2_btree_node_to_freelist(struct bch_fs *c, struct btree *b) +{ + mutex_lock(&c->btree_cache.lock); + list_move(&b->list, &c->btree_cache.freeable); + mutex_unlock(&c->btree_cache.lock); + + six_unlock_write(&b->c.lock); + six_unlock_intent(&b->c.lock); +} + /* Btree in memory cache - hash table */ void bch2_btree_node_hash_remove(struct btree_cache *bc, struct btree *b) @@ -736,6 +746,13 @@ out: start_time); memalloc_nofs_restore(flags); + + int ret = bch2_trans_relock(trans); + if (unlikely(ret)) { + bch2_btree_node_to_freelist(c, b); + return ERR_PTR(ret); + } + return b; err: mutex_lock(&bc->lock); diff --git a/fs/bcachefs/btree_cache.h b/fs/bcachefs/btree_cache.h index c0eb87a057cc..f82064007127 100644 --- a/fs/bcachefs/btree_cache.h +++ b/fs/bcachefs/btree_cache.h @@ -12,6 +12,8 @@ struct btree_iter; void bch2_recalc_btree_reserve(struct bch_fs *); +void bch2_btree_node_to_freelist(struct bch_fs *, struct btree *); + void bch2_btree_node_hash_remove(struct btree_cache *, struct btree *); int __bch2_btree_node_hash_insert(struct btree_cache *, struct btree *); int bch2_btree_node_hash_insert(struct btree_cache *, struct btree *, diff --git a/fs/bcachefs/btree_update_interior.c b/fs/bcachefs/btree_update_interior.c index b3454d4619e8..8fd112026e7a 100644 --- a/fs/bcachefs/btree_update_interior.c +++ b/fs/bcachefs/btree_update_interior.c @@ -317,6 +317,12 @@ static struct btree *__bch2_btree_node_alloc(struct btree_trans *trans, : 0; int ret; + b = bch2_btree_node_mem_alloc(trans, interior_node); + if (IS_ERR(b)) + return b; + + BUG_ON(b->ob.nr); + mutex_lock(&c->btree_reserve_cache_lock); if (c->btree_reserve_cache_nr > nr_reserve) { struct btree_alloc *a = @@ -325,10 +331,9 @@ static struct btree *__bch2_btree_node_alloc(struct btree_trans *trans, obs = a->ob; bkey_copy(&tmp.k, &a->k); mutex_unlock(&c->btree_reserve_cache_lock); - goto mem_alloc; + goto out; } mutex_unlock(&c->btree_reserve_cache_lock); - retry: ret = bch2_alloc_sectors_start_trans(trans, c->opts.metadata_target ?: @@ -341,7 +346,7 @@ retry: c->opts.metadata_replicas_required), watermark, 0, cl, &wp); if (unlikely(ret)) - return ERR_PTR(ret); + goto err; if (wp->sectors_free < btree_sectors(c)) { struct open_bucket *ob; @@ -360,19 +365,16 @@ retry: bch2_open_bucket_get(c, wp, &obs); bch2_alloc_sectors_done(c, wp); -mem_alloc: - b = bch2_btree_node_mem_alloc(trans, interior_node); - six_unlock_write(&b->c.lock); - six_unlock_intent(&b->c.lock); - - /* we hold cannibalize_lock: */ - BUG_ON(IS_ERR(b)); - BUG_ON(b->ob.nr); - +out: bkey_copy(&b->key, &tmp.k); b->ob = obs; + six_unlock_write(&b->c.lock); + six_unlock_intent(&b->c.lock); return b; +err: + bch2_btree_node_to_freelist(c, b); + return ERR_PTR(ret); } static struct btree *bch2_btree_node_alloc(struct btree_update *as, @@ -2439,6 +2441,9 @@ int bch2_btree_node_update_key(struct btree_trans *trans, struct btree_iter *ite } new_hash = bch2_btree_node_mem_alloc(trans, false); + ret = PTR_ERR_OR_ZERO(new_hash); + if (ret) + goto err; } path->intent_ref++; @@ -2446,14 +2451,9 @@ int bch2_btree_node_update_key(struct btree_trans *trans, struct btree_iter *ite commit_flags, skip_triggers); --path->intent_ref; - if (new_hash) { - mutex_lock(&c->btree_cache.lock); - list_move(&new_hash->list, &c->btree_cache.freeable); - mutex_unlock(&c->btree_cache.lock); - - six_unlock_write(&new_hash->c.lock); - six_unlock_intent(&new_hash->c.lock); - } + if (new_hash) + bch2_btree_node_to_freelist(c, new_hash); +err: closure_sync(&cl); bch2_btree_cache_cannibalize_unlock(trans); return ret; @@ -2522,6 +2522,10 @@ int bch2_btree_root_alloc_fake_trans(struct btree_trans *trans, enum btree_id id b = bch2_btree_node_mem_alloc(trans, false); bch2_btree_cache_cannibalize_unlock(trans); + ret = PTR_ERR_OR_ZERO(b); + if (ret) + return ret; + set_btree_node_fake(b); set_btree_node_need_rewrite(b); b->c.level = level; @@ -2553,7 +2557,7 @@ int bch2_btree_root_alloc_fake_trans(struct btree_trans *trans, enum btree_id id void bch2_btree_root_alloc_fake(struct bch_fs *c, enum btree_id id, unsigned level) { - bch2_trans_run(c, bch2_btree_root_alloc_fake_trans(trans, id, level)); + bch2_trans_run(c, lockrestart_do(trans, bch2_btree_root_alloc_fake_trans(trans, id, level))); } static void bch2_btree_update_to_text(struct printbuf *out, struct btree_update *as) -- cgit v1.2.3 From 5dbfc4ef72f15508882aff58c307b8425cf037a8 Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Tue, 20 Aug 2024 15:04:15 -0400 Subject: bcachefs: fix failure to relock in btree_node_fill() Signed-off-by: Kent Overstreet --- fs/bcachefs/btree_cache.c | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/fs/bcachefs/btree_cache.c b/fs/bcachefs/btree_cache.c index 063725ecb2b3..e52a06d3418c 100644 --- a/fs/bcachefs/btree_cache.c +++ b/fs/bcachefs/btree_cache.c @@ -873,6 +873,10 @@ static noinline struct btree *bch2_btree_node_fill(struct btree_trans *trans, bch2_btree_node_read(trans, b, sync); + int ret = bch2_trans_relock(trans); + if (ret) + return ERR_PTR(ret); + if (!sync) return NULL; -- cgit v1.2.3 From 6575b8c9877c3dd1f7db1d0d61bd250a0bf18b6d Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Tue, 20 Aug 2024 19:31:20 -0400 Subject: bcachefs: Fix locking in bch2_ioc_setlabel() Fixes: 7a254053a590 ("bcachefs: support FS_IOC_SETFSLABEL") Reported-by: syzbot+7e9efdfec27fbde0141d@syzkaller.appspotmail.com Signed-off-by: Kent Overstreet --- fs/bcachefs/fs-ioctl.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/fs/bcachefs/fs-ioctl.c b/fs/bcachefs/fs-ioctl.c index aea8132d2c40..99c7fe987c74 100644 --- a/fs/bcachefs/fs-ioctl.c +++ b/fs/bcachefs/fs-ioctl.c @@ -328,9 +328,8 @@ static int bch2_ioc_setlabel(struct bch_fs *c, mutex_lock(&c->sb_lock); strscpy(c->disk_sb.sb->label, label, BCH_SB_LABEL_SIZE); - mutex_unlock(&c->sb_lock); - ret = bch2_write_super(c); + mutex_unlock(&c->sb_lock); mnt_drop_write_file(file); return ret; -- cgit v1.2.3 From cab18be6957b6af8cbe3502fd5f6d7b9f02ccceb Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Wed, 21 Aug 2024 20:49:07 -0400 Subject: bcachefs: Fix replay_now_at() assert Journal replay, in the slowpath where we insert keys in journal order, was inserting keys in the wrong order; keys from early repair come last. Reported-by: syzbot+2c4fcb257ce2b6a29d0e@syzkaller.appspotmail.com Signed-off-by: Kent Overstreet --- fs/bcachefs/recovery.c | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-) diff --git a/fs/bcachefs/recovery.c b/fs/bcachefs/recovery.c index 11368dfa96b2..36de1c6fe8c3 100644 --- a/fs/bcachefs/recovery.c +++ b/fs/bcachefs/recovery.c @@ -241,7 +241,13 @@ static int journal_sort_seq_cmp(const void *_l, const void *_r) const struct journal_key *l = *((const struct journal_key **)_l); const struct journal_key *r = *((const struct journal_key **)_r); - return cmp_int(l->journal_seq, r->journal_seq); + /* + * Map 0 to U64_MAX, so that keys with journal_seq === 0 come last + * + * journal_seq == 0 means that the key comes from early repair, and + * should be inserted last so as to avoid overflowing the journal + */ + return cmp_int(l->journal_seq - 1, r->journal_seq - 1); } int bch2_journal_replay(struct bch_fs *c) -- cgit v1.2.3 From bdbdd4759f081ca2d0a5d9e8af21d742ffaf8439 Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Wed, 21 Aug 2024 21:10:45 -0400 Subject: bcachefs: Fix missing validation in bch2_sb_journal_v2_validate() Reported-by: syzbot+47ecc948aadfb2ab3efc@syzkaller.appspotmail.com Signed-off-by: Kent Overstreet --- fs/bcachefs/journal_sb.c | 15 +++++++++++++++ 1 file changed, 15 insertions(+) diff --git a/fs/bcachefs/journal_sb.c b/fs/bcachefs/journal_sb.c index db80e506e3ab..62b910f2fb27 100644 --- a/fs/bcachefs/journal_sb.c +++ b/fs/bcachefs/journal_sb.c @@ -104,6 +104,7 @@ static int bch2_sb_journal_v2_validate(struct bch_sb *sb, struct bch_sb_field *f struct bch_sb_field_journal_v2 *journal = field_to_type(f, journal_v2); struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx); int ret = -BCH_ERR_invalid_sb_journal; + u64 sum = 0; unsigned nr; unsigned i; struct u64_range *b; @@ -119,6 +120,15 @@ static int bch2_sb_journal_v2_validate(struct bch_sb *sb, struct bch_sb_field *f for (i = 0; i < nr; i++) { b[i].start = le64_to_cpu(journal->d[i].start); b[i].end = b[i].start + le64_to_cpu(journal->d[i].nr); + + if (b[i].end <= b[i].start) { + prt_printf(err, "journal buckets entry with bad nr: %llu+%llu", + le64_to_cpu(journal->d[i].start), + le64_to_cpu(journal->d[i].nr)); + goto err; + } + + sum += le64_to_cpu(journal->d[i].nr); } sort(b, nr, sizeof(*b), u64_range_cmp, NULL); @@ -148,6 +158,11 @@ static int bch2_sb_journal_v2_validate(struct bch_sb *sb, struct bch_sb_field *f } } + if (sum > UINT_MAX) { + prt_printf(err, "too many journal buckets: %llu > %u", sum, UINT_MAX); + goto err; + } + ret = 0; err: kfree(b); -- cgit v1.2.3 From 06f67437ab356e3140f51aea272d33ce28421f66 Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Wed, 21 Aug 2024 22:06:44 -0400 Subject: fs/super.c: improve get_tree() error message seeing an odd bug where we fail to correctly return an error from .get_tree(): https://syzkaller.appspot.com/bug?extid=c0360e8367d6d8d04a66 we need to be able to distinguish between accidently returning a positive error (as implied by the log) and no error. Cc: David Howells Signed-off-by: Kent Overstreet --- fs/super.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/fs/super.c b/fs/super.c index 38d72a3cf6fc..b7913b55debc 100644 --- a/fs/super.c +++ b/fs/super.c @@ -1802,8 +1802,8 @@ int vfs_get_tree(struct fs_context *fc) return error; if (!fc->root) { - pr_err("Filesystem %s get_tree() didn't set fc->root\n", - fc->fs_type->name); + pr_err("Filesystem %s get_tree() didn't set fc->root, returned %i\n", + fc->fs_type->name, error); /* We don't know what the locking state of the superblock is - * if there is a superblock. */ -- cgit v1.2.3 From 7f2de6947f92cfa4be8e5eaa1237e962bb8ee65f Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Wed, 21 Aug 2024 22:27:45 -0400 Subject: bcachefs: Fix warning in bch2_fs_journal_stop() j->last_empty_seq needs to match j->seq when the journal is empty Reported-by: syzbot+4093905737cf289b6b38@syzkaller.appspotmail.com Signed-off-by: Kent Overstreet --- fs/bcachefs/journal.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/fs/bcachefs/journal.c b/fs/bcachefs/journal.c index 649e3a01608a..f5f7db50ca31 100644 --- a/fs/bcachefs/journal.c +++ b/fs/bcachefs/journal.c @@ -1260,7 +1260,7 @@ int bch2_fs_journal_start(struct journal *j, u64 cur_seq) } if (!had_entries) - j->last_empty_seq = cur_seq; + j->last_empty_seq = cur_seq - 1; /* to match j->seq */ spin_lock(&j->lock); -- cgit v1.2.3 From 8ed823b19214e403ca485532f48c0e02035021ae Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Wed, 21 Aug 2024 22:57:56 -0400 Subject: bcachefs: Fix compat issue with old alloc_v4 keys we allow new fields to be added to existing key types, and new versions should treat them as being zeroed; this was not handled in alloc_v4_validate. Reported-by: syzbot+3b2968fa4953885dd66a@syzkaller.appspotmail.com Signed-off-by: Kent Overstreet --- fs/bcachefs/alloc_background.c | 50 ++++++++++++++++++----------------- fs/bcachefs/alloc_background_format.h | 1 + fs/bcachefs/btree_iter.h | 9 +++++++ 3 files changed, 36 insertions(+), 24 deletions(-) diff --git a/fs/bcachefs/alloc_background.c b/fs/bcachefs/alloc_background.c index ac933142aeda..dc97b1f8bc08 100644 --- a/fs/bcachefs/alloc_background.c +++ b/fs/bcachefs/alloc_background.c @@ -240,71 +240,73 @@ fsck_err: int bch2_alloc_v4_validate(struct bch_fs *c, struct bkey_s_c k, enum bch_validate_flags flags) { - struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k); + struct bch_alloc_v4 a; int ret = 0; - bkey_fsck_err_on(alloc_v4_u64s_noerror(a.v) > bkey_val_u64s(k.k), + bkey_val_copy(&a, bkey_s_c_to_alloc_v4(k)); + + bkey_fsck_err_on(alloc_v4_u64s_noerror(&a) > bkey_val_u64s(k.k), c, alloc_v4_val_size_bad, "bad val size (%u > %zu)", - alloc_v4_u64s_noerror(a.v), bkey_val_u64s(k.k)); + alloc_v4_u64s_noerror(&a), bkey_val_u64s(k.k)); - bkey_fsck_err_on(!BCH_ALLOC_V4_BACKPOINTERS_START(a.v) && - BCH_ALLOC_V4_NR_BACKPOINTERS(a.v), + bkey_fsck_err_on(!BCH_ALLOC_V4_BACKPOINTERS_START(&a) && + BCH_ALLOC_V4_NR_BACKPOINTERS(&a), c, alloc_v4_backpointers_start_bad, "invalid backpointers_start"); - bkey_fsck_err_on(alloc_data_type(*a.v, a.v->data_type) != a.v->data_type, + bkey_fsck_err_on(alloc_data_type(a, a.data_type) != a.data_type, c, alloc_key_data_type_bad, "invalid data type (got %u should be %u)", - a.v->data_type, alloc_data_type(*a.v, a.v->data_type)); + a.data_type, alloc_data_type(a, a.data_type)); for (unsigned i = 0; i < 2; i++) - bkey_fsck_err_on(a.v->io_time[i] > LRU_TIME_MAX, + bkey_fsck_err_on(a.io_time[i] > LRU_TIME_MAX, c, alloc_key_io_time_bad, "invalid io_time[%s]: %llu, max %llu", i == READ ? "read" : "write", - a.v->io_time[i], LRU_TIME_MAX); + a.io_time[i], LRU_TIME_MAX); - unsigned stripe_sectors = BCH_ALLOC_V4_BACKPOINTERS_START(a.v) * sizeof(u64) > + unsigned stripe_sectors = BCH_ALLOC_V4_BACKPOINTERS_START(&a) * sizeof(u64) > offsetof(struct bch_alloc_v4, stripe_sectors) - ? a.v->stripe_sectors + ? a.stripe_sectors : 0; - switch (a.v->data_type) { + switch (a.data_type) { case BCH_DATA_free: case BCH_DATA_need_gc_gens: case BCH_DATA_need_discard: bkey_fsck_err_on(stripe_sectors || - a.v->dirty_sectors || - a.v->cached_sectors || - a.v->stripe, + a.dirty_sectors || + a.cached_sectors || + a.stripe, c, alloc_key_empty_but_have_data, "empty data type free but have data %u.%u.%u %u", stripe_sectors, - a.v->dirty_sectors, - a.v->cached_sectors, - a.v->stripe); + a.dirty_sectors, + a.cached_sectors, + a.stripe); break; case BCH_DATA_sb: case BCH_DATA_journal: case BCH_DATA_btree: case BCH_DATA_user: case BCH_DATA_parity: - bkey_fsck_err_on(!a.v->dirty_sectors && + bkey_fsck_err_on(!a.dirty_sectors && !stripe_sectors, c, alloc_key_dirty_sectors_0, "data_type %s but dirty_sectors==0", - bch2_data_type_str(a.v->data_type)); + bch2_data_type_str(a.data_type)); break; case BCH_DATA_cached: - bkey_fsck_err_on(!a.v->cached_sectors || - a.v->dirty_sectors || + bkey_fsck_err_on(!a.cached_sectors || + a.dirty_sectors || stripe_sectors || - a.v->stripe, + a.stripe, c, alloc_key_cached_inconsistency, "data type inconsistency"); - bkey_fsck_err_on(!a.v->io_time[READ] && + bkey_fsck_err_on(!a.io_time[READ] && c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_to_lru_refs, c, alloc_key_cached_but_read_time_zero, "cached bucket with read_time == 0"); diff --git a/fs/bcachefs/alloc_background_format.h b/fs/bcachefs/alloc_background_format.h index 47d9d006502c..f754a2951d8a 100644 --- a/fs/bcachefs/alloc_background_format.h +++ b/fs/bcachefs/alloc_background_format.h @@ -69,6 +69,7 @@ struct bch_alloc_v4 { __u64 io_time[2]; __u32 stripe; __u32 nr_external_backpointers; + /* end of fields in original version of alloc_v4 */ __u64 fragmentation_lru; __u32 stripe_sectors; __u32 pad; diff --git a/fs/bcachefs/btree_iter.h b/fs/bcachefs/btree_iter.h index dca62375d7d3..222b7ce8a901 100644 --- a/fs/bcachefs/btree_iter.h +++ b/fs/bcachefs/btree_iter.h @@ -569,6 +569,15 @@ static inline struct bkey_s_c bch2_bkey_get_iter(struct btree_trans *trans, bkey_s_c_to_##_type(__bch2_bkey_get_iter(_trans, _iter, \ _btree_id, _pos, _flags, KEY_TYPE_##_type)) +#define bkey_val_copy(_dst_v, _src_k) \ +do { \ + unsigned b = min_t(unsigned, sizeof(*_dst_v), \ + bkey_val_bytes(_src_k.k)); \ + memcpy(_dst_v, _src_k.v, b); \ + if (b < sizeof(*_dst_v)) \ + memset((void *) (_dst_v) + b, 0, sizeof(*_dst_v) - b); \ +} while (0) + static inline int __bch2_bkey_get_val_typed(struct btree_trans *trans, unsigned btree_id, struct bpos pos, unsigned flags, unsigned type, -- cgit v1.2.3 From 0b50b7313ef2494926df30ce8e2ce284f1b847fc Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Wed, 21 Aug 2024 23:21:52 -0400 Subject: bcachefs: Fix refcounting in discard path bch_dev->io_ref does not protect against the filesystem going away; bch_fs->writes does. Thus the filesystem write ref needs to be the last ref we release. Reported-by: syzbot+9e0404b505e604f67e41@syzkaller.appspotmail.com Signed-off-by: Kent Overstreet --- fs/bcachefs/alloc_background.c | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) diff --git a/fs/bcachefs/alloc_background.c b/fs/bcachefs/alloc_background.c index dc97b1f8bc08..ba46f1c1d78a 100644 --- a/fs/bcachefs/alloc_background.c +++ b/fs/bcachefs/alloc_background.c @@ -1874,26 +1874,26 @@ static void bch2_do_discards_work(struct work_struct *work) trace_discard_buckets(c, s.seen, s.open, s.need_journal_commit, s.discarded, bch2_err_str(ret)); - bch2_write_ref_put(c, BCH_WRITE_REF_discard); percpu_ref_put(&ca->io_ref); + bch2_write_ref_put(c, BCH_WRITE_REF_discard); } void bch2_dev_do_discards(struct bch_dev *ca) { struct bch_fs *c = ca->fs; - if (!bch2_dev_get_ioref(c, ca->dev_idx, WRITE)) + if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_discard)) return; - if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_discard)) - goto put_ioref; + if (!bch2_dev_get_ioref(c, ca->dev_idx, WRITE)) + goto put_write_ref; if (queue_work(c->write_ref_wq, &ca->discard_work)) return; - bch2_write_ref_put(c, BCH_WRITE_REF_discard); -put_ioref: percpu_ref_put(&ca->io_ref); +put_write_ref: + bch2_write_ref_put(c, BCH_WRITE_REF_discard); } void bch2_do_discards(struct bch_fs *c) -- cgit v1.2.3 From dedb2fe37574857c84e9598b9f5272505dedf7af Mon Sep 17 00:00:00 2001 From: Yuesong Li Date: Thu, 22 Aug 2024 14:21:58 +0800 Subject: bcachefs: Fix double assignment in check_dirent_to_subvol() ret was assigned twice in check_dirent_to_subvol(). Reported by cocci. Signed-off-by: Yuesong Li Signed-off-by: Kent Overstreet --- fs/bcachefs/fsck.c | 1 - 1 file changed, 1 deletion(-) diff --git a/fs/bcachefs/fsck.c b/fs/bcachefs/fsck.c index 267c2336b115..6801c37ee803 100644 --- a/fs/bcachefs/fsck.c +++ b/fs/bcachefs/fsck.c @@ -2006,7 +2006,6 @@ static int check_dirent_to_subvol(struct btree_trans *trans, struct btree_iter * if (ret) { bch_err(c, "subvol %u points to missing inode root %llu", target_subvol, target_inum); ret = -BCH_ERR_fsck_repair_unimplemented; - ret = 0; goto err; } -- cgit v1.2.3 From 87313ac1f134d6ee1e7c858da8bdea9147b537a9 Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Thu, 22 Aug 2024 02:13:02 -0400 Subject: bcachefs: clear path->should_be_locked in bch2_btree_key_cache_drop() bch2_btree_key_cache_drop() evicts the key cache entry - it's used when we're doing an update that bypasses the key cache, because for cache coherency reasons a key can't be in the key cache unless it also exists in the btree - i.e. creates have to bypass the cache. After evicting, the path no longer points to a key cache key, and relock() will always fail if should_be_locked is true. Prep for improving path->should_be_locked assertions Signed-off-by: Kent Overstreet --- fs/bcachefs/btree_key_cache.c | 1 + 1 file changed, 1 insertion(+) diff --git a/fs/bcachefs/btree_key_cache.c b/fs/bcachefs/btree_key_cache.c index 79954490627c..9b3ec2a3b8ce 100644 --- a/fs/bcachefs/btree_key_cache.c +++ b/fs/bcachefs/btree_key_cache.c @@ -726,6 +726,7 @@ void bch2_btree_key_cache_drop(struct btree_trans *trans, mark_btree_node_locked(trans, path, 0, BTREE_NODE_UNLOCKED); btree_path_set_dirty(path, BTREE_ITER_NEED_TRAVERSE); + path->should_be_locked = false; } static unsigned long bch2_btree_key_cache_scan(struct shrinker *shrink, -- cgit v1.2.3 From 3e878fe5a0b139838a65f50a3df3caf3299dbc24 Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Thu, 22 Aug 2024 03:57:39 -0400 Subject: bcachefs: add missing inode_walker_exit() fix a small leak Signed-off-by: Kent Overstreet --- fs/bcachefs/fsck.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/fs/bcachefs/fsck.c b/fs/bcachefs/fsck.c index 6801c37ee803..83bd31b44aad 100644 --- a/fs/bcachefs/fsck.c +++ b/fs/bcachefs/fsck.c @@ -2215,6 +2215,8 @@ int bch2_check_xattrs(struct bch_fs *c) NULL, NULL, BCH_TRANS_COMMIT_no_enospc, check_xattr(trans, &iter, k, &hash_info, &inode))); + + inode_walker_exit(&inode); bch_err_fn(c, ret); return ret; } -- cgit v1.2.3 From a592cdf5164d3feb821085df71f63e70e8b8b08c Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Mon, 19 Aug 2024 16:41:00 -0400 Subject: bcachefs: don't use rht_bucket() in btree_key_cache_scan() rht_bucket() does strange complicated things when a rehash is in progress. Instead, just skip scanning when a rehash is in progress: scanning is going to be more expensive (many more empty slots to cover), and some sort of infinite loop is being observed Signed-off-by: Kent Overstreet --- fs/bcachefs/btree_key_cache.c | 30 +++++++++++++++++++++++++++--- 1 file changed, 27 insertions(+), 3 deletions(-) diff --git a/fs/bcachefs/btree_key_cache.c b/fs/bcachefs/btree_key_cache.c index 9b3ec2a3b8ce..fda7998734cb 100644 --- a/fs/bcachefs/btree_key_cache.c +++ b/fs/bcachefs/btree_key_cache.c @@ -778,6 +778,20 @@ static unsigned long bch2_btree_key_cache_scan(struct shrinker *shrink, rcu_read_lock(); tbl = rht_dereference_rcu(bc->table.tbl, &bc->table); + + /* + * Scanning is expensive while a rehash is in progress - most elements + * will be on the new hashtable, if it's in progress + * + * A rehash could still start while we're scanning - that's ok, we'll + * still see most elements. + */ + if (unlikely(tbl->nest)) { + rcu_read_unlock(); + srcu_read_unlock(&c->btree_trans_barrier, srcu_idx); + return SHRINK_STOP; + } + if (bc->shrink_iter >= tbl->size) bc->shrink_iter = 0; start = bc->shrink_iter; @@ -785,7 +799,7 @@ static unsigned long bch2_btree_key_cache_scan(struct shrinker *shrink, do { struct rhash_head *pos, *next; - pos = rht_ptr_rcu(rht_bucket(tbl, bc->shrink_iter)); + pos = rht_ptr_rcu(&tbl->buckets[bc->shrink_iter]); while (!rht_is_a_nulls(pos)) { next = rht_dereference_bucket_rcu(pos->next, tbl, bc->shrink_iter); @@ -866,12 +880,22 @@ void bch2_fs_btree_key_cache_exit(struct btree_key_cache *bc) while (atomic_long_read(&bc->nr_keys)) { rcu_read_lock(); tbl = rht_dereference_rcu(bc->table.tbl, &bc->table); - if (tbl) + if (tbl) { + if (tbl->nest) { + /* wait for in progress rehash */ + rcu_read_unlock(); + mutex_lock(&bc->table.mutex); + mutex_unlock(&bc->table.mutex); + rcu_read_lock(); + continue; + } for (i = 0; i < tbl->size; i++) - rht_for_each_entry_rcu(ck, pos, tbl, i, hash) { + while (pos = rht_ptr_rcu(&tbl->buckets[i]), !rht_is_a_nulls(pos)) { + ck = container_of(pos, struct bkey_cached, hash); bkey_cached_evict(bc, ck); list_add(&ck->list, &items); } + } rcu_read_unlock(); } -- cgit v1.2.3 From d3204616a67e53fdcad14c7026869330fb382fd4 Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Fri, 23 Aug 2024 17:38:41 -0400 Subject: bcachefs: Fix failure to flush moves before sleeping in copygc This fixes an apparent deadlock - rebalance would get stuck trying to take nocow locks because they weren't being released by copygc. Signed-off-by: Kent Overstreet --- fs/bcachefs/movinggc.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/fs/bcachefs/movinggc.c b/fs/bcachefs/movinggc.c index deef4f024d20..d86565bf07c8 100644 --- a/fs/bcachefs/movinggc.c +++ b/fs/bcachefs/movinggc.c @@ -383,7 +383,7 @@ static int bch2_copygc_thread(void *arg) if (min_member_capacity == U64_MAX) min_member_capacity = 128 * 2048; - bch2_trans_unlock_long(ctxt.trans); + move_buckets_wait(&ctxt, buckets, true); bch2_kthread_io_clock_wait(clock, last + (min_member_capacity >> 6), MAX_SCHEDULE_TIMEOUT); } -- cgit v1.2.3 From 49aa7830396bce33b00fa7ee734c35de36521138 Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Fri, 23 Aug 2024 15:35:22 -0400 Subject: bcachefs: Fix rebalance_work accounting rebalance_work was keying off of the presence of rebelance_opts in the extent - but that was incorrect, we keep those around after rebalance for indirect extents since the inode's options are not directly available Fixes: 20ac515a9cc7 ("bcachefs: bch_acct_rebalance_work") Signed-off-by: Kent Overstreet --- fs/bcachefs/bcachefs_format.h | 3 +- fs/bcachefs/buckets.c | 74 ++++++++++++++++++++++++++++--------------- fs/bcachefs/extents.c | 39 +++++++++++++++++++++++ fs/bcachefs/extents.h | 1 + fs/bcachefs/sb-downgrade.c | 8 ++++- 5 files changed, 98 insertions(+), 27 deletions(-) diff --git a/fs/bcachefs/bcachefs_format.h b/fs/bcachefs/bcachefs_format.h index c75f2e0f32bb..14ce726bf5a3 100644 --- a/fs/bcachefs/bcachefs_format.h +++ b/fs/bcachefs/bcachefs_format.h @@ -677,7 +677,8 @@ struct bch_sb_field_ext { x(bucket_stripe_sectors, BCH_VERSION(1, 8)) \ x(disk_accounting_v2, BCH_VERSION(1, 9)) \ x(disk_accounting_v3, BCH_VERSION(1, 10)) \ - x(disk_accounting_inum, BCH_VERSION(1, 11)) + x(disk_accounting_inum, BCH_VERSION(1, 11)) \ + x(rebalance_work_acct_fix, BCH_VERSION(1, 12)) enum bcachefs_metadata_version { bcachefs_metadata_version_min = 9, diff --git a/fs/bcachefs/buckets.c b/fs/bcachefs/buckets.c index be2bbd248631..a2274429e7f4 100644 --- a/fs/bcachefs/buckets.c +++ b/fs/bcachefs/buckets.c @@ -699,7 +699,8 @@ err: static int __trigger_extent(struct btree_trans *trans, enum btree_id btree_id, unsigned level, struct bkey_s_c k, - enum btree_iter_update_trigger_flags flags) + enum btree_iter_update_trigger_flags flags, + s64 *replicas_sectors) { bool gc = flags & BTREE_TRIGGER_gc; struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k); @@ -708,7 +709,6 @@ static int __trigger_extent(struct btree_trans *trans, enum bch_data_type data_type = bkey_is_btree_ptr(k.k) ? BCH_DATA_btree : BCH_DATA_user; - s64 replicas_sectors = 0; int ret = 0; struct disk_accounting_pos acc_replicas_key = { @@ -739,7 +739,7 @@ static int __trigger_extent(struct btree_trans *trans, if (ret) return ret; } else if (!p.has_ec) { - replicas_sectors += disk_sectors; + *replicas_sectors += disk_sectors; acc_replicas_key.replicas.devs[acc_replicas_key.replicas.nr_devs++] = p.ptr.dev; } else { ret = bch2_trigger_stripe_ptr(trans, k, p, data_type, disk_sectors, flags); @@ -777,7 +777,7 @@ static int __trigger_extent(struct btree_trans *trans, } if (acc_replicas_key.replicas.nr_devs) { - ret = bch2_disk_accounting_mod(trans, &acc_replicas_key, &replicas_sectors, 1, gc); + ret = bch2_disk_accounting_mod(trans, &acc_replicas_key, replicas_sectors, 1, gc); if (ret) return ret; } @@ -787,7 +787,7 @@ static int __trigger_extent(struct btree_trans *trans, .type = BCH_DISK_ACCOUNTING_snapshot, .snapshot.id = k.k->p.snapshot, }; - ret = bch2_disk_accounting_mod(trans, &acc_snapshot_key, &replicas_sectors, 1, gc); + ret = bch2_disk_accounting_mod(trans, &acc_snapshot_key, replicas_sectors, 1, gc); if (ret) return ret; } @@ -807,7 +807,7 @@ static int __trigger_extent(struct btree_trans *trans, .type = BCH_DISK_ACCOUNTING_btree, .btree.id = btree_id, }; - ret = bch2_disk_accounting_mod(trans, &acc_btree_key, &replicas_sectors, 1, gc); + ret = bch2_disk_accounting_mod(trans, &acc_btree_key, replicas_sectors, 1, gc); if (ret) return ret; } else { @@ -819,22 +819,13 @@ static int __trigger_extent(struct btree_trans *trans, s64 v[3] = { insert ? 1 : -1, insert ? k.k->size : -((s64) k.k->size), - replicas_sectors, + *replicas_sectors, }; ret = bch2_disk_accounting_mod(trans, &acc_inum_key, v, ARRAY_SIZE(v), gc); if (ret) return ret; } - if (bch2_bkey_rebalance_opts(k)) { - struct disk_accounting_pos acc = { - .type = BCH_DISK_ACCOUNTING_rebalance_work, - }; - ret = bch2_disk_accounting_mod(trans, &acc, &replicas_sectors, 1, gc); - if (ret) - return ret; - } - return 0; } @@ -843,6 +834,7 @@ int bch2_trigger_extent(struct btree_trans *trans, struct bkey_s_c old, struct bkey_s new, enum btree_iter_update_trigger_flags flags) { + struct bch_fs *c = trans->c; struct bkey_ptrs_c new_ptrs = bch2_bkey_ptrs_c(new.s_c); struct bkey_ptrs_c old_ptrs = bch2_bkey_ptrs_c(old); unsigned new_ptrs_bytes = (void *) new_ptrs.end - (void *) new_ptrs.start; @@ -858,21 +850,53 @@ int bch2_trigger_extent(struct btree_trans *trans, new_ptrs_bytes)) return 0; - if (flags & BTREE_TRIGGER_transactional) { - struct bch_fs *c = trans->c; - int mod = (int) bch2_bkey_needs_rebalance(c, new.s_c) - - (int) bch2_bkey_needs_rebalance(c, old); + if (flags & (BTREE_TRIGGER_transactional|BTREE_TRIGGER_gc)) { + s64 old_replicas_sectors = 0, new_replicas_sectors = 0; + + if (old.k->type) { + int ret = __trigger_extent(trans, btree, level, old, + flags & ~BTREE_TRIGGER_insert, + &old_replicas_sectors); + if (ret) + return ret; + } + + if (new.k->type) { + int ret = __trigger_extent(trans, btree, level, new.s_c, + flags & ~BTREE_TRIGGER_overwrite, + &new_replicas_sectors); + if (ret) + return ret; + } + + int need_rebalance_delta = 0; + s64 need_rebalance_sectors_delta = 0; + + s64 s = bch2_bkey_sectors_need_rebalance(c, old); + need_rebalance_delta -= s != 0; + need_rebalance_sectors_delta -= s; - if (mod) { + s = bch2_bkey_sectors_need_rebalance(c, old); + need_rebalance_delta += s != 0; + need_rebalance_sectors_delta += s; + + if ((flags & BTREE_TRIGGER_transactional) && need_rebalance_delta) { int ret = bch2_btree_bit_mod_buffered(trans, BTREE_ID_rebalance_work, - new.k->p, mod > 0); + new.k->p, need_rebalance_delta > 0); if (ret) return ret; } - } - if (flags & (BTREE_TRIGGER_transactional|BTREE_TRIGGER_gc)) - return trigger_run_overwrite_then_insert(__trigger_extent, trans, btree, level, old, new, flags); + if (need_rebalance_sectors_delta) { + struct disk_accounting_pos acc = { + .type = BCH_DISK_ACCOUNTING_rebalance_work, + }; + int ret = bch2_disk_accounting_mod(trans, &acc, &need_rebalance_sectors_delta, 1, + flags & BTREE_TRIGGER_gc); + if (ret) + return ret; + } + } return 0; } diff --git a/fs/bcachefs/extents.c b/fs/bcachefs/extents.c index 9406f82fc255..e317df3644a1 100644 --- a/fs/bcachefs/extents.c +++ b/fs/bcachefs/extents.c @@ -1379,6 +1379,45 @@ bool bch2_bkey_needs_rebalance(struct bch_fs *c, struct bkey_s_c k) return r != NULL; } +static u64 __bch2_bkey_sectors_need_rebalance(struct bch_fs *c, struct bkey_s_c k, + unsigned target, unsigned compression) +{ + struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k); + const union bch_extent_entry *entry; + struct extent_ptr_decoded p; + u64 sectors = 0; + + if (compression) { + unsigned compression_type = bch2_compression_opt_to_type(compression); + + bkey_for_each_ptr_decode(k.k, ptrs, p, entry) { + if (p.crc.compression_type == BCH_COMPRESSION_TYPE_incompressible || + p.ptr.unwritten) { + sectors = 0; + goto incompressible; + } + + if (!p.ptr.cached && p.crc.compression_type != compression_type) + sectors += p.crc.compressed_size; + } + } +incompressible: + if (target && bch2_target_accepts_data(c, BCH_DATA_user, target)) { + bkey_for_each_ptr_decode(k.k, ptrs, p, entry) + if (!p.ptr.cached && !bch2_dev_in_target(c, p.ptr.dev, target)) + sectors += p.crc.compressed_size; + } + + return sectors; +} + +u64 bch2_bkey_sectors_need_rebalance(struct bch_fs *c, struct bkey_s_c k) +{ + const struct bch_extent_rebalance *r = bch2_bkey_rebalance_opts(k); + + return r ? __bch2_bkey_sectors_need_rebalance(c, k, r->target, r->compression) : 0; +} + int bch2_bkey_set_needs_rebalance(struct bch_fs *c, struct bkey_i *_k, struct bch_io_opts *opts) { diff --git a/fs/bcachefs/extents.h b/fs/bcachefs/extents.h index 1a6ddee48041..709dd83183be 100644 --- a/fs/bcachefs/extents.h +++ b/fs/bcachefs/extents.h @@ -692,6 +692,7 @@ const struct bch_extent_rebalance *bch2_bkey_rebalance_opts(struct bkey_s_c); unsigned bch2_bkey_ptrs_need_rebalance(struct bch_fs *, struct bkey_s_c, unsigned, unsigned); bool bch2_bkey_needs_rebalance(struct bch_fs *, struct bkey_s_c); +u64 bch2_bkey_sectors_need_rebalance(struct bch_fs *, struct bkey_s_c); int bch2_bkey_set_needs_rebalance(struct bch_fs *, struct bkey_i *, struct bch_io_opts *); diff --git a/fs/bcachefs/sb-downgrade.c b/fs/bcachefs/sb-downgrade.c index 650a1f77ca40..c7e4cdd3f6a5 100644 --- a/fs/bcachefs/sb-downgrade.c +++ b/fs/bcachefs/sb-downgrade.c @@ -74,6 +74,9 @@ BCH_FSCK_ERR_accounting_key_replicas_devs_unsorted, \ BCH_FSCK_ERR_accounting_key_junk_at_end) \ x(disk_accounting_inum, \ + BIT_ULL(BCH_RECOVERY_PASS_check_allocations), \ + BCH_FSCK_ERR_accounting_mismatch) \ + x(rebalance_work_acct_fix, \ BIT_ULL(BCH_RECOVERY_PASS_check_allocations), \ BCH_FSCK_ERR_accounting_mismatch) @@ -108,7 +111,10 @@ BCH_FSCK_ERR_fs_usage_persistent_reserved_wrong, \ BCH_FSCK_ERR_fs_usage_replicas_wrong, \ BCH_FSCK_ERR_accounting_replicas_not_marked, \ - BCH_FSCK_ERR_bkey_version_in_future) + BCH_FSCK_ERR_bkey_version_in_future) \ + x(rebalance_work_acct_fix, \ + BIT_ULL(BCH_RECOVERY_PASS_check_allocations), \ + BCH_FSCK_ERR_accounting_mismatch) struct upgrade_downgrade_entry { u64 recovery_passes; -- cgit v1.2.3