From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from smtp.kernel.org (aws-us-west-2-korg-mail-alma10-1.taild15c8.ts.net [100.103.45.18]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id E1439233952 for ; Sun, 9 Aug 2026 01:26:45 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=100.103.45.18 ARC-Seal:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1786238807; cv=none; b=QiXGpnXgvXRKIbgsitWDDOOsJA47U6YMbJdppKiij6Z4HZmPksn+lBF3vlYUi1dAi7u47CundVFY8rD7hVjIeA9W7gpqYJAVmJG2m0QX1l/qZ+YO/ms5eR3tr6y1EhXTt5qSNP9EN4nSMLD5g1eMmunEz0/MwpvXk0X9jpq6GDY= ARC-Message-Signature:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1786238807; c=relaxed/simple; bh=ZJxm7w+NovV9GAOkrkywTC11DjtHlGBwkyywvutdkhc=; h=From:To:Cc:Subject:Date:Message-ID:In-Reply-To:References: MIME-Version; b=GbHUEm8lZA5QcWZON4m3lYw/1WLBa1zuPB+L3JfWCB0rtXZfBM/pU2ulkjKW4WWWbWx8smOJj4ihmDJ26zV/Nk1k8al7hLbj2QReq744E3E1+b40CKQcPvGW5by6BNu+bu229RR2V+d+0y890qwIiaQBP1OFjO9BAjGoyhI/jCg= ARC-Authentication-Results:i=1; smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=kernel.org header.i=@kernel.org header.b=nK4CLs6O; arc=none smtp.client-ip=100.103.45.18 Authentication-Results: smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=kernel.org header.i=@kernel.org header.b="nK4CLs6O" Received: by smtp.kernel.org (Postfix) with ESMTPSA id A4A531F00A3D; Sun, 9 Aug 2026 01:26:44 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=kernel.org; s=k20260515; t=1786238805; bh=GdyR95RgyXNdZtaoq952nJ4CVYObk7LTzOIlLzQE7oY=; h=From:To:Cc:Subject:Date:In-Reply-To:References; b=nK4CLs6OdPCvRlXBuEZC3AlYFEbPg6RUWXCAPji55t/VlEimtfSYx0feOIdiVwIIu TNyKInxIMSNYixxmzFjU6eo9movg6sH6pVzhzQOpmF8mqL0kgiq9rUIvJ566vUkrrU 1YaW8pywA1LiVYWX5EzjrK1gPFiFSWqT8iAB82awMAcgP5DZNbZjjb4SwFcgCPVd/1 Wry/9Nd8vd6uFW63GXLBjjrO2TT3E+3kAtnV/A5VngZXrOo+NIVRKPPcY6A1XCNX2q oTjbW3rJ78k1Rq+rlOwURqRRKxOvGWpUAGuSfTcKRTuSuoZZQO/CCO63sSBBmnw9uD l9tCoP7Le95kA== From: Chao Yu To: jaegeuk@kernel.org Cc: linux-f2fs-devel@lists.sourceforge.net, linux-kernel@vger.kernel.org, Chao Yu Subject: [PATCH v2 2/2] f2fs: reduce memory footprint of ino management Date: Sun, 9 Aug 2026 09:26:39 +0800 Message-ID: <20260809012639.568485-2-chao@kernel.org> X-Mailer: git-send-email 2.55.0.654.g21b8a5bc05-goog In-Reply-To: <20260809012639.568485-1-chao@kernel.org> References: <20260809012639.568485-1-chao@kernel.org> Precedence: bulk X-Mailing-List: linux-kernel@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 Content-Transfer-Encoding: 8bit Currently, ino entries for APPEND_INO, UPDATE_INO, TRANS_DIR_INO, and XATTR_DIR_INO allocate a 'struct ino_entry' slab object and attach it to both a list and a radix tree solely for existence checks via f2fs_exist_written_data(). Since these ino types only track binary existence status, we can embed the information directly into radix tree value entries as a bitmap: - The Linux radix tree/XArray supports in-place value entries via xa_mk_value() / xa_to_value(), which tag the least significant bit to store an unallocated integer value of BITS_PER_XA_VALUE bits (BITS_PER_LONG - 1) directly in the slot pointer. - For each inode, (ino / BITS_PER_XA_VALUE) serves as the radix tree slot index, and (ino % BITS_PER_XA_VALUE) is used as the bit offset within the slot's bitmap. For example, when tracking ino = 7: - Before: Allocate a 'struct ino_entry' ({ .ino = 7 }), insert its pointer into the radix tree at index = 7, and link it to im->ino_list. - After: Compute slot_index = 7 / BITS_PER_XA_VALUE (index 0) and bit_offset = 7 % BITS_PER_XA_VALUE (bit 7), then set bit 7 in the value entry via xa_mk_value(bitmap) at index 0, without allocating a slab object or linking to a list. Additionally: - In-place slot updates are performed via radix_tree_replace_slot(), and slots are deleted with radix_tree_delete() once the bitmap is zeroed. - Reorder the ino list enum so ORPHAN_INO and FLUSH_INO (which still require struct ino_entry and list traversal) remain separated, while bitmap-based trees are torn down using xa_destroy(). This eliminates 'struct ino_entry' slab allocations and linked-list tracking for these ino types, significantly reducing memory consumption. Signed-off-by: Chao Yu --- v2: - fix to call __remove_ino_entry correctly for FLUSH_INO fs/f2fs/checkpoint.c | 99 ++++++++++++++++++++++++++++++++++++++++---- fs/f2fs/f2fs.h | 6 ++- 2 files changed, 95 insertions(+), 10 deletions(-) diff --git a/fs/f2fs/checkpoint.c b/fs/f2fs/checkpoint.c index eb2f955b1e2a..f81ba8cc861a 100644 --- a/fs/f2fs/checkpoint.c +++ b/fs/f2fs/checkpoint.c @@ -766,6 +766,64 @@ static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type) spin_unlock(&im->ino_lock); } +static void __set_ino_bitmap(struct f2fs_sb_info *sbi, nid_t ino, int type) +{ + struct inode_management *im = &sbi->im[type]; + unsigned long index = INO_SLOT_INDEX(ino); + unsigned int ofs = INO_BIT_OFFSET(ino); + void **slot, *entry; + unsigned long bitmap = 0; + int ret; + + ret = radix_tree_preload(GFP_NOFS | __GFP_NOFAIL); + f2fs_bug_on(sbi, ret); + + spin_lock(&im->ino_lock); + slot = radix_tree_lookup_slot(&im->ino_root, index); + if (slot) { + entry = radix_tree_deref_slot(slot); + bitmap = xa_to_value(entry); + if (!(bitmap & (1UL << ofs))) { + bitmap |= (1UL << ofs); + entry = xa_mk_value(bitmap); + radix_tree_replace_slot(&im->ino_root, slot, entry); + } + } else { + bitmap |= (1UL << ofs); + entry = xa_mk_value(bitmap); + if (unlikely(radix_tree_insert(&im->ino_root, index, entry))) + f2fs_bug_on(sbi, 1); + } + spin_unlock(&im->ino_lock); + radix_tree_preload_end(); +} + +static void __clear_ino_bitmap(struct f2fs_sb_info *sbi, nid_t ino, int type) +{ + struct inode_management *im = &sbi->im[type]; + unsigned long index = INO_SLOT_INDEX(ino); + unsigned int ofs = INO_BIT_OFFSET(ino); + void **slot, *entry; + unsigned long bitmap; + + spin_lock(&im->ino_lock); + slot = radix_tree_lookup_slot(&im->ino_root, index); + if (slot) { + entry = radix_tree_deref_slot(slot); + bitmap = xa_to_value(entry); + if (bitmap & (1UL << ofs)) + bitmap &= ~(1UL << ofs); + + if (bitmap) { + entry = xa_mk_value(bitmap); + radix_tree_replace_slot(&im->ino_root, slot, entry); + } else { + radix_tree_delete(&im->ino_root, index); + } + } + spin_unlock(&im->ino_lock); +} + static void f2fs_wait_for_inode_record(struct f2fs_sb_info *sbi, int mode) { if (mode != APPEND_INO && mode != UPDATE_INO) @@ -778,8 +836,10 @@ static void f2fs_wait_for_inode_record(struct f2fs_sb_info *sbi, int mode) static void __f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, unsigned int devidx, int type) { - /* add new dirty ino entry into list */ - __add_ino_entry(sbi, ino, devidx, type); + if (type <= FLUSH_INO) + __add_ino_entry(sbi, ino, devidx, type); + else + __set_ino_bitmap(sbi, ino, type); } void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type) @@ -789,20 +849,33 @@ void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type) void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type) { - /* remove dirty ino entry from list */ - __remove_ino_entry(sbi, ino, type); + if (type <= FLUSH_INO) + __remove_ino_entry(sbi, ino, type); + else + __clear_ino_bitmap(sbi, ino, type); } -/* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */ +/* mode should be APPEND_INO, UPDATE_INO, TRANS_DIR_INO and XATTR_DIR_INO */ bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode) { struct inode_management *im = &sbi->im[mode]; - struct ino_entry *e; + unsigned long index = INO_SLOT_INDEX(ino); + unsigned int ofs = INO_BIT_OFFSET(ino); + void *entry; + unsigned long bitmap; + + f2fs_bug_on(sbi, mode <= FLUSH_INO); spin_lock(&im->ino_lock); - e = radix_tree_lookup(&im->ino_root, ino); + entry = radix_tree_lookup(&im->ino_root, index); + if (!entry) { + spin_unlock(&im->ino_lock); + return false; + } + bitmap = xa_to_value(entry); spin_unlock(&im->ino_lock); - return e ? true : false; + + return bitmap & (1UL << ofs); } void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all) @@ -810,7 +883,7 @@ void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all) struct ino_entry *e, *tmp; int i; - for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) { + for (i = all ? ORPHAN_INO : FLUSH_INO; i <= FLUSH_INO; i++) { struct inode_management *im = &sbi->im[i]; f2fs_wait_for_inode_record(sbi, i); @@ -824,6 +897,14 @@ void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all) } spin_unlock(&im->ino_lock); } + + for (i = APPEND_INO; i < MAX_INO_ENTRY; i++) { + struct inode_management *im = &sbi->im[i]; + + spin_lock(&im->ino_lock); + xa_destroy(&im->ino_root); + spin_unlock(&im->ino_lock); + } } void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino, diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h index 160e7db0674e..5ae2d347c3ce 100644 --- a/fs/f2fs/f2fs.h +++ b/fs/f2fs/f2fs.h @@ -389,14 +389,18 @@ enum { /* for the list of ino */ enum { ORPHAN_INO, /* for orphan ino list */ + FLUSH_INO, /* for multiple device flushing */ APPEND_INO, /* for append ino list */ UPDATE_INO, /* for update ino list */ TRANS_DIR_INO, /* for transactions dir ino list */ XATTR_DIR_INO, /* for xattr updated dir ino list */ - FLUSH_INO, /* for multiple device flushing */ MAX_INO_ENTRY, /* max. list */ }; +#define INO_BITS_PER_SLOT BITS_PER_XA_VALUE +#define INO_SLOT_INDEX(ino) ((ino) / INO_BITS_PER_SLOT) +#define INO_BIT_OFFSET(ino) ((ino) % INO_BITS_PER_SLOT) + struct ino_entry { struct list_head list; /* list head */ nid_t ino; /* inode number */ -- 2.49.0