vfs: introduce --atime-mode option to control atime update behavior

Currently juicefs doesn't update atime on access, so introduce
--atime-mode option to select how to update atime. And support three
modes:

- noatime: don't update atime
- relatime: update atime relative to motify or change time
- strictatime: always update atime

Fixes: #3240
Signed-off-by: Eryu Guan <[email protected]>

cmd/flags.go

@@ -214,6 +214,10 @@ func clientFlags() []cli.Flag {
 			Name:  "subdir",
 			Usage: "mount a sub-directory as root",
 		},
+		&cli.StringFlag{
+			Name:  "atime-mode",
+			Usage: "when to update atime, supported mode includes: noatime (default), relatime, strictatime",
+		},
 	}
 }
 

cmd/mount.go

@@ -259,6 +259,12 @@ func getVfsConf(c *cli.Context, metaConf *meta.Config, format *meta.Format, chun
 	if cfg.BackupMeta > 0 && cfg.BackupMeta < time.Minute*5 {
 		logger.Fatalf("backup-meta should not be less than 5 minutes: %s", cfg.BackupMeta)
 	}
+	atimeMode := c.String("atime-mode")
+	if atimeMode != vfs.RelAtime && atimeMode != vfs.StrictAtime {
+		logger.Warnf("unknown atime-mode \"%s\", changed to %s", atimeMode, vfs.NoAtime)
+		atimeMode = vfs.NoAtime
+	}
+	cfg.AtimeMode = atimeMode
 	return cfg
 }
 

docs/en/development/internals.md

@@ -173,7 +173,10 @@ type Attr struct {
 
 There are a few fields that need clarification.
 
-- Atime/Atimensec: set only when the file is created and when `SetAttr` is actively called, while accessing and modifying the file usually does not affect the Atime value
+- Atime/Atimensec: currently support three modes
+  - noatime: set only when the file is created and when `SetAttr` is actively called, while accessing and modifying the file usually does not affect the Atime value, this is the default behavior
+  - relatime: update  inode access times relative to modify or change time.  Access time is only updated if the previous access time was earlier than the current modify or change time, or the file's last access time is always updated if it  is more than 1 day old
+  - strictatime: always update atime on access
 - Nlink
   - Directory file: initial value is 2 ('.' and '..'), add 1 for each subdirectory
   - Other files: initial value is 1, add 1 for each hard link created

docs/en/reference/command_reference.md

@@ -318,6 +318,9 @@ interval (in seconds) to send heartbeat; it's recommended that all clients use t
 `--no-bgjob`<br />
 disable background jobs (clean-up, backup, etc.) (default: false)
 
+`--atime-mode value`<br />
+control atime behavior, support 3 modes, `noatime`, `relatime`, `strictatime`, default to `noatime`
+
 #### Examples
 
 ```bash

docs/zh_cn/development/internals.md

@@ -176,7 +176,10 @@ type Attr struct {
 
 其中几个需要说明的字段：
 
-- Atime/Atimensec：仅在文件创建和主动调用 `SetAttr` 时设置，平时访问与修改文件不影响 Atime 值
+- Atime/Atimensec：目前支持三种模式
+  - noatime: 仅在文件创建和主动调用 `SetAttr` 时设置，平时访问与修改文件不影响 Atime 值，这是默认行为
+  - relatime: Mtime 或者 Ctime 比 Atime 新时，或者 Atime 超过24小时没有更新时更新 Atime
+  - strictatime: 一直更新 Atime
 - Nlink：
   - 目录文件：初始值为 2（'.' 和 '..'），每有一个子目录 Nlink 值加 1
   - 其他文件：初始值为 1，每创建一个硬链接 Nlink 值加 1

docs/zh_cn/reference/command_reference.md

@@ -318,6 +318,9 @@ Consul 注册中心地址 (默认："127.0.0.1:8500")
 `--no-bgjob`<br />
 禁用后台作业（清理、备份等）（默认：false）
 
+`--atime-mode value`<br />
+控制如何更新atime，支持3种模式，`noatime`，`relatime`，`strictatime`，默认使用 `noatime`
+
 #### 示例
 
 ```shell

pkg/vfs/vfs.go

@@ -39,6 +39,10 @@ const (
 	maxName     = meta.MaxName
 	maxSymlink  = 4096
 	maxFileSize = meta.ChunkSize << 31
+
+	NoAtime     = "noatime"
+	RelAtime    = "relatime"
+	StrictAtime = "strictatime"
 )
 
 type Port struct {
@@ -61,6 +65,7 @@ type Config struct {
 	FastResolve     bool   `json:",omitempty"`
 	AccessLog       string `json:",omitempty"`
 	HideInternal    bool
+	AtimeMode       string
 }
 
 var (
@@ -564,6 +569,12 @@ func (v *VFS) Read(ctx Context, ino Ino, buf []byte, off uint64, fh uint64) (n i
 	if err == syscall.ENOENT {
 		err = syscall.EBADF
 	}
+
+	if err == 0 {
+		if aerr := v.touchAtime(ctx, ino); aerr != 0 {
+			logger.Warnf("read %d update atime: %v", ino, aerr)
+		}
+	}
 	h.removeOp(ctx)
 	return
 }
@@ -731,6 +742,9 @@ func (v *VFS) CopyFileRange(ctx Context, nodeIn Ino, fhIn, offIn uint64, nodeOut
 	if err == 0 {
 		v.reader.Invalidate(nodeOut, offOut, size)
 		v.invalidateLength(nodeOut)
+		if err := v.touchAtime(ctx, nodeIn); err != 0 {
+			logger.Warnf("copy_file_range %d update atime: %v", nodeIn, err)
+		}
 	}
 	return
 }
@@ -995,6 +1009,54 @@ func (v *VFS) cleanupModified() {
 	}
 }
 
+func (v *VFS) atimeNeedsUpdate(attr *Attr, now time.Time) bool {
+	if v.Conf.AtimeMode == RelAtime && relatimeNeedUpdate(attr, now) {
+		return true
+	}
+
+	return v.Conf.AtimeMode == StrictAtime && !now.Equal(time.Unix(attr.Atime, int64(attr.Atimensec)))
+}
+
+// caller makes sure inode is not special inode.
+func (v *VFS) touchAtime(ctx Context, ino Ino) syscall.Errno {
+	if (v.Conf.AtimeMode != StrictAtime && v.Conf.AtimeMode != RelAtime) || v.Conf.Meta.ReadOnly {
+		return 0
+	}
+
+	var origAttr = &Attr{}
+	if err := v.Meta.GetAttr(ctx, ino, origAttr); err != 0 {
+		return err
+	}
+	now := time.Now()
+	if !v.atimeNeedsUpdate(origAttr, now) {
+		return 0
+	}
+
+	return v.Meta.SetAttr(ctx, ino, meta.SetAttrAtimeNow, 0, &Attr{})
+}
+
+// With relative atime, only update atime if the previous atime is earlier than either the ctime or
+// mtime or if at least a day has passed since the last atime update.
+func relatimeNeedUpdate(attr *Attr, now time.Time) bool {
+	atime := time.Unix(attr.Atime, int64(attr.Atimensec))
+
+	// Is mtime younger than atime? If yes, update atime
+	if time.Unix(attr.Mtime, int64(attr.Mtimensec)).After(atime) {
+		return true
+	}
+
+	// Is ctime younger than atime? If yes, update atime
+	if time.Unix(attr.Ctime, int64(attr.Ctimensec)).After(atime) {
+		return true
+	}
+
+	// Is the previous atime value older than a day? If yes, update atime
+	if now.Sub(atime) > 24*time.Hour {
+		return true
+	}
+
+	return false
+}
 func initVFSMetrics(v *VFS, writer DataWriter, reader DataReader, registerer prometheus.Registerer) {
 	if registerer == nil {
 		return

pkg/vfs/vfs_test.go

@@ -830,3 +830,86 @@ func TestInternalFile(t *testing.T) {
 		t.Fatalf("result: %s", string(resp[:n]))
 	}
 }
+
+func TestAtime(t *testing.T) {
+	v, _ := createTestVFS()
+	ctx := NewLogContext(meta.NewContext(10, 1, []uint32{2}))
+
+	// noatime by default
+	fe, fh, e := v.Create(ctx, 1, "f1", 0755, 0, syscall.O_RDWR)
+	if e != 0 {
+		t.Fatalf("create file: %s", e)
+	}
+	if fe2, e := v.SetAttr(ctx, fe.Inode, meta.SetAttrAtime, 0, 0, 0, 0, 1234, 0, 5678, 0, 0); e != 0 {
+		t.Fatalf("setattr f1: %s", e)
+	} else if fe2.Attr.Atime != 1234 || fe2.Attr.Atimensec != 5678 {
+		t.Fatalf("setattr f1: %+v", fe2.Attr)
+	}
+	var buf = make([]byte, 4096)
+	if _, e := v.Read(ctx, fe.Inode, buf, 0, fh); e != 0 {
+		t.Fatalf("read file: %s", e)
+	}
+	if fe, e := v.GetAttr(ctx, fe.Inode, 0); e != 0 || fe.Attr.Atime != 1234 || fe.Attr.Atimensec != 5678 {
+		t.Fatalf("getattr after read f1: %s atime: %v, atimensec %v", e, fe.Attr.Atime, fe.Attr.Atimensec)
+	}
+
+	// set relatime
+	v.Conf.AtimeMode = RelAtime
+	fe, fh, e = v.Create(ctx, 1, "f2", 0755, 0, syscall.O_RDWR)
+	if e != 0 {
+		t.Fatalf("create file: %s", e)
+	}
+	if fe2, e := v.SetAttr(ctx, fe.Inode, meta.SetAttrAtime, 0, 0, 0, 0, 1234, 0, 5678, 0, 0); e != 0 {
+		t.Fatalf("setattr f2: %s", e)
+	} else if fe2.Attr.Atime != 1234 || fe2.Attr.Atimensec != 5678 {
+		t.Fatalf("setattr f2: %+v", fe2.Attr)
+	}
+	if _, e := v.Read(ctx, fe.Inode, buf, 0, fh); e != 0 {
+		t.Fatalf("read file: %s", e)
+	}
+	// older than 24h, update atime on read
+	if fe, e := v.GetAttr(ctx, fe.Inode, 0); e != 0 || fe.Attr.Atime == 1234 {
+		t.Fatalf("getattr after read f2: %s atime: %v, atimensec %v", e, fe.Attr.Atime, fe.Attr.Atimensec)
+	}
+	now := time.Now()
+	if fe2, e := v.SetAttr(ctx, fe.Inode, meta.SetAttrMtime, 0, 0, 0, 0, 0, now.Unix(), 0, uint32(now.Nanosecond()), 0); e != 0 {
+		t.Fatalf("setattr f2: %s", e)
+	} else if fe2.Attr.Mtime != now.Unix() || fe2.Attr.Mtimensec != uint32(now.Nanosecond()) {
+		t.Fatalf("setattr f2: %+v", fe2.Attr)
+	}
+	time.Sleep(time.Second)
+	if _, e := v.Read(ctx, fe.Inode, buf, 0, fh); e != 0 {
+		t.Fatalf("read file: %s", e)
+	}
+	// mtime > atime, update atime on access
+	if fe, e := v.GetAttr(ctx, fe.Inode, 0); e != 0 || fe.Attr.Atime < now.Unix() {
+		t.Fatalf("getattr after access f2: %s atime: %v, now %v", e, fe.Attr.Atime, now.Unix())
+	}
+
+	// set strictatime
+	v.Conf.AtimeMode = StrictAtime
+	fe, fh, e = v.Create(ctx, 1, "f3", 0755, 0, syscall.O_RDWR)
+	if e != 0 {
+		t.Fatalf("create file: %s", e)
+	}
+	if fe2, e := v.SetAttr(ctx, fe.Inode, meta.SetAttrAtime, 0, 0, 0, 0, 1234, 0, 5678, 0, 0); e != 0 {
+		t.Fatalf("setattr f3: %s", e)
+	} else if fe2.Attr.Atime != 1234 || fe2.Attr.Atimensec != 5678 {
+		t.Fatalf("setattr f3: %+v", fe2.Attr)
+	}
+	if _, e := v.Read(ctx, fe.Inode, buf, 0, fh); e != 0 {
+		t.Fatalf("read file: %s", e)
+	}
+	// always update atime on read
+	if fe, e := v.GetAttr(ctx, fe.Inode, 0); e != 0 || fe.Attr.Atime == 1234 {
+		t.Fatalf("getattr after read f3: %s atime: %v, atimensec %v", e, fe.Attr.Atime, fe.Attr.Atimensec)
+	}
+	now = time.Now()
+	time.Sleep(time.Second)
+	if _, e := v.Read(ctx, fe.Inode, buf, 0, fh); e != 0 {
+		t.Fatalf("read file: %s", e)
+	}
+	if fe, e := v.GetAttr(ctx, fe.Inode, 0); e != 0 || fe.Attr.Atime < now.Unix() {
+		t.Fatalf("getattr after access f3: %s atime: %v, now %v", e, fe.Attr.Atime, now.Unix())
+	}
+}