feat: refactor internal iam service

This moves the internal iam service from the posix backend so
that we can start implementing new iam services right in the auth
module.

The internal iam service has same behavior as before, but now
must be enabled with the --iam-dir cli option.

New single user service is the default when no other iam service
is selected. This just runs the gateway in single user mode with
just the root account.
This commit is contained in:
Ben McClelland
2023-10-02 11:12:18 -07:00
parent 4661af11dd
commit ae4e382e61
10 changed files with 301 additions and 275 deletions
-205
View File
@@ -29,9 +29,7 @@ import (
"sort"
"strconv"
"strings"
"sync"
"syscall"
"time"
"github.com/aws/aws-sdk-go-v2/service/s3"
"github.com/aws/aws-sdk-go-v2/service/s3/types"
@@ -48,19 +46,10 @@ type Posix struct {
rootfd *os.File
rootdir string
mu sync.RWMutex
iamcache []byte
iamvalid bool
iamexpire time.Time
}
var _ backend.Backend = &Posix{}
var (
cacheDuration = 5 * time.Minute
)
const (
metaTmpDir = ".sgwtmp"
metaTmpMultipartDir = metaTmpDir + "/multipart"
@@ -70,8 +59,6 @@ const (
contentTypeHdr = "content-type"
contentEncHdr = "content-encoding"
emptyMD5 = "d41d8cd98f00b204e9800998ecf8427e"
iamFile = "users.json"
iamBackupFile = "users.json.backup"
aclkey = "user.acl"
etagkey = "user.etag"
)
@@ -1656,198 +1643,6 @@ func (p *Posix) ListBucketsAndOwners(ctx context.Context) (buckets []s3response.
return buckets, nil
}
const (
iamMode = 0600
)
func (p *Posix) InitIAM() error {
p.mu.RLock()
defer p.mu.RUnlock()
_, err := os.ReadFile(iamFile)
if errors.Is(err, fs.ErrNotExist) {
b, err := json.Marshal(auth.IAMConfig{AccessAccounts: map[string]auth.Account{}})
if err != nil {
return fmt.Errorf("marshal default iam: %w", err)
}
err = os.WriteFile(iamFile, b, iamMode)
if err != nil {
return fmt.Errorf("write default iam: %w", err)
}
}
return nil
}
func (p *Posix) GetIAM() ([]byte, error) {
p.mu.RLock()
defer p.mu.RUnlock()
if !p.iamvalid || !p.iamexpire.After(time.Now()) {
p.mu.RUnlock()
err := p.refreshIAM()
p.mu.RLock()
if err != nil {
return nil, err
}
}
return p.iamcache, nil
}
const (
backoff = 100 * time.Millisecond
maxretry = 300
)
func (p *Posix) refreshIAM() error {
p.mu.Lock()
defer p.mu.Unlock()
// We are going to be racing with other running gateways without any
// coordination. So we might find the file does not exist at times.
// For this case we need to retry for a while assuming the other gateway
// will eventually write the file. If it doesn't after the max retries,
// then we will return the error.
retries := 0
for {
b, err := os.ReadFile(iamFile)
if errors.Is(err, fs.ErrNotExist) {
// racing with someone else updating
// keep retrying after backoff
retries++
if retries < maxretry {
time.Sleep(backoff)
continue
}
return fmt.Errorf("read iam file: %w", err)
}
if err != nil {
return err
}
p.iamcache = b
p.iamvalid = true
p.iamexpire = time.Now().Add(cacheDuration)
break
}
return nil
}
func (p *Posix) StoreIAM(update auth.UpdateAcctFunc) error {
p.mu.Lock()
defer p.mu.Unlock()
// We are going to be racing with other running gateways without any
// coordination. So the strategy here is to read the current file data.
// If the file doesn't exist, then we assume someone else is currently
// updating the file. So we just need to keep retrying. We also need
// to make sure the data is consistent within a single update. So racing
// writes to a file would possibly leave this in some invalid state.
// We can get atomic updates with rename. If we read the data, update
// the data, write to a temp file, then rename the tempfile back to the
// data file. This should always result in a complete data image.
// There is at least one unsolved failure mode here.
// If a gateway removes the data file and then crashes, all other
// gateways will retry forever thinking that the original will eventually
// write the file.
retries := 0
for {
b, err := os.ReadFile(iamFile)
if errors.Is(err, fs.ErrNotExist) {
// racing with someone else updating
// keep retrying after backoff
retries++
if retries < maxretry {
time.Sleep(backoff)
continue
}
// we have been unsuccessful trying to read the iam file
// so this must be the case where something happened and
// the file did not get updated successfully, and probably
// isn't going to be. The recovery procedure would be to
// copy the backup file into place of the original.
return fmt.Errorf("no iam file, needs backup recovery")
}
if err != nil && !errors.Is(err, fs.ErrNotExist) {
return fmt.Errorf("read iam file: %w", err)
}
// reset retries on successful read
retries = 0
err = os.Remove(iamFile)
if errors.Is(err, fs.ErrNotExist) {
// racing with someone else updating
// keep retrying after backoff
time.Sleep(backoff)
continue
}
if err != nil && !errors.Is(err, fs.ErrNotExist) {
return fmt.Errorf("remove old iam file: %w", err)
}
// save copy of data
datacopy := make([]byte, len(b))
copy(datacopy, b)
// make a backup copy in case we crash before update
// this is after remove, so there is a small window something
// can go wrong, but the remove should barrier other gateways
// from trying to write backup at the same time. Only one
// gateway will successfully remove the file.
os.WriteFile(iamBackupFile, b, iamMode)
b, err = update(b)
if err != nil {
// update failed, try to write old data back out
os.WriteFile(iamFile, datacopy, iamMode)
return fmt.Errorf("update iam data: %w", err)
}
err = writeTempFile(b)
if err != nil {
// update failed, try to write old data back out
os.WriteFile(iamFile, datacopy, iamMode)
return err
}
p.iamcache = b
p.iamvalid = true
p.iamexpire = time.Now().Add(cacheDuration)
break
}
return nil
}
func writeTempFile(b []byte) error {
f, err := os.CreateTemp(".", iamFile)
if err != nil {
return fmt.Errorf("create temp file: %w", err)
}
defer os.Remove(f.Name())
_, err = f.Write(b)
if err != nil {
return fmt.Errorf("write temp file: %w", err)
}
err = os.Rename(f.Name(), iamFile)
if err != nil {
return fmt.Errorf("rename temp file: %w", err)
}
return nil
}
func isNoAttr(err error) bool {
if err == nil {
return false