DRY of scoutfs integration, alignment testing for scoutfs.MoveData

backend/posix/posix.go

@@ -39,6 +39,7 @@ import (
 	"github.com/versity/versitygw/auth"
 	"github.com/versity/versitygw/backend"
 	"github.com/versity/versitygw/backend/meta"
+	"github.com/versity/versitygw/s3api/debuglogger"
 	"github.com/versity/versitygw/s3api/utils"
 	"github.com/versity/versitygw/s3err"
 	"github.com/versity/versitygw/s3response"
@@ -1360,6 +1361,10 @@ func getPartChecksum(algo types.ChecksumAlgorithm, part types.CompletedPart) str
 }
 
 func (p *Posix) CompleteMultipartUpload(ctx context.Context, input *s3.CompleteMultipartUploadInput) (s3response.CompleteMultipartUploadResult, string, error) {
+	return p.CompleteMultipartUploadWithCopy(ctx, input, nil)
+}
+
+func (p *Posix) CompleteMultipartUploadWithCopy(ctx context.Context, input *s3.CompleteMultipartUploadInput, customMove func(from *os.File, to *os.File) error) (s3response.CompleteMultipartUploadResult, string, error) {
 	acct, ok := ctx.Value("account").(auth.Account)
 	if !ok {
 		acct = auth.Account{}
@@ -1550,7 +1555,16 @@ func (p *Posix) CompleteMultipartUpload(ctx context.Context, input *s3.CompleteM
 			}
 		}
 
-		_, err = io.Copy(f.File(), rdr)
+		if customMove != nil {
+			err = customMove(pf, f.File())
+			if err != nil {
+				// Fail back to standard copy
+				debuglogger.Logf("Custom data block move failed (%w), failing back to io.Copy()", err)
+				_, err = io.Copy(f.File(), rdr)
+			}
+		} else {
+			_, err = io.Copy(f.File(), rdr)
+		}
 		pf.Close()
 		if err != nil {
 			if errors.Is(err, syscall.EDQUOT) {

backend/scoutfs/scoutfs.go

@@ -21,7 +21,6 @@ import (
 	"errors"
 	"fmt"
 	"io/fs"
-	"net/http"
 	"os"
 	"path/filepath"
 	"strings"
@@ -30,11 +29,9 @@ import (
 	"github.com/aws/aws-sdk-go-v2/service/s3"
 	"github.com/aws/aws-sdk-go-v2/service/s3/types"
 	"github.com/pkg/xattr"
-	"github.com/versity/versitygw/auth"
 	"github.com/versity/versitygw/backend"
 	"github.com/versity/versitygw/backend/meta"
 	"github.com/versity/versitygw/backend/posix"
-	"github.com/versity/versitygw/s3api/utils"
 	"github.com/versity/versitygw/s3err"
 	"github.com/versity/versitygw/s3response"
 )
@@ -67,19 +64,6 @@ type ScoutFS struct {
 	// RestoreObject: add batch stage request to file
 	glaciermode bool
 
-	// chownuid/gid enable chowning of files to the account uid/gid
-	// when objects are uploaded
-	chownuid bool
-	chowngid bool
-
-	// euid/egid are the effective uid/gid of the running versitygw process
-	// used to determine if chowning is needed
-	euid int
-	egid int
-
-	// newDirPerm is the permissions to use when creating new directories
-	newDirPerm fs.FileMode
-
 	// disableNoArchive is used to disable setting scoutam noarchive flag
 	// on mutlipart parts. This is enabled by default to prevent archive
 	// copies of temporary multipart parts.
@@ -91,19 +75,8 @@ var _ backend.Backend = &ScoutFS{}
 const (
 	metaTmpDir          = ".sgwtmp"
 	metaTmpMultipartDir = metaTmpDir + "/multipart"
-	tagHdr              = "X-Amz-Tagging"
-	metaHdr             = "X-Amz-Meta"
-	contentTypeHdr      = "content-type"
-	contentEncHdr       = "content-encoding"
-	contentLangHdr      = "content-language"
-	contentDispHdr      = "content-disposition"
-	cacheCtrlHdr        = "cache-control"
-	expiresHdr          = "expires"
-	emptyMD5            = "d41d8cd98f00b204e9800998ecf8427e"
 	etagkey             = "etag"
 	checksumsKey        = "checksums"
-	objectRetentionKey  = "object-retention"
-	objectLegalHoldKey  = "object-legal-hold"
 )
 
 var (
@@ -146,25 +119,6 @@ func (*ScoutFS) String() string {
 	return "ScoutFS Gateway"
 }
 
-// getChownIDs returns the uid and gid that should be used for chowning
-// the object to the account uid/gid. It also returns a boolean indicating
-// if chowning is needed.
-func (s *ScoutFS) getChownIDs(acct auth.Account) (int, int, bool) {
-	uid := s.euid
-	gid := s.egid
-	var needsChown bool
-	if s.chownuid && acct.UserID != s.euid {
-		uid = acct.UserID
-		needsChown = true
-	}
-	if s.chowngid && acct.GroupID != s.egid {
-		gid = acct.GroupID
-		needsChown = true
-	}
-
-	return uid, gid, needsChown
-}
-
 func (s *ScoutFS) UploadPart(ctx context.Context, input *s3.UploadPartInput) (*s3.UploadPartOutput, error) {
 	out, err := s.Posix.UploadPart(ctx, input)
 	if err != nil {
@@ -194,443 +148,7 @@ func (s *ScoutFS) UploadPart(ctx context.Context, input *s3.UploadPartInput) (*s
 // ioctl to not have to read and copy the part data to the final object. This
 // saves a read and write cycle for all mutlipart uploads.
 func (s *ScoutFS) CompleteMultipartUpload(ctx context.Context, input *s3.CompleteMultipartUploadInput) (s3response.CompleteMultipartUploadResult, string, error) {
-	acct, ok := ctx.Value("account").(auth.Account)
-	if !ok {
-		acct = auth.Account{}
-	}
-
-	var res s3response.CompleteMultipartUploadResult
-
-	if input.Key == nil {
-		return res, "", s3err.GetAPIError(s3err.ErrNoSuchKey)
-	}
-	if input.UploadId == nil {
-		return res, "", s3err.GetAPIError(s3err.ErrNoSuchUpload)
-	}
-	if input.MultipartUpload == nil {
-		return res, "", s3err.GetAPIError(s3err.ErrInvalidRequest)
-	}
-
-	bucket := *input.Bucket
-	object := *input.Key
-	uploadID := *input.UploadId
-	parts := input.MultipartUpload.Parts
-
-	_, err := os.Stat(bucket)
-	if errors.Is(err, fs.ErrNotExist) {
-		return res, "", s3err.GetAPIError(s3err.ErrNoSuchBucket)
-	}
-	if err != nil {
-		return res, "", fmt.Errorf("stat bucket: %w", err)
-	}
-
-	sum, err := s.checkUploadIDExists(bucket, object, uploadID)
-	if err != nil {
-		return res, "", err
-	}
-
-	objdir := filepath.Join(metaTmpMultipartDir, fmt.Sprintf("%x", sum))
-
-	checksums, err := s.retrieveChecksums(nil, bucket, filepath.Join(objdir, uploadID))
-	if err != nil && !errors.Is(err, meta.ErrNoSuchKey) {
-		return res, "", fmt.Errorf("get mp checksums: %w", err)
-	}
-
-	// ChecksumType should be the same as specified on CreateMultipartUpload
-	if input.ChecksumType != "" && checksums.Type != input.ChecksumType {
-		checksumType := checksums.Type
-		if checksumType == "" {
-			checksumType = types.ChecksumType("null")
-		}
-
-		return res, "", s3err.GetChecksumTypeMismatchOnMpErr(checksumType)
-	}
-
-	// check all parts ok
-	last := len(parts) - 1
-	var totalsize int64
-
-	// The initialie values is the lower limit of partNumber: 0
-	var partNumber int32
-	for i, part := range parts {
-		if part.PartNumber == nil {
-			return res, "", s3err.GetAPIError(s3err.ErrInvalidPart)
-		}
-		if *part.PartNumber < 1 {
-			return res, "", s3err.GetAPIError(s3err.ErrInvalidCompleteMpPartNumber)
-		}
-		if *part.PartNumber <= partNumber {
-			return res, "", s3err.GetAPIError(s3err.ErrInvalidPartOrder)
-		}
-
-		partNumber = *part.PartNumber
-
-		partObjPath := filepath.Join(objdir, uploadID, fmt.Sprintf("%v", *part.PartNumber))
-		fullPartPath := filepath.Join(bucket, partObjPath)
-		fi, err := os.Lstat(fullPartPath)
-		if err != nil {
-			return res, "", s3err.GetAPIError(s3err.ErrInvalidPart)
-		}
-
-		totalsize += fi.Size()
-		// all parts except the last need to be greater, thena
-		// the minimum allowed size (5 Mib)
-		if i < last && fi.Size() < backend.MinPartSize {
-			return res, "", s3err.GetAPIError(s3err.ErrEntityTooSmall)
-		}
-
-		b, err := s.meta.RetrieveAttribute(nil, bucket, partObjPath, etagkey)
-		etag := string(b)
-		if err != nil {
-			etag = ""
-		}
-		if parts[i].ETag == nil || !backend.AreEtagsSame(etag, *parts[i].ETag) {
-			return res, "", s3err.GetAPIError(s3err.ErrInvalidPart)
-		}
-
-		partChecksum, err := s.retrieveChecksums(nil, bucket, partObjPath)
-		if err != nil && !errors.Is(err, meta.ErrNoSuchKey) {
-			return res, "", fmt.Errorf("get part checksum: %w", err)
-		}
-
-		// If checksum has been provided on mp initalization
-		err = validatePartChecksum(partChecksum, part)
-		if err != nil {
-			return res, "", err
-		}
-	}
-
-	if input.MpuObjectSize != nil && totalsize != *input.MpuObjectSize {
-		return res, "", s3err.GetIncorrectMpObjectSizeErr(totalsize, *input.MpuObjectSize)
-	}
-
-	// use totalsize=0 because we wont be writing to the file, only moving
-	// extents around.  so we dont want to fallocate this.
-	f, err := s.openTmpFile(filepath.Join(bucket, metaTmpDir), bucket, object, 0, acct)
-	if err != nil {
-		if errors.Is(err, syscall.EDQUOT) {
-			return res, "", s3err.GetAPIError(s3err.ErrQuotaExceeded)
-		}
-		return res, "", fmt.Errorf("open temp file: %w", err)
-	}
-	defer f.cleanup()
-
-	for _, part := range parts {
-		if part.PartNumber == nil || *part.PartNumber < 1 {
-			return res, "", s3err.GetAPIError(s3err.ErrInvalidPart)
-		}
-
-		partObjPath := filepath.Join(objdir, uploadID, fmt.Sprintf("%v", *part.PartNumber))
-		fullPartPath := filepath.Join(bucket, partObjPath)
-		pf, err := os.Open(fullPartPath)
-		if err != nil {
-			return res, "", fmt.Errorf("open part %v: %v", *part.PartNumber, err)
-		}
-
-		// scoutfs move data is a metadata only operation that moves the data
-		// extent references from the source, appeding to the destination.
-		// this needs to be 4k aligned.
-		err = moveData(pf, f.File())
-		pf.Close()
-		if err != nil {
-			return res, "", fmt.Errorf("move blocks part %v: %v", *part.PartNumber, err)
-		}
-	}
-
-	userMetaData := make(map[string]string)
-	upiddir := filepath.Join(objdir, uploadID)
-	objMeta := s.loadUserMetaData(bucket, upiddir, userMetaData)
-	err = s.storeObjectMetadata(f.File(), bucket, object, objMeta)
-	if err != nil {
-		return res, "", err
-	}
-
-	objname := filepath.Join(bucket, object)
-	dir := filepath.Dir(objname)
-	if dir != "" {
-		uid, gid, doChown := s.getChownIDs(acct)
-		err = backend.MkdirAll(dir, uid, gid, doChown, s.newDirPerm)
-		if err != nil {
-			return res, "", err
-		}
-	}
-
-	for k, v := range userMetaData {
-		err = s.meta.StoreAttribute(f.File(), bucket, object, fmt.Sprintf("%v.%v", metaHdr, k), []byte(v))
-		if err != nil {
-			return res, "", fmt.Errorf("set user attr %q: %w", k, err)
-		}
-	}
-
-	// load and set tagging
-	tagging, err := s.meta.RetrieveAttribute(nil, bucket, upiddir, tagHdr)
-	if err != nil && !errors.Is(err, meta.ErrNoSuchKey) {
-		return res, "", fmt.Errorf("get object tagging: %w", err)
-	}
-	if err == nil {
-		err := s.meta.StoreAttribute(f.File(), bucket, object, tagHdr, tagging)
-		if err != nil {
-			return res, "", fmt.Errorf("set object tagging: %w", err)
-		}
-	}
-
-	// load and set legal hold
-	lHold, err := s.meta.RetrieveAttribute(nil, bucket, upiddir, objectLegalHoldKey)
-	if err != nil && !errors.Is(err, meta.ErrNoSuchKey) {
-		return res, "", fmt.Errorf("get object legal hold: %w", err)
-	}
-	if err == nil {
-		err := s.meta.StoreAttribute(f.File(), bucket, object, objectLegalHoldKey, lHold)
-		if err != nil {
-			return res, "", fmt.Errorf("set object legal hold: %w", err)
-		}
-	}
-
-	// load and set retention
-	ret, err := s.meta.RetrieveAttribute(nil, bucket, upiddir, objectRetentionKey)
-	if err != nil && !errors.Is(err, meta.ErrNoSuchKey) {
-		return res, "", fmt.Errorf("get object retention: %w", err)
-	}
-	if err == nil {
-		err := s.meta.StoreAttribute(f.File(), bucket, object, objectRetentionKey, ret)
-		if err != nil {
-			return res, "", fmt.Errorf("set object retention: %w", err)
-		}
-	}
-
-	// Calculate s3 compatible md5sum for complete multipart.
-	s3MD5 := backend.GetMultipartMD5(parts)
-
-	err = s.meta.StoreAttribute(f.File(), bucket, object, etagkey, []byte(s3MD5))
-	if err != nil {
-		return res, "", fmt.Errorf("set etag attr: %w", err)
-	}
-
-	err = f.link()
-	if err != nil {
-		return res, "", fmt.Errorf("link object in namespace: %w", err)
-	}
-
-	// cleanup tmp dirs
-	os.RemoveAll(filepath.Join(bucket, upiddir))
-	// use Remove for objdir in case there are still other uploads
-	// for same object name outstanding
-	os.Remove(filepath.Join(bucket, objdir))
-
-	return s3response.CompleteMultipartUploadResult{
-		Bucket: &bucket,
-		ETag:   &s3MD5,
-		Key:    &object,
-	}, "", nil
-}
-
-func (s *ScoutFS) storeObjectMetadata(f *os.File, bucket, object string, m objectMetadata) error {
-	if getString(m.ContentType) != "" {
-		err := s.meta.StoreAttribute(f, bucket, object, contentTypeHdr, []byte(*m.ContentType))
-		if err != nil {
-			return fmt.Errorf("set content-type: %w", err)
-		}
-	}
-	if getString(m.ContentEncoding) != "" {
-		err := s.meta.StoreAttribute(f, bucket, object, contentEncHdr, []byte(*m.ContentEncoding))
-		if err != nil {
-			return fmt.Errorf("set content-encoding: %w", err)
-		}
-	}
-	if getString(m.ContentDisposition) != "" {
-		err := s.meta.StoreAttribute(f, bucket, object, contentDispHdr, []byte(*m.ContentDisposition))
-		if err != nil {
-			return fmt.Errorf("set content-disposition: %w", err)
-		}
-	}
-	if getString(m.ContentLanguage) != "" {
-		err := s.meta.StoreAttribute(f, bucket, object, contentLangHdr, []byte(*m.ContentLanguage))
-		if err != nil {
-			return fmt.Errorf("set content-language: %w", err)
-		}
-	}
-	if getString(m.CacheControl) != "" {
-		err := s.meta.StoreAttribute(f, bucket, object, cacheCtrlHdr, []byte(*m.CacheControl))
-		if err != nil {
-			return fmt.Errorf("set cache-control: %w", err)
-		}
-	}
-	if getString(m.Expires) != "" {
-		err := s.meta.StoreAttribute(f, bucket, object, expiresHdr, []byte(*m.Expires))
-		if err != nil {
-			return fmt.Errorf("set cache-control: %w", err)
-		}
-	}
-
-	return nil
-}
-
-func validatePartChecksum(checksum s3response.Checksum, part types.CompletedPart) error {
-	n := numberOfChecksums(part)
-	if n > 1 {
-		return s3err.GetAPIError(s3err.ErrInvalidChecksumPart)
-	}
-	if checksum.Algorithm == "" {
-		if n != 0 {
-			return s3err.GetAPIError(s3err.ErrInvalidPart)
-		}
-
-		return nil
-	}
-
-	algo := checksum.Algorithm
-	if n == 0 {
-		return s3err.APIError{
-			Code:           "InvalidRequest",
-			Description:    fmt.Sprintf("The upload was created using a %v checksum. The complete request must include the checksum for each part. It was missing for part %v in the request.", strings.ToLower(string(algo)), *part.PartNumber),
-			HTTPStatusCode: http.StatusBadRequest,
-		}
-	}
-
-	for _, cs := range []struct {
-		checksum         *string
-		expectedChecksum string
-		algo             types.ChecksumAlgorithm
-	}{
-		{part.ChecksumCRC32, getString(checksum.CRC32), types.ChecksumAlgorithmCrc32},
-		{part.ChecksumCRC32C, getString(checksum.CRC32C), types.ChecksumAlgorithmCrc32c},
-		{part.ChecksumSHA1, getString(checksum.SHA1), types.ChecksumAlgorithmSha1},
-		{part.ChecksumSHA256, getString(checksum.SHA256), types.ChecksumAlgorithmSha256},
-		{part.ChecksumCRC64NVME, getString(checksum.CRC64NVME), types.ChecksumAlgorithmCrc64nvme},
-	} {
-		if cs.checksum == nil {
-			continue
-		}
-
-		if !utils.IsValidChecksum(*cs.checksum, cs.algo) {
-			return s3err.GetAPIError(s3err.ErrInvalidChecksumPart)
-		}
-
-		if *cs.checksum != cs.expectedChecksum {
-			if algo == cs.algo {
-				return s3err.GetAPIError(s3err.ErrInvalidPart)
-			}
-
-			return s3err.APIError{
-				Code:           "BadDigest",
-				Description:    fmt.Sprintf("The %v you specified for part %v did not match what we received.", strings.ToLower(string(cs.algo)), *part.PartNumber),
-				HTTPStatusCode: http.StatusBadRequest,
-			}
-		}
-	}
-
-	return nil
-}
-
-func numberOfChecksums(part types.CompletedPart) int {
-	counter := 0
-	if getString(part.ChecksumCRC32) != "" {
-		counter++
-	}
-	if getString(part.ChecksumCRC32C) != "" {
-		counter++
-	}
-	if getString(part.ChecksumSHA1) != "" {
-		counter++
-	}
-	if getString(part.ChecksumSHA256) != "" {
-		counter++
-	}
-	if getString(part.ChecksumCRC64NVME) != "" {
-		counter++
-	}
-
-	return counter
-}
-
-func (s *ScoutFS) checkUploadIDExists(bucket, object, uploadID string) ([32]byte, error) {
-	sum := sha256.Sum256([]byte(object))
-	objdir := filepath.Join(bucket, metaTmpMultipartDir, fmt.Sprintf("%x", sum))
-
-	_, err := os.Stat(filepath.Join(objdir, uploadID))
-	if errors.Is(err, fs.ErrNotExist) {
-		return [32]byte{}, s3err.GetAPIError(s3err.ErrNoSuchUpload)
-	}
-	if err != nil {
-		return [32]byte{}, fmt.Errorf("stat upload: %w", err)
-	}
-	return sum, nil
-}
-
-type objectMetadata struct {
-	ContentType        *string
-	ContentEncoding    *string
-	ContentDisposition *string
-	ContentLanguage    *string
-	CacheControl       *string
-	Expires            *string
-}
-
-// fll out the user metadata map with the metadata for the object
-// and return the content type and encoding
-func (s *ScoutFS) loadUserMetaData(bucket, object string, m map[string]string) objectMetadata {
-	ents, err := s.meta.ListAttributes(bucket, object)
-	if err != nil || len(ents) == 0 {
-		return objectMetadata{}
-	}
-	for _, e := range ents {
-		if !isValidMeta(e) {
-			continue
-		}
-		b, err := s.meta.RetrieveAttribute(nil, bucket, object, e)
-		if err != nil {
-			continue
-		}
-		if b == nil {
-			m[strings.TrimPrefix(e, fmt.Sprintf("%v.", metaHdr))] = ""
-			continue
-		}
-		m[strings.TrimPrefix(e, fmt.Sprintf("%v.", metaHdr))] = string(b)
-	}
-
-	var result objectMetadata
-
-	b, err := s.meta.RetrieveAttribute(nil, bucket, object, contentTypeHdr)
-	if err == nil {
-		result.ContentType = backend.GetPtrFromString(string(b))
-	}
-
-	b, err = s.meta.RetrieveAttribute(nil, bucket, object, contentEncHdr)
-	if err == nil {
-		result.ContentEncoding = backend.GetPtrFromString(string(b))
-	}
-
-	b, err = s.meta.RetrieveAttribute(nil, bucket, object, contentDispHdr)
-	if err == nil {
-		result.ContentDisposition = backend.GetPtrFromString(string(b))
-	}
-
-	b, err = s.meta.RetrieveAttribute(nil, bucket, object, contentLangHdr)
-	if err == nil {
-		result.ContentLanguage = backend.GetPtrFromString(string(b))
-	}
-
-	b, err = s.meta.RetrieveAttribute(nil, bucket, object, cacheCtrlHdr)
-	if err == nil {
-		result.CacheControl = backend.GetPtrFromString(string(b))
-	}
-
-	b, err = s.meta.RetrieveAttribute(nil, bucket, object, expiresHdr)
-	if err == nil {
-		result.Expires = backend.GetPtrFromString(string(b))
-	}
-
-	return result
-}
-
-func isValidMeta(val string) bool {
-	if strings.HasPrefix(val, metaHdr) {
-		return true
-	}
-	if strings.EqualFold(val, "Expires") {
-		return true
-	}
-	return false
+	return s.Posix.CompleteMultipartUploadWithCopy(ctx, input, moveData)
 }
 
 func (s *ScoutFS) HeadObject(ctx context.Context, input *s3.HeadObjectInput) (*s3.HeadObjectOutput, error) {
@@ -956,13 +474,6 @@ func (s *ScoutFS) RestoreObject(_ context.Context, input *s3.RestoreObjectInput)
 	return nil
 }
 
-func getString(str *string) string {
-	if str == nil {
-		return ""
-	}
-	return *str
-}
-
 func isStaging(objname string) (bool, error) {
 	b, err := xattr.Get(objname, flagskey)
 	if err != nil && !isNoAttr(err) {

backend/scoutfs/scoutfs_compat.go

@@ -17,22 +17,13 @@
 package scoutfs
 
 import (
-	"errors"
 	"fmt"
-	"io/fs"
 	"os"
-	"path/filepath"
-	"strconv"
-	"syscall"
-
-	"golang.org/x/sys/unix"
 
 	"github.com/versity/scoutfs-go"
-	"github.com/versity/versitygw/auth"
-	"github.com/versity/versitygw/backend"
 	"github.com/versity/versitygw/backend/meta"
 	"github.com/versity/versitygw/backend/posix"
-	"github.com/versity/versitygw/s3err"
+	"github.com/versity/versitygw/s3api/debuglogger"
 )
 
 func New(rootdir string, opts ScoutfsOpts) (*ScoutFS, error) {
@@ -58,147 +49,31 @@ func New(rootdir string, opts ScoutfsOpts) (*ScoutFS, error) {
 		rootfd:           f,
 		rootdir:          rootdir,
 		meta:             metastore,
-		chownuid:         opts.ChownUID,
-		chowngid:         opts.ChownGID,
 		glaciermode:      opts.GlacierMode,
-		newDirPerm:       opts.NewDirPerm,
 		disableNoArchive: opts.DisableNoArchive,
 	}, nil
 }
 
-const procfddir = "/proc/self/fd"
-
-type tmpfile struct {
-	f          *os.File
-	bucket     string
-	objname    string
-	size       int64
-	needsChown bool
-	uid        int
-	gid        int
-	newDirPerm fs.FileMode
-}
-
-var (
-	defaultFilePerm uint32 = 0644
-)
-
-func (s *ScoutFS) openTmpFile(dir, bucket, obj string, size int64, acct auth.Account) (*tmpfile, error) {
-	uid, gid, doChown := s.getChownIDs(acct)
-
-	// O_TMPFILE allows for a file handle to an unnamed file in the filesystem.
-	// This can help reduce contention within the namespace (parent directories),
-	// etc. And will auto cleanup the inode on close if we never link this
-	// file descriptor into the namespace.
-	fd, err := unix.Open(dir, unix.O_RDWR|unix.O_TMPFILE|unix.O_CLOEXEC, defaultFilePerm)
-	if err != nil {
-		if errors.Is(err, syscall.EROFS) {
-			return nil, s3err.GetAPIError(s3err.ErrMethodNotAllowed)
-		}
-		return nil, err
-	}
-
-	// for O_TMPFILE, filename is /proc/self/fd/<fd> to be used
-	// later to link file into namespace
-	f := os.NewFile(uintptr(fd), filepath.Join(procfddir, strconv.Itoa(fd)))
-
-	tmp := &tmpfile{
-		f:          f,
-		bucket:     bucket,
-		objname:    obj,
-		size:       size,
-		needsChown: doChown,
-		uid:        uid,
-		gid:        gid,
-		newDirPerm: s.newDirPerm,
-	}
-
-	if doChown {
-		err := f.Chown(uid, gid)
-		if err != nil {
-			return nil, fmt.Errorf("set temp file ownership: %w", err)
-		}
-	}
-
-	return tmp, nil
-}
-
-func (tmp *tmpfile) link() error {
-	// We use Linkat/Rename as the atomic operation for object puts. The
-	// upload is written to a temp (or unnamed/O_TMPFILE) file to not conflict
-	// with any other simultaneous uploads. The final operation is to move the
-	// temp file into place for the object. This ensures the object semantics
-	// of last upload completed wins and is not some combination of writes
-	// from simultaneous uploads.
-	objPath := filepath.Join(tmp.bucket, tmp.objname)
-	err := os.Remove(objPath)
-	if err != nil && !errors.Is(err, fs.ErrNotExist) {
-		return fmt.Errorf("remove stale path: %w", err)
-	}
-
-	dir := filepath.Dir(objPath)
-
-	err = backend.MkdirAll(dir, tmp.uid, tmp.gid, tmp.needsChown, tmp.newDirPerm)
-	if err != nil {
-		return fmt.Errorf("make parent dir: %w", err)
-	}
-
-	procdir, err := os.Open(procfddir)
-	if err != nil {
-		return fmt.Errorf("open proc dir: %w", err)
-	}
-	defer procdir.Close()
-
-	dirf, err := os.Open(dir)
-	if err != nil {
-		return fmt.Errorf("open parent dir: %w", err)
-	}
-	defer dirf.Close()
-
-	for {
-		err = unix.Linkat(int(procdir.Fd()), filepath.Base(tmp.f.Name()),
-			int(dirf.Fd()), filepath.Base(objPath), unix.AT_SYMLINK_FOLLOW)
-		if errors.Is(err, fs.ErrExist) {
-			err := os.Remove(objPath)
-			if err != nil && !errors.Is(err, fs.ErrNotExist) {
-				return fmt.Errorf("remove stale path: %w", err)
-			}
-			continue
-		}
-		if err != nil {
-			return fmt.Errorf("link tmpfile: %w", err)
-		}
-		break
-	}
-
-	err = tmp.f.Close()
-	if err != nil {
-		return fmt.Errorf("close tmpfile: %w", err)
-	}
-
-	return nil
-}
-
-func (tmp *tmpfile) Write(b []byte) (int, error) {
-	if int64(len(b)) > tmp.size {
-		return 0, fmt.Errorf("write exceeds content length %v", tmp.size)
-	}
-
-	n, err := tmp.f.Write(b)
-	tmp.size -= int64(n)
-	return n, err
-}
-
-func (tmp *tmpfile) cleanup() {
-	tmp.f.Close()
-}
-
-func (tmp *tmpfile) File() *os.File {
-	return tmp.f
-}
-
 func moveData(from *os.File, to *os.File) error {
-	return scoutfs.MoveData(from, to)
+	// May fail if the files are not 4K aligned; check for alignment
+	ffi, err := from.Stat()
+	if err != nil {
+		return fmt.Errorf("stat from: %v", err)
+	}
+	tfi, err := to.Stat()
+	if err != nil {
+		return fmt.Errorf("stat to: %v", err)
+	}
+	if ffi.Size()%4096 != 0 || tfi.Size()%4096 != 0 {
+		return os.ErrInvalid
+	}
+
+	//
+	err = scoutfs.MoveData(from, to)
+	if err != nil {
+		debuglogger.Logf("ScoutFs MoveData failed: %v", err)
+	}
+	return err
 }
 
 func statMore(path string) (stat, error) {

backend/scoutfs/scoutfs_incompat.go

@@ -20,44 +20,16 @@ import (
 	"errors"
 	"fmt"
 	"os"
-
-	"github.com/versity/versitygw/auth"
 )
 
 func New(rootdir string, opts ScoutfsOpts) (*ScoutFS, error) {
 	return nil, fmt.Errorf("scoutfs only available on linux")
 }
 
-type tmpfile struct{}
-
 var (
 	errNotSupported = errors.New("not supported")
 )
 
-func (s *ScoutFS) openTmpFile(_, _, _ string, _ int64, _ auth.Account) (*tmpfile, error) {
-	// make these look used for static check
-	_ = s.chownuid
-	_ = s.chowngid
-	_ = s.euid
-	_ = s.egid
-	return nil, errNotSupported
-}
-
-func (tmp *tmpfile) link() error {
-	return errNotSupported
-}
-
-func (tmp *tmpfile) Write(b []byte) (int, error) {
-	return 0, errNotSupported
-}
-
-func (tmp *tmpfile) cleanup() {
-}
-
-func (tmp *tmpfile) File() *os.File {
-	return nil
-}
-
 func moveData(_, _ *os.File) error {
 	return errNotSupported
 }