zrepl/replication/logic/replication_logic.go

641 lines
19 KiB
Go
Raw Permalink Normal View History

package logic
import (
"context"
"errors"
"fmt"
"io"
"sync"
"time"
"github.com/prometheus/client_golang/prometheus"
"github.com/zrepl/zrepl/daemon/logging/trace"
"github.com/zrepl/zrepl/logger"
"github.com/zrepl/zrepl/replication/driver"
. "github.com/zrepl/zrepl/replication/logic/diff"
"github.com/zrepl/zrepl/replication/logic/pdu"
"github.com/zrepl/zrepl/replication/report"
"github.com/zrepl/zrepl/util/bytecounter"
"github.com/zrepl/zrepl/util/chainlock"
"github.com/zrepl/zrepl/util/semaphore"
"github.com/zrepl/zrepl/zfs"
)
// Endpoint represents one side of the replication.
//
// An endpoint is either in Sender or Receiver mode, represented by the correspondingly
// named interfaces defined in this package.
type Endpoint interface {
// Does not include placeholder filesystems
ListFilesystems(ctx context.Context, req *pdu.ListFilesystemReq) (*pdu.ListFilesystemRes, error)
ListFilesystemVersions(ctx context.Context, req *pdu.ListFilesystemVersionsReq) (*pdu.ListFilesystemVersionsRes, error)
DestroySnapshots(ctx context.Context, req *pdu.DestroySnapshotsReq) (*pdu.DestroySnapshotsRes, error)
2019-03-22 19:41:12 +01:00
WaitForConnectivity(ctx context.Context) error
}
type Sender interface {
Endpoint
// If a non-nil io.ReadCloser is returned, it is guaranteed to be closed before
// any next call to the parent github.com/zrepl/zrepl/replication.Endpoint.
// If the send request is for dry run the io.ReadCloser will be nil
Send(ctx context.Context, r *pdu.SendReq) (*pdu.SendRes, io.ReadCloser, error)
SendDry(ctx context.Context, r *pdu.SendReq) (*pdu.SendRes, error)
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
SendCompleted(ctx context.Context, r *pdu.SendCompletedReq) (*pdu.SendCompletedRes, error)
ReplicationCursor(ctx context.Context, req *pdu.ReplicationCursorReq) (*pdu.ReplicationCursorRes, error)
}
type Receiver interface {
Endpoint
// Receive sends r and sendStream (the latter containing a ZFS send stream)
// to the parent github.com/zrepl/zrepl/replication.Endpoint.
Receive(ctx context.Context, req *pdu.ReceiveReq, receive io.ReadCloser) (*pdu.ReceiveRes, error)
}
type Planner struct {
sender Sender
receiver Receiver
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
policy PlannerPolicy
promSecsPerState *prometheus.HistogramVec // labels: state
promBytesReplicated *prometheus.CounterVec // labels: filesystem
}
func (p *Planner) Plan(ctx context.Context) ([]driver.FS, error) {
fss, err := p.doPlanning(ctx)
if err != nil {
return nil, err
}
dfss := make([]driver.FS, len(fss))
for i := range dfss {
dfss[i] = fss[i]
}
return dfss, nil
}
func (p *Planner) WaitForConnectivity(ctx context.Context) error {
var wg sync.WaitGroup
doPing := func(endpoint Endpoint, errOut *error) {
defer wg.Done()
ctx, endTask := trace.WithTaskFromStack(ctx)
defer endTask()
err := endpoint.WaitForConnectivity(ctx)
if err != nil {
*errOut = err
} else {
*errOut = nil
}
}
wg.Add(2)
var senderErr, receiverErr error
go doPing(p.sender, &senderErr)
go doPing(p.receiver, &receiverErr)
wg.Wait()
if senderErr == nil && receiverErr == nil {
return nil
} else if senderErr != nil && receiverErr != nil {
if senderErr.Error() == receiverErr.Error() {
return fmt.Errorf("sender and receiver are not reachable: %s", senderErr.Error())
} else {
return fmt.Errorf("sender and receiver are not reachable:\n sender: %s\n receiver: %s", senderErr, receiverErr)
}
} else {
var side string
var err *error
if senderErr != nil {
side = "sender"
err = &senderErr
} else {
side = "receiver"
err = &receiverErr
}
return fmt.Errorf("%s is not reachable: %s", side, *err)
}
}
type Filesystem struct {
sender Sender
receiver Receiver
policy PlannerPolicy // immutable, it's .ReplicationConfig member is a pointer and copied into messages
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
Path string // compat
receiverFS, senderFS *pdu.Filesystem // receiverFS may be nil, senderFS never nil
promBytesReplicated prometheus.Counter // compat
sizeEstimateRequestSem *semaphore.S
}
func (f *Filesystem) EqualToPreviousAttempt(other driver.FS) bool {
g, ok := other.(*Filesystem)
if !ok {
return false
}
// TODO: use GUIDs (issued by zrepl, not those from ZFS)
return f.Path == g.Path
}
func (f *Filesystem) PlanFS(ctx context.Context) ([]driver.Step, error) {
steps, err := f.doPlanning(ctx)
if err != nil {
return nil, err
}
dsteps := make([]driver.Step, len(steps))
for i := range dsteps {
dsteps[i] = steps[i]
}
return dsteps, nil
}
func (f *Filesystem) ReportInfo() *report.FilesystemInfo {
return &report.FilesystemInfo{Name: f.Path} // FIXME compat name
}
type Step struct {
sender Sender
receiver Receiver
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
parent *Filesystem
from, to *pdu.FilesystemVersion // from may be nil, indicating full send
rework resume token validation to allow resuming from raw sends of unencrypted datasets Before this change, resuming from an unencrypted dataset with send.raw=true specified wouldn't work with zrepl due to overly restrictive resume token checking. An initial PR to fix this was made in https://github.com/zrepl/zrepl/pull/503 but it didn't address the core of the problem. The core of the problem was that zrepl assumed that if a resume token contained `rawok=true, compressok=true`, the resulting send would be encrypted. But if the sender dataset was unencrypted, such a resume would actually result in an unencrypted send. Which could be totally legitimate but zrepl failed to recognize that. BACKGROUND ========== The following snippets of OpenZFS code are insightful regarding how the various ${X}ok values in the resume token are handled: - https://github.com/openzfs/zfs/blob/6c3c5fcfbe27d9193cd131753cc7e47ee2784621/module/zfs/dmu_send.c#L1947-L2012 - https://github.com/openzfs/zfs/blob/6c3c5fcfbe27d9193cd131753cc7e47ee2784621/module/zfs/dmu_recv.c#L877-L891 - https://github.com/openzfs/zfs/blob/6c3c5fc/lib/libzfs/libzfs_sendrecv.c#L1663-L1672 Basically, some zfs send flags make the DMU send code set some DMU send stream featureflags, although it's not a pure mapping, i.e, which DMU send stream flags are used depends somewhat on the dataset (e.g., is it encrypted or not, or, does it use zstd or not). Then, the receiver looks at some (but not all) feature flags and maps them to ${X}ok dataset zap attributes. These are funnelled back to the sender 1:1 through the resume_token. And the sender turns them into lzc flags. As an example, let's look at zfs send --raw. if the sender requests a raw send on an unencrypted dataset, the send stream (and hence the resume token) will not have the raw stream featureflag set, and hence the resume token will not have the rawok field set. Instead, it will have compressok, embedok, and depending on whether large blocks are present in the dataset, largeblockok set. WHAT'S ZREPL'S ROLE IN THIS? ============================ zrepl provides a virtual encrypted sendflag that is like `raw`, but further ensures that we only send encrypted datasets. For any other resume token stuff, it shoudn't do any checking, because it's a futile effort to keep up with ZFS send/recv features that are orthogonal to encryption. CHANGES MADE IN THIS COMMIT =========================== - Rip out a bunch of needless checking that zrepl would do during planning. These checks were there to give better error messages, but actually, the error messages created by the endpoint.Sender.Send RPC upon send args validation failure are good enough. - Add platformtests to validate all combinations of (Unencrypted/Encrypted FS) x (send.encrypted = true | false) x (send.raw = true | false) for cases both non-resuming and resuming send. Additional manual testing done: 1. With zrepl 0.5, setup with unencrypted dataset, send.raw=true specified, no send.encrypted specified. 2. Observe that regular non-resuming send works, but resuming doesn't work. 3. Upgrade zrepl to this change. 4. Observe that both regular and resuming send works. closes https://github.com/zrepl/zrepl/pull/613
2022-07-10 14:56:35 +02:00
resumeToken string // empty means no resume token shall be used
expectedSize uint64 // 0 means no size estimate present / possible
// byteCounter is nil initially, and set later in Step.doReplication
// => concurrent read of that pointer from Step.ReportInfo must be protected
byteCounter bytecounter.ReadCloser
byteCounterMtx chainlock.L
}
func (s *Step) TargetEquals(other driver.Step) bool {
t, ok := other.(*Step)
if !ok {
return false
}
if !s.parent.EqualToPreviousAttempt(t.parent) {
panic("Step interface promise broken: parent filesystems must be same")
}
return s.from.GetGuid() == t.from.GetGuid() &&
s.to.GetGuid() == t.to.GetGuid()
}
func (s *Step) TargetDate() time.Time {
return s.to.SnapshotTime() // FIXME compat name
}
func (s *Step) Step(ctx context.Context) error {
return s.doReplication(ctx)
}
func (s *Step) ReportInfo() *report.StepInfo {
// get current byteCounter value
var byteCounter uint64
s.byteCounterMtx.Lock()
if s.byteCounter != nil {
byteCounter = s.byteCounter.Count()
}
s.byteCounterMtx.Unlock()
from := ""
if s.from != nil {
from = s.from.RelName()
}
return &report.StepInfo{
From: from,
To: s.to.RelName(),
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
Resumed: s.resumeToken != "",
BytesExpected: s.expectedSize,
BytesReplicated: byteCounter,
}
}
// caller must ensure policy.Validate() == nil
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
func NewPlanner(secsPerState *prometheus.HistogramVec, bytesReplicated *prometheus.CounterVec, sender Sender, receiver Receiver, policy PlannerPolicy) *Planner {
if err := policy.Validate(); err != nil {
panic(err)
}
return &Planner{
sender: sender,
receiver: receiver,
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
policy: policy,
promSecsPerState: secsPerState,
promBytesReplicated: bytesReplicated,
}
}
func tryAutoresolveConflict(conflict error, policy ConflictResolution) (path []*pdu.FilesystemVersion, reason error) {
if _, ok := conflict.(*ConflictMostRecentSnapshotAlreadyPresent); ok {
// replicatoin is a no-op
return nil, nil
}
if noCommonAncestor, ok := conflict.(*ConflictNoCommonAncestor); ok {
if len(noCommonAncestor.SortedReceiverVersions) == 0 {
if len(noCommonAncestor.SortedSenderVersions) == 0 {
return nil, fmt.Errorf("no snapshots available on sender side")
}
switch policy.InitialReplication {
case InitialReplicationAutoResolutionMostRecent:
var mostRecentSnap *pdu.FilesystemVersion
for n := len(noCommonAncestor.SortedSenderVersions) - 1; n >= 0; n-- {
if noCommonAncestor.SortedSenderVersions[n].Type == pdu.FilesystemVersion_Snapshot {
mostRecentSnap = noCommonAncestor.SortedSenderVersions[n]
break
}
}
return []*pdu.FilesystemVersion{nil, mostRecentSnap}, nil
case InitialReplicationAutoResolutionAll:
path = append(path, nil)
for n := 0; n < len(noCommonAncestor.SortedSenderVersions); n++ {
if noCommonAncestor.SortedSenderVersions[n].Type == pdu.FilesystemVersion_Snapshot {
path = append(path, noCommonAncestor.SortedSenderVersions[n])
}
}
return path, nil
case InitialReplicationAutoResolutionFail:
return nil, fmt.Errorf("automatic conflict resolution for initial replication is disabled in config")
default:
panic(fmt.Sprintf("unimplemented: %#v", policy.InitialReplication))
}
}
}
return nil, conflict
}
func (p *Planner) doPlanning(ctx context.Context) ([]*Filesystem, error) {
log := getLogger(ctx)
log.Info("start planning")
slfssres, err := p.sender.ListFilesystems(ctx, &pdu.ListFilesystemReq{})
if err != nil {
log.WithError(err).WithField("errType", fmt.Sprintf("%T", err)).Error("error listing sender filesystems")
return nil, err
}
sfss := slfssres.GetFilesystems()
rlfssres, err := p.receiver.ListFilesystems(ctx, &pdu.ListFilesystemReq{})
if err != nil {
log.WithError(err).WithField("errType", fmt.Sprintf("%T", err)).Error("error listing receiver filesystems")
return nil, err
}
rfss := rlfssres.GetFilesystems()
sizeEstimateRequestSem := semaphore.New(int64(p.policy.SizeEstimationConcurrency))
q := make([]*Filesystem, 0, len(sfss))
for _, fs := range sfss {
var receiverFS *pdu.Filesystem
for _, rfs := range rfss {
if rfs.Path == fs.Path {
receiverFS = rfs
}
}
var ctr prometheus.Counter
if p.promBytesReplicated != nil {
ctr = p.promBytesReplicated.WithLabelValues(fs.Path)
}
q = append(q, &Filesystem{
sender: p.sender,
receiver: p.receiver,
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
policy: p.policy,
Path: fs.Path,
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
senderFS: fs,
receiverFS: receiverFS,
promBytesReplicated: ctr,
sizeEstimateRequestSem: sizeEstimateRequestSem,
})
}
return q, nil
}
func (fs *Filesystem) doPlanning(ctx context.Context) ([]*Step, error) {
log := func(ctx context.Context) logger.Logger {
return getLogger(ctx).WithField("filesystem", fs.Path)
}
log(ctx).Debug("assessing filesystem")
sfsvsres, err := fs.sender.ListFilesystemVersions(ctx, &pdu.ListFilesystemVersionsReq{Filesystem: fs.Path})
if err != nil {
log(ctx).WithError(err).Error("cannot get remote filesystem versions")
return nil, err
}
sfsvs := sfsvsres.GetVersions()
if len(sfsvs) < 1 {
err := errors.New("sender does not have any versions")
log(ctx).Error(err.Error())
return nil, err
}
var rfsvs []*pdu.FilesystemVersion
if fs.receiverFS != nil && !fs.receiverFS.GetIsPlaceholder() {
rfsvsres, err := fs.receiver.ListFilesystemVersions(ctx, &pdu.ListFilesystemVersionsReq{Filesystem: fs.Path})
if err != nil {
log(ctx).WithError(err).Error("receiver error")
return nil, err
}
rfsvs = rfsvsres.GetVersions()
} else {
rfsvs = []*pdu.FilesystemVersion{}
}
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
var resumeToken *zfs.ResumeToken
var resumeTokenRaw string
if fs.receiverFS != nil && fs.receiverFS.ResumeToken != "" {
resumeTokenRaw = fs.receiverFS.ResumeToken // shadow
log(ctx).WithField("receiverFS.ResumeToken", resumeTokenRaw).Debug("decode receiver fs resume token")
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
resumeToken, err = zfs.ParseResumeToken(ctx, resumeTokenRaw) // shadow
if err != nil {
// TODO in theory, we could do replication without resume token, but that would mean that
// we need to discard the resumable state on the receiver's side.
// Would be easy by setting UsedResumeToken=false in the RecvReq ...
// FIXME / CHECK semantics UsedResumeToken if SendReq.ResumeToken == ""
log(ctx).WithError(err).Error("cannot decode resume token, aborting")
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
return nil, err
}
log(ctx).WithField("token", resumeToken).Debug("decode resume token")
}
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
var steps []*Step
// build the list of replication steps
//
// prefer to resume any started replication instead of starting over with a normal IncrementalPath
//
// look for the step encoded in the resume token in the sender's version
// if we find that step:
// 1. use it as first step (including resume token)
// 2. compute subsequent steps by computing incremental path from the token.To version on
// ...
// that's actually equivalent to simply cutting off earlier versions from rfsvs and sfsvs
if resumeToken != nil {
sfsvs := SortVersionListByCreateTXGThenBookmarkLTSnapshot(sfsvs)
var fromVersion, toVersion *pdu.FilesystemVersion
var toVersionIdx int
for idx, sfsv := range sfsvs {
if resumeToken.HasFromGUID && sfsv.Guid == resumeToken.FromGUID {
if fromVersion != nil && fromVersion.Type == pdu.FilesystemVersion_Snapshot {
// prefer snapshots over bookmarks for size estimation
} else {
fromVersion = sfsv
}
}
if resumeToken.HasToGUID && sfsv.Guid == resumeToken.ToGUID && sfsv.Type == pdu.FilesystemVersion_Snapshot {
// `toversion` must always be a snapshot
toVersion, toVersionIdx = sfsv, idx
}
}
rework resume token validation to allow resuming from raw sends of unencrypted datasets Before this change, resuming from an unencrypted dataset with send.raw=true specified wouldn't work with zrepl due to overly restrictive resume token checking. An initial PR to fix this was made in https://github.com/zrepl/zrepl/pull/503 but it didn't address the core of the problem. The core of the problem was that zrepl assumed that if a resume token contained `rawok=true, compressok=true`, the resulting send would be encrypted. But if the sender dataset was unencrypted, such a resume would actually result in an unencrypted send. Which could be totally legitimate but zrepl failed to recognize that. BACKGROUND ========== The following snippets of OpenZFS code are insightful regarding how the various ${X}ok values in the resume token are handled: - https://github.com/openzfs/zfs/blob/6c3c5fcfbe27d9193cd131753cc7e47ee2784621/module/zfs/dmu_send.c#L1947-L2012 - https://github.com/openzfs/zfs/blob/6c3c5fcfbe27d9193cd131753cc7e47ee2784621/module/zfs/dmu_recv.c#L877-L891 - https://github.com/openzfs/zfs/blob/6c3c5fc/lib/libzfs/libzfs_sendrecv.c#L1663-L1672 Basically, some zfs send flags make the DMU send code set some DMU send stream featureflags, although it's not a pure mapping, i.e, which DMU send stream flags are used depends somewhat on the dataset (e.g., is it encrypted or not, or, does it use zstd or not). Then, the receiver looks at some (but not all) feature flags and maps them to ${X}ok dataset zap attributes. These are funnelled back to the sender 1:1 through the resume_token. And the sender turns them into lzc flags. As an example, let's look at zfs send --raw. if the sender requests a raw send on an unencrypted dataset, the send stream (and hence the resume token) will not have the raw stream featureflag set, and hence the resume token will not have the rawok field set. Instead, it will have compressok, embedok, and depending on whether large blocks are present in the dataset, largeblockok set. WHAT'S ZREPL'S ROLE IN THIS? ============================ zrepl provides a virtual encrypted sendflag that is like `raw`, but further ensures that we only send encrypted datasets. For any other resume token stuff, it shoudn't do any checking, because it's a futile effort to keep up with ZFS send/recv features that are orthogonal to encryption. CHANGES MADE IN THIS COMMIT =========================== - Rip out a bunch of needless checking that zrepl would do during planning. These checks were there to give better error messages, but actually, the error messages created by the endpoint.Sender.Send RPC upon send args validation failure are good enough. - Add platformtests to validate all combinations of (Unencrypted/Encrypted FS) x (send.encrypted = true | false) x (send.raw = true | false) for cases both non-resuming and resuming send. Additional manual testing done: 1. With zrepl 0.5, setup with unencrypted dataset, send.raw=true specified, no send.encrypted specified. 2. Observe that regular non-resuming send works, but resuming doesn't work. 3. Upgrade zrepl to this change. 4. Observe that both regular and resuming send works. closes https://github.com/zrepl/zrepl/pull/613
2022-07-10 14:56:35 +02:00
if toVersion == nil {
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
return nil, fmt.Errorf("resume token `toguid` = %v not found on sender (`toname` = %q)", resumeToken.ToGUID, resumeToken.ToName)
} else if fromVersion == toVersion {
return nil, fmt.Errorf("resume token `fromguid` and `toguid` match same version on sener")
}
// fromVersion may be nil, toVersion is no nil, encryption matches
// good to go this one step!
resumeStep := &Step{
parent: fs,
sender: fs.sender,
receiver: fs.receiver,
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
rework resume token validation to allow resuming from raw sends of unencrypted datasets Before this change, resuming from an unencrypted dataset with send.raw=true specified wouldn't work with zrepl due to overly restrictive resume token checking. An initial PR to fix this was made in https://github.com/zrepl/zrepl/pull/503 but it didn't address the core of the problem. The core of the problem was that zrepl assumed that if a resume token contained `rawok=true, compressok=true`, the resulting send would be encrypted. But if the sender dataset was unencrypted, such a resume would actually result in an unencrypted send. Which could be totally legitimate but zrepl failed to recognize that. BACKGROUND ========== The following snippets of OpenZFS code are insightful regarding how the various ${X}ok values in the resume token are handled: - https://github.com/openzfs/zfs/blob/6c3c5fcfbe27d9193cd131753cc7e47ee2784621/module/zfs/dmu_send.c#L1947-L2012 - https://github.com/openzfs/zfs/blob/6c3c5fcfbe27d9193cd131753cc7e47ee2784621/module/zfs/dmu_recv.c#L877-L891 - https://github.com/openzfs/zfs/blob/6c3c5fc/lib/libzfs/libzfs_sendrecv.c#L1663-L1672 Basically, some zfs send flags make the DMU send code set some DMU send stream featureflags, although it's not a pure mapping, i.e, which DMU send stream flags are used depends somewhat on the dataset (e.g., is it encrypted or not, or, does it use zstd or not). Then, the receiver looks at some (but not all) feature flags and maps them to ${X}ok dataset zap attributes. These are funnelled back to the sender 1:1 through the resume_token. And the sender turns them into lzc flags. As an example, let's look at zfs send --raw. if the sender requests a raw send on an unencrypted dataset, the send stream (and hence the resume token) will not have the raw stream featureflag set, and hence the resume token will not have the rawok field set. Instead, it will have compressok, embedok, and depending on whether large blocks are present in the dataset, largeblockok set. WHAT'S ZREPL'S ROLE IN THIS? ============================ zrepl provides a virtual encrypted sendflag that is like `raw`, but further ensures that we only send encrypted datasets. For any other resume token stuff, it shoudn't do any checking, because it's a futile effort to keep up with ZFS send/recv features that are orthogonal to encryption. CHANGES MADE IN THIS COMMIT =========================== - Rip out a bunch of needless checking that zrepl would do during planning. These checks were there to give better error messages, but actually, the error messages created by the endpoint.Sender.Send RPC upon send args validation failure are good enough. - Add platformtests to validate all combinations of (Unencrypted/Encrypted FS) x (send.encrypted = true | false) x (send.raw = true | false) for cases both non-resuming and resuming send. Additional manual testing done: 1. With zrepl 0.5, setup with unencrypted dataset, send.raw=true specified, no send.encrypted specified. 2. Observe that regular non-resuming send works, but resuming doesn't work. 3. Upgrade zrepl to this change. 4. Observe that both regular and resuming send works. closes https://github.com/zrepl/zrepl/pull/613
2022-07-10 14:56:35 +02:00
from: fromVersion,
to: toVersion,
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
resumeToken: resumeTokenRaw,
}
// by definition, the resume token _must_ be the receiver's most recent version, if they have any
// don't bother checking, zfs recv will produce an error if above assumption is wrong
//
// thus, subsequent steps are just incrementals on the sender's remaining _snapshots_ (not bookmarks)
var remainingSFSVs []*pdu.FilesystemVersion
for _, sfsv := range sfsvs[toVersionIdx:] {
if sfsv.Type == pdu.FilesystemVersion_Snapshot {
remainingSFSVs = append(remainingSFSVs, sfsv)
}
}
steps = make([]*Step, 0, len(remainingSFSVs)) // shadow
steps = append(steps, resumeStep)
for i := 0; i < len(remainingSFSVs)-1; i++ {
steps = append(steps, &Step{
parent: fs,
sender: fs.sender,
receiver: fs.receiver,
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
from: remainingSFSVs[i],
to: remainingSFSVs[i+1],
})
}
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
} else { // resumeToken == nil
path, conflict := IncrementalPath(rfsvs, sfsvs)
if conflict != nil {
updPath, updConflict := tryAutoresolveConflict(conflict, *fs.policy.ConflictResolution)
if updConflict == nil {
log(ctx).WithField("conflict", conflict).Info("conflict automatically resolved")
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
} else {
log(ctx).WithField("conflict", conflict).Error("cannot resolve conflict")
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
}
path, conflict = updPath, updConflict
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
}
if conflict != nil {
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
return nil, conflict
}
if len(path) == 0 {
steps = nil
} else if len(path) == 1 {
panic(fmt.Sprintf("len(path) must be two for incremental repl, and initial repl must start with nil, got path[0]=%#v", path[0]))
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
} else {
steps = make([]*Step, 0, len(path)) // shadow
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
for i := 0; i < len(path)-1; i++ {
steps = append(steps, &Step{
parent: fs,
sender: fs.sender,
receiver: fs.receiver,
rework resume token validation to allow resuming from raw sends of unencrypted datasets Before this change, resuming from an unencrypted dataset with send.raw=true specified wouldn't work with zrepl due to overly restrictive resume token checking. An initial PR to fix this was made in https://github.com/zrepl/zrepl/pull/503 but it didn't address the core of the problem. The core of the problem was that zrepl assumed that if a resume token contained `rawok=true, compressok=true`, the resulting send would be encrypted. But if the sender dataset was unencrypted, such a resume would actually result in an unencrypted send. Which could be totally legitimate but zrepl failed to recognize that. BACKGROUND ========== The following snippets of OpenZFS code are insightful regarding how the various ${X}ok values in the resume token are handled: - https://github.com/openzfs/zfs/blob/6c3c5fcfbe27d9193cd131753cc7e47ee2784621/module/zfs/dmu_send.c#L1947-L2012 - https://github.com/openzfs/zfs/blob/6c3c5fcfbe27d9193cd131753cc7e47ee2784621/module/zfs/dmu_recv.c#L877-L891 - https://github.com/openzfs/zfs/blob/6c3c5fc/lib/libzfs/libzfs_sendrecv.c#L1663-L1672 Basically, some zfs send flags make the DMU send code set some DMU send stream featureflags, although it's not a pure mapping, i.e, which DMU send stream flags are used depends somewhat on the dataset (e.g., is it encrypted or not, or, does it use zstd or not). Then, the receiver looks at some (but not all) feature flags and maps them to ${X}ok dataset zap attributes. These are funnelled back to the sender 1:1 through the resume_token. And the sender turns them into lzc flags. As an example, let's look at zfs send --raw. if the sender requests a raw send on an unencrypted dataset, the send stream (and hence the resume token) will not have the raw stream featureflag set, and hence the resume token will not have the rawok field set. Instead, it will have compressok, embedok, and depending on whether large blocks are present in the dataset, largeblockok set. WHAT'S ZREPL'S ROLE IN THIS? ============================ zrepl provides a virtual encrypted sendflag that is like `raw`, but further ensures that we only send encrypted datasets. For any other resume token stuff, it shoudn't do any checking, because it's a futile effort to keep up with ZFS send/recv features that are orthogonal to encryption. CHANGES MADE IN THIS COMMIT =========================== - Rip out a bunch of needless checking that zrepl would do during planning. These checks were there to give better error messages, but actually, the error messages created by the endpoint.Sender.Send RPC upon send args validation failure are good enough. - Add platformtests to validate all combinations of (Unencrypted/Encrypted FS) x (send.encrypted = true | false) x (send.raw = true | false) for cases both non-resuming and resuming send. Additional manual testing done: 1. With zrepl 0.5, setup with unencrypted dataset, send.raw=true specified, no send.encrypted specified. 2. Observe that regular non-resuming send works, but resuming doesn't work. 3. Upgrade zrepl to this change. 4. Observe that both regular and resuming send works. closes https://github.com/zrepl/zrepl/pull/613
2022-07-10 14:56:35 +02:00
from: path[i], // nil in case of initial repl
to: path[i+1],
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
})
}
}
}
if len(steps) == 0 {
log(ctx).Info("planning determined that no replication steps are required")
}
log(ctx).Debug("compute send size estimate")
errs := make(chan error, len(steps))
fanOutCtx, fanOutCancel := context.WithCancel(ctx)
_, fanOutAdd, fanOutWait := trace.WithTaskGroup(fanOutCtx, "compute-size-estimate")
defer fanOutCancel()
for _, step := range steps {
step := step // local copy that is moved into the closure
fanOutAdd(func(ctx context.Context) {
// TODO instead of the semaphore, rely on resource-exhaustion signaled by the remote endpoint to limit size-estimate requests
// Send is handled over rpc/dataconn ATM, which doesn't support the resource exhaustion status codes that gRPC defines
guard, err := fs.sizeEstimateRequestSem.Acquire(ctx)
if err != nil {
fanOutCancel()
return
}
defer guard.Release()
err = step.updateSizeEstimate(ctx)
if err != nil {
log(ctx).WithError(err).WithField("step", step).Error("error computing size estimate")
fanOutCancel()
}
errs <- err
})
}
fanOutWait()
close(errs)
var significantErr error = nil
for err := range errs {
if err != nil {
if significantErr == nil || significantErr == context.Canceled {
significantErr = err
}
}
}
if significantErr != nil {
return nil, significantErr
}
log(ctx).Debug("filesystem planning finished")
return steps, nil
}
func (s *Step) updateSizeEstimate(ctx context.Context) error {
log := getLogger(ctx)
sr := s.buildSendRequest()
log.Debug("initiate dry run send request")
sres, err := s.sender.SendDry(ctx, sr)
if err != nil {
log.WithError(err).Error("dry run send request failed")
return err
}
if sres == nil {
err := fmt.Errorf("dry run send request returned nil send result")
log.Error(err.Error())
return err
}
s.expectedSize = sres.GetExpectedSize()
return nil
}
func (s *Step) buildSendRequest() (sr *pdu.SendReq) {
fs := s.parent.Path
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
sr = &pdu.SendReq{
Filesystem: fs,
From: s.from, // may be nil
To: s.to,
ResumeToken: s.resumeToken,
ReplicationConfig: s.parent.policy.ReplicationConfig,
}
return sr
}
func (s *Step) doReplication(ctx context.Context) error {
fs := s.parent.Path
log := getLogger(ctx).WithField("filesystem", fs)
sr := s.buildSendRequest()
log.Debug("initiate send request")
sres, stream, err := s.sender.Send(ctx, sr)
if err != nil {
log.WithError(err).Error("send request failed")
return err
}
if sres == nil {
err := fmt.Errorf("send request returned nil send result")
log.Error(err.Error())
return err
}
if stream == nil {
err := errors.New("send request did not return a stream, broken endpoint implementation")
return err
}
defer stream.Close()
// Install a byte counter to track progress + for status report
byteCountingStream := bytecounter.NewReadCloser(stream)
s.byteCounterMtx.Lock()
s.byteCounter = byteCountingStream
s.byteCounterMtx.Unlock()
defer func() {
defer s.byteCounterMtx.Lock().Unlock()
if s.parent.promBytesReplicated != nil {
s.parent.promBytesReplicated.Add(float64(s.byteCounter.Count()))
}
}()
rr := &pdu.ReceiveReq{
Filesystem: fs,
To: sr.GetTo(),
ClearResumeToken: !sres.UsedResumeToken,
ReplicationConfig: s.parent.policy.ReplicationConfig,
}
log.Debug("initiate receive request")
_, err = s.receiver.Receive(ctx, rr, byteCountingStream)
if err != nil {
log.
WithError(err).
WithField("errType", fmt.Sprintf("%T", err)).
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
WithField("rr", fmt.Sprintf("%v", rr)).
Error("receive request failed (might also be error on sender)")
// This failure could be due to
// - an unexpected exit of ZFS on the sending side
// - an unexpected exit of ZFS on the receiving side
// - a connectivity issue
return err
}
log.Debug("receive finished")
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
log.Debug("tell sender replication completed")
_, err = s.sender.SendCompleted(ctx, &pdu.SendCompletedReq{
OriginalReq: sr,
})
if err != nil {
new features: {resumable,encrypted,hold-protected} send-recv, last-received-hold - **Resumable Send & Recv Support** No knobs required, automatically used where supported. - **Hold-Protected Send & Recv** Automatic ZFS holds to ensure that we can always resume a replication step. - **Encrypted Send & Recv Support** for OpenZFS native encryption. Configurable at the job level, i.e., for all filesystems a job is responsible for. - **Receive-side hold on last received dataset** The counterpart to the replication cursor bookmark on the send-side. Ensures that incremental replication will always be possible between a sender and receiver. Design Doc ---------- `replication/design.md` doc describes how we use ZFS holds and bookmarks to ensure that a single replication step is always resumable. The replication algorithm described in the design doc introduces the notion of job IDs (please read the details on this design doc). We reuse the job names for job IDs and use `JobID` type to ensure that a job name can be embedded into hold tags, bookmark names, etc. This might BREAK CONFIG on upgrade. Protocol Version Bump --------------------- This commit makes backwards-incompatible changes to the replication/pdu protobufs. Thus, bump the version number used in the protocol handshake. Replication Cursor Format Change -------------------------------- The new replication cursor bookmark format is: `#zrepl_CURSOR_G_${this.GUID}_J_${jobid}` Including the GUID enables transaction-safe moving-forward of the cursor. Including the job id enables that multiple sending jobs can send the same filesystem without interfering. The `zrepl migrate replication-cursor:v1-v2` subcommand can be used to safely destroy old-format cursors once zrepl has created new-format cursors. Changes in This Commit ---------------------- - package zfs - infrastructure for holds - infrastructure for resume token decoding - implement a variant of OpenZFS's `entity_namecheck` and use it for validation in new code - ZFSSendArgs to specify a ZFS send operation - validation code protects against malicious resume tokens by checking that the token encodes the same send parameters that the send-side would use if no resume token were available (i.e. same filesystem, `fromguid`, `toguid`) - RecvOptions support for `recv -s` flag - convert a bunch of ZFS operations to be idempotent - achieved through more differentiated error message scraping / additional pre-/post-checks - package replication/pdu - add field for encryption to send request messages - add fields for resume handling to send & recv request messages - receive requests now contain `FilesystemVersion To` in addition to the filesystem into which the stream should be `recv`d into - can use `zfs recv $root_fs/$client_id/path/to/dataset@${To.Name}`, which enables additional validation after recv (i.e. whether `To.Guid` matched what we received in the stream) - used to set `last-received-hold` - package replication/logic - introduce `PlannerPolicy` struct, currently only used to configure whether encrypted sends should be requested from the sender - integrate encryption and resume token support into `Step` struct - package endpoint - move the concepts that endpoint builds on top of ZFS to a single file `endpoint/endpoint_zfs.go` - step-holds + step-bookmarks - last-received-hold - new replication cursor + old replication cursor compat code - adjust `endpoint/endpoint.go` handlers for - encryption - resumability - new replication cursor - last-received-hold - client subcommand `zrepl holds list`: list all holds and hold-like bookmarks that zrepl thinks belong to it - client subcommand `zrepl migrate replication-cursor:v1-v2`
2019-09-11 17:19:17 +02:00
log.WithError(err).Error("error telling sender that replication completed successfully")
return err
}
return err
}
func (s *Step) String() string {
if s.from == nil { // FIXME: ZFS semantics are that to is nil on non-incremental send
return fmt.Sprintf("%s%s (full)", s.parent.Path, s.to.RelName())
} else {
return fmt.Sprintf("%s(%s => %s)", s.parent.Path, s.from.RelName(), s.to.RelName())
}
}