reserveVM's duplicate-name guard routed through Daemon.FindVM, which falls back to prefix-matching on both ids and names when no exact match is found. That turns the uniqueness check into a correctness bug: a brand-new VM name can be rejected because it happens to prefix an existing VM's id, or an existing VM's name. So `vm create --name beta` fails when `beta-sandbox` already exists. Swap in a dedicated store.GetVMByName that does a literal `WHERE name = ?` lookup, and use it from reserveVM. FindVM keeps its prefix-matching behaviour for user-facing lookup paths (`vm ssh <partial>`, `vm stop <partial>`) where "did you mean" semantics are the feature. Tests: - TestReserveVMAllowsNameThatPrefixesExistingVM — seeds a VM whose id + name both start with "longname", then reserves two new VMs named "longname" and "longname-sandbox". Both must succeed. Under the old FindVM-based check, both would fail. - TestReserveVMRejectsExactDuplicateName — actual collisions are still rejected after the swap. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
196 lines
6.2 KiB
Go
196 lines
6.2 KiB
Go
package daemon
|
|
|
|
import (
|
|
"context"
|
|
"database/sql"
|
|
"errors"
|
|
"fmt"
|
|
"os"
|
|
"path/filepath"
|
|
"strings"
|
|
|
|
"banger/internal/api"
|
|
"banger/internal/imagecat"
|
|
"banger/internal/model"
|
|
"banger/internal/vmdns"
|
|
)
|
|
|
|
// CreateVM is split into three phases so the global createVMMu guards
|
|
// only the narrow name+IP reservation window, not the slow image
|
|
// resolution or the multi-second boot flow:
|
|
//
|
|
// 1. Validate + resolve image. No global lock. Image auto-pull
|
|
// self-locks via imageOpsMu (which is also now publication-only).
|
|
// 2. Reserve a row: generate id, pick next IP, claim the name,
|
|
// UpsertVM the "created" record. Held under createVMMu so two
|
|
// concurrent `vm create --name foo` calls can't both think they
|
|
// won.
|
|
// 3. Boot. Only the per-VM lock is held — parallel creates against
|
|
// different VMs fully overlap.
|
|
func (d *Daemon) CreateVM(ctx context.Context, params api.VMCreateParams) (vm model.VMRecord, err error) {
|
|
op := d.beginOperation("vm.create")
|
|
defer func() {
|
|
if err != nil {
|
|
op.fail(err)
|
|
return
|
|
}
|
|
op.done(vmLogAttrs(vm)...)
|
|
}()
|
|
if err := validateOptionalPositiveSetting("vcpu", params.VCPUCount); err != nil {
|
|
return model.VMRecord{}, err
|
|
}
|
|
if err := validateOptionalPositiveSetting("memory", params.MemoryMiB); err != nil {
|
|
return model.VMRecord{}, err
|
|
}
|
|
|
|
imageName := params.ImageName
|
|
if imageName == "" {
|
|
imageName = d.config.DefaultImageName
|
|
}
|
|
vmCreateStage(ctx, "resolve_image", "resolving image")
|
|
image, err := d.findOrAutoPullImage(ctx, imageName)
|
|
if err != nil {
|
|
return model.VMRecord{}, err
|
|
}
|
|
vmCreateStage(ctx, "resolve_image", "using image "+image.Name)
|
|
op.stage("image_resolved", imageLogAttrs(image)...)
|
|
|
|
systemOverlaySize := int64(model.DefaultSystemOverlaySize)
|
|
if params.SystemOverlaySize != "" {
|
|
systemOverlaySize, err = model.ParseSize(params.SystemOverlaySize)
|
|
if err != nil {
|
|
return model.VMRecord{}, err
|
|
}
|
|
}
|
|
workDiskSize := int64(model.DefaultWorkDiskSize)
|
|
if params.WorkDiskSize != "" {
|
|
workDiskSize, err = model.ParseSize(params.WorkDiskSize)
|
|
if err != nil {
|
|
return model.VMRecord{}, err
|
|
}
|
|
}
|
|
spec := model.VMSpec{
|
|
VCPUCount: optionalIntOrDefault(params.VCPUCount, model.DefaultVCPUCount),
|
|
MemoryMiB: optionalIntOrDefault(params.MemoryMiB, model.DefaultMemoryMiB),
|
|
SystemOverlaySizeByte: systemOverlaySize,
|
|
WorkDiskSizeBytes: workDiskSize,
|
|
NATEnabled: params.NATEnabled,
|
|
}
|
|
|
|
vm, err = d.reserveVM(ctx, strings.TrimSpace(params.Name), image, spec)
|
|
if err != nil {
|
|
return model.VMRecord{}, err
|
|
}
|
|
op.stage("persisted", vmLogAttrs(vm)...)
|
|
vmCreateBindVM(ctx, vm)
|
|
vmCreateStage(ctx, "reserve_vm", fmt.Sprintf("allocated %s (%s)", vm.Name, vm.Runtime.GuestIP))
|
|
|
|
unlockVM := d.lockVMID(vm.ID)
|
|
defer unlockVM()
|
|
|
|
if params.NoStart {
|
|
vm.State = model.VMStateStopped
|
|
vm.Runtime.State = model.VMStateStopped
|
|
if err := d.store.UpsertVM(ctx, vm); err != nil {
|
|
return model.VMRecord{}, err
|
|
}
|
|
return vm, nil
|
|
}
|
|
return d.startVMLocked(ctx, vm, image)
|
|
}
|
|
|
|
// reserveVM holds createVMMu only long enough to verify the name is
|
|
// free, allocate a guest IP from the store, and persist the "created"
|
|
// reservation row. Everything else (image resolution upstream, boot
|
|
// downstream) runs outside this lock.
|
|
func (d *Daemon) reserveVM(ctx context.Context, requestedName string, image model.Image, spec model.VMSpec) (model.VMRecord, error) {
|
|
d.createVMMu.Lock()
|
|
defer d.createVMMu.Unlock()
|
|
|
|
name := requestedName
|
|
if name == "" {
|
|
generated, err := d.generateName(ctx)
|
|
if err != nil {
|
|
return model.VMRecord{}, err
|
|
}
|
|
name = generated
|
|
}
|
|
// Exact-name lookup. Using FindVM here would also match a new name
|
|
// that merely prefixes some existing VM's id or another VM's name,
|
|
// falsely rejecting perfectly valid names.
|
|
if _, err := d.store.GetVMByName(ctx, name); err == nil {
|
|
return model.VMRecord{}, fmt.Errorf("vm name already exists: %s", name)
|
|
} else if !errors.Is(err, sql.ErrNoRows) {
|
|
return model.VMRecord{}, err
|
|
}
|
|
|
|
id, err := model.NewID()
|
|
if err != nil {
|
|
return model.VMRecord{}, err
|
|
}
|
|
guestIP, err := d.store.NextGuestIP(ctx, bridgePrefix(d.config.BridgeIP))
|
|
if err != nil {
|
|
return model.VMRecord{}, err
|
|
}
|
|
vmDir := filepath.Join(d.layout.VMsDir, id)
|
|
if err := os.MkdirAll(vmDir, 0o755); err != nil {
|
|
return model.VMRecord{}, err
|
|
}
|
|
vsockCID, err := defaultVSockCID(guestIP)
|
|
if err != nil {
|
|
return model.VMRecord{}, err
|
|
}
|
|
now := model.Now()
|
|
vm := model.VMRecord{
|
|
ID: id,
|
|
Name: name,
|
|
ImageID: image.ID,
|
|
State: model.VMStateCreated,
|
|
CreatedAt: now,
|
|
UpdatedAt: now,
|
|
LastTouchedAt: now,
|
|
Spec: spec,
|
|
Runtime: model.VMRuntime{
|
|
State: model.VMStateCreated,
|
|
GuestIP: guestIP,
|
|
DNSName: vmdns.RecordName(name),
|
|
VMDir: vmDir,
|
|
VSockPath: defaultVSockPath(d.layout.RuntimeDir, id),
|
|
VSockCID: vsockCID,
|
|
SystemOverlay: filepath.Join(vmDir, "system.cow"),
|
|
WorkDiskPath: filepath.Join(vmDir, "root.ext4"),
|
|
LogPath: filepath.Join(vmDir, "firecracker.log"),
|
|
MetricsPath: filepath.Join(vmDir, "metrics.json"),
|
|
},
|
|
}
|
|
if err := d.store.UpsertVM(ctx, vm); err != nil {
|
|
return model.VMRecord{}, err
|
|
}
|
|
return vm, nil
|
|
}
|
|
|
|
// findOrAutoPullImage tries the local image store first; if the name
|
|
// isn't registered but matches an entry in the embedded imagecat
|
|
// catalog, it auto-pulls the bundle so `vm create --image foo` (and
|
|
// therefore `vm run`) works on a fresh host without the user having
|
|
// to run `image pull` first.
|
|
func (d *Daemon) findOrAutoPullImage(ctx context.Context, idOrName string) (model.Image, error) {
|
|
image, err := d.FindImage(ctx, idOrName)
|
|
if err == nil {
|
|
return image, nil
|
|
}
|
|
catalog, loadErr := imagecat.LoadEmbedded()
|
|
if loadErr != nil {
|
|
return model.Image{}, err
|
|
}
|
|
entry, lookupErr := catalog.Lookup(idOrName)
|
|
if lookupErr != nil {
|
|
// Not in the catalog either — surface the original not-found.
|
|
return model.Image{}, err
|
|
}
|
|
vmCreateStage(ctx, "auto_pull_image", fmt.Sprintf("pulling %s from image catalog", entry.Name))
|
|
if _, pullErr := d.PullImage(ctx, api.ImagePullParams{Ref: entry.Name}); pullErr != nil {
|
|
return model.Image{}, fmt.Errorf("auto-pull image %q: %w", entry.Name, pullErr)
|
|
}
|
|
return d.FindImage(ctx, idOrName)
|
|
}
|