images.go 28 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164
  1. package server
  2. import (
  3. "bytes"
  4. "cmp"
  5. "context"
  6. "crypto/sha256"
  7. "encoding/base64"
  8. "encoding/hex"
  9. "encoding/json"
  10. "errors"
  11. "fmt"
  12. "io"
  13. "log"
  14. "log/slog"
  15. "net/http"
  16. "net/url"
  17. "os"
  18. "path/filepath"
  19. "runtime"
  20. "slices"
  21. "strconv"
  22. "strings"
  23. "github.com/ollama/ollama/api"
  24. "github.com/ollama/ollama/auth"
  25. "github.com/ollama/ollama/envconfig"
  26. "github.com/ollama/ollama/format"
  27. "github.com/ollama/ollama/llm"
  28. "github.com/ollama/ollama/parser"
  29. "github.com/ollama/ollama/template"
  30. "github.com/ollama/ollama/types/errtypes"
  31. "github.com/ollama/ollama/types/model"
  32. "github.com/ollama/ollama/version"
  33. )
  34. var (
  35. errCapabilities = errors.New("does not support")
  36. errCapabilityCompletion = errors.New("completion")
  37. errCapabilityTools = errors.New("tools")
  38. errCapabilityInsert = errors.New("insert")
  39. )
  40. type Capability string
  41. const (
  42. CapabilityCompletion = Capability("completion")
  43. CapabilityTools = Capability("tools")
  44. CapabilityInsert = Capability("insert")
  45. )
  46. type registryOptions struct {
  47. Insecure bool
  48. Username string
  49. Password string
  50. Token string
  51. CheckRedirect func(req *http.Request, via []*http.Request) error
  52. }
  53. type Model struct {
  54. Name string `json:"name"`
  55. Config ConfigV2
  56. ShortName string
  57. ModelPath string
  58. ParentModel string
  59. AdapterPaths []string
  60. ProjectorPaths []string
  61. System string
  62. License []string
  63. Digest string
  64. Options map[string]interface{}
  65. Messages []api.Message
  66. Template *template.Template
  67. }
  68. // CheckCapabilities checks if the model has the specified capabilities returning an error describing
  69. // any missing or unknown capabilities
  70. func (m *Model) CheckCapabilities(caps ...Capability) error {
  71. var errs []error
  72. for _, cap := range caps {
  73. switch cap {
  74. case CapabilityCompletion:
  75. f, err := os.Open(m.ModelPath)
  76. if err != nil {
  77. slog.Error("couldn't open model file", "error", err)
  78. continue
  79. }
  80. defer f.Close()
  81. // TODO(mxyng): decode the GGML into model to avoid doing this multiple times
  82. ggml, _, err := llm.DecodeGGML(f, 0)
  83. if err != nil {
  84. slog.Error("couldn't decode ggml", "error", err)
  85. continue
  86. }
  87. if _, ok := ggml.KV()[fmt.Sprintf("%s.pooling_type", ggml.KV().Architecture())]; ok {
  88. errs = append(errs, errCapabilityCompletion)
  89. }
  90. case CapabilityTools:
  91. if !slices.Contains(m.Template.Vars(), "tools") {
  92. errs = append(errs, errCapabilityTools)
  93. }
  94. case CapabilityInsert:
  95. vars := m.Template.Vars()
  96. if !slices.Contains(vars, "suffix") {
  97. errs = append(errs, errCapabilityInsert)
  98. }
  99. default:
  100. slog.Error("unknown capability", "capability", cap)
  101. return fmt.Errorf("unknown capability: %s", cap)
  102. }
  103. }
  104. if err := errors.Join(errs...); err != nil {
  105. return fmt.Errorf("%w %w", errCapabilities, errors.Join(errs...))
  106. }
  107. return nil
  108. }
  109. func (m *Model) String() string {
  110. var modelfile parser.File
  111. modelfile.Commands = append(modelfile.Commands, parser.Command{
  112. Name: "model",
  113. Args: m.ModelPath,
  114. })
  115. for _, adapter := range m.AdapterPaths {
  116. modelfile.Commands = append(modelfile.Commands, parser.Command{
  117. Name: "adapter",
  118. Args: adapter,
  119. })
  120. }
  121. for _, projector := range m.ProjectorPaths {
  122. modelfile.Commands = append(modelfile.Commands, parser.Command{
  123. Name: "model",
  124. Args: projector,
  125. })
  126. }
  127. if m.Template != nil {
  128. modelfile.Commands = append(modelfile.Commands, parser.Command{
  129. Name: "template",
  130. Args: m.Template.String(),
  131. })
  132. }
  133. if m.System != "" {
  134. modelfile.Commands = append(modelfile.Commands, parser.Command{
  135. Name: "system",
  136. Args: m.System,
  137. })
  138. }
  139. for k, v := range m.Options {
  140. switch v := v.(type) {
  141. case []any:
  142. for _, s := range v {
  143. modelfile.Commands = append(modelfile.Commands, parser.Command{
  144. Name: k,
  145. Args: fmt.Sprintf("%v", s),
  146. })
  147. }
  148. default:
  149. modelfile.Commands = append(modelfile.Commands, parser.Command{
  150. Name: k,
  151. Args: fmt.Sprintf("%v", v),
  152. })
  153. }
  154. }
  155. for _, license := range m.License {
  156. modelfile.Commands = append(modelfile.Commands, parser.Command{
  157. Name: "license",
  158. Args: license,
  159. })
  160. }
  161. for _, msg := range m.Messages {
  162. modelfile.Commands = append(modelfile.Commands, parser.Command{
  163. Name: "message",
  164. Args: fmt.Sprintf("%s: %s", msg.Role, msg.Content),
  165. })
  166. }
  167. return modelfile.String()
  168. }
  169. type ConfigV2 struct {
  170. ModelFormat string `json:"model_format"`
  171. ModelFamily string `json:"model_family"`
  172. ModelFamilies []string `json:"model_families"`
  173. ModelType string `json:"model_type"`
  174. FileType string `json:"file_type"`
  175. // required by spec
  176. Architecture string `json:"architecture"`
  177. OS string `json:"os"`
  178. RootFS RootFS `json:"rootfs"`
  179. }
  180. type RootFS struct {
  181. Type string `json:"type"`
  182. DiffIDs []string `json:"diff_ids"`
  183. }
  184. func GetManifest(mp ModelPath) (*Manifest, string, error) {
  185. fp, err := mp.GetManifestPath()
  186. if err != nil {
  187. return nil, "", err
  188. }
  189. f, err := os.Open(fp)
  190. if err != nil {
  191. return nil, "", err
  192. }
  193. defer f.Close()
  194. sha256sum := sha256.New()
  195. var manifest Manifest
  196. if err := json.NewDecoder(io.TeeReader(f, sha256sum)).Decode(&manifest); err != nil {
  197. return nil, "", err
  198. }
  199. return &manifest, hex.EncodeToString(sha256sum.Sum(nil)), nil
  200. }
  201. func GetModel(name string) (*Model, error) {
  202. mp := ParseModelPath(name)
  203. manifest, digest, err := GetManifest(mp)
  204. if err != nil {
  205. return nil, err
  206. }
  207. model := &Model{
  208. Name: mp.GetFullTagname(),
  209. ShortName: mp.GetShortTagname(),
  210. Digest: digest,
  211. Template: template.DefaultTemplate,
  212. }
  213. if manifest.Config.Digest != "" {
  214. filename, err := GetBlobsPath(manifest.Config.Digest)
  215. if err != nil {
  216. return nil, err
  217. }
  218. configFile, err := os.Open(filename)
  219. if err != nil {
  220. return nil, err
  221. }
  222. defer configFile.Close()
  223. if err := json.NewDecoder(configFile).Decode(&model.Config); err != nil {
  224. return nil, err
  225. }
  226. }
  227. for _, layer := range manifest.Layers {
  228. filename, err := GetBlobsPath(layer.Digest)
  229. if err != nil {
  230. return nil, err
  231. }
  232. switch layer.MediaType {
  233. case "application/vnd.ollama.image.model":
  234. model.ModelPath = filename
  235. model.ParentModel = layer.From
  236. case "application/vnd.ollama.image.embed":
  237. // Deprecated in versions > 0.1.2
  238. // TODO: remove this warning in a future version
  239. slog.Info("WARNING: model contains embeddings, but embeddings in modelfiles have been deprecated and will be ignored.")
  240. case "application/vnd.ollama.image.adapter":
  241. model.AdapterPaths = append(model.AdapterPaths, filename)
  242. case "application/vnd.ollama.image.projector":
  243. model.ProjectorPaths = append(model.ProjectorPaths, filename)
  244. case "application/vnd.ollama.image.prompt",
  245. "application/vnd.ollama.image.template":
  246. bts, err := os.ReadFile(filename)
  247. if err != nil {
  248. return nil, err
  249. }
  250. model.Template, err = template.Parse(string(bts))
  251. if err != nil {
  252. return nil, err
  253. }
  254. case "application/vnd.ollama.image.system":
  255. bts, err := os.ReadFile(filename)
  256. if err != nil {
  257. return nil, err
  258. }
  259. model.System = string(bts)
  260. case "application/vnd.ollama.image.params":
  261. params, err := os.Open(filename)
  262. if err != nil {
  263. return nil, err
  264. }
  265. defer params.Close()
  266. // parse model options parameters into a map so that we can see which fields have been specified explicitly
  267. if err = json.NewDecoder(params).Decode(&model.Options); err != nil {
  268. return nil, err
  269. }
  270. case "application/vnd.ollama.image.messages":
  271. msgs, err := os.Open(filename)
  272. if err != nil {
  273. return nil, err
  274. }
  275. defer msgs.Close()
  276. if err = json.NewDecoder(msgs).Decode(&model.Messages); err != nil {
  277. return nil, err
  278. }
  279. case "application/vnd.ollama.image.license":
  280. bts, err := os.ReadFile(filename)
  281. if err != nil {
  282. return nil, err
  283. }
  284. model.License = append(model.License, string(bts))
  285. }
  286. }
  287. return model, nil
  288. }
  289. func realpath(rel, from string) string {
  290. abspath, err := filepath.Abs(from)
  291. if err != nil {
  292. return from
  293. }
  294. home, err := os.UserHomeDir()
  295. if err != nil {
  296. return abspath
  297. }
  298. if from == "~" {
  299. return home
  300. } else if strings.HasPrefix(from, "~/") {
  301. return filepath.Join(home, from[2:])
  302. }
  303. if _, err := os.Stat(filepath.Join(rel, from)); err == nil {
  304. // this is a file relative to the Modelfile
  305. return filepath.Join(rel, from)
  306. }
  307. return abspath
  308. }
  309. func CreateModel(ctx context.Context, name model.Name, modelFileDir, quantization string, modelfile *parser.File, fn func(resp api.ProgressResponse)) (err error) {
  310. config := ConfigV2{
  311. OS: "linux",
  312. Architecture: "amd64",
  313. RootFS: RootFS{
  314. Type: "layers",
  315. },
  316. }
  317. var messages []*api.Message
  318. parameters := make(map[string]any)
  319. var layers []Layer
  320. var baseLayers []*layerGGML
  321. for _, c := range modelfile.Commands {
  322. mediatype := fmt.Sprintf("application/vnd.ollama.image.%s", c.Name)
  323. command := c.Name
  324. switch command {
  325. case "model", "adapter":
  326. if name := model.ParseName(c.Args); name.IsValid() && command == "model" {
  327. baseLayers, err = parseFromModel(ctx, name, fn)
  328. if err != nil {
  329. return err
  330. }
  331. } else if strings.HasPrefix(c.Args, "@") {
  332. digest := strings.TrimPrefix(c.Args, "@")
  333. if ib, ok := intermediateBlobs[digest]; ok {
  334. p, err := GetBlobsPath(ib)
  335. if err != nil {
  336. return err
  337. }
  338. if _, err := os.Stat(p); errors.Is(err, os.ErrNotExist) {
  339. // pass
  340. } else if err != nil {
  341. return err
  342. } else {
  343. fn(api.ProgressResponse{Status: fmt.Sprintf("using cached layer %s", ib)})
  344. digest = ib
  345. }
  346. }
  347. blobpath, err := GetBlobsPath(digest)
  348. if err != nil {
  349. return err
  350. }
  351. blob, err := os.Open(blobpath)
  352. if err != nil {
  353. return err
  354. }
  355. defer blob.Close()
  356. baseLayers, err = parseFromFile(ctx, command, baseLayers, blob, digest, fn)
  357. if err != nil {
  358. return err
  359. }
  360. } else if file, err := os.Open(realpath(modelFileDir, c.Args)); err == nil {
  361. defer file.Close()
  362. baseLayers, err = parseFromFile(ctx, command, baseLayers, file, "", fn)
  363. if err != nil {
  364. return err
  365. }
  366. } else {
  367. return fmt.Errorf("invalid model reference: %s", c.Args)
  368. }
  369. for _, baseLayer := range baseLayers {
  370. if quantization != "" &&
  371. baseLayer.MediaType == "application/vnd.ollama.image.model" &&
  372. baseLayer.GGML != nil &&
  373. baseLayer.GGML.Name() == "gguf" {
  374. want, err := llm.ParseFileType(quantization)
  375. if err != nil {
  376. return err
  377. }
  378. ft := baseLayer.GGML.KV().FileType()
  379. if !slices.Contains([]string{"F16", "F32"}, ft.String()) {
  380. return errors.New("quantization is only supported for F16 and F32 models")
  381. } else if want != ft {
  382. fn(api.ProgressResponse{Status: fmt.Sprintf("quantizing %s model to %s", ft, quantization)})
  383. blob, err := GetBlobsPath(baseLayer.Digest)
  384. if err != nil {
  385. return err
  386. }
  387. temp, err := os.CreateTemp(filepath.Dir(blob), quantization)
  388. if err != nil {
  389. return err
  390. }
  391. defer temp.Close()
  392. defer os.Remove(temp.Name())
  393. if err := llm.Quantize(blob, temp.Name(), want); err != nil {
  394. return err
  395. }
  396. layer, err := NewLayer(temp, baseLayer.MediaType)
  397. if err != nil {
  398. return err
  399. }
  400. if _, err := temp.Seek(0, io.SeekStart); err != nil {
  401. return err
  402. }
  403. ggml, _, err := llm.DecodeGGML(temp, 0)
  404. if err != nil {
  405. return err
  406. }
  407. baseLayer.Layer = layer
  408. baseLayer.GGML = ggml
  409. }
  410. }
  411. if baseLayer.GGML != nil {
  412. config.ModelFormat = cmp.Or(config.ModelFormat, baseLayer.GGML.Name())
  413. config.ModelFamily = cmp.Or(config.ModelFamily, baseLayer.GGML.KV().Architecture())
  414. config.ModelType = cmp.Or(config.ModelType, format.HumanNumber(baseLayer.GGML.KV().ParameterCount()))
  415. config.FileType = cmp.Or(config.FileType, baseLayer.GGML.KV().FileType().String())
  416. config.ModelFamilies = append(config.ModelFamilies, baseLayer.GGML.KV().Architecture())
  417. }
  418. layers = append(layers, baseLayer.Layer)
  419. }
  420. case "license", "template", "system":
  421. if c.Name == "template" {
  422. if _, err := template.Parse(c.Args); err != nil {
  423. return fmt.Errorf("%w: %s", errBadTemplate, err)
  424. }
  425. }
  426. if c.Name != "license" {
  427. // replace
  428. layers = slices.DeleteFunc(layers, func(layer Layer) bool {
  429. if layer.MediaType != mediatype {
  430. return false
  431. }
  432. if err := layer.Remove(); err != nil {
  433. return false
  434. }
  435. return true
  436. })
  437. }
  438. blob := strings.NewReader(c.Args)
  439. layer, err := NewLayer(blob, mediatype)
  440. if err != nil {
  441. return err
  442. }
  443. layers = append(layers, layer)
  444. case "message":
  445. role, content, ok := strings.Cut(c.Args, ": ")
  446. if !ok {
  447. return fmt.Errorf("invalid message: %s", c.Args)
  448. }
  449. messages = append(messages, &api.Message{Role: role, Content: content})
  450. default:
  451. ps, err := api.FormatParams(map[string][]string{c.Name: {c.Args}})
  452. if err != nil {
  453. return err
  454. }
  455. for k, v := range ps {
  456. if ks, ok := parameters[k].([]string); ok {
  457. parameters[k] = append(ks, v.([]string)...)
  458. } else if vs, ok := v.([]string); ok {
  459. parameters[k] = vs
  460. } else {
  461. parameters[k] = v
  462. }
  463. }
  464. }
  465. }
  466. var err2 error
  467. layers = slices.DeleteFunc(layers, func(layer Layer) bool {
  468. switch layer.MediaType {
  469. case "application/vnd.ollama.image.message":
  470. // if there are new messages, remove the inherited ones
  471. if len(messages) > 0 {
  472. return true
  473. }
  474. return false
  475. case "application/vnd.ollama.image.params":
  476. // merge inherited parameters with new ones
  477. r, err := layer.Open()
  478. if err != nil {
  479. err2 = err
  480. return false
  481. }
  482. defer r.Close()
  483. var ps map[string]any
  484. if err := json.NewDecoder(r).Decode(&ps); err != nil {
  485. err2 = err
  486. return false
  487. }
  488. for k, v := range ps {
  489. if _, ok := parameters[k]; !ok {
  490. parameters[k] = v
  491. }
  492. }
  493. return true
  494. default:
  495. return false
  496. }
  497. })
  498. if err2 != nil {
  499. return err2
  500. }
  501. if len(messages) > 0 {
  502. var b bytes.Buffer
  503. if err := json.NewEncoder(&b).Encode(messages); err != nil {
  504. return err
  505. }
  506. layer, err := NewLayer(&b, "application/vnd.ollama.image.messages")
  507. if err != nil {
  508. return err
  509. }
  510. layers = append(layers, layer)
  511. }
  512. if len(parameters) > 0 {
  513. var b bytes.Buffer
  514. if err := json.NewEncoder(&b).Encode(parameters); err != nil {
  515. return err
  516. }
  517. layer, err := NewLayer(&b, "application/vnd.ollama.image.params")
  518. if err != nil {
  519. return err
  520. }
  521. layers = append(layers, layer)
  522. }
  523. digests := make([]string, len(layers))
  524. for i, layer := range layers {
  525. digests[i] = layer.Digest
  526. }
  527. config.RootFS.DiffIDs = digests
  528. var b bytes.Buffer
  529. if err := json.NewEncoder(&b).Encode(config); err != nil {
  530. return err
  531. }
  532. configLayer, err := NewLayer(&b, "application/vnd.docker.container.image.v1+json")
  533. if err != nil {
  534. return err
  535. }
  536. for _, layer := range append(layers, configLayer) {
  537. if layer.status != "" {
  538. fn(api.ProgressResponse{Status: layer.status})
  539. }
  540. }
  541. old, _ := ParseNamedManifest(name)
  542. fn(api.ProgressResponse{Status: "writing manifest"})
  543. if err := WriteManifest(name, configLayer, layers); err != nil {
  544. return err
  545. }
  546. if !envconfig.NoPrune() && old != nil {
  547. if err := old.RemoveLayers(); err != nil {
  548. return err
  549. }
  550. }
  551. fn(api.ProgressResponse{Status: "success"})
  552. return nil
  553. }
  554. func CopyModel(src, dst model.Name) error {
  555. if !dst.IsFullyQualified() {
  556. return model.Unqualified(dst)
  557. }
  558. if !src.IsFullyQualified() {
  559. return model.Unqualified(src)
  560. }
  561. if src.Filepath() == dst.Filepath() {
  562. return nil
  563. }
  564. manifests, err := GetManifestPath()
  565. if err != nil {
  566. return err
  567. }
  568. dstpath := filepath.Join(manifests, dst.Filepath())
  569. if err := os.MkdirAll(filepath.Dir(dstpath), 0o755); err != nil {
  570. return err
  571. }
  572. srcpath := filepath.Join(manifests, src.Filepath())
  573. srcfile, err := os.Open(srcpath)
  574. if err != nil {
  575. return err
  576. }
  577. defer srcfile.Close()
  578. dstfile, err := os.Create(dstpath)
  579. if err != nil {
  580. return err
  581. }
  582. defer dstfile.Close()
  583. _, err = io.Copy(dstfile, srcfile)
  584. return err
  585. }
  586. func deleteUnusedLayers(deleteMap map[string]struct{}) error {
  587. manifests, err := Manifests()
  588. if err != nil {
  589. return err
  590. }
  591. for _, manifest := range manifests {
  592. for _, layer := range manifest.Layers {
  593. delete(deleteMap, layer.Digest)
  594. }
  595. delete(deleteMap, manifest.Config.Digest)
  596. }
  597. // only delete the files which are still in the deleteMap
  598. for k := range deleteMap {
  599. fp, err := GetBlobsPath(k)
  600. if err != nil {
  601. slog.Info(fmt.Sprintf("couldn't get file path for '%s': %v", k, err))
  602. continue
  603. }
  604. if err := os.Remove(fp); err != nil {
  605. slog.Info(fmt.Sprintf("couldn't remove file '%s': %v", fp, err))
  606. continue
  607. }
  608. }
  609. return nil
  610. }
  611. func PruneLayers() error {
  612. deleteMap := make(map[string]struct{})
  613. p, err := GetBlobsPath("")
  614. if err != nil {
  615. return err
  616. }
  617. blobs, err := os.ReadDir(p)
  618. if err != nil {
  619. slog.Info(fmt.Sprintf("couldn't read dir '%s': %v", p, err))
  620. return err
  621. }
  622. for _, blob := range blobs {
  623. name := blob.Name()
  624. name = strings.ReplaceAll(name, "-", ":")
  625. _, err := GetBlobsPath(name)
  626. if err != nil {
  627. if errors.Is(err, ErrInvalidDigestFormat) {
  628. // remove invalid blobs (e.g. partial downloads)
  629. if err := os.Remove(filepath.Join(p, blob.Name())); err != nil {
  630. slog.Error("couldn't remove blob", "blob", blob.Name(), "error", err)
  631. }
  632. }
  633. continue
  634. }
  635. deleteMap[name] = struct{}{}
  636. }
  637. slog.Info(fmt.Sprintf("total blobs: %d", len(deleteMap)))
  638. if err := deleteUnusedLayers(deleteMap); err != nil {
  639. slog.Error(fmt.Sprintf("couldn't remove unused layers: %v", err))
  640. return nil
  641. }
  642. slog.Info(fmt.Sprintf("total unused blobs removed: %d", len(deleteMap)))
  643. return nil
  644. }
  645. func PruneDirectory(path string) error {
  646. info, err := os.Lstat(path)
  647. if err != nil {
  648. return err
  649. }
  650. if info.IsDir() && info.Mode()&os.ModeSymlink == 0 {
  651. entries, err := os.ReadDir(path)
  652. if err != nil {
  653. return err
  654. }
  655. for _, entry := range entries {
  656. if err := PruneDirectory(filepath.Join(path, entry.Name())); err != nil {
  657. return err
  658. }
  659. }
  660. entries, err = os.ReadDir(path)
  661. if err != nil {
  662. return err
  663. }
  664. if len(entries) > 0 {
  665. return nil
  666. }
  667. return os.Remove(path)
  668. }
  669. return nil
  670. }
  671. func PushModel(ctx context.Context, name string, regOpts *registryOptions, fn func(api.ProgressResponse)) error {
  672. mp := ParseModelPath(name)
  673. fn(api.ProgressResponse{Status: "retrieving manifest"})
  674. if mp.ProtocolScheme == "http" && !regOpts.Insecure {
  675. return errors.New("insecure protocol http")
  676. }
  677. manifest, _, err := GetManifest(mp)
  678. if err != nil {
  679. fn(api.ProgressResponse{Status: "couldn't retrieve manifest"})
  680. return err
  681. }
  682. var layers []Layer
  683. layers = append(layers, manifest.Layers...)
  684. if manifest.Config.Digest != "" {
  685. layers = append(layers, manifest.Config)
  686. }
  687. for _, layer := range layers {
  688. if err := uploadBlob(ctx, mp, layer, regOpts, fn); err != nil {
  689. slog.Info(fmt.Sprintf("error uploading blob: %v", err))
  690. return err
  691. }
  692. }
  693. fn(api.ProgressResponse{Status: "pushing manifest"})
  694. requestURL := mp.BaseURL()
  695. requestURL = requestURL.JoinPath("v2", mp.GetNamespaceRepository(), "manifests", mp.Tag)
  696. manifestJSON, err := json.Marshal(manifest)
  697. if err != nil {
  698. return err
  699. }
  700. headers := make(http.Header)
  701. headers.Set("Content-Type", "application/vnd.docker.distribution.manifest.v2+json")
  702. resp, err := makeRequestWithRetry(ctx, http.MethodPut, requestURL, headers, bytes.NewReader(manifestJSON), regOpts)
  703. if err != nil {
  704. return err
  705. }
  706. defer resp.Body.Close()
  707. fn(api.ProgressResponse{Status: "success"})
  708. return nil
  709. }
  710. func PullModel(ctx context.Context, name string, regOpts *registryOptions, fn func(api.ProgressResponse)) error {
  711. mp := ParseModelPath(name)
  712. // build deleteMap to prune unused layers
  713. deleteMap := make(map[string]struct{})
  714. manifest, _, err := GetManifest(mp)
  715. if errors.Is(err, os.ErrNotExist) {
  716. // noop
  717. } else if err != nil && !errors.Is(err, os.ErrNotExist) {
  718. return err
  719. } else {
  720. for _, l := range manifest.Layers {
  721. deleteMap[l.Digest] = struct{}{}
  722. }
  723. if manifest.Config.Digest != "" {
  724. deleteMap[manifest.Config.Digest] = struct{}{}
  725. }
  726. }
  727. if mp.ProtocolScheme == "http" && !regOpts.Insecure {
  728. return errors.New("insecure protocol http")
  729. }
  730. fn(api.ProgressResponse{Status: "pulling manifest"})
  731. manifest, err = pullModelManifest(ctx, mp, regOpts)
  732. if err != nil {
  733. return fmt.Errorf("pull model manifest: %s", err)
  734. }
  735. var layers []Layer
  736. layers = append(layers, manifest.Layers...)
  737. if manifest.Config.Digest != "" {
  738. layers = append(layers, manifest.Config)
  739. }
  740. skipVerify := make(map[string]bool)
  741. for _, layer := range layers {
  742. cacheHit, err := downloadBlob(ctx, downloadOpts{
  743. mp: mp,
  744. digest: layer.Digest,
  745. regOpts: regOpts,
  746. fn: fn,
  747. })
  748. if err != nil {
  749. return err
  750. }
  751. skipVerify[layer.Digest] = cacheHit
  752. delete(deleteMap, layer.Digest)
  753. }
  754. delete(deleteMap, manifest.Config.Digest)
  755. fn(api.ProgressResponse{Status: "verifying sha256 digest"})
  756. for _, layer := range layers {
  757. if skipVerify[layer.Digest] {
  758. continue
  759. }
  760. if err := verifyBlob(layer.Digest); err != nil {
  761. if errors.Is(err, errDigestMismatch) {
  762. // something went wrong, delete the blob
  763. fp, err := GetBlobsPath(layer.Digest)
  764. if err != nil {
  765. return err
  766. }
  767. if err := os.Remove(fp); err != nil {
  768. // log this, but return the original error
  769. slog.Info(fmt.Sprintf("couldn't remove file with digest mismatch '%s': %v", fp, err))
  770. }
  771. }
  772. return err
  773. }
  774. }
  775. fn(api.ProgressResponse{Status: "writing manifest"})
  776. manifestJSON, err := json.Marshal(manifest)
  777. if err != nil {
  778. return err
  779. }
  780. fp, err := mp.GetManifestPath()
  781. if err != nil {
  782. return err
  783. }
  784. if err := os.MkdirAll(filepath.Dir(fp), 0o755); err != nil {
  785. return err
  786. }
  787. err = os.WriteFile(fp, manifestJSON, 0o644)
  788. if err != nil {
  789. slog.Info(fmt.Sprintf("couldn't write to %s", fp))
  790. return err
  791. }
  792. if !envconfig.NoPrune() && len(deleteMap) > 0 {
  793. fn(api.ProgressResponse{Status: "removing unused layers"})
  794. if err := deleteUnusedLayers(deleteMap); err != nil {
  795. fn(api.ProgressResponse{Status: fmt.Sprintf("couldn't remove unused layers: %v", err)})
  796. }
  797. }
  798. fn(api.ProgressResponse{Status: "success"})
  799. return nil
  800. }
  801. func pullModelManifest(ctx context.Context, mp ModelPath, regOpts *registryOptions) (*Manifest, error) {
  802. requestURL := mp.BaseURL().JoinPath("v2", mp.GetNamespaceRepository(), "manifests", mp.Tag)
  803. headers := make(http.Header)
  804. headers.Set("Accept", "application/vnd.docker.distribution.manifest.v2+json")
  805. resp, err := makeRequestWithRetry(ctx, http.MethodGet, requestURL, headers, nil, regOpts)
  806. if err != nil {
  807. return nil, err
  808. }
  809. defer resp.Body.Close()
  810. var m Manifest
  811. if err := json.NewDecoder(resp.Body).Decode(&m); err != nil {
  812. return nil, err
  813. }
  814. return &m, err
  815. }
  816. // GetSHA256Digest returns the SHA256 hash of a given buffer and returns it, and the size of buffer
  817. func GetSHA256Digest(r io.Reader) (string, int64) {
  818. h := sha256.New()
  819. n, err := io.Copy(h, r)
  820. if err != nil {
  821. log.Fatal(err)
  822. }
  823. return fmt.Sprintf("sha256:%x", h.Sum(nil)), n
  824. }
  825. var errUnauthorized = errors.New("unauthorized: access denied")
  826. // getTokenSubject returns the subject of a JWT token, it does not validate the token
  827. func getTokenSubject(token string) string {
  828. parts := strings.Split(token, ".")
  829. if len(parts) != 3 {
  830. return ""
  831. }
  832. payload := parts[1]
  833. payloadBytes, err := base64.RawURLEncoding.DecodeString(payload)
  834. if err != nil {
  835. slog.Error(fmt.Sprintf("failed to decode jwt payload: %v", err))
  836. return ""
  837. }
  838. var payloadMap map[string]interface{}
  839. if err := json.Unmarshal(payloadBytes, &payloadMap); err != nil {
  840. slog.Error(fmt.Sprintf("failed to unmarshal payload JSON: %v", err))
  841. return ""
  842. }
  843. sub, ok := payloadMap["sub"]
  844. if !ok {
  845. slog.Error("jwt does not contain 'sub' field")
  846. return ""
  847. }
  848. return fmt.Sprintf("%s", sub)
  849. }
  850. func makeRequestWithRetry(ctx context.Context, method string, requestURL *url.URL, headers http.Header, body io.ReadSeeker, regOpts *registryOptions) (*http.Response, error) {
  851. anonymous := true // access will default to anonymous if no user is found associated with the public key
  852. for range 2 {
  853. resp, err := makeRequest(ctx, method, requestURL, headers, body, regOpts)
  854. if err != nil {
  855. if !errors.Is(err, context.Canceled) {
  856. slog.Info(fmt.Sprintf("request failed: %v", err))
  857. }
  858. return nil, err
  859. }
  860. switch {
  861. case resp.StatusCode == http.StatusUnauthorized:
  862. // Handle authentication error with one retry
  863. challenge := parseRegistryChallenge(resp.Header.Get("www-authenticate"))
  864. token, err := getAuthorizationToken(ctx, challenge)
  865. if err != nil {
  866. return nil, err
  867. }
  868. anonymous = getTokenSubject(token) == "anonymous"
  869. regOpts.Token = token
  870. if body != nil {
  871. _, err = body.Seek(0, io.SeekStart)
  872. if err != nil {
  873. return nil, err
  874. }
  875. }
  876. case resp.StatusCode == http.StatusNotFound:
  877. return nil, os.ErrNotExist
  878. case resp.StatusCode >= http.StatusBadRequest:
  879. responseBody, err := io.ReadAll(resp.Body)
  880. if err != nil {
  881. return nil, fmt.Errorf("%d: %s", resp.StatusCode, err)
  882. }
  883. return nil, fmt.Errorf("%d: %s", resp.StatusCode, responseBody)
  884. default:
  885. return resp, nil
  886. }
  887. }
  888. if anonymous {
  889. // no user is associated with the public key, and the request requires non-anonymous access
  890. pubKey, nestedErr := auth.GetPublicKey()
  891. if nestedErr != nil {
  892. slog.Error(fmt.Sprintf("couldn't get public key: %v", nestedErr))
  893. return nil, errUnauthorized
  894. }
  895. return nil, &errtypes.UnknownOllamaKey{Key: pubKey}
  896. }
  897. // user is associated with the public key, but is not authorized to make the request
  898. return nil, errUnauthorized
  899. }
  900. func makeRequest(ctx context.Context, method string, requestURL *url.URL, headers http.Header, body io.Reader, regOpts *registryOptions) (*http.Response, error) {
  901. if requestURL.Scheme != "http" && regOpts != nil && regOpts.Insecure {
  902. requestURL.Scheme = "http"
  903. }
  904. req, err := http.NewRequestWithContext(ctx, method, requestURL.String(), body)
  905. if err != nil {
  906. return nil, err
  907. }
  908. if headers != nil {
  909. req.Header = headers
  910. }
  911. if regOpts != nil {
  912. if regOpts.Token != "" {
  913. req.Header.Set("Authorization", "Bearer "+regOpts.Token)
  914. } else if regOpts.Username != "" && regOpts.Password != "" {
  915. req.SetBasicAuth(regOpts.Username, regOpts.Password)
  916. }
  917. }
  918. req.Header.Set("User-Agent", fmt.Sprintf("ollama/%s (%s %s) Go/%s", version.Version, runtime.GOARCH, runtime.GOOS, runtime.Version()))
  919. if s := req.Header.Get("Content-Length"); s != "" {
  920. contentLength, err := strconv.ParseInt(s, 10, 64)
  921. if err != nil {
  922. return nil, err
  923. }
  924. req.ContentLength = contentLength
  925. }
  926. resp, err := (&http.Client{
  927. CheckRedirect: regOpts.CheckRedirect,
  928. }).Do(req)
  929. if err != nil {
  930. return nil, err
  931. }
  932. return resp, nil
  933. }
  934. func getValue(header, key string) string {
  935. startIdx := strings.Index(header, key+"=")
  936. if startIdx == -1 {
  937. return ""
  938. }
  939. // Move the index to the starting quote after the key.
  940. startIdx += len(key) + 2
  941. endIdx := startIdx
  942. for endIdx < len(header) {
  943. if header[endIdx] == '"' {
  944. if endIdx+1 < len(header) && header[endIdx+1] != ',' { // If the next character isn't a comma, continue
  945. endIdx++
  946. continue
  947. }
  948. break
  949. }
  950. endIdx++
  951. }
  952. return header[startIdx:endIdx]
  953. }
  954. func parseRegistryChallenge(authStr string) registryChallenge {
  955. authStr = strings.TrimPrefix(authStr, "Bearer ")
  956. return registryChallenge{
  957. Realm: getValue(authStr, "realm"),
  958. Service: getValue(authStr, "service"),
  959. Scope: getValue(authStr, "scope"),
  960. }
  961. }
  962. var errDigestMismatch = errors.New("digest mismatch, file must be downloaded again")
  963. func verifyBlob(digest string) error {
  964. fp, err := GetBlobsPath(digest)
  965. if err != nil {
  966. return err
  967. }
  968. f, err := os.Open(fp)
  969. if err != nil {
  970. return err
  971. }
  972. defer f.Close()
  973. fileDigest, _ := GetSHA256Digest(f)
  974. if digest != fileDigest {
  975. return fmt.Errorf("%w: want %s, got %s", errDigestMismatch, digest, fileDigest)
  976. }
  977. return nil
  978. }