blob: 89be0f5f4e3229c7bbf2772d9bc367ab3764d857 [file] [log] [blame]
Jeff Gastonf1fd45e2017-08-09 18:25:28 -07001// Copyright 2017 Google Inc. All rights reserved.
2//
3// Licensed under the Apache License, Version 2.0 (the "License");
4// you may not use this file except in compliance with the License.
5// You may obtain a copy of the License at
6//
7// http://www.apache.org/licenses/LICENSE-2.0
8//
9// Unless required by applicable law or agreed to in writing, software
10// distributed under the License is distributed on an "AS IS" BASIS,
11// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12// See the License for the specific language governing permissions and
13// limitations under the License.
14
15package finder
16
17import (
18 "bufio"
19 "bytes"
20 "encoding/json"
Jeff Gastond3119522017-08-22 14:11:15 -070021 "errors"
Jeff Gastonf1fd45e2017-08-09 18:25:28 -070022 "fmt"
23 "io"
24 "os"
25 "path/filepath"
26 "runtime"
27 "sort"
28 "strings"
29 "sync"
30 "sync/atomic"
31 "time"
32
Colin Cross8d6395c2017-12-21 15:46:01 -080033 "android/soong/finder/fs"
Jeff Gastonf1fd45e2017-08-09 18:25:28 -070034)
35
36// This file provides a Finder struct that can quickly search for files satisfying
37// certain criteria.
38// This Finder gets its speed partially from parallelism and partially from caching.
39// If a Stat call returns the same result as last time, then it means Finder
40// can skip the ReadDir call for that dir.
41
42// The primary data structure used by the finder is the field Finder.nodes ,
43// which is a tree of nodes of type *pathMap .
44// Each node represents a directory on disk, along with its stats, subdirectories,
45// and contained files.
46
47// The common use case for the Finder is that the caller creates a Finder and gives
48// it the same query that was given to it in the previous execution.
49// In this situation, the major events that take place are:
50// 1. The Finder begins to load its db
51// 2. The Finder begins to stat the directories mentioned in its db (using multiple threads)
52// Calling Stat on each of these directories is generally a large fraction of the total time
53// 3. The Finder begins to construct a separate tree of nodes in each of its threads
54// 4. The Finder merges the individual node trees into the main node tree
55// 5. The Finder may call ReadDir a few times if there are a few directories that are out-of-date
56// These ReadDir calls might prompt additional Stat calls, etc
57// 6. The Finder waits for all loading to complete
58// 7. The Finder searches the cache for files matching the user's query (using multiple threads)
59
60// These are the invariants regarding concurrency:
61// 1. The public methods of Finder are threadsafe.
62// The public methods are only performance-optimized for one caller at a time, however.
63// For the moment, multiple concurrent callers shouldn't expect any better performance than
64// multiple serial callers.
65// 2. While building the node tree, only one thread may ever access the <children> collection of a
66// *pathMap at once.
67// a) The thread that accesses the <children> collection is the thread that discovers the
68// children (by reading them from the cache or by having received a response to ReadDir).
69// 1) Consequently, the thread that discovers the children also spawns requests to stat
70// subdirs.
71// b) Consequently, while building the node tree, no thread may do a lookup of its
72// *pathMap via filepath because another thread may be adding children to the
73// <children> collection of an ancestor node. Additionally, in rare cases, another thread
74// may be removing children from an ancestor node if the children were only discovered to
75// be irrelevant after calling ReadDir (which happens if a prune-file was just added).
76// 3. No query will begin to be serviced until all loading (both reading the db
77// and scanning the filesystem) is complete.
78// Tests indicate that it only takes about 10% as long to search the in-memory cache as to
79// generate it, making this not a huge loss in performance.
80// 4. The parsing of the db and the initial setup of the pathMap tree must complete before
81// beginning to call listDirSync (because listDirSync can create new entries in the pathMap)
82
83// see cmd/finder.go or finder_test.go for usage examples
84
85// Update versionString whenever making a backwards-incompatible change to the cache file format
86const versionString = "Android finder version 1"
87
88// a CacheParams specifies which files and directories the user wishes be scanned and
89// potentially added to the cache
90type CacheParams struct {
91 // WorkingDirectory is used as a base for any relative file paths given to the Finder
92 WorkingDirectory string
93
94 // RootDirs are the root directories used to initiate the search
95 RootDirs []string
96
97 // ExcludeDirs are directory names that if encountered are removed from the search
98 ExcludeDirs []string
99
100 // PruneFiles are file names that if encountered prune their entire directory
101 // (including siblings)
102 PruneFiles []string
103
104 // IncludeFiles are file names to include as matches
105 IncludeFiles []string
106}
107
108// a cacheConfig stores the inputs that determine what should be included in the cache
109type cacheConfig struct {
110 CacheParams
111
112 // FilesystemView is a unique identifier telling which parts of which file systems
113 // are readable by the Finder. In practice its value is essentially username@hostname.
114 // FilesystemView is set to ensure that a cache file copied to another host or
115 // found by another user doesn't inadvertently get reused.
116 FilesystemView string
117}
118
119func (p *cacheConfig) Dump() ([]byte, error) {
120 bytes, err := json.Marshal(p)
121 return bytes, err
122}
123
124// a cacheMetadata stores version information about the cache
125type cacheMetadata struct {
126 // The Version enables the Finder to determine whether it can even parse the file
127 // If the version changes, the entire cache file must be regenerated
128 Version string
129
130 // The CacheParams enables the Finder to determine whether the parameters match
131 // If the CacheParams change, the Finder can choose how much of the cache file to reuse
132 // (although in practice, the Finder will probably choose to ignore the entire file anyway)
133 Config cacheConfig
134}
135
136type Logger interface {
137 Output(calldepth int, s string) error
138}
139
140// the Finder is the main struct that callers will want to use
141type Finder struct {
142 // configuration
143 DbPath string
144 numDbLoadingThreads int
145 numSearchingThreads int
146 cacheMetadata cacheMetadata
147 logger Logger
148 filesystem fs.FileSystem
149
150 // temporary state
Jeff Gastonb64fc1c2017-08-04 12:30:12 -0700151 threadPool *threadPool
152 mutex sync.Mutex
153 fsErrs []fsErr
154 errlock sync.Mutex
155 shutdownWaitgroup sync.WaitGroup
Jeff Gastonf1fd45e2017-08-09 18:25:28 -0700156
157 // non-temporary state
158 modifiedFlag int32
159 nodes pathMap
160}
161
Jeff Gastond3119522017-08-22 14:11:15 -0700162var defaultNumThreads = runtime.NumCPU() * 2
163
Jeff Gastonf1fd45e2017-08-09 18:25:28 -0700164// New creates a new Finder for use
165func New(cacheParams CacheParams, filesystem fs.FileSystem,
Jeff Gastonb629e182017-08-14 16:49:18 -0700166 logger Logger, dbPath string) (f *Finder, err error) {
Jeff Gastond3119522017-08-22 14:11:15 -0700167 return newImpl(cacheParams, filesystem, logger, dbPath, defaultNumThreads)
168}
Jeff Gastonf1fd45e2017-08-09 18:25:28 -0700169
Jeff Gastond3119522017-08-22 14:11:15 -0700170// newImpl is like New but accepts more params
171func newImpl(cacheParams CacheParams, filesystem fs.FileSystem,
172 logger Logger, dbPath string, numThreads int) (f *Finder, err error) {
Jeff Gastonf1fd45e2017-08-09 18:25:28 -0700173 numDbLoadingThreads := numThreads
174 numSearchingThreads := numThreads
175
176 metadata := cacheMetadata{
177 Version: versionString,
178 Config: cacheConfig{
179 CacheParams: cacheParams,
180 FilesystemView: filesystem.ViewId(),
181 },
182 }
183
Jeff Gastonb629e182017-08-14 16:49:18 -0700184 f = &Finder{
Jeff Gastonf1fd45e2017-08-09 18:25:28 -0700185 numDbLoadingThreads: numDbLoadingThreads,
186 numSearchingThreads: numSearchingThreads,
187 cacheMetadata: metadata,
188 logger: logger,
189 filesystem: filesystem,
190
191 nodes: *newPathMap("/"),
192 DbPath: dbPath,
Jeff Gastonb64fc1c2017-08-04 12:30:12 -0700193
194 shutdownWaitgroup: sync.WaitGroup{},
Jeff Gastonf1fd45e2017-08-09 18:25:28 -0700195 }
196
Jeff Gastonb629e182017-08-14 16:49:18 -0700197 f.loadFromFilesystem()
Jeff Gastonf1fd45e2017-08-09 18:25:28 -0700198
Jeff Gastonb629e182017-08-14 16:49:18 -0700199 // check for any filesystem errors
200 err = f.getErr()
201 if err != nil {
202 return nil, err
203 }
204
205 // confirm that every path mentioned in the CacheConfig exists
206 for _, path := range cacheParams.RootDirs {
Jeff Gastonb64fc1c2017-08-04 12:30:12 -0700207 if !filepath.IsAbs(path) {
208 path = filepath.Join(f.cacheMetadata.Config.WorkingDirectory, path)
209 }
Jeff Gastonb629e182017-08-14 16:49:18 -0700210 node := f.nodes.GetNode(filepath.Clean(path), false)
211 if node == nil || node.ModTime == 0 {
Jeff Gastonb64fc1c2017-08-04 12:30:12 -0700212 return nil, fmt.Errorf("path %v was specified to be included in the cache but does not exist\n", path)
Jeff Gastonb629e182017-08-14 16:49:18 -0700213 }
214 }
215
216 return f, nil
Jeff Gastonf1fd45e2017-08-09 18:25:28 -0700217}
218
219// FindNamed searches for every cached file
220func (f *Finder) FindAll() []string {
221 return f.FindAt("/")
222}
223
224// FindNamed searches for every cached file under <rootDir>
225func (f *Finder) FindAt(rootDir string) []string {
226 filter := func(entries DirEntries) (dirNames []string, fileNames []string) {
227 return entries.DirNames, entries.FileNames
228 }
229 return f.FindMatching(rootDir, filter)
230}
231
232// FindNamed searches for every cached file named <fileName>
233func (f *Finder) FindNamed(fileName string) []string {
234 return f.FindNamedAt("/", fileName)
235}
236
237// FindNamedAt searches under <rootPath> for every file named <fileName>
238// The reason a caller might use FindNamedAt instead of FindNamed is if they want
239// to limit their search to a subset of the cache
240func (f *Finder) FindNamedAt(rootPath string, fileName string) []string {
241 filter := func(entries DirEntries) (dirNames []string, fileNames []string) {
242 matches := []string{}
243 for _, foundName := range entries.FileNames {
244 if foundName == fileName {
245 matches = append(matches, foundName)
246 }
247 }
248 return entries.DirNames, matches
249
250 }
251 return f.FindMatching(rootPath, filter)
252}
253
254// FindFirstNamed searches for every file named <fileName>
255// Whenever it finds a match, it stops search subdirectories
256func (f *Finder) FindFirstNamed(fileName string) []string {
257 return f.FindFirstNamedAt("/", fileName)
258}
259
260// FindFirstNamedAt searches for every file named <fileName>
261// Whenever it finds a match, it stops search subdirectories
262func (f *Finder) FindFirstNamedAt(rootPath string, fileName string) []string {
263 filter := func(entries DirEntries) (dirNames []string, fileNames []string) {
264 matches := []string{}
265 for _, foundName := range entries.FileNames {
266 if foundName == fileName {
267 matches = append(matches, foundName)
268 }
269 }
270
271 if len(matches) > 0 {
272 return []string{}, matches
273 }
274 return entries.DirNames, matches
275 }
276 return f.FindMatching(rootPath, filter)
277}
278
279// FindMatching is the most general exported function for searching for files in the cache
280// The WalkFunc will be invoked repeatedly and is expected to modify the provided DirEntries
281// in place, removing file paths and directories as desired.
282// WalkFunc will be invoked potentially many times in parallel, and must be threadsafe.
283func (f *Finder) FindMatching(rootPath string, filter WalkFunc) []string {
284 // set up some parameters
285 scanStart := time.Now()
286 var isRel bool
287 workingDir := f.cacheMetadata.Config.WorkingDirectory
288
289 isRel = !filepath.IsAbs(rootPath)
290 if isRel {
291 rootPath = filepath.Join(workingDir, rootPath)
292 }
293
294 rootPath = filepath.Clean(rootPath)
295
296 // ensure nothing else is using the Finder
297 f.verbosef("FindMatching waiting for finder to be idle\n")
298 f.lock()
299 defer f.unlock()
300
301 node := f.nodes.GetNode(rootPath, false)
302 if node == nil {
303 f.verbosef("No data for path %v ; apparently not included in cache params: %v\n",
304 rootPath, f.cacheMetadata.Config.CacheParams)
305 // path is not found; don't do a search
306 return []string{}
307 }
308
309 // search for matching files
310 f.verbosef("Finder finding %v using cache\n", rootPath)
311 results := f.findInCacheMultithreaded(node, filter, f.numSearchingThreads)
312
313 // format and return results
314 if isRel {
315 for i := 0; i < len(results); i++ {
316 results[i] = strings.Replace(results[i], workingDir+"/", "", 1)
317 }
318 }
319 sort.Strings(results)
320 f.verbosef("Found %v files under %v in %v using cache\n",
321 len(results), rootPath, time.Since(scanStart))
322 return results
323}
324
Jeff Gastonb64fc1c2017-08-04 12:30:12 -0700325// Shutdown declares that the finder is no longer needed and waits for its cleanup to complete
326// Currently, that only entails waiting for the database dump to complete.
Jeff Gastonf1fd45e2017-08-09 18:25:28 -0700327func (f *Finder) Shutdown() {
Jeff Gastonb64fc1c2017-08-04 12:30:12 -0700328 f.waitForDbDump()
329}
330
331// End of public api
332
333func (f *Finder) goDumpDb() {
Jeff Gastonf1fd45e2017-08-09 18:25:28 -0700334 if f.wasModified() {
Jeff Gastonb64fc1c2017-08-04 12:30:12 -0700335 f.shutdownWaitgroup.Add(1)
336 go func() {
337 err := f.dumpDb()
338 if err != nil {
339 f.verbosef("%v\n", err)
340 }
341 f.shutdownWaitgroup.Done()
342 }()
Jeff Gastonf1fd45e2017-08-09 18:25:28 -0700343 } else {
344 f.verbosef("Skipping dumping unmodified db\n")
345 }
346}
347
Jeff Gastonb64fc1c2017-08-04 12:30:12 -0700348func (f *Finder) waitForDbDump() {
349 f.shutdownWaitgroup.Wait()
350}
Jeff Gastonf1fd45e2017-08-09 18:25:28 -0700351
352// joinCleanPaths is like filepath.Join but is faster because
353// joinCleanPaths doesn't have to support paths ending in "/" or containing ".."
354func joinCleanPaths(base string, leaf string) string {
355 if base == "" {
356 return leaf
357 }
358 if base == "/" {
359 return base + leaf
360 }
361 if leaf == "" {
362 return base
363 }
364 return base + "/" + leaf
365}
366
367func (f *Finder) verbosef(format string, args ...interface{}) {
368 f.logger.Output(2, fmt.Sprintf(format, args...))
369}
370
371// loadFromFilesystem populates the in-memory cache based on the contents of the filesystem
372func (f *Finder) loadFromFilesystem() {
373 f.threadPool = newThreadPool(f.numDbLoadingThreads)
374
375 err := f.startFromExternalCache()
376 if err != nil {
377 f.startWithoutExternalCache()
378 }
379
Jeff Gastonb64fc1c2017-08-04 12:30:12 -0700380 f.goDumpDb()
381
Jeff Gastonf1fd45e2017-08-09 18:25:28 -0700382 f.threadPool = nil
383}
384
385func (f *Finder) startFind(path string) {
386 if !filepath.IsAbs(path) {
387 path = filepath.Join(f.cacheMetadata.Config.WorkingDirectory, path)
388 }
389 node := f.nodes.GetNode(path, true)
390 f.statDirAsync(node)
391}
392
393func (f *Finder) lock() {
394 f.mutex.Lock()
395}
396
397func (f *Finder) unlock() {
398 f.mutex.Unlock()
399}
400
401// a statResponse is the relevant portion of the response from the filesystem to a Stat call
402type statResponse struct {
403 ModTime int64
404 Inode uint64
405 Device uint64
406}
407
408// a pathAndStats stores a path and its stats
409type pathAndStats struct {
410 statResponse
411
412 Path string
413}
414
415// a dirFullInfo stores all of the relevant information we know about a directory
416type dirFullInfo struct {
417 pathAndStats
418
419 FileNames []string
420}
421
422// a PersistedDirInfo is the information about a dir that we save to our cache on disk
423type PersistedDirInfo struct {
424 // These field names are short because they are repeated many times in the output json file
425 P string // path
426 T int64 // modification time
427 I uint64 // inode number
428 F []string // relevant filenames contained
429}
430
431// a PersistedDirs is the information that we persist for a group of dirs
432type PersistedDirs struct {
433 // the device on which each directory is stored
434 Device uint64
435 // the common root path to which all contained dirs are relative
436 Root string
437 // the directories themselves
438 Dirs []PersistedDirInfo
439}
440
441// a CacheEntry is the smallest unit that can be read and parsed from the cache (on disk) at a time
442type CacheEntry []PersistedDirs
443
444// a DirEntries lists the files and directories contained directly within a specific directory
445type DirEntries struct {
446 Path string
447
448 // elements of DirNames are just the dir names; they don't include any '/' character
449 DirNames []string
450 // elements of FileNames are just the file names; they don't include '/' character
451 FileNames []string
452}
453
454// a WalkFunc is the type that is passed into various Find functions for determining which
455// directories the caller wishes be walked. The WalkFunc is expected to decide which
456// directories to walk and which files to consider as matches to the original query.
457type WalkFunc func(DirEntries) (dirs []string, files []string)
458
459// a mapNode stores the relevant stats about a directory to be stored in a pathMap
460type mapNode struct {
461 statResponse
462 FileNames []string
463}
464
465// a pathMap implements the directory tree structure of nodes
466type pathMap struct {
467 mapNode
468
469 path string
470
471 children map[string]*pathMap
472
473 // number of descendent nodes, including self
474 approximateNumDescendents int
475}
476
477func newPathMap(path string) *pathMap {
478 result := &pathMap{path: path, children: make(map[string]*pathMap, 4),
479 approximateNumDescendents: 1}
480 return result
481}
482
483// GetNode returns the node at <path>
484func (m *pathMap) GetNode(path string, createIfNotFound bool) *pathMap {
485 if len(path) > 0 && path[0] == '/' {
486 path = path[1:]
487 }
488
489 node := m
490 for {
491 if path == "" {
492 return node
493 }
494
495 index := strings.Index(path, "/")
496 var firstComponent string
497 if index >= 0 {
498 firstComponent = path[:index]
499 path = path[index+1:]
500 } else {
501 firstComponent = path
502 path = ""
503 }
504
505 child, found := node.children[firstComponent]
506
507 if !found {
508 if createIfNotFound {
509 child = node.newChild(firstComponent)
510 } else {
511 return nil
512 }
513 }
514
515 node = child
516 }
517}
518
519func (m *pathMap) newChild(name string) (child *pathMap) {
520 path := joinCleanPaths(m.path, name)
521 newChild := newPathMap(path)
522 m.children[name] = newChild
523
524 return m.children[name]
525}
526
527func (m *pathMap) UpdateNumDescendents() int {
528 count := 1
529 for _, child := range m.children {
530 count += child.approximateNumDescendents
531 }
532 m.approximateNumDescendents = count
533 return count
534}
535
536func (m *pathMap) UpdateNumDescendentsRecursive() {
537 for _, child := range m.children {
538 child.UpdateNumDescendentsRecursive()
539 }
540 m.UpdateNumDescendents()
541}
542
543func (m *pathMap) MergeIn(other *pathMap) {
544 for key, theirs := range other.children {
545 ours, found := m.children[key]
546 if found {
547 ours.MergeIn(theirs)
548 } else {
549 m.children[key] = theirs
550 }
551 }
552 if other.ModTime != 0 {
553 m.mapNode = other.mapNode
554 }
555 m.UpdateNumDescendents()
556}
557
558func (m *pathMap) DumpAll() []dirFullInfo {
559 results := []dirFullInfo{}
560 m.dumpInto("", &results)
561 return results
562}
563
564func (m *pathMap) dumpInto(path string, results *[]dirFullInfo) {
565 *results = append(*results,
566 dirFullInfo{
567 pathAndStats{statResponse: m.statResponse, Path: path},
568 m.FileNames},
569 )
570 for key, child := range m.children {
571 childPath := joinCleanPaths(path, key)
572 if len(childPath) == 0 || childPath[0] != '/' {
573 childPath = "/" + childPath
574 }
575 child.dumpInto(childPath, results)
576 }
577}
578
579// a semaphore can be locked by up to <capacity> callers at once
580type semaphore struct {
581 pool chan bool
582}
583
584func newSemaphore(capacity int) *semaphore {
585 return &semaphore{pool: make(chan bool, capacity)}
586}
587
588func (l *semaphore) Lock() {
589 l.pool <- true
590}
591
592func (l *semaphore) Unlock() {
593 <-l.pool
594}
595
596// A threadPool runs goroutines and supports throttling and waiting.
597// Without throttling, Go may exhaust the maximum number of various resources, such as
598// threads or file descriptors, and crash the program.
599type threadPool struct {
600 receivedRequests sync.WaitGroup
601 activeRequests semaphore
602}
603
604func newThreadPool(maxNumConcurrentThreads int) *threadPool {
605 return &threadPool{
606 receivedRequests: sync.WaitGroup{},
607 activeRequests: *newSemaphore(maxNumConcurrentThreads),
608 }
609}
610
611// Run requests to run the given function in its own goroutine
612func (p *threadPool) Run(function func()) {
613 p.receivedRequests.Add(1)
614 // If Run() was called from within a goroutine spawned by this threadPool,
615 // then we may need to return from Run() before having capacity to actually
616 // run <function>.
617 //
618 // It's possible that the body of <function> contains a statement (such as a syscall)
619 // that will cause Go to pin it to a thread, or will contain a statement that uses
620 // another resource that is in short supply (such as a file descriptor), so we can't
621 // actually run <function> until we have capacity.
622 //
623 // However, the semaphore used for synchronization is implemented via a channel and
624 // shouldn't require a new thread for each access.
625 go func() {
626 p.activeRequests.Lock()
627 function()
628 p.activeRequests.Unlock()
629 p.receivedRequests.Done()
630 }()
631}
632
633// Wait waits until all goroutines are done, just like sync.WaitGroup's Wait
634func (p *threadPool) Wait() {
635 p.receivedRequests.Wait()
636}
637
Jeff Gastonb629e182017-08-14 16:49:18 -0700638type fsErr struct {
639 path string
640 err error
641}
642
643func (e fsErr) String() string {
644 return e.path + ": " + e.err.Error()
645}
646
Jeff Gastonf1fd45e2017-08-09 18:25:28 -0700647func (f *Finder) serializeCacheEntry(dirInfos []dirFullInfo) ([]byte, error) {
648 // group each dirFullInfo by its Device, to avoid having to repeat it in the output
649 dirsByDevice := map[uint64][]PersistedDirInfo{}
650 for _, entry := range dirInfos {
651 _, found := dirsByDevice[entry.Device]
652 if !found {
653 dirsByDevice[entry.Device] = []PersistedDirInfo{}
654 }
655 dirsByDevice[entry.Device] = append(dirsByDevice[entry.Device],
656 PersistedDirInfo{P: entry.Path, T: entry.ModTime, I: entry.Inode, F: entry.FileNames})
657 }
658
659 cacheEntry := CacheEntry{}
660
661 for device, infos := range dirsByDevice {
662 // find common prefix
663 prefix := ""
664 if len(infos) > 0 {
665 prefix = infos[0].P
666 }
667 for _, info := range infos {
668 for !strings.HasPrefix(info.P+"/", prefix+"/") {
669 prefix = filepath.Dir(prefix)
Jeff Gaston996716a2017-08-22 13:33:19 -0700670 if prefix == "/" {
671 break
672 }
Jeff Gastonf1fd45e2017-08-09 18:25:28 -0700673 }
674 }
675 // remove common prefix
676 for i := range infos {
677 suffix := strings.Replace(infos[i].P, prefix, "", 1)
678 if len(suffix) > 0 && suffix[0] == '/' {
679 suffix = suffix[1:]
680 }
681 infos[i].P = suffix
682 }
683
684 // turn the map (keyed by device) into a list of structs with labeled fields
685 // this is to improve readability of the output
686 cacheEntry = append(cacheEntry, PersistedDirs{Device: device, Root: prefix, Dirs: infos})
687 }
688
689 // convert to json.
690 // it would save some space to use a different format than json for the db file,
691 // but the space and time savings are small, and json is easy for humans to read
692 bytes, err := json.Marshal(cacheEntry)
693 return bytes, err
694}
695
696func (f *Finder) parseCacheEntry(bytes []byte) ([]dirFullInfo, error) {
697 var cacheEntry CacheEntry
698 err := json.Unmarshal(bytes, &cacheEntry)
699 if err != nil {
700 return nil, err
701 }
702
703 // convert from a CacheEntry to a []dirFullInfo (by copying a few fields)
704 capacity := 0
705 for _, element := range cacheEntry {
706 capacity += len(element.Dirs)
707 }
708 nodes := make([]dirFullInfo, capacity)
709 count := 0
710 for _, element := range cacheEntry {
711 for _, dir := range element.Dirs {
712 path := joinCleanPaths(element.Root, dir.P)
713
714 nodes[count] = dirFullInfo{
715 pathAndStats: pathAndStats{
716 statResponse: statResponse{
717 ModTime: dir.T, Inode: dir.I, Device: element.Device,
718 },
719 Path: path},
720 FileNames: dir.F}
721 count++
722 }
723 }
724 return nodes, nil
725}
726
727// We use the following separator byte to distinguish individually parseable blocks of json
728// because we know this separator won't appear in the json that we're parsing.
729//
730// The newline byte can only appear in a UTF-8 stream if the newline character appears, because:
731// - The newline character is encoded as "0000 1010" in binary ("0a" in hex)
732// - UTF-8 dictates that bytes beginning with a "0" bit are never emitted as part of a multibyte
733// character.
734//
735// We know that the newline character will never appear in our json string, because:
736// - If a newline character appears as part of a data string, then json encoding will
737// emit two characters instead: '\' and 'n'.
738// - The json encoder that we use doesn't emit the optional newlines between any of its
739// other outputs.
740const lineSeparator = byte('\n')
741
742func (f *Finder) readLine(reader *bufio.Reader) ([]byte, error) {
743 return reader.ReadBytes(lineSeparator)
744}
745
746// validateCacheHeader reads the cache header from cacheReader and tells whether the cache is compatible with this Finder
747func (f *Finder) validateCacheHeader(cacheReader *bufio.Reader) bool {
748 cacheVersionBytes, err := f.readLine(cacheReader)
749 if err != nil {
750 f.verbosef("Failed to read database header; database is invalid\n")
751 return false
752 }
753 if len(cacheVersionBytes) > 0 && cacheVersionBytes[len(cacheVersionBytes)-1] == lineSeparator {
754 cacheVersionBytes = cacheVersionBytes[:len(cacheVersionBytes)-1]
755 }
756 cacheVersionString := string(cacheVersionBytes)
757 currentVersion := f.cacheMetadata.Version
758 if cacheVersionString != currentVersion {
759 f.verbosef("Version changed from %q to %q, database is not applicable\n", cacheVersionString, currentVersion)
760 return false
761 }
762
763 cacheParamBytes, err := f.readLine(cacheReader)
764 if err != nil {
765 f.verbosef("Failed to read database search params; database is invalid\n")
766 return false
767 }
768
769 if len(cacheParamBytes) > 0 && cacheParamBytes[len(cacheParamBytes)-1] == lineSeparator {
770 cacheParamBytes = cacheParamBytes[:len(cacheParamBytes)-1]
771 }
772
773 currentParamBytes, err := f.cacheMetadata.Config.Dump()
774 if err != nil {
775 panic("Finder failed to serialize its parameters")
776 }
777 cacheParamString := string(cacheParamBytes)
778 currentParamString := string(currentParamBytes)
779 if cacheParamString != currentParamString {
780 f.verbosef("Params changed from %q to %q, database is not applicable\n", cacheParamString, currentParamString)
781 return false
782 }
783 return true
784}
785
786// loadBytes compares the cache info in <data> to the state of the filesystem
787// loadBytes returns a map representing <data> and also a slice of dirs that need to be re-walked
788func (f *Finder) loadBytes(id int, data []byte) (m *pathMap, dirsToWalk []string, err error) {
789
790 helperStartTime := time.Now()
791
792 cachedNodes, err := f.parseCacheEntry(data)
793 if err != nil {
794 return nil, nil, fmt.Errorf("Failed to parse block %v: %v\n", id, err.Error())
795 }
796
797 unmarshalDate := time.Now()
798 f.verbosef("Unmarshaled %v objects for %v in %v\n",
799 len(cachedNodes), id, unmarshalDate.Sub(helperStartTime))
800
801 tempMap := newPathMap("/")
802 stats := make([]statResponse, len(cachedNodes))
803
804 for i, node := range cachedNodes {
805 // check the file system for an updated timestamp
806 stats[i] = f.statDirSync(node.Path)
807 }
808
809 dirsToWalk = []string{}
810 for i, cachedNode := range cachedNodes {
811 updated := stats[i]
812 // save the cached value
813 container := tempMap.GetNode(cachedNode.Path, true)
814 container.mapNode = mapNode{statResponse: updated}
815
816 // if the metadata changed and the directory still exists, then
817 // make a note to walk it later
818 if !f.isInfoUpToDate(cachedNode.statResponse, updated) && updated.ModTime != 0 {
819 f.setModified()
820 // make a note that the directory needs to be walked
821 dirsToWalk = append(dirsToWalk, cachedNode.Path)
822 } else {
823 container.mapNode.FileNames = cachedNode.FileNames
824 }
825 }
826 // count the number of nodes to improve our understanding of the shape of the tree,
827 // thereby improving parallelism of subsequent searches
828 tempMap.UpdateNumDescendentsRecursive()
829
830 f.verbosef("Statted inodes of block %v in %v\n", id, time.Now().Sub(unmarshalDate))
831 return tempMap, dirsToWalk, nil
832}
833
834// startFromExternalCache loads the cache database from disk
835// startFromExternalCache waits to return until the load of the cache db is complete, but
836// startFromExternalCache does not wait for all every listDir() or statDir() request to complete
837func (f *Finder) startFromExternalCache() (err error) {
838 startTime := time.Now()
839 dbPath := f.DbPath
840
841 // open cache file and validate its header
842 reader, err := f.filesystem.Open(dbPath)
843 if err != nil {
844 return errors.New("No data to load from database\n")
845 }
846 bufferedReader := bufio.NewReader(reader)
847 if !f.validateCacheHeader(bufferedReader) {
848 return errors.New("Cache header does not match")
849 }
850 f.verbosef("Database header matches, will attempt to use database %v\n", f.DbPath)
851
852 // read the file and spawn threads to process it
853 nodesToWalk := [][]*pathMap{}
854 mainTree := newPathMap("/")
855
856 // read the blocks and stream them into <blockChannel>
857 type dataBlock struct {
858 id int
859 err error
860 data []byte
861 }
862 blockChannel := make(chan dataBlock, f.numDbLoadingThreads)
863 readBlocks := func() {
864 index := 0
865 for {
866 // It takes some time to unmarshal the input from json, so we want
867 // to unmarshal it in parallel. In order to find valid places to
868 // break the input, we scan for the line separators that we inserted
869 // (for this purpose) when we dumped the database.
870 data, err := f.readLine(bufferedReader)
871 var response dataBlock
872 done := false
873 if err != nil && err != io.EOF {
874 response = dataBlock{id: index, err: err, data: nil}
875 done = true
876 } else {
877 done = (err == io.EOF)
878 response = dataBlock{id: index, err: nil, data: data}
879 }
880 blockChannel <- response
881 index++
882 duration := time.Since(startTime)
883 f.verbosef("Read block %v after %v\n", index, duration)
884 if done {
885 f.verbosef("Read %v blocks in %v\n", index, duration)
886 close(blockChannel)
887 return
888 }
889 }
890 }
891 go readBlocks()
892
893 // Read from <blockChannel> and stream the responses into <resultChannel>.
894 type workResponse struct {
895 id int
896 err error
897 tree *pathMap
898 updatedDirs []string
899 }
900 resultChannel := make(chan workResponse)
901 processBlocks := func() {
902 numProcessed := 0
903 threadPool := newThreadPool(f.numDbLoadingThreads)
904 for {
905 // get a block to process
906 block, received := <-blockChannel
907 if !received {
908 break
909 }
910
911 if block.err != nil {
912 resultChannel <- workResponse{err: block.err}
913 break
914 }
915 numProcessed++
916 // wait until there is CPU available to process it
917 threadPool.Run(
918 func() {
919 processStartTime := time.Now()
920 f.verbosef("Starting to process block %v after %v\n",
921 block.id, processStartTime.Sub(startTime))
922 tempMap, updatedDirs, err := f.loadBytes(block.id, block.data)
923 var response workResponse
924 if err != nil {
925 f.verbosef(
926 "Block %v failed to parse with error %v\n",
927 block.id, err)
928 response = workResponse{err: err}
929 } else {
930 response = workResponse{
931 id: block.id,
932 err: nil,
933 tree: tempMap,
934 updatedDirs: updatedDirs,
935 }
936 }
937 f.verbosef("Processed block %v in %v\n",
938 block.id, time.Since(processStartTime),
939 )
940 resultChannel <- response
941 },
942 )
943 }
944 threadPool.Wait()
945 f.verbosef("Finished processing %v blocks in %v\n",
946 numProcessed, time.Since(startTime))
947 close(resultChannel)
948 }
949 go processBlocks()
950
951 // Read from <resultChannel> and use the results
952 combineResults := func() (err error) {
953 for {
954 result, received := <-resultChannel
955 if !received {
956 break
957 }
958 if err != nil {
959 // In case of an error, wait for work to complete before
960 // returning the error. This ensures that any subsequent
961 // work doesn't need to compete for resources (and possibly
962 // fail due to, for example, a filesystem limit on the number of
963 // concurrently open files) with past work.
964 continue
965 }
966 if result.err != nil {
967 err = result.err
968 continue
969 }
970 // update main tree
971 mainTree.MergeIn(result.tree)
972 // record any new directories that we will need to Stat()
973 updatedNodes := make([]*pathMap, len(result.updatedDirs))
974 for j, dir := range result.updatedDirs {
975 node := mainTree.GetNode(dir, false)
976 updatedNodes[j] = node
977 }
978 nodesToWalk = append(nodesToWalk, updatedNodes)
979 }
980 return err
981 }
982 err = combineResults()
983 if err != nil {
984 return err
985 }
986
987 f.nodes = *mainTree
988
989 // after having loaded the entire db and therefore created entries for
990 // the directories we know of, now it's safe to start calling ReadDir on
991 // any updated directories
992 for i := range nodesToWalk {
993 f.listDirsAsync(nodesToWalk[i])
994 }
Jeff Gastonb629e182017-08-14 16:49:18 -0700995 f.verbosef("Loaded db and statted known dirs in %v\n", time.Since(startTime))
996 f.threadPool.Wait()
997 f.verbosef("Loaded db and statted all dirs in %v\n", time.Now().Sub(startTime))
998
Jeff Gastonf1fd45e2017-08-09 18:25:28 -0700999 return err
1000}
1001
1002// startWithoutExternalCache starts scanning the filesystem according to the cache config
1003// startWithoutExternalCache should be called if startFromExternalCache is not applicable
1004func (f *Finder) startWithoutExternalCache() {
Jeff Gastonb629e182017-08-14 16:49:18 -07001005 startTime := time.Now()
Jeff Gastonf1fd45e2017-08-09 18:25:28 -07001006 configDirs := f.cacheMetadata.Config.RootDirs
1007
1008 // clean paths
1009 candidates := make([]string, len(configDirs))
1010 for i, dir := range configDirs {
1011 candidates[i] = filepath.Clean(dir)
1012 }
1013 // remove duplicates
1014 dirsToScan := make([]string, 0, len(configDirs))
1015 for _, candidate := range candidates {
1016 include := true
1017 for _, included := range dirsToScan {
1018 if included == "/" || strings.HasPrefix(candidate+"/", included+"/") {
1019 include = false
1020 break
1021 }
1022 }
1023 if include {
1024 dirsToScan = append(dirsToScan, candidate)
1025 }
1026 }
1027
1028 // start searching finally
1029 for _, path := range dirsToScan {
1030 f.verbosef("Starting find of %v\n", path)
1031 f.startFind(path)
1032 }
Jeff Gastonb629e182017-08-14 16:49:18 -07001033
1034 f.threadPool.Wait()
1035
1036 f.verbosef("Scanned filesystem (not using cache) in %v\n", time.Now().Sub(startTime))
Jeff Gastonf1fd45e2017-08-09 18:25:28 -07001037}
1038
1039// isInfoUpToDate tells whether <new> can confirm that results computed at <old> are still valid
1040func (f *Finder) isInfoUpToDate(old statResponse, new statResponse) (equal bool) {
1041 if old.Inode != new.Inode {
1042 return false
1043 }
1044 if old.ModTime != new.ModTime {
1045 return false
1046 }
1047 if old.Device != new.Device {
1048 return false
1049 }
1050 return true
1051}
1052
1053func (f *Finder) wasModified() bool {
1054 return atomic.LoadInt32(&f.modifiedFlag) > 0
1055}
1056
1057func (f *Finder) setModified() {
1058 var newVal int32
1059 newVal = 1
1060 atomic.StoreInt32(&f.modifiedFlag, newVal)
1061}
1062
1063// sortedDirEntries exports directory entries to facilitate dumping them to the external cache
1064func (f *Finder) sortedDirEntries() []dirFullInfo {
1065 startTime := time.Now()
1066 nodes := make([]dirFullInfo, 0)
1067 for _, node := range f.nodes.DumpAll() {
1068 if node.ModTime != 0 {
1069 nodes = append(nodes, node)
1070 }
1071 }
1072 discoveryDate := time.Now()
1073 f.verbosef("Generated %v cache entries in %v\n", len(nodes), discoveryDate.Sub(startTime))
1074 less := func(i int, j int) bool {
1075 return nodes[i].Path < nodes[j].Path
1076 }
1077 sort.Slice(nodes, less)
1078 sortDate := time.Now()
1079 f.verbosef("Sorted %v cache entries in %v\n", len(nodes), sortDate.Sub(discoveryDate))
1080
1081 return nodes
1082}
1083
1084// serializeDb converts the cache database into a form to save to disk
1085func (f *Finder) serializeDb() ([]byte, error) {
1086 // sort dir entries
1087 var entryList = f.sortedDirEntries()
1088
1089 // Generate an output file that can be conveniently loaded using the same number of threads
1090 // as were used in this execution (because presumably that will be the number of threads
1091 // used in the next execution too)
1092
1093 // generate header
1094 header := []byte{}
1095 header = append(header, []byte(f.cacheMetadata.Version)...)
1096 header = append(header, lineSeparator)
1097 configDump, err := f.cacheMetadata.Config.Dump()
1098 if err != nil {
1099 return nil, err
1100 }
1101 header = append(header, configDump...)
1102
1103 // serialize individual blocks in parallel
1104 numBlocks := f.numDbLoadingThreads
1105 if numBlocks > len(entryList) {
1106 numBlocks = len(entryList)
1107 }
1108 blocks := make([][]byte, 1+numBlocks)
1109 blocks[0] = header
1110 blockMin := 0
1111 wg := sync.WaitGroup{}
1112 var errLock sync.Mutex
1113
1114 for i := 1; i <= numBlocks; i++ {
1115 // identify next block
1116 blockMax := len(entryList) * i / numBlocks
1117 block := entryList[blockMin:blockMax]
1118
1119 // process block
1120 wg.Add(1)
1121 go func(index int, block []dirFullInfo) {
1122 byteBlock, subErr := f.serializeCacheEntry(block)
1123 f.verbosef("Serialized block %v into %v bytes\n", index, len(byteBlock))
1124 if subErr != nil {
1125 f.verbosef("%v\n", subErr.Error())
1126 errLock.Lock()
1127 err = subErr
1128 errLock.Unlock()
1129 } else {
1130 blocks[index] = byteBlock
1131 }
1132 wg.Done()
1133 }(i, block)
1134
1135 blockMin = blockMax
1136 }
1137
1138 wg.Wait()
1139
1140 if err != nil {
1141 return nil, err
1142 }
1143
1144 content := bytes.Join(blocks, []byte{lineSeparator})
1145
1146 return content, nil
1147}
1148
1149// dumpDb saves the cache database to disk
1150func (f *Finder) dumpDb() error {
1151 startTime := time.Now()
1152 f.verbosef("Dumping db\n")
1153
1154 tempPath := f.DbPath + ".tmp"
1155
1156 bytes, err := f.serializeDb()
1157 if err != nil {
1158 return err
1159 }
1160 serializeDate := time.Now()
1161 f.verbosef("Serialized db in %v\n", serializeDate.Sub(startTime))
1162 // dump file and atomically move
1163 err = f.filesystem.WriteFile(tempPath, bytes, 0777)
1164 if err != nil {
1165 return err
1166 }
1167 err = f.filesystem.Rename(tempPath, f.DbPath)
1168 if err != nil {
1169 return err
1170 }
1171
1172 f.verbosef("Wrote db in %v\n", time.Now().Sub(serializeDate))
1173 return nil
Jeff Gastonb629e182017-08-14 16:49:18 -07001174
1175}
1176
1177// canIgnoreFsErr checks for certain classes of filesystem errors that are safe to ignore
1178func (f *Finder) canIgnoreFsErr(err error) bool {
1179 pathErr, isPathErr := err.(*os.PathError)
1180 if !isPathErr {
1181 // Don't recognize this error
1182 return false
1183 }
Jeff Gaston821271b2017-08-23 17:53:38 -07001184 if os.IsPermission(pathErr) {
Jeff Gastonb629e182017-08-14 16:49:18 -07001185 // Permission errors are ignored:
1186 // https://issuetracker.google.com/37553659
1187 // https://github.com/google/kati/pull/116
1188 return true
1189 }
1190 if pathErr.Err == os.ErrNotExist {
1191 // If a directory doesn't exist, that generally means the cache is out-of-date
1192 return true
1193 }
1194 // Don't recognize this error
1195 return false
1196}
1197
1198// onFsError should be called whenever a potentially fatal error is returned from a filesystem call
1199func (f *Finder) onFsError(path string, err error) {
1200 if !f.canIgnoreFsErr(err) {
1201 // We could send the errors through a channel instead, although that would cause this call
1202 // to block unless we preallocated a sufficient buffer or spawned a reader thread.
1203 // Although it wouldn't be too complicated to spawn a reader thread, it's still slightly
1204 // more convenient to use a lock. Only in an unusual situation should this code be
1205 // invoked anyway.
1206 f.errlock.Lock()
1207 f.fsErrs = append(f.fsErrs, fsErr{path: path, err: err})
1208 f.errlock.Unlock()
1209 }
1210}
1211
1212// discardErrsForPrunedPaths removes any errors for paths that are no longer included in the cache
1213func (f *Finder) discardErrsForPrunedPaths() {
1214 // This function could be somewhat inefficient due to being single-threaded,
1215 // but the length of f.fsErrs should be approximately 0, so it shouldn't take long anyway.
1216 relevantErrs := make([]fsErr, 0, len(f.fsErrs))
1217 for _, fsErr := range f.fsErrs {
1218 path := fsErr.path
1219 node := f.nodes.GetNode(path, false)
1220 if node != nil {
1221 // The path in question wasn't pruned due to a failure to process a parent directory.
1222 // So, the failure to process this path is important
1223 relevantErrs = append(relevantErrs, fsErr)
1224 }
1225 }
1226 f.fsErrs = relevantErrs
1227}
1228
1229// getErr returns an error based on previous calls to onFsErr, if any
1230func (f *Finder) getErr() error {
1231 f.discardErrsForPrunedPaths()
1232
1233 numErrs := len(f.fsErrs)
1234 if numErrs < 1 {
1235 return nil
1236 }
1237
1238 maxNumErrsToInclude := 10
1239 message := ""
1240 if numErrs > maxNumErrsToInclude {
1241 message = fmt.Sprintf("finder encountered %v errors: %v...", numErrs, f.fsErrs[:maxNumErrsToInclude])
1242 } else {
1243 message = fmt.Sprintf("finder encountered %v errors: %v", numErrs, f.fsErrs)
1244 }
1245
1246 return errors.New(message)
Jeff Gastonf1fd45e2017-08-09 18:25:28 -07001247}
1248
1249func (f *Finder) statDirAsync(dir *pathMap) {
1250 node := dir
1251 path := dir.path
1252 f.threadPool.Run(
1253 func() {
1254 updatedStats := f.statDirSync(path)
1255
1256 if !f.isInfoUpToDate(node.statResponse, updatedStats) {
1257 node.mapNode = mapNode{
1258 statResponse: updatedStats,
1259 FileNames: []string{},
1260 }
1261 f.setModified()
1262 if node.statResponse.ModTime != 0 {
1263 // modification time was updated, so re-scan for
1264 // child directories
1265 f.listDirAsync(dir)
1266 }
1267 }
1268 },
1269 )
1270}
1271
1272func (f *Finder) statDirSync(path string) statResponse {
1273
1274 fileInfo, err := f.filesystem.Lstat(path)
1275
1276 var stats statResponse
1277 if err != nil {
Jeff Gastonb629e182017-08-14 16:49:18 -07001278 // possibly record this error
1279 f.onFsError(path, err)
Jeff Gastonf1fd45e2017-08-09 18:25:28 -07001280 // in case of a failure to stat the directory, treat the directory as missing (modTime = 0)
1281 return stats
1282 }
1283 modTime := fileInfo.ModTime()
1284 stats = statResponse{}
1285 inode, err := f.filesystem.InodeNumber(fileInfo)
1286 if err != nil {
1287 panic(fmt.Sprintf("Could not get inode number of %v: %v\n", path, err.Error()))
1288 }
1289 stats.Inode = inode
1290 device, err := f.filesystem.DeviceNumber(fileInfo)
1291 if err != nil {
1292 panic(fmt.Sprintf("Could not get device number of %v: %v\n", path, err.Error()))
1293 }
1294 stats.Device = device
1295 permissionsChangeTime, err := f.filesystem.PermTime(fileInfo)
1296
1297 if err != nil {
1298 panic(fmt.Sprintf("Could not get permissions modification time (CTime) of %v: %v\n", path, err.Error()))
1299 }
1300 // We're only interested in knowing whether anything about the directory
1301 // has changed since last check, so we use the latest of the two
1302 // modification times (content modification (mtime) and
1303 // permission modification (ctime))
1304 if permissionsChangeTime.After(modTime) {
1305 modTime = permissionsChangeTime
1306 }
1307 stats.ModTime = modTime.UnixNano()
1308
1309 return stats
1310}
1311
1312// pruneCacheCandidates removes the items that we don't want to include in our persistent cache
1313func (f *Finder) pruneCacheCandidates(items *DirEntries) {
1314
1315 for _, fileName := range items.FileNames {
1316 for _, abortedName := range f.cacheMetadata.Config.PruneFiles {
1317 if fileName == abortedName {
1318 items.FileNames = []string{}
1319 items.DirNames = []string{}
1320 return
1321 }
1322 }
1323 }
1324
1325 // remove any files that aren't the ones we want to include
1326 writeIndex := 0
1327 for _, fileName := range items.FileNames {
1328 // include only these files
1329 for _, includedName := range f.cacheMetadata.Config.IncludeFiles {
1330 if fileName == includedName {
1331 items.FileNames[writeIndex] = fileName
1332 writeIndex++
1333 break
1334 }
1335 }
1336 }
1337 // resize
1338 items.FileNames = items.FileNames[:writeIndex]
1339
1340 writeIndex = 0
1341 for _, dirName := range items.DirNames {
1342 items.DirNames[writeIndex] = dirName
1343 // ignore other dirs that are known to not be inputs to the build process
1344 include := true
1345 for _, excludedName := range f.cacheMetadata.Config.ExcludeDirs {
1346 if dirName == excludedName {
1347 // don't include
1348 include = false
1349 break
1350 }
1351 }
1352 if include {
1353 writeIndex++
1354 }
1355 }
1356 // resize
1357 items.DirNames = items.DirNames[:writeIndex]
1358}
1359
1360func (f *Finder) listDirsAsync(nodes []*pathMap) {
1361 f.threadPool.Run(
1362 func() {
1363 for i := range nodes {
1364 f.listDirSync(nodes[i])
1365 }
1366 },
1367 )
1368}
1369
1370func (f *Finder) listDirAsync(node *pathMap) {
1371 f.threadPool.Run(
1372 func() {
1373 f.listDirSync(node)
1374 },
1375 )
1376}
1377
1378func (f *Finder) listDirSync(dir *pathMap) {
1379 path := dir.path
1380 children, err := f.filesystem.ReadDir(path)
1381
1382 if err != nil {
Jeff Gastonb629e182017-08-14 16:49:18 -07001383 // possibly record this error
1384 f.onFsError(path, err)
Jeff Gastonf1fd45e2017-08-09 18:25:28 -07001385 // if listing the contents of the directory fails (presumably due to
1386 // permission denied), then treat the directory as empty
Colin Crossa88c8832017-12-21 15:39:26 -08001387 children = nil
Jeff Gastonf1fd45e2017-08-09 18:25:28 -07001388 }
1389
1390 var subdirs []string
1391 var subfiles []string
1392
1393 for _, child := range children {
1394 linkBits := child.Mode() & os.ModeSymlink
1395 isLink := linkBits != 0
1396 if child.IsDir() {
1397 if !isLink {
1398 // Skip symlink dirs.
1399 // We don't have to support symlink dirs because
1400 // that would cause duplicates.
1401 subdirs = append(subdirs, child.Name())
1402 }
1403 } else {
1404 // We do have to support symlink files because the link name might be
1405 // different than the target name
1406 // (for example, Android.bp -> build/soong/root.bp)
1407 subfiles = append(subfiles, child.Name())
1408 }
1409
1410 }
1411 parentNode := dir
1412
1413 entry := &DirEntries{Path: path, DirNames: subdirs, FileNames: subfiles}
1414 f.pruneCacheCandidates(entry)
1415
1416 // create a pathMap node for each relevant subdirectory
1417 relevantChildren := map[string]*pathMap{}
1418 for _, subdirName := range entry.DirNames {
1419 childNode, found := parentNode.children[subdirName]
1420 // if we already knew of this directory, then we already have a request pending to Stat it
1421 // if we didn't already know of this directory, then we must Stat it now
1422 if !found {
1423 childNode = parentNode.newChild(subdirName)
1424 f.statDirAsync(childNode)
1425 }
1426 relevantChildren[subdirName] = childNode
1427 }
1428 // Note that in rare cases, it's possible that we're reducing the set of
1429 // children via this statement, if these are all true:
1430 // 1. we previously had a cache that knew about subdirectories of parentNode
1431 // 2. the user created a prune-file (described in pruneCacheCandidates)
1432 // inside <parentNode>, which specifies that the contents of parentNode
1433 // are to be ignored.
1434 // The fact that it's possible to remove children here means that *pathMap structs
1435 // must not be looked up from f.nodes by filepath (and instead must be accessed by
1436 // direct pointer) until after every listDirSync completes
1437 parentNode.FileNames = entry.FileNames
1438 parentNode.children = relevantChildren
1439
1440}
1441
1442// listMatches takes a node and a function that specifies which subdirectories and
1443// files to include, and listMatches returns the matches
1444func (f *Finder) listMatches(node *pathMap,
1445 filter WalkFunc) (subDirs []*pathMap, filePaths []string) {
1446 entries := DirEntries{
1447 FileNames: node.FileNames,
1448 }
1449 entries.DirNames = make([]string, 0, len(node.children))
1450 for childName := range node.children {
1451 entries.DirNames = append(entries.DirNames, childName)
1452 }
1453
1454 dirNames, fileNames := filter(entries)
1455
1456 subDirs = []*pathMap{}
1457 filePaths = make([]string, 0, len(fileNames))
1458 for _, fileName := range fileNames {
1459 filePaths = append(filePaths, joinCleanPaths(node.path, fileName))
1460 }
1461 subDirs = make([]*pathMap, 0, len(dirNames))
1462 for _, childName := range dirNames {
1463 child, ok := node.children[childName]
1464 if ok {
1465 subDirs = append(subDirs, child)
1466 }
1467 }
1468
1469 return subDirs, filePaths
1470}
1471
1472// findInCacheMultithreaded spawns potentially multiple goroutines with which to search the cache.
1473func (f *Finder) findInCacheMultithreaded(node *pathMap, filter WalkFunc,
1474 approxNumThreads int) []string {
1475
1476 if approxNumThreads < 2 {
1477 // Done spawning threads; process remaining directories
1478 return f.findInCacheSinglethreaded(node, filter)
1479 }
1480
1481 totalWork := 0
1482 for _, child := range node.children {
1483 totalWork += child.approximateNumDescendents
1484 }
1485 childrenResults := make(chan []string, len(node.children))
1486
1487 subDirs, filePaths := f.listMatches(node, filter)
1488
1489 // process child directories
1490 for _, child := range subDirs {
1491 numChildThreads := approxNumThreads * child.approximateNumDescendents / totalWork
1492 childProcessor := func(child *pathMap) {
1493 childResults := f.findInCacheMultithreaded(child, filter, numChildThreads)
1494 childrenResults <- childResults
1495 }
1496 // If we're allowed to use more than 1 thread to process this directory,
1497 // then instead we use 1 thread for each subdirectory.
1498 // It would be strange to spawn threads for only some subdirectories.
1499 go childProcessor(child)
1500 }
1501
1502 // collect results
1503 for i := 0; i < len(subDirs); i++ {
1504 childResults := <-childrenResults
1505 filePaths = append(filePaths, childResults...)
1506 }
1507 close(childrenResults)
1508
1509 return filePaths
1510}
1511
1512// findInCacheSinglethreaded synchronously searches the cache for all matching file paths
1513// note findInCacheSinglethreaded runs 2X to 4X as fast by being iterative rather than recursive
1514func (f *Finder) findInCacheSinglethreaded(node *pathMap, filter WalkFunc) []string {
1515 if node == nil {
1516 return []string{}
1517 }
1518
1519 nodes := []*pathMap{node}
1520 matches := []string{}
1521
1522 for len(nodes) > 0 {
1523 currentNode := nodes[0]
1524 nodes = nodes[1:]
1525
1526 subDirs, filePaths := f.listMatches(currentNode, filter)
1527
1528 nodes = append(nodes, subDirs...)
1529
1530 matches = append(matches, filePaths...)
1531 }
1532 return matches
1533}