rclone/fs/cache/cache.go

286 lines
8.3 KiB
Go
Raw Permalink Normal View History

// Package cache implements the Fs cache
package cache
import (
"context"
"runtime"
fs/cache: fix parent not getting pinned when remote is a file Before this change, when cache.GetFn was called on a file rather than a directory, two cache entries would be added (the file + its parent) but only one of them would get pinned if the caller then called Pin(f). This left the other one exposed to expiration if the ci.FsCacheExpireDuration was reached. This was problematic because both entries point to the same Fs, and if one entry expires while the other is pinned, the Shutdown method gets erroneously called on an Fs that is still in use. An example of the problem showed up in the Hasher backend, which uses the Shutdown method to stop the bolt db used to store hashes. If a command was run on a Hasher file (ex. `rclone md5sum --download hasher:somelargefile.zip`) and hashing the file took longer than the --fs-cache-expire-duration (5m by default), the bolt db was stopped before the hashing operation completed, resulting in an error. This change fixes the issue by ensuring that: 1. only one entry is added to the cache (the file's parent, not the file). 2. future lookups correctly find the entry regardless of whether they are called with the parent name or one of its children. 3. fs.ErrorIsFile is returned when (and only when) fsString points to a file (preserving the fix from https://github.com/rclone/rclone/commit/8d5bc7f28b8459a0e04fdbfd2dca43c5d9ec4555). Note that f.Root() should always point to the parent dir as of https://github.com/rclone/rclone/commit/c69eb84573c85206ab028eda2987180e049ef2e4
2024-03-10 12:36:57 +08:00
"strings"
"sync"
"github.com/rclone/rclone/fs"
"github.com/rclone/rclone/fs/filter"
2019-08-01 05:19:07 +08:00
"github.com/rclone/rclone/lib/cache"
)
var (
fs/cache: fix parent not getting pinned when remote is a file Before this change, when cache.GetFn was called on a file rather than a directory, two cache entries would be added (the file + its parent) but only one of them would get pinned if the caller then called Pin(f). This left the other one exposed to expiration if the ci.FsCacheExpireDuration was reached. This was problematic because both entries point to the same Fs, and if one entry expires while the other is pinned, the Shutdown method gets erroneously called on an Fs that is still in use. An example of the problem showed up in the Hasher backend, which uses the Shutdown method to stop the bolt db used to store hashes. If a command was run on a Hasher file (ex. `rclone md5sum --download hasher:somelargefile.zip`) and hashing the file took longer than the --fs-cache-expire-duration (5m by default), the bolt db was stopped before the hashing operation completed, resulting in an error. This change fixes the issue by ensuring that: 1. only one entry is added to the cache (the file's parent, not the file). 2. future lookups correctly find the entry regardless of whether they are called with the parent name or one of its children. 3. fs.ErrorIsFile is returned when (and only when) fsString points to a file (preserving the fix from https://github.com/rclone/rclone/commit/8d5bc7f28b8459a0e04fdbfd2dca43c5d9ec4555). Note that f.Root() should always point to the parent dir as of https://github.com/rclone/rclone/commit/c69eb84573c85206ab028eda2987180e049ef2e4
2024-03-10 12:36:57 +08:00
once sync.Once // creation
c *cache.Cache
mu sync.Mutex // mutex to protect remap
remap = map[string]string{} // map user supplied names to canonical names - [fsString]canonicalName
childParentMap = map[string]string{} // tracks a one-to-many relationship between parent dirs and their direct children files - [child]parent
)
// Create the cache just once
func createOnFirstUse() {
once.Do(func() {
ci := fs.GetConfig(context.Background())
c = cache.New()
c.SetExpireDuration(ci.FsCacheExpireDuration)
c.SetExpireInterval(ci.FsCacheExpireInterval)
c.SetFinalizer(func(value interface{}) {
if s, ok := value.(fs.Shutdowner); ok {
_ = fs.CountError(context.Background(), s.Shutdown(context.Background()))
}
})
})
}
// Canonicalize looks up fsString in the mapping from user supplied
// names to canonical names and return the canonical form
func Canonicalize(fsString string) string {
createOnFirstUse()
mu.Lock()
canonicalName, ok := remap[fsString]
mu.Unlock()
if !ok {
return fsString
}
fs.Debugf(nil, "fs cache: switching user supplied name %q for canonical name %q", fsString, canonicalName)
return canonicalName
}
// Put in a mapping from fsString => canonicalName if they are different
func addMapping(fsString, canonicalName string) {
if canonicalName == fsString {
return
}
mu.Lock()
remap[fsString] = canonicalName
mu.Unlock()
}
fs/cache: fix parent not getting pinned when remote is a file Before this change, when cache.GetFn was called on a file rather than a directory, two cache entries would be added (the file + its parent) but only one of them would get pinned if the caller then called Pin(f). This left the other one exposed to expiration if the ci.FsCacheExpireDuration was reached. This was problematic because both entries point to the same Fs, and if one entry expires while the other is pinned, the Shutdown method gets erroneously called on an Fs that is still in use. An example of the problem showed up in the Hasher backend, which uses the Shutdown method to stop the bolt db used to store hashes. If a command was run on a Hasher file (ex. `rclone md5sum --download hasher:somelargefile.zip`) and hashing the file took longer than the --fs-cache-expire-duration (5m by default), the bolt db was stopped before the hashing operation completed, resulting in an error. This change fixes the issue by ensuring that: 1. only one entry is added to the cache (the file's parent, not the file). 2. future lookups correctly find the entry regardless of whether they are called with the parent name or one of its children. 3. fs.ErrorIsFile is returned when (and only when) fsString points to a file (preserving the fix from https://github.com/rclone/rclone/commit/8d5bc7f28b8459a0e04fdbfd2dca43c5d9ec4555). Note that f.Root() should always point to the parent dir as of https://github.com/rclone/rclone/commit/c69eb84573c85206ab028eda2987180e049ef2e4
2024-03-10 12:36:57 +08:00
// addChild tracks known file (child) to directory (parent) relationships.
// Note that the canonicalName of a child will always equal that of its parent,
// but not everything with an equal canonicalName is a child.
// It could be an alias or overridden version of a directory.
func addChild(child, parent string) {
if child == parent {
return
}
mu.Lock()
childParentMap[child] = parent
mu.Unlock()
}
// returns true if name is definitely known to be a child (i.e. a file, not a dir).
// returns false if name is a dir or if we don't know.
func isChild(child string) bool {
mu.Lock()
_, found := childParentMap[child]
mu.Unlock()
return found
}
// ensures that we return fs.ErrorIsFile when necessary
func getError(fsString string, err error) error {
if err != nil && err != fs.ErrorIsFile {
return err
}
if isChild(fsString) {
return fs.ErrorIsFile
}
return nil
}
// GetFn gets an fs.Fs named fsString either from the cache or creates
2019-08-01 05:19:07 +08:00
// it afresh with the create function
func GetFn(ctx context.Context, fsString string, create func(ctx context.Context, fsString string) (fs.Fs, error)) (f fs.Fs, err error) {
createOnFirstUse()
canonicalFsString := Canonicalize(fsString)
created := false
value, err := c.Get(canonicalFsString, func(canonicalFsString string) (f interface{}, ok bool, err error) {
f, err = create(ctx, fsString) // always create the backend with the original non-canonicalised string
2019-08-01 05:19:07 +08:00
ok = err == nil || err == fs.ErrorIsFile
created = ok
2019-08-01 05:19:07 +08:00
return f, ok, err
})
fs/cache: fix parent not getting pinned when remote is a file Before this change, when cache.GetFn was called on a file rather than a directory, two cache entries would be added (the file + its parent) but only one of them would get pinned if the caller then called Pin(f). This left the other one exposed to expiration if the ci.FsCacheExpireDuration was reached. This was problematic because both entries point to the same Fs, and if one entry expires while the other is pinned, the Shutdown method gets erroneously called on an Fs that is still in use. An example of the problem showed up in the Hasher backend, which uses the Shutdown method to stop the bolt db used to store hashes. If a command was run on a Hasher file (ex. `rclone md5sum --download hasher:somelargefile.zip`) and hashing the file took longer than the --fs-cache-expire-duration (5m by default), the bolt db was stopped before the hashing operation completed, resulting in an error. This change fixes the issue by ensuring that: 1. only one entry is added to the cache (the file's parent, not the file). 2. future lookups correctly find the entry regardless of whether they are called with the parent name or one of its children. 3. fs.ErrorIsFile is returned when (and only when) fsString points to a file (preserving the fix from https://github.com/rclone/rclone/commit/8d5bc7f28b8459a0e04fdbfd2dca43c5d9ec4555). Note that f.Root() should always point to the parent dir as of https://github.com/rclone/rclone/commit/c69eb84573c85206ab028eda2987180e049ef2e4
2024-03-10 12:36:57 +08:00
f, ok := value.(fs.Fs)
if err != nil && err != fs.ErrorIsFile {
fs/cache: fix parent not getting pinned when remote is a file Before this change, when cache.GetFn was called on a file rather than a directory, two cache entries would be added (the file + its parent) but only one of them would get pinned if the caller then called Pin(f). This left the other one exposed to expiration if the ci.FsCacheExpireDuration was reached. This was problematic because both entries point to the same Fs, and if one entry expires while the other is pinned, the Shutdown method gets erroneously called on an Fs that is still in use. An example of the problem showed up in the Hasher backend, which uses the Shutdown method to stop the bolt db used to store hashes. If a command was run on a Hasher file (ex. `rclone md5sum --download hasher:somelargefile.zip`) and hashing the file took longer than the --fs-cache-expire-duration (5m by default), the bolt db was stopped before the hashing operation completed, resulting in an error. This change fixes the issue by ensuring that: 1. only one entry is added to the cache (the file's parent, not the file). 2. future lookups correctly find the entry regardless of whether they are called with the parent name or one of its children. 3. fs.ErrorIsFile is returned when (and only when) fsString points to a file (preserving the fix from https://github.com/rclone/rclone/commit/8d5bc7f28b8459a0e04fdbfd2dca43c5d9ec4555). Note that f.Root() should always point to the parent dir as of https://github.com/rclone/rclone/commit/c69eb84573c85206ab028eda2987180e049ef2e4
2024-03-10 12:36:57 +08:00
if ok {
return f, err // for possible future uses of PutErr
}
2019-08-01 05:19:07 +08:00
return nil, err
}
// Check we stored the Fs at the canonical name
if created {
canonicalName := fs.ConfigString(f)
if canonicalName != canonicalFsString {
fs/cache: fix parent not getting pinned when remote is a file Before this change, when cache.GetFn was called on a file rather than a directory, two cache entries would be added (the file + its parent) but only one of them would get pinned if the caller then called Pin(f). This left the other one exposed to expiration if the ci.FsCacheExpireDuration was reached. This was problematic because both entries point to the same Fs, and if one entry expires while the other is pinned, the Shutdown method gets erroneously called on an Fs that is still in use. An example of the problem showed up in the Hasher backend, which uses the Shutdown method to stop the bolt db used to store hashes. If a command was run on a Hasher file (ex. `rclone md5sum --download hasher:somelargefile.zip`) and hashing the file took longer than the --fs-cache-expire-duration (5m by default), the bolt db was stopped before the hashing operation completed, resulting in an error. This change fixes the issue by ensuring that: 1. only one entry is added to the cache (the file's parent, not the file). 2. future lookups correctly find the entry regardless of whether they are called with the parent name or one of its children. 3. fs.ErrorIsFile is returned when (and only when) fsString points to a file (preserving the fix from https://github.com/rclone/rclone/commit/8d5bc7f28b8459a0e04fdbfd2dca43c5d9ec4555). Note that f.Root() should always point to the parent dir as of https://github.com/rclone/rclone/commit/c69eb84573c85206ab028eda2987180e049ef2e4
2024-03-10 12:36:57 +08:00
if err == nil { // it's a dir
fs.Debugf(nil, "fs cache: renaming cache item %q to be canonical %q", canonicalFsString, canonicalName)
value, found := c.Rename(canonicalFsString, canonicalName)
if found {
f = value.(fs.Fs)
}
addMapping(canonicalFsString, canonicalName)
fs/cache: fix parent not getting pinned when remote is a file Before this change, when cache.GetFn was called on a file rather than a directory, two cache entries would be added (the file + its parent) but only one of them would get pinned if the caller then called Pin(f). This left the other one exposed to expiration if the ci.FsCacheExpireDuration was reached. This was problematic because both entries point to the same Fs, and if one entry expires while the other is pinned, the Shutdown method gets erroneously called on an Fs that is still in use. An example of the problem showed up in the Hasher backend, which uses the Shutdown method to stop the bolt db used to store hashes. If a command was run on a Hasher file (ex. `rclone md5sum --download hasher:somelargefile.zip`) and hashing the file took longer than the --fs-cache-expire-duration (5m by default), the bolt db was stopped before the hashing operation completed, resulting in an error. This change fixes the issue by ensuring that: 1. only one entry is added to the cache (the file's parent, not the file). 2. future lookups correctly find the entry regardless of whether they are called with the parent name or one of its children. 3. fs.ErrorIsFile is returned when (and only when) fsString points to a file (preserving the fix from https://github.com/rclone/rclone/commit/8d5bc7f28b8459a0e04fdbfd2dca43c5d9ec4555). Note that f.Root() should always point to the parent dir as of https://github.com/rclone/rclone/commit/c69eb84573c85206ab028eda2987180e049ef2e4
2024-03-10 12:36:57 +08:00
} else { // it's a file
// the fs we cache is always the file's parent, never the file,
// but we use the childParentMap to return the correct error status based on the fsString passed in.
fs.Debugf(nil, "fs cache: renaming child cache item %q to be canonical for parent %q", canonicalFsString, canonicalName)
value, found := c.Rename(canonicalFsString, canonicalName) // rename the file entry to parent
if found {
f = value.(fs.Fs) // if parent already exists, use it
}
Put(canonicalName, f) // force err == nil for the cache
addMapping(canonicalFsString, canonicalName) // note the fsString-canonicalName connection for future lookups
addChild(fsString, canonicalName) // note the file-directory connection for future lookups
}
}
}
fs/cache: fix parent not getting pinned when remote is a file Before this change, when cache.GetFn was called on a file rather than a directory, two cache entries would be added (the file + its parent) but only one of them would get pinned if the caller then called Pin(f). This left the other one exposed to expiration if the ci.FsCacheExpireDuration was reached. This was problematic because both entries point to the same Fs, and if one entry expires while the other is pinned, the Shutdown method gets erroneously called on an Fs that is still in use. An example of the problem showed up in the Hasher backend, which uses the Shutdown method to stop the bolt db used to store hashes. If a command was run on a Hasher file (ex. `rclone md5sum --download hasher:somelargefile.zip`) and hashing the file took longer than the --fs-cache-expire-duration (5m by default), the bolt db was stopped before the hashing operation completed, resulting in an error. This change fixes the issue by ensuring that: 1. only one entry is added to the cache (the file's parent, not the file). 2. future lookups correctly find the entry regardless of whether they are called with the parent name or one of its children. 3. fs.ErrorIsFile is returned when (and only when) fsString points to a file (preserving the fix from https://github.com/rclone/rclone/commit/8d5bc7f28b8459a0e04fdbfd2dca43c5d9ec4555). Note that f.Root() should always point to the parent dir as of https://github.com/rclone/rclone/commit/c69eb84573c85206ab028eda2987180e049ef2e4
2024-03-10 12:36:57 +08:00
return f, getError(fsString, err) // ensure fs.ErrorIsFile is returned when necessary
}
// Pin f into the cache until Unpin is called
func Pin(f fs.Fs) {
createOnFirstUse()
c.Pin(fs.ConfigString(f))
}
// PinUntilFinalized pins f into the cache until x is garbage collected
//
// This calls runtime.SetFinalizer on x so it shouldn't have a
// finalizer already.
func PinUntilFinalized(f fs.Fs, x interface{}) {
Pin(f)
runtime.SetFinalizer(x, func(_ interface{}) {
Unpin(f)
})
}
// Unpin f from the cache
func Unpin(f fs.Fs) {
createOnFirstUse()
2022-06-20 19:14:58 +08:00
c.Unpin(fs.ConfigString(f))
}
// To avoid circular dependencies these are filled in by fs/rc/jobs/job.go
var (
// JobGetJobID for internal use only
JobGetJobID func(context.Context) (int64, bool)
// JobOnFinish for internal use only
JobOnFinish func(int64, func()) (func(), error)
)
// Get gets an fs.Fs named fsString either from the cache or creates it afresh
func Get(ctx context.Context, fsString string) (f fs.Fs, err error) {
// If we are making a long lived backend which lives longer
// than this request, we want to disconnect it from the
// current context and in particular any WithCancel contexts,
// but we want to preserve the config embedded in the context.
newCtx := context.Background()
newCtx = fs.CopyConfig(newCtx, ctx)
newCtx = filter.CopyConfig(newCtx, ctx)
f, err = GetFn(newCtx, fsString, fs.NewFs)
if f == nil || (err != nil && err != fs.ErrorIsFile) {
return f, err
}
// If this is part of an rc job then pin the backend until it finishes
if JobOnFinish != nil && JobGetJobID != nil {
if jobID, ok := JobGetJobID(ctx); ok {
// fs.Debugf(f, "Pin for job %d", jobID)
Pin(f)
_, _ = JobOnFinish(jobID, func() {
// fs.Debugf(f, "Unpin for job %d", jobID)
Unpin(f)
})
}
}
return f, err
}
// GetArr gets []fs.Fs from []fsStrings either from the cache or creates it afresh
func GetArr(ctx context.Context, fsStrings []string) (f []fs.Fs, err error) {
var fArr []fs.Fs
for _, fsString := range fsStrings {
f1, err1 := GetFn(ctx, fsString, fs.NewFs)
if err1 != nil {
return fArr, err1
}
fArr = append(fArr, f1)
}
return fArr, nil
}
// PutErr puts an fs.Fs named fsString into the cache with err
func PutErr(fsString string, f fs.Fs, err error) {
createOnFirstUse()
canonicalName := fs.ConfigString(f)
c.PutErr(canonicalName, f, err)
addMapping(fsString, canonicalName)
fs/cache: fix parent not getting pinned when remote is a file Before this change, when cache.GetFn was called on a file rather than a directory, two cache entries would be added (the file + its parent) but only one of them would get pinned if the caller then called Pin(f). This left the other one exposed to expiration if the ci.FsCacheExpireDuration was reached. This was problematic because both entries point to the same Fs, and if one entry expires while the other is pinned, the Shutdown method gets erroneously called on an Fs that is still in use. An example of the problem showed up in the Hasher backend, which uses the Shutdown method to stop the bolt db used to store hashes. If a command was run on a Hasher file (ex. `rclone md5sum --download hasher:somelargefile.zip`) and hashing the file took longer than the --fs-cache-expire-duration (5m by default), the bolt db was stopped before the hashing operation completed, resulting in an error. This change fixes the issue by ensuring that: 1. only one entry is added to the cache (the file's parent, not the file). 2. future lookups correctly find the entry regardless of whether they are called with the parent name or one of its children. 3. fs.ErrorIsFile is returned when (and only when) fsString points to a file (preserving the fix from https://github.com/rclone/rclone/commit/8d5bc7f28b8459a0e04fdbfd2dca43c5d9ec4555). Note that f.Root() should always point to the parent dir as of https://github.com/rclone/rclone/commit/c69eb84573c85206ab028eda2987180e049ef2e4
2024-03-10 12:36:57 +08:00
if err == fs.ErrorIsFile {
addChild(fsString, canonicalName)
}
}
// Put puts an fs.Fs named fsString into the cache
func Put(fsString string, f fs.Fs) {
PutErr(fsString, f, nil)
}
// ClearConfig deletes all entries which were based on the config name passed in
//
// Returns number of entries deleted
func ClearConfig(name string) (deleted int) {
createOnFirstUse()
fs/cache: fix parent not getting pinned when remote is a file Before this change, when cache.GetFn was called on a file rather than a directory, two cache entries would be added (the file + its parent) but only one of them would get pinned if the caller then called Pin(f). This left the other one exposed to expiration if the ci.FsCacheExpireDuration was reached. This was problematic because both entries point to the same Fs, and if one entry expires while the other is pinned, the Shutdown method gets erroneously called on an Fs that is still in use. An example of the problem showed up in the Hasher backend, which uses the Shutdown method to stop the bolt db used to store hashes. If a command was run on a Hasher file (ex. `rclone md5sum --download hasher:somelargefile.zip`) and hashing the file took longer than the --fs-cache-expire-duration (5m by default), the bolt db was stopped before the hashing operation completed, resulting in an error. This change fixes the issue by ensuring that: 1. only one entry is added to the cache (the file's parent, not the file). 2. future lookups correctly find the entry regardless of whether they are called with the parent name or one of its children. 3. fs.ErrorIsFile is returned when (and only when) fsString points to a file (preserving the fix from https://github.com/rclone/rclone/commit/8d5bc7f28b8459a0e04fdbfd2dca43c5d9ec4555). Note that f.Root() should always point to the parent dir as of https://github.com/rclone/rclone/commit/c69eb84573c85206ab028eda2987180e049ef2e4
2024-03-10 12:36:57 +08:00
ClearMappingsPrefix(name)
return c.DeletePrefix(name + ":")
}
// Clear removes everything from the cache
func Clear() {
createOnFirstUse()
2019-08-01 05:19:07 +08:00
c.Clear()
fs/cache: fix parent not getting pinned when remote is a file Before this change, when cache.GetFn was called on a file rather than a directory, two cache entries would be added (the file + its parent) but only one of them would get pinned if the caller then called Pin(f). This left the other one exposed to expiration if the ci.FsCacheExpireDuration was reached. This was problematic because both entries point to the same Fs, and if one entry expires while the other is pinned, the Shutdown method gets erroneously called on an Fs that is still in use. An example of the problem showed up in the Hasher backend, which uses the Shutdown method to stop the bolt db used to store hashes. If a command was run on a Hasher file (ex. `rclone md5sum --download hasher:somelargefile.zip`) and hashing the file took longer than the --fs-cache-expire-duration (5m by default), the bolt db was stopped before the hashing operation completed, resulting in an error. This change fixes the issue by ensuring that: 1. only one entry is added to the cache (the file's parent, not the file). 2. future lookups correctly find the entry regardless of whether they are called with the parent name or one of its children. 3. fs.ErrorIsFile is returned when (and only when) fsString points to a file (preserving the fix from https://github.com/rclone/rclone/commit/8d5bc7f28b8459a0e04fdbfd2dca43c5d9ec4555). Note that f.Root() should always point to the parent dir as of https://github.com/rclone/rclone/commit/c69eb84573c85206ab028eda2987180e049ef2e4
2024-03-10 12:36:57 +08:00
ClearMappings()
}
// Entries returns the number of entries in the cache
func Entries() int {
createOnFirstUse()
return c.Entries()
}
fs/cache: fix parent not getting pinned when remote is a file Before this change, when cache.GetFn was called on a file rather than a directory, two cache entries would be added (the file + its parent) but only one of them would get pinned if the caller then called Pin(f). This left the other one exposed to expiration if the ci.FsCacheExpireDuration was reached. This was problematic because both entries point to the same Fs, and if one entry expires while the other is pinned, the Shutdown method gets erroneously called on an Fs that is still in use. An example of the problem showed up in the Hasher backend, which uses the Shutdown method to stop the bolt db used to store hashes. If a command was run on a Hasher file (ex. `rclone md5sum --download hasher:somelargefile.zip`) and hashing the file took longer than the --fs-cache-expire-duration (5m by default), the bolt db was stopped before the hashing operation completed, resulting in an error. This change fixes the issue by ensuring that: 1. only one entry is added to the cache (the file's parent, not the file). 2. future lookups correctly find the entry regardless of whether they are called with the parent name or one of its children. 3. fs.ErrorIsFile is returned when (and only when) fsString points to a file (preserving the fix from https://github.com/rclone/rclone/commit/8d5bc7f28b8459a0e04fdbfd2dca43c5d9ec4555). Note that f.Root() should always point to the parent dir as of https://github.com/rclone/rclone/commit/c69eb84573c85206ab028eda2987180e049ef2e4
2024-03-10 12:36:57 +08:00
// ClearMappings removes everything from remap and childParentMap
func ClearMappings() {
mu.Lock()
defer mu.Unlock()
remap = map[string]string{}
childParentMap = map[string]string{}
}
// ClearMappingsPrefix deletes all mappings to parents with given prefix
//
// Returns number of entries deleted
func ClearMappingsPrefix(prefix string) (deleted int) {
mu.Lock()
do := func(mapping map[string]string) {
for key, val := range mapping {
if !strings.HasPrefix(val, prefix) {
continue
}
delete(mapping, key)
deleted++
}
}
do(remap)
do(childParentMap)
mu.Unlock()
return deleted
}
// EntriesWithPinCount returns the number of pinned and unpinned entries in the cache
//
// Each entry is counted only once, regardless of entry.pinCount
func EntriesWithPinCount() (pinned, unpinned int) {
createOnFirstUse()
return c.EntriesWithPinCount()
}