147 lines
3.5 KiB
Go
147 lines
3.5 KiB
Go
package download
|
|
|
|
import (
|
|
"database/sql"
|
|
"os"
|
|
"sort"
|
|
"strings"
|
|
"time"
|
|
|
|
"github.com/tidwall/gjson"
|
|
_ "modernc.org/sqlite" // import sqlite3 driver
|
|
|
|
"github.com/moond4rk/hackbrowserdata/extractor"
|
|
"github.com/moond4rk/hackbrowserdata/log"
|
|
"github.com/moond4rk/hackbrowserdata/types"
|
|
"github.com/moond4rk/hackbrowserdata/utils/typeutil"
|
|
)
|
|
|
|
func init() {
|
|
extractor.RegisterExtractor(types.ChromiumDownload, func() extractor.Extractor {
|
|
return new(ChromiumDownload)
|
|
})
|
|
extractor.RegisterExtractor(types.FirefoxDownload, func() extractor.Extractor {
|
|
return new(FirefoxDownload)
|
|
})
|
|
}
|
|
|
|
type ChromiumDownload []download
|
|
|
|
type download struct {
|
|
TargetPath string
|
|
URL string
|
|
TotalBytes int64
|
|
StartTime time.Time
|
|
EndTime time.Time
|
|
MimeType string
|
|
}
|
|
|
|
const (
|
|
queryChromiumDownload = `SELECT target_path, tab_url, total_bytes, start_time, end_time, mime_type FROM downloads`
|
|
)
|
|
|
|
func (c *ChromiumDownload) Extract(_ []byte) error {
|
|
db, err := sql.Open("sqlite", types.ChromiumDownload.TempFilename())
|
|
if err != nil {
|
|
return err
|
|
}
|
|
defer os.Remove(types.ChromiumDownload.TempFilename())
|
|
defer db.Close()
|
|
rows, err := db.Query(queryChromiumDownload)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
defer rows.Close()
|
|
for rows.Next() {
|
|
var (
|
|
targetPath, tabURL, mimeType string
|
|
totalBytes, startTime, endTime int64
|
|
)
|
|
if err := rows.Scan(&targetPath, &tabURL, &totalBytes, &startTime, &endTime, &mimeType); err != nil {
|
|
log.Warnf("scan chromium download error: %v", err)
|
|
}
|
|
data := download{
|
|
TargetPath: targetPath,
|
|
URL: tabURL,
|
|
TotalBytes: totalBytes,
|
|
StartTime: typeutil.TimeEpoch(startTime),
|
|
EndTime: typeutil.TimeEpoch(endTime),
|
|
MimeType: mimeType,
|
|
}
|
|
*c = append(*c, data)
|
|
}
|
|
sort.Slice(*c, func(i, j int) bool {
|
|
return (*c)[i].TotalBytes > (*c)[j].TotalBytes
|
|
})
|
|
return nil
|
|
}
|
|
|
|
func (c *ChromiumDownload) Name() string {
|
|
return "download"
|
|
}
|
|
|
|
func (c *ChromiumDownload) Len() int {
|
|
return len(*c)
|
|
}
|
|
|
|
type FirefoxDownload []download
|
|
|
|
const (
|
|
queryFirefoxDownload = `SELECT place_id, GROUP_CONCAT(content), url, dateAdded FROM (SELECT * FROM moz_annos INNER JOIN moz_places ON moz_annos.place_id=moz_places.id) t GROUP BY place_id`
|
|
closeJournalMode = `PRAGMA journal_mode=off`
|
|
)
|
|
|
|
func (f *FirefoxDownload) Extract(_ []byte) error {
|
|
db, err := sql.Open("sqlite", types.FirefoxDownload.TempFilename())
|
|
if err != nil {
|
|
return err
|
|
}
|
|
defer os.Remove(types.FirefoxDownload.TempFilename())
|
|
defer db.Close()
|
|
|
|
_, err = db.Exec(closeJournalMode)
|
|
if err != nil {
|
|
log.Errorf("close journal mode error: %v", err)
|
|
}
|
|
rows, err := db.Query(queryFirefoxDownload)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
defer rows.Close()
|
|
for rows.Next() {
|
|
var (
|
|
content, url string
|
|
placeID, dateAdded int64
|
|
)
|
|
if err = rows.Scan(&placeID, &content, &url, &dateAdded); err != nil {
|
|
log.Warnf("scan firefox download error: %v", err)
|
|
}
|
|
contentList := strings.Split(content, ",{")
|
|
if len(contentList) > 1 {
|
|
path := contentList[0]
|
|
json := "{" + contentList[1]
|
|
endTime := gjson.Get(json, "endTime")
|
|
fileSize := gjson.Get(json, "fileSize")
|
|
*f = append(*f, download{
|
|
TargetPath: path,
|
|
URL: url,
|
|
TotalBytes: fileSize.Int(),
|
|
StartTime: typeutil.TimeStamp(dateAdded / 1000000),
|
|
EndTime: typeutil.TimeStamp(endTime.Int() / 1000),
|
|
})
|
|
}
|
|
}
|
|
sort.Slice(*f, func(i, j int) bool {
|
|
return (*f)[i].TotalBytes < (*f)[j].TotalBytes
|
|
})
|
|
return nil
|
|
}
|
|
|
|
func (f *FirefoxDownload) Name() string {
|
|
return "download"
|
|
}
|
|
|
|
func (f *FirefoxDownload) Len() int {
|
|
return len(*f)
|
|
}
|