mirror of
https://github.com/xuxiaobo-bobo/boda_jsEnv.git
synced 2025-04-20 03:19:56 +08:00
189 lines
7.8 KiB
JavaScript
189 lines
7.8 KiB
JavaScript
"use strict";
|
||
Object.defineProperty(exports, "__esModule", { value: true });
|
||
exports.parseMLSxDate = exports.transformList = exports.parseLine = exports.testLine = void 0;
|
||
const FileInfo_1 = require("./FileInfo");
|
||
function parseSize(value, info) {
|
||
info.size = parseInt(value, 10);
|
||
}
|
||
/**
|
||
* Parsers for MLSD facts.
|
||
*/
|
||
const factHandlersByName = {
|
||
"size": parseSize, // File size
|
||
"sizd": parseSize, // Directory size
|
||
"unique": (value, info) => {
|
||
info.uniqueID = value;
|
||
},
|
||
"modify": (value, info) => {
|
||
info.modifiedAt = parseMLSxDate(value);
|
||
info.rawModifiedAt = info.modifiedAt.toISOString();
|
||
},
|
||
"type": (value, info) => {
|
||
// There seems to be confusion on how to handle symbolic links for Unix. RFC 3659 doesn't describe
|
||
// this but mentions some examples using the syntax `type=OS.unix=slink:<target>`. But according to
|
||
// an entry in the Errata (https://www.rfc-editor.org/errata/eid1500) this syntax can't be valid.
|
||
// Instead it proposes to use `type=OS.unix=symlink` and to then list the actual target of the
|
||
// symbolic link as another entry in the directory listing. The unique identifiers can then be used
|
||
// to derive the connection between link(s) and target. We'll have to handle both cases as there
|
||
// are differing opinions on how to deal with this. Here are some links on this topic:
|
||
// - ProFTPD source: https://github.com/proftpd/proftpd/blob/56e6dfa598cbd4ef5c6cba439bcbcd53a63e3b21/modules/mod_facts.c#L531
|
||
// - ProFTPD bug: http://bugs.proftpd.org/show_bug.cgi?id=3318
|
||
// - ProFTPD statement: http://www.proftpd.org/docs/modules/mod_facts.html
|
||
// – FileZilla bug: https://trac.filezilla-project.org/ticket/9310
|
||
if (value.startsWith("OS.unix=slink")) {
|
||
info.type = FileInfo_1.FileType.SymbolicLink;
|
||
info.link = value.substr(value.indexOf(":") + 1);
|
||
return 1 /* FactHandlerResult.Continue */;
|
||
}
|
||
switch (value) {
|
||
case "file":
|
||
info.type = FileInfo_1.FileType.File;
|
||
break;
|
||
case "dir":
|
||
info.type = FileInfo_1.FileType.Directory;
|
||
break;
|
||
case "OS.unix=symlink":
|
||
info.type = FileInfo_1.FileType.SymbolicLink;
|
||
// The target of the symbolic link might be defined in another line in the directory listing.
|
||
// We'll handle this in `transformList()` below.
|
||
break;
|
||
case "cdir": // Current directory being listed
|
||
case "pdir": // Parent directory
|
||
return 2 /* FactHandlerResult.IgnoreFile */; // Don't include these entries in the listing
|
||
default:
|
||
info.type = FileInfo_1.FileType.Unknown;
|
||
}
|
||
return 1 /* FactHandlerResult.Continue */;
|
||
},
|
||
"unix.mode": (value, info) => {
|
||
const digits = value.substr(-3);
|
||
info.permissions = {
|
||
user: parseInt(digits[0], 10),
|
||
group: parseInt(digits[1], 10),
|
||
world: parseInt(digits[2], 10)
|
||
};
|
||
},
|
||
"unix.ownername": (value, info) => {
|
||
info.user = value;
|
||
},
|
||
"unix.owner": (value, info) => {
|
||
if (info.user === undefined)
|
||
info.user = value;
|
||
},
|
||
get "unix.uid"() {
|
||
return this["unix.owner"];
|
||
},
|
||
"unix.groupname": (value, info) => {
|
||
info.group = value;
|
||
},
|
||
"unix.group": (value, info) => {
|
||
if (info.group === undefined)
|
||
info.group = value;
|
||
},
|
||
get "unix.gid"() {
|
||
return this["unix.group"];
|
||
}
|
||
// Regarding the fact "perm":
|
||
// We don't handle permission information stored in "perm" because its information is conceptually
|
||
// different from what users of FTP clients usually associate with "permissions". Those that have
|
||
// some expectations (and probably want to edit them with a SITE command) often unknowingly expect
|
||
// the Unix permission system. The information passed by "perm" describes what FTP commands can be
|
||
// executed with a file/directory. But even this can be either incomplete or just meant as a "guide"
|
||
// as the spec mentions. From https://tools.ietf.org/html/rfc3659#section-7.5.5: "The permissions are
|
||
// described here as they apply to FTP commands. They may not map easily into particular permissions
|
||
// available on the server's operating system." The parser by Apache Commons tries to translate these
|
||
// to Unix permissions – this is misleading users and might not even be correct.
|
||
};
|
||
/**
|
||
* Split a string once at the first position of a delimiter. For example
|
||
* `splitStringOnce("a b c d", " ")` returns `["a", "b c d"]`.
|
||
*/
|
||
function splitStringOnce(str, delimiter) {
|
||
const pos = str.indexOf(delimiter);
|
||
const a = str.substr(0, pos);
|
||
const b = str.substr(pos + delimiter.length);
|
||
return [a, b];
|
||
}
|
||
/**
|
||
* Returns true if a given line might be part of an MLSD listing.
|
||
*
|
||
* - Example 1: `size=15227;type=dir;perm=el;modify=20190419065730; test one`
|
||
* - Example 2: ` file name` (leading space)
|
||
*/
|
||
function testLine(line) {
|
||
return /^\S+=\S+;/.test(line) || line.startsWith(" ");
|
||
}
|
||
exports.testLine = testLine;
|
||
/**
|
||
* Parse single line as MLSD listing, see specification at https://tools.ietf.org/html/rfc3659#section-7.
|
||
*/
|
||
function parseLine(line) {
|
||
const [packedFacts, name] = splitStringOnce(line, " ");
|
||
if (name === "" || name === "." || name === "..") {
|
||
return undefined;
|
||
}
|
||
const info = new FileInfo_1.FileInfo(name);
|
||
const facts = packedFacts.split(";");
|
||
for (const fact of facts) {
|
||
const [factName, factValue] = splitStringOnce(fact, "=");
|
||
if (!factValue) {
|
||
continue;
|
||
}
|
||
const factHandler = factHandlersByName[factName.toLowerCase()];
|
||
if (!factHandler) {
|
||
continue;
|
||
}
|
||
const result = factHandler(factValue, info);
|
||
if (result === 2 /* FactHandlerResult.IgnoreFile */) {
|
||
return undefined;
|
||
}
|
||
}
|
||
return info;
|
||
}
|
||
exports.parseLine = parseLine;
|
||
function transformList(files) {
|
||
// Create a map of all files that are not symbolic links by their unique ID
|
||
const nonLinksByID = new Map();
|
||
for (const file of files) {
|
||
if (!file.isSymbolicLink && file.uniqueID !== undefined) {
|
||
nonLinksByID.set(file.uniqueID, file);
|
||
}
|
||
}
|
||
const resolvedFiles = [];
|
||
for (const file of files) {
|
||
// Try to associate unresolved symbolic links with a target file/directory.
|
||
if (file.isSymbolicLink && file.uniqueID !== undefined && file.link === undefined) {
|
||
const target = nonLinksByID.get(file.uniqueID);
|
||
if (target !== undefined) {
|
||
file.link = target.name;
|
||
}
|
||
}
|
||
// The target of a symbolic link is listed as an entry in the directory listing but might
|
||
// have a path pointing outside of this directory. In that case we don't want this entry
|
||
// to be part of the listing. We generally don't want these kind of entries at all.
|
||
const isPartOfDirectory = !file.name.includes("/");
|
||
if (isPartOfDirectory) {
|
||
resolvedFiles.push(file);
|
||
}
|
||
}
|
||
return resolvedFiles;
|
||
}
|
||
exports.transformList = transformList;
|
||
/**
|
||
* Parse date as specified in https://tools.ietf.org/html/rfc3659#section-2.3.
|
||
*
|
||
* Message contains response code and modified time in the format: YYYYMMDDHHMMSS[.sss]
|
||
* For example `19991005213102` or `19980615100045.014`.
|
||
*/
|
||
function parseMLSxDate(fact) {
|
||
return new Date(Date.UTC(+fact.slice(0, 4), // Year
|
||
+fact.slice(4, 6) - 1, // Month
|
||
+fact.slice(6, 8), // Date
|
||
+fact.slice(8, 10), // Hours
|
||
+fact.slice(10, 12), // Minutes
|
||
+fact.slice(12, 14), // Seconds
|
||
+fact.slice(15, 18) // Milliseconds
|
||
));
|
||
}
|
||
exports.parseMLSxDate = parseMLSxDate;
|