* xfa, xfdf: refuse XML nested past 1024 levels (roxmltree recurses with nesting depth)
roxmltree parses by recursive descent, so its stack use grows with the nesting depth of the
document it is handed. Ten thousand nested elements overflow the main thread's stack and abort
the process ("fatal runtime error: stack overflow, aborting", SIGABRT, core dumped); three
thousand still parse on the default stack. A stack overflow is `abort()`, so `catch_unwind`
cannot intercept it and the whole application dies.
Two of our XML entry points hand attacker-supplied text straight to it:
- XFDF/FDF form data import. A 1.4 MB file with 50,000 nested `<field>` elements aborted
`pdfcraft-cli run doc_import_data` in 288 ms (SIGABRT, core dumped). The same file now returns
"the file nests XML elements 50003 deep, past the 1024 the reader accepts" in 54 ms, at 19 MB.
- The XFA template parser (`crates/xfa/src/parse.rs`), reached when a document with an XFA form
is opened or laid out. A 10,000-deep template aborted this crate's own test run before the fix.
`crates/xfa/src/data.rs` parses the datasets packet and the data written back into it, so it is
guarded the same way.
The bound has to be applied before parsing, not while walking the result: the overflow happens
inside the parser, so a depth cap on our own walks (64, unchanged) cannot prevent it. Both
crates measure the nesting with a small scanner that skips comments, CDATA, processing
instructions, the DOCTYPE and quoted attribute values, and counts a tag that closes itself
(`<draw/>`) as one level while it is read rather than as a level left open, so a template full of
styling elements is two levels deep rather than two thousand. It cannot under-count a real
element; a stray `<` in text is invalid XML and only makes it over-count, so it errs towards
refusing a file rather than trusting it.
The limit is 1024 because of measurement, not taste: 3000 levels still parse on the default stack
and 10000 abort, and `hostile_templates_fail_without_panicking` keeps asserting that a 500-deep
template parses, so the bound has to sit well above that.
Tests: `deeply_nested_fields_are_refused_instead_of_overflowing_the_stack` and
`deeply_nested_template_is_refused_instead_of_overflowing_the_stack` both abort with a stack
overflow before this change and pass after it; `nesting_depth_scanner_counts_what_matters` and
`self_closing_elements_do_not_accumulate_in_the_depth` cover the scanner (comments, CDATA, a
quoted attribute holding `<`, and 2,000 sibling `<draw/>` elements, which a scanner that leaves
self-closing tags open counts as 2,000 deep);
`many_self_closing_elements_do_not_accumulate_into_the_bound` parses a template holding two
thousand of them; and `nested_fields_within_the_limit_still_read` proves a 16-level field tree
still reads with the same dotted names. Both suites pass: xfa 36, xfdf 18.
The parse options also carry `nodes_limit` now, which was already the shape the XFA parser used:
a 93 MB XFDF with two million sibling elements went from 1082 MB and 2323 ms to 289 MB and
1299 ms ("nodes limit reached").
The scanner is duplicated between the two crates because nothing between them owns XML. If it
should live in one place, say where and it moves.
* xfa: the depth limit holds on a default-stack worker thread
---------
Co-authored-by: Brandon Thomas <echelon@gmail.com>
pdfcraft-xfa
XFA forms (XFA 3.3, the template and layout parts). Layer L3; depends on pdfcraft-cos and
pdfcraft-fonts only.
A dynamic XFA form is a PDF shell around an XML template (/AcroForm /XFA): one placeholder page
("requires Adobe Reader") and /NeedsRendering true. Viewers without an XFA engine show the
placeholder. This crate reads the template, lays it out, and writes ordinary pages and AcroForm
fields into the document, so everything else in PdfCraft (rendering, filling, saving, printing,
the automation tools) works on it unchanged. The engine does this when it opens such a form.
What it does
-
Packets (
read_packets): the XDP from one stream or the(name, stream)array; UTF-8 and UTF-16; capped at 64 MB. -
Template (
parse): subforms, areas, fields, draws, exclusion groups, page sets with page areas, content areas and media; measurements in mm, cm, in, pt and px; fonts, paragraphs, margins, borders (edge order rules), captions with reserves, check-button items, pictures, rich text (exDataXHTML: paragraphs,br, bold/italic/underline/size spans, embedded fields), inline JPEG, PNG and GIF images, lines, rectangles, occurrences,breakBefore/breakAfter, overflow leaders,columnWidthsandcolSpan. -
Layout (
layout): positioned,tb,lr-tb/rl-tb,tableandrowlayouts; a container without a width is as wide as its content; flow across content areas and pages with page masters chosen by occurrence; a table's header row repeats after a break; presencehiddentakes no space,invisibletakes space; repeating subforms get their initial instances, in flowed content and in a top-to-bottom container inside positioned content (a table in an area) alike; fields that compute the page number or count on layout show the numbers. -
PDF (
pdf::write_form): pages with content streams in the standard 14 fonts (Arial and friends → Helvetica, Courier New → Courier, serif faces → Times; nothing is embedded), JPEG XObjects, and widgets: text (multi-line, max length, alignment), date (Acrobat'sAFDateformat and keystroke actions from the picture clause), check boxes and radio groups (fromexclGroup), push buttons with Reset / Print / Save As / URI actions recognised from their scripts. Appearance streams are left empty for the forms layer to generate. Each field keeps its SOM path in/PCSom; the AcroForm gets/PCXfaLayoutso a saved form is not laid out twice. The XFA packets stay, so Adobe's viewers keep rendering the form from them. -
Data (
data): thedatasetspacket. On layout, a field takes its value from the data node at its data path: its SOM path without areas, which are named in SOM expressions but are not data scopes (data_path, as Adobe's viewers bind) (dates in ISO form, check and radio states by their on values), and a repeating subform or row gets as many instances as the data has.write_datasetsmerges the AcroForm fields' values (by/PCSom, or by the Designer field names of a static form) into the existingxfa:data: only the bound nodes' text changes and missing nodes are added, so unbound data, other namespaces, attributes and comments stay byte for byte. Values it can't write (a node holding structured content, absurd SOM indices, a packet that is neither UTF-8 nor UTF-16) come back as warnings; UTF-16 packets are written back as UTF-16.read_valuesgoes the other way, so a form filled here shows its values in Adobe's viewers and a form filled there shows them here. The engine does both: values on open (recorded in the document's warnings), the packet after every form edit.
Layout measures each container once per (node, width): width-less containers would otherwise be measured exponentially often in their nesting. A measurement budget backs this up.
-
Scripts (
script): the live form as scripts see it (form_tree: the instance tree with values, presence and access, and theScriptEvents in document order, built withinMAX_FORM_NODESobjects and cut short, withtruncatedset, past that;has_scriptssays whether a template has any, and the engine skips all of this for forms without), the presence and access overrides scripts make, rows added and removed in the data (DataOps, applied to the in-memoryDataNodewhile an event runs and written bywrite_data_opsas one rewrite of the datasets packet per event), andrerender, which lays the form out again from template, data and overrides. The scripts themselves run in thejscrate's XFA object model; the engine runs initialize and calculate on open, change, exit, validate and calculate after a field changes, and click for buttons.Private keys PdfCraft writes (other viewers ignore them):
Key Where Holds /PCSomgenerated fields the field's SOM path, so values go back to the data /PCXfaLayoutAcroForm what laying out produced, so a saved form is not laid out again /PCXfaOverridesAcroForm << /Presence << /som (hidden) … >> /Access << /som (readOnly) … >> >>: presence and access scripts set, by SOM path (instance 0 of a repeating subform stands for all); applied to the template before every layout, so undo, save and reopen keep them. Adobe's viewers draw from the XFA packets and don't read it; at most 10 000 entries are read or written/PCXfaClickgenerated push buttons true: the template has a click script for this button. Such buttons carry no/A: XFA JavaScript is not Acrobat JavaScript, so it is never written as a/S /JavaScriptaction other viewers would run. PdfCraft finds the script by the field's/PCSom
API sketch
if pdfcraft_xfa::is_dynamic(&doc) {
let report = pdfcraft_xfa::render_into(&mut doc)?; // pages, fields, warnings
for f in pdfcraft_forms::fields(&doc) { pdfcraft_forms::redraw_field(&mut doc, &f.name)?; }
}
let form = pdfcraft_xfa::layout_xml(template_xml)?; // pages of items, for tests
Deliberately not done (yet)
- Data binding is the default one only: explicit
bind refexpressions, global binding and data descriptions are not followed, and no standalone XML or XDP data file is imported or exported. - Scripting. JavaScript (boa) and FormCalc (a native interpreter,
pdfcraft_js::formcalc) share one object model, one set of effects and one set of budgets. FormCalc covers the language and the common built-ins (arithmetic, logical, string, date/time, financial, unit); not locale-aware pictures beyond simple number and date ones, orGet/Post/Put(refused). The object model covers what forms commonly use (xfa.formnavigation and SOM resolution,rawValue,presence,access, instance managers,xfa.hostmessages, reset, print, focus and URLs,xfa.layoutpage numbers,xfa.event); notxfa.template, data descriptions,xfa.connectionSet,border/font/uiproperties (reads give a sink that accepts writes),execEvent, or the change event per keystroke. Validate failures show their message and keep the value, as Acrobat does for scripts. A hidden instance of a repeating subform hides every instance (overrides apply to the template). A repeating subform shows what its data holds once the data holds any instance, else itsinitialcount, and never fewer than one (occur min="0"still shows one row). Each script runs in a fresh engine on its own thread; a form with hundreds of calculate scripts pays that on every field change (a shared engine per event is the obvious next step). - Script limits. Forms run their scripts on open without being asked, so the engine bounds each event: a script's effects (values set again on the same object merge, the last wins; 10 000 effects, 100 message boxes, 1 000 console lines per script and per event), its loop iterations (100 000 per call frame for initialize, calculate and validate, 1 000 000 for click and change), 2 000 scripts and 50 relayouts per event, and 3 s (open, changes) or 5 s (click) of scripts per event, after which the rest are skipped and reported. Only a click may print, save, open a link or move the focus; other events asking are noted in the console. What open-time scripts change is noted in the document's XFA warnings. Known gaps: the engine's loop limit is per call frame, so loops inside nested calls can run far longer; such a script is abandoned after 1 s (3 s for a click) on its thread, which keeps running until the engine's own limits end it, and the document's scripts are turned off. Memory a script allocates is not capped (the same holds for AcroForm JavaScript).
- Static XFA forms (
/NeedsRenderingabsent, AcroForm fields present) keep their AcroForm; only their data is read and written. - Field kinds. Choice lists become combo boxes (editable with
textEntry) or list boxes (open="always", multi-select withopen="multiSelect"), with the shown items and the saved values of<items save="1">as the options; the data holds saved values, several on separate lines. Password fields are password text fields. Signature fields are/Sigfields the app can sign. Image fields show the picture the data or template holds (JPEG, PNG, GIF) but a new picture can't be chosen; barcode fields show their value as text, not as bars. - Missing:
keepconstraints,subformSetrelations,rl-tbis mirroredlr-tb, font metrics are the approximate Helvetica ones, and line heights are 1.15 × size. - Acrobat cannot be an oracle here (clean-room rules): layout follows the specification, and fidelity has been checked by eye on real government forms, not pixel-compared.