2 * Copyright © 2007,2008,2009,2010 Red Hat, Inc.
3 * Copyright © 2012,2018 Google, Inc.
5 * This is part of HarfBuzz, a text shaping library.
7 * Permission is hereby granted, without written agreement and without
8 * license or royalty fees, to use, copy, modify, and distribute this
9 * software and its documentation for any purpose, provided that the
10 * above copyright notice and the following two paragraphs appear in
11 * all copies of this software.
13 * IN NO EVENT SHALL THE COPYRIGHT HOLDER BE LIABLE TO ANY PARTY FOR
14 * DIRECT, INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES
15 * ARISING OUT OF THE USE OF THIS SOFTWARE AND ITS DOCUMENTATION, EVEN
16 * IF THE COPYRIGHT HOLDER HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH
19 * THE COPYRIGHT HOLDER SPECIFICALLY DISCLAIMS ANY WARRANTIES, INCLUDING,
20 * BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
21 * FITNESS FOR A PARTICULAR PURPOSE. THE SOFTWARE PROVIDED HEREUNDER IS
22 * ON AN "AS IS" BASIS, AND THE COPYRIGHT HOLDER HAS NO OBLIGATION TO
23 * PROVIDE MAINTENANCE, SUPPORT, UPDATES, ENHANCEMENTS, OR MODIFICATIONS.
25 * Red Hat Author(s): Behdad Esfahbod
26 * Google Author(s): Behdad Esfahbod
29 #ifndef HB_SANITIZE_HH
30 #define HB_SANITIZE_HH
34 #include "hb-dispatch.hh"
41 * === Introduction ===
43 * The sanitize machinery is at the core of our zero-cost font loading. We
44 * mmap() font file into memory and create a blob out of it. Font subtables
45 * are returned as a readonly sub-blob of the main font blob. These table
46 * blobs are then sanitized before use, to ensure invalid memory access does
47 * not happen. The toplevel sanitize API use is like, eg. to load the 'head'
50 * hb_blob_t *head_blob = hb_sanitize_context_t ().reference_table<OT::head> (face);
52 * The blob then can be converted to a head table struct with:
54 * const head *head_table = head_blob->as<head> ();
56 * What the reference_table does is, to call hb_face_reference_table() to load
57 * the table blob, sanitize it and return either the sanitized blob, or empty
58 * blob if sanitization failed. The blob->as() function returns the null
59 * object of its template type argument if the blob is empty. Otherwise, it
60 * just casts the blob contents to the desired type.
62 * Sanitizing a blob of data with a type T works as follows (with minor
65 * - Cast blob content to T*, call sanitize() method of it,
66 * - If sanitize succeeded, return blob.
67 * - Otherwise, if blob is not writable, try making it writable,
68 * or copy if cannot be made writable in-place,
69 * - Call sanitize() again. Return blob if sanitize succeeded.
70 * - Return empty blob otherwise.
73 * === The sanitize() contract ===
75 * The sanitize() method of each object type shall return true if it's safe to
76 * call other methods of the object, and false otherwise.
78 * Note that what sanitize() checks for might align with what the specification
79 * describes as valid table data, but does not have to be. In particular, we
80 * do NOT want to be pedantic and concern ourselves with validity checks that
81 * are irrelevant to our use of the table. On the contrary, we want to be
82 * lenient with error handling and accept invalid data to the extent that it
83 * does not impose extra burden on us.
85 * Based on the sanitize contract, one can see that what we check for depends
86 * on how we use the data in other table methods. Ie. if other table methods
87 * assume that offsets do NOT point out of the table data block, then that's
88 * something sanitize() must check for (GSUB/GPOS/GDEF/etc work this way). On
89 * the other hand, if other methods do such checks themselves, then sanitize()
90 * does not have to bother with them (glyf/local work this way). The choice
91 * depends on the table structure and sanitize() performance. For example, to
92 * check glyf/loca offsets in sanitize() would cost O(num-glyphs). We try hard
93 * to avoid such costs during font loading. By postponing such checks to the
94 * actual glyph loading, we reduce the sanitize cost to O(1) and total runtime
95 * cost to O(used-glyphs). As such, this is preferred.
97 * The same argument can be made re GSUB/GPOS/GDEF, but there, the table
98 * structure is so complicated that by checking all offsets at sanitize() time,
99 * we make the code much simpler in other methods, as offsets and referenced
100 * objects do not need to be validated at each use site.
103 /* This limits sanitizing time on really broken fonts. */
104 #ifndef HB_SANITIZE_MAX_EDITS
105 #define HB_SANITIZE_MAX_EDITS 32
107 #ifndef HB_SANITIZE_MAX_OPS_FACTOR
108 #define HB_SANITIZE_MAX_OPS_FACTOR 8
110 #ifndef HB_SANITIZE_MAX_OPS_MIN
111 #define HB_SANITIZE_MAX_OPS_MIN 16384
113 #ifndef HB_SANITIZE_MAX_OPS_MAX
114 #define HB_SANITIZE_MAX_OPS_MAX 0x3FFFFFFF
117 struct hb_sanitize_context_t :
118 hb_dispatch_context_t<hb_sanitize_context_t, bool, HB_DEBUG_SANITIZE>
120 hb_sanitize_context_t () :
122 start (nullptr), end (nullptr),
124 writable (false), edit_count (0),
127 num_glyphs_set (false) {}
129 const char *get_name () { return "SANITIZE"; }
130 template <typename T, typename F>
131 bool may_dispatch (const T *obj HB_UNUSED, const F *format)
132 { return format->sanitize (this); }
133 static return_t default_return_value () { return true; }
134 static return_t no_dispatch_return_value () { return false; }
135 bool stop_sublookup_iteration (const return_t r) const { return !r; }
138 template <typename T, typename ...Ts> auto
139 _dispatch (const T &obj, hb_priority<1>, Ts&&... ds) HB_AUTO_RETURN
140 ( obj.sanitize (this, hb_forward<Ts> (ds)...) )
141 template <typename T, typename ...Ts> auto
142 _dispatch (const T &obj, hb_priority<0>, Ts&&... ds) HB_AUTO_RETURN
143 ( obj.dispatch (this, hb_forward<Ts> (ds)...) )
145 template <typename T, typename ...Ts> auto
146 dispatch (const T &obj, Ts&&... ds) HB_AUTO_RETURN
147 ( _dispatch (obj, hb_prioritize, hb_forward<Ts> (ds)...) )
150 void init (hb_blob_t *b)
152 this->blob = hb_blob_reference (b);
153 this->writable = false;
156 void set_num_glyphs (unsigned int num_glyphs_)
158 num_glyphs = num_glyphs_;
159 num_glyphs_set = true;
161 unsigned int get_num_glyphs () { return num_glyphs; }
163 void set_max_ops (int max_ops_) { max_ops = max_ops_; }
165 template <typename T>
166 void set_object (const T *obj)
172 const char *obj_start = (const char *) obj;
173 if (unlikely (obj_start < this->start || this->end <= obj_start))
174 this->start = this->end = nullptr;
177 this->start = obj_start;
178 this->end = obj_start + hb_min (size_t (this->end - obj_start), obj->get_size ());
184 this->start = this->blob->data;
185 this->end = this->start + this->blob->length;
186 assert (this->start <= this->end); /* Must not overflow. */
189 void start_processing ()
192 this->max_ops = hb_max ((unsigned int) (this->end - this->start) * HB_SANITIZE_MAX_OPS_FACTOR,
193 (unsigned) HB_SANITIZE_MAX_OPS_MIN);
194 this->edit_count = 0;
195 this->debug_depth = 0;
197 DEBUG_MSG_LEVEL (SANITIZE, start, 0, +1,
198 "start [%p..%p] (%lu bytes)",
199 this->start, this->end,
200 (unsigned long) (this->end - this->start));
203 void end_processing ()
205 DEBUG_MSG_LEVEL (SANITIZE, this->start, 0, -1,
206 "end [%p..%p] %u edit requests",
207 this->start, this->end, this->edit_count);
209 hb_blob_destroy (this->blob);
210 this->blob = nullptr;
211 this->start = this->end = nullptr;
214 unsigned get_edit_count () { return edit_count; }
216 bool check_range (const void *base,
217 unsigned int len) const
219 const char *p = (const char *) base;
223 (unsigned int) (this->end - p) >= len &&
224 this->max_ops-- > 0);
226 DEBUG_MSG_LEVEL (SANITIZE, p, this->debug_depth+1, 0,
227 "check_range [%p..%p]"
228 " (%d bytes) in [%p..%p] -> %s",
230 this->start, this->end,
231 ok ? "OK" : "OUT-OF-RANGE");
236 template <typename T>
237 bool check_range (const T *base,
239 unsigned int b) const
241 return !hb_unsigned_mul_overflows (a, b) &&
242 this->check_range (base, a * b);
245 template <typename T>
246 bool check_range (const T *base,
249 unsigned int c) const
251 return !hb_unsigned_mul_overflows (a, b) &&
252 this->check_range (base, a * b, c);
255 template <typename T>
256 bool check_array (const T *base, unsigned int len) const
258 return this->check_range (base, len, hb_static_size (T));
261 template <typename T>
262 bool check_array (const T *base,
264 unsigned int b) const
266 return this->check_range (base, a, b, hb_static_size (T));
269 template <typename Type>
270 bool check_struct (const Type *obj) const
271 { return likely (this->check_range (obj, obj->min_size)); }
273 bool may_edit (const void *base, unsigned int len)
275 if (this->edit_count >= HB_SANITIZE_MAX_EDITS)
278 const char *p = (const char *) base;
281 DEBUG_MSG_LEVEL (SANITIZE, p, this->debug_depth+1, 0,
282 "may_edit(%u) [%p..%p] (%d bytes) in [%p..%p] -> %s",
285 this->start, this->end,
286 this->writable ? "GRANTED" : "DENIED");
288 return this->writable;
291 template <typename Type, typename ValueType>
292 bool try_set (const Type *obj, const ValueType &v)
294 if (this->may_edit (obj, hb_static_size (Type)))
296 * const_cast<Type *> (obj) = v;
302 template <typename Type>
303 hb_blob_t *sanitize_blob (hb_blob_t *blob)
310 DEBUG_MSG_FUNC (SANITIZE, start, "start");
314 if (unlikely (!start))
320 Type *t = reinterpret_cast<Type *> (const_cast<char *> (start));
322 sane = t->sanitize (this);
327 DEBUG_MSG_FUNC (SANITIZE, start, "passed first round with %d edits; going for second round", edit_count);
329 /* sanitize again to ensure no toe-stepping */
331 sane = t->sanitize (this);
333 DEBUG_MSG_FUNC (SANITIZE, start, "requested %d edits in second round; FAILLING", edit_count);
340 if (edit_count && !writable) {
341 start = hb_blob_get_data_writable (blob, nullptr);
342 end = start + blob->length;
347 /* ok, we made it writable by relocating. try again */
348 DEBUG_MSG_FUNC (SANITIZE, start, "retry");
356 DEBUG_MSG_FUNC (SANITIZE, start, sane ? "PASSED" : "FAILED");
359 hb_blob_make_immutable (blob);
364 hb_blob_destroy (blob);
365 return hb_blob_get_empty ();
369 template <typename Type>
370 hb_blob_t *reference_table (const hb_face_t *face, hb_tag_t tableTag = Type::tableTag)
373 set_num_glyphs (hb_face_get_glyph_count (face));
374 return sanitize_blob<Type> (hb_face_reference_table (face, tableTag));
377 mutable unsigned int debug_depth;
378 const char *start, *end;
382 unsigned int edit_count;
384 unsigned int num_glyphs;
388 struct hb_sanitize_with_object_t
390 template <typename T>
391 hb_sanitize_with_object_t (hb_sanitize_context_t *c, const T& obj) : c (c)
392 { c->set_object (obj); }
393 ~hb_sanitize_with_object_t ()
394 { c->reset_object (); }
397 hb_sanitize_context_t *c;
401 #endif /* HB_SANITIZE_HH */