1:orphan: 2 3.. UBIFS Authentication 4.. sigma star gmbh 5.. 2018 6 7Introduction 8============ 9 10UBIFS utilizes the fscrypt framework to provide confidentiality for file 11contents and file names. This prevents attacks where an attacker is able to 12read contents of the filesystem on a single point in time. A classic example 13is a lost smartphone where the attacker is unable to read personal data stored 14on the device without the filesystem decryption key. 15 16At the current state, UBIFS encryption however does not prevent attacks where 17the attacker is able to modify the filesystem contents and the user uses the 18device afterwards. In such a scenario an attacker can modify filesystem 19contents arbitrarily without the user noticing. One example is to modify a 20binary to perform a malicious action when executed [DMC-CBC-ATTACK]. Since 21most of the filesystem metadata of UBIFS is stored in plain, this makes it 22fairly easy to swap files and replace their contents. 23 24Other full disk encryption systems like dm-crypt cover all filesystem metadata, 25which makes such kinds of attacks more complicated, but not impossible. 26Especially, if the attacker is given access to the device multiple points in 27time. For dm-crypt and other filesystems that build upon the Linux block IO 28layer, the dm-integrity or dm-verity subsystems [DM-INTEGRITY, DM-VERITY] 29can be used to get full data authentication at the block layer. 30These can also be combined with dm-crypt [CRYPTSETUP2]. 31 32This document describes an approach to get file contents _and_ full metadata 33authentication for UBIFS. Since UBIFS uses fscrypt for file contents and file 34name encryption, the authentication system could be tied into fscrypt such that 35existing features like key derivation can be utilized. It should however also 36be possible to use UBIFS authentication without using encryption. 37 38 39MTD, UBI & UBIFS 40---------------- 41 42On Linux, the MTD (Memory Technology Devices) subsystem provides a uniform 43interface to access raw flash devices. One of the more prominent subsystems that 44work on top of MTD is UBI (Unsorted Block Images). It provides volume management 45for flash devices and is thus somewhat similar to LVM for block devices. In 46addition, it deals with flash-specific wear-leveling and transparent I/O error 47handling. UBI offers logical erase blocks (LEBs) to the layers on top of it 48and maps them transparently to physical erase blocks (PEBs) on the flash. 49 50UBIFS is a filesystem for raw flash which operates on top of UBI. Thus, wear 51leveling and some flash specifics are left to UBI, while UBIFS focuses on 52scalability, performance and recoverability. 53 54:: 55 56 +------------+ +*******+ +-----------+ +-----+ 57 | | * UBIFS * | UBI-BLOCK | | ... | 58 | JFFS/JFFS2 | +*******+ +-----------+ +-----+ 59 | | +-----------------------------+ +-----------+ +-----+ 60 | | | UBI | | MTD-BLOCK | | ... | 61 +------------+ +-----------------------------+ +-----------+ +-----+ 62 +------------------------------------------------------------------+ 63 | MEMORY TECHNOLOGY DEVICES (MTD) | 64 +------------------------------------------------------------------+ 65 +-----------------------------+ +--------------------------+ +-----+ 66 | NAND DRIVERS | | NOR DRIVERS | | ... | 67 +-----------------------------+ +--------------------------+ +-----+ 68 69 Figure 1: Linux kernel subsystems for dealing with raw flash 70 71 72 73Internally, UBIFS maintains multiple data structures which are persisted on 74the flash: 75 76- *Index*: an on-flash B+ tree where the leaf nodes contain filesystem data 77- *Journal*: an additional data structure to collect FS changes before updating 78 the on-flash index and reduce flash wear. 79- *Tree Node Cache (TNC)*: an in-memory B+ tree that reflects the current FS 80 state to avoid frequent flash reads. It is basically the in-memory 81 representation of the index, but contains additional attributes. 82- *LEB property tree (LPT)*: an on-flash B+ tree for free space accounting per 83 UBI LEB. 84 85In the remainder of this section we will cover the on-flash UBIFS data 86structures in more detail. The TNC is of less importance here since it is never 87persisted onto the flash directly. More details on UBIFS can also be found in 88[UBIFS-WP]. 89 90 91UBIFS Index & Tree Node Cache 92~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 93 94Basic on-flash UBIFS entities are called *nodes*. UBIFS knows different types 95of nodes. Eg. data nodes (`struct ubifs_data_node`) which store chunks of file 96contents or inode nodes (`struct ubifs_ino_node`) which represent VFS inodes. 97Almost all types of nodes share a common header (`ubifs_ch`) containing basic 98information like node type, node length, a sequence number, etc. (see 99`fs/ubifs/ubifs-media.h`in kernel source). Exceptions are entries of the LPT 100and some less important node types like padding nodes which are used to pad 101unusable content at the end of LEBs. 102 103To avoid re-writing the whole B+ tree on every single change, it is implemented 104as *wandering tree*, where only the changed nodes are re-written and previous 105versions of them are obsoleted without erasing them right away. As a result, 106the index is not stored in a single place on the flash, but *wanders* around 107and there are obsolete parts on the flash as long as the LEB containing them is 108not reused by UBIFS. To find the most recent version of the index, UBIFS stores 109a special node called *master node* into UBI LEB 1 which always points to the 110most recent root node of the UBIFS index. For recoverability, the master node 111is additionally duplicated to LEB 2. Mounting UBIFS is thus a simple read of 112LEB 1 and 2 to get the current master node and from there get the location of 113the most recent on-flash index. 114 115The TNC is the in-memory representation of the on-flash index. It contains some 116additional runtime attributes per node which are not persisted. One of these is 117a dirty-flag which marks nodes that have to be persisted the next time the 118index is written onto the flash. The TNC acts as a write-back cache and all 119modifications of the on-flash index are done through the TNC. Like other caches, 120the TNC does not have to mirror the full index into memory, but reads parts of 121it from flash whenever needed. A *commit* is the UBIFS operation of updating the 122on-flash filesystem structures like the index. On every commit, the TNC nodes 123marked as dirty are written to the flash to update the persisted index. 124 125 126Journal 127~~~~~~~ 128 129To avoid wearing out the flash, the index is only persisted (*commited*) when 130certain conditions are met (eg. ``fsync(2)``). The journal is used to record 131any changes (in form of inode nodes, data nodes etc.) between commits 132of the index. During mount, the journal is read from the flash and replayed 133onto the TNC (which will be created on-demand from the on-flash index). 134 135UBIFS reserves a bunch of LEBs just for the journal called *log area*. The 136amount of log area LEBs is configured on filesystem creation (using 137``mkfs.ubifs``) and stored in the superblock node. The log area contains only 138two types of nodes: *reference nodes* and *commit start nodes*. A commit start 139node is written whenever an index commit is performed. Reference nodes are 140written on every journal update. Each reference node points to the position of 141other nodes (inode nodes, data nodes etc.) on the flash that are part of this 142journal entry. These nodes are called *buds* and describe the actual filesystem 143changes including their data. 144 145The log area is maintained as a ring. Whenever the journal is almost full, 146a commit is initiated. This also writes a commit start node so that during 147mount, UBIFS will seek for the most recent commit start node and just replay 148every reference node after that. Every reference node before the commit start 149node will be ignored as they are already part of the on-flash index. 150 151When writing a journal entry, UBIFS first ensures that enough space is 152available to write the reference node and buds part of this entry. Then, the 153reference node is written and afterwards the buds describing the file changes. 154On replay, UBIFS will record every reference node and inspect the location of 155the referenced LEBs to discover the buds. If these are corrupt or missing, 156UBIFS will attempt to recover them by re-reading the LEB. This is however only 157done for the last referenced LEB of the journal. Only this can become corrupt 158because of a power cut. If the recovery fails, UBIFS will not mount. An error 159for every other LEB will directly cause UBIFS to fail the mount operation. 160 161:: 162 163 | ---- LOG AREA ---- | ---------- MAIN AREA ------------ | 164 165 -----+------+-----+--------+---- ------+-----+-----+--------------- 166 \ | | | | / / | | | \ 167 / CS | REF | REF | | \ \ DENT | INO | INO | / 168 \ | | | | / / | | | \ 169 ----+------+-----+--------+--- -------+-----+-----+---------------- 170 | | ^ ^ 171 | | | | 172 +------------------------+ | 173 | | 174 +-------------------------------+ 175 176 177 Figure 2: UBIFS flash layout of log area with commit start nodes 178 (CS) and reference nodes (REF) pointing to main area 179 containing their buds 180 181 182LEB Property Tree/Table 183~~~~~~~~~~~~~~~~~~~~~~~ 184 185The LEB property tree is used to store per-LEB information. This includes the 186LEB type and amount of free and *dirty* (old, obsolete content) space [1]_ on 187the LEB. The type is important, because UBIFS never mixes index nodes with data 188nodes on a single LEB and thus each LEB has a specific purpose. This again is 189useful for free space calculations. See [UBIFS-WP] for more details. 190 191The LEB property tree again is a B+ tree, but it is much smaller than the 192index. Due to its smaller size it is always written as one chunk on every 193commit. Thus, saving the LPT is an atomic operation. 194 195 196.. [1] Since LEBs can only be appended and never overwritten, there is a 197 difference between free space ie. the remaining space left on the LEB to be 198 written to without erasing it and previously written content that is obsolete 199 but can't be overwritten without erasing the full LEB. 200 201 202UBIFS Authentication 203==================== 204 205This chapter introduces UBIFS authentication which enables UBIFS to verify 206the authenticity and integrity of metadata and file contents stored on flash. 207 208 209Threat Model 210------------ 211 212UBIFS authentication enables detection of offline data modification. While it 213does not prevent it, it enables (trusted) code to check the integrity and 214authenticity of on-flash file contents and filesystem metadata. This covers 215attacks where file contents are swapped. 216 217UBIFS authentication will not protect against rollback of full flash contents. 218Ie. an attacker can still dump the flash and restore it at a later time without 219detection. It will also not protect against partial rollback of individual 220index commits. That means that an attacker is able to partially undo changes. 221This is possible because UBIFS does not immediately overwrites obsolete 222versions of the index tree or the journal, but instead marks them as obsolete 223and garbage collection erases them at a later time. An attacker can use this by 224erasing parts of the current tree and restoring old versions that are still on 225the flash and have not yet been erased. This is possible, because every commit 226will always write a new version of the index root node and the master node 227without overwriting the previous version. This is further helped by the 228wear-leveling operations of UBI which copies contents from one physical 229eraseblock to another and does not atomically erase the first eraseblock. 230 231UBIFS authentication does not cover attacks where an attacker is able to 232execute code on the device after the authentication key was provided. 233Additional measures like secure boot and trusted boot have to be taken to 234ensure that only trusted code is executed on a device. 235 236 237Authentication 238-------------- 239 240To be able to fully trust data read from flash, all UBIFS data structures 241stored on flash are authenticated. That is: 242 243- The index which includes file contents, file metadata like extended 244 attributes, file length etc. 245- The journal which also contains file contents and metadata by recording changes 246 to the filesystem 247- The LPT which stores UBI LEB metadata which UBIFS uses for free space accounting 248 249 250Index Authentication 251~~~~~~~~~~~~~~~~~~~~ 252 253Through UBIFS' concept of a wandering tree, it already takes care of only 254updating and persisting changed parts from leaf node up to the root node 255of the full B+ tree. This enables us to augment the index nodes of the tree 256with a hash over each node's child nodes. As a result, the index basically also 257a Merkle tree. Since the leaf nodes of the index contain the actual filesystem 258data, the hashes of their parent index nodes thus cover all the file contents 259and file metadata. When a file changes, the UBIFS index is updated accordingly 260from the leaf nodes up to the root node including the master node. This process 261can be hooked to recompute the hash only for each changed node at the same time. 262Whenever a file is read, UBIFS can verify the hashes from each leaf node up to 263the root node to ensure the node's integrity. 264 265To ensure the authenticity of the whole index, the UBIFS master node stores a 266keyed hash (HMAC) over its own contents and a hash of the root node of the index 267tree. As mentioned above, the master node is always written to the flash whenever 268the index is persisted (ie. on index commit). 269 270Using this approach only UBIFS index nodes and the master node are changed to 271include a hash. All other types of nodes will remain unchanged. This reduces 272the storage overhead which is precious for users of UBIFS (ie. embedded 273devices). 274 275:: 276 277 +---------------+ 278 | Master Node | 279 | (hash) | 280 +---------------+ 281 | 282 v 283 +-------------------+ 284 | Index Node #1 | 285 | | 286 | branch0 branchn | 287 | (hash) (hash) | 288 +-------------------+ 289 | ... | (fanout: 8) 290 | | 291 +-------+ +------+ 292 | | 293 v v 294 +-------------------+ +-------------------+ 295 | Index Node #2 | | Index Node #3 | 296 | | | | 297 | branch0 branchn | | branch0 branchn | 298 | (hash) (hash) | | (hash) (hash) | 299 +-------------------+ +-------------------+ 300 | ... | ... | 301 v v v 302 +-----------+ +----------+ +-----------+ 303 | Data Node | | INO Node | | DENT Node | 304 +-----------+ +----------+ +-----------+ 305 306 307 Figure 3: Coverage areas of index node hash and master node HMAC 308 309 310 311The most important part for robustness and power-cut safety is to atomically 312persist the hash and file contents. Here the existing UBIFS logic for how 313changed nodes are persisted is already designed for this purpose such that 314UBIFS can safely recover if a power-cut occurs while persisting. Adding 315hashes to index nodes does not change this since each hash will be persisted 316atomically together with its respective node. 317 318 319Journal Authentication 320~~~~~~~~~~~~~~~~~~~~~~ 321 322The journal is authenticated too. Since the journal is continuously written 323it is necessary to also add authentication information frequently to the 324journal so that in case of a powercut not too much data can't be authenticated. 325This is done by creating a continuous hash beginning from the commit start node 326over the previous reference nodes, the current reference node, and the bud 327nodes. From time to time whenever it is suitable authentication nodes are added 328between the bud nodes. This new node type contains a HMAC over the current state 329of the hash chain. That way a journal can be authenticated up to the last 330authentication node. The tail of the journal which may not have a authentication 331node cannot be authenticated and is skipped during journal replay. 332 333We get this picture for journal authentication:: 334 335 ,,,,,,,, 336 ,......,........................................... 337 ,. CS , hash1.----. hash2.----. 338 ,. | , . |hmac . |hmac 339 ,. v , . v . v 340 ,.REF#0,-> bud -> bud -> bud.-> auth -> bud -> bud.-> auth ... 341 ,..|...,........................................... 342 , | , 343 , | ,,,,,,,,,,,,,,, 344 . | hash3,----. 345 , | , |hmac 346 , v , v 347 , REF#1 -> bud -> bud,-> auth ... 348 ,,,|,,,,,,,,,,,,,,,,,, 349 v 350 REF#2 -> ... 351 | 352 V 353 ... 354 355Since the hash also includes the reference nodes an attacker cannot reorder or 356skip any journal heads for replay. An attacker can only remove bud nodes or 357reference nodes from the end of the journal, effectively rewinding the 358filesystem at maximum back to the last commit. 359 360The location of the log area is stored in the master node. Since the master 361node is authenticated with a HMAC as described above, it is not possible to 362tamper with that without detection. The size of the log area is specified when 363the filesystem is created using `mkfs.ubifs` and stored in the superblock node. 364To avoid tampering with this and other values stored there, a HMAC is added to 365the superblock struct. The superblock node is stored in LEB 0 and is only 366modified on feature flag or similar changes, but never on file changes. 367 368 369LPT Authentication 370~~~~~~~~~~~~~~~~~~ 371 372The location of the LPT root node on the flash is stored in the UBIFS master 373node. Since the LPT is written and read atomically on every commit, there is 374no need to authenticate individual nodes of the tree. It suffices to 375protect the integrity of the full LPT by a simple hash stored in the master 376node. Since the master node itself is authenticated, the LPTs authenticity can 377be verified by verifying the authenticity of the master node and comparing the 378LTP hash stored there with the hash computed from the read on-flash LPT. 379 380 381Key Management 382-------------- 383 384For simplicity, UBIFS authentication uses a single key to compute the HMACs 385of superblock, master, commit start and reference nodes. This key has to be 386available on creation of the filesystem (`mkfs.ubifs`) to authenticate the 387superblock node. Further, it has to be available on mount of the filesystem 388to verify authenticated nodes and generate new HMACs for changes. 389 390UBIFS authentication is intended to operate side-by-side with UBIFS encryption 391(fscrypt) to provide confidentiality and authenticity. Since UBIFS encryption 392has a different approach of encryption policies per directory, there can be 393multiple fscrypt master keys and there might be folders without encryption. 394UBIFS authentication on the other hand has an all-or-nothing approach in the 395sense that it either authenticates everything of the filesystem or nothing. 396Because of this and because UBIFS authentication should also be usable without 397encryption, it does not share the same master key with fscrypt, but manages 398a dedicated authentication key. 399 400The API for providing the authentication key has yet to be defined, but the 401key can eg. be provided by userspace through a keyring similar to the way it 402is currently done in fscrypt. It should however be noted that the current 403fscrypt approach has shown its flaws and the userspace API will eventually 404change [FSCRYPT-POLICY2]. 405 406Nevertheless, it will be possible for a user to provide a single passphrase 407or key in userspace that covers UBIFS authentication and encryption. This can 408be solved by the corresponding userspace tools which derive a second key for 409authentication in addition to the derived fscrypt master key used for 410encryption. 411 412To be able to check if the proper key is available on mount, the UBIFS 413superblock node will additionally store a hash of the authentication key. This 414approach is similar to the approach proposed for fscrypt encryption policy v2 415[FSCRYPT-POLICY2]. 416 417 418Future Extensions 419================= 420 421In certain cases where a vendor wants to provide an authenticated filesystem 422image to customers, it should be possible to do so without sharing the secret 423UBIFS authentication key. Instead, in addition the each HMAC a digital 424signature could be stored where the vendor shares the public key alongside the 425filesystem image. In case this filesystem has to be modified afterwards, 426UBIFS can exchange all digital signatures with HMACs on first mount similar 427to the way the IMA/EVM subsystem deals with such situations. The HMAC key 428will then have to be provided beforehand in the normal way. 429 430 431References 432========== 433 434[CRYPTSETUP2] http://www.saout.de/pipermail/dm-crypt/2017-November/005745.html 435 436[DMC-CBC-ATTACK] http://www.jakoblell.com/blog/2013/12/22/practical-malleability-attack-against-cbc-encrypted-luks-partitions/ 437 438[DM-INTEGRITY] https://www.kernel.org/doc/Documentation/device-mapper/dm-integrity.rst 439 440[DM-VERITY] https://www.kernel.org/doc/Documentation/device-mapper/verity.rst 441 442[FSCRYPT-POLICY2] https://www.spinics.net/lists/linux-ext4/msg58710.html 443 444[UBIFS-WP] http://www.linux-mtd.infradead.org/doc/ubifs_whitepaper.pdf 445