1================================================================================ 2WHAT IS Flash-Friendly File System (F2FS)? 3================================================================================ 4 5NAND flash memory-based storage devices, such as SSD, eMMC, and SD cards, have 6been equipped on a variety systems ranging from mobile to server systems. Since 7they are known to have different characteristics from the conventional rotating 8disks, a file system, an upper layer to the storage device, should adapt to the 9changes from the sketch in the design level. 10 11F2FS is a file system exploiting NAND flash memory-based storage devices, which 12is based on Log-structured File System (LFS). The design has been focused on 13addressing the fundamental issues in LFS, which are snowball effect of wandering 14tree and high cleaning overhead. 15 16Since a NAND flash memory-based storage device shows different characteristic 17according to its internal geometry or flash memory management scheme, namely FTL, 18F2FS and its tools support various parameters not only for configuring on-disk 19layout, but also for selecting allocation and cleaning algorithms. 20 21The following git tree provides the file system formatting tool (mkfs.f2fs), 22a consistency checking tool (fsck.f2fs), and a debugging tool (dump.f2fs). 23>> git://git.kernel.org/pub/scm/linux/kernel/git/jaegeuk/f2fs-tools.git 24 25For reporting bugs and sending patches, please use the following mailing list: 26>> linux-f2fs-devel@lists.sourceforge.net 27 28================================================================================ 29BACKGROUND AND DESIGN ISSUES 30================================================================================ 31 32Log-structured File System (LFS) 33-------------------------------- 34"A log-structured file system writes all modifications to disk sequentially in 35a log-like structure, thereby speeding up both file writing and crash recovery. 36The log is the only structure on disk; it contains indexing information so that 37files can be read back from the log efficiently. In order to maintain large free 38areas on disk for fast writing, we divide the log into segments and use a 39segment cleaner to compress the live information from heavily fragmented 40segments." from Rosenblum, M. and Ousterhout, J. K., 1992, "The design and 41implementation of a log-structured file system", ACM Trans. Computer Systems 4210, 1, 26–52. 43 44Wandering Tree Problem 45---------------------- 46In LFS, when a file data is updated and written to the end of log, its direct 47pointer block is updated due to the changed location. Then the indirect pointer 48block is also updated due to the direct pointer block update. In this manner, 49the upper index structures such as inode, inode map, and checkpoint block are 50also updated recursively. This problem is called as wandering tree problem [1], 51and in order to enhance the performance, it should eliminate or relax the update 52propagation as much as possible. 53 54[1] Bityutskiy, A. 2005. JFFS3 design issues. http://www.linux-mtd.infradead.org/ 55 56Cleaning Overhead 57----------------- 58Since LFS is based on out-of-place writes, it produces so many obsolete blocks 59scattered across the whole storage. In order to serve new empty log space, it 60needs to reclaim these obsolete blocks seamlessly to users. This job is called 61as a cleaning process. 62 63The process consists of three operations as follows. 641. A victim segment is selected through referencing segment usage table. 652. It loads parent index structures of all the data in the victim identified by 66 segment summary blocks. 673. It checks the cross-reference between the data and its parent index structure. 684. It moves valid data selectively. 69 70This cleaning job may cause unexpected long delays, so the most important goal 71is to hide the latencies to users. And also definitely, it should reduce the 72amount of valid data to be moved, and move them quickly as well. 73 74================================================================================ 75KEY FEATURES 76================================================================================ 77 78Flash Awareness 79--------------- 80- Enlarge the random write area for better performance, but provide the high 81 spatial locality 82- Align FS data structures to the operational units in FTL as best efforts 83 84Wandering Tree Problem 85---------------------- 86- Use a term, “node”, that represents inodes as well as various pointer blocks 87- Introduce Node Address Table (NAT) containing the locations of all the “node” 88 blocks; this will cut off the update propagation. 89 90Cleaning Overhead 91----------------- 92- Support a background cleaning process 93- Support greedy and cost-benefit algorithms for victim selection policies 94- Support multi-head logs for static/dynamic hot and cold data separation 95- Introduce adaptive logging for efficient block allocation 96 97================================================================================ 98MOUNT OPTIONS 99================================================================================ 100 101background_gc=%s Turn on/off cleaning operations, namely garbage 102 collection, triggered in background when I/O subsystem is 103 idle. If background_gc=on, it will turn on the garbage 104 collection and if background_gc=off, garbage collection 105 will be truned off. 106 Default value for this option is on. So garbage 107 collection is on by default. 108disable_roll_forward Disable the roll-forward recovery routine 109discard Issue discard/TRIM commands when a segment is cleaned. 110no_heap Disable heap-style segment allocation which finds free 111 segments for data from the beginning of main area, while 112 for node from the end of main area. 113nouser_xattr Disable Extended User Attributes. Note: xattr is enabled 114 by default if CONFIG_F2FS_FS_XATTR is selected. 115noacl Disable POSIX Access Control List. Note: acl is enabled 116 by default if CONFIG_F2FS_FS_POSIX_ACL is selected. 117active_logs=%u Support configuring the number of active logs. In the 118 current design, f2fs supports only 2, 4, and 6 logs. 119 Default number is 6. 120disable_ext_identify Disable the extension list configured by mkfs, so f2fs 121 does not aware of cold files such as media files. 122inline_xattr Enable the inline xattrs feature. 123inline_data Enable the inline data feature: New created small(<~3.4k) 124 files can be written into inode block. 125flush_merge Merge concurrent cache_flush commands as much as possible 126 to eliminate redundant command issues. If the underlying 127 device handles the cache_flush command relatively slowly, 128 recommend to enable this option. 129nobarrier This option can be used if underlying storage guarantees 130 its cached data should be written to the novolatile area. 131 If this option is set, no cache_flush commands are issued 132 but f2fs still guarantees the write ordering of all the 133 data writes. 134fastboot This option is used when a system wants to reduce mount 135 time as much as possible, even though normal performance 136 can be sacrificed. 137extent_cache Enable an extent cache based on rb-tree, it can cache 138 as many as extent which map between contiguous logical 139 address and physical address per inode, resulting in 140 increasing the cache hit ratio. Set by default. 141noextent_cache Disable an extent cache based on rb-tree explicitly, see 142 the above extent_cache mount option. 143noinline_data Disable the inline data feature, inline data feature is 144 enabled by default. 145data_flush Enable data flushing before checkpoint in order to 146 persist data of regular and symlink. 147mode=%s Control block allocation mode which supports "adaptive" 148 and "lfs". In "lfs" mode, there should be no random 149 writes towards main area. 150io_bits=%u Set the bit size of write IO requests. It should be set 151 with "mode=lfs". 152usrquota Enable plain user disk quota accounting. 153grpquota Enable plain group disk quota accounting. 154prjquota Enable plain project quota accounting. 155usrjquota=<file> Appoint specified file and type during mount, so that quota 156grpjquota=<file> information can be properly updated during recovery flow, 157prjjquota=<file> <quota file>: must be in root directory; 158jqfmt=<quota type> <quota type>: [vfsold,vfsv0,vfsv1]. 159offusrjquota Turn off user journelled quota. 160offgrpjquota Turn off group journelled quota. 161offprjjquota Turn off project journelled quota. 162quota Enable plain user disk quota accounting. 163noquota Disable all plain disk quota option. 164 165================================================================================ 166DEBUGFS ENTRIES 167================================================================================ 168 169/sys/kernel/debug/f2fs/ contains information about all the partitions mounted as 170f2fs. Each file shows the whole f2fs information. 171 172/sys/kernel/debug/f2fs/status includes: 173 - major file system information managed by f2fs currently 174 - average SIT information about whole segments 175 - current memory footprint consumed by f2fs. 176 177================================================================================ 178SYSFS ENTRIES 179================================================================================ 180 181Information about mounted f2f2 file systems can be found in 182/sys/fs/f2fs. Each mounted filesystem will have a directory in 183/sys/fs/f2fs based on its device name (i.e., /sys/fs/f2fs/sda). 184The files in each per-device directory are shown in table below. 185 186Files in /sys/fs/f2fs/<devname> 187(see also Documentation/ABI/testing/sysfs-fs-f2fs) 188.............................................................................. 189 File Content 190 191 gc_max_sleep_time This tuning parameter controls the maximum sleep 192 time for the garbage collection thread. Time is 193 in milliseconds. 194 195 gc_min_sleep_time This tuning parameter controls the minimum sleep 196 time for the garbage collection thread. Time is 197 in milliseconds. 198 199 gc_no_gc_sleep_time This tuning parameter controls the default sleep 200 time for the garbage collection thread. Time is 201 in milliseconds. 202 203 gc_idle This parameter controls the selection of victim 204 policy for garbage collection. Setting gc_idle = 0 205 (default) will disable this option. Setting 206 gc_idle = 1 will select the Cost Benefit approach 207 & setting gc_idle = 2 will select the greedy aproach. 208 209 gc_urgent This parameter controls triggering background GCs 210 urgently or not. Setting gc_urgent = 0 [default] 211 makes back to default behavior, while if it is set 212 to 1, background thread starts to do GC by given 213 gc_urgent_sleep_time interval. 214 215 gc_urgent_sleep_time This parameter controls sleep time for gc_urgent. 216 500 ms is set by default. See above gc_urgent. 217 218 reclaim_segments This parameter controls the number of prefree 219 segments to be reclaimed. If the number of prefree 220 segments is larger than the number of segments 221 in the proportion to the percentage over total 222 volume size, f2fs tries to conduct checkpoint to 223 reclaim the prefree segments to free segments. 224 By default, 5% over total # of segments. 225 226 max_small_discards This parameter controls the number of discard 227 commands that consist small blocks less than 2MB. 228 The candidates to be discarded are cached until 229 checkpoint is triggered, and issued during the 230 checkpoint. By default, it is disabled with 0. 231 232 ipu_policy This parameter controls the policy of in-place 233 updates in f2fs. There are five policies: 234 0x01: F2FS_IPU_FORCE, 0x02: F2FS_IPU_SSR, 235 0x04: F2FS_IPU_UTIL, 0x08: F2FS_IPU_SSR_UTIL, 236 0x10: F2FS_IPU_FSYNC. 237 238 min_ipu_util This parameter controls the threshold to trigger 239 in-place-updates. The number indicates percentage 240 of the filesystem utilization, and used by 241 F2FS_IPU_UTIL and F2FS_IPU_SSR_UTIL policies. 242 243 min_fsync_blocks This parameter controls the threshold to trigger 244 in-place-updates when F2FS_IPU_FSYNC mode is set. 245 The number indicates the number of dirty pages 246 when fsync needs to flush on its call path. If 247 the number is less than this value, it triggers 248 in-place-updates. 249 250 max_victim_search This parameter controls the number of trials to 251 find a victim segment when conducting SSR and 252 cleaning operations. The default value is 4096 253 which covers 8GB block address range. 254 255 dir_level This parameter controls the directory level to 256 support large directory. If a directory has a 257 number of files, it can reduce the file lookup 258 latency by increasing this dir_level value. 259 Otherwise, it needs to decrease this value to 260 reduce the space overhead. The default value is 0. 261 262 ram_thresh This parameter controls the memory footprint used 263 by free nids and cached nat entries. By default, 264 10 is set, which indicates 10 MB / 1 GB RAM. 265 266================================================================================ 267USAGE 268================================================================================ 269 2701. Download userland tools and compile them. 271 2722. Skip, if f2fs was compiled statically inside kernel. 273 Otherwise, insert the f2fs.ko module. 274 # insmod f2fs.ko 275 2763. Create a directory trying to mount 277 # mkdir /mnt/f2fs 278 2794. Format the block device, and then mount as f2fs 280 # mkfs.f2fs -l label /dev/block_device 281 # mount -t f2fs /dev/block_device /mnt/f2fs 282 283mkfs.f2fs 284--------- 285The mkfs.f2fs is for the use of formatting a partition as the f2fs filesystem, 286which builds a basic on-disk layout. 287 288The options consist of: 289-l [label] : Give a volume label, up to 512 unicode name. 290-a [0 or 1] : Split start location of each area for heap-based allocation. 291 1 is set by default, which performs this. 292-o [int] : Set overprovision ratio in percent over volume size. 293 5 is set by default. 294-s [int] : Set the number of segments per section. 295 1 is set by default. 296-z [int] : Set the number of sections per zone. 297 1 is set by default. 298-e [str] : Set basic extension list. e.g. "mp3,gif,mov" 299-t [0 or 1] : Disable discard command or not. 300 1 is set by default, which conducts discard. 301 302fsck.f2fs 303--------- 304The fsck.f2fs is a tool to check the consistency of an f2fs-formatted 305partition, which examines whether the filesystem metadata and user-made data 306are cross-referenced correctly or not. 307Note that, initial version of the tool does not fix any inconsistency. 308 309The options consist of: 310 -d debug level [default:0] 311 312dump.f2fs 313--------- 314The dump.f2fs shows the information of specific inode and dumps SSA and SIT to 315file. Each file is dump_ssa and dump_sit. 316 317The dump.f2fs is used to debug on-disk data structures of the f2fs filesystem. 318It shows on-disk inode information reconized by a given inode number, and is 319able to dump all the SSA and SIT entries into predefined files, ./dump_ssa and 320./dump_sit respectively. 321 322The options consist of: 323 -d debug level [default:0] 324 -i inode no (hex) 325 -s [SIT dump segno from #1~#2 (decimal), for all 0~-1] 326 -a [SSA dump segno from #1~#2 (decimal), for all 0~-1] 327 328Examples: 329# dump.f2fs -i [ino] /dev/sdx 330# dump.f2fs -s 0~-1 /dev/sdx (SIT dump) 331# dump.f2fs -a 0~-1 /dev/sdx (SSA dump) 332 333================================================================================ 334DESIGN 335================================================================================ 336 337On-disk Layout 338-------------- 339 340F2FS divides the whole volume into a number of segments, each of which is fixed 341to 2MB in size. A section is composed of consecutive segments, and a zone 342consists of a set of sections. By default, section and zone sizes are set to one 343segment size identically, but users can easily modify the sizes by mkfs. 344 345F2FS splits the entire volume into six areas, and all the areas except superblock 346consists of multiple segments as described below. 347 348 align with the zone size <-| 349 |-> align with the segment size 350 _________________________________________________________________________ 351 | | | Segment | Node | Segment | | 352 | Superblock | Checkpoint | Info. | Address | Summary | Main | 353 | (SB) | (CP) | Table (SIT) | Table (NAT) | Area (SSA) | | 354 |____________|_____2______|______N______|______N______|______N_____|__N___| 355 . . 356 . . 357 . . 358 ._________________________________________. 359 |_Segment_|_..._|_Segment_|_..._|_Segment_| 360 . . 361 ._________._________ 362 |_section_|__...__|_ 363 . . 364 .________. 365 |__zone__| 366 367- Superblock (SB) 368 : It is located at the beginning of the partition, and there exist two copies 369 to avoid file system crash. It contains basic partition information and some 370 default parameters of f2fs. 371 372- Checkpoint (CP) 373 : It contains file system information, bitmaps for valid NAT/SIT sets, orphan 374 inode lists, and summary entries of current active segments. 375 376- Segment Information Table (SIT) 377 : It contains segment information such as valid block count and bitmap for the 378 validity of all the blocks. 379 380- Node Address Table (NAT) 381 : It is composed of a block address table for all the node blocks stored in 382 Main area. 383 384- Segment Summary Area (SSA) 385 : It contains summary entries which contains the owner information of all the 386 data and node blocks stored in Main area. 387 388- Main Area 389 : It contains file and directory data including their indices. 390 391In order to avoid misalignment between file system and flash-based storage, F2FS 392aligns the start block address of CP with the segment size. Also, it aligns the 393start block address of Main area with the zone size by reserving some segments 394in SSA area. 395 396Reference the following survey for additional technical details. 397https://wiki.linaro.org/WorkingGroups/Kernel/Projects/FlashCardSurvey 398 399File System Metadata Structure 400------------------------------ 401 402F2FS adopts the checkpointing scheme to maintain file system consistency. At 403mount time, F2FS first tries to find the last valid checkpoint data by scanning 404CP area. In order to reduce the scanning time, F2FS uses only two copies of CP. 405One of them always indicates the last valid data, which is called as shadow copy 406mechanism. In addition to CP, NAT and SIT also adopt the shadow copy mechanism. 407 408For file system consistency, each CP points to which NAT and SIT copies are 409valid, as shown as below. 410 411 +--------+----------+---------+ 412 | CP | SIT | NAT | 413 +--------+----------+---------+ 414 . . . . 415 . . . . 416 . . . . 417 +-------+-------+--------+--------+--------+--------+ 418 | CP #0 | CP #1 | SIT #0 | SIT #1 | NAT #0 | NAT #1 | 419 +-------+-------+--------+--------+--------+--------+ 420 | ^ ^ 421 | | | 422 `----------------------------------------' 423 424Index Structure 425--------------- 426 427The key data structure to manage the data locations is a "node". Similar to 428traditional file structures, F2FS has three types of node: inode, direct node, 429indirect node. F2FS assigns 4KB to an inode block which contains 923 data block 430indices, two direct node pointers, two indirect node pointers, and one double 431indirect node pointer as described below. One direct node block contains 1018 432data blocks, and one indirect node block contains also 1018 node blocks. Thus, 433one inode block (i.e., a file) covers: 434 435 4KB * (923 + 2 * 1018 + 2 * 1018 * 1018 + 1018 * 1018 * 1018) := 3.94TB. 436 437 Inode block (4KB) 438 |- data (923) 439 |- direct node (2) 440 | `- data (1018) 441 |- indirect node (2) 442 | `- direct node (1018) 443 | `- data (1018) 444 `- double indirect node (1) 445 `- indirect node (1018) 446 `- direct node (1018) 447 `- data (1018) 448 449Note that, all the node blocks are mapped by NAT which means the location of 450each node is translated by the NAT table. In the consideration of the wandering 451tree problem, F2FS is able to cut off the propagation of node updates caused by 452leaf data writes. 453 454Directory Structure 455------------------- 456 457A directory entry occupies 11 bytes, which consists of the following attributes. 458 459- hash hash value of the file name 460- ino inode number 461- len the length of file name 462- type file type such as directory, symlink, etc 463 464A dentry block consists of 214 dentry slots and file names. Therein a bitmap is 465used to represent whether each dentry is valid or not. A dentry block occupies 4664KB with the following composition. 467 468 Dentry Block(4 K) = bitmap (27 bytes) + reserved (3 bytes) + 469 dentries(11 * 214 bytes) + file name (8 * 214 bytes) 470 471 [Bucket] 472 +--------------------------------+ 473 |dentry block 1 | dentry block 2 | 474 +--------------------------------+ 475 . . 476 . . 477 . [Dentry Block Structure: 4KB] . 478 +--------+----------+----------+------------+ 479 | bitmap | reserved | dentries | file names | 480 +--------+----------+----------+------------+ 481 [Dentry Block: 4KB] . . 482 . . 483 . . 484 +------+------+-----+------+ 485 | hash | ino | len | type | 486 +------+------+-----+------+ 487 [Dentry Structure: 11 bytes] 488 489F2FS implements multi-level hash tables for directory structure. Each level has 490a hash table with dedicated number of hash buckets as shown below. Note that 491"A(2B)" means a bucket includes 2 data blocks. 492 493---------------------- 494A : bucket 495B : block 496N : MAX_DIR_HASH_DEPTH 497---------------------- 498 499level #0 | A(2B) 500 | 501level #1 | A(2B) - A(2B) 502 | 503level #2 | A(2B) - A(2B) - A(2B) - A(2B) 504 . | . . . . 505level #N/2 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) 506 . | . . . . 507level #N | A(4B) - A(4B) - A(4B) - A(4B) - A(4B) - ... - A(4B) 508 509The number of blocks and buckets are determined by, 510 511 ,- 2, if n < MAX_DIR_HASH_DEPTH / 2, 512 # of blocks in level #n = | 513 `- 4, Otherwise 514 515 ,- 2^(n + dir_level), 516 | if n + dir_level < MAX_DIR_HASH_DEPTH / 2, 517 # of buckets in level #n = | 518 `- 2^((MAX_DIR_HASH_DEPTH / 2) - 1), 519 Otherwise 520 521When F2FS finds a file name in a directory, at first a hash value of the file 522name is calculated. Then, F2FS scans the hash table in level #0 to find the 523dentry consisting of the file name and its inode number. If not found, F2FS 524scans the next hash table in level #1. In this way, F2FS scans hash tables in 525each levels incrementally from 1 to N. In each levels F2FS needs to scan only 526one bucket determined by the following equation, which shows O(log(# of files)) 527complexity. 528 529 bucket number to scan in level #n = (hash value) % (# of buckets in level #n) 530 531In the case of file creation, F2FS finds empty consecutive slots that cover the 532file name. F2FS searches the empty slots in the hash tables of whole levels from 5331 to N in the same way as the lookup operation. 534 535The following figure shows an example of two cases holding children. 536 --------------> Dir <-------------- 537 | | 538 child child 539 540 child - child [hole] - child 541 542 child - child - child [hole] - [hole] - child 543 544 Case 1: Case 2: 545 Number of children = 6, Number of children = 3, 546 File size = 7 File size = 7 547 548Default Block Allocation 549------------------------ 550 551At runtime, F2FS manages six active logs inside "Main" area: Hot/Warm/Cold node 552and Hot/Warm/Cold data. 553 554- Hot node contains direct node blocks of directories. 555- Warm node contains direct node blocks except hot node blocks. 556- Cold node contains indirect node blocks 557- Hot data contains dentry blocks 558- Warm data contains data blocks except hot and cold data blocks 559- Cold data contains multimedia data or migrated data blocks 560 561LFS has two schemes for free space management: threaded log and copy-and-compac- 562tion. The copy-and-compaction scheme which is known as cleaning, is well-suited 563for devices showing very good sequential write performance, since free segments 564are served all the time for writing new data. However, it suffers from cleaning 565overhead under high utilization. Contrarily, the threaded log scheme suffers 566from random writes, but no cleaning process is needed. F2FS adopts a hybrid 567scheme where the copy-and-compaction scheme is adopted by default, but the 568policy is dynamically changed to the threaded log scheme according to the file 569system status. 570 571In order to align F2FS with underlying flash-based storage, F2FS allocates a 572segment in a unit of section. F2FS expects that the section size would be the 573same as the unit size of garbage collection in FTL. Furthermore, with respect 574to the mapping granularity in FTL, F2FS allocates each section of the active 575logs from different zones as much as possible, since FTL can write the data in 576the active logs into one allocation unit according to its mapping granularity. 577 578Cleaning process 579---------------- 580 581F2FS does cleaning both on demand and in the background. On-demand cleaning is 582triggered when there are not enough free segments to serve VFS calls. Background 583cleaner is operated by a kernel thread, and triggers the cleaning job when the 584system is idle. 585 586F2FS supports two victim selection policies: greedy and cost-benefit algorithms. 587In the greedy algorithm, F2FS selects a victim segment having the smallest number 588of valid blocks. In the cost-benefit algorithm, F2FS selects a victim segment 589according to the segment age and the number of valid blocks in order to address 590log block thrashing problem in the greedy algorithm. F2FS adopts the greedy 591algorithm for on-demand cleaner, while background cleaner adopts cost-benefit 592algorithm. 593 594In order to identify whether the data in the victim segment are valid or not, 595F2FS manages a bitmap. Each bit represents the validity of a block, and the 596bitmap is composed of a bit stream covering whole blocks in main area. 597