1.. SPDX-License-Identifier: GPL-2.0 2 3================================= 4Flash-Friendly File System (F2FS) 5================================= 6 7Overview 8======== 9 10NAND flash memory-based storage devices, such as SSD, eMMC, and SD cards, have 11been equipped on a variety systems ranging from mobile to server systems. Since 12they are known to have different characteristics from the conventional rotating 13disks, a file system, an upper layer to the storage device, should adapt to the 14changes from the sketch in the design level. 15 16F2FS is a file system exploiting NAND flash memory-based storage devices, which 17is based on Log-structured File System (LFS). The design has been focused on 18addressing the fundamental issues in LFS, which are snowball effect of wandering 19tree and high cleaning overhead. 20 21Since a NAND flash memory-based storage device shows different characteristic 22according to its internal geometry or flash memory management scheme, namely FTL, 23F2FS and its tools support various parameters not only for configuring on-disk 24layout, but also for selecting allocation and cleaning algorithms. 25 26The following git tree provides the file system formatting tool (mkfs.f2fs), 27a consistency checking tool (fsck.f2fs), and a debugging tool (dump.f2fs). 28 29- git://git.kernel.org/pub/scm/linux/kernel/git/jaegeuk/f2fs-tools.git 30 31For sending patches, please use the following mailing list: 32 33- linux-f2fs-devel@lists.sourceforge.net 34 35For reporting bugs, please use the following f2fs bug tracker link: 36 37- https://bugzilla.kernel.org/enter_bug.cgi?product=File%20System&component=f2fs 38 39Background and Design issues 40============================ 41 42Log-structured File System (LFS) 43-------------------------------- 44"A log-structured file system writes all modifications to disk sequentially in 45a log-like structure, thereby speeding up both file writing and crash recovery. 46The log is the only structure on disk; it contains indexing information so that 47files can be read back from the log efficiently. In order to maintain large free 48areas on disk for fast writing, we divide the log into segments and use a 49segment cleaner to compress the live information from heavily fragmented 50segments." from Rosenblum, M. and Ousterhout, J. K., 1992, "The design and 51implementation of a log-structured file system", ACM Trans. Computer Systems 5210, 1, 26–52. 53 54Wandering Tree Problem 55---------------------- 56In LFS, when a file data is updated and written to the end of log, its direct 57pointer block is updated due to the changed location. Then the indirect pointer 58block is also updated due to the direct pointer block update. In this manner, 59the upper index structures such as inode, inode map, and checkpoint block are 60also updated recursively. This problem is called as wandering tree problem [1], 61and in order to enhance the performance, it should eliminate or relax the update 62propagation as much as possible. 63 64[1] Bityutskiy, A. 2005. JFFS3 design issues. http://www.linux-mtd.infradead.org/ 65 66Cleaning Overhead 67----------------- 68Since LFS is based on out-of-place writes, it produces so many obsolete blocks 69scattered across the whole storage. In order to serve new empty log space, it 70needs to reclaim these obsolete blocks seamlessly to users. This job is called 71as a cleaning process. 72 73The process consists of three operations as follows. 74 751. A victim segment is selected through referencing segment usage table. 762. It loads parent index structures of all the data in the victim identified by 77 segment summary blocks. 783. It checks the cross-reference between the data and its parent index structure. 794. It moves valid data selectively. 80 81This cleaning job may cause unexpected long delays, so the most important goal 82is to hide the latencies to users. And also definitely, it should reduce the 83amount of valid data to be moved, and move them quickly as well. 84 85Key Features 86============ 87 88Flash Awareness 89--------------- 90- Enlarge the random write area for better performance, but provide the high 91 spatial locality 92- Align FS data structures to the operational units in FTL as best efforts 93 94Wandering Tree Problem 95---------------------- 96- Use a term, “node”, that represents inodes as well as various pointer blocks 97- Introduce Node Address Table (NAT) containing the locations of all the “node” 98 blocks; this will cut off the update propagation. 99 100Cleaning Overhead 101----------------- 102- Support a background cleaning process 103- Support greedy and cost-benefit algorithms for victim selection policies 104- Support multi-head logs for static/dynamic hot and cold data separation 105- Introduce adaptive logging for efficient block allocation 106 107Mount Options 108============= 109 110 111======================== ============================================================ 112background_gc=%s Turn on/off cleaning operations, namely garbage 113 collection, triggered in background when I/O subsystem is 114 idle. If background_gc=on, it will turn on the garbage 115 collection and if background_gc=off, garbage collection 116 will be turned off. If background_gc=sync, it will turn 117 on synchronous garbage collection running in background. 118 Default value for this option is on. So garbage 119 collection is on by default. 120gc_merge When background_gc is on, this option can be enabled to 121 let background GC thread to handle foreground GC requests, 122 it can eliminate the sluggish issue caused by slow foreground 123 GC operation when GC is triggered from a process with limited 124 I/O and CPU resources. 125nogc_merge Disable GC merge feature. 126disable_roll_forward Disable the roll-forward recovery routine 127norecovery Disable the roll-forward recovery routine, mounted read- 128 only (i.e., -o ro,disable_roll_forward) 129discard/nodiscard Enable/disable real-time discard in f2fs, if discard is 130 enabled, f2fs will issue discard/TRIM commands when a 131 segment is cleaned. 132heap/no_heap Deprecated. 133nouser_xattr Disable Extended User Attributes. Note: xattr is enabled 134 by default if CONFIG_F2FS_FS_XATTR is selected. 135noacl Disable POSIX Access Control List. Note: acl is enabled 136 by default if CONFIG_F2FS_FS_POSIX_ACL is selected. 137active_logs=%u Support configuring the number of active logs. In the 138 current design, f2fs supports only 2, 4, and 6 logs. 139 Default number is 6. 140disable_ext_identify Disable the extension list configured by mkfs, so f2fs 141 is not aware of cold files such as media files. 142inline_xattr Enable the inline xattrs feature. 143noinline_xattr Disable the inline xattrs feature. 144inline_xattr_size=%u Support configuring inline xattr size, it depends on 145 flexible inline xattr feature. 146inline_data Enable the inline data feature: Newly created small (<~3.4k) 147 files can be written into inode block. 148inline_dentry Enable the inline dir feature: data in newly created 149 directory entries can be written into inode block. The 150 space of inode block which is used to store inline 151 dentries is limited to ~3.4k. 152noinline_dentry Disable the inline dentry feature. 153flush_merge Merge concurrent cache_flush commands as much as possible 154 to eliminate redundant command issues. If the underlying 155 device handles the cache_flush command relatively slowly, 156 recommend to enable this option. 157nobarrier This option can be used if underlying storage guarantees 158 its cached data should be written to the novolatile area. 159 If this option is set, no cache_flush commands are issued 160 but f2fs still guarantees the write ordering of all the 161 data writes. 162barrier If this option is set, cache_flush commands are allowed to be 163 issued. 164fastboot This option is used when a system wants to reduce mount 165 time as much as possible, even though normal performance 166 can be sacrificed. 167extent_cache Enable an extent cache based on rb-tree, it can cache 168 as many as extent which map between contiguous logical 169 address and physical address per inode, resulting in 170 increasing the cache hit ratio. Set by default. 171noextent_cache Disable an extent cache based on rb-tree explicitly, see 172 the above extent_cache mount option. 173noinline_data Disable the inline data feature, inline data feature is 174 enabled by default. 175data_flush Enable data flushing before checkpoint in order to 176 persist data of regular and symlink. 177reserve_root=%d Support configuring reserved space which is used for 178 allocation from a privileged user with specified uid or 179 gid, unit: 4KB, the default limit is 12.5% of user blocks. 180reserve_node=%d Support configuring reserved nodes which are used for 181 allocation from a privileged user with specified uid or 182 gid, the default limit is 12.5% of all nodes. 183resuid=%d The user ID which may use the reserved blocks and nodes. 184resgid=%d The group ID which may use the reserved blocks and nodes. 185fault_injection=%d Enable fault injection in all supported types with 186 specified injection rate. 187fault_type=%d Support configuring fault injection type, should be 188 enabled with fault_injection option, fault type value 189 is shown below, it supports single or combined type. 190 191 .. code-block:: none 192 193 =========================== ========== 194 Type_Name Type_Value 195 =========================== ========== 196 FAULT_KMALLOC 0x00000001 197 FAULT_KVMALLOC 0x00000002 198 FAULT_PAGE_ALLOC 0x00000004 199 FAULT_PAGE_GET 0x00000008 200 FAULT_ALLOC_BIO 0x00000010 (obsolete) 201 FAULT_ALLOC_NID 0x00000020 202 FAULT_ORPHAN 0x00000040 203 FAULT_BLOCK 0x00000080 204 FAULT_DIR_DEPTH 0x00000100 205 FAULT_EVICT_INODE 0x00000200 206 FAULT_TRUNCATE 0x00000400 207 FAULT_READ_IO 0x00000800 208 FAULT_CHECKPOINT 0x00001000 209 FAULT_DISCARD 0x00002000 (obsolete) 210 FAULT_WRITE_IO 0x00004000 211 FAULT_SLAB_ALLOC 0x00008000 212 FAULT_DQUOT_INIT 0x00010000 213 FAULT_LOCK_OP 0x00020000 214 FAULT_BLKADDR_VALIDITY 0x00040000 215 FAULT_BLKADDR_CONSISTENCE 0x00080000 216 FAULT_NO_SEGMENT 0x00100000 217 FAULT_INCONSISTENT_FOOTER 0x00200000 218 FAULT_ATOMIC_TIMEOUT 0x00400000 (1000ms) 219 FAULT_VMALLOC 0x00800000 220 FAULT_LOCK_TIMEOUT 0x01000000 (1000ms) 221 FAULT_SKIP_WRITE 0x02000000 222 =========================== ========== 223mode=%s Control block allocation mode which supports "adaptive" 224 and "lfs". In "lfs" mode, there should be no random 225 writes towards main area. 226 "fragment:segment" and "fragment:block" are newly added here. 227 These are developer options for experiments to simulate filesystem 228 fragmentation/after-GC situation itself. The developers use these 229 modes to understand filesystem fragmentation/after-GC condition well, 230 and eventually get some insights to handle them better. 231 In "fragment:segment", f2fs allocates a new segment in random 232 position. With this, we can simulate the after-GC condition. 233 In "fragment:block", we can scatter block allocation with 234 "max_fragment_chunk" and "max_fragment_hole" sysfs nodes. 235 We added some randomness to both chunk and hole size to make 236 it close to realistic IO pattern. So, in this mode, f2fs will allocate 237 1..<max_fragment_chunk> blocks in a chunk and make a hole in the 238 length of 1..<max_fragment_hole> by turns. With this, the newly 239 allocated blocks will be scattered throughout the whole partition. 240 Note that "fragment:block" implicitly enables "fragment:segment" 241 option for more randomness. 242 Please, use these options for your experiments and we strongly 243 recommend to re-format the filesystem after using these options. 244usrquota Enable plain user disk quota accounting. 245grpquota Enable plain group disk quota accounting. 246prjquota Enable plain project quota accounting. 247usrjquota=<file> Appoint specified file and type during mount, so that quota 248grpjquota=<file> information can be properly updated during recovery flow, 249prjjquota=<file> <quota file>: must be in root directory; 250jqfmt=<quota type> <quota type>: [vfsold,vfsv0,vfsv1]. 251usrjquota= Turn off user journalled quota. 252grpjquota= Turn off group journalled quota. 253prjjquota= Turn off project journalled quota. 254quota Enable plain user disk quota accounting. 255noquota Disable all plain disk quota option. 256alloc_mode=%s Adjust block allocation policy, which supports "reuse" 257 and "default". 258fsync_mode=%s Control the policy of fsync. Currently supports "posix", 259 "strict", and "nobarrier". In "posix" mode, which is 260 default, fsync will follow POSIX semantics and does a 261 light operation to improve the filesystem performance. 262 In "strict" mode, fsync will be heavy and behaves in line 263 with xfs, ext4 and btrfs, where xfstest generic/342 will 264 pass, but the performance will regress. "nobarrier" is 265 based on "posix", but doesn't issue flush command for 266 non-atomic files likewise "nobarrier" mount option. 267test_dummy_encryption 268test_dummy_encryption=%s 269 Enable dummy encryption, which provides a fake fscrypt 270 context. The fake fscrypt context is used by xfstests. 271 The argument may be either "v1" or "v2", in order to 272 select the corresponding fscrypt policy version. 273checkpoint=%s[:%u[%]] Set to "disable" to turn off checkpointing. Set to "enable" 274 to re-enable checkpointing. Is enabled by default. While 275 disabled, any unmounting or unexpected shutdowns will cause 276 the filesystem contents to appear as they did when the 277 filesystem was mounted with that option. 278 While mounting with checkpoint=disable, the filesystem must 279 run garbage collection to ensure that all available space can 280 be used. If this takes too much time, the mount may return 281 EAGAIN. You may optionally add a value to indicate how much 282 of the disk you would be willing to temporarily give up to 283 avoid additional garbage collection. This can be given as a 284 number of blocks, or as a percent. For instance, mounting 285 with checkpoint=disable:100% would always succeed, but it may 286 hide up to all remaining free space. The actual space that 287 would be unusable can be viewed at /sys/fs/f2fs/<disk>/unusable 288 This space is reclaimed once checkpoint=enable. 289checkpoint_merge When checkpoint is enabled, this can be used to create a kernel 290 daemon and make it to merge concurrent checkpoint requests as 291 much as possible to eliminate redundant checkpoint issues. Plus, 292 we can eliminate the sluggish issue caused by slow checkpoint 293 operation when the checkpoint is done in a process context in 294 a cgroup having low i/o budget and cpu shares. To make this 295 do better, we set the default i/o priority of the kernel daemon 296 to "3", to give one higher priority than other kernel threads. 297 This is the same way to give a I/O priority to the jbd2 298 journaling thread of ext4 filesystem. 299nocheckpoint_merge Disable checkpoint merge feature. 300compress_algorithm=%s Control compress algorithm, currently f2fs supports "lzo", 301 "lz4", "zstd" and "lzo-rle" algorithm. 302compress_algorithm=%s:%d Control compress algorithm and its compress level, now, only 303 "lz4" and "zstd" support compress level config:: 304 305 ========= =========== 306 algorithm level range 307 ========= =========== 308 lz4 3 - 16 309 zstd 1 - 22 310 ========= =========== 311 312compress_log_size=%u Support configuring compress cluster size. The size will 313 be 4KB * (1 << %u). The default and minimum sizes are 16KB. 314compress_extension=%s Support adding specified extension, so that f2fs can enable 315 compression on those corresponding files, e.g. if all files 316 with '.ext' has high compression rate, we can set the '.ext' 317 on compression extension list and enable compression on 318 these file by default rather than to enable it via ioctl. 319 For other files, we can still enable compression via ioctl. 320 Note that, there is one reserved special extension '*', it 321 can be set to enable compression for all files. 322nocompress_extension=%s Support adding specified extension, so that f2fs can disable 323 compression on those corresponding files, just contrary to compression extension. 324 If you know exactly which files cannot be compressed, you can use this. 325 The same extension name can't appear in both compress and nocompress 326 extension at the same time. 327 If the compress extension specifies all files, the types specified by the 328 nocompress extension will be treated as special cases and will not be compressed. 329 Don't allow use '*' to specifie all file in nocompress extension. 330 After add nocompress_extension, the priority should be: 331 dir_flag < comp_extention,nocompress_extension < comp_file_flag,no_comp_file_flag. 332 See more in compression sections. 333 334compress_chksum Support verifying chksum of raw data in compressed cluster. 335compress_mode=%s Control file compression mode. This supports "fs" and "user" 336 modes. In "fs" mode (default), f2fs does automatic compression 337 on the compression enabled files. In "user" mode, f2fs disables 338 the automaic compression and gives the user discretion of 339 choosing the target file and the timing. The user can do manual 340 compression/decompression on the compression enabled files using 341 ioctls. 342compress_cache Support to use address space of a filesystem managed inode to 343 cache compressed block, in order to improve cache hit ratio of 344 random read. 345inlinecrypt When possible, encrypt/decrypt the contents of encrypted 346 files using the blk-crypto framework rather than 347 filesystem-layer encryption. This allows the use of 348 inline encryption hardware. The on-disk format is 349 unaffected. For more details, see 350 Documentation/block/inline-encryption.rst. 351atgc Enable age-threshold garbage collection, it provides high 352 effectiveness and efficiency on background GC. 353discard_unit=%s Control discard unit, the argument can be "block", "segment" 354 and "section", issued discard command's offset/size will be 355 aligned to the unit, by default, "discard_unit=block" is set, 356 so that small discard functionality is enabled. 357 For blkzoned device, "discard_unit=section" will be set by 358 default, it is helpful for large sized SMR or ZNS devices to 359 reduce memory cost by getting rid of fs metadata supports small 360 discard. 361memory=%s Control memory mode. This supports "normal" and "low" modes. 362 "low" mode is introduced to support low memory devices. 363 Because of the nature of low memory devices, in this mode, f2fs 364 will try to save memory sometimes by sacrificing performance. 365 "normal" mode is the default mode and same as before. 366age_extent_cache Enable an age extent cache based on rb-tree. It records 367 data block update frequency of the extent per inode, in 368 order to provide better temperature hints for data block 369 allocation. 370errors=%s Specify f2fs behavior on critical errors. This supports modes: 371 "panic", "continue" and "remount-ro", respectively, trigger 372 panic immediately, continue without doing anything, and remount 373 the partition in read-only mode. By default it uses "continue" 374 mode. 375 376 .. code-block:: none 377 378 ====================== =============== =============== ======== 379 mode continue remount-ro panic 380 ====================== =============== =============== ======== 381 access ops normal normal N/A 382 syscall errors -EIO -EROFS N/A 383 mount option rw ro N/A 384 pending dir write keep keep N/A 385 pending non-dir write drop keep N/A 386 pending node write drop keep N/A 387 pending meta write keep keep N/A 388 ====================== =============== =============== ======== 389nat_bits Enable nat_bits feature to enhance full/empty nat blocks access, 390 by default it's disabled. 391lookup_mode=%s Control the directory lookup behavior for casefolded 392 directories. This option has no effect on directories 393 that do not have the casefold feature enabled. 394 395 .. code-block:: none 396 397 ================== ======================================== 398 Value Description 399 ================== ======================================== 400 perf (Default) Enforces a hash-only lookup. 401 The linear search fallback is always 402 disabled, ignoring the on-disk flag. 403 compat Enables the linear search fallback for 404 compatibility with directory entries 405 created by older kernel that used a 406 different case-folding algorithm. 407 This mode ignores the on-disk flag. 408 auto F2FS determines the mode based on the 409 on-disk `SB_ENC_NO_COMPAT_FALLBACK_FL` 410 flag. 411 ================== ======================================== 412======================== ============================================================ 413 414Debugfs Entries 415=============== 416 417/sys/kernel/debug/f2fs/ contains information about all the partitions mounted as 418f2fs. Each file shows the whole f2fs information. 419 420/sys/kernel/debug/f2fs/status includes: 421 422 - major file system information managed by f2fs currently 423 - average SIT information about whole segments 424 - current memory footprint consumed by f2fs. 425 426Sysfs Entries 427============= 428 429Information about mounted f2fs file systems can be found in 430/sys/fs/f2fs. Each mounted filesystem will have a directory in 431/sys/fs/f2fs based on its device name (i.e., /sys/fs/f2fs/sda). 432The files in each per-device directory are shown in table below. 433 434Files in /sys/fs/f2fs/<devname> 435(see also Documentation/ABI/testing/sysfs-fs-f2fs) 436 437Usage 438===== 439 4401. Download userland tools and compile them. 441 4422. Skip, if f2fs was compiled statically inside kernel. 443 Otherwise, insert the f2fs.ko module:: 444 445 # insmod f2fs.ko 446 4473. Create a directory to use when mounting:: 448 449 # mkdir /mnt/f2fs 450 4514. Format the block device, and then mount as f2fs:: 452 453 # mkfs.f2fs -l label /dev/block_device 454 # mount -t f2fs /dev/block_device /mnt/f2fs 455 456mkfs.f2fs 457--------- 458The mkfs.f2fs is for the use of formatting a partition as the f2fs filesystem, 459which builds a basic on-disk layout. 460 461The quick options consist of: 462 463=============== =========================================================== 464``-l [label]`` Give a volume label, up to 512 unicode name. 465``-a [0 or 1]`` Split start location of each area for heap-based allocation. 466 467 1 is set by default, which performs this. 468``-o [int]`` Set overprovision ratio in percent over volume size. 469 470 5 is set by default. 471``-s [int]`` Set the number of segments per section. 472 473 1 is set by default. 474``-z [int]`` Set the number of sections per zone. 475 476 1 is set by default. 477``-e [str]`` Set basic extension list. e.g. "mp3,gif,mov" 478``-t [0 or 1]`` Disable discard command or not. 479 480 1 is set by default, which conducts discard. 481=============== =========================================================== 482 483Note: please refer to the manpage of mkfs.f2fs(8) to get full option list. 484 485fsck.f2fs 486--------- 487The fsck.f2fs is a tool to check the consistency of an f2fs-formatted 488partition, which examines whether the filesystem metadata and user-made data 489are cross-referenced correctly or not. 490Note that, initial version of the tool does not fix any inconsistency. 491 492The quick options consist of:: 493 494 -d debug level [default:0] 495 496Note: please refer to the manpage of fsck.f2fs(8) to get full option list. 497 498dump.f2fs 499--------- 500The dump.f2fs shows the information of specific inode and dumps SSA and SIT to 501file. Each file is dump_ssa and dump_sit. 502 503The dump.f2fs is used to debug on-disk data structures of the f2fs filesystem. 504It shows on-disk inode information recognized by a given inode number, and is 505able to dump all the SSA and SIT entries into predefined files, ./dump_ssa and 506./dump_sit respectively. 507 508The options consist of:: 509 510 -d debug level [default:0] 511 -i inode no (hex) 512 -s [SIT dump segno from #1~#2 (decimal), for all 0~-1] 513 -a [SSA dump segno from #1~#2 (decimal), for all 0~-1] 514 515Examples:: 516 517 # dump.f2fs -i [ino] /dev/sdx 518 # dump.f2fs -s 0~-1 /dev/sdx (SIT dump) 519 # dump.f2fs -a 0~-1 /dev/sdx (SSA dump) 520 521Note: please refer to the manpage of dump.f2fs(8) to get full option list. 522 523sload.f2fs 524---------- 525The sload.f2fs gives a way to insert files and directories in the existing disk 526image. This tool is useful when building f2fs images given compiled files. 527 528Note: please refer to the manpage of sload.f2fs(8) to get full option list. 529 530resize.f2fs 531----------- 532The resize.f2fs lets a user resize the f2fs-formatted disk image, while preserving 533all the files and directories stored in the image. 534 535Note: please refer to the manpage of resize.f2fs(8) to get full option list. 536 537defrag.f2fs 538----------- 539The defrag.f2fs can be used to defragment scattered written data as well as 540filesystem metadata across the disk. This can improve the write speed by giving 541more free consecutive space. 542 543Note: please refer to the manpage of defrag.f2fs(8) to get full option list. 544 545f2fs_io 546------- 547The f2fs_io is a simple tool to issue various filesystem APIs as well as 548f2fs-specific ones, which is very useful for QA tests. 549 550Note: please refer to the manpage of f2fs_io(8) to get full option list. 551 552Design 553====== 554 555On-disk Layout 556-------------- 557 558F2FS divides the whole volume into a number of segments, each of which is fixed 559to 2MB in size. A section is composed of consecutive segments, and a zone 560consists of a set of sections. By default, section and zone sizes are set to one 561segment size identically, but users can easily modify the sizes by mkfs. 562 563F2FS splits the entire volume into six areas, and all the areas except superblock 564consist of multiple segments as described below:: 565 566 align with the zone size <-| 567 |-> align with the segment size 568 _________________________________________________________________________ 569 | | | Segment | Node | Segment | | 570 | Superblock | Checkpoint | Info. | Address | Summary | Main | 571 | (SB) | (CP) | Table (SIT) | Table (NAT) | Area (SSA) | | 572 |____________|_____2______|______N______|______N______|______N_____|__N___| 573 . . 574 . . 575 . . 576 ._________________________________________. 577 |_Segment_|_..._|_Segment_|_..._|_Segment_| 578 . . 579 ._________._________ 580 |_section_|__...__|_ 581 . . 582 .________. 583 |__zone__| 584 585- Superblock (SB) 586 It is located at the beginning of the partition, and there exist two copies 587 to avoid file system crash. It contains basic partition information and some 588 default parameters of f2fs. 589 590- Checkpoint (CP) 591 It contains file system information, bitmaps for valid NAT/SIT sets, orphan 592 inode lists, and summary entries of current active segments. 593 594- Segment Information Table (SIT) 595 It contains segment information such as valid block count and bitmap for the 596 validity of all the blocks. 597 598- Node Address Table (NAT) 599 It is composed of a block address table for all the node blocks stored in 600 Main area. 601 602- Segment Summary Area (SSA) 603 It contains summary entries which contains the owner information of all the 604 data and node blocks stored in Main area. 605 606- Main Area 607 It contains file and directory data including their indices. 608 609In order to avoid misalignment between file system and flash-based storage, F2FS 610aligns the start block address of CP with the segment size. Also, it aligns the 611start block address of Main area with the zone size by reserving some segments 612in SSA area. 613 614Reference the following survey for additional technical details. 615https://wiki.linaro.org/WorkingGroups/Kernel/Projects/FlashCardSurvey 616 617File System Metadata Structure 618------------------------------ 619 620F2FS adopts the checkpointing scheme to maintain file system consistency. At 621mount time, F2FS first tries to find the last valid checkpoint data by scanning 622CP area. In order to reduce the scanning time, F2FS uses only two copies of CP. 623One of them always indicates the last valid data, which is called as shadow copy 624mechanism. In addition to CP, NAT and SIT also adopt the shadow copy mechanism. 625 626For file system consistency, each CP points to which NAT and SIT copies are 627valid, as shown as below:: 628 629 +--------+----------+---------+ 630 | CP | SIT | NAT | 631 +--------+----------+---------+ 632 . . . . 633 . . . . 634 . . . . 635 +-------+-------+--------+--------+--------+--------+ 636 | CP #0 | CP #1 | SIT #0 | SIT #1 | NAT #0 | NAT #1 | 637 +-------+-------+--------+--------+--------+--------+ 638 | ^ ^ 639 | | | 640 `----------------------------------------' 641 642Index Structure 643--------------- 644 645The key data structure to manage the data locations is a "node". Similar to 646traditional file structures, F2FS has three types of node: inode, direct node, 647indirect node. F2FS assigns 4KB to an inode block which contains 923 data block 648indices, two direct node pointers, two indirect node pointers, and one double 649indirect node pointer as described below. One direct node block contains 1018 650data blocks, and one indirect node block contains also 1018 node blocks. Thus, 651one inode block (i.e., a file) covers:: 652 653 4KB * (923 + 2 * 1018 + 2 * 1018 * 1018 + 1018 * 1018 * 1018) := 3.94TB. 654 655 Inode block (4KB) 656 |- data (923) 657 |- direct node (2) 658 | `- data (1018) 659 |- indirect node (2) 660 | `- direct node (1018) 661 | `- data (1018) 662 `- double indirect node (1) 663 `- indirect node (1018) 664 `- direct node (1018) 665 `- data (1018) 666 667Note that all the node blocks are mapped by NAT which means the location of 668each node is translated by the NAT table. In the consideration of the wandering 669tree problem, F2FS is able to cut off the propagation of node updates caused by 670leaf data writes. 671 672Directory Structure 673------------------- 674 675A directory entry occupies 11 bytes, which consists of the following attributes. 676 677- hash hash value of the file name 678- ino inode number 679- len the length of file name 680- type file type such as directory, symlink, etc 681 682A dentry block consists of 214 dentry slots and file names. Therein a bitmap is 683used to represent whether each dentry is valid or not. A dentry block occupies 6844KB with the following composition. 685 686:: 687 688 Dentry Block(4 K) = bitmap (27 bytes) + reserved (3 bytes) + 689 dentries(11 * 214 bytes) + file name (8 * 214 bytes) 690 691 [Bucket] 692 +--------------------------------+ 693 |dentry block 1 | dentry block 2 | 694 +--------------------------------+ 695 . . 696 . . 697 . [Dentry Block Structure: 4KB] . 698 +--------+----------+----------+------------+ 699 | bitmap | reserved | dentries | file names | 700 +--------+----------+----------+------------+ 701 [Dentry Block: 4KB] . . 702 . . 703 . . 704 +------+------+-----+------+ 705 | hash | ino | len | type | 706 +------+------+-----+------+ 707 [Dentry Structure: 11 bytes] 708 709F2FS implements multi-level hash tables for directory structure. Each level has 710a hash table with dedicated number of hash buckets as shown below. Note that 711"A(2B)" means a bucket includes 2 data blocks. 712 713:: 714 715 ---------------------- 716 A : bucket 717 B : block 718 N : MAX_DIR_HASH_DEPTH 719 ---------------------- 720 721 level #0 | A(2B) 722 | 723 level #1 | A(2B) - A(2B) 724 | 725 level #2 | A(2B) - A(2B) - A(2B) - A(2B) 726 . | . . . . 727 level #N/2 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) 728 . | . . . . 729 level #N | A(4B) - A(4B) - A(4B) - A(4B) - A(4B) - ... - A(4B) 730 731The number of blocks and buckets are determined by:: 732 733 ,- 2, if n < MAX_DIR_HASH_DEPTH / 2, 734 # of blocks in level #n = | 735 `- 4, Otherwise 736 737 ,- 2^(n + dir_level), 738 | if n + dir_level < MAX_DIR_HASH_DEPTH / 2, 739 # of buckets in level #n = | 740 `- 2^((MAX_DIR_HASH_DEPTH / 2) - 1), 741 Otherwise 742 743When F2FS finds a file name in a directory, at first a hash value of the file 744name is calculated. Then, F2FS scans the hash table in level #0 to find the 745dentry consisting of the file name and its inode number. If not found, F2FS 746scans the next hash table in level #1. In this way, F2FS scans hash tables in 747each levels incrementally from 1 to N. In each level F2FS needs to scan only 748one bucket determined by the following equation, which shows O(log(# of files)) 749complexity:: 750 751 bucket number to scan in level #n = (hash value) % (# of buckets in level #n) 752 753In the case of file creation, F2FS finds empty consecutive slots that cover the 754file name. F2FS searches the empty slots in the hash tables of whole levels from 7551 to N in the same way as the lookup operation. 756 757The following figure shows an example of two cases holding children:: 758 759 --------------> Dir <-------------- 760 | | 761 child child 762 763 child - child [hole] - child 764 765 child - child - child [hole] - [hole] - child 766 767 Case 1: Case 2: 768 Number of children = 6, Number of children = 3, 769 File size = 7 File size = 7 770 771Default Block Allocation 772------------------------ 773 774At runtime, F2FS manages six active logs inside "Main" area: Hot/Warm/Cold node 775and Hot/Warm/Cold data. 776 777- Hot node contains direct node blocks of directories. 778- Warm node contains direct node blocks except hot node blocks. 779- Cold node contains indirect node blocks 780- Hot data contains dentry blocks 781- Warm data contains data blocks except hot and cold data blocks 782- Cold data contains multimedia data or migrated data blocks 783 784LFS has two schemes for free space management: threaded log and copy-and-compac- 785tion. The copy-and-compaction scheme which is known as cleaning, is well-suited 786for devices showing very good sequential write performance, since free segments 787are served all the time for writing new data. However, it suffers from cleaning 788overhead under high utilization. Contrarily, the threaded log scheme suffers 789from random writes, but no cleaning process is needed. F2FS adopts a hybrid 790scheme where the copy-and-compaction scheme is adopted by default, but the 791policy is dynamically changed to the threaded log scheme according to the file 792system status. 793 794In order to align F2FS with underlying flash-based storage, F2FS allocates a 795segment in a unit of section. F2FS expects that the section size would be the 796same as the unit size of garbage collection in FTL. Furthermore, with respect 797to the mapping granularity in FTL, F2FS allocates each section of the active 798logs from different zones as much as possible, since FTL can write the data in 799the active logs into one allocation unit according to its mapping granularity. 800 801Cleaning process 802---------------- 803 804F2FS does cleaning both on demand and in the background. On-demand cleaning is 805triggered when there are not enough free segments to serve VFS calls. Background 806cleaner is operated by a kernel thread, and triggers the cleaning job when the 807system is idle. 808 809F2FS supports two victim selection policies: greedy and cost-benefit algorithms. 810In the greedy algorithm, F2FS selects a victim segment having the smallest number 811of valid blocks. In the cost-benefit algorithm, F2FS selects a victim segment 812according to the segment age and the number of valid blocks in order to address 813log block thrashing problem in the greedy algorithm. F2FS adopts the greedy 814algorithm for on-demand cleaner, while background cleaner adopts cost-benefit 815algorithm. 816 817In order to identify whether the data in the victim segment are valid or not, 818F2FS manages a bitmap. Each bit represents the validity of a block, and the 819bitmap is composed of a bit stream covering whole blocks in main area. 820 821Write-hint Policy 822----------------- 823 824F2FS sets the whint all the time with the below policy. 825 826===================== ======================== =================== 827User F2FS Block 828===================== ======================== =================== 829N/A META WRITE_LIFE_NONE|REQ_META 830N/A HOT_NODE WRITE_LIFE_NONE 831N/A WARM_NODE WRITE_LIFE_MEDIUM 832N/A COLD_NODE WRITE_LIFE_LONG 833ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME 834extension list " " 835 836-- buffered io 837------------------------------------------------------------------ 838N/A COLD_DATA WRITE_LIFE_EXTREME 839N/A HOT_DATA WRITE_LIFE_SHORT 840N/A WARM_DATA WRITE_LIFE_NOT_SET 841 842-- direct io 843------------------------------------------------------------------ 844WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME 845WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT 846WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET 847WRITE_LIFE_NONE " WRITE_LIFE_NONE 848WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM 849WRITE_LIFE_LONG " WRITE_LIFE_LONG 850===================== ======================== =================== 851 852Fallocate(2) Policy 853------------------- 854 855The default policy follows the below POSIX rule. 856 857Allocating disk space 858 The default operation (i.e., mode is zero) of fallocate() allocates 859 the disk space within the range specified by offset and len. The 860 file size (as reported by stat(2)) will be changed if offset+len is 861 greater than the file size. Any subregion within the range specified 862 by offset and len that did not contain data before the call will be 863 initialized to zero. This default behavior closely resembles the 864 behavior of the posix_fallocate(3) library function, and is intended 865 as a method of optimally implementing that function. 866 867However, once F2FS receives ioctl(fd, F2FS_IOC_SET_PIN_FILE) in prior to 868fallocate(fd, DEFAULT_MODE), it allocates on-disk block addresses having 869zero or random data, which is useful to the below scenario where: 870 871 1. create(fd) 872 2. ioctl(fd, F2FS_IOC_SET_PIN_FILE) 873 3. fallocate(fd, 0, 0, size) 874 4. address = fibmap(fd, offset) 875 5. open(blkdev) 876 6. write(blkdev, address) 877 878Compression implementation 879-------------------------- 880 881- New term named cluster is defined as basic unit of compression, file can 882 be divided into multiple clusters logically. One cluster includes 4 << n 883 (n >= 0) logical pages, compression size is also cluster size, each of 884 cluster can be compressed or not. 885 886- In cluster metadata layout, one special block address is used to indicate 887 a cluster is a compressed one or normal one; for compressed cluster, following 888 metadata maps cluster to [1, 4 << n - 1] physical blocks, in where f2fs 889 stores data including compress header and compressed data. 890 891- In order to eliminate write amplification during overwrite, F2FS only 892 support compression on write-once file, data can be compressed only when 893 all logical blocks in cluster contain valid data and compress ratio of 894 cluster data is lower than specified threshold. 895 896- To enable compression on regular inode, there are four ways: 897 898 * chattr +c file 899 * chattr +c dir; touch dir/file 900 * mount w/ -o compress_extension=ext; touch file.ext 901 * mount w/ -o compress_extension=*; touch any_file 902 903- To disable compression on regular inode, there are two ways: 904 905 * chattr -c file 906 * mount w/ -o nocompress_extension=ext; touch file.ext 907 908- Priority in between FS_COMPR_FL, FS_NOCOMP_FS, extensions: 909 910 * compress_extension=so; nocompress_extension=zip; chattr +c dir; touch 911 dir/foo.so; touch dir/bar.zip; touch dir/baz.txt; then foo.so and baz.txt 912 should be compresse, bar.zip should be non-compressed. chattr +c dir/bar.zip 913 can enable compress on bar.zip. 914 * compress_extension=so; nocompress_extension=zip; chattr -c dir; touch 915 dir/foo.so; touch dir/bar.zip; touch dir/baz.txt; then foo.so should be 916 compresse, bar.zip and baz.txt should be non-compressed. 917 chattr+c dir/bar.zip; chattr+c dir/baz.txt; can enable compress on bar.zip 918 and baz.txt. 919 920- At this point, compression feature doesn't expose compressed space to user 921 directly in order to guarantee potential data updates later to the space. 922 Instead, the main goal is to reduce data writes to flash disk as much as 923 possible, resulting in extending disk life time as well as relaxing IO 924 congestion. Alternatively, we've added ioctl(F2FS_IOC_RELEASE_COMPRESS_BLOCKS) 925 interface to reclaim compressed space and show it to user after setting a 926 special flag to the inode. Once the compressed space is released, the flag 927 will block writing data to the file until either the compressed space is 928 reserved via ioctl(F2FS_IOC_RESERVE_COMPRESS_BLOCKS) or the file size is 929 truncated to zero. 930 931Compress metadata layout:: 932 933 [Dnode Structure] 934 +-----------------------------------------------+ 935 | cluster 1 | cluster 2 | ......... | cluster N | 936 +-----------------------------------------------+ 937 . . . . 938 . . . . 939 . Compressed Cluster . . Normal Cluster . 940 +----------+---------+---------+---------+ +---------+---------+---------+---------+ 941 |compr flag| block 1 | block 2 | block 3 | | block 1 | block 2 | block 3 | block 4 | 942 +----------+---------+---------+---------+ +---------+---------+---------+---------+ 943 . . 944 . . 945 . . 946 +-------------+-------------+----------+----------------------------+ 947 | data length | data chksum | reserved | compressed data | 948 +-------------+-------------+----------+----------------------------+ 949 950Compression mode 951-------------------------- 952 953f2fs supports "fs" and "user" compression modes with "compression_mode" mount option. 954With this option, f2fs provides a choice to select the way how to compress the 955compression enabled files (refer to "Compression implementation" section for how to 956enable compression on a regular inode). 957 9581) compress_mode=fs 959 960 This is the default option. f2fs does automatic compression in the writeback of the 961 compression enabled files. 962 9632) compress_mode=user 964 965 This disables the automatic compression and gives the user discretion of choosing the 966 target file and the timing. The user can do manual compression/decompression on the 967 compression enabled files using F2FS_IOC_DECOMPRESS_FILE and F2FS_IOC_COMPRESS_FILE 968 ioctls like the below. 969 970To decompress a file:: 971 972 fd = open(filename, O_WRONLY, 0); 973 ret = ioctl(fd, F2FS_IOC_DECOMPRESS_FILE); 974 975To compress a file:: 976 977 fd = open(filename, O_WRONLY, 0); 978 ret = ioctl(fd, F2FS_IOC_COMPRESS_FILE); 979 980NVMe Zoned Namespace devices 981---------------------------- 982 983- ZNS defines a per-zone capacity which can be equal or less than the 984 zone-size. Zone-capacity is the number of usable blocks in the zone. 985 F2FS checks if zone-capacity is less than zone-size, if it is, then any 986 segment which starts after the zone-capacity is marked as not-free in 987 the free segment bitmap at initial mount time. These segments are marked 988 as permanently used so they are not allocated for writes and 989 consequently are not needed to be garbage collected. In case the 990 zone-capacity is not aligned to default segment size(2MB), then a segment 991 can start before the zone-capacity and span across zone-capacity boundary. 992 Such spanning segments are also considered as usable segments. All blocks 993 past the zone-capacity are considered unusable in these segments. 994 995Device aliasing feature 996----------------------- 997 998f2fs can utilize a special file called a "device aliasing file." This file allows 999the entire storage device to be mapped with a single, large extent, not using 1000the usual f2fs node structures. This mapped area is pinned and primarily intended 1001for holding the space. 1002 1003Essentially, this mechanism allows a portion of the f2fs area to be temporarily 1004reserved and used by another filesystem or for different purposes. Once that 1005external usage is complete, the device aliasing file can be deleted, releasing 1006the reserved space back to F2FS for its own use. 1007 1008.. code-block:: 1009 1010 # ls /dev/vd* 1011 /dev/vdb (32GB) /dev/vdc (32GB) 1012 # mkfs.ext4 /dev/vdc 1013 # mkfs.f2fs -c /dev/vdc@vdc.file /dev/vdb 1014 # mount /dev/vdb /mnt/f2fs 1015 # ls -l /mnt/f2fs 1016 vdc.file 1017 # df -h 1018 /dev/vdb 64G 33G 32G 52% /mnt/f2fs 1019 1020 # mount -o loop /dev/vdc /mnt/ext4 1021 # df -h 1022 /dev/vdb 64G 33G 32G 52% /mnt/f2fs 1023 /dev/loop7 32G 24K 30G 1% /mnt/ext4 1024 # umount /mnt/ext4 1025 1026 # f2fs_io getflags /mnt/f2fs/vdc.file 1027 get a flag on /mnt/f2fs/vdc.file ret=0, flags=nocow(pinned),immutable 1028 # f2fs_io setflags noimmutable /mnt/f2fs/vdc.file 1029 get a flag on noimmutable ret=0, flags=800010 1030 set a flag on /mnt/f2fs/vdc.file ret=0, flags=noimmutable 1031 # rm /mnt/f2fs/vdc.file 1032 # df -h 1033 /dev/vdb 64G 753M 64G 2% /mnt/f2fs 1034 1035So, the key idea is, user can do any file operations on /dev/vdc, and 1036reclaim the space after the use, while the space is counted as /data. 1037That doesn't require modifying partition size and filesystem format. 1038 1039Per-file Read-Only Large Folio Support 1040-------------------------------------- 1041 1042F2FS implements large folio support on the read path to leverage high-order 1043page allocation for significant performance gains. To minimize code complexity, 1044this support is currently excluded from the write path, which requires handling 1045complex optimizations such as compression and block allocation modes. 1046 1047This optional feature is triggered only when a file's immutable bit is set. 1048Consequently, F2FS will return EOPNOTSUPP if a user attempts to open a cached 1049file with write permissions, even immediately after clearing the bit. Write 1050access is only restored once the cached inode is dropped. The usage flow is 1051demonstrated below: 1052 1053.. code-block:: 1054 1055 # f2fs_io setflags immutable /data/testfile_read_seq 1056 1057 /* flush and reload the inode to enable the large folio */ 1058 # sync && echo 3 > /proc/sys/vm/drop_caches 1059 1060 /* mmap(MAP_POPULATE) + mlock() */ 1061 # f2fs_io read 128 0 1024 mmap 1 0 /data/testfile_read_seq 1062 1063 /* mmap() + fadvise(POSIX_FADV_WILLNEED) + mlock() */ 1064 # f2fs_io read 128 0 1024 fadvise 1 0 /data/testfile_read_seq 1065 1066 /* mmap() + mlock2(MLOCK_ONFAULT) + madvise(MADV_POPULATE_READ) */ 1067 # f2fs_io read 128 0 1024 madvise 1 0 /data/testfile_read_seq 1068 1069 # f2fs_io clearflags immutable /data/testfile_read_seq 1070 1071 # f2fs_io write 1 0 1 zero buffered /data/testfile_read_seq 1072 Failed to open /mnt/test/test: Operation not supported 1073 1074 /* flush and reload the inode to disable the large folio */ 1075 # sync && echo 3 > /proc/sys/vm/drop_caches 1076 1077 # f2fs_io write 1 0 1 zero buffered /data/testfile_read_seq 1078 Written 4096 bytes with pattern = zero, total_time = 29 us, max_latency = 28 us 1079 1080 # rm /data/testfile_read_seq 1081