1.. SPDX-License-Identifier: GPL-2.0 2 3========================================== 4WHAT IS Flash-Friendly File System (F2FS)? 5========================================== 6 7NAND flash memory-based storage devices, such as SSD, eMMC, and SD cards, have 8been equipped on a variety systems ranging from mobile to server systems. Since 9they are known to have different characteristics from the conventional rotating 10disks, a file system, an upper layer to the storage device, should adapt to the 11changes from the sketch in the design level. 12 13F2FS is a file system exploiting NAND flash memory-based storage devices, which 14is based on Log-structured File System (LFS). The design has been focused on 15addressing the fundamental issues in LFS, which are snowball effect of wandering 16tree and high cleaning overhead. 17 18Since a NAND flash memory-based storage device shows different characteristic 19according to its internal geometry or flash memory management scheme, namely FTL, 20F2FS and its tools support various parameters not only for configuring on-disk 21layout, but also for selecting allocation and cleaning algorithms. 22 23The following git tree provides the file system formatting tool (mkfs.f2fs), 24a consistency checking tool (fsck.f2fs), and a debugging tool (dump.f2fs). 25 26- git://git.kernel.org/pub/scm/linux/kernel/git/jaegeuk/f2fs-tools.git 27 28For sending patches, please use the following mailing list: 29 30- linux-f2fs-devel@lists.sourceforge.net 31 32For reporting bugs, please use the following f2fs bug tracker link: 33 34- https://bugzilla.kernel.org/enter_bug.cgi?product=File%20System&component=f2fs 35 36Background and Design issues 37============================ 38 39Log-structured File System (LFS) 40-------------------------------- 41"A log-structured file system writes all modifications to disk sequentially in 42a log-like structure, thereby speeding up both file writing and crash recovery. 43The log is the only structure on disk; it contains indexing information so that 44files can be read back from the log efficiently. In order to maintain large free 45areas on disk for fast writing, we divide the log into segments and use a 46segment cleaner to compress the live information from heavily fragmented 47segments." from Rosenblum, M. and Ousterhout, J. K., 1992, "The design and 48implementation of a log-structured file system", ACM Trans. Computer Systems 4910, 1, 26–52. 50 51Wandering Tree Problem 52---------------------- 53In LFS, when a file data is updated and written to the end of log, its direct 54pointer block is updated due to the changed location. Then the indirect pointer 55block is also updated due to the direct pointer block update. In this manner, 56the upper index structures such as inode, inode map, and checkpoint block are 57also updated recursively. This problem is called as wandering tree problem [1], 58and in order to enhance the performance, it should eliminate or relax the update 59propagation as much as possible. 60 61[1] Bityutskiy, A. 2005. JFFS3 design issues. http://www.linux-mtd.infradead.org/ 62 63Cleaning Overhead 64----------------- 65Since LFS is based on out-of-place writes, it produces so many obsolete blocks 66scattered across the whole storage. In order to serve new empty log space, it 67needs to reclaim these obsolete blocks seamlessly to users. This job is called 68as a cleaning process. 69 70The process consists of three operations as follows. 71 721. A victim segment is selected through referencing segment usage table. 732. It loads parent index structures of all the data in the victim identified by 74 segment summary blocks. 753. It checks the cross-reference between the data and its parent index structure. 764. It moves valid data selectively. 77 78This cleaning job may cause unexpected long delays, so the most important goal 79is to hide the latencies to users. And also definitely, it should reduce the 80amount of valid data to be moved, and move them quickly as well. 81 82Key Features 83============ 84 85Flash Awareness 86--------------- 87- Enlarge the random write area for better performance, but provide the high 88 spatial locality 89- Align FS data structures to the operational units in FTL as best efforts 90 91Wandering Tree Problem 92---------------------- 93- Use a term, “node”, that represents inodes as well as various pointer blocks 94- Introduce Node Address Table (NAT) containing the locations of all the “node” 95 blocks; this will cut off the update propagation. 96 97Cleaning Overhead 98----------------- 99- Support a background cleaning process 100- Support greedy and cost-benefit algorithms for victim selection policies 101- Support multi-head logs for static/dynamic hot and cold data separation 102- Introduce adaptive logging for efficient block allocation 103 104Mount Options 105============= 106 107 108======================== ============================================================ 109background_gc=%s Turn on/off cleaning operations, namely garbage 110 collection, triggered in background when I/O subsystem is 111 idle. If background_gc=on, it will turn on the garbage 112 collection and if background_gc=off, garbage collection 113 will be turned off. If background_gc=sync, it will turn 114 on synchronous garbage collection running in background. 115 Default value for this option is on. So garbage 116 collection is on by default. 117gc_merge When background_gc is on, this option can be enabled to 118 let background GC thread to handle foreground GC requests, 119 it can eliminate the sluggish issue caused by slow foreground 120 GC operation when GC is triggered from a process with limited 121 I/O and CPU resources. 122nogc_merge Disable GC merge feature. 123disable_roll_forward Disable the roll-forward recovery routine 124norecovery Disable the roll-forward recovery routine, mounted read- 125 only (i.e., -o ro,disable_roll_forward) 126discard/nodiscard Enable/disable real-time discard in f2fs, if discard is 127 enabled, f2fs will issue discard/TRIM commands when a 128 segment is cleaned. 129no_heap Disable heap-style segment allocation which finds free 130 segments for data from the beginning of main area, while 131 for node from the end of main area. 132nouser_xattr Disable Extended User Attributes. Note: xattr is enabled 133 by default if CONFIG_F2FS_FS_XATTR is selected. 134noacl Disable POSIX Access Control List. Note: acl is enabled 135 by default if CONFIG_F2FS_FS_POSIX_ACL is selected. 136active_logs=%u Support configuring the number of active logs. In the 137 current design, f2fs supports only 2, 4, and 6 logs. 138 Default number is 6. 139disable_ext_identify Disable the extension list configured by mkfs, so f2fs 140 is not aware of cold files such as media files. 141inline_xattr Enable the inline xattrs feature. 142noinline_xattr Disable the inline xattrs feature. 143inline_xattr_size=%u Support configuring inline xattr size, it depends on 144 flexible inline xattr feature. 145inline_data Enable the inline data feature: Newly created small (<~3.4k) 146 files can be written into inode block. 147inline_dentry Enable the inline dir feature: data in newly created 148 directory entries can be written into inode block. The 149 space of inode block which is used to store inline 150 dentries is limited to ~3.4k. 151noinline_dentry Disable the inline dentry feature. 152flush_merge Merge concurrent cache_flush commands as much as possible 153 to eliminate redundant command issues. If the underlying 154 device handles the cache_flush command relatively slowly, 155 recommend to enable this option. 156nobarrier This option can be used if underlying storage guarantees 157 its cached data should be written to the novolatile area. 158 If this option is set, no cache_flush commands are issued 159 but f2fs still guarantees the write ordering of all the 160 data writes. 161barrier If this option is set, cache_flush commands are allowed to be 162 issued. 163fastboot This option is used when a system wants to reduce mount 164 time as much as possible, even though normal performance 165 can be sacrificed. 166extent_cache Enable an extent cache based on rb-tree, it can cache 167 as many as extent which map between contiguous logical 168 address and physical address per inode, resulting in 169 increasing the cache hit ratio. Set by default. 170noextent_cache Disable an extent cache based on rb-tree explicitly, see 171 the above extent_cache mount option. 172noinline_data Disable the inline data feature, inline data feature is 173 enabled by default. 174data_flush Enable data flushing before checkpoint in order to 175 persist data of regular and symlink. 176reserve_root=%d Support configuring reserved space which is used for 177 allocation from a privileged user with specified uid or 178 gid, unit: 4KB, the default limit is 0.2% of user blocks. 179resuid=%d The user ID which may use the reserved blocks. 180resgid=%d The group ID which may use the reserved blocks. 181fault_injection=%d Enable fault injection in all supported types with 182 specified injection rate. 183fault_type=%d Support configuring fault injection type, should be 184 enabled with fault_injection option, fault type value 185 is shown below, it supports single or combined type. 186 187 =================== =========== 188 Type_Name Type_Value 189 =================== =========== 190 FAULT_KMALLOC 0x000000001 191 FAULT_KVMALLOC 0x000000002 192 FAULT_PAGE_ALLOC 0x000000004 193 FAULT_PAGE_GET 0x000000008 194 FAULT_ALLOC_BIO 0x000000010 (obsolete) 195 FAULT_ALLOC_NID 0x000000020 196 FAULT_ORPHAN 0x000000040 197 FAULT_BLOCK 0x000000080 198 FAULT_DIR_DEPTH 0x000000100 199 FAULT_EVICT_INODE 0x000000200 200 FAULT_TRUNCATE 0x000000400 201 FAULT_READ_IO 0x000000800 202 FAULT_CHECKPOINT 0x000001000 203 FAULT_DISCARD 0x000002000 204 FAULT_WRITE_IO 0x000004000 205 FAULT_SLAB_ALLOC 0x000008000 206 FAULT_DQUOT_INIT 0x000010000 207 FAULT_LOCK_OP 0x000020000 208 FAULT_BLKADDR 0x000040000 209 =================== =========== 210mode=%s Control block allocation mode which supports "adaptive" 211 and "lfs". In "lfs" mode, there should be no random 212 writes towards main area. 213 "fragment:segment" and "fragment:block" are newly added here. 214 These are developer options for experiments to simulate filesystem 215 fragmentation/after-GC situation itself. The developers use these 216 modes to understand filesystem fragmentation/after-GC condition well, 217 and eventually get some insights to handle them better. 218 In "fragment:segment", f2fs allocates a new segment in ramdom 219 position. With this, we can simulate the after-GC condition. 220 In "fragment:block", we can scatter block allocation with 221 "max_fragment_chunk" and "max_fragment_hole" sysfs nodes. 222 We added some randomness to both chunk and hole size to make 223 it close to realistic IO pattern. So, in this mode, f2fs will allocate 224 1..<max_fragment_chunk> blocks in a chunk and make a hole in the 225 length of 1..<max_fragment_hole> by turns. With this, the newly 226 allocated blocks will be scattered throughout the whole partition. 227 Note that "fragment:block" implicitly enables "fragment:segment" 228 option for more randomness. 229 Please, use these options for your experiments and we strongly 230 recommend to re-format the filesystem after using these options. 231io_bits=%u Set the bit size of write IO requests. It should be set 232 with "mode=lfs". 233usrquota Enable plain user disk quota accounting. 234grpquota Enable plain group disk quota accounting. 235prjquota Enable plain project quota accounting. 236usrjquota=<file> Appoint specified file and type during mount, so that quota 237grpjquota=<file> information can be properly updated during recovery flow, 238prjjquota=<file> <quota file>: must be in root directory; 239jqfmt=<quota type> <quota type>: [vfsold,vfsv0,vfsv1]. 240offusrjquota Turn off user journalled quota. 241offgrpjquota Turn off group journalled quota. 242offprjjquota Turn off project journalled quota. 243quota Enable plain user disk quota accounting. 244noquota Disable all plain disk quota option. 245alloc_mode=%s Adjust block allocation policy, which supports "reuse" 246 and "default". 247fsync_mode=%s Control the policy of fsync. Currently supports "posix", 248 "strict", and "nobarrier". In "posix" mode, which is 249 default, fsync will follow POSIX semantics and does a 250 light operation to improve the filesystem performance. 251 In "strict" mode, fsync will be heavy and behaves in line 252 with xfs, ext4 and btrfs, where xfstest generic/342 will 253 pass, but the performance will regress. "nobarrier" is 254 based on "posix", but doesn't issue flush command for 255 non-atomic files likewise "nobarrier" mount option. 256test_dummy_encryption 257test_dummy_encryption=%s 258 Enable dummy encryption, which provides a fake fscrypt 259 context. The fake fscrypt context is used by xfstests. 260 The argument may be either "v1" or "v2", in order to 261 select the corresponding fscrypt policy version. 262checkpoint=%s[:%u[%]] Set to "disable" to turn off checkpointing. Set to "enable" 263 to reenable checkpointing. Is enabled by default. While 264 disabled, any unmounting or unexpected shutdowns will cause 265 the filesystem contents to appear as they did when the 266 filesystem was mounted with that option. 267 While mounting with checkpoint=disable, the filesystem must 268 run garbage collection to ensure that all available space can 269 be used. If this takes too much time, the mount may return 270 EAGAIN. You may optionally add a value to indicate how much 271 of the disk you would be willing to temporarily give up to 272 avoid additional garbage collection. This can be given as a 273 number of blocks, or as a percent. For instance, mounting 274 with checkpoint=disable:100% would always succeed, but it may 275 hide up to all remaining free space. The actual space that 276 would be unusable can be viewed at /sys/fs/f2fs/<disk>/unusable 277 This space is reclaimed once checkpoint=enable. 278checkpoint_merge When checkpoint is enabled, this can be used to create a kernel 279 daemon and make it to merge concurrent checkpoint requests as 280 much as possible to eliminate redundant checkpoint issues. Plus, 281 we can eliminate the sluggish issue caused by slow checkpoint 282 operation when the checkpoint is done in a process context in 283 a cgroup having low i/o budget and cpu shares. To make this 284 do better, we set the default i/o priority of the kernel daemon 285 to "3", to give one higher priority than other kernel threads. 286 This is the same way to give a I/O priority to the jbd2 287 journaling thread of ext4 filesystem. 288nocheckpoint_merge Disable checkpoint merge feature. 289compress_algorithm=%s Control compress algorithm, currently f2fs supports "lzo", 290 "lz4", "zstd" and "lzo-rle" algorithm. 291compress_algorithm=%s:%d Control compress algorithm and its compress level, now, only 292 "lz4" and "zstd" support compress level config. 293 algorithm level range 294 lz4 3 - 16 295 zstd 1 - 22 296compress_log_size=%u Support configuring compress cluster size. The size will 297 be 4KB * (1 << %u). The default and minimum sizes are 16KB. 298compress_extension=%s Support adding specified extension, so that f2fs can enable 299 compression on those corresponding files, e.g. if all files 300 with '.ext' has high compression rate, we can set the '.ext' 301 on compression extension list and enable compression on 302 these file by default rather than to enable it via ioctl. 303 For other files, we can still enable compression via ioctl. 304 Note that, there is one reserved special extension '*', it 305 can be set to enable compression for all files. 306nocompress_extension=%s Support adding specified extension, so that f2fs can disable 307 compression on those corresponding files, just contrary to compression extension. 308 If you know exactly which files cannot be compressed, you can use this. 309 The same extension name can't appear in both compress and nocompress 310 extension at the same time. 311 If the compress extension specifies all files, the types specified by the 312 nocompress extension will be treated as special cases and will not be compressed. 313 Don't allow use '*' to specifie all file in nocompress extension. 314 After add nocompress_extension, the priority should be: 315 dir_flag < comp_extention,nocompress_extension < comp_file_flag,no_comp_file_flag. 316 See more in compression sections. 317 318compress_chksum Support verifying chksum of raw data in compressed cluster. 319compress_mode=%s Control file compression mode. This supports "fs" and "user" 320 modes. In "fs" mode (default), f2fs does automatic compression 321 on the compression enabled files. In "user" mode, f2fs disables 322 the automaic compression and gives the user discretion of 323 choosing the target file and the timing. The user can do manual 324 compression/decompression on the compression enabled files using 325 ioctls. 326compress_cache Support to use address space of a filesystem managed inode to 327 cache compressed block, in order to improve cache hit ratio of 328 random read. 329inlinecrypt When possible, encrypt/decrypt the contents of encrypted 330 files using the blk-crypto framework rather than 331 filesystem-layer encryption. This allows the use of 332 inline encryption hardware. The on-disk format is 333 unaffected. For more details, see 334 Documentation/block/inline-encryption.rst. 335atgc Enable age-threshold garbage collection, it provides high 336 effectiveness and efficiency on background GC. 337discard_unit=%s Control discard unit, the argument can be "block", "segment" 338 and "section", issued discard command's offset/size will be 339 aligned to the unit, by default, "discard_unit=block" is set, 340 so that small discard functionality is enabled. 341 For blkzoned device, "discard_unit=section" will be set by 342 default, it is helpful for large sized SMR or ZNS devices to 343 reduce memory cost by getting rid of fs metadata supports small 344 discard. 345memory=%s Control memory mode. This supports "normal" and "low" modes. 346 "low" mode is introduced to support low memory devices. 347 Because of the nature of low memory devices, in this mode, f2fs 348 will try to save memory sometimes by sacrificing performance. 349 "normal" mode is the default mode and same as before. 350age_extent_cache Enable an age extent cache based on rb-tree. It records 351 data block update frequency of the extent per inode, in 352 order to provide better temperature hints for data block 353 allocation. 354errors=%s Specify f2fs behavior on critical errors. This supports modes: 355 "panic", "continue" and "remount-ro", respectively, trigger 356 panic immediately, continue without doing anything, and remount 357 the partition in read-only mode. By default it uses "continue" 358 mode. 359 ====================== =============== =============== ======== 360 mode continue remount-ro panic 361 ====================== =============== =============== ======== 362 access ops normal noraml N/A 363 syscall errors -EIO -EROFS N/A 364 mount option rw ro N/A 365 pending dir write keep keep N/A 366 pending non-dir write drop keep N/A 367 pending node write drop keep N/A 368 pending meta write keep keep N/A 369 ====================== =============== =============== ======== 370======================== ============================================================ 371 372Debugfs Entries 373=============== 374 375/sys/kernel/debug/f2fs/ contains information about all the partitions mounted as 376f2fs. Each file shows the whole f2fs information. 377 378/sys/kernel/debug/f2fs/status includes: 379 380 - major file system information managed by f2fs currently 381 - average SIT information about whole segments 382 - current memory footprint consumed by f2fs. 383 384Sysfs Entries 385============= 386 387Information about mounted f2fs file systems can be found in 388/sys/fs/f2fs. Each mounted filesystem will have a directory in 389/sys/fs/f2fs based on its device name (i.e., /sys/fs/f2fs/sda). 390The files in each per-device directory are shown in table below. 391 392Files in /sys/fs/f2fs/<devname> 393(see also Documentation/ABI/testing/sysfs-fs-f2fs) 394 395Usage 396===== 397 3981. Download userland tools and compile them. 399 4002. Skip, if f2fs was compiled statically inside kernel. 401 Otherwise, insert the f2fs.ko module:: 402 403 # insmod f2fs.ko 404 4053. Create a directory to use when mounting:: 406 407 # mkdir /mnt/f2fs 408 4094. Format the block device, and then mount as f2fs:: 410 411 # mkfs.f2fs -l label /dev/block_device 412 # mount -t f2fs /dev/block_device /mnt/f2fs 413 414mkfs.f2fs 415--------- 416The mkfs.f2fs is for the use of formatting a partition as the f2fs filesystem, 417which builds a basic on-disk layout. 418 419The quick options consist of: 420 421=============== =========================================================== 422``-l [label]`` Give a volume label, up to 512 unicode name. 423``-a [0 or 1]`` Split start location of each area for heap-based allocation. 424 425 1 is set by default, which performs this. 426``-o [int]`` Set overprovision ratio in percent over volume size. 427 428 5 is set by default. 429``-s [int]`` Set the number of segments per section. 430 431 1 is set by default. 432``-z [int]`` Set the number of sections per zone. 433 434 1 is set by default. 435``-e [str]`` Set basic extension list. e.g. "mp3,gif,mov" 436``-t [0 or 1]`` Disable discard command or not. 437 438 1 is set by default, which conducts discard. 439=============== =========================================================== 440 441Note: please refer to the manpage of mkfs.f2fs(8) to get full option list. 442 443fsck.f2fs 444--------- 445The fsck.f2fs is a tool to check the consistency of an f2fs-formatted 446partition, which examines whether the filesystem metadata and user-made data 447are cross-referenced correctly or not. 448Note that, initial version of the tool does not fix any inconsistency. 449 450The quick options consist of:: 451 452 -d debug level [default:0] 453 454Note: please refer to the manpage of fsck.f2fs(8) to get full option list. 455 456dump.f2fs 457--------- 458The dump.f2fs shows the information of specific inode and dumps SSA and SIT to 459file. Each file is dump_ssa and dump_sit. 460 461The dump.f2fs is used to debug on-disk data structures of the f2fs filesystem. 462It shows on-disk inode information recognized by a given inode number, and is 463able to dump all the SSA and SIT entries into predefined files, ./dump_ssa and 464./dump_sit respectively. 465 466The options consist of:: 467 468 -d debug level [default:0] 469 -i inode no (hex) 470 -s [SIT dump segno from #1~#2 (decimal), for all 0~-1] 471 -a [SSA dump segno from #1~#2 (decimal), for all 0~-1] 472 473Examples:: 474 475 # dump.f2fs -i [ino] /dev/sdx 476 # dump.f2fs -s 0~-1 /dev/sdx (SIT dump) 477 # dump.f2fs -a 0~-1 /dev/sdx (SSA dump) 478 479Note: please refer to the manpage of dump.f2fs(8) to get full option list. 480 481sload.f2fs 482---------- 483The sload.f2fs gives a way to insert files and directories in the exisiting disk 484image. This tool is useful when building f2fs images given compiled files. 485 486Note: please refer to the manpage of sload.f2fs(8) to get full option list. 487 488resize.f2fs 489----------- 490The resize.f2fs lets a user resize the f2fs-formatted disk image, while preserving 491all the files and directories stored in the image. 492 493Note: please refer to the manpage of resize.f2fs(8) to get full option list. 494 495defrag.f2fs 496----------- 497The defrag.f2fs can be used to defragment scattered written data as well as 498filesystem metadata across the disk. This can improve the write speed by giving 499more free consecutive space. 500 501Note: please refer to the manpage of defrag.f2fs(8) to get full option list. 502 503f2fs_io 504------- 505The f2fs_io is a simple tool to issue various filesystem APIs as well as 506f2fs-specific ones, which is very useful for QA tests. 507 508Note: please refer to the manpage of f2fs_io(8) to get full option list. 509 510Design 511====== 512 513On-disk Layout 514-------------- 515 516F2FS divides the whole volume into a number of segments, each of which is fixed 517to 2MB in size. A section is composed of consecutive segments, and a zone 518consists of a set of sections. By default, section and zone sizes are set to one 519segment size identically, but users can easily modify the sizes by mkfs. 520 521F2FS splits the entire volume into six areas, and all the areas except superblock 522consist of multiple segments as described below:: 523 524 align with the zone size <-| 525 |-> align with the segment size 526 _________________________________________________________________________ 527 | | | Segment | Node | Segment | | 528 | Superblock | Checkpoint | Info. | Address | Summary | Main | 529 | (SB) | (CP) | Table (SIT) | Table (NAT) | Area (SSA) | | 530 |____________|_____2______|______N______|______N______|______N_____|__N___| 531 . . 532 . . 533 . . 534 ._________________________________________. 535 |_Segment_|_..._|_Segment_|_..._|_Segment_| 536 . . 537 ._________._________ 538 |_section_|__...__|_ 539 . . 540 .________. 541 |__zone__| 542 543- Superblock (SB) 544 It is located at the beginning of the partition, and there exist two copies 545 to avoid file system crash. It contains basic partition information and some 546 default parameters of f2fs. 547 548- Checkpoint (CP) 549 It contains file system information, bitmaps for valid NAT/SIT sets, orphan 550 inode lists, and summary entries of current active segments. 551 552- Segment Information Table (SIT) 553 It contains segment information such as valid block count and bitmap for the 554 validity of all the blocks. 555 556- Node Address Table (NAT) 557 It is composed of a block address table for all the node blocks stored in 558 Main area. 559 560- Segment Summary Area (SSA) 561 It contains summary entries which contains the owner information of all the 562 data and node blocks stored in Main area. 563 564- Main Area 565 It contains file and directory data including their indices. 566 567In order to avoid misalignment between file system and flash-based storage, F2FS 568aligns the start block address of CP with the segment size. Also, it aligns the 569start block address of Main area with the zone size by reserving some segments 570in SSA area. 571 572Reference the following survey for additional technical details. 573https://wiki.linaro.org/WorkingGroups/Kernel/Projects/FlashCardSurvey 574 575File System Metadata Structure 576------------------------------ 577 578F2FS adopts the checkpointing scheme to maintain file system consistency. At 579mount time, F2FS first tries to find the last valid checkpoint data by scanning 580CP area. In order to reduce the scanning time, F2FS uses only two copies of CP. 581One of them always indicates the last valid data, which is called as shadow copy 582mechanism. In addition to CP, NAT and SIT also adopt the shadow copy mechanism. 583 584For file system consistency, each CP points to which NAT and SIT copies are 585valid, as shown as below:: 586 587 +--------+----------+---------+ 588 | CP | SIT | NAT | 589 +--------+----------+---------+ 590 . . . . 591 . . . . 592 . . . . 593 +-------+-------+--------+--------+--------+--------+ 594 | CP #0 | CP #1 | SIT #0 | SIT #1 | NAT #0 | NAT #1 | 595 +-------+-------+--------+--------+--------+--------+ 596 | ^ ^ 597 | | | 598 `----------------------------------------' 599 600Index Structure 601--------------- 602 603The key data structure to manage the data locations is a "node". Similar to 604traditional file structures, F2FS has three types of node: inode, direct node, 605indirect node. F2FS assigns 4KB to an inode block which contains 923 data block 606indices, two direct node pointers, two indirect node pointers, and one double 607indirect node pointer as described below. One direct node block contains 1018 608data blocks, and one indirect node block contains also 1018 node blocks. Thus, 609one inode block (i.e., a file) covers:: 610 611 4KB * (923 + 2 * 1018 + 2 * 1018 * 1018 + 1018 * 1018 * 1018) := 3.94TB. 612 613 Inode block (4KB) 614 |- data (923) 615 |- direct node (2) 616 | `- data (1018) 617 |- indirect node (2) 618 | `- direct node (1018) 619 | `- data (1018) 620 `- double indirect node (1) 621 `- indirect node (1018) 622 `- direct node (1018) 623 `- data (1018) 624 625Note that all the node blocks are mapped by NAT which means the location of 626each node is translated by the NAT table. In the consideration of the wandering 627tree problem, F2FS is able to cut off the propagation of node updates caused by 628leaf data writes. 629 630Directory Structure 631------------------- 632 633A directory entry occupies 11 bytes, which consists of the following attributes. 634 635- hash hash value of the file name 636- ino inode number 637- len the length of file name 638- type file type such as directory, symlink, etc 639 640A dentry block consists of 214 dentry slots and file names. Therein a bitmap is 641used to represent whether each dentry is valid or not. A dentry block occupies 6424KB with the following composition. 643 644:: 645 646 Dentry Block(4 K) = bitmap (27 bytes) + reserved (3 bytes) + 647 dentries(11 * 214 bytes) + file name (8 * 214 bytes) 648 649 [Bucket] 650 +--------------------------------+ 651 |dentry block 1 | dentry block 2 | 652 +--------------------------------+ 653 . . 654 . . 655 . [Dentry Block Structure: 4KB] . 656 +--------+----------+----------+------------+ 657 | bitmap | reserved | dentries | file names | 658 +--------+----------+----------+------------+ 659 [Dentry Block: 4KB] . . 660 . . 661 . . 662 +------+------+-----+------+ 663 | hash | ino | len | type | 664 +------+------+-----+------+ 665 [Dentry Structure: 11 bytes] 666 667F2FS implements multi-level hash tables for directory structure. Each level has 668a hash table with dedicated number of hash buckets as shown below. Note that 669"A(2B)" means a bucket includes 2 data blocks. 670 671:: 672 673 ---------------------- 674 A : bucket 675 B : block 676 N : MAX_DIR_HASH_DEPTH 677 ---------------------- 678 679 level #0 | A(2B) 680 | 681 level #1 | A(2B) - A(2B) 682 | 683 level #2 | A(2B) - A(2B) - A(2B) - A(2B) 684 . | . . . . 685 level #N/2 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) 686 . | . . . . 687 level #N | A(4B) - A(4B) - A(4B) - A(4B) - A(4B) - ... - A(4B) 688 689The number of blocks and buckets are determined by:: 690 691 ,- 2, if n < MAX_DIR_HASH_DEPTH / 2, 692 # of blocks in level #n = | 693 `- 4, Otherwise 694 695 ,- 2^(n + dir_level), 696 | if n + dir_level < MAX_DIR_HASH_DEPTH / 2, 697 # of buckets in level #n = | 698 `- 2^((MAX_DIR_HASH_DEPTH / 2) - 1), 699 Otherwise 700 701When F2FS finds a file name in a directory, at first a hash value of the file 702name is calculated. Then, F2FS scans the hash table in level #0 to find the 703dentry consisting of the file name and its inode number. If not found, F2FS 704scans the next hash table in level #1. In this way, F2FS scans hash tables in 705each levels incrementally from 1 to N. In each level F2FS needs to scan only 706one bucket determined by the following equation, which shows O(log(# of files)) 707complexity:: 708 709 bucket number to scan in level #n = (hash value) % (# of buckets in level #n) 710 711In the case of file creation, F2FS finds empty consecutive slots that cover the 712file name. F2FS searches the empty slots in the hash tables of whole levels from 7131 to N in the same way as the lookup operation. 714 715The following figure shows an example of two cases holding children:: 716 717 --------------> Dir <-------------- 718 | | 719 child child 720 721 child - child [hole] - child 722 723 child - child - child [hole] - [hole] - child 724 725 Case 1: Case 2: 726 Number of children = 6, Number of children = 3, 727 File size = 7 File size = 7 728 729Default Block Allocation 730------------------------ 731 732At runtime, F2FS manages six active logs inside "Main" area: Hot/Warm/Cold node 733and Hot/Warm/Cold data. 734 735- Hot node contains direct node blocks of directories. 736- Warm node contains direct node blocks except hot node blocks. 737- Cold node contains indirect node blocks 738- Hot data contains dentry blocks 739- Warm data contains data blocks except hot and cold data blocks 740- Cold data contains multimedia data or migrated data blocks 741 742LFS has two schemes for free space management: threaded log and copy-and-compac- 743tion. The copy-and-compaction scheme which is known as cleaning, is well-suited 744for devices showing very good sequential write performance, since free segments 745are served all the time for writing new data. However, it suffers from cleaning 746overhead under high utilization. Contrarily, the threaded log scheme suffers 747from random writes, but no cleaning process is needed. F2FS adopts a hybrid 748scheme where the copy-and-compaction scheme is adopted by default, but the 749policy is dynamically changed to the threaded log scheme according to the file 750system status. 751 752In order to align F2FS with underlying flash-based storage, F2FS allocates a 753segment in a unit of section. F2FS expects that the section size would be the 754same as the unit size of garbage collection in FTL. Furthermore, with respect 755to the mapping granularity in FTL, F2FS allocates each section of the active 756logs from different zones as much as possible, since FTL can write the data in 757the active logs into one allocation unit according to its mapping granularity. 758 759Cleaning process 760---------------- 761 762F2FS does cleaning both on demand and in the background. On-demand cleaning is 763triggered when there are not enough free segments to serve VFS calls. Background 764cleaner is operated by a kernel thread, and triggers the cleaning job when the 765system is idle. 766 767F2FS supports two victim selection policies: greedy and cost-benefit algorithms. 768In the greedy algorithm, F2FS selects a victim segment having the smallest number 769of valid blocks. In the cost-benefit algorithm, F2FS selects a victim segment 770according to the segment age and the number of valid blocks in order to address 771log block thrashing problem in the greedy algorithm. F2FS adopts the greedy 772algorithm for on-demand cleaner, while background cleaner adopts cost-benefit 773algorithm. 774 775In order to identify whether the data in the victim segment are valid or not, 776F2FS manages a bitmap. Each bit represents the validity of a block, and the 777bitmap is composed of a bit stream covering whole blocks in main area. 778 779Fallocate(2) Policy 780------------------- 781 782The default policy follows the below POSIX rule. 783 784Allocating disk space 785 The default operation (i.e., mode is zero) of fallocate() allocates 786 the disk space within the range specified by offset and len. The 787 file size (as reported by stat(2)) will be changed if offset+len is 788 greater than the file size. Any subregion within the range specified 789 by offset and len that did not contain data before the call will be 790 initialized to zero. This default behavior closely resembles the 791 behavior of the posix_fallocate(3) library function, and is intended 792 as a method of optimally implementing that function. 793 794However, once F2FS receives ioctl(fd, F2FS_IOC_SET_PIN_FILE) in prior to 795fallocate(fd, DEFAULT_MODE), it allocates on-disk block addressess having 796zero or random data, which is useful to the below scenario where: 797 798 1. create(fd) 799 2. ioctl(fd, F2FS_IOC_SET_PIN_FILE) 800 3. fallocate(fd, 0, 0, size) 801 4. address = fibmap(fd, offset) 802 5. open(blkdev) 803 6. write(blkdev, address) 804 805Compression implementation 806-------------------------- 807 808- New term named cluster is defined as basic unit of compression, file can 809 be divided into multiple clusters logically. One cluster includes 4 << n 810 (n >= 0) logical pages, compression size is also cluster size, each of 811 cluster can be compressed or not. 812 813- In cluster metadata layout, one special block address is used to indicate 814 a cluster is a compressed one or normal one; for compressed cluster, following 815 metadata maps cluster to [1, 4 << n - 1] physical blocks, in where f2fs 816 stores data including compress header and compressed data. 817 818- In order to eliminate write amplification during overwrite, F2FS only 819 support compression on write-once file, data can be compressed only when 820 all logical blocks in cluster contain valid data and compress ratio of 821 cluster data is lower than specified threshold. 822 823- To enable compression on regular inode, there are four ways: 824 825 * chattr +c file 826 * chattr +c dir; touch dir/file 827 * mount w/ -o compress_extension=ext; touch file.ext 828 * mount w/ -o compress_extension=*; touch any_file 829 830- To disable compression on regular inode, there are two ways: 831 832 * chattr -c file 833 * mount w/ -o nocompress_extension=ext; touch file.ext 834 835- Priority in between FS_COMPR_FL, FS_NOCOMP_FS, extensions: 836 837 * compress_extension=so; nocompress_extension=zip; chattr +c dir; touch 838 dir/foo.so; touch dir/bar.zip; touch dir/baz.txt; then foo.so and baz.txt 839 should be compresse, bar.zip should be non-compressed. chattr +c dir/bar.zip 840 can enable compress on bar.zip. 841 * compress_extension=so; nocompress_extension=zip; chattr -c dir; touch 842 dir/foo.so; touch dir/bar.zip; touch dir/baz.txt; then foo.so should be 843 compresse, bar.zip and baz.txt should be non-compressed. 844 chattr+c dir/bar.zip; chattr+c dir/baz.txt; can enable compress on bar.zip 845 and baz.txt. 846 847- At this point, compression feature doesn't expose compressed space to user 848 directly in order to guarantee potential data updates later to the space. 849 Instead, the main goal is to reduce data writes to flash disk as much as 850 possible, resulting in extending disk life time as well as relaxing IO 851 congestion. Alternatively, we've added ioctl(F2FS_IOC_RELEASE_COMPRESS_BLOCKS) 852 interface to reclaim compressed space and show it to user after setting a 853 special flag to the inode. Once the compressed space is released, the flag 854 will block writing data to the file until either the compressed space is 855 reserved via ioctl(F2FS_IOC_RESERVE_COMPRESS_BLOCKS) or the file size is 856 truncated to zero. 857 858Compress metadata layout:: 859 860 [Dnode Structure] 861 +-----------------------------------------------+ 862 | cluster 1 | cluster 2 | ......... | cluster N | 863 +-----------------------------------------------+ 864 . . . . 865 . . . . 866 . Compressed Cluster . . Normal Cluster . 867 +----------+---------+---------+---------+ +---------+---------+---------+---------+ 868 |compr flag| block 1 | block 2 | block 3 | | block 1 | block 2 | block 3 | block 4 | 869 +----------+---------+---------+---------+ +---------+---------+---------+---------+ 870 . . 871 . . 872 . . 873 +-------------+-------------+----------+----------------------------+ 874 | data length | data chksum | reserved | compressed data | 875 +-------------+-------------+----------+----------------------------+ 876 877Compression mode 878-------------------------- 879 880f2fs supports "fs" and "user" compression modes with "compression_mode" mount option. 881With this option, f2fs provides a choice to select the way how to compress the 882compression enabled files (refer to "Compression implementation" section for how to 883enable compression on a regular inode). 884 8851) compress_mode=fs 886This is the default option. f2fs does automatic compression in the writeback of the 887compression enabled files. 888 8892) compress_mode=user 890This disables the automatic compression and gives the user discretion of choosing the 891target file and the timing. The user can do manual compression/decompression on the 892compression enabled files using F2FS_IOC_DECOMPRESS_FILE and F2FS_IOC_COMPRESS_FILE 893ioctls like the below. 894 895To decompress a file, 896 897fd = open(filename, O_WRONLY, 0); 898ret = ioctl(fd, F2FS_IOC_DECOMPRESS_FILE); 899 900To compress a file, 901 902fd = open(filename, O_WRONLY, 0); 903ret = ioctl(fd, F2FS_IOC_COMPRESS_FILE); 904 905NVMe Zoned Namespace devices 906---------------------------- 907 908- ZNS defines a per-zone capacity which can be equal or less than the 909 zone-size. Zone-capacity is the number of usable blocks in the zone. 910 F2FS checks if zone-capacity is less than zone-size, if it is, then any 911 segment which starts after the zone-capacity is marked as not-free in 912 the free segment bitmap at initial mount time. These segments are marked 913 as permanently used so they are not allocated for writes and 914 consequently are not needed to be garbage collected. In case the 915 zone-capacity is not aligned to default segment size(2MB), then a segment 916 can start before the zone-capacity and span across zone-capacity boundary. 917 Such spanning segments are also considered as usable segments. All blocks 918 past the zone-capacity are considered unusable in these segments. 919