189272ca1SMauro Carvalho Chehab.. SPDX-License-Identifier: GPL-2.0 289272ca1SMauro Carvalho Chehab 389272ca1SMauro Carvalho Chehab========================================== 489272ca1SMauro Carvalho ChehabWHAT IS Flash-Friendly File System (F2FS)? 589272ca1SMauro Carvalho Chehab========================================== 689272ca1SMauro Carvalho Chehab 789272ca1SMauro Carvalho ChehabNAND flash memory-based storage devices, such as SSD, eMMC, and SD cards, have 889272ca1SMauro Carvalho Chehabbeen equipped on a variety systems ranging from mobile to server systems. Since 989272ca1SMauro Carvalho Chehabthey are known to have different characteristics from the conventional rotating 1089272ca1SMauro Carvalho Chehabdisks, a file system, an upper layer to the storage device, should adapt to the 1189272ca1SMauro Carvalho Chehabchanges from the sketch in the design level. 1289272ca1SMauro Carvalho Chehab 1389272ca1SMauro Carvalho ChehabF2FS is a file system exploiting NAND flash memory-based storage devices, which 1489272ca1SMauro Carvalho Chehabis based on Log-structured File System (LFS). The design has been focused on 1589272ca1SMauro Carvalho Chehabaddressing the fundamental issues in LFS, which are snowball effect of wandering 1689272ca1SMauro Carvalho Chehabtree and high cleaning overhead. 1789272ca1SMauro Carvalho Chehab 1889272ca1SMauro Carvalho ChehabSince a NAND flash memory-based storage device shows different characteristic 1989272ca1SMauro Carvalho Chehabaccording to its internal geometry or flash memory management scheme, namely FTL, 2089272ca1SMauro Carvalho ChehabF2FS and its tools support various parameters not only for configuring on-disk 2189272ca1SMauro Carvalho Chehablayout, but also for selecting allocation and cleaning algorithms. 2289272ca1SMauro Carvalho Chehab 2389272ca1SMauro Carvalho ChehabThe following git tree provides the file system formatting tool (mkfs.f2fs), 2489272ca1SMauro Carvalho Chehaba consistency checking tool (fsck.f2fs), and a debugging tool (dump.f2fs). 2589272ca1SMauro Carvalho Chehab 2689272ca1SMauro Carvalho Chehab- git://git.kernel.org/pub/scm/linux/kernel/git/jaegeuk/f2fs-tools.git 2789272ca1SMauro Carvalho Chehab 2889272ca1SMauro Carvalho ChehabFor reporting bugs and sending patches, please use the following mailing list: 2989272ca1SMauro Carvalho Chehab 3089272ca1SMauro Carvalho Chehab- linux-f2fs-devel@lists.sourceforge.net 3189272ca1SMauro Carvalho Chehab 3289272ca1SMauro Carvalho ChehabBackground and Design issues 3389272ca1SMauro Carvalho Chehab============================ 3489272ca1SMauro Carvalho Chehab 3589272ca1SMauro Carvalho ChehabLog-structured File System (LFS) 3689272ca1SMauro Carvalho Chehab-------------------------------- 3789272ca1SMauro Carvalho Chehab"A log-structured file system writes all modifications to disk sequentially in 3889272ca1SMauro Carvalho Chehaba log-like structure, thereby speeding up both file writing and crash recovery. 3989272ca1SMauro Carvalho ChehabThe log is the only structure on disk; it contains indexing information so that 4089272ca1SMauro Carvalho Chehabfiles can be read back from the log efficiently. In order to maintain large free 4189272ca1SMauro Carvalho Chehabareas on disk for fast writing, we divide the log into segments and use a 4289272ca1SMauro Carvalho Chehabsegment cleaner to compress the live information from heavily fragmented 4389272ca1SMauro Carvalho Chehabsegments." from Rosenblum, M. and Ousterhout, J. K., 1992, "The design and 4489272ca1SMauro Carvalho Chehabimplementation of a log-structured file system", ACM Trans. Computer Systems 4589272ca1SMauro Carvalho Chehab10, 1, 26–52. 4689272ca1SMauro Carvalho Chehab 4789272ca1SMauro Carvalho ChehabWandering Tree Problem 4889272ca1SMauro Carvalho Chehab---------------------- 4989272ca1SMauro Carvalho ChehabIn LFS, when a file data is updated and written to the end of log, its direct 5089272ca1SMauro Carvalho Chehabpointer block is updated due to the changed location. Then the indirect pointer 5189272ca1SMauro Carvalho Chehabblock is also updated due to the direct pointer block update. In this manner, 5289272ca1SMauro Carvalho Chehabthe upper index structures such as inode, inode map, and checkpoint block are 5389272ca1SMauro Carvalho Chehabalso updated recursively. This problem is called as wandering tree problem [1], 5489272ca1SMauro Carvalho Chehaband in order to enhance the performance, it should eliminate or relax the update 5589272ca1SMauro Carvalho Chehabpropagation as much as possible. 5689272ca1SMauro Carvalho Chehab 5789272ca1SMauro Carvalho Chehab[1] Bityutskiy, A. 2005. JFFS3 design issues. http://www.linux-mtd.infradead.org/ 5889272ca1SMauro Carvalho Chehab 5989272ca1SMauro Carvalho ChehabCleaning Overhead 6089272ca1SMauro Carvalho Chehab----------------- 6189272ca1SMauro Carvalho ChehabSince LFS is based on out-of-place writes, it produces so many obsolete blocks 6289272ca1SMauro Carvalho Chehabscattered across the whole storage. In order to serve new empty log space, it 6389272ca1SMauro Carvalho Chehabneeds to reclaim these obsolete blocks seamlessly to users. This job is called 6489272ca1SMauro Carvalho Chehabas a cleaning process. 6589272ca1SMauro Carvalho Chehab 6689272ca1SMauro Carvalho ChehabThe process consists of three operations as follows. 6789272ca1SMauro Carvalho Chehab 6889272ca1SMauro Carvalho Chehab1. A victim segment is selected through referencing segment usage table. 6989272ca1SMauro Carvalho Chehab2. It loads parent index structures of all the data in the victim identified by 7089272ca1SMauro Carvalho Chehab segment summary blocks. 7189272ca1SMauro Carvalho Chehab3. It checks the cross-reference between the data and its parent index structure. 7289272ca1SMauro Carvalho Chehab4. It moves valid data selectively. 7389272ca1SMauro Carvalho Chehab 7489272ca1SMauro Carvalho ChehabThis cleaning job may cause unexpected long delays, so the most important goal 7589272ca1SMauro Carvalho Chehabis to hide the latencies to users. And also definitely, it should reduce the 7689272ca1SMauro Carvalho Chehabamount of valid data to be moved, and move them quickly as well. 7789272ca1SMauro Carvalho Chehab 7889272ca1SMauro Carvalho ChehabKey Features 7989272ca1SMauro Carvalho Chehab============ 8089272ca1SMauro Carvalho Chehab 8189272ca1SMauro Carvalho ChehabFlash Awareness 8289272ca1SMauro Carvalho Chehab--------------- 8389272ca1SMauro Carvalho Chehab- Enlarge the random write area for better performance, but provide the high 8489272ca1SMauro Carvalho Chehab spatial locality 8589272ca1SMauro Carvalho Chehab- Align FS data structures to the operational units in FTL as best efforts 8689272ca1SMauro Carvalho Chehab 8789272ca1SMauro Carvalho ChehabWandering Tree Problem 8889272ca1SMauro Carvalho Chehab---------------------- 8989272ca1SMauro Carvalho Chehab- Use a term, “node”, that represents inodes as well as various pointer blocks 9089272ca1SMauro Carvalho Chehab- Introduce Node Address Table (NAT) containing the locations of all the “node” 9189272ca1SMauro Carvalho Chehab blocks; this will cut off the update propagation. 9289272ca1SMauro Carvalho Chehab 9389272ca1SMauro Carvalho ChehabCleaning Overhead 9489272ca1SMauro Carvalho Chehab----------------- 9589272ca1SMauro Carvalho Chehab- Support a background cleaning process 9689272ca1SMauro Carvalho Chehab- Support greedy and cost-benefit algorithms for victim selection policies 9789272ca1SMauro Carvalho Chehab- Support multi-head logs for static/dynamic hot and cold data separation 9889272ca1SMauro Carvalho Chehab- Introduce adaptive logging for efficient block allocation 9989272ca1SMauro Carvalho Chehab 10089272ca1SMauro Carvalho ChehabMount Options 10189272ca1SMauro Carvalho Chehab============= 10289272ca1SMauro Carvalho Chehab 10389272ca1SMauro Carvalho Chehab 1049aa1ccb4SJonathan Corbet======================== ============================================================ 10589272ca1SMauro Carvalho Chehabbackground_gc=%s Turn on/off cleaning operations, namely garbage 10689272ca1SMauro Carvalho Chehab collection, triggered in background when I/O subsystem is 10789272ca1SMauro Carvalho Chehab idle. If background_gc=on, it will turn on the garbage 10889272ca1SMauro Carvalho Chehab collection and if background_gc=off, garbage collection 10989272ca1SMauro Carvalho Chehab will be turned off. If background_gc=sync, it will turn 11089272ca1SMauro Carvalho Chehab on synchronous garbage collection running in background. 11189272ca1SMauro Carvalho Chehab Default value for this option is on. So garbage 11289272ca1SMauro Carvalho Chehab collection is on by default. 1135911d2d1SChao Yugc_merge When background_gc is on, this option can be enabled to 1145911d2d1SChao Yu let background GC thread to handle foreground GC requests, 1155911d2d1SChao Yu it can eliminate the sluggish issue caused by slow foreground 1165911d2d1SChao Yu GC operation when GC is triggered from a process with limited 1175911d2d1SChao Yu I/O and CPU resources. 1185911d2d1SChao Yunogc_merge Disable GC merge feature. 11989272ca1SMauro Carvalho Chehabdisable_roll_forward Disable the roll-forward recovery routine 12089272ca1SMauro Carvalho Chehabnorecovery Disable the roll-forward recovery routine, mounted read- 12189272ca1SMauro Carvalho Chehab only (i.e., -o ro,disable_roll_forward) 12289272ca1SMauro Carvalho Chehabdiscard/nodiscard Enable/disable real-time discard in f2fs, if discard is 12389272ca1SMauro Carvalho Chehab enabled, f2fs will issue discard/TRIM commands when a 12489272ca1SMauro Carvalho Chehab segment is cleaned. 12589272ca1SMauro Carvalho Chehabno_heap Disable heap-style segment allocation which finds free 12689272ca1SMauro Carvalho Chehab segments for data from the beginning of main area, while 12789272ca1SMauro Carvalho Chehab for node from the end of main area. 12889272ca1SMauro Carvalho Chehabnouser_xattr Disable Extended User Attributes. Note: xattr is enabled 12989272ca1SMauro Carvalho Chehab by default if CONFIG_F2FS_FS_XATTR is selected. 13089272ca1SMauro Carvalho Chehabnoacl Disable POSIX Access Control List. Note: acl is enabled 13189272ca1SMauro Carvalho Chehab by default if CONFIG_F2FS_FS_POSIX_ACL is selected. 13289272ca1SMauro Carvalho Chehabactive_logs=%u Support configuring the number of active logs. In the 13389272ca1SMauro Carvalho Chehab current design, f2fs supports only 2, 4, and 6 logs. 13489272ca1SMauro Carvalho Chehab Default number is 6. 13589272ca1SMauro Carvalho Chehabdisable_ext_identify Disable the extension list configured by mkfs, so f2fs 136ca313c82SRandy Dunlap is not aware of cold files such as media files. 13789272ca1SMauro Carvalho Chehabinline_xattr Enable the inline xattrs feature. 13889272ca1SMauro Carvalho Chehabnoinline_xattr Disable the inline xattrs feature. 13989272ca1SMauro Carvalho Chehabinline_xattr_size=%u Support configuring inline xattr size, it depends on 14089272ca1SMauro Carvalho Chehab flexible inline xattr feature. 141ca313c82SRandy Dunlapinline_data Enable the inline data feature: Newly created small (<~3.4k) 14289272ca1SMauro Carvalho Chehab files can be written into inode block. 143ca313c82SRandy Dunlapinline_dentry Enable the inline dir feature: data in newly created 14489272ca1SMauro Carvalho Chehab directory entries can be written into inode block. The 14589272ca1SMauro Carvalho Chehab space of inode block which is used to store inline 14689272ca1SMauro Carvalho Chehab dentries is limited to ~3.4k. 14789272ca1SMauro Carvalho Chehabnoinline_dentry Disable the inline dentry feature. 14889272ca1SMauro Carvalho Chehabflush_merge Merge concurrent cache_flush commands as much as possible 14989272ca1SMauro Carvalho Chehab to eliminate redundant command issues. If the underlying 15089272ca1SMauro Carvalho Chehab device handles the cache_flush command relatively slowly, 15189272ca1SMauro Carvalho Chehab recommend to enable this option. 15289272ca1SMauro Carvalho Chehabnobarrier This option can be used if underlying storage guarantees 15389272ca1SMauro Carvalho Chehab its cached data should be written to the novolatile area. 15489272ca1SMauro Carvalho Chehab If this option is set, no cache_flush commands are issued 15589272ca1SMauro Carvalho Chehab but f2fs still guarantees the write ordering of all the 15689272ca1SMauro Carvalho Chehab data writes. 15789272ca1SMauro Carvalho Chehabfastboot This option is used when a system wants to reduce mount 15889272ca1SMauro Carvalho Chehab time as much as possible, even though normal performance 15989272ca1SMauro Carvalho Chehab can be sacrificed. 16089272ca1SMauro Carvalho Chehabextent_cache Enable an extent cache based on rb-tree, it can cache 16189272ca1SMauro Carvalho Chehab as many as extent which map between contiguous logical 16289272ca1SMauro Carvalho Chehab address and physical address per inode, resulting in 16389272ca1SMauro Carvalho Chehab increasing the cache hit ratio. Set by default. 16489272ca1SMauro Carvalho Chehabnoextent_cache Disable an extent cache based on rb-tree explicitly, see 16589272ca1SMauro Carvalho Chehab the above extent_cache mount option. 16689272ca1SMauro Carvalho Chehabnoinline_data Disable the inline data feature, inline data feature is 16789272ca1SMauro Carvalho Chehab enabled by default. 16889272ca1SMauro Carvalho Chehabdata_flush Enable data flushing before checkpoint in order to 16989272ca1SMauro Carvalho Chehab persist data of regular and symlink. 17089272ca1SMauro Carvalho Chehabreserve_root=%d Support configuring reserved space which is used for 17189272ca1SMauro Carvalho Chehab allocation from a privileged user with specified uid or 17289272ca1SMauro Carvalho Chehab gid, unit: 4KB, the default limit is 0.2% of user blocks. 17389272ca1SMauro Carvalho Chehabresuid=%d The user ID which may use the reserved blocks. 17489272ca1SMauro Carvalho Chehabresgid=%d The group ID which may use the reserved blocks. 17589272ca1SMauro Carvalho Chehabfault_injection=%d Enable fault injection in all supported types with 17689272ca1SMauro Carvalho Chehab specified injection rate. 17789272ca1SMauro Carvalho Chehabfault_type=%d Support configuring fault injection type, should be 17889272ca1SMauro Carvalho Chehab enabled with fault_injection option, fault type value 17989272ca1SMauro Carvalho Chehab is shown below, it supports single or combined type. 18089272ca1SMauro Carvalho Chehab 18189272ca1SMauro Carvalho Chehab =================== =========== 18289272ca1SMauro Carvalho Chehab Type_Name Type_Value 18389272ca1SMauro Carvalho Chehab =================== =========== 18489272ca1SMauro Carvalho Chehab FAULT_KMALLOC 0x000000001 18589272ca1SMauro Carvalho Chehab FAULT_KVMALLOC 0x000000002 18689272ca1SMauro Carvalho Chehab FAULT_PAGE_ALLOC 0x000000004 18789272ca1SMauro Carvalho Chehab FAULT_PAGE_GET 0x000000008 188b96d9b3bSChao Yu FAULT_ALLOC_BIO 0x000000010 (obsolete) 18989272ca1SMauro Carvalho Chehab FAULT_ALLOC_NID 0x000000020 19089272ca1SMauro Carvalho Chehab FAULT_ORPHAN 0x000000040 19189272ca1SMauro Carvalho Chehab FAULT_BLOCK 0x000000080 19289272ca1SMauro Carvalho Chehab FAULT_DIR_DEPTH 0x000000100 19389272ca1SMauro Carvalho Chehab FAULT_EVICT_INODE 0x000000200 19489272ca1SMauro Carvalho Chehab FAULT_TRUNCATE 0x000000400 19589272ca1SMauro Carvalho Chehab FAULT_READ_IO 0x000000800 19689272ca1SMauro Carvalho Chehab FAULT_CHECKPOINT 0x000001000 19789272ca1SMauro Carvalho Chehab FAULT_DISCARD 0x000002000 19889272ca1SMauro Carvalho Chehab FAULT_WRITE_IO 0x000004000 19932410577SChao Yu FAULT_SLAB_ALLOC 0x000008000 20010a26878SChao Yu FAULT_DQUOT_INIT 0x000010000 2013e020389SChao Yu FAULT_LOCK_OP 0x000020000 20289272ca1SMauro Carvalho Chehab =================== =========== 20389272ca1SMauro Carvalho Chehabmode=%s Control block allocation mode which supports "adaptive" 20489272ca1SMauro Carvalho Chehab and "lfs". In "lfs" mode, there should be no random 20589272ca1SMauro Carvalho Chehab writes towards main area. 2066691d940SDaeho Jeong "fragment:segment" and "fragment:block" are newly added here. 2076691d940SDaeho Jeong These are developer options for experiments to simulate filesystem 2086691d940SDaeho Jeong fragmentation/after-GC situation itself. The developers use these 2096691d940SDaeho Jeong modes to understand filesystem fragmentation/after-GC condition well, 2106691d940SDaeho Jeong and eventually get some insights to handle them better. 2116691d940SDaeho Jeong In "fragment:segment", f2fs allocates a new segment in ramdom 2126691d940SDaeho Jeong position. With this, we can simulate the after-GC condition. 2136691d940SDaeho Jeong In "fragment:block", we can scatter block allocation with 2146691d940SDaeho Jeong "max_fragment_chunk" and "max_fragment_hole" sysfs nodes. 2156691d940SDaeho Jeong We added some randomness to both chunk and hole size to make 2166691d940SDaeho Jeong it close to realistic IO pattern. So, in this mode, f2fs will allocate 2176691d940SDaeho Jeong 1..<max_fragment_chunk> blocks in a chunk and make a hole in the 2186691d940SDaeho Jeong length of 1..<max_fragment_hole> by turns. With this, the newly 2196691d940SDaeho Jeong allocated blocks will be scattered throughout the whole partition. 2206691d940SDaeho Jeong Note that "fragment:block" implicitly enables "fragment:segment" 2216691d940SDaeho Jeong option for more randomness. 2226691d940SDaeho Jeong Please, use these options for your experiments and we strongly 2236691d940SDaeho Jeong recommend to re-format the filesystem after using these options. 22489272ca1SMauro Carvalho Chehabio_bits=%u Set the bit size of write IO requests. It should be set 22589272ca1SMauro Carvalho Chehab with "mode=lfs". 22689272ca1SMauro Carvalho Chehabusrquota Enable plain user disk quota accounting. 22789272ca1SMauro Carvalho Chehabgrpquota Enable plain group disk quota accounting. 22889272ca1SMauro Carvalho Chehabprjquota Enable plain project quota accounting. 22989272ca1SMauro Carvalho Chehabusrjquota=<file> Appoint specified file and type during mount, so that quota 23089272ca1SMauro Carvalho Chehabgrpjquota=<file> information can be properly updated during recovery flow, 23189272ca1SMauro Carvalho Chehabprjjquota=<file> <quota file>: must be in root directory; 23289272ca1SMauro Carvalho Chehabjqfmt=<quota type> <quota type>: [vfsold,vfsv0,vfsv1]. 233ca313c82SRandy Dunlapoffusrjquota Turn off user journalled quota. 234ca313c82SRandy Dunlapoffgrpjquota Turn off group journalled quota. 235ca313c82SRandy Dunlapoffprjjquota Turn off project journalled quota. 23689272ca1SMauro Carvalho Chehabquota Enable plain user disk quota accounting. 23789272ca1SMauro Carvalho Chehabnoquota Disable all plain disk quota option. 23889272ca1SMauro Carvalho Chehaballoc_mode=%s Adjust block allocation policy, which supports "reuse" 23989272ca1SMauro Carvalho Chehab and "default". 24089272ca1SMauro Carvalho Chehabfsync_mode=%s Control the policy of fsync. Currently supports "posix", 24189272ca1SMauro Carvalho Chehab "strict", and "nobarrier". In "posix" mode, which is 24289272ca1SMauro Carvalho Chehab default, fsync will follow POSIX semantics and does a 24389272ca1SMauro Carvalho Chehab light operation to improve the filesystem performance. 24489272ca1SMauro Carvalho Chehab In "strict" mode, fsync will be heavy and behaves in line 24589272ca1SMauro Carvalho Chehab with xfs, ext4 and btrfs, where xfstest generic/342 will 24689272ca1SMauro Carvalho Chehab pass, but the performance will regress. "nobarrier" is 24789272ca1SMauro Carvalho Chehab based on "posix", but doesn't issue flush command for 24889272ca1SMauro Carvalho Chehab non-atomic files likewise "nobarrier" mount option. 249ed318a6cSEric Biggerstest_dummy_encryption 250ed318a6cSEric Biggerstest_dummy_encryption=%s 251ed318a6cSEric Biggers Enable dummy encryption, which provides a fake fscrypt 25289272ca1SMauro Carvalho Chehab context. The fake fscrypt context is used by xfstests. 253ed318a6cSEric Biggers The argument may be either "v1" or "v2", in order to 254ed318a6cSEric Biggers select the corresponding fscrypt policy version. 25589272ca1SMauro Carvalho Chehabcheckpoint=%s[:%u[%]] Set to "disable" to turn off checkpointing. Set to "enable" 25689272ca1SMauro Carvalho Chehab to reenable checkpointing. Is enabled by default. While 25789272ca1SMauro Carvalho Chehab disabled, any unmounting or unexpected shutdowns will cause 25889272ca1SMauro Carvalho Chehab the filesystem contents to appear as they did when the 25989272ca1SMauro Carvalho Chehab filesystem was mounted with that option. 26089272ca1SMauro Carvalho Chehab While mounting with checkpoint=disabled, the filesystem must 26189272ca1SMauro Carvalho Chehab run garbage collection to ensure that all available space can 26289272ca1SMauro Carvalho Chehab be used. If this takes too much time, the mount may return 26389272ca1SMauro Carvalho Chehab EAGAIN. You may optionally add a value to indicate how much 26489272ca1SMauro Carvalho Chehab of the disk you would be willing to temporarily give up to 26589272ca1SMauro Carvalho Chehab avoid additional garbage collection. This can be given as a 26689272ca1SMauro Carvalho Chehab number of blocks, or as a percent. For instance, mounting 26789272ca1SMauro Carvalho Chehab with checkpoint=disable:100% would always succeed, but it may 26889272ca1SMauro Carvalho Chehab hide up to all remaining free space. The actual space that 26989272ca1SMauro Carvalho Chehab would be unusable can be viewed at /sys/fs/f2fs/<disk>/unusable 27089272ca1SMauro Carvalho Chehab This space is reclaimed once checkpoint=enable. 271261eeb9cSDaeho Jeongcheckpoint_merge When checkpoint is enabled, this can be used to create a kernel 272261eeb9cSDaeho Jeong daemon and make it to merge concurrent checkpoint requests as 273261eeb9cSDaeho Jeong much as possible to eliminate redundant checkpoint issues. Plus, 274261eeb9cSDaeho Jeong we can eliminate the sluggish issue caused by slow checkpoint 275261eeb9cSDaeho Jeong operation when the checkpoint is done in a process context in 276261eeb9cSDaeho Jeong a cgroup having low i/o budget and cpu shares. To make this 277261eeb9cSDaeho Jeong do better, we set the default i/o priority of the kernel daemon 278261eeb9cSDaeho Jeong to "3", to give one higher priority than other kernel threads. 279261eeb9cSDaeho Jeong This is the same way to give a I/O priority to the jbd2 280261eeb9cSDaeho Jeong journaling thread of ext4 filesystem. 281261eeb9cSDaeho Jeongnocheckpoint_merge Disable checkpoint merge feature. 282f40f31caSLinus Torvaldscompress_algorithm=%s Control compress algorithm, currently f2fs supports "lzo", 2836d92b201SChao Yu "lz4", "zstd" and "lzo-rle" algorithm. 2843fde13f8SChao Yucompress_algorithm=%s:%d Control compress algorithm and its compress level, now, only 2853fde13f8SChao Yu "lz4" and "zstd" support compress level config. 2863fde13f8SChao Yu algorithm level range 2873fde13f8SChao Yu lz4 3 - 16 2883fde13f8SChao Yu zstd 1 - 22 28989272ca1SMauro Carvalho Chehabcompress_log_size=%u Support configuring compress cluster size, the size will 29089272ca1SMauro Carvalho Chehab be 4KB * (1 << %u), 16KB is minimum size, also it's 29189272ca1SMauro Carvalho Chehab default size. 29289272ca1SMauro Carvalho Chehabcompress_extension=%s Support adding specified extension, so that f2fs can enable 29389272ca1SMauro Carvalho Chehab compression on those corresponding files, e.g. if all files 29489272ca1SMauro Carvalho Chehab with '.ext' has high compression rate, we can set the '.ext' 29589272ca1SMauro Carvalho Chehab on compression extension list and enable compression on 29689272ca1SMauro Carvalho Chehab these file by default rather than to enable it via ioctl. 29789272ca1SMauro Carvalho Chehab For other files, we can still enable compression via ioctl. 298fa679555SChao Yu Note that, there is one reserved special extension '*', it 299fa679555SChao Yu can be set to enable compression for all files. 300151b1982SFengnan Changnocompress_extension=%s Support adding specified extension, so that f2fs can disable 301151b1982SFengnan Chang compression on those corresponding files, just contrary to compression extension. 302151b1982SFengnan Chang If you know exactly which files cannot be compressed, you can use this. 303151b1982SFengnan Chang The same extension name can't appear in both compress and nocompress 304151b1982SFengnan Chang extension at the same time. 305151b1982SFengnan Chang If the compress extension specifies all files, the types specified by the 306151b1982SFengnan Chang nocompress extension will be treated as special cases and will not be compressed. 307151b1982SFengnan Chang Don't allow use '*' to specifie all file in nocompress extension. 308151b1982SFengnan Chang After add nocompress_extension, the priority should be: 309151b1982SFengnan Chang dir_flag < comp_extention,nocompress_extension < comp_file_flag,no_comp_file_flag. 310151b1982SFengnan Chang See more in compression sections. 311151b1982SFengnan Chang 312b28f047bSChao Yucompress_chksum Support verifying chksum of raw data in compressed cluster. 313602a16d5SDaeho Jeongcompress_mode=%s Control file compression mode. This supports "fs" and "user" 314602a16d5SDaeho Jeong modes. In "fs" mode (default), f2fs does automatic compression 315602a16d5SDaeho Jeong on the compression enabled files. In "user" mode, f2fs disables 316602a16d5SDaeho Jeong the automaic compression and gives the user discretion of 317602a16d5SDaeho Jeong choosing the target file and the timing. The user can do manual 318602a16d5SDaeho Jeong compression/decompression on the compression enabled files using 319602a16d5SDaeho Jeong ioctls. 3206ce19affSChao Yucompress_cache Support to use address space of a filesystem managed inode to 3216ce19affSChao Yu cache compressed block, in order to improve cache hit ratio of 3226ce19affSChao Yu random read. 3232324d50dSLinus Torvaldsinlinecrypt When possible, encrypt/decrypt the contents of encrypted 32427aacd28SSatya Tangirala files using the blk-crypto framework rather than 32527aacd28SSatya Tangirala filesystem-layer encryption. This allows the use of 32627aacd28SSatya Tangirala inline encryption hardware. The on-disk format is 32727aacd28SSatya Tangirala unaffected. For more details, see 32827aacd28SSatya Tangirala Documentation/block/inline-encryption.rst. 329093749e2SChao Yuatgc Enable age-threshold garbage collection, it provides high 330093749e2SChao Yu effectiveness and efficiency on background GC. 3314f993264SChao Yudiscard_unit=%s Control discard unit, the argument can be "block", "segment" 3324f993264SChao Yu and "section", issued discard command's offset/size will be 3334f993264SChao Yu aligned to the unit, by default, "discard_unit=block" is set, 3344f993264SChao Yu so that small discard functionality is enabled. 3354f993264SChao Yu For blkzoned device, "discard_unit=section" will be set by 3364f993264SChao Yu default, it is helpful for large sized SMR or ZNS devices to 3374f993264SChao Yu reduce memory cost by getting rid of fs metadata supports small 3384f993264SChao Yu discard. 3399aa1ccb4SJonathan Corbet======================== ============================================================ 34089272ca1SMauro Carvalho Chehab 34189272ca1SMauro Carvalho ChehabDebugfs Entries 34289272ca1SMauro Carvalho Chehab=============== 34389272ca1SMauro Carvalho Chehab 34489272ca1SMauro Carvalho Chehab/sys/kernel/debug/f2fs/ contains information about all the partitions mounted as 34589272ca1SMauro Carvalho Chehabf2fs. Each file shows the whole f2fs information. 34689272ca1SMauro Carvalho Chehab 34789272ca1SMauro Carvalho Chehab/sys/kernel/debug/f2fs/status includes: 34889272ca1SMauro Carvalho Chehab 34989272ca1SMauro Carvalho Chehab - major file system information managed by f2fs currently 35089272ca1SMauro Carvalho Chehab - average SIT information about whole segments 35189272ca1SMauro Carvalho Chehab - current memory footprint consumed by f2fs. 35289272ca1SMauro Carvalho Chehab 35389272ca1SMauro Carvalho ChehabSysfs Entries 35489272ca1SMauro Carvalho Chehab============= 35589272ca1SMauro Carvalho Chehab 35689272ca1SMauro Carvalho ChehabInformation about mounted f2fs file systems can be found in 35789272ca1SMauro Carvalho Chehab/sys/fs/f2fs. Each mounted filesystem will have a directory in 35889272ca1SMauro Carvalho Chehab/sys/fs/f2fs based on its device name (i.e., /sys/fs/f2fs/sda). 35989272ca1SMauro Carvalho ChehabThe files in each per-device directory are shown in table below. 36089272ca1SMauro Carvalho Chehab 36189272ca1SMauro Carvalho ChehabFiles in /sys/fs/f2fs/<devname> 36289272ca1SMauro Carvalho Chehab(see also Documentation/ABI/testing/sysfs-fs-f2fs) 36389272ca1SMauro Carvalho Chehab 36489272ca1SMauro Carvalho ChehabUsage 36589272ca1SMauro Carvalho Chehab===== 36689272ca1SMauro Carvalho Chehab 36789272ca1SMauro Carvalho Chehab1. Download userland tools and compile them. 36889272ca1SMauro Carvalho Chehab 36989272ca1SMauro Carvalho Chehab2. Skip, if f2fs was compiled statically inside kernel. 37089272ca1SMauro Carvalho Chehab Otherwise, insert the f2fs.ko module:: 37189272ca1SMauro Carvalho Chehab 37289272ca1SMauro Carvalho Chehab # insmod f2fs.ko 37389272ca1SMauro Carvalho Chehab 374ca313c82SRandy Dunlap3. Create a directory to use when mounting:: 37589272ca1SMauro Carvalho Chehab 37689272ca1SMauro Carvalho Chehab # mkdir /mnt/f2fs 37789272ca1SMauro Carvalho Chehab 37889272ca1SMauro Carvalho Chehab4. Format the block device, and then mount as f2fs:: 37989272ca1SMauro Carvalho Chehab 38089272ca1SMauro Carvalho Chehab # mkfs.f2fs -l label /dev/block_device 38189272ca1SMauro Carvalho Chehab # mount -t f2fs /dev/block_device /mnt/f2fs 38289272ca1SMauro Carvalho Chehab 38389272ca1SMauro Carvalho Chehabmkfs.f2fs 38489272ca1SMauro Carvalho Chehab--------- 38589272ca1SMauro Carvalho ChehabThe mkfs.f2fs is for the use of formatting a partition as the f2fs filesystem, 38689272ca1SMauro Carvalho Chehabwhich builds a basic on-disk layout. 38789272ca1SMauro Carvalho Chehab 388568d2a1eSJaegeuk KimThe quick options consist of: 38989272ca1SMauro Carvalho Chehab 39089272ca1SMauro Carvalho Chehab=============== =========================================================== 39189272ca1SMauro Carvalho Chehab``-l [label]`` Give a volume label, up to 512 unicode name. 39289272ca1SMauro Carvalho Chehab``-a [0 or 1]`` Split start location of each area for heap-based allocation. 39389272ca1SMauro Carvalho Chehab 39489272ca1SMauro Carvalho Chehab 1 is set by default, which performs this. 39589272ca1SMauro Carvalho Chehab``-o [int]`` Set overprovision ratio in percent over volume size. 39689272ca1SMauro Carvalho Chehab 39789272ca1SMauro Carvalho Chehab 5 is set by default. 39889272ca1SMauro Carvalho Chehab``-s [int]`` Set the number of segments per section. 39989272ca1SMauro Carvalho Chehab 40089272ca1SMauro Carvalho Chehab 1 is set by default. 40189272ca1SMauro Carvalho Chehab``-z [int]`` Set the number of sections per zone. 40289272ca1SMauro Carvalho Chehab 40389272ca1SMauro Carvalho Chehab 1 is set by default. 40489272ca1SMauro Carvalho Chehab``-e [str]`` Set basic extension list. e.g. "mp3,gif,mov" 40589272ca1SMauro Carvalho Chehab``-t [0 or 1]`` Disable discard command or not. 40689272ca1SMauro Carvalho Chehab 40789272ca1SMauro Carvalho Chehab 1 is set by default, which conducts discard. 40889272ca1SMauro Carvalho Chehab=============== =========================================================== 40989272ca1SMauro Carvalho Chehab 410ca313c82SRandy DunlapNote: please refer to the manpage of mkfs.f2fs(8) to get full option list. 411568d2a1eSJaegeuk Kim 41289272ca1SMauro Carvalho Chehabfsck.f2fs 41389272ca1SMauro Carvalho Chehab--------- 41489272ca1SMauro Carvalho ChehabThe fsck.f2fs is a tool to check the consistency of an f2fs-formatted 41589272ca1SMauro Carvalho Chehabpartition, which examines whether the filesystem metadata and user-made data 41689272ca1SMauro Carvalho Chehabare cross-referenced correctly or not. 41789272ca1SMauro Carvalho ChehabNote that, initial version of the tool does not fix any inconsistency. 41889272ca1SMauro Carvalho Chehab 419568d2a1eSJaegeuk KimThe quick options consist of:: 42089272ca1SMauro Carvalho Chehab 42189272ca1SMauro Carvalho Chehab -d debug level [default:0] 42289272ca1SMauro Carvalho Chehab 423ca313c82SRandy DunlapNote: please refer to the manpage of fsck.f2fs(8) to get full option list. 424568d2a1eSJaegeuk Kim 42589272ca1SMauro Carvalho Chehabdump.f2fs 42689272ca1SMauro Carvalho Chehab--------- 42789272ca1SMauro Carvalho ChehabThe dump.f2fs shows the information of specific inode and dumps SSA and SIT to 42889272ca1SMauro Carvalho Chehabfile. Each file is dump_ssa and dump_sit. 42989272ca1SMauro Carvalho Chehab 43089272ca1SMauro Carvalho ChehabThe dump.f2fs is used to debug on-disk data structures of the f2fs filesystem. 43189272ca1SMauro Carvalho ChehabIt shows on-disk inode information recognized by a given inode number, and is 43289272ca1SMauro Carvalho Chehabable to dump all the SSA and SIT entries into predefined files, ./dump_ssa and 43389272ca1SMauro Carvalho Chehab./dump_sit respectively. 43489272ca1SMauro Carvalho Chehab 43589272ca1SMauro Carvalho ChehabThe options consist of:: 43689272ca1SMauro Carvalho Chehab 43789272ca1SMauro Carvalho Chehab -d debug level [default:0] 43889272ca1SMauro Carvalho Chehab -i inode no (hex) 43989272ca1SMauro Carvalho Chehab -s [SIT dump segno from #1~#2 (decimal), for all 0~-1] 44089272ca1SMauro Carvalho Chehab -a [SSA dump segno from #1~#2 (decimal), for all 0~-1] 44189272ca1SMauro Carvalho Chehab 44289272ca1SMauro Carvalho ChehabExamples:: 44389272ca1SMauro Carvalho Chehab 44489272ca1SMauro Carvalho Chehab # dump.f2fs -i [ino] /dev/sdx 44589272ca1SMauro Carvalho Chehab # dump.f2fs -s 0~-1 /dev/sdx (SIT dump) 44689272ca1SMauro Carvalho Chehab # dump.f2fs -a 0~-1 /dev/sdx (SSA dump) 44789272ca1SMauro Carvalho Chehab 448ca313c82SRandy DunlapNote: please refer to the manpage of dump.f2fs(8) to get full option list. 449568d2a1eSJaegeuk Kim 450568d2a1eSJaegeuk Kimsload.f2fs 451568d2a1eSJaegeuk Kim---------- 452568d2a1eSJaegeuk KimThe sload.f2fs gives a way to insert files and directories in the exisiting disk 453568d2a1eSJaegeuk Kimimage. This tool is useful when building f2fs images given compiled files. 454568d2a1eSJaegeuk Kim 455ca313c82SRandy DunlapNote: please refer to the manpage of sload.f2fs(8) to get full option list. 456568d2a1eSJaegeuk Kim 457568d2a1eSJaegeuk Kimresize.f2fs 458568d2a1eSJaegeuk Kim----------- 459ca313c82SRandy DunlapThe resize.f2fs lets a user resize the f2fs-formatted disk image, while preserving 460568d2a1eSJaegeuk Kimall the files and directories stored in the image. 461568d2a1eSJaegeuk Kim 462ca313c82SRandy DunlapNote: please refer to the manpage of resize.f2fs(8) to get full option list. 463568d2a1eSJaegeuk Kim 464568d2a1eSJaegeuk Kimdefrag.f2fs 465568d2a1eSJaegeuk Kim----------- 466ca313c82SRandy DunlapThe defrag.f2fs can be used to defragment scattered written data as well as 467568d2a1eSJaegeuk Kimfilesystem metadata across the disk. This can improve the write speed by giving 468568d2a1eSJaegeuk Kimmore free consecutive space. 469568d2a1eSJaegeuk Kim 470ca313c82SRandy DunlapNote: please refer to the manpage of defrag.f2fs(8) to get full option list. 471568d2a1eSJaegeuk Kim 472568d2a1eSJaegeuk Kimf2fs_io 473568d2a1eSJaegeuk Kim------- 474568d2a1eSJaegeuk KimThe f2fs_io is a simple tool to issue various filesystem APIs as well as 475568d2a1eSJaegeuk Kimf2fs-specific ones, which is very useful for QA tests. 476568d2a1eSJaegeuk Kim 477ca313c82SRandy DunlapNote: please refer to the manpage of f2fs_io(8) to get full option list. 478568d2a1eSJaegeuk Kim 47989272ca1SMauro Carvalho ChehabDesign 48089272ca1SMauro Carvalho Chehab====== 48189272ca1SMauro Carvalho Chehab 48289272ca1SMauro Carvalho ChehabOn-disk Layout 48389272ca1SMauro Carvalho Chehab-------------- 48489272ca1SMauro Carvalho Chehab 48589272ca1SMauro Carvalho ChehabF2FS divides the whole volume into a number of segments, each of which is fixed 48689272ca1SMauro Carvalho Chehabto 2MB in size. A section is composed of consecutive segments, and a zone 48789272ca1SMauro Carvalho Chehabconsists of a set of sections. By default, section and zone sizes are set to one 48889272ca1SMauro Carvalho Chehabsegment size identically, but users can easily modify the sizes by mkfs. 48989272ca1SMauro Carvalho Chehab 49089272ca1SMauro Carvalho ChehabF2FS splits the entire volume into six areas, and all the areas except superblock 491ca313c82SRandy Dunlapconsist of multiple segments as described below:: 49289272ca1SMauro Carvalho Chehab 49389272ca1SMauro Carvalho Chehab align with the zone size <-| 49489272ca1SMauro Carvalho Chehab |-> align with the segment size 49589272ca1SMauro Carvalho Chehab _________________________________________________________________________ 49689272ca1SMauro Carvalho Chehab | | | Segment | Node | Segment | | 49789272ca1SMauro Carvalho Chehab | Superblock | Checkpoint | Info. | Address | Summary | Main | 49889272ca1SMauro Carvalho Chehab | (SB) | (CP) | Table (SIT) | Table (NAT) | Area (SSA) | | 49989272ca1SMauro Carvalho Chehab |____________|_____2______|______N______|______N______|______N_____|__N___| 50089272ca1SMauro Carvalho Chehab . . 50189272ca1SMauro Carvalho Chehab . . 50289272ca1SMauro Carvalho Chehab . . 50389272ca1SMauro Carvalho Chehab ._________________________________________. 50489272ca1SMauro Carvalho Chehab |_Segment_|_..._|_Segment_|_..._|_Segment_| 50589272ca1SMauro Carvalho Chehab . . 50689272ca1SMauro Carvalho Chehab ._________._________ 50789272ca1SMauro Carvalho Chehab |_section_|__...__|_ 50889272ca1SMauro Carvalho Chehab . . 50989272ca1SMauro Carvalho Chehab .________. 51089272ca1SMauro Carvalho Chehab |__zone__| 51189272ca1SMauro Carvalho Chehab 51289272ca1SMauro Carvalho Chehab- Superblock (SB) 51389272ca1SMauro Carvalho Chehab It is located at the beginning of the partition, and there exist two copies 51489272ca1SMauro Carvalho Chehab to avoid file system crash. It contains basic partition information and some 51589272ca1SMauro Carvalho Chehab default parameters of f2fs. 51689272ca1SMauro Carvalho Chehab 51789272ca1SMauro Carvalho Chehab- Checkpoint (CP) 51889272ca1SMauro Carvalho Chehab It contains file system information, bitmaps for valid NAT/SIT sets, orphan 51989272ca1SMauro Carvalho Chehab inode lists, and summary entries of current active segments. 52089272ca1SMauro Carvalho Chehab 52189272ca1SMauro Carvalho Chehab- Segment Information Table (SIT) 52289272ca1SMauro Carvalho Chehab It contains segment information such as valid block count and bitmap for the 52389272ca1SMauro Carvalho Chehab validity of all the blocks. 52489272ca1SMauro Carvalho Chehab 52589272ca1SMauro Carvalho Chehab- Node Address Table (NAT) 52689272ca1SMauro Carvalho Chehab It is composed of a block address table for all the node blocks stored in 52789272ca1SMauro Carvalho Chehab Main area. 52889272ca1SMauro Carvalho Chehab 52989272ca1SMauro Carvalho Chehab- Segment Summary Area (SSA) 53089272ca1SMauro Carvalho Chehab It contains summary entries which contains the owner information of all the 53189272ca1SMauro Carvalho Chehab data and node blocks stored in Main area. 53289272ca1SMauro Carvalho Chehab 53389272ca1SMauro Carvalho Chehab- Main Area 53489272ca1SMauro Carvalho Chehab It contains file and directory data including their indices. 53589272ca1SMauro Carvalho Chehab 53689272ca1SMauro Carvalho ChehabIn order to avoid misalignment between file system and flash-based storage, F2FS 53789272ca1SMauro Carvalho Chehabaligns the start block address of CP with the segment size. Also, it aligns the 53889272ca1SMauro Carvalho Chehabstart block address of Main area with the zone size by reserving some segments 53989272ca1SMauro Carvalho Chehabin SSA area. 54089272ca1SMauro Carvalho Chehab 54189272ca1SMauro Carvalho ChehabReference the following survey for additional technical details. 54289272ca1SMauro Carvalho Chehabhttps://wiki.linaro.org/WorkingGroups/Kernel/Projects/FlashCardSurvey 54389272ca1SMauro Carvalho Chehab 54489272ca1SMauro Carvalho ChehabFile System Metadata Structure 54589272ca1SMauro Carvalho Chehab------------------------------ 54689272ca1SMauro Carvalho Chehab 54789272ca1SMauro Carvalho ChehabF2FS adopts the checkpointing scheme to maintain file system consistency. At 54889272ca1SMauro Carvalho Chehabmount time, F2FS first tries to find the last valid checkpoint data by scanning 54989272ca1SMauro Carvalho ChehabCP area. In order to reduce the scanning time, F2FS uses only two copies of CP. 55089272ca1SMauro Carvalho ChehabOne of them always indicates the last valid data, which is called as shadow copy 55189272ca1SMauro Carvalho Chehabmechanism. In addition to CP, NAT and SIT also adopt the shadow copy mechanism. 55289272ca1SMauro Carvalho Chehab 55389272ca1SMauro Carvalho ChehabFor file system consistency, each CP points to which NAT and SIT copies are 55489272ca1SMauro Carvalho Chehabvalid, as shown as below:: 55589272ca1SMauro Carvalho Chehab 55689272ca1SMauro Carvalho Chehab +--------+----------+---------+ 55789272ca1SMauro Carvalho Chehab | CP | SIT | NAT | 55889272ca1SMauro Carvalho Chehab +--------+----------+---------+ 55989272ca1SMauro Carvalho Chehab . . . . 56089272ca1SMauro Carvalho Chehab . . . . 56189272ca1SMauro Carvalho Chehab . . . . 56289272ca1SMauro Carvalho Chehab +-------+-------+--------+--------+--------+--------+ 56389272ca1SMauro Carvalho Chehab | CP #0 | CP #1 | SIT #0 | SIT #1 | NAT #0 | NAT #1 | 56489272ca1SMauro Carvalho Chehab +-------+-------+--------+--------+--------+--------+ 56589272ca1SMauro Carvalho Chehab | ^ ^ 56689272ca1SMauro Carvalho Chehab | | | 56789272ca1SMauro Carvalho Chehab `----------------------------------------' 56889272ca1SMauro Carvalho Chehab 56989272ca1SMauro Carvalho ChehabIndex Structure 57089272ca1SMauro Carvalho Chehab--------------- 57189272ca1SMauro Carvalho Chehab 57289272ca1SMauro Carvalho ChehabThe key data structure to manage the data locations is a "node". Similar to 57389272ca1SMauro Carvalho Chehabtraditional file structures, F2FS has three types of node: inode, direct node, 57489272ca1SMauro Carvalho Chehabindirect node. F2FS assigns 4KB to an inode block which contains 923 data block 57589272ca1SMauro Carvalho Chehabindices, two direct node pointers, two indirect node pointers, and one double 57689272ca1SMauro Carvalho Chehabindirect node pointer as described below. One direct node block contains 1018 57789272ca1SMauro Carvalho Chehabdata blocks, and one indirect node block contains also 1018 node blocks. Thus, 57889272ca1SMauro Carvalho Chehabone inode block (i.e., a file) covers:: 57989272ca1SMauro Carvalho Chehab 58089272ca1SMauro Carvalho Chehab 4KB * (923 + 2 * 1018 + 2 * 1018 * 1018 + 1018 * 1018 * 1018) := 3.94TB. 58189272ca1SMauro Carvalho Chehab 58289272ca1SMauro Carvalho Chehab Inode block (4KB) 58389272ca1SMauro Carvalho Chehab |- data (923) 58489272ca1SMauro Carvalho Chehab |- direct node (2) 58589272ca1SMauro Carvalho Chehab | `- data (1018) 58689272ca1SMauro Carvalho Chehab |- indirect node (2) 58789272ca1SMauro Carvalho Chehab | `- direct node (1018) 58889272ca1SMauro Carvalho Chehab | `- data (1018) 58989272ca1SMauro Carvalho Chehab `- double indirect node (1) 59089272ca1SMauro Carvalho Chehab `- indirect node (1018) 59189272ca1SMauro Carvalho Chehab `- direct node (1018) 59289272ca1SMauro Carvalho Chehab `- data (1018) 59389272ca1SMauro Carvalho Chehab 594ca313c82SRandy DunlapNote that all the node blocks are mapped by NAT which means the location of 59589272ca1SMauro Carvalho Chehabeach node is translated by the NAT table. In the consideration of the wandering 59689272ca1SMauro Carvalho Chehabtree problem, F2FS is able to cut off the propagation of node updates caused by 59789272ca1SMauro Carvalho Chehableaf data writes. 59889272ca1SMauro Carvalho Chehab 59989272ca1SMauro Carvalho ChehabDirectory Structure 60089272ca1SMauro Carvalho Chehab------------------- 60189272ca1SMauro Carvalho Chehab 60289272ca1SMauro Carvalho ChehabA directory entry occupies 11 bytes, which consists of the following attributes. 60389272ca1SMauro Carvalho Chehab 60489272ca1SMauro Carvalho Chehab- hash hash value of the file name 60589272ca1SMauro Carvalho Chehab- ino inode number 60689272ca1SMauro Carvalho Chehab- len the length of file name 60789272ca1SMauro Carvalho Chehab- type file type such as directory, symlink, etc 60889272ca1SMauro Carvalho Chehab 60989272ca1SMauro Carvalho ChehabA dentry block consists of 214 dentry slots and file names. Therein a bitmap is 61089272ca1SMauro Carvalho Chehabused to represent whether each dentry is valid or not. A dentry block occupies 61189272ca1SMauro Carvalho Chehab4KB with the following composition. 61289272ca1SMauro Carvalho Chehab 61389272ca1SMauro Carvalho Chehab:: 61489272ca1SMauro Carvalho Chehab 61589272ca1SMauro Carvalho Chehab Dentry Block(4 K) = bitmap (27 bytes) + reserved (3 bytes) + 61689272ca1SMauro Carvalho Chehab dentries(11 * 214 bytes) + file name (8 * 214 bytes) 61789272ca1SMauro Carvalho Chehab 61889272ca1SMauro Carvalho Chehab [Bucket] 61989272ca1SMauro Carvalho Chehab +--------------------------------+ 62089272ca1SMauro Carvalho Chehab |dentry block 1 | dentry block 2 | 62189272ca1SMauro Carvalho Chehab +--------------------------------+ 62289272ca1SMauro Carvalho Chehab . . 62389272ca1SMauro Carvalho Chehab . . 62489272ca1SMauro Carvalho Chehab . [Dentry Block Structure: 4KB] . 62589272ca1SMauro Carvalho Chehab +--------+----------+----------+------------+ 62689272ca1SMauro Carvalho Chehab | bitmap | reserved | dentries | file names | 62789272ca1SMauro Carvalho Chehab +--------+----------+----------+------------+ 62889272ca1SMauro Carvalho Chehab [Dentry Block: 4KB] . . 62989272ca1SMauro Carvalho Chehab . . 63089272ca1SMauro Carvalho Chehab . . 63189272ca1SMauro Carvalho Chehab +------+------+-----+------+ 63289272ca1SMauro Carvalho Chehab | hash | ino | len | type | 63389272ca1SMauro Carvalho Chehab +------+------+-----+------+ 63489272ca1SMauro Carvalho Chehab [Dentry Structure: 11 bytes] 63589272ca1SMauro Carvalho Chehab 63689272ca1SMauro Carvalho ChehabF2FS implements multi-level hash tables for directory structure. Each level has 63789272ca1SMauro Carvalho Chehaba hash table with dedicated number of hash buckets as shown below. Note that 63889272ca1SMauro Carvalho Chehab"A(2B)" means a bucket includes 2 data blocks. 63989272ca1SMauro Carvalho Chehab 64089272ca1SMauro Carvalho Chehab:: 64189272ca1SMauro Carvalho Chehab 64289272ca1SMauro Carvalho Chehab ---------------------- 64389272ca1SMauro Carvalho Chehab A : bucket 64489272ca1SMauro Carvalho Chehab B : block 64589272ca1SMauro Carvalho Chehab N : MAX_DIR_HASH_DEPTH 64689272ca1SMauro Carvalho Chehab ---------------------- 64789272ca1SMauro Carvalho Chehab 64889272ca1SMauro Carvalho Chehab level #0 | A(2B) 64989272ca1SMauro Carvalho Chehab | 65089272ca1SMauro Carvalho Chehab level #1 | A(2B) - A(2B) 65189272ca1SMauro Carvalho Chehab | 65289272ca1SMauro Carvalho Chehab level #2 | A(2B) - A(2B) - A(2B) - A(2B) 65389272ca1SMauro Carvalho Chehab . | . . . . 65489272ca1SMauro Carvalho Chehab level #N/2 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) 65589272ca1SMauro Carvalho Chehab . | . . . . 65689272ca1SMauro Carvalho Chehab level #N | A(4B) - A(4B) - A(4B) - A(4B) - A(4B) - ... - A(4B) 65789272ca1SMauro Carvalho Chehab 65889272ca1SMauro Carvalho ChehabThe number of blocks and buckets are determined by:: 65989272ca1SMauro Carvalho Chehab 66089272ca1SMauro Carvalho Chehab ,- 2, if n < MAX_DIR_HASH_DEPTH / 2, 66189272ca1SMauro Carvalho Chehab # of blocks in level #n = | 66289272ca1SMauro Carvalho Chehab `- 4, Otherwise 66389272ca1SMauro Carvalho Chehab 66489272ca1SMauro Carvalho Chehab ,- 2^(n + dir_level), 66589272ca1SMauro Carvalho Chehab | if n + dir_level < MAX_DIR_HASH_DEPTH / 2, 66689272ca1SMauro Carvalho Chehab # of buckets in level #n = | 66789272ca1SMauro Carvalho Chehab `- 2^((MAX_DIR_HASH_DEPTH / 2) - 1), 66889272ca1SMauro Carvalho Chehab Otherwise 66989272ca1SMauro Carvalho Chehab 67089272ca1SMauro Carvalho ChehabWhen F2FS finds a file name in a directory, at first a hash value of the file 67189272ca1SMauro Carvalho Chehabname is calculated. Then, F2FS scans the hash table in level #0 to find the 67289272ca1SMauro Carvalho Chehabdentry consisting of the file name and its inode number. If not found, F2FS 67389272ca1SMauro Carvalho Chehabscans the next hash table in level #1. In this way, F2FS scans hash tables in 674ca313c82SRandy Dunlapeach levels incrementally from 1 to N. In each level F2FS needs to scan only 67589272ca1SMauro Carvalho Chehabone bucket determined by the following equation, which shows O(log(# of files)) 67689272ca1SMauro Carvalho Chehabcomplexity:: 67789272ca1SMauro Carvalho Chehab 67889272ca1SMauro Carvalho Chehab bucket number to scan in level #n = (hash value) % (# of buckets in level #n) 67989272ca1SMauro Carvalho Chehab 68089272ca1SMauro Carvalho ChehabIn the case of file creation, F2FS finds empty consecutive slots that cover the 68189272ca1SMauro Carvalho Chehabfile name. F2FS searches the empty slots in the hash tables of whole levels from 68289272ca1SMauro Carvalho Chehab1 to N in the same way as the lookup operation. 68389272ca1SMauro Carvalho Chehab 68489272ca1SMauro Carvalho ChehabThe following figure shows an example of two cases holding children:: 68589272ca1SMauro Carvalho Chehab 68689272ca1SMauro Carvalho Chehab --------------> Dir <-------------- 68789272ca1SMauro Carvalho Chehab | | 68889272ca1SMauro Carvalho Chehab child child 68989272ca1SMauro Carvalho Chehab 69089272ca1SMauro Carvalho Chehab child - child [hole] - child 69189272ca1SMauro Carvalho Chehab 69289272ca1SMauro Carvalho Chehab child - child - child [hole] - [hole] - child 69389272ca1SMauro Carvalho Chehab 69489272ca1SMauro Carvalho Chehab Case 1: Case 2: 69589272ca1SMauro Carvalho Chehab Number of children = 6, Number of children = 3, 69689272ca1SMauro Carvalho Chehab File size = 7 File size = 7 69789272ca1SMauro Carvalho Chehab 69889272ca1SMauro Carvalho ChehabDefault Block Allocation 69989272ca1SMauro Carvalho Chehab------------------------ 70089272ca1SMauro Carvalho Chehab 70189272ca1SMauro Carvalho ChehabAt runtime, F2FS manages six active logs inside "Main" area: Hot/Warm/Cold node 70289272ca1SMauro Carvalho Chehaband Hot/Warm/Cold data. 70389272ca1SMauro Carvalho Chehab 70489272ca1SMauro Carvalho Chehab- Hot node contains direct node blocks of directories. 70589272ca1SMauro Carvalho Chehab- Warm node contains direct node blocks except hot node blocks. 70689272ca1SMauro Carvalho Chehab- Cold node contains indirect node blocks 70789272ca1SMauro Carvalho Chehab- Hot data contains dentry blocks 70889272ca1SMauro Carvalho Chehab- Warm data contains data blocks except hot and cold data blocks 70989272ca1SMauro Carvalho Chehab- Cold data contains multimedia data or migrated data blocks 71089272ca1SMauro Carvalho Chehab 71189272ca1SMauro Carvalho ChehabLFS has two schemes for free space management: threaded log and copy-and-compac- 71289272ca1SMauro Carvalho Chehabtion. The copy-and-compaction scheme which is known as cleaning, is well-suited 71389272ca1SMauro Carvalho Chehabfor devices showing very good sequential write performance, since free segments 71489272ca1SMauro Carvalho Chehabare served all the time for writing new data. However, it suffers from cleaning 71589272ca1SMauro Carvalho Chehaboverhead under high utilization. Contrarily, the threaded log scheme suffers 71689272ca1SMauro Carvalho Chehabfrom random writes, but no cleaning process is needed. F2FS adopts a hybrid 71789272ca1SMauro Carvalho Chehabscheme where the copy-and-compaction scheme is adopted by default, but the 71889272ca1SMauro Carvalho Chehabpolicy is dynamically changed to the threaded log scheme according to the file 71989272ca1SMauro Carvalho Chehabsystem status. 72089272ca1SMauro Carvalho Chehab 72189272ca1SMauro Carvalho ChehabIn order to align F2FS with underlying flash-based storage, F2FS allocates a 72289272ca1SMauro Carvalho Chehabsegment in a unit of section. F2FS expects that the section size would be the 72389272ca1SMauro Carvalho Chehabsame as the unit size of garbage collection in FTL. Furthermore, with respect 72489272ca1SMauro Carvalho Chehabto the mapping granularity in FTL, F2FS allocates each section of the active 72589272ca1SMauro Carvalho Chehablogs from different zones as much as possible, since FTL can write the data in 72689272ca1SMauro Carvalho Chehabthe active logs into one allocation unit according to its mapping granularity. 72789272ca1SMauro Carvalho Chehab 72889272ca1SMauro Carvalho ChehabCleaning process 72989272ca1SMauro Carvalho Chehab---------------- 73089272ca1SMauro Carvalho Chehab 73189272ca1SMauro Carvalho ChehabF2FS does cleaning both on demand and in the background. On-demand cleaning is 73289272ca1SMauro Carvalho Chehabtriggered when there are not enough free segments to serve VFS calls. Background 73389272ca1SMauro Carvalho Chehabcleaner is operated by a kernel thread, and triggers the cleaning job when the 73489272ca1SMauro Carvalho Chehabsystem is idle. 73589272ca1SMauro Carvalho Chehab 73689272ca1SMauro Carvalho ChehabF2FS supports two victim selection policies: greedy and cost-benefit algorithms. 73789272ca1SMauro Carvalho ChehabIn the greedy algorithm, F2FS selects a victim segment having the smallest number 73889272ca1SMauro Carvalho Chehabof valid blocks. In the cost-benefit algorithm, F2FS selects a victim segment 73989272ca1SMauro Carvalho Chehabaccording to the segment age and the number of valid blocks in order to address 74089272ca1SMauro Carvalho Chehablog block thrashing problem in the greedy algorithm. F2FS adopts the greedy 74189272ca1SMauro Carvalho Chehabalgorithm for on-demand cleaner, while background cleaner adopts cost-benefit 74289272ca1SMauro Carvalho Chehabalgorithm. 74389272ca1SMauro Carvalho Chehab 74489272ca1SMauro Carvalho ChehabIn order to identify whether the data in the victim segment are valid or not, 74589272ca1SMauro Carvalho ChehabF2FS manages a bitmap. Each bit represents the validity of a block, and the 74689272ca1SMauro Carvalho Chehabbitmap is composed of a bit stream covering whole blocks in main area. 74789272ca1SMauro Carvalho Chehab 74889272ca1SMauro Carvalho ChehabFallocate(2) Policy 74989272ca1SMauro Carvalho Chehab------------------- 75089272ca1SMauro Carvalho Chehab 751ca313c82SRandy DunlapThe default policy follows the below POSIX rule. 75289272ca1SMauro Carvalho Chehab 75389272ca1SMauro Carvalho ChehabAllocating disk space 75489272ca1SMauro Carvalho Chehab The default operation (i.e., mode is zero) of fallocate() allocates 75589272ca1SMauro Carvalho Chehab the disk space within the range specified by offset and len. The 75689272ca1SMauro Carvalho Chehab file size (as reported by stat(2)) will be changed if offset+len is 75789272ca1SMauro Carvalho Chehab greater than the file size. Any subregion within the range specified 75889272ca1SMauro Carvalho Chehab by offset and len that did not contain data before the call will be 75989272ca1SMauro Carvalho Chehab initialized to zero. This default behavior closely resembles the 76089272ca1SMauro Carvalho Chehab behavior of the posix_fallocate(3) library function, and is intended 76189272ca1SMauro Carvalho Chehab as a method of optimally implementing that function. 76289272ca1SMauro Carvalho Chehab 76389272ca1SMauro Carvalho ChehabHowever, once F2FS receives ioctl(fd, F2FS_IOC_SET_PIN_FILE) in prior to 764ca313c82SRandy Dunlapfallocate(fd, DEFAULT_MODE), it allocates on-disk block addressess having 76589272ca1SMauro Carvalho Chehabzero or random data, which is useful to the below scenario where: 76689272ca1SMauro Carvalho Chehab 76789272ca1SMauro Carvalho Chehab 1. create(fd) 76889272ca1SMauro Carvalho Chehab 2. ioctl(fd, F2FS_IOC_SET_PIN_FILE) 76989272ca1SMauro Carvalho Chehab 3. fallocate(fd, 0, 0, size) 77089272ca1SMauro Carvalho Chehab 4. address = fibmap(fd, offset) 77189272ca1SMauro Carvalho Chehab 5. open(blkdev) 77289272ca1SMauro Carvalho Chehab 6. write(blkdev, address) 77389272ca1SMauro Carvalho Chehab 77489272ca1SMauro Carvalho ChehabCompression implementation 77589272ca1SMauro Carvalho Chehab-------------------------- 77689272ca1SMauro Carvalho Chehab 77789272ca1SMauro Carvalho Chehab- New term named cluster is defined as basic unit of compression, file can 77889272ca1SMauro Carvalho Chehab be divided into multiple clusters logically. One cluster includes 4 << n 77989272ca1SMauro Carvalho Chehab (n >= 0) logical pages, compression size is also cluster size, each of 78089272ca1SMauro Carvalho Chehab cluster can be compressed or not. 78189272ca1SMauro Carvalho Chehab 78289272ca1SMauro Carvalho Chehab- In cluster metadata layout, one special block address is used to indicate 783ca313c82SRandy Dunlap a cluster is a compressed one or normal one; for compressed cluster, following 78489272ca1SMauro Carvalho Chehab metadata maps cluster to [1, 4 << n - 1] physical blocks, in where f2fs 78589272ca1SMauro Carvalho Chehab stores data including compress header and compressed data. 78689272ca1SMauro Carvalho Chehab 78789272ca1SMauro Carvalho Chehab- In order to eliminate write amplification during overwrite, F2FS only 78889272ca1SMauro Carvalho Chehab support compression on write-once file, data can be compressed only when 7894fc781a3SChao Yu all logical blocks in cluster contain valid data and compress ratio of 7904fc781a3SChao Yu cluster data is lower than specified threshold. 79189272ca1SMauro Carvalho Chehab 792151b1982SFengnan Chang- To enable compression on regular inode, there are four ways: 79389272ca1SMauro Carvalho Chehab 79489272ca1SMauro Carvalho Chehab * chattr +c file 79589272ca1SMauro Carvalho Chehab * chattr +c dir; touch dir/file 79689272ca1SMauro Carvalho Chehab * mount w/ -o compress_extension=ext; touch file.ext 79738740707SChao Yu * mount w/ -o compress_extension=*; touch any_file 79838740707SChao Yu 799151b1982SFengnan Chang- To disable compression on regular inode, there are two ways: 800151b1982SFengnan Chang 801151b1982SFengnan Chang * chattr -c file 802151b1982SFengnan Chang * mount w/ -o nocompress_extension=ext; touch file.ext 803151b1982SFengnan Chang 804151b1982SFengnan Chang- Priority in between FS_COMPR_FL, FS_NOCOMP_FS, extensions: 805151b1982SFengnan Chang 806151b1982SFengnan Chang * compress_extension=so; nocompress_extension=zip; chattr +c dir; touch 807151b1982SFengnan Chang dir/foo.so; touch dir/bar.zip; touch dir/baz.txt; then foo.so and baz.txt 808151b1982SFengnan Chang should be compresse, bar.zip should be non-compressed. chattr +c dir/bar.zip 809151b1982SFengnan Chang can enable compress on bar.zip. 810151b1982SFengnan Chang * compress_extension=so; nocompress_extension=zip; chattr -c dir; touch 811151b1982SFengnan Chang dir/foo.so; touch dir/bar.zip; touch dir/baz.txt; then foo.so should be 812151b1982SFengnan Chang compresse, bar.zip and baz.txt should be non-compressed. 813151b1982SFengnan Chang chattr+c dir/bar.zip; chattr+c dir/baz.txt; can enable compress on bar.zip 814151b1982SFengnan Chang and baz.txt. 815151b1982SFengnan Chang 81638740707SChao Yu- At this point, compression feature doesn't expose compressed space to user 81738740707SChao Yu directly in order to guarantee potential data updates later to the space. 81838740707SChao Yu Instead, the main goal is to reduce data writes to flash disk as much as 81938740707SChao Yu possible, resulting in extending disk life time as well as relaxing IO 8204a4fc043SFengnan Chang congestion. Alternatively, we've added ioctl(F2FS_IOC_RELEASE_COMPRESS_BLOCKS) 821*d218bee8SChao Liu interface to reclaim compressed space and show it to user after setting a 822*d218bee8SChao Liu special flag to the inode. Once the compressed space is released, the flag 823*d218bee8SChao Liu will block writing data to the file until either the compressed space is 824*d218bee8SChao Liu reserved via ioctl(F2FS_IOC_RESERVE_COMPRESS_BLOCKS) or the file size is 825*d218bee8SChao Liu truncated to zero. 82689272ca1SMauro Carvalho Chehab 82789272ca1SMauro Carvalho ChehabCompress metadata layout:: 82889272ca1SMauro Carvalho Chehab 82989272ca1SMauro Carvalho Chehab [Dnode Structure] 83089272ca1SMauro Carvalho Chehab +-----------------------------------------------+ 83189272ca1SMauro Carvalho Chehab | cluster 1 | cluster 2 | ......... | cluster N | 83289272ca1SMauro Carvalho Chehab +-----------------------------------------------+ 83389272ca1SMauro Carvalho Chehab . . . . 83489272ca1SMauro Carvalho Chehab . . . . 83589272ca1SMauro Carvalho Chehab . Compressed Cluster . . Normal Cluster . 83689272ca1SMauro Carvalho Chehab +----------+---------+---------+---------+ +---------+---------+---------+---------+ 83789272ca1SMauro Carvalho Chehab |compr flag| block 1 | block 2 | block 3 | | block 1 | block 2 | block 3 | block 4 | 83889272ca1SMauro Carvalho Chehab +----------+---------+---------+---------+ +---------+---------+---------+---------+ 83989272ca1SMauro Carvalho Chehab . . 84089272ca1SMauro Carvalho Chehab . . 84189272ca1SMauro Carvalho Chehab . . 84289272ca1SMauro Carvalho Chehab +-------------+-------------+----------+----------------------------+ 84389272ca1SMauro Carvalho Chehab | data length | data chksum | reserved | compressed data | 84489272ca1SMauro Carvalho Chehab +-------------+-------------+----------+----------------------------+ 845de881df9SAravind Ramesh 846602a16d5SDaeho JeongCompression mode 847602a16d5SDaeho Jeong-------------------------- 848602a16d5SDaeho Jeong 849602a16d5SDaeho Jeongf2fs supports "fs" and "user" compression modes with "compression_mode" mount option. 850602a16d5SDaeho JeongWith this option, f2fs provides a choice to select the way how to compress the 851602a16d5SDaeho Jeongcompression enabled files (refer to "Compression implementation" section for how to 852602a16d5SDaeho Jeongenable compression on a regular inode). 853602a16d5SDaeho Jeong 854602a16d5SDaeho Jeong1) compress_mode=fs 855602a16d5SDaeho JeongThis is the default option. f2fs does automatic compression in the writeback of the 856602a16d5SDaeho Jeongcompression enabled files. 857602a16d5SDaeho Jeong 858602a16d5SDaeho Jeong2) compress_mode=user 859092af2ebSEd TsaiThis disables the automatic compression and gives the user discretion of choosing the 860602a16d5SDaeho Jeongtarget file and the timing. The user can do manual compression/decompression on the 861602a16d5SDaeho Jeongcompression enabled files using F2FS_IOC_DECOMPRESS_FILE and F2FS_IOC_COMPRESS_FILE 862602a16d5SDaeho Jeongioctls like the below. 863602a16d5SDaeho Jeong 864602a16d5SDaeho JeongTo decompress a file, 865602a16d5SDaeho Jeong 866602a16d5SDaeho Jeongfd = open(filename, O_WRONLY, 0); 867602a16d5SDaeho Jeongret = ioctl(fd, F2FS_IOC_DECOMPRESS_FILE); 868602a16d5SDaeho Jeong 869602a16d5SDaeho JeongTo compress a file, 870602a16d5SDaeho Jeong 871602a16d5SDaeho Jeongfd = open(filename, O_WRONLY, 0); 872602a16d5SDaeho Jeongret = ioctl(fd, F2FS_IOC_COMPRESS_FILE); 873602a16d5SDaeho Jeong 874de881df9SAravind RameshNVMe Zoned Namespace devices 875de881df9SAravind Ramesh---------------------------- 876de881df9SAravind Ramesh 877de881df9SAravind Ramesh- ZNS defines a per-zone capacity which can be equal or less than the 878de881df9SAravind Ramesh zone-size. Zone-capacity is the number of usable blocks in the zone. 879ca313c82SRandy Dunlap F2FS checks if zone-capacity is less than zone-size, if it is, then any 880de881df9SAravind Ramesh segment which starts after the zone-capacity is marked as not-free in 881de881df9SAravind Ramesh the free segment bitmap at initial mount time. These segments are marked 882de881df9SAravind Ramesh as permanently used so they are not allocated for writes and 883de881df9SAravind Ramesh consequently are not needed to be garbage collected. In case the 884de881df9SAravind Ramesh zone-capacity is not aligned to default segment size(2MB), then a segment 885de881df9SAravind Ramesh can start before the zone-capacity and span across zone-capacity boundary. 886de881df9SAravind Ramesh Such spanning segments are also considered as usable segments. All blocks 887de881df9SAravind Ramesh past the zone-capacity are considered unusable in these segments. 888