1 ================================================================================
2 WHAT IS Flash-Friendly File System (F2FS)?
3 ================================================================================
5 NAND flash memory-based storage devices, such as SSD, eMMC, and SD cards, have
6 been equipped on a variety systems ranging from mobile to server systems. Since
7 they are known to have different characteristics from the conventional rotating
8 disks, a file system, an upper layer to the storage device, should adapt to the
9 changes from the sketch in the design level.
11 F2FS is a file system exploiting NAND flash memory-based storage devices, which
12 is based on Log-structured File System (LFS). The design has been focused on
13 addressing the fundamental issues in LFS, which are snowball effect of wandering
14 tree and high cleaning overhead.
16 Since a NAND flash memory-based storage device shows different characteristic
17 according to its internal geometry or flash memory management scheme, namely FTL,
18 F2FS and its tools support various parameters not only for configuring on-disk
19 layout, but also for selecting allocation and cleaning algorithms.
21 The following git tree provides the file system formatting tool (mkfs.f2fs),
22 a consistency checking tool (fsck.f2fs), and a debugging tool (dump.f2fs).
23 >> git://git.kernel.org/pub/scm/linux/kernel/git/jaegeuk/f2fs-tools.git
25 For reporting bugs and sending patches, please use the following mailing list:
26 >> linux-f2fs-devel@lists.sourceforge.net
28 ================================================================================
29 BACKGROUND AND DESIGN ISSUES
30 ================================================================================
32 Log-structured File System (LFS)
33 --------------------------------
34 "A log-structured file system writes all modifications to disk sequentially in
35 a log-like structure, thereby speeding up both file writing and crash recovery.
36 The log is the only structure on disk; it contains indexing information so that
37 files can be read back from the log efficiently. In order to maintain large free
38 areas on disk for fast writing, we divide the log into segments and use a
39 segment cleaner to compress the live information from heavily fragmented
40 segments." from Rosenblum, M. and Ousterhout, J. K., 1992, "The design and
41 implementation of a log-structured file system", ACM Trans. Computer Systems
44 Wandering Tree Problem
45 ----------------------
46 In LFS, when a file data is updated and written to the end of log, its direct
47 pointer block is updated due to the changed location. Then the indirect pointer
48 block is also updated due to the direct pointer block update. In this manner,
49 the upper index structures such as inode, inode map, and checkpoint block are
50 also updated recursively. This problem is called as wandering tree problem [1],
51 and in order to enhance the performance, it should eliminate or relax the update
52 propagation as much as possible.
54 [1] Bityutskiy, A. 2005. JFFS3 design issues. http://www.linux-mtd.infradead.org/
58 Since LFS is based on out-of-place writes, it produces so many obsolete blocks
59 scattered across the whole storage. In order to serve new empty log space, it
60 needs to reclaim these obsolete blocks seamlessly to users. This job is called
61 as a cleaning process.
63 The process consists of three operations as follows.
64 1. A victim segment is selected through referencing segment usage table.
65 2. It loads parent index structures of all the data in the victim identified by
66 segment summary blocks.
67 3. It checks the cross-reference between the data and its parent index structure.
68 4. It moves valid data selectively.
70 This cleaning job may cause unexpected long delays, so the most important goal
71 is to hide the latencies to users. And also definitely, it should reduce the
72 amount of valid data to be moved, and move them quickly as well.
74 ================================================================================
76 ================================================================================
80 - Enlarge the random write area for better performance, but provide the high
82 - Align FS data structures to the operational units in FTL as best efforts
84 Wandering Tree Problem
85 ----------------------
86 - Use a term, “node”, that represents inodes as well as various pointer blocks
87 - Introduce Node Address Table (NAT) containing the locations of all the “node”
88 blocks; this will cut off the update propagation.
92 - Support a background cleaning process
93 - Support greedy and cost-benefit algorithms for victim selection policies
94 - Support multi-head logs for static/dynamic hot and cold data separation
95 - Introduce adaptive logging for efficient block allocation
97 ================================================================================
99 ================================================================================
101 background_gc=%s Turn on/off cleaning operations, namely garbage
102 collection, triggered in background when I/O subsystem is
103 idle. If background_gc=on, it will turn on the garbage
104 collection and if background_gc=off, garbage collection
106 Default value for this option is on. So garbage
107 collection is on by default.
108 disable_roll_forward Disable the roll-forward recovery routine
109 discard Issue discard/TRIM commands when a segment is cleaned.
110 no_heap Disable heap-style segment allocation which finds free
111 segments for data from the beginning of main area, while
112 for node from the end of main area.
113 nouser_xattr Disable Extended User Attributes. Note: xattr is enabled
114 by default if CONFIG_F2FS_FS_XATTR is selected.
115 noacl Disable POSIX Access Control List. Note: acl is enabled
116 by default if CONFIG_F2FS_FS_POSIX_ACL is selected.
117 active_logs=%u Support configuring the number of active logs. In the
118 current design, f2fs supports only 2, 4, and 6 logs.
120 disable_ext_identify Disable the extension list configured by mkfs, so f2fs
121 does not aware of cold files such as media files.
122 inline_xattr Enable the inline xattrs feature.
123 inline_data Enable the inline data feature: New created small(<~3.4k)
124 files can be written into inode block.
125 flush_merge Merge concurrent cache_flush commands as much as possible
126 to eliminate redundant command issues. If the underlying
127 device handles the cache_flush command relatively slowly,
128 recommend to enable this option.
129 nobarrier This option can be used if underlying storage guarantees
130 its cached data should be written to the novolatile area.
131 If this option is set, no cache_flush commands are issued
132 but f2fs still guarantees the write ordering of all the
135 ================================================================================
137 ================================================================================
139 /sys/kernel/debug/f2fs/ contains information about all the partitions mounted as
140 f2fs. Each file shows the whole f2fs information.
142 /sys/kernel/debug/f2fs/status includes:
143 - major file system information managed by f2fs currently
144 - average SIT information about whole segments
145 - current memory footprint consumed by f2fs.
147 ================================================================================
149 ================================================================================
151 Information about mounted f2f2 file systems can be found in
152 /sys/fs/f2fs. Each mounted filesystem will have a directory in
153 /sys/fs/f2fs based on its device name (i.e., /sys/fs/f2fs/sda).
154 The files in each per-device directory are shown in table below.
156 Files in /sys/fs/f2fs/<devname>
157 (see also Documentation/ABI/testing/sysfs-fs-f2fs)
158 ..............................................................................
161 gc_max_sleep_time This tuning parameter controls the maximum sleep
162 time for the garbage collection thread. Time is
165 gc_min_sleep_time This tuning parameter controls the minimum sleep
166 time for the garbage collection thread. Time is
169 gc_no_gc_sleep_time This tuning parameter controls the default sleep
170 time for the garbage collection thread. Time is
173 gc_idle This parameter controls the selection of victim
174 policy for garbage collection. Setting gc_idle = 0
175 (default) will disable this option. Setting
176 gc_idle = 1 will select the Cost Benefit approach
177 & setting gc_idle = 2 will select the greedy aproach.
179 reclaim_segments This parameter controls the number of prefree
180 segments to be reclaimed. If the number of prefree
181 segments is larger than the number of segments
182 in the proportion to the percentage over total
183 volume size, f2fs tries to conduct checkpoint to
184 reclaim the prefree segments to free segments.
185 By default, 5% over total # of segments.
187 max_small_discards This parameter controls the number of discard
188 commands that consist small blocks less than 2MB.
189 The candidates to be discarded are cached until
190 checkpoint is triggered, and issued during the
191 checkpoint. By default, it is disabled with 0.
193 ipu_policy This parameter controls the policy of in-place
194 updates in f2fs. There are five policies:
195 0: F2FS_IPU_FORCE, 1: F2FS_IPU_SSR,
196 2: F2FS_IPU_UTIL, 3: F2FS_IPU_SSR_UTIL,
199 min_ipu_util This parameter controls the threshold to trigger
200 in-place-updates. The number indicates percentage
201 of the filesystem utilization, and used by
202 F2FS_IPU_UTIL and F2FS_IPU_SSR_UTIL policies.
204 max_victim_search This parameter controls the number of trials to
205 find a victim segment when conducting SSR and
206 cleaning operations. The default value is 4096
207 which covers 8GB block address range.
209 dir_level This parameter controls the directory level to
210 support large directory. If a directory has a
211 number of files, it can reduce the file lookup
212 latency by increasing this dir_level value.
213 Otherwise, it needs to decrease this value to
214 reduce the space overhead. The default value is 0.
216 ram_thresh This parameter controls the memory footprint used
217 by free nids and cached nat entries. By default,
218 10 is set, which indicates 10 MB / 1 GB RAM.
220 ================================================================================
222 ================================================================================
224 1. Download userland tools and compile them.
226 2. Skip, if f2fs was compiled statically inside kernel.
227 Otherwise, insert the f2fs.ko module.
230 3. Create a directory trying to mount
233 4. Format the block device, and then mount as f2fs
234 # mkfs.f2fs -l label /dev/block_device
235 # mount -t f2fs /dev/block_device /mnt/f2fs
239 The mkfs.f2fs is for the use of formatting a partition as the f2fs filesystem,
240 which builds a basic on-disk layout.
242 The options consist of:
243 -l [label] : Give a volume label, up to 512 unicode name.
244 -a [0 or 1] : Split start location of each area for heap-based allocation.
245 1 is set by default, which performs this.
246 -o [int] : Set overprovision ratio in percent over volume size.
248 -s [int] : Set the number of segments per section.
250 -z [int] : Set the number of sections per zone.
252 -e [str] : Set basic extension list. e.g. "mp3,gif,mov"
253 -t [0 or 1] : Disable discard command or not.
254 1 is set by default, which conducts discard.
258 The fsck.f2fs is a tool to check the consistency of an f2fs-formatted
259 partition, which examines whether the filesystem metadata and user-made data
260 are cross-referenced correctly or not.
261 Note that, initial version of the tool does not fix any inconsistency.
263 The options consist of:
264 -d debug level [default:0]
268 The dump.f2fs shows the information of specific inode and dumps SSA and SIT to
269 file. Each file is dump_ssa and dump_sit.
271 The dump.f2fs is used to debug on-disk data structures of the f2fs filesystem.
272 It shows on-disk inode information reconized by a given inode number, and is
273 able to dump all the SSA and SIT entries into predefined files, ./dump_ssa and
274 ./dump_sit respectively.
276 The options consist of:
277 -d debug level [default:0]
279 -s [SIT dump segno from #1~#2 (decimal), for all 0~-1]
280 -a [SSA dump segno from #1~#2 (decimal), for all 0~-1]
283 # dump.f2fs -i [ino] /dev/sdx
284 # dump.f2fs -s 0~-1 /dev/sdx (SIT dump)
285 # dump.f2fs -a 0~-1 /dev/sdx (SSA dump)
287 ================================================================================
289 ================================================================================
294 F2FS divides the whole volume into a number of segments, each of which is fixed
295 to 2MB in size. A section is composed of consecutive segments, and a zone
296 consists of a set of sections. By default, section and zone sizes are set to one
297 segment size identically, but users can easily modify the sizes by mkfs.
299 F2FS splits the entire volume into six areas, and all the areas except superblock
300 consists of multiple segments as described below.
302 align with the zone size <-|
303 |-> align with the segment size
304 _________________________________________________________________________
305 | | | Segment | Node | Segment | |
306 | Superblock | Checkpoint | Info. | Address | Summary | Main |
307 | (SB) | (CP) | Table (SIT) | Table (NAT) | Area (SSA) | |
308 |____________|_____2______|______N______|______N______|______N_____|__N___|
312 ._________________________________________.
313 |_Segment_|_..._|_Segment_|_..._|_Segment_|
322 : It is located at the beginning of the partition, and there exist two copies
323 to avoid file system crash. It contains basic partition information and some
324 default parameters of f2fs.
327 : It contains file system information, bitmaps for valid NAT/SIT sets, orphan
328 inode lists, and summary entries of current active segments.
330 - Segment Information Table (SIT)
331 : It contains segment information such as valid block count and bitmap for the
332 validity of all the blocks.
334 - Node Address Table (NAT)
335 : It is composed of a block address table for all the node blocks stored in
338 - Segment Summary Area (SSA)
339 : It contains summary entries which contains the owner information of all the
340 data and node blocks stored in Main area.
343 : It contains file and directory data including their indices.
345 In order to avoid misalignment between file system and flash-based storage, F2FS
346 aligns the start block address of CP with the segment size. Also, it aligns the
347 start block address of Main area with the zone size by reserving some segments
350 Reference the following survey for additional technical details.
351 https://wiki.linaro.org/WorkingGroups/Kernel/Projects/FlashCardSurvey
353 File System Metadata Structure
354 ------------------------------
356 F2FS adopts the checkpointing scheme to maintain file system consistency. At
357 mount time, F2FS first tries to find the last valid checkpoint data by scanning
358 CP area. In order to reduce the scanning time, F2FS uses only two copies of CP.
359 One of them always indicates the last valid data, which is called as shadow copy
360 mechanism. In addition to CP, NAT and SIT also adopt the shadow copy mechanism.
362 For file system consistency, each CP points to which NAT and SIT copies are
363 valid, as shown as below.
365 +--------+----------+---------+
367 +--------+----------+---------+
371 +-------+-------+--------+--------+--------+--------+
372 | CP #0 | CP #1 | SIT #0 | SIT #1 | NAT #0 | NAT #1 |
373 +-------+-------+--------+--------+--------+--------+
376 `----------------------------------------'
381 The key data structure to manage the data locations is a "node". Similar to
382 traditional file structures, F2FS has three types of node: inode, direct node,
383 indirect node. F2FS assigns 4KB to an inode block which contains 923 data block
384 indices, two direct node pointers, two indirect node pointers, and one double
385 indirect node pointer as described below. One direct node block contains 1018
386 data blocks, and one indirect node block contains also 1018 node blocks. Thus,
387 one inode block (i.e., a file) covers:
389 4KB * (923 + 2 * 1018 + 2 * 1018 * 1018 + 1018 * 1018 * 1018) := 3.94TB.
396 | `- direct node (1018)
398 `- double indirect node (1)
399 `- indirect node (1018)
400 `- direct node (1018)
403 Note that, all the node blocks are mapped by NAT which means the location of
404 each node is translated by the NAT table. In the consideration of the wandering
405 tree problem, F2FS is able to cut off the propagation of node updates caused by
411 A directory entry occupies 11 bytes, which consists of the following attributes.
413 - hash hash value of the file name
415 - len the length of file name
416 - type file type such as directory, symlink, etc
418 A dentry block consists of 214 dentry slots and file names. Therein a bitmap is
419 used to represent whether each dentry is valid or not. A dentry block occupies
420 4KB with the following composition.
422 Dentry Block(4 K) = bitmap (27 bytes) + reserved (3 bytes) +
423 dentries(11 * 214 bytes) + file name (8 * 214 bytes)
426 +--------------------------------+
427 |dentry block 1 | dentry block 2 |
428 +--------------------------------+
431 . [Dentry Block Structure: 4KB] .
432 +--------+----------+----------+------------+
433 | bitmap | reserved | dentries | file names |
434 +--------+----------+----------+------------+
435 [Dentry Block: 4KB] . .
438 +------+------+-----+------+
439 | hash | ino | len | type |
440 +------+------+-----+------+
441 [Dentry Structure: 11 bytes]
443 F2FS implements multi-level hash tables for directory structure. Each level has
444 a hash table with dedicated number of hash buckets as shown below. Note that
445 "A(2B)" means a bucket includes 2 data blocks.
447 ----------------------
450 N : MAX_DIR_HASH_DEPTH
451 ----------------------
455 level #1 | A(2B) - A(2B)
457 level #2 | A(2B) - A(2B) - A(2B) - A(2B)
459 level #N/2 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B)
461 level #N | A(4B) - A(4B) - A(4B) - A(4B) - A(4B) - ... - A(4B)
463 The number of blocks and buckets are determined by,
465 ,- 2, if n < MAX_DIR_HASH_DEPTH / 2,
466 # of blocks in level #n = |
469 ,- 2^(n + dir_level),
470 | if n + dir_level < MAX_DIR_HASH_DEPTH / 2,
471 # of buckets in level #n = |
472 `- 2^((MAX_DIR_HASH_DEPTH / 2) - 1),
475 When F2FS finds a file name in a directory, at first a hash value of the file
476 name is calculated. Then, F2FS scans the hash table in level #0 to find the
477 dentry consisting of the file name and its inode number. If not found, F2FS
478 scans the next hash table in level #1. In this way, F2FS scans hash tables in
479 each levels incrementally from 1 to N. In each levels F2FS needs to scan only
480 one bucket determined by the following equation, which shows O(log(# of files))
483 bucket number to scan in level #n = (hash value) % (# of buckets in level #n)
485 In the case of file creation, F2FS finds empty consecutive slots that cover the
486 file name. F2FS searches the empty slots in the hash tables of whole levels from
487 1 to N in the same way as the lookup operation.
489 The following figure shows an example of two cases holding children.
490 --------------> Dir <--------------
494 child - child [hole] - child
496 child - child - child [hole] - [hole] - child
499 Number of children = 6, Number of children = 3,
500 File size = 7 File size = 7
502 Default Block Allocation
503 ------------------------
505 At runtime, F2FS manages six active logs inside "Main" area: Hot/Warm/Cold node
506 and Hot/Warm/Cold data.
508 - Hot node contains direct node blocks of directories.
509 - Warm node contains direct node blocks except hot node blocks.
510 - Cold node contains indirect node blocks
511 - Hot data contains dentry blocks
512 - Warm data contains data blocks except hot and cold data blocks
513 - Cold data contains multimedia data or migrated data blocks
515 LFS has two schemes for free space management: threaded log and copy-and-compac-
516 tion. The copy-and-compaction scheme which is known as cleaning, is well-suited
517 for devices showing very good sequential write performance, since free segments
518 are served all the time for writing new data. However, it suffers from cleaning
519 overhead under high utilization. Contrarily, the threaded log scheme suffers
520 from random writes, but no cleaning process is needed. F2FS adopts a hybrid
521 scheme where the copy-and-compaction scheme is adopted by default, but the
522 policy is dynamically changed to the threaded log scheme according to the file
525 In order to align F2FS with underlying flash-based storage, F2FS allocates a
526 segment in a unit of section. F2FS expects that the section size would be the
527 same as the unit size of garbage collection in FTL. Furthermore, with respect
528 to the mapping granularity in FTL, F2FS allocates each section of the active
529 logs from different zones as much as possible, since FTL can write the data in
530 the active logs into one allocation unit according to its mapping granularity.
535 F2FS does cleaning both on demand and in the background. On-demand cleaning is
536 triggered when there are not enough free segments to serve VFS calls. Background
537 cleaner is operated by a kernel thread, and triggers the cleaning job when the
540 F2FS supports two victim selection policies: greedy and cost-benefit algorithms.
541 In the greedy algorithm, F2FS selects a victim segment having the smallest number
542 of valid blocks. In the cost-benefit algorithm, F2FS selects a victim segment
543 according to the segment age and the number of valid blocks in order to address
544 log block thrashing problem in the greedy algorithm. F2FS adopts the greedy
545 algorithm for on-demand cleaner, while background cleaner adopts cost-benefit
548 In order to identify whether the data in the victim segment are valid or not,
549 F2FS manages a bitmap. Each bit represents the validity of a block, and the
550 bitmap is composed of a bit stream covering whole blocks in main area.