1 2Ext4 Filesystem 3=============== 4 5Ext4 is an an advanced level of the ext3 filesystem which incorporates 6scalability and reliability enhancements for supporting large filesystems 7(64 bit) in keeping with increasing disk capacities and state-of-the-art 8feature requirements. 9 10Mailing list: linux-ext4@vger.kernel.org 11Web site: http://ext4.wiki.kernel.org 12 13 141. Quick usage instructions: 15=========================== 16 17Note: More extensive information for getting started with ext4 can be 18 found at the ext4 wiki site at the URL: 19 http://ext4.wiki.kernel.org/index.php/Ext4_Howto 20 21 - Compile and install the latest version of e2fsprogs (as of this 22 writing version 1.41.3) from: 23 24 http://sourceforge.net/project/showfiles.php?group_id=2406 25 26 or 27 28 ftp://ftp.kernel.org/pub/linux/kernel/people/tytso/e2fsprogs/ 29 30 or grab the latest git repository from: 31 32 git://git.kernel.org/pub/scm/fs/ext2/e2fsprogs.git 33 34 - Note that it is highly important to install the mke2fs.conf file 35 that comes with the e2fsprogs 1.41.x sources in /etc/mke2fs.conf. If 36 you have edited the /etc/mke2fs.conf file installed on your system, 37 you will need to merge your changes with the version from e2fsprogs 38 1.41.x. 39 40 - Create a new filesystem using the ext4 filesystem type: 41 42 # mke2fs -t ext4 /dev/hda1 43 44 Or to configure an existing ext3 filesystem to support extents: 45 46 # tune2fs -O extents /dev/hda1 47 48 If the filesystem was created with 128 byte inodes, it can be 49 converted to use 256 byte for greater efficiency via: 50 51 # tune2fs -I 256 /dev/hda1 52 53 (Note: we currently do not have tools to convert an ext4 54 filesystem back to ext3; so please do not do try this on production 55 filesystems.) 56 57 - Mounting: 58 59 # mount -t ext4 /dev/hda1 /wherever 60 61 - When comparing performance with other filesystems, it's always 62 important to try multiple workloads; very often a subtle change in a 63 workload parameter can completely change the ranking of which 64 filesystems do well compared to others. When comparing versus ext3, 65 note that ext4 enables write barriers by default, while ext3 does 66 not enable write barriers by default. So it is useful to use 67 explicitly specify whether barriers are enabled or not when via the 68 '-o barriers=[0|1]' mount option for both ext3 and ext4 filesystems 69 for a fair comparison. When tuning ext3 for best benchmark numbers, 70 it is often worthwhile to try changing the data journaling mode; '-o 71 data=writeback,nobh' can be faster for some workloads. (Note 72 however that running mounted with data=writeback can potentially 73 leave stale data exposed in recently written files in case of an 74 unclean shutdown, which could be a security exposure in some 75 situations.) Configuring the filesystem with a large journal can 76 also be helpful for metadata-intensive workloads. 77 782. Features 79=========== 80 812.1 Currently available 82 83* ability to use filesystems > 16TB (e2fsprogs support not available yet) 84* extent format reduces metadata overhead (RAM, IO for access, transactions) 85* extent format more robust in face of on-disk corruption due to magics, 86* internal redundancy in tree 87* improved file allocation (multi-block alloc) 88* lift 32000 subdirectory limit imposed by i_links_count[1] 89* nsec timestamps for mtime, atime, ctime, create time 90* inode version field on disk (NFSv4, Lustre) 91* reduced e2fsck time via uninit_bg feature 92* journal checksumming for robustness, performance 93* persistent file preallocation (e.g for streaming media, databases) 94* ability to pack bitmaps and inode tables into larger virtual groups via the 95 flex_bg feature 96* large file support 97* Inode allocation using large virtual block groups via flex_bg 98* delayed allocation 99* large block (up to pagesize) support 100* efficent new ordered mode in JBD2 and ext4(avoid using buffer head to force 101 the ordering) 102 103[1] Filesystems with a block size of 1k may see a limit imposed by the 104directory hash tree having a maximum depth of two. 105 1062.2 Candidate features for future inclusion 107 108* Online defrag (patches available but not well tested) 109* reduced mke2fs time via lazy itable initialization in conjuction with 110 the uninit_bg feature (capability to do this is available in e2fsprogs 111 but a kernel thread to do lazy zeroing of unused inode table blocks 112 after filesystem is first mounted is required for safety) 113 114There are several others under discussion, whether they all make it in is 115partly a function of how much time everyone has to work on them. Features like 116metadata checksumming have been discussed and planned for a bit but no patches 117exist yet so I'm not sure they're in the near-term roadmap. 118 119The big performance win will come with mballoc, delalloc and flex_bg 120grouping of bitmaps and inode tables. Some test results available here: 121 122 - http://www.bullopensource.org/ext4/20080818-ffsb/ffsb-write-2.6.27-rc1.html 123 - http://www.bullopensource.org/ext4/20080818-ffsb/ffsb-readwrite-2.6.27-rc1.html 124 1253. Options 126========== 127 128When mounting an ext4 filesystem, the following option are accepted: 129(*) == default 130 131ro Mount filesystem read only. Note that ext4 will 132 replay the journal (and thus write to the 133 partition) even when mounted "read only". The 134 mount options "ro,noload" can be used to prevent 135 writes to the filesystem. 136 137journal_checksum Enable checksumming of the journal transactions. 138 This will allow the recovery code in e2fsck and the 139 kernel to detect corruption in the kernel. It is a 140 compatible change and will be ignored by older kernels. 141 142journal_async_commit Commit block can be written to disk without waiting 143 for descriptor blocks. If enabled older kernels cannot 144 mount the device. This will enable 'journal_checksum' 145 internally. 146 147journal=update Update the ext4 file system's journal to the current 148 format. 149 150journal_dev=devnum When the external journal device's major/minor numbers 151 have changed, this option allows the user to specify 152 the new journal location. The journal device is 153 identified through its new major/minor numbers encoded 154 in devnum. 155 156norecovery Don't load the journal on mounting. Note that 157noload if the filesystem was not unmounted cleanly, 158 skipping the journal replay will lead to the 159 filesystem containing inconsistencies that can 160 lead to any number of problems. 161 162data=journal All data are committed into the journal prior to being 163 written into the main file system. 164 165data=ordered (*) All data are forced directly out to the main file 166 system prior to its metadata being committed to the 167 journal. 168 169data=writeback Data ordering is not preserved, data may be written 170 into the main file system after its metadata has been 171 committed to the journal. 172 173commit=nrsec (*) Ext4 can be told to sync all its data and metadata 174 every 'nrsec' seconds. The default value is 5 seconds. 175 This means that if you lose your power, you will lose 176 as much as the latest 5 seconds of work (your 177 filesystem will not be damaged though, thanks to the 178 journaling). This default value (or any low value) 179 will hurt performance, but it's good for data-safety. 180 Setting it to 0 will have the same effect as leaving 181 it at the default (5 seconds). 182 Setting it to very large values will improve 183 performance. 184 185barrier=<0|1(*)> This enables/disables the use of write barriers in 186barrier(*) the jbd code. barrier=0 disables, barrier=1 enables. 187nobarrier This also requires an IO stack which can support 188 barriers, and if jbd gets an error on a barrier 189 write, it will disable again with a warning. 190 Write barriers enforce proper on-disk ordering 191 of journal commits, making volatile disk write caches 192 safe to use, at some performance penalty. If 193 your disks are battery-backed in one way or another, 194 disabling barriers may safely improve performance. 195 The mount options "barrier" and "nobarrier" can 196 also be used to enable or disable barriers, for 197 consistency with other ext4 mount options. 198 199inode_readahead_blks=n This tuning parameter controls the maximum 200 number of inode table blocks that ext4's inode 201 table readahead algorithm will pre-read into 202 the buffer cache. The default value is 32 blocks. 203 204orlov (*) This enables the new Orlov block allocator. It is 205 enabled by default. 206 207oldalloc This disables the Orlov block allocator and enables 208 the old block allocator. Orlov should have better 209 performance - we'd like to get some feedback if it's 210 the contrary for you. 211 212user_xattr Enables Extended User Attributes. Additionally, you 213 need to have extended attribute support enabled in the 214 kernel configuration (CONFIG_EXT4_FS_XATTR). See the 215 attr(5) manual page and http://acl.bestbits.at/ to 216 learn more about extended attributes. 217 218nouser_xattr Disables Extended User Attributes. 219 220acl Enables POSIX Access Control Lists support. 221 Additionally, you need to have ACL support enabled in 222 the kernel configuration (CONFIG_EXT4_FS_POSIX_ACL). 223 See the acl(5) manual page and http://acl.bestbits.at/ 224 for more information. 225 226noacl This option disables POSIX Access Control List 227 support. 228 229reservation 230 231noreservation 232 233bsddf (*) Make 'df' act like BSD. 234minixdf Make 'df' act like Minix. 235 236debug Extra debugging information is sent to syslog. 237 238abort Simulate the effects of calling ext4_abort() for 239 debugging purposes. This is normally used while 240 remounting a filesystem which is already mounted. 241 242errors=remount-ro Remount the filesystem read-only on an error. 243errors=continue Keep going on a filesystem error. 244errors=panic Panic and halt the machine if an error occurs. 245 (These mount options override the errors behavior 246 specified in the superblock, which can be configured 247 using tune2fs) 248 249data_err=ignore(*) Just print an error message if an error occurs 250 in a file data buffer in ordered mode. 251data_err=abort Abort the journal if an error occurs in a file 252 data buffer in ordered mode. 253 254grpid Give objects the same group ID as their creator. 255bsdgroups 256 257nogrpid (*) New objects have the group ID of their creator. 258sysvgroups 259 260resgid=n The group ID which may use the reserved blocks. 261 262resuid=n The user ID which may use the reserved blocks. 263 264sb=n Use alternate superblock at this location. 265 266quota These options are ignored by the filesystem. They 267noquota are used only by quota tools to recognize volumes 268grpquota where quota should be turned on. See documentation 269usrquota in the quota-tools package for more details 270 (http://sourceforge.net/projects/linuxquota). 271 272jqfmt=<quota type> These options tell filesystem details about quota 273usrjquota=<file> so that quota information can be properly updated 274grpjquota=<file> during journal replay. They replace the above 275 quota options. See documentation in the quota-tools 276 package for more details 277 (http://sourceforge.net/projects/linuxquota). 278 279bh (*) ext4 associates buffer heads to data pages to 280nobh (a) cache disk block mapping information 281 (b) link pages into transaction to provide 282 ordering guarantees. 283 "bh" option forces use of buffer heads. 284 "nobh" option tries to avoid associating buffer 285 heads (supported only for "writeback" mode). 286 287stripe=n Number of filesystem blocks that mballoc will try 288 to use for allocation size and alignment. For RAID5/6 289 systems this should be the number of data 290 disks * RAID chunk size in file system blocks. 291 292delalloc (*) Defer block allocation until just before ext4 293 writes out the block(s) in question. This 294 allows ext4 to better allocation decisions 295 more efficiently. 296nodelalloc Disable delayed allocation. Blocks are allocated 297 when the data is copied from userspace to the 298 page cache, either via the write(2) system call 299 or when an mmap'ed page which was previously 300 unallocated is written for the first time. 301 302max_batch_time=usec Maximum amount of time ext4 should wait for 303 additional filesystem operations to be batch 304 together with a synchronous write operation. 305 Since a synchronous write operation is going to 306 force a commit and then a wait for the I/O 307 complete, it doesn't cost much, and can be a 308 huge throughput win, we wait for a small amount 309 of time to see if any other transactions can 310 piggyback on the synchronous write. The 311 algorithm used is designed to automatically tune 312 for the speed of the disk, by measuring the 313 amount of time (on average) that it takes to 314 finish committing a transaction. Call this time 315 the "commit time". If the time that the 316 transaction has been running is less than the 317 commit time, ext4 will try sleeping for the 318 commit time to see if other operations will join 319 the transaction. The commit time is capped by 320 the max_batch_time, which defaults to 15000us 321 (15ms). This optimization can be turned off 322 entirely by setting max_batch_time to 0. 323 324min_batch_time=usec This parameter sets the commit time (as 325 described above) to be at least min_batch_time. 326 It defaults to zero microseconds. Increasing 327 this parameter may improve the throughput of 328 multi-threaded, synchronous workloads on very 329 fast disks, at the cost of increasing latency. 330 331journal_ioprio=prio The I/O priority (from 0 to 7, where 0 is the 332 highest priorty) which should be used for I/O 333 operations submitted by kjournald2 during a 334 commit operation. This defaults to 3, which is 335 a slightly higher priority than the default I/O 336 priority. 337 338auto_da_alloc(*) Many broken applications don't use fsync() when 339noauto_da_alloc replacing existing files via patterns such as 340 fd = open("foo.new")/write(fd,..)/close(fd)/ 341 rename("foo.new", "foo"), or worse yet, 342 fd = open("foo", O_TRUNC)/write(fd,..)/close(fd). 343 If auto_da_alloc is enabled, ext4 will detect 344 the replace-via-rename and replace-via-truncate 345 patterns and force that any delayed allocation 346 blocks are allocated such that at the next 347 journal commit, in the default data=ordered 348 mode, the data blocks of the new file are forced 349 to disk before the rename() operation is 350 committed. This provides roughly the same level 351 of guarantees as ext3, and avoids the 352 "zero-length" problem that can happen when a 353 system crashes before the delayed allocation 354 blocks are forced to disk. 355 356discard Controls whether ext4 should issue discard/TRIM 357nodiscard(*) commands to the underlying block device when 358 blocks are freed. This is useful for SSD devices 359 and sparse/thinly-provisioned LUNs, but it is off 360 by default until sufficient testing has been done. 361 362Data Mode 363========= 364There are 3 different data modes: 365 366* writeback mode 367In data=writeback mode, ext4 does not journal data at all. This mode provides 368a similar level of journaling as that of XFS, JFS, and ReiserFS in its default 369mode - metadata journaling. A crash+recovery can cause incorrect data to 370appear in files which were written shortly before the crash. This mode will 371typically provide the best ext4 performance. 372 373* ordered mode 374In data=ordered mode, ext4 only officially journals metadata, but it logically 375groups metadata information related to data changes with the data blocks into a 376single unit called a transaction. When it's time to write the new metadata 377out to disk, the associated data blocks are written first. In general, 378this mode performs slightly slower than writeback but significantly faster than journal mode. 379 380* journal mode 381data=journal mode provides full data and metadata journaling. All new data is 382written to the journal first, and then to its final location. 383In the event of a crash, the journal can be replayed, bringing both data and 384metadata into a consistent state. This mode is the slowest except when data 385needs to be read from and written to disk at the same time where it 386outperforms all others modes. Currently ext4 does not have delayed 387allocation support if this data journalling mode is selected. 388 389References 390========== 391 392kernel source: <file:fs/ext4/> 393 <file:fs/jbd2/> 394 395programs: http://e2fsprogs.sourceforge.net/ 396 397useful links: http://fedoraproject.org/wiki/ext3-devel 398 http://www.bullopensource.org/ext4/ 399 http://ext4.wiki.kernel.org/index.php/Main_Page 400 http://fedoraproject.org/wiki/Features/Ext4 401