| Message ID | 1313780769-21713-1-git-send-email-adityakali@google.com |
|---|---|
| State | Superseded, archived |
| Headers | show |
On 2011-08-19, at 1:06 PM, Aditya Kali <adityakali@google.com> wrote: > This patch is an attempt towards supporting quotas as first class > feature in ext4. It is based on the proposal at: > https://ext4.wiki.kernel.org/index.php/Design_For_1st_Class_Quota_in_Ext4 > This patch introduces a new feature - EXT4_FEATURE_RO_COMPAT_QUOTA which, when > turned on, enables quota accounting at mount time iteself. > 6) With the QUOTA feature set, since the quota inodes are hidden, some of the > utilities from quota-tools will no longer work correctly. Instead, e2fsprogs > will include support for fixing the quota files. It would be preferable from a usability point of view to have the existing quota tools to work with the new quota files instead of having to introduce new quota utilities. There are a number of ways that this could be facilitated: - make the quota files visible to userspace, but immutable - add the needed calls to the existing quota tools and use them when manipulating quotas on ext4 filesystems with this feature - add library interfaces for doing the quota changes to libext2fs (or a new very simple library) and have the quota tools link to that instead of interfacing to the kernel directly Cheers, Andreas-- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
> > 6) With the QUOTA feature set, since the quota inodes are hidden, some of the > > utilities from quota-tools will no longer work correctly. Instead, e2fsprogs > > will include support for fixing the quota files. > > It would be preferable from a usability point of view to have the existing quota tools to work with the new quota files instead of having to introduce new quota utilities. > There are a number of ways that this could be facilitated: > - make the quota files visible to userspace, but immutable IMHO direct quota file access is always bad thing, and unfortunately tools was designed that way, now we have a chance to fix that. More than less one of the goals of FCQ is to enable quota accounting on mount. So quota file's content will always be out of date due to page-cache tricks. But if we export virtual files instead of real ones this allow to solve both issues at a time. OpenVZ use this trick to handle per container quota via unmodified quotatools. -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
>> 6) With the QUOTA feature set, since the quota inodes are hidden, some of the >> utilities from quota-tools will no longer work correctly. Instead, e2fsprogs >> will include support for fixing the quota files. > > It would be preferable from a usability point of view to have the existing quota tools to work with the new quota files instead of having to introduce new quota utilities. > > There are a number of ways that this could be facilitated: > - make the quota files visible to userspace, but immutable > - add the needed calls to the existing quota tools and use them when manipulating quotas on ext4 filesystems with this feature There are only few operations of the quota tools that depend on accessing the quota files and, AFAIK, supporting hidden quota files is already in the works. I believe it is already done for some other filesystems like ocfs2. > - add library interfaces for doing the quota changes to libext2fs (or a new very simple library) and have the quota tools link to that instead of interfacing to the kernel directly > I have already sent patches to add a new quota library in e2fsprogs (this library too is majorly based on the current quota tools). With this, e2fsck can scan and fix quota files and tune2fs/mke2fs can be used to turn on/off the new quota feature. Thanks, -- Aditya > Cheers, Andreas -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Sat 20-08-11 09:58:38, Andreas Dilger wrote: > On 2011-08-19, at 1:06 PM, Aditya Kali <adityakali@google.com> wrote: > > > This patch is an attempt towards supporting quotas as first class > > feature in ext4. It is based on the proposal at: > > https://ext4.wiki.kernel.org/index.php/Design_For_1st_Class_Quota_in_Ext4 > > This patch introduces a new feature - EXT4_FEATURE_RO_COMPAT_QUOTA which, when > > turned on, enables quota accounting at mount time iteself. > > > 6) With the QUOTA feature set, since the quota inodes are hidden, some of the > > utilities from quota-tools will no longer work correctly. Instead, e2fsprogs > > will include support for fixing the quota files. > > It would be preferable from a usability point of view to have the > existing quota tools to work with the new quota files instead of having > to introduce new quota utilities. They will generally work - e.g. quota(1), edquota(8), setquota(8) commands will work just fine. Quotacheck(8) does not make sense when quota is handled as filesystem metadata and doing necessary things in e2fsck is not only logical but also faster since it reads all necessary data anyway. The only pain point is repquota(8) - that needs a new kernel interface to work reliably (for ocfs2 and xfs we workaround the inability to iterate over all quota structures from userspace by using getpwent but that's just ugly). Honza
On Fri 19-08-11 12:06:09, Aditya Kali wrote: > This patch is an attempt towards supporting quotas as first class > feature in ext4. It is based on the proposal at: > https://ext4.wiki.kernel.org/index.php/Design_For_1st_Class_Quota_in_Ext4 > This patch introduces a new feature - EXT4_FEATURE_RO_COMPAT_QUOTA which, when > turned on, enables quota accounting at mount time iteself. Also, the > quota inodes are stored in two additional superblock fields. > Some changes introduced by this patch that should be pointed out are: > 1) Two new ext4-superblock fields - s_usr_quota_inum and s_grp_quota_inum > for storing the quota inodes in use. > 2) If the QUOTA feature and corresponding quota inodes are set in superblock, > Quotas are turned on at mount time irrespective of the quota mount options. > Thus the mount options 'quota', 'usrquota' and 'grpquota' are completely > ignored with the new QUOTA feature flag. Hmm, cannot we automatically enable 'usrquota' and 'grpquota' options when we see QUOTA feature enabled? Quota tools still need these options to identify a filesystem they can work with (although another option would be to change quota tools to use GETFMT with ext4 filesystem to identify whether quotas are enabled or not). > 3) Default quota inodes are: inode#3 for tracking userquota and inode#4 for > tracking group quota. The superblock fields can be set to use other inodes > as well. > 4) mke2fs or tune2fs will initialize these inodes when quota feature is > being set. The default reserved inodes will not be visible to user as > regular files. > 5) Once quotas are turned on, they cannot be turned off while the FS is > mounted. This is because we do not want to let the quota get inconsistent. Both xfs and ocfs2 allow for mode in which usage is tracked (i.e. quota information is kept uptodate) but limits are not enforced. I imagine this might be useful in some cases and also from consistency point of view it would be nice. In ocfs2 I implemented it so that on mount, updating of quota information is enabled (when appropriate fs feature is enabled) but limits are not enforced. Quotaon/quotaoff commands then toggle enforcement of quota limits. XFS has mount options which decide whether quota information should be kept uptodate and whether it should be enforced or not. > 6) With the QUOTA feature set, since the quota inodes are hidden, some of the > utilities from quota-tools will no longer work correctly. Instead, e2fsprogs > will include support for fixing the quota files. I think it might be useful in some cases to enable only user quotas but leave group quotas disabled (it slightly reduces quota overhead - mainly because we have less quota structures to update on disk). So it would be useful to have USRQUOTA and GRPQUOTA features separate... Honza > 7) Support is only for the new V2 quota file format. > > Signed-off-by: Aditya Kali <adityakali@google.com> > --- > fs/ext4/ext4.h | 10 ++++- > fs/ext4/ext4_jbd2.h | 18 +++++++--- > fs/ext4/super.c | 89 +++++++++++++++++++++++++++++++++++++++++++++++++-- > 3 files changed, 106 insertions(+), 11 deletions(-) > > diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h > index e717dfd..5b57bc2 100644 > --- a/fs/ext4/ext4.h > +++ b/fs/ext4/ext4.h > @@ -1065,7 +1065,9 @@ struct ext4_super_block { > __u8 s_last_error_func[32]; /* function where the error happened */ > #define EXT4_S_ERR_END offsetof(struct ext4_super_block, s_mount_opts) > __u8 s_mount_opts[64]; > - __le32 s_reserved[112]; /* Padding to the end of the block */ > + __le32 s_usr_quota_inum; /* inode for tracking user quota */ > + __le32 s_grp_quota_inum; /* inode for tracking group quota */ > + __le32 s_reserved[110]; /* Padding to the end of the block */ > }; > > #define EXT4_S_ERR_LEN (EXT4_S_ERR_END - EXT4_S_ERR_START) > @@ -1142,6 +1144,7 @@ struct ext4_sb_info { > #ifdef CONFIG_QUOTA > char *s_qf_names[MAXQUOTAS]; /* Names of quota files with journalled quota */ > int s_jquota_fmt; /* Format of quota to use */ > + unsigned long s_qf_inums[MAXQUOTAS]; /* Quota file inodes */ > #endif > unsigned int s_want_extra_isize; /* New inodes should reserve # bytes */ > struct rb_root system_blks; > @@ -1241,6 +1244,8 @@ static inline struct timespec ext4_current_time(struct inode *inode) > static inline int ext4_valid_inum(struct super_block *sb, unsigned long ino) > { > return ino == EXT4_ROOT_INO || > + ino == EXT4_USR_QUOTA_INO || > + ino == EXT4_GRP_QUOTA_INO || > ino == EXT4_JOURNAL_INO || > ino == EXT4_RESIZE_INO || > (ino >= EXT4_FIRST_INO(sb) && > @@ -1401,7 +1406,8 @@ static inline void ext4_clear_state_flags(struct ext4_inode_info *ei) > EXT4_FEATURE_RO_COMPAT_DIR_NLINK | \ > EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE | \ > EXT4_FEATURE_RO_COMPAT_BTREE_DIR |\ > - EXT4_FEATURE_RO_COMPAT_HUGE_FILE) > + EXT4_FEATURE_RO_COMPAT_HUGE_FILE| \ > + EXT4_FEATURE_RO_COMPAT_QUOTA) > > /* > * Default values for user and/or group using reserved blocks > diff --git a/fs/ext4/ext4_jbd2.h b/fs/ext4/ext4_jbd2.h > index bb85757..0f097ed 100644 > --- a/fs/ext4/ext4_jbd2.h > +++ b/fs/ext4/ext4_jbd2.h > @@ -87,14 +87,20 @@ > #ifdef CONFIG_QUOTA > /* Amount of blocks needed for quota update - we know that the structure was > * allocated so we need to update only data block */ > -#define EXT4_QUOTA_TRANS_BLOCKS(sb) (test_opt(sb, QUOTA) ? 1 : 0) > +#define EXT4_QUOTA_TRANS_BLOCKS(sb) ((test_opt(sb, QUOTA) ||\ > + EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) ?\ > + 1 : 0) > /* Amount of blocks needed for quota insert/delete - we do some block writes > * but inode, sb and group updates are done only once */ > -#define EXT4_QUOTA_INIT_BLOCKS(sb) (test_opt(sb, QUOTA) ? (DQUOT_INIT_ALLOC*\ > - (EXT4_SINGLEDATA_TRANS_BLOCKS(sb)-3)+3+DQUOT_INIT_REWRITE) : 0) > - > -#define EXT4_QUOTA_DEL_BLOCKS(sb) (test_opt(sb, QUOTA) ? (DQUOT_DEL_ALLOC*\ > - (EXT4_SINGLEDATA_TRANS_BLOCKS(sb)-3)+3+DQUOT_DEL_REWRITE) : 0) > +#define EXT4_QUOTA_INIT_BLOCKS(sb) ((test_opt(sb, QUOTA) ||\ > + EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) ?\ > + (DQUOT_INIT_ALLOC*(EXT4_SINGLEDATA_TRANS_BLOCKS(sb)-3)\ > + +3+DQUOT_INIT_REWRITE) : 0) > + > +#define EXT4_QUOTA_DEL_BLOCKS(sb) ((test_opt(sb, QUOTA) ||\ > + EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) ?\ > + (DQUOT_DEL_ALLOC*(EXT4_SINGLEDATA_TRANS_BLOCKS(sb)-3)\ > + +3+DQUOT_DEL_REWRITE) : 0) > #else > #define EXT4_QUOTA_TRANS_BLOCKS(sb) 0 > #define EXT4_QUOTA_INIT_BLOCKS(sb) 0 > diff --git a/fs/ext4/super.c b/fs/ext4/super.c > index 4687fea..3f3d87e 100644 > --- a/fs/ext4/super.c > +++ b/fs/ext4/super.c > @@ -1225,6 +1225,7 @@ static int ext4_mark_dquot_dirty(struct dquot *dquot); > static int ext4_write_info(struct super_block *sb, int type); > static int ext4_quota_on(struct super_block *sb, int type, int format_id, > struct path *path); > +static int ext4_quota_on_meta(struct super_block *sb, int type, int format_id); > static int ext4_quota_off(struct super_block *sb, int type); > static int ext4_quota_on_mount(struct super_block *sb, int type); > static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, > @@ -1252,6 +1253,16 @@ static const struct quotactl_ops ext4_qctl_operations = { > .get_dqblk = dquot_get_dqblk, > .set_dqblk = dquot_set_dqblk > }; > + > +static const struct quotactl_ops ext4_new_qctl_operations = { > + .quota_on_meta = ext4_quota_on_meta, > + .quota_off = ext4_quota_off, > + .quota_sync = dquot_quota_sync, > + .get_info = dquot_get_dqinfo, > + .set_info = dquot_set_dqinfo, > + .get_dqblk = dquot_get_dqblk, > + .set_dqblk = dquot_set_dqblk > +}; > #endif > > static const struct super_operations ext4_sops = { > @@ -3518,6 +3529,15 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent) > #ifdef CONFIG_QUOTA > sb->s_qcop = &ext4_qctl_operations; > sb->dq_op = &ext4_quota_operations; > + > + if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) { > + /* Use new qctl operations with quota on function that does not > + * require user specified quota file path. */ > + sb->s_qcop = &ext4_new_qctl_operations; > + > + sbi->s_qf_inums[USRQUOTA] = es->s_usr_quota_inum; > + sbi->s_qf_inums[GRPQUOTA] = es->s_grp_quota_inum; > + } > #endif > memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid)); > > @@ -3745,8 +3765,19 @@ no_journal: > } else > descr = "out journal"; > > - ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. " > - "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts, > +#ifdef CONFIG_QUOTA > + /* Enable quotas during mount. */ > + if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) && > + !(sb->s_flags & MS_RDONLY)) { > + ext4_quota_on_meta(sb, USRQUOTA, QFMT_VFS_V1); > + ext4_quota_on_meta(sb, GRPQUOTA, QFMT_VFS_V1); > + } > +#endif /* CONFIG_QUOTA */ > + > + ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. quota=%s. " > + "Opts: %s%s%s", descr, > + sb_any_quota_loaded(sb) ? "on" : "off", > + sbi->s_es->s_mount_opts, > *sbi->s_es->s_mount_opts ? "; " : "", orig_data); > > if (es->s_error_count) > @@ -4106,6 +4137,12 @@ static int ext4_commit_super(struct super_block *sb, int sync) > es->s_free_inodes_count = > cpu_to_le32(percpu_counter_sum_positive( > &EXT4_SB(sb)->s_freeinodes_counter)); > +#ifdef CONFIG_QUOTA > + if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) { > + es->s_usr_quota_inum = EXT4_SB(sb)->s_qf_inums[USRQUOTA]; > + es->s_grp_quota_inum = EXT4_SB(sb)->s_qf_inums[GRPQUOTA]; > + } > +#endif > sb->s_dirt = 0; > BUFFER_TRACE(sbh, "marking dirty"); > mark_buffer_dirty(sbh); > @@ -4687,7 +4724,7 @@ static int ext4_quota_on_mount(struct super_block *sb, int type) > * Standard function to be called on quota_on > */ > static int ext4_quota_on(struct super_block *sb, int type, int format_id, > - struct path *path) > + struct path *path) > { > int err; > > @@ -4726,11 +4763,57 @@ static int ext4_quota_on(struct super_block *sb, int type, int format_id, > return dquot_quota_on(sb, type, format_id, path); > } > > +/* > + * New quota_on function that is used when QUOTA feature is set. > + */ > +static int ext4_quota_on_meta(struct super_block *sb, int type, int format_id) > +{ > + int err; > + struct inode *qf_inode; > + > + if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) || > + !EXT4_SB(sb)->s_qf_inums[type]) > + return -EINVAL; > + > + qf_inode = ext4_iget(sb, EXT4_SB(sb)->s_qf_inums[type]); > + if (IS_ERR(qf_inode)) { > + EXT4_SB(sb)->s_qf_inums[type] = 0; > + return PTR_ERR(qf_inode); > + } > + > + /* > + * When we journal data on quota file, we have to flush journal to see > + * all updates to the file when we bypass pagecache... > + */ > + if (EXT4_SB(sb)->s_journal && > + ext4_should_journal_data(qf_inode)) { > + /* > + * We don't need to lock updates but journal_flush() could > + * otherwise be livelocked... > + */ > + jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal); > + err = jbd2_journal_flush(EXT4_SB(sb)->s_journal); > + jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal); > + if (err) > + return err; > + } > + > + err = dquot_enable(qf_inode, type, format_id, > + DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED); > + iput(qf_inode); > + > + return err; > +} > + > static int ext4_quota_off(struct super_block *sb, int type) > { > struct inode *inode = sb_dqopt(sb)->files[type]; > handle_t *handle; > > + /* Do not allow turning quotas off if QUOTA feature is set. */ > + if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) > + return -EINVAL; > + > /* Force all delayed allocation blocks to be allocated. > * Caller already holds s_umount sem */ > if (test_opt(sb, DELALLOC)) > -- > 1.7.3.1 >
Thanks for your feedback. On Mon, Aug 29, 2011 at 2:57 PM, Jan Kara <jack@suse.cz> wrote: > On Fri 19-08-11 12:06:09, Aditya Kali wrote: >> This patch is an attempt towards supporting quotas as first class >> feature in ext4. It is based on the proposal at: >> https://ext4.wiki.kernel.org/index.php/Design_For_1st_Class_Quota_in_Ext4 >> This patch introduces a new feature - EXT4_FEATURE_RO_COMPAT_QUOTA which, when >> turned on, enables quota accounting at mount time iteself. Also, the >> quota inodes are stored in two additional superblock fields. >> Some changes introduced by this patch that should be pointed out are: >> 1) Two new ext4-superblock fields - s_usr_quota_inum and s_grp_quota_inum >> for storing the quota inodes in use. >> 2) If the QUOTA feature and corresponding quota inodes are set in superblock, >> Quotas are turned on at mount time irrespective of the quota mount options. >> Thus the mount options 'quota', 'usrquota' and 'grpquota' are completely >> ignored with the new QUOTA feature flag. > Hmm, cannot we automatically enable 'usrquota' and 'grpquota' options > when we see QUOTA feature enabled? Quota tools still need these options to > identify a filesystem they can work with (although another option would be > to change quota tools to use GETFMT with ext4 filesystem to identify > whether quotas are enabled or not). > In my current change, these options are just ignored, and the user can still set them if any user tools depend on them while mounting/remounting the filesystem. The design wiki (https://ext4.wiki.kernel.org/index.php/Design_For_1st_Class_Quota_in_Ext4#How_can_userspace_tell_if_quota_is_enabled.3F) mentions that 'usrquota' & 'grpquota' options won't be used (so I was under impression that they are being deprecated). Also, if these are set automatically on mount, it would mean that the user wont be able to set these options anymore. >> 3) Default quota inodes are: inode#3 for tracking userquota and inode#4 for >> tracking group quota. The superblock fields can be set to use other inodes >> as well. >> 4) mke2fs or tune2fs will initialize these inodes when quota feature is >> being set. The default reserved inodes will not be visible to user as >> regular files. >> 5) Once quotas are turned on, they cannot be turned off while the FS is >> mounted. This is because we do not want to let the quota get inconsistent. > Both xfs and ocfs2 allow for mode in which usage is tracked (i.e. quota > information is kept uptodate) but limits are not enforced. I imagine this > might be useful in some cases and also from consistency point of view it > would be nice. > > In ocfs2 I implemented it so that on mount, updating of quota information > is enabled (when appropriate fs feature is enabled) but limits are not > enforced. Quotaon/quotaoff commands then toggle enforcement of quota > limits. XFS has mount options which decide whether quota information should > be kept uptodate and whether it should be enforced or not. Enabling only usage on mount makes sense. The limits can then be turned on using quotaon. I assume then we will also need to allow turning the limits off by user (instead of completely denying quotaoff)? > >> 6) With the QUOTA feature set, since the quota inodes are hidden, some of the >> utilities from quota-tools will no longer work correctly. Instead, e2fsprogs >> will include support for fixing the quota files. > I think it might be useful in some cases to enable only user quotas but > leave group quotas disabled (it slightly reduces quota overhead - mainly > because we have less quota structures to update on disk). So it would be > useful to have USRQUOTA and GRPQUOTA features separate... > It is possible to enable only one type of quotas. Instead of having two filesystem features, there are already two superblock fields s_usr_quota_inum and s_grp_quota_inum. The enabling can be disabled by setting the corresponding field to '0'. This is also supported via tune2fs: # to enable only user quotas and not group quotas: $ tune2fs -Q usrquota /dev/ram1 # this will set the 'quota' feature in superblock and initialize only usrquota inode. # to enable both user and group quotas: $ tune2fs -Q usrquota,grpquota /dev/ram1 > Honza -- Aditya -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Mon 29-08-11 15:42:36, Aditya Kali wrote: > On Mon, Aug 29, 2011 at 2:57 PM, Jan Kara <jack@suse.cz> wrote: > > On Fri 19-08-11 12:06:09, Aditya Kali wrote: > >> This patch is an attempt towards supporting quotas as first class > >> feature in ext4. It is based on the proposal at: > >> https://ext4.wiki.kernel.org/index.php/Design_For_1st_Class_Quota_in_Ext4 > >> This patch introduces a new feature - EXT4_FEATURE_RO_COMPAT_QUOTA which, when > >> turned on, enables quota accounting at mount time iteself. Also, the > >> quota inodes are stored in two additional superblock fields. > >> Some changes introduced by this patch that should be pointed out are: > >> 1) Two new ext4-superblock fields - s_usr_quota_inum and s_grp_quota_inum > >> for storing the quota inodes in use. > >> 2) If the QUOTA feature and corresponding quota inodes are set in superblock, > >> Quotas are turned on at mount time irrespective of the quota mount options. > >> Thus the mount options 'quota', 'usrquota' and 'grpquota' are completely > >> ignored with the new QUOTA feature flag. > > Hmm, cannot we automatically enable 'usrquota' and 'grpquota' options > > when we see QUOTA feature enabled? Quota tools still need these options to > > identify a filesystem they can work with (although another option would be > > to change quota tools to use GETFMT with ext4 filesystem to identify > > whether quotas are enabled or not). > > > In my current change, these options are just ignored, and the user can > still set them if any user tools depend on them while > mounting/remounting the filesystem. > The design wiki > (https://ext4.wiki.kernel.org/index.php/Design_For_1st_Class_Quota_in_Ext4#How_can_userspace_tell_if_quota_is_enabled.3F) > mentions that 'usrquota' & 'grpquota' options won't be used (so I was > under impression that they are being deprecated). Also, if these are > set automatically on mount, it would mean that the user wont be able > to set these options anymore. OK, probably makes sense in the long run. I'll update quota tools to use GETFMT quotactl then... > >> 3) Default quota inodes are: inode#3 for tracking userquota and inode#4 for > >> tracking group quota. The superblock fields can be set to use other inodes > >> as well. > >> 4) mke2fs or tune2fs will initialize these inodes when quota feature is > >> being set. The default reserved inodes will not be visible to user as > >> regular files. > >> 5) Once quotas are turned on, they cannot be turned off while the FS is > >> mounted. This is because we do not want to let the quota get inconsistent. > > Both xfs and ocfs2 allow for mode in which usage is tracked (i.e. quota > > information is kept uptodate) but limits are not enforced. I imagine this > > might be useful in some cases and also from consistency point of view it > > would be nice. > > > > In ocfs2 I implemented it so that on mount, updating of quota information > > is enabled (when appropriate fs feature is enabled) but limits are not > > enforced. Quotaon/quotaoff commands then toggle enforcement of quota > > limits. XFS has mount options which decide whether quota information should > > be kept uptodate and whether it should be enforced or not. > > Enabling only usage on mount makes sense. The limits can then be > turned on using quotaon. I assume then we will also need to allow > turning the limits off by user (instead of completely denying > quotaoff)? Yes. > >> 6) With the QUOTA feature set, since the quota inodes are hidden, some of the > >> utilities from quota-tools will no longer work correctly. Instead, e2fsprogs > >> will include support for fixing the quota files. > > I think it might be useful in some cases to enable only user quotas but > > leave group quotas disabled (it slightly reduces quota overhead - mainly > > because we have less quota structures to update on disk). So it would be > > useful to have USRQUOTA and GRPQUOTA features separate... > > > It is possible to enable only one type of quotas. Instead of having > two filesystem features, there are already two superblock fields > s_usr_quota_inum and s_grp_quota_inum. The enabling can be disabled by > setting the corresponding field to '0'. This is also supported via > tune2fs: > > # to enable only user quotas and not group quotas: > $ tune2fs -Q usrquota /dev/ram1 # this will set the 'quota' feature in > superblock and initialize only usrquota inode. > > # to enable both user and group quotas: > $ tune2fs -Q usrquota,grpquota /dev/ram1 Ah, I see. OK. Honza
On Mon, Aug 29, 2011 at 10:46:02PM +0200, Jan Kara wrote: > The only pain point is repquota(8) - that needs a new kernel interface to > work reliably (for ocfs2 and xfs we workaround the inability to iterate > over all quota structures from userspace by using getpwent but that's just > ugly). Feel free to a new quotactl case to support it, but I don't reall see the problem of the getpwent loop. -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Fri 09-09-11 11:40:57, Christoph Hellwig wrote: > On Mon, Aug 29, 2011 at 10:46:02PM +0200, Jan Kara wrote: > > The only pain point is repquota(8) - that needs a new kernel interface to > > work reliably (for ocfs2 and xfs we workaround the inability to iterate > > over all quota structures from userspace by using getpwent but that's just > > ugly). > > Feel free to a new quotactl case to support it, but I don't reall see > the problem of the getpwent loop. Well, the trouble is with systems which have accounts stored e.g. in LDAP with thousands / tens thousands of users (although each machine is used only by a small subset of users). Then iterating with getpwent gets really inefficient... Honza
On Sat, Sep 10, 2011 at 10:24:09PM +0200, Jan Kara wrote: > Well, the trouble is with systems which have accounts stored e.g. in > LDAP with thousands / tens thousands of users (although each machine is > used only by a small subset of users). Then iterating with getpwent gets > really inefficient... Please make sure to Cc the xfs list when adding an interface like this, as it's trivial to implement for us. -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h index e717dfd..5b57bc2 100644 --- a/fs/ext4/ext4.h +++ b/fs/ext4/ext4.h @@ -1065,7 +1065,9 @@ struct ext4_super_block { __u8 s_last_error_func[32]; /* function where the error happened */ #define EXT4_S_ERR_END offsetof(struct ext4_super_block, s_mount_opts) __u8 s_mount_opts[64]; - __le32 s_reserved[112]; /* Padding to the end of the block */ + __le32 s_usr_quota_inum; /* inode for tracking user quota */ + __le32 s_grp_quota_inum; /* inode for tracking group quota */ + __le32 s_reserved[110]; /* Padding to the end of the block */ }; #define EXT4_S_ERR_LEN (EXT4_S_ERR_END - EXT4_S_ERR_START) @@ -1142,6 +1144,7 @@ struct ext4_sb_info { #ifdef CONFIG_QUOTA char *s_qf_names[MAXQUOTAS]; /* Names of quota files with journalled quota */ int s_jquota_fmt; /* Format of quota to use */ + unsigned long s_qf_inums[MAXQUOTAS]; /* Quota file inodes */ #endif unsigned int s_want_extra_isize; /* New inodes should reserve # bytes */ struct rb_root system_blks; @@ -1241,6 +1244,8 @@ static inline struct timespec ext4_current_time(struct inode *inode) static inline int ext4_valid_inum(struct super_block *sb, unsigned long ino) { return ino == EXT4_ROOT_INO || + ino == EXT4_USR_QUOTA_INO || + ino == EXT4_GRP_QUOTA_INO || ino == EXT4_JOURNAL_INO || ino == EXT4_RESIZE_INO || (ino >= EXT4_FIRST_INO(sb) && @@ -1401,7 +1406,8 @@ static inline void ext4_clear_state_flags(struct ext4_inode_info *ei) EXT4_FEATURE_RO_COMPAT_DIR_NLINK | \ EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE | \ EXT4_FEATURE_RO_COMPAT_BTREE_DIR |\ - EXT4_FEATURE_RO_COMPAT_HUGE_FILE) + EXT4_FEATURE_RO_COMPAT_HUGE_FILE| \ + EXT4_FEATURE_RO_COMPAT_QUOTA) /* * Default values for user and/or group using reserved blocks diff --git a/fs/ext4/ext4_jbd2.h b/fs/ext4/ext4_jbd2.h index bb85757..0f097ed 100644 --- a/fs/ext4/ext4_jbd2.h +++ b/fs/ext4/ext4_jbd2.h @@ -87,14 +87,20 @@ #ifdef CONFIG_QUOTA /* Amount of blocks needed for quota update - we know that the structure was * allocated so we need to update only data block */ -#define EXT4_QUOTA_TRANS_BLOCKS(sb) (test_opt(sb, QUOTA) ? 1 : 0) +#define EXT4_QUOTA_TRANS_BLOCKS(sb) ((test_opt(sb, QUOTA) ||\ + EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) ?\ + 1 : 0) /* Amount of blocks needed for quota insert/delete - we do some block writes * but inode, sb and group updates are done only once */ -#define EXT4_QUOTA_INIT_BLOCKS(sb) (test_opt(sb, QUOTA) ? (DQUOT_INIT_ALLOC*\ - (EXT4_SINGLEDATA_TRANS_BLOCKS(sb)-3)+3+DQUOT_INIT_REWRITE) : 0) - -#define EXT4_QUOTA_DEL_BLOCKS(sb) (test_opt(sb, QUOTA) ? (DQUOT_DEL_ALLOC*\ - (EXT4_SINGLEDATA_TRANS_BLOCKS(sb)-3)+3+DQUOT_DEL_REWRITE) : 0) +#define EXT4_QUOTA_INIT_BLOCKS(sb) ((test_opt(sb, QUOTA) ||\ + EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) ?\ + (DQUOT_INIT_ALLOC*(EXT4_SINGLEDATA_TRANS_BLOCKS(sb)-3)\ + +3+DQUOT_INIT_REWRITE) : 0) + +#define EXT4_QUOTA_DEL_BLOCKS(sb) ((test_opt(sb, QUOTA) ||\ + EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) ?\ + (DQUOT_DEL_ALLOC*(EXT4_SINGLEDATA_TRANS_BLOCKS(sb)-3)\ + +3+DQUOT_DEL_REWRITE) : 0) #else #define EXT4_QUOTA_TRANS_BLOCKS(sb) 0 #define EXT4_QUOTA_INIT_BLOCKS(sb) 0 diff --git a/fs/ext4/super.c b/fs/ext4/super.c index 4687fea..3f3d87e 100644 --- a/fs/ext4/super.c +++ b/fs/ext4/super.c @@ -1225,6 +1225,7 @@ static int ext4_mark_dquot_dirty(struct dquot *dquot); static int ext4_write_info(struct super_block *sb, int type); static int ext4_quota_on(struct super_block *sb, int type, int format_id, struct path *path); +static int ext4_quota_on_meta(struct super_block *sb, int type, int format_id); static int ext4_quota_off(struct super_block *sb, int type); static int ext4_quota_on_mount(struct super_block *sb, int type); static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, @@ -1252,6 +1253,16 @@ static const struct quotactl_ops ext4_qctl_operations = { .get_dqblk = dquot_get_dqblk, .set_dqblk = dquot_set_dqblk }; + +static const struct quotactl_ops ext4_new_qctl_operations = { + .quota_on_meta = ext4_quota_on_meta, + .quota_off = ext4_quota_off, + .quota_sync = dquot_quota_sync, + .get_info = dquot_get_dqinfo, + .set_info = dquot_set_dqinfo, + .get_dqblk = dquot_get_dqblk, + .set_dqblk = dquot_set_dqblk +}; #endif static const struct super_operations ext4_sops = { @@ -3518,6 +3529,15 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent) #ifdef CONFIG_QUOTA sb->s_qcop = &ext4_qctl_operations; sb->dq_op = &ext4_quota_operations; + + if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) { + /* Use new qctl operations with quota on function that does not + * require user specified quota file path. */ + sb->s_qcop = &ext4_new_qctl_operations; + + sbi->s_qf_inums[USRQUOTA] = es->s_usr_quota_inum; + sbi->s_qf_inums[GRPQUOTA] = es->s_grp_quota_inum; + } #endif memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid)); @@ -3745,8 +3765,19 @@ no_journal: } else descr = "out journal"; - ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. " - "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts, +#ifdef CONFIG_QUOTA + /* Enable quotas during mount. */ + if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) && + !(sb->s_flags & MS_RDONLY)) { + ext4_quota_on_meta(sb, USRQUOTA, QFMT_VFS_V1); + ext4_quota_on_meta(sb, GRPQUOTA, QFMT_VFS_V1); + } +#endif /* CONFIG_QUOTA */ + + ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. quota=%s. " + "Opts: %s%s%s", descr, + sb_any_quota_loaded(sb) ? "on" : "off", + sbi->s_es->s_mount_opts, *sbi->s_es->s_mount_opts ? "; " : "", orig_data); if (es->s_error_count) @@ -4106,6 +4137,12 @@ static int ext4_commit_super(struct super_block *sb, int sync) es->s_free_inodes_count = cpu_to_le32(percpu_counter_sum_positive( &EXT4_SB(sb)->s_freeinodes_counter)); +#ifdef CONFIG_QUOTA + if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) { + es->s_usr_quota_inum = EXT4_SB(sb)->s_qf_inums[USRQUOTA]; + es->s_grp_quota_inum = EXT4_SB(sb)->s_qf_inums[GRPQUOTA]; + } +#endif sb->s_dirt = 0; BUFFER_TRACE(sbh, "marking dirty"); mark_buffer_dirty(sbh); @@ -4687,7 +4724,7 @@ static int ext4_quota_on_mount(struct super_block *sb, int type) * Standard function to be called on quota_on */ static int ext4_quota_on(struct super_block *sb, int type, int format_id, - struct path *path) + struct path *path) { int err; @@ -4726,11 +4763,57 @@ static int ext4_quota_on(struct super_block *sb, int type, int format_id, return dquot_quota_on(sb, type, format_id, path); } +/* + * New quota_on function that is used when QUOTA feature is set. + */ +static int ext4_quota_on_meta(struct super_block *sb, int type, int format_id) +{ + int err; + struct inode *qf_inode; + + if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) || + !EXT4_SB(sb)->s_qf_inums[type]) + return -EINVAL; + + qf_inode = ext4_iget(sb, EXT4_SB(sb)->s_qf_inums[type]); + if (IS_ERR(qf_inode)) { + EXT4_SB(sb)->s_qf_inums[type] = 0; + return PTR_ERR(qf_inode); + } + + /* + * When we journal data on quota file, we have to flush journal to see + * all updates to the file when we bypass pagecache... + */ + if (EXT4_SB(sb)->s_journal && + ext4_should_journal_data(qf_inode)) { + /* + * We don't need to lock updates but journal_flush() could + * otherwise be livelocked... + */ + jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal); + err = jbd2_journal_flush(EXT4_SB(sb)->s_journal); + jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal); + if (err) + return err; + } + + err = dquot_enable(qf_inode, type, format_id, + DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED); + iput(qf_inode); + + return err; +} + static int ext4_quota_off(struct super_block *sb, int type) { struct inode *inode = sb_dqopt(sb)->files[type]; handle_t *handle; + /* Do not allow turning quotas off if QUOTA feature is set. */ + if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) + return -EINVAL; + /* Force all delayed allocation blocks to be allocated. * Caller already holds s_umount sem */ if (test_opt(sb, DELALLOC))
This patch is an attempt towards supporting quotas as first class feature in ext4. It is based on the proposal at: https://ext4.wiki.kernel.org/index.php/Design_For_1st_Class_Quota_in_Ext4 This patch introduces a new feature - EXT4_FEATURE_RO_COMPAT_QUOTA which, when turned on, enables quota accounting at mount time iteself. Also, the quota inodes are stored in two additional superblock fields. Some changes introduced by this patch that should be pointed out are: 1) Two new ext4-superblock fields - s_usr_quota_inum and s_grp_quota_inum for storing the quota inodes in use. 2) If the QUOTA feature and corresponding quota inodes are set in superblock, Quotas are turned on at mount time irrespective of the quota mount options. Thus the mount options 'quota', 'usrquota' and 'grpquota' are completely ignored with the new QUOTA feature flag. 3) Default quota inodes are: inode#3 for tracking userquota and inode#4 for tracking group quota. The superblock fields can be set to use other inodes as well. 4) mke2fs or tune2fs will initialize these inodes when quota feature is being set. The default reserved inodes will not be visible to user as regular files. 5) Once quotas are turned on, they cannot be turned off while the FS is mounted. This is because we do not want to let the quota get inconsistent. 6) With the QUOTA feature set, since the quota inodes are hidden, some of the utilities from quota-tools will no longer work correctly. Instead, e2fsprogs will include support for fixing the quota files. 7) Support is only for the new V2 quota file format. Signed-off-by: Aditya Kali <adityakali@google.com> --- fs/ext4/ext4.h | 10 ++++- fs/ext4/ext4_jbd2.h | 18 +++++++--- fs/ext4/super.c | 89 +++++++++++++++++++++++++++++++++++++++++++++++++-- 3 files changed, 106 insertions(+), 11 deletions(-)