From: Jan Kara Subject: Re: [PATCH V5 1/5] quota: Add reservation support for delayed block allocation Date: Tue, 13 Jan 2009 16:37:49 +0100 Message-ID: <20090113153748.GE10064@duck.suse.cz> References: <1231216808.9267.22.camel@mingming-laptop> <20090106100645.GH10705@duck.suse.cz> <1231805946.6752.17.camel@mingming-laptop> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Andrew Morton , tytso , linux-ext4 , linux-fsdevel To: Mingming Cao Return-path: Content-Disposition: inline In-Reply-To: <1231805946.6752.17.camel@mingming-laptop> Sender: linux-fsdevel-owner@vger.kernel.org List-Id: linux-ext4.vger.kernel.org On Mon 12-01-09 16:19:06, Mingming Cao wrote: > Thanks for your review and suggestions. All points are taken. I have > updated the quota patches.I am attaching the updated patch here just for > your review. > > I am waiting for the ext4 tree to updated to rebase the whole series > against 2.6.29-rc1 plus ext4 patch queue. > > Quota: Add quota reservation support > > Delayed allocation defers the block allocation at the dirty pages > flush-out time, doing quota charge/check at that time is too late. > But we can't charge the quota blocks until blocks are really allocated, > otherwise users could get overcharged after reboot from system crash. > > This patch adds quota reservation for delayed llocation. Quota blocks > are reserved in memory, inode and quota won't gets dirtied until later > block allocation time. > > Signed-off-by: Mingming Cao The patch is fine. You can add Acked-by: Jan Kara How do you want to merge the patches? Via ext4 patch queue? There's one generic quota patch that I also need to push to fix some OCFS2 issue and it collides with your patchset. And also there're further cleanups in quota code which are long overdue which I want to base on all other patches. So I've decided to setup quota git tree. I'll pull in your two VFS quota patches. Will that work for you? The tree should be soon at: git.kernel.org/pub/scm/linux/kernel/git/jack/linux-quota-2.6.git The branch you can pull from is for_next (or for_mm if there'll be some more long term experimental stuff but I'm not aware of anything like that now). Honza > --- > fs/dquot.c | 117 ++++++++++++++++++++++++++++++++++------------- > include/linux/quota.h | 3 + > include/linux/quotaops.h | 21 ++++++++ > 3 files changed, 110 insertions(+), 31 deletions(-) > > Index: linux-2.6.28-git7/fs/dquot.c > =================================================================== > --- linux-2.6.28-git7.orig/fs/dquot.c 2009-01-06 10:55:08.000000000 -0800 > +++ linux-2.6.28-git7/fs/dquot.c 2009-01-06 15:42:31.000000000 -0800 > @@ -898,6 +898,11 @@ static inline void dquot_incr_space(stru > dquot->dq_dqb.dqb_curspace += number; > } > > +static inline void dquot_resv_space(struct dquot *dquot, qsize_t number) > +{ > + dquot->dq_dqb.dqb_rsvspace += number; > +} > + > static inline void dquot_decr_inodes(struct dquot *dquot, qsize_t number) > { > if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE || > @@ -1067,7 +1072,11 @@ err_out: > kfree_skb(skb); > } > #endif > - > +/* > + * Write warnings to the console and send warning messages over netlink. > + * > + * Note that this function can sleep. > + */ > static inline void flush_warnings(struct dquot * const *dquots, char *warntype) > { > int i; > @@ -1128,13 +1137,18 @@ static int check_idq(struct dquot *dquot > /* needs dq_data_lock */ > static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc, char *warntype) > { > + qsize_t tspace; > + > *warntype = QUOTA_NL_NOWARN; > if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_type) || > test_bit(DQ_FAKE_B, &dquot->dq_flags)) > return QUOTA_OK; > > + tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace > + + space; > + > if (dquot->dq_dqb.dqb_bhardlimit && > - dquot->dq_dqb.dqb_curspace + space > dquot->dq_dqb.dqb_bhardlimit && > + tspace > dquot->dq_dqb.dqb_bhardlimit && > !ignore_hardlimit(dquot)) { > if (!prealloc) > *warntype = QUOTA_NL_BHARDWARN; > @@ -1142,7 +1156,7 @@ static int check_bdq(struct dquot *dquot > } > > if (dquot->dq_dqb.dqb_bsoftlimit && > - dquot->dq_dqb.dqb_curspace + space > dquot->dq_dqb.dqb_bsoftlimit && > + tspace > dquot->dq_dqb.dqb_bsoftlimit && > dquot->dq_dqb.dqb_btime && get_seconds() >= dquot->dq_dqb.dqb_btime && > !ignore_hardlimit(dquot)) { > if (!prealloc) > @@ -1151,7 +1165,7 @@ static int check_bdq(struct dquot *dquot > } > > if (dquot->dq_dqb.dqb_bsoftlimit && > - dquot->dq_dqb.dqb_curspace + space > dquot->dq_dqb.dqb_bsoftlimit && > + tspace > dquot->dq_dqb.dqb_bsoftlimit && > dquot->dq_dqb.dqb_btime == 0) { > if (!prealloc) { > *warntype = QUOTA_NL_BSOFTWARN; > @@ -1292,51 +1306,92 @@ void vfs_dq_drop(struct inode *inode) > /* > * This operation can block, but only after everything is updated > */ > -int dquot_alloc_space(struct inode *inode, qsize_t number, int warn) > +int __dquot_alloc_space(struct inode *inode, qsize_t number, > + int warn, int reserve) > { > - int cnt, ret = NO_QUOTA; > + int cnt, ret = QUOTA_OK; > char warntype[MAXQUOTAS]; > > - /* First test before acquiring mutex - solves deadlocks when we > - * re-enter the quota code and are already holding the mutex */ > - if (IS_NOQUOTA(inode)) { > -out_add: > - inode_add_bytes(inode, number); > - return QUOTA_OK; > - } > for (cnt = 0; cnt < MAXQUOTAS; cnt++) > warntype[cnt] = QUOTA_NL_NOWARN; > > - down_read(&sb_dqopt(inode->i_sb)->dqptr_sem); > - if (IS_NOQUOTA(inode)) { /* Now we can do reliable test... */ > - up_read(&sb_dqopt(inode->i_sb)->dqptr_sem); > - goto out_add; > - } > spin_lock(&dq_data_lock); > for (cnt = 0; cnt < MAXQUOTAS; cnt++) { > if (inode->i_dquot[cnt] == NODQUOT) > continue; > - if (check_bdq(inode->i_dquot[cnt], number, warn, warntype+cnt) == NO_QUOTA) > - goto warn_put_all; > + if (check_bdq(inode->i_dquot[cnt], number, warn, warntype+cnt) > + == NO_QUOTA) { > + ret = NO_QUOTA; > + goto out_unlock; > + } > } > for (cnt = 0; cnt < MAXQUOTAS; cnt++) { > if (inode->i_dquot[cnt] == NODQUOT) > continue; > - dquot_incr_space(inode->i_dquot[cnt], number); > + if (reserve) > + dquot_resv_space(inode->i_dquot[cnt], number); > + else > + dquot_incr_space(inode->i_dquot[cnt], number); > } > - inode_add_bytes(inode, number); > - ret = QUOTA_OK; > -warn_put_all: > + if (!reserve) > + inode_add_bytes(inode, number); > +out_unlock: > spin_unlock(&dq_data_lock); > - if (ret == QUOTA_OK) > - /* Dirtify all the dquots - this can block when journalling */ > - for (cnt = 0; cnt < MAXQUOTAS; cnt++) > - if (inode->i_dquot[cnt]) > - mark_dquot_dirty(inode->i_dquot[cnt]); > flush_warnings(inode->i_dquot, warntype); > + return ret; > +} > + > +int dquot_alloc_space(struct inode *inode, qsize_t number, int warn) > +{ > + int cnt, ret = QUOTA_OK; > + > + /* > + * First test before acquiring mutex - solves deadlocks when we > + * re-enter the quota code and are already holding the mutex > + */ > + if (IS_NOQUOTA(inode)) { > + inode_add_bytes(inode, number); > + goto out; > + } > + > + down_read(&sb_dqopt(inode->i_sb)->dqptr_sem); > + if (IS_NOQUOTA(inode)) { > + inode_add_bytes(inode, number); > + goto out_unlock; > + } > + > + ret = __dquot_alloc_space(inode, number, warn, 0); > + if (ret == NO_QUOTA) > + goto out_unlock; > + > + /* Dirtify all the dquots - this can block when journalling */ > + for (cnt = 0; cnt < MAXQUOTAS; cnt++) > + if (inode->i_dquot[cnt]) > + mark_dquot_dirty(inode->i_dquot[cnt]); > +out_unlock: > up_read(&sb_dqopt(inode->i_sb)->dqptr_sem); > +out: > + return ret; > +} > + > +int dquot_reserve_space(struct inode *inode, qsize_t number, int warn) > +{ > + int ret = QUOTA_OK; > + > + if (IS_NOQUOTA(inode)) > + goto out; > + > + down_read(&sb_dqopt(inode->i_sb)->dqptr_sem); > + if (IS_NOQUOTA(inode)) > + goto out_unlock; > + > + ret = __dquot_alloc_space(inode, number, warn, 1); > +out_unlock: > + up_read(&sb_dqopt(inode->i_sb)->dqptr_sem); > +out: > return ret; > } > +EXPORT_SYMBOL(dquot_reserve_space); > > /* > * This operation can block, but only after everything is updated > @@ -2025,7 +2080,7 @@ static void do_get_dqblk(struct dquot *d > spin_lock(&dq_data_lock); > di->dqb_bhardlimit = stoqb(dm->dqb_bhardlimit); > di->dqb_bsoftlimit = stoqb(dm->dqb_bsoftlimit); > - di->dqb_curspace = dm->dqb_curspace; > + di->dqb_curspace = dm->dqb_curspace + dm->dqb_rsvspace; > di->dqb_ihardlimit = dm->dqb_ihardlimit; > di->dqb_isoftlimit = dm->dqb_isoftlimit; > di->dqb_curinodes = dm->dqb_curinodes; > @@ -2067,7 +2122,7 @@ static int do_set_dqblk(struct dquot *dq > > spin_lock(&dq_data_lock); > if (di->dqb_valid & QIF_SPACE) { > - dm->dqb_curspace = di->dqb_curspace; > + dm->dqb_curspace = di->dqb_curspace - dm->dqb_rsvspace; > check_blim = 1; > __set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags); > } > Index: linux-2.6.28-git7/include/linux/quota.h > =================================================================== > --- linux-2.6.28-git7.orig/include/linux/quota.h 2009-01-06 10:55:08.000000000 -0800 > +++ linux-2.6.28-git7/include/linux/quota.h 2009-01-06 15:40:14.000000000 -0800 > @@ -198,6 +198,7 @@ struct mem_dqblk { > qsize_t dqb_bhardlimit; /* absolute limit on disk blks alloc */ > qsize_t dqb_bsoftlimit; /* preferred limit on disk blks */ > qsize_t dqb_curspace; /* current used space */ > + qsize_t dqb_rsvspace; /* current reserved space for delalloc*/ > qsize_t dqb_ihardlimit; /* absolute limit on allocated inodes */ > qsize_t dqb_isoftlimit; /* preferred inode limit */ > qsize_t dqb_curinodes; /* current # allocated inodes */ > @@ -308,6 +309,8 @@ struct dquot_operations { > int (*release_dquot) (struct dquot *); /* Quota is going to be deleted from disk */ > int (*mark_dirty) (struct dquot *); /* Dquot is marked dirty */ > int (*write_info) (struct super_block *, int); /* Write of quota "superblock" */ > + /* reserve quota for delayed block allocation */ > + int (*reserve_space) (struct inode *, qsize_t, int); > }; > > /* Operations handling requests from userspace */ > Index: linux-2.6.28-git7/include/linux/quotaops.h > =================================================================== > --- linux-2.6.28-git7.orig/include/linux/quotaops.h 2009-01-06 10:55:08.000000000 -0800 > +++ linux-2.6.28-git7/include/linux/quotaops.h 2009-01-06 15:40:14.000000000 -0800 > @@ -185,6 +185,16 @@ static inline int vfs_dq_alloc_space(str > return ret; > } > > +static inline int vfs_dq_reserve_space(struct inode *inode, qsize_t nr) > +{ > + if (sb_any_quota_active(inode->i_sb)) { > + /* Used space is updated in alloc_space() */ > + if (inode->i_sb->dq_op->reserve_space(inode, nr, 0) == NO_QUOTA) > + return 1; > + } > + return 0; > +} > + > static inline int vfs_dq_alloc_inode(struct inode *inode) > { > if (sb_any_quota_active(inode->i_sb)) { > @@ -341,6 +351,11 @@ static inline int vfs_dq_alloc_space(str > return 0; > } > > +static inline int vfs_dq_reserve_space(struct inode *inode, qsize_t nr) > +{ > + return 0; > +} > + > static inline void vfs_dq_free_space_nodirty(struct inode *inode, qsize_t nr) > { > inode_sub_bytes(inode, nr); > @@ -378,6 +393,12 @@ static inline int vfs_dq_alloc_block(str > nr << inode->i_sb->s_blocksize_bits); > } > > +static inline int vfs_dq_reserve_block(struct inode *inode, qsize_t nr) > +{ > + return vfs_dq_reserve_space(inode, > + nr << inode->i_blkbits); > +} > + > static inline void vfs_dq_free_block_nodirty(struct inode *inode, qsize_t nr) > { > vfs_dq_free_space_nodirty(inode, nr << inode->i_sb->s_blocksize_bits); -- Jan Kara SUSE Labs, CR