On Mon 05-01-09 20:40:08, Mingming Cao wrote:
> 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 <cmm@xxxxxxxxxx>
Just a few comments below:
> ---
> fs/dquot.c | 111 +++++++++++++++++++++++++++++++++--------------
> include/linux/quota.h | 3 +
> include/linux/quotaops.h | 22 +++++++++
> 3 files changed, 105 insertions(+), 31 deletions(-)
>
> Index: linux-2.6.28-git7/fs/dquot.c
> ===================================================================
> --- linux-2.6.28-git7.orig/fs/dquot.c 2009-01-05 17:09:50.000000000 -0800
> +++ linux-2.6.28-git7/fs/dquot.c 2009-01-05 20:07:08.000000000 -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 +1072,9 @@ err_out:
> kfree_skb(skb);
> }
> #endif
> -
> +/*
> + * Should called with dq_data_lock dropped, this function could sleep.
> + */
If you want to add a comment here I'd add something like:
/*
* Write warnings to the console and send warning messages over netlink.
*
* Note that this function can sleep.
*/
Because you cannot hold *any* spinlock when calling a function that can
sleep...
> static inline void flush_warnings(struct dquot * const *dquots, char *warntype)
> {
> int i;
> @@ -1128,13 +1135,18 @@ 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 +1154,7 @@ 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 +1163,7 @@ 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 +1304,88 @@ 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);
> + }
> }
> - inode_add_bytes(inode, number);
> - ret = QUOTA_OK;
> -warn_put_all:
> +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))
> + goto out;
> +
> + down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
> + if (IS_NOQUOTA(inode))
> + goto out_unlock;
We need to call inode_add_bytes(inode, number) even for IS_NOQUOTA()
inodes (but we don't have to hold dq_data_lock for such inodes). Your
rewrite has removed this call...
> +
> + 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 +2074,7 @@ 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 +2116,7 @@ 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 2009-01-05 17:09:37.000000000 -0800
> +++ linux-2.6.28-git7/include/linux/quota.h 2009-01-05 20:07:08.000000000 -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 2009-01-05 17:09:37.000000000 -0800
> +++ linux-2.6.28-git7/include/linux/quotaops.h 2009-01-05 20:07:08.000000000 -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);
> @@ -372,12 +387,19 @@ static inline int vfs_dq_alloc_block_nod
> nr << inode->i_sb->s_blocksize_bits);
> }
>
> +
^^ This empty line was added by accident I guess...
> static inline int vfs_dq_alloc_block(struct inode *inode, qsize_t nr)
> {
> return vfs_dq_alloc_space(inode,
> 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 <jack@xxxxxxx>
SUSE Labs, CR
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
