在 2008-12-03三的 20:06 +0100，Jan Kara写道：
> Hmm, I can't find the email with this patch so I've just copied it from
> some mail archive ;). Sorry if the CC's are wrong.
>
> > quota: Add quota reservation claim and released operations
> >
> > Reserved quota will be claimed at the block allocation time. Over-booked
> > quota could be returned back with the release callback function.
> >
> > Signed-off-by: Mingming Cao <cmm@xxxxxxxxxx>
> > ---
> > fs/dquot.c | 87 +++++++++++++++++++++++++++++++++++++++++++++++
> > include/linux/quota.h | 4 +-
> > include/linux/quotaops.h | 55 +++++++++++++++++++++++++++++
> > 3 files changed, 145 insertions(+), 1 deletion(-)
> >
> > Index: linux-2.6.28-rc2/include/linux/quota.h
> > ===================================================================
> > --- linux-2.6.28-rc2.orig/include/linux/quota.h 2008-11-06 13:36:42.000000000 -0800
> > +++ linux-2.6.28-rc2/include/linux/quota.h 2008-11-06 14:03:52.000000000 -0800
> > @@ -292,7 +292,9 @@ 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" */
> > - int (*reserve_space) (struct inode *, qsize_t, int); /* reserve quota for delayed block allocation */
> > + int (*reserve_space) (struct inode *, qsize_t, int); /* reserve quota for delayed alloc */
> > + int (*claim_space) (struct inode *, qsize_t); /* claim reserved quota for delayed alloc */
> > + void (*release_rsv) (struct inode *, qsize_t); /* release rsved quota for delayed alloc */
> > };
> >
> > /* Operations handling requests from userspace */
> > Index: linux-2.6.28-rc2/include/linux/quotaops.h
> > ===================================================================
> > --- linux-2.6.28-rc2.orig/include/linux/quotaops.h 2008-11-06 13:37:04.000000000 -0800
> > +++ linux-2.6.28-rc2/include/linux/quotaops.h 2008-11-06 14:03:52.000000000 -0800
> > @@ -28,6 +28,11 @@ int dquot_drop(struct inode *inode);
> > int dquot_alloc_space(struct inode *inode, qsize_t number, int prealloc);
> > int dquot_alloc_inode(const struct inode *inode, qsize_t number);
> >
> > +int dquot_reserve_space(struct inode *inode, qsize_t number, int prealloc);
> > +int dquot_claim_space(struct inode *inode, qsize_t number);
> > +void dquot_release_reserved_space(struct inode *inode, qsize_t number);
> > +
> > +
> > int dquot_free_space(struct inode *inode, qsize_t number);
> > int dquot_free_inode(const struct inode *inode, qsize_t number);
> >
> > @@ -196,6 +201,31 @@ static inline int vfs_dq_alloc_inode(str
> > return 0;
> > }
> >
> > +/*
> > + * Convert in-memory reserved quotas to real consumed quotas
> > + */
> > +static inline int vfs_dq_claim_space(struct inode *inode, qsize_t nr)
> > +{
> > + if (sb_any_quota_active(inode->i_sb)) {
> > + if (inode->i_sb->dq_op->claim_space(inode, nr) == NO_QUOTA)
> > + return 1;
> > + } else
> > + inode_add_bytes(inode, nr);
> > +
> > + mark_inode_dirty(inode);
> > + return 0;
> > +}
> > +
> > +/*
> > + * Release reserved (in-memory) quotas
> > + */
> > +static inline
> > +void vfs_dq_release_reservation_space(struct inode *inode, qsize_t nr)
> > +{
> > + if (sb_any_quota_active(inode->i_sb))
> > + inode->i_sb->dq_op->release_rsv(inode, nr);
> > +}
> > +
> > static inline void vfs_dq_free_space_nodirty(struct inode *inode, qsize_t nr)
> > {
> > if (sb_any_quota_active(inode->i_sb))
> > @@ -342,6 +372,17 @@ static inline int vfs_dq_reserve_space(s
> > return 0;
> > }
> >
> > +static inline int vfs_dq_claim_space(struct inode *inode, qsize_t nr)
> > +{
> > + return vfs_dq_alloc_space(inode, nr);
> > +}
> > +
> > +static inline
> > +int vfs_dq_release_reservation_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);
> > @@ -386,6 +427,18 @@ static inline int vfs_dq_reserve_block(s
> > nr << inode->i_blkbits);
> > }
> >
> > +static inline int vfs_dq_claim_block(struct inode *inode, qsize_t nr)
> > +{
> > + return vfs_dq_claim_space(inode,
> > + nr << inode->i_blkbits);
> > +}
> > +
> > +static inline
> > +void vfs_dq_release_reservation(struct inode *inode, qsize_t nr)
> > +{
> > + vfs_dq_release_reservation_space(inode, nr << inode->i_blkbits);
> > +}
> > +
> I guess the function should be called like vfs_dq_release_reservation_block().
> It's ugly long but we should not omit the "block" part. Maybe we could shorten
> reservation everywhere in function names to rsv?
>
> > 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);
> > @@ -415,6 +468,8 @@ static inline void vfs_dq_free_block(str
> > vfs_dq_alloc_block_nodirty(inode, nr)
> > #define DQUOT_ALLOC_BLOCK(inode, nr) vfs_dq_alloc_block(inode, nr)
> > #define DQUOT_RESERVE_BLOCK(inode, nr) vfs_dq_reserve_block(inode, nr)
> > +#define DQUOT_CLAIM_BLOCK(inode, nr) vfs_dq_claim_block(inode, nr)
> > +#define DQUOT_RELEASE_RSV_BLOCK(inode, nr) vfs_dq_release_reservation(inode, nr)
> Please call lowercase variants from ext4 and don't define these functions.
>
> > #define DQUOT_ALLOC_INODE(inode) vfs_dq_alloc_inode(inode)
> > #define DQUOT_FREE_SPACE_NODIRTY(inode, nr) \
> > vfs_dq_free_space_nodirty(inode, nr)
> > Index: linux-2.6.28-rc2/fs/dquot.c
> > ===================================================================
> > --- linux-2.6.28-rc2.orig/fs/dquot.c 2008-11-06 13:37:04.000000000 -0800
> > +++ linux-2.6.28-rc2/fs/dquot.c 2008-11-06 14:03:52.000000000 -0800
> > @@ -846,6 +846,24 @@ static inline void dquot_resv_space(stru
> > dquot->dq_dqb.dqb_rsvspace += number;
> > }
> >
> > +/*
> > + * Claim reserved quota space
> > + */
> > +static int dquot_claim_reserved_space(struct dquot *dquot,
> > + qsize_t number)
> > +{
> > + if (dquot->dq_dqb.dqb_rsvspace < number) {
> > + printk(KERN_WARNING "Reserved quota %llu is not enough for"
> > + "request %llu bytes\n",
> > + (unsigned long long)dquot->dq_dqb.dqb_rsvspace, number);
> > + return 1;
> Wouldn't a WARN_ON here be more appropriate? It's a filesystem bug to cause
> this AFAICS.
>
> > + }
> > +
> > + dquot->dq_dqb.dqb_curspace += number;
> > + dquot->dq_dqb.dqb_rsvspace -= number;
> > + return 0;
> > +}
> You should use dq_data_lock to protect these operations...
>
> > +
> > static inline void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
> > {
> > if (dquot->dq_dqb.dqb_curinodes > number)
> > @@ -1319,6 +1337,73 @@ out:
> > return ret;
> > }
> >
> > +int dquot_claim_space(struct inode *inode, qsize_t number)
> > +{
> > + int cnt;
> > + int ret = QUOTA_OK;
> > +
> > + if (IS_NOQUOTA(inode)) {
> > + inode_add_bytes(inode, number);
> > + return ret;
> > + }
> > +
> > + down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
> > + if (IS_NOQUOTA(inode)) {
> > + up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
> > + inode_add_bytes(inode, number);
> > + return ret;
> > + }
> > +
> > + /* Claim reserved quotas to allocated quotas */
> > + for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
> > + if (inode->i_dquot[cnt] != NODQUOT)
> > + ret = dquot_claim_reserved_space(inode->i_dquot[cnt],
> > + number);
> > + }
> > + if (ret == NO_QUOTA) {
> > + up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
> > + return ret;
> > + }
> It seems a bit silly to try to recover here from filesystem bugs. I'd just
> make dquot_claim_reserved_space() void and ignore possible failure here.
> We won't do anything harmful like loosing data. Just counters might become
> out of sync but given there's a bug in fs anyway it does not matter much.
>
> > + /* 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]);
> > + up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
> > +
> > + /* Update inode bytes */
> > + inode_add_bytes(inode, number);
> And this should be called from under dq_data_lock from
> dquot_claim_reserved_space().
> BTW: This reminds me that you should also modify dquot_transfer() function.
> Because that should not only transfer i_blocks sectors from one user
> to another but it has to also transfer the amount reserved for that inode...
> I think the easiest way around this would be to change i_blocks already when
> reservation is acquired (and than substract it when some of it is given back).
>
> > + return ret;
> > +}
> > +
> > +/*
> > + * Release reserved quota space
> > + */
> > +void dquot_release_reserved_space(struct inode *inode, qsize_t number)
> > +{
> > + int cnt;
> > + struct dquot *dquot;
> > +
> > + if (IS_NOQUOTA(inode))
> > + goto out;
> > +
> > + down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
> > + if (IS_NOQUOTA(inode))
> > + goto out_unlock;
> > +
> > + /* Release reserved dquots */
> > + for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
> > + if (inode->i_dquot[cnt] != NODQUOT) {
> > + dquot = inode->i_dquot[cnt];
> > + dquot->dq_dqb.dqb_rsvspace -= number;
> > + }
> > + }
> Again dq_data_lock...
>
> > +
> > +out_unlock:
> > + up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
> > +out:
> > + return;
> > +}
> > +
> > /*
> > * This operation can block, but only after everything is updated
> > */
> > @@ -2344,6 +2429,8 @@ EXPORT_SYMBOL(dquot_alloc_inode);
> > EXPORT_SYMBOL(dquot_free_space);
> > EXPORT_SYMBOL(dquot_free_inode);
> > EXPORT_SYMBOL(dquot_reserve_space);
> > +EXPORT_SYMBOL(dquot_claim_space);
> > +EXPORT_SYMBOL(dquot_release_reserved_space);
> > EXPORT_SYMBOL(dquot_transfer);
> > EXPORT_SYMBOL(vfs_dq_transfer);
> > EXPORT_SYMBOL(vfs_dq_quota_on_remount);
>
> Honza
Incremental fix
---
fs/dquot.c | 54 ++++++++++++++++++++++++++---------------------
include/linux/quotaops.h | 4 ---
2 files changed, 31 insertions(+), 27 deletions(-)
Index: linux-2.6.28-rc2/fs/dquot.c
===================================================================
--- linux-2.6.28-rc2.orig/fs/dquot.c 2008-12-09 15:56:29.000000000 -0800
+++ linux-2.6.28-rc2/fs/dquot.c 2008-12-09 17:09:14.000000000 -0800
@@ -849,19 +849,18 @@ static inline void dquot_resv_space(stru
/*
* Claim reserved quota space
*/
-static int dquot_claim_reserved_space(struct dquot *dquot,
+static void dquot_claim_reserved_space(struct dquot *dquot,
qsize_t number)
{
- if (dquot->dq_dqb.dqb_rsvspace < number) {
- printk(KERN_WARNING "Reserved quota %llu is not enough for"
- "request %llu bytes\n",
- (unsigned long long)dquot->dq_dqb.dqb_rsvspace, number);
- return 1;
- }
-
+ WARN_ON(dquot->dq_dqb.dqb_rsvspace < number);
dquot->dq_dqb.dqb_curspace += number;
dquot->dq_dqb.dqb_rsvspace -= number;
- return 0;
+}
+
+static inline
+void dquot_free_reserved_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)
@@ -1344,34 +1343,32 @@ int dquot_claim_space(struct inode *inod
if (IS_NOQUOTA(inode)) {
inode_add_bytes(inode, number);
- return ret;
+ goto out;
}
down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
if (IS_NOQUOTA(inode)) {
up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
inode_add_bytes(inode, number);
- return ret;
+ goto out;
}
+ spin_lock(&dq_data_lock);
/* Claim reserved quotas to allocated quotas */
for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
if (inode->i_dquot[cnt] != NODQUOT)
- ret = dquot_claim_reserved_space(inode->i_dquot[cnt],
+ dquot_claim_reserved_space(inode->i_dquot[cnt],
number);
}
- if (ret == NO_QUOTA) {
- up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
- return ret;
- }
+ /* Update inode bytes */
+ inode_add_bytes(inode, number);
+ spin_unlock(&dq_data_lock);
/* 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]);
up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
-
- /* Update inode bytes */
- inode_add_bytes(inode, number);
+out:
return ret;
}
@@ -1390,13 +1387,15 @@ void dquot_release_reserved_space(struct
if (IS_NOQUOTA(inode))
goto out_unlock;
+ spin_lock(&dq_data_lock);
/* Release reserved dquots */
for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
if (inode->i_dquot[cnt] != NODQUOT) {
dquot = inode->i_dquot[cnt];
- dquot->dq_dqb.dqb_rsvspace -= number;
+ dquot_free_reserved_space(dquot, number);
}
}
+ spin_unlock(&dq_data_lock);
out_unlock:
up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
@@ -1533,7 +1532,8 @@ int dquot_free_inode(const struct inode
*/
int dquot_transfer(struct inode *inode, struct iattr *iattr)
{
- qsize_t space;
+ qsize_t space, cur_space;
+ qsize_t rsv_space = 0;
struct dquot *transfer_from[MAXQUOTAS];
struct dquot *transfer_to[MAXQUOTAS];
int cnt, ret = NO_QUOTA, chuid = (iattr->ia_valid & ATTR_UID) && inode->i_uid != iattr->ia_uid,
@@ -1574,12 +1574,16 @@ int dquot_transfer(struct inode *inode,
}
}
spin_lock(&dq_data_lock);
- space = inode_get_bytes(inode);
+ space = cur_space = inode_get_bytes(inode);
/* Build the transfer_from list and check the limits */
for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
if (transfer_to[cnt] == NODQUOT)
continue;
transfer_from[cnt] = inode->i_dquot[cnt];
+ if (!rsv_space) {
+ rsv_space = transfer_from[cnt]->dq_dqb.dqb_rsvspace;
+ space += rsv_space;
+ }
if (check_idq(transfer_to[cnt], 1, warntype_to + cnt) ==
NO_QUOTA || check_bdq(transfer_to[cnt], space, 0,
warntype_to + cnt) == NO_QUOTA)
@@ -1603,11 +1607,13 @@ int dquot_transfer(struct inode *inode,
warntype_from_space[cnt] =
info_bdq_free(transfer_from[cnt], space);
dquot_decr_inodes(transfer_from[cnt], 1);
- dquot_decr_space(transfer_from[cnt], space);
+ dquot_decr_space(transfer_from[cnt], cur_space);
+ dquot_free_reserved_space(transfer_from[cnt], rsv_space);
}
dquot_incr_inodes(transfer_to[cnt], 1);
- dquot_incr_space(transfer_to[cnt], space);
+ dquot_incr_space(transfer_to[cnt], cur_space);
+ dquot_resv_space(transfer_to[cnt], rsv_space);
inode->i_dquot[cnt] = transfer_to[cnt];
}
Index: linux-2.6.28-rc2/include/linux/quotaops.h
===================================================================
--- linux-2.6.28-rc2.orig/include/linux/quotaops.h 2008-12-09 15:56:04.000000000 -0800
+++ linux-2.6.28-rc2/include/linux/quotaops.h 2008-12-09 15:58:09.000000000 -0800
@@ -434,7 +434,7 @@ static inline int vfs_dq_claim_block(str
}
static inline
-void vfs_dq_release_reservation(struct inode *inode, qsize_t nr)
+void vfs_dq_release_reservation_block(struct inode *inode, qsize_t nr)
{
vfs_dq_release_reservation_space(inode, nr << inode->i_blkbits);
}
@@ -467,8 +467,6 @@ static inline void vfs_dq_free_block(str
#define DQUOT_ALLOC_BLOCK_NODIRTY(inode, nr) \
vfs_dq_alloc_block_nodirty(inode, nr)
#define DQUOT_ALLOC_BLOCK(inode, nr) vfs_dq_alloc_block(inode, nr)
-#define DQUOT_CLAIM_BLOCK(inode, nr) vfs_dq_claim_block(inode, nr)
-#define DQUOT_RELEASE_RSV_BLOCK(inode, nr) vfs_dq_release_reservation(inode, nr)
#define DQUOT_ALLOC_INODE(inode) vfs_dq_alloc_inode(inode)
#define DQUOT_FREE_SPACE_NODIRTY(inode, nr) \
vfs_dq_free_space_nodirty(inode, nr)
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
