Skip to contentSkip to conversation

xfs: fix project quota ENOSPC vs EDQUOT

PATCH In review Eric Sandeen 5 participants 16 messages May 8, 2020 – May 14, 2020 XFS filesystem

lore
Share & tools
lore
#0 lore raw

cover xfs: fix project quota ENOSPC vs EDQUOT

Currently when project quota is enabled, group quota will return ENOSPC vs EDQUOT. This is the right error code for project quota but not for group quota; this is an old bug left over from when they were mutually exclusive.

#1 lore raw

PATCH 1/2 xfs: group quota should return EDQUOT when prj quota enabled

Long ago, group & project quota were mutually exclusive, and so when we turned on XFS_QMOPT_ENOSPC ("return ENOSPC if project quota is exceeded") when project quota was enabled, we only needed to disable it again for user quota.

When group & project quota got separated, this got missed, and as a result if project quota is enabled and group quota is exceeded, the error code returned is incorrectly returned as ENOSPC not EDQUOT.

Fix this by stripping XFS_QMOPT_ENOSPC out of flags for group quota when we try to reserve the space.

---

fs/xfs/xfs_trans_dquot.c +2 -1
index e4eca607e512..47c6e88c9db6 100644
@@ -757,7 +757,8 @@ xfs_trans_reserve_quota_bydquots(
 	}
 
 	if (gdqp) {
-		error = xfs_trans_dqresv(tp, mp, gdqp, nblks, ninos, flags);
+		error = xfs_trans_dqresv(tp, mp, gdqp, nblks, ninos,
+					(flags & ~XFS_QMOPT_ENOSPC));
 		if (error)
 			goto unwind_usr;
 	}
Reviewed-by on v1 1/2
#2 lore raw

Re: [PATCH 1/2] xfs: group quota should return EDQUOT when prj quota enabled

Quoted 11 lines · Eric Sandeen

Long ago, group & project quota were mutually exclusive, and so when we turned on XFS_QMOPT_ENOSPC ("return ENOSPC if project quota is exceeded") when project quota was enabled, we only needed to disable it again for user quota.

When group & project quota got separated, this got missed, and as a result if project quota is enabled and group quota is exceeded, the error code returned is incorrectly returned as ENOSPC not EDQUOT.

Fix this by stripping XFS_QMOPT_ENOSPC out of flags for group quota when we try to reserve the space.

Looks weird, but given that it has historical precedence, and you fix all this mess up in the next patch:

#3 lore raw

Re: [PATCH 1/2] xfs: group quota should return EDQUOT when prj quota enabled

Quoted 10 lines · Christoph Hellwig
Quoted 9 lines · Eric Sandeen

Long ago, group & project quota were mutually exclusive, and so when we turned on XFS_QMOPT_ENOSPC ("return ENOSPC if project quota is exceeded") when project quota was enabled, we only needed to disable it again for user quota.

When group & project quota got separated, this got missed, and as a result if project quota is enabled and group quota is exceeded, the error code returned is incorrectly returned as ENOSPC not EDQUOT.

Quoted 7 lines
Quoted 2 lines

Fix this by stripping XFS_QMOPT_ENOSPC out of flags for group quota when we try to reserve the space.

Looks weird, but given that it has historical precedence, and you fix all this mess up in the next patch:

Thanks. I could roll it all up into one patch if that's better, this seemed half weird / half more-obvious....

-Eric

Reviewed-by on v1 1/2
#4 lore raw

Re: [PATCH 1/2] xfs: group quota should return EDQUOT when prj quota enabled

In reply to Eric Sandeen
Quoted 9 lines · Eric Sandeen

Long ago, group & project quota were mutually exclusive, and so when we turned on XFS_QMOPT_ENOSPC ("return ENOSPC if project quota is exceeded") when project quota was enabled, we only needed to disable it again for user quota.

When group & project quota got separated, this got missed, and as a result if project quota is enabled and group quota is exceeded, the error code returned is incorrectly returned as ENOSPC not EDQUOT.

Quoted 5 lines

Fix this by stripping XFS_QMOPT_ENOSPC out of flags for group quota when we try to reserve the space.

---

Quoted 16 lines
fs/xfs/xfs_trans_dquot.c +2 -1
index e4eca607e512..47c6e88c9db6 100644
@@ -757,7 +757,8 @@ xfs_trans_reserve_quota_bydquots(
 	}
 
 	if (gdqp) {
-		error = xfs_trans_dqresv(tp, mp, gdqp, nblks, ninos, flags);
+		error = xfs_trans_dqresv(tp, mp, gdqp, nblks, ninos,
+					(flags & ~XFS_QMOPT_ENOSPC));
 		if (error)
 			goto unwind_usr;
 	}
#5 lore raw

PATCH 2/2 xfs: remove XFS_QMOPT_ENOSPC flag

In reply to Eric Sandeen

The only place we return -EDQUOT, and therefore need to make a decision about returning -ENOSPC for project quota instead, is in xfs_trans_dqresv().

So there's no reason to be setting and clearing XFS_QMOPT_ENOSPC at higher levels; if xfs_trans_dqresv has failed, test if the dqp we were were handed is a project quota and if so, return -ENOSPC instead of EDQUOT. The complexity is just a leftover from when project & group quota were mutually exclusive and shared some codepaths.

The prior patch was the trivial bugfix, this is the slightly more involved cleanup.

---

Patch 1 was the trivial bugfix, this is the slightly more involved cleanup.

fs/xfs/libxfs/xfs_quota_defs.h +0 -1
index b2113b17e53c..56d9dd787e7b 100644
@@ -100,7 +100,6 @@ typedef uint16_t	xfs_qwarncnt_t;
 #define XFS_QMOPT_FORCE_RES	0x0000010 /* ignore quota limits */
 #define XFS_QMOPT_SBVERSION	0x0000040 /* change superblock version num */
 #define XFS_QMOPT_GQUOTA	0x0002000 /* group dquot requested */
-#define XFS_QMOPT_ENOSPC	0x0004000 /* enospc instead of edquot (prj) */
 
 /*
  * flags to xfs_trans_mod_dquot to indicate which field needs to be
fs/xfs/xfs_qm.c +3 -6
index c225691fad15..591779aa2fd0 100644
@@ -1808,7 +1808,7 @@ xfs_qm_vop_chown_reserve(
 {
 	struct xfs_mount	*mp = ip->i_mount;
 	uint64_t		delblks;
-	unsigned int		blkflags, prjflags = 0;
+	unsigned int		blkflags;
 	struct xfs_dquot	*udq_unres = NULL;
 	struct xfs_dquot	*gdq_unres = NULL;
 	struct xfs_dquot	*pdq_unres = NULL;
@@ -1849,7 +1849,6 @@ xfs_qm_vop_chown_reserve(
 
 	if (XFS_IS_PQUOTA_ON(ip->i_mount) && pdqp &&
 	    ip->i_d.di_projid != be32_to_cpu(pdqp->q_core.d_id)) {
-		prjflags = XFS_QMOPT_ENOSPC;
 		pdq_delblks = pdqp;
 		if (delblks) {
 			ASSERT(ip->i_pdquot);
@@ -1859,8 +1858,7 @@ xfs_qm_vop_chown_reserve(
 
 	error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
 				udq_delblks, gdq_delblks, pdq_delblks,
-				ip->i_d.di_nblocks, 1,
-				flags | blkflags | prjflags);
+				ip->i_d.di_nblocks, 1, flags | blkflags);
 	if (error)
 		return error;
 
@@ -1878,8 +1876,7 @@ xfs_qm_vop_chown_reserve(
 		ASSERT(udq_unres || gdq_unres || pdq_unres);
 		error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
 			    udq_delblks, gdq_delblks, pdq_delblks,
-			    (xfs_qcnt_t)delblks, 0,
-			    flags | blkflags | prjflags);
+			    (xfs_qcnt_t)delblks, 0, flags | blkflags);
 		if (error)
 			return error;
 		xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
fs/xfs/xfs_trans_dquot.c +5 -11
index 2c3557a80e69..2c07897a3c37 100644
@@ -711,7 +711,7 @@ xfs_trans_dqresv(
 
 error_return:
 	xfs_dqunlock(dqp);
-	if (flags & XFS_QMOPT_ENOSPC)
+	if (XFS_QM_ISPDQ(dqp))
 		return -ENOSPC;
 	return -EDQUOT;
 }
@@ -751,15 +751,13 @@ xfs_trans_reserve_quota_bydquots(
 	ASSERT(flags & XFS_QMOPT_RESBLK_MASK);
 
 	if (udqp) {
-		error = xfs_trans_dqresv(tp, mp, udqp, nblks, ninos,
-					(flags & ~XFS_QMOPT_ENOSPC));
+		error = xfs_trans_dqresv(tp, mp, udqp, nblks, ninos, flags);
 		if (error)
 			return error;
 	}
 
 	if (gdqp) {
-		error = xfs_trans_dqresv(tp, mp, gdqp, nblks, ninos,
-					(flags & ~XFS_QMOPT_ENOSPC));
+		error = xfs_trans_dqresv(tp, mp, gdqp, nblks, ninos, flags);
 		if (error)
 			goto unwind_usr;
 	}
@@ -804,16 +802,12 @@ xfs_trans_reserve_quota_nblks(
 
 	if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
 		return 0;
-	if (XFS_IS_PQUOTA_ON(mp))
-		flags |= XFS_QMOPT_ENOSPC;
 
 	ASSERT(!xfs_is_quota_inode(&mp->m_sb, ip->i_ino));
 
 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
-	ASSERT((flags & ~(XFS_QMOPT_FORCE_RES | XFS_QMOPT_ENOSPC)) ==
-				XFS_TRANS_DQ_RES_RTBLKS ||
-	       (flags & ~(XFS_QMOPT_FORCE_RES | XFS_QMOPT_ENOSPC)) ==
-				XFS_TRANS_DQ_RES_BLKS);
+	ASSERT((flags & ~(XFS_QMOPT_FORCE_RES)) == XFS_TRANS_DQ_RES_RTBLKS ||
+	       (flags & ~(XFS_QMOPT_FORCE_RES)) == XFS_TRANS_DQ_RES_BLKS);
 
 	/*
 	 * Reserve nblks against these dquots, with trans as the mediator.
Reviewed-by on v1 2/2
#6 lore raw

Re: [PATCH 2/2] xfs: remove XFS_QMOPT_ENOSPC flag

Quoted 11 lines · Eric Sandeen

The only place we return -EDQUOT, and therefore need to make a decision about returning -ENOSPC for project quota instead, is in xfs_trans_dqresv().

So there's no reason to be setting and clearing XFS_QMOPT_ENOSPC at higher levels; if xfs_trans_dqresv has failed, test if the dqp we were were handed is a project quota and if so, return -ENOSPC instead of EDQUOT. The complexity is just a leftover from when project & group quota were mutually exclusive and shared some codepaths.

The prior patch was the trivial bugfix, this is the slightly more involved cleanup.

Nice:

#7 lore raw

Re: [PATCH 2/2] xfs: remove XFS_QMOPT_ENOSPC flag

In reply to Eric Sandeen
Quoted 9 lines · Eric Sandeen

The only place we return -EDQUOT, and therefore need to make a decision about returning -ENOSPC for project quota instead, is in xfs_trans_dqresv().

So there's no reason to be setting and clearing XFS_QMOPT_ENOSPC at higher levels; if xfs_trans_dqresv has failed, test if the dqp we were were handed is a project quota and if so, return -ENOSPC instead of EDQUOT. The complexity is just a leftover from when project & group quota were mutually exclusive and shared some codepaths.

Quoted 5 lines

The prior patch was the trivial bugfix, this is the slightly more involved cleanup.

---

Hmm so what about callers that don't pass QMOPT_ENOSPC? For example, it looks like various xfs_trans_reserve_quota() calls pass a pdqp but don't set the flag. Is the intent to change behavior such that -ENOSPC is unconditional for project quota reservation failures?

Brian

Quoted 108 lines

Patch 1 was the trivial bugfix, this is the slightly more involved cleanup.

fs/xfs/libxfs/xfs_quota_defs.h +0 -1
index b2113b17e53c..56d9dd787e7b 100644
@@ -100,7 +100,6 @@ typedef uint16_t	xfs_qwarncnt_t;
 #define XFS_QMOPT_FORCE_RES	0x0000010 /* ignore quota limits */
 #define XFS_QMOPT_SBVERSION	0x0000040 /* change superblock version num */
 #define XFS_QMOPT_GQUOTA	0x0002000 /* group dquot requested */
-#define XFS_QMOPT_ENOSPC	0x0004000 /* enospc instead of edquot (prj) */
 
 /*
  * flags to xfs_trans_mod_dquot to indicate which field needs to be
fs/xfs/xfs_qm.c +3 -6
index c225691fad15..591779aa2fd0 100644
@@ -1808,7 +1808,7 @@ xfs_qm_vop_chown_reserve(
 {
 	struct xfs_mount	*mp = ip->i_mount;
 	uint64_t		delblks;
-	unsigned int		blkflags, prjflags = 0;
+	unsigned int		blkflags;
 	struct xfs_dquot	*udq_unres = NULL;
 	struct xfs_dquot	*gdq_unres = NULL;
 	struct xfs_dquot	*pdq_unres = NULL;
@@ -1849,7 +1849,6 @@ xfs_qm_vop_chown_reserve(
 
 	if (XFS_IS_PQUOTA_ON(ip->i_mount) && pdqp &&
 	    ip->i_d.di_projid != be32_to_cpu(pdqp->q_core.d_id)) {
-		prjflags = XFS_QMOPT_ENOSPC;
 		pdq_delblks = pdqp;
 		if (delblks) {
 			ASSERT(ip->i_pdquot);
@@ -1859,8 +1858,7 @@ xfs_qm_vop_chown_reserve(
 
 	error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
 				udq_delblks, gdq_delblks, pdq_delblks,
-				ip->i_d.di_nblocks, 1,
-				flags | blkflags | prjflags);
+				ip->i_d.di_nblocks, 1, flags | blkflags);
 	if (error)
 		return error;
 
@@ -1878,8 +1876,7 @@ xfs_qm_vop_chown_reserve(
 		ASSERT(udq_unres || gdq_unres || pdq_unres);
 		error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
 			    udq_delblks, gdq_delblks, pdq_delblks,
-			    (xfs_qcnt_t)delblks, 0,
-			    flags | blkflags | prjflags);
+			    (xfs_qcnt_t)delblks, 0, flags | blkflags);
 		if (error)
 			return error;
 		xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
fs/xfs/xfs_trans_dquot.c +5 -11
index 2c3557a80e69..2c07897a3c37 100644
@@ -711,7 +711,7 @@ xfs_trans_dqresv(
 
 error_return:
 	xfs_dqunlock(dqp);
-	if (flags & XFS_QMOPT_ENOSPC)
+	if (XFS_QM_ISPDQ(dqp))
 		return -ENOSPC;
 	return -EDQUOT;
 }
@@ -751,15 +751,13 @@ xfs_trans_reserve_quota_bydquots(
 	ASSERT(flags & XFS_QMOPT_RESBLK_MASK);
 
 	if (udqp) {
-		error = xfs_trans_dqresv(tp, mp, udqp, nblks, ninos,
-					(flags & ~XFS_QMOPT_ENOSPC));
+		error = xfs_trans_dqresv(tp, mp, udqp, nblks, ninos, flags);
 		if (error)
 			return error;
 	}
 
 	if (gdqp) {
-		error = xfs_trans_dqresv(tp, mp, gdqp, nblks, ninos,
-					(flags & ~XFS_QMOPT_ENOSPC));
+		error = xfs_trans_dqresv(tp, mp, gdqp, nblks, ninos, flags);
 		if (error)
 			goto unwind_usr;
 	}
@@ -804,16 +802,12 @@ xfs_trans_reserve_quota_nblks(
 
 	if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
 		return 0;
-	if (XFS_IS_PQUOTA_ON(mp))
-		flags |= XFS_QMOPT_ENOSPC;
 
 	ASSERT(!xfs_is_quota_inode(&mp->m_sb, ip->i_ino));
 
 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
-	ASSERT((flags & ~(XFS_QMOPT_FORCE_RES | XFS_QMOPT_ENOSPC)) ==
-				XFS_TRANS_DQ_RES_RTBLKS ||
-	       (flags & ~(XFS_QMOPT_FORCE_RES | XFS_QMOPT_ENOSPC)) ==
-				XFS_TRANS_DQ_RES_BLKS);
+	ASSERT((flags & ~(XFS_QMOPT_FORCE_RES)) == XFS_TRANS_DQ_RES_RTBLKS ||
+	       (flags & ~(XFS_QMOPT_FORCE_RES)) == XFS_TRANS_DQ_RES_BLKS);
 
 	/*
 	 * Reserve nblks against these dquots, with trans as the mediator.
#8 lore raw

Re: [PATCH 2/2] xfs: remove XFS_QMOPT_ENOSPC flag

Quoted 13 lines · Brian Foster
Quoted 12 lines · Eric Sandeen

The only place we return -EDQUOT, and therefore need to make a decision about returning -ENOSPC for project quota instead, is in xfs_trans_dqresv().

So there's no reason to be setting and clearing XFS_QMOPT_ENOSPC at higher levels; if xfs_trans_dqresv has failed, test if the dqp we were were handed is a project quota and if so, return -ENOSPC instead of EDQUOT. The complexity is just a leftover from when project & group quota were mutually exclusive and shared some codepaths.

The prior patch was the trivial bugfix, this is the slightly more involved cleanup.

Quoted 6 lines
Quoted 2 lines

---

Hmm so what about callers that don't pass QMOPT_ENOSPC? For example, it looks like various xfs_trans_reserve_quota() calls pass a pdqp but don't set the flag.

Oh, interesting. I bet that was an oversight, tbh, but let's see.

<rewinds 14 years>

commit 9a2a7de268f67fea0c450ed3e99a2d31f43d7166

Author: Nathan Scott <[email protected]>
Date:   Fri Mar 31 13:04:49 2006 +1000

[XFS] Make project quota enforcement return an error code consistent with its use.

so yeah, even back then, stuff like xfs_symlink returned EDQUOT not ENOSPC.

Today, these call the reservation w/o the special ENOSPC flag:

xfs_unmap_extent
xfs_create
xfs_create_tmpfile
xfs_symlink

and so will return EDQUOT instead of ENOSPC even for project quota.

You're right that my patch changes these to ENOSPC.

Quoted 2 lines

Is the intent to change behavior such that -ENOSPC is unconditional for project quota reservation failures?

Now it's a conundrum. I /think/ the current behavior is due to an oversight, but

a) I'm not certain, and
b) can we change it now?

-Eric

#9 lore raw

Re: [PATCH 2/2] xfs: remove XFS_QMOPT_ENOSPC flag

Quoted 45 lines · Eric Sandeen
Quoted 13 lines · Brian Foster
Quoted 12 lines · Eric Sandeen

The only place we return -EDQUOT, and therefore need to make a decision about returning -ENOSPC for project quota instead, is in xfs_trans_dqresv().

So there's no reason to be setting and clearing XFS_QMOPT_ENOSPC at higher levels; if xfs_trans_dqresv has failed, test if the dqp we were were handed is a project quota and if so, return -ENOSPC instead of EDQUOT. The complexity is just a leftover from when project & group quota were mutually exclusive and shared some codepaths.

The prior patch was the trivial bugfix, this is the slightly more involved cleanup.

Quoted 6 lines
Quoted 2 lines

---

Hmm so what about callers that don't pass QMOPT_ENOSPC? For example, it looks like various xfs_trans_reserve_quota() calls pass a pdqp but don't set the flag.

Oh, interesting. I bet that was an oversight, tbh, but let's see.

<rewinds 14 years>

commit 9a2a7de268f67fea0c450ed3e99a2d31f43d7166

Author: Nathan Scott <[email protected]>
Date:   Fri Mar 31 13:04:49 2006 +1000

[XFS] Make project quota enforcement return an error code consistent with its use.

so yeah, even back then, stuff like xfs_symlink returned EDQUOT not ENOSPC.

Today, these call the reservation w/o the special ENOSPC flag:

xfs_unmap_extent
xfs_create
xfs_create_tmpfile
xfs_symlink

and so will return EDQUOT instead of ENOSPC even for project quota.

You're right that my patch changes these to ENOSPC.

Quoted 8 lines
Quoted 2 lines

Is the intent to change behavior such that -ENOSPC is unconditional for project quota reservation failures?

Now it's a conundrum. I /think/ the current behavior is due to an oversight, but

a) I'm not certain, and
b) can we change it now?

Heh, I can't really tell what the intended/expected behavior is. For whatever it's worth, it seems reasonable enough to me to change it based on the fact that project quotas have been expected to return -ENOSPC in at least some common cases for many years. It seems unlikely that users would know or care about the change in behavior in the subset noted above, but who knows. It might be good to get some other opinions. :P

Brian

Quoted 2 lines

-Eric

Reviewed-by
#10 lore raw

Re: [PATCH 2/2] xfs: remove XFS_QMOPT_ENOSPC flag

Quoted 55 lines · Brian Foster
Quoted 45 lines · Eric Sandeen
Quoted 13 lines · Brian Foster
Quoted 12 lines · Eric Sandeen

The only place we return -EDQUOT, and therefore need to make a decision about returning -ENOSPC for project quota instead, is in xfs_trans_dqresv().

So there's no reason to be setting and clearing XFS_QMOPT_ENOSPC at higher levels; if xfs_trans_dqresv has failed, test if the dqp we were were handed is a project quota and if so, return -ENOSPC instead of EDQUOT. The complexity is just a leftover from when project & group quota were mutually exclusive and shared some codepaths.

The prior patch was the trivial bugfix, this is the slightly more involved cleanup.

Quoted 6 lines
Quoted 2 lines

---

Hmm so what about callers that don't pass QMOPT_ENOSPC? For example, it looks like various xfs_trans_reserve_quota() calls pass a pdqp but don't set the flag.

Oh, interesting. I bet that was an oversight, tbh, but let's see.

<rewinds 14 years>

commit 9a2a7de268f67fea0c450ed3e99a2d31f43d7166

Author: Nathan Scott <[email protected]>
Date:   Fri Mar 31 13:04:49 2006 +1000

[XFS] Make project quota enforcement return an error code consistent with its use.

so yeah, even back then, stuff like xfs_symlink returned EDQUOT not ENOSPC.

Today, these call the reservation w/o the special ENOSPC flag:

xfs_unmap_extent
xfs_create
xfs_create_tmpfile
xfs_symlink

and so will return EDQUOT instead of ENOSPC even for project quota.

You're right that my patch changes these to ENOSPC.

Quoted 8 lines
Quoted 2 lines

Is the intent to change behavior such that -ENOSPC is unconditional for project quota reservation failures?

Now it's a conundrum. I /think/ the current behavior is due to an oversight, but

a) I'm not certain, and
b) can we change it now?

Quoted 6 lines

Heh, I can't really tell what the intended/expected behavior is. For whatever it's worth, it seems reasonable enough to me to change it based on the fact that project quotas have been expected to return -ENOSPC in at least some common cases for many years. It seems unlikely that users would know or care about the change in behavior in the subset noted above, but who knows. It might be good to get some other opinions. :P

"I bet you a beer at the next conference (if they ever happen again) that nobody will notice"? :P

TBH while I find it a little odd that project quota gets to return ENOSPC instead of EDQUOT, I find it more odd that sometimes it doesn't. This at least gets us to consistent behavior (EDQUOT for user/group, ENOSPC for project) so for the series:

(Let's see what an fstests run spits out...)

--D

Quoted 5 lines

Brian

Quoted 2 lines

-Eric

#11 lore raw

Re: [PATCH 2/2] xfs: remove XFS_QMOPT_ENOSPC flag

Quoted 63 lines · Darrick J. Wong
Quoted 55 lines · Brian Foster
Quoted 45 lines · Eric Sandeen
Quoted 13 lines · Brian Foster
Quoted 12 lines · Eric Sandeen

The only place we return -EDQUOT, and therefore need to make a decision about returning -ENOSPC for project quota instead, is in xfs_trans_dqresv().

So there's no reason to be setting and clearing XFS_QMOPT_ENOSPC at higher levels; if xfs_trans_dqresv has failed, test if the dqp we were were handed is a project quota and if so, return -ENOSPC instead of EDQUOT. The complexity is just a leftover from when project & group quota were mutually exclusive and shared some codepaths.

The prior patch was the trivial bugfix, this is the slightly more involved cleanup.

Quoted 6 lines
Quoted 2 lines

---

Hmm so what about callers that don't pass QMOPT_ENOSPC? For example, it looks like various xfs_trans_reserve_quota() calls pass a pdqp but don't set the flag.

Oh, interesting. I bet that was an oversight, tbh, but let's see.

<rewinds 14 years>

commit 9a2a7de268f67fea0c450ed3e99a2d31f43d7166

Author: Nathan Scott <[email protected]>
Date:   Fri Mar 31 13:04:49 2006 +1000

[XFS] Make project quota enforcement return an error code consistent with its use.

so yeah, even back then, stuff like xfs_symlink returned EDQUOT not ENOSPC.

Today, these call the reservation w/o the special ENOSPC flag:

xfs_unmap_extent
xfs_create
xfs_create_tmpfile
xfs_symlink

and so will return EDQUOT instead of ENOSPC even for project quota.

You're right that my patch changes these to ENOSPC.

Quoted 8 lines
Quoted 2 lines

Is the intent to change behavior such that -ENOSPC is unconditional for project quota reservation failures?

Now it's a conundrum. I /think/ the current behavior is due to an oversight, but

a) I'm not certain, and
b) can we change it now?

Quoted 6 lines

Heh, I can't really tell what the intended/expected behavior is. For whatever it's worth, it seems reasonable enough to me to change it based on the fact that project quotas have been expected to return -ENOSPC in at least some common cases for many years. It seems unlikely that users would know or care about the change in behavior in the subset noted above, but who knows. It might be good to get some other opinions. :P

Quoted 3 lines

"I bet you a beer at the next conference (if they ever happen again) that nobody will notice"? :P

Apocalypse aside, free beer is free beer. ;)

Quoted 5 lines

TBH while I find it a little odd that project quota gets to return ENOSPC instead of EDQUOT, I find it more odd that sometimes it doesn't. This at least gets us to consistent behavior (EDQUOT for user/group, ENOSPC for project) so for the series:

Works for me, but can we update the commit log to describe the behavior change before this goes in? In fact, it might even make sense to retitle the patch to something like "xfs: always return -ENOSPC on project quota reservation failure" and let the flag removal be a side effect of that.

Brian

Quoted 12 lines

(Let's see what an fstests run spits out...)

--D

Quoted 5 lines

Brian

Quoted 2 lines

-Eric

#12 lore raw

Re: [PATCH 2/2] xfs: remove XFS_QMOPT_ENOSPC flag

Quoted 4 lines · Brian Foster
Quoted 3 lines · Darrick J. Wong
Quoted 2 lines · Brian Foster
Quoted 1 line · Eric Sandeen

On 5/8/20 8:01 AM, Brian Foster wrote:

...

Quoted 30 lines
Quoted 18 lines
Quoted 11 lines
Quoted 10 lines

You're right that my patch changes these to ENOSPC.

Quoted 2 lines

Is the intent to change behavior such that -ENOSPC is unconditional for project quota reservation failures?

Now it's a conundrum. I /think/ the current behavior is due to an oversight, but

a) I'm not certain, and
b) can we change it now?

Quoted 6 lines

Heh, I can't really tell what the intended/expected behavior is. For whatever it's worth, it seems reasonable enough to me to change it based on the fact that project quotas have been expected to return -ENOSPC in at least some common cases for many years. It seems unlikely that users would know or care about the change in behavior in the subset noted above, but who knows. It might be good to get some other opinions. :P

Quoted 3 lines

"I bet you a beer at the next conference (if they ever happen again) that nobody will notice"? :P

Apocalypse aside, free beer is free beer. ;)

Quoted 5 lines

TBH while I find it a little odd that project quota gets to return ENOSPC instead of EDQUOT, I find it more odd that sometimes it doesn't. This at least gets us to consistent behavior (EDQUOT for user/group, ENOSPC for project) so for the series:

Quoted 4 lines

Works for me, but can we update the commit log to describe the behavior change before this goes in? In fact, it might even make sense to retitle the patch to something like "xfs: always return -ENOSPC on project quota reservation failure" and let the flag removal be a side effect of that.

Yes that's a good plan. I'm also happy to just combine the 2 patches if that's better. I'll sync up w/ Darrick to see if this can still happen.e

Thanks again for spotting the difference,

-Eric

Quoted 1 line

Brian

#13 lore raw

Re: [PATCH 2/2] xfs: remove XFS_QMOPT_ENOSPC flag

Quoted 43 lines · Eric Sandeen
Quoted 4 lines · Brian Foster
Quoted 3 lines · Darrick J. Wong
Quoted 2 lines · Brian Foster
Quoted 1 line · Eric Sandeen

On 5/8/20 8:01 AM, Brian Foster wrote:

...

Quoted 30 lines
Quoted 18 lines
Quoted 11 lines
Quoted 10 lines

You're right that my patch changes these to ENOSPC.

Quoted 2 lines

Is the intent to change behavior such that -ENOSPC is unconditional for project quota reservation failures?

Now it's a conundrum. I /think/ the current behavior is due to an oversight, but

a) I'm not certain, and
b) can we change it now?

Quoted 6 lines

Heh, I can't really tell what the intended/expected behavior is. For whatever it's worth, it seems reasonable enough to me to change it based on the fact that project quotas have been expected to return -ENOSPC in at least some common cases for many years. It seems unlikely that users would know or care about the change in behavior in the subset noted above, but who knows. It might be good to get some other opinions. :P

Quoted 3 lines

"I bet you a beer at the next conference (if they ever happen again) that nobody will notice"? :P

Apocalypse aside, free beer is free beer. ;)

Quoted 5 lines

TBH while I find it a little odd that project quota gets to return ENOSPC instead of EDQUOT, I find it more odd that sometimes it doesn't. This at least gets us to consistent behavior (EDQUOT for user/group, ENOSPC for project) so for the series:

Quoted 4 lines

Works for me, but can we update the commit log to describe the behavior change before this goes in? In fact, it might even make sense to retitle the patch to something like "xfs: always return -ENOSPC on project quota reservation failure" and let the flag removal be a side effect of that.

Quoted 4 lines

Yes that's a good plan. I'm also happy to just combine the 2 patches if that's better. I'll sync up w/ Darrick to see if this can still happen.e

Thanks again for spotting the difference,

Yeah, that's fine. I haven't even had a chance to find out if last night's testing passed... :$

--D

Quoted 3 lines

-Eric

Quoted 1 line

Brian

#14 lore raw

PATCH v2 2/2 xfs: always return -ENOSPC on project quota reservation failure

In reply to Eric Sandeen

XFS project quota treats project hierarchies as "mini filesysems" and so rather than -EDQUOT, the intent is to return -ENOSPC when a quota reservation fails, but this behavior is not consistent.

The only place we make a decision between -EDQUOT and -ENOSPC returns based on quota type is in xfs_trans_dqresv().

This behavior is currently controlled by whether or not the XFS_QMOPT_ENOSPC flag gets passed into the quota reservation. However, its use is not consistent; paths such as xfs_create() and xfs_symlink() don't set the flag, so a reservation failure will return -EDQUOT for project quota reservation failures rather than -ENOSPC for these sorts of operations, even for project quota:

# mkdir mnt/project
# xfs_quota -x -c "project -s -p mnt/project 42" mnt
# xfs_quota -x -c 'limit -p isoft=2 ihard=3 42' mnt
# touch mnt/project/file{1,2,3}
touch: cannot touch ‘mnt/project/file3’: Disk quota exceeded

We can make this consistent by not requiring the flag to be set at the top of the callchain; instead we can simply test whether we are reserving a project quota with XFS_QM_ISPDQ in xfs_trans_dqresv and if so, return -ENOSPC for that failure. This removes the need for the XFS_QMOPT_ENOSPC altogether and simplifies the code a fair bit.

---

V2: Change title and changelog to clearly reflect the bug and behavior change, which wasn't really identified as a bug and behavior change until Brian carefully reviewed V1. :)

No patch changes.

fs/xfs/libxfs/xfs_quota_defs.h +0 -1
index b2113b17e53c..56d9dd787e7b 100644
@@ -100,7 +100,6 @@ typedef uint16_t	xfs_qwarncnt_t;
 #define XFS_QMOPT_FORCE_RES	0x0000010 /* ignore quota limits */
 #define XFS_QMOPT_SBVERSION	0x0000040 /* change superblock version num */
 #define XFS_QMOPT_GQUOTA	0x0002000 /* group dquot requested */
-#define XFS_QMOPT_ENOSPC	0x0004000 /* enospc instead of edquot (prj) */
 
 /*
  * flags to xfs_trans_mod_dquot to indicate which field needs to be
fs/xfs/xfs_qm.c +3 -6
index c225691fad15..591779aa2fd0 100644
@@ -1808,7 +1808,7 @@ xfs_qm_vop_chown_reserve(
 {
 	struct xfs_mount	*mp = ip->i_mount;
 	uint64_t		delblks;
-	unsigned int		blkflags, prjflags = 0;
+	unsigned int		blkflags;
 	struct xfs_dquot	*udq_unres = NULL;
 	struct xfs_dquot	*gdq_unres = NULL;
 	struct xfs_dquot	*pdq_unres = NULL;
@@ -1849,7 +1849,6 @@ xfs_qm_vop_chown_reserve(
 
 	if (XFS_IS_PQUOTA_ON(ip->i_mount) && pdqp &&
 	    ip->i_d.di_projid != be32_to_cpu(pdqp->q_core.d_id)) {
-		prjflags = XFS_QMOPT_ENOSPC;
 		pdq_delblks = pdqp;
 		if (delblks) {
 			ASSERT(ip->i_pdquot);
@@ -1859,8 +1858,7 @@ xfs_qm_vop_chown_reserve(
 
 	error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
 				udq_delblks, gdq_delblks, pdq_delblks,
-				ip->i_d.di_nblocks, 1,
-				flags | blkflags | prjflags);
+				ip->i_d.di_nblocks, 1, flags | blkflags);
 	if (error)
 		return error;
 
@@ -1878,8 +1876,7 @@ xfs_qm_vop_chown_reserve(
 		ASSERT(udq_unres || gdq_unres || pdq_unres);
 		error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
 			    udq_delblks, gdq_delblks, pdq_delblks,
-			    (xfs_qcnt_t)delblks, 0,
-			    flags | blkflags | prjflags);
+			    (xfs_qcnt_t)delblks, 0, flags | blkflags);
 		if (error)
 			return error;
 		xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
fs/xfs/xfs_trans_dquot.c +5 -11
index 2c3557a80e69..2c07897a3c37 100644
@@ -711,7 +711,7 @@ xfs_trans_dqresv(
 
 error_return:
 	xfs_dqunlock(dqp);
-	if (flags & XFS_QMOPT_ENOSPC)
+	if (XFS_QM_ISPDQ(dqp))
 		return -ENOSPC;
 	return -EDQUOT;
 }
@@ -751,15 +751,13 @@ xfs_trans_reserve_quota_bydquots(
 	ASSERT(flags & XFS_QMOPT_RESBLK_MASK);
 
 	if (udqp) {
-		error = xfs_trans_dqresv(tp, mp, udqp, nblks, ninos,
-					(flags & ~XFS_QMOPT_ENOSPC));
+		error = xfs_trans_dqresv(tp, mp, udqp, nblks, ninos, flags);
 		if (error)
 			return error;
 	}
 
 	if (gdqp) {
-		error = xfs_trans_dqresv(tp, mp, gdqp, nblks, ninos,
-					(flags & ~XFS_QMOPT_ENOSPC));
+		error = xfs_trans_dqresv(tp, mp, gdqp, nblks, ninos, flags);
 		if (error)
 			goto unwind_usr;
 	}
@@ -804,16 +802,12 @@ xfs_trans_reserve_quota_nblks(
 
 	if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
 		return 0;
-	if (XFS_IS_PQUOTA_ON(mp))
-		flags |= XFS_QMOPT_ENOSPC;
 
 	ASSERT(!xfs_is_quota_inode(&mp->m_sb, ip->i_ino));
 
 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
-	ASSERT((flags & ~(XFS_QMOPT_FORCE_RES | XFS_QMOPT_ENOSPC)) ==
-				XFS_TRANS_DQ_RES_RTBLKS ||
-	       (flags & ~(XFS_QMOPT_FORCE_RES | XFS_QMOPT_ENOSPC)) ==
-				XFS_TRANS_DQ_RES_BLKS);
+	ASSERT((flags & ~(XFS_QMOPT_FORCE_RES)) == XFS_TRANS_DQ_RES_RTBLKS ||
+	       (flags & ~(XFS_QMOPT_FORCE_RES)) == XFS_TRANS_DQ_RES_BLKS);
 
 	/*
 	 * Reserve nblks against these dquots, with trans as the mediator.
Reviewed-by on v2 2/2
#15 lore raw

Re: [PATCH 2/2 V2] xfs: always return -ENOSPC on project quota reservation failure

Quoted 26 lines · Eric Sandeen

XFS project quota treats project hierarchies as "mini filesysems" and so rather than -EDQUOT, the intent is to return -ENOSPC when a quota reservation fails, but this behavior is not consistent.

The only place we make a decision between -EDQUOT and -ENOSPC returns based on quota type is in xfs_trans_dqresv().

This behavior is currently controlled by whether or not the XFS_QMOPT_ENOSPC flag gets passed into the quota reservation. However, its use is not consistent; paths such as xfs_create() and xfs_symlink() don't set the flag, so a reservation failure will return -EDQUOT for project quota reservation failures rather than -ENOSPC for these sorts of operations, even for project quota:

# mkdir mnt/project
# xfs_quota -x -c "project -s -p mnt/project 42" mnt
# xfs_quota -x -c 'limit -p isoft=2 ihard=3 42' mnt
# touch mnt/project/file{1,2,3}
touch: cannot touch ‘mnt/project/file3’: Disk quota exceeded

We can make this consistent by not requiring the flag to be set at the top of the callchain; instead we can simply test whether we are reserving a project quota with XFS_QM_ISPDQ in xfs_trans_dqresv and if so, return -ENOSPC for that failure. This removes the need for the XFS_QMOPT_ENOSPC altogether and simplifies the code a fair bit.

Quoted 3 lines

---

Thanks for the update.

Quoted 113 lines

V2: Change title and changelog to clearly reflect the bug and behavior change, which wasn't really identified as a bug and behavior change until Brian carefully reviewed V1. :)

No patch changes.

fs/xfs/libxfs/xfs_quota_defs.h +0 -1
index b2113b17e53c..56d9dd787e7b 100644
@@ -100,7 +100,6 @@ typedef uint16_t	xfs_qwarncnt_t;
 #define XFS_QMOPT_FORCE_RES	0x0000010 /* ignore quota limits */
 #define XFS_QMOPT_SBVERSION	0x0000040 /* change superblock version num */
 #define XFS_QMOPT_GQUOTA	0x0002000 /* group dquot requested */
-#define XFS_QMOPT_ENOSPC	0x0004000 /* enospc instead of edquot (prj) */
 
 /*
  * flags to xfs_trans_mod_dquot to indicate which field needs to be
fs/xfs/xfs_qm.c +3 -6
index c225691fad15..591779aa2fd0 100644
@@ -1808,7 +1808,7 @@ xfs_qm_vop_chown_reserve(
 {
 	struct xfs_mount	*mp = ip->i_mount;
 	uint64_t		delblks;
-	unsigned int		blkflags, prjflags = 0;
+	unsigned int		blkflags;
 	struct xfs_dquot	*udq_unres = NULL;
 	struct xfs_dquot	*gdq_unres = NULL;
 	struct xfs_dquot	*pdq_unres = NULL;
@@ -1849,7 +1849,6 @@ xfs_qm_vop_chown_reserve(
 
 	if (XFS_IS_PQUOTA_ON(ip->i_mount) && pdqp &&
 	    ip->i_d.di_projid != be32_to_cpu(pdqp->q_core.d_id)) {
-		prjflags = XFS_QMOPT_ENOSPC;
 		pdq_delblks = pdqp;
 		if (delblks) {
 			ASSERT(ip->i_pdquot);
@@ -1859,8 +1858,7 @@ xfs_qm_vop_chown_reserve(
 
 	error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
 				udq_delblks, gdq_delblks, pdq_delblks,
-				ip->i_d.di_nblocks, 1,
-				flags | blkflags | prjflags);
+				ip->i_d.di_nblocks, 1, flags | blkflags);
 	if (error)
 		return error;
 
@@ -1878,8 +1876,7 @@ xfs_qm_vop_chown_reserve(
 		ASSERT(udq_unres || gdq_unres || pdq_unres);
 		error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
 			    udq_delblks, gdq_delblks, pdq_delblks,
-			    (xfs_qcnt_t)delblks, 0,
-			    flags | blkflags | prjflags);
+			    (xfs_qcnt_t)delblks, 0, flags | blkflags);
 		if (error)
 			return error;
 		xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
fs/xfs/xfs_trans_dquot.c +5 -11
index 2c3557a80e69..2c07897a3c37 100644
@@ -711,7 +711,7 @@ xfs_trans_dqresv(
 
 error_return:
 	xfs_dqunlock(dqp);
-	if (flags & XFS_QMOPT_ENOSPC)
+	if (XFS_QM_ISPDQ(dqp))
 		return -ENOSPC;
 	return -EDQUOT;
 }
@@ -751,15 +751,13 @@ xfs_trans_reserve_quota_bydquots(
 	ASSERT(flags & XFS_QMOPT_RESBLK_MASK);
 
 	if (udqp) {
-		error = xfs_trans_dqresv(tp, mp, udqp, nblks, ninos,
-					(flags & ~XFS_QMOPT_ENOSPC));
+		error = xfs_trans_dqresv(tp, mp, udqp, nblks, ninos, flags);
 		if (error)
 			return error;
 	}
 
 	if (gdqp) {
-		error = xfs_trans_dqresv(tp, mp, gdqp, nblks, ninos,
-					(flags & ~XFS_QMOPT_ENOSPC));
+		error = xfs_trans_dqresv(tp, mp, gdqp, nblks, ninos, flags);
 		if (error)
 			goto unwind_usr;
 	}
@@ -804,16 +802,12 @@ xfs_trans_reserve_quota_nblks(
 
 	if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
 		return 0;
-	if (XFS_IS_PQUOTA_ON(mp))
-		flags |= XFS_QMOPT_ENOSPC;
 
 	ASSERT(!xfs_is_quota_inode(&mp->m_sb, ip->i_ino));
 
 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
-	ASSERT((flags & ~(XFS_QMOPT_FORCE_RES | XFS_QMOPT_ENOSPC)) ==
-				XFS_TRANS_DQ_RES_RTBLKS ||
-	       (flags & ~(XFS_QMOPT_FORCE_RES | XFS_QMOPT_ENOSPC)) ==
-				XFS_TRANS_DQ_RES_BLKS);
+	ASSERT((flags & ~(XFS_QMOPT_FORCE_RES)) == XFS_TRANS_DQ_RES_RTBLKS ||
+	       (flags & ~(XFS_QMOPT_FORCE_RES)) == XFS_TRANS_DQ_RES_BLKS);
 
 	/*
 	 * Reserve nblks against these dquots, with trans as the mediator.