Viewing: osd_quota.c
// SPDX-License-Identifier: GPL-2.0
/*
* Copyright (c) 2012, 2017, Intel Corporation.
* Use is subject to license terms.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
*
* Author: Johann Lombardi <johann@whamcloud.com>
* Author: Niu Yawei <niu@whamcloud.com>
*/
#include <dt_object.h>
#include <lustre_quota.h>
#include "osd_internal.h"
/**
* Helpers function to find out the quota type (USRQUOTA/GRPQUOTA) of a
* given object
*/
static inline int fid2type(const struct lu_fid *fid)
{
LASSERT(fid_is_acct(fid));
switch (fid_oid(fid)) {
case ACCT_USER_OID:
return USRQUOTA;
case ACCT_GROUP_OID:
return GRPQUOTA;
case ACCT_PROJECT_OID:
return PRJQUOTA;
}
LASSERTF(0, "invalid fid for quota type: %u\n", fid_oid(fid));
return USRQUOTA;
}
/**
* Space Accounting Management
*/
/**
* Look up an accounting object based on its fid.
*
* \param info - is the osd thread info passed by the caller
* \param osd - is the osd device
* \param fid - is the fid of the accounting object we want to look up
* \param id - is the osd_inode_id struct to fill with the inode number of
* the quota file if the lookup is successful
*/
int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
const struct lu_fid *fid, struct osd_inode_id *id)
{
struct super_block *sb = osd_sb(osd);
ENTRY;
LASSERT(fid_is_acct(fid));
if (!ldiskfs_has_feature_quota(sb))
RETURN(-ENOENT);
/**
* ldiskfs won't load quota inodes on RO mount,
* So disable it in osd-ldiskfs to keep same behavior
* like lower layer to avoid further confusions.
*/
if (osd->od_dt_dev.dd_rdonly)
RETURN(-ENOENT);
id->oii_gen = OSD_OII_NOGEN;
switch (fid2type(fid)) {
case USRQUOTA:
id->oii_ino =
le32_to_cpu(LDISKFS_SB(sb)->s_es->s_usr_quota_inum);
break;
case GRPQUOTA:
id->oii_ino =
le32_to_cpu(LDISKFS_SB(sb)->s_es->s_grp_quota_inum);
break;
case PRJQUOTA:
if (ldiskfs_has_feature_project(sb)) {
__le32 prj_quota;
prj_quota = LDISKFS_SB(sb)->s_es->s_prj_quota_inum;
id->oii_ino = le32_to_cpu(prj_quota);
} else {
RETURN(-ENOENT);
}
break;
}
if (!ldiskfs_valid_inum(sb, id->oii_ino) &&
id->oii_ino != LDISKFS_USR_QUOTA_INO &&
id->oii_ino != LDISKFS_GRP_QUOTA_INO)
RETURN(-ENOENT);
RETURN(0);
}
/**
* Return space usage (#blocks & #inodes) consumed by a given uid or gid.
*
* \param env - is the environment passed by the caller
* \param dtobj - is the accounting object
* \param dtrec - is the record to fill with space usage information
* \param dtkey - is the id of the user or group for which we would
* like to access disk usage.
*
* \retval +ve - success : exact match
* \retval -ve - failure
*/
static int osd_acct_index_lookup(const struct lu_env *env,
struct dt_object *dtobj,
struct dt_rec *dtrec,
const struct dt_key *dtkey)
{
struct osd_thread_info *info = osd_oti_get(env);
struct qc_dqblk *dqblk = &info->oti_qdq;
struct super_block *sb = osd_sb(osd_obj2dev(osd_dt_obj(dtobj)));
struct lquota_acct_rec *rec = (struct lquota_acct_rec *)dtrec;
__u64 id = *((__u64 *)dtkey);
int rc;
struct kqid qid;
int type;
ENTRY;
type = fid2type(lu_object_fid(&dtobj->do_lu));
memset(dqblk, 0, sizeof(*dqblk));
qid = make_kqid(&init_user_ns, type, id);
rc = sb->s_qcop->get_dqblk(sb, qid, dqblk);
if (rc)
RETURN(rc);
rec->bspace = dqblk->d_space;
rec->ispace = dqblk->d_ino_count;
RETURN(+1);
}
#define QUOTA_IT_READ_ERROR(it, rc) \
CERROR("%s: Error while trying to read quota information, " \
"failed with %d\n", \
osd_dev(it->oiq_obj->oo_dt.do_lu.lo_dev)->od_svname, rc) \
/**
* Initialize osd Iterator for given osd index object.
*
* \param dt - osd index object
* \param attr - not used
*/
static struct dt_it *osd_it_acct_init(const struct lu_env *env,
struct dt_object *dt,
__u32 attr)
{
struct osd_it_quota *it;
struct lu_object *lo = &dt->do_lu;
struct osd_object *obj = osd_dt_obj(dt);
ENTRY;
LASSERT(lu_object_exists(lo));
OBD_ALLOC_PTR(it);
if (it == NULL)
RETURN(ERR_PTR(-ENOMEM));
lu_object_get(lo);
it->oiq_obj = obj;
INIT_LIST_HEAD(&it->oiq_list);
/* LUSTRE_DQTREEOFF is the initial offset where the tree can be found */
it->oiq_blk[0] = LUSTRE_DQTREEOFF;
/*
* NB: we don't need to store the tree depth since it is always
* equal to LUSTRE_DQTREEDEPTH - 1 (root has depth = 0) for a leaf
* block.
*/
RETURN((struct dt_it *)it);
}
/**
* Free given iterator.
*
* \param di - osd iterator
*/
static void osd_it_acct_fini(const struct lu_env *env, struct dt_it *di)
{
struct osd_it_quota *it = (struct osd_it_quota *)di;
struct osd_quota_leaf *leaf, *tmp;
ENTRY;
osd_object_put(env, it->oiq_obj);
list_for_each_entry_safe(leaf, tmp, &it->oiq_list, oql_link) {
list_del_init(&leaf->oql_link);
OBD_FREE_PTR(leaf);
}
OBD_FREE_PTR(it);
EXIT;
}
/**
* Move Iterator to record specified by \a key, if the \a key isn't found,
* move to the first valid record.
*
* \param di - osd iterator
* \param key - uid or gid
*
* \retval +ve - di points to the first valid record
* \retval +1 - di points to exact matched key
* \retval -ve - failure
*/
static int osd_it_acct_get(const struct lu_env *env, struct dt_it *di,
const struct dt_key *key)
{
struct osd_it_quota *it = (struct osd_it_quota *)di;
const struct lu_fid *fid = lu_object_fid(&it->oiq_obj->oo_dt.do_lu);
int type;
qid_t dqid = *(qid_t *)key;
loff_t offset;
int rc;
ENTRY;
type = fid2type(fid);
offset = find_tree_dqentry(env, it->oiq_obj, type, dqid,
LUSTRE_DQTREEOFF, 0, it);
if (offset > 0) { /* Found */
RETURN(+1);
} else if (offset < 0) { /* Error */
QUOTA_IT_READ_ERROR(it, (int)offset);
RETURN((int)offset);
}
/* The @key is not found, move to the first valid entry */
rc = walk_tree_dqentry(env, it->oiq_obj, type, it->oiq_blk[0], 0,
0, it);
if (rc == 0)
rc = 1;
else if (rc > 0)
rc = -ENOENT;
RETURN(rc);
}
/**
* Release Iterator
*
* \param di - osd iterator
*/
static void osd_it_acct_put(const struct lu_env *env, struct dt_it *di)
{
return;
}
static int osd_it_add_processed(struct osd_it_quota *it, int depth)
{
struct osd_quota_leaf *leaf;
OBD_ALLOC_PTR(leaf);
if (leaf == NULL)
RETURN(-ENOMEM);
INIT_LIST_HEAD(&leaf->oql_link);
leaf->oql_blk = it->oiq_blk[depth];
list_add_tail(&leaf->oql_link, &it->oiq_list);
RETURN(0);
}
/**
* Move on to the next valid entry.
*
* \param di - osd iterator
*
* \retval +ve - iterator reached the end
* \retval 0 - iterator has not reached the end yet
* \retval -ve - unexpected failure
*/
static int osd_it_acct_next(const struct lu_env *env, struct dt_it *di)
{
struct osd_it_quota *it = (struct osd_it_quota *)di;
const struct lu_fid *fid = lu_object_fid(&it->oiq_obj->oo_dt.do_lu);
int type;
int depth, rc;
uint index;
ENTRY;
type = fid2type(fid);
/*
* Let's first check if there are any remaining valid entry in the
* current leaf block. Start with the next entry after the current one.
*/
depth = LUSTRE_DQTREEDEPTH;
index = it->oiq_index[depth];
if (++index < LUSTRE_DQSTRINBLK) {
/* Search for the next valid entry from current index */
rc = walk_block_dqentry(env, it->oiq_obj, type,
it->oiq_blk[depth], index, it);
if (rc < 0) {
QUOTA_IT_READ_ERROR(it, rc);
RETURN(rc);
} else if (rc == 0) {
/*
* Found on entry, @it is already updated to the
* new position in walk_block_dqentry().
*/
RETURN(0);
} else {
rc = osd_it_add_processed(it, depth);
if (rc)
RETURN(rc);
}
} else {
rc = osd_it_add_processed(it, depth);
if (rc)
RETURN(rc);
}
rc = 1;
/*
* We have consumed all the entries of the current leaf block, move on
* to the next one.
*/
depth--;
/*
* We keep searching as long as walk_tree_dqentry() returns +1
* (= no valid entry found).
*/
for (; depth >= 0 && rc > 0; depth--) {
index = it->oiq_index[depth];
if (++index > 0xff)
continue;
rc = walk_tree_dqentry(env, it->oiq_obj, type,
it->oiq_blk[depth], depth, index, it);
}
if (rc < 0)
QUOTA_IT_READ_ERROR(it, rc);
RETURN(rc);
}
/**
* Return pointer to the key under iterator.
*
* \param di - osd iterator
*/
static struct dt_key *osd_it_acct_key(const struct lu_env *env,
const struct dt_it *di)
{
struct osd_it_quota *it = (struct osd_it_quota *)di;
ENTRY;
RETURN((struct dt_key *)&it->oiq_id);
}
/**
* Return size of key under iterator (in bytes)
*
* \param di - osd iterator
*/
static int osd_it_acct_key_size(const struct lu_env *env,
const struct dt_it *di)
{
struct osd_it_quota *it = (struct osd_it_quota *)di;
ENTRY;
RETURN((int)sizeof(it->oiq_id));
}
/**
* Return pointer to the record under iterator.
*
* \param di - osd iterator
* \param attr - not used
*/
static int osd_it_acct_rec(const struct lu_env *env,
const struct dt_it *di,
struct dt_rec *dtrec, __u32 attr)
{
struct osd_it_quota *it = (struct osd_it_quota *)di;
const struct dt_key *key = osd_it_acct_key(env, di);
int rc;
ENTRY;
rc = osd_acct_index_lookup(env, &it->oiq_obj->oo_dt, dtrec, key);
RETURN(rc > 0 ? 0 : rc);
}
/**
* Returns cookie for current Iterator position.
*
* \param di - osd iterator
*/
static __u64 osd_it_acct_store(const struct lu_env *env,
const struct dt_it *di)
{
struct osd_it_quota *it = (struct osd_it_quota *)di;
ENTRY;
RETURN(it->oiq_id);
}
/**
* Restore iterator from cookie. if the \a hash isn't found,
* restore the first valid record.
*
* \param di - osd iterator
* \param hash - iterator location cookie
*
* \retval +ve - di points to the first valid record
* \retval +1 - di points to exact matched hash
* \retval -ve - failure
*/
static int osd_it_acct_load(const struct lu_env *env,
const struct dt_it *di, __u64 hash)
{
struct osd_it_quota *it = (struct osd_it_quota *)di;
ENTRY;
/*
* LU-8999 - If it is called to resume the iteration, calling
* osd_it_acct_get could change the block orders in the lower level
* of the quota tree, which are saved in osd_it_quota->oiq_blk.
*/
if (it->oiq_id != 0 && it->oiq_id == hash)
RETURN(1);
RETURN(osd_it_acct_get(env, (struct dt_it *)di,
(const struct dt_key *)&hash));
}
/**
* Index and Iterator operations for accounting objects
*/
const struct dt_index_operations osd_acct_index_ops = {
.dio_lookup = osd_acct_index_lookup,
.dio_it = {
.init = osd_it_acct_init,
.fini = osd_it_acct_fini,
.get = osd_it_acct_get,
.put = osd_it_acct_put,
.next = osd_it_acct_next,
.key = osd_it_acct_key,
.key_size = osd_it_acct_key_size,
.rec = osd_it_acct_rec,
.store = osd_it_acct_store,
.load = osd_it_acct_load
}
};
static inline void osd_quota_swab(char *ptr, size_t size)
{
int offset;
LASSERT((size & (sizeof(__u64) - 1)) == 0);
for (offset = 0; offset < size; offset += sizeof(__u64))
__swab64s((__u64 *)(ptr + offset));
}
const struct dt_rec *osd_quota_pack(struct osd_object *obj,
const struct dt_rec *rec,
union lquota_rec *quota_rec)
{
#ifdef __BIG_ENDIAN
struct iam_descr *descr;
LASSERT(obj->oo_dir != NULL);
descr = obj->oo_dir->od_container.ic_descr;
memcpy(quota_rec, rec, descr->id_rec_size);
osd_quota_swab((char *)quota_rec, descr->id_rec_size);
return (const struct dt_rec *)quota_rec;
#else
return rec;
#endif
}
void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec)
{
#ifdef __BIG_ENDIAN
struct iam_descr *descr;
LASSERT(obj->oo_dir != NULL);
descr = obj->oo_dir->od_container.ic_descr;
osd_quota_swab((char *)rec, descr->id_rec_size);
#endif
}
static inline int osd_qid_type(struct osd_thandle *oh, int i)
{
return oh->ot_id_types[i];
}
/**
* Reserve journal credits for quota files update first, then call
* ->op_begin() to perform quota enforcement.
*
* \param env - the environment passed by the caller
* \param oh - osd transaction handle
* \param qi - quota id & space required for this operation
* \param obj - osd object, could be NULL when it's under create
* \param enforce - whether to perform quota enforcement
* \param flags - if the operation is write, return no user quota, no
* group quota, or sync commit flags to the caller
*
* \retval 0 - success
* \retval -ve - failure
*/
int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
struct lquota_id_info *qi, struct osd_object *obj,
bool enforce, enum osd_quota_local_flags *local_flags)
{
struct osd_device *dev;
struct qsd_instance *qsd;
struct lu_fid fid = { 0 };
struct inode *inode = NULL;
unsigned long long ino = 0;
int i, rc = 0, crd;
__u8 res = qi->lqi_is_blk ? LQUOTA_RES_DT : LQUOTA_RES_MD;
bool found = false;
ENTRY;
if (obj) {
fid = *lu_object_fid(&obj->oo_dt.do_lu);
inode = obj->oo_inode;
ino = inode ? inode->i_ino : 0;
}
CDEBUG(D_QUOTA, "fid="DFID" %p ino=%llu type=%u, id=%llu\n",
PFID(&fid), inode, ino, qi->lqi_type, qi->lqi_id.qid_uid);
LASSERT(oh != NULL);
LASSERTF(oh->ot_id_cnt <= OSD_MAX_UGID_CNT, "count=%d\n",
oh->ot_id_cnt);
dev = osd_dt_dev(oh->ot_super.th_dev);
LASSERT(dev != NULL);
if (res == LQUOTA_RES_DT)
qsd = dev->od_quota_slave_dt;
else
qsd = dev->od_quota_slave_md;
for (i = 0; i < oh->ot_id_cnt; i++) {
if (oh->ot_id_array[i] == qi->lqi_id.qid_uid &&
oh->ot_id_res[i] == res &&
oh->ot_id_types[i] == qi->lqi_type) {
found = true;
break;
}
}
if (!found) {
/* we need to account for credits for this new ID */
if (i >= OSD_MAX_UGID_CNT) {
rc = -EOVERFLOW;
CERROR("%s: too many qids %u > %u on "DFID": rc = %d\n",
osd_name(dev), i + 1, OSD_MAX_UGID_CNT,
PFID(&fid), rc);
RETURN(rc);
}
if (qi->lqi_id.qid_uid == 0 && qi->lqi_space > 0) {
/* root ID should be always present in the quota file,
* also only "target" uid (where we add space) is
* guaranteed, the source one can change after the
* declaration */
crd = 1;
} else {
/* can't rely on the current state as it can change
* by the execution.
* if used space for this ID could be dropped to zero,
* reserve extra credits for removing ID entry from
* the quota file
*/
if (qi->lqi_space < 0)
crd = LDISKFS_QUOTA_DEL_BLOCKS(osd_sb(dev));
else
crd = LDISKFS_QUOTA_INIT_BLOCKS(osd_sb(dev));
}
osd_trans_declare_op(env, oh, OSD_OT_QUOTA, crd);
oh->ot_id_array[i] = qi->lqi_id.qid_uid;
oh->ot_id_types[i] = qi->lqi_type;
oh->ot_id_res[i] = res;
oh->ot_id_cnt++;
}
if (unlikely(qsd == NULL))
/* quota slave instance hasn't been allocated yet */
RETURN(0);
/* check quota */
if (enforce)
rc = qsd_op_begin(env, qsd, oh->ot_quota_trans, qi,
local_flags);
RETURN(rc);
}
/**
* Wrapper for osd_declare_qid()
*
* \param env - the environment passed by the caller
* \param uid - user id of the inode
* \param gid - group id of the inode
* \param space - how many blocks/inodes will be consumed/released
* \param oh - osd transaction handle
* \param obj - osd object, could be NULL when it's under create
* \param flags - if the operation is write, return no user quota, no
* group quota, or sync commit flags to the caller
* \param osd_qid_flags - indicate this is a inode/block accounting
* and whether changes are performed by root user
*
* \retval 0 - success
* \retval -ve - failure
*/
int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
__u32 projid, long long space, struct osd_thandle *oh,
struct osd_object *obj,
enum osd_quota_local_flags *local_flags,
enum osd_qid_declare_flags osd_qid_declare_flags)
{
struct osd_thread_info *info = osd_oti_get(env);
struct lquota_id_info *qi = &info->oti_qi;
int rcu, rcg, rcp = 0; /* user & group & project rc */
struct thandle *th = &oh->ot_super;
enum osd_quota_local_flags tmp_flags;
bool force = !!(osd_qid_declare_flags & OSD_QID_FORCE) ||
th->th_ignore_quota;
ENTRY;
/* very fast path for special files like llog */
if (uid == 0 && gid == 0 && projid == 0)
return 0;
/* let's start with user quota */
qi->lqi_id.qid_uid = uid;
qi->lqi_type = USRQUOTA;
qi->lqi_space = space;
qi->lqi_is_blk = !!(osd_qid_declare_flags & OSD_QID_BLK);
rcu = osd_declare_qid(env, oh, qi, obj, true, local_flags);
if (force && (rcu == -EDQUOT || rcu == -EINPROGRESS))
/* ignore EDQUOT & EINPROGRESS when changes are done by root */
rcu = 0;
/*
* For non-fatal error, we want to continue to get the noquota flags
* for group id. This is only for commit write, which has @flags passed
* in. See osd_declare_write_commit().
* When force is set to true, we also want to proceed with the gid
*/
if (rcu && (rcu != -EDQUOT || local_flags == NULL))
RETURN(rcu);
/* and now group quota */
qi->lqi_id.qid_gid = gid;
qi->lqi_type = GRPQUOTA;
rcg = osd_declare_qid(env, oh, qi, obj, true, local_flags);
if (force && (rcg == -EDQUOT || rcg == -EINPROGRESS))
/* as before, ignore EDQUOT & EINPROGRESS for root */
rcg = 0;
if (rcg && (rcg != -EDQUOT || local_flags == NULL))
RETURN(rcg);
/* and now project quota */
qi->lqi_id.qid_projid = projid;
qi->lqi_ignore_root_proj_quota = th->th_ignore_root_proj_quota;
qi->lqi_type = PRJQUOTA;
tmp_flags = 0;
if (local_flags)
tmp_flags = *local_flags;
rcp = osd_declare_qid(env, oh, qi, obj, true, &tmp_flags);
if (tmp_flags & QUOTA_FL_ROOT_PRJQUOTA &&
!(osd_qid_declare_flags & OSD_QID_IGNORE_ROOT_PRJ))
/* Currently, th_ignore_quota is only set for inode quota
* in mdd_trans_create if the user has CAP_SYS_RESOURCE,
* then it should be ignored if root_prj_enable is set.
*/
force = 0;
if (local_flags)
*local_flags = tmp_flags;
if (force && (rcp == -EDQUOT || rcp == -EINPROGRESS)) {
CDEBUG(D_QUOTA, "forced to ignore quota flags = %#x\n",
local_flags ? *local_flags : -1);
/* as before, ignore EDQUOT & EINPROGRESS for root */
rcp = 0;
}
RETURN(rcu ? rcu : (rcg ? rcg : rcp));
}
