* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, 2012, Intel Corporation.
+ * Copyright (c) 2011, 2013, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#include <linux/fs.h>
/* XATTR_{REPLACE,CREATE} */
#include <linux/xattr.h>
-/* simple_mkdir() */
-#include <lvfs.h>
/*
* struct OBD_{ALLOC,FREE}*()
/* struct ptlrpc_thread */
#include <lustre_net.h>
#include <lustre_fid.h>
+/* process_config */
+#include <lustre_param.h>
#include "osd_internal.h"
+#include "osd_dynlocks.h"
/* llo_* api support */
#include <md_object.h>
CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
"ldiskfs with parallel directory operations");
+int ldiskfs_track_declares_assert;
+CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644,
+ "LBUG during tracking of declares");
+
+/* Slab to allocate dynlocks */
+struct kmem_cache *dynlock_cachep;
+
+static struct lu_kmem_descr ldiskfs_caches[] = {
+ {
+ .ckd_cache = &dynlock_cachep,
+ .ckd_name = "dynlock_cache",
+ .ckd_size = sizeof(struct dynlock_handle)
+ },
+ {
+ .ckd_cache = NULL
+ }
+};
+
static const char dot[] = ".";
static const char dotdot[] = "..";
static const char remote_obj_dir[] = "REM_OBJ_DIR";
static const struct dt_index_operations osd_index_iam_ops;
static const struct dt_index_operations osd_index_ea_ops;
-#ifdef OSD_TRACK_DECLARES
int osd_trans_declare_op2rb[] = {
[OSD_OT_ATTR_SET] = OSD_OT_ATTR_SET,
[OSD_OT_PUNCH] = OSD_OT_MAX,
[OSD_OT_WRITE] = OSD_OT_WRITE,
[OSD_OT_INSERT] = OSD_OT_DELETE,
[OSD_OT_DELETE] = OSD_OT_INSERT,
+ [OSD_OT_UPDATE] = OSD_OT_MAX,
[OSD_OT_QUOTA] = OSD_OT_MAX,
};
-#endif
static int osd_has_index(const struct osd_object *obj)
{
}
}
-static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry,
- const char *name, void *buf, int len)
-{
- dentry->d_inode = inode;
- return inode->i_op->getxattr(dentry, name, buf, len);
-}
-
int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
struct dentry *dentry, struct lustre_mdt_attrs *lma)
{
int rc;
- rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA, (void *)lma,
- sizeof(*lma));
- if (rc == -ERANGE) {
- /* try with old lma size */
- rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA,
- info->oti_mdt_attrs_old,
- LMA_OLD_SIZE);
- if (rc > 0)
- memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
- }
+ CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
+ rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
+ info->oti_mdt_attrs_old, LMA_OLD_SIZE);
if (rc > 0) {
+ if ((void *)lma != (void *)info->oti_mdt_attrs_old)
+ memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
+ rc = 0;
+ lustre_lma_swab(lma);
/* Check LMA compatibility */
- if (lma->lma_incompat & ~cpu_to_le32(LMA_INCOMPAT_SUPP)) {
- CWARN("%.16s: unsupported incompat LMA feature(s) "
- "%lx/%#x\n",
+ if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
+ CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
+ "for fid = "DFID", ino = %lu\n",
LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
- inode->i_ino, le32_to_cpu(lma->lma_incompat) &
- ~LMA_INCOMPAT_SUPP);
- rc = -ENOSYS;
- } else {
- lustre_lma_swab(lma);
- rc = 0;
+ lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
+ PFID(&lma->lma_self_fid), inode->i_ino);
+ rc = -EOPNOTSUPP;
}
} else if (rc == 0) {
rc = -ENODATA;
return inode;
}
-static struct inode *
-osd_iget_verify(struct osd_thread_info *info, struct osd_device *dev,
- struct osd_inode_id *id, const struct lu_fid *fid)
+/**
+ * \retval +v: new filter_fid, does not contain self-fid
+ * \retval 0: filter_fid_old, contains self-fid
+ * \retval -v: other failure cases
+ */
+int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
+ struct dentry *dentry, struct lu_fid *fid)
{
- struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
- struct inode *inode;
+ struct filter_fid_old *ff = &info->oti_ff;
+ struct ost_id *ostid = &info->oti_ostid;
int rc;
- inode = osd_iget(info, dev, id);
- if (IS_ERR(inode))
- return inode;
+ rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
+ if (rc == sizeof(*ff)) {
+ rc = 0;
+ ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
+ ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
+ /* XXX: should use real OST index in the future. LU-3569 */
+ ostid_to_fid(fid, ostid, 0);
+ } else if (rc == sizeof(struct filter_fid)) {
+ rc = 1;
+ } else if (rc >= 0) {
+ rc = -EINVAL;
+ }
- rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
- if (rc == -ENODATA)
- return inode;
+ return rc;
+}
- if (rc != 0) {
- iput(inode);
- return ERR_PTR(rc);
+static int osd_lma_self_repair(struct osd_thread_info *info,
+ struct osd_device *osd, struct inode *inode,
+ const struct lu_fid *fid, __u32 compat)
+{
+ handle_t *jh;
+ int rc;
+
+ LASSERT(current->journal_info == NULL);
+
+ jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
+ osd_dto_credits_noquota[DTO_XATTR_SET]);
+ if (IS_ERR(jh)) {
+ rc = PTR_ERR(jh);
+ CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
+ osd_name(osd), rc);
+ return rc;
}
- if (!lu_fid_eq(fid, &lma->lma_self_fid)) {
- CDEBUG(D_LFSCK, "inconsistent obj: "DFID", %lu, "DFID"\n",
- PFID(&lma->lma_self_fid), inode->i_ino, PFID(fid));
- iput(inode);
- return ERR_PTR(-EREMCHG);
+ rc = osd_ea_fid_set(info, inode, fid, compat, 0);
+ if (rc != 0)
+ CWARN("%s: cannot self repair the LMA: rc = %d\n",
+ osd_name(osd), rc);
+ ldiskfs_journal_stop(jh);
+ return rc;
+}
+
+static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
+{
+ struct osd_thread_info *info = osd_oti_get(env);
+ struct osd_device *osd = osd_obj2dev(obj);
+ struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
+ struct inode *inode = obj->oo_inode;
+ struct dentry *dentry = &info->oti_obj_dentry;
+ struct lu_fid *fid = NULL;
+ const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
+ int rc;
+ ENTRY;
+
+ if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
+ RETURN(0);
+
+ CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
+ rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
+ info->oti_mdt_attrs_old, LMA_OLD_SIZE);
+ if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
+ fid = &lma->lma_self_fid;
+ rc = osd_get_idif(info, inode, dentry, fid);
+ if ((rc > 0) || (rc == -ENODATA && osd->od_lma_self_repair)) {
+ /* For the given OST-object, if it has neither LMA nor
+ * FID in XATTR_NAME_FID, then the given FID (which is
+ * contained in the @obj, from client RPC for locating
+ * the OST-object) is trusted. We use it to generate
+ * the LMA. */
+ osd_lma_self_repair(info, osd, inode, rfid,
+ fid_is_on_ost(info, osd, fid, OI_CHECK_FLD) ?
+ LMAC_FID_ON_OST : 0);
+ RETURN(0);
+ }
}
- return inode;
+ if (unlikely(rc == -ENODATA))
+ RETURN(0);
+
+ if (rc < 0)
+ RETURN(rc);
+
+ if (rc > 0) {
+ rc = 0;
+ lustre_lma_swab(lma);
+ if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
+ CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
+ CWARN("%s: unsupported incompat LMA feature(s) %#x for "
+ "fid = "DFID", ino = %lu\n", osd_name(osd),
+ lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
+ PFID(rfid), inode->i_ino);
+ rc = -EOPNOTSUPP;
+ } else if (!(lma->lma_compat & LMAC_NOT_IN_OI)) {
+ fid = &lma->lma_self_fid;
+ }
+ }
+
+ if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
+ if (fid_is_idif(rfid) && fid_is_idif(fid)) {
+ struct ost_id *oi = &info->oti_ostid;
+ struct lu_fid *fid1 = &info->oti_fid3;
+ __u32 idx = fid_idif_ost_idx(rfid);
+
+ /* For old IDIF, the OST index is not part of the IDIF,
+ * Means that different OSTs may have the same IDIFs.
+ * Under such case, we need to make some compatible
+ * check to make sure to trigger OI scrub properly. */
+ if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
+ /* Given @rfid is new, LMA is old. */
+ fid_to_ostid(fid, oi);
+ ostid_to_fid(fid1, oi, idx);
+ if (lu_fid_eq(fid1, rfid)) {
+ if (osd->od_lma_self_repair)
+ osd_lma_self_repair(info, osd,
+ inode, rfid,
+ LMAC_FID_ON_OST);
+ RETURN(0);
+ }
+ }
+ }
+
+ CDEBUG(D_INODE, "%s: FID "DFID" != self_fid "DFID"\n",
+ osd_name(osd), PFID(rfid), PFID(fid));
+ rc = -EREMCHG;
+ }
+
+ RETURN(rc);
}
static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
struct osd_scrub *scrub;
struct scrub_file *sf;
int result;
- int verify = 0;
+ int saved = 0;
+ bool in_oi = false;
+ bool triggered = false;
ENTRY;
LINVRNT(osd_invariant(obj));
if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
RETURN(-ENOENT);
+ /* For the object is created as locking anchor, or for the object to
+ * be created on disk. No need to osd_oi_lookup() at here because FID
+ * shouldn't never be re-used, if it's really a duplicate FID from
+ * unexpected reason, we should be able to detect it later by calling
+ * do_create->osd_oi_insert(). */
+ if (conf != NULL && conf->loc_flags & LOC_F_NEW)
+ GOTO(out, result = 0);
+
/* Search order: 1. per-thread cache. */
- if (lu_fid_eq(fid, &oic->oic_fid)) {
+ if (lu_fid_eq(fid, &oic->oic_fid) &&
+ likely(oic->oic_dev == dev)) {
id = &oic->oic_lid;
goto iget;
}
goto iget;
}
- if (sf->sf_flags & SF_INCONSISTENT)
- verify = 1;
-
- /*
- * Objects are created as locking anchors or place holders for objects
- * yet to be created. No need to osd_oi_lookup() at here because FID
- * shouldn't never be re-used, if it's really a duplicate FID from
- * unexpected reason, we should be able to detect it later by calling
- * do_create->osd_oi_insert()
- */
- if (conf != NULL && conf->loc_flags & LOC_F_NEW)
- GOTO(out, result = 0);
-
/* Search order: 3. OI files. */
- result = osd_oi_lookup(info, dev, fid, id, true);
+ result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
if (result == -ENOENT) {
- if (!fid_is_norm(fid) || fid_is_on_ost(info, dev, fid) ||
+ if (!fid_is_norm(fid) ||
+ fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
!ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
sf->sf_oi_bitmap))
GOTO(out, result = 0);
if (result != 0)
GOTO(out, result);
+ in_oi = true;
+
iget:
- if (verify == 0)
- inode = osd_iget(info, dev, id);
- else
- inode = osd_iget_verify(info, dev, id, fid);
+ inode = osd_iget(info, dev, id);
if (IS_ERR(inode)) {
result = PTR_ERR(inode);
if (result == -ENOENT || result == -ESTALE) {
- fid_zero(&oic->oic_fid);
- result = 0;
+ if (!in_oi) {
+ fid_zero(&oic->oic_fid);
+ GOTO(out, result = -ENOENT);
+ }
+
+ /* XXX: There are three possible cases:
+ * 1. Backup/restore caused the OI invalid.
+ * 2. Someone unlinked the object but NOT removed
+ * the OI mapping, such as mount target device
+ * as ldiskfs, and modify something directly.
+ * 3. Someone just removed the object between the
+ * former oi_lookup and the iget. It is normal.
+ *
+ * It is diffcult to distinguish the 2nd from the
+ * 1st case. Relatively speaking, the 1st case is
+ * common than the 2nd case, trigger OI scrub. */
+ result = osd_oi_lookup(info, dev, fid, id, true);
+ if (result == 0)
+ /* It is the case 1 or 2. */
+ goto trigger;
} else if (result == -EREMCHG) {
trigger:
+ if (unlikely(triggered))
+ GOTO(out, result = saved);
+
+ triggered = true;
if (thread_is_running(&scrub->os_thread)) {
result = -EINPROGRESS;
} else if (!dev->od_noscrub) {
else
result = -EREMCHG;
}
+
+ /* We still have chance to get the valid inode: for the
+ * object which is referenced by remote name entry, the
+ * object on the local MDT will be linked under the dir
+ * of "/REMOTE_PARENT_DIR" with its FID string as name.
+ *
+ * We do not know whether the object for the given FID
+ * is referenced by some remote name entry or not, and
+ * especially for DNE II, a multiple-linked object may
+ * have many name entries reside on many MDTs.
+ *
+ * To simplify the operation, OSD will not distinguish
+ * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
+ * only happened for the RPC from other MDT during the
+ * OI scrub, or for the client side RPC with FID only,
+ * such as FID to path, or from old connected client. */
+ saved = result;
+ result = osd_lookup_in_remote_parent(info, dev,
+ fid, id);
+ if (result == 0) {
+ in_oi = false;
+ goto iget;
+ }
+
+ result = saved;
}
GOTO(out, result);
obj->oo_inode = inode;
LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
+ result = osd_check_lma(env, obj);
+ if (result != 0) {
+ iput(inode);
+ obj->oo_inode = NULL;
+ if (result == -EREMCHG)
+ goto trigger;
+
+ GOTO(out, result);
+ }
+
obj->oo_compat_dot_created = 1;
obj->oo_compat_dotdot_created = 1;
dcb->dcb_func(NULL, th, dcb, error);
}
- lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th);
+ lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
lu_device_put(lud);
th->th_dev = NULL;
LASSERT(cfs_atomic_read(&iobuf->dr_numreqs) == 0);
th = ERR_PTR(-ENOMEM);
- OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
+ OBD_ALLOC_GFP(oh, sizeof *oh, __GFP_IO);
if (oh != NULL) {
oh->ot_quota_trans = &oti->oti_quota_trans;
memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
osd_th_alloced(oh);
- memset(oti->oti_declare_ops, 0, OSD_OT_MAX);
- memset(oti->oti_declare_ops_rb, 0, OSD_OT_MAX);
- memset(oti->oti_declare_ops_cred, 0, OSD_OT_MAX);
+ memset(oti->oti_declare_ops, 0,
+ sizeof(oti->oti_declare_ops));
+ memset(oti->oti_declare_ops_rb, 0,
+ sizeof(oti->oti_declare_ops_rb));
+ memset(oti->oti_declare_ops_cred, 0,
+ sizeof(oti->oti_declare_ops_cred));
oti->oti_rollback = false;
}
RETURN(th);
LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
oh->ot_credits,
osd_journal(dev)->j_max_transaction_buffers);
-#ifdef OSD_TRACK_DECLARES
CWARN(" create: %u/%u, delete: %u/%u, destroy: %u/%u\n",
oti->oti_declare_ops[OSD_OT_CREATE],
oti->oti_declare_ops_cred[OSD_OT_CREATE],
last_credits = oh->ot_credits;
last_printed = jiffies;
}
-#endif
/* XXX Limit the credits to 'max_transaction_buffers', and
* let the underlying filesystem to catch the error if
* we really need so many credits.
* XXX temporary stuff. Some abstraction layer should
* be used.
*/
- jh = ldiskfs_journal_start_sb(osd_sb(dev), oh->ot_credits);
+ jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
osd_th_started(oh);
if (!IS_ERR(jh)) {
oh->ot_handle = jh;
lu_context_enter(&th->th_ctx);
lu_device_get(&d->dd_lu_dev);
- oh->ot_dev_link = lu_ref_add(&d->dd_lu_dev.ld_reference,
- "osd-tx", th);
+ lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
+ "osd-tx", th);
oti->oti_txns++;
rc = 0;
} else {
RETURN(rc);
}
+static int osd_seq_exists(const struct lu_env *env,
+ struct osd_device *osd, obd_seq seq)
+{
+ struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
+ struct seq_server_site *ss = osd_seq_site(osd);
+ int rc;
+ ENTRY;
+
+ if (ss == NULL)
+ RETURN(1);
+
+ rc = osd_fld_lookup(env, osd, seq, range);
+ if (rc != 0) {
+ if (rc != -ENOENT)
+ CERROR("%s: can't lookup FLD sequence "LPX64
+ ": rc = %d\n", osd_name(osd), seq, rc);
+ RETURN(0);
+ }
+
+ RETURN(ss->ss_node_id == range->lsr_index);
+}
+
/*
* Concurrency: shouldn't matter.
*/
OBD_FREE_PTR(oh);
}
- /* as we want IO to journal and data IO be concurrent, we don't block
- * awaiting data IO completion in osd_do_bio(), instead we wait here
- * once transaction is submitted to the journal. all reqular requests
- * don't do direct IO (except read/write), thus this wait_event becomes
- * no-op for them.
- *
- * IMPORTANT: we have to wait till any IO submited by the thread is
- * completed otherwise iobuf may be corrupted by different request
- */
- cfs_wait_event(iobuf->dr_wait,
- cfs_atomic_read(&iobuf->dr_numreqs) == 0);
- if (!rc)
- rc = iobuf->dr_error;
+ /* as we want IO to journal and data IO be concurrent, we don't block
+ * awaiting data IO completion in osd_do_bio(), instead we wait here
+ * once transaction is submitted to the journal. all reqular requests
+ * don't do direct IO (except read/write), thus this wait_event becomes
+ * no-op for them.
+ *
+ * IMPORTANT: we have to wait till any IO submited by the thread is
+ * completed otherwise iobuf may be corrupted by different request
+ */
+ wait_event(iobuf->dr_wait,
+ cfs_atomic_read(&iobuf->dr_numreqs) == 0);
+ if (!rc)
+ rc = iobuf->dr_error;
- RETURN(rc);
+ RETURN(rc);
}
static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
/*
* XXX should be taken from not-yet-existing fs abstraction layer.
*/
- param->ddp_mnt = osd_dt_dev(dev)->od_mnt;
param->ddp_max_name_len = LDISKFS_NAME_LEN;
param->ddp_max_nlink = LDISKFS_LINK_MAX;
param->ddp_block_shift = sb->s_blocksize_bits;
static int osd_ro(const struct lu_env *env, struct dt_device *d)
{
- struct super_block *sb = osd_sb(osd_dt_dev(d));
- int rc;
- ENTRY;
+ struct super_block *sb = osd_sb(osd_dt_dev(d));
+ struct block_device *dev = sb->s_bdev;
+#ifdef HAVE_DEV_SET_RDONLY
+ struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
+ int rc = 0;
+#else
+ int rc = -EOPNOTSUPP;
+#endif
+ ENTRY;
+#ifdef HAVE_DEV_SET_RDONLY
CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
- rc = __lvfs_set_rdonly(sb->s_bdev, LDISKFS_SB(sb)->journal_bdev);
- RETURN(rc);
+ if (jdev && (jdev != dev)) {
+ CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
+ (long)jdev);
+ dev_set_rdonly(jdev);
+ }
+ CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
+ dev_set_rdonly(dev);
+#else
+ CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
+ osd_dt_dev(d)->od_svname, (long)dev, rc);
+#endif
+ RETURN(rc);
}
/*
}
int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
- struct lustre_capa *capa, __u64 opc)
+ struct lustre_capa *capa, __u64 opc)
{
- const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
- struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
- struct md_capainfo *ci;
- int rc;
+ const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
+ struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
+ struct lu_capainfo *lci;
+ int rc;
- if (!dev->od_fl_capa)
- return 0;
+ if (!osd->od_fl_capa)
+ return 0;
- if (capa == BYPASS_CAPA)
- return 0;
+ if (capa == BYPASS_CAPA)
+ return 0;
- ci = md_capainfo(env);
- if (unlikely(!ci))
- return 0;
+ lci = lu_capainfo_get(env);
+ if (unlikely(lci == NULL))
+ return 0;
- if (ci->mc_auth == LC_ID_NONE)
- return 0;
+ if (lci->lci_auth == LC_ID_NONE)
+ return 0;
- if (!capa) {
- CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
- return -EACCES;
- }
+ if (capa == NULL) {
+ CERROR("%s: no capability provided for FID "DFID": rc = %d\n",
+ osd_name(osd), PFID(fid), -EACCES);
+ return -EACCES;
+ }
- if (!lu_fid_eq(fid, &capa->lc_fid)) {
- DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
- PFID(fid));
- return -EACCES;
- }
+ if (!lu_fid_eq(fid, &capa->lc_fid)) {
+ DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
+ PFID(fid));
+ return -EACCES;
+ }
- if (!capa_opc_supported(capa, opc)) {
- DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
- return -EACCES;
- }
+ if (!capa_opc_supported(capa, opc)) {
+ DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
+ return -EACCES;
+ }
- if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
- DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
- return -EACCES;
- }
+ rc = capa_is_sane(env, osd, capa, osd->od_capa_keys);
+ if (rc != 0) {
+ DEBUG_CAPA(D_ERROR, capa, "insane: rc = %d", rc);
+ return -EACCES;
+ }
- return 0;
+ return 0;
}
static struct timespec *osd_inode_time(const struct lu_env *env,
spin_unlock(&obj->oo_guard);
if (!rc)
- inode->i_sb->s_op->dirty_inode(inode);
+ ll_dirty_inode(inode, I_DIRTY_DATASYNC);
return rc;
}
}
static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
- cfs_umode_t mode,
- struct dt_allocation_hint *hint,
- struct thandle *th)
+ umode_t mode, struct dt_allocation_hint *hint,
+ struct thandle *th)
{
int result;
struct osd_device *osd = osd_obj2dev(obj);
}
static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
- struct lu_attr *attr,
- struct dt_allocation_hint *hint,
- struct dt_object_format *dof,
- struct thandle *th)
+ struct lu_attr *attr,
+ struct dt_allocation_hint *hint,
+ struct dt_object_format *dof,
+ struct thandle *th)
{
- cfs_umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
- int result;
+ umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
+ int result;
- LINVRNT(osd_invariant(obj));
- LASSERT(obj->oo_inode == NULL);
+ LINVRNT(osd_invariant(obj));
+ LASSERT(obj->oo_inode == NULL);
LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
S_ISFIFO(mode) || S_ISSOCK(mode));
static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
struct dt_object *parent, struct dt_object *child,
- cfs_umode_t child_mode)
+ umode_t child_mode)
{
LASSERT(ah);
* enabled on ldiskfs (lquota takes care of it).
*/
LASSERTF(result == 0, "%d", result);
- inode->i_sb->s_op->dirty_inode(inode);
+ ll_dirty_inode(inode, I_DIRTY_DATASYNC);
}
attr->la_valid = valid;
struct thandle *th)
{
int result;
+ __u32 umask;
+
+ /* we drop umask so that permissions we pass are not affected */
+ umask = current->fs->umask;
+ current->fs->umask = 0;
result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
th);
unlock_new_inode(obj->oo_inode);
}
- return result;
+ /* restore previous umask value */
+ current->fs->umask = umask;
+
+ return result;
}
/**
* \retval 0, on success
*/
static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
- const struct lu_fid *fid, struct thandle *th)
+ const struct lu_fid *fid, struct thandle *th)
{
- struct osd_thread_info *info = osd_oti_get(env);
- struct osd_inode_id *id = &info->oti_id;
- struct osd_device *osd = osd_obj2dev(obj);
+ struct osd_thread_info *info = osd_oti_get(env);
+ struct osd_inode_id *id = &info->oti_id;
+ struct osd_device *osd = osd_obj2dev(obj);
+ struct osd_thandle *oh;
- LASSERT(obj->oo_inode != NULL);
+ LASSERT(obj->oo_inode != NULL);
+
+ oh = container_of0(th, struct osd_thandle, ot_super);
+ LASSERT(oh->ot_handle);
osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
- return osd_oi_insert(info, osd, fid, id, th);
+ return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
}
int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
- const struct lu_fid *fid, struct lu_seq_range *range)
+ obd_seq seq, struct lu_seq_range *range)
{
struct seq_server_site *ss = osd_seq_site(osd);
- int rc;
- if (fid_is_idif(fid)) {
- range->lsr_flags = LU_SEQ_RANGE_OST;
- range->lsr_index = fid_idif_ost_idx(fid);
+ if (fid_seq_is_idif(seq)) {
+ fld_range_set_ost(range);
+ range->lsr_index = idif_ost_idx(seq);
return 0;
}
- if (!fid_seq_in_fldb(fid_seq(fid))) {
- range->lsr_flags = LU_SEQ_RANGE_MDT;
+ if (!fid_seq_in_fldb(seq)) {
+ fld_range_set_mdt(range);
if (ss != NULL)
/* FIXME: If ss is NULL, it suppose not get lsr_index
* at all */
}
LASSERT(ss != NULL);
- range->lsr_flags = -1;
- rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
- if (rc != 0) {
- CERROR("%s can not find "DFID": rc = %d\n",
- osd_name(osd), PFID(fid), rc);
- }
- return rc;
+ fld_range_set_any(range);
+ /* OSD will only do local fld lookup */
+ return fld_local_lookup(env, ss->ss_server_fld, seq, range);
}
/*
struct dt_object_format *dof,
struct thandle *handle)
{
- struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
struct osd_thandle *oh;
int rc;
ENTRY;
osd_trans_declare_op(env, oh, OSD_OT_CREATE,
osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
if (!fid_is_on_ost(osd_oti_get(env), osd_dt_dev(handle->th_dev),
- lu_object_fid(&dt->do_lu)))
+ lu_object_fid(&dt->do_lu), OI_CHECK_FLD))
/* Reuse idle OI block may cause additional one OI block
* to be changed. */
osd_trans_declare_op(env, oh, OSD_OT_INSERT,
if (rc != 0)
RETURN(rc);
- /* It does fld look up inside declare, and the result will be
- * added to fld cache, so the following fld lookup inside insert
- * does not need send RPC anymore, so avoid send rpc with holding
- * transaction */
- if (fid_is_norm(lu_object_fid(&dt->do_lu)) &&
- !fid_is_last_id(lu_object_fid(&dt->do_lu)))
- osd_fld_lookup(env, osd_dt_dev(handle->th_dev),
- lu_object_fid(&dt->do_lu), range);
-
-
RETURN(rc);
}
LASSERT(oh->ot_handle == NULL);
LASSERT(inode);
- osd_trans_declare_op(env, oh, OSD_OT_DELETE,
+ osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
/* Recycle idle OI leaf may cause additional three OI blocks
* to be changed. */
- osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
+ osd_trans_declare_op(env, oh, OSD_OT_DELETE,
osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
-
/* one less inode */
rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
false, true, NULL, false);
if (unlikely(fid_is_acct(fid)))
RETURN(-EPERM);
- /* Parallel control for OI scrub. For most of cases, there is no
- * lock contention. So it will not affect unlink performance. */
- mutex_lock(&inode->i_mutex);
if (S_ISDIR(inode->i_mode)) {
- LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1);
- /* it will check/delete the agent inode for every dir
- * destory, how to optimize it? unlink performance
- * impaction XXX */
- result = osd_delete_from_agent(env, osd, obj, oh);
+ LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
+ inode->i_nlink == 2);
+ /* it will check/delete the inode from remote parent,
+ * how to optimize it? unlink performance impaction XXX */
+ result = osd_delete_from_remote_parent(env, osd, obj, oh);
if (result != 0 && result != -ENOENT) {
- CERROR("%s: delete agent inode "DFID": rc = %d\n",
+ CERROR("%s: delete inode "DFID": rc = %d\n",
osd_name(osd), PFID(fid), result);
}
spin_lock(&obj->oo_guard);
clear_nlink(inode);
spin_unlock(&obj->oo_guard);
- inode->i_sb->s_op->dirty_inode(inode);
+ ll_dirty_inode(inode, I_DIRTY_DATASYNC);
}
osd_trans_exec_op(env, th, OSD_OT_DESTROY);
- result = osd_oi_delete(osd_oti_get(env), osd, fid, th);
- mutex_unlock(&inode->i_mutex);
+ result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
+ OI_CHECK_FLD);
/* XXX: add to ext3 orphan list */
/* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
RETURN(0);
}
-static inline int __osd_xattr_set(struct osd_thread_info *info,
- struct inode *inode, const char *name,
- const void *buf, int buflen, int fl)
-{
- struct dentry *dentry = &info->oti_child_dentry;
-
- ll_vfs_dq_init(inode);
- dentry->d_inode = inode;
- return inode->i_op->setxattr(dentry, name, buf, buflen, fl);
-}
-
/**
* Put the fid into lustre_mdt_attrs, and then place the structure
* inode's ea. This fid should not be altered during the life time
* FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
*/
int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
- const struct lu_fid *fid)
+ const struct lu_fid *fid, __u32 compat, __u32 incompat)
{
struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
int rc;
+ ENTRY;
if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
- return 0;
-
- if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF) && fid_is_client_visible(fid))
- return 0;
+ RETURN(0);
- lustre_lma_init(lma, fid);
+ lustre_lma_init(lma, fid, compat, incompat);
lustre_lma_swab(lma);
rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
XATTR_CREATE);
- /* Someone may created the EA by race. */
- if (unlikely(rc == -EEXIST))
- rc = 0;
- return rc;
+ /* LMA may already exist, but we need to check that all the
+ * desired compat/incompat flags have been added. */
+ if (unlikely(rc == -EEXIST)) {
+ if (compat == 0 && incompat == 0)
+ RETURN(0);
+
+ rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
+ XATTR_NAME_LMA, info->oti_mdt_attrs_old,
+ LMA_OLD_SIZE);
+ if (rc <= 0)
+ RETURN(-EINVAL);
+
+ lustre_lma_swab(lma);
+ if (!(~lma->lma_compat & compat) &&
+ !(~lma->lma_incompat & incompat))
+ RETURN(0);
+
+ lma->lma_compat |= compat;
+ lma->lma_incompat |= incompat;
+ lustre_lma_swab(lma);
+ rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
+ sizeof(*lma), XATTR_REPLACE);
+ }
+
+ RETURN(rc);
}
/**
void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
const struct dt_rec *fid)
{
- if (!fid_is_client_mdt_visible((const struct lu_fid *)fid)) {
+ if (!fid_is_namespace_visible((const struct lu_fid *)fid) ||
+ OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
param->edp_magic = 0;
return;
}
}
/**
- * Create an local inode for remote entry
+ * Create an local agent inode for remote entry
*/
-static struct inode *osd_create_remote_inode(const struct lu_env *env,
- struct osd_device *osd,
- struct osd_object *pobj,
- const struct lu_fid *fid,
- struct thandle *th)
+static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
+ struct osd_device *osd,
+ struct osd_object *pobj,
+ const struct lu_fid *fid,
+ struct thandle *th)
{
struct osd_thread_info *info = osd_oti_get(env);
struct inode *local;
RETURN(local);
}
+ /* Set special LMA flag for local agent inode */
+ rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
+ if (rc != 0) {
+ CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
+ osd_name(osd), PFID(fid), rc);
+ RETURN(ERR_PTR(rc));
+ }
+
rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
(const struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
(const struct dt_rec *)fid, oh);
}
/**
- * Delete local inode for remote entry
+ * Delete local agent inode for remote entry
*/
-static int osd_delete_remote_inode(const struct lu_env *env,
- struct osd_device *osd,
- const struct lu_fid *fid,
- __u32 ino, struct osd_thandle *oh)
+static int osd_delete_local_agent_inode(const struct lu_env *env,
+ struct osd_device *osd,
+ const struct lu_fid *fid,
+ __u32 ino, struct osd_thandle *oh)
{
struct osd_thread_info *oti = osd_oti_get(env);
struct osd_inode_id *id = &oti->oti_id;
osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
result = __osd_object_create(info, obj, attr, hint, dof, th);
- if ((result == 0) &&
- (fid_is_last_id(fid) ||
- !fid_is_on_ost(info, osd_dt_dev(th->th_dev), fid)))
- result = osd_ea_fid_set(info, obj->oo_inode, fid);
+ if (result == 0)
+ result = osd_ea_fid_set(info, obj->oo_inode, fid,
+ fid_is_on_ost(info, osd_obj2dev(obj),
+ fid, OI_CHECK_FLD) ?
+ LMAC_FID_ON_OST : 0, 0);
- if (result == 0)
- result = __osd_oi_insert(env, obj, fid, th);
+ if (result == 0)
+ result = __osd_oi_insert(env, obj, fid, th);
LASSERT(ergo(result == 0,
dt_object_exists(dt) && !dt_object_remote(dt)));
static int osd_object_ref_add(const struct lu_env *env,
struct dt_object *dt, struct thandle *th)
{
- struct osd_object *obj = osd_dt_obj(dt);
- struct inode *inode = obj->oo_inode;
+ struct osd_object *obj = osd_dt_obj(dt);
+ struct inode *inode = obj->oo_inode;
+ struct osd_thandle *oh;
+ int rc = 0;
LINVRNT(osd_invariant(obj));
LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
LASSERT(osd_write_locked(env, obj));
LASSERT(th != NULL);
+ oh = container_of0(th, struct osd_thandle, ot_super);
+ LASSERT(oh->ot_handle != NULL);
+
osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
/*
- * DIR_NLINK feature is set for compatibility reasons if:
- * 1) nlinks > LDISKFS_LINK_MAX, or
- * 2) nlinks == 2, since this indicates i_nlink was previously 1.
+ * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
+ * (65000) subdirectories by storing "1" in i_nlink if the link count
+ * would otherwise overflow. Directory tranversal tools understand
+ * that (st_nlink == 1) indicates that the filesystem dose not track
+ * hard links count on the directory, and will not abort subdirectory
+ * scanning early once (st_nlink - 2) subdirs have been found.
*
- * It is easier to always set this flag (rather than check and set),
- * since it has less overhead, and the superblock will be dirtied
- * at some point. Both e2fsprogs and any Lustre-supported ldiskfs
- * do not actually care whether this flag is set or not.
+ * This also has to properly handle the case of inodes with nlink == 0
+ * in case they are being linked into the PENDING directory
*/
spin_lock(&obj->oo_guard);
- /* inc_nlink from 0 may cause WARN_ON */
- if(inode->i_nlink == 0)
- set_nlink(inode, 1);
- else
- inc_nlink(inode);
- if (S_ISDIR(inode->i_mode) && inode->i_nlink > 1) {
- if (inode->i_nlink >= LDISKFS_LINK_MAX ||
- inode->i_nlink == 2)
- set_nlink(inode, 1);
- }
+ ldiskfs_inc_count(oh->ot_handle, inode);
LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
spin_unlock(&obj->oo_guard);
- inode->i_sb->s_op->dirty_inode(inode);
+
+ ll_dirty_inode(inode, I_DIRTY_DATASYNC);
LINVRNT(osd_invariant(obj));
- return 0;
+ return rc;
}
static int osd_declare_object_ref_del(const struct lu_env *env,
static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
struct thandle *th)
{
- struct osd_object *obj = osd_dt_obj(dt);
- struct inode *inode = obj->oo_inode;
+ struct osd_object *obj = osd_dt_obj(dt);
+ struct inode *inode = obj->oo_inode;
+ struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
+ struct osd_thandle *oh;
- LINVRNT(osd_invariant(obj));
+ LINVRNT(osd_invariant(obj));
LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
- LASSERT(osd_write_locked(env, obj));
- LASSERT(th != NULL);
+ LASSERT(osd_write_locked(env, obj));
+ LASSERT(th != NULL);
+
+ oh = container_of0(th, struct osd_thandle, ot_super);
+ LASSERT(oh->ot_handle != NULL);
osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
spin_lock(&obj->oo_guard);
- LASSERT(inode->i_nlink > 0);
- drop_nlink(inode);
- /* If this is/was a many-subdir directory (nlink > LDISKFS_LINK_MAX)
- * then the nlink count is 1. Don't let it be set to 0 or the directory
- * inode will be deleted incorrectly. */
- if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
- set_nlink(inode, 1);
+ /* That can be result of upgrade from old Lustre version and
+ * applied only to local files. Just skip this ref_del call.
+ * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
+ if (inode->i_nlink == 0) {
+ CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
+ D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
+ ", maybe an upgraded file? (LU-3915)\n",
+ osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
+ spin_unlock(&obj->oo_guard);
+ return 0;
+ }
+
+ ldiskfs_dec_count(oh->ot_handle, inode);
spin_unlock(&obj->oo_guard);
- inode->i_sb->s_op->dirty_inode(inode);
+
+ ll_dirty_inode(inode, I_DIRTY_DATASYNC);
LINVRNT(osd_invariant(obj));
return 0;
LDISKFS_I(inode)->i_fs_version = *new_version;
/** Version is set after all inode operations are finished,
* so we should mark it dirty here */
- inode->i_sb->s_op->dirty_inode(inode);
+ ll_dirty_inode(inode, I_DIRTY_DATASYNC);
}
/*
struct inode *inode = obj->oo_inode;
struct osd_thread_info *info = osd_oti_get(env);
int fs_flags = 0;
+ ENTRY;
LASSERT(handle != NULL);
return -EACCES;
dentry->d_inode = inode;
+ dentry->d_sb = inode->i_sb;
return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
}
ll_vfs_dq_init(inode);
dentry->d_inode = inode;
+ dentry->d_sb = inode->i_sb;
rc = inode->i_op->removexattr(dentry, name);
return rc;
}
static struct obd_capa *osd_capa_get(const struct lu_env *env,
- struct dt_object *dt,
- struct lustre_capa *old,
- __u64 opc)
+ struct dt_object *dt,
+ struct lustre_capa *old, __u64 opc)
{
- struct osd_thread_info *info = osd_oti_get(env);
- const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
- struct osd_object *obj = osd_dt_obj(dt);
- struct osd_device *dev = osd_obj2dev(obj);
- struct lustre_capa_key *key = &info->oti_capa_key;
- struct lustre_capa *capa = &info->oti_capa;
- struct obd_capa *oc;
- struct md_capainfo *ci;
- int rc;
- ENTRY;
+ struct osd_thread_info *info = osd_oti_get(env);
+ const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
+ struct osd_object *obj = osd_dt_obj(dt);
+ struct osd_device *osd = osd_obj2dev(obj);
+ struct lustre_capa_key *key = &info->oti_capa_key;
+ struct lustre_capa *capa = &info->oti_capa;
+ struct obd_capa *oc;
+ struct lu_capainfo *lci;
+ int rc;
+ ENTRY;
- if (!dev->od_fl_capa)
- RETURN(ERR_PTR(-ENOENT));
+ if (!osd->od_fl_capa)
+ RETURN(ERR_PTR(-ENOENT));
LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
- LINVRNT(osd_invariant(obj));
+ LINVRNT(osd_invariant(obj));
- /* renewal sanity check */
- if (old && osd_object_auth(env, dt, old, opc))
- RETURN(ERR_PTR(-EACCES));
-
- ci = md_capainfo(env);
- if (unlikely(!ci))
- RETURN(ERR_PTR(-ENOENT));
-
- switch (ci->mc_auth) {
- case LC_ID_NONE:
- RETURN(NULL);
- case LC_ID_PLAIN:
- capa->lc_uid = obj->oo_inode->i_uid;
- capa->lc_gid = obj->oo_inode->i_gid;
- capa->lc_flags = LC_ID_PLAIN;
- break;
- case LC_ID_CONVERT: {
- __u32 d[4], s[4];
-
- s[0] = obj->oo_inode->i_uid;
- cfs_get_random_bytes(&(s[1]), sizeof(__u32));
- s[2] = obj->oo_inode->i_gid;
- cfs_get_random_bytes(&(s[3]), sizeof(__u32));
- rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
- if (unlikely(rc))
- RETURN(ERR_PTR(rc));
-
- capa->lc_uid = ((__u64)d[1] << 32) | d[0];
- capa->lc_gid = ((__u64)d[3] << 32) | d[2];
- capa->lc_flags = LC_ID_CONVERT;
- break;
- }
- default:
- RETURN(ERR_PTR(-EINVAL));
+ /* renewal sanity check */
+ if (old && osd_object_auth(env, dt, old, opc))
+ RETURN(ERR_PTR(-EACCES));
+
+ lci = lu_capainfo_get(env);
+ if (unlikely(lci == NULL))
+ RETURN(ERR_PTR(-ENOENT));
+
+ switch (lci->lci_auth) {
+ case LC_ID_NONE:
+ RETURN(NULL);
+ case LC_ID_PLAIN:
+ capa->lc_uid = obj->oo_inode->i_uid;
+ capa->lc_gid = obj->oo_inode->i_gid;
+ capa->lc_flags = LC_ID_PLAIN;
+ break;
+ case LC_ID_CONVERT: {
+ __u32 d[4], s[4];
+
+ s[0] = obj->oo_inode->i_uid;
+ cfs_get_random_bytes(&(s[1]), sizeof(__u32));
+ s[2] = obj->oo_inode->i_gid;
+ cfs_get_random_bytes(&(s[3]), sizeof(__u32));
+ rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
+ if (unlikely(rc))
+ RETURN(ERR_PTR(rc));
+
+ capa->lc_uid = ((__u64)d[1] << 32) | d[0];
+ capa->lc_gid = ((__u64)d[3] << 32) | d[2];
+ capa->lc_flags = LC_ID_CONVERT;
+ break;
}
+ default:
+ RETURN(ERR_PTR(-EINVAL));
+ }
- capa->lc_fid = *fid;
- capa->lc_opc = opc;
- capa->lc_flags |= dev->od_capa_alg << 24;
- capa->lc_timeout = dev->od_capa_timeout;
- capa->lc_expiry = 0;
+ capa->lc_fid = *fid;
+ capa->lc_opc = opc;
+ capa->lc_flags |= osd->od_capa_alg << 24;
+ capa->lc_timeout = osd->od_capa_timeout;
+ capa->lc_expiry = 0;
- oc = capa_lookup(dev->od_capa_hash, capa, 1);
- if (oc) {
- LASSERT(!capa_is_expired(oc));
- RETURN(oc);
- }
+ oc = capa_lookup(osd->od_capa_hash, capa, 1);
+ if (oc) {
+ LASSERT(!capa_is_expired(oc));
+ RETURN(oc);
+ }
spin_lock(&capa_lock);
- *key = dev->od_capa_keys[1];
+ *key = osd->od_capa_keys[1];
spin_unlock(&capa_lock);
- capa->lc_keyid = key->lk_keyid;
- capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
+ capa->lc_keyid = key->lk_keyid;
+ capa->lc_expiry = cfs_time_current_sec() + osd->od_capa_timeout;
- rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
- if (rc) {
- DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
- RETURN(ERR_PTR(rc));
- }
+ rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
+ if (rc) {
+ DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
+ RETURN(ERR_PTR(rc));
+ }
- oc = capa_add(dev->od_capa_hash, capa);
- RETURN(oc);
+ oc = capa_add(osd->od_capa_hash, capa);
+ RETURN(oc);
}
static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
ENTRY;
dentry->d_inode = inode;
+ dentry->d_sb = inode->i_sb;
file->f_dentry = dentry;
file->f_mapping = inode->i_mapping;
file->f_op = inode->i_fop;
+ set_file_inode(file, inode);
+#ifndef HAVE_FILE_FSYNC_4ARGS
mutex_lock(&inode->i_mutex);
- rc = file->f_op->fsync(file, dentry, 0);
+#endif
+ rc = do_fsync(file, 0);
+#ifndef HAVE_FILE_FSYNC_4ARGS
mutex_unlock(&inode->i_mutex);
+#endif
RETURN(rc);
}
}
LINVRNT(osd_invariant(obj));
- if (is_quota_glb_feat(feat))
+ if (result == 0 && is_quota_glb_feat(feat) &&
+ fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
result = osd_quota_migration(env, dt, feat);
return result;
static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
struct lu_fid *fid)
{
- struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
- struct seq_server_site *ss = osd_seq_site(osd);
- int rc;
ENTRY;
- /* Those FID seqs, which are not in FLDB, must be local seq */
- if (unlikely(!fid_seq_in_fldb(fid_seq(fid)) || ss == NULL))
+ /* FID seqs not in FLDB, must be local seq */
+ if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
RETURN(0);
- rc = osd_fld_lookup(env, osd, fid, range);
- if (rc != 0) {
- CERROR("%s: Can not lookup fld for "DFID"\n",
- osd_name(osd), PFID(fid));
- RETURN(rc);
- }
+ if (osd_seq_exists(env, osd, fid_seq(fid)))
+ RETURN(0);
- RETURN(ss->ss_node_id != range->lsr_index);
+ RETURN(1);
}
/**
down_write(&obj->oo_ext_idx_sem);
}
- bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
+ bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
if (bh) {
- rc = ldiskfs_delete_entry(oh->ot_handle,
- dir, de, bh);
+ __u32 ino = 0;
+
+ /* If this is not the ".." entry, it might be a remote DNE
+ * entry and we need to check if the FID is for a remote
+ * MDT. If the FID is not in the directory entry (e.g.
+ * upgraded 1.8 filesystem without dirdata enabled) then
+ * we need to get the FID from the LMA. For a remote directory
+ * there HAS to be an LMA, it cannot be an IGIF inode in this
+ * case.
+ *
+ * Delete the entry before the agent inode in order to
+ * simplify error handling. At worst an error after deleting
+ * the entry first might leak the agent inode afterward. The
+ * reverse would need filesystem abort in case of error deleting
+ * the entry after the agent had been removed, or leave a
+ * dangling entry pointing at a random inode. */
+ if (strcmp((char *)key, dotdot) != 0) {
+ LASSERT(de != NULL);
+ rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
+ /* If Fid is not in dentry, try to get it from LMA */
+ if (rc == -ENODATA) {
+ struct osd_inode_id *id;
+ struct inode *inode;
+
+ /* Before trying to get fid from the inode,
+ * check whether the inode is valid.
+ *
+ * If the inode has been deleted, do not go
+ * ahead to do osd_ea_fid_get, which will set
+ * the inode to bad inode, which might cause
+ * the inode to be deleted uncorrectly */
+ inode = ldiskfs_iget(osd_sb(osd),
+ le32_to_cpu(de->inode));
+ if (IS_ERR(inode)) {
+ CDEBUG(D_INODE, "%s: "DFID"get inode"
+ "error.\n", osd_name(osd),
+ PFID(fid));
+ rc = PTR_ERR(inode);
+ } else {
+ if (likely(inode->i_nlink != 0)) {
+ id = &osd_oti_get(env)->oti_id;
+ rc = osd_ea_fid_get(env, obj,
+ le32_to_cpu(de->inode),
+ fid, id);
+ } else {
+ CDEBUG(D_INFO, "%s: %u "DFID
+ "deleted.\n",
+ osd_name(osd),
+ le32_to_cpu(de->inode),
+ PFID(fid));
+ rc = -ESTALE;
+ }
+ iput(inode);
+ }
+ }
+ if (rc == 0 &&
+ unlikely(osd_remote_fid(env, osd, fid)))
+ /* Need to delete agent inode */
+ ino = le32_to_cpu(de->inode);
+ }
+ rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
brelse(bh);
+ if (rc == 0 && unlikely(ino != 0)) {
+ rc = osd_delete_local_agent_inode(env, osd, fid, ino,
+ oh);
+ if (rc != 0)
+ CERROR("%s: del local inode "DFID": rc = %d\n",
+ osd_name(osd), PFID(fid), rc);
+ }
} else {
rc = -ENOENT;
}
GOTO(out, rc);
/* For inode on the remote MDT, .. will point to
- * /Agent directory. So do not try to lookup/delete
- * remote inode for .. */
- if (strcmp((char *)key, dotdot) == 0)
- GOTO(out, rc = 0);
-
- LASSERT(de != NULL);
- rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
- if (rc == 0 && osd_remote_fid(env, osd, fid)) {
- __u32 ino = le32_to_cpu(de->inode);
+ * /Agent directory, Check whether it needs to delete
+ * from agent directory */
+ if (unlikely(strcmp((char *)key, dotdot) == 0)) {
+ rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
+ oh);
+ if (rc != 0 && rc != -ENOENT) {
+ CERROR("%s: delete agent inode "DFID": rc = %d\n",
+ osd_name(osd), PFID(fid), rc);
+ }
- rc = osd_delete_remote_inode(env, osd, fid, ino, oh);
- if (rc != 0)
- CERROR("%s: del local inode "DFID": rc = %d\n",
- osd_name(osd), PFID(fid), rc);
- } else {
- if (rc == -ENODATA)
+ if (rc == -ENOENT)
rc = 0;
+
+ GOTO(out, rc);
}
out:
RETURN_EXIT;
again:
- rc = osd_oi_lookup(oti, dev, fid, id, true);
+ rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD);
if (rc != 0 && rc != -ENOENT)
RETURN_EXIT;
return rc;
}
+int osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
+ struct osd_inode_id *id, const struct lu_fid *fid)
+{
+ CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
+ id->oii_ino, id->oii_gen, info);
+ info->oti_cache.oic_lid = *id;
+ info->oti_cache.oic_fid = *fid;
+ info->oti_cache.oic_dev = osd;
+
+ return 0;
+}
+
/**
* Calls ->lookup() to find dentry. From dentry get inode and
* read inode's ea to get fid. This is required for interoperability
down_read(&obj->oo_ext_idx_sem);
}
- bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
+ bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
if (bh) {
struct osd_thread_info *oti = osd_oti_get(env);
struct osd_inode_id *id = &oti->oti_id;
rc = osd_ea_fid_get(env, obj, ino, fid, id);
else
osd_id_gen(id, ino, OSD_OII_NOGEN);
- if (rc != 0 || osd_remote_fid(env, dev, fid)) {
+ if (rc != 0) {
fid_zero(&oic->oic_fid);
GOTO(out, rc);
}
- oic->oic_lid = *id;
- oic->oic_fid = *fid;
+ rc = osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id,
+ fid);
+ if (rc != 0)
+ GOTO(out, rc);
if ((scrub->os_pos_current <= ino) &&
((sf->sf_flags & SF_INCONSISTENT) ||
(sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
lu_object_put(env, luch);
child = ERR_PTR(-ENOENT);
}
- } else
- child = (void *)luch;
+ } else {
+ child = ERR_CAST(luch);
+ }
- return child;
+ return child;
}
/**
/* If parent on remote MDT, we need put this object
* under AGENT */
oh = container_of(th, typeof(*oh), ot_super);
- rc = osd_add_to_agent(env, osd, obj, oh);
+ rc = osd_add_to_remote_parent(env, osd, obj, oh);
if (rc != 0) {
- CERROR("%s: add agent "DFID" error: rc = %d\n",
+ CERROR("%s: add "DFID" error: rc = %d\n",
osd_name(osd),
PFID(lu_object_fid(&dt->do_lu)), rc);
RETURN(rc);
}
- child_inode = igrab(omm->omm_agent_dentry->d_inode);
+ child_inode = igrab(omm->omm_remote_parent->d_inode);
} else {
- child_inode = osd_create_remote_inode(env, osd, obj,
- fid, th);
+ child_inode = osd_create_local_agent_inode(env, osd,
+ obj, fid,
+ th);
if (IS_ERR(child_inode))
RETURN(PTR_ERR(child_inode));
}
CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
osd_name(osd), PFID(fid),
id->oii_ino, id->oii_gen,
- (int)PTR_ERR(child_inode));
- RETURN(PTR_ERR(child_inode));
+ (int)PTR_ERR(child));
+ RETURN(PTR_ERR(child));
}
child_inode = igrab(child->oo_inode);
}
{
/* check if file type is required */
if (ent->lde_attrs & LUDA_TYPE) {
- int align = sizeof(struct luda_type) - 1;
struct luda_type *lt;
+ int align = sizeof(*lt) - 1;
len = (len + align) & ~align;
lt = (struct luda_type *)(ent->lde_name + len);
- lt->lt_type = cpu_to_le16(CFS_DTTOIF(type));
+ lt->lt_type = cpu_to_le16(DTTOIF(type));
}
ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
strncpy(ent->lde_name, name, namelen);
+ ent->lde_name[namelen] = '\0';
ent->lde_namelen = cpu_to_le16(namelen);
/* append lustre attributes */
struct osd_object *obj = osd_dt_obj(dt);
struct osd_thread_info *info = osd_oti_get(env);
struct osd_it_ea *it = &info->oti_it_ea;
+ struct file *file = &it->oie_file;
struct lu_object *lo = &dt->do_lu;
struct dentry *obj_dentry = &info->oti_it_dentry;
ENTRY;
it->oie_dirent = NULL;
it->oie_buf = info->oti_it_ea_buf;
it->oie_obj = obj;
- it->oie_file.f_pos = 0;
- it->oie_file.f_dentry = obj_dentry;
- if (attr & LUDA_64BITHASH)
- it->oie_file.f_mode |= FMODE_64BITHASH;
- else
- it->oie_file.f_mode |= FMODE_32BITHASH;
- it->oie_file.f_mapping = obj->oo_inode->i_mapping;
- it->oie_file.f_op = obj->oo_inode->i_fop;
- it->oie_file.private_data = NULL;
- lu_object_get(lo);
- RETURN((struct dt_it *) it);
+
+ /* Reset the "file" totally to avoid to reuse any old value from
+ * former readdir handling, the "file->f_pos" should be zero. */
+ memset(file, 0, sizeof(*file));
+ /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
+ if (attr & LUDA_64BITHASH)
+ file->f_mode = FMODE_64BITHASH;
+ else
+ file->f_mode = FMODE_32BITHASH;
+ file->f_dentry = obj_dentry;
+ file->f_mapping = obj->oo_inode->i_mapping;
+ file->f_op = obj->oo_inode->i_fop;
+ set_file_inode(file, obj->oo_inode);
+
+ lu_object_get(lo);
+ RETURN((struct dt_it *) it);
}
/**
return 0;
}
+static inline int
+osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
+{
+ LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
+ "dot_dotdot = %d\n", dot_dotdot);
+
+ if (LDISKFS_DIR_REC_LEN(de) >=
+ __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
+ return 1;
+ else
+ return 0;
+}
+
static int
osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
struct inode *dir, struct inode *inode,
struct inode *inode;
int credits;
int rc;
+ int dot_dotdot = 0;
bool dirty = false;
- bool is_dotdot = false;
ENTRY;
if (ent->oied_name[0] == '.') {
- /* Skip dot entry, even if it has stale FID-in-dirent, because
- * we do not use such FID-in-dirent anymore, it is harmless. */
if (ent->oied_namelen == 1)
- RETURN(0);
-
- if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
- is_dotdot = true;
+ dot_dotdot = 1;
+ else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
+ dot_dotdot = 2;
}
dentry = osd_child_dentry_get(env, obj, ent->oied_name,
again:
if (dev->od_dirent_journal) {
- jh = ldiskfs_journal_start_sb(sb, credits);
+ jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
if (IS_ERR(jh)) {
rc = PTR_ERR(jh);
CERROR("%.16s: fail to start trans for dirent "
ent->oied_name, rc);
RETURN(rc);
}
- }
- if (obj->oo_hl_head != NULL) {
- hlock = osd_oti_get(env)->oti_hlock;
- ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
- LDISKFS_HLOCK_DEL);
+ if (obj->oo_hl_head != NULL) {
+ hlock = osd_oti_get(env)->oti_hlock;
+ /* "0" means exclusive lock for the whole directory.
+ * We need to prevent others access such name entry
+ * during the delete + insert. Neither HLOCK_ADD nor
+ * HLOCK_DEL cannot guarantee the atomicity. */
+ ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
+ } else {
+ down_write(&obj->oo_ext_idx_sem);
+ }
} else {
- down_write(&obj->oo_ext_idx_sem);
+ if (obj->oo_hl_head != NULL) {
+ hlock = osd_oti_get(env)->oti_hlock;
+ ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
+ LDISKFS_HLOCK_LOOKUP);
+ } else {
+ down_read(&obj->oo_ext_idx_sem);
+ }
}
- bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
- /* For dotdot entry, if there is not enough space to hold FID-in-dirent,
- * just keep it there. It only happens when the device upgraded from 1.8
- * or restored from MDT file-level backup. For the whole directory, only
- * dotdot entry has no FID-in-dirent and needs to get FID from LMA when
- * readdir, it will not affect the performance much. */
+ bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
+ /* For dot/dotdot entry, if there is not enough space to hold the
+ * FID-in-dirent, just keep them there. It only happens when the
+ * device upgraded from 1.8 or restored from MDT file-level backup.
+ * For the whole directory, only dot/dotdot entry have no FID-in-dirent
+ * and needs to get FID from LMA when readdir, it will not affect the
+ * performance much. */
if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
- (is_dotdot && !osd_dirent_has_space(de->rec_len,
- ent->oied_namelen,
- sb->s_blocksize))) {
+ (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
*attr |= LUDA_IGNORE;
GOTO(out_journal, rc = 0);
}
GOTO(out_journal, rc);
}
+ /* skip the REMOTE_PARENT_DIR. */
+ if (inode == dev->od_mdt_map->omm_remote_parent->d_inode)
+ GOTO(out_inode, rc = 0);
+
rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
if (rc == 0) {
+ LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI));
+
if (fid_is_sane(fid)) {
/* FID-in-dirent is valid. */
if (lu_fid_eq(fid, &lma->lma_self_fid))
if (hlock != NULL)
ldiskfs_htree_unlock(hlock);
else
- up_write(&obj->oo_ext_idx_sem);
+ up_read(&obj->oo_ext_idx_sem);
dev->od_dirent_journal = 1;
goto again;
}
} else {
/* Do not repair under dryrun mode. */
if (*attr & LUDA_VERIFY_DRYRUN) {
+ *fid = lma->lma_self_fid;
*attr |= LUDA_REPAIR;
GOTO(out_inode, rc = 0);
}
if (hlock != NULL)
ldiskfs_htree_unlock(hlock);
else
- up_write(&obj->oo_ext_idx_sem);
+ up_read(&obj->oo_ext_idx_sem);
dev->od_dirent_journal = 1;
goto again;
}
} else if (rc == -ENODATA) {
/* Do not repair under dryrun mode. */
if (*attr & LUDA_VERIFY_DRYRUN) {
- if (fid_is_sane(fid))
+ if (fid_is_sane(fid)) {
*attr |= LUDA_REPAIR;
- else
+ } else {
+ lu_igif_build(fid, inode->i_ino,
+ inode->i_generation);
*attr |= LUDA_UPGRADE;
+ }
GOTO(out_inode, rc = 0);
}
if (hlock != NULL)
ldiskfs_htree_unlock(hlock);
else
- up_write(&obj->oo_ext_idx_sem);
+ up_read(&obj->oo_ext_idx_sem);
dev->od_dirent_journal = 1;
goto again;
}
if (unlikely(fid_is_sane(fid))) {
/* FID-in-dirent exists, but FID-in-LMA is lost.
* Trust the FID-in-dirent, and add FID-in-LMA. */
- rc = osd_ea_fid_set(info, inode, fid);
+ rc = osd_ea_fid_set(info, inode, fid, 0, 0);
if (rc == 0)
*attr |= LUDA_REPAIR;
} else {
out_journal:
brelse(bh);
- if (hlock != NULL)
+ if (hlock != NULL) {
ldiskfs_htree_unlock(hlock);
- else
- up_write(&obj->oo_ext_idx_sem);
+ } else {
+ if (dev->od_dirent_journal)
+ up_write(&obj->oo_ext_idx_sem);
+ else
+ up_read(&obj->oo_ext_idx_sem);
+ }
if (jh != NULL)
ldiskfs_journal_stop(jh);
if (rc >= 0 && !dirty)
if (unlikely(ino == osd_sb(dev)->s_root->d_inode->i_ino)) {
attr |= LUDA_IGNORE;
rc = 0;
- goto pack;
+ } else {
+ rc = osd_dirent_check_repair(env, obj, it, fid, id,
+ &attr);
}
-
- rc = osd_dirent_check_repair(env, obj, it, fid, id, &attr);
} else {
attr &= ~LU_DIRENT_ATTRS_MASK;
if (!fid_is_sane(fid)) {
}
}
- if (rc < 0)
- RETURN(rc);
-
-pack:
+ /* Pack the entry anyway, at least the offset is right. */
osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
it->oie_dirent->oied_name,
it->oie_dirent->oied_namelen,
it->oie_dirent->oied_type, attr);
+ if (rc < 0)
+ RETURN(rc);
+
if (osd_remote_fid(env, dev, fid))
RETURN(0);
- if (likely(!(attr & LUDA_IGNORE))) {
- oic->oic_lid = *id;
- oic->oic_fid = *fid;
- }
+ if (likely(!(attr & LUDA_IGNORE)))
+ rc = osd_add_oi_cache(oti, dev, id, fid);
if (!(attr & LUDA_VERIFY) &&
(scrub->os_pos_current <= ino) &&
static void osd_key_fini(const struct lu_context *ctx,
struct lu_context_key *key, void* data)
{
- struct osd_thread_info *info = data;
+ struct osd_thread_info *info = data;
- if (info->oti_hlock != NULL)
- ldiskfs_htree_lock_free(info->oti_hlock);
- OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
- OBD_FREE_PTR(info);
+ if (info->oti_hlock != NULL)
+ ldiskfs_htree_lock_free(info->oti_hlock);
+ OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
+ lu_buf_free(&info->oti_iobuf.dr_pg_buf);
+ lu_buf_free(&info->oti_iobuf.dr_bl_buf);
+ OBD_FREE_PTR(info);
}
static void osd_key_exit(const struct lu_context *ctx,
{
struct osd_device *osd = osd_dev(d);
- strncpy(osd->od_svname, name, MAX_OBD_NAME);
+ if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
+ >= sizeof(osd->od_svname))
+ return -E2BIG;
return osd_procfs_init(osd, name);
}
{
ENTRY;
- osd_scrub_cleanup(env, o);
-
- if (o->od_fsops) {
- fsfilt_put_ops(o->od_fsops);
- o->od_fsops = NULL;
- }
-
/* shutdown quota slave instance associated with the device */
if (o->od_quota_slave != NULL) {
qsd_fini(env, o->od_quota_slave);
RETURN(0);
}
+static void osd_umount(const struct lu_env *env, struct osd_device *o)
+{
+ ENTRY;
+
+ if (o->od_mnt != NULL) {
+ shrink_dcache_sb(osd_sb(o));
+ osd_sync(env, &o->od_dt_dev);
+
+ mntput(o->od_mnt);
+ o->od_mnt = NULL;
+ }
+
+ EXIT;
+}
+
static int osd_mount(const struct lu_env *env,
struct osd_device *o, struct lustre_cfg *cfg)
{
struct file_system_type *type;
char *options = NULL;
char *str;
- int rc = 0;
+ struct osd_thread_info *info = osd_oti_get(env);
+ struct lu_fid *fid = &info->oti_fid;
+ struct inode *inode;
+ int rc = 0;
ENTRY;
if (o->od_mnt != NULL)
RETURN(-E2BIG);
strcpy(o->od_mntdev, dev);
- o->od_fsops = fsfilt_get_ops(mt_str(LDD_MT_LDISKFS));
- if (o->od_fsops == NULL) {
- CERROR("Can't find fsfilt_ldiskfs\n");
- RETURN(-ENOTSUPP);
- }
-
- OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
+ OBD_PAGE_ALLOC(__page, GFP_IOFS);
if (__page == NULL)
GOTO(out, rc = -ENOMEM);
if (str)
lmd_flags = simple_strtoul(str + 1, NULL, 0);
opts = lustre_cfg_string(cfg, 3);
- page = (unsigned long)cfs_page_address(__page);
+ page = (unsigned long)page_address(__page);
options = (char *)page;
*options = '\0';
if (opts == NULL)
/* Glom up mount options */
if (*options != '\0')
strcat(options, ",");
- strlcat(options, "no_mbcache", CFS_PAGE_SIZE);
+ strlcat(options, "no_mbcache", PAGE_CACHE_SIZE);
type = get_fs_type("ldiskfs");
if (!type) {
}
o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
- cfs_module_put(type->owner);
+ module_put(type->owner);
if (IS_ERR(o->od_mnt)) {
rc = PTR_ERR(o->od_mnt);
- CERROR("%s: can't mount %s: %d\n", name, dev, rc);
o->od_mnt = NULL;
+ CERROR("%s: can't mount %s: %d\n", name, dev, rc);
GOTO(out, rc);
}
- if (lvfs_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
+#ifdef HAVE_DEV_SET_RDONLY
+ if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
CERROR("%s: underlying device %s is marked as read-only. "
"Setup failed\n", name, dev);
- mntput(o->od_mnt);
- o->od_mnt = NULL;
- GOTO(out, rc = -EROFS);
+ GOTO(out_mnt, rc = -EROFS);
}
+#endif
if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
CERROR("%s: device %s is mounted w/o journal\n", name, dev);
- mntput(o->od_mnt);
- o->od_mnt = NULL;
- GOTO(out, rc = -EINVAL);
+ GOTO(out_mnt, rc = -EINVAL);
+ }
+
+ inode = osd_sb(o)->s_root->d_inode;
+ ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI);
+ lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
+ rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
+ if (rc != 0) {
+ CERROR("%s: failed to set lma on %s root inode\n", name, dev);
+ GOTO(out_mnt, rc);
}
- ldiskfs_set_inode_state(osd_sb(o)->s_root->d_inode,
- LDISKFS_STATE_LUSTRE_NO_OI);
if (lmd_flags & LMD_FLG_NOSCRUB)
o->od_noscrub = 1;
+ GOTO(out, rc = 0);
+
+out_mnt:
+ mntput(o->od_mnt);
+ o->od_mnt = NULL;
+
out:
if (__page)
OBD_PAGE_FREE(__page);
- if (rc)
- fsfilt_put_ops(o->od_fsops);
- RETURN(rc);
+ return rc;
}
static struct lu_device *osd_device_fini(const struct lu_env *env,
- struct lu_device *d)
+ struct lu_device *d)
{
- int rc;
- ENTRY;
-
- rc = osd_shutdown(env, osd_dev(d));
-
- osd_obj_map_fini(osd_dev(d));
-
- shrink_dcache_sb(osd_sb(osd_dev(d)));
- osd_sync(env, lu2dt_dev(d));
-
- rc = osd_procfs_fini(osd_dev(d));
- if (rc) {
- CERROR("proc fini error %d \n", rc);
- RETURN (ERR_PTR(rc));
- }
+ struct osd_device *o = osd_dev(d);
+ ENTRY;
- if (osd_dev(d)->od_mnt) {
- mntput(osd_dev(d)->od_mnt);
- osd_dev(d)->od_mnt = NULL;
- }
+ osd_shutdown(env, o);
+ osd_procfs_fini(o);
+ osd_scrub_cleanup(env, o);
+ osd_obj_map_fini(o);
+ osd_umount(env, o);
- RETURN(NULL);
+ RETURN(NULL);
}
static int osd_device_init0(const struct lu_env *env,
struct lu_device *l = osd2lu_dev(o);
struct osd_thread_info *info;
int rc;
+ int cplen = 0;
/* if the module was re-loaded, env can loose its keys */
rc = lu_env_refill((struct lu_env *) env);
if (rc)
GOTO(out_capa, rc);
- CFS_INIT_LIST_HEAD(&o->od_ios_list);
- /* setup scrub, including OI files initialization */
- rc = osd_scrub_setup(env, o);
- if (rc < 0)
+ cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
+ sizeof(o->od_svname));
+ if (cplen >= sizeof(o->od_svname)) {
+ rc = -E2BIG;
GOTO(out_mnt, rc);
+ }
- strncpy(o->od_svname, lustre_cfg_string(cfg, 4),
- sizeof(o->od_svname) - 1);
+ if (server_name_is_ost(o->od_svname))
+ o->od_is_ost = 1;
- rc = osd_obj_map_init(o);
+ rc = osd_obj_map_init(env, o);
if (rc != 0)
- GOTO(out_scrub, rc);
+ GOTO(out_mnt, rc);
rc = lu_site_init(&o->od_site, l);
- if (rc)
+ if (rc != 0)
GOTO(out_compat, rc);
o->od_site.ls_bottom_dev = l;
rc = lu_site_init_finish(&o->od_site);
- if (rc)
+ if (rc != 0)
+ GOTO(out_site, rc);
+
+ /* self-repair LMA by default */
+ o->od_lma_self_repair = 1;
+
+ CFS_INIT_LIST_HEAD(&o->od_ios_list);
+ /* setup scrub, including OI files initialization */
+ rc = osd_scrub_setup(env, o);
+ if (rc < 0)
GOTO(out_site, rc);
rc = osd_procfs_init(o, o->od_svname);
if (rc != 0) {
CERROR("%s: can't initialize procfs: rc = %d\n",
o->od_svname, rc);
- GOTO(out_site, rc);
+ GOTO(out_scrub, rc);
}
LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
}
RETURN(0);
+
out_procfs:
osd_procfs_fini(o);
+out_scrub:
+ osd_scrub_cleanup(env, o);
out_site:
lu_site_fini(&o->od_site);
out_compat:
osd_obj_map_fini(o);
-out_scrub:
- osd_scrub_cleanup(env, o);
out_mnt:
- osd_oi_fini(info, o);
- osd_shutdown(env, o);
- mntput(o->od_mnt);
- o->od_mnt = NULL;
+ osd_umount(env, o);
out_capa:
cleanup_capa_hash(o->od_capa_hash);
out:
- RETURN(rc);
+ return rc;
}
static struct lu_device *osd_device_alloc(const struct lu_env *env,
static int osd_process_config(const struct lu_env *env,
struct lu_device *d, struct lustre_cfg *cfg)
{
- struct osd_device *o = osd_dev(d);
- int err;
- ENTRY;
+ struct osd_device *o = osd_dev(d);
+ int rc;
+ ENTRY;
- switch(cfg->lcfg_command) {
- case LCFG_SETUP:
- err = osd_mount(env, o, cfg);
- break;
- case LCFG_CLEANUP:
+ switch (cfg->lcfg_command) {
+ case LCFG_SETUP:
+ rc = osd_mount(env, o, cfg);
+ break;
+ case LCFG_CLEANUP:
lu_dev_del_linkage(d->ld_site, d);
- err = osd_shutdown(env, o);
+ rc = osd_shutdown(env, o);
break;
- default:
- err = -ENOSYS;
- }
+ case LCFG_PARAM:
+ LASSERT(&o->od_dt_dev);
+ rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
+ cfg, &o->od_dt_dev);
+ if (rc > 0 || rc == -ENOSYS)
+ rc = class_process_proc_param(PARAM_OST,
+ lprocfs_osd_obd_vars,
+ cfg, &o->od_dt_dev);
+ break;
+ default:
+ rc = -ENOSYS;
+ }
- RETURN(err);
+ RETURN(rc);
}
static int osd_recovery_complete(const struct lu_env *env,
int result = 0;
ENTRY;
- if (dev->ld_site && lu_device_is_md(dev->ld_site->ls_top_dev)) {
- /* MDT/MDD still use old infrastructure to create
- * special files */
- result = llo_local_objects_setup(env, lu2md_dev(pdev),
- lu2dt_dev(dev));
- if (result)
- RETURN(result);
- }
-
if (osd->od_quota_slave != NULL)
/* set up quota slave objects */
result = qsd_prepare(env, osd->od_quota_slave);
static int __init osd_mod_init(void)
{
- struct lprocfs_static_vars lvars;
+ int rc;
+
+ osd_oi_mod_init();
- osd_oi_mod_init();
- lprocfs_osd_init_vars(&lvars);
- return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
- LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
+ rc = lu_kmem_init(ldiskfs_caches);
+ if (rc)
+ return rc;
+
+ rc = class_register_type(&osd_obd_device_ops, NULL, NULL,
+#ifndef HAVE_ONLY_PROCFS_SEQ
+ lprocfs_osd_module_vars,
+#endif
+ LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
+ if (rc)
+ lu_kmem_fini(ldiskfs_caches);
+ return rc;
}
static void __exit osd_mod_exit(void)
{
class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
+ lu_kmem_fini(ldiskfs_caches);
}
MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");