/* 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"
id->oii_ino, PTR_ERR(inode));
} else if (id->oii_gen != OSD_OII_NOGEN &&
inode->i_generation != id->oii_gen) {
- CDEBUG(D_INODE, "unmatched inode: ino = %u, gen0 = %u, "
- "gen1 = %u\n",
+ CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
+ "i_generation = %u\n",
id->oii_ino, id->oii_gen, inode->i_generation);
iput(inode);
inode = ERR_PTR(-ESTALE);
/* due to parallel readdir and unlink,
* we can have dead inode here. */
CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
- make_bad_inode(inode);
iput(inode);
inode = ERR_PTR(-ESTALE);
} else if (is_bad_inode(inode)) {
return inode;
}
+static struct inode *osd_iget_check(struct osd_thread_info *info,
+ struct osd_device *dev,
+ const struct lu_fid *fid,
+ struct osd_inode_id *id,
+ bool in_oi)
+{
+ struct inode *inode;
+ int rc = 0;
+ ENTRY;
+
+ inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
+ if (IS_ERR(inode)) {
+ rc = PTR_ERR(inode);
+ if (!in_oi || (rc != -ENOENT && rc != -ESTALE)) {
+ CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
+ id->oii_ino, rc);
+
+ GOTO(put, rc);
+ }
+
+ goto check_oi;
+ }
+
+ if (is_bad_inode(inode)) {
+ rc = -ENOENT;
+ if (!in_oi) {
+ CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
+
+ GOTO(put, rc);
+ }
+
+ goto check_oi;
+ }
+
+ if (id->oii_gen != OSD_OII_NOGEN &&
+ inode->i_generation != id->oii_gen) {
+ rc = -ESTALE;
+ if (!in_oi) {
+ CDEBUG(D_INODE, "unmatched inode: ino = %u, "
+ "oii_gen = %u, i_generation = %u\n",
+ id->oii_ino, id->oii_gen, inode->i_generation);
+
+ GOTO(put, rc);
+ }
+
+ goto check_oi;
+ }
+
+ if (inode->i_nlink == 0) {
+ rc = -ENOENT;
+ if (!in_oi) {
+ CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
+
+ GOTO(put, rc);
+ }
+
+ goto check_oi;
+ }
+
+check_oi:
+ if (rc != 0) {
+ LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
+
+ rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
+ /* XXX: There are some possible cases:
+ * 1. rc = 0.
+ * Backup/restore caused the OI invalid.
+ * 2. rc = 0.
+ * Someone unlinked the object but NOT removed
+ * the OI mapping, such as mount target device
+ * as ldiskfs, and modify something directly.
+ * 3. rc = -ENOENT.
+ * Someone just removed the object between the
+ * former oi_lookup and the iget. It is normal.
+ * 4. Other failure cases.
+ *
+ * Generally, when the device is mounted, it will
+ * auto check whether the system is restored from
+ * file-level backup or not. We trust such detect
+ * to distinguish the 1st case from the 2nd case. */
+ if (rc == 0) {
+ if (!IS_ERR(inode) && inode->i_generation != 0 &&
+ inode->i_generation == id->oii_gen)
+ rc = -ENOENT;
+ else
+ rc = -EREMCHG;
+ }
+ } else {
+ if (id->oii_gen == OSD_OII_NOGEN)
+ osd_id_gen(id, inode->i_ino, inode->i_generation);
+
+ /* Do not update file c/mtime in ldiskfs.
+ * NB: we don't have any lock to protect this because we don't
+ * have reference on osd_object now, but contention with
+ * another lookup + attr_set can't happen in the tiny window
+ * between if (...) and set S_NOCMTIME. */
+ if (!(inode->i_flags & S_NOCMTIME))
+ inode->i_flags |= S_NOCMTIME;
+ }
+
+ GOTO(put, rc);
+
+put:
+ if (rc != 0) {
+ if (!IS_ERR(inode))
+ iput(inode);
+
+ inode = ERR_PTR(rc);
+ }
+
+ return inode;
+}
+
/**
* \retval +v: new filter_fid, does not contain self-fid
* \retval 0: filter_fid_old, contains self-fid
return 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;
+ }
+
+ 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 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;
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(lu_object_fid(&obj->oo_dt.do_lu)) &&
- osd->od_check_ff) {
+ 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)) {
- handle_t *jh;
-
/* 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. */
-
- 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)) {
- CWARN("%s: cannot start journal for "
- "lma_self_repair: rc = %ld\n",
- osd_name(osd), PTR_ERR(jh));
- RETURN(0);
- }
-
- rc = osd_ea_fid_set(info, inode,
- lu_object_fid(&obj->oo_dt.do_lu),
- fid_is_on_ost(info, osd,
- lu_object_fid(&obj->oo_dt.do_lu),
- OI_CHECK_FLD) ?
- LMAC_FID_ON_OST : 0, 0);
- if (rc != 0)
- CWARN("%s: cannot self repair the LMA: "
- "rc = %d\n", osd_name(osd), rc);
- ldiskfs_journal_stop(jh);
+ 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);
}
}
CWARN("%s: unsupported incompat LMA feature(s) %#x for "
"fid = "DFID", ino = %lu\n", osd_name(osd),
lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
- PFID(lu_object_fid(&obj->oo_dt.do_lu)),
- inode->i_ino);
+ 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(lu_object_fid(&obj->oo_dt.do_lu), 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(lu_object_fid(&obj->oo_dt.do_lu)),
- PFID(&lma->lma_self_fid));
+ osd_name(osd), PFID(rfid), PFID(fid));
rc = -EREMCHG;
}
in_oi = true;
iget:
- inode = osd_iget(info, dev, id);
+ inode = osd_iget_check(info, dev, fid, id, in_oi);
if (IS_ERR(inode)) {
result = PTR_ERR(inode);
if (result == -ENOENT || result == -ESTALE) {
- if (!in_oi) {
+ 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;
+ GOTO(out, result = -ENOENT);
} else if (result == -EREMCHG) {
trigger:
+ if (!in_oi)
+ fid_zero(&oic->oic_fid);
+
if (unlikely(triggered))
GOTO(out, result = saved);
result = -EINPROGRESS;
} else if (!dev->od_noscrub) {
result = osd_scrub_start(dev);
- LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC "
- "for "DFID", rc = %d [1]\n",
- LDISKFS_SB(osd_sb(dev))->s_es->\
- s_volume_name,PFID(fid), result);
+ LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
+ "for "DFID", rc = %d [1]\n",
+ osd_name(dev), PFID(fid),result);
if (result == 0 || result == -EALREADY)
result = -EINPROGRESS;
else
if (result != 0) {
iput(inode);
obj->oo_inode = NULL;
- if (result == -EREMCHG)
+ if (result == -EREMCHG) {
+ if (!in_oi) {
+ result = osd_oi_lookup(info, dev, fid, id,
+ OI_CHECK_FLD);
+ if (result != 0) {
+ fid_zero(&oic->oic_fid);
+ GOTO(out, result);
+ }
+ }
+
goto trigger;
+ }
GOTO(out, result);
}
lu_context_exit(&th->th_ctx);
lu_context_fini(&th->th_ctx);
- OBD_FREE_PTR(oh);
+ thandle_put(th);
}
static struct thandle *osd_trans_create(const struct lu_env *env,
- struct dt_device *d)
+ struct dt_device *d)
{
- struct osd_thread_info *oti = osd_oti_get(env);
- struct osd_iobuf *iobuf = &oti->oti_iobuf;
- struct osd_thandle *oh;
- struct thandle *th;
- ENTRY;
+ struct osd_thread_info *oti = osd_oti_get(env);
+ struct osd_iobuf *iobuf = &oti->oti_iobuf;
+ struct osd_thandle *oh;
+ struct thandle *th;
+ ENTRY;
- /* on pending IO in this thread should left from prev. request */
- LASSERT(cfs_atomic_read(&iobuf->dr_numreqs) == 0);
+ /* on pending IO in this thread should left from prev. request */
+ LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
th = ERR_PTR(-ENOMEM);
OBD_ALLOC_GFP(oh, sizeof *oh, __GFP_IO);
th->th_result = 0;
th->th_tags = LCT_TX_HANDLE;
oh->ot_credits = 0;
- oti->oti_dev = osd_dt_dev(d);
+ atomic_set(&th->th_refc, 1);
+ th->th_alloc_size = sizeof(*oh);
+ oti->oti_dev = osd_dt_dev(d);
CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
osd_th_alloced(oh);
/*
* Concurrency: shouldn't matter.
*/
-static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
+static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
+ struct thandle *th)
{
int rc = 0;
struct osd_thandle *oh;
if (rc != 0)
CERROR("Failure to stop transaction: %d\n", rc);
} else {
- OBD_FREE_PTR(oh);
+ thandle_put(&oh->ot_super);
}
/* as we want IO to journal and data IO be concurrent, we don't block
* completed otherwise iobuf may be corrupted by different request
*/
wait_event(iobuf->dr_wait,
- cfs_atomic_read(&iobuf->dr_numreqs) == 0);
+ atomic_read(&iobuf->dr_numreqs) == 0);
if (!rc)
rc = iobuf->dr_error;
d ? d->id_ops->id_name : "plain");
}
+#define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
+
/*
* Concurrency: shouldn't matter.
*/
}
spin_lock(&osd->od_osfs_lock);
- /* cache 1 second */
- if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
- result = sb->s_op->statfs(sb->s_root, ksfs);
- if (likely(result == 0)) { /* N.B. statfs can't really fail */
- osd->od_osfs_age = cfs_time_current_64();
- statfs_pack(&osd->od_statfs, ksfs);
- if (sb->s_flags & MS_RDONLY)
- sfs->os_state = OS_STATE_READONLY;
- }
+ result = sb->s_op->statfs(sb->s_root, ksfs);
+ if (likely(result == 0)) { /* N.B. statfs can't really fail */
+ statfs_pack(sfs, ksfs);
+ if (sb->s_flags & MS_RDONLY)
+ sfs->os_state = OS_STATE_READONLY;
}
- if (likely(result == 0))
- *sfs = osd->od_statfs;
spin_unlock(&osd->od_osfs_lock);
- if (unlikely(env == NULL))
+ if (unlikely(env == NULL))
OBD_FREE_PTR(ksfs);
+ /* Reserve a small amount of space for local objects like last_rcvd,
+ * llog, quota files, ... */
+ if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
+ sfs->os_bavail = 0;
+ } else {
+ sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
+ /** Take out metadata overhead for indirect blocks */
+ sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
+ }
+
return result;
}
{
LASSERT(ah);
- memset(ah, 0, sizeof(*ah));
ah->dah_parent = parent;
ah->dah_mode = child_mode;
}
if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
return -EACCES;
+ CDEBUG(D_INODE, DFID" set xattr '%s' with size %zd\n",
+ PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
+
osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
if (fl & LU_XATTR_REPLACE)
fs_flags |= XATTR_REPLACE;
}
if (!dev->od_noscrub && ++once == 1) {
- CDEBUG(D_LFSCK, "Trigger OI scrub by RPC for "DFID"\n",
- PFID(fid));
rc = osd_scrub_start(dev);
- LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC for "DFID
- ", rc = %d [2]\n",
- LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
- PFID(fid), rc);
+ LCONSOLE_WARN("%.16s: trigger OI scrub by RPC for "DFID
+ ", rc = %d [2]\n",
+ LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
+ PFID(fid), rc);
if (rc == 0)
goto again;
}
GOTO(out, rc);
}
+ if (osd_remote_fid(env, dev, fid))
+ GOTO(out, rc = 0);
+
rc = osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id,
fid);
if (rc != 0)
else
up_read(&obj->oo_ext_idx_sem);
- if (it->oie_rd_dirent == 0) {
- result = -EIO;
- } else {
- it->oie_dirent = it->oie_buf;
- it->oie_it_dirent = 1;
- }
+ if (it->oie_rd_dirent == 0) {
+ /*If it does not get any dirent, it means it has been reached
+ *to the end of the dir */
+ it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
+ } else {
+ it->oie_dirent = it->oie_buf;
+ it->oie_it_dirent = 1;
+ }
- RETURN(result);
+ RETURN(result);
}
/**
#endif
if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
- LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
+ LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
CERROR("%s: device %s is mounted w/o journal\n", name, dev);
GOTO(out_mnt, rc = -EINVAL);
}
+#ifdef LDISKFS_MOUNT_DIRDATA
+ if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
+ LDISKFS_FEATURE_INCOMPAT_DIRDATA))
+ LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
+#endif
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) {
struct osd_device *o = osd_dev(d);
ENTRY;
- osd_procfs_fini(o);
osd_shutdown(env, o);
+ osd_procfs_fini(o);
osd_scrub_cleanup(env, o);
osd_obj_map_fini(o);
osd_umount(env, o);
spin_lock_init(&o->od_osfs_lock);
mutex_init(&o->od_otable_mutex);
- o->od_osfs_age = cfs_time_shift_64(-1000);
o->od_capa_hash = init_capa_hash();
if (o->od_capa_hash == NULL)
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,
static int __init osd_mod_init(void)
{
- struct lprocfs_static_vars lvars;
int rc;
osd_oi_mod_init();
- lprocfs_osd_init_vars(&lvars);
rc = lu_kmem_init(ldiskfs_caches);
if (rc)
rc = class_register_type(&osd_obd_device_ops, NULL, NULL,
#ifndef HAVE_ONLY_PROCFS_SEQ
- lvars.module_vars,
+ lprocfs_osd_module_vars,
#endif
LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
if (rc)