* Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, 2012, Whamcloud, Inc.
+ * Copyright (c) 2011, 2013, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
si_meminfo(&si);
pages = si.totalram - si.totalhigh;
if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
-#ifdef HAVE_BGL_SUPPORT
- lru_page_max = pages / 4;
-#else
lru_page_max = pages / 2;
-#endif
} else {
lru_page_max = (pages / 4) * 3;
}
OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
OBD_CONNECT_EINPROGRESS |
- OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE;
+ OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
+ OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
data->ocd_connect_flags |= OBD_CONNECT_SOM;
if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
- data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
+ data->ocd_brw_size = MD_MAX_BRW_SIZE;
err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
if (err == -EBUSY) {
GOTO(out, err);
}
- err = obd_fid_init(sbi->ll_md_exp);
- if (err) {
- CERROR("Can't init metadata layer FID infrastructure, "
- "rc %d\n", err);
- GOTO(out_md, err);
- }
+ sbi->ll_md_exp->exp_connect_data = *data;
- err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
- cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
- if (err)
- GOTO(out_md_fid, err);
-
- /* This needs to be after statfs to ensure connect has finished.
- * Note that "data" does NOT contain the valid connect reply.
- * If connecting to a 1.8 server there will be no LMV device, so
- * we can access the MDC export directly and exp_connect_flags will
- * be non-zero, but if accessing an upgraded 2.1 server it will
- * have the correct flags filled in.
- * XXX: fill in the LMV exp_connect_flags from MDC(s). */
- valid = sbi->ll_md_exp->exp_connect_flags & CLIENT_CONNECT_MDT_REQD;
- if (sbi->ll_md_exp->exp_connect_flags != 0 &&
- valid != CLIENT_CONNECT_MDT_REQD) {
- char *buf;
-
- OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
- obd_connect_flags2str(buf, CFS_PAGE_SIZE,
- valid ^ CLIENT_CONNECT_MDT_REQD, ",");
- LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
- "feature(s) needed for correct operation "
- "of this client (%s). Please upgrade "
- "server or downgrade client.\n",
- sbi->ll_md_exp->exp_obd->obd_name, buf);
- OBD_FREE(buf, CFS_PAGE_SIZE);
- GOTO(out_md, err = -EPROTO);
- }
+ err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
+ LUSTRE_SEQ_METADATA);
+ if (err) {
+ CERROR("%s: Can't init metadata layer FID infrastructure, "
+ "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
+ GOTO(out_md, err);
+ }
- size = sizeof(*data);
- err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
- KEY_CONN_DATA, &size, data, NULL);
- if (err) {
- CERROR("Get connect data failed: %d \n", err);
- GOTO(out_md, err);
- }
+ /* For mount, we only need fs info from MDT0, and also in DNE, it
+ * can make sure the client can be mounted as long as MDT0 is
+ * avaible */
+ err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
+ cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
+ OBD_STATFS_FOR_MDT0);
+ if (err)
+ GOTO(out_md_fid, err);
+
+ /* This needs to be after statfs to ensure connect has finished.
+ * Note that "data" does NOT contain the valid connect reply.
+ * If connecting to a 1.8 server there will be no LMV device, so
+ * we can access the MDC export directly and exp_connect_flags will
+ * be non-zero, but if accessing an upgraded 2.1 server it will
+ * have the correct flags filled in.
+ * XXX: fill in the LMV exp_connect_flags from MDC(s). */
+ valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
+ if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
+ valid != CLIENT_CONNECT_MDT_REQD) {
+ char *buf;
+
+ OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
+ obd_connect_flags2str(buf, CFS_PAGE_SIZE,
+ valid ^ CLIENT_CONNECT_MDT_REQD, ",");
+ LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
+ "feature(s) needed for correct operation "
+ "of this client (%s). Please upgrade "
+ "server or downgrade client.\n",
+ sbi->ll_md_exp->exp_obd->obd_name, buf);
+ OBD_FREE(buf, CFS_PAGE_SIZE);
+ GOTO(out_md_fid, err = -EPROTO);
+ }
+
+ size = sizeof(*data);
+ err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
+ KEY_CONN_DATA, &size, data, NULL);
+ if (err) {
+ CERROR("%s: Get connect data failed: rc = %d\n",
+ sbi->ll_md_exp->exp_obd->obd_name, err);
+ GOTO(out_md_fid, err);
+ }
- LASSERT(osfs->os_bsize);
+ LASSERT(osfs->os_bsize);
sb->s_blocksize = osfs->os_bsize;
sb->s_blocksize_bits = log2(osfs->os_bsize);
sb->s_magic = LL_SUPER_MAGIC;
if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
sbi->ll_flags |= LL_SBI_64BIT_HASH;
- if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
- sbi->ll_md_brw_size = data->ocd_brw_size;
- else
- sbi->ll_md_brw_size = CFS_PAGE_SIZE;
+ if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
+ sbi->ll_md_brw_size = data->ocd_brw_size;
+ else
+ sbi->ll_md_brw_size = CFS_PAGE_SIZE;
if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
LCONSOLE_INFO("Layout lock feature supported.\n");
sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
}
- obd = class_name2obd(dt);
- if (!obd) {
- CERROR("DT %s: not setup or attached\n", dt);
- GOTO(out_md_fid, err = -ENODEV);
- }
+ obd = class_name2obd(dt);
+ if (!obd) {
+ CERROR("DT %s: not setup or attached\n", dt);
+ GOTO(out_md_fid, err = -ENODEV);
+ }
data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
- OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
+ OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
OBD_CONNECT_MAXBYTES |
OBD_CONNECT_EINPROGRESS |
- OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE;
+ OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
+ OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
data->ocd_connect_flags |= OBD_CONNECT_SOM;
obd->obd_upcall.onu_owner = &sbi->ll_lco;
obd->obd_upcall.onu_upcall = cl_ocd_update;
- data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
+ data->ocd_brw_size = DT_MAX_BRW_SIZE;
+
+ err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
+ NULL);
+ if (err == -EBUSY) {
+ LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
+ "recovery, of which this client is not a "
+ "part. Please wait for recovery to "
+ "complete, abort, or time out.\n", dt);
+ GOTO(out_md, err);
+ } else if (err) {
+ CERROR("%s: Cannot connect to %s: rc = %d\n",
+ sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
+ GOTO(out_md, err);
+ }
- err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
- if (err == -EBUSY) {
- LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
- "recovery, of which this client is not a "
- "part. Please wait for recovery to "
- "complete, abort, or time out.\n", dt);
- GOTO(out_md_fid, err);
- } else if (err) {
- CERROR("Cannot connect to %s: rc = %d\n", dt, err);
- GOTO(out_md_fid, err);
- }
+ sbi->ll_dt_exp->exp_connect_data = *data;
- err = obd_fid_init(sbi->ll_dt_exp);
- if (err) {
- CERROR("Can't init data layer FID infrastructure, "
- "rc %d\n", err);
- GOTO(out_dt, err);
- }
+ err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
+ LUSTRE_SEQ_METADATA);
+ if (err) {
+ CERROR("%s: Can't init data layer FID infrastructure, "
+ "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
+ GOTO(out_dt, err);
+ }
mutex_lock(&sbi->ll_lco.lco_lock);
- sbi->ll_lco.lco_flags = data->ocd_connect_flags;
- sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
- sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
+ sbi->ll_lco.lco_flags = data->ocd_connect_flags;
+ sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
+ sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
mutex_unlock(&sbi->ll_lco.lco_lock);
- fid_zero(&sbi->ll_root_fid);
- err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
- if (err) {
- CERROR("cannot mds_connect: rc = %d\n", err);
- GOTO(out_lock_cn_cb, err);
- }
- if (!fid_is_sane(&sbi->ll_root_fid)) {
- CERROR("Invalid root fid during mount\n");
- GOTO(out_lock_cn_cb, err = -EINVAL);
- }
- CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
+ fid_zero(&sbi->ll_root_fid);
+ err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
+ if (err) {
+ CERROR("cannot mds_connect: rc = %d\n", err);
+ GOTO(out_lock_cn_cb, err);
+ }
+ if (!fid_is_sane(&sbi->ll_root_fid)) {
+ CERROR("%s: Invalid root fid "DFID" during mount\n",
+ sbi->ll_md_exp->exp_obd->obd_name,
+ PFID(&sbi->ll_root_fid));
+ GOTO(out_lock_cn_cb, err = -EINVAL);
+ }
+ CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
- sb->s_op = &lustre_super_operations;
+ sb->s_op = &lustre_super_operations;
#if THREAD_SIZE >= 8192 /*b=17630*/
sb->s_export_op = &lustre_export_operations;
#endif
- /* make root inode
- * XXX: move this to after cbd setup? */
- valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
- if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
- valid |= OBD_MD_FLRMTPERM;
- else if (sbi->ll_flags & LL_SBI_ACL)
- valid |= OBD_MD_FLACL;
+ /* make root inode
+ * XXX: move this to after cbd setup? */
+ valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
+ if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
+ valid |= OBD_MD_FLRMTPERM;
+ else if (sbi->ll_flags & LL_SBI_ACL)
+ valid |= OBD_MD_FLACL;
- OBD_ALLOC_PTR(op_data);
- if (op_data == NULL)
- GOTO(out_lock_cn_cb, err = -ENOMEM);
+ OBD_ALLOC_PTR(op_data);
+ if (op_data == NULL)
+ GOTO(out_lock_cn_cb, err = -ENOMEM);
- op_data->op_fid1 = sbi->ll_root_fid;
- op_data->op_mode = 0;
- op_data->op_capa1 = oc;
- op_data->op_valid = valid;
+ op_data->op_fid1 = sbi->ll_root_fid;
+ op_data->op_mode = 0;
+ op_data->op_capa1 = oc;
+ op_data->op_valid = valid;
+
+ err = md_getattr(sbi->ll_md_exp, op_data, &request);
+ if (oc)
+ capa_put(oc);
+ OBD_FREE_PTR(op_data);
+ if (err) {
+ CERROR("%s: md_getattr failed for root: rc = %d\n",
+ sbi->ll_md_exp->exp_obd->obd_name, err);
+ GOTO(out_lock_cn_cb, err);
+ }
- err = md_getattr(sbi->ll_md_exp, op_data, &request);
- if (oc)
- capa_put(oc);
- OBD_FREE_PTR(op_data);
- if (err) {
- CERROR("md_getattr failed for root: rc = %d\n", err);
- GOTO(out_lock_cn_cb, err);
- }
- err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
- sbi->ll_md_exp, &lmd);
- if (err) {
- CERROR("failed to understand root inode md: rc = %d\n", err);
- ptlrpc_req_finished (request);
- GOTO(out_lock_cn_cb, err);
- }
+ err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
+ sbi->ll_md_exp, &lmd);
+ if (err) {
+ CERROR("failed to understand root inode md: rc = %d\n", err);
+ ptlrpc_req_finished(request);
+ GOTO(out_lock_cn_cb, err);
+ }
LASSERT(fid_is_sane(&sbi->ll_root_fid));
- root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
+ root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
+ sbi->ll_flags & LL_SBI_32BIT_API),
+ &lmd);
md_free_lustre_md(sbi->ll_md_exp, &lmd);
ptlrpc_req_finished(request);
if (sb->s_root == NULL) {
CERROR("%s: can't make root dentry\n",
ll_get_fsname(sb, NULL, 0));
- GOTO(out_lock_cn_cb, err = -ENOMEM);
+ GOTO(out_root, err = -ENOMEM);
}
#ifdef HAVE_DCACHE_LOCK
if (root)
iput(root);
out_lock_cn_cb:
- obd_fid_fini(sbi->ll_dt_exp);
+ obd_fid_fini(sbi->ll_dt_exp->exp_obd);
out_dt:
obd_disconnect(sbi->ll_dt_exp);
sbi->ll_dt_exp = NULL;
+ /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
+ obd_zombie_barrier();
out_md_fid:
- obd_fid_fini(sbi->ll_md_exp);
+ obd_fid_fini(sbi->ll_md_exp->exp_obd);
out_md:
obd_disconnect(sbi->ll_md_exp);
sbi->ll_md_exp = NULL;
void ll_dump_inode(struct inode *inode)
{
- struct list_head *tmp;
- int dentry_count = 0;
+ struct ll_d_hlist_node *tmp;
+ int dentry_count = 0;
- LASSERT(inode != NULL);
+ LASSERT(inode != NULL);
- list_for_each(tmp, &inode->i_dentry)
- dentry_count++;
+ ll_d_hlist_for_each(tmp, &inode->i_dentry)
+ dentry_count++;
- CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
- inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
- inode->i_mode, atomic_read(&inode->i_count), dentry_count);
+ CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
+ inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
+ inode->i_mode, atomic_read(&inode->i_count), dentry_count);
}
void lustre_dump_dentry(struct dentry *dentry, int recur)
cfs_list_del(&sbi->ll_conn_chain);
- obd_fid_fini(sbi->ll_dt_exp);
+ obd_fid_fini(sbi->ll_dt_exp->exp_obd);
obd_disconnect(sbi->ll_dt_exp);
sbi->ll_dt_exp = NULL;
+ /* wait till all OSCs are gone, since cl_cache is accessing sbi.
+ * see LU-2543. */
+ obd_zombie_barrier();
lprocfs_unregister_mountpoint(sbi);
- obd_fid_fini(sbi->ll_md_exp);
+ obd_fid_fini(sbi->ll_md_exp->exp_obd);
obd_disconnect(sbi->ll_md_exp);
sbi->ll_md_exp = NULL;
/* we need restore s_dev from changed for clustred NFS before put_super
* because new kernels have cached s_dev and change sb->s_dev in
* put_super not affected real removing devices */
- if (sbi)
- sb->s_dev = sbi->ll_sdev_orig;
- EXIT;
+ if (sbi) {
+ sb->s_dev = sbi->ll_sdev_orig;
+ sbi->ll_umounting = 1;
+ }
+ EXIT;
}
char *ll_read_opt(const char *opt, char *data)
mutex_init(&lli->lli_och_mutex);
spin_lock_init(&lli->lli_agl_lock);
lli->lli_has_smd = false;
+ lli->lli_layout_gen = LL_LAYOUT_GEN_NONE;
lli->lli_clob = NULL;
LASSERT(lli->lli_vfs_inode.i_mode != 0);
CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
lli->lli_agl_index = 0;
lli->lli_async_rc = 0;
+ lli->lli_volatile = false;
}
mutex_init(&lli->lli_layout_mutex);
}
if (err)
GOTO(out_free, err);
- err = ll_bdi_init(&lsi->lsi_bdi);
- if (err)
- GOTO(out_free, err);
- lsi->lsi_flags |= LSI_BDI_INITIALIZED;
- lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
- err = ll_bdi_register(&lsi->lsi_bdi);
- if (err)
- GOTO(out_free, err);
+ err = bdi_init(&lsi->lsi_bdi);
+ if (err)
+ GOTO(out_free, err);
+ lsi->lsi_flags |= LSI_BDI_INITIALIZED;
+ lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
+ err = ll_bdi_register(&lsi->lsi_bdi);
+ if (err)
+ GOTO(out_free, err);
#ifdef HAVE_SB_BDI
sb->s_bdi = &lsi->lsi_bdi;
if (profilenm)
class_del_profile(profilenm);
- if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
- ll_bdi_destroy(&lsi->lsi_bdi);
- lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
- }
+ if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
+ bdi_destroy(&lsi->lsi_bdi);
+ lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
+ }
ll_free_sbi(sb);
lsi->lsi_llsbi = NULL;
lustre_common_put_super(sb);
- cl_env_cache_purge(~0);
-
cfs_module_put(THIS_MODULE);
EXIT;
} /* client_put_super */
+struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
+{
+ struct inode *inode = NULL;
+
+ /* NOTE: we depend on atomic igrab() -bzzz */
+ lock_res_and_lock(lock);
+ if (lock->l_resource->lr_lvb_inode) {
+ struct ll_inode_info * lli;
+ lli = ll_i2info(lock->l_resource->lr_lvb_inode);
+ if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
+ inode = igrab(lock->l_resource->lr_lvb_inode);
+ } else {
+ inode = lock->l_resource->lr_lvb_inode;
+ LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
+ D_WARNING, lock, "lr_lvb_inode %p is "
+ "bogus: magic %08x",
+ lock->l_resource->lr_lvb_inode,
+ lli->lli_inode_magic);
+ inode = NULL;
+ }
+ }
+ unlock_res_and_lock(lock);
+ return inode;
+}
+
struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
{
struct inode *inode = NULL;
return inode;
}
-static int null_if_equal(struct ldlm_lock *lock, void *data)
-{
- if (data == lock->l_ast_data) {
- lock->l_ast_data = NULL;
-
- if (lock->l_req_mode != lock->l_granted_mode)
- LDLM_ERROR(lock,"clearing inode with ungranted lock");
- }
-
- return LDLM_ITER_CONTINUE;
-}
-
void ll_clear_inode(struct inode *inode)
{
struct ll_inode_info *lli = ll_i2info(inode);
}
ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
- md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
- null_if_equal, inode);
+ md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
LASSERT(!lli->lli_open_fd_write_count);
LASSERT(!lli->lli_open_fd_read_count);
RETURN(rc);
}
- /* We want to adjust timestamps.
- * If there is at least some data in file, we cleared ATTR_SIZE
- * to avoid update size, otherwise it is important to do.(SOM case)
- * (bug 6196) */
- ia_valid = op_data->op_attr.ia_valid;
- /* Since we set ATTR_*_SET flags above, and already done permission
- * check, So don't let inode_change_ok() check it again. */
- op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
- rc = simple_setattr(dentry, &op_data->op_attr);
- op_data->op_attr.ia_valid = ia_valid;
+ ia_valid = op_data->op_attr.ia_valid;
+ /* inode size will be in ll_setattr_ost, can't do it now since dirty
+ * cache is not cleared yet. */
+ op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
+ rc = simple_setattr(dentry, &op_data->op_attr);
+ op_data->op_attr.ia_valid = ia_valid;
/* Extract epoch data if obtained. */
op_data->op_handle = md.body->handle;
*/
int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
{
- struct lov_stripe_md *lsm;
struct inode *inode = dentry->d_inode;
struct ll_inode_info *lli = ll_i2info(inode);
struct md_op_data *op_data = NULL;
struct md_open_data *mod = NULL;
- int ia_valid = attr->ia_valid;
int rc = 0, rc1 = 0;
ENTRY;
PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
attr->ia_valid);
- if (ia_valid & ATTR_SIZE) {
+ if (attr->ia_valid & ATTR_SIZE) {
/* Check new size against VFS/VM file size limit and rlimit */
rc = inode_newsize_ok(inode, attr->ia_size);
if (rc)
}
/* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
- if (ia_valid & TIMES_SET_FLAGS) {
+ if (attr->ia_valid & TIMES_SET_FLAGS) {
if (cfs_curproc_fsuid() != inode->i_uid &&
!cfs_capable(CFS_CAP_FOWNER))
RETURN(-EPERM);
attr->ia_ctime = CFS_CURRENT_TIME;
attr->ia_valid |= ATTR_CTIME_SET;
}
- if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
+ if (!(attr->ia_valid & ATTR_ATIME_SET) &&
+ (attr->ia_valid & ATTR_ATIME)) {
attr->ia_atime = CFS_CURRENT_TIME;
attr->ia_valid |= ATTR_ATIME_SET;
}
- if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
+ if (!(attr->ia_valid & ATTR_MTIME_SET) &&
+ (attr->ia_valid & ATTR_MTIME)) {
attr->ia_mtime = CFS_CURRENT_TIME;
attr->ia_valid |= ATTR_MTIME_SET;
}
LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
cfs_time_current_sec());
+ /* If we are changing file size, file content is modified, flag it. */
+ if (attr->ia_valid & ATTR_SIZE) {
+ attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
+ spin_lock(&lli->lli_lock);
+ lli->lli_flags |= LLIF_DATA_MODIFIED;
+ spin_unlock(&lli->lli_lock);
+ }
+
/* We always do an MDS RPC, even if we're only changing the size;
* only the MDS knows whether truncate() should fail with -ETXTBUSY */
RETURN(-ENOMEM);
if (!S_ISDIR(inode->i_mode)) {
- if (ia_valid & ATTR_SIZE)
+ if (attr->ia_valid & ATTR_SIZE)
inode_dio_write_done(inode);
mutex_unlock(&inode->i_mutex);
down_write(&lli->lli_trunc_sem);
- mutex_lock(&inode->i_mutex);
- if (ia_valid & ATTR_SIZE)
- inode_dio_wait(inode);
}
- /* We need a steady stripe configuration for setattr to avoid
- * confusion. */
- lsm = ccc_inode_lsm_get(inode);
-
- /* NB: ATTR_SIZE will only be set after this point if the size
- * resides on the MDS, ie, this file has no objects. */
- if (lsm != NULL)
- attr->ia_valid &= ~ATTR_SIZE;
- /* can't call ll_setattr_ost() while holding a refcount of lsm */
- ccc_inode_lsm_put(inode, lsm);
-
memcpy(&op_data->op_attr, attr, sizeof(*attr));
/* Open epoch for truncate. */
if (exp_connect_som(ll_i2mdexp(inode)) &&
- (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
+ (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
op_data->op_flags = MF_EPOCH_OPEN;
rc = ll_md_setattr(dentry, op_data, &mod);
if (rc)
GOTO(out, rc);
+ /* RPC to MDT is sent, cancel data modification flag */
+ if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
+ spin_lock(&lli->lli_lock);
+ lli->lli_flags &= ~LLIF_DATA_MODIFIED;
+ spin_unlock(&lli->lli_lock);
+ }
+
ll_ioepoch_open(lli, op_data->op_ioepoch);
if (!S_ISREG(inode->i_mode))
GOTO(out, rc = 0);
- if (ia_valid & ATTR_SIZE)
- attr->ia_valid |= ATTR_SIZE;
- if (ia_valid & (ATTR_SIZE |
- ATTR_ATIME | ATTR_ATIME_SET |
- ATTR_MTIME | ATTR_MTIME_SET))
+ if (attr->ia_valid & (ATTR_SIZE |
+ ATTR_ATIME | ATTR_ATIME_SET |
+ ATTR_MTIME | ATTR_MTIME_SET))
/* For truncate and utimes sending attributes to OSTs, setting
* mtime/atime to the past will be performed under PW [0:EOF]
* extent lock (new_size:EOF for truncate). It may seem
}
ll_finish_md_op_data(op_data);
}
- if (!S_ISDIR(inode->i_mode))
+ if (!S_ISDIR(inode->i_mode)) {
up_write(&lli->lli_trunc_sem);
+ mutex_lock(&inode->i_mutex);
+ if (attr->ia_valid & ATTR_SIZE)
+ inode_dio_wait(inode);
+ }
- ll_stats_ops_tally(ll_i2sbi(inode), (ia_valid & ATTR_SIZE) ?
- LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
+ ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
+ LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
- return rc;
+ return rc;
}
int ll_setattr(struct dentry *de, struct iattr *attr)
LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
if (lsm != NULL) {
- LASSERT(S_ISREG(inode->i_mode));
- CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
- lsm, inode->i_ino, inode->i_generation, inode);
- /* cl_file_inode_init must go before lli_has_smd or a race
- * is possible where client thinks the file has stripes,
- * but lov raid0 is not setup yet and parallel e.g.
- * glimpse would try to use uninitialized lov */
- if (cl_file_inode_init(inode, md) == 0)
- lli->lli_has_smd = true;
+ if (!lli->lli_has_smd &&
+ !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
+ cl_file_inode_init(inode, md);
lli->lli_maxbytes = lsm->lsm_maxbytes;
if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
lli->lli_maxbytes = MAX_LFS_FILESIZE;
- if (md->lsm != NULL)
- obd_free_memmd(ll_i2dtexp(inode), &md->lsm);
}
if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
spin_unlock(&lli->lli_lock);
}
#endif
- inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
+ inode->i_ino = cl_fid_build_ino(&body->fid1,
+ sbi->ll_flags & LL_SBI_32BIT_API);
inode->i_generation = cl_fid_build_gen(&body->fid1);
if (body->valid & OBD_MD_FLATIME) {
if (body->valid & OBD_MD_FLSIZE) {
if (exp_connect_som(ll_i2mdexp(inode)) &&
- S_ISREG(inode->i_mode) && lli->lli_has_smd) {
+ S_ISREG(inode->i_mode)) {
struct lustre_handle lockh;
ldlm_mode_t mode;
if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
/* discard all dirty pages before truncating them, required by
* osc_extent implementation at LU-1030. */
- cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_DISCARD);
+ cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
+ CL_FSYNC_DISCARD, 1);
truncate_inode_pages(&inode->i_data, 0);
RETURN(-ENOMEM);
}
oinfo.oi_md = lsm;
- oinfo.oi_oa->o_id = lsm->lsm_object_id;
- oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
+ oinfo.oi_oa->o_oi = lsm->lsm_oi;
oinfo.oi_oa->o_flags = flags;
oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
OBD_MD_FLGROUP;
obd->obd_force = 1;
OBD_ALLOC_PTR(ioc_data);
- if (ioc_data) {
- obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
- sizeof ioc_data, ioc_data, NULL);
+ if (ioc_data) {
+ obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
+ sizeof *ioc_data, ioc_data, NULL);
- obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
- sizeof ioc_data, ioc_data, NULL);
-
- OBD_FREE_PTR(ioc_data);
- }
+ obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
+ sizeof *ioc_data, ioc_data, NULL);
+ OBD_FREE_PTR(ioc_data);
+ }
/* Really, we'd like to wait until there are no requests outstanding,
* and then continue. For now, we just invalidate the requests,
return 0;
}
-int ll_prep_inode(struct inode **inode,
- struct ptlrpc_request *req,
- struct super_block *sb)
+int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
+ struct super_block *sb, struct lookup_intent *it)
{
struct ll_sb_info *sbi = NULL;
struct lustre_md md;
- __u64 ibits;
int rc;
ENTRY;
*/
LASSERT(fid_is_sane(&md.body->fid1));
- *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
+ *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
+ sbi->ll_flags & LL_SBI_32BIT_API),
+ &md);
if (*inode == NULL || IS_ERR(*inode)) {
- if (md.lsm)
- obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
#ifdef CONFIG_FS_POSIX_ACL
if (md.posix_acl) {
posix_acl_release(md.posix_acl);
}
}
- /* sanity check for LAYOUT lock. */
- ibits = MDS_INODELOCK_LAYOUT;
- if (S_ISREG(md.body->mode) && sbi->ll_flags & LL_SBI_LAYOUT_LOCK &&
- md.lsm != NULL && !ll_have_md_lock(*inode, &ibits, LCK_MINMODE)) {
- CERROR("%s: inode "DFID" (%p) layout lock not granted.\n",
- ll_get_fsname(sb, NULL, 0),
- PFID(ll_inode2fid(*inode)), *inode);
+ /* Handling piggyback layout lock.
+ * Layout lock can be piggybacked by getattr and open request.
+ * The lsm can be applied to inode only if it comes with a layout lock
+ * otherwise correct layout may be overwritten, for example:
+ * 1. proc1: mdt returns a lsm but not granting layout
+ * 2. layout was changed by another client
+ * 3. proc2: refresh layout and layout lock granted
+ * 4. proc1: to apply a stale layout */
+ if (it != NULL && it->d.lustre.it_lock_mode != 0) {
+ struct lustre_handle lockh;
+ struct ldlm_lock *lock;
+
+ lockh.cookie = it->d.lustre.it_lock_handle;
+ lock = ldlm_handle2lock(&lockh);
+ LASSERT(lock != NULL);
+ if (ldlm_has_layout(lock)) {
+ struct cl_object_conf conf;
+
+ memset(&conf, 0, sizeof(conf));
+ conf.coc_opc = OBJECT_CONF_SET;
+ conf.coc_inode = *inode;
+ conf.coc_lock = lock;
+ conf.u.coc_md = &md;
+ (void)ll_layout_conf(*inode, &conf);
+ }
+ LDLM_LOCK_PUT(lock);
}
out:
+ if (md.lsm != NULL)
+ obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
md_free_lustre_md(sbi->ll_md_exp, &md);
RETURN(rc);
}
op_data->op_capa2 = NULL;
}
- op_data->op_name = name;
- op_data->op_namelen = namelen;
- op_data->op_mode = mode;
- op_data->op_mod_time = cfs_time_current_sec();
- op_data->op_fsuid = cfs_curproc_fsuid();
- op_data->op_fsgid = cfs_curproc_fsgid();
- op_data->op_cap = cfs_curproc_cap_pack();
- op_data->op_bias = MDS_CHECK_SPLIT;
- op_data->op_opc = opc;
- op_data->op_mds = 0;
- op_data->op_data = data;
+ op_data->op_name = name;
+ op_data->op_namelen = namelen;
+ op_data->op_mode = mode;
+ op_data->op_mod_time = cfs_time_current_sec();
+ op_data->op_fsuid = cfs_curproc_fsuid();
+ op_data->op_fsgid = cfs_curproc_fsgid();
+ op_data->op_cap = cfs_curproc_cap_pack();
+ op_data->op_bias = 0;
+ op_data->op_cli_flags = 0;
+ if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
+ filename_is_volatile(name, namelen, NULL))
+ op_data->op_bias |= MDS_CREATE_VOLATILE;
+ op_data->op_opc = opc;
+ op_data->op_mds = 0;
+ op_data->op_data = data;
/* If the file is being opened after mknod() (normally due to NFS)
* try to use the default stripe data from parent directory for
/** We ignore parent's capability temporary. */
}
+ /* When called by ll_setattr_raw, file is i1. */
+ if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
+ op_data->op_bias |= MDS_DATA_MODIFIED;
+
return op_data;
}
struct dentry *dentry = NULL;
struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
- buf = (char *)__get_free_page(GFP_KERNEL);
+ /* this can be called inside spin lock so use GFP_ATOMIC. */
+ buf = (char *)__get_free_page(GFP_ATOMIC);
if (buf != NULL) {
dentry = d_find_alias(page->mapping->host);
if (dentry != NULL)