#define DEBUG_SUBSYSTEM S_LLITE
#include <linux/module.h>
+#include <linux/types.h>
#include <linux/random.h>
#include <linux/version.h>
#include <linux/lustre_ha.h>
#include <linux/lustre_dlm.h>
#include <linux/lprocfs_status.h>
+#include <linux/lustre_snap.h>
#include "llite_internal.h"
kmem_cache_t *ll_file_data_slab;
extern struct address_space_operations ll_aops;
extern struct address_space_operations ll_dir_aops;
-extern struct super_operations ll_super_operations;
#ifndef log2
#define log2(n) ffz(~(n))
#endif
-struct ll_sb_info *lustre_init_sbi(struct super_block *sb)
+struct ll_sb_info *lustre_init_sbi(struct super_block *sb)
{
struct ll_sb_info *sbi = NULL;
class_uuid_t uuid;
if (!sbi)
RETURN(NULL);
- spin_lock_init(&sbi->ll_pglist_lock);
+ spin_lock_init(&sbi->ll_lock);
INIT_LIST_HEAD(&sbi->ll_pglist);
sbi->ll_pglist_gen = 0;
+ sbi->ll_max_read_ahead_pages = SBI_DEFAULT_RA_MAX;
INIT_LIST_HEAD(&sbi->ll_conn_chain);
INIT_HLIST_HEAD(&sbi->ll_orphan_dentry_list);
- ll_s2sbi(sb) = sbi;
+ INIT_LIST_HEAD(&sbi->ll_mnt_list);
+ sema_init(&sbi->ll_gns_sem, 1);
+ init_completion(&sbi->ll_gns_completion);
+ sbi->ll_gns_state = LL_GNS_STATE_IDLE;
+ sbi->ll_gns_timer.data = (unsigned long)sbi;
+ sbi->ll_gns_timer.function = ll_gns_timer_callback;
+ init_timer(&sbi->ll_gns_timer);
+ INIT_LIST_HEAD(&sbi->ll_gns_sbi_head);
+
+ ll_set_sbi(sb, sbi);
generate_random_uuid(uuid);
class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
RETURN(sbi);
}
-void lustre_free_sbi(struct super_block *sb)
+void lustre_free_sbi(struct super_block *sb)
{
struct ll_sb_info *sbi = ll_s2sbi(sb);
ENTRY;
- if (sbi != NULL)
+ if (sbi != NULL) {
+ list_del(&sbi->ll_gns_sbi_head);
+ del_timer(&sbi->ll_gns_timer);
OBD_FREE(sbi, sizeof(*sbi));
- ll_s2sbi(sb) = NULL;
+ }
+ ll_set_sbi(sb, NULL);
EXIT;
}
+int lustre_init_ea_size(struct ll_sb_info *sbi)
+{
+ struct lov_desc desc;
+ int rc, valsize;
+
+ /* OSC may have larger ea size */
+ valsize = sizeof(desc);
+ rc = obd_get_info(sbi->ll_osc_exp, strlen("lovdesc") + 1, "lovdesc",
+ &valsize, &desc);
+ if (rc)
+ RETURN(rc);
+ obd_init_ea_size(sbi->ll_mdc_exp, obd_size_diskmd(sbi->ll_osc_exp, NULL),
+ desc.ld_tgt_count*sizeof(struct llog_cookie));
+
+ RETURN(rc);
+}
+
int lustre_common_fill_super(struct super_block *sb, char *mdc, char *osc)
{
struct inode *root = 0;
CERROR("could not register mount in /proc/lustre");
}
- mdc_init_ea_size(obd, osc);
-
- err = obd_connect(&mdc_conn, obd, &sbi->ll_sb_uuid);
+ err = obd_connect(&mdc_conn, obd, &sbi->ll_sb_uuid, 0);
if (err == -EBUSY) {
CERROR("An MDS (mdc %s) is performing recovery, of which this"
" client is not a part. Please wait for recovery to "
sb->s_blocksize_bits = log2(osfs.os_bsize);
sb->s_magic = LL_SUPER_MAGIC;
sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
-
- devno = get_uuid2int(sbi2mdc(sbi)->cl_import->imp_target_uuid.uuid,
- strlen(sbi2mdc(sbi)->cl_import->imp_target_uuid.uuid));
+
+ devno = get_uuid2int(sbi->ll_mdc_exp->exp_obd->obd_uuid.uuid,
+ strlen(sbi->ll_mdc_exp->exp_obd->obd_uuid.uuid));
+
sb->s_dev = devno;
obd = class_name2obd(osc);
GOTO(out_mdc, err);
}
- err = obd_connect(&osc_conn, obd, &sbi->ll_sb_uuid);
+ err = obd_connect(&osc_conn, obd, &sbi->ll_sb_uuid, 0);
if (err == -EBUSY) {
CERROR("An OST (osc %s) is performing recovery, of which this"
" client is not a part. Please wait for recovery to "
}
sbi->ll_osc_exp = class_conn2export(&osc_conn);
- err = mdc_getstatus(sbi->ll_mdc_exp, &rootfid);
+ lustre_init_ea_size(sbi);
+
+ err = md_getstatus(sbi->ll_mdc_exp, &rootfid);
if (err) {
CERROR("cannot mds_connect: rc = %d\n", err);
GOTO(out_osc, err);
sb->s_op = &lustre_super_operations;
- /* make root inode
+ /* make root inode
* XXX: move this to after cbd setup? */
- err = mdc_getattr(sbi->ll_mdc_exp, &rootfid,
- OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
+ err = md_getattr(sbi->ll_mdc_exp, &rootfid,
+ OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS, 0, &request);
if (err) {
- CERROR("mdc_getattr failed for root: rc = %d\n", err);
+ CERROR("md_getattr failed for root: rc = %d\n", err);
GOTO(out_osc, err);
}
- err = mdc_req2lustre_md(request, 0, sbi->ll_osc_exp, &md);
+ err = mdc_req2lustre_md(sbi->ll_mdc_exp, request, 0,
+ sbi->ll_osc_exp, &md);
if (err) {
CERROR("failed to understand root inode md: rc = %d\n",err);
ptlrpc_req_finished (request);
GOTO(out_root, err);
}
+ ll_gns_add_timer(sbi);
+
+ /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
+ backing dev info assigned to inode mapping is used for
+ determining maximal readahead. */
+#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
+ /* bug 2805 - set VM readahead to zero */
+ vm_max_readahead = vm_min_readahead = 0;
+#endif
+
sb->s_root = d_alloc_root(root);
+
+#ifdef S_PDIROPS
+ CWARN("Enabling PDIROPS\n");
+ sb->s_flags |= S_PDIROPS;
+#endif
+
RETURN(err);
out_root:
struct hlist_node *tmp, *next;
ENTRY;
+ ll_gns_del_timer(sbi);
+
ll_close_thread_shutdown(sbi->ll_lcq);
list_del(&sbi->ll_conn_chain);
spin_lock(&dcache_lock);
hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
+ CWARN("orphan dentry %*s (%p) at unmount\n",
+ dentry->d_name.len, dentry->d_name.name, dentry);
shrink_dcache_parent(dentry);
}
spin_unlock(&dcache_lock);
RETURN(fl);
}
-void ll_options(char *options, char **ost, char **mdc, int *flags)
+void ll_options(char *options, char **ost, char **mdc, int *flags,
+ char **clone_opts)
{
char *this_char;
#if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
continue;
if (!*mdc && (*mdc = ll_read_opt("mdc", this_char)))
continue;
+ if (!*clone_opts && (*clone_opts = ll_read_opt("clone",
+ this_char)))
+ continue;
if (!(*flags & LL_SBI_NOLCK) &&
((*flags) = (*flags) |
ll_set_opt("nolock", this_char,
LL_SBI_NOLCK)))
continue;
}
+
EXIT;
}
lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
spin_lock_init(&lli->lli_lock);
INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
+ lli->lli_inode_magic = LLI_INODE_MAGIC;
}
int ll_fill_super(struct super_block *sb, void *data, int silent)
struct ll_sb_info *sbi;
char *osc = NULL;
char *mdc = NULL;
+ char *clone_opts = NULL;
int err;
ENTRY;
RETURN(-ENOMEM);
sbi->ll_flags |= LL_SBI_READAHEAD;
- ll_options(data, &osc, &mdc, &sbi->ll_flags);
+ ll_options(data, &osc, &mdc, &sbi->ll_flags, &clone_opts);
if (!osc) {
CERROR("no osc\n");
OBD_FREE(mdc, strlen(mdc) + 1);
if (osc)
OBD_FREE(osc, strlen(osc) + 1);
+ if (clone_opts)
+ OBD_FREE(clone_opts, strlen(clone_opts) + 1);
RETURN(err);
} /* ll_read_super */
-void ll_put_super(struct super_block *sb)
-{
- ENTRY;
-
- CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
-
- lustre_common_put_super(sb);
-
- lustre_free_sbi(sb);
-
- EXIT;
-} /* ll_put_super */
-
-int lustre_process_log(struct lustre_mount_data *lmd, char * profile,
- struct config_llog_instance *cfg)
+static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
+ struct config_llog_instance *cfg, int allow_recov)
{
struct lustre_cfg lcfg;
struct portals_cfg pcfg;
int err;
ENTRY;
+ if (lmd_bad_magic(lmd))
+ RETURN(-EINVAL);
+
generate_random_uuid(uuid);
class_uuid_unparse(uuid, &mdc_uuid);
PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
pcfg.pcfg_nal = lmd->lmd_nal;
pcfg.pcfg_nid = lmd->lmd_local_nid;
- err = kportal_nal_cmd(&pcfg);
+ err = libcfs_nal_cmd(&pcfg);
if (err <0)
GOTO(out, err);
}
pcfg.pcfg_misc = lmd->lmd_port;
pcfg.pcfg_size = 8388608;
pcfg.pcfg_flags = 0x4; /*share*/
- err = kportal_nal_cmd(&pcfg);
+ err = libcfs_nal_cmd(&pcfg);
if (err <0)
GOTO(out, err);
}
err = class_process_config(&lcfg);
if (err < 0)
GOTO(out_detach, err);
-
+
obd = class_name2obd(name);
if (obd == NULL)
GOTO(out_cleanup, err = -EINVAL);
- err = obd_connect(&mdc_conn, obd, &mdc_uuid);
+ /* Disable initial recovery on this import */
+ err = obd_set_info(obd->obd_self_export,
+ strlen("initial_recov"), "initial_recov",
+ sizeof(allow_recov), &allow_recov);
+ if (err)
+ GOTO(out_cleanup, err);
+
+ err = obd_connect(&mdc_conn, obd, &mdc_uuid, 0);
if (err) {
CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, err);
GOTO(out_cleanup, err);
}
-
+
exp = class_conn2export(&mdc_conn);
-
- ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
- rc = class_config_parse_llog(ctxt, profile, cfg);
- if (rc) {
- CERROR("class_config_parse_llog failed: rc = %d\n", rc);
- }
+
+ ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
+ rc = class_config_process_llog(ctxt, profile, cfg);
+ if (rc)
+ CERROR("class_config_process_llog failed: rc = %d\n", rc);
err = obd_disconnect(exp, 0);
pcfg.pcfg_nid = lmd->lmd_server_nid;
pcfg.pcfg_id = lmd->lmd_server_ipaddr;
pcfg.pcfg_flags = 1; /*share*/
- err = kportal_nal_cmd(&pcfg);
+ err = libcfs_nal_cmd(&pcfg);
if (err <0)
GOTO(out, err);
}
out:
if (rc == 0)
rc = err;
-
+
RETURN(rc);
}
ENTRY;
CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
+ if (lmd_bad_magic(lmd))
+ RETURN(-EINVAL);
+
sbi = lustre_init_sbi(sb);
if (!sbi)
RETURN(-ENOMEM);
struct config_llog_instance cfg;
int len;
- if (!lmd->lmd_mds) {
+ if (lmd->lmd_mds[0] == '\0') {
CERROR("no mds name\n");
GOTO(out_free, err = -EINVAL);
}
OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
- if (sbi->ll_lmd == NULL)
+ if (sbi->ll_lmd == NULL)
GOTO(out_free, err = -ENOMEM);
memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
/* generate a string unique to this super, let's try
the address of the super itself.*/
- len = (sizeof(sb) * 2) + 1;
+ len = (sizeof(sb) * 2) + 1;
OBD_ALLOC(sbi->ll_instance, len);
- if (sbi->ll_instance == NULL)
+ if (sbi->ll_instance == NULL)
GOTO(out_free, err = -ENOMEM);
sprintf(sbi->ll_instance, "%p", sb);
cfg.cfg_instance = sbi->ll_instance;
cfg.cfg_uuid = sbi->ll_sb_uuid;
cfg.cfg_local_nid = lmd->lmd_local_nid;
- err = lustre_process_log(lmd, lmd->lmd_profile, &cfg);
+ err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
if (err < 0) {
CERROR("Unable to process log: %s\n", lmd->lmd_profile);
-
GOTO(out_free, err);
}
}
if (osc)
OBD_FREE(osc, strlen(osc) + 1);
- OBD_ALLOC(osc, strlen(lprof->lp_osc) +
+ OBD_ALLOC(osc, strlen(lprof->lp_osc) +
strlen(sbi->ll_instance) + 2);
sprintf(osc, "%s-%s", lprof->lp_osc, sbi->ll_instance);
if (mdc)
OBD_FREE(mdc, strlen(mdc) + 1);
- OBD_ALLOC(mdc, strlen(lprof->lp_mdc) +
+ OBD_ALLOC(mdc, strlen(lprof->lp_mdc) +
strlen(sbi->ll_instance) + 2);
sprintf(mdc, "%s-%s", lprof->lp_mdc, sbi->ll_instance);
}
CERROR("no mdc\n");
GOTO(out_free, err = -EINVAL);
}
-
+
err = lustre_common_fill_super(sb, mdc, osc);
-
+
if (err)
GOTO(out_free, err);
-
+
out_dev:
if (mdc)
OBD_FREE(mdc, strlen(mdc) + 1);
int err;
if (sbi->ll_instance != NULL) {
+ struct lustre_mount_data *lmd = sbi->ll_lmd;
char * cln_prof;
struct config_llog_instance cfg;
cfg.cfg_uuid = sbi->ll_sb_uuid;
OBD_ALLOC(cln_prof, len);
- sprintf(cln_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
+ sprintf(cln_prof, "%s-clean", lmd->lmd_profile);
- err = lustre_process_log(sbi->ll_lmd, cln_prof, &cfg);
- if (err < 0)
+ err = lustre_process_log(lmd, cln_prof, &cfg, 0);
+ if (err < 0)
CERROR("Unable to process log: %s\n", cln_prof);
OBD_FREE(cln_prof, len);
OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance)+ 1);
goto out_dev;
} /* lustre_fill_super */
+static void lustre_manual_cleanup(struct ll_sb_info *sbi)
+{
+ struct lustre_cfg lcfg;
+ struct obd_device *obd;
+ int next = 0;
+
+ while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
+ {
+ int err;
+
+ LCFG_INIT(lcfg, LCFG_CLEANUP, obd->obd_name);
+ err = class_process_config(&lcfg);
+ if (err) {
+ CERROR("cleanup failed: %s\n", obd->obd_name);
+ //continue;
+ }
+
+ LCFG_INIT(lcfg, LCFG_DETACH, obd->obd_name);
+ err = class_process_config(&lcfg);
+ if (err) {
+ CERROR("detach failed: %s\n", obd->obd_name);
+ //continue;
+ }
+ }
+
+ if (sbi->ll_lmd != NULL)
+ class_del_profile(sbi->ll_lmd->lmd_profile);
+}
+
void lustre_put_super(struct super_block *sb)
{
+ struct obd_device *obd;
struct ll_sb_info *sbi = ll_s2sbi(sb);
+ int force_umount = 0;
ENTRY;
CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
+ obd = class_exp2obd(sbi->ll_mdc_exp);
+ if (obd)
+ force_umount = obd->obd_no_recov;
+ obd = NULL;
lustre_common_put_super(sb);
-
if (sbi->ll_lmd != NULL) {
char * cln_prof;
int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
int err;
struct config_llog_instance cfg;
+ if (force_umount) {
+ CERROR("force umount, doing manual cleanup\n");
+ lustre_manual_cleanup(sbi);
+ GOTO(free_lmd, 0);
+ }
+
cfg.cfg_instance = sbi->ll_instance;
cfg.cfg_uuid = sbi->ll_sb_uuid;
OBD_ALLOC(cln_prof, len);
sprintf(cln_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
- err = lustre_process_log(sbi->ll_lmd, cln_prof, &cfg);
- if (err < 0)
- CERROR("Unable to process log: %s\n", cln_prof);
+ err = lustre_process_log(sbi->ll_lmd, cln_prof, &cfg, 0);
+ if (err < 0) {
+ CERROR("Unable to process log: %s, doing manual cleanup"
+ "\n", cln_prof);
+ lustre_manual_cleanup(sbi);
+ }
OBD_FREE(cln_prof, len);
+ free_lmd:
OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
}
EXIT;
} /* lustre_put_super */
+int ll_process_config_update(struct ll_sb_info *sbi, int clean)
+{
+ struct obd_export *mdc_exp = sbi->ll_mdc_exp;
+ struct lustre_mount_data *lmd = sbi->ll_lmd;
+ struct llog_ctxt *ctxt;
+ struct config_llog_instance cfg;
+ char *profile = lmd->lmd_profile, *name = NULL;
+ int rc, namelen = 0, version;
+ ENTRY;
+
+ if (profile == NULL)
+ RETURN(0);
+ if (lmd == NULL) {
+ CERROR("Client not mounted with zero-conf; cannot process "
+ "update log.\n");
+ RETURN(0);
+ }
+
+ rc = ldlm_cli_cancel_unused(mdc_exp->exp_obd->obd_namespace, NULL,
+ LDLM_FL_CONFIG_CHANGE, NULL);
+ if (rc != 0)
+ CWARN("ldlm_cli_cancel_unused(mdc): %d\n", rc);
+
+ rc = obd_cancel_unused(sbi->ll_osc_exp, NULL, LDLM_FL_CONFIG_CHANGE,
+ NULL);
+ if (rc != 0)
+ CWARN("obd_cancel_unused(lov): %d\n", rc);
+
+ cfg.cfg_instance = sbi->ll_instance;
+ cfg.cfg_uuid = sbi->ll_sb_uuid;
+ cfg.cfg_local_nid = lmd->lmd_local_nid;
+
+ namelen = strlen(profile) + 20; /* -clean-######### */
+ OBD_ALLOC(name, namelen);
+ if (name == NULL)
+ RETURN(-ENOMEM);
+
+ if (clean) {
+ version = sbi->ll_config_version - 1;
+ sprintf(name, "%s-clean-%d", profile, version);
+ } else {
+ version = sbi->ll_config_version + 1;
+ sprintf(name, "%s-%d", profile, version);
+ }
+
+ CWARN("Applying configuration log %s\n", name);
+
+ ctxt = llog_get_context(&mdc_exp->exp_obd->obd_llogs,
+ LLOG_CONFIG_REPL_CTXT);
+ rc = class_config_process_llog(ctxt, name, &cfg);
+ if (rc == 0)
+ sbi->ll_config_version = version;
+ CWARN("Finished applying configuration log %s: %d\n", name, rc);
+
+ if (rc == 0 && clean == 0) {
+ struct lov_desc desc;
+ int rc, valsize;
+ valsize = sizeof(desc);
+ rc = obd_get_info(sbi->ll_osc_exp, strlen("lovdesc") + 1,
+ "lovdesc", &valsize, &desc);
+
+ rc = obd_init_ea_size(mdc_exp,
+ obd_size_diskmd(sbi->ll_osc_exp, NULL),
+ (desc.ld_tgt_count *
+ sizeof(struct llog_cookie)));
+ }
+ OBD_FREE(name, namelen);
+ RETURN(rc);
+}
struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
{
- struct inode *inode;
+ struct inode *inode = NULL;
l_lock(&lock->l_resource->lr_namespace->ns_lock);
- if (lock->l_ast_data)
- inode = igrab(lock->l_ast_data);
- else
- inode = NULL;
+ if (lock->l_ast_data) {
+ struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
+ if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
+ inode = igrab(lock->l_ast_data);
+ } else {
+ CERROR("DEBUG: l_ast_data %p is bogus: magic %x\n",
+ lock->l_ast_data, lli->lli_inode_magic);
+ }
+ }
l_unlock(&lock->l_resource->lr_namespace->ns_lock);
return inode;
}
-static int null_if_equal(struct ldlm_lock *lock, void *data)
+int null_if_equal(struct ldlm_lock *lock, void *data)
{
- if (data == lock->l_ast_data)
+ if (data == lock->l_ast_data) {
lock->l_ast_data = NULL;
- if (lock->l_req_mode != lock->l_granted_mode)
- return LDLM_ITER_STOP;
+ if (lock->l_req_mode != lock->l_granted_mode)
+ LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
+ }
return LDLM_ITER_CONTINUE;
}
CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
inode->i_generation, inode);
+ lli->lli_inode_magic = LLI_INODE_DEAD;
ll_inode2fid(&fid, inode);
clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
- mdc_change_cbdata(sbi->ll_mdc_exp, &fid, null_if_equal, inode);
+ md_change_cbdata(sbi->ll_mdc_exp, &fid, null_if_equal, inode);
if (lli->lli_smd)
obd_change_cbdata(sbi->ll_osc_exp, lli->lli_smd,
lli->lli_smd = NULL;
}
+ if (lli->lli_mea) {
+ /* FIXME: change cbdata for mea here */
+ obd_free_memmd(sbi->ll_mdc_exp,
+ (struct lov_stripe_md **) &lli->lli_mea);
+ lli->lli_mea = NULL;
+ }
+
if (lli->lli_symlink_name) {
OBD_FREE(lli->lli_symlink_name,
strlen(lli->lli_symlink_name) + 1);
struct lustre_md md;
ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
- rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
- attr, NULL, 0, NULL, 0, &request);
-
+ rc = md_setattr(sbi->ll_mdc_exp, &op_data,
+ attr, NULL, 0, NULL, 0, &request);
if (rc) {
ptlrpc_req_finished(request);
if (rc != -EPERM && rc != -EACCES)
- CERROR("mdc_setattr fails: rc = %d\n", rc);
+ CERROR("md_setattr fails: rc = %d\n", rc);
RETURN(rc);
}
- rc = mdc_req2lustre_md(request, 0, sbi->ll_osc_exp, &md);
+ rc = mdc_req2lustre_md(sbi->ll_mdc_exp, request, 0,
+ sbi->ll_osc_exp, &md);
if (rc) {
ptlrpc_req_finished(request);
RETURN(rc);
/* Won't invoke vmtruncate as we already cleared ATTR_SIZE,
* but needed to set timestamps backwards on utime. */
inode_setattr(inode, attr);
- ll_update_inode(inode, md.body, md.lsm);
+ ll_update_inode(inode, &md);
ptlrpc_req_finished(request);
if (!lsm || !S_ISREG(inode->i_mode)) {
/* from sys_utime() */
if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
if (current->fsuid != inode->i_uid &&
- (rc = ll_permission(inode, MAY_WRITE, NULL)) != 0)
+ (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
RETURN(rc);
} else {
- /* from inode_change_ok() */
- if (current->fsuid != inode->i_uid &&
- !capable(CAP_FOWNER))
- RETURN(-EPERM);
+ /* from inode_change_ok() */
+ if (current->fsuid != inode->i_uid &&
+ !capable(CAP_FOWNER))
+ RETURN(-EPERM);
}
}
* If we don't we can race with other i_size updaters on our node, like
* ll_file_read. We can also race with i_size propogation to other
* nodes through dirtying and writeback of final cached pages. This
- * last one is especially bad for racing o_append users on other
+ * last one is especially bad for racing o_append users on other
* nodes. */
if (ia_valid & ATTR_SIZE) {
- struct ldlm_extent extent = { .start = attr->ia_size,
- .end = OBD_OBJECT_EOF };
+ ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
+ OBD_OBJECT_EOF } };
struct lustre_handle lockh = { 0 };
int err, ast_flags = 0;
/* XXX when we fix the AST intents to pass the discard-range
/* bug 1639: avoid write/truncate i_sem/DLM deadlock */
LASSERT(atomic_read(&inode->i_sem.count) <= 0);
up(&inode->i_sem);
- rc = ll_extent_lock_no_validate(NULL, inode, lsm, LCK_PW,
- &extent, &lockh, ast_flags);
+ rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
+ ast_flags);
down(&inode->i_sem);
- if (rc != ELDLM_OK)
+ if (rc != 0)
RETURN(rc);
rc = vmtruncate(inode, attr->ia_size);
- if (rc == 0)
- set_bit(LLI_F_HAVE_OST_SIZE_LOCK,
- &ll_i2info(inode)->lli_flags);
-
- //ll_try_done_writing(inode);
+ /* We need to drop the semaphore here, because this unlock may
+ * result in a cancellation, which will need the i_sem */
+ up(&inode->i_sem);
/* unlock now as we don't mind others file lockers racing with
* the mds updates below? */
err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
+ down(&inode->i_sem);
if (err) {
CERROR("ll_extent_unlock failed: %d\n", err);
if (!rc)
CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
inode->i_ino, LTIME_S(attr->ia_mtime));
oa.o_id = lsm->lsm_object_id;
- oa.o_valid = OBD_MD_FLID;
+ oa.o_gr = lsm->lsm_object_gr;
+ oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
obdo_from_inode(&oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
OBD_MD_FLMTIME | OBD_MD_FLCTIME);
rc = obd_setattr(sbi->ll_osc_exp, &oa, lsm, NULL);
RETURN(rc);
}
+ osfs->os_type = sb->s_magic;
+
CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
return 0;
}
-void dump_lsm(int level, struct lov_stripe_md *lsm)
-{
- CDEBUG(level, "objid "LPX64", maxbytes "LPX64", magic 0x%08X, "
- "stripe_size %u, stripe_count %u\n",
- lsm->lsm_object_id, lsm->lsm_maxbytes, lsm->lsm_magic,
- lsm->lsm_stripe_size, lsm->lsm_stripe_count);
-}
-
-void ll_update_inode(struct inode *inode, struct mds_body *body,
- struct lov_stripe_md *lsm)
+void ll_update_inode(struct inode *inode, struct lustre_md *md)
{
struct ll_inode_info *lli = ll_i2info(inode);
+ struct mds_body *body = md->body;
+ struct lov_stripe_md *lsm = md->lsm;
+ struct mea *mea = md->mea;
+ ENTRY;
- LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
+ LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
+ LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
if (lsm != NULL) {
+ LASSERT(lsm->lsm_object_gr > 0);
if (lli->lli_smd == NULL) {
lli->lli_smd = lsm;
lli->lli_maxbytes = lsm->lsm_maxbytes;
if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
} else {
+ int i;
if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
CERROR("lsm mismatch for inode %ld\n",
inode->i_ino);
dump_lsm(D_ERROR, lsm);
LBUG();
}
+ /* XXX FIXME -- We should decide on a safer (atomic) and
+ * more elegant way to update the lsm */
+ for (i = 0; i < lsm->lsm_stripe_count; i++) {
+ lli->lli_smd->lsm_oinfo[i].loi_id =
+ lsm->lsm_oinfo[i].loi_id;
+ lli->lli_smd->lsm_oinfo[i].loi_gr =
+ lsm->lsm_oinfo[i].loi_gr;
+ lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
+ lsm->lsm_oinfo[i].loi_ost_idx;
+ lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
+ lsm->lsm_oinfo[i].loi_ost_gen;
+ }
}
+ /* bug 2844 - limit i_blksize for broken user-space apps */
+ LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
+ inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
if (lli->lli_smd != lsm)
obd_free_memmd(ll_i2obdexp(inode), &lsm);
}
+ if (mea != NULL) {
+ if (lli->lli_mea == NULL) {
+ lli->lli_mea = mea;
+ } else {
+ if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
+ CERROR("mea mismatch for inode %lu\n",
+ inode->i_ino);
+ LBUG();
+ }
+ }
+ if (lli->lli_mea != mea)
+ obd_free_memmd(ll_i2mdcexp(inode),
+ (struct lov_stripe_md **) &mea);
+
+ }
+
if (body->valid & OBD_MD_FLID)
inode->i_ino = body->ino;
if (body->valid & OBD_MD_FLATIME)
if (body->valid & OBD_MD_FLSIZE)
set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
+
+ lli->lli_mds = body->mds;
+ inode->i_dev = (kdev_t) body->mds;
+ LASSERT(body->mds < 1000);
}
+#if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
+static struct backing_dev_info ll_backing_dev_info = {
+ .ra_pages = 0, /* No readahead */
+ .memory_backed = 0, /* Does contribute to dirty memory */
+};
+#endif
+
void ll_read_inode2(struct inode *inode, void *opaque)
{
struct lustre_md *md = opaque;
LTIME_S(inode->i_mtime) = 0;
LTIME_S(inode->i_atime) = 0;
LTIME_S(inode->i_ctime) = 0;
- ll_update_inode(inode, md->body, md->lsm);
+
+ inode->i_rdev = 0;
+ ll_update_inode(inode, md);
/* OIDEBUG(inode); */
EXIT;
} else {
inode->i_op = &ll_special_inode_operations;
+
#if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
init_special_inode(inode, inode->i_mode,
kdev_t_to_nr(inode->i_rdev));
+
+ /* initializing backing dev info. */
+ inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
#else
init_special_inode(inode, inode->i_mode, inode->i_rdev);
#endif
+ lli->ll_save_ifop = inode->i_fop;
+
+ if (S_ISCHR(inode->i_mode))
+ inode->i_fop = &ll_special_chr_inode_fops;
+ else if (S_ISBLK(inode->i_mode))
+ inode->i_fop = &ll_special_blk_inode_fops;
+ else if (S_ISFIFO(inode->i_mode))
+ inode->i_fop = &ll_special_fifo_inode_fops;
+ else if (S_ISSOCK(inode->i_mode))
+ inode->i_fop = &ll_special_sock_inode_fops;
+
+ CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
+ inode->i_fop);
+
+ if (lli->ll_save_ifop->owner) {
+ CWARN("%p has owner %p\n", lli->ll_save_ifop,
+ lli->ll_save_ifop->owner);
+ }
EXIT;
}
}
+void ll_delete_inode(struct inode *inode)
+{
+ int rc;
+ struct ll_fid fid;
+ struct ll_sb_info *sbi = ll_i2sbi(inode);
+ ENTRY;
+
+ ll_inode2fid(&fid, inode);
+
+ rc = md_delete_object(sbi->ll_mdc_exp, &fid);
+ if (rc) {
+ CERROR("md_delete_object() failed, error %d.\n",
+ rc);
+ }
+
+ clear_inode(inode);
+ EXIT;
+}
+
int ll_iocontrol(struct inode *inode, struct file *file,
unsigned int cmd, unsigned long arg)
{
struct mds_body *body;
ll_inode2fid(&fid, inode);
- rc = mdc_getattr(sbi->ll_mdc_exp, &fid, valid, 0, &req);
+ rc = md_getattr(sbi->ll_mdc_exp, &fid, valid, 0, &req);
if (rc) {
CERROR("failure %d inode %lu\n", rc, inode->i_ino);
RETURN(-abs(rc));
attr.ia_attr_flags = flags;
attr.ia_valid |= ATTR_ATTR_FLAG;
- rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
- &attr, NULL, 0, NULL, 0, &req);
+ rc = md_setattr(sbi->ll_mdc_exp, &op_data,
+ &attr, NULL, 0, NULL, 0, &req);
if (rc) {
ptlrpc_req_finished(req);
if (rc != -EPERM && rc != -EACCES)
- CERROR("mdc_setattr fails: rc = %d\n", rc);
+ CERROR("md_setattr fails: rc = %d\n", rc);
obdo_free(oa);
RETURN(rc);
}
ptlrpc_req_finished(req);
oa->o_id = lsm->lsm_object_id;
+ oa->o_gr = lsm->lsm_object_gr;
oa->o_flags = flags;
- oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS;
+ oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
rc = obd_setattr(sbi->ll_osc_exp, oa, lsm, NULL);
obdo_free(oa);
if (rc) {
if (rc != -EPERM && rc != -EACCES)
- CERROR("mdc_setattr fails: rc = %d\n", rc);
+ CERROR("md_setattr fails: rc = %d\n", rc);
RETURN(rc);
}
EXIT;
}
-int ll_prep_inode(struct obd_export *exp, struct inode **inode,
- struct ptlrpc_request *req, int offset,struct super_block *sb)
+int ll_prep_inode(struct obd_export *osc_exp, struct obd_export *mdc_exp,
+ struct inode **inode, struct ptlrpc_request *req,
+ int offset, struct super_block *sb)
{
struct lustre_md md;
int rc = 0;
- rc = mdc_req2lustre_md(req, offset, exp, &md);
+ rc = mdc_req2lustre_md(mdc_exp, req, offset, osc_exp, &md);
if (rc)
RETURN(rc);
if (*inode) {
- ll_update_inode(*inode, md.body, md.lsm);
+ ll_update_inode(*inode, &md);
} else {
LASSERT(sb);
*inode = ll_iget(sb, md.body->ino, &md);
if (*inode == NULL || is_bad_inode(*inode)) {
/* free the lsm if we allocated one above */
if (md.lsm != NULL)
- obd_free_memmd(exp, &md.lsm);
+ obd_free_memmd(osc_exp, &md.lsm);
+ if (md.mea != NULL)
+ obd_free_memmd(mdc_exp,
+ (struct lov_stripe_md**)&md.mea);
rc = -ENOMEM;
CERROR("new_inode -fatal: rc %d\n", rc);
}