#include <linux/dcache.h>
#include <linux/init.h>
#include <linux/version.h>
+#include <linux/jbd.h>
+#include <linux/ext3_fs.h>
#if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
# include <linux/mount.h>
# include <linux/buffer_head.h>
+# include <linux/bio.h>
#endif
#include <linux/obd_class.h>
#include <linux/obd_lov.h>
+#include <linux/obd_ost.h>
#include <linux/lustre_dlm.h>
#include <linux/lustre_fsfilt.h>
#include <linux/lprocfs_status.h>
#include <linux/lustre_log.h>
#include <linux/lustre_commit_confd.h>
+#include <libcfs/list.h>
+#include <linux/lustre_smfs.h>
+#include <linux/lustre_sec.h>
#include "filter_internal.h"
+/* Group 0 is no longer a legal group, to catch uninitialized IDs */
+#define FILTER_MIN_GROUPS 3
+
static struct lvfs_callback_ops filter_lvfs_ops;
static int filter_destroy(struct obd_export *exp, struct obdo *oa,
struct lov_stripe_md *ea, struct obd_trans_info *);
+struct obd_llogs *filter_grab_llog_for_group(struct obd_device *,
+ int, struct obd_export *);
static void filter_commit_cb(struct obd_device *obd, __u64 transno,
void *cb_data, int error)
obd_transno_commit_cb(obd, transno, error);
}
-
/* Assumes caller has already pushed us into the kernel context. */
int filter_finish_transno(struct obd_export *exp, struct obd_trans_info *oti,
int rc)
filter->fo_fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
spin_unlock(&filter->fo_translock);
oti->oti_transno = last_rcvd;
- } else {
+ } else {
spin_lock(&filter->fo_translock);
last_rcvd = oti->oti_transno;
if (last_rcvd > le64_to_cpu(filter->fo_fsd->fsd_last_transno))
cpu_to_le64(last_rcvd);
spin_unlock(&filter->fo_translock);
}
+
fcd->fcd_last_rcvd = cpu_to_le64(last_rcvd);
- fcd->fcd_mount_count = filter->fo_fsd->fsd_mount_count;
/* could get xid from oti, if it's ever needed */
fcd->fcd_last_xid = 0;
off = fed->fed_lr_off;
- fsfilt_add_journal_cb(exp->exp_obd, last_rcvd, oti->oti_handle,
- filter_commit_cb, NULL);
+
+ fsfilt_add_journal_cb(exp->exp_obd, filter->fo_sb, last_rcvd,
+ oti->oti_handle, filter_commit_cb, NULL);
+
err = fsfilt_write_record(exp->exp_obd, filter->fo_rcvd_filp, fcd,
sizeof(*fcd), &off, 0);
if (err) {
LASSERT(bitmap != NULL);
/* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
- if (!strcmp(fed->fed_fcd->fcd_uuid, obd->obd_uuid.uuid))
+ if (!strcmp((char *)fed->fed_fcd->fcd_uuid, (char *)obd->obd_uuid.uuid))
RETURN(0);
/* the bitmap operations can handle cl_idx > sizeof(long) * 8, so
fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
if (new_client) {
- struct obd_run_ctxt saved;
+ struct lvfs_run_ctxt saved;
loff_t off = fed->fed_lr_off;
int err;
void *handle;
CDEBUG(D_INFO, "writing client fcd at idx %u (%llu) (len %u)\n",
fed->fed_lr_idx,off,(unsigned int)sizeof(*fed->fed_fcd));
- push_ctxt(&saved, &obd->obd_ctxt, NULL);
+ push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
/* Transaction needed to fix bug 1403 */
handle = fsfilt_start(obd,
filter->fo_rcvd_filp->f_dentry->d_inode,
fed->fed_fcd,
sizeof(*fed->fed_fcd),
&off, 1);
- fsfilt_commit(obd,
+ fsfilt_commit(obd, filter->fo_sb,
filter->fo_rcvd_filp->f_dentry->d_inode,
handle, 1);
}
- pop_ctxt(&saved, &obd->obd_ctxt, NULL);
+ pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
if (err) {
CERROR("error writing %s client idx %u: rc %d\n",
struct filter_obd *filter = &exp->exp_obd->u.filter;
struct obd_device *obd = exp->exp_obd;
struct filter_client_data zero_fcd;
- struct obd_run_ctxt saved;
+ struct lvfs_run_ctxt saved;
int rc;
loff_t off;
ENTRY;
GOTO(free, 0);
/* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
- if (strcmp(fed->fed_fcd->fcd_uuid, obd->obd_uuid.uuid ) == 0)
+ if (!strcmp((char *)fed->fed_fcd->fcd_uuid, (char *)obd->obd_uuid.uuid))
GOTO(free, 0);
LASSERT(filter->fo_last_rcvd_slots != NULL);
CDEBUG(D_INFO, "freeing client at idx %u (%lld) with UUID '%s'\n",
fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
- if (!test_and_clear_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
+ /* Clear the bit _after_ zeroing out the client so we don't
+ race with filter_client_add and zero out new clients.*/
+ if (!test_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
CERROR("FILTER client %u: bit already clear in bitmap!!\n",
fed->fed_lr_idx);
LBUG();
}
memset(&zero_fcd, 0, sizeof zero_fcd);
- push_ctxt(&saved, &obd->obd_ctxt, NULL);
+ push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
rc = fsfilt_write_record(obd, filter->fo_rcvd_filp, &zero_fcd,
sizeof(zero_fcd), &off, 1);
- pop_ctxt(&saved, &obd->obd_ctxt, NULL);
+ if (rc == 0)
+ /* update server's transno */
+ filter_update_server_data(obd, filter->fo_rcvd_filp,
+ filter->fo_fsd, 1);
+ pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
CDEBUG(rc == 0 ? D_INFO : D_ERROR,
"zeroing disconnecting client %s at idx %u (%llu) in %s rc %d\n",
fed->fed_fcd->fcd_uuid, fed->fed_lr_idx, fed->fed_lr_off,
LAST_RCVD, rc);
+ if (!test_and_clear_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
+ CERROR("FILTER client %u: bit already clear in bitmap!!\n",
+ fed->fed_lr_idx);
+ LBUG();
+ }
+
free:
OBD_FREE(fed->fed_fcd, sizeof(*fed->fed_fcd));
{
OBD_FREE(filter->fo_fsd, sizeof(*filter->fo_fsd));
filter->fo_fsd = NULL;
- OBD_FREE(filter->fo_last_rcvd_slots,
- FILTER_LR_MAX_CLIENT_WORDS * sizeof(unsigned long));
+ OBD_FREE(filter->fo_last_rcvd_slots, FILTER_LR_MAX_CLIENTS/8);
filter->fo_last_rcvd_slots = NULL;
return 0;
}
int rc;
ENTRY;
+ if (filter->fo_last_objid_files[group] == NULL) {
+ CERROR("Object group "LPU64" not fully setup; not updating "
+ "last_objid\n", group);
+ RETURN(0);
+ }
+
CDEBUG(D_INODE, "server last_objid for group "LPU64": "LPU64"\n",
group, filter->fo_last_objids[group]);
RETURN(-ENOMEM);
filter->fo_fsd = fsd;
- OBD_ALLOC(filter->fo_last_rcvd_slots,
- FILTER_LR_MAX_CLIENT_WORDS * sizeof(unsigned long));
+ OBD_ALLOC(filter->fo_last_rcvd_slots, FILTER_LR_MAX_CLIENTS/8);
if (filter->fo_last_rcvd_slots == NULL) {
OBD_FREE(fsd, sizeof(*fsd));
RETURN(-ENOMEM);
LAST_RCVD, rc);
GOTO(err_fsd, rc);
}
- if (strcmp(fsd->fsd_uuid, obd->obd_uuid.uuid) != 0) {
+ if (strcmp((char *)fsd->fsd_uuid, (char *)obd->obd_uuid.uuid)) {
CERROR("OBD UUID %s does not match last_rcvd UUID %s\n",
obd->obd_uuid.uuid, fsd->fsd_uuid);
GOTO(err_fsd, rc = -EINVAL);
filter->fo_subdir_count = le16_to_cpu(fsd->fsd_subdir_count);
}
- if (fsd->fsd_feature_incompat & ~le32_to_cpu(FILTER_INCOMPAT_SUPP)) {
+ if (fsd->fsd_feature_incompat & ~cpu_to_le32(FILTER_INCOMPAT_SUPP)) {
CERROR("unsupported feature %x\n",
le32_to_cpu(fsd->fsd_feature_incompat) &
~FILTER_INCOMPAT_SUPP);
GOTO(err_fsd, rc = -EINVAL);
}
- if (fsd->fsd_feature_rocompat & ~le32_to_cpu(FILTER_ROCOMPAT_SUPP)) {
+ if (fsd->fsd_feature_rocompat & ~cpu_to_le32(FILTER_ROCOMPAT_SUPP)) {
CERROR("read-only feature %x\n",
le32_to_cpu(fsd->fsd_feature_rocompat) &
~FILTER_ROCOMPAT_SUPP);
CDEBUG(D_INODE, "%s: server last_rcvd : "LPU64"\n",
obd->obd_name, le64_to_cpu(fsd->fsd_last_transno));
- CDEBUG(D_INODE, "%s: server last_mount: "LPU64"\n",
- obd->obd_name, mount_count);
+ CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
+ obd->obd_name, mount_count + 1);
CDEBUG(D_INODE, "%s: server data size: %u\n",
obd->obd_name, le32_to_cpu(fsd->fsd_server_size));
CDEBUG(D_INODE, "%s: per-client data start: %u\n",
CDEBUG(D_INODE, "%s: server subdir_count: %u\n",
obd->obd_name, le16_to_cpu(fsd->fsd_subdir_count));
CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
- last_rcvd_size <= FILTER_LR_CLIENT_START ? 0 :
- (last_rcvd_size-FILTER_LR_CLIENT_START) /FILTER_LR_CLIENT_SIZE);
+ last_rcvd_size <= le32_to_cpu(fsd->fsd_client_start) ? 0 :
+ (last_rcvd_size - le32_to_cpu(fsd->fsd_client_start)) /
+ le16_to_cpu(fsd->fsd_client_size));
if (!obd->obd_replayable) {
CWARN("%s: recovery support OFF\n", obd->obd_name);
for (cl_idx = 0, off = le32_to_cpu(fsd->fsd_client_start);
off < last_rcvd_size; cl_idx++) {
__u64 last_rcvd;
- int mount_age;
+ struct obd_export *exp;
+ struct filter_export_data *fed;
if (!fcd) {
OBD_ALLOC(fcd, sizeof(*fcd));
/* These exports are cleaned up by filter_disconnect(), so they
* need to be set up like real exports as filter_connect() does.
*/
- mount_age = mount_count - le64_to_cpu(fcd->fcd_mount_count);
- if (mount_age < FILTER_MOUNT_RECOV) {
- struct obd_export *exp = class_new_export(obd);
- struct filter_export_data *fed;
- CERROR("RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64
- " srv lr: "LPU64" mnt: "LPU64" last mount: "
- LPU64"\n", fcd->fcd_uuid, cl_idx,
- last_rcvd, le64_to_cpu(fsd->fsd_last_transno),
- le64_to_cpu(fcd->fcd_mount_count), mount_count);
- if (exp == NULL)
- GOTO(err_client, rc = -ENOMEM);
-
- memcpy(&exp->exp_client_uuid.uuid, fcd->fcd_uuid,
- sizeof exp->exp_client_uuid.uuid);
- fed = &exp->exp_filter_data;
- fed->fed_fcd = fcd;
- filter_client_add(obd, filter, fed, cl_idx);
- /* create helper if export init gets more complex */
- spin_lock_init(&fed->fed_lock);
-
- fcd = NULL;
- obd->obd_recoverable_clients++;
- class_export_put(exp);
- } else {
- CDEBUG(D_INFO, "discarded client %d UUID '%s' count "
- LPU64"\n", cl_idx, fcd->fcd_uuid,
- le64_to_cpu(fcd->fcd_mount_count));
- }
+ exp = class_new_export(obd);
+ CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64" "
+ "srv lr: "LPU64" fcd_group %d \n", fcd->fcd_uuid, cl_idx,
+ last_rcvd, le64_to_cpu(fsd->fsd_last_transno),
+ le32_to_cpu(fcd->fcd_group));
+ if (exp == NULL)
+ GOTO(err_client, rc = -ENOMEM);
+
+ memcpy(&exp->exp_client_uuid.uuid, fcd->fcd_uuid,
+ sizeof exp->exp_client_uuid.uuid);
+ fed = &exp->exp_filter_data;
+ fed->fed_fcd = fcd;
+ fed->fed_group = le32_to_cpu(fcd->fcd_group);
+ filter_client_add(obd, filter, fed, cl_idx);
+ /* create helper if export init gets more complex */
+ spin_lock_init(&fed->fed_lock);
+
+ fcd = NULL;
+ exp->exp_connected = 0;
+ exp->exp_req_replay_needed = 1;
+ exp->exp_lock_replay_needed = 1;
+ atomic_inc(&obd->obd_req_replay_clients);
+ atomic_inc(&obd->obd_lock_replay_clients);
+ obd->obd_recoverable_clients++;
+ obd->obd_max_recoverable_clients++;
+ class_export_put(exp);
CDEBUG(D_OTHER, "client at idx %d has last_rcvd = "LPU64"\n",
cl_idx, last_rcvd);
- if (last_rcvd > le64_to_cpu(filter->fo_fsd->fsd_last_transno))
- filter->fo_fsd->fsd_last_transno=cpu_to_le64(last_rcvd);
+ if (last_rcvd > le64_to_cpu(fsd->fsd_last_transno))
+ fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
}
- obd->obd_last_committed = le64_to_cpu(filter->fo_fsd->fsd_last_transno);
+ obd->obd_last_committed = le64_to_cpu(fsd->fsd_last_transno);
if (obd->obd_recoverable_clients) {
- CERROR("RECOVERY: %d recoverable clients, last_rcvd "
- LPU64"\n", obd->obd_recoverable_clients,
- le64_to_cpu(filter->fo_fsd->fsd_last_transno));
+ CWARN("RECOVERY: service %s, %d recoverable clients, "
+ "last_transno "LPU64"\n", obd->obd_name,
+ obd->obd_recoverable_clients,
+ le64_to_cpu(fsd->fsd_last_transno));
obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
- obd->obd_recovering = 1;
+ target_start_recovery_thread(obd, ost_handle);
}
if (fcd)
OBD_FREE(fcd, sizeof(*fcd));
out:
- fsd->fsd_mount_count = cpu_to_le64(mount_count + 1);
+ filter->fo_mount_count = mount_count + 1;
+ fsd->fsd_mount_count = cpu_to_le64(filter->fo_mount_count);
/* save it, so mount count and last_transno is current */
rc = filter_update_server_data(obd, filp, filter->fo_fsd, 1);
-
- RETURN(rc);
+ if (rc)
+ GOTO(err_client, rc);
+ RETURN(0);
err_client:
class_disconnect_exports(obd, 0);
static int filter_cleanup_groups(struct obd_device *obd)
{
struct filter_obd *filter = &obd->u.filter;
- int i;
+ struct dentry *dentry;
+ int i, k;
ENTRY;
- if (filter->fo_subdir_count) {
- for (i = 0; i < filter->fo_subdir_count; i++) {
- struct dentry *dentry = filter->fo_dentry_O_sub[i];
- f_dput(dentry);
- filter->fo_dentry_O_sub[i] = NULL;
- }
- OBD_FREE(filter->fo_dentry_O_sub,
- filter->fo_subdir_count *
- sizeof(*filter->fo_dentry_O_sub));
- }
- if (filter->fo_dentry_O_groups != NULL &&
- filter->fo_last_objids != NULL &&
- filter->fo_last_objid_files != NULL) {
- for (i = 0; i < FILTER_GROUPS; i++) {
- struct dentry *dentry = filter->fo_dentry_O_groups[i];
- struct file *filp = filter->fo_last_objid_files[i];
- if (dentry != NULL) {
+ for (i = 0; i < filter->fo_group_count; i++) {
+ if (filter->fo_subdirs != NULL) {
+ for (k = 0; k < filter->fo_subdir_count; k++) {
+ dentry = filter->fo_subdirs[i].dentry[k];
+ if (dentry == NULL)
+ continue;
f_dput(dentry);
- filter->fo_dentry_O_groups[i] = NULL;
- }
- if (filp != NULL) {
- filp_close(filp, 0);
- filter->fo_last_objid_files[i] = NULL;
+ filter->fo_subdirs[i].dentry[k] = NULL;
}
}
+ if (filter->fo_last_objid_files[i] != NULL) {
+ filp_close(filter->fo_last_objid_files[i], 0);
+ filter->fo_last_objid_files[i] = NULL;
+ }
+ if (filter->fo_groups[i] != NULL) {
+ dput(filter->fo_groups[i]);
+ filter->fo_groups[i] = NULL;
+ }
}
- if (filter->fo_dentry_O_groups != NULL)
- OBD_FREE(filter->fo_dentry_O_groups,
- FILTER_GROUPS * sizeof(struct dentry *));
+ if (filter->fo_subdirs != NULL)
+ OBD_FREE(filter->fo_subdirs,
+ filter->fo_group_count * sizeof(*filter->fo_subdirs));
+ if (filter->fo_groups != NULL)
+ OBD_FREE(filter->fo_groups,
+ filter->fo_group_count * sizeof(*filter->fo_groups));
if (filter->fo_last_objids != NULL)
OBD_FREE(filter->fo_last_objids,
- FILTER_GROUPS * sizeof(__u64));
+ filter->fo_group_count * sizeof(__u64));
if (filter->fo_last_objid_files != NULL)
OBD_FREE(filter->fo_last_objid_files,
- FILTER_GROUPS * sizeof(struct file *));
+ filter->fo_group_count * sizeof(struct file *));
+ f_dput(filter->fo_dentry_O);
+ RETURN(0);
+}
+
+static int filter_update_last_group(struct obd_device *obd, int group)
+{
+ struct filter_obd *filter = &obd->u.filter;
+ struct file *filp = NULL;
+ int last_group = 0, rc;
+ loff_t off = 0;
+ ENTRY;
+
+ if (group <= filter->fo_committed_group)
+ RETURN(0);
+
+ filp = filp_open("LAST_GROUP", O_RDWR, 0700);
+ if (IS_ERR(filp)) {
+ rc = PTR_ERR(filp);
+ filp = NULL;
+ CERROR("cannot open LAST_GROUP: rc = %d\n", rc);
+ GOTO(cleanup, rc);
+ }
+
+ rc = fsfilt_read_record(obd, filp, &last_group, sizeof(__u32), &off);
+ if (rc) {
+ CDEBUG(D_INODE, "error reading LAST_GROUP: rc %d\n",rc);
+ GOTO(cleanup, rc);
+ }
+ LASSERT(off == 0 || last_group >= FILTER_MIN_GROUPS);
+ CDEBUG(D_INODE, "%s: previous %d, new %d\n",
+ obd->obd_name, last_group, group);
+
+ off = 0;
+ last_group = group;
+ /* must be sync: bXXXX */
+ rc = fsfilt_write_record(obd, filp, &last_group, sizeof(__u32), &off, 1);
+ if (rc) {
+ CDEBUG(D_INODE, "error updating LAST_GROUP: rc %d\n", rc);
+ GOTO(cleanup, rc);
+ }
+
+ filter->fo_committed_group = group;
+cleanup:
+ if (filp)
+ filp_close(filp, 0);
+ RETURN(rc);
+}
+
+static int filter_read_group_internal(struct obd_device *obd, int group,
+ int create)
+{
+ struct filter_obd *filter = &obd->u.filter;
+ __u64 *new_objids = NULL;
+ struct filter_subdirs *new_subdirs = NULL, *tmp_subdirs = NULL;
+ struct dentry **new_groups = NULL;
+ struct file **new_files = NULL;
+ struct dentry *dentry;
+ struct file *filp;
+ int old_count = filter->fo_group_count, rc, stage = 0, i;
+ char name[25];
+ __u64 last_objid;
+ loff_t off = 0;
+
+ snprintf(name, 24, "%d", group);
+ name[24] = '\0';
+
+ if (!create) {
+ dentry = ll_lookup_one_len(name, filter->fo_dentry_O,
+ strlen(name));
+ if (IS_ERR(dentry)) {
+ CERROR("Cannot lookup expected object group %d: %ld\n",
+ group, PTR_ERR(dentry));
+ RETURN(PTR_ERR(dentry));
+ }
+ } else {
+ dentry = simple_mkdir(filter->fo_dentry_O, name, 0700, 1);
+ if (IS_ERR(dentry)) {
+ CERROR("cannot lookup/create O/%s: rc = %ld\n", name,
+ PTR_ERR(dentry));
+ RETURN(PTR_ERR(dentry));
+ }
+ }
+ stage = 1;
+
+ snprintf(name, 24, "O/%d/LAST_ID", group);
+ name[24] = '\0';
+ filp = filp_open(name, O_CREAT | O_RDWR, 0700);
+ if (IS_ERR(filp)) {
+ CERROR("cannot create %s: rc = %ld\n", name, PTR_ERR(filp));
+ GOTO(cleanup, rc = PTR_ERR(filp));
+ }
+ stage = 2;
+
+ rc = fsfilt_read_record(obd, filp, &last_objid, sizeof(__u64), &off);
+ if (rc) {
+ CDEBUG(D_INODE, "error reading %s: rc %d\n", name, rc);
+ GOTO(cleanup, rc);
+ }
+
+ if (filter->fo_subdir_count) {
+ OBD_ALLOC(tmp_subdirs, sizeof(*tmp_subdirs));
+ if (tmp_subdirs == NULL)
+ GOTO(cleanup, rc = -ENOMEM);
+ stage = 3;
+
+ for (i = 0; i < filter->fo_subdir_count; i++) {
+ char dir[20];
+ snprintf(dir, sizeof(dir), "d%u", i);
+
+ tmp_subdirs->dentry[i] =
+ simple_mkdir(dentry, dir, 0700, 1);
+ if (IS_ERR(tmp_subdirs->dentry[i])) {
+ rc = PTR_ERR(tmp_subdirs->dentry[i]);
+ CERROR("can't lookup/create O/%d/%s: rc = %d\n",
+ group, dir, rc);
+ GOTO(cleanup, rc);
+ }
+ CDEBUG(D_INODE, "got/created O/%d/%s: %p\n", group, dir,
+ tmp_subdirs->dentry[i]);
+ }
+ }
+
+ /* 'group' is an index; we need an array of length 'group + 1' */
+ if (group + 1 > old_count) {
+ int len = group + 1;
+ OBD_ALLOC(new_objids, len * sizeof(*new_objids));
+ OBD_ALLOC(new_subdirs, len * sizeof(*new_subdirs));
+ OBD_ALLOC(new_groups, len * sizeof(*new_groups));
+ OBD_ALLOC(new_files, len * sizeof(*new_files));
+ stage = 4;
+ if (new_objids == NULL || new_subdirs == NULL ||
+ new_groups == NULL || new_files == NULL)
+ GOTO(cleanup, rc = -ENOMEM);
+
+ memcpy(new_objids, filter->fo_last_objids,
+ old_count * sizeof(*new_objids));
+ memcpy(new_subdirs, filter->fo_subdirs,
+ old_count * sizeof(*new_subdirs));
+ memcpy(new_groups, filter->fo_groups,
+ old_count * sizeof(*new_groups));
+ memcpy(new_files, filter->fo_last_objid_files,
+ old_count * sizeof(*new_files));
+
+ if (old_count) {
+ OBD_FREE(filter->fo_last_objids,
+ old_count * sizeof(*new_objids));
+ OBD_FREE(filter->fo_subdirs,
+ old_count * sizeof(*new_subdirs));
+ OBD_FREE(filter->fo_groups,
+ old_count * sizeof(*new_groups));
+ OBD_FREE(filter->fo_last_objid_files,
+ old_count * sizeof(*new_files));
+ }
+ filter->fo_last_objids = new_objids;
+ filter->fo_subdirs = new_subdirs;
+ filter->fo_groups = new_groups;
+ filter->fo_last_objid_files = new_files;
+ filter->fo_group_count = len;
+ }
+
+ filter->fo_groups[group] = dentry;
+ filter->fo_last_objid_files[group] = filp;
+ if (filter->fo_subdir_count) {
+ filter->fo_subdirs[group] = *tmp_subdirs;
+ OBD_FREE(tmp_subdirs, sizeof(*tmp_subdirs));
+ }
+
+ filter_update_last_group(obd, group);
+
+ if (filp->f_dentry->d_inode->i_size == 0) {
+ filter->fo_last_objids[group] = FILTER_INIT_OBJID;
+ RETURN(0);
+ }
+
+ filter->fo_last_objids[group] = le64_to_cpu(last_objid);
+ CDEBUG(D_INODE, "%s: server last_objid group %d: "LPU64"\n",
+ obd->obd_name, group, last_objid);
RETURN(0);
+ cleanup:
+ switch (stage) {
+ case 4:
+ if (new_objids != NULL)
+ OBD_FREE(new_objids, group * sizeof(*new_objids));
+ if (new_subdirs != NULL)
+ OBD_FREE(new_subdirs, group * sizeof(*new_subdirs));
+ if (new_groups != NULL)
+ OBD_FREE(new_groups, group * sizeof(*new_groups));
+ if (new_files != NULL)
+ OBD_FREE(new_files, group * sizeof(*new_files));
+ case 3:
+ if (filter->fo_subdir_count) {
+ for (i = 0; i < filter->fo_subdir_count; i++) {
+ if (tmp_subdirs->dentry[i] != NULL)
+ dput(tmp_subdirs->dentry[i]);
+ }
+ OBD_FREE(tmp_subdirs, sizeof(*tmp_subdirs));
+ }
+ case 2:
+ filp_close(filp, 0);
+ case 1:
+ dput(dentry);
+ }
+ RETURN(rc);
+}
+
+static int filter_read_groups(struct obd_device *obd, int last_group,
+ int create)
+{
+ struct filter_obd *filter = &obd->u.filter;
+ int old_count, group, rc = 0;
+
+ down(&filter->fo_init_lock);
+ old_count = filter->fo_group_count;
+ for (group = old_count; group <= last_group; group++) {
+ if (group == 0)
+ continue; /* no group zero */
+
+ rc = filter_read_group_internal(obd, group, create);
+ if (rc != 0)
+ break;
+ }
+ up(&filter->fo_init_lock);
+ return rc;
}
-/* FIXME: object groups */
static int filter_prep_groups(struct obd_device *obd)
{
struct filter_obd *filter = &obd->u.filter;
struct dentry *dentry, *O_dentry;
- struct file *filp;
- int i, rc = 0, cleanup_phase = 0;
+ int rc = 0, cleanup_phase = 0;
+ struct file *filp = NULL;
+ int last_group;
+ loff_t off = 0;
ENTRY;
- O_dentry = simple_mkdir(current->fs->pwd, "O", 0700);
+ O_dentry = simple_mkdir(current->fs->pwd, "O", 0700, 1);
CDEBUG(D_INODE, "got/created O: %p\n", O_dentry);
if (IS_ERR(O_dentry)) {
rc = PTR_ERR(O_dentry);
GOTO(cleanup_O0, rc);
cleanup_O0:
- dput(O0_dentry);
+ f_dput(O0_dentry);
cleanup_R:
- dput(dentry);
+ f_dput(dentry);
if (rc)
GOTO(cleanup, rc);
} else {
- dput(dentry);
+ f_dput(dentry);
}
- OBD_ALLOC(filter->fo_last_objids, FILTER_GROUPS * sizeof(__u64));
- if (filter->fo_last_objids == NULL)
- GOTO(cleanup, rc = -ENOMEM);
cleanup_phase = 2; /* groups */
- OBD_ALLOC(filter->fo_dentry_O_groups, FILTER_GROUPS * sizeof(dentry));
- if (filter->fo_dentry_O_groups == NULL)
- GOTO(cleanup, rc = -ENOMEM);
- OBD_ALLOC(filter->fo_last_objid_files, FILTER_GROUPS * sizeof(filp));
- if (filter->fo_last_objid_files == NULL)
- GOTO(cleanup, rc = -ENOMEM);
-
- for (i = 0; i < FILTER_GROUPS; i++) {
- char name[25];
- loff_t off = 0;
-
- sprintf(name, "%d", i);
- dentry = simple_mkdir(O_dentry, name, 0700);
- CDEBUG(D_INODE, "got/created O/%s: %p\n", name, dentry);
- if (IS_ERR(dentry)) {
- rc = PTR_ERR(dentry);
- CERROR("cannot create O/%s: rc = %d\n", name, rc);
- GOTO(cleanup, rc);
- }
- filter->fo_dentry_O_groups[i] = dentry;
-
- sprintf(name, "O/%d/LAST_ID", i);
- filp = filp_open(name, O_CREAT | O_RDWR, 0700);
- if (IS_ERR(dentry)) {
- rc = PTR_ERR(dentry);
- CERROR("cannot create %s: rc = %d\n", name, rc);
- GOTO(cleanup, rc);
- }
- filter->fo_last_objid_files[i] = filp;
-
- if (filp->f_dentry->d_inode->i_size == 0) {
- if (i == 0 && filter->fo_fsd->fsd_unused != 0) {
- /* OST conversion, remove sometime post 1.0 */
- filter->fo_last_objids[i] =
- le64_to_cpu(filter->fo_fsd->fsd_unused);
- CWARN("saving old objid "LPU64" to LAST_ID\n",
- filter->fo_last_objids[i]);
- rc = filter_update_last_objid(obd, 0, 1);
- if (rc)
- GOTO(cleanup, rc);
- } else {
- filter->fo_last_objids[i] = FILTER_INIT_OBJID;
- }
- continue;
- }
-
- rc = fsfilt_read_record(obd, filp, &filter->fo_last_objids[i],
- sizeof(__u64), &off);
- if (rc) {
- CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
- name, rc);
- GOTO(cleanup, rc);
- }
- filter->fo_last_objids[i] =
- le64_to_cpu(filter->fo_last_objids[i]);
- CDEBUG(D_INODE, "%s: server last_objid group %d: "LPU64"\n",
- obd->obd_name, i, filter->fo_last_objids[i]);
+ /* we have to initialize all groups before first connections from
+ * clients because they may send create/destroy for any group -bzzz */
+ filp = filp_open("LAST_GROUP", O_CREAT | O_RDWR, 0700);
+ if (IS_ERR(filp)) {
+ CERROR("cannot create LAST_GROUP: rc = %ld\n", PTR_ERR(filp));
+ GOTO(cleanup, rc = PTR_ERR(filp));
}
+ cleanup_phase = 3; /* filp */
- if (filter->fo_subdir_count) {
- O_dentry = filter->fo_dentry_O_groups[0];
- OBD_ALLOC(filter->fo_dentry_O_sub,
- filter->fo_subdir_count * sizeof(dentry));
- if (filter->fo_dentry_O_sub == NULL)
- GOTO(cleanup, rc = -ENOMEM);
-
- for (i = 0; i < filter->fo_subdir_count; i++) {
- char dir[20];
- snprintf(dir, sizeof(dir), "d%u", i);
-
- dentry = simple_mkdir(O_dentry, dir, 0700);
- CDEBUG(D_INODE, "got/created O/0/%s: %p\n", dir,dentry);
- if (IS_ERR(dentry)) {
- rc = PTR_ERR(dentry);
- CERROR("can't create O/0/%s: rc = %d\n",dir,rc);
- GOTO(cleanup, rc);
- }
- filter->fo_dentry_O_sub[i] = dentry;
- }
+ rc = fsfilt_read_record(obd, filp, &last_group, sizeof(__u32), &off);
+ if (rc) {
+ CDEBUG(D_INODE, "error reading LAST_GROUP: rc %d\n", rc);
+ GOTO(cleanup, rc);
+ }
+ if (off == 0) {
+ last_group = FILTER_MIN_GROUPS;
+ } else {
+ LASSERT(last_group >= FILTER_MIN_GROUPS);
}
+
+ CWARN("%s: initialize groups [%d,%d]\n", obd->obd_name,
+ FILTER_MIN_GROUPS, last_group);
+ filter->fo_committed_group = last_group;
+ rc = filter_read_groups(obd, last_group, 1);
+ if (rc)
+ GOTO(cleanup, rc);
+
+ filp_close(filp, 0);
RETURN(0);
cleanup:
switch (cleanup_phase) {
+ case 3:
+ filp_close(filp, 0);
case 2:
filter_cleanup_groups(obd);
case 1:
/* setup the object store with correct subdirectories */
static int filter_prep(struct obd_device *obd)
{
- struct obd_run_ctxt saved;
+ struct lvfs_run_ctxt saved;
struct filter_obd *filter = &obd->u.filter;
struct file *file;
struct inode *inode;
int rc = 0;
ENTRY;
- push_ctxt(&saved, &obd->obd_ctxt, NULL);
+ push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
file = filp_open(LAST_RCVD, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
if (!file || IS_ERR(file)) {
rc = PTR_ERR(file);
GOTO(err_server_data, rc);
out:
- pop_ctxt(&saved, &obd->obd_ctxt, NULL);
+ pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
return(rc);
/* cleanup the filter: write last used object id to status file */
static void filter_post(struct obd_device *obd)
{
- struct obd_run_ctxt saved;
+ struct lvfs_run_ctxt saved;
struct filter_obd *filter = &obd->u.filter;
int rc, i;
* best to start a transaction with h_sync, because we removed this
* from lastobjid */
- push_ctxt(&saved, &obd->obd_ctxt, NULL);
+ push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
filter->fo_fsd, 0);
if (rc)
CERROR("error writing server data: rc = %d\n", rc);
- for (i = 0; i < FILTER_GROUPS; i++) {
- rc = filter_update_last_objid(obd, i, (i == FILTER_GROUPS - 1));
+ for (i = 1; i < filter->fo_group_count; i++) {
+ rc = filter_update_last_objid(obd, i,
+ (i == filter->fo_group_count - 1));
if (rc)
CERROR("error writing group %d lastobjid: rc = %d\n",
i, rc);
}
- filp_close(filter->fo_rcvd_filp, 0);
+ rc = filp_close(filter->fo_rcvd_filp, 0);
filter->fo_rcvd_filp = NULL;
if (rc)
CERROR("error closing %s: rc = %d\n", LAST_RCVD, rc);
filter_cleanup_groups(obd);
- f_dput(filter->fo_dentry_O);
filter_free_server_data(filter);
- pop_ctxt(&saved, &obd->obd_ctxt, NULL);
+ pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
}
-static void filter_set_last_id(struct filter_obd *filter, struct obdo *oa,
- obd_id id)
+static void filter_set_last_id(struct filter_obd *filter, int group, obd_id id)
{
- obd_gr group = 0;
LASSERT(filter->fo_fsd != NULL);
-
- if (oa != NULL) {
- LASSERT(oa->o_gr <= FILTER_GROUPS);
- group = oa->o_gr;
- }
+ LASSERT(group > 0);
+ LASSERT(group < filter->fo_group_count);
spin_lock(&filter->fo_objidlock);
filter->fo_last_objids[group] = id;
spin_unlock(&filter->fo_objidlock);
}
-__u64 filter_last_id(struct filter_obd *filter, struct obdo *oa)
+__u64 filter_last_id(struct filter_obd *filter, int group)
{
obd_id id;
- obd_gr group = 0;
LASSERT(filter->fo_fsd != NULL);
+ LASSERT(group > 0);
+ LASSERT(group < filter->fo_group_count);
- if (oa != NULL) {
- LASSERT(oa->o_gr <= FILTER_GROUPS);
- group = oa->o_gr;
- }
-
- /* FIXME: object groups */
spin_lock(&filter->fo_objidlock);
id = filter->fo_last_objids[group];
spin_unlock(&filter->fo_objidlock);
RETURN(0);
}
-static int filter_lock_dentry(struct obd_device *obd, struct dentry *de,
- ldlm_mode_t lock_mode,struct lustre_handle *lockh)
-{
- struct ldlm_res_id res_id = { .name = {0} };
- int flags = 0, rc;
- ENTRY;
+extern void *lock_dir(struct inode *dir, struct qstr *name);
+extern void unlock_dir(struct inode *dir, void *lock);
- res_id.name[0] = de->d_inode->i_ino;
- res_id.name[1] = de->d_inode->i_generation;
- rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, NULL,
- res_id, LDLM_PLAIN, NULL, 0, lock_mode,
- &flags, ldlm_completion_ast,
- filter_blocking_ast, NULL, lockh);
+static void *filter_lock_dentry(struct obd_device *obd,
+ struct dentry *dparent,
+ obd_id id)
+{
+#ifdef S_PDIROPS
+ struct qstr qstr;
+ char name[32];
+ int len;
- RETURN(rc == ELDLM_OK ? 0 : -EIO); /* XXX translate ldlm code */
+ len = sprintf(name, LPU64, id);
+ qstr_assign(&qstr, (char *)name, len);
+ return lock_dir(dparent->d_inode, &qstr);
+#else
+ down(&dparent->d_inode->i_sem);
+#endif
+ return 0;
}
/* We never dget the object parent, so DON'T dput it either */
-static void filter_parent_unlock(struct dentry *dparent,
- struct lustre_handle *lockh,
- ldlm_mode_t lock_mode)
+static void filter_parent_unlock(struct dentry *dparent, void *lock)
{
- ldlm_lock_decref(lockh, lock_mode);
+#ifdef S_PDIROPS
+ LASSERT(lock != NULL);
+ unlock_dir(dparent->d_inode, lock);
+#else
+ up(&dparent->d_inode->i_sem);
+#endif
}
/* We never dget the object parent, so DON'T dput it either */
struct dentry *filter_parent(struct obd_device *obd, obd_gr group, obd_id objid)
{
struct filter_obd *filter = &obd->u.filter;
- LASSERT(group < FILTER_GROUPS); /* FIXME: object groups */
+ LASSERT(group < filter->fo_group_count);
+ LASSERT(group > 0);
- if (group > 0 || filter->fo_subdir_count == 0)
- return filter->fo_dentry_O_groups[group];
+ if (filter->fo_subdir_count == 0)
+ return filter->fo_groups[group];
- return filter->fo_dentry_O_sub[objid & (filter->fo_subdir_count - 1)];
+ return filter->fo_subdirs[group].dentry[objid & (filter->fo_subdir_count - 1)];
}
/* We never dget the object parent, so DON'T dput it either */
struct dentry *filter_parent_lock(struct obd_device *obd, obd_gr group,
- obd_id objid, ldlm_mode_t lock_mode,
- struct lustre_handle *lockh)
+ obd_id objid, void **lock)
{
unsigned long now = jiffies;
- struct dentry *de = filter_parent(obd, group, objid);
- int rc;
+ struct dentry *dparent = filter_parent(obd, group, objid);
+
+ if (IS_ERR(dparent))
+ return dparent;
- if (IS_ERR(de))
- return de;
+ LASSERT(dparent);
+ LASSERT(dparent->d_inode);
- rc = filter_lock_dentry(obd, de, lock_mode, lockh);
- if (time_after(jiffies, now + 15 * HZ))
- CERROR("slow parent lock %lus\n", (jiffies - now) / HZ);
- return rc ? ERR_PTR(rc) : de;
+ *lock = filter_lock_dentry(obd, dparent, objid);
+ fsfilt_check_slow(now, obd_timeout, "parent lock");
+ return dparent;
}
/* How to get files, dentries, inodes from object id's.
* dir_dentry is NULL, we do a read lock while we do the lookup to
* avoid races with create/destroy and such changing the directory
* internal to the filesystem code. */
-struct dentry *filter_fid2dentry(struct obd_device *obd,
- struct dentry *dir_dentry,
- obd_gr group, obd_id id)
+struct dentry *filter_id2dentry(struct obd_device *obd,
+ struct dentry *dir_dentry,
+ obd_gr group, obd_id id)
{
- struct lustre_handle lockh;
struct dentry *dparent = dir_dentry;
struct dentry *dchild;
+ void *lock = NULL;
char name[32];
int len;
ENTRY;
len = sprintf(name, LPU64, id);
if (dir_dentry == NULL) {
- dparent = filter_parent_lock(obd, group, id, LCK_PR, &lockh);
- if (IS_ERR(dparent))
+ dparent = filter_parent_lock(obd, group, id, &lock);
+ if (IS_ERR(dparent)) {
+ CERROR("%s: error getting object "LPU64":"LPU64
+ " parent: rc %ld\n", obd->obd_name,
+ id, group, PTR_ERR(dparent));
RETURN(dparent);
+ }
}
- CDEBUG(D_INODE, "looking up object O/%*s/%s\n",
+ CDEBUG(D_INODE, "looking up object O/%.*s/%s\n",
dparent->d_name.len, dparent->d_name.name, name);
- dchild = ll_lookup_one_len(name, dparent, len);
+ dchild = /*ll_*/lookup_one_len(name, dparent, len);
if (dir_dentry == NULL)
- filter_parent_unlock(dparent, &lockh, LCK_PR);
+ filter_parent_unlock(dparent, lock);
if (IS_ERR(dchild)) {
- CERROR("child lookup error %ld\n", PTR_ERR(dchild));
+ CERROR("%s: child lookup error %ld\n", obd->obd_name,
+ PTR_ERR(dchild));
RETURN(dchild);
}
+ if (dchild->d_inode != NULL && is_bad_inode(dchild->d_inode)) {
+ CERROR("%s: got bad inode "LPU64"\n", obd->obd_name, id);
+ f_dput(dchild);
+ RETURN(ERR_PTR(-ENOENT));
+ }
+
CDEBUG(D_INODE, "got child objid %s: %p, count = %d\n",
name, dchild, atomic_read(&dchild->d_count));
RETURN(dchild);
}
-static int filter_prepare_destroy(struct obd_device *obd, obd_id objid)
+static int filter_prepare_destroy(struct obd_device *obd, obd_id objid,
+ obd_id group)
{
struct lustre_handle lockh;
int flags = LDLM_AST_DISCARD_DATA, rc;
- struct ldlm_res_id res_id = { .name = { objid } };
- struct ldlm_extent extent = { 0, OBD_OBJECT_EOF };
- ENTRY;
+ struct ldlm_res_id res_id = { .name = { objid, 0, group, 0 } };
+ ldlm_policy_data_t policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
+ ENTRY;
/* Tell the clients that the object is gone now and that they should
- * throw away any cached pages. If we're the OST at stripe 0 in the
- * file then this enqueue will communicate the DISCARD to all the
- * clients. This assumes that we always destroy all the objects for
- * a file at a time, as is currently the case. If we're not the
- * OST at stripe 0 then we'll harmlessly get a very lonely lock in
- * the local DLM and immediately drop it. */
- rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, NULL,
- res_id, LDLM_EXTENT, &extent,
- sizeof(extent), LCK_PW, &flags,
- ldlm_completion_ast, filter_blocking_ast,
- NULL, &lockh);
+ * throw away any cached pages. */
+ rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, res_id,
+ LDLM_EXTENT, &policy, LCK_PW,
+ &flags, filter_blocking_ast, ldlm_completion_ast,
+ NULL, NULL, NULL, 0, NULL, &lockh);
/* We only care about the side-effects, just drop the lock. */
if (rc == ELDLM_OK)
ENTRY;
if (inode->i_nlink != 1 || atomic_read(&inode->i_count) != 1) {
- CERROR("destroying objid %*s nlink = %lu, count = %d\n",
+ CERROR("destroying objid %.*s nlink = %lu, count = %d\n",
dchild->d_name.len, dchild->d_name.name,
- (unsigned long)inode->i_nlink,
+ (unsigned long)inode->i_nlink,
atomic_read(&inode->i_count));
}
rc = vfs_unlink(dparent->d_inode, dchild);
if (rc)
- CERROR("error unlinking objid %*s: rc %d\n",
+ CERROR("error unlinking objid %.*s: rc %d\n",
dchild->d_name.len, dchild->d_name.name, rc);
RETURN(rc);
}
-/* mount the file system (secretly) */
-int filter_common_setup(struct obd_device *obd, obd_count len, void *buf,
- char *option)
+static int filter_intent_policy(struct ldlm_namespace *ns,
+ struct ldlm_lock **lockp, void *req_cookie,
+ ldlm_mode_t mode, int flags, void *data)
{
- struct lustre_cfg* lcfg = buf;
- struct filter_obd *filter = &obd->u.filter;
- struct vfsmount *mnt;
- int rc = 0;
+ struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
+ struct ptlrpc_request *req = req_cookie;
+ struct ldlm_lock *lock = *lockp, *l = NULL;
+ struct ldlm_resource *res = lock->l_resource;
+ ldlm_processing_policy policy;
+ struct ost_lvb *res_lvb, *reply_lvb;
+ struct ldlm_reply *rep;
+ struct list_head *tmp;
+ ldlm_error_t err;
+ int tmpflags = 0, rc, repsize[2] = {sizeof(*rep), sizeof(*reply_lvb)};
ENTRY;
- dev_clear_rdonly(2);
+ policy = ldlm_get_processing_policy(res);
+ LASSERT(policy != NULL);
+ LASSERT(req != NULL);
- if (!lcfg->lcfg_inlbuf1 || !lcfg->lcfg_inlbuf2)
- RETURN(-EINVAL);
+ rc = lustre_pack_reply(req, 2, repsize, NULL);
+ if (rc)
+ RETURN(req->rq_status = rc);
- obd->obd_fsops = fsfilt_get_ops(lcfg->lcfg_inlbuf2);
- if (IS_ERR(obd->obd_fsops))
- RETURN(PTR_ERR(obd->obd_fsops));
+ rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*rep));
+ LASSERT(rep != NULL);
- mnt = do_kern_mount(lcfg->lcfg_inlbuf2, MS_NOATIME | MS_NODIRATIME,
- lcfg->lcfg_inlbuf1, option);
- rc = PTR_ERR(mnt);
- if (IS_ERR(mnt))
- GOTO(err_ops, rc);
+ reply_lvb = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*reply_lvb));
+ LASSERT(reply_lvb != NULL);
- if (lcfg->lcfg_inllen3 > 0 && lcfg->lcfg_inlbuf3) {
- if (*lcfg->lcfg_inlbuf3 == 'f') {
- obd->obd_replayable = 1;
- obd_sync_filter = 1;
- CERROR("%s: recovery enabled\n", obd->obd_name);
- } else {
- if (*lcfg->lcfg_inlbuf3 != 'n') {
- CERROR("unrecognised flag '%c'\n",
- *lcfg->lcfg_inlbuf3);
- }
- // XXX Robert? Why do we get errors here
- // GOTO(err_mntput, rc = -EINVAL);
- }
- }
+ //fixup_handle_for_resent_req(req, lock, &lockh);
- filter->fo_vfsmnt = mnt;
- filter->fo_sb = mnt->mnt_sb;
- filter->fo_fstype = mnt->mnt_sb->s_type->name;
- CDEBUG(D_SUPER, "%s: mnt = %p\n", filter->fo_fstype, mnt);
+ /* If we grant any lock at all, it will be a whole-file read lock.
+ * Call the extent policy function to see if our request can be
+ * granted, or is blocked. */
+ lock->l_policy_data.l_extent.start = 0;
+ lock->l_policy_data.l_extent.end = OBD_OBJECT_EOF;
+ lock->l_req_mode = LCK_PR;
- OBD_SET_CTXT_MAGIC(&obd->obd_ctxt);
- obd->obd_ctxt.pwdmnt = mnt;
- obd->obd_ctxt.pwd = mnt->mnt_root;
- obd->obd_ctxt.fs = get_ds();
- obd->obd_ctxt.cb_ops = filter_lvfs_ops;
+ l_lock(&res->lr_namespace->ns_lock);
- rc = filter_prep(obd);
- if (rc)
- GOTO(err_mntput, rc);
+ res->lr_tmp = &rpc_list;
+ rc = policy(lock, &tmpflags, 0, &err);
+ res->lr_tmp = NULL;
- spin_lock_init(&filter->fo_translock);
- spin_lock_init(&filter->fo_objidlock);
- INIT_LIST_HEAD(&filter->fo_export_list);
- sema_init(&filter->fo_alloc_lock, 1);
+ /* FIXME: we should change the policy function slightly, to not make
+ * this list at all, since we just turn around and free it */
+ while (!list_empty(&rpc_list)) {
+ struct ldlm_ast_work *w =
+ list_entry(rpc_list.next, struct ldlm_ast_work, w_list);
+ list_del(&w->w_list);
+ LDLM_LOCK_PUT(w->w_lock);
+ OBD_FREE(w, sizeof(*w));
+ }
- obd->obd_namespace = ldlm_namespace_new("filter-tgt",
- LDLM_NAMESPACE_SERVER);
- if (obd->obd_namespace == NULL)
- GOTO(err_post, rc = -ENOMEM);
+ if (rc == LDLM_ITER_CONTINUE) {
+ /* The lock met with no resistance; we're finished. */
+ l_unlock(&res->lr_namespace->ns_lock);
+ RETURN(ELDLM_LOCK_REPLACED);
+ }
- ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
- "filter_ldlm_cb_client", &obd->obd_ldlm_client);
+ /* Do not grant any lock, but instead send GL callbacks. The extent
+ * policy nicely created a list of all PW locks for us. We will choose
+ * the highest of those which are larger than the size in the LVB, if
+ * any, and perform a glimpse callback. */
+ down(&res->lr_lvb_sem);
+ res_lvb = res->lr_lvb_data;
+ LASSERT(res_lvb != NULL);
+ *reply_lvb = *res_lvb;
+ up(&res->lr_lvb_sem);
- RETURN(0);
+ list_for_each(tmp, &res->lr_granted) {
+ struct ldlm_lock *tmplock =
+ list_entry(tmp, struct ldlm_lock, l_res_link);
-err_post:
- filter_post(obd);
-err_mntput:
- unlock_kernel();
- mntput(mnt);
- filter->fo_sb = 0;
- lock_kernel();
-err_ops:
- fsfilt_put_ops(obd->obd_fsops);
- return rc;
-}
+ if (tmplock->l_granted_mode == LCK_PR)
+ continue;
-static int filter_setup(struct obd_device *obd, obd_count len, void *buf)
-{
- struct lustre_cfg* lcfg = buf;
- const char *str = NULL;
- char *option = NULL;
- int n = 0;
- int rc;
+ if (tmplock->l_policy_data.l_extent.end <= reply_lvb->lvb_size)
+ continue;
- if (!strcmp(lcfg->lcfg_inlbuf2, "ext3")) {
-#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
- /* bug 1577: implement async-delete for 2.5 */
- str = "errors=remount-ro,asyncdel";
-#else
- str = "errors=remount-ro";
-#endif
- n = strlen(str) + 1;
- OBD_ALLOC(option, n);
- if (option == NULL)
- RETURN(-ENOMEM);
- strcpy(option, str);
+ if (l == NULL) {
+ l = LDLM_LOCK_GET(tmplock);
+ continue;
+ }
+
+ if (l->l_policy_data.l_extent.start >
+ tmplock->l_policy_data.l_extent.start)
+ continue;
+
+ LDLM_LOCK_PUT(l);
+ l = LDLM_LOCK_GET(tmplock);
}
+ l_unlock(&res->lr_namespace->ns_lock);
- rc = filter_common_setup(obd, len, buf, option);
- if (option)
- OBD_FREE(option, n);
- return rc;
-}
+ /* There were no PW locks beyond the size in the LVB; finished. */
+ if (l == NULL)
+ RETURN(ELDLM_LOCK_ABORTED);
-static int filter_postsetup(struct obd_device *obd)
+ if (l->l_glimpse_ast == NULL) {
+ /* We are racing with unlink(); just return -ENOENT */
+ rep->lock_policy_res1 = -ENOENT;
+ goto out;
+ }
+
+ LASSERTF(l->l_glimpse_ast != NULL, "l == %p", l);
+
+ rc = l->l_glimpse_ast(l, NULL); /* this will update the LVB */
+ if (rc != 0 && res->lr_namespace->ns_lvbo &&
+ res->lr_namespace->ns_lvbo->lvbo_update) {
+ res->lr_namespace->ns_lvbo->lvbo_update(res, NULL, 0, 1);
+ }
+
+ down(&res->lr_lvb_sem);
+ *reply_lvb = *res_lvb;
+ up(&res->lr_lvb_sem);
+out:
+ LDLM_LOCK_PUT(l);
+
+ RETURN(ELDLM_LOCK_ABORTED);
+}
+
+static int filter_post_fs_cleanup(struct obd_device *obd)
{
int rc = 0;
+
+ rc = fsfilt_post_cleanup(obd);
+
+ RETURN(rc);
+}
+
+#if 0
+static int filter_group_set_fs_flags(struct obd_device *obd, int group)
+{
+ struct filter_obd *filter = &obd->u.filter;
+ int rc = 0, i = 0;
+ ENTRY;
+
+ /* zero group is not longer valid. */
+ if (group== 0)
+ RETURN(rc);
+ for (i = 0; i < filter->fo_subdir_count; i++) {
+ struct dentry *dentry;
+ dentry = (filter->fo_subdirs + group)->dentry[i];
+ rc = fsfilt_set_fs_flags(obd, dentry->d_inode,
+ SM_DO_REC | SM_DO_COW);
+ if (rc)
+ RETURN(rc);
+ }
+ RETURN(rc);
+}
+#endif
+
+static int filter_post_fs_setup(struct obd_device *obd)
+{
+ struct filter_obd *filter = &obd->u.filter;
+ int rc = 0;
+
+ rc = fsfilt_post_setup(obd, filter->fo_dentry_O);
+
+ return rc;
+}
+
+/* mount the file system (secretly) */
+int filter_common_setup(struct obd_device *obd, obd_count len, void *buf,
+ char *option)
+{
+ struct lustre_cfg *lcfg = buf;
+ struct filter_obd *filter = &obd->u.filter;
+ struct lvfs_obd_ctxt *lvfs_ctxt = NULL;
+ struct vfsmount *mnt;
+ char *str;
+ char ns_name[48];
+ int rc = 0, i;
ENTRY;
- // XXX add a storage location for the logid for size changes
-#ifdef ENABLE_ORPHANS
- rc = llog_cat_initialize(obd, 1);
+ if ((LUSTRE_CFG_BUFLEN(lcfg, 1)) < 1 ||
+ (LUSTRE_CFG_BUFLEN(lcfg, 2) < 1))
+ RETURN(-EINVAL);
+
+ obd->obd_fsops = fsfilt_get_ops(lustre_cfg_string(lcfg, 2));
+ if (IS_ERR(obd->obd_fsops))
+ RETURN(PTR_ERR(obd->obd_fsops));
+
+ rc = lvfs_mount_fs(lustre_cfg_string(lcfg, 1),
+ lustre_cfg_string(lcfg, 2),
+ option, MS_NOATIME | MS_NODIRATIME, &lvfs_ctxt);
+ if (rc) {
+ CERROR("lvfs_mount_fs failed: rc = %d\n", rc);
+ GOTO(err_ops, rc);
+ }
+ LASSERT(lvfs_ctxt);
+
+ mnt = lvfs_ctxt->loc_mnt;
+ filter->fo_lvfs_ctxt = lvfs_ctxt;
+
+ if (LUSTRE_CFG_BUFLEN(lcfg, 3) > 0 && lustre_cfg_buf(lcfg, 3)) {
+ str = lustre_cfg_string(lcfg, 3);
+ if (*str == 'f') {
+ obd->obd_replayable = 1;
+ obd_sync_filter = 1;
+ CWARN("%s: recovery enabled\n", obd->obd_name);
+ } else {
+ if (*str != 'n') {
+ CERROR("unrecognised flag '%c'\n",
+ *str);
+ }
+ // XXX Robert? Why do we get errors here
+ // GOTO(err_mntput, rc = -EINVAL);
+ }
+ }
+
+ filter->fo_vfsmnt = mnt;
+ filter->fo_sb = mnt->mnt_sb;
+ filter->fo_fstype = mnt->mnt_sb->s_type->name;
+ CDEBUG(D_SUPER, "%s: mnt = %p\n", filter->fo_fstype, mnt);
+
+ OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
+ obd->obd_lvfs_ctxt.pwdmnt = mnt;
+ obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
+ obd->obd_lvfs_ctxt.fs = get_ds();
+ obd->obd_lvfs_ctxt.cb_ops = filter_lvfs_ops;
+
+ ll_clear_rdonly(ll_sbdev(filter->fo_sb));
+
+ rc = fsfilt_setup(obd, mnt->mnt_sb);
+ if (rc)
+ GOTO(err_mntput, rc);
+
+ sema_init(&filter->fo_init_lock, 1);
+ filter->fo_committed_group = 0;
+ rc = filter_prep(obd);
if (rc)
+ GOTO(err_mntput, rc);
+
+ filter->fo_destroys_in_progress = 0;
+ for (i = 0; i < 32; i++)
+ sema_init(&filter->fo_create_locks[i], 1);
+
+ spin_lock_init(&filter->fo_translock);
+ spin_lock_init(&filter->fo_objidlock);
+ INIT_LIST_HEAD(&filter->fo_export_list);
+ sema_init(&filter->fo_alloc_lock, 1);
+ spin_lock_init(&filter->fo_r_pages.oh_lock);
+ spin_lock_init(&filter->fo_w_pages.oh_lock);
+ spin_lock_init(&filter->fo_r_discont_pages.oh_lock);
+ spin_lock_init(&filter->fo_w_discont_pages.oh_lock);
+ spin_lock_init(&filter->fo_r_discont_blocks.oh_lock);
+ spin_lock_init(&filter->fo_w_discont_blocks.oh_lock);
+ filter->fo_readcache_max_filesize = FILTER_MAX_CACHE_SIZE;
+
+ INIT_LIST_HEAD(&filter->fo_llog_list);
+ spin_lock_init(&filter->fo_llog_list_lock);
+
+ sprintf(ns_name, "filter-%s", obd->obd_uuid.uuid);
+ obd->obd_namespace = ldlm_namespace_new(ns_name, LDLM_NAMESPACE_SERVER);
+
+ if (obd->obd_namespace == NULL)
+ GOTO(err_post, rc = -ENOMEM);
+ obd->obd_namespace->ns_lvbp = obd;
+ obd->obd_namespace->ns_lvbo = &filter_lvbo;
+ ldlm_register_intent(obd->obd_namespace, filter_intent_policy);
+
+ ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
+ "filter_ldlm_cb_client", &obd->obd_ldlm_client);
+
+ rc = obd_llog_cat_initialize(obd, &obd->obd_llogs, 1, CATLIST);
+ if (rc) {
CERROR("failed to setup llogging subsystems\n");
-#endif
+ GOTO(err_post, rc);
+ }
+ RETURN(0);
+
+err_post:
+ filter_post(obd);
+err_mntput:
+ unlock_kernel();
+ lvfs_umount_fs(filter->fo_lvfs_ctxt);
+ filter->fo_sb = 0;
+ lock_kernel();
+err_ops:
+ fsfilt_put_ops(obd->obd_fsops);
+ return rc;
+}
+
+static int filter_attach(struct obd_device *obd, obd_count len, void *data)
+{
+ struct lprocfs_static_vars lvars;
+ int rc;
+
+ lprocfs_init_vars(filter, &lvars);
+ rc = lprocfs_obd_attach(obd, lvars.obd_vars);
+ if (rc != 0)
+ return rc;
+
+ rc = lprocfs_alloc_obd_stats(obd, LPROC_FILTER_LAST);
+ if (rc != 0)
+ return rc;
+
+ /* Init obdfilter private stats here */
+ lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_READ_BYTES,
+ LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
+ lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_WRITE_BYTES,
+ LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
+
+ return lproc_filter_attach_seqstat(obd);
+}
+
+static int filter_detach(struct obd_device *dev)
+{
+ lprocfs_free_obd_stats(dev);
+ return lprocfs_obd_detach(dev);
+}
+
+static int filter_setup(struct obd_device *obd, obd_count len, void *buf)
+{
+ struct filter_obd *filter = &obd->u.filter;
+ struct lustre_cfg *lcfg = buf;
+ unsigned long page;
+ int rc;
+ ENTRY;
+
+ spin_lock_init(&filter->fo_denylist_lock);
+ INIT_LIST_HEAD(&filter->fo_denylist);
+
+ /* 2.6.9 selinux wants a full option page for do_kern_mount (bug6471) */
+ page = get_zeroed_page(GFP_KERNEL);
+ if (!page)
+ RETURN(-ENOMEM);
+
+ memcpy((void *)page, lustre_cfg_buf(lcfg, 4),
+ LUSTRE_CFG_BUFLEN(lcfg, 4));
+
+ /* all mount options including errors=remount-ro and asyncdel are passed
+ * using 4th lcfg param. And it is good, finally we have got rid of
+ * hardcoded fs types in the code. */
+ rc = filter_common_setup(obd, len, buf, (void *)page);
+ free_page(page);
+
+ if (rc)
+ RETURN(rc);
+ rc = filter_post_fs_setup(obd);
RETURN(rc);
}
static int filter_cleanup(struct obd_device *obd, int flags)
{
struct filter_obd *filter = &obd->u.filter;
+ ll_sbdev_type save_dev;
ENTRY;
if (flags & OBD_OPT_FAILOVER)
}
}
+ target_cleanup_recovery(obd);
+
ldlm_namespace_free(obd->obd_namespace, flags & OBD_OPT_FORCE);
if (filter->fo_sb == NULL)
RETURN(0);
+ save_dev = ll_sbdev(filter->fo_sb);
+ filter_post_fs_cleanup(obd);
filter_post(obd);
shrink_dcache_parent(filter->fo_sb->s_root);
filter->fo_sb = 0;
- if (atomic_read(&filter->fo_vfsmnt->mnt_count) > 1)
- CERROR("%s: mount point %p busy, mnt_count: %d\n",
- obd->obd_name, filter->fo_vfsmnt,
- atomic_read(&filter->fo_vfsmnt->mnt_count));
+ spin_lock(&filter->fo_denylist_lock);
+ while (!list_empty(&filter->fo_denylist)) {
+ deny_sec_t *p_deny_sec = list_entry(filter->fo_denylist.next,
+ deny_sec_t, list);
+ list_del(&p_deny_sec->list);
+ OBD_FREE(p_deny_sec, sizeof(*p_deny_sec));
+ }
+ spin_unlock(&filter->fo_denylist_lock);
unlock_kernel();
- mntput(filter->fo_vfsmnt);
+ lvfs_umount_fs(filter->fo_lvfs_ctxt);
//destroy_buffers(filter->fo_sb->s_dev);
filter->fo_sb = NULL;
fsfilt_put_ops(obd->obd_fsops);
lock_kernel();
- dev_clear_rdonly(2);
+ ll_clear_rdonly(save_dev);
RETURN(0);
}
-static int filter_attach(struct obd_device *obd, obd_count len, void *data)
+static int filter_process_config(struct obd_device *obd, obd_count len, void *buf)
{
- struct lprocfs_static_vars lvars;
- int rc;
-
- lprocfs_init_vars(filter, &lvars);
- rc = lprocfs_obd_attach(obd, lvars.obd_vars);
- if (rc != 0)
- return rc;
+ struct lustre_cfg *lcfg = buf;
+ struct filter_obd *filter = &obd->u.filter;
+ int rc = 0;
+ ENTRY;
- rc = lprocfs_alloc_obd_stats(obd, LPROC_FILTER_LAST);
- if (rc != 0)
- return rc;
+ switch(lcfg->lcfg_command) {
+ case LCFG_SET_SECURITY: {
+ if ((LUSTRE_CFG_BUFLEN(lcfg, 1) == 0) ||
+ (LUSTRE_CFG_BUFLEN(lcfg, 2) == 0))
+ GOTO(out, rc = -EINVAL);
+
+ if (!strcmp(lustre_cfg_string(lcfg, 1), "deny_sec")){
+ spin_lock(&filter->fo_denylist_lock);
+ rc = add_deny_security(lustre_cfg_string(lcfg, 2),
+ &filter->fo_denylist);
+ spin_unlock(&filter->fo_denylist_lock);
+ }else {
+ CERROR("Unrecognized key\n");
+ rc = -EINVAL;
+ }
+ break;
+ }
+ default: {
+ CERROR("Unknown command: %d\n", lcfg->lcfg_command);
+ GOTO(out, rc = -EINVAL);
- /* Init obdfilter private stats here */
- lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_READ_BYTES,
- LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
- lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_WRITE_BYTES,
- LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
- return rc;
+ }
+ }
+out:
+ RETURN(rc);
}
-static int filter_detach(struct obd_device *dev)
+static int filter_connect_post(struct obd_export *exp, unsigned initial,
+ unsigned long connect_flags)
{
- lprocfs_free_obd_stats(dev);
- return lprocfs_obd_detach(dev);
+ struct obd_device *obd = exp->exp_obd;
+ struct filter_export_data *fed;
+ char str[PTL_NALFMT_SIZE];
+ struct obd_llogs *llog;
+ struct llog_ctxt *ctxt;
+ int rc = 0;
+ ENTRY;
+
+ fed = &exp->exp_filter_data;
+ if (fed->fed_group < FILTER_MIN_GROUPS)
+ RETURN(0);
+
+ /* initialize llogs for connections from MDS */
+ llog = filter_grab_llog_for_group(obd, fed->fed_group, exp);
+ LASSERT(llog != NULL);
+
+ ctxt = llog_get_context(llog, LLOG_UNLINK_REPL_CTXT);
+ LASSERT(ctxt != NULL);
+
+ rc = llog_receptor_accept(ctxt, exp->exp_imp_reverse);
+
+ portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
+ exp->exp_connection->c_peer.peer_id.nid, str);
+
+ CDEBUG(D_OTHER, "%s: init llog ctxt for export "LPX64"/%s, group %d\n",
+ obd->obd_name, exp->exp_connection->c_peer.peer_id.nid,
+ str, fed->fed_group);
+
+ RETURN(rc);
}
/* nearly identical to mds_connect */
static int filter_connect(struct lustre_handle *conn, struct obd_device *obd,
- struct obd_uuid *cluuid)
+ struct obd_uuid *cluuid,
+ struct obd_connect_data *data,
+ unsigned long connect_flags)
{
struct obd_export *exp;
struct filter_export_data *fed;
struct filter_client_data *fcd = NULL;
struct filter_obd *filter = &obd->u.filter;
- int rc;
+ struct lvfs_run_ctxt saved;
+ int rc, group;
ENTRY;
if (conn == NULL || obd == NULL || cluuid == NULL)
spin_lock_init(&fed->fed_lock);
- if (!obd->obd_replayable)
- GOTO(cleanup, rc = 0);
+ /* connection from MDS */
+ group = connect_flags;
+ if (obd->obd_replayable) {
+ OBD_ALLOC(fcd, sizeof(*fcd));
+ if (!fcd) {
+ CERROR("filter: out of memory for client data\n");
+ GOTO(cleanup, rc = -ENOMEM);
+ }
- OBD_ALLOC(fcd, sizeof(*fcd));
- if (!fcd) {
- CERROR("filter: out of memory for client data\n");
- GOTO(cleanup, rc = -ENOMEM);
+ memcpy(fcd->fcd_uuid, cluuid, sizeof(fcd->fcd_uuid));
+ fed->fed_fcd = fcd;
+ fed->fed_fcd->fcd_group = group;
+ rc = filter_client_add(obd, filter, fed, -1);
+ if (rc)
+ GOTO(cleanup, rc);
}
-
- memcpy(fcd->fcd_uuid, cluuid, sizeof(fcd->fcd_uuid));
- fed->fed_fcd = fcd;
- fcd->fcd_mount_count = cpu_to_le64(filter->fo_fsd->fsd_mount_count);
-
- rc = filter_client_add(obd, filter, fed, -1);
-
+ CWARN("%s: Received MDS connection ("LPX64"); group %d\n",
+ obd->obd_name, exp->exp_handle.h_cookie, group);
+
+ if (group == 0)
+ GOTO(cleanup, rc);
+
+ if (fed->fed_group != 0 && fed->fed_group != group) {
+ char str[PTL_NALFMT_SIZE];
+ portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
+ exp->exp_connection->c_peer.peer_id.nid, str);
+ CERROR("!!! This export (nid "LPX64"/%s) used object group %d "
+ "earlier; now it's trying to use group %d! This could "
+ "be a bug in the MDS. Tell CFS.\n",
+ exp->exp_connection->c_peer.peer_id.nid, str,
+ fed->fed_group, group);
+ GOTO(cleanup, rc = -EPROTO);
+ }
+ fed->fed_group = group;
+
+ push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
+ rc = filter_read_groups(obd, group, 1);
+ pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
+ if (rc != 0) {
+ CERROR("can't read group %u\n", group);
+ GOTO(cleanup, rc);
+ }
+#if 0
+ rc = filter_group_set_fs_flags(obd, group);
+ if (rc != 0) {
+ CERROR("can't set kml flags %u\n", group);
+ GOTO(cleanup, rc);
+ }
+#endif
cleanup:
if (rc) {
if (fcd)
static int filter_precleanup(struct obd_device *obd, int flags)
{
+ struct filter_group_llog *log;
+ struct filter_obd *filter;
int rc = 0;
ENTRY;
-#ifdef ENABLE_ORPHANS
- rc = obd_llog_finish(obd, 0);
+ filter = &obd->u.filter;
+
+ spin_lock(&filter->fo_llog_list_lock);
+ while (!list_empty(&filter->fo_llog_list)) {
+ log = list_entry(filter->fo_llog_list.next,
+ struct filter_group_llog, list);
+ list_del(&log->list);
+ spin_unlock(&filter->fo_llog_list_lock);
+
+ rc = obd_llog_finish(obd, log->llogs, 0);
+ if (rc)
+ CERROR("failed to cleanup llogging subsystem for %u\n",
+ log->group);
+ OBD_FREE(log->llogs, sizeof(*(log->llogs)));
+ OBD_FREE(log, sizeof(*log));
+ spin_lock(&filter->fo_llog_list_lock);
+ }
+ spin_unlock(&filter->fo_llog_list_lock);
+
+ rc = obd_llog_finish(obd, &obd->obd_llogs, 0);
if (rc)
CERROR("failed to cleanup llogging subsystem\n");
-#endif
RETURN(rc);
}
+/* Do extra sanity checks for grant accounting. We do this at connect,
+ * disconnect, and statfs RPC time, so it shouldn't be too bad. We can
+ * always get rid of it or turn it off when we know accounting is good. */
+static void filter_grant_sanity_check(struct obd_device *obd, const char *func)
+{
+ struct filter_export_data *fed;
+ struct obd_export *exp;
+ obd_size maxsize = obd->obd_osfs.os_blocks * obd->obd_osfs.os_bsize;
+ obd_size tot_dirty = 0, tot_pending = 0, tot_granted = 0;
+ obd_size fo_tot_dirty, fo_tot_pending, fo_tot_granted;
+ int level = D_CACHE;
+
+ if (list_empty(&obd->obd_exports))
+ return;
+
+ spin_lock(&obd->obd_osfs_lock);
+ spin_lock(&obd->obd_dev_lock);
+ list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
+ fed = &exp->exp_filter_data;
+ if (fed->fed_grant < 0 || fed->fed_pending < 0 ||
+ fed->fed_dirty < 0)
+ level = D_ERROR;
+ if (maxsize > 0) { /* we may not have done a statfs yet */
+ LASSERTF(fed->fed_grant + fed->fed_pending <= maxsize,
+ "cli %s/%p %ld+%ld > "LPU64"\n",
+ exp->exp_client_uuid.uuid, exp,
+ fed->fed_grant, fed->fed_pending, maxsize);
+ LASSERTF(fed->fed_dirty <= maxsize,
+ "cli %s/%p %ld > "LPU64"\n",
+ exp->exp_client_uuid.uuid, exp,
+ fed->fed_dirty, maxsize);
+ }
+ CDEBUG(level, "%s: cli %s/%p dirty %ld pend %ld grant %ld\n",
+ obd->obd_name, exp->exp_client_uuid.uuid, exp,
+ fed->fed_dirty, fed->fed_pending, fed->fed_grant);
+ tot_granted += fed->fed_grant + fed->fed_pending;
+ tot_pending += fed->fed_pending;
+ tot_dirty += fed->fed_dirty;
+ }
+ fo_tot_granted = obd->u.filter.fo_tot_granted;
+ fo_tot_pending = obd->u.filter.fo_tot_pending;
+ fo_tot_dirty = obd->u.filter.fo_tot_dirty;
+ spin_unlock(&obd->obd_dev_lock);
+ spin_unlock(&obd->obd_osfs_lock);
+
+ /* Do these assertions outside the spinlocks so we don't kill system */
+ if (tot_granted != fo_tot_granted)
+ CERROR("%s: tot_granted "LPU64" != fo_tot_granted "LPU64"\n",
+ func, tot_granted, fo_tot_granted);
+ if (tot_pending != fo_tot_pending)
+ CERROR("%s: tot_pending "LPU64" != fo_tot_pending "LPU64"\n",
+ func, tot_pending, fo_tot_pending);
+ if (tot_dirty != fo_tot_dirty)
+ CERROR("%s: tot_dirty "LPU64" != fo_tot_dirty "LPU64"\n",
+ func, tot_dirty, fo_tot_dirty);
+ if (tot_pending > tot_granted)
+ CERROR("%s: tot_pending "LPU64" > tot_granted "LPU64"\n",
+ func, tot_pending, tot_granted);
+ if (tot_granted > maxsize)
+ CERROR("%s: tot_granted "LPU64" > maxsize "LPU64"\n",
+ func, tot_granted, maxsize);
+ if (tot_dirty > maxsize)
+ CERROR("%s: tot_dirty "LPU64" > maxsize "LPU64"\n",
+ func, tot_dirty, maxsize);
+}
+
+/* Remove this client from the grant accounting totals. We also remove
+ * the export from the obd device under the osfs and dev locks to ensure
+ * that the filter_grant_sanity_check() calculations are always valid.
+ * The client should do something similar when it invalidates its import. */
+static void filter_grant_discard(struct obd_export *exp)
+{
+ struct obd_device *obd = exp->exp_obd;
+ struct filter_obd *filter = &obd->u.filter;
+ struct filter_export_data *fed = &exp->exp_filter_data;
+ int level = D_CACHE;
+
+ spin_lock(&obd->obd_osfs_lock);
+ spin_lock(&exp->exp_obd->obd_dev_lock);
+ list_del_init(&exp->exp_obd_chain);
+ spin_unlock(&exp->exp_obd->obd_dev_lock);
+
+ if (fed->fed_dirty < 0 || fed->fed_grant < 0 || fed->fed_pending < 0)
+ level = D_ERROR;
+ CDEBUG(level, "%s: cli %s/%p dirty %ld pend %ld grant %ld\n",
+ obd->obd_name, exp->exp_client_uuid.uuid, exp,
+ fed->fed_dirty, fed->fed_pending, fed->fed_grant);
+
+ LASSERTF(filter->fo_tot_granted >= fed->fed_grant,
+ "%s: tot_granted "LPU64" cli %s/%p fed_grant %ld\n",
+ obd->obd_name, filter->fo_tot_granted,
+ exp->exp_client_uuid.uuid, exp, fed->fed_grant);
+ filter->fo_tot_granted -= fed->fed_grant;
+ LASSERTF(filter->fo_tot_pending >= fed->fed_pending,
+ "%s: tot_pending "LPU64" cli %s/%p fed_pending %ld\n",
+ obd->obd_name, filter->fo_tot_pending,
+ exp->exp_client_uuid.uuid, exp, fed->fed_pending);
+ LASSERTF(filter->fo_tot_dirty >= fed->fed_dirty,
+ "%s: tot_dirty "LPU64" cli %s/%p fed_dirty %ld\n",
+ obd->obd_name, filter->fo_tot_dirty,
+ exp->exp_client_uuid.uuid, exp, fed->fed_dirty);
+ filter->fo_tot_dirty -= fed->fed_dirty;
+ fed->fed_dirty = 0;
+ fed->fed_grant = 0;
+
+ spin_unlock(&obd->obd_osfs_lock);
+}
+
static int filter_destroy_export(struct obd_export *exp)
{
ENTRY;
+ if (exp->exp_filter_data.fed_pending)
+ CERROR("%s: cli %s/%p has %lu pending on destroyed export\n",
+ exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
+ exp, exp->exp_filter_data.fed_pending);
+
target_destroy_export(exp);
if (exp->exp_obd->obd_replayable)
filter_client_free(exp, exp->exp_flags);
+
+ filter_grant_discard(exp);
+ if (!(exp->exp_flags & OBD_OPT_FORCE))
+ filter_grant_sanity_check(exp->exp_obd, __FUNCTION__);
+
RETURN(0);
}
+static void filter_sync_llogs(struct obd_device *obd, struct obd_export *dexp)
+{
+ struct filter_group_llog *fglog, *nlog;
+ struct filter_obd *filter;
+ int worked = 0, group;
+ struct llog_ctxt *ctxt;
+ ENTRY;
+
+ filter = &obd->u.filter;
+
+ /* we can't sync log holding spinlock. also, we do not want to get
+ * into livelock. so we do following: loop over MDS's exports in
+ * group order and skip already synced llogs -bzzz */
+ do {
+ /* look for group with min. number, but > worked */
+ fglog = NULL;
+ group = 1 << 30;
+ spin_lock(&filter->fo_llog_list_lock);
+ list_for_each_entry(nlog, &filter->fo_llog_list, list) {
+
+ if (nlog->group <= worked) {
+ /* this group is already synced */
+ continue;
+ }
+
+ if (group < nlog->group) {
+ /* we have group with smaller number to sync */
+ continue;
+ }
+
+ /* store current minimal group */
+ fglog = nlog;
+ group = nlog->group;
+ }
+ spin_unlock(&filter->fo_llog_list_lock);
+
+ if (fglog == NULL)
+ break;
+
+ worked = fglog->group;
+ if (fglog->exp && (dexp == fglog->exp || dexp == NULL)) {
+ ctxt = llog_get_context(fglog->llogs,
+ LLOG_UNLINK_REPL_CTXT);
+ LASSERT(ctxt != NULL);
+ llog_sync(ctxt, fglog->exp);
+ }
+ } while (fglog != NULL);
+}
+
/* also incredibly similar to mds_disconnect */
-static int filter_disconnect(struct obd_export *exp, int flags)
+static int filter_disconnect(struct obd_export *exp, unsigned long flags)
{
+ struct obd_device *obd = exp->exp_obd;
unsigned long irqflags;
- struct llog_ctxt *ctxt;
int rc;
ENTRY;
LASSERT(exp);
- ldlm_cancel_locks_for_export(exp);
+ class_export_get(exp);
spin_lock_irqsave(&exp->exp_lock, irqflags);
exp->exp_flags = flags;
spin_unlock_irqrestore(&exp->exp_lock, irqflags);
- fsfilt_sync(exp->exp_obd, exp->exp_obd->u.filter.fo_sb);
- /* XXX cleanup preallocated inodes */
-
- /* flush any remaining cancel messages out to the target */
- ctxt = llog_get_context(exp->exp_obd, LLOG_UNLINK_REPL_CTXT);
- llog_sync(ctxt, exp);
+ if (!(flags & OBD_OPT_FORCE))
+ filter_grant_sanity_check(obd, __FUNCTION__);
+ filter_grant_discard(exp);
+ /* Disconnect early so that clients can't keep using export */
rc = class_disconnect(exp, flags);
+
+ ldlm_cancel_locks_for_export(exp);
+
+ fsfilt_sync(obd, obd->u.filter.fo_sb);
+
+ /* flush any remaining cancel messages out to the target */
+ filter_sync_llogs(obd, exp);
+ class_export_put(exp);
RETURN(rc);
}
{
struct dentry *dchild = NULL;
obd_gr group = 0;
+ ENTRY;
if (oa->o_valid & OBD_MD_FLGROUP)
group = oa->o_gr;
- dchild = filter_fid2dentry(obd, NULL, group, oa->o_id);
+ dchild = filter_id2dentry(obd, NULL, group, oa->o_id);
if (IS_ERR(dchild)) {
- CERROR("%s error looking up object: "LPU64"\n", what, oa->o_id);
+ CERROR("%s error looking up object: "LPU64"\n",
+ what, oa->o_id);
RETURN(dchild);
}
if (dchild->d_inode == NULL) {
- CERROR("%s on non-existent object: "LPU64"\n", what, oa->o_id);
+ CDEBUG(D_INFO, "%s: %s on non-existent object: "
+ LPU64"\n", obd->obd_name, what, oa->o_id);
f_dput(dchild);
RETURN(ERR_PTR(-ENOENT));
}
- return dchild;
+ RETURN(dchild);
}
static int filter_getattr(struct obd_export *exp, struct obdo *oa,
RETURN(rc);
}
-/* this is called from filter_truncate() until we have filter_punch() */
-static int filter_setattr(struct obd_export *exp, struct obdo *oa,
- struct lov_stripe_md *md, struct obd_trans_info *oti)
+int filter_setattr_internal(struct obd_export *exp, struct dentry *dentry,
+ struct obdo *oa, struct obd_trans_info *oti)
{
- struct obd_run_ctxt saved;
struct filter_obd *filter;
- struct dentry *dentry;
struct iattr iattr;
void *handle;
- int rc, rc2;
+ int rc, err;
ENTRY;
- LASSERT(oti != NULL);
-
- dentry = filter_oa2dentry(exp->exp_obd, oa);
- if (IS_ERR(dentry))
- RETURN(PTR_ERR(dentry));
-
+ LASSERT(dentry != NULL);
+ LASSERT(!IS_ERR(dentry));
+ LASSERT(dentry->d_inode != NULL);
+
filter = &exp->exp_obd->u.filter;
-
iattr_from_obdo(&iattr, oa, oa->o_valid);
- push_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
- lock_kernel();
-
if (iattr.ia_valid & ATTR_SIZE)
down(&dentry->d_inode->i_sem);
- handle = fsfilt_start(exp->exp_obd, dentry->d_inode, FSFILT_OP_SETATTR,
- oti);
+ handle = fsfilt_start(exp->exp_obd, dentry->d_inode,
+ FSFILT_OP_SETATTR, oti);
if (IS_ERR(handle))
GOTO(out_unlock, rc = PTR_ERR(handle));
/* XXX this could be a rwsem instead, if filter_preprw played along */
if (iattr.ia_valid & ATTR_ATTR_FLAG)
- rc = fsfilt_iocontrol(exp->exp_obd, dentry->d_inode, NULL,
- EXT3_IOC_SETFLAGS,
+ rc = fsfilt_iocontrol(exp->exp_obd, dentry->d_inode,
+ NULL, EXT3_IOC_SETFLAGS,
(long)&iattr.ia_attr_flags);
else
- rc = fsfilt_setattr(exp->exp_obd, dentry, handle, &iattr, 1);
+ rc = fsfilt_setattr(exp->exp_obd, dentry, handle,
+ &iattr, 1);
+
rc = filter_finish_transno(exp, oti, rc);
- rc2 = fsfilt_commit(exp->exp_obd, dentry->d_inode, handle, 0);
- if (rc2) {
- CERROR("error on commit, err = %d\n", rc2);
+
+ err = fsfilt_commit(exp->exp_obd, filter->fo_sb,
+ dentry->d_inode, handle,
+ exp->exp_sync);
+ if (err) {
+ CERROR("error on commit, err = %d\n", err);
if (!rc)
- rc = rc2;
+ rc = err;
}
+ EXIT;
+out_unlock:
+ if (iattr.ia_valid & ATTR_SIZE)
+ up(&dentry->d_inode->i_sem);
+ return rc;
+}
+
+/* this is called from filter_truncate() until we have filter_punch() */
+int filter_setattr(struct obd_export *exp, struct obdo *oa,
+ struct lov_stripe_md *md, struct obd_trans_info *oti)
+{
+ struct ldlm_res_id res_id = { .name = { oa->o_id, 0, oa->o_gr, 0 } };
+ struct ldlm_valblock_ops *ns_lvbo;
+ struct lvfs_run_ctxt saved;
+ struct filter_obd *filter;
+ struct ldlm_resource *res;
+ struct dentry *dentry;
+ int rc;
+ ENTRY;
+ LASSERT(oti != NULL);
+
+ filter = &exp->exp_obd->u.filter;
+ push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
+
+ /* make sure that object is allocated. */
+ dentry = filter_crow_object(exp->exp_obd,
+ oa->o_gr, oa->o_id);
+ if (IS_ERR(dentry))
+ GOTO(out_pop, rc = PTR_ERR(dentry));
+
+ lock_kernel();
+
+ /* setting objects attributes (including owner/group) */
+ rc = filter_setattr_internal(exp, dentry, oa, oti);
+ if (rc)
+ GOTO(out_unlock, rc);
+
+ res = ldlm_resource_get(exp->exp_obd->obd_namespace, NULL,
+ res_id, LDLM_EXTENT, 0);
+
+ if (res != NULL) {
+ ns_lvbo = res->lr_namespace->ns_lvbo;
+ if (ns_lvbo && ns_lvbo->lvbo_update)
+ rc = ns_lvbo->lvbo_update(res, NULL, 0, 0);
+ ldlm_resource_putref(res);
+ }
+
oa->o_valid = OBD_MD_FLID;
obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
+ EXIT;
out_unlock:
- if (iattr.ia_valid & ATTR_SIZE)
- up(&dentry->d_inode->i_sem);
unlock_kernel();
- pop_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
-
f_dput(dentry);
- RETURN(rc);
+out_pop:
+ pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
+ return rc;
}
/* XXX identical to osc_unpackmd */
RETURN(lsm_size);
}
-static void filter_destroy_precreated(struct obd_export *exp, struct obdo *oa,
- struct filter_obd *filter)
-{
- struct obdo doa; /* XXX obdo on stack */
- __u64 last, id;
- ENTRY;
- LASSERT(oa);
-
- memset(&doa, 0, sizeof(doa));
- if (oa->o_valid & OBD_MD_FLGROUP)
- doa.o_gr = oa->o_gr;
- else
- doa.o_gr = 0;
- doa.o_mode = S_IFREG;
- last = filter_last_id(filter, &doa); /* FIXME: object groups */
- CWARN("deleting orphan objects from "LPU64" to "LPU64"\n",
- oa->o_id + 1, last);
- for (id = oa->o_id + 1; id <= last; id++) {
- doa.o_id = id;
- filter_destroy(exp, &doa, NULL, NULL);
- }
- spin_lock(&filter->fo_objidlock);
- filter->fo_last_objids[0] = oa->o_id; /* FIXME: object groups */
- spin_unlock(&filter->fo_objidlock);
- EXIT;
-}
-
-/* returns a negative error or a nonnegative number of files to create */
-static int filter_should_precreate(struct obd_export *exp, struct obdo *oa,
- int group)
+static int filter_statfs(struct obd_device *obd, struct obd_statfs *osfs,
+ unsigned long max_age)
{
- struct obd_device *obd = exp->exp_obd;
struct filter_obd *filter = &obd->u.filter;
- int diff, rc;
+ int blockbits = filter->fo_sb->s_blocksize_bits;
+ int rc;
ENTRY;
- /* only precreate if group == 0 and o_id is specfied */
- if (group != 0 || oa->o_id == 0)
- RETURN(1);
+ /* at least try to account for cached pages. its still racey and
+ * might be under-reporting if clients haven't announced their
+ * caches with brw recently */
+ spin_lock(&obd->obd_osfs_lock);
+ rc = fsfilt_statfs(obd, filter->fo_sb, max_age);
+ memcpy(osfs, &obd->obd_osfs, sizeof(*osfs));
+ spin_unlock(&obd->obd_osfs_lock);
- diff = oa->o_id - filter_last_id(filter, oa);
- CDEBUG(D_INFO, "filter_last_id() = "LPU64" -> diff = %d\n",
- filter_last_id(filter, oa), diff);
- if (diff >= 0)
- RETURN(diff);
+ CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
+ " pending "LPU64" free "LPU64" avail "LPU64"\n",
+ filter->fo_tot_dirty, filter->fo_tot_granted,
+ filter->fo_tot_pending,
+ osfs->os_bfree << blockbits, osfs->os_bavail << blockbits);
- if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
- !(oa->o_flags & OBD_FL_DELORPHAN)) {
- CERROR("filter asked to delete %d objects, but DELORPHAN flag "
- "isn't set!\n", -diff);
- RETURN(0);
- }
+ filter_grant_sanity_check(obd, __FUNCTION__);
+
+ osfs->os_bavail -= min(osfs->os_bavail,
+ (filter->fo_tot_dirty + filter->fo_tot_pending +
+ osfs->os_bsize -1) >> blockbits);
- /* delete orphan request */
- filter_destroy_precreated(exp, oa, filter);
- rc = filter_update_last_objid(obd, group, 0);
- if (rc)
- CERROR("unable to write lastobjid, but orphans were deleted\n");
RETURN(rc);
}
-/* We rely on the fact that only one thread will be creating files in a given
- * group at a time, which is why we don't need an atomic filter_get_new_id.
- * Even if we had that atomic function, the following race would exist:
- *
- * thread 1: gets id x from filter_next_id
- * thread 2: gets id (x + 1) from filter_next_id
- * thread 2: creates object (x + 1)
- * thread 1: tries to create object x, gets -ENOSPC
- */
-static int filter_precreate(struct obd_device *obd, struct obdo *oa,
- obd_gr group, int *num)
+int filter_create_object(struct obd_device *obd, struct obdo *oa,
+ obd_gr group)
{
- struct lustre_handle parent_lockh;
+ struct dentry *dparent = NULL;
struct dentry *dchild = NULL;
struct filter_obd *filter;
- struct dentry *dparent;
- struct iattr attr;
- int err = 0, rc = 0, i;
- __u64 next_id;
- void *handle;
+ struct obd_statfs *osfs;
+ int cleanup_phase = 0;
+ int err = 0, rc = 0;
+ void *handle = NULL;
+ void *lock = NULL;
ENTRY;
filter = &obd->u.filter;
- for (i = 0; i < *num && err == 0; i++) {
- next_id = filter_last_id(filter, NULL) + 1;
- CDEBUG(D_INFO, "precreate objid "LPU64"\n", next_id);
+ OBD_ALLOC(osfs, sizeof(*osfs));
+ if (osfs == NULL)
+ RETURN(-ENOMEM);
+ rc = filter_statfs(obd, osfs, jiffies - HZ);
+ if (rc == 0 && osfs->os_bavail < (osfs->os_blocks >> 10)) {
+ CDEBUG(D_HA, "OST out of space! avail "LPU64"\n",
+ osfs->os_bavail << filter->fo_sb->s_blocksize_bits);
+ rc = -ENOSPC;
+ }
+ OBD_FREE(osfs, sizeof(*osfs));
+ if (rc)
+ RETURN(rc);
- dparent = filter_parent_lock(obd, group, next_id, LCK_PW,
- &parent_lockh);
- if (IS_ERR(dparent)) {
- rc = PTR_ERR(dparent);
- break;
- }
+ down(&filter->fo_create_locks[group]);
+
+ if (test_bit(group, &filter->fo_destroys_in_progress)) {
+ CWARN("%s: precreate aborted by destroy\n",
+ obd->obd_name);
+ GOTO(out, rc = -EALREADY);
+ }
- dchild = filter_fid2dentry(obd, dparent, group, next_id);
- if (IS_ERR(dchild))
- GOTO(cleanup_lock, rc = PTR_ERR(dchild));
+ CDEBUG(D_INFO, "precreate objid "LPU64"\n", oa->o_id);
- if (dchild->d_inode != NULL) {
- /* This would only happen if lastobjid was bad on disk*/
- CERROR("Serious error: objid %*s already exists; is "
- "this filesystem corrupt?\n",
- dchild->d_name.len, dchild->d_name.name);
- GOTO(cleanup_dchild, rc = -EEXIST);
- }
+ dparent = filter_parent_lock(obd, group, oa->o_id, &lock);
+ if (IS_ERR(dparent))
+ GOTO(cleanup, rc = PTR_ERR(dparent));
+ cleanup_phase = 1;
+
+ dchild = filter_id2dentry(obd, dparent, group, oa->o_id);
+ if (IS_ERR(dchild))
+ GOTO(cleanup, rc = PTR_ERR(dchild));
+ cleanup_phase = 2;
+
+ if (dchild->d_inode != NULL)
+ GOTO(cleanup, rc = 0);
+
+ handle = fsfilt_start_log(obd, dparent->d_inode,
+ FSFILT_OP_CREATE, NULL, 1);
+ if (IS_ERR(handle))
+ GOTO(cleanup, rc = PTR_ERR(handle));
+ cleanup_phase = 3;
- handle = fsfilt_start(obd, dparent->d_inode,
- FSFILT_OP_CREATE_LOG, NULL);
- if (IS_ERR(handle))
- GOTO(cleanup_dchild, rc = PTR_ERR(handle));
+ rc = ll_vfs_create(dparent->d_inode, dchild, S_IFREG, NULL);
+ if (rc) {
+ CERROR("create failed rc = %d\n", rc);
+ GOTO(cleanup, rc);
+ }
+
+ fsfilt_set_fs_flags(obd, dparent->d_inode, SM_DO_REC);
- rc = ll_vfs_create(dparent->d_inode, dchild, S_IFREG, NULL);
+ if (oa->o_id > filter_last_id(filter, group)) {
+ /*
+ * saving last created object id, it will be needed in recovery
+ * for deleting orphanes.
+ */
+ filter_set_last_id(filter, group, oa->o_id);
+
+ rc = filter_update_last_objid(obd, group, 0);
if (rc) {
- CERROR("create failed rc = %d\n", rc);
- GOTO(cleanup_commit, rc);
- } else if (oa != NULL &&
- (oa->o_valid & (OBD_MD_FLCTIME | OBD_MD_FLMTIME |
- OBD_MD_FLSIZE))) {
- iattr_from_obdo(&attr, oa, oa->o_valid);
- rc = fsfilt_setattr(obd, dchild, handle, &attr, 1);
- if (rc)
- CERROR("create setattr failed rc = %d\n", rc);
+ CERROR("unable to write lastobjid, but "
+ "orphans were deleted, err = %d\n",
+ rc);
+ rc = 0;
}
- filter_set_last_id(filter, NULL, next_id);
- err = filter_update_last_objid(obd, group, 0);
- if (err)
- CERROR("unable to write lastobjid but file created\n");
- cleanup_commit:
- err = fsfilt_commit(obd, dparent->d_inode, handle, 0);
+ }
+cleanup:
+ switch(cleanup_phase) {
+ case 3:
+ err = fsfilt_commit(obd, filter->fo_sb,
+ dparent->d_inode, handle, 0);
if (err) {
CERROR("error on commit, err = %d\n", err);
if (!rc)
rc = err;
}
- if (dchild->d_inode != NULL && oa != NULL)
- obdo_from_inode(oa, dchild->d_inode,
- FILTER_VALID_FLAGS);
- cleanup_dchild:
+ case 2:
f_dput(dchild);
- cleanup_lock:
- filter_parent_unlock(dparent, &parent_lockh, LCK_PW);
- if (rc)
- break;
- oa = NULL; /* oa applies for first iteration only */
+ case 1:
+ filter_parent_unlock(dparent, lock);
+ case 0:
+ break;
}
- *num = i;
- CDEBUG(D_INFO, "filter_precreate() created %d objects\n", i);
+ if (rc)
+ GOTO(out, rc);
+
+out:
+ up(&filter->fo_create_locks[group]);
RETURN(rc);
}
-static int filter_create(struct obd_export *exp, struct obdo *oa,
- struct lov_stripe_md **ea, struct obd_trans_info *oti)
+struct dentry *filter_crow_object(struct obd_device *obd,
+ __u64 ogr, __u64 oid)
{
- struct obd_device *obd = NULL;
- struct obd_run_ctxt saved;
- struct lov_stripe_md *lsm = NULL;
- obd_gr group = 0;
- int rc = 0, diff;
+ struct dentry *dentry;
+ struct obdo *oa;
+ int rc = 0;
ENTRY;
- if (oa->o_valid & OBD_MD_FLGROUP)
- group = oa->o_gr;
+ /* check if object is already allocated */
+ dentry = filter_id2dentry(obd, NULL, ogr, oid);
+ if (IS_ERR(dentry))
+ RETURN(dentry);
- CDEBUG(D_INFO, "filter_create(od->o_gr="LPU64",od->o_id="LPU64")\n",
- group, oa->o_id);
- if (ea != NULL) {
- lsm = *ea;
- if (lsm == NULL) {
- rc = obd_alloc_memmd(exp, &lsm);
- if (rc < 0)
- RETURN(rc);
- }
+ if (dentry->d_inode)
+ RETURN(dentry);
+
+ f_dput(dentry);
+
+ /* allocate object as it does not exist */
+ oa = obdo_alloc();
+ if (oa == NULL)
+ RETURN(ERR_PTR(-ENOMEM));
+
+ oa->o_id = oid;
+ oa->o_gr = ogr;
+ oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
+
+ CDEBUG(D_INODE, "OSS object "LPU64"/"LPU64
+ " does not exists - allocate now\n",
+ oid, ogr);
+
+ rc = filter_create_object(obd, oa, oa->o_gr);
+ if (rc) {
+ CERROR("cannot create OSS object "LPU64"/"LPU64
+ ", err = %d\n", oa->o_id, oa->o_gr, rc);
+ GOTO(out_free_oa, dentry = ERR_PTR(rc));
+ }
+
+ /* lookup for just created object and return it to caller */
+ dentry = filter_id2dentry(obd, NULL, ogr, oid);
+ if (IS_ERR(dentry))
+ GOTO(out_free_oa, dentry);
+
+ if (dentry->d_inode == NULL) {
+ f_dput(dentry);
+ dentry = ERR_PTR(-ENOENT);
+ CERROR("cannot find just created OSS object "
+ LPU64"/"LPU64" err = %d\n", oid,
+ ogr, (int)PTR_ERR(dentry));
+ GOTO(out_free_oa, dentry);
}
+ EXIT;
+out_free_oa:
+ obdo_free(oa);
+ return dentry;
+}
+
+static int
+filter_clear_orphans(struct obd_export *exp, struct obdo *oa)
+{
+ struct obd_device *obd = NULL;
+ struct filter_obd *filter;
+ struct obdo *doa = NULL;
+ int rc = 0, orphans;
+ __u64 last, id;
+ ENTRY;
+
+ LASSERT(oa);
+ LASSERT(oa->o_gr != 0);
+ LASSERT(oa->o_valid & OBD_MD_FLGROUP);
+
obd = exp->exp_obd;
- push_ctxt(&saved, &obd->obd_ctxt, NULL);
+ filter = &obd->u.filter;
+
+ last = filter_last_id(filter, oa->o_gr);
+ orphans = last - oa->o_id;
+
+ if (orphans <= 0)
+ RETURN(0);
+
+ doa = obdo_alloc();
+ if (doa == NULL)
+ RETURN(-ENOMEM);
+
+ doa->o_gr = oa->o_gr;
+ doa->o_mode = S_IFREG;
+ doa->o_valid = oa->o_valid & (OBD_MD_FLGROUP | OBD_MD_FLID);
- diff = filter_should_precreate(exp, oa, group);
- if (diff > 0) {
- oa->o_id = filter_last_id(&obd->u.filter, oa);
- rc = filter_precreate(obd, oa, group, &diff);
- oa->o_id += diff;
- oa->o_valid = OBD_MD_FLID;
+ set_bit(doa->o_gr, &filter->fo_destroys_in_progress);
+ down(&filter->fo_create_locks[doa->o_gr]);
+ if (!test_bit(doa->o_gr, &filter->fo_destroys_in_progress)) {
+ CERROR("%s:["LPU64"] destroy_in_progress already cleared\n",
+ exp->exp_obd->obd_name, doa->o_gr);
+ up(&filter->fo_create_locks[doa->o_gr]);
+ GOTO(out_free_doa, 0);
}
- pop_ctxt(&saved, &obd->obd_ctxt, NULL);
- if (rc && ea != NULL && *ea != lsm) {
- obd_free_memmd(exp, &lsm);
- } else if (rc == 0 && ea != NULL) {
- /* XXX LOV STACKING: the lsm that is passed to us from
- * LOV does not have valid lsm_oinfo data structs, so
- * don't go touching that. This needs to be fixed in a
- * big way. */
- lsm->lsm_object_id = oa->o_id;
- *ea = lsm;
+ CWARN("%s:["LPU64"] deleting orphan objects from "LPU64" to "
+ LPU64"\n", exp->exp_obd->obd_name, doa->o_gr,
+ oa->o_id + 1, last);
+
+ for (id = oa->o_id + 1; id <= last; id++) {
+ doa->o_id = id;
+ filter_destroy(exp, doa, NULL, NULL);
}
- RETURN(rc);
+ CDEBUG(D_HA, "%s:["LPU64"] after destroy: set last_objids = "
+ LPU64"\n", exp->exp_obd->obd_name, doa->o_gr, oa->o_id);
+
+ /* return next free id to be used as a new start of sequence -bzzz */
+ oa->o_id = last + 1;
+
+ filter_set_last_id(filter, oa->o_gr, oa->o_id);
+ clear_bit(doa->o_gr, &filter->fo_destroys_in_progress);
+ up(&filter->fo_create_locks[oa->o_gr]);
+
+ EXIT;
+out_free_doa:
+ obdo_free(doa);
+ return rc;
+}
+
+/*
+ * by now this function is only needed as entry point for deleting orphanes on
+ * OSS as objects are created on first write attempt. --umka
+ */
+static int
+filter_create(struct obd_export *exp, struct obdo *oa, void *acl,
+ int acl_size, struct lov_stripe_md **ea,
+ struct obd_trans_info *oti)
+{
+ struct filter_export_data *fed;
+ struct obd_device *obd = NULL;
+ int group = oa->o_gr, rc = 0;
+ struct lvfs_run_ctxt saved;
+ struct filter_obd *filter;
+ char str[PTL_NALFMT_SIZE];
+ ENTRY;
+
+ LASSERT(acl == NULL && acl_size == 0);
+
+ if (!(oa->o_valid & OBD_MD_FLGROUP) || group == 0) {
+ portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
+ exp->exp_connection->c_peer.peer_id.nid, str);
+ CERROR("!!! nid "LPX64"/%s sent invalid object group %d\n",
+ exp->exp_connection->c_peer.peer_id.nid, str, group);
+ RETURN(-EINVAL);
+ }
+
+ obd = exp->exp_obd;
+ fed = &exp->exp_filter_data;
+ filter = &obd->u.filter;
+
+ if (fed->fed_group != group) {
+ portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
+ exp->exp_connection->c_peer.peer_id.nid, str);
+ CERROR("!!! this export (nid "LPX64"/%s) used object group %d "
+ "earlier; now it's trying to use group %d! This could "
+ "be a bug in the MDS. Tell CFS.\n",
+ exp->exp_connection->c_peer.peer_id.nid, str,
+ fed->fed_group, group);
+ RETURN(-ENOTUNIQ);
+ }
+
+ CDEBUG(D_INFO, "filter_create(od->o_gr=%d,od->o_id="LPU64")\n",
+ group, oa->o_id);
+
+ obd = exp->exp_obd;
+ push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
+
+ LASSERT((oa->o_valid & OBD_MD_FLFLAGS) &&
+ (oa->o_flags == OBD_FL_DELORPHAN));
+
+ rc = filter_clear_orphans(exp, oa);
+ if (rc) {
+ CERROR("cannot clear orphanes starting from "
+ LPU64", err = %d\n", oa->o_id, rc);
+ } else {
+ rc = filter_update_last_objid(obd, group, 0);
+ if (rc) {
+ CERROR("unable to write lastobjid, but "
+ "orphans were deleted, err = %d\n",
+ rc);
+ }
+ }
+ pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
+
+ RETURN(0);
}
static int filter_destroy(struct obd_export *exp, struct obdo *oa,
struct lov_stripe_md *ea, struct obd_trans_info *oti)
{
- struct obd_device *obd;
- struct filter_obd *filter;
+ int rc, rc2, cleanup_phase = 0, have_prepared = 0;
struct dentry *dchild = NULL, *dparent = NULL;
- struct obd_run_ctxt saved;
- void *handle = NULL;
- struct lustre_handle parent_lockh;
struct llog_cookie *fcc = NULL;
- int rc, rc2, cleanup_phase = 0, have_prepared = 0;
- obd_gr group = 0;
+ struct lvfs_run_ctxt saved;
+ struct filter_obd *filter;
+ struct obd_device *obd;
+ void *handle = NULL;
+ void *lock = NULL;
+ struct iattr iattr;
ENTRY;
- if (oa->o_valid & OBD_MD_FLGROUP)
- group = oa->o_gr;
+ LASSERT(oa->o_valid & OBD_MD_FLGROUP);
obd = exp->exp_obd;
filter = &obd->u.filter;
- push_ctxt(&saved, &obd->obd_ctxt, NULL);
+ push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
acquire_locks:
- dparent = filter_parent_lock(obd, group, oa->o_id, LCK_PW,
- &parent_lockh);
+ dparent = filter_parent_lock(obd, oa->o_gr, oa->o_id, &lock);
if (IS_ERR(dparent))
GOTO(cleanup, rc = PTR_ERR(dparent));
cleanup_phase = 1;
- dchild = filter_fid2dentry(obd, dparent, group, oa->o_id);
+ dchild = filter_id2dentry(obd, dparent, oa->o_gr, oa->o_id);
if (IS_ERR(dchild))
- GOTO(cleanup, rc = -ENOENT);
+ GOTO(cleanup, rc = PTR_ERR(dchild));
cleanup_phase = 2;
if (dchild->d_inode == NULL) {
- CERROR("destroying non-existent object "LPU64"\n", oa->o_id);
+ CDEBUG(D_INODE, "destroying non-existent object "LPU64"\n",
+ oa->o_id);
GOTO(cleanup, rc = -ENOENT);
}
* complication of condition the above code to skip it on the
* second time through. */
f_dput(dchild);
- filter_parent_unlock(dparent, &parent_lockh, LCK_PW);
+ filter_parent_unlock(dparent, lock);
- filter_prepare_destroy(obd, oa->o_id);
+ filter_prepare_destroy(obd, oa->o_id, oa->o_gr);
have_prepared = 1;
goto acquire_locks;
}
- handle = fsfilt_start(obd, dparent->d_inode, FSFILT_OP_UNLINK_LOG, oti);
- if (IS_ERR(handle))
- GOTO(cleanup, rc = PTR_ERR(handle));
- cleanup_phase = 3;
-
/* Our MDC connection is established by the MDS to us */
if (oa->o_valid & OBD_MD_FLCOOKIE) {
OBD_ALLOC(fcc, sizeof(*fcc));
memcpy(fcc, obdo_logcookie(oa), sizeof(*fcc));
}
+ /* we're gonna truncate it first in order to avoid possible
+ * deadlock:
+ * P1 P2
+ * open trasaction open transaction
+ * down(i_zombie) down(i_zombie)
+ * restart transaction
+ * (see BUG 4180) -bzzz
+ */
+ down(&dchild->d_inode->i_sem);
+ handle = fsfilt_start_log(obd, dparent->d_inode,FSFILT_OP_SETATTR,NULL,1);
+ if (IS_ERR(handle)) {
+ up(&dchild->d_inode->i_sem);
+ GOTO(cleanup, rc = PTR_ERR(handle));
+ }
+
+ iattr.ia_valid = ATTR_SIZE;
+ iattr.ia_size = 0;
+ rc = fsfilt_setattr(obd, dchild, handle, &iattr, 1);
+
+ rc2 = fsfilt_commit(obd, filter->fo_sb, dparent->d_inode, handle, 0);
+ up(&dchild->d_inode->i_sem);
+ if (rc)
+ GOTO(cleanup, rc);
+ if (rc2)
+ GOTO(cleanup, rc = rc2);
+
+ handle = fsfilt_start_log(obd, dparent->d_inode,FSFILT_OP_UNLINK,oti,1);
+ if (IS_ERR(handle))
+ GOTO(cleanup, rc = PTR_ERR(handle));
+
+ cleanup_phase = 3;
+
rc = filter_destroy_internal(obd, oa->o_id, dparent, dchild);
cleanup:
switch(cleanup_phase) {
case 3:
- if (fcc != NULL)
- fsfilt_add_journal_cb(obd, 0, oti->oti_handle,
+ if (fcc != NULL) {
+ fsfilt_add_journal_cb(obd, filter->fo_sb, 0,
+ oti ? oti->oti_handle : handle,
filter_cancel_cookies_cb, fcc);
+ }
rc = filter_finish_transno(exp, oti, rc);
- rc2 = fsfilt_commit(obd, dparent->d_inode, handle, 0);
+ rc2 = fsfilt_commit(obd, filter->fo_sb, dparent->d_inode,
+ handle, exp->exp_sync);
if (rc2) {
CERROR("error on commit, err = %d\n", rc2);
if (!rc)
case 2:
f_dput(dchild);
case 1:
- if (rc || oti == NULL) {
- filter_parent_unlock(dparent, &parent_lockh, LCK_PW);
- } else {
- memcpy(&oti->oti_ack_locks[0].lock, &parent_lockh,
- sizeof(parent_lockh));
- oti->oti_ack_locks[0].mode = LCK_PW;
- }
+ filter_parent_unlock(dparent, lock);
case 0:
- pop_ctxt(&saved, &obd->obd_ctxt, NULL);
+ pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
break;
default:
CERROR("invalid cleanup_phase %d\n", cleanup_phase);
CERROR("PUNCH not supported, only truncate: end = "LPX64"\n",
end);
- CDEBUG(D_INODE, "calling truncate for object "LPU64", valid = %x, "
+ CDEBUG(D_INODE, "calling truncate for object "LPU64", valid = "LPU64", "
"o_size = "LPD64"\n", oa->o_id, oa->o_valid, start);
oa->o_size = start;
error = filter_setattr(exp, oa, NULL, oti);
static int filter_sync(struct obd_export *exp, struct obdo *oa,
struct lov_stripe_md *lsm, obd_off start, obd_off end)
{
- struct obd_run_ctxt saved;
+ struct obd_device *obd = exp->exp_obd;
+ struct lvfs_run_ctxt saved;
struct filter_obd *filter;
struct dentry *dentry;
int rc, rc2;
ENTRY;
- filter = &exp->exp_obd->u.filter;
+ filter = &obd->u.filter;
/* an objid of zero is taken to mean "sync whole filesystem" */
- if (!oa || !oa->o_valid & OBD_MD_FLID) {
- rc = fsfilt_sync(exp->exp_obd, filter->fo_sb);
- GOTO(out_exp, rc);
+ if (!oa || !(oa->o_valid & OBD_MD_FLID)) {
+ rc = fsfilt_sync(obd, filter->fo_sb);
+ /* flush any remaining cancel messages out to the target */
+ filter_sync_llogs(obd, NULL);
+ RETURN(rc);
}
- dentry = filter_oa2dentry(exp->exp_obd, oa);
+ dentry = filter_oa2dentry(obd, oa);
if (IS_ERR(dentry))
- GOTO(out_exp, rc = PTR_ERR(dentry));
+ RETURN(PTR_ERR(dentry));
- push_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
+ push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
down(&dentry->d_inode->i_sem);
- rc = filemap_fdatasync(dentry->d_inode->i_mapping);
+ rc = filemap_fdatawrite(dentry->d_inode->i_mapping);
if (rc == 0) {
/* just any file to grab fsync method - "file" arg unused */
struct file *file = filter->fo_rcvd_filp;
oa->o_valid = OBD_MD_FLID;
obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
- pop_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
+ pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
f_dput(dentry);
-out_exp:
RETURN(rc);
}
-static int filter_statfs(struct obd_device *obd, struct obd_statfs *osfs,
- unsigned long max_age)
-{
- ENTRY;
- RETURN(fsfilt_statfs(obd, obd->u.filter.fo_sb, osfs));
-}
-
static int filter_get_info(struct obd_export *exp, __u32 keylen,
void *key, __u32 *vallen, void *val)
{
+ struct filter_export_data *fed = &exp->exp_filter_data;
struct obd_device *obd;
ENTRY;
if (keylen >= strlen("last_id") && memcmp(key, "last_id", 7) == 0) {
obd_id *last_id = val;
- /* FIXME: object groups */
- *last_id = filter_last_id(&obd->u.filter, 0);
+ *last_id = filter_last_id(&obd->u.filter, fed->fed_group);
+ RETURN(0);
+ }
+ if (keylen >= strlen("reint_log") && memcmp(key, "reint_log", 9) == 0) {
+ /*Get log_context handle*/
+ unsigned long *llh_handle = val;
+ *vallen = sizeof(unsigned long);
+ *llh_handle = (unsigned long)obd->obd_llog_ctxt[LLOG_REINT_ORIG_CTXT];
+ RETURN(0);
+ }
+ if (keylen >= strlen("cache_sb") && memcmp(key, "cache_sb", 8) == 0) {
+ /*Get log_context handle*/
+ unsigned long *sb = val;
+ *vallen = sizeof(unsigned long);
+ *sb = (unsigned long)obd->u.filter.fo_sb;
RETURN(0);
}
+
CDEBUG(D_IOCTL, "invalid key\n");
RETURN(-EINVAL);
}
-static int filter_set_info(struct obd_export *exp, __u32 keylen,
- void *key, __u32 vallen, void *val)
+struct obd_llogs *filter_grab_llog_for_group(struct obd_device *obd, int group,
+ struct obd_export *export)
{
- struct obd_device *obd;
- struct lustre_handle conn;
-#ifdef ENABLE_ORPHANS
+ struct filter_group_llog *fglog, *nlog;
+ char name[32] = "CATLIST";
+ struct filter_obd *filter;
struct llog_ctxt *ctxt;
-#endif
- int rc = 0;
- ENTRY;
+ struct list_head *cur;
+ int rc;
- conn.cookie = exp->exp_handle.h_cookie;
+ filter = &obd->u.filter;
- obd = exp->exp_obd;
- if (obd == NULL) {
- CDEBUG(D_IOCTL, "invalid exp %p cookie "LPX64"\n",
- exp, conn.cookie);
- RETURN(-EINVAL);
+ spin_lock(&filter->fo_llog_list_lock);
+ list_for_each(cur, &filter->fo_llog_list) {
+ fglog = list_entry(cur, struct filter_group_llog, list);
+ if (fglog->group == group) {
+ if (!(fglog->exp == NULL || fglog->exp == export || export == NULL))
+ CWARN("%s: export for group %d changes: 0x%p -> 0x%p\n",
+ obd->obd_name, group, fglog->exp, export);
+ spin_unlock(&filter->fo_llog_list_lock);
+ goto init;
+ }
}
+ spin_unlock(&filter->fo_llog_list_lock);
- if (keylen < strlen("mds_conn") ||
- memcmp(key, "mds_conn", keylen) != 0)
- RETURN(-EINVAL);
+ if (export == NULL)
+ RETURN(NULL);
- CWARN("Received MDS connection ("LPX64")\n", conn.cookie);
- memcpy(&obd->u.filter.fo_mdc_conn, &conn, sizeof(conn));
-#ifdef ENABLE_ORPHANS
- ctxt = llog_get_context(obd, LLOG_UNLINK_REPL_CTXT);
- rc = llog_receptor_accept(ctxt, exp->exp_imp_reverse);
-#endif
- RETURN(rc);
+ OBD_ALLOC(fglog, sizeof(*fglog));
+ if (fglog == NULL)
+ RETURN(NULL);
+ fglog->group = group;
+
+ OBD_ALLOC(fglog->llogs, sizeof(struct obd_llogs));
+ if (fglog->llogs == NULL) {
+ OBD_FREE(fglog, sizeof(*fglog));
+ RETURN(NULL);
+ }
+
+ spin_lock(&filter->fo_llog_list_lock);
+ list_for_each(cur, &filter->fo_llog_list) {
+ nlog = list_entry(cur, struct filter_group_llog, list);
+ LASSERT(nlog->group != group);
+ }
+ list_add(&fglog->list, &filter->fo_llog_list);
+ spin_unlock(&filter->fo_llog_list_lock);
+
+ rc = obd_llog_cat_initialize(obd, fglog->llogs, 1, name);
+ if (rc) {
+ OBD_FREE(fglog->llogs, sizeof(*(fglog->llogs)));
+ OBD_FREE(fglog, sizeof(*fglog));
+ RETURN(NULL);
+ }
+
+init:
+ if (export) {
+ fglog->exp = export;
+ ctxt = llog_get_context(fglog->llogs, LLOG_UNLINK_REPL_CTXT);
+ LASSERT(ctxt != NULL);
+
+ llog_receptor_accept(ctxt, export->exp_imp_reverse);
+ }
+ CDEBUG(D_OTHER, "%s: new llog 0x%p for group %u\n",
+ obd->obd_name, fglog->llogs, group);
+
+ RETURN(fglog->llogs);
}
int filter_iocontrol(unsigned int cmd, struct obd_export *exp,
switch (cmd) {
case OBD_IOC_ABORT_RECOVERY:
- CERROR("aborting recovery for device %s\n", obd->obd_name);
- target_abort_recovery(obd);
+ target_stop_recovery_thread(obd);
RETURN(0);
case OBD_IOC_SET_READONLY: {
BDEVNAME_DECLARE_STORAGE(tmp);
CERROR("setting device %s read-only\n",
ll_bdevname(sb, tmp));
-
+
handle = fsfilt_start(obd, inode, FSFILT_OP_MKNOD, NULL);
LASSERT(handle);
- (void)fsfilt_commit(obd, inode, handle, 1);
+ (void)fsfilt_commit(obd, sb, inode, handle, 1);
- dev_set_rdonly(ll_sbdev(obd->u.filter.fo_sb), 2);
+ ll_set_rdonly(ll_sbdev(obd->u.filter.fo_sb));
RETURN(0);
}
case OBD_IOC_CATLOGLIST: {
- rc = llog_catlog_list(obd, 1, data);
+ rc = llog_catalog_list(obd, 1, data);
RETURN(rc);
}
-
case OBD_IOC_LLOG_CANCEL:
- case OBD_IOC_LLOG_REMOVE:
+ case OBD_IOC_LLOG_REMOVE:
case OBD_IOC_LLOG_INFO:
case OBD_IOC_LLOG_PRINT: {
/* FIXME to be finished */
RETURN(-EOPNOTSUPP);
/*
struct llog_ctxt *ctxt = NULL;
-
- push_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_ctxt, NULL);
+
+ push_ctxt(&saved, &ctxt->loc_ctxt, NULL);
rc = llog_ioctl(ctxt, cmd, data);
- pop_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_ctxt, NULL);
-
+ pop_ctxt(&saved, &ctxt->loc_ctxt, NULL);
+
RETURN(rc);
*/
}
static struct llog_operations filter_unlink_repl_logops;
static struct llog_operations filter_size_orig_logops = {
lop_setup: llog_obd_origin_setup,
- lop_cleanup: llog_obd_origin_cleanup,
- lop_add: llog_obd_origin_add
+ lop_cleanup: llog_catalog_cleanup,
+ lop_add: llog_catalog_add,
};
-static int filter_llog_init(struct obd_device *obd, struct obd_device *tgt,
- int count, struct llog_logid *logid)
+static int filter_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
+ struct obd_device *tgt, int count,
+ struct llog_catid *catid)
{
struct llog_ctxt *ctxt;
int rc;
ENTRY;
-
+
filter_unlink_repl_logops = llog_client_ops;
filter_unlink_repl_logops.lop_cancel = llog_obd_repl_cancel;
filter_unlink_repl_logops.lop_connect = llog_repl_connect;
filter_unlink_repl_logops.lop_sync = llog_obd_repl_sync;
- rc = llog_setup(obd, LLOG_UNLINK_REPL_CTXT, tgt, 0, NULL,
+ rc = obd_llog_setup(obd, llogs, LLOG_UNLINK_REPL_CTXT, tgt, 0, NULL,
&filter_unlink_repl_logops);
if (rc)
RETURN(rc);
/* FIXME - assign unlink_cb for filter's recovery */
- ctxt = llog_get_context(obd, LLOG_UNLINK_REPL_CTXT);
+ ctxt = llog_get_context(llogs, LLOG_UNLINK_REPL_CTXT);
ctxt->llog_proc_cb = filter_recov_log_unlink_cb;
- rc = llog_setup(obd, LLOG_SIZE_ORIG_CTXT, tgt, 0, NULL,
- &filter_size_orig_logops);
+ /* FIXME - count should be 1 to setup size log */
+ rc = obd_llog_setup(obd, llogs, LLOG_SIZE_ORIG_CTXT, tgt, 0,
+ &catid->lci_logid, &filter_size_orig_logops);
RETURN(rc);
}
-static int filter_llog_finish(struct obd_device *obd, int count)
+static int filter_llog_finish(struct obd_device *obd,
+ struct obd_llogs *llogs, int count)
{
int rc;
ENTRY;
-
- rc = llog_cleanup(llog_get_context(obd, LLOG_UNLINK_REPL_CTXT));
+
+ rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_UNLINK_REPL_CTXT));
if (rc)
RETURN(rc);
- rc = llog_cleanup(llog_get_context(obd, LLOG_SIZE_ORIG_CTXT));
+ rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_SIZE_ORIG_CTXT));
+ RETURN(rc);
+}
+
+static int filter_llog_connect(struct obd_export *exp,
+ struct llogd_conn_body *body)
+{
+ struct obd_device *obd = exp->exp_obd;
+ struct llog_ctxt *ctxt;
+ struct obd_llogs *llog;
+ int rc;
+ ENTRY;
+
+ CDEBUG(D_OTHER, "handle connect for %s: %u/%u/%u\n", obd->obd_name,
+ (unsigned) body->lgdc_logid.lgl_ogr,
+ (unsigned) body->lgdc_logid.lgl_oid,
+ (unsigned) body->lgdc_logid.lgl_ogen);
+ llog = filter_grab_llog_for_group(obd, body->lgdc_logid.lgl_ogr, exp);
+ LASSERT(llog != NULL);
+ ctxt = llog_get_context(llog, body->lgdc_ctxt_idx);
+ rc = llog_connect(ctxt, 1, &body->lgdc_logid,
+ &body->lgdc_gen, NULL);
+ if (rc != 0)
+ CERROR("failed to connect\n");
+
RETURN(rc);
}
-static struct dentry *filter_lvfs_fid2dentry(__u64 id, __u32 gen, __u64 gr,
- void *data)
+static struct dentry *filter_lvfs_id2dentry(__u64 id, __u32 gen,
+ __u64 gr, void *data)
{
- return filter_fid2dentry(data, NULL, gr, id);
+ return filter_id2dentry(data, NULL, gr, id);
}
static struct lvfs_callback_ops filter_lvfs_ops = {
- l_fid2dentry: filter_lvfs_fid2dentry,
+ l_id2dentry: filter_lvfs_id2dentry,
};
static struct obd_ops filter_obd_ops = {
- o_owner: THIS_MODULE,
- o_attach: filter_attach,
- o_detach: filter_detach,
- o_get_info: filter_get_info,
- o_set_info: filter_set_info,
- o_setup: filter_setup,
- o_postsetup: filter_postsetup,
- o_precleanup: filter_precleanup,
- o_cleanup: filter_cleanup,
- o_connect: filter_connect,
- o_disconnect: filter_disconnect,
- o_statfs: filter_statfs,
- o_getattr: filter_getattr,
- o_unpackmd: filter_unpackmd,
- o_create: filter_create,
- o_setattr: filter_setattr,
- o_destroy: filter_destroy,
- o_brw: filter_brw,
- o_punch: filter_truncate,
- o_sync: filter_sync,
- o_preprw: filter_preprw,
- o_commitrw: filter_commitrw,
- o_destroy_export: filter_destroy_export,
- o_llog_init: filter_llog_init,
- o_llog_finish: filter_llog_finish,
- o_iocontrol: filter_iocontrol,
+ .o_owner = THIS_MODULE,
+ .o_attach = filter_attach,
+ .o_detach = filter_detach,
+ .o_get_info = filter_get_info,
+ .o_setup = filter_setup,
+ .o_precleanup = filter_precleanup,
+ .o_cleanup = filter_cleanup,
+ .o_process_config = filter_process_config,
+ .o_connect = filter_connect,
+ .o_connect_post = filter_connect_post,
+ .o_disconnect = filter_disconnect,
+ .o_statfs = filter_statfs,
+ .o_getattr = filter_getattr,
+ .o_unpackmd = filter_unpackmd,
+ .o_create = filter_create,
+ .o_setattr = filter_setattr,
+ .o_destroy = filter_destroy,
+ .o_brw = filter_brw,
+ .o_punch = filter_truncate,
+ .o_sync = filter_sync,
+ .o_preprw = filter_preprw,
+ .o_commitrw = filter_commitrw,
+ .o_do_cow = filter_do_cow,
+ .o_write_extents = filter_write_extents,
+ .o_destroy_export = filter_destroy_export,
+ .o_llog_init = filter_llog_init,
+ .o_llog_finish = filter_llog_finish,
+ .o_llog_connect = filter_llog_connect,
+ .o_iocontrol = filter_iocontrol,
};
static struct obd_ops filter_sanobd_ops = {
- o_owner: THIS_MODULE,
- o_attach: filter_attach,
- o_detach: filter_detach,
- o_get_info: filter_get_info,
- o_set_info: filter_set_info,
- o_setup: filter_san_setup,
- o_precleanup: filter_precleanup,
- o_cleanup: filter_cleanup,
- o_connect: filter_connect,
- o_disconnect: filter_disconnect,
- o_statfs: filter_statfs,
- o_getattr: filter_getattr,
- o_unpackmd: filter_unpackmd,
- o_create: filter_create,
- o_setattr: filter_setattr,
- o_destroy: filter_destroy,
- o_brw: filter_brw,
- o_punch: filter_truncate,
- o_sync: filter_sync,
- o_preprw: filter_preprw,
- o_commitrw: filter_commitrw,
- o_san_preprw: filter_san_preprw,
- o_destroy_export: filter_destroy_export,
- o_llog_init: filter_llog_init,
- o_llog_finish: filter_llog_finish,
- o_iocontrol: filter_iocontrol,
+ .o_owner = THIS_MODULE,
+ .o_attach = filter_attach,
+ .o_detach = filter_detach,
+ .o_get_info = filter_get_info,
+ .o_setup = filter_san_setup,
+ .o_precleanup = filter_precleanup,
+ .o_cleanup = filter_cleanup,
+ .o_connect = filter_connect,
+ .o_connect_post = filter_connect_post,
+ .o_disconnect = filter_disconnect,
+ .o_statfs = filter_statfs,
+ .o_getattr = filter_getattr,
+ .o_unpackmd = filter_unpackmd,
+ .o_create = filter_create,
+ .o_setattr = filter_setattr,
+ .o_destroy = filter_destroy,
+ .o_brw = filter_brw,
+ .o_punch = filter_truncate,
+ .o_sync = filter_sync,
+ .o_preprw = filter_preprw,
+ .o_commitrw = filter_commitrw,
+ .o_do_cow = filter_do_cow,
+ .o_write_extents = filter_write_extents,
+ .o_san_preprw = filter_san_preprw,
+ .o_destroy_export = filter_destroy_export,
+ .o_llog_init = filter_llog_init,
+ .o_llog_finish = filter_llog_finish,
+ .o_llog_connect = filter_llog_connect,
+ .o_iocontrol = filter_iocontrol,
};
static int __init obdfilter_init(void)
{
struct lprocfs_static_vars lvars;
- int rc;
+ int size, rc;
printk(KERN_INFO "Lustre: Filtering OBD driver; info@clusterfs.com\n");
lprocfs_init_vars(filter, &lvars);
- rc = class_register_type(&filter_obd_ops, lvars.module_vars,
+ size = OBDFILTER_CREATED_SCRATCHPAD_ENTRIES *
+ sizeof(*obdfilter_created_scratchpad);
+
+ OBD_ALLOC(obdfilter_created_scratchpad, size);
+ if (obdfilter_created_scratchpad == NULL) {
+ CERROR ("Can't allocate scratchpad\n");
+ return -ENOMEM;
+ }
+
+ rc = class_register_type(&filter_obd_ops, NULL, lvars.module_vars,
OBD_FILTER_DEVICENAME);
- if (rc)
+ if (rc) {
+ OBD_FREE(obdfilter_created_scratchpad, size);
return rc;
+ }
- rc = class_register_type(&filter_sanobd_ops, lvars.module_vars,
+ rc = class_register_type(&filter_sanobd_ops, NULL, lvars.module_vars,
OBD_FILTER_SAN_DEVICENAME);
- if (rc)
+ if (rc) {
class_unregister_type(OBD_FILTER_DEVICENAME);
+ OBD_FREE(obdfilter_created_scratchpad, size);
+ }
return rc;
}
{
class_unregister_type(OBD_FILTER_SAN_DEVICENAME);
class_unregister_type(OBD_FILTER_DEVICENAME);
+ OBD_FREE(obdfilter_created_scratchpad,
+ OBDFILTER_CREATED_SCRATCHPAD_ENTRIES *
+ sizeof(*obdfilter_created_scratchpad));
}
MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
git://git.whamcloud.com / fs / lustre-release.git / blobdiff