* threaded operation on the OST.
*/
-#define EXPORT_SYMTAB
#define DEBUG_SUBSYSTEM S_FILTER
#include <linux/config.h>
#include <linux/module.h>
-#include <linux/pagemap.h> // XXX kill me soon
#include <linux/fs.h>
#include <linux/dcache.h>
-#include <linux/obd_class.h>
-#include <linux/lustre_dlm.h>
-#include <linux/obd_filter.h>
#include <linux/init.h>
-#include <linux/random.h>
-#include <linux/lustre_fsfilt.h>
-#include <linux/lprocfs_status.h>
#include <linux/version.h>
#if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
-#include <linux/mount.h>
-#endif
-
-enum {
- LPROC_FILTER_READS = 0,
- LPROC_FILTER_READ_BYTES = 1,
- LPROC_FILTER_WRITES = 2,
- LPROC_FILTER_WRITE_BYTES = 3,
- LPROC_FILTER_LAST = LPROC_FILTER_WRITE_BYTES +1
-};
-
-/* should be generic per-obd stats... */
-struct xprocfs_io_stat {
- __u64 st_read_bytes;
- __u64 st_read_reqs;
- __u64 st_write_bytes;
- __u64 st_write_reqs;
- __u64 st_getattr_reqs;
- __u64 st_setattr_reqs;
- __u64 st_create_reqs;
- __u64 st_destroy_reqs;
- __u64 st_statfs_reqs;
- __u64 st_syncfs_reqs;
- __u64 st_open_reqs;
- __u64 st_close_reqs;
- __u64 st_punch_reqs;
-};
-
-static struct xprocfs_io_stat xprocfs_iostats[NR_CPUS];
-static struct proc_dir_entry *xprocfs_dir;
-
-#define XPROCFS_BUMP_MYCPU_IOSTAT(field, count) \
-do { \
- xprocfs_iostats[smp_processor_id()].field += (count); \
-} while (0)
-
-#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
-#define DECLARE_XPROCFS_SUM_STAT(field) \
-static long long \
-xprocfs_sum_##field (void) \
-{ \
- long long stat = 0; \
- int i; \
- \
- for (i = 0; i < smp_num_cpus; i++) \
- stat += xprocfs_iostats[i].field; \
- return (stat); \
-}
-
-DECLARE_XPROCFS_SUM_STAT (st_read_bytes)
-DECLARE_XPROCFS_SUM_STAT (st_read_reqs)
-DECLARE_XPROCFS_SUM_STAT (st_write_bytes)
-DECLARE_XPROCFS_SUM_STAT (st_write_reqs)
-DECLARE_XPROCFS_SUM_STAT (st_getattr_reqs)
-DECLARE_XPROCFS_SUM_STAT (st_setattr_reqs)
-DECLARE_XPROCFS_SUM_STAT (st_create_reqs)
-DECLARE_XPROCFS_SUM_STAT (st_destroy_reqs)
-DECLARE_XPROCFS_SUM_STAT (st_statfs_reqs)
-DECLARE_XPROCFS_SUM_STAT (st_syncfs_reqs)
-DECLARE_XPROCFS_SUM_STAT (st_open_reqs)
-DECLARE_XPROCFS_SUM_STAT (st_close_reqs)
-DECLARE_XPROCFS_SUM_STAT (st_punch_reqs)
-#endif
-
-static int
-xprocfs_rd_stat (char *page, char **start, off_t off, int count,
- int *eof, void *data)
-{
- long long (*fn)(void) = (long long(*)(void))data;
- int len;
-
- *eof = 1;
- if (off != 0)
- return (0);
-
- len = snprintf (page, count, "%Ld\n", fn());
- *start = page;
- return (len);
-}
-
-
-static void
-xprocfs_add_stat(char *name, long long (*fn)(void))
-{
- struct proc_dir_entry *entry;
-
- entry = create_proc_entry (name, S_IFREG|S_IRUGO, xprocfs_dir);
- if (entry == NULL) {
- CERROR ("Can't add procfs stat %s\n", name);
- return;
- }
-
- entry->data = fn;
- entry->read_proc = xprocfs_rd_stat;
- entry->write_proc = NULL;
-}
-
-static void
-xprocfs_init (char *name)
-{
- char dirname[64];
-
- snprintf (dirname, sizeof (dirname), "sys/%s", name);
-
- xprocfs_dir = proc_mkdir (dirname, NULL);
- if (xprocfs_dir == NULL) {
- CERROR ("Can't make procfs dir %s\n", dirname);
- return;
- }
-
-#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
- xprocfs_add_stat ("read_bytes", xprocfs_sum_st_read_bytes);
- xprocfs_add_stat ("read_reqs", xprocfs_sum_st_read_reqs);
- xprocfs_add_stat ("write_bytes", xprocfs_sum_st_write_bytes);
- xprocfs_add_stat ("write_reqs", xprocfs_sum_st_write_reqs);
- xprocfs_add_stat ("getattr_reqs", xprocfs_sum_st_getattr_reqs);
- xprocfs_add_stat ("setattr_reqs", xprocfs_sum_st_setattr_reqs);
- xprocfs_add_stat ("create_reqs", xprocfs_sum_st_create_reqs);
- xprocfs_add_stat ("destroy_reqs", xprocfs_sum_st_destroy_reqs);
- xprocfs_add_stat ("statfs_reqs", xprocfs_sum_st_statfs_reqs);
- xprocfs_add_stat ("syncfs_reqs", xprocfs_sum_st_syncfs_reqs);
- xprocfs_add_stat ("open_reqs", xprocfs_sum_st_open_reqs);
- xprocfs_add_stat ("close_reqs", xprocfs_sum_st_close_reqs);
- xprocfs_add_stat ("punch_reqs", xprocfs_sum_st_punch_reqs);
+# include <linux/mount.h>
+# include <linux/buffer_head.h>
#endif
-}
-
-void xprocfs_fini (void)
-{
- if (xprocfs_dir == NULL)
- return;
-
- remove_proc_entry ("read_bytes", xprocfs_dir);
- remove_proc_entry ("read_reqs", xprocfs_dir);
- remove_proc_entry ("write_bytes", xprocfs_dir);
- remove_proc_entry ("write_reqs", xprocfs_dir);
- remove_proc_entry ("getattr_reqs", xprocfs_dir);
- remove_proc_entry ("setattr_reqs", xprocfs_dir);
- remove_proc_entry ("create_reqs", xprocfs_dir);
- remove_proc_entry ("destroy_reqs", xprocfs_dir);
- remove_proc_entry ("statfs_reqs", xprocfs_dir);
- remove_proc_entry ("syncfs_reqs", xprocfs_dir);
- remove_proc_entry ("open_reqs", xprocfs_dir);
- remove_proc_entry ("close_reqs", xprocfs_dir);
- remove_proc_entry ("punch_reqs", xprocfs_dir);
-
- remove_proc_entry (xprocfs_dir->name, xprocfs_dir->parent);
- xprocfs_dir = NULL;
-}
-
-#define S_SHIFT 12
-static char *obd_type_by_mode[S_IFMT >> S_SHIFT] = {
- [0] NULL,
- [S_IFREG >> S_SHIFT] "R",
- [S_IFDIR >> S_SHIFT] "D",
- [S_IFCHR >> S_SHIFT] "C",
- [S_IFBLK >> S_SHIFT] "B",
- [S_IFIFO >> S_SHIFT] "F",
- [S_IFSOCK >> S_SHIFT] "S",
- [S_IFLNK >> S_SHIFT] "L"
-};
-
-static inline const char *obd_mode_to_type(int mode)
-{
- return obd_type_by_mode[(mode & S_IFMT) >> S_SHIFT];
-}
-static void filter_ffd_addref(void *ffdp)
-{
- struct filter_file_data *ffd = ffdp;
-
- atomic_inc(&ffd->ffd_refcount);
- CDEBUG(D_INFO, "GETting ffd %p : new refcount %d\n", ffd,
- atomic_read(&ffd->ffd_refcount));
-}
-
-static struct filter_file_data *filter_ffd_new(void)
-{
- struct filter_file_data *ffd;
-
- OBD_ALLOC(ffd, sizeof *ffd);
- if (ffd == NULL) {
- CERROR("out of memory\n");
- return NULL;
- }
-
- atomic_set(&ffd->ffd_refcount, 2);
+#include <linux/obd_class.h>
+#include <linux/obd_lov.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 <portals/list.h>
- INIT_LIST_HEAD(&ffd->ffd_handle.h_link);
- class_handle_hash(&ffd->ffd_handle, filter_ffd_addref);
+#include "filter_internal.h"
- return ffd;
-}
+static struct lvfs_callback_ops filter_lvfs_ops;
-static struct filter_file_data *filter_handle2ffd(struct lustre_handle *handle)
-{
- struct filter_file_data *ffd = NULL;
- ENTRY;
- LASSERT(handle != NULL);
- ffd = class_handle2object(handle->cookie);
- if (ffd != NULL)
- LASSERT(ffd->ffd_file->private_data == ffd);
- RETURN(ffd);
-}
+static int filter_destroy(struct obd_export *exp, struct obdo *oa,
+ struct lov_stripe_md *ea, struct obd_trans_info *);
-static void filter_ffd_put(struct filter_file_data *ffd)
+static void filter_commit_cb(struct obd_device *obd, __u64 transno,
+ void *cb_data, int error)
{
- CDEBUG(D_INFO, "PUTting ffd %p : new refcount %d\n", ffd,
- atomic_read(&ffd->ffd_refcount) - 1);
- LASSERT(atomic_read(&ffd->ffd_refcount) > 0 &&
- atomic_read(&ffd->ffd_refcount) < 0x5a5a);
- if (atomic_dec_and_test(&ffd->ffd_refcount)) {
- LASSERT(list_empty(&ffd->ffd_handle.h_link));
- OBD_FREE(ffd, sizeof *ffd);
- }
+ obd_transno_commit_cb(obd, transno, error);
}
-static void filter_ffd_destroy(struct filter_file_data *ffd)
-{
- class_handle_unhash(&ffd->ffd_handle);
- filter_ffd_put(ffd);
-}
-static void filter_commit_cb(struct obd_device *obd, __u64 transno, 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 *export, void *handle,
- struct obd_trans_info *oti, int rc)
+int filter_finish_transno(struct obd_export *exp, struct obd_trans_info *oti,
+ int rc)
{
- __u64 last_rcvd;
- struct obd_device *obd = export->exp_obd;
- struct filter_obd *filter = &obd->u.filter;
- struct filter_export_data *fed = &export->exp_filter_data;
+ struct filter_obd *filter = &exp->exp_obd->u.filter;
+ struct filter_export_data *fed = &exp->exp_filter_data;
struct filter_client_data *fcd = fed->fed_fcd;
+ __u64 last_rcvd;
loff_t off;
- ssize_t written;
+ int err, log_pri = D_HA;
/* Propagate error code. */
if (rc)
RETURN(rc);
- if (!obd->obd_replayable)
+ if (!exp->exp_obd->obd_replayable || oti == NULL)
RETURN(rc);
/* we don't allocate new transnos for replayed requests */
-#if 0
- /* perhaps if transno already set? or should level be in oti? */
- if (req->rq_level == LUSTRE_CONN_RECOVD)
- GOTO(out, rc = 0);
-#endif
-
- off = fed->fed_lr_off;
-
- spin_lock(&filter->fo_translock);
- last_rcvd = le64_to_cpu(filter->fo_fsd->fsd_last_rcvd);
- filter->fo_fsd->fsd_last_rcvd = cpu_to_le64(last_rcvd + 1);
- spin_unlock(&filter->fo_translock);
- if (oti)
+ if (oti->oti_transno == 0) {
+ spin_lock(&filter->fo_translock);
+ last_rcvd = le64_to_cpu(filter->fo_fsd->fsd_last_transno) + 1;
+ filter->fo_fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
+ spin_unlock(&filter->fo_translock);
oti->oti_transno = last_rcvd;
+ } else {
+ spin_lock(&filter->fo_translock);
+ last_rcvd = oti->oti_transno;
+ if (last_rcvd > le64_to_cpu(filter->fo_fsd->fsd_last_transno))
+ 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;
- /* get this from oti */
-#if 0
- if (oti)
- fcd->fcd_last_xid = cpu_to_le64(oti->oti_xid);
- else
-#else
+ /* could get xid from oti, if it's ever needed */
fcd->fcd_last_xid = 0;
-#endif
- fsfilt_set_last_rcvd(obd, last_rcvd, handle, filter_commit_cb);
- written = lustre_fwrite(filter->fo_rcvd_filp, (char *)fcd, sizeof(*fcd),
- &off);
- CDEBUG(D_INODE, "wrote trans #"LPD64" for client %s at #%d: written = "
- LPSZ"\n", last_rcvd, fcd->fcd_uuid, fed->fed_lr_idx, written);
-
- if (written == sizeof(*fcd))
- RETURN(0);
- CERROR("error writing to last_rcvd file: rc = %d\n", (int)written);
- if (written >= 0)
- RETURN(-EIO);
- RETURN(written);
-}
+ off = fed->fed_lr_off;
+ fsfilt_add_journal_cb(exp->exp_obd, 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) {
+ log_pri = D_ERROR;
+ if (rc == 0)
+ rc = err;
+ }
-/* write the pathname into the string */
-static char *filter_id(char *buf, struct filter_obd *filter, obd_id id,
- obd_mode mode)
-{
- if (!S_ISREG(mode) || filter->fo_subdir_count == 0)
- sprintf(buf, "O/%s/"LPU64, obd_mode_to_type(mode), id);
- else
- sprintf(buf, "O/%s/d%d/"LPU64, obd_mode_to_type(mode),
- (int)id & (filter->fo_subdir_count - 1), id);
+ CDEBUG(log_pri, "wrote trans "LPU64" for client %s at #%d: err = %d\n",
+ last_rcvd, fcd->fcd_uuid, fed->fed_lr_idx, err);
- return buf;
+ RETURN(rc);
}
-static inline void f_dput(struct dentry *dentry)
+void f_dput(struct dentry *dentry)
{
/* Can't go inside filter_ddelete because it can block */
CDEBUG(D_INODE, "putting %s: %p, count = %d\n",
dput(dentry);
}
-/* Not racy w.r.t. others, because we are the only user of this dentry */
-static void filter_drelease(struct dentry *dentry)
-{
- if (dentry->d_fsdata)
- OBD_FREE(dentry->d_fsdata, sizeof(struct filter_dentry_data));
-}
-
-struct dentry_operations filter_dops = {
- .d_release = filter_drelease,
-};
-
-#define LAST_RCVD "last_rcvd"
-#define INIT_OBJID 2
-
-/* This limit is arbitrary, but for now we fit it in 1 page (32k clients) */
-#define FILTER_LR_MAX_CLIENTS (PAGE_SIZE * 8)
-#define FILTER_LR_MAX_CLIENT_WORDS (FILTER_LR_MAX_CLIENTS/sizeof(unsigned long))
-
/* Add client data to the FILTER. We use a bitmap to locate a free space
* in the last_rcvd file if cl_idx is -1 (i.e. a new client).
* Otherwise, we have just read the data from the last_rcvd file and
- * we know its offset.
- */
-int filter_client_add(struct obd_device *obd, struct filter_obd *filter,
- struct filter_export_data *fed, int cl_idx)
+ * we know its offset. */
+static int filter_client_add(struct obd_device *obd, struct filter_obd *filter,
+ struct filter_export_data *fed, int cl_idx)
{
unsigned long *bitmap = filter->fo_last_rcvd_slots;
int new_client = (cl_idx == -1);
+ ENTRY;
LASSERT(bitmap != NULL);
- /* XXX if mcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
- if (!strcmp(fed->fed_fcd->fcd_uuid, "OBD_CLASS_UUID"))
+ /* 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))
RETURN(0);
/* the bitmap operations can handle cl_idx > sizeof(long) * 8, so
repeat:
if (cl_idx >= FILTER_LR_MAX_CLIENTS) {
CERROR("no client slots - fix FILTER_LR_MAX_CLIENTS\n");
- return -ENOMEM;
+ RETURN(-ENOMEM);
}
if (test_and_set_bit(cl_idx, bitmap)) {
CERROR("FILTER client %d: found bit is set in bitmap\n",
if (new_client) {
struct obd_run_ctxt saved;
loff_t off = fed->fed_lr_off;
- ssize_t written;
+ 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, &filter->fo_ctxt, NULL);
- /* Transaction eeded to fix for bug 1403 */
+ push_ctxt(&saved, &obd->obd_ctxt, NULL);
+ /* Transaction needed to fix bug 1403 */
handle = fsfilt_start(obd,
filter->fo_rcvd_filp->f_dentry->d_inode,
- FSFILT_OP_SETATTR);
+ FSFILT_OP_SETATTR, NULL);
if (IS_ERR(handle)) {
- written = PTR_ERR(handle);
- CERROR("unable to start transaction: rc %d\n",
- (int)written);
+ err = PTR_ERR(handle);
+ CERROR("unable to start transaction: rc %d\n", err);
} else {
- written = lustre_fwrite(filter->fo_rcvd_filp,
- (char *)fed->fed_fcd,
- sizeof(*fed->fed_fcd), &off);
+ err = fsfilt_write_record(obd, filter->fo_rcvd_filp,
+ fed->fed_fcd,
+ sizeof(*fed->fed_fcd),
+ &off, 1);
fsfilt_commit(obd,
filter->fo_rcvd_filp->f_dentry->d_inode,
- handle, 0);
+ handle, 1);
}
- pop_ctxt(&saved, &filter->fo_ctxt, NULL);
+ pop_ctxt(&saved, &obd->obd_ctxt, NULL);
- if (written != sizeof(*fed->fed_fcd)) {
- if (written < 0)
- RETURN(written);
- RETURN(-EIO);
+ if (err) {
+ CERROR("error writing %s client idx %u: rc %d\n",
+ LAST_RCVD, fed->fed_lr_idx, err);
+ RETURN(err);
}
}
- return 0;
+ RETURN(0);
}
-int filter_client_free(struct obd_export *exp, int failover)
+static int filter_client_free(struct obd_export *exp, int flags)
{
struct filter_export_data *fed = &exp->exp_filter_data;
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;
- int written;
+ int rc;
loff_t off;
+ ENTRY;
- if (!fed->fed_fcd)
+ if (fed->fed_fcd == NULL)
RETURN(0);
- if (failover != 0) {
- OBD_FREE(fed->fed_fcd, sizeof(*fed->fed_fcd));
- RETURN(0);
- }
+ if (flags & OBD_OPT_FAILOVER)
+ 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)
+ 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, &filter->fo_ctxt, NULL);
- written = lustre_fwrite(filter->fo_rcvd_filp, (const char *)&zero_fcd,
- sizeof(zero_fcd), &off);
-
- /* XXX: this write gets lost sometimes, unless this sync is here. */
- if (written > 0)
- file_fsync(filter->fo_rcvd_filp,
- filter->fo_rcvd_filp->f_dentry, 1);
- pop_ctxt(&saved, &filter->fo_ctxt, NULL);
-
- if (written != sizeof(zero_fcd)) {
- CERROR("error zeroing out client %s idx %u (%llu) in %s: %d\n",
- fed->fed_fcd->fcd_uuid, fed->fed_lr_idx, fed->fed_lr_off,
- LAST_RCVD, written);
- } else {
- CDEBUG(D_INFO,
- "zeroed disconnecting client %s at idx %u (%llu)\n",
- fed->fed_fcd->fcd_uuid, fed->fed_lr_idx,fed->fed_lr_off);
+ push_ctxt(&saved, &obd->obd_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);
+
+ 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));
- return 0;
+ RETURN(0);
}
static int filter_free_server_data(struct filter_obd *filter)
return 0;
}
-
/* assumes caller is already in kernel ctxt */
-static int filter_update_server_data(struct file *filp,
- struct filter_server_data *fsd)
+int filter_update_server_data(struct obd_device *obd, struct file *filp,
+ struct filter_server_data *fsd, int force_sync)
{
loff_t off = 0;
int rc;
+ ENTRY;
CDEBUG(D_INODE, "server uuid : %s\n", fsd->fsd_uuid);
- CDEBUG(D_INODE, "server last_objid: "LPU64"\n",
- le64_to_cpu(fsd->fsd_last_objid));
CDEBUG(D_INODE, "server last_rcvd : "LPU64"\n",
- le64_to_cpu(fsd->fsd_last_rcvd));
+ le64_to_cpu(fsd->fsd_last_transno));
CDEBUG(D_INODE, "server last_mount: "LPU64"\n",
le64_to_cpu(fsd->fsd_mount_count));
- rc = lustre_fwrite(filp, (char *)fsd, sizeof(*fsd), &off);
- if (rc != sizeof(*fsd)) {
- CDEBUG(D_INODE, "error writing filter_server_data: rc = %d\n",
- rc);
- RETURN(-EIO);
- }
- RETURN(0);
+ rc = fsfilt_write_record(obd, filp, fsd, sizeof(*fsd), &off,force_sync);
+ if (rc)
+ CERROR("error writing filter_server_data: rc = %d\n", rc);
+
+ RETURN(rc);
+}
+
+int filter_update_last_objid(struct obd_device *obd, obd_gr group,
+ int force_sync)
+{
+ struct filter_obd *filter = &obd->u.filter;
+ __u64 tmp;
+ loff_t off = 0;
+ int rc;
+ ENTRY;
+
+ CDEBUG(D_INODE, "server last_objid for group "LPU64": "LPU64"\n",
+ group, filter->fo_last_objids[group]);
+
+ tmp = cpu_to_le64(filter->fo_last_objids[group]);
+ rc = fsfilt_write_record(obd, filter->fo_last_objid_files[group],
+ &tmp, sizeof(tmp), &off, force_sync);
+ if (rc)
+ CERROR("error writing group "LPU64" last objid: rc = %d\n",
+ group, rc);
+ RETURN(rc);
}
/* assumes caller has already in kernel ctxt */
-static int filter_init_server_data(struct obd_device *obd, struct file * filp,
- __u64 init_lastobjid)
+static int filter_init_server_data(struct obd_device *obd, struct file * filp)
{
struct filter_obd *filter = &obd->u.filter;
struct filter_server_data *fsd;
struct filter_client_data *fcd = NULL;
struct inode *inode = filp->f_dentry->d_inode;
unsigned long last_rcvd_size = inode->i_size;
- __u64 mount_count = 0;
+ __u64 mount_count;
int cl_idx;
loff_t off = 0;
int rc;
}
if (last_rcvd_size == 0) {
- CERROR("%s: initializing new last_rcvd\n", obd->obd_name);
+ CWARN("%s: initializing new %s\n", obd->obd_name, LAST_RCVD);
memcpy(fsd->fsd_uuid, obd->obd_uuid.uuid,sizeof(fsd->fsd_uuid));
- fsd->fsd_last_objid = cpu_to_le64(init_lastobjid);
- fsd->fsd_last_rcvd = 0;
+ fsd->fsd_last_transno = 0;
mount_count = fsd->fsd_mount_count = 0;
fsd->fsd_server_size = cpu_to_le32(FILTER_LR_SERVER_SIZE);
fsd->fsd_client_start = cpu_to_le32(FILTER_LR_CLIENT_START);
fsd->fsd_subdir_count = cpu_to_le16(FILTER_SUBDIR_COUNT);
filter->fo_subdir_count = FILTER_SUBDIR_COUNT;
} else {
- ssize_t retval = lustre_fread(filp, (char *)fsd, sizeof(*fsd),
- &off);
- if (retval != sizeof(*fsd)) {
- CDEBUG(D_INODE,"OBD filter: error reading %s\n",
- LAST_RCVD);
- GOTO(err_fsd, rc = -EIO);
+ rc = fsfilt_read_record(obd, filp, fsd, sizeof(*fsd), &off);
+ if (rc) {
+ CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
+ LAST_RCVD, rc);
+ GOTO(err_fsd, rc);
+ }
+ if (strcmp(fsd->fsd_uuid, obd->obd_uuid.uuid) != 0) {
+ CERROR("OBD UUID %s does not match last_rcvd UUID %s\n",
+ obd->obd_uuid.uuid, fsd->fsd_uuid);
+ GOTO(err_fsd, rc = -EINVAL);
}
mount_count = le64_to_cpu(fsd->fsd_mount_count);
filter->fo_subdir_count = le16_to_cpu(fsd->fsd_subdir_count);
}
- if (fsd->fsd_feature_incompat) {
+ if (fsd->fsd_feature_incompat & ~cpu_to_le32(FILTER_INCOMPAT_SUPP)) {
CERROR("unsupported feature %x\n",
- le32_to_cpu(fsd->fsd_feature_incompat));
+ le32_to_cpu(fsd->fsd_feature_incompat) &
+ ~FILTER_INCOMPAT_SUPP);
GOTO(err_fsd, rc = -EINVAL);
}
- if (fsd->fsd_feature_rocompat) {
+ if (fsd->fsd_feature_rocompat & ~cpu_to_le32(FILTER_ROCOMPAT_SUPP)) {
CERROR("read-only feature %x\n",
- le32_to_cpu(fsd->fsd_feature_rocompat));
+ le32_to_cpu(fsd->fsd_feature_rocompat) &
+ ~FILTER_ROCOMPAT_SUPP);
/* Do something like remount filesystem read-only */
GOTO(err_fsd, rc = -EINVAL);
}
- CDEBUG(D_INODE, "%s: server last_objid: "LPU64"\n",
- obd->obd_name, le64_to_cpu(fsd->fsd_last_objid));
CDEBUG(D_INODE, "%s: server last_rcvd : "LPU64"\n",
- obd->obd_name, le64_to_cpu(fsd->fsd_last_rcvd));
- CDEBUG(D_INODE, "%s: server last_mount: "LPU64"\n",
- obd->obd_name, mount_count);
+ obd->obd_name, le64_to_cpu(fsd->fsd_last_transno));
+ 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",
obd->obd_name, le32_to_cpu(fsd->fsd_client_size));
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);
- /*
- * When we do a clean FILTER shutdown, we save the last_rcvd into
- * the header. If we find clients with higher last_rcvd values
- * then those clients may need recovery done.
- */
if (!obd->obd_replayable) {
- CERROR("%s: recovery support OFF\n", obd->obd_name);
+ CWARN("%s: recovery support OFF\n", obd->obd_name);
GOTO(out, rc = 0);
}
- for (cl_idx = 0; off < last_rcvd_size; cl_idx++) {
+ 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));
if (!fcd)
- GOTO(err_fsd, rc = -ENOMEM);
+ GOTO(err_client, rc = -ENOMEM);
}
- /* Don't assume off is incremented properly, in case
- * sizeof(fsd) isn't the same as fsd->fsd_client_size.
- */
+ /* Don't assume off is incremented properly by
+ * fsfilt_read_record(), in case sizeof(*fcd)
+ * isn't the same as fsd->fsd_client_size. */
off = le32_to_cpu(fsd->fsd_client_start) +
cl_idx * le16_to_cpu(fsd->fsd_client_size);
- rc = lustre_fread(filp, (char *)fcd, sizeof(*fcd), &off);
- if (rc != sizeof(*fcd)) {
- CERROR("error reading FILTER %s offset %d: rc = %d\n",
- LAST_RCVD, cl_idx, rc);
- if (rc > 0) /* XXX fatal error or just abort reading? */
- rc = -EIO;
- break;
+ rc = fsfilt_read_record(obd, filp, fcd, sizeof(*fcd), &off);
+ if (rc) {
+ CERROR("error reading FILT %s idx %d off %llu: rc %d\n",
+ LAST_RCVD, cl_idx, off, rc);
+ break; /* read error shouldn't cause startup to fail */
}
if (fcd->fcd_uuid[0] == '\0') {
/* 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_rcvd),
- le64_to_cpu(fcd->fcd_mount_count), mount_count);
- if (exp == NULL) {
- /* XXX this rc is ignored */
- rc = -ENOMEM;
- break;
- }
- 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 */
- INIT_LIST_HEAD(&fed->fed_open_head);
- 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"\n", fcd->fcd_uuid, cl_idx,
+ last_rcvd, le64_to_cpu(fsd->fsd_last_transno));
+ 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);
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_rcvd))
- filter->fo_fsd->fsd_last_rcvd = cpu_to_le64(last_rcvd);
-
- obd->obd_last_committed =
- le64_to_cpu(filter->fo_fsd->fsd_last_rcvd);
- 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_rcvd));
- obd->obd_next_recovery_transno =
- obd->obd_last_committed + 1;
- obd->obd_recovering = 1;
- }
+ 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(fsd->fsd_last_transno);
+ if (obd->obd_recoverable_clients) {
+ CWARN("RECOVERY: %d recoverable clients, last_rcvd "
+ LPU64"\n", 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;
}
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_recvd is current */
- rc = filter_update_server_data(filp, filter->fo_fsd);
+ /* save it, so mount count and last_transno is current */
+ rc = filter_update_server_data(obd, filp, filter->fo_fsd, 1);
RETURN(rc);
+err_client:
+ class_disconnect_exports(obd, 0);
err_fsd:
filter_free_server_data(filter);
RETURN(rc);
}
-/* setup the object store with correct subdirectories */
-static int filter_prep(struct obd_device *obd)
+static int filter_cleanup_groups(struct obd_device *obd)
{
- struct obd_run_ctxt saved;
struct filter_obd *filter = &obd->u.filter;
- struct dentry *dentry, *O_dentry;
- struct file *file;
- struct inode *inode;
int i;
- int rc = 0;
- int mode = 0;
+ ENTRY;
+
+ 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) {
+ f_dput(dentry);
+ filter->fo_dentry_O_groups[i] = NULL;
+ }
+ if (filp != NULL) {
+ filp_close(filp, 0);
+ filter->fo_last_objid_files[i] = NULL;
+ }
+ }
+ }
+ if (filter->fo_dentry_O_sub != NULL && filter->fo_subdir_count) {
+ for (i = 0; i < filter->fo_subdir_count; i++) {
+ struct dentry *dentry = filter->fo_dentry_O_sub[i];
+ if (dentry != NULL) {
+ 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)
+ OBD_FREE(filter->fo_dentry_O_groups,
+ FILTER_GROUPS * sizeof(struct dentry *));
+ if (filter->fo_last_objids != NULL)
+ OBD_FREE(filter->fo_last_objids,
+ FILTER_GROUPS * sizeof(__u64));
+ if (filter->fo_last_objid_files != NULL)
+ OBD_FREE(filter->fo_last_objid_files,
+ FILTER_GROUPS * sizeof(struct file *));
+ RETURN(0);
+}
+
+/* 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;
+ ENTRY;
- push_ctxt(&saved, &filter->fo_ctxt, NULL);
- dentry = simple_mkdir(current->fs->pwd, "O", 0700);
- CDEBUG(D_INODE, "got/created O: %p\n", dentry);
- if (IS_ERR(dentry)) {
- rc = PTR_ERR(dentry);
+ 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);
CERROR("cannot open/create O: rc = %d\n", rc);
- GOTO(out, rc);
+ GOTO(cleanup, rc);
}
- filter->fo_dentry_O = dentry;
+ filter->fo_dentry_O = O_dentry;
+ cleanup_phase = 1; /* O_dentry */
- /*
- * Create directories and/or get dentries for each object type.
- * This saves us from having to do multiple lookups for each one.
- */
- O_dentry = filter->fo_dentry_O;
- for (mode = 0; mode < (S_IFMT >> S_SHIFT); mode++) {
- char *name = obd_type_by_mode[mode];
+ /* Lookup "R" to tell if we're on an old OST FS and need to convert
+ * from O/R/<dir>/<objid> to O/0/<dir>/<objid>. This can be removed
+ * some time post 1.0 when all old-style OSTs have converted along
+ * with the init_objid hack. */
+ dentry = ll_lookup_one_len("R", O_dentry, 1);
+ if (IS_ERR(dentry))
+ GOTO(cleanup, rc = PTR_ERR(dentry));
+ if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
+ struct dentry *O0_dentry = lookup_one_len("0", O_dentry, 1);
+ ENTRY;
+
+ CWARN("converting OST to new object layout\n");
+ if (IS_ERR(O0_dentry)) {
+ rc = PTR_ERR(O0_dentry);
+ CERROR("error looking up O/0: rc %d\n", rc);
+ GOTO(cleanup_R, rc);
+ }
- if (!name) {
- filter->fo_dentry_O_mode[mode] = NULL;
- continue;
+ if (O0_dentry->d_inode) {
+ CERROR("Both O/R and O/0 exist. Fix manually.\n");
+ GOTO(cleanup_O0, rc = -EEXIST);
}
- 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(err_O_mode, rc);
+
+ down(&O_dentry->d_inode->i_sem);
+ rc = vfs_rename(O_dentry->d_inode, dentry,
+ O_dentry->d_inode, O0_dentry);
+ up(&O_dentry->d_inode->i_sem);
+
+ if (rc) {
+ CERROR("error renaming O/R to O/0: rc %d\n", rc);
+ GOTO(cleanup_O0, rc);
}
- filter->fo_dentry_O_mode[mode] = dentry;
+ filter->fo_fsd->fsd_feature_incompat |=
+ cpu_to_le32(FILTER_INCOMPAT_GROUPS);
+ rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
+ filter->fo_fsd, 1);
+ GOTO(cleanup_O0, rc);
+
+ cleanup_O0:
+ dput(O0_dentry);
+ cleanup_R:
+ dput(dentry);
+ if (rc)
+ GOTO(cleanup, rc);
+ } else {
+ dput(dentry);
}
- file = filp_open(LAST_RCVD, O_RDWR | O_CREAT, 0700);
- if (!file || IS_ERR(file)) {
- rc = PTR_ERR(file);
- CERROR("OBD filter: cannot open/create %s: rc = %d\n",
- LAST_RCVD, rc);
- GOTO(err_O_mode, rc);
- }
+ OBD_ALLOC(filter->fo_last_objids, FILTER_GROUPS * sizeof(__u64));
+ if (filter->fo_last_objids == NULL)
+ GOTO(cleanup, rc = -ENOMEM);
+ cleanup_phase = 2; /* groups */
- if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
- CERROR("%s is not a regular file!: mode = %o\n", LAST_RCVD,
- file->f_dentry->d_inode->i_mode);
- GOTO(err_filp, rc = -ENOENT);
- }
+ 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);
- rc = fsfilt_journal_data(obd, file);
- if (rc) {
- CERROR("cannot journal data on %s: rc = %d\n", LAST_RCVD, rc);
- GOTO(err_filp, rc);
- }
- /* steal operations */
- inode = file->f_dentry->d_inode;
- filter->fo_fop = file->f_op;
- filter->fo_iop = inode->i_op;
- filter->fo_aops = inode->i_mapping->a_ops;
+ for (i = 0; i < FILTER_GROUPS; i++) {
+ char name[25];
+ loff_t off = 0;
- rc = filter_init_server_data(obd, file, INIT_OBJID);
- if (rc) {
- CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc);
- GOTO(err_client, rc);
+ sprintf(name, "%d", i);
+ dentry = simple_mkdir(O_dentry, name, 0700, 1);
+ CDEBUG(D_INODE, "got/created O/%s: %p\n", name, dentry);
+ if (IS_ERR(dentry)) {
+ rc = PTR_ERR(dentry);
+ CERROR("cannot lookup/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]);
}
- filter->fo_rcvd_filp = file;
if (filter->fo_subdir_count) {
- O_dentry = filter->fo_dentry_O_mode[S_IFREG >> S_SHIFT];
+ 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)
- GOTO(err_client, rc = -ENOMEM);
+ 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/R/%s: %p\n", dir,dentry);
+ dentry = simple_mkdir(O_dentry, dir, 0700, 1);
+ 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/R/%s: rc = %d\n",dir,rc);
- GOTO(err_O_sub, rc);
+ CERROR("can't lookup/create O/0/%s: rc = %d\n",
+ dir, rc);
+ GOTO(cleanup, rc);
}
filter->fo_dentry_O_sub[i] = dentry;
}
}
- rc = 0;
- out:
- pop_ctxt(&saved, &filter->fo_ctxt, NULL);
-
- return(rc);
+ RETURN(0);
-err_O_sub:
- while (i-- > 0) {
- struct dentry *dentry = filter->fo_dentry_O_sub[i];
- if (dentry) {
- f_dput(dentry);
- filter->fo_dentry_O_sub[i] = NULL;
- }
+ cleanup:
+ switch (cleanup_phase) {
+ case 2:
+ filter_cleanup_groups(obd);
+ case 1:
+ f_dput(filter->fo_dentry_O);
+ filter->fo_dentry_O = NULL;
+ default:
+ break;
}
- OBD_FREE(filter->fo_dentry_O_sub,
- filter->fo_subdir_count * sizeof(dentry));
-err_client:
- class_disconnect_exports(obd, 0);
-err_filp:
+ return rc;
+}
+
+/* setup the object store with correct subdirectories */
+static int filter_prep(struct obd_device *obd)
+{
+ struct obd_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);
+ file = filp_open(LAST_RCVD, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
+ if (!file || IS_ERR(file)) {
+ rc = PTR_ERR(file);
+ CERROR("OBD filter: cannot open/create %s: rc = %d\n",
+ LAST_RCVD, rc);
+ GOTO(out, rc);
+ }
+
+ if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
+ CERROR("%s is not a regular file!: mode = %o\n", LAST_RCVD,
+ file->f_dentry->d_inode->i_mode);
+ GOTO(err_filp, rc = -ENOENT);
+ }
+
+ /* steal operations */
+ inode = file->f_dentry->d_inode;
+ filter->fo_fop = file->f_op;
+ filter->fo_iop = inode->i_op;
+ filter->fo_aops = inode->i_mapping->a_ops;
+
+ rc = filter_init_server_data(obd, file);
+ if (rc) {
+ CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc);
+ GOTO(err_filp, rc);
+ }
+ filter->fo_rcvd_filp = file;
+
+ rc = filter_prep_groups(obd);
+ if (rc)
+ GOTO(err_server_data, rc);
+
+ out:
+ pop_ctxt(&saved, &obd->obd_ctxt, NULL);
+
+ return(rc);
+
+ err_server_data:
+ //class_disconnect_exports(obd, 0);
+ filter_free_server_data(filter);
+ err_filp:
if (filp_close(file, 0))
CERROR("can't close %s after error\n", LAST_RCVD);
filter->fo_rcvd_filp = NULL;
-err_O_mode:
- while (mode-- > 0) {
- struct dentry *dentry = filter->fo_dentry_O_mode[mode];
- if (dentry) {
- f_dput(dentry);
- filter->fo_dentry_O_mode[mode] = NULL;
- }
- }
- f_dput(filter->fo_dentry_O);
- filter->fo_dentry_O = NULL;
goto out;
}
{
struct obd_run_ctxt saved;
struct filter_obd *filter = &obd->u.filter;
- long rc;
- int mode;
+ int rc, i;
/* XXX: filter_update_lastobjid used to call fsync_dev. It might be
* best to start a transaction with h_sync, because we removed this
* from lastobjid */
- push_ctxt(&saved, &filter->fo_ctxt, NULL);
- rc = filter_update_server_data(filter->fo_rcvd_filp, filter->fo_fsd);
+ push_ctxt(&saved, &obd->obd_ctxt, NULL);
+ rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
+ filter->fo_fsd, 0);
if (rc)
- CERROR("OBD filter: error writing lastobjid: rc = %ld\n", rc);
-
+ CERROR("error writing server data: rc = %d\n", rc);
- if (filter->fo_rcvd_filp) {
- rc = file_fsync(filter->fo_rcvd_filp,
- filter->fo_rcvd_filp->f_dentry, 1);
- filp_close(filter->fo_rcvd_filp, 0);
- filter->fo_rcvd_filp = NULL;
+ for (i = 0; i < FILTER_GROUPS; i++) {
+ rc = filter_update_last_objid(obd, i, (i == FILTER_GROUPS - 1));
if (rc)
- CERROR("last_rcvd file won't closed rc = %ld\n", rc);
+ CERROR("error writing group %d lastobjid: rc = %d\n",
+ i, rc);
}
- if (filter->fo_subdir_count) {
- int i;
- 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));
- }
- for (mode = 0; mode < (S_IFMT >> S_SHIFT); mode++) {
- struct dentry *dentry = filter->fo_dentry_O_mode[mode];
- if (dentry) {
- f_dput(dentry);
- filter->fo_dentry_O_mode[mode] = NULL;
- }
- }
+ 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, &filter->fo_ctxt, NULL);
+ pop_ctxt(&saved, &obd->obd_ctxt, NULL);
}
-
-static __u64 filter_next_id(struct obd_device *obd)
+static void filter_set_last_id(struct filter_obd *filter, struct obdo *oa,
+ obd_id id)
{
- obd_id id;
- LASSERT(obd->u.filter.fo_fsd != NULL);
+ obd_gr group = 0;
+ LASSERT(filter->fo_fsd != NULL);
- spin_lock(&obd->u.filter.fo_objidlock);
- id = le64_to_cpu(obd->u.filter.fo_fsd->fsd_last_objid);
- obd->u.filter.fo_fsd->fsd_last_objid = cpu_to_le64(id + 1);
- spin_unlock(&obd->u.filter.fo_objidlock);
+ if (oa != NULL) {
+ LASSERT(oa->o_gr <= FILTER_GROUPS);
+ group = oa->o_gr;
+ }
- return id;
+ spin_lock(&filter->fo_objidlock);
+ filter->fo_last_objids[group] = id;
+ spin_unlock(&filter->fo_objidlock);
}
-/* how to get files, dentries, inodes from object id's */
-/* parent i_sem is already held if needed for exclusivity */
-static struct dentry *filter_fid2dentry(struct obd_device *obd,
- struct dentry *dparent,
- __u64 id, int lockit)
+__u64 filter_last_id(struct filter_obd *filter, struct obdo *oa)
{
- struct super_block *sb = obd->u.filter.fo_sb;
- struct dentry *dchild;
- char name[32];
- int len;
- ENTRY;
-
- if (!sb || !sb->s_dev) {
- CERROR("fatal: device not initialized.\n");
- RETURN(ERR_PTR(-ENXIO));
- }
-
- if (id == 0) {
- CERROR("fatal: invalid object id 0\n");
- LBUG();
- RETURN(ERR_PTR(-ESTALE));
- }
+ obd_id id;
+ obd_gr group = 0;
+ LASSERT(filter->fo_fsd != NULL);
- len = sprintf(name, LPU64, id);
- CDEBUG(D_INODE, "looking up object O/%*s/%s\n",
- dparent->d_name.len, dparent->d_name.name, name);
- if (lockit)
- down(&dparent->d_inode->i_sem);
- dchild = lookup_one_len(name, dparent, len);
- if (lockit)
- up(&dparent->d_inode->i_sem);
- if (IS_ERR(dchild)) {
- CERROR("child lookup error %ld\n", PTR_ERR(dchild));
- RETURN(dchild);
+ if (oa != NULL) {
+ LASSERT(oa->o_gr <= FILTER_GROUPS);
+ group = oa->o_gr;
}
- CDEBUG(D_INODE, "got child obj O/%*s/%s: %p, count = %d\n",
- dparent->d_name.len, dparent->d_name.name, name, dchild,
- atomic_read(&dchild->d_count));
-
- LASSERT(atomic_read(&dchild->d_count) > 0);
+ /* FIXME: object groups */
+ spin_lock(&filter->fo_objidlock);
+ id = filter->fo_last_objids[group];
+ spin_unlock(&filter->fo_objidlock);
- RETURN(dchild);
+ return id;
}
/* direct cut-n-paste of mds_blocking_ast() */
-int filter_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
- void *data, int flag)
+static int filter_blocking_ast(struct ldlm_lock *lock,
+ struct ldlm_lock_desc *desc,
+ void *data, int flag)
{
int do_ast;
ENTRY;
/* XXX layering violation! -phil */
l_lock(&lock->l_resource->lr_namespace->ns_lock);
/* Get this: if filter_blocking_ast is racing with ldlm_intent_policy,
- * such that mds_blocking_ast is called just before l_i_p takes the
+ * such that filter_blocking_ast is called just before l_i_p takes the
* ns_lock, then by the time we get the lock, we might not be the
* correct blocking function anymore. So check, and return early, if
* so. */
RETURN(0);
}
-static int filter_lock_dentry(struct obd_device *obd, struct dentry *de,
- int lock_mode, struct lustre_handle *lockh)
+static int filter_lock_dentry(struct obd_device *obd, struct dentry *dparent)
{
- struct ldlm_res_id res_id = { .name = {0} };
- int flags = 0, rc;
- ENTRY;
-
- 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);
+ down(&dparent->d_inode->i_sem);
+ return 0;
+}
- RETURN(rc == ELDLM_OK ? 0 : -ENOLCK); /* XXX translate ldlm code */
+/* We never dget the object parent, so DON'T dput it either */
+static void filter_parent_unlock(struct dentry *dparent)
+{
+ up(&dparent->d_inode->i_sem);
}
-static inline struct dentry *filter_parent(struct obd_device *obd,
- obd_mode mode, obd_id objid)
+/* 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(S_ISREG(mode)); /* only regular files for now */
- if (!S_ISREG(mode) || filter->fo_subdir_count == 0)
- return filter->fo_dentry_O_mode[(mode & S_IFMT) >> S_SHIFT];
+ if (group > 0 || filter->fo_subdir_count == 0)
+ return filter->fo_dentry_O_groups[group];
return filter->fo_dentry_O_sub[objid & (filter->fo_subdir_count - 1)];
}
-static inline struct dentry *filter_parent_lock(struct obd_device *obd,
- obd_mode mode, obd_id objid,
- int lock_mode,
- struct lustre_handle *lockh)
+/* 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)
{
- struct dentry *de = filter_parent(obd, mode, objid);
+ unsigned long now = jiffies;
+ struct dentry *dparent = filter_parent(obd, group, objid);
int rc;
- if (IS_ERR(de))
- return de;
+ if (IS_ERR(dparent))
+ return dparent;
- rc = filter_lock_dentry(obd, de, lock_mode, lockh);
- return rc ? ERR_PTR(rc) : de;
+ rc = filter_lock_dentry(obd, dparent);
+ if (time_after(jiffies, now + 15 * HZ))
+ CERROR("slow parent lock %lus\n", (jiffies - now) / HZ);
+ return rc ? ERR_PTR(rc) : dparent;
}
-static struct file *filter_obj_open(struct obd_export *export,
- __u64 id, __u32 type, int parent_mode,
- struct lustre_handle *parent_lockh)
-{
- struct obd_device *obd = export->exp_obd;
- struct filter_obd *filter = &obd->u.filter;
- struct super_block *sb = filter->fo_sb;
- struct dentry *dchild = NULL, *parent;
- struct filter_export_data *fed = &export->exp_filter_data;
- struct filter_dentry_data *fdd = NULL;
- struct filter_file_data *ffd = NULL;
- struct obd_run_ctxt saved;
- char name[24];
- struct file *file;
- int len, cleanup_phase = 0;
+/* How to get files, dentries, inodes from object id's.
+ *
+ * If dir_dentry is passed, the caller has already locked the parent
+ * appropriately for this operation (normally a write lock). If
+ * 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 *dparent = dir_dentry;
+ struct dentry *dchild;
+ char name[32];
+ int len;
ENTRY;
- push_ctxt(&saved, &filter->fo_ctxt, NULL);
-
- if (!sb || !sb->s_dev) {
- CERROR("fatal: device not initialized.\n");
- GOTO(cleanup, file = ERR_PTR(-ENXIO));
- }
-
- if (!id) {
- CERROR("fatal: invalid obdo "LPU64"\n", id);
- GOTO(cleanup, file = ERR_PTR(-ESTALE));
- }
-
- if (!(type & S_IFMT)) {
- CERROR("OBD %s, object "LPU64" has bad type: %o\n",
- __FUNCTION__, id, type);
- GOTO(cleanup, file = ERR_PTR(-EINVAL));
+ if (id == 0) {
+ CERROR("fatal: invalid object id 0\n");
+ RETURN(ERR_PTR(-ESTALE));
}
- ffd = filter_ffd_new();
- if (ffd == NULL) {
- CERROR("obdfilter: out of memory\n");
- GOTO(cleanup, file = ERR_PTR(-ENOMEM));
+ len = sprintf(name, LPU64, id);
+ if (dir_dentry == NULL) {
+ dparent = filter_parent_lock(obd, group, id);
+ if (IS_ERR(dparent))
+ RETURN(dparent);
}
-
- cleanup_phase = 1;
-
- /* We preallocate this to avoid blocking while holding fo_fddlock */
- OBD_ALLOC(fdd, sizeof *fdd);
- if (fdd == NULL) {
- CERROR("obdfilter: out of memory\n");
- GOTO(cleanup, file = ERR_PTR(-ENOMEM));
+ 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);
+ if (dir_dentry == NULL)
+ filter_parent_unlock(dparent);
+ if (IS_ERR(dchild)) {
+ CERROR("child lookup error %ld\n", PTR_ERR(dchild));
+ RETURN(dchild);
}
- cleanup_phase = 2;
-
- parent = filter_parent_lock(obd, type, id, parent_mode, parent_lockh);
- if (IS_ERR(parent))
- GOTO(cleanup, file = (void *)parent);
+ CDEBUG(D_INODE, "got child objid %s: %p, count = %d\n",
+ name, dchild, atomic_read(&dchild->d_count));
- cleanup_phase = 3;
+ LASSERT(atomic_read(&dchild->d_count) > 0);
- len = snprintf(name, sizeof(name), LPU64, id);
- dchild = lookup_one_len(name, parent, len);
- if (IS_ERR(dchild))
- GOTO(cleanup, file = (void *)dchild);
- LASSERT(dchild->d_inode);
-
- cleanup_phase = 4;
-
- /* dentry_open does a dput(de) and mntput(mds->mds_vfsmnt) on error */
- mntget(filter->fo_vfsmnt);
- file = dentry_open(dchild, filter->fo_vfsmnt, O_RDWR | O_LARGEFILE);
- if (IS_ERR(file)) {
- dchild = NULL; /* prevent a double dput in step 4 */
- CERROR("error opening %s: rc %ld\n", name, PTR_ERR(file));
- GOTO(cleanup, file);
- }
-
- spin_lock(&filter->fo_fddlock);
- if (dchild->d_fsdata) {
- spin_unlock(&filter->fo_fddlock);
- OBD_FREE(fdd, sizeof *fdd);
- fdd = dchild->d_fsdata;
- /* should only happen during client recovery */
- if (fdd->fdd_flags & FILTER_FLAG_DESTROY)
- CDEBUG(D_INODE,"opening destroyed object "LPU64"\n",id);
- atomic_inc(&fdd->fdd_open_count);
- } else {
- atomic_set(&fdd->fdd_open_count, 1);
- fdd->fdd_flags = 0;
- fdd->fdd_objid = id;
- /* If this is racy, then we can use {cmp}xchg and atomic_add */
- dchild->d_fsdata = fdd;
- spin_unlock(&filter->fo_fddlock);
- }
+ RETURN(dchild);
+}
- ffd->ffd_file = file;
- LASSERT(file->private_data == NULL);
- file->private_data = ffd;
+static int filter_prepare_destroy(struct obd_device *obd, obd_id objid)
+{
+ struct lustre_handle lockh;
+ int flags = LDLM_AST_DISCARD_DATA, rc;
+ struct ldlm_res_id res_id = { .name = { objid } };
+ ldlm_policy_data_t policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
- if (!dchild->d_op)
- dchild->d_op = &filter_dops;
- else
- LASSERT(dchild->d_op == &filter_dops);
+ ENTRY;
+ /* Tell the clients that the object is gone now and that they should
+ * 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);
- spin_lock(&fed->fed_lock);
- list_add(&ffd->ffd_export_list, &fed->fed_open_head);
- spin_unlock(&fed->fed_lock);
+ /* We only care about the side-effects, just drop the lock. */
+ if (rc == ELDLM_OK)
+ ldlm_lock_decref(&lockh, LCK_PW);
- CDEBUG(D_INODE, "opened objid "LPU64": rc = %p\n", id, file);
-cleanup:
- switch (cleanup_phase) {
- case 4:
- if (IS_ERR(file))
- l_dput(dchild);
- case 3:
- if (IS_ERR(file))
- ldlm_lock_decref(parent_lockh, parent_mode);
- case 2:
- if (IS_ERR(file))
- OBD_FREE(fdd, sizeof *fdd);
- case 1:
- if (IS_ERR(file))
- filter_ffd_destroy(ffd);
- filter_ffd_put(ffd);
- case 0:
- pop_ctxt(&saved, &filter->fo_ctxt, NULL);
- }
- RETURN(file);
+ RETURN(rc);
}
-/* Caller must hold i_sem on dir_dentry->d_inode */
-/* Caller must push us into kernel context */
-static int filter_destroy_internal(struct obd_device *obd,
- struct dentry *dir_dentry,
- struct dentry *object_dentry)
+/* Caller must hold LCK_PW on parent and push us into kernel context.
+ * Caller is also required to ensure that dchild->d_inode exists. */
+static int filter_destroy_internal(struct obd_device *obd, obd_id objid,
+ struct dentry *dparent,
+ struct dentry *dchild)
{
- struct inode *inode = object_dentry->d_inode;
+ struct inode *inode = dchild->d_inode;
int rc;
ENTRY;
if (inode->i_nlink != 1 || atomic_read(&inode->i_count) != 1) {
- CERROR("destroying objid %*s nlink = %d, count = %d\n",
- object_dentry->d_name.len,
- object_dentry->d_name.name,
- inode->i_nlink, atomic_read(&inode->i_count));
+ CERROR("destroying objid %*s nlink = %lu, count = %d\n",
+ dchild->d_name.len, dchild->d_name.name,
+ (unsigned long)inode->i_nlink,
+ atomic_read(&inode->i_count));
}
- rc = vfs_unlink(dir_dentry->d_inode, object_dentry);
+ rc = vfs_unlink(dparent->d_inode, dchild);
if (rc)
CERROR("error unlinking objid %*s: rc %d\n",
- object_dentry->d_name.len,
- object_dentry->d_name.name, rc);
+ dchild->d_name.len, dchild->d_name.name, rc);
RETURN(rc);
}
-/* If closing because we are failing this device, then
- don't do the unlink on close.
-*/
-static int filter_close_internal(struct obd_export *export,
- struct filter_file_data *ffd,
- struct obd_trans_info *oti,
- int failover)
-{
- struct obd_device *obd = export->exp_obd;
- struct filter_obd *filter = &obd->u.filter;
- struct file *filp = ffd->ffd_file;
- struct dentry *object_dentry = dget(filp->f_dentry);
- struct filter_dentry_data *fdd = object_dentry->d_fsdata;
- struct lustre_handle parent_lockh;
- int rc, rc2, cleanup_phase = 0;
- struct dentry *dir_dentry;
- struct obd_run_ctxt saved;
+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 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 list_head *tmp;
+ ldlm_error_t err;
+ int tmpflags = 0, rc, repsize[2] = {sizeof(struct ldlm_reply),
+ sizeof(struct ost_lvb) };
ENTRY;
- LASSERT(filp->private_data == ffd);
- LASSERT(fdd);
+ policy = ldlm_get_processing_policy(res);
+ LASSERT(policy != NULL);
+ LASSERT(req != NULL);
- rc = filp_close(filp, 0);
+ rc = lustre_pack_reply(req, 2, repsize, NULL);
+ if (rc)
+ RETURN(req->rq_status = rc);
- if (atomic_dec_and_test(&fdd->fdd_open_count) &&
- fdd->fdd_flags & FILTER_FLAG_DESTROY && !failover) {
- void *handle;
+ reply_lvb = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*reply_lvb));
+ LASSERT(reply_lvb != NULL);
- push_ctxt(&saved, &filter->fo_ctxt, NULL);
- cleanup_phase = 1;
+ //fixup_handle_for_resent_req(req, lock, &lockh);
- dir_dentry = filter_parent_lock(obd, S_IFREG, fdd->fdd_objid,
- LCK_PW, &parent_lockh);
- if (IS_ERR(dir_dentry))
- GOTO(cleanup, rc = PTR_ERR(dir_dentry));
- cleanup_phase = 2;
+ /* 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;
- handle = fsfilt_start(obd, dir_dentry->d_inode,
- FSFILT_OP_UNLINK);
- if (IS_ERR(handle))
- GOTO(cleanup, rc = PTR_ERR(handle));
+ l_lock(&res->lr_namespace->ns_lock);
- /* XXX unlink from PENDING directory now too */
- rc2 = filter_destroy_internal(obd, dir_dentry, object_dentry);
- if (rc2 && !rc)
- rc = rc2;
- rc = filter_finish_transno(export, handle, oti, rc);
- rc2 = fsfilt_commit(obd, dir_dentry->d_inode, handle, 0);
- if (rc2) {
- CERROR("error on commit, err = %d\n", rc2);
- if (!rc)
- rc = rc2;
- }
+ res->lr_tmp = &rpc_list;
+ rc = policy(lock, &tmpflags, 0, &err);
+ res->lr_tmp = NULL;
+
+ /* 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));
}
-cleanup:
- switch(cleanup_phase) {
- case 2:
- if (rc || oti == NULL) {
- ldlm_lock_decref(&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;
+ 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);
+ }
+
+ /* 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->lvb_size = res_lvb->lvb_size;
+ up(&res->lr_lvb_sem);
+
+ list_for_each(tmp, &res->lr_granted) {
+ struct ldlm_lock *tmplock =
+ list_entry(tmp, struct ldlm_lock, l_res_link);
+
+ if (tmplock->l_granted_mode == LCK_PR)
+ continue;
+
+ if (tmplock->l_policy_data.l_extent.end <=
+ reply_lvb->lvb_size)
+ continue;
+
+ if (l == NULL) {
+ l = LDLM_LOCK_GET(tmplock);
+ continue;
}
- case 1:
- pop_ctxt(&saved, &filter->fo_ctxt, NULL);
- case 0:
- f_dput(object_dentry);
- filter_ffd_destroy(ffd);
- break;
- default:
- CERROR("invalid cleanup_phase %d\n", cleanup_phase);
- LBUG();
+
+ 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);
- RETURN(rc);
+ /* There were no PW locks beyond the size in the LVB; finished. */
+ if (l == NULL)
+ RETURN(ELDLM_LOCK_ABORTED);
+
+ LASSERT(l->l_glimpse_ast != NULL);
+ 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->lvb_size = res_lvb->lvb_size;
+ up(&res->lr_lvb_sem);
+
+ LDLM_LOCK_PUT(l);
+
+ RETURN(ELDLM_LOCK_ABORTED);
}
-/* obd methods */
/* mount the file system (secretly) */
-static int filter_common_setup(struct obd_device *obd, obd_count len, void *buf,
- char *option)
+int filter_common_setup(struct obd_device *obd, obd_count len, void *buf,
+ char *option)
{
- struct obd_ioctl_data* data = buf;
- struct filter_obd *filter;
+ struct lustre_cfg* lcfg = buf;
+ struct filter_obd *filter = &obd->u.filter;
struct vfsmount *mnt;
int rc = 0;
ENTRY;
- if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2)
+ dev_clear_rdonly(2);
+
+ if (!lcfg->lcfg_inlbuf1 || !lcfg->lcfg_inlbuf2)
RETURN(-EINVAL);
- obd->obd_fsops = fsfilt_get_ops(data->ioc_inlbuf2);
+ obd->obd_fsops = fsfilt_get_ops(lcfg->lcfg_inlbuf2);
if (IS_ERR(obd->obd_fsops))
RETURN(PTR_ERR(obd->obd_fsops));
- mnt = do_kern_mount(data->ioc_inlbuf2, 0, data->ioc_inlbuf1, option);
+ 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);
- if (data->ioc_inllen3 > 0 && data->ioc_inlbuf3) {
- if (*data->ioc_inlbuf3 == 'f') {
+ if (lcfg->lcfg_inllen3 > 0 && lcfg->lcfg_inlbuf3) {
+ if (*lcfg->lcfg_inlbuf3 == 'f') {
obd->obd_replayable = 1;
obd_sync_filter = 1;
- CERROR("%s: configured for recovery and sync write\n",
- obd->obd_name);
+ CWARN("%s: recovery enabled\n", obd->obd_name);
} else {
- CERROR("unrecognised flag '%c'\n",
- *data->ioc_inlbuf3);
+ 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);
}
}
- filter = &obd->u.filter;
filter->fo_vfsmnt = mnt;
- filter->fo_fstype = strdup(data->ioc_inlbuf2);
- filter->fo_sb = mnt->mnt_root->d_inode->i_sb;
- CDEBUG(D_SUPER, "%s: mnt = %p\n", data->ioc_inlbuf1, 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(&filter->fo_ctxt);
- filter->fo_ctxt.pwdmnt = mnt;
- filter->fo_ctxt.pwd = mnt->mnt_root;
- filter->fo_ctxt.fs = get_ds();
+ 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;
rc = filter_prep(obd);
if (rc)
- GOTO(err_kfree, rc);
+ GOTO(err_mntput, rc);
spin_lock_init(&filter->fo_translock);
- spin_lock_init(&filter->fo_fddlock);
spin_lock_init(&filter->fo_objidlock);
INIT_LIST_HEAD(&filter->fo_export_list);
-
- obd->obd_namespace =
- ldlm_namespace_new("filter-tgt", LDLM_NAMESPACE_SERVER);
- if (!obd->obd_namespace)
+ 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;
+
+ obd->obd_namespace = ldlm_namespace_new("filter-tgt",
+ 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 = llog_cat_initialize(obd, 1);
+ if (rc) {
+ CERROR("failed to setup llogging subsystems\n");
+ GOTO(err_post, rc);
+ }
+
RETURN(0);
err_post:
filter_post(obd);
-err_kfree:
- kfree(filter->fo_fstype);
+err_mntput:
unlock_kernel();
- mntput(filter->fo_vfsmnt);
+ mntput(mnt);
filter->fo_sb = 0;
lock_kernel();
err_ops:
static int filter_setup(struct obd_device *obd, obd_count len, void *buf)
{
- struct obd_ioctl_data* data = buf;
- char *option = NULL;
-
- if (!strcmp(data->ioc_inlbuf2, "ext3"))
- option = "asyncdel";
-
- return filter_common_setup(obd, len, buf, option);
-}
-
-/* sanobd setup methods - use a specific mount option */
-static int filter_san_setup(struct obd_device *obd, obd_count len, void *buf)
-{
- struct obd_ioctl_data* data = buf;
+ struct lustre_cfg* lcfg = buf;
+ const char *str = NULL;
char *option = NULL;
+ int n = 0;
+ int rc;
- if (!data->ioc_inlbuf2)
- RETURN(-EINVAL);
-
- /* for extN/ext3 filesystem, we must mount it with 'writeback' mode */
- if (!strcmp(data->ioc_inlbuf2, "extN"))
- option = "data=writeback";
- else if (!strcmp(data->ioc_inlbuf2, "ext3"))
- option = "data=writeback,asyncdel";
- else
- LBUG(); /* just a reminder */
+ 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);
+ }
- return filter_common_setup(obd, len, buf, option);
+ rc = filter_common_setup(obd, len, buf, option);
+ if (option)
+ OBD_FREE(option, n);
+ return rc;
}
-static int filter_cleanup(struct obd_device *obd, int force, int failover)
+static int filter_cleanup(struct obd_device *obd, int flags)
{
- struct super_block *sb;
+ struct filter_obd *filter = &obd->u.filter;
ENTRY;
- if (failover)
+ if (flags & OBD_OPT_FAILOVER)
CERROR("%s: shutting down for failover; client state will"
" be preserved.\n", obd->obd_name);
if (!list_empty(&obd->obd_exports)) {
CERROR("%s: still has clients!\n", obd->obd_name);
- class_disconnect_exports(obd, failover);
+ class_disconnect_exports(obd, flags);
if (!list_empty(&obd->obd_exports)) {
CERROR("still has exports after forced cleanup?\n");
RETURN(-EBUSY);
}
}
- ldlm_namespace_free(obd->obd_namespace);
+ ldlm_namespace_free(obd->obd_namespace, flags & OBD_OPT_FORCE);
- sb = obd->u.filter.fo_sb;
- if (!obd->u.filter.fo_sb)
+ if (filter->fo_sb == NULL)
RETURN(0);
filter_post(obd);
- shrink_dcache_parent(sb->s_root);
- unlock_kernel();
-
- if (atomic_read(&obd->u.filter.fo_vfsmnt->mnt_count) > 1){
- CERROR("%s: mount point busy, mnt_count: %d\n", obd->obd_name,
- atomic_read(&obd->u.filter.fo_vfsmnt->mnt_count));
- }
+ shrink_dcache_parent(filter->fo_sb->s_root);
+ filter->fo_sb = 0;
- mntput(obd->u.filter.fo_vfsmnt);
- obd->u.filter.fo_sb = 0;
-/* destroy_buffers(obd->u.filter.fo_sb->s_dev);*/
+ 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));
- kfree(obd->u.filter.fo_fstype);
+ unlock_kernel();
+ mntput(filter->fo_vfsmnt);
+ //destroy_buffers(filter->fo_sb->s_dev);
+ filter->fo_sb = NULL;
fsfilt_put_ops(obd->obd_fsops);
-
lock_kernel();
+ dev_clear_rdonly(2);
+
RETURN(0);
}
-int filter_attach(struct obd_device *dev, obd_count len, void *data)
+static int filter_attach(struct obd_device *obd, obd_count len, void *data)
{
struct lprocfs_static_vars lvars;
- struct lprocfs_counters* cntrs;
int rc;
- lprocfs_init_vars(&lvars);
- rc = lprocfs_obd_attach(dev, lvars.obd_vars);
+ lprocfs_init_vars(filter, &lvars);
+ rc = lprocfs_obd_attach(obd, lvars.obd_vars);
if (rc != 0)
return rc;
- rc = lprocfs_alloc_obd_counters(dev, LPROC_FILTER_LAST);
+ rc = lprocfs_alloc_obd_stats(obd, LPROC_FILTER_LAST);
if (rc != 0)
return rc;
- /* Init obdfilter private counters here */
- cntrs = dev->counters;
- LPROCFS_COUNTER_INIT(&cntrs->cntr[LPROC_FILTER_READS],
- 0, NULL, "read", "reqs");
- LPROCFS_COUNTER_INIT(&cntrs->cntr[LPROC_FILTER_READ_BYTES],
- LPROCFS_CNTR_AVGMINMAX,
- NULL, "read_bytes", "bytes");
- LPROCFS_COUNTER_INIT(&cntrs->cntr[LPROC_FILTER_WRITES],
- 0, NULL, "write", "reqs");
-
- LPROCFS_COUNTER_INIT(&cntrs->cntr[LPROC_FILTER_WRITE_BYTES],
- LPROCFS_CNTR_AVGMINMAX,
- NULL, "write_bytes", "bytes");
- 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);
}
-int filter_detach(struct obd_device *dev)
+static int filter_detach(struct obd_device *dev)
{
- lprocfs_free_obd_counters(dev);
+ lprocfs_free_obd_stats(dev);
return lprocfs_obd_detach(dev);
}
{
struct obd_export *exp;
struct filter_export_data *fed;
- struct filter_client_data *fcd;
+ struct filter_client_data *fcd = NULL;
struct filter_obd *filter = &obd->u.filter;
int rc;
-
ENTRY;
- if (!conn || !obd || !cluuid)
+ if (conn == NULL || obd == NULL || cluuid == NULL)
RETURN(-EINVAL);
rc = class_connect(conn, obd, cluuid);
if (rc)
RETURN(rc);
exp = class_conn2export(conn);
- LASSERT(exp);
+ LASSERT(exp != NULL);
fed = &exp->exp_filter_data;
- class_export_put(exp);
- INIT_LIST_HEAD(&exp->exp_filter_data.fed_open_head);
- spin_lock_init(&exp->exp_filter_data.fed_lock);
+ spin_lock_init(&fed->fed_lock);
if (!obd->obd_replayable)
- RETURN(0);
+ GOTO(cleanup, rc = 0);
OBD_ALLOC(fcd, sizeof(*fcd));
if (!fcd) {
CERROR("filter: out of memory for client data\n");
- GOTO(out_export, rc = -ENOMEM);
+ GOTO(cleanup, rc = -ENOMEM);
}
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);
- if (rc)
- GOTO(out_fcd, rc);
- RETURN(rc);
+cleanup:
+ if (rc) {
+ if (fcd)
+ OBD_FREE(fcd, sizeof(*fcd));
+ class_disconnect(exp, 0);
+ } else {
+ class_export_put(exp);
+ }
+ return rc;
+}
+
+static int filter_precleanup(struct obd_device *obd, int flags)
+{
+ int rc = 0;
+ ENTRY;
-out_fcd:
- OBD_FREE(fcd, sizeof(*fcd));
-out_export:
- class_disconnect(conn, 0);
+ rc = obd_llog_finish(obd, 0);
+ if (rc)
+ CERROR("failed to cleanup llogging subsystem\n");
RETURN(rc);
}
-static void filter_destroy_export(struct obd_export *exp)
+/* 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, 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;
+
+ 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;
+ LASSERTF(fed->fed_grant + fed->fed_pending <= maxsize,
+ "cli %s/%p %lu+%lu > "LPU64"\n",
+ exp->exp_client_uuid.uuid, exp,
+ fed->fed_grant, fed->fed_pending, maxsize);
+ LASSERTF(fed->fed_dirty <= maxsize, "cli %s/%p %lu > "LPU64"\n",
+ exp->exp_client_uuid.uuid, exp,fed->fed_dirty,maxsize);
+ CDEBUG(D_CACHE,"%s: cli %s/%p dirty %lu pend %lu grant %lu\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;
+ 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);
+
+ CDEBUG(D_CACHE, "%s: cli %s/%p dirty %lu pend %lu grant %lu\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 %lu\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(exp->exp_obd->u.filter.fo_tot_pending >= fed->fed_pending,
+ "%s: tot_pending "LPU64" cli %s/%p fed_pending %lu\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 %lu\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;
- spin_lock(&fed->fed_lock);
- while (!list_empty(&fed->fed_open_head)) {
- struct filter_file_data *ffd;
-
- ffd = list_entry(fed->fed_open_head.next, typeof(*ffd),
- ffd_export_list);
- list_del(&ffd->ffd_export_list);
- spin_unlock(&fed->fed_lock);
- CERROR("force close file %*s (hdl %p:"LPX64") on disconnect\n",
- ffd->ffd_file->f_dentry->d_name.len,
- ffd->ffd_file->f_dentry->d_name.name,
- ffd, ffd->ffd_handle.h_cookie);
+ 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);
- filter_close_internal(exp, ffd, NULL, exp->exp_failover);
- spin_lock(&fed->fed_lock);
- }
- spin_unlock(&fed->fed_lock);
+ target_destroy_export(exp);
if (exp->exp_obd->obd_replayable)
- filter_client_free(exp, exp->exp_failover);
- EXIT;
+ 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);
}
/* also incredibly similar to mds_disconnect */
-static int filter_disconnect(struct lustre_handle *conn, int failover)
+static int filter_disconnect(struct obd_export *exp, int flags)
{
- struct obd_export *exp = class_conn2export(conn);
+ struct obd_device *obd = exp->exp_obd;
+ unsigned long irqflags;
+ struct llog_ctxt *ctxt;
int rc;
- unsigned long flags;
ENTRY;
LASSERT(exp);
- ldlm_cancel_locks_for_export(exp);
+ class_export_get(exp);
- spin_lock_irqsave(&exp->exp_lock, flags);
- exp->exp_failover = failover;
- spin_unlock_irqrestore(&exp->exp_lock, flags);
+ spin_lock_irqsave(&exp->exp_lock, irqflags);
+ exp->exp_flags = flags;
+ spin_unlock_irqrestore(&exp->exp_lock, irqflags);
- rc = class_disconnect(conn, failover);
+ if (!(flags & OBD_OPT_FORCE))
+ filter_grant_sanity_check(obd, __FUNCTION__);
+ filter_grant_discard(exp);
- fsfilt_sync(exp->exp_obd, exp->exp_obd->u.filter.fo_sb);
- class_export_put(exp);
- /* XXX cleanup preallocated inodes */
- RETURN(rc);
-}
+ /* Disconnect early so that clients can't keep using export */
+ rc = class_disconnect(exp, flags);
-static void filter_from_inode(struct obdo *oa, struct inode *inode, int valid)
-{
- int type = oa->o_mode & S_IFMT;
- ENTRY;
+ ldlm_cancel_locks_for_export(exp);
- CDEBUG(D_INFO, "src inode %lu (%p), dst obdo "LPU64" valid 0x%08x\n",
- inode->i_ino, inode, oa->o_id, valid);
- /* Don't copy the inode number in place of the object ID */
- obdo_from_inode(oa, inode, valid);
- oa->o_mode &= ~S_IFMT;
- oa->o_mode |= type;
+ fsfilt_sync(obd, obd->u.filter.fo_sb);
- if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
- obd_rdev rdev = kdev_t_to_nr(inode->i_rdev);
- oa->o_rdev = rdev;
- oa->o_valid |= OBD_MD_FLRDEV;
- }
+ /* flush any remaining cancel messages out to the target */
+ ctxt = llog_get_context(obd, LLOG_UNLINK_REPL_CTXT);
+ llog_sync(ctxt, exp);
- EXIT;
+ class_export_put(exp);
+ RETURN(rc);
}
-static struct dentry *__filter_oa2dentry(struct lustre_handle *conn,
- struct obdo *oa, int locked,char *what)
+struct dentry *__filter_oa2dentry(struct obd_device *obd,
+ struct obdo *oa, const char *what)
{
- struct dentry *dentry = NULL;
+ struct dentry *dchild = NULL;
+ obd_gr group = 0;
- if (oa->o_valid & OBD_MD_FLHANDLE) {
- struct lustre_handle *ost_handle = obdo_handle(oa);
- struct filter_file_data *ffd = filter_handle2ffd(ost_handle);
+ if (oa->o_valid & OBD_MD_FLGROUP)
+ group = oa->o_gr;
- if (ffd != NULL) {
- dentry = dget(ffd->ffd_file->f_dentry);
- filter_ffd_put(ffd);
- }
- }
+ dchild = filter_fid2dentry(obd, NULL, group, oa->o_id);
- if (!dentry) {
- struct obd_device *obd = class_conn2obd(conn);
- if (!obd) {
- CERROR("invalid client cookie "LPX64"\n", conn->cookie);
- RETURN(ERR_PTR(-EINVAL));
- }
- dentry = filter_fid2dentry(obd, filter_parent(obd, oa->o_mode,
- oa->o_id),
- oa->o_id, locked);
- }
-
- if (IS_ERR(dentry)) {
+ if (IS_ERR(dchild)) {
CERROR("%s error looking up object: "LPU64"\n", what, oa->o_id);
- RETURN(dentry);
+ RETURN(dchild);
}
- if (!dentry->d_inode) {
+ if (dchild->d_inode == NULL) {
CERROR("%s on non-existent object: "LPU64"\n", what, oa->o_id);
- f_dput(dentry);
+ f_dput(dchild);
RETURN(ERR_PTR(-ENOENT));
}
- return dentry;
+ return dchild;
}
-#define filter_oa2dentry(conn, oa, locked) __filter_oa2dentry(conn, oa, locked,\
- __FUNCTION__)
-
-static int filter_getattr(struct lustre_handle *conn, struct obdo *oa,
+static int filter_getattr(struct obd_export *exp, struct obdo *oa,
struct lov_stripe_md *md)
{
struct dentry *dentry = NULL;
+ struct obd_device *obd;
int rc = 0;
ENTRY;
- XPROCFS_BUMP_MYCPU_IOSTAT (st_getattr_reqs, 1);
+ obd = class_exp2obd(exp);
+ if (obd == NULL) {
+ CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
+ exp->exp_handle.h_cookie);
+ RETURN(-EINVAL);
+ }
- dentry = filter_oa2dentry(conn, oa, 1);
+ dentry = filter_oa2dentry(obd, oa);
if (IS_ERR(dentry))
RETURN(PTR_ERR(dentry));
- filter_from_inode(oa, dentry->d_inode, oa->o_valid);
+ /* Limit the valid bits in the return data to what we actually use */
+ oa->o_valid = OBD_MD_FLID;
+ obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
f_dput(dentry);
RETURN(rc);
}
/* this is called from filter_truncate() until we have filter_punch() */
-static int filter_setattr(struct lustre_handle *conn, struct obdo *oa,
+static int filter_setattr(struct obd_export *exp, struct obdo *oa,
struct lov_stripe_md *md, struct obd_trans_info *oti)
{
struct obd_run_ctxt saved;
- struct obd_export *export = class_conn2export(conn);
- struct obd_device *obd = class_conn2obd(conn);
- struct filter_obd *filter = &obd->u.filter;
+ struct filter_obd *filter;
struct dentry *dentry;
struct iattr iattr;
- struct inode *inode;
- void * handle;
+ struct ldlm_res_id res_id = { .name = { oa->o_id } };
+ struct ldlm_resource *res;
+ void *handle;
int rc, rc2;
ENTRY;
- XPROCFS_BUMP_MYCPU_IOSTAT (st_setattr_reqs, 1);
-
- dentry = filter_oa2dentry(conn, oa, 0);
+ LASSERT(oti != NULL);
+ dentry = filter_oa2dentry(exp->exp_obd, oa);
if (IS_ERR(dentry))
- GOTO(out_exp, rc = PTR_ERR(dentry));
+ RETURN(PTR_ERR(dentry));
+
+ filter = &exp->exp_obd->u.filter;
iattr_from_obdo(&iattr, oa, oa->o_valid);
- iattr.ia_mode = (iattr.ia_mode & ~S_IFMT) | S_IFREG;
- inode = dentry->d_inode;
- push_ctxt(&saved, &filter->fo_ctxt, NULL);
+ push_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
lock_kernel();
- if (iattr.ia_valid & ATTR_SIZE)
- down(&inode->i_sem);
- handle = fsfilt_start(obd, dentry->d_inode, FSFILT_OP_SETATTR);
+ 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);
if (IS_ERR(handle))
GOTO(out_unlock, rc = PTR_ERR(handle));
- if (inode->i_op->setattr)
- rc = inode->i_op->setattr(dentry, &iattr);
+ /* 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,
+ (long)&iattr.ia_attr_flags);
else
- rc = inode_setattr(inode, &iattr);
- rc = filter_finish_transno(export, handle, oti, rc);
- rc2 = fsfilt_commit(obd, dentry->d_inode, handle, 0);
+ 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);
if (!rc)
}
if (iattr.ia_valid & ATTR_SIZE) {
- up(&inode->i_sem);
- oa->o_valid = OBD_MD_FLBLOCKS | OBD_MD_FLCTIME | OBD_MD_FLMTIME;
- obdo_from_inode(oa, inode, oa->o_valid);
+ res = ldlm_resource_get(exp->exp_obd->obd_namespace, NULL,
+ res_id, LDLM_EXTENT, 0);
+ if (res == NULL) {
+ CERROR("!!! resource_get failed for object "LPU64" -- "
+ "filter_setattr with no lock?\n", oa->o_id);
+ } else {
+ if (res->lr_namespace->ns_lvbo &&
+ res->lr_namespace->ns_lvbo->lvbo_update) {
+ rc = res->lr_namespace->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);
+
out_unlock:
+ if (iattr.ia_valid & ATTR_SIZE)
+ up(&dentry->d_inode->i_sem);
unlock_kernel();
- pop_ctxt(&saved, &filter->fo_ctxt, NULL);
+ pop_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
f_dput(dentry);
- out_exp:
- class_export_put(export);
RETURN(rc);
}
-static int filter_open(struct lustre_handle *conn, struct obdo *oa,
- struct lov_stripe_md *ea, struct obd_trans_info *oti,
- struct obd_client_handle *och)
+/* XXX identical to osc_unpackmd */
+static int filter_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
+ struct lov_mds_md *lmm, int lmm_bytes)
{
- struct obd_export *export;
- struct lustre_handle *handle;
- struct filter_file_data *ffd;
- struct file *filp;
- struct lustre_handle parent_lockh;
- int rc = 0;
+ int lsm_size;
ENTRY;
- export = class_conn2export(conn);
- if (!export) {
- CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
- conn->cookie);
- GOTO(out, rc = -EINVAL);
+ if (lmm != NULL) {
+ if (lmm_bytes < sizeof (*lmm)) {
+ CERROR("lov_mds_md too small: %d, need %d\n",
+ lmm_bytes, (int)sizeof(*lmm));
+ RETURN(-EINVAL);
+ }
+ /* XXX LOV_MAGIC etc check? */
+
+ if (lmm->lmm_object_id == cpu_to_le64(0)) {
+ CERROR("lov_mds_md: zero lmm_object_id\n");
+ RETURN(-EINVAL);
+ }
}
- XPROCFS_BUMP_MYCPU_IOSTAT (st_open_reqs, 1);
+ lsm_size = lov_stripe_md_size(1);
+ if (lsmp == NULL)
+ RETURN(lsm_size);
- filp = filter_obj_open(export, oa->o_id, oa->o_mode,
- LCK_PR, &parent_lockh);
- if (IS_ERR(filp))
- GOTO(out, rc = PTR_ERR(filp));
+ if (*lsmp != NULL && lmm == NULL) {
+ OBD_FREE(*lsmp, lsm_size);
+ *lsmp = NULL;
+ RETURN(0);
+ }
- filter_from_inode(oa, filp->f_dentry->d_inode, oa->o_valid);
+ if (*lsmp == NULL) {
+ OBD_ALLOC(*lsmp, lsm_size);
+ if (*lsmp == NULL)
+ RETURN(-ENOMEM);
- ffd = filp->private_data;
- handle = obdo_handle(oa);
- handle->cookie = ffd->ffd_handle.h_cookie;
- oa->o_valid |= OBD_MD_FLHANDLE;
+ loi_init((*lsmp)->lsm_oinfo);
+ }
-out:
- class_export_put(export);
- if (!rc) {
- memcpy(&oti->oti_ack_locks[0].lock, &parent_lockh,
- sizeof(parent_lockh));
- oti->oti_ack_locks[0].mode = LCK_PR;
+ if (lmm != NULL) {
+ /* XXX zero *lsmp? */
+ (*lsmp)->lsm_object_id = le64_to_cpu (lmm->lmm_object_id);
+ LASSERT((*lsmp)->lsm_object_id);
}
- RETURN(rc);
+
+ (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
+
+ RETURN(lsm_size);
}
-static int filter_close(struct lustre_handle *conn, struct obdo *oa,
- struct lov_stripe_md *ea, struct obd_trans_info *oti)
+static void filter_destroy_precreated(struct obd_export *exp, struct obdo *oa,
+ struct filter_obd *filter)
{
- struct obd_export *exp = class_conn2export(conn);
- struct filter_file_data *ffd;
- struct filter_export_data *fed;
- int rc;
+ struct obdo doa; /* XXX obdo on stack */
+ __u64 last, id;
ENTRY;
+ LASSERT(oa);
- if (!exp) {
- CDEBUG(D_IOCTL, "invalid client cookie"LPX64"\n", conn->cookie);
- GOTO(out, rc = -EINVAL);
- }
+ 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;
+}
- XPROCFS_BUMP_MYCPU_IOSTAT (st_close_reqs, 1);
+/* 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)
+{
+ struct obd_device *obd = exp->exp_obd;
+ struct filter_obd *filter = &obd->u.filter;
+ int diff, rc;
+ ENTRY;
- if (!(oa->o_valid & OBD_MD_FLHANDLE)) {
- CERROR("no handle for close of objid "LPU64"\n", oa->o_id);
- GOTO(out, rc = -EINVAL);
- }
+ 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);
+
+ /* delete orphans request */
+ if ((oa->o_valid & OBD_MD_FLFLAGS) &&
+ (oa->o_flags & OBD_FL_DELORPHAN)) {
+ if (diff >= 0)
+ RETURN(diff);
+ if (-diff > 10000) { /* XXX make this smarter */
+ CERROR("ignoring bogus orphan destroy request: obdid "
+ LPU64" last_id "LPU64"\n",
+ oa->o_id, filter_last_id(filter, oa));
+ RETURN(-EINVAL);
+ }
+ 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(0);
+ } else {
+ /* only precreate if group == 0 and o_id is specfied */
+ if (!(oa->o_valid & OBD_FL_DELORPHAN) &&
+ (group != 0 || oa->o_id == 0))
+ RETURN(1);
- ffd = filter_handle2ffd(obdo_handle(oa));
- if (ffd == NULL) {
- CERROR("bad handle ("LPX64") for close\n",
- obdo_handle(oa)->cookie);
- GOTO(out, rc = -ESTALE);
+ LASSERT(diff >= 0);
+ RETURN(diff);
}
-
- fed = &exp->exp_filter_data;
- spin_lock(&fed->fed_lock);
- list_del(&ffd->ffd_export_list);
- spin_unlock(&fed->fed_lock);
-
- rc = filter_close_internal(exp, ffd, oti, 0);
- filter_ffd_put(ffd);
- GOTO(out, rc);
- out:
- class_export_put(exp);
- return rc;
}
-static int filter_create(struct lustre_handle *conn, struct obdo *oa,
- struct lov_stripe_md **ea, struct obd_trans_info *oti)
+/* 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)
{
- struct obd_export *export;
- struct obd_device *obd = class_conn2obd(conn);
- struct filter_obd *filter = &obd->u.filter;
- struct obd_run_ctxt saved;
- struct dentry *dir_dentry;
- struct lustre_handle parent_lockh;
- struct dentry *new = NULL;
- struct iattr;
- void *handle;
- int err, rc, cleanup_phase;
+ struct dentry *dchild = NULL;
+ struct filter_obd *filter;
+ struct dentry *dparent;
+ int err = 0, rc = 0, i;
+ __u64 next_id;
+ int recreate_obj = 0;
+ void *handle = NULL;
ENTRY;
- if (!obd) {
- CERROR("invalid client cookie "LPX64"\n", conn->cookie);
- RETURN(-EINVAL);
+ filter = &obd->u.filter;
+
+ if ((oa->o_valid & OBD_MD_FLFLAGS) &&
+ (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
+ recreate_obj = 1;
}
- export = class_conn2export(conn);
- XPROCFS_BUMP_MYCPU_IOSTAT (st_create_reqs, 1);
+ for (i = 0; i < *num && err == 0; i++) {
+ int cleanup_phase = 0;
- oa->o_id = filter_next_id(obd);
+ if (recreate_obj) {
+ __u64 last_id;
+ next_id = oa->o_id;
+ last_id = filter_last_id(filter, NULL);
+ if (next_id > last_id) {
+ CERROR("Error: Trying to recreate obj greater"
+ "than last id "LPD64" > "LPD64"\n",
+ next_id, last_id);
+ RETURN(-EINVAL);
+ }
+ } else
+ next_id = filter_last_id(filter, NULL) + 1;
- push_ctxt(&saved, &filter->fo_ctxt, NULL);
- retry:
- cleanup_phase = 0;
- dir_dentry = filter_parent_lock(obd, S_IFREG, oa->o_id, LCK_PW,
- &parent_lockh);
- if (IS_ERR(dir_dentry))
- GOTO(cleanup, rc = PTR_ERR(dir_dentry));
- cleanup_phase = 1;
+ CDEBUG(D_INFO, "precreate objid "LPU64"\n", next_id);
- new = filter_fid2dentry(obd, dir_dentry, oa->o_id, 0);
- if (IS_ERR(new))
- GOTO(cleanup, rc = PTR_ERR(new));
- if (new->d_inode) {
- char buf[32];
+ dparent = filter_parent_lock(obd, group, next_id);
+ if (IS_ERR(dparent))
+ GOTO(cleanup, rc = PTR_ERR(dparent));
+ cleanup_phase = 1;
- /* This would only happen if lastobjid was bad on disk */
- CERROR("Serious error: objid %s already exists; is this "
- "filesystem corrupt? I will try to work around it.\n",
- filter_id(buf, filter, oa->o_id, oa->o_mode));
- f_dput(new);
- ldlm_lock_decref(&parent_lockh, LCK_PW);
- oa->o_id = filter_next_id(obd);
- goto retry;
- }
+ dchild = filter_fid2dentry(obd, dparent, group, next_id);
+ if (IS_ERR(dchild))
+ GOTO(cleanup, rc = PTR_ERR(dchild));
+ cleanup_phase = 2;
- cleanup_phase = 2;
- handle = fsfilt_start(obd, dir_dentry->d_inode, FSFILT_OP_CREATE);
- if (IS_ERR(handle))
- GOTO(cleanup, rc = PTR_ERR(handle));
+ if (dchild->d_inode != NULL) {
+ /* This would only happen if lastobjid was bad on disk*/
+ /* Could also happen if recreating missing obj but
+ * already exists
+ */
+ if (recreate_obj) {
+ CERROR("Serious error: recreating obj %*s but "
+ "obj already exists \n",
+ dchild->d_name.len, dchild->d_name.name);
+ } else {
+ CERROR("Serious error: objid %*s already "
+ "exists; is this filesystem corrupt?\n",
+ dchild->d_name.len, dchild->d_name.name);
+ }
+ GOTO(cleanup, rc = -EEXIST);
+ }
- rc = vfs_create(dir_dentry->d_inode, new, oa->o_mode);
- if (rc)
- CERROR("create failed rc = %d\n", rc);
+ handle = fsfilt_start(obd, dparent->d_inode,
+ FSFILT_OP_CREATE_LOG, NULL);
+ if (IS_ERR(handle))
+ GOTO(cleanup, rc = PTR_ERR(handle));
+ cleanup_phase = 3;
- rc = filter_finish_transno(export, handle, oti, rc);
- err = filter_update_server_data(filter->fo_rcvd_filp, filter->fo_fsd);
- if (err) {
- CERROR("unable to write lastobjid but file created\n");
- if (!rc)
- rc = err;
- }
- err = fsfilt_commit(obd, dir_dentry->d_inode, handle, 0);
- if (err) {
- CERROR("error on commit, err = %d\n", err);
- if (!rc)
- rc = err;
+ rc = ll_vfs_create(dparent->d_inode, dchild, S_IFREG, NULL);
+ if (rc) {
+ CERROR("create failed rc = %d\n", rc);
+ GOTO(cleanup, rc);
+ }
+
+ if (!recreate_obj) {
+ 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:
+ switch(cleanup_phase) {
+ case 3:
+ err = fsfilt_commit(obd, dparent->d_inode, handle, 0);
+ if (err) {
+ CERROR("error on commit, err = %d\n", err);
+ if (!rc)
+ rc = err;
+ }
+ case 2:
+ f_dput(dchild);
+ case 1:
+ filter_parent_unlock(dparent);
+ case 0:
+ break;
+ }
+
+ if (rc)
+ break;
}
+ *num = i;
- if (rc)
- GOTO(cleanup, rc);
+ CDEBUG(D_INFO, "filter_precreate() created %d objects\n", i);
+ RETURN(rc);
+}
- /* Set flags for fields we have set in the inode struct */
- oa->o_valid = OBD_MD_FLID | OBD_MD_FLBLKSZ | OBD_MD_FLBLOCKS |
- OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME;
- filter_from_inode(oa, new->d_inode, oa->o_valid);
+static int filter_create(struct obd_export *exp, struct obdo *oa,
+ struct lov_stripe_md **ea, struct obd_trans_info *oti)
+{
+ struct obd_device *obd = NULL;
+ struct obd_run_ctxt saved;
+ struct lov_stripe_md *lsm = NULL;
+ obd_gr group = 0;
+ int rc = 0, diff;
+ ENTRY;
- EXIT;
-cleanup:
- switch(cleanup_phase) {
- case 2:
- f_dput(new);
- case 1: /* locked parent dentry */
- if (rc || oti == NULL) {
- ldlm_lock_decref(&parent_lockh, LCK_PW);
+ if (oa->o_valid & OBD_MD_FLGROUP)
+ group = oa->o_gr;
+
+ 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);
+ }
+ }
+
+ obd = exp->exp_obd;
+ push_ctxt(&saved, &obd->obd_ctxt, NULL);
+
+ if ((oa->o_valid & OBD_MD_FLFLAGS) &&
+ (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
+ if (oa->o_id > filter_last_id(&obd->u.filter, oa)) {
+ CERROR("recreate objid "LPU64" > last id "LPU64"\n",
+ oa->o_id, filter_last_id(&obd->u.filter, oa));
+ rc = -EINVAL;
} else {
- memcpy(&oti->oti_ack_locks[0].lock, &parent_lockh,
- sizeof(parent_lockh));
- oti->oti_ack_locks[0].mode = LCK_PW;
+ diff = 1;
+ rc = filter_precreate(obd, oa, group, &diff);
}
- case 0:
- pop_ctxt(&saved, &filter->fo_ctxt, NULL);
- class_export_put(export);
- break;
- default:
- CERROR("invalid cleanup_phase %d\n", cleanup_phase);
- LBUG();
+ } else {
+ 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;
+ }
+ }
+
+ 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;
}
RETURN(rc);
}
-static int filter_destroy(struct lustre_handle *conn, struct obdo *oa,
+static int filter_destroy(struct obd_export *exp, struct obdo *oa,
struct lov_stripe_md *ea, struct obd_trans_info *oti)
{
- struct obd_export *export;
- struct obd_device *obd = class_conn2obd(conn);
- struct filter_obd *filter = &obd->u.filter;
- struct dentry *dir_dentry, *object_dentry = NULL;
- struct filter_dentry_data *fdd;
+ struct obd_device *obd;
+ struct filter_obd *filter;
+ struct dentry *dchild = NULL, *dparent = NULL;
struct obd_run_ctxt saved;
void *handle = NULL;
- struct lustre_handle parent_lockh;
- int rc, rc2, cleanup_phase = 0;
+ struct llog_cookie *fcc = NULL;
+ int rc, rc2, cleanup_phase = 0, have_prepared = 0;
+ obd_gr group = 0;
ENTRY;
- if (!obd) {
- CERROR("invalid client cookie "LPX64"\n", conn->cookie);
- RETURN(-EINVAL);
- }
+ if (oa->o_valid & OBD_MD_FLGROUP)
+ group = oa->o_gr;
- export = class_conn2export(conn);
- XPROCFS_BUMP_MYCPU_IOSTAT (st_destroy_reqs, 1);
+ obd = exp->exp_obd;
+ filter = &obd->u.filter;
- CDEBUG(D_INODE, "destroying objid "LPU64"\n", oa->o_id);
+ push_ctxt(&saved, &obd->obd_ctxt, NULL);
- push_ctxt(&saved, &filter->fo_ctxt, NULL);
- dir_dentry = filter_parent_lock(obd, oa->o_mode, oa->o_id,
- LCK_PW, &parent_lockh);
- if (IS_ERR(dir_dentry))
- GOTO(cleanup, rc = PTR_ERR(dir_dentry));
+ acquire_locks:
+ dparent = filter_parent_lock(obd, group, oa->o_id);
+ if (IS_ERR(dparent))
+ GOTO(cleanup, rc = PTR_ERR(dparent));
cleanup_phase = 1;
- object_dentry = filter_oa2dentry(conn, oa, 0);
- if (IS_ERR(object_dentry))
+ dchild = filter_fid2dentry(obd, dparent, group, oa->o_id);
+ if (IS_ERR(dchild))
GOTO(cleanup, rc = -ENOENT);
cleanup_phase = 2;
- handle = fsfilt_start(obd, dir_dentry->d_inode, FSFILT_OP_UNLINK);
+ if (dchild->d_inode == NULL) {
+ CDEBUG(D_INODE, "destroying non-existent object "LPU64"\n",
+ oa->o_id);
+ GOTO(cleanup, rc = -ENOENT);
+ }
+
+ if (!have_prepared) {
+ /* If we're really going to destroy the object, get ready
+ * by getting the clients to discard their cached data.
+ *
+ * We have to drop the parent lock, because
+ * filter_prepare_destroy will acquire a PW on the object, and
+ * we don't want to deadlock with an incoming write to the
+ * object, which has the extent PW and then wants to get the
+ * parent dentry to do the lookup.
+ *
+ * We dput the child because it's not worth the extra
+ * complication of condition the above code to skip it on the
+ * second time through. */
+ f_dput(dchild);
+ filter_parent_unlock(dparent);
+
+ filter_prepare_destroy(obd, oa->o_id);
+ 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;
- fdd = object_dentry->d_fsdata;
- if (fdd && atomic_read(&fdd->fdd_open_count)) {
- if (!(fdd->fdd_flags & FILTER_FLAG_DESTROY)) {
- fdd->fdd_flags |= FILTER_FLAG_DESTROY;
- /* XXX put into PENDING directory in case of crash */
- CDEBUG(D_INODE,
- "defer destroy of %dx open objid "LPU64"\n",
- atomic_read(&fdd->fdd_open_count), oa->o_id);
- } else
- CDEBUG(D_INODE,
- "repeat destroy of %dx open objid "LPU64"\n",
- atomic_read(&fdd->fdd_open_count), oa->o_id);
- GOTO(cleanup, rc = 0);
+ /* Our MDC connection is established by the MDS to us */
+ if (oa->o_valid & OBD_MD_FLCOOKIE) {
+ OBD_ALLOC(fcc, sizeof(*fcc));
+ if (fcc != NULL)
+ memcpy(fcc, obdo_logcookie(oa), sizeof(*fcc));
}
- rc = filter_destroy_internal(obd, dir_dentry, object_dentry);
+ rc = filter_destroy_internal(obd, oa->o_id, dparent, dchild);
cleanup:
switch(cleanup_phase) {
case 3:
- rc = filter_finish_transno(export, handle, oti, rc);
- rc2 = fsfilt_commit(obd, dir_dentry->d_inode, handle, 0);
+ if (fcc != NULL) {
+ if (oti != NULL)
+ fsfilt_add_journal_cb(obd, 0, oti->oti_handle,
+ filter_cancel_cookies_cb,
+ fcc);
+ else
+ fsfilt_add_journal_cb(obd, 0, handle,
+ filter_cancel_cookies_cb,
+ fcc);
+ }
+ rc = filter_finish_transno(exp, oti, rc);
+ rc2 = fsfilt_commit(obd, dparent->d_inode, handle, 0);
if (rc2) {
CERROR("error on commit, err = %d\n", rc2);
if (!rc)
rc = rc2;
}
case 2:
- f_dput(object_dentry);
+ f_dput(dchild);
case 1:
- if (rc || oti == NULL) {
- ldlm_lock_decref(&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);
case 0:
- pop_ctxt(&saved, &filter->fo_ctxt, NULL);
- class_export_put(export);
+ pop_ctxt(&saved, &obd->obd_ctxt, NULL);
break;
default:
CERROR("invalid cleanup_phase %d\n", cleanup_phase);
}
/* NB start and end are used for punch, but not truncate */
-static int filter_truncate(struct lustre_handle *conn, struct obdo *oa,
+static int filter_truncate(struct obd_export *exp, struct obdo *oa,
struct lov_stripe_md *lsm,
obd_off start, obd_off end,
struct obd_trans_info *oti)
int error;
ENTRY;
- XPROCFS_BUMP_MYCPU_IOSTAT (st_punch_reqs, 1);
-
if (end != OBD_OBJECT_EOF)
CERROR("PUNCH not supported, only truncate: end = "LPX64"\n",
end);
CDEBUG(D_INODE, "calling truncate for object "LPU64", valid = %x, "
"o_size = "LPD64"\n", oa->o_id, oa->o_valid, start);
oa->o_size = start;
- error = filter_setattr(conn, oa, NULL, oti);
+ error = filter_setattr(exp, oa, NULL, oti);
RETURN(error);
}
-static inline void lustre_put_page(struct page *page)
+static int filter_sync(struct obd_export *exp, struct obdo *oa,
+ struct lov_stripe_md *lsm, obd_off start, obd_off end)
{
- page_cache_release(page);
-}
-
-static int filter_start_page_read(struct inode *inode, struct niobuf_local *lnb)
-{
- struct address_space *mapping = inode->i_mapping;
- struct page *page;
- unsigned long index = lnb->offset >> PAGE_SHIFT;
- int rc;
-
- page = grab_cache_page(mapping, index); /* locked page */
- if (IS_ERR(page))
- return lnb->rc = PTR_ERR(page);
-
- lnb->page = page;
-
- if (inode->i_size < lnb->offset + lnb->len - 1)
- lnb->rc = inode->i_size - lnb->offset;
- else
- lnb->rc = lnb->len;
-
- if (PageUptodate(page)) {
- unlock_page(page);
- return 0;
- }
-
- rc = mapping->a_ops->readpage(NULL, page);
- if (rc < 0) {
- CERROR("page index %lu, rc = %d\n", index, rc);
- lnb->page = NULL;
- lustre_put_page(page);
- return lnb->rc = rc;
- }
-
- return 0;
-}
-
-static int filter_finish_page_read(struct niobuf_local *lnb)
-{
- if (lnb->page == NULL)
- return 0;
+ struct obd_run_ctxt saved;
+ struct filter_obd *filter;
+ struct dentry *dentry;
+ struct llog_ctxt *ctxt;
+ int rc, rc2;
+ ENTRY;
- if (PageUptodate(lnb->page))
- return 0;
+ filter = &exp->exp_obd->u.filter;
- wait_on_page(lnb->page);
- if (!PageUptodate(lnb->page)) {
- CERROR("page index %lu/offset "LPX64" not uptodate\n",
- lnb->page->index, lnb->offset);
- GOTO(err_page, lnb->rc = -EIO);
- }
- if (PageError(lnb->page)) {
- CERROR("page index %lu/offset "LPX64" has error\n",
- lnb->page->index, lnb->offset);
- GOTO(err_page, lnb->rc = -EIO);
+ /* 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);
+ /* flush any remaining cancel messages out to the target */
+ ctxt = llog_get_context(exp->exp_obd, LLOG_UNLINK_REPL_CTXT);
+ llog_sync(ctxt, exp);
+ RETURN(rc);
}
- return 0;
+ dentry = filter_oa2dentry(exp->exp_obd, oa);
+ if (IS_ERR(dentry))
+ RETURN(PTR_ERR(dentry));
-err_page:
- lustre_put_page(lnb->page);
- lnb->page = NULL;
- return lnb->rc;
-}
+ push_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
-static struct page *lustre_get_page_write(struct inode *inode,
- unsigned long index)
-{
- struct address_space *mapping = inode->i_mapping;
- struct page *page;
- int rc;
+ down(&dentry->d_inode->i_sem);
+ rc = filemap_fdatasync(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;
- page = grab_cache_page(mapping, index); /* locked page */
+ if (file->f_op && file->f_op->fsync)
+ rc = file->f_op->fsync(NULL, dentry, 1);
- if (!IS_ERR(page)) {
- /* Note: Called with "O" and "PAGE_SIZE" this is essentially
- * a no-op for most filesystems, because we write the whole
- * page. For partial-page I/O this will read in the page.
- */
- rc = mapping->a_ops->prepare_write(NULL, page, 0, PAGE_SIZE);
- if (rc) {
- CERROR("page index %lu, rc = %d\n", index, rc);
- if (rc != -ENOSPC)
- LBUG();
- GOTO(err_unlock, rc);
- }
- /* XXX not sure if we need this if we are overwriting page */
- if (PageError(page)) {
- CERROR("error on page index %lu, rc = %d\n", index, rc);
- LBUG();
- GOTO(err_unlock, rc = -EIO);
- }
+ rc2 = filemap_fdatawait(dentry->d_inode->i_mapping);
+ if (!rc)
+ rc = rc2;
}
- return page;
-
-err_unlock:
- unlock_page(page);
- lustre_put_page(page);
- return ERR_PTR(rc);
-}
+ up(&dentry->d_inode->i_sem);
-#if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
-int waitfor_one_page(struct page *page)
-{
- wait_on_page_locked(page);
- return 0;
-}
-#endif
+ oa->o_valid = OBD_MD_FLID;
+ obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
-#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
-/* We should only change the file mtime (and not the ctime, like
- * update_inode_times() in generic_file_write()) when we only change data.
- */
-static inline void inode_update_time(struct inode *inode, int ctime_too)
-{
- time_t now = CURRENT_TIME;
- if (inode->i_mtime == now && (!ctime_too || inode->i_ctime == now))
- return;
- inode->i_mtime = now;
- if (ctime_too)
- inode->i_ctime = now;
- mark_inode_dirty_sync(inode);
-}
-#endif
+ pop_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
-static int lustre_commit_write(struct niobuf_local *lnb)
-{
- struct page *page = lnb->page;
- unsigned from = lnb->offset & ~PAGE_MASK;
- unsigned to = from + lnb->len;
- struct inode *inode = page->mapping->host;
- int err;
-
- LASSERT(to <= PAGE_SIZE);
- err = page->mapping->a_ops->commit_write(NULL, page, from, to);
- if (!err && IS_SYNC(inode))
- waitfor_one_page(page);
- //SetPageUptodate(page); // the client commit_write will do this
-
- SetPageReferenced(page);
- unlock_page(page);
- lustre_put_page(page);
- return err;
+ f_dput(dentry);
+ RETURN(rc);
}
-int filter_get_page_write(struct inode *inode, struct niobuf_local *lnb,
- int *pglocked)
+static int filter_statfs(struct obd_device *obd, struct obd_statfs *osfs,
+ unsigned long max_age)
{
- unsigned long index = lnb->offset >> PAGE_SHIFT;
- struct address_space *mapping = inode->i_mapping;
- struct page *page;
+ struct filter_obd *filter = &obd->u.filter;
+ int blockbits = filter->fo_sb->s_blocksize_bits;
int rc;
+ ENTRY;
- //ASSERT_PAGE_INDEX(index, GOTO(err, rc = -EINVAL));
- if (*pglocked)
- page = grab_cache_page_nowait(mapping, index); /* locked page */
- else
- page = grab_cache_page(mapping, index); /* locked page */
+ /* 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);
+ 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);
- /* This page is currently locked, so get a temporary page instead. */
- if (!page) {
- unsigned long addr;
- CDEBUG(D_ERROR,"ino %lu page %ld locked\n", inode->i_ino,index);
- addr = __get_free_pages(GFP_KERNEL, 0); /* locked page */
- if (!addr) {
- CERROR("no memory for a temp page\n");
- GOTO(err, rc = -ENOMEM);
- }
- POISON((void *)addr, 0xBA, PAGE_SIZE);
- page = virt_to_page(addr);
- page->index = index;
- lnb->page = page;
- lnb->flags |= N_LOCAL_TEMP_PAGE;
- } else if (!IS_ERR(page)) {
- (*pglocked)++;
-
- rc = mapping->a_ops->prepare_write(NULL, page,
- lnb->offset & ~PAGE_MASK,
- lnb->len);
- if (rc) {
- if (rc != -ENOSPC)
- CERROR("page index %lu, rc = %d\n", index, rc);
- GOTO(err_unlock, rc);
- }
- /* XXX not sure if we need this if we are overwriting page */
- if (PageError(page)) {
- CERROR("error on page index %lu, rc = %d\n", index, rc);
- LBUG();
- GOTO(err_unlock, rc = -EIO);
- }
- lnb->page = page;
- }
+ filter_grant_sanity_check(obd, __FUNCTION__);
- return 0;
+ osfs->os_bavail -= min(osfs->os_bavail,
+ (filter->fo_tot_dirty + filter->fo_tot_pending +
+ osfs->os_bsize -1) >> blockbits);
-err_unlock:
- unlock_page(page);
- lustre_put_page(page);
-err:
- return lnb->rc = rc;
-}
-
-/*
- * We need to balance prepare_write() calls with commit_write() calls.
- * If the page has been prepared, but we have no data for it, we don't
- * want to overwrite valid data on disk, but we still need to zero out
- * data for space which was newly allocated. Like part of what happens
- * in __block_prepare_write() for newly allocated blocks.
- *
- * XXX currently __block_prepare_write() creates buffers for all the
- * pages, and the filesystems mark these buffers as BH_New if they
- * were newly allocated from disk. We use the BH_New flag similarly.
- */
-static int filter_commit_write(struct niobuf_local *lnb, int err)
-{
-#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
- if (err) {
- unsigned block_start, block_end;
- struct buffer_head *bh, *head = lnb->page->buffers;
- unsigned blocksize = head->b_size;
-
- /* debugging: just seeing if this ever happens */
- CDEBUG(err == -ENOSPC ? D_INODE : D_ERROR,
- "called for ino %lu:%lu on err %d\n",
- lnb->page->mapping->host->i_ino, lnb->page->index, err);
-
- /* Currently one buffer per page, but in the future... */
- for (bh = head, block_start = 0; bh != head || !block_start;
- block_start = block_end, bh = bh->b_this_page) {
- block_end = block_start + blocksize;
- if (buffer_new(bh)) {
- memset(kmap(lnb->page) + block_start, 0,
- blocksize);
- kunmap(lnb->page);
- }
- }
- }
-#endif
- return lustre_commit_write(lnb);
+ RETURN(rc);
}
-static int filter_preprw(int cmd, struct obd_export *export,
- int objcount, struct obd_ioobj *obj,
- int niocount, struct niobuf_remote *nb,
- struct niobuf_local *res, void **desc_private,
- struct obd_trans_info *oti)
+static int filter_get_info(struct obd_export *exp, __u32 keylen,
+ void *key, __u32 *vallen, void *val)
{
- struct obd_run_ctxt saved;
struct obd_device *obd;
- struct obd_ioobj *o;
- struct niobuf_remote *rnb;
- struct niobuf_local *lnb;
- struct fsfilt_objinfo *fso;
- struct dentry *dentry;
- struct inode *inode;
- struct lprocfs_counters *cntrs;
- int pglocked = 0, rc = 0, i, j;
-
ENTRY;
- if ((cmd & OBD_BRW_WRITE) != 0)
- XPROCFS_BUMP_MYCPU_IOSTAT (st_write_reqs, 1);
- else
- XPROCFS_BUMP_MYCPU_IOSTAT (st_read_reqs, 1);
-
- memset(res, 0, niocount * sizeof(*res));
-
- obd = export->exp_obd;
- if (obd == NULL)
+ obd = class_exp2obd(exp);
+ if (obd == NULL) {
+ CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
+ exp->exp_handle.h_cookie);
RETURN(-EINVAL);
-
- cntrs = obd->counters;
- if ((cmd & OBD_BRW_WRITE) != 0)
- LPROCFS_COUNTER_INCBY1(&cntrs->cntr[LPROC_FILTER_WRITES]);
- else
- LPROCFS_COUNTER_INCBY1(&cntrs->cntr[LPROC_FILTER_READS]);
-
- // theoretically we support multi-obj BRW RPCs, but until then...
- LASSERT(objcount == 1);
-
- OBD_ALLOC(fso, objcount * sizeof(*fso));
- if (!fso)
- RETURN(-ENOMEM);
-
- push_ctxt(&saved, &obd->u.filter.fo_ctxt, NULL);
-
- for (i = 0, o = obj; i < objcount; i++, o++) {
- struct filter_dentry_data *fdd;
-
- LASSERT(o->ioo_bufcnt);
-
- dentry = filter_fid2dentry(obd, filter_parent(obd, S_IFREG,
- o->ioo_id),
- o->ioo_id, 0);
-
- if (IS_ERR(dentry))
- GOTO(out_objinfo, rc = PTR_ERR(dentry));
-
- fso[i].fso_dentry = dentry;
- fso[i].fso_bufcnt = o->ioo_bufcnt;
-
- if (!dentry->d_inode) {
- CERROR("trying to BRW to non-existent file "LPU64"\n",
- o->ioo_id);
- GOTO(out_objinfo, rc = -ENOENT);
- }
-
- /* If we ever start to support mutli-object BRW RPCs, we will
- * need to get locks on mulitple inodes (in order) or use the
- * DLM to do the locking for us (and use the same locking in
- * filter_setattr() for truncate). That isn't all, because
- * there still exists the possibility of a truncate starting
- * a new transaction while holding the ext3 rwsem = write
- * while some writes (which have started their transactions
- * here) blocking on the ext3 rwsem = read => lock inversion.
- *
- * The handling gets very ugly when dealing with locked pages.
- * It may be easier to just get rid of the locked page code
- * (which has problems of its own) and either discover we do
- * not need it anymore (i.e. it was a symptom of another bug)
- * or ensure we get the page locks in an appropriate order.
- */
- if (cmd & OBD_BRW_WRITE)
- down(&dentry->d_inode->i_sem);
- fdd = dentry->d_fsdata;
- if (!fdd || !atomic_read(&fdd->fdd_open_count))
- CDEBUG(D_PAGE, "I/O to unopened object "LPU64"\n",
- o->ioo_id);
- }
-
- if (cmd & OBD_BRW_WRITE) {
- *desc_private = fsfilt_brw_start(obd, objcount, fso,
- niocount, nb);
- if (IS_ERR(*desc_private)) {
- rc = PTR_ERR(*desc_private);
- CDEBUG(rc == -ENOSPC ? D_INODE : D_ERROR,
- "error starting transaction: rc = %d\n", rc);
- *desc_private = NULL;
- GOTO(out_objinfo, rc);
- }
}
- for (i = 0, o = obj, rnb = nb, lnb = res; i < objcount; i++, o++) {
- dentry = fso[i].fso_dentry;
- inode = dentry->d_inode;
-
- for (j = 0; j < o->ioo_bufcnt; j++, rnb++, lnb++) {
- if (j == 0)
- lnb->dentry = dentry;
- else
- lnb->dentry = dget(dentry);
-
- lnb->offset = rnb->offset;
- lnb->len = rnb->len;
- lnb->flags = rnb->flags;
-
- if (cmd & OBD_BRW_WRITE) {
- rc = filter_get_page_write(inode,lnb,&pglocked);
-
- XPROCFS_BUMP_MYCPU_IOSTAT(st_write_bytes,
- lnb->len);
- LPROCFS_COUNTER_INCR(&cntrs->cntr[LPROC_FILTER_WRITE_BYTES], lnb->len);
- } else if (inode->i_size <= rnb->offset) {
- /* If there's no more data, abort early.
- * lnb->page == NULL and lnb->rc == 0, so it's
- * easy to detect later. */
- f_dput(lnb->dentry);
- lnb->dentry = NULL;
- break;
- } else {
- rc = filter_start_page_read(inode, lnb);
-
- XPROCFS_BUMP_MYCPU_IOSTAT(st_read_bytes,
- lnb->len);
- LPROCFS_COUNTER_INCR(&cntrs->cntr[LPROC_FILTER_READ_BYTES], lnb->len);
- }
-
- if (rc) {
- CDEBUG(rc == -ENOSPC ? D_INODE : D_ERROR,
- "error on page @"LPU64"%u/%u: rc = %d\n",
- lnb->offset, j, o->ioo_bufcnt, rc);
- f_dput(dentry);
- GOTO(out_pages, rc);
- }
-
- if ((cmd & OBD_BRW_READ) && lnb->rc < lnb->len) {
- /* Likewise with a partial read */
- break;
- }
- }
- }
-
- while ((cmd & OBD_BRW_READ) && lnb-- > res) {
- rc = filter_finish_page_read(lnb);
- if (rc) {
- CERROR("error on page %u@"LPU64": rc = %d\n",
- lnb->len, lnb->offset, rc);
- f_dput(lnb->dentry);
- GOTO(out_pages, rc);
- }
+ if (keylen == strlen("blocksize") &&
+ memcmp(key, "blocksize", keylen) == 0) {
+ __u32 *blocksize = val;
+ *vallen = sizeof(*blocksize);
+ *blocksize = obd->u.filter.fo_sb->s_blocksize;
+ RETURN(0);
}
- EXIT;
-out:
- OBD_FREE(fso, objcount * sizeof(*fso));
- current->journal_info = NULL;
- pop_ctxt(&saved, &obd->u.filter.fo_ctxt, NULL);
- return rc;
-out_pages:
- while (lnb-- > res) {
- if (cmd & OBD_BRW_WRITE) {
- filter_commit_write(lnb, rc);
- up(&lnb->dentry->d_inode->i_sem);
- } else {
- lustre_put_page(lnb->page);
- }
- f_dput(lnb->dentry);
- }
- if (cmd & OBD_BRW_WRITE) {
- filter_finish_transno(export, *desc_private, oti, rc);
- fsfilt_commit(obd,
- filter_parent(obd,S_IFREG,obj->ioo_id)->d_inode,
- *desc_private, 0);
+ if (keylen == strlen("blocksize_bits") &&
+ memcmp(key, "blocksize_bits", keylen) == 0) {
+ __u32 *blocksize_bits = val;
+ *vallen = sizeof(*blocksize_bits);
+ *blocksize_bits = obd->u.filter.fo_sb->s_blocksize_bits;
+ RETURN(0);
}
- goto out; /* dropped the dentry refs already (one per page) */
-out_objinfo:
- for (i = 0; i < objcount && fso[i].fso_dentry; i++) {
- if (cmd & OBD_BRW_WRITE)
- up(&fso[i].fso_dentry->d_inode->i_sem);
- f_dput(fso[i].fso_dentry);
+ 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);
+ RETURN(0);
}
- goto out;
+ CDEBUG(D_IOCTL, "invalid key\n");
+ RETURN(-EINVAL);
}
-static int filter_write_locked_page(struct niobuf_local *lnb)
+static int filter_set_info(struct obd_export *exp, __u32 keylen,
+ void *key, __u32 vallen, void *val)
{
- struct page *lpage;
- void *lpage_addr;
- void *lnb_addr;
- int rc;
+ struct obd_device *obd;
+ struct lustre_handle conn;
+ struct llog_ctxt *ctxt;
+ int rc = 0;
ENTRY;
- lpage = lustre_get_page_write(lnb->dentry->d_inode, lnb->page->index);
- if (IS_ERR(lpage)) {
- /* It is highly unlikely that we would ever get an error here.
- * The page we want to get was previously locked, so it had to
- * have already allocated the space, and we were just writing
- * over the same data, so there would be no hole in the file.
- *
- * XXX: possibility of a race with truncate could exist, need
- * to check that. There are no guarantees w.r.t.
- * write order even on a local filesystem, although the
- * normal response would be to return the number of bytes
- * successfully written and leave the rest to the app.
- */
- rc = PTR_ERR(lpage);
- CERROR("error getting locked page index %ld: rc = %d\n",
- lnb->page->index, rc);
- LBUG();
- lustre_commit_write(lnb);
- RETURN(rc);
- }
-
- /* 2 kmaps == vanishingly small deadlock opportunity */
- lpage_addr = kmap(lpage);
- lnb_addr = kmap(lnb->page);
+ conn.cookie = exp->exp_handle.h_cookie;
- memcpy(lpage_addr, lnb_addr, PAGE_SIZE);
-
- kunmap(lnb->page);
- kunmap(lpage);
-
- lustre_put_page(lnb->page);
+ obd = exp->exp_obd;
+ if (obd == NULL) {
+ CDEBUG(D_IOCTL, "invalid exp %p cookie "LPX64"\n",
+ exp, conn.cookie);
+ RETURN(-EINVAL);
+ }
- lnb->page = lpage;
- rc = lustre_commit_write(lnb);
- if (rc)
- CERROR("error committing locked page %ld: rc = %d\n",
- lnb->page->index, rc);
+ if (keylen < strlen("mds_conn") ||
+ memcmp(key, "mds_conn", keylen) != 0)
+ RETURN(-EINVAL);
+ CWARN("Received MDS connection ("LPX64")\n", conn.cookie);
+ memcpy(&obd->u.filter.fo_mdc_conn, &conn, sizeof(conn));
+ ctxt = llog_get_context(obd, LLOG_UNLINK_REPL_CTXT);
+ rc = llog_receptor_accept(ctxt, exp->exp_imp_reverse);
RETURN(rc);
}
-static int filter_syncfs(struct obd_export *exp)
+int filter_iocontrol(unsigned int cmd, struct obd_export *exp,
+ int len, void *karg, void *uarg)
{
struct obd_device *obd = exp->exp_obd;
- ENTRY;
-
- XPROCFS_BUMP_MYCPU_IOSTAT (st_syncfs_reqs, 1);
-
- RETURN(fsfilt_sync(obd, obd->u.filter.fo_sb));
-}
-
-static int filter_commitrw(int cmd, struct obd_export *export,
- int objcount, struct obd_ioobj *obj,
- int niocount, struct niobuf_local *res,
- void *desc_private, struct obd_trans_info *oti)
-{
- struct obd_run_ctxt saved;
- struct obd_ioobj *o;
- struct niobuf_local *lnb;
- struct obd_device *obd = export->exp_obd;
- int found_locked = 0, rc = 0, i;
- ENTRY;
-
- push_ctxt(&saved, &obd->u.filter.fo_ctxt, NULL);
-
- LASSERT(!current->journal_info);
- current->journal_info = desc_private;
-
- for (i = 0, o = obj, lnb = res; i < objcount; i++, o++) {
- int j;
+ struct obd_ioctl_data *data = karg;
+ int rc = 0;
- if (cmd & OBD_BRW_WRITE) {
- inode_update_time(lnb->dentry->d_inode, 1);
- up(&lnb->dentry->d_inode->i_sem);
- }
- for (j = 0 ; j < o->ioo_bufcnt ; j++, lnb++) {
- if (lnb->page == NULL) {
- continue;
- }
- if (lnb->flags & N_LOCAL_TEMP_PAGE) {
- found_locked++;
- continue;
- }
+ switch (cmd) {
+ case OBD_IOC_ABORT_RECOVERY:
+ CERROR("aborting recovery for device %s\n", obd->obd_name);
+ target_abort_recovery(obd);
+ RETURN(0);
- if (cmd & OBD_BRW_WRITE) {
- int err = filter_commit_write(lnb, 0);
+ case OBD_IOC_SET_READONLY: {
+ void *handle;
+ struct super_block *sb = obd->u.filter.fo_sb;
+ struct inode *inode = sb->s_root->d_inode;
+ BDEVNAME_DECLARE_STORAGE(tmp);
+ CERROR("setting device %s read-only\n",
+ ll_bdevname(sb, tmp));
- if (!rc)
- rc = err;
- } else {
- lustre_put_page(lnb->page);
- }
+ handle = fsfilt_start(obd, inode, FSFILT_OP_MKNOD, NULL);
+ LASSERT(handle);
+ (void)fsfilt_commit(obd, inode, handle, 1);
- f_dput(lnb->dentry);
- }
+ dev_set_rdonly(ll_sbdev(obd->u.filter.fo_sb), 2);
+ RETURN(0);
}
- for (i = 0, o = obj, lnb = res; found_locked > 0 && i < objcount;
- i++, o++) {
- int j;
- for (j = 0 ; j < o->ioo_bufcnt ; j++, lnb++) {
- int err;
- if (!(lnb->flags & N_LOCAL_TEMP_PAGE))
- continue;
-
- err = filter_write_locked_page(lnb);
- if (!rc)
- rc = err;
- f_dput(lnb->dentry);
- found_locked--;
- }
+ case OBD_IOC_CATLOGLIST: {
+ rc = llog_catlog_list(obd, 1, data);
+ RETURN(rc);
}
- if (cmd & OBD_BRW_WRITE) {
- /* We just want any dentry for the commit, for now */
- struct dentry *dir_dentry = filter_parent(obd, S_IFREG, 0);
- int err;
+ case OBD_IOC_LLOG_CANCEL:
+ 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;
- rc = filter_finish_transno(export, desc_private, oti, rc);
- err = fsfilt_commit(obd, dir_dentry->d_inode, desc_private,
- obd_sync_filter);
- if (err)
- rc = err;
- if (obd_sync_filter)
- LASSERT(oti->oti_transno <= obd->obd_last_committed);
+ push_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_ctxt, NULL);
+ rc = llog_ioctl(ctxt, cmd, data);
+ pop_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_ctxt, NULL);
+ RETURN(rc);
+*/
}
- LASSERT(!current->journal_info);
- pop_ctxt(&saved, &obd->u.filter.fo_ctxt, NULL);
- RETURN(rc);
-}
-
-static int filter_brw(int cmd, struct lustre_handle *conn,
- struct lov_stripe_md *lsm, obd_count oa_bufs,
- struct brw_page *pga, struct obd_trans_info *oti)
-{
- struct obd_export *export = class_conn2export(conn);
- struct obd_ioobj ioo;
- struct niobuf_local *lnb;
- struct niobuf_remote *rnb;
- obd_count i;
- void *desc_private;
- int ret = 0;
- ENTRY;
-
- if (export == NULL)
+ default:
RETURN(-EINVAL);
-
- OBD_ALLOC(lnb, oa_bufs * sizeof(struct niobuf_local));
- OBD_ALLOC(rnb, oa_bufs * sizeof(struct niobuf_remote));
-
- if (lnb == NULL || rnb == NULL)
- GOTO(out, ret = -ENOMEM);
-
- for (i = 0; i < oa_bufs; i++) {
- rnb[i].offset = pga[i].off;
- rnb[i].len = pga[i].count;
- }
-
- ioo.ioo_id = lsm->lsm_object_id;
- ioo.ioo_gr = 0;
- ioo.ioo_type = S_IFREG;
- ioo.ioo_bufcnt = oa_bufs;
-
- ret = filter_preprw(cmd, export, 1, &ioo, oa_bufs, rnb, lnb,
- &desc_private, oti);
- if (ret != 0)
- GOTO(out, ret);
-
- for (i = 0; i < oa_bufs; i++) {
- void *virt = kmap(pga[i].pg);
- obd_off off = pga[i].off & ~PAGE_MASK;
- void *addr = kmap(lnb[i].page);
-
- /* 2 kmaps == vanishingly small deadlock opportunity */
-
- if (cmd & OBD_BRW_WRITE)
- memcpy(addr + off, virt + off, pga[i].count);
- else
- memcpy(virt + off, addr + off, pga[i].count);
-
- kunmap(addr);
- kunmap(virt);
}
-
- ret = filter_commitrw(cmd, export, 1, &ioo, oa_bufs, lnb, desc_private,
- oti);
-
-out:
- if (lnb)
- OBD_FREE(lnb, oa_bufs * sizeof(struct niobuf_local));
- if (rnb)
- OBD_FREE(rnb, oa_bufs * sizeof(struct niobuf_remote));
- class_export_put(export);
- RETURN(ret);
+ RETURN(0);
}
-static int filter_san_preprw(int cmd, struct lustre_handle *conn,
- int objcount, struct obd_ioobj *obj,
- int niocount, struct niobuf_remote *nb)
+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
+};
+
+static int filter_llog_init(struct obd_device *obd, struct obd_device *tgt,
+ int count, struct llog_catid *logid)
{
- struct obd_device *obd;
- struct obd_ioobj *o = obj;
- struct niobuf_remote *rnb = nb;
- int rc = 0;
- int i;
+ struct llog_ctxt *ctxt;
+ int rc;
ENTRY;
- if ((cmd & OBD_BRW_WRITE) != 0)
- XPROCFS_BUMP_MYCPU_IOSTAT (st_write_reqs, 1);
- else
- XPROCFS_BUMP_MYCPU_IOSTAT (st_read_reqs, 1);
-
- obd = class_conn2obd(conn);
- if (!obd) {
- CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
- conn->cookie);
- RETURN(-EINVAL);
- }
-
- for (i = 0; i < objcount; i++, o++) {
- struct dentry *dentry;
- struct inode *inode;
- int (*fs_bmap)(struct address_space *, long);
- int j;
-
- dentry = filter_fid2dentry(obd, filter_parent(obd, S_IFREG,
- o->ioo_id),
- o->ioo_id, 0);
- if (IS_ERR(dentry))
- GOTO(out, rc = PTR_ERR(dentry));
- inode = dentry->d_inode;
- if (!inode) {
- CERROR("trying to BRW to non-existent file "LPU64"\n",
- o->ioo_id);
- f_dput(dentry);
- GOTO(out, rc = -ENOENT);
- }
- fs_bmap = inode->i_mapping->a_ops->bmap;
-
- for (j = 0; j < o->ioo_bufcnt; j++, rnb++) {
- long block;
+ 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;
- block = rnb->offset >> inode->i_blkbits;
-
- if (cmd == OBD_BRW_READ) {
- block = fs_bmap(inode->i_mapping, block);
- } else {
- loff_t newsize = rnb->offset + rnb->len;
- /* fs_prep_san_write will also update inode
- * size for us:
- * (1) new alloced block
- * (2) existed block but size extented
- */
- /* FIXME We could call fs_prep_san_write()
- * only once for all the blocks allocation.
- * Now call it once for each block, for
- * simplicity. And if error happens, we
- * probably need to release previous alloced
- * block */
- rc = fs_prep_san_write(obd, inode, &block,
- 1, newsize);
- if (rc)
- break;
- }
+ rc = llog_setup(obd, 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_proc_cb = filter_recov_log_unlink_cb;
- rnb->offset = block;
- }
- f_dput(dentry);
- }
-out:
+ rc = llog_setup(obd, LLOG_SIZE_ORIG_CTXT, tgt, 0, NULL,
+ &filter_size_orig_logops);
RETURN(rc);
}
-static int filter_statfs(struct lustre_handle *conn, struct obd_statfs *osfs)
+static int filter_llog_finish(struct obd_device *obd, int count)
{
- struct obd_device *obd;
+ int rc;
ENTRY;
- obd = class_conn2obd(conn);
-
- XPROCFS_BUMP_MYCPU_IOSTAT (st_statfs_reqs, 1);
+ rc = llog_cleanup(llog_get_context(obd, LLOG_UNLINK_REPL_CTXT));
+ if (rc)
+ RETURN(rc);
- RETURN(fsfilt_statfs(obd, obd->u.filter.fo_sb, osfs));
+ rc = llog_cleanup(llog_get_context(obd, LLOG_SIZE_ORIG_CTXT));
+ RETURN(rc);
}
-static int filter_get_info(struct lustre_handle *conn, __u32 keylen,
- void *key, __u32 *vallen, void *val)
+static struct dentry *filter_lvfs_fid2dentry(__u64 id, __u32 gen, __u64 gr,
+ void *data)
{
- struct obd_device *obd;
- ENTRY;
-
- obd = class_conn2obd(conn);
- if (!obd) {
- CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
- conn->cookie);
- RETURN(-EINVAL);
- }
-
- if (keylen == strlen("blocksize") &&
- memcmp(key, "blocksize", keylen) == 0) {
- __u32 *blocksize = val;
- *vallen = sizeof(*blocksize);
- *blocksize = obd->u.filter.fo_sb->s_blocksize;
- RETURN(0);
- }
-
- if (keylen == strlen("blocksize_bits") &&
- memcmp(key, "blocksize_bits", keylen) == 0) {
- __u32 *blocksize_bits = val;
- *vallen = sizeof(*blocksize_bits);
- *blocksize_bits = obd->u.filter.fo_sb->s_blocksize_bits;
- RETURN(0);
- }
-
- CDEBUG(D_IOCTL, "invalid key\n");
- RETURN(-EINVAL);
+ return filter_fid2dentry(data, NULL, gr, id);
}
-int filter_copy_data(struct lustre_handle *dst_conn, struct obdo *dst,
- struct lustre_handle *src_conn, struct obdo *src,
- obd_size count, obd_off offset, struct obd_trans_info *oti)
-{
- struct page *page;
- struct lov_stripe_md srcmd, dstmd;
- unsigned long index = 0;
- int err = 0;
-
- LBUG(); /* THIS CODE IS NOT CORRECT -phil */
-
- memset(&srcmd, 0, sizeof(srcmd));
- memset(&dstmd, 0, sizeof(dstmd));
- srcmd.lsm_object_id = src->o_id;
- dstmd.lsm_object_id = dst->o_id;
-
- ENTRY;
- CDEBUG(D_INFO, "src: ino "LPU64" blocks "LPU64", size "LPU64
- ", dst: ino "LPU64"\n",
- src->o_id, src->o_blocks, src->o_size, dst->o_id);
- page = alloc_page(GFP_USER);
- if (page == NULL)
- RETURN(-ENOMEM);
-
- wait_on_page(page);
-
- /* XXX with brw vector I/O, we could batch up reads and writes here,
- * all we need to do is allocate multiple pages to handle the I/Os
- * and arrays to handle the request parameters.
- */
- while (index < ((src->o_size + PAGE_SIZE - 1) >> PAGE_SHIFT)) {
- struct brw_page pg;
-
- pg.pg = page;
- pg.count = PAGE_SIZE;
- pg.off = (page->index) << PAGE_SHIFT;
- pg.flag = 0;
-
- page->index = index;
- err = obd_brw(OBD_BRW_READ, src_conn, &srcmd, 1, &pg, NULL);
- if (err) {
- EXIT;
- break;
- }
-
- pg.flag = OBD_BRW_CREATE;
- CDEBUG(D_INFO, "Read page %ld ...\n", page->index);
-
- err = obd_brw(OBD_BRW_WRITE, dst_conn, &dstmd, 1, &pg, oti);
-
- /* XXX should handle dst->o_size, dst->o_blocks here */
- if (err) {
- EXIT;
- break;
- }
-
- CDEBUG(D_INFO, "Wrote page %ld ...\n", page->index);
-
- index++;
- }
- dst->o_size = src->o_size;
- dst->o_blocks = src->o_blocks;
- dst->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
- unlock_page(page);
- __free_page(page);
-
- RETURN(err);
-}
+static struct lvfs_callback_ops filter_lvfs_ops = {
+ l_fid2dentry: filter_lvfs_fid2dentry,
+};
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_precleanup: filter_precleanup,
o_cleanup: filter_cleanup,
o_connect: filter_connect,
o_disconnect: filter_disconnect,
o_statfs: filter_statfs,
- o_syncfs: filter_syncfs,
o_getattr: filter_getattr,
+ o_unpackmd: filter_unpackmd,
o_create: filter_create,
o_setattr: filter_setattr,
o_destroy: filter_destroy,
- o_open: filter_open,
- o_close: filter_close,
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,
-#if 0
- o_san_preprw: filter_san_preprw,
- o_preallocate: filter_preallocate_inodes,
- o_migrate: filter_migrate,
- o_copy: filter_copy_data,
- o_iterate: filter_iterate
-#endif
+ o_llog_init: filter_llog_init,
+ o_llog_finish: filter_llog_finish,
+ o_iocontrol: filter_iocontrol,
};
static struct obd_ops filter_sanobd_ops = {
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_open: filter_open,
- o_close: filter_close,
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
-#if 0
- o_preallocate: filter_preallocate_inodes,
- o_migrate: filter_migrate,
- o_copy: filter_copy_data,
- o_iterate: filter_iterate
-#endif
+ o_destroy_export: filter_destroy_export,
+ o_llog_init: filter_llog_init,
+ o_llog_finish: filter_llog_finish,
+ o_iocontrol: filter_iocontrol,
};
-
static int __init obdfilter_init(void)
{
struct lprocfs_static_vars lvars;
int rc;
- printk(KERN_INFO "Lustre Filtering OBD driver; info@clusterfs.com\n");
+ printk(KERN_INFO "Lustre: Filtering OBD driver; info@clusterfs.com\n");
- xprocfs_init ("filter");
- lprocfs_init_vars(&lvars);
+ lprocfs_init_vars(filter, &lvars);
rc = class_register_type(&filter_obd_ops, lvars.module_vars,
OBD_FILTER_DEVICENAME);
{
class_unregister_type(OBD_FILTER_SAN_DEVICENAME);
class_unregister_type(OBD_FILTER_DEVICENAME);
- xprocfs_fini ();
}
MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");