* Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2012, 2014, Intel Corporation.
+ * Copyright (c) 2012, 2015, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#define DEBUG_SUBSYSTEM S_LOG
#include <linux/kthread.h>
-#include <obd_class.h>
+#include <llog_swab.h>
#include <lustre_log.h>
+#include <obd_class.h>
#include "llog_internal.h"
-
/*
* Allocate a new log or catalog handle
* Used inside llog_open().
return NULL;
init_rwsem(&loghandle->lgh_lock);
- init_rwsem(&loghandle->lgh_hdr_lock);
+ mutex_init(&loghandle->lgh_hdr_mutex);
INIT_LIST_HEAD(&loghandle->u.phd.phd_entry);
atomic_set(&loghandle->lgh_refcount, 1);
llog_free_handle(loghandle);
}
+static int llog_declare_destroy(const struct lu_env *env,
+ struct llog_handle *handle,
+ struct thandle *th)
+{
+ struct llog_operations *lop;
+ int rc;
+
+ ENTRY;
+
+ rc = llog_handle2ops(handle, &lop);
+ if (rc)
+ RETURN(rc);
+ if (lop->lop_declare_destroy == NULL)
+ RETURN(-EOPNOTSUPP);
+
+ rc = lop->lop_declare_destroy(env, handle, th);
+
+ RETURN(rc);
+}
+
+int llog_trans_destroy(const struct lu_env *env, struct llog_handle *handle,
+ struct thandle *th)
+{
+ struct llog_operations *lop;
+ int rc;
+ ENTRY;
+
+ rc = llog_handle2ops(handle, &lop);
+ if (rc < 0)
+ RETURN(rc);
+ if (lop->lop_destroy == NULL)
+ RETURN(-EOPNOTSUPP);
+
+ LASSERT(handle->lgh_obj != NULL);
+ if (!dt_object_exists(handle->lgh_obj))
+ RETURN(0);
+
+ rc = lop->lop_destroy(env, handle, th);
+
+ RETURN(rc);
+}
+
+int llog_destroy(const struct lu_env *env, struct llog_handle *handle)
+{
+ struct llog_operations *lop;
+ struct dt_device *dt;
+ struct thandle *th;
+ int rc;
+
+ ENTRY;
+
+ rc = llog_handle2ops(handle, &lop);
+ if (rc < 0)
+ RETURN(rc);
+ if (lop->lop_destroy == NULL)
+ RETURN(-EOPNOTSUPP);
+
+ if (handle->lgh_obj == NULL) {
+ /* if lgh_obj == NULL, then it is from client side destroy */
+ rc = lop->lop_destroy(env, handle, NULL);
+ RETURN(rc);
+ }
+
+ if (!dt_object_exists(handle->lgh_obj))
+ RETURN(0);
+
+ dt = lu2dt_dev(handle->lgh_obj->do_lu.lo_dev);
+
+ th = dt_trans_create(env, dt);
+ if (IS_ERR(th))
+ RETURN(PTR_ERR(th));
+
+ rc = llog_declare_destroy(env, handle, th);
+ if (rc != 0)
+ GOTO(out_trans, rc);
+
+ rc = dt_trans_start_local(env, dt, th);
+ if (rc < 0)
+ GOTO(out_trans, rc);
+
+ rc = lop->lop_destroy(env, handle, th);
+
+out_trans:
+ dt_trans_stop(env, dt, th);
+
+ RETURN(rc);
+}
+EXPORT_SYMBOL(llog_destroy);
+
/* returns negative on error; 0 if success; 1 if success & log destroyed */
int llog_cancel_rec(const struct lu_env *env, struct llog_handle *loghandle,
int index)
{
- struct llog_log_hdr *llh = loghandle->lgh_hdr;
- int rc = 0;
- ENTRY;
+ struct llog_thread_info *lgi = llog_info(env);
+ struct dt_device *dt;
+ struct llog_log_hdr *llh = loghandle->lgh_hdr;
+ struct thandle *th;
+ int rc;
+ int rc1;
+ bool subtract_count = false;
- CDEBUG(D_RPCTRACE, "Canceling %d in log "DOSTID"\n",
- index, POSTID(&loghandle->lgh_id.lgl_oi));
+ ENTRY;
- if (index == 0) {
- CERROR("Can't cancel index 0 which is header\n");
- RETURN(-EINVAL);
- }
+ CDEBUG(D_RPCTRACE, "Canceling %d in log "DOSTID"\n", index,
+ POSTID(&loghandle->lgh_id.lgl_oi));
- down_write(&loghandle->lgh_hdr_lock);
+ if (index == 0) {
+ CERROR("Can't cancel index 0 which is header\n");
+ RETURN(-EINVAL);
+ }
+
+ LASSERT(loghandle != NULL);
+ LASSERT(loghandle->lgh_ctxt != NULL);
+ LASSERT(loghandle->lgh_obj != NULL);
+
+ dt = lu2dt_dev(loghandle->lgh_obj->do_lu.lo_dev);
+
+ th = dt_trans_create(env, dt);
+ if (IS_ERR(th))
+ RETURN(PTR_ERR(th));
+
+ rc = llog_declare_write_rec(env, loghandle, &llh->llh_hdr, index, th);
+ if (rc < 0)
+ GOTO(out_trans, rc);
+
+ if ((llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY)) {
+ rc = llog_declare_destroy(env, loghandle, th);
+ if (rc < 0)
+ GOTO(out_trans, rc);
+ }
+
+ th->th_wait_submit = 1;
+ rc = dt_trans_start_local(env, dt, th);
+ if (rc < 0)
+ GOTO(out_trans, rc);
+
+ down_write(&loghandle->lgh_lock);
+ /* clear bitmap */
+ mutex_lock(&loghandle->lgh_hdr_mutex);
if (!ext2_clear_bit(index, LLOG_HDR_BITMAP(llh))) {
- up_write(&loghandle->lgh_hdr_lock);
CDEBUG(D_RPCTRACE, "Catalog index %u already clear?\n", index);
- RETURN(-ENOENT);
+ GOTO(out_unlock, rc);
}
- llh->llh_count--;
+ loghandle->lgh_hdr->llh_count--;
+ subtract_count = true;
+ /* Pass this index to llog_osd_write_rec(), which will use the index
+ * to only update the necesary bitmap. */
+ lgi->lgi_cookie.lgc_index = index;
+ /* update header */
+ rc = llog_write_rec(env, loghandle, &llh->llh_hdr, &lgi->lgi_cookie,
+ LLOG_HEADER_IDX, th);
+ if (rc != 0)
+ GOTO(out_unlock, rc);
if ((llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) &&
(llh->llh_count == 1) &&
- (loghandle->lgh_last_idx == LLOG_HDR_BITMAP_SIZE(llh) - 1)) {
- up_write(&loghandle->lgh_hdr_lock);
- rc = llog_destroy(env, loghandle);
+ ((loghandle->lgh_last_idx == LLOG_HDR_BITMAP_SIZE(llh) - 1) ||
+ (loghandle->u.phd.phd_cat_handle != NULL &&
+ loghandle->u.phd.phd_cat_handle->u.chd.chd_current_log !=
+ loghandle))) {
+ /* never try to destroy it again */
+ llh->llh_flags &= ~LLOG_F_ZAP_WHEN_EMPTY;
+ rc = llog_trans_destroy(env, loghandle, th);
if (rc < 0) {
+ /* Sigh, can not destroy the final plain llog, but
+ * the bitmap has been clearly, so the record can not
+ * be accessed anymore, let's return 0 for now, and
+ * the orphan will be handled by LFSCK. */
CERROR("%s: can't destroy empty llog #"DOSTID
"#%08x: rc = %d\n",
loghandle->lgh_ctxt->loc_obd->obd_name,
POSTID(&loghandle->lgh_id.lgl_oi),
loghandle->lgh_id.lgl_ogen, rc);
- GOTO(out_err, rc);
+ GOTO(out_unlock, rc);
}
- RETURN(LLOG_DEL_PLAIN);
+ rc = LLOG_DEL_PLAIN;
}
- up_write(&loghandle->lgh_hdr_lock);
- rc = llog_write(env, loghandle, &llh->llh_hdr, LLOG_HEADER_IDX);
- if (rc < 0) {
- CERROR("%s: fail to write header for llog #"DOSTID
- "#%08x: rc = %d\n",
- loghandle->lgh_ctxt->loc_obd->obd_name,
- POSTID(&loghandle->lgh_id.lgl_oi),
- loghandle->lgh_id.lgl_ogen, rc);
- GOTO(out_err, rc);
+out_unlock:
+ mutex_unlock(&loghandle->lgh_hdr_mutex);
+ up_write(&loghandle->lgh_lock);
+out_trans:
+ rc1 = dt_trans_stop(env, dt, th);
+ if (rc == 0)
+ rc = rc1;
+ if (rc < 0 && subtract_count) {
+ mutex_lock(&loghandle->lgh_hdr_mutex);
+ loghandle->lgh_hdr->llh_count++;
+ ext2_set_bit(index, LLOG_HDR_BITMAP(llh));
+ mutex_unlock(&loghandle->lgh_hdr_mutex);
}
- RETURN(0);
-out_err:
- down_write(&loghandle->lgh_hdr_lock);
- ext2_set_bit(index, LLOG_HDR_BITMAP(llh));
- llh->llh_count++;
- up_write(&loghandle->lgh_hdr_lock);
- return rc;
+ RETURN(rc);
}
-static int llog_read_header(const struct lu_env *env,
- struct llog_handle *handle,
- struct obd_uuid *uuid)
+int llog_read_header(const struct lu_env *env, struct llog_handle *handle,
+ const struct obd_uuid *uuid)
{
struct llog_operations *lop;
int rc;
+ ENTRY;
rc = llog_handle2ops(handle, &lop);
if (rc)
/* lrh_len should be initialized in llog_init_handle */
handle->lgh_last_idx = 0; /* header is record with index 0 */
+ handle->lgh_write_offset = 0;
llh->llh_count = 1; /* for the header record */
llh->llh_hdr.lrh_type = LLOG_HDR_MAGIC;
LASSERT(handle->lgh_ctxt->loc_chunk_size >=
memcpy(&llh->llh_tgtuuid, uuid,
sizeof(llh->llh_tgtuuid));
llh->llh_bitmap_offset = offsetof(typeof(*llh), llh_bitmap);
+ /* Since update llog header might also call this function,
+ * let's reset the bitmap to 0 here */
+ memset(LLOG_HDR_BITMAP(llh), 0, llh->llh_hdr.lrh_len -
+ llh->llh_bitmap_offset -
+ sizeof(llh->llh_tail));
ext2_set_bit(0, LLOG_HDR_BITMAP(llh));
LLOG_HDR_TAIL(llh)->lrt_len = llh->llh_hdr.lrh_len;
LLOG_HDR_TAIL(llh)->lrt_index = llh->llh_hdr.lrh_index;
rc = 0;
}
- return rc;
+ RETURN(rc);
}
+EXPORT_SYMBOL(llog_read_header);
int llog_init_handle(const struct lu_env *env, struct llog_handle *handle,
int flags, struct obd_uuid *uuid)
LASSERT(list_empty(&handle->u.chd.chd_head));
INIT_LIST_HEAD(&handle->u.chd.chd_head);
llh->llh_size = sizeof(struct llog_logid_rec);
+ llh->llh_flags |= LLOG_F_IS_FIXSIZE;
} else if (!(flags & LLOG_F_IS_PLAIN)) {
CERROR("%s: unknown flags: %#x (expected %#x or %#x)\n",
handle->lgh_ctxt->loc_obd->obd_name,
struct llog_log_hdr *llh = loghandle->lgh_hdr;
struct llog_process_cat_data *cd = lpi->lpi_catdata;
char *buf;
- int chunk_size;
- __u64 cur_offset;
- __u64 last_offset;
+ size_t chunk_size;
+ __u64 cur_offset, tmp_offset;
int rc = 0, index = 1, last_index;
int saved_index = 0;
int last_called_index = 0;
RETURN(-EINVAL);
cur_offset = chunk_size = llh->llh_hdr.lrh_len;
+ /* expect chunk_size to be power of two */
+ LASSERT(is_power_of_2(chunk_size));
OBD_ALLOC_LARGE(buf, chunk_size);
if (buf == NULL) {
else
last_index = LLOG_HDR_BITMAP_SIZE(llh) - 1;
- if (index > last_index) /* Record is not in this buffer. */
- GOTO(out, rc);
-
while (rc == 0) {
struct llog_rec_hdr *rec;
+ off_t chunk_offset;
+ unsigned int buf_offset = 0;
+ bool partial_chunk;
/* skip records not set in bitmap */
while (index <= last_index &&
!ext2_test_bit(index, LLOG_HDR_BITMAP(llh)))
++index;
- LASSERT(index <= last_index + 1);
- if (index == last_index + 1)
- break;
-repeat:
- CDEBUG(D_OTHER, "index: %d last_index %d\n",
- index, last_index);
+ /* There are no indices prior the last_index */
+ if (index > last_index)
+ break;
+
+ CDEBUG(D_OTHER, "index: %d last_index %d\n", index,
+ last_index);
+repeat:
/* get the buf with our target record; avoid old garbage */
memset(buf, 0, chunk_size);
- last_offset = cur_offset;
rc = llog_next_block(lpi->lpi_env, loghandle, &saved_index,
index, &cur_offset, buf, chunk_size);
if (rc != 0)
GOTO(out, rc);
+ /* NB: after llog_next_block() call the cur_offset is the
+ * offset of the next block after read one.
+ * The absolute offset of the current chunk is calculated
+ * from cur_offset value and stored in chunk_offset variable.
+ */
+ tmp_offset = cur_offset;
+ if (do_div(tmp_offset, chunk_size) != 0) {
+ partial_chunk = true;
+ chunk_offset = cur_offset & ~(chunk_size - 1);
+ } else {
+ partial_chunk = false;
+ chunk_offset = cur_offset - chunk_size;
+ }
+
/* NB: when rec->lrh_len is accessed it is already swabbed
* since it is used at the "end" of the loop and the rec
* swabbing is done at the beginning of the loop. */
- for (rec = (struct llog_rec_hdr *)buf;
+ for (rec = (struct llog_rec_hdr *)(buf + buf_offset);
(char *)rec < buf + chunk_size;
rec = llog_rec_hdr_next(rec)) {
CDEBUG(D_OTHER, "processing rec 0x%p type %#x\n",
rec, rec->lrh_type);
- if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
+ if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
lustre_swab_llog_rec(rec);
- CDEBUG(D_OTHER, "after swabbing, type=%#x idx=%d\n",
- rec->lrh_type, rec->lrh_index);
-
- if (rec->lrh_index == 0) {
- /* probably another rec just got added? */
- if (index <= loghandle->lgh_last_idx)
- GOTO(repeat, rc = 0);
- GOTO(out, rc = 0); /* no more records */
+ CDEBUG(D_OTHER, "after swabbing, type=%#x idx=%d\n",
+ rec->lrh_type, rec->lrh_index);
+
+ /* for partial chunk the end of it is zeroed, check
+ * for index 0 to distinguish it. */
+ if (partial_chunk && rec->lrh_index == 0) {
+ /* concurrent llog_add() might add new records
+ * while llog_processing, check this is not
+ * the case and re-read the current chunk
+ * otherwise. */
+ if (index > loghandle->lgh_last_idx)
+ GOTO(out, rc = 0);
+ CDEBUG(D_OTHER, "Re-read last llog buffer for "
+ "new records, index %u, last %u\n",
+ index, loghandle->lgh_last_idx);
+ /* save offset inside buffer for the re-read */
+ buf_offset = (char *)rec - (char *)buf;
+ cur_offset = chunk_offset;
+ goto repeat;
}
+
if (rec->lrh_len == 0 || rec->lrh_len > chunk_size) {
- CWARN("invalid length %d in llog record for "
- "index %d/%d\n", rec->lrh_len,
- rec->lrh_index, index);
- GOTO(out, rc = -EINVAL);
- }
+ CWARN("invalid length %d in llog record for "
+ "index %d/%d\n", rec->lrh_len,
+ rec->lrh_index, index);
+ GOTO(out, rc = -EINVAL);
+ }
- if (rec->lrh_index < index) {
- CDEBUG(D_OTHER, "skipping lrh_index %d\n",
- rec->lrh_index);
- continue;
- }
+ if (rec->lrh_index < index) {
+ CDEBUG(D_OTHER, "skipping lrh_index %d\n",
+ rec->lrh_index);
+ continue;
+ }
+
+ if (rec->lrh_index != index) {
+ CERROR("%s: Invalid record: index %u but "
+ "expected %u\n",
+ loghandle->lgh_ctxt->loc_obd->obd_name,
+ rec->lrh_index, index);
+ GOTO(out, rc = -ERANGE);
+ }
CDEBUG(D_OTHER,
"lrh_index: %d lrh_len: %d (%d remains)\n",
rec->lrh_index, rec->lrh_len,
(int)(buf + chunk_size - (char *)rec));
- loghandle->lgh_cur_idx = rec->lrh_index;
- loghandle->lgh_cur_offset = (char *)rec - (char *)buf +
- last_offset;
+ loghandle->lgh_cur_idx = rec->lrh_index;
+ loghandle->lgh_cur_offset = (char *)rec - (char *)buf +
+ chunk_offset;
/* if set, process the callback on this record */
if (ext2_test_bit(index, LLOG_HDR_BITMAP(llh))) {
rc = llog_cancel_rec(lpi->lpi_env,
loghandle,
rec->lrh_index);
- }
- if (rc)
- GOTO(out, rc);
- } else {
- CDEBUG(D_OTHER, "Skipped index %d\n", index);
- }
-
- /* next record, still in buffer? */
- ++index;
- if (index > last_index)
- GOTO(out, rc = 0);
- }
- }
+ }
+ if (rc)
+ GOTO(out, rc);
+ }
+ /* exit if the last index is reached */
+ if (index >= last_index)
+ GOTO(out, rc = 0);
+ ++index;
+ }
+ }
out:
- if (cd != NULL)
- cd->lpcd_last_idx = last_called_index;
+ if (cd != NULL)
+ cd->lpcd_last_idx = last_called_index;
if (unlikely(rc == -EIO && loghandle->lgh_obj != NULL)) {
- /* something bad happened to the processing of a local
- * llog file, probably I/O error or the log got corrupted..
- * to be able to finally release the log we discard any
- * remaining bits in the header */
- CERROR("Local llog found corrupted\n");
- while (index <= last_index) {
- if (ext2_test_bit(index, LLOG_HDR_BITMAP(llh)) != 0)
- llog_cancel_rec(lpi->lpi_env, loghandle, index);
- index++;
+ if (dt_object_remote(loghandle->lgh_obj)) {
+ /* If it is remote object, then -EIO might means
+ * disconnection or eviction, let's return -EAGAIN,
+ * so for update recovery log processing, it will
+ * retry until the umount or abort recovery, see
+ * lod_sub_recovery_thread() */
+ CERROR("%s retry remote llog process\n",
+ loghandle->lgh_ctxt->loc_obd->obd_name);
+ rc = -EAGAIN;
+ } else {
+ /* something bad happened to the processing of a local
+ * llog file, probably I/O error or the log got
+ * corrupted to be able to finally release the log we
+ * discard any remaining bits in the header */
+ CERROR("Local llog found corrupted\n");
+ while (index <= last_index) {
+ if (ext2_test_bit(index,
+ LLOG_HDR_BITMAP(llh)) != 0)
+ llog_cancel_rec(lpi->lpi_env, loghandle,
+ index);
+ index++;
+ }
+ rc = 0;
}
- rc = 0;
}
OBD_FREE_LARGE(buf, chunk_size);
if (IS_ERR(th))
GOTO(out, rc = PTR_ERR(th));
+ /* Create update llog object synchronously, which
+ * happens during inialization process see
+ * lod_sub_prep_llog(), to make sure the update
+ * llog object is created before corss-MDT writing
+ * updates into the llog object */
+ if (ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID)
+ th->th_sync = 1;
+
th->th_wait_submit = 1;
rc = llog_declare_create(env, *res, th);
if (rc == 0) {
RETURN(rc);
}
EXPORT_SYMBOL(llog_backup);
+
+/* Get size of llog */
+__u64 llog_size(const struct lu_env *env, struct llog_handle *llh)
+{
+ int rc;
+ struct lu_attr la;
+
+ rc = llh->lgh_obj->do_ops->do_attr_get(env, llh->lgh_obj, &la);
+ if (rc) {
+ CERROR("%s: attr_get failed, rc = %d\n",
+ llh->lgh_ctxt->loc_obd->obd_name, rc);
+ return 0;
+ }
+
+ return la.la_size;
+}
+EXPORT_SYMBOL(llog_size);
+