Whamcloud - gitweb
LU-15481 llog: Add LLOG_SKIP_PLAIN to skip llog plain
[fs/lustre-release.git] / lustre / obdclass / llog.c
index c9f9616..1af4ae5 100644 (file)
@@ -27,7 +27,6 @@
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
- * Lustre is a trademark of Sun Microsystems, Inc.
  *
  * lustre/obdclass/llog.c
  *
@@ -50,6 +49,7 @@
 #include <obd_support.h>
 #include <obd_class.h>
 #include "llog_internal.h"
+
 /*
  * Allocate a new log or catalog handle
  * Used inside llog_open().
@@ -103,7 +103,7 @@ int llog_handle_put(const struct lu_env *env, struct llog_handle *loghandle)
        int rc = 0;
 
        if (refcount_dec_and_test(&loghandle->lgh_refcount)) {
-               struct llog_operations *lop;
+               const struct llog_operations *lop;
 
                rc = llog_handle2ops(loghandle, &lop);
                if (!rc) {
@@ -121,7 +121,7 @@ static int llog_declare_destroy(const struct lu_env *env,
                                struct llog_handle *handle,
                                struct thandle *th)
 {
-       struct llog_operations *lop;
+       const struct llog_operations *lop;
        int rc;
 
        ENTRY;
@@ -140,7 +140,7 @@ static int llog_declare_destroy(const struct lu_env *env,
 int llog_trans_destroy(const struct lu_env *env, struct llog_handle *handle,
                       struct thandle *th)
 {
-       struct llog_operations  *lop;
+       const struct llog_operations *lop;
        int rc;
        ENTRY;
 
@@ -161,9 +161,9 @@ int llog_trans_destroy(const struct lu_env *env, struct llog_handle *handle,
 
 int llog_destroy(const struct lu_env *env, struct llog_handle *handle)
 {
-       struct llog_operations  *lop;
-       struct dt_device        *dt;
-       struct thandle          *th;
+       const struct llog_operations *lop;
+       struct dt_device *dt;
+       struct thandle *th;
        int rc;
 
        ENTRY;
@@ -304,7 +304,7 @@ int llog_cancel_arr_rec(const struct lu_env *env, struct llog_handle *loghandle,
                         * be accessed anymore, let's return 0 for now, and
                         * the orphan will be handled by LFSCK. */
                        CERROR("%s: can't destroy empty llog "DFID": rc = %d\n",
-                              loghandle->lgh_ctxt->loc_obd->obd_name,
+                              loghandle2name(loghandle),
                               PFID(&loghandle->lgh_id.lgl_oi.oi_fid), rc);
                        GOTO(out_unlock, rc = 0);
                }
@@ -312,13 +312,22 @@ int llog_cancel_arr_rec(const struct lu_env *env, struct llog_handle *loghandle,
        }
 
 out_unlock:
+       if (rc < 0) {
+               /* restore bitmap while holding a mutex */
+               if (subtract_count) {
+                       loghandle->lgh_hdr->llh_count += num;
+                       subtract_count = false;
+               }
+               for (i = i - 1; i >= 0; i--)
+                       set_bit_le(index[i], LLOG_HDR_BITMAP(llh));
+       }
        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) {
+       if (rc1 < 0) {
                mutex_lock(&loghandle->lgh_hdr_mutex);
                if (subtract_count)
                        loghandle->lgh_hdr->llh_count += num;
@@ -338,7 +347,7 @@ int llog_cancel_rec(const struct lu_env *env, struct llog_handle *loghandle,
 int llog_read_header(const struct lu_env *env, struct llog_handle *handle,
                     const struct obd_uuid *uuid)
 {
-       struct llog_operations *lop;
+       const struct llog_operations *lop;
        int rc;
        ENTRY;
 
@@ -407,7 +416,7 @@ int llog_init_handle(const struct lu_env *env, struct llog_handle *handle,
                             (llh->llh_flags & LLOG_F_IS_CAT &&
                              flags & LLOG_F_IS_PLAIN))) {
                        CERROR("%s: llog type is %s but initializing %s\n",
-                              handle->lgh_ctxt->loc_obd->obd_name,
+                              loghandle2name(handle),
                               llh->llh_flags & LLOG_F_IS_CAT ?
                               "catalog" : "plain",
                               flags & LLOG_F_IS_CAT ? "catalog" : "plain");
@@ -427,7 +436,7 @@ int llog_init_handle(const struct lu_env *env, struct llog_handle *handle,
                if (unlikely(uuid &&
                             !obd_uuid_equals(uuid, &llh->llh_tgtuuid))) {
                        CERROR("%s: llog uuid mismatch: %s/%s\n",
-                              handle->lgh_ctxt->loc_obd->obd_name,
+                              loghandle2name(handle),
                               (char *)uuid->uuid,
                               (char *)llh->llh_tgtuuid.uuid);
                        GOTO(out, rc = -EEXIST);
@@ -440,8 +449,8 @@ int llog_init_handle(const struct lu_env *env, struct llog_handle *handle,
                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,
-                      flags, LLOG_F_IS_CAT, LLOG_F_IS_PLAIN);
+                      loghandle2name(handle), flags, LLOG_F_IS_CAT,
+                      LLOG_F_IS_PLAIN);
                rc = -EINVAL;
        }
        llh->llh_flags |= fmt;
@@ -454,6 +463,39 @@ out:
 }
 EXPORT_SYMBOL(llog_init_handle);
 
+int llog_verify_record(const struct llog_handle *llh, struct llog_rec_hdr *rec)
+{
+       int chunk_size = llh->lgh_hdr->llh_hdr.lrh_len;
+
+       if (rec->lrh_len == 0 || rec->lrh_len > chunk_size) {
+               CERROR("%s: record is too large: %d > %d\n",
+                      loghandle2name(llh), rec->lrh_len, chunk_size);
+               return -EINVAL;
+       }
+       if (rec->lrh_index >= LLOG_HDR_BITMAP_SIZE(llh->lgh_hdr)) {
+               CERROR("%s: index is too high: %d\n",
+                      loghandle2name(llh), rec->lrh_index);
+               return -EINVAL;
+       }
+       if ((rec->lrh_type & LLOG_OP_MASK) != LLOG_OP_MAGIC) {
+               CERROR("%s: magic %x is bad\n",
+                      loghandle2name(llh), rec->lrh_type);
+               return -EINVAL;
+       }
+
+       return 0;
+}
+EXPORT_SYMBOL(llog_verify_record);
+
+static inline bool llog_is_index_skipable(int idx, struct llog_log_hdr *llh,
+                                         struct llog_process_cat_data *cd)
+{
+       if (cd && (cd->lpcd_read_mode & LLOG_READ_MODE_RAW))
+               return false;
+
+       return !test_bit_le(idx, LLOG_HDR_BITMAP(llh));
+}
+
 static int llog_process_thread(void *arg)
 {
        struct llog_process_info        *lpi = arg;
@@ -468,6 +510,7 @@ static int llog_process_thread(void *arg)
        int                              saved_index = 0;
        int                              last_called_index = 0;
        bool                             repeated = false;
+       bool                            refresh_idx = false;
 
        ENTRY;
 
@@ -490,8 +533,10 @@ static int llog_process_thread(void *arg)
                last_called_index = cd->lpcd_first_idx;
                index = cd->lpcd_first_idx + 1;
        }
-       if (cd != NULL && cd->lpcd_last_idx)
+       if (cd && cd->lpcd_last_idx)
                last_index = cd->lpcd_last_idx;
+       else if (cd && (cd->lpcd_read_mode & LLOG_READ_MODE_RAW))
+               last_index = loghandle->lgh_last_idx;
        else
                last_index = LLOG_HDR_BITMAP_SIZE(llh) - 1;
 
@@ -499,13 +544,12 @@ static int llog_process_thread(void *arg)
                struct llog_rec_hdr *rec;
                off_t chunk_offset = 0;
                unsigned int buf_offset = 0;
-               bool partial_chunk;
                int     lh_last_idx;
                int     synced_idx = 0;
 
                /* skip records not set in bitmap */
                while (index <= last_index &&
-                      !test_bit_le(index, LLOG_HDR_BITMAP(llh)))
+                      llog_is_index_skipable(index, llh, cd))
                        ++index;
 
                /* There are no indices prior the last_index */
@@ -527,6 +571,8 @@ repeat:
                        CDEBUG(D_OTHER, "cur_offset %llu, chunk_offset %llu,"
                               " buf_offset %u, rc = %d\n", cur_offset,
                               (__u64)chunk_offset, buf_offset, rc);
+               if (rc == -ESTALE)
+                       GOTO(out, rc = 0);
                /* we`ve tried to reread the chunk, but there is no
                 * new records */
                if (rc == -EIO && repeated && (chunk_offset + buf_offset) ==
@@ -540,13 +586,10 @@ repeat:
                 * The absolute offset of the current chunk is calculated
                 * from cur_offset value and stored in chunk_offset variable.
                 */
-               if ((cur_offset & (chunk_size - 1)) != 0) {
-                       partial_chunk = true;
+               if ((cur_offset & (chunk_size - 1)) != 0)
                        chunk_offset = cur_offset & ~(chunk_size - 1);
-               } else {
-                       partial_chunk = false;
+               else
                        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
@@ -615,15 +658,21 @@ repeat:
 
                        repeated = false;
 
-                       if (rec->lrh_len == 0 || rec->lrh_len > chunk_size) {
-                               CWARN("%s: invalid length %d in llog "DFID
-                                     "record for index %d/%d\n",
-                                      loghandle->lgh_ctxt->loc_obd->obd_name,
-                                      rec->lrh_len,
+                       rc = llog_verify_record(loghandle, rec);
+                       if (rc) {
+                               CERROR("%s: invalid record in llog "DFID
+                                      " record for index %d/%d: rc = %d\n",
+                                      loghandle2name(loghandle),
                                       PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
-                                      rec->lrh_index, index);
-
-                               GOTO(out, rc = -EINVAL);
+                                      rec->lrh_index, index, rc);
+                               /*
+                                * the block seem to be corrupted, let's try
+                                * with the next one. reset rc to go to the
+                                * next chunk.
+                                */
+                               refresh_idx = true;
+                               index = 0;
+                               GOTO(repeat, rc = 0);
                        }
 
                        if (rec->lrh_index < index) {
@@ -632,13 +681,15 @@ repeat:
                                continue;
                        }
 
-                       if (rec->lrh_index != index) {
-                               CERROR("%s: "DFID" Invalid record: index %u"
-                                      " but expected %u\n",
-                                      loghandle->lgh_ctxt->loc_obd->obd_name,
+                       if (rec->lrh_index > index) {
+                               /* the record itself looks good, but we met a
+                                * gap which can be result of old bugs, just
+                                * keep going */
+                               CERROR("%s: "DFID" index %u, expected %u\n",
+                                      loghandle2name(loghandle),
                                       PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
                                       rec->lrh_index, index);
-                               GOTO(out, rc = -ERANGE);
+                               index = rec->lrh_index;
                        }
 
                        CDEBUG(D_OTHER,
@@ -650,8 +701,8 @@ repeat:
                        loghandle->lgh_cur_offset = (char *)rec - (char *)buf +
                                                    chunk_offset;
 
-                       /* if set, process the callback on this record */
-                       if (test_bit_le(index, LLOG_HDR_BITMAP(llh))) {
+                       /* if needed, process the callback on this record */
+                       if (!llog_is_index_skipable(index, llh, cd)) {
                                struct llog_cookie *lgc;
                                __u64   tmp_off;
                                int     tmp_idx;
@@ -684,12 +735,19 @@ repeat:
                                        lgc->lgc_index = tmp_idx;
                                }
 
-                               if (rc == LLOG_PROC_BREAK) {
+                               if (rc == LLOG_PROC_BREAK ||
+                                   rc == LLOG_SKIP_PLAIN) {
                                        GOTO(out, rc);
                                } else if (rc == LLOG_DEL_RECORD) {
                                        rc = llog_cancel_rec(lpi->lpi_env,
                                                             loghandle,
                                                             rec->lrh_index);
+                                       /* Allow parallel cancelling, ENOENT
+                                        * means record was canceled at another
+                                        * processing thread or callback
+                                        */
+                                       if (rc == -ENOENT)
+                                               rc = 0;
                                }
                                if (rc)
                                        GOTO(out, rc);
@@ -724,7 +782,7 @@ out:
                         * 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);
+                              loghandle2name(loghandle));
                        rc = -EAGAIN;
                } else {
                        /* something bad happened to the processing of a local
@@ -733,7 +791,7 @@ out:
                         * discard any remaining bits in the header */
                        CERROR("%s: Local llog found corrupted #"DOSTID":%x"
                               " %s index %d count %d\n",
-                              loghandle->lgh_ctxt->loc_obd->obd_name,
+                              loghandle2name(loghandle),
                               POSTID(&loghandle->lgh_id.lgl_oi),
                               loghandle->lgh_id.lgl_ogen,
                               ((llh->llh_flags & LLOG_F_IS_CAT) ? "catalog" :
@@ -778,8 +836,6 @@ static int llog_process_thread_daemonize(void *arg)
        }
        task_unlock(lpi->lpi_reftask);
 
-       unshare_fs_struct();
-
        /* client env has no keys, tags is just 0 */
        rc = lu_env_init(&env, LCT_LOCAL | LCT_MG_THREAD);
        if (rc)
@@ -801,7 +857,8 @@ int llog_process_or_fork(const struct lu_env *env,
        struct llog_process_info *lpi;
        struct llog_process_data *d = data;
        struct llog_process_cat_data *cd = catdata;
-       int                      rc;
+       __u32 flags = loghandle->lgh_hdr->llh_flags;
+       int rc;
 
        ENTRY;
 
@@ -815,11 +872,12 @@ int llog_process_or_fork(const struct lu_env *env,
        lpi->lpi_cbdata    = data;
        lpi->lpi_catdata   = catdata;
 
-       CDEBUG(D_OTHER, "Processing "DFID" flags 0x%03x startcat %d startidx %d first_idx %d last_idx %d\n",
-              PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
-              loghandle->lgh_hdr->llh_flags, d ? d->lpd_startcat : -1,
-              d ? d->lpd_startidx : -1, cd ? cd->lpcd_first_idx : -1,
-              cd ? cd->lpcd_last_idx : -1);
+       CDEBUG(D_OTHER, "Processing "DFID" flags 0x%03x startcat %d startidx %d first_idx %d last_idx %d read_mode %d\n",
+              PFID(&loghandle->lgh_id.lgl_oi.oi_fid), flags,
+              (flags & LLOG_F_IS_CAT) && d ? d->lpd_startcat : -1,
+              (flags & LLOG_F_IS_CAT) && d ? d->lpd_startidx : -1,
+              cd ? cd->lpcd_first_idx : -1, cd ? cd->lpcd_last_idx : -1,
+              cd ? cd->lpcd_read_mode : -1);
        if (fork) {
                struct task_struct *task;
 
@@ -836,7 +894,7 @@ int llog_process_or_fork(const struct lu_env *env,
                if (IS_ERR(task)) {
                        rc = PTR_ERR(task);
                        CERROR("%s: cannot start thread: rc = %d\n",
-                              loghandle->lgh_ctxt->loc_obd->obd_name, rc);
+                              loghandle2name(loghandle), rc);
                        GOTO(out_lpi, rc);
                }
                wait_for_completion(&lpi->lpi_completion);
@@ -910,7 +968,7 @@ int llog_reverse_process(const struct lu_env *env,
 
                /* skip records not set in bitmap */
                while (index >= first_index &&
-                      !test_bit_le(index, LLOG_HDR_BITMAP(llh)))
+                      llog_is_index_skipable(index, llh, cd))
                        --index;
 
                LASSERT(index >= first_index - 1);
@@ -940,13 +998,14 @@ int llog_reverse_process(const struct lu_env *env,
                        if (tail->lrt_index == 0)
                                GOTO(out, rc = 0); /* no more records */
 
-                       /* if set, process the callback on this record */
-                       if (test_bit_le(index, LLOG_HDR_BITMAP(llh))) {
+                       /* if needed, process the callback on this record */
+                       if (!llog_is_index_skipable(index, llh, cd)) {
                                rec = (void *)tail - tail->lrt_len +
                                      sizeof(*tail);
 
                                rc = cb(env, loghandle, rec, data);
-                               if (rc == LLOG_PROC_BREAK) {
+                               if (rc == LLOG_PROC_BREAK ||
+                                   rc == LLOG_SKIP_PLAIN) {
                                        GOTO(out, rc);
                                } else if (rc == LLOG_DEL_RECORD) {
                                        rc = llog_cancel_rec(env, loghandle,
@@ -987,8 +1046,8 @@ EXPORT_SYMBOL(llog_reverse_process);
  */
 int llog_exist(struct llog_handle *loghandle)
 {
-       struct llog_operations  *lop;
-       int                      rc;
+       const struct llog_operations *lop;
+       int rc;
 
        ENTRY;
 
@@ -1007,7 +1066,7 @@ int llog_declare_create(const struct lu_env *env,
                        struct llog_handle *loghandle, struct thandle *th)
 {
        const struct cred *old_cred;
-       struct llog_operations  *lop;
+       const struct llog_operations *lop;
        int rc;
 
        ENTRY;
@@ -1028,7 +1087,7 @@ int llog_create(const struct lu_env *env, struct llog_handle *handle,
                struct thandle *th)
 {
        const struct cred *old_cred;
-       struct llog_operations  *lop;
+       const struct llog_operations *lop;
        int rc;
 
        ENTRY;
@@ -1051,7 +1110,7 @@ int llog_declare_write_rec(const struct lu_env *env,
                           struct thandle *th)
 {
        const struct cred *old_cred;
-       struct llog_operations  *lop;
+       const struct llog_operations *lop;
        int rc;
 
        ENTRY;
@@ -1074,7 +1133,7 @@ int llog_write_rec(const struct lu_env *env, struct llog_handle *handle,
                   int idx, struct thandle *th)
 {
        const struct cred *old_cred;
-       struct llog_operations  *lop;
+       const struct llog_operations *lop;
        int rc, buflen;
 
        ENTRY;
@@ -1089,12 +1148,11 @@ int llog_write_rec(const struct lu_env *env, struct llog_handle *handle,
                RETURN(-EPROTO);
        } else if (th == NULL) {
                CERROR("%s: missed transaction handle\n",
-                       handle->lgh_obj->do_lu.lo_dev->ld_obd->obd_name);
+                      loghandle2name(handle));
                RETURN(-EPROTO);
        } else if (handle->lgh_hdr == NULL) {
                CERROR("%s: loghandle %p with no header\n",
-                       handle->lgh_obj->do_lu.lo_dev->ld_obd->obd_name,
-                       handle);
+                      loghandle2name(handle), handle);
                RETURN(-EPROTO);
        }
 
@@ -1373,13 +1431,24 @@ out:
 }
 EXPORT_SYMBOL(llog_is_empty);
 
+/* this callback run in raw read mode (canceled record are processed) */
 int llog_copy_handler(const struct lu_env *env, struct llog_handle *llh,
                      struct llog_rec_hdr *rec, void *data)
 {
-       struct llog_handle      *copy_llh = data;
+       struct llog_handle *copy_llh = data;
+       int idx = rec->lrh_index;
+       int rc;
+
+       ENTRY;
 
        /* Append all records */
-       return llog_write(env, copy_llh, rec, LLOG_NEXT_IDX);
+       rc = llog_write(env, copy_llh, rec, LLOG_NEXT_IDX);
+
+       /* Cancel the record if it is canceled on the source */
+       if (!rc && !test_bit_le(idx, LLOG_HDR_BITMAP(llh->lgh_hdr)))
+               rc = llog_cancel_rec(env, copy_llh, copy_llh->lgh_last_idx);
+
+       RETURN(rc);
 }
 
 /* backup plain llog */
@@ -1387,8 +1456,9 @@ int llog_backup(const struct lu_env *env, struct obd_device *obd,
                struct llog_ctxt *ctxt, struct llog_ctxt *bctxt,
                char *name, char *backup)
 {
-       struct llog_handle      *llh, *bllh;
-       int                      rc;
+       struct llog_handle *llh, *bllh;
+       struct llog_process_cat_data cd = {0};
+       int rc;
 
        ENTRY;
 
@@ -1433,9 +1503,11 @@ int llog_backup(const struct lu_env *env, struct obd_device *obd,
        if (rc)
                GOTO(out_backup, rc);
 
+       /* Read canceled records to have an exact copy */
+       cd.lpcd_read_mode = LLOG_READ_MODE_RAW;
        /* Copy log record by record */
        rc = llog_process_or_fork(env, llh, llog_copy_handler, (void *)bllh,
-                                 NULL, false);
+                                 &cd, false);
        if (rc)
                CERROR("%s: failed to backup log %s: rc = %d\n",
                       obd->obd_name, name, rc);
@@ -1456,8 +1528,8 @@ __u64 llog_size(const struct lu_env *env, struct llog_handle *llh)
        rc = llh->lgh_obj->do_ops->do_attr_get(env, llh->lgh_obj, &la);
        if (rc) {
                CERROR("%s: attr_get failed for "DFID": rc = %d\n",
-                      llh->lgh_ctxt->loc_obd->obd_name,
-                      PFID(&llh->lgh_id.lgl_oi.oi_fid), rc);
+                      loghandle2name(llh), PFID(&llh->lgh_id.lgl_oi.oi_fid),
+                      rc);
                return 0;
        }