X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fmds%2Fmds_fs.c;h=1dfc246bc1e2d78bf9617939fdf79785e4ffbb49;hb=f98de6c88bbea09960f6019c7ad92a03a2b44913;hp=77317189804c5a7b98e518d27b73917a07ee1aef;hpb=a7870d22e5b4acfc57c4b7c77abda39a98b1e886;p=fs%2Flustre-release.git diff --git a/lustre/mds/mds_fs.c b/lustre/mds/mds_fs.c index 7731718..1dfc246 100644 --- a/lustre/mds/mds_fs.c +++ b/lustre/mds/mds_fs.c @@ -91,7 +91,7 @@ int mds_client_add(struct obd_device *obd, struct mds_obd *mds, } } - CDEBUG(D_INFO, "client at index %d with UUID '%s' added\n", + CDEBUG(D_INFO, "client at idx %d with UUID '%s' added\n", cl_idx, med->med_mcd->mcd_uuid); med->med_idx = cl_idx; @@ -135,10 +135,12 @@ int mds_client_free(struct obd_export *exp, int clear_client) if (!strcmp(med->med_mcd->mcd_uuid, obd->obd_uuid.uuid)) GOTO(free_and_out, 0); - CDEBUG(D_INFO, "freeing client at index %u (%lld)with UUID '%s'\n", + CDEBUG(D_INFO, "freeing client at idx %u (%lld)with UUID '%s'\n", med->med_idx, med->med_off, med->med_mcd->mcd_uuid); - if (!test_and_clear_bit(med->med_idx, bitmap)) { + /* Clear the bit _after_ zeroing out the client so we don't + race with mds_client_add and zero out new clients.*/ + if (!test_bit(med->med_idx, bitmap)) { CERROR("MDS client %u: bit already clear in bitmap!!\n", med->med_idx); LBUG(); @@ -152,10 +154,16 @@ int mds_client_free(struct obd_export *exp, int clear_client) pop_ctxt(&saved, &obd->obd_ctxt, NULL); CDEBUG(rc == 0 ? D_INFO : D_ERROR, - "zeroing out client %s off %u in %s rc %d\n", + "zeroing out client %s idx %u in %s rc %d\n", med->med_mcd->mcd_uuid, med->med_idx, LAST_RCVD, rc); } + if (!test_and_clear_bit(med->med_idx, bitmap)) { + CERROR("MDS client %u: bit already clear in bitmap!!\n", + med->med_idx); + LBUG(); + } + free_and_out: OBD_FREE(med->med_mcd, sizeof(*med->med_mcd)); @@ -179,13 +187,15 @@ static int mds_read_last_rcvd(struct obd_device *obd, struct file *file) struct mds_client_data *mcd = NULL; loff_t off = 0; unsigned long last_rcvd_size = file->f_dentry->d_inode->i_size; - __u64 last_transno = 0; __u64 mount_count; int cl_idx, rc = 0; ENTRY; - LASSERT(sizeof(struct mds_client_data) == MDS_LR_CLIENT_SIZE); - LASSERT(sizeof(struct mds_server_data) <= MDS_LR_SERVER_SIZE); + /* ensure padding in the struct is the correct size */ + LASSERT (offsetof(struct mds_server_data, msd_padding) + + sizeof(msd->msd_padding) == MDS_LR_SERVER_SIZE); + LASSERT (offsetof(struct mds_client_data, mcd_padding) + + sizeof(mcd->mcd_padding) == MDS_LR_CLIENT_SIZE); OBD_ALLOC_WAIT(msd, sizeof(*msd)); if (!msd) @@ -202,32 +212,27 @@ static int mds_read_last_rcvd(struct obd_device *obd, struct file *file) if (last_rcvd_size == 0) { CWARN("%s: initializing new %s\n", obd->obd_name, LAST_RCVD); + memcpy(msd->msd_uuid, obd->obd_uuid.uuid,sizeof(msd->msd_uuid)); + msd->msd_last_transno = 0; + mount_count = msd->msd_mount_count = 0; msd->msd_server_size = cpu_to_le32(MDS_LR_SERVER_SIZE); msd->msd_client_start = cpu_to_le32(MDS_LR_CLIENT_START); msd->msd_client_size = cpu_to_le16(MDS_LR_CLIENT_SIZE); msd->msd_feature_rocompat = cpu_to_le32(MDS_ROCOMPAT_LOVOBJID); - rc = fsfilt_write_record(obd, file, msd, sizeof(*msd), &off, 1); - - if (rc == 0) - RETURN(0); - - CERROR("%s: error writing new MSD: %d\n", obd->obd_name, rc); - GOTO(err_msd, rc); - } - - rc = fsfilt_read_record(obd, file, msd, sizeof(*msd), &off); - if (rc) { - CERROR("error reading MDS %s: rc = %d\n", LAST_RCVD, rc); - GOTO(err_msd, rc); + } else { + rc = fsfilt_read_record(obd, file, msd, sizeof(*msd), &off); + if (rc) { + CERROR("error reading MDS %s: rc = %d\n", LAST_RCVD, rc); + GOTO(err_msd, rc); + } + if (strcmp(msd->msd_uuid, obd->obd_uuid.uuid) != 0) { + CERROR("OBD UUID %s does not match last_rcvd UUID %s\n", + obd->obd_uuid.uuid, msd->msd_uuid); + GOTO(err_msd, rc = -EINVAL); + } + mount_count = le64_to_cpu(msd->msd_mount_count); } - if (!msd->msd_server_size) - msd->msd_server_size = cpu_to_le32(MDS_LR_SERVER_SIZE); - if (!msd->msd_client_start) - msd->msd_client_start = cpu_to_le32(MDS_LR_CLIENT_START); - if (!msd->msd_client_size) - msd->msd_client_size = cpu_to_le16(MDS_LR_CLIENT_SIZE); - if (msd->msd_feature_incompat & ~cpu_to_le32(MDS_INCOMPAT_SUPP)) { CERROR("unsupported incompat feature %x\n", le32_to_cpu(msd->msd_feature_incompat) & @@ -244,16 +249,12 @@ static int mds_read_last_rcvd(struct obd_device *obd, struct file *file) GOTO(err_msd, rc = -EINVAL); } - last_transno = le64_to_cpu(msd->msd_last_transno); - mds->mds_last_transno = last_transno; - - mount_count = le64_to_cpu(msd->msd_mount_count); - mds->mds_mount_count = mount_count; + mds->mds_last_transno = le64_to_cpu(msd->msd_last_transno); CDEBUG(D_INODE, "%s: server last_transno: "LPU64"\n", - obd->obd_name, last_transno); + obd->obd_name, mds->mds_last_transno); CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n", - obd->obd_name, mount_count); + obd->obd_name, mount_count + 1); CDEBUG(D_INODE, "%s: server data size: %u\n", obd->obd_name, le32_to_cpu(msd->msd_server_size)); CDEBUG(D_INODE, "%s: per-client data start: %u\n", @@ -272,7 +273,8 @@ static int mds_read_last_rcvd(struct obd_device *obd, struct file *file) for (cl_idx = 0, off = le32_to_cpu(msd->msd_client_start); off < last_rcvd_size; cl_idx++) { __u64 last_transno; - int mount_age; + struct obd_export *exp; + struct mds_export_data *med; if (!mcd) { OBD_ALLOC_WAIT(mcd, sizeof(*mcd)); @@ -303,63 +305,60 @@ static int mds_read_last_rcvd(struct obd_device *obd, struct file *file) /* These exports are cleaned up by mds_disconnect(), so they * need to be set up like real exports as mds_connect() does. */ - mount_age = mount_count - le64_to_cpu(mcd->mcd_mount_count); - if (mount_age < MDS_MOUNT_RECOV) { - struct obd_export *exp = class_new_export(obd); - struct mds_export_data *med; - CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64 - " srv lr: "LPU64" mnt: "LPU64" last mount: " - LPU64"\n", mcd->mcd_uuid, cl_idx, - last_transno, le64_to_cpu(msd->msd_last_transno), - le64_to_cpu(mcd->mcd_mount_count), mount_count); - if (exp == NULL) - GOTO(err_client, rc = -ENOMEM); - - memcpy(&exp->exp_client_uuid.uuid, mcd->mcd_uuid, - sizeof exp->exp_client_uuid.uuid); - med = &exp->exp_mds_data; - med->med_mcd = mcd; - mds_client_add(obd, mds, med, cl_idx); - /* create helper if export init gets more complex */ - INIT_LIST_HEAD(&med->med_open_head); - spin_lock_init(&med->med_open_lock); - - mcd = NULL; - obd->obd_recoverable_clients++; - obd->obd_max_recoverable_clients++; - class_export_put(exp); - } else { - CDEBUG(D_INFO, "discarded client %d, UUID '%s', count " - LPU64"\n", cl_idx, mcd->mcd_uuid, - le64_to_cpu(mcd->mcd_mount_count)); - } + CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64 + " srv lr: "LPU64"\n", mcd->mcd_uuid, cl_idx, + last_transno, le64_to_cpu(msd->msd_last_transno)); + + exp = class_new_export(obd); + if (exp == NULL) + GOTO(err_client, rc = -ENOMEM); + + memcpy(&exp->exp_client_uuid.uuid, mcd->mcd_uuid, + sizeof exp->exp_client_uuid.uuid); + med = &exp->exp_mds_data; + med->med_mcd = mcd; + mds_client_add(obd, mds, med, cl_idx); + /* create helper if export init gets more complex */ + INIT_LIST_HEAD(&med->med_open_head); + spin_lock_init(&med->med_open_lock); + + mcd = NULL; + obd->obd_recoverable_clients++; + obd->obd_max_recoverable_clients++; + class_export_put(exp); CDEBUG(D_OTHER, "client at idx %d has last_transno = "LPU64"\n", cl_idx, last_transno); if (last_transno > mds->mds_last_transno) - mds->mds_last_transno = last_transno; + mds->mds_last_transno = last_transno; } obd->obd_last_committed = mds->mds_last_transno; if (obd->obd_recoverable_clients) { - CWARN("RECOVERY: %d recoverable clients, last_transno " - LPU64"\n", obd->obd_recoverable_clients, - mds->mds_last_transno); + CWARN("RECOVERY: service %s, %d recoverable clients, " + "last_transno "LPU64"\n", obd->obd_name, + obd->obd_recoverable_clients, mds->mds_last_transno); obd->obd_next_recovery_transno = obd->obd_last_committed + 1; obd->obd_recovering = 1; } if (mcd) OBD_FREE(mcd, sizeof(*mcd)); + + mds->mds_mount_count = mount_count + 1; + msd->msd_mount_count = cpu_to_le64(mds->mds_mount_count); - return 0; + /* save it, so mount count and last_transno is current */ + rc = mds_update_server_data(obd, 1); + + RETURN(rc); err_client: class_disconnect_exports(obd, 0); err_msd: mds_server_free_data(mds); - return rc; + RETURN(rc); } int mds_fs_setup(struct obd_device *obd, struct vfsmount *mnt)