/* Copyright (c) 2008-2012 Red Hat, Inc. This file is part of GlusterFS. This file is licensed to you under your choice of the GNU Lesser General Public License, version 3 or any later version (LGPLv3 or later), or the GNU General Public License, version 2 (GPLv2), in all cases as published by the Free Software Foundation. */ #include "tier.h" #include "dht-common.h" #include "xlator.h" #include "syscall.h" #include #include #include #include "events.h" #define GF_DISK_SECTOR_SIZE 512 #define DHT_REBALANCE_PID 4242 /* Change it if required */ #define DHT_REBALANCE_BLKSIZE (1024 * 1024) /* 1 MB */ #define MAX_MIGRATE_QUEUE_COUNT 500 #define MIN_MIGRATE_QUEUE_COUNT 200 #define MAX_REBAL_TYPE_SIZE 16 #define FILE_CNT_INTERVAL 600 /* 10 mins */ #define ESTIMATE_START_INTERVAL 600 /* 10 mins */ #define HARDLINK_MIG_INPROGRESS -2 #define SKIP_MIGRATION_FD_POSITIVE -3 #ifndef MAX #define MAX(a, b) (((a) > (b))?(a):(b)) #endif #define GF_CRAWL_INDEX_MOVE(idx, sv_cnt) { \ idx++; \ idx %= sv_cnt; \ } uint64_t g_totalfiles = 0; uint64_t g_totalsize = 0; void gf_defrag_free_dir_dfmeta (struct dir_dfmeta *meta, int local_subvols_cnt) { int i = 0; if (meta) { for (i = 0; i < local_subvols_cnt; i++) { gf_dirent_free (&meta->equeue[i]); } GF_FREE (meta->equeue); GF_FREE (meta->head); GF_FREE (meta->iterator); GF_FREE (meta->offset_var); GF_FREE (meta->fetch_entries); GF_FREE (meta); } } void gf_defrag_free_container (struct dht_container *container) { if (container) { gf_dirent_entry_free (container->df_entry); if (container->parent_loc) { loc_wipe (container->parent_loc); } GF_FREE (container->parent_loc); GF_FREE (container); } } void dht_set_global_defrag_error (gf_defrag_info_t *defrag, int ret) { LOCK (&defrag->lock); { defrag->global_error = ret; } UNLOCK (&defrag->lock); return; } static gf_boolean_t dht_is_tier_command (int cmd) { gf_boolean_t is_tier = _gf_false; switch (cmd) { case GF_DEFRAG_CMD_START_TIER: case GF_DEFRAG_CMD_STATUS_TIER: case GF_DEFRAG_CMD_START_DETACH_TIER: case GF_DEFRAG_CMD_STOP_DETACH_TIER: case GF_DEFRAG_CMD_PAUSE_TIER: case GF_DEFRAG_CMD_RESUME_TIER: is_tier = _gf_true; break; default: break; } return is_tier; } static int dht_send_rebalance_event (xlator_t *this, int cmd, gf_defrag_status_t status) { int ret = -1; char *volname = NULL; char *tmpstr = NULL; char *ptr = NULL; char *suffix = "-dht"; dht_conf_t *conf = NULL; gf_defrag_info_t *defrag = NULL; int len = 0; eventtypes_t event = EVENT_LAST; switch (status) { case GF_DEFRAG_STATUS_COMPLETE: event = EVENT_VOLUME_REBALANCE_COMPLETE; break; case GF_DEFRAG_STATUS_FAILED: event = EVENT_VOLUME_REBALANCE_FAILED; break; case GF_DEFRAG_STATUS_STOPPED: event = EVENT_VOLUME_REBALANCE_STOP; break; default: break; } if (dht_is_tier_command (cmd)) { /* We should have the tier volume name*/ conf = this->private; defrag = conf->defrag; volname = defrag->tier_conf.volname; } else { /* DHT volume */ len = strlen (this->name); tmpstr = gf_strdup (this->name); if (tmpstr) { ptr = tmpstr + (len - strlen (suffix)); if (!strcmp (ptr, suffix)) { tmpstr[len - strlen (suffix)] = '\0'; volname = tmpstr; } } } if (!volname) { /* Better than nothing */ volname = this->name; } if (event != EVENT_LAST) { gf_event (event, "volume=%s", volname); } GF_FREE (tmpstr); return ret; } static void dht_strip_out_acls (dict_t *dict) { if (dict) { dict_del (dict, "trusted.SGI_ACL_FILE"); dict_del (dict, POSIX_ACL_ACCESS_XATTR); } } static int dht_write_with_holes (xlator_t *to, fd_t *fd, struct iovec *vec, int count, int32_t size, off_t offset, struct iobref *iobref, int *fop_errno) { int i = 0; int ret = -1; int start_idx = 0; int tmp_offset = 0; int write_needed = 0; int buf_len = 0; int size_pending = 0; char *buf = NULL; /* loop through each vector */ for (i = 0; i < count; i++) { buf = vec[i].iov_base; buf_len = vec[i].iov_len; for (start_idx = 0; (start_idx + GF_DISK_SECTOR_SIZE) <= buf_len; start_idx += GF_DISK_SECTOR_SIZE) { if (mem_0filled (buf + start_idx, GF_DISK_SECTOR_SIZE) != 0) { write_needed = 1; continue; } if (write_needed) { ret = syncop_write (to, fd, (buf + tmp_offset), (start_idx - tmp_offset), (offset + tmp_offset), iobref, 0, NULL, NULL); /* 'path' will be logged in calling function */ if (ret < 0) { gf_log (THIS->name, GF_LOG_WARNING, "failed to write (%s)", strerror (-ret)); *fop_errno = -ret; ret = -1; goto out; } write_needed = 0; } tmp_offset = start_idx + GF_DISK_SECTOR_SIZE; } if ((start_idx < buf_len) || write_needed) { /* This means, last chunk is not yet written.. write it */ ret = syncop_write (to, fd, (buf + tmp_offset), (buf_len - tmp_offset), (offset + tmp_offset), iobref, 0, NULL, NULL); if (ret < 0) { /* 'path' will be logged in calling function */ gf_log (THIS->name, GF_LOG_WARNING, "failed to write (%s)", strerror (-ret)); *fop_errno = -ret; ret = -1; goto out; } } size_pending = (size - buf_len); if (!size_pending) break; } ret = size; out: return ret; } /* return values: -1 : failure -2 : success Hard link migration is carried out in three stages. (Say there are n hardlinks) Stage 1: Setting the new hashed subvol information on the 1st hardlink encountered (linkto setxattr) Stage 2: Creating hardlinks on new hashed subvol for the 2nd to (n-1)th hardlink Stage 3: Physical migration of the data file for nth hardlink Why to deem "-2" as success and not "0": dht_migrate_file expects return value "0" from _is_file_migratable if the file has to be migrated. _is_file_migratable returns zero only when it is called with the flag "GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS". gf_defrag_handle_hardlink calls dht_migrate_file for physical migration of the data file with the flag "GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS" Hence, gf_defrag_handle_hardlink returning "0" for success will force "dht_migrate_file" to migrate each of the hardlink which is not intended. For each of the three stage mentioned above "-2" will be returned and will be converted to "0" in dht_migrate_file. */ int32_t gf_defrag_handle_hardlink (xlator_t *this, loc_t *loc, int *fop_errno) { int32_t ret = -1; xlator_t *cached_subvol = NULL; xlator_t *hashed_subvol = NULL; xlator_t *linkto_subvol = NULL; data_t *data = NULL; struct iatt iatt = {0,}; int32_t op_errno = 0; dht_conf_t *conf = NULL; gf_loglevel_t loglevel = 0; dict_t *link_xattr = NULL; dict_t *dict = NULL; dict_t *xattr_rsp = NULL; struct iatt stbuf = {0,}; *fop_errno = EINVAL; GF_VALIDATE_OR_GOTO ("defrag", loc, out); GF_VALIDATE_OR_GOTO ("defrag", loc->name, out); GF_VALIDATE_OR_GOTO ("defrag", this, out); GF_VALIDATE_OR_GOTO ("defrag", this->private, out); conf = this->private; if (gf_uuid_is_null (loc->pargfid)) { gf_msg ("", GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed :" "loc->pargfid is NULL for %s", loc->path); *fop_errno = EINVAL; ret = -1; goto out; } if (gf_uuid_is_null (loc->gfid)) { gf_msg ("", GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed :" "loc->gfid is NULL for %s", loc->path); *fop_errno = EINVAL; ret = -1; goto out; } link_xattr = dict_new (); if (!link_xattr) { ret = -1; *fop_errno = ENOMEM; goto out; } /* Parallel migration can lead to migration of the hard link multiple times which can lead to data loss. Hence, adding a fresh lookup to decide whether migration is required or not. Elaborating the scenario for let say 10 hardlinks [link{1..10}]: Let say the first hard link "link1" does the setxattr of the new hashed subvolume info on the cached file. As there are multiple threads working, we might have already all the links created on the new hashed by the time we reach hardlink let say link5. Now the number of links on hashed is equal to that of cached. Hence, file migration will happen for link6. Cached Hashed --------T link6 rwxrwxrwx link6 Now post above state all the link file on the cached will be zero byte linkto files. Hence, if we still do migration for the following files link{7..10}, we will end up migrating 0 data leading to data loss. Hence, a lookup can make sure whether we need to migrate the file or not. */ dict = dict_new (); if (!dict) { ret = -1; *fop_errno = ENOMEM; gf_msg (this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY, "could not allocate memory for dict"); goto out; } ret = dict_set_int32 (dict, conf->link_xattr_name, 256); if (ret) { *fop_errno = ENOMEM; ret = -1; gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed:" "%s: failed to set 'linkto' key in dict", loc->path); goto out; } ret = syncop_lookup (this, loc, &stbuf, NULL, dict, &xattr_rsp); if (ret) { /*Ignore ENOENT and ESTALE as file might have been migrated already*/ if (-ret == ENOENT || -ret == ESTALE) { ret = -2; goto out; } gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed:%s lookup failed with ret = %d", loc->path, ret); *fop_errno = -ret; ret = -1; goto out; } cached_subvol = dht_subvol_get_cached (this, loc->inode); if (!cached_subvol) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed :" "Failed to get cached subvol" " for %s on %s", loc->name, this->name); *fop_errno = EINVAL; ret = -1; goto out; } hashed_subvol = dht_subvol_get_hashed (this, loc); if (!hashed_subvol) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed :" "Failed to get hashed subvol" " for %s on %s", loc->name, this->name); *fop_errno = EINVAL; ret = -1; goto out; } /* Hardlink migration happens only with remove-brick. So this condition will * be true only when the migration has happened. In case hardlinks are migrated * for rebalance case, remove this check. Having this check here avoid redundant * calls below*/ if (hashed_subvol == cached_subvol) { ret = -2; goto out; } gf_log (this->name, GF_LOG_INFO, "Attempting to migrate hardlink %s " "with gfid %s from %s -> %s", loc->name, uuid_utoa (loc->gfid), cached_subvol->name, hashed_subvol->name); data = dict_get (xattr_rsp, conf->link_xattr_name); /* set linkto on cached -> hashed if not present, else link it */ if (!data) { ret = dict_set_str (link_xattr, conf->link_xattr_name, hashed_subvol->name); if (ret) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed :" "Failed to set dictionary value:" " key = %s for %s", conf->link_xattr_name, loc->name); *fop_errno = ENOMEM; ret = -1; goto out; } ret = syncop_setxattr (cached_subvol, loc, link_xattr, 0, NULL, NULL); if (ret) { gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed :" "Linkto setxattr failed %s -> %s", cached_subvol->name, loc->name); *fop_errno = -ret; ret = -1; goto out; } gf_msg_debug (this->name, 0, "hardlink target subvol created on %s " ",cached %s, file %s", hashed_subvol->name, cached_subvol->name, loc->path); ret = -2; goto out; } else { linkto_subvol = dht_linkfile_subvol (this, NULL, NULL, xattr_rsp); if (!linkto_subvol) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_SUBVOL_ERROR, "Failed to get " "linkto subvol for %s", loc->name); } else { hashed_subvol = linkto_subvol; } ret = syncop_link (hashed_subvol, loc, loc, &iatt, NULL, NULL); if (ret) { op_errno = -ret; ret = -1; loglevel = (op_errno == EEXIST) ? GF_LOG_DEBUG : \ GF_LOG_ERROR; gf_msg (this->name, loglevel, op_errno, DHT_MSG_MIGRATE_HARDLINK_FILE_FAILED, "link of %s -> %s" " failed on subvol %s", loc->name, uuid_utoa(loc->gfid), hashed_subvol->name); if (op_errno != EEXIST) { *fop_errno = op_errno; goto out; } } else { gf_msg_debug (this->name, 0, "syncop_link successful for" " hardlink %s on subvol %s, cached %s", loc->path, hashed_subvol->name, cached_subvol->name); } } ret = syncop_lookup (hashed_subvol, loc, &iatt, NULL, NULL, NULL); if (ret) { gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed :Failed lookup %s on %s ", loc->name, hashed_subvol->name); *fop_errno = -ret; ret = -1; goto out; } /* There is a race where on the target subvol for the hardlink * (note: hash subvol for the hardlink might differ from this), some * other client(non-rebalance) would have created a linkto file for that * hardlink as part of lookup. So let say there are 10 hardlinks, on the * 5th hardlink it self the hardlinks might have migrated. Now for * (6..10th) hardlinks the cached and target would be same as the file * has already migrated. Hence this check is needed */ if (cached_subvol == hashed_subvol) { gf_msg_debug (this->name, 0, "source %s and destination %s " "for hardlink %s are same", cached_subvol->name, hashed_subvol->name, loc->path); ret = -2; goto out; } if (iatt.ia_nlink == stbuf.ia_nlink) { ret = dht_migrate_file (this, loc, cached_subvol, hashed_subvol, GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS, fop_errno); if (ret) { goto out; } } ret = -2; out: if (link_xattr) dict_unref (link_xattr); if (xattr_rsp) dict_unref (xattr_rsp); if (dict) dict_unref (dict); return ret; } static int __check_file_has_hardlink (xlator_t *this, loc_t *loc, struct iatt *stbuf, dict_t *xattrs, int flags, gf_defrag_info_t *defrag, dht_conf_t *conf, int *fop_errno) { int ret = 0; if (flags == GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS) { ret = 0; return ret; } if (stbuf->ia_nlink > 1) { /* support for decomission */ if (flags == GF_DHT_MIGRATE_HARDLINK) { synclock_lock (&conf->link_lock); ret = gf_defrag_handle_hardlink (this, loc, fop_errno); synclock_unlock (&conf->link_lock); /* Returning zero will force the file to be remigrated. Checkout gf_defrag_handle_hardlink for more information. */ if (ret && ret != -2) { gf_msg (this->name, GF_LOG_WARNING, 0, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed:" "%s: failed to migrate file with link", loc->path); } } else { gf_msg (this->name, GF_LOG_WARNING, 0, DHT_MSG_MIGRATE_FILE_FAILED, "Migration skipped for:" "%s: file has hardlinks", loc->path); *fop_errno = ENOTSUP; ret = 1; } } return ret; } /* return values 0 : File will be migrated -2 : File will not be migrated (This is the return value from gf_defrag_handle_hardlink. Checkout gf_defrag_handle_hardlink for description of "returning -2") -1 : failure */ static int __is_file_migratable (xlator_t *this, loc_t *loc, struct iatt *stbuf, dict_t *xattrs, int flags, gf_defrag_info_t *defrag, dht_conf_t *conf, int *fop_errno) { int ret = -1; int lock_count = 0; if (IA_ISDIR (stbuf->ia_type)) { gf_msg (this->name, GF_LOG_WARNING, 0, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed:" "%s: migrate-file called on directory", loc->path); *fop_errno = EISDIR; ret = -1; goto out; } if (!conf->lock_migration_enabled) { ret = dict_get_int32 (xattrs, GLUSTERFS_POSIXLK_COUNT, &lock_count); if (ret) { gf_msg (this->name, GF_LOG_WARNING, 0, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed:" "%s: Unable to get lock count for file", loc->path); *fop_errno = EINVAL; ret = -1; goto out; } if (lock_count) { gf_msg (this->name, GF_LOG_WARNING, 0, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed: %s: File has locks." " Skipping file migration", loc->path); *fop_errno = ENOTSUP; ret = 1; goto out; } } /* Check if file has hardlink*/ ret = __check_file_has_hardlink (this, loc, stbuf, xattrs, flags, defrag, conf, fop_errno); out: return ret; } static int __dht_rebalance_create_dst_file (xlator_t *this, xlator_t *to, xlator_t *from, loc_t *loc, struct iatt *stbuf, fd_t **dst_fd, int *fop_errno) { int ret = -1; int ret2 = -1; fd_t *fd = NULL; struct iatt new_stbuf = {0,}; struct iatt check_stbuf= {0,}; dht_conf_t *conf = NULL; dict_t *dict = NULL; dict_t *xdata = NULL; conf = this->private; dict = dict_new (); if (!dict) { *fop_errno = ENOMEM; ret = -1; gf_msg (this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY, "dictionary allocation failed for" "path:%s", loc->path); goto out; } ret = dict_set_gfuuid (dict, "gfid-req", stbuf->ia_gfid, true); if (ret) { *fop_errno = ENOMEM; ret = -1; gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED, "%s: failed to set dictionary value: key = gfid-req", loc->path); goto out; } ret = dict_set_str (dict, conf->link_xattr_name, from->name); if (ret) { *fop_errno = ENOMEM; ret = -1; gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED, "%s: failed to set dictionary value: key = %s ", loc->path, conf->link_xattr_name); goto out; } fd = fd_create (loc->inode, DHT_REBALANCE_PID); if (!fd) { *fop_errno = ENOMEM; ret = -1; gf_msg (this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_MIGRATE_FILE_FAILED, "%s: fd create failed (destination)", loc->path); goto out; } if (!!dht_is_tier_xlator (this)) { xdata = dict_new (); if (!xdata) { *fop_errno = ENOMEM; ret = -1; gf_msg (this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_MIGRATE_FILE_FAILED, "%s: dict_new failed)", loc->path); goto out; } ret = dict_set_int32 (xdata, GF_CLEAN_WRITE_PROTECTION, 1); if (ret) { *fop_errno = ENOMEM; ret = -1; gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED, "%s: failed to set dictionary value: key = %s ", loc->path, GF_CLEAN_WRITE_PROTECTION); goto out; } } ret = syncop_lookup (to, loc, &new_stbuf, NULL, xdata, NULL); if (!ret) { /* File exits in the destination, check if gfid matches */ if (gf_uuid_compare (stbuf->ia_gfid, new_stbuf.ia_gfid) != 0) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_GFID_MISMATCH, "file %s exists in %s with different gfid", loc->path, to->name); *fop_errno = EINVAL; ret = -1; goto out; } } if ((ret < 0) && (-ret != ENOENT)) { /* File exists in destination, but not accessible */ gf_msg (THIS->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "%s: failed to lookup file", loc->path); *fop_errno = -ret; ret = -1; goto out; } /* Create the destination with LINKFILE mode, and linkto xattr, if the linkfile already exists, just open the file */ if (!ret) { /* * File already present, just open the file. */ ret = syncop_open (to, loc, O_RDWR, fd, NULL, NULL); if (ret < 0) { gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "failed to open %s on %s", loc->path, to->name); *fop_errno = -ret; ret = -1; goto out; } } else { ret = syncop_create (to, loc, O_RDWR, DHT_LINKFILE_MODE, fd, &new_stbuf, dict, NULL); if (ret < 0) { gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "failed to create %s on %s", loc->path, to->name); *fop_errno = -ret; ret = -1; goto out; } } fd_bind (fd); /*Reason of doing lookup after create again: *In the create, there is some time-gap between opening fd at the *server (posix_layer) and binding it in server (incrementing fd count), *so if in that time-gap, if other process sends unlink considering it *as a linkto file, because inode->fd count will be 0, so file will be *unlinked at the backend. And because further operations are performed *on fd, so though migration will be done but will end with no file *at the backend. */ ret = syncop_lookup (to, loc, &check_stbuf, NULL, NULL, NULL); if (!ret) { if (gf_uuid_compare (stbuf->ia_gfid, check_stbuf.ia_gfid) != 0) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_GFID_MISMATCH, "file %s exists in %s with different gfid," "found in lookup after create", loc->path, to->name); *fop_errno = EINVAL; ret = -1; goto out; } } if (-ret == ENOENT) { gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "%s: file does not exist" "on %s", loc->path, to->name); *fop_errno = -ret; ret = -1; goto out; } ret = syncop_fsetattr (to, fd, stbuf, (GF_SET_ATTR_UID | GF_SET_ATTR_GID), NULL, NULL, NULL, NULL); if (ret < 0) { *fop_errno = -ret; gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "chown failed for %s on %s", loc->path, to->name); } /* No need to bother about 0 byte size files */ if (stbuf->ia_size > 0) { if (conf->use_fallocate) { ret = syncop_fallocate (to, fd, 0, 0, stbuf->ia_size, NULL, NULL); if (ret < 0) { if (ret == -EOPNOTSUPP || ret == -EINVAL || ret == -ENOSYS) { conf->use_fallocate = _gf_false; } else { gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "fallocate failed for %s on %s", loc->path, to->name); *fop_errno = -ret; /* fallocate does not release the space * in some cases */ ret2 = syncop_ftruncate (to, fd, 0, NULL, NULL, NULL, NULL); if (ret2 < 0) { gf_msg (this->name, GF_LOG_WARNING, -ret2, DHT_MSG_MIGRATE_FILE_FAILED, "ftruncate failed for " "%s on %s", loc->path, to->name); } goto out; } } } if (!conf->use_fallocate) { ret = syncop_ftruncate (to, fd, stbuf->ia_size, NULL, NULL, NULL, NULL); if (ret < 0) { *fop_errno = -ret; gf_msg (this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "ftruncate failed for %s on %s", loc->path, to->name); } } } /* success */ ret = 0; if (dst_fd) *dst_fd = fd; out: if (ret) { if (fd) { fd_unref (fd); } } if (dict) dict_unref (dict); if (xdata) dict_unref (dict); return ret; } static int __dht_check_free_space (xlator_t *this, xlator_t *to, xlator_t *from, loc_t *loc, struct iatt *stbuf, int flag, dht_conf_t *conf, gf_boolean_t *target_changed, xlator_t **new_subvol, int *fop_errno) { struct statvfs src_statfs = {0,}; struct statvfs dst_statfs = {0,}; int ret = -1; dict_t *xdata = NULL; dht_layout_t *layout = NULL; uint64_t src_statfs_blocks = 1; uint64_t dst_statfs_blocks = 1; double dst_post_availspacepercent = 0; double src_post_availspacepercent = 0; uint64_t file_blocks = 0; uint64_t src_total_blocks = 0; uint64_t dst_total_blocks = 0; xdata = dict_new (); if (!xdata) { *fop_errno = ENOMEM; ret = -1; gf_msg (this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY, "failed to allocate dictionary"); goto out; } ret = dict_set_int8 (xdata, GF_INTERNAL_IGNORE_DEEM_STATFS, 1); if (ret) { gf_log (this->name, GF_LOG_ERROR, "Failed to set " GF_INTERNAL_IGNORE_DEEM_STATFS" in dict"); ret = -1; *fop_errno = ENOMEM; goto out; } ret = syncop_statfs (from, loc, &src_statfs, xdata, NULL); if (ret) { gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "failed to get statfs of %s on %s", loc->path, from->name); *fop_errno = -ret; ret = -1; goto out; } ret = syncop_statfs (to, loc, &dst_statfs, xdata, NULL); if (ret) { gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "failed to get statfs of %s on %s", loc->path, to->name); *fop_errno = -ret; ret = -1; goto out; } gf_msg_debug (this->name, 0, "min_free_disk - %f , block available - " "%lu , block size - %lu ", conf->min_free_disk, dst_statfs.f_bavail, dst_statfs.f_bsize); dst_statfs_blocks = dst_statfs.f_bavail * (dst_statfs.f_frsize / GF_DISK_SECTOR_SIZE); src_statfs_blocks = src_statfs.f_bavail * (src_statfs.f_frsize / GF_DISK_SECTOR_SIZE); dst_total_blocks = dst_statfs.f_blocks * (dst_statfs.f_frsize / GF_DISK_SECTOR_SIZE); src_total_blocks = src_statfs.f_blocks * (src_statfs.f_frsize / GF_DISK_SECTOR_SIZE); /* if force option is given, do not check for space @ dst. * Check only if space is avail for the file */ if (flag != GF_DHT_MIGRATE_DATA) goto check_avail_space; /* Check: During rebalance `migrate-data` - Destination subvol experiences a `reduction` in 'blocks' of free space, at the same time source subvol gains certain 'blocks' of free space. A valid check is necessary here to avoid erroneous move to destination where the space could be scantily available. With heterogeneous brick support, an actual space comparison could prevent any files being migrated to newly added bricks if they are smaller then the free space available on the existing bricks. */ if (stbuf) { if (!conf->use_fallocate) { file_blocks = stbuf->ia_size + GF_DISK_SECTOR_SIZE - 1; file_blocks /= GF_DISK_SECTOR_SIZE; if (file_blocks >= dst_statfs_blocks) { dst_statfs_blocks = 0; } else { dst_statfs_blocks -= file_blocks; } } src_post_availspacepercent = ((src_statfs_blocks + file_blocks) * 100) / src_total_blocks; dst_post_availspacepercent = (dst_statfs_blocks * 100) / dst_total_blocks; if (dst_post_availspacepercent < src_post_availspacepercent) { gf_msg (this->name, GF_LOG_WARNING, 0, DHT_MSG_MIGRATE_FILE_FAILED, "data movement of file " "{blocks:%"PRIu64" name:(%s)} would result in " "dst node (%s:%"PRIu64") having lower disk " "space than the source node (%s:%"PRIu64")" ".Skipping file.", stbuf->ia_blocks, loc->path, to->name, dst_statfs_blocks, from->name, src_statfs_blocks); /* this is not a 'failure', but we don't want to consider this as 'success' too :-/ */ *fop_errno = ENOSPC; ret = 1; goto out; } } check_avail_space: if (conf->disk_unit == 'p' && dst_statfs.f_blocks) { dst_post_availspacepercent = (dst_statfs_blocks * 100) / dst_total_blocks; gf_msg_debug (this->name, 0, "file : %s, post_availspacepercent" " : %lf f_bavail : %lu min-free-disk: %lf", loc->path, dst_post_availspacepercent, dst_statfs.f_bavail, conf->min_free_disk); if (dst_post_availspacepercent < conf->min_free_disk) { gf_msg (this->name, GF_LOG_WARNING, 0, 0, "Write will cross min-free-disk for " "file - %s on subvol - %s. Looking " "for new subvol", loc->path, to->name); goto find_new_subvol; } else { ret = 0; goto out; } } if (conf->disk_unit != 'p') { if ((dst_statfs_blocks * GF_DISK_SECTOR_SIZE) < conf->min_free_disk) { gf_msg_debug (this->name, 0, "file : %s, destination " "frsize: %lu f_bavail : %lu " "min-free-disk: %lf", loc->path, dst_statfs.f_frsize, dst_statfs.f_bavail, conf->min_free_disk); gf_msg (this->name, GF_LOG_WARNING, 0, 0, "write will" " cross min-free-disk for file - %s on subvol -" " %s. looking for new subvol", loc->path, to->name); goto find_new_subvol; } else { ret = 0; goto out; } } find_new_subvol: layout = dht_layout_get (this, loc->parent); if (!layout) { gf_log (this->name, GF_LOG_ERROR, "Layout is NULL"); *fop_errno = EINVAL; ret = -1; goto out; } *new_subvol = dht_subvol_with_free_space_inodes (this, to, from, layout, stbuf->ia_size); if ((!(*new_subvol)) || (*new_subvol == from)) { gf_msg (this->name, GF_LOG_WARNING, 0, DHT_MSG_SUBVOL_INSUFF_SPACE, "Could not find any subvol" " with space accommodating the file - %s. Consider " "adding bricks", loc->path); *target_changed = _gf_false; *fop_errno = ENOSPC; ret = -1; } else { gf_msg (this->name, GF_LOG_INFO, 0, 0, "new target found - %s" " for file - %s", (*new_subvol)->name, loc->path); *target_changed = _gf_true; ret = 0; } out: if (xdata) dict_unref (xdata); return ret; } static int __dht_rebalance_migrate_data (xlator_t *this, gf_defrag_info_t *defrag, xlator_t *from, xlator_t *to, fd_t *src, fd_t *dst, uint64_t ia_size, int hole_exists, int *fop_errno) { int ret = 0; int count = 0; off_t offset = 0; struct iovec *vector = NULL; struct iobref *iobref = NULL; uint64_t total = 0; size_t read_size = 0; dict_t *xdata = NULL; dht_conf_t *conf = NULL; conf = this->private; /* if file size is '0', no need to enter this loop */ while (total < ia_size) { read_size = (((ia_size - total) > DHT_REBALANCE_BLKSIZE) ? DHT_REBALANCE_BLKSIZE : (ia_size - total)); ret = syncop_readv (from, src, read_size, offset, 0, &vector, &count, &iobref, NULL, NULL, NULL); if (!ret || (ret < 0)) { *fop_errno = -ret; break; } if (hole_exists) { ret = dht_write_with_holes (to, dst, vector, count, ret, offset, iobref, fop_errno); } else { if (!conf->force_migration && !dht_is_tier_xlator (this)) { xdata = dict_new (); if (!xdata) { gf_msg ("dht", GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED, "insufficient memory"); ret = -1; *fop_errno = ENOMEM; break; } /* Fail this write and abort rebalance if we * detect a write from client since migration of * this file started. This is done to avoid * potential data corruption due to out of order * writes from rebalance and client to the same * region (as compared between src and dst * files). See * https://github.com/gluster/glusterfs/issues/308 * for more details. */ ret = dict_set_int32 (xdata, GF_AVOID_OVERWRITE, 1); if (ret) { gf_msg ("dht", GF_LOG_ERROR, 0, ENOMEM, "failed to set dict"); ret = -1; *fop_errno = ENOMEM; break; } } ret = syncop_writev (to, dst, vector, count, offset, iobref, 0, NULL, NULL, xdata, NULL); if (ret < 0) { *fop_errno = -ret; } } if ((defrag && defrag->cmd == GF_DEFRAG_CMD_START_TIER) && (gf_defrag_get_pause_state (&defrag->tier_conf) != TIER_RUNNING)) { gf_msg ("tier", GF_LOG_INFO, 0, DHT_MSG_TIER_PAUSED, "Migrate file paused"); ret = -1; } if (ret < 0) { break; } offset += ret; total += ret; GF_FREE (vector); if (iobref) iobref_unref (iobref); iobref = NULL; vector = NULL; } if (iobref) iobref_unref (iobref); GF_FREE (vector); if (ret >= 0) ret = 0; else ret = -1; if (xdata) { dict_unref (xdata); } return ret; } static int __dht_rebalance_open_src_file (xlator_t *this, xlator_t *from, xlator_t *to, loc_t *loc, struct iatt *stbuf, fd_t **src_fd, gf_boolean_t *clean_src, int *fop_errno) { int ret = 0; fd_t *fd = NULL; dict_t *dict = NULL; struct iatt iatt = {0,}; dht_conf_t *conf = NULL; conf = this->private; *clean_src = _gf_false; fd = fd_create (loc->inode, DHT_REBALANCE_PID); if (!fd) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED, "%s: fd create failed (source)", loc->path); *fop_errno = ENOMEM; ret = -1; goto out; } ret = syncop_open (from, loc, O_RDWR, fd, NULL, NULL); if (ret < 0) { gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "failed to open file %s on %s", loc->path, from->name); *fop_errno = -ret; ret = -1; goto out; } fd_bind (fd); if (src_fd) *src_fd = fd; ret = -1; dict = dict_new (); if (!dict) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED, "%s: Could not allocate memory for dict", loc->path); *fop_errno = ENOMEM; ret = -1; goto out; } ret = dict_set_str (dict, conf->link_xattr_name, to->name); if (ret) { gf_log (this->name, GF_LOG_ERROR, "failed to set xattr in dict for %s (linkto:%s)", loc->path, to->name); *fop_errno = ENOMEM; ret = -1; goto out; } /* Once the migration starts, the source should have 'linkto' key set to show which is the target, so other clients can work around it */ ret = syncop_setxattr (from, loc, dict, 0, NULL, NULL); if (ret) { gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "failed to set xattr on %s in %s", loc->path, from->name); *fop_errno = -ret; ret = -1; goto out; } /* Reset source mode/xattr if migration fails*/ *clean_src = _gf_true; /* mode should be (+S+T) to indicate migration is in progress */ iatt.ia_prot = stbuf->ia_prot; iatt.ia_type = stbuf->ia_type; iatt.ia_prot.sticky = 1; iatt.ia_prot.sgid = 1; ret = syncop_setattr (from, loc, &iatt, GF_SET_ATTR_MODE, NULL, NULL, NULL, NULL); if (ret) { gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "failed to set mode on %s in %s", loc->path, from->name); *fop_errno = -ret; ret = -1; goto out; } /* success */ ret = 0; out: if (dict) dict_unref (dict); return ret; } int migrate_special_files (xlator_t *this, xlator_t *from, xlator_t *to, loc_t *loc, struct iatt *buf, int *fop_errno) { int ret = -1; dict_t *rsp_dict = NULL; dict_t *dict = NULL; char *link = NULL; struct iatt stbuf = {0,}; dht_conf_t *conf = this->private; dict = dict_new (); if (!dict) { *fop_errno = ENOMEM; ret = -1; goto out; } ret = dict_set_int32 (dict, conf->link_xattr_name, 256); if (ret) { *fop_errno = ENOMEM; ret = -1; gf_log (this->name, GF_LOG_ERROR, "%s: failed to set 'linkto' key in dict", loc->path); goto out; } /* check in the destination if the file is link file */ ret = syncop_lookup (to, loc, &stbuf, NULL, dict, &rsp_dict); if ((ret < 0) && (-ret != ENOENT)) { gf_msg (this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "%s: lookup failed", loc->path); *fop_errno = -ret; ret = -1; goto out; } /* we no more require this key */ dict_del (dict, conf->link_xattr_name); /* file exists in target node, only if it is 'linkfile' its valid, otherwise, error out */ if (!ret) { if (!check_is_linkfile (loc->inode, &stbuf, rsp_dict, conf->link_xattr_name)) { gf_msg (this->name, GF_LOG_WARNING, 0, DHT_MSG_MIGRATE_FILE_FAILED, "%s: file exists in destination", loc->path); *fop_errno = EINVAL; ret = -1; goto out; } /* as file is linkfile, delete it */ ret = syncop_unlink (to, loc, NULL, NULL); if (ret) { gf_msg (this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "%s: failed to delete the linkfile", loc->path); *fop_errno = -ret; ret = -1; goto out; } } /* Set the gfid of the source file in dict */ ret = dict_set_gfuuid (dict, "gfid-req", buf->ia_gfid, true); if (ret) { *fop_errno = ENOMEM; ret = -1; gf_log (this->name, GF_LOG_ERROR, "%s: failed to set gfid in dict for create", loc->path); goto out; } /* Create the file in target */ if (IA_ISLNK (buf->ia_type)) { /* Handle symlinks separately */ ret = syncop_readlink (from, loc, &link, buf->ia_size, NULL, NULL); if (ret < 0) { gf_msg (this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "%s: readlink on symlink failed", loc->path); *fop_errno = -ret; ret = -1; goto out; } ret = syncop_symlink (to, loc, link, 0, dict, NULL); if (ret) { gf_msg (this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "%s: creating symlink failed", loc->path); *fop_errno = -ret; ret = -1; goto out; } goto done; } ret = syncop_mknod (to, loc, st_mode_from_ia (buf->ia_prot, buf->ia_type), makedev (ia_major (buf->ia_rdev), ia_minor (buf->ia_rdev)), 0, dict, NULL); if (ret) { gf_msg (this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "%s: mknod failed", loc->path); *fop_errno = -ret; ret = -1; goto out; } done: ret = syncop_setattr (to, loc, buf, (GF_SET_ATTR_MTIME | GF_SET_ATTR_UID | GF_SET_ATTR_GID | GF_SET_ATTR_MODE), NULL, NULL, NULL, NULL); if (ret) { gf_msg (this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "%s: failed to perform setattr on %s", loc->path, to->name); *fop_errno = -ret; } ret = syncop_unlink (from, loc, NULL, NULL); if (ret) { gf_msg (this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "%s: unlink failed", loc->path); *fop_errno = -ret; ret = -1; } out: GF_FREE (link); if (dict) dict_unref (dict); if (rsp_dict) dict_unref (rsp_dict); return ret; } static int __dht_migration_cleanup_src_file (xlator_t *this, loc_t *loc, fd_t *fd, xlator_t *from, ia_prot_t *src_ia_prot) { int ret = -1; dht_conf_t *conf = NULL; struct iatt new_stbuf = {0,}; if (!this || !fd || !from || !src_ia_prot) { goto out; } conf = this->private; /*Revert source mode and xattr changes*/ ret = syncop_fstat (from, fd, &new_stbuf, NULL, NULL); if (ret < 0) { /* Failed to get the stat info */ gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file cleanup failed: failed to fstat " "file %s on %s ", loc->path, from->name); ret = -1; goto out; } /* Remove the sticky bit and sgid bit set, reset it to 0*/ if (!src_ia_prot->sticky) new_stbuf.ia_prot.sticky = 0; if (!src_ia_prot->sgid) new_stbuf.ia_prot.sgid = 0; ret = syncop_fsetattr (from, fd, &new_stbuf, (GF_SET_ATTR_GID | GF_SET_ATTR_MODE), NULL, NULL, NULL, NULL); if (ret) { gf_msg (this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file cleanup failed:" "%s: failed to perform fsetattr on %s ", loc->path, from->name); ret = -1; goto out; } ret = syncop_fremovexattr (from, fd, conf->link_xattr_name, 0, NULL); if (ret) { gf_log (this->name, GF_LOG_WARNING, "%s: failed to remove linkto xattr on %s (%s)", loc->path, from->name, strerror (-ret)); ret = -1; goto out; } ret = 0; out: return ret; } /* return values: -1 : failure 0 : successfully migrated data 1 : not a failure, but we can't migrate data as of now */ int dht_migrate_file (xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to, int flag, int *fop_errno) { int ret = -1; struct iatt new_stbuf = {0,}; struct iatt stbuf = {0,}; struct iatt empty_iatt = {0,}; ia_prot_t src_ia_prot = {0,}; fd_t *src_fd = NULL; fd_t *dst_fd = NULL; dict_t *dict = NULL; dict_t *xattr = NULL; dict_t *xattr_rsp = NULL; int file_has_holes = 0; dht_conf_t *conf = this->private; int rcvd_enoent_from_src = 0; struct gf_flock flock = {0, }; struct gf_flock plock = {0, }; loc_t tmp_loc = {0, }; loc_t parent_loc = {0, }; gf_boolean_t inodelk_locked = _gf_false; gf_boolean_t entrylk_locked = _gf_false; gf_boolean_t p_locked = _gf_false; int lk_ret = -1; gf_defrag_info_t *defrag = NULL; gf_boolean_t clean_src = _gf_false; gf_boolean_t clean_dst = _gf_false; int log_level = GF_LOG_INFO; gf_boolean_t delete_src_linkto = _gf_true; lock_migration_info_t locklist; dict_t *meta_dict = NULL; gf_boolean_t meta_locked = _gf_false; gf_boolean_t target_changed = _gf_false; xlator_t *new_target = NULL; xlator_t *old_target = NULL; xlator_t *hashed_subvol = NULL; fd_t *linkto_fd = NULL; if (from == to) { gf_msg_debug (this->name, 0, "destination and source are same. file %s" " might have migrated already", loc->path); ret = 0; goto out; } /* If defrag is NULL, it should be assumed that migration is triggered * from client */ defrag = conf->defrag; /* migration of files from clients is restricted to non-tiered clients * for now */ if (!defrag && dht_is_tier_xlator (this)) { ret = ENOTSUP; goto out; } if (defrag && defrag->tier_conf.is_tier) log_level = GF_LOG_TRACE; gf_log (this->name, log_level, "%s: attempting to move from %s to %s", loc->path, from->name, to->name); dict = dict_new (); if (!dict) { ret = -1; *fop_errno = ENOMEM; gf_msg (this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY, "Could not allocate memory for dict"); goto out; } ret = dict_set_int32 (dict, conf->link_xattr_name, 256); if (ret) { *fop_errno = ENOMEM; ret = -1; gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed:" "%s: failed to set 'linkto' key in dict", loc->path); goto out; } /* Do not migrate file in case lock migration is not enabled on the * volume*/ if (!conf->lock_migration_enabled) { ret = dict_set_int32 (dict, GLUSTERFS_POSIXLK_COUNT, sizeof(int32_t)); if (ret) { *fop_errno = ENOMEM; ret = -1; gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed: %s: failed to " "set "GLUSTERFS_POSIXLK_COUNT" key in dict", loc->path); goto out; } } else { gf_msg (this->name, GF_LOG_INFO, 0, 0, "locks will be migrated" " for file: %s", loc->path); } ret = dht_build_parent_loc (this, &parent_loc, loc, fop_errno); if (ret < 0) { ret = -1; gf_msg (this->name, GF_LOG_WARNING, *fop_errno, DHT_MSG_MIGRATE_FILE_FAILED, "%s: failed to build parent loc, which is needed to " "acquire entrylk to synchronize with renames on this " "path. Skipping migration", loc->path); goto out; } hashed_subvol = dht_subvol_get_hashed (this, loc); if (hashed_subvol == NULL) { ret = -1; gf_msg (this->name, GF_LOG_WARNING, EINVAL, DHT_MSG_MIGRATE_FILE_FAILED, "%s: cannot find hashed subvol which is needed to " "synchronize with renames on this path. " "Skipping migration", loc->path); goto out; } flock.l_type = F_WRLCK; tmp_loc.inode = inode_ref (loc->inode); gf_uuid_copy (tmp_loc.gfid, loc->gfid); tmp_loc.path = gf_strdup(loc->path); /* this inodelk happens with flock.owner being zero. But to synchronize * hardlink migration we need to have different lkowner for each migration * Filed a bug here: https://bugzilla.redhat.com/show_bug.cgi?id=1468202 to * track the fix for this. Currently synclock takes care of synchronizing * hardlink migration. Once this bug is fixed we can avoid taking synclock */ ret = syncop_inodelk (from, DHT_FILE_MIGRATE_DOMAIN, &tmp_loc, F_SETLKW, &flock, NULL, NULL); if (ret < 0) { *fop_errno = -ret; ret = -1; gf_msg (this->name, GF_LOG_WARNING, *fop_errno, DHT_MSG_MIGRATE_FILE_FAILED, "migrate file failed: " "%s: failed to lock file on %s", loc->path, from->name); goto out; } inodelk_locked = _gf_true; /* dht_rename has changed to use entrylk on hashed subvol for * synchronization. So, rebalance too has to acquire an entrylk on * hashed subvol. */ ret = syncop_entrylk (hashed_subvol, DHT_ENTRY_SYNC_DOMAIN, &parent_loc, loc->name, ENTRYLK_LOCK, ENTRYLK_WRLCK, NULL, NULL); if (ret < 0) { *fop_errno = -ret; ret = -1; gf_msg (this->name, GF_LOG_WARNING, *fop_errno, DHT_MSG_MIGRATE_FILE_FAILED, "%s: failed to acquire entrylk on subvol %s", loc->path, hashed_subvol->name); goto out; } entrylk_locked = _gf_true; /* Phase 1 - Data migration is in progress from now on */ ret = syncop_lookup (from, loc, &stbuf, NULL, dict, &xattr_rsp); if (ret) { *fop_errno = -ret; ret = -1; gf_msg (this->name, GF_LOG_ERROR, *fop_errno, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed:" "%s: lookup failed on %s", loc->path, from->name); goto out; } /* preserve source mode, so set the same to the destination */ src_ia_prot = stbuf.ia_prot; /* Check if file can be migrated */ ret = __is_file_migratable (this, loc, &stbuf, xattr_rsp, flag, defrag, conf, fop_errno); if (ret) { if (ret == HARDLINK_MIG_INPROGRESS) ret = 0; goto out; } /* Take care of the special files */ if (!IA_ISREG (stbuf.ia_type)) { /* Special files */ ret = migrate_special_files (this, from, to, loc, &stbuf, fop_errno); goto out; } /* create the destination, with required modes/xattr */ ret = __dht_rebalance_create_dst_file (this, to, from, loc, &stbuf, &dst_fd, fop_errno); if (ret) { gf_msg (this->name, GF_LOG_ERROR, 0, 0, "Create dst failed" " on - %s for file - %s", to->name, loc->path); goto out; } clean_dst = _gf_true; ret = __dht_check_free_space (this, to, from, loc, &stbuf, flag, conf, &target_changed, &new_target, fop_errno); if (target_changed) { /* Can't handle for hardlinks. Marking this as failure */ if (flag == GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS || stbuf.ia_nlink > 1) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_SUBVOL_INSUFF_SPACE, "Exiting migration for" " file - %s. flag - %d, stbuf.ia_nlink - %d", loc->path, flag, stbuf.ia_nlink); ret = -1; goto out; } ret = syncop_ftruncate (to, dst_fd, 0, NULL, NULL, NULL, NULL); if (ret) { gf_log (this->name, GF_LOG_WARNING, "%s: failed to perform truncate on %s (%s)", loc->path, to->name, strerror (-ret)); } syncop_close (dst_fd); dst_fd = NULL; old_target = to; to = new_target; clean_dst = _gf_false; /* if the file migration is successful to this new target, then * update the xattr on the old destination to point the new * destination. We need to do update this only post migration * as in case of failure the linkto needs to point to the source * subvol */ ret = __dht_rebalance_create_dst_file (this, to, from, loc, &stbuf, &dst_fd, fop_errno); if (ret) { gf_log (this->name, GF_LOG_ERROR, "Create dst failed" " on - %s for file - %s", to->name, loc->path); goto out; } else { gf_msg (this->name, GF_LOG_INFO, 0, 0, "destination for file " "- %s is changed to - %s", loc->path, to->name); clean_dst = _gf_true; } } if (ret) { goto out; } /* Open the source, and also update mode/xattr */ ret = __dht_rebalance_open_src_file (this, from, to, loc, &stbuf, &src_fd, &clean_src, fop_errno); if (ret) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed: failed to open %s on %s", loc->path, from->name); goto out; } /* TODO: move all xattr related operations to fd based operations */ ret = syncop_listxattr (from, loc, &xattr, NULL, NULL); if (ret < 0) { *fop_errno = -ret; gf_msg (this->name, GF_LOG_WARNING, *fop_errno, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed:" "%s: failed to get xattr from %s", loc->path, from->name); ret = -1; goto out; } /* Copying posix acls to the linkto file messes up the permissions*/ dht_strip_out_acls (xattr); /* Remove the linkto xattr as we don't want to overwrite the value * set on the dst. */ dict_del (xattr, conf->link_xattr_name); /* We need to error out if this fails as having the wrong shard xattrs * set on the dst could cause data corruption */ ret = syncop_fsetxattr (to, dst_fd, xattr, 0, NULL, NULL); if (ret < 0) { *fop_errno = -ret; gf_msg (this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "%s: failed to set xattr on %s", loc->path, to->name); ret = -1; goto out; } if (xattr_rsp) { /* we no more require this key */ dict_del (dict, conf->link_xattr_name); dict_unref (xattr_rsp); } ret = syncop_fstat (from, src_fd, &stbuf, dict, &xattr_rsp); if (ret) { gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed:failed to lookup %s on %s ", loc->path, from->name); *fop_errno = -ret; ret = -1; goto out; } /* Check again if file has hardlink */ ret = __check_file_has_hardlink (this, loc, &stbuf, xattr_rsp, flag, defrag, conf, fop_errno); if (ret) { if (ret == HARDLINK_MIG_INPROGRESS) ret = 0; goto out; } /* Try to preserve 'holes' while migrating data */ if (stbuf.ia_size > (stbuf.ia_blocks * GF_DISK_SECTOR_SIZE)) file_has_holes = 1; ret = __dht_rebalance_migrate_data (this, defrag, from, to, src_fd, dst_fd, stbuf.ia_size, file_has_holes, fop_errno); if (ret) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed: %s: failed to migrate data", loc->path); ret = -1; goto out; } /* TODO: Sync the locks */ ret = syncop_fsync (to, dst_fd, 0, NULL, NULL, NULL, NULL); if (ret) { gf_log (this->name, GF_LOG_WARNING, "%s: failed to fsync on %s (%s)", loc->path, to->name, strerror (-ret)); *fop_errno = -ret; } /* Phase 2 - Data-Migration Complete, Housekeeping updates pending */ ret = syncop_fstat (from, src_fd, &new_stbuf, NULL, NULL); if (ret < 0) { /* Failed to get the stat info */ gf_msg ( this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed: failed to fstat file %s on %s ", loc->path, from->name); *fop_errno = -ret; ret = -1; goto out; } /* Lock the entire source file to prevent clients from taking a lock on it as dht_lk does not handle file migration. This still leaves a small window where conflicting locks can be granted to different clients. If client1 requests a blocking lock on the src file, it will be granted after the migrating process releases its lock. If client2 requests a lock on the dst data file, it will also be granted, but all FOPs will be redirected to the dst data file. */ /* Take meta lock */ if (conf->lock_migration_enabled) { meta_dict = dict_new (); if (!meta_dict) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED, "dict_new failed"); *fop_errno = ENOMEM; ret = -1; goto out; } ret = dict_set_str (meta_dict, GLUSTERFS_INTERNAL_FOP_KEY, "yes"); if (ret) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED, "Failed to set dictionary value: key = %s," " path = %s", GLUSTERFS_INTERNAL_FOP_KEY, loc->path); *fop_errno = ENOMEM; ret = -1; goto out; } ret = dict_set_int32 (meta_dict, GF_META_LOCK_KEY, 1); if (ret) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED, "Trace dict_set failed"); *fop_errno = ENOMEM; ret = -1; goto out; } ret = syncop_setxattr (from, loc, meta_dict, 0, NULL, NULL); if (ret) { gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "Trace syncop_setxattr metalock failed"); *fop_errno = -ret; ret = -1; goto out; } else { meta_locked = _gf_true; } } if (!conf->lock_migration_enabled) { plock.l_type = F_WRLCK; plock.l_start = 0; plock.l_len = 0; plock.l_whence = SEEK_SET; ret = syncop_lk (from, src_fd, F_SETLK, &plock, NULL, NULL); if (ret) { gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed:" "%s: Failed to lock on %s", loc->path, from->name); *fop_errno = -ret; ret = -1; goto out; } p_locked = _gf_true; } else { INIT_LIST_HEAD (&locklist.list); ret = syncop_getactivelk (from, loc, &locklist, NULL, NULL); if (ret == 0) { gf_log (this->name, GF_LOG_INFO, "No active locks on:%s" , loc->path); } else if (ret > 0) { ret = syncop_setactivelk (to, loc, &locklist, NULL, NULL); if (ret) { gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_LOCK_MIGRATION_FAILED, "write lock failed on:%s", loc->path); *fop_errno = -ret; ret = -1; goto metaunlock; } } else { gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_LOCK_MIGRATION_FAILED, "getactivelk failed for file: %s", loc->path); *fop_errno = -ret; } } /* source would have both sticky bit and sgid bit set, reset it to 0, and set the source permission on destination, if it was not set prior to setting rebalance-modes in source */ if (!src_ia_prot.sticky) new_stbuf.ia_prot.sticky = 0; if (!src_ia_prot.sgid) new_stbuf.ia_prot.sgid = 0; /* TODO: if the source actually had sticky bit, or sgid bit set, we are not handling it */ ret = syncop_fsetattr (to, dst_fd, &new_stbuf, (GF_SET_ATTR_UID | GF_SET_ATTR_GID | GF_SET_ATTR_MODE), NULL, NULL, NULL, NULL); if (ret) { gf_msg (this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed:" "%s: failed to perform setattr on %s ", loc->path, to->name); *fop_errno = -ret; ret = -1; goto metaunlock; } /* Because 'futimes' is not portable */ ret = syncop_setattr (to, loc, &new_stbuf, (GF_SET_ATTR_MTIME | GF_SET_ATTR_ATIME), NULL, NULL, NULL, NULL); if (ret) { gf_log (this->name, GF_LOG_WARNING, "%s: failed to perform setattr on %s ", loc->path, to->name); *fop_errno = -ret; } if (target_changed) { dict_del (dict, conf->link_xattr_name); dict_del (dict, GLUSTERFS_POSIXLK_COUNT); ret = dict_set_str (dict, conf->link_xattr_name, to->name); if (ret) { gf_log (this->name, GF_LOG_ERROR, "failed to set xattr in dict for %s (linkto:%s)", loc->path, to->name); *fop_errno = ENOMEM; ret = -1; goto out; } ret = syncop_setxattr (old_target, loc, dict, 0, NULL, NULL); if (ret && -ret != ESTALE && -ret != ENOENT) { gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "failed to set xattr on %s in %s", loc->path, old_target->name); *fop_errno = -ret; ret = -1; goto out; } else if (-ret == ESTALE || -ret == ENOENT) { /* The failure ESTALE indicates that the linkto * file on the hashed subvol might have been deleted. * In this case will create a linkto file with new target * as linkto xattr value*/ linkto_fd = fd_create (loc->inode, DHT_REBALANCE_PID); if (!linkto_fd) { gf_msg (this->name, GF_LOG_ERROR, errno, DHT_MSG_MIGRATE_FILE_FAILED, "%s: fd create failed", loc->path); *fop_errno = ENOMEM; ret = -1; goto out; } ret = syncop_create (old_target, loc, O_RDWR, DHT_LINKFILE_MODE, linkto_fd, NULL, dict, NULL); if (ret != 0 && -ret != EEXIST && -ret != ESTALE) { *fop_errno = -ret; ret = -1; gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "failed to create linkto file on %s in %s", loc->path, old_target->name); goto out; } else if (ret == 0) { ret = syncop_fsetattr (old_target, linkto_fd, &stbuf, (GF_SET_ATTR_UID | GF_SET_ATTR_GID), NULL, NULL, NULL, NULL); if (ret < 0) { *fop_errno = -ret; gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "chown failed for %s on %s", loc->path, old_target->name); } } } } clean_dst = _gf_false; /* Posix acls are not set on DHT linkto files as part of the initial * initial xattrs set on the dst file, so these need * to be set on the dst file after the linkto attrs are removed. * TODO: Optimize this. */ if (xattr) { dict_unref (xattr); xattr = NULL; } /* Set only the Posix ACLs this time */ ret = syncop_getxattr (from, loc, &xattr, POSIX_ACL_ACCESS_XATTR, NULL, NULL); if (ret < 0) { if ((-ret != ENODATA) && (-ret != ENOATTR)) { gf_msg (this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed:" "%s: failed to get xattr from %s", loc->path, from->name); *fop_errno = -ret; } } else { ret = syncop_setxattr (to, loc, xattr, 0, NULL, NULL); if (ret < 0) { /* Potential problem here where Posix ACLs will * not be set on the target file */ gf_msg (this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed:" "%s: failed to set xattr on %s", loc->path, to->name); *fop_errno = -ret; } } /* store size of previous migrated file */ if (defrag && defrag->tier_conf.is_tier) { if (from != TIER_HASHED_SUBVOL) { defrag->tier_conf.st_last_promoted_size = stbuf.ia_size; } else { /* Don't delete the linkto file on the hashed subvol */ delete_src_linkto = _gf_false; defrag->tier_conf.st_last_demoted_size = stbuf.ia_size; } } /* The src file is being unlinked after this so we don't need to clean it up */ clean_src = _gf_false; /* Make the source as a linkfile first before deleting it */ empty_iatt.ia_prot.sticky = 1; ret = syncop_fsetattr (from, src_fd, &empty_iatt, GF_SET_ATTR_MODE, NULL, NULL, NULL, NULL); if (ret) { gf_msg (this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed:" "%s: failed to perform setattr on %s ", loc->path, from->name); *fop_errno = -ret; ret = -1; goto metaunlock; } /* Free up the data blocks on the source node, as the whole file is migrated */ ret = syncop_ftruncate (from, src_fd, 0, NULL, NULL, NULL, NULL); if (ret) { gf_log (this->name, GF_LOG_WARNING, "%s: failed to perform truncate on %s (%s)", loc->path, from->name, strerror (-ret)); *fop_errno = -ret; } /* remove the 'linkto' xattr from the destination */ ret = syncop_fremovexattr (to, dst_fd, conf->link_xattr_name, 0, NULL); if (ret) { gf_log (this->name, GF_LOG_WARNING, "%s: failed to perform removexattr on %s (%s)", loc->path, to->name, strerror (-ret)); *fop_errno = -ret; } /* Do a stat and check the gfid before unlink */ /* * Cached file changes its state from non-linkto to linkto file after * migrating data. If lookup from any other mount-point is performed, * converted-linkto-cached file will be treated as a stale and will be * unlinked. But by this time, file is already migrated. So further * failure because of ENOENT should not be treated as error */ ret = syncop_stat (from, loc, &empty_iatt, NULL, NULL); if (ret) { gf_msg (this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "%s: failed to do a stat on %s", loc->path, from->name); if (-ret != ENOENT) { *fop_errno = -ret; ret = -1; goto metaunlock; } rcvd_enoent_from_src = 1; } if ((gf_uuid_compare (empty_iatt.ia_gfid, loc->gfid) == 0 ) && (!rcvd_enoent_from_src) && delete_src_linkto) { /* take out the source from namespace */ ret = syncop_unlink (from, loc, NULL, NULL); if (ret) { gf_msg (this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "%s: failed to perform unlink on %s", loc->path, from->name); *fop_errno = -ret; ret = -1; goto metaunlock; } } ret = syncop_lookup (this, loc, NULL, NULL, NULL, NULL); if (ret) { gf_msg_debug (this->name, -ret, "%s: failed to lookup the file on subvolumes", loc->path); *fop_errno = -ret; } gf_msg (this->name, log_level, 0, DHT_MSG_MIGRATE_FILE_COMPLETE, "completed migration of %s from subvolume %s to %s", loc->path, from->name, to->name); ret = 0; metaunlock: if (conf->lock_migration_enabled && meta_locked) { dict_del (meta_dict, GF_META_LOCK_KEY); ret = dict_set_int32 (meta_dict, GF_META_UNLOCK_KEY, 1); if (ret) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED, "Trace dict_set failed"); *fop_errno = ENOMEM; ret = -1; goto out; } if (clean_dst == _gf_false) ret = dict_set_int32 (meta_dict, "status", 1); else ret = dict_set_int32 (meta_dict, "status", 0); if (ret) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED, "Trace dict_set failed"); *fop_errno = ENOMEM; ret = -1; goto out; } ret = syncop_setxattr (from, loc, meta_dict, 0, NULL, NULL); if (ret) { gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "Trace syncop_setxattr meta unlock failed"); *fop_errno = -ret; ret = -1; goto out; } } out: if (clean_src) { /* Revert source mode and xattr changes*/ lk_ret = __dht_migration_cleanup_src_file (this, loc, src_fd, from, &src_ia_prot); if (lk_ret) { gf_msg (this->name, GF_LOG_WARNING, 0, DHT_MSG_MIGRATE_FILE_FAILED, "%s: failed to cleanup source file on %s", loc->path, from->name); } } /* reset the destination back to 0 */ if (clean_dst) { lk_ret = syncop_ftruncate (to, dst_fd, 0, NULL, NULL, NULL, NULL); if (lk_ret) { gf_msg (this->name, GF_LOG_ERROR, -lk_ret, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed: " "%s: failed to reset target size back to 0", loc->path); } } if (inodelk_locked) { flock.l_type = F_UNLCK; lk_ret = syncop_inodelk (from, DHT_FILE_MIGRATE_DOMAIN, &tmp_loc, F_SETLK, &flock, NULL, NULL); if (lk_ret < 0) { gf_msg (this->name, GF_LOG_WARNING, -lk_ret, DHT_MSG_MIGRATE_FILE_FAILED, "%s: failed to unlock file on %s", loc->path, from->name); } } if (entrylk_locked) { lk_ret = syncop_entrylk (hashed_subvol, DHT_ENTRY_SYNC_DOMAIN, &parent_loc, loc->name, ENTRYLK_UNLOCK, ENTRYLK_UNLOCK, NULL, NULL); if (lk_ret < 0) { gf_msg (this->name, GF_LOG_WARNING, -lk_ret, DHT_MSG_MIGRATE_FILE_FAILED, "%s: failed to unlock entrylk on %s", loc->path, hashed_subvol->name); } } if (p_locked) { plock.l_type = F_UNLCK; lk_ret = syncop_lk (from, src_fd, F_SETLK, &plock, NULL, NULL); if (lk_ret < 0) { gf_msg (this->name, GF_LOG_WARNING, -lk_ret, DHT_MSG_MIGRATE_FILE_FAILED, "%s: failed to unlock file on %s", loc->path, from->name); } } if (!dht_is_tier_xlator (this)) { lk_ret = syncop_removexattr (to, loc, GF_PROTECT_FROM_EXTERNAL_WRITES, NULL, NULL); if (lk_ret && (lk_ret != -ENODATA) && (lk_ret != -ENOATTR)) { gf_msg (this->name, GF_LOG_WARNING, -lk_ret, 0, "%s: removexattr failed key %s", loc->path, GF_PROTECT_FROM_EXTERNAL_WRITES); } } if (dict) dict_unref (dict); if (xattr) dict_unref (xattr); if (xattr_rsp) dict_unref (xattr_rsp); if (dst_fd) syncop_close (dst_fd); if (src_fd) syncop_close (src_fd); if (linkto_fd) syncop_close (linkto_fd); loc_wipe (&tmp_loc); loc_wipe (&parent_loc); return ret; } static int rebalance_task (void *data) { int ret = -1; dht_local_t *local = NULL; call_frame_t *frame = NULL; int fop_errno = 0; frame = data; local = frame->local; /* This function is 'synchrounous', hence if it returns, we are done with the task */ ret = dht_migrate_file (THIS, &local->loc, local->rebalance.from_subvol, local->rebalance.target_node, local->flags, &fop_errno); return ret; } static int rebalance_task_completion (int op_ret, call_frame_t *sync_frame, void *data) { int32_t op_errno = EINVAL; if (op_ret == -1) { /* Failure of migration process, mostly due to write process. as we can't preserve the exact errno, lets say there was no space to migrate-data */ op_errno = ENOSPC; } else if (op_ret == 1) { /* migration didn't happen, but is not a failure, let the user understand that he doesn't have permission to migrate the file. */ op_ret = -1; op_errno = EPERM; } else if (op_ret != 0) { op_errno = -op_ret; op_ret = -1; } DHT_STACK_UNWIND (setxattr, sync_frame, op_ret, op_errno, NULL); return 0; } int dht_start_rebalance_task (xlator_t *this, call_frame_t *frame) { int ret = -1; ret = synctask_new (this->ctx->env, rebalance_task, rebalance_task_completion, frame, frame); return ret; } int gf_listener_stop (xlator_t *this) { glusterfs_ctx_t *ctx = NULL; cmd_args_t *cmd_args = NULL; int ret = 0; ctx = this->ctx; GF_ASSERT (ctx); cmd_args = &ctx->cmd_args; if (cmd_args->sock_file) { ret = sys_unlink (cmd_args->sock_file); if (ret && (ENOENT == errno)) { ret = 0; } } if (ret) { gf_msg (this->name, GF_LOG_ERROR, errno, DHT_MSG_SOCKET_ERROR, "Failed to unlink listener " "socket %s", cmd_args->sock_file); } return ret; } void dht_build_root_inode (xlator_t *this, inode_t **inode) { inode_table_t *itable = NULL; uuid_t root_gfid = {0, }; itable = inode_table_new (0, this); if (!itable) return; root_gfid[15] = 1; *inode = inode_find (itable, root_gfid); } void dht_build_root_loc (inode_t *inode, loc_t *loc) { loc->path = "/"; loc->inode = inode; loc->inode->ia_type = IA_IFDIR; memset (loc->gfid, 0, 16); loc->gfid[15] = 1; } /* return values: 1 -> error, bug ignore and continue 0 -> proceed -1 -> error, handle it */ int32_t gf_defrag_handle_migrate_error (int32_t op_errno, gf_defrag_info_t *defrag) { int ret = 0; /* if errno is not ENOTCONN, we can still continue with rebalance process */ if (op_errno != ENOTCONN) { ret = 1; goto out; } if (op_errno == ENOTCONN) { /* Most probably mount point went missing (mostly due to a brick down), say rebalance failure to user, let him restart it if everything is fine */ defrag->defrag_status = GF_DEFRAG_STATUS_FAILED; ret = -1; goto out; } out: return ret; } static gf_boolean_t gf_defrag_pattern_match (gf_defrag_info_t *defrag, char *name, uint64_t size) { gf_defrag_pattern_list_t *trav = NULL; gf_boolean_t match = _gf_false; gf_boolean_t ret = _gf_false; GF_VALIDATE_OR_GOTO ("dht", defrag, out); trav = defrag->defrag_pattern; while (trav) { if (!fnmatch (trav->path_pattern, name, FNM_NOESCAPE)) { match = _gf_true; break; } trav = trav->next; } if ((match == _gf_true) && (size >= trav->size)) ret = _gf_true; out: return ret; } int dht_dfreaddirp_done (dht_dfoffset_ctx_t *offset_var, int cnt) { int i; int result = 1; for (i = 0; i < cnt; i++) { if (offset_var[i].readdir_done == 0) { result = 0; break; } } return result; } int static gf_defrag_ctx_subvols_init (dht_dfoffset_ctx_t *offset_var, xlator_t *this) { int i; dht_conf_t *conf = NULL; conf = this->private; if (!conf) return -1; for (i = 0; i < conf->local_subvols_cnt; i++) { offset_var[i].this = conf->local_subvols[i]; offset_var[i].offset = (off_t) 0; offset_var[i].readdir_done = 0; } return 0; } static int dht_get_first_non_null_index (subvol_nodeuuids_info_t *entry) { int i = 0; int index = 0; for (i = 0; i < entry->count; i++) { if (!gf_uuid_is_null (entry->elements[i].uuid)) { index = i; goto out; } } if (i == entry->count) { index = -1; } out: return index; } /* Return value * 0 : this node does not migrate the file * 1 : this node migrates the file * * Use the hash value of the gfid to determine which node will migrate files. * Using the gfid instead of the name also ensures that the same node handles * all hardlinks. */ int gf_defrag_should_i_migrate (xlator_t *this, int local_subvol_index, uuid_t gfid) { int ret = 0; int i = local_subvol_index; char *str = NULL; uint32_t hashval = 0; int32_t index = 0; dht_conf_t *conf = NULL; char buf[UUID_CANONICAL_FORM_LEN + 1] = {0, }; subvol_nodeuuids_info_t *entry = NULL; conf = this->private; /* Pure distribute. A subvol in this case will be handled by only one node */ entry = &(conf->local_nodeuuids[i]); if (entry->count == 1) { return 1; } str = uuid_utoa_r (gfid, buf); ret = dht_hash_compute (this, 0, str, &hashval); if (ret == 0) { index = (hashval % entry->count); if (entry->elements[index].info == REBAL_NODEUUID_MINE) { /* Index matches this node's nodeuuid.*/ ret = 1; goto out; } /* Brick down - some other node has to migrate these files*/ if (gf_uuid_is_null (entry->elements[index].uuid)) { /* Fall back to the first non-null index */ index = dht_get_first_non_null_index (entry); if (index == -1) { /* None of the bricks in the subvol are up. * CHILD_DOWN will kill the process soon */ return 0; } if (entry->elements[index].info == REBAL_NODEUUID_MINE) { /* Index matches this node's nodeuuid.*/ ret = 1; goto out; } } } out: return ret; } int gf_defrag_migrate_single_file (void *opaque) { xlator_t *this = NULL; dht_conf_t *conf = NULL; gf_defrag_info_t *defrag = NULL; int ret = 0; gf_dirent_t *entry = NULL; struct timeval start = {0,}; loc_t entry_loc = {0,}; loc_t *loc = NULL; struct iatt iatt = {0,}; dict_t *migrate_data = NULL; struct timeval end = {0,}; double elapsed = {0,}; struct dht_container *rebal_entry = NULL; inode_t *inode = NULL; xlator_t *hashed_subvol = NULL; xlator_t *cached_subvol = NULL; call_frame_t *statfs_frame = NULL; xlator_t *old_THIS = NULL; data_t *tmp = NULL; int fop_errno = 0; gf_dht_migrate_data_type_t rebal_type = GF_DHT_MIGRATE_DATA; char value[MAX_REBAL_TYPE_SIZE] = {0,}; struct iatt *iatt_ptr = NULL; gf_boolean_t update_skippedcount = _gf_true; int i = 0; rebal_entry = (struct dht_container *)opaque; if (!rebal_entry) { gf_log ("DHT", GF_LOG_ERROR, "rebal_entry is NULL"); ret = -1; goto out; } this = rebal_entry->this; conf = this->private; defrag = conf->defrag; loc = rebal_entry->parent_loc; migrate_data = rebal_entry->migrate_data; entry = rebal_entry->df_entry; iatt_ptr = &entry->d_stat; if (defrag->defrag_status != GF_DEFRAG_STATUS_STARTED) { ret = -1; goto out; } if (defrag->stats == _gf_true) { gettimeofday (&start, NULL); } if (defrag->defrag_pattern && (gf_defrag_pattern_match (defrag, entry->d_name, entry->d_stat.ia_size) == _gf_false)) { gf_log (this->name, GF_LOG_ERROR, "pattern_match failed"); goto out; } memset (&entry_loc, 0, sizeof (entry_loc)); ret = dht_build_child_loc (this, &entry_loc, loc, entry->d_name); if (ret) { LOCK (&defrag->lock); { defrag->total_failures += 1; } UNLOCK (&defrag->lock); ret = 0; gf_log (this->name, GF_LOG_ERROR, "Child loc build failed"); goto out; } if (!gf_defrag_should_i_migrate (this, rebal_entry->local_subvol_index, entry->d_stat.ia_gfid)) { gf_msg_debug (this->name, 0, "Don't migrate %s ", entry_loc.path); goto out; } gf_uuid_copy (entry_loc.gfid, entry->d_stat.ia_gfid); gf_uuid_copy (entry_loc.pargfid, loc->gfid); ret = syncop_lookup (this, &entry_loc, &iatt, NULL, NULL, NULL); if (ret) { gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed: %s lookup failed", entry_loc.path); /* Increase failure count only for remove-brick op, so that * user is warned to check the removed-brick for any files left * unmigrated */ if (conf->decommission_subvols_cnt) { LOCK (&defrag->lock); { defrag->total_failures += 1; } UNLOCK (&defrag->lock); } ret = 0; goto out; } iatt_ptr = &iatt; hashed_subvol = dht_subvol_get_hashed (this, &entry_loc); if (!hashed_subvol) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_HASHED_SUBVOL_GET_FAILED, "Failed to get hashed subvol for %s", entry_loc.path); ret = 0; goto out; } cached_subvol = dht_subvol_get_cached (this, entry_loc.inode); if (!cached_subvol) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_CACHED_SUBVOL_GET_FAILED, "Failed to get cached subvol for %s", entry_loc.path); ret = 0; goto out; } if (hashed_subvol == cached_subvol) { ret = 0; goto out; } inode = inode_link (entry_loc.inode, entry_loc.parent, entry->d_name, &iatt); inode_unref (entry_loc.inode); /* use the inode returned by inode_link */ entry_loc.inode = inode; old_THIS = THIS; THIS = this; statfs_frame = create_frame (this, this->ctx->pool); if (!statfs_frame) { gf_msg (this->name, GF_LOG_ERROR, DHT_MSG_NO_MEMORY, ENOMEM, "Insufficient memory. Frame creation failed"); ret = -1; goto out; } /* async statfs information for honoring min-free-disk */ dht_get_du_info (statfs_frame, this, loc); THIS = old_THIS; tmp = dict_get (migrate_data, GF_XATTR_FILE_MIGRATE_KEY); if (tmp) { memcpy (value, tmp->data, tmp->len); if (strcmp (value, "force") == 0) rebal_type = GF_DHT_MIGRATE_DATA_EVEN_IF_LINK_EXISTS; if (conf->decommission_in_progress) rebal_type = GF_DHT_MIGRATE_HARDLINK; } ret = dht_migrate_file (this, &entry_loc, cached_subvol, hashed_subvol, rebal_type, &fop_errno); if (ret == 1) { if (fop_errno == ENOSPC) { gf_msg_debug (this->name, 0, "migrate-data skipped for" " %s due to space constraints", entry_loc.path); /* For remove-brick case if the source is not one of the * removed-brick, do not mark the error as failure */ if (conf->decommission_subvols_cnt) { for (i = 0; i < conf->subvolume_cnt; i++) { if (conf->decommissioned_bricks[i] == cached_subvol) { LOCK (&defrag->lock); { defrag->total_failures += 1; update_skippedcount = _gf_false; } UNLOCK (&defrag->lock); break; } } } if (update_skippedcount) { LOCK (&defrag->lock); { defrag->skipped += 1; } UNLOCK (&defrag->lock); gf_msg (this->name, GF_LOG_INFO, 0, DHT_MSG_MIGRATE_FILE_SKIPPED, "File migration skipped for %s.", entry_loc.path); } } else if (fop_errno == ENOTSUP) { gf_msg_debug (this->name, 0, "migrate-data skipped for" " hardlink %s ", entry_loc.path); LOCK (&defrag->lock); { defrag->skipped += 1; } UNLOCK (&defrag->lock); gf_msg (this->name, GF_LOG_INFO, 0, DHT_MSG_MIGRATE_FILE_SKIPPED, "File migration skipped for %s.", entry_loc.path); } ret = 0; } else if (ret < 0) { if (fop_errno != EEXIST) { gf_msg (this->name, GF_LOG_ERROR, fop_errno, DHT_MSG_MIGRATE_FILE_FAILED, "migrate-data failed for %s", entry_loc.path); LOCK (&defrag->lock); { defrag->total_failures += 1; } UNLOCK (&defrag->lock); } ret = gf_defrag_handle_migrate_error (fop_errno, defrag); if (!ret) { gf_msg(this->name, GF_LOG_ERROR, fop_errno, DHT_MSG_MIGRATE_FILE_FAILED, "migrate-data on %s failed:", entry_loc.path); } else if (ret == 1) { ret = 0; } goto out; } LOCK (&defrag->lock); { defrag->total_files += 1; defrag->total_data += iatt.ia_size; } UNLOCK (&defrag->lock); if (defrag->stats == _gf_true) { gettimeofday (&end, NULL); elapsed = (end.tv_sec - start.tv_sec) * 1e6 + (end.tv_usec - start.tv_usec); gf_log (this->name, GF_LOG_INFO, "Migration of " "file:%s size:%"PRIu64" bytes took %.2f" "secs and ret: %d", entry_loc.name, iatt.ia_size, elapsed/1e6, ret); } out: if (statfs_frame) { STACK_DESTROY (statfs_frame->root); } if (iatt_ptr) { LOCK (&defrag->lock); { defrag->size_processed += iatt_ptr->ia_size; } UNLOCK (&defrag->lock); } loc_wipe (&entry_loc); return ret; } void * gf_defrag_task (void *opaque) { struct list_head *q_head = NULL; struct dht_container *iterator = NULL; gf_defrag_info_t *defrag = NULL; int ret = 0; pid_t pid = GF_CLIENT_PID_DEFRAG; defrag = (gf_defrag_info_t *)opaque; if (!defrag) { gf_msg ("dht", GF_LOG_ERROR, 0, 0, "defrag is NULL"); goto out; } syncopctx_setfspid (&pid); q_head = &(defrag->queue[0].list); /* The following while loop will dequeue one entry from the defrag->queue under lock. We will update the defrag->global_error only when there is an error which is critical to stop the rebalance process. The stop message will be intimated to other migrator threads by setting the defrag->defrag_status to GF_DEFRAG_STATUS_FAILED. In defrag->queue, a low watermark (MIN_MIGRATE_QUEUE_COUNT) is maintained so that crawler does not starve the file migration workers and a high watermark (MAX_MIGRATE_QUEUE_COUNT) so that crawler does not go far ahead in filling up the queue. */ while (_gf_true) { if (defrag->defrag_status != GF_DEFRAG_STATUS_STARTED) { pthread_cond_broadcast ( &defrag->rebalance_crawler_alarm); pthread_cond_broadcast ( &defrag->parallel_migration_cond); goto out; } pthread_mutex_lock (&defrag->dfq_mutex); { /*Throttle down: If the reconfigured count is less than current thread count, then the current thread will sleep */ /*TODO: Need to refactor the following block to work *under defrag->lock. For now access * defrag->current_thread_count and rthcount under * dfq_mutex lock */ while (!defrag->crawl_done && (defrag->recon_thread_count < defrag->current_thread_count)) { defrag->current_thread_count--; gf_msg_debug ("DHT", 0, "Thread sleeping. " "current thread count: %d", defrag->current_thread_count); pthread_cond_wait ( &defrag->df_wakeup_thread, &defrag->dfq_mutex); defrag->current_thread_count++; gf_msg_debug ("DHT", 0, "Thread wokeup. " "current thread count: %d", defrag->current_thread_count); } if (defrag->q_entry_count) { iterator = list_entry (q_head->next, typeof(*iterator), list); gf_msg_debug ("DHT", 0, "picking entry " "%s", iterator->df_entry->d_name); list_del_init (&(iterator->list)); defrag->q_entry_count--; if ((defrag->q_entry_count < MIN_MIGRATE_QUEUE_COUNT) && defrag->wakeup_crawler) { pthread_cond_broadcast ( &defrag->rebalance_crawler_alarm); } pthread_mutex_unlock (&defrag->dfq_mutex); ret = gf_defrag_migrate_single_file ((void *)iterator); /*Critical errors: ENOTCONN and ENOSPACE*/ if (ret) { dht_set_global_defrag_error (defrag, ret); defrag->defrag_status = GF_DEFRAG_STATUS_FAILED; pthread_cond_broadcast ( &defrag->rebalance_crawler_alarm); pthread_cond_broadcast ( &defrag->parallel_migration_cond); goto out; } gf_defrag_free_container (iterator); continue; } else { /* defrag->crawl_done flag is set means crawling file system is done and hence a list_empty when the above flag is set indicates there are no more entries to be added to the queue and rebalance is finished */ if (!defrag->crawl_done) { defrag->current_thread_count--; gf_msg_debug ("DHT", 0, "Thread " "sleeping while waiting " "for migration entries. " "current thread count:%d", defrag->current_thread_count); pthread_cond_wait ( &defrag->parallel_migration_cond, &defrag->dfq_mutex); } if (defrag->crawl_done && !defrag->q_entry_count) { defrag->current_thread_count++; gf_msg_debug ("DHT", 0, "Exiting thread"); pthread_cond_broadcast ( &defrag->parallel_migration_cond); goto unlock; } else { defrag->current_thread_count++; gf_msg_debug ("DHT", 0, "Thread woke up" " as found migrating entries. " "current thread count:%d", defrag->current_thread_count); pthread_mutex_unlock (&defrag->dfq_mutex); continue; } } } unlock: pthread_mutex_unlock (&defrag->dfq_mutex); break; } out: return NULL; } int static gf_defrag_get_entry (xlator_t *this, int i, struct dht_container **container, loc_t *loc, dht_conf_t *conf, gf_defrag_info_t *defrag, fd_t *fd, dict_t *migrate_data, struct dir_dfmeta *dir_dfmeta, dict_t *xattr_req, int *should_commit_hash, int *perrno) { int ret = -1; char is_linkfile = 0; gf_dirent_t *df_entry = NULL; struct dht_container *tmp_container = NULL; if (defrag->defrag_status != GF_DEFRAG_STATUS_STARTED) { ret = -1; goto out; } if (dir_dfmeta->offset_var[i].readdir_done == 1) { ret = 0; goto out; } if (dir_dfmeta->fetch_entries[i] == 1) { ret = syncop_readdirp (conf->local_subvols[i], fd, 131072, dir_dfmeta->offset_var[i].offset, &(dir_dfmeta->equeue[i]), xattr_req, NULL); if (ret == 0) { dir_dfmeta->offset_var[i].readdir_done = 1; ret = 0; goto out; } if (ret < 0) { gf_msg (this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_DATA_FAILED, "Readdirp failed. Aborting data migration for " "directory: %s", loc->path); *perrno = -ret; ret = -1; goto out; } if (list_empty (&(dir_dfmeta->equeue[i].list))) { dir_dfmeta->offset_var[i].readdir_done = 1; ret = 0; goto out; } dir_dfmeta->fetch_entries[i] = 0; } while (1) { if (defrag->defrag_status != GF_DEFRAG_STATUS_STARTED) { ret = -1; goto out; } df_entry = list_entry (dir_dfmeta->iterator[i]->next, typeof (*df_entry), list); if (&df_entry->list == dir_dfmeta->head[i]) { gf_dirent_free (&(dir_dfmeta->equeue[i])); INIT_LIST_HEAD (&(dir_dfmeta->equeue[i].list)); dir_dfmeta->fetch_entries[i] = 1; dir_dfmeta->iterator[i] = dir_dfmeta->head[i]; ret = 0; goto out; } dir_dfmeta->iterator[i] = dir_dfmeta->iterator[i]->next; dir_dfmeta->offset_var[i].offset = df_entry->d_off; if (!strcmp (df_entry->d_name, ".") || !strcmp (df_entry->d_name, "..")) continue; if (IA_ISDIR (df_entry->d_stat.ia_type)) { defrag->size_processed += df_entry->d_stat.ia_size; continue; } defrag->num_files_lookedup++; if (defrag->defrag_pattern && (gf_defrag_pattern_match (defrag, df_entry->d_name, df_entry->d_stat.ia_size) == _gf_false)) { defrag->size_processed += df_entry->d_stat.ia_size; continue; } is_linkfile = check_is_linkfile (NULL, &df_entry->d_stat, df_entry->dict, conf->link_xattr_name); if (is_linkfile) { /* No need to add linkto file to the queue for migration. Only the actual data file need to be checked for migration criteria. */ gf_msg_debug (this->name, 0, "Skipping linkfile" " %s on subvol: %s", df_entry->d_name, conf->local_subvols[i]->name); continue; } /*Build Container Structure */ tmp_container = GF_CALLOC (1, sizeof(struct dht_container), gf_dht_mt_container_t); if (!tmp_container) { gf_log (this->name, GF_LOG_ERROR, "Failed to allocate " "memory for container"); ret = -1; goto out; } tmp_container->df_entry = gf_dirent_for_name (df_entry->d_name); if (!tmp_container->df_entry) { gf_log (this->name, GF_LOG_ERROR, "Failed to allocate " "memory for df_entry"); ret = -1; goto out; } tmp_container->local_subvol_index = i; tmp_container->df_entry->d_stat = df_entry->d_stat; tmp_container->df_entry->d_ino = df_entry->d_ino; tmp_container->df_entry->d_type = df_entry->d_type; tmp_container->df_entry->d_len = df_entry->d_len; tmp_container->parent_loc = GF_CALLOC(1, sizeof(*loc), gf_dht_mt_loc_t); if (!tmp_container->parent_loc) { gf_log (this->name, GF_LOG_ERROR, "Failed to allocate " "memory for loc"); ret = -1; goto out; } ret = loc_copy (tmp_container->parent_loc, loc); if (ret) { gf_log (this->name, GF_LOG_ERROR, "loc_copy failed"); ret = -1; goto out; } tmp_container->migrate_data = migrate_data; tmp_container->this = this; if (df_entry->dict) tmp_container->df_entry->dict = dict_ref (df_entry->dict); /*Build Container Structure >> END*/ ret = 0; goto out; } out: if (ret == 0) { *container = tmp_container; } else { if (tmp_container) { gf_defrag_free_container (tmp_container); } } return ret; } int gf_defrag_process_dir (xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, dict_t *migrate_data, int *perrno) { int ret = -1; fd_t *fd = NULL; dht_conf_t *conf = NULL; gf_dirent_t entries; dict_t *xattr_req = NULL; struct timeval dir_start = {0,}; struct timeval end = {0,}; double elapsed = {0,}; int local_subvols_cnt = 0; int i = 0; int j = 0; struct dht_container *container = NULL; int ldfq_count = 0; int dfc_index = 0; int throttle_up = 0; struct dir_dfmeta *dir_dfmeta = NULL; int should_commit_hash = 1; gf_log (this->name, GF_LOG_INFO, "migrate data called on %s", loc->path); gettimeofday (&dir_start, NULL); conf = this->private; local_subvols_cnt = conf->local_subvols_cnt; if (!local_subvols_cnt) { ret = 0; goto out; } fd = fd_create (loc->inode, defrag->pid); if (!fd) { gf_log (this->name, GF_LOG_ERROR, "Failed to create fd"); ret = -1; goto out; } ret = syncop_opendir (this, loc, fd, NULL, NULL); if (ret) { gf_msg (this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_DATA_FAILED, "Migrate data failed: Failed to open dir %s", loc->path); *perrno = -ret; ret = -1; goto out; } fd_bind (fd); dir_dfmeta = GF_CALLOC (1, sizeof (*dir_dfmeta), gf_common_mt_pointer); if (!dir_dfmeta) { gf_log (this->name, GF_LOG_ERROR, "dir_dfmeta is NULL"); ret = -1; goto out; } dir_dfmeta->head = GF_CALLOC (local_subvols_cnt, sizeof (*(dir_dfmeta->head)), gf_common_mt_pointer); if (!dir_dfmeta->head) { gf_log (this->name, GF_LOG_ERROR, "dir_dfmeta->head is NULL"); ret = -1; goto out; } dir_dfmeta->iterator = GF_CALLOC (local_subvols_cnt, sizeof (*(dir_dfmeta->iterator)), gf_common_mt_pointer); if (!dir_dfmeta->iterator) { gf_log (this->name, GF_LOG_ERROR, "dir_dfmeta->iterator is NULL"); ret = -1; goto out; } dir_dfmeta->equeue = GF_CALLOC (local_subvols_cnt, sizeof (entries), gf_dht_mt_dirent_t); if (!dir_dfmeta->equeue) { gf_log (this->name, GF_LOG_ERROR, "dir_dfmeta->equeue is NULL"); ret = -1; goto out; } dir_dfmeta->offset_var = GF_CALLOC (local_subvols_cnt, sizeof (dht_dfoffset_ctx_t), gf_dht_mt_octx_t); if (!dir_dfmeta->offset_var) { gf_log (this->name, GF_LOG_ERROR, "dir_dfmeta->offset_var is NULL"); ret = -1; goto out; } ret = gf_defrag_ctx_subvols_init (dir_dfmeta->offset_var, this); if (ret) { gf_log (this->name, GF_LOG_ERROR, "dht_dfoffset_ctx_t" "initialization failed"); ret = -1; goto out; } dir_dfmeta->fetch_entries = GF_CALLOC (local_subvols_cnt, sizeof (int), gf_common_mt_int); if (!dir_dfmeta->fetch_entries) { gf_log (this->name, GF_LOG_ERROR, "dir_dfmeta->fetch_entries is NULL"); ret = -1; goto out; } for (i = 0; i < local_subvols_cnt ; i++) { INIT_LIST_HEAD (&(dir_dfmeta->equeue[i].list)); dir_dfmeta->head[i] = &(dir_dfmeta->equeue[i].list); dir_dfmeta->iterator[i] = dir_dfmeta->head[i]; dir_dfmeta->fetch_entries[i] = 1; } xattr_req = dict_new (); if (!xattr_req) { gf_log (this->name, GF_LOG_ERROR, "dict_new failed"); ret = -1; goto out; } ret = dict_set_uint32 (xattr_req, conf->link_xattr_name, 256); if (ret) { gf_log (this->name, GF_LOG_ERROR, "failed to set dict for " "key: %s", conf->link_xattr_name); ret = -1; goto out; } /* Job: Read entries from each local subvol and store the entries in equeue array of linked list. Now pick one entry from the equeue array in a round robin basis and add them to defrag Queue. */ while (!dht_dfreaddirp_done(dir_dfmeta->offset_var, local_subvols_cnt)) { pthread_mutex_lock (&defrag->dfq_mutex); { /*Throttle up: If reconfigured count is higher than current thread count, wake up the sleeping threads TODO: Need to refactor this. Instead of making the thread sleep and wake, we should terminate and spawn threads on-demand*/ if (defrag->recon_thread_count > defrag->current_thread_count) { throttle_up = (defrag->recon_thread_count - defrag->current_thread_count); for (j = 0; j < throttle_up; j++) { pthread_cond_signal ( &defrag->df_wakeup_thread); } } while (defrag->q_entry_count > MAX_MIGRATE_QUEUE_COUNT) { defrag->wakeup_crawler = 1; pthread_cond_wait ( &defrag->rebalance_crawler_alarm, &defrag->dfq_mutex); } ldfq_count = defrag->q_entry_count; if (defrag->wakeup_crawler) { defrag->wakeup_crawler = 0; } } pthread_mutex_unlock (&defrag->dfq_mutex); while (ldfq_count <= MAX_MIGRATE_QUEUE_COUNT && !dht_dfreaddirp_done(dir_dfmeta->offset_var, local_subvols_cnt)) { ret = gf_defrag_get_entry (this, dfc_index, &container, loc, conf, defrag, fd, migrate_data, dir_dfmeta, xattr_req, &should_commit_hash, perrno); if (ret) { gf_log (this->name, GF_LOG_WARNING, "Found " "error from gf_defrag_get_entry"); ret = -1; goto out; } /* Check if we got an entry, else we need to move the index to the next subvol */ if (!container) { GF_CRAWL_INDEX_MOVE(dfc_index, local_subvols_cnt); continue; } /* Q this entry in the dfq */ pthread_mutex_lock (&defrag->dfq_mutex); { list_add_tail (&container->list, &(defrag->queue[0].list)); defrag->q_entry_count++; ldfq_count = defrag->q_entry_count; gf_msg_debug (this->name, 0, "added " "file:%s parent:%s to the queue ", container->df_entry->d_name, container->parent_loc->path); pthread_cond_signal ( &defrag->parallel_migration_cond); } pthread_mutex_unlock (&defrag->dfq_mutex); GF_CRAWL_INDEX_MOVE(dfc_index, local_subvols_cnt); } } gettimeofday (&end, NULL); elapsed = (end.tv_sec - dir_start.tv_sec) * 1e6 + (end.tv_usec - dir_start.tv_usec); gf_log (this->name, GF_LOG_INFO, "Migration operation on dir %s took " "%.2f secs", loc->path, elapsed/1e6); ret = 0; out: gf_defrag_free_dir_dfmeta (dir_dfmeta, local_subvols_cnt); if (xattr_req) dict_unref(xattr_req); if (fd) fd_unref (fd); if (ret == 0 && should_commit_hash == 0) { ret = 2; } /* It does not matter if it errored out - this number is * used to calculate rebalance estimated time to complete. * No locking required as dirs are processed by a single thread. */ defrag->num_dirs_processed++; return ret; } int gf_defrag_settle_hash (xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, dict_t *fix_layout) { int ret; dht_conf_t *conf = NULL; /* * Now we're ready to update the directory commit hash for the volume * root, so that hash miscompares and broadcast lookups can stop. * However, we want to skip that if fix-layout is all we did. In * that case, we want the miscompares etc. to continue until a real * rebalance is complete. */ if (defrag->cmd == GF_DEFRAG_CMD_START_LAYOUT_FIX || defrag->cmd == GF_DEFRAG_CMD_START_DETACH_TIER || defrag->cmd == GF_DEFRAG_CMD_DETACH_START) { return 0; } conf = this->private; if (!conf) { /*Uh oh */ return -1; } if (conf->local_subvols_cnt == 0 || !conf->lookup_optimize) { /* Commit hash updates are only done on local subvolumes and * only when lookup optimization is needed (for older client * support) */ return 0; } ret = dict_set_uint32 (fix_layout, "new-commit-hash", defrag->new_commit_hash); if (ret) { gf_log (this->name, GF_LOG_ERROR, "Failed to set new-commit-hash"); return -1; } ret = syncop_setxattr (this, loc, fix_layout, 0, NULL, NULL); if (ret) { gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_LAYOUT_FIX_FAILED, "fix layout on %s failed", loc->path); if (-ret == ENOENT || -ret == ESTALE) { /* Dir most likely is deleted */ return 0; } return -1; } /* TBD: find more efficient solution than adding/deleting every time */ dict_del(fix_layout, "new-commit-hash"); return 0; } /* Function for doing a named lookup on file inodes during an attach tier * So that a hardlink lookup heal i.e gfid to parent gfid lookup heal * happens on pre-existing data. This is required so that the ctr database has * hardlinks of all the exisitng file in the volume. CTR xlator on the * brick/server side does db update/insert of the hardlink on a namelookup. * Currently the namedlookup is done synchronous to the fixlayout that is * triggered by attach tier. This is not performant, adding more time to * fixlayout. The performant approach is record the hardlinks on a compressed * datastore and then do the namelookup asynchronously later, giving the ctr db * eventual consistency * */ int gf_fix_layout_tier_attach_lookup (xlator_t *this, loc_t *parent_loc, gf_dirent_t *file_dentry) { int ret = -1; dict_t *lookup_xdata = NULL; dht_conf_t *conf = NULL; loc_t file_loc = {0,}; struct iatt iatt = {0,}; GF_VALIDATE_OR_GOTO ("tier", this, out); GF_VALIDATE_OR_GOTO (this->name, parent_loc, out); GF_VALIDATE_OR_GOTO (this->name, file_dentry, out); GF_VALIDATE_OR_GOTO (this->name, this->private, out); if (!parent_loc->inode) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_LOG_TIER_ERROR, "%s/%s parent is NULL", parent_loc->path, file_dentry->d_name); goto out; } conf = this->private; loc_wipe (&file_loc); if (gf_uuid_is_null (file_dentry->d_stat.ia_gfid)) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_LOG_TIER_ERROR, "%s/%s gfid not present", parent_loc->path, file_dentry->d_name); goto out; } gf_uuid_copy (file_loc.gfid, file_dentry->d_stat.ia_gfid); if (gf_uuid_is_null (parent_loc->gfid)) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_LOG_TIER_ERROR, "%s/%s" " gfid not present", parent_loc->path, file_dentry->d_name); goto out; } gf_uuid_copy (file_loc.pargfid, parent_loc->gfid); ret = dht_build_child_loc (this, &file_loc, parent_loc, file_dentry->d_name); if (ret) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_LOG_TIER_ERROR, "Child loc build failed"); ret = -1; goto out; } lookup_xdata = dict_new (); if (!lookup_xdata) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_LOG_TIER_ERROR, "Failed creating lookup dict for %s", file_dentry->d_name); goto out; } ret = dict_set_int32 (lookup_xdata, CTR_ATTACH_TIER_LOOKUP, 1); if (ret) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_LOG_TIER_ERROR, "Failed to set lookup flag"); goto out; } gf_uuid_copy (file_loc.parent->gfid, parent_loc->gfid); /* Sending lookup to cold tier only */ ret = syncop_lookup (conf->subvolumes[0], &file_loc, &iatt, NULL, lookup_xdata, NULL); if (ret) { /* If the file does not exist on the cold tier than it must */ /* have been discovered on the hot tier. This is not an error. */ gf_msg (this->name, GF_LOG_INFO, 0, DHT_MSG_LOG_TIER_STATUS, "%s lookup to cold tier on attach heal failed", file_loc.path); goto out; } ret = 0; out: loc_wipe (&file_loc); if (lookup_xdata) dict_unref (lookup_xdata); return ret; } int gf_defrag_fix_layout (xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, dict_t *fix_layout, dict_t *migrate_data) { int ret = -1; loc_t entry_loc = {0,}; fd_t *fd = NULL; gf_dirent_t entries; gf_dirent_t *tmp = NULL; gf_dirent_t *entry = NULL; gf_boolean_t free_entries = _gf_false; off_t offset = 0; struct iatt iatt = {0,}; inode_t *linked_inode = NULL, *inode = NULL; dht_conf_t *conf = NULL; int should_commit_hash = 1; int perrno = 0; conf = this->private; if (!conf) { ret = -1; goto out; } ret = syncop_lookup (this, loc, &iatt, NULL, NULL, NULL); if (ret) { if (strcmp (loc->path, "/") == 0) { gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_DIR_LOOKUP_FAILED, "lookup failed for:%s", loc->path); defrag->total_failures++; ret = -1; goto out; } if (-ret == ENOENT || -ret == ESTALE) { gf_msg (this->name, GF_LOG_INFO, -ret, DHT_MSG_DIR_LOOKUP_FAILED, "Dir:%s renamed or removed. Skipping", loc->path); if (conf->decommission_subvols_cnt) { defrag->total_failures++; } ret = 0; goto out; } else { gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_DIR_LOOKUP_FAILED, "lookup failed for:%s", loc->path); defrag->total_failures++; goto out; } } fd = fd_create (loc->inode, defrag->pid); if (!fd) { gf_log (this->name, GF_LOG_ERROR, "Failed to create fd"); ret = -1; goto out; } ret = syncop_opendir (this, loc, fd, NULL, NULL); if (ret) { if (-ret == ENOENT || -ret == ESTALE) { if (conf->decommission_subvols_cnt) { defrag->total_failures++; } ret = 0; goto out; } gf_log (this->name, GF_LOG_ERROR, "Failed to open dir %s, " "err:%d", loc->path, -ret); ret = -1; goto out; } fd_bind (fd); INIT_LIST_HEAD (&entries.list); while ((ret = syncop_readdirp (this, fd, 131072, offset, &entries, NULL, NULL)) != 0) { if (ret < 0) { if (-ret == ENOENT || -ret == ESTALE) { if (conf->decommission_subvols_cnt) { defrag->total_failures++; } ret = 0; goto out; } gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_READDIR_ERROR, "readdirp failed for " "path %s. Aborting fix-layout", loc->path); ret = -1; goto out; } if (list_empty (&entries.list)) break; free_entries = _gf_true; list_for_each_entry_safe (entry, tmp, &entries.list, list) { if (defrag->defrag_status != GF_DEFRAG_STATUS_STARTED) { ret = 1; goto out; } offset = entry->d_off; if (!strcmp (entry->d_name, ".") || !strcmp (entry->d_name, "..")) continue; if (!IA_ISDIR (entry->d_stat.ia_type)) { /* If its a fix layout during the attach * tier operation do lookups on files * on cold subvolume so that there is a * CTR DB Lookup Heal triggered on existing * data. * */ if (defrag->cmd == GF_DEFRAG_CMD_START_TIER) { gf_fix_layout_tier_attach_lookup (this, loc, entry); } continue; } loc_wipe (&entry_loc); ret = dht_build_child_loc (this, &entry_loc, loc, entry->d_name); if (ret) { gf_log (this->name, GF_LOG_ERROR, "Child loc" " build failed for entry: %s", entry->d_name); if (conf->decommission_in_progress) { defrag->defrag_status = GF_DEFRAG_STATUS_FAILED; goto out; } else { should_commit_hash = 0; continue; } } if (gf_uuid_is_null (entry->d_stat.ia_gfid)) { gf_log (this->name, GF_LOG_ERROR, "%s/%s" " gfid not present", loc->path, entry->d_name); continue; } gf_uuid_copy (entry_loc.gfid, entry->d_stat.ia_gfid); /*In case the gfid stored in the inode by inode_link * and the gfid obtained in the lookup differs, then * client3_3_lookup_cbk will return ESTALE and proper * error will be captured */ linked_inode = inode_link (entry_loc.inode, loc->inode, entry->d_name, &entry->d_stat); inode = entry_loc.inode; entry_loc.inode = linked_inode; inode_unref (inode); if (gf_uuid_is_null (loc->gfid)) { gf_log (this->name, GF_LOG_ERROR, "%s/%s" " gfid not present", loc->path, entry->d_name); continue; } gf_uuid_copy (entry_loc.pargfid, loc->gfid); ret = syncop_lookup (this, &entry_loc, &iatt, NULL, NULL, NULL); if (ret) { if (-ret == ENOENT || -ret == ESTALE) { gf_msg (this->name, GF_LOG_INFO, -ret, DHT_MSG_DIR_LOOKUP_FAILED, "Dir:%s renamed or removed. " "Skipping", loc->path); ret = 0; if (conf->decommission_subvols_cnt) { defrag->total_failures++; } continue; } else { gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_DIR_LOOKUP_FAILED, "lookup failed for:%s", entry_loc.path); defrag->total_failures++; if (conf->decommission_in_progress) { defrag->defrag_status = GF_DEFRAG_STATUS_FAILED; ret = -1; goto out; } else { should_commit_hash = 0; continue; } } } /* A return value of 2 means, either process_dir or * lookup of a dir failed. Hence, don't commit hash * for the current directory*/ ret = gf_defrag_fix_layout (this, defrag, &entry_loc, fix_layout, migrate_data); if (ret && ret != 2) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_LAYOUT_FIX_FAILED, "Fix layout failed for %s", entry_loc.path); defrag->total_failures++; if (conf->decommission_in_progress) { defrag->defrag_status = GF_DEFRAG_STATUS_FAILED; goto out; } else { /* Let's not commit-hash if * gf_defrag_fix_layout failed*/ continue; } } } gf_dirent_free (&entries); free_entries = _gf_false; INIT_LIST_HEAD (&entries.list); } ret = syncop_setxattr (this, loc, fix_layout, 0, NULL, NULL); if (ret) { if (-ret == ENOENT || -ret == ESTALE) { gf_msg (this->name, GF_LOG_INFO, -ret, DHT_MSG_LAYOUT_FIX_FAILED, "Setxattr failed. Dir %s " "renamed or removed", loc->path); if (conf->decommission_subvols_cnt) { defrag->total_failures++; } ret = 0; } else { gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_LAYOUT_FIX_FAILED, "Setxattr failed for %s", loc->path); defrag->total_failures++; if (conf->decommission_in_progress) { defrag->defrag_status = GF_DEFRAG_STATUS_FAILED; ret = -1; goto out; } } } if ((defrag->cmd != GF_DEFRAG_CMD_START_TIER) && (defrag->cmd != GF_DEFRAG_CMD_START_LAYOUT_FIX)) { ret = gf_defrag_process_dir (this, defrag, loc, migrate_data, &perrno); if (ret && (ret != 2)) { if (perrno == ENOENT || perrno == ESTALE) { ret = 0; goto out; } else { defrag->total_failures++; gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_DEFRAG_PROCESS_DIR_FAILED, "gf_defrag_process_dir failed for " "directory: %s", loc->path); if (conf->decommission_in_progress) { goto out; } should_commit_hash = 0; } } else if (ret == 2) { should_commit_hash = 0; } } gf_msg_trace (this->name, 0, "fix layout called on %s", loc->path); if (should_commit_hash && gf_defrag_settle_hash (this, defrag, loc, fix_layout) != 0) { defrag->total_failures++; gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_SETTLE_HASH_FAILED, "Settle hash failed for %s", loc->path); ret = -1; if (conf->decommission_in_progress) { defrag->defrag_status = GF_DEFRAG_STATUS_FAILED; goto out; } } ret = 0; out: if (free_entries) gf_dirent_free (&entries); loc_wipe (&entry_loc); if (fd) fd_unref (fd); if (ret == 0 && should_commit_hash == 0) { ret = 2; } return ret; } /****************************************************************************** * Tier background Fix layout functions ******************************************************************************/ /* This is the background tier fixlayout thread */ void * gf_tier_do_fix_layout (void *args) { gf_tier_fix_layout_arg_t *tier_fix_layout_arg = args; int ret = -1; xlator_t *this = NULL; dht_conf_t *conf = NULL; gf_defrag_info_t *defrag = NULL; dict_t *dict = NULL; loc_t loc = {0,}; struct iatt iatt = {0,}; struct iatt parent = {0,}; GF_VALIDATE_OR_GOTO ("tier", tier_fix_layout_arg, out); GF_VALIDATE_OR_GOTO ("tier", tier_fix_layout_arg->this, out); this = tier_fix_layout_arg->this; conf = this->private; GF_VALIDATE_OR_GOTO (this->name, conf, out); defrag = conf->defrag; GF_VALIDATE_OR_GOTO (this->name, defrag, out); GF_VALIDATE_OR_GOTO (this->name, defrag->root_inode, out); GF_VALIDATE_OR_GOTO (this->name, tier_fix_layout_arg->fix_layout, out); /* Get Root loc_t */ dht_build_root_loc (defrag->root_inode, &loc); ret = syncop_lookup (this, &loc, &iatt, &parent, NULL, NULL); if (ret) { gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_REBALANCE_START_FAILED, "Lookup on root failed."); ret = -1; goto out; } /* Start the crawl */ gf_msg (this->name, GF_LOG_INFO, 0, DHT_MSG_LOG_TIER_STATUS, "Tiering Fixlayout started"); ret = gf_defrag_fix_layout (this, defrag, &loc, tier_fix_layout_arg->fix_layout, NULL); if (ret && ret != 2) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_REBALANCE_FAILED, "Tiering fixlayout failed."); ret = -1; goto out; } if (ret != 2 && gf_defrag_settle_hash (this, defrag, &loc, tier_fix_layout_arg->fix_layout) != 0) { defrag->total_failures++; ret = -1; goto out; } dict = dict_new (); if (!dict) { ret = -1; goto out; } ret = dict_set_str (dict, GF_XATTR_TIER_LAYOUT_FIXED_KEY, "yes"); if (ret) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_REBALANCE_FAILED, "Failed to set dictionary value: key = %s", GF_XATTR_TIER_LAYOUT_FIXED_KEY); ret = -1; goto out; } /* Marking the completion of tiering fix layout via a xattr on root */ ret = syncop_setxattr (this, &loc, dict, 0, NULL, NULL); if (ret) { gf_log (this->name, GF_LOG_ERROR, "Failed to set tiering fix " "layout completed xattr on %s", loc.path); ret = -1; goto out; } ret = 0; out: if (ret && defrag) defrag->total_failures++; if (dict) dict_unref (dict); return NULL; } int gf_tier_start_fix_layout (xlator_t *this, loc_t *loc, gf_defrag_info_t *defrag, dict_t *fix_layout) { int ret = -1; dict_t *tier_dict = NULL; gf_tier_fix_layout_arg_t *tier_fix_layout_arg = NULL; tier_dict = dict_new (); if (!tier_dict) { gf_log ("tier", GF_LOG_ERROR, "Tier fix layout failed :" "Creation of tier_dict failed"); ret = -1; goto out; } /* Check if layout is fixed already */ ret = syncop_getxattr (this, loc, &tier_dict, GF_XATTR_TIER_LAYOUT_FIXED_KEY, NULL, NULL); if (ret != 0) { tier_fix_layout_arg = &defrag->tier_conf.tier_fix_layout_arg; /*Fill crawl arguments */ tier_fix_layout_arg->this = this; tier_fix_layout_arg->fix_layout = fix_layout; /* Spawn the fix layout thread so that its done in the * background */ ret = gf_thread_create (&tier_fix_layout_arg->thread_id, NULL, gf_tier_do_fix_layout, tier_fix_layout_arg, "tierfixl"); if (ret) { gf_log ("tier", GF_LOG_ERROR, "Thread creation failed. " "Background fix layout for tiering will not " "work."); defrag->total_failures++; goto out; } } ret = 0; out: if (tier_dict) dict_unref (tier_dict); return ret; } void gf_tier_clear_fix_layout (xlator_t *this, loc_t *loc, gf_defrag_info_t *defrag) { int ret = -1; dict_t *dict = NULL; GF_VALIDATE_OR_GOTO ("tier", this, out); GF_VALIDATE_OR_GOTO (this->name, loc, out); GF_VALIDATE_OR_GOTO (this->name, defrag, out); /* Check if background fixlayout is completed. This is not * multi-process safe i.e there is a possibility that by the time * we move to remove the xattr there it might have been cleared by some * other detach process from other node. We ignore the error if such * a thing happens */ ret = syncop_getxattr (this, loc, &dict, GF_XATTR_TIER_LAYOUT_FIXED_KEY, NULL, NULL); if (ret) { /* Background fixlayout not complete - nothing to clear*/ gf_msg (this->name, GF_LOG_WARNING, -ret, DHT_MSG_LOG_TIER_STATUS, "Unable to retrieve fixlayout xattr." "Assume background fix layout not complete"); goto out; } ret = syncop_removexattr (this, loc, GF_XATTR_TIER_LAYOUT_FIXED_KEY, NULL, NULL); if (ret) { gf_msg (this->name, GF_LOG_WARNING, -ret, DHT_MSG_LOG_TIER_STATUS, "Failed removing tier fix layout " "xattr from %s", loc->path); goto out; } ret = 0; out: if (dict) dict_unref (dict); } void gf_tier_wait_fix_lookup (gf_defrag_info_t *defrag) { if (defrag->tier_conf.tier_fix_layout_arg.thread_id) { pthread_join (defrag->tier_conf.tier_fix_layout_arg.thread_id, NULL); } } /******************Tier background Fix layout functions END********************/ int dht_init_local_subvols_and_nodeuuids (xlator_t *this, dht_conf_t *conf, loc_t *loc) { dict_t *dict = NULL; gf_defrag_info_t *defrag = NULL; uuid_t *uuid_ptr = NULL; int ret = -1; int i = 0; int j = 0; defrag = conf->defrag; if (defrag->cmd != GF_DEFRAG_CMD_START_TIER) { /* Find local subvolumes */ ret = syncop_getxattr (this, loc, &dict, GF_REBAL_FIND_LOCAL_SUBVOL, NULL, NULL); if (ret && (ret != -ENODATA)) { gf_msg (this->name, GF_LOG_ERROR, -ret, 0, "local " "subvolume determination failed with error: %d", -ret); ret = -1; goto out; } if (!ret) goto out; } ret = syncop_getxattr (this, loc, &dict, GF_REBAL_OLD_FIND_LOCAL_SUBVOL, NULL, NULL); if (ret) { gf_msg (this->name, GF_LOG_ERROR, -ret, 0, "local " "subvolume determination failed with error: %d", -ret); ret = -1; goto out; } ret = 0; out: if (ret) { return ret; } for (i = 0 ; i < conf->local_subvols_cnt; i++) { gf_msg (this->name, GF_LOG_INFO, 0, 0, "local subvol: " "%s", conf->local_subvols[i]->name); for (j = 0; j < conf->local_nodeuuids[i].count; j++) { uuid_ptr = &(conf->local_nodeuuids[i].elements[j].uuid); gf_msg (this->name, GF_LOG_INFO, 0, 0, "node uuid : %s", uuid_utoa(*uuid_ptr)); } } return ret; } /* Functions for the rebalance estimates feature */ uint64_t gf_defrag_subvol_file_size (xlator_t *this, loc_t *root_loc) { int ret = -1; struct statvfs buf = {0,}; if (!this) return 0; ret = syncop_statfs (this, root_loc, &buf, NULL, NULL); if (ret) { /* Aargh! */ return 0; } return ((buf.f_blocks - buf.f_bfree) * buf.f_frsize); } uint64_t gf_defrag_total_file_size (xlator_t *this, loc_t *root_loc) { dht_conf_t *conf = NULL; int i = 0; uint64_t size_files = 0; uint64_t total_size = 0; conf = this->private; if (!conf) { return 0; } for (i = 0 ; i < conf->local_subvols_cnt; i++) { size_files = gf_defrag_subvol_file_size (conf->local_subvols[i], root_loc); total_size += size_files; gf_msg (this->name, GF_LOG_INFO, 0, 0, "local subvol: %s," "cnt = %"PRIu64, conf->local_subvols[i]->name, size_files); } gf_msg (this->name, GF_LOG_INFO, 0, 0, "Total size files = %"PRIu64, total_size); return total_size; } static void* dht_file_counter_thread (void *args) { gf_defrag_info_t *defrag = NULL; loc_t root_loc = {0,}; struct timespec time_to_wait = {0,}; struct timeval now = {0,}; uint64_t tmp_size = 0; if (!args) return NULL; defrag = (gf_defrag_info_t *) args; dht_build_root_loc (defrag->root_inode, &root_loc); while (defrag->defrag_status == GF_DEFRAG_STATUS_STARTED) { gettimeofday (&now, NULL); time_to_wait.tv_sec = now.tv_sec + 600; time_to_wait.tv_nsec = 0; pthread_mutex_lock (&defrag->fc_mutex); pthread_cond_timedwait (&defrag->fc_wakeup_cond, &defrag->fc_mutex, &time_to_wait); pthread_mutex_unlock (&defrag->fc_mutex); if (defrag->defrag_status != GF_DEFRAG_STATUS_STARTED) break; tmp_size = gf_defrag_total_file_size (defrag->this, &root_loc); gf_log ("dht", GF_LOG_INFO, "tmp data size =%"PRIu64, tmp_size); if (!tmp_size) { gf_msg ("dht", GF_LOG_ERROR, 0, 0, "Failed to get " "the total data size. Unable to estimate " "time to complete rebalance."); } else { g_totalsize = tmp_size; gf_msg_debug ("dht", 0, "total data size =%"PRIu64, g_totalsize); } } return NULL; } int gf_defrag_estimates_cleanup (xlator_t *this, gf_defrag_info_t *defrag, pthread_t filecnt_thread) { int ret = -1; /* Wake up the filecounter thread. * By now the defrag status will no longer be * GF_DEFRAG_STATUS_STARTED so the thread will exit the loop. */ pthread_mutex_lock (&defrag->fc_mutex); { pthread_cond_broadcast (&defrag->fc_wakeup_cond); } pthread_mutex_unlock (&defrag->fc_mutex); ret = pthread_join (filecnt_thread, NULL); if (ret) { gf_msg ("dht", GF_LOG_ERROR, ret, 0, "file_counter_thread: pthread_join failed."); ret = -1; } return ret; } int gf_defrag_estimates_init (xlator_t *this, loc_t *loc, pthread_t *filecnt_thread) { int ret = -1; dht_conf_t *conf = NULL; gf_defrag_info_t *defrag = NULL; conf = this->private; defrag = conf->defrag; g_totalsize = gf_defrag_total_file_size (this, loc); if (!g_totalsize) { gf_msg (this->name, GF_LOG_ERROR, 0, 0, "Failed to get " "the total data size. Unable to estimate " "time to complete rebalance."); goto out; } ret = gf_thread_create (filecnt_thread, NULL, &dht_file_counter_thread, (void *)defrag, "dhtfcnt"); if (ret) { gf_msg (this->name, GF_LOG_ERROR, ret, 0, "Failed to " "create the file counter thread "); ret = -1; } ret = 0; out: return ret; } /* Init and cleanup functions for parallel file migration*/ int gf_defrag_parallel_migration_init (xlator_t *this, gf_defrag_info_t *defrag, pthread_t **tid_array, int *thread_index) { int ret = -1; int thread_spawn_count = 0; int index = 0; pthread_t *tid = NULL; char thread_name[GF_THREAD_NAMEMAX] = {0,}; if (!defrag) goto out; /* Initialize global entry queue */ defrag->queue = GF_CALLOC (1, sizeof (struct dht_container), gf_dht_mt_container_t); if (!defrag->queue) { gf_msg (this->name, GF_LOG_ERROR, ENOMEM, 0, "Failed to initialise migration queue"); ret = -1; goto out; } INIT_LIST_HEAD (&(defrag->queue[0].list)); thread_spawn_count = MAX (MAX_REBAL_THREADS, 4); gf_msg_debug (this->name, 0, "thread_spawn_count: %d", thread_spawn_count); tid = GF_CALLOC (thread_spawn_count, sizeof (pthread_t), gf_common_mt_pthread_t); if (!tid) { gf_msg (this->name, GF_LOG_ERROR, ENOMEM, 0, "Failed to create migration threads"); ret = -1; goto out; } defrag->current_thread_count = thread_spawn_count; /*Spawn Threads Here*/ while (index < thread_spawn_count) { snprintf (thread_name, sizeof(thread_name), "dhtmig%d", ((index + 1) & 0x3ff)); ret = gf_thread_create (&(tid[index]), NULL, &gf_defrag_task, (void *)defrag, thread_name); if (ret != 0) { gf_msg ("DHT", GF_LOG_ERROR, ret, 0, "Thread[%d] creation failed. ", index); ret = -1; goto out; } else { gf_log ("DHT", GF_LOG_INFO, "Thread[%d] " "creation successful", index); } index++; } ret = 0; out: *thread_index = index; *tid_array = tid; return ret; } int gf_defrag_parallel_migration_cleanup (gf_defrag_info_t *defrag, pthread_t *tid_array, int thread_index) { int ret = -1; int i = 0; if (!defrag) goto out; /* Wake up all migration threads */ pthread_mutex_lock (&defrag->dfq_mutex); { defrag->crawl_done = 1; pthread_cond_broadcast (&defrag->parallel_migration_cond); pthread_cond_broadcast (&defrag->df_wakeup_thread); } pthread_mutex_unlock (&defrag->dfq_mutex); /*Wait for all the threads to complete their task*/ for (i = 0; i < thread_index; i++) { pthread_join (tid_array[i], NULL); } GF_FREE (tid_array); /* Cleanup the migration queue */ if (defrag->queue) { gf_dirent_free (defrag->queue[0].df_entry); INIT_LIST_HEAD (&(defrag->queue[0].list)); } GF_FREE (defrag->queue); ret = 0; out: return ret; } int gf_defrag_start_crawl (void *data) { xlator_t *this = NULL; dht_conf_t *conf = NULL; gf_defrag_info_t *defrag = NULL; dict_t *fix_layout = NULL; dict_t *migrate_data = NULL; dict_t *status = NULL; glusterfs_ctx_t *ctx = NULL; dht_methods_t *methods = NULL; call_frame_t *statfs_frame = NULL; xlator_t *old_THIS = NULL; int ret = -1; loc_t loc = {0,}; struct iatt iatt = {0,}; struct iatt parent = {0,}; int thread_index = 0; pthread_t *tid = NULL; pthread_t filecnt_thread; gf_boolean_t is_tier_detach = _gf_false; gf_boolean_t fc_thread_started = _gf_false; this = data; if (!this) goto exit; ctx = this->ctx; if (!ctx) goto exit; conf = this->private; if (!conf) goto exit; defrag = conf->defrag; if (!defrag) goto exit; gettimeofday (&defrag->start_time, NULL); dht_build_root_inode (this, &defrag->root_inode); if (!defrag->root_inode) goto out; dht_build_root_loc (defrag->root_inode, &loc); /* fix-layout on '/' first */ ret = syncop_lookup (this, &loc, &iatt, &parent, NULL, NULL); if (ret) { gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_REBALANCE_START_FAILED, "Failed to start rebalance: look up on / failed"); ret = -1; goto out; } old_THIS = THIS; THIS = this; statfs_frame = create_frame (this, this->ctx->pool); if (!statfs_frame) { gf_msg (this->name, GF_LOG_ERROR, DHT_MSG_NO_MEMORY, ENOMEM, "Insufficient memory. Frame creation failed"); ret = -1; goto out; } /* async statfs update for honoring min-free-disk */ dht_get_du_info (statfs_frame, this, &loc); THIS = old_THIS; fix_layout = dict_new (); if (!fix_layout) { ret = -1; goto out; } /* * Unfortunately, we can't do special xattrs (like fix.layout) and * real ones in the same call currently, and changing it seems * riskier than just doing two calls. */ gf_log (this->name, GF_LOG_INFO, "%s using commit hash %u", __func__, conf->vol_commit_hash); ret = dict_set_uint32 (fix_layout, conf->commithash_xattr_name, conf->vol_commit_hash); if (ret) { gf_log (this->name, GF_LOG_ERROR, "Failed to set %s", conf->commithash_xattr_name); defrag->total_failures++; ret = -1; goto out; } ret = syncop_setxattr (this, &loc, fix_layout, 0, NULL, NULL); if (ret) { gf_log (this->name, GF_LOG_ERROR, "Failed to set commit hash on %s. " "Rebalance cannot proceed.", loc.path); defrag->total_failures++; ret = -1; goto out; } /* We now return to our regularly scheduled program. */ ret = dict_set_str (fix_layout, GF_XATTR_FIX_LAYOUT_KEY, "yes"); if (ret) { gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_REBALANCE_START_FAILED, "Failed to start rebalance:" "Failed to set dictionary value: key = %s", GF_XATTR_FIX_LAYOUT_KEY); defrag->total_failures++; ret = -1; goto out; } defrag->new_commit_hash = conf->vol_commit_hash; ret = syncop_setxattr (this, &loc, fix_layout, 0, NULL, NULL); if (ret) { gf_msg (this->name, GF_LOG_ERROR, -ret, DHT_MSG_REBALANCE_FAILED, "fix layout on %s failed", loc.path); defrag->total_failures++; ret = -1; goto out; } if (defrag->cmd != GF_DEFRAG_CMD_START_LAYOUT_FIX) { /* We need to migrate files */ migrate_data = dict_new (); if (!migrate_data) { defrag->total_failures++; ret = -1; goto out; } ret = dict_set_str (migrate_data, GF_XATTR_FILE_MIGRATE_KEY, (defrag->cmd == GF_DEFRAG_CMD_START_FORCE) ? "force" : "non-force"); if (ret) { defrag->total_failures++; ret = -1; goto out; } ret = dht_init_local_subvols_and_nodeuuids (this, conf, &loc); if (ret) { ret = -1; goto out; } /* Initialise the structures required for parallel migration */ ret = gf_defrag_parallel_migration_init (this, defrag, &tid, &thread_index); if (ret) { gf_msg (this->name, GF_LOG_ERROR, 0, 0, "Aborting rebalance."); goto out; } ret = gf_defrag_estimates_init (this, &loc, &filecnt_thread); if (ret) { /* Not a fatal error. Allow the rebalance to proceed*/ ret = 0; } else { fc_thread_started = _gf_true; } } if (defrag->cmd == GF_DEFRAG_CMD_START_TIER) { /* Fix layout for attach tier */ ret = gf_tier_start_fix_layout (this, &loc, defrag, fix_layout); if (ret) { goto out; } methods = &(conf->methods); /* Calling tier_start of tier.c */ methods->migration_other(this, defrag); if (defrag->cmd == GF_DEFRAG_CMD_START_DETACH_TIER || defrag->cmd == GF_DEFRAG_CMD_DETACH_START) { ret = dict_set_str (migrate_data, GF_XATTR_FILE_MIGRATE_KEY, "force"); if (ret) goto out; } } else { ret = gf_defrag_fix_layout (this, defrag, &loc, fix_layout, migrate_data); if (ret && ret != 2) { defrag->total_failures++; ret = -1; goto out; } if (ret != 2 && gf_defrag_settle_hash (this, defrag, &loc, fix_layout) != 0) { defrag->total_failures++; ret = -1; goto out; } if (defrag->cmd == GF_DEFRAG_CMD_START_DETACH_TIER || defrag->cmd == GF_DEFRAG_CMD_DETACH_START) is_tier_detach = _gf_true; } gf_log ("DHT", GF_LOG_INFO, "crawling file-system completed"); out: /* We are here means crawling the entire file system is done or something failed. Set defrag->crawl_done flag to intimate the migrator threads to exhaust the defrag->queue and terminate*/ if (ret) { defrag->defrag_status = GF_DEFRAG_STATUS_FAILED; } if (defrag->cmd == GF_DEFRAG_CMD_START_TIER) { /* Wait for the tier fixlayout to * complete if its was started.*/ gf_tier_wait_fix_lookup (defrag); } if (is_tier_detach && ret == 0) { /* If it was a detach remove the tier fix-layout * xattr on root. Ignoring the failure, as nothing has to be * done, logging is done in gf_tier_clear_fix_layout */ gf_tier_clear_fix_layout (this, &loc, defrag); } gf_defrag_parallel_migration_cleanup (defrag, tid, thread_index); if ((defrag->defrag_status != GF_DEFRAG_STATUS_STOPPED) && (defrag->defrag_status != GF_DEFRAG_STATUS_FAILED)) { defrag->defrag_status = GF_DEFRAG_STATUS_COMPLETE; } if (fc_thread_started) { gf_defrag_estimates_cleanup (this, defrag, filecnt_thread); } dht_send_rebalance_event (this, defrag->cmd, defrag->defrag_status); LOCK (&defrag->lock); { status = dict_new (); gf_defrag_status_get (conf, status); if (ctx && ctx->notify) ctx->notify (GF_EN_DEFRAG_STATUS, status); if (status) dict_unref (status); defrag->is_exiting = 1; } UNLOCK (&defrag->lock); GF_FREE (defrag); conf->defrag = NULL; if (migrate_data) dict_unref (migrate_data); if (statfs_frame) { STACK_DESTROY (statfs_frame->root); } exit: return ret; } static int gf_defrag_done (int ret, call_frame_t *sync_frame, void *data) { gf_listener_stop (sync_frame->this); STACK_DESTROY (sync_frame->root); kill (getpid(), SIGTERM); return 0; } void * gf_defrag_start (void *data) { int ret = -1; call_frame_t *frame = NULL; dht_conf_t *conf = NULL; gf_defrag_info_t *defrag = NULL; xlator_t *this = NULL; xlator_t *old_THIS = NULL; this = data; conf = this->private; if (!conf) goto out; defrag = conf->defrag; if (!defrag) goto out; frame = create_frame (this, this->ctx->pool); if (!frame) goto out; frame->root->pid = GF_CLIENT_PID_DEFRAG; defrag->pid = frame->root->pid; defrag->defrag_status = GF_DEFRAG_STATUS_STARTED; old_THIS = THIS; THIS = this; ret = synctask_new (this->ctx->env, gf_defrag_start_crawl, gf_defrag_done, frame, this); if (ret) gf_msg (this->name, GF_LOG_ERROR, 0, DHT_MSG_REBALANCE_START_FAILED, "Could not create task for rebalance"); THIS = old_THIS; out: return NULL; } uint64_t gf_defrag_get_estimates_based_on_size (dht_conf_t *conf) { gf_defrag_info_t *defrag = NULL; double rate_processed = 0; uint64_t total_processed = 0; uint64_t tmp_count = 0; uint64_t time_to_complete = 0; struct timeval now = {0,}; double elapsed = 0; defrag = conf->defrag; if (!g_totalsize) goto out; gettimeofday (&now, NULL); elapsed = now.tv_sec - defrag->start_time.tv_sec; /* Don't calculate the estimates for the first 10 minutes. * It is unlikely to be accurate and estimates are not required * if the process finishes in less than 10 mins. */ if (elapsed < ESTIMATE_START_INTERVAL) { gf_msg (THIS->name, GF_LOG_INFO, 0, 0, "Rebalance estimates will not be available for the " "first %d seconds.", ESTIMATE_START_INTERVAL); goto out; } total_processed = defrag->size_processed; /* rate at which files processed */ rate_processed = (total_processed)/elapsed; tmp_count = g_totalsize; if (rate_processed) { time_to_complete = (tmp_count)/rate_processed; } else { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "Unable to calculate estimated time for rebalance"); } gf_log (THIS->name, GF_LOG_INFO, "TIME: (size) total_processed=%"PRIu64" tmp_cnt = %"PRIu64"," "rate_processed=%f, elapsed = %f", total_processed, tmp_count, rate_processed, elapsed); out: return time_to_complete; } int gf_defrag_status_get (dht_conf_t *conf, dict_t *dict) { int ret = 0; uint64_t files = 0; uint64_t size = 0; uint64_t lookup = 0; uint64_t failures = 0; uint64_t skipped = 0; uint64_t promoted = 0; uint64_t demoted = 0; char *status = ""; double elapsed = 0; struct timeval end = {0,}; uint64_t time_to_complete = 0; uint64_t time_left = 0; gf_defrag_info_t *defrag = conf->defrag; if (!defrag) goto out; ret = 0; if (defrag->defrag_status == GF_DEFRAG_STATUS_NOT_STARTED) goto out; files = defrag->total_files; size = defrag->total_data; lookup = defrag->num_files_lookedup; failures = defrag->total_failures; skipped = defrag->skipped; promoted = defrag->total_files_promoted; demoted = defrag->total_files_demoted; gettimeofday (&end, NULL); elapsed = end.tv_sec - defrag->start_time.tv_sec; /* The rebalance is still in progress */ if ((defrag->cmd != GF_DEFRAG_CMD_START_TIER) && (defrag->defrag_status == GF_DEFRAG_STATUS_STARTED)) { time_to_complete = gf_defrag_get_estimates_based_on_size (conf); if (time_to_complete && (time_to_complete > elapsed)) time_left = time_to_complete - elapsed; gf_log (THIS->name, GF_LOG_INFO, "TIME: Estimated total time to complete (size)= %"PRIu64 " seconds, seconds left = %"PRIu64"", time_to_complete, time_left); } if (!dict) goto log; ret = dict_set_uint64 (dict, "promoted", promoted); if (ret) gf_log (THIS->name, GF_LOG_WARNING, "failed to set promoted count"); ret = dict_set_uint64 (dict, "demoted", demoted); if (ret) gf_log (THIS->name, GF_LOG_WARNING, "failed to set demoted count"); ret = dict_set_uint64 (dict, "files", files); if (ret) gf_log (THIS->name, GF_LOG_WARNING, "failed to set file count"); ret = dict_set_uint64 (dict, "size", size); if (ret) gf_log (THIS->name, GF_LOG_WARNING, "failed to set size of xfer"); ret = dict_set_uint64 (dict, "lookups", lookup); if (ret) gf_log (THIS->name, GF_LOG_WARNING, "failed to set lookedup file count"); ret = dict_set_int32 (dict, "status", defrag->defrag_status); if (ret) gf_log (THIS->name, GF_LOG_WARNING, "failed to set status"); ret = dict_set_double (dict, "run-time", elapsed); if (ret) gf_log (THIS->name, GF_LOG_WARNING, "failed to set run-time"); ret = dict_set_uint64 (dict, "failures", failures); if (ret) gf_log (THIS->name, GF_LOG_WARNING, "failed to set failure count"); ret = dict_set_uint64 (dict, "skipped", skipped); if (ret) gf_log (THIS->name, GF_LOG_WARNING, "failed to set skipped file count"); ret = dict_set_uint64 (dict, "time-left", time_left); if (ret) gf_log (THIS->name, GF_LOG_WARNING, "failed to set time-left"); log: switch (defrag->defrag_status) { case GF_DEFRAG_STATUS_NOT_STARTED: status = "not started"; break; case GF_DEFRAG_STATUS_STARTED: status = "in progress"; break; case GF_DEFRAG_STATUS_STOPPED: status = "stopped"; break; case GF_DEFRAG_STATUS_COMPLETE: status = "completed"; break; case GF_DEFRAG_STATUS_FAILED: status = "failed"; break; default: break; } gf_msg (THIS->name, GF_LOG_INFO, 0, DHT_MSG_REBALANCE_STATUS, "Rebalance is %s. Time taken is %.2f secs", status, elapsed); gf_msg (THIS->name, GF_LOG_INFO, 0, DHT_MSG_REBALANCE_STATUS, "Files migrated: %"PRIu64", size: %" PRIu64", lookups: %"PRIu64", failures: %"PRIu64", skipped: " "%"PRIu64, files, size, lookup, failures, skipped); out: return 0; } void gf_defrag_set_pause_state (gf_tier_conf_t *tier_conf, tier_pause_state_t state) { pthread_mutex_lock (&tier_conf->pause_mutex); tier_conf->pause_state = state; pthread_mutex_unlock (&tier_conf->pause_mutex); } tier_pause_state_t gf_defrag_get_pause_state (gf_tier_conf_t *tier_conf) { int state; pthread_mutex_lock (&tier_conf->pause_mutex); state = tier_conf->pause_state; pthread_mutex_unlock (&tier_conf->pause_mutex); return state; } tier_pause_state_t gf_defrag_check_pause_tier (gf_tier_conf_t *tier_conf) { int woke = 0; int state = -1; pthread_mutex_lock (&tier_conf->pause_mutex); if (tier_conf->pause_state == TIER_RUNNING) goto out; if (tier_conf->pause_state == TIER_PAUSED) goto out; if (tier_conf->promote_in_progress || tier_conf->demote_in_progress) goto out; tier_conf->pause_state = TIER_PAUSED; if (tier_conf->pause_synctask) { synctask_wake (tier_conf->pause_synctask); tier_conf->pause_synctask = 0; woke = 1; } gf_msg ("tier", GF_LOG_DEBUG, 0, DHT_MSG_TIER_PAUSED, "woken %d", woke); gf_event (EVENT_TIER_PAUSE, "vol=%s", tier_conf->volname); out: state = tier_conf->pause_state; pthread_mutex_unlock (&tier_conf->pause_mutex); return state; } void gf_defrag_pause_tier_timeout (void *data) { xlator_t *this = NULL; dht_conf_t *conf = NULL; gf_defrag_info_t *defrag = NULL; this = (xlator_t *) data; GF_VALIDATE_OR_GOTO ("tier", this, out); conf = this->private; GF_VALIDATE_OR_GOTO (this->name, conf, out); defrag = conf->defrag; GF_VALIDATE_OR_GOTO (this->name, defrag, out); gf_msg (this->name, GF_LOG_DEBUG, 0, DHT_MSG_TIER_PAUSED, "Request pause timer timeout"); gf_defrag_check_pause_tier (&defrag->tier_conf); out: return; } int gf_defrag_pause_tier (xlator_t *this, gf_defrag_info_t *defrag) { int ret = 0; struct timespec delta = {0,}; int delay = 2; if (defrag->defrag_status != GF_DEFRAG_STATUS_STARTED) goto out; /* * Set flag requesting to pause tiering. Wait 'delay' seconds for * tiering to actually stop as indicated by the pause state * before returning success or failure. */ gf_defrag_set_pause_state (&defrag->tier_conf, TIER_REQUEST_PAUSE); /* * If migration is not underway, can pause immediately. */ gf_defrag_check_pause_tier (&defrag->tier_conf); if (gf_defrag_get_pause_state (&defrag->tier_conf) == TIER_PAUSED) goto out; gf_msg (this->name, GF_LOG_DEBUG, 0, DHT_MSG_TIER_PAUSED, "Request pause tier"); defrag->tier_conf.pause_synctask = synctask_get (); delta.tv_sec = delay; delta.tv_nsec = 0; defrag->tier_conf.pause_timer = gf_timer_call_after (this->ctx, delta, gf_defrag_pause_tier_timeout, this); synctask_yield (defrag->tier_conf.pause_synctask); if (gf_defrag_get_pause_state (&defrag->tier_conf) == TIER_PAUSED) goto out; gf_defrag_set_pause_state (&defrag->tier_conf, TIER_RUNNING); ret = -1; out: gf_msg (this->name, GF_LOG_DEBUG, 0, DHT_MSG_TIER_PAUSED, "Pause tiering ret=%d", ret); return ret; } int gf_defrag_resume_tier (xlator_t *this, gf_defrag_info_t *defrag) { gf_msg (this->name, GF_LOG_DEBUG, 0, DHT_MSG_TIER_RESUME, "Pause end. Resume tiering"); gf_defrag_set_pause_state (&defrag->tier_conf, TIER_RUNNING); gf_event (EVENT_TIER_RESUME, "vol=%s", defrag->tier_conf.volname); return 0; } int gf_defrag_start_detach_tier (gf_defrag_info_t *defrag) { defrag->cmd = GF_DEFRAG_CMD_START_DETACH_TIER; return 0; } int gf_defrag_stop (dht_conf_t *conf, gf_defrag_status_t status, dict_t *output) { /* TODO: set a variable 'stop_defrag' here, it should be checked in defrag loop */ int ret = -1; gf_defrag_info_t *defrag = conf->defrag; GF_ASSERT (defrag); if (defrag->defrag_status == GF_DEFRAG_STATUS_NOT_STARTED) { goto out; } gf_msg ("", GF_LOG_INFO, 0, DHT_MSG_REBALANCE_STOPPED, "Received stop command on rebalance"); defrag->defrag_status = status; if (output) gf_defrag_status_get (conf, output); ret = 0; out: gf_msg_debug ("", 0, "Returning %d", ret); return ret; }