diff options
Diffstat (limited to 'xlators/features/bit-rot/src/bitd')
| -rw-r--r-- | xlators/features/bit-rot/src/bitd/Makefile.am | 23 | ||||
| -rw-r--r-- | xlators/features/bit-rot/src/bitd/bit-rot-bitd-messages.h | 101 | ||||
| -rw-r--r-- | xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.c | 78 | ||||
| -rw-r--r-- | xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.h | 50 | ||||
| -rw-r--r-- | xlators/features/bit-rot/src/bitd/bit-rot-scrub.c | 2070 | ||||
| -rw-r--r-- | xlators/features/bit-rot/src/bitd/bit-rot-scrub.h | 46 | ||||
| -rw-r--r-- | xlators/features/bit-rot/src/bitd/bit-rot-ssm.c | 124 | ||||
| -rw-r--r-- | xlators/features/bit-rot/src/bitd/bit-rot-ssm.h | 38 | ||||
| -rw-r--r-- | xlators/features/bit-rot/src/bitd/bit-rot.c | 2232 | ||||
| -rw-r--r-- | xlators/features/bit-rot/src/bitd/bit-rot.h | 302 |
10 files changed, 5064 insertions, 0 deletions
diff --git a/xlators/features/bit-rot/src/bitd/Makefile.am b/xlators/features/bit-rot/src/bitd/Makefile.am new file mode 100644 index 00000000000..6db800e6565 --- /dev/null +++ b/xlators/features/bit-rot/src/bitd/Makefile.am @@ -0,0 +1,23 @@ +if WITH_SERVER +xlator_LTLIBRARIES = bit-rot.la +endif +xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features + +bit_rot_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS) + +AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \ + -I$(top_srcdir)/rpc/xdr/src/ -I$(top_builddir)/rpc/xdr/src/ \ + -I$(top_srcdir)/rpc/rpc-lib/src -I$(CONTRIBDIR)/timer-wheel \ + -I$(top_srcdir)/xlators/features/bit-rot/src/stub + +bit_rot_la_SOURCES = bit-rot.c bit-rot-scrub.c bit-rot-ssm.c \ + bit-rot-scrub-status.c +bit_rot_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la \ + $(top_builddir)/xlators/features/changelog/lib/src/libgfchangelog.la + +noinst_HEADERS = bit-rot.h bit-rot-scrub.h bit-rot-bitd-messages.h bit-rot-ssm.h \ + bit-rot-scrub-status.h + +AM_CFLAGS = -Wall -DBR_RATE_LIMIT_SIGNER $(GF_CFLAGS) + +CLEANFILES = diff --git a/xlators/features/bit-rot/src/bitd/bit-rot-bitd-messages.h b/xlators/features/bit-rot/src/bitd/bit-rot-bitd-messages.h new file mode 100644 index 00000000000..5bc5103a27c --- /dev/null +++ b/xlators/features/bit-rot/src/bitd/bit-rot-bitd-messages.h @@ -0,0 +1,101 @@ +/* + Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com> + 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. + */ + +#ifndef _BITROT_BITD_MESSAGES_H_ +#define _BITROT_BITD_MESSAGES_H_ + +#include <glusterfs/glfs-message-id.h> + +/* To add new message IDs, append new identifiers at the end of the list. + * + * Never remove a message ID. If it's not used anymore, you can rename it or + * leave it as it is, but not delete it. This is to prevent reutilization of + * IDs by other messages. + * + * The component name must match one of the entries defined in + * glfs-message-id.h. + */ + +GLFS_MSGID(BITROT_BITD, BRB_MSG_FD_CREATE_FAILED, BRB_MSG_READV_FAILED, + BRB_MSG_BLOCK_READ_FAILED, BRB_MSG_CALC_CHECKSUM_FAILED, + BRB_MSG_NO_MEMORY, BRB_MSG_GET_SIGN_FAILED, BRB_MSG_SET_SIGN_FAILED, + BRB_MSG_OP_FAILED, BRB_MSG_READ_AND_SIGN_FAILED, BRB_MSG_SIGN_FAILED, + BRB_MSG_GET_SUBVOL_FAILED, BRB_MSG_SET_TIMER_FAILED, + BRB_MSG_GET_INFO_FAILED, BRB_MSG_PATH_FAILED, BRB_MSG_MARK_BAD_FILE, + BRB_MSG_TRIGGER_SIGN, BRB_MSG_REGISTER_FAILED, + BRB_MSG_CRAWLING_START, BRB_MSG_SPAWN_FAILED, + BRB_MSG_INVALID_SUBVOL_CHILD, BRB_MSG_SKIP_OBJECT, BRB_MSG_NO_CHILD, + BRB_MSG_CHECKSUM_MISMATCH, BRB_MSG_MARK_CORRUPTED, + BRB_MSG_CRAWLING_FINISH, BRB_MSG_CALC_ERROR, BRB_MSG_LOOKUP_FAILED, + BRB_MSG_PARTIAL_VERSION_PRESENCE, BRB_MSG_MEM_ACNT_FAILED, + BRB_MSG_TIMER_WHEEL_UNAVAILABLE, BRB_MSG_BITROT_LOADED, + BRB_MSG_SCALE_DOWN_FAILED, BRB_MSG_SCALE_UP_FAILED, + BRB_MSG_SCALE_DOWN_SCRUBBER, BRB_MSG_SCALING_UP_SCRUBBER, + BRB_MSG_UNKNOWN_THROTTLE, BRB_MSG_RATE_LIMIT_INFO, + BRB_MSG_SCRUB_INFO, BRB_MSG_CONNECTED_TO_BRICK, BRB_MSG_BRICK_INFO, + BRB_MSG_SUBVOL_CONNECT_FAILED, BRB_MSG_INVALID_SUBVOL, + BRB_MSG_RESCHEDULE_SCRUBBER_FAILED, BRB_MSG_SCRUB_START, + BRB_MSG_SCRUB_FINISH, BRB_MSG_SCRUB_RUNNING, + BRB_MSG_SCRUB_RESCHEDULED, BRB_MSG_SCRUB_TUNABLE, + BRB_MSG_SCRUB_THREAD_CLEANUP, BRB_MSG_SCRUBBER_CLEANED, + BRB_MSG_GENERIC_SSM_INFO, BRB_MSG_ZERO_TIMEOUT_BUG, + BRB_MSG_BAD_OBJ_READDIR_FAIL, BRB_MSG_SSM_FAILED, + BRB_MSG_SCRUB_WAIT_FAILED, BRB_MSG_TRIGGER_SIGN_FAILED, + BRB_MSG_EVENT_UNHANDLED, BRB_MSG_COULD_NOT_SCHEDULE_SCRUB, + BRB_MSG_THREAD_CREATION_FAILED, BRB_MSG_MEM_POOL_ALLOC, + BRB_MSG_SAVING_HASH_FAILED); + +#define BRB_MSG_FD_CREATE_FAILED_STR "failed to create fd for the inode" +#define BRB_MSG_READV_FAILED_STR "readv failed" +#define BRB_MSG_BLOCK_READ_FAILED_STR "reading block failed" +#define BRB_MSG_NO_MEMORY_STR "failed to allocate memory" +#define BRB_MSG_CALC_CHECKSUM_FAILED_STR "calculating checksum failed" +#define BRB_MSG_GET_SIGN_FAILED_STR "failed to get the signature" +#define BRB_MSG_SET_SIGN_FAILED_STR "signing failed" +#define BRB_MSG_OP_FAILED_STR "failed on object" +#define BRB_MSG_TRIGGER_SIGN_FAILED_STR "Could not trigger signing" +#define BRB_MSG_READ_AND_SIGN_FAILED_STR "reading and signing of object failed" +#define BRB_MSG_SET_TIMER_FAILED_STR "Failed to allocate object expiry timer" +#define BRB_MSG_GET_SUBVOL_FAILED_STR \ + "failed to get the subvolume for the brick" +#define BRB_MSG_PATH_FAILED_STR "path failed" +#define BRB_MSG_SKIP_OBJECT_STR "Entry is marked corrupted. skipping" +#define BRB_MSG_PARTIAL_VERSION_PRESENCE_STR \ + "PArtial version xattr presence detected, ignoring" +#define BRB_MSG_TRIGGER_SIGN_STR "Triggering signing" +#define BRB_MSG_CRAWLING_START_STR \ + "Crawling brick, scanning for unsigned objects" +#define BRB_MSG_CRAWLING_FINISH_STR "Completed crawling brick" +#define BRB_MSG_REGISTER_FAILED_STR "Register to changelog failed" +#define BRB_MSG_SPAWN_FAILED_STR "failed to spawn" +#define BRB_MSG_CONNECTED_TO_BRICK_STR "Connected to brick" +#define BRB_MSG_LOOKUP_FAILED_STR "lookup on root failed" +#define BRB_MSG_GET_INFO_FAILED_STR "failed to get stub info" +#define BRB_MSG_SCRUB_THREAD_CLEANUP_STR "Error cleaning up scanner thread" +#define BRB_MSG_SCRUBBER_CLEANED_STR "clened up scrubber for brick" +#define BRB_MSG_SUBVOL_CONNECT_FAILED_STR \ + "callback handler for subvolume failed" +#define BRB_MSG_MEM_ACNT_FAILED_STR "Memory accounting init failed" +#define BRB_MSG_EVENT_UNHANDLED_STR "Event unhandled for child" +#define BRB_MSG_INVALID_SUBVOL_STR "Got event from invalid subvolume" +#define BRB_MSG_RESCHEDULE_SCRUBBER_FAILED_STR \ + "on demand scrub schedule failed. Scrubber is not in pending state." +#define BRB_MSG_COULD_NOT_SCHEDULE_SCRUB_STR \ + "Could not schedule ondemand scrubbing. Scrubbing will continue " \ + "according to old frequency." +#define BRB_MSG_THREAD_CREATION_FAILED_STR "thread creation failed" +#define BRB_MSG_RATE_LIMIT_INFO_STR "Rate Limit Info" +#define BRB_MSG_MEM_POOL_ALLOC_STR "failed to allocate mem-pool for timer" +#define BRB_MSG_NO_CHILD_STR "FATAL: no children" +#define BRB_MSG_TIMER_WHEEL_UNAVAILABLE_STR "global timer wheel unavailable" +#define BRB_MSG_BITROT_LOADED_STR "bit-rot xlator loaded" +#define BRB_MSG_SAVING_HASH_FAILED_STR \ + "failed to allocate memory for saving hash of the object" +#endif /* !_BITROT_BITD_MESSAGES_H_ */ diff --git a/xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.c b/xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.c new file mode 100644 index 00000000000..5cef2ffa5e5 --- /dev/null +++ b/xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.c @@ -0,0 +1,78 @@ +/* + Copyright (c) 2016 Red Hat, Inc. <http://www.redhat.com> + 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 <string.h> +#include <stdio.h> + +#include "bit-rot-scrub-status.h" + +void +br_inc_unsigned_file_count(br_scrub_stats_t *scrub_stat) +{ + if (!scrub_stat) + return; + + pthread_mutex_lock(&scrub_stat->lock); + { + scrub_stat->unsigned_files++; + } + pthread_mutex_unlock(&scrub_stat->lock); +} + +void +br_inc_scrubbed_file(br_scrub_stats_t *scrub_stat) +{ + if (!scrub_stat) + return; + + pthread_mutex_lock(&scrub_stat->lock); + { + scrub_stat->scrubbed_files++; + } + pthread_mutex_unlock(&scrub_stat->lock); +} + +void +br_update_scrub_start_time(br_scrub_stats_t *scrub_stat, time_t time) +{ + if (!scrub_stat) + return; + + pthread_mutex_lock(&scrub_stat->lock); + { + scrub_stat->scrub_start_time = time; + } + pthread_mutex_unlock(&scrub_stat->lock); +} + +void +br_update_scrub_finish_time(br_scrub_stats_t *scrub_stat, char *timestr, + time_t time) +{ + int lst_size = 0; + + if (!scrub_stat) + return; + + lst_size = sizeof(scrub_stat->last_scrub_time); + if (strlen(timestr) >= lst_size) + return; + + pthread_mutex_lock(&scrub_stat->lock); + { + scrub_stat->scrub_end_time = time; + + scrub_stat->scrub_duration = scrub_stat->scrub_end_time - + scrub_stat->scrub_start_time; + + snprintf(scrub_stat->last_scrub_time, lst_size, "%s", timestr); + } + pthread_mutex_unlock(&scrub_stat->lock); +} diff --git a/xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.h b/xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.h new file mode 100644 index 00000000000..f022aa831eb --- /dev/null +++ b/xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.h @@ -0,0 +1,50 @@ +/* + Copyright (c) 2016 Red Hat, Inc. <http://www.redhat.com> + 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. +*/ + +#ifndef __BIT_ROT_SCRUB_STATUS_H__ +#define __BIT_ROT_SCRUB_STATUS_H__ + +#include <stdint.h> +#include <sys/time.h> +#include <pthread.h> + +#include <glusterfs/common-utils.h> + +struct br_scrub_stats { + uint64_t scrubbed_files; /* Total number of scrubbed files. */ + + uint64_t unsigned_files; /* Total number of unsigned files. */ + + uint64_t scrub_duration; /* Duration of last scrub. */ + + char last_scrub_time[GF_TIMESTR_SIZE]; /* Last scrub completion time. */ + + time_t scrub_start_time; /* Scrubbing starting time. */ + + time_t scrub_end_time; /* Scrubbing finishing time. */ + + int8_t scrub_running; /* Whether scrub running or not. */ + + pthread_mutex_t lock; +}; + +typedef struct br_scrub_stats br_scrub_stats_t; + +void +br_inc_unsigned_file_count(br_scrub_stats_t *scrub_stat); +void +br_inc_scrubbed_file(br_scrub_stats_t *scrub_stat); +void +br_update_scrub_start_time(br_scrub_stats_t *scrub_stat, time_t time); +void +br_update_scrub_finish_time(br_scrub_stats_t *scrub_stat, char *timestr, + time_t time); + +#endif /* __BIT_ROT_SCRUB_STATUS_H__ */ diff --git a/xlators/features/bit-rot/src/bitd/bit-rot-scrub.c b/xlators/features/bit-rot/src/bitd/bit-rot-scrub.c new file mode 100644 index 00000000000..289dd53f610 --- /dev/null +++ b/xlators/features/bit-rot/src/bitd/bit-rot-scrub.c @@ -0,0 +1,2070 @@ +/* + Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com> + 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 <math.h> +#include <ctype.h> +#include <sys/uio.h> + +#include <glusterfs/glusterfs.h> +#include <glusterfs/logging.h> +#include <glusterfs/common-utils.h> + +#include "bit-rot-scrub.h" +#include <pthread.h> +#include "bit-rot-bitd-messages.h" +#include "bit-rot-scrub-status.h" +#include <glusterfs/events.h> + +struct br_scrubbers { + pthread_t scrubthread; + + struct list_head list; +}; + +struct br_fsscan_entry { + void *data; + + loc_t parent; + + gf_dirent_t *entry; + + struct br_scanfs *fsscan; /* backpointer to subvolume scanner */ + + struct list_head list; +}; + +/** + * fetch signature extended attribute from an object's fd. + * NOTE: On success @xattr is not unref'd as @sign points + * to the dictionary value. + */ +static int32_t +bitd_fetch_signature(xlator_t *this, br_child_t *child, fd_t *fd, + dict_t **xattr, br_isignature_out_t **sign) +{ + int32_t ret = -1; + + ret = syncop_fgetxattr(child->xl, fd, xattr, GLUSTERFS_GET_OBJECT_SIGNATURE, + NULL, NULL); + if (ret < 0) { + br_log_object(this, "fgetxattr", fd->inode->gfid, -ret); + goto out; + } + + ret = dict_get_ptr(*xattr, GLUSTERFS_GET_OBJECT_SIGNATURE, (void **)sign); + if (ret) { + gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_GET_SIGN_FAILED, + "failed to extract signature info [GFID: %s]", + uuid_utoa(fd->inode->gfid)); + goto unref_dict; + } + + return 0; + +unref_dict: + dict_unref(*xattr); +out: + return -1; +} + +/** + * POST COMPUTE CHECK + * + * Checks to be performed before verifying calculated signature + * Object is skipped if: + * - has stale signature + * - mismatches versions caches in pre-compute check + */ + +int32_t +bitd_scrub_post_compute_check(xlator_t *this, br_child_t *child, fd_t *fd, + unsigned long version, + br_isignature_out_t **signature, + br_scrub_stats_t *scrub_stat, + gf_boolean_t skip_stat) +{ + int32_t ret = 0; + size_t signlen = 0; + dict_t *xattr = NULL; + br_isignature_out_t *signptr = NULL; + + ret = bitd_fetch_signature(this, child, fd, &xattr, &signptr); + if (ret < 0) { + if (!skip_stat) + br_inc_unsigned_file_count(scrub_stat); + goto out; + } + + /** + * Either the object got dirtied during the time the signature was + * calculated OR the version we saved during pre-compute check does + * not match now, implying that the object got dirtied and signed in + * between scrubs pre & post compute checks (checksum window). + * + * The log entry looks pretty ugly, but helps in debugging.. + */ + if (signptr->stale || (signptr->version != version)) { + if (!skip_stat) + br_inc_unsigned_file_count(scrub_stat); + gf_msg_debug(this->name, 0, + "<STAGE: POST> Object [GFID: %s] " + "either has a stale signature OR underwent " + "signing during checksumming {Stale: %d | " + "Version: %lu,%lu}", + uuid_utoa(fd->inode->gfid), (signptr->stale) ? 1 : 0, + version, signptr->version); + ret = -1; + goto unref_dict; + } + + signlen = signptr->signaturelen; + *signature = GF_MALLOC(sizeof(br_isignature_out_t) + signlen, + gf_common_mt_char); + + (void)memcpy(*signature, signptr, sizeof(br_isignature_out_t) + signlen); + + (*signature)->signaturelen = signlen; + +unref_dict: + dict_unref(xattr); +out: + return ret; +} + +static int32_t +bitd_signature_staleness(xlator_t *this, br_child_t *child, fd_t *fd, + int *stale, unsigned long *version, + br_scrub_stats_t *scrub_stat, gf_boolean_t skip_stat) +{ + int32_t ret = -1; + dict_t *xattr = NULL; + br_isignature_out_t *signptr = NULL; + + ret = bitd_fetch_signature(this, child, fd, &xattr, &signptr); + if (ret < 0) { + if (!skip_stat) + br_inc_unsigned_file_count(scrub_stat); + goto out; + } + + /** + * save version for validation in post compute stage + * c.f. bitd_scrub_post_compute_check() + */ + *stale = signptr->stale ? 1 : 0; + *version = signptr->version; + + dict_unref(xattr); + +out: + return ret; +} + +/** + * PRE COMPUTE CHECK + * + * Checks to be performed before initiating object signature calculation. + * An object is skipped if: + * - it's already marked corrupted + * - has stale signature + */ +int32_t +bitd_scrub_pre_compute_check(xlator_t *this, br_child_t *child, fd_t *fd, + unsigned long *version, + br_scrub_stats_t *scrub_stat, + gf_boolean_t skip_stat) +{ + int stale = 0; + int32_t ret = -1; + + if (bitd_is_bad_file(this, child, NULL, fd)) { + gf_msg(this->name, GF_LOG_WARNING, 0, BRB_MSG_SKIP_OBJECT, + "Object [GFID: %s] is marked corrupted, skipping..", + uuid_utoa(fd->inode->gfid)); + goto out; + } + + ret = bitd_signature_staleness(this, child, fd, &stale, version, scrub_stat, + skip_stat); + if (!ret && stale) { + if (!skip_stat) + br_inc_unsigned_file_count(scrub_stat); + gf_msg_debug(this->name, 0, + "<STAGE: PRE> Object [GFID: %s] " + "has stale signature", + uuid_utoa(fd->inode->gfid)); + ret = -1; + } + +out: + return ret; +} + +/* static int */ +int +bitd_compare_ckum(xlator_t *this, br_isignature_out_t *sign, unsigned char *md, + inode_t *linked_inode, gf_dirent_t *entry, fd_t *fd, + br_child_t *child, loc_t *loc) +{ + int ret = -1; + dict_t *xattr = NULL; + + GF_VALIDATE_OR_GOTO("bit-rot", this, out); + GF_VALIDATE_OR_GOTO(this->name, sign, out); + GF_VALIDATE_OR_GOTO(this->name, fd, out); + GF_VALIDATE_OR_GOTO(this->name, child, out); + GF_VALIDATE_OR_GOTO(this->name, linked_inode, out); + GF_VALIDATE_OR_GOTO(this->name, md, out); + GF_VALIDATE_OR_GOTO(this->name, entry, out); + + if (strncmp(sign->signature, (char *)md, sign->signaturelen) == 0) { + gf_msg_debug(this->name, 0, + "%s [GFID: %s | Brick: %s] " + "matches calculated checksum", + loc->path, uuid_utoa(linked_inode->gfid), + child->brick_path); + return 0; + } + + gf_msg(this->name, GF_LOG_DEBUG, 0, BRB_MSG_CHECKSUM_MISMATCH, + "Object checksum mismatch: %s [GFID: %s | Brick: %s]", loc->path, + uuid_utoa(linked_inode->gfid), child->brick_path); + gf_msg(this->name, GF_LOG_ALERT, 0, BRB_MSG_CHECKSUM_MISMATCH, + "CORRUPTION DETECTED: Object %s {Brick: %s | GFID: %s}", loc->path, + child->brick_path, uuid_utoa(linked_inode->gfid)); + + /* Perform bad-file marking */ + xattr = dict_new(); + if (!xattr) { + ret = -1; + goto out; + } + + ret = dict_set_int32(xattr, BITROT_OBJECT_BAD_KEY, _gf_true); + if (ret) { + gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_MARK_BAD_FILE, + "Error setting bad-file marker for %s [GFID: %s | " + "Brick: %s]", + loc->path, uuid_utoa(linked_inode->gfid), child->brick_path); + goto dictfree; + } + + gf_msg(this->name, GF_LOG_ALERT, 0, BRB_MSG_MARK_CORRUPTED, + "Marking" + " %s [GFID: %s | Brick: %s] as corrupted..", + loc->path, uuid_utoa(linked_inode->gfid), child->brick_path); + gf_event(EVENT_BITROT_BAD_FILE, "gfid=%s;path=%s;brick=%s", + uuid_utoa(linked_inode->gfid), loc->path, child->brick_path); + ret = syncop_fsetxattr(child->xl, fd, xattr, 0, NULL, NULL); + if (ret) + gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_MARK_BAD_FILE, + "Error marking object %s [GFID: %s] as corrupted", loc->path, + uuid_utoa(linked_inode->gfid)); + +dictfree: + dict_unref(xattr); +out: + return ret; +} + +/** + * "The Scrubber" + * + * Perform signature validation for a given object with the assumption + * that the signature is SHA256 (because signer as of now _always_ + * signs with SHA256). + */ +int +br_scrubber_scrub_begin(xlator_t *this, struct br_fsscan_entry *fsentry) +{ + int32_t ret = -1; + fd_t *fd = NULL; + loc_t loc = { + 0, + }; + struct iatt iatt = { + 0, + }; + struct iatt parent_buf = { + 0, + }; + pid_t pid = 0; + br_child_t *child = NULL; + unsigned char *md = NULL; + inode_t *linked_inode = NULL; + br_isignature_out_t *sign = NULL; + unsigned long signedversion = 0; + gf_dirent_t *entry = NULL; + br_private_t *priv = NULL; + loc_t *parent = NULL; + gf_boolean_t skip_stat = _gf_false; + uuid_t shard_root_gfid = { + 0, + }; + + GF_VALIDATE_OR_GOTO("bit-rot", fsentry, out); + + entry = fsentry->entry; + parent = &fsentry->parent; + child = fsentry->data; + + priv = this->private; + + GF_VALIDATE_OR_GOTO("bit-rot", entry, out); + GF_VALIDATE_OR_GOTO("bit-rot", parent, out); + GF_VALIDATE_OR_GOTO("bit-rot", child, out); + GF_VALIDATE_OR_GOTO("bit-rot", priv, out); + + pid = GF_CLIENT_PID_SCRUB; + + ret = br_prepare_loc(this, child, parent, entry, &loc); + if (!ret) + goto out; + + syncopctx_setfspid(&pid); + + ret = syncop_lookup(child->xl, &loc, &iatt, &parent_buf, NULL, NULL); + if (ret) { + br_log_object_path(this, "lookup", loc.path, -ret); + goto out; + } + + linked_inode = inode_link(loc.inode, parent->inode, loc.name, &iatt); + if (linked_inode) + inode_lookup(linked_inode); + + gf_msg_debug(this->name, 0, "Scrubbing object %s [GFID: %s]", entry->d_name, + uuid_utoa(linked_inode->gfid)); + + if (iatt.ia_type != IA_IFREG) { + gf_msg_debug(this->name, 0, "%s is not a regular file", entry->d_name); + ret = 0; + goto unref_inode; + } + + if (IS_DHT_LINKFILE_MODE((&iatt))) { + gf_msg_debug(this->name, 0, "%s is a dht sticky bit file", + entry->d_name); + ret = 0; + goto unref_inode; + } + + /* skip updating scrub statistics for shard entries */ + gf_uuid_parse(SHARD_ROOT_GFID, shard_root_gfid); + if (gf_uuid_compare(loc.pargfid, shard_root_gfid) == 0) + skip_stat = _gf_true; + + /** + * open() an fd for subsequent operations + */ + fd = fd_create(linked_inode, 0); + if (!fd) { + gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_FD_CREATE_FAILED, + "failed to create fd for inode %s", + uuid_utoa(linked_inode->gfid)); + goto unref_inode; + } + + ret = syncop_open(child->xl, &loc, O_RDWR, fd, NULL, NULL); + if (ret) { + br_log_object(this, "open", linked_inode->gfid, -ret); + ret = -1; + goto unrefd; + } + + fd_bind(fd); + + /** + * perform pre compute checks before initiating checksum + * computation + * - presence of bad object + * - signature staleness + */ + ret = bitd_scrub_pre_compute_check(this, child, fd, &signedversion, + &priv->scrub_stat, skip_stat); + if (ret) + goto unrefd; /* skip this object */ + + /* if all's good, proceed to calculate the hash */ + md = GF_MALLOC(SHA256_DIGEST_LENGTH, gf_common_mt_char); + if (!md) + goto unrefd; + + ret = br_calculate_obj_checksum(md, child, fd, &iatt); + if (ret) { + gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_CALC_ERROR, + "error calculating hash for object [GFID: %s]", + uuid_utoa(fd->inode->gfid)); + ret = -1; + goto free_md; + } + + /** + * perform post compute checks as an object's signature may have + * become stale while scrubber calculated checksum. + */ + ret = bitd_scrub_post_compute_check(this, child, fd, signedversion, &sign, + &priv->scrub_stat, skip_stat); + if (ret) + goto free_md; + + ret = bitd_compare_ckum(this, sign, md, linked_inode, entry, fd, child, + &loc); + + if (!skip_stat) + br_inc_scrubbed_file(&priv->scrub_stat); + + GF_FREE(sign); /* allocated on post-compute */ + + /** fd_unref() takes care of closing fd.. like syncop_close() */ + +free_md: + GF_FREE(md); +unrefd: + fd_unref(fd); +unref_inode: + inode_unref(linked_inode); +out: + loc_wipe(&loc); + return ret; +} + +static void +_br_lock_cleaner(void *arg) +{ + pthread_mutex_t *mutex = arg; + + pthread_mutex_unlock(mutex); +} + +static void +wait_for_scrubbing(xlator_t *this, struct br_scanfs *fsscan) +{ + br_private_t *priv = NULL; + struct br_scrubber *fsscrub = NULL; + + priv = this->private; + fsscrub = &priv->fsscrub; + + pthread_cleanup_push(_br_lock_cleaner, &fsscan->waitlock); + pthread_mutex_lock(&fsscan->waitlock); + { + pthread_cleanup_push(_br_lock_cleaner, &fsscrub->mutex); + pthread_mutex_lock(&fsscrub->mutex); + { + list_replace_init(&fsscan->queued, &fsscan->ready); + + /* wake up scrubbers */ + pthread_cond_broadcast(&fsscrub->cond); + } + pthread_mutex_unlock(&fsscrub->mutex); + pthread_cleanup_pop(0); + + while (fsscan->entries != 0) + pthread_cond_wait(&fsscan->waitcond, &fsscan->waitlock); + } + pthread_mutex_unlock(&fsscan->waitlock); + pthread_cleanup_pop(0); +} + +static void +_br_fsscan_inc_entry_count(struct br_scanfs *fsscan) +{ + fsscan->entries++; +} + +static void +_br_fsscan_dec_entry_count(struct br_scanfs *fsscan) +{ + if (--fsscan->entries == 0) { + pthread_mutex_lock(&fsscan->waitlock); + { + pthread_cond_signal(&fsscan->waitcond); + } + pthread_mutex_unlock(&fsscan->waitlock); + } +} + +static void +_br_fsscan_collect_entry(struct br_scanfs *fsscan, + struct br_fsscan_entry *fsentry) +{ + list_add_tail(&fsentry->list, &fsscan->queued); + _br_fsscan_inc_entry_count(fsscan); +} + +#define NR_ENTRIES (1 << 7) /* ..bulk scrubbing */ + +int +br_fsscanner_handle_entry(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent, + void *data) +{ + int32_t ret = -1; + int scrub = 0; + br_child_t *child = NULL; + xlator_t *this = NULL; + struct br_scanfs *fsscan = NULL; + struct br_fsscan_entry *fsentry = NULL; + + GF_VALIDATE_OR_GOTO("bit-rot", subvol, error_return); + GF_VALIDATE_OR_GOTO("bit-rot", data, error_return); + + child = data; + this = child->this; + fsscan = &child->fsscan; + + _mask_cancellation(); + + fsentry = GF_CALLOC(1, sizeof(*fsentry), gf_br_mt_br_fsscan_entry_t); + if (!fsentry) + goto error_return; + + { + fsentry->data = data; + fsentry->fsscan = &child->fsscan; + + /* copy parent loc */ + ret = loc_copy(&fsentry->parent, parent); + if (ret) + goto dealloc; + + /* copy child entry */ + fsentry->entry = entry_copy(entry); + if (!fsentry->entry) + goto locwipe; + + INIT_LIST_HEAD(&fsentry->list); + } + + LOCK(&fsscan->entrylock); + { + _br_fsscan_collect_entry(fsscan, fsentry); + + /** + * need not be a equality check as entries may be pushed + * back onto the scanned queue when thread(s) are cleaned. + */ + if (fsscan->entries >= NR_ENTRIES) + scrub = 1; + } + UNLOCK(&fsscan->entrylock); + + _unmask_cancellation(); + + if (scrub) + wait_for_scrubbing(this, fsscan); + + return 0; + +locwipe: + loc_wipe(&fsentry->parent); +dealloc: + GF_FREE(fsentry); +error_return: + return -1; +} + +int32_t +br_fsscan_deactivate(xlator_t *this) +{ + int ret = 0; + br_private_t *priv = NULL; + br_scrub_state_t nstate = 0; + struct br_monitor *scrub_monitor = NULL; + + priv = this->private; + scrub_monitor = &priv->scrub_monitor; + + ret = gf_tw_del_timer(priv->timer_wheel, scrub_monitor->timer); + if (ret == 0) { + nstate = BR_SCRUB_STATE_STALLED; + gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO, + "Volume is under active scrubbing. Pausing scrub.."); + } else { + nstate = BR_SCRUB_STATE_PAUSED; + gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO, + "Scrubber paused"); + } + + _br_monitor_set_scrub_state(scrub_monitor, nstate); + + return 0; +} + +static void +br_scrubber_log_time(xlator_t *this, const char *sfx) +{ + char timestr[GF_TIMESTR_SIZE] = { + 0, + }; + br_private_t *priv = NULL; + time_t now = 0; + + now = gf_time(); + priv = this->private; + + gf_time_fmt(timestr, sizeof(timestr), now, gf_timefmt_FT); + + if (strcasecmp(sfx, "started") == 0) { + br_update_scrub_start_time(&priv->scrub_stat, now); + gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_START, + "Scrubbing %s at %s", sfx, timestr); + } else { + br_update_scrub_finish_time(&priv->scrub_stat, timestr, now); + gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_FINISH, + "Scrubbing %s at %s", sfx, timestr); + } +} + +static void +br_fsscanner_log_time(xlator_t *this, br_child_t *child, const char *sfx) +{ + char timestr[GF_TIMESTR_SIZE] = { + 0, + }; + time_t now = 0; + + now = gf_time(); + gf_time_fmt(timestr, sizeof(timestr), now, gf_timefmt_FT); + + if (strcasecmp(sfx, "started") == 0) { + gf_msg_debug(this->name, 0, "Scrubbing \"%s\" %s at %s", + child->brick_path, sfx, timestr); + } else { + gf_msg_debug(this->name, 0, "Scrubbing \"%s\" %s at %s", + child->brick_path, sfx, timestr); + } +} + +void +br_child_set_scrub_state(br_child_t *child, gf_boolean_t state) +{ + child->active_scrubbing = state; +} + +static void +br_fsscanner_wait_until_kicked(xlator_t *this, br_child_t *child) +{ + br_private_t *priv = NULL; + struct br_monitor *scrub_monitor = NULL; + + priv = this->private; + scrub_monitor = &priv->scrub_monitor; + + pthread_cleanup_push(_br_lock_cleaner, &scrub_monitor->wakelock); + pthread_mutex_lock(&scrub_monitor->wakelock); + { + while (!scrub_monitor->kick) + pthread_cond_wait(&scrub_monitor->wakecond, + &scrub_monitor->wakelock); + + /* Child lock is to synchronize with disconnect events */ + pthread_cleanup_push(_br_lock_cleaner, &child->lock); + pthread_mutex_lock(&child->lock); + { + scrub_monitor->active_child_count++; + br_child_set_scrub_state(child, _gf_true); + } + pthread_mutex_unlock(&child->lock); + pthread_cleanup_pop(0); + } + pthread_mutex_unlock(&scrub_monitor->wakelock); + pthread_cleanup_pop(0); +} + +static void +br_scrubber_entry_control(xlator_t *this) +{ + br_private_t *priv = NULL; + struct br_monitor *scrub_monitor = NULL; + + priv = this->private; + scrub_monitor = &priv->scrub_monitor; + + LOCK(&scrub_monitor->lock); + { + /* Move the state to BR_SCRUB_STATE_ACTIVE */ + if (scrub_monitor->state == BR_SCRUB_STATE_PENDING) + scrub_monitor->state = BR_SCRUB_STATE_ACTIVE; + br_scrubber_log_time(this, "started"); + priv->scrub_stat.scrub_running = 1; + } + UNLOCK(&scrub_monitor->lock); +} + +static void +br_scrubber_exit_control(xlator_t *this) +{ + br_private_t *priv = NULL; + struct br_monitor *scrub_monitor = NULL; + + priv = this->private; + scrub_monitor = &priv->scrub_monitor; + + LOCK(&scrub_monitor->lock); + { + br_scrubber_log_time(this, "finished"); + priv->scrub_stat.scrub_running = 0; + + if (scrub_monitor->state == BR_SCRUB_STATE_ACTIVE) { + (void)br_fsscan_activate(this); + } else { + UNLOCK(&scrub_monitor->lock); + gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO, + "Volume waiting to get rescheduled.."); + return; + } + } + UNLOCK(&scrub_monitor->lock); +} + +static void +br_fsscanner_entry_control(xlator_t *this, br_child_t *child) +{ + br_fsscanner_log_time(this, child, "started"); +} + +static void +br_fsscanner_exit_control(xlator_t *this, br_child_t *child) +{ + br_private_t *priv = NULL; + struct br_monitor *scrub_monitor = NULL; + + priv = this->private; + scrub_monitor = &priv->scrub_monitor; + + if (!_br_is_child_connected(child)) { + gf_msg(this->name, GF_LOG_WARNING, 0, BRB_MSG_SCRUB_INFO, + "Brick [%s] disconnected while scrubbing. Scrubbing " + "might be incomplete", + child->brick_path); + } + + br_fsscanner_log_time(this, child, "finished"); + + pthread_cleanup_push(_br_lock_cleaner, &scrub_monitor->wakelock); + pthread_mutex_lock(&scrub_monitor->wakelock); + { + scrub_monitor->active_child_count--; + pthread_cleanup_push(_br_lock_cleaner, &child->lock); + pthread_mutex_lock(&child->lock); + { + br_child_set_scrub_state(child, _gf_false); + } + pthread_mutex_unlock(&child->lock); + pthread_cleanup_pop(0); + + if (scrub_monitor->active_child_count == 0) { + /* The last child has finished scrubbing. + * Set the kick to false and wake up other + * children who are waiting for the last + * child to complete scrubbing. + */ + scrub_monitor->kick = _gf_false; + pthread_cond_broadcast(&scrub_monitor->wakecond); + + /* Signal monitor thread waiting for the all + * the children to finish scrubbing. + */ + pthread_cleanup_push(_br_lock_cleaner, &scrub_monitor->donelock); + pthread_mutex_lock(&scrub_monitor->donelock); + { + scrub_monitor->done = _gf_true; + pthread_cond_signal(&scrub_monitor->donecond); + } + pthread_mutex_unlock(&scrub_monitor->donelock); + pthread_cleanup_pop(0); + } else { + while (scrub_monitor->active_child_count) + pthread_cond_wait(&scrub_monitor->wakecond, + &scrub_monitor->wakelock); + } + } + pthread_mutex_unlock(&scrub_monitor->wakelock); + pthread_cleanup_pop(0); +} + +void * +br_fsscanner(void *arg) +{ + loc_t loc = { + 0, + }; + br_child_t *child = NULL; + xlator_t *this = NULL; + struct br_scanfs *fsscan = NULL; + + child = arg; + this = child->this; + fsscan = &child->fsscan; + + THIS = this; + loc.inode = child->table->root; + + while (1) { + br_fsscanner_wait_until_kicked(this, child); + { + /* precursor for scrub */ + br_fsscanner_entry_control(this, child); + + /* scrub */ + (void)syncop_ftw(child->xl, &loc, GF_CLIENT_PID_SCRUB, child, + br_fsscanner_handle_entry); + if (!list_empty(&fsscan->queued)) + wait_for_scrubbing(this, fsscan); + + /* scrub exit criteria */ + br_fsscanner_exit_control(this, child); + } + } + + return NULL; +} + +/** + * Keep this routine extremely simple and do not ever try to acquire + * child->lock here: it may lead to deadlock. Scrubber state is + * modified in br_fsscanner(). An intermediate state change to pause + * changes the scrub state to the _correct_ state by identifying a + * non-pending timer. + */ +void +br_kickstart_scanner(struct gf_tw_timer_list *timer, void *data, + unsigned long calltime) +{ + xlator_t *this = NULL; + struct br_monitor *scrub_monitor = data; + br_private_t *priv = NULL; + + THIS = this = scrub_monitor->this; + priv = this->private; + + /* Reset scrub statistics */ + priv->scrub_stat.scrubbed_files = 0; + priv->scrub_stat.unsigned_files = 0; + + /* Moves state from PENDING to ACTIVE */ + (void)br_scrubber_entry_control(this); + + /* kickstart scanning.. */ + pthread_mutex_lock(&scrub_monitor->wakelock); + { + scrub_monitor->kick = _gf_true; + GF_ASSERT(scrub_monitor->active_child_count == 0); + pthread_cond_broadcast(&scrub_monitor->wakecond); + } + pthread_mutex_unlock(&scrub_monitor->wakelock); + + return; +} + +static uint32_t +br_fsscan_calculate_delta(uint32_t times) +{ + return times; +} + +#define BR_SCRUB_ONDEMAND (1) +#define BR_SCRUB_MINUTE (60) +#define BR_SCRUB_HOURLY (60 * 60) +#define BR_SCRUB_DAILY (1 * 24 * 60 * 60) +#define BR_SCRUB_WEEKLY (7 * 24 * 60 * 60) +#define BR_SCRUB_BIWEEKLY (14 * 24 * 60 * 60) +#define BR_SCRUB_MONTHLY (30 * 24 * 60 * 60) + +static unsigned int +br_fsscan_calculate_timeout(scrub_freq_t freq) +{ + uint32_t timo = 0; + + switch (freq) { + case BR_FSSCRUB_FREQ_MINUTE: + timo = br_fsscan_calculate_delta(BR_SCRUB_MINUTE); + break; + case BR_FSSCRUB_FREQ_HOURLY: + timo = br_fsscan_calculate_delta(BR_SCRUB_HOURLY); + break; + case BR_FSSCRUB_FREQ_DAILY: + timo = br_fsscan_calculate_delta(BR_SCRUB_DAILY); + break; + case BR_FSSCRUB_FREQ_WEEKLY: + timo = br_fsscan_calculate_delta(BR_SCRUB_WEEKLY); + break; + case BR_FSSCRUB_FREQ_BIWEEKLY: + timo = br_fsscan_calculate_delta(BR_SCRUB_BIWEEKLY); + break; + case BR_FSSCRUB_FREQ_MONTHLY: + timo = br_fsscan_calculate_delta(BR_SCRUB_MONTHLY); + break; + default: + timo = 0; + } + + return timo; +} + +int32_t +br_fsscan_schedule(xlator_t *this) +{ + uint32_t timo = 0; + br_private_t *priv = NULL; + char timestr[GF_TIMESTR_SIZE] = { + 0, + }; + struct br_scrubber *fsscrub = NULL; + struct gf_tw_timer_list *timer = NULL; + struct br_monitor *scrub_monitor = NULL; + + priv = this->private; + fsscrub = &priv->fsscrub; + scrub_monitor = &priv->scrub_monitor; + + scrub_monitor->boot = gf_time(); + + timo = br_fsscan_calculate_timeout(fsscrub->frequency); + if (timo == 0) { + gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_ZERO_TIMEOUT_BUG, + "BUG: Zero schedule timeout"); + goto error_return; + } + + scrub_monitor->timer = GF_CALLOC(1, sizeof(*scrub_monitor->timer), + gf_br_stub_mt_br_scanner_freq_t); + if (!scrub_monitor->timer) + goto error_return; + + timer = scrub_monitor->timer; + INIT_LIST_HEAD(&timer->entry); + + timer->data = scrub_monitor; + timer->expires = timo; + timer->function = br_kickstart_scanner; + + gf_tw_add_timer(priv->timer_wheel, timer); + _br_monitor_set_scrub_state(scrub_monitor, BR_SCRUB_STATE_PENDING); + + gf_time_fmt(timestr, sizeof(timestr), (scrub_monitor->boot + timo), + gf_timefmt_FT); + gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO, + "Scrubbing is " + "scheduled to run at %s", + timestr); + + return 0; + +error_return: + return -1; +} + +int32_t +br_fsscan_activate(xlator_t *this) +{ + uint32_t timo = 0; + char timestr[GF_TIMESTR_SIZE] = { + 0, + }; + time_t now = 0; + br_private_t *priv = NULL; + struct br_scrubber *fsscrub = NULL; + struct br_monitor *scrub_monitor = NULL; + + priv = this->private; + fsscrub = &priv->fsscrub; + scrub_monitor = &priv->scrub_monitor; + + now = gf_time(); + timo = br_fsscan_calculate_timeout(fsscrub->frequency); + if (timo == 0) { + gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_ZERO_TIMEOUT_BUG, + "BUG: Zero schedule timeout"); + return -1; + } + + pthread_mutex_lock(&scrub_monitor->donelock); + { + scrub_monitor->done = _gf_false; + } + pthread_mutex_unlock(&scrub_monitor->donelock); + + gf_time_fmt(timestr, sizeof(timestr), now + timo, gf_timefmt_FT); + (void)gf_tw_mod_timer(priv->timer_wheel, scrub_monitor->timer, timo); + + _br_monitor_set_scrub_state(scrub_monitor, BR_SCRUB_STATE_PENDING); + gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO, + "Scrubbing is " + "rescheduled to run at %s", + timestr); + + return 0; +} + +int32_t +br_fsscan_reschedule(xlator_t *this) +{ + int32_t ret = 0; + uint32_t timo = 0; + char timestr[GF_TIMESTR_SIZE] = { + 0, + }; + time_t now = 0; + br_private_t *priv = NULL; + struct br_scrubber *fsscrub = NULL; + struct br_monitor *scrub_monitor = NULL; + + priv = this->private; + fsscrub = &priv->fsscrub; + scrub_monitor = &priv->scrub_monitor; + + if (!fsscrub->frequency_reconf) + return 0; + + now = gf_time(); + timo = br_fsscan_calculate_timeout(fsscrub->frequency); + if (timo == 0) { + gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_ZERO_TIMEOUT_BUG, + "BUG: Zero schedule timeout"); + return -1; + } + + gf_time_fmt(timestr, sizeof(timestr), now + timo, gf_timefmt_FT); + + pthread_mutex_lock(&scrub_monitor->donelock); + { + scrub_monitor->done = _gf_false; + } + pthread_mutex_unlock(&scrub_monitor->donelock); + + ret = gf_tw_mod_timer_pending(priv->timer_wheel, scrub_monitor->timer, + timo); + if (ret == 0) + gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO, + "Scrubber is currently running and would be " + "rescheduled after completion"); + else { + _br_monitor_set_scrub_state(scrub_monitor, BR_SCRUB_STATE_PENDING); + gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO, + "Scrubbing rescheduled to run at %s", timestr); + } + + return 0; +} + +int32_t +br_fsscan_ondemand(xlator_t *this) +{ + int32_t ret = 0; + uint32_t timo = 0; + char timestr[GF_TIMESTR_SIZE] = { + 0, + }; + time_t now = 0; + br_private_t *priv = NULL; + struct br_monitor *scrub_monitor = NULL; + + priv = this->private; + scrub_monitor = &priv->scrub_monitor; + + now = gf_time(); + timo = BR_SCRUB_ONDEMAND; + gf_time_fmt(timestr, sizeof(timestr), now + timo, gf_timefmt_FT); + + pthread_mutex_lock(&scrub_monitor->donelock); + { + scrub_monitor->done = _gf_false; + } + pthread_mutex_unlock(&scrub_monitor->donelock); + + ret = gf_tw_mod_timer_pending(priv->timer_wheel, scrub_monitor->timer, + timo); + if (ret == 0) + gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO, + "Scrubber is currently running and would be " + "rescheduled after completion"); + else { + _br_monitor_set_scrub_state(scrub_monitor, BR_SCRUB_STATE_PENDING); + gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO, + "Ondemand Scrubbing scheduled to run at %s", timestr); + } + + return 0; +} + +#define BR_SCRUB_THREAD_SCALE_LAZY 0 +#define BR_SCRUB_THREAD_SCALE_NORMAL 0.4 +#define BR_SCRUB_THREAD_SCALE_AGGRESSIVE 1.0 + +#ifndef M_E +#define M_E 2.718 +#endif + +/** + * This is just a simple exponential scale to a fixed value selected + * per throttle config. We probably need to be more smart and select + * the scale based on the number of processor cores too. + */ +static unsigned int +br_scrubber_calc_scale(xlator_t *this, br_private_t *priv, + scrub_throttle_t throttle) +{ + unsigned int scale = 0; + + switch (throttle) { + case BR_SCRUB_THROTTLE_VOID: + case BR_SCRUB_THROTTLE_STALLED: + scale = 0; + break; + case BR_SCRUB_THROTTLE_LAZY: + scale = priv->child_count * pow(M_E, BR_SCRUB_THREAD_SCALE_LAZY); + break; + case BR_SCRUB_THROTTLE_NORMAL: + scale = priv->child_count * pow(M_E, BR_SCRUB_THREAD_SCALE_NORMAL); + break; + case BR_SCRUB_THROTTLE_AGGRESSIVE: + scale = priv->child_count * + pow(M_E, BR_SCRUB_THREAD_SCALE_AGGRESSIVE); + break; + default: + gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_UNKNOWN_THROTTLE, + "Unknown throttle %d", throttle); + } + + return scale; +} + +static br_child_t * +_br_scrubber_get_next_child(struct br_scrubber *fsscrub) +{ + br_child_t *child = NULL; + + child = list_first_entry(&fsscrub->scrublist, br_child_t, list); + list_rotate_left(&fsscrub->scrublist); + + return child; +} + +static void +_br_scrubber_get_entry(br_child_t *child, struct br_fsscan_entry **fsentry) +{ + struct br_scanfs *fsscan = &child->fsscan; + + if (list_empty(&fsscan->ready)) + return; + *fsentry = list_first_entry(&fsscan->ready, struct br_fsscan_entry, list); + list_del_init(&(*fsentry)->list); +} + +static void +_br_scrubber_find_scrubbable_entry(struct br_scrubber *fsscrub, + struct br_fsscan_entry **fsentry) +{ + br_child_t *child = NULL; + br_child_t *firstchild = NULL; + + while (1) { + while (list_empty(&fsscrub->scrublist)) + pthread_cond_wait(&fsscrub->cond, &fsscrub->mutex); + + firstchild = NULL; + for (child = _br_scrubber_get_next_child(fsscrub); child != firstchild; + child = _br_scrubber_get_next_child(fsscrub)) { + if (!firstchild) + firstchild = child; + + _br_scrubber_get_entry(child, fsentry); + if (*fsentry) + break; + } + + if (*fsentry) + break; + + /* nothing to work on.. wait till available */ + pthread_cond_wait(&fsscrub->cond, &fsscrub->mutex); + } +} + +static void +br_scrubber_pick_entry(struct br_scrubber *fsscrub, + struct br_fsscan_entry **fsentry) +{ + pthread_cleanup_push(_br_lock_cleaner, &fsscrub->mutex); + + pthread_mutex_lock(&fsscrub->mutex); + { + *fsentry = NULL; + _br_scrubber_find_scrubbable_entry(fsscrub, fsentry); + } + pthread_mutex_unlock(&fsscrub->mutex); + + pthread_cleanup_pop(0); +} + +struct br_scrub_entry { + gf_boolean_t scrubbed; + struct br_fsscan_entry *fsentry; +}; + +/** + * We need to be a bit careful here. These thread(s) are prone to cancellations + * when threads are scaled down (depending on the thottling value configured) + * and pausing scrub. A thread can get cancelled while it's waiting for entries + * in the ->pending queue or when an object is undergoing scrubbing. + */ +static void +br_scrubber_entry_handle(void *arg) +{ + struct br_scanfs *fsscan = NULL; + struct br_scrub_entry *sentry = NULL; + struct br_fsscan_entry *fsentry = NULL; + + sentry = arg; + + fsentry = sentry->fsentry; + fsscan = fsentry->fsscan; + + LOCK(&fsscan->entrylock); + { + if (sentry->scrubbed) { + _br_fsscan_dec_entry_count(fsscan); + + /* cleanup ->entry */ + fsentry->data = NULL; + fsentry->fsscan = NULL; + loc_wipe(&fsentry->parent); + gf_dirent_entry_free(fsentry->entry); + + GF_FREE(sentry->fsentry); + } else { + /* (re)queue the entry again for scrub */ + _br_fsscan_collect_entry(fsscan, sentry->fsentry); + } + } + UNLOCK(&fsscan->entrylock); +} + +static void +br_scrubber_scrub_entry(xlator_t *this, struct br_fsscan_entry *fsentry) +{ + struct br_scrub_entry sentry = { + 0, + }; + + sentry.scrubbed = 0; + sentry.fsentry = fsentry; + + pthread_cleanup_push(br_scrubber_entry_handle, &sentry); + { + (void)br_scrubber_scrub_begin(this, fsentry); + sentry.scrubbed = 1; + } + pthread_cleanup_pop(1); +} + +void * +br_scrubber_proc(void *arg) +{ + xlator_t *this = NULL; + struct br_scrubber *fsscrub = NULL; + struct br_fsscan_entry *fsentry = NULL; + + fsscrub = arg; + THIS = this = fsscrub->this; + + while (1) { + br_scrubber_pick_entry(fsscrub, &fsentry); + br_scrubber_scrub_entry(this, fsentry); + sleep(1); + } + + return NULL; +} + +static int32_t +br_scrubber_scale_up(xlator_t *this, struct br_scrubber *fsscrub, + unsigned int v1, unsigned int v2) +{ + int i = 0; + int32_t ret = -1; + int diff = 0; + struct br_scrubbers *scrub = NULL; + + diff = (int)(v2 - v1); + + gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCALING_UP_SCRUBBER, + "Scaling up scrubbers [%d => %d]", v1, v2); + + for (i = 0; i < diff; i++) { + scrub = GF_CALLOC(diff, sizeof(*scrub), gf_br_mt_br_scrubber_t); + if (!scrub) + break; + + INIT_LIST_HEAD(&scrub->list); + ret = gf_thread_create(&scrub->scrubthread, NULL, br_scrubber_proc, + fsscrub, "brsproc"); + if (ret) + break; + + fsscrub->nr_scrubbers++; + list_add_tail(&scrub->list, &fsscrub->scrubbers); + } + + if ((i != diff) && !scrub) + goto error_return; + + if (i != diff) /* degraded scaling.. */ + gf_msg(this->name, GF_LOG_WARNING, 0, BRB_MSG_SCALE_UP_FAILED, + "Could not fully scale up to %d scrubber(s). Spawned " + "%d/%d [total scrubber(s): %d]", + v2, i, diff, (v1 + i)); + + return 0; + +error_return: + return -1; +} + +static int32_t +br_scrubber_scale_down(xlator_t *this, struct br_scrubber *fsscrub, + unsigned int v1, unsigned int v2) +{ + int i = 0; + int diff = 0; + int32_t ret = -1; + struct br_scrubbers *scrub = NULL; + + diff = (int)(v1 - v2); + + gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCALE_DOWN_SCRUBBER, + "Scaling down scrubbers [%d => %d]", v1, v2); + + for (i = 0; i < diff; i++) { + scrub = list_first_entry(&fsscrub->scrubbers, struct br_scrubbers, + list); + + list_del_init(&scrub->list); + ret = gf_thread_cleanup_xint(scrub->scrubthread); + if (ret) + break; + GF_FREE(scrub); + + fsscrub->nr_scrubbers--; + } + + if (ret) { + gf_msg(this->name, GF_LOG_WARNING, 0, BRB_MSG_SCALE_DOWN_FAILED, + "Could not fully scale down " + "to %d scrubber(s). Terminated %d/%d [total " + "scrubber(s): %d]", + v1, i, diff, (v2 - i)); + ret = 0; + } + + return ret; +} + +static int32_t +br_scrubber_configure(xlator_t *this, br_private_t *priv, + struct br_scrubber *fsscrub, scrub_throttle_t nthrottle) +{ + int32_t ret = 0; + unsigned int v1 = 0; + unsigned int v2 = 0; + + v1 = fsscrub->nr_scrubbers; + v2 = br_scrubber_calc_scale(this, priv, nthrottle); + + if (v1 == v2) + return 0; + + if (v1 > v2) + ret = br_scrubber_scale_down(this, fsscrub, v1, v2); + else + ret = br_scrubber_scale_up(this, fsscrub, v1, v2); + + return ret; +} + +static int32_t +br_scrubber_fetch_option(xlator_t *this, char *opt, dict_t *options, + char **value) +{ + if (options) + GF_OPTION_RECONF(opt, *value, options, str, error_return); + else + GF_OPTION_INIT(opt, *value, str, error_return); + + return 0; + +error_return: + return -1; +} + +/* internal "throttle" override */ +#define BR_SCRUB_STALLED "STALLED" + +/* TODO: token buket spec */ +static int32_t +br_scrubber_handle_throttle(xlator_t *this, br_private_t *priv, dict_t *options, + gf_boolean_t scrubstall) +{ + int32_t ret = 0; + char *tmp = NULL; + struct br_scrubber *fsscrub = NULL; + scrub_throttle_t nthrottle = BR_SCRUB_THROTTLE_VOID; + + fsscrub = &priv->fsscrub; + fsscrub->throttle_reconf = _gf_false; + + ret = br_scrubber_fetch_option(this, "scrub-throttle", options, &tmp); + if (ret) + goto error_return; + + if (scrubstall) + tmp = BR_SCRUB_STALLED; + + if (strcasecmp(tmp, "lazy") == 0) + nthrottle = BR_SCRUB_THROTTLE_LAZY; + else if (strcasecmp(tmp, "normal") == 0) + nthrottle = BR_SCRUB_THROTTLE_NORMAL; + else if (strcasecmp(tmp, "aggressive") == 0) + nthrottle = BR_SCRUB_THROTTLE_AGGRESSIVE; + else if (strcasecmp(tmp, BR_SCRUB_STALLED) == 0) + nthrottle = BR_SCRUB_THROTTLE_STALLED; + else + goto error_return; + + /* on failure old throttling value is preserved */ + ret = br_scrubber_configure(this, priv, fsscrub, nthrottle); + if (ret) + goto error_return; + + if (fsscrub->throttle != nthrottle) + fsscrub->throttle_reconf = _gf_true; + + fsscrub->throttle = nthrottle; + return 0; + +error_return: + return -1; +} + +static int32_t +br_scrubber_handle_stall(xlator_t *this, br_private_t *priv, dict_t *options, + gf_boolean_t *scrubstall) +{ + int32_t ret = 0; + char *tmp = NULL; + + ret = br_scrubber_fetch_option(this, "scrub-state", options, &tmp); + if (ret) + goto error_return; + + if (strcasecmp(tmp, "pause") == 0) /* anything else is active */ + *scrubstall = _gf_true; + + return 0; + +error_return: + return -1; +} + +static int32_t +br_scrubber_handle_freq(xlator_t *this, br_private_t *priv, dict_t *options, + gf_boolean_t scrubstall) +{ + int32_t ret = -1; + char *tmp = NULL; + scrub_freq_t frequency = BR_FSSCRUB_FREQ_HOURLY; + struct br_scrubber *fsscrub = NULL; + + fsscrub = &priv->fsscrub; + fsscrub->frequency_reconf = _gf_true; + + ret = br_scrubber_fetch_option(this, "scrub-freq", options, &tmp); + if (ret) + goto error_return; + + if (scrubstall) + tmp = BR_SCRUB_STALLED; + + if (strcasecmp(tmp, "hourly") == 0) { + frequency = BR_FSSCRUB_FREQ_HOURLY; + } else if (strcasecmp(tmp, "daily") == 0) { + frequency = BR_FSSCRUB_FREQ_DAILY; + } else if (strcasecmp(tmp, "weekly") == 0) { + frequency = BR_FSSCRUB_FREQ_WEEKLY; + } else if (strcasecmp(tmp, "biweekly") == 0) { + frequency = BR_FSSCRUB_FREQ_BIWEEKLY; + } else if (strcasecmp(tmp, "monthly") == 0) { + frequency = BR_FSSCRUB_FREQ_MONTHLY; + } else if (strcasecmp(tmp, "minute") == 0) { + frequency = BR_FSSCRUB_FREQ_MINUTE; + } else if (strcasecmp(tmp, BR_SCRUB_STALLED) == 0) { + frequency = BR_FSSCRUB_FREQ_STALLED; + } else + goto error_return; + + if (fsscrub->frequency == frequency) + fsscrub->frequency_reconf = _gf_false; + else + fsscrub->frequency = frequency; + + return 0; + +error_return: + return -1; +} + +static void +br_scrubber_log_option(xlator_t *this, br_private_t *priv, + gf_boolean_t scrubstall) +{ + struct br_scrubber *fsscrub = &priv->fsscrub; + char *scrub_throttle_str[] = { + [BR_SCRUB_THROTTLE_LAZY] = "lazy", + [BR_SCRUB_THROTTLE_NORMAL] = "normal", + [BR_SCRUB_THROTTLE_AGGRESSIVE] = "aggressive", + [BR_SCRUB_THROTTLE_STALLED] = "stalled", + }; + + char *scrub_freq_str[] = { + [0] = "", + [BR_FSSCRUB_FREQ_HOURLY] = "hourly", + [BR_FSSCRUB_FREQ_DAILY] = "daily", + [BR_FSSCRUB_FREQ_WEEKLY] = "weekly", + [BR_FSSCRUB_FREQ_BIWEEKLY] = "biweekly", + [BR_FSSCRUB_FREQ_MONTHLY] = "monthly (30 days)", + [BR_FSSCRUB_FREQ_MINUTE] = "every minute", + }; + + if (scrubstall) + return; /* logged as pause */ + + if (fsscrub->frequency_reconf || fsscrub->throttle_reconf) { + if (fsscrub->throttle == BR_SCRUB_THROTTLE_VOID) + return; + gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_TUNABLE, + "SCRUB TUNABLES:: [Frequency: %s, Throttle: %s]", + scrub_freq_str[fsscrub->frequency], + scrub_throttle_str[fsscrub->throttle]); + } +} + +int32_t +br_scrubber_handle_options(xlator_t *this, br_private_t *priv, dict_t *options) +{ + int32_t ret = 0; + gf_boolean_t scrubstall = _gf_false; /* not as dangerous as it sounds */ + + ret = br_scrubber_handle_stall(this, priv, options, &scrubstall); + if (ret) + goto error_return; + + ret = br_scrubber_handle_throttle(this, priv, options, scrubstall); + if (ret) + goto error_return; + + ret = br_scrubber_handle_freq(this, priv, options, scrubstall); + if (ret) + goto error_return; + + br_scrubber_log_option(this, priv, scrubstall); + + return 0; + +error_return: + return -1; +} + +inode_t * +br_lookup_bad_obj_dir(xlator_t *this, br_child_t *child, uuid_t gfid) +{ + struct iatt statbuf = { + 0, + }; + inode_table_t *table = NULL; + int32_t ret = -1; + loc_t loc = { + 0, + }; + inode_t *linked_inode = NULL; + int32_t op_errno = 0; + + GF_VALIDATE_OR_GOTO("bit-rot-scrubber", this, out); + GF_VALIDATE_OR_GOTO(this->name, this->private, out); + GF_VALIDATE_OR_GOTO(this->name, child, out); + + table = child->table; + + loc.inode = inode_new(table); + if (!loc.inode) { + gf_msg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_NO_MEMORY, + "failed to allocate a new inode for" + "bad object directory"); + goto out; + } + + gf_uuid_copy(loc.gfid, gfid); + + ret = syncop_lookup(child->xl, &loc, &statbuf, NULL, NULL, NULL); + if (ret < 0) { + op_errno = -ret; + gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_LOOKUP_FAILED, + "failed to lookup the bad " + "objects directory (gfid: %s (%s))", + uuid_utoa(gfid), strerror(op_errno)); + goto out; + } + + linked_inode = inode_link(loc.inode, NULL, NULL, &statbuf); + if (linked_inode) + inode_lookup(linked_inode); + +out: + loc_wipe(&loc); + return linked_inode; +} + +int32_t +br_read_bad_object_dir(xlator_t *this, br_child_t *child, fd_t *fd, + dict_t *dict) +{ + gf_dirent_t entries; + gf_dirent_t *entry = NULL; + int32_t ret = -1; + off_t offset = 0; + int32_t count = 0; + char key[32] = { + 0, + }; + dict_t *out_dict = NULL; + + INIT_LIST_HEAD(&entries.list); + + while ((ret = syncop_readdir(child->xl, fd, 131072, offset, &entries, NULL, + &out_dict))) { + if (ret < 0) + goto out; + + list_for_each_entry(entry, &entries.list, list) + { + offset = entry->d_off; + + snprintf(key, sizeof(key), "quarantine-%d", count); + + /* + * ignore the dict_set errors for now. The intention is + * to get as many bad objects as possible instead of + * erroring out at the first failure. + */ + ret = dict_set_dynstr_with_alloc(dict, key, entry->d_name); + if (!ret) + count++; + + if (out_dict) { + dict_copy(out_dict, dict); + dict_unref(out_dict); + out_dict = NULL; + } + } + + gf_dirent_free(&entries); + } + + ret = count; + ret = dict_set_int32_sizen(dict, "count", count); + +out: + return ret; +} + +int32_t +br_get_bad_objects_from_child(xlator_t *this, dict_t *dict, br_child_t *child) +{ + inode_t *inode = NULL; + inode_table_t *table = NULL; + fd_t *fd = NULL; + int32_t ret = -1; + loc_t loc = { + 0, + }; + int32_t op_errno = 0; + + GF_VALIDATE_OR_GOTO("bit-rot-scrubber", this, out); + GF_VALIDATE_OR_GOTO(this->name, this->private, out); + GF_VALIDATE_OR_GOTO(this->name, child, out); + GF_VALIDATE_OR_GOTO(this->name, dict, out); + + table = child->table; + + inode = inode_find(table, BR_BAD_OBJ_CONTAINER); + if (!inode) { + inode = br_lookup_bad_obj_dir(this, child, BR_BAD_OBJ_CONTAINER); + if (!inode) + goto out; + } + + fd = fd_create(inode, 0); + if (!fd) { + gf_msg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_FD_CREATE_FAILED, + "fd creation for the bad " + "objects directory failed (gfid: %s)", + uuid_utoa(BR_BAD_OBJ_CONTAINER)); + goto out; + } + + loc.inode = inode; + gf_uuid_copy(loc.gfid, inode->gfid); + + ret = syncop_opendir(child->xl, &loc, fd, NULL, NULL); + if (ret < 0) { + op_errno = -ret; + fd_unref(fd); + fd = NULL; + gf_msg(this->name, GF_LOG_ERROR, op_errno, BRB_MSG_FD_CREATE_FAILED, + "failed to open the bad " + "objects directory %s", + uuid_utoa(BR_BAD_OBJ_CONTAINER)); + goto out; + } + + fd_bind(fd); + + ret = br_read_bad_object_dir(this, child, fd, dict); + if (ret < 0) { + gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_BAD_OBJ_READDIR_FAIL, + "readdir of the bad " + "objects directory (%s) failed ", + uuid_utoa(BR_BAD_OBJ_CONTAINER)); + goto out; + } + + ret = 0; + +out: + loc_wipe(&loc); + if (fd) + fd_unref(fd); + return ret; +} + +int32_t +br_collect_bad_objects_of_child(xlator_t *this, br_child_t *child, dict_t *dict, + dict_t *child_dict, int32_t total_count) +{ + int32_t ret = -1; + int32_t count = 0; + char key[32] = { + 0, + }; + char main_key[32] = { + 0, + }; + int32_t j = 0; + int32_t tmp_count = 0; + char *entry = NULL; + char tmp[PATH_MAX] = { + 0, + }; + char *path = NULL; + int32_t len = 0; + + ret = dict_get_int32_sizen(child_dict, "count", &count); + if (ret) + goto out; + + tmp_count = total_count; + + for (j = 0; j < count; j++) { + len = snprintf(key, sizeof(key), "quarantine-%d", j); + ret = dict_get_strn(child_dict, key, len, &entry); + if (ret) + continue; + + ret = dict_get_str(child_dict, entry, &path); + len = snprintf(tmp, PATH_MAX, "%s ==> BRICK: %s\n path: %s", entry, + child->brick_path, path); + if ((len < 0) || (len >= PATH_MAX)) { + continue; + } + snprintf(main_key, sizeof(main_key), "quarantine-%d", tmp_count); + + ret = dict_set_dynstr_with_alloc(dict, main_key, tmp); + if (!ret) + tmp_count++; + path = NULL; + } + + ret = tmp_count; + +out: + return ret; +} + +int32_t +br_collect_bad_objects_from_children(xlator_t *this, dict_t *dict) +{ + int32_t ret = -1; + dict_t *child_dict = NULL; + int32_t i = 0; + int32_t total_count = 0; + br_child_t *child = NULL; + br_private_t *priv = NULL; + dict_t *tmp_dict = NULL; + + priv = this->private; + tmp_dict = dict; + + for (i = 0; i < priv->child_count; i++) { + child = &priv->children[i]; + GF_ASSERT(child); + if (!_br_is_child_connected(child)) + continue; + + child_dict = dict_new(); + if (!child_dict) { + gf_msg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_NO_MEMORY, + "failed to allocate dict"); + continue; + } + ret = br_get_bad_objects_from_child(this, child_dict, child); + /* + * Continue asking the remaining children for the list of + * bad objects even though getting the list from one of them + * fails. + */ + if (ret) { + dict_unref(child_dict); + continue; + } + + ret = br_collect_bad_objects_of_child(this, child, tmp_dict, child_dict, + total_count); + if (ret < 0) { + dict_unref(child_dict); + continue; + } + + total_count = ret; + dict_unref(child_dict); + child_dict = NULL; + } + + ret = dict_set_int32(tmp_dict, "total-count", total_count); + + return ret; +} + +int32_t +br_get_bad_objects_list(xlator_t *this, dict_t **dict) +{ + int32_t ret = -1; + dict_t *tmp_dict = NULL; + + GF_VALIDATE_OR_GOTO("bir-rot-scrubber", this, out); + GF_VALIDATE_OR_GOTO(this->name, dict, out); + + tmp_dict = *dict; + if (!tmp_dict) { + tmp_dict = dict_new(); + if (!tmp_dict) { + gf_msg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_NO_MEMORY, + "failed to allocate dict"); + goto out; + } + *dict = tmp_dict; + } + + ret = br_collect_bad_objects_from_children(this, tmp_dict); + +out: + return ret; +} + +static int +wait_for_scrub_to_finish(xlator_t *this) +{ + int ret = -1; + br_private_t *priv = NULL; + struct br_monitor *scrub_monitor = NULL; + + priv = this->private; + scrub_monitor = &priv->scrub_monitor; + + GF_VALIDATE_OR_GOTO("bit-rot", scrub_monitor, out); + GF_VALIDATE_OR_GOTO("bit-rot", this, out); + + gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO, + "Waiting for all children to start and finish scrub"); + + pthread_mutex_lock(&scrub_monitor->donelock); + { + while (!scrub_monitor->done) + pthread_cond_wait(&scrub_monitor->donecond, + &scrub_monitor->donelock); + } + pthread_mutex_unlock(&scrub_monitor->donelock); + ret = 0; +out: + return ret; +} + +/** + * This function is executed in a separate thread. This is scrubber monitor + * thread that takes care of state machine. + */ +void * +br_monitor_thread(void *arg) +{ + int32_t ret = 0; + xlator_t *this = NULL; + br_private_t *priv = NULL; + struct br_monitor *scrub_monitor = NULL; + + this = arg; + priv = this->private; + + /* + * Since, this is the topmost xlator, THIS has to be set by bit-rot + * xlator itself (STACK_WIND won't help in this case). Also it has + * to be done for each thread that gets spawned. Otherwise, a new + * thread will get global_xlator's pointer when it does "THIS". + */ + THIS = this; + + scrub_monitor = &priv->scrub_monitor; + + pthread_mutex_lock(&scrub_monitor->mutex); + { + while (!scrub_monitor->inited) + pthread_cond_wait(&scrub_monitor->cond, &scrub_monitor->mutex); + } + pthread_mutex_unlock(&scrub_monitor->mutex); + + /* this needs to be serialized with reconfigure() */ + pthread_mutex_lock(&priv->lock); + { + ret = br_scrub_state_machine(this, _gf_false); + } + pthread_mutex_unlock(&priv->lock); + if (ret) { + gf_msg(this->name, GF_LOG_ERROR, -ret, BRB_MSG_SSM_FAILED, + "Scrub state machine failed"); + goto out; + } + + while (1) { + /* Wait for all children to finish scrubbing */ + ret = wait_for_scrub_to_finish(this); + if (ret) { + gf_msg(this->name, GF_LOG_ERROR, -ret, BRB_MSG_SCRUB_WAIT_FAILED, + "Scrub wait failed"); + goto out; + } + + /* scrub exit criteria: Move the state to PENDING */ + br_scrubber_exit_control(this); + } + +out: + return NULL; +} + +static void +br_set_scrub_state(struct br_monitor *scrub_monitor, br_scrub_state_t state) +{ + LOCK(&scrub_monitor->lock); + { + _br_monitor_set_scrub_state(scrub_monitor, state); + } + UNLOCK(&scrub_monitor->lock); +} + +int32_t +br_scrubber_monitor_init(xlator_t *this, br_private_t *priv) +{ + struct br_monitor *scrub_monitor = NULL; + int ret = 0; + + scrub_monitor = &priv->scrub_monitor; + + LOCK_INIT(&scrub_monitor->lock); + scrub_monitor->this = this; + + scrub_monitor->inited = _gf_false; + pthread_mutex_init(&scrub_monitor->mutex, NULL); + pthread_cond_init(&scrub_monitor->cond, NULL); + + scrub_monitor->kick = _gf_false; + scrub_monitor->active_child_count = 0; + pthread_mutex_init(&scrub_monitor->wakelock, NULL); + pthread_cond_init(&scrub_monitor->wakecond, NULL); + + scrub_monitor->done = _gf_false; + pthread_mutex_init(&scrub_monitor->donelock, NULL); + pthread_cond_init(&scrub_monitor->donecond, NULL); + + /* Set the state to INACTIVE */ + br_set_scrub_state(&priv->scrub_monitor, BR_SCRUB_STATE_INACTIVE); + + /* Start the monitor thread */ + ret = gf_thread_create(&scrub_monitor->thread, NULL, br_monitor_thread, + this, "brmon"); + if (ret != 0) { + gf_msg(this->name, GF_LOG_ERROR, -ret, BRB_MSG_SPAWN_FAILED, + "monitor thread creation failed"); + ret = -1; + goto err; + } + + return 0; +err: + pthread_mutex_destroy(&scrub_monitor->mutex); + pthread_cond_destroy(&scrub_monitor->cond); + + pthread_mutex_destroy(&scrub_monitor->wakelock); + pthread_cond_destroy(&scrub_monitor->wakecond); + + pthread_mutex_destroy(&scrub_monitor->donelock); + pthread_cond_destroy(&scrub_monitor->donecond); + + LOCK_DESTROY(&scrub_monitor->lock); + + return ret; +} + +int32_t +br_scrubber_init(xlator_t *this, br_private_t *priv) +{ + struct br_scrubber *fsscrub = NULL; + int ret = 0; + + priv->tbf = tbf_init(NULL, 0); + if (!priv->tbf) + return -1; + + ret = br_scrubber_monitor_init(this, priv); + if (ret) + return -1; + + fsscrub = &priv->fsscrub; + + fsscrub->this = this; + fsscrub->throttle = BR_SCRUB_THROTTLE_VOID; + + pthread_mutex_init(&fsscrub->mutex, NULL); + pthread_cond_init(&fsscrub->cond, NULL); + + fsscrub->nr_scrubbers = 0; + INIT_LIST_HEAD(&fsscrub->scrubbers); + INIT_LIST_HEAD(&fsscrub->scrublist); + + return 0; +} diff --git a/xlators/features/bit-rot/src/bitd/bit-rot-scrub.h b/xlators/features/bit-rot/src/bitd/bit-rot-scrub.h new file mode 100644 index 00000000000..4e5f67bc021 --- /dev/null +++ b/xlators/features/bit-rot/src/bitd/bit-rot-scrub.h @@ -0,0 +1,46 @@ +/* + Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com> + 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. +*/ + +#ifndef __BIT_ROT_SCRUB_H__ +#define __BIT_ROT_SCRUB_H__ + +#include <glusterfs/xlator.h> +#include "bit-rot.h" + +void * +br_fsscanner(void *); + +int32_t +br_fsscan_schedule(xlator_t *); +int32_t +br_fsscan_reschedule(xlator_t *); +int32_t +br_fsscan_activate(xlator_t *); +int32_t +br_fsscan_deactivate(xlator_t *); +int32_t +br_fsscan_ondemand(xlator_t *); + +int32_t +br_scrubber_handle_options(xlator_t *, br_private_t *, dict_t *); + +int32_t +br_scrubber_monitor_init(xlator_t *, br_private_t *); + +int32_t +br_scrubber_init(xlator_t *, br_private_t *); + +int32_t +br_collect_bad_objects_from_children(xlator_t *this, dict_t *dict); + +void +br_child_set_scrub_state(br_child_t *, gf_boolean_t); + +#endif /* __BIT_ROT_SCRUB_H__ */ diff --git a/xlators/features/bit-rot/src/bitd/bit-rot-ssm.c b/xlators/features/bit-rot/src/bitd/bit-rot-ssm.c new file mode 100644 index 00000000000..753e31a3b23 --- /dev/null +++ b/xlators/features/bit-rot/src/bitd/bit-rot-ssm.c @@ -0,0 +1,124 @@ +/* + Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com> + 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 "bit-rot-ssm.h" +#include "bit-rot-scrub.h" +#include "bit-rot-bitd-messages.h" + +int +br_scrub_ssm_noop(xlator_t *this) +{ + return 0; +} + +int +br_scrub_ssm_state_pause(xlator_t *this) +{ + br_private_t *priv = NULL; + struct br_monitor *scrub_monitor = NULL; + + priv = this->private; + scrub_monitor = &priv->scrub_monitor; + + gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_GENERIC_SSM_INFO, + "Scrubber paused"); + _br_monitor_set_scrub_state(scrub_monitor, BR_SCRUB_STATE_PAUSED); + return 0; +} + +int +br_scrub_ssm_state_ipause(xlator_t *this) +{ + br_private_t *priv = NULL; + struct br_monitor *scrub_monitor = NULL; + + priv = this->private; + scrub_monitor = &priv->scrub_monitor; + + gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_GENERIC_SSM_INFO, + "Scrubber paused"); + _br_monitor_set_scrub_state(scrub_monitor, BR_SCRUB_STATE_IPAUSED); + return 0; +} + +int +br_scrub_ssm_state_active(xlator_t *this) +{ + br_private_t *priv = NULL; + struct br_monitor *scrub_monitor = NULL; + + priv = this->private; + scrub_monitor = &priv->scrub_monitor; + + if (scrub_monitor->done) { + (void)br_fsscan_activate(this); + } else { + gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_GENERIC_SSM_INFO, + "Scrubbing resumed"); + _br_monitor_set_scrub_state(scrub_monitor, BR_SCRUB_STATE_ACTIVE); + } + + return 0; +} + +int +br_scrub_ssm_state_stall(xlator_t *this) +{ + br_private_t *priv = NULL; + struct br_monitor *scrub_monitor = NULL; + + priv = this->private; + scrub_monitor = &priv->scrub_monitor; + + gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_GENERIC_SSM_INFO, + "Volume is under active scrubbing. Pausing scrub.."); + _br_monitor_set_scrub_state(scrub_monitor, BR_SCRUB_STATE_STALLED); + return 0; +} + +static br_scrub_ssm_call *br_scrub_ssm[BR_SCRUB_MAXSTATES][BR_SCRUB_MAXEVENTS] = + { + /* INACTIVE */ + {br_fsscan_schedule, br_scrub_ssm_state_ipause, br_scrub_ssm_noop}, + /* PENDING */ + {br_fsscan_reschedule, br_fsscan_deactivate, br_fsscan_ondemand}, + /* ACTIVE */ + {br_scrub_ssm_noop, br_scrub_ssm_state_stall, br_scrub_ssm_noop}, + /* PAUSED */ + {br_fsscan_activate, br_scrub_ssm_noop, br_scrub_ssm_noop}, + /* IPAUSED */ + {br_fsscan_schedule, br_scrub_ssm_noop, br_scrub_ssm_noop}, + /* STALLED */ + {br_scrub_ssm_state_active, br_scrub_ssm_noop, br_scrub_ssm_noop}, +}; + +int32_t +br_scrub_state_machine(xlator_t *this, gf_boolean_t scrub_ondemand) +{ + br_private_t *priv = NULL; + br_scrub_ssm_call *call = NULL; + struct br_scrubber *fsscrub = NULL; + br_scrub_state_t currstate = 0; + br_scrub_event_t event = 0; + struct br_monitor *scrub_monitor = NULL; + + priv = this->private; + fsscrub = &priv->fsscrub; + scrub_monitor = &priv->scrub_monitor; + + currstate = scrub_monitor->state; + if (scrub_ondemand) + event = BR_SCRUB_EVENT_ONDEMAND; + else + event = _br_child_get_scrub_event(fsscrub); + + call = br_scrub_ssm[currstate][event]; + return call(this); +} diff --git a/xlators/features/bit-rot/src/bitd/bit-rot-ssm.h b/xlators/features/bit-rot/src/bitd/bit-rot-ssm.h new file mode 100644 index 00000000000..37b45a42eac --- /dev/null +++ b/xlators/features/bit-rot/src/bitd/bit-rot-ssm.h @@ -0,0 +1,38 @@ +/* + Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com> + 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. +*/ + +#ifndef __BIT_ROT_SSM_H__ +#define __BIT_ROT_SSM_H__ + +#include <glusterfs/xlator.h> + +typedef enum br_scrub_state { + BR_SCRUB_STATE_INACTIVE = 0, + BR_SCRUB_STATE_PENDING, + BR_SCRUB_STATE_ACTIVE, + BR_SCRUB_STATE_PAUSED, + BR_SCRUB_STATE_IPAUSED, + BR_SCRUB_STATE_STALLED, + BR_SCRUB_MAXSTATES, +} br_scrub_state_t; + +typedef enum br_scrub_event { + BR_SCRUB_EVENT_SCHEDULE = 0, + BR_SCRUB_EVENT_PAUSE, + BR_SCRUB_EVENT_ONDEMAND, + BR_SCRUB_MAXEVENTS, +} br_scrub_event_t; + +struct br_monitor; + +int32_t +br_scrub_state_machine(xlator_t *, gf_boolean_t); + +#endif /* __BIT_ROT_SSM_H__ */ diff --git a/xlators/features/bit-rot/src/bitd/bit-rot.c b/xlators/features/bit-rot/src/bitd/bit-rot.c new file mode 100644 index 00000000000..a2f1c343a1d --- /dev/null +++ b/xlators/features/bit-rot/src/bitd/bit-rot.c @@ -0,0 +1,2232 @@ +/* + Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com> + 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 <ctype.h> + +#include <glusterfs/logging.h> +#include <glusterfs/compat-errno.h> + +#include "bit-rot.h" +#include "bit-rot-scrub.h" +#include <pthread.h> +#include "bit-rot-bitd-messages.h" + +#define BR_HASH_CALC_READ_SIZE (128 * 1024) + +typedef int32_t(br_child_handler)(xlator_t *, br_child_t *); + +struct br_child_event { + xlator_t *this; + + br_child_t *child; + + br_child_handler *call; + + struct list_head list; +}; + +static int +br_find_child_index(xlator_t *this, xlator_t *child) +{ + br_private_t *priv = NULL; + int i = -1; + int index = -1; + + GF_VALIDATE_OR_GOTO("bit-rot", this, out); + GF_VALIDATE_OR_GOTO(this->name, this->private, out); + GF_VALIDATE_OR_GOTO(this->name, child, out); + + priv = this->private; + + for (i = 0; i < priv->child_count; i++) { + if (child == priv->children[i].xl) { + index = i; + break; + } + } + +out: + return index; +} + +br_child_t * +br_get_child_from_brick_path(xlator_t *this, char *brick_path) +{ + br_private_t *priv = NULL; + br_child_t *child = NULL; + br_child_t *tmp = NULL; + int i = 0; + + GF_VALIDATE_OR_GOTO("bit-rot", this, out); + GF_VALIDATE_OR_GOTO(this->name, this->private, out); + GF_VALIDATE_OR_GOTO(this->name, brick_path, out); + + priv = this->private; + + pthread_mutex_lock(&priv->lock); + { + for (i = 0; i < priv->child_count; i++) { + tmp = &priv->children[i]; + if (!strcmp(tmp->brick_path, brick_path)) { + child = tmp; + break; + } + } + } + pthread_mutex_unlock(&priv->lock); + +out: + return child; +} + +/** + * probably we'll encapsulate brick inside our own structure when + * needed -- later. + */ +void * +br_brick_init(void *xl, struct gf_brick_spec *brick) +{ + return brick; +} + +/** + * and cleanup things here when allocated br_brick_init(). + */ +void +br_brick_fini(void *xl, char *brick, void *data) +{ + return; +} + +/** + * TODO: Signature can contain null terminators which causes bitrot + * stub to store truncated hash as it depends on string length of + * the hash. + * + * FIX: Send the string length as part of the signature struct and + * change stub to handle this change. + */ +static br_isignature_t * +br_prepare_signature(const unsigned char *sign, unsigned long hashlen, + int8_t hashtype, br_object_t *object) +{ + br_isignature_t *signature = NULL; + + /* TODO: use mem-pool */ + signature = GF_CALLOC(1, signature_size(hashlen + 1), + gf_br_stub_mt_signature_t); + if (!signature) + return NULL; + + /* object version */ + signature->signedversion = object->signedversion; + + /* signature length & type */ + signature->signaturelen = hashlen; + signature->signaturetype = hashtype; + + /* signature itself */ + memcpy(signature->signature, (char *)sign, hashlen); + signature->signature[hashlen + 1] = '\0'; + + return signature; +} + +gf_boolean_t +bitd_is_bad_file(xlator_t *this, br_child_t *child, loc_t *loc, fd_t *fd) +{ + int32_t ret = -1; + dict_t *xattr = NULL; + inode_t *inode = NULL; + gf_boolean_t bad_file = _gf_false; + + GF_VALIDATE_OR_GOTO("bit-rot", this, out); + + inode = (loc) ? loc->inode : fd->inode; + + if (fd) + ret = syncop_fgetxattr(child->xl, fd, &xattr, BITROT_OBJECT_BAD_KEY, + NULL, NULL); + else if (loc) + ret = syncop_getxattr(child->xl, loc, &xattr, BITROT_OBJECT_BAD_KEY, + NULL, NULL); + + if (!ret) { + gf_msg_debug(this->name, 0, "[GFID: %s] is marked corrupted", + uuid_utoa(inode->gfid)); + bad_file = _gf_true; + } + + if (xattr) + dict_unref(xattr); + +out: + return bad_file; +} + +/** + * Do a lookup on the gfid present within the object. + */ +static int32_t +br_object_lookup(xlator_t *this, br_object_t *object, struct iatt *iatt, + inode_t **linked_inode) +{ + int ret = -EINVAL; + loc_t loc = { + 0, + }; + inode_t *inode = NULL; + + GF_VALIDATE_OR_GOTO("bit-rot", this, out); + GF_VALIDATE_OR_GOTO(this->name, object, out); + + inode = inode_find(object->child->table, object->gfid); + + if (inode) + loc.inode = inode; + else + loc.inode = inode_new(object->child->table); + + if (!loc.inode) { + ret = -ENOMEM; + goto out; + } + + gf_uuid_copy(loc.gfid, object->gfid); + + ret = syncop_lookup(object->child->xl, &loc, iatt, NULL, NULL, NULL); + if (ret < 0) + goto out; + + /* + * The file might have been deleted by the application + * after getting the event, but before doing a lookup. + * So use linked_inode after inode_link is done. + */ + *linked_inode = inode_link(loc.inode, NULL, NULL, iatt); + if (*linked_inode) + inode_lookup(*linked_inode); + +out: + loc_wipe(&loc); + return ret; +} + +/** + * open the object with O_RDONLY flags and return the fd. How to let brick + * know that open is being done by bitd because syncop framework does not allow + * passing xdata -- may be use frame->root->pid itself. + */ +static int32_t +br_object_open(xlator_t *this, br_object_t *object, inode_t *inode, + fd_t **openfd) +{ + int32_t ret = -1; + fd_t *fd = NULL; + loc_t loc = { + 0, + }; + + GF_VALIDATE_OR_GOTO("bit-rot", this, out); + GF_VALIDATE_OR_GOTO(this->name, object, out); + GF_VALIDATE_OR_GOTO(this->name, inode, out); + + ret = -EINVAL; + fd = fd_create(inode, 0); + if (!fd) { + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_FD_CREATE_FAILED, + "gfid=%s", uuid_utoa(inode->gfid), NULL); + goto out; + } + + loc.inode = inode_ref(inode); + gf_uuid_copy(loc.gfid, inode->gfid); + + ret = syncop_open(object->child->xl, &loc, O_RDONLY, fd, NULL, NULL); + if (ret) { + br_log_object(this, "open", inode->gfid, -ret); + fd_unref(fd); + fd = NULL; + } else { + fd_bind(fd); + *openfd = fd; + } + + loc_wipe(&loc); + +out: + return ret; +} + +/** + * read 128k block from the object @object from the offset @offset + * and return the buffer. + */ +static int32_t +br_object_read_block_and_sign(xlator_t *this, fd_t *fd, br_child_t *child, + off_t offset, size_t size, SHA256_CTX *sha256) +{ + int32_t ret = -1; + tbf_t *tbf = NULL; + struct iovec *iovec = NULL; + struct iobref *iobref = NULL; + br_private_t *priv = NULL; + int count = 0; + int i = 0; + + GF_VALIDATE_OR_GOTO("bit-rot", this, out); + GF_VALIDATE_OR_GOTO(this->name, fd, out); + GF_VALIDATE_OR_GOTO(this->name, fd->inode, out); + GF_VALIDATE_OR_GOTO(this->name, child, out); + GF_VALIDATE_OR_GOTO(this->name, this->private, out); + + priv = this->private; + + GF_VALIDATE_OR_GOTO(this->name, priv->tbf, out); + tbf = priv->tbf; + + ret = syncop_readv(child->xl, fd, size, offset, 0, &iovec, &count, &iobref, + NULL, NULL, NULL); + + if (ret < 0) { + gf_smsg(this->name, GF_LOG_ERROR, errno, BRB_MSG_READV_FAILED, + "gfid=%s", uuid_utoa(fd->inode->gfid), NULL); + ret = -1; + goto out; + } + + if (ret == 0) + goto out; + + for (i = 0; i < count; i++) { + TBF_THROTTLE_BEGIN(tbf, TBF_OP_HASH, iovec[i].iov_len); + { + SHA256_Update(sha256, (const unsigned char *)(iovec[i].iov_base), + iovec[i].iov_len); + } + TBF_THROTTLE_BEGIN(tbf, TBF_OP_HASH, iovec[i].iov_len); + } + +out: + if (iovec) + GF_FREE(iovec); + + if (iobref) + iobref_unref(iobref); + + return ret; +} + +int32_t +br_calculate_obj_checksum(unsigned char *md, br_child_t *child, fd_t *fd, + struct iatt *iatt) +{ + int32_t ret = -1; + off_t offset = 0; + size_t block = BR_HASH_CALC_READ_SIZE; + xlator_t *this = NULL; + + SHA256_CTX sha256; + + GF_VALIDATE_OR_GOTO("bit-rot", child, out); + GF_VALIDATE_OR_GOTO("bit-rot", iatt, out); + GF_VALIDATE_OR_GOTO("bit-rot", fd, out); + + this = child->this; + + SHA256_Init(&sha256); + + while (1) { + ret = br_object_read_block_and_sign(this, fd, child, offset, block, + &sha256); + if (ret < 0) { + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_BLOCK_READ_FAILED, + "offset=%" PRIu64, offset, "object-gfid=%s", + uuid_utoa(fd->inode->gfid), NULL); + break; + } + + if (ret == 0) + break; + + offset += ret; + } + + if (ret == 0) + SHA256_Final(md, &sha256); + +out: + return ret; +} + +static int32_t +br_object_checksum(unsigned char *md, br_object_t *object, fd_t *fd, + struct iatt *iatt) +{ + return br_calculate_obj_checksum(md, object->child, fd, iatt); +} + +static int32_t +br_object_read_sign(inode_t *linked_inode, fd_t *fd, br_object_t *object, + struct iatt *iatt) +{ + int32_t ret = -1; + xlator_t *this = NULL; + dict_t *xattr = NULL; + unsigned char *md = NULL; + br_isignature_t *sign = NULL; + + GF_VALIDATE_OR_GOTO("bit-rot", object, out); + GF_VALIDATE_OR_GOTO("bit-rot", linked_inode, out); + GF_VALIDATE_OR_GOTO("bit-rot", fd, out); + + this = object->this; + + md = GF_MALLOC(SHA256_DIGEST_LENGTH, gf_common_mt_char); + if (!md) { + gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_SAVING_HASH_FAILED, + "object-gfid=%s", uuid_utoa(fd->inode->gfid), NULL); + goto out; + } + + ret = br_object_checksum(md, object, fd, iatt); + if (ret) { + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_CALC_CHECKSUM_FAILED, + "object-gfid=%s", uuid_utoa(linked_inode->gfid), NULL); + goto free_signature; + } + + sign = br_prepare_signature(md, SHA256_DIGEST_LENGTH, + BR_SIGNATURE_TYPE_SHA256, object); + if (!sign) { + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_GET_SIGN_FAILED, + "object-gfid=%s", uuid_utoa(fd->inode->gfid), NULL); + goto free_signature; + } + + xattr = dict_for_key_value(GLUSTERFS_SET_OBJECT_SIGNATURE, (void *)sign, + signature_size(SHA256_DIGEST_LENGTH), _gf_true); + + if (!xattr) { + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_SET_SIGN_FAILED, + "dict-allocation object-gfid=%s", uuid_utoa(fd->inode->gfid), + NULL); + goto free_isign; + } + + ret = syncop_fsetxattr(object->child->xl, fd, xattr, 0, NULL, NULL); + if (ret) { + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_SET_SIGN_FAILED, + "fsetxattr object-gfid=%s", uuid_utoa(fd->inode->gfid), NULL); + goto unref_dict; + } + + ret = 0; + +unref_dict: + dict_unref(xattr); +free_isign: + GF_FREE(sign); +free_signature: + GF_FREE(md); +out: + return ret; +} + +static int +br_object_sign_softerror(int32_t op_errno) +{ + return ((op_errno == ENOENT) || (op_errno == ESTALE) || + (op_errno == ENODATA)); +} + +void +br_log_object(xlator_t *this, char *op, uuid_t gfid, int32_t op_errno) +{ + int softerror = br_object_sign_softerror(op_errno); + if (softerror) { + gf_msg_debug(this->name, 0, + "%s() failed on object %s " + "[reason: %s]", + op, uuid_utoa(gfid), strerror(op_errno)); + } else { + gf_smsg(this->name, GF_LOG_ERROR, op_errno, BRB_MSG_OP_FAILED, "op=%s", + op, "gfid=%s", uuid_utoa(gfid), NULL); + } +} + +void +br_log_object_path(xlator_t *this, char *op, const char *path, int32_t op_errno) +{ + int softerror = br_object_sign_softerror(op_errno); + if (softerror) { + gf_msg_debug(this->name, 0, + "%s() failed on object %s " + "[reason: %s]", + op, path, strerror(op_errno)); + } else { + gf_smsg(this->name, GF_LOG_ERROR, op_errno, BRB_MSG_OP_FAILED, "op=%s", + op, "path=%s", path, NULL); + } +} + +static void +br_trigger_sign(xlator_t *this, br_child_t *child, inode_t *linked_inode, + loc_t *loc, gf_boolean_t need_reopen) +{ + fd_t *fd = NULL; + int32_t ret = -1; + uint32_t val = 0; + dict_t *dict = NULL; + pid_t pid = GF_CLIENT_PID_BITD; + + syncopctx_setfspid(&pid); + + val = (need_reopen == _gf_true) ? BR_OBJECT_REOPEN : BR_OBJECT_RESIGN; + + dict = dict_new(); + if (!dict) + goto out; + + ret = dict_set_uint32(dict, BR_REOPEN_SIGN_HINT_KEY, val); + if (ret) + goto cleanup_dict; + + ret = -1; + fd = fd_create(linked_inode, 0); + if (!fd) { + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_FD_CREATE_FAILED, + "gfid=%s", uuid_utoa(linked_inode->gfid), NULL); + goto cleanup_dict; + } + + ret = syncop_open(child->xl, loc, O_RDWR, fd, NULL, NULL); + if (ret) { + br_log_object(this, "open", linked_inode->gfid, -ret); + goto unref_fd; + } + + fd_bind(fd); + + ret = syncop_fsetxattr(child->xl, fd, dict, 0, NULL, NULL); + if (ret) + br_log_object(this, "fsetxattr", linked_inode->gfid, -ret); + + /* passthough: fd_unref() */ + +unref_fd: + fd_unref(fd); +cleanup_dict: + dict_unref(dict); +out: + if (ret) { + gf_smsg(this->name, GF_LOG_WARNING, 0, BRB_MSG_TRIGGER_SIGN_FAILED, + "gfid=%s", uuid_utoa(linked_inode->gfid), "reopen-hint-val=%d", + val, NULL); + } +} + +static void +br_object_resign(xlator_t *this, br_object_t *object, inode_t *linked_inode) +{ + loc_t loc = { + 0, + }; + + loc.inode = inode_ref(linked_inode); + gf_uuid_copy(loc.gfid, linked_inode->gfid); + + br_trigger_sign(this, object->child, linked_inode, &loc, _gf_false); + + loc_wipe(&loc); +} + +/** + * Sign a given object. This routine runs full throttle. There needs to be + * some form of priority scheduling and/or read burstness to avoid starving + * (or kicking) client I/O's. + */ +static int32_t +br_sign_object(br_object_t *object) +{ + int32_t ret = -1; + inode_t *linked_inode = NULL; + xlator_t *this = NULL; + fd_t *fd = NULL; + struct iatt iatt = { + 0, + }; + pid_t pid = GF_CLIENT_PID_BITD; + br_sign_state_t sign_info = BR_SIGN_NORMAL; + + GF_VALIDATE_OR_GOTO("bit-rot", object, out); + + this = object->this; + + /** + * FIXME: This is required as signing an object is restricted to + * clients with special frame->root->pid. Change the way client + * pid is set. + */ + syncopctx_setfspid(&pid); + + ret = br_object_lookup(this, object, &iatt, &linked_inode); + if (ret) { + br_log_object(this, "lookup", object->gfid, -ret); + goto out; + } + + /** + * For fd's that have notified for reopening, we send an explicit + * open() followed by a dummy write() call. This triggers the + * actual signing of the object. + */ + sign_info = ntohl(object->sign_info); + if (sign_info == BR_SIGN_REOPEN_WAIT) { + br_object_resign(this, object, linked_inode); + goto unref_inode; + } + + ret = br_object_open(this, object, linked_inode, &fd); + if (!fd) { + br_log_object(this, "open", object->gfid, -ret); + goto unref_inode; + } + + /** + * we have an open file descriptor on the object. from here on, + * do not be generous to file operation errors. + */ + gf_msg_debug(this->name, 0, "Signing object [%s]", + uuid_utoa(linked_inode->gfid)); + + ret = br_object_read_sign(linked_inode, fd, object, &iatt); + if (ret) { + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_READ_AND_SIGN_FAILED, + "gfid=%s", uuid_utoa(linked_inode->gfid), NULL); + goto unref_fd; + } + + ret = 0; + +unref_fd: + fd_unref(fd); +unref_inode: + inode_unref(linked_inode); +out: + return ret; +} + +static br_object_t * +__br_pick_object(br_private_t *priv) +{ + br_object_t *object = NULL; + + while (list_empty(&priv->obj_queue->objects)) { + pthread_cond_wait(&priv->object_cond, &priv->lock); + } + + object = list_first_entry(&priv->obj_queue->objects, br_object_t, list); + list_del_init(&object->list); + + return object; +} + +/** + * This is the place where the signing of the objects is triggered. + */ +void * +br_process_object(void *arg) +{ + xlator_t *this = NULL; + br_object_t *object = NULL; + br_private_t *priv = NULL; + int32_t ret = -1; + + this = arg; + priv = this->private; + + THIS = this; + + for (;;) { + pthread_mutex_lock(&priv->lock); + { + object = __br_pick_object(priv); + } + pthread_mutex_unlock(&priv->lock); + + ret = br_sign_object(object); + if (ret && !br_object_sign_softerror(-ret)) + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_SET_SIGN_FAILED, + "gfid=%s", uuid_utoa(object->gfid), NULL); + GF_FREE(object); + } + + return NULL; +} + +/** + * This function gets kicked in once the object is expired from the + * timer wheel. This actually adds the object received via notification + * from the changelog to the queue from where the objects gets picked + * up for signing. + * + * This routine can be made lightweight by introducing an alternate + * timer-wheel API that dispatches _all_ expired objects in one-shot + * rather than an object at-a-time. This routine can then just simply + * be a call to list_splice_tail(). + * + * NOTE: use call_time to instrument signing time in br_sign_object(). + */ +void +br_add_object_to_queue(struct gf_tw_timer_list *timer, void *data, + unsigned long call_time) +{ + br_object_t *object = NULL; + xlator_t *this = NULL; + br_private_t *priv = NULL; + + object = data; + this = object->this; + priv = this->private; + + THIS = this; + + pthread_mutex_lock(&priv->lock); + { + list_add_tail(&object->list, &priv->obj_queue->objects); + pthread_cond_broadcast(&priv->object_cond); + } + pthread_mutex_unlock(&priv->lock); + + if (timer) + mem_put(timer); + return; +} + +static br_object_t * +br_initialize_object(xlator_t *this, br_child_t *child, changelog_event_t *ev) +{ + br_object_t *object = NULL; + + object = GF_CALLOC(1, sizeof(*object), gf_br_mt_br_object_t); + if (!object) + goto out; + INIT_LIST_HEAD(&object->list); + + object->this = this; + object->child = child; + gf_uuid_copy(object->gfid, ev->u.releasebr.gfid); + + /* NOTE: it's BE, but no worry */ + object->signedversion = ev->u.releasebr.version; + object->sign_info = ev->u.releasebr.sign_info; + +out: + return object; +} + +static struct gf_tw_timer_list * +br_initialize_timer(xlator_t *this, br_object_t *object, br_child_t *child, + changelog_event_t *ev) +{ + br_private_t *priv = NULL; + struct gf_tw_timer_list *timer = NULL; + + priv = this->private; + + timer = mem_get0(child->timer_pool); + if (!timer) + goto out; + INIT_LIST_HEAD(&timer->entry); + + timer->expires = priv->expiry_time; + if (!timer->expires) + timer->expires = 1; + + timer->data = object; + timer->function = br_add_object_to_queue; + gf_tw_add_timer(priv->timer_wheel, timer); + +out: + return timer; +} + +static int32_t +br_schedule_object_reopen(xlator_t *this, br_object_t *object, + br_child_t *child, changelog_event_t *ev) +{ + struct gf_tw_timer_list *timer = NULL; + + timer = br_initialize_timer(this, object, child, ev); + if (!timer) + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_SET_TIMER_FAILED, + "gfid=%s", uuid_utoa(object->gfid), NULL); + return timer ? 0 : -1; +} + +static int32_t +br_object_quicksign(xlator_t *this, br_object_t *object) +{ + br_add_object_to_queue(NULL, object, 0ULL); + return 0; +} + +/** + * This callback function registered with the changelog is executed + * whenever a notification from the changelog is received. This should + * add the object (or the gfid) on which the notification has come to + * the timer-wheel with some expiry time. + * + * TODO: use mem-pool for allocations and maybe allocate timer and + * object as a single alloc and bifurcate their respective pointers. + */ +void +br_brick_callback(void *xl, char *brick, void *data, changelog_event_t *ev) +{ + int32_t ret = 0; + uuid_t gfid = { + 0, + }; + xlator_t *this = NULL; + br_object_t *object = NULL; + br_child_t *child = NULL; + br_sign_state_t sign_info = BR_SIGN_INVALID; + + this = xl; + + GF_VALIDATE_OR_GOTO(this->name, ev, out); + GF_VALIDATE_OR_GOTO("bit-rot", this, out); + GF_VALIDATE_OR_GOTO(this->name, this->private, out); + + GF_ASSERT(ev->ev_type == CHANGELOG_OP_TYPE_BR_RELEASE); + GF_ASSERT(!gf_uuid_is_null(ev->u.releasebr.gfid)); + + gf_uuid_copy(gfid, ev->u.releasebr.gfid); + + gf_msg_debug(this->name, 0, "RELEASE EVENT [GFID %s]", uuid_utoa(gfid)); + + child = br_get_child_from_brick_path(this, brick); + if (!child) { + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_GET_SUBVOL_FAILED, + "brick=%s", brick, NULL); + goto out; + } + + object = br_initialize_object(this, child, ev); + if (!object) { + gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_NO_MEMORY, + "object-gfid=%s", uuid_utoa(gfid), NULL); + goto out; + } + + /* sanity check */ + sign_info = ntohl(object->sign_info); + GF_ASSERT(sign_info != BR_SIGN_NORMAL); + + if (sign_info == BR_SIGN_REOPEN_WAIT) + ret = br_schedule_object_reopen(this, object, child, ev); + else + ret = br_object_quicksign(this, object); + + if (ret) + goto free_object; + + gf_msg_debug(this->name, 0, "->callback: brick [%s], type [%d]\n", brick, + ev->ev_type); + return; + +free_object: + GF_FREE(object); +out: + return; +} + +void +br_fill_brick_spec(struct gf_brick_spec *brick, char *path) +{ + brick->brick_path = gf_strdup(path); + brick->filter = CHANGELOG_OP_TYPE_BR_RELEASE; + + brick->init = br_brick_init; + brick->fini = br_brick_fini; + brick->callback = br_brick_callback; + brick->connected = NULL; + brick->disconnected = NULL; +} + +static gf_boolean_t +br_check_object_need_sign(xlator_t *this, dict_t *xattr, br_child_t *child) +{ + int32_t ret = -1; + gf_boolean_t need_sign = _gf_false; + br_isignature_out_t *sign = NULL; + + GF_VALIDATE_OR_GOTO("bit-rot", this, out); + GF_VALIDATE_OR_GOTO(this->name, xattr, out); + GF_VALIDATE_OR_GOTO(this->name, child, out); + + ret = dict_get_ptr(xattr, GLUSTERFS_GET_OBJECT_SIGNATURE, (void **)&sign); + if (ret) { + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_GET_SIGN_FAILED, + "object-info", NULL); + goto out; + } + + /* Object has been opened and hence dirty. Do not sign it */ + if (sign->stale) + need_sign = _gf_true; + +out: + return need_sign; +} + +int32_t +br_prepare_loc(xlator_t *this, br_child_t *child, loc_t *parent, + gf_dirent_t *entry, loc_t *loc) +{ + int32_t ret = -1; + inode_t *inode = NULL; + + inode = inode_grep(child->table, parent->inode, entry->d_name); + if (!inode) + loc->inode = inode_new(child->table); + else { + loc->inode = inode; + if (loc->inode->ia_type != IA_IFREG) { + gf_msg_debug(this->name, 0, + "%s is not a regular " + "file", + entry->d_name); + ret = 0; + goto out; + } + } + + loc->parent = inode_ref(parent->inode); + gf_uuid_copy(loc->pargfid, parent->inode->gfid); + + ret = inode_path(parent->inode, entry->d_name, (char **)&loc->path); + if (ret < 0 || !loc->path) { + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_PATH_FAILED, + "inode_path=%s", entry->d_name, "parent-gfid=%s", + uuid_utoa(parent->inode->gfid), NULL); + goto out; + } + + loc->name = strrchr(loc->path, '/'); + if (loc->name) + loc->name++; + + ret = 1; + +out: + return ret; +} + +/** + * Oneshot crawler + * --------------- + * This is a catchup mechanism. Objects that remained unsigned from the + * last run for whatever reason (node crashes, reboots, etc..) become + * candidates for signing. This allows the signature to "catch up" with + * the current state of the object. Triggering signing is easy: perform + * an open() followed by a close() thereby resulting in call boomerang. + * (though not back to itself :)) + */ +int +bitd_oneshot_crawl(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent, + void *data) +{ + int op_errno = 0; + br_child_t *child = NULL; + xlator_t *this = NULL; + loc_t loc = { + 0, + }; + struct iatt iatt = { + 0, + }; + struct iatt parent_buf = { + 0, + }; + dict_t *xattr = NULL; + int32_t ret = -1; + inode_t *linked_inode = NULL; + gf_boolean_t need_signing = _gf_false; + gf_boolean_t need_reopen = _gf_true; + + GF_VALIDATE_OR_GOTO("bit-rot", subvol, out); + GF_VALIDATE_OR_GOTO("bit-rot", data, out); + + child = data; + this = child->this; + + ret = br_prepare_loc(this, child, parent, entry, &loc); + if (!ret) + goto out; + + ret = syncop_lookup(child->xl, &loc, &iatt, &parent_buf, NULL, NULL); + if (ret) { + br_log_object_path(this, "lookup", loc.path, -ret); + goto out; + } + + linked_inode = inode_link(loc.inode, parent->inode, loc.name, &iatt); + if (linked_inode) + inode_lookup(linked_inode); + + if (iatt.ia_type != IA_IFREG) { + gf_msg_debug(this->name, 0, + "%s is not a regular file, " + "skipping..", + entry->d_name); + ret = 0; + goto unref_inode; + } + + /** + * As of now, 2 cases are possible and handled. + * 1) GlusterFS is upgraded from a previous version which does not + * have any idea about bit-rot and have data in the filesystem. + * In this case syncop_getxattr fails with ENODATA and the object + * is signed. (In real, when crawler sends lookup, bit-rot-stub + * creates the xattrs before returning lookup reply) + * 2) Bit-rot was not enabled or BitD was does for some reasons, during + * which some files were created, but since BitD was down, were not + * signed. + * If the file was just created and was being written some data when + * the down BitD came up, then bit-rot stub should be intelligent to + * identify this case (by comparing the ongoing version or by checking + * if there are any fds present for that inode) and handle properly. + */ + + if (bitd_is_bad_file(this, child, &loc, NULL)) { + gf_smsg(this->name, GF_LOG_WARNING, 0, BRB_MSG_SKIP_OBJECT, "path=%s", + loc.path, NULL); + goto unref_inode; + } + + ret = syncop_getxattr(child->xl, &loc, &xattr, + GLUSTERFS_GET_OBJECT_SIGNATURE, NULL, NULL); + if (ret < 0) { + op_errno = -ret; + br_log_object(this, "getxattr", linked_inode->gfid, op_errno); + + /** + * No need to sign the zero byte objects as the signing + * happens upon first modification of the object. + */ + if (op_errno == ENODATA && (iatt.ia_size != 0)) + need_signing = _gf_true; + if (op_errno == EINVAL) + gf_smsg(this->name, GF_LOG_WARNING, 0, + BRB_MSG_PARTIAL_VERSION_PRESENCE, "gfid=%s", + uuid_utoa(linked_inode->gfid), NULL); + } else { + need_signing = br_check_object_need_sign(this, xattr, child); + + /* + * If we are here means, bitrot daemon has started. Is it just + * a simple restart of the daemon or is it started because the + * feature is enabled is something hard to determine. Hence, + * if need_signing is false (because bit-rot version and signature + * are present), then still go ahead and sign it. + */ + if (!need_signing) { + need_signing = _gf_true; + need_reopen = _gf_true; + } + } + + if (!need_signing) + goto unref_dict; + + gf_smsg(this->name, GF_LOG_INFO, 0, BRB_MSG_TRIGGER_SIGN, "path=%s", + loc.path, "gfid=%s", uuid_utoa(linked_inode->gfid), "Brick-path=%s", + child->brick_path, NULL); + br_trigger_sign(this, child, linked_inode, &loc, need_reopen); + + ret = 0; + +unref_dict: + if (xattr) + dict_unref(xattr); +unref_inode: + inode_unref(linked_inode); +out: + loc_wipe(&loc); + + return ret; +} + +#define BR_CRAWL_THROTTLE_COUNT 50 +#define BR_CRAWL_THROTTLE_ZZZ 5 + +void * +br_oneshot_signer(void *arg) +{ + loc_t loc = { + 0, + }; + xlator_t *this = NULL; + br_child_t *child = NULL; + + child = arg; + this = child->this; + + THIS = this; + + gf_smsg(this->name, GF_LOG_INFO, 0, BRB_MSG_CRAWLING_START, "brick-path=%s", + child->brick_path, NULL); + + loc.inode = child->table->root; + (void)syncop_ftw_throttle(child->xl, &loc, GF_CLIENT_PID_BITD, child, + bitd_oneshot_crawl, BR_CRAWL_THROTTLE_COUNT, + BR_CRAWL_THROTTLE_ZZZ); + + gf_smsg(this->name, GF_LOG_INFO, 0, BRB_MSG_CRAWLING_FINISH, + "brick-path=%s", child->brick_path, NULL); + + return NULL; +} + +static void +br_set_child_state(br_child_t *child, br_child_state_t state) +{ + pthread_mutex_lock(&child->lock); + { + _br_set_child_state(child, state); + } + pthread_mutex_unlock(&child->lock); +} + +/** + * At this point a thread is spawned to crawl the filesystem (in + * tortoise pace) to sign objects that were not signed in previous run(s). + * Such objects are identified by examining it's dirtyness and timestamp. + * + * pick object: + * signature_is_stale() && (object_timestamp() <= stub_init_time()) + * + * Also, we register to the changelog library to subscribe for event + * notifications. + */ +static int32_t +br_enact_signer(xlator_t *this, br_child_t *child, br_stub_init_t *stub) +{ + int32_t ret = 0; + br_private_t *priv = NULL; + struct gf_brick_spec *brick = NULL; + + priv = this->private; + + brick = GF_CALLOC(1, sizeof(struct gf_brick_spec), + gf_common_mt_gf_brick_spec_t); + if (!brick) + goto error_return; + + br_fill_brick_spec(brick, stub->export); + ret = gf_changelog_register_generic(brick, 1, 1, + this->ctx->cmd_args.log_file, -1, this); + if (ret) { + gf_smsg(this->name, GF_LOG_ERROR, errno, BRB_MSG_REGISTER_FAILED, NULL); + goto dealloc; + } + + child->threadrunning = 0; + ret = gf_thread_create(&child->thread, NULL, br_oneshot_signer, child, + "brosign"); + if (ret) + gf_smsg(this->name, GF_LOG_WARNING, 0, BRB_MSG_SPAWN_FAILED, + "FS-crawler-thread", NULL); + else + child->threadrunning = 1; + + /* it's OK to continue, "old" objects would be signed when modified */ + list_add_tail(&child->list, &priv->signing); + return 0; + +dealloc: + GF_FREE(brick); +error_return: + return -1; +} + +static int32_t +br_launch_scrubber(xlator_t *this, br_child_t *child, struct br_scanfs *fsscan, + struct br_scrubber *fsscrub) +{ + int32_t ret = -1; + br_private_t *priv = NULL; + struct br_monitor *scrub_monitor = NULL; + + priv = this->private; + + scrub_monitor = &priv->scrub_monitor; + ret = gf_thread_create(&child->thread, NULL, br_fsscanner, child, + "brfsscan"); + if (ret != 0) { + gf_smsg(this->name, GF_LOG_ALERT, 0, BRB_MSG_SPAWN_FAILED, + "bitrot-scrubber-daemon Brick-path=%s", child->brick_path, + NULL); + goto error_return; + } + + /* Signal monitor to kick off state machine*/ + pthread_mutex_lock(&scrub_monitor->mutex); + { + if (!scrub_monitor->inited) + pthread_cond_signal(&scrub_monitor->cond); + scrub_monitor->inited = _gf_true; + } + pthread_mutex_unlock(&scrub_monitor->mutex); + + /** + * Everything has been setup.. add this subvolume to scrubbers + * list. + */ + pthread_mutex_lock(&fsscrub->mutex); + { + list_add_tail(&child->list, &fsscrub->scrublist); + pthread_cond_broadcast(&fsscrub->cond); + } + pthread_mutex_unlock(&fsscrub->mutex); + + return 0; + +error_return: + return -1; +} + +static int32_t +br_enact_scrubber(xlator_t *this, br_child_t *child) +{ + int32_t ret = 0; + br_private_t *priv = NULL; + struct br_scanfs *fsscan = NULL; + struct br_scrubber *fsscrub = NULL; + + priv = this->private; + + fsscan = &child->fsscan; + fsscrub = &priv->fsscrub; + + /** + * if this child already witnesses a successful connection earlier + * there's no need to initialize mutexes, condvars, etc.. + */ + if (_br_child_witnessed_connection(child)) + return br_launch_scrubber(this, child, fsscan, fsscrub); + + LOCK_INIT(&fsscan->entrylock); + pthread_mutex_init(&fsscan->waitlock, NULL); + pthread_cond_init(&fsscan->waitcond, NULL); + + fsscan->entries = 0; + INIT_LIST_HEAD(&fsscan->queued); + INIT_LIST_HEAD(&fsscan->ready); + + ret = br_launch_scrubber(this, child, fsscan, fsscrub); + if (ret) + goto error_return; + + return 0; + +error_return: + LOCK_DESTROY(&fsscan->entrylock); + pthread_mutex_destroy(&fsscan->waitlock); + pthread_cond_destroy(&fsscan->waitcond); + + return -1; +} + +static int32_t +br_child_enaction(xlator_t *this, br_child_t *child, br_stub_init_t *stub) +{ + int32_t ret = -1; + br_private_t *priv = this->private; + + pthread_mutex_lock(&child->lock); + { + if (priv->iamscrubber) + ret = br_enact_scrubber(this, child); + else + ret = br_enact_signer(this, child, stub); + + if (!ret) { + child->witnessed = 1; + _br_set_child_state(child, BR_CHILD_STATE_CONNECTED); + gf_smsg(this->name, GF_LOG_INFO, 0, BRB_MSG_CONNECTED_TO_BRICK, + "brick-path=%s", child->brick_path, NULL); + } + } + pthread_mutex_unlock(&child->lock); + + return ret; +} + +/** + * This routine fetches various attributes associated with a child which + * is basically a subvolume. Attributes include brick path and the stub + * birth time. This is done by performing a lookup on the root followed + * by getxattr() on a virtual key. Depending on the configuration, the + * process either acts as a signer or a scrubber. + */ +int32_t +br_brick_connect(xlator_t *this, br_child_t *child) +{ + int32_t ret = -1; + loc_t loc = { + 0, + }; + struct iatt buf = { + 0, + }; + struct iatt parent = { + 0, + }; + br_stub_init_t *stub = NULL; + dict_t *xattr = NULL; + int op_errno = 0; + + GF_VALIDATE_OR_GOTO("bit-rot", this, out); + GF_VALIDATE_OR_GOTO(this->name, child, out); + GF_VALIDATE_OR_GOTO(this->name, this->private, out); + + br_child_set_scrub_state(child, _gf_false); + br_set_child_state(child, BR_CHILD_STATE_INITIALIZING); + + loc.inode = inode_ref(child->table->root); + gf_uuid_copy(loc.gfid, loc.inode->gfid); + loc.path = gf_strdup("/"); + + ret = syncop_lookup(child->xl, &loc, &buf, &parent, NULL, NULL); + if (ret) { + op_errno = -ret; + ret = -1; + gf_smsg(this->name, GF_LOG_ERROR, op_errno, BRB_MSG_LOOKUP_FAILED, + NULL); + goto wipeloc; + } + + ret = syncop_getxattr(child->xl, &loc, &xattr, + GLUSTERFS_GET_BR_STUB_INIT_TIME, NULL, NULL); + if (ret) { + op_errno = -ret; + ret = -1; + gf_smsg(this->name, GF_LOG_ERROR, op_errno, BRB_MSG_GET_INFO_FAILED, + NULL); + goto wipeloc; + } + + ret = dict_get_ptr(xattr, GLUSTERFS_GET_BR_STUB_INIT_TIME, (void **)&stub); + if (ret) { + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_GET_INFO_FAILED, NULL); + goto free_dict; + } + + memcpy(child->brick_path, stub->export, strlen(stub->export) + 1); + child->tv.tv_sec = ntohl(stub->timebuf[0]); + child->tv.tv_usec = ntohl(stub->timebuf[1]); + + ret = br_child_enaction(this, child, stub); + +free_dict: + dict_unref(xattr); +wipeloc: + loc_wipe(&loc); +out: + if (ret) + br_set_child_state(child, BR_CHILD_STATE_CONNFAILED); + return ret; +} + +/* TODO: cleanup signer */ +static int32_t +br_cleanup_signer(xlator_t *this, br_child_t *child) +{ + return 0; +} + +static int32_t +br_cleanup_scrubber(xlator_t *this, br_child_t *child) +{ + int32_t ret = 0; + br_private_t *priv = NULL; + struct br_scrubber *fsscrub = NULL; + struct br_monitor *scrub_monitor = NULL; + + priv = this->private; + fsscrub = &priv->fsscrub; + scrub_monitor = &priv->scrub_monitor; + + if (_br_is_child_scrub_active(child)) { + scrub_monitor->active_child_count--; + br_child_set_scrub_state(child, _gf_false); + } + + /** + * 0x0: child (brick) goes out of rotation + * + * This is fully safe w.r.t. entries for this child being actively + * scrubbed. Each of the scrubber thread(s) would finish scrubbing + * the entry (probably failing due to disconnection) and either + * putting the entry back into the queue or continuing further. + * Either way, pending entries for this child's queue need not be + * drained; entries just sit there in the queued/ready list to be + * consumed later upon re-connection. + */ + pthread_mutex_lock(&fsscrub->mutex); + { + list_del_init(&child->list); + } + pthread_mutex_unlock(&fsscrub->mutex); + + /** + * 0x1: cleanup scanner thread + * + * The pending timer needs to be removed _after_ cleaning up the + * filesystem scanner (scheduling the next scrub time is not a + * cancellation point). + */ + ret = gf_thread_cleanup_xint(child->thread); + if (ret) + gf_smsg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_THREAD_CLEANUP, NULL); + + gf_smsg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUBBER_CLEANED, + "brick-path=%s", child->brick_path, NULL); + + return 0; +} + +/** + * OK.. this child has made it's mind to go down the drain. So, + * let's clean up what it touched. (NOTE: there's no need to clean + * the inode table, it's just reused taking care of stale inodes) + */ +int32_t +br_brick_disconnect(xlator_t *this, br_child_t *child) +{ + int32_t ret = 0; + struct br_monitor *scrub_monitor = NULL; + br_private_t *priv = this->private; + + scrub_monitor = &priv->scrub_monitor; + + /* Lock order should be wakelock and then child lock to + * dead locks. + */ + pthread_mutex_lock(&scrub_monitor->wakelock); + { + pthread_mutex_lock(&child->lock); + { + if (!_br_is_child_connected(child)) + goto unblock; + + /* child is on death row.. */ + _br_set_child_state(child, BR_CHILD_STATE_DISCONNECTED); + + if (priv->iamscrubber) + ret = br_cleanup_scrubber(this, child); + else + ret = br_cleanup_signer(this, child); + } + unblock: + pthread_mutex_unlock(&child->lock); + } + pthread_mutex_unlock(&scrub_monitor->wakelock); + + return ret; +} + +/** + * This function is executed in a separate thread. The thread gets the + * brick from where CHILD_UP has received from the queue and gets the + * information regarding that brick (such as brick path). + */ +void * +br_handle_events(void *arg) +{ + int32_t ret = 0; + xlator_t *this = NULL; + br_private_t *priv = NULL; + br_child_t *child = NULL; + struct br_child_event *childev = NULL; + + this = arg; + priv = this->private; + + /* + * Since, this is the topmost xlator, THIS has to be set by bit-rot + * xlator itself (STACK_WIND won't help in this case). Also it has + * to be done for each thread that gets spawned. Otherwise, a new + * thread will get global_xlator's pointer when it does "THIS". + */ + THIS = this; + + while (1) { + pthread_mutex_lock(&priv->lock); + { + while (list_empty(&priv->bricks)) + pthread_cond_wait(&priv->cond, &priv->lock); + + childev = list_first_entry(&priv->bricks, struct br_child_event, + list); + list_del_init(&childev->list); + } + pthread_mutex_unlock(&priv->lock); + + child = childev->child; + ret = childev->call(this, child); + if (ret) + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_SUBVOL_CONNECT_FAILED, + "name=%s", child->xl->name, NULL); + GF_FREE(childev); + } + + return NULL; +} + +int32_t +mem_acct_init(xlator_t *this) +{ + int32_t ret = -1; + + if (!this) + return ret; + + ret = xlator_mem_acct_init(this, gf_br_stub_mt_end + 1); + + if (ret != 0) { + gf_smsg(this->name, GF_LOG_WARNING, 0, BRB_MSG_MEM_ACNT_FAILED, NULL); + return ret; + } + + return ret; +} + +static void +_br_qchild_event(xlator_t *this, br_child_t *child, br_child_handler *call) +{ + br_private_t *priv = NULL; + struct br_child_event *childev = NULL; + + priv = this->private; + + childev = GF_CALLOC(1, sizeof(*childev), gf_br_mt_br_child_event_t); + if (!childev) { + gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_EVENT_UNHANDLED, + "Brick-name=%s", child->xl->name, NULL); + return; + } + + INIT_LIST_HEAD(&childev->list); + childev->this = this; + childev->child = child; + childev->call = call; + + list_add_tail(&childev->list, &priv->bricks); +} + +int +br_scrubber_status_get(xlator_t *this, dict_t **dict) +{ + int ret = -1; + br_private_t *priv = NULL; + struct br_scrub_stats *scrub_stats = NULL; + + priv = this->private; + + GF_VALIDATE_OR_GOTO("bit-rot", priv, out); + + scrub_stats = &priv->scrub_stat; + + ret = br_get_bad_objects_list(this, dict); + if (ret) { + gf_msg_debug(this->name, 0, + "Failed to collect corrupt " + "files"); + } + + ret = dict_set_int8(*dict, "scrub-running", scrub_stats->scrub_running); + if (ret) { + gf_msg_debug(this->name, 0, + "Failed setting scrub_running " + "entry to the dictionary"); + } + + ret = dict_set_uint64(*dict, "scrubbed-files", scrub_stats->scrubbed_files); + if (ret) { + gf_msg_debug(this->name, 0, + "Failed to setting scrubbed file " + "entry to the dictionary"); + } + + ret = dict_set_uint64(*dict, "unsigned-files", scrub_stats->unsigned_files); + if (ret) { + gf_msg_debug(this->name, 0, + "Failed to set unsigned file count" + " entry to the dictionary"); + } + + ret = dict_set_uint64(*dict, "scrub-duration", scrub_stats->scrub_duration); + if (ret) { + gf_msg_debug(this->name, 0, + "Failed to set scrub duration" + " entry to the dictionary"); + } + + ret = dict_set_dynstr_with_alloc(*dict, "last-scrub-time", + scrub_stats->last_scrub_time); + if (ret) { + gf_msg_debug(this->name, 0, + "Failed to set " + "last scrub time value"); + } + +out: + return ret; +} + +int +notify(xlator_t *this, int32_t event, void *data, ...) +{ + int idx = -1; + int ret = -1; + xlator_t *subvol = NULL; + br_child_t *child = NULL; + br_private_t *priv = NULL; + dict_t *output = NULL; + va_list ap; + struct br_monitor *scrub_monitor = NULL; + + subvol = (xlator_t *)data; + priv = this->private; + scrub_monitor = &priv->scrub_monitor; + + gf_msg_trace(this->name, 0, "Notification received: %d", event); + + idx = br_find_child_index(this, subvol); + + switch (event) { + case GF_EVENT_CHILD_UP: + if (idx < 0) { + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_INVALID_SUBVOL, + "event=%d", event, NULL); + goto out; + } + + pthread_mutex_lock(&priv->lock); + { + child = &priv->children[idx]; + if (child->child_up == 1) + goto unblock_0; + priv->up_children++; + + child->child_up = 1; + child->xl = subvol; + if (!child->table) + child->table = inode_table_new(4096, subvol); + + _br_qchild_event(this, child, br_brick_connect); + pthread_cond_signal(&priv->cond); + } + unblock_0: + pthread_mutex_unlock(&priv->lock); + + if (priv->up_children == priv->child_count) + default_notify(this, event, data); + break; + + case GF_EVENT_CHILD_DOWN: + if (idx < 0) { + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_INVALID_SUBVOL, + "event=%d", event, NULL); + goto out; + } + + pthread_mutex_lock(&priv->lock); + { + child = &priv->children[idx]; + if (child->child_up == 0) + goto unblock_1; + + child->child_up = 0; + priv->up_children--; + + _br_qchild_event(this, child, br_brick_disconnect); + pthread_cond_signal(&priv->cond); + } + unblock_1: + pthread_mutex_unlock(&priv->lock); + + if (priv->up_children == 0) + default_notify(this, event, data); + break; + + case GF_EVENT_SCRUB_STATUS: + gf_msg_debug(this->name, GF_LOG_INFO, + "BitRot scrub status " + "called"); + va_start(ap, data); + output = va_arg(ap, dict_t *); + va_end(ap); + + ret = br_scrubber_status_get(this, &output); + gf_msg_debug(this->name, 0, "returning %d", ret); + break; + + case GF_EVENT_SCRUB_ONDEMAND: + gf_log(this->name, GF_LOG_INFO, + "BitRot scrub ondemand " + "called"); + + if (scrub_monitor->state != BR_SCRUB_STATE_PENDING) { + gf_smsg(this->name, GF_LOG_ERROR, 0, + BRB_MSG_RESCHEDULE_SCRUBBER_FAILED, "Current-state=%d", + scrub_monitor->state, NULL); + return -2; + } + + /* Needs synchronization with reconfigure thread */ + pthread_mutex_lock(&priv->lock); + { + ret = br_scrub_state_machine(this, _gf_true); + } + pthread_mutex_unlock(&priv->lock); + + if (ret) { + gf_smsg(this->name, GF_LOG_ERROR, 0, + BRB_MSG_COULD_NOT_SCHEDULE_SCRUB, NULL); + } + gf_msg_debug(this->name, 0, "returning %d", ret); + break; + default: + default_notify(this, event, data); + } + +out: + return 0; +} + +static void +br_fini_signer(xlator_t *this, br_private_t *priv) +{ + int i = 0; + + if (priv == NULL) + return; + + for (; i < priv->signer_th_count; i++) { + (void)gf_thread_cleanup_xint(priv->obj_queue->workers[i]); + } + GF_FREE(priv->obj_queue->workers); + + pthread_cond_destroy(&priv->object_cond); +} + +/** + * Initialize signer specific structures, spawn worker threads. + */ + +static int32_t +br_init_signer(xlator_t *this, br_private_t *priv) +{ + int i = 0; + int32_t ret = -1; + + /* initialize gfchangelog xlator context */ + ret = gf_changelog_init(this); + if (ret) + goto out; + + pthread_cond_init(&priv->object_cond, NULL); + + priv->obj_queue = GF_CALLOC(1, sizeof(*priv->obj_queue), + gf_br_mt_br_ob_n_wk_t); + if (!priv->obj_queue) + goto cleanup_cond; + INIT_LIST_HEAD(&priv->obj_queue->objects); + + priv->obj_queue->workers = GF_CALLOC( + priv->signer_th_count, sizeof(pthread_t), gf_br_mt_br_worker_t); + if (!priv->obj_queue->workers) + goto cleanup_obj_queue; + + for (i = 0; i < priv->signer_th_count; i++) { + ret = gf_thread_create(&priv->obj_queue->workers[i], NULL, + br_process_object, this, "brpobj"); + if (ret != 0) { + gf_smsg(this->name, GF_LOG_ERROR, -ret, + BRB_MSG_THREAD_CREATION_FAILED, NULL); + ret = -1; + goto cleanup_threads; + } + } + + return 0; + +cleanup_threads: + for (i--; i >= 0; i--) { + (void)gf_thread_cleanup_xint(priv->obj_queue->workers[i]); + } + GF_FREE(priv->obj_queue->workers); + +cleanup_obj_queue: + GF_FREE(priv->obj_queue); + +cleanup_cond: + /* that's explicit */ + pthread_cond_destroy(&priv->object_cond); +out: + return -1; +} + +/** + * For signer, only rate limit CPU usage (during hash calculation) when + * compiled with -DBR_RATE_LIMIT_SIGNER cflags, else let it run full + * throttle. + */ +static int32_t +br_rate_limit_signer(xlator_t *this, int child_count, int numbricks) +{ + br_private_t *priv = NULL; + tbf_opspec_t spec = { + 0, + }; + + priv = this->private; + + spec.op = TBF_OP_HASH; + spec.rate = 0; + spec.maxlimit = 0; + + /** + * OK. Most implementations of TBF I've come across generate tokens + * every second (UML, etc..) and some chose sub-second granularity + * (blk-iothrottle cgroups). TBF algorithm itself does not enforce + * any logic for choosing generation interval and it seems pretty + * logical as one could jack up token count per interval w.r.t. + * generation rate. + * + * Value used here is chosen based on a series of test(s) performed + * to balance object signing time and not maxing out on all available + * CPU cores. It's obvious to have seconds granularity and jack up + * token count per interval, thereby achieving close to similar + * results. Let's stick to this as it seems to be working fine for + * the set of ops that are throttled. + **/ + spec.token_gen_interval = 600000; /* In usec */ + +#ifdef BR_RATE_LIMIT_SIGNER + + double contribution = 0; + contribution = ((double)1 - ((double)child_count / (double)numbricks)); + if (contribution == 0) + contribution = 1; + spec.rate = BR_HASH_CALC_READ_SIZE * contribution; + spec.maxlimit = priv->signer_th_count * BR_HASH_CALC_READ_SIZE; + +#endif + + if (!spec.rate) + gf_smsg(this->name, GF_LOG_INFO, 0, BRB_MSG_RATE_LIMIT_INFO, + "FULL THROTTLE", NULL); + else + gf_smsg(this->name, GF_LOG_INFO, 0, BRB_MSG_RATE_LIMIT_INFO, + "tokens/sec-rate=%lu", spec.rate, "maxlimit=%lu", spec.maxlimit, + NULL); + + priv->tbf = tbf_init(&spec, 1); + return priv->tbf ? 0 : -1; +} + +static int32_t +br_signer_handle_options(xlator_t *this, br_private_t *priv, dict_t *options) +{ + if (options) { + GF_OPTION_RECONF("expiry-time", priv->expiry_time, options, uint32, + error_return); + GF_OPTION_RECONF("signer-threads", priv->signer_th_count, options, + uint32, error_return); + } else { + GF_OPTION_INIT("expiry-time", priv->expiry_time, uint32, error_return); + GF_OPTION_INIT("signer-threads", priv->signer_th_count, uint32, + error_return); + } + + return 0; + +error_return: + return -1; +} + +static int32_t +br_signer_init(xlator_t *this, br_private_t *priv) +{ + int32_t ret = 0; + int numbricks = 0; + + GF_OPTION_INIT("expiry-time", priv->expiry_time, uint32, error_return); + GF_OPTION_INIT("brick-count", numbricks, int32, error_return); + GF_OPTION_INIT("signer-threads", priv->signer_th_count, uint32, + error_return); + + ret = br_rate_limit_signer(this, priv->child_count, numbricks); + if (ret) + goto error_return; + + ret = br_init_signer(this, priv); + if (ret) + goto cleanup_tbf; + + return 0; + +cleanup_tbf: + /* cleanup TBF */ +error_return: + return -1; +} + +static void +br_free_scrubber_monitor(xlator_t *this, br_private_t *priv) +{ + struct br_monitor *scrub_monitor = &priv->scrub_monitor; + + if (scrub_monitor->timer) { + (void)gf_tw_del_timer(priv->timer_wheel, scrub_monitor->timer); + + GF_FREE(scrub_monitor->timer); + scrub_monitor->timer = NULL; + } + + (void)gf_thread_cleanup_xint(scrub_monitor->thread); + + /* Clean up cond and mutex variables */ + pthread_mutex_destroy(&scrub_monitor->mutex); + pthread_cond_destroy(&scrub_monitor->cond); + + pthread_mutex_destroy(&scrub_monitor->wakelock); + pthread_cond_destroy(&scrub_monitor->wakecond); + + pthread_mutex_destroy(&scrub_monitor->donelock); + pthread_cond_destroy(&scrub_monitor->donecond); + + LOCK_DESTROY(&scrub_monitor->lock); +} + +static void +br_free_children(xlator_t *this, br_private_t *priv, int count) +{ + br_child_t *child = NULL; + + for (--count; count >= 0; count--) { + child = &priv->children[count]; + mem_pool_destroy(child->timer_pool); + pthread_mutex_destroy(&child->lock); + } + + GF_FREE(priv->children); + priv->children = NULL; +} + +static int +br_init_children(xlator_t *this, br_private_t *priv) +{ + int i = 0; + br_child_t *child = NULL; + xlator_list_t *trav = NULL; + + priv->child_count = xlator_subvolume_count(this); + priv->children = GF_CALLOC(priv->child_count, sizeof(*priv->children), + gf_br_mt_br_child_t); + if (!priv->children) + goto err; + + trav = this->children; + while (trav) { + child = &priv->children[i]; + + pthread_mutex_init(&child->lock, NULL); + child->witnessed = 0; + + br_set_child_state(child, BR_CHILD_STATE_DISCONNECTED); + + child->this = this; + child->xl = trav->xlator; + + child->timer_pool = mem_pool_new(struct gf_tw_timer_list, 4096); + if (!child->timer_pool) { + gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_MEM_POOL_ALLOC, + NULL); + errno = ENOMEM; + goto freechild; + } + + INIT_LIST_HEAD(&child->list); + + i++; + trav = trav->next; + } + + return 0; + +freechild: + br_free_children(this, priv, i); +err: + return -1; +} + +int32_t +init(xlator_t *this) +{ + int32_t ret = -1; + br_private_t *priv = NULL; + + if (!this->children) { + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_NO_CHILD, NULL); + goto out; + } + + priv = GF_CALLOC(1, sizeof(*priv), gf_br_mt_br_private_t); + if (!priv) { + gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_NO_MEMORY, NULL); + goto out; + } + + GF_OPTION_INIT("scrubber", priv->iamscrubber, bool, free_priv); + + ret = br_init_children(this, priv); + if (ret) + goto free_priv; + + pthread_mutex_init(&priv->lock, NULL); + pthread_cond_init(&priv->cond, NULL); + + INIT_LIST_HEAD(&priv->bricks); + INIT_LIST_HEAD(&priv->signing); + + priv->timer_wheel = glusterfs_ctx_tw_get(this->ctx); + if (!priv->timer_wheel) { + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_TIMER_WHEEL_UNAVAILABLE, + NULL); + goto cleanup; + } + + this->private = priv; + + if (!priv->iamscrubber) { + ret = br_signer_init(this, priv); + if (!ret) + ret = br_signer_handle_options(this, priv, NULL); + } else { + ret = br_scrubber_init(this, priv); + if (!ret) + ret = br_scrubber_handle_options(this, priv, NULL); + } + + if (ret) + goto cleanup; + + ret = gf_thread_create(&priv->thread, NULL, br_handle_events, this, + "brhevent"); + if (ret != 0) { + gf_smsg(this->name, GF_LOG_ERROR, -ret, BRB_MSG_THREAD_CREATION_FAILED, + NULL); + ret = -1; + } + + if (!ret) { + gf_smsg(this->name, GF_LOG_INFO, 0, BRB_MSG_BITROT_LOADED, "mode=%s", + (priv->iamscrubber) ? "SCRUBBER" : "SIGNER", NULL); + return 0; + } + +cleanup: + (void)pthread_cond_destroy(&priv->cond); + (void)pthread_mutex_destroy(&priv->lock); + + br_free_children(this, priv, priv->child_count); + +free_priv: + GF_FREE(priv); +out: + this->private = NULL; + return -1; +} + +void +fini(xlator_t *this) +{ + br_private_t *priv = this->private; + + if (!priv) + return; + + if (!priv->iamscrubber) + br_fini_signer(this, priv); + else + (void)br_free_scrubber_monitor(this, priv); + + br_free_children(this, priv, priv->child_count); + + this->private = NULL; + GF_FREE(priv); + + glusterfs_ctx_tw_put(this->ctx); + + return; +} + +static void +br_reconfigure_monitor(xlator_t *this) +{ + int32_t ret = 0; + + ret = br_scrub_state_machine(this, _gf_false); + if (ret) { + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_COULD_NOT_SCHEDULE_SCRUB, + NULL); + } +} + +static int +br_reconfigure_scrubber(xlator_t *this, dict_t *options) +{ + int32_t ret = -1; + br_private_t *priv = NULL; + + priv = this->private; + + pthread_mutex_lock(&priv->lock); + { + ret = br_scrubber_handle_options(this, priv, options); + } + pthread_mutex_unlock(&priv->lock); + + if (ret) + goto err; + + /* change state for all _up_ subvolume(s) */ + pthread_mutex_lock(&priv->lock); + { + br_reconfigure_monitor(this); + } + pthread_mutex_unlock(&priv->lock); + +err: + return ret; +} + +static int +br_reconfigure_signer(xlator_t *this, dict_t *options) +{ + br_private_t *priv = this->private; + + return br_signer_handle_options(this, priv, options); +} + +int +reconfigure(xlator_t *this, dict_t *options) +{ + int ret = 0; + br_private_t *priv = NULL; + + priv = this->private; + + if (priv->iamscrubber) + ret = br_reconfigure_scrubber(this, options); + else + ret = br_reconfigure_signer(this, options); + + return ret; +} + +struct xlator_fops fops; + +struct xlator_cbks cbks; + +struct volume_options options[] = { + { + .key = {"expiry-time"}, + .type = GF_OPTION_TYPE_INT, + .default_value = SIGNING_TIMEOUT, + .op_version = {GD_OP_VERSION_3_7_0}, + .flags = OPT_FLAG_SETTABLE, + .description = "Waiting time for an object on which it waits " + "before it is signed", + }, + { + .key = {"brick-count"}, + .type = GF_OPTION_TYPE_STR, + .description = "Total number of bricks for the current node for " + "all volumes in the trusted storage pool.", + }, + { + .key = {"scrubber", "scrub"}, + .type = GF_OPTION_TYPE_BOOL, + .default_value = "false", + .op_version = {GD_OP_VERSION_3_7_0}, + .flags = OPT_FLAG_SETTABLE | OPT_FLAG_FORCE, + .description = "option to run as a scrubber", + }, + { + .key = {"scrub-throttle"}, + .type = GF_OPTION_TYPE_STR, + .default_value = "lazy", + .op_version = {GD_OP_VERSION_3_7_0}, + .flags = OPT_FLAG_SETTABLE, + .description = "Scrub-throttle value is a measure of how fast " + "or slow the scrubber scrubs the filesystem for " + "volume <VOLNAME>", + }, + { + .key = {"scrub-freq"}, + .type = GF_OPTION_TYPE_STR, + .default_value = "biweekly", + .op_version = {GD_OP_VERSION_3_7_0}, + .flags = OPT_FLAG_SETTABLE, + .description = "Scrub frequency for volume <VOLNAME>", + }, + { + .key = {"scrub-state"}, + .type = GF_OPTION_TYPE_STR, + .default_value = "active", + .op_version = {GD_OP_VERSION_4_0_0}, + .flags = OPT_FLAG_SETTABLE, + .description = "Pause/Resume scrub. Upon resume, scrubber " + "continues from where it left off.", + }, + { + .key = {"signer-threads"}, + .type = GF_OPTION_TYPE_INT, + .default_value = BR_WORKERS, + .op_version = {GD_OP_VERSION_8_0}, + .flags = OPT_FLAG_SETTABLE, + .description = "Number of signing process threads. As a best " + "practice, set this to the number of processor cores", + }, + {.key = {NULL}}, +}; + +xlator_api_t xlator_api = { + .init = init, + .fini = fini, + .notify = notify, + .reconfigure = reconfigure, + .mem_acct_init = mem_acct_init, + .op_version = {1}, /* Present from the initial version */ + .fops = &fops, + .cbks = &cbks, + .options = options, + .identifier = "bit-rot", + .category = GF_MAINTAINED, +}; diff --git a/xlators/features/bit-rot/src/bitd/bit-rot.h b/xlators/features/bit-rot/src/bitd/bit-rot.h new file mode 100644 index 00000000000..8ac7dcdac3d --- /dev/null +++ b/xlators/features/bit-rot/src/bitd/bit-rot.h @@ -0,0 +1,302 @@ +/* + Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com> + 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. +*/ + +#ifndef __BIT_ROT_H__ +#define __BIT_ROT_H__ + +#include <glusterfs/glusterfs.h> +#include <glusterfs/logging.h> +#include <glusterfs/dict.h> +#include <glusterfs/xlator.h> +#include <glusterfs/defaults.h> +#include <glusterfs/syncop.h> +#include <glusterfs/syncop-utils.h> +#include "changelog.h" +#include "timer-wheel.h" + +#include <glusterfs/throttle-tbf.h> +#include "bit-rot-ssm.h" + +#include "bit-rot-common.h" +#include "bit-rot-stub-mem-types.h" +#include "bit-rot-scrub-status.h" + +#include <openssl/sha.h> + +typedef enum scrub_throttle { + BR_SCRUB_THROTTLE_VOID = -1, + BR_SCRUB_THROTTLE_LAZY = 0, + BR_SCRUB_THROTTLE_NORMAL = 1, + BR_SCRUB_THROTTLE_AGGRESSIVE = 2, + BR_SCRUB_THROTTLE_STALLED = 3, +} scrub_throttle_t; + +typedef enum scrub_freq { + BR_FSSCRUB_FREQ_HOURLY = 1, + BR_FSSCRUB_FREQ_DAILY, + BR_FSSCRUB_FREQ_WEEKLY, + BR_FSSCRUB_FREQ_BIWEEKLY, + BR_FSSCRUB_FREQ_MONTHLY, + BR_FSSCRUB_FREQ_MINUTE, + BR_FSSCRUB_FREQ_STALLED, +} scrub_freq_t; + +#define signature_size(hl) (sizeof(br_isignature_t) + hl + 1) + +struct br_scanfs { + gf_lock_t entrylock; + + pthread_mutex_t waitlock; + pthread_cond_t waitcond; + + unsigned int entries; + struct list_head queued; + struct list_head ready; +}; + +/* just need three states to track child status */ +typedef enum br_child_state { + BR_CHILD_STATE_CONNECTED = 1, + BR_CHILD_STATE_INITIALIZING, + BR_CHILD_STATE_CONNFAILED, + BR_CHILD_STATE_DISCONNECTED, +} br_child_state_t; + +struct br_child { + pthread_mutex_t lock; /* protects child state */ + char witnessed; /* witnessed at least one successful + connection */ + br_child_state_t c_state; /* current state of this child */ + + char child_up; /* Indicates whether this child is + up or not */ + xlator_t *xl; /* client xlator corresponding to + this child */ + inode_table_t *table; /* inode table for this child */ + char brick_path[PATH_MAX]; /* brick export directory of this + child */ + struct list_head list; /* hook to attach to the list of + UP children */ + xlator_t *this; /* Bit rot xlator */ + + pthread_t thread; /* initial crawler for unsigned + object(s) or scrub crawler */ + int threadrunning; /* active thread */ + + struct mem_pool *timer_pool; /* timer-wheel's timer mem-pool */ + + struct timeval tv; + + struct br_scanfs fsscan; /* per subvolume FS scanner */ + + gf_boolean_t active_scrubbing; /* Actively scrubbing or not */ +}; + +typedef struct br_child br_child_t; + +struct br_obj_n_workers { + struct list_head objects; /* queue of objects expired from the + timer wheel and ready to be picked + up for signing */ + pthread_t *workers; /* Threads which pick up the objects + from the above queue and start + signing each object */ +}; + +struct br_scrubber { + xlator_t *this; + + scrub_throttle_t throttle; + + /** + * frequency of scanning for this subvolume. this should + * normally be per-child, but since all children follow the + * same frequency for a volume, this option ends up here + * instead of br_child_t. + */ + scrub_freq_t frequency; + + gf_boolean_t frequency_reconf; + gf_boolean_t throttle_reconf; + + pthread_mutex_t mutex; + pthread_cond_t cond; + + unsigned int nr_scrubbers; + struct list_head scrubbers; + + /** + * list of "rotatable" subvolume(s) undergoing scrubbing + */ + struct list_head scrublist; +}; + +struct br_monitor { + gf_lock_t lock; + pthread_t thread; /* Monitor thread */ + + gf_boolean_t inited; + pthread_mutex_t mutex; + pthread_cond_t cond; /* Thread starts and will be waiting on cond. + First child which is up wakes this up */ + + xlator_t *this; + /* scheduler */ + uint32_t boot; + + int32_t active_child_count; /* Number of children currently scrubbing */ + gf_boolean_t kick; /* This variable tracks the scrubber is + * kicked or not. Both 'kick' and + * 'active_child_count' uses the same pair + * of mutex-cond variable, i.e, wakelock and + * wakecond. */ + + pthread_mutex_t wakelock; + pthread_cond_t wakecond; + + gf_boolean_t done; + pthread_mutex_t donelock; + pthread_cond_t donecond; + + struct gf_tw_timer_list *timer; + br_scrub_state_t state; /* current scrub state */ +}; + +typedef struct br_obj_n_workers br_obj_n_workers_t; + +typedef struct br_private br_private_t; + +typedef void (*br_scrubbed_file_update)(br_private_t *priv); + +struct br_private { + pthread_mutex_t lock; + + struct list_head bricks; /* list of bricks from which enents + have been received */ + + struct list_head signing; + + pthread_cond_t object_cond; /* handling signing of objects */ + int child_count; + br_child_t *children; /* list of subvolumes */ + int up_children; + + pthread_cond_t cond; /* handling CHILD_UP notifications */ + pthread_t thread; /* thread for connecting each UP + child with changelog */ + + struct tvec_base *timer_wheel; /* timer wheel where the objects which + changelog has sent sits and waits + for expiry */ + br_obj_n_workers_t *obj_queue; /* place holder for all the objects + that are expired from timer wheel + and ready to be picked up for + signing and the workers which sign + the objects */ + + uint32_t expiry_time; /* objects "wait" time */ + + uint32_t signer_th_count; /* Number of signing process threads */ + + tbf_t *tbf; /* token bucket filter */ + + gf_boolean_t iamscrubber; /* function as a fs scrubber */ + + struct br_scrub_stats scrub_stat; /* statistics of scrub*/ + + struct br_scrubber fsscrub; /* scrubbers for this subvolume */ + + struct br_monitor scrub_monitor; /* scrubber monitor */ +}; + +struct br_object { + xlator_t *this; + + uuid_t gfid; + + unsigned long signedversion; /* version against which this object will + be signed */ + br_child_t *child; /* object's subvolume */ + + int sign_info; + + struct list_head list; /* hook to add to the queue once the + object is expired from timer wheel */ + void *data; +}; + +typedef struct br_object br_object_t; +typedef int32_t(br_scrub_ssm_call)(xlator_t *); + +void +br_log_object(xlator_t *, char *, uuid_t, int32_t); + +void +br_log_object_path(xlator_t *, char *, const char *, int32_t); + +int32_t +br_calculate_obj_checksum(unsigned char *, br_child_t *, fd_t *, struct iatt *); + +int32_t +br_prepare_loc(xlator_t *, br_child_t *, loc_t *, gf_dirent_t *, loc_t *); + +gf_boolean_t +bitd_is_bad_file(xlator_t *, br_child_t *, loc_t *, fd_t *); + +static inline void +_br_set_child_state(br_child_t *child, br_child_state_t state) +{ + child->c_state = state; +} + +static inline int +_br_is_child_connected(br_child_t *child) +{ + return (child->c_state == BR_CHILD_STATE_CONNECTED); +} + +static inline int +_br_is_child_scrub_active(br_child_t *child) +{ + return child->active_scrubbing; +} + +static inline int +_br_child_failed_conn(br_child_t *child) +{ + return (child->c_state == BR_CHILD_STATE_CONNFAILED); +} + +static inline int +_br_child_witnessed_connection(br_child_t *child) +{ + return (child->witnessed == 1); +} + +/* scrub state */ +static inline void +_br_monitor_set_scrub_state(struct br_monitor *scrub_monitor, + br_scrub_state_t state) +{ + scrub_monitor->state = state; +} + +static inline br_scrub_event_t +_br_child_get_scrub_event(struct br_scrubber *fsscrub) +{ + return (fsscrub->frequency == BR_FSSCRUB_FREQ_STALLED) + ? BR_SCRUB_EVENT_PAUSE + : BR_SCRUB_EVENT_SCHEDULE; +} + +int32_t +br_get_bad_objects_list(xlator_t *this, dict_t **dict); + +#endif /* __BIT_ROT_H__ */ |