1
0
mirror of git://sourceware.org/git/lvm2.git synced 2024-12-22 17:35:59 +03:00
lvm2/tools/polldaemon.c
David Teigland c7fdacbc50 pvmove: in fork mode destroy bcache in child
When pvmove was run in background mode and forks
instead of using lvmpolld, the child pvmove process
was not clearing the bcache from the parent, so all
the aio ops in the child were failing.
2018-04-25 16:40:36 -05:00

705 lines
19 KiB
C

/*
* Copyright (C) 2003-2004 Sistina Software, Inc. All rights reserved.
* Copyright (C) 2004-2015 Red Hat, Inc. All rights reserved.
*
* This file is part of LVM2.
*
* This copyrighted material is made available to anyone wishing to use,
* modify, copy, or redistribute it subject to the terms and conditions
* of the GNU Lesser General Public License v.2.1.
*
* You should have received a copy of the GNU Lesser General Public License
* along with this program; if not, write to the Free Software Foundation,
* Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include "tools.h"
#include "polldaemon.h"
#include "lvm2cmdline.h"
#include "lvmpolld-client.h"
#include <time.h>
#define WAIT_AT_LEAST_NANOSECS 100000
progress_t poll_mirror_progress(struct cmd_context *cmd,
struct logical_volume *lv, const char *name,
struct daemon_parms *parms)
{
dm_percent_t segment_percent = DM_PERCENT_0, overall_percent = DM_PERCENT_0;
uint32_t event_nr = 0;
if (!lv_is_mirrored(lv) ||
!lv_mirror_percent(cmd, lv, !parms->interval, &segment_percent,
&event_nr) ||
(segment_percent == DM_PERCENT_INVALID)) {
log_error("ABORTING: Mirror percentage check failed.");
return PROGRESS_CHECK_FAILED;
}
overall_percent = copy_percent(lv);
if (parms->progress_display)
log_print_unless_silent("%s: %s: %s%%", name, parms->progress_title,
display_percent(cmd, overall_percent));
else
log_verbose("%s: %s: %s%%", name, parms->progress_title,
display_percent(cmd, overall_percent));
if (segment_percent != DM_PERCENT_100)
return PROGRESS_UNFINISHED;
if (overall_percent == DM_PERCENT_100)
return PROGRESS_FINISHED_ALL;
return PROGRESS_FINISHED_SEGMENT;
}
static int _check_lv_status(struct cmd_context *cmd,
struct volume_group *vg,
struct logical_volume *lv,
const char *name, struct daemon_parms *parms,
int *finished)
{
struct dm_list *lvs_changed;
progress_t progress;
/* By default, caller should not retry */
*finished = 1;
if (parms->aborting) {
if (!(lvs_changed = lvs_using_lv(cmd, vg, lv))) {
log_error("Failed to generate list of copied LVs: "
"can't abort.");
return 0;
}
if (!parms->poll_fns->finish_copy(cmd, vg, lv, lvs_changed))
return_0;
return 1;
}
progress = parms->poll_fns->poll_progress(cmd, lv, name, parms);
if (progress == PROGRESS_CHECK_FAILED)
return_0;
if (progress == PROGRESS_UNFINISHED) {
/* The only case the caller *should* try again later */
*finished = 0;
return 1;
}
if (!(lvs_changed = lvs_using_lv(cmd, vg, lv))) {
log_error("ABORTING: Failed to generate list of copied LVs");
return 0;
}
/* Finished? Or progress to next segment? */
if (progress == PROGRESS_FINISHED_ALL) {
if (!parms->poll_fns->finish_copy(cmd, vg, lv, lvs_changed))
return_0;
} else {
if (parms->poll_fns->update_metadata &&
!parms->poll_fns->update_metadata(cmd, vg, lv, lvs_changed, 0)) {
log_error("ABORTING: Segment progression failed.");
parms->poll_fns->finish_copy(cmd, vg, lv, lvs_changed);
return 0;
}
*finished = 0; /* Another segment */
}
return 1;
}
static void _nanosleep(unsigned secs, unsigned allow_zero_time)
{
struct timespec wtime = {
.tv_sec = secs,
};
if (!secs && !allow_zero_time)
wtime.tv_nsec = WAIT_AT_LEAST_NANOSECS;
while (!nanosleep(&wtime, &wtime) && errno == EINTR) {}
}
static void _sleep_and_rescan_devices(struct cmd_context *cmd, struct daemon_parms *parms)
{
if (parms->interval && !parms->aborting) {
/*
* FIXME: do we really need to drop everything and then rescan
* everything between each iteration? What change exactly does
* each iteration check for, and does seeing that require
* rescanning everything?
*/
lvmcache_destroy(cmd, 1, 0);
label_scan_destroy(cmd);
dev_close_all();
_nanosleep(parms->interval, 1);
lvmcache_label_scan(cmd);
}
}
int wait_for_single_lv(struct cmd_context *cmd, struct poll_operation_id *id,
struct daemon_parms *parms)
{
struct volume_group *vg = NULL;
struct logical_volume *lv;
int finished = 0;
uint32_t lockd_state = 0;
int ret;
if (!parms->wait_before_testing)
lvmcache_label_scan(cmd);
/* Poll for completion */
while (!finished) {
if (parms->wait_before_testing)
_sleep_and_rescan_devices(cmd, parms);
/*
* An ex VG lock is needed because the check can call finish_copy
* which writes the VG.
*/
if (!lockd_vg(cmd, id->vg_name, "ex", 0, &lockd_state)) {
log_error("ABORTING: Can't lock VG for %s.", id->display_name);
return 0;
}
/* Locks the (possibly renamed) VG again */
vg = vg_read(cmd, id->vg_name, NULL, READ_FOR_UPDATE, lockd_state);
if (vg_read_error(vg)) {
/* What more could we do here? */
log_error("ABORTING: Can't reread VG for %s.", id->display_name);
release_vg(vg);
vg = NULL;
ret = 0;
goto out;
}
lv = find_lv(vg, id->lv_name);
if (lv && id->uuid && strcmp(id->uuid, (char *)&lv->lvid))
lv = NULL;
if (lv && parms->lv_type && !(lv->status & parms->lv_type))
lv = NULL;
if (!lv) {
if (parms->lv_type == PVMOVE)
log_print_unless_silent("%s: No pvmove in progress - already finished or aborted.",
id->display_name);
else
log_print_unless_silent("Can't find LV in %s for %s.",
vg->name, id->display_name);
ret = 1;
goto out;
}
/*
* If the LV is not active locally, the kernel cannot be
* queried for its status. We must exit in this case.
*/
if (!lv_is_active_locally(lv)) {
log_print_unless_silent("%s: Interrupted: No longer active.", id->display_name);
ret = 1;
goto out;
}
if (!_check_lv_status(cmd, vg, lv, id->display_name, parms, &finished)) {
ret = 0;
goto_out;
}
unlock_and_release_vg(cmd, vg, vg->name);
if (!lockd_vg(cmd, id->vg_name, "un", 0, &lockd_state))
stack;
/*
* FIXME Sleeping after testing, while preferred, also works around
* unreliable "finished" state checking in _percent_run. If the
* above _check_lv_status is deferred until after the first sleep it
* may be that a polldaemon will run without ever completing.
*
* This happens when one snapshot-merge polldaemon is racing with
* another (polling the same LV). The first to see the LV status
* reach the "finished" state will alter the LV that the other
* polldaemon(s) are polling. These other polldaemon(s) can then
* continue polling an LV that doesn't have a "status".
*/
if (!parms->wait_before_testing && !finished)
_sleep_and_rescan_devices(cmd, parms);
}
return 1;
out:
if (vg)
unlock_and_release_vg(cmd, vg, vg->name);
if (!lockd_vg(cmd, id->vg_name, "un", 0, &lockd_state))
stack;
return ret;
}
struct poll_id_list {
struct dm_list list;
struct poll_operation_id *id;
};
static struct poll_operation_id *_copy_poll_operation_id(struct dm_pool *mem,
const struct poll_operation_id *id)
{
struct poll_operation_id *copy;
if (!id || !id->vg_name || !id->lv_name || !id->display_name || !id->uuid) {
log_error(INTERNAL_ERROR "Wrong params for _copy_poll_operation_id.");
return NULL;
}
if (!(copy = dm_pool_alloc(mem, sizeof(*copy)))) {
log_error("Poll operation ID allocation failed.");
return NULL;
}
if (!(copy->display_name = dm_pool_strdup(mem, id->display_name)) ||
!(copy->lv_name = dm_pool_strdup(mem, id->lv_name)) ||
!(copy->vg_name = dm_pool_strdup(mem, id->vg_name)) ||
!(copy->uuid = dm_pool_strdup(mem, id->uuid))) {
log_error("Failed to copy one or more poll_operation_id members.");
dm_pool_free(mem, copy);
return NULL;
}
return copy;
}
static struct poll_id_list* _poll_id_list_create(struct dm_pool *mem,
const struct poll_operation_id *id)
{
struct poll_id_list *idl = (struct poll_id_list *) dm_pool_alloc(mem, sizeof(struct poll_id_list));
if (!idl) {
log_error("Poll ID list allocation failed.");
return NULL;
}
if (!(idl->id = _copy_poll_operation_id(mem, id))) {
dm_pool_free(mem, idl);
return NULL;
}
return idl;
}
static int _poll_vg(struct cmd_context *cmd, const char *vgname,
struct volume_group *vg, struct processing_handle *handle)
{
struct daemon_parms *parms;
struct lv_list *lvl;
struct dm_list idls;
struct poll_id_list *idl;
struct poll_operation_id id;
struct logical_volume *lv;
int finished;
if (!handle || !(parms = (struct daemon_parms *) handle->custom_handle)) {
log_error(INTERNAL_ERROR "Handle is undefined.");
return ECMD_FAILED;
}
dm_list_init(&idls);
/*
* first iterate all LVs in a VG and collect LVs suitable
* for polling (or an abort) which takes place below
*/
dm_list_iterate_items(lvl, &vg->lvs) {
lv = lvl->lv;
if (!(lv->status & parms->lv_type))
continue;
id.display_name = parms->poll_fns->get_copy_name_from_lv(lv);
if (!id.display_name && !parms->aborting)
continue;
if (!id.display_name) {
log_error("Device name for LV %s not found in metadata. "
"(unfinished pvmove mirror removal?)", display_lvname(lv));
goto err;
}
/* FIXME Need to do the activation from _set_up_pvmove here
* if it's not running and we're not aborting. */
if (!lv_is_active(lv)) {
log_print_unless_silent("%s: Skipping inactive LV. Try lvchange or vgchange.", id.display_name);
continue;
}
id.lv_name = lv->name;
id.vg_name = vg->name;
id.uuid = lv->lvid.s;
idl = _poll_id_list_create(cmd->mem, &id);
if (!idl) {
log_error("Failed to create poll_id_list.");
goto err;
}
dm_list_add(&idls, &idl->list);
}
/* perform the poll operation on LVs collected in previous cycle */
dm_list_iterate_items(idl, &idls) {
if (!(lv = find_lv(vg, idl->id->lv_name)))
continue;
if (idl->id->uuid && strcmp(idl->id->uuid, (char *)&lv->lvid))
continue;
if (parms->lv_type && !(lv->status & parms->lv_type))
continue;
if (_check_lv_status(cmd, vg, lv, idl->id->display_name, parms, &finished) && !finished)
parms->outstanding_count++;
}
err:
if (!dm_list_empty(&idls))
dm_pool_free(cmd->mem, dm_list_item(dm_list_first(&idls), struct poll_id_list));
return ECMD_PROCESSED;
}
static void _poll_for_all_vgs(struct cmd_context *cmd,
struct processing_handle *handle)
{
struct daemon_parms *parms = (struct daemon_parms *) handle->custom_handle;
while (1) {
parms->outstanding_count = 0;
process_each_vg(cmd, 0, NULL, NULL, NULL, READ_FOR_UPDATE, 0, handle, _poll_vg);
if (!parms->outstanding_count)
break;
_nanosleep(parms->interval, 1);
}
}
#ifdef LVMPOLLD_SUPPORT
typedef struct {
struct daemon_parms *parms;
struct dm_list idls;
} lvmpolld_parms_t;
static int _report_progress(struct cmd_context *cmd, struct poll_operation_id *id,
struct daemon_parms *parms)
{
struct volume_group *vg;
struct logical_volume *lv;
uint32_t lockd_state = 0;
int ret;
/*
* It's reasonable to expect a lockd_vg("sh") here, but it should
* not actually be needed, because we only report the progress on
* the same host where the pvmove/lvconvert is happening. This means
* that the local pvmove/lvconvert/lvpoll commands are updating the
* local lvmetad with the latest info they have, and we just need to
* read the latest info that they have put into lvmetad about their
* progress. No VG lock is needed to protect anything here (we're
* just reading the VG), and no VG lock is needed to force a VG read
* from disk to get changes from other hosts, because the only change
* to the VG we're interested in is the change done locally.
*/
vg = vg_read(cmd, id->vg_name, NULL, 0, lockd_state);
if (vg_read_error(vg)) {
release_vg(vg);
log_error("Can't reread VG for %s", id->display_name);
ret = 0;
goto out_ret;
}
lv = find_lv(vg, id->lv_name);
if (lv && id->uuid && strcmp(id->uuid, (char *)&lv->lvid))
lv = NULL;
/*
* CONVERTING is set only during mirror upconversion but we may need to
* read LV's progress info even when it's not converting (linear->mirror)
*/
if (lv && (parms->lv_type ^ CONVERTING) && !(lv->status & parms->lv_type))
lv = NULL;
if (!lv) {
if (parms->lv_type == PVMOVE)
log_verbose("%s: No pvmove in progress - already finished or aborted.",
id->display_name);
else
log_verbose("Can't find LV in %s for %s. Already finished or removed.",
vg->name, id->display_name);
ret = 1;
goto out;
}
if (!lv_is_active_locally(lv)) {
log_verbose("%s: Interrupted: No longer active.", id->display_name);
ret = 1;
goto out;
}
if (parms->poll_fns->poll_progress(cmd, lv, id->display_name, parms) == PROGRESS_CHECK_FAILED) {
ret = 0;
goto out;
}
ret = 1;
out:
unlock_and_release_vg(cmd, vg, vg->name);
out_ret:
return ret;
}
static int _lvmpolld_init_poll_vg(struct cmd_context *cmd, const char *vgname,
struct volume_group *vg, struct processing_handle *handle)
{
int r;
struct lv_list *lvl;
struct logical_volume *lv;
struct poll_id_list *idl;
struct poll_operation_id id;
lvmpolld_parms_t *lpdp = (lvmpolld_parms_t *) handle->custom_handle;
dm_list_iterate_items(lvl, &vg->lvs) {
lv = lvl->lv;
if (!(lv->status & lpdp->parms->lv_type))
continue;
id.display_name = lpdp->parms->poll_fns->get_copy_name_from_lv(lv);
if (!id.display_name && !lpdp->parms->aborting)
continue;
id.vg_name = lv->vg->name;
id.lv_name = lv->name;
if (!*lv->lvid.s) {
log_print_unless_silent("Missing LV uuid within: %s/%s", id.vg_name, id.lv_name);
continue;
}
id.uuid = lv->lvid.s;
r = lvmpolld_poll_init(cmd, &id, lpdp->parms);
if (r && !lpdp->parms->background) {
if (!(idl = _poll_id_list_create(cmd->mem, &id)))
return ECMD_FAILED;
dm_list_add(&lpdp->idls, &idl->list);
}
}
return ECMD_PROCESSED;
}
static void _lvmpolld_poll_for_all_vgs(struct cmd_context *cmd,
struct daemon_parms *parms,
struct processing_handle *handle)
{
int r;
struct dm_list *first;
struct poll_id_list *idl, *tlv;
unsigned finished;
lvmpolld_parms_t lpdp = {
.parms = parms
};
dm_list_init(&lpdp.idls);
handle->custom_handle = &lpdp;
process_each_vg(cmd, 0, NULL, NULL, NULL, 0, 0, handle, _lvmpolld_init_poll_vg);
first = dm_list_first(&lpdp.idls);
while (!dm_list_empty(&lpdp.idls)) {
dm_list_iterate_items_safe(idl, tlv, &lpdp.idls) {
r = lvmpolld_request_info(idl->id, lpdp.parms,
&finished);
if (!r || finished)
dm_list_del(&idl->list);
else if (!parms->aborting)
_report_progress(cmd, idl->id, lpdp.parms);
}
if (lpdp.parms->interval)
dev_close_all();
_nanosleep(lpdp.parms->interval, 0);
}
if (first)
dm_pool_free(cmd->mem, dm_list_item(first, struct poll_id_list));
}
static int _lvmpoll_daemon(struct cmd_context *cmd, struct poll_operation_id *id,
struct daemon_parms *parms)
{
int r;
struct processing_handle *handle = NULL;
unsigned finished = 0;
if (parms->aborting)
parms->interval = 0;
if (id) {
r = lvmpolld_poll_init(cmd, id, parms);
if (r && !parms->background) {
while (1) {
if (!(r = lvmpolld_request_info(id, parms, &finished)) ||
finished ||
(!parms->aborting && !(r = _report_progress(cmd, id, parms))))
break;
if (parms->interval)
dev_close_all();
_nanosleep(parms->interval, 0);
}
}
return r ? ECMD_PROCESSED : ECMD_FAILED;
}
/* process all in-flight operations */
if (!(handle = init_processing_handle(cmd, NULL))) {
log_error("Failed to initialize processing handle.");
return ECMD_FAILED;
}
_lvmpolld_poll_for_all_vgs(cmd, parms, handle);
destroy_processing_handle(cmd, handle);
return ECMD_PROCESSED;
}
#else
# define _lvmpoll_daemon(cmd, id, parms) (ECMD_FAILED)
#endif /* LVMPOLLD_SUPPORT */
/*
* Only allow *one* return from poll_daemon() (the parent).
* If there is a child it must exit (ignoring the memory leak messages).
* - 'background' is advisory so a child polldaemon may not be used even
* if it was requested.
*/
static int _poll_daemon(struct cmd_context *cmd, struct poll_operation_id *id,
struct daemon_parms *parms)
{
struct processing_handle *handle = NULL;
int daemon_mode = 0;
int ret = ECMD_PROCESSED;
if (parms->background) {
daemon_mode = become_daemon(cmd, 0);
if (daemon_mode == 0)
return ECMD_PROCESSED; /* Parent */
if (daemon_mode == 1)
parms->progress_display = 0; /* Child */
/* FIXME Use wait_event (i.e. interval = 0) and */
/* fork one daemon per copy? */
if ((daemon_mode == 1) && find_config_tree_bool(cmd, global_use_lvmetad_CFG, NULL)) {
if (!lvmetad_connect(cmd))
log_warn("WARNING: lvm polling process %d cannot connect to lvmetad.", getpid());
}
}
/*
* Process one specific task or all incomplete tasks?
*/
/* clear lvmcache/bcache/fds from the parent */
lvmcache_destroy(cmd, 1, 0);
label_scan_destroy(cmd);
dev_close_all();
if (id) {
if (!wait_for_single_lv(cmd, id, parms)) {
stack;
ret = ECMD_FAILED;
}
} else {
if (!parms->interval)
parms->interval = find_config_tree_int(cmd, activation_polling_interval_CFG, NULL);
if (!(handle = init_processing_handle(cmd, NULL))) {
log_error("Failed to initialize processing handle.");
ret = ECMD_FAILED;
} else {
handle->custom_handle = parms;
_poll_for_all_vgs(cmd, handle);
}
}
if (parms->background && daemon_mode == 1) {
destroy_processing_handle(cmd, handle);
/*
* child was successfully forked:
* background polldaemon must not return to the caller
* because it will redundantly continue performing the
* caller's task (that the parent already performed)
*/
/* FIXME Attempt proper cleanup */
_exit(lvm_return_code(ret));
}
destroy_processing_handle(cmd, handle);
return ret;
}
static int _daemon_parms_init(struct cmd_context *cmd, struct daemon_parms *parms,
unsigned background, struct poll_functions *poll_fns,
const char *progress_title, uint64_t lv_type)
{
sign_t interval_sign;
parms->aborting = arg_is_set(cmd, abort_ARG);
parms->background = background;
interval_sign = arg_sign_value(cmd, interval_ARG, SIGN_NONE);
if (interval_sign == SIGN_MINUS) {
log_error("Argument to --interval cannot be negative.");
return 0;
}
parms->interval = arg_uint_value(cmd, interval_ARG,
find_config_tree_int(cmd, activation_polling_interval_CFG, NULL));
parms->wait_before_testing = (interval_sign == SIGN_PLUS);
parms->progress_title = progress_title;
parms->lv_type = lv_type;
parms->poll_fns = poll_fns;
if (parms->interval && !parms->aborting)
log_verbose("Checking progress %s waiting every %u seconds.",
(parms->wait_before_testing ? "after" : "before"),
parms->interval);
parms->progress_display = parms->interval ? 1 : 0;
return 1;
}
int poll_daemon(struct cmd_context *cmd, unsigned background,
uint64_t lv_type, struct poll_functions *poll_fns,
const char *progress_title, struct poll_operation_id *id)
{
struct daemon_parms parms;
if (!_daemon_parms_init(cmd, &parms, background, poll_fns, progress_title, lv_type))
return_EINVALID_CMD_LINE;
if (lvmpolld_use())
return _lvmpoll_daemon(cmd, id, &parms);
/* classical polling allows only PMVOVE or 0 values */
parms.lv_type &= PVMOVE;
return _poll_daemon(cmd, id, &parms);
}