static int open_log(struct asfd *asfd,
struct sdirs *sdirs, struct conf **cconfs)
{
int ret=-1;
char *logpath=NULL;
const char *peer_version=get_string(cconfs[OPT_PEER_VERSION]);
if(!(logpath=prepend_s(sdirs->rworking, "log"))) goto end;
if(log_fzp_set(logpath, cconfs))
{
logp("could not open log file: %s\n", logpath);
goto end;
}
logp("Client version: %s\n", peer_version?:"");
logp("Protocol: %d\n", (int)get_protocol(cconfs));
if(get_int(cconfs[OPT_CLIENT_IS_WINDOWS]))
logp("Client is Windows\n");
// Make sure a warning appears in the backup log.
// The client will already have been sent a message with logw.
// This time, prevent it sending a logw to the client by specifying
// NULL for cntr.
if(get_int(cconfs[OPT_VERSION_WARN])) version_warn(asfd, NULL, cconfs);
ret=0;
end:
free_w(&logpath);
return ret;
}
int reload(struct conf **confs, const char *conffile, bool firsttime)
{
if(!firsttime) logp("Reloading config\n");
if(confs_init(confs)) return -1;
if(conf_load_global_only(conffile, confs)) return -1;
umask(get_mode_t(confs[OPT_UMASK]));
// This will turn on syslogging which could not be turned on before
// conf_load.
log_fzp_set(NULL, confs);
#ifndef HAVE_WIN32
if(get_e_burp_mode(confs[OPT_BURP_MODE])==BURP_MODE_SERVER)
setup_signals();
#endif
// Do not try to change user or group after the first time.
if(firsttime && chuser_and_or_chgrp(
get_string(confs[OPT_USER]), get_string(confs[OPT_GROUP])))
return -1;
return 0;
}
static int backup_phase4_server(struct sdirs *sdirs, struct conf **cconfs)
{
int breaking=get_int(cconfs[OPT_BREAKPOINT]);
if(breaking==4)
return breakpoint(breaking, __func__);
log_fzp_set(NULL, cconfs);
// Phase4 will open logfp again (in case it is resuming).
switch(get_protocol(cconfs))
{
case PROTO_1:
return backup_phase4_server_protocol1(sdirs, cconfs);
default:
return backup_phase4_server_protocol2(sdirs, cconfs);
}
}
static int champ_chooser_fork(struct sdirs *sdirs, struct conf **confs,
int resume)
{
pid_t childpid=-1;
int cret;
if(!get_int(confs[OPT_FORK]))
{
logp("Not forking a champ chooser process.\n");
// They need to manually run a separate process.
return 0;
}
switch((childpid=fork()))
{
case -1:
logp("fork failed in %s: %s\n",
__func__, strerror(errno));
return -1;
case 0:
// Child.
log_fzp_set(NULL, confs);
switch(champ_chooser_server(sdirs, confs, resume))
{
case 0:
cret=0;
break;
default:
cret=1;
break;
}
exit(cret);
default:
// Parent.
logp("forked champ chooser pid %d\n", childpid);
return 0;
}
return -1; // Not reached.
}
static void run_test(
int expected_result,
int entries,
void setup_datadir_tmp_callback(
struct slist *slist, struct fdirs *fdirs, struct conf **confs))
{
struct conf **confs;
struct sdirs *sdirs;
struct fdirs *fdirs;
struct slist *slist;
setup(&sdirs, &fdirs, &confs);
build_storage_dirs(sdirs, sd1, ARR_LEN(sd1));
slist=build_manifest(
fdirs->manifest,
PROTO_1,
entries,
/*phase*/ 3);
setup_datadir_tmp_callback(slist, fdirs, confs);
clock_t start;
clock_t diff;
start = clock();
fail_unless(backup_phase4_server_protocol1(sdirs, confs)
==expected_result);
diff = clock() - start;
int msec = diff * 1000 / CLOCKS_PER_SEC;
printf("%d.%d\n", msec/1000, msec%1000);
log_fzp_set(NULL, confs);
assert_datadir(slist, fdirs);
slist_free(&slist);
tear_down(&sdirs, &fdirs, &confs);
}
static int do_backup_server(struct async *as, struct sdirs *sdirs,
struct conf **cconfs, const char *incexc, int resume)
{
int ret=0;
int do_phase2=1;
struct asfd *asfd=as->asfd;
enum protocol protocol=get_protocol(cconfs);
struct cntr *cntr=get_cntr(cconfs);
logp("in do_backup_server\n");
log_rshash(cconfs);
if(resume)
{
if(sdirs_get_real_working_from_symlink(sdirs)
|| sdirs_get_real_manifest(sdirs, protocol)
|| open_log(asfd, sdirs, cconfs))
goto error;
}
else
{
// Not resuming - need to set everything up fresh.
if(sdirs_create_real_working(sdirs,
get_string(cconfs[OPT_TIMESTAMP_FORMAT]))
|| sdirs_get_real_manifest(sdirs, protocol)
|| open_log(asfd, sdirs, cconfs))
goto error;
if(write_incexc(sdirs->rworking, incexc))
{
logp("unable to write incexc\n");
goto error;
}
if(backup_phase1_server(as, sdirs, cconfs))
{
logp("error in phase 1\n");
goto error;
}
}
if(resume)
{
struct stat statp;
if(lstat(sdirs->phase1data, &statp)
&& !lstat(sdirs->changed, &statp)
&& !lstat(sdirs->unchanged, &statp))
{
// In this condition, it looks like there was an
// interruption during phase3. Skip phase2.
do_phase2=0;
}
}
if(do_phase2)
{
if(backup_phase2_server(as, sdirs, incexc, resume, cconfs))
{
logp("error in backup phase 2\n");
goto error;
}
asfd->write_str(asfd, CMD_GEN, "okbackupend");
}
// Close the connection with the client, the rest of the job we can do
// by ourselves.
logp("Backup ending - disconnect from client.\n");
if(asfd_flush_asio(asfd)) goto end;
as->asfd_remove(as, asfd);
asfd_close(asfd);
if(backup_phase3_server(sdirs, cconfs))
{
logp("error in backup phase 3\n");
goto error;
}
if(do_rename(sdirs->working, sdirs->finishing))
goto error;
if(backup_phase4_server(sdirs, cconfs))
{
logp("error in backup phase 4\n");
goto error;
}
cntr_print(cntr, ACTION_BACKUP, asfd);
cntr_stats_to_file(cntr, sdirs->rworking, ACTION_BACKUP);
if(protocol==PROTO_2)
{
// Regenerate dindex before the symlink is renamed, so that the
// champ chooser cleanup does not try to remove data files
// whilst the dindex regeneration is happening.
if(regenerate_client_dindex(sdirs))
goto error;
}
// Move the symlink to indicate that we are now in the end phase. The
// rename() race condition is automatically recoverable here.
if(do_rename(sdirs->finishing, sdirs->current))
goto error;
logp("Backup completed.\n");
log_fzp_set(NULL, cconfs);
compress_filename(sdirs->rworking,
"log", "log.gz", get_int(cconfs[OPT_COMPRESSION]));
goto end;
error:
ret=-1;
end:
log_fzp_set(NULL, cconfs);
return ret;
}
static int recover_working(struct async *as,
struct sdirs *sdirs, const char *incexc,
int *resume, struct conf **cconfs)
{
int ret=-1;
char msg[256]="";
char *logpath=NULL;
struct stat statp;
char *phase1datatmp=NULL;
enum recovery_method recovery_method=get_e_recovery_method(
cconfs[OPT_WORKING_DIR_RECOVERY_METHOD]);
// The working directory has not finished being populated.
// Check what to do.
if(get_fullrealwork(as->asfd, sdirs, cconfs)) goto end;
if(!sdirs->rworking) goto end;
log_recovery_method(sdirs, recovery_method);
if(!(phase1datatmp=get_tmp_filename(sdirs->phase1data)))
goto end;
// If there is still a phase1 tmp file...
if(!lstat(phase1datatmp, &statp)
||
// ...or phase1 has not even got underway yet...
(lstat(phase1datatmp, &statp)
&& lstat(sdirs->phase1data, &statp)
&& lstat(sdirs->changed, &statp)
&& lstat(sdirs->unchanged, &statp)))
{
// ...phase 1 did not complete - delete everything.
logp("Phase 1 has not completed.\n");
recovery_method=RECOVERY_METHOD_DELETE;
}
if(recovery_method==RECOVERY_METHOD_DELETE)
{
ret=working_delete(as, sdirs, cconfs);
goto end;
}
// We are not deleting the old working directory - open the log inside
// for appending.
if(!(logpath=prepend_s(sdirs->rworking, "log"))
|| log_fzp_set(logpath, cconfs))
goto end;
switch(recovery_method)
{
case RECOVERY_METHOD_DELETE:
// Dealt with above.
break;
case RECOVERY_METHOD_RESUME:
ret=working_resume(as, sdirs, incexc, resume, cconfs);
break;
case RECOVERY_METHOD_UNSET:
default:
snprintf(msg, sizeof(msg),
"Unknown working_dir_recovery_method: %d\n",
(int)recovery_method);
log_and_send(as->asfd, msg);
break;
}
end:
free_w(&logpath);
free_w(&phase1datatmp);
log_fzp_set(NULL, cconfs); // fclose the logfzp
return ret;
}
int champ_chooser_server(struct sdirs *sdirs, struct conf **confs,
int resume)
{
int s;
int ret=-1;
int len;
struct asfd *asfd=NULL;
struct sockaddr_un local;
struct lock *lock=NULL;
struct async *as=NULL;
int started=0;
struct scores *scores=NULL;
const char *directory=get_string(confs[OPT_DIRECTORY]);
if(!(lock=lock_alloc_and_init(sdirs->champlock))
|| build_path_w(sdirs->champlock))
goto end;
lock_get(lock);
switch(lock->status)
{
case GET_LOCK_GOT:
log_fzp_set(sdirs->champlog, confs);
logp("Got champ lock for dedup_group: %s\n",
get_string(confs[OPT_DEDUP_GROUP]));
break;
case GET_LOCK_NOT_GOT:
case GET_LOCK_ERROR:
default:
//logp("Did not get champ lock\n");
goto end;
}
if((s=socket(AF_UNIX, SOCK_STREAM, 0))<0)
{
logp("socket error in %s: %s\n", __func__, strerror(errno));
goto end;
}
memset(&local, 0, sizeof(struct sockaddr_un));
local.sun_family=AF_UNIX;
snprintf(local.sun_path, sizeof(local.sun_path),
"%s", sdirs->champsock);
len=strlen(local.sun_path)+sizeof(local.sun_family)+1;
unlink(sdirs->champsock);
if(bind(s, (struct sockaddr *)&local, len)<0)
{
logp("bind error in %s: %s\n", __func__, strerror(errno));
goto end;
}
if(listen(s, 5)<0)
{
logp("listen error in %s: %s\n", __func__, strerror(errno));
goto end;
}
if(!(as=async_alloc())
|| as->init(as, 0)
|| !(asfd=setup_asfd(as, "champ chooser main socket", &s,
/*listen*/"")))
goto end;
asfd->fdtype=ASFD_FD_SERVER_LISTEN_MAIN;
// I think that this is probably the best point at which to run a
// cleanup job to delete unused data files, because no other process
// can fiddle with the dedup_group at this point.
// Cannot do it on a resume, or it will delete files that are
// referenced in the backup we are resuming.
if(delete_unused_data_files(sdirs, resume))
goto end;
// Load the sparse indexes for this dedup group.
if(!(scores=champ_chooser_init(sdirs->data)))
goto end;
while(1)
{
for(asfd=as->asfd->next; asfd; asfd=asfd->next)
{
if(!asfd->blist->head
|| asfd->blist->head->got==BLK_INCOMING) continue;
if(results_to_fd(asfd)) goto end;
}
int removed;
switch(as->read_write(as))
{
case 0:
// Check the main socket last, as it might add
// a new client to the list.
for(asfd=as->asfd->next; asfd; asfd=asfd->next)
{
while(asfd->rbuf->buf)
{
if(deal_with_client_rbuf(asfd,
directory, scores))
goto end;
// Get as much out of the
// readbuf as possible.
if(asfd->parse_readbuf(asfd))
goto end;
}
}
if(as->asfd->new_client)
{
// Incoming client.
as->asfd->new_client=0;
if(champ_chooser_new_client(as, confs))
goto end;
started=1;
}
break;
default:
removed=0;
// Maybe one of the fds had a problem.
// Find and remove it and carry on if possible.
for(asfd=as->asfd->next; asfd; )
{
struct asfd *a;
if(!asfd->want_to_remove)
{
asfd=asfd->next;
continue;
}
as->asfd_remove(as, asfd);
logp("%s: disconnected fd %d\n",
asfd->desc, asfd->fd);
a=asfd->next;
asfd_free(&asfd);
asfd=a;
removed++;
}
if(removed) break;
// If we got here, there was no fd to remove.
// It is a fatal error.
goto end;
}
if(started && !as->asfd->next)
{
logp("All clients disconnected.\n");
ret=0;
break;
}
}
end:
logp("champ chooser exiting: %d\n", ret);
champ_chooser_free(&scores);
log_fzp_set(NULL, confs);
async_free(&as);
asfd_free(&asfd); // This closes s for us.
close_fd(&s);
unlink(sdirs->champsock);
// FIX THIS: free asfds.
lock_release(lock);
lock_free(&lock);
return ret;
}
int backup_phase4_server_protocol1(struct sdirs *sdirs, struct conf **cconfs)
{
int ret=-1;
struct stat statp;
char realcurrent[256]="";
uint64_t bno=0;
int hardlinked_current=0;
char tstmp[64]="";
int previous_backup=0;
struct fdirs *fdirs=NULL;
readlink_w(sdirs->current, realcurrent, sizeof(realcurrent));
if(!(fdirs=fdirs_alloc())
|| fdirs_init(fdirs, sdirs, realcurrent))
goto end;
if(log_fzp_set(fdirs->logpath, cconfs))
goto end;
logp("Begin phase4 (shuffle files)\n");
if(write_status(CNTR_STATUS_SHUFFLING, NULL, get_cntr(cconfs)))
goto end;
if(!lstat(sdirs->current, &statp)) // Had a previous backup.
{
previous_backup++;
if(lstat(fdirs->hlinkedcurrent, &statp))
{
hardlinked_current=0;
logp("Previous backup is not a hardlinked_archive\n");
logp(" will generate reverse deltas\n");
}
else
{
hardlinked_current=1;
logp("Previous backup is a hardlinked_archive\n");
logp(" will not generate reverse deltas\n");
}
// If current was not a hardlinked_archive, need to duplicate
// it.
if(!hardlinked_current && lstat(fdirs->currentdup, &statp))
{
// Have not duplicated the current backup yet.
if(!lstat(fdirs->currentduptmp, &statp))
{
logp("Removing previous directory: %s\n",
fdirs->currentduptmp);
if(recursive_delete(fdirs->currentduptmp))
{
logp("Could not delete %s\n",
fdirs->currentduptmp);
goto end;
}
}
logp("Duplicating current backup.\n");
if(recursive_hardlink(sdirs->current,
fdirs->currentduptmp, cconfs)
// The rename race condition is of no consequence here
// because currentdup does not exist.
|| do_rename(fdirs->currentduptmp, fdirs->currentdup))
goto end;
}
}
if(timestamp_read(fdirs->timestamp, tstmp, sizeof(tstmp)))
{
logp("could not read timestamp file: %s\n",
fdirs->timestamp);
goto end;
}
// Get the backup number.
bno=strtoull(tstmp, NULL, 10);
// Determine whether the new backup should be a hardlinked
// archive or not, from the confs and the backup number...
if(need_hardlinked_archive(cconfs, bno))
{
// Create a file to indicate that the previous backup
// does not have others depending on it.
struct fzp *hfp=NULL;
if(!(hfp=fzp_open(fdirs->hlinked, "wb"))) goto end;
// Stick the next backup timestamp in it. It might
// be useful one day when wondering when the next
// backup, now deleted, was made.
fzp_printf(hfp, "%s\n", tstmp);
if(fzp_close(&hfp))
{
logp("error closing hardlinked indication\n");
goto end;
}
}
else
unlink(fdirs->hlinked);
if(atomic_data_jiggle(sdirs, fdirs, hardlinked_current, cconfs))
{
logp("could not finish up backup.\n");
goto end;
}
if(write_status(CNTR_STATUS_SHUFFLING,
"deleting temporary files", get_cntr(cconfs)))
goto end;
// Remove the temporary data directory, we have now removed
// everything useful from it.
recursive_delete(fdirs->datadirtmp);
// Clean up the currentdata directory - this is now the 'old'
// currentdata directory. Any files that were deleted from
// the client will be left in there, so call recursive_delete
// with the option that makes it not delete files.
// This will have the effect of getting rid of unnecessary
// directories.
recursive_delete_dirs_only(fdirs->currentdupdata);
// Rename the old current to something that we know to delete.
if(previous_backup && !hardlinked_current)
{
if(deleteme_move(sdirs,
fdirs->fullrealcurrent, realcurrent, cconfs)
// I have tested that potential race conditions on the
// rename() are automatically recoverable here.
|| do_rename(fdirs->currentdup, fdirs->fullrealcurrent))
goto end;
}
if(deleteme_maybe_delete(cconfs, sdirs))
goto end;
logp("End phase4 (shuffle files)\n");
ret=0;
end:
fdirs_free(&fdirs);
return ret;
}
static int restore_manifest(struct asfd *asfd, struct bu *bu,
regex_t *regex, int srestore, enum action act, struct sdirs *sdirs,
char **dir_for_notify, struct conf **cconfs)
{
int ret=-1;
char *manifest=NULL;
char *logpath=NULL;
char *logpathz=NULL;
// For sending status information up to the server.
enum cntr_status cntr_status=CNTR_STATUS_RESTORING;
if(act==ACTION_RESTORE) cntr_status=CNTR_STATUS_RESTORING;
else if(act==ACTION_VERIFY) cntr_status=CNTR_STATUS_VERIFYING;
if((act==ACTION_RESTORE && get_logpaths(bu, "restorelog",
&logpath, &logpathz))
|| (act==ACTION_VERIFY && get_logpaths(bu, "verifylog",
&logpath, &logpathz))
|| !(manifest=prepend_s(bu->path,
get_protocol(cconfs)==PROTO_1?
"manifest.gz":"manifest")))
{
log_and_send_oom(asfd, __func__);
goto end;
}
if(log_fzp_set(logpath, cconfs))
{
char msg[256]="";
snprintf(msg, sizeof(msg),
"could not open log file: %s", logpath);
log_and_send(asfd, msg);
goto end;
}
*dir_for_notify=strdup_w(bu->path, __func__);
log_restore_settings(cconfs, srestore);
// First, do a pass through the manifest to set up cntr.
// This is the equivalent of a phase1 scan during backup.
if(setup_cntr(asfd, manifest,
regex, srestore, act, cntr_status, cconfs))
goto end;
if(get_int(cconfs[OPT_SEND_CLIENT_CNTR])
&& cntr_send(get_cntr(cconfs)))
goto end;
// Now, do the actual restore.
ret=actual_restore(asfd, bu, manifest,
regex, srestore, act, sdirs, cntr_status, cconfs);
end:
log_fzp_set(NULL, cconfs);
compress_file(logpath, logpathz, get_int(cconfs[OPT_COMPRESSION]));
if(manifest) free(manifest);
if(logpath) free(logpath);
if(logpathz) free(logpathz);
return ret;
}
