scsi/qemu-pr-helper.c

   1 /*
   2  * Privileged helper to handle persistent reservation commands for QEMU
   3  *
   4  * Copyright (C) 2017 Red Hat, Inc. <pbonzini@redhat.com>
   5  *
   6  * Author: Paolo Bonzini <pbonzini@redhat.com>
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License as published by
  10  * the Free Software Foundation; under version 2 of the License.
  11  *
  12  * This program is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  * GNU General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU General Public License
  18  * along with this program; if not, see <http://www.gnu.org/licenses/>.
  19  */
  20
  21 #include "qemu/osdep.h"
  22 #include <getopt.h>
  23 #include <sys/ioctl.h>
  24 #include <linux/dm-ioctl.h>
  25 #include <scsi/sg.h>
  26
  27 #ifdef CONFIG_LIBCAP
  28 #include <cap-ng.h>
  29 #endif
  30 #include <pwd.h>
  31 #include <grp.h>
  32
  33 #ifdef CONFIG_MPATH
  34 #include <libudev.h>
  35 #include <mpath_cmd.h>
  36 #include <mpath_persist.h>
  37 #endif
  38
  39 #include "qapi/error.h"
  40 #include "qemu-common.h"
  41 #include "qemu/cutils.h"
  42 #include "qemu/main-loop.h"
  43 #include "qemu/error-report.h"
  44 #include "qemu/config-file.h"
  45 #include "qemu/bswap.h"
  46 #include "qemu/log.h"
  47 #include "qemu/systemd.h"
  48 #include "qapi/util.h"
  49 #include "qapi/qmp/qstring.h"
  50 #include "io/channel-socket.h"
  51 #include "trace/control.h"
  52 #include "qemu-version.h"
  53
  54 #include "block/aio.h"
  55 #include "block/thread-pool.h"
  56
  57 #include "scsi/constants.h"
  58 #include "scsi/utils.h"
  59 #include "pr-helper.h"
  60
  61 #define PR_OUT_FIXED_PARAM_SIZE 24
  62
  63 static char *socket_path;
  64 static char *pidfile;
  65 static enum { RUNNING, TERMINATE, TERMINATING } state;
  66 static QIOChannelSocket *server_ioc;
  67 static int server_watch;
  68 static int num_active_sockets = 1;
  69 static int noisy;
  70 static int verbose;
  71
  72 #ifdef CONFIG_LIBCAP
  73 static int uid = -1;
  74 static int gid = -1;
  75 #endif
  76
  77 static void compute_default_paths(void)
  78 {
  79     socket_path = qemu_get_local_state_pathname("run/qemu-pr-helper.sock");
  80     pidfile = qemu_get_local_state_pathname("run/qemu-pr-helper.pid");
  81 }
  82
  83 static void usage(const char *name)
  84 {
  85     (printf) (
  86 "Usage: %s [OPTIONS] FILE\n"
  87 "Persistent Reservation helper program for QEMU\n"
  88 "\n"
  89 "  -h, --help                display this help and exit\n"
  90 "  -V, --version             output version information and exit\n"
  91 "\n"
  92 "  -d, --daemon              run in the background\n"
  93 "  -f, --pidfile=PATH        PID file when running as a daemon\n"
  94 "                            (default '%s')\n"
  95 "  -k, --socket=PATH         path to the unix socket\n"
  96 "                            (default '%s')\n"
  97 "  -T, --trace [[enable=]<pattern>][,events=<file>][,file=<file>]\n"
  98 "                            specify tracing options\n"
  99 #ifdef CONFIG_LIBCAP
 100 "  -u, --user=USER           user to drop privileges to\n"
 101 "  -g, --group=GROUP         group to drop privileges to\n"
 102 #endif
 103 "\n"
 104 QEMU_HELP_BOTTOM "\n"
 105     , name, pidfile, socket_path);
 106 }
 107
 108 static void version(const char *name)
 109 {
 110     printf(
 111 "%s " QEMU_FULL_VERSION "\n"
 112 "Written by Paolo Bonzini.\n"
 113 "\n"
 114 QEMU_COPYRIGHT "\n"
 115 "This is free software; see the source for copying conditions.  There is NO\n"
 116 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
 117     , name);
 118 }
 119
 120 static void write_pidfile(void)
 121 {
 122     int pidfd;
 123     char pidstr[32];
 124
 125     pidfd = qemu_open(pidfile, O_CREAT|O_WRONLY, S_IRUSR|S_IWUSR);
 126     if (pidfd == -1) {
 127         error_report("Cannot open pid file, %s", strerror(errno));
 128         exit(EXIT_FAILURE);
 129     }
 130
 131     if (lockf(pidfd, F_TLOCK, 0)) {
 132         error_report("Cannot lock pid file, %s", strerror(errno));
 133         goto fail;
 134     }
 135     if (ftruncate(pidfd, 0)) {
 136         error_report("Failed to truncate pid file");
 137         goto fail;
 138     }
 139
 140     snprintf(pidstr, sizeof(pidstr), "%d\n", getpid());
 141     if (write(pidfd, pidstr, strlen(pidstr)) != strlen(pidstr)) {
 142         error_report("Failed to write pid file");
 143         goto fail;
 144     }
 145     return;
 146
 147 fail:
 148     unlink(pidfile);
 149     close(pidfd);
 150     exit(EXIT_FAILURE);
 151 }
 152
 153 /* SG_IO support */
 154
 155 typedef struct PRHelperSGIOData {
 156     int fd;
 157     const uint8_t *cdb;
 158     uint8_t *sense;
 159     uint8_t *buf;
 160     int sz;              /* input/output */
 161     int dir;
 162 } PRHelperSGIOData;
 163
 164 static int do_sgio_worker(void *opaque)
 165 {
 166     PRHelperSGIOData *data = opaque;
 167     struct sg_io_hdr io_hdr;
 168     int ret;
 169     int status;
 170     SCSISense sense_code;
 171
 172     memset(data->sense, 0, PR_HELPER_SENSE_SIZE);
 173     memset(&io_hdr, 0, sizeof(io_hdr));
 174     io_hdr.interface_id = 'S';
 175     io_hdr.cmd_len = PR_HELPER_CDB_SIZE;
 176     io_hdr.cmdp = (uint8_t *)data->cdb;
 177     io_hdr.sbp = data->sense;
 178     io_hdr.mx_sb_len = PR_HELPER_SENSE_SIZE;
 179     io_hdr.timeout = 1;
 180     io_hdr.dxfer_direction = data->dir;
 181     io_hdr.dxferp = (char *)data->buf;
 182     io_hdr.dxfer_len = data->sz;
 183     ret = ioctl(data->fd, SG_IO, &io_hdr);
 184     status = sg_io_sense_from_errno(ret < 0 ? errno : 0, &io_hdr,
 185                                     &sense_code);
 186     if (status == GOOD) {
 187         data->sz -= io_hdr.resid;
 188     } else {
 189         data->sz = 0;
 190     }
 191
 192     if (status == CHECK_CONDITION &&
 193         !(io_hdr.driver_status & SG_ERR_DRIVER_SENSE)) {
 194         scsi_build_sense(data->sense, sense_code);
 195     }
 196
 197     return status;
 198 }
 199
 200 static int do_sgio(int fd, const uint8_t *cdb, uint8_t *sense,
 201                     uint8_t *buf, int *sz, int dir)
 202 {
 203     ThreadPool *pool = aio_get_thread_pool(qemu_get_aio_context());
 204     int r;
 205
 206     PRHelperSGIOData data = {
 207         .fd = fd,
 208         .cdb = cdb,
 209         .sense = sense,
 210         .buf = buf,
 211         .sz = *sz,
 212         .dir = dir,
 213     };
 214
 215     r = thread_pool_submit_co(pool, do_sgio_worker, &data);
 216     *sz = data.sz;
 217     return r;
 218 }
 219
 220 /* Device mapper interface */
 221
 222 #ifdef CONFIG_MPATH
 223 #define CONTROL_PATH "/dev/mapper/control"
 224
 225 typedef struct DMData {
 226     struct dm_ioctl dm;
 227     uint8_t data[1024];
 228 } DMData;
 229
 230 static int control_fd;
 231
 232 static void *dm_ioctl(int ioc, struct dm_ioctl *dm)
 233 {
 234     static DMData d;
 235     memcpy(&d.dm, dm, sizeof(d.dm));
 236     QEMU_BUILD_BUG_ON(sizeof(d.data) < sizeof(struct dm_target_spec));
 237
 238     d.dm.version[0] = DM_VERSION_MAJOR;
 239     d.dm.version[1] = 0;
 240     d.dm.version[2] = 0;
 241     d.dm.data_size = 1024;
 242     d.dm.data_start = offsetof(DMData, data);
 243     if (ioctl(control_fd, ioc, &d) < 0) {
 244         return NULL;
 245     }
 246     memcpy(dm, &d.dm, sizeof(d.dm));
 247     return &d.data;
 248 }
 249
 250 static void *dm_dev_ioctl(int fd, int ioc, struct dm_ioctl *dm)
 251 {
 252     struct stat st;
 253     int r;
 254
 255     r = fstat(fd, &st);
 256     if (r < 0) {
 257         perror("fstat");
 258         exit(1);
 259     }
 260
 261     dm->dev = st.st_rdev;
 262     return dm_ioctl(ioc, dm);
 263 }
 264
 265 static void dm_init(void)
 266 {
 267     control_fd = open(CONTROL_PATH, O_RDWR);
 268     if (control_fd < 0) {
 269         perror("Cannot open " CONTROL_PATH);
 270         exit(1);
 271     }
 272     struct dm_ioctl dm = { 0 };
 273     if (!dm_ioctl(DM_VERSION, &dm)) {
 274         perror("ioctl");
 275         exit(1);
 276     }
 277     if (dm.version[0] != DM_VERSION_MAJOR) {
 278         fprintf(stderr, "Unsupported device mapper interface");
 279         exit(1);
 280     }
 281 }
 282
 283 /* Variables required by libmultipath and libmpathpersist.  */
 284 QEMU_BUILD_BUG_ON(PR_HELPER_DATA_SIZE > MPATH_MAX_PARAM_LEN);
 285 static struct config *multipath_conf;
 286 unsigned mpath_mx_alloc_len = PR_HELPER_DATA_SIZE;
 287 int logsink;
 288 struct udev *udev;
 289
 290 extern struct config *get_multipath_config(void);
 291 struct config *get_multipath_config(void)
 292 {
 293     return multipath_conf;
 294 }
 295
 296 extern void put_multipath_config(struct config *conf);
 297 void put_multipath_config(struct config *conf)
 298 {
 299 }
 300
 301 static void multipath_pr_init(void)
 302 {
 303     udev = udev_new();
 304     multipath_conf = mpath_lib_init();
 305 }
 306
 307 static int is_mpath(int fd)
 308 {
 309     struct dm_ioctl dm = { .flags = DM_NOFLUSH_FLAG };
 310     struct dm_target_spec *tgt;
 311
 312     tgt = dm_dev_ioctl(fd, DM_TABLE_STATUS, &dm);
 313     if (!tgt) {
 314         if (errno == ENXIO) {
 315             return 0;
 316         }
 317         perror("ioctl");
 318         exit(EXIT_FAILURE);
 319     }
 320     return !strncmp(tgt->target_type, "multipath", DM_MAX_TYPE_NAME);
 321 }
 322
 323 static SCSISense mpath_generic_sense(int r)
 324 {
 325     switch (r) {
 326     case MPATH_PR_SENSE_NOT_READY:
 327          return SENSE_CODE(NOT_READY);
 328     case MPATH_PR_SENSE_MEDIUM_ERROR:
 329          return SENSE_CODE(READ_ERROR);
 330     case MPATH_PR_SENSE_HARDWARE_ERROR:
 331          return SENSE_CODE(TARGET_FAILURE);
 332     case MPATH_PR_SENSE_ABORTED_COMMAND:
 333          return SENSE_CODE(IO_ERROR);
 334     default:
 335          abort();
 336     }
 337 }
 338
 339 static int mpath_reconstruct_sense(int fd, int r, uint8_t *sense)
 340 {
 341     switch (r) {
 342     case MPATH_PR_SUCCESS:
 343         return GOOD;
 344     case MPATH_PR_SENSE_NOT_READY:
 345     case MPATH_PR_SENSE_MEDIUM_ERROR:
 346     case MPATH_PR_SENSE_HARDWARE_ERROR:
 347     case MPATH_PR_SENSE_ABORTED_COMMAND:
 348         {
 349             /* libmpathpersist ate the exact sense.  Try to find it by
 350              * issuing TEST UNIT READY.
 351              */
 352             uint8_t cdb[6] = { TEST_UNIT_READY };
 353             int sz = 0;
 354             int r = do_sgio(fd, cdb, sense, NULL, &sz, SG_DXFER_NONE);
 355
 356             if (r != GOOD) {
 357                 return r;
 358             }
 359             scsi_build_sense(sense, mpath_generic_sense(r));
 360             return CHECK_CONDITION;
 361         }
 362
 363     case MPATH_PR_SENSE_UNIT_ATTENTION:
 364         /* Congratulations libmpathpersist, you ruined the Unit Attention...
 365          * Return a heavyweight one.
 366          */
 367         scsi_build_sense(sense, SENSE_CODE(SCSI_BUS_RESET));
 368         return CHECK_CONDITION;
 369     case MPATH_PR_SENSE_INVALID_OP:
 370         /* Only one valid sense.  */
 371         scsi_build_sense(sense, SENSE_CODE(INVALID_OPCODE));
 372         return CHECK_CONDITION;
 373     case MPATH_PR_ILLEGAL_REQ:
 374         /* Guess.  */
 375         scsi_build_sense(sense, SENSE_CODE(INVALID_PARAM));
 376         return CHECK_CONDITION;
 377     case MPATH_PR_NO_SENSE:
 378         scsi_build_sense(sense, SENSE_CODE(NO_SENSE));
 379         return CHECK_CONDITION;
 380
 381     case MPATH_PR_RESERV_CONFLICT:
 382         return RESERVATION_CONFLICT;
 383
 384     case MPATH_PR_OTHER:
 385     default:
 386         scsi_build_sense(sense, SENSE_CODE(LUN_COMM_FAILURE));
 387         return CHECK_CONDITION;
 388     }
 389 }
 390
 391 static int multipath_pr_in(int fd, const uint8_t *cdb, uint8_t *sense,
 392                            uint8_t *data, int sz)
 393 {
 394     int rq_servact = cdb[1];
 395     struct prin_resp resp;
 396     size_t written;
 397     int r;
 398
 399     switch (rq_servact) {
 400     case MPATH_PRIN_RKEY_SA:
 401     case MPATH_PRIN_RRES_SA:
 402     case MPATH_PRIN_RCAP_SA:
 403         break;
 404     case MPATH_PRIN_RFSTAT_SA:
 405         /* Nobody implements it anyway, so bail out. */
 406     default:
 407         /* Cannot parse any other output.  */
 408         scsi_build_sense(sense, SENSE_CODE(INVALID_FIELD));
 409         return CHECK_CONDITION;
 410     }
 411
 412     r = mpath_persistent_reserve_in(fd, rq_servact, &resp, noisy, verbose);
 413     if (r == MPATH_PR_SUCCESS) {
 414         switch (rq_servact) {
 415         case MPATH_PRIN_RKEY_SA:
 416         case MPATH_PRIN_RRES_SA: {
 417             struct prin_readdescr *out = &resp.prin_descriptor.prin_readkeys;
 418             assert(sz >= 8);
 419             written = MIN(out->additional_length + 8, sz);
 420             stl_be_p(&data[0], out->prgeneration);
 421             stl_be_p(&data[4], out->additional_length);
 422             memcpy(&data[8], out->key_list, written - 8);
 423             break;
 424         }
 425         case MPATH_PRIN_RCAP_SA: {
 426             struct prin_capdescr *out = &resp.prin_descriptor.prin_readcap;
 427             assert(sz >= 6);
 428             written = 6;
 429             stw_be_p(&data[0], out->length);
 430             data[2] = out->flags[0];
 431             data[3] = out->flags[1];
 432             stw_be_p(&data[4], out->pr_type_mask);
 433             break;
 434         }
 435         default:
 436             scsi_build_sense(sense, SENSE_CODE(INVALID_OPCODE));
 437             return CHECK_CONDITION;
 438         }
 439         assert(written <= sz);
 440         memset(data + written, 0, sz - written);
 441     }
 442
 443     return mpath_reconstruct_sense(fd, r, sense);
 444 }
 445
 446 static int multipath_pr_out(int fd, const uint8_t *cdb, uint8_t *sense,
 447                             const uint8_t *param, int sz)
 448 {
 449     int rq_servact = cdb[1];
 450     int rq_scope = cdb[2] >> 4;
 451     int rq_type = cdb[2] & 0xf;
 452     struct prout_param_descriptor paramp;
 453     char transportids[PR_HELPER_DATA_SIZE];
 454     int r;
 455
 456     if (sz < PR_OUT_FIXED_PARAM_SIZE) {
 457         /* Illegal request, Parameter list length error.  This isn't fatal;
 458          * we have read the data, send an error without closing the socket.
 459          */
 460         scsi_build_sense(sense, SENSE_CODE(INVALID_PARAM_LEN));
 461         return CHECK_CONDITION;
 462     }
 463
 464     switch (rq_servact) {
 465     case MPATH_PROUT_REG_SA:
 466     case MPATH_PROUT_RES_SA:
 467     case MPATH_PROUT_REL_SA:
 468     case MPATH_PROUT_CLEAR_SA:
 469     case MPATH_PROUT_PREE_SA:
 470     case MPATH_PROUT_PREE_AB_SA:
 471     case MPATH_PROUT_REG_IGN_SA:
 472         break;
 473     case MPATH_PROUT_REG_MOV_SA:
 474         /* Not supported by struct prout_param_descriptor.  */
 475     default:
 476         /* Cannot parse any other input.  */
 477         scsi_build_sense(sense, SENSE_CODE(INVALID_FIELD));
 478         return CHECK_CONDITION;
 479     }
 480
 481     /* Convert input data, especially transport IDs, to the structs
 482      * used by libmpathpersist (which, of course, will immediately
 483      * do the opposite).
 484      */
 485     memset(&paramp, 0, sizeof(paramp));
 486     memcpy(&paramp.key, &param[0], 8);
 487     memcpy(&paramp.sa_key, &param[8], 8);
 488     paramp.sa_flags = param[20];
 489     if (sz > PR_OUT_FIXED_PARAM_SIZE) {
 490         size_t transportid_len;
 491         int i, j;
 492         if (sz < PR_OUT_FIXED_PARAM_SIZE + 4) {
 493             scsi_build_sense(sense, SENSE_CODE(INVALID_PARAM_LEN));
 494             return CHECK_CONDITION;
 495         }
 496         transportid_len = ldl_be_p(&param[24]) + PR_OUT_FIXED_PARAM_SIZE + 4;
 497         if (transportid_len > sz) {
 498             scsi_build_sense(sense, SENSE_CODE(INVALID_PARAM));
 499             return CHECK_CONDITION;
 500         }
 501         for (i = PR_OUT_FIXED_PARAM_SIZE + 4, j = 0; i < transportid_len; ) {
 502             struct transportid *id = (struct transportid *) &transportids[j];
 503             int len;
 504
 505             id->format_code = param[i] & 0xc0;
 506             id->protocol_id = param[i] & 0x0f;
 507             switch (param[i] & 0xcf) {
 508             case 0:
 509                 /* FC transport.  */
 510                 if (i + 24 > transportid_len) {
 511                     goto illegal_req;
 512                 }
 513                 memcpy(id->n_port_name, &param[i + 8], 8);
 514                 j += offsetof(struct transportid, n_port_name[8]);
 515                 i += 24;
 516                 break;
 517             case 5:
 518             case 0x45:
 519                 /* iSCSI transport.  */
 520                 len = lduw_be_p(&param[i + 2]);
 521                 if (len > 252 || (len & 3) || i + len + 4 > transportid_len) {
 522                     /* For format code 00, the standard says the maximum is 223
 523                      * plus the NUL terminator.  For format code 01 there is no
 524                      * maximum length, but libmpathpersist ignores the first
 525                      * byte of id->iscsi_name so our maximum is 252.
 526                      */
 527                     goto illegal_req;
 528                 }
 529                 if (memchr(&param[i + 4], 0, len) == NULL) {
 530                     goto illegal_req;
 531                 }
 532                 memcpy(id->iscsi_name, &param[i + 2], len + 2);
 533                 j += offsetof(struct transportid, iscsi_name[len + 2]);
 534                 i += len + 4;
 535                 break;
 536             case 6:
 537                 /* SAS transport.  */
 538                 if (i + 24 > transportid_len) {
 539                     goto illegal_req;
 540                 }
 541                 memcpy(id->sas_address, &param[i + 4], 8);
 542                 j += offsetof(struct transportid, sas_address[8]);
 543                 i += 24;
 544                 break;
 545             default:
 546             illegal_req:
 547                 scsi_build_sense(sense, SENSE_CODE(INVALID_PARAM));
 548                 return CHECK_CONDITION;
 549             }
 550
 551             paramp.trnptid_list[paramp.num_transportid++] = id;
 552         }
 553     }
 554
 555     r = mpath_persistent_reserve_out(fd, rq_servact, rq_scope, rq_type,
 556                                      &paramp, noisy, verbose);
 557     return mpath_reconstruct_sense(fd, r, sense);
 558 }
 559 #endif
 560
 561 static int do_pr_in(int fd, const uint8_t *cdb, uint8_t *sense,
 562                     uint8_t *data, int *resp_sz)
 563 {
 564 #ifdef CONFIG_MPATH
 565     if (is_mpath(fd)) {
 566         /* multipath_pr_in fills the whole input buffer.  */
 567         int r = multipath_pr_in(fd, cdb, sense, data, *resp_sz);
 568         if (r != GOOD) {
 569             *resp_sz = 0;
 570         }
 571         return r;
 572     }
 573 #endif
 574
 575     return do_sgio(fd, cdb, sense, data, resp_sz,
 576                    SG_DXFER_FROM_DEV);
 577 }
 578
 579 static int do_pr_out(int fd, const uint8_t *cdb, uint8_t *sense,
 580                      const uint8_t *param, int sz)
 581 {
 582     int resp_sz;
 583
 584     if ((fcntl(fd, F_GETFL) & O_ACCMODE) == O_RDONLY) {
 585         scsi_build_sense(sense, SENSE_CODE(INVALID_OPCODE));
 586         return CHECK_CONDITION;
 587     }
 588
 589 #ifdef CONFIG_MPATH
 590     if (is_mpath(fd)) {
 591         return multipath_pr_out(fd, cdb, sense, param, sz);
 592     }
 593 #endif
 594
 595     resp_sz = sz;
 596     return do_sgio(fd, cdb, sense, (uint8_t *)param, &resp_sz,
 597                    SG_DXFER_TO_DEV);
 598 }
 599
 600 /* Client */
 601
 602 typedef struct PRHelperClient {
 603     QIOChannelSocket *ioc;
 604     Coroutine *co;
 605     int fd;
 606     uint8_t data[PR_HELPER_DATA_SIZE];
 607 } PRHelperClient;
 608
 609 typedef struct PRHelperRequest {
 610     int fd;
 611     size_t sz;
 612     uint8_t cdb[PR_HELPER_CDB_SIZE];
 613 } PRHelperRequest;
 614
 615 static int coroutine_fn prh_read(PRHelperClient *client, void *buf, int sz,
 616                                  Error **errp)
 617 {
 618     int ret = 0;
 619
 620     while (sz > 0) {
 621         int *fds = NULL;
 622         size_t nfds = 0;
 623         int i;
 624         struct iovec iov;
 625         ssize_t n_read;
 626
 627         iov.iov_base = buf;
 628         iov.iov_len = sz;
 629         n_read = qio_channel_readv_full(QIO_CHANNEL(client->ioc), &iov, 1,
 630                                         &fds, &nfds, errp);
 631
 632         if (n_read == QIO_CHANNEL_ERR_BLOCK) {
 633             qio_channel_yield(QIO_CHANNEL(client->ioc), G_IO_IN);
 634             continue;
 635         }
 636         if (n_read <= 0) {
 637             ret = n_read ? n_read : -1;
 638             goto err;
 639         }
 640
 641         /* Stash one file descriptor per request.  */
 642         if (nfds) {
 643             bool too_many = false;
 644             for (i = 0; i < nfds; i++) {
 645                 if (client->fd == -1) {
 646                     client->fd = fds[i];
 647                 } else {
 648                     close(fds[i]);
 649                     too_many = true;
 650                 }
 651             }
 652             g_free(fds);
 653             if (too_many) {
 654                 ret = -1;
 655                 goto err;
 656             }
 657         }
 658
 659         buf += n_read;
 660         sz -= n_read;
 661     }
 662
 663     return 0;
 664
 665 err:
 666     if (client->fd != -1) {
 667         close(client->fd);
 668         client->fd = -1;
 669     }
 670     return ret;
 671 }
 672
 673 static int coroutine_fn prh_read_request(PRHelperClient *client,
 674                                          PRHelperRequest *req,
 675                                          PRHelperResponse *resp, Error **errp)
 676 {
 677     uint32_t sz;
 678
 679     if (prh_read(client, req->cdb, sizeof(req->cdb), NULL) < 0) {
 680         return -1;
 681     }
 682
 683     if (client->fd == -1) {
 684         error_setg(errp, "No file descriptor in request.");
 685         return -1;
 686     }
 687
 688     if (req->cdb[0] != PERSISTENT_RESERVE_OUT &&
 689         req->cdb[0] != PERSISTENT_RESERVE_IN) {
 690         error_setg(errp, "Invalid CDB, closing socket.");
 691         goto out_close;
 692     }
 693
 694     sz = scsi_cdb_xfer(req->cdb);
 695     if (sz > sizeof(client->data)) {
 696         goto out_close;
 697     }
 698
 699     if (req->cdb[0] == PERSISTENT_RESERVE_OUT) {
 700         if (qio_channel_read_all(QIO_CHANNEL(client->ioc),
 701                                  (char *)client->data, sz,
 702                                  errp) < 0) {
 703             goto out_close;
 704         }
 705     }
 706
 707     req->fd = client->fd;
 708     req->sz = sz;
 709     client->fd = -1;
 710     return sz;
 711
 712 out_close:
 713     close(client->fd);
 714     client->fd = -1;
 715     return -1;
 716 }
 717
 718 static int coroutine_fn prh_write_response(PRHelperClient *client,
 719                                            PRHelperRequest *req,
 720                                            PRHelperResponse *resp, Error **errp)
 721 {
 722     ssize_t r;
 723     size_t sz;
 724
 725     if (req->cdb[0] == PERSISTENT_RESERVE_IN && resp->result == GOOD) {
 726         assert(resp->sz <= req->sz && resp->sz <= sizeof(client->data));
 727     } else {
 728         assert(resp->sz == 0);
 729     }
 730
 731     sz = resp->sz;
 732
 733     resp->result = cpu_to_be32(resp->result);
 734     resp->sz = cpu_to_be32(resp->sz);
 735     r = qio_channel_write_all(QIO_CHANNEL(client->ioc),
 736                               (char *) resp, sizeof(*resp), errp);
 737     if (r < 0) {
 738         return r;
 739     }
 740
 741     r = qio_channel_write_all(QIO_CHANNEL(client->ioc),
 742                               (char *) client->data,
 743                               sz, errp);
 744     return r < 0 ? r : 0;
 745 }
 746
 747 static void coroutine_fn prh_co_entry(void *opaque)
 748 {
 749     PRHelperClient *client = opaque;
 750     Error *local_err = NULL;
 751     uint32_t flags;
 752     int r;
 753
 754     qio_channel_set_blocking(QIO_CHANNEL(client->ioc),
 755                              false, NULL);
 756     qio_channel_attach_aio_context(QIO_CHANNEL(client->ioc),
 757                                    qemu_get_aio_context());
 758
 759     /* A very simple negotiation for future extensibility.  No features
 760      * are defined so write 0.
 761      */
 762     flags = cpu_to_be32(0);
 763     r = qio_channel_write_all(QIO_CHANNEL(client->ioc),
 764                              (char *) &flags, sizeof(flags), NULL);
 765     if (r < 0) {
 766         goto out;
 767     }
 768
 769     r = qio_channel_read_all(QIO_CHANNEL(client->ioc),
 770                              (char *) &flags, sizeof(flags), NULL);
 771     if (be32_to_cpu(flags) != 0 || r < 0) {
 772         goto out;
 773     }
 774
 775     while (atomic_read(&state) == RUNNING) {
 776         PRHelperRequest req;
 777         PRHelperResponse resp;
 778         int sz;
 779
 780         sz = prh_read_request(client, &req, &resp, &local_err);
 781         if (sz < 0) {
 782             break;
 783         }
 784
 785         num_active_sockets++;
 786         if (req.cdb[0] == PERSISTENT_RESERVE_OUT) {
 787             r = do_pr_out(req.fd, req.cdb, resp.sense,
 788                           client->data, sz);
 789             resp.sz = 0;
 790         } else {
 791             resp.sz = sizeof(client->data);
 792             r = do_pr_in(req.fd, req.cdb, resp.sense,
 793                          client->data, &resp.sz);
 794             resp.sz = MIN(resp.sz, sz);
 795         }
 796         num_active_sockets--;
 797         close(req.fd);
 798         if (r == -1) {
 799             break;
 800         }
 801         resp.result = r;
 802
 803         if (prh_write_response(client, &req, &resp, &local_err) < 0) {
 804             break;
 805         }
 806     }
 807
 808     if (local_err) {
 809         if (verbose == 0) {
 810             error_free(local_err);
 811         } else {
 812             error_report_err(local_err);
 813         }
 814     }
 815
 816 out:
 817     qio_channel_detach_aio_context(QIO_CHANNEL(client->ioc));
 818     object_unref(OBJECT(client->ioc));
 819     g_free(client);
 820 }
 821
 822 static gboolean accept_client(QIOChannel *ioc, GIOCondition cond, gpointer opaque)
 823 {
 824     QIOChannelSocket *cioc;
 825     PRHelperClient *prh;
 826
 827     cioc = qio_channel_socket_accept(QIO_CHANNEL_SOCKET(ioc),
 828                                      NULL);
 829     if (!cioc) {
 830         return TRUE;
 831     }
 832
 833     prh = g_new(PRHelperClient, 1);
 834     prh->ioc = cioc;
 835     prh->fd = -1;
 836     prh->co = qemu_coroutine_create(prh_co_entry, prh);
 837     qemu_coroutine_enter(prh->co);
 838
 839     return TRUE;
 840 }
 841
 842 static void termsig_handler(int signum)
 843 {
 844     atomic_cmpxchg(&state, RUNNING, TERMINATE);
 845     qemu_notify_event();
 846 }
 847
 848 static void close_server_socket(void)
 849 {
 850     assert(server_ioc);
 851
 852     g_source_remove(server_watch);
 853     server_watch = -1;
 854     object_unref(OBJECT(server_ioc));
 855     num_active_sockets--;
 856 }
 857
 858 #ifdef CONFIG_LIBCAP
 859 static int drop_privileges(void)
 860 {
 861     /* clear all capabilities */
 862     capng_clear(CAPNG_SELECT_BOTH);
 863
 864     if (capng_update(CAPNG_ADD, CAPNG_EFFECTIVE | CAPNG_PERMITTED,
 865                      CAP_SYS_RAWIO) < 0) {
 866         return -1;
 867     }
 868
 869 #ifdef CONFIG_MPATH
 870     /* For /dev/mapper/control ioctls */
 871     if (capng_update(CAPNG_ADD, CAPNG_EFFECTIVE | CAPNG_PERMITTED,
 872                      CAP_SYS_ADMIN) < 0) {
 873         return -1;
 874     }
 875 #endif
 876
 877     /* Change user/group id, retaining the capabilities.  Because file descriptors
 878      * are passed via SCM_RIGHTS, we don't need supplementary groups (and in
 879      * fact the helper can run as "nobody").
 880      */
 881     if (capng_change_id(uid != -1 ? uid : getuid(),
 882                         gid != -1 ? gid : getgid(),
 883                         CAPNG_DROP_SUPP_GRP | CAPNG_CLEAR_BOUNDING)) {
 884         return -1;
 885     }
 886
 887     return 0;
 888 }
 889 #endif
 890
 891 int main(int argc, char **argv)
 892 {
 893     const char *sopt = "hVk:f:dT:u:g:vq";
 894     struct option lopt[] = {
 895         { "help", no_argument, NULL, 'h' },
 896         { "version", no_argument, NULL, 'V' },
 897         { "socket", required_argument, NULL, 'k' },
 898         { "pidfile", required_argument, NULL, 'f' },
 899         { "daemon", no_argument, NULL, 'd' },
 900         { "trace", required_argument, NULL, 'T' },
 901         { "user", required_argument, NULL, 'u' },
 902         { "group", required_argument, NULL, 'g' },
 903         { "verbose", no_argument, NULL, 'v' },
 904         { "quiet", no_argument, NULL, 'q' },
 905         { NULL, 0, NULL, 0 }
 906     };
 907     int opt_ind = 0;
 908     int loglevel = 1;
 909     int quiet = 0;
 910     int ch;
 911     Error *local_err = NULL;
 912     char *trace_file = NULL;
 913     bool daemonize = false;
 914     bool pidfile_specified = false;
 915     bool socket_path_specified = false;
 916     unsigned socket_activation;
 917
 918     struct sigaction sa_sigterm;
 919     memset(&sa_sigterm, 0, sizeof(sa_sigterm));
 920     sa_sigterm.sa_handler = termsig_handler;
 921     sigaction(SIGTERM, &sa_sigterm, NULL);
 922     sigaction(SIGINT, &sa_sigterm, NULL);
 923     sigaction(SIGHUP, &sa_sigterm, NULL);
 924
 925     signal(SIGPIPE, SIG_IGN);
 926
 927     module_call_init(MODULE_INIT_TRACE);
 928     module_call_init(MODULE_INIT_QOM);
 929     qemu_add_opts(&qemu_trace_opts);
 930     qemu_init_exec_dir(argv[0]);
 931
 932     compute_default_paths();
 933
 934     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 935         switch (ch) {
 936         case 'k':
 937             g_free(socket_path);
 938             socket_path = g_strdup(optarg);
 939             socket_path_specified = true;
 940             if (socket_path[0] != '/') {
 941                 error_report("socket path must be absolute");
 942                 exit(EXIT_FAILURE);
 943             }
 944             break;
 945         case 'f':
 946             g_free(pidfile);
 947             pidfile = g_strdup(optarg);
 948             pidfile_specified = true;
 949             break;
 950 #ifdef CONFIG_LIBCAP
 951         case 'u': {
 952             unsigned long res;
 953             struct passwd *userinfo = getpwnam(optarg);
 954             if (userinfo) {
 955                 uid = userinfo->pw_uid;
 956             } else if (qemu_strtoul(optarg, NULL, 10, &res) == 0 &&
 957                        (uid_t)res == res) {
 958                 uid = res;
 959             } else {
 960                 error_report("invalid user '%s'", optarg);
 961                 exit(EXIT_FAILURE);
 962             }
 963             break;
 964         }
 965         case 'g': {
 966             unsigned long res;
 967             struct group *groupinfo = getgrnam(optarg);
 968             if (groupinfo) {
 969                 gid = groupinfo->gr_gid;
 970             } else if (qemu_strtoul(optarg, NULL, 10, &res) == 0 &&
 971                        (gid_t)res == res) {
 972                 gid = res;
 973             } else {
 974                 error_report("invalid group '%s'", optarg);
 975                 exit(EXIT_FAILURE);
 976             }
 977             break;
 978         }
 979 #else
 980         case 'u':
 981         case 'g':
 982             error_report("-%c not supported by this %s", ch, argv[0]);
 983             exit(1);
 984 #endif
 985         case 'd':
 986             daemonize = true;
 987             break;
 988         case 'q':
 989             quiet = 1;
 990             break;
 991         case 'v':
 992             ++loglevel;
 993             break;
 994         case 'T':
 995             g_free(trace_file);
 996             trace_file = trace_opt_parse(optarg);
 997             break;
 998         case 'V':
 999             version(argv[0]);
1000             exit(EXIT_SUCCESS);
1001             break;
1002         case 'h':
1003             usage(argv[0]);
1004             exit(EXIT_SUCCESS);
1005             break;
1006         case '?':
1007             error_report("Try `%s --help' for more information.", argv[0]);
1008             exit(EXIT_FAILURE);
1009         }
1010     }
1011
1012     /* set verbosity */
1013     noisy = !quiet && (loglevel >= 3);
1014     verbose = quiet ? 0 : MIN(loglevel, 3);
1015
1016     if (!trace_init_backends()) {
1017         exit(EXIT_FAILURE);
1018     }
1019     trace_init_file(trace_file);
1020     qemu_set_log(LOG_TRACE);
1021
1022 #ifdef CONFIG_MPATH
1023     dm_init();
1024     multipath_pr_init();
1025 #endif
1026
1027     socket_activation = check_socket_activation();
1028     if (socket_activation == 0) {
1029         SocketAddress saddr;
1030         saddr = (SocketAddress){
1031             .type = SOCKET_ADDRESS_TYPE_UNIX,
1032             .u.q_unix.path = socket_path,
1033         };
1034         server_ioc = qio_channel_socket_new();
1035         if (qio_channel_socket_listen_sync(server_ioc, &saddr, &local_err) < 0) {
1036             object_unref(OBJECT(server_ioc));
1037             error_report_err(local_err);
1038             return 1;
1039         }
1040     } else {
1041         /* Using socket activation - check user didn't use -p etc. */
1042         if (socket_path_specified) {
1043             error_report("Unix socket can't be set when using socket activation");
1044             exit(EXIT_FAILURE);
1045         }
1046
1047         /* Can only listen on a single socket.  */
1048         if (socket_activation > 1) {
1049             error_report("%s does not support socket activation with LISTEN_FDS > 1",
1050                          argv[0]);
1051             exit(EXIT_FAILURE);
1052         }
1053         server_ioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD,
1054                                                &local_err);
1055         if (server_ioc == NULL) {
1056             error_report("Failed to use socket activation: %s",
1057                          error_get_pretty(local_err));
1058             exit(EXIT_FAILURE);
1059         }
1060     }
1061
1062     if (qemu_init_main_loop(&local_err)) {
1063         error_report_err(local_err);
1064         exit(EXIT_FAILURE);
1065     }
1066
1067     server_watch = qio_channel_add_watch(QIO_CHANNEL(server_ioc),
1068                                          G_IO_IN,
1069                                          accept_client,
1070                                          NULL, NULL);
1071
1072     if (daemonize) {
1073         if (daemon(0, 0) < 0) {
1074             error_report("Failed to daemonize: %s", strerror(errno));
1075             exit(EXIT_FAILURE);
1076         }
1077     }
1078
1079     if (daemonize || pidfile_specified)
1080         write_pidfile();
1081
1082 #ifdef CONFIG_LIBCAP
1083     if (drop_privileges() < 0) {
1084         error_report("Failed to drop privileges: %s", strerror(errno));
1085         exit(EXIT_FAILURE);
1086     }
1087 #endif
1088
1089     state = RUNNING;
1090     do {
1091         main_loop_wait(false);
1092         if (state == TERMINATE) {
1093             state = TERMINATING;
1094             close_server_socket();
1095         }
1096     } while (num_active_sockets > 0);
1097
1098     exit(EXIT_SUCCESS);
1099 }