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     if (!socket_path) {
  80         socket_path = qemu_get_local_state_pathname("run/qemu-pr-helper.sock");
  81     }
  82 }
  83
  84 static void usage(const char *name)
  85 {
  86     compute_default_paths();
  87     (printf) (
  88 "Usage: %s [OPTIONS] FILE\n"
  89 "Persistent Reservation helper program for QEMU\n"
  90 "\n"
  91 "  -h, --help                display this help and exit\n"
  92 "  -V, --version             output version information and exit\n"
  93 "\n"
  94 "  -d, --daemon              run in the background\n"
  95 "  -f, --pidfile=PATH        PID file when running as a daemon\n"
  96 "                            (default '%s')\n"
  97 "  -k, --socket=PATH         path to the unix socket\n"
  98 "                            (default '%s')\n"
  99 "  -T, --trace [[enable=]<pattern>][,events=<file>][,file=<file>]\n"
 100 "                            specify tracing options\n"
 101 #ifdef CONFIG_LIBCAP
 102 "  -u, --user=USER           user to drop privileges to\n"
 103 "  -g, --group=GROUP         group to drop privileges to\n"
 104 #endif
 105 "\n"
 106 QEMU_HELP_BOTTOM "\n"
 107     , name, pidfile, socket_path);
 108 }
 109
 110 static void version(const char *name)
 111 {
 112     printf(
 113 "%s " QEMU_FULL_VERSION "\n"
 114 "Written by Paolo Bonzini.\n"
 115 "\n"
 116 QEMU_COPYRIGHT "\n"
 117 "This is free software; see the source for copying conditions.  There is NO\n"
 118 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
 119     , name);
 120 }
 121
 122 static void write_pidfile(void)
 123 {
 124     int pidfd;
 125     char pidstr[32];
 126
 127     pidfd = qemu_open(pidfile, O_CREAT|O_WRONLY, S_IRUSR|S_IWUSR);
 128     if (pidfd == -1) {
 129         error_report("Cannot open pid file, %s", strerror(errno));
 130         exit(EXIT_FAILURE);
 131     }
 132
 133     if (lockf(pidfd, F_TLOCK, 0)) {
 134         error_report("Cannot lock pid file, %s", strerror(errno));
 135         goto fail;
 136     }
 137     if (ftruncate(pidfd, 0)) {
 138         error_report("Failed to truncate pid file");
 139         goto fail;
 140     }
 141
 142     snprintf(pidstr, sizeof(pidstr), "%d\n", getpid());
 143     if (write(pidfd, pidstr, strlen(pidstr)) != strlen(pidstr)) {
 144         error_report("Failed to write pid file");
 145         goto fail;
 146     }
 147     return;
 148
 149 fail:
 150     unlink(pidfile);
 151     close(pidfd);
 152     exit(EXIT_FAILURE);
 153 }
 154
 155 /* SG_IO support */
 156
 157 typedef struct PRHelperSGIOData {
 158     int fd;
 159     const uint8_t *cdb;
 160     uint8_t *sense;
 161     uint8_t *buf;
 162     int sz;              /* input/output */
 163     int dir;
 164 } PRHelperSGIOData;
 165
 166 static int do_sgio_worker(void *opaque)
 167 {
 168     PRHelperSGIOData *data = opaque;
 169     struct sg_io_hdr io_hdr;
 170     int ret;
 171     int status;
 172     SCSISense sense_code;
 173
 174     memset(data->sense, 0, PR_HELPER_SENSE_SIZE);
 175     memset(&io_hdr, 0, sizeof(io_hdr));
 176     io_hdr.interface_id = 'S';
 177     io_hdr.cmd_len = PR_HELPER_CDB_SIZE;
 178     io_hdr.cmdp = (uint8_t *)data->cdb;
 179     io_hdr.sbp = data->sense;
 180     io_hdr.mx_sb_len = PR_HELPER_SENSE_SIZE;
 181     io_hdr.timeout = 1;
 182     io_hdr.dxfer_direction = data->dir;
 183     io_hdr.dxferp = (char *)data->buf;
 184     io_hdr.dxfer_len = data->sz;
 185     ret = ioctl(data->fd, SG_IO, &io_hdr);
 186     status = sg_io_sense_from_errno(ret < 0 ? errno : 0, &io_hdr,
 187                                     &sense_code);
 188     if (status == GOOD) {
 189         data->sz -= io_hdr.resid;
 190     } else {
 191         data->sz = 0;
 192     }
 193
 194     if (status == CHECK_CONDITION &&
 195         !(io_hdr.driver_status & SG_ERR_DRIVER_SENSE)) {
 196         scsi_build_sense(data->sense, sense_code);
 197     }
 198
 199     return status;
 200 }
 201
 202 static int do_sgio(int fd, const uint8_t *cdb, uint8_t *sense,
 203                     uint8_t *buf, int *sz, int dir)
 204 {
 205     ThreadPool *pool = aio_get_thread_pool(qemu_get_aio_context());
 206     int r;
 207
 208     PRHelperSGIOData data = {
 209         .fd = fd,
 210         .cdb = cdb,
 211         .sense = sense,
 212         .buf = buf,
 213         .sz = *sz,
 214         .dir = dir,
 215     };
 216
 217     r = thread_pool_submit_co(pool, do_sgio_worker, &data);
 218     *sz = data.sz;
 219     return r;
 220 }
 221
 222 /* Device mapper interface */
 223
 224 #ifdef CONFIG_MPATH
 225 #define CONTROL_PATH "/dev/mapper/control"
 226
 227 typedef struct DMData {
 228     struct dm_ioctl dm;
 229     uint8_t data[1024];
 230 } DMData;
 231
 232 static int control_fd;
 233
 234 static void *dm_ioctl(int ioc, struct dm_ioctl *dm)
 235 {
 236     static DMData d;
 237     memcpy(&d.dm, dm, sizeof(d.dm));
 238     QEMU_BUILD_BUG_ON(sizeof(d.data) < sizeof(struct dm_target_spec));
 239
 240     d.dm.version[0] = DM_VERSION_MAJOR;
 241     d.dm.version[1] = 0;
 242     d.dm.version[2] = 0;
 243     d.dm.data_size = 1024;
 244     d.dm.data_start = offsetof(DMData, data);
 245     if (ioctl(control_fd, ioc, &d) < 0) {
 246         return NULL;
 247     }
 248     memcpy(dm, &d.dm, sizeof(d.dm));
 249     return &d.data;
 250 }
 251
 252 static void *dm_dev_ioctl(int fd, int ioc, struct dm_ioctl *dm)
 253 {
 254     struct stat st;
 255     int r;
 256
 257     r = fstat(fd, &st);
 258     if (r < 0) {
 259         perror("fstat");
 260         exit(1);
 261     }
 262
 263     dm->dev = st.st_rdev;
 264     return dm_ioctl(ioc, dm);
 265 }
 266
 267 static void dm_init(void)
 268 {
 269     control_fd = open(CONTROL_PATH, O_RDWR);
 270     if (control_fd < 0) {
 271         perror("Cannot open " CONTROL_PATH);
 272         exit(1);
 273     }
 274     struct dm_ioctl dm = { 0 };
 275     if (!dm_ioctl(DM_VERSION, &dm)) {
 276         perror("ioctl");
 277         exit(1);
 278     }
 279     if (dm.version[0] != DM_VERSION_MAJOR) {
 280         fprintf(stderr, "Unsupported device mapper interface");
 281         exit(1);
 282     }
 283 }
 284
 285 /* Variables required by libmultipath and libmpathpersist.  */
 286 QEMU_BUILD_BUG_ON(PR_HELPER_DATA_SIZE > MPATH_MAX_PARAM_LEN);
 287 static struct config *multipath_conf;
 288 unsigned mpath_mx_alloc_len = PR_HELPER_DATA_SIZE;
 289 int logsink;
 290 struct udev *udev;
 291
 292 extern struct config *get_multipath_config(void);
 293 struct config *get_multipath_config(void)
 294 {
 295     return multipath_conf;
 296 }
 297
 298 extern void put_multipath_config(struct config *conf);
 299 void put_multipath_config(struct config *conf)
 300 {
 301 }
 302
 303 static void multipath_pr_init(void)
 304 {
 305     udev = udev_new();
 306     multipath_conf = mpath_lib_init();
 307 }
 308
 309 static int is_mpath(int fd)
 310 {
 311     struct dm_ioctl dm = { .flags = DM_NOFLUSH_FLAG };
 312     struct dm_target_spec *tgt;
 313
 314     tgt = dm_dev_ioctl(fd, DM_TABLE_STATUS, &dm);
 315     if (!tgt) {
 316         if (errno == ENXIO) {
 317             return 0;
 318         }
 319         perror("ioctl");
 320         exit(EXIT_FAILURE);
 321     }
 322     return !strncmp(tgt->target_type, "multipath", DM_MAX_TYPE_NAME);
 323 }
 324
 325 static SCSISense mpath_generic_sense(int r)
 326 {
 327     switch (r) {
 328     case MPATH_PR_SENSE_NOT_READY:
 329          return SENSE_CODE(NOT_READY);
 330     case MPATH_PR_SENSE_MEDIUM_ERROR:
 331          return SENSE_CODE(READ_ERROR);
 332     case MPATH_PR_SENSE_HARDWARE_ERROR:
 333          return SENSE_CODE(TARGET_FAILURE);
 334     case MPATH_PR_SENSE_ABORTED_COMMAND:
 335          return SENSE_CODE(IO_ERROR);
 336     default:
 337          abort();
 338     }
 339 }
 340
 341 static int mpath_reconstruct_sense(int fd, int r, uint8_t *sense)
 342 {
 343     switch (r) {
 344     case MPATH_PR_SUCCESS:
 345         return GOOD;
 346     case MPATH_PR_SENSE_NOT_READY:
 347     case MPATH_PR_SENSE_MEDIUM_ERROR:
 348     case MPATH_PR_SENSE_HARDWARE_ERROR:
 349     case MPATH_PR_SENSE_ABORTED_COMMAND:
 350         {
 351             /* libmpathpersist ate the exact sense.  Try to find it by
 352              * issuing TEST UNIT READY.
 353              */
 354             uint8_t cdb[6] = { TEST_UNIT_READY };
 355             int sz = 0;
 356             int r = do_sgio(fd, cdb, sense, NULL, &sz, SG_DXFER_NONE);
 357
 358             if (r != GOOD) {
 359                 return r;
 360             }
 361             scsi_build_sense(sense, mpath_generic_sense(r));
 362             return CHECK_CONDITION;
 363         }
 364
 365     case MPATH_PR_SENSE_UNIT_ATTENTION:
 366         /* Congratulations libmpathpersist, you ruined the Unit Attention...
 367          * Return a heavyweight one.
 368          */
 369         scsi_build_sense(sense, SENSE_CODE(SCSI_BUS_RESET));
 370         return CHECK_CONDITION;
 371     case MPATH_PR_SENSE_INVALID_OP:
 372         /* Only one valid sense.  */
 373         scsi_build_sense(sense, SENSE_CODE(INVALID_OPCODE));
 374         return CHECK_CONDITION;
 375     case MPATH_PR_ILLEGAL_REQ:
 376         /* Guess.  */
 377         scsi_build_sense(sense, SENSE_CODE(INVALID_PARAM));
 378         return CHECK_CONDITION;
 379     case MPATH_PR_NO_SENSE:
 380         scsi_build_sense(sense, SENSE_CODE(NO_SENSE));
 381         return CHECK_CONDITION;
 382
 383     case MPATH_PR_RESERV_CONFLICT:
 384         return RESERVATION_CONFLICT;
 385
 386     case MPATH_PR_OTHER:
 387     default:
 388         scsi_build_sense(sense, SENSE_CODE(LUN_COMM_FAILURE));
 389         return CHECK_CONDITION;
 390     }
 391 }
 392
 393 static int multipath_pr_in(int fd, const uint8_t *cdb, uint8_t *sense,
 394                            uint8_t *data, int sz)
 395 {
 396     int rq_servact = cdb[1];
 397     struct prin_resp resp;
 398     size_t written;
 399     int r;
 400
 401     switch (rq_servact) {
 402     case MPATH_PRIN_RKEY_SA:
 403     case MPATH_PRIN_RRES_SA:
 404     case MPATH_PRIN_RCAP_SA:
 405         break;
 406     case MPATH_PRIN_RFSTAT_SA:
 407         /* Nobody implements it anyway, so bail out. */
 408     default:
 409         /* Cannot parse any other output.  */
 410         scsi_build_sense(sense, SENSE_CODE(INVALID_FIELD));
 411         return CHECK_CONDITION;
 412     }
 413
 414     r = mpath_persistent_reserve_in(fd, rq_servact, &resp, noisy, verbose);
 415     if (r == MPATH_PR_SUCCESS) {
 416         switch (rq_servact) {
 417         case MPATH_PRIN_RKEY_SA:
 418         case MPATH_PRIN_RRES_SA: {
 419             struct prin_readdescr *out = &resp.prin_descriptor.prin_readkeys;
 420             assert(sz >= 8);
 421             written = MIN(out->additional_length + 8, sz);
 422             stl_be_p(&data[0], out->prgeneration);
 423             stl_be_p(&data[4], out->additional_length);
 424             memcpy(&data[8], out->key_list, written - 8);
 425             break;
 426         }
 427         case MPATH_PRIN_RCAP_SA: {
 428             struct prin_capdescr *out = &resp.prin_descriptor.prin_readcap;
 429             assert(sz >= 6);
 430             written = 6;
 431             stw_be_p(&data[0], out->length);
 432             data[2] = out->flags[0];
 433             data[3] = out->flags[1];
 434             stw_be_p(&data[4], out->pr_type_mask);
 435             break;
 436         }
 437         default:
 438             scsi_build_sense(sense, SENSE_CODE(INVALID_OPCODE));
 439             return CHECK_CONDITION;
 440         }
 441         assert(written <= sz);
 442         memset(data + written, 0, sz - written);
 443     }
 444
 445     return mpath_reconstruct_sense(fd, r, sense);
 446 }
 447
 448 static int multipath_pr_out(int fd, const uint8_t *cdb, uint8_t *sense,
 449                             const uint8_t *param, int sz)
 450 {
 451     int rq_servact = cdb[1];
 452     int rq_scope = cdb[2] >> 4;
 453     int rq_type = cdb[2] & 0xf;
 454     struct prout_param_descriptor paramp;
 455     char transportids[PR_HELPER_DATA_SIZE];
 456     int r;
 457
 458     if (sz < PR_OUT_FIXED_PARAM_SIZE) {
 459         /* Illegal request, Parameter list length error.  This isn't fatal;
 460          * we have read the data, send an error without closing the socket.
 461          */
 462         scsi_build_sense(sense, SENSE_CODE(INVALID_PARAM_LEN));
 463         return CHECK_CONDITION;
 464     }
 465
 466     switch (rq_servact) {
 467     case MPATH_PROUT_REG_SA:
 468     case MPATH_PROUT_RES_SA:
 469     case MPATH_PROUT_REL_SA:
 470     case MPATH_PROUT_CLEAR_SA:
 471     case MPATH_PROUT_PREE_SA:
 472     case MPATH_PROUT_PREE_AB_SA:
 473     case MPATH_PROUT_REG_IGN_SA:
 474         break;
 475     case MPATH_PROUT_REG_MOV_SA:
 476         /* Not supported by struct prout_param_descriptor.  */
 477     default:
 478         /* Cannot parse any other input.  */
 479         scsi_build_sense(sense, SENSE_CODE(INVALID_FIELD));
 480         return CHECK_CONDITION;
 481     }
 482
 483     /* Convert input data, especially transport IDs, to the structs
 484      * used by libmpathpersist (which, of course, will immediately
 485      * do the opposite).
 486      */
 487     memset(&paramp, 0, sizeof(paramp));
 488     memcpy(&paramp.key, &param[0], 8);
 489     memcpy(&paramp.sa_key, &param[8], 8);
 490     paramp.sa_flags = param[20];
 491     if (sz > PR_OUT_FIXED_PARAM_SIZE) {
 492         size_t transportid_len;
 493         int i, j;
 494         if (sz < PR_OUT_FIXED_PARAM_SIZE + 4) {
 495             scsi_build_sense(sense, SENSE_CODE(INVALID_PARAM_LEN));
 496             return CHECK_CONDITION;
 497         }
 498         transportid_len = ldl_be_p(&param[24]) + PR_OUT_FIXED_PARAM_SIZE + 4;
 499         if (transportid_len > sz) {
 500             scsi_build_sense(sense, SENSE_CODE(INVALID_PARAM));
 501             return CHECK_CONDITION;
 502         }
 503         for (i = PR_OUT_FIXED_PARAM_SIZE + 4, j = 0; i < transportid_len; ) {
 504             struct transportid *id = (struct transportid *) &transportids[j];
 505             int len;
 506
 507             id->format_code = param[i] & 0xc0;
 508             id->protocol_id = param[i] & 0x0f;
 509             switch (param[i] & 0xcf) {
 510             case 0:
 511                 /* FC transport.  */
 512                 if (i + 24 > transportid_len) {
 513                     goto illegal_req;
 514                 }
 515                 memcpy(id->n_port_name, &param[i + 8], 8);
 516                 j += offsetof(struct transportid, n_port_name[8]);
 517                 i += 24;
 518                 break;
 519             case 5:
 520             case 0x45:
 521                 /* iSCSI transport.  */
 522                 len = lduw_be_p(&param[i + 2]);
 523                 if (len > 252 || (len & 3) || i + len + 4 > transportid_len) {
 524                     /* For format code 00, the standard says the maximum is 223
 525                      * plus the NUL terminator.  For format code 01 there is no
 526                      * maximum length, but libmpathpersist ignores the first
 527                      * byte of id->iscsi_name so our maximum is 252.
 528                      */
 529                     goto illegal_req;
 530                 }
 531                 if (memchr(&param[i + 4], 0, len) == NULL) {
 532                     goto illegal_req;
 533                 }
 534                 memcpy(id->iscsi_name, &param[i + 2], len + 2);
 535                 j += offsetof(struct transportid, iscsi_name[len + 2]);
 536                 i += len + 4;
 537                 break;
 538             case 6:
 539                 /* SAS transport.  */
 540                 if (i + 24 > transportid_len) {
 541                     goto illegal_req;
 542                 }
 543                 memcpy(id->sas_address, &param[i + 4], 8);
 544                 j += offsetof(struct transportid, sas_address[8]);
 545                 i += 24;
 546                 break;
 547             default:
 548             illegal_req:
 549                 scsi_build_sense(sense, SENSE_CODE(INVALID_PARAM));
 550                 return CHECK_CONDITION;
 551             }
 552
 553             paramp.trnptid_list[paramp.num_transportid++] = id;
 554         }
 555     }
 556
 557     r = mpath_persistent_reserve_out(fd, rq_servact, rq_scope, rq_type,
 558                                      &paramp, noisy, verbose);
 559     return mpath_reconstruct_sense(fd, r, sense);
 560 }
 561 #endif
 562
 563 static int do_pr_in(int fd, const uint8_t *cdb, uint8_t *sense,
 564                     uint8_t *data, int *resp_sz)
 565 {
 566 #ifdef CONFIG_MPATH
 567     if (is_mpath(fd)) {
 568         /* multipath_pr_in fills the whole input buffer.  */
 569         int r = multipath_pr_in(fd, cdb, sense, data, *resp_sz);
 570         if (r != GOOD) {
 571             *resp_sz = 0;
 572         }
 573         return r;
 574     }
 575 #endif
 576
 577     return do_sgio(fd, cdb, sense, data, resp_sz,
 578                    SG_DXFER_FROM_DEV);
 579 }
 580
 581 static int do_pr_out(int fd, const uint8_t *cdb, uint8_t *sense,
 582                      const uint8_t *param, int sz)
 583 {
 584     int resp_sz;
 585
 586     if ((fcntl(fd, F_GETFL) & O_ACCMODE) == O_RDONLY) {
 587         scsi_build_sense(sense, SENSE_CODE(INVALID_OPCODE));
 588         return CHECK_CONDITION;
 589     }
 590
 591 #ifdef CONFIG_MPATH
 592     if (is_mpath(fd)) {
 593         return multipath_pr_out(fd, cdb, sense, param, sz);
 594     }
 595 #endif
 596
 597     resp_sz = sz;
 598     return do_sgio(fd, cdb, sense, (uint8_t *)param, &resp_sz,
 599                    SG_DXFER_TO_DEV);
 600 }
 601
 602 /* Client */
 603
 604 typedef struct PRHelperClient {
 605     QIOChannelSocket *ioc;
 606     Coroutine *co;
 607     int fd;
 608     uint8_t data[PR_HELPER_DATA_SIZE];
 609 } PRHelperClient;
 610
 611 typedef struct PRHelperRequest {
 612     int fd;
 613     size_t sz;
 614     uint8_t cdb[PR_HELPER_CDB_SIZE];
 615 } PRHelperRequest;
 616
 617 static int coroutine_fn prh_read(PRHelperClient *client, void *buf, int sz,
 618                                  Error **errp)
 619 {
 620     int ret = 0;
 621
 622     while (sz > 0) {
 623         int *fds = NULL;
 624         size_t nfds = 0;
 625         int i;
 626         struct iovec iov;
 627         ssize_t n_read;
 628
 629         iov.iov_base = buf;
 630         iov.iov_len = sz;
 631         n_read = qio_channel_readv_full(QIO_CHANNEL(client->ioc), &iov, 1,
 632                                         &fds, &nfds, errp);
 633
 634         if (n_read == QIO_CHANNEL_ERR_BLOCK) {
 635             qio_channel_yield(QIO_CHANNEL(client->ioc), G_IO_IN);
 636             continue;
 637         }
 638         if (n_read <= 0) {
 639             ret = n_read ? n_read : -1;
 640             goto err;
 641         }
 642
 643         /* Stash one file descriptor per request.  */
 644         if (nfds) {
 645             bool too_many = false;
 646             for (i = 0; i < nfds; i++) {
 647                 if (client->fd == -1) {
 648                     client->fd = fds[i];
 649                 } else {
 650                     close(fds[i]);
 651                     too_many = true;
 652                 }
 653             }
 654             g_free(fds);
 655             if (too_many) {
 656                 ret = -1;
 657                 goto err;
 658             }
 659         }
 660
 661         buf += n_read;
 662         sz -= n_read;
 663     }
 664
 665     return 0;
 666
 667 err:
 668     if (client->fd != -1) {
 669         close(client->fd);
 670         client->fd = -1;
 671     }
 672     return ret;
 673 }
 674
 675 static int coroutine_fn prh_read_request(PRHelperClient *client,
 676                                          PRHelperRequest *req,
 677                                          PRHelperResponse *resp, Error **errp)
 678 {
 679     uint32_t sz;
 680
 681     if (prh_read(client, req->cdb, sizeof(req->cdb), NULL) < 0) {
 682         return -1;
 683     }
 684
 685     if (client->fd == -1) {
 686         error_setg(errp, "No file descriptor in request.");
 687         return -1;
 688     }
 689
 690     if (req->cdb[0] != PERSISTENT_RESERVE_OUT &&
 691         req->cdb[0] != PERSISTENT_RESERVE_IN) {
 692         error_setg(errp, "Invalid CDB, closing socket.");
 693         goto out_close;
 694     }
 695
 696     sz = scsi_cdb_xfer(req->cdb);
 697     if (sz > sizeof(client->data)) {
 698         goto out_close;
 699     }
 700
 701     if (req->cdb[0] == PERSISTENT_RESERVE_OUT) {
 702         if (qio_channel_read_all(QIO_CHANNEL(client->ioc),
 703                                  (char *)client->data, sz,
 704                                  errp) < 0) {
 705             goto out_close;
 706         }
 707     }
 708
 709     req->fd = client->fd;
 710     req->sz = sz;
 711     client->fd = -1;
 712     return sz;
 713
 714 out_close:
 715     close(client->fd);
 716     client->fd = -1;
 717     return -1;
 718 }
 719
 720 static int coroutine_fn prh_write_response(PRHelperClient *client,
 721                                            PRHelperRequest *req,
 722                                            PRHelperResponse *resp, Error **errp)
 723 {
 724     ssize_t r;
 725     size_t sz;
 726
 727     if (req->cdb[0] == PERSISTENT_RESERVE_IN && resp->result == GOOD) {
 728         assert(resp->sz <= req->sz && resp->sz <= sizeof(client->data));
 729     } else {
 730         assert(resp->sz == 0);
 731     }
 732
 733     sz = resp->sz;
 734
 735     resp->result = cpu_to_be32(resp->result);
 736     resp->sz = cpu_to_be32(resp->sz);
 737     r = qio_channel_write_all(QIO_CHANNEL(client->ioc),
 738                               (char *) resp, sizeof(*resp), errp);
 739     if (r < 0) {
 740         return r;
 741     }
 742
 743     r = qio_channel_write_all(QIO_CHANNEL(client->ioc),
 744                               (char *) client->data,
 745                               sz, errp);
 746     return r < 0 ? r : 0;
 747 }
 748
 749 static void coroutine_fn prh_co_entry(void *opaque)
 750 {
 751     PRHelperClient *client = opaque;
 752     Error *local_err = NULL;
 753     uint32_t flags;
 754     int r;
 755
 756     qio_channel_set_blocking(QIO_CHANNEL(client->ioc),
 757                              false, NULL);
 758     qio_channel_attach_aio_context(QIO_CHANNEL(client->ioc),
 759                                    qemu_get_aio_context());
 760
 761     /* A very simple negotiation for future extensibility.  No features
 762      * are defined so write 0.
 763      */
 764     flags = cpu_to_be32(0);
 765     r = qio_channel_write_all(QIO_CHANNEL(client->ioc),
 766                              (char *) &flags, sizeof(flags), NULL);
 767     if (r < 0) {
 768         goto out;
 769     }
 770
 771     r = qio_channel_read_all(QIO_CHANNEL(client->ioc),
 772                              (char *) &flags, sizeof(flags), NULL);
 773     if (be32_to_cpu(flags) != 0 || r < 0) {
 774         goto out;
 775     }
 776
 777     while (atomic_read(&state) == RUNNING) {
 778         PRHelperRequest req;
 779         PRHelperResponse resp;
 780         int sz;
 781
 782         sz = prh_read_request(client, &req, &resp, &local_err);
 783         if (sz < 0) {
 784             break;
 785         }
 786
 787         num_active_sockets++;
 788         if (req.cdb[0] == PERSISTENT_RESERVE_OUT) {
 789             r = do_pr_out(req.fd, req.cdb, resp.sense,
 790                           client->data, sz);
 791             resp.sz = 0;
 792         } else {
 793             resp.sz = sizeof(client->data);
 794             r = do_pr_in(req.fd, req.cdb, resp.sense,
 795                          client->data, &resp.sz);
 796             resp.sz = MIN(resp.sz, sz);
 797         }
 798         num_active_sockets--;
 799         close(req.fd);
 800         if (r == -1) {
 801             break;
 802         }
 803         resp.result = r;
 804
 805         if (prh_write_response(client, &req, &resp, &local_err) < 0) {
 806             break;
 807         }
 808     }
 809
 810     if (local_err) {
 811         if (verbose == 0) {
 812             error_free(local_err);
 813         } else {
 814             error_report_err(local_err);
 815         }
 816     }
 817
 818 out:
 819     qio_channel_detach_aio_context(QIO_CHANNEL(client->ioc));
 820     object_unref(OBJECT(client->ioc));
 821     g_free(client);
 822 }
 823
 824 static gboolean accept_client(QIOChannel *ioc, GIOCondition cond, gpointer opaque)
 825 {
 826     QIOChannelSocket *cioc;
 827     PRHelperClient *prh;
 828
 829     cioc = qio_channel_socket_accept(QIO_CHANNEL_SOCKET(ioc),
 830                                      NULL);
 831     if (!cioc) {
 832         return TRUE;
 833     }
 834
 835     prh = g_new(PRHelperClient, 1);
 836     prh->ioc = cioc;
 837     prh->fd = -1;
 838     prh->co = qemu_coroutine_create(prh_co_entry, prh);
 839     qemu_coroutine_enter(prh->co);
 840
 841     return TRUE;
 842 }
 843
 844
 845 /*
 846  * Check socket parameters compatibility when socket activation is used.
 847  */
 848 static const char *socket_activation_validate_opts(void)
 849 {
 850     if (socket_path != NULL) {
 851         return "Unix socket can't be set when using socket activation";
 852     }
 853
 854     return NULL;
 855 }
 856
 857 static void termsig_handler(int signum)
 858 {
 859     atomic_cmpxchg(&state, RUNNING, TERMINATE);
 860     qemu_notify_event();
 861 }
 862
 863 static void close_server_socket(void)
 864 {
 865     assert(server_ioc);
 866
 867     g_source_remove(server_watch);
 868     server_watch = -1;
 869     object_unref(OBJECT(server_ioc));
 870     num_active_sockets--;
 871 }
 872
 873 #ifdef CONFIG_LIBCAP
 874 static int drop_privileges(void)
 875 {
 876     /* clear all capabilities */
 877     capng_clear(CAPNG_SELECT_BOTH);
 878
 879     if (capng_update(CAPNG_ADD, CAPNG_EFFECTIVE | CAPNG_PERMITTED,
 880                      CAP_SYS_RAWIO) < 0) {
 881         return -1;
 882     }
 883
 884 #ifdef CONFIG_MPATH
 885     /* For /dev/mapper/control ioctls */
 886     if (capng_update(CAPNG_ADD, CAPNG_EFFECTIVE | CAPNG_PERMITTED,
 887                      CAP_SYS_ADMIN) < 0) {
 888         return -1;
 889     }
 890 #endif
 891
 892     /* Change user/group id, retaining the capabilities.  Because file descriptors
 893      * are passed via SCM_RIGHTS, we don't need supplementary groups (and in
 894      * fact the helper can run as "nobody").
 895      */
 896     if (capng_change_id(uid != -1 ? uid : getuid(),
 897                         gid != -1 ? gid : getgid(),
 898                         CAPNG_DROP_SUPP_GRP | CAPNG_CLEAR_BOUNDING)) {
 899         return -1;
 900     }
 901
 902     return 0;
 903 }
 904 #endif
 905
 906 int main(int argc, char **argv)
 907 {
 908     const char *sopt = "hVk:f:dT:u:g:vq";
 909     struct option lopt[] = {
 910         { "help", no_argument, NULL, 'h' },
 911         { "version", no_argument, NULL, 'V' },
 912         { "socket", required_argument, NULL, 'k' },
 913         { "pidfile", required_argument, NULL, 'f' },
 914         { "daemon", no_argument, NULL, 'd' },
 915         { "trace", required_argument, NULL, 'T' },
 916         { "user", required_argument, NULL, 'u' },
 917         { "group", required_argument, NULL, 'g' },
 918         { "verbose", no_argument, NULL, 'v' },
 919         { "quiet", no_argument, NULL, 'q' },
 920         { NULL, 0, NULL, 0 }
 921     };
 922     int opt_ind = 0;
 923     int loglevel = 1;
 924     int quiet = 0;
 925     int ch;
 926     Error *local_err = NULL;
 927     char *trace_file = NULL;
 928     bool daemonize = false;
 929     bool pidfile_specified = false;
 930     unsigned socket_activation;
 931
 932     struct sigaction sa_sigterm;
 933     memset(&sa_sigterm, 0, sizeof(sa_sigterm));
 934     sa_sigterm.sa_handler = termsig_handler;
 935     sigaction(SIGTERM, &sa_sigterm, NULL);
 936     sigaction(SIGINT, &sa_sigterm, NULL);
 937     sigaction(SIGHUP, &sa_sigterm, NULL);
 938
 939     signal(SIGPIPE, SIG_IGN);
 940
 941     module_call_init(MODULE_INIT_TRACE);
 942     module_call_init(MODULE_INIT_QOM);
 943     qemu_add_opts(&qemu_trace_opts);
 944     qemu_init_exec_dir(argv[0]);
 945
 946     pidfile = qemu_get_local_state_pathname("run/qemu-pr-helper.pid");
 947
 948     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 949         switch (ch) {
 950         case 'k':
 951             socket_path = optarg;
 952             if (socket_path[0] != '/') {
 953                 error_report("socket path must be absolute");
 954                 exit(EXIT_FAILURE);
 955             }
 956             break;
 957         case 'f':
 958             g_free(pidfile);
 959             pidfile = g_strdup(optarg);
 960             pidfile_specified = true;
 961             break;
 962 #ifdef CONFIG_LIBCAP
 963         case 'u': {
 964             unsigned long res;
 965             struct passwd *userinfo = getpwnam(optarg);
 966             if (userinfo) {
 967                 uid = userinfo->pw_uid;
 968             } else if (qemu_strtoul(optarg, NULL, 10, &res) == 0 &&
 969                        (uid_t)res == res) {
 970                 uid = res;
 971             } else {
 972                 error_report("invalid user '%s'", optarg);
 973                 exit(EXIT_FAILURE);
 974             }
 975             break;
 976         }
 977         case 'g': {
 978             unsigned long res;
 979             struct group *groupinfo = getgrnam(optarg);
 980             if (groupinfo) {
 981                 gid = groupinfo->gr_gid;
 982             } else if (qemu_strtoul(optarg, NULL, 10, &res) == 0 &&
 983                        (gid_t)res == res) {
 984                 gid = res;
 985             } else {
 986                 error_report("invalid group '%s'", optarg);
 987                 exit(EXIT_FAILURE);
 988             }
 989             break;
 990         }
 991 #else
 992         case 'u':
 993         case 'g':
 994             error_report("-%c not supported by this %s", ch, argv[0]);
 995             exit(1);
 996 #endif
 997         case 'd':
 998             daemonize = true;
 999             break;
1000         case 'q':
1001             quiet = 1;
1002             break;
1003         case 'v':
1004             ++loglevel;
1005             break;
1006         case 'T':
1007             g_free(trace_file);
1008             trace_file = trace_opt_parse(optarg);
1009             break;
1010         case 'V':
1011             version(argv[0]);
1012             exit(EXIT_SUCCESS);
1013             break;
1014         case 'h':
1015             usage(argv[0]);
1016             exit(EXIT_SUCCESS);
1017             break;
1018         case '?':
1019             error_report("Try `%s --help' for more information.", argv[0]);
1020             exit(EXIT_FAILURE);
1021         }
1022     }
1023
1024     /* set verbosity */
1025     noisy = !quiet && (loglevel >= 3);
1026     verbose = quiet ? 0 : MIN(loglevel, 3);
1027
1028     if (!trace_init_backends()) {
1029         exit(EXIT_FAILURE);
1030     }
1031     trace_init_file(trace_file);
1032     qemu_set_log(LOG_TRACE);
1033
1034 #ifdef CONFIG_MPATH
1035     dm_init();
1036     multipath_pr_init();
1037 #endif
1038
1039     socket_activation = check_socket_activation();
1040     if (socket_activation == 0) {
1041         SocketAddress saddr;
1042         compute_default_paths();
1043         saddr = (SocketAddress){
1044             .type = SOCKET_ADDRESS_TYPE_UNIX,
1045             .u.q_unix.path = g_strdup(socket_path)
1046         };
1047         server_ioc = qio_channel_socket_new();
1048         if (qio_channel_socket_listen_sync(server_ioc, &saddr, &local_err) < 0) {
1049             object_unref(OBJECT(server_ioc));
1050             error_report_err(local_err);
1051             return 1;
1052         }
1053         g_free(saddr.u.q_unix.path);
1054     } else {
1055         /* Using socket activation - check user didn't use -p etc. */
1056         const char *err_msg = socket_activation_validate_opts();
1057         if (err_msg != NULL) {
1058             error_report("%s", err_msg);
1059             exit(EXIT_FAILURE);
1060         }
1061
1062         /* Can only listen on a single socket.  */
1063         if (socket_activation > 1) {
1064             error_report("%s does not support socket activation with LISTEN_FDS > 1",
1065                          argv[0]);
1066             exit(EXIT_FAILURE);
1067         }
1068         server_ioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD,
1069                                                &local_err);
1070         if (server_ioc == NULL) {
1071             error_report("Failed to use socket activation: %s",
1072                          error_get_pretty(local_err));
1073             exit(EXIT_FAILURE);
1074         }
1075         socket_path = NULL;
1076     }
1077
1078     if (qemu_init_main_loop(&local_err)) {
1079         error_report_err(local_err);
1080         exit(EXIT_FAILURE);
1081     }
1082
1083     server_watch = qio_channel_add_watch(QIO_CHANNEL(server_ioc),
1084                                          G_IO_IN,
1085                                          accept_client,
1086                                          NULL, NULL);
1087
1088     if (daemonize) {
1089         if (daemon(0, 0) < 0) {
1090             error_report("Failed to daemonize: %s", strerror(errno));
1091             exit(EXIT_FAILURE);
1092         }
1093     }
1094
1095     if (daemonize || pidfile_specified)
1096         write_pidfile();
1097
1098 #ifdef CONFIG_LIBCAP
1099     if (drop_privileges() < 0) {
1100         error_report("Failed to drop privileges: %s", strerror(errno));
1101         exit(EXIT_FAILURE);
1102     }
1103 #endif
1104
1105     state = RUNNING;
1106     do {
1107         main_loop_wait(false);
1108         if (state == TERMINATE) {
1109             state = TERMINATING;
1110             close_server_socket();
1111         }
1112     } while (num_active_sockets > 0);
1113
1114     exit(EXIT_SUCCESS);
1115 }