xref: /openbmc/qemu/scsi/qemu-pr-helper.c (revision 1770b2f2)
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_NG
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 "qemu/help-texts.h"
40 #include "qapi/error.h"
41 #include "qemu/cutils.h"
42 #include "qemu/main-loop.h"
43 #include "qemu/module.h"
44 #include "qemu/error-report.h"
45 #include "qemu/config-file.h"
46 #include "qemu/bswap.h"
47 #include "qemu/log.h"
48 #include "qemu/systemd.h"
49 #include "qapi/util.h"
50 #include "qapi/qmp/qstring.h"
51 #include "io/channel-socket.h"
52 #include "trace/control.h"
53 #include "qemu-version.h"
54 
55 #include "block/aio.h"
56 #include "block/thread-pool.h"
57 
58 #include "scsi/constants.h"
59 #include "scsi/utils.h"
60 #include "pr-helper.h"
61 
62 #define PR_OUT_FIXED_PARAM_SIZE 24
63 
64 static char *socket_path;
65 static char *pidfile;
66 static enum { RUNNING, TERMINATE, TERMINATING } state;
67 static QIOChannelSocket *server_ioc;
68 static int server_watch;
69 static int num_active_sockets = 1;
70 static int noisy;
71 static int verbose;
72 
73 #ifdef CONFIG_LIBCAP_NG
74 static int uid = -1;
75 static int gid = -1;
76 #endif
77 
78 static void compute_default_paths(void)
79 {
80     g_autofree char *state = qemu_get_local_state_dir();
81 
82     socket_path = g_build_filename(state, "run", "qemu-pr-helper.sock", NULL);
83     pidfile = g_build_filename(state, "run", "qemu-pr-helper.pid", NULL);
84 }
85 
86 static void usage(const char *name)
87 {
88     (printf) (
89 "Usage: %s [OPTIONS] FILE\n"
90 "Persistent Reservation helper program for QEMU\n"
91 "\n"
92 "  -h, --help                display this help and exit\n"
93 "  -V, --version             output version information and exit\n"
94 "\n"
95 "  -d, --daemon              run in the background\n"
96 "  -f, --pidfile=PATH        PID file when running as a daemon\n"
97 "                            (default '%s')\n"
98 "  -k, --socket=PATH         path to the unix socket\n"
99 "                            (default '%s')\n"
100 "  -T, --trace [[enable=]<pattern>][,events=<file>][,file=<file>]\n"
101 "                            specify tracing options\n"
102 #ifdef CONFIG_LIBCAP_NG
103 "  -u, --user=USER           user to drop privileges to\n"
104 "  -g, --group=GROUP         group to drop privileges to\n"
105 #endif
106 "\n"
107 QEMU_HELP_BOTTOM "\n"
108     , name, pidfile, socket_path);
109 }
110 
111 static void version(const char *name)
112 {
113     printf(
114 "%s " QEMU_FULL_VERSION "\n"
115 "Written by Paolo Bonzini.\n"
116 "\n"
117 QEMU_COPYRIGHT "\n"
118 "This is free software; see the source for copying conditions.  There is NO\n"
119 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
120     , name);
121 }
122 
123 /* SG_IO support */
124 
125 typedef struct PRHelperSGIOData {
126     int fd;
127     const uint8_t *cdb;
128     uint8_t *sense;
129     uint8_t *buf;
130     int sz;              /* input/output */
131     int dir;
132 } PRHelperSGIOData;
133 
134 static int do_sgio_worker(void *opaque)
135 {
136     PRHelperSGIOData *data = opaque;
137     struct sg_io_hdr io_hdr;
138     int ret;
139     int status;
140     SCSISense sense_code;
141 
142     memset(data->sense, 0, PR_HELPER_SENSE_SIZE);
143     memset(&io_hdr, 0, sizeof(io_hdr));
144     io_hdr.interface_id = 'S';
145     io_hdr.cmd_len = PR_HELPER_CDB_SIZE;
146     io_hdr.cmdp = (uint8_t *)data->cdb;
147     io_hdr.sbp = data->sense;
148     io_hdr.mx_sb_len = PR_HELPER_SENSE_SIZE;
149     io_hdr.timeout = 1;
150     io_hdr.dxfer_direction = data->dir;
151     io_hdr.dxferp = (char *)data->buf;
152     io_hdr.dxfer_len = data->sz;
153     ret = ioctl(data->fd, SG_IO, &io_hdr);
154 
155     if (ret < 0) {
156         status = scsi_sense_from_errno(errno, &sense_code);
157         if (status == CHECK_CONDITION) {
158             scsi_build_sense(data->sense, sense_code);
159         }
160     } else if (io_hdr.host_status != SCSI_HOST_OK) {
161         status = scsi_sense_from_host_status(io_hdr.host_status, &sense_code);
162         if (status == CHECK_CONDITION) {
163             scsi_build_sense(data->sense, sense_code);
164         }
165     } else if (io_hdr.driver_status & SG_ERR_DRIVER_TIMEOUT) {
166         status = BUSY;
167     } else {
168         status = io_hdr.status;
169     }
170 
171     if (status == GOOD) {
172         data->sz -= io_hdr.resid;
173     } else {
174         data->sz = 0;
175     }
176 
177     return status;
178 }
179 
180 static int do_sgio(int fd, const uint8_t *cdb, uint8_t *sense,
181                     uint8_t *buf, int *sz, int dir)
182 {
183     ThreadPool *pool = aio_get_thread_pool(qemu_get_aio_context());
184     int r;
185 
186     PRHelperSGIOData data = {
187         .fd = fd,
188         .cdb = cdb,
189         .sense = sense,
190         .buf = buf,
191         .sz = *sz,
192         .dir = dir,
193     };
194 
195     r = thread_pool_submit_co(pool, do_sgio_worker, &data);
196     *sz = data.sz;
197     return r;
198 }
199 
200 /* Device mapper interface */
201 
202 #ifdef CONFIG_MPATH
203 #define CONTROL_PATH "/dev/mapper/control"
204 
205 typedef struct DMData {
206     struct dm_ioctl dm;
207     uint8_t data[1024];
208 } DMData;
209 
210 static int control_fd;
211 
212 static void *dm_ioctl(int ioc, struct dm_ioctl *dm)
213 {
214     static DMData d;
215     memcpy(&d.dm, dm, sizeof(d.dm));
216     QEMU_BUILD_BUG_ON(sizeof(d.data) < sizeof(struct dm_target_spec));
217 
218     d.dm.version[0] = DM_VERSION_MAJOR;
219     d.dm.version[1] = 0;
220     d.dm.version[2] = 0;
221     d.dm.data_size = 1024;
222     d.dm.data_start = offsetof(DMData, data);
223     if (ioctl(control_fd, ioc, &d) < 0) {
224         return NULL;
225     }
226     memcpy(dm, &d.dm, sizeof(d.dm));
227     return &d.data;
228 }
229 
230 static void *dm_dev_ioctl(int fd, int ioc, struct dm_ioctl *dm)
231 {
232     struct stat st;
233     int r;
234 
235     r = fstat(fd, &st);
236     if (r < 0) {
237         perror("fstat");
238         exit(1);
239     }
240 
241     dm->dev = st.st_rdev;
242     return dm_ioctl(ioc, dm);
243 }
244 
245 static void dm_init(void)
246 {
247     control_fd = open(CONTROL_PATH, O_RDWR);
248     if (control_fd < 0) {
249         perror("Cannot open " CONTROL_PATH);
250         exit(1);
251     }
252     struct dm_ioctl dm = { };
253     if (!dm_ioctl(DM_VERSION, &dm)) {
254         perror("ioctl");
255         exit(1);
256     }
257     if (dm.version[0] != DM_VERSION_MAJOR) {
258         fprintf(stderr, "Unsupported device mapper interface");
259         exit(1);
260     }
261 }
262 
263 /* Variables required by libmultipath and libmpathpersist.  */
264 QEMU_BUILD_BUG_ON(PR_HELPER_DATA_SIZE > MPATH_MAX_PARAM_LEN);
265 static struct config *multipath_conf;
266 unsigned mpath_mx_alloc_len = PR_HELPER_DATA_SIZE;
267 int logsink;
268 struct udev *udev;
269 
270 extern struct config *get_multipath_config(void);
271 struct config *get_multipath_config(void)
272 {
273     return multipath_conf;
274 }
275 
276 extern void put_multipath_config(struct config *conf);
277 void put_multipath_config(struct config *conf)
278 {
279 }
280 
281 static void multipath_pr_init(void)
282 {
283     udev = udev_new();
284 #ifdef CONFIG_MPATH_NEW_API
285     multipath_conf = mpath_lib_init();
286 #else
287     mpath_lib_init(udev);
288 #endif
289 }
290 
291 static int is_mpath(int fd)
292 {
293     struct dm_ioctl dm = { .flags = DM_NOFLUSH_FLAG };
294     struct dm_target_spec *tgt;
295 
296     tgt = dm_dev_ioctl(fd, DM_TABLE_STATUS, &dm);
297     if (!tgt) {
298         if (errno == ENXIO) {
299             return 0;
300         }
301         perror("ioctl");
302         exit(EXIT_FAILURE);
303     }
304     return !strncmp(tgt->target_type, "multipath", DM_MAX_TYPE_NAME);
305 }
306 
307 static SCSISense mpath_generic_sense(int r)
308 {
309     switch (r) {
310     case MPATH_PR_SENSE_NOT_READY:
311          return SENSE_CODE(NOT_READY);
312     case MPATH_PR_SENSE_MEDIUM_ERROR:
313          return SENSE_CODE(READ_ERROR);
314     case MPATH_PR_SENSE_HARDWARE_ERROR:
315          return SENSE_CODE(TARGET_FAILURE);
316     case MPATH_PR_SENSE_ABORTED_COMMAND:
317          return SENSE_CODE(IO_ERROR);
318     default:
319          abort();
320     }
321 }
322 
323 static int mpath_reconstruct_sense(int fd, int r, uint8_t *sense)
324 {
325     switch (r) {
326     case MPATH_PR_SUCCESS:
327         return GOOD;
328     case MPATH_PR_SENSE_NOT_READY:
329     case MPATH_PR_SENSE_MEDIUM_ERROR:
330     case MPATH_PR_SENSE_HARDWARE_ERROR:
331     case MPATH_PR_SENSE_ABORTED_COMMAND:
332         {
333             /* libmpathpersist ate the exact sense.  Try to find it by
334              * issuing TEST UNIT READY.
335              */
336             uint8_t cdb[6] = { TEST_UNIT_READY };
337             int sz = 0;
338             int ret = do_sgio(fd, cdb, sense, NULL, &sz, SG_DXFER_NONE);
339 
340             if (ret != GOOD) {
341                 return ret;
342             }
343             scsi_build_sense(sense, mpath_generic_sense(r));
344             return CHECK_CONDITION;
345         }
346 
347     case MPATH_PR_SENSE_UNIT_ATTENTION:
348         /* Congratulations libmpathpersist, you ruined the Unit Attention...
349          * Return a heavyweight one.
350          */
351         scsi_build_sense(sense, SENSE_CODE(SCSI_BUS_RESET));
352         return CHECK_CONDITION;
353     case MPATH_PR_SENSE_INVALID_OP:
354         /* Only one valid sense.  */
355         scsi_build_sense(sense, SENSE_CODE(INVALID_OPCODE));
356         return CHECK_CONDITION;
357     case MPATH_PR_ILLEGAL_REQ:
358         /* Guess.  */
359         scsi_build_sense(sense, SENSE_CODE(INVALID_PARAM));
360         return CHECK_CONDITION;
361     case MPATH_PR_NO_SENSE:
362         scsi_build_sense(sense, SENSE_CODE(NO_SENSE));
363         return CHECK_CONDITION;
364 
365     case MPATH_PR_RESERV_CONFLICT:
366         return RESERVATION_CONFLICT;
367 
368     case MPATH_PR_OTHER:
369     default:
370         scsi_build_sense(sense, SENSE_CODE(LUN_COMM_FAILURE));
371         return CHECK_CONDITION;
372     }
373 }
374 
375 static int multipath_pr_in(int fd, const uint8_t *cdb, uint8_t *sense,
376                            uint8_t *data, int sz)
377 {
378     int rq_servact = cdb[1];
379     struct prin_resp resp;
380     size_t written;
381     int r;
382 
383     switch (rq_servact) {
384     case MPATH_PRIN_RKEY_SA:
385     case MPATH_PRIN_RRES_SA:
386     case MPATH_PRIN_RCAP_SA:
387         break;
388     case MPATH_PRIN_RFSTAT_SA:
389         /* Nobody implements it anyway, so bail out. */
390     default:
391         /* Cannot parse any other output.  */
392         scsi_build_sense(sense, SENSE_CODE(INVALID_FIELD));
393         return CHECK_CONDITION;
394     }
395 
396     r = mpath_persistent_reserve_in(fd, rq_servact, &resp, noisy, verbose);
397     if (r == MPATH_PR_SUCCESS) {
398         switch (rq_servact) {
399         case MPATH_PRIN_RKEY_SA:
400         case MPATH_PRIN_RRES_SA: {
401             struct prin_readdescr *out = &resp.prin_descriptor.prin_readkeys;
402             assert(sz >= 8);
403             written = MIN(out->additional_length + 8, sz);
404             stl_be_p(&data[0], out->prgeneration);
405             stl_be_p(&data[4], out->additional_length);
406             memcpy(&data[8], out->key_list, written - 8);
407             break;
408         }
409         case MPATH_PRIN_RCAP_SA: {
410             struct prin_capdescr *out = &resp.prin_descriptor.prin_readcap;
411             assert(sz >= 6);
412             written = 6;
413             stw_be_p(&data[0], out->length);
414             data[2] = out->flags[0];
415             data[3] = out->flags[1];
416             stw_be_p(&data[4], out->pr_type_mask);
417             break;
418         }
419         default:
420             scsi_build_sense(sense, SENSE_CODE(INVALID_OPCODE));
421             return CHECK_CONDITION;
422         }
423         assert(written <= sz);
424         memset(data + written, 0, sz - written);
425     }
426 
427     return mpath_reconstruct_sense(fd, r, sense);
428 }
429 
430 static int multipath_pr_out(int fd, const uint8_t *cdb, uint8_t *sense,
431                             const uint8_t *param, int sz)
432 {
433     int rq_servact = cdb[1];
434     int rq_scope = cdb[2] >> 4;
435     int rq_type = cdb[2] & 0xf;
436     g_autofree struct prout_param_descriptor *paramp = NULL;
437     char transportids[PR_HELPER_DATA_SIZE];
438     int r;
439 
440     paramp = g_malloc0(sizeof(struct prout_param_descriptor)
441                        + sizeof(struct transportid *) * MPATH_MX_TIDS);
442 
443     if (sz < PR_OUT_FIXED_PARAM_SIZE) {
444         /* Illegal request, Parameter list length error.  This isn't fatal;
445          * we have read the data, send an error without closing the socket.
446          */
447         scsi_build_sense(sense, SENSE_CODE(INVALID_PARAM_LEN));
448         return CHECK_CONDITION;
449     }
450 
451     switch (rq_servact) {
452     case MPATH_PROUT_REG_SA:
453     case MPATH_PROUT_RES_SA:
454     case MPATH_PROUT_REL_SA:
455     case MPATH_PROUT_CLEAR_SA:
456     case MPATH_PROUT_PREE_SA:
457     case MPATH_PROUT_PREE_AB_SA:
458     case MPATH_PROUT_REG_IGN_SA:
459         break;
460     case MPATH_PROUT_REG_MOV_SA:
461         /* Not supported by struct prout_param_descriptor.  */
462     default:
463         /* Cannot parse any other input.  */
464         scsi_build_sense(sense, SENSE_CODE(INVALID_FIELD));
465         return CHECK_CONDITION;
466     }
467 
468     /* Convert input data, especially transport IDs, to the structs
469      * used by libmpathpersist (which, of course, will immediately
470      * do the opposite).
471      */
472     memcpy(&paramp->key, &param[0], 8);
473     memcpy(&paramp->sa_key, &param[8], 8);
474     paramp->sa_flags = param[20];
475     if (sz > PR_OUT_FIXED_PARAM_SIZE) {
476         size_t transportid_len;
477         int i, j;
478         if (sz < PR_OUT_FIXED_PARAM_SIZE + 4) {
479             scsi_build_sense(sense, SENSE_CODE(INVALID_PARAM_LEN));
480             return CHECK_CONDITION;
481         }
482         transportid_len = ldl_be_p(&param[24]) + PR_OUT_FIXED_PARAM_SIZE + 4;
483         if (transportid_len > sz) {
484             scsi_build_sense(sense, SENSE_CODE(INVALID_PARAM));
485             return CHECK_CONDITION;
486         }
487         for (i = PR_OUT_FIXED_PARAM_SIZE + 4, j = 0; i < transportid_len; ) {
488             struct transportid *id = (struct transportid *) &transportids[j];
489             int len;
490 
491             id->format_code = param[i] & 0xc0;
492             id->protocol_id = param[i] & 0x0f;
493             switch (param[i] & 0xcf) {
494             case 0:
495                 /* FC transport.  */
496                 if (i + 24 > transportid_len) {
497                     goto illegal_req;
498                 }
499                 memcpy(id->n_port_name, &param[i + 8], 8);
500                 j += offsetof(struct transportid, n_port_name[8]);
501                 i += 24;
502                 break;
503             case 5:
504             case 0x45:
505                 /* iSCSI transport.  */
506                 len = lduw_be_p(&param[i + 2]);
507                 if (len > 252 || (len & 3) || i + len + 4 > transportid_len) {
508                     /* For format code 00, the standard says the maximum is 223
509                      * plus the NUL terminator.  For format code 01 there is no
510                      * maximum length, but libmpathpersist ignores the first
511                      * byte of id->iscsi_name so our maximum is 252.
512                      */
513                     goto illegal_req;
514                 }
515                 if (memchr(&param[i + 4], 0, len) == NULL) {
516                     goto illegal_req;
517                 }
518                 memcpy(id->iscsi_name, &param[i + 2], len + 2);
519                 j += offsetof(struct transportid, iscsi_name[len + 2]);
520                 i += len + 4;
521                 break;
522             case 6:
523                 /* SAS transport.  */
524                 if (i + 24 > transportid_len) {
525                     goto illegal_req;
526                 }
527                 memcpy(id->sas_address, &param[i + 4], 8);
528                 j += offsetof(struct transportid, sas_address[8]);
529                 i += 24;
530                 break;
531             default:
532             illegal_req:
533                 scsi_build_sense(sense, SENSE_CODE(INVALID_PARAM));
534                 return CHECK_CONDITION;
535             }
536 
537             assert(paramp->num_transportid < MPATH_MX_TIDS);
538             paramp->trnptid_list[paramp->num_transportid++] = id;
539         }
540     }
541 
542     r = mpath_persistent_reserve_out(fd, rq_servact, rq_scope, rq_type,
543                                      paramp, noisy, verbose);
544     return mpath_reconstruct_sense(fd, r, sense);
545 }
546 #endif
547 
548 static int do_pr_in(int fd, const uint8_t *cdb, uint8_t *sense,
549                     uint8_t *data, int *resp_sz)
550 {
551 #ifdef CONFIG_MPATH
552     if (is_mpath(fd)) {
553         /* multipath_pr_in fills the whole input buffer.  */
554         int r = multipath_pr_in(fd, cdb, sense, data, *resp_sz);
555         if (r != GOOD) {
556             *resp_sz = 0;
557         }
558         return r;
559     }
560 #endif
561 
562     return do_sgio(fd, cdb, sense, data, resp_sz,
563                    SG_DXFER_FROM_DEV);
564 }
565 
566 static int do_pr_out(int fd, const uint8_t *cdb, uint8_t *sense,
567                      const uint8_t *param, int sz)
568 {
569     int resp_sz;
570 
571     if ((fcntl(fd, F_GETFL) & O_ACCMODE) == O_RDONLY) {
572         scsi_build_sense(sense, SENSE_CODE(INVALID_OPCODE));
573         return CHECK_CONDITION;
574     }
575 
576 #ifdef CONFIG_MPATH
577     if (is_mpath(fd)) {
578         return multipath_pr_out(fd, cdb, sense, param, sz);
579     }
580 #endif
581 
582     resp_sz = sz;
583     return do_sgio(fd, cdb, sense, (uint8_t *)param, &resp_sz,
584                    SG_DXFER_TO_DEV);
585 }
586 
587 /* Client */
588 
589 typedef struct PRHelperClient {
590     QIOChannelSocket *ioc;
591     Coroutine *co;
592     int fd;
593     uint8_t data[PR_HELPER_DATA_SIZE];
594 } PRHelperClient;
595 
596 typedef struct PRHelperRequest {
597     int fd;
598     size_t sz;
599     uint8_t cdb[PR_HELPER_CDB_SIZE];
600 } PRHelperRequest;
601 
602 static int coroutine_fn prh_read(PRHelperClient *client, void *buf, int sz,
603                                  Error **errp)
604 {
605     int ret = 0;
606 
607     while (sz > 0) {
608         int *fds = NULL;
609         size_t nfds = 0;
610         int i;
611         struct iovec iov;
612         ssize_t n_read;
613 
614         iov.iov_base = buf;
615         iov.iov_len = sz;
616         n_read = qio_channel_readv_full(QIO_CHANNEL(client->ioc), &iov, 1,
617                                         &fds, &nfds, 0, errp);
618 
619         if (n_read == QIO_CHANNEL_ERR_BLOCK) {
620             qio_channel_yield(QIO_CHANNEL(client->ioc), G_IO_IN);
621             continue;
622         }
623         if (n_read <= 0) {
624             ret = n_read ? n_read : -1;
625             goto err;
626         }
627 
628         /* Stash one file descriptor per request.  */
629         if (nfds) {
630             bool too_many = false;
631             for (i = 0; i < nfds; i++) {
632                 if (client->fd == -1) {
633                     client->fd = fds[i];
634                 } else {
635                     close(fds[i]);
636                     too_many = true;
637                 }
638             }
639             g_free(fds);
640             if (too_many) {
641                 ret = -1;
642                 goto err;
643             }
644         }
645 
646         buf += n_read;
647         sz -= n_read;
648     }
649 
650     return 0;
651 
652 err:
653     if (client->fd != -1) {
654         close(client->fd);
655         client->fd = -1;
656     }
657     return ret;
658 }
659 
660 static int coroutine_fn prh_read_request(PRHelperClient *client,
661                                          PRHelperRequest *req,
662                                          PRHelperResponse *resp, Error **errp)
663 {
664     uint32_t sz;
665 
666     if (prh_read(client, req->cdb, sizeof(req->cdb), NULL) < 0) {
667         return -1;
668     }
669 
670     if (client->fd == -1) {
671         error_setg(errp, "No file descriptor in request.");
672         return -1;
673     }
674 
675     if (req->cdb[0] != PERSISTENT_RESERVE_OUT &&
676         req->cdb[0] != PERSISTENT_RESERVE_IN) {
677         error_setg(errp, "Invalid CDB, closing socket.");
678         goto out_close;
679     }
680 
681     sz = scsi_cdb_xfer(req->cdb);
682     if (sz > sizeof(client->data)) {
683         goto out_close;
684     }
685 
686     if (req->cdb[0] == PERSISTENT_RESERVE_OUT) {
687         if (qio_channel_read_all(QIO_CHANNEL(client->ioc),
688                                  (char *)client->data, sz,
689                                  errp) < 0) {
690             goto out_close;
691         }
692     }
693 
694     req->fd = client->fd;
695     req->sz = sz;
696     client->fd = -1;
697     return sz;
698 
699 out_close:
700     close(client->fd);
701     client->fd = -1;
702     return -1;
703 }
704 
705 static int coroutine_fn prh_write_response(PRHelperClient *client,
706                                            PRHelperRequest *req,
707                                            PRHelperResponse *resp, Error **errp)
708 {
709     ssize_t r;
710     size_t sz;
711 
712     if (req->cdb[0] == PERSISTENT_RESERVE_IN && resp->result == GOOD) {
713         assert(resp->sz <= req->sz && resp->sz <= sizeof(client->data));
714     } else {
715         assert(resp->sz == 0);
716     }
717 
718     sz = resp->sz;
719 
720     resp->result = cpu_to_be32(resp->result);
721     resp->sz = cpu_to_be32(resp->sz);
722     r = qio_channel_write_all(QIO_CHANNEL(client->ioc),
723                               (char *) resp, sizeof(*resp), errp);
724     if (r < 0) {
725         return r;
726     }
727 
728     r = qio_channel_write_all(QIO_CHANNEL(client->ioc),
729                               (char *) client->data,
730                               sz, errp);
731     return r < 0 ? r : 0;
732 }
733 
734 static void coroutine_fn prh_co_entry(void *opaque)
735 {
736     PRHelperClient *client = opaque;
737     Error *local_err = NULL;
738     uint32_t flags;
739     int r;
740 
741     qio_channel_set_blocking(QIO_CHANNEL(client->ioc),
742                              false, NULL);
743     qio_channel_attach_aio_context(QIO_CHANNEL(client->ioc),
744                                    qemu_get_aio_context());
745 
746     /* A very simple negotiation for future extensibility.  No features
747      * are defined so write 0.
748      */
749     flags = cpu_to_be32(0);
750     r = qio_channel_write_all(QIO_CHANNEL(client->ioc),
751                              (char *) &flags, sizeof(flags), NULL);
752     if (r < 0) {
753         goto out;
754     }
755 
756     r = qio_channel_read_all(QIO_CHANNEL(client->ioc),
757                              (char *) &flags, sizeof(flags), NULL);
758     if (be32_to_cpu(flags) != 0 || r < 0) {
759         goto out;
760     }
761 
762     while (qatomic_read(&state) == RUNNING) {
763         PRHelperRequest req;
764         PRHelperResponse resp;
765         int sz;
766 
767         sz = prh_read_request(client, &req, &resp, &local_err);
768         if (sz < 0) {
769             break;
770         }
771 
772         num_active_sockets++;
773         if (req.cdb[0] == PERSISTENT_RESERVE_OUT) {
774             r = do_pr_out(req.fd, req.cdb, resp.sense,
775                           client->data, sz);
776             resp.sz = 0;
777         } else {
778             resp.sz = sizeof(client->data);
779             r = do_pr_in(req.fd, req.cdb, resp.sense,
780                          client->data, &resp.sz);
781             resp.sz = MIN(resp.sz, sz);
782         }
783         num_active_sockets--;
784         close(req.fd);
785         if (r == -1) {
786             break;
787         }
788         resp.result = r;
789 
790         if (prh_write_response(client, &req, &resp, &local_err) < 0) {
791             break;
792         }
793     }
794 
795     if (local_err) {
796         if (verbose == 0) {
797             error_free(local_err);
798         } else {
799             error_report_err(local_err);
800         }
801     }
802 
803 out:
804     qio_channel_detach_aio_context(QIO_CHANNEL(client->ioc));
805     object_unref(OBJECT(client->ioc));
806     g_free(client);
807 }
808 
809 static gboolean accept_client(QIOChannel *ioc, GIOCondition cond, gpointer opaque)
810 {
811     QIOChannelSocket *cioc;
812     PRHelperClient *prh;
813 
814     cioc = qio_channel_socket_accept(QIO_CHANNEL_SOCKET(ioc),
815                                      NULL);
816     if (!cioc) {
817         return TRUE;
818     }
819 
820     prh = g_new(PRHelperClient, 1);
821     prh->ioc = cioc;
822     prh->fd = -1;
823     prh->co = qemu_coroutine_create(prh_co_entry, prh);
824     qemu_coroutine_enter(prh->co);
825 
826     return TRUE;
827 }
828 
829 static void termsig_handler(int signum)
830 {
831     qatomic_cmpxchg(&state, RUNNING, TERMINATE);
832     qemu_notify_event();
833 }
834 
835 static void close_server_socket(void)
836 {
837     assert(server_ioc);
838 
839     g_source_remove(server_watch);
840     server_watch = -1;
841     object_unref(OBJECT(server_ioc));
842     num_active_sockets--;
843 }
844 
845 #ifdef CONFIG_LIBCAP_NG
846 static int drop_privileges(void)
847 {
848     /* clear all capabilities */
849     capng_clear(CAPNG_SELECT_BOTH);
850 
851     if (capng_update(CAPNG_ADD, CAPNG_EFFECTIVE | CAPNG_PERMITTED,
852                      CAP_SYS_RAWIO) < 0) {
853         return -1;
854     }
855 
856 #ifdef CONFIG_MPATH
857     /* For /dev/mapper/control ioctls */
858     if (capng_update(CAPNG_ADD, CAPNG_EFFECTIVE | CAPNG_PERMITTED,
859                      CAP_SYS_ADMIN) < 0) {
860         return -1;
861     }
862 #endif
863 
864     /* Change user/group id, retaining the capabilities.  Because file descriptors
865      * are passed via SCM_RIGHTS, we don't need supplementary groups (and in
866      * fact the helper can run as "nobody").
867      */
868     if (capng_change_id(uid != -1 ? uid : getuid(),
869                         gid != -1 ? gid : getgid(),
870                         CAPNG_DROP_SUPP_GRP | CAPNG_CLEAR_BOUNDING)) {
871         return -1;
872     }
873 
874     return 0;
875 }
876 #endif
877 
878 int main(int argc, char **argv)
879 {
880     const char *sopt = "hVk:f:dT:u:g:vq";
881     struct option lopt[] = {
882         { "help", no_argument, NULL, 'h' },
883         { "version", no_argument, NULL, 'V' },
884         { "socket", required_argument, NULL, 'k' },
885         { "pidfile", required_argument, NULL, 'f' },
886         { "daemon", no_argument, NULL, 'd' },
887         { "trace", required_argument, NULL, 'T' },
888         { "user", required_argument, NULL, 'u' },
889         { "group", required_argument, NULL, 'g' },
890         { "verbose", no_argument, NULL, 'v' },
891         { "quiet", no_argument, NULL, 'q' },
892         { NULL, 0, NULL, 0 }
893     };
894     int opt_ind = 0;
895     int loglevel = 1;
896     int quiet = 0;
897     int ch;
898     Error *local_err = NULL;
899     bool daemonize = false;
900     bool pidfile_specified = false;
901     bool socket_path_specified = false;
902     unsigned socket_activation;
903 
904     struct sigaction sa_sigterm;
905     memset(&sa_sigterm, 0, sizeof(sa_sigterm));
906     sa_sigterm.sa_handler = termsig_handler;
907     sigaction(SIGTERM, &sa_sigterm, NULL);
908     sigaction(SIGINT, &sa_sigterm, NULL);
909     sigaction(SIGHUP, &sa_sigterm, NULL);
910 
911     signal(SIGPIPE, SIG_IGN);
912 
913     error_init(argv[0]);
914     module_call_init(MODULE_INIT_TRACE);
915     module_call_init(MODULE_INIT_QOM);
916     qemu_add_opts(&qemu_trace_opts);
917     qemu_init_exec_dir(argv[0]);
918 
919     compute_default_paths();
920 
921     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
922         switch (ch) {
923         case 'k':
924             g_free(socket_path);
925             socket_path = g_strdup(optarg);
926             socket_path_specified = true;
927             if (socket_path[0] != '/') {
928                 error_report("socket path must be absolute");
929                 exit(EXIT_FAILURE);
930             }
931             break;
932         case 'f':
933             g_free(pidfile);
934             pidfile = g_strdup(optarg);
935             pidfile_specified = true;
936             break;
937 #ifdef CONFIG_LIBCAP_NG
938         case 'u': {
939             unsigned long res;
940             struct passwd *userinfo = getpwnam(optarg);
941             if (userinfo) {
942                 uid = userinfo->pw_uid;
943             } else if (qemu_strtoul(optarg, NULL, 10, &res) == 0 &&
944                        (uid_t)res == res) {
945                 uid = res;
946             } else {
947                 error_report("invalid user '%s'", optarg);
948                 exit(EXIT_FAILURE);
949             }
950             break;
951         }
952         case 'g': {
953             unsigned long res;
954             struct group *groupinfo = getgrnam(optarg);
955             if (groupinfo) {
956                 gid = groupinfo->gr_gid;
957             } else if (qemu_strtoul(optarg, NULL, 10, &res) == 0 &&
958                        (gid_t)res == res) {
959                 gid = res;
960             } else {
961                 error_report("invalid group '%s'", optarg);
962                 exit(EXIT_FAILURE);
963             }
964             break;
965         }
966 #else
967         case 'u':
968         case 'g':
969             error_report("-%c not supported by this %s", ch, argv[0]);
970             exit(1);
971 #endif
972         case 'd':
973             daemonize = true;
974             break;
975         case 'q':
976             quiet = 1;
977             break;
978         case 'v':
979             ++loglevel;
980             break;
981         case 'T':
982             trace_opt_parse(optarg);
983             break;
984         case 'V':
985             version(argv[0]);
986             exit(EXIT_SUCCESS);
987             break;
988         case 'h':
989             usage(argv[0]);
990             exit(EXIT_SUCCESS);
991             break;
992         case '?':
993             error_report("Try `%s --help' for more information.", argv[0]);
994             exit(EXIT_FAILURE);
995         }
996     }
997 
998     /* set verbosity */
999     noisy = !quiet && (loglevel >= 3);
1000     verbose = quiet ? 0 : MIN(loglevel, 3);
1001 
1002     if (!trace_init_backends()) {
1003         exit(EXIT_FAILURE);
1004     }
1005     trace_init_file();
1006     qemu_set_log(LOG_TRACE, &error_fatal);
1007 
1008 #ifdef CONFIG_MPATH
1009     dm_init();
1010     multipath_pr_init();
1011 #endif
1012 
1013     socket_activation = check_socket_activation();
1014     if (socket_activation == 0) {
1015         SocketAddress saddr;
1016         saddr = (SocketAddress){
1017             .type = SOCKET_ADDRESS_TYPE_UNIX,
1018             .u.q_unix.path = socket_path,
1019         };
1020         server_ioc = qio_channel_socket_new();
1021         if (qio_channel_socket_listen_sync(server_ioc, &saddr,
1022                                            1, &local_err) < 0) {
1023             object_unref(OBJECT(server_ioc));
1024             error_report_err(local_err);
1025             return 1;
1026         }
1027     } else {
1028         /* Using socket activation - check user didn't use -p etc. */
1029         if (socket_path_specified) {
1030             error_report("Unix socket can't be set when using socket activation");
1031             exit(EXIT_FAILURE);
1032         }
1033 
1034         /* Can only listen on a single socket.  */
1035         if (socket_activation > 1) {
1036             error_report("%s does not support socket activation with LISTEN_FDS > 1",
1037                          argv[0]);
1038             exit(EXIT_FAILURE);
1039         }
1040         server_ioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD,
1041                                                &local_err);
1042         if (server_ioc == NULL) {
1043             error_reportf_err(local_err,
1044                               "Failed to use socket activation: ");
1045             exit(EXIT_FAILURE);
1046         }
1047     }
1048 
1049     qemu_init_main_loop(&error_fatal);
1050 
1051     server_watch = qio_channel_add_watch(QIO_CHANNEL(server_ioc),
1052                                          G_IO_IN,
1053                                          accept_client,
1054                                          NULL, NULL);
1055 
1056     if (daemonize) {
1057         if (daemon(0, 0) < 0) {
1058             error_report("Failed to daemonize: %s", strerror(errno));
1059             exit(EXIT_FAILURE);
1060         }
1061     }
1062 
1063     if (daemonize || pidfile_specified) {
1064         qemu_write_pidfile(pidfile, &error_fatal);
1065     }
1066 
1067 #ifdef CONFIG_LIBCAP_NG
1068     if (drop_privileges() < 0) {
1069         error_report("Failed to drop privileges: %s", strerror(errno));
1070         exit(EXIT_FAILURE);
1071     }
1072 #endif
1073 
1074     state = RUNNING;
1075     do {
1076         main_loop_wait(false);
1077         if (state == TERMINATE) {
1078             state = TERMINATING;
1079             close_server_socket();
1080         }
1081     } while (num_active_sockets > 0);
1082 
1083     exit(EXIT_SUCCESS);
1084 }
1085