xref: /openbmc/qemu/scsi/qemu-pr-helper.c (revision fc8c745d)
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-common.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     socket_path = qemu_get_local_state_pathname("run/qemu-pr-helper.sock");
81     pidfile = qemu_get_local_state_pathname("run/qemu-pr-helper.pid");
82 }
83 
84 static void usage(const char *name)
85 {
86     (printf) (
87 "Usage: %s [OPTIONS] FILE\n"
88 "Persistent Reservation helper program for QEMU\n"
89 "\n"
90 "  -h, --help                display this help and exit\n"
91 "  -V, --version             output version information and exit\n"
92 "\n"
93 "  -d, --daemon              run in the background\n"
94 "  -f, --pidfile=PATH        PID file when running as a daemon\n"
95 "                            (default '%s')\n"
96 "  -k, --socket=PATH         path to the unix socket\n"
97 "                            (default '%s')\n"
98 "  -T, --trace [[enable=]<pattern>][,events=<file>][,file=<file>]\n"
99 "                            specify tracing options\n"
100 #ifdef CONFIG_LIBCAP_NG
101 "  -u, --user=USER           user to drop privileges to\n"
102 "  -g, --group=GROUP         group to drop privileges to\n"
103 #endif
104 "\n"
105 QEMU_HELP_BOTTOM "\n"
106     , name, pidfile, socket_path);
107 }
108 
109 static void version(const char *name)
110 {
111     printf(
112 "%s " QEMU_FULL_VERSION "\n"
113 "Written by Paolo Bonzini.\n"
114 "\n"
115 QEMU_COPYRIGHT "\n"
116 "This is free software; see the source for copying conditions.  There is NO\n"
117 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
118     , name);
119 }
120 
121 /* SG_IO support */
122 
123 typedef struct PRHelperSGIOData {
124     int fd;
125     const uint8_t *cdb;
126     uint8_t *sense;
127     uint8_t *buf;
128     int sz;              /* input/output */
129     int dir;
130 } PRHelperSGIOData;
131 
132 static int do_sgio_worker(void *opaque)
133 {
134     PRHelperSGIOData *data = opaque;
135     struct sg_io_hdr io_hdr;
136     int ret;
137     int status;
138     SCSISense sense_code;
139 
140     memset(data->sense, 0, PR_HELPER_SENSE_SIZE);
141     memset(&io_hdr, 0, sizeof(io_hdr));
142     io_hdr.interface_id = 'S';
143     io_hdr.cmd_len = PR_HELPER_CDB_SIZE;
144     io_hdr.cmdp = (uint8_t *)data->cdb;
145     io_hdr.sbp = data->sense;
146     io_hdr.mx_sb_len = PR_HELPER_SENSE_SIZE;
147     io_hdr.timeout = 1;
148     io_hdr.dxfer_direction = data->dir;
149     io_hdr.dxferp = (char *)data->buf;
150     io_hdr.dxfer_len = data->sz;
151     ret = ioctl(data->fd, SG_IO, &io_hdr);
152 
153     if (ret < 0) {
154         status = scsi_sense_from_errno(errno, &sense_code);
155         if (status == CHECK_CONDITION) {
156             scsi_build_sense(data->sense, sense_code);
157         }
158     } else if (io_hdr.host_status != SCSI_HOST_OK) {
159         status = scsi_sense_from_host_status(io_hdr.host_status, &sense_code);
160         if (status == CHECK_CONDITION) {
161             scsi_build_sense(data->sense, sense_code);
162         }
163     } else if (io_hdr.driver_status & SG_ERR_DRIVER_TIMEOUT) {
164         status = BUSY;
165     } else {
166         status = io_hdr.status;
167     }
168 
169     if (status == GOOD) {
170         data->sz -= io_hdr.resid;
171     } else {
172         data->sz = 0;
173     }
174 
175     return status;
176 }
177 
178 static int do_sgio(int fd, const uint8_t *cdb, uint8_t *sense,
179                     uint8_t *buf, int *sz, int dir)
180 {
181     ThreadPool *pool = aio_get_thread_pool(qemu_get_aio_context());
182     int r;
183 
184     PRHelperSGIOData data = {
185         .fd = fd,
186         .cdb = cdb,
187         .sense = sense,
188         .buf = buf,
189         .sz = *sz,
190         .dir = dir,
191     };
192 
193     r = thread_pool_submit_co(pool, do_sgio_worker, &data);
194     *sz = data.sz;
195     return r;
196 }
197 
198 /* Device mapper interface */
199 
200 #ifdef CONFIG_MPATH
201 #define CONTROL_PATH "/dev/mapper/control"
202 
203 typedef struct DMData {
204     struct dm_ioctl dm;
205     uint8_t data[1024];
206 } DMData;
207 
208 static int control_fd;
209 
210 static void *dm_ioctl(int ioc, struct dm_ioctl *dm)
211 {
212     static DMData d;
213     memcpy(&d.dm, dm, sizeof(d.dm));
214     QEMU_BUILD_BUG_ON(sizeof(d.data) < sizeof(struct dm_target_spec));
215 
216     d.dm.version[0] = DM_VERSION_MAJOR;
217     d.dm.version[1] = 0;
218     d.dm.version[2] = 0;
219     d.dm.data_size = 1024;
220     d.dm.data_start = offsetof(DMData, data);
221     if (ioctl(control_fd, ioc, &d) < 0) {
222         return NULL;
223     }
224     memcpy(dm, &d.dm, sizeof(d.dm));
225     return &d.data;
226 }
227 
228 static void *dm_dev_ioctl(int fd, int ioc, struct dm_ioctl *dm)
229 {
230     struct stat st;
231     int r;
232 
233     r = fstat(fd, &st);
234     if (r < 0) {
235         perror("fstat");
236         exit(1);
237     }
238 
239     dm->dev = st.st_rdev;
240     return dm_ioctl(ioc, dm);
241 }
242 
243 static void dm_init(void)
244 {
245     control_fd = open(CONTROL_PATH, O_RDWR);
246     if (control_fd < 0) {
247         perror("Cannot open " CONTROL_PATH);
248         exit(1);
249     }
250     struct dm_ioctl dm = { };
251     if (!dm_ioctl(DM_VERSION, &dm)) {
252         perror("ioctl");
253         exit(1);
254     }
255     if (dm.version[0] != DM_VERSION_MAJOR) {
256         fprintf(stderr, "Unsupported device mapper interface");
257         exit(1);
258     }
259 }
260 
261 /* Variables required by libmultipath and libmpathpersist.  */
262 QEMU_BUILD_BUG_ON(PR_HELPER_DATA_SIZE > MPATH_MAX_PARAM_LEN);
263 static struct config *multipath_conf;
264 unsigned mpath_mx_alloc_len = PR_HELPER_DATA_SIZE;
265 int logsink;
266 struct udev *udev;
267 
268 extern struct config *get_multipath_config(void);
269 struct config *get_multipath_config(void)
270 {
271     return multipath_conf;
272 }
273 
274 extern void put_multipath_config(struct config *conf);
275 void put_multipath_config(struct config *conf)
276 {
277 }
278 
279 static void multipath_pr_init(void)
280 {
281     udev = udev_new();
282 #ifdef CONFIG_MPATH_NEW_API
283     multipath_conf = mpath_lib_init();
284 #else
285     mpath_lib_init(udev);
286 #endif
287 }
288 
289 static int is_mpath(int fd)
290 {
291     struct dm_ioctl dm = { .flags = DM_NOFLUSH_FLAG };
292     struct dm_target_spec *tgt;
293 
294     tgt = dm_dev_ioctl(fd, DM_TABLE_STATUS, &dm);
295     if (!tgt) {
296         if (errno == ENXIO) {
297             return 0;
298         }
299         perror("ioctl");
300         exit(EXIT_FAILURE);
301     }
302     return !strncmp(tgt->target_type, "multipath", DM_MAX_TYPE_NAME);
303 }
304 
305 static SCSISense mpath_generic_sense(int r)
306 {
307     switch (r) {
308     case MPATH_PR_SENSE_NOT_READY:
309          return SENSE_CODE(NOT_READY);
310     case MPATH_PR_SENSE_MEDIUM_ERROR:
311          return SENSE_CODE(READ_ERROR);
312     case MPATH_PR_SENSE_HARDWARE_ERROR:
313          return SENSE_CODE(TARGET_FAILURE);
314     case MPATH_PR_SENSE_ABORTED_COMMAND:
315          return SENSE_CODE(IO_ERROR);
316     default:
317          abort();
318     }
319 }
320 
321 static int mpath_reconstruct_sense(int fd, int r, uint8_t *sense)
322 {
323     switch (r) {
324     case MPATH_PR_SUCCESS:
325         return GOOD;
326     case MPATH_PR_SENSE_NOT_READY:
327     case MPATH_PR_SENSE_MEDIUM_ERROR:
328     case MPATH_PR_SENSE_HARDWARE_ERROR:
329     case MPATH_PR_SENSE_ABORTED_COMMAND:
330         {
331             /* libmpathpersist ate the exact sense.  Try to find it by
332              * issuing TEST UNIT READY.
333              */
334             uint8_t cdb[6] = { TEST_UNIT_READY };
335             int sz = 0;
336             int ret = do_sgio(fd, cdb, sense, NULL, &sz, SG_DXFER_NONE);
337 
338             if (ret != GOOD) {
339                 return ret;
340             }
341             scsi_build_sense(sense, mpath_generic_sense(r));
342             return CHECK_CONDITION;
343         }
344 
345     case MPATH_PR_SENSE_UNIT_ATTENTION:
346         /* Congratulations libmpathpersist, you ruined the Unit Attention...
347          * Return a heavyweight one.
348          */
349         scsi_build_sense(sense, SENSE_CODE(SCSI_BUS_RESET));
350         return CHECK_CONDITION;
351     case MPATH_PR_SENSE_INVALID_OP:
352         /* Only one valid sense.  */
353         scsi_build_sense(sense, SENSE_CODE(INVALID_OPCODE));
354         return CHECK_CONDITION;
355     case MPATH_PR_ILLEGAL_REQ:
356         /* Guess.  */
357         scsi_build_sense(sense, SENSE_CODE(INVALID_PARAM));
358         return CHECK_CONDITION;
359     case MPATH_PR_NO_SENSE:
360         scsi_build_sense(sense, SENSE_CODE(NO_SENSE));
361         return CHECK_CONDITION;
362 
363     case MPATH_PR_RESERV_CONFLICT:
364         return RESERVATION_CONFLICT;
365 
366     case MPATH_PR_OTHER:
367     default:
368         scsi_build_sense(sense, SENSE_CODE(LUN_COMM_FAILURE));
369         return CHECK_CONDITION;
370     }
371 }
372 
373 static int multipath_pr_in(int fd, const uint8_t *cdb, uint8_t *sense,
374                            uint8_t *data, int sz)
375 {
376     int rq_servact = cdb[1];
377     struct prin_resp resp;
378     size_t written;
379     int r;
380 
381     switch (rq_servact) {
382     case MPATH_PRIN_RKEY_SA:
383     case MPATH_PRIN_RRES_SA:
384     case MPATH_PRIN_RCAP_SA:
385         break;
386     case MPATH_PRIN_RFSTAT_SA:
387         /* Nobody implements it anyway, so bail out. */
388     default:
389         /* Cannot parse any other output.  */
390         scsi_build_sense(sense, SENSE_CODE(INVALID_FIELD));
391         return CHECK_CONDITION;
392     }
393 
394     r = mpath_persistent_reserve_in(fd, rq_servact, &resp, noisy, verbose);
395     if (r == MPATH_PR_SUCCESS) {
396         switch (rq_servact) {
397         case MPATH_PRIN_RKEY_SA:
398         case MPATH_PRIN_RRES_SA: {
399             struct prin_readdescr *out = &resp.prin_descriptor.prin_readkeys;
400             assert(sz >= 8);
401             written = MIN(out->additional_length + 8, sz);
402             stl_be_p(&data[0], out->prgeneration);
403             stl_be_p(&data[4], out->additional_length);
404             memcpy(&data[8], out->key_list, written - 8);
405             break;
406         }
407         case MPATH_PRIN_RCAP_SA: {
408             struct prin_capdescr *out = &resp.prin_descriptor.prin_readcap;
409             assert(sz >= 6);
410             written = 6;
411             stw_be_p(&data[0], out->length);
412             data[2] = out->flags[0];
413             data[3] = out->flags[1];
414             stw_be_p(&data[4], out->pr_type_mask);
415             break;
416         }
417         default:
418             scsi_build_sense(sense, SENSE_CODE(INVALID_OPCODE));
419             return CHECK_CONDITION;
420         }
421         assert(written <= sz);
422         memset(data + written, 0, sz - written);
423     }
424 
425     return mpath_reconstruct_sense(fd, r, sense);
426 }
427 
428 static int multipath_pr_out(int fd, const uint8_t *cdb, uint8_t *sense,
429                             const uint8_t *param, int sz)
430 {
431     int rq_servact = cdb[1];
432     int rq_scope = cdb[2] >> 4;
433     int rq_type = cdb[2] & 0xf;
434     g_autofree struct prout_param_descriptor *paramp = NULL;
435     char transportids[PR_HELPER_DATA_SIZE];
436     int r;
437 
438     paramp = g_malloc0(sizeof(struct prout_param_descriptor)
439                        + sizeof(struct transportid *) * MPATH_MX_TIDS);
440 
441     if (sz < PR_OUT_FIXED_PARAM_SIZE) {
442         /* Illegal request, Parameter list length error.  This isn't fatal;
443          * we have read the data, send an error without closing the socket.
444          */
445         scsi_build_sense(sense, SENSE_CODE(INVALID_PARAM_LEN));
446         return CHECK_CONDITION;
447     }
448 
449     switch (rq_servact) {
450     case MPATH_PROUT_REG_SA:
451     case MPATH_PROUT_RES_SA:
452     case MPATH_PROUT_REL_SA:
453     case MPATH_PROUT_CLEAR_SA:
454     case MPATH_PROUT_PREE_SA:
455     case MPATH_PROUT_PREE_AB_SA:
456     case MPATH_PROUT_REG_IGN_SA:
457         break;
458     case MPATH_PROUT_REG_MOV_SA:
459         /* Not supported by struct prout_param_descriptor.  */
460     default:
461         /* Cannot parse any other input.  */
462         scsi_build_sense(sense, SENSE_CODE(INVALID_FIELD));
463         return CHECK_CONDITION;
464     }
465 
466     /* Convert input data, especially transport IDs, to the structs
467      * used by libmpathpersist (which, of course, will immediately
468      * do the opposite).
469      */
470     memcpy(&paramp->key, &param[0], 8);
471     memcpy(&paramp->sa_key, &param[8], 8);
472     paramp->sa_flags = param[20];
473     if (sz > PR_OUT_FIXED_PARAM_SIZE) {
474         size_t transportid_len;
475         int i, j;
476         if (sz < PR_OUT_FIXED_PARAM_SIZE + 4) {
477             scsi_build_sense(sense, SENSE_CODE(INVALID_PARAM_LEN));
478             return CHECK_CONDITION;
479         }
480         transportid_len = ldl_be_p(&param[24]) + PR_OUT_FIXED_PARAM_SIZE + 4;
481         if (transportid_len > sz) {
482             scsi_build_sense(sense, SENSE_CODE(INVALID_PARAM));
483             return CHECK_CONDITION;
484         }
485         for (i = PR_OUT_FIXED_PARAM_SIZE + 4, j = 0; i < transportid_len; ) {
486             struct transportid *id = (struct transportid *) &transportids[j];
487             int len;
488 
489             id->format_code = param[i] & 0xc0;
490             id->protocol_id = param[i] & 0x0f;
491             switch (param[i] & 0xcf) {
492             case 0:
493                 /* FC transport.  */
494                 if (i + 24 > transportid_len) {
495                     goto illegal_req;
496                 }
497                 memcpy(id->n_port_name, &param[i + 8], 8);
498                 j += offsetof(struct transportid, n_port_name[8]);
499                 i += 24;
500                 break;
501             case 5:
502             case 0x45:
503                 /* iSCSI transport.  */
504                 len = lduw_be_p(&param[i + 2]);
505                 if (len > 252 || (len & 3) || i + len + 4 > transportid_len) {
506                     /* For format code 00, the standard says the maximum is 223
507                      * plus the NUL terminator.  For format code 01 there is no
508                      * maximum length, but libmpathpersist ignores the first
509                      * byte of id->iscsi_name so our maximum is 252.
510                      */
511                     goto illegal_req;
512                 }
513                 if (memchr(&param[i + 4], 0, len) == NULL) {
514                     goto illegal_req;
515                 }
516                 memcpy(id->iscsi_name, &param[i + 2], len + 2);
517                 j += offsetof(struct transportid, iscsi_name[len + 2]);
518                 i += len + 4;
519                 break;
520             case 6:
521                 /* SAS transport.  */
522                 if (i + 24 > transportid_len) {
523                     goto illegal_req;
524                 }
525                 memcpy(id->sas_address, &param[i + 4], 8);
526                 j += offsetof(struct transportid, sas_address[8]);
527                 i += 24;
528                 break;
529             default:
530             illegal_req:
531                 scsi_build_sense(sense, SENSE_CODE(INVALID_PARAM));
532                 return CHECK_CONDITION;
533             }
534 
535             assert(paramp->num_transportid < MPATH_MX_TIDS);
536             paramp->trnptid_list[paramp->num_transportid++] = id;
537         }
538     }
539 
540     r = mpath_persistent_reserve_out(fd, rq_servact, rq_scope, rq_type,
541                                      paramp, noisy, verbose);
542     return mpath_reconstruct_sense(fd, r, sense);
543 }
544 #endif
545 
546 static int do_pr_in(int fd, const uint8_t *cdb, uint8_t *sense,
547                     uint8_t *data, int *resp_sz)
548 {
549 #ifdef CONFIG_MPATH
550     if (is_mpath(fd)) {
551         /* multipath_pr_in fills the whole input buffer.  */
552         int r = multipath_pr_in(fd, cdb, sense, data, *resp_sz);
553         if (r != GOOD) {
554             *resp_sz = 0;
555         }
556         return r;
557     }
558 #endif
559 
560     return do_sgio(fd, cdb, sense, data, resp_sz,
561                    SG_DXFER_FROM_DEV);
562 }
563 
564 static int do_pr_out(int fd, const uint8_t *cdb, uint8_t *sense,
565                      const uint8_t *param, int sz)
566 {
567     int resp_sz;
568 
569     if ((fcntl(fd, F_GETFL) & O_ACCMODE) == O_RDONLY) {
570         scsi_build_sense(sense, SENSE_CODE(INVALID_OPCODE));
571         return CHECK_CONDITION;
572     }
573 
574 #ifdef CONFIG_MPATH
575     if (is_mpath(fd)) {
576         return multipath_pr_out(fd, cdb, sense, param, sz);
577     }
578 #endif
579 
580     resp_sz = sz;
581     return do_sgio(fd, cdb, sense, (uint8_t *)param, &resp_sz,
582                    SG_DXFER_TO_DEV);
583 }
584 
585 /* Client */
586 
587 typedef struct PRHelperClient {
588     QIOChannelSocket *ioc;
589     Coroutine *co;
590     int fd;
591     uint8_t data[PR_HELPER_DATA_SIZE];
592 } PRHelperClient;
593 
594 typedef struct PRHelperRequest {
595     int fd;
596     size_t sz;
597     uint8_t cdb[PR_HELPER_CDB_SIZE];
598 } PRHelperRequest;
599 
600 static int coroutine_fn prh_read(PRHelperClient *client, void *buf, int sz,
601                                  Error **errp)
602 {
603     int ret = 0;
604 
605     while (sz > 0) {
606         int *fds = NULL;
607         size_t nfds = 0;
608         int i;
609         struct iovec iov;
610         ssize_t n_read;
611 
612         iov.iov_base = buf;
613         iov.iov_len = sz;
614         n_read = qio_channel_readv_full(QIO_CHANNEL(client->ioc), &iov, 1,
615                                         &fds, &nfds, errp);
616 
617         if (n_read == QIO_CHANNEL_ERR_BLOCK) {
618             qio_channel_yield(QIO_CHANNEL(client->ioc), G_IO_IN);
619             continue;
620         }
621         if (n_read <= 0) {
622             ret = n_read ? n_read : -1;
623             goto err;
624         }
625 
626         /* Stash one file descriptor per request.  */
627         if (nfds) {
628             bool too_many = false;
629             for (i = 0; i < nfds; i++) {
630                 if (client->fd == -1) {
631                     client->fd = fds[i];
632                 } else {
633                     close(fds[i]);
634                     too_many = true;
635                 }
636             }
637             g_free(fds);
638             if (too_many) {
639                 ret = -1;
640                 goto err;
641             }
642         }
643 
644         buf += n_read;
645         sz -= n_read;
646     }
647 
648     return 0;
649 
650 err:
651     if (client->fd != -1) {
652         close(client->fd);
653         client->fd = -1;
654     }
655     return ret;
656 }
657 
658 static int coroutine_fn prh_read_request(PRHelperClient *client,
659                                          PRHelperRequest *req,
660                                          PRHelperResponse *resp, Error **errp)
661 {
662     uint32_t sz;
663 
664     if (prh_read(client, req->cdb, sizeof(req->cdb), NULL) < 0) {
665         return -1;
666     }
667 
668     if (client->fd == -1) {
669         error_setg(errp, "No file descriptor in request.");
670         return -1;
671     }
672 
673     if (req->cdb[0] != PERSISTENT_RESERVE_OUT &&
674         req->cdb[0] != PERSISTENT_RESERVE_IN) {
675         error_setg(errp, "Invalid CDB, closing socket.");
676         goto out_close;
677     }
678 
679     sz = scsi_cdb_xfer(req->cdb);
680     if (sz > sizeof(client->data)) {
681         goto out_close;
682     }
683 
684     if (req->cdb[0] == PERSISTENT_RESERVE_OUT) {
685         if (qio_channel_read_all(QIO_CHANNEL(client->ioc),
686                                  (char *)client->data, sz,
687                                  errp) < 0) {
688             goto out_close;
689         }
690     }
691 
692     req->fd = client->fd;
693     req->sz = sz;
694     client->fd = -1;
695     return sz;
696 
697 out_close:
698     close(client->fd);
699     client->fd = -1;
700     return -1;
701 }
702 
703 static int coroutine_fn prh_write_response(PRHelperClient *client,
704                                            PRHelperRequest *req,
705                                            PRHelperResponse *resp, Error **errp)
706 {
707     ssize_t r;
708     size_t sz;
709 
710     if (req->cdb[0] == PERSISTENT_RESERVE_IN && resp->result == GOOD) {
711         assert(resp->sz <= req->sz && resp->sz <= sizeof(client->data));
712     } else {
713         assert(resp->sz == 0);
714     }
715 
716     sz = resp->sz;
717 
718     resp->result = cpu_to_be32(resp->result);
719     resp->sz = cpu_to_be32(resp->sz);
720     r = qio_channel_write_all(QIO_CHANNEL(client->ioc),
721                               (char *) resp, sizeof(*resp), errp);
722     if (r < 0) {
723         return r;
724     }
725 
726     r = qio_channel_write_all(QIO_CHANNEL(client->ioc),
727                               (char *) client->data,
728                               sz, errp);
729     return r < 0 ? r : 0;
730 }
731 
732 static void coroutine_fn prh_co_entry(void *opaque)
733 {
734     PRHelperClient *client = opaque;
735     Error *local_err = NULL;
736     uint32_t flags;
737     int r;
738 
739     qio_channel_set_blocking(QIO_CHANNEL(client->ioc),
740                              false, NULL);
741     qio_channel_attach_aio_context(QIO_CHANNEL(client->ioc),
742                                    qemu_get_aio_context());
743 
744     /* A very simple negotiation for future extensibility.  No features
745      * are defined so write 0.
746      */
747     flags = cpu_to_be32(0);
748     r = qio_channel_write_all(QIO_CHANNEL(client->ioc),
749                              (char *) &flags, sizeof(flags), NULL);
750     if (r < 0) {
751         goto out;
752     }
753 
754     r = qio_channel_read_all(QIO_CHANNEL(client->ioc),
755                              (char *) &flags, sizeof(flags), NULL);
756     if (be32_to_cpu(flags) != 0 || r < 0) {
757         goto out;
758     }
759 
760     while (qatomic_read(&state) == RUNNING) {
761         PRHelperRequest req;
762         PRHelperResponse resp;
763         int sz;
764 
765         sz = prh_read_request(client, &req, &resp, &local_err);
766         if (sz < 0) {
767             break;
768         }
769 
770         num_active_sockets++;
771         if (req.cdb[0] == PERSISTENT_RESERVE_OUT) {
772             r = do_pr_out(req.fd, req.cdb, resp.sense,
773                           client->data, sz);
774             resp.sz = 0;
775         } else {
776             resp.sz = sizeof(client->data);
777             r = do_pr_in(req.fd, req.cdb, resp.sense,
778                          client->data, &resp.sz);
779             resp.sz = MIN(resp.sz, sz);
780         }
781         num_active_sockets--;
782         close(req.fd);
783         if (r == -1) {
784             break;
785         }
786         resp.result = r;
787 
788         if (prh_write_response(client, &req, &resp, &local_err) < 0) {
789             break;
790         }
791     }
792 
793     if (local_err) {
794         if (verbose == 0) {
795             error_free(local_err);
796         } else {
797             error_report_err(local_err);
798         }
799     }
800 
801 out:
802     qio_channel_detach_aio_context(QIO_CHANNEL(client->ioc));
803     object_unref(OBJECT(client->ioc));
804     g_free(client);
805 }
806 
807 static gboolean accept_client(QIOChannel *ioc, GIOCondition cond, gpointer opaque)
808 {
809     QIOChannelSocket *cioc;
810     PRHelperClient *prh;
811 
812     cioc = qio_channel_socket_accept(QIO_CHANNEL_SOCKET(ioc),
813                                      NULL);
814     if (!cioc) {
815         return TRUE;
816     }
817 
818     prh = g_new(PRHelperClient, 1);
819     prh->ioc = cioc;
820     prh->fd = -1;
821     prh->co = qemu_coroutine_create(prh_co_entry, prh);
822     qemu_coroutine_enter(prh->co);
823 
824     return TRUE;
825 }
826 
827 static void termsig_handler(int signum)
828 {
829     qatomic_cmpxchg(&state, RUNNING, TERMINATE);
830     qemu_notify_event();
831 }
832 
833 static void close_server_socket(void)
834 {
835     assert(server_ioc);
836 
837     g_source_remove(server_watch);
838     server_watch = -1;
839     object_unref(OBJECT(server_ioc));
840     num_active_sockets--;
841 }
842 
843 #ifdef CONFIG_LIBCAP_NG
844 static int drop_privileges(void)
845 {
846     /* clear all capabilities */
847     capng_clear(CAPNG_SELECT_BOTH);
848 
849     if (capng_update(CAPNG_ADD, CAPNG_EFFECTIVE | CAPNG_PERMITTED,
850                      CAP_SYS_RAWIO) < 0) {
851         return -1;
852     }
853 
854 #ifdef CONFIG_MPATH
855     /* For /dev/mapper/control ioctls */
856     if (capng_update(CAPNG_ADD, CAPNG_EFFECTIVE | CAPNG_PERMITTED,
857                      CAP_SYS_ADMIN) < 0) {
858         return -1;
859     }
860 #endif
861 
862     /* Change user/group id, retaining the capabilities.  Because file descriptors
863      * are passed via SCM_RIGHTS, we don't need supplementary groups (and in
864      * fact the helper can run as "nobody").
865      */
866     if (capng_change_id(uid != -1 ? uid : getuid(),
867                         gid != -1 ? gid : getgid(),
868                         CAPNG_DROP_SUPP_GRP | CAPNG_CLEAR_BOUNDING)) {
869         return -1;
870     }
871 
872     return 0;
873 }
874 #endif
875 
876 int main(int argc, char **argv)
877 {
878     const char *sopt = "hVk:f:dT:u:g:vq";
879     struct option lopt[] = {
880         { "help", no_argument, NULL, 'h' },
881         { "version", no_argument, NULL, 'V' },
882         { "socket", required_argument, NULL, 'k' },
883         { "pidfile", required_argument, NULL, 'f' },
884         { "daemon", no_argument, NULL, 'd' },
885         { "trace", required_argument, NULL, 'T' },
886         { "user", required_argument, NULL, 'u' },
887         { "group", required_argument, NULL, 'g' },
888         { "verbose", no_argument, NULL, 'v' },
889         { "quiet", no_argument, NULL, 'q' },
890         { NULL, 0, NULL, 0 }
891     };
892     int opt_ind = 0;
893     int loglevel = 1;
894     int quiet = 0;
895     int ch;
896     Error *local_err = NULL;
897     bool daemonize = false;
898     bool pidfile_specified = false;
899     bool socket_path_specified = false;
900     unsigned socket_activation;
901 
902     struct sigaction sa_sigterm;
903     memset(&sa_sigterm, 0, sizeof(sa_sigterm));
904     sa_sigterm.sa_handler = termsig_handler;
905     sigaction(SIGTERM, &sa_sigterm, NULL);
906     sigaction(SIGINT, &sa_sigterm, NULL);
907     sigaction(SIGHUP, &sa_sigterm, NULL);
908 
909     signal(SIGPIPE, SIG_IGN);
910 
911     error_init(argv[0]);
912     module_call_init(MODULE_INIT_TRACE);
913     module_call_init(MODULE_INIT_QOM);
914     qemu_add_opts(&qemu_trace_opts);
915     qemu_init_exec_dir(argv[0]);
916 
917     compute_default_paths();
918 
919     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
920         switch (ch) {
921         case 'k':
922             g_free(socket_path);
923             socket_path = g_strdup(optarg);
924             socket_path_specified = true;
925             if (socket_path[0] != '/') {
926                 error_report("socket path must be absolute");
927                 exit(EXIT_FAILURE);
928             }
929             break;
930         case 'f':
931             g_free(pidfile);
932             pidfile = g_strdup(optarg);
933             pidfile_specified = true;
934             break;
935 #ifdef CONFIG_LIBCAP_NG
936         case 'u': {
937             unsigned long res;
938             struct passwd *userinfo = getpwnam(optarg);
939             if (userinfo) {
940                 uid = userinfo->pw_uid;
941             } else if (qemu_strtoul(optarg, NULL, 10, &res) == 0 &&
942                        (uid_t)res == res) {
943                 uid = res;
944             } else {
945                 error_report("invalid user '%s'", optarg);
946                 exit(EXIT_FAILURE);
947             }
948             break;
949         }
950         case 'g': {
951             unsigned long res;
952             struct group *groupinfo = getgrnam(optarg);
953             if (groupinfo) {
954                 gid = groupinfo->gr_gid;
955             } else if (qemu_strtoul(optarg, NULL, 10, &res) == 0 &&
956                        (gid_t)res == res) {
957                 gid = res;
958             } else {
959                 error_report("invalid group '%s'", optarg);
960                 exit(EXIT_FAILURE);
961             }
962             break;
963         }
964 #else
965         case 'u':
966         case 'g':
967             error_report("-%c not supported by this %s", ch, argv[0]);
968             exit(1);
969 #endif
970         case 'd':
971             daemonize = true;
972             break;
973         case 'q':
974             quiet = 1;
975             break;
976         case 'v':
977             ++loglevel;
978             break;
979         case 'T':
980             trace_opt_parse(optarg);
981             break;
982         case 'V':
983             version(argv[0]);
984             exit(EXIT_SUCCESS);
985             break;
986         case 'h':
987             usage(argv[0]);
988             exit(EXIT_SUCCESS);
989             break;
990         case '?':
991             error_report("Try `%s --help' for more information.", argv[0]);
992             exit(EXIT_FAILURE);
993         }
994     }
995 
996     /* set verbosity */
997     noisy = !quiet && (loglevel >= 3);
998     verbose = quiet ? 0 : MIN(loglevel, 3);
999 
1000     if (!trace_init_backends()) {
1001         exit(EXIT_FAILURE);
1002     }
1003     trace_init_file();
1004     qemu_set_log(LOG_TRACE);
1005 
1006 #ifdef CONFIG_MPATH
1007     dm_init();
1008     multipath_pr_init();
1009 #endif
1010 
1011     socket_activation = check_socket_activation();
1012     if (socket_activation == 0) {
1013         SocketAddress saddr;
1014         saddr = (SocketAddress){
1015             .type = SOCKET_ADDRESS_TYPE_UNIX,
1016             .u.q_unix.path = socket_path,
1017         };
1018         server_ioc = qio_channel_socket_new();
1019         if (qio_channel_socket_listen_sync(server_ioc, &saddr,
1020                                            1, &local_err) < 0) {
1021             object_unref(OBJECT(server_ioc));
1022             error_report_err(local_err);
1023             return 1;
1024         }
1025     } else {
1026         /* Using socket activation - check user didn't use -p etc. */
1027         if (socket_path_specified) {
1028             error_report("Unix socket can't be set when using socket activation");
1029             exit(EXIT_FAILURE);
1030         }
1031 
1032         /* Can only listen on a single socket.  */
1033         if (socket_activation > 1) {
1034             error_report("%s does not support socket activation with LISTEN_FDS > 1",
1035                          argv[0]);
1036             exit(EXIT_FAILURE);
1037         }
1038         server_ioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD,
1039                                                &local_err);
1040         if (server_ioc == NULL) {
1041             error_reportf_err(local_err,
1042                               "Failed to use socket activation: ");
1043             exit(EXIT_FAILURE);
1044         }
1045     }
1046 
1047     if (qemu_init_main_loop(&local_err)) {
1048         error_report_err(local_err);
1049         exit(EXIT_FAILURE);
1050     }
1051 
1052     server_watch = qio_channel_add_watch(QIO_CHANNEL(server_ioc),
1053                                          G_IO_IN,
1054                                          accept_client,
1055                                          NULL, NULL);
1056 
1057     if (daemonize) {
1058         if (daemon(0, 0) < 0) {
1059             error_report("Failed to daemonize: %s", strerror(errno));
1060             exit(EXIT_FAILURE);
1061         }
1062     }
1063 
1064     if ((daemonize || pidfile_specified) &&
1065         !qemu_write_pidfile(pidfile, &local_err)) {
1066         error_report_err(local_err);
1067         exit(EXIT_FAILURE);
1068     }
1069 
1070 #ifdef CONFIG_LIBCAP_NG
1071     if (drop_privileges() < 0) {
1072         error_report("Failed to drop privileges: %s", strerror(errno));
1073         exit(EXIT_FAILURE);
1074     }
1075 #endif
1076 
1077     state = RUNNING;
1078     do {
1079         main_loop_wait(false);
1080         if (state == TERMINATE) {
1081             state = TERMINATING;
1082             close_server_socket();
1083         }
1084     } while (num_active_sockets > 0);
1085 
1086     exit(EXIT_SUCCESS);
1087 }
1088