Attachment #201388 for bug #234576

View | Details | Raw Unified | Return to bug 234576
Collapse All | Expand All




/* The is only one remote component for now. */
#define	ISREMOTE(no)	((no) == 1)

/* There appears to be an issue with ZFS where you get MAXPHYS + 0x200 back in geom_gate
   on occasion - usually when using raidz2.  This causes a lot of issues; so we "fix" this
   by allowing an additional 0x800 bytes before croaking.                                  */

/* Upgrading to 12.0 leads to a request for 0x100000 bytes.  This makes this fix even more
   fragile but as a workaround, allow ggate_recv to receive this much data.                */ 
#define GGRMAXLEN 0x100000 

struct hio {
	/*
	 * Number of components we are still waiting for.

			    sizeof(hio->hio_next[0]) * ncomps);
		}
		hio->hio_ggio.gctl_version = G_GATE_VERSION;
		hio->hio_ggio.gctl_data = malloc(GGRMAXLEN);
		if (hio->hio_ggio.gctl_data == NULL) {
			primary_exitx(EX_TEMPFAIL,
			    "Unable to allocate %zu bytes of memory for gctl_data.",
			    GGRMAXLEN);
		}
		hio->hio_ggio.gctl_length = GGRMAXLEN;
		hio->hio_ggio.gctl_error = 0;
		TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_free_next);
		hio_free_list_size++;

		pjdlog_debug(2, "ggate_recv: (%p) Got free request.", hio);
		ggio = &hio->hio_ggio;
		ggio->gctl_unit = res->hr_ggateunit;
		ggio->gctl_length = GGRMAXLEN;
		ggio->gctl_error = 0;
		hio->hio_done = false;
		hio->hio_replication = res->hr_replication;

			 * Buffer too small? Impossible, we allocate MAXPHYS
			 * bytes - request can't be bigger than that.
			 */
			primary_exitx(EX_OSERR, "G_GATE_CMD_START failed: %s.  This is in the source as being 'impossible' and appears to have happened.  allocated=%x length=%x maxphys=%x", strerror(error), GGRMAXLEN, ggio->gctl_length, MAXPHYS);

			/* FALLTHROUGH */
		case ENXIO:
		default:

			    strerror(error));
		}

		if (ggio->gctl_length > MAXPHYS) {
			pjdlog_warning("gctl_length is larger than MAXPHYS (%x) - would have been ENOMEM.", ggio->gctl_length);
		}
	
		ncomp = 0;
		ncomps = HAST_NCOMPONENTS;


Lines 131-136 Link Here

(-)secondary.c (-5 / +8 lines)
131	mtx_unlock(&hio_##name##_list_lock); \	131	mtx_unlock(&hio_##name##_list_lock); \
132	} while (0)	132	} while (0)
133		133
		134	/* See comments in primary.c about this. */
		135	#define GGRMAXLEN 0x100000
		136
134	static void	137	static void
135	output_status_aux(struct nv *nvout)	138	output_status_aux(struct nv *nvout)
136	{	139	{
Lines 196-206 Link Here
196	"Unable to allocate memory (%zu bytes) for hio request.",	199	"Unable to allocate memory (%zu bytes) for hio request.",
197	sizeof(*hio));	200	sizeof(*hio));
198	}	201	}
199	hio->hio_data = malloc(MAXPHYS);	202	hio->hio_data = malloc(GGRMAXLEN);
200	if (hio->hio_data == NULL) {	203	if (hio->hio_data == NULL) {
201	pjdlog_exitx(EX_TEMPFAIL,	204	pjdlog_exitx(EX_TEMPFAIL,
202	"Unable to allocate memory (%zu bytes) for gctl_data.",	205	"Unable to allocate memory (%zu bytes) for gctl_data.",
203	(size_t)MAXPHYS);	206	(size_t)GGRMAXLEN);
204	}	207	}
205	hio_clear(hio);	208	hio_clear(hio);
206	TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_next);	209	TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_next);
Lines 597-605 Link Here
597	hio->hio_error = EINVAL;	600	hio->hio_error = EINVAL;
598	goto end;	601	goto end;
599	}	602	}
600	if (hio->hio_cmd != HIO_DELETE && hio->hio_length > MAXPHYS) {	603	if (hio->hio_cmd != HIO_DELETE && hio->hio_length > GGRMAXLEN) {
601	pjdlog_error("Data length is too large (%ju > %ju).",	604	pjdlog_error("Data length is too large (%ju > %ju).",
602	(uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS);	605	(uintmax_t)hio->hio_length, (uintmax_t)GGRMAXLEN);
603	hio->hio_error = EINVAL;	606	hio->hio_error = EINVAL;
604	goto end;	607	goto end;
605	}	608	}
Lines 705-711 Link Here
705	continue;	708	continue;
706	} else if (hio->hio_cmd == HIO_WRITE) {	709	} else if (hio->hio_cmd == HIO_WRITE) {
707	if (hast_proto_recv_data(res, res->hr_remotein, nv,	710	if (hast_proto_recv_data(res, res->hr_remotein, nv,
708	hio->hio_data, MAXPHYS) == -1) {	711	hio->hio_data, GGRMAXLEN) == -1) {
709	secondary_exit(EX_TEMPFAIL,	712	secondary_exit(EX_TEMPFAIL,
710	"Unable to receive request data");	713	"Unable to receive request data");
711	}	714	}

Return to bug 234576

Lines 73-78 Link Here

(-)primary.c (-4 / +18 lines)
73	/* The is only one remote component for now. */	73	/* The is only one remote component for now. */
74	#define ISREMOTE(no) ((no) == 1)	74	#define ISREMOTE(no) ((no) == 1)
75		75
		76	/* There appears to be an issue with ZFS where you get MAXPHYS + 0x200 back in geom_gate
		77	on occasion - usually when using raidz2. This causes a lot of issues; so we "fix" this
		78	by allowing an additional 0x800 bytes before croaking. */
		79
		80	/* Upgrading to 12.0 leads to a request for 0x100000 bytes. This makes this fix even more
		81	fragile but as a workaround, allow ggate_recv to receive this much data. */
		82	#define GGRMAXLEN 0x100000
		83
76	struct hio {	84	struct hio {
77	/*	85	/*
78	* Number of components we are still waiting for.	86	* Number of components we are still waiting for.
Lines 490-502 Link Here
490	sizeof(hio->hio_next[0]) * ncomps);	498	sizeof(hio->hio_next[0]) * ncomps);
491	}	499	}
492	hio->hio_ggio.gctl_version = G_GATE_VERSION;	500	hio->hio_ggio.gctl_version = G_GATE_VERSION;
493	hio->hio_ggio.gctl_data = malloc(MAXPHYS);	501	hio->hio_ggio.gctl_data = malloc(GGRMAXLEN);
494	if (hio->hio_ggio.gctl_data == NULL) {	502	if (hio->hio_ggio.gctl_data == NULL) {
495	primary_exitx(EX_TEMPFAIL,	503	primary_exitx(EX_TEMPFAIL,
496	"Unable to allocate %zu bytes of memory for gctl_data.",	504	"Unable to allocate %zu bytes of memory for gctl_data.",
497	MAXPHYS);	505	GGRMAXLEN);
498	}	506	}
499	hio->hio_ggio.gctl_length = MAXPHYS;	507	hio->hio_ggio.gctl_length = GGRMAXLEN;
500	hio->hio_ggio.gctl_error = 0;	508	hio->hio_ggio.gctl_error = 0;
501	TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_free_next);	509	TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_free_next);
502	hio_free_list_size++;	510	hio_free_list_size++;
Lines 1236-1242 Link Here
1236	pjdlog_debug(2, "ggate_recv: (%p) Got free request.", hio);	1244	pjdlog_debug(2, "ggate_recv: (%p) Got free request.", hio);
1237	ggio = &hio->hio_ggio;	1245	ggio = &hio->hio_ggio;
1238	ggio->gctl_unit = res->hr_ggateunit;	1246	ggio->gctl_unit = res->hr_ggateunit;
1239	ggio->gctl_length = MAXPHYS;	1247	ggio->gctl_length = GGRMAXLEN;
1240	ggio->gctl_error = 0;	1248	ggio->gctl_error = 0;
1241	hio->hio_done = false;	1249	hio->hio_done = false;
1242	hio->hio_replication = res->hr_replication;	1250	hio->hio_replication = res->hr_replication;
Lines 1266-1271 Link Here
1266	* Buffer too small? Impossible, we allocate MAXPHYS	1274	* Buffer too small? Impossible, we allocate MAXPHYS
1267	* bytes - request can't be bigger than that.	1275	* bytes - request can't be bigger than that.
1268	*/	1276	*/
		1277	primary_exitx(EX_OSERR, "G_GATE_CMD_START failed: %s. This is in the source as being 'impossible' and appears to have happened. allocated=%x length=%x maxphys=%x", strerror(error), GGRMAXLEN, ggio->gctl_length, MAXPHYS);
		1278
1269	/* FALLTHROUGH */	1279	/* FALLTHROUGH */
1270	case ENXIO:	1280	case ENXIO:
1271	default:	1281	default:
Lines 1273-1278 Link Here
1273	strerror(error));	1283	strerror(error));
1274	}	1284	}
1275		1285
		1286	if (ggio->gctl_length > MAXPHYS) {
		1287	pjdlog_warning("gctl_length is larger than MAXPHYS (%x) - would have been ENOMEM.", ggio->gctl_length);
		1288	}
		1289
1276	ncomp = 0;	1290	ncomp = 0;
1277	ncomps = HAST_NCOMPONENTS;	1291	ncomps = HAST_NCOMPONENTS;
1278		1292