1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
23 */
24 /*
25 * Copyright 2012, Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2013 by Delphix. All rights reserved.
27 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
28 */
29
30 #include <sys/conf.h>
31 #include <sys/file.h>
32 #include <sys/ddi.h>
33 #include <sys/sunddi.h>
34 #include <sys/modctl.h>
35 #include <sys/scsi/scsi.h>
36 #include <sys/scsi/generic/persist.h>
37 #include <sys/scsi/impl/scsi_reset_notify.h>
38 #include <sys/disp.h>
39 #include <sys/byteorder.h>
40 #include <sys/atomic.h>
41 #include <sys/ethernet.h>
42 #include <sys/sdt.h>
43 #include <sys/nvpair.h>
44 #include <sys/zone.h>
45 #include <sys/id_space.h>
107 static void stmf_svc_kill_obj_requests(void *obj);
108 static void stmf_svc_timeout(struct stmf_svc_clocks *);
109 void stmf_check_freetask();
110 void stmf_abort_target_reset(scsi_task_t *task);
111 stmf_status_t stmf_lun_reset_poll(stmf_lu_t *lu, struct scsi_task *task,
112 int target_reset);
113 void stmf_target_reset_poll(struct scsi_task *task);
114 void stmf_handle_lun_reset(scsi_task_t *task);
115 void stmf_handle_target_reset(scsi_task_t *task);
116 void stmf_xd_to_dbuf(stmf_data_buf_t *dbuf, int set_rel_off);
117 int stmf_load_ppd_ioctl(stmf_ppioctl_data_t *ppi, uint64_t *ppi_token,
118 uint32_t *err_ret);
119 int stmf_delete_ppd_ioctl(stmf_ppioctl_data_t *ppi);
120 int stmf_get_ppd_ioctl(stmf_ppioctl_data_t *ppi, stmf_ppioctl_data_t *ppi_out,
121 uint32_t *err_ret);
122 void stmf_delete_ppd(stmf_pp_data_t *ppd);
123 void stmf_delete_all_ppds();
124 void stmf_trace_clear();
125 void stmf_worker_init();
126 stmf_status_t stmf_worker_fini();
127 void stmf_worker_mgmt();
128 void stmf_worker_task(void *arg);
129 static void stmf_task_lu_free(scsi_task_t *task, stmf_i_scsi_session_t *iss);
130 static stmf_status_t stmf_ic_lu_reg(stmf_ic_reg_dereg_lun_msg_t *msg,
131 uint32_t type);
132 static stmf_status_t stmf_ic_lu_dereg(stmf_ic_reg_dereg_lun_msg_t *msg);
133 static stmf_status_t stmf_ic_rx_scsi_status(stmf_ic_scsi_status_msg_t *msg);
134 static stmf_status_t stmf_ic_rx_status(stmf_ic_status_msg_t *msg);
135 static stmf_status_t stmf_ic_rx_scsi_data(stmf_ic_scsi_data_msg_t *msg);
136 void stmf_task_lu_killall(stmf_lu_t *lu, scsi_task_t *tm_task, stmf_status_t s);
137
138 /* pppt modhandle */
139 ddi_modhandle_t pppt_mod;
140
141 /* pppt modload imported functions */
142 stmf_ic_reg_port_msg_alloc_func_t ic_reg_port_msg_alloc;
143 stmf_ic_dereg_port_msg_alloc_func_t ic_dereg_port_msg_alloc;
144 stmf_ic_reg_lun_msg_alloc_func_t ic_reg_lun_msg_alloc;
145 stmf_ic_dereg_lun_msg_alloc_func_t ic_dereg_lun_msg_alloc;
146 stmf_ic_lun_active_msg_alloc_func_t ic_lun_active_msg_alloc;
147 stmf_ic_scsi_cmd_msg_alloc_func_t ic_scsi_cmd_msg_alloc;
148 stmf_ic_scsi_data_xfer_done_msg_alloc_func_t ic_scsi_data_xfer_done_msg_alloc;
149 stmf_ic_session_create_msg_alloc_func_t ic_session_reg_msg_alloc;
150 stmf_ic_session_destroy_msg_alloc_func_t ic_session_dereg_msg_alloc;
151 stmf_ic_tx_msg_func_t ic_tx_msg;
152 stmf_ic_msg_free_func_t ic_msg_free;
153
154 static void stmf_itl_task_start(stmf_i_scsi_task_t *itask);
155 static void stmf_itl_lu_new_task(stmf_i_scsi_task_t *itask);
156 static void stmf_itl_task_done(stmf_i_scsi_task_t *itask);
157
158 static void stmf_lport_xfer_start(stmf_i_scsi_task_t *itask,
159 stmf_data_buf_t *dbuf);
160 static void stmf_lport_xfer_done(stmf_i_scsi_task_t *itask,
161 stmf_data_buf_t *dbuf);
162
163 static void stmf_update_kstat_lu_q(scsi_task_t *, void());
164 static void stmf_update_kstat_lport_q(scsi_task_t *, void());
165 static void stmf_update_kstat_lu_io(scsi_task_t *, stmf_data_buf_t *);
166 static void stmf_update_kstat_lport_io(scsi_task_t *, stmf_data_buf_t *);
167
168 static int stmf_irport_compare(const void *void_irport1,
169 const void *void_irport2);
170 static stmf_i_remote_port_t *stmf_irport_create(scsi_devid_desc_t *rport_devid);
171 static void stmf_irport_destroy(stmf_i_remote_port_t *irport);
172 static stmf_i_remote_port_t *stmf_irport_register(
173 scsi_devid_desc_t *rport_devid);
174 static stmf_i_remote_port_t *stmf_irport_lookup_locked(
175 scsi_devid_desc_t *rport_devid);
176 static void stmf_irport_deregister(stmf_i_remote_port_t *irport);
177
178 extern struct mod_ops mod_driverops;
179
180 /* =====[ Tunables ]===== */
181 /* Internal tracing */
182 volatile int stmf_trace_on = 1;
183 volatile int stmf_trace_buf_size = (1 * 1024 * 1024);
184 /*
185 * The reason default task timeout is 75 is because we want the
186 * host to timeout 1st and mostly host timeout is 60 seconds.
187 */
188 volatile int stmf_default_task_timeout = 75;
189 /*
190 * Setting this to one means, you are responsible for config load and keeping
191 * things in sync with persistent database.
192 */
193 volatile int stmf_allow_modunload = 0;
194
195 volatile int stmf_max_nworkers = 256;
196 volatile int stmf_min_nworkers = 4;
197 volatile int stmf_worker_scale_down_delay = 20;
198
199 /* === [ Debugging and fault injection ] === */
200 #ifdef DEBUG
201 volatile uint32_t stmf_drop_task_counter = 0;
202 volatile uint32_t stmf_drop_buf_counter = 0;
203
204 #endif
205
206 stmf_state_t stmf_state;
207 static stmf_lu_t *dlun0;
208
209 static uint8_t stmf_first_zero[] =
210 { 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0, 0xff };
211 static uint8_t stmf_first_one[] =
212 { 0xff, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0 };
213
214 static kmutex_t trace_buf_lock;
215 static int trace_buf_size;
216 static int trace_buf_curndx;
217 caddr_t stmf_trace_buf;
218
219 static enum {
220 STMF_WORKERS_DISABLED = 0,
221 STMF_WORKERS_ENABLING,
222 STMF_WORKERS_ENABLED
223 } stmf_workers_state = STMF_WORKERS_DISABLED;
224 static int stmf_i_max_nworkers;
225 static int stmf_i_min_nworkers;
226 static int stmf_nworkers_cur; /* # of workers currently running */
227 static int stmf_nworkers_needed; /* # of workers need to be running */
228 static int stmf_worker_sel_counter = 0;
229 static uint32_t stmf_cur_ntasks = 0;
230 static clock_t stmf_wm_last = 0;
231 /*
232 * This is equal to stmf_nworkers_cur while we are increasing # workers and
233 * stmf_nworkers_needed while we are decreasing the worker count.
234 */
235 static int stmf_nworkers_accepting_cmds;
236 static stmf_worker_t *stmf_workers = NULL;
237 static clock_t stmf_worker_mgmt_delay = 2;
238 static clock_t stmf_worker_scale_down_timer = 0;
239 static int stmf_worker_scale_down_qd = 0;
240
241 static struct cb_ops stmf_cb_ops = {
242 stmf_open, /* open */
243 stmf_close, /* close */
244 nodev, /* strategy */
245 nodev, /* print */
246 nodev, /* dump */
247 nodev, /* read */
248 nodev, /* write */
249 stmf_ioctl, /* ioctl */
250 nodev, /* devmap */
251 nodev, /* mmap */
252 nodev, /* segmap */
253 nochpoll, /* chpoll */
254 ddi_prop_op, /* cb_prop_op */
255 0, /* streamtab */
256 D_NEW | D_MP, /* cb_flag */
257 CB_REV, /* rev */
258 nodev, /* aread */
259 nodev /* awrite */
260 };
261
262 static struct dev_ops stmf_ops = {
263 DEVO_REV,
264 0,
265 stmf_getinfo,
266 nulldev, /* identify */
267 nulldev, /* probe */
268 stmf_attach,
269 stmf_detach,
270 nodev, /* reset */
271 &stmf_cb_ops,
272 NULL, /* bus_ops */
273 NULL /* power */
274 };
275
276 #define STMF_NAME "COMSTAR STMF"
277 #define STMF_MODULE_NAME "stmf"
278
279 static struct modldrv modldrv = {
280 &mod_driverops,
281 STMF_NAME,
282 &stmf_ops
283 };
284
285 static struct modlinkage modlinkage = {
286 MODREV_1,
287 &modldrv,
288 NULL
289 };
290
291 int
292 _init(void)
293 {
294 int ret;
295
296 ret = mod_install(&modlinkage);
297 if (ret)
298 return (ret);
299 stmf_trace_buf = kmem_zalloc(stmf_trace_buf_size, KM_SLEEP);
300 trace_buf_size = stmf_trace_buf_size;
301 trace_buf_curndx = 0;
302 mutex_init(&trace_buf_lock, NULL, MUTEX_DRIVER, 0);
303 bzero(&stmf_state, sizeof (stmf_state_t));
304 /* STMF service is off by default */
305 stmf_state.stmf_service_running = 0;
306 /* default lu/lport states are online */
307 stmf_state.stmf_default_lu_state = STMF_STATE_ONLINE;
308 stmf_state.stmf_default_lport_state = STMF_STATE_ONLINE;
309 mutex_init(&stmf_state.stmf_lock, NULL, MUTEX_DRIVER, NULL);
310 cv_init(&stmf_state.stmf_cv, NULL, CV_DRIVER, NULL);
311 stmf_session_counter = (uint64_t)ddi_get_lbolt();
312 avl_create(&stmf_state.stmf_irportlist,
313 stmf_irport_compare, sizeof (stmf_i_remote_port_t),
314 offsetof(stmf_i_remote_port_t, irport_ln));
315 stmf_state.stmf_ilport_inst_space =
316 id_space_create("lport-instances", 0, MAX_ILPORT);
317 stmf_state.stmf_irport_inst_space =
318 id_space_create("rport-instances", 0, MAX_IRPORT);
319 stmf_view_init();
320 stmf_svc_init();
321 stmf_dlun_init();
322 return (ret);
353 ret = mod_remove(&modlinkage);
354 if (ret) {
355 stmf_svc_init();
356 stmf_dlun_init();
357 stmf_worker_init();
358 return (ret);
359 }
360
361 stmf_view_clear_config();
362
363 while ((irport = avl_destroy_nodes(&stmf_state.stmf_irportlist,
364 &avl_dest_cookie)) != NULL)
365 stmf_irport_destroy(irport);
366 avl_destroy(&stmf_state.stmf_irportlist);
367 id_space_destroy(stmf_state.stmf_ilport_inst_space);
368 id_space_destroy(stmf_state.stmf_irport_inst_space);
369
370 kmem_free(stmf_trace_buf, stmf_trace_buf_size);
371 mutex_destroy(&trace_buf_lock);
372 mutex_destroy(&stmf_state.stmf_lock);
373 cv_destroy(&stmf_state.stmf_cv);
374 return (ret);
375 }
376
377 int
378 _info(struct modinfo *modinfop)
379 {
380 return (mod_info(&modlinkage, modinfop));
381 }
382
383 /* ARGSUSED */
384 static int
385 stmf_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg, void **result)
386 {
387 switch (cmd) {
388 case DDI_INFO_DEVT2DEVINFO:
389 *result = stmf_state.stmf_dip;
390 break;
391 case DDI_INFO_DEVT2INSTANCE:
392 *result =
1638 find_task_from_msgid(uint8_t *lu_id, stmf_ic_msgid_t task_msgid)
1639 {
1640 stmf_i_lu_t *ilu;
1641 stmf_i_scsi_task_t *itask;
1642
1643 mutex_enter(&stmf_state.stmf_lock);
1644 for (ilu = stmf_state.stmf_ilulist; ilu != NULL; ilu = ilu->ilu_next) {
1645 if (bcmp(lu_id, ilu->ilu_lu->lu_id->ident, 16) == 0) {
1646 break;
1647 }
1648 }
1649
1650 if (ilu == NULL) {
1651 mutex_exit(&stmf_state.stmf_lock);
1652 return (NULL);
1653 }
1654
1655 mutex_enter(&ilu->ilu_task_lock);
1656 for (itask = ilu->ilu_tasks; itask != NULL;
1657 itask = itask->itask_lu_next) {
1658 if (itask->itask_flags & (ITASK_IN_FREE_LIST |
1659 ITASK_BEING_ABORTED)) {
1660 continue;
1661 }
1662 if (itask->itask_proxy_msg_id == task_msgid) {
1663 break;
1664 }
1665 }
1666 mutex_exit(&ilu->ilu_task_lock);
1667 mutex_exit(&stmf_state.stmf_lock);
1668
1669 if (itask != NULL) {
1670 return (itask->itask_task);
1671 } else {
1672 /* task not found. Likely already aborted. */
1673 return (NULL);
1674 }
1675 }
1676
1677 /*
1678 * message received from pppt/ic
1679 */
1680 stmf_status_t
1681 stmf_msg_rx(stmf_ic_msg_t *msg)
1886 }
1887
1888 if (ilport->ilport_proxy_registered == 0) {
1889 return (STMF_FAILURE);
1890 }
1891
1892 mutex_enter(&stmf_state.stmf_lock);
1893 itask->itask_proxy_msg_id = stmf_proxy_msg_id++;
1894 mutex_exit(&stmf_state.stmf_lock);
1895 itask->itask_proxy_dbuf = dbuf;
1896
1897 /*
1898 * stmf will now take over the task handling for this task
1899 * but it still needs to be treated differently from other
1900 * default handled tasks, hence the ITASK_PROXY_TASK.
1901 * If this is a task management function, we're really just
1902 * duping the command to the peer. Set the TM bit so that
1903 * we can recognize this on return since we won't be completing
1904 * the proxied task in that case.
1905 */
1906 if (task->task_mgmt_function) {
1907 itask->itask_proxy_msg_id |= MSG_ID_TM_BIT;
1908 } else {
1909 uint32_t new, old;
1910 do {
1911 new = old = itask->itask_flags;
1912 if (new & ITASK_BEING_ABORTED)
1913 return (STMF_FAILURE);
1914 new |= ITASK_DEFAULT_HANDLING | ITASK_PROXY_TASK;
1915 } while (atomic_cas_32(&itask->itask_flags, old, new) != old);
1916 }
1917 if (dbuf) {
1918 ic_cmd_msg = ic_scsi_cmd_msg_alloc(itask->itask_proxy_msg_id,
1919 task, dbuf->db_data_size, dbuf->db_sglist[0].seg_addr,
1920 itask->itask_proxy_msg_id);
1921 } else {
1922 ic_cmd_msg = ic_scsi_cmd_msg_alloc(itask->itask_proxy_msg_id,
1923 task, 0, NULL, itask->itask_proxy_msg_id);
1924 }
1925 if (ic_cmd_msg) {
1926 ic_ret = ic_tx_msg(ic_cmd_msg);
1927 if (ic_ret == STMF_IC_MSG_SUCCESS) {
1928 ret = STMF_SUCCESS;
1929 }
1930 }
1931 return (ret);
1932 }
1933
1934
1935 stmf_status_t
1936 pppt_modload()
1937 {
1938 int error;
1939
1940 if (pppt_mod == NULL && ((pppt_mod =
1941 ddi_modopen("drv/pppt", KRTLD_MODE_FIRST, &error)) == NULL)) {
1942 cmn_err(CE_WARN, "Unable to load pppt");
1943 return (STMF_FAILURE);
1944 }
2508 /*
2509 * User is requesting that the token be checked.
2510 * If there was another set after the user's get
2511 * it's an error
2512 */
2513 if (ppi->ppi_token_valid) {
2514 if (ppi->ppi_token != ppd->ppd_token) {
2515 *err_ret = STMF_IOCERR_PPD_UPDATED;
2516 mutex_exit(&stmf_state.stmf_lock);
2517 return (EINVAL);
2518 }
2519 }
2520
2521 if ((ret = nvlist_unpack((char *)ppi->ppi_data,
2522 (size_t)ppi->ppi_data_size, &nv, KM_NOSLEEP)) != 0) {
2523 mutex_exit(&stmf_state.stmf_lock);
2524 return (ret);
2525 }
2526
2527 /* Free any existing lists and add this one to the ppd */
2528 nvlist_free(ppd->ppd_nv);
2529 ppd->ppd_nv = nv;
2530
2531 /* set the token for writes */
2532 ppd->ppd_token++;
2533 /* return token to caller */
2534 if (ppi_token) {
2535 *ppi_token = ppd->ppd_token;
2536 }
2537
2538 /* If there is a provider registered, do the notifications */
2539 if (ppd->ppd_provider) {
2540 uint32_t cb_flags = 0;
2541
2542 if (stmf_state.stmf_config_state == STMF_CONFIG_INIT)
2543 cb_flags |= STMF_PCB_STMF_ONLINING;
2544 if (ppi->ppi_lu_provider) {
2545 ilp = (stmf_i_lu_provider_t *)ppd->ppd_provider;
2546 if (ilp->ilp_lp->lp_cb == NULL)
2547 goto bail_out;
2580 if (ppd->ppd_lu_provider) {
2581 ((stmf_i_lu_provider_t *)
2582 ppd->ppd_provider)->ilp_ppd = NULL;
2583 } else {
2584 ((stmf_i_port_provider_t *)
2585 ppd->ppd_provider)->ipp_ppd = NULL;
2586 }
2587 ppd->ppd_provider = NULL;
2588 }
2589
2590 for (pppd = &stmf_state.stmf_ppdlist; *pppd != NULL;
2591 pppd = &((*pppd)->ppd_next)) {
2592 if (*pppd == ppd)
2593 break;
2594 }
2595
2596 if (*pppd == NULL)
2597 return;
2598
2599 *pppd = ppd->ppd_next;
2600 nvlist_free(ppd->ppd_nv);
2601
2602 kmem_free(ppd, ppd->ppd_alloc_size);
2603 }
2604
2605 int
2606 stmf_delete_ppd_ioctl(stmf_ppioctl_data_t *ppi)
2607 {
2608 stmf_pp_data_t *ppd;
2609 int ret = ENOENT;
2610
2611 if ((ppi->ppi_lu_provider + ppi->ppi_port_provider) != 1) {
2612 return (EINVAL);
2613 }
2614
2615 mutex_enter(&stmf_state.stmf_lock);
2616
2617 for (ppd = stmf_state.stmf_ppdlist; ppd != NULL; ppd = ppd->ppd_next) {
2618 if (ppi->ppi_lu_provider) {
2619 if (!ppd->ppd_lu_provider)
2689 }
2690
2691 void
2692 stmf_delete_all_ppds()
2693 {
2694 stmf_pp_data_t *ppd, *nppd;
2695
2696 ASSERT(mutex_owned(&stmf_state.stmf_lock));
2697 for (ppd = stmf_state.stmf_ppdlist; ppd != NULL; ppd = nppd) {
2698 nppd = ppd->ppd_next;
2699 stmf_delete_ppd(ppd);
2700 }
2701 }
2702
2703 /*
2704 * 16 is the max string length of a protocol_ident, increase
2705 * the size if needed.
2706 */
2707 #define STMF_KSTAT_LU_SZ (STMF_GUID_INPUT + 1 + 256)
2708 #define STMF_KSTAT_TGT_SZ (256 * 2 + 16)
2709
2710 /*
2711 * This array matches the Protocol Identifier in stmf_ioctl.h
2712 */
2713 #define MAX_PROTO_STR_LEN 32
2714
2715 char *protocol_ident[PROTOCOL_ANY] = {
2716 "Fibre Channel",
2717 "Parallel SCSI",
2718 "SSA",
2719 "IEEE_1394",
2720 "SRP",
2721 "iSCSI",
2722 "SAS",
2723 "ADT",
2724 "ATAPI",
2725 "UNKNOWN", "UNKNOWN", "UNKNOWN", "UNKNOWN", "UNKNOWN", "UNKNOWN"
2726 };
2727
2728 /*
2766 }
2767
2768 static void
2769 stmf_update_kstat_lport_io(scsi_task_t *task, stmf_data_buf_t *dbuf)
2770 {
2771 stmf_i_local_port_t *ilp;
2772 kstat_io_t *kip;
2773
2774 ilp = (stmf_i_local_port_t *)task->task_lport->lport_stmf_private;
2775 if (ilp != NULL && ilp->ilport_kstat_io != NULL) {
2776 kip = KSTAT_IO_PTR(ilp->ilport_kstat_io);
2777 if (kip != NULL) {
2778 mutex_enter(ilp->ilport_kstat_io->ks_lock);
2779 STMF_UPDATE_KSTAT_IO(kip, dbuf);
2780 mutex_exit(ilp->ilport_kstat_io->ks_lock);
2781 }
2782 }
2783 }
2784
2785 static void
2786 stmf_update_kstat_lu_io(scsi_task_t *task, stmf_data_buf_t *dbuf)
2787 {
2788 stmf_i_lu_t *ilu;
2789 kstat_io_t *kip;
2790
2791 ilu = (stmf_i_lu_t *)task->task_lu->lu_stmf_private;
2792 if (ilu != NULL && ilu->ilu_kstat_io != NULL) {
2793 kip = KSTAT_IO_PTR(ilu->ilu_kstat_io);
2794 if (kip != NULL) {
2795 mutex_enter(ilu->ilu_kstat_io->ks_lock);
2796 STMF_UPDATE_KSTAT_IO(kip, dbuf);
2797 mutex_exit(ilu->ilu_kstat_io->ks_lock);
2798 }
2799 }
2800 }
2801
2802 static void
2803 stmf_create_kstat_lu(stmf_i_lu_t *ilu)
2804 {
2805 char ks_nm[KSTAT_STRLEN];
3426 alloc_len = sizeof (*irport) + sizeof (scsi_devid_desc_t) +
3427 rport_devid->ident_length - 1;
3428 irport = kmem_zalloc(alloc_len, KM_NOSLEEP);
3429 if (irport == NULL) {
3430 return (NULL);
3431 }
3432
3433 irport->irport_instance =
3434 id_alloc_nosleep(stmf_state.stmf_irport_inst_space);
3435 if (irport->irport_instance == -1) {
3436 kmem_free(irport, alloc_len);
3437 return (NULL);
3438 }
3439
3440 irport->irport_id =
3441 (struct scsi_devid_desc *)(irport + 1); /* Ptr. Arith. */
3442 bcopy(rport_devid, irport->irport_id,
3443 sizeof (scsi_devid_desc_t) + rport_devid->ident_length - 1);
3444 irport->irport_refcnt = 1;
3445 mutex_init(&irport->irport_mutex, NULL, MUTEX_DEFAULT, NULL);
3446
3447 return (irport);
3448 }
3449
3450 static void
3451 stmf_irport_destroy(stmf_i_remote_port_t *irport)
3452 {
3453 id_free(stmf_state.stmf_irport_inst_space, irport->irport_instance);
3454 mutex_destroy(&irport->irport_mutex);
3455 kmem_free(irport, sizeof (*irport) + sizeof (scsi_devid_desc_t) +
3456 irport->irport_id->ident_length - 1);
3457 }
3458
3459 static stmf_i_remote_port_t *
3460 stmf_irport_register(scsi_devid_desc_t *rport_devid)
3461 {
3462 stmf_i_remote_port_t *irport;
3463
3464 mutex_enter(&stmf_state.stmf_lock);
3465
3466 /*
3467 * Lookup will bump the refcnt if there's an existing rport
3468 * context for this identifier.
3469 */
3470 if ((irport = stmf_irport_lookup_locked(rport_devid)) != NULL) {
3471 mutex_exit(&stmf_state.stmf_lock);
3472 return (irport);
3473 }
3474
3475 irport = stmf_irport_create(rport_devid);
3476 if (irport == NULL) {
3477 mutex_exit(&stmf_state.stmf_lock);
3478 return (NULL);
3479 }
3480
3481 avl_add(&stmf_state.stmf_irportlist, irport);
3482 mutex_exit(&stmf_state.stmf_lock);
3483
3484 return (irport);
3485 }
3486
3487 static stmf_i_remote_port_t *
3488 stmf_irport_lookup_locked(scsi_devid_desc_t *rport_devid)
3489 {
3490 stmf_i_remote_port_t *irport;
3491 stmf_i_remote_port_t tmp_irport;
3492
3493 ASSERT(mutex_owned(&stmf_state.stmf_lock));
3494 tmp_irport.irport_id = rport_devid;
3495 irport = avl_find(&stmf_state.stmf_irportlist, &tmp_irport, NULL);
3496 if (irport != NULL) {
3497 mutex_enter(&irport->irport_mutex);
3498 irport->irport_refcnt++;
3499 mutex_exit(&irport->irport_mutex);
3500 }
3584
3585 mutex_enter(&stmf_state.stmf_lock);
3586 rw_enter(&ilport->ilport_lock, RW_WRITER);
3587 (void) stmf_session_create_lun_map(ilport, iss);
3588 ilport->ilport_nsessions++;
3589 iss->iss_next = ilport->ilport_ss_list;
3590 ilport->ilport_ss_list = iss;
3591 rw_exit(&ilport->ilport_lock);
3592 mutex_exit(&stmf_state.stmf_lock);
3593
3594 iss->iss_creation_time = ddi_get_time();
3595 ss->ss_session_id = atomic_inc_64_nv(&stmf_session_counter);
3596 iss->iss_flags &= ~ISS_BEING_CREATED;
3597 /* XXX should we remove ISS_LUN_INVENTORY_CHANGED on new session? */
3598 iss->iss_flags &= ~ISS_LUN_INVENTORY_CHANGED;
3599 DTRACE_PROBE2(session__online, stmf_local_port_t *, lport,
3600 stmf_scsi_session_t *, ss);
3601 return (STMF_SUCCESS);
3602 }
3603
3604 void
3605 stmf_deregister_scsi_session(stmf_local_port_t *lport, stmf_scsi_session_t *ss)
3606 {
3607 stmf_i_local_port_t *ilport = (stmf_i_local_port_t *)
3608 lport->lport_stmf_private;
3609 stmf_i_scsi_session_t *iss, **ppss;
3610 int found = 0;
3611 stmf_ic_msg_t *ic_session_dereg;
3612 stmf_status_t ic_ret = STMF_FAILURE;
3613
3614 DTRACE_PROBE2(session__offline, stmf_local_port_t *, lport,
3615 stmf_scsi_session_t *, ss);
3616
3617 iss = (stmf_i_scsi_session_t *)ss->ss_stmf_private;
3618 if (ss->ss_rport_alias) {
3619 ss->ss_rport_alias = NULL;
3620 }
3621
3622 try_dereg_ss_again:
3623 mutex_enter(&stmf_state.stmf_lock);
3624 atomic_and_32(&iss->iss_flags,
3625 ~(ISS_LUN_INVENTORY_CHANGED | ISS_GOT_INITIAL_LUNS));
3626 if (iss->iss_flags & ISS_EVENT_ACTIVE) {
3627 mutex_exit(&stmf_state.stmf_lock);
3628 delay(1);
3629 goto try_dereg_ss_again;
3630 }
3631
3632 /* dereg proxy session if not standby port */
3642 }
3643 }
3644 }
3645
3646 rw_enter(&ilport->ilport_lock, RW_WRITER);
3647 for (ppss = &ilport->ilport_ss_list; *ppss != NULL;
3648 ppss = &((*ppss)->iss_next)) {
3649 if (iss == (*ppss)) {
3650 *ppss = (*ppss)->iss_next;
3651 found = 1;
3652 break;
3653 }
3654 }
3655 if (!found) {
3656 cmn_err(CE_PANIC, "Deregister session called for non existent"
3657 " session");
3658 }
3659 ilport->ilport_nsessions--;
3660
3661 stmf_irport_deregister(iss->iss_irport);
3662 (void) stmf_session_destroy_lun_map(ilport, iss);
3663 rw_exit(&ilport->ilport_lock);
3664 mutex_exit(&stmf_state.stmf_lock);
3665
3666 if (iss->iss_flags & ISS_NULL_TPTID) {
3667 stmf_remote_port_free(ss->ss_rport);
3668 }
3669 }
3670
3671 stmf_i_scsi_session_t *
3672 stmf_session_id_to_issptr(uint64_t session_id, int stay_locked)
3673 {
3674 stmf_i_local_port_t *ilport;
3675 stmf_i_scsi_session_t *iss;
3676
3677 mutex_enter(&stmf_state.stmf_lock);
3678 for (ilport = stmf_state.stmf_ilportlist; ilport != NULL;
3679 ilport = ilport->ilport_next) {
3680 rw_enter(&ilport->ilport_lock, RW_WRITER);
3681 for (iss = ilport->ilport_ss_list; iss != NULL;
3682 iss = iss->iss_next) {
3683 if (iss->iss_ss->ss_session_id == session_id) {
3684 if (!stay_locked)
3685 rw_exit(&ilport->ilport_lock);
3686 mutex_exit(&stmf_state.stmf_lock);
3687 return (iss);
3688 }
3689 }
3690 rw_exit(&ilport->ilport_lock);
3846 goto dai_scan_done;
3847 }
3848 }
3849 } /* lun table for a session */
3850 } /* sessions */
3851 rw_exit(&ilport->ilport_lock);
3852 } /* ports */
3853
3854 dai_scan_done:
3855 mutex_exit(&stmf_state.stmf_lock);
3856
3857 for (i = 0; i < nu; i++) {
3858 stmf_do_itl_dereg(lu, itl_list[i],
3859 STMF_ITL_REASON_DEREG_REQUEST);
3860 }
3861 kmem_free(itl_list, nmaps * sizeof (stmf_itl_data_t *));
3862
3863 return (STMF_SUCCESS);
3864 }
3865
3866 stmf_status_t
3867 stmf_get_itl_handle(stmf_lu_t *lu, uint8_t *lun, stmf_scsi_session_t *ss,
3868 uint64_t session_id, void **itl_handle_retp)
3869 {
3870 stmf_i_scsi_session_t *iss;
3871 stmf_lun_map_ent_t *ent;
3872 stmf_lun_map_t *lm;
3873 stmf_status_t ret;
3874 int i;
3875 uint16_t n;
3876
3877 if (ss == NULL) {
3878 iss = stmf_session_id_to_issptr(session_id, 1);
3879 if (iss == NULL)
3880 return (STMF_NOT_FOUND);
3881 } else {
3882 iss = (stmf_i_scsi_session_t *)ss->ss_stmf_private;
3883 rw_enter(iss->iss_lockp, RW_WRITER);
3884 }
3885
3886 ent = NULL;
3887 if (lun == NULL) {
3888 lm = iss->iss_sm;
3889 for (i = 0; i < lm->lm_nentries; i++) {
3890 if (lm->lm_plus[i] == NULL)
3891 continue;
3892 ent = (stmf_lun_map_ent_t *)lm->lm_plus[i];
3893 if (ent->ent_lu == lu)
3894 break;
3895 }
3896 } else {
3897 n = ((uint16_t)lun[1] | (((uint16_t)(lun[0] & 0x3F)) << 8));
3898 ent = (stmf_lun_map_ent_t *)
3899 stmf_get_ent_from_map(iss->iss_sm, n);
3900 if (lu && (ent->ent_lu != lu))
3901 ent = NULL;
3902 }
3903 if (ent && ent->ent_itl_datap) {
3904 *itl_handle_retp = ent->ent_itl_datap->itl_handle;
3905 ret = STMF_SUCCESS;
3906 } else {
3907 ret = STMF_NOT_FOUND;
3908 }
3909
3910 rw_exit(iss->iss_lockp);
3911 return (ret);
3912 }
3913
3914 stmf_data_buf_t *
3915 stmf_alloc_dbuf(scsi_task_t *task, uint32_t size, uint32_t *pminsize,
3916 uint32_t flags)
3917 {
3918 stmf_i_scsi_task_t *itask =
3919 (stmf_i_scsi_task_t *)task->task_stmf_private;
3920 stmf_local_port_t *lport = task->task_lport;
3921 stmf_data_buf_t *dbuf;
3922 uint8_t ndx;
3923
3924 ndx = stmf_first_zero[itask->itask_allocated_buf_map];
3925 if (ndx == 0xff)
3926 return (NULL);
3927 dbuf = itask->itask_dbufs[ndx] = lport->lport_ds->ds_alloc_data_buf(
3928 task, size, pminsize, flags);
3929 if (dbuf) {
3930 task->task_cur_nbufs++;
3931 itask->itask_allocated_buf_map |= (1 << ndx);
3932 dbuf->db_flags &= ~DB_LPORT_XFER_ACTIVE;
3933 dbuf->db_handle = ndx;
4027
4028 /*
4029 * We allocate 7 extra bytes for CDB to provide a cdb pointer which
4030 * is guaranteed to be 8 byte aligned. Some LU providers like OSD
4031 * depend upon this alignment.
4032 */
4033 if (cdb_length_in >= 16)
4034 cdb_length = cdb_length_in + 7;
4035 else
4036 cdb_length = 16 + 7;
4037 iss = (stmf_i_scsi_session_t *)ss->ss_stmf_private;
4038 luNbr = ((uint16_t)lun[1] | (((uint16_t)(lun[0] & 0x3F)) << 8));
4039 rw_enter(iss->iss_lockp, RW_READER);
4040 lun_map_ent =
4041 (stmf_lun_map_ent_t *)stmf_get_ent_from_map(iss->iss_sm, luNbr);
4042 if (!lun_map_ent) {
4043 lu = dlun0;
4044 } else {
4045 lu = lun_map_ent->ent_lu;
4046 }
4047 ilu = lu->lu_stmf_private;
4048 if (ilu->ilu_flags & ILU_RESET_ACTIVE) {
4049 rw_exit(iss->iss_lockp);
4050 return (NULL);
4051 }
4052 ASSERT(lu == dlun0 || (ilu->ilu_state != STMF_STATE_OFFLINING &&
4053 ilu->ilu_state != STMF_STATE_OFFLINE));
4054 do {
4055 if (ilu->ilu_free_tasks == NULL) {
4056 new_task = 1;
4057 break;
4058 }
4059 mutex_enter(&ilu->ilu_task_lock);
4060 for (ppitask = &ilu->ilu_free_tasks; (*ppitask != NULL) &&
4061 ((*ppitask)->itask_cdb_buf_size < cdb_length);
4062 ppitask = &((*ppitask)->itask_lu_free_next))
4063 ;
4064 if (*ppitask) {
4065 itask = *ppitask;
4066 *ppitask = (*ppitask)->itask_lu_free_next;
4067 ilu->ilu_ntasks_free--;
4068 if (ilu->ilu_ntasks_free < ilu->ilu_ntasks_min_free)
4069 ilu->ilu_ntasks_min_free = ilu->ilu_ntasks_free;
4070 } else {
4071 new_task = 1;
4072 }
4073 mutex_exit(&ilu->ilu_task_lock);
4081 * selection process above.
4082 */
4083 uint8_t *save_cdb;
4084 uintptr_t t_start, t_end;
4085
4086 task = itask->itask_task;
4087 save_cdb = task->task_cdb; /* save */
4088 t_start = (uintptr_t)&task->task_flags;
4089 t_end = (uintptr_t)&task->task_extended_cmd;
4090 bzero((void *)t_start, (size_t)(t_end - t_start));
4091 task->task_cdb = save_cdb; /* restore */
4092 itask->itask_ncmds = 0;
4093 } else {
4094 task = (scsi_task_t *)stmf_alloc(STMF_STRUCT_SCSI_TASK,
4095 cdb_length, AF_FORCE_NOSLEEP);
4096 if (task == NULL) {
4097 rw_exit(iss->iss_lockp);
4098 return (NULL);
4099 }
4100 task->task_lu = lu;
4101 l = task->task_lun_no;
4102 l[0] = lun[0];
4103 l[1] = lun[1];
4104 l[2] = lun[2];
4105 l[3] = lun[3];
4106 l[4] = lun[4];
4107 l[5] = lun[5];
4108 l[6] = lun[6];
4109 l[7] = lun[7];
4110 task->task_cdb = (uint8_t *)task->task_port_private;
4111 if ((ulong_t)(task->task_cdb) & 7ul) {
4112 task->task_cdb = (uint8_t *)(((ulong_t)
4113 (task->task_cdb) + 7ul) & ~(7ul));
4114 }
4115 itask = (stmf_i_scsi_task_t *)task->task_stmf_private;
4116 itask->itask_cdb_buf_size = cdb_length;
4117 mutex_init(&itask->itask_audit_mutex, NULL, MUTEX_DRIVER, NULL);
4118 }
4119 task->task_session = ss;
4120 task->task_lport = lport;
4121 task->task_cdb_length = cdb_length_in;
4122 itask->itask_flags = ITASK_IN_TRANSITION;
4123 itask->itask_waitq_time = 0;
4124 itask->itask_lu_read_time = itask->itask_lu_write_time = 0;
4125 itask->itask_lport_read_time = itask->itask_lport_write_time = 0;
4126 itask->itask_read_xfer = itask->itask_write_xfer = 0;
4127 itask->itask_audit_index = 0;
4128
4129 if (new_task) {
4130 if (lu->lu_task_alloc(task) != STMF_SUCCESS) {
4131 rw_exit(iss->iss_lockp);
4132 stmf_free(task);
4133 return (NULL);
4134 }
4135 mutex_enter(&ilu->ilu_task_lock);
4136 if (ilu->ilu_flags & ILU_RESET_ACTIVE) {
4137 mutex_exit(&ilu->ilu_task_lock);
4138 rw_exit(iss->iss_lockp);
4139 stmf_free(task);
4140 return (NULL);
4141 }
4142 itask->itask_lu_next = ilu->ilu_tasks;
4143 if (ilu->ilu_tasks)
4144 ilu->ilu_tasks->itask_lu_prev = itask;
4145 ilu->ilu_tasks = itask;
4146 /* kmem_zalloc automatically makes itask->itask_lu_prev NULL */
4147 ilu->ilu_ntasks++;
4148 mutex_exit(&ilu->ilu_task_lock);
4149 }
4150
4151 itask->itask_ilu_task_cntr = ilu->ilu_cur_task_cntr;
4152 atomic_inc_32(itask->itask_ilu_task_cntr);
4153 itask->itask_start_time = ddi_get_lbolt();
4154
4155 if ((lun_map_ent != NULL) && ((itask->itask_itl_datap =
4156 lun_map_ent->ent_itl_datap) != NULL)) {
4157 atomic_inc_32(&itask->itask_itl_datap->itl_counter);
4158 task->task_lu_itl_handle = itask->itask_itl_datap->itl_handle;
4159 } else {
4160 itask->itask_itl_datap = NULL;
4161 task->task_lu_itl_handle = NULL;
4162 }
4163
4164 rw_exit(iss->iss_lockp);
4165 return (task);
4166 }
4167
4168 static void
4169 stmf_task_lu_free(scsi_task_t *task, stmf_i_scsi_session_t *iss)
4170 {
4171 stmf_i_scsi_task_t *itask =
4172 (stmf_i_scsi_task_t *)task->task_stmf_private;
4173 stmf_i_lu_t *ilu = (stmf_i_lu_t *)task->task_lu->lu_stmf_private;
4174
4175 ASSERT(rw_lock_held(iss->iss_lockp));
4176 itask->itask_flags = ITASK_IN_FREE_LIST;
4177 itask->itask_proxy_msg_id = 0;
4178 mutex_enter(&ilu->ilu_task_lock);
4179 itask->itask_lu_free_next = ilu->ilu_free_tasks;
4180 ilu->ilu_free_tasks = itask;
4181 ilu->ilu_ntasks_free++;
4182 if (ilu->ilu_ntasks == ilu->ilu_ntasks_free)
4183 cv_signal(&ilu->ilu_offline_pending_cv);
4184 mutex_exit(&ilu->ilu_task_lock);
4185 atomic_dec_32(itask->itask_ilu_task_cntr);
4186 }
4187
4188 void
4189 stmf_task_lu_check_freelist(stmf_i_lu_t *ilu)
4190 {
4191 uint32_t num_to_release, ndx;
4192 stmf_i_scsi_task_t *itask;
4193 stmf_lu_t *lu = ilu->ilu_lu;
4194
4195 ASSERT(ilu->ilu_ntasks_min_free <= ilu->ilu_ntasks_free);
4196
4197 /* free half of the minimal free of the free tasks */
4198 num_to_release = (ilu->ilu_ntasks_min_free + 1) / 2;
4199 if (!num_to_release) {
4200 return;
4201 }
4202 for (ndx = 0; ndx < num_to_release; ndx++) {
4203 mutex_enter(&ilu->ilu_task_lock);
4204 itask = ilu->ilu_free_tasks;
4205 if (itask == NULL) {
4242 if (!ilu->ilu_ntasks_min_free) {
4243 ilu->ilu_ntasks_min_free = ilu->ilu_ntasks_free;
4244 continue;
4245 }
4246 ilu->ilu_flags |= ILU_STALL_DEREGISTER;
4247 mutex_exit(&stmf_state.stmf_lock);
4248 stmf_task_lu_check_freelist(ilu);
4249 /*
4250 * we do not care about the accuracy of
4251 * ilu_ntasks_min_free, so we don't lock here
4252 */
4253 ilu->ilu_ntasks_min_free = ilu->ilu_ntasks_free;
4254 mutex_enter(&stmf_state.stmf_lock);
4255 ilu->ilu_flags &= ~ILU_STALL_DEREGISTER;
4256 cv_broadcast(&stmf_state.stmf_cv);
4257 if (ddi_get_lbolt() >= endtime)
4258 break;
4259 }
4260 }
4261
4262 void
4263 stmf_do_ilu_timeouts(stmf_i_lu_t *ilu)
4264 {
4265 clock_t l = ddi_get_lbolt();
4266 clock_t ps = drv_usectohz(1000000);
4267 stmf_i_scsi_task_t *itask;
4268 scsi_task_t *task;
4269 uint32_t to;
4270
4271 mutex_enter(&ilu->ilu_task_lock);
4272 for (itask = ilu->ilu_tasks; itask != NULL;
4273 itask = itask->itask_lu_next) {
4274 if (itask->itask_flags & (ITASK_IN_FREE_LIST |
4275 ITASK_BEING_ABORTED)) {
4276 continue;
4277 }
4278 task = itask->itask_task;
4279 if (task->task_timeout == 0)
4280 to = stmf_default_task_timeout;
4281 else
4282 to = task->task_timeout;
4283 if ((itask->itask_start_time + (to * ps)) > l)
4284 continue;
4285 stmf_abort(STMF_QUEUE_TASK_ABORT, task,
4286 STMF_TIMEOUT, NULL);
4287 }
4288 mutex_exit(&ilu->ilu_task_lock);
4289 }
4290
4291 /*
4292 * Called with stmf_lock held
4293 */
4294 void
4295 stmf_check_ilu_timing()
4296 {
4297 stmf_i_lu_t *ilu;
4298 clock_t endtime = ddi_get_lbolt() + drv_usectohz(10000);
4299
4300 /* stmf_svc_ilu_timing may get changed after stmf_lock is released */
4301 while ((ilu = stmf_state.stmf_svc_ilu_timing) != NULL) {
4302 stmf_state.stmf_svc_ilu_timing = ilu->ilu_next;
4303 if (ilu->ilu_cur_task_cntr == (&ilu->ilu_task_cntr1)) {
4304 if (ilu->ilu_task_cntr2 == 0) {
4319 mutex_exit(&stmf_state.stmf_lock);
4320 stmf_do_ilu_timeouts(ilu);
4321 mutex_enter(&stmf_state.stmf_lock);
4322 ilu->ilu_flags &= ~ILU_STALL_DEREGISTER;
4323 cv_broadcast(&stmf_state.stmf_cv);
4324 if (ddi_get_lbolt() >= endtime)
4325 break;
4326 }
4327 }
4328
4329 /*
4330 * Kills all tasks on a lu except tm_task
4331 */
4332 void
4333 stmf_task_lu_killall(stmf_lu_t *lu, scsi_task_t *tm_task, stmf_status_t s)
4334 {
4335 stmf_i_lu_t *ilu = (stmf_i_lu_t *)lu->lu_stmf_private;
4336 stmf_i_scsi_task_t *itask;
4337
4338 mutex_enter(&ilu->ilu_task_lock);
4339
4340 for (itask = ilu->ilu_tasks; itask != NULL;
4341 itask = itask->itask_lu_next) {
4342 if (itask->itask_flags & ITASK_IN_FREE_LIST)
4343 continue;
4344 if (itask->itask_task == tm_task)
4345 continue;
4346 stmf_abort(STMF_QUEUE_TASK_ABORT, itask->itask_task, s, NULL);
4347 }
4348 mutex_exit(&ilu->ilu_task_lock);
4349 }
4350
4351 void
4352 stmf_free_task_bufs(stmf_i_scsi_task_t *itask, stmf_local_port_t *lport)
4353 {
4354 int i;
4355 uint8_t map;
4356
4357 if ((map = itask->itask_allocated_buf_map) == 0)
4358 return;
4359 for (i = 0; i < 4; i++) {
4360 if (map & 1) {
4361 stmf_data_buf_t *dbuf;
4362
4363 dbuf = itask->itask_dbufs[i];
4379 } else {
4380 ASSERT(dbuf->db_lu_private == NULL);
4381 dbuf->db_lu_private = NULL;
4382 lport->lport_ds->ds_free_data_buf(
4383 lport->lport_ds, dbuf);
4384 }
4385 }
4386 map >>= 1;
4387 }
4388 itask->itask_allocated_buf_map = 0;
4389 }
4390
4391 void
4392 stmf_task_free(scsi_task_t *task)
4393 {
4394 stmf_local_port_t *lport = task->task_lport;
4395 stmf_i_scsi_task_t *itask = (stmf_i_scsi_task_t *)
4396 task->task_stmf_private;
4397 stmf_i_scsi_session_t *iss = (stmf_i_scsi_session_t *)
4398 task->task_session->ss_stmf_private;
4399
4400 stmf_task_audit(itask, TE_TASK_FREE, CMD_OR_IOF_NA, NULL);
4401
4402 stmf_free_task_bufs(itask, lport);
4403 stmf_itl_task_done(itask);
4404 DTRACE_PROBE2(stmf__task__end, scsi_task_t *, task,
4405 hrtime_t,
4406 itask->itask_done_timestamp - itask->itask_start_timestamp);
4407 if (itask->itask_itl_datap) {
4408 if (atomic_dec_32_nv(&itask->itask_itl_datap->itl_counter) ==
4409 0) {
4410 stmf_release_itl_handle(task->task_lu,
4411 itask->itask_itl_datap);
4412 }
4413 }
4414
4415 rw_enter(iss->iss_lockp, RW_READER);
4416 lport->lport_task_free(task);
4417 if (itask->itask_worker) {
4418 atomic_dec_32(&stmf_cur_ntasks);
4419 atomic_dec_32(&itask->itask_worker->worker_ref_count);
4420 }
4421 /*
4422 * After calling stmf_task_lu_free, the task pointer can no longer
4423 * be trusted.
4424 */
4425 stmf_task_lu_free(task, iss);
4426 rw_exit(iss->iss_lockp);
4427 }
4428
4429 void
4430 stmf_post_task(scsi_task_t *task, stmf_data_buf_t *dbuf)
4431 {
4432 stmf_i_scsi_task_t *itask = (stmf_i_scsi_task_t *)
4433 task->task_stmf_private;
4434 stmf_i_lu_t *ilu = (stmf_i_lu_t *)task->task_lu->lu_stmf_private;
4435 int nv;
4436 uint32_t old, new;
4437 uint32_t ct;
4438 stmf_worker_t *w, *w1;
4439 uint8_t tm;
4440
4441 if (task->task_max_nbufs > 4)
4442 task->task_max_nbufs = 4;
4443 task->task_cur_nbufs = 0;
4444 /* Latest value of currently running tasks */
4445 ct = atomic_inc_32_nv(&stmf_cur_ntasks);
4446
4447 /* Select the next worker using round robin */
4448 nv = (int)atomic_inc_32_nv((uint32_t *)&stmf_worker_sel_counter);
4449 if (nv >= stmf_nworkers_accepting_cmds) {
4450 int s = nv;
4451 do {
4452 nv -= stmf_nworkers_accepting_cmds;
4453 } while (nv >= stmf_nworkers_accepting_cmds);
4454 if (nv < 0)
4455 nv = 0;
4456 /* Its ok if this cas fails */
4457 (void) atomic_cas_32((uint32_t *)&stmf_worker_sel_counter,
4458 s, nv);
4459 }
4460 w = &stmf_workers[nv];
4461
4462 /*
4463 * A worker can be pinned by interrupt. So select the next one
4464 * if it has lower load.
4465 */
4466 if ((nv + 1) >= stmf_nworkers_accepting_cmds) {
4467 w1 = stmf_workers;
4468 } else {
4469 w1 = &stmf_workers[nv + 1];
4470 }
4471 if (w1->worker_queue_depth < w->worker_queue_depth)
4472 w = w1;
4473
4474 mutex_enter(&w->worker_lock);
4475 if (((w->worker_flags & STMF_WORKER_STARTED) == 0) ||
4476 (w->worker_flags & STMF_WORKER_TERMINATE)) {
4477 /*
4478 * Maybe we are in the middle of a change. Just go to
4479 * the 1st worker.
4480 */
4481 mutex_exit(&w->worker_lock);
4482 w = stmf_workers;
4483 mutex_enter(&w->worker_lock);
4484 }
4485 itask->itask_worker = w;
4486 /*
4487 * Track max system load inside the worker as we already have the
4488 * worker lock (no point implementing another lock). The service
4489 * thread will do the comparisons and figure out the max overall
4490 * system load.
4491 */
4492 if (w->worker_max_sys_qdepth_pu < ct)
4493 w->worker_max_sys_qdepth_pu = ct;
4494
4495 do {
4496 old = new = itask->itask_flags;
4497 new |= ITASK_KNOWN_TO_TGT_PORT | ITASK_IN_WORKER_QUEUE;
4498 if (task->task_mgmt_function) {
4499 tm = task->task_mgmt_function;
4500 if ((tm == TM_TARGET_RESET) ||
4501 (tm == TM_TARGET_COLD_RESET) ||
4502 (tm == TM_TARGET_WARM_RESET)) {
4503 new |= ITASK_DEFAULT_HANDLING;
4504 }
4505 } else if (task->task_cdb[0] == SCMD_REPORT_LUNS) {
4506 new |= ITASK_DEFAULT_HANDLING;
4507 }
4508 new &= ~ITASK_IN_TRANSITION;
4509 } while (atomic_cas_32(&itask->itask_flags, old, new) != old);
4510
4511 stmf_itl_task_start(itask);
4512
4513 itask->itask_worker_next = NULL;
4514 if (w->worker_task_tail) {
4515 w->worker_task_tail->itask_worker_next = itask;
4516 } else {
4517 w->worker_task_head = itask;
4518 }
4519 w->worker_task_tail = itask;
4520 if (++(w->worker_queue_depth) > w->worker_max_qdepth_pu) {
4521 w->worker_max_qdepth_pu = w->worker_queue_depth;
4522 }
4523 /* Measure task waitq time */
4524 itask->itask_waitq_enter_timestamp = gethrtime();
4525 atomic_inc_32(&w->worker_ref_count);
4526 itask->itask_cmd_stack[0] = ITASK_CMD_NEW_TASK;
4527 itask->itask_ncmds = 1;
4528 stmf_task_audit(itask, TE_TASK_START, CMD_OR_IOF_NA, dbuf);
4529 if (dbuf) {
4530 itask->itask_allocated_buf_map = 1;
4531 itask->itask_dbufs[0] = dbuf;
4532 dbuf->db_handle = 0;
4533 } else {
4534 itask->itask_allocated_buf_map = 0;
4535 itask->itask_dbufs[0] = NULL;
4536 }
4537
4538 if ((w->worker_flags & STMF_WORKER_ACTIVE) == 0) {
4539 w->worker_signal_timestamp = gethrtime();
4540 DTRACE_PROBE2(worker__signal, stmf_worker_t *, w,
4541 scsi_task_t *, task);
4542 cv_signal(&w->worker_cv);
4543 }
4544 mutex_exit(&w->worker_lock);
4545
4546 /*
4547 * This can only happen if during stmf_task_alloc(), ILU_RESET_ACTIVE
4548 * was set between checking of ILU_RESET_ACTIVE and clearing of the
4549 * ITASK_IN_FREE_LIST flag. Take care of these "sneaked-in" tasks here.
4550 */
4551 if (ilu->ilu_flags & ILU_RESET_ACTIVE) {
4552 stmf_abort(STMF_QUEUE_TASK_ABORT, task, STMF_ABORTED, NULL);
4553 }
4554 }
4555
4556 static void
4557 stmf_task_audit(stmf_i_scsi_task_t *itask,
4558 task_audit_event_t te, uint32_t cmd_or_iof, stmf_data_buf_t *dbuf)
4559 {
4560 stmf_task_audit_rec_t *ar;
4561
4562 mutex_enter(&itask->itask_audit_mutex);
4563 ar = &itask->itask_audit_records[itask->itask_audit_index++];
4564 itask->itask_audit_index &= (ITASK_TASK_AUDIT_DEPTH - 1);
4580 * we will only call that entry point if ITASK_KNOWN_TO_LU was set.
4581 *
4582 * Same logic applies for the port.
4583 *
4584 * Also ITASK_BEING_ABORTED will not be allowed to set if both KNOWN_TO_LU
4585 * and KNOWN_TO_TGT_PORT are reset.
4586 *
4587 * +++++++++++++++++++++++++++++++++++++++++++++++
4588 */
4589
4590 stmf_status_t
4591 stmf_xfer_data(scsi_task_t *task, stmf_data_buf_t *dbuf, uint32_t ioflags)
4592 {
4593 stmf_status_t ret = STMF_SUCCESS;
4594
4595 stmf_i_scsi_task_t *itask =
4596 (stmf_i_scsi_task_t *)task->task_stmf_private;
4597
4598 stmf_task_audit(itask, TE_XFER_START, ioflags, dbuf);
4599
4600 if (ioflags & STMF_IOF_LU_DONE) {
4601 uint32_t new, old;
4602 do {
4603 new = old = itask->itask_flags;
4604 if (new & ITASK_BEING_ABORTED)
4605 return (STMF_ABORTED);
4606 new &= ~ITASK_KNOWN_TO_LU;
4607 } while (atomic_cas_32(&itask->itask_flags, old, new) != old);
4608 }
4609 if (itask->itask_flags & ITASK_BEING_ABORTED)
4610 return (STMF_ABORTED);
4611 #ifdef DEBUG
4612 if (!(ioflags & STMF_IOF_STATS_ONLY) && stmf_drop_buf_counter > 0) {
4613 if (atomic_dec_32_nv(&stmf_drop_buf_counter) == 1)
4614 return (STMF_SUCCESS);
4615 }
4616 #endif
4617
4618 stmf_update_kstat_lu_io(task, dbuf);
4619 stmf_update_kstat_lport_io(task, dbuf);
4620 stmf_lport_xfer_start(itask, dbuf);
4621 if (ioflags & STMF_IOF_STATS_ONLY) {
4622 stmf_lport_xfer_done(itask, dbuf);
4623 return (STMF_SUCCESS);
4624 }
4625
4626 dbuf->db_flags |= DB_LPORT_XFER_ACTIVE;
4627 ret = task->task_lport->lport_xfer_data(task, dbuf, ioflags);
4628
4629 /*
4630 * Port provider may have already called the buffer callback in
4631 * which case dbuf->db_xfer_start_timestamp will be 0.
4632 */
4633 if (ret != STMF_SUCCESS) {
4634 dbuf->db_flags &= ~DB_LPORT_XFER_ACTIVE;
4635 if (dbuf->db_xfer_start_timestamp != 0)
4636 stmf_lport_xfer_done(itask, dbuf);
4637 }
4638
4639 return (ret);
4640 }
4641
4642 void
4643 stmf_data_xfer_done(scsi_task_t *task, stmf_data_buf_t *dbuf, uint32_t iof)
4644 {
4645 stmf_i_scsi_task_t *itask =
4646 (stmf_i_scsi_task_t *)task->task_stmf_private;
4647 stmf_i_local_port_t *ilport;
4648 stmf_worker_t *w = itask->itask_worker;
4649 uint32_t new, old;
4650 uint8_t update_queue_flags, free_it, queue_it;
4651
4652 stmf_lport_xfer_done(itask, dbuf);
4653
4654 stmf_task_audit(itask, TE_XFER_DONE, iof, dbuf);
4655
4656 /* Guard against unexpected completions from the lport */
4657 if (dbuf->db_flags & DB_LPORT_XFER_ACTIVE) {
4658 dbuf->db_flags &= ~DB_LPORT_XFER_ACTIVE;
4659 } else {
4660 /*
4661 * This should never happen.
4662 */
4663 ilport = task->task_lport->lport_stmf_private;
4664 ilport->ilport_unexpected_comp++;
4665 cmn_err(CE_PANIC, "Unexpected xfer completion task %p dbuf %p",
4666 (void *)task, (void *)dbuf);
4667 return;
4668 }
4669
4670 mutex_enter(&w->worker_lock);
4671 do {
4672 new = old = itask->itask_flags;
4673 if (old & ITASK_BEING_ABORTED) {
4674 mutex_exit(&w->worker_lock);
4675 return;
4676 }
4677 free_it = 0;
4678 if (iof & STMF_IOF_LPORT_DONE) {
4679 new &= ~ITASK_KNOWN_TO_TGT_PORT;
4680 task->task_completion_status = dbuf->db_xfer_status;
4681 free_it = 1;
4682 }
4683 /*
4684 * If the task is known to LU then queue it. But if
4685 * it is already queued (multiple completions) then
4686 * just update the buffer information by grabbing the
4687 * worker lock. If the task is not known to LU,
4688 * completed/aborted, then see if we need to
4689 * free this task.
4690 */
4691 if (old & ITASK_KNOWN_TO_LU) {
4692 free_it = 0;
4693 update_queue_flags = 1;
4694 if (old & ITASK_IN_WORKER_QUEUE) {
4695 queue_it = 0;
4696 } else {
4697 queue_it = 1;
4698 new |= ITASK_IN_WORKER_QUEUE;
4699 }
4700 } else {
4701 update_queue_flags = 0;
4702 queue_it = 0;
4703 }
4704 } while (atomic_cas_32(&itask->itask_flags, old, new) != old);
4705
4706 if (update_queue_flags) {
4707 uint8_t cmd = (dbuf->db_handle << 5) | ITASK_CMD_DATA_XFER_DONE;
4708
4709 ASSERT(itask->itask_ncmds < ITASK_MAX_NCMDS);
4710 itask->itask_cmd_stack[itask->itask_ncmds++] = cmd;
4711 if (queue_it) {
4712 itask->itask_worker_next = NULL;
4713 if (w->worker_task_tail) {
4714 w->worker_task_tail->itask_worker_next = itask;
4715 } else {
4716 w->worker_task_head = itask;
4717 }
4718 w->worker_task_tail = itask;
4719 /* Measure task waitq time */
4720 itask->itask_waitq_enter_timestamp = gethrtime();
4721 if (++(w->worker_queue_depth) >
4722 w->worker_max_qdepth_pu) {
4723 w->worker_max_qdepth_pu = w->worker_queue_depth;
4724 }
4725 if ((w->worker_flags & STMF_WORKER_ACTIVE) == 0)
4726 cv_signal(&w->worker_cv);
4727 }
4728 }
4729 mutex_exit(&w->worker_lock);
4730
4731 if (free_it) {
4732 if ((itask->itask_flags & (ITASK_KNOWN_TO_LU |
4733 ITASK_KNOWN_TO_TGT_PORT | ITASK_IN_WORKER_QUEUE |
4734 ITASK_BEING_ABORTED)) == 0) {
4735 stmf_task_free(task);
4736 }
4737 }
4738 }
4739
4740 stmf_status_t
4741 stmf_send_scsi_status(scsi_task_t *task, uint32_t ioflags)
4742 {
4743 DTRACE_PROBE1(scsi__send__status, scsi_task_t *, task);
4744
4745 stmf_i_scsi_task_t *itask =
4746 (stmf_i_scsi_task_t *)task->task_stmf_private;
4747
4748 stmf_task_audit(itask, TE_SEND_STATUS, ioflags, NULL);
4749
4750 if (ioflags & STMF_IOF_LU_DONE) {
4751 uint32_t new, old;
4752 do {
4753 new = old = itask->itask_flags;
4754 if (new & ITASK_BEING_ABORTED)
4755 return (STMF_ABORTED);
4756 new &= ~ITASK_KNOWN_TO_LU;
4757 } while (atomic_cas_32(&itask->itask_flags, old, new) != old);
4758 }
4759
4760 if (!(itask->itask_flags & ITASK_KNOWN_TO_TGT_PORT)) {
4761 return (STMF_SUCCESS);
4762 }
4763
4764 if (itask->itask_flags & ITASK_BEING_ABORTED)
4765 return (STMF_ABORTED);
4766
4767 if (task->task_additional_flags & TASK_AF_NO_EXPECTED_XFER_LENGTH) {
4768 task->task_status_ctrl = 0;
4769 task->task_resid = 0;
4770 } else if (task->task_cmd_xfer_length >
4771 task->task_expected_xfer_length) {
4772 task->task_status_ctrl = TASK_SCTRL_OVER;
4773 task->task_resid = task->task_cmd_xfer_length -
4774 task->task_expected_xfer_length;
4775 } else if (task->task_nbytes_transferred <
4776 task->task_expected_xfer_length) {
4777 task->task_status_ctrl = TASK_SCTRL_UNDER;
4778 task->task_resid = task->task_expected_xfer_length -
4779 task->task_nbytes_transferred;
4780 } else {
4781 task->task_status_ctrl = 0;
4782 task->task_resid = 0;
4783 }
4784 return (task->task_lport->lport_send_status(task, ioflags));
4785 }
4786
4787 void
4788 stmf_send_status_done(scsi_task_t *task, stmf_status_t s, uint32_t iof)
4789 {
4790 stmf_i_scsi_task_t *itask =
4791 (stmf_i_scsi_task_t *)task->task_stmf_private;
4792 stmf_worker_t *w = itask->itask_worker;
4793 uint32_t new, old;
4794 uint8_t free_it, queue_it;
4795
4796 stmf_task_audit(itask, TE_SEND_STATUS_DONE, iof, NULL);
4797
4798 mutex_enter(&w->worker_lock);
4799 do {
4800 new = old = itask->itask_flags;
4801 if (old & ITASK_BEING_ABORTED) {
4802 mutex_exit(&w->worker_lock);
4803 return;
4804 }
4805 free_it = 0;
4806 if (iof & STMF_IOF_LPORT_DONE) {
4807 new &= ~ITASK_KNOWN_TO_TGT_PORT;
4808 free_it = 1;
4809 }
4810 /*
4811 * If the task is known to LU then queue it. But if
4812 * it is already queued (multiple completions) then
4813 * just update the buffer information by grabbing the
4814 * worker lock. If the task is not known to LU,
4815 * completed/aborted, then see if we need to
4816 * free this task.
4817 */
4818 if (old & ITASK_KNOWN_TO_LU) {
4819 free_it = 0;
4820 queue_it = 1;
4821 if (old & ITASK_IN_WORKER_QUEUE) {
4822 cmn_err(CE_PANIC, "status completion received"
4823 " when task is already in worker queue "
4824 " task = %p", (void *)task);
4825 }
4826 new |= ITASK_IN_WORKER_QUEUE;
4827 } else {
4828 queue_it = 0;
4829 }
4830 } while (atomic_cas_32(&itask->itask_flags, old, new) != old);
4831 task->task_completion_status = s;
4832
4833
4834 if (queue_it) {
4835 ASSERT(itask->itask_ncmds < ITASK_MAX_NCMDS);
4836 itask->itask_cmd_stack[itask->itask_ncmds++] =
4837 ITASK_CMD_STATUS_DONE;
4838 itask->itask_worker_next = NULL;
4839 if (w->worker_task_tail) {
4840 w->worker_task_tail->itask_worker_next = itask;
4841 } else {
4842 w->worker_task_head = itask;
4843 }
4844 w->worker_task_tail = itask;
4845 /* Measure task waitq time */
4846 itask->itask_waitq_enter_timestamp = gethrtime();
4847 if (++(w->worker_queue_depth) > w->worker_max_qdepth_pu) {
4848 w->worker_max_qdepth_pu = w->worker_queue_depth;
4849 }
4850 if ((w->worker_flags & STMF_WORKER_ACTIVE) == 0)
4851 cv_signal(&w->worker_cv);
4852 }
4853 mutex_exit(&w->worker_lock);
4854
4855 if (free_it) {
4856 if ((itask->itask_flags & (ITASK_KNOWN_TO_LU |
4857 ITASK_KNOWN_TO_TGT_PORT | ITASK_IN_WORKER_QUEUE |
4858 ITASK_BEING_ABORTED)) == 0) {
4859 stmf_task_free(task);
4860 } else {
4861 cmn_err(CE_PANIC, "LU is done with the task but LPORT "
4862 " is not done, itask %p itask_flags %x",
4863 (void *)itask, itask->itask_flags);
4864 }
4865 }
4866 }
4867
4868 void
4869 stmf_task_lu_done(scsi_task_t *task)
4870 {
4871 stmf_i_scsi_task_t *itask =
4872 (stmf_i_scsi_task_t *)task->task_stmf_private;
4873 stmf_worker_t *w = itask->itask_worker;
4874 uint32_t new, old;
4875
4876 mutex_enter(&w->worker_lock);
4877 do {
4878 new = old = itask->itask_flags;
4879 if (old & ITASK_BEING_ABORTED) {
4880 mutex_exit(&w->worker_lock);
4881 return;
4882 }
4883 if (old & ITASK_IN_WORKER_QUEUE) {
4884 cmn_err(CE_PANIC, "task_lu_done received"
4885 " when task is in worker queue "
4886 " task = %p", (void *)task);
4887 }
4888 new &= ~ITASK_KNOWN_TO_LU;
4889 } while (atomic_cas_32(&itask->itask_flags, old, new) != old);
4890
4891 mutex_exit(&w->worker_lock);
4892
4893 if ((itask->itask_flags & (ITASK_KNOWN_TO_LU |
4894 ITASK_KNOWN_TO_TGT_PORT | ITASK_IN_WORKER_QUEUE |
4895 ITASK_BEING_ABORTED)) == 0) {
4896 stmf_task_free(task);
4897 } else {
4898 cmn_err(CE_PANIC, "stmf_lu_done should be the last stage but "
4899 " the task is still not done, task = %p", (void *)task);
4900 }
4901 }
4902
4903 void
4904 stmf_queue_task_for_abort(scsi_task_t *task, stmf_status_t s)
4905 {
4906 stmf_i_scsi_task_t *itask =
4907 (stmf_i_scsi_task_t *)task->task_stmf_private;
4908 stmf_worker_t *w;
4909 uint32_t old, new;
4910
4911 stmf_task_audit(itask, TE_TASK_ABORT, CMD_OR_IOF_NA, NULL);
4912
4913 do {
4914 old = new = itask->itask_flags;
4915 if ((old & ITASK_BEING_ABORTED) ||
4916 ((old & (ITASK_KNOWN_TO_TGT_PORT |
4917 ITASK_KNOWN_TO_LU)) == 0)) {
4918 return;
4919 }
4920 new |= ITASK_BEING_ABORTED;
4921 } while (atomic_cas_32(&itask->itask_flags, old, new) != old);
4922 task->task_completion_status = s;
4923 itask->itask_start_time = ddi_get_lbolt();
4924
4925 if (((w = itask->itask_worker) == NULL) ||
4926 (itask->itask_flags & ITASK_IN_TRANSITION)) {
4927 return;
4928 }
4929
4930 /* Queue it and get out */
4931 mutex_enter(&w->worker_lock);
4932 if (itask->itask_flags & ITASK_IN_WORKER_QUEUE) {
4933 mutex_exit(&w->worker_lock);
4934 return;
4935 }
4936 atomic_or_32(&itask->itask_flags, ITASK_IN_WORKER_QUEUE);
4937 itask->itask_worker_next = NULL;
4938 if (w->worker_task_tail) {
4939 w->worker_task_tail->itask_worker_next = itask;
4940 } else {
4941 w->worker_task_head = itask;
4942 }
4943 w->worker_task_tail = itask;
4944 if (++(w->worker_queue_depth) > w->worker_max_qdepth_pu) {
4945 w->worker_max_qdepth_pu = w->worker_queue_depth;
4946 }
4947 if ((w->worker_flags & STMF_WORKER_ACTIVE) == 0)
4948 cv_signal(&w->worker_cv);
4949 mutex_exit(&w->worker_lock);
4950 }
4951
4952 void
4953 stmf_abort(int abort_cmd, scsi_task_t *task, stmf_status_t s, void *arg)
4954 {
4955 stmf_i_scsi_task_t *itask = NULL;
4956 uint32_t old, new, f, rf;
4957
4958 DTRACE_PROBE2(scsi__task__abort, scsi_task_t *, task,
4959 stmf_status_t, s);
4960
4961 switch (abort_cmd) {
4962 case STMF_QUEUE_ABORT_LU:
4963 stmf_task_lu_killall((stmf_lu_t *)arg, task, s);
4964 return;
4965 case STMF_QUEUE_TASK_ABORT:
4966 stmf_queue_task_for_abort(task, s);
4967 return;
4968 case STMF_REQUEUE_TASK_ABORT_LPORT:
4969 rf = ITASK_TGT_PORT_ABORT_CALLED;
4970 f = ITASK_KNOWN_TO_TGT_PORT;
4971 break;
4972 case STMF_REQUEUE_TASK_ABORT_LU:
4973 rf = ITASK_LU_ABORT_CALLED;
4974 f = ITASK_KNOWN_TO_LU;
4975 break;
4976 default:
4977 return;
4978 }
4979 itask = (stmf_i_scsi_task_t *)task->task_stmf_private;
4980 f |= ITASK_BEING_ABORTED | rf;
4981 do {
4982 old = new = itask->itask_flags;
4983 if ((old & f) != f) {
4984 return;
4985 }
4986 new &= ~rf;
4987 } while (atomic_cas_32(&itask->itask_flags, old, new) != old);
4988 }
4989
4990 void
4991 stmf_task_lu_aborted(scsi_task_t *task, stmf_status_t s, uint32_t iof)
4992 {
4993 char info[STMF_CHANGE_INFO_LEN];
4994 stmf_i_scsi_task_t *itask = TASK_TO_ITASK(task);
4995 unsigned long long st;
4996
4997 stmf_task_audit(itask, TE_TASK_LU_ABORTED, iof, NULL);
4998
4999 st = s; /* gcc fix */
5000 if ((s != STMF_ABORT_SUCCESS) && (s != STMF_NOT_FOUND)) {
5001 (void) snprintf(info, sizeof (info),
5002 "task %p, lu failed to abort ret=%llx", (void *)task, st);
5003 } else if ((iof & STMF_IOF_LU_DONE) == 0) {
5004 (void) snprintf(info, sizeof (info),
5005 "Task aborted but LU is not finished, task ="
5006 "%p, s=%llx, iof=%x", (void *)task, st, iof);
5007 } else {
5008 /*
5009 * LU abort successfully
5010 */
5011 atomic_and_32(&itask->itask_flags, ~ITASK_KNOWN_TO_LU);
5012 return;
5013 }
5014
5015 stmf_abort_task_offline(task, 1, info);
5016 }
5017
5018 void
5019 stmf_task_lport_aborted(scsi_task_t *task, stmf_status_t s, uint32_t iof)
5020 {
5021 char info[STMF_CHANGE_INFO_LEN];
5022 stmf_i_scsi_task_t *itask = TASK_TO_ITASK(task);
5023 unsigned long long st;
5024 uint32_t old, new;
5025
5026 stmf_task_audit(itask, TE_TASK_LPORT_ABORTED, iof, NULL);
5027
5028 st = s;
5029 if ((s != STMF_ABORT_SUCCESS) && (s != STMF_NOT_FOUND)) {
5030 (void) snprintf(info, sizeof (info),
5031 "task %p, tgt port failed to abort ret=%llx", (void *)task,
5032 st);
5033 } else if ((iof & STMF_IOF_LPORT_DONE) == 0) {
5034 (void) snprintf(info, sizeof (info),
5035 "Task aborted but tgt port is not finished, "
5036 "task=%p, s=%llx, iof=%x", (void *)task, st, iof);
5037 } else {
5038 /*
5039 * LPORT abort successfully
5040 */
5041 do {
5042 old = new = itask->itask_flags;
5043 if (!(old & ITASK_KNOWN_TO_TGT_PORT))
5044 return;
5045 new &= ~ITASK_KNOWN_TO_TGT_PORT;
5046 } while (atomic_cas_32(&itask->itask_flags, old, new) != old);
5047 return;
5048 }
5049
5050 stmf_abort_task_offline(task, 0, info);
5051 }
5052
5053 stmf_status_t
5054 stmf_task_poll_lu(scsi_task_t *task, uint32_t timeout)
5055 {
5056 stmf_i_scsi_task_t *itask = (stmf_i_scsi_task_t *)
5057 task->task_stmf_private;
5058 stmf_worker_t *w = itask->itask_worker;
5059 int i;
5060
5061 ASSERT(itask->itask_flags & ITASK_KNOWN_TO_LU);
5062 mutex_enter(&w->worker_lock);
5063 if (itask->itask_ncmds >= ITASK_MAX_NCMDS) {
5064 mutex_exit(&w->worker_lock);
5065 return (STMF_BUSY);
5066 }
5067 for (i = 0; i < itask->itask_ncmds; i++) {
5068 if (itask->itask_cmd_stack[i] == ITASK_CMD_POLL_LU) {
5069 mutex_exit(&w->worker_lock);
5070 return (STMF_SUCCESS);
5071 }
5072 }
5073 itask->itask_cmd_stack[itask->itask_ncmds++] = ITASK_CMD_POLL_LU;
5074 if (timeout == ITASK_DEFAULT_POLL_TIMEOUT) {
5075 itask->itask_poll_timeout = ddi_get_lbolt() + 1;
5076 } else {
5077 clock_t t = drv_usectohz(timeout * 1000);
5078 if (t == 0)
5079 t = 1;
5080 itask->itask_poll_timeout = ddi_get_lbolt() + t;
5081 }
5082 if ((itask->itask_flags & ITASK_IN_WORKER_QUEUE) == 0) {
5083 itask->itask_worker_next = NULL;
5084 if (w->worker_task_tail) {
5085 w->worker_task_tail->itask_worker_next = itask;
5086 } else {
5087 w->worker_task_head = itask;
5088 }
5089 w->worker_task_tail = itask;
5090 if (++(w->worker_queue_depth) > w->worker_max_qdepth_pu) {
5091 w->worker_max_qdepth_pu = w->worker_queue_depth;
5092 }
5093 atomic_or_32(&itask->itask_flags, ITASK_IN_WORKER_QUEUE);
5094 if ((w->worker_flags & STMF_WORKER_ACTIVE) == 0)
5095 cv_signal(&w->worker_cv);
5096 }
5097 mutex_exit(&w->worker_lock);
5098 return (STMF_SUCCESS);
5099 }
5100
5101 stmf_status_t
5102 stmf_task_poll_lport(scsi_task_t *task, uint32_t timeout)
5103 {
5104 stmf_i_scsi_task_t *itask = (stmf_i_scsi_task_t *)
5105 task->task_stmf_private;
5106 stmf_worker_t *w = itask->itask_worker;
5107 int i;
5108
5109 ASSERT(itask->itask_flags & ITASK_KNOWN_TO_TGT_PORT);
5110 mutex_enter(&w->worker_lock);
5111 if (itask->itask_ncmds >= ITASK_MAX_NCMDS) {
5112 mutex_exit(&w->worker_lock);
5113 return (STMF_BUSY);
5114 }
5115 for (i = 0; i < itask->itask_ncmds; i++) {
5116 if (itask->itask_cmd_stack[i] == ITASK_CMD_POLL_LPORT) {
5117 mutex_exit(&w->worker_lock);
5118 return (STMF_SUCCESS);
5119 }
5120 }
5121 itask->itask_cmd_stack[itask->itask_ncmds++] = ITASK_CMD_POLL_LPORT;
5122 if (timeout == ITASK_DEFAULT_POLL_TIMEOUT) {
5123 itask->itask_poll_timeout = ddi_get_lbolt() + 1;
5124 } else {
5125 clock_t t = drv_usectohz(timeout * 1000);
5126 if (t == 0)
5127 t = 1;
5128 itask->itask_poll_timeout = ddi_get_lbolt() + t;
5129 }
5130 if ((itask->itask_flags & ITASK_IN_WORKER_QUEUE) == 0) {
5131 itask->itask_worker_next = NULL;
5132 if (w->worker_task_tail) {
5133 w->worker_task_tail->itask_worker_next = itask;
5134 } else {
5135 w->worker_task_head = itask;
5136 }
5137 w->worker_task_tail = itask;
5138 if (++(w->worker_queue_depth) > w->worker_max_qdepth_pu) {
5139 w->worker_max_qdepth_pu = w->worker_queue_depth;
5140 }
5141 if ((w->worker_flags & STMF_WORKER_ACTIVE) == 0)
5142 cv_signal(&w->worker_cv);
5143 }
5144 mutex_exit(&w->worker_lock);
5145 return (STMF_SUCCESS);
5146 }
5147
5148 void
5149 stmf_do_task_abort(scsi_task_t *task)
5150 {
5151 stmf_i_scsi_task_t *itask = TASK_TO_ITASK(task);
5152 stmf_lu_t *lu;
5153 stmf_local_port_t *lport;
5154 unsigned long long ret;
5155 uint32_t old, new;
5156 uint8_t call_lu_abort, call_port_abort;
5157 char info[STMF_CHANGE_INFO_LEN];
5158
5159 lu = task->task_lu;
5160 lport = task->task_lport;
5161 do {
5162 old = new = itask->itask_flags;
5163 if ((old & (ITASK_KNOWN_TO_LU | ITASK_LU_ABORT_CALLED)) ==
5164 ITASK_KNOWN_TO_LU) {
5165 new |= ITASK_LU_ABORT_CALLED;
5166 call_lu_abort = 1;
5167 } else {
5168 call_lu_abort = 0;
5169 }
5170 } while (atomic_cas_32(&itask->itask_flags, old, new) != old);
5171
5172 if (call_lu_abort) {
5173 if ((itask->itask_flags & ITASK_DEFAULT_HANDLING) == 0) {
5174 ret = lu->lu_abort(lu, STMF_LU_ABORT_TASK, task, 0);
5175 } else {
5176 ret = dlun0->lu_abort(lu, STMF_LU_ABORT_TASK, task, 0);
5177 }
5178 if ((ret == STMF_ABORT_SUCCESS) || (ret == STMF_NOT_FOUND)) {
5179 stmf_task_lu_aborted(task, ret, STMF_IOF_LU_DONE);
5180 } else if (ret == STMF_BUSY) {
5181 atomic_and_32(&itask->itask_flags,
5182 ~ITASK_LU_ABORT_CALLED);
5183 } else if (ret != STMF_SUCCESS) {
5184 (void) snprintf(info, sizeof (info),
5185 "Abort failed by LU %p, ret %llx", (void *)lu, ret);
5186 stmf_abort_task_offline(task, 1, info);
5187 }
5188 } else if (itask->itask_flags & ITASK_KNOWN_TO_LU) {
5189 if (ddi_get_lbolt() > (itask->itask_start_time +
5190 STMF_SEC2TICK(lu->lu_abort_timeout?
5191 lu->lu_abort_timeout : ITASK_DEFAULT_ABORT_TIMEOUT))) {
5192 (void) snprintf(info, sizeof (info),
5193 "lu abort timed out");
5194 stmf_abort_task_offline(itask->itask_task, 1, info);
5195 }
5196 }
5197
5198 do {
5199 old = new = itask->itask_flags;
5200 if ((old & (ITASK_KNOWN_TO_TGT_PORT |
5201 ITASK_TGT_PORT_ABORT_CALLED)) == ITASK_KNOWN_TO_TGT_PORT) {
5202 new |= ITASK_TGT_PORT_ABORT_CALLED;
5203 call_port_abort = 1;
5204 } else {
5205 call_port_abort = 0;
5206 }
5207 } while (atomic_cas_32(&itask->itask_flags, old, new) != old);
5208 if (call_port_abort) {
5209 ret = lport->lport_abort(lport, STMF_LPORT_ABORT_TASK, task, 0);
5210 if ((ret == STMF_ABORT_SUCCESS) || (ret == STMF_NOT_FOUND)) {
5211 stmf_task_lport_aborted(task, ret, STMF_IOF_LPORT_DONE);
5212 } else if (ret == STMF_BUSY) {
5213 atomic_and_32(&itask->itask_flags,
5214 ~ITASK_TGT_PORT_ABORT_CALLED);
5215 } else if (ret != STMF_SUCCESS) {
5216 (void) snprintf(info, sizeof (info),
5217 "Abort failed by tgt port %p ret %llx",
5218 (void *)lport, ret);
5219 stmf_abort_task_offline(task, 0, info);
5220 }
5221 } else if (itask->itask_flags & ITASK_KNOWN_TO_TGT_PORT) {
5222 if (ddi_get_lbolt() > (itask->itask_start_time +
5223 STMF_SEC2TICK(lport->lport_abort_timeout?
5224 lport->lport_abort_timeout :
5225 ITASK_DEFAULT_ABORT_TIMEOUT))) {
5226 (void) snprintf(info, sizeof (info),
5227 "lport abort timed out");
5228 stmf_abort_task_offline(itask->itask_task, 0, info);
5229 }
5230 }
5231 }
5232
5233 stmf_status_t
5234 stmf_ctl(int cmd, void *obj, void *arg)
5235 {
5236 stmf_status_t ret;
5237 stmf_i_lu_t *ilu;
5238 stmf_i_local_port_t *ilport;
5239 stmf_state_change_info_t *ssci = (stmf_state_change_info_t *)arg;
5240
5241 mutex_enter(&stmf_state.stmf_lock);
5242 ret = STMF_INVALID_ARG;
5243 if (cmd & STMF_CMD_LU_OP) {
5244 ilu = stmf_lookup_lu((stmf_lu_t *)obj);
5245 if (ilu == NULL) {
5246 goto stmf_ctl_lock_exit;
5247 }
5248 DTRACE_PROBE3(lu__state__change,
5249 stmf_lu_t *, ilu->ilu_lu,
5250 int, cmd, stmf_state_change_info_t *, ssci);
5542 sdid->association = ID_IS_TARGET_PORT;
5543 sdid->ident_length = 20;
5544 /* Convert wwn value to "wwn.XXXXXXXXXXXXXXXX" format */
5545 (void) snprintf(wwn_str, sizeof (wwn_str),
5546 "wwn.%02X%02X%02X%02X%02X%02X%02X%02X",
5547 wwn[0], wwn[1], wwn[2], wwn[3], wwn[4], wwn[5], wwn[6], wwn[7]);
5548 bcopy(wwn_str, (char *)sdid->ident, 20);
5549 }
5550
5551
5552 stmf_xfer_data_t *
5553 stmf_prepare_tpgs_data(uint8_t ilu_alua)
5554 {
5555 stmf_xfer_data_t *xd;
5556 stmf_i_local_port_t *ilport;
5557 uint8_t *p;
5558 uint32_t sz, asz, nports = 0, nports_standby = 0;
5559
5560 mutex_enter(&stmf_state.stmf_lock);
5561 /* check if any ports are standby and create second group */
5562 for (ilport = stmf_state.stmf_ilportlist; ilport;
5563 ilport = ilport->ilport_next) {
5564 if (ilport->ilport_standby == 1) {
5565 nports_standby++;
5566 } else {
5567 nports++;
5568 }
5569 }
5570
5571 /* The spec only allows for 255 ports to be reported per group */
5572 nports = min(nports, 255);
5573 nports_standby = min(nports_standby, 255);
5574 sz = (nports * 4) + 12;
5575 if (nports_standby && ilu_alua) {
5576 sz += (nports_standby * 4) + 8;
5577 }
5578 asz = sz + sizeof (*xd) - 4;
5579 xd = (stmf_xfer_data_t *)kmem_zalloc(asz, KM_NOSLEEP);
5580 if (xd == NULL) {
5581 mutex_exit(&stmf_state.stmf_lock);
5582 return (NULL);
5583 }
5584 xd->alloc_size = asz;
5585 xd->size_left = sz;
5586
5587 p = xd->buf;
5588
5589 *((uint32_t *)p) = BE_32(sz - 4);
5590 p += 4;
5591 p[0] = 0x80; /* PREF */
5592 p[1] = 5; /* AO_SUP, S_SUP */
5593 if (stmf_state.stmf_alua_node == 1) {
5594 p[3] = 1; /* Group 1 */
5595 } else {
5596 p[3] = 0; /* Group 0 */
5597 }
5598 p[7] = nports & 0xff;
5599 p += 8;
5600 for (ilport = stmf_state.stmf_ilportlist; ilport;
5601 ilport = ilport->ilport_next) {
5602 if (ilport->ilport_standby == 1) {
5603 continue;
5604 }
5605 ((uint16_t *)p)[1] = BE_16(ilport->ilport_rtpid);
5606 p += 4;
5607 }
5608 if (nports_standby && ilu_alua) {
5609 p[0] = 0x02; /* Non PREF, Standby */
5610 p[1] = 5; /* AO_SUP, S_SUP */
5611 if (stmf_state.stmf_alua_node == 1) {
5612 p[3] = 0; /* Group 0 */
5613 } else {
5614 p[3] = 1; /* Group 1 */
5615 }
5616 p[7] = nports_standby & 0xff;
5617 p += 8;
5618 for (ilport = stmf_state.stmf_ilportlist; ilport;
5619 ilport = ilport->ilport_next) {
5620 if (ilport->ilport_standby == 0) {
5621 continue;
5622 }
5623 ((uint16_t *)p)[1] = BE_16(ilport->ilport_rtpid);
5624 p += 4;
5625 }
5626 }
5627
5628 mutex_exit(&stmf_state.stmf_lock);
5629
5630 return (xd);
5631 }
5632
5633 struct scsi_devid_desc *
5634 stmf_scsilib_get_devid_desc(uint16_t rtpid)
5635 {
5636 scsi_devid_desc_t *devid = NULL;
5637 stmf_i_local_port_t *ilport;
5638
5639 mutex_enter(&stmf_state.stmf_lock);
5640
5641 for (ilport = stmf_state.stmf_ilportlist; ilport;
5642 ilport = ilport->ilport_next) {
5643 if (ilport->ilport_rtpid == rtpid) {
5644 scsi_devid_desc_t *id = ilport->ilport_lport->lport_id;
5845 break;
5846 }
5847 }
5848
5849 page[2] = (m >> 8) & 0xff;
5850 page[3] = m & 0xff;
5851
5852 return (n);
5853 }
5854
5855 void
5856 stmf_scsilib_handle_report_tpgs(scsi_task_t *task, stmf_data_buf_t *dbuf)
5857 {
5858 stmf_i_scsi_task_t *itask =
5859 (stmf_i_scsi_task_t *)task->task_stmf_private;
5860 stmf_i_lu_t *ilu =
5861 (stmf_i_lu_t *)task->task_lu->lu_stmf_private;
5862 stmf_xfer_data_t *xd;
5863 uint32_t sz, minsz;
5864
5865 itask->itask_flags |= ITASK_DEFAULT_HANDLING;
5866 task->task_cmd_xfer_length =
5867 ((((uint32_t)task->task_cdb[6]) << 24) |
5868 (((uint32_t)task->task_cdb[7]) << 16) |
5869 (((uint32_t)task->task_cdb[8]) << 8) |
5870 ((uint32_t)task->task_cdb[9]));
5871
5872 if (task->task_additional_flags &
5873 TASK_AF_NO_EXPECTED_XFER_LENGTH) {
5874 task->task_expected_xfer_length =
5875 task->task_cmd_xfer_length;
5876 }
5877
5878 if (task->task_cmd_xfer_length == 0) {
5879 stmf_scsilib_send_status(task, STATUS_GOOD, 0);
5880 return;
5881 }
5882 if (task->task_cmd_xfer_length < 4) {
5883 stmf_scsilib_send_status(task, STATUS_CHECK,
5884 STMF_SAA_INVALID_FIELD_IN_CDB);
5885 return;
5886 }
5887
5888 sz = min(task->task_expected_xfer_length,
5889 task->task_cmd_xfer_length);
5890
5891 xd = stmf_prepare_tpgs_data(ilu->ilu_alua);
5892
5893 if (xd == NULL) {
5894 stmf_abort(STMF_QUEUE_TASK_ABORT, task,
5895 STMF_ALLOC_FAILURE, NULL);
5896 return;
5962 /*
5963 * To sync with target reset, grab this lock. The LU is not going
5964 * anywhere as there is atleast one task pending (this task).
5965 */
5966 mutex_enter(&stmf_state.stmf_lock);
5967
5968 if (ilu->ilu_flags & ILU_RESET_ACTIVE) {
5969 mutex_exit(&stmf_state.stmf_lock);
5970 stmf_scsilib_send_status(task, STATUS_CHECK,
5971 STMF_SAA_OPERATION_IN_PROGRESS);
5972 return;
5973 }
5974 atomic_or_32(&ilu->ilu_flags, ILU_RESET_ACTIVE);
5975 mutex_exit(&stmf_state.stmf_lock);
5976
5977 /*
5978 * Mark this task as the one causing LU reset so that we know who
5979 * was responsible for setting the ILU_RESET_ACTIVE. In case this
5980 * task itself gets aborted, we will clear ILU_RESET_ACTIVE.
5981 */
5982 itask->itask_flags |= ITASK_DEFAULT_HANDLING | ITASK_CAUSING_LU_RESET;
5983
5984 /* Initiatiate abort on all commands on this LU except this one */
5985 stmf_abort(STMF_QUEUE_ABORT_LU, task, STMF_ABORTED, task->task_lu);
5986
5987 /* Start polling on this task */
5988 if (stmf_task_poll_lu(task, ITASK_DEFAULT_POLL_TIMEOUT)
5989 != STMF_SUCCESS) {
5990 stmf_abort(STMF_QUEUE_TASK_ABORT, task, STMF_ALLOC_FAILURE,
5991 NULL);
5992 return;
5993 }
5994 }
5995
5996 void
5997 stmf_handle_target_reset(scsi_task_t *task)
5998 {
5999 stmf_i_scsi_task_t *itask;
6000 stmf_i_lu_t *ilu;
6001 stmf_i_scsi_session_t *iss;
6002 stmf_lun_map_t *lm;
6038 ilu = (stmf_i_lu_t *)(lm_ent->ent_lu->lu_stmf_private);
6039 if (ilu->ilu_flags & ILU_RESET_ACTIVE) {
6040 atomic_and_32(&iss->iss_flags, ~ISS_RESET_ACTIVE);
6041 rw_exit(iss->iss_lockp);
6042 mutex_exit(&stmf_state.stmf_lock);
6043 stmf_scsilib_send_status(task, STATUS_CHECK,
6044 STMF_SAA_OPERATION_IN_PROGRESS);
6045 return;
6046 }
6047 }
6048 if (lf == 0) {
6049 /* No luns in this session */
6050 atomic_and_32(&iss->iss_flags, ~ISS_RESET_ACTIVE);
6051 rw_exit(iss->iss_lockp);
6052 mutex_exit(&stmf_state.stmf_lock);
6053 stmf_scsilib_send_status(task, STATUS_GOOD, 0);
6054 return;
6055 }
6056
6057 /* ok, start the damage */
6058 itask->itask_flags |= ITASK_DEFAULT_HANDLING |
6059 ITASK_CAUSING_TARGET_RESET;
6060 for (i = 0; i < lm->lm_nentries; i++) {
6061 if (lm->lm_plus[i] == NULL)
6062 continue;
6063 lm_ent = (stmf_lun_map_ent_t *)lm->lm_plus[i];
6064 ilu = (stmf_i_lu_t *)(lm_ent->ent_lu->lu_stmf_private);
6065 atomic_or_32(&ilu->ilu_flags, ILU_RESET_ACTIVE);
6066 }
6067
6068 for (i = 0; i < lm->lm_nentries; i++) {
6069 if (lm->lm_plus[i] == NULL)
6070 continue;
6071 lm_ent = (stmf_lun_map_ent_t *)lm->lm_plus[i];
6072 stmf_abort(STMF_QUEUE_ABORT_LU, task, STMF_ABORTED,
6073 lm_ent->ent_lu);
6074 }
6075
6076 rw_exit(iss->iss_lockp);
6077 mutex_exit(&stmf_state.stmf_lock);
6078
6079 /* Start polling on this task */
6097 (task->task_cdb[0] == SCMD_INQUIRY)) {
6098 rw_exit(iss->iss_lockp);
6099 return (0);
6100 }
6101 atomic_and_32(&iss->iss_flags,
6102 ~(ISS_LUN_INVENTORY_CHANGED | ISS_GOT_INITIAL_LUNS));
6103 rw_exit(iss->iss_lockp);
6104
6105 if (task->task_cdb[0] == SCMD_REPORT_LUNS) {
6106 return (0);
6107 }
6108 stmf_scsilib_send_status(task, STATUS_CHECK,
6109 STMF_SAA_REPORT_LUN_DATA_HAS_CHANGED);
6110 return (1);
6111 }
6112
6113 void
6114 stmf_worker_init()
6115 {
6116 uint32_t i;
6117
6118 /* Make local copy of global tunables */
6119 stmf_i_max_nworkers = stmf_max_nworkers;
6120 stmf_i_min_nworkers = stmf_min_nworkers;
6121
6122 ASSERT(stmf_workers == NULL);
6123 if (stmf_i_min_nworkers < 4) {
6124 stmf_i_min_nworkers = 4;
6125 }
6126 if (stmf_i_max_nworkers < stmf_i_min_nworkers) {
6127 stmf_i_max_nworkers = stmf_i_min_nworkers;
6128 }
6129 stmf_workers = (stmf_worker_t *)kmem_zalloc(
6130 sizeof (stmf_worker_t) * stmf_i_max_nworkers, KM_SLEEP);
6131 for (i = 0; i < stmf_i_max_nworkers; i++) {
6132 stmf_worker_t *w = &stmf_workers[i];
6133 mutex_init(&w->worker_lock, NULL, MUTEX_DRIVER, NULL);
6134 cv_init(&w->worker_cv, NULL, CV_DRIVER, NULL);
6135 }
6136 stmf_worker_mgmt_delay = drv_usectohz(20 * 1000);
6137 stmf_workers_state = STMF_WORKERS_ENABLED;
6138
6139 /* Workers will be started by stmf_worker_mgmt() */
6140
6141 /* Lets wait for atleast one worker to start */
6142 while (stmf_nworkers_cur == 0)
6143 delay(drv_usectohz(20 * 1000));
6144 stmf_worker_mgmt_delay = drv_usectohz(3 * 1000 * 1000);
6145 }
6146
6147 stmf_status_t
6148 stmf_worker_fini()
6149 {
6150 int i;
6151 clock_t sb;
6152
6153 if (stmf_workers_state == STMF_WORKERS_DISABLED)
6154 return (STMF_SUCCESS);
6155 ASSERT(stmf_workers);
6156 stmf_workers_state = STMF_WORKERS_DISABLED;
6157 stmf_worker_mgmt_delay = drv_usectohz(20 * 1000);
6158 cv_signal(&stmf_state.stmf_cv);
6159
6160 sb = ddi_get_lbolt() + drv_usectohz(10 * 1000 * 1000);
6161 /* Wait for all the threads to die */
6162 while (stmf_nworkers_cur != 0) {
6163 if (ddi_get_lbolt() > sb) {
6164 stmf_workers_state = STMF_WORKERS_ENABLED;
6165 return (STMF_BUSY);
6166 }
6167 delay(drv_usectohz(100 * 1000));
6168 }
6169 for (i = 0; i < stmf_i_max_nworkers; i++) {
6170 stmf_worker_t *w = &stmf_workers[i];
6171 mutex_destroy(&w->worker_lock);
6172 cv_destroy(&w->worker_cv);
6173 }
6174 kmem_free(stmf_workers, sizeof (stmf_worker_t) * stmf_i_max_nworkers);
6175 stmf_workers = NULL;
6176
6177 return (STMF_SUCCESS);
6178 }
6179
6180 void
6181 stmf_worker_task(void *arg)
6182 {
6183 stmf_worker_t *w;
6184 stmf_i_scsi_session_t *iss;
6185 scsi_task_t *task;
6186 stmf_i_scsi_task_t *itask;
6187 stmf_data_buf_t *dbuf;
6188 stmf_lu_t *lu;
6189 clock_t wait_timer = 0;
6190 clock_t wait_ticks, wait_delta = 0;
6191 uint32_t old, new;
6192 uint8_t curcmd;
6193 uint8_t abort_free;
6194 uint8_t wait_queue;
6195 uint8_t dec_qdepth;
6196
6197 w = (stmf_worker_t *)arg;
6198 wait_ticks = drv_usectohz(10000);
6199
6200 DTRACE_PROBE1(worker__create, stmf_worker_t, w);
6201 mutex_enter(&w->worker_lock);
6202 w->worker_flags |= STMF_WORKER_STARTED | STMF_WORKER_ACTIVE;
6203 stmf_worker_loop:;
6204 if ((w->worker_ref_count == 0) &&
6205 (w->worker_flags & STMF_WORKER_TERMINATE)) {
6206 w->worker_flags &= ~(STMF_WORKER_STARTED |
6207 STMF_WORKER_ACTIVE | STMF_WORKER_TERMINATE);
6208 w->worker_tid = NULL;
6209 mutex_exit(&w->worker_lock);
6210 DTRACE_PROBE1(worker__destroy, stmf_worker_t, w);
6211 thread_exit();
6212 }
6213 /* CONSTCOND */
6214 while (1) {
6215 dec_qdepth = 0;
6216 if (wait_timer && (ddi_get_lbolt() >= wait_timer)) {
6217 wait_timer = 0;
6218 wait_delta = 0;
6219 if (w->worker_wait_head) {
6220 ASSERT(w->worker_wait_tail);
6221 if (w->worker_task_head == NULL)
6222 w->worker_task_head =
6223 w->worker_wait_head;
6224 else
6225 w->worker_task_tail->itask_worker_next =
6226 w->worker_wait_head;
6227 w->worker_task_tail = w->worker_wait_tail;
6228 w->worker_wait_head = w->worker_wait_tail =
6229 NULL;
6230 }
6231 }
6232 if ((itask = w->worker_task_head) == NULL) {
6233 break;
6234 }
6235 task = itask->itask_task;
6236 DTRACE_PROBE2(worker__active, stmf_worker_t, w,
6237 scsi_task_t *, task);
6238 w->worker_task_head = itask->itask_worker_next;
6239 if (w->worker_task_head == NULL)
6240 w->worker_task_tail = NULL;
6241
6242 wait_queue = 0;
6243 abort_free = 0;
6244 if (itask->itask_ncmds > 0) {
6245 curcmd = itask->itask_cmd_stack[itask->itask_ncmds - 1];
6246 } else {
6247 ASSERT(itask->itask_flags & ITASK_BEING_ABORTED);
6248 }
6249 do {
6250 old = itask->itask_flags;
6251 if (old & ITASK_BEING_ABORTED) {
6252 itask->itask_ncmds = 1;
6253 curcmd = itask->itask_cmd_stack[0] =
6254 ITASK_CMD_ABORT;
6255 goto out_itask_flag_loop;
6256 } else if ((curcmd & ITASK_CMD_MASK) ==
6257 ITASK_CMD_NEW_TASK) {
6258 /*
6259 * set ITASK_KSTAT_IN_RUNQ, this flag
6260 * will not reset until task completed
6261 */
6262 new = old | ITASK_KNOWN_TO_LU |
6263 ITASK_KSTAT_IN_RUNQ;
6264 } else {
6265 goto out_itask_flag_loop;
6266 }
6267 } while (atomic_cas_32(&itask->itask_flags, old, new) != old);
6268
6269 out_itask_flag_loop:
6270
6271 /*
6272 * Decide if this task needs to go to a queue and/or if
6273 * we can decrement the itask_cmd_stack.
6274 */
6275 if (curcmd == ITASK_CMD_ABORT) {
6276 if (itask->itask_flags & (ITASK_KNOWN_TO_LU |
6277 ITASK_KNOWN_TO_TGT_PORT)) {
6278 wait_queue = 1;
6279 } else {
6280 abort_free = 1;
6281 }
6282 } else if ((curcmd & ITASK_CMD_POLL) &&
6283 (itask->itask_poll_timeout > ddi_get_lbolt())) {
6284 wait_queue = 1;
6285 }
6286
6287 if (wait_queue) {
6306 w->worker_task_tail = itask;
6307 } else {
6308 atomic_and_32(&itask->itask_flags,
6309 ~ITASK_IN_WORKER_QUEUE);
6310 /*
6311 * This is where the queue depth should go down by
6312 * one but we delay that on purpose to account for
6313 * the call into the provider. The actual decrement
6314 * happens after the worker has done its job.
6315 */
6316 dec_qdepth = 1;
6317 itask->itask_waitq_time +=
6318 gethrtime() - itask->itask_waitq_enter_timestamp;
6319 }
6320
6321 /* We made it here means we are going to call LU */
6322 if ((itask->itask_flags & ITASK_DEFAULT_HANDLING) == 0)
6323 lu = task->task_lu;
6324 else
6325 lu = dlun0;
6326 dbuf = itask->itask_dbufs[ITASK_CMD_BUF_NDX(curcmd)];
6327 mutex_exit(&w->worker_lock);
6328 curcmd &= ITASK_CMD_MASK;
6329 stmf_task_audit(itask, TE_PROCESS_CMD, curcmd, dbuf);
6330 switch (curcmd) {
6331 case ITASK_CMD_NEW_TASK:
6332 iss = (stmf_i_scsi_session_t *)
6333 task->task_session->ss_stmf_private;
6334 stmf_itl_lu_new_task(itask);
6335 if (iss->iss_flags & ISS_LUN_INVENTORY_CHANGED) {
6336 if (stmf_handle_cmd_during_ic(itask))
6337 break;
6338 }
6339 #ifdef DEBUG
6340 if (stmf_drop_task_counter > 0) {
6341 if (atomic_dec_32_nv(&stmf_drop_task_counter)
6342 == 1)
6343 break;
6344 }
6345 #endif
6346 DTRACE_PROBE1(scsi__task__start, scsi_task_t *, task);
6347 lu->lu_new_task(task, dbuf);
6348 break;
6349 case ITASK_CMD_DATA_XFER_DONE:
6350 lu->lu_dbuf_xfer_done(task, dbuf);
6351 break;
6352 case ITASK_CMD_STATUS_DONE:
6353 lu->lu_send_status_done(task);
6354 break;
6355 case ITASK_CMD_ABORT:
6356 if (abort_free) {
6357 stmf_task_free(task);
6358 } else {
6359 stmf_do_task_abort(task);
6360 }
6361 break;
6362 case ITASK_CMD_POLL_LU:
6363 if (!wait_queue) {
6364 lu->lu_task_poll(task);
6365 }
6366 break;
6367 case ITASK_CMD_POLL_LPORT:
6368 if (!wait_queue)
6369 task->task_lport->lport_task_poll(task);
6370 break;
6371 case ITASK_CMD_SEND_STATUS:
6372 /* case ITASK_CMD_XFER_DATA: */
6373 break;
6374 }
6375 mutex_enter(&w->worker_lock);
6376 if (dec_qdepth) {
6377 w->worker_queue_depth--;
6378 }
6379 }
6380 if ((w->worker_flags & STMF_WORKER_TERMINATE) && (wait_timer == 0)) {
6381 if (w->worker_ref_count == 0)
6382 goto stmf_worker_loop;
6383 else {
6384 wait_timer = ddi_get_lbolt() + 1;
6385 wait_delta = 1;
6386 }
6387 }
6388 w->worker_flags &= ~STMF_WORKER_ACTIVE;
6389 if (wait_timer) {
6390 DTRACE_PROBE1(worker__timed__sleep, stmf_worker_t, w);
6391 (void) cv_reltimedwait(&w->worker_cv, &w->worker_lock,
6392 wait_delta, TR_CLOCK_TICK);
6393 } else {
6394 DTRACE_PROBE1(worker__sleep, stmf_worker_t, w);
6395 cv_wait(&w->worker_cv, &w->worker_lock);
6396 }
6397 DTRACE_PROBE1(worker__wakeup, stmf_worker_t, w);
6398 w->worker_flags |= STMF_WORKER_ACTIVE;
6399 goto stmf_worker_loop;
6400 }
6401
6402 void
6403 stmf_worker_mgmt()
6404 {
6405 int i;
6406 int workers_needed;
6407 uint32_t qd;
6408 clock_t tps, d = 0;
6409 uint32_t cur_max_ntasks = 0;
6410 stmf_worker_t *w;
6411
6412 /* Check if we are trying to increase the # of threads */
6413 for (i = stmf_nworkers_cur; i < stmf_nworkers_needed; i++) {
6414 if (stmf_workers[i].worker_flags & STMF_WORKER_STARTED) {
6415 stmf_nworkers_cur++;
6416 stmf_nworkers_accepting_cmds++;
6417 } else {
6418 /* Wait for transition to complete */
6419 return;
6420 }
6421 }
6422 /* Check if we are trying to decrease the # of workers */
6423 for (i = (stmf_nworkers_cur - 1); i >= stmf_nworkers_needed; i--) {
6424 if ((stmf_workers[i].worker_flags & STMF_WORKER_STARTED) == 0) {
6425 stmf_nworkers_cur--;
6426 /*
6427 * stmf_nworkers_accepting_cmds has already been
6428 * updated by the request to reduce the # of workers.
6429 */
6430 } else {
6431 /* Wait for transition to complete */
6432 return;
6433 }
6434 }
6435 /* Check if we are being asked to quit */
6436 if (stmf_workers_state != STMF_WORKERS_ENABLED) {
6437 if (stmf_nworkers_cur) {
6438 workers_needed = 0;
6439 goto worker_mgmt_trigger_change;
6440 }
6441 return;
6442 }
6443 /* Check if we are starting */
6444 if (stmf_nworkers_cur < stmf_i_min_nworkers) {
6445 workers_needed = stmf_i_min_nworkers;
6446 goto worker_mgmt_trigger_change;
6447 }
6448
6449 tps = drv_usectohz(1 * 1000 * 1000);
6450 if ((stmf_wm_last != 0) &&
6451 ((d = ddi_get_lbolt() - stmf_wm_last) > tps)) {
6452 qd = 0;
6453 for (i = 0; i < stmf_nworkers_accepting_cmds; i++) {
6454 qd += stmf_workers[i].worker_max_qdepth_pu;
6455 stmf_workers[i].worker_max_qdepth_pu = 0;
6456 if (stmf_workers[i].worker_max_sys_qdepth_pu >
6457 cur_max_ntasks) {
6458 cur_max_ntasks =
6459 stmf_workers[i].worker_max_sys_qdepth_pu;
6460 }
6461 stmf_workers[i].worker_max_sys_qdepth_pu = 0;
6462 }
6463 }
6464 stmf_wm_last = ddi_get_lbolt();
6465 if (d <= tps) {
6466 /* still ramping up */
6467 return;
6468 }
6469 /* max qdepth cannot be more than max tasks */
6470 if (qd > cur_max_ntasks)
6471 qd = cur_max_ntasks;
6472
6473 /* See if we have more workers */
6474 if (qd < stmf_nworkers_accepting_cmds) {
6475 /*
6476 * Since we dont reduce the worker count right away, monitor
6477 * the highest load during the scale_down_delay.
6478 */
6479 if (qd > stmf_worker_scale_down_qd)
6480 stmf_worker_scale_down_qd = qd;
6481 if (stmf_worker_scale_down_timer == 0) {
6482 stmf_worker_scale_down_timer = ddi_get_lbolt() +
6483 drv_usectohz(stmf_worker_scale_down_delay *
6484 1000 * 1000);
6485 return;
6486 }
6487 if (ddi_get_lbolt() < stmf_worker_scale_down_timer) {
6488 return;
6489 }
6490 /* Its time to reduce the workers */
6491 if (stmf_worker_scale_down_qd < stmf_i_min_nworkers)
6492 stmf_worker_scale_down_qd = stmf_i_min_nworkers;
6493 if (stmf_worker_scale_down_qd > stmf_i_max_nworkers)
6494 stmf_worker_scale_down_qd = stmf_i_max_nworkers;
6495 if (stmf_worker_scale_down_qd == stmf_nworkers_cur)
6496 return;
6497 workers_needed = stmf_worker_scale_down_qd;
6498 stmf_worker_scale_down_qd = 0;
6499 goto worker_mgmt_trigger_change;
6500 }
6501 stmf_worker_scale_down_qd = 0;
6502 stmf_worker_scale_down_timer = 0;
6503 if (qd > stmf_i_max_nworkers)
6504 qd = stmf_i_max_nworkers;
6505 if (qd < stmf_i_min_nworkers)
6506 qd = stmf_i_min_nworkers;
6507 if (qd == stmf_nworkers_cur)
6508 return;
6509 workers_needed = qd;
6510 goto worker_mgmt_trigger_change;
6511
6512 /* NOTREACHED */
6513 return;
6514
6515 worker_mgmt_trigger_change:
6516 ASSERT(workers_needed != stmf_nworkers_cur);
6517 if (workers_needed > stmf_nworkers_cur) {
6518 stmf_nworkers_needed = workers_needed;
6519 for (i = stmf_nworkers_cur; i < workers_needed; i++) {
6520 w = &stmf_workers[i];
6521 w->worker_tid = thread_create(NULL, 0, stmf_worker_task,
6522 (void *)&stmf_workers[i], 0, &p0, TS_RUN,
6523 minclsyspri);
6524 }
6525 return;
6526 }
6527 /* At this point we know that we are decreasing the # of workers */
6528 stmf_nworkers_accepting_cmds = workers_needed;
6529 stmf_nworkers_needed = workers_needed;
6530 /* Signal the workers that its time to quit */
6531 for (i = (stmf_nworkers_cur - 1); i >= stmf_nworkers_needed; i--) {
6532 w = &stmf_workers[i];
6533 ASSERT(w && (w->worker_flags & STMF_WORKER_STARTED));
6534 mutex_enter(&w->worker_lock);
6535 w->worker_flags |= STMF_WORKER_TERMINATE;
6536 if ((w->worker_flags & STMF_WORKER_ACTIVE) == 0)
6537 cv_signal(&w->worker_cv);
6538 mutex_exit(&w->worker_lock);
6539 }
6540 }
6541
6542 /*
6543 * Fills out a dbuf from stmf_xfer_data_t (contained in the db_lu_private).
6544 * If all the data has been filled out, frees the xd and makes
6545 * db_lu_private NULL.
6546 */
6547 void
6548 stmf_xd_to_dbuf(stmf_data_buf_t *dbuf, int set_rel_off)
6549 {
6550 stmf_xfer_data_t *xd;
6551 uint8_t *p;
6552 int i;
6553 uint32_t s;
6554
6555 xd = (stmf_xfer_data_t *)dbuf->db_lu_private;
6556 dbuf->db_data_size = 0;
6557 if (set_rel_off)
6558 dbuf->db_relative_offset = xd->size_done;
6559 for (i = 0; i < dbuf->db_sglist_length; i++) {
6560 s = min(xd->size_left, dbuf->db_sglist[i].seg_length);
6561 p = &xd->buf[xd->size_done];
6638 stmf_abort(STMF_QUEUE_TASK_ABORT, task,
6639 STMF_ALLOC_FAILURE, NULL);
6640 return;
6641 }
6642 dbuf->db_lu_private = NULL;
6643
6644 p = dbuf->db_sglist[0].seg_addr;
6645
6646 /*
6647 * Standard inquiry handling only.
6648 */
6649
6650 bzero(p, inq_page_length + 5);
6651
6652 p[0] = DPQ_SUPPORTED | DTYPE_UNKNOWN;
6653 p[2] = 5;
6654 p[3] = 0x12;
6655 p[4] = inq_page_length;
6656 p[6] = 0x80;
6657
6658 (void) strncpy((char *)p+8, "SUN ", 8);
6659 (void) strncpy((char *)p+16, "COMSTAR ", 16);
6660 (void) strncpy((char *)p+32, "1.0 ", 4);
6661
6662 dbuf->db_data_size = sz;
6663 dbuf->db_relative_offset = 0;
6664 dbuf->db_flags = DB_DIRECTION_TO_RPORT;
6665 (void) stmf_xfer_data(task, dbuf, 0);
6666
6667 return;
6668
6669 case SCMD_REPORT_LUNS:
6670 task->task_cmd_xfer_length =
6671 ((((uint32_t)task->task_cdb[6]) << 24) |
6672 (((uint32_t)task->task_cdb[7]) << 16) |
6673 (((uint32_t)task->task_cdb[8]) << 8) |
6674 ((uint32_t)task->task_cdb[9]));
6675
6676 if (task->task_additional_flags &
6677 TASK_AF_NO_EXPECTED_XFER_LENGTH) {
6678 task->task_expected_xfer_length =
6679 task->task_cmd_xfer_length;
6680 }
6850 case TM_CLEAR_TASK_SET:
6851 case TM_LUN_RESET:
6852 (void) stmf_lun_reset_poll(task->task_lu, task, 0);
6853 return;
6854 case TM_TARGET_RESET:
6855 case TM_TARGET_COLD_RESET:
6856 case TM_TARGET_WARM_RESET:
6857 stmf_target_reset_poll(task);
6858 return;
6859 }
6860 }
6861
6862 /* ARGSUSED */
6863 void
6864 stmf_dlun0_ctl(struct stmf_lu *lu, int cmd, void *arg)
6865 {
6866 /* This function will never be called */
6867 cmn_err(CE_WARN, "stmf_dlun0_ctl called with cmd %x", cmd);
6868 }
6869
6870 void
6871 stmf_dlun_init()
6872 {
6873 stmf_i_lu_t *ilu;
6874
6875 dlun0 = stmf_alloc(STMF_STRUCT_STMF_LU, 0, 0);
6876 dlun0->lu_task_alloc = stmf_dlun0_task_alloc;
6877 dlun0->lu_new_task = stmf_dlun0_new_task;
6878 dlun0->lu_dbuf_xfer_done = stmf_dlun0_dbuf_done;
6879 dlun0->lu_send_status_done = stmf_dlun0_status_done;
6880 dlun0->lu_task_free = stmf_dlun0_task_free;
6881 dlun0->lu_abort = stmf_dlun0_abort;
6882 dlun0->lu_task_poll = stmf_dlun0_task_poll;
6883 dlun0->lu_ctl = stmf_dlun0_ctl;
6884
6885 ilu = (stmf_i_lu_t *)dlun0->lu_stmf_private;
6886 ilu->ilu_cur_task_cntr = &ilu->ilu_task_cntr1;
6887 }
6888
6889 stmf_status_t
6890 stmf_dlun_fini()
6891 {
6892 stmf_i_lu_t *ilu;
6893
6894 ilu = (stmf_i_lu_t *)dlun0->lu_stmf_private;
6895
6896 ASSERT(ilu->ilu_ntasks == ilu->ilu_ntasks_free);
6897 if (ilu->ilu_ntasks) {
6898 stmf_i_scsi_task_t *itask, *nitask;
6899
6900 nitask = ilu->ilu_tasks;
6901 do {
6902 itask = nitask;
7127 {
7128 if (STMF_EVENT_ENABLED(ilport->ilport_event_hdl, eventid) &&
7129 (ilport->ilport_lport->lport_event_handler != NULL)) {
7130 ilport->ilport_lport->lport_event_handler(
7131 ilport->ilport_lport, eventid, arg, flags);
7132 }
7133 }
7134
7135 /*
7136 * With the possibility of having multiple itl sessions pointing to the
7137 * same itl_kstat_info, the ilu_kstat_lock mutex is used to synchronize
7138 * the kstat update of the ilu_kstat_io, itl_kstat_taskq and itl_kstat_lu_xfer
7139 * statistics.
7140 */
7141 void
7142 stmf_itl_task_start(stmf_i_scsi_task_t *itask)
7143 {
7144 stmf_itl_data_t *itl = itask->itask_itl_datap;
7145 scsi_task_t *task = itask->itask_task;
7146 stmf_i_lu_t *ilu;
7147
7148 if (itl == NULL || task->task_lu == dlun0)
7149 return;
7150 ilu = (stmf_i_lu_t *)task->task_lu->lu_stmf_private;
7151 itask->itask_start_timestamp = gethrtime();
7152 if (ilu->ilu_kstat_io != NULL) {
7153 mutex_enter(ilu->ilu_kstat_io->ks_lock);
7154 stmf_update_kstat_lu_q(itask->itask_task, kstat_waitq_enter);
7155 mutex_exit(ilu->ilu_kstat_io->ks_lock);
7156 }
7157
7158 stmf_update_kstat_lport_q(itask->itask_task, kstat_waitq_enter);
7159 }
7160
7161 void
7162 stmf_itl_lu_new_task(stmf_i_scsi_task_t *itask)
7163 {
7164 stmf_itl_data_t *itl = itask->itask_itl_datap;
7165 scsi_task_t *task = itask->itask_task;
7166 stmf_i_lu_t *ilu;
7167
7168 if (itl == NULL || task->task_lu == dlun0)
7169 return;
7170 ilu = (stmf_i_lu_t *)task->task_lu->lu_stmf_private;
7171 if (ilu->ilu_kstat_io != NULL) {
7172 mutex_enter(ilu->ilu_kstat_io->ks_lock);
7173 stmf_update_kstat_lu_q(itask->itask_task, kstat_waitq_to_runq);
7174 mutex_exit(ilu->ilu_kstat_io->ks_lock);
7175 }
7176
7177 stmf_update_kstat_lport_q(itask->itask_task, kstat_waitq_to_runq);
7178 }
7179
7180 void
7181 stmf_itl_task_done(stmf_i_scsi_task_t *itask)
7182 {
7183 stmf_itl_data_t *itl = itask->itask_itl_datap;
7184 scsi_task_t *task = itask->itask_task;
7185 stmf_i_lu_t *ilu;
7186
7187 itask->itask_done_timestamp = gethrtime();
7188
7189 if (itl == NULL || task->task_lu == dlun0)
7190 return;
7191 ilu = (stmf_i_lu_t *)task->task_lu->lu_stmf_private;
7192
7193 if (ilu->ilu_kstat_io == NULL)
7194 return;
7195
7196 mutex_enter(ilu->ilu_kstat_io->ks_lock);
7197
7198 if (itask->itask_flags & ITASK_KSTAT_IN_RUNQ) {
7199 stmf_update_kstat_lu_q(task, kstat_runq_exit);
7200 mutex_exit(ilu->ilu_kstat_io->ks_lock);
7201 stmf_update_kstat_lport_q(task, kstat_runq_exit);
7202 } else {
7203 stmf_update_kstat_lu_q(task, kstat_waitq_exit);
7204 mutex_exit(ilu->ilu_kstat_io->ks_lock);
7205 stmf_update_kstat_lport_q(task, kstat_waitq_exit);
7206 }
7207 }
7208
7209 static void
7210 stmf_lport_xfer_start(stmf_i_scsi_task_t *itask, stmf_data_buf_t *dbuf)
7211 {
7212 stmf_itl_data_t *itl = itask->itask_itl_datap;
7213
7214 if (itl == NULL)
7215 return;
7216
7217 DTRACE_PROBE2(scsi__xfer__start, scsi_task_t *, itask->itask_task,
7218 stmf_data_buf_t *, dbuf);
7219
7220 dbuf->db_xfer_start_timestamp = gethrtime();
7221 }
7222
7223 static void
7224 stmf_lport_xfer_done(stmf_i_scsi_task_t *itask, stmf_data_buf_t *dbuf)
7225 {
7226 stmf_itl_data_t *itl = itask->itask_itl_datap;
7227 hrtime_t elapsed_time;
7228 uint64_t xfer_size;
7229
7230 if (itl == NULL)
7231 return;
7232
7233 xfer_size = (dbuf->db_xfer_status == STMF_SUCCESS) ?
7234 dbuf->db_data_size : 0;
7235
7236 elapsed_time = gethrtime() - dbuf->db_xfer_start_timestamp;
7237 if (dbuf->db_flags & DB_DIRECTION_TO_RPORT) {
7238 atomic_add_64((uint64_t *)&itask->itask_lport_read_time,
7239 elapsed_time);
7240 atomic_add_64((uint64_t *)&itask->itask_read_xfer,
7241 xfer_size);
7242 } else {
7243 atomic_add_64((uint64_t *)&itask->itask_lport_write_time,
7244 elapsed_time);
7245 atomic_add_64((uint64_t *)&itask->itask_write_xfer,
7246 xfer_size);
7247 }
7248
7249 DTRACE_PROBE3(scsi__xfer__end, scsi_task_t *, itask->itask_task,
7250 stmf_data_buf_t *, dbuf, hrtime_t, elapsed_time);
7251
7252 dbuf->db_xfer_start_timestamp = 0;
7253 }
7254
7255 void
7256 stmf_svc_init()
7257 {
7258 if (stmf_state.stmf_svc_flags & STMF_SVC_STARTED)
7259 return;
7260 stmf_state.stmf_svc_tailp = &stmf_state.stmf_svc_active;
7261 stmf_state.stmf_svc_taskq = ddi_taskq_create(0, "STMF_SVC_TASKQ", 1,
7262 TASKQ_DEFAULTPRI, 0);
7263 (void) ddi_taskq_dispatch(stmf_state.stmf_svc_taskq,
7264 stmf_svc, 0, DDI_SLEEP);
7265 }
7266
7267 stmf_status_t
7268 stmf_svc_fini()
7269 {
7270 uint32_t i;
7271
7272 mutex_enter(&stmf_state.stmf_lock);
7273 if (stmf_state.stmf_svc_flags & STMF_SVC_STARTED) {
7274 stmf_state.stmf_svc_flags |= STMF_SVC_TERMINATE;
7275 cv_signal(&stmf_state.stmf_cv);
7276 }
7277 mutex_exit(&stmf_state.stmf_lock);
7278
7279 /* Wait for 5 seconds */
7280 for (i = 0; i < 500; i++) {
7281 if (stmf_state.stmf_svc_flags & STMF_SVC_STARTED)
7282 delay(drv_usectohz(10000));
7283 else
7284 break;
7285 }
7286 if (i == 500)
7287 return (STMF_BUSY);
7288
7289 ddi_taskq_destroy(stmf_state.stmf_svc_taskq);
7290
7291 return (STMF_SUCCESS);
7292 }
7293
7294 struct stmf_svc_clocks {
7295 clock_t drain_start, drain_next;
7296 clock_t timing_start, timing_next;
7297 clock_t worker_delay;
7298 };
7299
7300 /* ARGSUSED */
7301 void
7302 stmf_svc(void *arg)
7303 {
7304 stmf_svc_req_t *req;
7305 stmf_lu_t *lu;
7306 stmf_i_lu_t *ilu;
7307 stmf_local_port_t *lport;
7308 struct stmf_svc_clocks clks = { 0 };
7309
7310 mutex_enter(&stmf_state.stmf_lock);
7311 stmf_state.stmf_svc_flags |= STMF_SVC_STARTED | STMF_SVC_ACTIVE;
7312
7313 while (!(stmf_state.stmf_svc_flags & STMF_SVC_TERMINATE)) {
7314 if (stmf_state.stmf_svc_active == NULL) {
7315 stmf_svc_timeout(&clks);
7316 continue;
7317 }
7318
7319 /*
7320 * Pop the front request from the active list. After this,
7321 * the request will no longer be referenced by global state,
7322 * so it should be safe to access it without holding the
7323 * stmf state lock.
7324 */
7325 req = stmf_state.stmf_svc_active;
7326 stmf_state.stmf_svc_active = req->svc_next;
7327
7328 if (stmf_state.stmf_svc_active == NULL)
7329 stmf_state.stmf_svc_tailp = &stmf_state.stmf_svc_active;
7330
7331 switch (req->svc_cmd) {
7332 case STMF_CMD_LPORT_ONLINE:
7333 /* Fallthrough */
7334 case STMF_CMD_LPORT_OFFLINE:
7335 mutex_exit(&stmf_state.stmf_lock);
7336 lport = (stmf_local_port_t *)req->svc_obj;
7337 lport->lport_ctl(lport, req->svc_cmd, &req->svc_info);
7338 break;
7339 case STMF_CMD_LU_ONLINE:
7340 mutex_exit(&stmf_state.stmf_lock);
7341 lu = (stmf_lu_t *)req->svc_obj;
7342 lu->lu_ctl(lu, req->svc_cmd, &req->svc_info);
7343 break;
7344 case STMF_CMD_LU_OFFLINE:
7345 /* Remove all mappings of this LU */
7346 stmf_session_lu_unmapall((stmf_lu_t *)req->svc_obj);
7347 /* Kill all the pending I/Os for this LU */
7348 mutex_exit(&stmf_state.stmf_lock);
7349 stmf_task_lu_killall((stmf_lu_t *)req->svc_obj, NULL,
7350 STMF_ABORTED);
7381 if (stmf_state.stmf_nlus &&
7382 ((!clks->timing_next) || (ddi_get_lbolt() >= clks->timing_next))) {
7383 if (!stmf_state.stmf_svc_ilu_timing) {
7384 /* we are starting a new round */
7385 stmf_state.stmf_svc_ilu_timing =
7386 stmf_state.stmf_ilulist;
7387 clks->timing_start = ddi_get_lbolt();
7388 }
7389
7390 stmf_check_ilu_timing();
7391 if (!stmf_state.stmf_svc_ilu_timing) {
7392 /* we finished a complete round */
7393 clks->timing_next =
7394 clks->timing_start + drv_usectohz(5*1000*1000);
7395 } else {
7396 /* we still have some ilu items to check */
7397 clks->timing_next =
7398 ddi_get_lbolt() + drv_usectohz(1*1000*1000);
7399 }
7400
7401 if (stmf_state.stmf_svc_active)
7402 return;
7403 }
7404
7405 /* Check if there are free tasks to clear */
7406 if (stmf_state.stmf_nlus &&
7407 ((!clks->drain_next) || (ddi_get_lbolt() >= clks->drain_next))) {
7408 if (!stmf_state.stmf_svc_ilu_draining) {
7409 /* we are starting a new round */
7410 stmf_state.stmf_svc_ilu_draining =
7411 stmf_state.stmf_ilulist;
7412 clks->drain_start = ddi_get_lbolt();
7413 }
7414
7415 stmf_check_freetask();
7416 if (!stmf_state.stmf_svc_ilu_draining) {
7417 /* we finished a complete round */
7418 clks->drain_next =
7419 clks->drain_start + drv_usectohz(10*1000*1000);
7420 } else {
7421 /* we still have some ilu items to check */
7422 clks->drain_next =
7423 ddi_get_lbolt() + drv_usectohz(1*1000*1000);
7424 }
7425
7426 if (stmf_state.stmf_svc_active)
7427 return;
7428 }
7429
7430 /* Check if we need to run worker_mgmt */
7431 if (ddi_get_lbolt() > clks->worker_delay) {
7432 stmf_worker_mgmt();
7433 clks->worker_delay = ddi_get_lbolt() +
7434 stmf_worker_mgmt_delay;
7435 }
7436
7437 /* Check if any active session got its 1st LUN */
7438 if (stmf_state.stmf_process_initial_luns) {
7439 int stmf_level = 0;
7440 int port_level;
7441
7442 for (ilport = stmf_state.stmf_ilportlist; ilport;
7443 ilport = next_ilport) {
7444 int ilport_lock_held;
7445 next_ilport = ilport->ilport_next;
7446
7447 if ((ilport->ilport_flags &
7448 ILPORT_SS_GOT_INITIAL_LUNS) == 0)
7449 continue;
7450
7451 port_level = 0;
7452 rw_enter(&ilport->ilport_lock, RW_READER);
7453 ilport_lock_held = 1;
7454
7455 for (iss = ilport->ilport_ss_list; iss;
7456 iss = iss->iss_next) {
7543 stmf_svc_req_t *req;
7544 int s;
7545
7546 ASSERT(!mutex_owned(&stmf_state.stmf_lock));
7547 s = sizeof (stmf_svc_req_t);
7548 if (info->st_additional_info) {
7549 s += strlen(info->st_additional_info) + 1;
7550 }
7551 req = kmem_zalloc(s, KM_SLEEP);
7552
7553 req->svc_cmd = cmd;
7554 req->svc_obj = obj;
7555 req->svc_info.st_rflags = info->st_rflags;
7556 if (info->st_additional_info) {
7557 req->svc_info.st_additional_info = (char *)(GET_BYTE_OFFSET(req,
7558 sizeof (stmf_svc_req_t)));
7559 (void) strcpy(req->svc_info.st_additional_info,
7560 info->st_additional_info);
7561 }
7562 req->svc_req_alloc_size = s;
7563 req->svc_next = NULL;
7564
7565 mutex_enter(&stmf_state.stmf_lock);
7566 *stmf_state.stmf_svc_tailp = req;
7567 stmf_state.stmf_svc_tailp = &req->svc_next;
7568 if ((stmf_state.stmf_svc_flags & STMF_SVC_ACTIVE) == 0) {
7569 cv_signal(&stmf_state.stmf_cv);
7570 }
7571 mutex_exit(&stmf_state.stmf_lock);
7572 }
7573
7574 static void
7575 stmf_svc_kill_obj_requests(void *obj)
7576 {
7577 stmf_svc_req_t *prev_req = NULL;
7578 stmf_svc_req_t *next_req;
7579 stmf_svc_req_t *req;
7580
7581 ASSERT(mutex_owned(&stmf_state.stmf_lock));
7582
7583 for (req = stmf_state.stmf_svc_active; req != NULL; req = next_req) {
7584 next_req = req->svc_next;
7585
7586 if (req->svc_obj == obj) {
7587 if (prev_req != NULL)
7588 prev_req->svc_next = next_req;
7589 else
7590 stmf_state.stmf_svc_active = next_req;
7591
7592 if (next_req == NULL)
7593 stmf_state.stmf_svc_tailp = (prev_req != NULL) ?
7594 &prev_req->svc_next :
7595 &stmf_state.stmf_svc_active;
7596
7597 kmem_free(req, req->svc_req_alloc_size);
7598 } else {
7599 prev_req = req;
7600 }
7601 }
7602 }
7603
7604 void
7605 stmf_trace(caddr_t ident, const char *fmt, ...)
7606 {
7607 va_list args;
7608 char tbuf[160];
7609 int len;
7610
7611 if (!stmf_trace_on)
7612 return;
7613 len = snprintf(tbuf, 158, "%s:%07lu: ", ident ? ident : "",
7614 ddi_get_lbolt());
7615 va_start(args, fmt);
7616 len += vsnprintf(tbuf + len, 158 - len, fmt, args);
7617 va_end(args);
7618
7619 if (len > 158) {
7625 mutex_enter(&trace_buf_lock);
7626 bcopy(tbuf, &stmf_trace_buf[trace_buf_curndx], len+1);
7627 trace_buf_curndx += len;
7628 if (trace_buf_curndx > (trace_buf_size - 320))
7629 trace_buf_curndx = 0;
7630 mutex_exit(&trace_buf_lock);
7631 }
7632
7633 void
7634 stmf_trace_clear()
7635 {
7636 if (!stmf_trace_on)
7637 return;
7638 mutex_enter(&trace_buf_lock);
7639 trace_buf_curndx = 0;
7640 if (trace_buf_size > 0)
7641 stmf_trace_buf[0] = 0;
7642 mutex_exit(&trace_buf_lock);
7643 }
7644
7645 static void
7646 stmf_abort_task_offline(scsi_task_t *task, int offline_lu, char *info)
7647 {
7648 stmf_state_change_info_t change_info;
7649 void *ctl_private;
7650 uint32_t ctl_cmd;
7651 int msg = 0;
7652
7653 stmf_trace("FROM STMF", "abort_task_offline called for %s: %s",
7654 offline_lu ? "LU" : "LPORT", info ? info : "no additional info");
7655 change_info.st_additional_info = info;
7656 if (offline_lu) {
7657 change_info.st_rflags = STMF_RFLAG_RESET |
7658 STMF_RFLAG_LU_ABORT;
7659 ctl_private = task->task_lu;
7660 if (((stmf_i_lu_t *)
7661 task->task_lu->lu_stmf_private)->ilu_state ==
7662 STMF_STATE_ONLINE) {
7663 msg = 1;
7664 }
7665 ctl_cmd = STMF_CMD_LU_OFFLINE;
7666 } else {
7667 change_info.st_rflags = STMF_RFLAG_RESET |
7668 STMF_RFLAG_LPORT_ABORT;
7669 ctl_private = task->task_lport;
7670 if (((stmf_i_local_port_t *)
7671 task->task_lport->lport_stmf_private)->ilport_state ==
7672 STMF_STATE_ONLINE) {
7673 msg = 1;
7674 }
7675 ctl_cmd = STMF_CMD_LPORT_OFFLINE;
7676 }
7677
7678 if (msg) {
7679 stmf_trace(0, "Calling stmf_ctl to offline %s : %s",
7680 offline_lu ? "LU" : "LPORT", info ? info :
7681 "<no additional info>");
7682 }
7683 (void) stmf_ctl(ctl_cmd, ctl_private, &change_info);
7684 }
7685
7686 static char
7687 stmf_ctoi(char c)
7688 {
7689 if ((c >= '0') && (c <= '9'))
7690 c -= '0';
7691 else if ((c >= 'A') && (c <= 'F'))
7692 c = c - 'A' + 10;
7693 else if ((c >= 'a') && (c <= 'f'))
7694 c = c - 'a' + 10;
7695 else
7696 c = -1;
7697 return (c);
7698 }
7699
7700 /* Convert from Hex value in ASCII format to the equivalent bytes */
7701 static boolean_t
7702 stmf_base16_str_to_binary(char *c, int dplen, uint8_t *dp)
7703 {
7720
7721 boolean_t
7722 stmf_scsilib_tptid_validate(scsi_transport_id_t *tptid, uint32_t total_sz,
7723 uint16_t *tptid_sz)
7724 {
7725 uint16_t tpd_len = SCSI_TPTID_SIZE;
7726
7727 if (tptid_sz)
7728 *tptid_sz = 0;
7729 if (total_sz < sizeof (scsi_transport_id_t))
7730 return (B_FALSE);
7731
7732 switch (tptid->protocol_id) {
7733
7734 case PROTOCOL_FIBRE_CHANNEL:
7735 /* FC Transport ID validation checks. SPC3 rev23, Table 284 */
7736 if (total_sz < tpd_len || tptid->format_code != 0)
7737 return (B_FALSE);
7738 break;
7739
7740 case PROTOCOL_iSCSI:
7741 {
7742 iscsi_transport_id_t *iscsiid;
7743 uint16_t adn_len, name_len;
7744
7745 /* Check for valid format code, SPC3 rev 23 Table 288 */
7746 if ((total_sz < tpd_len) ||
7747 (tptid->format_code != 0 && tptid->format_code != 1))
7748 return (B_FALSE);
7749
7750 iscsiid = (iscsi_transport_id_t *)tptid;
7751 adn_len = READ_SCSI16(iscsiid->add_len, uint16_t);
7752 tpd_len = sizeof (iscsi_transport_id_t) + adn_len - 1;
7753
7754 /*
7755 * iSCSI Transport ID validation checks.
7756 * As per SPC3 rev 23 Section 7.5.4.6 and Table 289 & Table 290
7757 */
7758 if (adn_len < 20 || (adn_len % 4 != 0))
7759 return (B_FALSE);
7760
7765 /* If the format_code is 1 check for ISID seperator */
7766 if ((tptid->format_code == 1) && (strstr(iscsiid->iscsi_name,
7767 SCSI_TPTID_ISCSI_ISID_SEPERATOR) == NULL))
7768 return (B_FALSE);
7769
7770 }
7771 break;
7772
7773 case PROTOCOL_SRP:
7774 /* SRP Transport ID validation checks. SPC3 rev23, Table 287 */
7775 if (total_sz < tpd_len || tptid->format_code != 0)
7776 return (B_FALSE);
7777 break;
7778
7779 case PROTOCOL_PARALLEL_SCSI:
7780 case PROTOCOL_SSA:
7781 case PROTOCOL_IEEE_1394:
7782 case PROTOCOL_SAS:
7783 case PROTOCOL_ADT:
7784 case PROTOCOL_ATAPI:
7785 default:
7786 {
7787 stmf_dflt_scsi_tptid_t *dflttpd;
7788
7789 tpd_len = sizeof (stmf_dflt_scsi_tptid_t);
7790 if (total_sz < tpd_len)
7791 return (B_FALSE);
7792 dflttpd = (stmf_dflt_scsi_tptid_t *)tptid;
7793 tpd_len = tpd_len + SCSI_READ16(&dflttpd->ident_len) - 1;
7794 if (total_sz < tpd_len)
7795 return (B_FALSE);
7796 }
7797 break;
7798 }
7799 if (tptid_sz)
7800 *tptid_sz = tpd_len;
7801 return (B_TRUE);
7802 }
7803
7804 boolean_t
7805 stmf_scsilib_tptid_compare(scsi_transport_id_t *tpd1,
7806 scsi_transport_id_t *tpd2)
7807 {
7808 if ((tpd1->protocol_id != tpd2->protocol_id) ||
7809 (tpd1->format_code != tpd2->format_code))
7810 return (B_FALSE);
7811
7812 switch (tpd1->protocol_id) {
7813
7814 case PROTOCOL_iSCSI:
7815 {
7816 iscsi_transport_id_t *iscsitpd1, *iscsitpd2;
7817 uint16_t len;
7818
7819 iscsitpd1 = (iscsi_transport_id_t *)tpd1;
7820 iscsitpd2 = (iscsi_transport_id_t *)tpd2;
7821 len = SCSI_READ16(&iscsitpd1->add_len);
7822 if ((memcmp(iscsitpd1->add_len, iscsitpd2->add_len, 2) != 0) ||
7823 (memcmp(iscsitpd1->iscsi_name, iscsitpd2->iscsi_name, len)
7824 != 0))
7825 return (B_FALSE);
7826 }
7827 break;
7828
7829 case PROTOCOL_SRP:
7830 {
7831 scsi_srp_transport_id_t *srptpd1, *srptpd2;
7832
7833 srptpd1 = (scsi_srp_transport_id_t *)tpd1;
7834 srptpd2 = (scsi_srp_transport_id_t *)tpd2;
7835 if (memcmp(srptpd1->srp_name, srptpd2->srp_name,
7836 sizeof (srptpd1->srp_name)) != 0)
7837 return (B_FALSE);
7838 }
7839 break;
7840
7841 case PROTOCOL_FIBRE_CHANNEL:
7842 {
7843 scsi_fc_transport_id_t *fctpd1, *fctpd2;
7844
7845 fctpd1 = (scsi_fc_transport_id_t *)tpd1;
7846 fctpd2 = (scsi_fc_transport_id_t *)tpd2;
7847 if (memcmp(fctpd1->port_name, fctpd2->port_name,
7848 sizeof (fctpd1->port_name)) != 0)
7849 return (B_FALSE);
7850 }
7851 break;
7852
7853 case PROTOCOL_PARALLEL_SCSI:
7854 case PROTOCOL_SSA:
7855 case PROTOCOL_IEEE_1394:
7856 case PROTOCOL_SAS:
7857 case PROTOCOL_ADT:
7858 case PROTOCOL_ATAPI:
7859 default:
7860 {
7861 stmf_dflt_scsi_tptid_t *dflt1, *dflt2;
7862 uint16_t len;
7863
7864 dflt1 = (stmf_dflt_scsi_tptid_t *)tpd1;
7865 dflt2 = (stmf_dflt_scsi_tptid_t *)tpd2;
7866 len = SCSI_READ16(&dflt1->ident_len);
7867 if ((memcmp(dflt1->ident_len, dflt2->ident_len, 2) != 0) ||
7868 (memcmp(dflt1->ident, dflt2->ident, len) != 0))
7869 return (B_FALSE);
7870 }
7871 break;
7872 }
7873 return (B_TRUE);
7874 }
7875
7876 /*
7877 * Changes devid_desc to corresponding TransportID format
7878 * Returns :- pointer to stmf_remote_port_t
7879 * Note :- Allocates continous memory for stmf_remote_port_t and TransportID,
7950 ident_len = devid->ident_length;
7951 sz = ALIGNED_TO_8BYTE_BOUNDARY(sizeof (stmf_dflt_scsi_tptid_t) +
7952 ident_len - 1);
7953 rpt = stmf_remote_port_alloc(sz);
7954 rpt->rport_tptid->format_code = 0;
7955 rpt->rport_tptid->protocol_id = devid->protocol_id;
7956 dflt_tpd = (stmf_dflt_scsi_tptid_t *)rpt->rport_tptid;
7957 SCSI_WRITE16(dflt_tpd->ident_len, ident_len);
7958 (void) memcpy(dflt_tpd->ident, devid->ident, ident_len);
7959 break;
7960 }
7961 return (rpt);
7962
7963 devid_to_remote_port_fail:
7964 stmf_remote_port_free(rpt);
7965 return (NULL);
7966
7967 }
7968
7969 stmf_remote_port_t *
7970 stmf_remote_port_alloc(uint16_t tptid_sz) {
7971 stmf_remote_port_t *rpt;
7972 rpt = (stmf_remote_port_t *)kmem_zalloc(
7973 sizeof (stmf_remote_port_t) + tptid_sz, KM_SLEEP);
7974 rpt->rport_tptid_sz = tptid_sz;
7975 rpt->rport_tptid = (scsi_transport_id_t *)(rpt + 1);
7976 return (rpt);
7977 }
7978
7979 void
7980 stmf_remote_port_free(stmf_remote_port_t *rpt)
7981 {
7982 /*
7983 * Note: stmf_scsilib_devid_to_remote_port() function allocates
7984 * remote port structures for all transports in the same way, So
7985 * it is safe to deallocate it in a protocol independent manner.
7986 * If any of the allocation method changes, corresponding changes
7987 * need to be made here too.
7988 */
7989 kmem_free(rpt, sizeof (stmf_remote_port_t) + rpt->rport_tptid_sz);
7990 }
|
1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
24 */
25 /*
26 * Copyright 2019 Nexenta Systems, Inc. All rights reserved.
27 * Copyright (c) 2013 by Delphix. All rights reserved.
28 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
29 */
30
31 #include <sys/conf.h>
32 #include <sys/file.h>
33 #include <sys/ddi.h>
34 #include <sys/sunddi.h>
35 #include <sys/modctl.h>
36 #include <sys/scsi/scsi.h>
37 #include <sys/scsi/generic/persist.h>
38 #include <sys/scsi/impl/scsi_reset_notify.h>
39 #include <sys/disp.h>
40 #include <sys/byteorder.h>
41 #include <sys/atomic.h>
42 #include <sys/ethernet.h>
43 #include <sys/sdt.h>
44 #include <sys/nvpair.h>
45 #include <sys/zone.h>
46 #include <sys/id_space.h>
108 static void stmf_svc_kill_obj_requests(void *obj);
109 static void stmf_svc_timeout(struct stmf_svc_clocks *);
110 void stmf_check_freetask();
111 void stmf_abort_target_reset(scsi_task_t *task);
112 stmf_status_t stmf_lun_reset_poll(stmf_lu_t *lu, struct scsi_task *task,
113 int target_reset);
114 void stmf_target_reset_poll(struct scsi_task *task);
115 void stmf_handle_lun_reset(scsi_task_t *task);
116 void stmf_handle_target_reset(scsi_task_t *task);
117 void stmf_xd_to_dbuf(stmf_data_buf_t *dbuf, int set_rel_off);
118 int stmf_load_ppd_ioctl(stmf_ppioctl_data_t *ppi, uint64_t *ppi_token,
119 uint32_t *err_ret);
120 int stmf_delete_ppd_ioctl(stmf_ppioctl_data_t *ppi);
121 int stmf_get_ppd_ioctl(stmf_ppioctl_data_t *ppi, stmf_ppioctl_data_t *ppi_out,
122 uint32_t *err_ret);
123 void stmf_delete_ppd(stmf_pp_data_t *ppd);
124 void stmf_delete_all_ppds();
125 void stmf_trace_clear();
126 void stmf_worker_init();
127 stmf_status_t stmf_worker_fini();
128 void stmf_worker_task(void *arg);
129 static void stmf_task_lu_free(scsi_task_t *task, stmf_i_scsi_session_t *iss);
130 static stmf_status_t stmf_ic_lu_reg(stmf_ic_reg_dereg_lun_msg_t *msg,
131 uint32_t type);
132 static stmf_status_t stmf_ic_lu_dereg(stmf_ic_reg_dereg_lun_msg_t *msg);
133 static stmf_status_t stmf_ic_rx_scsi_status(stmf_ic_scsi_status_msg_t *msg);
134 static stmf_status_t stmf_ic_rx_status(stmf_ic_status_msg_t *msg);
135 static stmf_status_t stmf_ic_rx_scsi_data(stmf_ic_scsi_data_msg_t *msg);
136 void stmf_task_lu_killall(stmf_lu_t *lu, scsi_task_t *tm_task, stmf_status_t s);
137
138 /* pppt modhandle */
139 ddi_modhandle_t pppt_mod;
140
141 /* pppt modload imported functions */
142 stmf_ic_reg_port_msg_alloc_func_t ic_reg_port_msg_alloc;
143 stmf_ic_dereg_port_msg_alloc_func_t ic_dereg_port_msg_alloc;
144 stmf_ic_reg_lun_msg_alloc_func_t ic_reg_lun_msg_alloc;
145 stmf_ic_dereg_lun_msg_alloc_func_t ic_dereg_lun_msg_alloc;
146 stmf_ic_lun_active_msg_alloc_func_t ic_lun_active_msg_alloc;
147 stmf_ic_scsi_cmd_msg_alloc_func_t ic_scsi_cmd_msg_alloc;
148 stmf_ic_scsi_data_xfer_done_msg_alloc_func_t ic_scsi_data_xfer_done_msg_alloc;
149 stmf_ic_session_create_msg_alloc_func_t ic_session_reg_msg_alloc;
150 stmf_ic_session_destroy_msg_alloc_func_t ic_session_dereg_msg_alloc;
151 stmf_ic_tx_msg_func_t ic_tx_msg;
152 stmf_ic_msg_free_func_t ic_msg_free;
153
154 static void stmf_itl_task_start(stmf_i_scsi_task_t *itask);
155 static void stmf_itl_lu_new_task(stmf_i_scsi_task_t *itask);
156 static void stmf_itl_task_done(stmf_i_scsi_task_t *itask);
157
158 static void stmf_lport_xfer_start(stmf_i_scsi_task_t *itask,
159 stmf_data_buf_t *dbuf);
160 static void stmf_lport_xfer_done(stmf_i_scsi_task_t *itask,
161 stmf_data_buf_t *dbuf);
162
163 static void stmf_update_kstat_lu_q(scsi_task_t *, void());
164 static void stmf_update_kstat_lport_q(scsi_task_t *, void());
165 static void stmf_update_kstat_lu_io(scsi_task_t *, stmf_data_buf_t *);
166 static void stmf_update_kstat_lport_io(scsi_task_t *, stmf_data_buf_t *);
167 static hrtime_t stmf_update_rport_timestamps(hrtime_t *start_tstamp,
168 hrtime_t *done_tstamp, stmf_i_scsi_task_t *itask);
169
170 static int stmf_irport_compare(const void *void_irport1,
171 const void *void_irport2);
172 static void stmf_create_kstat_rport(stmf_i_remote_port_t *irport);
173 static void stmf_destroy_kstat_rport(stmf_i_remote_port_t *irport);
174 static int stmf_kstat_rport_update(kstat_t *ksp, int rw);
175 static stmf_i_remote_port_t *stmf_irport_create(scsi_devid_desc_t *rport_devid);
176 static void stmf_irport_destroy(stmf_i_remote_port_t *irport);
177 static stmf_i_remote_port_t *stmf_irport_register(
178 scsi_devid_desc_t *rport_devid);
179 static stmf_i_remote_port_t *stmf_irport_lookup_locked(
180 scsi_devid_desc_t *rport_devid);
181 static void stmf_irport_deregister(stmf_i_remote_port_t *irport);
182
183 extern struct mod_ops mod_driverops;
184
185 /* =====[ Tunables ]===== */
186 /* Internal tracing */
187 volatile int stmf_trace_on = 0;
188 volatile int stmf_trace_buf_size = (1 * 1024 * 1024);
189 /*
190 * The reason default task timeout is 75 is because we want the
191 * host to timeout 1st and mostly host timeout is 60 seconds.
192 */
193 volatile int stmf_default_task_timeout = 75;
194 /*
195 * Setting this to one means, you are responsible for config load and keeping
196 * things in sync with persistent database.
197 */
198 volatile int stmf_allow_modunload = 0;
199
200 volatile int stmf_nworkers = 512;
201 volatile int stmf_worker_warn = 0;
202
203 /* === [ Debugging and fault injection ] === */
204 #ifdef DEBUG
205 volatile int stmf_drop_task_counter = 0;
206 volatile int stmf_drop_buf_counter = 0;
207
208 #endif
209
210 stmf_state_t stmf_state;
211 static stmf_lu_t *dlun0;
212
213 static uint8_t stmf_first_zero[] =
214 { 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0, 0xff };
215 static uint8_t stmf_first_one[] =
216 { 0xff, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0 };
217
218 static kmutex_t trace_buf_lock;
219 static int trace_buf_size;
220 static int trace_buf_curndx;
221 caddr_t stmf_trace_buf;
222
223 static enum {
224 STMF_WORKERS_DISABLED = 0,
225 STMF_WORKERS_ENABLING,
226 STMF_WORKERS_ENABLED
227 } stmf_workers_state = STMF_WORKERS_DISABLED;
228 static kmutex_t stmf_worker_sel_mx;
229 volatile uint32_t stmf_nworkers_cur = 0; /* # of workers currently running */
230 static int stmf_worker_sel_counter = 0;
231 static uint32_t stmf_cur_ntasks = 0;
232 static clock_t stmf_wm_next = 0;
233 static int stmf_nworkers_accepting_cmds;
234 static stmf_worker_t *stmf_workers = NULL;
235 static clock_t stmf_worker_scale_down_timer = 0;
236 static int stmf_worker_scale_down_qd = 0;
237
238 static struct cb_ops stmf_cb_ops = {
239 stmf_open, /* open */
240 stmf_close, /* close */
241 nodev, /* strategy */
242 nodev, /* print */
243 nodev, /* dump */
244 nodev, /* read */
245 nodev, /* write */
246 stmf_ioctl, /* ioctl */
247 nodev, /* devmap */
248 nodev, /* mmap */
249 nodev, /* segmap */
250 nochpoll, /* chpoll */
251 ddi_prop_op, /* cb_prop_op */
252 0, /* streamtab */
253 D_NEW | D_MP, /* cb_flag */
254 CB_REV, /* rev */
255 nodev, /* aread */
256 nodev /* awrite */
257 };
258
259 static struct dev_ops stmf_ops = {
260 DEVO_REV,
261 0,
262 stmf_getinfo,
263 nulldev, /* identify */
264 nulldev, /* probe */
265 stmf_attach,
266 stmf_detach,
267 nodev, /* reset */
268 &stmf_cb_ops,
269 NULL, /* bus_ops */
270 NULL /* power */
271 };
272
273
274 #define STMF_MODULE_NAME "stmf"
275
276 #ifdef DEBUG
277 #define STMF_NAME "COMSTAR STMF D " __DATE__ " " __TIME__
278 #else
279 #define STMF_NAME "COMSTAR STMF"
280 #endif
281
282 static struct modldrv modldrv = {
283 &mod_driverops,
284 STMF_NAME,
285 &stmf_ops
286 };
287
288 static struct modlinkage modlinkage = {
289 MODREV_1,
290 &modldrv,
291 NULL
292 };
293
294 int
295 _init(void)
296 {
297 int ret;
298
299 ret = mod_install(&modlinkage);
300 if (ret)
301 return (ret);
302 stmf_trace_buf = kmem_zalloc(stmf_trace_buf_size, KM_SLEEP);
303 trace_buf_size = stmf_trace_buf_size;
304 trace_buf_curndx = 0;
305 mutex_init(&trace_buf_lock, NULL, MUTEX_DRIVER, 0);
306 mutex_init(&stmf_worker_sel_mx, NULL, MUTEX_ADAPTIVE, 0);
307 bzero(&stmf_state, sizeof (stmf_state_t));
308 /* STMF service is off by default */
309 stmf_state.stmf_service_running = 0;
310 /* default lu/lport states are online */
311 stmf_state.stmf_default_lu_state = STMF_STATE_ONLINE;
312 stmf_state.stmf_default_lport_state = STMF_STATE_ONLINE;
313 mutex_init(&stmf_state.stmf_lock, NULL, MUTEX_DRIVER, NULL);
314 cv_init(&stmf_state.stmf_cv, NULL, CV_DRIVER, NULL);
315 stmf_session_counter = (uint64_t)ddi_get_lbolt();
316 avl_create(&stmf_state.stmf_irportlist,
317 stmf_irport_compare, sizeof (stmf_i_remote_port_t),
318 offsetof(stmf_i_remote_port_t, irport_ln));
319 stmf_state.stmf_ilport_inst_space =
320 id_space_create("lport-instances", 0, MAX_ILPORT);
321 stmf_state.stmf_irport_inst_space =
322 id_space_create("rport-instances", 0, MAX_IRPORT);
323 stmf_view_init();
324 stmf_svc_init();
325 stmf_dlun_init();
326 return (ret);
357 ret = mod_remove(&modlinkage);
358 if (ret) {
359 stmf_svc_init();
360 stmf_dlun_init();
361 stmf_worker_init();
362 return (ret);
363 }
364
365 stmf_view_clear_config();
366
367 while ((irport = avl_destroy_nodes(&stmf_state.stmf_irportlist,
368 &avl_dest_cookie)) != NULL)
369 stmf_irport_destroy(irport);
370 avl_destroy(&stmf_state.stmf_irportlist);
371 id_space_destroy(stmf_state.stmf_ilport_inst_space);
372 id_space_destroy(stmf_state.stmf_irport_inst_space);
373
374 kmem_free(stmf_trace_buf, stmf_trace_buf_size);
375 mutex_destroy(&trace_buf_lock);
376 mutex_destroy(&stmf_state.stmf_lock);
377 mutex_destroy(&stmf_worker_sel_mx);
378 cv_destroy(&stmf_state.stmf_cv);
379 return (ret);
380 }
381
382 int
383 _info(struct modinfo *modinfop)
384 {
385 return (mod_info(&modlinkage, modinfop));
386 }
387
388 /* ARGSUSED */
389 static int
390 stmf_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg, void **result)
391 {
392 switch (cmd) {
393 case DDI_INFO_DEVT2DEVINFO:
394 *result = stmf_state.stmf_dip;
395 break;
396 case DDI_INFO_DEVT2INSTANCE:
397 *result =
1643 find_task_from_msgid(uint8_t *lu_id, stmf_ic_msgid_t task_msgid)
1644 {
1645 stmf_i_lu_t *ilu;
1646 stmf_i_scsi_task_t *itask;
1647
1648 mutex_enter(&stmf_state.stmf_lock);
1649 for (ilu = stmf_state.stmf_ilulist; ilu != NULL; ilu = ilu->ilu_next) {
1650 if (bcmp(lu_id, ilu->ilu_lu->lu_id->ident, 16) == 0) {
1651 break;
1652 }
1653 }
1654
1655 if (ilu == NULL) {
1656 mutex_exit(&stmf_state.stmf_lock);
1657 return (NULL);
1658 }
1659
1660 mutex_enter(&ilu->ilu_task_lock);
1661 for (itask = ilu->ilu_tasks; itask != NULL;
1662 itask = itask->itask_lu_next) {
1663 mutex_enter(&itask->itask_mutex);
1664 if (itask->itask_flags & (ITASK_IN_FREE_LIST |
1665 ITASK_BEING_ABORTED)) {
1666 mutex_exit(&itask->itask_mutex);
1667 continue;
1668 }
1669 mutex_exit(&itask->itask_mutex);
1670 if (itask->itask_proxy_msg_id == task_msgid) {
1671 break;
1672 }
1673 }
1674 mutex_exit(&ilu->ilu_task_lock);
1675 mutex_exit(&stmf_state.stmf_lock);
1676
1677 if (itask != NULL) {
1678 return (itask->itask_task);
1679 } else {
1680 /* task not found. Likely already aborted. */
1681 return (NULL);
1682 }
1683 }
1684
1685 /*
1686 * message received from pppt/ic
1687 */
1688 stmf_status_t
1689 stmf_msg_rx(stmf_ic_msg_t *msg)
1894 }
1895
1896 if (ilport->ilport_proxy_registered == 0) {
1897 return (STMF_FAILURE);
1898 }
1899
1900 mutex_enter(&stmf_state.stmf_lock);
1901 itask->itask_proxy_msg_id = stmf_proxy_msg_id++;
1902 mutex_exit(&stmf_state.stmf_lock);
1903 itask->itask_proxy_dbuf = dbuf;
1904
1905 /*
1906 * stmf will now take over the task handling for this task
1907 * but it still needs to be treated differently from other
1908 * default handled tasks, hence the ITASK_PROXY_TASK.
1909 * If this is a task management function, we're really just
1910 * duping the command to the peer. Set the TM bit so that
1911 * we can recognize this on return since we won't be completing
1912 * the proxied task in that case.
1913 */
1914 mutex_enter(&itask->itask_mutex);
1915 if (task->task_mgmt_function) {
1916 itask->itask_proxy_msg_id |= MSG_ID_TM_BIT;
1917 } else {
1918 if (itask->itask_flags & ITASK_BEING_ABORTED) {
1919 mutex_exit(&itask->itask_mutex);
1920 return (STMF_FAILURE);
1921 }
1922 itask->itask_flags |= ITASK_DEFAULT_HANDLING | ITASK_PROXY_TASK;
1923 }
1924 if (dbuf) {
1925 ic_cmd_msg = ic_scsi_cmd_msg_alloc(itask->itask_proxy_msg_id,
1926 task, dbuf->db_data_size, dbuf->db_sglist[0].seg_addr,
1927 itask->itask_proxy_msg_id);
1928 } else {
1929 ic_cmd_msg = ic_scsi_cmd_msg_alloc(itask->itask_proxy_msg_id,
1930 task, 0, NULL, itask->itask_proxy_msg_id);
1931 }
1932 mutex_exit(&itask->itask_mutex);
1933 if (ic_cmd_msg) {
1934 ic_ret = ic_tx_msg(ic_cmd_msg);
1935 if (ic_ret == STMF_IC_MSG_SUCCESS) {
1936 ret = STMF_SUCCESS;
1937 }
1938 }
1939 return (ret);
1940 }
1941
1942
1943 stmf_status_t
1944 pppt_modload()
1945 {
1946 int error;
1947
1948 if (pppt_mod == NULL && ((pppt_mod =
1949 ddi_modopen("drv/pppt", KRTLD_MODE_FIRST, &error)) == NULL)) {
1950 cmn_err(CE_WARN, "Unable to load pppt");
1951 return (STMF_FAILURE);
1952 }
2516 /*
2517 * User is requesting that the token be checked.
2518 * If there was another set after the user's get
2519 * it's an error
2520 */
2521 if (ppi->ppi_token_valid) {
2522 if (ppi->ppi_token != ppd->ppd_token) {
2523 *err_ret = STMF_IOCERR_PPD_UPDATED;
2524 mutex_exit(&stmf_state.stmf_lock);
2525 return (EINVAL);
2526 }
2527 }
2528
2529 if ((ret = nvlist_unpack((char *)ppi->ppi_data,
2530 (size_t)ppi->ppi_data_size, &nv, KM_NOSLEEP)) != 0) {
2531 mutex_exit(&stmf_state.stmf_lock);
2532 return (ret);
2533 }
2534
2535 /* Free any existing lists and add this one to the ppd */
2536 if (ppd->ppd_nv)
2537 nvlist_free(ppd->ppd_nv);
2538 ppd->ppd_nv = nv;
2539
2540 /* set the token for writes */
2541 ppd->ppd_token++;
2542 /* return token to caller */
2543 if (ppi_token) {
2544 *ppi_token = ppd->ppd_token;
2545 }
2546
2547 /* If there is a provider registered, do the notifications */
2548 if (ppd->ppd_provider) {
2549 uint32_t cb_flags = 0;
2550
2551 if (stmf_state.stmf_config_state == STMF_CONFIG_INIT)
2552 cb_flags |= STMF_PCB_STMF_ONLINING;
2553 if (ppi->ppi_lu_provider) {
2554 ilp = (stmf_i_lu_provider_t *)ppd->ppd_provider;
2555 if (ilp->ilp_lp->lp_cb == NULL)
2556 goto bail_out;
2589 if (ppd->ppd_lu_provider) {
2590 ((stmf_i_lu_provider_t *)
2591 ppd->ppd_provider)->ilp_ppd = NULL;
2592 } else {
2593 ((stmf_i_port_provider_t *)
2594 ppd->ppd_provider)->ipp_ppd = NULL;
2595 }
2596 ppd->ppd_provider = NULL;
2597 }
2598
2599 for (pppd = &stmf_state.stmf_ppdlist; *pppd != NULL;
2600 pppd = &((*pppd)->ppd_next)) {
2601 if (*pppd == ppd)
2602 break;
2603 }
2604
2605 if (*pppd == NULL)
2606 return;
2607
2608 *pppd = ppd->ppd_next;
2609 if (ppd->ppd_nv)
2610 nvlist_free(ppd->ppd_nv);
2611
2612 kmem_free(ppd, ppd->ppd_alloc_size);
2613 }
2614
2615 int
2616 stmf_delete_ppd_ioctl(stmf_ppioctl_data_t *ppi)
2617 {
2618 stmf_pp_data_t *ppd;
2619 int ret = ENOENT;
2620
2621 if ((ppi->ppi_lu_provider + ppi->ppi_port_provider) != 1) {
2622 return (EINVAL);
2623 }
2624
2625 mutex_enter(&stmf_state.stmf_lock);
2626
2627 for (ppd = stmf_state.stmf_ppdlist; ppd != NULL; ppd = ppd->ppd_next) {
2628 if (ppi->ppi_lu_provider) {
2629 if (!ppd->ppd_lu_provider)
2699 }
2700
2701 void
2702 stmf_delete_all_ppds()
2703 {
2704 stmf_pp_data_t *ppd, *nppd;
2705
2706 ASSERT(mutex_owned(&stmf_state.stmf_lock));
2707 for (ppd = stmf_state.stmf_ppdlist; ppd != NULL; ppd = nppd) {
2708 nppd = ppd->ppd_next;
2709 stmf_delete_ppd(ppd);
2710 }
2711 }
2712
2713 /*
2714 * 16 is the max string length of a protocol_ident, increase
2715 * the size if needed.
2716 */
2717 #define STMF_KSTAT_LU_SZ (STMF_GUID_INPUT + 1 + 256)
2718 #define STMF_KSTAT_TGT_SZ (256 * 2 + 16)
2719 #define STMF_KSTAT_RPORT_DATAMAX (sizeof (stmf_kstat_rport_info_t) / \
2720 sizeof (kstat_named_t))
2721
2722 /*
2723 * This array matches the Protocol Identifier in stmf_ioctl.h
2724 */
2725 #define MAX_PROTO_STR_LEN 32
2726
2727 char *protocol_ident[PROTOCOL_ANY] = {
2728 "Fibre Channel",
2729 "Parallel SCSI",
2730 "SSA",
2731 "IEEE_1394",
2732 "SRP",
2733 "iSCSI",
2734 "SAS",
2735 "ADT",
2736 "ATAPI",
2737 "UNKNOWN", "UNKNOWN", "UNKNOWN", "UNKNOWN", "UNKNOWN", "UNKNOWN"
2738 };
2739
2740 /*
2778 }
2779
2780 static void
2781 stmf_update_kstat_lport_io(scsi_task_t *task, stmf_data_buf_t *dbuf)
2782 {
2783 stmf_i_local_port_t *ilp;
2784 kstat_io_t *kip;
2785
2786 ilp = (stmf_i_local_port_t *)task->task_lport->lport_stmf_private;
2787 if (ilp != NULL && ilp->ilport_kstat_io != NULL) {
2788 kip = KSTAT_IO_PTR(ilp->ilport_kstat_io);
2789 if (kip != NULL) {
2790 mutex_enter(ilp->ilport_kstat_io->ks_lock);
2791 STMF_UPDATE_KSTAT_IO(kip, dbuf);
2792 mutex_exit(ilp->ilport_kstat_io->ks_lock);
2793 }
2794 }
2795 }
2796
2797 static void
2798 stmf_update_kstat_rport_io(scsi_task_t *task, stmf_data_buf_t *dbuf)
2799 {
2800 stmf_i_scsi_session_t *iss;
2801 stmf_i_remote_port_t *irport;
2802 kstat_io_t *kip;
2803
2804 iss = task->task_session->ss_stmf_private;
2805 irport = iss->iss_irport;
2806 if (irport->irport_kstat_io != NULL) {
2807 kip = KSTAT_IO_PTR(irport->irport_kstat_io);
2808 mutex_enter(irport->irport_kstat_io->ks_lock);
2809 STMF_UPDATE_KSTAT_IO(kip, dbuf);
2810 mutex_exit(irport->irport_kstat_io->ks_lock);
2811 }
2812 }
2813
2814 static void
2815 stmf_update_kstat_rport_estat(scsi_task_t *task)
2816 {
2817 stmf_i_scsi_task_t *itask;
2818 stmf_i_scsi_session_t *iss;
2819 stmf_i_remote_port_t *irport;
2820 stmf_kstat_rport_estat_t *ks_estat;
2821 hrtime_t lat = 0;
2822 uint32_t n = 0;
2823
2824 itask = task->task_stmf_private;
2825 iss = task->task_session->ss_stmf_private;
2826 irport = iss->iss_irport;
2827
2828 if (irport->irport_kstat_estat == NULL)
2829 return;
2830
2831 ks_estat = (stmf_kstat_rport_estat_t *)KSTAT_NAMED_PTR(
2832 irport->irport_kstat_estat);
2833
2834 mutex_enter(irport->irport_kstat_estat->ks_lock);
2835
2836 if (task->task_flags & TF_READ_DATA)
2837 n = atomic_dec_32_nv(&irport->irport_nread_tasks);
2838 else if (task->task_flags & TF_WRITE_DATA)
2839 n = atomic_dec_32_nv(&irport->irport_nwrite_tasks);
2840
2841 if (itask->itask_read_xfer > 0) {
2842 ks_estat->i_nread_tasks.value.ui64++;
2843 lat = stmf_update_rport_timestamps(
2844 &irport->irport_rdstart_timestamp,
2845 &irport->irport_rddone_timestamp, itask);
2846 if (n == 0)
2847 ks_estat->i_rport_read_latency.value.ui64 += lat;
2848 } else if ((itask->itask_write_xfer > 0) ||
2849 (task->task_flags & TF_INITIAL_BURST)) {
2850 ks_estat->i_nwrite_tasks.value.ui64++;
2851 lat = stmf_update_rport_timestamps(
2852 &irport->irport_wrstart_timestamp,
2853 &irport->irport_wrdone_timestamp, itask);
2854 if (n == 0)
2855 ks_estat->i_rport_write_latency.value.ui64 += lat;
2856 }
2857
2858 if (n == 0) {
2859 if (task->task_flags & TF_READ_DATA) {
2860 irport->irport_rdstart_timestamp = LLONG_MAX;
2861 irport->irport_rddone_timestamp = 0;
2862 } else if (task->task_flags & TF_WRITE_DATA) {
2863 irport->irport_wrstart_timestamp = LLONG_MAX;
2864 irport->irport_wrdone_timestamp = 0;
2865 }
2866 }
2867
2868 mutex_exit(irport->irport_kstat_estat->ks_lock);
2869 }
2870
2871 static hrtime_t
2872 stmf_update_rport_timestamps(hrtime_t *start_tstamp, hrtime_t *done_tstamp,
2873 stmf_i_scsi_task_t *itask)
2874 {
2875 *start_tstamp = MIN(*start_tstamp, itask->itask_start_timestamp);
2876 if ((*done_tstamp == 0) &&
2877 (itask->itask_xfer_done_timestamp == 0)) {
2878 *done_tstamp = *start_tstamp;
2879 } else {
2880 *done_tstamp = MAX(*done_tstamp,
2881 itask->itask_xfer_done_timestamp);
2882 }
2883
2884 return (*done_tstamp - *start_tstamp);
2885 }
2886
2887 static void
2888 stmf_update_kstat_lu_io(scsi_task_t *task, stmf_data_buf_t *dbuf)
2889 {
2890 stmf_i_lu_t *ilu;
2891 kstat_io_t *kip;
2892
2893 ilu = (stmf_i_lu_t *)task->task_lu->lu_stmf_private;
2894 if (ilu != NULL && ilu->ilu_kstat_io != NULL) {
2895 kip = KSTAT_IO_PTR(ilu->ilu_kstat_io);
2896 if (kip != NULL) {
2897 mutex_enter(ilu->ilu_kstat_io->ks_lock);
2898 STMF_UPDATE_KSTAT_IO(kip, dbuf);
2899 mutex_exit(ilu->ilu_kstat_io->ks_lock);
2900 }
2901 }
2902 }
2903
2904 static void
2905 stmf_create_kstat_lu(stmf_i_lu_t *ilu)
2906 {
2907 char ks_nm[KSTAT_STRLEN];
3528 alloc_len = sizeof (*irport) + sizeof (scsi_devid_desc_t) +
3529 rport_devid->ident_length - 1;
3530 irport = kmem_zalloc(alloc_len, KM_NOSLEEP);
3531 if (irport == NULL) {
3532 return (NULL);
3533 }
3534
3535 irport->irport_instance =
3536 id_alloc_nosleep(stmf_state.stmf_irport_inst_space);
3537 if (irport->irport_instance == -1) {
3538 kmem_free(irport, alloc_len);
3539 return (NULL);
3540 }
3541
3542 irport->irport_id =
3543 (struct scsi_devid_desc *)(irport + 1); /* Ptr. Arith. */
3544 bcopy(rport_devid, irport->irport_id,
3545 sizeof (scsi_devid_desc_t) + rport_devid->ident_length - 1);
3546 irport->irport_refcnt = 1;
3547 mutex_init(&irport->irport_mutex, NULL, MUTEX_DEFAULT, NULL);
3548 irport->irport_rdstart_timestamp = LLONG_MAX;
3549 irport->irport_wrstart_timestamp = LLONG_MAX;
3550
3551 return (irport);
3552 }
3553
3554 static void
3555 stmf_irport_destroy(stmf_i_remote_port_t *irport)
3556 {
3557 stmf_destroy_kstat_rport(irport);
3558 id_free(stmf_state.stmf_irport_inst_space, irport->irport_instance);
3559 mutex_destroy(&irport->irport_mutex);
3560 kmem_free(irport, sizeof (*irport) + sizeof (scsi_devid_desc_t) +
3561 irport->irport_id->ident_length - 1);
3562 }
3563
3564 static void
3565 stmf_create_kstat_rport(stmf_i_remote_port_t *irport)
3566 {
3567 scsi_devid_desc_t *id = irport->irport_id;
3568 char ks_nm[KSTAT_STRLEN];
3569 stmf_kstat_rport_info_t *ks_info;
3570 stmf_kstat_rport_estat_t *ks_estat;
3571 char *ident = NULL;
3572
3573 ks_info = kmem_zalloc(sizeof (*ks_info), KM_NOSLEEP);
3574 if (ks_info == NULL)
3575 goto err_out;
3576
3577 (void) snprintf(ks_nm, KSTAT_STRLEN, "stmf_rport_%"PRIxPTR"",
3578 (uintptr_t)irport);
3579 irport->irport_kstat_info = kstat_create(STMF_MODULE_NAME, 0,
3580 ks_nm, "misc", KSTAT_TYPE_NAMED,
3581 STMF_KSTAT_RPORT_DATAMAX - STMF_RPORT_INFO_LIMIT,
3582 KSTAT_FLAG_VIRTUAL | KSTAT_FLAG_VAR_SIZE);
3583 if (irport->irport_kstat_info == NULL) {
3584 kmem_free(ks_info, sizeof (*ks_info));
3585 goto err_out;
3586 }
3587
3588 irport->irport_kstat_info->ks_data = ks_info;
3589 irport->irport_kstat_info->ks_private = irport;
3590 irport->irport_kstat_info->ks_update = stmf_kstat_rport_update;
3591 ident = kmem_alloc(id->ident_length + 1, KM_NOSLEEP);
3592 if (ident == NULL) {
3593 kstat_delete(irport->irport_kstat_info);
3594 irport->irport_kstat_info = NULL;
3595 kmem_free(ks_info, sizeof (*ks_info));
3596 goto err_out;
3597 }
3598
3599 (void) memcpy(ident, id->ident, id->ident_length);
3600 ident[id->ident_length] = '\0';
3601 kstat_named_init(&ks_info->i_rport_name, "name", KSTAT_DATA_STRING);
3602 kstat_named_init(&ks_info->i_protocol, "protocol",
3603 KSTAT_DATA_STRING);
3604
3605 kstat_named_setstr(&ks_info->i_rport_name, ident);
3606 kstat_named_setstr(&ks_info->i_protocol,
3607 protocol_ident[irport->irport_id->protocol_id]);
3608 irport->irport_kstat_info->ks_lock = &irport->irport_mutex;
3609 irport->irport_info_dirty = B_TRUE;
3610 kstat_install(irport->irport_kstat_info);
3611
3612 (void) snprintf(ks_nm, KSTAT_STRLEN, "stmf_rport_io_%"PRIxPTR"",
3613 (uintptr_t)irport);
3614 irport->irport_kstat_io = kstat_create(STMF_MODULE_NAME, 0, ks_nm,
3615 "io", KSTAT_TYPE_IO, 1, 0);
3616 if (irport->irport_kstat_io == NULL)
3617 goto err_out;
3618
3619 irport->irport_kstat_io->ks_lock = &irport->irport_mutex;
3620 kstat_install(irport->irport_kstat_io);
3621
3622 (void) snprintf(ks_nm, KSTAT_STRLEN, "stmf_rport_st_%"PRIxPTR"",
3623 (uintptr_t)irport);
3624 irport->irport_kstat_estat = kstat_create(STMF_MODULE_NAME, 0, ks_nm,
3625 "misc", KSTAT_TYPE_NAMED,
3626 sizeof (*ks_estat) / sizeof (kstat_named_t), 0);
3627 if (irport->irport_kstat_estat == NULL)
3628 goto err_out;
3629
3630 ks_estat = (stmf_kstat_rport_estat_t *)KSTAT_NAMED_PTR(
3631 irport->irport_kstat_estat);
3632 kstat_named_init(&ks_estat->i_rport_read_latency,
3633 "rlatency", KSTAT_DATA_UINT64);
3634 kstat_named_init(&ks_estat->i_rport_write_latency,
3635 "wlatency", KSTAT_DATA_UINT64);
3636 kstat_named_init(&ks_estat->i_nread_tasks, "rntasks",
3637 KSTAT_DATA_UINT64);
3638 kstat_named_init(&ks_estat->i_nwrite_tasks, "wntasks",
3639 KSTAT_DATA_UINT64);
3640 irport->irport_kstat_estat->ks_lock = &irport->irport_mutex;
3641 kstat_install(irport->irport_kstat_estat);
3642
3643 return;
3644
3645 err_out:
3646 (void) memcpy(ks_nm, id->ident, MAX(KSTAT_STRLEN - 1,
3647 id->ident_length));
3648 ks_nm[id->ident_length] = '\0';
3649 cmn_err(CE_WARN, "STMF: remote port kstat creation failed: %s", ks_nm);
3650 }
3651
3652 static void
3653 stmf_destroy_kstat_rport(stmf_i_remote_port_t *irport)
3654 {
3655 if (irport->irport_kstat_io != NULL) {
3656 kstat_delete(irport->irport_kstat_io);
3657 }
3658 if (irport->irport_kstat_info != NULL) {
3659 stmf_kstat_rport_info_t *ks_info;
3660 kstat_named_t *knp;
3661 void *ptr;
3662 int i;
3663
3664 ks_info = (stmf_kstat_rport_info_t *)KSTAT_NAMED_PTR(
3665 irport->irport_kstat_info);
3666 kstat_delete(irport->irport_kstat_info);
3667 ptr = KSTAT_NAMED_STR_PTR(&ks_info->i_rport_name);
3668 kmem_free(ptr, KSTAT_NAMED_STR_BUFLEN(&ks_info->i_rport_name));
3669
3670 for (i = 0, knp = ks_info->i_rport_uinfo;
3671 i < STMF_RPORT_INFO_LIMIT; i++, knp++) {
3672 ptr = KSTAT_NAMED_STR_PTR(knp);
3673 if (ptr != NULL)
3674 kmem_free(ptr, KSTAT_NAMED_STR_BUFLEN(knp));
3675 }
3676 kmem_free(ks_info, sizeof (*ks_info));
3677 }
3678 }
3679
3680 static stmf_i_remote_port_t *
3681 stmf_irport_register(scsi_devid_desc_t *rport_devid)
3682 {
3683 stmf_i_remote_port_t *irport;
3684
3685 mutex_enter(&stmf_state.stmf_lock);
3686
3687 /*
3688 * Lookup will bump the refcnt if there's an existing rport
3689 * context for this identifier.
3690 */
3691 if ((irport = stmf_irport_lookup_locked(rport_devid)) != NULL) {
3692 mutex_exit(&stmf_state.stmf_lock);
3693 return (irport);
3694 }
3695
3696 irport = stmf_irport_create(rport_devid);
3697 if (irport == NULL) {
3698 mutex_exit(&stmf_state.stmf_lock);
3699 return (NULL);
3700 }
3701
3702 stmf_create_kstat_rport(irport);
3703 avl_add(&stmf_state.stmf_irportlist, irport);
3704 mutex_exit(&stmf_state.stmf_lock);
3705
3706 return (irport);
3707 }
3708
3709 static stmf_i_remote_port_t *
3710 stmf_irport_lookup_locked(scsi_devid_desc_t *rport_devid)
3711 {
3712 stmf_i_remote_port_t *irport;
3713 stmf_i_remote_port_t tmp_irport;
3714
3715 ASSERT(mutex_owned(&stmf_state.stmf_lock));
3716 tmp_irport.irport_id = rport_devid;
3717 irport = avl_find(&stmf_state.stmf_irportlist, &tmp_irport, NULL);
3718 if (irport != NULL) {
3719 mutex_enter(&irport->irport_mutex);
3720 irport->irport_refcnt++;
3721 mutex_exit(&irport->irport_mutex);
3722 }
3806
3807 mutex_enter(&stmf_state.stmf_lock);
3808 rw_enter(&ilport->ilport_lock, RW_WRITER);
3809 (void) stmf_session_create_lun_map(ilport, iss);
3810 ilport->ilport_nsessions++;
3811 iss->iss_next = ilport->ilport_ss_list;
3812 ilport->ilport_ss_list = iss;
3813 rw_exit(&ilport->ilport_lock);
3814 mutex_exit(&stmf_state.stmf_lock);
3815
3816 iss->iss_creation_time = ddi_get_time();
3817 ss->ss_session_id = atomic_inc_64_nv(&stmf_session_counter);
3818 iss->iss_flags &= ~ISS_BEING_CREATED;
3819 /* XXX should we remove ISS_LUN_INVENTORY_CHANGED on new session? */
3820 iss->iss_flags &= ~ISS_LUN_INVENTORY_CHANGED;
3821 DTRACE_PROBE2(session__online, stmf_local_port_t *, lport,
3822 stmf_scsi_session_t *, ss);
3823 return (STMF_SUCCESS);
3824 }
3825
3826 stmf_status_t
3827 stmf_add_rport_info(stmf_scsi_session_t *ss,
3828 const char *prop_name, const char *prop_value)
3829 {
3830 stmf_i_scsi_session_t *iss = ss->ss_stmf_private;
3831 stmf_i_remote_port_t *irport = iss->iss_irport;
3832 kstat_named_t *knp;
3833 char *s;
3834 int i;
3835
3836 s = strdup(prop_value);
3837
3838 mutex_enter(irport->irport_kstat_info->ks_lock);
3839 /* Make sure the caller doesn't try to add already existing property */
3840 knp = KSTAT_NAMED_PTR(irport->irport_kstat_info);
3841 for (i = 0; i < STMF_KSTAT_RPORT_DATAMAX; i++, knp++) {
3842 if (KSTAT_NAMED_STR_PTR(knp) == NULL)
3843 break;
3844
3845 ASSERT(strcmp(knp->name, prop_name) != 0);
3846 }
3847
3848 if (i == STMF_KSTAT_RPORT_DATAMAX) {
3849 mutex_exit(irport->irport_kstat_info->ks_lock);
3850 kmem_free(s, strlen(s) + 1);
3851 return (STMF_FAILURE);
3852 }
3853
3854 irport->irport_info_dirty = B_TRUE;
3855 kstat_named_init(knp, prop_name, KSTAT_DATA_STRING);
3856 kstat_named_setstr(knp, s);
3857 mutex_exit(irport->irport_kstat_info->ks_lock);
3858
3859 return (STMF_SUCCESS);
3860 }
3861
3862 void
3863 stmf_remove_rport_info(stmf_scsi_session_t *ss,
3864 const char *prop_name)
3865 {
3866 stmf_i_scsi_session_t *iss = ss->ss_stmf_private;
3867 stmf_i_remote_port_t *irport = iss->iss_irport;
3868 kstat_named_t *knp;
3869 char *s;
3870 int i;
3871 uint32_t len;
3872
3873 mutex_enter(irport->irport_kstat_info->ks_lock);
3874 knp = KSTAT_NAMED_PTR(irport->irport_kstat_info);
3875 for (i = 0; i < STMF_KSTAT_RPORT_DATAMAX; i++, knp++) {
3876 if ((knp->name != NULL) && (strcmp(knp->name, prop_name) == 0))
3877 break;
3878 }
3879
3880 if (i == STMF_KSTAT_RPORT_DATAMAX) {
3881 mutex_exit(irport->irport_kstat_info->ks_lock);
3882 return;
3883 }
3884
3885 s = KSTAT_NAMED_STR_PTR(knp);
3886 len = KSTAT_NAMED_STR_BUFLEN(knp);
3887
3888 for (; i < STMF_KSTAT_RPORT_DATAMAX - 1; i++, knp++) {
3889 kstat_named_init(knp, knp[1].name, KSTAT_DATA_STRING);
3890 kstat_named_setstr(knp, KSTAT_NAMED_STR_PTR(&knp[1]));
3891 }
3892 kstat_named_init(knp, "", KSTAT_DATA_STRING);
3893
3894 irport->irport_info_dirty = B_TRUE;
3895 mutex_exit(irport->irport_kstat_info->ks_lock);
3896 kmem_free(s, len);
3897 }
3898
3899 static int
3900 stmf_kstat_rport_update(kstat_t *ksp, int rw)
3901 {
3902 stmf_i_remote_port_t *irport = ksp->ks_private;
3903 kstat_named_t *knp;
3904 uint_t ndata = 0;
3905 size_t dsize = 0;
3906 int i;
3907
3908 if (rw == KSTAT_WRITE)
3909 return (EACCES);
3910
3911 if (!irport->irport_info_dirty)
3912 return (0);
3913
3914 knp = KSTAT_NAMED_PTR(ksp);
3915 for (i = 0; i < STMF_KSTAT_RPORT_DATAMAX; i++, knp++) {
3916 if (KSTAT_NAMED_STR_PTR(knp) == NULL)
3917 break;
3918 ndata++;
3919 dsize += KSTAT_NAMED_STR_BUFLEN(knp);
3920 }
3921
3922 ksp->ks_ndata = ndata;
3923 ksp->ks_data_size = sizeof (kstat_named_t) * ndata + dsize;
3924 irport->irport_info_dirty = B_FALSE;
3925
3926 return (0);
3927 }
3928
3929 void
3930 stmf_deregister_scsi_session(stmf_local_port_t *lport, stmf_scsi_session_t *ss)
3931 {
3932 stmf_i_local_port_t *ilport = (stmf_i_local_port_t *)
3933 lport->lport_stmf_private;
3934 stmf_i_scsi_session_t *iss, **ppss;
3935 int found = 0;
3936 stmf_ic_msg_t *ic_session_dereg;
3937 stmf_status_t ic_ret = STMF_FAILURE;
3938 stmf_lun_map_t *sm;
3939 stmf_i_lu_t *ilu;
3940 uint16_t n;
3941 stmf_lun_map_ent_t *ent;
3942
3943 DTRACE_PROBE2(session__offline, stmf_local_port_t *, lport,
3944 stmf_scsi_session_t *, ss);
3945
3946 iss = (stmf_i_scsi_session_t *)ss->ss_stmf_private;
3947 if (ss->ss_rport_alias) {
3948 ss->ss_rport_alias = NULL;
3949 }
3950
3951 try_dereg_ss_again:
3952 mutex_enter(&stmf_state.stmf_lock);
3953 atomic_and_32(&iss->iss_flags,
3954 ~(ISS_LUN_INVENTORY_CHANGED | ISS_GOT_INITIAL_LUNS));
3955 if (iss->iss_flags & ISS_EVENT_ACTIVE) {
3956 mutex_exit(&stmf_state.stmf_lock);
3957 delay(1);
3958 goto try_dereg_ss_again;
3959 }
3960
3961 /* dereg proxy session if not standby port */
3971 }
3972 }
3973 }
3974
3975 rw_enter(&ilport->ilport_lock, RW_WRITER);
3976 for (ppss = &ilport->ilport_ss_list; *ppss != NULL;
3977 ppss = &((*ppss)->iss_next)) {
3978 if (iss == (*ppss)) {
3979 *ppss = (*ppss)->iss_next;
3980 found = 1;
3981 break;
3982 }
3983 }
3984 if (!found) {
3985 cmn_err(CE_PANIC, "Deregister session called for non existent"
3986 " session");
3987 }
3988 ilport->ilport_nsessions--;
3989
3990 stmf_irport_deregister(iss->iss_irport);
3991 /*
3992 * to avoid conflict with updating session's map,
3993 * which only grab stmf_lock
3994 */
3995 sm = iss->iss_sm;
3996 iss->iss_sm = NULL;
3997 iss->iss_hg = NULL;
3998
3999 rw_exit(&ilport->ilport_lock);
4000
4001 if (sm->lm_nentries) {
4002 for (n = 0; n < sm->lm_nentries; n++) {
4003 if ((ent = (stmf_lun_map_ent_t *)sm->lm_plus[n])
4004 != NULL) {
4005 if (ent->ent_itl_datap) {
4006 stmf_do_itl_dereg(ent->ent_lu,
4007 ent->ent_itl_datap,
4008 STMF_ITL_REASON_IT_NEXUS_LOSS);
4009 }
4010 ilu = (stmf_i_lu_t *)
4011 ent->ent_lu->lu_stmf_private;
4012 atomic_dec_32(&ilu->ilu_ref_cnt);
4013 kmem_free(sm->lm_plus[n],
4014 sizeof (stmf_lun_map_ent_t));
4015 }
4016 }
4017 kmem_free(sm->lm_plus,
4018 sizeof (stmf_lun_map_ent_t *) * sm->lm_nentries);
4019 }
4020 kmem_free(sm, sizeof (*sm));
4021
4022 if (iss->iss_flags & ISS_NULL_TPTID) {
4023 stmf_remote_port_free(ss->ss_rport);
4024 }
4025
4026 mutex_exit(&stmf_state.stmf_lock);
4027 }
4028
4029
4030
4031 stmf_i_scsi_session_t *
4032 stmf_session_id_to_issptr(uint64_t session_id, int stay_locked)
4033 {
4034 stmf_i_local_port_t *ilport;
4035 stmf_i_scsi_session_t *iss;
4036
4037 mutex_enter(&stmf_state.stmf_lock);
4038 for (ilport = stmf_state.stmf_ilportlist; ilport != NULL;
4039 ilport = ilport->ilport_next) {
4040 rw_enter(&ilport->ilport_lock, RW_WRITER);
4041 for (iss = ilport->ilport_ss_list; iss != NULL;
4042 iss = iss->iss_next) {
4043 if (iss->iss_ss->ss_session_id == session_id) {
4044 if (!stay_locked)
4045 rw_exit(&ilport->ilport_lock);
4046 mutex_exit(&stmf_state.stmf_lock);
4047 return (iss);
4048 }
4049 }
4050 rw_exit(&ilport->ilport_lock);
4206 goto dai_scan_done;
4207 }
4208 }
4209 } /* lun table for a session */
4210 } /* sessions */
4211 rw_exit(&ilport->ilport_lock);
4212 } /* ports */
4213
4214 dai_scan_done:
4215 mutex_exit(&stmf_state.stmf_lock);
4216
4217 for (i = 0; i < nu; i++) {
4218 stmf_do_itl_dereg(lu, itl_list[i],
4219 STMF_ITL_REASON_DEREG_REQUEST);
4220 }
4221 kmem_free(itl_list, nmaps * sizeof (stmf_itl_data_t *));
4222
4223 return (STMF_SUCCESS);
4224 }
4225
4226 stmf_data_buf_t *
4227 stmf_alloc_dbuf(scsi_task_t *task, uint32_t size, uint32_t *pminsize,
4228 uint32_t flags)
4229 {
4230 stmf_i_scsi_task_t *itask =
4231 (stmf_i_scsi_task_t *)task->task_stmf_private;
4232 stmf_local_port_t *lport = task->task_lport;
4233 stmf_data_buf_t *dbuf;
4234 uint8_t ndx;
4235
4236 ndx = stmf_first_zero[itask->itask_allocated_buf_map];
4237 if (ndx == 0xff)
4238 return (NULL);
4239 dbuf = itask->itask_dbufs[ndx] = lport->lport_ds->ds_alloc_data_buf(
4240 task, size, pminsize, flags);
4241 if (dbuf) {
4242 task->task_cur_nbufs++;
4243 itask->itask_allocated_buf_map |= (1 << ndx);
4244 dbuf->db_flags &= ~DB_LPORT_XFER_ACTIVE;
4245 dbuf->db_handle = ndx;
4339
4340 /*
4341 * We allocate 7 extra bytes for CDB to provide a cdb pointer which
4342 * is guaranteed to be 8 byte aligned. Some LU providers like OSD
4343 * depend upon this alignment.
4344 */
4345 if (cdb_length_in >= 16)
4346 cdb_length = cdb_length_in + 7;
4347 else
4348 cdb_length = 16 + 7;
4349 iss = (stmf_i_scsi_session_t *)ss->ss_stmf_private;
4350 luNbr = ((uint16_t)lun[1] | (((uint16_t)(lun[0] & 0x3F)) << 8));
4351 rw_enter(iss->iss_lockp, RW_READER);
4352 lun_map_ent =
4353 (stmf_lun_map_ent_t *)stmf_get_ent_from_map(iss->iss_sm, luNbr);
4354 if (!lun_map_ent) {
4355 lu = dlun0;
4356 } else {
4357 lu = lun_map_ent->ent_lu;
4358 }
4359
4360 ilu = lu->lu_stmf_private;
4361 if (ilu->ilu_flags & ILU_RESET_ACTIVE) {
4362 rw_exit(iss->iss_lockp);
4363 return (NULL);
4364 }
4365
4366 /*
4367 * if the LUN is being offlined or is offline then only command
4368 * that are to query the LUN are allowed. These are handled in
4369 * stmf via the dlun0 vector. It is possible that a race condition
4370 * will cause other commands to arrive while the lun is in the
4371 * process of being offlined. Check for those and just let the
4372 * protocol stack handle the error.
4373 */
4374 if ((ilu->ilu_state == STMF_STATE_OFFLINING) ||
4375 (ilu->ilu_state == STMF_STATE_OFFLINE)) {
4376 if (lu != dlun0) {
4377 rw_exit(iss->iss_lockp);
4378 return (NULL);
4379 }
4380 }
4381
4382 do {
4383 if (ilu->ilu_free_tasks == NULL) {
4384 new_task = 1;
4385 break;
4386 }
4387 mutex_enter(&ilu->ilu_task_lock);
4388 for (ppitask = &ilu->ilu_free_tasks; (*ppitask != NULL) &&
4389 ((*ppitask)->itask_cdb_buf_size < cdb_length);
4390 ppitask = &((*ppitask)->itask_lu_free_next))
4391 ;
4392 if (*ppitask) {
4393 itask = *ppitask;
4394 *ppitask = (*ppitask)->itask_lu_free_next;
4395 ilu->ilu_ntasks_free--;
4396 if (ilu->ilu_ntasks_free < ilu->ilu_ntasks_min_free)
4397 ilu->ilu_ntasks_min_free = ilu->ilu_ntasks_free;
4398 } else {
4399 new_task = 1;
4400 }
4401 mutex_exit(&ilu->ilu_task_lock);
4409 * selection process above.
4410 */
4411 uint8_t *save_cdb;
4412 uintptr_t t_start, t_end;
4413
4414 task = itask->itask_task;
4415 save_cdb = task->task_cdb; /* save */
4416 t_start = (uintptr_t)&task->task_flags;
4417 t_end = (uintptr_t)&task->task_extended_cmd;
4418 bzero((void *)t_start, (size_t)(t_end - t_start));
4419 task->task_cdb = save_cdb; /* restore */
4420 itask->itask_ncmds = 0;
4421 } else {
4422 task = (scsi_task_t *)stmf_alloc(STMF_STRUCT_SCSI_TASK,
4423 cdb_length, AF_FORCE_NOSLEEP);
4424 if (task == NULL) {
4425 rw_exit(iss->iss_lockp);
4426 return (NULL);
4427 }
4428 task->task_lu = lu;
4429 task->task_cdb = (uint8_t *)task->task_port_private;
4430 if ((ulong_t)(task->task_cdb) & 7ul) {
4431 task->task_cdb = (uint8_t *)(((ulong_t)
4432 (task->task_cdb) + 7ul) & ~(7ul));
4433 }
4434 itask = (stmf_i_scsi_task_t *)task->task_stmf_private;
4435 itask->itask_cdb_buf_size = cdb_length;
4436 mutex_init(&itask->itask_audit_mutex, NULL, MUTEX_DRIVER, NULL);
4437 mutex_init(&itask->itask_mutex, NULL, MUTEX_DRIVER, NULL);
4438 }
4439
4440 /*
4441 * Since a LUN can be mapped as different LUN ids to different initiator
4442 * groups, we need to set LUN id for a new task and reset LUN id for
4443 * a reused task.
4444 */
4445 l = task->task_lun_no;
4446 l[0] = lun[0];
4447 l[1] = lun[1];
4448 l[2] = lun[2];
4449 l[3] = lun[3];
4450 l[4] = lun[4];
4451 l[5] = lun[5];
4452 l[6] = lun[6];
4453 l[7] = lun[7];
4454
4455 mutex_enter(&itask->itask_mutex);
4456 task->task_session = ss;
4457 task->task_lport = lport;
4458 task->task_cdb_length = cdb_length_in;
4459 itask->itask_flags = ITASK_IN_TRANSITION;
4460 itask->itask_waitq_time = 0;
4461 itask->itask_lu_read_time = itask->itask_lu_write_time = 0;
4462 itask->itask_lport_read_time = itask->itask_lport_write_time = 0;
4463 itask->itask_read_xfer = itask->itask_write_xfer = 0;
4464 itask->itask_audit_index = 0;
4465 bzero(&itask->itask_audit_records[0],
4466 sizeof (stmf_task_audit_rec_t) * ITASK_TASK_AUDIT_DEPTH);
4467 mutex_exit(&itask->itask_mutex);
4468
4469 if (new_task) {
4470 if (lu->lu_task_alloc(task) != STMF_SUCCESS) {
4471 rw_exit(iss->iss_lockp);
4472 stmf_free(task);
4473 return (NULL);
4474 }
4475 mutex_enter(&ilu->ilu_task_lock);
4476 if (ilu->ilu_flags & ILU_RESET_ACTIVE) {
4477 mutex_exit(&ilu->ilu_task_lock);
4478 rw_exit(iss->iss_lockp);
4479 stmf_free(task);
4480 return (NULL);
4481 }
4482 itask->itask_lu_next = ilu->ilu_tasks;
4483 if (ilu->ilu_tasks)
4484 ilu->ilu_tasks->itask_lu_prev = itask;
4485 ilu->ilu_tasks = itask;
4486 /* kmem_zalloc automatically makes itask->itask_lu_prev NULL */
4487 ilu->ilu_ntasks++;
4488 mutex_exit(&ilu->ilu_task_lock);
4489 }
4490
4491 itask->itask_ilu_task_cntr = ilu->ilu_cur_task_cntr;
4492 atomic_inc_32(itask->itask_ilu_task_cntr);
4493 itask->itask_start_time = ddi_get_lbolt();
4494
4495 if ((lun_map_ent != NULL) && ((itask->itask_itl_datap =
4496 lun_map_ent->ent_itl_datap) != NULL)) {
4497 atomic_inc_32(&itask->itask_itl_datap->itl_counter);
4498 task->task_lu_itl_handle = itask->itask_itl_datap->itl_handle;
4499 } else {
4500 itask->itask_itl_datap = NULL;
4501 task->task_lu_itl_handle = NULL;
4502 }
4503
4504 rw_exit(iss->iss_lockp);
4505 return (task);
4506 }
4507
4508 /* ARGSUSED */
4509 static void
4510 stmf_task_lu_free(scsi_task_t *task, stmf_i_scsi_session_t *iss)
4511 {
4512 stmf_i_scsi_task_t *itask =
4513 (stmf_i_scsi_task_t *)task->task_stmf_private;
4514 stmf_i_lu_t *ilu = (stmf_i_lu_t *)task->task_lu->lu_stmf_private;
4515
4516 ASSERT(rw_lock_held(iss->iss_lockp));
4517 ASSERT((itask->itask_flags & ITASK_IN_FREE_LIST) == 0);
4518 ASSERT((itask->itask_flags & ITASK_IN_WORKER_QUEUE) == 0);
4519 ASSERT((itask->itask_flags & ITASK_IN_TRANSITION) == 0);
4520 ASSERT((itask->itask_flags & ITASK_KNOWN_TO_LU) == 0);
4521 ASSERT(mutex_owned(&itask->itask_mutex));
4522
4523 itask->itask_flags = ITASK_IN_FREE_LIST;
4524 itask->itask_ncmds = 0;
4525 itask->itask_proxy_msg_id = 0;
4526 atomic_dec_32(itask->itask_ilu_task_cntr);
4527 itask->itask_worker_next = NULL;
4528 mutex_exit(&itask->itask_mutex);
4529
4530 mutex_enter(&ilu->ilu_task_lock);
4531 itask->itask_lu_free_next = ilu->ilu_free_tasks;
4532 ilu->ilu_free_tasks = itask;
4533 ilu->ilu_ntasks_free++;
4534 if (ilu->ilu_ntasks == ilu->ilu_ntasks_free)
4535 cv_signal(&ilu->ilu_offline_pending_cv);
4536 mutex_exit(&ilu->ilu_task_lock);
4537 }
4538
4539 void
4540 stmf_task_lu_check_freelist(stmf_i_lu_t *ilu)
4541 {
4542 uint32_t num_to_release, ndx;
4543 stmf_i_scsi_task_t *itask;
4544 stmf_lu_t *lu = ilu->ilu_lu;
4545
4546 ASSERT(ilu->ilu_ntasks_min_free <= ilu->ilu_ntasks_free);
4547
4548 /* free half of the minimal free of the free tasks */
4549 num_to_release = (ilu->ilu_ntasks_min_free + 1) / 2;
4550 if (!num_to_release) {
4551 return;
4552 }
4553 for (ndx = 0; ndx < num_to_release; ndx++) {
4554 mutex_enter(&ilu->ilu_task_lock);
4555 itask = ilu->ilu_free_tasks;
4556 if (itask == NULL) {
4593 if (!ilu->ilu_ntasks_min_free) {
4594 ilu->ilu_ntasks_min_free = ilu->ilu_ntasks_free;
4595 continue;
4596 }
4597 ilu->ilu_flags |= ILU_STALL_DEREGISTER;
4598 mutex_exit(&stmf_state.stmf_lock);
4599 stmf_task_lu_check_freelist(ilu);
4600 /*
4601 * we do not care about the accuracy of
4602 * ilu_ntasks_min_free, so we don't lock here
4603 */
4604 ilu->ilu_ntasks_min_free = ilu->ilu_ntasks_free;
4605 mutex_enter(&stmf_state.stmf_lock);
4606 ilu->ilu_flags &= ~ILU_STALL_DEREGISTER;
4607 cv_broadcast(&stmf_state.stmf_cv);
4608 if (ddi_get_lbolt() >= endtime)
4609 break;
4610 }
4611 }
4612
4613 /*
4614 * Since this method is looking to find tasks that are stuck, lost, or senile
4615 * it should be more willing to give up scaning during this time period. This
4616 * is why mutex_tryenter is now used instead of the standard mutex_enter.
4617 * There has been at least one case were the following occurred.
4618 *
4619 * 1) The iscsit_deferred() method is trying to register a session and
4620 * needs the global lock which is held.
4621 * 2) Another thread which holds the global lock is trying to deregister a
4622 * session and needs the session lock.
4623 * 3) A third thread is allocating a stmf task that has grabbed the session
4624 * lock and is trying to grab the lun task lock.
4625 * 4) There's a timeout thread that has the lun task lock and is trying to grab
4626 * a specific task lock.
4627 * 5) The thread that has the task lock is waiting for the ref count to go to
4628 * zero.
4629 * 6) There's a task that would drop the count to zero, but it's in the task
4630 * queue waiting to run and is stuck because of #1 is currently block.
4631 *
4632 * This method is number 4 in the above chain of events. Had this code
4633 * originally used mutex_tryenter the chain would have been broken and the
4634 * system wouldn't have hung. So, now this method uses mutex_tryenter and
4635 * you know why it does so.
4636 */
4637 /* ---- Only one thread calls stmf_do_ilu_timeouts so no lock required ---- */
4638 typedef struct stmf_bailout_cnt {
4639 int no_ilu_lock;
4640 int no_task_lock;
4641 int tasks_checked;
4642 } stmf_bailout_cnt_t;
4643
4644 stmf_bailout_cnt_t stmf_bailout;
4645
4646 static void
4647 stmf_do_ilu_timeouts(stmf_i_lu_t *ilu)
4648 {
4649 clock_t l = ddi_get_lbolt();
4650 clock_t ps = drv_usectohz(1000000);
4651 stmf_i_scsi_task_t *itask;
4652 scsi_task_t *task;
4653 uint32_t to;
4654
4655 if (mutex_tryenter(&ilu->ilu_task_lock) == 0) {
4656 stmf_bailout.no_ilu_lock++;
4657 return;
4658 }
4659
4660 for (itask = ilu->ilu_tasks; itask != NULL;
4661 itask = itask->itask_lu_next) {
4662 if (mutex_tryenter(&itask->itask_mutex) == 0) {
4663 stmf_bailout.no_task_lock++;
4664 continue;
4665 }
4666 stmf_bailout.tasks_checked++;
4667 if (itask->itask_flags & (ITASK_IN_FREE_LIST |
4668 ITASK_BEING_ABORTED)) {
4669 mutex_exit(&itask->itask_mutex);
4670 continue;
4671 }
4672 task = itask->itask_task;
4673 if (task->task_timeout == 0)
4674 to = stmf_default_task_timeout;
4675 else
4676 to = task->task_timeout;
4677
4678 if ((itask->itask_start_time + (to * ps)) > l) {
4679 mutex_exit(&itask->itask_mutex);
4680 continue;
4681 }
4682 mutex_exit(&itask->itask_mutex);
4683 stmf_abort(STMF_QUEUE_TASK_ABORT, task,
4684 STMF_TIMEOUT, NULL);
4685 }
4686 mutex_exit(&ilu->ilu_task_lock);
4687 }
4688
4689 /*
4690 * Called with stmf_lock held
4691 */
4692 void
4693 stmf_check_ilu_timing()
4694 {
4695 stmf_i_lu_t *ilu;
4696 clock_t endtime = ddi_get_lbolt() + drv_usectohz(10000);
4697
4698 /* stmf_svc_ilu_timing may get changed after stmf_lock is released */
4699 while ((ilu = stmf_state.stmf_svc_ilu_timing) != NULL) {
4700 stmf_state.stmf_svc_ilu_timing = ilu->ilu_next;
4701 if (ilu->ilu_cur_task_cntr == (&ilu->ilu_task_cntr1)) {
4702 if (ilu->ilu_task_cntr2 == 0) {
4717 mutex_exit(&stmf_state.stmf_lock);
4718 stmf_do_ilu_timeouts(ilu);
4719 mutex_enter(&stmf_state.stmf_lock);
4720 ilu->ilu_flags &= ~ILU_STALL_DEREGISTER;
4721 cv_broadcast(&stmf_state.stmf_cv);
4722 if (ddi_get_lbolt() >= endtime)
4723 break;
4724 }
4725 }
4726
4727 /*
4728 * Kills all tasks on a lu except tm_task
4729 */
4730 void
4731 stmf_task_lu_killall(stmf_lu_t *lu, scsi_task_t *tm_task, stmf_status_t s)
4732 {
4733 stmf_i_lu_t *ilu = (stmf_i_lu_t *)lu->lu_stmf_private;
4734 stmf_i_scsi_task_t *itask;
4735
4736 mutex_enter(&ilu->ilu_task_lock);
4737 for (itask = ilu->ilu_tasks; itask != NULL;
4738 itask = itask->itask_lu_next) {
4739 mutex_enter(&itask->itask_mutex);
4740 if (itask->itask_flags & ITASK_IN_FREE_LIST) {
4741 mutex_exit(&itask->itask_mutex);
4742 continue;
4743 }
4744 mutex_exit(&itask->itask_mutex);
4745 if (itask->itask_task == tm_task)
4746 continue;
4747 stmf_abort(STMF_QUEUE_TASK_ABORT, itask->itask_task, s, NULL);
4748 }
4749 mutex_exit(&ilu->ilu_task_lock);
4750 }
4751
4752 void
4753 stmf_free_task_bufs(stmf_i_scsi_task_t *itask, stmf_local_port_t *lport)
4754 {
4755 int i;
4756 uint8_t map;
4757
4758 if ((map = itask->itask_allocated_buf_map) == 0)
4759 return;
4760 for (i = 0; i < 4; i++) {
4761 if (map & 1) {
4762 stmf_data_buf_t *dbuf;
4763
4764 dbuf = itask->itask_dbufs[i];
4780 } else {
4781 ASSERT(dbuf->db_lu_private == NULL);
4782 dbuf->db_lu_private = NULL;
4783 lport->lport_ds->ds_free_data_buf(
4784 lport->lport_ds, dbuf);
4785 }
4786 }
4787 map >>= 1;
4788 }
4789 itask->itask_allocated_buf_map = 0;
4790 }
4791
4792 void
4793 stmf_task_free(scsi_task_t *task)
4794 {
4795 stmf_local_port_t *lport = task->task_lport;
4796 stmf_i_scsi_task_t *itask = (stmf_i_scsi_task_t *)
4797 task->task_stmf_private;
4798 stmf_i_scsi_session_t *iss = (stmf_i_scsi_session_t *)
4799 task->task_session->ss_stmf_private;
4800 stmf_lu_t *lu = task->task_lu;
4801
4802 stmf_task_audit(itask, TE_TASK_FREE, CMD_OR_IOF_NA, NULL);
4803 ASSERT(mutex_owned(&itask->itask_mutex));
4804 if ((lu != NULL) && (lu->lu_task_done != NULL))
4805 lu->lu_task_done(task);
4806 stmf_free_task_bufs(itask, lport);
4807 stmf_itl_task_done(itask);
4808 DTRACE_PROBE2(stmf__task__end, scsi_task_t *, task,
4809 hrtime_t,
4810 itask->itask_done_timestamp - itask->itask_start_timestamp);
4811 if (itask->itask_itl_datap) {
4812 if (atomic_dec_32_nv(&itask->itask_itl_datap->itl_counter) ==
4813 0) {
4814 stmf_release_itl_handle(task->task_lu,
4815 itask->itask_itl_datap);
4816 }
4817 }
4818
4819 /*
4820 * To prevent a deadlock condition must release the itask_mutex,
4821 * grab a reader lock on iss_lockp and then reacquire the itask_mutex.
4822 */
4823 mutex_exit(&itask->itask_mutex);
4824 rw_enter(iss->iss_lockp, RW_READER);
4825 mutex_enter(&itask->itask_mutex);
4826
4827 lport->lport_task_free(task);
4828 if (itask->itask_worker) {
4829 atomic_dec_32(&stmf_cur_ntasks);
4830 atomic_dec_32(&itask->itask_worker->worker_ref_count);
4831 }
4832 /*
4833 * After calling stmf_task_lu_free, the task pointer can no longer
4834 * be trusted.
4835 */
4836 stmf_task_lu_free(task, iss);
4837 rw_exit(iss->iss_lockp);
4838 }
4839
4840 void
4841 stmf_post_task(scsi_task_t *task, stmf_data_buf_t *dbuf)
4842 {
4843 stmf_i_scsi_task_t *itask = (stmf_i_scsi_task_t *)
4844 task->task_stmf_private;
4845 stmf_i_lu_t *ilu = (stmf_i_lu_t *)task->task_lu->lu_stmf_private;
4846 int nv;
4847 uint32_t new;
4848 uint32_t ct;
4849 stmf_worker_t *w;
4850 uint8_t tm;
4851
4852 if (task->task_max_nbufs > 4)
4853 task->task_max_nbufs = 4;
4854 task->task_cur_nbufs = 0;
4855 /* Latest value of currently running tasks */
4856 ct = atomic_inc_32_nv(&stmf_cur_ntasks);
4857
4858 /* Select the next worker using round robin */
4859 mutex_enter(&stmf_worker_sel_mx);
4860 stmf_worker_sel_counter++;
4861 if (stmf_worker_sel_counter >= stmf_nworkers)
4862 stmf_worker_sel_counter = 0;
4863 nv = stmf_worker_sel_counter;
4864
4865 /* if the selected worker is not idle then bump to the next worker */
4866 if (stmf_workers[nv].worker_queue_depth > 0) {
4867 stmf_worker_sel_counter++;
4868 if (stmf_worker_sel_counter >= stmf_nworkers)
4869 stmf_worker_sel_counter = 0;
4870 nv = stmf_worker_sel_counter;
4871 }
4872 mutex_exit(&stmf_worker_sel_mx);
4873
4874 w = &stmf_workers[nv];
4875
4876 mutex_enter(&itask->itask_mutex);
4877 mutex_enter(&w->worker_lock);
4878
4879 itask->itask_worker = w;
4880
4881 /*
4882 * Track max system load inside the worker as we already have the
4883 * worker lock (no point implementing another lock). The service
4884 * thread will do the comparisons and figure out the max overall
4885 * system load.
4886 */
4887 if (w->worker_max_sys_qdepth_pu < ct)
4888 w->worker_max_sys_qdepth_pu = ct;
4889
4890 new = itask->itask_flags;
4891 new |= ITASK_KNOWN_TO_TGT_PORT;
4892 if (task->task_mgmt_function) {
4893 tm = task->task_mgmt_function;
4894 if ((tm == TM_TARGET_RESET) ||
4895 (tm == TM_TARGET_COLD_RESET) ||
4896 (tm == TM_TARGET_WARM_RESET)) {
4897 new |= ITASK_DEFAULT_HANDLING;
4898 }
4899 } else if (task->task_cdb[0] == SCMD_REPORT_LUNS) {
4900 new |= ITASK_DEFAULT_HANDLING;
4901 }
4902 new &= ~ITASK_IN_TRANSITION;
4903 itask->itask_flags = new;
4904
4905 stmf_itl_task_start(itask);
4906
4907 itask->itask_cmd_stack[0] = ITASK_CMD_NEW_TASK;
4908 itask->itask_ncmds = 1;
4909
4910 if ((task->task_flags & TF_INITIAL_BURST) &&
4911 !(curthread->t_flag & T_INTR_THREAD)) {
4912 stmf_update_kstat_lu_io(task, dbuf);
4913 stmf_update_kstat_lport_io(task, dbuf);
4914 stmf_update_kstat_rport_io(task, dbuf);
4915 }
4916
4917 stmf_task_audit(itask, TE_TASK_START, CMD_OR_IOF_NA, dbuf);
4918 if (dbuf) {
4919 itask->itask_allocated_buf_map = 1;
4920 itask->itask_dbufs[0] = dbuf;
4921 dbuf->db_handle = 0;
4922 } else {
4923 itask->itask_allocated_buf_map = 0;
4924 itask->itask_dbufs[0] = NULL;
4925 }
4926
4927 STMF_ENQUEUE_ITASK(w, itask);
4928
4929 mutex_exit(&w->worker_lock);
4930 mutex_exit(&itask->itask_mutex);
4931
4932 /*
4933 * This can only happen if during stmf_task_alloc(), ILU_RESET_ACTIVE
4934 * was set between checking of ILU_RESET_ACTIVE and clearing of the
4935 * ITASK_IN_FREE_LIST flag. Take care of these "sneaked-in" tasks here.
4936 */
4937 if (ilu->ilu_flags & ILU_RESET_ACTIVE) {
4938 stmf_abort(STMF_QUEUE_TASK_ABORT, task, STMF_ABORTED, NULL);
4939 }
4940 }
4941
4942 static void
4943 stmf_task_audit(stmf_i_scsi_task_t *itask,
4944 task_audit_event_t te, uint32_t cmd_or_iof, stmf_data_buf_t *dbuf)
4945 {
4946 stmf_task_audit_rec_t *ar;
4947
4948 mutex_enter(&itask->itask_audit_mutex);
4949 ar = &itask->itask_audit_records[itask->itask_audit_index++];
4950 itask->itask_audit_index &= (ITASK_TASK_AUDIT_DEPTH - 1);
4966 * we will only call that entry point if ITASK_KNOWN_TO_LU was set.
4967 *
4968 * Same logic applies for the port.
4969 *
4970 * Also ITASK_BEING_ABORTED will not be allowed to set if both KNOWN_TO_LU
4971 * and KNOWN_TO_TGT_PORT are reset.
4972 *
4973 * +++++++++++++++++++++++++++++++++++++++++++++++
4974 */
4975
4976 stmf_status_t
4977 stmf_xfer_data(scsi_task_t *task, stmf_data_buf_t *dbuf, uint32_t ioflags)
4978 {
4979 stmf_status_t ret = STMF_SUCCESS;
4980
4981 stmf_i_scsi_task_t *itask =
4982 (stmf_i_scsi_task_t *)task->task_stmf_private;
4983
4984 stmf_task_audit(itask, TE_XFER_START, ioflags, dbuf);
4985
4986 mutex_enter(&itask->itask_mutex);
4987 if (ioflags & STMF_IOF_LU_DONE) {
4988 if (itask->itask_flags & ITASK_BEING_ABORTED) {
4989 mutex_exit(&itask->itask_mutex);
4990 return (STMF_ABORTED);
4991 }
4992 itask->itask_flags &= ~ITASK_KNOWN_TO_LU;
4993 }
4994 if ((itask->itask_flags & ITASK_BEING_ABORTED) != 0) {
4995 mutex_exit(&itask->itask_mutex);
4996 return (STMF_ABORTED);
4997 }
4998 mutex_exit(&itask->itask_mutex);
4999
5000 #ifdef DEBUG
5001 if (!(ioflags & STMF_IOF_STATS_ONLY) && stmf_drop_buf_counter > 0) {
5002 if (atomic_dec_32_nv((uint32_t *)&stmf_drop_buf_counter) == 1)
5003 return (STMF_SUCCESS);
5004 }
5005 #endif
5006
5007 stmf_update_kstat_lu_io(task, dbuf);
5008 stmf_update_kstat_lport_io(task, dbuf);
5009 stmf_update_kstat_rport_io(task, dbuf);
5010 stmf_lport_xfer_start(itask, dbuf);
5011 if (ioflags & STMF_IOF_STATS_ONLY) {
5012 stmf_lport_xfer_done(itask, dbuf);
5013 return (STMF_SUCCESS);
5014 }
5015
5016 dbuf->db_flags |= DB_LPORT_XFER_ACTIVE;
5017 ret = task->task_lport->lport_xfer_data(task, dbuf, ioflags);
5018
5019 /*
5020 * Port provider may have already called the buffer callback in
5021 * which case dbuf->db_xfer_start_timestamp will be 0.
5022 */
5023 if (ret != STMF_SUCCESS) {
5024 dbuf->db_flags &= ~DB_LPORT_XFER_ACTIVE;
5025 if (dbuf->db_xfer_start_timestamp != 0)
5026 stmf_lport_xfer_done(itask, dbuf);
5027 }
5028
5029 return (ret);
5030 }
5031
5032 void
5033 stmf_data_xfer_done(scsi_task_t *task, stmf_data_buf_t *dbuf, uint32_t iof)
5034 {
5035 stmf_i_scsi_task_t *itask =
5036 (stmf_i_scsi_task_t *)task->task_stmf_private;
5037 stmf_i_local_port_t *ilport;
5038 stmf_worker_t *w = itask->itask_worker;
5039 uint32_t new;
5040 uint8_t update_queue_flags, free_it, queue_it;
5041
5042 stmf_lport_xfer_done(itask, dbuf);
5043
5044 stmf_task_audit(itask, TE_XFER_DONE, iof, dbuf);
5045
5046 /* Guard against unexpected completions from the lport */
5047 if (dbuf->db_flags & DB_LPORT_XFER_ACTIVE) {
5048 dbuf->db_flags &= ~DB_LPORT_XFER_ACTIVE;
5049 } else {
5050 /*
5051 * This should never happen.
5052 */
5053 ilport = task->task_lport->lport_stmf_private;
5054 ilport->ilport_unexpected_comp++;
5055 cmn_err(CE_PANIC, "Unexpected xfer completion task %p dbuf %p",
5056 (void *)task, (void *)dbuf);
5057 return;
5058 }
5059
5060 mutex_enter(&itask->itask_mutex);
5061 mutex_enter(&w->worker_lock);
5062 new = itask->itask_flags;
5063 if (itask->itask_flags & ITASK_BEING_ABORTED) {
5064 mutex_exit(&w->worker_lock);
5065 mutex_exit(&itask->itask_mutex);
5066 return;
5067 }
5068 free_it = 0;
5069 if (iof & STMF_IOF_LPORT_DONE) {
5070 new &= ~ITASK_KNOWN_TO_TGT_PORT;
5071 task->task_completion_status = dbuf->db_xfer_status;
5072 free_it = 1;
5073 }
5074 /*
5075 * If the task is known to LU then queue it. But if
5076 * it is already queued (multiple completions) then
5077 * just update the buffer information by grabbing the
5078 * worker lock. If the task is not known to LU,
5079 * completed/aborted, then see if we need to
5080 * free this task.
5081 */
5082 if (itask->itask_flags & ITASK_KNOWN_TO_LU) {
5083 free_it = 0;
5084 update_queue_flags = 1;
5085 if (itask->itask_flags & ITASK_IN_WORKER_QUEUE) {
5086 queue_it = 0;
5087 } else {
5088 queue_it = 1;
5089 }
5090 } else {
5091 update_queue_flags = 0;
5092 queue_it = 0;
5093 }
5094 itask->itask_flags = new;
5095
5096 if (update_queue_flags) {
5097 uint8_t cmd = (dbuf->db_handle << 5) | ITASK_CMD_DATA_XFER_DONE;
5098
5099 ASSERT((itask->itask_flags & ITASK_IN_FREE_LIST) == 0);
5100 ASSERT(itask->itask_ncmds < ITASK_MAX_NCMDS);
5101
5102 itask->itask_cmd_stack[itask->itask_ncmds++] = cmd;
5103 if (queue_it) {
5104 STMF_ENQUEUE_ITASK(w, itask);
5105 }
5106 mutex_exit(&w->worker_lock);
5107 mutex_exit(&itask->itask_mutex);
5108 return;
5109 }
5110
5111 mutex_exit(&w->worker_lock);
5112 if (free_it) {
5113 if ((itask->itask_flags & (ITASK_KNOWN_TO_LU |
5114 ITASK_KNOWN_TO_TGT_PORT | ITASK_IN_WORKER_QUEUE |
5115 ITASK_BEING_ABORTED)) == 0) {
5116 stmf_task_free(task);
5117 return;
5118 }
5119 }
5120 mutex_exit(&itask->itask_mutex);
5121 }
5122
5123 stmf_status_t
5124 stmf_send_scsi_status(scsi_task_t *task, uint32_t ioflags)
5125 {
5126 DTRACE_PROBE1(scsi__send__status, scsi_task_t *, task);
5127
5128 stmf_i_scsi_task_t *itask =
5129 (stmf_i_scsi_task_t *)task->task_stmf_private;
5130
5131 stmf_task_audit(itask, TE_SEND_STATUS, ioflags, NULL);
5132
5133 mutex_enter(&itask->itask_mutex);
5134 if (ioflags & STMF_IOF_LU_DONE) {
5135 if (itask->itask_flags & ITASK_BEING_ABORTED) {
5136 mutex_exit(&itask->itask_mutex);
5137 return (STMF_ABORTED);
5138 }
5139 itask->itask_flags &= ~ITASK_KNOWN_TO_LU;
5140 }
5141
5142 if (!(itask->itask_flags & ITASK_KNOWN_TO_TGT_PORT)) {
5143 mutex_exit(&itask->itask_mutex);
5144 return (STMF_SUCCESS);
5145 }
5146
5147 if (itask->itask_flags & ITASK_BEING_ABORTED) {
5148 mutex_exit(&itask->itask_mutex);
5149 return (STMF_ABORTED);
5150 }
5151 mutex_exit(&itask->itask_mutex);
5152
5153 if (task->task_additional_flags & TASK_AF_NO_EXPECTED_XFER_LENGTH) {
5154 task->task_status_ctrl = 0;
5155 task->task_resid = 0;
5156 } else if (task->task_cmd_xfer_length >
5157 task->task_expected_xfer_length) {
5158 task->task_status_ctrl = TASK_SCTRL_OVER;
5159 task->task_resid = task->task_cmd_xfer_length -
5160 task->task_expected_xfer_length;
5161 } else if (task->task_nbytes_transferred <
5162 task->task_expected_xfer_length) {
5163 task->task_status_ctrl = TASK_SCTRL_UNDER;
5164 task->task_resid = task->task_expected_xfer_length -
5165 task->task_nbytes_transferred;
5166 } else {
5167 task->task_status_ctrl = 0;
5168 task->task_resid = 0;
5169 }
5170 return (task->task_lport->lport_send_status(task, ioflags));
5171 }
5172
5173 void
5174 stmf_send_status_done(scsi_task_t *task, stmf_status_t s, uint32_t iof)
5175 {
5176 stmf_i_scsi_task_t *itask =
5177 (stmf_i_scsi_task_t *)task->task_stmf_private;
5178 stmf_worker_t *w = itask->itask_worker;
5179 uint32_t new;
5180 uint8_t free_it, queue_it;
5181
5182 stmf_task_audit(itask, TE_SEND_STATUS_DONE, iof, NULL);
5183
5184 mutex_enter(&itask->itask_mutex);
5185 mutex_enter(&w->worker_lock);
5186 new = itask->itask_flags;
5187 if (itask->itask_flags & ITASK_BEING_ABORTED) {
5188 mutex_exit(&w->worker_lock);
5189 mutex_exit(&itask->itask_mutex);
5190 return;
5191 }
5192 free_it = 0;
5193 if (iof & STMF_IOF_LPORT_DONE) {
5194 new &= ~ITASK_KNOWN_TO_TGT_PORT;
5195 free_it = 1;
5196 }
5197 /*
5198 * If the task is known to LU then queue it. But if
5199 * it is already queued (multiple completions) then
5200 * just update the buffer information by grabbing the
5201 * worker lock. If the task is not known to LU,
5202 * completed/aborted, then see if we need to
5203 * free this task.
5204 */
5205 if (itask->itask_flags & ITASK_KNOWN_TO_LU) {
5206 free_it = 0;
5207 queue_it = 1;
5208 if (itask->itask_flags & ITASK_IN_WORKER_QUEUE) {
5209 cmn_err(CE_PANIC, "status completion received"
5210 " when task is already in worker queue "
5211 " task = %p", (void *)task);
5212 }
5213 } else {
5214 queue_it = 0;
5215 }
5216 itask->itask_flags = new;
5217 task->task_completion_status = s;
5218
5219 if (queue_it) {
5220 ASSERT(itask->itask_ncmds < ITASK_MAX_NCMDS);
5221 itask->itask_cmd_stack[itask->itask_ncmds++] =
5222 ITASK_CMD_STATUS_DONE;
5223
5224 STMF_ENQUEUE_ITASK(w, itask);
5225 mutex_exit(&w->worker_lock);
5226 mutex_exit(&itask->itask_mutex);
5227 return;
5228 }
5229
5230 mutex_exit(&w->worker_lock);
5231
5232 if (free_it) {
5233 if ((itask->itask_flags & (ITASK_KNOWN_TO_LU |
5234 ITASK_KNOWN_TO_TGT_PORT | ITASK_IN_WORKER_QUEUE |
5235 ITASK_BEING_ABORTED)) == 0) {
5236 stmf_task_free(task);
5237 return;
5238 } else {
5239 cmn_err(CE_PANIC, "LU is done with the task but LPORT "
5240 " is not done, itask %p itask_flags %x",
5241 (void *)itask, itask->itask_flags);
5242 }
5243 }
5244 mutex_exit(&itask->itask_mutex);
5245 }
5246
5247 void
5248 stmf_task_lu_done(scsi_task_t *task)
5249 {
5250 stmf_i_scsi_task_t *itask =
5251 (stmf_i_scsi_task_t *)task->task_stmf_private;
5252 stmf_worker_t *w = itask->itask_worker;
5253
5254 mutex_enter(&itask->itask_mutex);
5255 mutex_enter(&w->worker_lock);
5256 if (itask->itask_flags & ITASK_BEING_ABORTED) {
5257 mutex_exit(&w->worker_lock);
5258 mutex_exit(&itask->itask_mutex);
5259 return;
5260 }
5261 if (itask->itask_flags & ITASK_IN_WORKER_QUEUE) {
5262 cmn_err(CE_PANIC, "task_lu_done received"
5263 " when task is in worker queue "
5264 " task = %p", (void *)task);
5265 }
5266 itask->itask_flags &= ~ITASK_KNOWN_TO_LU;
5267
5268 mutex_exit(&w->worker_lock);
5269 if ((itask->itask_flags & (ITASK_KNOWN_TO_LU |
5270 ITASK_KNOWN_TO_TGT_PORT | ITASK_IN_WORKER_QUEUE |
5271 ITASK_BEING_ABORTED)) == 0) {
5272 stmf_task_free(task);
5273 return;
5274 } else {
5275 cmn_err(CE_PANIC, "stmf_lu_done should be the last stage but "
5276 " the task is still not done, task = %p", (void *)task);
5277 }
5278 mutex_exit(&itask->itask_mutex);
5279 }
5280
5281 void
5282 stmf_queue_task_for_abort(scsi_task_t *task, stmf_status_t s)
5283 {
5284 stmf_i_scsi_task_t *itask =
5285 (stmf_i_scsi_task_t *)task->task_stmf_private;
5286 stmf_worker_t *w;
5287
5288 stmf_task_audit(itask, TE_TASK_ABORT, CMD_OR_IOF_NA, NULL);
5289
5290 mutex_enter(&itask->itask_mutex);
5291 if ((itask->itask_flags & ITASK_BEING_ABORTED) ||
5292 ((itask->itask_flags & (ITASK_KNOWN_TO_TGT_PORT |
5293 ITASK_KNOWN_TO_LU)) == 0)) {
5294 mutex_exit(&itask->itask_mutex);
5295 return;
5296 }
5297 itask->itask_flags |= ITASK_BEING_ABORTED;
5298 task->task_completion_status = s;
5299
5300 if (((w = itask->itask_worker) == NULL) ||
5301 (itask->itask_flags & ITASK_IN_TRANSITION)) {
5302 mutex_exit(&itask->itask_mutex);
5303 return;
5304 }
5305
5306 /* Queue it and get out */
5307 if (itask->itask_flags & ITASK_IN_WORKER_QUEUE) {
5308 mutex_exit(&itask->itask_mutex);
5309 return;
5310 }
5311 mutex_enter(&w->worker_lock);
5312 STMF_ENQUEUE_ITASK(w, itask);
5313 mutex_exit(&w->worker_lock);
5314 mutex_exit(&itask->itask_mutex);
5315 }
5316
5317 void
5318 stmf_abort(int abort_cmd, scsi_task_t *task, stmf_status_t s, void *arg)
5319 {
5320 stmf_i_scsi_task_t *itask = NULL;
5321 uint32_t f, rf;
5322
5323 DTRACE_PROBE2(scsi__task__abort, scsi_task_t *, task,
5324 stmf_status_t, s);
5325
5326 switch (abort_cmd) {
5327 case STMF_QUEUE_ABORT_LU:
5328 stmf_task_lu_killall((stmf_lu_t *)arg, task, s);
5329 return;
5330 case STMF_QUEUE_TASK_ABORT:
5331 stmf_queue_task_for_abort(task, s);
5332 return;
5333 case STMF_REQUEUE_TASK_ABORT_LPORT:
5334 rf = ITASK_TGT_PORT_ABORT_CALLED;
5335 f = ITASK_KNOWN_TO_TGT_PORT;
5336 break;
5337 case STMF_REQUEUE_TASK_ABORT_LU:
5338 rf = ITASK_LU_ABORT_CALLED;
5339 f = ITASK_KNOWN_TO_LU;
5340 break;
5341 default:
5342 return;
5343 }
5344
5345 itask = (stmf_i_scsi_task_t *)task->task_stmf_private;
5346 mutex_enter(&itask->itask_mutex);
5347 f |= ITASK_BEING_ABORTED | rf;
5348
5349 if ((itask->itask_flags & f) != f) {
5350 mutex_exit(&itask->itask_mutex);
5351 return;
5352 }
5353 itask->itask_flags &= ~rf;
5354 mutex_exit(&itask->itask_mutex);
5355
5356 }
5357
5358 /*
5359 * NOTE: stmf_abort_task_offline will release and then reacquire the
5360 * itask_mutex. This is required to prevent a lock order violation.
5361 */
5362 void
5363 stmf_task_lu_aborted(scsi_task_t *task, stmf_status_t s, uint32_t iof)
5364 {
5365 char info[STMF_CHANGE_INFO_LEN];
5366 stmf_i_scsi_task_t *itask = TASK_TO_ITASK(task);
5367 unsigned long long st;
5368
5369 stmf_task_audit(itask, TE_TASK_LU_ABORTED, iof, NULL);
5370 ASSERT(mutex_owned(&itask->itask_mutex));
5371 st = s; /* gcc fix */
5372 if ((s != STMF_ABORT_SUCCESS) && (s != STMF_NOT_FOUND)) {
5373 (void) snprintf(info, sizeof (info),
5374 "task %p, lu failed to abort ret=%llx", (void *)task, st);
5375 } else if ((iof & STMF_IOF_LU_DONE) == 0) {
5376 (void) snprintf(info, sizeof (info),
5377 "Task aborted but LU is not finished, task ="
5378 "%p, s=%llx, iof=%x", (void *)task, st, iof);
5379 } else {
5380 /*
5381 * LU abort successfully
5382 */
5383 atomic_and_32(&itask->itask_flags, ~ITASK_KNOWN_TO_LU);
5384 return;
5385 }
5386
5387 stmf_abort_task_offline(task, 1, info);
5388 }
5389
5390 /*
5391 * NOTE: stmf_abort_task_offline will release and then reacquire the
5392 * itask_mutex. This is required to prevent a lock order violation.
5393 */
5394 void
5395 stmf_task_lport_aborted(scsi_task_t *task, stmf_status_t s, uint32_t iof)
5396 {
5397 char info[STMF_CHANGE_INFO_LEN];
5398 stmf_i_scsi_task_t *itask = TASK_TO_ITASK(task);
5399 unsigned long long st;
5400
5401 ASSERT(mutex_owned(&itask->itask_mutex));
5402 stmf_task_audit(itask, TE_TASK_LPORT_ABORTED, iof, NULL);
5403 st = s;
5404 if ((s != STMF_ABORT_SUCCESS) && (s != STMF_NOT_FOUND)) {
5405 (void) snprintf(info, sizeof (info),
5406 "task %p, tgt port failed to abort ret=%llx", (void *)task,
5407 st);
5408 } else if ((iof & STMF_IOF_LPORT_DONE) == 0) {
5409 (void) snprintf(info, sizeof (info),
5410 "Task aborted but tgt port is not finished, "
5411 "task=%p, s=%llx, iof=%x", (void *)task, st, iof);
5412 } else {
5413 /*
5414 * LPORT abort successfully
5415 */
5416 atomic_and_32(&itask->itask_flags, ~ITASK_KNOWN_TO_TGT_PORT);
5417 return;
5418 }
5419
5420 stmf_abort_task_offline(task, 0, info);
5421 }
5422
5423 void
5424 stmf_task_lport_aborted_unlocked(scsi_task_t *task, stmf_status_t s,
5425 uint32_t iof)
5426 {
5427 stmf_i_scsi_task_t *itask = TASK_TO_ITASK(task);
5428
5429 mutex_enter(&itask->itask_mutex);
5430 stmf_task_lport_aborted(task, s, iof);
5431 mutex_exit(&itask->itask_mutex);
5432 }
5433
5434 stmf_status_t
5435 stmf_task_poll_lu(scsi_task_t *task, uint32_t timeout)
5436 {
5437 stmf_i_scsi_task_t *itask = (stmf_i_scsi_task_t *)
5438 task->task_stmf_private;
5439 stmf_worker_t *w = itask->itask_worker;
5440 int i;
5441
5442 mutex_enter(&itask->itask_mutex);
5443 ASSERT(itask->itask_flags & ITASK_KNOWN_TO_LU);
5444 mutex_enter(&w->worker_lock);
5445 if (itask->itask_ncmds >= ITASK_MAX_NCMDS) {
5446 mutex_exit(&w->worker_lock);
5447 mutex_exit(&itask->itask_mutex);
5448 return (STMF_BUSY);
5449 }
5450 for (i = 0; i < itask->itask_ncmds; i++) {
5451 if (itask->itask_cmd_stack[i] == ITASK_CMD_POLL_LU) {
5452 mutex_exit(&w->worker_lock);
5453 mutex_exit(&itask->itask_mutex);
5454 return (STMF_SUCCESS);
5455 }
5456 }
5457 itask->itask_cmd_stack[itask->itask_ncmds++] = ITASK_CMD_POLL_LU;
5458 if (timeout == ITASK_DEFAULT_POLL_TIMEOUT) {
5459 itask->itask_poll_timeout = ddi_get_lbolt() + 1;
5460 } else {
5461 clock_t t = drv_usectohz(timeout * 1000);
5462 if (t == 0)
5463 t = 1;
5464 itask->itask_poll_timeout = ddi_get_lbolt() + t;
5465 }
5466 if ((itask->itask_flags & ITASK_IN_WORKER_QUEUE) == 0) {
5467 STMF_ENQUEUE_ITASK(w, itask);
5468 }
5469 mutex_exit(&w->worker_lock);
5470 mutex_exit(&itask->itask_mutex);
5471 return (STMF_SUCCESS);
5472 }
5473
5474 stmf_status_t
5475 stmf_task_poll_lport(scsi_task_t *task, uint32_t timeout)
5476 {
5477 stmf_i_scsi_task_t *itask = (stmf_i_scsi_task_t *)
5478 task->task_stmf_private;
5479 stmf_worker_t *w = itask->itask_worker;
5480 int i;
5481
5482 mutex_enter(&itask->itask_mutex);
5483 ASSERT(itask->itask_flags & ITASK_KNOWN_TO_TGT_PORT);
5484 mutex_enter(&w->worker_lock);
5485 if (itask->itask_ncmds >= ITASK_MAX_NCMDS) {
5486 mutex_exit(&w->worker_lock);
5487 mutex_exit(&itask->itask_mutex);
5488 return (STMF_BUSY);
5489 }
5490 for (i = 0; i < itask->itask_ncmds; i++) {
5491 if (itask->itask_cmd_stack[i] == ITASK_CMD_POLL_LPORT) {
5492 mutex_exit(&w->worker_lock);
5493 mutex_exit(&itask->itask_mutex);
5494 return (STMF_SUCCESS);
5495 }
5496 }
5497 itask->itask_cmd_stack[itask->itask_ncmds++] = ITASK_CMD_POLL_LPORT;
5498 if (timeout == ITASK_DEFAULT_POLL_TIMEOUT) {
5499 itask->itask_poll_timeout = ddi_get_lbolt() + 1;
5500 } else {
5501 clock_t t = drv_usectohz(timeout * 1000);
5502 if (t == 0)
5503 t = 1;
5504 itask->itask_poll_timeout = ddi_get_lbolt() + t;
5505 }
5506 if ((itask->itask_flags & ITASK_IN_WORKER_QUEUE) == 0) {
5507 STMF_ENQUEUE_ITASK(w, itask);
5508 }
5509 mutex_exit(&w->worker_lock);
5510 mutex_exit(&itask->itask_mutex);
5511 return (STMF_SUCCESS);
5512 }
5513
5514 void
5515 stmf_do_task_abort(scsi_task_t *task)
5516 {
5517 stmf_i_scsi_task_t *itask = TASK_TO_ITASK(task);
5518 stmf_lu_t *lu;
5519 stmf_local_port_t *lport;
5520 unsigned long long ret;
5521 uint32_t new = 0;
5522 uint8_t call_lu_abort, call_port_abort;
5523 char info[STMF_CHANGE_INFO_LEN];
5524
5525 lu = task->task_lu;
5526 lport = task->task_lport;
5527 mutex_enter(&itask->itask_mutex);
5528 new = itask->itask_flags;
5529 if ((itask->itask_flags & (ITASK_KNOWN_TO_LU |
5530 ITASK_LU_ABORT_CALLED)) == ITASK_KNOWN_TO_LU) {
5531 new |= ITASK_LU_ABORT_CALLED;
5532 call_lu_abort = 1;
5533 } else {
5534 call_lu_abort = 0;
5535 }
5536 itask->itask_flags = new;
5537
5538 if (call_lu_abort) {
5539 if ((itask->itask_flags & ITASK_DEFAULT_HANDLING) == 0) {
5540 ret = lu->lu_abort(lu, STMF_LU_ABORT_TASK, task, 0);
5541 } else {
5542 ret = dlun0->lu_abort(lu, STMF_LU_ABORT_TASK, task, 0);
5543 }
5544 if ((ret == STMF_ABORT_SUCCESS) || (ret == STMF_NOT_FOUND)) {
5545 stmf_task_lu_aborted(task, ret, STMF_IOF_LU_DONE);
5546 } else if (ret == STMF_BUSY) {
5547 atomic_and_32(&itask->itask_flags,
5548 ~ITASK_LU_ABORT_CALLED);
5549 } else if (ret != STMF_SUCCESS) {
5550 (void) snprintf(info, sizeof (info),
5551 "Abort failed by LU %p, ret %llx", (void *)lu, ret);
5552 stmf_abort_task_offline(task, 1, info);
5553 }
5554 } else if (itask->itask_flags & ITASK_KNOWN_TO_LU) {
5555 if (ddi_get_lbolt() > (itask->itask_start_time +
5556 STMF_SEC2TICK(lu->lu_abort_timeout?
5557 lu->lu_abort_timeout : ITASK_DEFAULT_ABORT_TIMEOUT))) {
5558 (void) snprintf(info, sizeof (info),
5559 "lu abort timed out");
5560 stmf_abort_task_offline(itask->itask_task, 1, info);
5561 }
5562 }
5563
5564 /*
5565 * NOTE: After the call to either stmf_abort_task_offline() or
5566 * stmf_task_lu_abort() the itask_mutex was dropped and reacquired
5567 * to avoid a deadlock situation with stmf_state.stmf_lock.
5568 */
5569
5570 new = itask->itask_flags;
5571 if ((itask->itask_flags & (ITASK_KNOWN_TO_TGT_PORT |
5572 ITASK_TGT_PORT_ABORT_CALLED)) == ITASK_KNOWN_TO_TGT_PORT) {
5573 new |= ITASK_TGT_PORT_ABORT_CALLED;
5574 call_port_abort = 1;
5575 } else {
5576 call_port_abort = 0;
5577 }
5578 itask->itask_flags = new;
5579
5580 if (call_port_abort) {
5581 ret = lport->lport_abort(lport, STMF_LPORT_ABORT_TASK, task, 0);
5582 if ((ret == STMF_ABORT_SUCCESS) || (ret == STMF_NOT_FOUND)) {
5583 stmf_task_lport_aborted(task, ret, STMF_IOF_LPORT_DONE);
5584 } else if (ret == STMF_BUSY) {
5585 atomic_and_32(&itask->itask_flags,
5586 ~ITASK_TGT_PORT_ABORT_CALLED);
5587 } else if (ret != STMF_SUCCESS) {
5588 (void) snprintf(info, sizeof (info),
5589 "Abort failed by tgt port %p ret %llx",
5590 (void *)lport, ret);
5591 stmf_abort_task_offline(task, 0, info);
5592 }
5593 } else if (itask->itask_flags & ITASK_KNOWN_TO_TGT_PORT) {
5594 if (ddi_get_lbolt() > (itask->itask_start_time +
5595 STMF_SEC2TICK(lport->lport_abort_timeout?
5596 lport->lport_abort_timeout :
5597 ITASK_DEFAULT_ABORT_TIMEOUT))) {
5598 (void) snprintf(info, sizeof (info),
5599 "lport abort timed out");
5600 stmf_abort_task_offline(itask->itask_task, 0, info);
5601 }
5602 }
5603 mutex_exit(&itask->itask_mutex);
5604 }
5605
5606 stmf_status_t
5607 stmf_ctl(int cmd, void *obj, void *arg)
5608 {
5609 stmf_status_t ret;
5610 stmf_i_lu_t *ilu;
5611 stmf_i_local_port_t *ilport;
5612 stmf_state_change_info_t *ssci = (stmf_state_change_info_t *)arg;
5613
5614 mutex_enter(&stmf_state.stmf_lock);
5615 ret = STMF_INVALID_ARG;
5616 if (cmd & STMF_CMD_LU_OP) {
5617 ilu = stmf_lookup_lu((stmf_lu_t *)obj);
5618 if (ilu == NULL) {
5619 goto stmf_ctl_lock_exit;
5620 }
5621 DTRACE_PROBE3(lu__state__change,
5622 stmf_lu_t *, ilu->ilu_lu,
5623 int, cmd, stmf_state_change_info_t *, ssci);
5915 sdid->association = ID_IS_TARGET_PORT;
5916 sdid->ident_length = 20;
5917 /* Convert wwn value to "wwn.XXXXXXXXXXXXXXXX" format */
5918 (void) snprintf(wwn_str, sizeof (wwn_str),
5919 "wwn.%02X%02X%02X%02X%02X%02X%02X%02X",
5920 wwn[0], wwn[1], wwn[2], wwn[3], wwn[4], wwn[5], wwn[6], wwn[7]);
5921 bcopy(wwn_str, (char *)sdid->ident, 20);
5922 }
5923
5924
5925 stmf_xfer_data_t *
5926 stmf_prepare_tpgs_data(uint8_t ilu_alua)
5927 {
5928 stmf_xfer_data_t *xd;
5929 stmf_i_local_port_t *ilport;
5930 uint8_t *p;
5931 uint32_t sz, asz, nports = 0, nports_standby = 0;
5932
5933 mutex_enter(&stmf_state.stmf_lock);
5934 /* check if any ports are standby and create second group */
5935 for (ilport = stmf_state.stmf_ilportlist; ilport != NULL;
5936 ilport = ilport->ilport_next) {
5937 if (ilport->ilport_standby == 1) {
5938 nports_standby++;
5939 } else {
5940 nports++;
5941 }
5942 }
5943
5944 /*
5945 * Section 6.25 REPORT TARGET PORT GROUPS
5946 * The reply can contain many group replies. Each group is limited
5947 * to 255 port identifiers so we'll need to limit the amount of
5948 * data returned. For FC ports there's a physical limitation in
5949 * machines that make reaching 255 ports very, very unlikely. For
5950 * iSCSI on the other hand recent changes mean the port count could
5951 * be as high as 4096 (current limit). Limiting the data returned
5952 * for iSCSI isn't as bad as it sounds. This information is only
5953 * important for ALUA, which isn't supported for iSCSI. iSCSI uses
5954 * virtual IP addresses to deal with node fail over in a cluster.
5955 */
5956 nports = min(nports, 255);
5957 nports_standby = min(nports_standby, 255);
5958
5959 /*
5960 * The first 4 bytes of the returned data is the length. The
5961 * size of the Target Port Group header is 8 bytes. So, that's where
5962 * the 12 comes from. Each port entry is 4 bytes in size.
5963 */
5964 sz = (nports * 4) + 12;
5965 if (nports_standby != 0 && ilu_alua != 0) {
5966 /* --- Only add 8 bytes since it's just the Group header ----*/
5967 sz += (nports_standby * 4) + 8;
5968 }
5969
5970 /*
5971 * The stmf_xfer_data structure contains 4 bytes that will be
5972 * part of the data buffer. So, subtract the 4 bytes from the space
5973 * needed.
5974 */
5975 asz = sizeof (*xd) + sz - 4;
5976 xd = (stmf_xfer_data_t *)kmem_zalloc(asz, KM_NOSLEEP);
5977 if (xd == NULL) {
5978 mutex_exit(&stmf_state.stmf_lock);
5979 return (NULL);
5980 }
5981 xd->alloc_size = asz;
5982 xd->size_left = sz;
5983
5984 p = xd->buf;
5985
5986 /* ---- length values never include the field that holds the size ----*/
5987 *((uint32_t *)p) = BE_32(sz - 4);
5988 p += 4;
5989
5990 /* ---- Now fill out the first Target Group header ---- */
5991 p[0] = 0x80; /* PREF */
5992 p[1] = 5; /* AO_SUP, S_SUP */
5993 if (stmf_state.stmf_alua_node == 1) {
5994 p[3] = 1; /* Group 1 */
5995 } else {
5996 p[3] = 0; /* Group 0 */
5997 }
5998 p[7] = nports & 0xff;
5999 p += 8;
6000 for (ilport = stmf_state.stmf_ilportlist; ilport != NULL && nports != 0;
6001 ilport = ilport->ilport_next) {
6002 if (ilport->ilport_standby == 1) {
6003 continue;
6004 }
6005 ((uint16_t *)p)[1] = BE_16(ilport->ilport_rtpid);
6006 p += 4;
6007 nports--;
6008 }
6009 if (nports_standby != 0 && ilu_alua != 0) {
6010 p[0] = 0x02; /* Non PREF, Standby */
6011 p[1] = 5; /* AO_SUP, S_SUP */
6012 if (stmf_state.stmf_alua_node == 1) {
6013 p[3] = 0; /* Group 0 */
6014 } else {
6015 p[3] = 1; /* Group 1 */
6016 }
6017 p[7] = nports_standby & 0xff;
6018 p += 8;
6019 for (ilport = stmf_state.stmf_ilportlist; ilport != NULL &&
6020 nports_standby != 0; ilport = ilport->ilport_next) {
6021 if (ilport->ilport_standby == 0) {
6022 continue;
6023 }
6024 ((uint16_t *)p)[1] = BE_16(ilport->ilport_rtpid);
6025 p += 4;
6026 nports_standby--;
6027 }
6028 }
6029
6030 mutex_exit(&stmf_state.stmf_lock);
6031
6032 return (xd);
6033 }
6034
6035 struct scsi_devid_desc *
6036 stmf_scsilib_get_devid_desc(uint16_t rtpid)
6037 {
6038 scsi_devid_desc_t *devid = NULL;
6039 stmf_i_local_port_t *ilport;
6040
6041 mutex_enter(&stmf_state.stmf_lock);
6042
6043 for (ilport = stmf_state.stmf_ilportlist; ilport;
6044 ilport = ilport->ilport_next) {
6045 if (ilport->ilport_rtpid == rtpid) {
6046 scsi_devid_desc_t *id = ilport->ilport_lport->lport_id;
6247 break;
6248 }
6249 }
6250
6251 page[2] = (m >> 8) & 0xff;
6252 page[3] = m & 0xff;
6253
6254 return (n);
6255 }
6256
6257 void
6258 stmf_scsilib_handle_report_tpgs(scsi_task_t *task, stmf_data_buf_t *dbuf)
6259 {
6260 stmf_i_scsi_task_t *itask =
6261 (stmf_i_scsi_task_t *)task->task_stmf_private;
6262 stmf_i_lu_t *ilu =
6263 (stmf_i_lu_t *)task->task_lu->lu_stmf_private;
6264 stmf_xfer_data_t *xd;
6265 uint32_t sz, minsz;
6266
6267 mutex_enter(&itask->itask_mutex);
6268 itask->itask_flags |= ITASK_DEFAULT_HANDLING;
6269
6270 task->task_cmd_xfer_length =
6271 ((((uint32_t)task->task_cdb[6]) << 24) |
6272 (((uint32_t)task->task_cdb[7]) << 16) |
6273 (((uint32_t)task->task_cdb[8]) << 8) |
6274 ((uint32_t)task->task_cdb[9]));
6275
6276 if (task->task_additional_flags &
6277 TASK_AF_NO_EXPECTED_XFER_LENGTH) {
6278 task->task_expected_xfer_length =
6279 task->task_cmd_xfer_length;
6280 }
6281 mutex_exit(&itask->itask_mutex);
6282
6283 if (task->task_cmd_xfer_length == 0) {
6284 stmf_scsilib_send_status(task, STATUS_GOOD, 0);
6285 return;
6286 }
6287 if (task->task_cmd_xfer_length < 4) {
6288 stmf_scsilib_send_status(task, STATUS_CHECK,
6289 STMF_SAA_INVALID_FIELD_IN_CDB);
6290 return;
6291 }
6292
6293 sz = min(task->task_expected_xfer_length,
6294 task->task_cmd_xfer_length);
6295
6296 xd = stmf_prepare_tpgs_data(ilu->ilu_alua);
6297
6298 if (xd == NULL) {
6299 stmf_abort(STMF_QUEUE_TASK_ABORT, task,
6300 STMF_ALLOC_FAILURE, NULL);
6301 return;
6367 /*
6368 * To sync with target reset, grab this lock. The LU is not going
6369 * anywhere as there is atleast one task pending (this task).
6370 */
6371 mutex_enter(&stmf_state.stmf_lock);
6372
6373 if (ilu->ilu_flags & ILU_RESET_ACTIVE) {
6374 mutex_exit(&stmf_state.stmf_lock);
6375 stmf_scsilib_send_status(task, STATUS_CHECK,
6376 STMF_SAA_OPERATION_IN_PROGRESS);
6377 return;
6378 }
6379 atomic_or_32(&ilu->ilu_flags, ILU_RESET_ACTIVE);
6380 mutex_exit(&stmf_state.stmf_lock);
6381
6382 /*
6383 * Mark this task as the one causing LU reset so that we know who
6384 * was responsible for setting the ILU_RESET_ACTIVE. In case this
6385 * task itself gets aborted, we will clear ILU_RESET_ACTIVE.
6386 */
6387 mutex_enter(&itask->itask_mutex);
6388 itask->itask_flags |= ITASK_DEFAULT_HANDLING | ITASK_CAUSING_LU_RESET;
6389 mutex_exit(&itask->itask_mutex);
6390
6391 /* Initiatiate abort on all commands on this LU except this one */
6392 stmf_abort(STMF_QUEUE_ABORT_LU, task, STMF_ABORTED, task->task_lu);
6393
6394 /* Start polling on this task */
6395 if (stmf_task_poll_lu(task, ITASK_DEFAULT_POLL_TIMEOUT)
6396 != STMF_SUCCESS) {
6397 stmf_abort(STMF_QUEUE_TASK_ABORT, task, STMF_ALLOC_FAILURE,
6398 NULL);
6399 return;
6400 }
6401 }
6402
6403 void
6404 stmf_handle_target_reset(scsi_task_t *task)
6405 {
6406 stmf_i_scsi_task_t *itask;
6407 stmf_i_lu_t *ilu;
6408 stmf_i_scsi_session_t *iss;
6409 stmf_lun_map_t *lm;
6445 ilu = (stmf_i_lu_t *)(lm_ent->ent_lu->lu_stmf_private);
6446 if (ilu->ilu_flags & ILU_RESET_ACTIVE) {
6447 atomic_and_32(&iss->iss_flags, ~ISS_RESET_ACTIVE);
6448 rw_exit(iss->iss_lockp);
6449 mutex_exit(&stmf_state.stmf_lock);
6450 stmf_scsilib_send_status(task, STATUS_CHECK,
6451 STMF_SAA_OPERATION_IN_PROGRESS);
6452 return;
6453 }
6454 }
6455 if (lf == 0) {
6456 /* No luns in this session */
6457 atomic_and_32(&iss->iss_flags, ~ISS_RESET_ACTIVE);
6458 rw_exit(iss->iss_lockp);
6459 mutex_exit(&stmf_state.stmf_lock);
6460 stmf_scsilib_send_status(task, STATUS_GOOD, 0);
6461 return;
6462 }
6463
6464 /* ok, start the damage */
6465 mutex_enter(&itask->itask_mutex);
6466 itask->itask_flags |= ITASK_DEFAULT_HANDLING |
6467 ITASK_CAUSING_TARGET_RESET;
6468 mutex_exit(&itask->itask_mutex);
6469 for (i = 0; i < lm->lm_nentries; i++) {
6470 if (lm->lm_plus[i] == NULL)
6471 continue;
6472 lm_ent = (stmf_lun_map_ent_t *)lm->lm_plus[i];
6473 ilu = (stmf_i_lu_t *)(lm_ent->ent_lu->lu_stmf_private);
6474 atomic_or_32(&ilu->ilu_flags, ILU_RESET_ACTIVE);
6475 }
6476
6477 for (i = 0; i < lm->lm_nentries; i++) {
6478 if (lm->lm_plus[i] == NULL)
6479 continue;
6480 lm_ent = (stmf_lun_map_ent_t *)lm->lm_plus[i];
6481 stmf_abort(STMF_QUEUE_ABORT_LU, task, STMF_ABORTED,
6482 lm_ent->ent_lu);
6483 }
6484
6485 rw_exit(iss->iss_lockp);
6486 mutex_exit(&stmf_state.stmf_lock);
6487
6488 /* Start polling on this task */
6506 (task->task_cdb[0] == SCMD_INQUIRY)) {
6507 rw_exit(iss->iss_lockp);
6508 return (0);
6509 }
6510 atomic_and_32(&iss->iss_flags,
6511 ~(ISS_LUN_INVENTORY_CHANGED | ISS_GOT_INITIAL_LUNS));
6512 rw_exit(iss->iss_lockp);
6513
6514 if (task->task_cdb[0] == SCMD_REPORT_LUNS) {
6515 return (0);
6516 }
6517 stmf_scsilib_send_status(task, STATUS_CHECK,
6518 STMF_SAA_REPORT_LUN_DATA_HAS_CHANGED);
6519 return (1);
6520 }
6521
6522 void
6523 stmf_worker_init()
6524 {
6525 uint32_t i;
6526 stmf_worker_t *w;
6527
6528 /* Make local copy of global tunables */
6529
6530 /*
6531 * Allow workers to be scaled down to a very low number for cases
6532 * where the load is light. If the number of threads gets below
6533 * 4 assume it is a mistake and force the threads back to a
6534 * reasonable number. The low limit of 4 is simply legacy and
6535 * may be too low.
6536 */
6537 ASSERT(stmf_workers == NULL);
6538 if (stmf_nworkers < 4) {
6539 stmf_nworkers = 64;
6540 }
6541
6542 stmf_workers = (stmf_worker_t *)kmem_zalloc(
6543 sizeof (stmf_worker_t) * stmf_nworkers, KM_SLEEP);
6544 for (i = 0; i < stmf_nworkers; i++) {
6545 stmf_worker_t *w = &stmf_workers[i];
6546 mutex_init(&w->worker_lock, NULL, MUTEX_DRIVER, NULL);
6547 cv_init(&w->worker_cv, NULL, CV_DRIVER, NULL);
6548 }
6549 stmf_workers_state = STMF_WORKERS_ENABLED;
6550
6551 /* Check if we are starting */
6552 if (stmf_nworkers_cur < stmf_nworkers - 1) {
6553 for (i = stmf_nworkers_cur; i < stmf_nworkers; i++) {
6554 w = &stmf_workers[i];
6555 w->worker_tid = thread_create(NULL, 0, stmf_worker_task,
6556 (void *)&stmf_workers[i], 0, &p0, TS_RUN,
6557 minclsyspri);
6558 stmf_nworkers_accepting_cmds++;
6559 }
6560 return;
6561 }
6562
6563 /* Lets wait for atleast one worker to start */
6564 while (stmf_nworkers_cur == 0)
6565 delay(drv_usectohz(20 * 1000));
6566 }
6567
6568 stmf_status_t
6569 stmf_worker_fini()
6570 {
6571 int i;
6572 clock_t sb;
6573
6574 if (stmf_workers_state == STMF_WORKERS_DISABLED)
6575 return (STMF_SUCCESS);
6576 ASSERT(stmf_workers);
6577 stmf_workers_state = STMF_WORKERS_DISABLED;
6578 cv_signal(&stmf_state.stmf_cv);
6579
6580 sb = ddi_get_lbolt() + drv_usectohz(10 * 1000 * 1000);
6581 /* Wait for all the threads to die */
6582 while (stmf_nworkers_cur != 0) {
6583 if (ddi_get_lbolt() > sb) {
6584 stmf_workers_state = STMF_WORKERS_ENABLED;
6585 return (STMF_BUSY);
6586 }
6587 delay(drv_usectohz(100 * 1000));
6588 }
6589 for (i = 0; i < stmf_nworkers; i++) {
6590 stmf_worker_t *w = &stmf_workers[i];
6591 mutex_destroy(&w->worker_lock);
6592 cv_destroy(&w->worker_cv);
6593 }
6594 kmem_free(stmf_workers, sizeof (stmf_worker_t) * stmf_nworkers);
6595 stmf_workers = NULL;
6596
6597 return (STMF_SUCCESS);
6598 }
6599
6600 void
6601 stmf_worker_task(void *arg)
6602 {
6603 stmf_worker_t *w;
6604 stmf_i_scsi_session_t *iss;
6605 scsi_task_t *task;
6606 stmf_i_scsi_task_t *itask;
6607 stmf_data_buf_t *dbuf;
6608 stmf_lu_t *lu;
6609 clock_t wait_timer = 0;
6610 clock_t wait_ticks, wait_delta = 0;
6611 uint8_t curcmd;
6612 uint8_t abort_free;
6613 uint8_t wait_queue;
6614 uint8_t dec_qdepth;
6615
6616 w = (stmf_worker_t *)arg;
6617 wait_ticks = drv_usectohz(10000);
6618
6619 DTRACE_PROBE1(worker__create, stmf_worker_t, w);
6620 mutex_enter(&w->worker_lock);
6621 w->worker_flags |= STMF_WORKER_STARTED | STMF_WORKER_ACTIVE;
6622 atomic_inc_32(&stmf_nworkers_cur);
6623
6624 stmf_worker_loop:
6625 if ((w->worker_ref_count == 0) &&
6626 (w->worker_flags & STMF_WORKER_TERMINATE)) {
6627 w->worker_flags &= ~(STMF_WORKER_STARTED |
6628 STMF_WORKER_ACTIVE | STMF_WORKER_TERMINATE);
6629 w->worker_tid = NULL;
6630 mutex_exit(&w->worker_lock);
6631 DTRACE_PROBE1(worker__destroy, stmf_worker_t, w);
6632 atomic_dec_32(&stmf_nworkers_cur);
6633 thread_exit();
6634 }
6635
6636 /* CONSTCOND */
6637 while (1) {
6638 /* worker lock is held at this point */
6639 dec_qdepth = 0;
6640 if (wait_timer && (ddi_get_lbolt() >= wait_timer)) {
6641 wait_timer = 0;
6642 wait_delta = 0;
6643 if (w->worker_wait_head) {
6644 ASSERT(w->worker_wait_tail);
6645 if (w->worker_task_head == NULL)
6646 w->worker_task_head =
6647 w->worker_wait_head;
6648 else
6649 w->worker_task_tail->itask_worker_next =
6650 w->worker_wait_head;
6651 w->worker_task_tail = w->worker_wait_tail;
6652 w->worker_wait_head = w->worker_wait_tail =
6653 NULL;
6654 }
6655 }
6656
6657 STMF_DEQUEUE_ITASK(w, itask);
6658 if (itask == NULL)
6659 break;
6660
6661 ASSERT((itask->itask_flags & ITASK_IN_FREE_LIST) == 0);
6662 task = itask->itask_task;
6663 DTRACE_PROBE2(worker__active, stmf_worker_t, w,
6664 scsi_task_t *, task);
6665 wait_queue = 0;
6666 abort_free = 0;
6667 mutex_exit(&w->worker_lock);
6668 mutex_enter(&itask->itask_mutex);
6669 mutex_enter(&w->worker_lock);
6670
6671 if (itask->itask_ncmds > 0) {
6672 curcmd = itask->itask_cmd_stack[itask->itask_ncmds - 1];
6673 } else {
6674 ASSERT(itask->itask_flags & ITASK_BEING_ABORTED);
6675 }
6676 if (itask->itask_flags & ITASK_BEING_ABORTED) {
6677 itask->itask_ncmds = 1;
6678 curcmd = itask->itask_cmd_stack[0] =
6679 ITASK_CMD_ABORT;
6680 goto out_itask_flag_loop;
6681 } else if ((curcmd & ITASK_CMD_MASK) == ITASK_CMD_NEW_TASK) {
6682 /*
6683 * set ITASK_KSTAT_IN_RUNQ, this flag
6684 * will not reset until task completed
6685 */
6686 itask->itask_flags |= ITASK_KNOWN_TO_LU |
6687 ITASK_KSTAT_IN_RUNQ;
6688 } else {
6689 goto out_itask_flag_loop;
6690 }
6691
6692 out_itask_flag_loop:
6693
6694 /*
6695 * Decide if this task needs to go to a queue and/or if
6696 * we can decrement the itask_cmd_stack.
6697 */
6698 if (curcmd == ITASK_CMD_ABORT) {
6699 if (itask->itask_flags & (ITASK_KNOWN_TO_LU |
6700 ITASK_KNOWN_TO_TGT_PORT)) {
6701 wait_queue = 1;
6702 } else {
6703 abort_free = 1;
6704 }
6705 } else if ((curcmd & ITASK_CMD_POLL) &&
6706 (itask->itask_poll_timeout > ddi_get_lbolt())) {
6707 wait_queue = 1;
6708 }
6709
6710 if (wait_queue) {
6729 w->worker_task_tail = itask;
6730 } else {
6731 atomic_and_32(&itask->itask_flags,
6732 ~ITASK_IN_WORKER_QUEUE);
6733 /*
6734 * This is where the queue depth should go down by
6735 * one but we delay that on purpose to account for
6736 * the call into the provider. The actual decrement
6737 * happens after the worker has done its job.
6738 */
6739 dec_qdepth = 1;
6740 itask->itask_waitq_time +=
6741 gethrtime() - itask->itask_waitq_enter_timestamp;
6742 }
6743
6744 /* We made it here means we are going to call LU */
6745 if ((itask->itask_flags & ITASK_DEFAULT_HANDLING) == 0)
6746 lu = task->task_lu;
6747 else
6748 lu = dlun0;
6749
6750 dbuf = itask->itask_dbufs[ITASK_CMD_BUF_NDX(curcmd)];
6751 mutex_exit(&w->worker_lock);
6752 curcmd &= ITASK_CMD_MASK;
6753 stmf_task_audit(itask, TE_PROCESS_CMD, curcmd, dbuf);
6754 mutex_exit(&itask->itask_mutex);
6755
6756 switch (curcmd) {
6757 case ITASK_CMD_NEW_TASK:
6758 iss = (stmf_i_scsi_session_t *)
6759 task->task_session->ss_stmf_private;
6760 stmf_itl_lu_new_task(itask);
6761 if (iss->iss_flags & ISS_LUN_INVENTORY_CHANGED) {
6762 if (stmf_handle_cmd_during_ic(itask)) {
6763 break;
6764 }
6765 }
6766 #ifdef DEBUG
6767 if (stmf_drop_task_counter > 0) {
6768 if (atomic_dec_32_nv(
6769 (uint32_t *)&stmf_drop_task_counter) == 1) {
6770 break;
6771 }
6772 }
6773 #endif
6774 DTRACE_PROBE1(scsi__task__start, scsi_task_t *, task);
6775 lu->lu_new_task(task, dbuf);
6776 break;
6777 case ITASK_CMD_DATA_XFER_DONE:
6778 lu->lu_dbuf_xfer_done(task, dbuf);
6779 break;
6780 case ITASK_CMD_STATUS_DONE:
6781 lu->lu_send_status_done(task);
6782 break;
6783 case ITASK_CMD_ABORT:
6784 if (abort_free) {
6785 mutex_enter(&itask->itask_mutex);
6786 stmf_task_free(task);
6787 } else {
6788 stmf_do_task_abort(task);
6789 }
6790 break;
6791 case ITASK_CMD_POLL_LU:
6792 if (!wait_queue) {
6793 lu->lu_task_poll(task);
6794 }
6795 break;
6796 case ITASK_CMD_POLL_LPORT:
6797 if (!wait_queue)
6798 task->task_lport->lport_task_poll(task);
6799 break;
6800 case ITASK_CMD_SEND_STATUS:
6801 /* case ITASK_CMD_XFER_DATA: */
6802 break;
6803 }
6804
6805 mutex_enter(&w->worker_lock);
6806 if (dec_qdepth) {
6807 w->worker_queue_depth--;
6808 }
6809 }
6810 if ((w->worker_flags & STMF_WORKER_TERMINATE) && (wait_timer == 0)) {
6811 if (w->worker_ref_count == 0)
6812 goto stmf_worker_loop;
6813 else {
6814 wait_timer = ddi_get_lbolt() + 1;
6815 wait_delta = 1;
6816 }
6817 }
6818 w->worker_flags &= ~STMF_WORKER_ACTIVE;
6819 if (wait_timer) {
6820 DTRACE_PROBE1(worker__timed__sleep, stmf_worker_t, w);
6821 (void) cv_reltimedwait(&w->worker_cv, &w->worker_lock,
6822 wait_delta, TR_CLOCK_TICK);
6823 } else {
6824 DTRACE_PROBE1(worker__sleep, stmf_worker_t, w);
6825 cv_wait(&w->worker_cv, &w->worker_lock);
6826 }
6827 DTRACE_PROBE1(worker__wakeup, stmf_worker_t, w);
6828 w->worker_flags |= STMF_WORKER_ACTIVE;
6829 goto stmf_worker_loop;
6830 }
6831
6832 /*
6833 * Fills out a dbuf from stmf_xfer_data_t (contained in the db_lu_private).
6834 * If all the data has been filled out, frees the xd and makes
6835 * db_lu_private NULL.
6836 */
6837 void
6838 stmf_xd_to_dbuf(stmf_data_buf_t *dbuf, int set_rel_off)
6839 {
6840 stmf_xfer_data_t *xd;
6841 uint8_t *p;
6842 int i;
6843 uint32_t s;
6844
6845 xd = (stmf_xfer_data_t *)dbuf->db_lu_private;
6846 dbuf->db_data_size = 0;
6847 if (set_rel_off)
6848 dbuf->db_relative_offset = xd->size_done;
6849 for (i = 0; i < dbuf->db_sglist_length; i++) {
6850 s = min(xd->size_left, dbuf->db_sglist[i].seg_length);
6851 p = &xd->buf[xd->size_done];
6928 stmf_abort(STMF_QUEUE_TASK_ABORT, task,
6929 STMF_ALLOC_FAILURE, NULL);
6930 return;
6931 }
6932 dbuf->db_lu_private = NULL;
6933
6934 p = dbuf->db_sglist[0].seg_addr;
6935
6936 /*
6937 * Standard inquiry handling only.
6938 */
6939
6940 bzero(p, inq_page_length + 5);
6941
6942 p[0] = DPQ_SUPPORTED | DTYPE_UNKNOWN;
6943 p[2] = 5;
6944 p[3] = 0x12;
6945 p[4] = inq_page_length;
6946 p[6] = 0x80;
6947
6948 (void) strncpy((char *)p+8, "NONE ", 8);
6949 (void) strncpy((char *)p+16, "NONE ", 16);
6950 (void) strncpy((char *)p+32, "NONE", 4);
6951
6952 dbuf->db_data_size = sz;
6953 dbuf->db_relative_offset = 0;
6954 dbuf->db_flags = DB_DIRECTION_TO_RPORT;
6955 (void) stmf_xfer_data(task, dbuf, 0);
6956
6957 return;
6958
6959 case SCMD_REPORT_LUNS:
6960 task->task_cmd_xfer_length =
6961 ((((uint32_t)task->task_cdb[6]) << 24) |
6962 (((uint32_t)task->task_cdb[7]) << 16) |
6963 (((uint32_t)task->task_cdb[8]) << 8) |
6964 ((uint32_t)task->task_cdb[9]));
6965
6966 if (task->task_additional_flags &
6967 TASK_AF_NO_EXPECTED_XFER_LENGTH) {
6968 task->task_expected_xfer_length =
6969 task->task_cmd_xfer_length;
6970 }
7140 case TM_CLEAR_TASK_SET:
7141 case TM_LUN_RESET:
7142 (void) stmf_lun_reset_poll(task->task_lu, task, 0);
7143 return;
7144 case TM_TARGET_RESET:
7145 case TM_TARGET_COLD_RESET:
7146 case TM_TARGET_WARM_RESET:
7147 stmf_target_reset_poll(task);
7148 return;
7149 }
7150 }
7151
7152 /* ARGSUSED */
7153 void
7154 stmf_dlun0_ctl(struct stmf_lu *lu, int cmd, void *arg)
7155 {
7156 /* This function will never be called */
7157 cmn_err(CE_WARN, "stmf_dlun0_ctl called with cmd %x", cmd);
7158 }
7159
7160 /* ARGSUSED */
7161 void
7162 stmf_dlun0_task_done(struct scsi_task *task)
7163 {
7164 }
7165
7166 void
7167 stmf_dlun_init()
7168 {
7169 stmf_i_lu_t *ilu;
7170
7171 dlun0 = stmf_alloc(STMF_STRUCT_STMF_LU, 0, 0);
7172 dlun0->lu_task_alloc = stmf_dlun0_task_alloc;
7173 dlun0->lu_new_task = stmf_dlun0_new_task;
7174 dlun0->lu_dbuf_xfer_done = stmf_dlun0_dbuf_done;
7175 dlun0->lu_send_status_done = stmf_dlun0_status_done;
7176 dlun0->lu_task_free = stmf_dlun0_task_free;
7177 dlun0->lu_abort = stmf_dlun0_abort;
7178 dlun0->lu_task_poll = stmf_dlun0_task_poll;
7179 dlun0->lu_task_done = stmf_dlun0_task_done;
7180 dlun0->lu_ctl = stmf_dlun0_ctl;
7181
7182 ilu = (stmf_i_lu_t *)dlun0->lu_stmf_private;
7183 ilu->ilu_cur_task_cntr = &ilu->ilu_task_cntr1;
7184 }
7185
7186 stmf_status_t
7187 stmf_dlun_fini()
7188 {
7189 stmf_i_lu_t *ilu;
7190
7191 ilu = (stmf_i_lu_t *)dlun0->lu_stmf_private;
7192
7193 ASSERT(ilu->ilu_ntasks == ilu->ilu_ntasks_free);
7194 if (ilu->ilu_ntasks) {
7195 stmf_i_scsi_task_t *itask, *nitask;
7196
7197 nitask = ilu->ilu_tasks;
7198 do {
7199 itask = nitask;
7424 {
7425 if (STMF_EVENT_ENABLED(ilport->ilport_event_hdl, eventid) &&
7426 (ilport->ilport_lport->lport_event_handler != NULL)) {
7427 ilport->ilport_lport->lport_event_handler(
7428 ilport->ilport_lport, eventid, arg, flags);
7429 }
7430 }
7431
7432 /*
7433 * With the possibility of having multiple itl sessions pointing to the
7434 * same itl_kstat_info, the ilu_kstat_lock mutex is used to synchronize
7435 * the kstat update of the ilu_kstat_io, itl_kstat_taskq and itl_kstat_lu_xfer
7436 * statistics.
7437 */
7438 void
7439 stmf_itl_task_start(stmf_i_scsi_task_t *itask)
7440 {
7441 stmf_itl_data_t *itl = itask->itask_itl_datap;
7442 scsi_task_t *task = itask->itask_task;
7443 stmf_i_lu_t *ilu;
7444 stmf_i_scsi_session_t *iss =
7445 itask->itask_task->task_session->ss_stmf_private;
7446 stmf_i_remote_port_t *irport = iss->iss_irport;
7447
7448 if (itl == NULL || task->task_lu == dlun0)
7449 return;
7450 ilu = (stmf_i_lu_t *)task->task_lu->lu_stmf_private;
7451 itask->itask_start_timestamp = gethrtime();
7452 itask->itask_xfer_done_timestamp = 0;
7453 if (ilu->ilu_kstat_io != NULL) {
7454 mutex_enter(ilu->ilu_kstat_io->ks_lock);
7455 stmf_update_kstat_lu_q(itask->itask_task, kstat_waitq_enter);
7456 mutex_exit(ilu->ilu_kstat_io->ks_lock);
7457 }
7458
7459 if (irport->irport_kstat_estat != NULL) {
7460 if (task->task_flags & TF_READ_DATA)
7461 atomic_inc_32(&irport->irport_nread_tasks);
7462 else if (task->task_flags & TF_WRITE_DATA)
7463 atomic_inc_32(&irport->irport_nwrite_tasks);
7464 }
7465
7466 stmf_update_kstat_lport_q(itask->itask_task, kstat_waitq_enter);
7467 }
7468
7469 void
7470 stmf_itl_lu_new_task(stmf_i_scsi_task_t *itask)
7471 {
7472 stmf_itl_data_t *itl = itask->itask_itl_datap;
7473 scsi_task_t *task = itask->itask_task;
7474 stmf_i_lu_t *ilu;
7475
7476 if (itl == NULL || task->task_lu == dlun0)
7477 return;
7478 ilu = (stmf_i_lu_t *)task->task_lu->lu_stmf_private;
7479 if (ilu->ilu_kstat_io != NULL) {
7480 mutex_enter(ilu->ilu_kstat_io->ks_lock);
7481 stmf_update_kstat_lu_q(itask->itask_task, kstat_waitq_to_runq);
7482 mutex_exit(ilu->ilu_kstat_io->ks_lock);
7483 }
7484
7485 stmf_update_kstat_lport_q(itask->itask_task, kstat_waitq_to_runq);
7486 }
7487
7488 void
7489 stmf_itl_task_done(stmf_i_scsi_task_t *itask)
7490 {
7491 stmf_itl_data_t *itl = itask->itask_itl_datap;
7492 scsi_task_t *task = itask->itask_task;
7493 stmf_i_lu_t *ilu;
7494
7495 itask->itask_done_timestamp = gethrtime();
7496
7497 if (itl == NULL || task->task_lu == dlun0)
7498 return;
7499 ilu = (stmf_i_lu_t *)task->task_lu->lu_stmf_private;
7500
7501 if (ilu->ilu_kstat_io == NULL)
7502 return;
7503
7504 stmf_update_kstat_rport_estat(task);
7505
7506 mutex_enter(ilu->ilu_kstat_io->ks_lock);
7507
7508 if (itask->itask_flags & ITASK_KSTAT_IN_RUNQ) {
7509 stmf_update_kstat_lu_q(task, kstat_runq_exit);
7510 mutex_exit(ilu->ilu_kstat_io->ks_lock);
7511 stmf_update_kstat_lport_q(task, kstat_runq_exit);
7512 } else {
7513 stmf_update_kstat_lu_q(task, kstat_waitq_exit);
7514 mutex_exit(ilu->ilu_kstat_io->ks_lock);
7515 stmf_update_kstat_lport_q(task, kstat_waitq_exit);
7516 }
7517 }
7518
7519 void
7520 stmf_lu_xfer_done(scsi_task_t *task, boolean_t read, hrtime_t elapsed_time)
7521 {
7522 stmf_i_scsi_task_t *itask = task->task_stmf_private;
7523
7524 if (task->task_lu == dlun0)
7525 return;
7526
7527 if (read) {
7528 atomic_add_64((uint64_t *)&itask->itask_lu_read_time,
7529 elapsed_time);
7530 } else {
7531 atomic_add_64((uint64_t *)&itask->itask_lu_write_time,
7532 elapsed_time);
7533 }
7534 }
7535
7536 static void
7537 stmf_lport_xfer_start(stmf_i_scsi_task_t *itask, stmf_data_buf_t *dbuf)
7538 {
7539 stmf_itl_data_t *itl = itask->itask_itl_datap;
7540
7541 if (itl == NULL)
7542 return;
7543
7544 DTRACE_PROBE2(scsi__xfer__start, scsi_task_t *, itask->itask_task,
7545 stmf_data_buf_t *, dbuf);
7546
7547 dbuf->db_xfer_start_timestamp = gethrtime();
7548 }
7549
7550 static void
7551 stmf_lport_xfer_done(stmf_i_scsi_task_t *itask, stmf_data_buf_t *dbuf)
7552 {
7553 stmf_itl_data_t *itl = itask->itask_itl_datap;
7554 hrtime_t elapsed_time;
7555 uint64_t xfer_size;
7556
7557 if (itl == NULL)
7558 return;
7559
7560 xfer_size = (dbuf->db_xfer_status == STMF_SUCCESS) ?
7561 dbuf->db_data_size : 0;
7562
7563 itask->itask_xfer_done_timestamp = gethrtime();
7564 elapsed_time = itask->itask_xfer_done_timestamp -
7565 dbuf->db_xfer_start_timestamp;
7566 if (dbuf->db_flags & DB_DIRECTION_TO_RPORT) {
7567 atomic_add_64((uint64_t *)&itask->itask_lport_read_time,
7568 elapsed_time);
7569 atomic_add_64((uint64_t *)&itask->itask_read_xfer,
7570 xfer_size);
7571 } else {
7572 atomic_add_64((uint64_t *)&itask->itask_lport_write_time,
7573 elapsed_time);
7574 atomic_add_64((uint64_t *)&itask->itask_write_xfer,
7575 xfer_size);
7576 }
7577
7578 DTRACE_PROBE3(scsi__xfer__end, scsi_task_t *, itask->itask_task,
7579 stmf_data_buf_t *, dbuf, hrtime_t, elapsed_time);
7580
7581 dbuf->db_xfer_start_timestamp = 0;
7582 }
7583
7584 void
7585 stmf_svc_init()
7586 {
7587 if (stmf_state.stmf_svc_flags & STMF_SVC_STARTED)
7588 return;
7589 list_create(&stmf_state.stmf_svc_list, sizeof (stmf_svc_req_t),
7590 offsetof(stmf_svc_req_t, svc_list_entry));
7591 stmf_state.stmf_svc_taskq = ddi_taskq_create(0, "STMF_SVC_TASKQ", 1,
7592 TASKQ_DEFAULTPRI, 0);
7593 (void) ddi_taskq_dispatch(stmf_state.stmf_svc_taskq,
7594 stmf_svc, 0, DDI_SLEEP);
7595 }
7596
7597 stmf_status_t
7598 stmf_svc_fini()
7599 {
7600 uint32_t i;
7601
7602 mutex_enter(&stmf_state.stmf_lock);
7603 if (stmf_state.stmf_svc_flags & STMF_SVC_STARTED) {
7604 stmf_state.stmf_svc_flags |= STMF_SVC_TERMINATE;
7605 cv_signal(&stmf_state.stmf_cv);
7606 }
7607 mutex_exit(&stmf_state.stmf_lock);
7608
7609 /* Wait for 5 seconds */
7610 for (i = 0; i < 500; i++) {
7611 if (stmf_state.stmf_svc_flags & STMF_SVC_STARTED)
7612 delay(drv_usectohz(10000));
7613 else
7614 break;
7615 }
7616 if (i == 500)
7617 return (STMF_BUSY);
7618
7619 list_destroy(&stmf_state.stmf_svc_list);
7620 ddi_taskq_destroy(stmf_state.stmf_svc_taskq);
7621
7622 return (STMF_SUCCESS);
7623 }
7624
7625 struct stmf_svc_clocks {
7626 clock_t drain_start, drain_next;
7627 clock_t timing_start, timing_next;
7628 clock_t worker_delay;
7629 };
7630
7631 /* ARGSUSED */
7632 void
7633 stmf_svc(void *arg)
7634 {
7635 stmf_svc_req_t *req;
7636 stmf_lu_t *lu;
7637 stmf_i_lu_t *ilu;
7638 stmf_local_port_t *lport;
7639 struct stmf_svc_clocks clks = { 0 };
7640
7641 mutex_enter(&stmf_state.stmf_lock);
7642 stmf_state.stmf_svc_flags |= STMF_SVC_STARTED | STMF_SVC_ACTIVE;
7643
7644 while (!(stmf_state.stmf_svc_flags & STMF_SVC_TERMINATE)) {
7645 if (list_is_empty(&stmf_state.stmf_svc_list)) {
7646 stmf_svc_timeout(&clks);
7647 continue;
7648 }
7649
7650 /*
7651 * Pop the front request from the active list. After this,
7652 * the request will no longer be referenced by global state,
7653 * so it should be safe to access it without holding the
7654 * stmf state lock.
7655 */
7656 req = list_remove_head(&stmf_state.stmf_svc_list);
7657 if (req == NULL)
7658 continue;
7659
7660 switch (req->svc_cmd) {
7661 case STMF_CMD_LPORT_ONLINE:
7662 /* Fallthrough */
7663 case STMF_CMD_LPORT_OFFLINE:
7664 mutex_exit(&stmf_state.stmf_lock);
7665 lport = (stmf_local_port_t *)req->svc_obj;
7666 lport->lport_ctl(lport, req->svc_cmd, &req->svc_info);
7667 break;
7668 case STMF_CMD_LU_ONLINE:
7669 mutex_exit(&stmf_state.stmf_lock);
7670 lu = (stmf_lu_t *)req->svc_obj;
7671 lu->lu_ctl(lu, req->svc_cmd, &req->svc_info);
7672 break;
7673 case STMF_CMD_LU_OFFLINE:
7674 /* Remove all mappings of this LU */
7675 stmf_session_lu_unmapall((stmf_lu_t *)req->svc_obj);
7676 /* Kill all the pending I/Os for this LU */
7677 mutex_exit(&stmf_state.stmf_lock);
7678 stmf_task_lu_killall((stmf_lu_t *)req->svc_obj, NULL,
7679 STMF_ABORTED);
7710 if (stmf_state.stmf_nlus &&
7711 ((!clks->timing_next) || (ddi_get_lbolt() >= clks->timing_next))) {
7712 if (!stmf_state.stmf_svc_ilu_timing) {
7713 /* we are starting a new round */
7714 stmf_state.stmf_svc_ilu_timing =
7715 stmf_state.stmf_ilulist;
7716 clks->timing_start = ddi_get_lbolt();
7717 }
7718
7719 stmf_check_ilu_timing();
7720 if (!stmf_state.stmf_svc_ilu_timing) {
7721 /* we finished a complete round */
7722 clks->timing_next =
7723 clks->timing_start + drv_usectohz(5*1000*1000);
7724 } else {
7725 /* we still have some ilu items to check */
7726 clks->timing_next =
7727 ddi_get_lbolt() + drv_usectohz(1*1000*1000);
7728 }
7729
7730 if (!list_is_empty(&stmf_state.stmf_svc_list))
7731 return;
7732 }
7733
7734 /* Check if there are free tasks to clear */
7735 if (stmf_state.stmf_nlus &&
7736 ((!clks->drain_next) || (ddi_get_lbolt() >= clks->drain_next))) {
7737 if (!stmf_state.stmf_svc_ilu_draining) {
7738 /* we are starting a new round */
7739 stmf_state.stmf_svc_ilu_draining =
7740 stmf_state.stmf_ilulist;
7741 clks->drain_start = ddi_get_lbolt();
7742 }
7743
7744 stmf_check_freetask();
7745 if (!stmf_state.stmf_svc_ilu_draining) {
7746 /* we finished a complete round */
7747 clks->drain_next =
7748 clks->drain_start + drv_usectohz(10*1000*1000);
7749 } else {
7750 /* we still have some ilu items to check */
7751 clks->drain_next =
7752 ddi_get_lbolt() + drv_usectohz(1*1000*1000);
7753 }
7754
7755 if (!list_is_empty(&stmf_state.stmf_svc_list))
7756 return;
7757 }
7758
7759 /* Check if any active session got its 1st LUN */
7760 if (stmf_state.stmf_process_initial_luns) {
7761 int stmf_level = 0;
7762 int port_level;
7763
7764 for (ilport = stmf_state.stmf_ilportlist; ilport;
7765 ilport = next_ilport) {
7766 int ilport_lock_held;
7767 next_ilport = ilport->ilport_next;
7768
7769 if ((ilport->ilport_flags &
7770 ILPORT_SS_GOT_INITIAL_LUNS) == 0)
7771 continue;
7772
7773 port_level = 0;
7774 rw_enter(&ilport->ilport_lock, RW_READER);
7775 ilport_lock_held = 1;
7776
7777 for (iss = ilport->ilport_ss_list; iss;
7778 iss = iss->iss_next) {
7865 stmf_svc_req_t *req;
7866 int s;
7867
7868 ASSERT(!mutex_owned(&stmf_state.stmf_lock));
7869 s = sizeof (stmf_svc_req_t);
7870 if (info->st_additional_info) {
7871 s += strlen(info->st_additional_info) + 1;
7872 }
7873 req = kmem_zalloc(s, KM_SLEEP);
7874
7875 req->svc_cmd = cmd;
7876 req->svc_obj = obj;
7877 req->svc_info.st_rflags = info->st_rflags;
7878 if (info->st_additional_info) {
7879 req->svc_info.st_additional_info = (char *)(GET_BYTE_OFFSET(req,
7880 sizeof (stmf_svc_req_t)));
7881 (void) strcpy(req->svc_info.st_additional_info,
7882 info->st_additional_info);
7883 }
7884 req->svc_req_alloc_size = s;
7885
7886 mutex_enter(&stmf_state.stmf_lock);
7887 list_insert_tail(&stmf_state.stmf_svc_list, req);
7888 if ((stmf_state.stmf_svc_flags & STMF_SVC_ACTIVE) == 0) {
7889 cv_signal(&stmf_state.stmf_cv);
7890 }
7891 mutex_exit(&stmf_state.stmf_lock);
7892 }
7893
7894 static void
7895 stmf_svc_kill_obj_requests(void *obj)
7896 {
7897 stmf_svc_req_t *req;
7898
7899 ASSERT(mutex_owned(&stmf_state.stmf_lock));
7900
7901 for (req = list_head(&stmf_state.stmf_svc_list); req != NULL;
7902 req = list_next(&stmf_state.stmf_svc_list, req)) {
7903 if (req->svc_obj == obj) {
7904 list_remove(&stmf_state.stmf_svc_list, req);
7905 kmem_free(req, req->svc_req_alloc_size);
7906 }
7907 }
7908 }
7909
7910 void
7911 stmf_trace(caddr_t ident, const char *fmt, ...)
7912 {
7913 va_list args;
7914 char tbuf[160];
7915 int len;
7916
7917 if (!stmf_trace_on)
7918 return;
7919 len = snprintf(tbuf, 158, "%s:%07lu: ", ident ? ident : "",
7920 ddi_get_lbolt());
7921 va_start(args, fmt);
7922 len += vsnprintf(tbuf + len, 158 - len, fmt, args);
7923 va_end(args);
7924
7925 if (len > 158) {
7931 mutex_enter(&trace_buf_lock);
7932 bcopy(tbuf, &stmf_trace_buf[trace_buf_curndx], len+1);
7933 trace_buf_curndx += len;
7934 if (trace_buf_curndx > (trace_buf_size - 320))
7935 trace_buf_curndx = 0;
7936 mutex_exit(&trace_buf_lock);
7937 }
7938
7939 void
7940 stmf_trace_clear()
7941 {
7942 if (!stmf_trace_on)
7943 return;
7944 mutex_enter(&trace_buf_lock);
7945 trace_buf_curndx = 0;
7946 if (trace_buf_size > 0)
7947 stmf_trace_buf[0] = 0;
7948 mutex_exit(&trace_buf_lock);
7949 }
7950
7951 /*
7952 * NOTE: Due to lock order problems that are not possible to fix this
7953 * method drops and reacquires the itask_mutex around the call to stmf_ctl.
7954 * Another possible work around would be to use a dispatch queue and have
7955 * the call to stmf_ctl run on another thread that's not holding the
7956 * itask_mutex. The problem with that approach is that it's difficult to
7957 * determine what impact an asynchronous change would have on the system state.
7958 */
7959 static void
7960 stmf_abort_task_offline(scsi_task_t *task, int offline_lu, char *info)
7961 {
7962 stmf_state_change_info_t change_info;
7963 void *ctl_private;
7964 uint32_t ctl_cmd;
7965 int msg = 0;
7966 stmf_i_scsi_task_t *itask =
7967 (stmf_i_scsi_task_t *)task->task_stmf_private;
7968
7969 stmf_trace("FROM STMF", "abort_task_offline called for %s: %s",
7970 offline_lu ? "LU" : "LPORT", info ? info : "no additional info");
7971 change_info.st_additional_info = info;
7972 ASSERT(mutex_owned(&itask->itask_mutex));
7973
7974 if (offline_lu) {
7975 change_info.st_rflags = STMF_RFLAG_RESET |
7976 STMF_RFLAG_LU_ABORT;
7977 ctl_private = task->task_lu;
7978 if (((stmf_i_lu_t *)
7979 task->task_lu->lu_stmf_private)->ilu_state ==
7980 STMF_STATE_ONLINE) {
7981 msg = 1;
7982 }
7983 ctl_cmd = STMF_CMD_LU_OFFLINE;
7984 } else {
7985 change_info.st_rflags = STMF_RFLAG_RESET |
7986 STMF_RFLAG_LPORT_ABORT;
7987 ctl_private = task->task_lport;
7988 if (((stmf_i_local_port_t *)
7989 task->task_lport->lport_stmf_private)->ilport_state ==
7990 STMF_STATE_ONLINE) {
7991 msg = 1;
7992 }
7993 ctl_cmd = STMF_CMD_LPORT_OFFLINE;
7994 }
7995
7996 if (msg) {
7997 stmf_trace(0, "Calling stmf_ctl to offline %s : %s",
7998 offline_lu ? "LU" : "LPORT", info ? info :
7999 "<no additional info>");
8000 }
8001 mutex_exit(&itask->itask_mutex);
8002 (void) stmf_ctl(ctl_cmd, ctl_private, &change_info);
8003 mutex_enter(&itask->itask_mutex);
8004 }
8005
8006 static char
8007 stmf_ctoi(char c)
8008 {
8009 if ((c >= '0') && (c <= '9'))
8010 c -= '0';
8011 else if ((c >= 'A') && (c <= 'F'))
8012 c = c - 'A' + 10;
8013 else if ((c >= 'a') && (c <= 'f'))
8014 c = c - 'a' + 10;
8015 else
8016 c = -1;
8017 return (c);
8018 }
8019
8020 /* Convert from Hex value in ASCII format to the equivalent bytes */
8021 static boolean_t
8022 stmf_base16_str_to_binary(char *c, int dplen, uint8_t *dp)
8023 {
8040
8041 boolean_t
8042 stmf_scsilib_tptid_validate(scsi_transport_id_t *tptid, uint32_t total_sz,
8043 uint16_t *tptid_sz)
8044 {
8045 uint16_t tpd_len = SCSI_TPTID_SIZE;
8046
8047 if (tptid_sz)
8048 *tptid_sz = 0;
8049 if (total_sz < sizeof (scsi_transport_id_t))
8050 return (B_FALSE);
8051
8052 switch (tptid->protocol_id) {
8053
8054 case PROTOCOL_FIBRE_CHANNEL:
8055 /* FC Transport ID validation checks. SPC3 rev23, Table 284 */
8056 if (total_sz < tpd_len || tptid->format_code != 0)
8057 return (B_FALSE);
8058 break;
8059
8060 case PROTOCOL_iSCSI: /* CSTYLED */
8061 {
8062 iscsi_transport_id_t *iscsiid;
8063 uint16_t adn_len, name_len;
8064
8065 /* Check for valid format code, SPC3 rev 23 Table 288 */
8066 if ((total_sz < tpd_len) ||
8067 (tptid->format_code != 0 && tptid->format_code != 1))
8068 return (B_FALSE);
8069
8070 iscsiid = (iscsi_transport_id_t *)tptid;
8071 adn_len = READ_SCSI16(iscsiid->add_len, uint16_t);
8072 tpd_len = sizeof (iscsi_transport_id_t) + adn_len - 1;
8073
8074 /*
8075 * iSCSI Transport ID validation checks.
8076 * As per SPC3 rev 23 Section 7.5.4.6 and Table 289 & Table 290
8077 */
8078 if (adn_len < 20 || (adn_len % 4 != 0))
8079 return (B_FALSE);
8080
8085 /* If the format_code is 1 check for ISID seperator */
8086 if ((tptid->format_code == 1) && (strstr(iscsiid->iscsi_name,
8087 SCSI_TPTID_ISCSI_ISID_SEPERATOR) == NULL))
8088 return (B_FALSE);
8089
8090 }
8091 break;
8092
8093 case PROTOCOL_SRP:
8094 /* SRP Transport ID validation checks. SPC3 rev23, Table 287 */
8095 if (total_sz < tpd_len || tptid->format_code != 0)
8096 return (B_FALSE);
8097 break;
8098
8099 case PROTOCOL_PARALLEL_SCSI:
8100 case PROTOCOL_SSA:
8101 case PROTOCOL_IEEE_1394:
8102 case PROTOCOL_SAS:
8103 case PROTOCOL_ADT:
8104 case PROTOCOL_ATAPI:
8105 default: /* CSTYLED */
8106 {
8107 stmf_dflt_scsi_tptid_t *dflttpd;
8108
8109 tpd_len = sizeof (stmf_dflt_scsi_tptid_t);
8110 if (total_sz < tpd_len)
8111 return (B_FALSE);
8112 dflttpd = (stmf_dflt_scsi_tptid_t *)tptid;
8113 tpd_len = tpd_len + SCSI_READ16(&dflttpd->ident_len) - 1;
8114 if (total_sz < tpd_len)
8115 return (B_FALSE);
8116 }
8117 break;
8118 }
8119 if (tptid_sz)
8120 *tptid_sz = tpd_len;
8121 return (B_TRUE);
8122 }
8123
8124 boolean_t
8125 stmf_scsilib_tptid_compare(scsi_transport_id_t *tpd1,
8126 scsi_transport_id_t *tpd2)
8127 {
8128 if ((tpd1->protocol_id != tpd2->protocol_id) ||
8129 (tpd1->format_code != tpd2->format_code))
8130 return (B_FALSE);
8131
8132 switch (tpd1->protocol_id) {
8133
8134 case PROTOCOL_iSCSI: /* CSTYLED */
8135 {
8136 iscsi_transport_id_t *iscsitpd1, *iscsitpd2;
8137 uint16_t len;
8138
8139 iscsitpd1 = (iscsi_transport_id_t *)tpd1;
8140 iscsitpd2 = (iscsi_transport_id_t *)tpd2;
8141 len = SCSI_READ16(&iscsitpd1->add_len);
8142 if ((memcmp(iscsitpd1->add_len, iscsitpd2->add_len, 2) != 0) ||
8143 (memcmp(iscsitpd1->iscsi_name, iscsitpd2->iscsi_name, len)
8144 != 0))
8145 return (B_FALSE);
8146 }
8147 break;
8148
8149 case PROTOCOL_SRP: /* CSTYLED */
8150 {
8151 scsi_srp_transport_id_t *srptpd1, *srptpd2;
8152
8153 srptpd1 = (scsi_srp_transport_id_t *)tpd1;
8154 srptpd2 = (scsi_srp_transport_id_t *)tpd2;
8155 if (memcmp(srptpd1->srp_name, srptpd2->srp_name,
8156 sizeof (srptpd1->srp_name)) != 0)
8157 return (B_FALSE);
8158 }
8159 break;
8160
8161 case PROTOCOL_FIBRE_CHANNEL: /* CSTYLED */
8162 {
8163 scsi_fc_transport_id_t *fctpd1, *fctpd2;
8164
8165 fctpd1 = (scsi_fc_transport_id_t *)tpd1;
8166 fctpd2 = (scsi_fc_transport_id_t *)tpd2;
8167 if (memcmp(fctpd1->port_name, fctpd2->port_name,
8168 sizeof (fctpd1->port_name)) != 0)
8169 return (B_FALSE);
8170 }
8171 break;
8172
8173 case PROTOCOL_PARALLEL_SCSI:
8174 case PROTOCOL_SSA:
8175 case PROTOCOL_IEEE_1394:
8176 case PROTOCOL_SAS:
8177 case PROTOCOL_ADT:
8178 case PROTOCOL_ATAPI:
8179 default: /* CSTYLED */
8180 {
8181 stmf_dflt_scsi_tptid_t *dflt1, *dflt2;
8182 uint16_t len;
8183
8184 dflt1 = (stmf_dflt_scsi_tptid_t *)tpd1;
8185 dflt2 = (stmf_dflt_scsi_tptid_t *)tpd2;
8186 len = SCSI_READ16(&dflt1->ident_len);
8187 if ((memcmp(dflt1->ident_len, dflt2->ident_len, 2) != 0) ||
8188 (memcmp(dflt1->ident, dflt2->ident, len) != 0))
8189 return (B_FALSE);
8190 }
8191 break;
8192 }
8193 return (B_TRUE);
8194 }
8195
8196 /*
8197 * Changes devid_desc to corresponding TransportID format
8198 * Returns :- pointer to stmf_remote_port_t
8199 * Note :- Allocates continous memory for stmf_remote_port_t and TransportID,
8270 ident_len = devid->ident_length;
8271 sz = ALIGNED_TO_8BYTE_BOUNDARY(sizeof (stmf_dflt_scsi_tptid_t) +
8272 ident_len - 1);
8273 rpt = stmf_remote_port_alloc(sz);
8274 rpt->rport_tptid->format_code = 0;
8275 rpt->rport_tptid->protocol_id = devid->protocol_id;
8276 dflt_tpd = (stmf_dflt_scsi_tptid_t *)rpt->rport_tptid;
8277 SCSI_WRITE16(dflt_tpd->ident_len, ident_len);
8278 (void) memcpy(dflt_tpd->ident, devid->ident, ident_len);
8279 break;
8280 }
8281 return (rpt);
8282
8283 devid_to_remote_port_fail:
8284 stmf_remote_port_free(rpt);
8285 return (NULL);
8286
8287 }
8288
8289 stmf_remote_port_t *
8290 stmf_remote_port_alloc(uint16_t tptid_sz)
8291 {
8292 stmf_remote_port_t *rpt;
8293 rpt = (stmf_remote_port_t *)kmem_zalloc(
8294 sizeof (stmf_remote_port_t) + tptid_sz, KM_SLEEP);
8295 rpt->rport_tptid_sz = tptid_sz;
8296 rpt->rport_tptid = (scsi_transport_id_t *)(rpt + 1);
8297 return (rpt);
8298 }
8299
8300 void
8301 stmf_remote_port_free(stmf_remote_port_t *rpt)
8302 {
8303 /*
8304 * Note: stmf_scsilib_devid_to_remote_port() function allocates
8305 * remote port structures for all transports in the same way, So
8306 * it is safe to deallocate it in a protocol independent manner.
8307 * If any of the allocation method changes, corresponding changes
8308 * need to be made here too.
8309 */
8310 kmem_free(rpt, sizeof (stmf_remote_port_t) + rpt->rport_tptid_sz);
8311 }
8312
8313 stmf_lu_t *
8314 stmf_check_and_hold_lu(scsi_task_t *task, uint8_t *guid)
8315 {
8316 stmf_i_scsi_session_t *iss;
8317 stmf_lu_t *lu;
8318 stmf_i_lu_t *ilu = NULL;
8319 stmf_lun_map_t *sm;
8320 stmf_lun_map_ent_t *lme;
8321 int i;
8322
8323 iss = (stmf_i_scsi_session_t *)task->task_session->ss_stmf_private;
8324 rw_enter(iss->iss_lockp, RW_READER);
8325 sm = iss->iss_sm;
8326
8327 for (i = 0; i < sm->lm_nentries; i++) {
8328 if (sm->lm_plus[i] == NULL)
8329 continue;
8330 lme = (stmf_lun_map_ent_t *)sm->lm_plus[i];
8331 lu = lme->ent_lu;
8332 if (bcmp(lu->lu_id->ident, guid, 16) == 0) {
8333 break;
8334 }
8335 lu = NULL;
8336 }
8337
8338 if (!lu) {
8339 goto hold_lu_done;
8340 }
8341
8342 ilu = lu->lu_stmf_private;
8343 mutex_enter(&ilu->ilu_task_lock);
8344 ilu->ilu_additional_ref++;
8345 mutex_exit(&ilu->ilu_task_lock);
8346
8347 hold_lu_done:
8348 rw_exit(iss->iss_lockp);
8349 return (lu);
8350 }
8351
8352 void
8353 stmf_release_lu(stmf_lu_t *lu)
8354 {
8355 stmf_i_lu_t *ilu;
8356
8357 ilu = lu->lu_stmf_private;
8358 ASSERT(ilu->ilu_additional_ref != 0);
8359 mutex_enter(&ilu->ilu_task_lock);
8360 ilu->ilu_additional_ref--;
8361 mutex_exit(&ilu->ilu_task_lock);
8362 }
8363
8364 int
8365 stmf_is_task_being_aborted(scsi_task_t *task)
8366 {
8367 stmf_i_scsi_task_t *itask;
8368
8369 itask = (stmf_i_scsi_task_t *)task->task_stmf_private;
8370 if (itask->itask_flags & ITASK_BEING_ABORTED)
8371 return (1);
8372
8373 return (0);
8374 }
8375
8376 volatile boolean_t stmf_pgr_aptpl_always = B_FALSE;
8377
8378 boolean_t
8379 stmf_is_pgr_aptpl_always()
8380 {
8381 return (stmf_pgr_aptpl_always);
8382 }
|