xref: /xnu-12377.41.6/bsd/skywalk/nexus/flowswitch/flow/flow_owner.c (revision bbb1b6f9e71b8cdde6e5cd6f4841f207dee3d828)
1 /*
2  * Copyright (c) 2016-2021 Apple Inc. All rights reserved.
3  *
4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5  *
6  * This file contains Original Code and/or Modifications of Original Code
7  * as defined in and that are subject to the Apple Public Source License
8  * Version 2.0 (the 'License'). You may not use this file except in
9  * compliance with the License. The rights granted to you under the License
10  * may not be used to create, or enable the creation or redistribution of,
11  * unlawful or unlicensed copies of an Apple operating system, or to
12  * circumvent, violate, or enable the circumvention or violation of, any
13  * terms of an Apple operating system software license agreement.
14  *
15  * Please obtain a copy of the License at
16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
17  *
18  * The Original Code and all software distributed under the License are
19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23  * Please see the License for the specific language governing rights and
24  * limitations under the License.
25  *
26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27  */
28 
29 #include <skywalk/os_skywalk_private.h>
30 #include <skywalk/nexus/flowswitch/fsw_var.h>
31 #include <skywalk/nexus/flowswitch/flow/flow_var.h>
32 
33 static uint32_t flow_owner_bucket_purge_common(struct flow_owner_bucket *,
34     nexus_port_t, boolean_t);
35 static int fo_cmp(const struct flow_owner *, const struct flow_owner *);
36 static struct flow_owner *fo_alloc(boolean_t);
37 static void fo_free(struct flow_owner *);
38 
39 static LCK_GRP_DECLARE(flow_owner_lock_group, "sk_flow_owner_lock");
40 static LCK_ATTR_DECLARE(flow_owner_lock_attr, 0, 0);
41 
42 RB_GENERATE_PREV(flow_owner_tree, flow_owner, fo_link, fo_cmp);
43 
44 KALLOC_TYPE_VAR_DEFINE(KT_SK_FOB, struct flow_owner_bucket, KT_DEFAULT);
45 
46 struct flow_owner_bucket *
47 __sized_by(*tot_sz)
flow_owner_buckets_alloc(size_t fob_cnt,size_t * fob_sz,size_t * tot_sz)48 flow_owner_buckets_alloc(size_t fob_cnt, size_t * fob_sz, size_t * tot_sz){
49 	size_t cache_sz = skmem_cpu_cache_line_size();
50 	struct flow_owner_bucket *fob;
51 	size_t fob_tot_sz;
52 
53 	/* each bucket is CPU cache-aligned */
54 	*fob_sz = P2ROUNDUP(sizeof(*fob), cache_sz);
55 	*tot_sz = fob_tot_sz = fob_cnt * (*fob_sz);
56 	fob = sk_alloc_type_hash(KT_SK_FOB, fob_tot_sz, Z_WAITOK, skmem_tag_fsw_fob_hash);
57 	if (__improbable(fob == NULL)) {
58 		return NULL;
59 	}
60 
61 #if !KASAN_CLASSIC
62 	/*
63 	 * except in KASAN_CLASSIC mode, kalloc will always maintain cacheline
64 	 * size alignment if the requested size is a multiple of a cacheline
65 	 * size (this is true for any size that is a power of two from 16 to
66 	 * PAGE_SIZE).
67 	 *
68 	 * Because this is an optimization only, it is OK to leave KASAN_CLASSIC
69 	 * not respect this.
70 	 */
71 	ASSERT(IS_P2ALIGNED(fob, cache_sz));
72 #endif
73 
74 	SK_DF(SK_VERB_MEM, "fob %p fob_cnt %zu fob_sz %zu "
75 	    "(total %zu bytes) ALLOC", SK_KVA(fob), fob_cnt,
76 	    *fob_sz, fob_tot_sz);
77 
78 	return fob;
79 }
80 
81 void
flow_owner_buckets_free(struct flow_owner_bucket * fob,size_t tot_sz)82 flow_owner_buckets_free(struct flow_owner_bucket *fob, size_t tot_sz)
83 {
84 	SK_DF(SK_VERB_MEM, "fob %p FREE", SK_KVA(fob));
85 	sk_free_type_hash(KT_SK_FOB, tot_sz, fob);
86 }
87 
88 void
flow_owner_bucket_init(struct flow_owner_bucket * fob)89 flow_owner_bucket_init(struct flow_owner_bucket *fob)
90 {
91 #if !KASAN_CLASSIC
92 	ASSERT(IS_P2ALIGNED(fob, skmem_cpu_cache_line_size()));
93 #endif /* !KASAN_CLASSIC */
94 	lck_mtx_init(&fob->fob_lock, &flow_owner_lock_group,
95 	    &flow_owner_lock_attr);
96 	RB_INIT(&fob->fob_owner_head);
97 }
98 
99 void
flow_owner_bucket_destroy(struct flow_owner_bucket * fob)100 flow_owner_bucket_destroy(struct flow_owner_bucket *fob)
101 {
102 	/*
103 	 * In the event we are called as part of the nexus destructor,
104 	 * we need to wait until all threads have exited the flow close
105 	 * critical section, and that the flow_owner_bucket is empty.
106 	 * By the time we get here, the module initiating the request
107 	 * (e.g. NECP) has been quiesced, so any flow open requests would
108 	 * have been rejected.
109 	 */
110 	FOB_LOCK(fob);
111 	while (!RB_EMPTY(&fob->fob_owner_head)) {
112 		SK_ERR("waiting for fob %p to go idle", SK_KVA(fob));
113 		if (++(fob->fob_dtor_waiters) == 0) {   /* wraparound */
114 			fob->fob_dtor_waiters++;
115 		}
116 		(void) msleep(&fob->fob_dtor_waiters, &fob->fob_lock,
117 		    (PZERO - 1), __FUNCTION__, NULL);
118 	}
119 	while (fob->fob_busy_flags & FOBF_CLOSE_BUSY) {
120 		if (++(fob->fob_close_waiters) == 0) {  /* wraparound */
121 			fob->fob_close_waiters++;
122 		}
123 		(void) msleep(&fob->fob_close_waiters, &fob->fob_lock,
124 		    (PZERO - 1), __FUNCTION__, NULL);
125 	}
126 	ASSERT(RB_EMPTY(&fob->fob_owner_head));
127 	ASSERT(!(fob->fob_busy_flags & FOBF_OPEN_BUSY));
128 	ASSERT(!(fob->fob_busy_flags & FOBF_CLOSE_BUSY));
129 	FOB_UNLOCK(fob);
130 	lck_mtx_destroy(&fob->fob_lock, &flow_owner_lock_group);
131 }
132 
133 static uint32_t
flow_owner_bucket_purge_common(struct flow_owner_bucket * fob,nexus_port_t nx_port,boolean_t if_idle)134 flow_owner_bucket_purge_common(struct flow_owner_bucket *fob,
135     nexus_port_t nx_port, boolean_t if_idle)
136 {
137 	/* called by flow_owner_bucket_purge_all()? */
138 	boolean_t locked = (nx_port == NEXUS_PORT_ANY);
139 	struct flow_owner *fo, *tfo;
140 	struct flow_entry *fe, *tfe;
141 	uint32_t cnt = 0;
142 
143 	if (!locked) {
144 		FOB_LOCK(fob);
145 	}
146 	FOB_LOCK_ASSERT_HELD(fob);
147 
148 	RB_FOREACH_SAFE(fo, flow_owner_tree, &fob->fob_owner_head, tfo) {
149 		if (fo->fo_nx_port != nx_port && nx_port != NEXUS_PORT_ANY) {
150 			continue;
151 		}
152 
153 		if (!if_idle || nx_port == NEXUS_PORT_ANY) {
154 			RB_FOREACH_SAFE(fe, flow_entry_id_tree,
155 			    &fo->fo_flow_entry_id_head, tfe) {
156 				ASSERT(fe->fe_nx_port == fo->fo_nx_port);
157 				flow_entry_retain(fe);
158 				flow_entry_destroy(fo, fe, FALSE, NULL);
159 			}
160 		}
161 
162 		ASSERT(nx_port != NEXUS_PORT_ANY ||
163 		    RB_EMPTY(&fo->fo_flow_entry_id_head));
164 
165 		if (RB_EMPTY(&fo->fo_flow_entry_id_head)) {
166 			flow_owner_free(fob, fo);
167 			++cnt;
168 		} else if (nx_port != NEXUS_PORT_ANY) {
169 			/* let ms_flow_unbind() know this port is gone */
170 			fo->fo_nx_port_destroyed = TRUE;
171 			VERIFY(fo->fo_nx_port_na == NULL);
172 		}
173 	}
174 
175 	if (!locked) {
176 		FOB_UNLOCK(fob);
177 	}
178 
179 	return cnt;
180 }
181 
182 void
flow_owner_bucket_purge_all(struct flow_owner_bucket * fob)183 flow_owner_bucket_purge_all(struct flow_owner_bucket *fob)
184 {
185 	(void) flow_owner_bucket_purge_common(fob, NEXUS_PORT_ANY, TRUE);
186 }
187 
188 static uint32_t
flow_owner_bucket_activate_nx_port_common(struct flow_owner_bucket * fob,nexus_port_t nx_port,struct nexus_adapter * nx_port_na,na_activate_mode_t mode)189 flow_owner_bucket_activate_nx_port_common(struct flow_owner_bucket *fob,
190     nexus_port_t nx_port, struct nexus_adapter *nx_port_na,
191     na_activate_mode_t mode)
192 {
193 	struct flow_owner *fo;
194 	struct flow_entry *fe;
195 	uint32_t cnt = 0;
196 
197 	VERIFY(nx_port != NEXUS_PORT_ANY);
198 	FOB_LOCK(fob);
199 
200 	RB_FOREACH(fo, flow_owner_tree, &fob->fob_owner_head) {
201 		if (fo->fo_nx_port_destroyed || (fo->fo_nx_port != nx_port)) {
202 			continue;
203 		}
204 
205 		if (mode == NA_ACTIVATE_MODE_ON) {
206 			VERIFY(fo->fo_nx_port_na == NULL);
207 			*(struct nexus_adapter **)(uintptr_t)&fo->fo_nx_port_na = nx_port_na;
208 		}
209 
210 		RB_FOREACH(fe, flow_entry_id_tree,
211 		    &fo->fo_flow_entry_id_head) {
212 			if (fe->fe_flags & FLOWENTF_TORN_DOWN) {
213 				continue;
214 			}
215 			VERIFY(fe->fe_nx_port == fo->fo_nx_port);
216 			if (fe->fe_adv_idx != FLOWADV_IDX_NONE) {
217 				if (mode == NA_ACTIVATE_MODE_ON) {
218 					na_flowadv_entry_alloc(
219 						fo->fo_nx_port_na, fe->fe_uuid,
220 						fe->fe_adv_idx, fe->fe_flowid);
221 				} else if (fo->fo_nx_port_na != NULL) {
222 					na_flowadv_entry_free(fo->fo_nx_port_na,
223 					    fe->fe_uuid, fe->fe_adv_idx,
224 					    fe->fe_flowid);
225 				}
226 			}
227 		}
228 
229 		if (mode != NA_ACTIVATE_MODE_ON && fo->fo_nx_port_na != NULL) {
230 			*(struct nexus_adapter **)(uintptr_t)&fo->fo_nx_port_na = NULL;
231 		}
232 
233 		++cnt;
234 	}
235 
236 	FOB_UNLOCK(fob);
237 	return cnt;
238 }
239 
240 uint32_t
flow_owner_activate_nexus_port(struct flow_mgr * fm,boolean_t pid_bound,pid_t pid,nexus_port_t nx_port,struct nexus_adapter * nx_port_na,na_activate_mode_t mode)241 flow_owner_activate_nexus_port(struct flow_mgr *fm,
242     boolean_t pid_bound, pid_t pid, nexus_port_t nx_port,
243     struct nexus_adapter *nx_port_na, na_activate_mode_t mode)
244 {
245 	struct flow_owner_bucket *fob;
246 	uint32_t fo_cnt = 0;
247 
248 	VERIFY(nx_port != NEXUS_PORT_ANY);
249 	VERIFY(nx_port_na != NULL);
250 
251 	if (pid_bound) {
252 		fob = flow_mgr_get_fob_by_pid(fm, pid);
253 		fo_cnt = flow_owner_bucket_activate_nx_port_common(fob, nx_port,
254 		    nx_port_na, mode);
255 	} else {
256 		uint32_t i;
257 		/*
258 		 * Otherwise, this can get expensive since we need to search
259 		 * thru all proc-mapping buckets to find the flows that are
260 		 * related to this nexus port.
261 		 */
262 		for (i = 0; i < fm->fm_owner_buckets_cnt; i++) {
263 			fob = flow_mgr_get_fob_at_idx(fm, i);
264 			fo_cnt += flow_owner_bucket_activate_nx_port_common(fob,
265 			    nx_port, nx_port_na, mode);
266 		}
267 	}
268 	/* There shouldn't be more than one flow owners on a nexus port */
269 	VERIFY(fo_cnt <= 1);
270 	return fo_cnt;
271 }
272 
273 static void
flow_owner_bucket_attach_common(struct flow_owner_bucket * fob,nexus_port_t nx_port)274 flow_owner_bucket_attach_common(struct flow_owner_bucket *fob,
275     nexus_port_t nx_port)
276 {
277 	struct flow_owner *fo;
278 
279 	VERIFY(nx_port != NEXUS_PORT_ANY);
280 	FOB_LOCK(fob);
281 
282 	RB_FOREACH(fo, flow_owner_tree, &fob->fob_owner_head) {
283 		if (fo->fo_nx_port_destroyed && (fo->fo_nx_port == nx_port)) {
284 			fo->fo_nx_port_destroyed = FALSE;
285 		}
286 	}
287 
288 	FOB_UNLOCK(fob);
289 }
290 
291 void
flow_owner_attach_nexus_port(struct flow_mgr * fm,boolean_t pid_bound,pid_t pid,nexus_port_t nx_port)292 flow_owner_attach_nexus_port(struct flow_mgr *fm, boolean_t pid_bound,
293     pid_t pid, nexus_port_t nx_port)
294 {
295 	struct flow_owner_bucket *fob;
296 	ASSERT(nx_port != NEXUS_PORT_ANY);
297 
298 	if (pid_bound) {
299 		fob = flow_mgr_get_fob_by_pid(fm, pid);
300 		flow_owner_bucket_attach_common(fob, nx_port);
301 	} else {
302 		uint32_t i;
303 		/*
304 		 * Otherwise, this can get expensive since we need to search
305 		 * thru all proc-mapping buckets to find the flows that are
306 		 * related to this nexus port.
307 		 */
308 		for (i = 0; i < fm->fm_owner_buckets_cnt; i++) {
309 			fob = flow_mgr_get_fob_at_idx(fm, i);
310 			flow_owner_bucket_attach_common(fob, nx_port);
311 		}
312 	}
313 }
314 
315 uint32_t
flow_owner_detach_nexus_port(struct flow_mgr * fm,boolean_t pid_bound,pid_t pid,nexus_port_t nx_port,boolean_t if_idle)316 flow_owner_detach_nexus_port(struct flow_mgr *fm, boolean_t pid_bound,
317     pid_t pid, nexus_port_t nx_port, boolean_t if_idle)
318 {
319 	struct flow_owner_bucket *fob;
320 	uint32_t purged = 0;
321 	ASSERT(nx_port != NEXUS_PORT_ANY);
322 
323 	if (pid_bound) {
324 		fob = flow_mgr_get_fob_by_pid(fm, pid);
325 		purged = flow_owner_bucket_purge_common(fob, nx_port, if_idle);
326 	} else {
327 		uint32_t i;
328 		/*
329 		 * Otherwise, this can get expensive since we need to search
330 		 * thru all proc-mapping buckets to find the flows that are
331 		 * related to this nexus port.
332 		 */
333 		for (i = 0; i < fm->fm_owner_buckets_cnt; i++) {
334 			fob = flow_mgr_get_fob_at_idx(fm, i);
335 			purged += flow_owner_bucket_purge_common(fob,
336 			    nx_port, if_idle);
337 		}
338 	}
339 	return purged;
340 }
341 
342 /* 64-bit mask with range */
343 #define FO_BMASK64(_beg, _end)  \
344 	((((uint64_t)0xffffffffffffffff) >>     \
345 	    (63 - (_end))) & ~((1ULL << (_beg)) - 1))
346 
347 struct flow_owner *
flow_owner_alloc(struct flow_owner_bucket * fob,struct proc * p,nexus_port_t nx_port,bool nx_port_pid_bound,bool flowadv,struct nx_flowswitch * fsw,struct nexus_adapter * nx_port_na,void * context,bool low_latency)348 flow_owner_alloc(struct flow_owner_bucket *fob, struct proc *p,
349     nexus_port_t nx_port, bool nx_port_pid_bound, bool flowadv,
350     struct nx_flowswitch *fsw, struct nexus_adapter *nx_port_na,
351     void *context, bool low_latency)
352 {
353 	struct flow_owner *fo;
354 	const pid_t pid = proc_pid(p);
355 
356 	static_assert(true == 1);
357 	static_assert(false == 0);
358 	ASSERT(low_latency == true || low_latency == false);
359 	ASSERT(nx_port != NEXUS_PORT_ANY);
360 	FOB_LOCK_ASSERT_HELD(fob);
361 
362 #if DEBUG
363 	ASSERT(flow_owner_find_by_pid(fob, pid, context, low_latency) == NULL);
364 	RB_FOREACH(fo, flow_owner_tree, &fob->fob_owner_head) {
365 		if (!fo->fo_nx_port_destroyed && (fo->fo_nx_port == nx_port)) {
366 			VERIFY(0);
367 			/* NOTREACHED */
368 			__builtin_unreachable();
369 		}
370 	}
371 #endif /* DEBUG */
372 
373 	fo = fo_alloc(TRUE);
374 	if (fo != NULL) {
375 		if (flowadv) {
376 			uint32_t i;
377 			bitmap_t *bmap;
378 
379 			bmap = skmem_cache_alloc(sk_fab_cache, SKMEM_SLEEP);
380 			if (bmap == NULL) {
381 				SK_ERR("failed to alloc flow advisory bitmap");
382 				fo_free(fo);
383 				return NULL;
384 			}
385 			bzero(bmap, sk_fab_size);
386 			fo->fo_flowadv_bmap = bmap;
387 			fo->fo_num_flowadv_bmaps = sk_fadv_nchunks;
388 			fo->fo_flowadv_max = sk_max_flows;
389 
390 			/* set the bits for free indices */
391 			for (i = 0; i < sk_fadv_nchunks; i++) {
392 				uint32_t end = 63;
393 
394 				if (i == (sk_fadv_nchunks - 1)) {
395 					end = ((sk_max_flows - 1) %
396 					    FO_FLOWADV_CHUNK);
397 				}
398 
399 				fo->fo_flowadv_bmap[i] = FO_BMASK64(0, end);
400 			}
401 		}
402 		RB_INIT(&fo->fo_flow_entry_id_head);
403 		/* const override */
404 		*(struct flow_owner_bucket **)(uintptr_t)&fo->fo_bucket = fob;
405 		fo->fo_context = context;
406 		fo->fo_pid = pid;
407 		(void) snprintf(fo->fo_name, sizeof(fo->fo_name), "%s",
408 		    proc_name_address(p));
409 		fo->fo_nx_port_pid_bound = nx_port_pid_bound;
410 		fo->fo_low_latency = low_latency;
411 		fo->fo_nx_port = nx_port;
412 		*(struct nexus_adapter **)(uintptr_t)&fo->fo_nx_port_na = nx_port_na;
413 		*(struct nx_flowswitch **)(uintptr_t)&fo->fo_fsw = fsw;
414 		RB_INSERT(flow_owner_tree, &fob->fob_owner_head, fo);
415 
416 		SK_DF(SK_VERB_FLOW, "%s(%d) fob %p added fo %p "
417 		    "nx_port %d nx_port_pid_bound %d ll %d nx_port_na %p",
418 		    fo->fo_name, fo->fo_pid, SK_KVA(fob), SK_KVA(fo),
419 		    (int)nx_port, nx_port_pid_bound, fo->fo_low_latency,
420 		    SK_KVA(nx_port_na));
421 	}
422 
423 	return fo;
424 }
425 
426 void
flow_owner_free(struct flow_owner_bucket * fob,struct flow_owner * fo)427 flow_owner_free(struct flow_owner_bucket *fob, struct flow_owner *fo)
428 {
429 	FOB_LOCK_ASSERT_HELD(fob);
430 
431 	ASSERT(fo->fo_bucket == fob);
432 	*(struct flow_owner_bucket **)(uintptr_t)&fo->fo_bucket = NULL;
433 	RB_REMOVE(flow_owner_tree, &fob->fob_owner_head, fo);
434 
435 	ASSERT(fo->fo_num_flowadv == 0);
436 	skmem_cache_free(sk_fab_cache, fo->fo_flowadv_bmap);
437 	fo->fo_flowadv_bmap = NULL;
438 	fo->fo_num_flowadv_bmaps = 0;
439 
440 	/* wake up any thread blocked in flow_owner_bucket_destroy() */
441 	if (RB_EMPTY(&fob->fob_owner_head) && fob->fob_dtor_waiters > 0) {
442 		fob->fob_dtor_waiters = 0;
443 		wakeup(&fob->fob_dtor_waiters);
444 	}
445 
446 	SK_DF(SK_VERB_FLOW, "%s(%d) fob %p removed fo %p nx_port %d",
447 	    fo->fo_name, fo->fo_pid, SK_KVA(fob), SK_KVA(fo),
448 	    (int)fo->fo_nx_port);
449 
450 	fo_free(fo);
451 }
452 
453 int
flow_owner_flowadv_index_alloc(struct flow_owner * fo,flowadv_idx_t * fadv_idx)454 flow_owner_flowadv_index_alloc(struct flow_owner *fo, flowadv_idx_t *fadv_idx)
455 {
456 	bitmap_t *bmap = fo->fo_flowadv_bmap;
457 	size_t nchunks, i, j, idx = FLOWADV_IDX_NONE;
458 
459 	FOB_LOCK_ASSERT_HELD(FO_BUCKET(fo));
460 	ASSERT(fo->fo_flowadv_max != 0);
461 
462 	nchunks = P2ROUNDUP(fo->fo_flowadv_max, FO_FLOWADV_CHUNK) /
463 	    FO_FLOWADV_CHUNK;
464 
465 	for (i = 0; i < nchunks; i++) {
466 		j = ffsll(bmap[i]);
467 		if (j == 0) {
468 			/* All indices in this chunk are in use */
469 			continue;
470 		}
471 		--j;
472 		/* mark the index as in use */
473 		bit_clear(bmap[i], j);
474 		idx = (i * FO_FLOWADV_CHUNK) + j;
475 		break;
476 	}
477 
478 	if (idx == FLOWADV_IDX_NONE) {
479 		SK_ERR("%s(%d) flow advisory table full: num %u max %u",
480 		    fo->fo_name, fo->fo_pid, fo->fo_num_flowadv,
481 		    fo->fo_flowadv_max);
482 		VERIFY(fo->fo_num_flowadv == fo->fo_flowadv_max);
483 		*fadv_idx = FLOWADV_IDX_NONE;
484 		return ENOMEM;
485 	}
486 
487 	fo->fo_num_flowadv++;
488 	ASSERT(idx < ((flowadv_idx_t) -1));
489 	*fadv_idx = (flowadv_idx_t)idx;
490 	ASSERT(*fadv_idx < fo->fo_flowadv_max);
491 	return 0;
492 }
493 
494 void
flow_owner_flowadv_index_free(struct flow_owner * fo,flowadv_idx_t fadv_idx)495 flow_owner_flowadv_index_free(struct flow_owner *fo, flowadv_idx_t fadv_idx)
496 {
497 	uint32_t chunk_idx, bit_pos;
498 	bitmap_t *bmap = fo->fo_flowadv_bmap;
499 
500 	FOB_LOCK_ASSERT_HELD(FO_BUCKET(fo));
501 	ASSERT(fo->fo_num_flowadv != 0);
502 	ASSERT((fo->fo_flowadv_max != 0) && (fadv_idx < fo->fo_flowadv_max));
503 
504 	chunk_idx = fadv_idx / FO_FLOWADV_CHUNK;
505 	bit_pos = fadv_idx % FO_FLOWADV_CHUNK;
506 	ASSERT(!bit_test(bmap[chunk_idx], bit_pos));
507 	/* mark the index as free */
508 	bit_set(bmap[chunk_idx], bit_pos);
509 	fo->fo_num_flowadv--;
510 }
511 
512 int
flow_owner_destroy_entry(struct flow_owner * fo,uuid_t uuid,bool nolinger,void * close_params)513 flow_owner_destroy_entry(struct flow_owner *fo, uuid_t uuid,
514     bool nolinger, void *close_params)
515 {
516 	struct flow_entry *fe = NULL;
517 	int err = 0;
518 
519 	FOB_LOCK_ASSERT_HELD(FO_BUCKET(fo));
520 
521 	/* lookup such flow for this process */
522 	fe = flow_entry_find_by_uuid(fo, uuid);
523 	if (fe == NULL) {
524 		err = ENOENT;
525 	} else {
526 		/* free flow entry (OK to linger if caller asked) */
527 		flow_entry_destroy(fo, fe, nolinger, close_params);
528 	}
529 
530 	return err;
531 }
532 
533 static inline int
fo_cmp(const struct flow_owner * a,const struct flow_owner * b)534 fo_cmp(const struct flow_owner *a, const struct flow_owner *b)
535 {
536 	if (a->fo_pid > b->fo_pid) {
537 		return 1;
538 	}
539 	if (a->fo_pid < b->fo_pid) {
540 		return -1;
541 	}
542 	if ((intptr_t)a->fo_context > (intptr_t)b->fo_context) {
543 		return 1;
544 	} else if ((intptr_t)a->fo_context < (intptr_t)b->fo_context) {
545 		return -1;
546 	}
547 	if (a->fo_low_latency != b->fo_low_latency) {
548 		if (a->fo_low_latency) {
549 			return 1;
550 		} else {
551 			return -1;
552 		}
553 	}
554 	return 0;
555 }
556 
557 static struct flow_owner *
fo_alloc(boolean_t can_block)558 fo_alloc(boolean_t can_block)
559 {
560 	struct flow_owner *fo;
561 
562 	fo = skmem_cache_alloc(sk_fo_cache,
563 	    can_block ? SKMEM_SLEEP : SKMEM_NOSLEEP);
564 	if (fo == NULL) {
565 		return NULL;
566 	}
567 
568 	bzero(fo, sk_fo_size);
569 
570 	SK_DF(SK_VERB_MEM, "fo %p ALLOC", SK_KVA(fo));
571 
572 	return fo;
573 }
574 
575 static void
fo_free(struct flow_owner * fo)576 fo_free(struct flow_owner *fo)
577 {
578 	ASSERT(fo->fo_bucket == NULL);
579 	ASSERT(RB_EMPTY(&fo->fo_flow_entry_id_head));
580 	ASSERT(fo->fo_flowadv_bmap == NULL);
581 
582 	SK_DF(SK_VERB_MEM, "fo %p FREE", SK_KVA(fo));
583 
584 	skmem_cache_free(sk_fo_cache, fo);
585 }
586