1 // SPDX-License-Identifier: GPL-2.0
3 * Multipath support for RPC
5 * Copyright (c) 2015, 2016, Primary Data, Inc. All rights reserved.
7 * Trond Myklebust <trond.myklebust@primarydata.com>
10 #include <linux/atomic.h>
11 #include <linux/types.h>
12 #include <linux/kref.h>
13 #include <linux/list.h>
14 #include <linux/rcupdate.h>
15 #include <linux/rculist.h>
16 #include <linux/slab.h>
17 #include <linux/spinlock.h>
18 #include <linux/sunrpc/xprt.h>
19 #include <linux/sunrpc/addr.h>
20 #include <linux/sunrpc/xprtmultipath.h>
24 typedef struct rpc_xprt *(*xprt_switch_find_xprt_t)(struct rpc_xprt_switch *xps,
25 const struct rpc_xprt *cur);
27 static const struct rpc_xprt_iter_ops rpc_xprt_iter_singular;
28 static const struct rpc_xprt_iter_ops rpc_xprt_iter_roundrobin;
29 static const struct rpc_xprt_iter_ops rpc_xprt_iter_listall;
30 static const struct rpc_xprt_iter_ops rpc_xprt_iter_listoffline;
32 static void xprt_switch_add_xprt_locked(struct rpc_xprt_switch *xps,
33 struct rpc_xprt *xprt)
35 if (unlikely(xprt_get(xprt) == NULL))
37 list_add_tail_rcu(&xprt->xprt_switch, &xps->xps_xprt_list);
39 if (xps->xps_nxprts == 0)
40 xps->xps_net = xprt->xprt_net;
46 * rpc_xprt_switch_add_xprt - Add a new rpc_xprt to an rpc_xprt_switch
47 * @xps: pointer to struct rpc_xprt_switch
48 * @xprt: pointer to struct rpc_xprt
50 * Adds xprt to the end of the list of struct rpc_xprt in xps.
52 void rpc_xprt_switch_add_xprt(struct rpc_xprt_switch *xps,
53 struct rpc_xprt *xprt)
57 spin_lock(&xps->xps_lock);
58 if (xps->xps_net == xprt->xprt_net || xps->xps_net == NULL)
59 xprt_switch_add_xprt_locked(xps, xprt);
60 spin_unlock(&xps->xps_lock);
61 rpc_sysfs_xprt_setup(xps, xprt, GFP_KERNEL);
64 static void xprt_switch_remove_xprt_locked(struct rpc_xprt_switch *xps,
65 struct rpc_xprt *xprt, bool offline)
67 if (unlikely(xprt == NULL))
69 if (!test_bit(XPRT_OFFLINE, &xprt->state) && offline)
72 if (xps->xps_nxprts == 0)
75 list_del_rcu(&xprt->xprt_switch);
79 * rpc_xprt_switch_remove_xprt - Removes an rpc_xprt from a rpc_xprt_switch
80 * @xps: pointer to struct rpc_xprt_switch
81 * @xprt: pointer to struct rpc_xprt
82 * @offline: indicates if the xprt that's being removed is in an offline state
84 * Removes xprt from the list of struct rpc_xprt in xps.
86 void rpc_xprt_switch_remove_xprt(struct rpc_xprt_switch *xps,
87 struct rpc_xprt *xprt, bool offline)
89 spin_lock(&xps->xps_lock);
90 xprt_switch_remove_xprt_locked(xps, xprt, offline);
91 spin_unlock(&xps->xps_lock);
95 static DEFINE_IDA(rpc_xprtswitch_ids);
97 void xprt_multipath_cleanup_ids(void)
99 ida_destroy(&rpc_xprtswitch_ids);
102 static int xprt_switch_alloc_id(struct rpc_xprt_switch *xps, gfp_t gfp_flags)
106 id = ida_alloc(&rpc_xprtswitch_ids, gfp_flags);
114 static void xprt_switch_free_id(struct rpc_xprt_switch *xps)
116 ida_free(&rpc_xprtswitch_ids, xps->xps_id);
120 * xprt_switch_alloc - Allocate a new struct rpc_xprt_switch
121 * @xprt: pointer to struct rpc_xprt
122 * @gfp_flags: allocation flags
124 * On success, returns an initialised struct rpc_xprt_switch, containing
125 * the entry xprt. Returns NULL on failure.
127 struct rpc_xprt_switch *xprt_switch_alloc(struct rpc_xprt *xprt,
130 struct rpc_xprt_switch *xps;
132 xps = kmalloc(sizeof(*xps), gfp_flags);
134 spin_lock_init(&xps->xps_lock);
135 kref_init(&xps->xps_kref);
136 xprt_switch_alloc_id(xps, gfp_flags);
137 xps->xps_nxprts = xps->xps_nactive = 0;
138 atomic_long_set(&xps->xps_queuelen, 0);
140 INIT_LIST_HEAD(&xps->xps_xprt_list);
141 xps->xps_iter_ops = &rpc_xprt_iter_singular;
142 rpc_sysfs_xprt_switch_setup(xps, xprt, gfp_flags);
143 xprt_switch_add_xprt_locked(xps, xprt);
144 xps->xps_nunique_destaddr_xprts = 1;
145 rpc_sysfs_xprt_setup(xps, xprt, gfp_flags);
151 static void xprt_switch_free_entries(struct rpc_xprt_switch *xps)
153 spin_lock(&xps->xps_lock);
154 while (!list_empty(&xps->xps_xprt_list)) {
155 struct rpc_xprt *xprt;
157 xprt = list_first_entry(&xps->xps_xprt_list,
158 struct rpc_xprt, xprt_switch);
159 xprt_switch_remove_xprt_locked(xps, xprt, true);
160 spin_unlock(&xps->xps_lock);
162 spin_lock(&xps->xps_lock);
164 spin_unlock(&xps->xps_lock);
167 static void xprt_switch_free(struct kref *kref)
169 struct rpc_xprt_switch *xps = container_of(kref,
170 struct rpc_xprt_switch, xps_kref);
172 xprt_switch_free_entries(xps);
173 rpc_sysfs_xprt_switch_destroy(xps);
174 xprt_switch_free_id(xps);
175 kfree_rcu(xps, xps_rcu);
179 * xprt_switch_get - Return a reference to a rpc_xprt_switch
180 * @xps: pointer to struct rpc_xprt_switch
182 * Returns a reference to xps unless the refcount is already zero.
184 struct rpc_xprt_switch *xprt_switch_get(struct rpc_xprt_switch *xps)
186 if (xps != NULL && kref_get_unless_zero(&xps->xps_kref))
192 * xprt_switch_put - Release a reference to a rpc_xprt_switch
193 * @xps: pointer to struct rpc_xprt_switch
195 * Release the reference to xps, and free it once the refcount is zero.
197 void xprt_switch_put(struct rpc_xprt_switch *xps)
200 kref_put(&xps->xps_kref, xprt_switch_free);
204 * rpc_xprt_switch_set_roundrobin - Set a round-robin policy on rpc_xprt_switch
205 * @xps: pointer to struct rpc_xprt_switch
207 * Sets a round-robin default policy for iterators acting on xps.
209 void rpc_xprt_switch_set_roundrobin(struct rpc_xprt_switch *xps)
211 if (READ_ONCE(xps->xps_iter_ops) != &rpc_xprt_iter_roundrobin)
212 WRITE_ONCE(xps->xps_iter_ops, &rpc_xprt_iter_roundrobin);
216 const struct rpc_xprt_iter_ops *xprt_iter_ops(const struct rpc_xprt_iter *xpi)
218 if (xpi->xpi_ops != NULL)
220 return rcu_dereference(xpi->xpi_xpswitch)->xps_iter_ops;
224 void xprt_iter_no_rewind(struct rpc_xprt_iter *xpi)
229 void xprt_iter_default_rewind(struct rpc_xprt_iter *xpi)
231 WRITE_ONCE(xpi->xpi_cursor, NULL);
235 bool xprt_is_active(const struct rpc_xprt *xprt)
237 return (kref_read(&xprt->kref) != 0 &&
238 !test_bit(XPRT_OFFLINE, &xprt->state));
242 struct rpc_xprt *xprt_switch_find_first_entry(struct list_head *head)
244 struct rpc_xprt *pos;
246 list_for_each_entry_rcu(pos, head, xprt_switch) {
247 if (xprt_is_active(pos))
254 struct rpc_xprt *xprt_switch_find_first_entry_offline(struct list_head *head)
256 struct rpc_xprt *pos;
258 list_for_each_entry_rcu(pos, head, xprt_switch) {
259 if (!xprt_is_active(pos))
266 struct rpc_xprt *xprt_iter_first_entry(struct rpc_xprt_iter *xpi)
268 struct rpc_xprt_switch *xps = rcu_dereference(xpi->xpi_xpswitch);
272 return xprt_switch_find_first_entry(&xps->xps_xprt_list);
276 struct rpc_xprt *_xprt_switch_find_current_entry(struct list_head *head,
277 const struct rpc_xprt *cur,
280 struct rpc_xprt *pos;
283 list_for_each_entry_rcu(pos, head, xprt_switch) {
286 if (found && ((find_active && xprt_is_active(pos)) ||
287 (!find_active && !xprt_is_active(pos))))
294 struct rpc_xprt *xprt_switch_find_current_entry(struct list_head *head,
295 const struct rpc_xprt *cur)
297 return _xprt_switch_find_current_entry(head, cur, true);
301 struct rpc_xprt * _xprt_iter_current_entry(struct rpc_xprt_iter *xpi,
302 struct rpc_xprt *first_entry(struct list_head *head),
303 struct rpc_xprt *current_entry(struct list_head *head,
304 const struct rpc_xprt *cur))
306 struct rpc_xprt_switch *xps = rcu_dereference(xpi->xpi_xpswitch);
307 struct list_head *head;
311 head = &xps->xps_xprt_list;
312 if (xpi->xpi_cursor == NULL || xps->xps_nxprts < 2)
313 return first_entry(head);
314 return current_entry(head, xpi->xpi_cursor);
318 struct rpc_xprt *xprt_iter_current_entry(struct rpc_xprt_iter *xpi)
320 return _xprt_iter_current_entry(xpi, xprt_switch_find_first_entry,
321 xprt_switch_find_current_entry);
325 struct rpc_xprt *xprt_switch_find_current_entry_offline(struct list_head *head,
326 const struct rpc_xprt *cur)
328 return _xprt_switch_find_current_entry(head, cur, false);
332 struct rpc_xprt *xprt_iter_current_entry_offline(struct rpc_xprt_iter *xpi)
334 return _xprt_iter_current_entry(xpi,
335 xprt_switch_find_first_entry_offline,
336 xprt_switch_find_current_entry_offline);
340 bool __rpc_xprt_switch_has_addr(struct rpc_xprt_switch *xps,
341 const struct sockaddr *sap)
343 struct list_head *head;
344 struct rpc_xprt *pos;
346 if (xps == NULL || sap == NULL)
349 head = &xps->xps_xprt_list;
350 list_for_each_entry_rcu(pos, head, xprt_switch) {
351 if (rpc_cmp_addr_port(sap, (struct sockaddr *)&pos->addr)) {
352 pr_info("RPC: addr %s already in xprt switch\n",
353 pos->address_strings[RPC_DISPLAY_ADDR]);
360 bool rpc_xprt_switch_has_addr(struct rpc_xprt_switch *xps,
361 const struct sockaddr *sap)
366 res = __rpc_xprt_switch_has_addr(xps, sap);
373 struct rpc_xprt *xprt_switch_find_next_entry(struct list_head *head,
374 const struct rpc_xprt *cur, bool check_active)
376 struct rpc_xprt *pos, *prev = NULL;
379 list_for_each_entry_rcu(pos, head, xprt_switch) {
382 /* for request to return active transports return only
383 * active, for request to return offline transports
384 * return only offline
386 if (found && ((check_active && xprt_is_active(pos)) ||
387 (!check_active && !xprt_is_active(pos))))
395 struct rpc_xprt *xprt_switch_set_next_cursor(struct rpc_xprt_switch *xps,
396 struct rpc_xprt **cursor,
397 xprt_switch_find_xprt_t find_next)
399 struct rpc_xprt *pos, *old;
401 old = smp_load_acquire(cursor);
402 pos = find_next(xps, old);
403 smp_store_release(cursor, pos);
408 struct rpc_xprt *xprt_iter_next_entry_multiple(struct rpc_xprt_iter *xpi,
409 xprt_switch_find_xprt_t find_next)
411 struct rpc_xprt_switch *xps = rcu_dereference(xpi->xpi_xpswitch);
415 return xprt_switch_set_next_cursor(xps, &xpi->xpi_cursor, find_next);
419 struct rpc_xprt *__xprt_switch_find_next_entry_roundrobin(struct list_head *head,
420 const struct rpc_xprt *cur)
422 struct rpc_xprt *ret;
424 ret = xprt_switch_find_next_entry(head, cur, true);
427 return xprt_switch_find_first_entry(head);
431 struct rpc_xprt *xprt_switch_find_next_entry_roundrobin(struct rpc_xprt_switch *xps,
432 const struct rpc_xprt *cur)
434 struct list_head *head = &xps->xps_xprt_list;
435 struct rpc_xprt *xprt;
436 unsigned int nactive;
439 unsigned long xprt_queuelen, xps_queuelen;
441 xprt = __xprt_switch_find_next_entry_roundrobin(head, cur);
444 xprt_queuelen = atomic_long_read(&xprt->queuelen);
445 xps_queuelen = atomic_long_read(&xps->xps_queuelen);
446 nactive = READ_ONCE(xps->xps_nactive);
447 /* Exit loop if xprt_queuelen <= average queue length */
448 if (xprt_queuelen * nactive <= xps_queuelen)
456 struct rpc_xprt *xprt_iter_next_entry_roundrobin(struct rpc_xprt_iter *xpi)
458 return xprt_iter_next_entry_multiple(xpi,
459 xprt_switch_find_next_entry_roundrobin);
463 struct rpc_xprt *xprt_switch_find_next_entry_all(struct rpc_xprt_switch *xps,
464 const struct rpc_xprt *cur)
466 return xprt_switch_find_next_entry(&xps->xps_xprt_list, cur, true);
470 struct rpc_xprt *xprt_switch_find_next_entry_offline(struct rpc_xprt_switch *xps,
471 const struct rpc_xprt *cur)
473 return xprt_switch_find_next_entry(&xps->xps_xprt_list, cur, false);
477 struct rpc_xprt *xprt_iter_next_entry_all(struct rpc_xprt_iter *xpi)
479 return xprt_iter_next_entry_multiple(xpi,
480 xprt_switch_find_next_entry_all);
484 struct rpc_xprt *xprt_iter_next_entry_offline(struct rpc_xprt_iter *xpi)
486 return xprt_iter_next_entry_multiple(xpi,
487 xprt_switch_find_next_entry_offline);
491 * xprt_iter_rewind - Resets the xprt iterator
492 * @xpi: pointer to rpc_xprt_iter
494 * Resets xpi to ensure that it points to the first entry in the list
497 void xprt_iter_rewind(struct rpc_xprt_iter *xpi)
500 xprt_iter_ops(xpi)->xpi_rewind(xpi);
504 static void __xprt_iter_init(struct rpc_xprt_iter *xpi,
505 struct rpc_xprt_switch *xps,
506 const struct rpc_xprt_iter_ops *ops)
508 rcu_assign_pointer(xpi->xpi_xpswitch, xprt_switch_get(xps));
509 xpi->xpi_cursor = NULL;
514 * xprt_iter_init - Initialise an xprt iterator
515 * @xpi: pointer to rpc_xprt_iter
516 * @xps: pointer to rpc_xprt_switch
518 * Initialises the iterator to use the default iterator ops
519 * as set in xps. This function is mainly intended for internal
520 * use in the rpc_client.
522 void xprt_iter_init(struct rpc_xprt_iter *xpi,
523 struct rpc_xprt_switch *xps)
525 __xprt_iter_init(xpi, xps, NULL);
529 * xprt_iter_init_listall - Initialise an xprt iterator
530 * @xpi: pointer to rpc_xprt_iter
531 * @xps: pointer to rpc_xprt_switch
533 * Initialises the iterator to iterate once through the entire list
536 void xprt_iter_init_listall(struct rpc_xprt_iter *xpi,
537 struct rpc_xprt_switch *xps)
539 __xprt_iter_init(xpi, xps, &rpc_xprt_iter_listall);
542 void xprt_iter_init_listoffline(struct rpc_xprt_iter *xpi,
543 struct rpc_xprt_switch *xps)
545 __xprt_iter_init(xpi, xps, &rpc_xprt_iter_listoffline);
549 * xprt_iter_xchg_switch - Atomically swap out the rpc_xprt_switch
550 * @xpi: pointer to rpc_xprt_iter
551 * @newswitch: pointer to a new rpc_xprt_switch or NULL
553 * Swaps out the existing xpi->xpi_xpswitch with a new value.
555 struct rpc_xprt_switch *xprt_iter_xchg_switch(struct rpc_xprt_iter *xpi,
556 struct rpc_xprt_switch *newswitch)
558 struct rpc_xprt_switch __rcu *oldswitch;
560 /* Atomically swap out the old xpswitch */
561 oldswitch = xchg(&xpi->xpi_xpswitch, RCU_INITIALIZER(newswitch));
562 if (newswitch != NULL)
563 xprt_iter_rewind(xpi);
564 return rcu_dereference_protected(oldswitch, true);
568 * xprt_iter_destroy - Destroys the xprt iterator
569 * @xpi: pointer to rpc_xprt_iter
571 void xprt_iter_destroy(struct rpc_xprt_iter *xpi)
573 xprt_switch_put(xprt_iter_xchg_switch(xpi, NULL));
577 * xprt_iter_xprt - Returns the rpc_xprt pointed to by the cursor
578 * @xpi: pointer to rpc_xprt_iter
580 * Returns a pointer to the struct rpc_xprt that is currently
581 * pointed to by the cursor.
582 * Caller must be holding rcu_read_lock().
584 struct rpc_xprt *xprt_iter_xprt(struct rpc_xprt_iter *xpi)
586 WARN_ON_ONCE(!rcu_read_lock_held());
587 return xprt_iter_ops(xpi)->xpi_xprt(xpi);
591 struct rpc_xprt *xprt_iter_get_helper(struct rpc_xprt_iter *xpi,
592 struct rpc_xprt *(*fn)(struct rpc_xprt_iter *))
594 struct rpc_xprt *ret;
601 } while (ret == NULL);
606 * xprt_iter_get_xprt - Returns the rpc_xprt pointed to by the cursor
607 * @xpi: pointer to rpc_xprt_iter
609 * Returns a reference to the struct rpc_xprt that is currently
610 * pointed to by the cursor.
612 struct rpc_xprt *xprt_iter_get_xprt(struct rpc_xprt_iter *xpi)
614 struct rpc_xprt *xprt;
617 xprt = xprt_iter_get_helper(xpi, xprt_iter_ops(xpi)->xpi_xprt);
623 * xprt_iter_get_next - Returns the next rpc_xprt following the cursor
624 * @xpi: pointer to rpc_xprt_iter
626 * Returns a reference to the struct rpc_xprt that immediately follows the
627 * entry pointed to by the cursor.
629 struct rpc_xprt *xprt_iter_get_next(struct rpc_xprt_iter *xpi)
631 struct rpc_xprt *xprt;
634 xprt = xprt_iter_get_helper(xpi, xprt_iter_ops(xpi)->xpi_next);
639 /* Policy for always returning the first entry in the rpc_xprt_switch */
641 const struct rpc_xprt_iter_ops rpc_xprt_iter_singular = {
642 .xpi_rewind = xprt_iter_no_rewind,
643 .xpi_xprt = xprt_iter_first_entry,
644 .xpi_next = xprt_iter_first_entry,
647 /* Policy for round-robin iteration of entries in the rpc_xprt_switch */
649 const struct rpc_xprt_iter_ops rpc_xprt_iter_roundrobin = {
650 .xpi_rewind = xprt_iter_default_rewind,
651 .xpi_xprt = xprt_iter_current_entry,
652 .xpi_next = xprt_iter_next_entry_roundrobin,
655 /* Policy for once-through iteration of entries in the rpc_xprt_switch */
657 const struct rpc_xprt_iter_ops rpc_xprt_iter_listall = {
658 .xpi_rewind = xprt_iter_default_rewind,
659 .xpi_xprt = xprt_iter_current_entry,
660 .xpi_next = xprt_iter_next_entry_all,
664 const struct rpc_xprt_iter_ops rpc_xprt_iter_listoffline = {
665 .xpi_rewind = xprt_iter_default_rewind,
666 .xpi_xprt = xprt_iter_current_entry_offline,
667 .xpi_next = xprt_iter_next_entry_offline,