1 /******************************************************************************
2  *
3  * Back-end of the driver for virtual block devices. This portion of the
4  * driver exports a 'unified' block-device interface that can be accessed
5  * by any operating system that implements a compatible front end. A
6  * reference front-end implementation can be found in:
7  *  drivers/block/xen-blkfront.c
8  *
9  * Copyright (c) 2003-2004, Keir Fraser & Steve Hand
10  * Copyright (c) 2005, Christopher Clark
11  *
12  * This program is free software; you can redistribute it and/or
13  * modify it under the terms of the GNU General Public License version 2
14  * as published by the Free Software Foundation; or, when distributed
15  * separately from the Linux kernel or incorporated into other
16  * software packages, subject to the following license:
17  *
18  * Permission is hereby granted, free of charge, to any person obtaining a copy
19  * of this source file (the "Software"), to deal in the Software without
20  * restriction, including without limitation the rights to use, copy, modify,
21  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
22  * and to permit persons to whom the Software is furnished to do so, subject to
23  * the following conditions:
24  *
25  * The above copyright notice and this permission notice shall be included in
26  * all copies or substantial portions of the Software.
27  *
28  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
29  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
30  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
31  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
32  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
33  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
34  * IN THE SOFTWARE.
35  */
36 
37 #define pr_fmt(fmt) "xen-blkback: " fmt
38 
39 #include <linux/spinlock.h>
40 #include <linux/kthread.h>
41 #include <linux/list.h>
42 #include <linux/delay.h>
43 #include <linux/freezer.h>
44 #include <linux/bitmap.h>
45 
46 #include <xen/events.h>
47 #include <xen/page.h>
48 #include <xen/xen.h>
49 #include <asm/xen/hypervisor.h>
50 #include <asm/xen/hypercall.h>
51 #include <xen/balloon.h>
52 #include <xen/grant_table.h>
53 #include "common.h"
54 
55 /*
56  * Maximum number of unused free pages to keep in the internal buffer.
57  * Setting this to a value too low will reduce memory used in each backend,
58  * but can have a performance penalty.
59  *
60  * A sane value is xen_blkif_reqs * BLKIF_MAX_SEGMENTS_PER_REQUEST, but can
61  * be set to a lower value that might degrade performance on some intensive
62  * IO workloads.
63  */
64 
65 static int max_buffer_pages = 1024;
66 module_param_named(max_buffer_pages, max_buffer_pages, int, 0644);
67 MODULE_PARM_DESC(max_buffer_pages,
68 "Maximum number of free pages to keep in each block backend buffer");
69 
70 /*
71  * Maximum number of grants to map persistently in blkback. For maximum
72  * performance this should be the total numbers of grants that can be used
73  * to fill the ring, but since this might become too high, specially with
74  * the use of indirect descriptors, we set it to a value that provides good
75  * performance without using too much memory.
76  *
77  * When the list of persistent grants is full we clean it up using a LRU
78  * algorithm.
79  */
80 
81 static int max_pgrants = 1056;
82 module_param_named(max_persistent_grants, max_pgrants, int, 0644);
83 MODULE_PARM_DESC(max_persistent_grants,
84                  "Maximum number of grants to map persistently");
85 
86 /*
87  * How long a persistent grant is allowed to remain allocated without being in
88  * use. The time is in seconds, 0 means indefinitely long.
89  */
90 
91 static unsigned int pgrant_timeout = 60;
92 module_param_named(persistent_grant_unused_seconds, pgrant_timeout,
93 		   uint, 0644);
94 MODULE_PARM_DESC(persistent_grant_unused_seconds,
95 		 "Time in seconds an unused persistent grant is allowed to "
96 		 "remain allocated. Default is 60, 0 means unlimited.");
97 
98 /*
99  * Maximum number of rings/queues blkback supports, allow as many queues as there
100  * are CPUs if user has not specified a value.
101  */
102 unsigned int xenblk_max_queues;
103 module_param_named(max_queues, xenblk_max_queues, uint, 0644);
104 MODULE_PARM_DESC(max_queues,
105 		 "Maximum number of hardware queues per virtual disk." \
106 		 "By default it is the number of online CPUs.");
107 
108 /*
109  * Maximum order of pages to be used for the shared ring between front and
110  * backend, 4KB page granularity is used.
111  */
112 unsigned int xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
113 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, 0444);
114 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
115 /*
116  * The LRU mechanism to clean the lists of persistent grants needs to
117  * be executed periodically. The time interval between consecutive executions
118  * of the purge mechanism is set in ms.
119  */
120 #define LRU_INTERVAL 100
121 
122 /*
123  * When the persistent grants list is full we will remove unused grants
124  * from the list. The percent number of grants to be removed at each LRU
125  * execution.
126  */
127 #define LRU_PERCENT_CLEAN 5
128 
129 /* Run-time switchable: /sys/module/blkback/parameters/ */
130 static unsigned int log_stats;
131 module_param(log_stats, int, 0644);
132 
133 #define BLKBACK_INVALID_HANDLE (~0)
134 
135 static inline bool persistent_gnt_timeout(struct persistent_gnt *persistent_gnt)
136 {
137 	return pgrant_timeout && (jiffies - persistent_gnt->last_used >=
138 			HZ * pgrant_timeout);
139 }
140 
141 #define vaddr(page) ((unsigned long)pfn_to_kaddr(page_to_pfn(page)))
142 
143 static int do_block_io_op(struct xen_blkif_ring *ring, unsigned int *eoi_flags);
144 static int dispatch_rw_block_io(struct xen_blkif_ring *ring,
145 				struct blkif_request *req,
146 				struct pending_req *pending_req);
147 static void make_response(struct xen_blkif_ring *ring, u64 id,
148 			  unsigned short op, int st);
149 
150 #define foreach_grant_safe(pos, n, rbtree, node) \
151 	for ((pos) = container_of(rb_first((rbtree)), typeof(*(pos)), node), \
152 	     (n) = (&(pos)->node != NULL) ? rb_next(&(pos)->node) : NULL; \
153 	     &(pos)->node != NULL; \
154 	     (pos) = container_of(n, typeof(*(pos)), node), \
155 	     (n) = (&(pos)->node != NULL) ? rb_next(&(pos)->node) : NULL)
156 
157 
158 /*
159  * We don't need locking around the persistent grant helpers
160  * because blkback uses a single-thread for each backend, so we
161  * can be sure that this functions will never be called recursively.
162  *
163  * The only exception to that is put_persistent_grant, that can be called
164  * from interrupt context (by xen_blkbk_unmap), so we have to use atomic
165  * bit operations to modify the flags of a persistent grant and to count
166  * the number of used grants.
167  */
168 static int add_persistent_gnt(struct xen_blkif_ring *ring,
169 			       struct persistent_gnt *persistent_gnt)
170 {
171 	struct rb_node **new = NULL, *parent = NULL;
172 	struct persistent_gnt *this;
173 	struct xen_blkif *blkif = ring->blkif;
174 
175 	if (ring->persistent_gnt_c >= max_pgrants) {
176 		if (!blkif->vbd.overflow_max_grants)
177 			blkif->vbd.overflow_max_grants = 1;
178 		return -EBUSY;
179 	}
180 	/* Figure out where to put new node */
181 	new = &ring->persistent_gnts.rb_node;
182 	while (*new) {
183 		this = container_of(*new, struct persistent_gnt, node);
184 
185 		parent = *new;
186 		if (persistent_gnt->gnt < this->gnt)
187 			new = &((*new)->rb_left);
188 		else if (persistent_gnt->gnt > this->gnt)
189 			new = &((*new)->rb_right);
190 		else {
191 			pr_alert_ratelimited("trying to add a gref that's already in the tree\n");
192 			return -EINVAL;
193 		}
194 	}
195 
196 	persistent_gnt->active = true;
197 	/* Add new node and rebalance tree. */
198 	rb_link_node(&(persistent_gnt->node), parent, new);
199 	rb_insert_color(&(persistent_gnt->node), &ring->persistent_gnts);
200 	ring->persistent_gnt_c++;
201 	atomic_inc(&ring->persistent_gnt_in_use);
202 	return 0;
203 }
204 
205 static struct persistent_gnt *get_persistent_gnt(struct xen_blkif_ring *ring,
206 						 grant_ref_t gref)
207 {
208 	struct persistent_gnt *data;
209 	struct rb_node *node = NULL;
210 
211 	node = ring->persistent_gnts.rb_node;
212 	while (node) {
213 		data = container_of(node, struct persistent_gnt, node);
214 
215 		if (gref < data->gnt)
216 			node = node->rb_left;
217 		else if (gref > data->gnt)
218 			node = node->rb_right;
219 		else {
220 			if (data->active) {
221 				pr_alert_ratelimited("requesting a grant already in use\n");
222 				return NULL;
223 			}
224 			data->active = true;
225 			atomic_inc(&ring->persistent_gnt_in_use);
226 			return data;
227 		}
228 	}
229 	return NULL;
230 }
231 
232 static void put_persistent_gnt(struct xen_blkif_ring *ring,
233                                struct persistent_gnt *persistent_gnt)
234 {
235 	if (!persistent_gnt->active)
236 		pr_alert_ratelimited("freeing a grant already unused\n");
237 	persistent_gnt->last_used = jiffies;
238 	persistent_gnt->active = false;
239 	atomic_dec(&ring->persistent_gnt_in_use);
240 }
241 
242 static void free_persistent_gnts(struct xen_blkif_ring *ring, struct rb_root *root,
243                                  unsigned int num)
244 {
245 	struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST];
246 	struct page *pages[BLKIF_MAX_SEGMENTS_PER_REQUEST];
247 	struct persistent_gnt *persistent_gnt;
248 	struct rb_node *n;
249 	int segs_to_unmap = 0;
250 	struct gntab_unmap_queue_data unmap_data;
251 
252 	unmap_data.pages = pages;
253 	unmap_data.unmap_ops = unmap;
254 	unmap_data.kunmap_ops = NULL;
255 
256 	foreach_grant_safe(persistent_gnt, n, root, node) {
257 		BUG_ON(persistent_gnt->handle ==
258 			BLKBACK_INVALID_HANDLE);
259 		gnttab_set_unmap_op(&unmap[segs_to_unmap],
260 			(unsigned long) pfn_to_kaddr(page_to_pfn(
261 				persistent_gnt->page)),
262 			GNTMAP_host_map,
263 			persistent_gnt->handle);
264 
265 		pages[segs_to_unmap] = persistent_gnt->page;
266 
267 		if (++segs_to_unmap == BLKIF_MAX_SEGMENTS_PER_REQUEST ||
268 			!rb_next(&persistent_gnt->node)) {
269 
270 			unmap_data.count = segs_to_unmap;
271 			BUG_ON(gnttab_unmap_refs_sync(&unmap_data));
272 
273 			gnttab_page_cache_put(&ring->free_pages, pages,
274 					      segs_to_unmap);
275 			segs_to_unmap = 0;
276 		}
277 
278 		rb_erase(&persistent_gnt->node, root);
279 		kfree(persistent_gnt);
280 		num--;
281 	}
282 	BUG_ON(num != 0);
283 }
284 
285 void xen_blkbk_unmap_purged_grants(struct work_struct *work)
286 {
287 	struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST];
288 	struct page *pages[BLKIF_MAX_SEGMENTS_PER_REQUEST];
289 	struct persistent_gnt *persistent_gnt;
290 	int segs_to_unmap = 0;
291 	struct xen_blkif_ring *ring = container_of(work, typeof(*ring), persistent_purge_work);
292 	struct gntab_unmap_queue_data unmap_data;
293 
294 	unmap_data.pages = pages;
295 	unmap_data.unmap_ops = unmap;
296 	unmap_data.kunmap_ops = NULL;
297 
298 	while(!list_empty(&ring->persistent_purge_list)) {
299 		persistent_gnt = list_first_entry(&ring->persistent_purge_list,
300 		                                  struct persistent_gnt,
301 		                                  remove_node);
302 		list_del(&persistent_gnt->remove_node);
303 
304 		gnttab_set_unmap_op(&unmap[segs_to_unmap],
305 			vaddr(persistent_gnt->page),
306 			GNTMAP_host_map,
307 			persistent_gnt->handle);
308 
309 		pages[segs_to_unmap] = persistent_gnt->page;
310 
311 		if (++segs_to_unmap == BLKIF_MAX_SEGMENTS_PER_REQUEST) {
312 			unmap_data.count = segs_to_unmap;
313 			BUG_ON(gnttab_unmap_refs_sync(&unmap_data));
314 			gnttab_page_cache_put(&ring->free_pages, pages,
315 					      segs_to_unmap);
316 			segs_to_unmap = 0;
317 		}
318 		kfree(persistent_gnt);
319 	}
320 	if (segs_to_unmap > 0) {
321 		unmap_data.count = segs_to_unmap;
322 		BUG_ON(gnttab_unmap_refs_sync(&unmap_data));
323 		gnttab_page_cache_put(&ring->free_pages, pages, segs_to_unmap);
324 	}
325 }
326 
327 static void purge_persistent_gnt(struct xen_blkif_ring *ring)
328 {
329 	struct persistent_gnt *persistent_gnt;
330 	struct rb_node *n;
331 	unsigned int num_clean, total;
332 	bool scan_used = false;
333 	struct rb_root *root;
334 
335 	if (work_busy(&ring->persistent_purge_work)) {
336 		pr_alert_ratelimited("Scheduled work from previous purge is still busy, cannot purge list\n");
337 		goto out;
338 	}
339 
340 	if (ring->persistent_gnt_c < max_pgrants ||
341 	    (ring->persistent_gnt_c == max_pgrants &&
342 	    !ring->blkif->vbd.overflow_max_grants)) {
343 		num_clean = 0;
344 	} else {
345 		num_clean = (max_pgrants / 100) * LRU_PERCENT_CLEAN;
346 		num_clean = ring->persistent_gnt_c - max_pgrants + num_clean;
347 		num_clean = min(ring->persistent_gnt_c, num_clean);
348 		pr_debug("Going to purge at least %u persistent grants\n",
349 			 num_clean);
350 	}
351 
352 	/*
353 	 * At this point, we can assure that there will be no calls
354          * to get_persistent_grant (because we are executing this code from
355          * xen_blkif_schedule), there can only be calls to put_persistent_gnt,
356          * which means that the number of currently used grants will go down,
357          * but never up, so we will always be able to remove the requested
358          * number of grants.
359 	 */
360 
361 	total = 0;
362 
363 	BUG_ON(!list_empty(&ring->persistent_purge_list));
364 	root = &ring->persistent_gnts;
365 purge_list:
366 	foreach_grant_safe(persistent_gnt, n, root, node) {
367 		BUG_ON(persistent_gnt->handle ==
368 			BLKBACK_INVALID_HANDLE);
369 
370 		if (persistent_gnt->active)
371 			continue;
372 		if (!scan_used && !persistent_gnt_timeout(persistent_gnt))
373 			continue;
374 		if (scan_used && total >= num_clean)
375 			continue;
376 
377 		rb_erase(&persistent_gnt->node, root);
378 		list_add(&persistent_gnt->remove_node,
379 			 &ring->persistent_purge_list);
380 		total++;
381 	}
382 	/*
383 	 * Check whether we also need to start cleaning
384 	 * grants that were used since last purge in order to cope
385 	 * with the requested num
386 	 */
387 	if (!scan_used && total < num_clean) {
388 		pr_debug("Still missing %u purged frames\n", num_clean - total);
389 		scan_used = true;
390 		goto purge_list;
391 	}
392 
393 	if (total) {
394 		ring->persistent_gnt_c -= total;
395 		ring->blkif->vbd.overflow_max_grants = 0;
396 
397 		/* We can defer this work */
398 		schedule_work(&ring->persistent_purge_work);
399 		pr_debug("Purged %u/%u\n", num_clean, total);
400 	}
401 
402 out:
403 	return;
404 }
405 
406 /*
407  * Retrieve from the 'pending_reqs' a free pending_req structure to be used.
408  */
409 static struct pending_req *alloc_req(struct xen_blkif_ring *ring)
410 {
411 	struct pending_req *req = NULL;
412 	unsigned long flags;
413 
414 	spin_lock_irqsave(&ring->pending_free_lock, flags);
415 	if (!list_empty(&ring->pending_free)) {
416 		req = list_entry(ring->pending_free.next, struct pending_req,
417 				 free_list);
418 		list_del(&req->free_list);
419 	}
420 	spin_unlock_irqrestore(&ring->pending_free_lock, flags);
421 	return req;
422 }
423 
424 /*
425  * Return the 'pending_req' structure back to the freepool. We also
426  * wake up the thread if it was waiting for a free page.
427  */
428 static void free_req(struct xen_blkif_ring *ring, struct pending_req *req)
429 {
430 	unsigned long flags;
431 	int was_empty;
432 
433 	spin_lock_irqsave(&ring->pending_free_lock, flags);
434 	was_empty = list_empty(&ring->pending_free);
435 	list_add(&req->free_list, &ring->pending_free);
436 	spin_unlock_irqrestore(&ring->pending_free_lock, flags);
437 	if (was_empty)
438 		wake_up(&ring->pending_free_wq);
439 }
440 
441 /*
442  * Routines for managing virtual block devices (vbds).
443  */
444 static int xen_vbd_translate(struct phys_req *req, struct xen_blkif *blkif,
445 			     int operation)
446 {
447 	struct xen_vbd *vbd = &blkif->vbd;
448 	int rc = -EACCES;
449 
450 	if ((operation != REQ_OP_READ) && vbd->readonly)
451 		goto out;
452 
453 	if (likely(req->nr_sects)) {
454 		blkif_sector_t end = req->sector_number + req->nr_sects;
455 
456 		if (unlikely(end < req->sector_number))
457 			goto out;
458 		if (unlikely(end > vbd_sz(vbd)))
459 			goto out;
460 	}
461 
462 	req->dev  = vbd->pdevice;
463 	req->bdev = vbd->bdev;
464 	rc = 0;
465 
466  out:
467 	return rc;
468 }
469 
470 static void xen_vbd_resize(struct xen_blkif *blkif)
471 {
472 	struct xen_vbd *vbd = &blkif->vbd;
473 	struct xenbus_transaction xbt;
474 	int err;
475 	struct xenbus_device *dev = xen_blkbk_xenbus(blkif->be);
476 	unsigned long long new_size = vbd_sz(vbd);
477 
478 	pr_info("VBD Resize: Domid: %d, Device: (%d, %d)\n",
479 		blkif->domid, MAJOR(vbd->pdevice), MINOR(vbd->pdevice));
480 	pr_info("VBD Resize: new size %llu\n", new_size);
481 	vbd->size = new_size;
482 again:
483 	err = xenbus_transaction_start(&xbt);
484 	if (err) {
485 		pr_warn("Error starting transaction\n");
486 		return;
487 	}
488 	err = xenbus_printf(xbt, dev->nodename, "sectors", "%llu",
489 			    (unsigned long long)vbd_sz(vbd));
490 	if (err) {
491 		pr_warn("Error writing new size\n");
492 		goto abort;
493 	}
494 	/*
495 	 * Write the current state; we will use this to synchronize
496 	 * the front-end. If the current state is "connected" the
497 	 * front-end will get the new size information online.
498 	 */
499 	err = xenbus_printf(xbt, dev->nodename, "state", "%d", dev->state);
500 	if (err) {
501 		pr_warn("Error writing the state\n");
502 		goto abort;
503 	}
504 
505 	err = xenbus_transaction_end(xbt, 0);
506 	if (err == -EAGAIN)
507 		goto again;
508 	if (err)
509 		pr_warn("Error ending transaction\n");
510 	return;
511 abort:
512 	xenbus_transaction_end(xbt, 1);
513 }
514 
515 /*
516  * Notification from the guest OS.
517  */
518 static void blkif_notify_work(struct xen_blkif_ring *ring)
519 {
520 	ring->waiting_reqs = 1;
521 	wake_up(&ring->wq);
522 }
523 
524 irqreturn_t xen_blkif_be_int(int irq, void *dev_id)
525 {
526 	blkif_notify_work(dev_id);
527 	return IRQ_HANDLED;
528 }
529 
530 /*
531  * SCHEDULER FUNCTIONS
532  */
533 
534 static void print_stats(struct xen_blkif_ring *ring)
535 {
536 	pr_info("(%s): oo %3llu  |  rd %4llu  |  wr %4llu  |  f %4llu"
537 		 "  |  ds %4llu | pg: %4u/%4d\n",
538 		 current->comm, ring->st_oo_req,
539 		 ring->st_rd_req, ring->st_wr_req,
540 		 ring->st_f_req, ring->st_ds_req,
541 		 ring->persistent_gnt_c, max_pgrants);
542 	ring->st_print = jiffies + msecs_to_jiffies(10 * 1000);
543 	ring->st_rd_req = 0;
544 	ring->st_wr_req = 0;
545 	ring->st_oo_req = 0;
546 	ring->st_ds_req = 0;
547 }
548 
549 int xen_blkif_schedule(void *arg)
550 {
551 	struct xen_blkif_ring *ring = arg;
552 	struct xen_blkif *blkif = ring->blkif;
553 	struct xen_vbd *vbd = &blkif->vbd;
554 	unsigned long timeout;
555 	int ret;
556 	bool do_eoi;
557 	unsigned int eoi_flags = XEN_EOI_FLAG_SPURIOUS;
558 
559 	set_freezable();
560 	while (!kthread_should_stop()) {
561 		if (try_to_freeze())
562 			continue;
563 		if (unlikely(vbd->size != vbd_sz(vbd)))
564 			xen_vbd_resize(blkif);
565 
566 		timeout = msecs_to_jiffies(LRU_INTERVAL);
567 
568 		timeout = wait_event_interruptible_timeout(
569 			ring->wq,
570 			ring->waiting_reqs || kthread_should_stop(),
571 			timeout);
572 		if (timeout == 0)
573 			goto purge_gnt_list;
574 		timeout = wait_event_interruptible_timeout(
575 			ring->pending_free_wq,
576 			!list_empty(&ring->pending_free) ||
577 			kthread_should_stop(),
578 			timeout);
579 		if (timeout == 0)
580 			goto purge_gnt_list;
581 
582 		do_eoi = ring->waiting_reqs;
583 
584 		ring->waiting_reqs = 0;
585 		smp_mb(); /* clear flag *before* checking for work */
586 
587 		ret = do_block_io_op(ring, &eoi_flags);
588 		if (ret > 0)
589 			ring->waiting_reqs = 1;
590 		if (ret == -EACCES)
591 			wait_event_interruptible(ring->shutdown_wq,
592 						 kthread_should_stop());
593 
594 		if (do_eoi && !ring->waiting_reqs) {
595 			xen_irq_lateeoi(ring->irq, eoi_flags);
596 			eoi_flags |= XEN_EOI_FLAG_SPURIOUS;
597 		}
598 
599 purge_gnt_list:
600 		if (blkif->vbd.feature_gnt_persistent &&
601 		    time_after(jiffies, ring->next_lru)) {
602 			purge_persistent_gnt(ring);
603 			ring->next_lru = jiffies + msecs_to_jiffies(LRU_INTERVAL);
604 		}
605 
606 		/* Shrink the free pages pool if it is too large. */
607 		if (time_before(jiffies, blkif->buffer_squeeze_end))
608 			gnttab_page_cache_shrink(&ring->free_pages, 0);
609 		else
610 			gnttab_page_cache_shrink(&ring->free_pages,
611 						 max_buffer_pages);
612 
613 		if (log_stats && time_after(jiffies, ring->st_print))
614 			print_stats(ring);
615 	}
616 
617 	/* Drain pending purge work */
618 	flush_work(&ring->persistent_purge_work);
619 
620 	if (log_stats)
621 		print_stats(ring);
622 
623 	ring->xenblkd = NULL;
624 
625 	return 0;
626 }
627 
628 /*
629  * Remove persistent grants and empty the pool of free pages
630  */
631 void xen_blkbk_free_caches(struct xen_blkif_ring *ring)
632 {
633 	/* Free all persistent grant pages */
634 	if (!RB_EMPTY_ROOT(&ring->persistent_gnts))
635 		free_persistent_gnts(ring, &ring->persistent_gnts,
636 			ring->persistent_gnt_c);
637 
638 	BUG_ON(!RB_EMPTY_ROOT(&ring->persistent_gnts));
639 	ring->persistent_gnt_c = 0;
640 
641 	/* Since we are shutting down remove all pages from the buffer */
642 	gnttab_page_cache_shrink(&ring->free_pages, 0 /* All */);
643 }
644 
645 static unsigned int xen_blkbk_unmap_prepare(
646 	struct xen_blkif_ring *ring,
647 	struct grant_page **pages,
648 	unsigned int num,
649 	struct gnttab_unmap_grant_ref *unmap_ops,
650 	struct page **unmap_pages)
651 {
652 	unsigned int i, invcount = 0;
653 
654 	for (i = 0; i < num; i++) {
655 		if (pages[i]->persistent_gnt != NULL) {
656 			put_persistent_gnt(ring, pages[i]->persistent_gnt);
657 			continue;
658 		}
659 		if (pages[i]->handle == BLKBACK_INVALID_HANDLE)
660 			continue;
661 		unmap_pages[invcount] = pages[i]->page;
662 		gnttab_set_unmap_op(&unmap_ops[invcount], vaddr(pages[i]->page),
663 				    GNTMAP_host_map, pages[i]->handle);
664 		pages[i]->handle = BLKBACK_INVALID_HANDLE;
665 		invcount++;
666 	}
667 
668 	return invcount;
669 }
670 
671 static void xen_blkbk_unmap_and_respond_callback(int result, struct gntab_unmap_queue_data *data)
672 {
673 	struct pending_req *pending_req = (struct pending_req *)(data->data);
674 	struct xen_blkif_ring *ring = pending_req->ring;
675 	struct xen_blkif *blkif = ring->blkif;
676 
677 	/* BUG_ON used to reproduce existing behaviour,
678 	   but is this the best way to deal with this? */
679 	BUG_ON(result);
680 
681 	gnttab_page_cache_put(&ring->free_pages, data->pages, data->count);
682 	make_response(ring, pending_req->id,
683 		      pending_req->operation, pending_req->status);
684 	free_req(ring, pending_req);
685 	/*
686 	 * Make sure the request is freed before releasing blkif,
687 	 * or there could be a race between free_req and the
688 	 * cleanup done in xen_blkif_free during shutdown.
689 	 *
690 	 * NB: The fact that we might try to wake up pending_free_wq
691 	 * before drain_complete (in case there's a drain going on)
692 	 * it's not a problem with our current implementation
693 	 * because we can assure there's no thread waiting on
694 	 * pending_free_wq if there's a drain going on, but it has
695 	 * to be taken into account if the current model is changed.
696 	 */
697 	if (atomic_dec_and_test(&ring->inflight) && atomic_read(&blkif->drain)) {
698 		complete(&blkif->drain_complete);
699 	}
700 	xen_blkif_put(blkif);
701 }
702 
703 static void xen_blkbk_unmap_and_respond(struct pending_req *req)
704 {
705 	struct gntab_unmap_queue_data* work = &req->gnttab_unmap_data;
706 	struct xen_blkif_ring *ring = req->ring;
707 	struct grant_page **pages = req->segments;
708 	unsigned int invcount;
709 
710 	invcount = xen_blkbk_unmap_prepare(ring, pages, req->nr_segs,
711 					   req->unmap, req->unmap_pages);
712 
713 	work->data = req;
714 	work->done = xen_blkbk_unmap_and_respond_callback;
715 	work->unmap_ops = req->unmap;
716 	work->kunmap_ops = NULL;
717 	work->pages = req->unmap_pages;
718 	work->count = invcount;
719 
720 	gnttab_unmap_refs_async(&req->gnttab_unmap_data);
721 }
722 
723 
724 /*
725  * Unmap the grant references.
726  *
727  * This could accumulate ops up to the batch size to reduce the number
728  * of hypercalls, but since this is only used in error paths there's
729  * no real need.
730  */
731 static void xen_blkbk_unmap(struct xen_blkif_ring *ring,
732                             struct grant_page *pages[],
733                             int num)
734 {
735 	struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST];
736 	struct page *unmap_pages[BLKIF_MAX_SEGMENTS_PER_REQUEST];
737 	unsigned int invcount = 0;
738 	int ret;
739 
740 	while (num) {
741 		unsigned int batch = min(num, BLKIF_MAX_SEGMENTS_PER_REQUEST);
742 
743 		invcount = xen_blkbk_unmap_prepare(ring, pages, batch,
744 						   unmap, unmap_pages);
745 		if (invcount) {
746 			ret = gnttab_unmap_refs(unmap, NULL, unmap_pages, invcount);
747 			BUG_ON(ret);
748 			gnttab_page_cache_put(&ring->free_pages, unmap_pages,
749 					      invcount);
750 		}
751 		pages += batch;
752 		num -= batch;
753 	}
754 }
755 
756 static int xen_blkbk_map(struct xen_blkif_ring *ring,
757 			 struct grant_page *pages[],
758 			 int num, bool ro)
759 {
760 	struct gnttab_map_grant_ref map[BLKIF_MAX_SEGMENTS_PER_REQUEST];
761 	struct page *pages_to_gnt[BLKIF_MAX_SEGMENTS_PER_REQUEST];
762 	struct persistent_gnt *persistent_gnt = NULL;
763 	phys_addr_t addr = 0;
764 	int i, seg_idx, new_map_idx;
765 	int segs_to_map = 0;
766 	int ret = 0;
767 	int last_map = 0, map_until = 0;
768 	int use_persistent_gnts;
769 	struct xen_blkif *blkif = ring->blkif;
770 
771 	use_persistent_gnts = (blkif->vbd.feature_gnt_persistent);
772 
773 	/*
774 	 * Fill out preq.nr_sects with proper amount of sectors, and setup
775 	 * assign map[..] with the PFN of the page in our domain with the
776 	 * corresponding grant reference for each page.
777 	 */
778 again:
779 	for (i = map_until; i < num; i++) {
780 		uint32_t flags;
781 
782 		if (use_persistent_gnts) {
783 			persistent_gnt = get_persistent_gnt(
784 				ring,
785 				pages[i]->gref);
786 		}
787 
788 		if (persistent_gnt) {
789 			/*
790 			 * We are using persistent grants and
791 			 * the grant is already mapped
792 			 */
793 			pages[i]->page = persistent_gnt->page;
794 			pages[i]->persistent_gnt = persistent_gnt;
795 		} else {
796 			if (gnttab_page_cache_get(&ring->free_pages,
797 						  &pages[i]->page)) {
798 				gnttab_page_cache_put(&ring->free_pages,
799 						      pages_to_gnt,
800 						      segs_to_map);
801 				ret = -ENOMEM;
802 				goto out;
803 			}
804 			addr = vaddr(pages[i]->page);
805 			pages_to_gnt[segs_to_map] = pages[i]->page;
806 			pages[i]->persistent_gnt = NULL;
807 			flags = GNTMAP_host_map;
808 			if (!use_persistent_gnts && ro)
809 				flags |= GNTMAP_readonly;
810 			gnttab_set_map_op(&map[segs_to_map++], addr,
811 					  flags, pages[i]->gref,
812 					  blkif->domid);
813 		}
814 		map_until = i + 1;
815 		if (segs_to_map == BLKIF_MAX_SEGMENTS_PER_REQUEST)
816 			break;
817 	}
818 
819 	if (segs_to_map)
820 		ret = gnttab_map_refs(map, NULL, pages_to_gnt, segs_to_map);
821 
822 	/*
823 	 * Now swizzle the MFN in our domain with the MFN from the other domain
824 	 * so that when we access vaddr(pending_req,i) it has the contents of
825 	 * the page from the other domain.
826 	 */
827 	for (seg_idx = last_map, new_map_idx = 0; seg_idx < map_until; seg_idx++) {
828 		if (!pages[seg_idx]->persistent_gnt) {
829 			/* This is a newly mapped grant */
830 			BUG_ON(new_map_idx >= segs_to_map);
831 			if (unlikely(map[new_map_idx].status != 0)) {
832 				pr_debug("invalid buffer -- could not remap it\n");
833 				gnttab_page_cache_put(&ring->free_pages,
834 						      &pages[seg_idx]->page, 1);
835 				pages[seg_idx]->handle = BLKBACK_INVALID_HANDLE;
836 				ret |= !ret;
837 				goto next;
838 			}
839 			pages[seg_idx]->handle = map[new_map_idx].handle;
840 		} else {
841 			continue;
842 		}
843 		if (use_persistent_gnts &&
844 		    ring->persistent_gnt_c < max_pgrants) {
845 			/*
846 			 * We are using persistent grants, the grant is
847 			 * not mapped but we might have room for it.
848 			 */
849 			persistent_gnt = kmalloc(sizeof(struct persistent_gnt),
850 				                 GFP_KERNEL);
851 			if (!persistent_gnt) {
852 				/*
853 				 * If we don't have enough memory to
854 				 * allocate the persistent_gnt struct
855 				 * map this grant non-persistenly
856 				 */
857 				goto next;
858 			}
859 			persistent_gnt->gnt = map[new_map_idx].ref;
860 			persistent_gnt->handle = map[new_map_idx].handle;
861 			persistent_gnt->page = pages[seg_idx]->page;
862 			if (add_persistent_gnt(ring,
863 			                       persistent_gnt)) {
864 				kfree(persistent_gnt);
865 				persistent_gnt = NULL;
866 				goto next;
867 			}
868 			pages[seg_idx]->persistent_gnt = persistent_gnt;
869 			pr_debug("grant %u added to the tree of persistent grants, using %u/%u\n",
870 				 persistent_gnt->gnt, ring->persistent_gnt_c,
871 				 max_pgrants);
872 			goto next;
873 		}
874 		if (use_persistent_gnts && !blkif->vbd.overflow_max_grants) {
875 			blkif->vbd.overflow_max_grants = 1;
876 			pr_debug("domain %u, device %#x is using maximum number of persistent grants\n",
877 			         blkif->domid, blkif->vbd.handle);
878 		}
879 		/*
880 		 * We could not map this grant persistently, so use it as
881 		 * a non-persistent grant.
882 		 */
883 next:
884 		new_map_idx++;
885 	}
886 	segs_to_map = 0;
887 	last_map = map_until;
888 	if (!ret && map_until != num)
889 		goto again;
890 
891 out:
892 	for (i = last_map; i < num; i++) {
893 		/* Don't zap current batch's valid persistent grants. */
894 		if(i >= map_until)
895 			pages[i]->persistent_gnt = NULL;
896 		pages[i]->handle = BLKBACK_INVALID_HANDLE;
897 	}
898 
899 	return ret;
900 }
901 
902 static int xen_blkbk_map_seg(struct pending_req *pending_req)
903 {
904 	int rc;
905 
906 	rc = xen_blkbk_map(pending_req->ring, pending_req->segments,
907 			   pending_req->nr_segs,
908 	                   (pending_req->operation != BLKIF_OP_READ));
909 
910 	return rc;
911 }
912 
913 static int xen_blkbk_parse_indirect(struct blkif_request *req,
914 				    struct pending_req *pending_req,
915 				    struct seg_buf seg[],
916 				    struct phys_req *preq)
917 {
918 	struct grant_page **pages = pending_req->indirect_pages;
919 	struct xen_blkif_ring *ring = pending_req->ring;
920 	int indirect_grefs, rc, n, nseg, i;
921 	struct blkif_request_segment *segments = NULL;
922 
923 	nseg = pending_req->nr_segs;
924 	indirect_grefs = INDIRECT_PAGES(nseg);
925 	BUG_ON(indirect_grefs > BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST);
926 
927 	for (i = 0; i < indirect_grefs; i++)
928 		pages[i]->gref = req->u.indirect.indirect_grefs[i];
929 
930 	rc = xen_blkbk_map(ring, pages, indirect_grefs, true);
931 	if (rc)
932 		goto unmap;
933 
934 	for (n = 0; n < nseg; n++) {
935 		uint8_t first_sect, last_sect;
936 
937 		if ((n % SEGS_PER_INDIRECT_FRAME) == 0) {
938 			/* Map indirect segments */
939 			if (segments)
940 				kunmap_atomic(segments);
941 			segments = kmap_atomic(pages[n/SEGS_PER_INDIRECT_FRAME]->page);
942 		}
943 		i = n % SEGS_PER_INDIRECT_FRAME;
944 
945 		pending_req->segments[n]->gref = segments[i].gref;
946 
947 		first_sect = READ_ONCE(segments[i].first_sect);
948 		last_sect = READ_ONCE(segments[i].last_sect);
949 		if (last_sect >= (XEN_PAGE_SIZE >> 9) || last_sect < first_sect) {
950 			rc = -EINVAL;
951 			goto unmap;
952 		}
953 
954 		seg[n].nsec = last_sect - first_sect + 1;
955 		seg[n].offset = first_sect << 9;
956 		preq->nr_sects += seg[n].nsec;
957 	}
958 
959 unmap:
960 	if (segments)
961 		kunmap_atomic(segments);
962 	xen_blkbk_unmap(ring, pages, indirect_grefs);
963 	return rc;
964 }
965 
966 static int dispatch_discard_io(struct xen_blkif_ring *ring,
967 				struct blkif_request *req)
968 {
969 	int err = 0;
970 	int status = BLKIF_RSP_OKAY;
971 	struct xen_blkif *blkif = ring->blkif;
972 	struct block_device *bdev = blkif->vbd.bdev;
973 	unsigned long secure;
974 	struct phys_req preq;
975 
976 	xen_blkif_get(blkif);
977 
978 	preq.sector_number = req->u.discard.sector_number;
979 	preq.nr_sects      = req->u.discard.nr_sectors;
980 
981 	err = xen_vbd_translate(&preq, blkif, REQ_OP_WRITE);
982 	if (err) {
983 		pr_warn("access denied: DISCARD [%llu->%llu] on dev=%04x\n",
984 			preq.sector_number,
985 			preq.sector_number + preq.nr_sects, blkif->vbd.pdevice);
986 		goto fail_response;
987 	}
988 	ring->st_ds_req++;
989 
990 	secure = (blkif->vbd.discard_secure &&
991 		 (req->u.discard.flag & BLKIF_DISCARD_SECURE)) ?
992 		 BLKDEV_DISCARD_SECURE : 0;
993 
994 	err = blkdev_issue_discard(bdev, req->u.discard.sector_number,
995 				   req->u.discard.nr_sectors,
996 				   GFP_KERNEL, secure);
997 fail_response:
998 	if (err == -EOPNOTSUPP) {
999 		pr_debug("discard op failed, not supported\n");
1000 		status = BLKIF_RSP_EOPNOTSUPP;
1001 	} else if (err)
1002 		status = BLKIF_RSP_ERROR;
1003 
1004 	make_response(ring, req->u.discard.id, req->operation, status);
1005 	xen_blkif_put(blkif);
1006 	return err;
1007 }
1008 
1009 static int dispatch_other_io(struct xen_blkif_ring *ring,
1010 			     struct blkif_request *req,
1011 			     struct pending_req *pending_req)
1012 {
1013 	free_req(ring, pending_req);
1014 	make_response(ring, req->u.other.id, req->operation,
1015 		      BLKIF_RSP_EOPNOTSUPP);
1016 	return -EIO;
1017 }
1018 
1019 static void xen_blk_drain_io(struct xen_blkif_ring *ring)
1020 {
1021 	struct xen_blkif *blkif = ring->blkif;
1022 
1023 	atomic_set(&blkif->drain, 1);
1024 	do {
1025 		if (atomic_read(&ring->inflight) == 0)
1026 			break;
1027 		wait_for_completion_interruptible_timeout(
1028 				&blkif->drain_complete, HZ);
1029 
1030 		if (!atomic_read(&blkif->drain))
1031 			break;
1032 	} while (!kthread_should_stop());
1033 	atomic_set(&blkif->drain, 0);
1034 }
1035 
1036 static void __end_block_io_op(struct pending_req *pending_req,
1037 		blk_status_t error)
1038 {
1039 	/* An error fails the entire request. */
1040 	if (pending_req->operation == BLKIF_OP_FLUSH_DISKCACHE &&
1041 	    error == BLK_STS_NOTSUPP) {
1042 		pr_debug("flush diskcache op failed, not supported\n");
1043 		xen_blkbk_flush_diskcache(XBT_NIL, pending_req->ring->blkif->be, 0);
1044 		pending_req->status = BLKIF_RSP_EOPNOTSUPP;
1045 	} else if (pending_req->operation == BLKIF_OP_WRITE_BARRIER &&
1046 		   error == BLK_STS_NOTSUPP) {
1047 		pr_debug("write barrier op failed, not supported\n");
1048 		xen_blkbk_barrier(XBT_NIL, pending_req->ring->blkif->be, 0);
1049 		pending_req->status = BLKIF_RSP_EOPNOTSUPP;
1050 	} else if (error) {
1051 		pr_debug("Buffer not up-to-date at end of operation,"
1052 			 " error=%d\n", error);
1053 		pending_req->status = BLKIF_RSP_ERROR;
1054 	}
1055 
1056 	/*
1057 	 * If all of the bio's have completed it is time to unmap
1058 	 * the grant references associated with 'request' and provide
1059 	 * the proper response on the ring.
1060 	 */
1061 	if (atomic_dec_and_test(&pending_req->pendcnt))
1062 		xen_blkbk_unmap_and_respond(pending_req);
1063 }
1064 
1065 /*
1066  * bio callback.
1067  */
1068 static void end_block_io_op(struct bio *bio)
1069 {
1070 	__end_block_io_op(bio->bi_private, bio->bi_status);
1071 	bio_put(bio);
1072 }
1073 
1074 
1075 
1076 /*
1077  * Function to copy the from the ring buffer the 'struct blkif_request'
1078  * (which has the sectors we want, number of them, grant references, etc),
1079  * and transmute  it to the block API to hand it over to the proper block disk.
1080  */
1081 static int
1082 __do_block_io_op(struct xen_blkif_ring *ring, unsigned int *eoi_flags)
1083 {
1084 	union blkif_back_rings *blk_rings = &ring->blk_rings;
1085 	struct blkif_request req;
1086 	struct pending_req *pending_req;
1087 	RING_IDX rc, rp;
1088 	int more_to_do = 0;
1089 
1090 	rc = blk_rings->common.req_cons;
1091 	rp = blk_rings->common.sring->req_prod;
1092 	rmb(); /* Ensure we see queued requests up to 'rp'. */
1093 
1094 	if (RING_REQUEST_PROD_OVERFLOW(&blk_rings->common, rp)) {
1095 		rc = blk_rings->common.rsp_prod_pvt;
1096 		pr_warn("Frontend provided bogus ring requests (%d - %d = %d). Halting ring processing on dev=%04x\n",
1097 			rp, rc, rp - rc, ring->blkif->vbd.pdevice);
1098 		return -EACCES;
1099 	}
1100 	while (rc != rp) {
1101 
1102 		if (RING_REQUEST_CONS_OVERFLOW(&blk_rings->common, rc))
1103 			break;
1104 
1105 		/* We've seen a request, so clear spurious eoi flag. */
1106 		*eoi_flags &= ~XEN_EOI_FLAG_SPURIOUS;
1107 
1108 		if (kthread_should_stop()) {
1109 			more_to_do = 1;
1110 			break;
1111 		}
1112 
1113 		pending_req = alloc_req(ring);
1114 		if (NULL == pending_req) {
1115 			ring->st_oo_req++;
1116 			more_to_do = 1;
1117 			break;
1118 		}
1119 
1120 		switch (ring->blkif->blk_protocol) {
1121 		case BLKIF_PROTOCOL_NATIVE:
1122 			memcpy(&req, RING_GET_REQUEST(&blk_rings->native, rc), sizeof(req));
1123 			break;
1124 		case BLKIF_PROTOCOL_X86_32:
1125 			blkif_get_x86_32_req(&req, RING_GET_REQUEST(&blk_rings->x86_32, rc));
1126 			break;
1127 		case BLKIF_PROTOCOL_X86_64:
1128 			blkif_get_x86_64_req(&req, RING_GET_REQUEST(&blk_rings->x86_64, rc));
1129 			break;
1130 		default:
1131 			BUG();
1132 		}
1133 		blk_rings->common.req_cons = ++rc; /* before make_response() */
1134 
1135 		/* Apply all sanity checks to /private copy/ of request. */
1136 		barrier();
1137 
1138 		switch (req.operation) {
1139 		case BLKIF_OP_READ:
1140 		case BLKIF_OP_WRITE:
1141 		case BLKIF_OP_WRITE_BARRIER:
1142 		case BLKIF_OP_FLUSH_DISKCACHE:
1143 		case BLKIF_OP_INDIRECT:
1144 			if (dispatch_rw_block_io(ring, &req, pending_req))
1145 				goto done;
1146 			break;
1147 		case BLKIF_OP_DISCARD:
1148 			free_req(ring, pending_req);
1149 			if (dispatch_discard_io(ring, &req))
1150 				goto done;
1151 			break;
1152 		default:
1153 			if (dispatch_other_io(ring, &req, pending_req))
1154 				goto done;
1155 			break;
1156 		}
1157 
1158 		/* Yield point for this unbounded loop. */
1159 		cond_resched();
1160 	}
1161 done:
1162 	return more_to_do;
1163 }
1164 
1165 static int
1166 do_block_io_op(struct xen_blkif_ring *ring, unsigned int *eoi_flags)
1167 {
1168 	union blkif_back_rings *blk_rings = &ring->blk_rings;
1169 	int more_to_do;
1170 
1171 	do {
1172 		more_to_do = __do_block_io_op(ring, eoi_flags);
1173 		if (more_to_do)
1174 			break;
1175 
1176 		RING_FINAL_CHECK_FOR_REQUESTS(&blk_rings->common, more_to_do);
1177 	} while (more_to_do);
1178 
1179 	return more_to_do;
1180 }
1181 /*
1182  * Transmutation of the 'struct blkif_request' to a proper 'struct bio'
1183  * and call the 'submit_bio' to pass it to the underlying storage.
1184  */
1185 static int dispatch_rw_block_io(struct xen_blkif_ring *ring,
1186 				struct blkif_request *req,
1187 				struct pending_req *pending_req)
1188 {
1189 	struct phys_req preq;
1190 	struct seg_buf *seg = pending_req->seg;
1191 	unsigned int nseg;
1192 	struct bio *bio = NULL;
1193 	struct bio **biolist = pending_req->biolist;
1194 	int i, nbio = 0;
1195 	int operation;
1196 	int operation_flags = 0;
1197 	struct blk_plug plug;
1198 	bool drain = false;
1199 	struct grant_page **pages = pending_req->segments;
1200 	unsigned short req_operation;
1201 
1202 	req_operation = req->operation == BLKIF_OP_INDIRECT ?
1203 			req->u.indirect.indirect_op : req->operation;
1204 
1205 	if ((req->operation == BLKIF_OP_INDIRECT) &&
1206 	    (req_operation != BLKIF_OP_READ) &&
1207 	    (req_operation != BLKIF_OP_WRITE)) {
1208 		pr_debug("Invalid indirect operation (%u)\n", req_operation);
1209 		goto fail_response;
1210 	}
1211 
1212 	switch (req_operation) {
1213 	case BLKIF_OP_READ:
1214 		ring->st_rd_req++;
1215 		operation = REQ_OP_READ;
1216 		break;
1217 	case BLKIF_OP_WRITE:
1218 		ring->st_wr_req++;
1219 		operation = REQ_OP_WRITE;
1220 		operation_flags = REQ_SYNC | REQ_IDLE;
1221 		break;
1222 	case BLKIF_OP_WRITE_BARRIER:
1223 		drain = true;
1224 		fallthrough;
1225 	case BLKIF_OP_FLUSH_DISKCACHE:
1226 		ring->st_f_req++;
1227 		operation = REQ_OP_WRITE;
1228 		operation_flags = REQ_PREFLUSH;
1229 		break;
1230 	default:
1231 		operation = 0; /* make gcc happy */
1232 		goto fail_response;
1233 		break;
1234 	}
1235 
1236 	/* Check that the number of segments is sane. */
1237 	nseg = req->operation == BLKIF_OP_INDIRECT ?
1238 	       req->u.indirect.nr_segments : req->u.rw.nr_segments;
1239 
1240 	if (unlikely(nseg == 0 && operation_flags != REQ_PREFLUSH) ||
1241 	    unlikely((req->operation != BLKIF_OP_INDIRECT) &&
1242 		     (nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST)) ||
1243 	    unlikely((req->operation == BLKIF_OP_INDIRECT) &&
1244 		     (nseg > MAX_INDIRECT_SEGMENTS))) {
1245 		pr_debug("Bad number of segments in request (%d)\n", nseg);
1246 		/* Haven't submitted any bio's yet. */
1247 		goto fail_response;
1248 	}
1249 
1250 	preq.nr_sects      = 0;
1251 
1252 	pending_req->ring      = ring;
1253 	pending_req->id        = req->u.rw.id;
1254 	pending_req->operation = req_operation;
1255 	pending_req->status    = BLKIF_RSP_OKAY;
1256 	pending_req->nr_segs   = nseg;
1257 
1258 	if (req->operation != BLKIF_OP_INDIRECT) {
1259 		preq.dev               = req->u.rw.handle;
1260 		preq.sector_number     = req->u.rw.sector_number;
1261 		for (i = 0; i < nseg; i++) {
1262 			pages[i]->gref = req->u.rw.seg[i].gref;
1263 			seg[i].nsec = req->u.rw.seg[i].last_sect -
1264 				req->u.rw.seg[i].first_sect + 1;
1265 			seg[i].offset = (req->u.rw.seg[i].first_sect << 9);
1266 			if ((req->u.rw.seg[i].last_sect >= (XEN_PAGE_SIZE >> 9)) ||
1267 			    (req->u.rw.seg[i].last_sect <
1268 			     req->u.rw.seg[i].first_sect))
1269 				goto fail_response;
1270 			preq.nr_sects += seg[i].nsec;
1271 		}
1272 	} else {
1273 		preq.dev               = req->u.indirect.handle;
1274 		preq.sector_number     = req->u.indirect.sector_number;
1275 		if (xen_blkbk_parse_indirect(req, pending_req, seg, &preq))
1276 			goto fail_response;
1277 	}
1278 
1279 	if (xen_vbd_translate(&preq, ring->blkif, operation) != 0) {
1280 		pr_debug("access denied: %s of [%llu,%llu] on dev=%04x\n",
1281 			 operation == REQ_OP_READ ? "read" : "write",
1282 			 preq.sector_number,
1283 			 preq.sector_number + preq.nr_sects,
1284 			 ring->blkif->vbd.pdevice);
1285 		goto fail_response;
1286 	}
1287 
1288 	/*
1289 	 * This check _MUST_ be done after xen_vbd_translate as the preq.bdev
1290 	 * is set there.
1291 	 */
1292 	for (i = 0; i < nseg; i++) {
1293 		if (((int)preq.sector_number|(int)seg[i].nsec) &
1294 		    ((bdev_logical_block_size(preq.bdev) >> 9) - 1)) {
1295 			pr_debug("Misaligned I/O request from domain %d\n",
1296 				 ring->blkif->domid);
1297 			goto fail_response;
1298 		}
1299 	}
1300 
1301 	/* Wait on all outstanding I/O's and once that has been completed
1302 	 * issue the flush.
1303 	 */
1304 	if (drain)
1305 		xen_blk_drain_io(pending_req->ring);
1306 
1307 	/*
1308 	 * If we have failed at this point, we need to undo the M2P override,
1309 	 * set gnttab_set_unmap_op on all of the grant references and perform
1310 	 * the hypercall to unmap the grants - that is all done in
1311 	 * xen_blkbk_unmap.
1312 	 */
1313 	if (xen_blkbk_map_seg(pending_req))
1314 		goto fail_flush;
1315 
1316 	/*
1317 	 * This corresponding xen_blkif_put is done in __end_block_io_op, or
1318 	 * below (in "!bio") if we are handling a BLKIF_OP_DISCARD.
1319 	 */
1320 	xen_blkif_get(ring->blkif);
1321 	atomic_inc(&ring->inflight);
1322 
1323 	for (i = 0; i < nseg; i++) {
1324 		while ((bio == NULL) ||
1325 		       (bio_add_page(bio,
1326 				     pages[i]->page,
1327 				     seg[i].nsec << 9,
1328 				     seg[i].offset) == 0)) {
1329 			bio = bio_alloc(preq.bdev, bio_max_segs(nseg - i),
1330 					operation | operation_flags,
1331 					GFP_KERNEL);
1332 			biolist[nbio++] = bio;
1333 			bio->bi_private = pending_req;
1334 			bio->bi_end_io  = end_block_io_op;
1335 			bio->bi_iter.bi_sector  = preq.sector_number;
1336 		}
1337 
1338 		preq.sector_number += seg[i].nsec;
1339 	}
1340 
1341 	/* This will be hit if the operation was a flush or discard. */
1342 	if (!bio) {
1343 		BUG_ON(operation_flags != REQ_PREFLUSH);
1344 
1345 		bio = bio_alloc(preq.bdev, 0, operation | operation_flags,
1346 				GFP_KERNEL);
1347 		biolist[nbio++] = bio;
1348 		bio->bi_private = pending_req;
1349 		bio->bi_end_io  = end_block_io_op;
1350 	}
1351 
1352 	atomic_set(&pending_req->pendcnt, nbio);
1353 	blk_start_plug(&plug);
1354 
1355 	for (i = 0; i < nbio; i++)
1356 		submit_bio(biolist[i]);
1357 
1358 	/* Let the I/Os go.. */
1359 	blk_finish_plug(&plug);
1360 
1361 	if (operation == REQ_OP_READ)
1362 		ring->st_rd_sect += preq.nr_sects;
1363 	else if (operation == REQ_OP_WRITE)
1364 		ring->st_wr_sect += preq.nr_sects;
1365 
1366 	return 0;
1367 
1368  fail_flush:
1369 	xen_blkbk_unmap(ring, pending_req->segments,
1370 	                pending_req->nr_segs);
1371  fail_response:
1372 	/* Haven't submitted any bio's yet. */
1373 	make_response(ring, req->u.rw.id, req_operation, BLKIF_RSP_ERROR);
1374 	free_req(ring, pending_req);
1375 	msleep(1); /* back off a bit */
1376 	return -EIO;
1377 }
1378 
1379 
1380 
1381 /*
1382  * Put a response on the ring on how the operation fared.
1383  */
1384 static void make_response(struct xen_blkif_ring *ring, u64 id,
1385 			  unsigned short op, int st)
1386 {
1387 	struct blkif_response *resp;
1388 	unsigned long     flags;
1389 	union blkif_back_rings *blk_rings;
1390 	int notify;
1391 
1392 	spin_lock_irqsave(&ring->blk_ring_lock, flags);
1393 	blk_rings = &ring->blk_rings;
1394 	/* Place on the response ring for the relevant domain. */
1395 	switch (ring->blkif->blk_protocol) {
1396 	case BLKIF_PROTOCOL_NATIVE:
1397 		resp = RING_GET_RESPONSE(&blk_rings->native,
1398 					 blk_rings->native.rsp_prod_pvt);
1399 		break;
1400 	case BLKIF_PROTOCOL_X86_32:
1401 		resp = RING_GET_RESPONSE(&blk_rings->x86_32,
1402 					 blk_rings->x86_32.rsp_prod_pvt);
1403 		break;
1404 	case BLKIF_PROTOCOL_X86_64:
1405 		resp = RING_GET_RESPONSE(&blk_rings->x86_64,
1406 					 blk_rings->x86_64.rsp_prod_pvt);
1407 		break;
1408 	default:
1409 		BUG();
1410 	}
1411 
1412 	resp->id        = id;
1413 	resp->operation = op;
1414 	resp->status    = st;
1415 
1416 	blk_rings->common.rsp_prod_pvt++;
1417 	RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blk_rings->common, notify);
1418 	spin_unlock_irqrestore(&ring->blk_ring_lock, flags);
1419 	if (notify)
1420 		notify_remote_via_irq(ring->irq);
1421 }
1422 
1423 static int __init xen_blkif_init(void)
1424 {
1425 	int rc = 0;
1426 
1427 	if (!xen_domain())
1428 		return -ENODEV;
1429 
1430 	if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
1431 		pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
1432 			xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
1433 		xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
1434 	}
1435 
1436 	if (xenblk_max_queues == 0)
1437 		xenblk_max_queues = num_online_cpus();
1438 
1439 	rc = xen_blkif_interface_init();
1440 	if (rc)
1441 		goto failed_init;
1442 
1443 	rc = xen_blkif_xenbus_init();
1444 	if (rc)
1445 		goto failed_init;
1446 
1447  failed_init:
1448 	return rc;
1449 }
1450 
1451 module_init(xen_blkif_init);
1452 
1453 static void __exit xen_blkif_fini(void)
1454 {
1455 	xen_blkif_xenbus_fini();
1456 	xen_blkif_interface_fini();
1457 }
1458 
1459 module_exit(xen_blkif_fini);
1460 
1461 MODULE_LICENSE("Dual BSD/GPL");
1462 MODULE_ALIAS("xen-backend:vbd");
1463