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