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