xref: /linux/drivers/block/xen-blkback/blkback.c (revision 148f9bb87745ed45f7a11b2cbd3bc0f017d5d257)
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  * These are rather arbitrary. They are fairly large because adjacent requests
54  * pulled from a communication ring are quite likely to end up being part of
55  * the same scatter/gather request at the disc.
56  *
57  * ** TRY INCREASING 'xen_blkif_reqs' IF WRITE SPEEDS SEEM TOO LOW **
58  *
59  * This will increase the chances of being able to write whole tracks.
60  * 64 should be enough to keep us competitive with Linux.
61  */
62 static int xen_blkif_reqs = 64;
63 module_param_named(reqs, xen_blkif_reqs, int, 0);
64 MODULE_PARM_DESC(reqs, "Number of blkback requests to allocate");
65 
66 /* Run-time switchable: /sys/module/blkback/parameters/ */
67 static unsigned int log_stats;
68 module_param(log_stats, int, 0644);
69 
70 /*
71  * Each outstanding request that we've passed to the lower device layers has a
72  * 'pending_req' allocated to it. Each buffer_head that completes decrements
73  * the pendcnt towards zero. When it hits zero, the specified domain has a
74  * response queued for it, with the saved 'id' passed back.
75  */
76 struct pending_req {
77 	struct xen_blkif	*blkif;
78 	u64			id;
79 	int			nr_pages;
80 	atomic_t		pendcnt;
81 	unsigned short		operation;
82 	int			status;
83 	struct list_head	free_list;
84 	DECLARE_BITMAP(unmap_seg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
85 };
86 
87 #define BLKBACK_INVALID_HANDLE (~0)
88 
89 struct xen_blkbk {
90 	struct pending_req	*pending_reqs;
91 	/* List of all 'pending_req' available */
92 	struct list_head	pending_free;
93 	/* And its spinlock. */
94 	spinlock_t		pending_free_lock;
95 	wait_queue_head_t	pending_free_wq;
96 	/* The list of all pages that are available. */
97 	struct page		**pending_pages;
98 	/* And the grant handles that are available. */
99 	grant_handle_t		*pending_grant_handles;
100 };
101 
102 static struct xen_blkbk *blkbk;
103 
104 /*
105  * Maximum number of grant pages that can be mapped in blkback.
106  * BLKIF_MAX_SEGMENTS_PER_REQUEST * RING_SIZE is the maximum number of
107  * pages that blkback will persistently map.
108  * Currently, this is:
109  * RING_SIZE = 32 (for all known ring types)
110  * BLKIF_MAX_SEGMENTS_PER_REQUEST = 11
111  * sizeof(struct persistent_gnt) = 48
112  * So the maximum memory used to store the grants is:
113  * 32 * 11 * 48 = 16896 bytes
114  */
115 static inline unsigned int max_mapped_grant_pages(enum blkif_protocol protocol)
116 {
117 	switch (protocol) {
118 	case BLKIF_PROTOCOL_NATIVE:
119 		return __CONST_RING_SIZE(blkif, PAGE_SIZE) *
120 			   BLKIF_MAX_SEGMENTS_PER_REQUEST;
121 	case BLKIF_PROTOCOL_X86_32:
122 		return __CONST_RING_SIZE(blkif_x86_32, PAGE_SIZE) *
123 			   BLKIF_MAX_SEGMENTS_PER_REQUEST;
124 	case BLKIF_PROTOCOL_X86_64:
125 		return __CONST_RING_SIZE(blkif_x86_64, PAGE_SIZE) *
126 			   BLKIF_MAX_SEGMENTS_PER_REQUEST;
127 	default:
128 		BUG();
129 	}
130 	return 0;
131 }
132 
133 
134 /*
135  * Little helpful macro to figure out the index and virtual address of the
136  * pending_pages[..]. For each 'pending_req' we have have up to
137  * BLKIF_MAX_SEGMENTS_PER_REQUEST (11) pages. The seg would be from 0 through
138  * 10 and would index in the pending_pages[..].
139  */
140 static inline int vaddr_pagenr(struct pending_req *req, int seg)
141 {
142 	return (req - blkbk->pending_reqs) *
143 		BLKIF_MAX_SEGMENTS_PER_REQUEST + seg;
144 }
145 
146 #define pending_page(req, seg) pending_pages[vaddr_pagenr(req, seg)]
147 
148 static inline unsigned long vaddr(struct pending_req *req, int seg)
149 {
150 	unsigned long pfn = page_to_pfn(blkbk->pending_page(req, seg));
151 	return (unsigned long)pfn_to_kaddr(pfn);
152 }
153 
154 #define pending_handle(_req, _seg) \
155 	(blkbk->pending_grant_handles[vaddr_pagenr(_req, _seg)])
156 
157 
158 static int do_block_io_op(struct xen_blkif *blkif);
159 static int dispatch_rw_block_io(struct xen_blkif *blkif,
160 				struct blkif_request *req,
161 				struct pending_req *pending_req);
162 static void make_response(struct xen_blkif *blkif, u64 id,
163 			  unsigned short op, int st);
164 
165 #define foreach_grant_safe(pos, n, rbtree, node) \
166 	for ((pos) = container_of(rb_first((rbtree)), typeof(*(pos)), node), \
167 	     (n) = (&(pos)->node != NULL) ? rb_next(&(pos)->node) : NULL; \
168 	     &(pos)->node != NULL; \
169 	     (pos) = container_of(n, typeof(*(pos)), node), \
170 	     (n) = (&(pos)->node != NULL) ? rb_next(&(pos)->node) : NULL)
171 
172 
173 static void add_persistent_gnt(struct rb_root *root,
174 			       struct persistent_gnt *persistent_gnt)
175 {
176 	struct rb_node **new = &(root->rb_node), *parent = NULL;
177 	struct persistent_gnt *this;
178 
179 	/* Figure out where to put new node */
180 	while (*new) {
181 		this = container_of(*new, struct persistent_gnt, node);
182 
183 		parent = *new;
184 		if (persistent_gnt->gnt < this->gnt)
185 			new = &((*new)->rb_left);
186 		else if (persistent_gnt->gnt > this->gnt)
187 			new = &((*new)->rb_right);
188 		else {
189 			pr_alert(DRV_PFX " trying to add a gref that's already in the tree\n");
190 			BUG();
191 		}
192 	}
193 
194 	/* Add new node and rebalance tree. */
195 	rb_link_node(&(persistent_gnt->node), parent, new);
196 	rb_insert_color(&(persistent_gnt->node), root);
197 }
198 
199 static struct persistent_gnt *get_persistent_gnt(struct rb_root *root,
200 						 grant_ref_t gref)
201 {
202 	struct persistent_gnt *data;
203 	struct rb_node *node = root->rb_node;
204 
205 	while (node) {
206 		data = container_of(node, struct persistent_gnt, node);
207 
208 		if (gref < data->gnt)
209 			node = node->rb_left;
210 		else if (gref > data->gnt)
211 			node = node->rb_right;
212 		else
213 			return data;
214 	}
215 	return NULL;
216 }
217 
218 static void free_persistent_gnts(struct rb_root *root, unsigned int num)
219 {
220 	struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST];
221 	struct page *pages[BLKIF_MAX_SEGMENTS_PER_REQUEST];
222 	struct persistent_gnt *persistent_gnt;
223 	struct rb_node *n;
224 	int ret = 0;
225 	int segs_to_unmap = 0;
226 
227 	foreach_grant_safe(persistent_gnt, n, root, node) {
228 		BUG_ON(persistent_gnt->handle ==
229 			BLKBACK_INVALID_HANDLE);
230 		gnttab_set_unmap_op(&unmap[segs_to_unmap],
231 			(unsigned long) pfn_to_kaddr(page_to_pfn(
232 				persistent_gnt->page)),
233 			GNTMAP_host_map,
234 			persistent_gnt->handle);
235 
236 		pages[segs_to_unmap] = persistent_gnt->page;
237 
238 		if (++segs_to_unmap == BLKIF_MAX_SEGMENTS_PER_REQUEST ||
239 			!rb_next(&persistent_gnt->node)) {
240 			ret = gnttab_unmap_refs(unmap, NULL, pages,
241 				segs_to_unmap);
242 			BUG_ON(ret);
243 			free_xenballooned_pages(segs_to_unmap, pages);
244 			segs_to_unmap = 0;
245 		}
246 
247 		rb_erase(&persistent_gnt->node, root);
248 		kfree(persistent_gnt);
249 		num--;
250 	}
251 	BUG_ON(num != 0);
252 }
253 
254 /*
255  * Retrieve from the 'pending_reqs' a free pending_req structure to be used.
256  */
257 static struct pending_req *alloc_req(void)
258 {
259 	struct pending_req *req = NULL;
260 	unsigned long flags;
261 
262 	spin_lock_irqsave(&blkbk->pending_free_lock, flags);
263 	if (!list_empty(&blkbk->pending_free)) {
264 		req = list_entry(blkbk->pending_free.next, struct pending_req,
265 				 free_list);
266 		list_del(&req->free_list);
267 	}
268 	spin_unlock_irqrestore(&blkbk->pending_free_lock, flags);
269 	return req;
270 }
271 
272 /*
273  * Return the 'pending_req' structure back to the freepool. We also
274  * wake up the thread if it was waiting for a free page.
275  */
276 static void free_req(struct pending_req *req)
277 {
278 	unsigned long flags;
279 	int was_empty;
280 
281 	spin_lock_irqsave(&blkbk->pending_free_lock, flags);
282 	was_empty = list_empty(&blkbk->pending_free);
283 	list_add(&req->free_list, &blkbk->pending_free);
284 	spin_unlock_irqrestore(&blkbk->pending_free_lock, flags);
285 	if (was_empty)
286 		wake_up(&blkbk->pending_free_wq);
287 }
288 
289 /*
290  * Routines for managing virtual block devices (vbds).
291  */
292 static int xen_vbd_translate(struct phys_req *req, struct xen_blkif *blkif,
293 			     int operation)
294 {
295 	struct xen_vbd *vbd = &blkif->vbd;
296 	int rc = -EACCES;
297 
298 	if ((operation != READ) && vbd->readonly)
299 		goto out;
300 
301 	if (likely(req->nr_sects)) {
302 		blkif_sector_t end = req->sector_number + req->nr_sects;
303 
304 		if (unlikely(end < req->sector_number))
305 			goto out;
306 		if (unlikely(end > vbd_sz(vbd)))
307 			goto out;
308 	}
309 
310 	req->dev  = vbd->pdevice;
311 	req->bdev = vbd->bdev;
312 	rc = 0;
313 
314  out:
315 	return rc;
316 }
317 
318 static void xen_vbd_resize(struct xen_blkif *blkif)
319 {
320 	struct xen_vbd *vbd = &blkif->vbd;
321 	struct xenbus_transaction xbt;
322 	int err;
323 	struct xenbus_device *dev = xen_blkbk_xenbus(blkif->be);
324 	unsigned long long new_size = vbd_sz(vbd);
325 
326 	pr_info(DRV_PFX "VBD Resize: Domid: %d, Device: (%d, %d)\n",
327 		blkif->domid, MAJOR(vbd->pdevice), MINOR(vbd->pdevice));
328 	pr_info(DRV_PFX "VBD Resize: new size %llu\n", new_size);
329 	vbd->size = new_size;
330 again:
331 	err = xenbus_transaction_start(&xbt);
332 	if (err) {
333 		pr_warn(DRV_PFX "Error starting transaction");
334 		return;
335 	}
336 	err = xenbus_printf(xbt, dev->nodename, "sectors", "%llu",
337 			    (unsigned long long)vbd_sz(vbd));
338 	if (err) {
339 		pr_warn(DRV_PFX "Error writing new size");
340 		goto abort;
341 	}
342 	/*
343 	 * Write the current state; we will use this to synchronize
344 	 * the front-end. If the current state is "connected" the
345 	 * front-end will get the new size information online.
346 	 */
347 	err = xenbus_printf(xbt, dev->nodename, "state", "%d", dev->state);
348 	if (err) {
349 		pr_warn(DRV_PFX "Error writing the state");
350 		goto abort;
351 	}
352 
353 	err = xenbus_transaction_end(xbt, 0);
354 	if (err == -EAGAIN)
355 		goto again;
356 	if (err)
357 		pr_warn(DRV_PFX "Error ending transaction");
358 	return;
359 abort:
360 	xenbus_transaction_end(xbt, 1);
361 }
362 
363 /*
364  * Notification from the guest OS.
365  */
366 static void blkif_notify_work(struct xen_blkif *blkif)
367 {
368 	blkif->waiting_reqs = 1;
369 	wake_up(&blkif->wq);
370 }
371 
372 irqreturn_t xen_blkif_be_int(int irq, void *dev_id)
373 {
374 	blkif_notify_work(dev_id);
375 	return IRQ_HANDLED;
376 }
377 
378 /*
379  * SCHEDULER FUNCTIONS
380  */
381 
382 static void print_stats(struct xen_blkif *blkif)
383 {
384 	pr_info("xen-blkback (%s): oo %3llu  |  rd %4llu  |  wr %4llu  |  f %4llu"
385 		 "  |  ds %4llu\n",
386 		 current->comm, blkif->st_oo_req,
387 		 blkif->st_rd_req, blkif->st_wr_req,
388 		 blkif->st_f_req, blkif->st_ds_req);
389 	blkif->st_print = jiffies + msecs_to_jiffies(10 * 1000);
390 	blkif->st_rd_req = 0;
391 	blkif->st_wr_req = 0;
392 	blkif->st_oo_req = 0;
393 	blkif->st_ds_req = 0;
394 }
395 
396 int xen_blkif_schedule(void *arg)
397 {
398 	struct xen_blkif *blkif = arg;
399 	struct xen_vbd *vbd = &blkif->vbd;
400 
401 	xen_blkif_get(blkif);
402 
403 	while (!kthread_should_stop()) {
404 		if (try_to_freeze())
405 			continue;
406 		if (unlikely(vbd->size != vbd_sz(vbd)))
407 			xen_vbd_resize(blkif);
408 
409 		wait_event_interruptible(
410 			blkif->wq,
411 			blkif->waiting_reqs || kthread_should_stop());
412 		wait_event_interruptible(
413 			blkbk->pending_free_wq,
414 			!list_empty(&blkbk->pending_free) ||
415 			kthread_should_stop());
416 
417 		blkif->waiting_reqs = 0;
418 		smp_mb(); /* clear flag *before* checking for work */
419 
420 		if (do_block_io_op(blkif))
421 			blkif->waiting_reqs = 1;
422 
423 		if (log_stats && time_after(jiffies, blkif->st_print))
424 			print_stats(blkif);
425 	}
426 
427 	/* Free all persistent grant pages */
428 	if (!RB_EMPTY_ROOT(&blkif->persistent_gnts))
429 		free_persistent_gnts(&blkif->persistent_gnts,
430 			blkif->persistent_gnt_c);
431 
432 	BUG_ON(!RB_EMPTY_ROOT(&blkif->persistent_gnts));
433 	blkif->persistent_gnt_c = 0;
434 
435 	if (log_stats)
436 		print_stats(blkif);
437 
438 	blkif->xenblkd = NULL;
439 	xen_blkif_put(blkif);
440 
441 	return 0;
442 }
443 
444 struct seg_buf {
445 	unsigned int offset;
446 	unsigned int nsec;
447 };
448 /*
449  * Unmap the grant references, and also remove the M2P over-rides
450  * used in the 'pending_req'.
451  */
452 static void xen_blkbk_unmap(struct pending_req *req)
453 {
454 	struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST];
455 	struct page *pages[BLKIF_MAX_SEGMENTS_PER_REQUEST];
456 	unsigned int i, invcount = 0;
457 	grant_handle_t handle;
458 	int ret;
459 
460 	for (i = 0; i < req->nr_pages; i++) {
461 		if (!test_bit(i, req->unmap_seg))
462 			continue;
463 		handle = pending_handle(req, i);
464 		if (handle == BLKBACK_INVALID_HANDLE)
465 			continue;
466 		gnttab_set_unmap_op(&unmap[invcount], vaddr(req, i),
467 				    GNTMAP_host_map, handle);
468 		pending_handle(req, i) = BLKBACK_INVALID_HANDLE;
469 		pages[invcount] = virt_to_page(vaddr(req, i));
470 		invcount++;
471 	}
472 
473 	ret = gnttab_unmap_refs(unmap, NULL, pages, invcount);
474 	BUG_ON(ret);
475 }
476 
477 static int xen_blkbk_map(struct blkif_request *req,
478 			 struct pending_req *pending_req,
479 			 struct seg_buf seg[],
480 			 struct page *pages[])
481 {
482 	struct gnttab_map_grant_ref map[BLKIF_MAX_SEGMENTS_PER_REQUEST];
483 	struct persistent_gnt *persistent_gnts[BLKIF_MAX_SEGMENTS_PER_REQUEST];
484 	struct page *pages_to_gnt[BLKIF_MAX_SEGMENTS_PER_REQUEST];
485 	struct persistent_gnt *persistent_gnt = NULL;
486 	struct xen_blkif *blkif = pending_req->blkif;
487 	phys_addr_t addr = 0;
488 	int i, j;
489 	bool new_map;
490 	int nseg = req->u.rw.nr_segments;
491 	int segs_to_map = 0;
492 	int ret = 0;
493 	int use_persistent_gnts;
494 
495 	use_persistent_gnts = (blkif->vbd.feature_gnt_persistent);
496 
497 	BUG_ON(blkif->persistent_gnt_c >
498 		   max_mapped_grant_pages(pending_req->blkif->blk_protocol));
499 
500 	/*
501 	 * Fill out preq.nr_sects with proper amount of sectors, and setup
502 	 * assign map[..] with the PFN of the page in our domain with the
503 	 * corresponding grant reference for each page.
504 	 */
505 	for (i = 0; i < nseg; i++) {
506 		uint32_t flags;
507 
508 		if (use_persistent_gnts)
509 			persistent_gnt = get_persistent_gnt(
510 				&blkif->persistent_gnts,
511 				req->u.rw.seg[i].gref);
512 
513 		if (persistent_gnt) {
514 			/*
515 			 * We are using persistent grants and
516 			 * the grant is already mapped
517 			 */
518 			new_map = false;
519 		} else if (use_persistent_gnts &&
520 			   blkif->persistent_gnt_c <
521 			   max_mapped_grant_pages(blkif->blk_protocol)) {
522 			/*
523 			 * We are using persistent grants, the grant is
524 			 * not mapped but we have room for it
525 			 */
526 			new_map = true;
527 			persistent_gnt = kmalloc(
528 				sizeof(struct persistent_gnt),
529 				GFP_KERNEL);
530 			if (!persistent_gnt)
531 				return -ENOMEM;
532 			if (alloc_xenballooned_pages(1, &persistent_gnt->page,
533 			    false)) {
534 				kfree(persistent_gnt);
535 				return -ENOMEM;
536 			}
537 			persistent_gnt->gnt = req->u.rw.seg[i].gref;
538 			persistent_gnt->handle = BLKBACK_INVALID_HANDLE;
539 
540 			pages_to_gnt[segs_to_map] =
541 				persistent_gnt->page;
542 			addr = (unsigned long) pfn_to_kaddr(
543 				page_to_pfn(persistent_gnt->page));
544 
545 			add_persistent_gnt(&blkif->persistent_gnts,
546 				persistent_gnt);
547 			blkif->persistent_gnt_c++;
548 			pr_debug(DRV_PFX " grant %u added to the tree of persistent grants, using %u/%u\n",
549 				 persistent_gnt->gnt, blkif->persistent_gnt_c,
550 				 max_mapped_grant_pages(blkif->blk_protocol));
551 		} else {
552 			/*
553 			 * We are either using persistent grants and
554 			 * hit the maximum limit of grants mapped,
555 			 * or we are not using persistent grants.
556 			 */
557 			if (use_persistent_gnts &&
558 				!blkif->vbd.overflow_max_grants) {
559 				blkif->vbd.overflow_max_grants = 1;
560 				pr_alert(DRV_PFX " domain %u, device %#x is using maximum number of persistent grants\n",
561 					 blkif->domid, blkif->vbd.handle);
562 			}
563 			new_map = true;
564 			pages[i] = blkbk->pending_page(pending_req, i);
565 			addr = vaddr(pending_req, i);
566 			pages_to_gnt[segs_to_map] =
567 				blkbk->pending_page(pending_req, i);
568 		}
569 
570 		if (persistent_gnt) {
571 			pages[i] = persistent_gnt->page;
572 			persistent_gnts[i] = persistent_gnt;
573 		} else {
574 			persistent_gnts[i] = NULL;
575 		}
576 
577 		if (new_map) {
578 			flags = GNTMAP_host_map;
579 			if (!persistent_gnt &&
580 			    (pending_req->operation != BLKIF_OP_READ))
581 				flags |= GNTMAP_readonly;
582 			gnttab_set_map_op(&map[segs_to_map++], addr,
583 					  flags, req->u.rw.seg[i].gref,
584 					  blkif->domid);
585 		}
586 	}
587 
588 	if (segs_to_map) {
589 		ret = gnttab_map_refs(map, NULL, pages_to_gnt, segs_to_map);
590 		BUG_ON(ret);
591 	}
592 
593 	/*
594 	 * Now swizzle the MFN in our domain with the MFN from the other domain
595 	 * so that when we access vaddr(pending_req,i) it has the contents of
596 	 * the page from the other domain.
597 	 */
598 	bitmap_zero(pending_req->unmap_seg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
599 	for (i = 0, j = 0; i < nseg; i++) {
600 		if (!persistent_gnts[i] ||
601 		    persistent_gnts[i]->handle == BLKBACK_INVALID_HANDLE) {
602 			/* This is a newly mapped grant */
603 			BUG_ON(j >= segs_to_map);
604 			if (unlikely(map[j].status != 0)) {
605 				pr_debug(DRV_PFX "invalid buffer -- could not remap it\n");
606 				map[j].handle = BLKBACK_INVALID_HANDLE;
607 				ret |= 1;
608 				if (persistent_gnts[i]) {
609 					rb_erase(&persistent_gnts[i]->node,
610 						 &blkif->persistent_gnts);
611 					blkif->persistent_gnt_c--;
612 					kfree(persistent_gnts[i]);
613 					persistent_gnts[i] = NULL;
614 				}
615 			}
616 		}
617 		if (persistent_gnts[i]) {
618 			if (persistent_gnts[i]->handle ==
619 			    BLKBACK_INVALID_HANDLE) {
620 				/*
621 				 * If this is a new persistent grant
622 				 * save the handler
623 				 */
624 				persistent_gnts[i]->handle = map[j++].handle;
625 			}
626 			pending_handle(pending_req, i) =
627 				persistent_gnts[i]->handle;
628 
629 			if (ret)
630 				continue;
631 		} else {
632 			pending_handle(pending_req, i) = map[j++].handle;
633 			bitmap_set(pending_req->unmap_seg, i, 1);
634 
635 			if (ret)
636 				continue;
637 		}
638 		seg[i].offset = (req->u.rw.seg[i].first_sect << 9);
639 	}
640 	return ret;
641 }
642 
643 static int dispatch_discard_io(struct xen_blkif *blkif,
644 				struct blkif_request *req)
645 {
646 	int err = 0;
647 	int status = BLKIF_RSP_OKAY;
648 	struct block_device *bdev = blkif->vbd.bdev;
649 	unsigned long secure;
650 
651 	blkif->st_ds_req++;
652 
653 	xen_blkif_get(blkif);
654 	secure = (blkif->vbd.discard_secure &&
655 		 (req->u.discard.flag & BLKIF_DISCARD_SECURE)) ?
656 		 BLKDEV_DISCARD_SECURE : 0;
657 
658 	err = blkdev_issue_discard(bdev, req->u.discard.sector_number,
659 				   req->u.discard.nr_sectors,
660 				   GFP_KERNEL, secure);
661 
662 	if (err == -EOPNOTSUPP) {
663 		pr_debug(DRV_PFX "discard op failed, not supported\n");
664 		status = BLKIF_RSP_EOPNOTSUPP;
665 	} else if (err)
666 		status = BLKIF_RSP_ERROR;
667 
668 	make_response(blkif, req->u.discard.id, req->operation, status);
669 	xen_blkif_put(blkif);
670 	return err;
671 }
672 
673 static int dispatch_other_io(struct xen_blkif *blkif,
674 			     struct blkif_request *req,
675 			     struct pending_req *pending_req)
676 {
677 	free_req(pending_req);
678 	make_response(blkif, req->u.other.id, req->operation,
679 		      BLKIF_RSP_EOPNOTSUPP);
680 	return -EIO;
681 }
682 
683 static void xen_blk_drain_io(struct xen_blkif *blkif)
684 {
685 	atomic_set(&blkif->drain, 1);
686 	do {
687 		/* The initial value is one, and one refcnt taken at the
688 		 * start of the xen_blkif_schedule thread. */
689 		if (atomic_read(&blkif->refcnt) <= 2)
690 			break;
691 		wait_for_completion_interruptible_timeout(
692 				&blkif->drain_complete, HZ);
693 
694 		if (!atomic_read(&blkif->drain))
695 			break;
696 	} while (!kthread_should_stop());
697 	atomic_set(&blkif->drain, 0);
698 }
699 
700 /*
701  * Completion callback on the bio's. Called as bh->b_end_io()
702  */
703 
704 static void __end_block_io_op(struct pending_req *pending_req, int error)
705 {
706 	/* An error fails the entire request. */
707 	if ((pending_req->operation == BLKIF_OP_FLUSH_DISKCACHE) &&
708 	    (error == -EOPNOTSUPP)) {
709 		pr_debug(DRV_PFX "flush diskcache op failed, not supported\n");
710 		xen_blkbk_flush_diskcache(XBT_NIL, pending_req->blkif->be, 0);
711 		pending_req->status = BLKIF_RSP_EOPNOTSUPP;
712 	} else if ((pending_req->operation == BLKIF_OP_WRITE_BARRIER) &&
713 		    (error == -EOPNOTSUPP)) {
714 		pr_debug(DRV_PFX "write barrier op failed, not supported\n");
715 		xen_blkbk_barrier(XBT_NIL, pending_req->blkif->be, 0);
716 		pending_req->status = BLKIF_RSP_EOPNOTSUPP;
717 	} else if (error) {
718 		pr_debug(DRV_PFX "Buffer not up-to-date at end of operation,"
719 			 " error=%d\n", error);
720 		pending_req->status = BLKIF_RSP_ERROR;
721 	}
722 
723 	/*
724 	 * If all of the bio's have completed it is time to unmap
725 	 * the grant references associated with 'request' and provide
726 	 * the proper response on the ring.
727 	 */
728 	if (atomic_dec_and_test(&pending_req->pendcnt)) {
729 		xen_blkbk_unmap(pending_req);
730 		make_response(pending_req->blkif, pending_req->id,
731 			      pending_req->operation, pending_req->status);
732 		xen_blkif_put(pending_req->blkif);
733 		if (atomic_read(&pending_req->blkif->refcnt) <= 2) {
734 			if (atomic_read(&pending_req->blkif->drain))
735 				complete(&pending_req->blkif->drain_complete);
736 		}
737 		free_req(pending_req);
738 	}
739 }
740 
741 /*
742  * bio callback.
743  */
744 static void end_block_io_op(struct bio *bio, int error)
745 {
746 	__end_block_io_op(bio->bi_private, error);
747 	bio_put(bio);
748 }
749 
750 
751 
752 /*
753  * Function to copy the from the ring buffer the 'struct blkif_request'
754  * (which has the sectors we want, number of them, grant references, etc),
755  * and transmute  it to the block API to hand it over to the proper block disk.
756  */
757 static int
758 __do_block_io_op(struct xen_blkif *blkif)
759 {
760 	union blkif_back_rings *blk_rings = &blkif->blk_rings;
761 	struct blkif_request req;
762 	struct pending_req *pending_req;
763 	RING_IDX rc, rp;
764 	int more_to_do = 0;
765 
766 	rc = blk_rings->common.req_cons;
767 	rp = blk_rings->common.sring->req_prod;
768 	rmb(); /* Ensure we see queued requests up to 'rp'. */
769 
770 	while (rc != rp) {
771 
772 		if (RING_REQUEST_CONS_OVERFLOW(&blk_rings->common, rc))
773 			break;
774 
775 		if (kthread_should_stop()) {
776 			more_to_do = 1;
777 			break;
778 		}
779 
780 		pending_req = alloc_req();
781 		if (NULL == pending_req) {
782 			blkif->st_oo_req++;
783 			more_to_do = 1;
784 			break;
785 		}
786 
787 		switch (blkif->blk_protocol) {
788 		case BLKIF_PROTOCOL_NATIVE:
789 			memcpy(&req, RING_GET_REQUEST(&blk_rings->native, rc), sizeof(req));
790 			break;
791 		case BLKIF_PROTOCOL_X86_32:
792 			blkif_get_x86_32_req(&req, RING_GET_REQUEST(&blk_rings->x86_32, rc));
793 			break;
794 		case BLKIF_PROTOCOL_X86_64:
795 			blkif_get_x86_64_req(&req, RING_GET_REQUEST(&blk_rings->x86_64, rc));
796 			break;
797 		default:
798 			BUG();
799 		}
800 		blk_rings->common.req_cons = ++rc; /* before make_response() */
801 
802 		/* Apply all sanity checks to /private copy/ of request. */
803 		barrier();
804 
805 		switch (req.operation) {
806 		case BLKIF_OP_READ:
807 		case BLKIF_OP_WRITE:
808 		case BLKIF_OP_WRITE_BARRIER:
809 		case BLKIF_OP_FLUSH_DISKCACHE:
810 			if (dispatch_rw_block_io(blkif, &req, pending_req))
811 				goto done;
812 			break;
813 		case BLKIF_OP_DISCARD:
814 			free_req(pending_req);
815 			if (dispatch_discard_io(blkif, &req))
816 				goto done;
817 			break;
818 		default:
819 			if (dispatch_other_io(blkif, &req, pending_req))
820 				goto done;
821 			break;
822 		}
823 
824 		/* Yield point for this unbounded loop. */
825 		cond_resched();
826 	}
827 done:
828 	return more_to_do;
829 }
830 
831 static int
832 do_block_io_op(struct xen_blkif *blkif)
833 {
834 	union blkif_back_rings *blk_rings = &blkif->blk_rings;
835 	int more_to_do;
836 
837 	do {
838 		more_to_do = __do_block_io_op(blkif);
839 		if (more_to_do)
840 			break;
841 
842 		RING_FINAL_CHECK_FOR_REQUESTS(&blk_rings->common, more_to_do);
843 	} while (more_to_do);
844 
845 	return more_to_do;
846 }
847 /*
848  * Transmutation of the 'struct blkif_request' to a proper 'struct bio'
849  * and call the 'submit_bio' to pass it to the underlying storage.
850  */
851 static int dispatch_rw_block_io(struct xen_blkif *blkif,
852 				struct blkif_request *req,
853 				struct pending_req *pending_req)
854 {
855 	struct phys_req preq;
856 	struct seg_buf seg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
857 	unsigned int nseg;
858 	struct bio *bio = NULL;
859 	struct bio *biolist[BLKIF_MAX_SEGMENTS_PER_REQUEST];
860 	int i, nbio = 0;
861 	int operation;
862 	struct blk_plug plug;
863 	bool drain = false;
864 	struct page *pages[BLKIF_MAX_SEGMENTS_PER_REQUEST];
865 
866 	switch (req->operation) {
867 	case BLKIF_OP_READ:
868 		blkif->st_rd_req++;
869 		operation = READ;
870 		break;
871 	case BLKIF_OP_WRITE:
872 		blkif->st_wr_req++;
873 		operation = WRITE_ODIRECT;
874 		break;
875 	case BLKIF_OP_WRITE_BARRIER:
876 		drain = true;
877 	case BLKIF_OP_FLUSH_DISKCACHE:
878 		blkif->st_f_req++;
879 		operation = WRITE_FLUSH;
880 		break;
881 	default:
882 		operation = 0; /* make gcc happy */
883 		goto fail_response;
884 		break;
885 	}
886 
887 	/* Check that the number of segments is sane. */
888 	nseg = req->u.rw.nr_segments;
889 
890 	if (unlikely(nseg == 0 && operation != WRITE_FLUSH) ||
891 	    unlikely(nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
892 		pr_debug(DRV_PFX "Bad number of segments in request (%d)\n",
893 			 nseg);
894 		/* Haven't submitted any bio's yet. */
895 		goto fail_response;
896 	}
897 
898 	preq.sector_number = req->u.rw.sector_number;
899 	preq.nr_sects      = 0;
900 
901 	pending_req->blkif     = blkif;
902 	pending_req->id        = req->u.rw.id;
903 	pending_req->operation = req->operation;
904 	pending_req->status    = BLKIF_RSP_OKAY;
905 	pending_req->nr_pages  = nseg;
906 
907 	for (i = 0; i < nseg; i++) {
908 		seg[i].nsec = req->u.rw.seg[i].last_sect -
909 			req->u.rw.seg[i].first_sect + 1;
910 		if ((req->u.rw.seg[i].last_sect >= (PAGE_SIZE >> 9)) ||
911 		    (req->u.rw.seg[i].last_sect < req->u.rw.seg[i].first_sect))
912 			goto fail_response;
913 		preq.nr_sects += seg[i].nsec;
914 
915 	}
916 
917 	if (xen_vbd_translate(&preq, blkif, operation) != 0) {
918 		pr_debug(DRV_PFX "access denied: %s of [%llu,%llu] on dev=%04x\n",
919 			 operation == READ ? "read" : "write",
920 			 preq.sector_number,
921 			 preq.sector_number + preq.nr_sects,
922 			 blkif->vbd.pdevice);
923 		goto fail_response;
924 	}
925 
926 	/*
927 	 * This check _MUST_ be done after xen_vbd_translate as the preq.bdev
928 	 * is set there.
929 	 */
930 	for (i = 0; i < nseg; i++) {
931 		if (((int)preq.sector_number|(int)seg[i].nsec) &
932 		    ((bdev_logical_block_size(preq.bdev) >> 9) - 1)) {
933 			pr_debug(DRV_PFX "Misaligned I/O request from domain %d",
934 				 blkif->domid);
935 			goto fail_response;
936 		}
937 	}
938 
939 	/* Wait on all outstanding I/O's and once that has been completed
940 	 * issue the WRITE_FLUSH.
941 	 */
942 	if (drain)
943 		xen_blk_drain_io(pending_req->blkif);
944 
945 	/*
946 	 * If we have failed at this point, we need to undo the M2P override,
947 	 * set gnttab_set_unmap_op on all of the grant references and perform
948 	 * the hypercall to unmap the grants - that is all done in
949 	 * xen_blkbk_unmap.
950 	 */
951 	if (xen_blkbk_map(req, pending_req, seg, pages))
952 		goto fail_flush;
953 
954 	/*
955 	 * This corresponding xen_blkif_put is done in __end_block_io_op, or
956 	 * below (in "!bio") if we are handling a BLKIF_OP_DISCARD.
957 	 */
958 	xen_blkif_get(blkif);
959 
960 	for (i = 0; i < nseg; i++) {
961 		while ((bio == NULL) ||
962 		       (bio_add_page(bio,
963 				     pages[i],
964 				     seg[i].nsec << 9,
965 				     seg[i].offset) == 0)) {
966 
967 			bio = bio_alloc(GFP_KERNEL, nseg-i);
968 			if (unlikely(bio == NULL))
969 				goto fail_put_bio;
970 
971 			biolist[nbio++] = bio;
972 			bio->bi_bdev    = preq.bdev;
973 			bio->bi_private = pending_req;
974 			bio->bi_end_io  = end_block_io_op;
975 			bio->bi_sector  = preq.sector_number;
976 		}
977 
978 		preq.sector_number += seg[i].nsec;
979 	}
980 
981 	/* This will be hit if the operation was a flush or discard. */
982 	if (!bio) {
983 		BUG_ON(operation != WRITE_FLUSH);
984 
985 		bio = bio_alloc(GFP_KERNEL, 0);
986 		if (unlikely(bio == NULL))
987 			goto fail_put_bio;
988 
989 		biolist[nbio++] = bio;
990 		bio->bi_bdev    = preq.bdev;
991 		bio->bi_private = pending_req;
992 		bio->bi_end_io  = end_block_io_op;
993 	}
994 
995 	atomic_set(&pending_req->pendcnt, nbio);
996 	blk_start_plug(&plug);
997 
998 	for (i = 0; i < nbio; i++)
999 		submit_bio(operation, biolist[i]);
1000 
1001 	/* Let the I/Os go.. */
1002 	blk_finish_plug(&plug);
1003 
1004 	if (operation == READ)
1005 		blkif->st_rd_sect += preq.nr_sects;
1006 	else if (operation & WRITE)
1007 		blkif->st_wr_sect += preq.nr_sects;
1008 
1009 	return 0;
1010 
1011  fail_flush:
1012 	xen_blkbk_unmap(pending_req);
1013  fail_response:
1014 	/* Haven't submitted any bio's yet. */
1015 	make_response(blkif, req->u.rw.id, req->operation, BLKIF_RSP_ERROR);
1016 	free_req(pending_req);
1017 	msleep(1); /* back off a bit */
1018 	return -EIO;
1019 
1020  fail_put_bio:
1021 	for (i = 0; i < nbio; i++)
1022 		bio_put(biolist[i]);
1023 	atomic_set(&pending_req->pendcnt, 1);
1024 	__end_block_io_op(pending_req, -EINVAL);
1025 	msleep(1); /* back off a bit */
1026 	return -EIO;
1027 }
1028 
1029 
1030 
1031 /*
1032  * Put a response on the ring on how the operation fared.
1033  */
1034 static void make_response(struct xen_blkif *blkif, u64 id,
1035 			  unsigned short op, int st)
1036 {
1037 	struct blkif_response  resp;
1038 	unsigned long     flags;
1039 	union blkif_back_rings *blk_rings = &blkif->blk_rings;
1040 	int notify;
1041 
1042 	resp.id        = id;
1043 	resp.operation = op;
1044 	resp.status    = st;
1045 
1046 	spin_lock_irqsave(&blkif->blk_ring_lock, flags);
1047 	/* Place on the response ring for the relevant domain. */
1048 	switch (blkif->blk_protocol) {
1049 	case BLKIF_PROTOCOL_NATIVE:
1050 		memcpy(RING_GET_RESPONSE(&blk_rings->native, blk_rings->native.rsp_prod_pvt),
1051 		       &resp, sizeof(resp));
1052 		break;
1053 	case BLKIF_PROTOCOL_X86_32:
1054 		memcpy(RING_GET_RESPONSE(&blk_rings->x86_32, blk_rings->x86_32.rsp_prod_pvt),
1055 		       &resp, sizeof(resp));
1056 		break;
1057 	case BLKIF_PROTOCOL_X86_64:
1058 		memcpy(RING_GET_RESPONSE(&blk_rings->x86_64, blk_rings->x86_64.rsp_prod_pvt),
1059 		       &resp, sizeof(resp));
1060 		break;
1061 	default:
1062 		BUG();
1063 	}
1064 	blk_rings->common.rsp_prod_pvt++;
1065 	RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blk_rings->common, notify);
1066 	spin_unlock_irqrestore(&blkif->blk_ring_lock, flags);
1067 	if (notify)
1068 		notify_remote_via_irq(blkif->irq);
1069 }
1070 
1071 static int __init xen_blkif_init(void)
1072 {
1073 	int i, mmap_pages;
1074 	int rc = 0;
1075 
1076 	if (!xen_domain())
1077 		return -ENODEV;
1078 
1079 	blkbk = kzalloc(sizeof(struct xen_blkbk), GFP_KERNEL);
1080 	if (!blkbk) {
1081 		pr_alert(DRV_PFX "%s: out of memory!\n", __func__);
1082 		return -ENOMEM;
1083 	}
1084 
1085 	mmap_pages = xen_blkif_reqs * BLKIF_MAX_SEGMENTS_PER_REQUEST;
1086 
1087 	blkbk->pending_reqs          = kzalloc(sizeof(blkbk->pending_reqs[0]) *
1088 					xen_blkif_reqs, GFP_KERNEL);
1089 	blkbk->pending_grant_handles = kmalloc(sizeof(blkbk->pending_grant_handles[0]) *
1090 					mmap_pages, GFP_KERNEL);
1091 	blkbk->pending_pages         = kzalloc(sizeof(blkbk->pending_pages[0]) *
1092 					mmap_pages, GFP_KERNEL);
1093 
1094 	if (!blkbk->pending_reqs || !blkbk->pending_grant_handles ||
1095 	    !blkbk->pending_pages) {
1096 		rc = -ENOMEM;
1097 		goto out_of_memory;
1098 	}
1099 
1100 	for (i = 0; i < mmap_pages; i++) {
1101 		blkbk->pending_grant_handles[i] = BLKBACK_INVALID_HANDLE;
1102 		blkbk->pending_pages[i] = alloc_page(GFP_KERNEL);
1103 		if (blkbk->pending_pages[i] == NULL) {
1104 			rc = -ENOMEM;
1105 			goto out_of_memory;
1106 		}
1107 	}
1108 	rc = xen_blkif_interface_init();
1109 	if (rc)
1110 		goto failed_init;
1111 
1112 	INIT_LIST_HEAD(&blkbk->pending_free);
1113 	spin_lock_init(&blkbk->pending_free_lock);
1114 	init_waitqueue_head(&blkbk->pending_free_wq);
1115 
1116 	for (i = 0; i < xen_blkif_reqs; i++)
1117 		list_add_tail(&blkbk->pending_reqs[i].free_list,
1118 			      &blkbk->pending_free);
1119 
1120 	rc = xen_blkif_xenbus_init();
1121 	if (rc)
1122 		goto failed_init;
1123 
1124 	return 0;
1125 
1126  out_of_memory:
1127 	pr_alert(DRV_PFX "%s: out of memory\n", __func__);
1128  failed_init:
1129 	kfree(blkbk->pending_reqs);
1130 	kfree(blkbk->pending_grant_handles);
1131 	if (blkbk->pending_pages) {
1132 		for (i = 0; i < mmap_pages; i++) {
1133 			if (blkbk->pending_pages[i])
1134 				__free_page(blkbk->pending_pages[i]);
1135 		}
1136 		kfree(blkbk->pending_pages);
1137 	}
1138 	kfree(blkbk);
1139 	blkbk = NULL;
1140 	return rc;
1141 }
1142 
1143 module_init(xen_blkif_init);
1144 
1145 MODULE_LICENSE("Dual BSD/GPL");
1146 MODULE_ALIAS("xen-backend:vbd");
1147