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 
43 #include <xen/events.h>
44 #include <xen/page.h>
45 #include <xen/xen.h>
46 #include <asm/xen/hypervisor.h>
47 #include <asm/xen/hypercall.h>
48 #include "common.h"
49 
50 /*
51  * These are rather arbitrary. They are fairly large because adjacent requests
52  * pulled from a communication ring are quite likely to end up being part of
53  * the same scatter/gather request at the disc.
54  *
55  * ** TRY INCREASING 'xen_blkif_reqs' IF WRITE SPEEDS SEEM TOO LOW **
56  *
57  * This will increase the chances of being able to write whole tracks.
58  * 64 should be enough to keep us competitive with Linux.
59  */
60 static int xen_blkif_reqs = 64;
61 module_param_named(reqs, xen_blkif_reqs, int, 0);
62 MODULE_PARM_DESC(reqs, "Number of blkback requests to allocate");
63 
64 /* Run-time switchable: /sys/module/blkback/parameters/ */
65 static unsigned int log_stats;
66 module_param(log_stats, int, 0644);
67 
68 /*
69  * Each outstanding request that we've passed to the lower device layers has a
70  * 'pending_req' allocated to it. Each buffer_head that completes decrements
71  * the pendcnt towards zero. When it hits zero, the specified domain has a
72  * response queued for it, with the saved 'id' passed back.
73  */
74 struct pending_req {
75 	struct xen_blkif	*blkif;
76 	u64			id;
77 	int			nr_pages;
78 	atomic_t		pendcnt;
79 	unsigned short		operation;
80 	int			status;
81 	struct list_head	free_list;
82 };
83 
84 #define BLKBACK_INVALID_HANDLE (~0)
85 
86 struct xen_blkbk {
87 	struct pending_req	*pending_reqs;
88 	/* List of all 'pending_req' available */
89 	struct list_head	pending_free;
90 	/* And its spinlock. */
91 	spinlock_t		pending_free_lock;
92 	wait_queue_head_t	pending_free_wq;
93 	/* The list of all pages that are available. */
94 	struct page		**pending_pages;
95 	/* And the grant handles that are available. */
96 	grant_handle_t		*pending_grant_handles;
97 };
98 
99 static struct xen_blkbk *blkbk;
100 
101 /*
102  * Little helpful macro to figure out the index and virtual address of the
103  * pending_pages[..]. For each 'pending_req' we have have up to
104  * BLKIF_MAX_SEGMENTS_PER_REQUEST (11) pages. The seg would be from 0 through
105  * 10 and would index in the pending_pages[..].
106  */
107 static inline int vaddr_pagenr(struct pending_req *req, int seg)
108 {
109 	return (req - blkbk->pending_reqs) *
110 		BLKIF_MAX_SEGMENTS_PER_REQUEST + seg;
111 }
112 
113 #define pending_page(req, seg) pending_pages[vaddr_pagenr(req, seg)]
114 
115 static inline unsigned long vaddr(struct pending_req *req, int seg)
116 {
117 	unsigned long pfn = page_to_pfn(blkbk->pending_page(req, seg));
118 	return (unsigned long)pfn_to_kaddr(pfn);
119 }
120 
121 #define pending_handle(_req, _seg) \
122 	(blkbk->pending_grant_handles[vaddr_pagenr(_req, _seg)])
123 
124 
125 static int do_block_io_op(struct xen_blkif *blkif);
126 static int dispatch_rw_block_io(struct xen_blkif *blkif,
127 				struct blkif_request *req,
128 				struct pending_req *pending_req);
129 static void make_response(struct xen_blkif *blkif, u64 id,
130 			  unsigned short op, int st);
131 
132 /*
133  * Retrieve from the 'pending_reqs' a free pending_req structure to be used.
134  */
135 static struct pending_req *alloc_req(void)
136 {
137 	struct pending_req *req = NULL;
138 	unsigned long flags;
139 
140 	spin_lock_irqsave(&blkbk->pending_free_lock, flags);
141 	if (!list_empty(&blkbk->pending_free)) {
142 		req = list_entry(blkbk->pending_free.next, struct pending_req,
143 				 free_list);
144 		list_del(&req->free_list);
145 	}
146 	spin_unlock_irqrestore(&blkbk->pending_free_lock, flags);
147 	return req;
148 }
149 
150 /*
151  * Return the 'pending_req' structure back to the freepool. We also
152  * wake up the thread if it was waiting for a free page.
153  */
154 static void free_req(struct pending_req *req)
155 {
156 	unsigned long flags;
157 	int was_empty;
158 
159 	spin_lock_irqsave(&blkbk->pending_free_lock, flags);
160 	was_empty = list_empty(&blkbk->pending_free);
161 	list_add(&req->free_list, &blkbk->pending_free);
162 	spin_unlock_irqrestore(&blkbk->pending_free_lock, flags);
163 	if (was_empty)
164 		wake_up(&blkbk->pending_free_wq);
165 }
166 
167 /*
168  * Routines for managing virtual block devices (vbds).
169  */
170 static int xen_vbd_translate(struct phys_req *req, struct xen_blkif *blkif,
171 			     int operation)
172 {
173 	struct xen_vbd *vbd = &blkif->vbd;
174 	int rc = -EACCES;
175 
176 	if ((operation != READ) && vbd->readonly)
177 		goto out;
178 
179 	if (likely(req->nr_sects)) {
180 		blkif_sector_t end = req->sector_number + req->nr_sects;
181 
182 		if (unlikely(end < req->sector_number))
183 			goto out;
184 		if (unlikely(end > vbd_sz(vbd)))
185 			goto out;
186 	}
187 
188 	req->dev  = vbd->pdevice;
189 	req->bdev = vbd->bdev;
190 	rc = 0;
191 
192  out:
193 	return rc;
194 }
195 
196 static void xen_vbd_resize(struct xen_blkif *blkif)
197 {
198 	struct xen_vbd *vbd = &blkif->vbd;
199 	struct xenbus_transaction xbt;
200 	int err;
201 	struct xenbus_device *dev = xen_blkbk_xenbus(blkif->be);
202 	unsigned long long new_size = vbd_sz(vbd);
203 
204 	pr_info(DRV_PFX "VBD Resize: Domid: %d, Device: (%d, %d)\n",
205 		blkif->domid, MAJOR(vbd->pdevice), MINOR(vbd->pdevice));
206 	pr_info(DRV_PFX "VBD Resize: new size %llu\n", new_size);
207 	vbd->size = new_size;
208 again:
209 	err = xenbus_transaction_start(&xbt);
210 	if (err) {
211 		pr_warn(DRV_PFX "Error starting transaction");
212 		return;
213 	}
214 	err = xenbus_printf(xbt, dev->nodename, "sectors", "%llu",
215 			    (unsigned long long)vbd_sz(vbd));
216 	if (err) {
217 		pr_warn(DRV_PFX "Error writing new size");
218 		goto abort;
219 	}
220 	/*
221 	 * Write the current state; we will use this to synchronize
222 	 * the front-end. If the current state is "connected" the
223 	 * front-end will get the new size information online.
224 	 */
225 	err = xenbus_printf(xbt, dev->nodename, "state", "%d", dev->state);
226 	if (err) {
227 		pr_warn(DRV_PFX "Error writing the state");
228 		goto abort;
229 	}
230 
231 	err = xenbus_transaction_end(xbt, 0);
232 	if (err == -EAGAIN)
233 		goto again;
234 	if (err)
235 		pr_warn(DRV_PFX "Error ending transaction");
236 	return;
237 abort:
238 	xenbus_transaction_end(xbt, 1);
239 }
240 
241 /*
242  * Notification from the guest OS.
243  */
244 static void blkif_notify_work(struct xen_blkif *blkif)
245 {
246 	blkif->waiting_reqs = 1;
247 	wake_up(&blkif->wq);
248 }
249 
250 irqreturn_t xen_blkif_be_int(int irq, void *dev_id)
251 {
252 	blkif_notify_work(dev_id);
253 	return IRQ_HANDLED;
254 }
255 
256 /*
257  * SCHEDULER FUNCTIONS
258  */
259 
260 static void print_stats(struct xen_blkif *blkif)
261 {
262 	pr_info("xen-blkback (%s): oo %3d  |  rd %4d  |  wr %4d  |  f %4d"
263 		 "  |  ds %4d\n",
264 		 current->comm, blkif->st_oo_req,
265 		 blkif->st_rd_req, blkif->st_wr_req,
266 		 blkif->st_f_req, blkif->st_ds_req);
267 	blkif->st_print = jiffies + msecs_to_jiffies(10 * 1000);
268 	blkif->st_rd_req = 0;
269 	blkif->st_wr_req = 0;
270 	blkif->st_oo_req = 0;
271 	blkif->st_ds_req = 0;
272 }
273 
274 int xen_blkif_schedule(void *arg)
275 {
276 	struct xen_blkif *blkif = arg;
277 	struct xen_vbd *vbd = &blkif->vbd;
278 
279 	xen_blkif_get(blkif);
280 
281 	while (!kthread_should_stop()) {
282 		if (try_to_freeze())
283 			continue;
284 		if (unlikely(vbd->size != vbd_sz(vbd)))
285 			xen_vbd_resize(blkif);
286 
287 		wait_event_interruptible(
288 			blkif->wq,
289 			blkif->waiting_reqs || kthread_should_stop());
290 		wait_event_interruptible(
291 			blkbk->pending_free_wq,
292 			!list_empty(&blkbk->pending_free) ||
293 			kthread_should_stop());
294 
295 		blkif->waiting_reqs = 0;
296 		smp_mb(); /* clear flag *before* checking for work */
297 
298 		if (do_block_io_op(blkif))
299 			blkif->waiting_reqs = 1;
300 
301 		if (log_stats && time_after(jiffies, blkif->st_print))
302 			print_stats(blkif);
303 	}
304 
305 	if (log_stats)
306 		print_stats(blkif);
307 
308 	blkif->xenblkd = NULL;
309 	xen_blkif_put(blkif);
310 
311 	return 0;
312 }
313 
314 struct seg_buf {
315 	unsigned long buf;
316 	unsigned int nsec;
317 };
318 /*
319  * Unmap the grant references, and also remove the M2P over-rides
320  * used in the 'pending_req'.
321  */
322 static void xen_blkbk_unmap(struct pending_req *req)
323 {
324 	struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST];
325 	struct page *pages[BLKIF_MAX_SEGMENTS_PER_REQUEST];
326 	unsigned int i, invcount = 0;
327 	grant_handle_t handle;
328 	int ret;
329 
330 	for (i = 0; i < req->nr_pages; i++) {
331 		handle = pending_handle(req, i);
332 		if (handle == BLKBACK_INVALID_HANDLE)
333 			continue;
334 		gnttab_set_unmap_op(&unmap[invcount], vaddr(req, i),
335 				    GNTMAP_host_map, handle);
336 		pending_handle(req, i) = BLKBACK_INVALID_HANDLE;
337 		pages[invcount] = virt_to_page(vaddr(req, i));
338 		invcount++;
339 	}
340 
341 	ret = gnttab_unmap_refs(unmap, NULL, pages, invcount);
342 	BUG_ON(ret);
343 }
344 
345 static int xen_blkbk_map(struct blkif_request *req,
346 			 struct pending_req *pending_req,
347 			 struct seg_buf seg[])
348 {
349 	struct gnttab_map_grant_ref map[BLKIF_MAX_SEGMENTS_PER_REQUEST];
350 	int i;
351 	int nseg = req->u.rw.nr_segments;
352 	int ret = 0;
353 
354 	/*
355 	 * Fill out preq.nr_sects with proper amount of sectors, and setup
356 	 * assign map[..] with the PFN of the page in our domain with the
357 	 * corresponding grant reference for each page.
358 	 */
359 	for (i = 0; i < nseg; i++) {
360 		uint32_t flags;
361 
362 		flags = GNTMAP_host_map;
363 		if (pending_req->operation != BLKIF_OP_READ)
364 			flags |= GNTMAP_readonly;
365 		gnttab_set_map_op(&map[i], vaddr(pending_req, i), flags,
366 				  req->u.rw.seg[i].gref,
367 				  pending_req->blkif->domid);
368 	}
369 
370 	ret = gnttab_map_refs(map, NULL, &blkbk->pending_page(pending_req, 0), nseg);
371 	BUG_ON(ret);
372 
373 	/*
374 	 * Now swizzle the MFN in our domain with the MFN from the other domain
375 	 * so that when we access vaddr(pending_req,i) it has the contents of
376 	 * the page from the other domain.
377 	 */
378 	for (i = 0; i < nseg; i++) {
379 		if (unlikely(map[i].status != 0)) {
380 			pr_debug(DRV_PFX "invalid buffer -- could not remap it\n");
381 			map[i].handle = BLKBACK_INVALID_HANDLE;
382 			ret |= 1;
383 		}
384 
385 		pending_handle(pending_req, i) = map[i].handle;
386 
387 		if (ret)
388 			continue;
389 
390 		seg[i].buf  = map[i].dev_bus_addr |
391 			(req->u.rw.seg[i].first_sect << 9);
392 	}
393 	return ret;
394 }
395 
396 static int dispatch_discard_io(struct xen_blkif *blkif,
397 				struct blkif_request *req)
398 {
399 	int err = 0;
400 	int status = BLKIF_RSP_OKAY;
401 	struct block_device *bdev = blkif->vbd.bdev;
402 	unsigned long secure;
403 
404 	blkif->st_ds_req++;
405 
406 	xen_blkif_get(blkif);
407 	secure = (blkif->vbd.discard_secure &&
408 		 (req->u.discard.flag & BLKIF_DISCARD_SECURE)) ?
409 		 BLKDEV_DISCARD_SECURE : 0;
410 
411 	err = blkdev_issue_discard(bdev, req->u.discard.sector_number,
412 				   req->u.discard.nr_sectors,
413 				   GFP_KERNEL, secure);
414 
415 	if (err == -EOPNOTSUPP) {
416 		pr_debug(DRV_PFX "discard op failed, not supported\n");
417 		status = BLKIF_RSP_EOPNOTSUPP;
418 	} else if (err)
419 		status = BLKIF_RSP_ERROR;
420 
421 	make_response(blkif, req->u.discard.id, req->operation, status);
422 	xen_blkif_put(blkif);
423 	return err;
424 }
425 
426 static void xen_blk_drain_io(struct xen_blkif *blkif)
427 {
428 	atomic_set(&blkif->drain, 1);
429 	do {
430 		/* The initial value is one, and one refcnt taken at the
431 		 * start of the xen_blkif_schedule thread. */
432 		if (atomic_read(&blkif->refcnt) <= 2)
433 			break;
434 		wait_for_completion_interruptible_timeout(
435 				&blkif->drain_complete, HZ);
436 
437 		if (!atomic_read(&blkif->drain))
438 			break;
439 	} while (!kthread_should_stop());
440 	atomic_set(&blkif->drain, 0);
441 }
442 
443 /*
444  * Completion callback on the bio's. Called as bh->b_end_io()
445  */
446 
447 static void __end_block_io_op(struct pending_req *pending_req, int error)
448 {
449 	/* An error fails the entire request. */
450 	if ((pending_req->operation == BLKIF_OP_FLUSH_DISKCACHE) &&
451 	    (error == -EOPNOTSUPP)) {
452 		pr_debug(DRV_PFX "flush diskcache op failed, not supported\n");
453 		xen_blkbk_flush_diskcache(XBT_NIL, pending_req->blkif->be, 0);
454 		pending_req->status = BLKIF_RSP_EOPNOTSUPP;
455 	} else if ((pending_req->operation == BLKIF_OP_WRITE_BARRIER) &&
456 		    (error == -EOPNOTSUPP)) {
457 		pr_debug(DRV_PFX "write barrier op failed, not supported\n");
458 		xen_blkbk_barrier(XBT_NIL, pending_req->blkif->be, 0);
459 		pending_req->status = BLKIF_RSP_EOPNOTSUPP;
460 	} else if (error) {
461 		pr_debug(DRV_PFX "Buffer not up-to-date at end of operation,"
462 			 " error=%d\n", error);
463 		pending_req->status = BLKIF_RSP_ERROR;
464 	}
465 
466 	/*
467 	 * If all of the bio's have completed it is time to unmap
468 	 * the grant references associated with 'request' and provide
469 	 * the proper response on the ring.
470 	 */
471 	if (atomic_dec_and_test(&pending_req->pendcnt)) {
472 		xen_blkbk_unmap(pending_req);
473 		make_response(pending_req->blkif, pending_req->id,
474 			      pending_req->operation, pending_req->status);
475 		xen_blkif_put(pending_req->blkif);
476 		if (atomic_read(&pending_req->blkif->refcnt) <= 2) {
477 			if (atomic_read(&pending_req->blkif->drain))
478 				complete(&pending_req->blkif->drain_complete);
479 		}
480 		free_req(pending_req);
481 	}
482 }
483 
484 /*
485  * bio callback.
486  */
487 static void end_block_io_op(struct bio *bio, int error)
488 {
489 	__end_block_io_op(bio->bi_private, error);
490 	bio_put(bio);
491 }
492 
493 
494 
495 /*
496  * Function to copy the from the ring buffer the 'struct blkif_request'
497  * (which has the sectors we want, number of them, grant references, etc),
498  * and transmute  it to the block API to hand it over to the proper block disk.
499  */
500 static int
501 __do_block_io_op(struct xen_blkif *blkif)
502 {
503 	union blkif_back_rings *blk_rings = &blkif->blk_rings;
504 	struct blkif_request req;
505 	struct pending_req *pending_req;
506 	RING_IDX rc, rp;
507 	int more_to_do = 0;
508 
509 	rc = blk_rings->common.req_cons;
510 	rp = blk_rings->common.sring->req_prod;
511 	rmb(); /* Ensure we see queued requests up to 'rp'. */
512 
513 	while (rc != rp) {
514 
515 		if (RING_REQUEST_CONS_OVERFLOW(&blk_rings->common, rc))
516 			break;
517 
518 		if (kthread_should_stop()) {
519 			more_to_do = 1;
520 			break;
521 		}
522 
523 		pending_req = alloc_req();
524 		if (NULL == pending_req) {
525 			blkif->st_oo_req++;
526 			more_to_do = 1;
527 			break;
528 		}
529 
530 		switch (blkif->blk_protocol) {
531 		case BLKIF_PROTOCOL_NATIVE:
532 			memcpy(&req, RING_GET_REQUEST(&blk_rings->native, rc), sizeof(req));
533 			break;
534 		case BLKIF_PROTOCOL_X86_32:
535 			blkif_get_x86_32_req(&req, RING_GET_REQUEST(&blk_rings->x86_32, rc));
536 			break;
537 		case BLKIF_PROTOCOL_X86_64:
538 			blkif_get_x86_64_req(&req, RING_GET_REQUEST(&blk_rings->x86_64, rc));
539 			break;
540 		default:
541 			BUG();
542 		}
543 		blk_rings->common.req_cons = ++rc; /* before make_response() */
544 
545 		/* Apply all sanity checks to /private copy/ of request. */
546 		barrier();
547 		if (unlikely(req.operation == BLKIF_OP_DISCARD)) {
548 			free_req(pending_req);
549 			if (dispatch_discard_io(blkif, &req))
550 				break;
551 		} else if (dispatch_rw_block_io(blkif, &req, pending_req))
552 			break;
553 
554 		/* Yield point for this unbounded loop. */
555 		cond_resched();
556 	}
557 
558 	return more_to_do;
559 }
560 
561 static int
562 do_block_io_op(struct xen_blkif *blkif)
563 {
564 	union blkif_back_rings *blk_rings = &blkif->blk_rings;
565 	int more_to_do;
566 
567 	do {
568 		more_to_do = __do_block_io_op(blkif);
569 		if (more_to_do)
570 			break;
571 
572 		RING_FINAL_CHECK_FOR_REQUESTS(&blk_rings->common, more_to_do);
573 	} while (more_to_do);
574 
575 	return more_to_do;
576 }
577 /*
578  * Transmutation of the 'struct blkif_request' to a proper 'struct bio'
579  * and call the 'submit_bio' to pass it to the underlying storage.
580  */
581 static int dispatch_rw_block_io(struct xen_blkif *blkif,
582 				struct blkif_request *req,
583 				struct pending_req *pending_req)
584 {
585 	struct phys_req preq;
586 	struct seg_buf seg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
587 	unsigned int nseg;
588 	struct bio *bio = NULL;
589 	struct bio *biolist[BLKIF_MAX_SEGMENTS_PER_REQUEST];
590 	int i, nbio = 0;
591 	int operation;
592 	struct blk_plug plug;
593 	bool drain = false;
594 
595 	switch (req->operation) {
596 	case BLKIF_OP_READ:
597 		blkif->st_rd_req++;
598 		operation = READ;
599 		break;
600 	case BLKIF_OP_WRITE:
601 		blkif->st_wr_req++;
602 		operation = WRITE_ODIRECT;
603 		break;
604 	case BLKIF_OP_WRITE_BARRIER:
605 		drain = true;
606 	case BLKIF_OP_FLUSH_DISKCACHE:
607 		blkif->st_f_req++;
608 		operation = WRITE_FLUSH;
609 		break;
610 	default:
611 		operation = 0; /* make gcc happy */
612 		goto fail_response;
613 		break;
614 	}
615 
616 	/* Check that the number of segments is sane. */
617 	nseg = req->u.rw.nr_segments;
618 
619 	if (unlikely(nseg == 0 && operation != WRITE_FLUSH) ||
620 	    unlikely(nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
621 		pr_debug(DRV_PFX "Bad number of segments in request (%d)\n",
622 			 nseg);
623 		/* Haven't submitted any bio's yet. */
624 		goto fail_response;
625 	}
626 
627 	preq.dev           = req->u.rw.handle;
628 	preq.sector_number = req->u.rw.sector_number;
629 	preq.nr_sects      = 0;
630 
631 	pending_req->blkif     = blkif;
632 	pending_req->id        = req->u.rw.id;
633 	pending_req->operation = req->operation;
634 	pending_req->status    = BLKIF_RSP_OKAY;
635 	pending_req->nr_pages  = nseg;
636 
637 	for (i = 0; i < nseg; i++) {
638 		seg[i].nsec = req->u.rw.seg[i].last_sect -
639 			req->u.rw.seg[i].first_sect + 1;
640 		if ((req->u.rw.seg[i].last_sect >= (PAGE_SIZE >> 9)) ||
641 		    (req->u.rw.seg[i].last_sect < req->u.rw.seg[i].first_sect))
642 			goto fail_response;
643 		preq.nr_sects += seg[i].nsec;
644 
645 	}
646 
647 	if (xen_vbd_translate(&preq, blkif, operation) != 0) {
648 		pr_debug(DRV_PFX "access denied: %s of [%llu,%llu] on dev=%04x\n",
649 			 operation == READ ? "read" : "write",
650 			 preq.sector_number,
651 			 preq.sector_number + preq.nr_sects, preq.dev);
652 		goto fail_response;
653 	}
654 
655 	/*
656 	 * This check _MUST_ be done after xen_vbd_translate as the preq.bdev
657 	 * is set there.
658 	 */
659 	for (i = 0; i < nseg; i++) {
660 		if (((int)preq.sector_number|(int)seg[i].nsec) &
661 		    ((bdev_logical_block_size(preq.bdev) >> 9) - 1)) {
662 			pr_debug(DRV_PFX "Misaligned I/O request from domain %d",
663 				 blkif->domid);
664 			goto fail_response;
665 		}
666 	}
667 
668 	/* Wait on all outstanding I/O's and once that has been completed
669 	 * issue the WRITE_FLUSH.
670 	 */
671 	if (drain)
672 		xen_blk_drain_io(pending_req->blkif);
673 
674 	/*
675 	 * If we have failed at this point, we need to undo the M2P override,
676 	 * set gnttab_set_unmap_op on all of the grant references and perform
677 	 * the hypercall to unmap the grants - that is all done in
678 	 * xen_blkbk_unmap.
679 	 */
680 	if (xen_blkbk_map(req, pending_req, seg))
681 		goto fail_flush;
682 
683 	/*
684 	 * This corresponding xen_blkif_put is done in __end_block_io_op, or
685 	 * below (in "!bio") if we are handling a BLKIF_OP_DISCARD.
686 	 */
687 	xen_blkif_get(blkif);
688 
689 	for (i = 0; i < nseg; i++) {
690 		while ((bio == NULL) ||
691 		       (bio_add_page(bio,
692 				     blkbk->pending_page(pending_req, i),
693 				     seg[i].nsec << 9,
694 				     seg[i].buf & ~PAGE_MASK) == 0)) {
695 
696 			bio = bio_alloc(GFP_KERNEL, nseg-i);
697 			if (unlikely(bio == NULL))
698 				goto fail_put_bio;
699 
700 			biolist[nbio++] = bio;
701 			bio->bi_bdev    = preq.bdev;
702 			bio->bi_private = pending_req;
703 			bio->bi_end_io  = end_block_io_op;
704 			bio->bi_sector  = preq.sector_number;
705 		}
706 
707 		preq.sector_number += seg[i].nsec;
708 	}
709 
710 	/* This will be hit if the operation was a flush or discard. */
711 	if (!bio) {
712 		BUG_ON(operation != WRITE_FLUSH);
713 
714 		bio = bio_alloc(GFP_KERNEL, 0);
715 		if (unlikely(bio == NULL))
716 			goto fail_put_bio;
717 
718 		biolist[nbio++] = bio;
719 		bio->bi_bdev    = preq.bdev;
720 		bio->bi_private = pending_req;
721 		bio->bi_end_io  = end_block_io_op;
722 	}
723 
724 	/*
725 	 * We set it one so that the last submit_bio does not have to call
726 	 * atomic_inc.
727 	 */
728 	atomic_set(&pending_req->pendcnt, nbio);
729 
730 	/* Get a reference count for the disk queue and start sending I/O */
731 	blk_start_plug(&plug);
732 
733 	for (i = 0; i < nbio; i++)
734 		submit_bio(operation, biolist[i]);
735 
736 	/* Let the I/Os go.. */
737 	blk_finish_plug(&plug);
738 
739 	if (operation == READ)
740 		blkif->st_rd_sect += preq.nr_sects;
741 	else if (operation & WRITE)
742 		blkif->st_wr_sect += preq.nr_sects;
743 
744 	return 0;
745 
746  fail_flush:
747 	xen_blkbk_unmap(pending_req);
748  fail_response:
749 	/* Haven't submitted any bio's yet. */
750 	make_response(blkif, req->u.rw.id, req->operation, BLKIF_RSP_ERROR);
751 	free_req(pending_req);
752 	msleep(1); /* back off a bit */
753 	return -EIO;
754 
755  fail_put_bio:
756 	for (i = 0; i < nbio; i++)
757 		bio_put(biolist[i]);
758 	__end_block_io_op(pending_req, -EINVAL);
759 	msleep(1); /* back off a bit */
760 	return -EIO;
761 }
762 
763 
764 
765 /*
766  * Put a response on the ring on how the operation fared.
767  */
768 static void make_response(struct xen_blkif *blkif, u64 id,
769 			  unsigned short op, int st)
770 {
771 	struct blkif_response  resp;
772 	unsigned long     flags;
773 	union blkif_back_rings *blk_rings = &blkif->blk_rings;
774 	int notify;
775 
776 	resp.id        = id;
777 	resp.operation = op;
778 	resp.status    = st;
779 
780 	spin_lock_irqsave(&blkif->blk_ring_lock, flags);
781 	/* Place on the response ring for the relevant domain. */
782 	switch (blkif->blk_protocol) {
783 	case BLKIF_PROTOCOL_NATIVE:
784 		memcpy(RING_GET_RESPONSE(&blk_rings->native, blk_rings->native.rsp_prod_pvt),
785 		       &resp, sizeof(resp));
786 		break;
787 	case BLKIF_PROTOCOL_X86_32:
788 		memcpy(RING_GET_RESPONSE(&blk_rings->x86_32, blk_rings->x86_32.rsp_prod_pvt),
789 		       &resp, sizeof(resp));
790 		break;
791 	case BLKIF_PROTOCOL_X86_64:
792 		memcpy(RING_GET_RESPONSE(&blk_rings->x86_64, blk_rings->x86_64.rsp_prod_pvt),
793 		       &resp, sizeof(resp));
794 		break;
795 	default:
796 		BUG();
797 	}
798 	blk_rings->common.rsp_prod_pvt++;
799 	RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blk_rings->common, notify);
800 	spin_unlock_irqrestore(&blkif->blk_ring_lock, flags);
801 	if (notify)
802 		notify_remote_via_irq(blkif->irq);
803 }
804 
805 static int __init xen_blkif_init(void)
806 {
807 	int i, mmap_pages;
808 	int rc = 0;
809 
810 	if (!xen_domain())
811 		return -ENODEV;
812 
813 	blkbk = kzalloc(sizeof(struct xen_blkbk), GFP_KERNEL);
814 	if (!blkbk) {
815 		pr_alert(DRV_PFX "%s: out of memory!\n", __func__);
816 		return -ENOMEM;
817 	}
818 
819 	mmap_pages = xen_blkif_reqs * BLKIF_MAX_SEGMENTS_PER_REQUEST;
820 
821 	blkbk->pending_reqs          = kzalloc(sizeof(blkbk->pending_reqs[0]) *
822 					xen_blkif_reqs, GFP_KERNEL);
823 	blkbk->pending_grant_handles = kmalloc(sizeof(blkbk->pending_grant_handles[0]) *
824 					mmap_pages, GFP_KERNEL);
825 	blkbk->pending_pages         = kzalloc(sizeof(blkbk->pending_pages[0]) *
826 					mmap_pages, GFP_KERNEL);
827 
828 	if (!blkbk->pending_reqs || !blkbk->pending_grant_handles ||
829 	    !blkbk->pending_pages) {
830 		rc = -ENOMEM;
831 		goto out_of_memory;
832 	}
833 
834 	for (i = 0; i < mmap_pages; i++) {
835 		blkbk->pending_grant_handles[i] = BLKBACK_INVALID_HANDLE;
836 		blkbk->pending_pages[i] = alloc_page(GFP_KERNEL);
837 		if (blkbk->pending_pages[i] == NULL) {
838 			rc = -ENOMEM;
839 			goto out_of_memory;
840 		}
841 	}
842 	rc = xen_blkif_interface_init();
843 	if (rc)
844 		goto failed_init;
845 
846 	INIT_LIST_HEAD(&blkbk->pending_free);
847 	spin_lock_init(&blkbk->pending_free_lock);
848 	init_waitqueue_head(&blkbk->pending_free_wq);
849 
850 	for (i = 0; i < xen_blkif_reqs; i++)
851 		list_add_tail(&blkbk->pending_reqs[i].free_list,
852 			      &blkbk->pending_free);
853 
854 	rc = xen_blkif_xenbus_init();
855 	if (rc)
856 		goto failed_init;
857 
858 	return 0;
859 
860  out_of_memory:
861 	pr_alert(DRV_PFX "%s: out of memory\n", __func__);
862  failed_init:
863 	kfree(blkbk->pending_reqs);
864 	kfree(blkbk->pending_grant_handles);
865 	if (blkbk->pending_pages) {
866 		for (i = 0; i < mmap_pages; i++) {
867 			if (blkbk->pending_pages[i])
868 				__free_page(blkbk->pending_pages[i]);
869 		}
870 		kfree(blkbk->pending_pages);
871 	}
872 	kfree(blkbk);
873 	blkbk = NULL;
874 	return rc;
875 }
876 
877 module_init(xen_blkif_init);
878 
879 MODULE_LICENSE("Dual BSD/GPL");
880 MODULE_ALIAS("xen-backend:vbd");
881