1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
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
54
55
56
57
58
59
60
61
62static int xen_blkif_reqs = 64;
63module_param_named(reqs, xen_blkif_reqs, int, 0);
64MODULE_PARM_DESC(reqs, "Number of blkback requests to allocate");
65
66
67static unsigned int log_stats;
68module_param(log_stats, int, 0644);
69
70
71
72
73
74
75
76struct 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
89struct xen_blkbk {
90 struct pending_req *pending_reqs;
91
92 struct list_head pending_free;
93
94 spinlock_t pending_free_lock;
95 wait_queue_head_t pending_free_wq;
96
97 struct page **pending_pages;
98
99 grant_handle_t *pending_grant_handles;
100};
101
102static struct xen_blkbk *blkbk;
103
104
105
106
107
108
109
110
111
112
113
114
115static 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
136
137
138
139
140static 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
148static 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
158static int do_block_io_op(struct xen_blkif *blkif);
159static int dispatch_rw_block_io(struct xen_blkif *blkif,
160 struct blkif_request *req,
161 struct pending_req *pending_req);
162static 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
173static 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
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
195 rb_link_node(&(persistent_gnt->node), parent, new);
196 rb_insert_color(&(persistent_gnt->node), root);
197}
198
199static 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
218static 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
256
257static 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
274
275
276static 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
291
292static 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
318static 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;
330again:
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
344
345
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;
359abort:
360 xenbus_transaction_end(xbt, 1);
361}
362
363
364
365
366static void blkif_notify_work(struct xen_blkif *blkif)
367{
368 blkif->waiting_reqs = 1;
369 wake_up(&blkif->wq);
370}
371
372irqreturn_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
380
381
382static 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
396int 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();
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
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
444struct seg_buf {
445 unsigned int offset;
446 unsigned int nsec;
447};
448
449
450
451
452static 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
477static 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
502
503
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
516
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
524
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
554
555
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
595
596
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
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
622
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
643static 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
673static 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
683static void xen_blk_drain_io(struct xen_blkif *blkif)
684{
685 atomic_set(&blkif->drain, 1);
686 do {
687
688
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
702
703
704static void __end_block_io_op(struct pending_req *pending_req, int error)
705{
706
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
725
726
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
743
744static 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
754
755
756
757static 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();
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;
801
802
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
825 cond_resched();
826 }
827done:
828 return more_to_do;
829}
830
831static int
832do_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
849
850
851static 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;
883 goto fail_response;
884 break;
885 }
886
887
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
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
928
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
940
941
942 if (drain)
943 xen_blk_drain_io(pending_req->blkif);
944
945
946
947
948
949
950
951 if (xen_blkbk_map(req, pending_req, seg, pages))
952 goto fail_flush;
953
954
955
956
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
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
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
1015 make_response(blkif, req->u.rw.id, req->operation, BLKIF_RSP_ERROR);
1016 free_req(pending_req);
1017 msleep(1);
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);
1026 return -EIO;
1027}
1028
1029
1030
1031
1032
1033
1034static 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
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
1071static 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
1143module_init(xen_blkif_init);
1144
1145MODULE_LICENSE("Dual BSD/GPL");
1146MODULE_ALIAS("xen-backend:vbd");
1147