1 /******************************************************************************
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
9 * Copyright (c) 2003-2004, Keir Fraser & Steve Hand
10 * Copyright (c) 2005, Christopher Clark
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:
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:
25 * The above copyright notice and this permission notice shall be included in
26 * all copies or substantial portions of the Software.
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
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>
44 #include <xen/events.h>
47 #include <asm/xen/hypervisor.h>
48 #include <asm/xen/hypercall.h>
49 #include <xen/balloon.h>
53 * Maximum number of unused free pages to keep in the internal buffer.
54 * Setting this to a value too low will reduce memory used in each backend,
55 * but can have a performance penalty.
57 * A sane value is xen_blkif_reqs * BLKIF_MAX_SEGMENTS_PER_REQUEST, but can
58 * be set to a lower value that might degrade performance on some intensive
62 static int xen_blkif_max_buffer_pages = 1024;
63 module_param_named(max_buffer_pages, xen_blkif_max_buffer_pages, int, 0644);
64 MODULE_PARM_DESC(max_buffer_pages,
65 "Maximum number of free pages to keep in each block backend buffer");
68 * Maximum number of grants to map persistently in blkback. For maximum
69 * performance this should be the total numbers of grants that can be used
70 * to fill the ring, but since this might become too high, specially with
71 * the use of indirect descriptors, we set it to a value that provides good
72 * performance without using too much memory.
74 * When the list of persistent grants is full we clean it up using a LRU
78 static int xen_blkif_max_pgrants = 1056;
79 module_param_named(max_persistent_grants, xen_blkif_max_pgrants, int, 0644);
80 MODULE_PARM_DESC(max_persistent_grants,
81 "Maximum number of grants to map persistently");
84 * The LRU mechanism to clean the lists of persistent grants needs to
85 * be executed periodically. The time interval between consecutive executions
86 * of the purge mechanism is set in ms.
88 #define LRU_INTERVAL 100
91 * When the persistent grants list is full we will remove unused grants
92 * from the list. The percent number of grants to be removed at each LRU
95 #define LRU_PERCENT_CLEAN 5
97 /* Run-time switchable: /sys/module/blkback/parameters/ */
98 static unsigned int log_stats;
99 module_param(log_stats, int, 0644);
101 #define BLKBACK_INVALID_HANDLE (~0)
103 /* Number of free pages to remove on each call to free_xenballooned_pages */
104 #define NUM_BATCH_FREE_PAGES 10
106 static inline int get_free_page(struct xen_blkif *blkif, struct page **page)
110 spin_lock_irqsave(&blkif->free_pages_lock, flags);
111 if (list_empty(&blkif->free_pages)) {
112 BUG_ON(blkif->free_pages_num != 0);
113 spin_unlock_irqrestore(&blkif->free_pages_lock, flags);
114 return alloc_xenballooned_pages(1, page, false);
116 BUG_ON(blkif->free_pages_num == 0);
117 page[0] = list_first_entry(&blkif->free_pages, struct page, lru);
118 list_del(&page[0]->lru);
119 blkif->free_pages_num--;
120 spin_unlock_irqrestore(&blkif->free_pages_lock, flags);
125 static inline void put_free_pages(struct xen_blkif *blkif, struct page **page,
131 spin_lock_irqsave(&blkif->free_pages_lock, flags);
132 for (i = 0; i < num; i++)
133 list_add(&page[i]->lru, &blkif->free_pages);
134 blkif->free_pages_num += num;
135 spin_unlock_irqrestore(&blkif->free_pages_lock, flags);
138 static inline void shrink_free_pagepool(struct xen_blkif *blkif, int num)
140 /* Remove requested pages in batches of NUM_BATCH_FREE_PAGES */
141 struct page *page[NUM_BATCH_FREE_PAGES];
142 unsigned int num_pages = 0;
145 spin_lock_irqsave(&blkif->free_pages_lock, flags);
146 while (blkif->free_pages_num > num) {
147 BUG_ON(list_empty(&blkif->free_pages));
148 page[num_pages] = list_first_entry(&blkif->free_pages,
150 list_del(&page[num_pages]->lru);
151 blkif->free_pages_num--;
152 if (++num_pages == NUM_BATCH_FREE_PAGES) {
153 spin_unlock_irqrestore(&blkif->free_pages_lock, flags);
154 free_xenballooned_pages(num_pages, page);
155 spin_lock_irqsave(&blkif->free_pages_lock, flags);
159 spin_unlock_irqrestore(&blkif->free_pages_lock, flags);
161 free_xenballooned_pages(num_pages, page);
164 #define vaddr(page) ((unsigned long)pfn_to_kaddr(page_to_pfn(page)))
166 static int do_block_io_op(struct xen_blkif *blkif);
167 static int dispatch_rw_block_io(struct xen_blkif *blkif,
168 struct blkif_request *req,
169 struct pending_req *pending_req);
170 static void make_response(struct xen_blkif *blkif, u64 id,
171 unsigned short op, int st);
173 #define foreach_grant_safe(pos, n, rbtree, node) \
174 for ((pos) = container_of(rb_first((rbtree)), typeof(*(pos)), node), \
175 (n) = (&(pos)->node != NULL) ? rb_next(&(pos)->node) : NULL; \
176 &(pos)->node != NULL; \
177 (pos) = container_of(n, typeof(*(pos)), node), \
178 (n) = (&(pos)->node != NULL) ? rb_next(&(pos)->node) : NULL)
182 * We don't need locking around the persistent grant helpers
183 * because blkback uses a single-thread for each backed, so we
184 * can be sure that this functions will never be called recursively.
186 * The only exception to that is put_persistent_grant, that can be called
187 * from interrupt context (by xen_blkbk_unmap), so we have to use atomic
188 * bit operations to modify the flags of a persistent grant and to count
189 * the number of used grants.
191 static int add_persistent_gnt(struct xen_blkif *blkif,
192 struct persistent_gnt *persistent_gnt)
194 struct rb_node **new = NULL, *parent = NULL;
195 struct persistent_gnt *this;
197 if (blkif->persistent_gnt_c >= xen_blkif_max_pgrants) {
198 if (!blkif->vbd.overflow_max_grants)
199 blkif->vbd.overflow_max_grants = 1;
202 /* Figure out where to put new node */
203 new = &blkif->persistent_gnts.rb_node;
205 this = container_of(*new, struct persistent_gnt, node);
208 if (persistent_gnt->gnt < this->gnt)
209 new = &((*new)->rb_left);
210 else if (persistent_gnt->gnt > this->gnt)
211 new = &((*new)->rb_right);
213 pr_alert_ratelimited(DRV_PFX " trying to add a gref that's already in the tree\n");
218 bitmap_zero(persistent_gnt->flags, PERSISTENT_GNT_FLAGS_SIZE);
219 set_bit(PERSISTENT_GNT_ACTIVE, persistent_gnt->flags);
220 /* Add new node and rebalance tree. */
221 rb_link_node(&(persistent_gnt->node), parent, new);
222 rb_insert_color(&(persistent_gnt->node), &blkif->persistent_gnts);
223 blkif->persistent_gnt_c++;
224 atomic_inc(&blkif->persistent_gnt_in_use);
228 static struct persistent_gnt *get_persistent_gnt(struct xen_blkif *blkif,
231 struct persistent_gnt *data;
232 struct rb_node *node = NULL;
234 node = blkif->persistent_gnts.rb_node;
236 data = container_of(node, struct persistent_gnt, node);
238 if (gref < data->gnt)
239 node = node->rb_left;
240 else if (gref > data->gnt)
241 node = node->rb_right;
243 if(test_bit(PERSISTENT_GNT_ACTIVE, data->flags)) {
244 pr_alert_ratelimited(DRV_PFX " requesting a grant already in use\n");
247 set_bit(PERSISTENT_GNT_ACTIVE, data->flags);
248 atomic_inc(&blkif->persistent_gnt_in_use);
255 static void put_persistent_gnt(struct xen_blkif *blkif,
256 struct persistent_gnt *persistent_gnt)
258 if(!test_bit(PERSISTENT_GNT_ACTIVE, persistent_gnt->flags))
259 pr_alert_ratelimited(DRV_PFX " freeing a grant already unused");
260 set_bit(PERSISTENT_GNT_WAS_ACTIVE, persistent_gnt->flags);
261 clear_bit(PERSISTENT_GNT_ACTIVE, persistent_gnt->flags);
262 atomic_dec(&blkif->persistent_gnt_in_use);
265 static void free_persistent_gnts(struct xen_blkif *blkif, struct rb_root *root,
268 struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST];
269 struct page *pages[BLKIF_MAX_SEGMENTS_PER_REQUEST];
270 struct persistent_gnt *persistent_gnt;
273 int segs_to_unmap = 0;
275 foreach_grant_safe(persistent_gnt, n, root, node) {
276 BUG_ON(persistent_gnt->handle ==
277 BLKBACK_INVALID_HANDLE);
278 gnttab_set_unmap_op(&unmap[segs_to_unmap],
279 (unsigned long) pfn_to_kaddr(page_to_pfn(
280 persistent_gnt->page)),
282 persistent_gnt->handle);
284 pages[segs_to_unmap] = persistent_gnt->page;
286 if (++segs_to_unmap == BLKIF_MAX_SEGMENTS_PER_REQUEST ||
287 !rb_next(&persistent_gnt->node)) {
288 ret = gnttab_unmap_refs(unmap, pages, segs_to_unmap);
290 put_free_pages(blkif, pages, segs_to_unmap);
294 rb_erase(&persistent_gnt->node, root);
295 kfree(persistent_gnt);
301 static void unmap_purged_grants(struct work_struct *work)
303 struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST];
304 struct page *pages[BLKIF_MAX_SEGMENTS_PER_REQUEST];
305 struct persistent_gnt *persistent_gnt;
306 int ret, segs_to_unmap = 0;
307 struct xen_blkif *blkif = container_of(work, typeof(*blkif), persistent_purge_work);
309 while(!list_empty(&blkif->persistent_purge_list)) {
310 persistent_gnt = list_first_entry(&blkif->persistent_purge_list,
311 struct persistent_gnt,
313 list_del(&persistent_gnt->remove_node);
315 gnttab_set_unmap_op(&unmap[segs_to_unmap],
316 vaddr(persistent_gnt->page),
318 persistent_gnt->handle);
320 pages[segs_to_unmap] = persistent_gnt->page;
322 if (++segs_to_unmap == BLKIF_MAX_SEGMENTS_PER_REQUEST) {
323 ret = gnttab_unmap_refs(unmap, pages, segs_to_unmap);
325 put_free_pages(blkif, pages, segs_to_unmap);
328 kfree(persistent_gnt);
330 if (segs_to_unmap > 0) {
331 ret = gnttab_unmap_refs(unmap, pages, segs_to_unmap);
333 put_free_pages(blkif, pages, segs_to_unmap);
337 static void purge_persistent_gnt(struct xen_blkif *blkif)
339 struct persistent_gnt *persistent_gnt;
341 unsigned int num_clean, total;
342 bool scan_used = false, clean_used = false;
343 struct rb_root *root;
345 if (blkif->persistent_gnt_c < xen_blkif_max_pgrants ||
346 (blkif->persistent_gnt_c == xen_blkif_max_pgrants &&
347 !blkif->vbd.overflow_max_grants)) {
351 if (work_pending(&blkif->persistent_purge_work)) {
352 pr_alert_ratelimited(DRV_PFX "Scheduled work from previous purge is still pending, cannot purge list\n");
356 num_clean = (xen_blkif_max_pgrants / 100) * LRU_PERCENT_CLEAN;
357 num_clean = blkif->persistent_gnt_c - xen_blkif_max_pgrants + num_clean;
358 num_clean = min(blkif->persistent_gnt_c, num_clean);
359 if ((num_clean == 0) ||
360 (num_clean > (blkif->persistent_gnt_c - atomic_read(&blkif->persistent_gnt_in_use))))
364 * At this point, we can assure that there will be no calls
365 * to get_persistent_grant (because we are executing this code from
366 * xen_blkif_schedule), there can only be calls to put_persistent_gnt,
367 * which means that the number of currently used grants will go down,
368 * but never up, so we will always be able to remove the requested
374 pr_debug(DRV_PFX "Going to purge %u persistent grants\n", num_clean);
376 INIT_LIST_HEAD(&blkif->persistent_purge_list);
377 root = &blkif->persistent_gnts;
379 foreach_grant_safe(persistent_gnt, n, root, node) {
380 BUG_ON(persistent_gnt->handle ==
381 BLKBACK_INVALID_HANDLE);
384 clear_bit(PERSISTENT_GNT_WAS_ACTIVE, persistent_gnt->flags);
388 if (test_bit(PERSISTENT_GNT_ACTIVE, persistent_gnt->flags))
391 (test_bit(PERSISTENT_GNT_WAS_ACTIVE, persistent_gnt->flags)))
394 rb_erase(&persistent_gnt->node, root);
395 list_add(&persistent_gnt->remove_node,
396 &blkif->persistent_purge_list);
397 if (--num_clean == 0)
401 * If we get here it means we also need to start cleaning
402 * grants that were used since last purge in order to cope
403 * with the requested num
405 if (!scan_used && !clean_used) {
406 pr_debug(DRV_PFX "Still missing %u purged frames\n", num_clean);
412 pr_debug(DRV_PFX "Finished scanning for grants to clean, removing used flag\n");
417 blkif->persistent_gnt_c -= (total - num_clean);
418 blkif->vbd.overflow_max_grants = 0;
420 /* We can defer this work */
421 INIT_WORK(&blkif->persistent_purge_work, unmap_purged_grants);
422 schedule_work(&blkif->persistent_purge_work);
423 pr_debug(DRV_PFX "Purged %u/%u\n", (total - num_clean), total);
428 * Retrieve from the 'pending_reqs' a free pending_req structure to be used.
430 static struct pending_req *alloc_req(struct xen_blkif *blkif)
432 struct pending_req *req = NULL;
435 spin_lock_irqsave(&blkif->pending_free_lock, flags);
436 if (!list_empty(&blkif->pending_free)) {
437 req = list_entry(blkif->pending_free.next, struct pending_req,
439 list_del(&req->free_list);
441 spin_unlock_irqrestore(&blkif->pending_free_lock, flags);
446 * Return the 'pending_req' structure back to the freepool. We also
447 * wake up the thread if it was waiting for a free page.
449 static void free_req(struct xen_blkif *blkif, struct pending_req *req)
454 spin_lock_irqsave(&blkif->pending_free_lock, flags);
455 was_empty = list_empty(&blkif->pending_free);
456 list_add(&req->free_list, &blkif->pending_free);
457 spin_unlock_irqrestore(&blkif->pending_free_lock, flags);
459 wake_up(&blkif->pending_free_wq);
463 * Routines for managing virtual block devices (vbds).
465 static int xen_vbd_translate(struct phys_req *req, struct xen_blkif *blkif,
468 struct xen_vbd *vbd = &blkif->vbd;
471 if ((operation != READ) && vbd->readonly)
474 if (likely(req->nr_sects)) {
475 blkif_sector_t end = req->sector_number + req->nr_sects;
477 if (unlikely(end < req->sector_number))
479 if (unlikely(end > vbd_sz(vbd)))
483 req->dev = vbd->pdevice;
484 req->bdev = vbd->bdev;
491 static void xen_vbd_resize(struct xen_blkif *blkif)
493 struct xen_vbd *vbd = &blkif->vbd;
494 struct xenbus_transaction xbt;
496 struct xenbus_device *dev = xen_blkbk_xenbus(blkif->be);
497 unsigned long long new_size = vbd_sz(vbd);
499 pr_info(DRV_PFX "VBD Resize: Domid: %d, Device: (%d, %d)\n",
500 blkif->domid, MAJOR(vbd->pdevice), MINOR(vbd->pdevice));
501 pr_info(DRV_PFX "VBD Resize: new size %llu\n", new_size);
502 vbd->size = new_size;
504 err = xenbus_transaction_start(&xbt);
506 pr_warn(DRV_PFX "Error starting transaction");
509 err = xenbus_printf(xbt, dev->nodename, "sectors", "%llu",
510 (unsigned long long)vbd_sz(vbd));
512 pr_warn(DRV_PFX "Error writing new size");
516 * Write the current state; we will use this to synchronize
517 * the front-end. If the current state is "connected" the
518 * front-end will get the new size information online.
520 err = xenbus_printf(xbt, dev->nodename, "state", "%d", dev->state);
522 pr_warn(DRV_PFX "Error writing the state");
526 err = xenbus_transaction_end(xbt, 0);
530 pr_warn(DRV_PFX "Error ending transaction");
533 xenbus_transaction_end(xbt, 1);
537 * Notification from the guest OS.
539 static void blkif_notify_work(struct xen_blkif *blkif)
541 blkif->waiting_reqs = 1;
545 irqreturn_t xen_blkif_be_int(int irq, void *dev_id)
547 blkif_notify_work(dev_id);
552 * SCHEDULER FUNCTIONS
555 static void print_stats(struct xen_blkif *blkif)
557 pr_info("xen-blkback (%s): oo %3llu | rd %4llu | wr %4llu | f %4llu"
558 " | ds %4llu | pg: %4u/%4d\n",
559 current->comm, blkif->st_oo_req,
560 blkif->st_rd_req, blkif->st_wr_req,
561 blkif->st_f_req, blkif->st_ds_req,
562 blkif->persistent_gnt_c,
563 xen_blkif_max_pgrants);
564 blkif->st_print = jiffies + msecs_to_jiffies(10 * 1000);
565 blkif->st_rd_req = 0;
566 blkif->st_wr_req = 0;
567 blkif->st_oo_req = 0;
568 blkif->st_ds_req = 0;
571 int xen_blkif_schedule(void *arg)
573 struct xen_blkif *blkif = arg;
574 struct xen_vbd *vbd = &blkif->vbd;
575 unsigned long timeout;
578 xen_blkif_get(blkif);
580 while (!kthread_should_stop()) {
583 if (unlikely(vbd->size != vbd_sz(vbd)))
584 xen_vbd_resize(blkif);
586 timeout = msecs_to_jiffies(LRU_INTERVAL);
588 timeout = wait_event_interruptible_timeout(
590 blkif->waiting_reqs || kthread_should_stop(),
594 timeout = wait_event_interruptible_timeout(
595 blkif->pending_free_wq,
596 !list_empty(&blkif->pending_free) ||
597 kthread_should_stop(),
602 blkif->waiting_reqs = 0;
603 smp_mb(); /* clear flag *before* checking for work */
605 ret = do_block_io_op(blkif);
607 blkif->waiting_reqs = 1;
609 wait_event_interruptible(blkif->shutdown_wq,
610 kthread_should_stop());
613 if (blkif->vbd.feature_gnt_persistent &&
614 time_after(jiffies, blkif->next_lru)) {
615 purge_persistent_gnt(blkif);
616 blkif->next_lru = jiffies + msecs_to_jiffies(LRU_INTERVAL);
619 /* Shrink if we have more than xen_blkif_max_buffer_pages */
620 shrink_free_pagepool(blkif, xen_blkif_max_buffer_pages);
622 if (log_stats && time_after(jiffies, blkif->st_print))
626 /* Since we are shutting down remove all pages from the buffer */
627 shrink_free_pagepool(blkif, 0 /* All */);
629 /* Free all persistent grant pages */
630 if (!RB_EMPTY_ROOT(&blkif->persistent_gnts))
631 free_persistent_gnts(blkif, &blkif->persistent_gnts,
632 blkif->persistent_gnt_c);
634 BUG_ON(!RB_EMPTY_ROOT(&blkif->persistent_gnts));
635 blkif->persistent_gnt_c = 0;
640 blkif->xenblkd = NULL;
641 xen_blkif_put(blkif);
647 * Unmap the grant references, and also remove the M2P over-rides
648 * used in the 'pending_req'.
650 static void xen_blkbk_unmap(struct xen_blkif *blkif,
651 struct grant_page *pages[],
654 struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST];
655 struct page *unmap_pages[BLKIF_MAX_SEGMENTS_PER_REQUEST];
656 unsigned int i, invcount = 0;
659 for (i = 0; i < num; i++) {
660 if (pages[i]->persistent_gnt != NULL) {
661 put_persistent_gnt(blkif, pages[i]->persistent_gnt);
664 if (pages[i]->handle == BLKBACK_INVALID_HANDLE)
666 unmap_pages[invcount] = pages[i]->page;
667 gnttab_set_unmap_op(&unmap[invcount], vaddr(pages[i]->page),
668 GNTMAP_host_map, pages[i]->handle);
669 pages[i]->handle = BLKBACK_INVALID_HANDLE;
670 if (++invcount == BLKIF_MAX_SEGMENTS_PER_REQUEST) {
671 ret = gnttab_unmap_refs(unmap, unmap_pages, invcount);
673 put_free_pages(blkif, unmap_pages, invcount);
678 ret = gnttab_unmap_refs(unmap, unmap_pages, invcount);
680 put_free_pages(blkif, unmap_pages, invcount);
684 static int xen_blkbk_map(struct xen_blkif *blkif,
685 struct grant_page *pages[],
688 struct gnttab_map_grant_ref map[BLKIF_MAX_SEGMENTS_PER_REQUEST];
689 struct page *pages_to_gnt[BLKIF_MAX_SEGMENTS_PER_REQUEST];
690 struct persistent_gnt *persistent_gnt = NULL;
691 phys_addr_t addr = 0;
692 int i, seg_idx, new_map_idx;
695 int last_map = 0, map_until = 0;
696 int use_persistent_gnts;
698 use_persistent_gnts = (blkif->vbd.feature_gnt_persistent);
701 * Fill out preq.nr_sects with proper amount of sectors, and setup
702 * assign map[..] with the PFN of the page in our domain with the
703 * corresponding grant reference for each page.
706 for (i = map_until; i < num; i++) {
709 if (use_persistent_gnts)
710 persistent_gnt = get_persistent_gnt(
714 if (persistent_gnt) {
716 * We are using persistent grants and
717 * the grant is already mapped
719 pages[i]->page = persistent_gnt->page;
720 pages[i]->persistent_gnt = persistent_gnt;
722 if (get_free_page(blkif, &pages[i]->page))
724 addr = vaddr(pages[i]->page);
725 pages_to_gnt[segs_to_map] = pages[i]->page;
726 pages[i]->persistent_gnt = NULL;
727 flags = GNTMAP_host_map;
728 if (!use_persistent_gnts && ro)
729 flags |= GNTMAP_readonly;
730 gnttab_set_map_op(&map[segs_to_map++], addr,
731 flags, pages[i]->gref,
735 if (segs_to_map == BLKIF_MAX_SEGMENTS_PER_REQUEST)
740 ret = gnttab_map_refs(map, pages_to_gnt, segs_to_map);
745 * Now swizzle the MFN in our domain with the MFN from the other domain
746 * so that when we access vaddr(pending_req,i) it has the contents of
747 * the page from the other domain.
749 for (seg_idx = last_map, new_map_idx = 0; seg_idx < map_until; seg_idx++) {
750 if (!pages[seg_idx]->persistent_gnt) {
751 /* This is a newly mapped grant */
752 BUG_ON(new_map_idx >= segs_to_map);
753 if (unlikely(map[new_map_idx].status != 0)) {
754 pr_debug(DRV_PFX "invalid buffer -- could not remap it\n");
755 pages[seg_idx]->handle = BLKBACK_INVALID_HANDLE;
759 pages[seg_idx]->handle = map[new_map_idx].handle;
763 if (use_persistent_gnts &&
764 blkif->persistent_gnt_c < xen_blkif_max_pgrants) {
766 * We are using persistent grants, the grant is
767 * not mapped but we might have room for it.
769 persistent_gnt = kmalloc(sizeof(struct persistent_gnt),
771 if (!persistent_gnt) {
773 * If we don't have enough memory to
774 * allocate the persistent_gnt struct
775 * map this grant non-persistenly
779 persistent_gnt->gnt = map[new_map_idx].ref;
780 persistent_gnt->handle = map[new_map_idx].handle;
781 persistent_gnt->page = pages[seg_idx]->page;
782 if (add_persistent_gnt(blkif,
784 kfree(persistent_gnt);
785 persistent_gnt = NULL;
788 pages[seg_idx]->persistent_gnt = persistent_gnt;
789 pr_debug(DRV_PFX " grant %u added to the tree of persistent grants, using %u/%u\n",
790 persistent_gnt->gnt, blkif->persistent_gnt_c,
791 xen_blkif_max_pgrants);
794 if (use_persistent_gnts && !blkif->vbd.overflow_max_grants) {
795 blkif->vbd.overflow_max_grants = 1;
796 pr_debug(DRV_PFX " domain %u, device %#x is using maximum number of persistent grants\n",
797 blkif->domid, blkif->vbd.handle);
800 * We could not map this grant persistently, so use it as
801 * a non-persistent grant.
807 last_map = map_until;
808 if (map_until != num)
814 pr_alert(DRV_PFX "%s: out of memory\n", __func__);
815 put_free_pages(blkif, pages_to_gnt, segs_to_map);
819 static int xen_blkbk_map_seg(struct pending_req *pending_req)
823 rc = xen_blkbk_map(pending_req->blkif, pending_req->segments,
824 pending_req->nr_pages,
825 (pending_req->operation != BLKIF_OP_READ));
830 static int xen_blkbk_parse_indirect(struct blkif_request *req,
831 struct pending_req *pending_req,
832 struct seg_buf seg[],
833 struct phys_req *preq)
835 struct grant_page **pages = pending_req->indirect_pages;
836 struct xen_blkif *blkif = pending_req->blkif;
837 int indirect_grefs, rc, n, nseg, i;
838 struct blkif_request_segment_aligned *segments = NULL;
840 nseg = pending_req->nr_pages;
841 indirect_grefs = INDIRECT_PAGES(nseg);
842 BUG_ON(indirect_grefs > BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST);
844 for (i = 0; i < indirect_grefs; i++)
845 pages[i]->gref = req->u.indirect.indirect_grefs[i];
847 rc = xen_blkbk_map(blkif, pages, indirect_grefs, true);
851 for (n = 0, i = 0; n < nseg; n++) {
852 if ((n % SEGS_PER_INDIRECT_FRAME) == 0) {
853 /* Map indirect segments */
855 kunmap_atomic(segments);
856 segments = kmap_atomic(pages[n/SEGS_PER_INDIRECT_FRAME]->page);
858 i = n % SEGS_PER_INDIRECT_FRAME;
859 pending_req->segments[n]->gref = segments[i].gref;
860 seg[n].nsec = segments[i].last_sect -
861 segments[i].first_sect + 1;
862 seg[n].offset = (segments[i].first_sect << 9);
863 if ((segments[i].last_sect >= (PAGE_SIZE >> 9)) ||
864 (segments[i].last_sect < segments[i].first_sect)) {
868 preq->nr_sects += seg[n].nsec;
873 kunmap_atomic(segments);
874 xen_blkbk_unmap(blkif, pages, indirect_grefs);
878 static int dispatch_discard_io(struct xen_blkif *blkif,
879 struct blkif_request *req)
882 int status = BLKIF_RSP_OKAY;
883 struct block_device *bdev = blkif->vbd.bdev;
884 unsigned long secure;
885 struct phys_req preq;
887 xen_blkif_get(blkif);
889 preq.sector_number = req->u.discard.sector_number;
890 preq.nr_sects = req->u.discard.nr_sectors;
892 err = xen_vbd_translate(&preq, blkif, WRITE);
894 pr_warn(DRV_PFX "access denied: DISCARD [%llu->%llu] on dev=%04x\n",
896 preq.sector_number + preq.nr_sects, blkif->vbd.pdevice);
901 secure = (blkif->vbd.discard_secure &&
902 (req->u.discard.flag & BLKIF_DISCARD_SECURE)) ?
903 BLKDEV_DISCARD_SECURE : 0;
905 err = blkdev_issue_discard(bdev, req->u.discard.sector_number,
906 req->u.discard.nr_sectors,
909 if (err == -EOPNOTSUPP) {
910 pr_debug(DRV_PFX "discard op failed, not supported\n");
911 status = BLKIF_RSP_EOPNOTSUPP;
913 status = BLKIF_RSP_ERROR;
915 make_response(blkif, req->u.discard.id, req->operation, status);
916 xen_blkif_put(blkif);
920 static int dispatch_other_io(struct xen_blkif *blkif,
921 struct blkif_request *req,
922 struct pending_req *pending_req)
924 free_req(blkif, pending_req);
925 make_response(blkif, req->u.other.id, req->operation,
926 BLKIF_RSP_EOPNOTSUPP);
930 static void xen_blk_drain_io(struct xen_blkif *blkif)
932 atomic_set(&blkif->drain, 1);
934 /* The initial value is one, and one refcnt taken at the
935 * start of the xen_blkif_schedule thread. */
936 if (atomic_read(&blkif->refcnt) <= 2)
938 wait_for_completion_interruptible_timeout(
939 &blkif->drain_complete, HZ);
941 if (!atomic_read(&blkif->drain))
943 } while (!kthread_should_stop());
944 atomic_set(&blkif->drain, 0);
948 * Completion callback on the bio's. Called as bh->b_end_io()
951 static void __end_block_io_op(struct pending_req *pending_req, int error)
953 /* An error fails the entire request. */
954 if ((pending_req->operation == BLKIF_OP_FLUSH_DISKCACHE) &&
955 (error == -EOPNOTSUPP)) {
956 pr_debug(DRV_PFX "flush diskcache op failed, not supported\n");
957 xen_blkbk_flush_diskcache(XBT_NIL, pending_req->blkif->be, 0);
958 pending_req->status = BLKIF_RSP_EOPNOTSUPP;
959 } else if ((pending_req->operation == BLKIF_OP_WRITE_BARRIER) &&
960 (error == -EOPNOTSUPP)) {
961 pr_debug(DRV_PFX "write barrier op failed, not supported\n");
962 xen_blkbk_barrier(XBT_NIL, pending_req->blkif->be, 0);
963 pending_req->status = BLKIF_RSP_EOPNOTSUPP;
965 pr_debug(DRV_PFX "Buffer not up-to-date at end of operation,"
966 " error=%d\n", error);
967 pending_req->status = BLKIF_RSP_ERROR;
971 * If all of the bio's have completed it is time to unmap
972 * the grant references associated with 'request' and provide
973 * the proper response on the ring.
975 if (atomic_dec_and_test(&pending_req->pendcnt)) {
976 xen_blkbk_unmap(pending_req->blkif,
977 pending_req->segments,
978 pending_req->nr_pages);
979 make_response(pending_req->blkif, pending_req->id,
980 pending_req->operation, pending_req->status);
981 xen_blkif_put(pending_req->blkif);
982 if (atomic_read(&pending_req->blkif->refcnt) <= 2) {
983 if (atomic_read(&pending_req->blkif->drain))
984 complete(&pending_req->blkif->drain_complete);
986 free_req(pending_req->blkif, pending_req);
993 static void end_block_io_op(struct bio *bio, int error)
995 __end_block_io_op(bio->bi_private, error);
1002 * Function to copy the from the ring buffer the 'struct blkif_request'
1003 * (which has the sectors we want, number of them, grant references, etc),
1004 * and transmute it to the block API to hand it over to the proper block disk.
1007 __do_block_io_op(struct xen_blkif *blkif)
1009 union blkif_back_rings *blk_rings = &blkif->blk_rings;
1010 struct blkif_request req;
1011 struct pending_req *pending_req;
1015 rc = blk_rings->common.req_cons;
1016 rp = blk_rings->common.sring->req_prod;
1017 rmb(); /* Ensure we see queued requests up to 'rp'. */
1019 if (RING_REQUEST_PROD_OVERFLOW(&blk_rings->common, rp)) {
1020 rc = blk_rings->common.rsp_prod_pvt;
1021 pr_warn(DRV_PFX "Frontend provided bogus ring requests (%d - %d = %d). Halting ring processing on dev=%04x\n",
1022 rp, rc, rp - rc, blkif->vbd.pdevice);
1027 if (RING_REQUEST_CONS_OVERFLOW(&blk_rings->common, rc))
1030 if (kthread_should_stop()) {
1035 pending_req = alloc_req(blkif);
1036 if (NULL == pending_req) {
1042 switch (blkif->blk_protocol) {
1043 case BLKIF_PROTOCOL_NATIVE:
1044 memcpy(&req, RING_GET_REQUEST(&blk_rings->native, rc), sizeof(req));
1046 case BLKIF_PROTOCOL_X86_32:
1047 blkif_get_x86_32_req(&req, RING_GET_REQUEST(&blk_rings->x86_32, rc));
1049 case BLKIF_PROTOCOL_X86_64:
1050 blkif_get_x86_64_req(&req, RING_GET_REQUEST(&blk_rings->x86_64, rc));
1055 blk_rings->common.req_cons = ++rc; /* before make_response() */
1057 /* Apply all sanity checks to /private copy/ of request. */
1060 switch (req.operation) {
1062 case BLKIF_OP_WRITE:
1063 case BLKIF_OP_WRITE_BARRIER:
1064 case BLKIF_OP_FLUSH_DISKCACHE:
1065 case BLKIF_OP_INDIRECT:
1066 if (dispatch_rw_block_io(blkif, &req, pending_req))
1069 case BLKIF_OP_DISCARD:
1070 free_req(blkif, pending_req);
1071 if (dispatch_discard_io(blkif, &req))
1075 if (dispatch_other_io(blkif, &req, pending_req))
1080 /* Yield point for this unbounded loop. */
1088 do_block_io_op(struct xen_blkif *blkif)
1090 union blkif_back_rings *blk_rings = &blkif->blk_rings;
1094 more_to_do = __do_block_io_op(blkif);
1098 RING_FINAL_CHECK_FOR_REQUESTS(&blk_rings->common, more_to_do);
1099 } while (more_to_do);
1104 * Transmutation of the 'struct blkif_request' to a proper 'struct bio'
1105 * and call the 'submit_bio' to pass it to the underlying storage.
1107 static int dispatch_rw_block_io(struct xen_blkif *blkif,
1108 struct blkif_request *req,
1109 struct pending_req *pending_req)
1111 struct phys_req preq;
1112 struct seg_buf *seg = pending_req->seg;
1114 struct bio *bio = NULL;
1115 struct bio **biolist = pending_req->biolist;
1118 struct blk_plug plug;
1120 struct grant_page **pages = pending_req->segments;
1121 unsigned short req_operation;
1123 req_operation = req->operation == BLKIF_OP_INDIRECT ?
1124 req->u.indirect.indirect_op : req->operation;
1125 if ((req->operation == BLKIF_OP_INDIRECT) &&
1126 (req_operation != BLKIF_OP_READ) &&
1127 (req_operation != BLKIF_OP_WRITE)) {
1128 pr_debug(DRV_PFX "Invalid indirect operation (%u)\n",
1133 switch (req_operation) {
1138 case BLKIF_OP_WRITE:
1140 operation = WRITE_ODIRECT;
1142 case BLKIF_OP_WRITE_BARRIER:
1144 case BLKIF_OP_FLUSH_DISKCACHE:
1146 operation = WRITE_FLUSH;
1149 operation = 0; /* make gcc happy */
1154 /* Check that the number of segments is sane. */
1155 nseg = req->operation == BLKIF_OP_INDIRECT ?
1156 req->u.indirect.nr_segments : req->u.rw.nr_segments;
1158 if (unlikely(nseg == 0 && operation != WRITE_FLUSH) ||
1159 unlikely((req->operation != BLKIF_OP_INDIRECT) &&
1160 (nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST)) ||
1161 unlikely((req->operation == BLKIF_OP_INDIRECT) &&
1162 (nseg > MAX_INDIRECT_SEGMENTS))) {
1163 pr_debug(DRV_PFX "Bad number of segments in request (%d)\n",
1165 /* Haven't submitted any bio's yet. */
1171 pending_req->blkif = blkif;
1172 pending_req->id = req->u.rw.id;
1173 pending_req->operation = req_operation;
1174 pending_req->status = BLKIF_RSP_OKAY;
1175 pending_req->nr_pages = nseg;
1177 if (req->operation != BLKIF_OP_INDIRECT) {
1178 preq.dev = req->u.rw.handle;
1179 preq.sector_number = req->u.rw.sector_number;
1180 for (i = 0; i < nseg; i++) {
1181 pages[i]->gref = req->u.rw.seg[i].gref;
1182 seg[i].nsec = req->u.rw.seg[i].last_sect -
1183 req->u.rw.seg[i].first_sect + 1;
1184 seg[i].offset = (req->u.rw.seg[i].first_sect << 9);
1185 if ((req->u.rw.seg[i].last_sect >= (PAGE_SIZE >> 9)) ||
1186 (req->u.rw.seg[i].last_sect <
1187 req->u.rw.seg[i].first_sect))
1189 preq.nr_sects += seg[i].nsec;
1192 preq.dev = req->u.indirect.handle;
1193 preq.sector_number = req->u.indirect.sector_number;
1194 if (xen_blkbk_parse_indirect(req, pending_req, seg, &preq))
1198 if (xen_vbd_translate(&preq, blkif, operation) != 0) {
1199 pr_debug(DRV_PFX "access denied: %s of [%llu,%llu] on dev=%04x\n",
1200 operation == READ ? "read" : "write",
1202 preq.sector_number + preq.nr_sects,
1203 blkif->vbd.pdevice);
1208 * This check _MUST_ be done after xen_vbd_translate as the preq.bdev
1211 for (i = 0; i < nseg; i++) {
1212 if (((int)preq.sector_number|(int)seg[i].nsec) &
1213 ((bdev_logical_block_size(preq.bdev) >> 9) - 1)) {
1214 pr_debug(DRV_PFX "Misaligned I/O request from domain %d",
1220 /* Wait on all outstanding I/O's and once that has been completed
1221 * issue the WRITE_FLUSH.
1224 xen_blk_drain_io(pending_req->blkif);
1227 * If we have failed at this point, we need to undo the M2P override,
1228 * set gnttab_set_unmap_op on all of the grant references and perform
1229 * the hypercall to unmap the grants - that is all done in
1232 if (xen_blkbk_map_seg(pending_req))
1236 * This corresponding xen_blkif_put is done in __end_block_io_op, or
1237 * below (in "!bio") if we are handling a BLKIF_OP_DISCARD.
1239 xen_blkif_get(blkif);
1241 for (i = 0; i < nseg; i++) {
1242 while ((bio == NULL) ||
1246 seg[i].offset) == 0)) {
1248 int nr_iovecs = min_t(int, (nseg-i), BIO_MAX_PAGES);
1249 bio = bio_alloc(GFP_KERNEL, nr_iovecs);
1250 if (unlikely(bio == NULL))
1253 biolist[nbio++] = bio;
1254 bio->bi_bdev = preq.bdev;
1255 bio->bi_private = pending_req;
1256 bio->bi_end_io = end_block_io_op;
1257 bio->bi_sector = preq.sector_number;
1260 preq.sector_number += seg[i].nsec;
1263 /* This will be hit if the operation was a flush or discard. */
1265 BUG_ON(operation != WRITE_FLUSH);
1267 bio = bio_alloc(GFP_KERNEL, 0);
1268 if (unlikely(bio == NULL))
1271 biolist[nbio++] = bio;
1272 bio->bi_bdev = preq.bdev;
1273 bio->bi_private = pending_req;
1274 bio->bi_end_io = end_block_io_op;
1277 atomic_set(&pending_req->pendcnt, nbio);
1278 blk_start_plug(&plug);
1280 for (i = 0; i < nbio; i++)
1281 submit_bio(operation, biolist[i]);
1283 /* Let the I/Os go.. */
1284 blk_finish_plug(&plug);
1286 if (operation == READ)
1287 blkif->st_rd_sect += preq.nr_sects;
1288 else if (operation & WRITE)
1289 blkif->st_wr_sect += preq.nr_sects;
1294 xen_blkbk_unmap(blkif, pending_req->segments,
1295 pending_req->nr_pages);
1297 /* Haven't submitted any bio's yet. */
1298 make_response(blkif, req->u.rw.id, req_operation, BLKIF_RSP_ERROR);
1299 free_req(blkif, pending_req);
1300 msleep(1); /* back off a bit */
1304 for (i = 0; i < nbio; i++)
1305 bio_put(biolist[i]);
1306 atomic_set(&pending_req->pendcnt, 1);
1307 __end_block_io_op(pending_req, -EINVAL);
1308 msleep(1); /* back off a bit */
1315 * Put a response on the ring on how the operation fared.
1317 static void make_response(struct xen_blkif *blkif, u64 id,
1318 unsigned short op, int st)
1320 struct blkif_response resp;
1321 unsigned long flags;
1322 union blkif_back_rings *blk_rings = &blkif->blk_rings;
1326 resp.operation = op;
1329 spin_lock_irqsave(&blkif->blk_ring_lock, flags);
1330 /* Place on the response ring for the relevant domain. */
1331 switch (blkif->blk_protocol) {
1332 case BLKIF_PROTOCOL_NATIVE:
1333 memcpy(RING_GET_RESPONSE(&blk_rings->native, blk_rings->native.rsp_prod_pvt),
1334 &resp, sizeof(resp));
1336 case BLKIF_PROTOCOL_X86_32:
1337 memcpy(RING_GET_RESPONSE(&blk_rings->x86_32, blk_rings->x86_32.rsp_prod_pvt),
1338 &resp, sizeof(resp));
1340 case BLKIF_PROTOCOL_X86_64:
1341 memcpy(RING_GET_RESPONSE(&blk_rings->x86_64, blk_rings->x86_64.rsp_prod_pvt),
1342 &resp, sizeof(resp));
1347 blk_rings->common.rsp_prod_pvt++;
1348 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blk_rings->common, notify);
1349 spin_unlock_irqrestore(&blkif->blk_ring_lock, flags);
1351 notify_remote_via_irq(blkif->irq);
1354 static int __init xen_blkif_init(void)
1361 rc = xen_blkif_interface_init();
1365 rc = xen_blkif_xenbus_init();
1373 module_init(xen_blkif_init);
1375 MODULE_LICENSE("Dual BSD/GPL");
1376 MODULE_ALIAS("xen-backend:vbd");