Commit | Line | Data |
---|---|---|
4d05a28d | 1 | /****************************************************************************** |
4d05a28d KRW |
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: | |
a1397fa3 | 7 | * drivers/block/xen-blkfront.c |
4d05a28d KRW |
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 | ||
77387b82 TC |
37 | #define pr_fmt(fmt) "xen-blkback: " fmt |
38 | ||
4d05a28d KRW |
39 | #include <linux/spinlock.h> |
40 | #include <linux/kthread.h> | |
41 | #include <linux/list.h> | |
42 | #include <linux/delay.h> | |
88122933 | 43 | #include <linux/freezer.h> |
0a8704a5 | 44 | #include <linux/bitmap.h> |
afd91d07 | 45 | |
88122933 JF |
46 | #include <xen/events.h> |
47 | #include <xen/page.h> | |
e79affc3 | 48 | #include <xen/xen.h> |
88122933 JF |
49 | #include <asm/xen/hypervisor.h> |
50 | #include <asm/xen/hypercall.h> | |
087ffecd | 51 | #include <xen/balloon.h> |
c43cf3ea | 52 | #include <xen/grant_table.h> |
4d05a28d KRW |
53 | #include "common.h" |
54 | ||
c6cc142d RPM |
55 | /* |
56 | * Maximum number of unused free pages to keep in the internal buffer. | |
57 | * Setting this to a value too low will reduce memory used in each backend, | |
58 | * but can have a performance penalty. | |
59 | * | |
60 | * A sane value is xen_blkif_reqs * BLKIF_MAX_SEGMENTS_PER_REQUEST, but can | |
61 | * be set to a lower value that might degrade performance on some intensive | |
62 | * IO workloads. | |
63 | */ | |
64 | ||
402b27f9 | 65 | static int xen_blkif_max_buffer_pages = 1024; |
c6cc142d RPM |
66 | module_param_named(max_buffer_pages, xen_blkif_max_buffer_pages, int, 0644); |
67 | MODULE_PARM_DESC(max_buffer_pages, | |
68 | "Maximum number of free pages to keep in each block backend buffer"); | |
69 | ||
3f3aad5e RPM |
70 | /* |
71 | * Maximum number of grants to map persistently in blkback. For maximum | |
72 | * performance this should be the total numbers of grants that can be used | |
73 | * to fill the ring, but since this might become too high, specially with | |
74 | * the use of indirect descriptors, we set it to a value that provides good | |
75 | * performance without using too much memory. | |
76 | * | |
77 | * When the list of persistent grants is full we clean it up using a LRU | |
78 | * algorithm. | |
79 | */ | |
80 | ||
402b27f9 | 81 | static int xen_blkif_max_pgrants = 1056; |
3f3aad5e RPM |
82 | module_param_named(max_persistent_grants, xen_blkif_max_pgrants, int, 0644); |
83 | MODULE_PARM_DESC(max_persistent_grants, | |
84 | "Maximum number of grants to map persistently"); | |
85 | ||
86 | /* | |
87 | * The LRU mechanism to clean the lists of persistent grants needs to | |
88 | * be executed periodically. The time interval between consecutive executions | |
89 | * of the purge mechanism is set in ms. | |
90 | */ | |
91 | #define LRU_INTERVAL 100 | |
92 | ||
93 | /* | |
94 | * When the persistent grants list is full we will remove unused grants | |
95 | * from the list. The percent number of grants to be removed at each LRU | |
96 | * execution. | |
97 | */ | |
98 | #define LRU_PERCENT_CLEAN 5 | |
99 | ||
4d05a28d | 100 | /* Run-time switchable: /sys/module/blkback/parameters/ */ |
2e9977c2 | 101 | static unsigned int log_stats; |
4d05a28d | 102 | module_param(log_stats, int, 0644); |
4d05a28d | 103 | |
4d05a28d KRW |
104 | #define BLKBACK_INVALID_HANDLE (~0) |
105 | ||
ff4b156f | 106 | /* Number of free pages to remove on each call to gnttab_free_pages */ |
c6cc142d RPM |
107 | #define NUM_BATCH_FREE_PAGES 10 |
108 | ||
c6cc142d RPM |
109 | static inline int get_free_page(struct xen_blkif *blkif, struct page **page) |
110 | { | |
111 | unsigned long flags; | |
112 | ||
113 | spin_lock_irqsave(&blkif->free_pages_lock, flags); | |
114 | if (list_empty(&blkif->free_pages)) { | |
115 | BUG_ON(blkif->free_pages_num != 0); | |
116 | spin_unlock_irqrestore(&blkif->free_pages_lock, flags); | |
ff4b156f | 117 | return gnttab_alloc_pages(1, page); |
c6cc142d RPM |
118 | } |
119 | BUG_ON(blkif->free_pages_num == 0); | |
120 | page[0] = list_first_entry(&blkif->free_pages, struct page, lru); | |
121 | list_del(&page[0]->lru); | |
122 | blkif->free_pages_num--; | |
123 | spin_unlock_irqrestore(&blkif->free_pages_lock, flags); | |
efe08a3e | 124 | |
c6cc142d RPM |
125 | return 0; |
126 | } | |
127 | ||
128 | static inline void put_free_pages(struct xen_blkif *blkif, struct page **page, | |
129 | int num) | |
4d05a28d | 130 | { |
c6cc142d RPM |
131 | unsigned long flags; |
132 | int i; | |
133 | ||
134 | spin_lock_irqsave(&blkif->free_pages_lock, flags); | |
135 | for (i = 0; i < num; i++) | |
136 | list_add(&page[i]->lru, &blkif->free_pages); | |
137 | blkif->free_pages_num += num; | |
138 | spin_unlock_irqrestore(&blkif->free_pages_lock, flags); | |
139 | } | |
140 | ||
141 | static inline void shrink_free_pagepool(struct xen_blkif *blkif, int num) | |
142 | { | |
143 | /* Remove requested pages in batches of NUM_BATCH_FREE_PAGES */ | |
144 | struct page *page[NUM_BATCH_FREE_PAGES]; | |
145 | unsigned int num_pages = 0; | |
146 | unsigned long flags; | |
147 | ||
148 | spin_lock_irqsave(&blkif->free_pages_lock, flags); | |
149 | while (blkif->free_pages_num > num) { | |
150 | BUG_ON(list_empty(&blkif->free_pages)); | |
151 | page[num_pages] = list_first_entry(&blkif->free_pages, | |
152 | struct page, lru); | |
153 | list_del(&page[num_pages]->lru); | |
154 | blkif->free_pages_num--; | |
155 | if (++num_pages == NUM_BATCH_FREE_PAGES) { | |
156 | spin_unlock_irqrestore(&blkif->free_pages_lock, flags); | |
ff4b156f | 157 | gnttab_free_pages(num_pages, page); |
c6cc142d RPM |
158 | spin_lock_irqsave(&blkif->free_pages_lock, flags); |
159 | num_pages = 0; | |
160 | } | |
161 | } | |
162 | spin_unlock_irqrestore(&blkif->free_pages_lock, flags); | |
163 | if (num_pages != 0) | |
ff4b156f | 164 | gnttab_free_pages(num_pages, page); |
4d05a28d KRW |
165 | } |
166 | ||
c6cc142d RPM |
167 | #define vaddr(page) ((unsigned long)pfn_to_kaddr(page_to_pfn(page))) |
168 | ||
30fd1502 KRW |
169 | static int do_block_io_op(struct xen_blkif *blkif); |
170 | static int dispatch_rw_block_io(struct xen_blkif *blkif, | |
fc53bf75 KRW |
171 | struct blkif_request *req, |
172 | struct pending_req *pending_req); | |
30fd1502 | 173 | static void make_response(struct xen_blkif *blkif, u64 id, |
4d05a28d KRW |
174 | unsigned short op, int st); |
175 | ||
7dc34117 RPM |
176 | #define foreach_grant_safe(pos, n, rbtree, node) \ |
177 | for ((pos) = container_of(rb_first((rbtree)), typeof(*(pos)), node), \ | |
217fd5e7 | 178 | (n) = (&(pos)->node != NULL) ? rb_next(&(pos)->node) : NULL; \ |
0a8704a5 | 179 | &(pos)->node != NULL; \ |
7dc34117 RPM |
180 | (pos) = container_of(n, typeof(*(pos)), node), \ |
181 | (n) = (&(pos)->node != NULL) ? rb_next(&(pos)->node) : NULL) | |
0a8704a5 RPM |
182 | |
183 | ||
3f3aad5e RPM |
184 | /* |
185 | * We don't need locking around the persistent grant helpers | |
186 | * because blkback uses a single-thread for each backed, so we | |
187 | * can be sure that this functions will never be called recursively. | |
188 | * | |
189 | * The only exception to that is put_persistent_grant, that can be called | |
190 | * from interrupt context (by xen_blkbk_unmap), so we have to use atomic | |
191 | * bit operations to modify the flags of a persistent grant and to count | |
192 | * the number of used grants. | |
193 | */ | |
194 | static int add_persistent_gnt(struct xen_blkif *blkif, | |
0a8704a5 RPM |
195 | struct persistent_gnt *persistent_gnt) |
196 | { | |
3f3aad5e | 197 | struct rb_node **new = NULL, *parent = NULL; |
0a8704a5 RPM |
198 | struct persistent_gnt *this; |
199 | ||
3f3aad5e RPM |
200 | if (blkif->persistent_gnt_c >= xen_blkif_max_pgrants) { |
201 | if (!blkif->vbd.overflow_max_grants) | |
202 | blkif->vbd.overflow_max_grants = 1; | |
203 | return -EBUSY; | |
204 | } | |
0a8704a5 | 205 | /* Figure out where to put new node */ |
3f3aad5e | 206 | new = &blkif->persistent_gnts.rb_node; |
0a8704a5 RPM |
207 | while (*new) { |
208 | this = container_of(*new, struct persistent_gnt, node); | |
209 | ||
210 | parent = *new; | |
211 | if (persistent_gnt->gnt < this->gnt) | |
212 | new = &((*new)->rb_left); | |
213 | else if (persistent_gnt->gnt > this->gnt) | |
214 | new = &((*new)->rb_right); | |
215 | else { | |
77387b82 | 216 | pr_alert_ratelimited("trying to add a gref that's already in the tree\n"); |
c6cc142d | 217 | return -EINVAL; |
0a8704a5 RPM |
218 | } |
219 | } | |
220 | ||
3f3aad5e RPM |
221 | bitmap_zero(persistent_gnt->flags, PERSISTENT_GNT_FLAGS_SIZE); |
222 | set_bit(PERSISTENT_GNT_ACTIVE, persistent_gnt->flags); | |
0a8704a5 RPM |
223 | /* Add new node and rebalance tree. */ |
224 | rb_link_node(&(persistent_gnt->node), parent, new); | |
3f3aad5e RPM |
225 | rb_insert_color(&(persistent_gnt->node), &blkif->persistent_gnts); |
226 | blkif->persistent_gnt_c++; | |
227 | atomic_inc(&blkif->persistent_gnt_in_use); | |
c6cc142d | 228 | return 0; |
0a8704a5 RPM |
229 | } |
230 | ||
3f3aad5e | 231 | static struct persistent_gnt *get_persistent_gnt(struct xen_blkif *blkif, |
0a8704a5 RPM |
232 | grant_ref_t gref) |
233 | { | |
234 | struct persistent_gnt *data; | |
3f3aad5e | 235 | struct rb_node *node = NULL; |
0a8704a5 | 236 | |
3f3aad5e | 237 | node = blkif->persistent_gnts.rb_node; |
0a8704a5 RPM |
238 | while (node) { |
239 | data = container_of(node, struct persistent_gnt, node); | |
240 | ||
241 | if (gref < data->gnt) | |
242 | node = node->rb_left; | |
243 | else if (gref > data->gnt) | |
244 | node = node->rb_right; | |
3f3aad5e RPM |
245 | else { |
246 | if(test_bit(PERSISTENT_GNT_ACTIVE, data->flags)) { | |
77387b82 | 247 | pr_alert_ratelimited("requesting a grant already in use\n"); |
3f3aad5e RPM |
248 | return NULL; |
249 | } | |
250 | set_bit(PERSISTENT_GNT_ACTIVE, data->flags); | |
251 | atomic_inc(&blkif->persistent_gnt_in_use); | |
0a8704a5 | 252 | return data; |
3f3aad5e | 253 | } |
0a8704a5 RPM |
254 | } |
255 | return NULL; | |
256 | } | |
257 | ||
3f3aad5e RPM |
258 | static void put_persistent_gnt(struct xen_blkif *blkif, |
259 | struct persistent_gnt *persistent_gnt) | |
260 | { | |
261 | if(!test_bit(PERSISTENT_GNT_ACTIVE, persistent_gnt->flags)) | |
77387b82 | 262 | pr_alert_ratelimited("freeing a grant already unused\n"); |
3f3aad5e RPM |
263 | set_bit(PERSISTENT_GNT_WAS_ACTIVE, persistent_gnt->flags); |
264 | clear_bit(PERSISTENT_GNT_ACTIVE, persistent_gnt->flags); | |
265 | atomic_dec(&blkif->persistent_gnt_in_use); | |
266 | } | |
267 | ||
c6cc142d RPM |
268 | static void free_persistent_gnts(struct xen_blkif *blkif, struct rb_root *root, |
269 | unsigned int num) | |
4d4f270f RPM |
270 | { |
271 | struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
272 | struct page *pages[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
273 | struct persistent_gnt *persistent_gnt; | |
7dc34117 | 274 | struct rb_node *n; |
4d4f270f | 275 | int segs_to_unmap = 0; |
c43cf3ea | 276 | struct gntab_unmap_queue_data unmap_data; |
c43cf3ea | 277 | |
c43cf3ea JH |
278 | unmap_data.pages = pages; |
279 | unmap_data.unmap_ops = unmap; | |
280 | unmap_data.kunmap_ops = NULL; | |
4d4f270f | 281 | |
7dc34117 | 282 | foreach_grant_safe(persistent_gnt, n, root, node) { |
4d4f270f RPM |
283 | BUG_ON(persistent_gnt->handle == |
284 | BLKBACK_INVALID_HANDLE); | |
285 | gnttab_set_unmap_op(&unmap[segs_to_unmap], | |
286 | (unsigned long) pfn_to_kaddr(page_to_pfn( | |
287 | persistent_gnt->page)), | |
288 | GNTMAP_host_map, | |
289 | persistent_gnt->handle); | |
290 | ||
291 | pages[segs_to_unmap] = persistent_gnt->page; | |
4d4f270f RPM |
292 | |
293 | if (++segs_to_unmap == BLKIF_MAX_SEGMENTS_PER_REQUEST || | |
294 | !rb_next(&persistent_gnt->node)) { | |
c43cf3ea JH |
295 | |
296 | unmap_data.count = segs_to_unmap; | |
b44166cd | 297 | BUG_ON(gnttab_unmap_refs_sync(&unmap_data)); |
c43cf3ea | 298 | |
c6cc142d | 299 | put_free_pages(blkif, pages, segs_to_unmap); |
4d4f270f RPM |
300 | segs_to_unmap = 0; |
301 | } | |
7dc34117 RPM |
302 | |
303 | rb_erase(&persistent_gnt->node, root); | |
304 | kfree(persistent_gnt); | |
305 | num--; | |
4d4f270f RPM |
306 | } |
307 | BUG_ON(num != 0); | |
308 | } | |
309 | ||
abb97b8c | 310 | void xen_blkbk_unmap_purged_grants(struct work_struct *work) |
3f3aad5e RPM |
311 | { |
312 | struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
313 | struct page *pages[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
314 | struct persistent_gnt *persistent_gnt; | |
325d73bf | 315 | int segs_to_unmap = 0; |
3f3aad5e | 316 | struct xen_blkif *blkif = container_of(work, typeof(*blkif), persistent_purge_work); |
325d73bf | 317 | struct gntab_unmap_queue_data unmap_data; |
325d73bf | 318 | |
325d73bf BL |
319 | unmap_data.pages = pages; |
320 | unmap_data.unmap_ops = unmap; | |
321 | unmap_data.kunmap_ops = NULL; | |
3f3aad5e RPM |
322 | |
323 | while(!list_empty(&blkif->persistent_purge_list)) { | |
324 | persistent_gnt = list_first_entry(&blkif->persistent_purge_list, | |
325 | struct persistent_gnt, | |
326 | remove_node); | |
327 | list_del(&persistent_gnt->remove_node); | |
328 | ||
329 | gnttab_set_unmap_op(&unmap[segs_to_unmap], | |
330 | vaddr(persistent_gnt->page), | |
331 | GNTMAP_host_map, | |
332 | persistent_gnt->handle); | |
333 | ||
334 | pages[segs_to_unmap] = persistent_gnt->page; | |
335 | ||
336 | if (++segs_to_unmap == BLKIF_MAX_SEGMENTS_PER_REQUEST) { | |
325d73bf | 337 | unmap_data.count = segs_to_unmap; |
b44166cd | 338 | BUG_ON(gnttab_unmap_refs_sync(&unmap_data)); |
3f3aad5e RPM |
339 | put_free_pages(blkif, pages, segs_to_unmap); |
340 | segs_to_unmap = 0; | |
341 | } | |
342 | kfree(persistent_gnt); | |
343 | } | |
344 | if (segs_to_unmap > 0) { | |
325d73bf | 345 | unmap_data.count = segs_to_unmap; |
b44166cd | 346 | BUG_ON(gnttab_unmap_refs_sync(&unmap_data)); |
3f3aad5e RPM |
347 | put_free_pages(blkif, pages, segs_to_unmap); |
348 | } | |
349 | } | |
350 | ||
351 | static void purge_persistent_gnt(struct xen_blkif *blkif) | |
352 | { | |
353 | struct persistent_gnt *persistent_gnt; | |
354 | struct rb_node *n; | |
355 | unsigned int num_clean, total; | |
2d910543 | 356 | bool scan_used = false, clean_used = false; |
3f3aad5e RPM |
357 | struct rb_root *root; |
358 | ||
359 | if (blkif->persistent_gnt_c < xen_blkif_max_pgrants || | |
360 | (blkif->persistent_gnt_c == xen_blkif_max_pgrants && | |
361 | !blkif->vbd.overflow_max_grants)) { | |
362 | return; | |
363 | } | |
364 | ||
365 | if (work_pending(&blkif->persistent_purge_work)) { | |
77387b82 | 366 | pr_alert_ratelimited("Scheduled work from previous purge is still pending, cannot purge list\n"); |
3f3aad5e RPM |
367 | return; |
368 | } | |
369 | ||
370 | num_clean = (xen_blkif_max_pgrants / 100) * LRU_PERCENT_CLEAN; | |
371 | num_clean = blkif->persistent_gnt_c - xen_blkif_max_pgrants + num_clean; | |
372 | num_clean = min(blkif->persistent_gnt_c, num_clean); | |
2d910543 RPM |
373 | if ((num_clean == 0) || |
374 | (num_clean > (blkif->persistent_gnt_c - atomic_read(&blkif->persistent_gnt_in_use)))) | |
3f3aad5e RPM |
375 | return; |
376 | ||
377 | /* | |
378 | * At this point, we can assure that there will be no calls | |
379 | * to get_persistent_grant (because we are executing this code from | |
380 | * xen_blkif_schedule), there can only be calls to put_persistent_gnt, | |
381 | * which means that the number of currently used grants will go down, | |
382 | * but never up, so we will always be able to remove the requested | |
383 | * number of grants. | |
384 | */ | |
385 | ||
386 | total = num_clean; | |
387 | ||
77387b82 | 388 | pr_debug("Going to purge %u persistent grants\n", num_clean); |
3f3aad5e | 389 | |
ef753411 | 390 | BUG_ON(!list_empty(&blkif->persistent_purge_list)); |
3f3aad5e RPM |
391 | root = &blkif->persistent_gnts; |
392 | purge_list: | |
393 | foreach_grant_safe(persistent_gnt, n, root, node) { | |
394 | BUG_ON(persistent_gnt->handle == | |
395 | BLKBACK_INVALID_HANDLE); | |
396 | ||
2d910543 RPM |
397 | if (clean_used) { |
398 | clear_bit(PERSISTENT_GNT_WAS_ACTIVE, persistent_gnt->flags); | |
399 | continue; | |
400 | } | |
401 | ||
3f3aad5e RPM |
402 | if (test_bit(PERSISTENT_GNT_ACTIVE, persistent_gnt->flags)) |
403 | continue; | |
404 | if (!scan_used && | |
405 | (test_bit(PERSISTENT_GNT_WAS_ACTIVE, persistent_gnt->flags))) | |
406 | continue; | |
407 | ||
408 | rb_erase(&persistent_gnt->node, root); | |
409 | list_add(&persistent_gnt->remove_node, | |
410 | &blkif->persistent_purge_list); | |
411 | if (--num_clean == 0) | |
412 | goto finished; | |
413 | } | |
414 | /* | |
415 | * If we get here it means we also need to start cleaning | |
416 | * grants that were used since last purge in order to cope | |
417 | * with the requested num | |
418 | */ | |
2d910543 | 419 | if (!scan_used && !clean_used) { |
77387b82 | 420 | pr_debug("Still missing %u purged frames\n", num_clean); |
3f3aad5e RPM |
421 | scan_used = true; |
422 | goto purge_list; | |
423 | } | |
424 | finished: | |
2d910543 | 425 | if (!clean_used) { |
77387b82 | 426 | pr_debug("Finished scanning for grants to clean, removing used flag\n"); |
2d910543 RPM |
427 | clean_used = true; |
428 | goto purge_list; | |
3f3aad5e | 429 | } |
2d910543 | 430 | |
3f3aad5e RPM |
431 | blkif->persistent_gnt_c -= (total - num_clean); |
432 | blkif->vbd.overflow_max_grants = 0; | |
433 | ||
434 | /* We can defer this work */ | |
3f3aad5e | 435 | schedule_work(&blkif->persistent_purge_work); |
77387b82 | 436 | pr_debug("Purged %u/%u\n", (total - num_clean), total); |
3f3aad5e RPM |
437 | return; |
438 | } | |
439 | ||
a1397fa3 KRW |
440 | /* |
441 | * Retrieve from the 'pending_reqs' a free pending_req structure to be used. | |
4d05a28d | 442 | */ |
bf0720c4 | 443 | static struct pending_req *alloc_req(struct xen_blkif *blkif) |
4d05a28d | 444 | { |
2e9977c2 | 445 | struct pending_req *req = NULL; |
4d05a28d KRW |
446 | unsigned long flags; |
447 | ||
bf0720c4 RPM |
448 | spin_lock_irqsave(&blkif->pending_free_lock, flags); |
449 | if (!list_empty(&blkif->pending_free)) { | |
450 | req = list_entry(blkif->pending_free.next, struct pending_req, | |
2e9977c2 | 451 | free_list); |
4d05a28d KRW |
452 | list_del(&req->free_list); |
453 | } | |
bf0720c4 | 454 | spin_unlock_irqrestore(&blkif->pending_free_lock, flags); |
4d05a28d KRW |
455 | return req; |
456 | } | |
457 | ||
a1397fa3 KRW |
458 | /* |
459 | * Return the 'pending_req' structure back to the freepool. We also | |
460 | * wake up the thread if it was waiting for a free page. | |
461 | */ | |
bf0720c4 | 462 | static void free_req(struct xen_blkif *blkif, struct pending_req *req) |
4d05a28d KRW |
463 | { |
464 | unsigned long flags; | |
465 | int was_empty; | |
466 | ||
bf0720c4 RPM |
467 | spin_lock_irqsave(&blkif->pending_free_lock, flags); |
468 | was_empty = list_empty(&blkif->pending_free); | |
469 | list_add(&req->free_list, &blkif->pending_free); | |
470 | spin_unlock_irqrestore(&blkif->pending_free_lock, flags); | |
4d05a28d | 471 | if (was_empty) |
bf0720c4 | 472 | wake_up(&blkif->pending_free_wq); |
4d05a28d KRW |
473 | } |
474 | ||
ee9ff853 KRW |
475 | /* |
476 | * Routines for managing virtual block devices (vbds). | |
477 | */ | |
3d814731 KRW |
478 | static int xen_vbd_translate(struct phys_req *req, struct xen_blkif *blkif, |
479 | int operation) | |
ee9ff853 | 480 | { |
3d814731 | 481 | struct xen_vbd *vbd = &blkif->vbd; |
ee9ff853 KRW |
482 | int rc = -EACCES; |
483 | ||
484 | if ((operation != READ) && vbd->readonly) | |
485 | goto out; | |
486 | ||
8ab52150 JB |
487 | if (likely(req->nr_sects)) { |
488 | blkif_sector_t end = req->sector_number + req->nr_sects; | |
489 | ||
490 | if (unlikely(end < req->sector_number)) | |
491 | goto out; | |
492 | if (unlikely(end > vbd_sz(vbd))) | |
493 | goto out; | |
494 | } | |
ee9ff853 KRW |
495 | |
496 | req->dev = vbd->pdevice; | |
497 | req->bdev = vbd->bdev; | |
498 | rc = 0; | |
499 | ||
500 | out: | |
501 | return rc; | |
502 | } | |
503 | ||
3d814731 | 504 | static void xen_vbd_resize(struct xen_blkif *blkif) |
ee9ff853 | 505 | { |
3d814731 | 506 | struct xen_vbd *vbd = &blkif->vbd; |
ee9ff853 KRW |
507 | struct xenbus_transaction xbt; |
508 | int err; | |
8b6bf747 | 509 | struct xenbus_device *dev = xen_blkbk_xenbus(blkif->be); |
42c7841d | 510 | unsigned long long new_size = vbd_sz(vbd); |
ee9ff853 | 511 | |
77387b82 | 512 | pr_info("VBD Resize: Domid: %d, Device: (%d, %d)\n", |
ee9ff853 | 513 | blkif->domid, MAJOR(vbd->pdevice), MINOR(vbd->pdevice)); |
77387b82 | 514 | pr_info("VBD Resize: new size %llu\n", new_size); |
ee9ff853 KRW |
515 | vbd->size = new_size; |
516 | again: | |
517 | err = xenbus_transaction_start(&xbt); | |
518 | if (err) { | |
77387b82 | 519 | pr_warn("Error starting transaction\n"); |
ee9ff853 KRW |
520 | return; |
521 | } | |
522 | err = xenbus_printf(xbt, dev->nodename, "sectors", "%llu", | |
42c7841d | 523 | (unsigned long long)vbd_sz(vbd)); |
ee9ff853 | 524 | if (err) { |
77387b82 | 525 | pr_warn("Error writing new size\n"); |
ee9ff853 KRW |
526 | goto abort; |
527 | } | |
528 | /* | |
529 | * Write the current state; we will use this to synchronize | |
530 | * the front-end. If the current state is "connected" the | |
531 | * front-end will get the new size information online. | |
532 | */ | |
533 | err = xenbus_printf(xbt, dev->nodename, "state", "%d", dev->state); | |
534 | if (err) { | |
77387b82 | 535 | pr_warn("Error writing the state\n"); |
ee9ff853 KRW |
536 | goto abort; |
537 | } | |
538 | ||
539 | err = xenbus_transaction_end(xbt, 0); | |
540 | if (err == -EAGAIN) | |
541 | goto again; | |
542 | if (err) | |
77387b82 | 543 | pr_warn("Error ending transaction\n"); |
496b318e | 544 | return; |
ee9ff853 KRW |
545 | abort: |
546 | xenbus_transaction_end(xbt, 1); | |
547 | } | |
548 | ||
a1397fa3 | 549 | /* |
b0aef179 KRW |
550 | * Notification from the guest OS. |
551 | */ | |
30fd1502 | 552 | static void blkif_notify_work(struct xen_blkif *blkif) |
4d05a28d | 553 | { |
b0aef179 KRW |
554 | blkif->waiting_reqs = 1; |
555 | wake_up(&blkif->wq); | |
556 | } | |
4d05a28d | 557 | |
8b6bf747 | 558 | irqreturn_t xen_blkif_be_int(int irq, void *dev_id) |
b0aef179 KRW |
559 | { |
560 | blkif_notify_work(dev_id); | |
561 | return IRQ_HANDLED; | |
4d05a28d KRW |
562 | } |
563 | ||
2e9977c2 | 564 | /* |
4d05a28d KRW |
565 | * SCHEDULER FUNCTIONS |
566 | */ | |
567 | ||
30fd1502 | 568 | static void print_stats(struct xen_blkif *blkif) |
4d05a28d | 569 | { |
77387b82 | 570 | pr_info("(%s): oo %3llu | rd %4llu | wr %4llu | f %4llu" |
3f3aad5e | 571 | " | ds %4llu | pg: %4u/%4d\n", |
ebe81906 | 572 | current->comm, blkif->st_oo_req, |
b3cb0d6a | 573 | blkif->st_rd_req, blkif->st_wr_req, |
c1a15d08 RPM |
574 | blkif->st_f_req, blkif->st_ds_req, |
575 | blkif->persistent_gnt_c, | |
3f3aad5e | 576 | xen_blkif_max_pgrants); |
4d05a28d KRW |
577 | blkif->st_print = jiffies + msecs_to_jiffies(10 * 1000); |
578 | blkif->st_rd_req = 0; | |
579 | blkif->st_wr_req = 0; | |
580 | blkif->st_oo_req = 0; | |
b3cb0d6a | 581 | blkif->st_ds_req = 0; |
4d05a28d KRW |
582 | } |
583 | ||
8b6bf747 | 584 | int xen_blkif_schedule(void *arg) |
4d05a28d | 585 | { |
30fd1502 | 586 | struct xen_blkif *blkif = arg; |
3d814731 | 587 | struct xen_vbd *vbd = &blkif->vbd; |
3f3aad5e | 588 | unsigned long timeout; |
8e3f8755 | 589 | int ret; |
4d05a28d | 590 | |
8b6bf747 | 591 | xen_blkif_get(blkif); |
4d05a28d | 592 | |
4d05a28d KRW |
593 | while (!kthread_should_stop()) { |
594 | if (try_to_freeze()) | |
595 | continue; | |
42c7841d | 596 | if (unlikely(vbd->size != vbd_sz(vbd))) |
3d814731 | 597 | xen_vbd_resize(blkif); |
4d05a28d | 598 | |
3f3aad5e RPM |
599 | timeout = msecs_to_jiffies(LRU_INTERVAL); |
600 | ||
601 | timeout = wait_event_interruptible_timeout( | |
4d05a28d | 602 | blkif->wq, |
3f3aad5e RPM |
603 | blkif->waiting_reqs || kthread_should_stop(), |
604 | timeout); | |
605 | if (timeout == 0) | |
606 | goto purge_gnt_list; | |
607 | timeout = wait_event_interruptible_timeout( | |
bf0720c4 RPM |
608 | blkif->pending_free_wq, |
609 | !list_empty(&blkif->pending_free) || | |
3f3aad5e RPM |
610 | kthread_should_stop(), |
611 | timeout); | |
612 | if (timeout == 0) | |
613 | goto purge_gnt_list; | |
4d05a28d KRW |
614 | |
615 | blkif->waiting_reqs = 0; | |
616 | smp_mb(); /* clear flag *before* checking for work */ | |
617 | ||
8e3f8755 KRW |
618 | ret = do_block_io_op(blkif); |
619 | if (ret > 0) | |
4d05a28d | 620 | blkif->waiting_reqs = 1; |
8e3f8755 KRW |
621 | if (ret == -EACCES) |
622 | wait_event_interruptible(blkif->shutdown_wq, | |
623 | kthread_should_stop()); | |
4d05a28d | 624 | |
3f3aad5e RPM |
625 | purge_gnt_list: |
626 | if (blkif->vbd.feature_gnt_persistent && | |
627 | time_after(jiffies, blkif->next_lru)) { | |
628 | purge_persistent_gnt(blkif); | |
629 | blkif->next_lru = jiffies + msecs_to_jiffies(LRU_INTERVAL); | |
630 | } | |
631 | ||
c6cc142d RPM |
632 | /* Shrink if we have more than xen_blkif_max_buffer_pages */ |
633 | shrink_free_pagepool(blkif, xen_blkif_max_buffer_pages); | |
634 | ||
4d05a28d KRW |
635 | if (log_stats && time_after(jiffies, blkif->st_print)) |
636 | print_stats(blkif); | |
637 | } | |
638 | ||
ef753411 RPM |
639 | /* Drain pending purge work */ |
640 | flush_work(&blkif->persistent_purge_work); | |
c6cc142d | 641 | |
ef753411 RPM |
642 | if (log_stats) |
643 | print_stats(blkif); | |
644 | ||
645 | blkif->xenblkd = NULL; | |
646 | xen_blkif_put(blkif); | |
647 | ||
648 | return 0; | |
649 | } | |
650 | ||
651 | /* | |
652 | * Remove persistent grants and empty the pool of free pages | |
653 | */ | |
654 | void xen_blkbk_free_caches(struct xen_blkif *blkif) | |
655 | { | |
0a8704a5 | 656 | /* Free all persistent grant pages */ |
4d4f270f | 657 | if (!RB_EMPTY_ROOT(&blkif->persistent_gnts)) |
c6cc142d | 658 | free_persistent_gnts(blkif, &blkif->persistent_gnts, |
4d4f270f | 659 | blkif->persistent_gnt_c); |
0a8704a5 | 660 | |
0a8704a5 | 661 | BUG_ON(!RB_EMPTY_ROOT(&blkif->persistent_gnts)); |
4d4f270f | 662 | blkif->persistent_gnt_c = 0; |
0a8704a5 | 663 | |
2ed22e3c MR |
664 | /* Since we are shutting down remove all pages from the buffer */ |
665 | shrink_free_pagepool(blkif, 0 /* All */); | |
4d05a28d KRW |
666 | } |
667 | ||
c43cf3ea JH |
668 | static unsigned int xen_blkbk_unmap_prepare( |
669 | struct xen_blkif *blkif, | |
670 | struct grant_page **pages, | |
671 | unsigned int num, | |
672 | struct gnttab_unmap_grant_ref *unmap_ops, | |
673 | struct page **unmap_pages) | |
b0aef179 | 674 | { |
b0aef179 | 675 | unsigned int i, invcount = 0; |
b0aef179 | 676 | |
31552ee3 | 677 | for (i = 0; i < num; i++) { |
bb642e83 RPM |
678 | if (pages[i]->persistent_gnt != NULL) { |
679 | put_persistent_gnt(blkif, pages[i]->persistent_gnt); | |
0a8704a5 | 680 | continue; |
3f3aad5e | 681 | } |
bb642e83 | 682 | if (pages[i]->handle == BLKBACK_INVALID_HANDLE) |
b0aef179 | 683 | continue; |
bb642e83 | 684 | unmap_pages[invcount] = pages[i]->page; |
c43cf3ea | 685 | gnttab_set_unmap_op(&unmap_ops[invcount], vaddr(pages[i]->page), |
bb642e83 RPM |
686 | GNTMAP_host_map, pages[i]->handle); |
687 | pages[i]->handle = BLKBACK_INVALID_HANDLE; | |
c43cf3ea JH |
688 | invcount++; |
689 | } | |
690 | ||
691 | return invcount; | |
692 | } | |
693 | ||
694 | static void xen_blkbk_unmap_and_respond_callback(int result, struct gntab_unmap_queue_data *data) | |
695 | { | |
696 | struct pending_req* pending_req = (struct pending_req*) (data->data); | |
697 | struct xen_blkif *blkif = pending_req->blkif; | |
698 | ||
699 | /* BUG_ON used to reproduce existing behaviour, | |
700 | but is this the best way to deal with this? */ | |
701 | BUG_ON(result); | |
702 | ||
703 | put_free_pages(blkif, data->pages, data->count); | |
704 | make_response(blkif, pending_req->id, | |
705 | pending_req->operation, pending_req->status); | |
706 | free_req(blkif, pending_req); | |
707 | /* | |
708 | * Make sure the request is freed before releasing blkif, | |
709 | * or there could be a race between free_req and the | |
710 | * cleanup done in xen_blkif_free during shutdown. | |
711 | * | |
712 | * NB: The fact that we might try to wake up pending_free_wq | |
713 | * before drain_complete (in case there's a drain going on) | |
714 | * it's not a problem with our current implementation | |
715 | * because we can assure there's no thread waiting on | |
716 | * pending_free_wq if there's a drain going on, but it has | |
717 | * to be taken into account if the current model is changed. | |
718 | */ | |
719 | if (atomic_dec_and_test(&blkif->inflight) && atomic_read(&blkif->drain)) { | |
720 | complete(&blkif->drain_complete); | |
721 | } | |
722 | xen_blkif_put(blkif); | |
723 | } | |
724 | ||
725 | static void xen_blkbk_unmap_and_respond(struct pending_req *req) | |
726 | { | |
727 | struct gntab_unmap_queue_data* work = &req->gnttab_unmap_data; | |
728 | struct xen_blkif *blkif = req->blkif; | |
729 | struct grant_page **pages = req->segments; | |
730 | unsigned int invcount; | |
731 | ||
732 | invcount = xen_blkbk_unmap_prepare(blkif, pages, req->nr_pages, | |
733 | req->unmap, req->unmap_pages); | |
734 | ||
735 | work->data = req; | |
736 | work->done = xen_blkbk_unmap_and_respond_callback; | |
737 | work->unmap_ops = req->unmap; | |
738 | work->kunmap_ops = NULL; | |
739 | work->pages = req->unmap_pages; | |
740 | work->count = invcount; | |
741 | ||
742 | gnttab_unmap_refs_async(&req->gnttab_unmap_data); | |
743 | } | |
744 | ||
745 | ||
746 | /* | |
747 | * Unmap the grant references. | |
748 | * | |
749 | * This could accumulate ops up to the batch size to reduce the number | |
750 | * of hypercalls, but since this is only used in error paths there's | |
751 | * no real need. | |
752 | */ | |
753 | static void xen_blkbk_unmap(struct xen_blkif *blkif, | |
754 | struct grant_page *pages[], | |
755 | int num) | |
756 | { | |
757 | struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
758 | struct page *unmap_pages[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
759 | unsigned int invcount = 0; | |
760 | int ret; | |
761 | ||
762 | while (num) { | |
763 | unsigned int batch = min(num, BLKIF_MAX_SEGMENTS_PER_REQUEST); | |
764 | ||
765 | invcount = xen_blkbk_unmap_prepare(blkif, pages, batch, | |
766 | unmap, unmap_pages); | |
767 | if (invcount) { | |
768 | ret = gnttab_unmap_refs(unmap, NULL, unmap_pages, invcount); | |
31552ee3 RPM |
769 | BUG_ON(ret); |
770 | put_free_pages(blkif, unmap_pages, invcount); | |
31552ee3 | 771 | } |
c43cf3ea JH |
772 | pages += batch; |
773 | num -= batch; | |
b0aef179 | 774 | } |
b0aef179 | 775 | } |
01f37f2d | 776 | |
bb642e83 RPM |
777 | static int xen_blkbk_map(struct xen_blkif *blkif, |
778 | struct grant_page *pages[], | |
31552ee3 | 779 | int num, bool ro) |
1a95fe6e KRW |
780 | { |
781 | struct gnttab_map_grant_ref map[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
0a8704a5 RPM |
782 | struct page *pages_to_gnt[BLKIF_MAX_SEGMENTS_PER_REQUEST]; |
783 | struct persistent_gnt *persistent_gnt = NULL; | |
0a8704a5 | 784 | phys_addr_t addr = 0; |
c6cc142d | 785 | int i, seg_idx, new_map_idx; |
0a8704a5 | 786 | int segs_to_map = 0; |
1a95fe6e | 787 | int ret = 0; |
31552ee3 | 788 | int last_map = 0, map_until = 0; |
0a8704a5 RPM |
789 | int use_persistent_gnts; |
790 | ||
791 | use_persistent_gnts = (blkif->vbd.feature_gnt_persistent); | |
792 | ||
01f37f2d KRW |
793 | /* |
794 | * Fill out preq.nr_sects with proper amount of sectors, and setup | |
1a95fe6e KRW |
795 | * assign map[..] with the PFN of the page in our domain with the |
796 | * corresponding grant reference for each page. | |
797 | */ | |
31552ee3 RPM |
798 | again: |
799 | for (i = map_until; i < num; i++) { | |
1a95fe6e KRW |
800 | uint32_t flags; |
801 | ||
0a8704a5 RPM |
802 | if (use_persistent_gnts) |
803 | persistent_gnt = get_persistent_gnt( | |
3f3aad5e | 804 | blkif, |
bb642e83 | 805 | pages[i]->gref); |
0a8704a5 RPM |
806 | |
807 | if (persistent_gnt) { | |
808 | /* | |
809 | * We are using persistent grants and | |
810 | * the grant is already mapped | |
811 | */ | |
bb642e83 RPM |
812 | pages[i]->page = persistent_gnt->page; |
813 | pages[i]->persistent_gnt = persistent_gnt; | |
0a8704a5 | 814 | } else { |
bb642e83 | 815 | if (get_free_page(blkif, &pages[i]->page)) |
c6cc142d | 816 | goto out_of_memory; |
bb642e83 RPM |
817 | addr = vaddr(pages[i]->page); |
818 | pages_to_gnt[segs_to_map] = pages[i]->page; | |
819 | pages[i]->persistent_gnt = NULL; | |
0a8704a5 | 820 | flags = GNTMAP_host_map; |
31552ee3 | 821 | if (!use_persistent_gnts && ro) |
0a8704a5 RPM |
822 | flags |= GNTMAP_readonly; |
823 | gnttab_set_map_op(&map[segs_to_map++], addr, | |
bb642e83 | 824 | flags, pages[i]->gref, |
0a8704a5 RPM |
825 | blkif->domid); |
826 | } | |
31552ee3 RPM |
827 | map_until = i + 1; |
828 | if (segs_to_map == BLKIF_MAX_SEGMENTS_PER_REQUEST) | |
829 | break; | |
1a95fe6e KRW |
830 | } |
831 | ||
0a8704a5 RPM |
832 | if (segs_to_map) { |
833 | ret = gnttab_map_refs(map, NULL, pages_to_gnt, segs_to_map); | |
834 | BUG_ON(ret); | |
835 | } | |
1a95fe6e | 836 | |
01f37f2d KRW |
837 | /* |
838 | * Now swizzle the MFN in our domain with the MFN from the other domain | |
1a95fe6e KRW |
839 | * so that when we access vaddr(pending_req,i) it has the contents of |
840 | * the page from the other domain. | |
841 | */ | |
31552ee3 | 842 | for (seg_idx = last_map, new_map_idx = 0; seg_idx < map_until; seg_idx++) { |
bb642e83 | 843 | if (!pages[seg_idx]->persistent_gnt) { |
0a8704a5 | 844 | /* This is a newly mapped grant */ |
c6cc142d RPM |
845 | BUG_ON(new_map_idx >= segs_to_map); |
846 | if (unlikely(map[new_map_idx].status != 0)) { | |
77387b82 | 847 | pr_debug("invalid buffer -- could not remap it\n"); |
61cecca8 | 848 | put_free_pages(blkif, &pages[seg_idx]->page, 1); |
bb642e83 | 849 | pages[seg_idx]->handle = BLKBACK_INVALID_HANDLE; |
0a8704a5 | 850 | ret |= 1; |
31552ee3 | 851 | goto next; |
0a8704a5 | 852 | } |
bb642e83 | 853 | pages[seg_idx]->handle = map[new_map_idx].handle; |
c6cc142d | 854 | } else { |
31552ee3 | 855 | continue; |
0a8704a5 | 856 | } |
c6cc142d | 857 | if (use_persistent_gnts && |
3f3aad5e | 858 | blkif->persistent_gnt_c < xen_blkif_max_pgrants) { |
c6cc142d RPM |
859 | /* |
860 | * We are using persistent grants, the grant is | |
3f3aad5e | 861 | * not mapped but we might have room for it. |
c6cc142d RPM |
862 | */ |
863 | persistent_gnt = kmalloc(sizeof(struct persistent_gnt), | |
864 | GFP_KERNEL); | |
865 | if (!persistent_gnt) { | |
0a8704a5 | 866 | /* |
c6cc142d RPM |
867 | * If we don't have enough memory to |
868 | * allocate the persistent_gnt struct | |
869 | * map this grant non-persistenly | |
0a8704a5 | 870 | */ |
31552ee3 | 871 | goto next; |
0a8704a5 | 872 | } |
c6cc142d RPM |
873 | persistent_gnt->gnt = map[new_map_idx].ref; |
874 | persistent_gnt->handle = map[new_map_idx].handle; | |
bb642e83 | 875 | persistent_gnt->page = pages[seg_idx]->page; |
3f3aad5e | 876 | if (add_persistent_gnt(blkif, |
c6cc142d RPM |
877 | persistent_gnt)) { |
878 | kfree(persistent_gnt); | |
879 | persistent_gnt = NULL; | |
31552ee3 | 880 | goto next; |
c6cc142d | 881 | } |
bb642e83 | 882 | pages[seg_idx]->persistent_gnt = persistent_gnt; |
77387b82 | 883 | pr_debug("grant %u added to the tree of persistent grants, using %u/%u\n", |
c6cc142d | 884 | persistent_gnt->gnt, blkif->persistent_gnt_c, |
3f3aad5e | 885 | xen_blkif_max_pgrants); |
c6cc142d RPM |
886 | goto next; |
887 | } | |
888 | if (use_persistent_gnts && !blkif->vbd.overflow_max_grants) { | |
889 | blkif->vbd.overflow_max_grants = 1; | |
77387b82 | 890 | pr_debug("domain %u, device %#x is using maximum number of persistent grants\n", |
c6cc142d | 891 | blkif->domid, blkif->vbd.handle); |
1a95fe6e | 892 | } |
c6cc142d RPM |
893 | /* |
894 | * We could not map this grant persistently, so use it as | |
895 | * a non-persistent grant. | |
896 | */ | |
c6cc142d | 897 | next: |
31552ee3 | 898 | new_map_idx++; |
1a95fe6e | 899 | } |
31552ee3 RPM |
900 | segs_to_map = 0; |
901 | last_map = map_until; | |
902 | if (map_until != num) | |
903 | goto again; | |
904 | ||
1a95fe6e | 905 | return ret; |
c6cc142d RPM |
906 | |
907 | out_of_memory: | |
77387b82 | 908 | pr_alert("%s: out of memory\n", __func__); |
c6cc142d RPM |
909 | put_free_pages(blkif, pages_to_gnt, segs_to_map); |
910 | return -ENOMEM; | |
1a95fe6e KRW |
911 | } |
912 | ||
bb642e83 | 913 | static int xen_blkbk_map_seg(struct pending_req *pending_req) |
31552ee3 | 914 | { |
402b27f9 | 915 | int rc; |
31552ee3 | 916 | |
bb642e83 | 917 | rc = xen_blkbk_map(pending_req->blkif, pending_req->segments, |
402b27f9 | 918 | pending_req->nr_pages, |
31552ee3 | 919 | (pending_req->operation != BLKIF_OP_READ)); |
31552ee3 | 920 | |
402b27f9 RPM |
921 | return rc; |
922 | } | |
31552ee3 | 923 | |
402b27f9 RPM |
924 | static int xen_blkbk_parse_indirect(struct blkif_request *req, |
925 | struct pending_req *pending_req, | |
926 | struct seg_buf seg[], | |
927 | struct phys_req *preq) | |
928 | { | |
bb642e83 | 929 | struct grant_page **pages = pending_req->indirect_pages; |
402b27f9 RPM |
930 | struct xen_blkif *blkif = pending_req->blkif; |
931 | int indirect_grefs, rc, n, nseg, i; | |
80bfa2f6 | 932 | struct blkif_request_segment *segments = NULL; |
402b27f9 RPM |
933 | |
934 | nseg = pending_req->nr_pages; | |
935 | indirect_grefs = INDIRECT_PAGES(nseg); | |
936 | BUG_ON(indirect_grefs > BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST); | |
937 | ||
bb642e83 RPM |
938 | for (i = 0; i < indirect_grefs; i++) |
939 | pages[i]->gref = req->u.indirect.indirect_grefs[i]; | |
940 | ||
941 | rc = xen_blkbk_map(blkif, pages, indirect_grefs, true); | |
402b27f9 RPM |
942 | if (rc) |
943 | goto unmap; | |
944 | ||
945 | for (n = 0, i = 0; n < nseg; n++) { | |
946 | if ((n % SEGS_PER_INDIRECT_FRAME) == 0) { | |
947 | /* Map indirect segments */ | |
948 | if (segments) | |
949 | kunmap_atomic(segments); | |
bb642e83 | 950 | segments = kmap_atomic(pages[n/SEGS_PER_INDIRECT_FRAME]->page); |
402b27f9 RPM |
951 | } |
952 | i = n % SEGS_PER_INDIRECT_FRAME; | |
bb642e83 | 953 | pending_req->segments[n]->gref = segments[i].gref; |
402b27f9 RPM |
954 | seg[n].nsec = segments[i].last_sect - |
955 | segments[i].first_sect + 1; | |
956 | seg[n].offset = (segments[i].first_sect << 9); | |
957 | if ((segments[i].last_sect >= (PAGE_SIZE >> 9)) || | |
958 | (segments[i].last_sect < segments[i].first_sect)) { | |
959 | rc = -EINVAL; | |
960 | goto unmap; | |
961 | } | |
962 | preq->nr_sects += seg[n].nsec; | |
963 | } | |
964 | ||
965 | unmap: | |
966 | if (segments) | |
967 | kunmap_atomic(segments); | |
bb642e83 | 968 | xen_blkbk_unmap(blkif, pages, indirect_grefs); |
402b27f9 | 969 | return rc; |
31552ee3 RPM |
970 | } |
971 | ||
42146352 KRW |
972 | static int dispatch_discard_io(struct xen_blkif *blkif, |
973 | struct blkif_request *req) | |
b3cb0d6a LD |
974 | { |
975 | int err = 0; | |
976 | int status = BLKIF_RSP_OKAY; | |
977 | struct block_device *bdev = blkif->vbd.bdev; | |
4dae7670 | 978 | unsigned long secure; |
604c499c | 979 | struct phys_req preq; |
b3cb0d6a | 980 | |
ea5ec76d VN |
981 | xen_blkif_get(blkif); |
982 | ||
604c499c KRW |
983 | preq.sector_number = req->u.discard.sector_number; |
984 | preq.nr_sects = req->u.discard.nr_sectors; | |
985 | ||
986 | err = xen_vbd_translate(&preq, blkif, WRITE); | |
987 | if (err) { | |
77387b82 | 988 | pr_warn("access denied: DISCARD [%llu->%llu] on dev=%04x\n", |
604c499c KRW |
989 | preq.sector_number, |
990 | preq.sector_number + preq.nr_sects, blkif->vbd.pdevice); | |
991 | goto fail_response; | |
992 | } | |
42146352 KRW |
993 | blkif->st_ds_req++; |
994 | ||
4dae7670 KRW |
995 | secure = (blkif->vbd.discard_secure && |
996 | (req->u.discard.flag & BLKIF_DISCARD_SECURE)) ? | |
997 | BLKDEV_DISCARD_SECURE : 0; | |
998 | ||
999 | err = blkdev_issue_discard(bdev, req->u.discard.sector_number, | |
1000 | req->u.discard.nr_sectors, | |
1001 | GFP_KERNEL, secure); | |
604c499c | 1002 | fail_response: |
b3cb0d6a | 1003 | if (err == -EOPNOTSUPP) { |
77387b82 | 1004 | pr_debug("discard op failed, not supported\n"); |
b3cb0d6a LD |
1005 | status = BLKIF_RSP_EOPNOTSUPP; |
1006 | } else if (err) | |
1007 | status = BLKIF_RSP_ERROR; | |
1008 | ||
97e36834 | 1009 | make_response(blkif, req->u.discard.id, req->operation, status); |
42146352 KRW |
1010 | xen_blkif_put(blkif); |
1011 | return err; | |
b3cb0d6a LD |
1012 | } |
1013 | ||
0e367ae4 DV |
1014 | static int dispatch_other_io(struct xen_blkif *blkif, |
1015 | struct blkif_request *req, | |
1016 | struct pending_req *pending_req) | |
1017 | { | |
bf0720c4 | 1018 | free_req(blkif, pending_req); |
0e367ae4 DV |
1019 | make_response(blkif, req->u.other.id, req->operation, |
1020 | BLKIF_RSP_EOPNOTSUPP); | |
1021 | return -EIO; | |
1022 | } | |
1023 | ||
29bde093 KRW |
1024 | static void xen_blk_drain_io(struct xen_blkif *blkif) |
1025 | { | |
1026 | atomic_set(&blkif->drain, 1); | |
1027 | do { | |
c05f3e3c | 1028 | if (atomic_read(&blkif->inflight) == 0) |
6927d920 | 1029 | break; |
29bde093 KRW |
1030 | wait_for_completion_interruptible_timeout( |
1031 | &blkif->drain_complete, HZ); | |
1032 | ||
1033 | if (!atomic_read(&blkif->drain)) | |
1034 | break; | |
29bde093 KRW |
1035 | } while (!kthread_should_stop()); |
1036 | atomic_set(&blkif->drain, 0); | |
1037 | } | |
1038 | ||
a1397fa3 KRW |
1039 | /* |
1040 | * Completion callback on the bio's. Called as bh->b_end_io() | |
4d05a28d KRW |
1041 | */ |
1042 | ||
2e9977c2 | 1043 | static void __end_block_io_op(struct pending_req *pending_req, int error) |
4d05a28d KRW |
1044 | { |
1045 | /* An error fails the entire request. */ | |
24f567f9 | 1046 | if ((pending_req->operation == BLKIF_OP_FLUSH_DISKCACHE) && |
4d05a28d | 1047 | (error == -EOPNOTSUPP)) { |
77387b82 | 1048 | pr_debug("flush diskcache op failed, not supported\n"); |
24f567f9 | 1049 | xen_blkbk_flush_diskcache(XBT_NIL, pending_req->blkif->be, 0); |
4d05a28d | 1050 | pending_req->status = BLKIF_RSP_EOPNOTSUPP; |
29bde093 KRW |
1051 | } else if ((pending_req->operation == BLKIF_OP_WRITE_BARRIER) && |
1052 | (error == -EOPNOTSUPP)) { | |
77387b82 | 1053 | pr_debug("write barrier op failed, not supported\n"); |
29bde093 KRW |
1054 | xen_blkbk_barrier(XBT_NIL, pending_req->blkif->be, 0); |
1055 | pending_req->status = BLKIF_RSP_EOPNOTSUPP; | |
4d05a28d | 1056 | } else if (error) { |
77387b82 | 1057 | pr_debug("Buffer not up-to-date at end of operation," |
ebe81906 | 1058 | " error=%d\n", error); |
4d05a28d KRW |
1059 | pending_req->status = BLKIF_RSP_ERROR; |
1060 | } | |
1061 | ||
01f37f2d KRW |
1062 | /* |
1063 | * If all of the bio's have completed it is time to unmap | |
a1397fa3 | 1064 | * the grant references associated with 'request' and provide |
2e9977c2 KRW |
1065 | * the proper response on the ring. |
1066 | */ | |
c43cf3ea JH |
1067 | if (atomic_dec_and_test(&pending_req->pendcnt)) |
1068 | xen_blkbk_unmap_and_respond(pending_req); | |
4d05a28d KRW |
1069 | } |
1070 | ||
a1397fa3 KRW |
1071 | /* |
1072 | * bio callback. | |
1073 | */ | |
88122933 | 1074 | static void end_block_io_op(struct bio *bio, int error) |
4d05a28d | 1075 | { |
4d05a28d KRW |
1076 | __end_block_io_op(bio->bi_private, error); |
1077 | bio_put(bio); | |
4d05a28d KRW |
1078 | } |
1079 | ||
1080 | ||
4d05a28d | 1081 | |
a1397fa3 KRW |
1082 | /* |
1083 | * Function to copy the from the ring buffer the 'struct blkif_request' | |
1084 | * (which has the sectors we want, number of them, grant references, etc), | |
1085 | * and transmute it to the block API to hand it over to the proper block disk. | |
4d05a28d | 1086 | */ |
b4726a9d DS |
1087 | static int |
1088 | __do_block_io_op(struct xen_blkif *blkif) | |
4d05a28d | 1089 | { |
88122933 JF |
1090 | union blkif_back_rings *blk_rings = &blkif->blk_rings; |
1091 | struct blkif_request req; | |
2e9977c2 | 1092 | struct pending_req *pending_req; |
4d05a28d KRW |
1093 | RING_IDX rc, rp; |
1094 | int more_to_do = 0; | |
1095 | ||
1096 | rc = blk_rings->common.req_cons; | |
1097 | rp = blk_rings->common.sring->req_prod; | |
1098 | rmb(); /* Ensure we see queued requests up to 'rp'. */ | |
1099 | ||
8e3f8755 KRW |
1100 | if (RING_REQUEST_PROD_OVERFLOW(&blk_rings->common, rp)) { |
1101 | rc = blk_rings->common.rsp_prod_pvt; | |
77387b82 | 1102 | pr_warn("Frontend provided bogus ring requests (%d - %d = %d). Halting ring processing on dev=%04x\n", |
8e3f8755 KRW |
1103 | rp, rc, rp - rc, blkif->vbd.pdevice); |
1104 | return -EACCES; | |
1105 | } | |
4d05a28d KRW |
1106 | while (rc != rp) { |
1107 | ||
1108 | if (RING_REQUEST_CONS_OVERFLOW(&blk_rings->common, rc)) | |
1109 | break; | |
1110 | ||
8270b45b | 1111 | if (kthread_should_stop()) { |
4d05a28d KRW |
1112 | more_to_do = 1; |
1113 | break; | |
1114 | } | |
1115 | ||
bf0720c4 | 1116 | pending_req = alloc_req(blkif); |
8270b45b KF |
1117 | if (NULL == pending_req) { |
1118 | blkif->st_oo_req++; | |
4d05a28d KRW |
1119 | more_to_do = 1; |
1120 | break; | |
1121 | } | |
1122 | ||
1123 | switch (blkif->blk_protocol) { | |
1124 | case BLKIF_PROTOCOL_NATIVE: | |
1125 | memcpy(&req, RING_GET_REQUEST(&blk_rings->native, rc), sizeof(req)); | |
1126 | break; | |
1127 | case BLKIF_PROTOCOL_X86_32: | |
1128 | blkif_get_x86_32_req(&req, RING_GET_REQUEST(&blk_rings->x86_32, rc)); | |
1129 | break; | |
1130 | case BLKIF_PROTOCOL_X86_64: | |
1131 | blkif_get_x86_64_req(&req, RING_GET_REQUEST(&blk_rings->x86_64, rc)); | |
1132 | break; | |
1133 | default: | |
1134 | BUG(); | |
1135 | } | |
1136 | blk_rings->common.req_cons = ++rc; /* before make_response() */ | |
1137 | ||
1138 | /* Apply all sanity checks to /private copy/ of request. */ | |
1139 | barrier(); | |
0e367ae4 DV |
1140 | |
1141 | switch (req.operation) { | |
1142 | case BLKIF_OP_READ: | |
1143 | case BLKIF_OP_WRITE: | |
1144 | case BLKIF_OP_WRITE_BARRIER: | |
1145 | case BLKIF_OP_FLUSH_DISKCACHE: | |
402b27f9 | 1146 | case BLKIF_OP_INDIRECT: |
0e367ae4 DV |
1147 | if (dispatch_rw_block_io(blkif, &req, pending_req)) |
1148 | goto done; | |
1149 | break; | |
1150 | case BLKIF_OP_DISCARD: | |
bf0720c4 | 1151 | free_req(blkif, pending_req); |
42146352 | 1152 | if (dispatch_discard_io(blkif, &req)) |
0e367ae4 | 1153 | goto done; |
4d05a28d | 1154 | break; |
0e367ae4 DV |
1155 | default: |
1156 | if (dispatch_other_io(blkif, &req, pending_req)) | |
1157 | goto done; | |
1158 | break; | |
1159 | } | |
4d05a28d KRW |
1160 | |
1161 | /* Yield point for this unbounded loop. */ | |
1162 | cond_resched(); | |
1163 | } | |
0e367ae4 | 1164 | done: |
4d05a28d KRW |
1165 | return more_to_do; |
1166 | } | |
1167 | ||
b4726a9d DS |
1168 | static int |
1169 | do_block_io_op(struct xen_blkif *blkif) | |
1170 | { | |
1171 | union blkif_back_rings *blk_rings = &blkif->blk_rings; | |
1172 | int more_to_do; | |
1173 | ||
1174 | do { | |
1175 | more_to_do = __do_block_io_op(blkif); | |
1176 | if (more_to_do) | |
1177 | break; | |
1178 | ||
1179 | RING_FINAL_CHECK_FOR_REQUESTS(&blk_rings->common, more_to_do); | |
1180 | } while (more_to_do); | |
1181 | ||
1182 | return more_to_do; | |
1183 | } | |
a1397fa3 | 1184 | /* |
01f37f2d KRW |
1185 | * Transmutation of the 'struct blkif_request' to a proper 'struct bio' |
1186 | * and call the 'submit_bio' to pass it to the underlying storage. | |
a1397fa3 | 1187 | */ |
30fd1502 KRW |
1188 | static int dispatch_rw_block_io(struct xen_blkif *blkif, |
1189 | struct blkif_request *req, | |
1190 | struct pending_req *pending_req) | |
4d05a28d | 1191 | { |
4d05a28d | 1192 | struct phys_req preq; |
402b27f9 | 1193 | struct seg_buf *seg = pending_req->seg; |
4d05a28d KRW |
1194 | unsigned int nseg; |
1195 | struct bio *bio = NULL; | |
402b27f9 | 1196 | struct bio **biolist = pending_req->biolist; |
1a95fe6e | 1197 | int i, nbio = 0; |
4d05a28d | 1198 | int operation; |
a19be5f0 | 1199 | struct blk_plug plug; |
29bde093 | 1200 | bool drain = false; |
bb642e83 | 1201 | struct grant_page **pages = pending_req->segments; |
402b27f9 RPM |
1202 | unsigned short req_operation; |
1203 | ||
1204 | req_operation = req->operation == BLKIF_OP_INDIRECT ? | |
1205 | req->u.indirect.indirect_op : req->operation; | |
1206 | if ((req->operation == BLKIF_OP_INDIRECT) && | |
1207 | (req_operation != BLKIF_OP_READ) && | |
1208 | (req_operation != BLKIF_OP_WRITE)) { | |
77387b82 | 1209 | pr_debug("Invalid indirect operation (%u)\n", req_operation); |
402b27f9 RPM |
1210 | goto fail_response; |
1211 | } | |
4d05a28d | 1212 | |
402b27f9 | 1213 | switch (req_operation) { |
4d05a28d | 1214 | case BLKIF_OP_READ: |
fc53bf75 | 1215 | blkif->st_rd_req++; |
4d05a28d KRW |
1216 | operation = READ; |
1217 | break; | |
1218 | case BLKIF_OP_WRITE: | |
fc53bf75 | 1219 | blkif->st_wr_req++; |
013c3ca1 | 1220 | operation = WRITE_ODIRECT; |
4d05a28d | 1221 | break; |
29bde093 KRW |
1222 | case BLKIF_OP_WRITE_BARRIER: |
1223 | drain = true; | |
24f567f9 | 1224 | case BLKIF_OP_FLUSH_DISKCACHE: |
fc53bf75 | 1225 | blkif->st_f_req++; |
24f567f9 | 1226 | operation = WRITE_FLUSH; |
4d05a28d KRW |
1227 | break; |
1228 | default: | |
1229 | operation = 0; /* make gcc happy */ | |
fc53bf75 KRW |
1230 | goto fail_response; |
1231 | break; | |
4d05a28d KRW |
1232 | } |
1233 | ||
42146352 | 1234 | /* Check that the number of segments is sane. */ |
402b27f9 RPM |
1235 | nseg = req->operation == BLKIF_OP_INDIRECT ? |
1236 | req->u.indirect.nr_segments : req->u.rw.nr_segments; | |
97e36834 | 1237 | |
42146352 | 1238 | if (unlikely(nseg == 0 && operation != WRITE_FLUSH) || |
402b27f9 RPM |
1239 | unlikely((req->operation != BLKIF_OP_INDIRECT) && |
1240 | (nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST)) || | |
1241 | unlikely((req->operation == BLKIF_OP_INDIRECT) && | |
1242 | (nseg > MAX_INDIRECT_SEGMENTS))) { | |
77387b82 | 1243 | pr_debug("Bad number of segments in request (%d)\n", nseg); |
1a95fe6e | 1244 | /* Haven't submitted any bio's yet. */ |
4d05a28d KRW |
1245 | goto fail_response; |
1246 | } | |
1247 | ||
4d05a28d KRW |
1248 | preq.nr_sects = 0; |
1249 | ||
1250 | pending_req->blkif = blkif; | |
97e36834 | 1251 | pending_req->id = req->u.rw.id; |
402b27f9 | 1252 | pending_req->operation = req_operation; |
4d05a28d KRW |
1253 | pending_req->status = BLKIF_RSP_OKAY; |
1254 | pending_req->nr_pages = nseg; | |
e9350493 | 1255 | |
402b27f9 RPM |
1256 | if (req->operation != BLKIF_OP_INDIRECT) { |
1257 | preq.dev = req->u.rw.handle; | |
1258 | preq.sector_number = req->u.rw.sector_number; | |
1259 | for (i = 0; i < nseg; i++) { | |
bb642e83 | 1260 | pages[i]->gref = req->u.rw.seg[i].gref; |
402b27f9 RPM |
1261 | seg[i].nsec = req->u.rw.seg[i].last_sect - |
1262 | req->u.rw.seg[i].first_sect + 1; | |
1263 | seg[i].offset = (req->u.rw.seg[i].first_sect << 9); | |
1264 | if ((req->u.rw.seg[i].last_sect >= (PAGE_SIZE >> 9)) || | |
1265 | (req->u.rw.seg[i].last_sect < | |
1266 | req->u.rw.seg[i].first_sect)) | |
1267 | goto fail_response; | |
1268 | preq.nr_sects += seg[i].nsec; | |
1269 | } | |
1270 | } else { | |
1271 | preq.dev = req->u.indirect.handle; | |
1272 | preq.sector_number = req->u.indirect.sector_number; | |
1273 | if (xen_blkbk_parse_indirect(req, pending_req, seg, &preq)) | |
4d05a28d | 1274 | goto fail_response; |
4d05a28d KRW |
1275 | } |
1276 | ||
3d814731 | 1277 | if (xen_vbd_translate(&preq, blkif, operation) != 0) { |
77387b82 | 1278 | pr_debug("access denied: %s of [%llu,%llu] on dev=%04x\n", |
ebe81906 KRW |
1279 | operation == READ ? "read" : "write", |
1280 | preq.sector_number, | |
a72d9002 CG |
1281 | preq.sector_number + preq.nr_sects, |
1282 | blkif->vbd.pdevice); | |
1a95fe6e | 1283 | goto fail_response; |
4d05a28d | 1284 | } |
01f37f2d KRW |
1285 | |
1286 | /* | |
3d814731 | 1287 | * This check _MUST_ be done after xen_vbd_translate as the preq.bdev |
01f37f2d KRW |
1288 | * is set there. |
1289 | */ | |
e9350493 KRW |
1290 | for (i = 0; i < nseg; i++) { |
1291 | if (((int)preq.sector_number|(int)seg[i].nsec) & | |
1292 | ((bdev_logical_block_size(preq.bdev) >> 9) - 1)) { | |
77387b82 | 1293 | pr_debug("Misaligned I/O request from domain %d\n", |
ebe81906 | 1294 | blkif->domid); |
e9350493 KRW |
1295 | goto fail_response; |
1296 | } | |
1297 | } | |
01f37f2d | 1298 | |
29bde093 KRW |
1299 | /* Wait on all outstanding I/O's and once that has been completed |
1300 | * issue the WRITE_FLUSH. | |
1301 | */ | |
1302 | if (drain) | |
1303 | xen_blk_drain_io(pending_req->blkif); | |
1304 | ||
01f37f2d KRW |
1305 | /* |
1306 | * If we have failed at this point, we need to undo the M2P override, | |
2e9977c2 KRW |
1307 | * set gnttab_set_unmap_op on all of the grant references and perform |
1308 | * the hypercall to unmap the grants - that is all done in | |
9f3aedf5 | 1309 | * xen_blkbk_unmap. |
2e9977c2 | 1310 | */ |
bb642e83 | 1311 | if (xen_blkbk_map_seg(pending_req)) |
4d05a28d KRW |
1312 | goto fail_flush; |
1313 | ||
b3cb0d6a LD |
1314 | /* |
1315 | * This corresponding xen_blkif_put is done in __end_block_io_op, or | |
1316 | * below (in "!bio") if we are handling a BLKIF_OP_DISCARD. | |
1317 | */ | |
8b6bf747 | 1318 | xen_blkif_get(blkif); |
c05f3e3c | 1319 | atomic_inc(&blkif->inflight); |
4d05a28d KRW |
1320 | |
1321 | for (i = 0; i < nseg; i++) { | |
4d05a28d KRW |
1322 | while ((bio == NULL) || |
1323 | (bio_add_page(bio, | |
bb642e83 | 1324 | pages[i]->page, |
4d05a28d | 1325 | seg[i].nsec << 9, |
ffb1dabd | 1326 | seg[i].offset) == 0)) { |
2e9977c2 | 1327 | |
1e0f7a21 RPM |
1328 | int nr_iovecs = min_t(int, (nseg-i), BIO_MAX_PAGES); |
1329 | bio = bio_alloc(GFP_KERNEL, nr_iovecs); | |
4d05a28d KRW |
1330 | if (unlikely(bio == NULL)) |
1331 | goto fail_put_bio; | |
1332 | ||
03e0edf9 | 1333 | biolist[nbio++] = bio; |
4d05a28d KRW |
1334 | bio->bi_bdev = preq.bdev; |
1335 | bio->bi_private = pending_req; | |
1336 | bio->bi_end_io = end_block_io_op; | |
4f024f37 | 1337 | bio->bi_iter.bi_sector = preq.sector_number; |
4d05a28d KRW |
1338 | } |
1339 | ||
1340 | preq.sector_number += seg[i].nsec; | |
1341 | } | |
1342 | ||
b3cb0d6a | 1343 | /* This will be hit if the operation was a flush or discard. */ |
4d05a28d | 1344 | if (!bio) { |
42146352 | 1345 | BUG_ON(operation != WRITE_FLUSH); |
b0f80127 | 1346 | |
42146352 KRW |
1347 | bio = bio_alloc(GFP_KERNEL, 0); |
1348 | if (unlikely(bio == NULL)) | |
1349 | goto fail_put_bio; | |
4d05a28d | 1350 | |
42146352 KRW |
1351 | biolist[nbio++] = bio; |
1352 | bio->bi_bdev = preq.bdev; | |
1353 | bio->bi_private = pending_req; | |
1354 | bio->bi_end_io = end_block_io_op; | |
4d05a28d KRW |
1355 | } |
1356 | ||
77089926 | 1357 | atomic_set(&pending_req->pendcnt, nbio); |
a19be5f0 KRW |
1358 | blk_start_plug(&plug); |
1359 | ||
77089926 KRW |
1360 | for (i = 0; i < nbio; i++) |
1361 | submit_bio(operation, biolist[i]); | |
1362 | ||
a19be5f0 | 1363 | /* Let the I/Os go.. */ |
3d68b399 | 1364 | blk_finish_plug(&plug); |
a19be5f0 | 1365 | |
4d05a28d KRW |
1366 | if (operation == READ) |
1367 | blkif->st_rd_sect += preq.nr_sects; | |
5c62cb48 | 1368 | else if (operation & WRITE) |
4d05a28d KRW |
1369 | blkif->st_wr_sect += preq.nr_sects; |
1370 | ||
fc53bf75 | 1371 | return 0; |
4d05a28d KRW |
1372 | |
1373 | fail_flush: | |
bb642e83 | 1374 | xen_blkbk_unmap(blkif, pending_req->segments, |
31552ee3 | 1375 | pending_req->nr_pages); |
4d05a28d | 1376 | fail_response: |
0faa8cca | 1377 | /* Haven't submitted any bio's yet. */ |
402b27f9 | 1378 | make_response(blkif, req->u.rw.id, req_operation, BLKIF_RSP_ERROR); |
bf0720c4 | 1379 | free_req(blkif, pending_req); |
4d05a28d | 1380 | msleep(1); /* back off a bit */ |
fc53bf75 | 1381 | return -EIO; |
4d05a28d KRW |
1382 | |
1383 | fail_put_bio: | |
03e0edf9 | 1384 | for (i = 0; i < nbio; i++) |
77089926 | 1385 | bio_put(biolist[i]); |
0e5e098a | 1386 | atomic_set(&pending_req->pendcnt, 1); |
4d05a28d | 1387 | __end_block_io_op(pending_req, -EINVAL); |
4d05a28d | 1388 | msleep(1); /* back off a bit */ |
fc53bf75 | 1389 | return -EIO; |
4d05a28d KRW |
1390 | } |
1391 | ||
1392 | ||
1393 | ||
a1397fa3 KRW |
1394 | /* |
1395 | * Put a response on the ring on how the operation fared. | |
4d05a28d | 1396 | */ |
30fd1502 | 1397 | static void make_response(struct xen_blkif *blkif, u64 id, |
4d05a28d KRW |
1398 | unsigned short op, int st) |
1399 | { | |
88122933 | 1400 | struct blkif_response resp; |
4d05a28d | 1401 | unsigned long flags; |
88122933 | 1402 | union blkif_back_rings *blk_rings = &blkif->blk_rings; |
4d05a28d KRW |
1403 | int notify; |
1404 | ||
1405 | resp.id = id; | |
1406 | resp.operation = op; | |
1407 | resp.status = st; | |
1408 | ||
1409 | spin_lock_irqsave(&blkif->blk_ring_lock, flags); | |
1410 | /* Place on the response ring for the relevant domain. */ | |
1411 | switch (blkif->blk_protocol) { | |
1412 | case BLKIF_PROTOCOL_NATIVE: | |
1413 | memcpy(RING_GET_RESPONSE(&blk_rings->native, blk_rings->native.rsp_prod_pvt), | |
1414 | &resp, sizeof(resp)); | |
1415 | break; | |
1416 | case BLKIF_PROTOCOL_X86_32: | |
1417 | memcpy(RING_GET_RESPONSE(&blk_rings->x86_32, blk_rings->x86_32.rsp_prod_pvt), | |
1418 | &resp, sizeof(resp)); | |
1419 | break; | |
1420 | case BLKIF_PROTOCOL_X86_64: | |
1421 | memcpy(RING_GET_RESPONSE(&blk_rings->x86_64, blk_rings->x86_64.rsp_prod_pvt), | |
1422 | &resp, sizeof(resp)); | |
1423 | break; | |
1424 | default: | |
1425 | BUG(); | |
1426 | } | |
1427 | blk_rings->common.rsp_prod_pvt++; | |
1428 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blk_rings->common, notify); | |
4d05a28d | 1429 | spin_unlock_irqrestore(&blkif->blk_ring_lock, flags); |
4d05a28d KRW |
1430 | if (notify) |
1431 | notify_remote_via_irq(blkif->irq); | |
1432 | } | |
1433 | ||
8b6bf747 | 1434 | static int __init xen_blkif_init(void) |
4d05a28d | 1435 | { |
8770b268 | 1436 | int rc = 0; |
4d05a28d | 1437 | |
b2167ba6 | 1438 | if (!xen_domain()) |
4d05a28d KRW |
1439 | return -ENODEV; |
1440 | ||
8b6bf747 | 1441 | rc = xen_blkif_interface_init(); |
8770b268 KRW |
1442 | if (rc) |
1443 | goto failed_init; | |
4d05a28d | 1444 | |
8b6bf747 | 1445 | rc = xen_blkif_xenbus_init(); |
8770b268 KRW |
1446 | if (rc) |
1447 | goto failed_init; | |
4d05a28d | 1448 | |
8770b268 | 1449 | failed_init: |
8770b268 | 1450 | return rc; |
4d05a28d KRW |
1451 | } |
1452 | ||
8b6bf747 | 1453 | module_init(xen_blkif_init); |
4d05a28d KRW |
1454 | |
1455 | MODULE_LICENSE("Dual BSD/GPL"); | |
a7e9357f | 1456 | MODULE_ALIAS("xen-backend:vbd"); |