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 | ||
973e5405 JG |
86 | /* |
87 | * How long a persistent grant is allowed to remain allocated without being in | |
88 | * use. The time is in seconds, 0 means indefinitely long. | |
89 | */ | |
90 | ||
91 | static unsigned int xen_blkif_pgrant_timeout = 60; | |
92 | module_param_named(persistent_grant_unused_seconds, xen_blkif_pgrant_timeout, | |
93 | uint, 0644); | |
94 | MODULE_PARM_DESC(persistent_grant_unused_seconds, | |
95 | "Time in seconds an unused persistent grant is allowed to " | |
96 | "remain allocated. Default is 60, 0 means unlimited."); | |
97 | ||
d62d8600 BL |
98 | /* |
99 | * Maximum number of rings/queues blkback supports, allow as many queues as there | |
100 | * are CPUs if user has not specified a value. | |
101 | */ | |
102 | unsigned int xenblk_max_queues; | |
103 | module_param_named(max_queues, xenblk_max_queues, uint, 0644); | |
104 | MODULE_PARM_DESC(max_queues, | |
105 | "Maximum number of hardware queues per virtual disk." \ | |
106 | "By default it is the number of online CPUs."); | |
107 | ||
86839c56 BL |
108 | /* |
109 | * Maximum order of pages to be used for the shared ring between front and | |
110 | * backend, 4KB page granularity is used. | |
111 | */ | |
9cce2914 | 112 | unsigned int xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER; |
5657a819 | 113 | module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, 0444); |
86839c56 | 114 | MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring"); |
3f3aad5e RPM |
115 | /* |
116 | * The LRU mechanism to clean the lists of persistent grants needs to | |
117 | * be executed periodically. The time interval between consecutive executions | |
118 | * of the purge mechanism is set in ms. | |
119 | */ | |
120 | #define LRU_INTERVAL 100 | |
121 | ||
122 | /* | |
123 | * When the persistent grants list is full we will remove unused grants | |
124 | * from the list. The percent number of grants to be removed at each LRU | |
125 | * execution. | |
126 | */ | |
127 | #define LRU_PERCENT_CLEAN 5 | |
128 | ||
4d05a28d | 129 | /* Run-time switchable: /sys/module/blkback/parameters/ */ |
2e9977c2 | 130 | static unsigned int log_stats; |
4d05a28d | 131 | module_param(log_stats, int, 0644); |
4d05a28d | 132 | |
4d05a28d KRW |
133 | #define BLKBACK_INVALID_HANDLE (~0) |
134 | ||
ff4b156f | 135 | /* Number of free pages to remove on each call to gnttab_free_pages */ |
c6cc142d RPM |
136 | #define NUM_BATCH_FREE_PAGES 10 |
137 | ||
973e5405 JG |
138 | static inline bool persistent_gnt_timeout(struct persistent_gnt *persistent_gnt) |
139 | { | |
140 | return xen_blkif_pgrant_timeout && | |
141 | (jiffies - persistent_gnt->last_used >= | |
142 | HZ * xen_blkif_pgrant_timeout); | |
143 | } | |
144 | ||
d4bf0065 | 145 | static inline int get_free_page(struct xen_blkif_ring *ring, struct page **page) |
c6cc142d RPM |
146 | { |
147 | unsigned long flags; | |
148 | ||
d4bf0065 BL |
149 | spin_lock_irqsave(&ring->free_pages_lock, flags); |
150 | if (list_empty(&ring->free_pages)) { | |
151 | BUG_ON(ring->free_pages_num != 0); | |
152 | spin_unlock_irqrestore(&ring->free_pages_lock, flags); | |
ff4b156f | 153 | return gnttab_alloc_pages(1, page); |
c6cc142d | 154 | } |
d4bf0065 BL |
155 | BUG_ON(ring->free_pages_num == 0); |
156 | page[0] = list_first_entry(&ring->free_pages, struct page, lru); | |
c6cc142d | 157 | list_del(&page[0]->lru); |
d4bf0065 BL |
158 | ring->free_pages_num--; |
159 | spin_unlock_irqrestore(&ring->free_pages_lock, flags); | |
efe08a3e | 160 | |
c6cc142d RPM |
161 | return 0; |
162 | } | |
163 | ||
d4bf0065 | 164 | static inline void put_free_pages(struct xen_blkif_ring *ring, struct page **page, |
c6cc142d | 165 | int num) |
4d05a28d | 166 | { |
c6cc142d RPM |
167 | unsigned long flags; |
168 | int i; | |
169 | ||
d4bf0065 | 170 | spin_lock_irqsave(&ring->free_pages_lock, flags); |
c6cc142d | 171 | for (i = 0; i < num; i++) |
d4bf0065 BL |
172 | list_add(&page[i]->lru, &ring->free_pages); |
173 | ring->free_pages_num += num; | |
174 | spin_unlock_irqrestore(&ring->free_pages_lock, flags); | |
c6cc142d RPM |
175 | } |
176 | ||
d4bf0065 | 177 | static inline void shrink_free_pagepool(struct xen_blkif_ring *ring, int num) |
c6cc142d RPM |
178 | { |
179 | /* Remove requested pages in batches of NUM_BATCH_FREE_PAGES */ | |
180 | struct page *page[NUM_BATCH_FREE_PAGES]; | |
181 | unsigned int num_pages = 0; | |
182 | unsigned long flags; | |
183 | ||
d4bf0065 BL |
184 | spin_lock_irqsave(&ring->free_pages_lock, flags); |
185 | while (ring->free_pages_num > num) { | |
186 | BUG_ON(list_empty(&ring->free_pages)); | |
187 | page[num_pages] = list_first_entry(&ring->free_pages, | |
c6cc142d RPM |
188 | struct page, lru); |
189 | list_del(&page[num_pages]->lru); | |
d4bf0065 | 190 | ring->free_pages_num--; |
c6cc142d | 191 | if (++num_pages == NUM_BATCH_FREE_PAGES) { |
d4bf0065 | 192 | spin_unlock_irqrestore(&ring->free_pages_lock, flags); |
ff4b156f | 193 | gnttab_free_pages(num_pages, page); |
d4bf0065 | 194 | spin_lock_irqsave(&ring->free_pages_lock, flags); |
c6cc142d RPM |
195 | num_pages = 0; |
196 | } | |
197 | } | |
d4bf0065 | 198 | spin_unlock_irqrestore(&ring->free_pages_lock, flags); |
c6cc142d | 199 | if (num_pages != 0) |
ff4b156f | 200 | gnttab_free_pages(num_pages, page); |
4d05a28d KRW |
201 | } |
202 | ||
c6cc142d RPM |
203 | #define vaddr(page) ((unsigned long)pfn_to_kaddr(page_to_pfn(page))) |
204 | ||
59795700 BL |
205 | static int do_block_io_op(struct xen_blkif_ring *ring); |
206 | static int dispatch_rw_block_io(struct xen_blkif_ring *ring, | |
fc53bf75 KRW |
207 | struct blkif_request *req, |
208 | struct pending_req *pending_req); | |
59795700 | 209 | static void make_response(struct xen_blkif_ring *ring, u64 id, |
4d05a28d KRW |
210 | unsigned short op, int st); |
211 | ||
7dc34117 RPM |
212 | #define foreach_grant_safe(pos, n, rbtree, node) \ |
213 | for ((pos) = container_of(rb_first((rbtree)), typeof(*(pos)), node), \ | |
217fd5e7 | 214 | (n) = (&(pos)->node != NULL) ? rb_next(&(pos)->node) : NULL; \ |
0a8704a5 | 215 | &(pos)->node != NULL; \ |
7dc34117 RPM |
216 | (pos) = container_of(n, typeof(*(pos)), node), \ |
217 | (n) = (&(pos)->node != NULL) ? rb_next(&(pos)->node) : NULL) | |
0a8704a5 RPM |
218 | |
219 | ||
3f3aad5e RPM |
220 | /* |
221 | * We don't need locking around the persistent grant helpers | |
d4bf0065 | 222 | * because blkback uses a single-thread for each backend, so we |
3f3aad5e RPM |
223 | * can be sure that this functions will never be called recursively. |
224 | * | |
225 | * The only exception to that is put_persistent_grant, that can be called | |
226 | * from interrupt context (by xen_blkbk_unmap), so we have to use atomic | |
227 | * bit operations to modify the flags of a persistent grant and to count | |
228 | * the number of used grants. | |
229 | */ | |
d4bf0065 | 230 | static int add_persistent_gnt(struct xen_blkif_ring *ring, |
0a8704a5 RPM |
231 | struct persistent_gnt *persistent_gnt) |
232 | { | |
3f3aad5e | 233 | struct rb_node **new = NULL, *parent = NULL; |
0a8704a5 | 234 | struct persistent_gnt *this; |
d4bf0065 | 235 | struct xen_blkif *blkif = ring->blkif; |
0a8704a5 | 236 | |
d4bf0065 | 237 | if (ring->persistent_gnt_c >= xen_blkif_max_pgrants) { |
3f3aad5e RPM |
238 | if (!blkif->vbd.overflow_max_grants) |
239 | blkif->vbd.overflow_max_grants = 1; | |
240 | return -EBUSY; | |
241 | } | |
0a8704a5 | 242 | /* Figure out where to put new node */ |
d4bf0065 | 243 | new = &ring->persistent_gnts.rb_node; |
0a8704a5 RPM |
244 | while (*new) { |
245 | this = container_of(*new, struct persistent_gnt, node); | |
246 | ||
247 | parent = *new; | |
248 | if (persistent_gnt->gnt < this->gnt) | |
249 | new = &((*new)->rb_left); | |
250 | else if (persistent_gnt->gnt > this->gnt) | |
251 | new = &((*new)->rb_right); | |
252 | else { | |
77387b82 | 253 | pr_alert_ratelimited("trying to add a gref that's already in the tree\n"); |
c6cc142d | 254 | return -EINVAL; |
0a8704a5 RPM |
255 | } |
256 | } | |
257 | ||
d77ff24e | 258 | persistent_gnt->active = true; |
0a8704a5 RPM |
259 | /* Add new node and rebalance tree. */ |
260 | rb_link_node(&(persistent_gnt->node), parent, new); | |
d4bf0065 BL |
261 | rb_insert_color(&(persistent_gnt->node), &ring->persistent_gnts); |
262 | ring->persistent_gnt_c++; | |
263 | atomic_inc(&ring->persistent_gnt_in_use); | |
c6cc142d | 264 | return 0; |
0a8704a5 RPM |
265 | } |
266 | ||
d4bf0065 | 267 | static struct persistent_gnt *get_persistent_gnt(struct xen_blkif_ring *ring, |
0a8704a5 RPM |
268 | grant_ref_t gref) |
269 | { | |
270 | struct persistent_gnt *data; | |
3f3aad5e | 271 | struct rb_node *node = NULL; |
0a8704a5 | 272 | |
d4bf0065 | 273 | node = ring->persistent_gnts.rb_node; |
0a8704a5 RPM |
274 | while (node) { |
275 | data = container_of(node, struct persistent_gnt, node); | |
276 | ||
277 | if (gref < data->gnt) | |
278 | node = node->rb_left; | |
279 | else if (gref > data->gnt) | |
280 | node = node->rb_right; | |
3f3aad5e | 281 | else { |
d77ff24e | 282 | if (data->active) { |
77387b82 | 283 | pr_alert_ratelimited("requesting a grant already in use\n"); |
3f3aad5e RPM |
284 | return NULL; |
285 | } | |
d77ff24e | 286 | data->active = true; |
d4bf0065 | 287 | atomic_inc(&ring->persistent_gnt_in_use); |
0a8704a5 | 288 | return data; |
3f3aad5e | 289 | } |
0a8704a5 RPM |
290 | } |
291 | return NULL; | |
292 | } | |
293 | ||
d4bf0065 | 294 | static void put_persistent_gnt(struct xen_blkif_ring *ring, |
3f3aad5e RPM |
295 | struct persistent_gnt *persistent_gnt) |
296 | { | |
d77ff24e | 297 | if (!persistent_gnt->active) |
77387b82 | 298 | pr_alert_ratelimited("freeing a grant already unused\n"); |
973e5405 | 299 | persistent_gnt->last_used = jiffies; |
d77ff24e | 300 | persistent_gnt->active = false; |
d4bf0065 | 301 | atomic_dec(&ring->persistent_gnt_in_use); |
3f3aad5e RPM |
302 | } |
303 | ||
d4bf0065 | 304 | static void free_persistent_gnts(struct xen_blkif_ring *ring, struct rb_root *root, |
c6cc142d | 305 | unsigned int num) |
4d4f270f RPM |
306 | { |
307 | struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
308 | struct page *pages[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
309 | struct persistent_gnt *persistent_gnt; | |
7dc34117 | 310 | struct rb_node *n; |
4d4f270f | 311 | int segs_to_unmap = 0; |
c43cf3ea | 312 | struct gntab_unmap_queue_data unmap_data; |
c43cf3ea | 313 | |
c43cf3ea JH |
314 | unmap_data.pages = pages; |
315 | unmap_data.unmap_ops = unmap; | |
316 | unmap_data.kunmap_ops = NULL; | |
4d4f270f | 317 | |
7dc34117 | 318 | foreach_grant_safe(persistent_gnt, n, root, node) { |
4d4f270f RPM |
319 | BUG_ON(persistent_gnt->handle == |
320 | BLKBACK_INVALID_HANDLE); | |
321 | gnttab_set_unmap_op(&unmap[segs_to_unmap], | |
322 | (unsigned long) pfn_to_kaddr(page_to_pfn( | |
323 | persistent_gnt->page)), | |
324 | GNTMAP_host_map, | |
325 | persistent_gnt->handle); | |
326 | ||
327 | pages[segs_to_unmap] = persistent_gnt->page; | |
4d4f270f RPM |
328 | |
329 | if (++segs_to_unmap == BLKIF_MAX_SEGMENTS_PER_REQUEST || | |
330 | !rb_next(&persistent_gnt->node)) { | |
c43cf3ea JH |
331 | |
332 | unmap_data.count = segs_to_unmap; | |
b44166cd | 333 | BUG_ON(gnttab_unmap_refs_sync(&unmap_data)); |
c43cf3ea | 334 | |
d4bf0065 | 335 | put_free_pages(ring, pages, segs_to_unmap); |
4d4f270f RPM |
336 | segs_to_unmap = 0; |
337 | } | |
7dc34117 RPM |
338 | |
339 | rb_erase(&persistent_gnt->node, root); | |
340 | kfree(persistent_gnt); | |
341 | num--; | |
4d4f270f RPM |
342 | } |
343 | BUG_ON(num != 0); | |
344 | } | |
345 | ||
abb97b8c | 346 | void xen_blkbk_unmap_purged_grants(struct work_struct *work) |
3f3aad5e RPM |
347 | { |
348 | struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
349 | struct page *pages[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
350 | struct persistent_gnt *persistent_gnt; | |
325d73bf | 351 | int segs_to_unmap = 0; |
d4bf0065 | 352 | struct xen_blkif_ring *ring = container_of(work, typeof(*ring), persistent_purge_work); |
325d73bf | 353 | struct gntab_unmap_queue_data unmap_data; |
325d73bf | 354 | |
325d73bf BL |
355 | unmap_data.pages = pages; |
356 | unmap_data.unmap_ops = unmap; | |
357 | unmap_data.kunmap_ops = NULL; | |
3f3aad5e | 358 | |
d4bf0065 BL |
359 | while(!list_empty(&ring->persistent_purge_list)) { |
360 | persistent_gnt = list_first_entry(&ring->persistent_purge_list, | |
3f3aad5e RPM |
361 | struct persistent_gnt, |
362 | remove_node); | |
363 | list_del(&persistent_gnt->remove_node); | |
364 | ||
365 | gnttab_set_unmap_op(&unmap[segs_to_unmap], | |
366 | vaddr(persistent_gnt->page), | |
367 | GNTMAP_host_map, | |
368 | persistent_gnt->handle); | |
369 | ||
370 | pages[segs_to_unmap] = persistent_gnt->page; | |
371 | ||
372 | if (++segs_to_unmap == BLKIF_MAX_SEGMENTS_PER_REQUEST) { | |
325d73bf | 373 | unmap_data.count = segs_to_unmap; |
b44166cd | 374 | BUG_ON(gnttab_unmap_refs_sync(&unmap_data)); |
d4bf0065 | 375 | put_free_pages(ring, pages, segs_to_unmap); |
3f3aad5e RPM |
376 | segs_to_unmap = 0; |
377 | } | |
378 | kfree(persistent_gnt); | |
379 | } | |
380 | if (segs_to_unmap > 0) { | |
325d73bf | 381 | unmap_data.count = segs_to_unmap; |
b44166cd | 382 | BUG_ON(gnttab_unmap_refs_sync(&unmap_data)); |
d4bf0065 | 383 | put_free_pages(ring, pages, segs_to_unmap); |
3f3aad5e RPM |
384 | } |
385 | } | |
386 | ||
d4bf0065 | 387 | static void purge_persistent_gnt(struct xen_blkif_ring *ring) |
3f3aad5e RPM |
388 | { |
389 | struct persistent_gnt *persistent_gnt; | |
390 | struct rb_node *n; | |
391 | unsigned int num_clean, total; | |
973e5405 | 392 | bool scan_used = false; |
3f3aad5e RPM |
393 | struct rb_root *root; |
394 | ||
d4bf0065 | 395 | if (work_busy(&ring->persistent_purge_work)) { |
53bc7dc0 | 396 | pr_alert_ratelimited("Scheduled work from previous purge is still busy, cannot purge list\n"); |
59795700 | 397 | goto out; |
3f3aad5e RPM |
398 | } |
399 | ||
973e5405 JG |
400 | if (ring->persistent_gnt_c < xen_blkif_max_pgrants || |
401 | (ring->persistent_gnt_c == xen_blkif_max_pgrants && | |
402 | !ring->blkif->vbd.overflow_max_grants)) { | |
403 | num_clean = 0; | |
404 | } else { | |
405 | num_clean = (xen_blkif_max_pgrants / 100) * LRU_PERCENT_CLEAN; | |
406 | num_clean = ring->persistent_gnt_c - xen_blkif_max_pgrants + | |
407 | num_clean; | |
408 | num_clean = min(ring->persistent_gnt_c, num_clean); | |
409 | pr_debug("Going to purge at least %u persistent grants\n", | |
410 | num_clean); | |
411 | } | |
3f3aad5e RPM |
412 | |
413 | /* | |
414 | * At this point, we can assure that there will be no calls | |
415 | * to get_persistent_grant (because we are executing this code from | |
416 | * xen_blkif_schedule), there can only be calls to put_persistent_gnt, | |
417 | * which means that the number of currently used grants will go down, | |
418 | * but never up, so we will always be able to remove the requested | |
419 | * number of grants. | |
420 | */ | |
421 | ||
973e5405 | 422 | total = 0; |
3f3aad5e | 423 | |
d4bf0065 BL |
424 | BUG_ON(!list_empty(&ring->persistent_purge_list)); |
425 | root = &ring->persistent_gnts; | |
3f3aad5e RPM |
426 | purge_list: |
427 | foreach_grant_safe(persistent_gnt, n, root, node) { | |
428 | BUG_ON(persistent_gnt->handle == | |
429 | BLKBACK_INVALID_HANDLE); | |
430 | ||
d77ff24e | 431 | if (persistent_gnt->active) |
3f3aad5e | 432 | continue; |
973e5405 JG |
433 | if (!scan_used && !persistent_gnt_timeout(persistent_gnt)) |
434 | continue; | |
435 | if (scan_used && total >= num_clean) | |
3f3aad5e RPM |
436 | continue; |
437 | ||
438 | rb_erase(&persistent_gnt->node, root); | |
439 | list_add(&persistent_gnt->remove_node, | |
d4bf0065 | 440 | &ring->persistent_purge_list); |
973e5405 | 441 | total++; |
3f3aad5e RPM |
442 | } |
443 | /* | |
973e5405 | 444 | * Check whether we also need to start cleaning |
3f3aad5e RPM |
445 | * grants that were used since last purge in order to cope |
446 | * with the requested num | |
447 | */ | |
973e5405 JG |
448 | if (!scan_used && total < num_clean) { |
449 | pr_debug("Still missing %u purged frames\n", num_clean - total); | |
3f3aad5e RPM |
450 | scan_used = true; |
451 | goto purge_list; | |
452 | } | |
2d910543 | 453 | |
973e5405 JG |
454 | if (total) { |
455 | ring->persistent_gnt_c -= total; | |
456 | ring->blkif->vbd.overflow_max_grants = 0; | |
3f3aad5e | 457 | |
973e5405 JG |
458 | /* We can defer this work */ |
459 | schedule_work(&ring->persistent_purge_work); | |
460 | pr_debug("Purged %u/%u\n", num_clean, total); | |
461 | } | |
59795700 BL |
462 | |
463 | out: | |
3f3aad5e RPM |
464 | return; |
465 | } | |
466 | ||
a1397fa3 KRW |
467 | /* |
468 | * Retrieve from the 'pending_reqs' a free pending_req structure to be used. | |
4d05a28d | 469 | */ |
59795700 | 470 | static struct pending_req *alloc_req(struct xen_blkif_ring *ring) |
4d05a28d | 471 | { |
2e9977c2 | 472 | struct pending_req *req = NULL; |
4d05a28d KRW |
473 | unsigned long flags; |
474 | ||
59795700 BL |
475 | spin_lock_irqsave(&ring->pending_free_lock, flags); |
476 | if (!list_empty(&ring->pending_free)) { | |
477 | req = list_entry(ring->pending_free.next, struct pending_req, | |
2e9977c2 | 478 | free_list); |
4d05a28d KRW |
479 | list_del(&req->free_list); |
480 | } | |
59795700 | 481 | spin_unlock_irqrestore(&ring->pending_free_lock, flags); |
4d05a28d KRW |
482 | return req; |
483 | } | |
484 | ||
a1397fa3 KRW |
485 | /* |
486 | * Return the 'pending_req' structure back to the freepool. We also | |
487 | * wake up the thread if it was waiting for a free page. | |
488 | */ | |
59795700 | 489 | static void free_req(struct xen_blkif_ring *ring, struct pending_req *req) |
4d05a28d KRW |
490 | { |
491 | unsigned long flags; | |
492 | int was_empty; | |
493 | ||
59795700 BL |
494 | spin_lock_irqsave(&ring->pending_free_lock, flags); |
495 | was_empty = list_empty(&ring->pending_free); | |
496 | list_add(&req->free_list, &ring->pending_free); | |
497 | spin_unlock_irqrestore(&ring->pending_free_lock, flags); | |
4d05a28d | 498 | if (was_empty) |
59795700 | 499 | wake_up(&ring->pending_free_wq); |
4d05a28d KRW |
500 | } |
501 | ||
ee9ff853 KRW |
502 | /* |
503 | * Routines for managing virtual block devices (vbds). | |
504 | */ | |
3d814731 KRW |
505 | static int xen_vbd_translate(struct phys_req *req, struct xen_blkif *blkif, |
506 | int operation) | |
ee9ff853 | 507 | { |
3d814731 | 508 | struct xen_vbd *vbd = &blkif->vbd; |
ee9ff853 KRW |
509 | int rc = -EACCES; |
510 | ||
a022606e | 511 | if ((operation != REQ_OP_READ) && vbd->readonly) |
ee9ff853 KRW |
512 | goto out; |
513 | ||
8ab52150 JB |
514 | if (likely(req->nr_sects)) { |
515 | blkif_sector_t end = req->sector_number + req->nr_sects; | |
516 | ||
517 | if (unlikely(end < req->sector_number)) | |
518 | goto out; | |
519 | if (unlikely(end > vbd_sz(vbd))) | |
520 | goto out; | |
521 | } | |
ee9ff853 KRW |
522 | |
523 | req->dev = vbd->pdevice; | |
524 | req->bdev = vbd->bdev; | |
525 | rc = 0; | |
526 | ||
527 | out: | |
528 | return rc; | |
529 | } | |
530 | ||
3d814731 | 531 | static void xen_vbd_resize(struct xen_blkif *blkif) |
ee9ff853 | 532 | { |
3d814731 | 533 | struct xen_vbd *vbd = &blkif->vbd; |
ee9ff853 KRW |
534 | struct xenbus_transaction xbt; |
535 | int err; | |
8b6bf747 | 536 | struct xenbus_device *dev = xen_blkbk_xenbus(blkif->be); |
42c7841d | 537 | unsigned long long new_size = vbd_sz(vbd); |
ee9ff853 | 538 | |
77387b82 | 539 | pr_info("VBD Resize: Domid: %d, Device: (%d, %d)\n", |
ee9ff853 | 540 | blkif->domid, MAJOR(vbd->pdevice), MINOR(vbd->pdevice)); |
77387b82 | 541 | pr_info("VBD Resize: new size %llu\n", new_size); |
ee9ff853 KRW |
542 | vbd->size = new_size; |
543 | again: | |
544 | err = xenbus_transaction_start(&xbt); | |
545 | if (err) { | |
77387b82 | 546 | pr_warn("Error starting transaction\n"); |
ee9ff853 KRW |
547 | return; |
548 | } | |
549 | err = xenbus_printf(xbt, dev->nodename, "sectors", "%llu", | |
42c7841d | 550 | (unsigned long long)vbd_sz(vbd)); |
ee9ff853 | 551 | if (err) { |
77387b82 | 552 | pr_warn("Error writing new size\n"); |
ee9ff853 KRW |
553 | goto abort; |
554 | } | |
555 | /* | |
556 | * Write the current state; we will use this to synchronize | |
557 | * the front-end. If the current state is "connected" the | |
558 | * front-end will get the new size information online. | |
559 | */ | |
560 | err = xenbus_printf(xbt, dev->nodename, "state", "%d", dev->state); | |
561 | if (err) { | |
77387b82 | 562 | pr_warn("Error writing the state\n"); |
ee9ff853 KRW |
563 | goto abort; |
564 | } | |
565 | ||
566 | err = xenbus_transaction_end(xbt, 0); | |
567 | if (err == -EAGAIN) | |
568 | goto again; | |
569 | if (err) | |
77387b82 | 570 | pr_warn("Error ending transaction\n"); |
496b318e | 571 | return; |
ee9ff853 KRW |
572 | abort: |
573 | xenbus_transaction_end(xbt, 1); | |
574 | } | |
575 | ||
a1397fa3 | 576 | /* |
b0aef179 KRW |
577 | * Notification from the guest OS. |
578 | */ | |
59795700 | 579 | static void blkif_notify_work(struct xen_blkif_ring *ring) |
4d05a28d | 580 | { |
59795700 BL |
581 | ring->waiting_reqs = 1; |
582 | wake_up(&ring->wq); | |
b0aef179 | 583 | } |
4d05a28d | 584 | |
8b6bf747 | 585 | irqreturn_t xen_blkif_be_int(int irq, void *dev_id) |
b0aef179 KRW |
586 | { |
587 | blkif_notify_work(dev_id); | |
588 | return IRQ_HANDLED; | |
4d05a28d KRW |
589 | } |
590 | ||
2e9977c2 | 591 | /* |
4d05a28d KRW |
592 | * SCHEDULER FUNCTIONS |
593 | */ | |
594 | ||
d4bf0065 | 595 | static void print_stats(struct xen_blkif_ring *ring) |
4d05a28d | 596 | { |
77387b82 | 597 | pr_info("(%s): oo %3llu | rd %4llu | wr %4llu | f %4llu" |
3f3aad5e | 598 | " | ds %4llu | pg: %4u/%4d\n", |
db6fbc10 BL |
599 | current->comm, ring->st_oo_req, |
600 | ring->st_rd_req, ring->st_wr_req, | |
601 | ring->st_f_req, ring->st_ds_req, | |
d4bf0065 | 602 | ring->persistent_gnt_c, |
3f3aad5e | 603 | xen_blkif_max_pgrants); |
db6fbc10 BL |
604 | ring->st_print = jiffies + msecs_to_jiffies(10 * 1000); |
605 | ring->st_rd_req = 0; | |
606 | ring->st_wr_req = 0; | |
607 | ring->st_oo_req = 0; | |
608 | ring->st_ds_req = 0; | |
4d05a28d KRW |
609 | } |
610 | ||
8b6bf747 | 611 | int xen_blkif_schedule(void *arg) |
4d05a28d | 612 | { |
59795700 BL |
613 | struct xen_blkif_ring *ring = arg; |
614 | struct xen_blkif *blkif = ring->blkif; | |
3d814731 | 615 | struct xen_vbd *vbd = &blkif->vbd; |
3f3aad5e | 616 | unsigned long timeout; |
8e3f8755 | 617 | int ret; |
4d05a28d | 618 | |
a6e7af12 | 619 | set_freezable(); |
4d05a28d KRW |
620 | while (!kthread_should_stop()) { |
621 | if (try_to_freeze()) | |
622 | continue; | |
42c7841d | 623 | if (unlikely(vbd->size != vbd_sz(vbd))) |
3d814731 | 624 | xen_vbd_resize(blkif); |
4d05a28d | 625 | |
3f3aad5e RPM |
626 | timeout = msecs_to_jiffies(LRU_INTERVAL); |
627 | ||
628 | timeout = wait_event_interruptible_timeout( | |
59795700 BL |
629 | ring->wq, |
630 | ring->waiting_reqs || kthread_should_stop(), | |
3f3aad5e RPM |
631 | timeout); |
632 | if (timeout == 0) | |
633 | goto purge_gnt_list; | |
634 | timeout = wait_event_interruptible_timeout( | |
59795700 BL |
635 | ring->pending_free_wq, |
636 | !list_empty(&ring->pending_free) || | |
3f3aad5e RPM |
637 | kthread_should_stop(), |
638 | timeout); | |
639 | if (timeout == 0) | |
640 | goto purge_gnt_list; | |
4d05a28d | 641 | |
59795700 | 642 | ring->waiting_reqs = 0; |
4d05a28d KRW |
643 | smp_mb(); /* clear flag *before* checking for work */ |
644 | ||
59795700 | 645 | ret = do_block_io_op(ring); |
8e3f8755 | 646 | if (ret > 0) |
59795700 | 647 | ring->waiting_reqs = 1; |
8e3f8755 | 648 | if (ret == -EACCES) |
59795700 | 649 | wait_event_interruptible(ring->shutdown_wq, |
8e3f8755 | 650 | kthread_should_stop()); |
4d05a28d | 651 | |
3f3aad5e RPM |
652 | purge_gnt_list: |
653 | if (blkif->vbd.feature_gnt_persistent && | |
d4bf0065 BL |
654 | time_after(jiffies, ring->next_lru)) { |
655 | purge_persistent_gnt(ring); | |
656 | ring->next_lru = jiffies + msecs_to_jiffies(LRU_INTERVAL); | |
3f3aad5e RPM |
657 | } |
658 | ||
c6cc142d | 659 | /* Shrink if we have more than xen_blkif_max_buffer_pages */ |
d4bf0065 | 660 | shrink_free_pagepool(ring, xen_blkif_max_buffer_pages); |
c6cc142d | 661 | |
db6fbc10 | 662 | if (log_stats && time_after(jiffies, ring->st_print)) |
d4bf0065 | 663 | print_stats(ring); |
4d05a28d KRW |
664 | } |
665 | ||
ef753411 | 666 | /* Drain pending purge work */ |
d4bf0065 | 667 | flush_work(&ring->persistent_purge_work); |
c6cc142d | 668 | |
ef753411 | 669 | if (log_stats) |
d4bf0065 | 670 | print_stats(ring); |
ef753411 | 671 | |
59795700 | 672 | ring->xenblkd = NULL; |
ef753411 RPM |
673 | |
674 | return 0; | |
675 | } | |
676 | ||
677 | /* | |
678 | * Remove persistent grants and empty the pool of free pages | |
679 | */ | |
59795700 | 680 | void xen_blkbk_free_caches(struct xen_blkif_ring *ring) |
ef753411 | 681 | { |
0a8704a5 | 682 | /* Free all persistent grant pages */ |
d4bf0065 BL |
683 | if (!RB_EMPTY_ROOT(&ring->persistent_gnts)) |
684 | free_persistent_gnts(ring, &ring->persistent_gnts, | |
685 | ring->persistent_gnt_c); | |
0a8704a5 | 686 | |
d4bf0065 BL |
687 | BUG_ON(!RB_EMPTY_ROOT(&ring->persistent_gnts)); |
688 | ring->persistent_gnt_c = 0; | |
0a8704a5 | 689 | |
2ed22e3c | 690 | /* Since we are shutting down remove all pages from the buffer */ |
d4bf0065 | 691 | shrink_free_pagepool(ring, 0 /* All */); |
4d05a28d KRW |
692 | } |
693 | ||
c43cf3ea | 694 | static unsigned int xen_blkbk_unmap_prepare( |
59795700 | 695 | struct xen_blkif_ring *ring, |
c43cf3ea JH |
696 | struct grant_page **pages, |
697 | unsigned int num, | |
698 | struct gnttab_unmap_grant_ref *unmap_ops, | |
699 | struct page **unmap_pages) | |
b0aef179 | 700 | { |
b0aef179 | 701 | unsigned int i, invcount = 0; |
b0aef179 | 702 | |
31552ee3 | 703 | for (i = 0; i < num; i++) { |
bb642e83 | 704 | if (pages[i]->persistent_gnt != NULL) { |
d4bf0065 | 705 | put_persistent_gnt(ring, pages[i]->persistent_gnt); |
0a8704a5 | 706 | continue; |
3f3aad5e | 707 | } |
bb642e83 | 708 | if (pages[i]->handle == BLKBACK_INVALID_HANDLE) |
b0aef179 | 709 | continue; |
bb642e83 | 710 | unmap_pages[invcount] = pages[i]->page; |
c43cf3ea | 711 | gnttab_set_unmap_op(&unmap_ops[invcount], vaddr(pages[i]->page), |
bb642e83 RPM |
712 | GNTMAP_host_map, pages[i]->handle); |
713 | pages[i]->handle = BLKBACK_INVALID_HANDLE; | |
c43cf3ea | 714 | invcount++; |
306b82a8 | 715 | } |
c43cf3ea | 716 | |
306b82a8 | 717 | return invcount; |
c43cf3ea JH |
718 | } |
719 | ||
720 | static void xen_blkbk_unmap_and_respond_callback(int result, struct gntab_unmap_queue_data *data) | |
721 | { | |
59795700 BL |
722 | struct pending_req *pending_req = (struct pending_req *)(data->data); |
723 | struct xen_blkif_ring *ring = pending_req->ring; | |
724 | struct xen_blkif *blkif = ring->blkif; | |
c43cf3ea JH |
725 | |
726 | /* BUG_ON used to reproduce existing behaviour, | |
727 | but is this the best way to deal with this? */ | |
728 | BUG_ON(result); | |
729 | ||
d4bf0065 | 730 | put_free_pages(ring, data->pages, data->count); |
59795700 | 731 | make_response(ring, pending_req->id, |
c43cf3ea | 732 | pending_req->operation, pending_req->status); |
59795700 | 733 | free_req(ring, pending_req); |
c43cf3ea JH |
734 | /* |
735 | * Make sure the request is freed before releasing blkif, | |
736 | * or there could be a race between free_req and the | |
737 | * cleanup done in xen_blkif_free during shutdown. | |
738 | * | |
739 | * NB: The fact that we might try to wake up pending_free_wq | |
740 | * before drain_complete (in case there's a drain going on) | |
741 | * it's not a problem with our current implementation | |
742 | * because we can assure there's no thread waiting on | |
743 | * pending_free_wq if there's a drain going on, but it has | |
744 | * to be taken into account if the current model is changed. | |
745 | */ | |
59795700 | 746 | if (atomic_dec_and_test(&ring->inflight) && atomic_read(&blkif->drain)) { |
c43cf3ea JH |
747 | complete(&blkif->drain_complete); |
748 | } | |
749 | xen_blkif_put(blkif); | |
750 | } | |
751 | ||
752 | static void xen_blkbk_unmap_and_respond(struct pending_req *req) | |
753 | { | |
754 | struct gntab_unmap_queue_data* work = &req->gnttab_unmap_data; | |
59795700 | 755 | struct xen_blkif_ring *ring = req->ring; |
c43cf3ea JH |
756 | struct grant_page **pages = req->segments; |
757 | unsigned int invcount; | |
758 | ||
59795700 | 759 | invcount = xen_blkbk_unmap_prepare(ring, pages, req->nr_segs, |
c43cf3ea JH |
760 | req->unmap, req->unmap_pages); |
761 | ||
762 | work->data = req; | |
763 | work->done = xen_blkbk_unmap_and_respond_callback; | |
764 | work->unmap_ops = req->unmap; | |
765 | work->kunmap_ops = NULL; | |
766 | work->pages = req->unmap_pages; | |
767 | work->count = invcount; | |
768 | ||
769 | gnttab_unmap_refs_async(&req->gnttab_unmap_data); | |
770 | } | |
771 | ||
772 | ||
773 | /* | |
774 | * Unmap the grant references. | |
775 | * | |
776 | * This could accumulate ops up to the batch size to reduce the number | |
777 | * of hypercalls, but since this is only used in error paths there's | |
778 | * no real need. | |
779 | */ | |
59795700 | 780 | static void xen_blkbk_unmap(struct xen_blkif_ring *ring, |
c43cf3ea JH |
781 | struct grant_page *pages[], |
782 | int num) | |
783 | { | |
784 | struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
785 | struct page *unmap_pages[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
786 | unsigned int invcount = 0; | |
787 | int ret; | |
788 | ||
789 | while (num) { | |
790 | unsigned int batch = min(num, BLKIF_MAX_SEGMENTS_PER_REQUEST); | |
59795700 BL |
791 | |
792 | invcount = xen_blkbk_unmap_prepare(ring, pages, batch, | |
c43cf3ea JH |
793 | unmap, unmap_pages); |
794 | if (invcount) { | |
795 | ret = gnttab_unmap_refs(unmap, NULL, unmap_pages, invcount); | |
31552ee3 | 796 | BUG_ON(ret); |
d4bf0065 | 797 | put_free_pages(ring, unmap_pages, invcount); |
31552ee3 | 798 | } |
c43cf3ea JH |
799 | pages += batch; |
800 | num -= batch; | |
b0aef179 | 801 | } |
b0aef179 | 802 | } |
01f37f2d | 803 | |
59795700 | 804 | static int xen_blkbk_map(struct xen_blkif_ring *ring, |
bb642e83 | 805 | struct grant_page *pages[], |
31552ee3 | 806 | int num, bool ro) |
1a95fe6e KRW |
807 | { |
808 | struct gnttab_map_grant_ref map[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
0a8704a5 RPM |
809 | struct page *pages_to_gnt[BLKIF_MAX_SEGMENTS_PER_REQUEST]; |
810 | struct persistent_gnt *persistent_gnt = NULL; | |
0a8704a5 | 811 | phys_addr_t addr = 0; |
c6cc142d | 812 | int i, seg_idx, new_map_idx; |
0a8704a5 | 813 | int segs_to_map = 0; |
1a95fe6e | 814 | int ret = 0; |
31552ee3 | 815 | int last_map = 0, map_until = 0; |
0a8704a5 | 816 | int use_persistent_gnts; |
59795700 | 817 | struct xen_blkif *blkif = ring->blkif; |
0a8704a5 RPM |
818 | |
819 | use_persistent_gnts = (blkif->vbd.feature_gnt_persistent); | |
820 | ||
01f37f2d KRW |
821 | /* |
822 | * Fill out preq.nr_sects with proper amount of sectors, and setup | |
1a95fe6e KRW |
823 | * assign map[..] with the PFN of the page in our domain with the |
824 | * corresponding grant reference for each page. | |
825 | */ | |
31552ee3 RPM |
826 | again: |
827 | for (i = map_until; i < num; i++) { | |
1a95fe6e KRW |
828 | uint32_t flags; |
829 | ||
59795700 | 830 | if (use_persistent_gnts) { |
0a8704a5 | 831 | persistent_gnt = get_persistent_gnt( |
d4bf0065 | 832 | ring, |
bb642e83 | 833 | pages[i]->gref); |
59795700 | 834 | } |
0a8704a5 RPM |
835 | |
836 | if (persistent_gnt) { | |
837 | /* | |
838 | * We are using persistent grants and | |
839 | * the grant is already mapped | |
840 | */ | |
bb642e83 RPM |
841 | pages[i]->page = persistent_gnt->page; |
842 | pages[i]->persistent_gnt = persistent_gnt; | |
0a8704a5 | 843 | } else { |
d4bf0065 | 844 | if (get_free_page(ring, &pages[i]->page)) |
c6cc142d | 845 | goto out_of_memory; |
bb642e83 RPM |
846 | addr = vaddr(pages[i]->page); |
847 | pages_to_gnt[segs_to_map] = pages[i]->page; | |
848 | pages[i]->persistent_gnt = NULL; | |
0a8704a5 | 849 | flags = GNTMAP_host_map; |
31552ee3 | 850 | if (!use_persistent_gnts && ro) |
0a8704a5 RPM |
851 | flags |= GNTMAP_readonly; |
852 | gnttab_set_map_op(&map[segs_to_map++], addr, | |
bb642e83 | 853 | flags, pages[i]->gref, |
0a8704a5 RPM |
854 | blkif->domid); |
855 | } | |
31552ee3 RPM |
856 | map_until = i + 1; |
857 | if (segs_to_map == BLKIF_MAX_SEGMENTS_PER_REQUEST) | |
858 | break; | |
1a95fe6e KRW |
859 | } |
860 | ||
0a8704a5 RPM |
861 | if (segs_to_map) { |
862 | ret = gnttab_map_refs(map, NULL, pages_to_gnt, segs_to_map); | |
863 | BUG_ON(ret); | |
864 | } | |
1a95fe6e | 865 | |
01f37f2d KRW |
866 | /* |
867 | * Now swizzle the MFN in our domain with the MFN from the other domain | |
1a95fe6e KRW |
868 | * so that when we access vaddr(pending_req,i) it has the contents of |
869 | * the page from the other domain. | |
870 | */ | |
31552ee3 | 871 | for (seg_idx = last_map, new_map_idx = 0; seg_idx < map_until; seg_idx++) { |
bb642e83 | 872 | if (!pages[seg_idx]->persistent_gnt) { |
0a8704a5 | 873 | /* This is a newly mapped grant */ |
c6cc142d RPM |
874 | BUG_ON(new_map_idx >= segs_to_map); |
875 | if (unlikely(map[new_map_idx].status != 0)) { | |
77387b82 | 876 | pr_debug("invalid buffer -- could not remap it\n"); |
d4bf0065 | 877 | put_free_pages(ring, &pages[seg_idx]->page, 1); |
bb642e83 | 878 | pages[seg_idx]->handle = BLKBACK_INVALID_HANDLE; |
0a8704a5 | 879 | ret |= 1; |
31552ee3 | 880 | goto next; |
0a8704a5 | 881 | } |
bb642e83 | 882 | pages[seg_idx]->handle = map[new_map_idx].handle; |
c6cc142d | 883 | } else { |
31552ee3 | 884 | continue; |
0a8704a5 | 885 | } |
c6cc142d | 886 | if (use_persistent_gnts && |
d4bf0065 | 887 | ring->persistent_gnt_c < xen_blkif_max_pgrants) { |
c6cc142d RPM |
888 | /* |
889 | * We are using persistent grants, the grant is | |
3f3aad5e | 890 | * not mapped but we might have room for it. |
c6cc142d RPM |
891 | */ |
892 | persistent_gnt = kmalloc(sizeof(struct persistent_gnt), | |
893 | GFP_KERNEL); | |
894 | if (!persistent_gnt) { | |
0a8704a5 | 895 | /* |
c6cc142d RPM |
896 | * If we don't have enough memory to |
897 | * allocate the persistent_gnt struct | |
898 | * map this grant non-persistenly | |
0a8704a5 | 899 | */ |
31552ee3 | 900 | goto next; |
0a8704a5 | 901 | } |
c6cc142d RPM |
902 | persistent_gnt->gnt = map[new_map_idx].ref; |
903 | persistent_gnt->handle = map[new_map_idx].handle; | |
bb642e83 | 904 | persistent_gnt->page = pages[seg_idx]->page; |
d4bf0065 | 905 | if (add_persistent_gnt(ring, |
c6cc142d RPM |
906 | persistent_gnt)) { |
907 | kfree(persistent_gnt); | |
908 | persistent_gnt = NULL; | |
31552ee3 | 909 | goto next; |
c6cc142d | 910 | } |
bb642e83 | 911 | pages[seg_idx]->persistent_gnt = persistent_gnt; |
77387b82 | 912 | pr_debug("grant %u added to the tree of persistent grants, using %u/%u\n", |
d4bf0065 | 913 | persistent_gnt->gnt, ring->persistent_gnt_c, |
3f3aad5e | 914 | xen_blkif_max_pgrants); |
c6cc142d RPM |
915 | goto next; |
916 | } | |
917 | if (use_persistent_gnts && !blkif->vbd.overflow_max_grants) { | |
918 | blkif->vbd.overflow_max_grants = 1; | |
77387b82 | 919 | pr_debug("domain %u, device %#x is using maximum number of persistent grants\n", |
c6cc142d | 920 | blkif->domid, blkif->vbd.handle); |
1a95fe6e | 921 | } |
c6cc142d RPM |
922 | /* |
923 | * We could not map this grant persistently, so use it as | |
924 | * a non-persistent grant. | |
925 | */ | |
c6cc142d | 926 | next: |
31552ee3 | 927 | new_map_idx++; |
1a95fe6e | 928 | } |
31552ee3 RPM |
929 | segs_to_map = 0; |
930 | last_map = map_until; | |
931 | if (map_until != num) | |
932 | goto again; | |
933 | ||
1a95fe6e | 934 | return ret; |
c6cc142d RPM |
935 | |
936 | out_of_memory: | |
77387b82 | 937 | pr_alert("%s: out of memory\n", __func__); |
d4bf0065 | 938 | put_free_pages(ring, pages_to_gnt, segs_to_map); |
c6cc142d | 939 | return -ENOMEM; |
1a95fe6e KRW |
940 | } |
941 | ||
bb642e83 | 942 | static int xen_blkbk_map_seg(struct pending_req *pending_req) |
31552ee3 | 943 | { |
402b27f9 | 944 | int rc; |
31552ee3 | 945 | |
59795700 | 946 | rc = xen_blkbk_map(pending_req->ring, pending_req->segments, |
6684fa1c | 947 | pending_req->nr_segs, |
31552ee3 | 948 | (pending_req->operation != BLKIF_OP_READ)); |
31552ee3 | 949 | |
402b27f9 RPM |
950 | return rc; |
951 | } | |
31552ee3 | 952 | |
402b27f9 RPM |
953 | static int xen_blkbk_parse_indirect(struct blkif_request *req, |
954 | struct pending_req *pending_req, | |
955 | struct seg_buf seg[], | |
956 | struct phys_req *preq) | |
957 | { | |
bb642e83 | 958 | struct grant_page **pages = pending_req->indirect_pages; |
59795700 | 959 | struct xen_blkif_ring *ring = pending_req->ring; |
402b27f9 | 960 | int indirect_grefs, rc, n, nseg, i; |
80bfa2f6 | 961 | struct blkif_request_segment *segments = NULL; |
402b27f9 | 962 | |
6684fa1c | 963 | nseg = pending_req->nr_segs; |
402b27f9 RPM |
964 | indirect_grefs = INDIRECT_PAGES(nseg); |
965 | BUG_ON(indirect_grefs > BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST); | |
966 | ||
bb642e83 RPM |
967 | for (i = 0; i < indirect_grefs; i++) |
968 | pages[i]->gref = req->u.indirect.indirect_grefs[i]; | |
969 | ||
59795700 | 970 | rc = xen_blkbk_map(ring, pages, indirect_grefs, true); |
402b27f9 RPM |
971 | if (rc) |
972 | goto unmap; | |
973 | ||
974 | for (n = 0, i = 0; n < nseg; n++) { | |
18779149 RPM |
975 | uint8_t first_sect, last_sect; |
976 | ||
402b27f9 RPM |
977 | if ((n % SEGS_PER_INDIRECT_FRAME) == 0) { |
978 | /* Map indirect segments */ | |
979 | if (segments) | |
980 | kunmap_atomic(segments); | |
bb642e83 | 981 | segments = kmap_atomic(pages[n/SEGS_PER_INDIRECT_FRAME]->page); |
402b27f9 RPM |
982 | } |
983 | i = n % SEGS_PER_INDIRECT_FRAME; | |
18779149 | 984 | |
bb642e83 | 985 | pending_req->segments[n]->gref = segments[i].gref; |
18779149 RPM |
986 | |
987 | first_sect = READ_ONCE(segments[i].first_sect); | |
988 | last_sect = READ_ONCE(segments[i].last_sect); | |
989 | if (last_sect >= (XEN_PAGE_SIZE >> 9) || last_sect < first_sect) { | |
402b27f9 RPM |
990 | rc = -EINVAL; |
991 | goto unmap; | |
992 | } | |
18779149 RPM |
993 | |
994 | seg[n].nsec = last_sect - first_sect + 1; | |
995 | seg[n].offset = first_sect << 9; | |
402b27f9 RPM |
996 | preq->nr_sects += seg[n].nsec; |
997 | } | |
998 | ||
999 | unmap: | |
1000 | if (segments) | |
1001 | kunmap_atomic(segments); | |
59795700 | 1002 | xen_blkbk_unmap(ring, pages, indirect_grefs); |
402b27f9 | 1003 | return rc; |
31552ee3 RPM |
1004 | } |
1005 | ||
59795700 | 1006 | static int dispatch_discard_io(struct xen_blkif_ring *ring, |
42146352 | 1007 | struct blkif_request *req) |
b3cb0d6a LD |
1008 | { |
1009 | int err = 0; | |
1010 | int status = BLKIF_RSP_OKAY; | |
59795700 | 1011 | struct xen_blkif *blkif = ring->blkif; |
b3cb0d6a | 1012 | struct block_device *bdev = blkif->vbd.bdev; |
4dae7670 | 1013 | unsigned long secure; |
604c499c | 1014 | struct phys_req preq; |
b3cb0d6a | 1015 | |
ea5ec76d VN |
1016 | xen_blkif_get(blkif); |
1017 | ||
604c499c KRW |
1018 | preq.sector_number = req->u.discard.sector_number; |
1019 | preq.nr_sects = req->u.discard.nr_sectors; | |
1020 | ||
a022606e | 1021 | err = xen_vbd_translate(&preq, blkif, REQ_OP_WRITE); |
604c499c | 1022 | if (err) { |
77387b82 | 1023 | pr_warn("access denied: DISCARD [%llu->%llu] on dev=%04x\n", |
604c499c KRW |
1024 | preq.sector_number, |
1025 | preq.sector_number + preq.nr_sects, blkif->vbd.pdevice); | |
1026 | goto fail_response; | |
1027 | } | |
db6fbc10 | 1028 | ring->st_ds_req++; |
42146352 | 1029 | |
4dae7670 KRW |
1030 | secure = (blkif->vbd.discard_secure && |
1031 | (req->u.discard.flag & BLKIF_DISCARD_SECURE)) ? | |
1032 | BLKDEV_DISCARD_SECURE : 0; | |
1033 | ||
1034 | err = blkdev_issue_discard(bdev, req->u.discard.sector_number, | |
1035 | req->u.discard.nr_sectors, | |
1036 | GFP_KERNEL, secure); | |
604c499c | 1037 | fail_response: |
b3cb0d6a | 1038 | if (err == -EOPNOTSUPP) { |
77387b82 | 1039 | pr_debug("discard op failed, not supported\n"); |
b3cb0d6a LD |
1040 | status = BLKIF_RSP_EOPNOTSUPP; |
1041 | } else if (err) | |
1042 | status = BLKIF_RSP_ERROR; | |
1043 | ||
59795700 | 1044 | make_response(ring, req->u.discard.id, req->operation, status); |
42146352 KRW |
1045 | xen_blkif_put(blkif); |
1046 | return err; | |
b3cb0d6a LD |
1047 | } |
1048 | ||
59795700 | 1049 | static int dispatch_other_io(struct xen_blkif_ring *ring, |
0e367ae4 DV |
1050 | struct blkif_request *req, |
1051 | struct pending_req *pending_req) | |
1052 | { | |
59795700 BL |
1053 | free_req(ring, pending_req); |
1054 | make_response(ring, req->u.other.id, req->operation, | |
0e367ae4 DV |
1055 | BLKIF_RSP_EOPNOTSUPP); |
1056 | return -EIO; | |
1057 | } | |
1058 | ||
59795700 | 1059 | static void xen_blk_drain_io(struct xen_blkif_ring *ring) |
29bde093 | 1060 | { |
59795700 BL |
1061 | struct xen_blkif *blkif = ring->blkif; |
1062 | ||
29bde093 KRW |
1063 | atomic_set(&blkif->drain, 1); |
1064 | do { | |
59795700 | 1065 | if (atomic_read(&ring->inflight) == 0) |
6927d920 | 1066 | break; |
29bde093 KRW |
1067 | wait_for_completion_interruptible_timeout( |
1068 | &blkif->drain_complete, HZ); | |
1069 | ||
1070 | if (!atomic_read(&blkif->drain)) | |
1071 | break; | |
29bde093 KRW |
1072 | } while (!kthread_should_stop()); |
1073 | atomic_set(&blkif->drain, 0); | |
1074 | } | |
1075 | ||
4e4cbee9 CH |
1076 | static void __end_block_io_op(struct pending_req *pending_req, |
1077 | blk_status_t error) | |
4d05a28d KRW |
1078 | { |
1079 | /* An error fails the entire request. */ | |
4e4cbee9 CH |
1080 | if (pending_req->operation == BLKIF_OP_FLUSH_DISKCACHE && |
1081 | error == BLK_STS_NOTSUPP) { | |
77387b82 | 1082 | pr_debug("flush diskcache op failed, not supported\n"); |
59795700 | 1083 | xen_blkbk_flush_diskcache(XBT_NIL, pending_req->ring->blkif->be, 0); |
4d05a28d | 1084 | pending_req->status = BLKIF_RSP_EOPNOTSUPP; |
4e4cbee9 CH |
1085 | } else if (pending_req->operation == BLKIF_OP_WRITE_BARRIER && |
1086 | error == BLK_STS_NOTSUPP) { | |
77387b82 | 1087 | pr_debug("write barrier op failed, not supported\n"); |
59795700 | 1088 | xen_blkbk_barrier(XBT_NIL, pending_req->ring->blkif->be, 0); |
29bde093 | 1089 | pending_req->status = BLKIF_RSP_EOPNOTSUPP; |
4d05a28d | 1090 | } else if (error) { |
77387b82 | 1091 | pr_debug("Buffer not up-to-date at end of operation," |
ebe81906 | 1092 | " error=%d\n", error); |
4d05a28d KRW |
1093 | pending_req->status = BLKIF_RSP_ERROR; |
1094 | } | |
1095 | ||
01f37f2d KRW |
1096 | /* |
1097 | * If all of the bio's have completed it is time to unmap | |
a1397fa3 | 1098 | * the grant references associated with 'request' and provide |
2e9977c2 KRW |
1099 | * the proper response on the ring. |
1100 | */ | |
c43cf3ea JH |
1101 | if (atomic_dec_and_test(&pending_req->pendcnt)) |
1102 | xen_blkbk_unmap_and_respond(pending_req); | |
4d05a28d KRW |
1103 | } |
1104 | ||
a1397fa3 KRW |
1105 | /* |
1106 | * bio callback. | |
1107 | */ | |
4246a0b6 | 1108 | static void end_block_io_op(struct bio *bio) |
4d05a28d | 1109 | { |
4e4cbee9 | 1110 | __end_block_io_op(bio->bi_private, bio->bi_status); |
4d05a28d | 1111 | bio_put(bio); |
4d05a28d KRW |
1112 | } |
1113 | ||
1114 | ||
4d05a28d | 1115 | |
a1397fa3 KRW |
1116 | /* |
1117 | * Function to copy the from the ring buffer the 'struct blkif_request' | |
1118 | * (which has the sectors we want, number of them, grant references, etc), | |
1119 | * and transmute it to the block API to hand it over to the proper block disk. | |
4d05a28d | 1120 | */ |
b4726a9d | 1121 | static int |
59795700 | 1122 | __do_block_io_op(struct xen_blkif_ring *ring) |
4d05a28d | 1123 | { |
59795700 | 1124 | union blkif_back_rings *blk_rings = &ring->blk_rings; |
88122933 | 1125 | struct blkif_request req; |
2e9977c2 | 1126 | struct pending_req *pending_req; |
4d05a28d KRW |
1127 | RING_IDX rc, rp; |
1128 | int more_to_do = 0; | |
1129 | ||
1130 | rc = blk_rings->common.req_cons; | |
1131 | rp = blk_rings->common.sring->req_prod; | |
1132 | rmb(); /* Ensure we see queued requests up to 'rp'. */ | |
1133 | ||
8e3f8755 KRW |
1134 | if (RING_REQUEST_PROD_OVERFLOW(&blk_rings->common, rp)) { |
1135 | rc = blk_rings->common.rsp_prod_pvt; | |
77387b82 | 1136 | pr_warn("Frontend provided bogus ring requests (%d - %d = %d). Halting ring processing on dev=%04x\n", |
59795700 | 1137 | rp, rc, rp - rc, ring->blkif->vbd.pdevice); |
8e3f8755 KRW |
1138 | return -EACCES; |
1139 | } | |
4d05a28d KRW |
1140 | while (rc != rp) { |
1141 | ||
1142 | if (RING_REQUEST_CONS_OVERFLOW(&blk_rings->common, rc)) | |
1143 | break; | |
1144 | ||
8270b45b | 1145 | if (kthread_should_stop()) { |
4d05a28d KRW |
1146 | more_to_do = 1; |
1147 | break; | |
1148 | } | |
1149 | ||
59795700 | 1150 | pending_req = alloc_req(ring); |
8270b45b | 1151 | if (NULL == pending_req) { |
db6fbc10 | 1152 | ring->st_oo_req++; |
4d05a28d KRW |
1153 | more_to_do = 1; |
1154 | break; | |
1155 | } | |
1156 | ||
59795700 | 1157 | switch (ring->blkif->blk_protocol) { |
4d05a28d KRW |
1158 | case BLKIF_PROTOCOL_NATIVE: |
1159 | memcpy(&req, RING_GET_REQUEST(&blk_rings->native, rc), sizeof(req)); | |
1160 | break; | |
1161 | case BLKIF_PROTOCOL_X86_32: | |
1162 | blkif_get_x86_32_req(&req, RING_GET_REQUEST(&blk_rings->x86_32, rc)); | |
1163 | break; | |
1164 | case BLKIF_PROTOCOL_X86_64: | |
1165 | blkif_get_x86_64_req(&req, RING_GET_REQUEST(&blk_rings->x86_64, rc)); | |
1166 | break; | |
1167 | default: | |
1168 | BUG(); | |
1169 | } | |
1170 | blk_rings->common.req_cons = ++rc; /* before make_response() */ | |
1171 | ||
1172 | /* Apply all sanity checks to /private copy/ of request. */ | |
1173 | barrier(); | |
0e367ae4 DV |
1174 | |
1175 | switch (req.operation) { | |
1176 | case BLKIF_OP_READ: | |
1177 | case BLKIF_OP_WRITE: | |
1178 | case BLKIF_OP_WRITE_BARRIER: | |
1179 | case BLKIF_OP_FLUSH_DISKCACHE: | |
402b27f9 | 1180 | case BLKIF_OP_INDIRECT: |
59795700 | 1181 | if (dispatch_rw_block_io(ring, &req, pending_req)) |
0e367ae4 DV |
1182 | goto done; |
1183 | break; | |
1184 | case BLKIF_OP_DISCARD: | |
59795700 BL |
1185 | free_req(ring, pending_req); |
1186 | if (dispatch_discard_io(ring, &req)) | |
0e367ae4 | 1187 | goto done; |
4d05a28d | 1188 | break; |
0e367ae4 | 1189 | default: |
59795700 | 1190 | if (dispatch_other_io(ring, &req, pending_req)) |
0e367ae4 DV |
1191 | goto done; |
1192 | break; | |
1193 | } | |
4d05a28d KRW |
1194 | |
1195 | /* Yield point for this unbounded loop. */ | |
1196 | cond_resched(); | |
1197 | } | |
0e367ae4 | 1198 | done: |
4d05a28d KRW |
1199 | return more_to_do; |
1200 | } | |
1201 | ||
b4726a9d | 1202 | static int |
59795700 | 1203 | do_block_io_op(struct xen_blkif_ring *ring) |
b4726a9d | 1204 | { |
59795700 | 1205 | union blkif_back_rings *blk_rings = &ring->blk_rings; |
b4726a9d DS |
1206 | int more_to_do; |
1207 | ||
1208 | do { | |
59795700 | 1209 | more_to_do = __do_block_io_op(ring); |
b4726a9d DS |
1210 | if (more_to_do) |
1211 | break; | |
1212 | ||
1213 | RING_FINAL_CHECK_FOR_REQUESTS(&blk_rings->common, more_to_do); | |
1214 | } while (more_to_do); | |
1215 | ||
1216 | return more_to_do; | |
1217 | } | |
a1397fa3 | 1218 | /* |
01f37f2d KRW |
1219 | * Transmutation of the 'struct blkif_request' to a proper 'struct bio' |
1220 | * and call the 'submit_bio' to pass it to the underlying storage. | |
a1397fa3 | 1221 | */ |
59795700 | 1222 | static int dispatch_rw_block_io(struct xen_blkif_ring *ring, |
30fd1502 KRW |
1223 | struct blkif_request *req, |
1224 | struct pending_req *pending_req) | |
4d05a28d | 1225 | { |
4d05a28d | 1226 | struct phys_req preq; |
402b27f9 | 1227 | struct seg_buf *seg = pending_req->seg; |
4d05a28d KRW |
1228 | unsigned int nseg; |
1229 | struct bio *bio = NULL; | |
402b27f9 | 1230 | struct bio **biolist = pending_req->biolist; |
1a95fe6e | 1231 | int i, nbio = 0; |
4d05a28d | 1232 | int operation; |
a022606e | 1233 | int operation_flags = 0; |
a19be5f0 | 1234 | struct blk_plug plug; |
29bde093 | 1235 | bool drain = false; |
bb642e83 | 1236 | struct grant_page **pages = pending_req->segments; |
402b27f9 RPM |
1237 | unsigned short req_operation; |
1238 | ||
1239 | req_operation = req->operation == BLKIF_OP_INDIRECT ? | |
1240 | req->u.indirect.indirect_op : req->operation; | |
67de5dfb | 1241 | |
402b27f9 RPM |
1242 | if ((req->operation == BLKIF_OP_INDIRECT) && |
1243 | (req_operation != BLKIF_OP_READ) && | |
1244 | (req_operation != BLKIF_OP_WRITE)) { | |
77387b82 | 1245 | pr_debug("Invalid indirect operation (%u)\n", req_operation); |
402b27f9 RPM |
1246 | goto fail_response; |
1247 | } | |
4d05a28d | 1248 | |
402b27f9 | 1249 | switch (req_operation) { |
4d05a28d | 1250 | case BLKIF_OP_READ: |
db6fbc10 | 1251 | ring->st_rd_req++; |
a022606e | 1252 | operation = REQ_OP_READ; |
4d05a28d KRW |
1253 | break; |
1254 | case BLKIF_OP_WRITE: | |
db6fbc10 | 1255 | ring->st_wr_req++; |
a022606e | 1256 | operation = REQ_OP_WRITE; |
70fd7614 | 1257 | operation_flags = REQ_SYNC | REQ_IDLE; |
4d05a28d | 1258 | break; |
29bde093 KRW |
1259 | case BLKIF_OP_WRITE_BARRIER: |
1260 | drain = true; | |
3f2c9405 | 1261 | /* fall through */ |
24f567f9 | 1262 | case BLKIF_OP_FLUSH_DISKCACHE: |
db6fbc10 | 1263 | ring->st_f_req++; |
a022606e | 1264 | operation = REQ_OP_WRITE; |
70fd7614 | 1265 | operation_flags = REQ_PREFLUSH; |
4d05a28d KRW |
1266 | break; |
1267 | default: | |
1268 | operation = 0; /* make gcc happy */ | |
fc53bf75 KRW |
1269 | goto fail_response; |
1270 | break; | |
4d05a28d KRW |
1271 | } |
1272 | ||
42146352 | 1273 | /* Check that the number of segments is sane. */ |
402b27f9 RPM |
1274 | nseg = req->operation == BLKIF_OP_INDIRECT ? |
1275 | req->u.indirect.nr_segments : req->u.rw.nr_segments; | |
97e36834 | 1276 | |
70fd7614 | 1277 | if (unlikely(nseg == 0 && operation_flags != REQ_PREFLUSH) || |
402b27f9 RPM |
1278 | unlikely((req->operation != BLKIF_OP_INDIRECT) && |
1279 | (nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST)) || | |
1280 | unlikely((req->operation == BLKIF_OP_INDIRECT) && | |
1281 | (nseg > MAX_INDIRECT_SEGMENTS))) { | |
77387b82 | 1282 | pr_debug("Bad number of segments in request (%d)\n", nseg); |
1a95fe6e | 1283 | /* Haven't submitted any bio's yet. */ |
4d05a28d KRW |
1284 | goto fail_response; |
1285 | } | |
1286 | ||
4d05a28d KRW |
1287 | preq.nr_sects = 0; |
1288 | ||
59795700 | 1289 | pending_req->ring = ring; |
97e36834 | 1290 | pending_req->id = req->u.rw.id; |
402b27f9 | 1291 | pending_req->operation = req_operation; |
4d05a28d | 1292 | pending_req->status = BLKIF_RSP_OKAY; |
6684fa1c | 1293 | pending_req->nr_segs = nseg; |
e9350493 | 1294 | |
402b27f9 RPM |
1295 | if (req->operation != BLKIF_OP_INDIRECT) { |
1296 | preq.dev = req->u.rw.handle; | |
1297 | preq.sector_number = req->u.rw.sector_number; | |
1298 | for (i = 0; i < nseg; i++) { | |
bb642e83 | 1299 | pages[i]->gref = req->u.rw.seg[i].gref; |
402b27f9 RPM |
1300 | seg[i].nsec = req->u.rw.seg[i].last_sect - |
1301 | req->u.rw.seg[i].first_sect + 1; | |
1302 | seg[i].offset = (req->u.rw.seg[i].first_sect << 9); | |
67de5dfb | 1303 | if ((req->u.rw.seg[i].last_sect >= (XEN_PAGE_SIZE >> 9)) || |
402b27f9 RPM |
1304 | (req->u.rw.seg[i].last_sect < |
1305 | req->u.rw.seg[i].first_sect)) | |
1306 | goto fail_response; | |
1307 | preq.nr_sects += seg[i].nsec; | |
1308 | } | |
1309 | } else { | |
1310 | preq.dev = req->u.indirect.handle; | |
1311 | preq.sector_number = req->u.indirect.sector_number; | |
1312 | if (xen_blkbk_parse_indirect(req, pending_req, seg, &preq)) | |
4d05a28d | 1313 | goto fail_response; |
4d05a28d KRW |
1314 | } |
1315 | ||
59795700 | 1316 | if (xen_vbd_translate(&preq, ring->blkif, operation) != 0) { |
77387b82 | 1317 | pr_debug("access denied: %s of [%llu,%llu] on dev=%04x\n", |
a022606e | 1318 | operation == REQ_OP_READ ? "read" : "write", |
ebe81906 | 1319 | preq.sector_number, |
a72d9002 | 1320 | preq.sector_number + preq.nr_sects, |
59795700 | 1321 | ring->blkif->vbd.pdevice); |
1a95fe6e | 1322 | goto fail_response; |
4d05a28d | 1323 | } |
01f37f2d KRW |
1324 | |
1325 | /* | |
3d814731 | 1326 | * This check _MUST_ be done after xen_vbd_translate as the preq.bdev |
01f37f2d KRW |
1327 | * is set there. |
1328 | */ | |
e9350493 KRW |
1329 | for (i = 0; i < nseg; i++) { |
1330 | if (((int)preq.sector_number|(int)seg[i].nsec) & | |
1331 | ((bdev_logical_block_size(preq.bdev) >> 9) - 1)) { | |
77387b82 | 1332 | pr_debug("Misaligned I/O request from domain %d\n", |
59795700 | 1333 | ring->blkif->domid); |
e9350493 KRW |
1334 | goto fail_response; |
1335 | } | |
1336 | } | |
01f37f2d | 1337 | |
29bde093 | 1338 | /* Wait on all outstanding I/O's and once that has been completed |
70fd7614 | 1339 | * issue the flush. |
29bde093 KRW |
1340 | */ |
1341 | if (drain) | |
59795700 | 1342 | xen_blk_drain_io(pending_req->ring); |
29bde093 | 1343 | |
01f37f2d KRW |
1344 | /* |
1345 | * If we have failed at this point, we need to undo the M2P override, | |
2e9977c2 KRW |
1346 | * set gnttab_set_unmap_op on all of the grant references and perform |
1347 | * the hypercall to unmap the grants - that is all done in | |
9f3aedf5 | 1348 | * xen_blkbk_unmap. |
2e9977c2 | 1349 | */ |
bb642e83 | 1350 | if (xen_blkbk_map_seg(pending_req)) |
4d05a28d KRW |
1351 | goto fail_flush; |
1352 | ||
b3cb0d6a LD |
1353 | /* |
1354 | * This corresponding xen_blkif_put is done in __end_block_io_op, or | |
1355 | * below (in "!bio") if we are handling a BLKIF_OP_DISCARD. | |
1356 | */ | |
59795700 BL |
1357 | xen_blkif_get(ring->blkif); |
1358 | atomic_inc(&ring->inflight); | |
4d05a28d KRW |
1359 | |
1360 | for (i = 0; i < nseg; i++) { | |
4d05a28d KRW |
1361 | while ((bio == NULL) || |
1362 | (bio_add_page(bio, | |
bb642e83 | 1363 | pages[i]->page, |
4d05a28d | 1364 | seg[i].nsec << 9, |
ffb1dabd | 1365 | seg[i].offset) == 0)) { |
2e9977c2 | 1366 | |
1e0f7a21 RPM |
1367 | int nr_iovecs = min_t(int, (nseg-i), BIO_MAX_PAGES); |
1368 | bio = bio_alloc(GFP_KERNEL, nr_iovecs); | |
4d05a28d KRW |
1369 | if (unlikely(bio == NULL)) |
1370 | goto fail_put_bio; | |
1371 | ||
03e0edf9 | 1372 | biolist[nbio++] = bio; |
74d46992 | 1373 | bio_set_dev(bio, preq.bdev); |
4d05a28d KRW |
1374 | bio->bi_private = pending_req; |
1375 | bio->bi_end_io = end_block_io_op; | |
4f024f37 | 1376 | bio->bi_iter.bi_sector = preq.sector_number; |
a022606e | 1377 | bio_set_op_attrs(bio, operation, operation_flags); |
4d05a28d KRW |
1378 | } |
1379 | ||
1380 | preq.sector_number += seg[i].nsec; | |
1381 | } | |
1382 | ||
b3cb0d6a | 1383 | /* This will be hit if the operation was a flush or discard. */ |
4d05a28d | 1384 | if (!bio) { |
70fd7614 | 1385 | BUG_ON(operation_flags != REQ_PREFLUSH); |
b0f80127 | 1386 | |
42146352 KRW |
1387 | bio = bio_alloc(GFP_KERNEL, 0); |
1388 | if (unlikely(bio == NULL)) | |
1389 | goto fail_put_bio; | |
4d05a28d | 1390 | |
42146352 | 1391 | biolist[nbio++] = bio; |
74d46992 | 1392 | bio_set_dev(bio, preq.bdev); |
42146352 KRW |
1393 | bio->bi_private = pending_req; |
1394 | bio->bi_end_io = end_block_io_op; | |
a022606e | 1395 | bio_set_op_attrs(bio, operation, operation_flags); |
4d05a28d KRW |
1396 | } |
1397 | ||
77089926 | 1398 | atomic_set(&pending_req->pendcnt, nbio); |
a19be5f0 KRW |
1399 | blk_start_plug(&plug); |
1400 | ||
77089926 | 1401 | for (i = 0; i < nbio; i++) |
4e49ea4a | 1402 | submit_bio(biolist[i]); |
77089926 | 1403 | |
a19be5f0 | 1404 | /* Let the I/Os go.. */ |
3d68b399 | 1405 | blk_finish_plug(&plug); |
a19be5f0 | 1406 | |
a022606e | 1407 | if (operation == REQ_OP_READ) |
db6fbc10 | 1408 | ring->st_rd_sect += preq.nr_sects; |
a022606e | 1409 | else if (operation == REQ_OP_WRITE) |
db6fbc10 | 1410 | ring->st_wr_sect += preq.nr_sects; |
4d05a28d | 1411 | |
fc53bf75 | 1412 | return 0; |
4d05a28d KRW |
1413 | |
1414 | fail_flush: | |
59795700 | 1415 | xen_blkbk_unmap(ring, pending_req->segments, |
6684fa1c | 1416 | pending_req->nr_segs); |
4d05a28d | 1417 | fail_response: |
0faa8cca | 1418 | /* Haven't submitted any bio's yet. */ |
59795700 BL |
1419 | make_response(ring, req->u.rw.id, req_operation, BLKIF_RSP_ERROR); |
1420 | free_req(ring, pending_req); | |
4d05a28d | 1421 | msleep(1); /* back off a bit */ |
fc53bf75 | 1422 | return -EIO; |
4d05a28d KRW |
1423 | |
1424 | fail_put_bio: | |
03e0edf9 | 1425 | for (i = 0; i < nbio; i++) |
77089926 | 1426 | bio_put(biolist[i]); |
0e5e098a | 1427 | atomic_set(&pending_req->pendcnt, 1); |
4e4cbee9 | 1428 | __end_block_io_op(pending_req, BLK_STS_RESOURCE); |
4d05a28d | 1429 | msleep(1); /* back off a bit */ |
fc53bf75 | 1430 | return -EIO; |
4d05a28d KRW |
1431 | } |
1432 | ||
1433 | ||
1434 | ||
a1397fa3 KRW |
1435 | /* |
1436 | * Put a response on the ring on how the operation fared. | |
4d05a28d | 1437 | */ |
59795700 | 1438 | static void make_response(struct xen_blkif_ring *ring, u64 id, |
4d05a28d KRW |
1439 | unsigned short op, int st) |
1440 | { | |
089bc014 | 1441 | struct blkif_response *resp; |
4d05a28d | 1442 | unsigned long flags; |
59795700 | 1443 | union blkif_back_rings *blk_rings; |
4d05a28d KRW |
1444 | int notify; |
1445 | ||
59795700 BL |
1446 | spin_lock_irqsave(&ring->blk_ring_lock, flags); |
1447 | blk_rings = &ring->blk_rings; | |
4d05a28d | 1448 | /* Place on the response ring for the relevant domain. */ |
59795700 | 1449 | switch (ring->blkif->blk_protocol) { |
4d05a28d | 1450 | case BLKIF_PROTOCOL_NATIVE: |
089bc014 JB |
1451 | resp = RING_GET_RESPONSE(&blk_rings->native, |
1452 | blk_rings->native.rsp_prod_pvt); | |
4d05a28d KRW |
1453 | break; |
1454 | case BLKIF_PROTOCOL_X86_32: | |
089bc014 JB |
1455 | resp = RING_GET_RESPONSE(&blk_rings->x86_32, |
1456 | blk_rings->x86_32.rsp_prod_pvt); | |
4d05a28d KRW |
1457 | break; |
1458 | case BLKIF_PROTOCOL_X86_64: | |
089bc014 JB |
1459 | resp = RING_GET_RESPONSE(&blk_rings->x86_64, |
1460 | blk_rings->x86_64.rsp_prod_pvt); | |
4d05a28d KRW |
1461 | break; |
1462 | default: | |
1463 | BUG(); | |
1464 | } | |
089bc014 JB |
1465 | |
1466 | resp->id = id; | |
1467 | resp->operation = op; | |
1468 | resp->status = st; | |
1469 | ||
4d05a28d KRW |
1470 | blk_rings->common.rsp_prod_pvt++; |
1471 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blk_rings->common, notify); | |
59795700 | 1472 | spin_unlock_irqrestore(&ring->blk_ring_lock, flags); |
4d05a28d | 1473 | if (notify) |
59795700 | 1474 | notify_remote_via_irq(ring->irq); |
4d05a28d KRW |
1475 | } |
1476 | ||
8b6bf747 | 1477 | static int __init xen_blkif_init(void) |
4d05a28d | 1478 | { |
8770b268 | 1479 | int rc = 0; |
4d05a28d | 1480 | |
b2167ba6 | 1481 | if (!xen_domain()) |
4d05a28d KRW |
1482 | return -ENODEV; |
1483 | ||
9cce2914 | 1484 | if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) { |
86839c56 | 1485 | pr_info("Invalid max_ring_order (%d), will use default max: %d.\n", |
9cce2914 JG |
1486 | xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER); |
1487 | xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER; | |
86839c56 BL |
1488 | } |
1489 | ||
d62d8600 BL |
1490 | if (xenblk_max_queues == 0) |
1491 | xenblk_max_queues = num_online_cpus(); | |
1492 | ||
8b6bf747 | 1493 | rc = xen_blkif_interface_init(); |
8770b268 KRW |
1494 | if (rc) |
1495 | goto failed_init; | |
4d05a28d | 1496 | |
8b6bf747 | 1497 | rc = xen_blkif_xenbus_init(); |
8770b268 KRW |
1498 | if (rc) |
1499 | goto failed_init; | |
4d05a28d | 1500 | |
8770b268 | 1501 | failed_init: |
8770b268 | 1502 | return rc; |
4d05a28d KRW |
1503 | } |
1504 | ||
8b6bf747 | 1505 | module_init(xen_blkif_init); |
4d05a28d KRW |
1506 | |
1507 | MODULE_LICENSE("Dual BSD/GPL"); | |
a7e9357f | 1508 | MODULE_ALIAS("xen-backend:vbd"); |