]>
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 | ||
37 | #include <linux/spinlock.h> | |
38 | #include <linux/kthread.h> | |
39 | #include <linux/list.h> | |
40 | #include <linux/delay.h> | |
88122933 | 41 | #include <linux/freezer.h> |
0a8704a5 | 42 | #include <linux/bitmap.h> |
afd91d07 | 43 | |
88122933 JF |
44 | #include <xen/events.h> |
45 | #include <xen/page.h> | |
e79affc3 | 46 | #include <xen/xen.h> |
88122933 JF |
47 | #include <asm/xen/hypervisor.h> |
48 | #include <asm/xen/hypercall.h> | |
087ffecd | 49 | #include <xen/balloon.h> |
4d05a28d KRW |
50 | #include "common.h" |
51 | ||
52 | /* | |
53 | * These are rather arbitrary. They are fairly large because adjacent requests | |
54 | * pulled from a communication ring are quite likely to end up being part of | |
55 | * the same scatter/gather request at the disc. | |
56 | * | |
8b6bf747 | 57 | * ** TRY INCREASING 'xen_blkif_reqs' IF WRITE SPEEDS SEEM TOO LOW ** |
4d05a28d KRW |
58 | * |
59 | * This will increase the chances of being able to write whole tracks. | |
60 | * 64 should be enough to keep us competitive with Linux. | |
61 | */ | |
8b6bf747 KRW |
62 | static int xen_blkif_reqs = 64; |
63 | module_param_named(reqs, xen_blkif_reqs, int, 0); | |
4d05a28d KRW |
64 | MODULE_PARM_DESC(reqs, "Number of blkback requests to allocate"); |
65 | ||
66 | /* Run-time switchable: /sys/module/blkback/parameters/ */ | |
2e9977c2 | 67 | static unsigned int log_stats; |
4d05a28d | 68 | module_param(log_stats, int, 0644); |
4d05a28d KRW |
69 | |
70 | /* | |
71 | * Each outstanding request that we've passed to the lower device layers has a | |
72 | * 'pending_req' allocated to it. Each buffer_head that completes decrements | |
73 | * the pendcnt towards zero. When it hits zero, the specified domain has a | |
74 | * response queued for it, with the saved 'id' passed back. | |
75 | */ | |
2e9977c2 | 76 | struct pending_req { |
30fd1502 | 77 | struct xen_blkif *blkif; |
01f37f2d KRW |
78 | u64 id; |
79 | int nr_pages; | |
80 | atomic_t pendcnt; | |
81 | unsigned short operation; | |
82 | int status; | |
83 | struct list_head free_list; | |
0a8704a5 | 84 | DECLARE_BITMAP(unmap_seg, BLKIF_MAX_SEGMENTS_PER_REQUEST); |
2e9977c2 | 85 | }; |
4d05a28d | 86 | |
4d05a28d KRW |
87 | #define BLKBACK_INVALID_HANDLE (~0) |
88 | ||
e8e28871 | 89 | struct xen_blkbk { |
2e9977c2 | 90 | struct pending_req *pending_reqs; |
a1397fa3 | 91 | /* List of all 'pending_req' available */ |
e8e28871 | 92 | struct list_head pending_free; |
a1397fa3 | 93 | /* And its spinlock. */ |
e8e28871 KRW |
94 | spinlock_t pending_free_lock; |
95 | wait_queue_head_t pending_free_wq; | |
a1397fa3 | 96 | /* The list of all pages that are available. */ |
e8e28871 | 97 | struct page **pending_pages; |
a1397fa3 | 98 | /* And the grant handles that are available. */ |
e8e28871 KRW |
99 | grant_handle_t *pending_grant_handles; |
100 | }; | |
101 | ||
102 | static struct xen_blkbk *blkbk; | |
4d05a28d | 103 | |
0a8704a5 RPM |
104 | /* |
105 | * Maximum number of grant pages that can be mapped in blkback. | |
106 | * BLKIF_MAX_SEGMENTS_PER_REQUEST * RING_SIZE is the maximum number of | |
107 | * pages that blkback will persistently map. | |
108 | * Currently, this is: | |
109 | * RING_SIZE = 32 (for all known ring types) | |
110 | * BLKIF_MAX_SEGMENTS_PER_REQUEST = 11 | |
111 | * sizeof(struct persistent_gnt) = 48 | |
112 | * So the maximum memory used to store the grants is: | |
113 | * 32 * 11 * 48 = 16896 bytes | |
114 | */ | |
115 | static inline unsigned int max_mapped_grant_pages(enum blkif_protocol protocol) | |
116 | { | |
117 | switch (protocol) { | |
118 | case BLKIF_PROTOCOL_NATIVE: | |
119 | return __CONST_RING_SIZE(blkif, PAGE_SIZE) * | |
120 | BLKIF_MAX_SEGMENTS_PER_REQUEST; | |
121 | case BLKIF_PROTOCOL_X86_32: | |
122 | return __CONST_RING_SIZE(blkif_x86_32, PAGE_SIZE) * | |
123 | BLKIF_MAX_SEGMENTS_PER_REQUEST; | |
124 | case BLKIF_PROTOCOL_X86_64: | |
125 | return __CONST_RING_SIZE(blkif_x86_64, PAGE_SIZE) * | |
126 | BLKIF_MAX_SEGMENTS_PER_REQUEST; | |
127 | default: | |
128 | BUG(); | |
129 | } | |
130 | return 0; | |
131 | } | |
132 | ||
133 | ||
a1397fa3 KRW |
134 | /* |
135 | * Little helpful macro to figure out the index and virtual address of the | |
136 | * pending_pages[..]. For each 'pending_req' we have have up to | |
137 | * BLKIF_MAX_SEGMENTS_PER_REQUEST (11) pages. The seg would be from 0 through | |
01f37f2d KRW |
138 | * 10 and would index in the pending_pages[..]. |
139 | */ | |
2e9977c2 | 140 | static inline int vaddr_pagenr(struct pending_req *req, int seg) |
4d05a28d | 141 | { |
2e9977c2 KRW |
142 | return (req - blkbk->pending_reqs) * |
143 | BLKIF_MAX_SEGMENTS_PER_REQUEST + seg; | |
4d05a28d KRW |
144 | } |
145 | ||
efe08a3e JB |
146 | #define pending_page(req, seg) pending_pages[vaddr_pagenr(req, seg)] |
147 | ||
2e9977c2 | 148 | static inline unsigned long vaddr(struct pending_req *req, int seg) |
4d05a28d | 149 | { |
e8e28871 | 150 | unsigned long pfn = page_to_pfn(blkbk->pending_page(req, seg)); |
4d05a28d KRW |
151 | return (unsigned long)pfn_to_kaddr(pfn); |
152 | } | |
153 | ||
154 | #define pending_handle(_req, _seg) \ | |
e8e28871 | 155 | (blkbk->pending_grant_handles[vaddr_pagenr(_req, _seg)]) |
4d05a28d KRW |
156 | |
157 | ||
30fd1502 KRW |
158 | static int do_block_io_op(struct xen_blkif *blkif); |
159 | static int dispatch_rw_block_io(struct xen_blkif *blkif, | |
fc53bf75 KRW |
160 | struct blkif_request *req, |
161 | struct pending_req *pending_req); | |
30fd1502 | 162 | static void make_response(struct xen_blkif *blkif, u64 id, |
4d05a28d KRW |
163 | unsigned short op, int st); |
164 | ||
7dc34117 RPM |
165 | #define foreach_grant_safe(pos, n, rbtree, node) \ |
166 | for ((pos) = container_of(rb_first((rbtree)), typeof(*(pos)), node), \ | |
217fd5e7 | 167 | (n) = (&(pos)->node != NULL) ? rb_next(&(pos)->node) : NULL; \ |
0a8704a5 | 168 | &(pos)->node != NULL; \ |
7dc34117 RPM |
169 | (pos) = container_of(n, typeof(*(pos)), node), \ |
170 | (n) = (&(pos)->node != NULL) ? rb_next(&(pos)->node) : NULL) | |
0a8704a5 RPM |
171 | |
172 | ||
173 | static void add_persistent_gnt(struct rb_root *root, | |
174 | struct persistent_gnt *persistent_gnt) | |
175 | { | |
176 | struct rb_node **new = &(root->rb_node), *parent = NULL; | |
177 | struct persistent_gnt *this; | |
178 | ||
179 | /* Figure out where to put new node */ | |
180 | while (*new) { | |
181 | this = container_of(*new, struct persistent_gnt, node); | |
182 | ||
183 | parent = *new; | |
184 | if (persistent_gnt->gnt < this->gnt) | |
185 | new = &((*new)->rb_left); | |
186 | else if (persistent_gnt->gnt > this->gnt) | |
187 | new = &((*new)->rb_right); | |
188 | else { | |
189 | pr_alert(DRV_PFX " trying to add a gref that's already in the tree\n"); | |
190 | BUG(); | |
191 | } | |
192 | } | |
193 | ||
194 | /* Add new node and rebalance tree. */ | |
195 | rb_link_node(&(persistent_gnt->node), parent, new); | |
196 | rb_insert_color(&(persistent_gnt->node), root); | |
197 | } | |
198 | ||
199 | static struct persistent_gnt *get_persistent_gnt(struct rb_root *root, | |
200 | grant_ref_t gref) | |
201 | { | |
202 | struct persistent_gnt *data; | |
203 | struct rb_node *node = root->rb_node; | |
204 | ||
205 | while (node) { | |
206 | data = container_of(node, struct persistent_gnt, node); | |
207 | ||
208 | if (gref < data->gnt) | |
209 | node = node->rb_left; | |
210 | else if (gref > data->gnt) | |
211 | node = node->rb_right; | |
212 | else | |
213 | return data; | |
214 | } | |
215 | return NULL; | |
216 | } | |
217 | ||
4d4f270f RPM |
218 | static void free_persistent_gnts(struct rb_root *root, unsigned int num) |
219 | { | |
220 | struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
221 | struct page *pages[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
222 | struct persistent_gnt *persistent_gnt; | |
7dc34117 | 223 | struct rb_node *n; |
4d4f270f RPM |
224 | int ret = 0; |
225 | int segs_to_unmap = 0; | |
226 | ||
7dc34117 | 227 | foreach_grant_safe(persistent_gnt, n, root, node) { |
4d4f270f RPM |
228 | BUG_ON(persistent_gnt->handle == |
229 | BLKBACK_INVALID_HANDLE); | |
230 | gnttab_set_unmap_op(&unmap[segs_to_unmap], | |
231 | (unsigned long) pfn_to_kaddr(page_to_pfn( | |
232 | persistent_gnt->page)), | |
233 | GNTMAP_host_map, | |
234 | persistent_gnt->handle); | |
235 | ||
236 | pages[segs_to_unmap] = persistent_gnt->page; | |
4d4f270f RPM |
237 | |
238 | if (++segs_to_unmap == BLKIF_MAX_SEGMENTS_PER_REQUEST || | |
239 | !rb_next(&persistent_gnt->node)) { | |
240 | ret = gnttab_unmap_refs(unmap, NULL, pages, | |
241 | segs_to_unmap); | |
242 | BUG_ON(ret); | |
087ffecd | 243 | free_xenballooned_pages(segs_to_unmap, pages); |
4d4f270f RPM |
244 | segs_to_unmap = 0; |
245 | } | |
7dc34117 RPM |
246 | |
247 | rb_erase(&persistent_gnt->node, root); | |
248 | kfree(persistent_gnt); | |
249 | num--; | |
4d4f270f RPM |
250 | } |
251 | BUG_ON(num != 0); | |
252 | } | |
253 | ||
a1397fa3 KRW |
254 | /* |
255 | * Retrieve from the 'pending_reqs' a free pending_req structure to be used. | |
4d05a28d | 256 | */ |
2e9977c2 | 257 | static struct pending_req *alloc_req(void) |
4d05a28d | 258 | { |
2e9977c2 | 259 | struct pending_req *req = NULL; |
4d05a28d KRW |
260 | unsigned long flags; |
261 | ||
e8e28871 KRW |
262 | spin_lock_irqsave(&blkbk->pending_free_lock, flags); |
263 | if (!list_empty(&blkbk->pending_free)) { | |
2e9977c2 KRW |
264 | req = list_entry(blkbk->pending_free.next, struct pending_req, |
265 | free_list); | |
4d05a28d KRW |
266 | list_del(&req->free_list); |
267 | } | |
e8e28871 | 268 | spin_unlock_irqrestore(&blkbk->pending_free_lock, flags); |
4d05a28d KRW |
269 | return req; |
270 | } | |
271 | ||
a1397fa3 KRW |
272 | /* |
273 | * Return the 'pending_req' structure back to the freepool. We also | |
274 | * wake up the thread if it was waiting for a free page. | |
275 | */ | |
2e9977c2 | 276 | static void free_req(struct pending_req *req) |
4d05a28d KRW |
277 | { |
278 | unsigned long flags; | |
279 | int was_empty; | |
280 | ||
e8e28871 KRW |
281 | spin_lock_irqsave(&blkbk->pending_free_lock, flags); |
282 | was_empty = list_empty(&blkbk->pending_free); | |
283 | list_add(&req->free_list, &blkbk->pending_free); | |
284 | spin_unlock_irqrestore(&blkbk->pending_free_lock, flags); | |
4d05a28d | 285 | if (was_empty) |
e8e28871 | 286 | wake_up(&blkbk->pending_free_wq); |
4d05a28d KRW |
287 | } |
288 | ||
ee9ff853 KRW |
289 | /* |
290 | * Routines for managing virtual block devices (vbds). | |
291 | */ | |
3d814731 KRW |
292 | static int xen_vbd_translate(struct phys_req *req, struct xen_blkif *blkif, |
293 | int operation) | |
ee9ff853 | 294 | { |
3d814731 | 295 | struct xen_vbd *vbd = &blkif->vbd; |
ee9ff853 KRW |
296 | int rc = -EACCES; |
297 | ||
298 | if ((operation != READ) && vbd->readonly) | |
299 | goto out; | |
300 | ||
8ab52150 JB |
301 | if (likely(req->nr_sects)) { |
302 | blkif_sector_t end = req->sector_number + req->nr_sects; | |
303 | ||
304 | if (unlikely(end < req->sector_number)) | |
305 | goto out; | |
306 | if (unlikely(end > vbd_sz(vbd))) | |
307 | goto out; | |
308 | } | |
ee9ff853 KRW |
309 | |
310 | req->dev = vbd->pdevice; | |
311 | req->bdev = vbd->bdev; | |
312 | rc = 0; | |
313 | ||
314 | out: | |
315 | return rc; | |
316 | } | |
317 | ||
3d814731 | 318 | static void xen_vbd_resize(struct xen_blkif *blkif) |
ee9ff853 | 319 | { |
3d814731 | 320 | struct xen_vbd *vbd = &blkif->vbd; |
ee9ff853 KRW |
321 | struct xenbus_transaction xbt; |
322 | int err; | |
8b6bf747 | 323 | struct xenbus_device *dev = xen_blkbk_xenbus(blkif->be); |
42c7841d | 324 | unsigned long long new_size = vbd_sz(vbd); |
ee9ff853 | 325 | |
22b20f2d | 326 | pr_info(DRV_PFX "VBD Resize: Domid: %d, Device: (%d, %d)\n", |
ee9ff853 | 327 | blkif->domid, MAJOR(vbd->pdevice), MINOR(vbd->pdevice)); |
22b20f2d | 328 | pr_info(DRV_PFX "VBD Resize: new size %llu\n", new_size); |
ee9ff853 KRW |
329 | vbd->size = new_size; |
330 | again: | |
331 | err = xenbus_transaction_start(&xbt); | |
332 | if (err) { | |
22b20f2d | 333 | pr_warn(DRV_PFX "Error starting transaction"); |
ee9ff853 KRW |
334 | return; |
335 | } | |
336 | err = xenbus_printf(xbt, dev->nodename, "sectors", "%llu", | |
42c7841d | 337 | (unsigned long long)vbd_sz(vbd)); |
ee9ff853 | 338 | if (err) { |
22b20f2d | 339 | pr_warn(DRV_PFX "Error writing new size"); |
ee9ff853 KRW |
340 | goto abort; |
341 | } | |
342 | /* | |
343 | * Write the current state; we will use this to synchronize | |
344 | * the front-end. If the current state is "connected" the | |
345 | * front-end will get the new size information online. | |
346 | */ | |
347 | err = xenbus_printf(xbt, dev->nodename, "state", "%d", dev->state); | |
348 | if (err) { | |
22b20f2d | 349 | pr_warn(DRV_PFX "Error writing the state"); |
ee9ff853 KRW |
350 | goto abort; |
351 | } | |
352 | ||
353 | err = xenbus_transaction_end(xbt, 0); | |
354 | if (err == -EAGAIN) | |
355 | goto again; | |
356 | if (err) | |
22b20f2d | 357 | pr_warn(DRV_PFX "Error ending transaction"); |
496b318e | 358 | return; |
ee9ff853 KRW |
359 | abort: |
360 | xenbus_transaction_end(xbt, 1); | |
361 | } | |
362 | ||
a1397fa3 | 363 | /* |
b0aef179 KRW |
364 | * Notification from the guest OS. |
365 | */ | |
30fd1502 | 366 | static void blkif_notify_work(struct xen_blkif *blkif) |
4d05a28d | 367 | { |
b0aef179 KRW |
368 | blkif->waiting_reqs = 1; |
369 | wake_up(&blkif->wq); | |
370 | } | |
4d05a28d | 371 | |
8b6bf747 | 372 | irqreturn_t xen_blkif_be_int(int irq, void *dev_id) |
b0aef179 KRW |
373 | { |
374 | blkif_notify_work(dev_id); | |
375 | return IRQ_HANDLED; | |
4d05a28d KRW |
376 | } |
377 | ||
2e9977c2 | 378 | /* |
4d05a28d KRW |
379 | * SCHEDULER FUNCTIONS |
380 | */ | |
381 | ||
30fd1502 | 382 | static void print_stats(struct xen_blkif *blkif) |
4d05a28d | 383 | { |
986cacbd ZK |
384 | pr_info("xen-blkback (%s): oo %3llu | rd %4llu | wr %4llu | f %4llu" |
385 | " | ds %4llu\n", | |
ebe81906 | 386 | current->comm, blkif->st_oo_req, |
b3cb0d6a LD |
387 | blkif->st_rd_req, blkif->st_wr_req, |
388 | blkif->st_f_req, blkif->st_ds_req); | |
4d05a28d KRW |
389 | blkif->st_print = jiffies + msecs_to_jiffies(10 * 1000); |
390 | blkif->st_rd_req = 0; | |
391 | blkif->st_wr_req = 0; | |
392 | blkif->st_oo_req = 0; | |
b3cb0d6a | 393 | blkif->st_ds_req = 0; |
4d05a28d KRW |
394 | } |
395 | ||
8b6bf747 | 396 | int xen_blkif_schedule(void *arg) |
4d05a28d | 397 | { |
30fd1502 | 398 | struct xen_blkif *blkif = arg; |
3d814731 | 399 | struct xen_vbd *vbd = &blkif->vbd; |
4d05a28d | 400 | |
8b6bf747 | 401 | xen_blkif_get(blkif); |
4d05a28d | 402 | |
4d05a28d KRW |
403 | while (!kthread_should_stop()) { |
404 | if (try_to_freeze()) | |
405 | continue; | |
42c7841d | 406 | if (unlikely(vbd->size != vbd_sz(vbd))) |
3d814731 | 407 | xen_vbd_resize(blkif); |
4d05a28d KRW |
408 | |
409 | wait_event_interruptible( | |
410 | blkif->wq, | |
411 | blkif->waiting_reqs || kthread_should_stop()); | |
412 | wait_event_interruptible( | |
e8e28871 | 413 | blkbk->pending_free_wq, |
2e9977c2 KRW |
414 | !list_empty(&blkbk->pending_free) || |
415 | kthread_should_stop()); | |
4d05a28d KRW |
416 | |
417 | blkif->waiting_reqs = 0; | |
418 | smp_mb(); /* clear flag *before* checking for work */ | |
419 | ||
420 | if (do_block_io_op(blkif)) | |
421 | blkif->waiting_reqs = 1; | |
4d05a28d KRW |
422 | |
423 | if (log_stats && time_after(jiffies, blkif->st_print)) | |
424 | print_stats(blkif); | |
425 | } | |
426 | ||
0a8704a5 | 427 | /* Free all persistent grant pages */ |
4d4f270f RPM |
428 | if (!RB_EMPTY_ROOT(&blkif->persistent_gnts)) |
429 | free_persistent_gnts(&blkif->persistent_gnts, | |
430 | blkif->persistent_gnt_c); | |
0a8704a5 | 431 | |
0a8704a5 | 432 | BUG_ON(!RB_EMPTY_ROOT(&blkif->persistent_gnts)); |
4d4f270f | 433 | blkif->persistent_gnt_c = 0; |
0a8704a5 | 434 | |
4d05a28d KRW |
435 | if (log_stats) |
436 | print_stats(blkif); | |
4d05a28d KRW |
437 | |
438 | blkif->xenblkd = NULL; | |
8b6bf747 | 439 | xen_blkif_put(blkif); |
4d05a28d KRW |
440 | |
441 | return 0; | |
442 | } | |
443 | ||
1a95fe6e | 444 | struct seg_buf { |
ffb1dabd | 445 | unsigned int offset; |
1a95fe6e KRW |
446 | unsigned int nsec; |
447 | }; | |
b0aef179 KRW |
448 | /* |
449 | * Unmap the grant references, and also remove the M2P over-rides | |
450 | * used in the 'pending_req'. | |
01f37f2d | 451 | */ |
9f3aedf5 | 452 | static void xen_blkbk_unmap(struct pending_req *req) |
b0aef179 KRW |
453 | { |
454 | struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
4f14faaa | 455 | struct page *pages[BLKIF_MAX_SEGMENTS_PER_REQUEST]; |
b0aef179 KRW |
456 | unsigned int i, invcount = 0; |
457 | grant_handle_t handle; | |
458 | int ret; | |
459 | ||
460 | for (i = 0; i < req->nr_pages; i++) { | |
0a8704a5 RPM |
461 | if (!test_bit(i, req->unmap_seg)) |
462 | continue; | |
b0aef179 KRW |
463 | handle = pending_handle(req, i); |
464 | if (handle == BLKBACK_INVALID_HANDLE) | |
465 | continue; | |
466 | gnttab_set_unmap_op(&unmap[invcount], vaddr(req, i), | |
467 | GNTMAP_host_map, handle); | |
468 | pending_handle(req, i) = BLKBACK_INVALID_HANDLE; | |
4f14faaa | 469 | pages[invcount] = virt_to_page(vaddr(req, i)); |
b0aef179 KRW |
470 | invcount++; |
471 | } | |
472 | ||
2fc136ee | 473 | ret = gnttab_unmap_refs(unmap, NULL, pages, invcount); |
b0aef179 | 474 | BUG_ON(ret); |
b0aef179 | 475 | } |
01f37f2d KRW |
476 | |
477 | static int xen_blkbk_map(struct blkif_request *req, | |
478 | struct pending_req *pending_req, | |
0a8704a5 RPM |
479 | struct seg_buf seg[], |
480 | struct page *pages[]) | |
1a95fe6e KRW |
481 | { |
482 | struct gnttab_map_grant_ref map[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
0a8704a5 RPM |
483 | struct persistent_gnt *persistent_gnts[BLKIF_MAX_SEGMENTS_PER_REQUEST]; |
484 | struct page *pages_to_gnt[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
485 | struct persistent_gnt *persistent_gnt = NULL; | |
486 | struct xen_blkif *blkif = pending_req->blkif; | |
487 | phys_addr_t addr = 0; | |
488 | int i, j; | |
489 | bool new_map; | |
97e36834 | 490 | int nseg = req->u.rw.nr_segments; |
0a8704a5 | 491 | int segs_to_map = 0; |
1a95fe6e | 492 | int ret = 0; |
0a8704a5 RPM |
493 | int use_persistent_gnts; |
494 | ||
495 | use_persistent_gnts = (blkif->vbd.feature_gnt_persistent); | |
496 | ||
497 | BUG_ON(blkif->persistent_gnt_c > | |
498 | max_mapped_grant_pages(pending_req->blkif->blk_protocol)); | |
01f37f2d KRW |
499 | |
500 | /* | |
501 | * Fill out preq.nr_sects with proper amount of sectors, and setup | |
1a95fe6e KRW |
502 | * assign map[..] with the PFN of the page in our domain with the |
503 | * corresponding grant reference for each page. | |
504 | */ | |
505 | for (i = 0; i < nseg; i++) { | |
506 | uint32_t flags; | |
507 | ||
0a8704a5 RPM |
508 | if (use_persistent_gnts) |
509 | persistent_gnt = get_persistent_gnt( | |
510 | &blkif->persistent_gnts, | |
511 | req->u.rw.seg[i].gref); | |
512 | ||
513 | if (persistent_gnt) { | |
514 | /* | |
515 | * We are using persistent grants and | |
516 | * the grant is already mapped | |
517 | */ | |
518 | new_map = false; | |
519 | } else if (use_persistent_gnts && | |
520 | blkif->persistent_gnt_c < | |
521 | max_mapped_grant_pages(blkif->blk_protocol)) { | |
522 | /* | |
523 | * We are using persistent grants, the grant is | |
524 | * not mapped but we have room for it | |
525 | */ | |
526 | new_map = true; | |
cb5bd4d1 | 527 | persistent_gnt = kmalloc( |
0a8704a5 RPM |
528 | sizeof(struct persistent_gnt), |
529 | GFP_KERNEL); | |
530 | if (!persistent_gnt) | |
531 | return -ENOMEM; | |
087ffecd RPM |
532 | if (alloc_xenballooned_pages(1, &persistent_gnt->page, |
533 | false)) { | |
0a8704a5 RPM |
534 | kfree(persistent_gnt); |
535 | return -ENOMEM; | |
536 | } | |
537 | persistent_gnt->gnt = req->u.rw.seg[i].gref; | |
cb5bd4d1 | 538 | persistent_gnt->handle = BLKBACK_INVALID_HANDLE; |
0a8704a5 RPM |
539 | |
540 | pages_to_gnt[segs_to_map] = | |
541 | persistent_gnt->page; | |
542 | addr = (unsigned long) pfn_to_kaddr( | |
543 | page_to_pfn(persistent_gnt->page)); | |
544 | ||
545 | add_persistent_gnt(&blkif->persistent_gnts, | |
546 | persistent_gnt); | |
547 | blkif->persistent_gnt_c++; | |
548 | pr_debug(DRV_PFX " grant %u added to the tree of persistent grants, using %u/%u\n", | |
549 | persistent_gnt->gnt, blkif->persistent_gnt_c, | |
550 | max_mapped_grant_pages(blkif->blk_protocol)); | |
551 | } else { | |
552 | /* | |
553 | * We are either using persistent grants and | |
554 | * hit the maximum limit of grants mapped, | |
555 | * or we are not using persistent grants. | |
556 | */ | |
557 | if (use_persistent_gnts && | |
558 | !blkif->vbd.overflow_max_grants) { | |
559 | blkif->vbd.overflow_max_grants = 1; | |
560 | pr_alert(DRV_PFX " domain %u, device %#x is using maximum number of persistent grants\n", | |
561 | blkif->domid, blkif->vbd.handle); | |
562 | } | |
563 | new_map = true; | |
564 | pages[i] = blkbk->pending_page(pending_req, i); | |
565 | addr = vaddr(pending_req, i); | |
566 | pages_to_gnt[segs_to_map] = | |
567 | blkbk->pending_page(pending_req, i); | |
568 | } | |
569 | ||
570 | if (persistent_gnt) { | |
571 | pages[i] = persistent_gnt->page; | |
572 | persistent_gnts[i] = persistent_gnt; | |
573 | } else { | |
574 | persistent_gnts[i] = NULL; | |
575 | } | |
576 | ||
577 | if (new_map) { | |
578 | flags = GNTMAP_host_map; | |
579 | if (!persistent_gnt && | |
580 | (pending_req->operation != BLKIF_OP_READ)) | |
581 | flags |= GNTMAP_readonly; | |
582 | gnttab_set_map_op(&map[segs_to_map++], addr, | |
583 | flags, req->u.rw.seg[i].gref, | |
584 | blkif->domid); | |
585 | } | |
1a95fe6e KRW |
586 | } |
587 | ||
0a8704a5 RPM |
588 | if (segs_to_map) { |
589 | ret = gnttab_map_refs(map, NULL, pages_to_gnt, segs_to_map); | |
590 | BUG_ON(ret); | |
591 | } | |
1a95fe6e | 592 | |
01f37f2d KRW |
593 | /* |
594 | * Now swizzle the MFN in our domain with the MFN from the other domain | |
1a95fe6e KRW |
595 | * so that when we access vaddr(pending_req,i) it has the contents of |
596 | * the page from the other domain. | |
597 | */ | |
0a8704a5 RPM |
598 | bitmap_zero(pending_req->unmap_seg, BLKIF_MAX_SEGMENTS_PER_REQUEST); |
599 | for (i = 0, j = 0; i < nseg; i++) { | |
cb5bd4d1 RPM |
600 | if (!persistent_gnts[i] || |
601 | persistent_gnts[i]->handle == BLKBACK_INVALID_HANDLE) { | |
0a8704a5 RPM |
602 | /* This is a newly mapped grant */ |
603 | BUG_ON(j >= segs_to_map); | |
604 | if (unlikely(map[j].status != 0)) { | |
605 | pr_debug(DRV_PFX "invalid buffer -- could not remap it\n"); | |
606 | map[j].handle = BLKBACK_INVALID_HANDLE; | |
607 | ret |= 1; | |
608 | if (persistent_gnts[i]) { | |
609 | rb_erase(&persistent_gnts[i]->node, | |
610 | &blkif->persistent_gnts); | |
611 | blkif->persistent_gnt_c--; | |
612 | kfree(persistent_gnts[i]); | |
613 | persistent_gnts[i] = NULL; | |
614 | } | |
615 | } | |
616 | } | |
617 | if (persistent_gnts[i]) { | |
cb5bd4d1 RPM |
618 | if (persistent_gnts[i]->handle == |
619 | BLKBACK_INVALID_HANDLE) { | |
0a8704a5 RPM |
620 | /* |
621 | * If this is a new persistent grant | |
622 | * save the handler | |
623 | */ | |
ffb1dabd | 624 | persistent_gnts[i]->handle = map[j++].handle; |
0a8704a5 RPM |
625 | } |
626 | pending_handle(pending_req, i) = | |
627 | persistent_gnts[i]->handle; | |
628 | ||
629 | if (ret) | |
630 | continue; | |
0a8704a5 | 631 | } else { |
ffb1dabd | 632 | pending_handle(pending_req, i) = map[j++].handle; |
0a8704a5 RPM |
633 | bitmap_set(pending_req->unmap_seg, i, 1); |
634 | ||
ffb1dabd | 635 | if (ret) |
0a8704a5 | 636 | continue; |
1a95fe6e | 637 | } |
ffb1dabd | 638 | seg[i].offset = (req->u.rw.seg[i].first_sect << 9); |
1a95fe6e KRW |
639 | } |
640 | return ret; | |
641 | } | |
642 | ||
42146352 KRW |
643 | static int dispatch_discard_io(struct xen_blkif *blkif, |
644 | struct blkif_request *req) | |
b3cb0d6a LD |
645 | { |
646 | int err = 0; | |
647 | int status = BLKIF_RSP_OKAY; | |
648 | struct block_device *bdev = blkif->vbd.bdev; | |
4dae7670 | 649 | unsigned long secure; |
b3cb0d6a | 650 | |
42146352 KRW |
651 | blkif->st_ds_req++; |
652 | ||
653 | xen_blkif_get(blkif); | |
4dae7670 KRW |
654 | secure = (blkif->vbd.discard_secure && |
655 | (req->u.discard.flag & BLKIF_DISCARD_SECURE)) ? | |
656 | BLKDEV_DISCARD_SECURE : 0; | |
657 | ||
658 | err = blkdev_issue_discard(bdev, req->u.discard.sector_number, | |
659 | req->u.discard.nr_sectors, | |
660 | GFP_KERNEL, secure); | |
b3cb0d6a LD |
661 | |
662 | if (err == -EOPNOTSUPP) { | |
663 | pr_debug(DRV_PFX "discard op failed, not supported\n"); | |
664 | status = BLKIF_RSP_EOPNOTSUPP; | |
665 | } else if (err) | |
666 | status = BLKIF_RSP_ERROR; | |
667 | ||
97e36834 | 668 | make_response(blkif, req->u.discard.id, req->operation, status); |
42146352 KRW |
669 | xen_blkif_put(blkif); |
670 | return err; | |
b3cb0d6a LD |
671 | } |
672 | ||
0e367ae4 DV |
673 | static int dispatch_other_io(struct xen_blkif *blkif, |
674 | struct blkif_request *req, | |
675 | struct pending_req *pending_req) | |
676 | { | |
677 | free_req(pending_req); | |
678 | make_response(blkif, req->u.other.id, req->operation, | |
679 | BLKIF_RSP_EOPNOTSUPP); | |
680 | return -EIO; | |
681 | } | |
682 | ||
29bde093 KRW |
683 | static void xen_blk_drain_io(struct xen_blkif *blkif) |
684 | { | |
685 | atomic_set(&blkif->drain, 1); | |
686 | do { | |
6927d920 KRW |
687 | /* The initial value is one, and one refcnt taken at the |
688 | * start of the xen_blkif_schedule thread. */ | |
689 | if (atomic_read(&blkif->refcnt) <= 2) | |
690 | break; | |
29bde093 KRW |
691 | wait_for_completion_interruptible_timeout( |
692 | &blkif->drain_complete, HZ); | |
693 | ||
694 | if (!atomic_read(&blkif->drain)) | |
695 | break; | |
29bde093 KRW |
696 | } while (!kthread_should_stop()); |
697 | atomic_set(&blkif->drain, 0); | |
698 | } | |
699 | ||
a1397fa3 KRW |
700 | /* |
701 | * Completion callback on the bio's. Called as bh->b_end_io() | |
4d05a28d KRW |
702 | */ |
703 | ||
2e9977c2 | 704 | static void __end_block_io_op(struct pending_req *pending_req, int error) |
4d05a28d KRW |
705 | { |
706 | /* An error fails the entire request. */ | |
24f567f9 | 707 | if ((pending_req->operation == BLKIF_OP_FLUSH_DISKCACHE) && |
4d05a28d | 708 | (error == -EOPNOTSUPP)) { |
22b20f2d | 709 | pr_debug(DRV_PFX "flush diskcache op failed, not supported\n"); |
24f567f9 | 710 | xen_blkbk_flush_diskcache(XBT_NIL, pending_req->blkif->be, 0); |
4d05a28d | 711 | pending_req->status = BLKIF_RSP_EOPNOTSUPP; |
29bde093 KRW |
712 | } else if ((pending_req->operation == BLKIF_OP_WRITE_BARRIER) && |
713 | (error == -EOPNOTSUPP)) { | |
714 | pr_debug(DRV_PFX "write barrier op failed, not supported\n"); | |
715 | xen_blkbk_barrier(XBT_NIL, pending_req->blkif->be, 0); | |
716 | pending_req->status = BLKIF_RSP_EOPNOTSUPP; | |
4d05a28d | 717 | } else if (error) { |
22b20f2d | 718 | pr_debug(DRV_PFX "Buffer not up-to-date at end of operation," |
ebe81906 | 719 | " error=%d\n", error); |
4d05a28d KRW |
720 | pending_req->status = BLKIF_RSP_ERROR; |
721 | } | |
722 | ||
01f37f2d KRW |
723 | /* |
724 | * If all of the bio's have completed it is time to unmap | |
a1397fa3 | 725 | * the grant references associated with 'request' and provide |
2e9977c2 KRW |
726 | * the proper response on the ring. |
727 | */ | |
4d05a28d | 728 | if (atomic_dec_and_test(&pending_req->pendcnt)) { |
9f3aedf5 | 729 | xen_blkbk_unmap(pending_req); |
4d05a28d KRW |
730 | make_response(pending_req->blkif, pending_req->id, |
731 | pending_req->operation, pending_req->status); | |
8b6bf747 | 732 | xen_blkif_put(pending_req->blkif); |
29bde093 KRW |
733 | if (atomic_read(&pending_req->blkif->refcnt) <= 2) { |
734 | if (atomic_read(&pending_req->blkif->drain)) | |
735 | complete(&pending_req->blkif->drain_complete); | |
736 | } | |
4d05a28d KRW |
737 | free_req(pending_req); |
738 | } | |
739 | } | |
740 | ||
a1397fa3 KRW |
741 | /* |
742 | * bio callback. | |
743 | */ | |
88122933 | 744 | static void end_block_io_op(struct bio *bio, int error) |
4d05a28d | 745 | { |
4d05a28d KRW |
746 | __end_block_io_op(bio->bi_private, error); |
747 | bio_put(bio); | |
4d05a28d KRW |
748 | } |
749 | ||
750 | ||
4d05a28d | 751 | |
a1397fa3 KRW |
752 | /* |
753 | * Function to copy the from the ring buffer the 'struct blkif_request' | |
754 | * (which has the sectors we want, number of them, grant references, etc), | |
755 | * and transmute it to the block API to hand it over to the proper block disk. | |
4d05a28d | 756 | */ |
b4726a9d DS |
757 | static int |
758 | __do_block_io_op(struct xen_blkif *blkif) | |
4d05a28d | 759 | { |
88122933 JF |
760 | union blkif_back_rings *blk_rings = &blkif->blk_rings; |
761 | struct blkif_request req; | |
2e9977c2 | 762 | struct pending_req *pending_req; |
4d05a28d KRW |
763 | RING_IDX rc, rp; |
764 | int more_to_do = 0; | |
765 | ||
766 | rc = blk_rings->common.req_cons; | |
767 | rp = blk_rings->common.sring->req_prod; | |
768 | rmb(); /* Ensure we see queued requests up to 'rp'. */ | |
769 | ||
770 | while (rc != rp) { | |
771 | ||
772 | if (RING_REQUEST_CONS_OVERFLOW(&blk_rings->common, rc)) | |
773 | break; | |
774 | ||
8270b45b | 775 | if (kthread_should_stop()) { |
4d05a28d KRW |
776 | more_to_do = 1; |
777 | break; | |
778 | } | |
779 | ||
8270b45b KF |
780 | pending_req = alloc_req(); |
781 | if (NULL == pending_req) { | |
782 | blkif->st_oo_req++; | |
4d05a28d KRW |
783 | more_to_do = 1; |
784 | break; | |
785 | } | |
786 | ||
787 | switch (blkif->blk_protocol) { | |
788 | case BLKIF_PROTOCOL_NATIVE: | |
789 | memcpy(&req, RING_GET_REQUEST(&blk_rings->native, rc), sizeof(req)); | |
790 | break; | |
791 | case BLKIF_PROTOCOL_X86_32: | |
792 | blkif_get_x86_32_req(&req, RING_GET_REQUEST(&blk_rings->x86_32, rc)); | |
793 | break; | |
794 | case BLKIF_PROTOCOL_X86_64: | |
795 | blkif_get_x86_64_req(&req, RING_GET_REQUEST(&blk_rings->x86_64, rc)); | |
796 | break; | |
797 | default: | |
798 | BUG(); | |
799 | } | |
800 | blk_rings->common.req_cons = ++rc; /* before make_response() */ | |
801 | ||
802 | /* Apply all sanity checks to /private copy/ of request. */ | |
803 | barrier(); | |
0e367ae4 DV |
804 | |
805 | switch (req.operation) { | |
806 | case BLKIF_OP_READ: | |
807 | case BLKIF_OP_WRITE: | |
808 | case BLKIF_OP_WRITE_BARRIER: | |
809 | case BLKIF_OP_FLUSH_DISKCACHE: | |
810 | if (dispatch_rw_block_io(blkif, &req, pending_req)) | |
811 | goto done; | |
812 | break; | |
813 | case BLKIF_OP_DISCARD: | |
42146352 KRW |
814 | free_req(pending_req); |
815 | if (dispatch_discard_io(blkif, &req)) | |
0e367ae4 | 816 | goto done; |
4d05a28d | 817 | break; |
0e367ae4 DV |
818 | default: |
819 | if (dispatch_other_io(blkif, &req, pending_req)) | |
820 | goto done; | |
821 | break; | |
822 | } | |
4d05a28d KRW |
823 | |
824 | /* Yield point for this unbounded loop. */ | |
825 | cond_resched(); | |
826 | } | |
0e367ae4 | 827 | done: |
4d05a28d KRW |
828 | return more_to_do; |
829 | } | |
830 | ||
b4726a9d DS |
831 | static int |
832 | do_block_io_op(struct xen_blkif *blkif) | |
833 | { | |
834 | union blkif_back_rings *blk_rings = &blkif->blk_rings; | |
835 | int more_to_do; | |
836 | ||
837 | do { | |
838 | more_to_do = __do_block_io_op(blkif); | |
839 | if (more_to_do) | |
840 | break; | |
841 | ||
842 | RING_FINAL_CHECK_FOR_REQUESTS(&blk_rings->common, more_to_do); | |
843 | } while (more_to_do); | |
844 | ||
845 | return more_to_do; | |
846 | } | |
a1397fa3 | 847 | /* |
01f37f2d KRW |
848 | * Transmutation of the 'struct blkif_request' to a proper 'struct bio' |
849 | * and call the 'submit_bio' to pass it to the underlying storage. | |
a1397fa3 | 850 | */ |
30fd1502 KRW |
851 | static int dispatch_rw_block_io(struct xen_blkif *blkif, |
852 | struct blkif_request *req, | |
853 | struct pending_req *pending_req) | |
4d05a28d | 854 | { |
4d05a28d | 855 | struct phys_req preq; |
1a95fe6e | 856 | struct seg_buf seg[BLKIF_MAX_SEGMENTS_PER_REQUEST]; |
4d05a28d KRW |
857 | unsigned int nseg; |
858 | struct bio *bio = NULL; | |
77089926 | 859 | struct bio *biolist[BLKIF_MAX_SEGMENTS_PER_REQUEST]; |
1a95fe6e | 860 | int i, nbio = 0; |
4d05a28d | 861 | int operation; |
a19be5f0 | 862 | struct blk_plug plug; |
29bde093 | 863 | bool drain = false; |
0a8704a5 | 864 | struct page *pages[BLKIF_MAX_SEGMENTS_PER_REQUEST]; |
4d05a28d KRW |
865 | |
866 | switch (req->operation) { | |
867 | case BLKIF_OP_READ: | |
fc53bf75 | 868 | blkif->st_rd_req++; |
4d05a28d KRW |
869 | operation = READ; |
870 | break; | |
871 | case BLKIF_OP_WRITE: | |
fc53bf75 | 872 | blkif->st_wr_req++; |
013c3ca1 | 873 | operation = WRITE_ODIRECT; |
4d05a28d | 874 | break; |
29bde093 KRW |
875 | case BLKIF_OP_WRITE_BARRIER: |
876 | drain = true; | |
24f567f9 | 877 | case BLKIF_OP_FLUSH_DISKCACHE: |
fc53bf75 | 878 | blkif->st_f_req++; |
24f567f9 | 879 | operation = WRITE_FLUSH; |
4d05a28d KRW |
880 | break; |
881 | default: | |
882 | operation = 0; /* make gcc happy */ | |
fc53bf75 KRW |
883 | goto fail_response; |
884 | break; | |
4d05a28d KRW |
885 | } |
886 | ||
42146352 KRW |
887 | /* Check that the number of segments is sane. */ |
888 | nseg = req->u.rw.nr_segments; | |
97e36834 | 889 | |
42146352 | 890 | if (unlikely(nseg == 0 && operation != WRITE_FLUSH) || |
4d05a28d | 891 | unlikely(nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST)) { |
22b20f2d | 892 | pr_debug(DRV_PFX "Bad number of segments in request (%d)\n", |
ebe81906 | 893 | nseg); |
1a95fe6e | 894 | /* Haven't submitted any bio's yet. */ |
4d05a28d KRW |
895 | goto fail_response; |
896 | } | |
897 | ||
c35950bf | 898 | preq.sector_number = req->u.rw.sector_number; |
4d05a28d KRW |
899 | preq.nr_sects = 0; |
900 | ||
901 | pending_req->blkif = blkif; | |
97e36834 | 902 | pending_req->id = req->u.rw.id; |
4d05a28d KRW |
903 | pending_req->operation = req->operation; |
904 | pending_req->status = BLKIF_RSP_OKAY; | |
905 | pending_req->nr_pages = nseg; | |
e9350493 | 906 | |
4d05a28d | 907 | for (i = 0; i < nseg; i++) { |
c35950bf KRW |
908 | seg[i].nsec = req->u.rw.seg[i].last_sect - |
909 | req->u.rw.seg[i].first_sect + 1; | |
c35950bf KRW |
910 | if ((req->u.rw.seg[i].last_sect >= (PAGE_SIZE >> 9)) || |
911 | (req->u.rw.seg[i].last_sect < req->u.rw.seg[i].first_sect)) | |
4d05a28d KRW |
912 | goto fail_response; |
913 | preq.nr_sects += seg[i].nsec; | |
976222e0 | 914 | |
4d05a28d KRW |
915 | } |
916 | ||
3d814731 | 917 | if (xen_vbd_translate(&preq, blkif, operation) != 0) { |
22b20f2d | 918 | pr_debug(DRV_PFX "access denied: %s of [%llu,%llu] on dev=%04x\n", |
ebe81906 KRW |
919 | operation == READ ? "read" : "write", |
920 | preq.sector_number, | |
a72d9002 CG |
921 | preq.sector_number + preq.nr_sects, |
922 | blkif->vbd.pdevice); | |
1a95fe6e | 923 | goto fail_response; |
4d05a28d | 924 | } |
01f37f2d KRW |
925 | |
926 | /* | |
3d814731 | 927 | * This check _MUST_ be done after xen_vbd_translate as the preq.bdev |
01f37f2d KRW |
928 | * is set there. |
929 | */ | |
e9350493 KRW |
930 | for (i = 0; i < nseg; i++) { |
931 | if (((int)preq.sector_number|(int)seg[i].nsec) & | |
932 | ((bdev_logical_block_size(preq.bdev) >> 9) - 1)) { | |
22b20f2d | 933 | pr_debug(DRV_PFX "Misaligned I/O request from domain %d", |
ebe81906 | 934 | blkif->domid); |
e9350493 KRW |
935 | goto fail_response; |
936 | } | |
937 | } | |
01f37f2d | 938 | |
29bde093 KRW |
939 | /* Wait on all outstanding I/O's and once that has been completed |
940 | * issue the WRITE_FLUSH. | |
941 | */ | |
942 | if (drain) | |
943 | xen_blk_drain_io(pending_req->blkif); | |
944 | ||
01f37f2d KRW |
945 | /* |
946 | * If we have failed at this point, we need to undo the M2P override, | |
2e9977c2 KRW |
947 | * set gnttab_set_unmap_op on all of the grant references and perform |
948 | * the hypercall to unmap the grants - that is all done in | |
9f3aedf5 | 949 | * xen_blkbk_unmap. |
2e9977c2 | 950 | */ |
0a8704a5 | 951 | if (xen_blkbk_map(req, pending_req, seg, pages)) |
4d05a28d KRW |
952 | goto fail_flush; |
953 | ||
b3cb0d6a LD |
954 | /* |
955 | * This corresponding xen_blkif_put is done in __end_block_io_op, or | |
956 | * below (in "!bio") if we are handling a BLKIF_OP_DISCARD. | |
957 | */ | |
8b6bf747 | 958 | xen_blkif_get(blkif); |
4d05a28d KRW |
959 | |
960 | for (i = 0; i < nseg; i++) { | |
4d05a28d KRW |
961 | while ((bio == NULL) || |
962 | (bio_add_page(bio, | |
0a8704a5 | 963 | pages[i], |
4d05a28d | 964 | seg[i].nsec << 9, |
ffb1dabd | 965 | seg[i].offset) == 0)) { |
2e9977c2 | 966 | |
03e0edf9 | 967 | bio = bio_alloc(GFP_KERNEL, nseg-i); |
4d05a28d KRW |
968 | if (unlikely(bio == NULL)) |
969 | goto fail_put_bio; | |
970 | ||
03e0edf9 | 971 | biolist[nbio++] = bio; |
4d05a28d KRW |
972 | bio->bi_bdev = preq.bdev; |
973 | bio->bi_private = pending_req; | |
974 | bio->bi_end_io = end_block_io_op; | |
975 | bio->bi_sector = preq.sector_number; | |
976 | } | |
977 | ||
978 | preq.sector_number += seg[i].nsec; | |
979 | } | |
980 | ||
b3cb0d6a | 981 | /* This will be hit if the operation was a flush or discard. */ |
4d05a28d | 982 | if (!bio) { |
42146352 | 983 | BUG_ON(operation != WRITE_FLUSH); |
b0f80127 | 984 | |
42146352 KRW |
985 | bio = bio_alloc(GFP_KERNEL, 0); |
986 | if (unlikely(bio == NULL)) | |
987 | goto fail_put_bio; | |
4d05a28d | 988 | |
42146352 KRW |
989 | biolist[nbio++] = bio; |
990 | bio->bi_bdev = preq.bdev; | |
991 | bio->bi_private = pending_req; | |
992 | bio->bi_end_io = end_block_io_op; | |
4d05a28d KRW |
993 | } |
994 | ||
77089926 | 995 | atomic_set(&pending_req->pendcnt, nbio); |
a19be5f0 KRW |
996 | blk_start_plug(&plug); |
997 | ||
77089926 KRW |
998 | for (i = 0; i < nbio; i++) |
999 | submit_bio(operation, biolist[i]); | |
1000 | ||
a19be5f0 | 1001 | /* Let the I/Os go.. */ |
3d68b399 | 1002 | blk_finish_plug(&plug); |
a19be5f0 | 1003 | |
4d05a28d KRW |
1004 | if (operation == READ) |
1005 | blkif->st_rd_sect += preq.nr_sects; | |
5c62cb48 | 1006 | else if (operation & WRITE) |
4d05a28d KRW |
1007 | blkif->st_wr_sect += preq.nr_sects; |
1008 | ||
fc53bf75 | 1009 | return 0; |
4d05a28d KRW |
1010 | |
1011 | fail_flush: | |
9f3aedf5 | 1012 | xen_blkbk_unmap(pending_req); |
4d05a28d | 1013 | fail_response: |
0faa8cca | 1014 | /* Haven't submitted any bio's yet. */ |
97e36834 | 1015 | make_response(blkif, req->u.rw.id, req->operation, BLKIF_RSP_ERROR); |
4d05a28d KRW |
1016 | free_req(pending_req); |
1017 | msleep(1); /* back off a bit */ | |
fc53bf75 | 1018 | return -EIO; |
4d05a28d KRW |
1019 | |
1020 | fail_put_bio: | |
03e0edf9 | 1021 | for (i = 0; i < nbio; i++) |
77089926 | 1022 | bio_put(biolist[i]); |
0e5e098a | 1023 | atomic_set(&pending_req->pendcnt, 1); |
4d05a28d | 1024 | __end_block_io_op(pending_req, -EINVAL); |
4d05a28d | 1025 | msleep(1); /* back off a bit */ |
fc53bf75 | 1026 | return -EIO; |
4d05a28d KRW |
1027 | } |
1028 | ||
1029 | ||
1030 | ||
a1397fa3 KRW |
1031 | /* |
1032 | * Put a response on the ring on how the operation fared. | |
4d05a28d | 1033 | */ |
30fd1502 | 1034 | static void make_response(struct xen_blkif *blkif, u64 id, |
4d05a28d KRW |
1035 | unsigned short op, int st) |
1036 | { | |
88122933 | 1037 | struct blkif_response resp; |
4d05a28d | 1038 | unsigned long flags; |
88122933 | 1039 | union blkif_back_rings *blk_rings = &blkif->blk_rings; |
4d05a28d KRW |
1040 | int notify; |
1041 | ||
1042 | resp.id = id; | |
1043 | resp.operation = op; | |
1044 | resp.status = st; | |
1045 | ||
1046 | spin_lock_irqsave(&blkif->blk_ring_lock, flags); | |
1047 | /* Place on the response ring for the relevant domain. */ | |
1048 | switch (blkif->blk_protocol) { | |
1049 | case BLKIF_PROTOCOL_NATIVE: | |
1050 | memcpy(RING_GET_RESPONSE(&blk_rings->native, blk_rings->native.rsp_prod_pvt), | |
1051 | &resp, sizeof(resp)); | |
1052 | break; | |
1053 | case BLKIF_PROTOCOL_X86_32: | |
1054 | memcpy(RING_GET_RESPONSE(&blk_rings->x86_32, blk_rings->x86_32.rsp_prod_pvt), | |
1055 | &resp, sizeof(resp)); | |
1056 | break; | |
1057 | case BLKIF_PROTOCOL_X86_64: | |
1058 | memcpy(RING_GET_RESPONSE(&blk_rings->x86_64, blk_rings->x86_64.rsp_prod_pvt), | |
1059 | &resp, sizeof(resp)); | |
1060 | break; | |
1061 | default: | |
1062 | BUG(); | |
1063 | } | |
1064 | blk_rings->common.rsp_prod_pvt++; | |
1065 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blk_rings->common, notify); | |
4d05a28d | 1066 | spin_unlock_irqrestore(&blkif->blk_ring_lock, flags); |
4d05a28d KRW |
1067 | if (notify) |
1068 | notify_remote_via_irq(blkif->irq); | |
1069 | } | |
1070 | ||
8b6bf747 | 1071 | static int __init xen_blkif_init(void) |
4d05a28d KRW |
1072 | { |
1073 | int i, mmap_pages; | |
8770b268 | 1074 | int rc = 0; |
4d05a28d | 1075 | |
b2167ba6 | 1076 | if (!xen_domain()) |
4d05a28d KRW |
1077 | return -ENODEV; |
1078 | ||
2e9977c2 | 1079 | blkbk = kzalloc(sizeof(struct xen_blkbk), GFP_KERNEL); |
e8e28871 | 1080 | if (!blkbk) { |
22b20f2d | 1081 | pr_alert(DRV_PFX "%s: out of memory!\n", __func__); |
e8e28871 KRW |
1082 | return -ENOMEM; |
1083 | } | |
1084 | ||
8b6bf747 | 1085 | mmap_pages = xen_blkif_reqs * BLKIF_MAX_SEGMENTS_PER_REQUEST; |
4d05a28d | 1086 | |
8e6dc6fe | 1087 | blkbk->pending_reqs = kzalloc(sizeof(blkbk->pending_reqs[0]) * |
8b6bf747 | 1088 | xen_blkif_reqs, GFP_KERNEL); |
8e6dc6fe | 1089 | blkbk->pending_grant_handles = kmalloc(sizeof(blkbk->pending_grant_handles[0]) * |
a742b02c KRW |
1090 | mmap_pages, GFP_KERNEL); |
1091 | blkbk->pending_pages = kzalloc(sizeof(blkbk->pending_pages[0]) * | |
1092 | mmap_pages, GFP_KERNEL); | |
4d05a28d | 1093 | |
2e9977c2 KRW |
1094 | if (!blkbk->pending_reqs || !blkbk->pending_grant_handles || |
1095 | !blkbk->pending_pages) { | |
8770b268 | 1096 | rc = -ENOMEM; |
4d05a28d | 1097 | goto out_of_memory; |
8770b268 | 1098 | } |
4d05a28d | 1099 | |
464fb419 | 1100 | for (i = 0; i < mmap_pages; i++) { |
e8e28871 | 1101 | blkbk->pending_grant_handles[i] = BLKBACK_INVALID_HANDLE; |
a742b02c | 1102 | blkbk->pending_pages[i] = alloc_page(GFP_KERNEL); |
464fb419 KRW |
1103 | if (blkbk->pending_pages[i] == NULL) { |
1104 | rc = -ENOMEM; | |
1105 | goto out_of_memory; | |
1106 | } | |
1107 | } | |
8b6bf747 | 1108 | rc = xen_blkif_interface_init(); |
8770b268 KRW |
1109 | if (rc) |
1110 | goto failed_init; | |
4d05a28d | 1111 | |
e8e28871 KRW |
1112 | INIT_LIST_HEAD(&blkbk->pending_free); |
1113 | spin_lock_init(&blkbk->pending_free_lock); | |
1114 | init_waitqueue_head(&blkbk->pending_free_wq); | |
4d05a28d | 1115 | |
8b6bf747 | 1116 | for (i = 0; i < xen_blkif_reqs; i++) |
2e9977c2 KRW |
1117 | list_add_tail(&blkbk->pending_reqs[i].free_list, |
1118 | &blkbk->pending_free); | |
4d05a28d | 1119 | |
8b6bf747 | 1120 | rc = xen_blkif_xenbus_init(); |
8770b268 KRW |
1121 | if (rc) |
1122 | goto failed_init; | |
4d05a28d KRW |
1123 | |
1124 | return 0; | |
1125 | ||
1126 | out_of_memory: | |
22b20f2d | 1127 | pr_alert(DRV_PFX "%s: out of memory\n", __func__); |
8770b268 | 1128 | failed_init: |
e8e28871 | 1129 | kfree(blkbk->pending_reqs); |
a742b02c | 1130 | kfree(blkbk->pending_grant_handles); |
9b83c771 DC |
1131 | if (blkbk->pending_pages) { |
1132 | for (i = 0; i < mmap_pages; i++) { | |
1133 | if (blkbk->pending_pages[i]) | |
1134 | __free_page(blkbk->pending_pages[i]); | |
1135 | } | |
1136 | kfree(blkbk->pending_pages); | |
464fb419 | 1137 | } |
a742b02c | 1138 | kfree(blkbk); |
e8e28871 | 1139 | blkbk = NULL; |
8770b268 | 1140 | return rc; |
4d05a28d KRW |
1141 | } |
1142 | ||
8b6bf747 | 1143 | module_init(xen_blkif_init); |
4d05a28d KRW |
1144 | |
1145 | MODULE_LICENSE("Dual BSD/GPL"); | |
a7e9357f | 1146 | MODULE_ALIAS("xen-backend:vbd"); |