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> |
afd91d07 | 42 | |
88122933 JF |
43 | #include <xen/events.h> |
44 | #include <xen/page.h> | |
45 | #include <asm/xen/hypervisor.h> | |
46 | #include <asm/xen/hypercall.h> | |
4d05a28d KRW |
47 | #include "common.h" |
48 | ||
49 | /* | |
50 | * These are rather arbitrary. They are fairly large because adjacent requests | |
51 | * pulled from a communication ring are quite likely to end up being part of | |
52 | * the same scatter/gather request at the disc. | |
53 | * | |
8b6bf747 | 54 | * ** TRY INCREASING 'xen_blkif_reqs' IF WRITE SPEEDS SEEM TOO LOW ** |
4d05a28d KRW |
55 | * |
56 | * This will increase the chances of being able to write whole tracks. | |
57 | * 64 should be enough to keep us competitive with Linux. | |
58 | */ | |
8b6bf747 KRW |
59 | static int xen_blkif_reqs = 64; |
60 | module_param_named(reqs, xen_blkif_reqs, int, 0); | |
4d05a28d KRW |
61 | MODULE_PARM_DESC(reqs, "Number of blkback requests to allocate"); |
62 | ||
63 | /* Run-time switchable: /sys/module/blkback/parameters/ */ | |
2e9977c2 | 64 | static unsigned int log_stats; |
4d05a28d | 65 | module_param(log_stats, int, 0644); |
4d05a28d KRW |
66 | |
67 | /* | |
68 | * Each outstanding request that we've passed to the lower device layers has a | |
69 | * 'pending_req' allocated to it. Each buffer_head that completes decrements | |
70 | * the pendcnt towards zero. When it hits zero, the specified domain has a | |
71 | * response queued for it, with the saved 'id' passed back. | |
72 | */ | |
2e9977c2 | 73 | struct pending_req { |
30fd1502 | 74 | struct xen_blkif *blkif; |
01f37f2d KRW |
75 | u64 id; |
76 | int nr_pages; | |
77 | atomic_t pendcnt; | |
78 | unsigned short operation; | |
79 | int status; | |
80 | struct list_head free_list; | |
2e9977c2 | 81 | }; |
4d05a28d | 82 | |
4d05a28d KRW |
83 | #define BLKBACK_INVALID_HANDLE (~0) |
84 | ||
e8e28871 | 85 | struct xen_blkbk { |
2e9977c2 | 86 | struct pending_req *pending_reqs; |
a1397fa3 | 87 | /* List of all 'pending_req' available */ |
e8e28871 | 88 | struct list_head pending_free; |
a1397fa3 | 89 | /* And its spinlock. */ |
e8e28871 KRW |
90 | spinlock_t pending_free_lock; |
91 | wait_queue_head_t pending_free_wq; | |
a1397fa3 | 92 | /* The list of all pages that are available. */ |
e8e28871 | 93 | struct page **pending_pages; |
a1397fa3 | 94 | /* And the grant handles that are available. */ |
e8e28871 KRW |
95 | grant_handle_t *pending_grant_handles; |
96 | }; | |
97 | ||
98 | static struct xen_blkbk *blkbk; | |
4d05a28d | 99 | |
a1397fa3 KRW |
100 | /* |
101 | * Little helpful macro to figure out the index and virtual address of the | |
102 | * pending_pages[..]. For each 'pending_req' we have have up to | |
103 | * BLKIF_MAX_SEGMENTS_PER_REQUEST (11) pages. The seg would be from 0 through | |
01f37f2d KRW |
104 | * 10 and would index in the pending_pages[..]. |
105 | */ | |
2e9977c2 | 106 | static inline int vaddr_pagenr(struct pending_req *req, int seg) |
4d05a28d | 107 | { |
2e9977c2 KRW |
108 | return (req - blkbk->pending_reqs) * |
109 | BLKIF_MAX_SEGMENTS_PER_REQUEST + seg; | |
4d05a28d KRW |
110 | } |
111 | ||
efe08a3e JB |
112 | #define pending_page(req, seg) pending_pages[vaddr_pagenr(req, seg)] |
113 | ||
2e9977c2 | 114 | static inline unsigned long vaddr(struct pending_req *req, int seg) |
4d05a28d | 115 | { |
e8e28871 | 116 | unsigned long pfn = page_to_pfn(blkbk->pending_page(req, seg)); |
4d05a28d KRW |
117 | return (unsigned long)pfn_to_kaddr(pfn); |
118 | } | |
119 | ||
120 | #define pending_handle(_req, _seg) \ | |
e8e28871 | 121 | (blkbk->pending_grant_handles[vaddr_pagenr(_req, _seg)]) |
4d05a28d KRW |
122 | |
123 | ||
30fd1502 KRW |
124 | static int do_block_io_op(struct xen_blkif *blkif); |
125 | static int dispatch_rw_block_io(struct xen_blkif *blkif, | |
fc53bf75 KRW |
126 | struct blkif_request *req, |
127 | struct pending_req *pending_req); | |
30fd1502 | 128 | static void make_response(struct xen_blkif *blkif, u64 id, |
4d05a28d KRW |
129 | unsigned short op, int st); |
130 | ||
a1397fa3 KRW |
131 | /* |
132 | * Retrieve from the 'pending_reqs' a free pending_req structure to be used. | |
4d05a28d | 133 | */ |
2e9977c2 | 134 | static struct pending_req *alloc_req(void) |
4d05a28d | 135 | { |
2e9977c2 | 136 | struct pending_req *req = NULL; |
4d05a28d KRW |
137 | unsigned long flags; |
138 | ||
e8e28871 KRW |
139 | spin_lock_irqsave(&blkbk->pending_free_lock, flags); |
140 | if (!list_empty(&blkbk->pending_free)) { | |
2e9977c2 KRW |
141 | req = list_entry(blkbk->pending_free.next, struct pending_req, |
142 | free_list); | |
4d05a28d KRW |
143 | list_del(&req->free_list); |
144 | } | |
e8e28871 | 145 | spin_unlock_irqrestore(&blkbk->pending_free_lock, flags); |
4d05a28d KRW |
146 | return req; |
147 | } | |
148 | ||
a1397fa3 KRW |
149 | /* |
150 | * Return the 'pending_req' structure back to the freepool. We also | |
151 | * wake up the thread if it was waiting for a free page. | |
152 | */ | |
2e9977c2 | 153 | static void free_req(struct pending_req *req) |
4d05a28d KRW |
154 | { |
155 | unsigned long flags; | |
156 | int was_empty; | |
157 | ||
e8e28871 KRW |
158 | spin_lock_irqsave(&blkbk->pending_free_lock, flags); |
159 | was_empty = list_empty(&blkbk->pending_free); | |
160 | list_add(&req->free_list, &blkbk->pending_free); | |
161 | spin_unlock_irqrestore(&blkbk->pending_free_lock, flags); | |
4d05a28d | 162 | if (was_empty) |
e8e28871 | 163 | wake_up(&blkbk->pending_free_wq); |
4d05a28d KRW |
164 | } |
165 | ||
ee9ff853 KRW |
166 | /* |
167 | * Routines for managing virtual block devices (vbds). | |
168 | */ | |
3d814731 KRW |
169 | static int xen_vbd_translate(struct phys_req *req, struct xen_blkif *blkif, |
170 | int operation) | |
ee9ff853 | 171 | { |
3d814731 | 172 | struct xen_vbd *vbd = &blkif->vbd; |
ee9ff853 KRW |
173 | int rc = -EACCES; |
174 | ||
175 | if ((operation != READ) && vbd->readonly) | |
176 | goto out; | |
177 | ||
178 | if (unlikely((req->sector_number + req->nr_sects) > vbd_sz(vbd))) | |
179 | goto out; | |
180 | ||
181 | req->dev = vbd->pdevice; | |
182 | req->bdev = vbd->bdev; | |
183 | rc = 0; | |
184 | ||
185 | out: | |
186 | return rc; | |
187 | } | |
188 | ||
3d814731 | 189 | static void xen_vbd_resize(struct xen_blkif *blkif) |
ee9ff853 | 190 | { |
3d814731 | 191 | struct xen_vbd *vbd = &blkif->vbd; |
ee9ff853 KRW |
192 | struct xenbus_transaction xbt; |
193 | int err; | |
8b6bf747 | 194 | struct xenbus_device *dev = xen_blkbk_xenbus(blkif->be); |
42c7841d | 195 | unsigned long long new_size = vbd_sz(vbd); |
ee9ff853 | 196 | |
22b20f2d | 197 | pr_info(DRV_PFX "VBD Resize: Domid: %d, Device: (%d, %d)\n", |
ee9ff853 | 198 | blkif->domid, MAJOR(vbd->pdevice), MINOR(vbd->pdevice)); |
22b20f2d | 199 | pr_info(DRV_PFX "VBD Resize: new size %llu\n", new_size); |
ee9ff853 KRW |
200 | vbd->size = new_size; |
201 | again: | |
202 | err = xenbus_transaction_start(&xbt); | |
203 | if (err) { | |
22b20f2d | 204 | pr_warn(DRV_PFX "Error starting transaction"); |
ee9ff853 KRW |
205 | return; |
206 | } | |
207 | err = xenbus_printf(xbt, dev->nodename, "sectors", "%llu", | |
42c7841d | 208 | (unsigned long long)vbd_sz(vbd)); |
ee9ff853 | 209 | if (err) { |
22b20f2d | 210 | pr_warn(DRV_PFX "Error writing new size"); |
ee9ff853 KRW |
211 | goto abort; |
212 | } | |
213 | /* | |
214 | * Write the current state; we will use this to synchronize | |
215 | * the front-end. If the current state is "connected" the | |
216 | * front-end will get the new size information online. | |
217 | */ | |
218 | err = xenbus_printf(xbt, dev->nodename, "state", "%d", dev->state); | |
219 | if (err) { | |
22b20f2d | 220 | pr_warn(DRV_PFX "Error writing the state"); |
ee9ff853 KRW |
221 | goto abort; |
222 | } | |
223 | ||
224 | err = xenbus_transaction_end(xbt, 0); | |
225 | if (err == -EAGAIN) | |
226 | goto again; | |
227 | if (err) | |
22b20f2d | 228 | pr_warn(DRV_PFX "Error ending transaction"); |
496b318e | 229 | return; |
ee9ff853 KRW |
230 | abort: |
231 | xenbus_transaction_end(xbt, 1); | |
232 | } | |
233 | ||
a1397fa3 | 234 | /* |
b0aef179 KRW |
235 | * Notification from the guest OS. |
236 | */ | |
30fd1502 | 237 | static void blkif_notify_work(struct xen_blkif *blkif) |
4d05a28d | 238 | { |
b0aef179 KRW |
239 | blkif->waiting_reqs = 1; |
240 | wake_up(&blkif->wq); | |
241 | } | |
4d05a28d | 242 | |
8b6bf747 | 243 | irqreturn_t xen_blkif_be_int(int irq, void *dev_id) |
b0aef179 KRW |
244 | { |
245 | blkif_notify_work(dev_id); | |
246 | return IRQ_HANDLED; | |
4d05a28d KRW |
247 | } |
248 | ||
2e9977c2 | 249 | /* |
4d05a28d KRW |
250 | * SCHEDULER FUNCTIONS |
251 | */ | |
252 | ||
30fd1502 | 253 | static void print_stats(struct xen_blkif *blkif) |
4d05a28d | 254 | { |
cca537af | 255 | pr_info("xen-blkback (%s): oo %3d | rd %4d | wr %4d | f %4d\n", |
ebe81906 KRW |
256 | current->comm, blkif->st_oo_req, |
257 | blkif->st_rd_req, blkif->st_wr_req, blkif->st_f_req); | |
4d05a28d KRW |
258 | blkif->st_print = jiffies + msecs_to_jiffies(10 * 1000); |
259 | blkif->st_rd_req = 0; | |
260 | blkif->st_wr_req = 0; | |
261 | blkif->st_oo_req = 0; | |
262 | } | |
263 | ||
8b6bf747 | 264 | int xen_blkif_schedule(void *arg) |
4d05a28d | 265 | { |
30fd1502 | 266 | struct xen_blkif *blkif = arg; |
3d814731 | 267 | struct xen_vbd *vbd = &blkif->vbd; |
4d05a28d | 268 | |
8b6bf747 | 269 | xen_blkif_get(blkif); |
4d05a28d | 270 | |
4d05a28d KRW |
271 | while (!kthread_should_stop()) { |
272 | if (try_to_freeze()) | |
273 | continue; | |
42c7841d | 274 | if (unlikely(vbd->size != vbd_sz(vbd))) |
3d814731 | 275 | xen_vbd_resize(blkif); |
4d05a28d KRW |
276 | |
277 | wait_event_interruptible( | |
278 | blkif->wq, | |
279 | blkif->waiting_reqs || kthread_should_stop()); | |
280 | wait_event_interruptible( | |
e8e28871 | 281 | blkbk->pending_free_wq, |
2e9977c2 KRW |
282 | !list_empty(&blkbk->pending_free) || |
283 | kthread_should_stop()); | |
4d05a28d KRW |
284 | |
285 | blkif->waiting_reqs = 0; | |
286 | smp_mb(); /* clear flag *before* checking for work */ | |
287 | ||
288 | if (do_block_io_op(blkif)) | |
289 | blkif->waiting_reqs = 1; | |
4d05a28d KRW |
290 | |
291 | if (log_stats && time_after(jiffies, blkif->st_print)) | |
292 | print_stats(blkif); | |
293 | } | |
294 | ||
295 | if (log_stats) | |
296 | print_stats(blkif); | |
4d05a28d KRW |
297 | |
298 | blkif->xenblkd = NULL; | |
8b6bf747 | 299 | xen_blkif_put(blkif); |
4d05a28d KRW |
300 | |
301 | return 0; | |
302 | } | |
303 | ||
1a95fe6e KRW |
304 | struct seg_buf { |
305 | unsigned long buf; | |
306 | unsigned int nsec; | |
307 | }; | |
b0aef179 KRW |
308 | /* |
309 | * Unmap the grant references, and also remove the M2P over-rides | |
310 | * used in the 'pending_req'. | |
01f37f2d | 311 | */ |
9f3aedf5 | 312 | static void xen_blkbk_unmap(struct pending_req *req) |
b0aef179 KRW |
313 | { |
314 | struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
315 | unsigned int i, invcount = 0; | |
316 | grant_handle_t handle; | |
317 | int ret; | |
318 | ||
319 | for (i = 0; i < req->nr_pages; i++) { | |
320 | handle = pending_handle(req, i); | |
321 | if (handle == BLKBACK_INVALID_HANDLE) | |
322 | continue; | |
323 | gnttab_set_unmap_op(&unmap[invcount], vaddr(req, i), | |
324 | GNTMAP_host_map, handle); | |
325 | pending_handle(req, i) = BLKBACK_INVALID_HANDLE; | |
326 | invcount++; | |
327 | } | |
328 | ||
329 | ret = HYPERVISOR_grant_table_op( | |
330 | GNTTABOP_unmap_grant_ref, unmap, invcount); | |
331 | BUG_ON(ret); | |
01f37f2d KRW |
332 | /* |
333 | * Note, we use invcount, so nr->pages, so we can't index | |
b0aef179 KRW |
334 | * using vaddr(req, i). |
335 | */ | |
336 | for (i = 0; i < invcount; i++) { | |
337 | ret = m2p_remove_override( | |
338 | virt_to_page(unmap[i].host_addr), false); | |
339 | if (ret) { | |
22b20f2d | 340 | pr_alert(DRV_PFX "Failed to remove M2P override for %lx\n", |
ebe81906 | 341 | (unsigned long)unmap[i].host_addr); |
b0aef179 KRW |
342 | continue; |
343 | } | |
344 | } | |
345 | } | |
01f37f2d KRW |
346 | |
347 | static int xen_blkbk_map(struct blkif_request *req, | |
348 | struct pending_req *pending_req, | |
9f3aedf5 | 349 | struct seg_buf seg[]) |
1a95fe6e KRW |
350 | { |
351 | struct gnttab_map_grant_ref map[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
352 | int i; | |
353 | int nseg = req->nr_segments; | |
354 | int ret = 0; | |
01f37f2d KRW |
355 | |
356 | /* | |
357 | * Fill out preq.nr_sects with proper amount of sectors, and setup | |
1a95fe6e KRW |
358 | * assign map[..] with the PFN of the page in our domain with the |
359 | * corresponding grant reference for each page. | |
360 | */ | |
361 | for (i = 0; i < nseg; i++) { | |
362 | uint32_t flags; | |
363 | ||
364 | flags = GNTMAP_host_map; | |
365 | if (pending_req->operation != BLKIF_OP_READ) | |
366 | flags |= GNTMAP_readonly; | |
367 | gnttab_set_map_op(&map[i], vaddr(pending_req, i), flags, | |
01f37f2d KRW |
368 | req->u.rw.seg[i].gref, |
369 | pending_req->blkif->domid); | |
1a95fe6e KRW |
370 | } |
371 | ||
372 | ret = HYPERVISOR_grant_table_op(GNTTABOP_map_grant_ref, map, nseg); | |
373 | BUG_ON(ret); | |
374 | ||
01f37f2d KRW |
375 | /* |
376 | * Now swizzle the MFN in our domain with the MFN from the other domain | |
1a95fe6e KRW |
377 | * so that when we access vaddr(pending_req,i) it has the contents of |
378 | * the page from the other domain. | |
379 | */ | |
380 | for (i = 0; i < nseg; i++) { | |
381 | if (unlikely(map[i].status != 0)) { | |
22b20f2d | 382 | pr_debug(DRV_PFX "invalid buffer -- could not remap it\n"); |
1a95fe6e KRW |
383 | map[i].handle = BLKBACK_INVALID_HANDLE; |
384 | ret |= 1; | |
385 | } | |
386 | ||
387 | pending_handle(pending_req, i) = map[i].handle; | |
388 | ||
389 | if (ret) | |
390 | continue; | |
391 | ||
392 | ret = m2p_add_override(PFN_DOWN(map[i].dev_bus_addr), | |
393 | blkbk->pending_page(pending_req, i), false); | |
394 | if (ret) { | |
22b20f2d | 395 | pr_alert(DRV_PFX "Failed to install M2P override for %lx (ret: %d)\n", |
ebe81906 | 396 | (unsigned long)map[i].dev_bus_addr, ret); |
1a95fe6e KRW |
397 | /* We could switch over to GNTTABOP_copy */ |
398 | continue; | |
399 | } | |
400 | ||
401 | seg[i].buf = map[i].dev_bus_addr | | |
402 | (req->u.rw.seg[i].first_sect << 9); | |
403 | } | |
404 | return ret; | |
405 | } | |
406 | ||
a1397fa3 KRW |
407 | /* |
408 | * Completion callback on the bio's. Called as bh->b_end_io() | |
4d05a28d KRW |
409 | */ |
410 | ||
2e9977c2 | 411 | static void __end_block_io_op(struct pending_req *pending_req, int error) |
4d05a28d KRW |
412 | { |
413 | /* An error fails the entire request. */ | |
24f567f9 | 414 | if ((pending_req->operation == BLKIF_OP_FLUSH_DISKCACHE) && |
4d05a28d | 415 | (error == -EOPNOTSUPP)) { |
22b20f2d | 416 | pr_debug(DRV_PFX "flush diskcache op failed, not supported\n"); |
24f567f9 | 417 | xen_blkbk_flush_diskcache(XBT_NIL, pending_req->blkif->be, 0); |
4d05a28d KRW |
418 | pending_req->status = BLKIF_RSP_EOPNOTSUPP; |
419 | } else if (error) { | |
22b20f2d | 420 | pr_debug(DRV_PFX "Buffer not up-to-date at end of operation," |
ebe81906 | 421 | " error=%d\n", error); |
4d05a28d KRW |
422 | pending_req->status = BLKIF_RSP_ERROR; |
423 | } | |
424 | ||
01f37f2d KRW |
425 | /* |
426 | * If all of the bio's have completed it is time to unmap | |
a1397fa3 | 427 | * the grant references associated with 'request' and provide |
2e9977c2 KRW |
428 | * the proper response on the ring. |
429 | */ | |
4d05a28d | 430 | if (atomic_dec_and_test(&pending_req->pendcnt)) { |
9f3aedf5 | 431 | xen_blkbk_unmap(pending_req); |
4d05a28d KRW |
432 | make_response(pending_req->blkif, pending_req->id, |
433 | pending_req->operation, pending_req->status); | |
8b6bf747 | 434 | xen_blkif_put(pending_req->blkif); |
4d05a28d KRW |
435 | free_req(pending_req); |
436 | } | |
437 | } | |
438 | ||
a1397fa3 KRW |
439 | /* |
440 | * bio callback. | |
441 | */ | |
88122933 | 442 | static void end_block_io_op(struct bio *bio, int error) |
4d05a28d | 443 | { |
4d05a28d KRW |
444 | __end_block_io_op(bio->bi_private, error); |
445 | bio_put(bio); | |
4d05a28d KRW |
446 | } |
447 | ||
448 | ||
4d05a28d | 449 | |
a1397fa3 KRW |
450 | /* |
451 | * Function to copy the from the ring buffer the 'struct blkif_request' | |
452 | * (which has the sectors we want, number of them, grant references, etc), | |
453 | * and transmute it to the block API to hand it over to the proper block disk. | |
4d05a28d | 454 | */ |
30fd1502 | 455 | static int do_block_io_op(struct xen_blkif *blkif) |
4d05a28d | 456 | { |
88122933 JF |
457 | union blkif_back_rings *blk_rings = &blkif->blk_rings; |
458 | struct blkif_request req; | |
2e9977c2 | 459 | struct pending_req *pending_req; |
4d05a28d KRW |
460 | RING_IDX rc, rp; |
461 | int more_to_do = 0; | |
462 | ||
463 | rc = blk_rings->common.req_cons; | |
464 | rp = blk_rings->common.sring->req_prod; | |
465 | rmb(); /* Ensure we see queued requests up to 'rp'. */ | |
466 | ||
467 | while (rc != rp) { | |
468 | ||
469 | if (RING_REQUEST_CONS_OVERFLOW(&blk_rings->common, rc)) | |
470 | break; | |
471 | ||
8270b45b | 472 | if (kthread_should_stop()) { |
4d05a28d KRW |
473 | more_to_do = 1; |
474 | break; | |
475 | } | |
476 | ||
8270b45b KF |
477 | pending_req = alloc_req(); |
478 | if (NULL == pending_req) { | |
479 | blkif->st_oo_req++; | |
4d05a28d KRW |
480 | more_to_do = 1; |
481 | break; | |
482 | } | |
483 | ||
484 | switch (blkif->blk_protocol) { | |
485 | case BLKIF_PROTOCOL_NATIVE: | |
486 | memcpy(&req, RING_GET_REQUEST(&blk_rings->native, rc), sizeof(req)); | |
487 | break; | |
488 | case BLKIF_PROTOCOL_X86_32: | |
489 | blkif_get_x86_32_req(&req, RING_GET_REQUEST(&blk_rings->x86_32, rc)); | |
490 | break; | |
491 | case BLKIF_PROTOCOL_X86_64: | |
492 | blkif_get_x86_64_req(&req, RING_GET_REQUEST(&blk_rings->x86_64, rc)); | |
493 | break; | |
494 | default: | |
495 | BUG(); | |
496 | } | |
497 | blk_rings->common.req_cons = ++rc; /* before make_response() */ | |
498 | ||
499 | /* Apply all sanity checks to /private copy/ of request. */ | |
500 | barrier(); | |
501 | ||
fc53bf75 | 502 | if (dispatch_rw_block_io(blkif, &req, pending_req)) |
4d05a28d | 503 | break; |
4d05a28d KRW |
504 | |
505 | /* Yield point for this unbounded loop. */ | |
506 | cond_resched(); | |
507 | } | |
508 | ||
509 | return more_to_do; | |
510 | } | |
511 | ||
a1397fa3 | 512 | /* |
01f37f2d KRW |
513 | * Transmutation of the 'struct blkif_request' to a proper 'struct bio' |
514 | * and call the 'submit_bio' to pass it to the underlying storage. | |
a1397fa3 | 515 | */ |
30fd1502 KRW |
516 | static int dispatch_rw_block_io(struct xen_blkif *blkif, |
517 | struct blkif_request *req, | |
518 | struct pending_req *pending_req) | |
4d05a28d | 519 | { |
4d05a28d | 520 | struct phys_req preq; |
1a95fe6e | 521 | struct seg_buf seg[BLKIF_MAX_SEGMENTS_PER_REQUEST]; |
4d05a28d KRW |
522 | unsigned int nseg; |
523 | struct bio *bio = NULL; | |
77089926 | 524 | struct bio *biolist[BLKIF_MAX_SEGMENTS_PER_REQUEST]; |
1a95fe6e | 525 | int i, nbio = 0; |
4d05a28d | 526 | int operation; |
a19be5f0 | 527 | struct blk_plug plug; |
4d05a28d KRW |
528 | |
529 | switch (req->operation) { | |
530 | case BLKIF_OP_READ: | |
fc53bf75 | 531 | blkif->st_rd_req++; |
4d05a28d KRW |
532 | operation = READ; |
533 | break; | |
534 | case BLKIF_OP_WRITE: | |
fc53bf75 | 535 | blkif->st_wr_req++; |
013c3ca1 | 536 | operation = WRITE_ODIRECT; |
4d05a28d | 537 | break; |
24f567f9 | 538 | case BLKIF_OP_FLUSH_DISKCACHE: |
fc53bf75 | 539 | blkif->st_f_req++; |
24f567f9 | 540 | operation = WRITE_FLUSH; |
01f37f2d | 541 | /* |
3d814731 | 542 | * The frontend likes to set this to -1, which xen_vbd_translate |
01f37f2d KRW |
543 | * is alergic too. |
544 | */ | |
24f567f9 | 545 | req->u.rw.sector_number = 0; |
4d05a28d | 546 | break; |
24f567f9 | 547 | case BLKIF_OP_WRITE_BARRIER: |
4d05a28d KRW |
548 | default: |
549 | operation = 0; /* make gcc happy */ | |
fc53bf75 KRW |
550 | goto fail_response; |
551 | break; | |
4d05a28d KRW |
552 | } |
553 | ||
a1397fa3 | 554 | /* Check that the number of segments is sane. */ |
4d05a28d | 555 | nseg = req->nr_segments; |
24f567f9 | 556 | if (unlikely(nseg == 0 && operation != WRITE_FLUSH) || |
4d05a28d | 557 | unlikely(nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST)) { |
22b20f2d | 558 | pr_debug(DRV_PFX "Bad number of segments in request (%d)\n", |
ebe81906 | 559 | nseg); |
1a95fe6e | 560 | /* Haven't submitted any bio's yet. */ |
4d05a28d KRW |
561 | goto fail_response; |
562 | } | |
563 | ||
564 | preq.dev = req->handle; | |
c35950bf | 565 | preq.sector_number = req->u.rw.sector_number; |
4d05a28d KRW |
566 | preq.nr_sects = 0; |
567 | ||
568 | pending_req->blkif = blkif; | |
569 | pending_req->id = req->id; | |
570 | pending_req->operation = req->operation; | |
571 | pending_req->status = BLKIF_RSP_OKAY; | |
572 | pending_req->nr_pages = nseg; | |
e9350493 | 573 | |
4d05a28d | 574 | for (i = 0; i < nseg; i++) { |
c35950bf KRW |
575 | seg[i].nsec = req->u.rw.seg[i].last_sect - |
576 | req->u.rw.seg[i].first_sect + 1; | |
c35950bf KRW |
577 | if ((req->u.rw.seg[i].last_sect >= (PAGE_SIZE >> 9)) || |
578 | (req->u.rw.seg[i].last_sect < req->u.rw.seg[i].first_sect)) | |
4d05a28d KRW |
579 | goto fail_response; |
580 | preq.nr_sects += seg[i].nsec; | |
976222e0 | 581 | |
4d05a28d KRW |
582 | } |
583 | ||
3d814731 | 584 | if (xen_vbd_translate(&preq, blkif, operation) != 0) { |
22b20f2d | 585 | pr_debug(DRV_PFX "access denied: %s of [%llu,%llu] on dev=%04x\n", |
ebe81906 KRW |
586 | operation == READ ? "read" : "write", |
587 | preq.sector_number, | |
588 | preq.sector_number + preq.nr_sects, preq.dev); | |
1a95fe6e | 589 | goto fail_response; |
4d05a28d | 590 | } |
01f37f2d KRW |
591 | |
592 | /* | |
3d814731 | 593 | * This check _MUST_ be done after xen_vbd_translate as the preq.bdev |
01f37f2d KRW |
594 | * is set there. |
595 | */ | |
e9350493 KRW |
596 | for (i = 0; i < nseg; i++) { |
597 | if (((int)preq.sector_number|(int)seg[i].nsec) & | |
598 | ((bdev_logical_block_size(preq.bdev) >> 9) - 1)) { | |
22b20f2d | 599 | pr_debug(DRV_PFX "Misaligned I/O request from domain %d", |
ebe81906 | 600 | blkif->domid); |
e9350493 KRW |
601 | goto fail_response; |
602 | } | |
603 | } | |
01f37f2d KRW |
604 | |
605 | /* | |
606 | * If we have failed at this point, we need to undo the M2P override, | |
2e9977c2 KRW |
607 | * set gnttab_set_unmap_op on all of the grant references and perform |
608 | * the hypercall to unmap the grants - that is all done in | |
9f3aedf5 | 609 | * xen_blkbk_unmap. |
2e9977c2 | 610 | */ |
9f3aedf5 | 611 | if (xen_blkbk_map(req, pending_req, seg)) |
4d05a28d KRW |
612 | goto fail_flush; |
613 | ||
3d68b399 | 614 | /* This corresponding xen_blkif_put is done in __end_block_io_op */ |
8b6bf747 | 615 | xen_blkif_get(blkif); |
4d05a28d KRW |
616 | |
617 | for (i = 0; i < nseg; i++) { | |
4d05a28d KRW |
618 | while ((bio == NULL) || |
619 | (bio_add_page(bio, | |
e8e28871 | 620 | blkbk->pending_page(pending_req, i), |
4d05a28d KRW |
621 | seg[i].nsec << 9, |
622 | seg[i].buf & ~PAGE_MASK) == 0)) { | |
2e9977c2 | 623 | |
03e0edf9 | 624 | bio = bio_alloc(GFP_KERNEL, nseg-i); |
4d05a28d KRW |
625 | if (unlikely(bio == NULL)) |
626 | goto fail_put_bio; | |
627 | ||
03e0edf9 | 628 | biolist[nbio++] = bio; |
4d05a28d KRW |
629 | bio->bi_bdev = preq.bdev; |
630 | bio->bi_private = pending_req; | |
631 | bio->bi_end_io = end_block_io_op; | |
632 | bio->bi_sector = preq.sector_number; | |
633 | } | |
634 | ||
635 | preq.sector_number += seg[i].nsec; | |
636 | } | |
637 | ||
3d68b399 | 638 | /* This will be hit if the operation was a flush. */ |
4d05a28d | 639 | if (!bio) { |
24f567f9 | 640 | BUG_ON(operation != WRITE_FLUSH); |
b0f80127 | 641 | |
03e0edf9 | 642 | bio = bio_alloc(GFP_KERNEL, 0); |
4d05a28d KRW |
643 | if (unlikely(bio == NULL)) |
644 | goto fail_put_bio; | |
645 | ||
03e0edf9 | 646 | biolist[nbio++] = bio; |
4d05a28d KRW |
647 | bio->bi_bdev = preq.bdev; |
648 | bio->bi_private = pending_req; | |
649 | bio->bi_end_io = end_block_io_op; | |
4d05a28d KRW |
650 | } |
651 | ||
01f37f2d KRW |
652 | /* |
653 | * We set it one so that the last submit_bio does not have to call | |
77089926 KRW |
654 | * atomic_inc. |
655 | */ | |
656 | atomic_set(&pending_req->pendcnt, nbio); | |
657 | ||
a19be5f0 KRW |
658 | /* Get a reference count for the disk queue and start sending I/O */ |
659 | blk_start_plug(&plug); | |
660 | ||
77089926 KRW |
661 | for (i = 0; i < nbio; i++) |
662 | submit_bio(operation, biolist[i]); | |
663 | ||
a19be5f0 | 664 | /* Let the I/Os go.. */ |
3d68b399 | 665 | blk_finish_plug(&plug); |
a19be5f0 | 666 | |
4d05a28d KRW |
667 | if (operation == READ) |
668 | blkif->st_rd_sect += preq.nr_sects; | |
24f567f9 | 669 | else if (operation == WRITE || operation == WRITE_FLUSH) |
4d05a28d KRW |
670 | blkif->st_wr_sect += preq.nr_sects; |
671 | ||
fc53bf75 | 672 | return 0; |
4d05a28d KRW |
673 | |
674 | fail_flush: | |
9f3aedf5 | 675 | xen_blkbk_unmap(pending_req); |
4d05a28d | 676 | fail_response: |
0faa8cca | 677 | /* Haven't submitted any bio's yet. */ |
4d05a28d KRW |
678 | make_response(blkif, req->id, req->operation, BLKIF_RSP_ERROR); |
679 | free_req(pending_req); | |
680 | msleep(1); /* back off a bit */ | |
fc53bf75 | 681 | return -EIO; |
4d05a28d KRW |
682 | |
683 | fail_put_bio: | |
03e0edf9 | 684 | for (i = 0; i < nbio; i++) |
77089926 | 685 | bio_put(biolist[i]); |
4d05a28d | 686 | __end_block_io_op(pending_req, -EINVAL); |
4d05a28d | 687 | msleep(1); /* back off a bit */ |
fc53bf75 | 688 | return -EIO; |
4d05a28d KRW |
689 | } |
690 | ||
691 | ||
692 | ||
a1397fa3 KRW |
693 | /* |
694 | * Put a response on the ring on how the operation fared. | |
4d05a28d | 695 | */ |
30fd1502 | 696 | static void make_response(struct xen_blkif *blkif, u64 id, |
4d05a28d KRW |
697 | unsigned short op, int st) |
698 | { | |
88122933 | 699 | struct blkif_response resp; |
4d05a28d | 700 | unsigned long flags; |
88122933 | 701 | union blkif_back_rings *blk_rings = &blkif->blk_rings; |
4d05a28d KRW |
702 | int more_to_do = 0; |
703 | int notify; | |
704 | ||
705 | resp.id = id; | |
706 | resp.operation = op; | |
707 | resp.status = st; | |
708 | ||
709 | spin_lock_irqsave(&blkif->blk_ring_lock, flags); | |
710 | /* Place on the response ring for the relevant domain. */ | |
711 | switch (blkif->blk_protocol) { | |
712 | case BLKIF_PROTOCOL_NATIVE: | |
713 | memcpy(RING_GET_RESPONSE(&blk_rings->native, blk_rings->native.rsp_prod_pvt), | |
714 | &resp, sizeof(resp)); | |
715 | break; | |
716 | case BLKIF_PROTOCOL_X86_32: | |
717 | memcpy(RING_GET_RESPONSE(&blk_rings->x86_32, blk_rings->x86_32.rsp_prod_pvt), | |
718 | &resp, sizeof(resp)); | |
719 | break; | |
720 | case BLKIF_PROTOCOL_X86_64: | |
721 | memcpy(RING_GET_RESPONSE(&blk_rings->x86_64, blk_rings->x86_64.rsp_prod_pvt), | |
722 | &resp, sizeof(resp)); | |
723 | break; | |
724 | default: | |
725 | BUG(); | |
726 | } | |
727 | blk_rings->common.rsp_prod_pvt++; | |
728 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blk_rings->common, notify); | |
729 | if (blk_rings->common.rsp_prod_pvt == blk_rings->common.req_cons) { | |
730 | /* | |
731 | * Tail check for pending requests. Allows frontend to avoid | |
732 | * notifications if requests are already in flight (lower | |
733 | * overheads and promotes batching). | |
734 | */ | |
735 | RING_FINAL_CHECK_FOR_REQUESTS(&blk_rings->common, more_to_do); | |
736 | ||
737 | } else if (RING_HAS_UNCONSUMED_REQUESTS(&blk_rings->common)) { | |
738 | more_to_do = 1; | |
739 | } | |
740 | ||
741 | spin_unlock_irqrestore(&blkif->blk_ring_lock, flags); | |
742 | ||
743 | if (more_to_do) | |
744 | blkif_notify_work(blkif); | |
745 | if (notify) | |
746 | notify_remote_via_irq(blkif->irq); | |
747 | } | |
748 | ||
8b6bf747 | 749 | static int __init xen_blkif_init(void) |
4d05a28d KRW |
750 | { |
751 | int i, mmap_pages; | |
8770b268 | 752 | int rc = 0; |
4d05a28d | 753 | |
88122933 | 754 | if (!xen_pv_domain()) |
4d05a28d KRW |
755 | return -ENODEV; |
756 | ||
2e9977c2 | 757 | blkbk = kzalloc(sizeof(struct xen_blkbk), GFP_KERNEL); |
e8e28871 | 758 | if (!blkbk) { |
22b20f2d | 759 | pr_alert(DRV_PFX "%s: out of memory!\n", __func__); |
e8e28871 KRW |
760 | return -ENOMEM; |
761 | } | |
762 | ||
8b6bf747 | 763 | mmap_pages = xen_blkif_reqs * BLKIF_MAX_SEGMENTS_PER_REQUEST; |
4d05a28d | 764 | |
e8e28871 | 765 | blkbk->pending_reqs = kmalloc(sizeof(blkbk->pending_reqs[0]) * |
8b6bf747 | 766 | xen_blkif_reqs, GFP_KERNEL); |
a742b02c KRW |
767 | blkbk->pending_grant_handles = kzalloc(sizeof(blkbk->pending_grant_handles[0]) * |
768 | mmap_pages, GFP_KERNEL); | |
769 | blkbk->pending_pages = kzalloc(sizeof(blkbk->pending_pages[0]) * | |
770 | mmap_pages, GFP_KERNEL); | |
4d05a28d | 771 | |
2e9977c2 KRW |
772 | if (!blkbk->pending_reqs || !blkbk->pending_grant_handles || |
773 | !blkbk->pending_pages) { | |
8770b268 | 774 | rc = -ENOMEM; |
4d05a28d | 775 | goto out_of_memory; |
8770b268 | 776 | } |
4d05a28d | 777 | |
464fb419 | 778 | for (i = 0; i < mmap_pages; i++) { |
e8e28871 | 779 | blkbk->pending_grant_handles[i] = BLKBACK_INVALID_HANDLE; |
a742b02c | 780 | blkbk->pending_pages[i] = alloc_page(GFP_KERNEL); |
464fb419 KRW |
781 | if (blkbk->pending_pages[i] == NULL) { |
782 | rc = -ENOMEM; | |
783 | goto out_of_memory; | |
784 | } | |
785 | } | |
8b6bf747 | 786 | rc = xen_blkif_interface_init(); |
8770b268 KRW |
787 | if (rc) |
788 | goto failed_init; | |
4d05a28d | 789 | |
e8e28871 KRW |
790 | memset(blkbk->pending_reqs, 0, sizeof(blkbk->pending_reqs)); |
791 | ||
792 | INIT_LIST_HEAD(&blkbk->pending_free); | |
793 | spin_lock_init(&blkbk->pending_free_lock); | |
794 | init_waitqueue_head(&blkbk->pending_free_wq); | |
4d05a28d | 795 | |
8b6bf747 | 796 | for (i = 0; i < xen_blkif_reqs; i++) |
2e9977c2 KRW |
797 | list_add_tail(&blkbk->pending_reqs[i].free_list, |
798 | &blkbk->pending_free); | |
4d05a28d | 799 | |
8b6bf747 | 800 | rc = xen_blkif_xenbus_init(); |
8770b268 KRW |
801 | if (rc) |
802 | goto failed_init; | |
4d05a28d KRW |
803 | |
804 | return 0; | |
805 | ||
806 | out_of_memory: | |
22b20f2d | 807 | pr_alert(DRV_PFX "%s: out of memory\n", __func__); |
8770b268 | 808 | failed_init: |
e8e28871 | 809 | kfree(blkbk->pending_reqs); |
a742b02c | 810 | kfree(blkbk->pending_grant_handles); |
464fb419 KRW |
811 | for (i = 0; i < mmap_pages; i++) { |
812 | if (blkbk->pending_pages[i]) | |
813 | __free_page(blkbk->pending_pages[i]); | |
814 | } | |
a742b02c KRW |
815 | kfree(blkbk->pending_pages); |
816 | kfree(blkbk); | |
e8e28871 | 817 | blkbk = NULL; |
8770b268 | 818 | return rc; |
4d05a28d KRW |
819 | } |
820 | ||
8b6bf747 | 821 | module_init(xen_blkif_init); |
4d05a28d KRW |
822 | |
823 | MODULE_LICENSE("Dual BSD/GPL"); |