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"); |
ee9ff853 KRW |
229 | abort: |
230 | xenbus_transaction_end(xbt, 1); | |
231 | } | |
232 | ||
a1397fa3 | 233 | /* |
b0aef179 KRW |
234 | * Notification from the guest OS. |
235 | */ | |
30fd1502 | 236 | static void blkif_notify_work(struct xen_blkif *blkif) |
4d05a28d | 237 | { |
b0aef179 KRW |
238 | blkif->waiting_reqs = 1; |
239 | wake_up(&blkif->wq); | |
240 | } | |
4d05a28d | 241 | |
8b6bf747 | 242 | irqreturn_t xen_blkif_be_int(int irq, void *dev_id) |
b0aef179 KRW |
243 | { |
244 | blkif_notify_work(dev_id); | |
245 | return IRQ_HANDLED; | |
4d05a28d KRW |
246 | } |
247 | ||
2e9977c2 | 248 | /* |
4d05a28d KRW |
249 | * SCHEDULER FUNCTIONS |
250 | */ | |
251 | ||
30fd1502 | 252 | static void print_stats(struct xen_blkif *blkif) |
4d05a28d | 253 | { |
cca537af | 254 | pr_info("xen-blkback (%s): oo %3d | rd %4d | wr %4d | f %4d\n", |
ebe81906 KRW |
255 | current->comm, blkif->st_oo_req, |
256 | blkif->st_rd_req, blkif->st_wr_req, blkif->st_f_req); | |
4d05a28d KRW |
257 | blkif->st_print = jiffies + msecs_to_jiffies(10 * 1000); |
258 | blkif->st_rd_req = 0; | |
259 | blkif->st_wr_req = 0; | |
260 | blkif->st_oo_req = 0; | |
261 | } | |
262 | ||
8b6bf747 | 263 | int xen_blkif_schedule(void *arg) |
4d05a28d | 264 | { |
30fd1502 | 265 | struct xen_blkif *blkif = arg; |
3d814731 | 266 | struct xen_vbd *vbd = &blkif->vbd; |
4d05a28d | 267 | |
8b6bf747 | 268 | xen_blkif_get(blkif); |
4d05a28d | 269 | |
4d05a28d KRW |
270 | while (!kthread_should_stop()) { |
271 | if (try_to_freeze()) | |
272 | continue; | |
42c7841d | 273 | if (unlikely(vbd->size != vbd_sz(vbd))) |
3d814731 | 274 | xen_vbd_resize(blkif); |
4d05a28d KRW |
275 | |
276 | wait_event_interruptible( | |
277 | blkif->wq, | |
278 | blkif->waiting_reqs || kthread_should_stop()); | |
279 | wait_event_interruptible( | |
e8e28871 | 280 | blkbk->pending_free_wq, |
2e9977c2 KRW |
281 | !list_empty(&blkbk->pending_free) || |
282 | kthread_should_stop()); | |
4d05a28d KRW |
283 | |
284 | blkif->waiting_reqs = 0; | |
285 | smp_mb(); /* clear flag *before* checking for work */ | |
286 | ||
287 | if (do_block_io_op(blkif)) | |
288 | blkif->waiting_reqs = 1; | |
4d05a28d KRW |
289 | |
290 | if (log_stats && time_after(jiffies, blkif->st_print)) | |
291 | print_stats(blkif); | |
292 | } | |
293 | ||
294 | if (log_stats) | |
295 | print_stats(blkif); | |
4d05a28d KRW |
296 | |
297 | blkif->xenblkd = NULL; | |
8b6bf747 | 298 | xen_blkif_put(blkif); |
4d05a28d KRW |
299 | |
300 | return 0; | |
301 | } | |
302 | ||
1a95fe6e KRW |
303 | struct seg_buf { |
304 | unsigned long buf; | |
305 | unsigned int nsec; | |
306 | }; | |
b0aef179 KRW |
307 | /* |
308 | * Unmap the grant references, and also remove the M2P over-rides | |
309 | * used in the 'pending_req'. | |
01f37f2d | 310 | */ |
9f3aedf5 | 311 | static void xen_blkbk_unmap(struct pending_req *req) |
b0aef179 KRW |
312 | { |
313 | struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
314 | unsigned int i, invcount = 0; | |
315 | grant_handle_t handle; | |
316 | int ret; | |
317 | ||
318 | for (i = 0; i < req->nr_pages; i++) { | |
319 | handle = pending_handle(req, i); | |
320 | if (handle == BLKBACK_INVALID_HANDLE) | |
321 | continue; | |
322 | gnttab_set_unmap_op(&unmap[invcount], vaddr(req, i), | |
323 | GNTMAP_host_map, handle); | |
324 | pending_handle(req, i) = BLKBACK_INVALID_HANDLE; | |
325 | invcount++; | |
326 | } | |
327 | ||
328 | ret = HYPERVISOR_grant_table_op( | |
329 | GNTTABOP_unmap_grant_ref, unmap, invcount); | |
330 | BUG_ON(ret); | |
01f37f2d KRW |
331 | /* |
332 | * Note, we use invcount, so nr->pages, so we can't index | |
b0aef179 KRW |
333 | * using vaddr(req, i). |
334 | */ | |
335 | for (i = 0; i < invcount; i++) { | |
336 | ret = m2p_remove_override( | |
337 | virt_to_page(unmap[i].host_addr), false); | |
338 | if (ret) { | |
22b20f2d | 339 | pr_alert(DRV_PFX "Failed to remove M2P override for %lx\n", |
ebe81906 | 340 | (unsigned long)unmap[i].host_addr); |
b0aef179 KRW |
341 | continue; |
342 | } | |
343 | } | |
344 | } | |
01f37f2d KRW |
345 | |
346 | static int xen_blkbk_map(struct blkif_request *req, | |
347 | struct pending_req *pending_req, | |
9f3aedf5 | 348 | struct seg_buf seg[]) |
1a95fe6e KRW |
349 | { |
350 | struct gnttab_map_grant_ref map[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
351 | int i; | |
352 | int nseg = req->nr_segments; | |
353 | int ret = 0; | |
01f37f2d KRW |
354 | |
355 | /* | |
356 | * Fill out preq.nr_sects with proper amount of sectors, and setup | |
1a95fe6e KRW |
357 | * assign map[..] with the PFN of the page in our domain with the |
358 | * corresponding grant reference for each page. | |
359 | */ | |
360 | for (i = 0; i < nseg; i++) { | |
361 | uint32_t flags; | |
362 | ||
363 | flags = GNTMAP_host_map; | |
364 | if (pending_req->operation != BLKIF_OP_READ) | |
365 | flags |= GNTMAP_readonly; | |
366 | gnttab_set_map_op(&map[i], vaddr(pending_req, i), flags, | |
01f37f2d KRW |
367 | req->u.rw.seg[i].gref, |
368 | pending_req->blkif->domid); | |
1a95fe6e KRW |
369 | } |
370 | ||
371 | ret = HYPERVISOR_grant_table_op(GNTTABOP_map_grant_ref, map, nseg); | |
372 | BUG_ON(ret); | |
373 | ||
01f37f2d KRW |
374 | /* |
375 | * Now swizzle the MFN in our domain with the MFN from the other domain | |
1a95fe6e KRW |
376 | * so that when we access vaddr(pending_req,i) it has the contents of |
377 | * the page from the other domain. | |
378 | */ | |
379 | for (i = 0; i < nseg; i++) { | |
380 | if (unlikely(map[i].status != 0)) { | |
22b20f2d | 381 | pr_debug(DRV_PFX "invalid buffer -- could not remap it\n"); |
1a95fe6e KRW |
382 | map[i].handle = BLKBACK_INVALID_HANDLE; |
383 | ret |= 1; | |
384 | } | |
385 | ||
386 | pending_handle(pending_req, i) = map[i].handle; | |
387 | ||
388 | if (ret) | |
389 | continue; | |
390 | ||
391 | ret = m2p_add_override(PFN_DOWN(map[i].dev_bus_addr), | |
392 | blkbk->pending_page(pending_req, i), false); | |
393 | if (ret) { | |
22b20f2d | 394 | pr_alert(DRV_PFX "Failed to install M2P override for %lx (ret: %d)\n", |
ebe81906 | 395 | (unsigned long)map[i].dev_bus_addr, ret); |
1a95fe6e KRW |
396 | /* We could switch over to GNTTABOP_copy */ |
397 | continue; | |
398 | } | |
399 | ||
400 | seg[i].buf = map[i].dev_bus_addr | | |
401 | (req->u.rw.seg[i].first_sect << 9); | |
402 | } | |
403 | return ret; | |
404 | } | |
405 | ||
a1397fa3 KRW |
406 | /* |
407 | * Completion callback on the bio's. Called as bh->b_end_io() | |
4d05a28d KRW |
408 | */ |
409 | ||
2e9977c2 | 410 | static void __end_block_io_op(struct pending_req *pending_req, int error) |
4d05a28d KRW |
411 | { |
412 | /* An error fails the entire request. */ | |
24f567f9 | 413 | if ((pending_req->operation == BLKIF_OP_FLUSH_DISKCACHE) && |
4d05a28d | 414 | (error == -EOPNOTSUPP)) { |
22b20f2d | 415 | pr_debug(DRV_PFX "flush diskcache op failed, not supported\n"); |
24f567f9 | 416 | xen_blkbk_flush_diskcache(XBT_NIL, pending_req->blkif->be, 0); |
4d05a28d KRW |
417 | pending_req->status = BLKIF_RSP_EOPNOTSUPP; |
418 | } else if (error) { | |
22b20f2d | 419 | pr_debug(DRV_PFX "Buffer not up-to-date at end of operation," |
ebe81906 | 420 | " error=%d\n", error); |
4d05a28d KRW |
421 | pending_req->status = BLKIF_RSP_ERROR; |
422 | } | |
423 | ||
01f37f2d KRW |
424 | /* |
425 | * If all of the bio's have completed it is time to unmap | |
a1397fa3 | 426 | * the grant references associated with 'request' and provide |
2e9977c2 KRW |
427 | * the proper response on the ring. |
428 | */ | |
4d05a28d | 429 | if (atomic_dec_and_test(&pending_req->pendcnt)) { |
9f3aedf5 | 430 | xen_blkbk_unmap(pending_req); |
4d05a28d KRW |
431 | make_response(pending_req->blkif, pending_req->id, |
432 | pending_req->operation, pending_req->status); | |
8b6bf747 | 433 | xen_blkif_put(pending_req->blkif); |
4d05a28d KRW |
434 | free_req(pending_req); |
435 | } | |
436 | } | |
437 | ||
a1397fa3 KRW |
438 | /* |
439 | * bio callback. | |
440 | */ | |
88122933 | 441 | static void end_block_io_op(struct bio *bio, int error) |
4d05a28d | 442 | { |
4d05a28d KRW |
443 | __end_block_io_op(bio->bi_private, error); |
444 | bio_put(bio); | |
4d05a28d KRW |
445 | } |
446 | ||
447 | ||
4d05a28d | 448 | |
a1397fa3 KRW |
449 | /* |
450 | * Function to copy the from the ring buffer the 'struct blkif_request' | |
451 | * (which has the sectors we want, number of them, grant references, etc), | |
452 | * and transmute it to the block API to hand it over to the proper block disk. | |
4d05a28d | 453 | */ |
30fd1502 | 454 | static int do_block_io_op(struct xen_blkif *blkif) |
4d05a28d | 455 | { |
88122933 JF |
456 | union blkif_back_rings *blk_rings = &blkif->blk_rings; |
457 | struct blkif_request req; | |
2e9977c2 | 458 | struct pending_req *pending_req; |
4d05a28d KRW |
459 | RING_IDX rc, rp; |
460 | int more_to_do = 0; | |
461 | ||
462 | rc = blk_rings->common.req_cons; | |
463 | rp = blk_rings->common.sring->req_prod; | |
464 | rmb(); /* Ensure we see queued requests up to 'rp'. */ | |
465 | ||
466 | while (rc != rp) { | |
467 | ||
468 | if (RING_REQUEST_CONS_OVERFLOW(&blk_rings->common, rc)) | |
469 | break; | |
470 | ||
8270b45b | 471 | if (kthread_should_stop()) { |
4d05a28d KRW |
472 | more_to_do = 1; |
473 | break; | |
474 | } | |
475 | ||
8270b45b KF |
476 | pending_req = alloc_req(); |
477 | if (NULL == pending_req) { | |
478 | blkif->st_oo_req++; | |
4d05a28d KRW |
479 | more_to_do = 1; |
480 | break; | |
481 | } | |
482 | ||
483 | switch (blkif->blk_protocol) { | |
484 | case BLKIF_PROTOCOL_NATIVE: | |
485 | memcpy(&req, RING_GET_REQUEST(&blk_rings->native, rc), sizeof(req)); | |
486 | break; | |
487 | case BLKIF_PROTOCOL_X86_32: | |
488 | blkif_get_x86_32_req(&req, RING_GET_REQUEST(&blk_rings->x86_32, rc)); | |
489 | break; | |
490 | case BLKIF_PROTOCOL_X86_64: | |
491 | blkif_get_x86_64_req(&req, RING_GET_REQUEST(&blk_rings->x86_64, rc)); | |
492 | break; | |
493 | default: | |
494 | BUG(); | |
495 | } | |
496 | blk_rings->common.req_cons = ++rc; /* before make_response() */ | |
497 | ||
498 | /* Apply all sanity checks to /private copy/ of request. */ | |
499 | barrier(); | |
500 | ||
fc53bf75 | 501 | if (dispatch_rw_block_io(blkif, &req, pending_req)) |
4d05a28d | 502 | break; |
4d05a28d KRW |
503 | |
504 | /* Yield point for this unbounded loop. */ | |
505 | cond_resched(); | |
506 | } | |
507 | ||
508 | return more_to_do; | |
509 | } | |
510 | ||
a1397fa3 | 511 | /* |
01f37f2d KRW |
512 | * Transmutation of the 'struct blkif_request' to a proper 'struct bio' |
513 | * and call the 'submit_bio' to pass it to the underlying storage. | |
a1397fa3 | 514 | */ |
30fd1502 KRW |
515 | static int dispatch_rw_block_io(struct xen_blkif *blkif, |
516 | struct blkif_request *req, | |
517 | struct pending_req *pending_req) | |
4d05a28d | 518 | { |
4d05a28d | 519 | struct phys_req preq; |
1a95fe6e | 520 | struct seg_buf seg[BLKIF_MAX_SEGMENTS_PER_REQUEST]; |
4d05a28d KRW |
521 | unsigned int nseg; |
522 | struct bio *bio = NULL; | |
77089926 | 523 | struct bio *biolist[BLKIF_MAX_SEGMENTS_PER_REQUEST]; |
1a95fe6e | 524 | int i, nbio = 0; |
4d05a28d | 525 | int operation; |
a19be5f0 | 526 | struct blk_plug plug; |
4d05a28d KRW |
527 | |
528 | switch (req->operation) { | |
529 | case BLKIF_OP_READ: | |
fc53bf75 | 530 | blkif->st_rd_req++; |
4d05a28d KRW |
531 | operation = READ; |
532 | break; | |
533 | case BLKIF_OP_WRITE: | |
fc53bf75 | 534 | blkif->st_wr_req++; |
013c3ca1 | 535 | operation = WRITE_ODIRECT; |
4d05a28d | 536 | break; |
24f567f9 | 537 | case BLKIF_OP_FLUSH_DISKCACHE: |
fc53bf75 | 538 | blkif->st_f_req++; |
24f567f9 | 539 | operation = WRITE_FLUSH; |
01f37f2d | 540 | /* |
3d814731 | 541 | * The frontend likes to set this to -1, which xen_vbd_translate |
01f37f2d KRW |
542 | * is alergic too. |
543 | */ | |
24f567f9 | 544 | req->u.rw.sector_number = 0; |
4d05a28d | 545 | break; |
24f567f9 | 546 | case BLKIF_OP_WRITE_BARRIER: |
4d05a28d KRW |
547 | default: |
548 | operation = 0; /* make gcc happy */ | |
fc53bf75 KRW |
549 | goto fail_response; |
550 | break; | |
4d05a28d KRW |
551 | } |
552 | ||
a1397fa3 | 553 | /* Check that the number of segments is sane. */ |
4d05a28d | 554 | nseg = req->nr_segments; |
24f567f9 | 555 | if (unlikely(nseg == 0 && operation != WRITE_FLUSH) || |
4d05a28d | 556 | unlikely(nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST)) { |
22b20f2d | 557 | pr_debug(DRV_PFX "Bad number of segments in request (%d)\n", |
ebe81906 | 558 | nseg); |
1a95fe6e | 559 | /* Haven't submitted any bio's yet. */ |
4d05a28d KRW |
560 | goto fail_response; |
561 | } | |
562 | ||
563 | preq.dev = req->handle; | |
c35950bf | 564 | preq.sector_number = req->u.rw.sector_number; |
4d05a28d KRW |
565 | preq.nr_sects = 0; |
566 | ||
567 | pending_req->blkif = blkif; | |
568 | pending_req->id = req->id; | |
569 | pending_req->operation = req->operation; | |
570 | pending_req->status = BLKIF_RSP_OKAY; | |
571 | pending_req->nr_pages = nseg; | |
e9350493 | 572 | |
4d05a28d | 573 | for (i = 0; i < nseg; i++) { |
c35950bf KRW |
574 | seg[i].nsec = req->u.rw.seg[i].last_sect - |
575 | req->u.rw.seg[i].first_sect + 1; | |
c35950bf KRW |
576 | if ((req->u.rw.seg[i].last_sect >= (PAGE_SIZE >> 9)) || |
577 | (req->u.rw.seg[i].last_sect < req->u.rw.seg[i].first_sect)) | |
4d05a28d KRW |
578 | goto fail_response; |
579 | preq.nr_sects += seg[i].nsec; | |
976222e0 | 580 | |
4d05a28d KRW |
581 | } |
582 | ||
3d814731 | 583 | if (xen_vbd_translate(&preq, blkif, operation) != 0) { |
22b20f2d | 584 | pr_debug(DRV_PFX "access denied: %s of [%llu,%llu] on dev=%04x\n", |
ebe81906 KRW |
585 | operation == READ ? "read" : "write", |
586 | preq.sector_number, | |
587 | preq.sector_number + preq.nr_sects, preq.dev); | |
1a95fe6e | 588 | goto fail_response; |
4d05a28d | 589 | } |
01f37f2d KRW |
590 | |
591 | /* | |
3d814731 | 592 | * This check _MUST_ be done after xen_vbd_translate as the preq.bdev |
01f37f2d KRW |
593 | * is set there. |
594 | */ | |
e9350493 KRW |
595 | for (i = 0; i < nseg; i++) { |
596 | if (((int)preq.sector_number|(int)seg[i].nsec) & | |
597 | ((bdev_logical_block_size(preq.bdev) >> 9) - 1)) { | |
22b20f2d | 598 | pr_debug(DRV_PFX "Misaligned I/O request from domain %d", |
ebe81906 | 599 | blkif->domid); |
e9350493 KRW |
600 | goto fail_response; |
601 | } | |
602 | } | |
01f37f2d KRW |
603 | |
604 | /* | |
605 | * If we have failed at this point, we need to undo the M2P override, | |
2e9977c2 KRW |
606 | * set gnttab_set_unmap_op on all of the grant references and perform |
607 | * the hypercall to unmap the grants - that is all done in | |
9f3aedf5 | 608 | * xen_blkbk_unmap. |
2e9977c2 | 609 | */ |
9f3aedf5 | 610 | if (xen_blkbk_map(req, pending_req, seg)) |
4d05a28d KRW |
611 | goto fail_flush; |
612 | ||
3d68b399 | 613 | /* This corresponding xen_blkif_put is done in __end_block_io_op */ |
8b6bf747 | 614 | xen_blkif_get(blkif); |
4d05a28d KRW |
615 | |
616 | for (i = 0; i < nseg; i++) { | |
4d05a28d KRW |
617 | while ((bio == NULL) || |
618 | (bio_add_page(bio, | |
e8e28871 | 619 | blkbk->pending_page(pending_req, i), |
4d05a28d KRW |
620 | seg[i].nsec << 9, |
621 | seg[i].buf & ~PAGE_MASK) == 0)) { | |
2e9977c2 | 622 | |
03e0edf9 | 623 | bio = bio_alloc(GFP_KERNEL, nseg-i); |
4d05a28d KRW |
624 | if (unlikely(bio == NULL)) |
625 | goto fail_put_bio; | |
626 | ||
03e0edf9 | 627 | biolist[nbio++] = bio; |
4d05a28d KRW |
628 | bio->bi_bdev = preq.bdev; |
629 | bio->bi_private = pending_req; | |
630 | bio->bi_end_io = end_block_io_op; | |
631 | bio->bi_sector = preq.sector_number; | |
632 | } | |
633 | ||
634 | preq.sector_number += seg[i].nsec; | |
635 | } | |
636 | ||
3d68b399 | 637 | /* This will be hit if the operation was a flush. */ |
4d05a28d | 638 | if (!bio) { |
24f567f9 | 639 | BUG_ON(operation != WRITE_FLUSH); |
b0f80127 | 640 | |
03e0edf9 | 641 | bio = bio_alloc(GFP_KERNEL, 0); |
4d05a28d KRW |
642 | if (unlikely(bio == NULL)) |
643 | goto fail_put_bio; | |
644 | ||
03e0edf9 | 645 | biolist[nbio++] = bio; |
4d05a28d KRW |
646 | bio->bi_bdev = preq.bdev; |
647 | bio->bi_private = pending_req; | |
648 | bio->bi_end_io = end_block_io_op; | |
4d05a28d KRW |
649 | } |
650 | ||
01f37f2d KRW |
651 | /* |
652 | * We set it one so that the last submit_bio does not have to call | |
77089926 KRW |
653 | * atomic_inc. |
654 | */ | |
655 | atomic_set(&pending_req->pendcnt, nbio); | |
656 | ||
a19be5f0 KRW |
657 | /* Get a reference count for the disk queue and start sending I/O */ |
658 | blk_start_plug(&plug); | |
659 | ||
77089926 KRW |
660 | for (i = 0; i < nbio; i++) |
661 | submit_bio(operation, biolist[i]); | |
662 | ||
a19be5f0 | 663 | /* Let the I/Os go.. */ |
3d68b399 | 664 | blk_finish_plug(&plug); |
a19be5f0 | 665 | |
4d05a28d KRW |
666 | if (operation == READ) |
667 | blkif->st_rd_sect += preq.nr_sects; | |
24f567f9 | 668 | else if (operation == WRITE || operation == WRITE_FLUSH) |
4d05a28d KRW |
669 | blkif->st_wr_sect += preq.nr_sects; |
670 | ||
fc53bf75 | 671 | return 0; |
4d05a28d KRW |
672 | |
673 | fail_flush: | |
9f3aedf5 | 674 | xen_blkbk_unmap(pending_req); |
4d05a28d | 675 | fail_response: |
0faa8cca | 676 | /* Haven't submitted any bio's yet. */ |
4d05a28d KRW |
677 | make_response(blkif, req->id, req->operation, BLKIF_RSP_ERROR); |
678 | free_req(pending_req); | |
679 | msleep(1); /* back off a bit */ | |
fc53bf75 | 680 | return -EIO; |
4d05a28d KRW |
681 | |
682 | fail_put_bio: | |
03e0edf9 | 683 | for (i = 0; i < nbio; i++) |
77089926 | 684 | bio_put(biolist[i]); |
4d05a28d | 685 | __end_block_io_op(pending_req, -EINVAL); |
4d05a28d | 686 | msleep(1); /* back off a bit */ |
fc53bf75 | 687 | return -EIO; |
4d05a28d KRW |
688 | } |
689 | ||
690 | ||
691 | ||
a1397fa3 KRW |
692 | /* |
693 | * Put a response on the ring on how the operation fared. | |
4d05a28d | 694 | */ |
30fd1502 | 695 | static void make_response(struct xen_blkif *blkif, u64 id, |
4d05a28d KRW |
696 | unsigned short op, int st) |
697 | { | |
88122933 | 698 | struct blkif_response resp; |
4d05a28d | 699 | unsigned long flags; |
88122933 | 700 | union blkif_back_rings *blk_rings = &blkif->blk_rings; |
4d05a28d KRW |
701 | int more_to_do = 0; |
702 | int notify; | |
703 | ||
704 | resp.id = id; | |
705 | resp.operation = op; | |
706 | resp.status = st; | |
707 | ||
708 | spin_lock_irqsave(&blkif->blk_ring_lock, flags); | |
709 | /* Place on the response ring for the relevant domain. */ | |
710 | switch (blkif->blk_protocol) { | |
711 | case BLKIF_PROTOCOL_NATIVE: | |
712 | memcpy(RING_GET_RESPONSE(&blk_rings->native, blk_rings->native.rsp_prod_pvt), | |
713 | &resp, sizeof(resp)); | |
714 | break; | |
715 | case BLKIF_PROTOCOL_X86_32: | |
716 | memcpy(RING_GET_RESPONSE(&blk_rings->x86_32, blk_rings->x86_32.rsp_prod_pvt), | |
717 | &resp, sizeof(resp)); | |
718 | break; | |
719 | case BLKIF_PROTOCOL_X86_64: | |
720 | memcpy(RING_GET_RESPONSE(&blk_rings->x86_64, blk_rings->x86_64.rsp_prod_pvt), | |
721 | &resp, sizeof(resp)); | |
722 | break; | |
723 | default: | |
724 | BUG(); | |
725 | } | |
726 | blk_rings->common.rsp_prod_pvt++; | |
727 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blk_rings->common, notify); | |
728 | if (blk_rings->common.rsp_prod_pvt == blk_rings->common.req_cons) { | |
729 | /* | |
730 | * Tail check for pending requests. Allows frontend to avoid | |
731 | * notifications if requests are already in flight (lower | |
732 | * overheads and promotes batching). | |
733 | */ | |
734 | RING_FINAL_CHECK_FOR_REQUESTS(&blk_rings->common, more_to_do); | |
735 | ||
736 | } else if (RING_HAS_UNCONSUMED_REQUESTS(&blk_rings->common)) { | |
737 | more_to_do = 1; | |
738 | } | |
739 | ||
740 | spin_unlock_irqrestore(&blkif->blk_ring_lock, flags); | |
741 | ||
742 | if (more_to_do) | |
743 | blkif_notify_work(blkif); | |
744 | if (notify) | |
745 | notify_remote_via_irq(blkif->irq); | |
746 | } | |
747 | ||
8b6bf747 | 748 | static int __init xen_blkif_init(void) |
4d05a28d KRW |
749 | { |
750 | int i, mmap_pages; | |
8770b268 | 751 | int rc = 0; |
4d05a28d | 752 | |
88122933 | 753 | if (!xen_pv_domain()) |
4d05a28d KRW |
754 | return -ENODEV; |
755 | ||
2e9977c2 | 756 | blkbk = kzalloc(sizeof(struct xen_blkbk), GFP_KERNEL); |
e8e28871 | 757 | if (!blkbk) { |
22b20f2d | 758 | pr_alert(DRV_PFX "%s: out of memory!\n", __func__); |
e8e28871 KRW |
759 | return -ENOMEM; |
760 | } | |
761 | ||
8b6bf747 | 762 | mmap_pages = xen_blkif_reqs * BLKIF_MAX_SEGMENTS_PER_REQUEST; |
4d05a28d | 763 | |
e8e28871 | 764 | blkbk->pending_reqs = kmalloc(sizeof(blkbk->pending_reqs[0]) * |
8b6bf747 | 765 | xen_blkif_reqs, GFP_KERNEL); |
a742b02c KRW |
766 | blkbk->pending_grant_handles = kzalloc(sizeof(blkbk->pending_grant_handles[0]) * |
767 | mmap_pages, GFP_KERNEL); | |
768 | blkbk->pending_pages = kzalloc(sizeof(blkbk->pending_pages[0]) * | |
769 | mmap_pages, GFP_KERNEL); | |
4d05a28d | 770 | |
2e9977c2 KRW |
771 | if (!blkbk->pending_reqs || !blkbk->pending_grant_handles || |
772 | !blkbk->pending_pages) { | |
8770b268 | 773 | rc = -ENOMEM; |
4d05a28d | 774 | goto out_of_memory; |
8770b268 | 775 | } |
4d05a28d | 776 | |
464fb419 | 777 | for (i = 0; i < mmap_pages; i++) { |
e8e28871 | 778 | blkbk->pending_grant_handles[i] = BLKBACK_INVALID_HANDLE; |
a742b02c | 779 | blkbk->pending_pages[i] = alloc_page(GFP_KERNEL); |
464fb419 KRW |
780 | if (blkbk->pending_pages[i] == NULL) { |
781 | rc = -ENOMEM; | |
782 | goto out_of_memory; | |
783 | } | |
784 | } | |
8b6bf747 | 785 | rc = xen_blkif_interface_init(); |
8770b268 KRW |
786 | if (rc) |
787 | goto failed_init; | |
4d05a28d | 788 | |
e8e28871 KRW |
789 | memset(blkbk->pending_reqs, 0, sizeof(blkbk->pending_reqs)); |
790 | ||
791 | INIT_LIST_HEAD(&blkbk->pending_free); | |
792 | spin_lock_init(&blkbk->pending_free_lock); | |
793 | init_waitqueue_head(&blkbk->pending_free_wq); | |
4d05a28d | 794 | |
8b6bf747 | 795 | for (i = 0; i < xen_blkif_reqs; i++) |
2e9977c2 KRW |
796 | list_add_tail(&blkbk->pending_reqs[i].free_list, |
797 | &blkbk->pending_free); | |
4d05a28d | 798 | |
8b6bf747 | 799 | rc = xen_blkif_xenbus_init(); |
8770b268 KRW |
800 | if (rc) |
801 | goto failed_init; | |
4d05a28d KRW |
802 | |
803 | return 0; | |
804 | ||
805 | out_of_memory: | |
22b20f2d | 806 | pr_alert(DRV_PFX "%s: out of memory\n", __func__); |
8770b268 | 807 | failed_init: |
e8e28871 | 808 | kfree(blkbk->pending_reqs); |
a742b02c | 809 | kfree(blkbk->pending_grant_handles); |
464fb419 KRW |
810 | for (i = 0; i < mmap_pages; i++) { |
811 | if (blkbk->pending_pages[i]) | |
812 | __free_page(blkbk->pending_pages[i]); | |
813 | } | |
a742b02c KRW |
814 | kfree(blkbk->pending_pages); |
815 | kfree(blkbk); | |
e8e28871 | 816 | blkbk = NULL; |
8770b268 | 817 | return rc; |
4d05a28d KRW |
818 | } |
819 | ||
8b6bf747 | 820 | module_init(xen_blkif_init); |
4d05a28d KRW |
821 | |
822 | MODULE_LICENSE("Dual BSD/GPL"); |