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 | ||
314146e5 TG |
49 | #define WRITE_BARRIER (REQ_WRITE | REQ_FLUSH | REQ_FUA) |
50 | ||
4d05a28d KRW |
51 | /* |
52 | * These are rather arbitrary. They are fairly large because adjacent requests | |
53 | * pulled from a communication ring are quite likely to end up being part of | |
54 | * the same scatter/gather request at the disc. | |
55 | * | |
56 | * ** TRY INCREASING 'blkif_reqs' IF WRITE SPEEDS SEEM TOO LOW ** | |
57 | * | |
58 | * This will increase the chances of being able to write whole tracks. | |
59 | * 64 should be enough to keep us competitive with Linux. | |
60 | */ | |
61 | static int blkif_reqs = 64; | |
62 | module_param_named(reqs, blkif_reqs, int, 0); | |
63 | MODULE_PARM_DESC(reqs, "Number of blkback requests to allocate"); | |
64 | ||
65 | /* Run-time switchable: /sys/module/blkback/parameters/ */ | |
2e9977c2 KRW |
66 | static unsigned int log_stats; |
67 | static unsigned int debug_lvl; | |
4d05a28d KRW |
68 | module_param(log_stats, int, 0644); |
69 | module_param(debug_lvl, int, 0644); | |
70 | ||
71 | /* | |
72 | * Each outstanding request that we've passed to the lower device layers has a | |
73 | * 'pending_req' allocated to it. Each buffer_head that completes decrements | |
74 | * the pendcnt towards zero. When it hits zero, the specified domain has a | |
75 | * response queued for it, with the saved 'id' passed back. | |
76 | */ | |
2e9977c2 | 77 | struct pending_req { |
5489377c | 78 | struct blkif_st *blkif; |
4d05a28d KRW |
79 | u64 id; |
80 | int nr_pages; | |
81 | atomic_t pendcnt; | |
82 | unsigned short operation; | |
83 | int status; | |
84 | struct list_head free_list; | |
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 | |
a1397fa3 KRW |
104 | /* |
105 | * Little helpful macro to figure out the index and virtual address of the | |
106 | * pending_pages[..]. For each 'pending_req' we have have up to | |
107 | * BLKIF_MAX_SEGMENTS_PER_REQUEST (11) pages. The seg would be from 0 through | |
108 | * 10 and would index in the pending_pages[..]. */ | |
2e9977c2 | 109 | static inline int vaddr_pagenr(struct pending_req *req, int seg) |
4d05a28d | 110 | { |
2e9977c2 KRW |
111 | return (req - blkbk->pending_reqs) * |
112 | BLKIF_MAX_SEGMENTS_PER_REQUEST + seg; | |
4d05a28d KRW |
113 | } |
114 | ||
efe08a3e JB |
115 | #define pending_page(req, seg) pending_pages[vaddr_pagenr(req, seg)] |
116 | ||
2e9977c2 | 117 | static inline unsigned long vaddr(struct pending_req *req, int seg) |
4d05a28d | 118 | { |
e8e28871 | 119 | unsigned long pfn = page_to_pfn(blkbk->pending_page(req, seg)); |
4d05a28d KRW |
120 | return (unsigned long)pfn_to_kaddr(pfn); |
121 | } | |
122 | ||
123 | #define pending_handle(_req, _seg) \ | |
e8e28871 | 124 | (blkbk->pending_grant_handles[vaddr_pagenr(_req, _seg)]) |
4d05a28d KRW |
125 | |
126 | ||
5489377c KRW |
127 | static int do_block_io_op(struct blkif_st *blkif); |
128 | static void dispatch_rw_block_io(struct blkif_st *blkif, | |
88122933 | 129 | struct blkif_request *req, |
2e9977c2 | 130 | struct pending_req *pending_req); |
5489377c | 131 | static void make_response(struct blkif_st *blkif, u64 id, |
4d05a28d KRW |
132 | unsigned short op, int st); |
133 | ||
a1397fa3 KRW |
134 | /* |
135 | * Retrieve from the 'pending_reqs' a free pending_req structure to be used. | |
4d05a28d | 136 | */ |
2e9977c2 | 137 | static struct pending_req *alloc_req(void) |
4d05a28d | 138 | { |
2e9977c2 | 139 | struct pending_req *req = NULL; |
4d05a28d KRW |
140 | unsigned long flags; |
141 | ||
e8e28871 KRW |
142 | spin_lock_irqsave(&blkbk->pending_free_lock, flags); |
143 | if (!list_empty(&blkbk->pending_free)) { | |
2e9977c2 KRW |
144 | req = list_entry(blkbk->pending_free.next, struct pending_req, |
145 | free_list); | |
4d05a28d KRW |
146 | list_del(&req->free_list); |
147 | } | |
e8e28871 | 148 | spin_unlock_irqrestore(&blkbk->pending_free_lock, flags); |
4d05a28d KRW |
149 | return req; |
150 | } | |
151 | ||
a1397fa3 KRW |
152 | /* |
153 | * Return the 'pending_req' structure back to the freepool. We also | |
154 | * wake up the thread if it was waiting for a free page. | |
155 | */ | |
2e9977c2 | 156 | static void free_req(struct pending_req *req) |
4d05a28d KRW |
157 | { |
158 | unsigned long flags; | |
159 | int was_empty; | |
160 | ||
e8e28871 KRW |
161 | spin_lock_irqsave(&blkbk->pending_free_lock, flags); |
162 | was_empty = list_empty(&blkbk->pending_free); | |
163 | list_add(&req->free_list, &blkbk->pending_free); | |
164 | spin_unlock_irqrestore(&blkbk->pending_free_lock, flags); | |
4d05a28d | 165 | if (was_empty) |
e8e28871 | 166 | wake_up(&blkbk->pending_free_wq); |
4d05a28d KRW |
167 | } |
168 | ||
a1397fa3 KRW |
169 | /* |
170 | * Unmap the grant references, and also remove the M2P over-rides | |
171 | * used in the 'pending_req'. | |
172 | */ | |
2e9977c2 | 173 | static void fast_flush_area(struct pending_req *req) |
4d05a28d KRW |
174 | { |
175 | struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
176 | unsigned int i, invcount = 0; | |
177 | grant_handle_t handle; | |
178 | int ret; | |
179 | ||
180 | for (i = 0; i < req->nr_pages; i++) { | |
181 | handle = pending_handle(req, i); | |
182 | if (handle == BLKBACK_INVALID_HANDLE) | |
183 | continue; | |
184 | gnttab_set_unmap_op(&unmap[invcount], vaddr(req, i), | |
185 | GNTMAP_host_map, handle); | |
186 | pending_handle(req, i) = BLKBACK_INVALID_HANDLE; | |
187 | invcount++; | |
188 | } | |
189 | ||
190 | ret = HYPERVISOR_grant_table_op( | |
191 | GNTTABOP_unmap_grant_ref, unmap, invcount); | |
192 | BUG_ON(ret); | |
5dc03639 | 193 | /* Note, we use invcount, so nr->pages, so we can't index |
2e9977c2 KRW |
194 | * using vaddr(req, i). |
195 | */ | |
5dc03639 KRW |
196 | for (i = 0; i < invcount; i++) { |
197 | ret = m2p_remove_override( | |
198 | virt_to_page(unmap[i].host_addr), false); | |
199 | if (ret) { | |
200 | printk(KERN_ALERT "Failed to remove M2P override for " \ | |
201 | "%lx\n", (unsigned long)unmap[i].host_addr); | |
202 | continue; | |
203 | } | |
204 | } | |
4d05a28d KRW |
205 | } |
206 | ||
2e9977c2 | 207 | /* |
4d05a28d KRW |
208 | * SCHEDULER FUNCTIONS |
209 | */ | |
210 | ||
5489377c | 211 | static void print_stats(struct blkif_st *blkif) |
4d05a28d KRW |
212 | { |
213 | printk(KERN_DEBUG "%s: oo %3d | rd %4d | wr %4d | br %4d\n", | |
214 | current->comm, blkif->st_oo_req, | |
215 | blkif->st_rd_req, blkif->st_wr_req, blkif->st_br_req); | |
216 | blkif->st_print = jiffies + msecs_to_jiffies(10 * 1000); | |
217 | blkif->st_rd_req = 0; | |
218 | blkif->st_wr_req = 0; | |
219 | blkif->st_oo_req = 0; | |
220 | } | |
221 | ||
222 | int blkif_schedule(void *arg) | |
223 | { | |
5489377c | 224 | struct blkif_st *blkif = arg; |
2ccbfe26 | 225 | struct vbd *vbd = &blkif->vbd; |
4d05a28d KRW |
226 | |
227 | blkif_get(blkif); | |
228 | ||
229 | if (debug_lvl) | |
230 | printk(KERN_DEBUG "%s: started\n", current->comm); | |
231 | ||
232 | while (!kthread_should_stop()) { | |
233 | if (try_to_freeze()) | |
234 | continue; | |
2ccbfe26 S |
235 | if (unlikely(vbd->size != vbd_size(vbd))) |
236 | vbd_resize(blkif); | |
4d05a28d KRW |
237 | |
238 | wait_event_interruptible( | |
239 | blkif->wq, | |
240 | blkif->waiting_reqs || kthread_should_stop()); | |
241 | wait_event_interruptible( | |
e8e28871 | 242 | blkbk->pending_free_wq, |
2e9977c2 KRW |
243 | !list_empty(&blkbk->pending_free) || |
244 | kthread_should_stop()); | |
4d05a28d KRW |
245 | |
246 | blkif->waiting_reqs = 0; | |
247 | smp_mb(); /* clear flag *before* checking for work */ | |
248 | ||
249 | if (do_block_io_op(blkif)) | |
250 | blkif->waiting_reqs = 1; | |
4d05a28d KRW |
251 | |
252 | if (log_stats && time_after(jiffies, blkif->st_print)) | |
253 | print_stats(blkif); | |
254 | } | |
255 | ||
256 | if (log_stats) | |
257 | print_stats(blkif); | |
258 | if (debug_lvl) | |
259 | printk(KERN_DEBUG "%s: exiting\n", current->comm); | |
260 | ||
261 | blkif->xenblkd = NULL; | |
262 | blkif_put(blkif); | |
263 | ||
264 | return 0; | |
265 | } | |
266 | ||
a1397fa3 KRW |
267 | /* |
268 | * Completion callback on the bio's. Called as bh->b_end_io() | |
4d05a28d KRW |
269 | */ |
270 | ||
2e9977c2 | 271 | static void __end_block_io_op(struct pending_req *pending_req, int error) |
4d05a28d KRW |
272 | { |
273 | /* An error fails the entire request. */ | |
274 | if ((pending_req->operation == BLKIF_OP_WRITE_BARRIER) && | |
275 | (error == -EOPNOTSUPP)) { | |
276 | DPRINTK("blkback: write barrier op failed, not supported\n"); | |
277 | blkback_barrier(XBT_NIL, pending_req->blkif->be, 0); | |
278 | pending_req->status = BLKIF_RSP_EOPNOTSUPP; | |
279 | } else if (error) { | |
280 | DPRINTK("Buffer not up-to-date at end of operation, " | |
281 | "error=%d\n", error); | |
282 | pending_req->status = BLKIF_RSP_ERROR; | |
283 | } | |
284 | ||
a1397fa3 KRW |
285 | /* If all of the bio's have completed it is time to unmap |
286 | * the grant references associated with 'request' and provide | |
2e9977c2 KRW |
287 | * the proper response on the ring. |
288 | */ | |
4d05a28d KRW |
289 | if (atomic_dec_and_test(&pending_req->pendcnt)) { |
290 | fast_flush_area(pending_req); | |
291 | make_response(pending_req->blkif, pending_req->id, | |
292 | pending_req->operation, pending_req->status); | |
293 | blkif_put(pending_req->blkif); | |
294 | free_req(pending_req); | |
295 | } | |
296 | } | |
297 | ||
a1397fa3 KRW |
298 | /* |
299 | * bio callback. | |
300 | */ | |
88122933 | 301 | static void end_block_io_op(struct bio *bio, int error) |
4d05a28d | 302 | { |
4d05a28d KRW |
303 | __end_block_io_op(bio->bi_private, error); |
304 | bio_put(bio); | |
4d05a28d KRW |
305 | } |
306 | ||
307 | ||
a1397fa3 KRW |
308 | /* |
309 | * Notification from the guest OS. | |
4d05a28d KRW |
310 | */ |
311 | ||
5489377c | 312 | static void blkif_notify_work(struct blkif_st *blkif) |
4d05a28d KRW |
313 | { |
314 | blkif->waiting_reqs = 1; | |
315 | wake_up(&blkif->wq); | |
316 | } | |
317 | ||
88122933 | 318 | irqreturn_t blkif_be_int(int irq, void *dev_id) |
4d05a28d KRW |
319 | { |
320 | blkif_notify_work(dev_id); | |
321 | return IRQ_HANDLED; | |
322 | } | |
323 | ||
324 | ||
325 | ||
a1397fa3 KRW |
326 | /* |
327 | * Function to copy the from the ring buffer the 'struct blkif_request' | |
328 | * (which has the sectors we want, number of them, grant references, etc), | |
329 | * and transmute it to the block API to hand it over to the proper block disk. | |
4d05a28d | 330 | */ |
5489377c | 331 | static int do_block_io_op(struct blkif_st *blkif) |
4d05a28d | 332 | { |
88122933 JF |
333 | union blkif_back_rings *blk_rings = &blkif->blk_rings; |
334 | struct blkif_request req; | |
2e9977c2 | 335 | struct pending_req *pending_req; |
4d05a28d KRW |
336 | RING_IDX rc, rp; |
337 | int more_to_do = 0; | |
338 | ||
339 | rc = blk_rings->common.req_cons; | |
340 | rp = blk_rings->common.sring->req_prod; | |
341 | rmb(); /* Ensure we see queued requests up to 'rp'. */ | |
342 | ||
343 | while (rc != rp) { | |
344 | ||
345 | if (RING_REQUEST_CONS_OVERFLOW(&blk_rings->common, rc)) | |
346 | break; | |
347 | ||
8270b45b | 348 | if (kthread_should_stop()) { |
4d05a28d KRW |
349 | more_to_do = 1; |
350 | break; | |
351 | } | |
352 | ||
8270b45b KF |
353 | pending_req = alloc_req(); |
354 | if (NULL == pending_req) { | |
355 | blkif->st_oo_req++; | |
4d05a28d KRW |
356 | more_to_do = 1; |
357 | break; | |
358 | } | |
359 | ||
360 | switch (blkif->blk_protocol) { | |
361 | case BLKIF_PROTOCOL_NATIVE: | |
362 | memcpy(&req, RING_GET_REQUEST(&blk_rings->native, rc), sizeof(req)); | |
363 | break; | |
364 | case BLKIF_PROTOCOL_X86_32: | |
365 | blkif_get_x86_32_req(&req, RING_GET_REQUEST(&blk_rings->x86_32, rc)); | |
366 | break; | |
367 | case BLKIF_PROTOCOL_X86_64: | |
368 | blkif_get_x86_64_req(&req, RING_GET_REQUEST(&blk_rings->x86_64, rc)); | |
369 | break; | |
370 | default: | |
371 | BUG(); | |
372 | } | |
373 | blk_rings->common.req_cons = ++rc; /* before make_response() */ | |
374 | ||
375 | /* Apply all sanity checks to /private copy/ of request. */ | |
376 | barrier(); | |
377 | ||
378 | switch (req.operation) { | |
379 | case BLKIF_OP_READ: | |
380 | blkif->st_rd_req++; | |
381 | dispatch_rw_block_io(blkif, &req, pending_req); | |
382 | break; | |
383 | case BLKIF_OP_WRITE_BARRIER: | |
384 | blkif->st_br_req++; | |
385 | /* fall through */ | |
386 | case BLKIF_OP_WRITE: | |
387 | blkif->st_wr_req++; | |
388 | dispatch_rw_block_io(blkif, &req, pending_req); | |
389 | break; | |
390 | default: | |
391 | /* A good sign something is wrong: sleep for a while to | |
392 | * avoid excessive CPU consumption by a bad guest. */ | |
393 | msleep(1); | |
394 | DPRINTK("error: unknown block io operation [%d]\n", | |
395 | req.operation); | |
396 | make_response(blkif, req.id, req.operation, | |
397 | BLKIF_RSP_ERROR); | |
398 | free_req(pending_req); | |
399 | break; | |
400 | } | |
401 | ||
402 | /* Yield point for this unbounded loop. */ | |
403 | cond_resched(); | |
404 | } | |
405 | ||
406 | return more_to_do; | |
407 | } | |
408 | ||
a1397fa3 KRW |
409 | /* |
410 | * Transumation of the 'struct blkif_request' to a proper 'struct bio' | |
411 | * and call the 'submit_bio' to pass it to the underlaying storage. | |
412 | */ | |
5489377c | 413 | static void dispatch_rw_block_io(struct blkif_st *blkif, |
88122933 | 414 | struct blkif_request *req, |
2e9977c2 | 415 | struct pending_req *pending_req) |
4d05a28d | 416 | { |
4d05a28d KRW |
417 | struct gnttab_map_grant_ref map[BLKIF_MAX_SEGMENTS_PER_REQUEST]; |
418 | struct phys_req preq; | |
419 | struct { | |
420 | unsigned long buf; unsigned int nsec; | |
421 | } seg[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
422 | unsigned int nseg; | |
423 | struct bio *bio = NULL; | |
77089926 KRW |
424 | struct bio *biolist[BLKIF_MAX_SEGMENTS_PER_REQUEST]; |
425 | int ret, i, nbio = 0; | |
4d05a28d | 426 | int operation; |
0faa8cca KRW |
427 | struct blk_plug plug; |
428 | struct request_queue *q; | |
4d05a28d KRW |
429 | |
430 | switch (req->operation) { | |
431 | case BLKIF_OP_READ: | |
432 | operation = READ; | |
433 | break; | |
434 | case BLKIF_OP_WRITE: | |
435 | operation = WRITE; | |
436 | break; | |
437 | case BLKIF_OP_WRITE_BARRIER: | |
314146e5 | 438 | operation = WRITE_BARRIER; |
4d05a28d KRW |
439 | break; |
440 | default: | |
441 | operation = 0; /* make gcc happy */ | |
442 | BUG(); | |
443 | } | |
444 | ||
a1397fa3 | 445 | /* Check that the number of segments is sane. */ |
4d05a28d | 446 | nseg = req->nr_segments; |
314146e5 | 447 | if (unlikely(nseg == 0 && operation != WRITE_BARRIER) || |
4d05a28d KRW |
448 | unlikely(nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST)) { |
449 | DPRINTK("Bad number of segments in request (%d)\n", nseg); | |
450 | goto fail_response; | |
451 | } | |
452 | ||
453 | preq.dev = req->handle; | |
c35950bf | 454 | preq.sector_number = req->u.rw.sector_number; |
4d05a28d KRW |
455 | preq.nr_sects = 0; |
456 | ||
457 | pending_req->blkif = blkif; | |
458 | pending_req->id = req->id; | |
459 | pending_req->operation = req->operation; | |
460 | pending_req->status = BLKIF_RSP_OKAY; | |
461 | pending_req->nr_pages = nseg; | |
462 | ||
a1397fa3 KRW |
463 | /* Fill out preq.nr_sects with proper amount of sectors, and setup |
464 | * assign map[..] with the PFN of the page in our domain with the | |
2e9977c2 KRW |
465 | * corresponding grant reference for each page. |
466 | */ | |
4d05a28d KRW |
467 | for (i = 0; i < nseg; i++) { |
468 | uint32_t flags; | |
469 | ||
c35950bf KRW |
470 | seg[i].nsec = req->u.rw.seg[i].last_sect - |
471 | req->u.rw.seg[i].first_sect + 1; | |
c35950bf KRW |
472 | if ((req->u.rw.seg[i].last_sect >= (PAGE_SIZE >> 9)) || |
473 | (req->u.rw.seg[i].last_sect < req->u.rw.seg[i].first_sect)) | |
4d05a28d KRW |
474 | goto fail_response; |
475 | preq.nr_sects += seg[i].nsec; | |
476 | ||
477 | flags = GNTMAP_host_map; | |
478 | if (operation != READ) | |
479 | flags |= GNTMAP_readonly; | |
480 | gnttab_set_map_op(&map[i], vaddr(pending_req, i), flags, | |
c35950bf | 481 | req->u.rw.seg[i].gref, blkif->domid); |
4d05a28d KRW |
482 | } |
483 | ||
484 | ret = HYPERVISOR_grant_table_op(GNTTABOP_map_grant_ref, map, nseg); | |
485 | BUG_ON(ret); | |
486 | ||
a1397fa3 | 487 | /* Now swizzel the MFN in our domain with the MFN from the other domain |
2e9977c2 KRW |
488 | * so that when we access vaddr(pending_req,i) it has the contents of |
489 | * the page from the other domain. | |
490 | */ | |
4d05a28d KRW |
491 | for (i = 0; i < nseg; i++) { |
492 | if (unlikely(map[i].status != 0)) { | |
493 | DPRINTK("invalid buffer -- could not remap it\n"); | |
494 | map[i].handle = BLKBACK_INVALID_HANDLE; | |
495 | ret |= 1; | |
496 | } | |
497 | ||
498 | pending_handle(pending_req, i) = map[i].handle; | |
499 | ||
500 | if (ret) | |
501 | continue; | |
2e9977c2 | 502 | |
5dc03639 KRW |
503 | ret = m2p_add_override(PFN_DOWN(map[i].dev_bus_addr), |
504 | blkbk->pending_page(pending_req, i), false); | |
505 | if (ret) { | |
506 | printk(KERN_ALERT "Failed to install M2P override for"\ | |
2e9977c2 KRW |
507 | " %lx (ret: %d)\n", (unsigned long) |
508 | map[i].dev_bus_addr, ret); | |
a1397fa3 | 509 | /* We could switch over to GNTTABOP_copy */ |
5dc03639 KRW |
510 | continue; |
511 | } | |
4d05a28d | 512 | |
4d05a28d | 513 | seg[i].buf = map[i].dev_bus_addr | |
c35950bf | 514 | (req->u.rw.seg[i].first_sect << 9); |
4d05a28d KRW |
515 | } |
516 | ||
2e9977c2 KRW |
517 | /* If we have failed at this point, we need to undo the M2P override, |
518 | * set gnttab_set_unmap_op on all of the grant references and perform | |
519 | * the hypercall to unmap the grants - that is all done in | |
520 | * fast_flush_area. | |
521 | */ | |
4d05a28d KRW |
522 | if (ret) |
523 | goto fail_flush; | |
524 | ||
525 | if (vbd_translate(&preq, blkif, operation) != 0) { | |
526 | DPRINTK("access denied: %s of [%llu,%llu] on dev=%04x\n", | |
527 | operation == READ ? "read" : "write", | |
528 | preq.sector_number, | |
529 | preq.sector_number + preq.nr_sects, preq.dev); | |
530 | goto fail_flush; | |
531 | } | |
532 | ||
77089926 | 533 | /* This corresponding blkif_put is done in __end_block_io_op */ |
4d05a28d KRW |
534 | blkif_get(blkif); |
535 | ||
536 | for (i = 0; i < nseg; i++) { | |
537 | if (((int)preq.sector_number|(int)seg[i].nsec) & | |
05d43865 | 538 | ((bdev_logical_block_size(preq.bdev) >> 9) - 1)) { |
4d05a28d KRW |
539 | DPRINTK("Misaligned I/O request from domain %d", |
540 | blkif->domid); | |
541 | goto fail_put_bio; | |
542 | } | |
543 | ||
544 | while ((bio == NULL) || | |
545 | (bio_add_page(bio, | |
e8e28871 | 546 | blkbk->pending_page(pending_req, i), |
4d05a28d KRW |
547 | seg[i].nsec << 9, |
548 | seg[i].buf & ~PAGE_MASK) == 0)) { | |
2e9977c2 | 549 | |
77089926 | 550 | bio = biolist[nbio++] = bio_alloc(GFP_KERNEL, nseg-i); |
4d05a28d KRW |
551 | if (unlikely(bio == NULL)) |
552 | goto fail_put_bio; | |
553 | ||
554 | bio->bi_bdev = preq.bdev; | |
555 | bio->bi_private = pending_req; | |
556 | bio->bi_end_io = end_block_io_op; | |
557 | bio->bi_sector = preq.sector_number; | |
558 | } | |
559 | ||
560 | preq.sector_number += seg[i].nsec; | |
561 | } | |
562 | ||
a1397fa3 | 563 | /* This will be hit if the operation was a barrier. */ |
4d05a28d | 564 | if (!bio) { |
314146e5 | 565 | BUG_ON(operation != WRITE_BARRIER); |
77089926 | 566 | bio = biolist[nbio++] = bio_alloc(GFP_KERNEL, 0); |
4d05a28d KRW |
567 | if (unlikely(bio == NULL)) |
568 | goto fail_put_bio; | |
569 | ||
570 | bio->bi_bdev = preq.bdev; | |
571 | bio->bi_private = pending_req; | |
572 | bio->bi_end_io = end_block_io_op; | |
573 | bio->bi_sector = -1; | |
574 | } | |
575 | ||
77089926 KRW |
576 | |
577 | /* We set it one so that the last submit_bio does not have to call | |
578 | * atomic_inc. | |
579 | */ | |
580 | atomic_set(&pending_req->pendcnt, nbio); | |
581 | ||
582 | /* Get a reference count for the disk queue and start sending I/O */ | |
583 | blk_get_queue(q); | |
584 | blk_start_plug(&plug); | |
585 | ||
586 | for (i = 0; i < nbio; i++) | |
587 | submit_bio(operation, biolist[i]); | |
588 | ||
589 | blk_finish_plug(&plug); | |
590 | /* Let the I/Os go.. */ | |
591 | blk_put_queue(q); | |
4d05a28d KRW |
592 | |
593 | if (operation == READ) | |
594 | blkif->st_rd_sect += preq.nr_sects; | |
314146e5 | 595 | else if (operation == WRITE || operation == WRITE_BARRIER) |
4d05a28d KRW |
596 | blkif->st_wr_sect += preq.nr_sects; |
597 | ||
598 | return; | |
599 | ||
600 | fail_flush: | |
601 | fast_flush_area(pending_req); | |
602 | fail_response: | |
0faa8cca | 603 | /* Haven't submitted any bio's yet. */ |
4d05a28d KRW |
604 | make_response(blkif, req->id, req->operation, BLKIF_RSP_ERROR); |
605 | free_req(pending_req); | |
606 | msleep(1); /* back off a bit */ | |
607 | return; | |
608 | ||
609 | fail_put_bio: | |
77089926 KRW |
610 | for (i = 0; i < (nbio-1); i++) |
611 | bio_put(biolist[i]); | |
4d05a28d | 612 | __end_block_io_op(pending_req, -EINVAL); |
4d05a28d KRW |
613 | msleep(1); /* back off a bit */ |
614 | return; | |
615 | } | |
616 | ||
617 | ||
618 | ||
a1397fa3 KRW |
619 | /* |
620 | * Put a response on the ring on how the operation fared. | |
4d05a28d | 621 | */ |
5489377c | 622 | static void make_response(struct blkif_st *blkif, u64 id, |
4d05a28d KRW |
623 | unsigned short op, int st) |
624 | { | |
88122933 | 625 | struct blkif_response resp; |
4d05a28d | 626 | unsigned long flags; |
88122933 | 627 | union blkif_back_rings *blk_rings = &blkif->blk_rings; |
4d05a28d KRW |
628 | int more_to_do = 0; |
629 | int notify; | |
630 | ||
631 | resp.id = id; | |
632 | resp.operation = op; | |
633 | resp.status = st; | |
634 | ||
635 | spin_lock_irqsave(&blkif->blk_ring_lock, flags); | |
636 | /* Place on the response ring for the relevant domain. */ | |
637 | switch (blkif->blk_protocol) { | |
638 | case BLKIF_PROTOCOL_NATIVE: | |
639 | memcpy(RING_GET_RESPONSE(&blk_rings->native, blk_rings->native.rsp_prod_pvt), | |
640 | &resp, sizeof(resp)); | |
641 | break; | |
642 | case BLKIF_PROTOCOL_X86_32: | |
643 | memcpy(RING_GET_RESPONSE(&blk_rings->x86_32, blk_rings->x86_32.rsp_prod_pvt), | |
644 | &resp, sizeof(resp)); | |
645 | break; | |
646 | case BLKIF_PROTOCOL_X86_64: | |
647 | memcpy(RING_GET_RESPONSE(&blk_rings->x86_64, blk_rings->x86_64.rsp_prod_pvt), | |
648 | &resp, sizeof(resp)); | |
649 | break; | |
650 | default: | |
651 | BUG(); | |
652 | } | |
653 | blk_rings->common.rsp_prod_pvt++; | |
654 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blk_rings->common, notify); | |
655 | if (blk_rings->common.rsp_prod_pvt == blk_rings->common.req_cons) { | |
656 | /* | |
657 | * Tail check for pending requests. Allows frontend to avoid | |
658 | * notifications if requests are already in flight (lower | |
659 | * overheads and promotes batching). | |
660 | */ | |
661 | RING_FINAL_CHECK_FOR_REQUESTS(&blk_rings->common, more_to_do); | |
662 | ||
663 | } else if (RING_HAS_UNCONSUMED_REQUESTS(&blk_rings->common)) { | |
664 | more_to_do = 1; | |
665 | } | |
666 | ||
667 | spin_unlock_irqrestore(&blkif->blk_ring_lock, flags); | |
668 | ||
669 | if (more_to_do) | |
670 | blkif_notify_work(blkif); | |
671 | if (notify) | |
672 | notify_remote_via_irq(blkif->irq); | |
673 | } | |
674 | ||
675 | static int __init blkif_init(void) | |
676 | { | |
677 | int i, mmap_pages; | |
8770b268 | 678 | int rc = 0; |
4d05a28d | 679 | |
88122933 | 680 | if (!xen_pv_domain()) |
4d05a28d KRW |
681 | return -ENODEV; |
682 | ||
2e9977c2 | 683 | blkbk = kzalloc(sizeof(struct xen_blkbk), GFP_KERNEL); |
e8e28871 KRW |
684 | if (!blkbk) { |
685 | printk(KERN_ALERT "%s: out of memory!\n", __func__); | |
686 | return -ENOMEM; | |
687 | } | |
688 | ||
4d05a28d KRW |
689 | mmap_pages = blkif_reqs * BLKIF_MAX_SEGMENTS_PER_REQUEST; |
690 | ||
e8e28871 | 691 | blkbk->pending_reqs = kmalloc(sizeof(blkbk->pending_reqs[0]) * |
4d05a28d | 692 | blkif_reqs, GFP_KERNEL); |
a742b02c KRW |
693 | blkbk->pending_grant_handles = kzalloc(sizeof(blkbk->pending_grant_handles[0]) * |
694 | mmap_pages, GFP_KERNEL); | |
695 | blkbk->pending_pages = kzalloc(sizeof(blkbk->pending_pages[0]) * | |
696 | mmap_pages, GFP_KERNEL); | |
4d05a28d | 697 | |
2e9977c2 KRW |
698 | if (!blkbk->pending_reqs || !blkbk->pending_grant_handles || |
699 | !blkbk->pending_pages) { | |
8770b268 | 700 | rc = -ENOMEM; |
4d05a28d | 701 | goto out_of_memory; |
8770b268 | 702 | } |
4d05a28d | 703 | |
464fb419 | 704 | for (i = 0; i < mmap_pages; i++) { |
e8e28871 | 705 | blkbk->pending_grant_handles[i] = BLKBACK_INVALID_HANDLE; |
a742b02c | 706 | blkbk->pending_pages[i] = alloc_page(GFP_KERNEL); |
464fb419 KRW |
707 | if (blkbk->pending_pages[i] == NULL) { |
708 | rc = -ENOMEM; | |
709 | goto out_of_memory; | |
710 | } | |
711 | } | |
8770b268 KRW |
712 | rc = blkif_interface_init(); |
713 | if (rc) | |
714 | goto failed_init; | |
4d05a28d | 715 | |
e8e28871 KRW |
716 | memset(blkbk->pending_reqs, 0, sizeof(blkbk->pending_reqs)); |
717 | ||
718 | INIT_LIST_HEAD(&blkbk->pending_free); | |
719 | spin_lock_init(&blkbk->pending_free_lock); | |
720 | init_waitqueue_head(&blkbk->pending_free_wq); | |
4d05a28d KRW |
721 | |
722 | for (i = 0; i < blkif_reqs; i++) | |
2e9977c2 KRW |
723 | list_add_tail(&blkbk->pending_reqs[i].free_list, |
724 | &blkbk->pending_free); | |
4d05a28d | 725 | |
8770b268 KRW |
726 | rc = blkif_xenbus_init(); |
727 | if (rc) | |
728 | goto failed_init; | |
4d05a28d KRW |
729 | |
730 | return 0; | |
731 | ||
732 | out_of_memory: | |
8770b268 KRW |
733 | printk(KERN_ERR "%s: out of memory\n", __func__); |
734 | failed_init: | |
e8e28871 | 735 | kfree(blkbk->pending_reqs); |
a742b02c | 736 | kfree(blkbk->pending_grant_handles); |
464fb419 KRW |
737 | for (i = 0; i < mmap_pages; i++) { |
738 | if (blkbk->pending_pages[i]) | |
739 | __free_page(blkbk->pending_pages[i]); | |
740 | } | |
a742b02c KRW |
741 | kfree(blkbk->pending_pages); |
742 | kfree(blkbk); | |
e8e28871 | 743 | blkbk = NULL; |
8770b268 | 744 | return rc; |
4d05a28d KRW |
745 | } |
746 | ||
747 | module_init(blkif_init); | |
748 | ||
749 | MODULE_LICENSE("Dual BSD/GPL"); |