1 /*
2  * This program is free software; you can redistribute it and/or
3  * modify it under the terms of the GNU General Public License version 2
4  * as published by the Free Software Foundation; or, when distributed
5  * separately from the Linux kernel or incorporated into other
6  * software packages, subject to the following license:
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this source file (the "Software"), to deal in the Software without
10  * restriction, including without limitation the rights to use, copy, modify,
11  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
12  * and to permit persons to whom the Software is furnished to do so, subject to
13  * the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
21  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
23  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24  * IN THE SOFTWARE.
25  */
26 
27 #ifndef __XEN_BLKIF__BACKEND__COMMON_H__
28 #define __XEN_BLKIF__BACKEND__COMMON_H__
29 
30 #include <linux/module.h>
31 #include <linux/interrupt.h>
32 #include <linux/slab.h>
33 #include <linux/blkdev.h>
34 #include <linux/vmalloc.h>
35 #include <linux/wait.h>
36 #include <linux/io.h>
37 #include <linux/rbtree.h>
38 #include <asm/setup.h>
39 #include <asm/pgalloc.h>
40 #include <asm/hypervisor.h>
41 #include <xen/grant_table.h>
42 #include <xen/xenbus.h>
43 #include <xen/interface/io/ring.h>
44 #include <xen/interface/io/blkif.h>
45 #include <xen/interface/io/protocols.h>
46 
47 /*
48  * This is the maximum number of segments that would be allowed in indirect
49  * requests. This value will also be passed to the frontend.
50  */
51 #define MAX_INDIRECT_SEGMENTS 256
52 
53 #define SEGS_PER_INDIRECT_FRAME \
54 	(PAGE_SIZE/sizeof(struct blkif_request_segment))
55 #define MAX_INDIRECT_PAGES \
56 	((MAX_INDIRECT_SEGMENTS + SEGS_PER_INDIRECT_FRAME - 1)/SEGS_PER_INDIRECT_FRAME)
57 #define INDIRECT_PAGES(_segs) \
58 	((_segs + SEGS_PER_INDIRECT_FRAME - 1)/SEGS_PER_INDIRECT_FRAME)
59 
60 /* Not a real protocol.  Used to generate ring structs which contain
61  * the elements common to all protocols only.  This way we get a
62  * compiler-checkable way to use common struct elements, so we can
63  * avoid using switch(protocol) in a number of places.  */
64 struct blkif_common_request {
65 	char dummy;
66 };
67 struct blkif_common_response {
68 	char dummy;
69 };
70 
71 struct blkif_x86_32_request_rw {
72 	uint8_t        nr_segments;  /* number of segments                   */
73 	blkif_vdev_t   handle;       /* only for read/write requests         */
74 	uint64_t       id;           /* private guest value, echoed in resp  */
75 	blkif_sector_t sector_number;/* start sector idx on disk (r/w only)  */
76 	struct blkif_request_segment seg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
77 } __attribute__((__packed__));
78 
79 struct blkif_x86_32_request_discard {
80 	uint8_t        flag;         /* BLKIF_DISCARD_SECURE or zero         */
81 	blkif_vdev_t   _pad1;        /* was "handle" for read/write requests */
82 	uint64_t       id;           /* private guest value, echoed in resp  */
83 	blkif_sector_t sector_number;/* start sector idx on disk (r/w only)  */
84 	uint64_t       nr_sectors;
85 } __attribute__((__packed__));
86 
87 struct blkif_x86_32_request_other {
88 	uint8_t        _pad1;
89 	blkif_vdev_t   _pad2;
90 	uint64_t       id;           /* private guest value, echoed in resp  */
91 } __attribute__((__packed__));
92 
93 struct blkif_x86_32_request_indirect {
94 	uint8_t        indirect_op;
95 	uint16_t       nr_segments;
96 	uint64_t       id;
97 	blkif_sector_t sector_number;
98 	blkif_vdev_t   handle;
99 	uint16_t       _pad1;
100 	grant_ref_t    indirect_grefs[BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST];
101 	/*
102 	 * The maximum number of indirect segments (and pages) that will
103 	 * be used is determined by MAX_INDIRECT_SEGMENTS, this value
104 	 * is also exported to the guest (via xenstore
105 	 * feature-max-indirect-segments entry), so the frontend knows how
106 	 * many indirect segments the backend supports.
107 	 */
108 	uint64_t       _pad2;        /* make it 64 byte aligned */
109 } __attribute__((__packed__));
110 
111 struct blkif_x86_32_request {
112 	uint8_t        operation;    /* BLKIF_OP_???                         */
113 	union {
114 		struct blkif_x86_32_request_rw rw;
115 		struct blkif_x86_32_request_discard discard;
116 		struct blkif_x86_32_request_other other;
117 		struct blkif_x86_32_request_indirect indirect;
118 	} u;
119 } __attribute__((__packed__));
120 
121 /* i386 protocol version */
122 #pragma pack(push, 4)
123 struct blkif_x86_32_response {
124 	uint64_t        id;              /* copied from request */
125 	uint8_t         operation;       /* copied from request */
126 	int16_t         status;          /* BLKIF_RSP_???       */
127 };
128 #pragma pack(pop)
129 /* x86_64 protocol version */
130 
131 struct blkif_x86_64_request_rw {
132 	uint8_t        nr_segments;  /* number of segments                   */
133 	blkif_vdev_t   handle;       /* only for read/write requests         */
134 	uint32_t       _pad1;        /* offsetof(blkif_reqest..,u.rw.id)==8  */
135 	uint64_t       id;
136 	blkif_sector_t sector_number;/* start sector idx on disk (r/w only)  */
137 	struct blkif_request_segment seg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
138 } __attribute__((__packed__));
139 
140 struct blkif_x86_64_request_discard {
141 	uint8_t        flag;         /* BLKIF_DISCARD_SECURE or zero         */
142 	blkif_vdev_t   _pad1;        /* was "handle" for read/write requests */
143         uint32_t       _pad2;        /* offsetof(blkif_..,u.discard.id)==8   */
144 	uint64_t       id;
145 	blkif_sector_t sector_number;/* start sector idx on disk (r/w only)  */
146 	uint64_t       nr_sectors;
147 } __attribute__((__packed__));
148 
149 struct blkif_x86_64_request_other {
150 	uint8_t        _pad1;
151 	blkif_vdev_t   _pad2;
152 	uint32_t       _pad3;        /* offsetof(blkif_..,u.discard.id)==8   */
153 	uint64_t       id;           /* private guest value, echoed in resp  */
154 } __attribute__((__packed__));
155 
156 struct blkif_x86_64_request_indirect {
157 	uint8_t        indirect_op;
158 	uint16_t       nr_segments;
159 	uint32_t       _pad1;        /* offsetof(blkif_..,u.indirect.id)==8   */
160 	uint64_t       id;
161 	blkif_sector_t sector_number;
162 	blkif_vdev_t   handle;
163 	uint16_t       _pad2;
164 	grant_ref_t    indirect_grefs[BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST];
165 	/*
166 	 * The maximum number of indirect segments (and pages) that will
167 	 * be used is determined by MAX_INDIRECT_SEGMENTS, this value
168 	 * is also exported to the guest (via xenstore
169 	 * feature-max-indirect-segments entry), so the frontend knows how
170 	 * many indirect segments the backend supports.
171 	 */
172 	uint32_t       _pad3;        /* make it 64 byte aligned */
173 } __attribute__((__packed__));
174 
175 struct blkif_x86_64_request {
176 	uint8_t        operation;    /* BLKIF_OP_???                         */
177 	union {
178 		struct blkif_x86_64_request_rw rw;
179 		struct blkif_x86_64_request_discard discard;
180 		struct blkif_x86_64_request_other other;
181 		struct blkif_x86_64_request_indirect indirect;
182 	} u;
183 } __attribute__((__packed__));
184 
185 struct blkif_x86_64_response {
186 	uint64_t       __attribute__((__aligned__(8))) id;
187 	uint8_t         operation;       /* copied from request */
188 	int16_t         status;          /* BLKIF_RSP_???       */
189 };
190 
191 DEFINE_RING_TYPES(blkif_common, struct blkif_common_request,
192 		  struct blkif_common_response);
193 DEFINE_RING_TYPES(blkif_x86_32, struct blkif_x86_32_request,
194 		  struct blkif_x86_32_response);
195 DEFINE_RING_TYPES(blkif_x86_64, struct blkif_x86_64_request,
196 		  struct blkif_x86_64_response);
197 
198 union blkif_back_rings {
199 	struct blkif_back_ring        native;
200 	struct blkif_common_back_ring common;
201 	struct blkif_x86_32_back_ring x86_32;
202 	struct blkif_x86_64_back_ring x86_64;
203 };
204 
205 enum blkif_protocol {
206 	BLKIF_PROTOCOL_NATIVE = 1,
207 	BLKIF_PROTOCOL_X86_32 = 2,
208 	BLKIF_PROTOCOL_X86_64 = 3,
209 };
210 
211 /*
212  * Default protocol if the frontend doesn't specify one.
213  */
214 #ifdef CONFIG_X86
215 #  define BLKIF_PROTOCOL_DEFAULT BLKIF_PROTOCOL_X86_32
216 #else
217 #  define BLKIF_PROTOCOL_DEFAULT BLKIF_PROTOCOL_NATIVE
218 #endif
219 
220 struct xen_vbd {
221 	/* What the domain refers to this vbd as. */
222 	blkif_vdev_t		handle;
223 	/* Non-zero -> read-only */
224 	unsigned char		readonly;
225 	/* VDISK_xxx */
226 	unsigned char		type;
227 	/* phys device that this vbd maps to. */
228 	u32			pdevice;
229 	struct block_device	*bdev;
230 	/* Cached size parameter. */
231 	sector_t		size;
232 	unsigned int		flush_support:1;
233 	unsigned int		discard_secure:1;
234 	unsigned int		feature_gnt_persistent:1;
235 	unsigned int		overflow_max_grants:1;
236 };
237 
238 struct backend_info;
239 
240 /* Number of available flags */
241 #define PERSISTENT_GNT_FLAGS_SIZE	2
242 /* This persistent grant is currently in use */
243 #define PERSISTENT_GNT_ACTIVE		0
244 /*
245  * This persistent grant has been used, this flag is set when we remove the
246  * PERSISTENT_GNT_ACTIVE, to know that this grant has been used recently.
247  */
248 #define PERSISTENT_GNT_WAS_ACTIVE	1
249 
250 /* Number of requests that we can fit in a ring */
251 #define XEN_BLKIF_REQS			32
252 
253 struct persistent_gnt {
254 	struct page *page;
255 	grant_ref_t gnt;
256 	grant_handle_t handle;
257 	DECLARE_BITMAP(flags, PERSISTENT_GNT_FLAGS_SIZE);
258 	struct rb_node node;
259 	struct list_head remove_node;
260 };
261 
262 struct xen_blkif {
263 	/* Unique identifier for this interface. */
264 	domid_t			domid;
265 	unsigned int		handle;
266 	/* Physical parameters of the comms window. */
267 	unsigned int		irq;
268 	/* Comms information. */
269 	enum blkif_protocol	blk_protocol;
270 	union blkif_back_rings	blk_rings;
271 	void			*blk_ring;
272 	/* The VBD attached to this interface. */
273 	struct xen_vbd		vbd;
274 	/* Back pointer to the backend_info. */
275 	struct backend_info	*be;
276 	/* Private fields. */
277 	spinlock_t		blk_ring_lock;
278 	atomic_t		refcnt;
279 
280 	wait_queue_head_t	wq;
281 	/* for barrier (drain) requests */
282 	struct completion	drain_complete;
283 	atomic_t		drain;
284 	atomic_t		inflight;
285 	/* One thread per one blkif. */
286 	struct task_struct	*xenblkd;
287 	unsigned int		waiting_reqs;
288 
289 	/* tree to store persistent grants */
290 	struct rb_root		persistent_gnts;
291 	unsigned int		persistent_gnt_c;
292 	atomic_t		persistent_gnt_in_use;
293 	unsigned long           next_lru;
294 
295 	/* used by the kworker that offload work from the persistent purge */
296 	struct list_head	persistent_purge_list;
297 	struct work_struct	persistent_purge_work;
298 
299 	/* buffer of free pages to map grant refs */
300 	spinlock_t		free_pages_lock;
301 	int			free_pages_num;
302 	struct list_head	free_pages;
303 
304 	/* List of all 'pending_req' available */
305 	struct list_head	pending_free;
306 	/* And its spinlock. */
307 	spinlock_t		pending_free_lock;
308 	wait_queue_head_t	pending_free_wq;
309 
310 	/* statistics */
311 	unsigned long		st_print;
312 	unsigned long long			st_rd_req;
313 	unsigned long long			st_wr_req;
314 	unsigned long long			st_oo_req;
315 	unsigned long long			st_f_req;
316 	unsigned long long			st_ds_req;
317 	unsigned long long			st_rd_sect;
318 	unsigned long long			st_wr_sect;
319 
320 	struct work_struct	free_work;
321 	/* Thread shutdown wait queue. */
322 	wait_queue_head_t	shutdown_wq;
323 };
324 
325 struct seg_buf {
326 	unsigned long offset;
327 	unsigned int nsec;
328 };
329 
330 struct grant_page {
331 	struct page 		*page;
332 	struct persistent_gnt	*persistent_gnt;
333 	grant_handle_t		handle;
334 	grant_ref_t		gref;
335 };
336 
337 /*
338  * Each outstanding request that we've passed to the lower device layers has a
339  * 'pending_req' allocated to it. Each buffer_head that completes decrements
340  * the pendcnt towards zero. When it hits zero, the specified domain has a
341  * response queued for it, with the saved 'id' passed back.
342  */
343 struct pending_req {
344 	struct xen_blkif	*blkif;
345 	u64			id;
346 	int			nr_pages;
347 	atomic_t		pendcnt;
348 	unsigned short		operation;
349 	int			status;
350 	struct list_head	free_list;
351 	struct grant_page	*segments[MAX_INDIRECT_SEGMENTS];
352 	/* Indirect descriptors */
353 	struct grant_page	*indirect_pages[MAX_INDIRECT_PAGES];
354 	struct seg_buf		seg[MAX_INDIRECT_SEGMENTS];
355 	struct bio		*biolist[MAX_INDIRECT_SEGMENTS];
356 	struct gnttab_unmap_grant_ref unmap[MAX_INDIRECT_SEGMENTS];
357 	struct page                   *unmap_pages[MAX_INDIRECT_SEGMENTS];
358 	struct gntab_unmap_queue_data gnttab_unmap_data;
359 };
360 
361 
362 #define vbd_sz(_v)	((_v)->bdev->bd_part ? \
363 			 (_v)->bdev->bd_part->nr_sects : \
364 			  get_capacity((_v)->bdev->bd_disk))
365 
366 #define xen_blkif_get(_b) (atomic_inc(&(_b)->refcnt))
367 #define xen_blkif_put(_b)				\
368 	do {						\
369 		if (atomic_dec_and_test(&(_b)->refcnt))	\
370 			schedule_work(&(_b)->free_work);\
371 	} while (0)
372 
373 struct phys_req {
374 	unsigned short		dev;
375 	blkif_sector_t		nr_sects;
376 	struct block_device	*bdev;
377 	blkif_sector_t		sector_number;
378 };
379 int xen_blkif_interface_init(void);
380 
381 int xen_blkif_xenbus_init(void);
382 
383 irqreturn_t xen_blkif_be_int(int irq, void *dev_id);
384 int xen_blkif_schedule(void *arg);
385 int xen_blkif_purge_persistent(void *arg);
386 void xen_blkbk_free_caches(struct xen_blkif *blkif);
387 
388 int xen_blkbk_flush_diskcache(struct xenbus_transaction xbt,
389 			      struct backend_info *be, int state);
390 
391 int xen_blkbk_barrier(struct xenbus_transaction xbt,
392 		      struct backend_info *be, int state);
393 struct xenbus_device *xen_blkbk_xenbus(struct backend_info *be);
394 void xen_blkbk_unmap_purged_grants(struct work_struct *work);
395 
396 static inline void blkif_get_x86_32_req(struct blkif_request *dst,
397 					struct blkif_x86_32_request *src)
398 {
399 	int i, n = BLKIF_MAX_SEGMENTS_PER_REQUEST, j;
400 	dst->operation = src->operation;
401 	switch (src->operation) {
402 	case BLKIF_OP_READ:
403 	case BLKIF_OP_WRITE:
404 	case BLKIF_OP_WRITE_BARRIER:
405 	case BLKIF_OP_FLUSH_DISKCACHE:
406 		dst->u.rw.nr_segments = src->u.rw.nr_segments;
407 		dst->u.rw.handle = src->u.rw.handle;
408 		dst->u.rw.id = src->u.rw.id;
409 		dst->u.rw.sector_number = src->u.rw.sector_number;
410 		barrier();
411 		if (n > dst->u.rw.nr_segments)
412 			n = dst->u.rw.nr_segments;
413 		for (i = 0; i < n; i++)
414 			dst->u.rw.seg[i] = src->u.rw.seg[i];
415 		break;
416 	case BLKIF_OP_DISCARD:
417 		dst->u.discard.flag = src->u.discard.flag;
418 		dst->u.discard.id = src->u.discard.id;
419 		dst->u.discard.sector_number = src->u.discard.sector_number;
420 		dst->u.discard.nr_sectors = src->u.discard.nr_sectors;
421 		break;
422 	case BLKIF_OP_INDIRECT:
423 		dst->u.indirect.indirect_op = src->u.indirect.indirect_op;
424 		dst->u.indirect.nr_segments = src->u.indirect.nr_segments;
425 		dst->u.indirect.handle = src->u.indirect.handle;
426 		dst->u.indirect.id = src->u.indirect.id;
427 		dst->u.indirect.sector_number = src->u.indirect.sector_number;
428 		barrier();
429 		j = min(MAX_INDIRECT_PAGES, INDIRECT_PAGES(dst->u.indirect.nr_segments));
430 		for (i = 0; i < j; i++)
431 			dst->u.indirect.indirect_grefs[i] =
432 				src->u.indirect.indirect_grefs[i];
433 		break;
434 	default:
435 		/*
436 		 * Don't know how to translate this op. Only get the
437 		 * ID so failure can be reported to the frontend.
438 		 */
439 		dst->u.other.id = src->u.other.id;
440 		break;
441 	}
442 }
443 
444 static inline void blkif_get_x86_64_req(struct blkif_request *dst,
445 					struct blkif_x86_64_request *src)
446 {
447 	int i, n = BLKIF_MAX_SEGMENTS_PER_REQUEST, j;
448 	dst->operation = src->operation;
449 	switch (src->operation) {
450 	case BLKIF_OP_READ:
451 	case BLKIF_OP_WRITE:
452 	case BLKIF_OP_WRITE_BARRIER:
453 	case BLKIF_OP_FLUSH_DISKCACHE:
454 		dst->u.rw.nr_segments = src->u.rw.nr_segments;
455 		dst->u.rw.handle = src->u.rw.handle;
456 		dst->u.rw.id = src->u.rw.id;
457 		dst->u.rw.sector_number = src->u.rw.sector_number;
458 		barrier();
459 		if (n > dst->u.rw.nr_segments)
460 			n = dst->u.rw.nr_segments;
461 		for (i = 0; i < n; i++)
462 			dst->u.rw.seg[i] = src->u.rw.seg[i];
463 		break;
464 	case BLKIF_OP_DISCARD:
465 		dst->u.discard.flag = src->u.discard.flag;
466 		dst->u.discard.id = src->u.discard.id;
467 		dst->u.discard.sector_number = src->u.discard.sector_number;
468 		dst->u.discard.nr_sectors = src->u.discard.nr_sectors;
469 		break;
470 	case BLKIF_OP_INDIRECT:
471 		dst->u.indirect.indirect_op = src->u.indirect.indirect_op;
472 		dst->u.indirect.nr_segments = src->u.indirect.nr_segments;
473 		dst->u.indirect.handle = src->u.indirect.handle;
474 		dst->u.indirect.id = src->u.indirect.id;
475 		dst->u.indirect.sector_number = src->u.indirect.sector_number;
476 		barrier();
477 		j = min(MAX_INDIRECT_PAGES, INDIRECT_PAGES(dst->u.indirect.nr_segments));
478 		for (i = 0; i < j; i++)
479 			dst->u.indirect.indirect_grefs[i] =
480 				src->u.indirect.indirect_grefs[i];
481 		break;
482 	default:
483 		/*
484 		 * Don't know how to translate this op. Only get the
485 		 * ID so failure can be reported to the frontend.
486 		 */
487 		dst->u.other.id = src->u.other.id;
488 		break;
489 	}
490 }
491 
492 #endif /* __XEN_BLKIF__BACKEND__COMMON_H__ */
493