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/page.h>
43 #include <xen/xenbus.h>
44 #include <xen/interface/io/ring.h>
45 #include <xen/interface/io/blkif.h>
46 #include <xen/interface/io/protocols.h>
47 
48 extern unsigned int xen_blkif_max_ring_order;
49 extern unsigned int xenblk_max_queues;
50 /*
51  * This is the maximum number of segments that would be allowed in indirect
52  * requests. This value will also be passed to the frontend.
53  */
54 #define MAX_INDIRECT_SEGMENTS 256
55 
56 /*
57  * Xen use 4K pages. The guest may use different page size (4K or 64K)
58  * Number of Xen pages per segment
59  */
60 #define XEN_PAGES_PER_SEGMENT   (PAGE_SIZE / XEN_PAGE_SIZE)
61 
62 #define XEN_PAGES_PER_INDIRECT_FRAME \
63 	(XEN_PAGE_SIZE/sizeof(struct blkif_request_segment))
64 #define SEGS_PER_INDIRECT_FRAME	\
65 	(XEN_PAGES_PER_INDIRECT_FRAME / XEN_PAGES_PER_SEGMENT)
66 
67 #define MAX_INDIRECT_PAGES \
68 	((MAX_INDIRECT_SEGMENTS + SEGS_PER_INDIRECT_FRAME - 1)/SEGS_PER_INDIRECT_FRAME)
69 #define INDIRECT_PAGES(_segs) DIV_ROUND_UP(_segs, XEN_PAGES_PER_INDIRECT_FRAME)
70 
71 /* Not a real protocol.  Used to generate ring structs which contain
72  * the elements common to all protocols only.  This way we get a
73  * compiler-checkable way to use common struct elements, so we can
74  * avoid using switch(protocol) in a number of places.  */
75 struct blkif_common_request {
76 	char dummy;
77 };
78 struct blkif_common_response {
79 	char dummy;
80 };
81 
82 struct blkif_x86_32_request_rw {
83 	uint8_t        nr_segments;  /* number of segments                   */
84 	blkif_vdev_t   handle;       /* only for read/write requests         */
85 	uint64_t       id;           /* private guest value, echoed in resp  */
86 	blkif_sector_t sector_number;/* start sector idx on disk (r/w only)  */
87 	struct blkif_request_segment seg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
88 } __attribute__((__packed__));
89 
90 struct blkif_x86_32_request_discard {
91 	uint8_t        flag;         /* BLKIF_DISCARD_SECURE or zero         */
92 	blkif_vdev_t   _pad1;        /* was "handle" for read/write requests */
93 	uint64_t       id;           /* private guest value, echoed in resp  */
94 	blkif_sector_t sector_number;/* start sector idx on disk (r/w only)  */
95 	uint64_t       nr_sectors;
96 } __attribute__((__packed__));
97 
98 struct blkif_x86_32_request_other {
99 	uint8_t        _pad1;
100 	blkif_vdev_t   _pad2;
101 	uint64_t       id;           /* private guest value, echoed in resp  */
102 } __attribute__((__packed__));
103 
104 struct blkif_x86_32_request_indirect {
105 	uint8_t        indirect_op;
106 	uint16_t       nr_segments;
107 	uint64_t       id;
108 	blkif_sector_t sector_number;
109 	blkif_vdev_t   handle;
110 	uint16_t       _pad1;
111 	grant_ref_t    indirect_grefs[BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST];
112 	/*
113 	 * The maximum number of indirect segments (and pages) that will
114 	 * be used is determined by MAX_INDIRECT_SEGMENTS, this value
115 	 * is also exported to the guest (via xenstore
116 	 * feature-max-indirect-segments entry), so the frontend knows how
117 	 * many indirect segments the backend supports.
118 	 */
119 	uint64_t       _pad2;        /* make it 64 byte aligned */
120 } __attribute__((__packed__));
121 
122 struct blkif_x86_32_request {
123 	uint8_t        operation;    /* BLKIF_OP_???                         */
124 	union {
125 		struct blkif_x86_32_request_rw rw;
126 		struct blkif_x86_32_request_discard discard;
127 		struct blkif_x86_32_request_other other;
128 		struct blkif_x86_32_request_indirect indirect;
129 	} u;
130 } __attribute__((__packed__));
131 
132 /* i386 protocol version */
133 #pragma pack(push, 4)
134 struct blkif_x86_32_response {
135 	uint64_t        id;              /* copied from request */
136 	uint8_t         operation;       /* copied from request */
137 	int16_t         status;          /* BLKIF_RSP_???       */
138 };
139 #pragma pack(pop)
140 /* x86_64 protocol version */
141 
142 struct blkif_x86_64_request_rw {
143 	uint8_t        nr_segments;  /* number of segments                   */
144 	blkif_vdev_t   handle;       /* only for read/write requests         */
145 	uint32_t       _pad1;        /* offsetof(blkif_reqest..,u.rw.id)==8  */
146 	uint64_t       id;
147 	blkif_sector_t sector_number;/* start sector idx on disk (r/w only)  */
148 	struct blkif_request_segment seg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
149 } __attribute__((__packed__));
150 
151 struct blkif_x86_64_request_discard {
152 	uint8_t        flag;         /* BLKIF_DISCARD_SECURE or zero         */
153 	blkif_vdev_t   _pad1;        /* was "handle" for read/write requests */
154         uint32_t       _pad2;        /* offsetof(blkif_..,u.discard.id)==8   */
155 	uint64_t       id;
156 	blkif_sector_t sector_number;/* start sector idx on disk (r/w only)  */
157 	uint64_t       nr_sectors;
158 } __attribute__((__packed__));
159 
160 struct blkif_x86_64_request_other {
161 	uint8_t        _pad1;
162 	blkif_vdev_t   _pad2;
163 	uint32_t       _pad3;        /* offsetof(blkif_..,u.discard.id)==8   */
164 	uint64_t       id;           /* private guest value, echoed in resp  */
165 } __attribute__((__packed__));
166 
167 struct blkif_x86_64_request_indirect {
168 	uint8_t        indirect_op;
169 	uint16_t       nr_segments;
170 	uint32_t       _pad1;        /* offsetof(blkif_..,u.indirect.id)==8   */
171 	uint64_t       id;
172 	blkif_sector_t sector_number;
173 	blkif_vdev_t   handle;
174 	uint16_t       _pad2;
175 	grant_ref_t    indirect_grefs[BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST];
176 	/*
177 	 * The maximum number of indirect segments (and pages) that will
178 	 * be used is determined by MAX_INDIRECT_SEGMENTS, this value
179 	 * is also exported to the guest (via xenstore
180 	 * feature-max-indirect-segments entry), so the frontend knows how
181 	 * many indirect segments the backend supports.
182 	 */
183 	uint32_t       _pad3;        /* make it 64 byte aligned */
184 } __attribute__((__packed__));
185 
186 struct blkif_x86_64_request {
187 	uint8_t        operation;    /* BLKIF_OP_???                         */
188 	union {
189 		struct blkif_x86_64_request_rw rw;
190 		struct blkif_x86_64_request_discard discard;
191 		struct blkif_x86_64_request_other other;
192 		struct blkif_x86_64_request_indirect indirect;
193 	} u;
194 } __attribute__((__packed__));
195 
196 struct blkif_x86_64_response {
197 	uint64_t       __attribute__((__aligned__(8))) id;
198 	uint8_t         operation;       /* copied from request */
199 	int16_t         status;          /* BLKIF_RSP_???       */
200 };
201 
202 DEFINE_RING_TYPES(blkif_common, struct blkif_common_request,
203 		  struct blkif_common_response);
204 DEFINE_RING_TYPES(blkif_x86_32, struct blkif_x86_32_request,
205 		  struct blkif_x86_32_response);
206 DEFINE_RING_TYPES(blkif_x86_64, struct blkif_x86_64_request,
207 		  struct blkif_x86_64_response);
208 
209 union blkif_back_rings {
210 	struct blkif_back_ring        native;
211 	struct blkif_common_back_ring common;
212 	struct blkif_x86_32_back_ring x86_32;
213 	struct blkif_x86_64_back_ring x86_64;
214 };
215 
216 enum blkif_protocol {
217 	BLKIF_PROTOCOL_NATIVE = 1,
218 	BLKIF_PROTOCOL_X86_32 = 2,
219 	BLKIF_PROTOCOL_X86_64 = 3,
220 };
221 
222 /*
223  * Default protocol if the frontend doesn't specify one.
224  */
225 #ifdef CONFIG_X86
226 #  define BLKIF_PROTOCOL_DEFAULT BLKIF_PROTOCOL_X86_32
227 #else
228 #  define BLKIF_PROTOCOL_DEFAULT BLKIF_PROTOCOL_NATIVE
229 #endif
230 
231 struct xen_vbd {
232 	/* What the domain refers to this vbd as. */
233 	blkif_vdev_t		handle;
234 	/* Non-zero -> read-only */
235 	unsigned char		readonly;
236 	/* VDISK_xxx */
237 	unsigned char		type;
238 	/* phys device that this vbd maps to. */
239 	u32			pdevice;
240 	struct block_device	*bdev;
241 	/* Cached size parameter. */
242 	sector_t		size;
243 	unsigned int		flush_support:1;
244 	unsigned int		discard_secure:1;
245 	unsigned int		feature_gnt_persistent:1;
246 	unsigned int		overflow_max_grants:1;
247 };
248 
249 struct backend_info;
250 
251 /* Number of available flags */
252 #define PERSISTENT_GNT_FLAGS_SIZE	2
253 /* This persistent grant is currently in use */
254 #define PERSISTENT_GNT_ACTIVE		0
255 /*
256  * This persistent grant has been used, this flag is set when we remove the
257  * PERSISTENT_GNT_ACTIVE, to know that this grant has been used recently.
258  */
259 #define PERSISTENT_GNT_WAS_ACTIVE	1
260 
261 /* Number of requests that we can fit in a ring */
262 #define XEN_BLKIF_REQS_PER_PAGE		32
263 
264 struct persistent_gnt {
265 	struct page *page;
266 	grant_ref_t gnt;
267 	grant_handle_t handle;
268 	DECLARE_BITMAP(flags, PERSISTENT_GNT_FLAGS_SIZE);
269 	struct rb_node node;
270 	struct list_head remove_node;
271 };
272 
273 /* Per-ring information. */
274 struct xen_blkif_ring {
275 	/* Physical parameters of the comms window. */
276 	unsigned int		irq;
277 	union blkif_back_rings	blk_rings;
278 	void			*blk_ring;
279 	/* Private fields. */
280 	spinlock_t		blk_ring_lock;
281 
282 	wait_queue_head_t	wq;
283 	atomic_t		inflight;
284 	/* One thread per blkif ring. */
285 	struct task_struct	*xenblkd;
286 	unsigned int		waiting_reqs;
287 
288 	/* List of all 'pending_req' available */
289 	struct list_head	pending_free;
290 	/* And its spinlock. */
291 	spinlock_t		pending_free_lock;
292 	wait_queue_head_t	pending_free_wq;
293 
294 	/* Tree to store persistent grants. */
295 	spinlock_t		pers_gnts_lock;
296 	struct rb_root		persistent_gnts;
297 	unsigned int		persistent_gnt_c;
298 	atomic_t		persistent_gnt_in_use;
299 	unsigned long           next_lru;
300 
301 	/* Statistics. */
302 	unsigned long		st_print;
303 	unsigned long long	st_rd_req;
304 	unsigned long long	st_wr_req;
305 	unsigned long long	st_oo_req;
306 	unsigned long long	st_f_req;
307 	unsigned long long	st_ds_req;
308 	unsigned long long	st_rd_sect;
309 	unsigned long long	st_wr_sect;
310 
311 	/* Used by the kworker that offload work from the persistent purge. */
312 	struct list_head	persistent_purge_list;
313 	struct work_struct	persistent_purge_work;
314 
315 	/* Buffer of free pages to map grant refs. */
316 	spinlock_t		free_pages_lock;
317 	int			free_pages_num;
318 	struct list_head	free_pages;
319 
320 	struct work_struct	free_work;
321 	/* Thread shutdown wait queue. */
322 	wait_queue_head_t	shutdown_wq;
323 	struct xen_blkif 	*blkif;
324 };
325 
326 struct xen_blkif {
327 	/* Unique identifier for this interface. */
328 	domid_t			domid;
329 	unsigned int		handle;
330 	/* Comms information. */
331 	enum blkif_protocol	blk_protocol;
332 	/* The VBD attached to this interface. */
333 	struct xen_vbd		vbd;
334 	/* Back pointer to the backend_info. */
335 	struct backend_info	*be;
336 	atomic_t		refcnt;
337 	/* for barrier (drain) requests */
338 	struct completion	drain_complete;
339 	atomic_t		drain;
340 
341 	struct work_struct	free_work;
342 	unsigned int 		nr_ring_pages;
343 	/* All rings for this device. */
344 	struct xen_blkif_ring	*rings;
345 	unsigned int		nr_rings;
346 };
347 
348 struct seg_buf {
349 	unsigned long offset;
350 	unsigned int nsec;
351 };
352 
353 struct grant_page {
354 	struct page 		*page;
355 	struct persistent_gnt	*persistent_gnt;
356 	grant_handle_t		handle;
357 	grant_ref_t		gref;
358 };
359 
360 /*
361  * Each outstanding request that we've passed to the lower device layers has a
362  * 'pending_req' allocated to it. Each buffer_head that completes decrements
363  * the pendcnt towards zero. When it hits zero, the specified domain has a
364  * response queued for it, with the saved 'id' passed back.
365  */
366 struct pending_req {
367 	struct xen_blkif_ring   *ring;
368 	u64			id;
369 	int			nr_segs;
370 	atomic_t		pendcnt;
371 	unsigned short		operation;
372 	int			status;
373 	struct list_head	free_list;
374 	struct grant_page	*segments[MAX_INDIRECT_SEGMENTS];
375 	/* Indirect descriptors */
376 	struct grant_page	*indirect_pages[MAX_INDIRECT_PAGES];
377 	struct seg_buf		seg[MAX_INDIRECT_SEGMENTS];
378 	struct bio		*biolist[MAX_INDIRECT_SEGMENTS];
379 	struct gnttab_unmap_grant_ref unmap[MAX_INDIRECT_SEGMENTS];
380 	struct page                   *unmap_pages[MAX_INDIRECT_SEGMENTS];
381 	struct gntab_unmap_queue_data gnttab_unmap_data;
382 };
383 
384 
385 #define vbd_sz(_v)	((_v)->bdev->bd_part ? \
386 			 (_v)->bdev->bd_part->nr_sects : \
387 			  get_capacity((_v)->bdev->bd_disk))
388 
389 #define xen_blkif_get(_b) (atomic_inc(&(_b)->refcnt))
390 #define xen_blkif_put(_b)				\
391 	do {						\
392 		if (atomic_dec_and_test(&(_b)->refcnt))	\
393 			schedule_work(&(_b)->free_work);\
394 	} while (0)
395 
396 struct phys_req {
397 	unsigned short		dev;
398 	blkif_sector_t		nr_sects;
399 	struct block_device	*bdev;
400 	blkif_sector_t		sector_number;
401 };
402 int xen_blkif_interface_init(void);
403 
404 int xen_blkif_xenbus_init(void);
405 
406 irqreturn_t xen_blkif_be_int(int irq, void *dev_id);
407 int xen_blkif_schedule(void *arg);
408 int xen_blkif_purge_persistent(void *arg);
409 void xen_blkbk_free_caches(struct xen_blkif_ring *ring);
410 
411 int xen_blkbk_flush_diskcache(struct xenbus_transaction xbt,
412 			      struct backend_info *be, int state);
413 
414 int xen_blkbk_barrier(struct xenbus_transaction xbt,
415 		      struct backend_info *be, int state);
416 struct xenbus_device *xen_blkbk_xenbus(struct backend_info *be);
417 void xen_blkbk_unmap_purged_grants(struct work_struct *work);
418 
419 static inline void blkif_get_x86_32_req(struct blkif_request *dst,
420 					struct blkif_x86_32_request *src)
421 {
422 	int i, n = BLKIF_MAX_SEGMENTS_PER_REQUEST, j;
423 	dst->operation = READ_ONCE(src->operation);
424 	switch (dst->operation) {
425 	case BLKIF_OP_READ:
426 	case BLKIF_OP_WRITE:
427 	case BLKIF_OP_WRITE_BARRIER:
428 	case BLKIF_OP_FLUSH_DISKCACHE:
429 		dst->u.rw.nr_segments = src->u.rw.nr_segments;
430 		dst->u.rw.handle = src->u.rw.handle;
431 		dst->u.rw.id = src->u.rw.id;
432 		dst->u.rw.sector_number = src->u.rw.sector_number;
433 		barrier();
434 		if (n > dst->u.rw.nr_segments)
435 			n = dst->u.rw.nr_segments;
436 		for (i = 0; i < n; i++)
437 			dst->u.rw.seg[i] = src->u.rw.seg[i];
438 		break;
439 	case BLKIF_OP_DISCARD:
440 		dst->u.discard.flag = src->u.discard.flag;
441 		dst->u.discard.id = src->u.discard.id;
442 		dst->u.discard.sector_number = src->u.discard.sector_number;
443 		dst->u.discard.nr_sectors = src->u.discard.nr_sectors;
444 		break;
445 	case BLKIF_OP_INDIRECT:
446 		dst->u.indirect.indirect_op = src->u.indirect.indirect_op;
447 		dst->u.indirect.nr_segments = src->u.indirect.nr_segments;
448 		dst->u.indirect.handle = src->u.indirect.handle;
449 		dst->u.indirect.id = src->u.indirect.id;
450 		dst->u.indirect.sector_number = src->u.indirect.sector_number;
451 		barrier();
452 		j = min(MAX_INDIRECT_PAGES, INDIRECT_PAGES(dst->u.indirect.nr_segments));
453 		for (i = 0; i < j; i++)
454 			dst->u.indirect.indirect_grefs[i] =
455 				src->u.indirect.indirect_grefs[i];
456 		break;
457 	default:
458 		/*
459 		 * Don't know how to translate this op. Only get the
460 		 * ID so failure can be reported to the frontend.
461 		 */
462 		dst->u.other.id = src->u.other.id;
463 		break;
464 	}
465 }
466 
467 static inline void blkif_get_x86_64_req(struct blkif_request *dst,
468 					struct blkif_x86_64_request *src)
469 {
470 	int i, n = BLKIF_MAX_SEGMENTS_PER_REQUEST, j;
471 	dst->operation = READ_ONCE(src->operation);
472 	switch (dst->operation) {
473 	case BLKIF_OP_READ:
474 	case BLKIF_OP_WRITE:
475 	case BLKIF_OP_WRITE_BARRIER:
476 	case BLKIF_OP_FLUSH_DISKCACHE:
477 		dst->u.rw.nr_segments = src->u.rw.nr_segments;
478 		dst->u.rw.handle = src->u.rw.handle;
479 		dst->u.rw.id = src->u.rw.id;
480 		dst->u.rw.sector_number = src->u.rw.sector_number;
481 		barrier();
482 		if (n > dst->u.rw.nr_segments)
483 			n = dst->u.rw.nr_segments;
484 		for (i = 0; i < n; i++)
485 			dst->u.rw.seg[i] = src->u.rw.seg[i];
486 		break;
487 	case BLKIF_OP_DISCARD:
488 		dst->u.discard.flag = src->u.discard.flag;
489 		dst->u.discard.id = src->u.discard.id;
490 		dst->u.discard.sector_number = src->u.discard.sector_number;
491 		dst->u.discard.nr_sectors = src->u.discard.nr_sectors;
492 		break;
493 	case BLKIF_OP_INDIRECT:
494 		dst->u.indirect.indirect_op = src->u.indirect.indirect_op;
495 		dst->u.indirect.nr_segments = src->u.indirect.nr_segments;
496 		dst->u.indirect.handle = src->u.indirect.handle;
497 		dst->u.indirect.id = src->u.indirect.id;
498 		dst->u.indirect.sector_number = src->u.indirect.sector_number;
499 		barrier();
500 		j = min(MAX_INDIRECT_PAGES, INDIRECT_PAGES(dst->u.indirect.nr_segments));
501 		for (i = 0; i < j; i++)
502 			dst->u.indirect.indirect_grefs[i] =
503 				src->u.indirect.indirect_grefs[i];
504 		break;
505 	default:
506 		/*
507 		 * Don't know how to translate this op. Only get the
508 		 * ID so failure can be reported to the frontend.
509 		 */
510 		dst->u.other.id = src->u.other.id;
511 		break;
512 	}
513 }
514 
515 #endif /* __XEN_BLKIF__BACKEND__COMMON_H__ */
516