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