rbd.c 153.7 KB
Newer Older
1

2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
/*
   rbd.c -- Export ceph rados objects as a Linux block device


   based on drivers/block/osdblk.c:

   Copyright 2009 Red Hat, Inc.

   This program is free software; you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
   the Free Software Foundation.

   This program is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   GNU General Public License for more details.

   You should have received a copy of the GNU General Public License
   along with this program; see the file COPYING.  If not, write to
   the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.



25
   For usage instructions, please refer to:
26

27
                 Documentation/ABI/testing/sysfs-bus-rbd
28 29 30 31 32 33

 */

#include <linux/ceph/libceph.h>
#include <linux/ceph/osd_client.h>
#include <linux/ceph/mon_client.h>
34
#include <linux/ceph/cls_lock_client.h>
35
#include <linux/ceph/decode.h>
36
#include <linux/parser.h>
37
#include <linux/bsearch.h>
38 39 40 41

#include <linux/kernel.h>
#include <linux/device.h>
#include <linux/module.h>
C
Christoph Hellwig 已提交
42
#include <linux/blk-mq.h>
43 44
#include <linux/fs.h>
#include <linux/blkdev.h>
45
#include <linux/slab.h>
46
#include <linux/idr.h>
I
Ilya Dryomov 已提交
47
#include <linux/workqueue.h>
48 49 50

#include "rbd_types.h"

A
Alex Elder 已提交
51 52
#define RBD_DEBUG	/* Activate rbd_assert() calls */

A
Alex Elder 已提交
53 54 55 56 57 58 59 60 61
/*
 * The basic unit of block I/O is a sector.  It is interpreted in a
 * number of contexts in Linux (blk, bio, genhd), but the default is
 * universally 512 bytes.  These symbols are just slightly more
 * meaningful than the bare numbers they represent.
 */
#define	SECTOR_SHIFT	9
#define	SECTOR_SIZE	(1ULL << SECTOR_SHIFT)

62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94
/*
 * Increment the given counter and return its updated value.
 * If the counter is already 0 it will not be incremented.
 * If the counter is already at its maximum value returns
 * -EINVAL without updating it.
 */
static int atomic_inc_return_safe(atomic_t *v)
{
	unsigned int counter;

	counter = (unsigned int)__atomic_add_unless(v, 1, 0);
	if (counter <= (unsigned int)INT_MAX)
		return (int)counter;

	atomic_dec(v);

	return -EINVAL;
}

/* Decrement the counter.  Return the resulting value, or -EINVAL */
static int atomic_dec_return_safe(atomic_t *v)
{
	int counter;

	counter = atomic_dec_return(v);
	if (counter >= 0)
		return counter;

	atomic_inc(v);

	return -EINVAL;
}

A
Alex Elder 已提交
95
#define RBD_DRV_NAME "rbd"
96

97 98
#define RBD_MINORS_PER_MAJOR		256
#define RBD_SINGLE_MAJOR_PART_SHIFT	4
99

100 101
#define RBD_MAX_PARENT_CHAIN_LEN	16

102 103 104 105
#define RBD_SNAP_DEV_NAME_PREFIX	"snap_"
#define RBD_MAX_SNAP_NAME_LEN	\
			(NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1))

106
#define RBD_MAX_SNAP_COUNT	510	/* allows max snapc to fit in 4KB */
107 108 109

#define RBD_SNAP_HEAD_NAME	"-"

110 111
#define	BAD_SNAP_INDEX	U32_MAX		/* invalid index into snap array */

112 113
/* This allows a single page to hold an image name sent by OSD */
#define RBD_IMAGE_NAME_LEN_MAX	(PAGE_SIZE - sizeof (__le32) - 1)
114
#define RBD_IMAGE_ID_LEN_MAX	64
115

116
#define RBD_OBJ_PREFIX_LEN_MAX	64
A
Alex Elder 已提交
117

118
#define RBD_NOTIFY_TIMEOUT	5	/* seconds */
119 120
#define RBD_RETRY_DELAY		msecs_to_jiffies(1000)

A
Alex Elder 已提交
121 122
/* Feature bits */

123 124 125 126
#define RBD_FEATURE_LAYERING		(1ULL<<0)
#define RBD_FEATURE_STRIPINGV2		(1ULL<<1)
#define RBD_FEATURE_EXCLUSIVE_LOCK	(1ULL<<2)
#define RBD_FEATURE_DATA_POOL		(1ULL<<7)
127
#define RBD_FEATURE_OPERATIONS		(1ULL<<8)
128

129 130
#define RBD_FEATURES_ALL	(RBD_FEATURE_LAYERING |		\
				 RBD_FEATURE_STRIPINGV2 |	\
131
				 RBD_FEATURE_EXCLUSIVE_LOCK |	\
132 133
				 RBD_FEATURE_DATA_POOL |	\
				 RBD_FEATURE_OPERATIONS)
A
Alex Elder 已提交
134 135 136

/* Features supported by this (client software) implementation. */

137
#define RBD_FEATURES_SUPPORTED	(RBD_FEATURES_ALL)
A
Alex Elder 已提交
138

A
Alex Elder 已提交
139 140 141 142
/*
 * An RBD device name will be "rbd#", where the "rbd" comes from
 * RBD_DRV_NAME above, and # is a unique integer identifier.
 */
143 144 145 146 147 148
#define DEV_NAME_LEN		32

/*
 * block device image metadata (in-memory version)
 */
struct rbd_image_header {
149
	/* These six fields never change for a given rbd image */
150
	char *object_prefix;
151
	__u8 obj_order;
152 153
	u64 stripe_unit;
	u64 stripe_count;
154
	s64 data_pool_id;
155
	u64 features;		/* Might be changeable someday? */
156

A
Alex Elder 已提交
157 158 159
	/* The remaining fields need to be updated occasionally */
	u64 image_size;
	struct ceph_snap_context *snapc;
160 161
	char *snap_names;	/* format 1 only */
	u64 *snap_sizes;	/* format 1 only */
162 163
};

164 165 166 167
/*
 * An rbd image specification.
 *
 * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely
A
Alex Elder 已提交
168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187
 * identify an image.  Each rbd_dev structure includes a pointer to
 * an rbd_spec structure that encapsulates this identity.
 *
 * Each of the id's in an rbd_spec has an associated name.  For a
 * user-mapped image, the names are supplied and the id's associated
 * with them are looked up.  For a layered image, a parent image is
 * defined by the tuple, and the names are looked up.
 *
 * An rbd_dev structure contains a parent_spec pointer which is
 * non-null if the image it represents is a child in a layered
 * image.  This pointer will refer to the rbd_spec structure used
 * by the parent rbd_dev for its own identity (i.e., the structure
 * is shared between the parent and child).
 *
 * Since these structures are populated once, during the discovery
 * phase of image construction, they are effectively immutable so
 * we make no effort to synchronize access to them.
 *
 * Note that code herein does not assume the image name is known (it
 * could be a null pointer).
188 189 190
 */
struct rbd_spec {
	u64		pool_id;
191
	const char	*pool_name;
192

193 194
	const char	*image_id;
	const char	*image_name;
195 196

	u64		snap_id;
197
	const char	*snap_name;
198 199 200 201

	struct kref	kref;
};

202
/*
A
Alex Elder 已提交
203
 * an instance of the client.  multiple devices may share an rbd client.
204 205 206 207 208 209 210
 */
struct rbd_client {
	struct ceph_client	*client;
	struct kref		kref;
	struct list_head	node;
};

A
Alex Elder 已提交
211 212 213 214 215 216 217 218
struct rbd_img_request;
typedef void (*rbd_img_callback_t)(struct rbd_img_request *);

#define	BAD_WHICH	U32_MAX		/* Good which or bad which, which? */

struct rbd_obj_request;
typedef void (*rbd_obj_callback_t)(struct rbd_obj_request *);

219
enum obj_request_type {
220
	OBJ_REQUEST_NODATA = 1,
221
	OBJ_REQUEST_BIO,	/* pointer into provided bio (list) */
222
	OBJ_REQUEST_BVECS,	/* pointer into provided bio_vec array */
223
};
A
Alex Elder 已提交
224

G
Guangliang Zhao 已提交
225
enum obj_operation_type {
226
	OBJ_OP_READ = 1,
G
Guangliang Zhao 已提交
227
	OBJ_OP_WRITE,
228
	OBJ_OP_DISCARD,
G
Guangliang Zhao 已提交
229 230
};

231 232
enum obj_req_flags {
	OBJ_REQ_DONE,		/* completion flag: not done = 0, done = 1 */
233
	OBJ_REQ_IMG_DATA,	/* object usage: standalone = 0, image = 1 */
234 235
};

I
Ilya Dryomov 已提交
236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257
/*
 * Writes go through the following state machine to deal with
 * layering:
 *
 *                       need copyup
 * RBD_OBJ_WRITE_GUARD ---------------> RBD_OBJ_WRITE_COPYUP
 *        |     ^                              |
 *        v     \------------------------------/
 *      done
 *        ^
 *        |
 * RBD_OBJ_WRITE_FLAT
 *
 * Writes start in RBD_OBJ_WRITE_GUARD or _FLAT, depending on whether
 * there is a parent or not.
 */
enum rbd_obj_write_state {
	RBD_OBJ_WRITE_FLAT = 1,
	RBD_OBJ_WRITE_GUARD,
	RBD_OBJ_WRITE_COPYUP,
};

A
Alex Elder 已提交
258
struct rbd_obj_request {
259
	u64			object_no;
A
Alex Elder 已提交
260 261
	u64			offset;		/* object start byte */
	u64			length;		/* bytes from offset */
262
	unsigned long		flags;
I
Ilya Dryomov 已提交
263 264 265 266
	union {
		bool			tried_parent;	/* for reads */
		enum rbd_obj_write_state write_state;	/* for writes */
	};
A
Alex Elder 已提交
267

268 269 270 271 272 273 274 275 276
	/*
	 * An object request associated with an image will have its
	 * img_data flag set; a standalone object request will not.
	 *
	 * Finally, an object request for rbd image data will have
	 * which != BAD_WHICH, and will have a non-null img_request
	 * pointer.  The value of which will be in the range
	 * 0..(img_request->obj_request_count-1).
	 */
277 278 279 280
	struct rbd_img_request	*img_request;
	u64			img_offset;
	/* links for img_request->obj_requests list */
	struct list_head	links;
A
Alex Elder 已提交
281 282 283
	u32			which;		/* posn image request list */

	enum obj_request_type	type;
284
	union {
285
		struct ceph_bio_iter	bio_pos;
286
		struct {
287 288
			struct ceph_bvec_iter	bvec_pos;
			u32			bvec_count;
289 290
		};
	};
291 292
	struct bio_vec		*copyup_bvecs;
	u32			copyup_bvec_count;
A
Alex Elder 已提交
293 294 295 296

	struct ceph_osd_request	*osd_req;

	u64			xferred;	/* bytes transferred */
297
	int			result;
A
Alex Elder 已提交
298 299 300 301 302 303

	rbd_obj_callback_t	callback;

	struct kref		kref;
};

A
Alex Elder 已提交
304
enum img_req_flags {
305 306
	IMG_REQ_WRITE,		/* I/O direction: read = 0, write = 1 */
	IMG_REQ_CHILD,		/* initiator: block = 0, child image = 1 */
307
	IMG_REQ_LAYERED,	/* ENOENT handling: normal = 0, layered = 1 */
308
	IMG_REQ_DISCARD,	/* discard: normal = 0, discard request = 1 */
A
Alex Elder 已提交
309 310
};

A
Alex Elder 已提交
311 312 313 314
struct rbd_img_request {
	struct rbd_device	*rbd_dev;
	u64			offset;	/* starting image byte offset */
	u64			length;	/* byte count from offset */
A
Alex Elder 已提交
315
	unsigned long		flags;
A
Alex Elder 已提交
316
	union {
317
		u64			snap_id;	/* for reads */
A
Alex Elder 已提交
318
		struct ceph_snap_context *snapc;	/* for writes */
319 320 321 322
	};
	union {
		struct request		*rq;		/* block request */
		struct rbd_obj_request	*obj_request;	/* obj req initiator */
A
Alex Elder 已提交
323 324 325 326
	};
	spinlock_t		completion_lock;/* protects next_completion */
	u32			next_completion;
	rbd_img_callback_t	callback;
327
	u64			xferred;/* aggregate bytes transferred */
328
	int			result;	/* first nonzero obj_request result */
A
Alex Elder 已提交
329 330 331 332 333 334 335 336

	u32			obj_request_count;
	struct list_head	obj_requests;	/* rbd_obj_request structs */

	struct kref		kref;
};

#define for_each_obj_request(ireq, oreq) \
337
	list_for_each_entry(oreq, &(ireq)->obj_requests, links)
A
Alex Elder 已提交
338
#define for_each_obj_request_from(ireq, oreq) \
339
	list_for_each_entry_from(oreq, &(ireq)->obj_requests, links)
A
Alex Elder 已提交
340
#define for_each_obj_request_safe(ireq, oreq, n) \
341
	list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links)
A
Alex Elder 已提交
342

343 344 345 346 347 348
enum rbd_watch_state {
	RBD_WATCH_STATE_UNREGISTERED,
	RBD_WATCH_STATE_REGISTERED,
	RBD_WATCH_STATE_ERROR,
};

349 350 351 352 353 354 355 356 357 358 359 360
enum rbd_lock_state {
	RBD_LOCK_STATE_UNLOCKED,
	RBD_LOCK_STATE_LOCKED,
	RBD_LOCK_STATE_RELEASING,
};

/* WatchNotify::ClientId */
struct rbd_client_id {
	u64 gid;
	u64 handle;
};

A
Alex Elder 已提交
361
struct rbd_mapping {
A
Alex Elder 已提交
362
	u64                     size;
A
Alex Elder 已提交
363
	u64                     features;
A
Alex Elder 已提交
364 365
};

366 367 368 369
/*
 * a single device
 */
struct rbd_device {
A
Alex Elder 已提交
370
	int			dev_id;		/* blkdev unique id */
371 372

	int			major;		/* blkdev assigned major */
373
	int			minor;
374 375
	struct gendisk		*disk;		/* blkdev's gendisk and rq */

376
	u32			image_format;	/* Either 1 or 2 */
377 378 379 380
	struct rbd_client	*rbd_client;

	char			name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */

381
	spinlock_t		lock;		/* queue, flags, open_count */
382 383

	struct rbd_image_header	header;
384
	unsigned long		flags;		/* possibly lock protected */
385
	struct rbd_spec		*spec;
386
	struct rbd_options	*opts;
387
	char			*config_info;	/* add{,_single_major} string */
388

389
	struct ceph_object_id	header_oid;
390
	struct ceph_object_locator header_oloc;
391

392
	struct ceph_file_layout	layout;		/* used for all rbd requests */
393

394 395
	struct mutex		watch_mutex;
	enum rbd_watch_state	watch_state;
396
	struct ceph_osd_linger_request *watch_handle;
397 398
	u64			watch_cookie;
	struct delayed_work	watch_dwork;
399

400 401
	struct rw_semaphore	lock_rwsem;
	enum rbd_lock_state	lock_state;
I
Ilya Dryomov 已提交
402
	char			lock_cookie[32];
403 404 405 406 407 408 409
	struct rbd_client_id	owner_cid;
	struct work_struct	acquired_lock_work;
	struct work_struct	released_lock_work;
	struct delayed_work	lock_dwork;
	struct work_struct	unlock_work;
	wait_queue_head_t	lock_waitq;

410
	struct workqueue_struct	*task_wq;
411

412 413
	struct rbd_spec		*parent_spec;
	u64			parent_overlap;
414
	atomic_t		parent_ref;
415
	struct rbd_device	*parent;
416

C
Christoph Hellwig 已提交
417 418 419
	/* Block layer tags. */
	struct blk_mq_tag_set	tag_set;

420 421
	/* protects updating the header */
	struct rw_semaphore     header_rwsem;
A
Alex Elder 已提交
422 423

	struct rbd_mapping	mapping;
424 425

	struct list_head	node;
426 427 428

	/* sysfs related */
	struct device		dev;
429
	unsigned long		open_count;	/* protected by lock */
430 431
};

432
/*
433 434 435 436
 * Flag bits for rbd_dev->flags:
 * - REMOVING (which is coupled with rbd_dev->open_count) is protected
 *   by rbd_dev->lock
 * - BLACKLISTED is protected by rbd_dev->lock_rwsem
437
 */
438 439
enum rbd_dev_flags {
	RBD_DEV_FLAG_EXISTS,	/* mapped snapshot has not been deleted */
440
	RBD_DEV_FLAG_REMOVING,	/* this mapping is being removed */
441
	RBD_DEV_FLAG_BLACKLISTED, /* our ceph_client is blacklisted */
442 443
};

444
static DEFINE_MUTEX(client_mutex);	/* Serialize client creation */
445

446
static LIST_HEAD(rbd_dev_list);    /* devices */
447 448
static DEFINE_SPINLOCK(rbd_dev_list_lock);

A
Alex Elder 已提交
449 450
static LIST_HEAD(rbd_client_list);		/* clients */
static DEFINE_SPINLOCK(rbd_client_list_lock);
451

452 453
/* Slab caches for frequently-allocated structures */

454
static struct kmem_cache	*rbd_img_request_cache;
455
static struct kmem_cache	*rbd_obj_request_cache;
456

457
static int rbd_major;
458 459
static DEFINE_IDA(rbd_dev_id_ida);

460 461
static struct workqueue_struct *rbd_wq;

462
/*
463
 * single-major requires >= 0.75 version of userspace rbd utility.
464
 */
465
static bool single_major = true;
466
module_param(single_major, bool, S_IRUGO);
467
MODULE_PARM_DESC(single_major, "Use a single major number for all rbd devices (default: true)");
468

A
Alex Elder 已提交
469 470 471 472
static ssize_t rbd_add(struct bus_type *bus, const char *buf,
		       size_t count);
static ssize_t rbd_remove(struct bus_type *bus, const char *buf,
			  size_t count);
473 474 475 476
static ssize_t rbd_add_single_major(struct bus_type *bus, const char *buf,
				    size_t count);
static ssize_t rbd_remove_single_major(struct bus_type *bus, const char *buf,
				       size_t count);
477
static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth);
478
static void rbd_spec_put(struct rbd_spec *spec);
A
Alex Elder 已提交
479

480 481
static int rbd_dev_id_to_minor(int dev_id)
{
482
	return dev_id << RBD_SINGLE_MAJOR_PART_SHIFT;
483 484 485 486
}

static int minor_to_rbd_dev_id(int minor)
{
487
	return minor >> RBD_SINGLE_MAJOR_PART_SHIFT;
488 489
}

490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505
static bool __rbd_is_lock_owner(struct rbd_device *rbd_dev)
{
	return rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED ||
	       rbd_dev->lock_state == RBD_LOCK_STATE_RELEASING;
}

static bool rbd_is_lock_owner(struct rbd_device *rbd_dev)
{
	bool is_lock_owner;

	down_read(&rbd_dev->lock_rwsem);
	is_lock_owner = __rbd_is_lock_owner(rbd_dev);
	up_read(&rbd_dev->lock_rwsem);
	return is_lock_owner;
}

506 507 508 509 510
static ssize_t rbd_supported_features_show(struct bus_type *bus, char *buf)
{
	return sprintf(buf, "0x%llx\n", RBD_FEATURES_SUPPORTED);
}

511 512
static BUS_ATTR(add, S_IWUSR, NULL, rbd_add);
static BUS_ATTR(remove, S_IWUSR, NULL, rbd_remove);
513 514
static BUS_ATTR(add_single_major, S_IWUSR, NULL, rbd_add_single_major);
static BUS_ATTR(remove_single_major, S_IWUSR, NULL, rbd_remove_single_major);
515
static BUS_ATTR(supported_features, S_IRUGO, rbd_supported_features_show, NULL);
516 517 518 519

static struct attribute *rbd_bus_attrs[] = {
	&bus_attr_add.attr,
	&bus_attr_remove.attr,
520 521
	&bus_attr_add_single_major.attr,
	&bus_attr_remove_single_major.attr,
522
	&bus_attr_supported_features.attr,
523
	NULL,
A
Alex Elder 已提交
524
};
525 526 527 528

static umode_t rbd_bus_is_visible(struct kobject *kobj,
				  struct attribute *attr, int index)
{
529 530 531 532 533
	if (!single_major &&
	    (attr == &bus_attr_add_single_major.attr ||
	     attr == &bus_attr_remove_single_major.attr))
		return 0;

534 535 536 537 538 539 540 541
	return attr->mode;
}

static const struct attribute_group rbd_bus_group = {
	.attrs = rbd_bus_attrs,
	.is_visible = rbd_bus_is_visible,
};
__ATTRIBUTE_GROUPS(rbd_bus);
A
Alex Elder 已提交
542 543 544

static struct bus_type rbd_bus_type = {
	.name		= "rbd",
545
	.bus_groups	= rbd_bus_groups,
A
Alex Elder 已提交
546 547 548 549 550 551 552 553 554 555 556
};

static void rbd_root_dev_release(struct device *dev)
{
}

static struct device rbd_root_dev = {
	.init_name =    "rbd",
	.release =      rbd_root_dev_release,
};

A
Alex Elder 已提交
557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583
static __printf(2, 3)
void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...)
{
	struct va_format vaf;
	va_list args;

	va_start(args, fmt);
	vaf.fmt = fmt;
	vaf.va = &args;

	if (!rbd_dev)
		printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf);
	else if (rbd_dev->disk)
		printk(KERN_WARNING "%s: %s: %pV\n",
			RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf);
	else if (rbd_dev->spec && rbd_dev->spec->image_name)
		printk(KERN_WARNING "%s: image %s: %pV\n",
			RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf);
	else if (rbd_dev->spec && rbd_dev->spec->image_id)
		printk(KERN_WARNING "%s: id %s: %pV\n",
			RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf);
	else	/* punt */
		printk(KERN_WARNING "%s: rbd_dev %p: %pV\n",
			RBD_DRV_NAME, rbd_dev, &vaf);
	va_end(args);
}

A
Alex Elder 已提交
584 585 586 587 588 589 590 591 592 593 594 595
#ifdef RBD_DEBUG
#define rbd_assert(expr)						\
		if (unlikely(!(expr))) {				\
			printk(KERN_ERR "\nAssertion failure in %s() "	\
						"at line %d:\n\n"	\
					"\trbd_assert(%s);\n\n",	\
					__func__, __LINE__, #expr);	\
			BUG();						\
		}
#else /* !RBD_DEBUG */
#  define rbd_assert(expr)	((void) 0)
#endif /* !RBD_DEBUG */
596

597
static void rbd_dev_remove_parent(struct rbd_device *rbd_dev);
A
Alex Elder 已提交
598

A
Alex Elder 已提交
599
static int rbd_dev_refresh(struct rbd_device *rbd_dev);
600
static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev);
601
static int rbd_dev_header_info(struct rbd_device *rbd_dev);
602
static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev);
603 604
static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
					u64 snap_id);
605 606 607 608
static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
				u8 *order, u64 *snap_size);
static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
		u64 *snap_features);
609

610 611
static int rbd_open(struct block_device *bdev, fmode_t mode)
{
A
Alex Elder 已提交
612
	struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
613
	bool removing = false;
614

615
	spin_lock_irq(&rbd_dev->lock);
616 617 618 619
	if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags))
		removing = true;
	else
		rbd_dev->open_count++;
620
	spin_unlock_irq(&rbd_dev->lock);
621 622 623
	if (removing)
		return -ENOENT;

A
Alex Elder 已提交
624
	(void) get_device(&rbd_dev->dev);
625

626 627 628
	return 0;
}

629
static void rbd_release(struct gendisk *disk, fmode_t mode)
630 631
{
	struct rbd_device *rbd_dev = disk->private_data;
632 633
	unsigned long open_count_before;

634
	spin_lock_irq(&rbd_dev->lock);
635
	open_count_before = rbd_dev->open_count--;
636
	spin_unlock_irq(&rbd_dev->lock);
637
	rbd_assert(open_count_before > 0);
638

A
Alex Elder 已提交
639
	put_device(&rbd_dev->dev);
640 641
}

G
Guangliang Zhao 已提交
642 643
static int rbd_ioctl_set_ro(struct rbd_device *rbd_dev, unsigned long arg)
{
644
	int ro;
G
Guangliang Zhao 已提交
645

646
	if (get_user(ro, (int __user *)arg))
G
Guangliang Zhao 已提交
647 648
		return -EFAULT;

649
	/* Snapshots can't be marked read-write */
G
Guangliang Zhao 已提交
650 651 652
	if (rbd_dev->spec->snap_id != CEPH_NOSNAP && !ro)
		return -EROFS;

653 654
	/* Let blkdev_roset() handle it */
	return -ENOTTY;
G
Guangliang Zhao 已提交
655 656 657 658 659 660
}

static int rbd_ioctl(struct block_device *bdev, fmode_t mode,
			unsigned int cmd, unsigned long arg)
{
	struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
661
	int ret;
G
Guangliang Zhao 已提交
662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681

	switch (cmd) {
	case BLKROSET:
		ret = rbd_ioctl_set_ro(rbd_dev, arg);
		break;
	default:
		ret = -ENOTTY;
	}

	return ret;
}

#ifdef CONFIG_COMPAT
static int rbd_compat_ioctl(struct block_device *bdev, fmode_t mode,
				unsigned int cmd, unsigned long arg)
{
	return rbd_ioctl(bdev, mode, cmd, arg);
}
#endif /* CONFIG_COMPAT */

682 683 684
static const struct block_device_operations rbd_bd_ops = {
	.owner			= THIS_MODULE,
	.open			= rbd_open,
685
	.release		= rbd_release,
G
Guangliang Zhao 已提交
686 687 688 689
	.ioctl			= rbd_ioctl,
#ifdef CONFIG_COMPAT
	.compat_ioctl		= rbd_compat_ioctl,
#endif
690 691 692
};

/*
693
 * Initialize an rbd client instance.  Success or not, this function
694
 * consumes ceph_opts.  Caller holds client_mutex.
695
 */
696
static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts)
697 698 699 700
{
	struct rbd_client *rbdc;
	int ret = -ENOMEM;

A
Alex Elder 已提交
701
	dout("%s:\n", __func__);
702 703 704 705 706 707 708
	rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL);
	if (!rbdc)
		goto out_opt;

	kref_init(&rbdc->kref);
	INIT_LIST_HEAD(&rbdc->node);

709
	rbdc->client = ceph_create_client(ceph_opts, rbdc);
710
	if (IS_ERR(rbdc->client))
711
		goto out_rbdc;
A
Alex Elder 已提交
712
	ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */
713 714 715

	ret = ceph_open_session(rbdc->client);
	if (ret < 0)
716
		goto out_client;
717

A
Alex Elder 已提交
718
	spin_lock(&rbd_client_list_lock);
719
	list_add_tail(&rbdc->node, &rbd_client_list);
A
Alex Elder 已提交
720
	spin_unlock(&rbd_client_list_lock);
721

A
Alex Elder 已提交
722
	dout("%s: rbdc %p\n", __func__, rbdc);
723

724
	return rbdc;
725
out_client:
726
	ceph_destroy_client(rbdc->client);
727
out_rbdc:
728 729
	kfree(rbdc);
out_opt:
A
Alex Elder 已提交
730 731
	if (ceph_opts)
		ceph_destroy_options(ceph_opts);
A
Alex Elder 已提交
732 733
	dout("%s: error %d\n", __func__, ret);

V
Vasiliy Kulikov 已提交
734
	return ERR_PTR(ret);
735 736
}

737 738 739 740 741 742 743
static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc)
{
	kref_get(&rbdc->kref);

	return rbdc;
}

744
/*
745 746
 * Find a ceph client with specific addr and configuration.  If
 * found, bump its reference count.
747
 */
748
static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts)
749 750
{
	struct rbd_client *client_node;
751
	bool found = false;
752

A
Alex Elder 已提交
753
	if (ceph_opts->flags & CEPH_OPT_NOSHARE)
754 755
		return NULL;

756 757 758
	spin_lock(&rbd_client_list_lock);
	list_for_each_entry(client_node, &rbd_client_list, node) {
		if (!ceph_compare_options(ceph_opts, client_node->client)) {
759 760
			__rbd_get_client(client_node);

761 762 763 764 765 766 767
			found = true;
			break;
		}
	}
	spin_unlock(&rbd_client_list_lock);

	return found ? client_node : NULL;
768 769
}

770
/*
771
 * (Per device) rbd map options
772 773
 */
enum {
I
Ilya Dryomov 已提交
774
	Opt_queue_depth,
775 776 777 778
	Opt_last_int,
	/* int args above */
	Opt_last_string,
	/* string args above */
A
Alex Elder 已提交
779 780
	Opt_read_only,
	Opt_read_write,
I
Ilya Dryomov 已提交
781
	Opt_lock_on_read,
I
Ilya Dryomov 已提交
782
	Opt_exclusive,
783
	Opt_err
784 785
};

A
Alex Elder 已提交
786
static match_table_t rbd_opts_tokens = {
I
Ilya Dryomov 已提交
787
	{Opt_queue_depth, "queue_depth=%d"},
788 789
	/* int args above */
	/* string args above */
A
Alex Elder 已提交
790
	{Opt_read_only, "read_only"},
A
Alex Elder 已提交
791 792 793
	{Opt_read_only, "ro"},		/* Alternate spelling */
	{Opt_read_write, "read_write"},
	{Opt_read_write, "rw"},		/* Alternate spelling */
I
Ilya Dryomov 已提交
794
	{Opt_lock_on_read, "lock_on_read"},
I
Ilya Dryomov 已提交
795
	{Opt_exclusive, "exclusive"},
796
	{Opt_err, NULL}
797 798
};

A
Alex Elder 已提交
799
struct rbd_options {
I
Ilya Dryomov 已提交
800
	int	queue_depth;
A
Alex Elder 已提交
801
	bool	read_only;
I
Ilya Dryomov 已提交
802
	bool	lock_on_read;
I
Ilya Dryomov 已提交
803
	bool	exclusive;
A
Alex Elder 已提交
804 805
};

I
Ilya Dryomov 已提交
806
#define RBD_QUEUE_DEPTH_DEFAULT	BLKDEV_MAX_RQ
A
Alex Elder 已提交
807
#define RBD_READ_ONLY_DEFAULT	false
I
Ilya Dryomov 已提交
808
#define RBD_LOCK_ON_READ_DEFAULT false
I
Ilya Dryomov 已提交
809
#define RBD_EXCLUSIVE_DEFAULT	false
A
Alex Elder 已提交
810

811 812
static int parse_rbd_opts_token(char *c, void *private)
{
A
Alex Elder 已提交
813
	struct rbd_options *rbd_opts = private;
814 815 816
	substring_t argstr[MAX_OPT_ARGS];
	int token, intval, ret;

A
Alex Elder 已提交
817
	token = match_token(c, rbd_opts_tokens, argstr);
818 819 820
	if (token < Opt_last_int) {
		ret = match_int(&argstr[0], &intval);
		if (ret < 0) {
821
			pr_err("bad mount option arg (not int) at '%s'\n", c);
822 823 824 825
			return ret;
		}
		dout("got int token %d val %d\n", token, intval);
	} else if (token > Opt_last_int && token < Opt_last_string) {
826
		dout("got string token %d val %s\n", token, argstr[0].from);
827 828 829 830 831
	} else {
		dout("got token %d\n", token);
	}

	switch (token) {
I
Ilya Dryomov 已提交
832 833 834 835 836 837 838
	case Opt_queue_depth:
		if (intval < 1) {
			pr_err("queue_depth out of range\n");
			return -EINVAL;
		}
		rbd_opts->queue_depth = intval;
		break;
A
Alex Elder 已提交
839 840 841 842 843 844
	case Opt_read_only:
		rbd_opts->read_only = true;
		break;
	case Opt_read_write:
		rbd_opts->read_only = false;
		break;
I
Ilya Dryomov 已提交
845 846 847
	case Opt_lock_on_read:
		rbd_opts->lock_on_read = true;
		break;
I
Ilya Dryomov 已提交
848 849 850
	case Opt_exclusive:
		rbd_opts->exclusive = true;
		break;
851
	default:
852 853
		/* libceph prints "bad option" msg */
		return -EINVAL;
854
	}
855

856 857 858
	return 0;
}

G
Guangliang Zhao 已提交
859 860 861 862 863 864 865
static char* obj_op_name(enum obj_operation_type op_type)
{
	switch (op_type) {
	case OBJ_OP_READ:
		return "read";
	case OBJ_OP_WRITE:
		return "write";
866 867
	case OBJ_OP_DISCARD:
		return "discard";
G
Guangliang Zhao 已提交
868 869 870 871 872
	default:
		return "???";
	}
}

873 874
/*
 * Get a ceph client with specific addr and configuration, if one does
875 876
 * not exist create it.  Either way, ceph_opts is consumed by this
 * function.
877
 */
878
static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts)
879
{
880
	struct rbd_client *rbdc;
881

882
	mutex_lock_nested(&client_mutex, SINGLE_DEPTH_NESTING);
883
	rbdc = rbd_client_find(ceph_opts);
884
	if (rbdc)	/* using an existing client */
A
Alex Elder 已提交
885
		ceph_destroy_options(ceph_opts);
886
	else
887
		rbdc = rbd_client_create(ceph_opts);
888
	mutex_unlock(&client_mutex);
889

890
	return rbdc;
891 892 893 894
}

/*
 * Destroy ceph client
A
Alex Elder 已提交
895
 *
A
Alex Elder 已提交
896
 * Caller must hold rbd_client_list_lock.
897 898 899 900 901
 */
static void rbd_client_release(struct kref *kref)
{
	struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref);

A
Alex Elder 已提交
902
	dout("%s: rbdc %p\n", __func__, rbdc);
903
	spin_lock(&rbd_client_list_lock);
904
	list_del(&rbdc->node);
905
	spin_unlock(&rbd_client_list_lock);
906 907 908 909 910 911 912 913 914

	ceph_destroy_client(rbdc->client);
	kfree(rbdc);
}

/*
 * Drop reference to ceph client node. If it's not referenced anymore, release
 * it.
 */
915
static void rbd_put_client(struct rbd_client *rbdc)
916
{
917 918
	if (rbdc)
		kref_put(&rbdc->kref, rbd_client_release);
919 920
}

921 922 923 924 925
static bool rbd_image_format_valid(u32 image_format)
{
	return image_format == 1 || image_format == 2;
}

926 927
static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk)
{
928 929 930 931 932 933 934
	size_t size;
	u32 snap_count;

	/* The header has to start with the magic rbd header text */
	if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT)))
		return false;

A
Alex Elder 已提交
935 936 937 938 939 940 941 942 943 944
	/* The bio layer requires at least sector-sized I/O */

	if (ondisk->options.order < SECTOR_SHIFT)
		return false;

	/* If we use u64 in a few spots we may be able to loosen this */

	if (ondisk->options.order > 8 * sizeof (int) - 1)
		return false;

945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962
	/*
	 * The size of a snapshot header has to fit in a size_t, and
	 * that limits the number of snapshots.
	 */
	snap_count = le32_to_cpu(ondisk->snap_count);
	size = SIZE_MAX - sizeof (struct ceph_snap_context);
	if (snap_count > size / sizeof (__le64))
		return false;

	/*
	 * Not only that, but the size of the entire the snapshot
	 * header must also be representable in a size_t.
	 */
	size -= snap_count * sizeof (__le64);
	if ((u64) size < le64_to_cpu(ondisk->snap_names_len))
		return false;

	return true;
963 964
}

I
Ilya Dryomov 已提交
965 966 967 968 969 970 971 972
/*
 * returns the size of an object in the image
 */
static u32 rbd_obj_bytes(struct rbd_image_header *header)
{
	return 1U << header->obj_order;
}

I
Ilya Dryomov 已提交
973 974 975 976 977 978 979 980 981 982 983
static void rbd_init_layout(struct rbd_device *rbd_dev)
{
	if (rbd_dev->header.stripe_unit == 0 ||
	    rbd_dev->header.stripe_count == 0) {
		rbd_dev->header.stripe_unit = rbd_obj_bytes(&rbd_dev->header);
		rbd_dev->header.stripe_count = 1;
	}

	rbd_dev->layout.stripe_unit = rbd_dev->header.stripe_unit;
	rbd_dev->layout.stripe_count = rbd_dev->header.stripe_count;
	rbd_dev->layout.object_size = rbd_obj_bytes(&rbd_dev->header);
984 985
	rbd_dev->layout.pool_id = rbd_dev->header.data_pool_id == CEPH_NOPOOL ?
			  rbd_dev->spec->pool_id : rbd_dev->header.data_pool_id;
I
Ilya Dryomov 已提交
986 987 988
	RCU_INIT_POINTER(rbd_dev->layout.pool_ns, NULL);
}

989
/*
990 991
 * Fill an rbd image header with information from the given format 1
 * on-disk header.
992
 */
A
Alex Elder 已提交
993
static int rbd_header_from_disk(struct rbd_device *rbd_dev,
994
				 struct rbd_image_header_ondisk *ondisk)
995
{
A
Alex Elder 已提交
996
	struct rbd_image_header *header = &rbd_dev->header;
997 998 999 1000 1001
	bool first_time = header->object_prefix == NULL;
	struct ceph_snap_context *snapc;
	char *object_prefix = NULL;
	char *snap_names = NULL;
	u64 *snap_sizes = NULL;
1002
	u32 snap_count;
1003
	int ret = -ENOMEM;
1004
	u32 i;
1005

1006
	/* Allocate this now to avoid having to handle failure below */
A
Alex Elder 已提交
1007

1008
	if (first_time) {
1009 1010 1011
		object_prefix = kstrndup(ondisk->object_prefix,
					 sizeof(ondisk->object_prefix),
					 GFP_KERNEL);
1012 1013 1014
		if (!object_prefix)
			return -ENOMEM;
	}
A
Alex Elder 已提交
1015

1016
	/* Allocate the snapshot context and fill it in */
A
Alex Elder 已提交
1017

1018 1019 1020 1021 1022
	snap_count = le32_to_cpu(ondisk->snap_count);
	snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
	if (!snapc)
		goto out_err;
	snapc->seq = le64_to_cpu(ondisk->snap_seq);
1023
	if (snap_count) {
1024
		struct rbd_image_snap_ondisk *snaps;
A
Alex Elder 已提交
1025 1026
		u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len);

1027
		/* We'll keep a copy of the snapshot names... */
1028

1029 1030 1031 1032
		if (snap_names_len > (u64)SIZE_MAX)
			goto out_2big;
		snap_names = kmalloc(snap_names_len, GFP_KERNEL);
		if (!snap_names)
A
Alex Elder 已提交
1033 1034
			goto out_err;

1035
		/* ...as well as the array of their sizes. */
1036 1037 1038
		snap_sizes = kmalloc_array(snap_count,
					   sizeof(*header->snap_sizes),
					   GFP_KERNEL);
1039
		if (!snap_sizes)
A
Alex Elder 已提交
1040
			goto out_err;
1041

A
Alex Elder 已提交
1042
		/*
1043 1044 1045
		 * Copy the names, and fill in each snapshot's id
		 * and size.
		 *
1046
		 * Note that rbd_dev_v1_header_info() guarantees the
1047
		 * ondisk buffer we're working with has
A
Alex Elder 已提交
1048 1049 1050
		 * snap_names_len bytes beyond the end of the
		 * snapshot id array, this memcpy() is safe.
		 */
1051 1052 1053 1054 1055 1056
		memcpy(snap_names, &ondisk->snaps[snap_count], snap_names_len);
		snaps = ondisk->snaps;
		for (i = 0; i < snap_count; i++) {
			snapc->snaps[i] = le64_to_cpu(snaps[i].id);
			snap_sizes[i] = le64_to_cpu(snaps[i].image_size);
		}
1057
	}
A
Alex Elder 已提交
1058

1059
	/* We won't fail any more, fill in the header */
1060

1061 1062 1063
	if (first_time) {
		header->object_prefix = object_prefix;
		header->obj_order = ondisk->options.order;
I
Ilya Dryomov 已提交
1064
		rbd_init_layout(rbd_dev);
1065
	} else {
A
Alex Elder 已提交
1066 1067 1068
		ceph_put_snap_context(header->snapc);
		kfree(header->snap_names);
		kfree(header->snap_sizes);
1069
	}
1070

1071
	/* The remaining fields always get updated (when we refresh) */
1072

A
Alex Elder 已提交
1073
	header->image_size = le64_to_cpu(ondisk->image_size);
1074 1075 1076
	header->snapc = snapc;
	header->snap_names = snap_names;
	header->snap_sizes = snap_sizes;
1077

1078
	return 0;
1079 1080
out_2big:
	ret = -EIO;
A
Alex Elder 已提交
1081
out_err:
1082 1083 1084 1085
	kfree(snap_sizes);
	kfree(snap_names);
	ceph_put_snap_context(snapc);
	kfree(object_prefix);
1086

1087
	return ret;
1088 1089
}

1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104
static const char *_rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, u32 which)
{
	const char *snap_name;

	rbd_assert(which < rbd_dev->header.snapc->num_snaps);

	/* Skip over names until we find the one we are looking for */

	snap_name = rbd_dev->header.snap_names;
	while (which--)
		snap_name += strlen(snap_name) + 1;

	return kstrdup(snap_name, GFP_KERNEL);
}

1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128
/*
 * Snapshot id comparison function for use with qsort()/bsearch().
 * Note that result is for snapshots in *descending* order.
 */
static int snapid_compare_reverse(const void *s1, const void *s2)
{
	u64 snap_id1 = *(u64 *)s1;
	u64 snap_id2 = *(u64 *)s2;

	if (snap_id1 < snap_id2)
		return 1;
	return snap_id1 == snap_id2 ? 0 : -1;
}

/*
 * Search a snapshot context to see if the given snapshot id is
 * present.
 *
 * Returns the position of the snapshot id in the array if it's found,
 * or BAD_SNAP_INDEX otherwise.
 *
 * Note: The snapshot array is in kept sorted (by the osd) in
 * reverse order, highest snapshot id first.
 */
1129 1130 1131
static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id)
{
	struct ceph_snap_context *snapc = rbd_dev->header.snapc;
1132
	u64 *found;
1133

1134 1135
	found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps,
				sizeof (snap_id), snapid_compare_reverse);
1136

1137
	return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX;
1138 1139
}

1140 1141
static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev,
					u64 snap_id)
1142
{
1143
	u32 which;
1144
	const char *snap_name;
1145

1146 1147
	which = rbd_dev_snap_index(rbd_dev, snap_id);
	if (which == BAD_SNAP_INDEX)
1148
		return ERR_PTR(-ENOENT);
1149

1150 1151
	snap_name = _rbd_dev_v1_snap_name(rbd_dev, which);
	return snap_name ? snap_name : ERR_PTR(-ENOMEM);
1152 1153 1154 1155
}

static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id)
{
1156 1157 1158
	if (snap_id == CEPH_NOSNAP)
		return RBD_SNAP_HEAD_NAME;

1159 1160 1161
	rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
	if (rbd_dev->image_format == 1)
		return rbd_dev_v1_snap_name(rbd_dev, snap_id);
1162

1163
	return rbd_dev_v2_snap_name(rbd_dev, snap_id);
1164 1165
}

1166 1167
static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
				u64 *snap_size)
1168
{
1169 1170 1171 1172 1173
	rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
	if (snap_id == CEPH_NOSNAP) {
		*snap_size = rbd_dev->header.image_size;
	} else if (rbd_dev->image_format == 1) {
		u32 which;
1174

1175 1176 1177
		which = rbd_dev_snap_index(rbd_dev, snap_id);
		if (which == BAD_SNAP_INDEX)
			return -ENOENT;
1178

1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190
		*snap_size = rbd_dev->header.snap_sizes[which];
	} else {
		u64 size = 0;
		int ret;

		ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size);
		if (ret)
			return ret;

		*snap_size = size;
	}
	return 0;
1191 1192
}

1193 1194
static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
			u64 *snap_features)
1195
{
1196 1197 1198 1199 1200
	rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
	if (snap_id == CEPH_NOSNAP) {
		*snap_features = rbd_dev->header.features;
	} else if (rbd_dev->image_format == 1) {
		*snap_features = 0;	/* No features for format 1 */
1201
	} else {
1202 1203
		u64 features = 0;
		int ret;
1204

1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215
		ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features);
		if (ret)
			return ret;

		*snap_features = features;
	}
	return 0;
}

static int rbd_dev_mapping_set(struct rbd_device *rbd_dev)
{
1216
	u64 snap_id = rbd_dev->spec->snap_id;
1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230
	u64 size = 0;
	u64 features = 0;
	int ret;

	ret = rbd_snap_size(rbd_dev, snap_id, &size);
	if (ret)
		return ret;
	ret = rbd_snap_features(rbd_dev, snap_id, &features);
	if (ret)
		return ret;

	rbd_dev->mapping.size = size;
	rbd_dev->mapping.features = features;

1231
	return 0;
1232 1233
}

A
Alex Elder 已提交
1234 1235 1236 1237
static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev)
{
	rbd_dev->mapping.size = 0;
	rbd_dev->mapping.features = 0;
1238 1239
}

A
Alex Elder 已提交
1240 1241
static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset)
{
I
Ilya Dryomov 已提交
1242
	u64 segment_size = rbd_obj_bytes(&rbd_dev->header);
1243

A
Alex Elder 已提交
1244 1245 1246 1247 1248 1249
	return offset & (segment_size - 1);
}

static u64 rbd_segment_length(struct rbd_device *rbd_dev,
				u64 offset, u64 length)
{
I
Ilya Dryomov 已提交
1250
	u64 segment_size = rbd_obj_bytes(&rbd_dev->header);
A
Alex Elder 已提交
1251 1252 1253

	offset &= segment_size - 1;

A
Alex Elder 已提交
1254
	rbd_assert(length <= U64_MAX - offset);
A
Alex Elder 已提交
1255 1256 1257 1258
	if (offset + length > segment_size)
		length = segment_size - offset;

	return length;
1259 1260
}

1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281
static void zero_bvec(struct bio_vec *bv)
{
	void *buf;
	unsigned long flags;

	buf = bvec_kmap_irq(bv, &flags);
	memset(buf, 0, bv->bv_len);
	flush_dcache_page(bv->bv_page);
	bvec_kunmap_irq(buf, &flags);
}

static void zero_bios(struct ceph_bio_iter *bio_pos, u32 off, u32 bytes)
{
	struct ceph_bio_iter it = *bio_pos;

	ceph_bio_iter_advance(&it, off);
	ceph_bio_iter_advance_step(&it, bytes, ({
		zero_bvec(&bv);
	}));
}

1282 1283 1284 1285 1286 1287 1288 1289
static void zero_bvecs(struct ceph_bvec_iter *bvec_pos, u32 off, u32 bytes)
{
	struct ceph_bvec_iter it = *bvec_pos;

	ceph_bvec_iter_advance(&it, off);
	ceph_bvec_iter_advance_step(&it, bytes, ({
		zero_bvec(&bv);
	}));
A
Alex Elder 已提交
1290 1291
}

I
Ilya Dryomov 已提交
1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312
/*
 * Zero a range in @obj_req data buffer defined by a bio (list) or
 * bio_vec array.
 *
 * @off is relative to the start of the data buffer.
 */
static void rbd_obj_zero_range(struct rbd_obj_request *obj_req, u32 off,
			       u32 bytes)
{
	switch (obj_req->type) {
	case OBJ_REQUEST_BIO:
		zero_bios(&obj_req->bio_pos, off, bytes);
		break;
	case OBJ_REQUEST_BVECS:
		zero_bvecs(&obj_req->bvec_pos, off, bytes);
		break;
	default:
		rbd_assert(0);
	}
}

1313 1314 1315 1316 1317
/*
 * The default/initial value for all object request flags is 0.  For
 * each flag, once its value is set to 1 it is never reset to 0
 * again.
 */
A
Alex Elder 已提交
1318
static void obj_request_img_data_set(struct rbd_obj_request *obj_request)
1319
{
A
Alex Elder 已提交
1320
	if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) {
1321 1322
		struct rbd_device *rbd_dev;

A
Alex Elder 已提交
1323
		rbd_dev = obj_request->img_request->rbd_dev;
1324
		rbd_warn(rbd_dev, "obj_request %p already marked img_data",
1325 1326 1327 1328
			obj_request);
	}
}

A
Alex Elder 已提交
1329
static bool obj_request_img_data_test(struct rbd_obj_request *obj_request)
1330 1331
{
	smp_mb();
A
Alex Elder 已提交
1332
	return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0;
1333 1334
}

A
Alex Elder 已提交
1335
static void obj_request_done_set(struct rbd_obj_request *obj_request)
1336
{
A
Alex Elder 已提交
1337 1338
	if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) {
		struct rbd_device *rbd_dev = NULL;
1339

A
Alex Elder 已提交
1340 1341
		if (obj_request_img_data_test(obj_request))
			rbd_dev = obj_request->img_request->rbd_dev;
1342
		rbd_warn(rbd_dev, "obj_request %p already marked done",
1343 1344 1345 1346
			obj_request);
	}
}

A
Alex Elder 已提交
1347
static bool obj_request_done_test(struct rbd_obj_request *obj_request)
1348 1349
{
	smp_mb();
A
Alex Elder 已提交
1350
	return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0;
1351 1352
}

1353 1354 1355 1356 1357 1358 1359 1360
static bool obj_request_overlaps_parent(struct rbd_obj_request *obj_request)
{
	struct rbd_device *rbd_dev = obj_request->img_request->rbd_dev;

	return obj_request->img_offset <
	    round_up(rbd_dev->parent_overlap, rbd_obj_bytes(&rbd_dev->header));
}

A
Alex Elder 已提交
1361 1362
static void rbd_obj_request_get(struct rbd_obj_request *obj_request)
{
A
Alex Elder 已提交
1363
	dout("%s: obj %p (was %d)\n", __func__, obj_request,
1364
		kref_read(&obj_request->kref));
A
Alex Elder 已提交
1365 1366 1367 1368 1369 1370 1371
	kref_get(&obj_request->kref);
}

static void rbd_obj_request_destroy(struct kref *kref);
static void rbd_obj_request_put(struct rbd_obj_request *obj_request)
{
	rbd_assert(obj_request != NULL);
A
Alex Elder 已提交
1372
	dout("%s: obj %p (was %d)\n", __func__, obj_request,
1373
		kref_read(&obj_request->kref));
A
Alex Elder 已提交
1374 1375 1376
	kref_put(&obj_request->kref, rbd_obj_request_destroy);
}

1377 1378 1379
static void rbd_img_request_get(struct rbd_img_request *img_request)
{
	dout("%s: img %p (was %d)\n", __func__, img_request,
1380
	     kref_read(&img_request->kref));
1381 1382 1383
	kref_get(&img_request->kref);
}

1384 1385
static bool img_request_child_test(struct rbd_img_request *img_request);
static void rbd_parent_request_destroy(struct kref *kref);
A
Alex Elder 已提交
1386 1387 1388 1389
static void rbd_img_request_destroy(struct kref *kref);
static void rbd_img_request_put(struct rbd_img_request *img_request)
{
	rbd_assert(img_request != NULL);
A
Alex Elder 已提交
1390
	dout("%s: img %p (was %d)\n", __func__, img_request,
1391
		kref_read(&img_request->kref));
1392 1393 1394 1395
	if (img_request_child_test(img_request))
		kref_put(&img_request->kref, rbd_parent_request_destroy);
	else
		kref_put(&img_request->kref, rbd_img_request_destroy);
A
Alex Elder 已提交
1396 1397 1398 1399 1400
}

static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request,
					struct rbd_obj_request *obj_request)
{
1401 1402
	rbd_assert(obj_request->img_request == NULL);

1403
	/* Image request now owns object's original reference */
A
Alex Elder 已提交
1404
	obj_request->img_request = img_request;
1405
	obj_request->which = img_request->obj_request_count;
1406 1407
	rbd_assert(!obj_request_img_data_test(obj_request));
	obj_request_img_data_set(obj_request);
A
Alex Elder 已提交
1408
	rbd_assert(obj_request->which != BAD_WHICH);
1409 1410
	img_request->obj_request_count++;
	list_add_tail(&obj_request->links, &img_request->obj_requests);
A
Alex Elder 已提交
1411 1412
	dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
		obj_request->which);
A
Alex Elder 已提交
1413 1414 1415 1416 1417 1418
}

static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request,
					struct rbd_obj_request *obj_request)
{
	rbd_assert(obj_request->which != BAD_WHICH);
1419

A
Alex Elder 已提交
1420 1421
	dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
		obj_request->which);
A
Alex Elder 已提交
1422
	list_del(&obj_request->links);
1423 1424 1425 1426
	rbd_assert(img_request->obj_request_count > 0);
	img_request->obj_request_count--;
	rbd_assert(obj_request->which == img_request->obj_request_count);
	obj_request->which = BAD_WHICH;
1427
	rbd_assert(obj_request_img_data_test(obj_request));
A
Alex Elder 已提交
1428 1429
	rbd_assert(obj_request->img_request == img_request);
	obj_request->img_request = NULL;
1430
	obj_request->callback = NULL;
A
Alex Elder 已提交
1431 1432 1433 1434 1435 1436
	rbd_obj_request_put(obj_request);
}

static bool obj_request_type_valid(enum obj_request_type type)
{
	switch (type) {
1437
	case OBJ_REQUEST_NODATA:
A
Alex Elder 已提交
1438
	case OBJ_REQUEST_BIO:
1439
	case OBJ_REQUEST_BVECS:
A
Alex Elder 已提交
1440 1441 1442 1443 1444 1445
		return true;
	default:
		return false;
	}
}

1446 1447
static void rbd_img_obj_callback(struct rbd_obj_request *obj_request);

1448
static void rbd_obj_request_submit(struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1449
{
1450 1451
	struct ceph_osd_request *osd_req = obj_request->osd_req;

1452 1453
	dout("%s %p object_no %016llx %llu~%llu osd_req %p\n", __func__,
	     obj_request, obj_request->object_no, obj_request->offset,
1454
	     obj_request->length, osd_req);
1455 1456 1457 1458
	if (obj_request_img_data_test(obj_request)) {
		WARN_ON(obj_request->callback != rbd_img_obj_callback);
		rbd_img_request_get(obj_request->img_request);
	}
1459
	ceph_osdc_start_request(osd_req->r_osdc, osd_req, false);
A
Alex Elder 已提交
1460 1461 1462 1463
}

static void rbd_img_request_complete(struct rbd_img_request *img_request)
{
1464

A
Alex Elder 已提交
1465
	dout("%s: img %p\n", __func__, img_request);
1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481

	/*
	 * If no error occurred, compute the aggregate transfer
	 * count for the image request.  We could instead use
	 * atomic64_cmpxchg() to update it as each object request
	 * completes; not clear which way is better off hand.
	 */
	if (!img_request->result) {
		struct rbd_obj_request *obj_request;
		u64 xferred = 0;

		for_each_obj_request(img_request, obj_request)
			xferred += obj_request->xferred;
		img_request->xferred = xferred;
	}

A
Alex Elder 已提交
1482 1483 1484 1485 1486 1487
	if (img_request->callback)
		img_request->callback(img_request);
	else
		rbd_img_request_put(img_request);
}

A
Alex Elder 已提交
1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504
/*
 * The default/initial value for all image request flags is 0.  Each
 * is conditionally set to 1 at image request initialization time
 * and currently never change thereafter.
 */
static void img_request_write_set(struct rbd_img_request *img_request)
{
	set_bit(IMG_REQ_WRITE, &img_request->flags);
	smp_mb();
}

static bool img_request_write_test(struct rbd_img_request *img_request)
{
	smp_mb();
	return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0;
}

1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519
/*
 * Set the discard flag when the img_request is an discard request
 */
static void img_request_discard_set(struct rbd_img_request *img_request)
{
	set_bit(IMG_REQ_DISCARD, &img_request->flags);
	smp_mb();
}

static bool img_request_discard_test(struct rbd_img_request *img_request)
{
	smp_mb();
	return test_bit(IMG_REQ_DISCARD, &img_request->flags) != 0;
}

1520 1521 1522 1523 1524 1525
static void img_request_child_set(struct rbd_img_request *img_request)
{
	set_bit(IMG_REQ_CHILD, &img_request->flags);
	smp_mb();
}

1526 1527 1528 1529 1530 1531
static void img_request_child_clear(struct rbd_img_request *img_request)
{
	clear_bit(IMG_REQ_CHILD, &img_request->flags);
	smp_mb();
}

1532 1533 1534 1535 1536 1537
static bool img_request_child_test(struct rbd_img_request *img_request)
{
	smp_mb();
	return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0;
}

1538 1539 1540 1541 1542 1543
static void img_request_layered_set(struct rbd_img_request *img_request)
{
	set_bit(IMG_REQ_LAYERED, &img_request->flags);
	smp_mb();
}

1544 1545 1546 1547 1548 1549
static void img_request_layered_clear(struct rbd_img_request *img_request)
{
	clear_bit(IMG_REQ_LAYERED, &img_request->flags);
	smp_mb();
}

1550 1551 1552 1553 1554 1555
static bool img_request_layered_test(struct rbd_img_request *img_request)
{
	smp_mb();
	return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0;
}

1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566
static enum obj_operation_type
rbd_img_request_op_type(struct rbd_img_request *img_request)
{
	if (img_request_write_test(img_request))
		return OBJ_OP_WRITE;
	else if (img_request_discard_test(img_request))
		return OBJ_OP_DISCARD;
	else
		return OBJ_OP_READ;
}

I
Ilya Dryomov 已提交
1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595
static bool rbd_obj_is_entire(struct rbd_obj_request *obj_req)
{
	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;

	return !obj_req->offset &&
	       obj_req->length == rbd_dev->layout.object_size;
}

static bool rbd_obj_is_tail(struct rbd_obj_request *obj_req)
{
	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;

	return obj_req->offset + obj_req->length ==
					rbd_dev->layout.object_size;
}

static bool rbd_img_is_write(struct rbd_img_request *img_req)
{
	switch (rbd_img_request_op_type(img_req)) {
	case OBJ_OP_READ:
		return false;
	case OBJ_OP_WRITE:
	case OBJ_OP_DISCARD:
		return true;
	default:
		rbd_assert(0);
	}
}

A
Alex Elder 已提交
1596 1597
static void rbd_obj_request_complete(struct rbd_obj_request *obj_request)
{
A
Alex Elder 已提交
1598 1599
	dout("%s: obj %p cb %p\n", __func__, obj_request,
		obj_request->callback);
1600
	obj_request->callback(obj_request);
A
Alex Elder 已提交
1601 1602
}

I
Ilya Dryomov 已提交
1603 1604
static void rbd_obj_handle_request(struct rbd_obj_request *obj_req);

1605
static void rbd_osd_req_callback(struct ceph_osd_request *osd_req)
A
Alex Elder 已提交
1606
{
I
Ilya Dryomov 已提交
1607
	struct rbd_obj_request *obj_req = osd_req->r_priv;
A
Alex Elder 已提交
1608

I
Ilya Dryomov 已提交
1609 1610 1611
	dout("%s osd_req %p result %d for obj_req %p\n", __func__, osd_req,
	     osd_req->r_result, obj_req);
	rbd_assert(osd_req == obj_req->osd_req);
1612

I
Ilya Dryomov 已提交
1613 1614 1615 1616 1617 1618 1619 1620 1621 1622
	obj_req->result = osd_req->r_result < 0 ? osd_req->r_result : 0;
	if (!obj_req->result && !rbd_img_is_write(obj_req->img_request))
		obj_req->xferred = osd_req->r_result;
	else
		/*
		 * Writes aren't allowed to return a data payload.  In some
		 * guarded write cases (e.g. stat + zero on an empty object)
		 * a stat response makes it through, but we don't care.
		 */
		obj_req->xferred = 0;
A
Alex Elder 已提交
1623

I
Ilya Dryomov 已提交
1624
	rbd_obj_handle_request(obj_req);
A
Alex Elder 已提交
1625 1626
}

1627
static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1628
{
1629
	struct ceph_osd_request *osd_req = obj_request->osd_req;
A
Alex Elder 已提交
1630

1631 1632
	rbd_assert(obj_request_img_data_test(obj_request));
	osd_req->r_snapid = obj_request->img_request->snap_id;
1633 1634 1635 1636 1637 1638
}

static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request)
{
	struct ceph_osd_request *osd_req = obj_request->osd_req;

1639
	ktime_get_real_ts(&osd_req->r_mtime);
1640
	osd_req->r_data_offset = obj_request->offset;
A
Alex Elder 已提交
1641 1642
}

1643 1644 1645 1646 1647 1648 1649 1650
static struct ceph_osd_request *
__rbd_osd_req_create(struct rbd_device *rbd_dev,
		     struct ceph_snap_context *snapc,
		     int num_ops, unsigned int flags,
		     struct rbd_obj_request *obj_request)
{
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
	struct ceph_osd_request *req;
1651 1652
	const char *name_format = rbd_dev->image_format == 1 ?
				      RBD_V1_DATA_FORMAT : RBD_V2_DATA_FORMAT;
1653 1654 1655 1656 1657 1658 1659 1660 1661 1662

	req = ceph_osdc_alloc_request(osdc, snapc, num_ops, false, GFP_NOIO);
	if (!req)
		return NULL;

	req->r_flags = flags;
	req->r_callback = rbd_osd_req_callback;
	req->r_priv = obj_request;

	req->r_base_oloc.pool = rbd_dev->layout.pool_id;
1663 1664
	if (ceph_oid_aprintf(&req->r_base_oid, GFP_NOIO, name_format,
			rbd_dev->header.object_prefix, obj_request->object_no))
1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676
		goto err_req;

	if (ceph_osdc_alloc_messages(req, GFP_NOIO))
		goto err_req;

	return req;

err_req:
	ceph_osdc_put_request(req);
	return NULL;
}

A
Alex Elder 已提交
1677 1678
static struct ceph_osd_request *rbd_osd_req_create(
					struct rbd_device *rbd_dev,
G
Guangliang Zhao 已提交
1679
					enum obj_operation_type op_type,
1680
					unsigned int num_ops,
A
Alex Elder 已提交
1681
					struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1682 1683 1684
{
	struct ceph_snap_context *snapc = NULL;

1685 1686
	if (obj_request_img_data_test(obj_request) &&
		(op_type == OBJ_OP_DISCARD || op_type == OBJ_OP_WRITE)) {
1687
		struct rbd_img_request *img_request = obj_request->img_request;
1688 1689 1690 1691 1692
		if (op_type == OBJ_OP_WRITE) {
			rbd_assert(img_request_write_test(img_request));
		} else {
			rbd_assert(img_request_discard_test(img_request));
		}
G
Guangliang Zhao 已提交
1693
		snapc = img_request->snapc;
A
Alex Elder 已提交
1694 1695
	}

1696 1697
	return __rbd_osd_req_create(rbd_dev, snapc, num_ops,
	    (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD) ?
1698
	    CEPH_OSD_FLAG_WRITE : CEPH_OSD_FLAG_READ, obj_request);
A
Alex Elder 已提交
1699 1700 1701 1702 1703 1704 1705
}

static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req)
{
	ceph_osdc_put_request(osd_req);
}

1706 1707
static struct rbd_obj_request *
rbd_obj_request_create(enum obj_request_type type)
A
Alex Elder 已提交
1708 1709 1710 1711 1712
{
	struct rbd_obj_request *obj_request;

	rbd_assert(obj_request_type_valid(type));

1713
	obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_NOIO);
1714
	if (!obj_request)
1715 1716
		return NULL;

A
Alex Elder 已提交
1717 1718 1719 1720 1721
	obj_request->which = BAD_WHICH;
	obj_request->type = type;
	INIT_LIST_HEAD(&obj_request->links);
	kref_init(&obj_request->kref);

1722
	dout("%s %p\n", __func__, obj_request);
A
Alex Elder 已提交
1723 1724 1725 1726 1727 1728
	return obj_request;
}

static void rbd_obj_request_destroy(struct kref *kref)
{
	struct rbd_obj_request *obj_request;
1729
	u32 i;
A
Alex Elder 已提交
1730 1731 1732

	obj_request = container_of(kref, struct rbd_obj_request, kref);

A
Alex Elder 已提交
1733 1734
	dout("%s: obj %p\n", __func__, obj_request);

A
Alex Elder 已提交
1735 1736 1737 1738 1739 1740 1741
	rbd_assert(obj_request->img_request == NULL);
	rbd_assert(obj_request->which == BAD_WHICH);

	if (obj_request->osd_req)
		rbd_osd_req_destroy(obj_request->osd_req);

	switch (obj_request->type) {
1742
	case OBJ_REQUEST_NODATA:
A
Alex Elder 已提交
1743
	case OBJ_REQUEST_BIO:
1744
	case OBJ_REQUEST_BVECS:
1745
		break;		/* Nothing to do */
1746 1747
	default:
		rbd_assert(0);
A
Alex Elder 已提交
1748 1749
	}

1750 1751 1752 1753 1754 1755 1756
	if (obj_request->copyup_bvecs) {
		for (i = 0; i < obj_request->copyup_bvec_count; i++) {
			if (obj_request->copyup_bvecs[i].bv_page)
				__free_page(obj_request->copyup_bvecs[i].bv_page);
		}
		kfree(obj_request->copyup_bvecs);
	}
1757

1758
	kmem_cache_free(rbd_obj_request_cache, obj_request);
A
Alex Elder 已提交
1759 1760
}

A
Alex Elder 已提交
1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771
/* It's OK to call this for a device with no parent */

static void rbd_spec_put(struct rbd_spec *spec);
static void rbd_dev_unparent(struct rbd_device *rbd_dev)
{
	rbd_dev_remove_parent(rbd_dev);
	rbd_spec_put(rbd_dev->parent_spec);
	rbd_dev->parent_spec = NULL;
	rbd_dev->parent_overlap = 0;
}

1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793
/*
 * Parent image reference counting is used to determine when an
 * image's parent fields can be safely torn down--after there are no
 * more in-flight requests to the parent image.  When the last
 * reference is dropped, cleaning them up is safe.
 */
static void rbd_dev_parent_put(struct rbd_device *rbd_dev)
{
	int counter;

	if (!rbd_dev->parent_spec)
		return;

	counter = atomic_dec_return_safe(&rbd_dev->parent_ref);
	if (counter > 0)
		return;

	/* Last reference; clean up parent data structures */

	if (!counter)
		rbd_dev_unparent(rbd_dev);
	else
1794
		rbd_warn(rbd_dev, "parent reference underflow");
1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806
}

/*
 * If an image has a non-zero parent overlap, get a reference to its
 * parent.
 *
 * Returns true if the rbd device has a parent with a non-zero
 * overlap and a reference for it was successfully taken, or
 * false otherwise.
 */
static bool rbd_dev_parent_get(struct rbd_device *rbd_dev)
{
1807
	int counter = 0;
1808 1809 1810 1811

	if (!rbd_dev->parent_spec)
		return false;

1812 1813 1814 1815
	down_read(&rbd_dev->header_rwsem);
	if (rbd_dev->parent_overlap)
		counter = atomic_inc_return_safe(&rbd_dev->parent_ref);
	up_read(&rbd_dev->header_rwsem);
1816 1817

	if (counter < 0)
1818
		rbd_warn(rbd_dev, "parent reference overflow");
1819

1820
	return counter > 0;
1821 1822
}

A
Alex Elder 已提交
1823 1824 1825 1826 1827
/*
 * Caller is responsible for filling in the list of object requests
 * that comprises the image request, and the Linux request pointer
 * (if there is one).
 */
A
Alex Elder 已提交
1828 1829
static struct rbd_img_request *rbd_img_request_create(
					struct rbd_device *rbd_dev,
A
Alex Elder 已提交
1830
					u64 offset, u64 length,
G
Guangliang Zhao 已提交
1831
					enum obj_operation_type op_type,
1832
					struct ceph_snap_context *snapc)
A
Alex Elder 已提交
1833 1834 1835
{
	struct rbd_img_request *img_request;

1836
	img_request = kmem_cache_zalloc(rbd_img_request_cache, GFP_NOIO);
A
Alex Elder 已提交
1837 1838 1839 1840 1841 1842
	if (!img_request)
		return NULL;

	img_request->rbd_dev = rbd_dev;
	img_request->offset = offset;
	img_request->length = length;
1843 1844 1845 1846
	if (op_type == OBJ_OP_DISCARD) {
		img_request_discard_set(img_request);
		img_request->snapc = snapc;
	} else if (op_type == OBJ_OP_WRITE) {
A
Alex Elder 已提交
1847
		img_request_write_set(img_request);
1848
		img_request->snapc = snapc;
A
Alex Elder 已提交
1849
	} else {
A
Alex Elder 已提交
1850
		img_request->snap_id = rbd_dev->spec->snap_id;
A
Alex Elder 已提交
1851
	}
1852
	if (rbd_dev_parent_get(rbd_dev))
1853
		img_request_layered_set(img_request);
1854

A
Alex Elder 已提交
1855 1856 1857 1858
	spin_lock_init(&img_request->completion_lock);
	INIT_LIST_HEAD(&img_request->obj_requests);
	kref_init(&img_request->kref);

A
Alex Elder 已提交
1859
	dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev,
G
Guangliang Zhao 已提交
1860
		obj_op_name(op_type), offset, length, img_request);
A
Alex Elder 已提交
1861

A
Alex Elder 已提交
1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872
	return img_request;
}

static void rbd_img_request_destroy(struct kref *kref)
{
	struct rbd_img_request *img_request;
	struct rbd_obj_request *obj_request;
	struct rbd_obj_request *next_obj_request;

	img_request = container_of(kref, struct rbd_img_request, kref);

A
Alex Elder 已提交
1873 1874
	dout("%s: img %p\n", __func__, img_request);

A
Alex Elder 已提交
1875 1876
	for_each_obj_request_safe(img_request, obj_request, next_obj_request)
		rbd_img_obj_request_del(img_request, obj_request);
1877
	rbd_assert(img_request->obj_request_count == 0);
A
Alex Elder 已提交
1878

1879 1880 1881 1882 1883
	if (img_request_layered_test(img_request)) {
		img_request_layered_clear(img_request);
		rbd_dev_parent_put(img_request->rbd_dev);
	}

1884 1885
	if (img_request_write_test(img_request) ||
		img_request_discard_test(img_request))
1886
		ceph_put_snap_context(img_request->snapc);
A
Alex Elder 已提交
1887

1888
	kmem_cache_free(rbd_img_request_cache, img_request);
A
Alex Elder 已提交
1889 1890
}

1891 1892 1893 1894 1895 1896 1897 1898 1899 1900
static struct rbd_img_request *rbd_parent_request_create(
					struct rbd_obj_request *obj_request,
					u64 img_offset, u64 length)
{
	struct rbd_img_request *parent_request;
	struct rbd_device *rbd_dev;

	rbd_assert(obj_request->img_request);
	rbd_dev = obj_request->img_request->rbd_dev;

1901
	parent_request = rbd_img_request_create(rbd_dev->parent, img_offset,
G
Guangliang Zhao 已提交
1902
						length, OBJ_OP_READ, NULL);
1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927
	if (!parent_request)
		return NULL;

	img_request_child_set(parent_request);
	rbd_obj_request_get(obj_request);
	parent_request->obj_request = obj_request;

	return parent_request;
}

static void rbd_parent_request_destroy(struct kref *kref)
{
	struct rbd_img_request *parent_request;
	struct rbd_obj_request *orig_request;

	parent_request = container_of(kref, struct rbd_img_request, kref);
	orig_request = parent_request->obj_request;

	parent_request->obj_request = NULL;
	rbd_obj_request_put(orig_request);
	img_request_child_clear(parent_request);

	rbd_img_request_destroy(kref);
}

1928 1929
static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request)
{
1930
	struct rbd_img_request *img_request;
1931 1932
	unsigned int xferred;
	int result;
A
Alex Elder 已提交
1933
	bool more;
1934

1935 1936 1937
	rbd_assert(obj_request_img_data_test(obj_request));
	img_request = obj_request->img_request;

1938 1939 1940 1941 1942
	rbd_assert(obj_request->xferred <= (u64)UINT_MAX);
	xferred = (unsigned int)obj_request->xferred;
	result = obj_request->result;
	if (result) {
		struct rbd_device *rbd_dev = img_request->rbd_dev;
G
Guangliang Zhao 已提交
1943 1944
		enum obj_operation_type op_type;

1945 1946 1947 1948 1949 1950
		if (img_request_discard_test(img_request))
			op_type = OBJ_OP_DISCARD;
		else if (img_request_write_test(img_request))
			op_type = OBJ_OP_WRITE;
		else
			op_type = OBJ_OP_READ;
1951

1952
		rbd_warn(rbd_dev, "%s %llx at %llx (%llx)",
G
Guangliang Zhao 已提交
1953 1954
			obj_op_name(op_type), obj_request->length,
			obj_request->img_offset, obj_request->offset);
1955
		rbd_warn(rbd_dev, "  result %d xferred %x",
1956 1957 1958
			result, xferred);
		if (!img_request->result)
			img_request->result = result;
1959 1960 1961 1962 1963
		/*
		 * Need to end I/O on the entire obj_request worth of
		 * bytes in case of error.
		 */
		xferred = obj_request->length;
1964 1965
	}

A
Alex Elder 已提交
1966 1967 1968 1969
	if (img_request_child_test(img_request)) {
		rbd_assert(img_request->obj_request != NULL);
		more = obj_request->which < img_request->obj_request_count - 1;
	} else {
1970 1971
		blk_status_t status = errno_to_blk_status(result);

A
Alex Elder 已提交
1972
		rbd_assert(img_request->rq != NULL);
C
Christoph Hellwig 已提交
1973

1974
		more = blk_update_request(img_request->rq, status, xferred);
C
Christoph Hellwig 已提交
1975
		if (!more)
1976
			__blk_mq_end_request(img_request->rq, status);
A
Alex Elder 已提交
1977 1978 1979
	}

	return more;
1980 1981
}

1982 1983 1984 1985 1986 1987
static void rbd_img_obj_callback(struct rbd_obj_request *obj_request)
{
	struct rbd_img_request *img_request;
	u32 which = obj_request->which;
	bool more = true;

1988
	rbd_assert(obj_request_img_data_test(obj_request));
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006
	img_request = obj_request->img_request;

	dout("%s: img %p obj %p\n", __func__, img_request, obj_request);
	rbd_assert(img_request != NULL);
	rbd_assert(img_request->obj_request_count > 0);
	rbd_assert(which != BAD_WHICH);
	rbd_assert(which < img_request->obj_request_count);

	spin_lock_irq(&img_request->completion_lock);
	if (which != img_request->next_completion)
		goto out;

	for_each_obj_request_from(img_request, obj_request) {
		rbd_assert(more);
		rbd_assert(which < img_request->obj_request_count);

		if (!obj_request_done_test(obj_request))
			break;
2007
		more = rbd_img_obj_end_request(obj_request);
2008 2009 2010 2011 2012 2013 2014
		which++;
	}

	rbd_assert(more ^ (which == img_request->obj_request_count));
	img_request->next_completion = which;
out:
	spin_unlock_irq(&img_request->completion_lock);
2015
	rbd_img_request_put(img_request);
2016 2017 2018 2019 2020

	if (!more)
		rbd_img_request_complete(img_request);
}

I
Ilya Dryomov 已提交
2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225
static void rbd_osd_req_setup_data(struct rbd_obj_request *obj_req, u32 which)
{
	switch (obj_req->type) {
	case OBJ_REQUEST_BIO:
		osd_req_op_extent_osd_data_bio(obj_req->osd_req, which,
					       &obj_req->bio_pos,
					       obj_req->length);
		break;
	case OBJ_REQUEST_BVECS:
		rbd_assert(obj_req->bvec_pos.iter.bi_size ==
							obj_req->length);
		osd_req_op_extent_osd_data_bvec_pos(obj_req->osd_req, which,
						    &obj_req->bvec_pos);
		break;
	default:
		rbd_assert(0);
	}
}

static int rbd_obj_setup_read(struct rbd_obj_request *obj_req)
{
	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;

	obj_req->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, obj_req);
	if (!obj_req->osd_req)
		return -ENOMEM;

	osd_req_op_extent_init(obj_req->osd_req, 0, CEPH_OSD_OP_READ,
			       obj_req->offset, obj_req->length, 0, 0);
	rbd_osd_req_setup_data(obj_req, 0);

	rbd_osd_req_format_read(obj_req);
	return 0;
}

static int __rbd_obj_setup_stat(struct rbd_obj_request *obj_req,
				unsigned int which)
{
	struct page **pages;

	/*
	 * The response data for a STAT call consists of:
	 *     le64 length;
	 *     struct {
	 *         le32 tv_sec;
	 *         le32 tv_nsec;
	 *     } mtime;
	 */
	pages = ceph_alloc_page_vector(1, GFP_NOIO);
	if (IS_ERR(pages))
		return PTR_ERR(pages);

	osd_req_op_init(obj_req->osd_req, which, CEPH_OSD_OP_STAT, 0);
	osd_req_op_raw_data_in_pages(obj_req->osd_req, which, pages,
				     8 + sizeof(struct ceph_timespec),
				     0, false, true);
	return 0;
}

static void __rbd_obj_setup_write(struct rbd_obj_request *obj_req,
				  unsigned int which)
{
	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
	u16 opcode;

	osd_req_op_alloc_hint_init(obj_req->osd_req, which++,
				   rbd_dev->layout.object_size,
				   rbd_dev->layout.object_size);

	if (rbd_obj_is_entire(obj_req))
		opcode = CEPH_OSD_OP_WRITEFULL;
	else
		opcode = CEPH_OSD_OP_WRITE;

	osd_req_op_extent_init(obj_req->osd_req, which, opcode,
			       obj_req->offset, obj_req->length, 0, 0);
	rbd_osd_req_setup_data(obj_req, which++);

	rbd_assert(which == obj_req->osd_req->r_num_ops);
	rbd_osd_req_format_write(obj_req);
}

static int rbd_obj_setup_write(struct rbd_obj_request *obj_req)
{
	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
	unsigned int num_osd_ops, which = 0;
	int ret;

	if (obj_request_overlaps_parent(obj_req)) {
		obj_req->write_state = RBD_OBJ_WRITE_GUARD;
		num_osd_ops = 3; /* stat + setallochint + write/writefull */
	} else {
		obj_req->write_state = RBD_OBJ_WRITE_FLAT;
		num_osd_ops = 2; /* setallochint + write/writefull */
	}

	obj_req->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_WRITE,
					      num_osd_ops, obj_req);
	if (!obj_req->osd_req)
		return -ENOMEM;

	if (obj_request_overlaps_parent(obj_req)) {
		ret = __rbd_obj_setup_stat(obj_req, which++);
		if (ret)
			return ret;
	}

	__rbd_obj_setup_write(obj_req, which);
	return 0;
}

static void __rbd_obj_setup_discard(struct rbd_obj_request *obj_req,
				    unsigned int which)
{
	u16 opcode;

	if (rbd_obj_is_entire(obj_req)) {
		if (obj_request_overlaps_parent(obj_req)) {
			opcode = CEPH_OSD_OP_TRUNCATE;
		} else {
			osd_req_op_init(obj_req->osd_req, which++,
					CEPH_OSD_OP_DELETE, 0);
			opcode = 0;
		}
	} else if (rbd_obj_is_tail(obj_req)) {
		opcode = CEPH_OSD_OP_TRUNCATE;
	} else {
		opcode = CEPH_OSD_OP_ZERO;
	}

	if (opcode)
		osd_req_op_extent_init(obj_req->osd_req, which++, opcode,
				       obj_req->offset, obj_req->length,
				       0, 0);

	rbd_assert(which == obj_req->osd_req->r_num_ops);
	rbd_osd_req_format_write(obj_req);
}

static int rbd_obj_setup_discard(struct rbd_obj_request *obj_req)
{
	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
	unsigned int num_osd_ops, which = 0;
	int ret;

	if (rbd_obj_is_entire(obj_req)) {
		obj_req->write_state = RBD_OBJ_WRITE_FLAT;
		num_osd_ops = 1; /* truncate/delete */
	} else {
		if (obj_request_overlaps_parent(obj_req)) {
			obj_req->write_state = RBD_OBJ_WRITE_GUARD;
			num_osd_ops = 2; /* stat + truncate/zero */
		} else {
			obj_req->write_state = RBD_OBJ_WRITE_FLAT;
			num_osd_ops = 1; /* truncate/zero */
		}
	}

	obj_req->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_DISCARD,
					      num_osd_ops, obj_req);
	if (!obj_req->osd_req)
		return -ENOMEM;

	if (!rbd_obj_is_entire(obj_req) &&
	    obj_request_overlaps_parent(obj_req)) {
		ret = __rbd_obj_setup_stat(obj_req, which++);
		if (ret)
			return ret;
	}

	__rbd_obj_setup_discard(obj_req, which);
	return 0;
}

/*
 * For each object request in @img_req, allocate an OSD request, add
 * individual OSD ops and prepare them for submission.  The number of
 * OSD ops depends on op_type and the overlap point (if any).
 */
static int __rbd_img_fill_request(struct rbd_img_request *img_req)
{
	struct rbd_obj_request *obj_req;
	int ret;

	for_each_obj_request(img_req, obj_req) {
		switch (rbd_img_request_op_type(img_req)) {
		case OBJ_OP_READ:
			ret = rbd_obj_setup_read(obj_req);
			break;
		case OBJ_OP_WRITE:
			ret = rbd_obj_setup_write(obj_req);
			break;
		case OBJ_OP_DISCARD:
			ret = rbd_obj_setup_discard(obj_req);
			break;
		default:
			rbd_assert(0);
		}
		if (ret)
			return ret;
	}

	return 0;
}

2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236
/*
 * Split up an image request into one or more object requests, each
 * to a different object.  The "type" parameter indicates whether
 * "data_desc" is the pointer to the head of a list of bio
 * structures, or the base of a page array.  In either case this
 * function assumes data_desc describes memory sufficient to hold
 * all data described by the image request.
 */
static int rbd_img_request_fill(struct rbd_img_request *img_request,
					enum obj_request_type type,
					void *data_desc)
A
Alex Elder 已提交
2237 2238 2239 2240
{
	struct rbd_device *rbd_dev = img_request->rbd_dev;
	struct rbd_obj_request *obj_request = NULL;
	struct rbd_obj_request *next_obj_request;
2241
	struct ceph_bio_iter bio_it;
2242
	struct ceph_bvec_iter bvec_it;
2243
	u64 img_offset;
A
Alex Elder 已提交
2244 2245
	u64 resid;

2246 2247
	dout("%s: img %p type %d data_desc %p\n", __func__, img_request,
		(int)type, data_desc);
A
Alex Elder 已提交
2248

2249
	img_offset = img_request->offset;
A
Alex Elder 已提交
2250
	resid = img_request->length;
A
Alex Elder 已提交
2251
	rbd_assert(resid > 0);
2252 2253

	if (type == OBJ_REQUEST_BIO) {
2254
		bio_it = *(struct ceph_bio_iter *)data_desc;
2255
		rbd_assert(img_offset ==
2256
			   bio_it.iter.bi_sector << SECTOR_SHIFT);
2257 2258
	} else if (type == OBJ_REQUEST_BVECS) {
		bvec_it = *(struct ceph_bvec_iter *)data_desc;
2259 2260
	}

A
Alex Elder 已提交
2261
	while (resid) {
2262
		u64 object_no = img_offset >> rbd_dev->header.obj_order;
2263 2264
		u64 offset = rbd_segment_offset(rbd_dev, img_offset);
		u64 length = rbd_segment_length(rbd_dev, img_offset, resid);
A
Alex Elder 已提交
2265

2266
		obj_request = rbd_obj_request_create(type);
A
Alex Elder 已提交
2267 2268
		if (!obj_request)
			goto out_unwind;
2269

2270
		obj_request->object_no = object_no;
2271 2272 2273
		obj_request->offset = offset;
		obj_request->length = length;

2274 2275 2276 2277 2278
		/*
		 * set obj_request->img_request before creating the
		 * osd_request so that it gets the right snapc
		 */
		rbd_img_obj_request_add(img_request, obj_request);
A
Alex Elder 已提交
2279

2280
		if (type == OBJ_REQUEST_BIO) {
2281 2282
			obj_request->bio_pos = bio_it;
			ceph_bio_iter_advance(&bio_it, length);
2283 2284 2285 2286
		} else if (type == OBJ_REQUEST_BVECS) {
			obj_request->bvec_pos = bvec_it;
			ceph_bvec_iter_shorten(&obj_request->bvec_pos, length);
			ceph_bvec_iter_advance(&bvec_it, length);
2287
		}
A
Alex Elder 已提交
2288

2289
		obj_request->callback = rbd_img_obj_callback;
2290
		obj_request->img_offset = img_offset;
2291

2292
		img_offset += length;
A
Alex Elder 已提交
2293 2294 2295
		resid -= length;
	}

I
Ilya Dryomov 已提交
2296
	return __rbd_img_fill_request(img_request);
A
Alex Elder 已提交
2297 2298 2299

out_unwind:
	for_each_obj_request_safe(img_request, obj_request, next_obj_request)
2300
		rbd_img_obj_request_del(img_request, obj_request);
A
Alex Elder 已提交
2301 2302 2303 2304 2305 2306 2307

	return -ENOMEM;
}

static int rbd_img_request_submit(struct rbd_img_request *img_request)
{
	struct rbd_obj_request *obj_request;
2308
	struct rbd_obj_request *next_obj_request;
2309
	int ret = 0;
A
Alex Elder 已提交
2310

A
Alex Elder 已提交
2311
	dout("%s: img %p\n", __func__, img_request);
A
Alex Elder 已提交
2312

2313 2314
	rbd_img_request_get(img_request);
	for_each_obj_request_safe(img_request, obj_request, next_obj_request) {
I
Ilya Dryomov 已提交
2315
		rbd_obj_request_submit(obj_request);
A
Alex Elder 已提交
2316 2317
	}

2318 2319
	rbd_img_request_put(img_request);
	return ret;
A
Alex Elder 已提交
2320
}
A
Alex Elder 已提交
2321

I
Ilya Dryomov 已提交
2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479
static void rbd_img_end_child_request(struct rbd_img_request *img_req);

static int rbd_obj_read_from_parent(struct rbd_obj_request *obj_req,
				    u64 img_offset, u32 bytes)
{
	struct rbd_img_request *img_req = obj_req->img_request;
	struct rbd_img_request *child_img_req;
	int ret;

	child_img_req = rbd_parent_request_create(obj_req, img_offset, bytes);
	if (!child_img_req)
		return -ENOMEM;

	child_img_req->callback = rbd_img_end_child_request;

	if (!rbd_img_is_write(img_req)) {
		switch (obj_req->type) {
		case OBJ_REQUEST_BIO:
			ret = rbd_img_request_fill(child_img_req,
						   OBJ_REQUEST_BIO,
						   &obj_req->bio_pos);
			break;
		case OBJ_REQUEST_BVECS:
			ret = rbd_img_request_fill(child_img_req,
						   OBJ_REQUEST_BVECS,
						   &obj_req->bvec_pos);
			break;
		default:
			rbd_assert(0);
		}
	} else {
		struct ceph_bvec_iter it = {
			.bvecs = obj_req->copyup_bvecs,
			.iter = { .bi_size = bytes },
		};

		ret = rbd_img_request_fill(child_img_req, OBJ_REQUEST_BVECS,
					   &it);
	}
	if (ret) {
		rbd_img_request_put(child_img_req);
		return ret;
	}

	rbd_img_request_submit(child_img_req);
	return 0;
}

static bool rbd_obj_handle_read(struct rbd_obj_request *obj_req)
{
	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
	int ret;

	if (obj_req->result == -ENOENT &&
	    obj_req->img_offset < rbd_dev->parent_overlap &&
	    !obj_req->tried_parent) {
		u64 obj_overlap = min(obj_req->length,
			      rbd_dev->parent_overlap - obj_req->img_offset);

		obj_req->tried_parent = true;
		ret = rbd_obj_read_from_parent(obj_req, obj_req->img_offset,
					       obj_overlap);
		if (ret) {
			obj_req->result = ret;
			return true;
		}
		return false;
	}

	/*
	 * -ENOENT means a hole in the image -- zero-fill the entire
	 * length of the request.  A short read also implies zero-fill
	 * to the end of the request.  In both cases we update xferred
	 * count to indicate the whole request was satisfied.
	 */
	if (obj_req->result == -ENOENT ||
	    (!obj_req->result && obj_req->xferred < obj_req->length)) {
		rbd_assert(!obj_req->xferred || !obj_req->result);
		rbd_obj_zero_range(obj_req, obj_req->xferred,
				   obj_req->length - obj_req->xferred);
		obj_req->result = 0;
		obj_req->xferred = obj_req->length;
	}

	return true;
}

/*
 * copyup_bvecs pages are never highmem pages
 */
static bool is_zero_bvecs(struct bio_vec *bvecs, u32 bytes)
{
	struct ceph_bvec_iter it = {
		.bvecs = bvecs,
		.iter = { .bi_size = bytes },
	};

	ceph_bvec_iter_advance_step(&it, bytes, ({
		if (memchr_inv(page_address(bv.bv_page) + bv.bv_offset, 0,
			       bv.bv_len))
			return false;
	}));
	return true;
}

static int rbd_obj_issue_copyup(struct rbd_obj_request *obj_req, u32 bytes)
{
	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
	unsigned int num_osd_ops = obj_req->osd_req->r_num_ops;

	dout("%s obj_req %p bytes %u\n", __func__, obj_req, bytes);
	rbd_assert(obj_req->osd_req->r_ops[0].op == CEPH_OSD_OP_STAT);
	rbd_osd_req_destroy(obj_req->osd_req);

	/*
	 * Create a copyup request with the same number of OSD ops as
	 * the original request.  The original request was stat + op(s),
	 * the new copyup request will be copyup + the same op(s).
	 */
	obj_req->osd_req = rbd_osd_req_create(rbd_dev,
				rbd_img_request_op_type(obj_req->img_request),
				num_osd_ops, obj_req);
	if (!obj_req->osd_req)
		return -ENOMEM;

	/*
	 * Only send non-zero copyup data to save some I/O and network
	 * bandwidth -- zero copyup data is equivalent to the object not
	 * existing.
	 */
	if (is_zero_bvecs(obj_req->copyup_bvecs, bytes)) {
		dout("%s obj_req %p detected zeroes\n", __func__, obj_req);
		bytes = 0;
	}

	osd_req_op_cls_init(obj_req->osd_req, 0, CEPH_OSD_OP_CALL, "rbd",
			    "copyup");
	osd_req_op_cls_request_data_bvecs(obj_req->osd_req, 0,
					  obj_req->copyup_bvecs, bytes);

	switch (rbd_img_request_op_type(obj_req->img_request)) {
	case OBJ_OP_WRITE:
		__rbd_obj_setup_write(obj_req, 1);
		break;
	case OBJ_OP_DISCARD:
		rbd_assert(!rbd_obj_is_entire(obj_req));
		__rbd_obj_setup_discard(obj_req, 1);
		break;
	default:
		rbd_assert(0);
	}

	rbd_obj_request_submit(obj_req);
	/* FIXME: in lieu of rbd_img_obj_callback() */
	rbd_img_request_put(obj_req->img_request);
	return 0;
}

2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507
static int setup_copyup_bvecs(struct rbd_obj_request *obj_req, u64 obj_overlap)
{
	u32 i;

	rbd_assert(!obj_req->copyup_bvecs);
	obj_req->copyup_bvec_count = calc_pages_for(0, obj_overlap);
	obj_req->copyup_bvecs = kcalloc(obj_req->copyup_bvec_count,
					sizeof(*obj_req->copyup_bvecs),
					GFP_NOIO);
	if (!obj_req->copyup_bvecs)
		return -ENOMEM;

	for (i = 0; i < obj_req->copyup_bvec_count; i++) {
		unsigned int len = min(obj_overlap, (u64)PAGE_SIZE);

		obj_req->copyup_bvecs[i].bv_page = alloc_page(GFP_NOIO);
		if (!obj_req->copyup_bvecs[i].bv_page)
			return -ENOMEM;

		obj_req->copyup_bvecs[i].bv_offset = 0;
		obj_req->copyup_bvecs[i].bv_len = len;
		obj_overlap -= len;
	}

	rbd_assert(!obj_overlap);
	return 0;
}

I
Ilya Dryomov 已提交
2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650
static int rbd_obj_handle_write_guard(struct rbd_obj_request *obj_req)
{
	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
	u64 img_offset;
	u64 obj_overlap;
	int ret;

	if (!obj_request_overlaps_parent(obj_req)) {
		/*
		 * The overlap has become 0 (most likely because the
		 * image has been flattened).  Use rbd_obj_issue_copyup()
		 * to re-submit the original write request -- the copyup
		 * operation itself will be a no-op, since someone must
		 * have populated the child object while we weren't
		 * looking.  Move to WRITE_FLAT state as we'll be done
		 * with the operation once the null copyup completes.
		 */
		obj_req->write_state = RBD_OBJ_WRITE_FLAT;
		return rbd_obj_issue_copyup(obj_req, 0);
	}

	/*
	 * Determine the byte range covered by the object in the
	 * child image to which the original request was to be sent.
	 */
	img_offset = obj_req->img_offset - obj_req->offset;
	obj_overlap = rbd_dev->layout.object_size;

	/*
	 * There is no defined parent data beyond the parent
	 * overlap, so limit what we read at that boundary if
	 * necessary.
	 */
	if (img_offset + obj_overlap > rbd_dev->parent_overlap) {
		rbd_assert(img_offset < rbd_dev->parent_overlap);
		obj_overlap = rbd_dev->parent_overlap - img_offset;
	}

	ret = setup_copyup_bvecs(obj_req, obj_overlap);
	if (ret)
		return ret;

	obj_req->write_state = RBD_OBJ_WRITE_COPYUP;
	return rbd_obj_read_from_parent(obj_req, img_offset, obj_overlap);
}

static bool rbd_obj_handle_write(struct rbd_obj_request *obj_req)
{
	int ret;

again:
	switch (obj_req->write_state) {
	case RBD_OBJ_WRITE_GUARD:
		rbd_assert(!obj_req->xferred);
		if (obj_req->result == -ENOENT) {
			/*
			 * The target object doesn't exist.  Read the data for
			 * the entire target object up to the overlap point (if
			 * any) from the parent, so we can use it for a copyup.
			 */
			ret = rbd_obj_handle_write_guard(obj_req);
			if (ret) {
				obj_req->result = ret;
				return true;
			}
			return false;
		}
		/* fall through */
	case RBD_OBJ_WRITE_FLAT:
		if (!obj_req->result)
			/*
			 * There is no such thing as a successful short
			 * write -- indicate the whole request was satisfied.
			 */
			obj_req->xferred = obj_req->length;
		return true;
	case RBD_OBJ_WRITE_COPYUP:
		obj_req->write_state = RBD_OBJ_WRITE_GUARD;
		if (obj_req->result)
			goto again;

		rbd_assert(obj_req->xferred);
		ret = rbd_obj_issue_copyup(obj_req, obj_req->xferred);
		if (ret) {
			obj_req->result = ret;
			return true;
		}
		return false;
	default:
		rbd_assert(0);
	}
}

/*
 * Returns true if @obj_req is completed, or false otherwise.
 */
static bool __rbd_obj_handle_request(struct rbd_obj_request *obj_req)
{
	switch (rbd_img_request_op_type(obj_req->img_request)) {
	case OBJ_OP_READ:
		return rbd_obj_handle_read(obj_req);
	case OBJ_OP_WRITE:
		return rbd_obj_handle_write(obj_req);
	case OBJ_OP_DISCARD:
		if (rbd_obj_handle_write(obj_req)) {
			/*
			 * Hide -ENOENT from delete/truncate/zero -- discarding
			 * a non-existent object is not a problem.
			 */
			if (obj_req->result == -ENOENT) {
				obj_req->result = 0;
				obj_req->xferred = obj_req->length;
			}
			return true;
		}
		return false;
	default:
		rbd_assert(0);
	}
}

static void rbd_img_end_child_request(struct rbd_img_request *img_req)
{
	struct rbd_obj_request *obj_req = img_req->obj_request;

	rbd_assert(test_bit(IMG_REQ_CHILD, &img_req->flags));

	obj_req->result = img_req->result;
	obj_req->xferred = img_req->xferred;
	rbd_img_request_put(img_req);

	rbd_obj_handle_request(obj_req);
}

static void rbd_obj_handle_request(struct rbd_obj_request *obj_req)
{
	if (!__rbd_obj_handle_request(obj_req))
		return;

	obj_request_done_set(obj_req);
	rbd_obj_request_complete(obj_req);
}

2651
static const struct rbd_client_id rbd_empty_cid;
A
Alex Elder 已提交
2652

2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688
static bool rbd_cid_equal(const struct rbd_client_id *lhs,
			  const struct rbd_client_id *rhs)
{
	return lhs->gid == rhs->gid && lhs->handle == rhs->handle;
}

static struct rbd_client_id rbd_get_cid(struct rbd_device *rbd_dev)
{
	struct rbd_client_id cid;

	mutex_lock(&rbd_dev->watch_mutex);
	cid.gid = ceph_client_gid(rbd_dev->rbd_client->client);
	cid.handle = rbd_dev->watch_cookie;
	mutex_unlock(&rbd_dev->watch_mutex);
	return cid;
}

/*
 * lock_rwsem must be held for write
 */
static void rbd_set_owner_cid(struct rbd_device *rbd_dev,
			      const struct rbd_client_id *cid)
{
	dout("%s rbd_dev %p %llu-%llu -> %llu-%llu\n", __func__, rbd_dev,
	     rbd_dev->owner_cid.gid, rbd_dev->owner_cid.handle,
	     cid->gid, cid->handle);
	rbd_dev->owner_cid = *cid; /* struct */
}

static void format_lock_cookie(struct rbd_device *rbd_dev, char *buf)
{
	mutex_lock(&rbd_dev->watch_mutex);
	sprintf(buf, "%s %llu", RBD_LOCK_COOKIE_PREFIX, rbd_dev->watch_cookie);
	mutex_unlock(&rbd_dev->watch_mutex);
}

2689 2690 2691 2692 2693 2694 2695 2696 2697
static void __rbd_lock(struct rbd_device *rbd_dev, const char *cookie)
{
	struct rbd_client_id cid = rbd_get_cid(rbd_dev);

	strcpy(rbd_dev->lock_cookie, cookie);
	rbd_set_owner_cid(rbd_dev, &cid);
	queue_work(rbd_dev->task_wq, &rbd_dev->acquired_lock_work);
}

2698 2699 2700 2701
/*
 * lock_rwsem must be held for write
 */
static int rbd_lock(struct rbd_device *rbd_dev)
A
Alex Elder 已提交
2702
{
2703
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2704
	char cookie[32];
2705
	int ret;
A
Alex Elder 已提交
2706

I
Ilya Dryomov 已提交
2707 2708
	WARN_ON(__rbd_is_lock_owner(rbd_dev) ||
		rbd_dev->lock_cookie[0] != '\0');
2709

2710 2711 2712 2713
	format_lock_cookie(rbd_dev, cookie);
	ret = ceph_cls_lock(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc,
			    RBD_LOCK_NAME, CEPH_CLS_LOCK_EXCLUSIVE, cookie,
			    RBD_LOCK_TAG, "", 0);
2714
	if (ret)
2715
		return ret;
A
Alex Elder 已提交
2716

2717
	rbd_dev->lock_state = RBD_LOCK_STATE_LOCKED;
2718
	__rbd_lock(rbd_dev, cookie);
2719
	return 0;
A
Alex Elder 已提交
2720 2721
}

2722 2723 2724
/*
 * lock_rwsem must be held for write
 */
I
Ilya Dryomov 已提交
2725
static void rbd_unlock(struct rbd_device *rbd_dev)
2726
{
2727
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2728 2729
	int ret;

I
Ilya Dryomov 已提交
2730 2731
	WARN_ON(!__rbd_is_lock_owner(rbd_dev) ||
		rbd_dev->lock_cookie[0] == '\0');
2732

2733
	ret = ceph_cls_unlock(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc,
I
Ilya Dryomov 已提交
2734
			      RBD_LOCK_NAME, rbd_dev->lock_cookie);
I
Ilya Dryomov 已提交
2735 2736
	if (ret && ret != -ENOENT)
		rbd_warn(rbd_dev, "failed to unlock: %d", ret);
2737

I
Ilya Dryomov 已提交
2738 2739
	/* treat errors as the image is unlocked */
	rbd_dev->lock_state = RBD_LOCK_STATE_UNLOCKED;
I
Ilya Dryomov 已提交
2740
	rbd_dev->lock_cookie[0] = '\0';
2741 2742
	rbd_set_owner_cid(rbd_dev, &rbd_empty_cid);
	queue_work(rbd_dev->task_wq, &rbd_dev->released_lock_work);
2743 2744
}

2745 2746 2747 2748
static int __rbd_notify_op_lock(struct rbd_device *rbd_dev,
				enum rbd_notify_op notify_op,
				struct page ***preply_pages,
				size_t *preply_len)
2749 2750
{
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2751 2752 2753 2754
	struct rbd_client_id cid = rbd_get_cid(rbd_dev);
	int buf_size = 4 + 8 + 8 + CEPH_ENCODING_START_BLK_LEN;
	char buf[buf_size];
	void *p = buf;
2755

2756
	dout("%s rbd_dev %p notify_op %d\n", __func__, rbd_dev, notify_op);
2757

2758 2759 2760 2761 2762
	/* encode *LockPayload NotifyMessage (op + ClientId) */
	ceph_start_encoding(&p, 2, 1, buf_size - CEPH_ENCODING_START_BLK_LEN);
	ceph_encode_32(&p, notify_op);
	ceph_encode_64(&p, cid.gid);
	ceph_encode_64(&p, cid.handle);
2763

2764 2765 2766
	return ceph_osdc_notify(osdc, &rbd_dev->header_oid,
				&rbd_dev->header_oloc, buf, buf_size,
				RBD_NOTIFY_TIMEOUT, preply_pages, preply_len);
2767 2768
}

2769 2770
static void rbd_notify_op_lock(struct rbd_device *rbd_dev,
			       enum rbd_notify_op notify_op)
2771
{
2772 2773
	struct page **reply_pages;
	size_t reply_len;
2774

2775 2776 2777
	__rbd_notify_op_lock(rbd_dev, notify_op, &reply_pages, &reply_len);
	ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len));
}
2778

2779 2780 2781 2782
static void rbd_notify_acquired_lock(struct work_struct *work)
{
	struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
						  acquired_lock_work);
2783

2784
	rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_ACQUIRED_LOCK);
2785 2786
}

2787
static void rbd_notify_released_lock(struct work_struct *work)
2788
{
2789 2790
	struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
						  released_lock_work);
I
Ilya Dryomov 已提交
2791

2792
	rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_RELEASED_LOCK);
2793 2794
}

2795
static int rbd_request_lock(struct rbd_device *rbd_dev)
2796
{
2797 2798 2799
	struct page **reply_pages;
	size_t reply_len;
	bool lock_owner_responded = false;
2800 2801
	int ret;

2802
	dout("%s rbd_dev %p\n", __func__, rbd_dev);
2803

2804 2805 2806 2807
	ret = __rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_REQUEST_LOCK,
				   &reply_pages, &reply_len);
	if (ret && ret != -ETIMEDOUT) {
		rbd_warn(rbd_dev, "failed to request lock: %d", ret);
2808
		goto out;
2809
	}
2810

2811 2812 2813 2814
	if (reply_len > 0 && reply_len <= PAGE_SIZE) {
		void *p = page_address(reply_pages[0]);
		void *const end = p + reply_len;
		u32 n;
2815

2816 2817 2818 2819
		ceph_decode_32_safe(&p, end, n, e_inval); /* num_acks */
		while (n--) {
			u8 struct_v;
			u32 len;
2820

2821 2822
			ceph_decode_need(&p, end, 8 + 8, e_inval);
			p += 8 + 8; /* skip gid and cookie */
2823

2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055
			ceph_decode_32_safe(&p, end, len, e_inval);
			if (!len)
				continue;

			if (lock_owner_responded) {
				rbd_warn(rbd_dev,
					 "duplicate lock owners detected");
				ret = -EIO;
				goto out;
			}

			lock_owner_responded = true;
			ret = ceph_start_decoding(&p, end, 1, "ResponseMessage",
						  &struct_v, &len);
			if (ret) {
				rbd_warn(rbd_dev,
					 "failed to decode ResponseMessage: %d",
					 ret);
				goto e_inval;
			}

			ret = ceph_decode_32(&p);
		}
	}

	if (!lock_owner_responded) {
		rbd_warn(rbd_dev, "no lock owners detected");
		ret = -ETIMEDOUT;
	}

out:
	ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len));
	return ret;

e_inval:
	ret = -EINVAL;
	goto out;
}

static void wake_requests(struct rbd_device *rbd_dev, bool wake_all)
{
	dout("%s rbd_dev %p wake_all %d\n", __func__, rbd_dev, wake_all);

	cancel_delayed_work(&rbd_dev->lock_dwork);
	if (wake_all)
		wake_up_all(&rbd_dev->lock_waitq);
	else
		wake_up(&rbd_dev->lock_waitq);
}

static int get_lock_owner_info(struct rbd_device *rbd_dev,
			       struct ceph_locker **lockers, u32 *num_lockers)
{
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
	u8 lock_type;
	char *lock_tag;
	int ret;

	dout("%s rbd_dev %p\n", __func__, rbd_dev);

	ret = ceph_cls_lock_info(osdc, &rbd_dev->header_oid,
				 &rbd_dev->header_oloc, RBD_LOCK_NAME,
				 &lock_type, &lock_tag, lockers, num_lockers);
	if (ret)
		return ret;

	if (*num_lockers == 0) {
		dout("%s rbd_dev %p no lockers detected\n", __func__, rbd_dev);
		goto out;
	}

	if (strcmp(lock_tag, RBD_LOCK_TAG)) {
		rbd_warn(rbd_dev, "locked by external mechanism, tag %s",
			 lock_tag);
		ret = -EBUSY;
		goto out;
	}

	if (lock_type == CEPH_CLS_LOCK_SHARED) {
		rbd_warn(rbd_dev, "shared lock type detected");
		ret = -EBUSY;
		goto out;
	}

	if (strncmp((*lockers)[0].id.cookie, RBD_LOCK_COOKIE_PREFIX,
		    strlen(RBD_LOCK_COOKIE_PREFIX))) {
		rbd_warn(rbd_dev, "locked by external mechanism, cookie %s",
			 (*lockers)[0].id.cookie);
		ret = -EBUSY;
		goto out;
	}

out:
	kfree(lock_tag);
	return ret;
}

static int find_watcher(struct rbd_device *rbd_dev,
			const struct ceph_locker *locker)
{
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
	struct ceph_watch_item *watchers;
	u32 num_watchers;
	u64 cookie;
	int i;
	int ret;

	ret = ceph_osdc_list_watchers(osdc, &rbd_dev->header_oid,
				      &rbd_dev->header_oloc, &watchers,
				      &num_watchers);
	if (ret)
		return ret;

	sscanf(locker->id.cookie, RBD_LOCK_COOKIE_PREFIX " %llu", &cookie);
	for (i = 0; i < num_watchers; i++) {
		if (!memcmp(&watchers[i].addr, &locker->info.addr,
			    sizeof(locker->info.addr)) &&
		    watchers[i].cookie == cookie) {
			struct rbd_client_id cid = {
				.gid = le64_to_cpu(watchers[i].name.num),
				.handle = cookie,
			};

			dout("%s rbd_dev %p found cid %llu-%llu\n", __func__,
			     rbd_dev, cid.gid, cid.handle);
			rbd_set_owner_cid(rbd_dev, &cid);
			ret = 1;
			goto out;
		}
	}

	dout("%s rbd_dev %p no watchers\n", __func__, rbd_dev);
	ret = 0;
out:
	kfree(watchers);
	return ret;
}

/*
 * lock_rwsem must be held for write
 */
static int rbd_try_lock(struct rbd_device *rbd_dev)
{
	struct ceph_client *client = rbd_dev->rbd_client->client;
	struct ceph_locker *lockers;
	u32 num_lockers;
	int ret;

	for (;;) {
		ret = rbd_lock(rbd_dev);
		if (ret != -EBUSY)
			return ret;

		/* determine if the current lock holder is still alive */
		ret = get_lock_owner_info(rbd_dev, &lockers, &num_lockers);
		if (ret)
			return ret;

		if (num_lockers == 0)
			goto again;

		ret = find_watcher(rbd_dev, lockers);
		if (ret) {
			if (ret > 0)
				ret = 0; /* have to request lock */
			goto out;
		}

		rbd_warn(rbd_dev, "%s%llu seems dead, breaking lock",
			 ENTITY_NAME(lockers[0].id.name));

		ret = ceph_monc_blacklist_add(&client->monc,
					      &lockers[0].info.addr);
		if (ret) {
			rbd_warn(rbd_dev, "blacklist of %s%llu failed: %d",
				 ENTITY_NAME(lockers[0].id.name), ret);
			goto out;
		}

		ret = ceph_cls_break_lock(&client->osdc, &rbd_dev->header_oid,
					  &rbd_dev->header_oloc, RBD_LOCK_NAME,
					  lockers[0].id.cookie,
					  &lockers[0].id.name);
		if (ret && ret != -ENOENT)
			goto out;

again:
		ceph_free_lockers(lockers, num_lockers);
	}

out:
	ceph_free_lockers(lockers, num_lockers);
	return ret;
}

/*
 * ret is set only if lock_state is RBD_LOCK_STATE_UNLOCKED
 */
static enum rbd_lock_state rbd_try_acquire_lock(struct rbd_device *rbd_dev,
						int *pret)
{
	enum rbd_lock_state lock_state;

	down_read(&rbd_dev->lock_rwsem);
	dout("%s rbd_dev %p read lock_state %d\n", __func__, rbd_dev,
	     rbd_dev->lock_state);
	if (__rbd_is_lock_owner(rbd_dev)) {
		lock_state = rbd_dev->lock_state;
		up_read(&rbd_dev->lock_rwsem);
		return lock_state;
	}

	up_read(&rbd_dev->lock_rwsem);
	down_write(&rbd_dev->lock_rwsem);
	dout("%s rbd_dev %p write lock_state %d\n", __func__, rbd_dev,
	     rbd_dev->lock_state);
	if (!__rbd_is_lock_owner(rbd_dev)) {
		*pret = rbd_try_lock(rbd_dev);
		if (*pret)
			rbd_warn(rbd_dev, "failed to acquire lock: %d", *pret);
	}

	lock_state = rbd_dev->lock_state;
	up_write(&rbd_dev->lock_rwsem);
	return lock_state;
}

static void rbd_acquire_lock(struct work_struct *work)
{
	struct rbd_device *rbd_dev = container_of(to_delayed_work(work),
					    struct rbd_device, lock_dwork);
	enum rbd_lock_state lock_state;
3056
	int ret = 0;
3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071

	dout("%s rbd_dev %p\n", __func__, rbd_dev);
again:
	lock_state = rbd_try_acquire_lock(rbd_dev, &ret);
	if (lock_state != RBD_LOCK_STATE_UNLOCKED || ret == -EBLACKLISTED) {
		if (lock_state == RBD_LOCK_STATE_LOCKED)
			wake_requests(rbd_dev, true);
		dout("%s rbd_dev %p lock_state %d ret %d - done\n", __func__,
		     rbd_dev, lock_state, ret);
		return;
	}

	ret = rbd_request_lock(rbd_dev);
	if (ret == -ETIMEDOUT) {
		goto again; /* treat this as a dead client */
I
Ilya Dryomov 已提交
3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082 3083
	} else if (ret == -EROFS) {
		rbd_warn(rbd_dev, "peer will not release lock");
		/*
		 * If this is rbd_add_acquire_lock(), we want to fail
		 * immediately -- reuse BLACKLISTED flag.  Otherwise we
		 * want to block.
		 */
		if (!(rbd_dev->disk->flags & GENHD_FL_UP)) {
			set_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags);
			/* wake "rbd map --exclusive" process */
			wake_requests(rbd_dev, false);
		}
3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111
	} else if (ret < 0) {
		rbd_warn(rbd_dev, "error requesting lock: %d", ret);
		mod_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork,
				 RBD_RETRY_DELAY);
	} else {
		/*
		 * lock owner acked, but resend if we don't see them
		 * release the lock
		 */
		dout("%s rbd_dev %p requeueing lock_dwork\n", __func__,
		     rbd_dev);
		mod_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork,
		    msecs_to_jiffies(2 * RBD_NOTIFY_TIMEOUT * MSEC_PER_SEC));
	}
}

/*
 * lock_rwsem must be held for write
 */
static bool rbd_release_lock(struct rbd_device *rbd_dev)
{
	dout("%s rbd_dev %p read lock_state %d\n", __func__, rbd_dev,
	     rbd_dev->lock_state);
	if (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED)
		return false;

	rbd_dev->lock_state = RBD_LOCK_STATE_RELEASING;
	downgrade_write(&rbd_dev->lock_rwsem);
3112
	/*
3113
	 * Ensure that all in-flight IO is flushed.
3114
	 *
3115 3116
	 * FIXME: ceph_osdc_sync() flushes the entire OSD client, which
	 * may be shared with other devices.
3117
	 */
3118 3119 3120 3121 3122 3123 3124 3125 3126
	ceph_osdc_sync(&rbd_dev->rbd_client->client->osdc);
	up_read(&rbd_dev->lock_rwsem);

	down_write(&rbd_dev->lock_rwsem);
	dout("%s rbd_dev %p write lock_state %d\n", __func__, rbd_dev,
	     rbd_dev->lock_state);
	if (rbd_dev->lock_state != RBD_LOCK_STATE_RELEASING)
		return false;

I
Ilya Dryomov 已提交
3127 3128 3129 3130 3131 3132 3133 3134 3135
	rbd_unlock(rbd_dev);
	/*
	 * Give others a chance to grab the lock - we would re-acquire
	 * almost immediately if we got new IO during ceph_osdc_sync()
	 * otherwise.  We need to ack our own notifications, so this
	 * lock_dwork will be requeued from rbd_wait_state_locked()
	 * after wake_requests() in rbd_handle_released_lock().
	 */
	cancel_delayed_work(&rbd_dev->lock_dwork);
3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215
	return true;
}

static void rbd_release_lock_work(struct work_struct *work)
{
	struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
						  unlock_work);

	down_write(&rbd_dev->lock_rwsem);
	rbd_release_lock(rbd_dev);
	up_write(&rbd_dev->lock_rwsem);
}

static void rbd_handle_acquired_lock(struct rbd_device *rbd_dev, u8 struct_v,
				     void **p)
{
	struct rbd_client_id cid = { 0 };

	if (struct_v >= 2) {
		cid.gid = ceph_decode_64(p);
		cid.handle = ceph_decode_64(p);
	}

	dout("%s rbd_dev %p cid %llu-%llu\n", __func__, rbd_dev, cid.gid,
	     cid.handle);
	if (!rbd_cid_equal(&cid, &rbd_empty_cid)) {
		down_write(&rbd_dev->lock_rwsem);
		if (rbd_cid_equal(&cid, &rbd_dev->owner_cid)) {
			/*
			 * we already know that the remote client is
			 * the owner
			 */
			up_write(&rbd_dev->lock_rwsem);
			return;
		}

		rbd_set_owner_cid(rbd_dev, &cid);
		downgrade_write(&rbd_dev->lock_rwsem);
	} else {
		down_read(&rbd_dev->lock_rwsem);
	}

	if (!__rbd_is_lock_owner(rbd_dev))
		wake_requests(rbd_dev, false);
	up_read(&rbd_dev->lock_rwsem);
}

static void rbd_handle_released_lock(struct rbd_device *rbd_dev, u8 struct_v,
				     void **p)
{
	struct rbd_client_id cid = { 0 };

	if (struct_v >= 2) {
		cid.gid = ceph_decode_64(p);
		cid.handle = ceph_decode_64(p);
	}

	dout("%s rbd_dev %p cid %llu-%llu\n", __func__, rbd_dev, cid.gid,
	     cid.handle);
	if (!rbd_cid_equal(&cid, &rbd_empty_cid)) {
		down_write(&rbd_dev->lock_rwsem);
		if (!rbd_cid_equal(&cid, &rbd_dev->owner_cid)) {
			dout("%s rbd_dev %p unexpected owner, cid %llu-%llu != owner_cid %llu-%llu\n",
			     __func__, rbd_dev, cid.gid, cid.handle,
			     rbd_dev->owner_cid.gid, rbd_dev->owner_cid.handle);
			up_write(&rbd_dev->lock_rwsem);
			return;
		}

		rbd_set_owner_cid(rbd_dev, &rbd_empty_cid);
		downgrade_write(&rbd_dev->lock_rwsem);
	} else {
		down_read(&rbd_dev->lock_rwsem);
	}

	if (!__rbd_is_lock_owner(rbd_dev))
		wake_requests(rbd_dev, false);
	up_read(&rbd_dev->lock_rwsem);
}

3216 3217 3218 3219 3220 3221
/*
 * Returns result for ResponseMessage to be encoded (<= 0), or 1 if no
 * ResponseMessage is needed.
 */
static int rbd_handle_request_lock(struct rbd_device *rbd_dev, u8 struct_v,
				   void **p)
3222 3223 3224
{
	struct rbd_client_id my_cid = rbd_get_cid(rbd_dev);
	struct rbd_client_id cid = { 0 };
3225
	int result = 1;
3226 3227 3228 3229 3230 3231 3232 3233 3234

	if (struct_v >= 2) {
		cid.gid = ceph_decode_64(p);
		cid.handle = ceph_decode_64(p);
	}

	dout("%s rbd_dev %p cid %llu-%llu\n", __func__, rbd_dev, cid.gid,
	     cid.handle);
	if (rbd_cid_equal(&cid, &my_cid))
3235
		return result;
3236 3237

	down_read(&rbd_dev->lock_rwsem);
3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249
	if (__rbd_is_lock_owner(rbd_dev)) {
		if (rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED &&
		    rbd_cid_equal(&rbd_dev->owner_cid, &rbd_empty_cid))
			goto out_unlock;

		/*
		 * encode ResponseMessage(0) so the peer can detect
		 * a missing owner
		 */
		result = 0;

		if (rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED) {
I
Ilya Dryomov 已提交
3250 3251 3252 3253 3254 3255 3256 3257 3258
			if (!rbd_dev->opts->exclusive) {
				dout("%s rbd_dev %p queueing unlock_work\n",
				     __func__, rbd_dev);
				queue_work(rbd_dev->task_wq,
					   &rbd_dev->unlock_work);
			} else {
				/* refuse to release the lock */
				result = -EROFS;
			}
3259 3260
		}
	}
3261 3262

out_unlock:
3263
	up_read(&rbd_dev->lock_rwsem);
3264
	return result;
3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283 3284
}

static void __rbd_acknowledge_notify(struct rbd_device *rbd_dev,
				     u64 notify_id, u64 cookie, s32 *result)
{
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
	int buf_size = 4 + CEPH_ENCODING_START_BLK_LEN;
	char buf[buf_size];
	int ret;

	if (result) {
		void *p = buf;

		/* encode ResponseMessage */
		ceph_start_encoding(&p, 1, 1,
				    buf_size - CEPH_ENCODING_START_BLK_LEN);
		ceph_encode_32(&p, *result);
	} else {
		buf_size = 0;
	}
A
Alex Elder 已提交
3285

3286 3287
	ret = ceph_osdc_notify_ack(osdc, &rbd_dev->header_oid,
				   &rbd_dev->header_oloc, notify_id, cookie,
3288
				   buf, buf_size);
3289
	if (ret)
3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312
		rbd_warn(rbd_dev, "acknowledge_notify failed: %d", ret);
}

static void rbd_acknowledge_notify(struct rbd_device *rbd_dev, u64 notify_id,
				   u64 cookie)
{
	dout("%s rbd_dev %p\n", __func__, rbd_dev);
	__rbd_acknowledge_notify(rbd_dev, notify_id, cookie, NULL);
}

static void rbd_acknowledge_notify_result(struct rbd_device *rbd_dev,
					  u64 notify_id, u64 cookie, s32 result)
{
	dout("%s rbd_dev %p result %d\n", __func__, rbd_dev, result);
	__rbd_acknowledge_notify(rbd_dev, notify_id, cookie, &result);
}

static void rbd_watch_cb(void *arg, u64 notify_id, u64 cookie,
			 u64 notifier_id, void *data, size_t data_len)
{
	struct rbd_device *rbd_dev = arg;
	void *p = data;
	void *const end = p + data_len;
3313
	u8 struct_v = 0;
3314 3315 3316 3317 3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346
	u32 len;
	u32 notify_op;
	int ret;

	dout("%s rbd_dev %p cookie %llu notify_id %llu data_len %zu\n",
	     __func__, rbd_dev, cookie, notify_id, data_len);
	if (data_len) {
		ret = ceph_start_decoding(&p, end, 1, "NotifyMessage",
					  &struct_v, &len);
		if (ret) {
			rbd_warn(rbd_dev, "failed to decode NotifyMessage: %d",
				 ret);
			return;
		}

		notify_op = ceph_decode_32(&p);
	} else {
		/* legacy notification for header updates */
		notify_op = RBD_NOTIFY_OP_HEADER_UPDATE;
		len = 0;
	}

	dout("%s rbd_dev %p notify_op %u\n", __func__, rbd_dev, notify_op);
	switch (notify_op) {
	case RBD_NOTIFY_OP_ACQUIRED_LOCK:
		rbd_handle_acquired_lock(rbd_dev, struct_v, &p);
		rbd_acknowledge_notify(rbd_dev, notify_id, cookie);
		break;
	case RBD_NOTIFY_OP_RELEASED_LOCK:
		rbd_handle_released_lock(rbd_dev, struct_v, &p);
		rbd_acknowledge_notify(rbd_dev, notify_id, cookie);
		break;
	case RBD_NOTIFY_OP_REQUEST_LOCK:
3347 3348
		ret = rbd_handle_request_lock(rbd_dev, struct_v, &p);
		if (ret <= 0)
3349
			rbd_acknowledge_notify_result(rbd_dev, notify_id,
3350
						      cookie, ret);
3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368
		else
			rbd_acknowledge_notify(rbd_dev, notify_id, cookie);
		break;
	case RBD_NOTIFY_OP_HEADER_UPDATE:
		ret = rbd_dev_refresh(rbd_dev);
		if (ret)
			rbd_warn(rbd_dev, "refresh failed: %d", ret);

		rbd_acknowledge_notify(rbd_dev, notify_id, cookie);
		break;
	default:
		if (rbd_is_lock_owner(rbd_dev))
			rbd_acknowledge_notify_result(rbd_dev, notify_id,
						      cookie, -EOPNOTSUPP);
		else
			rbd_acknowledge_notify(rbd_dev, notify_id, cookie);
		break;
	}
A
Alex Elder 已提交
3369 3370
}

3371 3372
static void __rbd_unregister_watch(struct rbd_device *rbd_dev);

3373
static void rbd_watch_errcb(void *arg, u64 cookie, int err)
3374
{
3375
	struct rbd_device *rbd_dev = arg;
3376

3377
	rbd_warn(rbd_dev, "encountered watch error: %d", err);
3378

3379 3380 3381 3382
	down_write(&rbd_dev->lock_rwsem);
	rbd_set_owner_cid(rbd_dev, &rbd_empty_cid);
	up_write(&rbd_dev->lock_rwsem);

3383 3384 3385 3386
	mutex_lock(&rbd_dev->watch_mutex);
	if (rbd_dev->watch_state == RBD_WATCH_STATE_REGISTERED) {
		__rbd_unregister_watch(rbd_dev);
		rbd_dev->watch_state = RBD_WATCH_STATE_ERROR;
3387

3388
		queue_delayed_work(rbd_dev->task_wq, &rbd_dev->watch_dwork, 0);
3389
	}
3390
	mutex_unlock(&rbd_dev->watch_mutex);
3391 3392
}

3393
/*
3394
 * watch_mutex must be locked
3395
 */
3396
static int __rbd_register_watch(struct rbd_device *rbd_dev)
3397 3398
{
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3399
	struct ceph_osd_linger_request *handle;
3400

3401
	rbd_assert(!rbd_dev->watch_handle);
3402
	dout("%s rbd_dev %p\n", __func__, rbd_dev);
3403

3404 3405 3406 3407 3408
	handle = ceph_osdc_watch(osdc, &rbd_dev->header_oid,
				 &rbd_dev->header_oloc, rbd_watch_cb,
				 rbd_watch_errcb, rbd_dev);
	if (IS_ERR(handle))
		return PTR_ERR(handle);
3409

3410
	rbd_dev->watch_handle = handle;
3411 3412 3413
	return 0;
}

3414 3415 3416 3417
/*
 * watch_mutex must be locked
 */
static void __rbd_unregister_watch(struct rbd_device *rbd_dev)
3418
{
3419 3420
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
	int ret;
3421

3422 3423
	rbd_assert(rbd_dev->watch_handle);
	dout("%s rbd_dev %p\n", __func__, rbd_dev);
3424

3425 3426 3427
	ret = ceph_osdc_unwatch(osdc, rbd_dev->watch_handle);
	if (ret)
		rbd_warn(rbd_dev, "failed to unwatch: %d", ret);
3428

3429
	rbd_dev->watch_handle = NULL;
3430 3431
}

3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450
static int rbd_register_watch(struct rbd_device *rbd_dev)
{
	int ret;

	mutex_lock(&rbd_dev->watch_mutex);
	rbd_assert(rbd_dev->watch_state == RBD_WATCH_STATE_UNREGISTERED);
	ret = __rbd_register_watch(rbd_dev);
	if (ret)
		goto out;

	rbd_dev->watch_state = RBD_WATCH_STATE_REGISTERED;
	rbd_dev->watch_cookie = rbd_dev->watch_handle->linger_id;

out:
	mutex_unlock(&rbd_dev->watch_mutex);
	return ret;
}

static void cancel_tasks_sync(struct rbd_device *rbd_dev)
3451
{
3452 3453 3454
	dout("%s rbd_dev %p\n", __func__, rbd_dev);

	cancel_delayed_work_sync(&rbd_dev->watch_dwork);
3455 3456 3457 3458
	cancel_work_sync(&rbd_dev->acquired_lock_work);
	cancel_work_sync(&rbd_dev->released_lock_work);
	cancel_delayed_work_sync(&rbd_dev->lock_dwork);
	cancel_work_sync(&rbd_dev->unlock_work);
3459 3460 3461 3462
}

static void rbd_unregister_watch(struct rbd_device *rbd_dev)
{
3463
	WARN_ON(waitqueue_active(&rbd_dev->lock_waitq));
3464 3465 3466 3467 3468 3469 3470
	cancel_tasks_sync(rbd_dev);

	mutex_lock(&rbd_dev->watch_mutex);
	if (rbd_dev->watch_state == RBD_WATCH_STATE_REGISTERED)
		__rbd_unregister_watch(rbd_dev);
	rbd_dev->watch_state = RBD_WATCH_STATE_UNREGISTERED;
	mutex_unlock(&rbd_dev->watch_mutex);
I
Ilya Dryomov 已提交
3471 3472

	ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc);
3473 3474
}

3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503
/*
 * lock_rwsem must be held for write
 */
static void rbd_reacquire_lock(struct rbd_device *rbd_dev)
{
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
	char cookie[32];
	int ret;

	WARN_ON(rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED);

	format_lock_cookie(rbd_dev, cookie);
	ret = ceph_cls_set_cookie(osdc, &rbd_dev->header_oid,
				  &rbd_dev->header_oloc, RBD_LOCK_NAME,
				  CEPH_CLS_LOCK_EXCLUSIVE, rbd_dev->lock_cookie,
				  RBD_LOCK_TAG, cookie);
	if (ret) {
		if (ret != -EOPNOTSUPP)
			rbd_warn(rbd_dev, "failed to update lock cookie: %d",
				 ret);

		/*
		 * Lock cookie cannot be updated on older OSDs, so do
		 * a manual release and queue an acquire.
		 */
		if (rbd_release_lock(rbd_dev))
			queue_delayed_work(rbd_dev->task_wq,
					   &rbd_dev->lock_dwork, 0);
	} else {
3504
		__rbd_lock(rbd_dev, cookie);
3505 3506 3507
	}
}

3508 3509 3510 3511 3512 3513 3514 3515 3516
static void rbd_reregister_watch(struct work_struct *work)
{
	struct rbd_device *rbd_dev = container_of(to_delayed_work(work),
					    struct rbd_device, watch_dwork);
	int ret;

	dout("%s rbd_dev %p\n", __func__, rbd_dev);

	mutex_lock(&rbd_dev->watch_mutex);
3517 3518
	if (rbd_dev->watch_state != RBD_WATCH_STATE_ERROR) {
		mutex_unlock(&rbd_dev->watch_mutex);
3519
		return;
3520
	}
3521 3522 3523 3524

	ret = __rbd_register_watch(rbd_dev);
	if (ret) {
		rbd_warn(rbd_dev, "failed to reregister watch: %d", ret);
3525
		if (ret == -EBLACKLISTED || ret == -ENOENT) {
3526
			set_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags);
3527
			wake_requests(rbd_dev, true);
3528
		} else {
3529 3530 3531
			queue_delayed_work(rbd_dev->task_wq,
					   &rbd_dev->watch_dwork,
					   RBD_RETRY_DELAY);
3532 3533
		}
		mutex_unlock(&rbd_dev->watch_mutex);
3534
		return;
3535 3536 3537 3538 3539 3540
	}

	rbd_dev->watch_state = RBD_WATCH_STATE_REGISTERED;
	rbd_dev->watch_cookie = rbd_dev->watch_handle->linger_id;
	mutex_unlock(&rbd_dev->watch_mutex);

3541 3542 3543 3544 3545
	down_write(&rbd_dev->lock_rwsem);
	if (rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED)
		rbd_reacquire_lock(rbd_dev);
	up_write(&rbd_dev->lock_rwsem);

3546 3547 3548 3549 3550
	ret = rbd_dev_refresh(rbd_dev);
	if (ret)
		rbd_warn(rbd_dev, "reregisteration refresh failed: %d", ret);
}

3551
/*
3552 3553
 * Synchronous osd object method call.  Returns the number of bytes
 * returned in the outbound buffer, or a negative error code.
3554 3555
 */
static int rbd_obj_method_sync(struct rbd_device *rbd_dev,
3556 3557
			     struct ceph_object_id *oid,
			     struct ceph_object_locator *oloc,
3558
			     const char *method_name,
3559
			     const void *outbound,
3560
			     size_t outbound_size,
3561
			     void *inbound,
3562
			     size_t inbound_size)
3563
{
3564 3565 3566
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
	struct page *req_page = NULL;
	struct page *reply_page;
3567 3568 3569
	int ret;

	/*
3570 3571 3572 3573 3574
	 * Method calls are ultimately read operations.  The result
	 * should placed into the inbound buffer provided.  They
	 * also supply outbound data--parameters for the object
	 * method.  Currently if this is present it will be a
	 * snapshot id.
3575
	 */
3576 3577 3578
	if (outbound) {
		if (outbound_size > PAGE_SIZE)
			return -E2BIG;
3579

3580 3581 3582
		req_page = alloc_page(GFP_KERNEL);
		if (!req_page)
			return -ENOMEM;
3583

3584
		memcpy(page_address(req_page), outbound, outbound_size);
3585
	}
3586

3587 3588 3589 3590 3591 3592
	reply_page = alloc_page(GFP_KERNEL);
	if (!reply_page) {
		if (req_page)
			__free_page(req_page);
		return -ENOMEM;
	}
3593

3594 3595 3596 3597 3598 3599 3600
	ret = ceph_osdc_call(osdc, oid, oloc, RBD_DRV_NAME, method_name,
			     CEPH_OSD_FLAG_READ, req_page, outbound_size,
			     reply_page, &inbound_size);
	if (!ret) {
		memcpy(inbound, page_address(reply_page), inbound_size);
		ret = inbound_size;
	}
3601

3602 3603 3604
	if (req_page)
		__free_page(req_page);
	__free_page(reply_page);
3605 3606 3607
	return ret;
}

3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 3620 3621 3622 3623 3624 3625 3626
/*
 * lock_rwsem must be held for read
 */
static void rbd_wait_state_locked(struct rbd_device *rbd_dev)
{
	DEFINE_WAIT(wait);

	do {
		/*
		 * Note the use of mod_delayed_work() in rbd_acquire_lock()
		 * and cancel_delayed_work() in wake_requests().
		 */
		dout("%s rbd_dev %p queueing lock_dwork\n", __func__, rbd_dev);
		queue_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork, 0);
		prepare_to_wait_exclusive(&rbd_dev->lock_waitq, &wait,
					  TASK_UNINTERRUPTIBLE);
		up_read(&rbd_dev->lock_rwsem);
		schedule();
		down_read(&rbd_dev->lock_rwsem);
3627 3628 3629
	} while (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED &&
		 !test_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags));

3630 3631 3632
	finish_wait(&rbd_dev->lock_waitq, &wait);
}

C
Christoph Hellwig 已提交
3633
static void rbd_queue_workfn(struct work_struct *work)
A
Alex Elder 已提交
3634
{
C
Christoph Hellwig 已提交
3635 3636
	struct request *rq = blk_mq_rq_from_pdu(work);
	struct rbd_device *rbd_dev = rq->q->queuedata;
I
Ilya Dryomov 已提交
3637
	struct rbd_img_request *img_request;
3638
	struct ceph_snap_context *snapc = NULL;
I
Ilya Dryomov 已提交
3639 3640
	u64 offset = (u64)blk_rq_pos(rq) << SECTOR_SHIFT;
	u64 length = blk_rq_bytes(rq);
G
Guangliang Zhao 已提交
3641
	enum obj_operation_type op_type;
3642
	u64 mapping_size;
I
Ilya Dryomov 已提交
3643
	bool must_be_locked;
A
Alex Elder 已提交
3644 3645
	int result;

3646 3647
	switch (req_op(rq)) {
	case REQ_OP_DISCARD:
3648
	case REQ_OP_WRITE_ZEROES:
3649
		op_type = OBJ_OP_DISCARD;
3650 3651
		break;
	case REQ_OP_WRITE:
G
Guangliang Zhao 已提交
3652
		op_type = OBJ_OP_WRITE;
3653 3654
		break;
	case REQ_OP_READ:
G
Guangliang Zhao 已提交
3655
		op_type = OBJ_OP_READ;
3656 3657 3658 3659 3660 3661
		break;
	default:
		dout("%s: non-fs request type %d\n", __func__, req_op(rq));
		result = -EIO;
		goto err;
	}
G
Guangliang Zhao 已提交
3662

I
Ilya Dryomov 已提交
3663
	/* Ignore/skip any zero-length requests */
A
Alex Elder 已提交
3664

I
Ilya Dryomov 已提交
3665 3666 3667 3668 3669
	if (!length) {
		dout("%s: zero-length request\n", __func__);
		result = 0;
		goto err_rq;
	}
A
Alex Elder 已提交
3670

3671 3672
	rbd_assert(op_type == OBJ_OP_READ ||
		   rbd_dev->spec->snap_id == CEPH_NOSNAP);
A
Alex Elder 已提交
3673

I
Ilya Dryomov 已提交
3674 3675 3676 3677 3678 3679 3680 3681 3682 3683 3684 3685
	/*
	 * Quit early if the mapped snapshot no longer exists.  It's
	 * still possible the snapshot will have disappeared by the
	 * time our request arrives at the osd, but there's no sense in
	 * sending it if we already know.
	 */
	if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) {
		dout("request for non-existent snapshot");
		rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP);
		result = -ENXIO;
		goto err_rq;
	}
A
Alex Elder 已提交
3686

I
Ilya Dryomov 已提交
3687 3688 3689 3690 3691 3692
	if (offset && length > U64_MAX - offset + 1) {
		rbd_warn(rbd_dev, "bad request range (%llu~%llu)", offset,
			 length);
		result = -EINVAL;
		goto err_rq;	/* Shouldn't happen */
	}
A
Alex Elder 已提交
3693

C
Christoph Hellwig 已提交
3694 3695
	blk_mq_start_request(rq);

3696 3697
	down_read(&rbd_dev->header_rwsem);
	mapping_size = rbd_dev->mapping.size;
G
Guangliang Zhao 已提交
3698
	if (op_type != OBJ_OP_READ) {
3699 3700 3701 3702 3703 3704
		snapc = rbd_dev->header.snapc;
		ceph_get_snap_context(snapc);
	}
	up_read(&rbd_dev->header_rwsem);

	if (offset + length > mapping_size) {
I
Ilya Dryomov 已提交
3705
		rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)", offset,
3706
			 length, mapping_size);
I
Ilya Dryomov 已提交
3707 3708 3709
		result = -EIO;
		goto err_rq;
	}
A
Alex Elder 已提交
3710

I
Ilya Dryomov 已提交
3711 3712 3713
	must_be_locked =
	    (rbd_dev->header.features & RBD_FEATURE_EXCLUSIVE_LOCK) &&
	    (op_type != OBJ_OP_READ || rbd_dev->opts->lock_on_read);
3714 3715
	if (must_be_locked) {
		down_read(&rbd_dev->lock_rwsem);
3716
		if (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED &&
I
Ilya Dryomov 已提交
3717 3718 3719 3720 3721 3722
		    !test_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags)) {
			if (rbd_dev->opts->exclusive) {
				rbd_warn(rbd_dev, "exclusive lock required");
				result = -EROFS;
				goto err_unlock;
			}
3723
			rbd_wait_state_locked(rbd_dev);
I
Ilya Dryomov 已提交
3724
		}
3725 3726 3727 3728
		if (test_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags)) {
			result = -EBLACKLISTED;
			goto err_unlock;
		}
3729 3730
	}

G
Guangliang Zhao 已提交
3731
	img_request = rbd_img_request_create(rbd_dev, offset, length, op_type,
3732
					     snapc);
I
Ilya Dryomov 已提交
3733 3734
	if (!img_request) {
		result = -ENOMEM;
3735
		goto err_unlock;
I
Ilya Dryomov 已提交
3736 3737
	}
	img_request->rq = rq;
3738
	snapc = NULL; /* img_request consumes a ref */
A
Alex Elder 已提交
3739

3740 3741 3742
	if (op_type == OBJ_OP_DISCARD)
		result = rbd_img_request_fill(img_request, OBJ_REQUEST_NODATA,
					      NULL);
3743 3744 3745 3746
	else {
		struct ceph_bio_iter bio_it = { .bio = rq->bio,
						.iter = rq->bio->bi_iter };

3747
		result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO,
3748 3749
					      &bio_it);
	}
I
Ilya Dryomov 已提交
3750 3751
	if (result)
		goto err_img_request;
A
Alex Elder 已提交
3752

I
Ilya Dryomov 已提交
3753 3754 3755
	result = rbd_img_request_submit(img_request);
	if (result)
		goto err_img_request;
A
Alex Elder 已提交
3756

3757 3758
	if (must_be_locked)
		up_read(&rbd_dev->lock_rwsem);
I
Ilya Dryomov 已提交
3759
	return;
A
Alex Elder 已提交
3760

I
Ilya Dryomov 已提交
3761 3762
err_img_request:
	rbd_img_request_put(img_request);
3763 3764 3765
err_unlock:
	if (must_be_locked)
		up_read(&rbd_dev->lock_rwsem);
I
Ilya Dryomov 已提交
3766 3767 3768
err_rq:
	if (result)
		rbd_warn(rbd_dev, "%s %llx at %llx result %d",
G
Guangliang Zhao 已提交
3769
			 obj_op_name(op_type), length, offset, result);
3770
	ceph_put_snap_context(snapc);
C
Christoph Hellwig 已提交
3771
err:
3772
	blk_mq_end_request(rq, errno_to_blk_status(result));
I
Ilya Dryomov 已提交
3773
}
A
Alex Elder 已提交
3774

3775
static blk_status_t rbd_queue_rq(struct blk_mq_hw_ctx *hctx,
C
Christoph Hellwig 已提交
3776
		const struct blk_mq_queue_data *bd)
I
Ilya Dryomov 已提交
3777
{
C
Christoph Hellwig 已提交
3778 3779
	struct request *rq = bd->rq;
	struct work_struct *work = blk_mq_rq_to_pdu(rq);
A
Alex Elder 已提交
3780

C
Christoph Hellwig 已提交
3781
	queue_work(rbd_wq, work);
3782
	return BLK_STS_OK;
A
Alex Elder 已提交
3783 3784
}

3785 3786
static void rbd_free_disk(struct rbd_device *rbd_dev)
{
3787 3788 3789
	blk_cleanup_queue(rbd_dev->disk->queue);
	blk_mq_free_tag_set(&rbd_dev->tag_set);
	put_disk(rbd_dev->disk);
3790
	rbd_dev->disk = NULL;
3791 3792
}

3793
static int rbd_obj_read_sync(struct rbd_device *rbd_dev,
3794 3795 3796
			     struct ceph_object_id *oid,
			     struct ceph_object_locator *oloc,
			     void *buf, int buf_len)
3797 3798

{
3799 3800 3801 3802
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
	struct ceph_osd_request *req;
	struct page **pages;
	int num_pages = calc_pages_for(0, buf_len);
3803 3804
	int ret;

3805 3806 3807
	req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_KERNEL);
	if (!req)
		return -ENOMEM;
3808

3809 3810 3811
	ceph_oid_copy(&req->r_base_oid, oid);
	ceph_oloc_copy(&req->r_base_oloc, oloc);
	req->r_flags = CEPH_OSD_FLAG_READ;
A
Alex Elder 已提交
3812

3813
	ret = ceph_osdc_alloc_messages(req, GFP_KERNEL);
3814
	if (ret)
3815
		goto out_req;
3816

3817 3818 3819 3820 3821
	pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
	if (IS_ERR(pages)) {
		ret = PTR_ERR(pages);
		goto out_req;
	}
3822

3823 3824 3825 3826 3827 3828 3829 3830
	osd_req_op_extent_init(req, 0, CEPH_OSD_OP_READ, 0, buf_len, 0, 0);
	osd_req_op_extent_osd_data_pages(req, 0, pages, buf_len, 0, false,
					 true);

	ceph_osdc_start_request(osdc, req, false);
	ret = ceph_osdc_wait_request(osdc, req);
	if (ret >= 0)
		ceph_copy_from_page_vector(pages, buf, 0, ret);
3831

3832 3833
out_req:
	ceph_osdc_put_request(req);
3834 3835 3836
	return ret;
}

3837
/*
A
Alex Elder 已提交
3838 3839 3840
 * Read the complete header for the given rbd device.  On successful
 * return, the rbd_dev->header field will contain up-to-date
 * information about the image.
3841
 */
3842
static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev)
3843
{
3844
	struct rbd_image_header_ondisk *ondisk = NULL;
3845
	u32 snap_count = 0;
3846 3847 3848
	u64 names_size = 0;
	u32 want_count;
	int ret;
3849

A
Alex Elder 已提交
3850
	/*
3851 3852 3853 3854 3855
	 * The complete header will include an array of its 64-bit
	 * snapshot ids, followed by the names of those snapshots as
	 * a contiguous block of NUL-terminated strings.  Note that
	 * the number of snapshots could change by the time we read
	 * it in, in which case we re-read it.
A
Alex Elder 已提交
3856
	 */
3857 3858 3859 3860 3861 3862 3863 3864 3865 3866
	do {
		size_t size;

		kfree(ondisk);

		size = sizeof (*ondisk);
		size += snap_count * sizeof (struct rbd_image_snap_ondisk);
		size += names_size;
		ondisk = kmalloc(size, GFP_KERNEL);
		if (!ondisk)
A
Alex Elder 已提交
3867
			return -ENOMEM;
3868

3869 3870
		ret = rbd_obj_read_sync(rbd_dev, &rbd_dev->header_oid,
					&rbd_dev->header_oloc, ondisk, size);
3871
		if (ret < 0)
A
Alex Elder 已提交
3872
			goto out;
A
Alex Elder 已提交
3873
		if ((size_t)ret < size) {
3874
			ret = -ENXIO;
A
Alex Elder 已提交
3875 3876
			rbd_warn(rbd_dev, "short header read (want %zd got %d)",
				size, ret);
A
Alex Elder 已提交
3877
			goto out;
3878 3879 3880
		}
		if (!rbd_dev_ondisk_valid(ondisk)) {
			ret = -ENXIO;
A
Alex Elder 已提交
3881
			rbd_warn(rbd_dev, "invalid header");
A
Alex Elder 已提交
3882
			goto out;
3883
		}
3884

3885 3886 3887 3888
		names_size = le64_to_cpu(ondisk->snap_names_len);
		want_count = snap_count;
		snap_count = le32_to_cpu(ondisk->snap_count);
	} while (snap_count != want_count);
A
Alex Elder 已提交
3889

A
Alex Elder 已提交
3890 3891
	ret = rbd_header_from_disk(rbd_dev, ondisk);
out:
3892 3893 3894
	kfree(ondisk);

	return ret;
3895 3896
}

3897 3898 3899 3900 3901 3902 3903 3904 3905 3906 3907 3908 3909 3910 3911 3912 3913 3914 3915
/*
 * Clear the rbd device's EXISTS flag if the snapshot it's mapped to
 * has disappeared from the (just updated) snapshot context.
 */
static void rbd_exists_validate(struct rbd_device *rbd_dev)
{
	u64 snap_id;

	if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags))
		return;

	snap_id = rbd_dev->spec->snap_id;
	if (snap_id == CEPH_NOSNAP)
		return;

	if (rbd_dev_snap_index(rbd_dev, snap_id) == BAD_SNAP_INDEX)
		clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
}

3916 3917 3918 3919 3920
static void rbd_dev_update_size(struct rbd_device *rbd_dev)
{
	sector_t size;

	/*
I
Ilya Dryomov 已提交
3921 3922 3923
	 * If EXISTS is not set, rbd_dev->disk may be NULL, so don't
	 * try to update its size.  If REMOVING is set, updating size
	 * is just useless work since the device can't be opened.
3924
	 */
I
Ilya Dryomov 已提交
3925 3926
	if (test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags) &&
	    !test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) {
3927 3928 3929 3930 3931 3932 3933
		size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE;
		dout("setting size to %llu sectors", (unsigned long long)size);
		set_capacity(rbd_dev->disk, size);
		revalidate_disk(rbd_dev->disk);
	}
}

A
Alex Elder 已提交
3934
static int rbd_dev_refresh(struct rbd_device *rbd_dev)
A
Alex Elder 已提交
3935
{
3936
	u64 mapping_size;
A
Alex Elder 已提交
3937 3938
	int ret;

3939
	down_write(&rbd_dev->header_rwsem);
3940
	mapping_size = rbd_dev->mapping.size;
3941 3942

	ret = rbd_dev_header_info(rbd_dev);
3943
	if (ret)
3944
		goto out;
3945

3946 3947 3948 3949 3950 3951 3952
	/*
	 * If there is a parent, see if it has disappeared due to the
	 * mapped image getting flattened.
	 */
	if (rbd_dev->parent) {
		ret = rbd_dev_v2_parent_info(rbd_dev);
		if (ret)
3953
			goto out;
3954 3955
	}

3956
	if (rbd_dev->spec->snap_id == CEPH_NOSNAP) {
3957
		rbd_dev->mapping.size = rbd_dev->header.image_size;
3958 3959 3960 3961
	} else {
		/* validate mapped snapshot's EXISTS flag */
		rbd_exists_validate(rbd_dev);
	}
3962

3963
out:
3964
	up_write(&rbd_dev->header_rwsem);
3965
	if (!ret && mapping_size != rbd_dev->mapping.size)
3966
		rbd_dev_update_size(rbd_dev);
A
Alex Elder 已提交
3967

3968
	return ret;
A
Alex Elder 已提交
3969 3970
}

3971 3972
static int rbd_init_request(struct blk_mq_tag_set *set, struct request *rq,
		unsigned int hctx_idx, unsigned int numa_node)
C
Christoph Hellwig 已提交
3973 3974 3975 3976 3977 3978 3979
{
	struct work_struct *work = blk_mq_rq_to_pdu(rq);

	INIT_WORK(work, rbd_queue_workfn);
	return 0;
}

3980
static const struct blk_mq_ops rbd_mq_ops = {
C
Christoph Hellwig 已提交
3981 3982 3983 3984
	.queue_rq	= rbd_queue_rq,
	.init_request	= rbd_init_request,
};

3985 3986 3987 3988
static int rbd_init_disk(struct rbd_device *rbd_dev)
{
	struct gendisk *disk;
	struct request_queue *q;
A
Alex Elder 已提交
3989
	u64 segment_size;
C
Christoph Hellwig 已提交
3990
	int err;
3991 3992

	/* create gendisk info */
3993 3994 3995
	disk = alloc_disk(single_major ?
			  (1 << RBD_SINGLE_MAJOR_PART_SHIFT) :
			  RBD_MINORS_PER_MAJOR);
3996
	if (!disk)
A
Alex Elder 已提交
3997
		return -ENOMEM;
3998

A
Alex Elder 已提交
3999
	snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
A
Alex Elder 已提交
4000
		 rbd_dev->dev_id);
4001
	disk->major = rbd_dev->major;
4002
	disk->first_minor = rbd_dev->minor;
4003 4004
	if (single_major)
		disk->flags |= GENHD_FL_EXT_DEVT;
4005 4006 4007
	disk->fops = &rbd_bd_ops;
	disk->private_data = rbd_dev;

C
Christoph Hellwig 已提交
4008 4009
	memset(&rbd_dev->tag_set, 0, sizeof(rbd_dev->tag_set));
	rbd_dev->tag_set.ops = &rbd_mq_ops;
I
Ilya Dryomov 已提交
4010
	rbd_dev->tag_set.queue_depth = rbd_dev->opts->queue_depth;
C
Christoph Hellwig 已提交
4011
	rbd_dev->tag_set.numa_node = NUMA_NO_NODE;
I
Ilya Dryomov 已提交
4012
	rbd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
C
Christoph Hellwig 已提交
4013 4014 4015 4016 4017
	rbd_dev->tag_set.nr_hw_queues = 1;
	rbd_dev->tag_set.cmd_size = sizeof(struct work_struct);

	err = blk_mq_alloc_tag_set(&rbd_dev->tag_set);
	if (err)
4018
		goto out_disk;
4019

C
Christoph Hellwig 已提交
4020 4021 4022 4023 4024 4025
	q = blk_mq_init_queue(&rbd_dev->tag_set);
	if (IS_ERR(q)) {
		err = PTR_ERR(q);
		goto out_tag_set;
	}

4026 4027
	queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
	/* QUEUE_FLAG_ADD_RANDOM is off by default for blk-mq */
A
Alex Elder 已提交
4028

4029
	/* set io sizes to object size */
A
Alex Elder 已提交
4030 4031
	segment_size = rbd_obj_bytes(&rbd_dev->header);
	blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE);
I
Ilya Dryomov 已提交
4032
	q->limits.max_sectors = queue_max_hw_sectors(q);
4033
	blk_queue_max_segments(q, USHRT_MAX);
4034
	blk_queue_max_segment_size(q, UINT_MAX);
A
Alex Elder 已提交
4035 4036
	blk_queue_io_min(q, segment_size);
	blk_queue_io_opt(q, segment_size);
4037

4038 4039 4040
	/* enable the discard support */
	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
	q->limits.discard_granularity = segment_size;
4041
	blk_queue_max_discard_sectors(q, segment_size / SECTOR_SIZE);
4042
	blk_queue_max_write_zeroes_sectors(q, segment_size / SECTOR_SIZE);
4043

4044
	if (!ceph_test_opt(rbd_dev->rbd_client->client, NOCRC))
4045
		q->backing_dev_info->capabilities |= BDI_CAP_STABLE_WRITES;
4046

4047 4048 4049 4050 4051
	/*
	 * disk_release() expects a queue ref from add_disk() and will
	 * put it.  Hold an extra ref until add_disk() is called.
	 */
	WARN_ON(!blk_get_queue(q));
4052 4053 4054 4055 4056 4057
	disk->queue = q;
	q->queuedata = rbd_dev;

	rbd_dev->disk = disk;

	return 0;
C
Christoph Hellwig 已提交
4058 4059
out_tag_set:
	blk_mq_free_tag_set(&rbd_dev->tag_set);
4060 4061
out_disk:
	put_disk(disk);
C
Christoph Hellwig 已提交
4062
	return err;
4063 4064
}

4065 4066 4067 4068
/*
  sysfs
*/

A
Alex Elder 已提交
4069 4070 4071 4072 4073
static struct rbd_device *dev_to_rbd_dev(struct device *dev)
{
	return container_of(dev, struct rbd_device, dev);
}

4074 4075 4076
static ssize_t rbd_size_show(struct device *dev,
			     struct device_attribute *attr, char *buf)
{
A
Alex Elder 已提交
4077
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4078

A
Alex Elder 已提交
4079 4080
	return sprintf(buf, "%llu\n",
		(unsigned long long)rbd_dev->mapping.size);
4081 4082
}

A
Alex Elder 已提交
4083 4084 4085 4086 4087 4088 4089 4090 4091 4092
/*
 * Note this shows the features for whatever's mapped, which is not
 * necessarily the base image.
 */
static ssize_t rbd_features_show(struct device *dev,
			     struct device_attribute *attr, char *buf)
{
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);

	return sprintf(buf, "0x%016llx\n",
A
Alex Elder 已提交
4093
			(unsigned long long)rbd_dev->mapping.features);
A
Alex Elder 已提交
4094 4095
}

4096 4097 4098
static ssize_t rbd_major_show(struct device *dev,
			      struct device_attribute *attr, char *buf)
{
A
Alex Elder 已提交
4099
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4100

A
Alex Elder 已提交
4101 4102 4103 4104
	if (rbd_dev->major)
		return sprintf(buf, "%d\n", rbd_dev->major);

	return sprintf(buf, "(none)\n");
4105 4106 4107 4108 4109 4110
}

static ssize_t rbd_minor_show(struct device *dev,
			      struct device_attribute *attr, char *buf)
{
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
A
Alex Elder 已提交
4111

4112
	return sprintf(buf, "%d\n", rbd_dev->minor);
4113 4114
}

4115 4116 4117 4118 4119 4120 4121 4122 4123 4124 4125
static ssize_t rbd_client_addr_show(struct device *dev,
				    struct device_attribute *attr, char *buf)
{
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
	struct ceph_entity_addr *client_addr =
	    ceph_client_addr(rbd_dev->rbd_client->client);

	return sprintf(buf, "%pISpc/%u\n", &client_addr->in_addr,
		       le32_to_cpu(client_addr->nonce));
}

4126 4127
static ssize_t rbd_client_id_show(struct device *dev,
				  struct device_attribute *attr, char *buf)
4128
{
A
Alex Elder 已提交
4129
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4130

4131
	return sprintf(buf, "client%lld\n",
4132
		       ceph_client_gid(rbd_dev->rbd_client->client));
4133 4134
}

4135 4136 4137 4138 4139 4140 4141 4142
static ssize_t rbd_cluster_fsid_show(struct device *dev,
				     struct device_attribute *attr, char *buf)
{
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);

	return sprintf(buf, "%pU\n", &rbd_dev->rbd_client->client->fsid);
}

4143 4144 4145 4146 4147 4148
static ssize_t rbd_config_info_show(struct device *dev,
				    struct device_attribute *attr, char *buf)
{
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);

	return sprintf(buf, "%s\n", rbd_dev->config_info);
4149 4150
}

4151 4152
static ssize_t rbd_pool_show(struct device *dev,
			     struct device_attribute *attr, char *buf)
4153
{
A
Alex Elder 已提交
4154
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4155

4156
	return sprintf(buf, "%s\n", rbd_dev->spec->pool_name);
4157 4158
}

4159 4160 4161 4162 4163
static ssize_t rbd_pool_id_show(struct device *dev,
			     struct device_attribute *attr, char *buf)
{
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);

4164
	return sprintf(buf, "%llu\n",
A
Alex Elder 已提交
4165
			(unsigned long long) rbd_dev->spec->pool_id);
4166 4167
}

4168 4169 4170
static ssize_t rbd_name_show(struct device *dev,
			     struct device_attribute *attr, char *buf)
{
A
Alex Elder 已提交
4171
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4172

A
Alex Elder 已提交
4173 4174 4175 4176
	if (rbd_dev->spec->image_name)
		return sprintf(buf, "%s\n", rbd_dev->spec->image_name);

	return sprintf(buf, "(unknown)\n");
4177 4178
}

A
Alex Elder 已提交
4179 4180 4181 4182 4183
static ssize_t rbd_image_id_show(struct device *dev,
			     struct device_attribute *attr, char *buf)
{
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);

4184
	return sprintf(buf, "%s\n", rbd_dev->spec->image_id);
A
Alex Elder 已提交
4185 4186
}

A
Alex Elder 已提交
4187 4188 4189 4190
/*
 * Shows the name of the currently-mapped snapshot (or
 * RBD_SNAP_HEAD_NAME for the base image).
 */
4191 4192 4193 4194
static ssize_t rbd_snap_show(struct device *dev,
			     struct device_attribute *attr,
			     char *buf)
{
A
Alex Elder 已提交
4195
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4196

4197
	return sprintf(buf, "%s\n", rbd_dev->spec->snap_name);
4198 4199
}

4200 4201 4202 4203 4204 4205 4206 4207
static ssize_t rbd_snap_id_show(struct device *dev,
				struct device_attribute *attr, char *buf)
{
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);

	return sprintf(buf, "%llu\n", rbd_dev->spec->snap_id);
}

4208
/*
4209 4210 4211
 * For a v2 image, shows the chain of parent images, separated by empty
 * lines.  For v1 images or if there is no parent, shows "(no parent
 * image)".
4212 4213
 */
static ssize_t rbd_parent_show(struct device *dev,
4214 4215
			       struct device_attribute *attr,
			       char *buf)
4216 4217
{
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4218
	ssize_t count = 0;
4219

4220
	if (!rbd_dev->parent)
4221 4222
		return sprintf(buf, "(no parent image)\n");

4223 4224 4225 4226 4227 4228 4229 4230 4231 4232 4233 4234 4235 4236 4237 4238
	for ( ; rbd_dev->parent; rbd_dev = rbd_dev->parent) {
		struct rbd_spec *spec = rbd_dev->parent_spec;

		count += sprintf(&buf[count], "%s"
			    "pool_id %llu\npool_name %s\n"
			    "image_id %s\nimage_name %s\n"
			    "snap_id %llu\nsnap_name %s\n"
			    "overlap %llu\n",
			    !count ? "" : "\n", /* first? */
			    spec->pool_id, spec->pool_name,
			    spec->image_id, spec->image_name ?: "(unknown)",
			    spec->snap_id, spec->snap_name,
			    rbd_dev->parent_overlap);
	}

	return count;
4239 4240
}

4241 4242 4243 4244 4245
static ssize_t rbd_image_refresh(struct device *dev,
				 struct device_attribute *attr,
				 const char *buf,
				 size_t size)
{
A
Alex Elder 已提交
4246
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4247
	int ret;
4248

A
Alex Elder 已提交
4249
	ret = rbd_dev_refresh(rbd_dev);
4250
	if (ret)
4251
		return ret;
4252

4253
	return size;
4254
}
4255

4256
static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
A
Alex Elder 已提交
4257
static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL);
4258
static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
4259
static DEVICE_ATTR(minor, S_IRUGO, rbd_minor_show, NULL);
4260
static DEVICE_ATTR(client_addr, S_IRUGO, rbd_client_addr_show, NULL);
4261
static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
4262
static DEVICE_ATTR(cluster_fsid, S_IRUGO, rbd_cluster_fsid_show, NULL);
4263
static DEVICE_ATTR(config_info, S_IRUSR, rbd_config_info_show, NULL);
4264
static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
4265
static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL);
4266
static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
A
Alex Elder 已提交
4267
static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL);
4268 4269
static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
4270
static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL);
4271
static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL);
4272 4273 4274

static struct attribute *rbd_attrs[] = {
	&dev_attr_size.attr,
A
Alex Elder 已提交
4275
	&dev_attr_features.attr,
4276
	&dev_attr_major.attr,
4277
	&dev_attr_minor.attr,
4278
	&dev_attr_client_addr.attr,
4279
	&dev_attr_client_id.attr,
4280
	&dev_attr_cluster_fsid.attr,
4281
	&dev_attr_config_info.attr,
4282
	&dev_attr_pool.attr,
4283
	&dev_attr_pool_id.attr,
4284
	&dev_attr_name.attr,
A
Alex Elder 已提交
4285
	&dev_attr_image_id.attr,
4286
	&dev_attr_current_snap.attr,
4287
	&dev_attr_snap_id.attr,
4288
	&dev_attr_parent.attr,
4289 4290 4291 4292 4293 4294 4295 4296 4297 4298 4299 4300 4301
	&dev_attr_refresh.attr,
	NULL
};

static struct attribute_group rbd_attr_group = {
	.attrs = rbd_attrs,
};

static const struct attribute_group *rbd_attr_groups[] = {
	&rbd_attr_group,
	NULL
};

4302
static void rbd_dev_release(struct device *dev);
4303

4304
static const struct device_type rbd_device_type = {
4305 4306
	.name		= "rbd",
	.groups		= rbd_attr_groups,
4307
	.release	= rbd_dev_release,
4308 4309
};

4310 4311 4312 4313 4314 4315 4316 4317 4318 4319 4320 4321 4322 4323 4324 4325 4326 4327 4328 4329 4330
static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec)
{
	kref_get(&spec->kref);

	return spec;
}

static void rbd_spec_free(struct kref *kref);
static void rbd_spec_put(struct rbd_spec *spec)
{
	if (spec)
		kref_put(&spec->kref, rbd_spec_free);
}

static struct rbd_spec *rbd_spec_alloc(void)
{
	struct rbd_spec *spec;

	spec = kzalloc(sizeof (*spec), GFP_KERNEL);
	if (!spec)
		return NULL;
4331 4332 4333

	spec->pool_id = CEPH_NOPOOL;
	spec->snap_id = CEPH_NOSNAP;
4334 4335 4336 4337 4338 4339 4340 4341 4342 4343 4344 4345 4346 4347 4348 4349
	kref_init(&spec->kref);

	return spec;
}

static void rbd_spec_free(struct kref *kref)
{
	struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref);

	kfree(spec->pool_name);
	kfree(spec->image_id);
	kfree(spec->image_name);
	kfree(spec->snap_name);
	kfree(spec);
}

4350
static void rbd_dev_free(struct rbd_device *rbd_dev)
4351
{
4352
	WARN_ON(rbd_dev->watch_state != RBD_WATCH_STATE_UNREGISTERED);
4353
	WARN_ON(rbd_dev->lock_state != RBD_LOCK_STATE_UNLOCKED);
4354

4355
	ceph_oid_destroy(&rbd_dev->header_oid);
4356
	ceph_oloc_destroy(&rbd_dev->header_oloc);
4357
	kfree(rbd_dev->config_info);
4358

4359 4360 4361 4362
	rbd_put_client(rbd_dev->rbd_client);
	rbd_spec_put(rbd_dev->spec);
	kfree(rbd_dev->opts);
	kfree(rbd_dev);
4363 4364 4365 4366 4367 4368 4369 4370 4371 4372 4373 4374 4375
}

static void rbd_dev_release(struct device *dev)
{
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
	bool need_put = !!rbd_dev->opts;

	if (need_put) {
		destroy_workqueue(rbd_dev->task_wq);
		ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id);
	}

	rbd_dev_free(rbd_dev);
4376 4377 4378 4379 4380 4381 4382 4383 4384 4385

	/*
	 * This is racy, but way better than putting module outside of
	 * the release callback.  The race window is pretty small, so
	 * doing something similar to dm (dm-builtin.c) is overkill.
	 */
	if (need_put)
		module_put(THIS_MODULE);
}

4386 4387
static struct rbd_device *__rbd_dev_create(struct rbd_client *rbdc,
					   struct rbd_spec *spec)
4388 4389 4390
{
	struct rbd_device *rbd_dev;

4391
	rbd_dev = kzalloc(sizeof(*rbd_dev), GFP_KERNEL);
4392 4393 4394 4395 4396 4397 4398
	if (!rbd_dev)
		return NULL;

	spin_lock_init(&rbd_dev->lock);
	INIT_LIST_HEAD(&rbd_dev->node);
	init_rwsem(&rbd_dev->header_rwsem);

4399
	rbd_dev->header.data_pool_id = CEPH_NOPOOL;
4400
	ceph_oid_init(&rbd_dev->header_oid);
4401
	rbd_dev->header_oloc.pool = spec->pool_id;
4402

4403 4404 4405 4406
	mutex_init(&rbd_dev->watch_mutex);
	rbd_dev->watch_state = RBD_WATCH_STATE_UNREGISTERED;
	INIT_DELAYED_WORK(&rbd_dev->watch_dwork, rbd_reregister_watch);

4407 4408 4409 4410 4411 4412 4413 4414
	init_rwsem(&rbd_dev->lock_rwsem);
	rbd_dev->lock_state = RBD_LOCK_STATE_UNLOCKED;
	INIT_WORK(&rbd_dev->acquired_lock_work, rbd_notify_acquired_lock);
	INIT_WORK(&rbd_dev->released_lock_work, rbd_notify_released_lock);
	INIT_DELAYED_WORK(&rbd_dev->lock_dwork, rbd_acquire_lock);
	INIT_WORK(&rbd_dev->unlock_work, rbd_release_lock_work);
	init_waitqueue_head(&rbd_dev->lock_waitq);

4415 4416 4417 4418 4419
	rbd_dev->dev.bus = &rbd_bus_type;
	rbd_dev->dev.type = &rbd_device_type;
	rbd_dev->dev.parent = &rbd_root_dev;
	device_initialize(&rbd_dev->dev);

4420
	rbd_dev->rbd_client = rbdc;
4421
	rbd_dev->spec = spec;
4422

4423 4424 4425 4426 4427 4428 4429 4430 4431 4432 4433 4434 4435 4436 4437 4438 4439 4440 4441 4442 4443 4444 4445 4446 4447 4448 4449 4450 4451 4452
	return rbd_dev;
}

/*
 * Create a mapping rbd_dev.
 */
static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc,
					 struct rbd_spec *spec,
					 struct rbd_options *opts)
{
	struct rbd_device *rbd_dev;

	rbd_dev = __rbd_dev_create(rbdc, spec);
	if (!rbd_dev)
		return NULL;

	rbd_dev->opts = opts;

	/* get an id and fill in device name */
	rbd_dev->dev_id = ida_simple_get(&rbd_dev_id_ida, 0,
					 minor_to_rbd_dev_id(1 << MINORBITS),
					 GFP_KERNEL);
	if (rbd_dev->dev_id < 0)
		goto fail_rbd_dev;

	sprintf(rbd_dev->name, RBD_DRV_NAME "%d", rbd_dev->dev_id);
	rbd_dev->task_wq = alloc_ordered_workqueue("%s-tasks", WQ_MEM_RECLAIM,
						   rbd_dev->name);
	if (!rbd_dev->task_wq)
		goto fail_dev_id;
4453

4454 4455
	/* we have a ref from do_rbd_add() */
	__module_get(THIS_MODULE);
4456

4457
	dout("%s rbd_dev %p dev_id %d\n", __func__, rbd_dev, rbd_dev->dev_id);
4458
	return rbd_dev;
4459 4460 4461 4462 4463 4464

fail_dev_id:
	ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id);
fail_rbd_dev:
	rbd_dev_free(rbd_dev);
	return NULL;
4465 4466 4467 4468
}

static void rbd_dev_destroy(struct rbd_device *rbd_dev)
{
4469 4470
	if (rbd_dev)
		put_device(&rbd_dev->dev);
4471 4472
}

4473 4474 4475 4476 4477 4478 4479 4480 4481 4482 4483 4484 4485 4486 4487
/*
 * Get the size and object order for an image snapshot, or if
 * snap_id is CEPH_NOSNAP, gets this information for the base
 * image.
 */
static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
				u8 *order, u64 *snap_size)
{
	__le64 snapid = cpu_to_le64(snap_id);
	int ret;
	struct {
		u8 order;
		__le64 size;
	} __attribute__ ((packed)) size_buf = { 0 };

4488 4489 4490 4491
	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
				  &rbd_dev->header_oloc, "get_size",
				  &snapid, sizeof(snapid),
				  &size_buf, sizeof(size_buf));
4492
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4493 4494
	if (ret < 0)
		return ret;
4495 4496
	if (ret < sizeof (size_buf))
		return -ERANGE;
4497

J
Josh Durgin 已提交
4498
	if (order) {
4499
		*order = size_buf.order;
J
Josh Durgin 已提交
4500 4501
		dout("  order %u", (unsigned int)*order);
	}
4502 4503
	*snap_size = le64_to_cpu(size_buf.size);

J
Josh Durgin 已提交
4504 4505
	dout("  snap_id 0x%016llx snap_size = %llu\n",
		(unsigned long long)snap_id,
4506
		(unsigned long long)*snap_size);
4507 4508 4509 4510 4511 4512 4513 4514 4515 4516 4517

	return 0;
}

static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev)
{
	return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP,
					&rbd_dev->header.obj_order,
					&rbd_dev->header.image_size);
}

4518 4519 4520 4521 4522 4523 4524 4525 4526 4527
static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev)
{
	void *reply_buf;
	int ret;
	void *p;

	reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL);
	if (!reply_buf)
		return -ENOMEM;

4528 4529 4530
	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
				  &rbd_dev->header_oloc, "get_object_prefix",
				  NULL, 0, reply_buf, RBD_OBJ_PREFIX_LEN_MAX);
4531
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4532 4533 4534 4535 4536
	if (ret < 0)
		goto out;

	p = reply_buf;
	rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p,
4537 4538
						p + ret, NULL, GFP_NOIO);
	ret = 0;
4539 4540 4541 4542 4543 4544 4545 4546 4547 4548 4549 4550 4551

	if (IS_ERR(rbd_dev->header.object_prefix)) {
		ret = PTR_ERR(rbd_dev->header.object_prefix);
		rbd_dev->header.object_prefix = NULL;
	} else {
		dout("  object_prefix = %s\n", rbd_dev->header.object_prefix);
	}
out:
	kfree(reply_buf);

	return ret;
}

4552 4553 4554 4555 4556 4557 4558
static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
		u64 *snap_features)
{
	__le64 snapid = cpu_to_le64(snap_id);
	struct {
		__le64 features;
		__le64 incompat;
4559
	} __attribute__ ((packed)) features_buf = { 0 };
4560
	u64 unsup;
4561 4562
	int ret;

4563 4564 4565 4566
	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
				  &rbd_dev->header_oloc, "get_features",
				  &snapid, sizeof(snapid),
				  &features_buf, sizeof(features_buf));
4567
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4568 4569
	if (ret < 0)
		return ret;
4570 4571
	if (ret < sizeof (features_buf))
		return -ERANGE;
A
Alex Elder 已提交
4572

4573 4574 4575 4576
	unsup = le64_to_cpu(features_buf.incompat) & ~RBD_FEATURES_SUPPORTED;
	if (unsup) {
		rbd_warn(rbd_dev, "image uses unsupported features: 0x%llx",
			 unsup);
A
Alex Elder 已提交
4577
		return -ENXIO;
4578
	}
A
Alex Elder 已提交
4579

4580 4581 4582
	*snap_features = le64_to_cpu(features_buf.features);

	dout("  snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
4583 4584 4585
		(unsigned long long)snap_id,
		(unsigned long long)*snap_features,
		(unsigned long long)le64_to_cpu(features_buf.incompat));
4586 4587 4588 4589 4590 4591 4592 4593 4594 4595

	return 0;
}

static int rbd_dev_v2_features(struct rbd_device *rbd_dev)
{
	return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP,
						&rbd_dev->header.features);
}

4596 4597 4598 4599 4600 4601 4602 4603
static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev)
{
	struct rbd_spec *parent_spec;
	size_t size;
	void *reply_buf = NULL;
	__le64 snapid;
	void *p;
	void *end;
A
Alex Elder 已提交
4604
	u64 pool_id;
4605
	char *image_id;
4606
	u64 snap_id;
4607 4608 4609 4610 4611 4612 4613 4614 4615 4616 4617 4618 4619 4620 4621 4622 4623
	u64 overlap;
	int ret;

	parent_spec = rbd_spec_alloc();
	if (!parent_spec)
		return -ENOMEM;

	size = sizeof (__le64) +				/* pool_id */
		sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX +	/* image_id */
		sizeof (__le64) +				/* snap_id */
		sizeof (__le64);				/* overlap */
	reply_buf = kmalloc(size, GFP_KERNEL);
	if (!reply_buf) {
		ret = -ENOMEM;
		goto out_err;
	}

4624
	snapid = cpu_to_le64(rbd_dev->spec->snap_id);
4625 4626 4627
	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
				  &rbd_dev->header_oloc, "get_parent",
				  &snapid, sizeof(snapid), reply_buf, size);
4628
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4629 4630 4631 4632
	if (ret < 0)
		goto out_err;

	p = reply_buf;
4633 4634
	end = reply_buf + ret;
	ret = -ERANGE;
A
Alex Elder 已提交
4635
	ceph_decode_64_safe(&p, end, pool_id, out_err);
4636 4637 4638 4639 4640 4641 4642 4643 4644 4645 4646 4647 4648 4649 4650 4651 4652
	if (pool_id == CEPH_NOPOOL) {
		/*
		 * Either the parent never existed, or we have
		 * record of it but the image got flattened so it no
		 * longer has a parent.  When the parent of a
		 * layered image disappears we immediately set the
		 * overlap to 0.  The effect of this is that all new
		 * requests will be treated as if the image had no
		 * parent.
		 */
		if (rbd_dev->parent_overlap) {
			rbd_dev->parent_overlap = 0;
			rbd_dev_parent_put(rbd_dev);
			pr_info("%s: clone image has been flattened\n",
				rbd_dev->disk->disk_name);
		}

4653
		goto out;	/* No parent?  No problem. */
4654
	}
4655

4656 4657 4658
	/* The ceph file layout needs to fit pool id in 32 bits */

	ret = -EIO;
A
Alex Elder 已提交
4659
	if (pool_id > (u64)U32_MAX) {
4660
		rbd_warn(NULL, "parent pool id too large (%llu > %u)",
A
Alex Elder 已提交
4661
			(unsigned long long)pool_id, U32_MAX);
4662
		goto out_err;
A
Alex Elder 已提交
4663
	}
4664

A
Alex Elder 已提交
4665
	image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
4666 4667 4668 4669
	if (IS_ERR(image_id)) {
		ret = PTR_ERR(image_id);
		goto out_err;
	}
4670
	ceph_decode_64_safe(&p, end, snap_id, out_err);
4671 4672
	ceph_decode_64_safe(&p, end, overlap, out_err);

4673 4674 4675 4676 4677 4678 4679 4680 4681
	/*
	 * The parent won't change (except when the clone is
	 * flattened, already handled that).  So we only need to
	 * record the parent spec we have not already done so.
	 */
	if (!rbd_dev->parent_spec) {
		parent_spec->pool_id = pool_id;
		parent_spec->image_id = image_id;
		parent_spec->snap_id = snap_id;
A
Alex Elder 已提交
4682 4683
		rbd_dev->parent_spec = parent_spec;
		parent_spec = NULL;	/* rbd_dev now owns this */
4684 4685
	} else {
		kfree(image_id);
4686 4687 4688
	}

	/*
4689 4690
	 * We always update the parent overlap.  If it's zero we issue
	 * a warning, as we will proceed as if there was no parent.
4691 4692 4693
	 */
	if (!overlap) {
		if (parent_spec) {
4694 4695 4696 4697
			/* refresh, careful to warn just once */
			if (rbd_dev->parent_overlap)
				rbd_warn(rbd_dev,
				    "clone now standalone (overlap became 0)");
4698
		} else {
4699 4700
			/* initial probe */
			rbd_warn(rbd_dev, "clone is standalone (overlap 0)");
4701
		}
A
Alex Elder 已提交
4702
	}
4703 4704
	rbd_dev->parent_overlap = overlap;

4705 4706 4707 4708 4709 4710 4711 4712 4713
out:
	ret = 0;
out_err:
	kfree(reply_buf);
	rbd_spec_put(parent_spec);

	return ret;
}

4714 4715 4716 4717 4718 4719 4720 4721 4722 4723 4724 4725 4726
static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev)
{
	struct {
		__le64 stripe_unit;
		__le64 stripe_count;
	} __attribute__ ((packed)) striping_info_buf = { 0 };
	size_t size = sizeof (striping_info_buf);
	void *p;
	u64 obj_size;
	u64 stripe_unit;
	u64 stripe_count;
	int ret;

4727 4728 4729
	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
				&rbd_dev->header_oloc, "get_stripe_unit_count",
				NULL, 0, &striping_info_buf, size);
4730 4731 4732 4733 4734 4735 4736 4737 4738 4739 4740 4741 4742
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
	if (ret < 0)
		return ret;
	if (ret < size)
		return -ERANGE;

	/*
	 * We don't actually support the "fancy striping" feature
	 * (STRIPINGV2) yet, but if the striping sizes are the
	 * defaults the behavior is the same as before.  So find
	 * out, and only fail if the image has non-default values.
	 */
	ret = -EINVAL;
I
Ilya Dryomov 已提交
4743
	obj_size = rbd_obj_bytes(&rbd_dev->header);
4744 4745 4746 4747 4748 4749 4750 4751 4752 4753 4754 4755 4756 4757
	p = &striping_info_buf;
	stripe_unit = ceph_decode_64(&p);
	if (stripe_unit != obj_size) {
		rbd_warn(rbd_dev, "unsupported stripe unit "
				"(got %llu want %llu)",
				stripe_unit, obj_size);
		return -EINVAL;
	}
	stripe_count = ceph_decode_64(&p);
	if (stripe_count != 1) {
		rbd_warn(rbd_dev, "unsupported stripe count "
				"(got %llu want 1)", stripe_count);
		return -EINVAL;
	}
4758 4759
	rbd_dev->header.stripe_unit = stripe_unit;
	rbd_dev->header.stripe_count = stripe_count;
4760 4761 4762 4763

	return 0;
}

4764 4765 4766 4767 4768 4769 4770 4771 4772 4773 4774 4775 4776 4777 4778 4779 4780 4781
static int rbd_dev_v2_data_pool(struct rbd_device *rbd_dev)
{
	__le64 data_pool_id;
	int ret;

	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
				  &rbd_dev->header_oloc, "get_data_pool",
				  NULL, 0, &data_pool_id, sizeof(data_pool_id));
	if (ret < 0)
		return ret;
	if (ret < sizeof(data_pool_id))
		return -EBADMSG;

	rbd_dev->header.data_pool_id = le64_to_cpu(data_pool_id);
	WARN_ON(rbd_dev->header.data_pool_id == CEPH_NOPOOL);
	return 0;
}

4782 4783
static char *rbd_dev_image_name(struct rbd_device *rbd_dev)
{
4784
	CEPH_DEFINE_OID_ONSTACK(oid);
4785 4786 4787 4788 4789 4790 4791 4792 4793 4794 4795 4796
	size_t image_id_size;
	char *image_id;
	void *p;
	void *end;
	size_t size;
	void *reply_buf = NULL;
	size_t len = 0;
	char *image_name = NULL;
	int ret;

	rbd_assert(!rbd_dev->spec->image_name);

A
Alex Elder 已提交
4797 4798
	len = strlen(rbd_dev->spec->image_id);
	image_id_size = sizeof (__le32) + len;
4799 4800 4801 4802 4803
	image_id = kmalloc(image_id_size, GFP_KERNEL);
	if (!image_id)
		return NULL;

	p = image_id;
4804
	end = image_id + image_id_size;
4805
	ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len);
4806 4807 4808 4809 4810 4811

	size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX;
	reply_buf = kmalloc(size, GFP_KERNEL);
	if (!reply_buf)
		goto out;

4812 4813 4814 4815
	ceph_oid_printf(&oid, "%s", RBD_DIRECTORY);
	ret = rbd_obj_method_sync(rbd_dev, &oid, &rbd_dev->header_oloc,
				  "dir_get_name", image_id, image_id_size,
				  reply_buf, size);
4816 4817 4818
	if (ret < 0)
		goto out;
	p = reply_buf;
4819 4820
	end = reply_buf + ret;

4821 4822 4823 4824 4825 4826 4827 4828 4829 4830 4831 4832
	image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL);
	if (IS_ERR(image_name))
		image_name = NULL;
	else
		dout("%s: name is %s len is %zd\n", __func__, image_name, len);
out:
	kfree(reply_buf);
	kfree(image_id);

	return image_name;
}

4833 4834 4835 4836 4837 4838 4839 4840 4841 4842 4843 4844 4845 4846 4847 4848 4849 4850 4851 4852 4853 4854 4855 4856 4857 4858 4859 4860 4861 4862
static u64 rbd_v1_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
{
	struct ceph_snap_context *snapc = rbd_dev->header.snapc;
	const char *snap_name;
	u32 which = 0;

	/* Skip over names until we find the one we are looking for */

	snap_name = rbd_dev->header.snap_names;
	while (which < snapc->num_snaps) {
		if (!strcmp(name, snap_name))
			return snapc->snaps[which];
		snap_name += strlen(snap_name) + 1;
		which++;
	}
	return CEPH_NOSNAP;
}

static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
{
	struct ceph_snap_context *snapc = rbd_dev->header.snapc;
	u32 which;
	bool found = false;
	u64 snap_id;

	for (which = 0; !found && which < snapc->num_snaps; which++) {
		const char *snap_name;

		snap_id = snapc->snaps[which];
		snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id);
4863 4864 4865 4866 4867 4868 4869
		if (IS_ERR(snap_name)) {
			/* ignore no-longer existing snapshots */
			if (PTR_ERR(snap_name) == -ENOENT)
				continue;
			else
				break;
		}
4870 4871 4872 4873 4874 4875 4876 4877 4878 4879 4880 4881 4882 4883 4884 4885 4886 4887
		found = !strcmp(name, snap_name);
		kfree(snap_name);
	}
	return found ? snap_id : CEPH_NOSNAP;
}

/*
 * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if
 * no snapshot by that name is found, or if an error occurs.
 */
static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
{
	if (rbd_dev->image_format == 1)
		return rbd_v1_snap_id_by_name(rbd_dev, name);

	return rbd_v2_snap_id_by_name(rbd_dev, name);
}

4888
/*
4889 4890 4891 4892 4893 4894 4895 4896 4897 4898 4899 4900 4901 4902 4903 4904 4905 4906 4907 4908 4909 4910 4911 4912 4913 4914 4915
 * An image being mapped will have everything but the snap id.
 */
static int rbd_spec_fill_snap_id(struct rbd_device *rbd_dev)
{
	struct rbd_spec *spec = rbd_dev->spec;

	rbd_assert(spec->pool_id != CEPH_NOPOOL && spec->pool_name);
	rbd_assert(spec->image_id && spec->image_name);
	rbd_assert(spec->snap_name);

	if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) {
		u64 snap_id;

		snap_id = rbd_snap_id_by_name(rbd_dev, spec->snap_name);
		if (snap_id == CEPH_NOSNAP)
			return -ENOENT;

		spec->snap_id = snap_id;
	} else {
		spec->snap_id = CEPH_NOSNAP;
	}

	return 0;
}

/*
 * A parent image will have all ids but none of the names.
4916
 *
4917 4918
 * All names in an rbd spec are dynamically allocated.  It's OK if we
 * can't figure out the name for an image id.
4919
 */
4920
static int rbd_spec_fill_names(struct rbd_device *rbd_dev)
4921
{
4922 4923 4924 4925 4926
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
	struct rbd_spec *spec = rbd_dev->spec;
	const char *pool_name;
	const char *image_name;
	const char *snap_name;
4927 4928
	int ret;

4929 4930 4931
	rbd_assert(spec->pool_id != CEPH_NOPOOL);
	rbd_assert(spec->image_id);
	rbd_assert(spec->snap_id != CEPH_NOSNAP);
4932

4933
	/* Get the pool name; we have to make our own copy of this */
4934

4935 4936 4937
	pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id);
	if (!pool_name) {
		rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id);
4938 4939
		return -EIO;
	}
4940 4941
	pool_name = kstrdup(pool_name, GFP_KERNEL);
	if (!pool_name)
4942 4943 4944 4945
		return -ENOMEM;

	/* Fetch the image name; tolerate failure here */

4946 4947
	image_name = rbd_dev_image_name(rbd_dev);
	if (!image_name)
A
Alex Elder 已提交
4948
		rbd_warn(rbd_dev, "unable to get image name");
4949

4950
	/* Fetch the snapshot name */
4951

4952
	snap_name = rbd_snap_name(rbd_dev, spec->snap_id);
4953 4954
	if (IS_ERR(snap_name)) {
		ret = PTR_ERR(snap_name);
4955
		goto out_err;
4956 4957 4958 4959 4960
	}

	spec->pool_name = pool_name;
	spec->image_name = image_name;
	spec->snap_name = snap_name;
4961 4962

	return 0;
4963

4964
out_err:
4965 4966
	kfree(image_name);
	kfree(pool_name);
4967 4968 4969
	return ret;
}

A
Alex Elder 已提交
4970
static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev)
4971 4972 4973 4974 4975 4976 4977 4978 4979 4980 4981 4982 4983 4984 4985 4986 4987 4988 4989 4990 4991 4992 4993
{
	size_t size;
	int ret;
	void *reply_buf;
	void *p;
	void *end;
	u64 seq;
	u32 snap_count;
	struct ceph_snap_context *snapc;
	u32 i;

	/*
	 * We'll need room for the seq value (maximum snapshot id),
	 * snapshot count, and array of that many snapshot ids.
	 * For now we have a fixed upper limit on the number we're
	 * prepared to receive.
	 */
	size = sizeof (__le64) + sizeof (__le32) +
			RBD_MAX_SNAP_COUNT * sizeof (__le64);
	reply_buf = kzalloc(size, GFP_KERNEL);
	if (!reply_buf)
		return -ENOMEM;

4994 4995 4996
	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
				  &rbd_dev->header_oloc, "get_snapcontext",
				  NULL, 0, reply_buf, size);
4997
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4998 4999 5000 5001
	if (ret < 0)
		goto out;

	p = reply_buf;
5002 5003
	end = reply_buf + ret;
	ret = -ERANGE;
5004 5005 5006 5007 5008 5009 5010 5011 5012 5013 5014 5015 5016 5017 5018 5019
	ceph_decode_64_safe(&p, end, seq, out);
	ceph_decode_32_safe(&p, end, snap_count, out);

	/*
	 * Make sure the reported number of snapshot ids wouldn't go
	 * beyond the end of our buffer.  But before checking that,
	 * make sure the computed size of the snapshot context we
	 * allocate is representable in a size_t.
	 */
	if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context))
				 / sizeof (u64)) {
		ret = -EINVAL;
		goto out;
	}
	if (!ceph_has_room(&p, end, snap_count * sizeof (__le64)))
		goto out;
5020
	ret = 0;
5021

5022
	snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
5023 5024 5025 5026 5027 5028 5029 5030
	if (!snapc) {
		ret = -ENOMEM;
		goto out;
	}
	snapc->seq = seq;
	for (i = 0; i < snap_count; i++)
		snapc->snaps[i] = ceph_decode_64(&p);

5031
	ceph_put_snap_context(rbd_dev->header.snapc);
5032 5033 5034
	rbd_dev->header.snapc = snapc;

	dout("  snap context seq = %llu, snap_count = %u\n",
5035
		(unsigned long long)seq, (unsigned int)snap_count);
5036 5037 5038
out:
	kfree(reply_buf);

5039
	return ret;
5040 5041
}

5042 5043
static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
					u64 snap_id)
A
Alex Elder 已提交
5044 5045 5046
{
	size_t size;
	void *reply_buf;
5047
	__le64 snapid;
A
Alex Elder 已提交
5048 5049 5050 5051 5052 5053 5054 5055 5056 5057
	int ret;
	void *p;
	void *end;
	char *snap_name;

	size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN;
	reply_buf = kmalloc(size, GFP_KERNEL);
	if (!reply_buf)
		return ERR_PTR(-ENOMEM);

5058
	snapid = cpu_to_le64(snap_id);
5059 5060 5061
	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
				  &rbd_dev->header_oloc, "get_snapshot_name",
				  &snapid, sizeof(snapid), reply_buf, size);
5062
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
5063 5064
	if (ret < 0) {
		snap_name = ERR_PTR(ret);
A
Alex Elder 已提交
5065
		goto out;
5066
	}
A
Alex Elder 已提交
5067 5068

	p = reply_buf;
5069
	end = reply_buf + ret;
A
Alex Elder 已提交
5070
	snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
5071
	if (IS_ERR(snap_name))
A
Alex Elder 已提交
5072 5073
		goto out;

5074
	dout("  snap_id 0x%016llx snap_name = %s\n",
5075
		(unsigned long long)snap_id, snap_name);
A
Alex Elder 已提交
5076 5077 5078
out:
	kfree(reply_buf);

5079
	return snap_name;
A
Alex Elder 已提交
5080 5081
}

5082
static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev)
A
Alex Elder 已提交
5083
{
5084
	bool first_time = rbd_dev->header.object_prefix == NULL;
A
Alex Elder 已提交
5085 5086
	int ret;

5087 5088
	ret = rbd_dev_v2_image_size(rbd_dev);
	if (ret)
5089
		return ret;
5090

5091 5092 5093
	if (first_time) {
		ret = rbd_dev_v2_header_onetime(rbd_dev);
		if (ret)
5094
			return ret;
5095 5096
	}

A
Alex Elder 已提交
5097
	ret = rbd_dev_v2_snap_context(rbd_dev);
5098 5099 5100 5101
	if (ret && first_time) {
		kfree(rbd_dev->header.object_prefix);
		rbd_dev->header.object_prefix = NULL;
	}
A
Alex Elder 已提交
5102 5103 5104 5105

	return ret;
}

5106 5107 5108 5109 5110 5111 5112 5113 5114 5115
static int rbd_dev_header_info(struct rbd_device *rbd_dev)
{
	rbd_assert(rbd_image_format_valid(rbd_dev->image_format));

	if (rbd_dev->image_format == 1)
		return rbd_dev_v1_header_info(rbd_dev);

	return rbd_dev_v2_header_info(rbd_dev);
}

5116 5117 5118
/*
 * Skips over white space at *buf, and updates *buf to point to the
 * first found non-space character (if any). Returns the length of
A
Alex Elder 已提交
5119 5120
 * the token (string of non-white space characters) found.  Note
 * that *buf must be terminated with '\0'.
5121 5122 5123 5124 5125 5126 5127 5128 5129 5130 5131 5132 5133 5134
 */
static inline size_t next_token(const char **buf)
{
        /*
        * These are the characters that produce nonzero for
        * isspace() in the "C" and "POSIX" locales.
        */
        const char *spaces = " \f\n\r\t\v";

        *buf += strspn(*buf, spaces);	/* Find start of token */

	return strcspn(*buf, spaces);   /* Return token length */
}

A
Alex Elder 已提交
5135 5136 5137 5138 5139 5140 5141 5142 5143 5144 5145 5146 5147 5148 5149 5150 5151 5152 5153 5154 5155 5156
/*
 * Finds the next token in *buf, dynamically allocates a buffer big
 * enough to hold a copy of it, and copies the token into the new
 * buffer.  The copy is guaranteed to be terminated with '\0'.  Note
 * that a duplicate buffer is created even for a zero-length token.
 *
 * Returns a pointer to the newly-allocated duplicate, or a null
 * pointer if memory for the duplicate was not available.  If
 * the lenp argument is a non-null pointer, the length of the token
 * (not including the '\0') is returned in *lenp.
 *
 * If successful, the *buf pointer will be updated to point beyond
 * the end of the found token.
 *
 * Note: uses GFP_KERNEL for allocation.
 */
static inline char *dup_token(const char **buf, size_t *lenp)
{
	char *dup;
	size_t len;

	len = next_token(buf);
A
Alex Elder 已提交
5157
	dup = kmemdup(*buf, len + 1, GFP_KERNEL);
A
Alex Elder 已提交
5158 5159 5160 5161 5162 5163 5164 5165 5166 5167 5168
	if (!dup)
		return NULL;
	*(dup + len) = '\0';
	*buf += len;

	if (lenp)
		*lenp = len;

	return dup;
}

5169
/*
5170 5171 5172 5173
 * Parse the options provided for an "rbd add" (i.e., rbd image
 * mapping) request.  These arrive via a write to /sys/bus/rbd/add,
 * and the data written is passed here via a NUL-terminated buffer.
 * Returns 0 if successful or an error code otherwise.
A
Alex Elder 已提交
5174
 *
5175 5176 5177 5178 5179 5180 5181 5182 5183 5184 5185 5186 5187 5188 5189 5190 5191 5192 5193 5194 5195 5196 5197 5198 5199 5200 5201 5202 5203 5204 5205 5206 5207 5208
 * The information extracted from these options is recorded in
 * the other parameters which return dynamically-allocated
 * structures:
 *  ceph_opts
 *      The address of a pointer that will refer to a ceph options
 *      structure.  Caller must release the returned pointer using
 *      ceph_destroy_options() when it is no longer needed.
 *  rbd_opts
 *	Address of an rbd options pointer.  Fully initialized by
 *	this function; caller must release with kfree().
 *  spec
 *	Address of an rbd image specification pointer.  Fully
 *	initialized by this function based on parsed options.
 *	Caller must release with rbd_spec_put().
 *
 * The options passed take this form:
 *  <mon_addrs> <options> <pool_name> <image_name> [<snap_id>]
 * where:
 *  <mon_addrs>
 *      A comma-separated list of one or more monitor addresses.
 *      A monitor address is an ip address, optionally followed
 *      by a port number (separated by a colon).
 *        I.e.:  ip1[:port1][,ip2[:port2]...]
 *  <options>
 *      A comma-separated list of ceph and/or rbd options.
 *  <pool_name>
 *      The name of the rados pool containing the rbd image.
 *  <image_name>
 *      The name of the image in that pool to map.
 *  <snap_id>
 *      An optional snapshot id.  If provided, the mapping will
 *      present data from the image at the time that snapshot was
 *      created.  The image head is used if no snapshot id is
 *      provided.  Snapshot mappings are always read-only.
5209
 */
5210
static int rbd_add_parse_args(const char *buf,
5211
				struct ceph_options **ceph_opts,
5212 5213
				struct rbd_options **opts,
				struct rbd_spec **rbd_spec)
5214
{
A
Alex Elder 已提交
5215
	size_t len;
5216
	char *options;
5217
	const char *mon_addrs;
5218
	char *snap_name;
5219
	size_t mon_addrs_size;
5220
	struct rbd_spec *spec = NULL;
5221
	struct rbd_options *rbd_opts = NULL;
5222
	struct ceph_options *copts;
5223
	int ret;
5224 5225 5226

	/* The first four tokens are required */

5227
	len = next_token(&buf);
5228 5229 5230 5231
	if (!len) {
		rbd_warn(NULL, "no monitor address(es) provided");
		return -EINVAL;
	}
5232
	mon_addrs = buf;
5233
	mon_addrs_size = len + 1;
5234
	buf += len;
5235

5236
	ret = -EINVAL;
5237 5238
	options = dup_token(&buf, NULL);
	if (!options)
5239
		return -ENOMEM;
5240 5241 5242 5243
	if (!*options) {
		rbd_warn(NULL, "no options provided");
		goto out_err;
	}
5244

5245 5246
	spec = rbd_spec_alloc();
	if (!spec)
5247
		goto out_mem;
5248 5249 5250 5251

	spec->pool_name = dup_token(&buf, NULL);
	if (!spec->pool_name)
		goto out_mem;
5252 5253 5254 5255
	if (!*spec->pool_name) {
		rbd_warn(NULL, "no pool name provided");
		goto out_err;
	}
5256

A
Alex Elder 已提交
5257
	spec->image_name = dup_token(&buf, NULL);
5258
	if (!spec->image_name)
5259
		goto out_mem;
5260 5261 5262 5263
	if (!*spec->image_name) {
		rbd_warn(NULL, "no image name provided");
		goto out_err;
	}
5264

5265 5266 5267 5268
	/*
	 * Snapshot name is optional; default is to use "-"
	 * (indicating the head/no snapshot).
	 */
5269
	len = next_token(&buf);
5270
	if (!len) {
5271 5272
		buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */
		len = sizeof (RBD_SNAP_HEAD_NAME) - 1;
5273
	} else if (len > RBD_MAX_SNAP_NAME_LEN) {
5274
		ret = -ENAMETOOLONG;
5275
		goto out_err;
5276
	}
5277 5278
	snap_name = kmemdup(buf, len + 1, GFP_KERNEL);
	if (!snap_name)
5279
		goto out_mem;
5280 5281
	*(snap_name + len) = '\0';
	spec->snap_name = snap_name;
A
Alex Elder 已提交
5282

5283
	/* Initialize all rbd options to the defaults */
5284

5285 5286 5287 5288 5289
	rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL);
	if (!rbd_opts)
		goto out_mem;

	rbd_opts->read_only = RBD_READ_ONLY_DEFAULT;
I
Ilya Dryomov 已提交
5290
	rbd_opts->queue_depth = RBD_QUEUE_DEPTH_DEFAULT;
I
Ilya Dryomov 已提交
5291
	rbd_opts->lock_on_read = RBD_LOCK_ON_READ_DEFAULT;
I
Ilya Dryomov 已提交
5292
	rbd_opts->exclusive = RBD_EXCLUSIVE_DEFAULT;
A
Alex Elder 已提交
5293

5294
	copts = ceph_parse_options(options, mon_addrs,
5295
					mon_addrs + mon_addrs_size - 1,
5296
					parse_rbd_opts_token, rbd_opts);
5297 5298
	if (IS_ERR(copts)) {
		ret = PTR_ERR(copts);
5299 5300
		goto out_err;
	}
5301 5302 5303
	kfree(options);

	*ceph_opts = copts;
5304
	*opts = rbd_opts;
5305
	*rbd_spec = spec;
5306

5307
	return 0;
5308
out_mem:
5309
	ret = -ENOMEM;
A
Alex Elder 已提交
5310
out_err:
5311 5312
	kfree(rbd_opts);
	rbd_spec_put(spec);
5313
	kfree(options);
A
Alex Elder 已提交
5314

5315
	return ret;
5316 5317
}

5318 5319 5320 5321 5322
/*
 * Return pool id (>= 0) or a negative error code.
 */
static int rbd_add_get_pool_id(struct rbd_client *rbdc, const char *pool_name)
{
5323
	struct ceph_options *opts = rbdc->client->options;
5324 5325 5326 5327 5328 5329 5330
	u64 newest_epoch;
	int tries = 0;
	int ret;

again:
	ret = ceph_pg_poolid_by_name(rbdc->client->osdc.osdmap, pool_name);
	if (ret == -ENOENT && tries++ < 1) {
5331 5332
		ret = ceph_monc_get_version(&rbdc->client->monc, "osdmap",
					    &newest_epoch);
5333 5334 5335 5336
		if (ret < 0)
			return ret;

		if (rbdc->client->osdc.osdmap->epoch < newest_epoch) {
5337
			ceph_osdc_maybe_request_map(&rbdc->client->osdc);
5338
			(void) ceph_monc_wait_osdmap(&rbdc->client->monc,
5339 5340
						     newest_epoch,
						     opts->mount_timeout);
5341 5342 5343 5344 5345 5346 5347 5348 5349 5350
			goto again;
		} else {
			/* the osdmap we have is new enough */
			return -ENOENT;
		}
	}

	return ret;
}

I
Ilya Dryomov 已提交
5351 5352 5353 5354 5355 5356 5357 5358 5359 5360 5361 5362 5363 5364 5365 5366 5367 5368 5369 5370 5371 5372 5373 5374 5375 5376 5377
static void rbd_dev_image_unlock(struct rbd_device *rbd_dev)
{
	down_write(&rbd_dev->lock_rwsem);
	if (__rbd_is_lock_owner(rbd_dev))
		rbd_unlock(rbd_dev);
	up_write(&rbd_dev->lock_rwsem);
}

static int rbd_add_acquire_lock(struct rbd_device *rbd_dev)
{
	if (!(rbd_dev->header.features & RBD_FEATURE_EXCLUSIVE_LOCK)) {
		rbd_warn(rbd_dev, "exclusive-lock feature is not enabled");
		return -EINVAL;
	}

	/* FIXME: "rbd map --exclusive" should be in interruptible */
	down_read(&rbd_dev->lock_rwsem);
	rbd_wait_state_locked(rbd_dev);
	up_read(&rbd_dev->lock_rwsem);
	if (test_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags)) {
		rbd_warn(rbd_dev, "failed to acquire exclusive lock");
		return -EROFS;
	}

	return 0;
}

A
Alex Elder 已提交
5378 5379 5380 5381 5382 5383 5384 5385 5386 5387 5388 5389 5390 5391 5392 5393 5394 5395
/*
 * An rbd format 2 image has a unique identifier, distinct from the
 * name given to it by the user.  Internally, that identifier is
 * what's used to specify the names of objects related to the image.
 *
 * A special "rbd id" object is used to map an rbd image name to its
 * id.  If that object doesn't exist, then there is no v2 rbd image
 * with the supplied name.
 *
 * This function will record the given rbd_dev's image_id field if
 * it can be determined, and in that case will return 0.  If any
 * errors occur a negative errno will be returned and the rbd_dev's
 * image_id field will be unchanged (and should be NULL).
 */
static int rbd_dev_image_id(struct rbd_device *rbd_dev)
{
	int ret;
	size_t size;
5396
	CEPH_DEFINE_OID_ONSTACK(oid);
A
Alex Elder 已提交
5397
	void *response;
5398
	char *image_id;
5399

A
Alex Elder 已提交
5400 5401 5402
	/*
	 * When probing a parent image, the image id is already
	 * known (and the image name likely is not).  There's no
5403 5404
	 * need to fetch the image id again in this case.  We
	 * do still need to set the image format though.
A
Alex Elder 已提交
5405
	 */
5406 5407 5408
	if (rbd_dev->spec->image_id) {
		rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1;

A
Alex Elder 已提交
5409
		return 0;
5410
	}
A
Alex Elder 已提交
5411

A
Alex Elder 已提交
5412 5413 5414 5415
	/*
	 * First, see if the format 2 image id file exists, and if
	 * so, get the image's persistent id from it.
	 */
5416 5417 5418 5419 5420 5421
	ret = ceph_oid_aprintf(&oid, GFP_KERNEL, "%s%s", RBD_ID_PREFIX,
			       rbd_dev->spec->image_name);
	if (ret)
		return ret;

	dout("rbd id object name is %s\n", oid.name);
A
Alex Elder 已提交
5422 5423 5424 5425 5426 5427 5428 5429 5430 5431

	/* Response will be an encoded string, which includes a length */

	size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX;
	response = kzalloc(size, GFP_NOIO);
	if (!response) {
		ret = -ENOMEM;
		goto out;
	}

5432 5433
	/* If it doesn't exist we'll assume it's a format 1 image */

5434 5435 5436
	ret = rbd_obj_method_sync(rbd_dev, &oid, &rbd_dev->header_oloc,
				  "get_id", NULL, 0,
				  response, RBD_IMAGE_ID_LEN_MAX);
5437
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
5438 5439 5440 5441 5442
	if (ret == -ENOENT) {
		image_id = kstrdup("", GFP_KERNEL);
		ret = image_id ? 0 : -ENOMEM;
		if (!ret)
			rbd_dev->image_format = 1;
5443
	} else if (ret >= 0) {
5444 5445 5446
		void *p = response;

		image_id = ceph_extract_encoded_string(&p, p + ret,
A
Alex Elder 已提交
5447
						NULL, GFP_NOIO);
5448
		ret = PTR_ERR_OR_ZERO(image_id);
5449 5450 5451 5452 5453 5454 5455
		if (!ret)
			rbd_dev->image_format = 2;
	}

	if (!ret) {
		rbd_dev->spec->image_id = image_id;
		dout("image_id is %s\n", image_id);
A
Alex Elder 已提交
5456 5457 5458
	}
out:
	kfree(response);
5459
	ceph_oid_destroy(&oid);
A
Alex Elder 已提交
5460 5461 5462
	return ret;
}

A
Alex Elder 已提交
5463 5464 5465 5466
/*
 * Undo whatever state changes are made by v1 or v2 header info
 * call.
 */
A
Alex Elder 已提交
5467 5468 5469 5470
static void rbd_dev_unprobe(struct rbd_device *rbd_dev)
{
	struct rbd_image_header	*header;

5471
	rbd_dev_parent_put(rbd_dev);
A
Alex Elder 已提交
5472 5473 5474 5475

	/* Free dynamic fields from the header, then zero it out */

	header = &rbd_dev->header;
5476
	ceph_put_snap_context(header->snapc);
A
Alex Elder 已提交
5477 5478 5479 5480 5481 5482
	kfree(header->snap_sizes);
	kfree(header->snap_names);
	kfree(header->object_prefix);
	memset(header, 0, sizeof (*header));
}

5483
static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev)
5484 5485 5486
{
	int ret;

5487
	ret = rbd_dev_v2_object_prefix(rbd_dev);
5488
	if (ret)
5489 5490
		goto out_err;

5491 5492 5493 5494
	/*
	 * Get the and check features for the image.  Currently the
	 * features are assumed to never change.
	 */
5495
	ret = rbd_dev_v2_features(rbd_dev);
5496
	if (ret)
5497
		goto out_err;
5498

5499 5500 5501 5502 5503 5504 5505
	/* If the image supports fancy striping, get its parameters */

	if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) {
		ret = rbd_dev_v2_striping_info(rbd_dev);
		if (ret < 0)
			goto out_err;
	}
5506

5507 5508 5509 5510 5511 5512
	if (rbd_dev->header.features & RBD_FEATURE_DATA_POOL) {
		ret = rbd_dev_v2_data_pool(rbd_dev);
		if (ret)
			goto out_err;
	}

I
Ilya Dryomov 已提交
5513
	rbd_init_layout(rbd_dev);
A
Alex Elder 已提交
5514
	return 0;
I
Ilya Dryomov 已提交
5515

5516
out_err:
A
Alex Elder 已提交
5517
	rbd_dev->header.features = 0;
5518 5519
	kfree(rbd_dev->header.object_prefix);
	rbd_dev->header.object_prefix = NULL;
5520
	return ret;
5521 5522
}

5523 5524 5525 5526 5527 5528
/*
 * @depth is rbd_dev_image_probe() -> rbd_dev_probe_parent() ->
 * rbd_dev_image_probe() recursion depth, which means it's also the
 * length of the already discovered part of the parent chain.
 */
static int rbd_dev_probe_parent(struct rbd_device *rbd_dev, int depth)
A
Alex Elder 已提交
5529
{
5530
	struct rbd_device *parent = NULL;
5531 5532 5533 5534 5535
	int ret;

	if (!rbd_dev->parent_spec)
		return 0;

5536 5537 5538 5539 5540 5541
	if (++depth > RBD_MAX_PARENT_CHAIN_LEN) {
		pr_info("parent chain is too long (%d)\n", depth);
		ret = -EINVAL;
		goto out_err;
	}

5542
	parent = __rbd_dev_create(rbd_dev->rbd_client, rbd_dev->parent_spec);
5543 5544
	if (!parent) {
		ret = -ENOMEM;
5545
		goto out_err;
5546 5547 5548 5549 5550 5551 5552 5553
	}

	/*
	 * Images related by parent/child relationships always share
	 * rbd_client and spec/parent_spec, so bump their refcounts.
	 */
	__rbd_get_client(rbd_dev->rbd_client);
	rbd_spec_get(rbd_dev->parent_spec);
5554

5555
	ret = rbd_dev_image_probe(parent, depth);
5556 5557
	if (ret < 0)
		goto out_err;
5558

5559
	rbd_dev->parent = parent;
5560
	atomic_set(&rbd_dev->parent_ref, 1);
5561
	return 0;
5562

5563
out_err:
5564
	rbd_dev_unparent(rbd_dev);
5565
	rbd_dev_destroy(parent);
5566 5567 5568
	return ret;
}

5569 5570 5571 5572 5573 5574 5575 5576 5577
static void rbd_dev_device_release(struct rbd_device *rbd_dev)
{
	clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
	rbd_dev_mapping_clear(rbd_dev);
	rbd_free_disk(rbd_dev);
	if (!single_major)
		unregister_blkdev(rbd_dev->major, rbd_dev->name);
}

I
Ilya Dryomov 已提交
5578 5579 5580 5581
/*
 * rbd_dev->header_rwsem must be locked for write and will be unlocked
 * upon return.
 */
5582
static int rbd_dev_device_setup(struct rbd_device *rbd_dev)
5583
{
A
Alex Elder 已提交
5584
	int ret;
A
Alex Elder 已提交
5585

5586
	/* Record our major and minor device numbers. */
A
Alex Elder 已提交
5587

5588 5589 5590
	if (!single_major) {
		ret = register_blkdev(0, rbd_dev->name);
		if (ret < 0)
5591
			goto err_out_unlock;
5592 5593 5594 5595 5596 5597 5598

		rbd_dev->major = ret;
		rbd_dev->minor = 0;
	} else {
		rbd_dev->major = rbd_major;
		rbd_dev->minor = rbd_dev_id_to_minor(rbd_dev->dev_id);
	}
A
Alex Elder 已提交
5599 5600 5601 5602 5603 5604 5605

	/* Set up the blkdev mapping. */

	ret = rbd_init_disk(rbd_dev);
	if (ret)
		goto err_out_blkdev;

5606
	ret = rbd_dev_mapping_set(rbd_dev);
A
Alex Elder 已提交
5607 5608
	if (ret)
		goto err_out_disk;
I
Ilya Dryomov 已提交
5609

5610
	set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE);
5611
	set_disk_ro(rbd_dev->disk, rbd_dev->opts->read_only);
5612

5613
	ret = dev_set_name(&rbd_dev->dev, "%d", rbd_dev->dev_id);
5614
	if (ret)
5615
		goto err_out_mapping;
A
Alex Elder 已提交
5616

5617
	set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
I
Ilya Dryomov 已提交
5618
	up_write(&rbd_dev->header_rwsem);
5619
	return 0;
5620

5621 5622
err_out_mapping:
	rbd_dev_mapping_clear(rbd_dev);
A
Alex Elder 已提交
5623 5624 5625
err_out_disk:
	rbd_free_disk(rbd_dev);
err_out_blkdev:
5626 5627
	if (!single_major)
		unregister_blkdev(rbd_dev->major, rbd_dev->name);
I
Ilya Dryomov 已提交
5628 5629
err_out_unlock:
	up_write(&rbd_dev->header_rwsem);
A
Alex Elder 已提交
5630 5631 5632
	return ret;
}

A
Alex Elder 已提交
5633 5634 5635
static int rbd_dev_header_name(struct rbd_device *rbd_dev)
{
	struct rbd_spec *spec = rbd_dev->spec;
5636
	int ret;
A
Alex Elder 已提交
5637 5638 5639 5640 5641

	/* Record the header object name for this rbd image. */

	rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
	if (rbd_dev->image_format == 1)
5642 5643
		ret = ceph_oid_aprintf(&rbd_dev->header_oid, GFP_KERNEL, "%s%s",
				       spec->image_name, RBD_SUFFIX);
A
Alex Elder 已提交
5644
	else
5645 5646
		ret = ceph_oid_aprintf(&rbd_dev->header_oid, GFP_KERNEL, "%s%s",
				       RBD_HEADER_PREFIX, spec->image_id);
A
Alex Elder 已提交
5647

5648
	return ret;
A
Alex Elder 已提交
5649 5650
}

5651 5652
static void rbd_dev_image_release(struct rbd_device *rbd_dev)
{
A
Alex Elder 已提交
5653
	rbd_dev_unprobe(rbd_dev);
5654 5655
	if (rbd_dev->opts)
		rbd_unregister_watch(rbd_dev);
A
Alex Elder 已提交
5656 5657 5658
	rbd_dev->image_format = 0;
	kfree(rbd_dev->spec->image_id);
	rbd_dev->spec->image_id = NULL;
5659 5660
}

5661 5662
/*
 * Probe for the existence of the header object for the given rbd
5663 5664 5665
 * device.  If this image is the one being mapped (i.e., not a
 * parent), initiate a watch on its header object before using that
 * object to get detailed information about the rbd image.
5666
 */
5667
static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth)
5668 5669 5670 5671
{
	int ret;

	/*
A
Alex Elder 已提交
5672 5673 5674 5675
	 * Get the id from the image id object.  Unless there's an
	 * error, rbd_dev->spec->image_id will be filled in with
	 * a dynamically-allocated string, and rbd_dev->image_format
	 * will be set to either 1 or 2.
5676 5677 5678
	 */
	ret = rbd_dev_image_id(rbd_dev);
	if (ret)
5679 5680
		return ret;

A
Alex Elder 已提交
5681 5682 5683 5684
	ret = rbd_dev_header_name(rbd_dev);
	if (ret)
		goto err_out_format;

5685
	if (!depth) {
5686
		ret = rbd_register_watch(rbd_dev);
5687 5688 5689 5690 5691
		if (ret) {
			if (ret == -ENOENT)
				pr_info("image %s/%s does not exist\n",
					rbd_dev->spec->pool_name,
					rbd_dev->spec->image_name);
5692
			goto err_out_format;
5693
		}
5694
	}
5695

5696
	ret = rbd_dev_header_info(rbd_dev);
5697
	if (ret)
5698
		goto err_out_watch;
A
Alex Elder 已提交
5699

5700 5701 5702 5703 5704 5705
	/*
	 * If this image is the one being mapped, we have pool name and
	 * id, image name and id, and snap name - need to fill snap id.
	 * Otherwise this is a parent image, identified by pool, image
	 * and snap ids - need to fill in names for those ids.
	 */
5706
	if (!depth)
5707 5708 5709
		ret = rbd_spec_fill_snap_id(rbd_dev);
	else
		ret = rbd_spec_fill_names(rbd_dev);
5710 5711 5712 5713 5714 5715
	if (ret) {
		if (ret == -ENOENT)
			pr_info("snap %s/%s@%s does not exist\n",
				rbd_dev->spec->pool_name,
				rbd_dev->spec->image_name,
				rbd_dev->spec->snap_name);
A
Alex Elder 已提交
5716
		goto err_out_probe;
5717
	}
5718

5719 5720 5721 5722 5723 5724 5725 5726 5727
	if (rbd_dev->header.features & RBD_FEATURE_LAYERING) {
		ret = rbd_dev_v2_parent_info(rbd_dev);
		if (ret)
			goto err_out_probe;

		/*
		 * Need to warn users if this image is the one being
		 * mapped and has a parent.
		 */
5728
		if (!depth && rbd_dev->parent_spec)
5729 5730 5731 5732
			rbd_warn(rbd_dev,
				 "WARNING: kernel layering is EXPERIMENTAL!");
	}

5733
	ret = rbd_dev_probe_parent(rbd_dev, depth);
A
Alex Elder 已提交
5734 5735 5736 5737
	if (ret)
		goto err_out_probe;

	dout("discovered format %u image, header name is %s\n",
5738
		rbd_dev->image_format, rbd_dev->header_oid.name);
A
Alex Elder 已提交
5739
	return 0;
5740

A
Alex Elder 已提交
5741 5742
err_out_probe:
	rbd_dev_unprobe(rbd_dev);
5743
err_out_watch:
5744
	if (!depth)
5745
		rbd_unregister_watch(rbd_dev);
A
Alex Elder 已提交
5746 5747
err_out_format:
	rbd_dev->image_format = 0;
5748 5749
	kfree(rbd_dev->spec->image_id);
	rbd_dev->spec->image_id = NULL;
5750 5751 5752
	return ret;
}

5753 5754 5755
static ssize_t do_rbd_add(struct bus_type *bus,
			  const char *buf,
			  size_t count)
5756
{
5757
	struct rbd_device *rbd_dev = NULL;
5758
	struct ceph_options *ceph_opts = NULL;
5759
	struct rbd_options *rbd_opts = NULL;
5760
	struct rbd_spec *spec = NULL;
5761
	struct rbd_client *rbdc;
5762
	int rc;
5763 5764 5765 5766 5767

	if (!try_module_get(THIS_MODULE))
		return -ENODEV;

	/* parse add command */
5768
	rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec);
5769
	if (rc < 0)
5770
		goto out;
5771

5772 5773 5774
	rbdc = rbd_get_client(ceph_opts);
	if (IS_ERR(rbdc)) {
		rc = PTR_ERR(rbdc);
5775
		goto err_out_args;
5776
	}
5777 5778

	/* pick the pool */
5779
	rc = rbd_add_get_pool_id(rbdc, spec->pool_name);
5780 5781 5782
	if (rc < 0) {
		if (rc == -ENOENT)
			pr_info("pool %s does not exist\n", spec->pool_name);
5783
		goto err_out_client;
5784
	}
A
Alex Elder 已提交
5785
	spec->pool_id = (u64)rc;
5786

5787
	rbd_dev = rbd_dev_create(rbdc, spec, rbd_opts);
5788 5789
	if (!rbd_dev) {
		rc = -ENOMEM;
5790
		goto err_out_client;
5791
	}
5792 5793
	rbdc = NULL;		/* rbd_dev now owns this */
	spec = NULL;		/* rbd_dev now owns this */
5794
	rbd_opts = NULL;	/* rbd_dev now owns this */
5795

5796 5797 5798 5799 5800 5801
	rbd_dev->config_info = kstrdup(buf, GFP_KERNEL);
	if (!rbd_dev->config_info) {
		rc = -ENOMEM;
		goto err_out_rbd_dev;
	}

I
Ilya Dryomov 已提交
5802
	down_write(&rbd_dev->header_rwsem);
5803
	rc = rbd_dev_image_probe(rbd_dev, 0);
5804 5805
	if (rc < 0) {
		up_write(&rbd_dev->header_rwsem);
5806
		goto err_out_rbd_dev;
5807
	}
5808

5809 5810
	/* If we are mapping a snapshot it must be marked read-only */
	if (rbd_dev->spec->snap_id != CEPH_NOSNAP)
5811
		rbd_dev->opts->read_only = true;
5812

5813
	rc = rbd_dev_device_setup(rbd_dev);
5814
	if (rc)
5815
		goto err_out_image_probe;
A
Alex Elder 已提交
5816

I
Ilya Dryomov 已提交
5817 5818 5819 5820
	if (rbd_dev->opts->exclusive) {
		rc = rbd_add_acquire_lock(rbd_dev);
		if (rc)
			goto err_out_device_setup;
A
Alex Elder 已提交
5821 5822
	}

5823 5824 5825 5826
	/* Everything's ready.  Announce the disk to the world. */

	rc = device_add(&rbd_dev->dev);
	if (rc)
I
Ilya Dryomov 已提交
5827
		goto err_out_image_lock;
5828 5829 5830 5831 5832 5833 5834 5835 5836 5837 5838 5839

	add_disk(rbd_dev->disk);
	/* see rbd_init_disk() */
	blk_put_queue(rbd_dev->disk->queue);

	spin_lock(&rbd_dev_list_lock);
	list_add_tail(&rbd_dev->node, &rbd_dev_list);
	spin_unlock(&rbd_dev_list_lock);

	pr_info("%s: capacity %llu features 0x%llx\n", rbd_dev->disk->disk_name,
		(unsigned long long)get_capacity(rbd_dev->disk) << SECTOR_SHIFT,
		rbd_dev->header.features);
5840 5841 5842 5843
	rc = count;
out:
	module_put(THIS_MODULE);
	return rc;
5844

I
Ilya Dryomov 已提交
5845 5846
err_out_image_lock:
	rbd_dev_image_unlock(rbd_dev);
5847 5848
err_out_device_setup:
	rbd_dev_device_release(rbd_dev);
5849 5850
err_out_image_probe:
	rbd_dev_image_release(rbd_dev);
5851 5852
err_out_rbd_dev:
	rbd_dev_destroy(rbd_dev);
5853
err_out_client:
5854
	rbd_put_client(rbdc);
5855
err_out_args:
5856
	rbd_spec_put(spec);
5857
	kfree(rbd_opts);
5858
	goto out;
5859 5860
}

5861 5862 5863 5864 5865 5866 5867 5868 5869 5870 5871 5872 5873 5874 5875 5876 5877
static ssize_t rbd_add(struct bus_type *bus,
		       const char *buf,
		       size_t count)
{
	if (single_major)
		return -EINVAL;

	return do_rbd_add(bus, buf, count);
}

static ssize_t rbd_add_single_major(struct bus_type *bus,
				    const char *buf,
				    size_t count)
{
	return do_rbd_add(bus, buf, count);
}

5878 5879
static void rbd_dev_remove_parent(struct rbd_device *rbd_dev)
{
A
Alex Elder 已提交
5880
	while (rbd_dev->parent) {
5881 5882 5883 5884 5885 5886 5887 5888 5889 5890 5891 5892
		struct rbd_device *first = rbd_dev;
		struct rbd_device *second = first->parent;
		struct rbd_device *third;

		/*
		 * Follow to the parent with no grandparent and
		 * remove it.
		 */
		while (second && (third = second->parent)) {
			first = second;
			second = third;
		}
A
Alex Elder 已提交
5893
		rbd_assert(second);
5894
		rbd_dev_image_release(second);
5895
		rbd_dev_destroy(second);
A
Alex Elder 已提交
5896 5897 5898 5899
		first->parent = NULL;
		first->parent_overlap = 0;

		rbd_assert(first->parent_spec);
5900 5901 5902 5903 5904
		rbd_spec_put(first->parent_spec);
		first->parent_spec = NULL;
	}
}

5905 5906 5907
static ssize_t do_rbd_remove(struct bus_type *bus,
			     const char *buf,
			     size_t count)
5908 5909
{
	struct rbd_device *rbd_dev = NULL;
5910 5911
	struct list_head *tmp;
	int dev_id;
M
Mike Christie 已提交
5912
	char opt_buf[6];
5913
	bool already = false;
M
Mike Christie 已提交
5914
	bool force = false;
5915
	int ret;
5916

M
Mike Christie 已提交
5917 5918 5919 5920 5921
	dev_id = -1;
	opt_buf[0] = '\0';
	sscanf(buf, "%d %5s", &dev_id, opt_buf);
	if (dev_id < 0) {
		pr_err("dev_id out of range\n");
5922
		return -EINVAL;
M
Mike Christie 已提交
5923 5924 5925 5926 5927 5928 5929 5930 5931
	}
	if (opt_buf[0] != '\0') {
		if (!strcmp(opt_buf, "force")) {
			force = true;
		} else {
			pr_err("bad remove option at '%s'\n", opt_buf);
			return -EINVAL;
		}
	}
5932

5933 5934 5935 5936 5937 5938 5939 5940
	ret = -ENOENT;
	spin_lock(&rbd_dev_list_lock);
	list_for_each(tmp, &rbd_dev_list) {
		rbd_dev = list_entry(tmp, struct rbd_device, node);
		if (rbd_dev->dev_id == dev_id) {
			ret = 0;
			break;
		}
5941
	}
5942 5943
	if (!ret) {
		spin_lock_irq(&rbd_dev->lock);
M
Mike Christie 已提交
5944
		if (rbd_dev->open_count && !force)
5945 5946
			ret = -EBUSY;
		else
5947 5948
			already = test_and_set_bit(RBD_DEV_FLAG_REMOVING,
							&rbd_dev->flags);
5949 5950 5951
		spin_unlock_irq(&rbd_dev->lock);
	}
	spin_unlock(&rbd_dev_list_lock);
5952
	if (ret < 0 || already)
5953
		return ret;
5954

M
Mike Christie 已提交
5955 5956 5957 5958 5959 5960 5961 5962 5963
	if (force) {
		/*
		 * Prevent new IO from being queued and wait for existing
		 * IO to complete/fail.
		 */
		blk_mq_freeze_queue(rbd_dev->disk->queue);
		blk_set_queue_dying(rbd_dev->disk->queue);
	}

5964 5965 5966 5967 5968
	del_gendisk(rbd_dev->disk);
	spin_lock(&rbd_dev_list_lock);
	list_del_init(&rbd_dev->node);
	spin_unlock(&rbd_dev_list_lock);
	device_del(&rbd_dev->dev);
5969

I
Ilya Dryomov 已提交
5970
	rbd_dev_image_unlock(rbd_dev);
5971
	rbd_dev_device_release(rbd_dev);
5972
	rbd_dev_image_release(rbd_dev);
5973
	rbd_dev_destroy(rbd_dev);
5974
	return count;
5975 5976
}

5977 5978 5979 5980 5981 5982 5983 5984 5985 5986 5987 5988 5989 5990 5991 5992 5993
static ssize_t rbd_remove(struct bus_type *bus,
			  const char *buf,
			  size_t count)
{
	if (single_major)
		return -EINVAL;

	return do_rbd_remove(bus, buf, count);
}

static ssize_t rbd_remove_single_major(struct bus_type *bus,
				       const char *buf,
				       size_t count)
{
	return do_rbd_remove(bus, buf, count);
}

5994 5995
/*
 * create control files in sysfs
5996
 * /sys/bus/rbd/...
5997 5998 5999
 */
static int rbd_sysfs_init(void)
{
6000
	int ret;
6001

6002
	ret = device_register(&rbd_root_dev);
A
Alex Elder 已提交
6003
	if (ret < 0)
6004
		return ret;
6005

6006 6007 6008
	ret = bus_register(&rbd_bus_type);
	if (ret < 0)
		device_unregister(&rbd_root_dev);
6009 6010 6011 6012 6013 6014

	return ret;
}

static void rbd_sysfs_cleanup(void)
{
6015
	bus_unregister(&rbd_bus_type);
6016
	device_unregister(&rbd_root_dev);
6017 6018
}

6019 6020 6021
static int rbd_slab_init(void)
{
	rbd_assert(!rbd_img_request_cache);
G
Geliang Tang 已提交
6022
	rbd_img_request_cache = KMEM_CACHE(rbd_img_request, 0);
6023 6024 6025 6026
	if (!rbd_img_request_cache)
		return -ENOMEM;

	rbd_assert(!rbd_obj_request_cache);
G
Geliang Tang 已提交
6027
	rbd_obj_request_cache = KMEM_CACHE(rbd_obj_request, 0);
6028 6029 6030
	if (!rbd_obj_request_cache)
		goto out_err;

6031
	return 0;
6032

6033
out_err:
6034 6035
	kmem_cache_destroy(rbd_img_request_cache);
	rbd_img_request_cache = NULL;
6036 6037 6038 6039 6040
	return -ENOMEM;
}

static void rbd_slab_exit(void)
{
6041 6042 6043 6044
	rbd_assert(rbd_obj_request_cache);
	kmem_cache_destroy(rbd_obj_request_cache);
	rbd_obj_request_cache = NULL;

6045 6046 6047 6048 6049
	rbd_assert(rbd_img_request_cache);
	kmem_cache_destroy(rbd_img_request_cache);
	rbd_img_request_cache = NULL;
}

A
Alex Elder 已提交
6050
static int __init rbd_init(void)
6051 6052 6053
{
	int rc;

6054 6055 6056 6057
	if (!libceph_compatible(NULL)) {
		rbd_warn(NULL, "libceph incompatibility (quitting)");
		return -EINVAL;
	}
I
Ilya Dryomov 已提交
6058

6059
	rc = rbd_slab_init();
6060 6061
	if (rc)
		return rc;
I
Ilya Dryomov 已提交
6062

6063 6064
	/*
	 * The number of active work items is limited by the number of
I
Ilya Dryomov 已提交
6065
	 * rbd devices * queue depth, so leave @max_active at default.
6066 6067 6068 6069 6070 6071 6072
	 */
	rbd_wq = alloc_workqueue(RBD_DRV_NAME, WQ_MEM_RECLAIM, 0);
	if (!rbd_wq) {
		rc = -ENOMEM;
		goto err_out_slab;
	}

6073 6074 6075 6076
	if (single_major) {
		rbd_major = register_blkdev(0, RBD_DRV_NAME);
		if (rbd_major < 0) {
			rc = rbd_major;
6077
			goto err_out_wq;
6078 6079 6080
		}
	}

6081 6082
	rc = rbd_sysfs_init();
	if (rc)
6083 6084 6085 6086 6087 6088
		goto err_out_blkdev;

	if (single_major)
		pr_info("loaded (major %d)\n", rbd_major);
	else
		pr_info("loaded\n");
6089

I
Ilya Dryomov 已提交
6090 6091
	return 0;

6092 6093 6094
err_out_blkdev:
	if (single_major)
		unregister_blkdev(rbd_major, RBD_DRV_NAME);
6095 6096
err_out_wq:
	destroy_workqueue(rbd_wq);
I
Ilya Dryomov 已提交
6097 6098
err_out_slab:
	rbd_slab_exit();
6099
	return rc;
6100 6101
}

A
Alex Elder 已提交
6102
static void __exit rbd_exit(void)
6103
{
I
Ilya Dryomov 已提交
6104
	ida_destroy(&rbd_dev_id_ida);
6105
	rbd_sysfs_cleanup();
6106 6107
	if (single_major)
		unregister_blkdev(rbd_major, RBD_DRV_NAME);
6108
	destroy_workqueue(rbd_wq);
6109
	rbd_slab_exit();
6110 6111 6112 6113 6114
}

module_init(rbd_init);
module_exit(rbd_exit);

A
Alex Elder 已提交
6115
MODULE_AUTHOR("Alex Elder <elder@inktank.com>");
6116 6117 6118 6119 6120
MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
/* following authorship retained from original osdblk.c */
MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");

6121
MODULE_DESCRIPTION("RADOS Block Device (RBD) driver");
6122
MODULE_LICENSE("GPL");