rbd.c 150.1 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/striper.h>
36
#include <linux/ceph/decode.h>
37
#include <linux/parser.h>
38
#include <linux/bsearch.h>
39 40 41 42

#include <linux/kernel.h>
#include <linux/device.h>
#include <linux/module.h>
C
Christoph Hellwig 已提交
43
#include <linux/blk-mq.h>
44 45
#include <linux/fs.h>
#include <linux/blkdev.h>
46
#include <linux/slab.h>
47
#include <linux/idr.h>
I
Ilya Dryomov 已提交
48
#include <linux/workqueue.h>
49 50 51

#include "rbd_types.h"

A
Alex Elder 已提交
52 53
#define RBD_DEBUG	/* Activate rbd_assert() calls */

54 55 56 57 58 59 60 61 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
/*
 * 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
87
#define RBD_DRV_NAME "rbd"
88

89 90
#define RBD_MINORS_PER_MAJOR		256
#define RBD_SINGLE_MAJOR_PART_SHIFT	4
91

92 93
#define RBD_MAX_PARENT_CHAIN_LEN	16

94 95 96 97
#define RBD_SNAP_DEV_NAME_PREFIX	"snap_"
#define RBD_MAX_SNAP_NAME_LEN	\
			(NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1))

98
#define RBD_MAX_SNAP_COUNT	510	/* allows max snapc to fit in 4KB */
99 100 101

#define RBD_SNAP_HEAD_NAME	"-"

102 103
#define	BAD_SNAP_INDEX	U32_MAX		/* invalid index into snap array */

104 105
/* This allows a single page to hold an image name sent by OSD */
#define RBD_IMAGE_NAME_LEN_MAX	(PAGE_SIZE - sizeof (__le32) - 1)
106
#define RBD_IMAGE_ID_LEN_MAX	64
107

108
#define RBD_OBJ_PREFIX_LEN_MAX	64
A
Alex Elder 已提交
109

110
#define RBD_NOTIFY_TIMEOUT	5	/* seconds */
111 112
#define RBD_RETRY_DELAY		msecs_to_jiffies(1000)

A
Alex Elder 已提交
113 114
/* Feature bits */

115 116 117 118
#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)
119
#define RBD_FEATURE_OPERATIONS		(1ULL<<8)
120

121 122
#define RBD_FEATURES_ALL	(RBD_FEATURE_LAYERING |		\
				 RBD_FEATURE_STRIPINGV2 |	\
123
				 RBD_FEATURE_EXCLUSIVE_LOCK |	\
124 125
				 RBD_FEATURE_DATA_POOL |	\
				 RBD_FEATURE_OPERATIONS)
A
Alex Elder 已提交
126 127 128

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

129
#define RBD_FEATURES_SUPPORTED	(RBD_FEATURES_ALL)
A
Alex Elder 已提交
130

A
Alex Elder 已提交
131 132 133 134
/*
 * An RBD device name will be "rbd#", where the "rbd" comes from
 * RBD_DRV_NAME above, and # is a unique integer identifier.
 */
135 136 137 138 139 140
#define DEV_NAME_LEN		32

/*
 * block device image metadata (in-memory version)
 */
struct rbd_image_header {
141
	/* These six fields never change for a given rbd image */
142
	char *object_prefix;
143
	__u8 obj_order;
144 145
	u64 stripe_unit;
	u64 stripe_count;
146
	s64 data_pool_id;
147
	u64 features;		/* Might be changeable someday? */
148

A
Alex Elder 已提交
149 150 151
	/* The remaining fields need to be updated occasionally */
	u64 image_size;
	struct ceph_snap_context *snapc;
152 153
	char *snap_names;	/* format 1 only */
	u64 *snap_sizes;	/* format 1 only */
154 155
};

156 157 158 159
/*
 * An rbd image specification.
 *
 * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely
A
Alex Elder 已提交
160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179
 * 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).
180 181 182
 */
struct rbd_spec {
	u64		pool_id;
183
	const char	*pool_name;
184

185 186
	const char	*image_id;
	const char	*image_name;
187 188

	u64		snap_id;
189
	const char	*snap_name;
190 191 192 193

	struct kref	kref;
};

194
/*
A
Alex Elder 已提交
195
 * an instance of the client.  multiple devices may share an rbd client.
196 197 198 199 200 201 202
 */
struct rbd_client {
	struct ceph_client	*client;
	struct kref		kref;
	struct list_head	node;
};

A
Alex Elder 已提交
203 204
struct rbd_img_request;

205
enum obj_request_type {
206
	OBJ_REQUEST_NODATA = 1,
207
	OBJ_REQUEST_BIO,	/* pointer into provided bio (list) */
208
	OBJ_REQUEST_BVECS,	/* pointer into provided bio_vec array */
209
	OBJ_REQUEST_OWN_BVECS,	/* private bio_vec array, doesn't own pages */
210
};
A
Alex Elder 已提交
211

G
Guangliang Zhao 已提交
212
enum obj_operation_type {
213
	OBJ_OP_READ = 1,
G
Guangliang Zhao 已提交
214
	OBJ_OP_WRITE,
215
	OBJ_OP_DISCARD,
G
Guangliang Zhao 已提交
216 217
};

I
Ilya Dryomov 已提交
218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237
/*
 * 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,
238 239
};

A
Alex Elder 已提交
240
struct rbd_obj_request {
241
	struct ceph_object_extent ex;
242
	union {
I
Ilya Dryomov 已提交
243 244
		bool			tried_parent;	/* for reads */
		enum rbd_obj_write_state write_state;	/* for writes */
245
	};
A
Alex Elder 已提交
246

247
	struct rbd_img_request	*img_request;
I
Ilya Dryomov 已提交
248 249
	struct ceph_file_extent	*img_extents;
	u32			num_img_extents;
A
Alex Elder 已提交
250

251
	union {
252
		struct ceph_bio_iter	bio_pos;
253
		struct {
254 255
			struct ceph_bvec_iter	bvec_pos;
			u32			bvec_count;
256
			u32			bvec_idx;
257 258
		};
	};
259 260
	struct bio_vec		*copyup_bvecs;
	u32			copyup_bvec_count;
A
Alex Elder 已提交
261 262 263 264

	struct ceph_osd_request	*osd_req;

	u64			xferred;	/* bytes transferred */
265
	int			result;
A
Alex Elder 已提交
266 267 268 269

	struct kref		kref;
};

A
Alex Elder 已提交
270
enum img_req_flags {
271
	IMG_REQ_CHILD,		/* initiator: block = 0, child image = 1 */
272
	IMG_REQ_LAYERED,	/* ENOENT handling: normal = 0, layered = 1 */
A
Alex Elder 已提交
273 274
};

A
Alex Elder 已提交
275 276
struct rbd_img_request {
	struct rbd_device	*rbd_dev;
I
Ilya Dryomov 已提交
277
	enum obj_operation_type	op_type;
278
	enum obj_request_type	data_type;
A
Alex Elder 已提交
279
	unsigned long		flags;
A
Alex Elder 已提交
280
	union {
281
		u64			snap_id;	/* for reads */
A
Alex Elder 已提交
282
		struct ceph_snap_context *snapc;	/* for writes */
283 284 285 286
	};
	union {
		struct request		*rq;		/* block request */
		struct rbd_obj_request	*obj_request;	/* obj req initiator */
A
Alex Elder 已提交
287
	};
288
	spinlock_t		completion_lock;
289
	u64			xferred;/* aggregate bytes transferred */
290
	int			result;	/* first nonzero obj_request result */
A
Alex Elder 已提交
291

292
	struct list_head	object_extents;	/* obj_req.ex structs */
A
Alex Elder 已提交
293
	u32			obj_request_count;
I
Ilya Dryomov 已提交
294
	u32			pending_count;
A
Alex Elder 已提交
295 296 297 298 299

	struct kref		kref;
};

#define for_each_obj_request(ireq, oreq) \
300
	list_for_each_entry(oreq, &(ireq)->object_extents, ex.oe_item)
A
Alex Elder 已提交
301
#define for_each_obj_request_safe(ireq, oreq, n) \
302
	list_for_each_entry_safe(oreq, n, &(ireq)->object_extents, ex.oe_item)
A
Alex Elder 已提交
303

304 305 306 307 308 309
enum rbd_watch_state {
	RBD_WATCH_STATE_UNREGISTERED,
	RBD_WATCH_STATE_REGISTERED,
	RBD_WATCH_STATE_ERROR,
};

310 311 312 313 314 315 316 317 318 319 320 321
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
322
struct rbd_mapping {
A
Alex Elder 已提交
323
	u64                     size;
A
Alex Elder 已提交
324
	u64                     features;
A
Alex Elder 已提交
325 326
};

327 328 329 330
/*
 * a single device
 */
struct rbd_device {
A
Alex Elder 已提交
331
	int			dev_id;		/* blkdev unique id */
332 333

	int			major;		/* blkdev assigned major */
334
	int			minor;
335 336
	struct gendisk		*disk;		/* blkdev's gendisk and rq */

337
	u32			image_format;	/* Either 1 or 2 */
338 339 340 341
	struct rbd_client	*rbd_client;

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

342
	spinlock_t		lock;		/* queue, flags, open_count */
343 344

	struct rbd_image_header	header;
345
	unsigned long		flags;		/* possibly lock protected */
346
	struct rbd_spec		*spec;
347
	struct rbd_options	*opts;
348
	char			*config_info;	/* add{,_single_major} string */
349

350
	struct ceph_object_id	header_oid;
351
	struct ceph_object_locator header_oloc;
352

353
	struct ceph_file_layout	layout;		/* used for all rbd requests */
354

355 356
	struct mutex		watch_mutex;
	enum rbd_watch_state	watch_state;
357
	struct ceph_osd_linger_request *watch_handle;
358 359
	u64			watch_cookie;
	struct delayed_work	watch_dwork;
360

361 362
	struct rw_semaphore	lock_rwsem;
	enum rbd_lock_state	lock_state;
I
Ilya Dryomov 已提交
363
	char			lock_cookie[32];
364 365 366 367 368 369 370
	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;

371
	struct workqueue_struct	*task_wq;
372

373 374
	struct rbd_spec		*parent_spec;
	u64			parent_overlap;
375
	atomic_t		parent_ref;
376
	struct rbd_device	*parent;
377

C
Christoph Hellwig 已提交
378 379 380
	/* Block layer tags. */
	struct blk_mq_tag_set	tag_set;

381 382
	/* protects updating the header */
	struct rw_semaphore     header_rwsem;
A
Alex Elder 已提交
383 384

	struct rbd_mapping	mapping;
385 386

	struct list_head	node;
387 388 389

	/* sysfs related */
	struct device		dev;
390
	unsigned long		open_count;	/* protected by lock */
391 392
};

393
/*
394 395 396 397
 * 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
398
 */
399 400
enum rbd_dev_flags {
	RBD_DEV_FLAG_EXISTS,	/* mapped snapshot has not been deleted */
401
	RBD_DEV_FLAG_REMOVING,	/* this mapping is being removed */
402
	RBD_DEV_FLAG_BLACKLISTED, /* our ceph_client is blacklisted */
403 404
};

405
static DEFINE_MUTEX(client_mutex);	/* Serialize client creation */
406

407
static LIST_HEAD(rbd_dev_list);    /* devices */
408 409
static DEFINE_SPINLOCK(rbd_dev_list_lock);

A
Alex Elder 已提交
410 411
static LIST_HEAD(rbd_client_list);		/* clients */
static DEFINE_SPINLOCK(rbd_client_list_lock);
412

413 414
/* Slab caches for frequently-allocated structures */

415
static struct kmem_cache	*rbd_img_request_cache;
416
static struct kmem_cache	*rbd_obj_request_cache;
417

418
static int rbd_major;
419 420
static DEFINE_IDA(rbd_dev_id_ida);

421 422
static struct workqueue_struct *rbd_wq;

423
/*
424
 * single-major requires >= 0.75 version of userspace rbd utility.
425
 */
426
static bool single_major = true;
427
module_param(single_major, bool, S_IRUGO);
428
MODULE_PARM_DESC(single_major, "Use a single major number for all rbd devices (default: true)");
429

A
Alex Elder 已提交
430 431 432 433
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);
434 435 436 437
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);
438
static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth);
A
Alex Elder 已提交
439

440 441
static int rbd_dev_id_to_minor(int dev_id)
{
442
	return dev_id << RBD_SINGLE_MAJOR_PART_SHIFT;
443 444 445 446
}

static int minor_to_rbd_dev_id(int minor)
{
447
	return minor >> RBD_SINGLE_MAJOR_PART_SHIFT;
448 449
}

450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465
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;
}

466 467 468 469 470
static ssize_t rbd_supported_features_show(struct bus_type *bus, char *buf)
{
	return sprintf(buf, "0x%llx\n", RBD_FEATURES_SUPPORTED);
}

471 472
static BUS_ATTR(add, S_IWUSR, NULL, rbd_add);
static BUS_ATTR(remove, S_IWUSR, NULL, rbd_remove);
473 474
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);
475
static BUS_ATTR(supported_features, S_IRUGO, rbd_supported_features_show, NULL);
476 477 478 479

static struct attribute *rbd_bus_attrs[] = {
	&bus_attr_add.attr,
	&bus_attr_remove.attr,
480 481
	&bus_attr_add_single_major.attr,
	&bus_attr_remove_single_major.attr,
482
	&bus_attr_supported_features.attr,
483
	NULL,
A
Alex Elder 已提交
484
};
485 486 487 488

static umode_t rbd_bus_is_visible(struct kobject *kobj,
				  struct attribute *attr, int index)
{
489 490 491 492 493
	if (!single_major &&
	    (attr == &bus_attr_add_single_major.attr ||
	     attr == &bus_attr_remove_single_major.attr))
		return 0;

494 495 496 497 498 499 500 501
	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
502 503 504

static struct bus_type rbd_bus_type = {
	.name		= "rbd",
505
	.bus_groups	= rbd_bus_groups,
A
Alex Elder 已提交
506 507 508 509 510 511 512 513 514 515 516
};

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
517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543
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
544 545 546 547 548 549 550 551 552 553 554 555
#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 */
556

557
static void rbd_dev_remove_parent(struct rbd_device *rbd_dev);
A
Alex Elder 已提交
558

A
Alex Elder 已提交
559
static int rbd_dev_refresh(struct rbd_device *rbd_dev);
560
static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev);
561
static int rbd_dev_header_info(struct rbd_device *rbd_dev);
562
static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev);
563 564
static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
					u64 snap_id);
565 566 567 568
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);
569

570 571
static int rbd_open(struct block_device *bdev, fmode_t mode)
{
A
Alex Elder 已提交
572
	struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
573
	bool removing = false;
574

575
	spin_lock_irq(&rbd_dev->lock);
576 577 578 579
	if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags))
		removing = true;
	else
		rbd_dev->open_count++;
580
	spin_unlock_irq(&rbd_dev->lock);
581 582 583
	if (removing)
		return -ENOENT;

A
Alex Elder 已提交
584
	(void) get_device(&rbd_dev->dev);
585

586 587 588
	return 0;
}

589
static void rbd_release(struct gendisk *disk, fmode_t mode)
590 591
{
	struct rbd_device *rbd_dev = disk->private_data;
592 593
	unsigned long open_count_before;

594
	spin_lock_irq(&rbd_dev->lock);
595
	open_count_before = rbd_dev->open_count--;
596
	spin_unlock_irq(&rbd_dev->lock);
597
	rbd_assert(open_count_before > 0);
598

A
Alex Elder 已提交
599
	put_device(&rbd_dev->dev);
600 601
}

G
Guangliang Zhao 已提交
602 603
static int rbd_ioctl_set_ro(struct rbd_device *rbd_dev, unsigned long arg)
{
604
	int ro;
G
Guangliang Zhao 已提交
605

606
	if (get_user(ro, (int __user *)arg))
G
Guangliang Zhao 已提交
607 608
		return -EFAULT;

609
	/* Snapshots can't be marked read-write */
G
Guangliang Zhao 已提交
610 611 612
	if (rbd_dev->spec->snap_id != CEPH_NOSNAP && !ro)
		return -EROFS;

613 614
	/* Let blkdev_roset() handle it */
	return -ENOTTY;
G
Guangliang Zhao 已提交
615 616 617 618 619 620
}

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;
621
	int ret;
G
Guangliang Zhao 已提交
622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641

	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 */

642 643 644
static const struct block_device_operations rbd_bd_ops = {
	.owner			= THIS_MODULE,
	.open			= rbd_open,
645
	.release		= rbd_release,
G
Guangliang Zhao 已提交
646 647 648 649
	.ioctl			= rbd_ioctl,
#ifdef CONFIG_COMPAT
	.compat_ioctl		= rbd_compat_ioctl,
#endif
650 651 652
};

/*
653
 * Initialize an rbd client instance.  Success or not, this function
654
 * consumes ceph_opts.  Caller holds client_mutex.
655
 */
656
static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts)
657 658 659 660
{
	struct rbd_client *rbdc;
	int ret = -ENOMEM;

A
Alex Elder 已提交
661
	dout("%s:\n", __func__);
662 663 664 665 666 667 668
	rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL);
	if (!rbdc)
		goto out_opt;

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

669
	rbdc->client = ceph_create_client(ceph_opts, rbdc);
670
	if (IS_ERR(rbdc->client))
671
		goto out_rbdc;
A
Alex Elder 已提交
672
	ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */
673 674 675

	ret = ceph_open_session(rbdc->client);
	if (ret < 0)
676
		goto out_client;
677

A
Alex Elder 已提交
678
	spin_lock(&rbd_client_list_lock);
679
	list_add_tail(&rbdc->node, &rbd_client_list);
A
Alex Elder 已提交
680
	spin_unlock(&rbd_client_list_lock);
681

A
Alex Elder 已提交
682
	dout("%s: rbdc %p\n", __func__, rbdc);
683

684
	return rbdc;
685
out_client:
686
	ceph_destroy_client(rbdc->client);
687
out_rbdc:
688 689
	kfree(rbdc);
out_opt:
A
Alex Elder 已提交
690 691
	if (ceph_opts)
		ceph_destroy_options(ceph_opts);
A
Alex Elder 已提交
692 693
	dout("%s: error %d\n", __func__, ret);

V
Vasiliy Kulikov 已提交
694
	return ERR_PTR(ret);
695 696
}

697 698 699 700 701 702 703
static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc)
{
	kref_get(&rbdc->kref);

	return rbdc;
}

704
/*
705 706
 * Find a ceph client with specific addr and configuration.  If
 * found, bump its reference count.
707
 */
708
static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts)
709 710
{
	struct rbd_client *client_node;
711
	bool found = false;
712

A
Alex Elder 已提交
713
	if (ceph_opts->flags & CEPH_OPT_NOSHARE)
714 715
		return NULL;

716 717 718
	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)) {
719 720
			__rbd_get_client(client_node);

721 722 723 724 725 726 727
			found = true;
			break;
		}
	}
	spin_unlock(&rbd_client_list_lock);

	return found ? client_node : NULL;
728 729
}

730
/*
731
 * (Per device) rbd map options
732 733
 */
enum {
I
Ilya Dryomov 已提交
734
	Opt_queue_depth,
735
	Opt_lock_timeout,
736 737 738 739
	Opt_last_int,
	/* int args above */
	Opt_last_string,
	/* string args above */
A
Alex Elder 已提交
740 741
	Opt_read_only,
	Opt_read_write,
I
Ilya Dryomov 已提交
742
	Opt_lock_on_read,
I
Ilya Dryomov 已提交
743
	Opt_exclusive,
744
	Opt_err
745 746
};

A
Alex Elder 已提交
747
static match_table_t rbd_opts_tokens = {
I
Ilya Dryomov 已提交
748
	{Opt_queue_depth, "queue_depth=%d"},
749
	{Opt_lock_timeout, "lock_timeout=%d"},
750 751
	/* int args above */
	/* string args above */
A
Alex Elder 已提交
752
	{Opt_read_only, "read_only"},
A
Alex Elder 已提交
753 754 755
	{Opt_read_only, "ro"},		/* Alternate spelling */
	{Opt_read_write, "read_write"},
	{Opt_read_write, "rw"},		/* Alternate spelling */
I
Ilya Dryomov 已提交
756
	{Opt_lock_on_read, "lock_on_read"},
I
Ilya Dryomov 已提交
757
	{Opt_exclusive, "exclusive"},
758
	{Opt_err, NULL}
759 760
};

A
Alex Elder 已提交
761
struct rbd_options {
I
Ilya Dryomov 已提交
762
	int	queue_depth;
763
	unsigned long	lock_timeout;
A
Alex Elder 已提交
764
	bool	read_only;
I
Ilya Dryomov 已提交
765
	bool	lock_on_read;
I
Ilya Dryomov 已提交
766
	bool	exclusive;
A
Alex Elder 已提交
767 768
};

I
Ilya Dryomov 已提交
769
#define RBD_QUEUE_DEPTH_DEFAULT	BLKDEV_MAX_RQ
770
#define RBD_LOCK_TIMEOUT_DEFAULT 0  /* no timeout */
A
Alex Elder 已提交
771
#define RBD_READ_ONLY_DEFAULT	false
I
Ilya Dryomov 已提交
772
#define RBD_LOCK_ON_READ_DEFAULT false
I
Ilya Dryomov 已提交
773
#define RBD_EXCLUSIVE_DEFAULT	false
A
Alex Elder 已提交
774

775 776
static int parse_rbd_opts_token(char *c, void *private)
{
A
Alex Elder 已提交
777
	struct rbd_options *rbd_opts = private;
778 779 780
	substring_t argstr[MAX_OPT_ARGS];
	int token, intval, ret;

A
Alex Elder 已提交
781
	token = match_token(c, rbd_opts_tokens, argstr);
782 783 784
	if (token < Opt_last_int) {
		ret = match_int(&argstr[0], &intval);
		if (ret < 0) {
785
			pr_err("bad mount option arg (not int) at '%s'\n", c);
786 787 788 789
			return ret;
		}
		dout("got int token %d val %d\n", token, intval);
	} else if (token > Opt_last_int && token < Opt_last_string) {
790
		dout("got string token %d val %s\n", token, argstr[0].from);
791 792 793 794 795
	} else {
		dout("got token %d\n", token);
	}

	switch (token) {
I
Ilya Dryomov 已提交
796 797 798 799 800 801 802
	case Opt_queue_depth:
		if (intval < 1) {
			pr_err("queue_depth out of range\n");
			return -EINVAL;
		}
		rbd_opts->queue_depth = intval;
		break;
803 804 805 806 807 808 809 810
	case Opt_lock_timeout:
		/* 0 is "wait forever" (i.e. infinite timeout) */
		if (intval < 0 || intval > INT_MAX / 1000) {
			pr_err("lock_timeout out of range\n");
			return -EINVAL;
		}
		rbd_opts->lock_timeout = msecs_to_jiffies(intval * 1000);
		break;
A
Alex Elder 已提交
811 812 813 814 815 816
	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
817 818 819
	case Opt_lock_on_read:
		rbd_opts->lock_on_read = true;
		break;
I
Ilya Dryomov 已提交
820 821 822
	case Opt_exclusive:
		rbd_opts->exclusive = true;
		break;
823
	default:
824 825
		/* libceph prints "bad option" msg */
		return -EINVAL;
826
	}
827

828 829 830
	return 0;
}

G
Guangliang Zhao 已提交
831 832 833 834 835 836 837
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";
838 839
	case OBJ_OP_DISCARD:
		return "discard";
G
Guangliang Zhao 已提交
840 841 842 843 844
	default:
		return "???";
	}
}

845 846
/*
 * Destroy ceph client
A
Alex Elder 已提交
847
 *
A
Alex Elder 已提交
848
 * Caller must hold rbd_client_list_lock.
849 850 851 852 853
 */
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
854
	dout("%s: rbdc %p\n", __func__, rbdc);
855
	spin_lock(&rbd_client_list_lock);
856
	list_del(&rbdc->node);
857
	spin_unlock(&rbd_client_list_lock);
858 859 860 861 862 863 864 865 866

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

/*
 * Drop reference to ceph client node. If it's not referenced anymore, release
 * it.
 */
867
static void rbd_put_client(struct rbd_client *rbdc)
868
{
869 870
	if (rbdc)
		kref_put(&rbdc->kref, rbd_client_release);
871 872
}

873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889
static int wait_for_latest_osdmap(struct ceph_client *client)
{
	u64 newest_epoch;
	int ret;

	ret = ceph_monc_get_version(&client->monc, "osdmap", &newest_epoch);
	if (ret)
		return ret;

	if (client->osdc.osdmap->epoch >= newest_epoch)
		return 0;

	ceph_osdc_maybe_request_map(&client->osdc);
	return ceph_monc_wait_osdmap(&client->monc, newest_epoch,
				     client->options->mount_timeout);
}

890 891 892 893 894 895 896 897
/*
 * Get a ceph client with specific addr and configuration, if one does
 * not exist create it.  Either way, ceph_opts is consumed by this
 * function.
 */
static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts)
{
	struct rbd_client *rbdc;
898
	int ret;
899 900 901

	mutex_lock_nested(&client_mutex, SINGLE_DEPTH_NESTING);
	rbdc = rbd_client_find(ceph_opts);
902
	if (rbdc) {
903
		ceph_destroy_options(ceph_opts);
904 905 906 907 908 909 910 911 912 913 914 915

		/*
		 * Using an existing client.  Make sure ->pg_pools is up to
		 * date before we look up the pool id in do_rbd_add().
		 */
		ret = wait_for_latest_osdmap(rbdc->client);
		if (ret) {
			rbd_warn(NULL, "failed to get latest osdmap: %d", ret);
			rbd_put_client(rbdc);
			rbdc = ERR_PTR(ret);
		}
	} else {
916
		rbdc = rbd_client_create(ceph_opts);
917
	}
918 919 920 921 922
	mutex_unlock(&client_mutex);

	return rbdc;
}

923 924 925 926 927
static bool rbd_image_format_valid(u32 image_format)
{
	return image_format == 1 || image_format == 2;
}

928 929
static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk)
{
930 931 932 933 934 935 936
	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
937 938 939 940 941 942 943 944 945 946
	/* 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;

947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964
	/*
	 * 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;
965 966
}

I
Ilya Dryomov 已提交
967 968 969 970 971 972 973 974
/*
 * 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
975 976 977 978 979 980 981 982 983 984 985
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);
986 987
	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
988 989 990
	RCU_INIT_POINTER(rbd_dev->layout.pool_ns, NULL);
}

991
/*
992 993
 * Fill an rbd image header with information from the given format 1
 * on-disk header.
994
 */
A
Alex Elder 已提交
995
static int rbd_header_from_disk(struct rbd_device *rbd_dev,
996
				 struct rbd_image_header_ondisk *ondisk)
997
{
A
Alex Elder 已提交
998
	struct rbd_image_header *header = &rbd_dev->header;
999 1000 1001 1002 1003
	bool first_time = header->object_prefix == NULL;
	struct ceph_snap_context *snapc;
	char *object_prefix = NULL;
	char *snap_names = NULL;
	u64 *snap_sizes = NULL;
1004
	u32 snap_count;
1005
	int ret = -ENOMEM;
1006
	u32 i;
1007

1008
	/* Allocate this now to avoid having to handle failure below */
A
Alex Elder 已提交
1009

1010
	if (first_time) {
1011 1012 1013
		object_prefix = kstrndup(ondisk->object_prefix,
					 sizeof(ondisk->object_prefix),
					 GFP_KERNEL);
1014 1015 1016
		if (!object_prefix)
			return -ENOMEM;
	}
A
Alex Elder 已提交
1017

1018
	/* Allocate the snapshot context and fill it in */
A
Alex Elder 已提交
1019

1020 1021 1022 1023 1024
	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);
1025
	if (snap_count) {
1026
		struct rbd_image_snap_ondisk *snaps;
A
Alex Elder 已提交
1027 1028
		u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len);

1029
		/* We'll keep a copy of the snapshot names... */
1030

1031 1032 1033 1034
		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
1035 1036
			goto out_err;

1037
		/* ...as well as the array of their sizes. */
1038 1039 1040
		snap_sizes = kmalloc_array(snap_count,
					   sizeof(*header->snap_sizes),
					   GFP_KERNEL);
1041
		if (!snap_sizes)
A
Alex Elder 已提交
1042
			goto out_err;
1043

A
Alex Elder 已提交
1044
		/*
1045 1046 1047
		 * Copy the names, and fill in each snapshot's id
		 * and size.
		 *
1048
		 * Note that rbd_dev_v1_header_info() guarantees the
1049
		 * ondisk buffer we're working with has
A
Alex Elder 已提交
1050 1051 1052
		 * snap_names_len bytes beyond the end of the
		 * snapshot id array, this memcpy() is safe.
		 */
1053 1054 1055 1056 1057 1058
		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);
		}
1059
	}
A
Alex Elder 已提交
1060

1061
	/* We won't fail any more, fill in the header */
1062

1063 1064 1065
	if (first_time) {
		header->object_prefix = object_prefix;
		header->obj_order = ondisk->options.order;
I
Ilya Dryomov 已提交
1066
		rbd_init_layout(rbd_dev);
1067
	} else {
A
Alex Elder 已提交
1068 1069 1070
		ceph_put_snap_context(header->snapc);
		kfree(header->snap_names);
		kfree(header->snap_sizes);
1071
	}
1072

1073
	/* The remaining fields always get updated (when we refresh) */
1074

A
Alex Elder 已提交
1075
	header->image_size = le64_to_cpu(ondisk->image_size);
1076 1077 1078
	header->snapc = snapc;
	header->snap_names = snap_names;
	header->snap_sizes = snap_sizes;
1079

1080
	return 0;
1081 1082
out_2big:
	ret = -EIO;
A
Alex Elder 已提交
1083
out_err:
1084 1085 1086 1087
	kfree(snap_sizes);
	kfree(snap_names);
	ceph_put_snap_context(snapc);
	kfree(object_prefix);
1088

1089
	return ret;
1090 1091
}

1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106
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);
}

1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130
/*
 * 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.
 */
1131 1132 1133
static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id)
{
	struct ceph_snap_context *snapc = rbd_dev->header.snapc;
1134
	u64 *found;
1135

1136 1137
	found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps,
				sizeof (snap_id), snapid_compare_reverse);
1138

1139
	return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX;
1140 1141
}

1142 1143
static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev,
					u64 snap_id)
1144
{
1145
	u32 which;
1146
	const char *snap_name;
1147

1148 1149
	which = rbd_dev_snap_index(rbd_dev, snap_id);
	if (which == BAD_SNAP_INDEX)
1150
		return ERR_PTR(-ENOENT);
1151

1152 1153
	snap_name = _rbd_dev_v1_snap_name(rbd_dev, which);
	return snap_name ? snap_name : ERR_PTR(-ENOMEM);
1154 1155 1156 1157
}

static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id)
{
1158 1159 1160
	if (snap_id == CEPH_NOSNAP)
		return RBD_SNAP_HEAD_NAME;

1161 1162 1163
	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);
1164

1165
	return rbd_dev_v2_snap_name(rbd_dev, snap_id);
1166 1167
}

1168 1169
static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
				u64 *snap_size)
1170
{
1171 1172 1173 1174 1175
	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;
1176

1177 1178 1179
		which = rbd_dev_snap_index(rbd_dev, snap_id);
		if (which == BAD_SNAP_INDEX)
			return -ENOENT;
1180

1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192
		*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;
1193 1194
}

1195 1196
static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
			u64 *snap_features)
1197
{
1198 1199 1200 1201 1202
	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 */
1203
	} else {
1204 1205
		u64 features = 0;
		int ret;
1206

1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217
		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)
{
1218
	u64 snap_id = rbd_dev->spec->snap_id;
1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232
	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;

1233
	return 0;
1234 1235
}

A
Alex Elder 已提交
1236 1237 1238 1239
static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev)
{
	rbd_dev->mapping.size = 0;
	rbd_dev->mapping.features = 0;
1240 1241
}

1242
static void zero_bvec(struct bio_vec *bv)
1243 1244
{
	void *buf;
1245
	unsigned long flags;
1246

1247 1248 1249 1250
	buf = bvec_kmap_irq(bv, &flags);
	memset(buf, 0, bv->bv_len);
	flush_dcache_page(bv->bv_page);
	bvec_kunmap_irq(buf, &flags);
1251 1252
}

1253
static void zero_bios(struct ceph_bio_iter *bio_pos, u32 off, u32 bytes)
A
Alex Elder 已提交
1254
{
1255
	struct ceph_bio_iter it = *bio_pos;
A
Alex Elder 已提交
1256

1257 1258 1259 1260
	ceph_bio_iter_advance(&it, off);
	ceph_bio_iter_advance_step(&it, bytes, ({
		zero_bvec(&bv);
	}));
A
Alex Elder 已提交
1261 1262
}

1263
static void zero_bvecs(struct ceph_bvec_iter *bvec_pos, u32 off, u32 bytes)
1264
{
1265
	struct ceph_bvec_iter it = *bvec_pos;
1266

1267 1268 1269 1270
	ceph_bvec_iter_advance(&it, off);
	ceph_bvec_iter_advance_step(&it, bytes, ({
		zero_bvec(&bv);
	}));
A
Alex Elder 已提交
1271 1272 1273
}

/*
I
Ilya Dryomov 已提交
1274
 * Zero a range in @obj_req data buffer defined by a bio (list) or
1275
 * (private) bio_vec array.
A
Alex Elder 已提交
1276
 *
I
Ilya Dryomov 已提交
1277
 * @off is relative to the start of the data buffer.
1278
 */
I
Ilya Dryomov 已提交
1279 1280
static void rbd_obj_zero_range(struct rbd_obj_request *obj_req, u32 off,
			       u32 bytes)
1281
{
1282
	switch (obj_req->img_request->data_type) {
I
Ilya Dryomov 已提交
1283 1284 1285 1286
	case OBJ_REQUEST_BIO:
		zero_bios(&obj_req->bio_pos, off, bytes);
		break;
	case OBJ_REQUEST_BVECS:
1287
	case OBJ_REQUEST_OWN_BVECS:
I
Ilya Dryomov 已提交
1288 1289 1290 1291
		zero_bvecs(&obj_req->bvec_pos, off, bytes);
		break;
	default:
		rbd_assert(0);
1292 1293 1294
	}
}

A
Alex Elder 已提交
1295 1296 1297 1298
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
1299
	dout("%s: obj %p (was %d)\n", __func__, obj_request,
1300
		kref_read(&obj_request->kref));
A
Alex Elder 已提交
1301 1302 1303
	kref_put(&obj_request->kref, rbd_obj_request_destroy);
}

1304 1305 1306
static void rbd_img_request_get(struct rbd_img_request *img_request)
{
	dout("%s: img %p (was %d)\n", __func__, img_request,
1307
	     kref_read(&img_request->kref));
1308 1309 1310
	kref_get(&img_request->kref);
}

A
Alex Elder 已提交
1311 1312 1313 1314
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
1315
	dout("%s: img %p (was %d)\n", __func__, img_request,
1316
		kref_read(&img_request->kref));
1317
	kref_put(&img_request->kref, rbd_img_request_destroy);
A
Alex Elder 已提交
1318 1319 1320 1321 1322
}

static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request,
					struct rbd_obj_request *obj_request)
{
1323 1324
	rbd_assert(obj_request->img_request == NULL);

1325
	/* Image request now owns object's original reference */
A
Alex Elder 已提交
1326
	obj_request->img_request = img_request;
1327
	img_request->obj_request_count++;
I
Ilya Dryomov 已提交
1328
	img_request->pending_count++;
1329
	dout("%s: img %p obj %p\n", __func__, img_request, obj_request);
A
Alex Elder 已提交
1330 1331 1332 1333 1334
}

static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request,
					struct rbd_obj_request *obj_request)
{
1335
	dout("%s: img %p obj %p\n", __func__, img_request, obj_request);
1336
	list_del(&obj_request->ex.oe_item);
1337 1338
	rbd_assert(img_request->obj_request_count > 0);
	img_request->obj_request_count--;
A
Alex Elder 已提交
1339 1340 1341 1342
	rbd_assert(obj_request->img_request == img_request);
	rbd_obj_request_put(obj_request);
}

1343
static void rbd_obj_request_submit(struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1344
{
1345 1346
	struct ceph_osd_request *osd_req = obj_request->osd_req;

1347
	dout("%s %p object_no %016llx %llu~%llu osd_req %p\n", __func__,
1348 1349
	     obj_request, obj_request->ex.oe_objno, obj_request->ex.oe_off,
	     obj_request->ex.oe_len, osd_req);
1350
	ceph_osdc_start_request(osd_req->r_osdc, osd_req, false);
A
Alex Elder 已提交
1351 1352
}

A
Alex Elder 已提交
1353 1354 1355 1356 1357
/*
 * 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.
 */
1358 1359 1360 1361 1362 1363
static void img_request_layered_set(struct rbd_img_request *img_request)
{
	set_bit(IMG_REQ_LAYERED, &img_request->flags);
	smp_mb();
}

1364 1365 1366 1367 1368 1369
static void img_request_layered_clear(struct rbd_img_request *img_request)
{
	clear_bit(IMG_REQ_LAYERED, &img_request->flags);
	smp_mb();
}

1370 1371 1372 1373 1374 1375
static bool img_request_layered_test(struct rbd_img_request *img_request)
{
	smp_mb();
	return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0;
}

I
Ilya Dryomov 已提交
1376
static bool rbd_obj_is_entire(struct rbd_obj_request *obj_req)
1377
{
I
Ilya Dryomov 已提交
1378
	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
A
Alex Elder 已提交
1379

1380 1381
	return !obj_req->ex.oe_off &&
	       obj_req->ex.oe_len == rbd_dev->layout.object_size;
1382 1383
}

I
Ilya Dryomov 已提交
1384
static bool rbd_obj_is_tail(struct rbd_obj_request *obj_req)
A
Alex Elder 已提交
1385
{
I
Ilya Dryomov 已提交
1386
	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
A
Alex Elder 已提交
1387

1388
	return obj_req->ex.oe_off + obj_req->ex.oe_len ==
I
Ilya Dryomov 已提交
1389
					rbd_dev->layout.object_size;
1390 1391
}

I
Ilya Dryomov 已提交
1392
static u64 rbd_obj_img_extents_bytes(struct rbd_obj_request *obj_req)
A
Alex Elder 已提交
1393
{
I
Ilya Dryomov 已提交
1394 1395
	return ceph_file_extents_bytes(obj_req->img_extents,
				       obj_req->num_img_extents);
A
Alex Elder 已提交
1396 1397
}

I
Ilya Dryomov 已提交
1398
static bool rbd_img_is_write(struct rbd_img_request *img_req)
A
Alex Elder 已提交
1399
{
I
Ilya Dryomov 已提交
1400
	switch (img_req->op_type) {
I
Ilya Dryomov 已提交
1401 1402 1403 1404 1405 1406
	case OBJ_OP_READ:
		return false;
	case OBJ_OP_WRITE:
	case OBJ_OP_DISCARD:
		return true;
	default:
A
Arnd Bergmann 已提交
1407
		BUG();
I
Ilya Dryomov 已提交
1408
	}
1409 1410
}

I
Ilya Dryomov 已提交
1411
static void rbd_obj_handle_request(struct rbd_obj_request *obj_req);
I
Ilya Dryomov 已提交
1412

1413
static void rbd_osd_req_callback(struct ceph_osd_request *osd_req)
A
Alex Elder 已提交
1414
{
I
Ilya Dryomov 已提交
1415
	struct rbd_obj_request *obj_req = osd_req->r_priv;
A
Alex Elder 已提交
1416

I
Ilya Dryomov 已提交
1417 1418 1419
	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);
A
Alex Elder 已提交
1420

I
Ilya Dryomov 已提交
1421 1422 1423 1424 1425 1426 1427 1428 1429 1430
	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
1431

I
Ilya Dryomov 已提交
1432
	rbd_obj_handle_request(obj_req);
A
Alex Elder 已提交
1433 1434
}

1435
static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1436
{
1437
	struct ceph_osd_request *osd_req = obj_request->osd_req;
A
Alex Elder 已提交
1438

1439
	osd_req->r_flags = CEPH_OSD_FLAG_READ;
1440
	osd_req->r_snapid = obj_request->img_request->snap_id;
1441 1442 1443 1444 1445 1446
}

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

1447
	osd_req->r_flags = CEPH_OSD_FLAG_WRITE;
1448
	ktime_get_real_ts(&osd_req->r_mtime);
1449
	osd_req->r_data_offset = obj_request->ex.oe_off;
A
Alex Elder 已提交
1450 1451
}

1452
static struct ceph_osd_request *
1453
rbd_osd_req_create(struct rbd_obj_request *obj_req, unsigned int num_ops)
1454
{
1455 1456
	struct rbd_img_request *img_req = obj_req->img_request;
	struct rbd_device *rbd_dev = img_req->rbd_dev;
1457 1458
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
	struct ceph_osd_request *req;
1459 1460
	const char *name_format = rbd_dev->image_format == 1 ?
				      RBD_V1_DATA_FORMAT : RBD_V2_DATA_FORMAT;
1461

1462 1463 1464
	req = ceph_osdc_alloc_request(osdc,
			(rbd_img_is_write(img_req) ? img_req->snapc : NULL),
			num_ops, false, GFP_NOIO);
1465 1466 1467 1468
	if (!req)
		return NULL;

	req->r_callback = rbd_osd_req_callback;
1469
	req->r_priv = obj_req;
1470 1471

	req->r_base_oloc.pool = rbd_dev->layout.pool_id;
1472
	if (ceph_oid_aprintf(&req->r_base_oid, GFP_NOIO, name_format,
1473
			rbd_dev->header.object_prefix, obj_req->ex.oe_objno))
1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485
		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
1486 1487 1488 1489 1490
static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req)
{
	ceph_osdc_put_request(osd_req);
}

1491
static struct rbd_obj_request *rbd_obj_request_create(void)
A
Alex Elder 已提交
1492 1493 1494
{
	struct rbd_obj_request *obj_request;

1495
	obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_NOIO);
1496
	if (!obj_request)
1497 1498
		return NULL;

1499
	ceph_object_extent_init(&obj_request->ex);
A
Alex Elder 已提交
1500 1501
	kref_init(&obj_request->kref);

1502
	dout("%s %p\n", __func__, obj_request);
A
Alex Elder 已提交
1503 1504 1505 1506 1507 1508
	return obj_request;
}

static void rbd_obj_request_destroy(struct kref *kref)
{
	struct rbd_obj_request *obj_request;
1509
	u32 i;
A
Alex Elder 已提交
1510 1511 1512

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

A
Alex Elder 已提交
1513 1514
	dout("%s: obj %p\n", __func__, obj_request);

A
Alex Elder 已提交
1515 1516 1517
	if (obj_request->osd_req)
		rbd_osd_req_destroy(obj_request->osd_req);

1518
	switch (obj_request->img_request->data_type) {
1519
	case OBJ_REQUEST_NODATA:
A
Alex Elder 已提交
1520
	case OBJ_REQUEST_BIO:
1521
	case OBJ_REQUEST_BVECS:
1522
		break;		/* Nothing to do */
1523 1524
	case OBJ_REQUEST_OWN_BVECS:
		kfree(obj_request->bvec_pos.bvecs);
1525
		break;
1526 1527
	default:
		rbd_assert(0);
A
Alex Elder 已提交
1528 1529
	}

I
Ilya Dryomov 已提交
1530
	kfree(obj_request->img_extents);
1531 1532 1533 1534 1535 1536
	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);
A
Alex Elder 已提交
1537 1538
	}

1539
	kmem_cache_free(rbd_obj_request_cache, obj_request);
A
Alex Elder 已提交
1540 1541
}

A
Alex Elder 已提交
1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552
/* 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;
}

1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574
/*
 * 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
1575
		rbd_warn(rbd_dev, "parent reference underflow");
1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587
}

/*
 * 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)
{
1588
	int counter = 0;
1589 1590 1591 1592

	if (!rbd_dev->parent_spec)
		return false;

1593 1594 1595 1596
	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);
1597 1598

	if (counter < 0)
1599
		rbd_warn(rbd_dev, "parent reference overflow");
1600

1601
	return counter > 0;
1602 1603
}

A
Alex Elder 已提交
1604 1605 1606 1607 1608
/*
 * 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
1609 1610
static struct rbd_img_request *rbd_img_request_create(
					struct rbd_device *rbd_dev,
G
Guangliang Zhao 已提交
1611
					enum obj_operation_type op_type,
1612
					struct ceph_snap_context *snapc)
A
Alex Elder 已提交
1613 1614 1615
{
	struct rbd_img_request *img_request;

1616
	img_request = kmem_cache_zalloc(rbd_img_request_cache, GFP_NOIO);
A
Alex Elder 已提交
1617 1618 1619 1620
	if (!img_request)
		return NULL;

	img_request->rbd_dev = rbd_dev;
I
Ilya Dryomov 已提交
1621 1622
	img_request->op_type = op_type;
	if (!rbd_img_is_write(img_request))
A
Alex Elder 已提交
1623
		img_request->snap_id = rbd_dev->spec->snap_id;
I
Ilya Dryomov 已提交
1624 1625 1626
	else
		img_request->snapc = snapc;

1627
	if (rbd_dev_parent_get(rbd_dev))
1628
		img_request_layered_set(img_request);
1629

A
Alex Elder 已提交
1630
	spin_lock_init(&img_request->completion_lock);
1631
	INIT_LIST_HEAD(&img_request->object_extents);
A
Alex Elder 已提交
1632 1633
	kref_init(&img_request->kref);

1634 1635
	dout("%s: rbd_dev %p %s -> img %p\n", __func__, rbd_dev,
	     obj_op_name(op_type), img_request);
A
Alex Elder 已提交
1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646
	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
1647 1648
	dout("%s: img %p\n", __func__, img_request);

A
Alex Elder 已提交
1649 1650
	for_each_obj_request_safe(img_request, obj_request, next_obj_request)
		rbd_img_obj_request_del(img_request, obj_request);
1651
	rbd_assert(img_request->obj_request_count == 0);
A
Alex Elder 已提交
1652

1653 1654 1655 1656 1657
	if (img_request_layered_test(img_request)) {
		img_request_layered_clear(img_request);
		rbd_dev_parent_put(img_request->rbd_dev);
	}

I
Ilya Dryomov 已提交
1658
	if (rbd_img_is_write(img_request))
1659
		ceph_put_snap_context(img_request->snapc);
A
Alex Elder 已提交
1660

1661
	kmem_cache_free(rbd_img_request_cache, img_request);
A
Alex Elder 已提交
1662 1663
}

I
Ilya Dryomov 已提交
1664 1665
static void prune_extents(struct ceph_file_extent *img_extents,
			  u32 *num_img_extents, u64 overlap)
1666
{
I
Ilya Dryomov 已提交
1667
	u32 cnt = *num_img_extents;
1668

I
Ilya Dryomov 已提交
1669 1670 1671
	/* drop extents completely beyond the overlap */
	while (cnt && img_extents[cnt - 1].fe_off >= overlap)
		cnt--;
1672

I
Ilya Dryomov 已提交
1673 1674
	if (cnt) {
		struct ceph_file_extent *ex = &img_extents[cnt - 1];
1675

I
Ilya Dryomov 已提交
1676 1677 1678 1679
		/* trim final overlapping extent */
		if (ex->fe_off + ex->fe_len > overlap)
			ex->fe_len = overlap - ex->fe_off;
	}
1680

I
Ilya Dryomov 已提交
1681
	*num_img_extents = cnt;
1682 1683
}

I
Ilya Dryomov 已提交
1684 1685 1686 1687 1688 1689
/*
 * Determine the byte range(s) covered by either just the object extent
 * or the entire object in the parent image.
 */
static int rbd_obj_calc_img_extents(struct rbd_obj_request *obj_req,
				    bool entire)
1690
{
I
Ilya Dryomov 已提交
1691 1692
	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
	int ret;
1693

I
Ilya Dryomov 已提交
1694 1695
	if (!rbd_dev->parent_overlap)
		return 0;
1696

I
Ilya Dryomov 已提交
1697 1698 1699 1700 1701 1702 1703 1704
	ret = ceph_extent_to_file(&rbd_dev->layout, obj_req->ex.oe_objno,
				  entire ? 0 : obj_req->ex.oe_off,
				  entire ? rbd_dev->layout.object_size :
							obj_req->ex.oe_len,
				  &obj_req->img_extents,
				  &obj_req->num_img_extents);
	if (ret)
		return ret;
1705

I
Ilya Dryomov 已提交
1706 1707 1708
	prune_extents(obj_req->img_extents, &obj_req->num_img_extents,
		      rbd_dev->parent_overlap);
	return 0;
1709 1710
}

I
Ilya Dryomov 已提交
1711
static void rbd_osd_req_setup_data(struct rbd_obj_request *obj_req, u32 which)
1712
{
1713
	switch (obj_req->img_request->data_type) {
I
Ilya Dryomov 已提交
1714 1715 1716
	case OBJ_REQUEST_BIO:
		osd_req_op_extent_osd_data_bio(obj_req->osd_req, which,
					       &obj_req->bio_pos,
1717
					       obj_req->ex.oe_len);
I
Ilya Dryomov 已提交
1718 1719
		break;
	case OBJ_REQUEST_BVECS:
1720
	case OBJ_REQUEST_OWN_BVECS:
I
Ilya Dryomov 已提交
1721
		rbd_assert(obj_req->bvec_pos.iter.bi_size ==
1722
							obj_req->ex.oe_len);
1723
		rbd_assert(obj_req->bvec_idx == obj_req->bvec_count);
I
Ilya Dryomov 已提交
1724 1725 1726 1727 1728
		osd_req_op_extent_osd_data_bvec_pos(obj_req->osd_req, which,
						    &obj_req->bvec_pos);
		break;
	default:
		rbd_assert(0);
1729
	}
I
Ilya Dryomov 已提交
1730
}
1731

I
Ilya Dryomov 已提交
1732 1733
static int rbd_obj_setup_read(struct rbd_obj_request *obj_req)
{
1734
	obj_req->osd_req = rbd_osd_req_create(obj_req, 1);
I
Ilya Dryomov 已提交
1735 1736
	if (!obj_req->osd_req)
		return -ENOMEM;
1737

I
Ilya Dryomov 已提交
1738
	osd_req_op_extent_init(obj_req->osd_req, 0, CEPH_OSD_OP_READ,
1739
			       obj_req->ex.oe_off, obj_req->ex.oe_len, 0, 0);
I
Ilya Dryomov 已提交
1740
	rbd_osd_req_setup_data(obj_req, 0);
C
Christoph Hellwig 已提交
1741

I
Ilya Dryomov 已提交
1742 1743 1744 1745 1746 1747 1748 1749
	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;
A
Alex Elder 已提交
1750

I
Ilya Dryomov 已提交
1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767
	/*
	 * 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;
1768 1769
}

I
Ilya Dryomov 已提交
1770 1771
static void __rbd_obj_setup_write(struct rbd_obj_request *obj_req,
				  unsigned int which)
1772
{
I
Ilya Dryomov 已提交
1773 1774
	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
	u16 opcode;
1775

I
Ilya Dryomov 已提交
1776 1777 1778
	osd_req_op_alloc_hint_init(obj_req->osd_req, which++,
				   rbd_dev->layout.object_size,
				   rbd_dev->layout.object_size);
1779

I
Ilya Dryomov 已提交
1780 1781 1782 1783
	if (rbd_obj_is_entire(obj_req))
		opcode = CEPH_OSD_OP_WRITEFULL;
	else
		opcode = CEPH_OSD_OP_WRITE;
1784

I
Ilya Dryomov 已提交
1785
	osd_req_op_extent_init(obj_req->osd_req, which, opcode,
1786
			       obj_req->ex.oe_off, obj_req->ex.oe_len, 0, 0);
I
Ilya Dryomov 已提交
1787
	rbd_osd_req_setup_data(obj_req, which++);
1788

I
Ilya Dryomov 已提交
1789 1790 1791
	rbd_assert(which == obj_req->osd_req->r_num_ops);
	rbd_osd_req_format_write(obj_req);
}
1792

I
Ilya Dryomov 已提交
1793 1794 1795 1796 1797
static int rbd_obj_setup_write(struct rbd_obj_request *obj_req)
{
	unsigned int num_osd_ops, which = 0;
	int ret;

I
Ilya Dryomov 已提交
1798 1799 1800 1801 1802 1803
	/* reverse map the entire object onto the parent */
	ret = rbd_obj_calc_img_extents(obj_req, true);
	if (ret)
		return ret;

	if (obj_req->num_img_extents) {
I
Ilya Dryomov 已提交
1804 1805 1806 1807 1808
		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 */
1809 1810
	}

1811
	obj_req->osd_req = rbd_osd_req_create(obj_req, num_osd_ops);
I
Ilya Dryomov 已提交
1812 1813
	if (!obj_req->osd_req)
		return -ENOMEM;
1814

I
Ilya Dryomov 已提交
1815
	if (obj_req->num_img_extents) {
I
Ilya Dryomov 已提交
1816 1817 1818 1819 1820 1821 1822
		ret = __rbd_obj_setup_stat(obj_req, which++);
		if (ret)
			return ret;
	}

	__rbd_obj_setup_write(obj_req, which);
	return 0;
1823 1824
}

I
Ilya Dryomov 已提交
1825 1826 1827
static void __rbd_obj_setup_discard(struct rbd_obj_request *obj_req,
				    unsigned int which)
{
1828 1829
	u16 opcode;

I
Ilya Dryomov 已提交
1830
	if (rbd_obj_is_entire(obj_req)) {
I
Ilya Dryomov 已提交
1831
		if (obj_req->num_img_extents) {
1832 1833
			osd_req_op_init(obj_req->osd_req, which++,
					CEPH_OSD_OP_CREATE, 0);
1834 1835
			opcode = CEPH_OSD_OP_TRUNCATE;
		} else {
I
Ilya Dryomov 已提交
1836 1837 1838
			osd_req_op_init(obj_req->osd_req, which++,
					CEPH_OSD_OP_DELETE, 0);
			opcode = 0;
1839
		}
I
Ilya Dryomov 已提交
1840 1841
	} else if (rbd_obj_is_tail(obj_req)) {
		opcode = CEPH_OSD_OP_TRUNCATE;
1842
	} else {
I
Ilya Dryomov 已提交
1843
		opcode = CEPH_OSD_OP_ZERO;
1844 1845
	}

I
Ilya Dryomov 已提交
1846 1847
	if (opcode)
		osd_req_op_extent_init(obj_req->osd_req, which++, opcode,
1848
				       obj_req->ex.oe_off, obj_req->ex.oe_len,
I
Ilya Dryomov 已提交
1849 1850 1851 1852
				       0, 0);

	rbd_assert(which == obj_req->osd_req->r_num_ops);
	rbd_osd_req_format_write(obj_req);
1853 1854
}

I
Ilya Dryomov 已提交
1855
static int rbd_obj_setup_discard(struct rbd_obj_request *obj_req)
A
Alex Elder 已提交
1856
{
I
Ilya Dryomov 已提交
1857 1858
	unsigned int num_osd_ops, which = 0;
	int ret;
A
Alex Elder 已提交
1859

I
Ilya Dryomov 已提交
1860 1861 1862 1863
	/* reverse map the entire object onto the parent */
	ret = rbd_obj_calc_img_extents(obj_req, true);
	if (ret)
		return ret;
1864

I
Ilya Dryomov 已提交
1865 1866
	if (rbd_obj_is_entire(obj_req)) {
		obj_req->write_state = RBD_OBJ_WRITE_FLAT;
1867 1868 1869 1870
		if (obj_req->num_img_extents)
			num_osd_ops = 2; /* create + truncate */
		else
			num_osd_ops = 1; /* delete */
I
Ilya Dryomov 已提交
1871
	} else {
I
Ilya Dryomov 已提交
1872
		if (obj_req->num_img_extents) {
I
Ilya Dryomov 已提交
1873 1874 1875 1876 1877 1878
			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 */
		}
1879 1880
	}

1881
	obj_req->osd_req = rbd_osd_req_create(obj_req, num_osd_ops);
I
Ilya Dryomov 已提交
1882 1883
	if (!obj_req->osd_req)
		return -ENOMEM;
A
Alex Elder 已提交
1884

I
Ilya Dryomov 已提交
1885
	if (!rbd_obj_is_entire(obj_req) && obj_req->num_img_extents) {
I
Ilya Dryomov 已提交
1886 1887 1888 1889
		ret = __rbd_obj_setup_stat(obj_req, which++);
		if (ret)
			return ret;
	}
1890

I
Ilya Dryomov 已提交
1891 1892 1893
	__rbd_obj_setup_discard(obj_req, which);
	return 0;
}
1894

I
Ilya Dryomov 已提交
1895 1896 1897 1898 1899 1900 1901 1902 1903
/*
 * 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;
A
Alex Elder 已提交
1904

I
Ilya Dryomov 已提交
1905
	for_each_obj_request(img_req, obj_req) {
I
Ilya Dryomov 已提交
1906
		switch (img_req->op_type) {
I
Ilya Dryomov 已提交
1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920
		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;
A
Alex Elder 已提交
1921 1922 1923
	}

	return 0;
I
Ilya Dryomov 已提交
1924
}
A
Alex Elder 已提交
1925

1926 1927 1928 1929
union rbd_img_fill_iter {
	struct ceph_bio_iter	bio_iter;
	struct ceph_bvec_iter	bvec_iter;
};
A
Alex Elder 已提交
1930

1931 1932 1933 1934 1935
struct rbd_img_fill_ctx {
	enum obj_request_type	pos_type;
	union rbd_img_fill_iter	*pos;
	union rbd_img_fill_iter	iter;
	ceph_object_extent_fn_t	set_pos_fn;
1936 1937
	ceph_object_extent_fn_t	count_fn;
	ceph_object_extent_fn_t	copy_fn;
1938
};
A
Alex Elder 已提交
1939

1940
static struct ceph_object_extent *alloc_object_extent(void *arg)
1941
{
1942 1943
	struct rbd_img_request *img_req = arg;
	struct rbd_obj_request *obj_req;
1944

1945 1946 1947
	obj_req = rbd_obj_request_create();
	if (!obj_req)
		return NULL;
I
Ilya Dryomov 已提交
1948

1949 1950 1951
	rbd_img_obj_request_add(img_req, obj_req);
	return &obj_req->ex;
}
1952

1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963
/*
 * While su != os && sc == 1 is technically not fancy (it's the same
 * layout as su == os && sc == 1), we can't use the nocopy path for it
 * because ->set_pos_fn() should be called only once per object.
 * ceph_file_to_extents() invokes action_fn once per stripe unit, so
 * treat su != os && sc == 1 as fancy.
 */
static bool rbd_layout_is_fancy(struct ceph_file_layout *l)
{
	return l->stripe_unit != l->object_size;
}
1964

1965 1966 1967 1968 1969 1970 1971 1972 1973
static int rbd_img_fill_request_nocopy(struct rbd_img_request *img_req,
				       struct ceph_file_extent *img_extents,
				       u32 num_img_extents,
				       struct rbd_img_fill_ctx *fctx)
{
	u32 i;
	int ret;

	img_req->data_type = fctx->pos_type;
1974 1975

	/*
1976 1977
	 * Create object requests and set each object request's starting
	 * position in the provided bio (list) or bio_vec array.
1978
	 */
1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989
	fctx->iter = *fctx->pos;
	for (i = 0; i < num_img_extents; i++) {
		ret = ceph_file_to_extents(&img_req->rbd_dev->layout,
					   img_extents[i].fe_off,
					   img_extents[i].fe_len,
					   &img_req->object_extents,
					   alloc_object_extent, img_req,
					   fctx->set_pos_fn, &fctx->iter);
		if (ret)
			return ret;
	}
1990

1991
	return __rbd_img_fill_request(img_req);
1992 1993
}

1994 1995 1996 1997
/*
 * Map a list of image extents to a list of object extents, create the
 * corresponding object requests (normally each to a different object,
 * but not always) and add them to @img_req.  For each object request,
1998
 * set up its data descriptor to point to the corresponding chunk(s) of
1999 2000
 * @fctx->pos data buffer.
 *
2001 2002 2003 2004
 * Because ceph_file_to_extents() will merge adjacent object extents
 * together, each object request's data descriptor may point to multiple
 * different chunks of @fctx->pos data buffer.
 *
2005 2006 2007 2008 2009 2010
 * @fctx->pos data buffer is assumed to be large enough.
 */
static int rbd_img_fill_request(struct rbd_img_request *img_req,
				struct ceph_file_extent *img_extents,
				u32 num_img_extents,
				struct rbd_img_fill_ctx *fctx)
2011
{
2012 2013
	struct rbd_device *rbd_dev = img_req->rbd_dev;
	struct rbd_obj_request *obj_req;
2014 2015 2016
	u32 i;
	int ret;

2017 2018 2019 2020
	if (fctx->pos_type == OBJ_REQUEST_NODATA ||
	    !rbd_layout_is_fancy(&rbd_dev->layout))
		return rbd_img_fill_request_nocopy(img_req, img_extents,
						   num_img_extents, fctx);
2021

2022
	img_req->data_type = OBJ_REQUEST_OWN_BVECS;
2023

2024
	/*
2025 2026 2027 2028 2029
	 * Create object requests and determine ->bvec_count for each object
	 * request.  Note that ->bvec_count sum over all object requests may
	 * be greater than the number of bio_vecs in the provided bio (list)
	 * or bio_vec array because when mapped, those bio_vecs can straddle
	 * stripe unit boundaries.
2030
	 */
2031 2032
	fctx->iter = *fctx->pos;
	for (i = 0; i < num_img_extents; i++) {
2033
		ret = ceph_file_to_extents(&rbd_dev->layout,
2034 2035 2036 2037
					   img_extents[i].fe_off,
					   img_extents[i].fe_len,
					   &img_req->object_extents,
					   alloc_object_extent, img_req,
2038 2039 2040
					   fctx->count_fn, &fctx->iter);
		if (ret)
			return ret;
2041
	}
2042

2043 2044 2045 2046 2047 2048 2049
	for_each_obj_request(img_req, obj_req) {
		obj_req->bvec_pos.bvecs = kmalloc_array(obj_req->bvec_count,
					      sizeof(*obj_req->bvec_pos.bvecs),
					      GFP_NOIO);
		if (!obj_req->bvec_pos.bvecs)
			return -ENOMEM;
	}
2050

2051
	/*
2052 2053
	 * Fill in each object request's private bio_vec array, splitting and
	 * rearranging the provided bio_vecs in stripe unit chunks as needed.
2054
	 */
2055 2056 2057 2058 2059 2060 2061
	fctx->iter = *fctx->pos;
	for (i = 0; i < num_img_extents; i++) {
		ret = ceph_iterate_extents(&rbd_dev->layout,
					   img_extents[i].fe_off,
					   img_extents[i].fe_len,
					   &img_req->object_extents,
					   fctx->copy_fn, &fctx->iter);
2062 2063 2064
		if (ret)
			return ret;
	}
2065

2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086
	return __rbd_img_fill_request(img_req);
}

static int rbd_img_fill_nodata(struct rbd_img_request *img_req,
			       u64 off, u64 len)
{
	struct ceph_file_extent ex = { off, len };
	union rbd_img_fill_iter dummy;
	struct rbd_img_fill_ctx fctx = {
		.pos_type = OBJ_REQUEST_NODATA,
		.pos = &dummy,
	};

	return rbd_img_fill_request(img_req, &ex, 1, &fctx);
}

static void set_bio_pos(struct ceph_object_extent *ex, u32 bytes, void *arg)
{
	struct rbd_obj_request *obj_req =
	    container_of(ex, struct rbd_obj_request, ex);
	struct ceph_bio_iter *it = arg;
2087

2088 2089 2090 2091
	dout("%s objno %llu bytes %u\n", __func__, ex->oe_objno, bytes);
	obj_req->bio_pos = *it;
	ceph_bio_iter_advance(it, bytes);
}
2092

2093 2094 2095 2096 2097
static void count_bio_bvecs(struct ceph_object_extent *ex, u32 bytes, void *arg)
{
	struct rbd_obj_request *obj_req =
	    container_of(ex, struct rbd_obj_request, ex);
	struct ceph_bio_iter *it = arg;
2098

2099 2100 2101 2102
	dout("%s objno %llu bytes %u\n", __func__, ex->oe_objno, bytes);
	ceph_bio_iter_advance_step(it, bytes, ({
		obj_req->bvec_count++;
	}));
2103

2104
}
2105

2106 2107 2108 2109 2110
static void copy_bio_bvecs(struct ceph_object_extent *ex, u32 bytes, void *arg)
{
	struct rbd_obj_request *obj_req =
	    container_of(ex, struct rbd_obj_request, ex);
	struct ceph_bio_iter *it = arg;
2111

2112 2113 2114 2115 2116
	dout("%s objno %llu bytes %u\n", __func__, ex->oe_objno, bytes);
	ceph_bio_iter_advance_step(it, bytes, ({
		obj_req->bvec_pos.bvecs[obj_req->bvec_idx++] = bv;
		obj_req->bvec_pos.iter.bi_size += bv.bv_len;
	}));
2117 2118
}

2119 2120 2121 2122 2123 2124 2125 2126 2127
static int __rbd_img_fill_from_bio(struct rbd_img_request *img_req,
				   struct ceph_file_extent *img_extents,
				   u32 num_img_extents,
				   struct ceph_bio_iter *bio_pos)
{
	struct rbd_img_fill_ctx fctx = {
		.pos_type = OBJ_REQUEST_BIO,
		.pos = (union rbd_img_fill_iter *)bio_pos,
		.set_pos_fn = set_bio_pos,
2128 2129
		.count_fn = count_bio_bvecs,
		.copy_fn = copy_bio_bvecs,
2130
	};
2131

2132 2133 2134
	return rbd_img_fill_request(img_req, img_extents, num_img_extents,
				    &fctx);
}
2135

2136 2137 2138 2139 2140
static int rbd_img_fill_from_bio(struct rbd_img_request *img_req,
				 u64 off, u64 len, struct bio *bio)
{
	struct ceph_file_extent ex = { off, len };
	struct ceph_bio_iter it = { .bio = bio, .iter = bio->bi_iter };
2141

2142 2143
	return __rbd_img_fill_from_bio(img_req, &ex, 1, &it);
}
A
Alex Elder 已提交
2144

2145 2146 2147 2148 2149
static void set_bvec_pos(struct ceph_object_extent *ex, u32 bytes, void *arg)
{
	struct rbd_obj_request *obj_req =
	    container_of(ex, struct rbd_obj_request, ex);
	struct ceph_bvec_iter *it = arg;
2150

2151 2152 2153 2154
	obj_req->bvec_pos = *it;
	ceph_bvec_iter_shorten(&obj_req->bvec_pos, bytes);
	ceph_bvec_iter_advance(it, bytes);
}
2155

2156 2157 2158 2159 2160
static void count_bvecs(struct ceph_object_extent *ex, u32 bytes, void *arg)
{
	struct rbd_obj_request *obj_req =
	    container_of(ex, struct rbd_obj_request, ex);
	struct ceph_bvec_iter *it = arg;
2161

2162 2163 2164 2165
	ceph_bvec_iter_advance_step(it, bytes, ({
		obj_req->bvec_count++;
	}));
}
2166

2167 2168 2169 2170 2171
static void copy_bvecs(struct ceph_object_extent *ex, u32 bytes, void *arg)
{
	struct rbd_obj_request *obj_req =
	    container_of(ex, struct rbd_obj_request, ex);
	struct ceph_bvec_iter *it = arg;
2172

2173 2174 2175 2176
	ceph_bvec_iter_advance_step(it, bytes, ({
		obj_req->bvec_pos.bvecs[obj_req->bvec_idx++] = bv;
		obj_req->bvec_pos.iter.bi_size += bv.bv_len;
	}));
2177 2178
}

2179 2180 2181 2182
static int __rbd_img_fill_from_bvecs(struct rbd_img_request *img_req,
				     struct ceph_file_extent *img_extents,
				     u32 num_img_extents,
				     struct ceph_bvec_iter *bvec_pos)
2183
{
2184 2185 2186 2187
	struct rbd_img_fill_ctx fctx = {
		.pos_type = OBJ_REQUEST_BVECS,
		.pos = (union rbd_img_fill_iter *)bvec_pos,
		.set_pos_fn = set_bvec_pos,
2188 2189
		.count_fn = count_bvecs,
		.copy_fn = copy_bvecs,
2190
	};
2191

2192 2193 2194
	return rbd_img_fill_request(img_req, img_extents, num_img_extents,
				    &fctx);
}
2195

2196 2197 2198 2199 2200 2201 2202 2203 2204 2205
static int rbd_img_fill_from_bvecs(struct rbd_img_request *img_req,
				   struct ceph_file_extent *img_extents,
				   u32 num_img_extents,
				   struct bio_vec *bvecs)
{
	struct ceph_bvec_iter it = {
		.bvecs = bvecs,
		.iter = { .bi_size = ceph_file_extents_bytes(img_extents,
							     num_img_extents) },
	};
2206

2207 2208 2209
	return __rbd_img_fill_from_bvecs(img_req, img_extents, num_img_extents,
					 &it);
}
2210

2211
static void rbd_img_request_submit(struct rbd_img_request *img_request)
A
Alex Elder 已提交
2212 2213
{
	struct rbd_obj_request *obj_request;
2214

A
Alex Elder 已提交
2215
	dout("%s: img %p\n", __func__, img_request);
2216

2217
	rbd_img_request_get(img_request);
2218
	for_each_obj_request(img_request, obj_request)
I
Ilya Dryomov 已提交
2219
		rbd_obj_request_submit(obj_request);
2220

2221
	rbd_img_request_put(img_request);
2222 2223
}

I
Ilya Dryomov 已提交
2224
static int rbd_obj_read_from_parent(struct rbd_obj_request *obj_req)
2225
{
I
Ilya Dryomov 已提交
2226 2227
	struct rbd_img_request *img_req = obj_req->img_request;
	struct rbd_img_request *child_img_req;
2228 2229
	int ret;

2230 2231
	child_img_req = rbd_img_request_create(img_req->rbd_dev->parent,
					       OBJ_OP_READ, NULL);
I
Ilya Dryomov 已提交
2232
	if (!child_img_req)
2233 2234
		return -ENOMEM;

2235 2236
	__set_bit(IMG_REQ_CHILD, &child_img_req->flags);
	child_img_req->obj_request = obj_req;
2237

I
Ilya Dryomov 已提交
2238
	if (!rbd_img_is_write(img_req)) {
2239
		switch (img_req->data_type) {
I
Ilya Dryomov 已提交
2240
		case OBJ_REQUEST_BIO:
2241 2242 2243 2244
			ret = __rbd_img_fill_from_bio(child_img_req,
						      obj_req->img_extents,
						      obj_req->num_img_extents,
						      &obj_req->bio_pos);
I
Ilya Dryomov 已提交
2245 2246
			break;
		case OBJ_REQUEST_BVECS:
2247
		case OBJ_REQUEST_OWN_BVECS:
2248 2249 2250 2251
			ret = __rbd_img_fill_from_bvecs(child_img_req,
						      obj_req->img_extents,
						      obj_req->num_img_extents,
						      &obj_req->bvec_pos);
I
Ilya Dryomov 已提交
2252 2253 2254 2255 2256
			break;
		default:
			rbd_assert(0);
		}
	} else {
2257 2258 2259 2260
		ret = rbd_img_fill_from_bvecs(child_img_req,
					      obj_req->img_extents,
					      obj_req->num_img_extents,
					      obj_req->copyup_bvecs);
I
Ilya Dryomov 已提交
2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276
	}
	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 &&
I
Ilya Dryomov 已提交
2277 2278 2279
	    rbd_dev->parent_overlap && !obj_req->tried_parent) {
		/* reverse map this object extent onto the parent */
		ret = rbd_obj_calc_img_extents(obj_req, false);
I
Ilya Dryomov 已提交
2280 2281 2282 2283
		if (ret) {
			obj_req->result = ret;
			return true;
		}
I
Ilya Dryomov 已提交
2284 2285 2286 2287 2288 2289 2290 2291 2292 2293

		if (obj_req->num_img_extents) {
			obj_req->tried_parent = true;
			ret = rbd_obj_read_from_parent(obj_req);
			if (ret) {
				obj_req->result = ret;
				return true;
			}
			return false;
		}
2294 2295
	}

2296
	/*
I
Ilya Dryomov 已提交
2297 2298 2299 2300
	 * -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.
2301
	 */
I
Ilya Dryomov 已提交
2302
	if (obj_req->result == -ENOENT ||
2303
	    (!obj_req->result && obj_req->xferred < obj_req->ex.oe_len)) {
I
Ilya Dryomov 已提交
2304 2305
		rbd_assert(!obj_req->xferred || !obj_req->result);
		rbd_obj_zero_range(obj_req, obj_req->xferred,
2306
				   obj_req->ex.oe_len - obj_req->xferred);
I
Ilya Dryomov 已提交
2307
		obj_req->result = 0;
2308
		obj_req->xferred = obj_req->ex.oe_len;
2309
	}
2310

I
Ilya Dryomov 已提交
2311 2312
	return true;
}
2313

I
Ilya Dryomov 已提交
2314 2315 2316 2317 2318 2319 2320 2321 2322
/*
 * 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 },
	};
2323

I
Ilya Dryomov 已提交
2324 2325 2326 2327 2328 2329
	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;
2330 2331
}

I
Ilya Dryomov 已提交
2332
static int rbd_obj_issue_copyup(struct rbd_obj_request *obj_req, u32 bytes)
2333
{
I
Ilya Dryomov 已提交
2334
	unsigned int num_osd_ops = obj_req->osd_req->r_num_ops;
2335

I
Ilya Dryomov 已提交
2336 2337 2338
	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);
2339

2340
	/*
I
Ilya Dryomov 已提交
2341 2342 2343
	 * 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).
2344
	 */
2345
	obj_req->osd_req = rbd_osd_req_create(obj_req, num_osd_ops);
I
Ilya Dryomov 已提交
2346 2347
	if (!obj_req->osd_req)
		return -ENOMEM;
2348

2349
	/*
I
Ilya Dryomov 已提交
2350 2351 2352
	 * Only send non-zero copyup data to save some I/O and network
	 * bandwidth -- zero copyup data is equivalent to the object not
	 * existing.
2353
	 */
I
Ilya Dryomov 已提交
2354 2355 2356 2357
	if (is_zero_bvecs(obj_req->copyup_bvecs, bytes)) {
		dout("%s obj_req %p detected zeroes\n", __func__, obj_req);
		bytes = 0;
	}
2358

I
Ilya Dryomov 已提交
2359 2360 2361 2362 2363
	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);

I
Ilya Dryomov 已提交
2364
	switch (obj_req->img_request->op_type) {
I
Ilya Dryomov 已提交
2365 2366 2367 2368 2369 2370 2371 2372 2373 2374
	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);
	}
2375

I
Ilya Dryomov 已提交
2376 2377
	rbd_obj_request_submit(obj_req);
	return 0;
2378 2379
}

2380
static int setup_copyup_bvecs(struct rbd_obj_request *obj_req, u64 obj_overlap)
2381
{
2382
	u32 i;
2383

2384 2385 2386 2387 2388 2389 2390
	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;
2391

2392 2393 2394 2395 2396 2397
	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;
2398

2399 2400 2401 2402
		obj_req->copyup_bvecs[i].bv_offset = 0;
		obj_req->copyup_bvecs[i].bv_len = len;
		obj_overlap -= len;
	}
2403

2404 2405
	rbd_assert(!obj_overlap);
	return 0;
2406 2407
}

I
Ilya Dryomov 已提交
2408
static int rbd_obj_handle_write_guard(struct rbd_obj_request *obj_req)
A
Alex Elder 已提交
2409
{
I
Ilya Dryomov 已提交
2410 2411
	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
	int ret;
A
Alex Elder 已提交
2412

I
Ilya Dryomov 已提交
2413 2414 2415 2416
	rbd_assert(obj_req->num_img_extents);
	prune_extents(obj_req->img_extents, &obj_req->num_img_extents,
		      rbd_dev->parent_overlap);
	if (!obj_req->num_img_extents) {
I
Ilya Dryomov 已提交
2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427
		/*
		 * 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);
A
Alex Elder 已提交
2428 2429
	}

I
Ilya Dryomov 已提交
2430
	ret = setup_copyup_bvecs(obj_req, rbd_obj_img_extents_bytes(obj_req));
I
Ilya Dryomov 已提交
2431 2432 2433 2434
	if (ret)
		return ret;

	obj_req->write_state = RBD_OBJ_WRITE_COPYUP;
I
Ilya Dryomov 已提交
2435
	return rbd_obj_read_from_parent(obj_req);
A
Alex Elder 已提交
2436
}
A
Alex Elder 已提交
2437

I
Ilya Dryomov 已提交
2438
static bool rbd_obj_handle_write(struct rbd_obj_request *obj_req)
A
Alex Elder 已提交
2439
{
I
Ilya Dryomov 已提交
2440
	int ret;
A
Alex Elder 已提交
2441

I
Ilya Dryomov 已提交
2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465
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.
			 */
2466
			obj_req->xferred = obj_req->ex.oe_len;
I
Ilya Dryomov 已提交
2467 2468 2469 2470 2471
		return true;
	case RBD_OBJ_WRITE_COPYUP:
		obj_req->write_state = RBD_OBJ_WRITE_GUARD;
		if (obj_req->result)
			goto again;
A
Alex Elder 已提交
2472

I
Ilya Dryomov 已提交
2473 2474 2475 2476 2477 2478 2479 2480
		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:
A
Arnd Bergmann 已提交
2481
		BUG();
I
Ilya Dryomov 已提交
2482 2483
	}
}
2484

I
Ilya Dryomov 已提交
2485 2486 2487 2488 2489
/*
 * Returns true if @obj_req is completed, or false otherwise.
 */
static bool __rbd_obj_handle_request(struct rbd_obj_request *obj_req)
{
I
Ilya Dryomov 已提交
2490
	switch (obj_req->img_request->op_type) {
I
Ilya Dryomov 已提交
2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502
	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;
2503
				obj_req->xferred = obj_req->ex.oe_len;
I
Ilya Dryomov 已提交
2504 2505 2506 2507 2508
			}
			return true;
		}
		return false;
	default:
A
Arnd Bergmann 已提交
2509
		BUG();
I
Ilya Dryomov 已提交
2510 2511
	}
}
2512

I
Ilya Dryomov 已提交
2513 2514 2515 2516 2517
static void rbd_obj_end_request(struct rbd_obj_request *obj_req)
{
	struct rbd_img_request *img_req = obj_req->img_request;

	rbd_assert((!obj_req->result &&
2518
		    obj_req->xferred == obj_req->ex.oe_len) ||
I
Ilya Dryomov 已提交
2519 2520 2521
		   (obj_req->result < 0 && !obj_req->xferred));
	if (!obj_req->result) {
		img_req->xferred += obj_req->xferred;
2522
		return;
2523
	}
A
Alex Elder 已提交
2524

I
Ilya Dryomov 已提交
2525 2526
	rbd_warn(img_req->rbd_dev,
		 "%s at objno %llu %llu~%llu result %d xferred %llu",
2527 2528
		 obj_op_name(img_req->op_type), obj_req->ex.oe_objno,
		 obj_req->ex.oe_off, obj_req->ex.oe_len, obj_req->result,
I
Ilya Dryomov 已提交
2529 2530 2531 2532 2533 2534
		 obj_req->xferred);
	if (!img_req->result) {
		img_req->result = obj_req->result;
		img_req->xferred = 0;
	}
}
A
Alex Elder 已提交
2535

I
Ilya Dryomov 已提交
2536 2537 2538
static void rbd_img_end_child_request(struct rbd_img_request *img_req)
{
	struct rbd_obj_request *obj_req = img_req->obj_request;
A
Alex Elder 已提交
2539

I
Ilya Dryomov 已提交
2540
	rbd_assert(test_bit(IMG_REQ_CHILD, &img_req->flags));
I
Ilya Dryomov 已提交
2541 2542 2543
	rbd_assert((!img_req->result &&
		    img_req->xferred == rbd_obj_img_extents_bytes(obj_req)) ||
		   (img_req->result < 0 && !img_req->xferred));
A
Alex Elder 已提交
2544

I
Ilya Dryomov 已提交
2545 2546 2547
	obj_req->result = img_req->result;
	obj_req->xferred = img_req->xferred;
	rbd_img_request_put(img_req);
A
Alex Elder 已提交
2548 2549
}

I
Ilya Dryomov 已提交
2550
static void rbd_img_end_request(struct rbd_img_request *img_req)
A
Alex Elder 已提交
2551
{
I
Ilya Dryomov 已提交
2552 2553 2554 2555
	rbd_assert(!test_bit(IMG_REQ_CHILD, &img_req->flags));
	rbd_assert((!img_req->result &&
		    img_req->xferred == blk_rq_bytes(img_req->rq)) ||
		   (img_req->result < 0 && !img_req->xferred));
A
Alex Elder 已提交
2556

I
Ilya Dryomov 已提交
2557 2558 2559
	blk_mq_end_request(img_req->rq,
			   errno_to_blk_status(img_req->result));
	rbd_img_request_put(img_req);
I
Ilya Dryomov 已提交
2560
}
A
Alex Elder 已提交
2561

I
Ilya Dryomov 已提交
2562 2563
static void rbd_obj_handle_request(struct rbd_obj_request *obj_req)
{
I
Ilya Dryomov 已提交
2564
	struct rbd_img_request *img_req;
A
Alex Elder 已提交
2565

I
Ilya Dryomov 已提交
2566
again:
I
Ilya Dryomov 已提交
2567 2568
	if (!__rbd_obj_handle_request(obj_req))
		return;
A
Alex Elder 已提交
2569

I
Ilya Dryomov 已提交
2570 2571 2572 2573 2574 2575 2576 2577
	img_req = obj_req->img_request;
	spin_lock(&img_req->completion_lock);
	rbd_obj_end_request(obj_req);
	rbd_assert(img_req->pending_count);
	if (--img_req->pending_count) {
		spin_unlock(&img_req->completion_lock);
		return;
	}
A
Alex Elder 已提交
2578

I
Ilya Dryomov 已提交
2579 2580 2581 2582 2583 2584 2585
	spin_unlock(&img_req->completion_lock);
	if (test_bit(IMG_REQ_CHILD, &img_req->flags)) {
		obj_req = img_req->obj_request;
		rbd_img_end_child_request(img_req);
		goto again;
	}
	rbd_img_end_request(img_req);
A
Alex Elder 已提交
2586
}
A
Alex Elder 已提交
2587

2588
static const struct rbd_client_id rbd_empty_cid;
A
Alex Elder 已提交
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
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);
}

2626 2627 2628 2629 2630 2631 2632 2633 2634
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);
}

2635 2636 2637 2638
/*
 * lock_rwsem must be held for write
 */
static int rbd_lock(struct rbd_device *rbd_dev)
A
Alex Elder 已提交
2639
{
2640
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2641
	char cookie[32];
2642
	int ret;
A
Alex Elder 已提交
2643

I
Ilya Dryomov 已提交
2644 2645
	WARN_ON(__rbd_is_lock_owner(rbd_dev) ||
		rbd_dev->lock_cookie[0] != '\0');
2646

2647 2648 2649 2650
	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);
2651
	if (ret)
2652
		return ret;
A
Alex Elder 已提交
2653

2654
	rbd_dev->lock_state = RBD_LOCK_STATE_LOCKED;
2655
	__rbd_lock(rbd_dev, cookie);
2656
	return 0;
A
Alex Elder 已提交
2657 2658
}

2659 2660 2661
/*
 * lock_rwsem must be held for write
 */
I
Ilya Dryomov 已提交
2662
static void rbd_unlock(struct rbd_device *rbd_dev)
2663
{
2664
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2665 2666
	int ret;

I
Ilya Dryomov 已提交
2667 2668
	WARN_ON(!__rbd_is_lock_owner(rbd_dev) ||
		rbd_dev->lock_cookie[0] == '\0');
2669

2670
	ret = ceph_cls_unlock(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc,
I
Ilya Dryomov 已提交
2671
			      RBD_LOCK_NAME, rbd_dev->lock_cookie);
I
Ilya Dryomov 已提交
2672 2673
	if (ret && ret != -ENOENT)
		rbd_warn(rbd_dev, "failed to unlock: %d", ret);
2674

I
Ilya Dryomov 已提交
2675 2676
	/* treat errors as the image is unlocked */
	rbd_dev->lock_state = RBD_LOCK_STATE_UNLOCKED;
I
Ilya Dryomov 已提交
2677
	rbd_dev->lock_cookie[0] = '\0';
2678 2679
	rbd_set_owner_cid(rbd_dev, &rbd_empty_cid);
	queue_work(rbd_dev->task_wq, &rbd_dev->released_lock_work);
2680 2681
}

2682 2683 2684 2685
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)
2686 2687
{
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2688
	struct rbd_client_id cid = rbd_get_cid(rbd_dev);
K
Kyle Spiers 已提交
2689 2690
	char buf[4 + 8 + 8 + CEPH_ENCODING_START_BLK_LEN];
	int buf_size = sizeof(buf);
2691
	void *p = buf;
2692

2693
	dout("%s rbd_dev %p notify_op %d\n", __func__, rbd_dev, notify_op);
2694

2695 2696 2697 2698 2699
	/* 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);
2700

2701 2702 2703
	return ceph_osdc_notify(osdc, &rbd_dev->header_oid,
				&rbd_dev->header_oloc, buf, buf_size,
				RBD_NOTIFY_TIMEOUT, preply_pages, preply_len);
2704 2705
}

2706 2707
static void rbd_notify_op_lock(struct rbd_device *rbd_dev,
			       enum rbd_notify_op notify_op)
2708
{
2709 2710
	struct page **reply_pages;
	size_t reply_len;
2711

2712 2713 2714
	__rbd_notify_op_lock(rbd_dev, notify_op, &reply_pages, &reply_len);
	ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len));
}
2715

2716 2717 2718 2719
static void rbd_notify_acquired_lock(struct work_struct *work)
{
	struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
						  acquired_lock_work);
2720

2721
	rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_ACQUIRED_LOCK);
2722 2723
}

2724
static void rbd_notify_released_lock(struct work_struct *work)
2725
{
2726 2727
	struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
						  released_lock_work);
I
Ilya Dryomov 已提交
2728

2729
	rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_RELEASED_LOCK);
2730 2731
}

2732
static int rbd_request_lock(struct rbd_device *rbd_dev)
2733
{
2734 2735 2736
	struct page **reply_pages;
	size_t reply_len;
	bool lock_owner_responded = false;
2737 2738
	int ret;

2739
	dout("%s rbd_dev %p\n", __func__, rbd_dev);
2740

2741 2742 2743 2744
	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);
2745
		goto out;
2746
	}
2747

2748 2749 2750 2751
	if (reply_len > 0 && reply_len <= PAGE_SIZE) {
		void *p = page_address(reply_pages[0]);
		void *const end = p + reply_len;
		u32 n;
2752

2753 2754 2755 2756
		ceph_decode_32_safe(&p, end, n, e_inval); /* num_acks */
		while (n--) {
			u8 struct_v;
			u32 len;
2757

2758 2759
			ceph_decode_need(&p, end, 8 + 8, e_inval);
			p += 8 + 8; /* skip gid and cookie */
2760

2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 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
			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;
2993
	int ret = 0;
2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008

	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
3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020
	} 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);
		}
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
	} 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);
3049
	/*
3050
	 * Ensure that all in-flight IO is flushed.
3051
	 *
3052 3053
	 * FIXME: ceph_osdc_sync() flushes the entire OSD client, which
	 * may be shared with other devices.
3054
	 */
3055 3056 3057 3058 3059 3060 3061 3062 3063
	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
3064 3065 3066 3067 3068 3069 3070 3071 3072
	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);
3073 3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 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 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152
	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);
}

3153 3154 3155 3156 3157 3158
/*
 * 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)
3159 3160 3161
{
	struct rbd_client_id my_cid = rbd_get_cid(rbd_dev);
	struct rbd_client_id cid = { 0 };
3162
	int result = 1;
3163 3164 3165 3166 3167 3168 3169 3170 3171

	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))
3172
		return result;
3173 3174

	down_read(&rbd_dev->lock_rwsem);
3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186
	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
3187 3188 3189 3190 3191 3192 3193 3194 3195
			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;
			}
3196 3197
		}
	}
3198 3199

out_unlock:
3200
	up_read(&rbd_dev->lock_rwsem);
3201
	return result;
3202 3203 3204 3205 3206 3207
}

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;
K
Kyle Spiers 已提交
3208 3209
	char buf[4 + CEPH_ENCODING_START_BLK_LEN];
	int buf_size = sizeof(buf);
3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220 3221
	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
3222

3223 3224
	ret = ceph_osdc_notify_ack(osdc, &rbd_dev->header_oid,
				   &rbd_dev->header_oloc, notify_id, cookie,
3225
				   buf, buf_size);
3226
	if (ret)
3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249
		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;
3250
	u8 struct_v = 0;
3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283
	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:
3284 3285
		ret = rbd_handle_request_lock(rbd_dev, struct_v, &p);
		if (ret <= 0)
3286
			rbd_acknowledge_notify_result(rbd_dev, notify_id,
3287
						      cookie, ret);
3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305
		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
3306 3307
}

3308 3309
static void __rbd_unregister_watch(struct rbd_device *rbd_dev);

3310
static void rbd_watch_errcb(void *arg, u64 cookie, int err)
3311
{
3312
	struct rbd_device *rbd_dev = arg;
3313

3314
	rbd_warn(rbd_dev, "encountered watch error: %d", err);
3315

3316 3317 3318 3319
	down_write(&rbd_dev->lock_rwsem);
	rbd_set_owner_cid(rbd_dev, &rbd_empty_cid);
	up_write(&rbd_dev->lock_rwsem);

3320 3321 3322 3323
	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;
3324

3325
		queue_delayed_work(rbd_dev->task_wq, &rbd_dev->watch_dwork, 0);
3326
	}
3327
	mutex_unlock(&rbd_dev->watch_mutex);
3328 3329
}

3330
/*
3331
 * watch_mutex must be locked
3332
 */
3333
static int __rbd_register_watch(struct rbd_device *rbd_dev)
3334 3335
{
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3336
	struct ceph_osd_linger_request *handle;
3337

3338
	rbd_assert(!rbd_dev->watch_handle);
3339
	dout("%s rbd_dev %p\n", __func__, rbd_dev);
3340

3341 3342 3343 3344 3345
	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);
3346

3347
	rbd_dev->watch_handle = handle;
3348 3349 3350
	return 0;
}

3351 3352 3353 3354
/*
 * watch_mutex must be locked
 */
static void __rbd_unregister_watch(struct rbd_device *rbd_dev)
3355
{
3356 3357
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
	int ret;
3358

3359 3360
	rbd_assert(rbd_dev->watch_handle);
	dout("%s rbd_dev %p\n", __func__, rbd_dev);
3361

3362 3363 3364
	ret = ceph_osdc_unwatch(osdc, rbd_dev->watch_handle);
	if (ret)
		rbd_warn(rbd_dev, "failed to unwatch: %d", ret);
3365

3366
	rbd_dev->watch_handle = NULL;
3367 3368
}

3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387
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)
3388
{
3389 3390 3391
	dout("%s rbd_dev %p\n", __func__, rbd_dev);

	cancel_delayed_work_sync(&rbd_dev->watch_dwork);
3392 3393 3394 3395
	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);
3396 3397 3398 3399
}

static void rbd_unregister_watch(struct rbd_device *rbd_dev)
{
3400
	WARN_ON(waitqueue_active(&rbd_dev->lock_waitq));
3401 3402 3403 3404 3405 3406 3407
	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
3408 3409

	ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc);
3410 3411
}

3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440
/*
 * 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 {
3441
		__rbd_lock(rbd_dev, cookie);
3442 3443 3444
	}
}

3445 3446 3447 3448 3449 3450 3451 3452 3453
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);
3454 3455
	if (rbd_dev->watch_state != RBD_WATCH_STATE_ERROR) {
		mutex_unlock(&rbd_dev->watch_mutex);
3456
		return;
3457
	}
3458 3459 3460 3461

	ret = __rbd_register_watch(rbd_dev);
	if (ret) {
		rbd_warn(rbd_dev, "failed to reregister watch: %d", ret);
3462
		if (ret == -EBLACKLISTED || ret == -ENOENT) {
3463
			set_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags);
3464
			wake_requests(rbd_dev, true);
3465
		} else {
3466 3467 3468
			queue_delayed_work(rbd_dev->task_wq,
					   &rbd_dev->watch_dwork,
					   RBD_RETRY_DELAY);
3469 3470
		}
		mutex_unlock(&rbd_dev->watch_mutex);
3471
		return;
3472 3473 3474 3475 3476 3477
	}

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

3478 3479 3480 3481 3482
	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);

3483 3484
	ret = rbd_dev_refresh(rbd_dev);
	if (ret)
3485
		rbd_warn(rbd_dev, "reregistration refresh failed: %d", ret);
3486 3487
}

3488
/*
3489 3490
 * Synchronous osd object method call.  Returns the number of bytes
 * returned in the outbound buffer, or a negative error code.
3491 3492
 */
static int rbd_obj_method_sync(struct rbd_device *rbd_dev,
3493 3494
			     struct ceph_object_id *oid,
			     struct ceph_object_locator *oloc,
3495
			     const char *method_name,
3496
			     const void *outbound,
3497
			     size_t outbound_size,
3498
			     void *inbound,
3499
			     size_t inbound_size)
3500
{
3501 3502 3503
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
	struct page *req_page = NULL;
	struct page *reply_page;
3504 3505 3506
	int ret;

	/*
3507 3508 3509 3510 3511
	 * 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.
3512
	 */
3513 3514 3515
	if (outbound) {
		if (outbound_size > PAGE_SIZE)
			return -E2BIG;
3516

3517 3518 3519
		req_page = alloc_page(GFP_KERNEL);
		if (!req_page)
			return -ENOMEM;
3520

3521
		memcpy(page_address(req_page), outbound, outbound_size);
3522
	}
3523

3524 3525 3526 3527 3528 3529
	reply_page = alloc_page(GFP_KERNEL);
	if (!reply_page) {
		if (req_page)
			__free_page(req_page);
		return -ENOMEM;
	}
3530

3531 3532 3533 3534 3535 3536 3537
	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;
	}
3538

3539 3540 3541
	if (req_page)
		__free_page(req_page);
	__free_page(reply_page);
3542 3543 3544
	return ret;
}

3545 3546 3547
/*
 * lock_rwsem must be held for read
 */
3548
static int rbd_wait_state_locked(struct rbd_device *rbd_dev, bool may_acquire)
3549 3550
{
	DEFINE_WAIT(wait);
3551
	unsigned long timeout;
3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563
	int ret = 0;

	if (test_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags))
		return -EBLACKLISTED;

	if (rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED)
		return 0;

	if (!may_acquire) {
		rbd_warn(rbd_dev, "exclusive lock required");
		return -EROFS;
	}
3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574

	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);
3575 3576
		timeout = schedule_timeout(ceph_timeout_jiffies(
						rbd_dev->opts->lock_timeout));
3577
		down_read(&rbd_dev->lock_rwsem);
3578 3579 3580 3581
		if (test_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags)) {
			ret = -EBLACKLISTED;
			break;
		}
3582 3583 3584 3585 3586
		if (!timeout) {
			rbd_warn(rbd_dev, "timed out waiting for lock");
			ret = -ETIMEDOUT;
			break;
		}
3587
	} while (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED);
3588

3589
	finish_wait(&rbd_dev->lock_waitq, &wait);
3590
	return ret;
3591 3592
}

C
Christoph Hellwig 已提交
3593
static void rbd_queue_workfn(struct work_struct *work)
A
Alex Elder 已提交
3594
{
C
Christoph Hellwig 已提交
3595 3596
	struct request *rq = blk_mq_rq_from_pdu(work);
	struct rbd_device *rbd_dev = rq->q->queuedata;
I
Ilya Dryomov 已提交
3597
	struct rbd_img_request *img_request;
3598
	struct ceph_snap_context *snapc = NULL;
I
Ilya Dryomov 已提交
3599 3600
	u64 offset = (u64)blk_rq_pos(rq) << SECTOR_SHIFT;
	u64 length = blk_rq_bytes(rq);
G
Guangliang Zhao 已提交
3601
	enum obj_operation_type op_type;
3602
	u64 mapping_size;
I
Ilya Dryomov 已提交
3603
	bool must_be_locked;
A
Alex Elder 已提交
3604 3605
	int result;

3606 3607
	switch (req_op(rq)) {
	case REQ_OP_DISCARD:
3608
	case REQ_OP_WRITE_ZEROES:
3609
		op_type = OBJ_OP_DISCARD;
3610 3611
		break;
	case REQ_OP_WRITE:
G
Guangliang Zhao 已提交
3612
		op_type = OBJ_OP_WRITE;
3613 3614
		break;
	case REQ_OP_READ:
G
Guangliang Zhao 已提交
3615
		op_type = OBJ_OP_READ;
3616 3617 3618 3619 3620 3621
		break;
	default:
		dout("%s: non-fs request type %d\n", __func__, req_op(rq));
		result = -EIO;
		goto err;
	}
G
Guangliang Zhao 已提交
3622

I
Ilya Dryomov 已提交
3623
	/* Ignore/skip any zero-length requests */
A
Alex Elder 已提交
3624

I
Ilya Dryomov 已提交
3625 3626 3627 3628 3629
	if (!length) {
		dout("%s: zero-length request\n", __func__);
		result = 0;
		goto err_rq;
	}
A
Alex Elder 已提交
3630

3631 3632
	rbd_assert(op_type == OBJ_OP_READ ||
		   rbd_dev->spec->snap_id == CEPH_NOSNAP);
A
Alex Elder 已提交
3633

I
Ilya Dryomov 已提交
3634 3635 3636 3637 3638 3639 3640 3641 3642 3643 3644 3645
	/*
	 * 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
3646

I
Ilya Dryomov 已提交
3647 3648 3649 3650 3651 3652
	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
3653

C
Christoph Hellwig 已提交
3654 3655
	blk_mq_start_request(rq);

3656 3657
	down_read(&rbd_dev->header_rwsem);
	mapping_size = rbd_dev->mapping.size;
G
Guangliang Zhao 已提交
3658
	if (op_type != OBJ_OP_READ) {
3659 3660 3661 3662 3663 3664
		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
3665
		rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)", offset,
3666
			 length, mapping_size);
I
Ilya Dryomov 已提交
3667 3668 3669
		result = -EIO;
		goto err_rq;
	}
A
Alex Elder 已提交
3670

I
Ilya Dryomov 已提交
3671 3672 3673
	must_be_locked =
	    (rbd_dev->header.features & RBD_FEATURE_EXCLUSIVE_LOCK) &&
	    (op_type != OBJ_OP_READ || rbd_dev->opts->lock_on_read);
3674 3675
	if (must_be_locked) {
		down_read(&rbd_dev->lock_rwsem);
3676 3677 3678
		result = rbd_wait_state_locked(rbd_dev,
					       !rbd_dev->opts->exclusive);
		if (result)
3679
			goto err_unlock;
3680 3681
	}

3682
	img_request = rbd_img_request_create(rbd_dev, op_type, snapc);
I
Ilya Dryomov 已提交
3683 3684
	if (!img_request) {
		result = -ENOMEM;
3685
		goto err_unlock;
I
Ilya Dryomov 已提交
3686 3687
	}
	img_request->rq = rq;
3688
	snapc = NULL; /* img_request consumes a ref */
A
Alex Elder 已提交
3689

3690
	if (op_type == OBJ_OP_DISCARD)
3691
		result = rbd_img_fill_nodata(img_request, offset, length);
3692
	else
3693 3694
		result = rbd_img_fill_from_bio(img_request, offset, length,
					       rq->bio);
I
Ilya Dryomov 已提交
3695 3696
	if (result)
		goto err_img_request;
A
Alex Elder 已提交
3697

3698
	rbd_img_request_submit(img_request);
3699 3700
	if (must_be_locked)
		up_read(&rbd_dev->lock_rwsem);
I
Ilya Dryomov 已提交
3701
	return;
A
Alex Elder 已提交
3702

I
Ilya Dryomov 已提交
3703 3704
err_img_request:
	rbd_img_request_put(img_request);
3705 3706 3707
err_unlock:
	if (must_be_locked)
		up_read(&rbd_dev->lock_rwsem);
I
Ilya Dryomov 已提交
3708 3709 3710
err_rq:
	if (result)
		rbd_warn(rbd_dev, "%s %llx at %llx result %d",
G
Guangliang Zhao 已提交
3711
			 obj_op_name(op_type), length, offset, result);
3712
	ceph_put_snap_context(snapc);
C
Christoph Hellwig 已提交
3713
err:
3714
	blk_mq_end_request(rq, errno_to_blk_status(result));
I
Ilya Dryomov 已提交
3715
}
A
Alex Elder 已提交
3716

3717
static blk_status_t rbd_queue_rq(struct blk_mq_hw_ctx *hctx,
C
Christoph Hellwig 已提交
3718
		const struct blk_mq_queue_data *bd)
I
Ilya Dryomov 已提交
3719
{
C
Christoph Hellwig 已提交
3720 3721
	struct request *rq = bd->rq;
	struct work_struct *work = blk_mq_rq_to_pdu(rq);
A
Alex Elder 已提交
3722

C
Christoph Hellwig 已提交
3723
	queue_work(rbd_wq, work);
3724
	return BLK_STS_OK;
A
Alex Elder 已提交
3725 3726
}

3727 3728
static void rbd_free_disk(struct rbd_device *rbd_dev)
{
3729 3730 3731
	blk_cleanup_queue(rbd_dev->disk->queue);
	blk_mq_free_tag_set(&rbd_dev->tag_set);
	put_disk(rbd_dev->disk);
3732
	rbd_dev->disk = NULL;
3733 3734
}

3735
static int rbd_obj_read_sync(struct rbd_device *rbd_dev,
3736 3737 3738
			     struct ceph_object_id *oid,
			     struct ceph_object_locator *oloc,
			     void *buf, int buf_len)
3739 3740

{
3741 3742 3743 3744
	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);
3745 3746
	int ret;

3747 3748 3749
	req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_KERNEL);
	if (!req)
		return -ENOMEM;
3750

3751 3752 3753
	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
3754

3755
	ret = ceph_osdc_alloc_messages(req, GFP_KERNEL);
3756
	if (ret)
3757
		goto out_req;
3758

3759 3760 3761 3762 3763
	pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
	if (IS_ERR(pages)) {
		ret = PTR_ERR(pages);
		goto out_req;
	}
3764

3765 3766 3767 3768 3769 3770 3771 3772
	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);
3773

3774 3775
out_req:
	ceph_osdc_put_request(req);
3776 3777 3778
	return ret;
}

3779
/*
A
Alex Elder 已提交
3780 3781 3782
 * 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.
3783
 */
3784
static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev)
3785
{
3786
	struct rbd_image_header_ondisk *ondisk = NULL;
3787
	u32 snap_count = 0;
3788 3789 3790
	u64 names_size = 0;
	u32 want_count;
	int ret;
3791

A
Alex Elder 已提交
3792
	/*
3793 3794 3795 3796 3797
	 * 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
3798
	 */
3799 3800 3801 3802 3803 3804 3805 3806 3807 3808
	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
3809
			return -ENOMEM;
3810

3811 3812
		ret = rbd_obj_read_sync(rbd_dev, &rbd_dev->header_oid,
					&rbd_dev->header_oloc, ondisk, size);
3813
		if (ret < 0)
A
Alex Elder 已提交
3814
			goto out;
A
Alex Elder 已提交
3815
		if ((size_t)ret < size) {
3816
			ret = -ENXIO;
A
Alex Elder 已提交
3817 3818
			rbd_warn(rbd_dev, "short header read (want %zd got %d)",
				size, ret);
A
Alex Elder 已提交
3819
			goto out;
3820 3821 3822
		}
		if (!rbd_dev_ondisk_valid(ondisk)) {
			ret = -ENXIO;
A
Alex Elder 已提交
3823
			rbd_warn(rbd_dev, "invalid header");
A
Alex Elder 已提交
3824
			goto out;
3825
		}
3826

3827 3828 3829 3830
		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
3831

A
Alex Elder 已提交
3832 3833
	ret = rbd_header_from_disk(rbd_dev, ondisk);
out:
3834 3835 3836
	kfree(ondisk);

	return ret;
3837 3838
}

3839 3840 3841 3842 3843 3844 3845 3846 3847 3848 3849 3850 3851 3852 3853 3854 3855 3856 3857
/*
 * 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);
}

3858 3859 3860 3861 3862
static void rbd_dev_update_size(struct rbd_device *rbd_dev)
{
	sector_t size;

	/*
I
Ilya Dryomov 已提交
3863 3864 3865
	 * 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.
3866
	 */
I
Ilya Dryomov 已提交
3867 3868
	if (test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags) &&
	    !test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) {
3869 3870 3871 3872 3873 3874 3875
		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
3876
static int rbd_dev_refresh(struct rbd_device *rbd_dev)
A
Alex Elder 已提交
3877
{
3878
	u64 mapping_size;
A
Alex Elder 已提交
3879 3880
	int ret;

3881
	down_write(&rbd_dev->header_rwsem);
3882
	mapping_size = rbd_dev->mapping.size;
3883 3884

	ret = rbd_dev_header_info(rbd_dev);
3885
	if (ret)
3886
		goto out;
3887

3888 3889 3890 3891 3892 3893 3894
	/*
	 * 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)
3895
			goto out;
3896 3897
	}

3898
	if (rbd_dev->spec->snap_id == CEPH_NOSNAP) {
3899
		rbd_dev->mapping.size = rbd_dev->header.image_size;
3900 3901 3902 3903
	} else {
		/* validate mapped snapshot's EXISTS flag */
		rbd_exists_validate(rbd_dev);
	}
3904

3905
out:
3906
	up_write(&rbd_dev->header_rwsem);
3907
	if (!ret && mapping_size != rbd_dev->mapping.size)
3908
		rbd_dev_update_size(rbd_dev);
A
Alex Elder 已提交
3909

3910
	return ret;
A
Alex Elder 已提交
3911 3912
}

3913 3914
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
3915 3916 3917 3918 3919 3920 3921
{
	struct work_struct *work = blk_mq_rq_to_pdu(rq);

	INIT_WORK(work, rbd_queue_workfn);
	return 0;
}

3922
static const struct blk_mq_ops rbd_mq_ops = {
C
Christoph Hellwig 已提交
3923 3924 3925 3926
	.queue_rq	= rbd_queue_rq,
	.init_request	= rbd_init_request,
};

3927 3928 3929 3930
static int rbd_init_disk(struct rbd_device *rbd_dev)
{
	struct gendisk *disk;
	struct request_queue *q;
3931 3932
	unsigned int objset_bytes =
	    rbd_dev->layout.object_size * rbd_dev->layout.stripe_count;
C
Christoph Hellwig 已提交
3933
	int err;
3934 3935

	/* create gendisk info */
3936 3937 3938
	disk = alloc_disk(single_major ?
			  (1 << RBD_SINGLE_MAJOR_PART_SHIFT) :
			  RBD_MINORS_PER_MAJOR);
3939
	if (!disk)
A
Alex Elder 已提交
3940
		return -ENOMEM;
3941

A
Alex Elder 已提交
3942
	snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
A
Alex Elder 已提交
3943
		 rbd_dev->dev_id);
3944
	disk->major = rbd_dev->major;
3945
	disk->first_minor = rbd_dev->minor;
3946 3947
	if (single_major)
		disk->flags |= GENHD_FL_EXT_DEVT;
3948 3949 3950
	disk->fops = &rbd_bd_ops;
	disk->private_data = rbd_dev;

C
Christoph Hellwig 已提交
3951 3952
	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
3953
	rbd_dev->tag_set.queue_depth = rbd_dev->opts->queue_depth;
C
Christoph Hellwig 已提交
3954
	rbd_dev->tag_set.numa_node = NUMA_NO_NODE;
I
Ilya Dryomov 已提交
3955
	rbd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
C
Christoph Hellwig 已提交
3956 3957 3958 3959 3960
	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)
3961
		goto out_disk;
3962

C
Christoph Hellwig 已提交
3963 3964 3965 3966 3967 3968
	q = blk_mq_init_queue(&rbd_dev->tag_set);
	if (IS_ERR(q)) {
		err = PTR_ERR(q);
		goto out_tag_set;
	}

3969
	blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
3970
	/* QUEUE_FLAG_ADD_RANDOM is off by default for blk-mq */
A
Alex Elder 已提交
3971

3972
	blk_queue_max_hw_sectors(q, objset_bytes >> SECTOR_SHIFT);
I
Ilya Dryomov 已提交
3973
	q->limits.max_sectors = queue_max_hw_sectors(q);
3974
	blk_queue_max_segments(q, USHRT_MAX);
3975
	blk_queue_max_segment_size(q, UINT_MAX);
3976 3977
	blk_queue_io_min(q, objset_bytes);
	blk_queue_io_opt(q, objset_bytes);
3978

3979
	/* enable the discard support */
3980
	blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
3981 3982 3983
	q->limits.discard_granularity = objset_bytes;
	blk_queue_max_discard_sectors(q, objset_bytes >> SECTOR_SHIFT);
	blk_queue_max_write_zeroes_sectors(q, objset_bytes >> SECTOR_SHIFT);
3984

3985
	if (!ceph_test_opt(rbd_dev->rbd_client->client, NOCRC))
3986
		q->backing_dev_info->capabilities |= BDI_CAP_STABLE_WRITES;
3987

3988 3989 3990 3991 3992
	/*
	 * 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));
3993 3994 3995 3996 3997 3998
	disk->queue = q;
	q->queuedata = rbd_dev;

	rbd_dev->disk = disk;

	return 0;
C
Christoph Hellwig 已提交
3999 4000
out_tag_set:
	blk_mq_free_tag_set(&rbd_dev->tag_set);
4001 4002
out_disk:
	put_disk(disk);
C
Christoph Hellwig 已提交
4003
	return err;
4004 4005
}

4006 4007 4008 4009
/*
  sysfs
*/

A
Alex Elder 已提交
4010 4011 4012 4013 4014
static struct rbd_device *dev_to_rbd_dev(struct device *dev)
{
	return container_of(dev, struct rbd_device, dev);
}

4015 4016 4017
static ssize_t rbd_size_show(struct device *dev,
			     struct device_attribute *attr, char *buf)
{
A
Alex Elder 已提交
4018
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4019

A
Alex Elder 已提交
4020 4021
	return sprintf(buf, "%llu\n",
		(unsigned long long)rbd_dev->mapping.size);
4022 4023
}

A
Alex Elder 已提交
4024 4025 4026 4027 4028 4029 4030 4031 4032 4033
/*
 * 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
4034
			(unsigned long long)rbd_dev->mapping.features);
A
Alex Elder 已提交
4035 4036
}

4037 4038 4039
static ssize_t rbd_major_show(struct device *dev,
			      struct device_attribute *attr, char *buf)
{
A
Alex Elder 已提交
4040
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4041

A
Alex Elder 已提交
4042 4043 4044 4045
	if (rbd_dev->major)
		return sprintf(buf, "%d\n", rbd_dev->major);

	return sprintf(buf, "(none)\n");
4046 4047 4048 4049 4050 4051
}

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
4052

4053
	return sprintf(buf, "%d\n", rbd_dev->minor);
4054 4055
}

4056 4057 4058 4059 4060 4061 4062 4063 4064 4065 4066
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));
}

4067 4068
static ssize_t rbd_client_id_show(struct device *dev,
				  struct device_attribute *attr, char *buf)
4069
{
A
Alex Elder 已提交
4070
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4071

4072
	return sprintf(buf, "client%lld\n",
4073
		       ceph_client_gid(rbd_dev->rbd_client->client));
4074 4075
}

4076 4077 4078 4079 4080 4081 4082 4083
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);
}

4084 4085 4086 4087 4088 4089
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);
4090 4091
}

4092 4093
static ssize_t rbd_pool_show(struct device *dev,
			     struct device_attribute *attr, char *buf)
4094
{
A
Alex Elder 已提交
4095
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4096

4097
	return sprintf(buf, "%s\n", rbd_dev->spec->pool_name);
4098 4099
}

4100 4101 4102 4103 4104
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);

4105
	return sprintf(buf, "%llu\n",
A
Alex Elder 已提交
4106
			(unsigned long long) rbd_dev->spec->pool_id);
4107 4108
}

4109 4110 4111
static ssize_t rbd_name_show(struct device *dev,
			     struct device_attribute *attr, char *buf)
{
A
Alex Elder 已提交
4112
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4113

A
Alex Elder 已提交
4114 4115 4116 4117
	if (rbd_dev->spec->image_name)
		return sprintf(buf, "%s\n", rbd_dev->spec->image_name);

	return sprintf(buf, "(unknown)\n");
4118 4119
}

A
Alex Elder 已提交
4120 4121 4122 4123 4124
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);

4125
	return sprintf(buf, "%s\n", rbd_dev->spec->image_id);
A
Alex Elder 已提交
4126 4127
}

A
Alex Elder 已提交
4128 4129 4130 4131
/*
 * Shows the name of the currently-mapped snapshot (or
 * RBD_SNAP_HEAD_NAME for the base image).
 */
4132 4133 4134 4135
static ssize_t rbd_snap_show(struct device *dev,
			     struct device_attribute *attr,
			     char *buf)
{
A
Alex Elder 已提交
4136
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4137

4138
	return sprintf(buf, "%s\n", rbd_dev->spec->snap_name);
4139 4140
}

4141 4142 4143 4144 4145 4146 4147 4148
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);
}

4149
/*
4150 4151 4152
 * 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)".
4153 4154
 */
static ssize_t rbd_parent_show(struct device *dev,
4155 4156
			       struct device_attribute *attr,
			       char *buf)
4157 4158
{
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4159
	ssize_t count = 0;
4160

4161
	if (!rbd_dev->parent)
4162 4163
		return sprintf(buf, "(no parent image)\n");

4164 4165 4166 4167 4168 4169 4170 4171 4172 4173 4174 4175 4176 4177 4178 4179
	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;
4180 4181
}

4182 4183 4184 4185 4186
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
4187
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4188
	int ret;
4189

A
Alex Elder 已提交
4190
	ret = rbd_dev_refresh(rbd_dev);
4191
	if (ret)
4192
		return ret;
4193

4194
	return size;
4195
}
4196

4197
static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
A
Alex Elder 已提交
4198
static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL);
4199
static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
4200
static DEVICE_ATTR(minor, S_IRUGO, rbd_minor_show, NULL);
4201
static DEVICE_ATTR(client_addr, S_IRUGO, rbd_client_addr_show, NULL);
4202
static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
4203
static DEVICE_ATTR(cluster_fsid, S_IRUGO, rbd_cluster_fsid_show, NULL);
4204
static DEVICE_ATTR(config_info, S_IRUSR, rbd_config_info_show, NULL);
4205
static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
4206
static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL);
4207
static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
A
Alex Elder 已提交
4208
static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL);
4209 4210
static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
4211
static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL);
4212
static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL);
4213 4214 4215

static struct attribute *rbd_attrs[] = {
	&dev_attr_size.attr,
A
Alex Elder 已提交
4216
	&dev_attr_features.attr,
4217
	&dev_attr_major.attr,
4218
	&dev_attr_minor.attr,
4219
	&dev_attr_client_addr.attr,
4220
	&dev_attr_client_id.attr,
4221
	&dev_attr_cluster_fsid.attr,
4222
	&dev_attr_config_info.attr,
4223
	&dev_attr_pool.attr,
4224
	&dev_attr_pool_id.attr,
4225
	&dev_attr_name.attr,
A
Alex Elder 已提交
4226
	&dev_attr_image_id.attr,
4227
	&dev_attr_current_snap.attr,
4228
	&dev_attr_snap_id.attr,
4229
	&dev_attr_parent.attr,
4230 4231 4232 4233 4234 4235 4236 4237 4238 4239 4240 4241 4242
	&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
};

4243
static void rbd_dev_release(struct device *dev);
4244

4245
static const struct device_type rbd_device_type = {
4246 4247
	.name		= "rbd",
	.groups		= rbd_attr_groups,
4248
	.release	= rbd_dev_release,
4249 4250
};

4251 4252 4253 4254 4255 4256 4257 4258 4259 4260 4261 4262 4263 4264 4265 4266 4267 4268 4269 4270 4271
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;
4272 4273 4274

	spec->pool_id = CEPH_NOPOOL;
	spec->snap_id = CEPH_NOSNAP;
4275 4276 4277 4278 4279 4280 4281 4282 4283 4284 4285 4286 4287 4288 4289 4290
	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);
}

4291
static void rbd_dev_free(struct rbd_device *rbd_dev)
4292
{
4293
	WARN_ON(rbd_dev->watch_state != RBD_WATCH_STATE_UNREGISTERED);
4294
	WARN_ON(rbd_dev->lock_state != RBD_LOCK_STATE_UNLOCKED);
4295

4296
	ceph_oid_destroy(&rbd_dev->header_oid);
4297
	ceph_oloc_destroy(&rbd_dev->header_oloc);
4298
	kfree(rbd_dev->config_info);
4299

4300 4301 4302 4303
	rbd_put_client(rbd_dev->rbd_client);
	rbd_spec_put(rbd_dev->spec);
	kfree(rbd_dev->opts);
	kfree(rbd_dev);
4304 4305 4306 4307 4308 4309 4310 4311 4312 4313 4314 4315 4316
}

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);
4317 4318 4319 4320 4321 4322 4323 4324 4325 4326

	/*
	 * 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);
}

4327 4328
static struct rbd_device *__rbd_dev_create(struct rbd_client *rbdc,
					   struct rbd_spec *spec)
4329 4330 4331
{
	struct rbd_device *rbd_dev;

4332
	rbd_dev = kzalloc(sizeof(*rbd_dev), GFP_KERNEL);
4333 4334 4335 4336 4337 4338 4339
	if (!rbd_dev)
		return NULL;

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

4340
	rbd_dev->header.data_pool_id = CEPH_NOPOOL;
4341
	ceph_oid_init(&rbd_dev->header_oid);
4342
	rbd_dev->header_oloc.pool = spec->pool_id;
4343

4344 4345 4346 4347
	mutex_init(&rbd_dev->watch_mutex);
	rbd_dev->watch_state = RBD_WATCH_STATE_UNREGISTERED;
	INIT_DELAYED_WORK(&rbd_dev->watch_dwork, rbd_reregister_watch);

4348 4349 4350 4351 4352 4353 4354 4355
	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);

4356 4357 4358 4359 4360
	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);

4361
	rbd_dev->rbd_client = rbdc;
4362
	rbd_dev->spec = spec;
4363

4364 4365 4366 4367 4368 4369 4370 4371 4372 4373 4374 4375 4376 4377 4378 4379 4380 4381 4382 4383 4384 4385 4386 4387 4388 4389 4390 4391 4392 4393
	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;
4394

4395 4396
	/* we have a ref from do_rbd_add() */
	__module_get(THIS_MODULE);
4397

4398
	dout("%s rbd_dev %p dev_id %d\n", __func__, rbd_dev, rbd_dev->dev_id);
4399
	return rbd_dev;
4400 4401 4402 4403 4404 4405

fail_dev_id:
	ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id);
fail_rbd_dev:
	rbd_dev_free(rbd_dev);
	return NULL;
4406 4407 4408 4409
}

static void rbd_dev_destroy(struct rbd_device *rbd_dev)
{
4410 4411
	if (rbd_dev)
		put_device(&rbd_dev->dev);
4412 4413
}

4414 4415 4416 4417 4418 4419 4420 4421 4422 4423 4424 4425 4426 4427 4428
/*
 * 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 };

4429 4430 4431 4432
	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));
4433
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4434 4435
	if (ret < 0)
		return ret;
4436 4437
	if (ret < sizeof (size_buf))
		return -ERANGE;
4438

J
Josh Durgin 已提交
4439
	if (order) {
4440
		*order = size_buf.order;
J
Josh Durgin 已提交
4441 4442
		dout("  order %u", (unsigned int)*order);
	}
4443 4444
	*snap_size = le64_to_cpu(size_buf.size);

J
Josh Durgin 已提交
4445 4446
	dout("  snap_id 0x%016llx snap_size = %llu\n",
		(unsigned long long)snap_id,
4447
		(unsigned long long)*snap_size);
4448 4449 4450 4451 4452 4453 4454 4455 4456 4457 4458

	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);
}

4459 4460 4461 4462 4463 4464 4465 4466 4467 4468
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;

4469 4470 4471
	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);
4472
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4473 4474 4475 4476 4477
	if (ret < 0)
		goto out;

	p = reply_buf;
	rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p,
4478 4479
						p + ret, NULL, GFP_NOIO);
	ret = 0;
4480 4481 4482 4483 4484 4485 4486 4487 4488 4489 4490 4491 4492

	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;
}

4493 4494 4495 4496 4497 4498 4499
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;
4500
	} __attribute__ ((packed)) features_buf = { 0 };
4501
	u64 unsup;
4502 4503
	int ret;

4504 4505 4506 4507
	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));
4508
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4509 4510
	if (ret < 0)
		return ret;
4511 4512
	if (ret < sizeof (features_buf))
		return -ERANGE;
A
Alex Elder 已提交
4513

4514 4515 4516 4517
	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
4518
		return -ENXIO;
4519
	}
A
Alex Elder 已提交
4520

4521 4522 4523
	*snap_features = le64_to_cpu(features_buf.features);

	dout("  snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
4524 4525 4526
		(unsigned long long)snap_id,
		(unsigned long long)*snap_features,
		(unsigned long long)le64_to_cpu(features_buf.incompat));
4527 4528 4529 4530 4531 4532 4533 4534 4535 4536

	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);
}

4537 4538 4539 4540 4541 4542 4543 4544
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
4545
	u64 pool_id;
4546
	char *image_id;
4547
	u64 snap_id;
4548 4549 4550 4551 4552 4553 4554 4555 4556 4557 4558 4559 4560 4561 4562 4563 4564
	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;
	}

4565
	snapid = cpu_to_le64(rbd_dev->spec->snap_id);
4566 4567 4568
	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
				  &rbd_dev->header_oloc, "get_parent",
				  &snapid, sizeof(snapid), reply_buf, size);
4569
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4570 4571 4572 4573
	if (ret < 0)
		goto out_err;

	p = reply_buf;
4574 4575
	end = reply_buf + ret;
	ret = -ERANGE;
A
Alex Elder 已提交
4576
	ceph_decode_64_safe(&p, end, pool_id, out_err);
4577 4578 4579 4580 4581 4582 4583 4584 4585 4586 4587 4588 4589 4590 4591 4592 4593
	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);
		}

4594
		goto out;	/* No parent?  No problem. */
4595
	}
4596

4597 4598 4599
	/* The ceph file layout needs to fit pool id in 32 bits */

	ret = -EIO;
A
Alex Elder 已提交
4600
	if (pool_id > (u64)U32_MAX) {
4601
		rbd_warn(NULL, "parent pool id too large (%llu > %u)",
A
Alex Elder 已提交
4602
			(unsigned long long)pool_id, U32_MAX);
4603
		goto out_err;
A
Alex Elder 已提交
4604
	}
4605

A
Alex Elder 已提交
4606
	image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
4607 4608 4609 4610
	if (IS_ERR(image_id)) {
		ret = PTR_ERR(image_id);
		goto out_err;
	}
4611
	ceph_decode_64_safe(&p, end, snap_id, out_err);
4612 4613
	ceph_decode_64_safe(&p, end, overlap, out_err);

4614 4615 4616 4617 4618 4619 4620 4621 4622
	/*
	 * 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
4623 4624
		rbd_dev->parent_spec = parent_spec;
		parent_spec = NULL;	/* rbd_dev now owns this */
4625 4626
	} else {
		kfree(image_id);
4627 4628 4629
	}

	/*
4630 4631
	 * We always update the parent overlap.  If it's zero we issue
	 * a warning, as we will proceed as if there was no parent.
4632 4633 4634
	 */
	if (!overlap) {
		if (parent_spec) {
4635 4636 4637 4638
			/* refresh, careful to warn just once */
			if (rbd_dev->parent_overlap)
				rbd_warn(rbd_dev,
				    "clone now standalone (overlap became 0)");
4639
		} else {
4640 4641
			/* initial probe */
			rbd_warn(rbd_dev, "clone is standalone (overlap 0)");
4642
		}
A
Alex Elder 已提交
4643
	}
4644 4645
	rbd_dev->parent_overlap = overlap;

4646 4647 4648 4649 4650 4651 4652 4653 4654
out:
	ret = 0;
out_err:
	kfree(reply_buf);
	rbd_spec_put(parent_spec);

	return ret;
}

4655 4656 4657 4658 4659 4660 4661 4662 4663 4664
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;
	int ret;

4665 4666 4667
	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);
4668 4669 4670 4671 4672 4673 4674
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
	if (ret < 0)
		return ret;
	if (ret < size)
		return -ERANGE;

	p = &striping_info_buf;
I
Ilya Dryomov 已提交
4675 4676
	rbd_dev->header.stripe_unit = ceph_decode_64(&p);
	rbd_dev->header.stripe_count = ceph_decode_64(&p);
4677 4678 4679
	return 0;
}

4680 4681 4682 4683 4684 4685 4686 4687 4688 4689 4690 4691 4692 4693 4694 4695 4696 4697
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;
}

4698 4699
static char *rbd_dev_image_name(struct rbd_device *rbd_dev)
{
4700
	CEPH_DEFINE_OID_ONSTACK(oid);
4701 4702 4703 4704 4705 4706 4707 4708 4709 4710 4711 4712
	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
4713 4714
	len = strlen(rbd_dev->spec->image_id);
	image_id_size = sizeof (__le32) + len;
4715 4716 4717 4718 4719
	image_id = kmalloc(image_id_size, GFP_KERNEL);
	if (!image_id)
		return NULL;

	p = image_id;
4720
	end = image_id + image_id_size;
4721
	ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len);
4722 4723 4724 4725 4726 4727

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

4728 4729 4730 4731
	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);
4732 4733 4734
	if (ret < 0)
		goto out;
	p = reply_buf;
4735 4736
	end = reply_buf + ret;

4737 4738 4739 4740 4741 4742 4743 4744 4745 4746 4747 4748
	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;
}

4749 4750 4751 4752 4753 4754 4755 4756 4757 4758 4759 4760 4761 4762 4763 4764 4765 4766 4767 4768 4769 4770 4771 4772 4773 4774 4775 4776 4777 4778
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);
4779 4780 4781 4782 4783 4784 4785
		if (IS_ERR(snap_name)) {
			/* ignore no-longer existing snapshots */
			if (PTR_ERR(snap_name) == -ENOENT)
				continue;
			else
				break;
		}
4786 4787 4788 4789 4790 4791 4792 4793 4794 4795 4796 4797 4798 4799 4800 4801 4802 4803
		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);
}

4804
/*
4805 4806 4807 4808 4809 4810 4811 4812 4813 4814 4815 4816 4817 4818 4819 4820 4821 4822 4823 4824 4825 4826 4827 4828 4829 4830 4831
 * 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.
4832
 *
4833 4834
 * All names in an rbd spec are dynamically allocated.  It's OK if we
 * can't figure out the name for an image id.
4835
 */
4836
static int rbd_spec_fill_names(struct rbd_device *rbd_dev)
4837
{
4838 4839 4840 4841 4842
	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;
4843 4844
	int ret;

4845 4846 4847
	rbd_assert(spec->pool_id != CEPH_NOPOOL);
	rbd_assert(spec->image_id);
	rbd_assert(spec->snap_id != CEPH_NOSNAP);
4848

4849
	/* Get the pool name; we have to make our own copy of this */
4850

4851 4852 4853
	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);
4854 4855
		return -EIO;
	}
4856 4857
	pool_name = kstrdup(pool_name, GFP_KERNEL);
	if (!pool_name)
4858 4859 4860 4861
		return -ENOMEM;

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

4862 4863
	image_name = rbd_dev_image_name(rbd_dev);
	if (!image_name)
A
Alex Elder 已提交
4864
		rbd_warn(rbd_dev, "unable to get image name");
4865

4866
	/* Fetch the snapshot name */
4867

4868
	snap_name = rbd_snap_name(rbd_dev, spec->snap_id);
4869 4870
	if (IS_ERR(snap_name)) {
		ret = PTR_ERR(snap_name);
4871
		goto out_err;
4872 4873 4874 4875 4876
	}

	spec->pool_name = pool_name;
	spec->image_name = image_name;
	spec->snap_name = snap_name;
4877 4878

	return 0;
4879

4880
out_err:
4881 4882
	kfree(image_name);
	kfree(pool_name);
4883 4884 4885
	return ret;
}

A
Alex Elder 已提交
4886
static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev)
4887 4888 4889 4890 4891 4892 4893 4894 4895 4896 4897 4898 4899 4900 4901 4902 4903 4904 4905 4906 4907 4908 4909
{
	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;

4910 4911 4912
	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
				  &rbd_dev->header_oloc, "get_snapcontext",
				  NULL, 0, reply_buf, size);
4913
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4914 4915 4916 4917
	if (ret < 0)
		goto out;

	p = reply_buf;
4918 4919
	end = reply_buf + ret;
	ret = -ERANGE;
4920 4921 4922 4923 4924 4925 4926 4927 4928 4929 4930 4931 4932 4933 4934 4935
	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;
4936
	ret = 0;
4937

4938
	snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
4939 4940 4941 4942 4943 4944 4945 4946
	if (!snapc) {
		ret = -ENOMEM;
		goto out;
	}
	snapc->seq = seq;
	for (i = 0; i < snap_count; i++)
		snapc->snaps[i] = ceph_decode_64(&p);

4947
	ceph_put_snap_context(rbd_dev->header.snapc);
4948 4949 4950
	rbd_dev->header.snapc = snapc;

	dout("  snap context seq = %llu, snap_count = %u\n",
4951
		(unsigned long long)seq, (unsigned int)snap_count);
4952 4953 4954
out:
	kfree(reply_buf);

4955
	return ret;
4956 4957
}

4958 4959
static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
					u64 snap_id)
A
Alex Elder 已提交
4960 4961 4962
{
	size_t size;
	void *reply_buf;
4963
	__le64 snapid;
A
Alex Elder 已提交
4964 4965 4966 4967 4968 4969 4970 4971 4972 4973
	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);

4974
	snapid = cpu_to_le64(snap_id);
4975 4976 4977
	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
				  &rbd_dev->header_oloc, "get_snapshot_name",
				  &snapid, sizeof(snapid), reply_buf, size);
4978
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4979 4980
	if (ret < 0) {
		snap_name = ERR_PTR(ret);
A
Alex Elder 已提交
4981
		goto out;
4982
	}
A
Alex Elder 已提交
4983 4984

	p = reply_buf;
4985
	end = reply_buf + ret;
A
Alex Elder 已提交
4986
	snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
4987
	if (IS_ERR(snap_name))
A
Alex Elder 已提交
4988 4989
		goto out;

4990
	dout("  snap_id 0x%016llx snap_name = %s\n",
4991
		(unsigned long long)snap_id, snap_name);
A
Alex Elder 已提交
4992 4993 4994
out:
	kfree(reply_buf);

4995
	return snap_name;
A
Alex Elder 已提交
4996 4997
}

4998
static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev)
A
Alex Elder 已提交
4999
{
5000
	bool first_time = rbd_dev->header.object_prefix == NULL;
A
Alex Elder 已提交
5001 5002
	int ret;

5003 5004
	ret = rbd_dev_v2_image_size(rbd_dev);
	if (ret)
5005
		return ret;
5006

5007 5008 5009
	if (first_time) {
		ret = rbd_dev_v2_header_onetime(rbd_dev);
		if (ret)
5010
			return ret;
5011 5012
	}

A
Alex Elder 已提交
5013
	ret = rbd_dev_v2_snap_context(rbd_dev);
5014 5015 5016 5017
	if (ret && first_time) {
		kfree(rbd_dev->header.object_prefix);
		rbd_dev->header.object_prefix = NULL;
	}
A
Alex Elder 已提交
5018 5019 5020 5021

	return ret;
}

5022 5023 5024 5025 5026 5027 5028 5029 5030 5031
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);
}

5032 5033 5034
/*
 * 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
5035 5036
 * the token (string of non-white space characters) found.  Note
 * that *buf must be terminated with '\0'.
5037 5038 5039 5040 5041 5042 5043 5044 5045 5046 5047 5048 5049 5050
 */
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
5051 5052 5053 5054 5055 5056 5057 5058 5059 5060 5061 5062 5063 5064 5065 5066 5067 5068 5069 5070 5071 5072
/*
 * 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
5073
	dup = kmemdup(*buf, len + 1, GFP_KERNEL);
A
Alex Elder 已提交
5074 5075 5076 5077 5078 5079 5080 5081 5082 5083 5084
	if (!dup)
		return NULL;
	*(dup + len) = '\0';
	*buf += len;

	if (lenp)
		*lenp = len;

	return dup;
}

5085
/*
5086 5087 5088 5089
 * 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
5090
 *
5091 5092 5093 5094 5095 5096 5097 5098 5099 5100 5101 5102 5103 5104 5105 5106 5107 5108 5109 5110 5111 5112 5113 5114 5115 5116 5117 5118 5119 5120 5121 5122 5123 5124
 * 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.
5125
 */
5126
static int rbd_add_parse_args(const char *buf,
5127
				struct ceph_options **ceph_opts,
5128 5129
				struct rbd_options **opts,
				struct rbd_spec **rbd_spec)
5130
{
A
Alex Elder 已提交
5131
	size_t len;
5132
	char *options;
5133
	const char *mon_addrs;
5134
	char *snap_name;
5135
	size_t mon_addrs_size;
5136
	struct rbd_spec *spec = NULL;
5137
	struct rbd_options *rbd_opts = NULL;
5138
	struct ceph_options *copts;
5139
	int ret;
5140 5141 5142

	/* The first four tokens are required */

5143
	len = next_token(&buf);
5144 5145 5146 5147
	if (!len) {
		rbd_warn(NULL, "no monitor address(es) provided");
		return -EINVAL;
	}
5148
	mon_addrs = buf;
5149
	mon_addrs_size = len + 1;
5150
	buf += len;
5151

5152
	ret = -EINVAL;
5153 5154
	options = dup_token(&buf, NULL);
	if (!options)
5155
		return -ENOMEM;
5156 5157 5158 5159
	if (!*options) {
		rbd_warn(NULL, "no options provided");
		goto out_err;
	}
5160

5161 5162
	spec = rbd_spec_alloc();
	if (!spec)
5163
		goto out_mem;
5164 5165 5166 5167

	spec->pool_name = dup_token(&buf, NULL);
	if (!spec->pool_name)
		goto out_mem;
5168 5169 5170 5171
	if (!*spec->pool_name) {
		rbd_warn(NULL, "no pool name provided");
		goto out_err;
	}
5172

A
Alex Elder 已提交
5173
	spec->image_name = dup_token(&buf, NULL);
5174
	if (!spec->image_name)
5175
		goto out_mem;
5176 5177 5178 5179
	if (!*spec->image_name) {
		rbd_warn(NULL, "no image name provided");
		goto out_err;
	}
5180

5181 5182 5183 5184
	/*
	 * Snapshot name is optional; default is to use "-"
	 * (indicating the head/no snapshot).
	 */
5185
	len = next_token(&buf);
5186
	if (!len) {
5187 5188
		buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */
		len = sizeof (RBD_SNAP_HEAD_NAME) - 1;
5189
	} else if (len > RBD_MAX_SNAP_NAME_LEN) {
5190
		ret = -ENAMETOOLONG;
5191
		goto out_err;
5192
	}
5193 5194
	snap_name = kmemdup(buf, len + 1, GFP_KERNEL);
	if (!snap_name)
5195
		goto out_mem;
5196 5197
	*(snap_name + len) = '\0';
	spec->snap_name = snap_name;
A
Alex Elder 已提交
5198

5199
	/* Initialize all rbd options to the defaults */
5200

5201 5202 5203 5204 5205
	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
5206
	rbd_opts->queue_depth = RBD_QUEUE_DEPTH_DEFAULT;
5207
	rbd_opts->lock_timeout = RBD_LOCK_TIMEOUT_DEFAULT;
I
Ilya Dryomov 已提交
5208
	rbd_opts->lock_on_read = RBD_LOCK_ON_READ_DEFAULT;
I
Ilya Dryomov 已提交
5209
	rbd_opts->exclusive = RBD_EXCLUSIVE_DEFAULT;
A
Alex Elder 已提交
5210

5211
	copts = ceph_parse_options(options, mon_addrs,
5212
					mon_addrs + mon_addrs_size - 1,
5213
					parse_rbd_opts_token, rbd_opts);
5214 5215
	if (IS_ERR(copts)) {
		ret = PTR_ERR(copts);
5216 5217
		goto out_err;
	}
5218 5219 5220
	kfree(options);

	*ceph_opts = copts;
5221
	*opts = rbd_opts;
5222
	*rbd_spec = spec;
5223

5224
	return 0;
5225
out_mem:
5226
	ret = -ENOMEM;
A
Alex Elder 已提交
5227
out_err:
5228 5229
	kfree(rbd_opts);
	rbd_spec_put(spec);
5230
	kfree(options);
A
Alex Elder 已提交
5231

5232
	return ret;
5233 5234
}

I
Ilya Dryomov 已提交
5235 5236 5237 5238 5239 5240 5241 5242 5243 5244
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)
{
5245 5246
	int ret;

I
Ilya Dryomov 已提交
5247 5248 5249 5250 5251 5252 5253
	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);
5254
	ret = rbd_wait_state_locked(rbd_dev, true);
I
Ilya Dryomov 已提交
5255
	up_read(&rbd_dev->lock_rwsem);
5256
	if (ret) {
I
Ilya Dryomov 已提交
5257 5258 5259 5260 5261 5262 5263
		rbd_warn(rbd_dev, "failed to acquire exclusive lock");
		return -EROFS;
	}

	return 0;
}

A
Alex Elder 已提交
5264 5265 5266 5267 5268 5269 5270 5271 5272 5273 5274 5275 5276 5277 5278 5279 5280 5281
/*
 * 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;
5282
	CEPH_DEFINE_OID_ONSTACK(oid);
A
Alex Elder 已提交
5283
	void *response;
5284
	char *image_id;
5285

A
Alex Elder 已提交
5286 5287 5288
	/*
	 * When probing a parent image, the image id is already
	 * known (and the image name likely is not).  There's no
5289 5290
	 * 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
5291
	 */
5292 5293 5294
	if (rbd_dev->spec->image_id) {
		rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1;

A
Alex Elder 已提交
5295
		return 0;
5296
	}
A
Alex Elder 已提交
5297

A
Alex Elder 已提交
5298 5299 5300 5301
	/*
	 * First, see if the format 2 image id file exists, and if
	 * so, get the image's persistent id from it.
	 */
5302 5303 5304 5305 5306 5307
	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
5308 5309 5310 5311 5312 5313 5314 5315 5316 5317

	/* 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;
	}

5318 5319
	/* If it doesn't exist we'll assume it's a format 1 image */

5320 5321 5322
	ret = rbd_obj_method_sync(rbd_dev, &oid, &rbd_dev->header_oloc,
				  "get_id", NULL, 0,
				  response, RBD_IMAGE_ID_LEN_MAX);
5323
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
5324 5325 5326 5327 5328
	if (ret == -ENOENT) {
		image_id = kstrdup("", GFP_KERNEL);
		ret = image_id ? 0 : -ENOMEM;
		if (!ret)
			rbd_dev->image_format = 1;
5329
	} else if (ret >= 0) {
5330 5331 5332
		void *p = response;

		image_id = ceph_extract_encoded_string(&p, p + ret,
A
Alex Elder 已提交
5333
						NULL, GFP_NOIO);
5334
		ret = PTR_ERR_OR_ZERO(image_id);
5335 5336 5337 5338 5339 5340 5341
		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
5342 5343 5344
	}
out:
	kfree(response);
5345
	ceph_oid_destroy(&oid);
A
Alex Elder 已提交
5346 5347 5348
	return ret;
}

A
Alex Elder 已提交
5349 5350 5351 5352
/*
 * Undo whatever state changes are made by v1 or v2 header info
 * call.
 */
A
Alex Elder 已提交
5353 5354 5355 5356
static void rbd_dev_unprobe(struct rbd_device *rbd_dev)
{
	struct rbd_image_header	*header;

5357
	rbd_dev_parent_put(rbd_dev);
A
Alex Elder 已提交
5358 5359 5360 5361

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

	header = &rbd_dev->header;
5362
	ceph_put_snap_context(header->snapc);
A
Alex Elder 已提交
5363 5364 5365 5366 5367 5368
	kfree(header->snap_sizes);
	kfree(header->snap_names);
	kfree(header->object_prefix);
	memset(header, 0, sizeof (*header));
}

5369
static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev)
5370 5371 5372
{
	int ret;

5373
	ret = rbd_dev_v2_object_prefix(rbd_dev);
5374
	if (ret)
5375 5376
		goto out_err;

5377 5378 5379 5380
	/*
	 * Get the and check features for the image.  Currently the
	 * features are assumed to never change.
	 */
5381
	ret = rbd_dev_v2_features(rbd_dev);
5382
	if (ret)
5383
		goto out_err;
5384

5385 5386 5387 5388 5389 5390 5391
	/* 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;
	}
5392

5393 5394 5395 5396 5397 5398
	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
5399
	rbd_init_layout(rbd_dev);
A
Alex Elder 已提交
5400
	return 0;
I
Ilya Dryomov 已提交
5401

5402
out_err:
A
Alex Elder 已提交
5403
	rbd_dev->header.features = 0;
5404 5405
	kfree(rbd_dev->header.object_prefix);
	rbd_dev->header.object_prefix = NULL;
5406
	return ret;
5407 5408
}

5409 5410 5411 5412 5413 5414
/*
 * @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
5415
{
5416
	struct rbd_device *parent = NULL;
5417 5418 5419 5420 5421
	int ret;

	if (!rbd_dev->parent_spec)
		return 0;

5422 5423 5424 5425 5426 5427
	if (++depth > RBD_MAX_PARENT_CHAIN_LEN) {
		pr_info("parent chain is too long (%d)\n", depth);
		ret = -EINVAL;
		goto out_err;
	}

5428
	parent = __rbd_dev_create(rbd_dev->rbd_client, rbd_dev->parent_spec);
5429 5430
	if (!parent) {
		ret = -ENOMEM;
5431
		goto out_err;
5432 5433 5434 5435 5436 5437 5438 5439
	}

	/*
	 * 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);
5440

5441
	ret = rbd_dev_image_probe(parent, depth);
5442 5443
	if (ret < 0)
		goto out_err;
5444

5445
	rbd_dev->parent = parent;
5446
	atomic_set(&rbd_dev->parent_ref, 1);
5447
	return 0;
5448

5449
out_err:
5450
	rbd_dev_unparent(rbd_dev);
5451
	rbd_dev_destroy(parent);
5452 5453 5454
	return ret;
}

5455 5456 5457 5458 5459 5460 5461 5462 5463
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
5464 5465 5466 5467
/*
 * rbd_dev->header_rwsem must be locked for write and will be unlocked
 * upon return.
 */
5468
static int rbd_dev_device_setup(struct rbd_device *rbd_dev)
5469
{
A
Alex Elder 已提交
5470
	int ret;
A
Alex Elder 已提交
5471

5472
	/* Record our major and minor device numbers. */
A
Alex Elder 已提交
5473

5474 5475 5476
	if (!single_major) {
		ret = register_blkdev(0, rbd_dev->name);
		if (ret < 0)
5477
			goto err_out_unlock;
5478 5479 5480 5481 5482 5483 5484

		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
5485 5486 5487 5488 5489 5490 5491

	/* Set up the blkdev mapping. */

	ret = rbd_init_disk(rbd_dev);
	if (ret)
		goto err_out_blkdev;

5492
	ret = rbd_dev_mapping_set(rbd_dev);
A
Alex Elder 已提交
5493 5494
	if (ret)
		goto err_out_disk;
I
Ilya Dryomov 已提交
5495

5496
	set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE);
5497
	set_disk_ro(rbd_dev->disk, rbd_dev->opts->read_only);
5498

5499
	ret = dev_set_name(&rbd_dev->dev, "%d", rbd_dev->dev_id);
5500
	if (ret)
5501
		goto err_out_mapping;
A
Alex Elder 已提交
5502

5503
	set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
I
Ilya Dryomov 已提交
5504
	up_write(&rbd_dev->header_rwsem);
5505
	return 0;
5506

5507 5508
err_out_mapping:
	rbd_dev_mapping_clear(rbd_dev);
A
Alex Elder 已提交
5509 5510 5511
err_out_disk:
	rbd_free_disk(rbd_dev);
err_out_blkdev:
5512 5513
	if (!single_major)
		unregister_blkdev(rbd_dev->major, rbd_dev->name);
I
Ilya Dryomov 已提交
5514 5515
err_out_unlock:
	up_write(&rbd_dev->header_rwsem);
A
Alex Elder 已提交
5516 5517 5518
	return ret;
}

A
Alex Elder 已提交
5519 5520 5521
static int rbd_dev_header_name(struct rbd_device *rbd_dev)
{
	struct rbd_spec *spec = rbd_dev->spec;
5522
	int ret;
A
Alex Elder 已提交
5523 5524 5525 5526 5527

	/* 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)
5528 5529
		ret = ceph_oid_aprintf(&rbd_dev->header_oid, GFP_KERNEL, "%s%s",
				       spec->image_name, RBD_SUFFIX);
A
Alex Elder 已提交
5530
	else
5531 5532
		ret = ceph_oid_aprintf(&rbd_dev->header_oid, GFP_KERNEL, "%s%s",
				       RBD_HEADER_PREFIX, spec->image_id);
A
Alex Elder 已提交
5533

5534
	return ret;
A
Alex Elder 已提交
5535 5536
}

5537 5538
static void rbd_dev_image_release(struct rbd_device *rbd_dev)
{
A
Alex Elder 已提交
5539
	rbd_dev_unprobe(rbd_dev);
5540 5541
	if (rbd_dev->opts)
		rbd_unregister_watch(rbd_dev);
A
Alex Elder 已提交
5542 5543 5544
	rbd_dev->image_format = 0;
	kfree(rbd_dev->spec->image_id);
	rbd_dev->spec->image_id = NULL;
5545 5546
}

5547 5548
/*
 * Probe for the existence of the header object for the given rbd
5549 5550 5551
 * 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.
5552
 */
5553
static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth)
5554 5555 5556 5557
{
	int ret;

	/*
A
Alex Elder 已提交
5558 5559 5560 5561
	 * 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.
5562 5563 5564
	 */
	ret = rbd_dev_image_id(rbd_dev);
	if (ret)
5565 5566
		return ret;

A
Alex Elder 已提交
5567 5568 5569 5570
	ret = rbd_dev_header_name(rbd_dev);
	if (ret)
		goto err_out_format;

5571
	if (!depth) {
5572
		ret = rbd_register_watch(rbd_dev);
5573 5574 5575 5576 5577
		if (ret) {
			if (ret == -ENOENT)
				pr_info("image %s/%s does not exist\n",
					rbd_dev->spec->pool_name,
					rbd_dev->spec->image_name);
5578
			goto err_out_format;
5579
		}
5580
	}
5581

5582
	ret = rbd_dev_header_info(rbd_dev);
5583
	if (ret)
5584
		goto err_out_watch;
A
Alex Elder 已提交
5585

5586 5587 5588 5589 5590 5591
	/*
	 * 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.
	 */
5592
	if (!depth)
5593 5594 5595
		ret = rbd_spec_fill_snap_id(rbd_dev);
	else
		ret = rbd_spec_fill_names(rbd_dev);
5596 5597 5598 5599 5600 5601
	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
5602
		goto err_out_probe;
5603
	}
5604

5605 5606 5607 5608 5609 5610 5611 5612 5613
	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.
		 */
5614
		if (!depth && rbd_dev->parent_spec)
5615 5616 5617 5618
			rbd_warn(rbd_dev,
				 "WARNING: kernel layering is EXPERIMENTAL!");
	}

5619
	ret = rbd_dev_probe_parent(rbd_dev, depth);
A
Alex Elder 已提交
5620 5621 5622 5623
	if (ret)
		goto err_out_probe;

	dout("discovered format %u image, header name is %s\n",
5624
		rbd_dev->image_format, rbd_dev->header_oid.name);
A
Alex Elder 已提交
5625
	return 0;
5626

A
Alex Elder 已提交
5627 5628
err_out_probe:
	rbd_dev_unprobe(rbd_dev);
5629
err_out_watch:
5630
	if (!depth)
5631
		rbd_unregister_watch(rbd_dev);
A
Alex Elder 已提交
5632 5633
err_out_format:
	rbd_dev->image_format = 0;
5634 5635
	kfree(rbd_dev->spec->image_id);
	rbd_dev->spec->image_id = NULL;
5636 5637 5638
	return ret;
}

5639 5640 5641
static ssize_t do_rbd_add(struct bus_type *bus,
			  const char *buf,
			  size_t count)
5642
{
5643
	struct rbd_device *rbd_dev = NULL;
5644
	struct ceph_options *ceph_opts = NULL;
5645
	struct rbd_options *rbd_opts = NULL;
5646
	struct rbd_spec *spec = NULL;
5647
	struct rbd_client *rbdc;
5648
	int rc;
5649 5650 5651 5652 5653

	if (!try_module_get(THIS_MODULE))
		return -ENODEV;

	/* parse add command */
5654
	rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec);
5655
	if (rc < 0)
5656
		goto out;
5657

5658 5659 5660
	rbdc = rbd_get_client(ceph_opts);
	if (IS_ERR(rbdc)) {
		rc = PTR_ERR(rbdc);
5661
		goto err_out_args;
5662
	}
5663 5664

	/* pick the pool */
5665
	rc = ceph_pg_poolid_by_name(rbdc->client->osdc.osdmap, spec->pool_name);
5666 5667 5668
	if (rc < 0) {
		if (rc == -ENOENT)
			pr_info("pool %s does not exist\n", spec->pool_name);
5669
		goto err_out_client;
5670
	}
A
Alex Elder 已提交
5671
	spec->pool_id = (u64)rc;
5672

5673
	rbd_dev = rbd_dev_create(rbdc, spec, rbd_opts);
5674 5675
	if (!rbd_dev) {
		rc = -ENOMEM;
5676
		goto err_out_client;
5677
	}
5678 5679
	rbdc = NULL;		/* rbd_dev now owns this */
	spec = NULL;		/* rbd_dev now owns this */
5680
	rbd_opts = NULL;	/* rbd_dev now owns this */
5681

5682 5683 5684 5685 5686 5687
	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
5688
	down_write(&rbd_dev->header_rwsem);
5689
	rc = rbd_dev_image_probe(rbd_dev, 0);
5690 5691
	if (rc < 0) {
		up_write(&rbd_dev->header_rwsem);
5692
		goto err_out_rbd_dev;
5693
	}
5694

5695 5696
	/* If we are mapping a snapshot it must be marked read-only */
	if (rbd_dev->spec->snap_id != CEPH_NOSNAP)
5697
		rbd_dev->opts->read_only = true;
5698

5699
	rc = rbd_dev_device_setup(rbd_dev);
5700
	if (rc)
5701
		goto err_out_image_probe;
A
Alex Elder 已提交
5702

I
Ilya Dryomov 已提交
5703 5704 5705 5706
	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
5707 5708
	}

5709 5710 5711 5712
	/* Everything's ready.  Announce the disk to the world. */

	rc = device_add(&rbd_dev->dev);
	if (rc)
I
Ilya Dryomov 已提交
5713
		goto err_out_image_lock;
5714 5715 5716 5717 5718 5719 5720 5721 5722 5723 5724 5725

	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);
5726 5727 5728 5729
	rc = count;
out:
	module_put(THIS_MODULE);
	return rc;
5730

I
Ilya Dryomov 已提交
5731 5732
err_out_image_lock:
	rbd_dev_image_unlock(rbd_dev);
5733 5734
err_out_device_setup:
	rbd_dev_device_release(rbd_dev);
5735 5736
err_out_image_probe:
	rbd_dev_image_release(rbd_dev);
5737 5738
err_out_rbd_dev:
	rbd_dev_destroy(rbd_dev);
5739
err_out_client:
5740
	rbd_put_client(rbdc);
5741
err_out_args:
5742
	rbd_spec_put(spec);
5743
	kfree(rbd_opts);
5744
	goto out;
5745 5746
}

5747 5748 5749 5750 5751 5752 5753 5754 5755 5756 5757 5758 5759 5760 5761 5762 5763
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);
}

5764 5765
static void rbd_dev_remove_parent(struct rbd_device *rbd_dev)
{
A
Alex Elder 已提交
5766
	while (rbd_dev->parent) {
5767 5768 5769 5770 5771 5772 5773 5774 5775 5776 5777 5778
		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
5779
		rbd_assert(second);
5780
		rbd_dev_image_release(second);
5781
		rbd_dev_destroy(second);
A
Alex Elder 已提交
5782 5783 5784 5785
		first->parent = NULL;
		first->parent_overlap = 0;

		rbd_assert(first->parent_spec);
5786 5787 5788 5789 5790
		rbd_spec_put(first->parent_spec);
		first->parent_spec = NULL;
	}
}

5791 5792 5793
static ssize_t do_rbd_remove(struct bus_type *bus,
			     const char *buf,
			     size_t count)
5794 5795
{
	struct rbd_device *rbd_dev = NULL;
5796 5797
	struct list_head *tmp;
	int dev_id;
M
Mike Christie 已提交
5798
	char opt_buf[6];
5799
	bool already = false;
M
Mike Christie 已提交
5800
	bool force = false;
5801
	int ret;
5802

M
Mike Christie 已提交
5803 5804 5805 5806 5807
	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");
5808
		return -EINVAL;
M
Mike Christie 已提交
5809 5810 5811 5812 5813 5814 5815 5816 5817
	}
	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;
		}
	}
5818

5819 5820 5821 5822 5823 5824 5825 5826
	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;
		}
5827
	}
5828 5829
	if (!ret) {
		spin_lock_irq(&rbd_dev->lock);
M
Mike Christie 已提交
5830
		if (rbd_dev->open_count && !force)
5831 5832
			ret = -EBUSY;
		else
5833 5834
			already = test_and_set_bit(RBD_DEV_FLAG_REMOVING,
							&rbd_dev->flags);
5835 5836 5837
		spin_unlock_irq(&rbd_dev->lock);
	}
	spin_unlock(&rbd_dev_list_lock);
5838
	if (ret < 0 || already)
5839
		return ret;
5840

M
Mike Christie 已提交
5841 5842 5843 5844 5845 5846 5847 5848 5849
	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);
	}

5850 5851 5852 5853 5854
	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);
5855

I
Ilya Dryomov 已提交
5856
	rbd_dev_image_unlock(rbd_dev);
5857
	rbd_dev_device_release(rbd_dev);
5858
	rbd_dev_image_release(rbd_dev);
5859
	rbd_dev_destroy(rbd_dev);
5860
	return count;
5861 5862
}

5863 5864 5865 5866 5867 5868 5869 5870 5871 5872 5873 5874 5875 5876 5877 5878 5879
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);
}

5880 5881
/*
 * create control files in sysfs
5882
 * /sys/bus/rbd/...
5883 5884 5885
 */
static int rbd_sysfs_init(void)
{
5886
	int ret;
5887

5888
	ret = device_register(&rbd_root_dev);
A
Alex Elder 已提交
5889
	if (ret < 0)
5890
		return ret;
5891

5892 5893 5894
	ret = bus_register(&rbd_bus_type);
	if (ret < 0)
		device_unregister(&rbd_root_dev);
5895 5896 5897 5898 5899 5900

	return ret;
}

static void rbd_sysfs_cleanup(void)
{
5901
	bus_unregister(&rbd_bus_type);
5902
	device_unregister(&rbd_root_dev);
5903 5904
}

5905 5906 5907
static int rbd_slab_init(void)
{
	rbd_assert(!rbd_img_request_cache);
G
Geliang Tang 已提交
5908
	rbd_img_request_cache = KMEM_CACHE(rbd_img_request, 0);
5909 5910 5911 5912
	if (!rbd_img_request_cache)
		return -ENOMEM;

	rbd_assert(!rbd_obj_request_cache);
G
Geliang Tang 已提交
5913
	rbd_obj_request_cache = KMEM_CACHE(rbd_obj_request, 0);
5914 5915 5916
	if (!rbd_obj_request_cache)
		goto out_err;

5917
	return 0;
5918

5919
out_err:
5920 5921
	kmem_cache_destroy(rbd_img_request_cache);
	rbd_img_request_cache = NULL;
5922 5923 5924 5925 5926
	return -ENOMEM;
}

static void rbd_slab_exit(void)
{
5927 5928 5929 5930
	rbd_assert(rbd_obj_request_cache);
	kmem_cache_destroy(rbd_obj_request_cache);
	rbd_obj_request_cache = NULL;

5931 5932 5933 5934 5935
	rbd_assert(rbd_img_request_cache);
	kmem_cache_destroy(rbd_img_request_cache);
	rbd_img_request_cache = NULL;
}

A
Alex Elder 已提交
5936
static int __init rbd_init(void)
5937 5938 5939
{
	int rc;

5940 5941 5942 5943
	if (!libceph_compatible(NULL)) {
		rbd_warn(NULL, "libceph incompatibility (quitting)");
		return -EINVAL;
	}
I
Ilya Dryomov 已提交
5944

5945
	rc = rbd_slab_init();
5946 5947
	if (rc)
		return rc;
I
Ilya Dryomov 已提交
5948

5949 5950
	/*
	 * The number of active work items is limited by the number of
I
Ilya Dryomov 已提交
5951
	 * rbd devices * queue depth, so leave @max_active at default.
5952 5953 5954 5955 5956 5957 5958
	 */
	rbd_wq = alloc_workqueue(RBD_DRV_NAME, WQ_MEM_RECLAIM, 0);
	if (!rbd_wq) {
		rc = -ENOMEM;
		goto err_out_slab;
	}

5959 5960 5961 5962
	if (single_major) {
		rbd_major = register_blkdev(0, RBD_DRV_NAME);
		if (rbd_major < 0) {
			rc = rbd_major;
5963
			goto err_out_wq;
5964 5965 5966
		}
	}

5967 5968
	rc = rbd_sysfs_init();
	if (rc)
5969 5970 5971 5972 5973 5974
		goto err_out_blkdev;

	if (single_major)
		pr_info("loaded (major %d)\n", rbd_major);
	else
		pr_info("loaded\n");
5975

I
Ilya Dryomov 已提交
5976 5977
	return 0;

5978 5979 5980
err_out_blkdev:
	if (single_major)
		unregister_blkdev(rbd_major, RBD_DRV_NAME);
5981 5982
err_out_wq:
	destroy_workqueue(rbd_wq);
I
Ilya Dryomov 已提交
5983 5984
err_out_slab:
	rbd_slab_exit();
5985
	return rc;
5986 5987
}

A
Alex Elder 已提交
5988
static void __exit rbd_exit(void)
5989
{
I
Ilya Dryomov 已提交
5990
	ida_destroy(&rbd_dev_id_ida);
5991
	rbd_sysfs_cleanup();
5992 5993
	if (single_major)
		unregister_blkdev(rbd_major, RBD_DRV_NAME);
5994
	destroy_workqueue(rbd_wq);
5995
	rbd_slab_exit();
5996 5997 5998 5999 6000
}

module_init(rbd_init);
module_exit(rbd_exit);

A
Alex Elder 已提交
6001
MODULE_AUTHOR("Alex Elder <elder@inktank.com>");
6002 6003 6004 6005 6006
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>");

6007
MODULE_DESCRIPTION("RADOS Block Device (RBD) driver");
6008
MODULE_LICENSE("GPL");