rbd.c 151.8 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
/*
 * 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;

64
	counter = (unsigned int)atomic_fetch_add_unless(v, 1, 0);
65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86
	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
	const char	*pool_ns;	/* NULL if default, never "" */
185

186 187
	const char	*image_id;
	const char	*image_name;
188 189

	u64		snap_id;
190
	const char	*snap_name;
191 192 193 194

	struct kref	kref;
};

195
/*
A
Alex Elder 已提交
196
 * an instance of the client.  multiple devices may share an rbd client.
197 198 199 200 201 202 203
 */
struct rbd_client {
	struct ceph_client	*client;
	struct kref		kref;
	struct list_head	node;
};

A
Alex Elder 已提交
204 205
struct rbd_img_request;

206
enum obj_request_type {
207
	OBJ_REQUEST_NODATA = 1,
208
	OBJ_REQUEST_BIO,	/* pointer into provided bio (list) */
209
	OBJ_REQUEST_BVECS,	/* pointer into provided bio_vec array */
210
	OBJ_REQUEST_OWN_BVECS,	/* private bio_vec array, doesn't own pages */
211
};
A
Alex Elder 已提交
212

G
Guangliang Zhao 已提交
213
enum obj_operation_type {
214
	OBJ_OP_READ = 1,
G
Guangliang Zhao 已提交
215
	OBJ_OP_WRITE,
216
	OBJ_OP_DISCARD,
G
Guangliang Zhao 已提交
217 218
};

I
Ilya Dryomov 已提交
219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238
/*
 * 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,
239 240
};

A
Alex Elder 已提交
241
struct rbd_obj_request {
242
	struct ceph_object_extent ex;
243
	union {
I
Ilya Dryomov 已提交
244 245
		bool			tried_parent;	/* for reads */
		enum rbd_obj_write_state write_state;	/* for writes */
246
	};
A
Alex Elder 已提交
247

248
	struct rbd_img_request	*img_request;
I
Ilya Dryomov 已提交
249 250
	struct ceph_file_extent	*img_extents;
	u32			num_img_extents;
A
Alex Elder 已提交
251

252
	union {
253
		struct ceph_bio_iter	bio_pos;
254
		struct {
255 256
			struct ceph_bvec_iter	bvec_pos;
			u32			bvec_count;
257
			u32			bvec_idx;
258 259
		};
	};
260 261
	struct bio_vec		*copyup_bvecs;
	u32			copyup_bvec_count;
A
Alex Elder 已提交
262 263 264 265

	struct ceph_osd_request	*osd_req;

	u64			xferred;	/* bytes transferred */
266
	int			result;
A
Alex Elder 已提交
267 268 269 270

	struct kref		kref;
};

A
Alex Elder 已提交
271
enum img_req_flags {
272
	IMG_REQ_CHILD,		/* initiator: block = 0, child image = 1 */
273
	IMG_REQ_LAYERED,	/* ENOENT handling: normal = 0, layered = 1 */
A
Alex Elder 已提交
274 275
};

A
Alex Elder 已提交
276 277
struct rbd_img_request {
	struct rbd_device	*rbd_dev;
I
Ilya Dryomov 已提交
278
	enum obj_operation_type	op_type;
279
	enum obj_request_type	data_type;
A
Alex Elder 已提交
280
	unsigned long		flags;
A
Alex Elder 已提交
281
	union {
282
		u64			snap_id;	/* for reads */
A
Alex Elder 已提交
283
		struct ceph_snap_context *snapc;	/* for writes */
284 285 286 287
	};
	union {
		struct request		*rq;		/* block request */
		struct rbd_obj_request	*obj_request;	/* obj req initiator */
A
Alex Elder 已提交
288
	};
289
	spinlock_t		completion_lock;
290
	u64			xferred;/* aggregate bytes transferred */
291
	int			result;	/* first nonzero obj_request result */
A
Alex Elder 已提交
292

293
	struct list_head	object_extents;	/* obj_req.ex structs */
A
Alex Elder 已提交
294
	u32			obj_request_count;
I
Ilya Dryomov 已提交
295
	u32			pending_count;
A
Alex Elder 已提交
296 297 298 299 300

	struct kref		kref;
};

#define for_each_obj_request(ireq, oreq) \
301
	list_for_each_entry(oreq, &(ireq)->object_extents, ex.oe_item)
A
Alex Elder 已提交
302
#define for_each_obj_request_safe(ireq, oreq, n) \
303
	list_for_each_entry_safe(oreq, n, &(ireq)->object_extents, ex.oe_item)
A
Alex Elder 已提交
304

305 306 307 308 309 310
enum rbd_watch_state {
	RBD_WATCH_STATE_UNREGISTERED,
	RBD_WATCH_STATE_REGISTERED,
	RBD_WATCH_STATE_ERROR,
};

311 312 313 314 315 316 317 318 319 320 321 322
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
323
struct rbd_mapping {
A
Alex Elder 已提交
324
	u64                     size;
A
Alex Elder 已提交
325
	u64                     features;
A
Alex Elder 已提交
326 327
};

328 329 330 331
/*
 * a single device
 */
struct rbd_device {
A
Alex Elder 已提交
332
	int			dev_id;		/* blkdev unique id */
333 334

	int			major;		/* blkdev assigned major */
335
	int			minor;
336 337
	struct gendisk		*disk;		/* blkdev's gendisk and rq */

338
	u32			image_format;	/* Either 1 or 2 */
339 340 341 342
	struct rbd_client	*rbd_client;

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

343
	spinlock_t		lock;		/* queue, flags, open_count */
344 345

	struct rbd_image_header	header;
346
	unsigned long		flags;		/* possibly lock protected */
347
	struct rbd_spec		*spec;
348
	struct rbd_options	*opts;
349
	char			*config_info;	/* add{,_single_major} string */
350

351
	struct ceph_object_id	header_oid;
352
	struct ceph_object_locator header_oloc;
353

354
	struct ceph_file_layout	layout;		/* used for all rbd requests */
355

356 357
	struct mutex		watch_mutex;
	enum rbd_watch_state	watch_state;
358
	struct ceph_osd_linger_request *watch_handle;
359 360
	u64			watch_cookie;
	struct delayed_work	watch_dwork;
361

362 363
	struct rw_semaphore	lock_rwsem;
	enum rbd_lock_state	lock_state;
I
Ilya Dryomov 已提交
364
	char			lock_cookie[32];
365 366 367 368 369 370 371
	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;

372
	struct workqueue_struct	*task_wq;
373

374 375
	struct rbd_spec		*parent_spec;
	u64			parent_overlap;
376
	atomic_t		parent_ref;
377
	struct rbd_device	*parent;
378

C
Christoph Hellwig 已提交
379 380 381
	/* Block layer tags. */
	struct blk_mq_tag_set	tag_set;

382 383
	/* protects updating the header */
	struct rw_semaphore     header_rwsem;
A
Alex Elder 已提交
384 385

	struct rbd_mapping	mapping;
386 387

	struct list_head	node;
388 389 390

	/* sysfs related */
	struct device		dev;
391
	unsigned long		open_count;	/* protected by lock */
392 393
};

394
/*
395 396 397 398
 * 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
399
 */
400 401
enum rbd_dev_flags {
	RBD_DEV_FLAG_EXISTS,	/* mapped snapshot has not been deleted */
402
	RBD_DEV_FLAG_REMOVING,	/* this mapping is being removed */
403
	RBD_DEV_FLAG_BLACKLISTED, /* our ceph_client is blacklisted */
404 405
};

406
static DEFINE_MUTEX(client_mutex);	/* Serialize client creation */
407

408
static LIST_HEAD(rbd_dev_list);    /* devices */
409 410
static DEFINE_SPINLOCK(rbd_dev_list_lock);

A
Alex Elder 已提交
411 412
static LIST_HEAD(rbd_client_list);		/* clients */
static DEFINE_SPINLOCK(rbd_client_list_lock);
413

414 415
/* Slab caches for frequently-allocated structures */

416
static struct kmem_cache	*rbd_img_request_cache;
417
static struct kmem_cache	*rbd_obj_request_cache;
418

419
static int rbd_major;
420 421
static DEFINE_IDA(rbd_dev_id_ida);

422 423
static struct workqueue_struct *rbd_wq;

424
/*
425
 * single-major requires >= 0.75 version of userspace rbd utility.
426
 */
427
static bool single_major = true;
428
module_param(single_major, bool, 0444);
429
MODULE_PARM_DESC(single_major, "Use a single major number for all rbd devices (default: true)");
430

A
Alex Elder 已提交
431 432 433 434
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);
435 436 437 438
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);
439
static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth);
A
Alex Elder 已提交
440

441 442
static int rbd_dev_id_to_minor(int dev_id)
{
443
	return dev_id << RBD_SINGLE_MAJOR_PART_SHIFT;
444 445 446 447
}

static int minor_to_rbd_dev_id(int minor)
{
448
	return minor >> RBD_SINGLE_MAJOR_PART_SHIFT;
449 450
}

451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466
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;
}

467 468 469 470 471
static ssize_t rbd_supported_features_show(struct bus_type *bus, char *buf)
{
	return sprintf(buf, "0x%llx\n", RBD_FEATURES_SUPPORTED);
}

472 473 474 475 476
static BUS_ATTR(add, 0200, NULL, rbd_add);
static BUS_ATTR(remove, 0200, NULL, rbd_remove);
static BUS_ATTR(add_single_major, 0200, NULL, rbd_add_single_major);
static BUS_ATTR(remove_single_major, 0200, NULL, rbd_remove_single_major);
static BUS_ATTR(supported_features, 0444, rbd_supported_features_show, NULL);
477 478 479 480

static struct attribute *rbd_bus_attrs[] = {
	&bus_attr_add.attr,
	&bus_attr_remove.attr,
481 482
	&bus_attr_add_single_major.attr,
	&bus_attr_remove_single_major.attr,
483
	&bus_attr_supported_features.attr,
484
	NULL,
A
Alex Elder 已提交
485
};
486 487 488 489

static umode_t rbd_bus_is_visible(struct kobject *kobj,
				  struct attribute *attr, int index)
{
490 491 492 493 494
	if (!single_major &&
	    (attr == &bus_attr_add_single_major.attr ||
	     attr == &bus_attr_remove_single_major.attr))
		return 0;

495 496 497 498 499 500 501 502
	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
503 504 505

static struct bus_type rbd_bus_type = {
	.name		= "rbd",
506
	.bus_groups	= rbd_bus_groups,
A
Alex Elder 已提交
507 508 509 510 511 512 513 514 515 516 517
};

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
518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544
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
545 546 547 548 549 550 551 552 553 554 555 556
#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 */
557

558
static void rbd_dev_remove_parent(struct rbd_device *rbd_dev);
A
Alex Elder 已提交
559

A
Alex Elder 已提交
560
static int rbd_dev_refresh(struct rbd_device *rbd_dev);
561
static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev);
562
static int rbd_dev_header_info(struct rbd_device *rbd_dev);
563
static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev);
564 565
static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
					u64 snap_id);
566 567 568 569
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);
570

571 572
static int rbd_open(struct block_device *bdev, fmode_t mode)
{
A
Alex Elder 已提交
573
	struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
574
	bool removing = false;
575

576
	spin_lock_irq(&rbd_dev->lock);
577 578 579 580
	if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags))
		removing = true;
	else
		rbd_dev->open_count++;
581
	spin_unlock_irq(&rbd_dev->lock);
582 583 584
	if (removing)
		return -ENOENT;

A
Alex Elder 已提交
585
	(void) get_device(&rbd_dev->dev);
586

587 588 589
	return 0;
}

590
static void rbd_release(struct gendisk *disk, fmode_t mode)
591 592
{
	struct rbd_device *rbd_dev = disk->private_data;
593 594
	unsigned long open_count_before;

595
	spin_lock_irq(&rbd_dev->lock);
596
	open_count_before = rbd_dev->open_count--;
597
	spin_unlock_irq(&rbd_dev->lock);
598
	rbd_assert(open_count_before > 0);
599

A
Alex Elder 已提交
600
	put_device(&rbd_dev->dev);
601 602
}

G
Guangliang Zhao 已提交
603 604
static int rbd_ioctl_set_ro(struct rbd_device *rbd_dev, unsigned long arg)
{
605
	int ro;
G
Guangliang Zhao 已提交
606

607
	if (get_user(ro, (int __user *)arg))
G
Guangliang Zhao 已提交
608 609
		return -EFAULT;

610
	/* Snapshots can't be marked read-write */
G
Guangliang Zhao 已提交
611 612 613
	if (rbd_dev->spec->snap_id != CEPH_NOSNAP && !ro)
		return -EROFS;

614 615
	/* Let blkdev_roset() handle it */
	return -ENOTTY;
G
Guangliang Zhao 已提交
616 617 618 619 620 621
}

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;
622
	int ret;
G
Guangliang Zhao 已提交
623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642

	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 */

643 644 645
static const struct block_device_operations rbd_bd_ops = {
	.owner			= THIS_MODULE,
	.open			= rbd_open,
646
	.release		= rbd_release,
G
Guangliang Zhao 已提交
647 648 649 650
	.ioctl			= rbd_ioctl,
#ifdef CONFIG_COMPAT
	.compat_ioctl		= rbd_compat_ioctl,
#endif
651 652 653
};

/*
654
 * Initialize an rbd client instance.  Success or not, this function
655
 * consumes ceph_opts.  Caller holds client_mutex.
656
 */
657
static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts)
658 659 660 661
{
	struct rbd_client *rbdc;
	int ret = -ENOMEM;

A
Alex Elder 已提交
662
	dout("%s:\n", __func__);
663 664 665 666 667 668 669
	rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL);
	if (!rbdc)
		goto out_opt;

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

670
	rbdc->client = ceph_create_client(ceph_opts, rbdc);
671
	if (IS_ERR(rbdc->client))
672
		goto out_rbdc;
A
Alex Elder 已提交
673
	ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */
674 675 676

	ret = ceph_open_session(rbdc->client);
	if (ret < 0)
677
		goto out_client;
678

A
Alex Elder 已提交
679
	spin_lock(&rbd_client_list_lock);
680
	list_add_tail(&rbdc->node, &rbd_client_list);
A
Alex Elder 已提交
681
	spin_unlock(&rbd_client_list_lock);
682

A
Alex Elder 已提交
683
	dout("%s: rbdc %p\n", __func__, rbdc);
684

685
	return rbdc;
686
out_client:
687
	ceph_destroy_client(rbdc->client);
688
out_rbdc:
689 690
	kfree(rbdc);
out_opt:
A
Alex Elder 已提交
691 692
	if (ceph_opts)
		ceph_destroy_options(ceph_opts);
A
Alex Elder 已提交
693 694
	dout("%s: error %d\n", __func__, ret);

V
Vasiliy Kulikov 已提交
695
	return ERR_PTR(ret);
696 697
}

698 699 700 701 702 703 704
static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc)
{
	kref_get(&rbdc->kref);

	return rbdc;
}

705
/*
706 707
 * Find a ceph client with specific addr and configuration.  If
 * found, bump its reference count.
708
 */
709
static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts)
710 711
{
	struct rbd_client *client_node;
712
	bool found = false;
713

A
Alex Elder 已提交
714
	if (ceph_opts->flags & CEPH_OPT_NOSHARE)
715 716
		return NULL;

717 718 719
	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)) {
720 721
			__rbd_get_client(client_node);

722 723 724 725 726 727 728
			found = true;
			break;
		}
	}
	spin_unlock(&rbd_client_list_lock);

	return found ? client_node : NULL;
729 730
}

731
/*
732
 * (Per device) rbd map options
733 734
 */
enum {
I
Ilya Dryomov 已提交
735
	Opt_queue_depth,
736
	Opt_lock_timeout,
737 738
	Opt_last_int,
	/* int args above */
739
	Opt_pool_ns,
740 741
	Opt_last_string,
	/* string args above */
A
Alex Elder 已提交
742 743
	Opt_read_only,
	Opt_read_write,
I
Ilya Dryomov 已提交
744
	Opt_lock_on_read,
I
Ilya Dryomov 已提交
745
	Opt_exclusive,
I
Ilya Dryomov 已提交
746
	Opt_notrim,
747
	Opt_err
748 749
};

A
Alex Elder 已提交
750
static match_table_t rbd_opts_tokens = {
I
Ilya Dryomov 已提交
751
	{Opt_queue_depth, "queue_depth=%d"},
752
	{Opt_lock_timeout, "lock_timeout=%d"},
753
	/* int args above */
754
	{Opt_pool_ns, "_pool_ns=%s"},
755
	/* string args above */
A
Alex Elder 已提交
756
	{Opt_read_only, "read_only"},
A
Alex Elder 已提交
757 758 759
	{Opt_read_only, "ro"},		/* Alternate spelling */
	{Opt_read_write, "read_write"},
	{Opt_read_write, "rw"},		/* Alternate spelling */
I
Ilya Dryomov 已提交
760
	{Opt_lock_on_read, "lock_on_read"},
I
Ilya Dryomov 已提交
761
	{Opt_exclusive, "exclusive"},
I
Ilya Dryomov 已提交
762
	{Opt_notrim, "notrim"},
763
	{Opt_err, NULL}
764 765
};

A
Alex Elder 已提交
766
struct rbd_options {
I
Ilya Dryomov 已提交
767
	int	queue_depth;
768
	unsigned long	lock_timeout;
A
Alex Elder 已提交
769
	bool	read_only;
I
Ilya Dryomov 已提交
770
	bool	lock_on_read;
I
Ilya Dryomov 已提交
771
	bool	exclusive;
I
Ilya Dryomov 已提交
772
	bool	trim;
A
Alex Elder 已提交
773 774
};

I
Ilya Dryomov 已提交
775
#define RBD_QUEUE_DEPTH_DEFAULT	BLKDEV_MAX_RQ
776
#define RBD_LOCK_TIMEOUT_DEFAULT 0  /* no timeout */
A
Alex Elder 已提交
777
#define RBD_READ_ONLY_DEFAULT	false
I
Ilya Dryomov 已提交
778
#define RBD_LOCK_ON_READ_DEFAULT false
I
Ilya Dryomov 已提交
779
#define RBD_EXCLUSIVE_DEFAULT	false
I
Ilya Dryomov 已提交
780
#define RBD_TRIM_DEFAULT	true
A
Alex Elder 已提交
781

782 783 784 785 786
struct parse_rbd_opts_ctx {
	struct rbd_spec		*spec;
	struct rbd_options	*opts;
};

787 788
static int parse_rbd_opts_token(char *c, void *private)
{
789
	struct parse_rbd_opts_ctx *pctx = private;
790 791 792
	substring_t argstr[MAX_OPT_ARGS];
	int token, intval, ret;

A
Alex Elder 已提交
793
	token = match_token(c, rbd_opts_tokens, argstr);
794 795 796
	if (token < Opt_last_int) {
		ret = match_int(&argstr[0], &intval);
		if (ret < 0) {
797
			pr_err("bad option arg (not int) at '%s'\n", c);
798 799 800 801
			return ret;
		}
		dout("got int token %d val %d\n", token, intval);
	} else if (token > Opt_last_int && token < Opt_last_string) {
802
		dout("got string token %d val %s\n", token, argstr[0].from);
803 804 805 806 807
	} else {
		dout("got token %d\n", token);
	}

	switch (token) {
I
Ilya Dryomov 已提交
808 809 810 811 812
	case Opt_queue_depth:
		if (intval < 1) {
			pr_err("queue_depth out of range\n");
			return -EINVAL;
		}
813
		pctx->opts->queue_depth = intval;
I
Ilya Dryomov 已提交
814
		break;
815 816 817 818 819 820
	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;
		}
821
		pctx->opts->lock_timeout = msecs_to_jiffies(intval * 1000);
822
		break;
823 824 825 826 827
	case Opt_pool_ns:
		kfree(pctx->spec->pool_ns);
		pctx->spec->pool_ns = match_strdup(argstr);
		if (!pctx->spec->pool_ns)
			return -ENOMEM;
828
		break;
A
Alex Elder 已提交
829
	case Opt_read_only:
830
		pctx->opts->read_only = true;
A
Alex Elder 已提交
831 832
		break;
	case Opt_read_write:
833
		pctx->opts->read_only = false;
A
Alex Elder 已提交
834
		break;
I
Ilya Dryomov 已提交
835
	case Opt_lock_on_read:
836
		pctx->opts->lock_on_read = true;
I
Ilya Dryomov 已提交
837
		break;
I
Ilya Dryomov 已提交
838
	case Opt_exclusive:
839
		pctx->opts->exclusive = true;
I
Ilya Dryomov 已提交
840
		break;
I
Ilya Dryomov 已提交
841
	case Opt_notrim:
842
		pctx->opts->trim = false;
I
Ilya Dryomov 已提交
843
		break;
844
	default:
845 846
		/* libceph prints "bad option" msg */
		return -EINVAL;
847
	}
848

849 850 851
	return 0;
}

G
Guangliang Zhao 已提交
852 853 854 855 856 857 858
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";
859 860
	case OBJ_OP_DISCARD:
		return "discard";
G
Guangliang Zhao 已提交
861 862 863 864 865
	default:
		return "???";
	}
}

866 867
/*
 * Destroy ceph client
A
Alex Elder 已提交
868
 *
A
Alex Elder 已提交
869
 * Caller must hold rbd_client_list_lock.
870 871 872 873 874
 */
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
875
	dout("%s: rbdc %p\n", __func__, rbdc);
876
	spin_lock(&rbd_client_list_lock);
877
	list_del(&rbdc->node);
878
	spin_unlock(&rbd_client_list_lock);
879 880 881 882 883 884 885 886 887

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

/*
 * Drop reference to ceph client node. If it's not referenced anymore, release
 * it.
 */
888
static void rbd_put_client(struct rbd_client *rbdc)
889
{
890 891
	if (rbdc)
		kref_put(&rbdc->kref, rbd_client_release);
892 893
}

894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910
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);
}

911 912 913 914 915 916 917 918
/*
 * 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;
919
	int ret;
920 921 922

	mutex_lock_nested(&client_mutex, SINGLE_DEPTH_NESTING);
	rbdc = rbd_client_find(ceph_opts);
923
	if (rbdc) {
924
		ceph_destroy_options(ceph_opts);
925 926 927 928 929 930 931 932 933 934 935 936

		/*
		 * 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 {
937
		rbdc = rbd_client_create(ceph_opts);
938
	}
939 940 941 942 943
	mutex_unlock(&client_mutex);

	return rbdc;
}

944 945 946 947 948
static bool rbd_image_format_valid(u32 image_format)
{
	return image_format == 1 || image_format == 2;
}

949 950
static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk)
{
951 952 953 954 955 956 957
	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
958 959 960 961 962 963 964 965 966 967
	/* 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;

968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985
	/*
	 * 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;
986 987
}

I
Ilya Dryomov 已提交
988 989 990 991 992 993 994 995
/*
 * 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
996 997 998 999 1000 1001 1002 1003 1004 1005 1006
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);
1007 1008
	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
1009 1010 1011
	RCU_INIT_POINTER(rbd_dev->layout.pool_ns, NULL);
}

1012
/*
1013 1014
 * Fill an rbd image header with information from the given format 1
 * on-disk header.
1015
 */
A
Alex Elder 已提交
1016
static int rbd_header_from_disk(struct rbd_device *rbd_dev,
1017
				 struct rbd_image_header_ondisk *ondisk)
1018
{
A
Alex Elder 已提交
1019
	struct rbd_image_header *header = &rbd_dev->header;
1020 1021 1022 1023 1024
	bool first_time = header->object_prefix == NULL;
	struct ceph_snap_context *snapc;
	char *object_prefix = NULL;
	char *snap_names = NULL;
	u64 *snap_sizes = NULL;
1025
	u32 snap_count;
1026
	int ret = -ENOMEM;
1027
	u32 i;
1028

1029
	/* Allocate this now to avoid having to handle failure below */
A
Alex Elder 已提交
1030

1031
	if (first_time) {
1032 1033 1034
		object_prefix = kstrndup(ondisk->object_prefix,
					 sizeof(ondisk->object_prefix),
					 GFP_KERNEL);
1035 1036 1037
		if (!object_prefix)
			return -ENOMEM;
	}
A
Alex Elder 已提交
1038

1039
	/* Allocate the snapshot context and fill it in */
A
Alex Elder 已提交
1040

1041 1042 1043 1044 1045
	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);
1046
	if (snap_count) {
1047
		struct rbd_image_snap_ondisk *snaps;
A
Alex Elder 已提交
1048 1049
		u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len);

1050
		/* We'll keep a copy of the snapshot names... */
1051

1052 1053 1054 1055
		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
1056 1057
			goto out_err;

1058
		/* ...as well as the array of their sizes. */
1059 1060 1061
		snap_sizes = kmalloc_array(snap_count,
					   sizeof(*header->snap_sizes),
					   GFP_KERNEL);
1062
		if (!snap_sizes)
A
Alex Elder 已提交
1063
			goto out_err;
1064

A
Alex Elder 已提交
1065
		/*
1066 1067 1068
		 * Copy the names, and fill in each snapshot's id
		 * and size.
		 *
1069
		 * Note that rbd_dev_v1_header_info() guarantees the
1070
		 * ondisk buffer we're working with has
A
Alex Elder 已提交
1071 1072 1073
		 * snap_names_len bytes beyond the end of the
		 * snapshot id array, this memcpy() is safe.
		 */
1074 1075 1076 1077 1078 1079
		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);
		}
1080
	}
A
Alex Elder 已提交
1081

1082
	/* We won't fail any more, fill in the header */
1083

1084 1085 1086
	if (first_time) {
		header->object_prefix = object_prefix;
		header->obj_order = ondisk->options.order;
I
Ilya Dryomov 已提交
1087
		rbd_init_layout(rbd_dev);
1088
	} else {
A
Alex Elder 已提交
1089 1090 1091
		ceph_put_snap_context(header->snapc);
		kfree(header->snap_names);
		kfree(header->snap_sizes);
1092
	}
1093

1094
	/* The remaining fields always get updated (when we refresh) */
1095

A
Alex Elder 已提交
1096
	header->image_size = le64_to_cpu(ondisk->image_size);
1097 1098 1099
	header->snapc = snapc;
	header->snap_names = snap_names;
	header->snap_sizes = snap_sizes;
1100

1101
	return 0;
1102 1103
out_2big:
	ret = -EIO;
A
Alex Elder 已提交
1104
out_err:
1105 1106 1107 1108
	kfree(snap_sizes);
	kfree(snap_names);
	ceph_put_snap_context(snapc);
	kfree(object_prefix);
1109

1110
	return ret;
1111 1112
}

1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127
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);
}

1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151
/*
 * 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.
 */
1152 1153 1154
static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id)
{
	struct ceph_snap_context *snapc = rbd_dev->header.snapc;
1155
	u64 *found;
1156

1157 1158
	found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps,
				sizeof (snap_id), snapid_compare_reverse);
1159

1160
	return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX;
1161 1162
}

1163 1164
static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev,
					u64 snap_id)
1165
{
1166
	u32 which;
1167
	const char *snap_name;
1168

1169 1170
	which = rbd_dev_snap_index(rbd_dev, snap_id);
	if (which == BAD_SNAP_INDEX)
1171
		return ERR_PTR(-ENOENT);
1172

1173 1174
	snap_name = _rbd_dev_v1_snap_name(rbd_dev, which);
	return snap_name ? snap_name : ERR_PTR(-ENOMEM);
1175 1176 1177 1178
}

static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id)
{
1179 1180 1181
	if (snap_id == CEPH_NOSNAP)
		return RBD_SNAP_HEAD_NAME;

1182 1183 1184
	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);
1185

1186
	return rbd_dev_v2_snap_name(rbd_dev, snap_id);
1187 1188
}

1189 1190
static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
				u64 *snap_size)
1191
{
1192 1193 1194 1195 1196
	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;
1197

1198 1199 1200
		which = rbd_dev_snap_index(rbd_dev, snap_id);
		if (which == BAD_SNAP_INDEX)
			return -ENOENT;
1201

1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213
		*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;
1214 1215
}

1216 1217
static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
			u64 *snap_features)
1218
{
1219 1220 1221 1222 1223
	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 */
1224
	} else {
1225 1226
		u64 features = 0;
		int ret;
1227

1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238
		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)
{
1239
	u64 snap_id = rbd_dev->spec->snap_id;
1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253
	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;

1254
	return 0;
1255 1256
}

A
Alex Elder 已提交
1257 1258 1259 1260
static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev)
{
	rbd_dev->mapping.size = 0;
	rbd_dev->mapping.features = 0;
1261 1262
}

1263
static void zero_bvec(struct bio_vec *bv)
1264 1265
{
	void *buf;
1266
	unsigned long flags;
1267

1268 1269 1270 1271
	buf = bvec_kmap_irq(bv, &flags);
	memset(buf, 0, bv->bv_len);
	flush_dcache_page(bv->bv_page);
	bvec_kunmap_irq(buf, &flags);
1272 1273
}

1274
static void zero_bios(struct ceph_bio_iter *bio_pos, u32 off, u32 bytes)
A
Alex Elder 已提交
1275
{
1276
	struct ceph_bio_iter it = *bio_pos;
A
Alex Elder 已提交
1277

1278 1279 1280 1281
	ceph_bio_iter_advance(&it, off);
	ceph_bio_iter_advance_step(&it, bytes, ({
		zero_bvec(&bv);
	}));
A
Alex Elder 已提交
1282 1283
}

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

1288 1289 1290 1291
	ceph_bvec_iter_advance(&it, off);
	ceph_bvec_iter_advance_step(&it, bytes, ({
		zero_bvec(&bv);
	}));
A
Alex Elder 已提交
1292 1293 1294
}

/*
I
Ilya Dryomov 已提交
1295
 * Zero a range in @obj_req data buffer defined by a bio (list) or
1296
 * (private) bio_vec array.
A
Alex Elder 已提交
1297
 *
I
Ilya Dryomov 已提交
1298
 * @off is relative to the start of the data buffer.
1299
 */
I
Ilya Dryomov 已提交
1300 1301
static void rbd_obj_zero_range(struct rbd_obj_request *obj_req, u32 off,
			       u32 bytes)
1302
{
1303
	switch (obj_req->img_request->data_type) {
I
Ilya Dryomov 已提交
1304 1305 1306 1307
	case OBJ_REQUEST_BIO:
		zero_bios(&obj_req->bio_pos, off, bytes);
		break;
	case OBJ_REQUEST_BVECS:
1308
	case OBJ_REQUEST_OWN_BVECS:
I
Ilya Dryomov 已提交
1309 1310 1311 1312
		zero_bvecs(&obj_req->bvec_pos, off, bytes);
		break;
	default:
		rbd_assert(0);
1313 1314 1315
	}
}

A
Alex Elder 已提交
1316 1317 1318 1319
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
1320
	dout("%s: obj %p (was %d)\n", __func__, obj_request,
1321
		kref_read(&obj_request->kref));
A
Alex Elder 已提交
1322 1323 1324
	kref_put(&obj_request->kref, rbd_obj_request_destroy);
}

1325 1326 1327
static void rbd_img_request_get(struct rbd_img_request *img_request)
{
	dout("%s: img %p (was %d)\n", __func__, img_request,
1328
	     kref_read(&img_request->kref));
1329 1330 1331
	kref_get(&img_request->kref);
}

A
Alex Elder 已提交
1332 1333 1334 1335
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
1336
	dout("%s: img %p (was %d)\n", __func__, img_request,
1337
		kref_read(&img_request->kref));
1338
	kref_put(&img_request->kref, rbd_img_request_destroy);
A
Alex Elder 已提交
1339 1340 1341 1342 1343
}

static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request,
					struct rbd_obj_request *obj_request)
{
1344 1345
	rbd_assert(obj_request->img_request == NULL);

1346
	/* Image request now owns object's original reference */
A
Alex Elder 已提交
1347
	obj_request->img_request = img_request;
1348
	img_request->obj_request_count++;
I
Ilya Dryomov 已提交
1349
	img_request->pending_count++;
1350
	dout("%s: img %p obj %p\n", __func__, img_request, obj_request);
A
Alex Elder 已提交
1351 1352 1353 1354 1355
}

static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request,
					struct rbd_obj_request *obj_request)
{
1356
	dout("%s: img %p obj %p\n", __func__, img_request, obj_request);
1357
	list_del(&obj_request->ex.oe_item);
1358 1359
	rbd_assert(img_request->obj_request_count > 0);
	img_request->obj_request_count--;
A
Alex Elder 已提交
1360 1361 1362 1363
	rbd_assert(obj_request->img_request == img_request);
	rbd_obj_request_put(obj_request);
}

1364
static void rbd_obj_request_submit(struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1365
{
1366 1367
	struct ceph_osd_request *osd_req = obj_request->osd_req;

1368
	dout("%s %p object_no %016llx %llu~%llu osd_req %p\n", __func__,
1369 1370
	     obj_request, obj_request->ex.oe_objno, obj_request->ex.oe_off,
	     obj_request->ex.oe_len, osd_req);
1371
	ceph_osdc_start_request(osd_req->r_osdc, osd_req, false);
A
Alex Elder 已提交
1372 1373
}

A
Alex Elder 已提交
1374 1375 1376 1377 1378
/*
 * 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.
 */
1379 1380 1381 1382 1383 1384
static void img_request_layered_set(struct rbd_img_request *img_request)
{
	set_bit(IMG_REQ_LAYERED, &img_request->flags);
	smp_mb();
}

1385 1386 1387 1388 1389 1390
static void img_request_layered_clear(struct rbd_img_request *img_request)
{
	clear_bit(IMG_REQ_LAYERED, &img_request->flags);
	smp_mb();
}

1391 1392 1393 1394 1395 1396
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
1397
static bool rbd_obj_is_entire(struct rbd_obj_request *obj_req)
1398
{
I
Ilya Dryomov 已提交
1399
	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
A
Alex Elder 已提交
1400

1401 1402
	return !obj_req->ex.oe_off &&
	       obj_req->ex.oe_len == rbd_dev->layout.object_size;
1403 1404
}

I
Ilya Dryomov 已提交
1405
static bool rbd_obj_is_tail(struct rbd_obj_request *obj_req)
A
Alex Elder 已提交
1406
{
I
Ilya Dryomov 已提交
1407
	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
A
Alex Elder 已提交
1408

1409
	return obj_req->ex.oe_off + obj_req->ex.oe_len ==
I
Ilya Dryomov 已提交
1410
					rbd_dev->layout.object_size;
1411 1412
}

I
Ilya Dryomov 已提交
1413
static u64 rbd_obj_img_extents_bytes(struct rbd_obj_request *obj_req)
A
Alex Elder 已提交
1414
{
I
Ilya Dryomov 已提交
1415 1416
	return ceph_file_extents_bytes(obj_req->img_extents,
				       obj_req->num_img_extents);
A
Alex Elder 已提交
1417 1418
}

I
Ilya Dryomov 已提交
1419
static bool rbd_img_is_write(struct rbd_img_request *img_req)
A
Alex Elder 已提交
1420
{
I
Ilya Dryomov 已提交
1421
	switch (img_req->op_type) {
I
Ilya Dryomov 已提交
1422 1423 1424 1425 1426 1427
	case OBJ_OP_READ:
		return false;
	case OBJ_OP_WRITE:
	case OBJ_OP_DISCARD:
		return true;
	default:
A
Arnd Bergmann 已提交
1428
		BUG();
I
Ilya Dryomov 已提交
1429
	}
1430 1431
}

I
Ilya Dryomov 已提交
1432
static void rbd_obj_handle_request(struct rbd_obj_request *obj_req);
I
Ilya Dryomov 已提交
1433

1434
static void rbd_osd_req_callback(struct ceph_osd_request *osd_req)
A
Alex Elder 已提交
1435
{
I
Ilya Dryomov 已提交
1436
	struct rbd_obj_request *obj_req = osd_req->r_priv;
A
Alex Elder 已提交
1437

I
Ilya Dryomov 已提交
1438 1439 1440
	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
1441

I
Ilya Dryomov 已提交
1442 1443 1444 1445 1446 1447 1448 1449 1450 1451
	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
1452

I
Ilya Dryomov 已提交
1453
	rbd_obj_handle_request(obj_req);
A
Alex Elder 已提交
1454 1455
}

1456
static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1457
{
1458
	struct ceph_osd_request *osd_req = obj_request->osd_req;
A
Alex Elder 已提交
1459

1460
	osd_req->r_flags = CEPH_OSD_FLAG_READ;
1461
	osd_req->r_snapid = obj_request->img_request->snap_id;
1462 1463 1464 1465 1466 1467
}

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

1468
	osd_req->r_flags = CEPH_OSD_FLAG_WRITE;
1469
	ktime_get_real_ts64(&osd_req->r_mtime);
1470
	osd_req->r_data_offset = obj_request->ex.oe_off;
A
Alex Elder 已提交
1471 1472
}

1473
static struct ceph_osd_request *
1474
rbd_osd_req_create(struct rbd_obj_request *obj_req, unsigned int num_ops)
1475
{
1476 1477
	struct rbd_img_request *img_req = obj_req->img_request;
	struct rbd_device *rbd_dev = img_req->rbd_dev;
1478 1479
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
	struct ceph_osd_request *req;
1480 1481
	const char *name_format = rbd_dev->image_format == 1 ?
				      RBD_V1_DATA_FORMAT : RBD_V2_DATA_FORMAT;
1482

1483 1484 1485
	req = ceph_osdc_alloc_request(osdc,
			(rbd_img_is_write(img_req) ? img_req->snapc : NULL),
			num_ops, false, GFP_NOIO);
1486 1487 1488 1489
	if (!req)
		return NULL;

	req->r_callback = rbd_osd_req_callback;
1490
	req->r_priv = obj_req;
1491

1492 1493 1494 1495 1496
	/*
	 * Data objects may be stored in a separate pool, but always in
	 * the same namespace in that pool as the header in its pool.
	 */
	ceph_oloc_copy(&req->r_base_oloc, &rbd_dev->header_oloc);
1497
	req->r_base_oloc.pool = rbd_dev->layout.pool_id;
1498

1499
	if (ceph_oid_aprintf(&req->r_base_oid, GFP_NOIO, name_format,
1500
			rbd_dev->header.object_prefix, obj_req->ex.oe_objno))
1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512
		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
1513 1514 1515 1516 1517
static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req)
{
	ceph_osdc_put_request(osd_req);
}

1518
static struct rbd_obj_request *rbd_obj_request_create(void)
A
Alex Elder 已提交
1519 1520 1521
{
	struct rbd_obj_request *obj_request;

1522
	obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_NOIO);
1523
	if (!obj_request)
1524 1525
		return NULL;

1526
	ceph_object_extent_init(&obj_request->ex);
A
Alex Elder 已提交
1527 1528
	kref_init(&obj_request->kref);

1529
	dout("%s %p\n", __func__, obj_request);
A
Alex Elder 已提交
1530 1531 1532 1533 1534 1535
	return obj_request;
}

static void rbd_obj_request_destroy(struct kref *kref)
{
	struct rbd_obj_request *obj_request;
1536
	u32 i;
A
Alex Elder 已提交
1537 1538 1539

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

A
Alex Elder 已提交
1540 1541
	dout("%s: obj %p\n", __func__, obj_request);

A
Alex Elder 已提交
1542 1543 1544
	if (obj_request->osd_req)
		rbd_osd_req_destroy(obj_request->osd_req);

1545
	switch (obj_request->img_request->data_type) {
1546
	case OBJ_REQUEST_NODATA:
A
Alex Elder 已提交
1547
	case OBJ_REQUEST_BIO:
1548
	case OBJ_REQUEST_BVECS:
1549
		break;		/* Nothing to do */
1550 1551
	case OBJ_REQUEST_OWN_BVECS:
		kfree(obj_request->bvec_pos.bvecs);
1552
		break;
1553 1554
	default:
		rbd_assert(0);
A
Alex Elder 已提交
1555 1556
	}

I
Ilya Dryomov 已提交
1557
	kfree(obj_request->img_extents);
1558 1559 1560 1561 1562 1563
	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
1564 1565
	}

1566
	kmem_cache_free(rbd_obj_request_cache, obj_request);
A
Alex Elder 已提交
1567 1568
}

A
Alex Elder 已提交
1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579
/* 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;
}

1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601
/*
 * 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
1602
		rbd_warn(rbd_dev, "parent reference underflow");
1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614
}

/*
 * 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)
{
1615
	int counter = 0;
1616 1617 1618 1619

	if (!rbd_dev->parent_spec)
		return false;

1620 1621 1622 1623
	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);
1624 1625

	if (counter < 0)
1626
		rbd_warn(rbd_dev, "parent reference overflow");
1627

1628
	return counter > 0;
1629 1630
}

A
Alex Elder 已提交
1631 1632 1633 1634 1635
/*
 * 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
1636 1637
static struct rbd_img_request *rbd_img_request_create(
					struct rbd_device *rbd_dev,
G
Guangliang Zhao 已提交
1638
					enum obj_operation_type op_type,
1639
					struct ceph_snap_context *snapc)
A
Alex Elder 已提交
1640 1641 1642
{
	struct rbd_img_request *img_request;

1643
	img_request = kmem_cache_zalloc(rbd_img_request_cache, GFP_NOIO);
A
Alex Elder 已提交
1644 1645 1646 1647
	if (!img_request)
		return NULL;

	img_request->rbd_dev = rbd_dev;
I
Ilya Dryomov 已提交
1648 1649
	img_request->op_type = op_type;
	if (!rbd_img_is_write(img_request))
A
Alex Elder 已提交
1650
		img_request->snap_id = rbd_dev->spec->snap_id;
I
Ilya Dryomov 已提交
1651 1652 1653
	else
		img_request->snapc = snapc;

1654
	if (rbd_dev_parent_get(rbd_dev))
1655
		img_request_layered_set(img_request);
1656

A
Alex Elder 已提交
1657
	spin_lock_init(&img_request->completion_lock);
1658
	INIT_LIST_HEAD(&img_request->object_extents);
A
Alex Elder 已提交
1659 1660
	kref_init(&img_request->kref);

1661 1662
	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
1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673
	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
1674 1675
	dout("%s: img %p\n", __func__, img_request);

A
Alex Elder 已提交
1676 1677
	for_each_obj_request_safe(img_request, obj_request, next_obj_request)
		rbd_img_obj_request_del(img_request, obj_request);
1678
	rbd_assert(img_request->obj_request_count == 0);
A
Alex Elder 已提交
1679

1680 1681 1682 1683 1684
	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
1685
	if (rbd_img_is_write(img_request))
1686
		ceph_put_snap_context(img_request->snapc);
A
Alex Elder 已提交
1687

1688
	kmem_cache_free(rbd_img_request_cache, img_request);
A
Alex Elder 已提交
1689 1690
}

I
Ilya Dryomov 已提交
1691 1692
static void prune_extents(struct ceph_file_extent *img_extents,
			  u32 *num_img_extents, u64 overlap)
1693
{
I
Ilya Dryomov 已提交
1694
	u32 cnt = *num_img_extents;
1695

I
Ilya Dryomov 已提交
1696 1697 1698
	/* drop extents completely beyond the overlap */
	while (cnt && img_extents[cnt - 1].fe_off >= overlap)
		cnt--;
1699

I
Ilya Dryomov 已提交
1700 1701
	if (cnt) {
		struct ceph_file_extent *ex = &img_extents[cnt - 1];
1702

I
Ilya Dryomov 已提交
1703 1704 1705 1706
		/* trim final overlapping extent */
		if (ex->fe_off + ex->fe_len > overlap)
			ex->fe_len = overlap - ex->fe_off;
	}
1707

I
Ilya Dryomov 已提交
1708
	*num_img_extents = cnt;
1709 1710
}

I
Ilya Dryomov 已提交
1711 1712 1713 1714 1715 1716
/*
 * 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)
1717
{
I
Ilya Dryomov 已提交
1718 1719
	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
	int ret;
1720

I
Ilya Dryomov 已提交
1721 1722
	if (!rbd_dev->parent_overlap)
		return 0;
1723

I
Ilya Dryomov 已提交
1724 1725 1726 1727 1728 1729 1730 1731
	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;
1732

I
Ilya Dryomov 已提交
1733 1734 1735
	prune_extents(obj_req->img_extents, &obj_req->num_img_extents,
		      rbd_dev->parent_overlap);
	return 0;
1736 1737
}

I
Ilya Dryomov 已提交
1738
static void rbd_osd_req_setup_data(struct rbd_obj_request *obj_req, u32 which)
1739
{
1740
	switch (obj_req->img_request->data_type) {
I
Ilya Dryomov 已提交
1741 1742 1743
	case OBJ_REQUEST_BIO:
		osd_req_op_extent_osd_data_bio(obj_req->osd_req, which,
					       &obj_req->bio_pos,
1744
					       obj_req->ex.oe_len);
I
Ilya Dryomov 已提交
1745 1746
		break;
	case OBJ_REQUEST_BVECS:
1747
	case OBJ_REQUEST_OWN_BVECS:
I
Ilya Dryomov 已提交
1748
		rbd_assert(obj_req->bvec_pos.iter.bi_size ==
1749
							obj_req->ex.oe_len);
1750
		rbd_assert(obj_req->bvec_idx == obj_req->bvec_count);
I
Ilya Dryomov 已提交
1751 1752 1753 1754 1755
		osd_req_op_extent_osd_data_bvec_pos(obj_req->osd_req, which,
						    &obj_req->bvec_pos);
		break;
	default:
		rbd_assert(0);
1756
	}
I
Ilya Dryomov 已提交
1757
}
1758

I
Ilya Dryomov 已提交
1759 1760
static int rbd_obj_setup_read(struct rbd_obj_request *obj_req)
{
1761
	obj_req->osd_req = rbd_osd_req_create(obj_req, 1);
I
Ilya Dryomov 已提交
1762 1763
	if (!obj_req->osd_req)
		return -ENOMEM;
1764

I
Ilya Dryomov 已提交
1765
	osd_req_op_extent_init(obj_req->osd_req, 0, CEPH_OSD_OP_READ,
1766
			       obj_req->ex.oe_off, obj_req->ex.oe_len, 0, 0);
I
Ilya Dryomov 已提交
1767
	rbd_osd_req_setup_data(obj_req, 0);
C
Christoph Hellwig 已提交
1768

I
Ilya Dryomov 已提交
1769 1770 1771 1772 1773 1774 1775 1776
	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
1777

I
Ilya Dryomov 已提交
1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794
	/*
	 * 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;
1795 1796
}

I
Ilya Dryomov 已提交
1797 1798
static void __rbd_obj_setup_write(struct rbd_obj_request *obj_req,
				  unsigned int which)
1799
{
I
Ilya Dryomov 已提交
1800 1801
	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
	u16 opcode;
1802

I
Ilya Dryomov 已提交
1803 1804 1805
	osd_req_op_alloc_hint_init(obj_req->osd_req, which++,
				   rbd_dev->layout.object_size,
				   rbd_dev->layout.object_size);
1806

I
Ilya Dryomov 已提交
1807 1808 1809 1810
	if (rbd_obj_is_entire(obj_req))
		opcode = CEPH_OSD_OP_WRITEFULL;
	else
		opcode = CEPH_OSD_OP_WRITE;
1811

I
Ilya Dryomov 已提交
1812
	osd_req_op_extent_init(obj_req->osd_req, which, opcode,
1813
			       obj_req->ex.oe_off, obj_req->ex.oe_len, 0, 0);
I
Ilya Dryomov 已提交
1814
	rbd_osd_req_setup_data(obj_req, which++);
1815

I
Ilya Dryomov 已提交
1816 1817 1818
	rbd_assert(which == obj_req->osd_req->r_num_ops);
	rbd_osd_req_format_write(obj_req);
}
1819

I
Ilya Dryomov 已提交
1820 1821 1822 1823 1824
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
1825 1826 1827 1828 1829 1830
	/* 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
1831 1832 1833 1834 1835
		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 */
1836 1837
	}

1838
	obj_req->osd_req = rbd_osd_req_create(obj_req, num_osd_ops);
I
Ilya Dryomov 已提交
1839 1840
	if (!obj_req->osd_req)
		return -ENOMEM;
1841

I
Ilya Dryomov 已提交
1842
	if (obj_req->num_img_extents) {
I
Ilya Dryomov 已提交
1843 1844 1845 1846 1847 1848 1849
		ret = __rbd_obj_setup_stat(obj_req, which++);
		if (ret)
			return ret;
	}

	__rbd_obj_setup_write(obj_req, which);
	return 0;
1850 1851
}

I
Ilya Dryomov 已提交
1852 1853 1854
static void __rbd_obj_setup_discard(struct rbd_obj_request *obj_req,
				    unsigned int which)
{
1855 1856
	u16 opcode;

I
Ilya Dryomov 已提交
1857
	if (rbd_obj_is_entire(obj_req)) {
I
Ilya Dryomov 已提交
1858
		if (obj_req->num_img_extents) {
1859 1860
			osd_req_op_init(obj_req->osd_req, which++,
					CEPH_OSD_OP_CREATE, 0);
1861 1862
			opcode = CEPH_OSD_OP_TRUNCATE;
		} else {
I
Ilya Dryomov 已提交
1863 1864 1865
			osd_req_op_init(obj_req->osd_req, which++,
					CEPH_OSD_OP_DELETE, 0);
			opcode = 0;
1866
		}
I
Ilya Dryomov 已提交
1867 1868
	} else if (rbd_obj_is_tail(obj_req)) {
		opcode = CEPH_OSD_OP_TRUNCATE;
1869
	} else {
I
Ilya Dryomov 已提交
1870
		opcode = CEPH_OSD_OP_ZERO;
1871 1872
	}

I
Ilya Dryomov 已提交
1873 1874
	if (opcode)
		osd_req_op_extent_init(obj_req->osd_req, which++, opcode,
1875
				       obj_req->ex.oe_off, obj_req->ex.oe_len,
I
Ilya Dryomov 已提交
1876 1877 1878 1879
				       0, 0);

	rbd_assert(which == obj_req->osd_req->r_num_ops);
	rbd_osd_req_format_write(obj_req);
1880 1881
}

I
Ilya Dryomov 已提交
1882
static int rbd_obj_setup_discard(struct rbd_obj_request *obj_req)
A
Alex Elder 已提交
1883
{
I
Ilya Dryomov 已提交
1884 1885
	unsigned int num_osd_ops, which = 0;
	int ret;
A
Alex Elder 已提交
1886

I
Ilya Dryomov 已提交
1887 1888 1889 1890
	/* reverse map the entire object onto the parent */
	ret = rbd_obj_calc_img_extents(obj_req, true);
	if (ret)
		return ret;
1891

I
Ilya Dryomov 已提交
1892 1893
	if (rbd_obj_is_entire(obj_req)) {
		obj_req->write_state = RBD_OBJ_WRITE_FLAT;
1894 1895 1896 1897
		if (obj_req->num_img_extents)
			num_osd_ops = 2; /* create + truncate */
		else
			num_osd_ops = 1; /* delete */
I
Ilya Dryomov 已提交
1898
	} else {
I
Ilya Dryomov 已提交
1899
		if (obj_req->num_img_extents) {
I
Ilya Dryomov 已提交
1900 1901 1902 1903 1904 1905
			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 */
		}
1906 1907
	}

1908
	obj_req->osd_req = rbd_osd_req_create(obj_req, num_osd_ops);
I
Ilya Dryomov 已提交
1909 1910
	if (!obj_req->osd_req)
		return -ENOMEM;
A
Alex Elder 已提交
1911

I
Ilya Dryomov 已提交
1912
	if (!rbd_obj_is_entire(obj_req) && obj_req->num_img_extents) {
I
Ilya Dryomov 已提交
1913 1914 1915 1916
		ret = __rbd_obj_setup_stat(obj_req, which++);
		if (ret)
			return ret;
	}
1917

I
Ilya Dryomov 已提交
1918 1919 1920
	__rbd_obj_setup_discard(obj_req, which);
	return 0;
}
1921

I
Ilya Dryomov 已提交
1922 1923 1924 1925 1926 1927 1928 1929 1930
/*
 * 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
1931

I
Ilya Dryomov 已提交
1932
	for_each_obj_request(img_req, obj_req) {
I
Ilya Dryomov 已提交
1933
		switch (img_req->op_type) {
I
Ilya Dryomov 已提交
1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947
		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
1948 1949 1950
	}

	return 0;
I
Ilya Dryomov 已提交
1951
}
A
Alex Elder 已提交
1952

1953 1954 1955 1956
union rbd_img_fill_iter {
	struct ceph_bio_iter	bio_iter;
	struct ceph_bvec_iter	bvec_iter;
};
A
Alex Elder 已提交
1957

1958 1959 1960 1961 1962
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;
1963 1964
	ceph_object_extent_fn_t	count_fn;
	ceph_object_extent_fn_t	copy_fn;
1965
};
A
Alex Elder 已提交
1966

1967
static struct ceph_object_extent *alloc_object_extent(void *arg)
1968
{
1969 1970
	struct rbd_img_request *img_req = arg;
	struct rbd_obj_request *obj_req;
1971

1972 1973 1974
	obj_req = rbd_obj_request_create();
	if (!obj_req)
		return NULL;
I
Ilya Dryomov 已提交
1975

1976 1977 1978
	rbd_img_obj_request_add(img_req, obj_req);
	return &obj_req->ex;
}
1979

1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990
/*
 * 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;
}
1991

1992 1993 1994 1995 1996 1997 1998 1999 2000
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;
2001 2002

	/*
2003 2004
	 * Create object requests and set each object request's starting
	 * position in the provided bio (list) or bio_vec array.
2005
	 */
2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016
	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;
	}
2017

2018
	return __rbd_img_fill_request(img_req);
2019 2020
}

2021 2022 2023 2024
/*
 * 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,
2025
 * set up its data descriptor to point to the corresponding chunk(s) of
2026 2027
 * @fctx->pos data buffer.
 *
2028 2029 2030 2031
 * 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.
 *
2032 2033 2034 2035 2036 2037
 * @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)
2038
{
2039 2040
	struct rbd_device *rbd_dev = img_req->rbd_dev;
	struct rbd_obj_request *obj_req;
2041 2042 2043
	u32 i;
	int ret;

2044 2045 2046 2047
	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);
2048

2049
	img_req->data_type = OBJ_REQUEST_OWN_BVECS;
2050

2051
	/*
2052 2053 2054 2055 2056
	 * 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.
2057
	 */
2058 2059
	fctx->iter = *fctx->pos;
	for (i = 0; i < num_img_extents; i++) {
2060
		ret = ceph_file_to_extents(&rbd_dev->layout,
2061 2062 2063 2064
					   img_extents[i].fe_off,
					   img_extents[i].fe_len,
					   &img_req->object_extents,
					   alloc_object_extent, img_req,
2065 2066 2067
					   fctx->count_fn, &fctx->iter);
		if (ret)
			return ret;
2068
	}
2069

2070 2071 2072 2073 2074 2075 2076
	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;
	}
2077

2078
	/*
2079 2080
	 * Fill in each object request's private bio_vec array, splitting and
	 * rearranging the provided bio_vecs in stripe unit chunks as needed.
2081
	 */
2082 2083 2084 2085 2086 2087 2088
	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);
2089 2090 2091
		if (ret)
			return ret;
	}
2092

2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113
	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;
2114

2115 2116 2117 2118
	dout("%s objno %llu bytes %u\n", __func__, ex->oe_objno, bytes);
	obj_req->bio_pos = *it;
	ceph_bio_iter_advance(it, bytes);
}
2119

2120 2121 2122 2123 2124
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;
2125

2126 2127 2128 2129
	dout("%s objno %llu bytes %u\n", __func__, ex->oe_objno, bytes);
	ceph_bio_iter_advance_step(it, bytes, ({
		obj_req->bvec_count++;
	}));
2130

2131
}
2132

2133 2134 2135 2136 2137
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;
2138

2139 2140 2141 2142 2143
	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;
	}));
2144 2145
}

2146 2147 2148 2149 2150 2151 2152 2153 2154
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,
2155 2156
		.count_fn = count_bio_bvecs,
		.copy_fn = copy_bio_bvecs,
2157
	};
2158

2159 2160 2161
	return rbd_img_fill_request(img_req, img_extents, num_img_extents,
				    &fctx);
}
2162

2163 2164 2165 2166 2167
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 };
2168

2169 2170
	return __rbd_img_fill_from_bio(img_req, &ex, 1, &it);
}
A
Alex Elder 已提交
2171

2172 2173 2174 2175 2176
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;
2177

2178 2179 2180 2181
	obj_req->bvec_pos = *it;
	ceph_bvec_iter_shorten(&obj_req->bvec_pos, bytes);
	ceph_bvec_iter_advance(it, bytes);
}
2182

2183 2184 2185 2186 2187
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;
2188

2189 2190 2191 2192
	ceph_bvec_iter_advance_step(it, bytes, ({
		obj_req->bvec_count++;
	}));
}
2193

2194 2195 2196 2197 2198
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;
2199

2200 2201 2202 2203
	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;
	}));
2204 2205
}

2206 2207 2208 2209
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)
2210
{
2211 2212 2213 2214
	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,
2215 2216
		.count_fn = count_bvecs,
		.copy_fn = copy_bvecs,
2217
	};
2218

2219 2220 2221
	return rbd_img_fill_request(img_req, img_extents, num_img_extents,
				    &fctx);
}
2222

2223 2224 2225 2226 2227 2228 2229 2230 2231 2232
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) },
	};
2233

2234 2235 2236
	return __rbd_img_fill_from_bvecs(img_req, img_extents, num_img_extents,
					 &it);
}
2237

2238
static void rbd_img_request_submit(struct rbd_img_request *img_request)
A
Alex Elder 已提交
2239 2240
{
	struct rbd_obj_request *obj_request;
2241

A
Alex Elder 已提交
2242
	dout("%s: img %p\n", __func__, img_request);
2243

2244
	rbd_img_request_get(img_request);
2245
	for_each_obj_request(img_request, obj_request)
I
Ilya Dryomov 已提交
2246
		rbd_obj_request_submit(obj_request);
2247

2248
	rbd_img_request_put(img_request);
2249 2250
}

I
Ilya Dryomov 已提交
2251
static int rbd_obj_read_from_parent(struct rbd_obj_request *obj_req)
2252
{
I
Ilya Dryomov 已提交
2253 2254
	struct rbd_img_request *img_req = obj_req->img_request;
	struct rbd_img_request *child_img_req;
2255 2256
	int ret;

2257 2258
	child_img_req = rbd_img_request_create(img_req->rbd_dev->parent,
					       OBJ_OP_READ, NULL);
I
Ilya Dryomov 已提交
2259
	if (!child_img_req)
2260 2261
		return -ENOMEM;

2262 2263
	__set_bit(IMG_REQ_CHILD, &child_img_req->flags);
	child_img_req->obj_request = obj_req;
2264

I
Ilya Dryomov 已提交
2265
	if (!rbd_img_is_write(img_req)) {
2266
		switch (img_req->data_type) {
I
Ilya Dryomov 已提交
2267
		case OBJ_REQUEST_BIO:
2268 2269 2270 2271
			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
2272 2273
			break;
		case OBJ_REQUEST_BVECS:
2274
		case OBJ_REQUEST_OWN_BVECS:
2275 2276 2277 2278
			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
2279 2280 2281 2282 2283
			break;
		default:
			rbd_assert(0);
		}
	} else {
2284 2285 2286 2287
		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
2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303
	}
	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
2304 2305 2306
	    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
2307 2308 2309 2310
		if (ret) {
			obj_req->result = ret;
			return true;
		}
I
Ilya Dryomov 已提交
2311 2312 2313 2314 2315 2316 2317 2318 2319 2320

		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;
		}
2321 2322
	}

2323
	/*
I
Ilya Dryomov 已提交
2324 2325 2326 2327
	 * -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.
2328
	 */
I
Ilya Dryomov 已提交
2329
	if (obj_req->result == -ENOENT ||
2330
	    (!obj_req->result && obj_req->xferred < obj_req->ex.oe_len)) {
I
Ilya Dryomov 已提交
2331 2332
		rbd_assert(!obj_req->xferred || !obj_req->result);
		rbd_obj_zero_range(obj_req, obj_req->xferred,
2333
				   obj_req->ex.oe_len - obj_req->xferred);
I
Ilya Dryomov 已提交
2334
		obj_req->result = 0;
2335
		obj_req->xferred = obj_req->ex.oe_len;
2336
	}
2337

I
Ilya Dryomov 已提交
2338 2339
	return true;
}
2340

I
Ilya Dryomov 已提交
2341 2342 2343 2344 2345 2346 2347 2348 2349
/*
 * 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 },
	};
2350

I
Ilya Dryomov 已提交
2351 2352 2353 2354 2355 2356
	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;
2357 2358
}

I
Ilya Dryomov 已提交
2359
static int rbd_obj_issue_copyup(struct rbd_obj_request *obj_req, u32 bytes)
2360
{
I
Ilya Dryomov 已提交
2361
	unsigned int num_osd_ops = obj_req->osd_req->r_num_ops;
2362
	int ret;
2363

I
Ilya Dryomov 已提交
2364 2365 2366
	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);
2367

2368
	/*
I
Ilya Dryomov 已提交
2369 2370 2371
	 * 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).
2372
	 */
2373
	obj_req->osd_req = rbd_osd_req_create(obj_req, num_osd_ops);
I
Ilya Dryomov 已提交
2374 2375
	if (!obj_req->osd_req)
		return -ENOMEM;
2376

2377 2378 2379 2380 2381
	ret = osd_req_op_cls_init(obj_req->osd_req, 0, CEPH_OSD_OP_CALL, "rbd",
				  "copyup");
	if (ret)
		return ret;

2382
	/*
I
Ilya Dryomov 已提交
2383 2384 2385
	 * Only send non-zero copyup data to save some I/O and network
	 * bandwidth -- zero copyup data is equivalent to the object not
	 * existing.
2386
	 */
I
Ilya Dryomov 已提交
2387 2388 2389 2390 2391
	if (is_zero_bvecs(obj_req->copyup_bvecs, bytes)) {
		dout("%s obj_req %p detected zeroes\n", __func__, obj_req);
		bytes = 0;
	}
	osd_req_op_cls_request_data_bvecs(obj_req->osd_req, 0,
2392 2393 2394
					  obj_req->copyup_bvecs,
					  obj_req->copyup_bvec_count,
					  bytes);
I
Ilya Dryomov 已提交
2395

I
Ilya Dryomov 已提交
2396
	switch (obj_req->img_request->op_type) {
I
Ilya Dryomov 已提交
2397 2398 2399 2400 2401 2402 2403 2404 2405 2406
	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);
	}
2407

I
Ilya Dryomov 已提交
2408 2409
	rbd_obj_request_submit(obj_req);
	return 0;
2410 2411
}

2412
static int setup_copyup_bvecs(struct rbd_obj_request *obj_req, u64 obj_overlap)
2413
{
2414
	u32 i;
2415

2416 2417 2418 2419 2420 2421 2422
	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;
2423

2424 2425 2426 2427 2428 2429
	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;
2430

2431 2432 2433 2434
		obj_req->copyup_bvecs[i].bv_offset = 0;
		obj_req->copyup_bvecs[i].bv_len = len;
		obj_overlap -= len;
	}
2435

2436 2437
	rbd_assert(!obj_overlap);
	return 0;
2438 2439
}

I
Ilya Dryomov 已提交
2440
static int rbd_obj_handle_write_guard(struct rbd_obj_request *obj_req)
A
Alex Elder 已提交
2441
{
I
Ilya Dryomov 已提交
2442 2443
	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
	int ret;
A
Alex Elder 已提交
2444

I
Ilya Dryomov 已提交
2445 2446 2447 2448
	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
2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459
		/*
		 * 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
2460 2461
	}

I
Ilya Dryomov 已提交
2462
	ret = setup_copyup_bvecs(obj_req, rbd_obj_img_extents_bytes(obj_req));
I
Ilya Dryomov 已提交
2463 2464 2465 2466
	if (ret)
		return ret;

	obj_req->write_state = RBD_OBJ_WRITE_COPYUP;
I
Ilya Dryomov 已提交
2467
	return rbd_obj_read_from_parent(obj_req);
A
Alex Elder 已提交
2468
}
A
Alex Elder 已提交
2469

I
Ilya Dryomov 已提交
2470
static bool rbd_obj_handle_write(struct rbd_obj_request *obj_req)
A
Alex Elder 已提交
2471
{
I
Ilya Dryomov 已提交
2472
	int ret;
A
Alex Elder 已提交
2473

I
Ilya Dryomov 已提交
2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497
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.
			 */
2498
			obj_req->xferred = obj_req->ex.oe_len;
I
Ilya Dryomov 已提交
2499 2500 2501 2502 2503
		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
2504

I
Ilya Dryomov 已提交
2505 2506 2507 2508 2509 2510 2511 2512
		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
2513
		BUG();
I
Ilya Dryomov 已提交
2514 2515
	}
}
2516

I
Ilya Dryomov 已提交
2517 2518 2519 2520 2521
/*
 * 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
2522
	switch (obj_req->img_request->op_type) {
I
Ilya Dryomov 已提交
2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534
	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;
2535
				obj_req->xferred = obj_req->ex.oe_len;
I
Ilya Dryomov 已提交
2536 2537 2538 2539 2540
			}
			return true;
		}
		return false;
	default:
A
Arnd Bergmann 已提交
2541
		BUG();
I
Ilya Dryomov 已提交
2542 2543
	}
}
2544

I
Ilya Dryomov 已提交
2545 2546 2547 2548 2549
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 &&
2550
		    obj_req->xferred == obj_req->ex.oe_len) ||
I
Ilya Dryomov 已提交
2551 2552 2553
		   (obj_req->result < 0 && !obj_req->xferred));
	if (!obj_req->result) {
		img_req->xferred += obj_req->xferred;
2554
		return;
2555
	}
A
Alex Elder 已提交
2556

I
Ilya Dryomov 已提交
2557 2558
	rbd_warn(img_req->rbd_dev,
		 "%s at objno %llu %llu~%llu result %d xferred %llu",
2559 2560
		 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
2561 2562 2563 2564 2565 2566
		 obj_req->xferred);
	if (!img_req->result) {
		img_req->result = obj_req->result;
		img_req->xferred = 0;
	}
}
A
Alex Elder 已提交
2567

I
Ilya Dryomov 已提交
2568 2569 2570
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
2571

I
Ilya Dryomov 已提交
2572
	rbd_assert(test_bit(IMG_REQ_CHILD, &img_req->flags));
I
Ilya Dryomov 已提交
2573 2574 2575
	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
2576

I
Ilya Dryomov 已提交
2577 2578 2579
	obj_req->result = img_req->result;
	obj_req->xferred = img_req->xferred;
	rbd_img_request_put(img_req);
A
Alex Elder 已提交
2580 2581
}

I
Ilya Dryomov 已提交
2582
static void rbd_img_end_request(struct rbd_img_request *img_req)
A
Alex Elder 已提交
2583
{
I
Ilya Dryomov 已提交
2584 2585 2586 2587
	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
2588

I
Ilya Dryomov 已提交
2589 2590 2591
	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
2592
}
A
Alex Elder 已提交
2593

I
Ilya Dryomov 已提交
2594 2595
static void rbd_obj_handle_request(struct rbd_obj_request *obj_req)
{
I
Ilya Dryomov 已提交
2596
	struct rbd_img_request *img_req;
A
Alex Elder 已提交
2597

I
Ilya Dryomov 已提交
2598
again:
I
Ilya Dryomov 已提交
2599 2600
	if (!__rbd_obj_handle_request(obj_req))
		return;
A
Alex Elder 已提交
2601

I
Ilya Dryomov 已提交
2602 2603 2604 2605 2606 2607 2608 2609
	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
2610

I
Ilya Dryomov 已提交
2611 2612 2613 2614 2615 2616 2617
	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
2618
}
A
Alex Elder 已提交
2619

2620
static const struct rbd_client_id rbd_empty_cid;
A
Alex Elder 已提交
2621

2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657
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);
}

2658 2659 2660 2661 2662 2663 2664 2665 2666
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);
}

2667 2668 2669 2670
/*
 * lock_rwsem must be held for write
 */
static int rbd_lock(struct rbd_device *rbd_dev)
A
Alex Elder 已提交
2671
{
2672
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2673
	char cookie[32];
2674
	int ret;
A
Alex Elder 已提交
2675

I
Ilya Dryomov 已提交
2676 2677
	WARN_ON(__rbd_is_lock_owner(rbd_dev) ||
		rbd_dev->lock_cookie[0] != '\0');
2678

2679 2680 2681 2682
	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);
2683
	if (ret)
2684
		return ret;
A
Alex Elder 已提交
2685

2686
	rbd_dev->lock_state = RBD_LOCK_STATE_LOCKED;
2687
	__rbd_lock(rbd_dev, cookie);
2688
	return 0;
A
Alex Elder 已提交
2689 2690
}

2691 2692 2693
/*
 * lock_rwsem must be held for write
 */
I
Ilya Dryomov 已提交
2694
static void rbd_unlock(struct rbd_device *rbd_dev)
2695
{
2696
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2697 2698
	int ret;

I
Ilya Dryomov 已提交
2699 2700
	WARN_ON(!__rbd_is_lock_owner(rbd_dev) ||
		rbd_dev->lock_cookie[0] == '\0');
2701

2702
	ret = ceph_cls_unlock(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc,
I
Ilya Dryomov 已提交
2703
			      RBD_LOCK_NAME, rbd_dev->lock_cookie);
I
Ilya Dryomov 已提交
2704 2705
	if (ret && ret != -ENOENT)
		rbd_warn(rbd_dev, "failed to unlock: %d", ret);
2706

I
Ilya Dryomov 已提交
2707 2708
	/* treat errors as the image is unlocked */
	rbd_dev->lock_state = RBD_LOCK_STATE_UNLOCKED;
I
Ilya Dryomov 已提交
2709
	rbd_dev->lock_cookie[0] = '\0';
2710 2711
	rbd_set_owner_cid(rbd_dev, &rbd_empty_cid);
	queue_work(rbd_dev->task_wq, &rbd_dev->released_lock_work);
2712 2713
}

2714 2715 2716 2717
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)
2718 2719
{
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2720
	struct rbd_client_id cid = rbd_get_cid(rbd_dev);
K
Kyle Spiers 已提交
2721 2722
	char buf[4 + 8 + 8 + CEPH_ENCODING_START_BLK_LEN];
	int buf_size = sizeof(buf);
2723
	void *p = buf;
2724

2725
	dout("%s rbd_dev %p notify_op %d\n", __func__, rbd_dev, notify_op);
2726

2727 2728 2729 2730 2731
	/* 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);
2732

2733 2734 2735
	return ceph_osdc_notify(osdc, &rbd_dev->header_oid,
				&rbd_dev->header_oloc, buf, buf_size,
				RBD_NOTIFY_TIMEOUT, preply_pages, preply_len);
2736 2737
}

2738 2739
static void rbd_notify_op_lock(struct rbd_device *rbd_dev,
			       enum rbd_notify_op notify_op)
2740
{
2741 2742
	struct page **reply_pages;
	size_t reply_len;
2743

2744 2745 2746
	__rbd_notify_op_lock(rbd_dev, notify_op, &reply_pages, &reply_len);
	ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len));
}
2747

2748 2749 2750 2751
static void rbd_notify_acquired_lock(struct work_struct *work)
{
	struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
						  acquired_lock_work);
2752

2753
	rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_ACQUIRED_LOCK);
2754 2755
}

2756
static void rbd_notify_released_lock(struct work_struct *work)
2757
{
2758 2759
	struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
						  released_lock_work);
I
Ilya Dryomov 已提交
2760

2761
	rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_RELEASED_LOCK);
2762 2763
}

2764
static int rbd_request_lock(struct rbd_device *rbd_dev)
2765
{
2766 2767 2768
	struct page **reply_pages;
	size_t reply_len;
	bool lock_owner_responded = false;
2769 2770
	int ret;

2771
	dout("%s rbd_dev %p\n", __func__, rbd_dev);
2772

2773 2774 2775 2776
	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);
2777
		goto out;
2778
	}
2779

2780 2781 2782 2783
	if (reply_len > 0 && reply_len <= PAGE_SIZE) {
		void *p = page_address(reply_pages[0]);
		void *const end = p + reply_len;
		u32 n;
2784

2785 2786 2787 2788
		ceph_decode_32_safe(&p, end, n, e_inval); /* num_acks */
		while (n--) {
			u8 struct_v;
			u32 len;
2789

2790 2791
			ceph_decode_need(&p, end, 8 + 8, e_inval);
			p += 8 + 8; /* skip gid and cookie */
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 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024
			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;
3025
	int ret = 0;
3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040

	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
3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052
	} 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);
		}
3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079 3080
	} 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);
3081
	/*
3082
	 * Ensure that all in-flight IO is flushed.
3083
	 *
3084 3085
	 * FIXME: ceph_osdc_sync() flushes the entire OSD client, which
	 * may be shared with other devices.
3086
	 */
3087 3088 3089 3090 3091 3092 3093 3094 3095
	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
3096 3097 3098 3099 3100 3101 3102 3103 3104
	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);
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 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184
	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);
}

3185 3186 3187 3188 3189 3190
/*
 * 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)
3191 3192 3193
{
	struct rbd_client_id my_cid = rbd_get_cid(rbd_dev);
	struct rbd_client_id cid = { 0 };
3194
	int result = 1;
3195 3196 3197 3198 3199 3200 3201 3202 3203

	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))
3204
		return result;
3205 3206

	down_read(&rbd_dev->lock_rwsem);
3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218
	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
3219 3220 3221 3222 3223 3224 3225 3226 3227
			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;
			}
3228 3229
		}
	}
3230 3231

out_unlock:
3232
	up_read(&rbd_dev->lock_rwsem);
3233
	return result;
3234 3235 3236 3237 3238 3239
}

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
3240 3241
	char buf[4 + CEPH_ENCODING_START_BLK_LEN];
	int buf_size = sizeof(buf);
3242 3243 3244 3245 3246 3247 3248 3249 3250 3251 3252 3253
	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
3254

3255 3256
	ret = ceph_osdc_notify_ack(osdc, &rbd_dev->header_oid,
				   &rbd_dev->header_oloc, notify_id, cookie,
3257
				   buf, buf_size);
3258
	if (ret)
3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281
		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;
3282
	u8 struct_v = 0;
3283 3284 3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313 3314 3315
	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:
3316 3317
		ret = rbd_handle_request_lock(rbd_dev, struct_v, &p);
		if (ret <= 0)
3318
			rbd_acknowledge_notify_result(rbd_dev, notify_id,
3319
						      cookie, ret);
3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337
		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
3338 3339
}

3340 3341
static void __rbd_unregister_watch(struct rbd_device *rbd_dev);

3342
static void rbd_watch_errcb(void *arg, u64 cookie, int err)
3343
{
3344
	struct rbd_device *rbd_dev = arg;
3345

3346
	rbd_warn(rbd_dev, "encountered watch error: %d", err);
3347

3348 3349 3350 3351
	down_write(&rbd_dev->lock_rwsem);
	rbd_set_owner_cid(rbd_dev, &rbd_empty_cid);
	up_write(&rbd_dev->lock_rwsem);

3352 3353 3354 3355
	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;
3356

3357
		queue_delayed_work(rbd_dev->task_wq, &rbd_dev->watch_dwork, 0);
3358
	}
3359
	mutex_unlock(&rbd_dev->watch_mutex);
3360 3361
}

3362
/*
3363
 * watch_mutex must be locked
3364
 */
3365
static int __rbd_register_watch(struct rbd_device *rbd_dev)
3366 3367
{
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3368
	struct ceph_osd_linger_request *handle;
3369

3370
	rbd_assert(!rbd_dev->watch_handle);
3371
	dout("%s rbd_dev %p\n", __func__, rbd_dev);
3372

3373 3374 3375 3376 3377
	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);
3378

3379
	rbd_dev->watch_handle = handle;
3380 3381 3382
	return 0;
}

3383 3384 3385 3386
/*
 * watch_mutex must be locked
 */
static void __rbd_unregister_watch(struct rbd_device *rbd_dev)
3387
{
3388 3389
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
	int ret;
3390

3391 3392
	rbd_assert(rbd_dev->watch_handle);
	dout("%s rbd_dev %p\n", __func__, rbd_dev);
3393

3394 3395 3396
	ret = ceph_osdc_unwatch(osdc, rbd_dev->watch_handle);
	if (ret)
		rbd_warn(rbd_dev, "failed to unwatch: %d", ret);
3397

3398
	rbd_dev->watch_handle = NULL;
3399 3400
}

3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419
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)
3420
{
3421 3422
	dout("%s rbd_dev %p\n", __func__, rbd_dev);

3423 3424 3425 3426
	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);
3427 3428 3429 3430
}

static void rbd_unregister_watch(struct rbd_device *rbd_dev)
{
3431
	WARN_ON(waitqueue_active(&rbd_dev->lock_waitq));
3432 3433 3434 3435 3436 3437 3438
	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
3439

3440
	cancel_delayed_work_sync(&rbd_dev->watch_dwork);
I
Ilya Dryomov 已提交
3441
	ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc);
3442 3443
}

3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471 3472
/*
 * 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 {
3473
		__rbd_lock(rbd_dev, cookie);
3474 3475 3476
	}
}

3477 3478 3479 3480 3481 3482 3483 3484 3485
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);
3486 3487
	if (rbd_dev->watch_state != RBD_WATCH_STATE_ERROR) {
		mutex_unlock(&rbd_dev->watch_mutex);
3488
		return;
3489
	}
3490 3491 3492 3493

	ret = __rbd_register_watch(rbd_dev);
	if (ret) {
		rbd_warn(rbd_dev, "failed to reregister watch: %d", ret);
3494
		if (ret == -EBLACKLISTED || ret == -ENOENT) {
3495
			set_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags);
3496
			wake_requests(rbd_dev, true);
3497
		} else {
3498 3499 3500
			queue_delayed_work(rbd_dev->task_wq,
					   &rbd_dev->watch_dwork,
					   RBD_RETRY_DELAY);
3501 3502
		}
		mutex_unlock(&rbd_dev->watch_mutex);
3503
		return;
3504 3505 3506 3507 3508 3509
	}

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

3510 3511 3512 3513 3514
	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);

3515 3516
	ret = rbd_dev_refresh(rbd_dev);
	if (ret)
3517
		rbd_warn(rbd_dev, "reregistration refresh failed: %d", ret);
3518 3519
}

3520
/*
3521 3522
 * Synchronous osd object method call.  Returns the number of bytes
 * returned in the outbound buffer, or a negative error code.
3523 3524
 */
static int rbd_obj_method_sync(struct rbd_device *rbd_dev,
3525 3526
			     struct ceph_object_id *oid,
			     struct ceph_object_locator *oloc,
3527
			     const char *method_name,
3528
			     const void *outbound,
3529
			     size_t outbound_size,
3530
			     void *inbound,
3531
			     size_t inbound_size)
3532
{
3533 3534 3535
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
	struct page *req_page = NULL;
	struct page *reply_page;
3536 3537 3538
	int ret;

	/*
3539 3540 3541 3542 3543
	 * 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.
3544
	 */
3545 3546 3547
	if (outbound) {
		if (outbound_size > PAGE_SIZE)
			return -E2BIG;
3548

3549 3550 3551
		req_page = alloc_page(GFP_KERNEL);
		if (!req_page)
			return -ENOMEM;
3552

3553
		memcpy(page_address(req_page), outbound, outbound_size);
3554
	}
3555

3556 3557 3558 3559 3560 3561
	reply_page = alloc_page(GFP_KERNEL);
	if (!reply_page) {
		if (req_page)
			__free_page(req_page);
		return -ENOMEM;
	}
3562

3563 3564 3565 3566 3567 3568 3569
	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;
	}
3570

3571 3572 3573
	if (req_page)
		__free_page(req_page);
	__free_page(reply_page);
3574 3575 3576
	return ret;
}

3577 3578 3579
/*
 * lock_rwsem must be held for read
 */
3580
static int rbd_wait_state_locked(struct rbd_device *rbd_dev, bool may_acquire)
3581 3582
{
	DEFINE_WAIT(wait);
3583
	unsigned long timeout;
3584 3585 3586 3587 3588 3589 3590 3591 3592 3593 3594 3595
	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;
	}
3596 3597 3598 3599 3600 3601 3602 3603 3604 3605 3606

	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);
3607 3608
		timeout = schedule_timeout(ceph_timeout_jiffies(
						rbd_dev->opts->lock_timeout));
3609
		down_read(&rbd_dev->lock_rwsem);
3610 3611 3612 3613
		if (test_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags)) {
			ret = -EBLACKLISTED;
			break;
		}
3614 3615 3616 3617 3618
		if (!timeout) {
			rbd_warn(rbd_dev, "timed out waiting for lock");
			ret = -ETIMEDOUT;
			break;
		}
3619
	} while (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED);
3620

3621
	finish_wait(&rbd_dev->lock_waitq, &wait);
3622
	return ret;
3623 3624
}

C
Christoph Hellwig 已提交
3625
static void rbd_queue_workfn(struct work_struct *work)
A
Alex Elder 已提交
3626
{
C
Christoph Hellwig 已提交
3627 3628
	struct request *rq = blk_mq_rq_from_pdu(work);
	struct rbd_device *rbd_dev = rq->q->queuedata;
I
Ilya Dryomov 已提交
3629
	struct rbd_img_request *img_request;
3630
	struct ceph_snap_context *snapc = NULL;
I
Ilya Dryomov 已提交
3631 3632
	u64 offset = (u64)blk_rq_pos(rq) << SECTOR_SHIFT;
	u64 length = blk_rq_bytes(rq);
G
Guangliang Zhao 已提交
3633
	enum obj_operation_type op_type;
3634
	u64 mapping_size;
I
Ilya Dryomov 已提交
3635
	bool must_be_locked;
A
Alex Elder 已提交
3636 3637
	int result;

3638 3639
	switch (req_op(rq)) {
	case REQ_OP_DISCARD:
3640
	case REQ_OP_WRITE_ZEROES:
3641
		op_type = OBJ_OP_DISCARD;
3642 3643
		break;
	case REQ_OP_WRITE:
G
Guangliang Zhao 已提交
3644
		op_type = OBJ_OP_WRITE;
3645 3646
		break;
	case REQ_OP_READ:
G
Guangliang Zhao 已提交
3647
		op_type = OBJ_OP_READ;
3648 3649 3650 3651 3652 3653
		break;
	default:
		dout("%s: non-fs request type %d\n", __func__, req_op(rq));
		result = -EIO;
		goto err;
	}
G
Guangliang Zhao 已提交
3654

I
Ilya Dryomov 已提交
3655
	/* Ignore/skip any zero-length requests */
A
Alex Elder 已提交
3656

I
Ilya Dryomov 已提交
3657 3658 3659 3660 3661
	if (!length) {
		dout("%s: zero-length request\n", __func__);
		result = 0;
		goto err_rq;
	}
A
Alex Elder 已提交
3662

3663 3664
	rbd_assert(op_type == OBJ_OP_READ ||
		   rbd_dev->spec->snap_id == CEPH_NOSNAP);
A
Alex Elder 已提交
3665

I
Ilya Dryomov 已提交
3666 3667 3668 3669 3670 3671 3672 3673 3674 3675 3676 3677
	/*
	 * 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
3678

I
Ilya Dryomov 已提交
3679 3680 3681 3682 3683 3684
	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
3685

C
Christoph Hellwig 已提交
3686 3687
	blk_mq_start_request(rq);

3688 3689
	down_read(&rbd_dev->header_rwsem);
	mapping_size = rbd_dev->mapping.size;
G
Guangliang Zhao 已提交
3690
	if (op_type != OBJ_OP_READ) {
3691 3692 3693 3694 3695 3696
		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
3697
		rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)", offset,
3698
			 length, mapping_size);
I
Ilya Dryomov 已提交
3699 3700 3701
		result = -EIO;
		goto err_rq;
	}
A
Alex Elder 已提交
3702

I
Ilya Dryomov 已提交
3703 3704 3705
	must_be_locked =
	    (rbd_dev->header.features & RBD_FEATURE_EXCLUSIVE_LOCK) &&
	    (op_type != OBJ_OP_READ || rbd_dev->opts->lock_on_read);
3706 3707
	if (must_be_locked) {
		down_read(&rbd_dev->lock_rwsem);
3708 3709 3710
		result = rbd_wait_state_locked(rbd_dev,
					       !rbd_dev->opts->exclusive);
		if (result)
3711
			goto err_unlock;
3712 3713
	}

3714
	img_request = rbd_img_request_create(rbd_dev, op_type, snapc);
I
Ilya Dryomov 已提交
3715 3716
	if (!img_request) {
		result = -ENOMEM;
3717
		goto err_unlock;
I
Ilya Dryomov 已提交
3718 3719
	}
	img_request->rq = rq;
3720
	snapc = NULL; /* img_request consumes a ref */
A
Alex Elder 已提交
3721

3722
	if (op_type == OBJ_OP_DISCARD)
3723
		result = rbd_img_fill_nodata(img_request, offset, length);
3724
	else
3725 3726
		result = rbd_img_fill_from_bio(img_request, offset, length,
					       rq->bio);
I
Ilya Dryomov 已提交
3727 3728
	if (result)
		goto err_img_request;
A
Alex Elder 已提交
3729

3730
	rbd_img_request_submit(img_request);
3731 3732
	if (must_be_locked)
		up_read(&rbd_dev->lock_rwsem);
I
Ilya Dryomov 已提交
3733
	return;
A
Alex Elder 已提交
3734

I
Ilya Dryomov 已提交
3735 3736
err_img_request:
	rbd_img_request_put(img_request);
3737 3738 3739
err_unlock:
	if (must_be_locked)
		up_read(&rbd_dev->lock_rwsem);
I
Ilya Dryomov 已提交
3740 3741 3742
err_rq:
	if (result)
		rbd_warn(rbd_dev, "%s %llx at %llx result %d",
G
Guangliang Zhao 已提交
3743
			 obj_op_name(op_type), length, offset, result);
3744
	ceph_put_snap_context(snapc);
C
Christoph Hellwig 已提交
3745
err:
3746
	blk_mq_end_request(rq, errno_to_blk_status(result));
I
Ilya Dryomov 已提交
3747
}
A
Alex Elder 已提交
3748

3749
static blk_status_t rbd_queue_rq(struct blk_mq_hw_ctx *hctx,
C
Christoph Hellwig 已提交
3750
		const struct blk_mq_queue_data *bd)
I
Ilya Dryomov 已提交
3751
{
C
Christoph Hellwig 已提交
3752 3753
	struct request *rq = bd->rq;
	struct work_struct *work = blk_mq_rq_to_pdu(rq);
A
Alex Elder 已提交
3754

C
Christoph Hellwig 已提交
3755
	queue_work(rbd_wq, work);
3756
	return BLK_STS_OK;
A
Alex Elder 已提交
3757 3758
}

3759 3760
static void rbd_free_disk(struct rbd_device *rbd_dev)
{
3761 3762 3763
	blk_cleanup_queue(rbd_dev->disk->queue);
	blk_mq_free_tag_set(&rbd_dev->tag_set);
	put_disk(rbd_dev->disk);
3764
	rbd_dev->disk = NULL;
3765 3766
}

3767
static int rbd_obj_read_sync(struct rbd_device *rbd_dev,
3768 3769 3770
			     struct ceph_object_id *oid,
			     struct ceph_object_locator *oloc,
			     void *buf, int buf_len)
3771 3772

{
3773 3774 3775 3776
	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);
3777 3778
	int ret;

3779 3780 3781
	req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_KERNEL);
	if (!req)
		return -ENOMEM;
3782

3783 3784 3785
	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
3786

3787
	ret = ceph_osdc_alloc_messages(req, GFP_KERNEL);
3788
	if (ret)
3789
		goto out_req;
3790

3791 3792 3793 3794 3795
	pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
	if (IS_ERR(pages)) {
		ret = PTR_ERR(pages);
		goto out_req;
	}
3796

3797 3798 3799 3800 3801 3802 3803 3804
	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);
3805

3806 3807
out_req:
	ceph_osdc_put_request(req);
3808 3809 3810
	return ret;
}

3811
/*
A
Alex Elder 已提交
3812 3813 3814
 * 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.
3815
 */
3816
static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev)
3817
{
3818
	struct rbd_image_header_ondisk *ondisk = NULL;
3819
	u32 snap_count = 0;
3820 3821 3822
	u64 names_size = 0;
	u32 want_count;
	int ret;
3823

A
Alex Elder 已提交
3824
	/*
3825 3826 3827 3828 3829
	 * 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
3830
	 */
3831 3832 3833 3834 3835 3836 3837 3838 3839 3840
	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
3841
			return -ENOMEM;
3842

3843 3844
		ret = rbd_obj_read_sync(rbd_dev, &rbd_dev->header_oid,
					&rbd_dev->header_oloc, ondisk, size);
3845
		if (ret < 0)
A
Alex Elder 已提交
3846
			goto out;
A
Alex Elder 已提交
3847
		if ((size_t)ret < size) {
3848
			ret = -ENXIO;
A
Alex Elder 已提交
3849 3850
			rbd_warn(rbd_dev, "short header read (want %zd got %d)",
				size, ret);
A
Alex Elder 已提交
3851
			goto out;
3852 3853 3854
		}
		if (!rbd_dev_ondisk_valid(ondisk)) {
			ret = -ENXIO;
A
Alex Elder 已提交
3855
			rbd_warn(rbd_dev, "invalid header");
A
Alex Elder 已提交
3856
			goto out;
3857
		}
3858

3859 3860 3861 3862
		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
3863

A
Alex Elder 已提交
3864 3865
	ret = rbd_header_from_disk(rbd_dev, ondisk);
out:
3866 3867 3868
	kfree(ondisk);

	return ret;
3869 3870
}

3871 3872 3873 3874 3875 3876 3877 3878 3879 3880 3881 3882 3883 3884 3885 3886 3887 3888 3889
/*
 * 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);
}

3890 3891 3892 3893 3894
static void rbd_dev_update_size(struct rbd_device *rbd_dev)
{
	sector_t size;

	/*
I
Ilya Dryomov 已提交
3895 3896 3897
	 * 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.
3898
	 */
I
Ilya Dryomov 已提交
3899 3900
	if (test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags) &&
	    !test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) {
3901 3902 3903 3904 3905 3906 3907
		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
3908
static int rbd_dev_refresh(struct rbd_device *rbd_dev)
A
Alex Elder 已提交
3909
{
3910
	u64 mapping_size;
A
Alex Elder 已提交
3911 3912
	int ret;

3913
	down_write(&rbd_dev->header_rwsem);
3914
	mapping_size = rbd_dev->mapping.size;
3915 3916

	ret = rbd_dev_header_info(rbd_dev);
3917
	if (ret)
3918
		goto out;
3919

3920 3921 3922 3923 3924 3925 3926
	/*
	 * 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)
3927
			goto out;
3928 3929
	}

3930
	if (rbd_dev->spec->snap_id == CEPH_NOSNAP) {
3931
		rbd_dev->mapping.size = rbd_dev->header.image_size;
3932 3933 3934 3935
	} else {
		/* validate mapped snapshot's EXISTS flag */
		rbd_exists_validate(rbd_dev);
	}
3936

3937
out:
3938
	up_write(&rbd_dev->header_rwsem);
3939
	if (!ret && mapping_size != rbd_dev->mapping.size)
3940
		rbd_dev_update_size(rbd_dev);
A
Alex Elder 已提交
3941

3942
	return ret;
A
Alex Elder 已提交
3943 3944
}

3945 3946
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
3947 3948 3949 3950 3951 3952 3953
{
	struct work_struct *work = blk_mq_rq_to_pdu(rq);

	INIT_WORK(work, rbd_queue_workfn);
	return 0;
}

3954
static const struct blk_mq_ops rbd_mq_ops = {
C
Christoph Hellwig 已提交
3955 3956 3957 3958
	.queue_rq	= rbd_queue_rq,
	.init_request	= rbd_init_request,
};

3959 3960 3961 3962
static int rbd_init_disk(struct rbd_device *rbd_dev)
{
	struct gendisk *disk;
	struct request_queue *q;
3963 3964
	unsigned int objset_bytes =
	    rbd_dev->layout.object_size * rbd_dev->layout.stripe_count;
C
Christoph Hellwig 已提交
3965
	int err;
3966 3967

	/* create gendisk info */
3968 3969 3970
	disk = alloc_disk(single_major ?
			  (1 << RBD_SINGLE_MAJOR_PART_SHIFT) :
			  RBD_MINORS_PER_MAJOR);
3971
	if (!disk)
A
Alex Elder 已提交
3972
		return -ENOMEM;
3973

A
Alex Elder 已提交
3974
	snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
A
Alex Elder 已提交
3975
		 rbd_dev->dev_id);
3976
	disk->major = rbd_dev->major;
3977
	disk->first_minor = rbd_dev->minor;
3978 3979
	if (single_major)
		disk->flags |= GENHD_FL_EXT_DEVT;
3980 3981 3982
	disk->fops = &rbd_bd_ops;
	disk->private_data = rbd_dev;

C
Christoph Hellwig 已提交
3983 3984
	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
3985
	rbd_dev->tag_set.queue_depth = rbd_dev->opts->queue_depth;
C
Christoph Hellwig 已提交
3986
	rbd_dev->tag_set.numa_node = NUMA_NO_NODE;
I
Ilya Dryomov 已提交
3987
	rbd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
C
Christoph Hellwig 已提交
3988 3989 3990 3991 3992
	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)
3993
		goto out_disk;
3994

C
Christoph Hellwig 已提交
3995 3996 3997 3998 3999 4000
	q = blk_mq_init_queue(&rbd_dev->tag_set);
	if (IS_ERR(q)) {
		err = PTR_ERR(q);
		goto out_tag_set;
	}

4001
	blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
4002
	/* QUEUE_FLAG_ADD_RANDOM is off by default for blk-mq */
A
Alex Elder 已提交
4003

4004
	blk_queue_max_hw_sectors(q, objset_bytes >> SECTOR_SHIFT);
I
Ilya Dryomov 已提交
4005
	q->limits.max_sectors = queue_max_hw_sectors(q);
4006
	blk_queue_max_segments(q, USHRT_MAX);
4007
	blk_queue_max_segment_size(q, UINT_MAX);
4008 4009
	blk_queue_io_min(q, objset_bytes);
	blk_queue_io_opt(q, objset_bytes);
4010

I
Ilya Dryomov 已提交
4011 4012 4013 4014 4015 4016
	if (rbd_dev->opts->trim) {
		blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
		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);
	}
4017

4018
	if (!ceph_test_opt(rbd_dev->rbd_client->client, NOCRC))
4019
		q->backing_dev_info->capabilities |= BDI_CAP_STABLE_WRITES;
4020

4021 4022 4023 4024 4025
	/*
	 * 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));
4026 4027 4028 4029 4030 4031
	disk->queue = q;
	q->queuedata = rbd_dev;

	rbd_dev->disk = disk;

	return 0;
C
Christoph Hellwig 已提交
4032 4033
out_tag_set:
	blk_mq_free_tag_set(&rbd_dev->tag_set);
4034 4035
out_disk:
	put_disk(disk);
C
Christoph Hellwig 已提交
4036
	return err;
4037 4038
}

4039 4040 4041 4042
/*
  sysfs
*/

A
Alex Elder 已提交
4043 4044 4045 4046 4047
static struct rbd_device *dev_to_rbd_dev(struct device *dev)
{
	return container_of(dev, struct rbd_device, dev);
}

4048 4049 4050
static ssize_t rbd_size_show(struct device *dev,
			     struct device_attribute *attr, char *buf)
{
A
Alex Elder 已提交
4051
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4052

A
Alex Elder 已提交
4053 4054
	return sprintf(buf, "%llu\n",
		(unsigned long long)rbd_dev->mapping.size);
4055 4056
}

A
Alex Elder 已提交
4057 4058 4059 4060 4061 4062 4063 4064 4065 4066
/*
 * 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
4067
			(unsigned long long)rbd_dev->mapping.features);
A
Alex Elder 已提交
4068 4069
}

4070 4071 4072
static ssize_t rbd_major_show(struct device *dev,
			      struct device_attribute *attr, char *buf)
{
A
Alex Elder 已提交
4073
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4074

A
Alex Elder 已提交
4075 4076 4077 4078
	if (rbd_dev->major)
		return sprintf(buf, "%d\n", rbd_dev->major);

	return sprintf(buf, "(none)\n");
4079 4080 4081 4082 4083 4084
}

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
4085

4086
	return sprintf(buf, "%d\n", rbd_dev->minor);
4087 4088
}

4089 4090 4091 4092 4093 4094 4095 4096 4097 4098 4099
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));
}

4100 4101
static ssize_t rbd_client_id_show(struct device *dev,
				  struct device_attribute *attr, char *buf)
4102
{
A
Alex Elder 已提交
4103
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4104

4105
	return sprintf(buf, "client%lld\n",
4106
		       ceph_client_gid(rbd_dev->rbd_client->client));
4107 4108
}

4109 4110 4111 4112 4113 4114 4115 4116
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);
}

4117 4118 4119 4120 4121 4122
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);
4123 4124
}

4125 4126
static ssize_t rbd_pool_show(struct device *dev,
			     struct device_attribute *attr, char *buf)
4127
{
A
Alex Elder 已提交
4128
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4129

4130
	return sprintf(buf, "%s\n", rbd_dev->spec->pool_name);
4131 4132
}

4133 4134 4135 4136 4137
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);

4138
	return sprintf(buf, "%llu\n",
A
Alex Elder 已提交
4139
			(unsigned long long) rbd_dev->spec->pool_id);
4140 4141
}

4142 4143 4144 4145 4146 4147 4148 4149
static ssize_t rbd_pool_ns_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->spec->pool_ns ?: "");
}

4150 4151 4152
static ssize_t rbd_name_show(struct device *dev,
			     struct device_attribute *attr, char *buf)
{
A
Alex Elder 已提交
4153
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4154

A
Alex Elder 已提交
4155 4156 4157 4158
	if (rbd_dev->spec->image_name)
		return sprintf(buf, "%s\n", rbd_dev->spec->image_name);

	return sprintf(buf, "(unknown)\n");
4159 4160
}

A
Alex Elder 已提交
4161 4162 4163 4164 4165
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);

4166
	return sprintf(buf, "%s\n", rbd_dev->spec->image_id);
A
Alex Elder 已提交
4167 4168
}

A
Alex Elder 已提交
4169 4170 4171 4172
/*
 * Shows the name of the currently-mapped snapshot (or
 * RBD_SNAP_HEAD_NAME for the base image).
 */
4173 4174 4175 4176
static ssize_t rbd_snap_show(struct device *dev,
			     struct device_attribute *attr,
			     char *buf)
{
A
Alex Elder 已提交
4177
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4178

4179
	return sprintf(buf, "%s\n", rbd_dev->spec->snap_name);
4180 4181
}

4182 4183 4184 4185 4186 4187 4188 4189
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);
}

4190
/*
4191 4192 4193
 * 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)".
4194 4195
 */
static ssize_t rbd_parent_show(struct device *dev,
4196 4197
			       struct device_attribute *attr,
			       char *buf)
4198 4199
{
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4200
	ssize_t count = 0;
4201

4202
	if (!rbd_dev->parent)
4203 4204
		return sprintf(buf, "(no parent image)\n");

4205 4206 4207 4208 4209 4210 4211 4212 4213 4214 4215 4216 4217 4218 4219 4220
	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;
4221 4222
}

4223 4224 4225 4226 4227
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
4228
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4229
	int ret;
4230

A
Alex Elder 已提交
4231
	ret = rbd_dev_refresh(rbd_dev);
4232
	if (ret)
4233
		return ret;
4234

4235
	return size;
4236
}
4237

4238 4239 4240 4241 4242 4243 4244 4245 4246 4247
static DEVICE_ATTR(size, 0444, rbd_size_show, NULL);
static DEVICE_ATTR(features, 0444, rbd_features_show, NULL);
static DEVICE_ATTR(major, 0444, rbd_major_show, NULL);
static DEVICE_ATTR(minor, 0444, rbd_minor_show, NULL);
static DEVICE_ATTR(client_addr, 0444, rbd_client_addr_show, NULL);
static DEVICE_ATTR(client_id, 0444, rbd_client_id_show, NULL);
static DEVICE_ATTR(cluster_fsid, 0444, rbd_cluster_fsid_show, NULL);
static DEVICE_ATTR(config_info, 0400, rbd_config_info_show, NULL);
static DEVICE_ATTR(pool, 0444, rbd_pool_show, NULL);
static DEVICE_ATTR(pool_id, 0444, rbd_pool_id_show, NULL);
4248
static DEVICE_ATTR(pool_ns, 0444, rbd_pool_ns_show, NULL);
4249 4250 4251 4252 4253 4254
static DEVICE_ATTR(name, 0444, rbd_name_show, NULL);
static DEVICE_ATTR(image_id, 0444, rbd_image_id_show, NULL);
static DEVICE_ATTR(refresh, 0200, NULL, rbd_image_refresh);
static DEVICE_ATTR(current_snap, 0444, rbd_snap_show, NULL);
static DEVICE_ATTR(snap_id, 0444, rbd_snap_id_show, NULL);
static DEVICE_ATTR(parent, 0444, rbd_parent_show, NULL);
4255 4256 4257

static struct attribute *rbd_attrs[] = {
	&dev_attr_size.attr,
A
Alex Elder 已提交
4258
	&dev_attr_features.attr,
4259
	&dev_attr_major.attr,
4260
	&dev_attr_minor.attr,
4261
	&dev_attr_client_addr.attr,
4262
	&dev_attr_client_id.attr,
4263
	&dev_attr_cluster_fsid.attr,
4264
	&dev_attr_config_info.attr,
4265
	&dev_attr_pool.attr,
4266
	&dev_attr_pool_id.attr,
4267
	&dev_attr_pool_ns.attr,
4268
	&dev_attr_name.attr,
A
Alex Elder 已提交
4269
	&dev_attr_image_id.attr,
4270
	&dev_attr_current_snap.attr,
4271
	&dev_attr_snap_id.attr,
4272
	&dev_attr_parent.attr,
4273 4274 4275 4276 4277 4278 4279 4280 4281 4282 4283 4284 4285
	&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
};

4286
static void rbd_dev_release(struct device *dev);
4287

4288
static const struct device_type rbd_device_type = {
4289 4290
	.name		= "rbd",
	.groups		= rbd_attr_groups,
4291
	.release	= rbd_dev_release,
4292 4293
};

4294 4295 4296 4297 4298 4299 4300 4301 4302 4303 4304 4305 4306 4307 4308 4309 4310 4311 4312 4313 4314
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;
4315 4316 4317

	spec->pool_id = CEPH_NOPOOL;
	spec->snap_id = CEPH_NOSNAP;
4318 4319 4320 4321 4322 4323 4324 4325 4326 4327
	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);
4328
	kfree(spec->pool_ns);
4329 4330 4331 4332 4333 4334
	kfree(spec->image_id);
	kfree(spec->image_name);
	kfree(spec->snap_name);
	kfree(spec);
}

4335
static void rbd_dev_free(struct rbd_device *rbd_dev)
4336
{
4337
	WARN_ON(rbd_dev->watch_state != RBD_WATCH_STATE_UNREGISTERED);
4338
	WARN_ON(rbd_dev->lock_state != RBD_LOCK_STATE_UNLOCKED);
4339

4340
	ceph_oid_destroy(&rbd_dev->header_oid);
4341
	ceph_oloc_destroy(&rbd_dev->header_oloc);
4342
	kfree(rbd_dev->config_info);
4343

4344 4345 4346 4347
	rbd_put_client(rbd_dev->rbd_client);
	rbd_spec_put(rbd_dev->spec);
	kfree(rbd_dev->opts);
	kfree(rbd_dev);
4348 4349 4350 4351 4352 4353 4354 4355 4356 4357 4358 4359 4360
}

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);
4361 4362 4363 4364 4365 4366 4367 4368 4369 4370

	/*
	 * 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);
}

4371 4372
static struct rbd_device *__rbd_dev_create(struct rbd_client *rbdc,
					   struct rbd_spec *spec)
4373 4374 4375
{
	struct rbd_device *rbd_dev;

4376
	rbd_dev = kzalloc(sizeof(*rbd_dev), GFP_KERNEL);
4377 4378 4379 4380 4381 4382 4383
	if (!rbd_dev)
		return NULL;

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

4384
	rbd_dev->header.data_pool_id = CEPH_NOPOOL;
4385
	ceph_oid_init(&rbd_dev->header_oid);
4386
	rbd_dev->header_oloc.pool = spec->pool_id;
4387 4388 4389 4390 4391 4392
	if (spec->pool_ns) {
		WARN_ON(!*spec->pool_ns);
		rbd_dev->header_oloc.pool_ns =
		    ceph_find_or_create_string(spec->pool_ns,
					       strlen(spec->pool_ns));
	}
4393

4394 4395 4396 4397
	mutex_init(&rbd_dev->watch_mutex);
	rbd_dev->watch_state = RBD_WATCH_STATE_UNREGISTERED;
	INIT_DELAYED_WORK(&rbd_dev->watch_dwork, rbd_reregister_watch);

4398 4399 4400 4401 4402 4403 4404 4405
	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);

4406 4407 4408 4409 4410
	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);

4411
	rbd_dev->rbd_client = rbdc;
4412
	rbd_dev->spec = spec;
4413

4414 4415 4416 4417 4418 4419 4420 4421 4422 4423 4424 4425 4426 4427 4428 4429 4430 4431 4432 4433 4434 4435 4436 4437 4438 4439 4440 4441 4442 4443
	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;
4444

4445 4446
	/* we have a ref from do_rbd_add() */
	__module_get(THIS_MODULE);
4447

4448
	dout("%s rbd_dev %p dev_id %d\n", __func__, rbd_dev, rbd_dev->dev_id);
4449
	return rbd_dev;
4450 4451 4452 4453 4454 4455

fail_dev_id:
	ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id);
fail_rbd_dev:
	rbd_dev_free(rbd_dev);
	return NULL;
4456 4457 4458 4459
}

static void rbd_dev_destroy(struct rbd_device *rbd_dev)
{
4460 4461
	if (rbd_dev)
		put_device(&rbd_dev->dev);
4462 4463
}

4464 4465 4466 4467 4468 4469 4470 4471 4472 4473 4474 4475 4476 4477 4478
/*
 * 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 };

4479 4480 4481 4482
	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));
4483
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4484 4485
	if (ret < 0)
		return ret;
4486 4487
	if (ret < sizeof (size_buf))
		return -ERANGE;
4488

J
Josh Durgin 已提交
4489
	if (order) {
4490
		*order = size_buf.order;
J
Josh Durgin 已提交
4491 4492
		dout("  order %u", (unsigned int)*order);
	}
4493 4494
	*snap_size = le64_to_cpu(size_buf.size);

J
Josh Durgin 已提交
4495 4496
	dout("  snap_id 0x%016llx snap_size = %llu\n",
		(unsigned long long)snap_id,
4497
		(unsigned long long)*snap_size);
4498 4499 4500 4501 4502 4503 4504 4505 4506 4507 4508

	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);
}

4509 4510 4511 4512 4513 4514 4515 4516 4517 4518
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;

4519 4520 4521
	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);
4522
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4523 4524 4525 4526 4527
	if (ret < 0)
		goto out;

	p = reply_buf;
	rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p,
4528 4529
						p + ret, NULL, GFP_NOIO);
	ret = 0;
4530 4531 4532 4533 4534 4535 4536 4537 4538 4539 4540 4541 4542

	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;
}

4543 4544 4545 4546 4547 4548 4549
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;
4550
	} __attribute__ ((packed)) features_buf = { 0 };
4551
	u64 unsup;
4552 4553
	int ret;

4554 4555 4556 4557
	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));
4558
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4559 4560
	if (ret < 0)
		return ret;
4561 4562
	if (ret < sizeof (features_buf))
		return -ERANGE;
A
Alex Elder 已提交
4563

4564 4565 4566 4567
	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
4568
		return -ENXIO;
4569
	}
A
Alex Elder 已提交
4570

4571 4572 4573
	*snap_features = le64_to_cpu(features_buf.features);

	dout("  snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
4574 4575 4576
		(unsigned long long)snap_id,
		(unsigned long long)*snap_features,
		(unsigned long long)le64_to_cpu(features_buf.incompat));
4577 4578 4579 4580 4581 4582 4583 4584 4585 4586

	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);
}

4587 4588 4589 4590 4591 4592 4593 4594
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
4595
	u64 pool_id;
4596
	char *image_id;
4597
	u64 snap_id;
4598 4599 4600 4601 4602 4603 4604 4605 4606 4607 4608 4609 4610 4611 4612 4613 4614
	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;
	}

4615
	snapid = cpu_to_le64(rbd_dev->spec->snap_id);
4616 4617 4618
	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
				  &rbd_dev->header_oloc, "get_parent",
				  &snapid, sizeof(snapid), reply_buf, size);
4619
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4620 4621 4622 4623
	if (ret < 0)
		goto out_err;

	p = reply_buf;
4624 4625
	end = reply_buf + ret;
	ret = -ERANGE;
A
Alex Elder 已提交
4626
	ceph_decode_64_safe(&p, end, pool_id, out_err);
4627 4628 4629 4630 4631 4632 4633 4634 4635 4636 4637 4638 4639 4640 4641 4642 4643
	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);
		}

4644
		goto out;	/* No parent?  No problem. */
4645
	}
4646

4647 4648 4649
	/* The ceph file layout needs to fit pool id in 32 bits */

	ret = -EIO;
A
Alex Elder 已提交
4650
	if (pool_id > (u64)U32_MAX) {
4651
		rbd_warn(NULL, "parent pool id too large (%llu > %u)",
A
Alex Elder 已提交
4652
			(unsigned long long)pool_id, U32_MAX);
4653
		goto out_err;
A
Alex Elder 已提交
4654
	}
4655

A
Alex Elder 已提交
4656
	image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
4657 4658 4659 4660
	if (IS_ERR(image_id)) {
		ret = PTR_ERR(image_id);
		goto out_err;
	}
4661
	ceph_decode_64_safe(&p, end, snap_id, out_err);
4662 4663
	ceph_decode_64_safe(&p, end, overlap, out_err);

4664 4665 4666 4667 4668 4669 4670 4671 4672
	/*
	 * 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;
4673 4674 4675 4676 4677 4678 4679 4680 4681 4682 4683

		/* TODO: support cloning across namespaces */
		if (rbd_dev->spec->pool_ns) {
			parent_spec->pool_ns = kstrdup(rbd_dev->spec->pool_ns,
						       GFP_KERNEL);
			if (!parent_spec->pool_ns) {
				ret = -ENOMEM;
				goto out_err;
			}
		}

A
Alex Elder 已提交
4684 4685
		rbd_dev->parent_spec = parent_spec;
		parent_spec = NULL;	/* rbd_dev now owns this */
4686 4687
	} else {
		kfree(image_id);
4688 4689 4690
	}

	/*
4691 4692
	 * We always update the parent overlap.  If it's zero we issue
	 * a warning, as we will proceed as if there was no parent.
4693 4694 4695
	 */
	if (!overlap) {
		if (parent_spec) {
4696 4697 4698 4699
			/* refresh, careful to warn just once */
			if (rbd_dev->parent_overlap)
				rbd_warn(rbd_dev,
				    "clone now standalone (overlap became 0)");
4700
		} else {
4701 4702
			/* initial probe */
			rbd_warn(rbd_dev, "clone is standalone (overlap 0)");
4703
		}
A
Alex Elder 已提交
4704
	}
4705 4706
	rbd_dev->parent_overlap = overlap;

4707 4708 4709 4710 4711 4712 4713 4714 4715
out:
	ret = 0;
out_err:
	kfree(reply_buf);
	rbd_spec_put(parent_spec);

	return ret;
}

4716 4717 4718 4719 4720 4721 4722 4723 4724 4725
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;

4726 4727 4728
	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);
4729 4730 4731 4732 4733 4734 4735
	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
4736 4737
	rbd_dev->header.stripe_unit = ceph_decode_64(&p);
	rbd_dev->header.stripe_count = ceph_decode_64(&p);
4738 4739 4740
	return 0;
}

4741 4742 4743 4744 4745 4746 4747 4748 4749 4750 4751 4752 4753 4754 4755 4756 4757 4758
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;
}

4759 4760
static char *rbd_dev_image_name(struct rbd_device *rbd_dev)
{
4761
	CEPH_DEFINE_OID_ONSTACK(oid);
4762 4763 4764 4765 4766 4767 4768 4769 4770 4771 4772 4773
	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
4774 4775
	len = strlen(rbd_dev->spec->image_id);
	image_id_size = sizeof (__le32) + len;
4776 4777 4778 4779 4780
	image_id = kmalloc(image_id_size, GFP_KERNEL);
	if (!image_id)
		return NULL;

	p = image_id;
4781
	end = image_id + image_id_size;
4782
	ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len);
4783 4784 4785 4786 4787 4788

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

4789 4790 4791 4792
	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);
4793 4794 4795
	if (ret < 0)
		goto out;
	p = reply_buf;
4796 4797
	end = reply_buf + ret;

4798 4799 4800 4801 4802 4803 4804 4805 4806 4807 4808 4809
	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;
}

4810 4811 4812 4813 4814 4815 4816 4817 4818 4819 4820 4821 4822 4823 4824 4825 4826 4827 4828 4829 4830 4831 4832 4833 4834 4835 4836 4837 4838 4839
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);
4840 4841 4842 4843 4844 4845 4846
		if (IS_ERR(snap_name)) {
			/* ignore no-longer existing snapshots */
			if (PTR_ERR(snap_name) == -ENOENT)
				continue;
			else
				break;
		}
4847 4848 4849 4850 4851 4852 4853 4854 4855 4856 4857 4858 4859 4860 4861 4862 4863 4864
		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);
}

4865
/*
4866 4867 4868 4869 4870 4871 4872 4873 4874 4875 4876 4877 4878 4879 4880 4881 4882 4883 4884 4885 4886 4887 4888 4889 4890 4891 4892
 * 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.
4893
 *
4894 4895
 * All names in an rbd spec are dynamically allocated.  It's OK if we
 * can't figure out the name for an image id.
4896
 */
4897
static int rbd_spec_fill_names(struct rbd_device *rbd_dev)
4898
{
4899 4900 4901 4902 4903
	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;
4904 4905
	int ret;

4906 4907 4908
	rbd_assert(spec->pool_id != CEPH_NOPOOL);
	rbd_assert(spec->image_id);
	rbd_assert(spec->snap_id != CEPH_NOSNAP);
4909

4910
	/* Get the pool name; we have to make our own copy of this */
4911

4912 4913 4914
	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);
4915 4916
		return -EIO;
	}
4917 4918
	pool_name = kstrdup(pool_name, GFP_KERNEL);
	if (!pool_name)
4919 4920 4921 4922
		return -ENOMEM;

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

4923 4924
	image_name = rbd_dev_image_name(rbd_dev);
	if (!image_name)
A
Alex Elder 已提交
4925
		rbd_warn(rbd_dev, "unable to get image name");
4926

4927
	/* Fetch the snapshot name */
4928

4929
	snap_name = rbd_snap_name(rbd_dev, spec->snap_id);
4930 4931
	if (IS_ERR(snap_name)) {
		ret = PTR_ERR(snap_name);
4932
		goto out_err;
4933 4934 4935 4936 4937
	}

	spec->pool_name = pool_name;
	spec->image_name = image_name;
	spec->snap_name = snap_name;
4938 4939

	return 0;
4940

4941
out_err:
4942 4943
	kfree(image_name);
	kfree(pool_name);
4944 4945 4946
	return ret;
}

A
Alex Elder 已提交
4947
static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev)
4948 4949 4950 4951 4952 4953 4954 4955 4956 4957 4958 4959 4960 4961 4962 4963 4964 4965 4966 4967 4968 4969 4970
{
	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;

4971 4972 4973
	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
				  &rbd_dev->header_oloc, "get_snapcontext",
				  NULL, 0, reply_buf, size);
4974
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4975 4976 4977 4978
	if (ret < 0)
		goto out;

	p = reply_buf;
4979 4980
	end = reply_buf + ret;
	ret = -ERANGE;
4981 4982 4983 4984 4985 4986 4987 4988 4989 4990 4991 4992 4993 4994 4995 4996
	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;
4997
	ret = 0;
4998

4999
	snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
5000 5001 5002 5003 5004 5005 5006 5007
	if (!snapc) {
		ret = -ENOMEM;
		goto out;
	}
	snapc->seq = seq;
	for (i = 0; i < snap_count; i++)
		snapc->snaps[i] = ceph_decode_64(&p);

5008
	ceph_put_snap_context(rbd_dev->header.snapc);
5009 5010 5011
	rbd_dev->header.snapc = snapc;

	dout("  snap context seq = %llu, snap_count = %u\n",
5012
		(unsigned long long)seq, (unsigned int)snap_count);
5013 5014 5015
out:
	kfree(reply_buf);

5016
	return ret;
5017 5018
}

5019 5020
static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
					u64 snap_id)
A
Alex Elder 已提交
5021 5022 5023
{
	size_t size;
	void *reply_buf;
5024
	__le64 snapid;
A
Alex Elder 已提交
5025 5026 5027 5028 5029 5030 5031 5032 5033 5034
	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);

5035
	snapid = cpu_to_le64(snap_id);
5036 5037 5038
	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
				  &rbd_dev->header_oloc, "get_snapshot_name",
				  &snapid, sizeof(snapid), reply_buf, size);
5039
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
5040 5041
	if (ret < 0) {
		snap_name = ERR_PTR(ret);
A
Alex Elder 已提交
5042
		goto out;
5043
	}
A
Alex Elder 已提交
5044 5045

	p = reply_buf;
5046
	end = reply_buf + ret;
A
Alex Elder 已提交
5047
	snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
5048
	if (IS_ERR(snap_name))
A
Alex Elder 已提交
5049 5050
		goto out;

5051
	dout("  snap_id 0x%016llx snap_name = %s\n",
5052
		(unsigned long long)snap_id, snap_name);
A
Alex Elder 已提交
5053 5054 5055
out:
	kfree(reply_buf);

5056
	return snap_name;
A
Alex Elder 已提交
5057 5058
}

5059
static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev)
A
Alex Elder 已提交
5060
{
5061
	bool first_time = rbd_dev->header.object_prefix == NULL;
A
Alex Elder 已提交
5062 5063
	int ret;

5064 5065
	ret = rbd_dev_v2_image_size(rbd_dev);
	if (ret)
5066
		return ret;
5067

5068 5069 5070
	if (first_time) {
		ret = rbd_dev_v2_header_onetime(rbd_dev);
		if (ret)
5071
			return ret;
5072 5073
	}

A
Alex Elder 已提交
5074
	ret = rbd_dev_v2_snap_context(rbd_dev);
5075 5076 5077 5078
	if (ret && first_time) {
		kfree(rbd_dev->header.object_prefix);
		rbd_dev->header.object_prefix = NULL;
	}
A
Alex Elder 已提交
5079 5080 5081 5082

	return ret;
}

5083 5084 5085 5086 5087 5088 5089 5090 5091 5092
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);
}

5093 5094 5095
/*
 * 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
5096 5097
 * the token (string of non-white space characters) found.  Note
 * that *buf must be terminated with '\0'.
5098 5099 5100 5101 5102 5103 5104 5105 5106 5107 5108 5109 5110 5111
 */
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
5112 5113 5114 5115 5116 5117 5118 5119 5120 5121 5122 5123 5124 5125 5126 5127 5128 5129 5130 5131 5132 5133
/*
 * 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
5134
	dup = kmemdup(*buf, len + 1, GFP_KERNEL);
A
Alex Elder 已提交
5135 5136 5137 5138 5139 5140 5141 5142 5143 5144 5145
	if (!dup)
		return NULL;
	*(dup + len) = '\0';
	*buf += len;

	if (lenp)
		*lenp = len;

	return dup;
}

5146
/*
5147 5148 5149 5150
 * 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
5151
 *
5152 5153 5154 5155 5156 5157 5158 5159 5160 5161 5162 5163 5164 5165 5166 5167 5168 5169 5170 5171 5172 5173 5174 5175 5176 5177 5178 5179 5180 5181 5182 5183 5184 5185
 * 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.
5186
 */
5187
static int rbd_add_parse_args(const char *buf,
5188
				struct ceph_options **ceph_opts,
5189 5190
				struct rbd_options **opts,
				struct rbd_spec **rbd_spec)
5191
{
A
Alex Elder 已提交
5192
	size_t len;
5193
	char *options;
5194
	const char *mon_addrs;
5195
	char *snap_name;
5196
	size_t mon_addrs_size;
5197
	struct parse_rbd_opts_ctx pctx = { 0 };
5198
	struct ceph_options *copts;
5199
	int ret;
5200 5201 5202

	/* The first four tokens are required */

5203
	len = next_token(&buf);
5204 5205 5206 5207
	if (!len) {
		rbd_warn(NULL, "no monitor address(es) provided");
		return -EINVAL;
	}
5208
	mon_addrs = buf;
5209
	mon_addrs_size = len + 1;
5210
	buf += len;
5211

5212
	ret = -EINVAL;
5213 5214
	options = dup_token(&buf, NULL);
	if (!options)
5215
		return -ENOMEM;
5216 5217 5218 5219
	if (!*options) {
		rbd_warn(NULL, "no options provided");
		goto out_err;
	}
5220

5221 5222
	pctx.spec = rbd_spec_alloc();
	if (!pctx.spec)
5223
		goto out_mem;
5224

5225 5226
	pctx.spec->pool_name = dup_token(&buf, NULL);
	if (!pctx.spec->pool_name)
5227
		goto out_mem;
5228
	if (!*pctx.spec->pool_name) {
5229 5230 5231
		rbd_warn(NULL, "no pool name provided");
		goto out_err;
	}
5232

5233 5234
	pctx.spec->image_name = dup_token(&buf, NULL);
	if (!pctx.spec->image_name)
5235
		goto out_mem;
5236
	if (!*pctx.spec->image_name) {
5237 5238 5239
		rbd_warn(NULL, "no image name provided");
		goto out_err;
	}
5240

5241 5242 5243 5244
	/*
	 * Snapshot name is optional; default is to use "-"
	 * (indicating the head/no snapshot).
	 */
5245
	len = next_token(&buf);
5246
	if (!len) {
5247 5248
		buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */
		len = sizeof (RBD_SNAP_HEAD_NAME) - 1;
5249
	} else if (len > RBD_MAX_SNAP_NAME_LEN) {
5250
		ret = -ENAMETOOLONG;
5251
		goto out_err;
5252
	}
5253 5254
	snap_name = kmemdup(buf, len + 1, GFP_KERNEL);
	if (!snap_name)
5255
		goto out_mem;
5256
	*(snap_name + len) = '\0';
5257
	pctx.spec->snap_name = snap_name;
A
Alex Elder 已提交
5258

5259
	/* Initialize all rbd options to the defaults */
5260

5261 5262
	pctx.opts = kzalloc(sizeof(*pctx.opts), GFP_KERNEL);
	if (!pctx.opts)
5263 5264
		goto out_mem;

5265 5266 5267 5268 5269 5270
	pctx.opts->read_only = RBD_READ_ONLY_DEFAULT;
	pctx.opts->queue_depth = RBD_QUEUE_DEPTH_DEFAULT;
	pctx.opts->lock_timeout = RBD_LOCK_TIMEOUT_DEFAULT;
	pctx.opts->lock_on_read = RBD_LOCK_ON_READ_DEFAULT;
	pctx.opts->exclusive = RBD_EXCLUSIVE_DEFAULT;
	pctx.opts->trim = RBD_TRIM_DEFAULT;
A
Alex Elder 已提交
5271

5272
	copts = ceph_parse_options(options, mon_addrs,
5273 5274
				   mon_addrs + mon_addrs_size - 1,
				   parse_rbd_opts_token, &pctx);
5275 5276
	if (IS_ERR(copts)) {
		ret = PTR_ERR(copts);
5277 5278
		goto out_err;
	}
5279 5280 5281
	kfree(options);

	*ceph_opts = copts;
5282 5283
	*opts = pctx.opts;
	*rbd_spec = pctx.spec;
5284

5285
	return 0;
5286
out_mem:
5287
	ret = -ENOMEM;
A
Alex Elder 已提交
5288
out_err:
5289 5290
	kfree(pctx.opts);
	rbd_spec_put(pctx.spec);
5291
	kfree(options);
A
Alex Elder 已提交
5292

5293
	return ret;
5294 5295
}

I
Ilya Dryomov 已提交
5296 5297 5298 5299 5300 5301 5302 5303 5304 5305
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)
{
5306 5307
	int ret;

I
Ilya Dryomov 已提交
5308 5309 5310 5311 5312 5313 5314
	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);
5315
	ret = rbd_wait_state_locked(rbd_dev, true);
I
Ilya Dryomov 已提交
5316
	up_read(&rbd_dev->lock_rwsem);
5317
	if (ret) {
I
Ilya Dryomov 已提交
5318 5319 5320 5321 5322 5323 5324
		rbd_warn(rbd_dev, "failed to acquire exclusive lock");
		return -EROFS;
	}

	return 0;
}

A
Alex Elder 已提交
5325 5326 5327 5328 5329 5330 5331 5332 5333 5334 5335 5336 5337 5338 5339 5340 5341 5342
/*
 * 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;
5343
	CEPH_DEFINE_OID_ONSTACK(oid);
A
Alex Elder 已提交
5344
	void *response;
5345
	char *image_id;
5346

A
Alex Elder 已提交
5347 5348 5349
	/*
	 * When probing a parent image, the image id is already
	 * known (and the image name likely is not).  There's no
5350 5351
	 * 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
5352
	 */
5353 5354 5355
	if (rbd_dev->spec->image_id) {
		rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1;

A
Alex Elder 已提交
5356
		return 0;
5357
	}
A
Alex Elder 已提交
5358

A
Alex Elder 已提交
5359 5360 5361 5362
	/*
	 * First, see if the format 2 image id file exists, and if
	 * so, get the image's persistent id from it.
	 */
5363 5364 5365 5366 5367 5368
	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
5369 5370 5371 5372 5373 5374 5375 5376 5377 5378

	/* 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;
	}

5379 5380
	/* If it doesn't exist we'll assume it's a format 1 image */

5381 5382 5383
	ret = rbd_obj_method_sync(rbd_dev, &oid, &rbd_dev->header_oloc,
				  "get_id", NULL, 0,
				  response, RBD_IMAGE_ID_LEN_MAX);
5384
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
5385 5386 5387 5388 5389
	if (ret == -ENOENT) {
		image_id = kstrdup("", GFP_KERNEL);
		ret = image_id ? 0 : -ENOMEM;
		if (!ret)
			rbd_dev->image_format = 1;
5390
	} else if (ret >= 0) {
5391 5392 5393
		void *p = response;

		image_id = ceph_extract_encoded_string(&p, p + ret,
A
Alex Elder 已提交
5394
						NULL, GFP_NOIO);
5395
		ret = PTR_ERR_OR_ZERO(image_id);
5396 5397 5398 5399 5400 5401 5402
		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
5403 5404 5405
	}
out:
	kfree(response);
5406
	ceph_oid_destroy(&oid);
A
Alex Elder 已提交
5407 5408 5409
	return ret;
}

A
Alex Elder 已提交
5410 5411 5412 5413
/*
 * Undo whatever state changes are made by v1 or v2 header info
 * call.
 */
A
Alex Elder 已提交
5414 5415 5416 5417
static void rbd_dev_unprobe(struct rbd_device *rbd_dev)
{
	struct rbd_image_header	*header;

5418
	rbd_dev_parent_put(rbd_dev);
A
Alex Elder 已提交
5419 5420 5421 5422

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

	header = &rbd_dev->header;
5423
	ceph_put_snap_context(header->snapc);
A
Alex Elder 已提交
5424 5425 5426 5427 5428 5429
	kfree(header->snap_sizes);
	kfree(header->snap_names);
	kfree(header->object_prefix);
	memset(header, 0, sizeof (*header));
}

5430
static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev)
5431 5432 5433
{
	int ret;

5434
	ret = rbd_dev_v2_object_prefix(rbd_dev);
5435
	if (ret)
5436 5437
		goto out_err;

5438 5439 5440 5441
	/*
	 * Get the and check features for the image.  Currently the
	 * features are assumed to never change.
	 */
5442
	ret = rbd_dev_v2_features(rbd_dev);
5443
	if (ret)
5444
		goto out_err;
5445

5446 5447 5448 5449 5450 5451 5452
	/* 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;
	}
5453

5454 5455 5456 5457 5458 5459
	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
5460
	rbd_init_layout(rbd_dev);
A
Alex Elder 已提交
5461
	return 0;
I
Ilya Dryomov 已提交
5462

5463
out_err:
A
Alex Elder 已提交
5464
	rbd_dev->header.features = 0;
5465 5466
	kfree(rbd_dev->header.object_prefix);
	rbd_dev->header.object_prefix = NULL;
5467
	return ret;
5468 5469
}

5470 5471 5472 5473 5474 5475
/*
 * @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
5476
{
5477
	struct rbd_device *parent = NULL;
5478 5479 5480 5481 5482
	int ret;

	if (!rbd_dev->parent_spec)
		return 0;

5483 5484 5485 5486 5487 5488
	if (++depth > RBD_MAX_PARENT_CHAIN_LEN) {
		pr_info("parent chain is too long (%d)\n", depth);
		ret = -EINVAL;
		goto out_err;
	}

5489
	parent = __rbd_dev_create(rbd_dev->rbd_client, rbd_dev->parent_spec);
5490 5491
	if (!parent) {
		ret = -ENOMEM;
5492
		goto out_err;
5493 5494 5495 5496 5497 5498 5499 5500
	}

	/*
	 * 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);
5501

5502
	ret = rbd_dev_image_probe(parent, depth);
5503 5504
	if (ret < 0)
		goto out_err;
5505

5506
	rbd_dev->parent = parent;
5507
	atomic_set(&rbd_dev->parent_ref, 1);
5508
	return 0;
5509

5510
out_err:
5511
	rbd_dev_unparent(rbd_dev);
5512
	rbd_dev_destroy(parent);
5513 5514 5515
	return ret;
}

5516 5517 5518 5519 5520 5521 5522 5523 5524
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
5525 5526 5527 5528
/*
 * rbd_dev->header_rwsem must be locked for write and will be unlocked
 * upon return.
 */
5529
static int rbd_dev_device_setup(struct rbd_device *rbd_dev)
5530
{
A
Alex Elder 已提交
5531
	int ret;
A
Alex Elder 已提交
5532

5533
	/* Record our major and minor device numbers. */
A
Alex Elder 已提交
5534

5535 5536 5537
	if (!single_major) {
		ret = register_blkdev(0, rbd_dev->name);
		if (ret < 0)
5538
			goto err_out_unlock;
5539 5540 5541 5542 5543 5544 5545

		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
5546 5547 5548 5549 5550 5551 5552

	/* Set up the blkdev mapping. */

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

5553
	ret = rbd_dev_mapping_set(rbd_dev);
A
Alex Elder 已提交
5554 5555
	if (ret)
		goto err_out_disk;
I
Ilya Dryomov 已提交
5556

5557
	set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE);
5558
	set_disk_ro(rbd_dev->disk, rbd_dev->opts->read_only);
5559

5560
	ret = dev_set_name(&rbd_dev->dev, "%d", rbd_dev->dev_id);
5561
	if (ret)
5562
		goto err_out_mapping;
A
Alex Elder 已提交
5563

5564
	set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
I
Ilya Dryomov 已提交
5565
	up_write(&rbd_dev->header_rwsem);
5566
	return 0;
5567

5568 5569
err_out_mapping:
	rbd_dev_mapping_clear(rbd_dev);
A
Alex Elder 已提交
5570 5571 5572
err_out_disk:
	rbd_free_disk(rbd_dev);
err_out_blkdev:
5573 5574
	if (!single_major)
		unregister_blkdev(rbd_dev->major, rbd_dev->name);
I
Ilya Dryomov 已提交
5575 5576
err_out_unlock:
	up_write(&rbd_dev->header_rwsem);
A
Alex Elder 已提交
5577 5578 5579
	return ret;
}

A
Alex Elder 已提交
5580 5581 5582
static int rbd_dev_header_name(struct rbd_device *rbd_dev)
{
	struct rbd_spec *spec = rbd_dev->spec;
5583
	int ret;
A
Alex Elder 已提交
5584 5585 5586 5587 5588

	/* 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)
5589 5590
		ret = ceph_oid_aprintf(&rbd_dev->header_oid, GFP_KERNEL, "%s%s",
				       spec->image_name, RBD_SUFFIX);
A
Alex Elder 已提交
5591
	else
5592 5593
		ret = ceph_oid_aprintf(&rbd_dev->header_oid, GFP_KERNEL, "%s%s",
				       RBD_HEADER_PREFIX, spec->image_id);
A
Alex Elder 已提交
5594

5595
	return ret;
A
Alex Elder 已提交
5596 5597
}

5598 5599
static void rbd_dev_image_release(struct rbd_device *rbd_dev)
{
A
Alex Elder 已提交
5600
	rbd_dev_unprobe(rbd_dev);
5601 5602
	if (rbd_dev->opts)
		rbd_unregister_watch(rbd_dev);
A
Alex Elder 已提交
5603 5604 5605
	rbd_dev->image_format = 0;
	kfree(rbd_dev->spec->image_id);
	rbd_dev->spec->image_id = NULL;
5606 5607
}

5608 5609
/*
 * Probe for the existence of the header object for the given rbd
5610 5611 5612
 * 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.
5613
 */
5614
static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth)
5615 5616 5617 5618
{
	int ret;

	/*
A
Alex Elder 已提交
5619 5620 5621 5622
	 * 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.
5623 5624 5625
	 */
	ret = rbd_dev_image_id(rbd_dev);
	if (ret)
5626 5627
		return ret;

A
Alex Elder 已提交
5628 5629 5630 5631
	ret = rbd_dev_header_name(rbd_dev);
	if (ret)
		goto err_out_format;

5632
	if (!depth) {
5633
		ret = rbd_register_watch(rbd_dev);
5634 5635
		if (ret) {
			if (ret == -ENOENT)
5636
				pr_info("image %s/%s%s%s does not exist\n",
5637
					rbd_dev->spec->pool_name,
5638 5639
					rbd_dev->spec->pool_ns ?: "",
					rbd_dev->spec->pool_ns ? "/" : "",
5640
					rbd_dev->spec->image_name);
5641
			goto err_out_format;
5642
		}
5643
	}
5644

5645
	ret = rbd_dev_header_info(rbd_dev);
5646
	if (ret)
5647
		goto err_out_watch;
A
Alex Elder 已提交
5648

5649 5650 5651 5652 5653 5654
	/*
	 * 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.
	 */
5655
	if (!depth)
5656 5657 5658
		ret = rbd_spec_fill_snap_id(rbd_dev);
	else
		ret = rbd_spec_fill_names(rbd_dev);
5659 5660
	if (ret) {
		if (ret == -ENOENT)
5661
			pr_info("snap %s/%s%s%s@%s does not exist\n",
5662
				rbd_dev->spec->pool_name,
5663 5664
				rbd_dev->spec->pool_ns ?: "",
				rbd_dev->spec->pool_ns ? "/" : "",
5665 5666
				rbd_dev->spec->image_name,
				rbd_dev->spec->snap_name);
A
Alex Elder 已提交
5667
		goto err_out_probe;
5668
	}
5669

5670 5671 5672 5673 5674 5675 5676 5677 5678
	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.
		 */
5679
		if (!depth && rbd_dev->parent_spec)
5680 5681 5682 5683
			rbd_warn(rbd_dev,
				 "WARNING: kernel layering is EXPERIMENTAL!");
	}

5684
	ret = rbd_dev_probe_parent(rbd_dev, depth);
A
Alex Elder 已提交
5685 5686 5687 5688
	if (ret)
		goto err_out_probe;

	dout("discovered format %u image, header name is %s\n",
5689
		rbd_dev->image_format, rbd_dev->header_oid.name);
A
Alex Elder 已提交
5690
	return 0;
5691

A
Alex Elder 已提交
5692 5693
err_out_probe:
	rbd_dev_unprobe(rbd_dev);
5694
err_out_watch:
5695
	if (!depth)
5696
		rbd_unregister_watch(rbd_dev);
A
Alex Elder 已提交
5697 5698
err_out_format:
	rbd_dev->image_format = 0;
5699 5700
	kfree(rbd_dev->spec->image_id);
	rbd_dev->spec->image_id = NULL;
5701 5702 5703
	return ret;
}

5704 5705 5706
static ssize_t do_rbd_add(struct bus_type *bus,
			  const char *buf,
			  size_t count)
5707
{
5708
	struct rbd_device *rbd_dev = NULL;
5709
	struct ceph_options *ceph_opts = NULL;
5710
	struct rbd_options *rbd_opts = NULL;
5711
	struct rbd_spec *spec = NULL;
5712
	struct rbd_client *rbdc;
5713
	int rc;
5714 5715 5716 5717 5718

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

	/* parse add command */
5719
	rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec);
5720
	if (rc < 0)
5721
		goto out;
5722

5723 5724 5725
	rbdc = rbd_get_client(ceph_opts);
	if (IS_ERR(rbdc)) {
		rc = PTR_ERR(rbdc);
5726
		goto err_out_args;
5727
	}
5728 5729

	/* pick the pool */
5730
	rc = ceph_pg_poolid_by_name(rbdc->client->osdc.osdmap, spec->pool_name);
5731 5732 5733
	if (rc < 0) {
		if (rc == -ENOENT)
			pr_info("pool %s does not exist\n", spec->pool_name);
5734
		goto err_out_client;
5735
	}
A
Alex Elder 已提交
5736
	spec->pool_id = (u64)rc;
5737

5738
	rbd_dev = rbd_dev_create(rbdc, spec, rbd_opts);
5739 5740
	if (!rbd_dev) {
		rc = -ENOMEM;
5741
		goto err_out_client;
5742
	}
5743 5744
	rbdc = NULL;		/* rbd_dev now owns this */
	spec = NULL;		/* rbd_dev now owns this */
5745
	rbd_opts = NULL;	/* rbd_dev now owns this */
5746

5747 5748 5749 5750 5751 5752
	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
5753
	down_write(&rbd_dev->header_rwsem);
5754
	rc = rbd_dev_image_probe(rbd_dev, 0);
5755 5756
	if (rc < 0) {
		up_write(&rbd_dev->header_rwsem);
5757
		goto err_out_rbd_dev;
5758
	}
5759

5760 5761
	/* If we are mapping a snapshot it must be marked read-only */
	if (rbd_dev->spec->snap_id != CEPH_NOSNAP)
5762
		rbd_dev->opts->read_only = true;
5763

5764
	rc = rbd_dev_device_setup(rbd_dev);
5765
	if (rc)
5766
		goto err_out_image_probe;
A
Alex Elder 已提交
5767

I
Ilya Dryomov 已提交
5768 5769 5770 5771
	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
5772 5773
	}

5774 5775 5776 5777
	/* Everything's ready.  Announce the disk to the world. */

	rc = device_add(&rbd_dev->dev);
	if (rc)
I
Ilya Dryomov 已提交
5778
		goto err_out_image_lock;
5779 5780 5781 5782 5783 5784 5785 5786 5787 5788 5789 5790

	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);
5791 5792 5793 5794
	rc = count;
out:
	module_put(THIS_MODULE);
	return rc;
5795

I
Ilya Dryomov 已提交
5796 5797
err_out_image_lock:
	rbd_dev_image_unlock(rbd_dev);
5798 5799
err_out_device_setup:
	rbd_dev_device_release(rbd_dev);
5800 5801
err_out_image_probe:
	rbd_dev_image_release(rbd_dev);
5802 5803
err_out_rbd_dev:
	rbd_dev_destroy(rbd_dev);
5804
err_out_client:
5805
	rbd_put_client(rbdc);
5806
err_out_args:
5807
	rbd_spec_put(spec);
5808
	kfree(rbd_opts);
5809
	goto out;
5810 5811
}

5812 5813 5814 5815 5816 5817 5818 5819 5820 5821 5822 5823 5824 5825 5826 5827 5828
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);
}

5829 5830
static void rbd_dev_remove_parent(struct rbd_device *rbd_dev)
{
A
Alex Elder 已提交
5831
	while (rbd_dev->parent) {
5832 5833 5834 5835 5836 5837 5838 5839 5840 5841 5842 5843
		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
5844
		rbd_assert(second);
5845
		rbd_dev_image_release(second);
5846
		rbd_dev_destroy(second);
A
Alex Elder 已提交
5847 5848 5849 5850
		first->parent = NULL;
		first->parent_overlap = 0;

		rbd_assert(first->parent_spec);
5851 5852 5853 5854 5855
		rbd_spec_put(first->parent_spec);
		first->parent_spec = NULL;
	}
}

5856 5857 5858
static ssize_t do_rbd_remove(struct bus_type *bus,
			     const char *buf,
			     size_t count)
5859 5860
{
	struct rbd_device *rbd_dev = NULL;
5861 5862
	struct list_head *tmp;
	int dev_id;
M
Mike Christie 已提交
5863
	char opt_buf[6];
5864
	bool already = false;
M
Mike Christie 已提交
5865
	bool force = false;
5866
	int ret;
5867

M
Mike Christie 已提交
5868 5869 5870 5871 5872
	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");
5873
		return -EINVAL;
M
Mike Christie 已提交
5874 5875 5876 5877 5878 5879 5880 5881 5882
	}
	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;
		}
	}
5883

5884 5885 5886 5887 5888 5889 5890 5891
	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;
		}
5892
	}
5893 5894
	if (!ret) {
		spin_lock_irq(&rbd_dev->lock);
M
Mike Christie 已提交
5895
		if (rbd_dev->open_count && !force)
5896 5897
			ret = -EBUSY;
		else
5898 5899
			already = test_and_set_bit(RBD_DEV_FLAG_REMOVING,
							&rbd_dev->flags);
5900 5901 5902
		spin_unlock_irq(&rbd_dev->lock);
	}
	spin_unlock(&rbd_dev_list_lock);
5903
	if (ret < 0 || already)
5904
		return ret;
5905

M
Mike Christie 已提交
5906 5907 5908 5909 5910 5911 5912 5913 5914
	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);
	}

5915 5916 5917 5918 5919
	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);
5920

I
Ilya Dryomov 已提交
5921
	rbd_dev_image_unlock(rbd_dev);
5922
	rbd_dev_device_release(rbd_dev);
5923
	rbd_dev_image_release(rbd_dev);
5924
	rbd_dev_destroy(rbd_dev);
5925
	return count;
5926 5927
}

5928 5929 5930 5931 5932 5933 5934 5935 5936 5937 5938 5939 5940 5941 5942 5943 5944
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);
}

5945 5946
/*
 * create control files in sysfs
5947
 * /sys/bus/rbd/...
5948 5949 5950
 */
static int rbd_sysfs_init(void)
{
5951
	int ret;
5952

5953
	ret = device_register(&rbd_root_dev);
A
Alex Elder 已提交
5954
	if (ret < 0)
5955
		return ret;
5956

5957 5958 5959
	ret = bus_register(&rbd_bus_type);
	if (ret < 0)
		device_unregister(&rbd_root_dev);
5960 5961 5962 5963 5964 5965

	return ret;
}

static void rbd_sysfs_cleanup(void)
{
5966
	bus_unregister(&rbd_bus_type);
5967
	device_unregister(&rbd_root_dev);
5968 5969
}

5970 5971 5972
static int rbd_slab_init(void)
{
	rbd_assert(!rbd_img_request_cache);
G
Geliang Tang 已提交
5973
	rbd_img_request_cache = KMEM_CACHE(rbd_img_request, 0);
5974 5975 5976 5977
	if (!rbd_img_request_cache)
		return -ENOMEM;

	rbd_assert(!rbd_obj_request_cache);
G
Geliang Tang 已提交
5978
	rbd_obj_request_cache = KMEM_CACHE(rbd_obj_request, 0);
5979 5980 5981
	if (!rbd_obj_request_cache)
		goto out_err;

5982
	return 0;
5983

5984
out_err:
5985 5986
	kmem_cache_destroy(rbd_img_request_cache);
	rbd_img_request_cache = NULL;
5987 5988 5989 5990 5991
	return -ENOMEM;
}

static void rbd_slab_exit(void)
{
5992 5993 5994 5995
	rbd_assert(rbd_obj_request_cache);
	kmem_cache_destroy(rbd_obj_request_cache);
	rbd_obj_request_cache = NULL;

5996 5997 5998 5999 6000
	rbd_assert(rbd_img_request_cache);
	kmem_cache_destroy(rbd_img_request_cache);
	rbd_img_request_cache = NULL;
}

A
Alex Elder 已提交
6001
static int __init rbd_init(void)
6002 6003 6004
{
	int rc;

6005 6006 6007 6008
	if (!libceph_compatible(NULL)) {
		rbd_warn(NULL, "libceph incompatibility (quitting)");
		return -EINVAL;
	}
I
Ilya Dryomov 已提交
6009

6010
	rc = rbd_slab_init();
6011 6012
	if (rc)
		return rc;
I
Ilya Dryomov 已提交
6013

6014 6015
	/*
	 * The number of active work items is limited by the number of
I
Ilya Dryomov 已提交
6016
	 * rbd devices * queue depth, so leave @max_active at default.
6017 6018 6019 6020 6021 6022 6023
	 */
	rbd_wq = alloc_workqueue(RBD_DRV_NAME, WQ_MEM_RECLAIM, 0);
	if (!rbd_wq) {
		rc = -ENOMEM;
		goto err_out_slab;
	}

6024 6025 6026 6027
	if (single_major) {
		rbd_major = register_blkdev(0, RBD_DRV_NAME);
		if (rbd_major < 0) {
			rc = rbd_major;
6028
			goto err_out_wq;
6029 6030 6031
		}
	}

6032 6033
	rc = rbd_sysfs_init();
	if (rc)
6034 6035 6036 6037 6038 6039
		goto err_out_blkdev;

	if (single_major)
		pr_info("loaded (major %d)\n", rbd_major);
	else
		pr_info("loaded\n");
6040

I
Ilya Dryomov 已提交
6041 6042
	return 0;

6043 6044 6045
err_out_blkdev:
	if (single_major)
		unregister_blkdev(rbd_major, RBD_DRV_NAME);
6046 6047
err_out_wq:
	destroy_workqueue(rbd_wq);
I
Ilya Dryomov 已提交
6048 6049
err_out_slab:
	rbd_slab_exit();
6050
	return rc;
6051 6052
}

A
Alex Elder 已提交
6053
static void __exit rbd_exit(void)
6054
{
I
Ilya Dryomov 已提交
6055
	ida_destroy(&rbd_dev_id_ida);
6056
	rbd_sysfs_cleanup();
6057 6058
	if (single_major)
		unregister_blkdev(rbd_major, RBD_DRV_NAME);
6059
	destroy_workqueue(rbd_wq);
6060
	rbd_slab_exit();
6061 6062 6063 6064 6065
}

module_init(rbd_init);
module_exit(rbd_exit);

A
Alex Elder 已提交
6066
MODULE_AUTHOR("Alex Elder <elder@inktank.com>");
6067 6068 6069 6070 6071
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>");

6072
MODULE_DESCRIPTION("RADOS Block Device (RBD) driver");
6073
MODULE_LICENSE("GPL");