rbd.c 163.6 KB
Newer Older
1

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


   based on drivers/block/osdblk.c:

   Copyright 2009 Red Hat, Inc.

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

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

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



25
   For usage instructions, please refer to:
26

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

 */

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

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

#include "rbd_types.h"

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

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

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

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

	atomic_dec(v);

	return -EINVAL;
}

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

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

	atomic_inc(v);

	return -EINVAL;
}

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

97 98
#define RBD_MINORS_PER_MAJOR		256
#define RBD_SINGLE_MAJOR_PART_SHIFT	4
99

100 101
#define RBD_MAX_PARENT_CHAIN_LEN	16

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

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

#define RBD_SNAP_HEAD_NAME	"-"

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

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

116
#define RBD_OBJ_PREFIX_LEN_MAX	64
A
Alex Elder 已提交
117

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

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

123 124 125 126 127
#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)

128 129
#define RBD_FEATURES_ALL	(RBD_FEATURE_LAYERING |		\
				 RBD_FEATURE_STRIPINGV2 |	\
130 131
				 RBD_FEATURE_EXCLUSIVE_LOCK |	\
				 RBD_FEATURE_DATA_POOL)
A
Alex Elder 已提交
132 133 134

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

135
#define RBD_FEATURES_SUPPORTED	(RBD_FEATURES_ALL)
A
Alex Elder 已提交
136

A
Alex Elder 已提交
137 138 139 140
/*
 * An RBD device name will be "rbd#", where the "rbd" comes from
 * RBD_DRV_NAME above, and # is a unique integer identifier.
 */
141 142 143 144 145 146
#define DEV_NAME_LEN		32

/*
 * block device image metadata (in-memory version)
 */
struct rbd_image_header {
147
	/* These six fields never change for a given rbd image */
148
	char *object_prefix;
149
	__u8 obj_order;
150 151
	u64 stripe_unit;
	u64 stripe_count;
152
	s64 data_pool_id;
153
	u64 features;		/* Might be changeable someday? */
154

A
Alex Elder 已提交
155 156 157
	/* The remaining fields need to be updated occasionally */
	u64 image_size;
	struct ceph_snap_context *snapc;
158 159
	char *snap_names;	/* format 1 only */
	u64 *snap_sizes;	/* format 1 only */
160 161
};

162 163 164 165
/*
 * An rbd image specification.
 *
 * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely
A
Alex Elder 已提交
166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185
 * 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).
186 187 188
 */
struct rbd_spec {
	u64		pool_id;
189
	const char	*pool_name;
190

191 192
	const char	*image_id;
	const char	*image_name;
193 194

	u64		snap_id;
195
	const char	*snap_name;
196 197 198 199

	struct kref	kref;
};

200
/*
A
Alex Elder 已提交
201
 * an instance of the client.  multiple devices may share an rbd client.
202 203 204 205 206 207 208
 */
struct rbd_client {
	struct ceph_client	*client;
	struct kref		kref;
	struct list_head	node;
};

A
Alex Elder 已提交
209 210 211 212 213 214 215 216
struct rbd_img_request;
typedef void (*rbd_img_callback_t)(struct rbd_img_request *);

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

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

217 218 219
enum obj_request_type {
	OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES
};
A
Alex Elder 已提交
220

G
Guangliang Zhao 已提交
221 222 223
enum obj_operation_type {
	OBJ_OP_WRITE,
	OBJ_OP_READ,
224
	OBJ_OP_DISCARD,
G
Guangliang Zhao 已提交
225 226
};

227 228
enum obj_req_flags {
	OBJ_REQ_DONE,		/* completion flag: not done = 0, done = 1 */
229
	OBJ_REQ_IMG_DATA,	/* object usage: standalone = 0, image = 1 */
230 231
	OBJ_REQ_KNOWN,		/* EXISTS flag valid: no = 0, yes = 1 */
	OBJ_REQ_EXISTS,		/* target exists: no = 0, yes = 1 */
232 233
};

A
Alex Elder 已提交
234
struct rbd_obj_request {
235
	u64			object_no;
A
Alex Elder 已提交
236 237
	u64			offset;		/* object start byte */
	u64			length;		/* bytes from offset */
238
	unsigned long		flags;
A
Alex Elder 已提交
239

240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264
	/*
	 * An object request associated with an image will have its
	 * img_data flag set; a standalone object request will not.
	 *
	 * A standalone object request will have which == BAD_WHICH
	 * and a null obj_request pointer.
	 *
	 * An object request initiated in support of a layered image
	 * object (to check for its existence before a write) will
	 * have which == BAD_WHICH and a non-null obj_request pointer.
	 *
	 * Finally, an object request for rbd image data will have
	 * which != BAD_WHICH, and will have a non-null img_request
	 * pointer.  The value of which will be in the range
	 * 0..(img_request->obj_request_count-1).
	 */
	union {
		struct rbd_obj_request	*obj_request;	/* STAT op */
		struct {
			struct rbd_img_request	*img_request;
			u64			img_offset;
			/* links for img_request->obj_requests list */
			struct list_head	links;
		};
	};
A
Alex Elder 已提交
265 266 267
	u32			which;		/* posn image request list */

	enum obj_request_type	type;
268 269 270 271 272 273 274
	union {
		struct bio	*bio_list;
		struct {
			struct page	**pages;
			u32		page_count;
		};
	};
275
	struct page		**copyup_pages;
276
	u32			copyup_page_count;
A
Alex Elder 已提交
277 278 279 280

	struct ceph_osd_request	*osd_req;

	u64			xferred;	/* bytes transferred */
281
	int			result;
A
Alex Elder 已提交
282 283

	rbd_obj_callback_t	callback;
284
	struct completion	completion;
A
Alex Elder 已提交
285 286 287 288

	struct kref		kref;
};

A
Alex Elder 已提交
289
enum img_req_flags {
290 291
	IMG_REQ_WRITE,		/* I/O direction: read = 0, write = 1 */
	IMG_REQ_CHILD,		/* initiator: block = 0, child image = 1 */
292
	IMG_REQ_LAYERED,	/* ENOENT handling: normal = 0, layered = 1 */
293
	IMG_REQ_DISCARD,	/* discard: normal = 0, discard request = 1 */
A
Alex Elder 已提交
294 295
};

A
Alex Elder 已提交
296 297 298 299
struct rbd_img_request {
	struct rbd_device	*rbd_dev;
	u64			offset;	/* starting image byte offset */
	u64			length;	/* byte count from offset */
A
Alex Elder 已提交
300
	unsigned long		flags;
A
Alex Elder 已提交
301
	union {
302
		u64			snap_id;	/* for reads */
A
Alex Elder 已提交
303
		struct ceph_snap_context *snapc;	/* for writes */
304 305 306 307
	};
	union {
		struct request		*rq;		/* block request */
		struct rbd_obj_request	*obj_request;	/* obj req initiator */
A
Alex Elder 已提交
308
	};
309
	struct page		**copyup_pages;
310
	u32			copyup_page_count;
A
Alex Elder 已提交
311 312 313
	spinlock_t		completion_lock;/* protects next_completion */
	u32			next_completion;
	rbd_img_callback_t	callback;
314
	u64			xferred;/* aggregate bytes transferred */
315
	int			result;	/* first nonzero obj_request result */
A
Alex Elder 已提交
316 317 318 319 320 321 322 323

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

	struct kref		kref;
};

#define for_each_obj_request(ireq, oreq) \
324
	list_for_each_entry(oreq, &(ireq)->obj_requests, links)
A
Alex Elder 已提交
325
#define for_each_obj_request_from(ireq, oreq) \
326
	list_for_each_entry_from(oreq, &(ireq)->obj_requests, links)
A
Alex Elder 已提交
327
#define for_each_obj_request_safe(ireq, oreq, n) \
328
	list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links)
A
Alex Elder 已提交
329

330 331 332 333 334 335
enum rbd_watch_state {
	RBD_WATCH_STATE_UNREGISTERED,
	RBD_WATCH_STATE_REGISTERED,
	RBD_WATCH_STATE_ERROR,
};

336 337 338 339 340 341 342 343 344 345 346 347
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
348
struct rbd_mapping {
A
Alex Elder 已提交
349
	u64                     size;
A
Alex Elder 已提交
350
	u64                     features;
A
Alex Elder 已提交
351 352 353
	bool			read_only;
};

354 355 356 357
/*
 * a single device
 */
struct rbd_device {
A
Alex Elder 已提交
358
	int			dev_id;		/* blkdev unique id */
359 360

	int			major;		/* blkdev assigned major */
361
	int			minor;
362 363
	struct gendisk		*disk;		/* blkdev's gendisk and rq */

364
	u32			image_format;	/* Either 1 or 2 */
365 366 367 368
	struct rbd_client	*rbd_client;

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

369
	spinlock_t		lock;		/* queue, flags, open_count */
370 371

	struct rbd_image_header	header;
372
	unsigned long		flags;		/* possibly lock protected */
373
	struct rbd_spec		*spec;
374
	struct rbd_options	*opts;
375
	char			*config_info;	/* add{,_single_major} string */
376

377
	struct ceph_object_id	header_oid;
378
	struct ceph_object_locator header_oloc;
379

380
	struct ceph_file_layout	layout;		/* used for all rbd requests */
381

382 383
	struct mutex		watch_mutex;
	enum rbd_watch_state	watch_state;
384
	struct ceph_osd_linger_request *watch_handle;
385 386
	u64			watch_cookie;
	struct delayed_work	watch_dwork;
387

388 389 390 391 392 393 394 395 396
	struct rw_semaphore	lock_rwsem;
	enum rbd_lock_state	lock_state;
	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;

397
	struct workqueue_struct	*task_wq;
398

399 400
	struct rbd_spec		*parent_spec;
	u64			parent_overlap;
401
	atomic_t		parent_ref;
402
	struct rbd_device	*parent;
403

C
Christoph Hellwig 已提交
404 405 406
	/* Block layer tags. */
	struct blk_mq_tag_set	tag_set;

407 408
	/* protects updating the header */
	struct rw_semaphore     header_rwsem;
A
Alex Elder 已提交
409 410

	struct rbd_mapping	mapping;
411 412

	struct list_head	node;
413 414 415

	/* sysfs related */
	struct device		dev;
416
	unsigned long		open_count;	/* protected by lock */
417 418
};

419
/*
420 421 422 423
 * 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
424
 */
425 426
enum rbd_dev_flags {
	RBD_DEV_FLAG_EXISTS,	/* mapped snapshot has not been deleted */
427
	RBD_DEV_FLAG_REMOVING,	/* this mapping is being removed */
428
	RBD_DEV_FLAG_BLACKLISTED, /* our ceph_client is blacklisted */
429 430
};

431
static DEFINE_MUTEX(client_mutex);	/* Serialize client creation */
432

433
static LIST_HEAD(rbd_dev_list);    /* devices */
434 435
static DEFINE_SPINLOCK(rbd_dev_list_lock);

A
Alex Elder 已提交
436 437
static LIST_HEAD(rbd_client_list);		/* clients */
static DEFINE_SPINLOCK(rbd_client_list_lock);
438

439 440
/* Slab caches for frequently-allocated structures */

441
static struct kmem_cache	*rbd_img_request_cache;
442
static struct kmem_cache	*rbd_obj_request_cache;
443

444
static int rbd_major;
445 446
static DEFINE_IDA(rbd_dev_id_ida);

447 448
static struct workqueue_struct *rbd_wq;

449 450 451 452 453 454 455 456
/*
 * Default to false for now, as single-major requires >= 0.75 version of
 * userspace rbd utility.
 */
static bool single_major = false;
module_param(single_major, bool, S_IRUGO);
MODULE_PARM_DESC(single_major, "Use a single major number for all rbd devices (default: false)");

457 458
static int rbd_img_request_submit(struct rbd_img_request *img_request);

A
Alex Elder 已提交
459 460 461 462
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);
463 464 465 466
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);
467
static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth);
468
static void rbd_spec_put(struct rbd_spec *spec);
A
Alex Elder 已提交
469

470 471
static int rbd_dev_id_to_minor(int dev_id)
{
472
	return dev_id << RBD_SINGLE_MAJOR_PART_SHIFT;
473 474 475 476
}

static int minor_to_rbd_dev_id(int minor)
{
477
	return minor >> RBD_SINGLE_MAJOR_PART_SHIFT;
478 479
}

480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502
static bool rbd_is_lock_supported(struct rbd_device *rbd_dev)
{
	return (rbd_dev->header.features & RBD_FEATURE_EXCLUSIVE_LOCK) &&
	       rbd_dev->spec->snap_id == CEPH_NOSNAP &&
	       !rbd_dev->mapping.read_only;
}

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;
}

503 504 505 506 507
static ssize_t rbd_supported_features_show(struct bus_type *bus, char *buf)
{
	return sprintf(buf, "0x%llx\n", RBD_FEATURES_SUPPORTED);
}

508 509
static BUS_ATTR(add, S_IWUSR, NULL, rbd_add);
static BUS_ATTR(remove, S_IWUSR, NULL, rbd_remove);
510 511
static BUS_ATTR(add_single_major, S_IWUSR, NULL, rbd_add_single_major);
static BUS_ATTR(remove_single_major, S_IWUSR, NULL, rbd_remove_single_major);
512
static BUS_ATTR(supported_features, S_IRUGO, rbd_supported_features_show, NULL);
513 514 515 516

static struct attribute *rbd_bus_attrs[] = {
	&bus_attr_add.attr,
	&bus_attr_remove.attr,
517 518
	&bus_attr_add_single_major.attr,
	&bus_attr_remove_single_major.attr,
519
	&bus_attr_supported_features.attr,
520
	NULL,
A
Alex Elder 已提交
521
};
522 523 524 525

static umode_t rbd_bus_is_visible(struct kobject *kobj,
				  struct attribute *attr, int index)
{
526 527 528 529 530
	if (!single_major &&
	    (attr == &bus_attr_add_single_major.attr ||
	     attr == &bus_attr_remove_single_major.attr))
		return 0;

531 532 533 534 535 536 537 538
	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
539 540 541

static struct bus_type rbd_bus_type = {
	.name		= "rbd",
542
	.bus_groups	= rbd_bus_groups,
A
Alex Elder 已提交
543 544 545 546 547 548 549 550 551 552 553
};

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
554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580
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
581 582 583 584 585 586 587 588 589 590 591 592
#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 */
593

I
Ilya Dryomov 已提交
594
static void rbd_osd_copyup_callback(struct rbd_obj_request *obj_request);
595
static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request);
596 597
static void rbd_img_parent_read(struct rbd_obj_request *obj_request);
static void rbd_dev_remove_parent(struct rbd_device *rbd_dev);
A
Alex Elder 已提交
598

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

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

A
Alex Elder 已提交
615
	if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only)
616 617
		return -EROFS;

618
	spin_lock_irq(&rbd_dev->lock);
619 620 621 622
	if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags))
		removing = true;
	else
		rbd_dev->open_count++;
623
	spin_unlock_irq(&rbd_dev->lock);
624 625 626
	if (removing)
		return -ENOENT;

A
Alex Elder 已提交
627
	(void) get_device(&rbd_dev->dev);
628

629 630 631
	return 0;
}

632
static void rbd_release(struct gendisk *disk, fmode_t mode)
633 634
{
	struct rbd_device *rbd_dev = disk->private_data;
635 636
	unsigned long open_count_before;

637
	spin_lock_irq(&rbd_dev->lock);
638
	open_count_before = rbd_dev->open_count--;
639
	spin_unlock_irq(&rbd_dev->lock);
640
	rbd_assert(open_count_before > 0);
641

A
Alex Elder 已提交
642
	put_device(&rbd_dev->dev);
643 644
}

G
Guangliang Zhao 已提交
645 646
static int rbd_ioctl_set_ro(struct rbd_device *rbd_dev, unsigned long arg)
{
647
	int ret = 0;
G
Guangliang Zhao 已提交
648 649
	int val;
	bool ro;
650
	bool ro_changed = false;
G
Guangliang Zhao 已提交
651

652
	/* get_user() may sleep, so call it before taking rbd_dev->lock */
G
Guangliang Zhao 已提交
653 654 655 656 657 658 659 660
	if (get_user(val, (int __user *)(arg)))
		return -EFAULT;

	ro = val ? true : false;
	/* Snapshot doesn't allow to write*/
	if (rbd_dev->spec->snap_id != CEPH_NOSNAP && !ro)
		return -EROFS;

661 662 663 664 665 666 667
	spin_lock_irq(&rbd_dev->lock);
	/* prevent others open this device */
	if (rbd_dev->open_count > 1) {
		ret = -EBUSY;
		goto out;
	}

G
Guangliang Zhao 已提交
668 669
	if (rbd_dev->mapping.read_only != ro) {
		rbd_dev->mapping.read_only = ro;
670
		ro_changed = true;
G
Guangliang Zhao 已提交
671 672
	}

673 674 675 676 677 678 679
out:
	spin_unlock_irq(&rbd_dev->lock);
	/* set_disk_ro() may sleep, so call it after releasing rbd_dev->lock */
	if (ret == 0 && ro_changed)
		set_disk_ro(rbd_dev->disk, ro ? 1 : 0);

	return ret;
G
Guangliang Zhao 已提交
680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706
}

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;
	int ret = 0;

	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 */

707 708 709
static const struct block_device_operations rbd_bd_ops = {
	.owner			= THIS_MODULE,
	.open			= rbd_open,
710
	.release		= rbd_release,
G
Guangliang Zhao 已提交
711 712 713 714
	.ioctl			= rbd_ioctl,
#ifdef CONFIG_COMPAT
	.compat_ioctl		= rbd_compat_ioctl,
#endif
715 716 717
};

/*
718
 * Initialize an rbd client instance.  Success or not, this function
719
 * consumes ceph_opts.  Caller holds client_mutex.
720
 */
721
static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts)
722 723 724 725
{
	struct rbd_client *rbdc;
	int ret = -ENOMEM;

A
Alex Elder 已提交
726
	dout("%s:\n", __func__);
727 728 729 730 731 732 733
	rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL);
	if (!rbdc)
		goto out_opt;

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

734
	rbdc->client = ceph_create_client(ceph_opts, rbdc);
735
	if (IS_ERR(rbdc->client))
736
		goto out_rbdc;
A
Alex Elder 已提交
737
	ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */
738 739 740

	ret = ceph_open_session(rbdc->client);
	if (ret < 0)
741
		goto out_client;
742

A
Alex Elder 已提交
743
	spin_lock(&rbd_client_list_lock);
744
	list_add_tail(&rbdc->node, &rbd_client_list);
A
Alex Elder 已提交
745
	spin_unlock(&rbd_client_list_lock);
746

A
Alex Elder 已提交
747
	dout("%s: rbdc %p\n", __func__, rbdc);
748

749
	return rbdc;
750
out_client:
751
	ceph_destroy_client(rbdc->client);
752
out_rbdc:
753 754
	kfree(rbdc);
out_opt:
A
Alex Elder 已提交
755 756
	if (ceph_opts)
		ceph_destroy_options(ceph_opts);
A
Alex Elder 已提交
757 758
	dout("%s: error %d\n", __func__, ret);

V
Vasiliy Kulikov 已提交
759
	return ERR_PTR(ret);
760 761
}

762 763 764 765 766 767 768
static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc)
{
	kref_get(&rbdc->kref);

	return rbdc;
}

769
/*
770 771
 * Find a ceph client with specific addr and configuration.  If
 * found, bump its reference count.
772
 */
773
static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts)
774 775
{
	struct rbd_client *client_node;
776
	bool found = false;
777

A
Alex Elder 已提交
778
	if (ceph_opts->flags & CEPH_OPT_NOSHARE)
779 780
		return NULL;

781 782 783
	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)) {
784 785
			__rbd_get_client(client_node);

786 787 788 789 790 791 792
			found = true;
			break;
		}
	}
	spin_unlock(&rbd_client_list_lock);

	return found ? client_node : NULL;
793 794
}

795
/*
796
 * (Per device) rbd map options
797 798
 */
enum {
I
Ilya Dryomov 已提交
799
	Opt_queue_depth,
800 801 802 803
	Opt_last_int,
	/* int args above */
	Opt_last_string,
	/* string args above */
A
Alex Elder 已提交
804 805
	Opt_read_only,
	Opt_read_write,
I
Ilya Dryomov 已提交
806
	Opt_lock_on_read,
807
	Opt_err
808 809
};

A
Alex Elder 已提交
810
static match_table_t rbd_opts_tokens = {
I
Ilya Dryomov 已提交
811
	{Opt_queue_depth, "queue_depth=%d"},
812 813
	/* int args above */
	/* string args above */
A
Alex Elder 已提交
814
	{Opt_read_only, "read_only"},
A
Alex Elder 已提交
815 816 817
	{Opt_read_only, "ro"},		/* Alternate spelling */
	{Opt_read_write, "read_write"},
	{Opt_read_write, "rw"},		/* Alternate spelling */
I
Ilya Dryomov 已提交
818
	{Opt_lock_on_read, "lock_on_read"},
819
	{Opt_err, NULL}
820 821
};

A
Alex Elder 已提交
822
struct rbd_options {
I
Ilya Dryomov 已提交
823
	int	queue_depth;
A
Alex Elder 已提交
824
	bool	read_only;
I
Ilya Dryomov 已提交
825
	bool	lock_on_read;
A
Alex Elder 已提交
826 827
};

I
Ilya Dryomov 已提交
828
#define RBD_QUEUE_DEPTH_DEFAULT	BLKDEV_MAX_RQ
A
Alex Elder 已提交
829
#define RBD_READ_ONLY_DEFAULT	false
I
Ilya Dryomov 已提交
830
#define RBD_LOCK_ON_READ_DEFAULT false
A
Alex Elder 已提交
831

832 833
static int parse_rbd_opts_token(char *c, void *private)
{
A
Alex Elder 已提交
834
	struct rbd_options *rbd_opts = private;
835 836 837
	substring_t argstr[MAX_OPT_ARGS];
	int token, intval, ret;

A
Alex Elder 已提交
838
	token = match_token(c, rbd_opts_tokens, argstr);
839 840 841
	if (token < Opt_last_int) {
		ret = match_int(&argstr[0], &intval);
		if (ret < 0) {
842
			pr_err("bad mount option arg (not int) at '%s'\n", c);
843 844 845 846
			return ret;
		}
		dout("got int token %d val %d\n", token, intval);
	} else if (token > Opt_last_int && token < Opt_last_string) {
847
		dout("got string token %d val %s\n", token, argstr[0].from);
848 849 850 851 852
	} else {
		dout("got token %d\n", token);
	}

	switch (token) {
I
Ilya Dryomov 已提交
853 854 855 856 857 858 859
	case Opt_queue_depth:
		if (intval < 1) {
			pr_err("queue_depth out of range\n");
			return -EINVAL;
		}
		rbd_opts->queue_depth = intval;
		break;
A
Alex Elder 已提交
860 861 862 863 864 865
	case Opt_read_only:
		rbd_opts->read_only = true;
		break;
	case Opt_read_write:
		rbd_opts->read_only = false;
		break;
I
Ilya Dryomov 已提交
866 867 868
	case Opt_lock_on_read:
		rbd_opts->lock_on_read = true;
		break;
869
	default:
870 871
		/* libceph prints "bad option" msg */
		return -EINVAL;
872
	}
873

874 875 876
	return 0;
}

G
Guangliang Zhao 已提交
877 878 879 880 881 882 883
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";
884 885
	case OBJ_OP_DISCARD:
		return "discard";
G
Guangliang Zhao 已提交
886 887 888 889 890
	default:
		return "???";
	}
}

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

900
	mutex_lock_nested(&client_mutex, SINGLE_DEPTH_NESTING);
901
	rbdc = rbd_client_find(ceph_opts);
902
	if (rbdc)	/* using an existing client */
A
Alex Elder 已提交
903
		ceph_destroy_options(ceph_opts);
904
	else
905
		rbdc = rbd_client_create(ceph_opts);
906
	mutex_unlock(&client_mutex);
907

908
	return rbdc;
909 910 911 912
}

/*
 * Destroy ceph client
A
Alex Elder 已提交
913
 *
A
Alex Elder 已提交
914
 * Caller must hold rbd_client_list_lock.
915 916 917 918 919
 */
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
920
	dout("%s: rbdc %p\n", __func__, rbdc);
921
	spin_lock(&rbd_client_list_lock);
922
	list_del(&rbdc->node);
923
	spin_unlock(&rbd_client_list_lock);
924 925 926 927 928 929 930 931 932

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

/*
 * Drop reference to ceph client node. If it's not referenced anymore, release
 * it.
 */
933
static void rbd_put_client(struct rbd_client *rbdc)
934
{
935 936
	if (rbdc)
		kref_put(&rbdc->kref, rbd_client_release);
937 938
}

939 940 941 942 943
static bool rbd_image_format_valid(u32 image_format)
{
	return image_format == 1 || image_format == 2;
}

944 945
static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk)
{
946 947 948 949 950 951 952
	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
953 954 955 956 957 958 959 960 961 962
	/* 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;

963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980
	/*
	 * 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;
981 982
}

I
Ilya Dryomov 已提交
983 984 985 986 987 988 989 990
/*
 * 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
991 992 993 994 995 996 997 998 999 1000 1001
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);
1002 1003
	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
1004 1005 1006
	RCU_INIT_POINTER(rbd_dev->layout.pool_ns, NULL);
}

1007
/*
1008 1009
 * Fill an rbd image header with information from the given format 1
 * on-disk header.
1010
 */
A
Alex Elder 已提交
1011
static int rbd_header_from_disk(struct rbd_device *rbd_dev,
1012
				 struct rbd_image_header_ondisk *ondisk)
1013
{
A
Alex Elder 已提交
1014
	struct rbd_image_header *header = &rbd_dev->header;
1015 1016 1017 1018 1019
	bool first_time = header->object_prefix == NULL;
	struct ceph_snap_context *snapc;
	char *object_prefix = NULL;
	char *snap_names = NULL;
	u64 *snap_sizes = NULL;
1020
	u32 snap_count;
1021
	int ret = -ENOMEM;
1022
	u32 i;
1023

1024
	/* Allocate this now to avoid having to handle failure below */
A
Alex Elder 已提交
1025

1026
	if (first_time) {
1027 1028 1029
		object_prefix = kstrndup(ondisk->object_prefix,
					 sizeof(ondisk->object_prefix),
					 GFP_KERNEL);
1030 1031 1032
		if (!object_prefix)
			return -ENOMEM;
	}
A
Alex Elder 已提交
1033

1034
	/* Allocate the snapshot context and fill it in */
A
Alex Elder 已提交
1035

1036 1037 1038 1039 1040
	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);
1041
	if (snap_count) {
1042
		struct rbd_image_snap_ondisk *snaps;
A
Alex Elder 已提交
1043 1044
		u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len);

1045
		/* We'll keep a copy of the snapshot names... */
1046

1047 1048 1049 1050
		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
1051 1052
			goto out_err;

1053
		/* ...as well as the array of their sizes. */
1054 1055 1056
		snap_sizes = kmalloc_array(snap_count,
					   sizeof(*header->snap_sizes),
					   GFP_KERNEL);
1057
		if (!snap_sizes)
A
Alex Elder 已提交
1058
			goto out_err;
1059

A
Alex Elder 已提交
1060
		/*
1061 1062 1063
		 * Copy the names, and fill in each snapshot's id
		 * and size.
		 *
1064
		 * Note that rbd_dev_v1_header_info() guarantees the
1065
		 * ondisk buffer we're working with has
A
Alex Elder 已提交
1066 1067 1068
		 * snap_names_len bytes beyond the end of the
		 * snapshot id array, this memcpy() is safe.
		 */
1069 1070 1071 1072 1073 1074
		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);
		}
1075
	}
A
Alex Elder 已提交
1076

1077
	/* We won't fail any more, fill in the header */
1078

1079 1080 1081
	if (first_time) {
		header->object_prefix = object_prefix;
		header->obj_order = ondisk->options.order;
I
Ilya Dryomov 已提交
1082
		rbd_init_layout(rbd_dev);
1083
	} else {
A
Alex Elder 已提交
1084 1085 1086
		ceph_put_snap_context(header->snapc);
		kfree(header->snap_names);
		kfree(header->snap_sizes);
1087
	}
1088

1089
	/* The remaining fields always get updated (when we refresh) */
1090

A
Alex Elder 已提交
1091
	header->image_size = le64_to_cpu(ondisk->image_size);
1092 1093 1094
	header->snapc = snapc;
	header->snap_names = snap_names;
	header->snap_sizes = snap_sizes;
1095

1096
	return 0;
1097 1098
out_2big:
	ret = -EIO;
A
Alex Elder 已提交
1099
out_err:
1100 1101 1102 1103
	kfree(snap_sizes);
	kfree(snap_names);
	ceph_put_snap_context(snapc);
	kfree(object_prefix);
1104

1105
	return ret;
1106 1107
}

1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122
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);
}

1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146
/*
 * 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.
 */
1147 1148 1149
static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id)
{
	struct ceph_snap_context *snapc = rbd_dev->header.snapc;
1150
	u64 *found;
1151

1152 1153
	found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps,
				sizeof (snap_id), snapid_compare_reverse);
1154

1155
	return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX;
1156 1157
}

1158 1159
static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev,
					u64 snap_id)
1160
{
1161
	u32 which;
1162
	const char *snap_name;
1163

1164 1165
	which = rbd_dev_snap_index(rbd_dev, snap_id);
	if (which == BAD_SNAP_INDEX)
1166
		return ERR_PTR(-ENOENT);
1167

1168 1169
	snap_name = _rbd_dev_v1_snap_name(rbd_dev, which);
	return snap_name ? snap_name : ERR_PTR(-ENOMEM);
1170 1171 1172 1173
}

static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id)
{
1174 1175 1176
	if (snap_id == CEPH_NOSNAP)
		return RBD_SNAP_HEAD_NAME;

1177 1178 1179
	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);
1180

1181
	return rbd_dev_v2_snap_name(rbd_dev, snap_id);
1182 1183
}

1184 1185
static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
				u64 *snap_size)
1186
{
1187 1188 1189 1190 1191
	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;
1192

1193 1194 1195
		which = rbd_dev_snap_index(rbd_dev, snap_id);
		if (which == BAD_SNAP_INDEX)
			return -ENOENT;
1196

1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208
		*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;
1209 1210
}

1211 1212
static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
			u64 *snap_features)
1213
{
1214 1215 1216 1217 1218
	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 */
1219
	} else {
1220 1221
		u64 features = 0;
		int ret;
1222

1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233
		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)
{
1234
	u64 snap_id = rbd_dev->spec->snap_id;
1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248
	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;

1249
	return 0;
1250 1251
}

A
Alex Elder 已提交
1252 1253 1254 1255
static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev)
{
	rbd_dev->mapping.size = 0;
	rbd_dev->mapping.features = 0;
1256 1257
}

A
Alex Elder 已提交
1258 1259
static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset)
{
I
Ilya Dryomov 已提交
1260
	u64 segment_size = rbd_obj_bytes(&rbd_dev->header);
1261

A
Alex Elder 已提交
1262 1263 1264 1265 1266 1267
	return offset & (segment_size - 1);
}

static u64 rbd_segment_length(struct rbd_device *rbd_dev,
				u64 offset, u64 length)
{
I
Ilya Dryomov 已提交
1268
	u64 segment_size = rbd_obj_bytes(&rbd_dev->header);
A
Alex Elder 已提交
1269 1270 1271

	offset &= segment_size - 1;

A
Alex Elder 已提交
1272
	rbd_assert(length <= U64_MAX - offset);
A
Alex Elder 已提交
1273 1274 1275 1276
	if (offset + length > segment_size)
		length = segment_size - offset;

	return length;
1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298
}

/*
 * bio helpers
 */

static void bio_chain_put(struct bio *chain)
{
	struct bio *tmp;

	while (chain) {
		tmp = chain;
		chain = chain->bi_next;
		bio_put(tmp);
	}
}

/*
 * zeros a bio chain, starting at specific offset
 */
static void zero_bio_chain(struct bio *chain, int start_ofs)
{
1299 1300
	struct bio_vec bv;
	struct bvec_iter iter;
1301 1302 1303 1304 1305
	unsigned long flags;
	void *buf;
	int pos = 0;

	while (chain) {
1306 1307
		bio_for_each_segment(bv, chain, iter) {
			if (pos + bv.bv_len > start_ofs) {
1308
				int remainder = max(start_ofs - pos, 0);
1309
				buf = bvec_kmap_irq(&bv, &flags);
1310
				memset(buf + remainder, 0,
1311 1312
				       bv.bv_len - remainder);
				flush_dcache_page(bv.bv_page);
1313
				bvec_kunmap_irq(buf, &flags);
1314
			}
1315
			pos += bv.bv_len;
1316 1317 1318 1319 1320 1321
		}

		chain = chain->bi_next;
	}
}

A
Alex Elder 已提交
1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339
/*
 * similar to zero_bio_chain(), zeros data defined by a page array,
 * starting at the given byte offset from the start of the array and
 * continuing up to the given end offset.  The pages array is
 * assumed to be big enough to hold all bytes up to the end.
 */
static void zero_pages(struct page **pages, u64 offset, u64 end)
{
	struct page **page = &pages[offset >> PAGE_SHIFT];

	rbd_assert(end > offset);
	rbd_assert(end - offset <= (u64)SIZE_MAX);
	while (offset < end) {
		size_t page_offset;
		size_t length;
		unsigned long flags;
		void *kaddr;

1340 1341
		page_offset = offset & ~PAGE_MASK;
		length = min_t(size_t, PAGE_SIZE - page_offset, end - offset);
A
Alex Elder 已提交
1342 1343 1344
		local_irq_save(flags);
		kaddr = kmap_atomic(*page);
		memset(kaddr + page_offset, 0, length);
1345
		flush_dcache_page(*page);
A
Alex Elder 已提交
1346 1347 1348 1349 1350 1351 1352 1353
		kunmap_atomic(kaddr);
		local_irq_restore(flags);

		offset += length;
		page++;
	}
}

1354
/*
A
Alex Elder 已提交
1355 1356
 * Clone a portion of a bio, starting at the given byte offset
 * and continuing for the number of bytes indicated.
1357
 */
A
Alex Elder 已提交
1358 1359 1360 1361
static struct bio *bio_clone_range(struct bio *bio_src,
					unsigned int offset,
					unsigned int len,
					gfp_t gfpmask)
1362
{
A
Alex Elder 已提交
1363 1364
	struct bio *bio;

K
Kent Overstreet 已提交
1365
	bio = bio_clone(bio_src, gfpmask);
A
Alex Elder 已提交
1366 1367
	if (!bio)
		return NULL;	/* ENOMEM */
1368

K
Kent Overstreet 已提交
1369
	bio_advance(bio, offset);
1370
	bio->bi_iter.bi_size = len;
A
Alex Elder 已提交
1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400

	return bio;
}

/*
 * Clone a portion of a bio chain, starting at the given byte offset
 * into the first bio in the source chain and continuing for the
 * number of bytes indicated.  The result is another bio chain of
 * exactly the given length, or a null pointer on error.
 *
 * The bio_src and offset parameters are both in-out.  On entry they
 * refer to the first source bio and the offset into that bio where
 * the start of data to be cloned is located.
 *
 * On return, bio_src is updated to refer to the bio in the source
 * chain that contains first un-cloned byte, and *offset will
 * contain the offset of that byte within that bio.
 */
static struct bio *bio_chain_clone_range(struct bio **bio_src,
					unsigned int *offset,
					unsigned int len,
					gfp_t gfpmask)
{
	struct bio *bi = *bio_src;
	unsigned int off = *offset;
	struct bio *chain = NULL;
	struct bio **end;

	/* Build up a chain of clone bios up to the limit */

1401
	if (!bi || off >= bi->bi_iter.bi_size || !len)
A
Alex Elder 已提交
1402
		return NULL;		/* Nothing to clone */
1403

A
Alex Elder 已提交
1404 1405 1406 1407 1408
	end = &chain;
	while (len) {
		unsigned int bi_size;
		struct bio *bio;

1409 1410
		if (!bi) {
			rbd_warn(NULL, "bio_chain exhausted with %u left", len);
A
Alex Elder 已提交
1411
			goto out_err;	/* EINVAL; ran out of bio's */
1412
		}
1413
		bi_size = min_t(unsigned int, bi->bi_iter.bi_size - off, len);
A
Alex Elder 已提交
1414 1415 1416 1417 1418 1419
		bio = bio_clone_range(bi, off, bi_size, gfpmask);
		if (!bio)
			goto out_err;	/* ENOMEM */

		*end = bio;
		end = &bio->bi_next;
1420

A
Alex Elder 已提交
1421
		off += bi_size;
1422
		if (off == bi->bi_iter.bi_size) {
A
Alex Elder 已提交
1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433
			bi = bi->bi_next;
			off = 0;
		}
		len -= bi_size;
	}
	*bio_src = bi;
	*offset = off;

	return chain;
out_err:
	bio_chain_put(chain);
1434 1435 1436 1437

	return NULL;
}

1438 1439 1440 1441 1442
/*
 * The default/initial value for all object request flags is 0.  For
 * each flag, once its value is set to 1 it is never reset to 0
 * again.
 */
A
Alex Elder 已提交
1443
static void obj_request_img_data_set(struct rbd_obj_request *obj_request)
1444
{
A
Alex Elder 已提交
1445
	if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) {
1446 1447
		struct rbd_device *rbd_dev;

A
Alex Elder 已提交
1448
		rbd_dev = obj_request->img_request->rbd_dev;
1449
		rbd_warn(rbd_dev, "obj_request %p already marked img_data",
1450 1451 1452 1453
			obj_request);
	}
}

A
Alex Elder 已提交
1454
static bool obj_request_img_data_test(struct rbd_obj_request *obj_request)
1455 1456
{
	smp_mb();
A
Alex Elder 已提交
1457
	return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0;
1458 1459
}

A
Alex Elder 已提交
1460
static void obj_request_done_set(struct rbd_obj_request *obj_request)
1461
{
A
Alex Elder 已提交
1462 1463
	if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) {
		struct rbd_device *rbd_dev = NULL;
1464

A
Alex Elder 已提交
1465 1466
		if (obj_request_img_data_test(obj_request))
			rbd_dev = obj_request->img_request->rbd_dev;
1467
		rbd_warn(rbd_dev, "obj_request %p already marked done",
1468 1469 1470 1471
			obj_request);
	}
}

A
Alex Elder 已提交
1472
static bool obj_request_done_test(struct rbd_obj_request *obj_request)
1473 1474
{
	smp_mb();
A
Alex Elder 已提交
1475
	return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0;
1476 1477
}

1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508
/*
 * This sets the KNOWN flag after (possibly) setting the EXISTS
 * flag.  The latter is set based on the "exists" value provided.
 *
 * Note that for our purposes once an object exists it never goes
 * away again.  It's possible that the response from two existence
 * checks are separated by the creation of the target object, and
 * the first ("doesn't exist") response arrives *after* the second
 * ("does exist").  In that case we ignore the second one.
 */
static void obj_request_existence_set(struct rbd_obj_request *obj_request,
				bool exists)
{
	if (exists)
		set_bit(OBJ_REQ_EXISTS, &obj_request->flags);
	set_bit(OBJ_REQ_KNOWN, &obj_request->flags);
	smp_mb();
}

static bool obj_request_known_test(struct rbd_obj_request *obj_request)
{
	smp_mb();
	return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0;
}

static bool obj_request_exists_test(struct rbd_obj_request *obj_request)
{
	smp_mb();
	return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0;
}

1509 1510 1511 1512 1513 1514 1515 1516
static bool obj_request_overlaps_parent(struct rbd_obj_request *obj_request)
{
	struct rbd_device *rbd_dev = obj_request->img_request->rbd_dev;

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

A
Alex Elder 已提交
1517 1518
static void rbd_obj_request_get(struct rbd_obj_request *obj_request)
{
A
Alex Elder 已提交
1519
	dout("%s: obj %p (was %d)\n", __func__, obj_request,
1520
		kref_read(&obj_request->kref));
A
Alex Elder 已提交
1521 1522 1523 1524 1525 1526 1527
	kref_get(&obj_request->kref);
}

static void rbd_obj_request_destroy(struct kref *kref);
static void rbd_obj_request_put(struct rbd_obj_request *obj_request)
{
	rbd_assert(obj_request != NULL);
A
Alex Elder 已提交
1528
	dout("%s: obj %p (was %d)\n", __func__, obj_request,
1529
		kref_read(&obj_request->kref));
A
Alex Elder 已提交
1530 1531 1532
	kref_put(&obj_request->kref, rbd_obj_request_destroy);
}

1533 1534 1535
static void rbd_img_request_get(struct rbd_img_request *img_request)
{
	dout("%s: img %p (was %d)\n", __func__, img_request,
1536
	     kref_read(&img_request->kref));
1537 1538 1539
	kref_get(&img_request->kref);
}

1540 1541
static bool img_request_child_test(struct rbd_img_request *img_request);
static void rbd_parent_request_destroy(struct kref *kref);
A
Alex Elder 已提交
1542 1543 1544 1545
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
1546
	dout("%s: img %p (was %d)\n", __func__, img_request,
1547
		kref_read(&img_request->kref));
1548 1549 1550 1551
	if (img_request_child_test(img_request))
		kref_put(&img_request->kref, rbd_parent_request_destroy);
	else
		kref_put(&img_request->kref, rbd_img_request_destroy);
A
Alex Elder 已提交
1552 1553 1554 1555 1556
}

static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request,
					struct rbd_obj_request *obj_request)
{
1557 1558
	rbd_assert(obj_request->img_request == NULL);

1559
	/* Image request now owns object's original reference */
A
Alex Elder 已提交
1560
	obj_request->img_request = img_request;
1561
	obj_request->which = img_request->obj_request_count;
1562 1563
	rbd_assert(!obj_request_img_data_test(obj_request));
	obj_request_img_data_set(obj_request);
A
Alex Elder 已提交
1564
	rbd_assert(obj_request->which != BAD_WHICH);
1565 1566
	img_request->obj_request_count++;
	list_add_tail(&obj_request->links, &img_request->obj_requests);
A
Alex Elder 已提交
1567 1568
	dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
		obj_request->which);
A
Alex Elder 已提交
1569 1570 1571 1572 1573 1574
}

static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request,
					struct rbd_obj_request *obj_request)
{
	rbd_assert(obj_request->which != BAD_WHICH);
1575

A
Alex Elder 已提交
1576 1577
	dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
		obj_request->which);
A
Alex Elder 已提交
1578
	list_del(&obj_request->links);
1579 1580 1581 1582
	rbd_assert(img_request->obj_request_count > 0);
	img_request->obj_request_count--;
	rbd_assert(obj_request->which == img_request->obj_request_count);
	obj_request->which = BAD_WHICH;
1583
	rbd_assert(obj_request_img_data_test(obj_request));
A
Alex Elder 已提交
1584 1585
	rbd_assert(obj_request->img_request == img_request);
	obj_request->img_request = NULL;
1586
	obj_request->callback = NULL;
A
Alex Elder 已提交
1587 1588 1589 1590 1591 1592
	rbd_obj_request_put(obj_request);
}

static bool obj_request_type_valid(enum obj_request_type type)
{
	switch (type) {
1593
	case OBJ_REQUEST_NODATA:
A
Alex Elder 已提交
1594
	case OBJ_REQUEST_BIO:
1595
	case OBJ_REQUEST_PAGES:
A
Alex Elder 已提交
1596 1597 1598 1599 1600 1601
		return true;
	default:
		return false;
	}
}

1602 1603
static void rbd_img_obj_callback(struct rbd_obj_request *obj_request);

1604
static void rbd_obj_request_submit(struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1605
{
1606 1607
	struct ceph_osd_request *osd_req = obj_request->osd_req;

1608 1609
	dout("%s %p object_no %016llx %llu~%llu osd_req %p\n", __func__,
	     obj_request, obj_request->object_no, obj_request->offset,
1610
	     obj_request->length, osd_req);
1611 1612 1613 1614
	if (obj_request_img_data_test(obj_request)) {
		WARN_ON(obj_request->callback != rbd_img_obj_callback);
		rbd_img_request_get(obj_request->img_request);
	}
1615
	ceph_osdc_start_request(osd_req->r_osdc, osd_req, false);
A
Alex Elder 已提交
1616 1617 1618 1619
}

static void rbd_img_request_complete(struct rbd_img_request *img_request)
{
1620

A
Alex Elder 已提交
1621
	dout("%s: img %p\n", __func__, img_request);
1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637

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

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

A
Alex Elder 已提交
1638 1639 1640 1641 1642 1643
	if (img_request->callback)
		img_request->callback(img_request);
	else
		rbd_img_request_put(img_request);
}

A
Alex Elder 已提交
1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660
/*
 * The default/initial value for all image request flags is 0.  Each
 * is conditionally set to 1 at image request initialization time
 * and currently never change thereafter.
 */
static void img_request_write_set(struct rbd_img_request *img_request)
{
	set_bit(IMG_REQ_WRITE, &img_request->flags);
	smp_mb();
}

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

1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675
/*
 * Set the discard flag when the img_request is an discard request
 */
static void img_request_discard_set(struct rbd_img_request *img_request)
{
	set_bit(IMG_REQ_DISCARD, &img_request->flags);
	smp_mb();
}

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

1676 1677 1678 1679 1680 1681
static void img_request_child_set(struct rbd_img_request *img_request)
{
	set_bit(IMG_REQ_CHILD, &img_request->flags);
	smp_mb();
}

1682 1683 1684 1685 1686 1687
static void img_request_child_clear(struct rbd_img_request *img_request)
{
	clear_bit(IMG_REQ_CHILD, &img_request->flags);
	smp_mb();
}

1688 1689 1690 1691 1692 1693
static bool img_request_child_test(struct rbd_img_request *img_request)
{
	smp_mb();
	return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0;
}

1694 1695 1696 1697 1698 1699
static void img_request_layered_set(struct rbd_img_request *img_request)
{
	set_bit(IMG_REQ_LAYERED, &img_request->flags);
	smp_mb();
}

1700 1701 1702 1703 1704 1705
static void img_request_layered_clear(struct rbd_img_request *img_request)
{
	clear_bit(IMG_REQ_LAYERED, &img_request->flags);
	smp_mb();
}

1706 1707 1708 1709 1710 1711
static bool img_request_layered_test(struct rbd_img_request *img_request)
{
	smp_mb();
	return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0;
}

1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722
static enum obj_operation_type
rbd_img_request_op_type(struct rbd_img_request *img_request)
{
	if (img_request_write_test(img_request))
		return OBJ_OP_WRITE;
	else if (img_request_discard_test(img_request))
		return OBJ_OP_DISCARD;
	else
		return OBJ_OP_READ;
}

1723 1724 1725
static void
rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request)
{
A
Alex Elder 已提交
1726 1727 1728
	u64 xferred = obj_request->xferred;
	u64 length = obj_request->length;

1729 1730
	dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
		obj_request, obj_request->img_request, obj_request->result,
A
Alex Elder 已提交
1731
		xferred, length);
1732
	/*
1733 1734 1735 1736 1737 1738
	 * ENOENT means a hole in the image.  We zero-fill the entire
	 * length of the request.  A short read also implies zero-fill
	 * to the end of the request.  An error requires the whole
	 * length of the request to be reported finished with an error
	 * to the block layer.  In each case we update the xferred
	 * count to indicate the whole request was satisfied.
1739
	 */
A
Alex Elder 已提交
1740
	rbd_assert(obj_request->type != OBJ_REQUEST_NODATA);
1741
	if (obj_request->result == -ENOENT) {
A
Alex Elder 已提交
1742 1743 1744 1745
		if (obj_request->type == OBJ_REQUEST_BIO)
			zero_bio_chain(obj_request->bio_list, 0);
		else
			zero_pages(obj_request->pages, 0, length);
1746
		obj_request->result = 0;
A
Alex Elder 已提交
1747 1748 1749 1750 1751
	} else if (xferred < length && !obj_request->result) {
		if (obj_request->type == OBJ_REQUEST_BIO)
			zero_bio_chain(obj_request->bio_list, xferred);
		else
			zero_pages(obj_request->pages, xferred, length);
1752
	}
1753
	obj_request->xferred = length;
1754 1755 1756
	obj_request_done_set(obj_request);
}

A
Alex Elder 已提交
1757 1758
static void rbd_obj_request_complete(struct rbd_obj_request *obj_request)
{
A
Alex Elder 已提交
1759 1760
	dout("%s: obj %p cb %p\n", __func__, obj_request,
		obj_request->callback);
A
Alex Elder 已提交
1761 1762
	if (obj_request->callback)
		obj_request->callback(obj_request);
1763 1764
	else
		complete_all(&obj_request->completion);
A
Alex Elder 已提交
1765 1766
}

1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782
static void rbd_obj_request_error(struct rbd_obj_request *obj_request, int err)
{
	obj_request->result = err;
	obj_request->xferred = 0;
	/*
	 * kludge - mirror rbd_obj_request_submit() to match a put in
	 * rbd_img_obj_callback()
	 */
	if (obj_request_img_data_test(obj_request)) {
		WARN_ON(obj_request->callback != rbd_img_obj_callback);
		rbd_img_request_get(obj_request->img_request);
	}
	obj_request_done_set(obj_request);
	rbd_obj_request_complete(obj_request);
}

1783
static void rbd_osd_read_callback(struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1784
{
A
Alex Elder 已提交
1785
	struct rbd_img_request *img_request = NULL;
A
Alex Elder 已提交
1786
	struct rbd_device *rbd_dev = NULL;
A
Alex Elder 已提交
1787 1788 1789 1790 1791
	bool layered = false;

	if (obj_request_img_data_test(obj_request)) {
		img_request = obj_request->img_request;
		layered = img_request && img_request_layered_test(img_request);
A
Alex Elder 已提交
1792
		rbd_dev = img_request->rbd_dev;
A
Alex Elder 已提交
1793
	}
A
Alex Elder 已提交
1794 1795 1796 1797

	dout("%s: obj %p img %p result %d %llu/%llu\n", __func__,
		obj_request, img_request, obj_request->result,
		obj_request->xferred, obj_request->length);
A
Alex Elder 已提交
1798 1799
	if (layered && obj_request->result == -ENOENT &&
			obj_request->img_offset < rbd_dev->parent_overlap)
A
Alex Elder 已提交
1800 1801
		rbd_img_parent_read(obj_request);
	else if (img_request)
1802 1803 1804
		rbd_img_obj_request_read_callback(obj_request);
	else
		obj_request_done_set(obj_request);
A
Alex Elder 已提交
1805 1806
}

1807
static void rbd_osd_write_callback(struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1808
{
1809 1810 1811
	dout("%s: obj %p result %d %llu\n", __func__, obj_request,
		obj_request->result, obj_request->length);
	/*
A
Alex Elder 已提交
1812 1813
	 * There is no such thing as a successful short write.  Set
	 * it to our originally-requested length.
1814 1815
	 */
	obj_request->xferred = obj_request->length;
1816
	obj_request_done_set(obj_request);
A
Alex Elder 已提交
1817 1818
}

1819 1820 1821 1822 1823 1824 1825 1826 1827
static void rbd_osd_discard_callback(struct rbd_obj_request *obj_request)
{
	dout("%s: obj %p result %d %llu\n", __func__, obj_request,
		obj_request->result, obj_request->length);
	/*
	 * There is no such thing as a successful short discard.  Set
	 * it to our originally-requested length.
	 */
	obj_request->xferred = obj_request->length;
1828 1829 1830
	/* discarding a non-existent object is not a problem */
	if (obj_request->result == -ENOENT)
		obj_request->result = 0;
1831 1832 1833
	obj_request_done_set(obj_request);
}

A
Alex Elder 已提交
1834 1835 1836 1837
/*
 * For a simple stat call there's nothing to do.  We'll do more if
 * this is part of a write sequence for a layered image.
 */
1838
static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1839
{
A
Alex Elder 已提交
1840
	dout("%s: obj %p\n", __func__, obj_request);
A
Alex Elder 已提交
1841 1842 1843
	obj_request_done_set(obj_request);
}

I
Ilya Dryomov 已提交
1844 1845 1846 1847 1848 1849 1850 1851 1852 1853
static void rbd_osd_call_callback(struct rbd_obj_request *obj_request)
{
	dout("%s: obj %p\n", __func__, obj_request);

	if (obj_request_img_data_test(obj_request))
		rbd_osd_copyup_callback(obj_request);
	else
		obj_request_done_set(obj_request);
}

1854
static void rbd_osd_req_callback(struct ceph_osd_request *osd_req)
A
Alex Elder 已提交
1855 1856 1857 1858
{
	struct rbd_obj_request *obj_request = osd_req->r_priv;
	u16 opcode;

1859
	dout("%s: osd_req %p\n", __func__, osd_req);
A
Alex Elder 已提交
1860
	rbd_assert(osd_req == obj_request->osd_req);
A
Alex Elder 已提交
1861 1862 1863 1864 1865 1866
	if (obj_request_img_data_test(obj_request)) {
		rbd_assert(obj_request->img_request);
		rbd_assert(obj_request->which != BAD_WHICH);
	} else {
		rbd_assert(obj_request->which == BAD_WHICH);
	}
A
Alex Elder 已提交
1867

1868 1869
	if (osd_req->r_result < 0)
		obj_request->result = osd_req->r_result;
A
Alex Elder 已提交
1870

1871 1872
	/*
	 * We support a 64-bit length, but ultimately it has to be
C
Christoph Hellwig 已提交
1873 1874
	 * passed to the block layer, which just supports a 32-bit
	 * length field.
1875
	 */
1876
	obj_request->xferred = osd_req->r_ops[0].outdata_len;
A
Alex Elder 已提交
1877
	rbd_assert(obj_request->xferred < (u64)UINT_MAX);
1878

1879
	opcode = osd_req->r_ops[0].op;
A
Alex Elder 已提交
1880 1881
	switch (opcode) {
	case CEPH_OSD_OP_READ:
1882
		rbd_osd_read_callback(obj_request);
A
Alex Elder 已提交
1883
		break;
1884
	case CEPH_OSD_OP_SETALLOCHINT:
1885 1886
		rbd_assert(osd_req->r_ops[1].op == CEPH_OSD_OP_WRITE ||
			   osd_req->r_ops[1].op == CEPH_OSD_OP_WRITEFULL);
1887
		/* fall through */
A
Alex Elder 已提交
1888
	case CEPH_OSD_OP_WRITE:
1889
	case CEPH_OSD_OP_WRITEFULL:
1890
		rbd_osd_write_callback(obj_request);
A
Alex Elder 已提交
1891
		break;
A
Alex Elder 已提交
1892
	case CEPH_OSD_OP_STAT:
1893
		rbd_osd_stat_callback(obj_request);
A
Alex Elder 已提交
1894
		break;
1895 1896 1897 1898 1899
	case CEPH_OSD_OP_DELETE:
	case CEPH_OSD_OP_TRUNCATE:
	case CEPH_OSD_OP_ZERO:
		rbd_osd_discard_callback(obj_request);
		break;
1900
	case CEPH_OSD_OP_CALL:
I
Ilya Dryomov 已提交
1901 1902
		rbd_osd_call_callback(obj_request);
		break;
A
Alex Elder 已提交
1903
	default:
1904 1905
		rbd_warn(NULL, "unexpected OSD op: object_no %016llx opcode %d",
			 obj_request->object_no, opcode);
A
Alex Elder 已提交
1906 1907 1908
		break;
	}

1909
	if (obj_request_done_test(obj_request))
A
Alex Elder 已提交
1910 1911 1912
		rbd_obj_request_complete(obj_request);
}

1913
static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1914
{
1915
	struct ceph_osd_request *osd_req = obj_request->osd_req;
A
Alex Elder 已提交
1916

1917 1918
	rbd_assert(obj_request_img_data_test(obj_request));
	osd_req->r_snapid = obj_request->img_request->snap_id;
1919 1920 1921 1922 1923 1924
}

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

1925 1926
	osd_req->r_mtime = CURRENT_TIME;
	osd_req->r_data_offset = obj_request->offset;
A
Alex Elder 已提交
1927 1928
}

1929 1930 1931 1932 1933 1934 1935 1936
static struct ceph_osd_request *
__rbd_osd_req_create(struct rbd_device *rbd_dev,
		     struct ceph_snap_context *snapc,
		     int num_ops, unsigned int flags,
		     struct rbd_obj_request *obj_request)
{
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
	struct ceph_osd_request *req;
1937 1938
	const char *name_format = rbd_dev->image_format == 1 ?
				      RBD_V1_DATA_FORMAT : RBD_V2_DATA_FORMAT;
1939 1940 1941 1942 1943 1944 1945 1946 1947 1948

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

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

	req->r_base_oloc.pool = rbd_dev->layout.pool_id;
1949 1950
	if (ceph_oid_aprintf(&req->r_base_oid, GFP_NOIO, name_format,
			rbd_dev->header.object_prefix, obj_request->object_no))
1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962
		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;
}

1963 1964 1965 1966 1967 1968
/*
 * Create an osd request.  A read request has one osd op (read).
 * A write request has either one (watch) or two (hint+write) osd ops.
 * (All rbd data writes are prefixed with an allocation hint op, but
 * technically osd watch is a write request, hence this distinction.)
 */
A
Alex Elder 已提交
1969 1970
static struct ceph_osd_request *rbd_osd_req_create(
					struct rbd_device *rbd_dev,
G
Guangliang Zhao 已提交
1971
					enum obj_operation_type op_type,
1972
					unsigned int num_ops,
A
Alex Elder 已提交
1973
					struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1974 1975 1976
{
	struct ceph_snap_context *snapc = NULL;

1977 1978
	if (obj_request_img_data_test(obj_request) &&
		(op_type == OBJ_OP_DISCARD || op_type == OBJ_OP_WRITE)) {
1979
		struct rbd_img_request *img_request = obj_request->img_request;
1980 1981 1982 1983 1984
		if (op_type == OBJ_OP_WRITE) {
			rbd_assert(img_request_write_test(img_request));
		} else {
			rbd_assert(img_request_discard_test(img_request));
		}
G
Guangliang Zhao 已提交
1985
		snapc = img_request->snapc;
A
Alex Elder 已提交
1986 1987
	}

G
Guangliang Zhao 已提交
1988
	rbd_assert(num_ops == 1 || ((op_type == OBJ_OP_WRITE) && num_ops == 2));
1989

1990 1991
	return __rbd_osd_req_create(rbd_dev, snapc, num_ops,
	    (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD) ?
1992
	    CEPH_OSD_FLAG_WRITE : CEPH_OSD_FLAG_READ, obj_request);
A
Alex Elder 已提交
1993 1994
}

1995
/*
1996 1997 1998 1999
 * Create a copyup osd request based on the information in the object
 * request supplied.  A copyup request has two or three osd ops, a
 * copyup method call, potentially a hint op, and a write or truncate
 * or zero op.
2000 2001 2002 2003 2004
 */
static struct ceph_osd_request *
rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request)
{
	struct rbd_img_request *img_request;
2005
	int num_osd_ops = 3;
2006 2007 2008 2009

	rbd_assert(obj_request_img_data_test(obj_request));
	img_request = obj_request->img_request;
	rbd_assert(img_request);
2010 2011
	rbd_assert(img_request_write_test(img_request) ||
			img_request_discard_test(img_request));
2012

2013 2014 2015
	if (img_request_discard_test(img_request))
		num_osd_ops = 2;

2016 2017
	return __rbd_osd_req_create(img_request->rbd_dev,
				    img_request->snapc, num_osd_ops,
2018
				    CEPH_OSD_FLAG_WRITE, obj_request);
2019 2020
}

A
Alex Elder 已提交
2021 2022 2023 2024 2025
static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req)
{
	ceph_osdc_put_request(osd_req);
}

2026 2027
static struct rbd_obj_request *
rbd_obj_request_create(enum obj_request_type type)
A
Alex Elder 已提交
2028 2029 2030 2031 2032
{
	struct rbd_obj_request *obj_request;

	rbd_assert(obj_request_type_valid(type));

2033
	obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_NOIO);
2034
	if (!obj_request)
2035 2036
		return NULL;

A
Alex Elder 已提交
2037 2038 2039
	obj_request->which = BAD_WHICH;
	obj_request->type = type;
	INIT_LIST_HEAD(&obj_request->links);
2040
	init_completion(&obj_request->completion);
A
Alex Elder 已提交
2041 2042
	kref_init(&obj_request->kref);

2043
	dout("%s %p\n", __func__, obj_request);
A
Alex Elder 已提交
2044 2045 2046 2047 2048 2049 2050 2051 2052
	return obj_request;
}

static void rbd_obj_request_destroy(struct kref *kref)
{
	struct rbd_obj_request *obj_request;

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

A
Alex Elder 已提交
2053 2054
	dout("%s: obj %p\n", __func__, obj_request);

A
Alex Elder 已提交
2055 2056 2057 2058 2059 2060 2061 2062
	rbd_assert(obj_request->img_request == NULL);
	rbd_assert(obj_request->which == BAD_WHICH);

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

	rbd_assert(obj_request_type_valid(obj_request->type));
	switch (obj_request->type) {
2063 2064
	case OBJ_REQUEST_NODATA:
		break;		/* Nothing to do */
A
Alex Elder 已提交
2065 2066 2067 2068
	case OBJ_REQUEST_BIO:
		if (obj_request->bio_list)
			bio_chain_put(obj_request->bio_list);
		break;
2069
	case OBJ_REQUEST_PAGES:
2070 2071 2072
		/* img_data requests don't own their page array */
		if (obj_request->pages &&
		    !obj_request_img_data_test(obj_request))
2073 2074 2075
			ceph_release_page_vector(obj_request->pages,
						obj_request->page_count);
		break;
A
Alex Elder 已提交
2076 2077
	}

2078
	kmem_cache_free(rbd_obj_request_cache, obj_request);
A
Alex Elder 已提交
2079 2080
}

A
Alex Elder 已提交
2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091
/* 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;
}

2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113
/*
 * 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
2114
		rbd_warn(rbd_dev, "parent reference underflow");
2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126
}

/*
 * 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)
{
2127
	int counter = 0;
2128 2129 2130 2131

	if (!rbd_dev->parent_spec)
		return false;

2132 2133 2134 2135
	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);
2136 2137

	if (counter < 0)
2138
		rbd_warn(rbd_dev, "parent reference overflow");
2139

2140
	return counter > 0;
2141 2142
}

A
Alex Elder 已提交
2143 2144 2145 2146 2147
/*
 * 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
2148 2149
static struct rbd_img_request *rbd_img_request_create(
					struct rbd_device *rbd_dev,
A
Alex Elder 已提交
2150
					u64 offset, u64 length,
G
Guangliang Zhao 已提交
2151
					enum obj_operation_type op_type,
2152
					struct ceph_snap_context *snapc)
A
Alex Elder 已提交
2153 2154 2155
{
	struct rbd_img_request *img_request;

2156
	img_request = kmem_cache_alloc(rbd_img_request_cache, GFP_NOIO);
A
Alex Elder 已提交
2157 2158 2159 2160 2161 2162 2163
	if (!img_request)
		return NULL;

	img_request->rq = NULL;
	img_request->rbd_dev = rbd_dev;
	img_request->offset = offset;
	img_request->length = length;
A
Alex Elder 已提交
2164
	img_request->flags = 0;
2165 2166 2167 2168
	if (op_type == OBJ_OP_DISCARD) {
		img_request_discard_set(img_request);
		img_request->snapc = snapc;
	} else if (op_type == OBJ_OP_WRITE) {
A
Alex Elder 已提交
2169
		img_request_write_set(img_request);
2170
		img_request->snapc = snapc;
A
Alex Elder 已提交
2171
	} else {
A
Alex Elder 已提交
2172
		img_request->snap_id = rbd_dev->spec->snap_id;
A
Alex Elder 已提交
2173
	}
2174
	if (rbd_dev_parent_get(rbd_dev))
2175
		img_request_layered_set(img_request);
A
Alex Elder 已提交
2176 2177 2178
	spin_lock_init(&img_request->completion_lock);
	img_request->next_completion = 0;
	img_request->callback = NULL;
2179
	img_request->result = 0;
A
Alex Elder 已提交
2180 2181 2182 2183
	img_request->obj_request_count = 0;
	INIT_LIST_HEAD(&img_request->obj_requests);
	kref_init(&img_request->kref);

A
Alex Elder 已提交
2184
	dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev,
G
Guangliang Zhao 已提交
2185
		obj_op_name(op_type), offset, length, img_request);
A
Alex Elder 已提交
2186

A
Alex Elder 已提交
2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197
	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
2198 2199
	dout("%s: img %p\n", __func__, img_request);

A
Alex Elder 已提交
2200 2201
	for_each_obj_request_safe(img_request, obj_request, next_obj_request)
		rbd_img_obj_request_del(img_request, obj_request);
2202
	rbd_assert(img_request->obj_request_count == 0);
A
Alex Elder 已提交
2203

2204 2205 2206 2207 2208
	if (img_request_layered_test(img_request)) {
		img_request_layered_clear(img_request);
		rbd_dev_parent_put(img_request->rbd_dev);
	}

2209 2210
	if (img_request_write_test(img_request) ||
		img_request_discard_test(img_request))
2211
		ceph_put_snap_context(img_request->snapc);
A
Alex Elder 已提交
2212

2213
	kmem_cache_free(rbd_img_request_cache, img_request);
A
Alex Elder 已提交
2214 2215
}

2216 2217 2218 2219 2220 2221 2222 2223 2224 2225
static struct rbd_img_request *rbd_parent_request_create(
					struct rbd_obj_request *obj_request,
					u64 img_offset, u64 length)
{
	struct rbd_img_request *parent_request;
	struct rbd_device *rbd_dev;

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

2226
	parent_request = rbd_img_request_create(rbd_dev->parent, img_offset,
G
Guangliang Zhao 已提交
2227
						length, OBJ_OP_READ, NULL);
2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252
	if (!parent_request)
		return NULL;

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

	return parent_request;
}

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

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

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

	rbd_img_request_destroy(kref);
}

2253 2254
static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request)
{
2255
	struct rbd_img_request *img_request;
2256 2257
	unsigned int xferred;
	int result;
A
Alex Elder 已提交
2258
	bool more;
2259

2260 2261 2262
	rbd_assert(obj_request_img_data_test(obj_request));
	img_request = obj_request->img_request;

2263 2264 2265 2266 2267
	rbd_assert(obj_request->xferred <= (u64)UINT_MAX);
	xferred = (unsigned int)obj_request->xferred;
	result = obj_request->result;
	if (result) {
		struct rbd_device *rbd_dev = img_request->rbd_dev;
G
Guangliang Zhao 已提交
2268 2269
		enum obj_operation_type op_type;

2270 2271 2272 2273 2274 2275
		if (img_request_discard_test(img_request))
			op_type = OBJ_OP_DISCARD;
		else if (img_request_write_test(img_request))
			op_type = OBJ_OP_WRITE;
		else
			op_type = OBJ_OP_READ;
2276

2277
		rbd_warn(rbd_dev, "%s %llx at %llx (%llx)",
G
Guangliang Zhao 已提交
2278 2279
			obj_op_name(op_type), obj_request->length,
			obj_request->img_offset, obj_request->offset);
2280
		rbd_warn(rbd_dev, "  result %d xferred %x",
2281 2282 2283
			result, xferred);
		if (!img_request->result)
			img_request->result = result;
2284 2285 2286 2287 2288
		/*
		 * Need to end I/O on the entire obj_request worth of
		 * bytes in case of error.
		 */
		xferred = obj_request->length;
2289 2290
	}

A
Alex Elder 已提交
2291 2292 2293 2294 2295
	if (img_request_child_test(img_request)) {
		rbd_assert(img_request->obj_request != NULL);
		more = obj_request->which < img_request->obj_request_count - 1;
	} else {
		rbd_assert(img_request->rq != NULL);
C
Christoph Hellwig 已提交
2296 2297 2298 2299

		more = blk_update_request(img_request->rq, result, xferred);
		if (!more)
			__blk_mq_end_request(img_request->rq, result);
A
Alex Elder 已提交
2300 2301 2302
	}

	return more;
2303 2304
}

2305 2306 2307 2308 2309 2310
static void rbd_img_obj_callback(struct rbd_obj_request *obj_request)
{
	struct rbd_img_request *img_request;
	u32 which = obj_request->which;
	bool more = true;

2311
	rbd_assert(obj_request_img_data_test(obj_request));
2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329
	img_request = obj_request->img_request;

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

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

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

		if (!obj_request_done_test(obj_request))
			break;
2330
		more = rbd_img_obj_end_request(obj_request);
2331 2332 2333 2334 2335 2336 2337
		which++;
	}

	rbd_assert(more ^ (which == img_request->obj_request_count));
	img_request->next_completion = which;
out:
	spin_unlock_irq(&img_request->completion_lock);
2338
	rbd_img_request_put(img_request);
2339 2340 2341 2342 2343

	if (!more)
		rbd_img_request_complete(img_request);
}

2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362
/*
 * Add individual osd ops to the given ceph_osd_request and prepare
 * them for submission. num_ops is the current number of
 * osd operations already to the object request.
 */
static void rbd_img_obj_request_fill(struct rbd_obj_request *obj_request,
				struct ceph_osd_request *osd_request,
				enum obj_operation_type op_type,
				unsigned int num_ops)
{
	struct rbd_img_request *img_request = obj_request->img_request;
	struct rbd_device *rbd_dev = img_request->rbd_dev;
	u64 object_size = rbd_obj_bytes(&rbd_dev->header);
	u64 offset = obj_request->offset;
	u64 length = obj_request->length;
	u64 img_end;
	u16 opcode;

	if (op_type == OBJ_OP_DISCARD) {
2363 2364 2365
		if (!offset && length == object_size &&
		    (!img_request_layered_test(img_request) ||
		     !obj_request_overlaps_parent(obj_request))) {
2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379
			opcode = CEPH_OSD_OP_DELETE;
		} else if ((offset + length == object_size)) {
			opcode = CEPH_OSD_OP_TRUNCATE;
		} else {
			down_read(&rbd_dev->header_rwsem);
			img_end = rbd_dev->header.image_size;
			up_read(&rbd_dev->header_rwsem);

			if (obj_request->img_offset + length == img_end)
				opcode = CEPH_OSD_OP_TRUNCATE;
			else
				opcode = CEPH_OSD_OP_ZERO;
		}
	} else if (op_type == OBJ_OP_WRITE) {
2380 2381 2382 2383
		if (!offset && length == object_size)
			opcode = CEPH_OSD_OP_WRITEFULL;
		else
			opcode = CEPH_OSD_OP_WRITE;
2384 2385 2386 2387 2388 2389 2390
		osd_req_op_alloc_hint_init(osd_request, num_ops,
					object_size, object_size);
		num_ops++;
	} else {
		opcode = CEPH_OSD_OP_READ;
	}

2391
	if (opcode == CEPH_OSD_OP_DELETE)
2392
		osd_req_op_init(osd_request, num_ops, opcode, 0);
2393 2394 2395 2396
	else
		osd_req_op_extent_init(osd_request, num_ops, opcode,
				       offset, length, 0, 0);

2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411
	if (obj_request->type == OBJ_REQUEST_BIO)
		osd_req_op_extent_osd_data_bio(osd_request, num_ops,
					obj_request->bio_list, length);
	else if (obj_request->type == OBJ_REQUEST_PAGES)
		osd_req_op_extent_osd_data_pages(osd_request, num_ops,
					obj_request->pages, length,
					offset & ~PAGE_MASK, false, false);

	/* Discards are also writes */
	if (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD)
		rbd_osd_req_format_write(obj_request);
	else
		rbd_osd_req_format_read(obj_request);
}

2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422
/*
 * Split up an image request into one or more object requests, each
 * to a different object.  The "type" parameter indicates whether
 * "data_desc" is the pointer to the head of a list of bio
 * structures, or the base of a page array.  In either case this
 * function assumes data_desc describes memory sufficient to hold
 * all data described by the image request.
 */
static int rbd_img_request_fill(struct rbd_img_request *img_request,
					enum obj_request_type type,
					void *data_desc)
A
Alex Elder 已提交
2423 2424 2425 2426
{
	struct rbd_device *rbd_dev = img_request->rbd_dev;
	struct rbd_obj_request *obj_request = NULL;
	struct rbd_obj_request *next_obj_request;
J
Jingoo Han 已提交
2427
	struct bio *bio_list = NULL;
2428
	unsigned int bio_offset = 0;
J
Jingoo Han 已提交
2429
	struct page **pages = NULL;
G
Guangliang Zhao 已提交
2430
	enum obj_operation_type op_type;
2431
	u64 img_offset;
A
Alex Elder 已提交
2432 2433
	u64 resid;

2434 2435
	dout("%s: img %p type %d data_desc %p\n", __func__, img_request,
		(int)type, data_desc);
A
Alex Elder 已提交
2436

2437
	img_offset = img_request->offset;
A
Alex Elder 已提交
2438
	resid = img_request->length;
A
Alex Elder 已提交
2439
	rbd_assert(resid > 0);
2440
	op_type = rbd_img_request_op_type(img_request);
2441 2442 2443

	if (type == OBJ_REQUEST_BIO) {
		bio_list = data_desc;
2444 2445
		rbd_assert(img_offset ==
			   bio_list->bi_iter.bi_sector << SECTOR_SHIFT);
2446
	} else if (type == OBJ_REQUEST_PAGES) {
2447 2448 2449
		pages = data_desc;
	}

A
Alex Elder 已提交
2450
	while (resid) {
2451
		struct ceph_osd_request *osd_req;
2452
		u64 object_no = img_offset >> rbd_dev->header.obj_order;
2453 2454
		u64 offset = rbd_segment_offset(rbd_dev, img_offset);
		u64 length = rbd_segment_length(rbd_dev, img_offset, resid);
A
Alex Elder 已提交
2455

2456
		obj_request = rbd_obj_request_create(type);
A
Alex Elder 已提交
2457 2458
		if (!obj_request)
			goto out_unwind;
2459

2460
		obj_request->object_no = object_no;
2461 2462 2463
		obj_request->offset = offset;
		obj_request->length = length;

2464 2465 2466 2467 2468
		/*
		 * set obj_request->img_request before creating the
		 * osd_request so that it gets the right snapc
		 */
		rbd_img_obj_request_add(img_request, obj_request);
A
Alex Elder 已提交
2469

2470 2471 2472 2473 2474 2475 2476 2477 2478
		if (type == OBJ_REQUEST_BIO) {
			unsigned int clone_size;

			rbd_assert(length <= (u64)UINT_MAX);
			clone_size = (unsigned int)length;
			obj_request->bio_list =
					bio_chain_clone_range(&bio_list,
								&bio_offset,
								clone_size,
2479
								GFP_NOIO);
2480
			if (!obj_request->bio_list)
2481
				goto out_unwind;
2482
		} else if (type == OBJ_REQUEST_PAGES) {
2483 2484 2485 2486 2487 2488 2489 2490 2491
			unsigned int page_count;

			obj_request->pages = pages;
			page_count = (u32)calc_pages_for(offset, length);
			obj_request->page_count = page_count;
			if ((offset + length) & ~PAGE_MASK)
				page_count--;	/* more on last page */
			pages += page_count;
		}
A
Alex Elder 已提交
2492

G
Guangliang Zhao 已提交
2493 2494 2495
		osd_req = rbd_osd_req_create(rbd_dev, op_type,
					(op_type == OBJ_OP_WRITE) ? 2 : 1,
					obj_request);
2496
		if (!osd_req)
2497
			goto out_unwind;
2498

2499
		obj_request->osd_req = osd_req;
2500
		obj_request->callback = rbd_img_obj_callback;
2501
		obj_request->img_offset = img_offset;
2502

2503
		rbd_img_obj_request_fill(obj_request, osd_req, op_type, 0);
A
Alex Elder 已提交
2504

2505
		img_offset += length;
A
Alex Elder 已提交
2506 2507 2508 2509 2510 2511 2512
		resid -= length;
	}

	return 0;

out_unwind:
	for_each_obj_request_safe(img_request, obj_request, next_obj_request)
2513
		rbd_img_obj_request_del(img_request, obj_request);
A
Alex Elder 已提交
2514 2515 2516 2517

	return -ENOMEM;
}

2518
static void
I
Ilya Dryomov 已提交
2519
rbd_osd_copyup_callback(struct rbd_obj_request *obj_request)
2520 2521 2522
{
	struct rbd_img_request *img_request;
	struct rbd_device *rbd_dev;
2523
	struct page **pages;
2524 2525
	u32 page_count;

I
Ilya Dryomov 已提交
2526 2527
	dout("%s: obj %p\n", __func__, obj_request);

2528 2529
	rbd_assert(obj_request->type == OBJ_REQUEST_BIO ||
		obj_request->type == OBJ_REQUEST_NODATA);
2530 2531 2532 2533 2534 2535 2536
	rbd_assert(obj_request_img_data_test(obj_request));
	img_request = obj_request->img_request;
	rbd_assert(img_request);

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

2537 2538
	pages = obj_request->copyup_pages;
	rbd_assert(pages != NULL);
2539
	obj_request->copyup_pages = NULL;
2540 2541 2542 2543
	page_count = obj_request->copyup_page_count;
	rbd_assert(page_count);
	obj_request->copyup_page_count = 0;
	ceph_release_page_vector(pages, page_count);
2544 2545 2546 2547 2548 2549 2550 2551 2552 2553

	/*
	 * We want the transfer count to reflect the size of the
	 * original write request.  There is no such thing as a
	 * successful short write, so if the request was successful
	 * we can just set it to the originally-requested length.
	 */
	if (!obj_request->result)
		obj_request->xferred = obj_request->length;

I
Ilya Dryomov 已提交
2554
	obj_request_done_set(obj_request);
2555 2556
}

2557 2558 2559 2560
static void
rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request)
{
	struct rbd_obj_request *orig_request;
2561 2562
	struct ceph_osd_request *osd_req;
	struct rbd_device *rbd_dev;
2563
	struct page **pages;
2564
	enum obj_operation_type op_type;
2565
	u32 page_count;
2566
	int img_result;
2567
	u64 parent_length;
2568 2569 2570 2571 2572 2573 2574 2575

	rbd_assert(img_request_child_test(img_request));

	/* First get what we need from the image request */

	pages = img_request->copyup_pages;
	rbd_assert(pages != NULL);
	img_request->copyup_pages = NULL;
2576 2577 2578
	page_count = img_request->copyup_page_count;
	rbd_assert(page_count);
	img_request->copyup_page_count = 0;
2579 2580 2581

	orig_request = img_request->obj_request;
	rbd_assert(orig_request != NULL);
2582
	rbd_assert(obj_request_type_valid(orig_request->type));
2583
	img_result = img_request->result;
2584
	parent_length = img_request->length;
2585
	rbd_assert(img_result || parent_length == img_request->xferred);
2586
	rbd_img_request_put(img_request);
2587

2588 2589
	rbd_assert(orig_request->img_request);
	rbd_dev = orig_request->img_request->rbd_dev;
2590 2591
	rbd_assert(rbd_dev);

2592 2593 2594 2595 2596 2597 2598
	/*
	 * If the overlap has become 0 (most likely because the
	 * image has been flattened) we need to free the pages
	 * and re-submit the original write request.
	 */
	if (!rbd_dev->parent_overlap) {
		ceph_release_page_vector(pages, page_count);
2599 2600
		rbd_obj_request_submit(orig_request);
		return;
2601
	}
2602

2603
	if (img_result)
2604 2605
		goto out_err;

2606 2607
	/*
	 * The original osd request is of no use to use any more.
2608
	 * We need a new one that can hold the three ops in a copyup
2609 2610 2611
	 * request.  Allocate the new copyup osd request for the
	 * original request, and release the old one.
	 */
2612
	img_result = -ENOMEM;
2613 2614 2615
	osd_req = rbd_osd_req_create_copyup(orig_request);
	if (!osd_req)
		goto out_err;
2616
	rbd_osd_req_destroy(orig_request->osd_req);
2617 2618
	orig_request->osd_req = osd_req;
	orig_request->copyup_pages = pages;
2619
	orig_request->copyup_page_count = page_count;
2620

2621
	/* Initialize the copyup op */
2622

2623
	osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup");
2624
	osd_req_op_cls_request_data_pages(osd_req, 0, pages, parent_length, 0,
2625
						false, false);
2626

2627
	/* Add the other op(s) */
2628

2629 2630
	op_type = rbd_img_request_op_type(orig_request->img_request);
	rbd_img_obj_request_fill(orig_request, osd_req, op_type, 1);
2631 2632 2633

	/* All set, send it off. */

2634 2635
	rbd_obj_request_submit(orig_request);
	return;
2636 2637

out_err:
2638
	ceph_release_page_vector(pages, page_count);
2639
	rbd_obj_request_error(orig_request, img_result);
2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652
}

/*
 * Read from the parent image the range of data that covers the
 * entire target of the given object request.  This is used for
 * satisfying a layered image write request when the target of an
 * object request from the image request does not exist.
 *
 * A page array big enough to hold the returned data is allocated
 * and supplied to rbd_img_request_fill() as the "data descriptor."
 * When the read completes, this page array will be transferred to
 * the original object request for the copyup operation.
 *
2653 2654
 * If an error occurs, it is recorded as the result of the original
 * object request in rbd_img_obj_exists_callback().
2655 2656 2657
 */
static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request)
{
2658
	struct rbd_device *rbd_dev = obj_request->img_request->rbd_dev;
2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672
	struct rbd_img_request *parent_request = NULL;
	u64 img_offset;
	u64 length;
	struct page **pages = NULL;
	u32 page_count;
	int result;

	rbd_assert(rbd_dev->parent != NULL);

	/*
	 * Determine the byte range covered by the object in the
	 * child image to which the original request was to be sent.
	 */
	img_offset = obj_request->img_offset - obj_request->offset;
I
Ilya Dryomov 已提交
2673
	length = rbd_obj_bytes(&rbd_dev->header);
2674

A
Alex Elder 已提交
2675 2676 2677 2678 2679 2680 2681 2682 2683 2684
	/*
	 * There is no defined parent data beyond the parent
	 * overlap, so limit what we read at that boundary if
	 * necessary.
	 */
	if (img_offset + length > rbd_dev->parent_overlap) {
		rbd_assert(img_offset < rbd_dev->parent_overlap);
		length = rbd_dev->parent_overlap - img_offset;
	}

2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697
	/*
	 * Allocate a page array big enough to receive the data read
	 * from the parent.
	 */
	page_count = (u32)calc_pages_for(0, length);
	pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
	if (IS_ERR(pages)) {
		result = PTR_ERR(pages);
		pages = NULL;
		goto out_err;
	}

	result = -ENOMEM;
2698 2699
	parent_request = rbd_parent_request_create(obj_request,
						img_offset, length);
2700 2701 2702 2703 2704 2705
	if (!parent_request)
		goto out_err;

	result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages);
	if (result)
		goto out_err;
2706

2707
	parent_request->copyup_pages = pages;
2708
	parent_request->copyup_page_count = page_count;
2709
	parent_request->callback = rbd_img_obj_parent_read_full_callback;
2710

2711 2712 2713 2714 2715
	result = rbd_img_request_submit(parent_request);
	if (!result)
		return 0;

	parent_request->copyup_pages = NULL;
2716
	parent_request->copyup_page_count = 0;
2717 2718 2719 2720 2721 2722 2723 2724 2725 2726
	parent_request->obj_request = NULL;
	rbd_obj_request_put(obj_request);
out_err:
	if (pages)
		ceph_release_page_vector(pages, page_count);
	if (parent_request)
		rbd_img_request_put(parent_request);
	return result;
}

2727 2728 2729
static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request)
{
	struct rbd_obj_request *orig_request;
2730
	struct rbd_device *rbd_dev;
2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741
	int result;

	rbd_assert(!obj_request_img_data_test(obj_request));

	/*
	 * All we need from the object request is the original
	 * request and the result of the STAT op.  Grab those, then
	 * we're done with the request.
	 */
	orig_request = obj_request->obj_request;
	obj_request->obj_request = NULL;
2742
	rbd_obj_request_put(orig_request);
2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753
	rbd_assert(orig_request);
	rbd_assert(orig_request->img_request);

	result = obj_request->result;
	obj_request->result = 0;

	dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__,
		obj_request, orig_request, result,
		obj_request->xferred, obj_request->length);
	rbd_obj_request_put(obj_request);

2754 2755
	/*
	 * If the overlap has become 0 (most likely because the
2756 2757
	 * image has been flattened) we need to re-submit the
	 * original request.
2758 2759 2760
	 */
	rbd_dev = orig_request->img_request->rbd_dev;
	if (!rbd_dev->parent_overlap) {
2761 2762
		rbd_obj_request_submit(orig_request);
		return;
2763
	}
2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774

	/*
	 * Our only purpose here is to determine whether the object
	 * exists, and we don't want to treat the non-existence as
	 * an error.  If something else comes back, transfer the
	 * error to the original request and complete it now.
	 */
	if (!result) {
		obj_request_existence_set(orig_request, true);
	} else if (result == -ENOENT) {
		obj_request_existence_set(orig_request, false);
2775 2776
	} else {
		goto fail_orig_request;
2777 2778 2779 2780 2781 2782
	}

	/*
	 * Resubmit the original request now that we have recorded
	 * whether the target object exists.
	 */
2783 2784 2785 2786 2787 2788 2789
	result = rbd_img_obj_request_submit(orig_request);
	if (result)
		goto fail_orig_request;

	return;

fail_orig_request:
2790
	rbd_obj_request_error(orig_request, result);
2791 2792 2793 2794
}

static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request)
{
2795
	struct rbd_device *rbd_dev = obj_request->img_request->rbd_dev;
2796
	struct rbd_obj_request *stat_request;
2797
	struct page **pages;
2798 2799 2800 2801
	u32 page_count;
	size_t size;
	int ret;

2802
	stat_request = rbd_obj_request_create(OBJ_REQUEST_PAGES);
2803 2804 2805
	if (!stat_request)
		return -ENOMEM;

2806 2807
	stat_request->object_no = obj_request->object_no;

2808 2809 2810 2811 2812 2813 2814
	stat_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1,
						   stat_request);
	if (!stat_request->osd_req) {
		ret = -ENOMEM;
		goto fail_stat_request;
	}

2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825
	/*
	 * The response data for a STAT call consists of:
	 *     le64 length;
	 *     struct {
	 *         le32 tv_sec;
	 *         le32 tv_nsec;
	 *     } mtime;
	 */
	size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32);
	page_count = (u32)calc_pages_for(0, size);
	pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
2826 2827 2828 2829
	if (IS_ERR(pages)) {
		ret = PTR_ERR(pages);
		goto fail_stat_request;
	}
2830

2831 2832 2833
	osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT, 0);
	osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0,
				     false, false);
2834 2835 2836 2837 2838 2839 2840

	rbd_obj_request_get(obj_request);
	stat_request->obj_request = obj_request;
	stat_request->pages = pages;
	stat_request->page_count = page_count;
	stat_request->callback = rbd_img_obj_exists_callback;

2841 2842
	rbd_obj_request_submit(stat_request);
	return 0;
2843

2844 2845
fail_stat_request:
	rbd_obj_request_put(stat_request);
2846 2847 2848
	return ret;
}

2849
static bool img_obj_request_simple(struct rbd_obj_request *obj_request)
2850
{
2851 2852
	struct rbd_img_request *img_request = obj_request->img_request;
	struct rbd_device *rbd_dev = img_request->rbd_dev;
2853

2854
	/* Reads */
2855 2856
	if (!img_request_write_test(img_request) &&
	    !img_request_discard_test(img_request))
2857 2858 2859 2860 2861 2862
		return true;

	/* Non-layered writes */
	if (!img_request_layered_test(img_request))
		return true;

2863
	/*
2864 2865
	 * Layered writes outside of the parent overlap range don't
	 * share any data with the parent.
2866
	 */
2867 2868
	if (!obj_request_overlaps_parent(obj_request))
		return true;
2869

2870 2871 2872 2873 2874 2875 2876 2877
	/*
	 * Entire-object layered writes - we will overwrite whatever
	 * parent data there is anyway.
	 */
	if (!obj_request->offset &&
	    obj_request->length == rbd_obj_bytes(&rbd_dev->header))
		return true;

2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890
	/*
	 * If the object is known to already exist, its parent data has
	 * already been copied.
	 */
	if (obj_request_known_test(obj_request) &&
	    obj_request_exists_test(obj_request))
		return true;

	return false;
}

static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request)
{
2891 2892 2893
	rbd_assert(obj_request_img_data_test(obj_request));
	rbd_assert(obj_request_type_valid(obj_request->type));
	rbd_assert(obj_request->img_request);
2894

2895
	if (img_obj_request_simple(obj_request)) {
2896 2897
		rbd_obj_request_submit(obj_request);
		return 0;
2898 2899 2900
	}

	/*
2901 2902 2903 2904
	 * It's a layered write.  The target object might exist but
	 * we may not know that yet.  If we know it doesn't exist,
	 * start by reading the data for the full target object from
	 * the parent so we can use it for a copyup to the target.
2905
	 */
2906
	if (obj_request_known_test(obj_request))
2907 2908 2909
		return rbd_img_obj_parent_read_full(obj_request);

	/* We don't know whether the target exists.  Go find out. */
2910 2911 2912 2913

	return rbd_img_obj_exists_submit(obj_request);
}

A
Alex Elder 已提交
2914 2915 2916
static int rbd_img_request_submit(struct rbd_img_request *img_request)
{
	struct rbd_obj_request *obj_request;
2917
	struct rbd_obj_request *next_obj_request;
2918
	int ret = 0;
A
Alex Elder 已提交
2919

A
Alex Elder 已提交
2920
	dout("%s: img %p\n", __func__, img_request);
A
Alex Elder 已提交
2921

2922 2923
	rbd_img_request_get(img_request);
	for_each_obj_request_safe(img_request, obj_request, next_obj_request) {
2924
		ret = rbd_img_obj_request_submit(obj_request);
A
Alex Elder 已提交
2925
		if (ret)
2926
			goto out_put_ireq;
A
Alex Elder 已提交
2927 2928
	}

2929 2930 2931
out_put_ireq:
	rbd_img_request_put(img_request);
	return ret;
A
Alex Elder 已提交
2932
}
A
Alex Elder 已提交
2933 2934 2935 2936

static void rbd_img_parent_read_callback(struct rbd_img_request *img_request)
{
	struct rbd_obj_request *obj_request;
A
Alex Elder 已提交
2937 2938
	struct rbd_device *rbd_dev;
	u64 obj_end;
2939 2940
	u64 img_xferred;
	int img_result;
A
Alex Elder 已提交
2941 2942 2943

	rbd_assert(img_request_child_test(img_request));

2944 2945
	/* First get what we need from the image request and release it */

A
Alex Elder 已提交
2946
	obj_request = img_request->obj_request;
2947 2948 2949 2950 2951 2952 2953 2954 2955
	img_xferred = img_request->xferred;
	img_result = img_request->result;
	rbd_img_request_put(img_request);

	/*
	 * If the overlap has become 0 (most likely because the
	 * image has been flattened) we need to re-submit the
	 * original request.
	 */
A
Alex Elder 已提交
2956 2957
	rbd_assert(obj_request);
	rbd_assert(obj_request->img_request);
2958 2959
	rbd_dev = obj_request->img_request->rbd_dev;
	if (!rbd_dev->parent_overlap) {
2960 2961
		rbd_obj_request_submit(obj_request);
		return;
2962
	}
A
Alex Elder 已提交
2963

2964
	obj_request->result = img_result;
A
Alex Elder 已提交
2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982
	if (obj_request->result)
		goto out;

	/*
	 * We need to zero anything beyond the parent overlap
	 * boundary.  Since rbd_img_obj_request_read_callback()
	 * will zero anything beyond the end of a short read, an
	 * easy way to do this is to pretend the data from the
	 * parent came up short--ending at the overlap boundary.
	 */
	rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length);
	obj_end = obj_request->img_offset + obj_request->length;
	if (obj_end > rbd_dev->parent_overlap) {
		u64 xferred = 0;

		if (obj_request->img_offset < rbd_dev->parent_overlap)
			xferred = rbd_dev->parent_overlap -
					obj_request->img_offset;
A
Alex Elder 已提交
2983

2984
		obj_request->xferred = min(img_xferred, xferred);
A
Alex Elder 已提交
2985
	} else {
2986
		obj_request->xferred = img_xferred;
A
Alex Elder 已提交
2987 2988
	}
out:
A
Alex Elder 已提交
2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000
	rbd_img_obj_request_read_callback(obj_request);
	rbd_obj_request_complete(obj_request);
}

static void rbd_img_parent_read(struct rbd_obj_request *obj_request)
{
	struct rbd_img_request *img_request;
	int result;

	rbd_assert(obj_request_img_data_test(obj_request));
	rbd_assert(obj_request->img_request != NULL);
	rbd_assert(obj_request->result == (s32) -ENOENT);
3001
	rbd_assert(obj_request_type_valid(obj_request->type));
A
Alex Elder 已提交
3002 3003

	/* rbd_read_finish(obj_request, obj_request->length); */
3004
	img_request = rbd_parent_request_create(obj_request,
A
Alex Elder 已提交
3005
						obj_request->img_offset,
3006
						obj_request->length);
A
Alex Elder 已提交
3007 3008 3009 3010
	result = -ENOMEM;
	if (!img_request)
		goto out_err;

3011 3012 3013 3014 3015 3016
	if (obj_request->type == OBJ_REQUEST_BIO)
		result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO,
						obj_request->bio_list);
	else
		result = rbd_img_request_fill(img_request, OBJ_REQUEST_PAGES,
						obj_request->pages);
A
Alex Elder 已提交
3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032
	if (result)
		goto out_err;

	img_request->callback = rbd_img_parent_read_callback;
	result = rbd_img_request_submit(img_request);
	if (result)
		goto out_err;

	return;
out_err:
	if (img_request)
		rbd_img_request_put(img_request);
	obj_request->result = result;
	obj_request->xferred = 0;
	obj_request_done_set(obj_request);
}
A
Alex Elder 已提交
3033

3034
static const struct rbd_client_id rbd_empty_cid;
A
Alex Elder 已提交
3035

3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075
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);
}

/*
 * lock_rwsem must be held for write
 */
static int rbd_lock(struct rbd_device *rbd_dev)
A
Alex Elder 已提交
3076
{
3077
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3078 3079
	struct rbd_client_id cid = rbd_get_cid(rbd_dev);
	char cookie[32];
3080
	int ret;
A
Alex Elder 已提交
3081

3082
	WARN_ON(__rbd_is_lock_owner(rbd_dev));
3083

3084 3085 3086 3087
	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);
3088
	if (ret)
3089
		return ret;
A
Alex Elder 已提交
3090

3091 3092 3093 3094
	rbd_dev->lock_state = RBD_LOCK_STATE_LOCKED;
	rbd_set_owner_cid(rbd_dev, &cid);
	queue_work(rbd_dev->task_wq, &rbd_dev->acquired_lock_work);
	return 0;
A
Alex Elder 已提交
3095 3096
}

3097 3098 3099 3100
/*
 * lock_rwsem must be held for write
 */
static int rbd_unlock(struct rbd_device *rbd_dev)
3101
{
3102
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3103
	char cookie[32];
3104 3105
	int ret;

3106
	WARN_ON(!__rbd_is_lock_owner(rbd_dev));
3107

3108
	rbd_dev->lock_state = RBD_LOCK_STATE_UNLOCKED;
3109

3110 3111 3112 3113 3114 3115
	format_lock_cookie(rbd_dev, cookie);
	ret = ceph_cls_unlock(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc,
			      RBD_LOCK_NAME, cookie);
	if (ret && ret != -ENOENT) {
		rbd_warn(rbd_dev, "cls_unlock failed: %d", ret);
		return ret;
3116 3117
	}

3118 3119 3120
	rbd_set_owner_cid(rbd_dev, &rbd_empty_cid);
	queue_work(rbd_dev->task_wq, &rbd_dev->released_lock_work);
	return 0;
3121 3122
}

3123 3124 3125 3126
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)
3127 3128
{
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3129 3130 3131 3132
	struct rbd_client_id cid = rbd_get_cid(rbd_dev);
	int buf_size = 4 + 8 + 8 + CEPH_ENCODING_START_BLK_LEN;
	char buf[buf_size];
	void *p = buf;
3133

3134
	dout("%s rbd_dev %p notify_op %d\n", __func__, rbd_dev, notify_op);
3135

3136 3137 3138 3139 3140
	/* 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);
3141

3142 3143 3144
	return ceph_osdc_notify(osdc, &rbd_dev->header_oid,
				&rbd_dev->header_oloc, buf, buf_size,
				RBD_NOTIFY_TIMEOUT, preply_pages, preply_len);
3145 3146
}

3147 3148
static void rbd_notify_op_lock(struct rbd_device *rbd_dev,
			       enum rbd_notify_op notify_op)
3149
{
3150 3151
	struct page **reply_pages;
	size_t reply_len;
3152

3153 3154 3155
	__rbd_notify_op_lock(rbd_dev, notify_op, &reply_pages, &reply_len);
	ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len));
}
3156

3157 3158 3159 3160
static void rbd_notify_acquired_lock(struct work_struct *work)
{
	struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
						  acquired_lock_work);
3161

3162
	rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_ACQUIRED_LOCK);
3163 3164
}

3165
static void rbd_notify_released_lock(struct work_struct *work)
3166
{
3167 3168
	struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
						  released_lock_work);
I
Ilya Dryomov 已提交
3169

3170
	rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_RELEASED_LOCK);
3171 3172
}

3173
static int rbd_request_lock(struct rbd_device *rbd_dev)
3174
{
3175 3176 3177
	struct page **reply_pages;
	size_t reply_len;
	bool lock_owner_responded = false;
3178 3179
	int ret;

3180
	dout("%s rbd_dev %p\n", __func__, rbd_dev);
3181

3182 3183 3184 3185
	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);
3186
		goto out;
3187
	}
3188

3189 3190 3191 3192
	if (reply_len > 0 && reply_len <= PAGE_SIZE) {
		void *p = page_address(reply_pages[0]);
		void *const end = p + reply_len;
		u32 n;
3193

3194 3195 3196 3197
		ceph_decode_32_safe(&p, end, n, e_inval); /* num_acks */
		while (n--) {
			u8 struct_v;
			u32 len;
3198

3199 3200
			ceph_decode_need(&p, end, 8 + 8, e_inval);
			p += 8 + 8; /* skip gid and cookie */
3201

3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283 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 3316 3317 3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 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 3473 3474 3475 3476 3477
			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;
	int ret;

	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 */
	} 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);
3478
	/*
3479
	 * Ensure that all in-flight IO is flushed.
3480
	 *
3481 3482
	 * FIXME: ceph_osdc_sync() flushes the entire OSD client, which
	 * may be shared with other devices.
3483
	 */
3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585 3586 3587 3588 3589 3590 3591 3592 3593 3594 3595 3596 3597 3598 3599 3600 3601 3602 3603 3604 3605 3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630
	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;

	if (!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);

	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);
}

static bool rbd_handle_request_lock(struct rbd_device *rbd_dev, u8 struct_v,
				    void **p)
{
	struct rbd_client_id my_cid = rbd_get_cid(rbd_dev);
	struct rbd_client_id cid = { 0 };
	bool need_to_send;

	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))
		return false;

	down_read(&rbd_dev->lock_rwsem);
	need_to_send = __rbd_is_lock_owner(rbd_dev);
	if (rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED) {
		if (!rbd_cid_equal(&rbd_dev->owner_cid, &rbd_empty_cid)) {
			dout("%s rbd_dev %p queueing unlock_work\n", __func__,
			     rbd_dev);
			queue_work(rbd_dev->task_wq, &rbd_dev->unlock_work);
		}
	}
	up_read(&rbd_dev->lock_rwsem);
	return need_to_send;
}

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

	if (result) {
		void *p = buf;

		/* encode ResponseMessage */
		ceph_start_encoding(&p, 1, 1,
				    buf_size - CEPH_ENCODING_START_BLK_LEN);
		ceph_encode_32(&p, *result);
	} else {
		buf_size = 0;
	}
A
Alex Elder 已提交
3631

3632 3633
	ret = ceph_osdc_notify_ack(osdc, &rbd_dev->header_oid,
				   &rbd_dev->header_oloc, notify_id, cookie,
3634
				   buf, buf_size);
3635
	if (ret)
3636 3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 3650 3651 3652 3653 3654 3655 3656 3657 3658
		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;
3659
	u8 struct_v = 0;
3660 3661 3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682 3683 3684 3685 3686 3687 3688 3689 3690 3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714 3715 3716 3717
	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:
		if (rbd_handle_request_lock(rbd_dev, struct_v, &p))
			/*
			 * send ResponseMessage(0) back so the client
			 * can detect a missing owner
			 */
			rbd_acknowledge_notify_result(rbd_dev, notify_id,
						      cookie, 0);
		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
3718 3719
}

3720 3721
static void __rbd_unregister_watch(struct rbd_device *rbd_dev);

3722
static void rbd_watch_errcb(void *arg, u64 cookie, int err)
3723
{
3724
	struct rbd_device *rbd_dev = arg;
3725

3726
	rbd_warn(rbd_dev, "encountered watch error: %d", err);
3727

3728 3729 3730 3731
	down_write(&rbd_dev->lock_rwsem);
	rbd_set_owner_cid(rbd_dev, &rbd_empty_cid);
	up_write(&rbd_dev->lock_rwsem);

3732 3733 3734 3735
	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;
3736

3737
		queue_delayed_work(rbd_dev->task_wq, &rbd_dev->watch_dwork, 0);
3738
	}
3739
	mutex_unlock(&rbd_dev->watch_mutex);
3740 3741
}

3742
/*
3743
 * watch_mutex must be locked
3744
 */
3745
static int __rbd_register_watch(struct rbd_device *rbd_dev)
3746 3747
{
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3748
	struct ceph_osd_linger_request *handle;
3749

3750
	rbd_assert(!rbd_dev->watch_handle);
3751
	dout("%s rbd_dev %p\n", __func__, rbd_dev);
3752

3753 3754 3755 3756 3757
	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);
3758

3759
	rbd_dev->watch_handle = handle;
3760 3761 3762
	return 0;
}

3763 3764 3765 3766
/*
 * watch_mutex must be locked
 */
static void __rbd_unregister_watch(struct rbd_device *rbd_dev)
3767
{
3768 3769
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
	int ret;
3770

3771 3772
	rbd_assert(rbd_dev->watch_handle);
	dout("%s rbd_dev %p\n", __func__, rbd_dev);
3773

3774 3775 3776
	ret = ceph_osdc_unwatch(osdc, rbd_dev->watch_handle);
	if (ret)
		rbd_warn(rbd_dev, "failed to unwatch: %d", ret);
3777

3778
	rbd_dev->watch_handle = NULL;
3779 3780
}

3781 3782 3783 3784 3785 3786 3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799
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)
3800
{
3801 3802 3803
	dout("%s rbd_dev %p\n", __func__, rbd_dev);

	cancel_delayed_work_sync(&rbd_dev->watch_dwork);
3804 3805 3806 3807
	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);
3808 3809 3810 3811
}

static void rbd_unregister_watch(struct rbd_device *rbd_dev)
{
3812
	WARN_ON(waitqueue_active(&rbd_dev->lock_waitq));
3813 3814 3815 3816 3817 3818 3819
	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
3820 3821

	ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc);
3822 3823
}

3824 3825 3826 3827
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);
3828
	bool was_lock_owner = false;
3829
	bool need_to_wake = false;
3830 3831 3832 3833
	int ret;

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

3834 3835 3836 3837
	down_write(&rbd_dev->lock_rwsem);
	if (rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED)
		was_lock_owner = rbd_release_lock(rbd_dev);

3838
	mutex_lock(&rbd_dev->watch_mutex);
3839 3840 3841 3842
	if (rbd_dev->watch_state != RBD_WATCH_STATE_ERROR) {
		mutex_unlock(&rbd_dev->watch_mutex);
		goto out;
	}
3843 3844 3845 3846

	ret = __rbd_register_watch(rbd_dev);
	if (ret) {
		rbd_warn(rbd_dev, "failed to reregister watch: %d", ret);
3847
		if (ret == -EBLACKLISTED || ret == -ENOENT) {
3848 3849 3850
			set_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags);
			need_to_wake = true;
		} else {
3851 3852 3853
			queue_delayed_work(rbd_dev->task_wq,
					   &rbd_dev->watch_dwork,
					   RBD_RETRY_DELAY);
3854 3855 3856
		}
		mutex_unlock(&rbd_dev->watch_mutex);
		goto out;
3857 3858
	}

3859
	need_to_wake = true;
3860 3861 3862 3863 3864 3865 3866 3867
	rbd_dev->watch_state = RBD_WATCH_STATE_REGISTERED;
	rbd_dev->watch_cookie = rbd_dev->watch_handle->linger_id;
	mutex_unlock(&rbd_dev->watch_mutex);

	ret = rbd_dev_refresh(rbd_dev);
	if (ret)
		rbd_warn(rbd_dev, "reregisteration refresh failed: %d", ret);

3868 3869 3870 3871 3872 3873 3874
	if (was_lock_owner) {
		ret = rbd_try_lock(rbd_dev);
		if (ret)
			rbd_warn(rbd_dev, "reregisteration lock failed: %d",
				 ret);
	}

3875
out:
3876
	up_write(&rbd_dev->lock_rwsem);
3877 3878
	if (need_to_wake)
		wake_requests(rbd_dev, true);
3879 3880
}

3881
/*
3882 3883
 * Synchronous osd object method call.  Returns the number of bytes
 * returned in the outbound buffer, or a negative error code.
3884 3885
 */
static int rbd_obj_method_sync(struct rbd_device *rbd_dev,
3886 3887
			     struct ceph_object_id *oid,
			     struct ceph_object_locator *oloc,
3888
			     const char *method_name,
3889
			     const void *outbound,
3890
			     size_t outbound_size,
3891
			     void *inbound,
3892
			     size_t inbound_size)
3893
{
3894 3895 3896
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
	struct page *req_page = NULL;
	struct page *reply_page;
3897 3898 3899
	int ret;

	/*
3900 3901 3902 3903 3904
	 * 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.
3905
	 */
3906 3907 3908
	if (outbound) {
		if (outbound_size > PAGE_SIZE)
			return -E2BIG;
3909

3910 3911 3912
		req_page = alloc_page(GFP_KERNEL);
		if (!req_page)
			return -ENOMEM;
3913

3914
		memcpy(page_address(req_page), outbound, outbound_size);
3915
	}
3916

3917 3918 3919 3920 3921 3922
	reply_page = alloc_page(GFP_KERNEL);
	if (!reply_page) {
		if (req_page)
			__free_page(req_page);
		return -ENOMEM;
	}
3923

3924 3925 3926 3927 3928 3929 3930
	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;
	}
3931

3932 3933 3934
	if (req_page)
		__free_page(req_page);
	__free_page(reply_page);
3935 3936 3937
	return ret;
}

3938 3939 3940 3941 3942 3943 3944 3945 3946 3947 3948 3949 3950 3951 3952 3953 3954 3955 3956
/*
 * lock_rwsem must be held for read
 */
static void rbd_wait_state_locked(struct rbd_device *rbd_dev)
{
	DEFINE_WAIT(wait);

	do {
		/*
		 * Note the use of mod_delayed_work() in rbd_acquire_lock()
		 * and cancel_delayed_work() in wake_requests().
		 */
		dout("%s rbd_dev %p queueing lock_dwork\n", __func__, rbd_dev);
		queue_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork, 0);
		prepare_to_wait_exclusive(&rbd_dev->lock_waitq, &wait,
					  TASK_UNINTERRUPTIBLE);
		up_read(&rbd_dev->lock_rwsem);
		schedule();
		down_read(&rbd_dev->lock_rwsem);
3957 3958 3959
	} while (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED &&
		 !test_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags));

3960 3961 3962
	finish_wait(&rbd_dev->lock_waitq, &wait);
}

C
Christoph Hellwig 已提交
3963
static void rbd_queue_workfn(struct work_struct *work)
A
Alex Elder 已提交
3964
{
C
Christoph Hellwig 已提交
3965 3966
	struct request *rq = blk_mq_rq_from_pdu(work);
	struct rbd_device *rbd_dev = rq->q->queuedata;
I
Ilya Dryomov 已提交
3967
	struct rbd_img_request *img_request;
3968
	struct ceph_snap_context *snapc = NULL;
I
Ilya Dryomov 已提交
3969 3970
	u64 offset = (u64)blk_rq_pos(rq) << SECTOR_SHIFT;
	u64 length = blk_rq_bytes(rq);
G
Guangliang Zhao 已提交
3971
	enum obj_operation_type op_type;
3972
	u64 mapping_size;
I
Ilya Dryomov 已提交
3973
	bool must_be_locked;
A
Alex Elder 已提交
3974 3975
	int result;

3976 3977
	switch (req_op(rq)) {
	case REQ_OP_DISCARD:
3978
		op_type = OBJ_OP_DISCARD;
3979 3980
		break;
	case REQ_OP_WRITE:
G
Guangliang Zhao 已提交
3981
		op_type = OBJ_OP_WRITE;
3982 3983
		break;
	case REQ_OP_READ:
G
Guangliang Zhao 已提交
3984
		op_type = OBJ_OP_READ;
3985 3986 3987 3988 3989 3990
		break;
	default:
		dout("%s: non-fs request type %d\n", __func__, req_op(rq));
		result = -EIO;
		goto err;
	}
G
Guangliang Zhao 已提交
3991

I
Ilya Dryomov 已提交
3992
	/* Ignore/skip any zero-length requests */
A
Alex Elder 已提交
3993

I
Ilya Dryomov 已提交
3994 3995 3996 3997 3998
	if (!length) {
		dout("%s: zero-length request\n", __func__);
		result = 0;
		goto err_rq;
	}
A
Alex Elder 已提交
3999

G
Guangliang Zhao 已提交
4000
	/* Only reads are allowed to a read-only device */
I
Ilya Dryomov 已提交
4001

G
Guangliang Zhao 已提交
4002
	if (op_type != OBJ_OP_READ) {
I
Ilya Dryomov 已提交
4003 4004 4005
		if (rbd_dev->mapping.read_only) {
			result = -EROFS;
			goto err_rq;
A
Alex Elder 已提交
4006
		}
I
Ilya Dryomov 已提交
4007 4008
		rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP);
	}
A
Alex Elder 已提交
4009

I
Ilya Dryomov 已提交
4010 4011 4012 4013 4014 4015 4016 4017 4018 4019 4020 4021
	/*
	 * 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
4022

I
Ilya Dryomov 已提交
4023 4024 4025 4026 4027 4028
	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
4029

C
Christoph Hellwig 已提交
4030 4031
	blk_mq_start_request(rq);

4032 4033
	down_read(&rbd_dev->header_rwsem);
	mapping_size = rbd_dev->mapping.size;
G
Guangliang Zhao 已提交
4034
	if (op_type != OBJ_OP_READ) {
4035 4036
		snapc = rbd_dev->header.snapc;
		ceph_get_snap_context(snapc);
4037
		must_be_locked = rbd_is_lock_supported(rbd_dev);
I
Ilya Dryomov 已提交
4038 4039 4040
	} else {
		must_be_locked = rbd_dev->opts->lock_on_read &&
					rbd_is_lock_supported(rbd_dev);
4041 4042 4043 4044
	}
	up_read(&rbd_dev->header_rwsem);

	if (offset + length > mapping_size) {
I
Ilya Dryomov 已提交
4045
		rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)", offset,
4046
			 length, mapping_size);
I
Ilya Dryomov 已提交
4047 4048 4049
		result = -EIO;
		goto err_rq;
	}
A
Alex Elder 已提交
4050

4051 4052
	if (must_be_locked) {
		down_read(&rbd_dev->lock_rwsem);
4053 4054
		if (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED &&
		    !test_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags))
4055
			rbd_wait_state_locked(rbd_dev);
4056 4057 4058 4059 4060 4061 4062

		WARN_ON((rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED) ^
			!test_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags));
		if (test_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags)) {
			result = -EBLACKLISTED;
			goto err_unlock;
		}
4063 4064
	}

G
Guangliang Zhao 已提交
4065
	img_request = rbd_img_request_create(rbd_dev, offset, length, op_type,
4066
					     snapc);
I
Ilya Dryomov 已提交
4067 4068
	if (!img_request) {
		result = -ENOMEM;
4069
		goto err_unlock;
I
Ilya Dryomov 已提交
4070 4071
	}
	img_request->rq = rq;
4072
	snapc = NULL; /* img_request consumes a ref */
A
Alex Elder 已提交
4073

4074 4075 4076 4077 4078 4079
	if (op_type == OBJ_OP_DISCARD)
		result = rbd_img_request_fill(img_request, OBJ_REQUEST_NODATA,
					      NULL);
	else
		result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO,
					      rq->bio);
I
Ilya Dryomov 已提交
4080 4081
	if (result)
		goto err_img_request;
A
Alex Elder 已提交
4082

I
Ilya Dryomov 已提交
4083 4084 4085
	result = rbd_img_request_submit(img_request);
	if (result)
		goto err_img_request;
A
Alex Elder 已提交
4086

4087 4088
	if (must_be_locked)
		up_read(&rbd_dev->lock_rwsem);
I
Ilya Dryomov 已提交
4089
	return;
A
Alex Elder 已提交
4090

I
Ilya Dryomov 已提交
4091 4092
err_img_request:
	rbd_img_request_put(img_request);
4093 4094 4095
err_unlock:
	if (must_be_locked)
		up_read(&rbd_dev->lock_rwsem);
I
Ilya Dryomov 已提交
4096 4097 4098
err_rq:
	if (result)
		rbd_warn(rbd_dev, "%s %llx at %llx result %d",
G
Guangliang Zhao 已提交
4099
			 obj_op_name(op_type), length, offset, result);
4100
	ceph_put_snap_context(snapc);
C
Christoph Hellwig 已提交
4101 4102
err:
	blk_mq_end_request(rq, result);
I
Ilya Dryomov 已提交
4103
}
A
Alex Elder 已提交
4104

C
Christoph Hellwig 已提交
4105 4106
static int rbd_queue_rq(struct blk_mq_hw_ctx *hctx,
		const struct blk_mq_queue_data *bd)
I
Ilya Dryomov 已提交
4107
{
C
Christoph Hellwig 已提交
4108 4109
	struct request *rq = bd->rq;
	struct work_struct *work = blk_mq_rq_to_pdu(rq);
A
Alex Elder 已提交
4110

C
Christoph Hellwig 已提交
4111 4112
	queue_work(rbd_wq, work);
	return BLK_MQ_RQ_QUEUE_OK;
A
Alex Elder 已提交
4113 4114
}

4115 4116 4117 4118 4119 4120 4121
static void rbd_free_disk(struct rbd_device *rbd_dev)
{
	struct gendisk *disk = rbd_dev->disk;

	if (!disk)
		return;

4122 4123
	rbd_dev->disk = NULL;
	if (disk->flags & GENHD_FL_UP) {
4124
		del_gendisk(disk);
4125 4126
		if (disk->queue)
			blk_cleanup_queue(disk->queue);
C
Christoph Hellwig 已提交
4127
		blk_mq_free_tag_set(&rbd_dev->tag_set);
4128
	}
4129 4130 4131
	put_disk(disk);
}

4132
static int rbd_obj_read_sync(struct rbd_device *rbd_dev,
4133 4134 4135
			     struct ceph_object_id *oid,
			     struct ceph_object_locator *oloc,
			     void *buf, int buf_len)
4136 4137

{
4138 4139 4140 4141
	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);
4142 4143
	int ret;

4144 4145 4146
	req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_KERNEL);
	if (!req)
		return -ENOMEM;
4147

4148 4149 4150
	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
4151

4152
	ret = ceph_osdc_alloc_messages(req, GFP_KERNEL);
4153
	if (ret)
4154
		goto out_req;
4155

4156 4157 4158 4159 4160
	pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
	if (IS_ERR(pages)) {
		ret = PTR_ERR(pages);
		goto out_req;
	}
4161

4162 4163 4164 4165 4166 4167 4168 4169
	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);
4170

4171 4172
out_req:
	ceph_osdc_put_request(req);
4173 4174 4175
	return ret;
}

4176
/*
A
Alex Elder 已提交
4177 4178 4179
 * 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.
4180
 */
4181
static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev)
4182
{
4183
	struct rbd_image_header_ondisk *ondisk = NULL;
4184
	u32 snap_count = 0;
4185 4186 4187
	u64 names_size = 0;
	u32 want_count;
	int ret;
4188

A
Alex Elder 已提交
4189
	/*
4190 4191 4192 4193 4194
	 * 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
4195
	 */
4196 4197 4198 4199 4200 4201 4202 4203 4204 4205
	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
4206
			return -ENOMEM;
4207

4208 4209
		ret = rbd_obj_read_sync(rbd_dev, &rbd_dev->header_oid,
					&rbd_dev->header_oloc, ondisk, size);
4210
		if (ret < 0)
A
Alex Elder 已提交
4211
			goto out;
A
Alex Elder 已提交
4212
		if ((size_t)ret < size) {
4213
			ret = -ENXIO;
A
Alex Elder 已提交
4214 4215
			rbd_warn(rbd_dev, "short header read (want %zd got %d)",
				size, ret);
A
Alex Elder 已提交
4216
			goto out;
4217 4218 4219
		}
		if (!rbd_dev_ondisk_valid(ondisk)) {
			ret = -ENXIO;
A
Alex Elder 已提交
4220
			rbd_warn(rbd_dev, "invalid header");
A
Alex Elder 已提交
4221
			goto out;
4222
		}
4223

4224 4225 4226 4227
		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
4228

A
Alex Elder 已提交
4229 4230
	ret = rbd_header_from_disk(rbd_dev, ondisk);
out:
4231 4232 4233
	kfree(ondisk);

	return ret;
4234 4235
}

4236 4237 4238 4239 4240 4241 4242 4243 4244 4245 4246 4247 4248 4249 4250 4251 4252 4253 4254
/*
 * 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);
}

4255 4256 4257 4258 4259
static void rbd_dev_update_size(struct rbd_device *rbd_dev)
{
	sector_t size;

	/*
I
Ilya Dryomov 已提交
4260 4261 4262
	 * 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.
4263
	 */
I
Ilya Dryomov 已提交
4264 4265
	if (test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags) &&
	    !test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) {
4266 4267 4268 4269 4270 4271 4272
		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
4273
static int rbd_dev_refresh(struct rbd_device *rbd_dev)
A
Alex Elder 已提交
4274
{
4275
	u64 mapping_size;
A
Alex Elder 已提交
4276 4277
	int ret;

4278
	down_write(&rbd_dev->header_rwsem);
4279
	mapping_size = rbd_dev->mapping.size;
4280 4281

	ret = rbd_dev_header_info(rbd_dev);
4282
	if (ret)
4283
		goto out;
4284

4285 4286 4287 4288 4289 4290 4291
	/*
	 * 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)
4292
			goto out;
4293 4294
	}

4295
	if (rbd_dev->spec->snap_id == CEPH_NOSNAP) {
4296
		rbd_dev->mapping.size = rbd_dev->header.image_size;
4297 4298 4299 4300
	} else {
		/* validate mapped snapshot's EXISTS flag */
		rbd_exists_validate(rbd_dev);
	}
4301

4302
out:
4303
	up_write(&rbd_dev->header_rwsem);
4304
	if (!ret && mapping_size != rbd_dev->mapping.size)
4305
		rbd_dev_update_size(rbd_dev);
A
Alex Elder 已提交
4306

4307
	return ret;
A
Alex Elder 已提交
4308 4309
}

C
Christoph Hellwig 已提交
4310 4311 4312 4313 4314 4315 4316 4317 4318 4319 4320 4321 4322 4323 4324
static int rbd_init_request(void *data, struct request *rq,
		unsigned int hctx_idx, unsigned int request_idx,
		unsigned int numa_node)
{
	struct work_struct *work = blk_mq_rq_to_pdu(rq);

	INIT_WORK(work, rbd_queue_workfn);
	return 0;
}

static struct blk_mq_ops rbd_mq_ops = {
	.queue_rq	= rbd_queue_rq,
	.init_request	= rbd_init_request,
};

4325 4326 4327 4328
static int rbd_init_disk(struct rbd_device *rbd_dev)
{
	struct gendisk *disk;
	struct request_queue *q;
A
Alex Elder 已提交
4329
	u64 segment_size;
C
Christoph Hellwig 已提交
4330
	int err;
4331 4332

	/* create gendisk info */
4333 4334 4335
	disk = alloc_disk(single_major ?
			  (1 << RBD_SINGLE_MAJOR_PART_SHIFT) :
			  RBD_MINORS_PER_MAJOR);
4336
	if (!disk)
A
Alex Elder 已提交
4337
		return -ENOMEM;
4338

A
Alex Elder 已提交
4339
	snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
A
Alex Elder 已提交
4340
		 rbd_dev->dev_id);
4341
	disk->major = rbd_dev->major;
4342
	disk->first_minor = rbd_dev->minor;
4343 4344
	if (single_major)
		disk->flags |= GENHD_FL_EXT_DEVT;
4345 4346 4347
	disk->fops = &rbd_bd_ops;
	disk->private_data = rbd_dev;

C
Christoph Hellwig 已提交
4348 4349
	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
4350
	rbd_dev->tag_set.queue_depth = rbd_dev->opts->queue_depth;
C
Christoph Hellwig 已提交
4351
	rbd_dev->tag_set.numa_node = NUMA_NO_NODE;
I
Ilya Dryomov 已提交
4352
	rbd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
C
Christoph Hellwig 已提交
4353 4354 4355 4356 4357
	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)
4358
		goto out_disk;
4359

C
Christoph Hellwig 已提交
4360 4361 4362 4363 4364 4365
	q = blk_mq_init_queue(&rbd_dev->tag_set);
	if (IS_ERR(q)) {
		err = PTR_ERR(q);
		goto out_tag_set;
	}

4366 4367
	queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
	/* QUEUE_FLAG_ADD_RANDOM is off by default for blk-mq */
A
Alex Elder 已提交
4368

4369
	/* set io sizes to object size */
A
Alex Elder 已提交
4370 4371
	segment_size = rbd_obj_bytes(&rbd_dev->header);
	blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE);
I
Ilya Dryomov 已提交
4372
	q->limits.max_sectors = queue_max_hw_sectors(q);
I
Ilya Dryomov 已提交
4373
	blk_queue_max_segments(q, segment_size / SECTOR_SIZE);
A
Alex Elder 已提交
4374 4375 4376
	blk_queue_max_segment_size(q, segment_size);
	blk_queue_io_min(q, segment_size);
	blk_queue_io_opt(q, segment_size);
4377

4378 4379 4380 4381
	/* enable the discard support */
	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
	q->limits.discard_granularity = segment_size;
	q->limits.discard_alignment = segment_size;
4382
	blk_queue_max_discard_sectors(q, segment_size / SECTOR_SIZE);
4383
	q->limits.discard_zeroes_data = 1;
4384

4385
	if (!ceph_test_opt(rbd_dev->rbd_client->client, NOCRC))
4386
		q->backing_dev_info->capabilities |= BDI_CAP_STABLE_WRITES;
4387

4388 4389 4390 4391 4392 4393 4394
	disk->queue = q;

	q->queuedata = rbd_dev;

	rbd_dev->disk = disk;

	return 0;
C
Christoph Hellwig 已提交
4395 4396
out_tag_set:
	blk_mq_free_tag_set(&rbd_dev->tag_set);
4397 4398
out_disk:
	put_disk(disk);
C
Christoph Hellwig 已提交
4399
	return err;
4400 4401
}

4402 4403 4404 4405
/*
  sysfs
*/

A
Alex Elder 已提交
4406 4407 4408 4409 4410
static struct rbd_device *dev_to_rbd_dev(struct device *dev)
{
	return container_of(dev, struct rbd_device, dev);
}

4411 4412 4413
static ssize_t rbd_size_show(struct device *dev,
			     struct device_attribute *attr, char *buf)
{
A
Alex Elder 已提交
4414
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4415

A
Alex Elder 已提交
4416 4417
	return sprintf(buf, "%llu\n",
		(unsigned long long)rbd_dev->mapping.size);
4418 4419
}

A
Alex Elder 已提交
4420 4421 4422 4423 4424 4425 4426 4427 4428 4429
/*
 * 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
4430
			(unsigned long long)rbd_dev->mapping.features);
A
Alex Elder 已提交
4431 4432
}

4433 4434 4435
static ssize_t rbd_major_show(struct device *dev,
			      struct device_attribute *attr, char *buf)
{
A
Alex Elder 已提交
4436
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4437

A
Alex Elder 已提交
4438 4439 4440 4441
	if (rbd_dev->major)
		return sprintf(buf, "%d\n", rbd_dev->major);

	return sprintf(buf, "(none)\n");
4442 4443 4444 4445 4446 4447
}

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
4448

4449
	return sprintf(buf, "%d\n", rbd_dev->minor);
4450 4451
}

4452 4453 4454 4455 4456 4457 4458 4459 4460 4461 4462
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));
}

4463 4464
static ssize_t rbd_client_id_show(struct device *dev,
				  struct device_attribute *attr, char *buf)
4465
{
A
Alex Elder 已提交
4466
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4467

4468
	return sprintf(buf, "client%lld\n",
4469
		       ceph_client_gid(rbd_dev->rbd_client->client));
4470 4471
}

4472 4473 4474 4475 4476 4477 4478 4479
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);
}

4480 4481 4482 4483 4484 4485
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);
4486 4487
}

4488 4489
static ssize_t rbd_pool_show(struct device *dev,
			     struct device_attribute *attr, char *buf)
4490
{
A
Alex Elder 已提交
4491
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4492

4493
	return sprintf(buf, "%s\n", rbd_dev->spec->pool_name);
4494 4495
}

4496 4497 4498 4499 4500
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);

4501
	return sprintf(buf, "%llu\n",
A
Alex Elder 已提交
4502
			(unsigned long long) rbd_dev->spec->pool_id);
4503 4504
}

4505 4506 4507
static ssize_t rbd_name_show(struct device *dev,
			     struct device_attribute *attr, char *buf)
{
A
Alex Elder 已提交
4508
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4509

A
Alex Elder 已提交
4510 4511 4512 4513
	if (rbd_dev->spec->image_name)
		return sprintf(buf, "%s\n", rbd_dev->spec->image_name);

	return sprintf(buf, "(unknown)\n");
4514 4515
}

A
Alex Elder 已提交
4516 4517 4518 4519 4520
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);

4521
	return sprintf(buf, "%s\n", rbd_dev->spec->image_id);
A
Alex Elder 已提交
4522 4523
}

A
Alex Elder 已提交
4524 4525 4526 4527
/*
 * Shows the name of the currently-mapped snapshot (or
 * RBD_SNAP_HEAD_NAME for the base image).
 */
4528 4529 4530 4531
static ssize_t rbd_snap_show(struct device *dev,
			     struct device_attribute *attr,
			     char *buf)
{
A
Alex Elder 已提交
4532
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4533

4534
	return sprintf(buf, "%s\n", rbd_dev->spec->snap_name);
4535 4536
}

4537 4538 4539 4540 4541 4542 4543 4544
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);
}

4545
/*
4546 4547 4548
 * 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)".
4549 4550
 */
static ssize_t rbd_parent_show(struct device *dev,
4551 4552
			       struct device_attribute *attr,
			       char *buf)
4553 4554
{
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4555
	ssize_t count = 0;
4556

4557
	if (!rbd_dev->parent)
4558 4559
		return sprintf(buf, "(no parent image)\n");

4560 4561 4562 4563 4564 4565 4566 4567 4568 4569 4570 4571 4572 4573 4574 4575
	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;
4576 4577
}

4578 4579 4580 4581 4582
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
4583
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4584
	int ret;
4585

A
Alex Elder 已提交
4586
	ret = rbd_dev_refresh(rbd_dev);
4587
	if (ret)
4588
		return ret;
4589

4590
	return size;
4591
}
4592

4593
static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
A
Alex Elder 已提交
4594
static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL);
4595
static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
4596
static DEVICE_ATTR(minor, S_IRUGO, rbd_minor_show, NULL);
4597
static DEVICE_ATTR(client_addr, S_IRUGO, rbd_client_addr_show, NULL);
4598
static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
4599
static DEVICE_ATTR(cluster_fsid, S_IRUGO, rbd_cluster_fsid_show, NULL);
4600
static DEVICE_ATTR(config_info, S_IRUSR, rbd_config_info_show, NULL);
4601
static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
4602
static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL);
4603
static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
A
Alex Elder 已提交
4604
static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL);
4605 4606
static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
4607
static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL);
4608
static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL);
4609 4610 4611

static struct attribute *rbd_attrs[] = {
	&dev_attr_size.attr,
A
Alex Elder 已提交
4612
	&dev_attr_features.attr,
4613
	&dev_attr_major.attr,
4614
	&dev_attr_minor.attr,
4615
	&dev_attr_client_addr.attr,
4616
	&dev_attr_client_id.attr,
4617
	&dev_attr_cluster_fsid.attr,
4618
	&dev_attr_config_info.attr,
4619
	&dev_attr_pool.attr,
4620
	&dev_attr_pool_id.attr,
4621
	&dev_attr_name.attr,
A
Alex Elder 已提交
4622
	&dev_attr_image_id.attr,
4623
	&dev_attr_current_snap.attr,
4624
	&dev_attr_snap_id.attr,
4625
	&dev_attr_parent.attr,
4626 4627 4628 4629 4630 4631 4632 4633 4634 4635 4636 4637 4638
	&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
};

4639
static void rbd_dev_release(struct device *dev);
4640

4641
static const struct device_type rbd_device_type = {
4642 4643
	.name		= "rbd",
	.groups		= rbd_attr_groups,
4644
	.release	= rbd_dev_release,
4645 4646
};

4647 4648 4649 4650 4651 4652 4653 4654 4655 4656 4657 4658 4659 4660 4661 4662 4663 4664 4665 4666 4667
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;
4668 4669 4670

	spec->pool_id = CEPH_NOPOOL;
	spec->snap_id = CEPH_NOSNAP;
4671 4672 4673 4674 4675 4676 4677 4678 4679 4680 4681 4682 4683 4684 4685 4686
	kref_init(&spec->kref);

	return spec;
}

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

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

4687
static void rbd_dev_free(struct rbd_device *rbd_dev)
4688
{
4689
	WARN_ON(rbd_dev->watch_state != RBD_WATCH_STATE_UNREGISTERED);
4690
	WARN_ON(rbd_dev->lock_state != RBD_LOCK_STATE_UNLOCKED);
4691

4692
	ceph_oid_destroy(&rbd_dev->header_oid);
4693
	ceph_oloc_destroy(&rbd_dev->header_oloc);
4694
	kfree(rbd_dev->config_info);
4695

4696 4697 4698 4699
	rbd_put_client(rbd_dev->rbd_client);
	rbd_spec_put(rbd_dev->spec);
	kfree(rbd_dev->opts);
	kfree(rbd_dev);
4700 4701 4702 4703 4704 4705 4706 4707 4708 4709 4710 4711 4712
}

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);
4713 4714 4715 4716 4717 4718 4719 4720 4721 4722

	/*
	 * 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);
}

4723 4724
static struct rbd_device *__rbd_dev_create(struct rbd_client *rbdc,
					   struct rbd_spec *spec)
4725 4726 4727
{
	struct rbd_device *rbd_dev;

4728
	rbd_dev = kzalloc(sizeof(*rbd_dev), GFP_KERNEL);
4729 4730 4731 4732 4733 4734 4735
	if (!rbd_dev)
		return NULL;

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

4736
	rbd_dev->header.data_pool_id = CEPH_NOPOOL;
4737
	ceph_oid_init(&rbd_dev->header_oid);
4738
	rbd_dev->header_oloc.pool = spec->pool_id;
4739

4740 4741 4742 4743
	mutex_init(&rbd_dev->watch_mutex);
	rbd_dev->watch_state = RBD_WATCH_STATE_UNREGISTERED;
	INIT_DELAYED_WORK(&rbd_dev->watch_dwork, rbd_reregister_watch);

4744 4745 4746 4747 4748 4749 4750 4751
	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);

4752 4753 4754 4755 4756
	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);

4757
	rbd_dev->rbd_client = rbdc;
4758
	rbd_dev->spec = spec;
4759

4760 4761 4762 4763 4764 4765 4766 4767 4768 4769 4770 4771 4772 4773 4774 4775 4776 4777 4778 4779 4780 4781 4782 4783 4784 4785 4786 4787 4788 4789
	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;
4790

4791 4792
	/* we have a ref from do_rbd_add() */
	__module_get(THIS_MODULE);
4793

4794
	dout("%s rbd_dev %p dev_id %d\n", __func__, rbd_dev, rbd_dev->dev_id);
4795
	return rbd_dev;
4796 4797 4798 4799 4800 4801

fail_dev_id:
	ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id);
fail_rbd_dev:
	rbd_dev_free(rbd_dev);
	return NULL;
4802 4803 4804 4805
}

static void rbd_dev_destroy(struct rbd_device *rbd_dev)
{
4806 4807
	if (rbd_dev)
		put_device(&rbd_dev->dev);
4808 4809
}

4810 4811 4812 4813 4814 4815 4816 4817 4818 4819 4820 4821 4822 4823 4824
/*
 * 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 };

4825 4826 4827 4828
	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));
4829
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4830 4831
	if (ret < 0)
		return ret;
4832 4833
	if (ret < sizeof (size_buf))
		return -ERANGE;
4834

J
Josh Durgin 已提交
4835
	if (order) {
4836
		*order = size_buf.order;
J
Josh Durgin 已提交
4837 4838
		dout("  order %u", (unsigned int)*order);
	}
4839 4840
	*snap_size = le64_to_cpu(size_buf.size);

J
Josh Durgin 已提交
4841 4842
	dout("  snap_id 0x%016llx snap_size = %llu\n",
		(unsigned long long)snap_id,
4843
		(unsigned long long)*snap_size);
4844 4845 4846 4847 4848 4849 4850 4851 4852 4853 4854

	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);
}

4855 4856 4857 4858 4859 4860 4861 4862 4863 4864
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;

4865 4866 4867
	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);
4868
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4869 4870 4871 4872 4873
	if (ret < 0)
		goto out;

	p = reply_buf;
	rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p,
4874 4875
						p + ret, NULL, GFP_NOIO);
	ret = 0;
4876 4877 4878 4879 4880 4881 4882 4883 4884 4885 4886 4887 4888

	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;
}

4889 4890 4891 4892 4893 4894 4895
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;
4896
	} __attribute__ ((packed)) features_buf = { 0 };
4897
	u64 unsup;
4898 4899
	int ret;

4900 4901 4902 4903
	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));
4904
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4905 4906
	if (ret < 0)
		return ret;
4907 4908
	if (ret < sizeof (features_buf))
		return -ERANGE;
A
Alex Elder 已提交
4909

4910 4911 4912 4913
	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
4914
		return -ENXIO;
4915
	}
A
Alex Elder 已提交
4916

4917 4918 4919
	*snap_features = le64_to_cpu(features_buf.features);

	dout("  snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
4920 4921 4922
		(unsigned long long)snap_id,
		(unsigned long long)*snap_features,
		(unsigned long long)le64_to_cpu(features_buf.incompat));
4923 4924 4925 4926 4927 4928 4929 4930 4931 4932

	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);
}

4933 4934 4935 4936 4937 4938 4939 4940
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
4941
	u64 pool_id;
4942
	char *image_id;
4943
	u64 snap_id;
4944 4945 4946 4947 4948 4949 4950 4951 4952 4953 4954 4955 4956 4957 4958 4959 4960
	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;
	}

4961
	snapid = cpu_to_le64(rbd_dev->spec->snap_id);
4962 4963 4964
	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
				  &rbd_dev->header_oloc, "get_parent",
				  &snapid, sizeof(snapid), reply_buf, size);
4965
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4966 4967 4968 4969
	if (ret < 0)
		goto out_err;

	p = reply_buf;
4970 4971
	end = reply_buf + ret;
	ret = -ERANGE;
A
Alex Elder 已提交
4972
	ceph_decode_64_safe(&p, end, pool_id, out_err);
4973 4974 4975 4976 4977 4978 4979 4980 4981 4982 4983 4984 4985 4986 4987 4988 4989
	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);
		}

4990
		goto out;	/* No parent?  No problem. */
4991
	}
4992

4993 4994 4995
	/* The ceph file layout needs to fit pool id in 32 bits */

	ret = -EIO;
A
Alex Elder 已提交
4996
	if (pool_id > (u64)U32_MAX) {
4997
		rbd_warn(NULL, "parent pool id too large (%llu > %u)",
A
Alex Elder 已提交
4998
			(unsigned long long)pool_id, U32_MAX);
4999
		goto out_err;
A
Alex Elder 已提交
5000
	}
5001

A
Alex Elder 已提交
5002
	image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
5003 5004 5005 5006
	if (IS_ERR(image_id)) {
		ret = PTR_ERR(image_id);
		goto out_err;
	}
5007
	ceph_decode_64_safe(&p, end, snap_id, out_err);
5008 5009
	ceph_decode_64_safe(&p, end, overlap, out_err);

5010 5011 5012 5013 5014 5015 5016 5017 5018
	/*
	 * The parent won't change (except when the clone is
	 * flattened, already handled that).  So we only need to
	 * record the parent spec we have not already done so.
	 */
	if (!rbd_dev->parent_spec) {
		parent_spec->pool_id = pool_id;
		parent_spec->image_id = image_id;
		parent_spec->snap_id = snap_id;
A
Alex Elder 已提交
5019 5020
		rbd_dev->parent_spec = parent_spec;
		parent_spec = NULL;	/* rbd_dev now owns this */
5021 5022
	} else {
		kfree(image_id);
5023 5024 5025
	}

	/*
5026 5027
	 * We always update the parent overlap.  If it's zero we issue
	 * a warning, as we will proceed as if there was no parent.
5028 5029 5030
	 */
	if (!overlap) {
		if (parent_spec) {
5031 5032 5033 5034
			/* refresh, careful to warn just once */
			if (rbd_dev->parent_overlap)
				rbd_warn(rbd_dev,
				    "clone now standalone (overlap became 0)");
5035
		} else {
5036 5037
			/* initial probe */
			rbd_warn(rbd_dev, "clone is standalone (overlap 0)");
5038
		}
A
Alex Elder 已提交
5039
	}
5040 5041
	rbd_dev->parent_overlap = overlap;

5042 5043 5044 5045 5046 5047 5048 5049 5050
out:
	ret = 0;
out_err:
	kfree(reply_buf);
	rbd_spec_put(parent_spec);

	return ret;
}

5051 5052 5053 5054 5055 5056 5057 5058 5059 5060 5061 5062 5063
static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev)
{
	struct {
		__le64 stripe_unit;
		__le64 stripe_count;
	} __attribute__ ((packed)) striping_info_buf = { 0 };
	size_t size = sizeof (striping_info_buf);
	void *p;
	u64 obj_size;
	u64 stripe_unit;
	u64 stripe_count;
	int ret;

5064 5065 5066
	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);
5067 5068 5069 5070 5071 5072 5073 5074 5075 5076 5077 5078 5079
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
	if (ret < 0)
		return ret;
	if (ret < size)
		return -ERANGE;

	/*
	 * We don't actually support the "fancy striping" feature
	 * (STRIPINGV2) yet, but if the striping sizes are the
	 * defaults the behavior is the same as before.  So find
	 * out, and only fail if the image has non-default values.
	 */
	ret = -EINVAL;
I
Ilya Dryomov 已提交
5080
	obj_size = rbd_obj_bytes(&rbd_dev->header);
5081 5082 5083 5084 5085 5086 5087 5088 5089 5090 5091 5092 5093 5094
	p = &striping_info_buf;
	stripe_unit = ceph_decode_64(&p);
	if (stripe_unit != obj_size) {
		rbd_warn(rbd_dev, "unsupported stripe unit "
				"(got %llu want %llu)",
				stripe_unit, obj_size);
		return -EINVAL;
	}
	stripe_count = ceph_decode_64(&p);
	if (stripe_count != 1) {
		rbd_warn(rbd_dev, "unsupported stripe count "
				"(got %llu want 1)", stripe_count);
		return -EINVAL;
	}
5095 5096
	rbd_dev->header.stripe_unit = stripe_unit;
	rbd_dev->header.stripe_count = stripe_count;
5097 5098 5099 5100

	return 0;
}

5101 5102 5103 5104 5105 5106 5107 5108 5109 5110 5111 5112 5113 5114 5115 5116 5117 5118
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;
}

5119 5120
static char *rbd_dev_image_name(struct rbd_device *rbd_dev)
{
5121
	CEPH_DEFINE_OID_ONSTACK(oid);
5122 5123 5124 5125 5126 5127 5128 5129 5130 5131 5132 5133
	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
5134 5135
	len = strlen(rbd_dev->spec->image_id);
	image_id_size = sizeof (__le32) + len;
5136 5137 5138 5139 5140
	image_id = kmalloc(image_id_size, GFP_KERNEL);
	if (!image_id)
		return NULL;

	p = image_id;
5141
	end = image_id + image_id_size;
5142
	ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len);
5143 5144 5145 5146 5147 5148

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

5149 5150 5151 5152
	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);
5153 5154 5155
	if (ret < 0)
		goto out;
	p = reply_buf;
5156 5157
	end = reply_buf + ret;

5158 5159 5160 5161 5162 5163 5164 5165 5166 5167 5168 5169
	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;
}

5170 5171 5172 5173 5174 5175 5176 5177 5178 5179 5180 5181 5182 5183 5184 5185 5186 5187 5188 5189 5190 5191 5192 5193 5194 5195 5196 5197 5198 5199
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);
5200 5201 5202 5203 5204 5205 5206
		if (IS_ERR(snap_name)) {
			/* ignore no-longer existing snapshots */
			if (PTR_ERR(snap_name) == -ENOENT)
				continue;
			else
				break;
		}
5207 5208 5209 5210 5211 5212 5213 5214 5215 5216 5217 5218 5219 5220 5221 5222 5223 5224
		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);
}

5225
/*
5226 5227 5228 5229 5230 5231 5232 5233 5234 5235 5236 5237 5238 5239 5240 5241 5242 5243 5244 5245 5246 5247 5248 5249 5250 5251 5252
 * 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.
5253
 *
5254 5255
 * All names in an rbd spec are dynamically allocated.  It's OK if we
 * can't figure out the name for an image id.
5256
 */
5257
static int rbd_spec_fill_names(struct rbd_device *rbd_dev)
5258
{
5259 5260 5261 5262 5263
	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;
5264 5265
	int ret;

5266 5267 5268
	rbd_assert(spec->pool_id != CEPH_NOPOOL);
	rbd_assert(spec->image_id);
	rbd_assert(spec->snap_id != CEPH_NOSNAP);
5269

5270
	/* Get the pool name; we have to make our own copy of this */
5271

5272 5273 5274
	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);
5275 5276
		return -EIO;
	}
5277 5278
	pool_name = kstrdup(pool_name, GFP_KERNEL);
	if (!pool_name)
5279 5280 5281 5282
		return -ENOMEM;

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

5283 5284
	image_name = rbd_dev_image_name(rbd_dev);
	if (!image_name)
A
Alex Elder 已提交
5285
		rbd_warn(rbd_dev, "unable to get image name");
5286

5287
	/* Fetch the snapshot name */
5288

5289
	snap_name = rbd_snap_name(rbd_dev, spec->snap_id);
5290 5291
	if (IS_ERR(snap_name)) {
		ret = PTR_ERR(snap_name);
5292
		goto out_err;
5293 5294 5295 5296 5297
	}

	spec->pool_name = pool_name;
	spec->image_name = image_name;
	spec->snap_name = snap_name;
5298 5299

	return 0;
5300

5301
out_err:
5302 5303
	kfree(image_name);
	kfree(pool_name);
5304 5305 5306
	return ret;
}

A
Alex Elder 已提交
5307
static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev)
5308 5309 5310 5311 5312 5313 5314 5315 5316 5317 5318 5319 5320 5321 5322 5323 5324 5325 5326 5327 5328 5329 5330
{
	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;

5331 5332 5333
	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
				  &rbd_dev->header_oloc, "get_snapcontext",
				  NULL, 0, reply_buf, size);
5334
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
5335 5336 5337 5338
	if (ret < 0)
		goto out;

	p = reply_buf;
5339 5340
	end = reply_buf + ret;
	ret = -ERANGE;
5341 5342 5343 5344 5345 5346 5347 5348 5349 5350 5351 5352 5353 5354 5355 5356
	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;
5357
	ret = 0;
5358

5359
	snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
5360 5361 5362 5363 5364 5365 5366 5367
	if (!snapc) {
		ret = -ENOMEM;
		goto out;
	}
	snapc->seq = seq;
	for (i = 0; i < snap_count; i++)
		snapc->snaps[i] = ceph_decode_64(&p);

5368
	ceph_put_snap_context(rbd_dev->header.snapc);
5369 5370 5371
	rbd_dev->header.snapc = snapc;

	dout("  snap context seq = %llu, snap_count = %u\n",
5372
		(unsigned long long)seq, (unsigned int)snap_count);
5373 5374 5375
out:
	kfree(reply_buf);

5376
	return ret;
5377 5378
}

5379 5380
static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
					u64 snap_id)
A
Alex Elder 已提交
5381 5382 5383
{
	size_t size;
	void *reply_buf;
5384
	__le64 snapid;
A
Alex Elder 已提交
5385 5386 5387 5388 5389 5390 5391 5392 5393 5394
	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);

5395
	snapid = cpu_to_le64(snap_id);
5396 5397 5398
	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
				  &rbd_dev->header_oloc, "get_snapshot_name",
				  &snapid, sizeof(snapid), reply_buf, size);
5399
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
5400 5401
	if (ret < 0) {
		snap_name = ERR_PTR(ret);
A
Alex Elder 已提交
5402
		goto out;
5403
	}
A
Alex Elder 已提交
5404 5405

	p = reply_buf;
5406
	end = reply_buf + ret;
A
Alex Elder 已提交
5407
	snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
5408
	if (IS_ERR(snap_name))
A
Alex Elder 已提交
5409 5410
		goto out;

5411
	dout("  snap_id 0x%016llx snap_name = %s\n",
5412
		(unsigned long long)snap_id, snap_name);
A
Alex Elder 已提交
5413 5414 5415
out:
	kfree(reply_buf);

5416
	return snap_name;
A
Alex Elder 已提交
5417 5418
}

5419
static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev)
A
Alex Elder 已提交
5420
{
5421
	bool first_time = rbd_dev->header.object_prefix == NULL;
A
Alex Elder 已提交
5422 5423
	int ret;

5424 5425
	ret = rbd_dev_v2_image_size(rbd_dev);
	if (ret)
5426
		return ret;
5427

5428 5429 5430
	if (first_time) {
		ret = rbd_dev_v2_header_onetime(rbd_dev);
		if (ret)
5431
			return ret;
5432 5433
	}

A
Alex Elder 已提交
5434
	ret = rbd_dev_v2_snap_context(rbd_dev);
5435 5436 5437 5438
	if (ret && first_time) {
		kfree(rbd_dev->header.object_prefix);
		rbd_dev->header.object_prefix = NULL;
	}
A
Alex Elder 已提交
5439 5440 5441 5442

	return ret;
}

5443 5444 5445 5446 5447 5448 5449 5450 5451 5452
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);
}

5453 5454 5455
/*
 * 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
5456 5457
 * the token (string of non-white space characters) found.  Note
 * that *buf must be terminated with '\0'.
5458 5459 5460 5461 5462 5463 5464 5465 5466 5467 5468 5469 5470 5471
 */
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
5472 5473 5474 5475 5476 5477 5478 5479 5480 5481 5482 5483 5484 5485 5486 5487 5488 5489 5490 5491 5492 5493
/*
 * 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
5494
	dup = kmemdup(*buf, len + 1, GFP_KERNEL);
A
Alex Elder 已提交
5495 5496 5497 5498 5499 5500 5501 5502 5503 5504 5505
	if (!dup)
		return NULL;
	*(dup + len) = '\0';
	*buf += len;

	if (lenp)
		*lenp = len;

	return dup;
}

5506
/*
5507 5508 5509 5510
 * 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
5511
 *
5512 5513 5514 5515 5516 5517 5518 5519 5520 5521 5522 5523 5524 5525 5526 5527 5528 5529 5530 5531 5532 5533 5534 5535 5536 5537 5538 5539 5540 5541 5542 5543 5544 5545
 * 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.
5546
 */
5547
static int rbd_add_parse_args(const char *buf,
5548
				struct ceph_options **ceph_opts,
5549 5550
				struct rbd_options **opts,
				struct rbd_spec **rbd_spec)
5551
{
A
Alex Elder 已提交
5552
	size_t len;
5553
	char *options;
5554
	const char *mon_addrs;
5555
	char *snap_name;
5556
	size_t mon_addrs_size;
5557
	struct rbd_spec *spec = NULL;
5558
	struct rbd_options *rbd_opts = NULL;
5559
	struct ceph_options *copts;
5560
	int ret;
5561 5562 5563

	/* The first four tokens are required */

5564
	len = next_token(&buf);
5565 5566 5567 5568
	if (!len) {
		rbd_warn(NULL, "no monitor address(es) provided");
		return -EINVAL;
	}
5569
	mon_addrs = buf;
5570
	mon_addrs_size = len + 1;
5571
	buf += len;
5572

5573
	ret = -EINVAL;
5574 5575
	options = dup_token(&buf, NULL);
	if (!options)
5576
		return -ENOMEM;
5577 5578 5579 5580
	if (!*options) {
		rbd_warn(NULL, "no options provided");
		goto out_err;
	}
5581

5582 5583
	spec = rbd_spec_alloc();
	if (!spec)
5584
		goto out_mem;
5585 5586 5587 5588

	spec->pool_name = dup_token(&buf, NULL);
	if (!spec->pool_name)
		goto out_mem;
5589 5590 5591 5592
	if (!*spec->pool_name) {
		rbd_warn(NULL, "no pool name provided");
		goto out_err;
	}
5593

A
Alex Elder 已提交
5594
	spec->image_name = dup_token(&buf, NULL);
5595
	if (!spec->image_name)
5596
		goto out_mem;
5597 5598 5599 5600
	if (!*spec->image_name) {
		rbd_warn(NULL, "no image name provided");
		goto out_err;
	}
5601

5602 5603 5604 5605
	/*
	 * Snapshot name is optional; default is to use "-"
	 * (indicating the head/no snapshot).
	 */
5606
	len = next_token(&buf);
5607
	if (!len) {
5608 5609
		buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */
		len = sizeof (RBD_SNAP_HEAD_NAME) - 1;
5610
	} else if (len > RBD_MAX_SNAP_NAME_LEN) {
5611
		ret = -ENAMETOOLONG;
5612
		goto out_err;
5613
	}
5614 5615
	snap_name = kmemdup(buf, len + 1, GFP_KERNEL);
	if (!snap_name)
5616
		goto out_mem;
5617 5618
	*(snap_name + len) = '\0';
	spec->snap_name = snap_name;
A
Alex Elder 已提交
5619

5620
	/* Initialize all rbd options to the defaults */
5621

5622 5623 5624 5625 5626
	rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL);
	if (!rbd_opts)
		goto out_mem;

	rbd_opts->read_only = RBD_READ_ONLY_DEFAULT;
I
Ilya Dryomov 已提交
5627
	rbd_opts->queue_depth = RBD_QUEUE_DEPTH_DEFAULT;
I
Ilya Dryomov 已提交
5628
	rbd_opts->lock_on_read = RBD_LOCK_ON_READ_DEFAULT;
A
Alex Elder 已提交
5629

5630
	copts = ceph_parse_options(options, mon_addrs,
5631
					mon_addrs + mon_addrs_size - 1,
5632
					parse_rbd_opts_token, rbd_opts);
5633 5634
	if (IS_ERR(copts)) {
		ret = PTR_ERR(copts);
5635 5636
		goto out_err;
	}
5637 5638 5639
	kfree(options);

	*ceph_opts = copts;
5640
	*opts = rbd_opts;
5641
	*rbd_spec = spec;
5642

5643
	return 0;
5644
out_mem:
5645
	ret = -ENOMEM;
A
Alex Elder 已提交
5646
out_err:
5647 5648
	kfree(rbd_opts);
	rbd_spec_put(spec);
5649
	kfree(options);
A
Alex Elder 已提交
5650

5651
	return ret;
5652 5653
}

5654 5655 5656 5657 5658
/*
 * Return pool id (>= 0) or a negative error code.
 */
static int rbd_add_get_pool_id(struct rbd_client *rbdc, const char *pool_name)
{
5659
	struct ceph_options *opts = rbdc->client->options;
5660 5661 5662 5663 5664 5665 5666
	u64 newest_epoch;
	int tries = 0;
	int ret;

again:
	ret = ceph_pg_poolid_by_name(rbdc->client->osdc.osdmap, pool_name);
	if (ret == -ENOENT && tries++ < 1) {
5667 5668
		ret = ceph_monc_get_version(&rbdc->client->monc, "osdmap",
					    &newest_epoch);
5669 5670 5671 5672
		if (ret < 0)
			return ret;

		if (rbdc->client->osdc.osdmap->epoch < newest_epoch) {
5673
			ceph_osdc_maybe_request_map(&rbdc->client->osdc);
5674
			(void) ceph_monc_wait_osdmap(&rbdc->client->monc,
5675 5676
						     newest_epoch,
						     opts->mount_timeout);
5677 5678 5679 5680 5681 5682 5683 5684 5685 5686
			goto again;
		} else {
			/* the osdmap we have is new enough */
			return -ENOENT;
		}
	}

	return ret;
}

A
Alex Elder 已提交
5687 5688 5689 5690 5691 5692 5693 5694 5695 5696 5697 5698 5699 5700 5701 5702 5703 5704
/*
 * 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;
5705
	CEPH_DEFINE_OID_ONSTACK(oid);
A
Alex Elder 已提交
5706
	void *response;
5707
	char *image_id;
5708

A
Alex Elder 已提交
5709 5710 5711
	/*
	 * When probing a parent image, the image id is already
	 * known (and the image name likely is not).  There's no
5712 5713
	 * 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
5714
	 */
5715 5716 5717
	if (rbd_dev->spec->image_id) {
		rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1;

A
Alex Elder 已提交
5718
		return 0;
5719
	}
A
Alex Elder 已提交
5720

A
Alex Elder 已提交
5721 5722 5723 5724
	/*
	 * First, see if the format 2 image id file exists, and if
	 * so, get the image's persistent id from it.
	 */
5725 5726 5727 5728 5729 5730
	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
5731 5732 5733 5734 5735 5736 5737 5738 5739 5740

	/* 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;
	}

5741 5742
	/* If it doesn't exist we'll assume it's a format 1 image */

5743 5744 5745
	ret = rbd_obj_method_sync(rbd_dev, &oid, &rbd_dev->header_oloc,
				  "get_id", NULL, 0,
				  response, RBD_IMAGE_ID_LEN_MAX);
5746
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
5747 5748 5749 5750 5751
	if (ret == -ENOENT) {
		image_id = kstrdup("", GFP_KERNEL);
		ret = image_id ? 0 : -ENOMEM;
		if (!ret)
			rbd_dev->image_format = 1;
5752
	} else if (ret >= 0) {
5753 5754 5755
		void *p = response;

		image_id = ceph_extract_encoded_string(&p, p + ret,
A
Alex Elder 已提交
5756
						NULL, GFP_NOIO);
5757
		ret = PTR_ERR_OR_ZERO(image_id);
5758 5759 5760 5761 5762 5763 5764
		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
5765 5766 5767
	}
out:
	kfree(response);
5768
	ceph_oid_destroy(&oid);
A
Alex Elder 已提交
5769 5770 5771
	return ret;
}

A
Alex Elder 已提交
5772 5773 5774 5775
/*
 * Undo whatever state changes are made by v1 or v2 header info
 * call.
 */
A
Alex Elder 已提交
5776 5777 5778 5779
static void rbd_dev_unprobe(struct rbd_device *rbd_dev)
{
	struct rbd_image_header	*header;

5780
	rbd_dev_parent_put(rbd_dev);
A
Alex Elder 已提交
5781 5782 5783 5784

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

	header = &rbd_dev->header;
5785
	ceph_put_snap_context(header->snapc);
A
Alex Elder 已提交
5786 5787 5788 5789 5790 5791
	kfree(header->snap_sizes);
	kfree(header->snap_names);
	kfree(header->object_prefix);
	memset(header, 0, sizeof (*header));
}

5792
static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev)
5793 5794 5795
{
	int ret;

5796
	ret = rbd_dev_v2_object_prefix(rbd_dev);
5797
	if (ret)
5798 5799
		goto out_err;

5800 5801 5802 5803
	/*
	 * Get the and check features for the image.  Currently the
	 * features are assumed to never change.
	 */
5804
	ret = rbd_dev_v2_features(rbd_dev);
5805
	if (ret)
5806
		goto out_err;
5807

5808 5809 5810 5811 5812 5813 5814
	/* 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;
	}
5815

5816 5817 5818 5819 5820 5821
	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
5822
	rbd_init_layout(rbd_dev);
A
Alex Elder 已提交
5823
	return 0;
I
Ilya Dryomov 已提交
5824

5825
out_err:
A
Alex Elder 已提交
5826
	rbd_dev->header.features = 0;
5827 5828
	kfree(rbd_dev->header.object_prefix);
	rbd_dev->header.object_prefix = NULL;
5829
	return ret;
5830 5831
}

5832 5833 5834 5835 5836 5837
/*
 * @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
5838
{
5839
	struct rbd_device *parent = NULL;
5840 5841 5842 5843 5844
	int ret;

	if (!rbd_dev->parent_spec)
		return 0;

5845 5846 5847 5848 5849 5850
	if (++depth > RBD_MAX_PARENT_CHAIN_LEN) {
		pr_info("parent chain is too long (%d)\n", depth);
		ret = -EINVAL;
		goto out_err;
	}

5851
	parent = __rbd_dev_create(rbd_dev->rbd_client, rbd_dev->parent_spec);
5852 5853
	if (!parent) {
		ret = -ENOMEM;
5854
		goto out_err;
5855 5856 5857 5858 5859 5860 5861 5862
	}

	/*
	 * 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);
5863

5864
	ret = rbd_dev_image_probe(parent, depth);
5865 5866
	if (ret < 0)
		goto out_err;
5867

5868
	rbd_dev->parent = parent;
5869
	atomic_set(&rbd_dev->parent_ref, 1);
5870
	return 0;
5871

5872
out_err:
5873
	rbd_dev_unparent(rbd_dev);
5874
	rbd_dev_destroy(parent);
5875 5876 5877
	return ret;
}

I
Ilya Dryomov 已提交
5878 5879 5880 5881
/*
 * rbd_dev->header_rwsem must be locked for write and will be unlocked
 * upon return.
 */
5882
static int rbd_dev_device_setup(struct rbd_device *rbd_dev)
5883
{
A
Alex Elder 已提交
5884
	int ret;
A
Alex Elder 已提交
5885

5886
	/* Record our major and minor device numbers. */
A
Alex Elder 已提交
5887

5888 5889 5890
	if (!single_major) {
		ret = register_blkdev(0, rbd_dev->name);
		if (ret < 0)
5891
			goto err_out_unlock;
5892 5893 5894 5895 5896 5897 5898

		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
5899 5900 5901 5902 5903 5904 5905

	/* Set up the blkdev mapping. */

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

5906
	ret = rbd_dev_mapping_set(rbd_dev);
A
Alex Elder 已提交
5907 5908
	if (ret)
		goto err_out_disk;
I
Ilya Dryomov 已提交
5909

5910
	set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE);
5911
	set_disk_ro(rbd_dev->disk, rbd_dev->mapping.read_only);
5912

5913 5914
	dev_set_name(&rbd_dev->dev, "%d", rbd_dev->dev_id);
	ret = device_add(&rbd_dev->dev);
5915
	if (ret)
5916
		goto err_out_mapping;
A
Alex Elder 已提交
5917 5918 5919

	/* Everything's ready.  Announce the disk to the world. */

5920
	set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
I
Ilya Dryomov 已提交
5921
	up_write(&rbd_dev->header_rwsem);
A
Alex Elder 已提交
5922

5923 5924 5925 5926
	spin_lock(&rbd_dev_list_lock);
	list_add_tail(&rbd_dev->node, &rbd_dev_list);
	spin_unlock(&rbd_dev_list_lock);

I
Ilya Dryomov 已提交
5927
	add_disk(rbd_dev->disk);
5928 5929 5930
	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);
A
Alex Elder 已提交
5931 5932

	return ret;
5933

5934 5935
err_out_mapping:
	rbd_dev_mapping_clear(rbd_dev);
A
Alex Elder 已提交
5936 5937 5938
err_out_disk:
	rbd_free_disk(rbd_dev);
err_out_blkdev:
5939 5940
	if (!single_major)
		unregister_blkdev(rbd_dev->major, rbd_dev->name);
I
Ilya Dryomov 已提交
5941 5942
err_out_unlock:
	up_write(&rbd_dev->header_rwsem);
A
Alex Elder 已提交
5943 5944 5945
	return ret;
}

A
Alex Elder 已提交
5946 5947 5948
static int rbd_dev_header_name(struct rbd_device *rbd_dev)
{
	struct rbd_spec *spec = rbd_dev->spec;
5949
	int ret;
A
Alex Elder 已提交
5950 5951 5952 5953 5954

	/* 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)
5955 5956
		ret = ceph_oid_aprintf(&rbd_dev->header_oid, GFP_KERNEL, "%s%s",
				       spec->image_name, RBD_SUFFIX);
A
Alex Elder 已提交
5957
	else
5958 5959
		ret = ceph_oid_aprintf(&rbd_dev->header_oid, GFP_KERNEL, "%s%s",
				       RBD_HEADER_PREFIX, spec->image_id);
A
Alex Elder 已提交
5960

5961
	return ret;
A
Alex Elder 已提交
5962 5963
}

5964 5965
static void rbd_dev_image_release(struct rbd_device *rbd_dev)
{
A
Alex Elder 已提交
5966
	rbd_dev_unprobe(rbd_dev);
5967 5968
	if (rbd_dev->opts)
		rbd_unregister_watch(rbd_dev);
A
Alex Elder 已提交
5969 5970 5971
	rbd_dev->image_format = 0;
	kfree(rbd_dev->spec->image_id);
	rbd_dev->spec->image_id = NULL;
5972 5973
}

5974 5975
/*
 * Probe for the existence of the header object for the given rbd
5976 5977 5978
 * 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.
5979
 */
5980
static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth)
5981 5982 5983 5984
{
	int ret;

	/*
A
Alex Elder 已提交
5985 5986 5987 5988
	 * 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.
5989 5990 5991
	 */
	ret = rbd_dev_image_id(rbd_dev);
	if (ret)
5992 5993
		return ret;

A
Alex Elder 已提交
5994 5995 5996 5997
	ret = rbd_dev_header_name(rbd_dev);
	if (ret)
		goto err_out_format;

5998
	if (!depth) {
5999
		ret = rbd_register_watch(rbd_dev);
6000 6001 6002 6003 6004
		if (ret) {
			if (ret == -ENOENT)
				pr_info("image %s/%s does not exist\n",
					rbd_dev->spec->pool_name,
					rbd_dev->spec->image_name);
6005
			goto err_out_format;
6006
		}
6007
	}
6008

6009
	ret = rbd_dev_header_info(rbd_dev);
6010
	if (ret)
6011
		goto err_out_watch;
A
Alex Elder 已提交
6012

6013 6014 6015 6016 6017 6018
	/*
	 * 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.
	 */
6019
	if (!depth)
6020 6021 6022
		ret = rbd_spec_fill_snap_id(rbd_dev);
	else
		ret = rbd_spec_fill_names(rbd_dev);
6023 6024 6025 6026 6027 6028
	if (ret) {
		if (ret == -ENOENT)
			pr_info("snap %s/%s@%s does not exist\n",
				rbd_dev->spec->pool_name,
				rbd_dev->spec->image_name,
				rbd_dev->spec->snap_name);
A
Alex Elder 已提交
6029
		goto err_out_probe;
6030
	}
6031

6032 6033 6034 6035 6036 6037 6038 6039 6040
	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.
		 */
6041
		if (!depth && rbd_dev->parent_spec)
6042 6043 6044 6045
			rbd_warn(rbd_dev,
				 "WARNING: kernel layering is EXPERIMENTAL!");
	}

6046
	ret = rbd_dev_probe_parent(rbd_dev, depth);
A
Alex Elder 已提交
6047 6048 6049 6050
	if (ret)
		goto err_out_probe;

	dout("discovered format %u image, header name is %s\n",
6051
		rbd_dev->image_format, rbd_dev->header_oid.name);
A
Alex Elder 已提交
6052
	return 0;
6053

A
Alex Elder 已提交
6054 6055
err_out_probe:
	rbd_dev_unprobe(rbd_dev);
6056
err_out_watch:
6057
	if (!depth)
6058
		rbd_unregister_watch(rbd_dev);
A
Alex Elder 已提交
6059 6060
err_out_format:
	rbd_dev->image_format = 0;
6061 6062
	kfree(rbd_dev->spec->image_id);
	rbd_dev->spec->image_id = NULL;
6063 6064 6065
	return ret;
}

6066 6067 6068
static ssize_t do_rbd_add(struct bus_type *bus,
			  const char *buf,
			  size_t count)
6069
{
6070
	struct rbd_device *rbd_dev = NULL;
6071
	struct ceph_options *ceph_opts = NULL;
6072
	struct rbd_options *rbd_opts = NULL;
6073
	struct rbd_spec *spec = NULL;
6074
	struct rbd_client *rbdc;
6075
	bool read_only;
6076
	int rc;
6077 6078 6079 6080 6081

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

	/* parse add command */
6082
	rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec);
6083
	if (rc < 0)
6084
		goto out;
6085

6086 6087 6088
	rbdc = rbd_get_client(ceph_opts);
	if (IS_ERR(rbdc)) {
		rc = PTR_ERR(rbdc);
6089
		goto err_out_args;
6090
	}
6091 6092

	/* pick the pool */
6093
	rc = rbd_add_get_pool_id(rbdc, spec->pool_name);
6094 6095 6096
	if (rc < 0) {
		if (rc == -ENOENT)
			pr_info("pool %s does not exist\n", spec->pool_name);
6097
		goto err_out_client;
6098
	}
A
Alex Elder 已提交
6099
	spec->pool_id = (u64)rc;
6100

6101
	rbd_dev = rbd_dev_create(rbdc, spec, rbd_opts);
6102 6103
	if (!rbd_dev) {
		rc = -ENOMEM;
6104
		goto err_out_client;
6105
	}
6106 6107
	rbdc = NULL;		/* rbd_dev now owns this */
	spec = NULL;		/* rbd_dev now owns this */
6108
	rbd_opts = NULL;	/* rbd_dev now owns this */
6109

6110 6111 6112 6113 6114 6115
	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
6116
	down_write(&rbd_dev->header_rwsem);
6117
	rc = rbd_dev_image_probe(rbd_dev, 0);
6118 6119
	if (rc < 0) {
		up_write(&rbd_dev->header_rwsem);
6120
		goto err_out_rbd_dev;
6121
	}
6122

6123 6124
	/* If we are mapping a snapshot it must be marked read-only */

6125
	read_only = rbd_dev->opts->read_only;
6126 6127 6128 6129
	if (rbd_dev->spec->snap_id != CEPH_NOSNAP)
		read_only = true;
	rbd_dev->mapping.read_only = read_only;

6130
	rc = rbd_dev_device_setup(rbd_dev);
6131
	if (rc)
6132
		goto err_out_image_probe;
A
Alex Elder 已提交
6133

6134 6135 6136 6137
	rc = count;
out:
	module_put(THIS_MODULE);
	return rc;
6138

6139 6140
err_out_image_probe:
	rbd_dev_image_release(rbd_dev);
6141 6142
err_out_rbd_dev:
	rbd_dev_destroy(rbd_dev);
6143
err_out_client:
6144
	rbd_put_client(rbdc);
6145
err_out_args:
6146
	rbd_spec_put(spec);
6147
	kfree(rbd_opts);
6148
	goto out;
6149 6150
}

6151 6152 6153 6154 6155 6156 6157 6158 6159 6160 6161 6162 6163 6164 6165 6166 6167
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);
}

6168
static void rbd_dev_device_release(struct rbd_device *rbd_dev)
6169 6170
{
	rbd_free_disk(rbd_dev);
6171 6172 6173 6174 6175

	spin_lock(&rbd_dev_list_lock);
	list_del_init(&rbd_dev->node);
	spin_unlock(&rbd_dev_list_lock);

6176
	clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
6177
	device_del(&rbd_dev->dev);
A
Alex Elder 已提交
6178
	rbd_dev_mapping_clear(rbd_dev);
6179 6180
	if (!single_major)
		unregister_blkdev(rbd_dev->major, rbd_dev->name);
6181 6182
}

6183 6184
static void rbd_dev_remove_parent(struct rbd_device *rbd_dev)
{
A
Alex Elder 已提交
6185
	while (rbd_dev->parent) {
6186 6187 6188 6189 6190 6191 6192 6193 6194 6195 6196 6197
		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
6198
		rbd_assert(second);
6199
		rbd_dev_image_release(second);
6200
		rbd_dev_destroy(second);
A
Alex Elder 已提交
6201 6202 6203 6204
		first->parent = NULL;
		first->parent_overlap = 0;

		rbd_assert(first->parent_spec);
6205 6206 6207 6208 6209
		rbd_spec_put(first->parent_spec);
		first->parent_spec = NULL;
	}
}

6210 6211 6212
static ssize_t do_rbd_remove(struct bus_type *bus,
			     const char *buf,
			     size_t count)
6213 6214
{
	struct rbd_device *rbd_dev = NULL;
6215 6216
	struct list_head *tmp;
	int dev_id;
M
Mike Christie 已提交
6217
	char opt_buf[6];
6218
	bool already = false;
M
Mike Christie 已提交
6219
	bool force = false;
6220
	int ret;
6221

M
Mike Christie 已提交
6222 6223 6224 6225 6226
	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");
6227
		return -EINVAL;
M
Mike Christie 已提交
6228 6229 6230 6231 6232 6233 6234 6235 6236
	}
	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;
		}
	}
6237

6238 6239 6240 6241 6242 6243 6244 6245
	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;
		}
6246
	}
6247 6248
	if (!ret) {
		spin_lock_irq(&rbd_dev->lock);
M
Mike Christie 已提交
6249
		if (rbd_dev->open_count && !force)
6250 6251
			ret = -EBUSY;
		else
6252 6253
			already = test_and_set_bit(RBD_DEV_FLAG_REMOVING,
							&rbd_dev->flags);
6254 6255 6256
		spin_unlock_irq(&rbd_dev->lock);
	}
	spin_unlock(&rbd_dev_list_lock);
6257
	if (ret < 0 || already)
6258
		return ret;
6259

M
Mike Christie 已提交
6260 6261 6262 6263 6264 6265 6266 6267 6268
	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);
	}

6269 6270 6271 6272
	down_write(&rbd_dev->lock_rwsem);
	if (__rbd_is_lock_owner(rbd_dev))
		rbd_unlock(rbd_dev);
	up_write(&rbd_dev->lock_rwsem);
6273

6274
	rbd_dev_device_release(rbd_dev);
6275
	rbd_dev_image_release(rbd_dev);
6276
	rbd_dev_destroy(rbd_dev);
6277
	return count;
6278 6279
}

6280 6281 6282 6283 6284 6285 6286 6287 6288 6289 6290 6291 6292 6293 6294 6295 6296
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);
}

6297 6298
/*
 * create control files in sysfs
6299
 * /sys/bus/rbd/...
6300 6301 6302
 */
static int rbd_sysfs_init(void)
{
6303
	int ret;
6304

6305
	ret = device_register(&rbd_root_dev);
A
Alex Elder 已提交
6306
	if (ret < 0)
6307
		return ret;
6308

6309 6310 6311
	ret = bus_register(&rbd_bus_type);
	if (ret < 0)
		device_unregister(&rbd_root_dev);
6312 6313 6314 6315 6316 6317

	return ret;
}

static void rbd_sysfs_cleanup(void)
{
6318
	bus_unregister(&rbd_bus_type);
6319
	device_unregister(&rbd_root_dev);
6320 6321
}

6322 6323 6324
static int rbd_slab_init(void)
{
	rbd_assert(!rbd_img_request_cache);
G
Geliang Tang 已提交
6325
	rbd_img_request_cache = KMEM_CACHE(rbd_img_request, 0);
6326 6327 6328 6329
	if (!rbd_img_request_cache)
		return -ENOMEM;

	rbd_assert(!rbd_obj_request_cache);
G
Geliang Tang 已提交
6330
	rbd_obj_request_cache = KMEM_CACHE(rbd_obj_request, 0);
6331 6332 6333
	if (!rbd_obj_request_cache)
		goto out_err;

6334
	return 0;
6335

6336
out_err:
6337 6338
	kmem_cache_destroy(rbd_img_request_cache);
	rbd_img_request_cache = NULL;
6339 6340 6341 6342 6343
	return -ENOMEM;
}

static void rbd_slab_exit(void)
{
6344 6345 6346 6347
	rbd_assert(rbd_obj_request_cache);
	kmem_cache_destroy(rbd_obj_request_cache);
	rbd_obj_request_cache = NULL;

6348 6349 6350 6351 6352
	rbd_assert(rbd_img_request_cache);
	kmem_cache_destroy(rbd_img_request_cache);
	rbd_img_request_cache = NULL;
}

A
Alex Elder 已提交
6353
static int __init rbd_init(void)
6354 6355 6356
{
	int rc;

6357 6358 6359 6360
	if (!libceph_compatible(NULL)) {
		rbd_warn(NULL, "libceph incompatibility (quitting)");
		return -EINVAL;
	}
I
Ilya Dryomov 已提交
6361

6362
	rc = rbd_slab_init();
6363 6364
	if (rc)
		return rc;
I
Ilya Dryomov 已提交
6365

6366 6367
	/*
	 * The number of active work items is limited by the number of
I
Ilya Dryomov 已提交
6368
	 * rbd devices * queue depth, so leave @max_active at default.
6369 6370 6371 6372 6373 6374 6375
	 */
	rbd_wq = alloc_workqueue(RBD_DRV_NAME, WQ_MEM_RECLAIM, 0);
	if (!rbd_wq) {
		rc = -ENOMEM;
		goto err_out_slab;
	}

6376 6377 6378 6379
	if (single_major) {
		rbd_major = register_blkdev(0, RBD_DRV_NAME);
		if (rbd_major < 0) {
			rc = rbd_major;
6380
			goto err_out_wq;
6381 6382 6383
		}
	}

6384 6385
	rc = rbd_sysfs_init();
	if (rc)
6386 6387 6388 6389 6390 6391
		goto err_out_blkdev;

	if (single_major)
		pr_info("loaded (major %d)\n", rbd_major);
	else
		pr_info("loaded\n");
6392

I
Ilya Dryomov 已提交
6393 6394
	return 0;

6395 6396 6397
err_out_blkdev:
	if (single_major)
		unregister_blkdev(rbd_major, RBD_DRV_NAME);
6398 6399
err_out_wq:
	destroy_workqueue(rbd_wq);
I
Ilya Dryomov 已提交
6400 6401
err_out_slab:
	rbd_slab_exit();
6402
	return rc;
6403 6404
}

A
Alex Elder 已提交
6405
static void __exit rbd_exit(void)
6406
{
I
Ilya Dryomov 已提交
6407
	ida_destroy(&rbd_dev_id_ida);
6408
	rbd_sysfs_cleanup();
6409 6410
	if (single_major)
		unregister_blkdev(rbd_major, RBD_DRV_NAME);
6411
	destroy_workqueue(rbd_wq);
6412
	rbd_slab_exit();
6413 6414 6415 6416 6417
}

module_init(rbd_init);
module_exit(rbd_exit);

A
Alex Elder 已提交
6418
MODULE_AUTHOR("Alex Elder <elder@inktank.com>");
6419 6420 6421 6422 6423
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>");

6424
MODULE_DESCRIPTION("RADOS Block Device (RBD) driver");
6425
MODULE_LICENSE("GPL");