rbd.c 164.8 KB
Newer Older
1

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


   based on drivers/block/osdblk.c:

   Copyright 2009 Red Hat, Inc.

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

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

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



25
   For usage instructions, please refer to:
26

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

 */

#include <linux/ceph/libceph.h>
#include <linux/ceph/osd_client.h>
#include <linux/ceph/mon_client.h>
34
#include <linux/ceph/cls_lock_client.h>
35
#include <linux/ceph/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 */

A
Alex Elder 已提交
123 124
#define RBD_FEATURE_LAYERING	(1<<0)
#define RBD_FEATURE_STRIPINGV2	(1<<1)
125 126 127 128
#define RBD_FEATURE_EXCLUSIVE_LOCK (1<<2)
#define RBD_FEATURES_ALL	(RBD_FEATURE_LAYERING |		\
				 RBD_FEATURE_STRIPINGV2 |	\
				 RBD_FEATURE_EXCLUSIVE_LOCK)
A
Alex Elder 已提交
129 130 131

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

132
#define RBD_FEATURES_SUPPORTED	(RBD_FEATURES_ALL)
A
Alex Elder 已提交
133

A
Alex Elder 已提交
134 135 136 137
/*
 * An RBD device name will be "rbd#", where the "rbd" comes from
 * RBD_DRV_NAME above, and # is a unique integer identifier.
 */
138 139 140 141 142 143
#define DEV_NAME_LEN		32

/*
 * block device image metadata (in-memory version)
 */
struct rbd_image_header {
144
	/* These six fields never change for a given rbd image */
145
	char *object_prefix;
146
	__u8 obj_order;
147 148 149
	u64 stripe_unit;
	u64 stripe_count;
	u64 features;		/* Might be changeable someday? */
150

A
Alex Elder 已提交
151 152 153
	/* The remaining fields need to be updated occasionally */
	u64 image_size;
	struct ceph_snap_context *snapc;
154 155
	char *snap_names;	/* format 1 only */
	u64 *snap_sizes;	/* format 1 only */
156 157
};

158 159 160 161
/*
 * An rbd image specification.
 *
 * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely
A
Alex Elder 已提交
162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181
 * 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).
182 183 184
 */
struct rbd_spec {
	u64		pool_id;
185
	const char	*pool_name;
186

187 188
	const char	*image_id;
	const char	*image_name;
189 190

	u64		snap_id;
191
	const char	*snap_name;
192 193 194 195

	struct kref	kref;
};

196
/*
A
Alex Elder 已提交
197
 * an instance of the client.  multiple devices may share an rbd client.
198 199 200 201 202 203 204
 */
struct rbd_client {
	struct ceph_client	*client;
	struct kref		kref;
	struct list_head	node;
};

A
Alex Elder 已提交
205 206 207 208 209 210 211 212
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 *);

213 214 215
enum obj_request_type {
	OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES
};
A
Alex Elder 已提交
216

G
Guangliang Zhao 已提交
217 218 219
enum obj_operation_type {
	OBJ_OP_WRITE,
	OBJ_OP_READ,
220
	OBJ_OP_DISCARD,
G
Guangliang Zhao 已提交
221 222
};

223 224
enum obj_req_flags {
	OBJ_REQ_DONE,		/* completion flag: not done = 0, done = 1 */
225
	OBJ_REQ_IMG_DATA,	/* object usage: standalone = 0, image = 1 */
226 227
	OBJ_REQ_KNOWN,		/* EXISTS flag valid: no = 0, yes = 1 */
	OBJ_REQ_EXISTS,		/* target exists: no = 0, yes = 1 */
228 229
};

A
Alex Elder 已提交
230 231 232 233
struct rbd_obj_request {
	const char		*object_name;
	u64			offset;		/* object start byte */
	u64			length;		/* bytes from offset */
234
	unsigned long		flags;
A
Alex Elder 已提交
235

236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260
	/*
	 * 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
261 262 263
	u32			which;		/* posn image request list */

	enum obj_request_type	type;
264 265 266 267 268 269 270
	union {
		struct bio	*bio_list;
		struct {
			struct page	**pages;
			u32		page_count;
		};
	};
271
	struct page		**copyup_pages;
272
	u32			copyup_page_count;
A
Alex Elder 已提交
273 274 275 276

	struct ceph_osd_request	*osd_req;

	u64			xferred;	/* bytes transferred */
277
	int			result;
A
Alex Elder 已提交
278 279

	rbd_obj_callback_t	callback;
280
	struct completion	completion;
A
Alex Elder 已提交
281 282 283 284

	struct kref		kref;
};

A
Alex Elder 已提交
285
enum img_req_flags {
286 287
	IMG_REQ_WRITE,		/* I/O direction: read = 0, write = 1 */
	IMG_REQ_CHILD,		/* initiator: block = 0, child image = 1 */
288
	IMG_REQ_LAYERED,	/* ENOENT handling: normal = 0, layered = 1 */
289
	IMG_REQ_DISCARD,	/* discard: normal = 0, discard request = 1 */
A
Alex Elder 已提交
290 291
};

A
Alex Elder 已提交
292 293 294 295
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
296
	unsigned long		flags;
A
Alex Elder 已提交
297
	union {
298
		u64			snap_id;	/* for reads */
A
Alex Elder 已提交
299
		struct ceph_snap_context *snapc;	/* for writes */
300 301 302 303
	};
	union {
		struct request		*rq;		/* block request */
		struct rbd_obj_request	*obj_request;	/* obj req initiator */
A
Alex Elder 已提交
304
	};
305
	struct page		**copyup_pages;
306
	u32			copyup_page_count;
A
Alex Elder 已提交
307 308 309
	spinlock_t		completion_lock;/* protects next_completion */
	u32			next_completion;
	rbd_img_callback_t	callback;
310
	u64			xferred;/* aggregate bytes transferred */
311
	int			result;	/* first nonzero obj_request result */
A
Alex Elder 已提交
312 313 314 315 316 317 318 319

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

	struct kref		kref;
};

#define for_each_obj_request(ireq, oreq) \
320
	list_for_each_entry(oreq, &(ireq)->obj_requests, links)
A
Alex Elder 已提交
321
#define for_each_obj_request_from(ireq, oreq) \
322
	list_for_each_entry_from(oreq, &(ireq)->obj_requests, links)
A
Alex Elder 已提交
323
#define for_each_obj_request_safe(ireq, oreq, n) \
324
	list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links)
A
Alex Elder 已提交
325

326 327 328 329 330 331
enum rbd_watch_state {
	RBD_WATCH_STATE_UNREGISTERED,
	RBD_WATCH_STATE_REGISTERED,
	RBD_WATCH_STATE_ERROR,
};

332 333 334 335 336 337 338 339 340 341 342 343
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
344
struct rbd_mapping {
A
Alex Elder 已提交
345
	u64                     size;
A
Alex Elder 已提交
346
	u64                     features;
A
Alex Elder 已提交
347 348 349
	bool			read_only;
};

350 351 352 353
/*
 * a single device
 */
struct rbd_device {
A
Alex Elder 已提交
354
	int			dev_id;		/* blkdev unique id */
355 356

	int			major;		/* blkdev assigned major */
357
	int			minor;
358 359
	struct gendisk		*disk;		/* blkdev's gendisk and rq */

360
	u32			image_format;	/* Either 1 or 2 */
361 362 363 364
	struct rbd_client	*rbd_client;

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

365
	spinlock_t		lock;		/* queue, flags, open_count */
366 367

	struct rbd_image_header	header;
368
	unsigned long		flags;		/* possibly lock protected */
369
	struct rbd_spec		*spec;
370
	struct rbd_options	*opts;
371
	char			*config_info;	/* add{,_single_major} string */
372

373
	struct ceph_object_id	header_oid;
374
	struct ceph_object_locator header_oloc;
375

376
	struct ceph_file_layout	layout;		/* used for all rbd requests */
377

378 379
	struct mutex		watch_mutex;
	enum rbd_watch_state	watch_state;
380
	struct ceph_osd_linger_request *watch_handle;
381 382
	u64			watch_cookie;
	struct delayed_work	watch_dwork;
383

384 385 386 387 388 389 390 391 392
	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;

393
	struct workqueue_struct	*task_wq;
394

395 396
	struct rbd_spec		*parent_spec;
	u64			parent_overlap;
397
	atomic_t		parent_ref;
398
	struct rbd_device	*parent;
399

C
Christoph Hellwig 已提交
400 401 402
	/* Block layer tags. */
	struct blk_mq_tag_set	tag_set;

403 404
	/* protects updating the header */
	struct rw_semaphore     header_rwsem;
A
Alex Elder 已提交
405 406

	struct rbd_mapping	mapping;
407 408

	struct list_head	node;
409 410 411

	/* sysfs related */
	struct device		dev;
412
	unsigned long		open_count;	/* protected by lock */
413 414
};

415
/*
416 417 418 419
 * 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
420
 */
421 422
enum rbd_dev_flags {
	RBD_DEV_FLAG_EXISTS,	/* mapped snapshot has not been deleted */
423
	RBD_DEV_FLAG_REMOVING,	/* this mapping is being removed */
424
	RBD_DEV_FLAG_BLACKLISTED, /* our ceph_client is blacklisted */
425 426
};

427
static DEFINE_MUTEX(client_mutex);	/* Serialize client creation */
428

429
static LIST_HEAD(rbd_dev_list);    /* devices */
430 431
static DEFINE_SPINLOCK(rbd_dev_list_lock);

A
Alex Elder 已提交
432 433
static LIST_HEAD(rbd_client_list);		/* clients */
static DEFINE_SPINLOCK(rbd_client_list_lock);
434

435 436
/* Slab caches for frequently-allocated structures */

437
static struct kmem_cache	*rbd_img_request_cache;
438
static struct kmem_cache	*rbd_obj_request_cache;
439
static struct kmem_cache	*rbd_segment_name_cache;
440

441
static int rbd_major;
442 443
static DEFINE_IDA(rbd_dev_id_ida);

444 445
static struct workqueue_struct *rbd_wq;

446 447 448 449 450 451 452 453
/*
 * 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)");

454 455
static int rbd_img_request_submit(struct rbd_img_request *img_request);

A
Alex Elder 已提交
456 457 458 459
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);
460 461 462 463
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);
464
static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth);
465
static void rbd_spec_put(struct rbd_spec *spec);
A
Alex Elder 已提交
466

467 468
static int rbd_dev_id_to_minor(int dev_id)
{
469
	return dev_id << RBD_SINGLE_MAJOR_PART_SHIFT;
470 471 472 473
}

static int minor_to_rbd_dev_id(int minor)
{
474
	return minor >> RBD_SINGLE_MAJOR_PART_SHIFT;
475 476
}

477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499
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;
}

500 501
static BUS_ATTR(add, S_IWUSR, NULL, rbd_add);
static BUS_ATTR(remove, S_IWUSR, NULL, rbd_remove);
502 503
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);
504 505 506 507

static struct attribute *rbd_bus_attrs[] = {
	&bus_attr_add.attr,
	&bus_attr_remove.attr,
508 509
	&bus_attr_add_single_major.attr,
	&bus_attr_remove_single_major.attr,
510
	NULL,
A
Alex Elder 已提交
511
};
512 513 514 515

static umode_t rbd_bus_is_visible(struct kobject *kobj,
				  struct attribute *attr, int index)
{
516 517 518 519 520
	if (!single_major &&
	    (attr == &bus_attr_add_single_major.attr ||
	     attr == &bus_attr_remove_single_major.attr))
		return 0;

521 522 523 524 525 526 527 528
	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
529 530 531

static struct bus_type rbd_bus_type = {
	.name		= "rbd",
532
	.bus_groups	= rbd_bus_groups,
A
Alex Elder 已提交
533 534 535 536 537 538 539 540 541 542 543
};

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
544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570
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
571 572 573 574 575 576 577 578 579 580 581 582
#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 */
583

I
Ilya Dryomov 已提交
584
static void rbd_osd_copyup_callback(struct rbd_obj_request *obj_request);
585
static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request);
586 587
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
588

A
Alex Elder 已提交
589
static int rbd_dev_refresh(struct rbd_device *rbd_dev);
590
static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev);
591
static int rbd_dev_header_info(struct rbd_device *rbd_dev);
592
static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev);
593 594
static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
					u64 snap_id);
595 596 597 598
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);
599

600 601
static int rbd_open(struct block_device *bdev, fmode_t mode)
{
A
Alex Elder 已提交
602
	struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
603
	bool removing = false;
604

A
Alex Elder 已提交
605
	if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only)
606 607
		return -EROFS;

608
	spin_lock_irq(&rbd_dev->lock);
609 610 611 612
	if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags))
		removing = true;
	else
		rbd_dev->open_count++;
613
	spin_unlock_irq(&rbd_dev->lock);
614 615 616
	if (removing)
		return -ENOENT;

A
Alex Elder 已提交
617
	(void) get_device(&rbd_dev->dev);
618

619 620 621
	return 0;
}

622
static void rbd_release(struct gendisk *disk, fmode_t mode)
623 624
{
	struct rbd_device *rbd_dev = disk->private_data;
625 626
	unsigned long open_count_before;

627
	spin_lock_irq(&rbd_dev->lock);
628
	open_count_before = rbd_dev->open_count--;
629
	spin_unlock_irq(&rbd_dev->lock);
630
	rbd_assert(open_count_before > 0);
631

A
Alex Elder 已提交
632
	put_device(&rbd_dev->dev);
633 634
}

G
Guangliang Zhao 已提交
635 636
static int rbd_ioctl_set_ro(struct rbd_device *rbd_dev, unsigned long arg)
{
637
	int ret = 0;
G
Guangliang Zhao 已提交
638 639
	int val;
	bool ro;
640
	bool ro_changed = false;
G
Guangliang Zhao 已提交
641

642
	/* get_user() may sleep, so call it before taking rbd_dev->lock */
G
Guangliang Zhao 已提交
643 644 645 646 647 648 649 650
	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;

651 652 653 654 655 656 657
	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
658 659
	if (rbd_dev->mapping.read_only != ro) {
		rbd_dev->mapping.read_only = ro;
660
		ro_changed = true;
G
Guangliang Zhao 已提交
661 662
	}

663 664 665 666 667 668 669
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
670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696
}

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 */

697 698 699
static const struct block_device_operations rbd_bd_ops = {
	.owner			= THIS_MODULE,
	.open			= rbd_open,
700
	.release		= rbd_release,
G
Guangliang Zhao 已提交
701 702 703 704
	.ioctl			= rbd_ioctl,
#ifdef CONFIG_COMPAT
	.compat_ioctl		= rbd_compat_ioctl,
#endif
705 706 707
};

/*
708
 * Initialize an rbd client instance.  Success or not, this function
709
 * consumes ceph_opts.  Caller holds client_mutex.
710
 */
711
static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts)
712 713 714 715
{
	struct rbd_client *rbdc;
	int ret = -ENOMEM;

A
Alex Elder 已提交
716
	dout("%s:\n", __func__);
717 718 719 720 721 722 723
	rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL);
	if (!rbdc)
		goto out_opt;

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

A
Alex Elder 已提交
724
	rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0);
725
	if (IS_ERR(rbdc->client))
726
		goto out_rbdc;
A
Alex Elder 已提交
727
	ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */
728 729 730

	ret = ceph_open_session(rbdc->client);
	if (ret < 0)
731
		goto out_client;
732

A
Alex Elder 已提交
733
	spin_lock(&rbd_client_list_lock);
734
	list_add_tail(&rbdc->node, &rbd_client_list);
A
Alex Elder 已提交
735
	spin_unlock(&rbd_client_list_lock);
736

A
Alex Elder 已提交
737
	dout("%s: rbdc %p\n", __func__, rbdc);
738

739
	return rbdc;
740
out_client:
741
	ceph_destroy_client(rbdc->client);
742
out_rbdc:
743 744
	kfree(rbdc);
out_opt:
A
Alex Elder 已提交
745 746
	if (ceph_opts)
		ceph_destroy_options(ceph_opts);
A
Alex Elder 已提交
747 748
	dout("%s: error %d\n", __func__, ret);

V
Vasiliy Kulikov 已提交
749
	return ERR_PTR(ret);
750 751
}

752 753 754 755 756 757 758
static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc)
{
	kref_get(&rbdc->kref);

	return rbdc;
}

759
/*
760 761
 * Find a ceph client with specific addr and configuration.  If
 * found, bump its reference count.
762
 */
763
static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts)
764 765
{
	struct rbd_client *client_node;
766
	bool found = false;
767

A
Alex Elder 已提交
768
	if (ceph_opts->flags & CEPH_OPT_NOSHARE)
769 770
		return NULL;

771 772 773
	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)) {
774 775
			__rbd_get_client(client_node);

776 777 778 779 780 781 782
			found = true;
			break;
		}
	}
	spin_unlock(&rbd_client_list_lock);

	return found ? client_node : NULL;
783 784
}

785
/*
786
 * (Per device) rbd map options
787 788
 */
enum {
I
Ilya Dryomov 已提交
789
	Opt_queue_depth,
790 791 792 793
	Opt_last_int,
	/* int args above */
	Opt_last_string,
	/* string args above */
A
Alex Elder 已提交
794 795
	Opt_read_only,
	Opt_read_write,
I
Ilya Dryomov 已提交
796
	Opt_lock_on_read,
797
	Opt_err
798 799
};

A
Alex Elder 已提交
800
static match_table_t rbd_opts_tokens = {
I
Ilya Dryomov 已提交
801
	{Opt_queue_depth, "queue_depth=%d"},
802 803
	/* int args above */
	/* string args above */
A
Alex Elder 已提交
804
	{Opt_read_only, "read_only"},
A
Alex Elder 已提交
805 806 807
	{Opt_read_only, "ro"},		/* Alternate spelling */
	{Opt_read_write, "read_write"},
	{Opt_read_write, "rw"},		/* Alternate spelling */
I
Ilya Dryomov 已提交
808
	{Opt_lock_on_read, "lock_on_read"},
809
	{Opt_err, NULL}
810 811
};

A
Alex Elder 已提交
812
struct rbd_options {
I
Ilya Dryomov 已提交
813
	int	queue_depth;
A
Alex Elder 已提交
814
	bool	read_only;
I
Ilya Dryomov 已提交
815
	bool	lock_on_read;
A
Alex Elder 已提交
816 817
};

I
Ilya Dryomov 已提交
818
#define RBD_QUEUE_DEPTH_DEFAULT	BLKDEV_MAX_RQ
A
Alex Elder 已提交
819
#define RBD_READ_ONLY_DEFAULT	false
I
Ilya Dryomov 已提交
820
#define RBD_LOCK_ON_READ_DEFAULT false
A
Alex Elder 已提交
821

822 823
static int parse_rbd_opts_token(char *c, void *private)
{
A
Alex Elder 已提交
824
	struct rbd_options *rbd_opts = private;
825 826 827
	substring_t argstr[MAX_OPT_ARGS];
	int token, intval, ret;

A
Alex Elder 已提交
828
	token = match_token(c, rbd_opts_tokens, argstr);
829 830 831
	if (token < Opt_last_int) {
		ret = match_int(&argstr[0], &intval);
		if (ret < 0) {
832
			pr_err("bad mount option arg (not int) at '%s'\n", c);
833 834 835 836
			return ret;
		}
		dout("got int token %d val %d\n", token, intval);
	} else if (token > Opt_last_int && token < Opt_last_string) {
837
		dout("got string token %d val %s\n", token, argstr[0].from);
838 839 840 841 842
	} else {
		dout("got token %d\n", token);
	}

	switch (token) {
I
Ilya Dryomov 已提交
843 844 845 846 847 848 849
	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
850 851 852 853 854 855
	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
856 857 858
	case Opt_lock_on_read:
		rbd_opts->lock_on_read = true;
		break;
859
	default:
860 861
		/* libceph prints "bad option" msg */
		return -EINVAL;
862
	}
863

864 865 866
	return 0;
}

G
Guangliang Zhao 已提交
867 868 869 870 871 872 873
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";
874 875
	case OBJ_OP_DISCARD:
		return "discard";
G
Guangliang Zhao 已提交
876 877 878 879 880
	default:
		return "???";
	}
}

881 882
/*
 * Get a ceph client with specific addr and configuration, if one does
883 884
 * not exist create it.  Either way, ceph_opts is consumed by this
 * function.
885
 */
886
static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts)
887
{
888
	struct rbd_client *rbdc;
889

890
	mutex_lock_nested(&client_mutex, SINGLE_DEPTH_NESTING);
891
	rbdc = rbd_client_find(ceph_opts);
892
	if (rbdc)	/* using an existing client */
A
Alex Elder 已提交
893
		ceph_destroy_options(ceph_opts);
894
	else
895
		rbdc = rbd_client_create(ceph_opts);
896
	mutex_unlock(&client_mutex);
897

898
	return rbdc;
899 900 901 902
}

/*
 * Destroy ceph client
A
Alex Elder 已提交
903
 *
A
Alex Elder 已提交
904
 * Caller must hold rbd_client_list_lock.
905 906 907 908 909
 */
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
910
	dout("%s: rbdc %p\n", __func__, rbdc);
911
	spin_lock(&rbd_client_list_lock);
912
	list_del(&rbdc->node);
913
	spin_unlock(&rbd_client_list_lock);
914 915 916 917 918 919 920 921 922

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

/*
 * Drop reference to ceph client node. If it's not referenced anymore, release
 * it.
 */
923
static void rbd_put_client(struct rbd_client *rbdc)
924
{
925 926
	if (rbdc)
		kref_put(&rbdc->kref, rbd_client_release);
927 928
}

929 930 931 932 933
static bool rbd_image_format_valid(u32 image_format)
{
	return image_format == 1 || image_format == 2;
}

934 935
static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk)
{
936 937 938 939 940 941 942
	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
943 944 945 946 947 948 949 950 951 952
	/* 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;

953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970
	/*
	 * 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;
971 972
}

I
Ilya Dryomov 已提交
973 974 975 976 977 978 979 980
/*
 * returns the size of an object in the image
 */
static u32 rbd_obj_bytes(struct rbd_image_header *header)
{
	return 1U << header->obj_order;
}

981
/*
982 983
 * Fill an rbd image header with information from the given format 1
 * on-disk header.
984
 */
A
Alex Elder 已提交
985
static int rbd_header_from_disk(struct rbd_device *rbd_dev,
986
				 struct rbd_image_header_ondisk *ondisk)
987
{
A
Alex Elder 已提交
988
	struct rbd_image_header *header = &rbd_dev->header;
989 990 991 992 993
	bool first_time = header->object_prefix == NULL;
	struct ceph_snap_context *snapc;
	char *object_prefix = NULL;
	char *snap_names = NULL;
	u64 *snap_sizes = NULL;
994
	u32 snap_count;
995
	int ret = -ENOMEM;
996
	u32 i;
997

998
	/* Allocate this now to avoid having to handle failure below */
A
Alex Elder 已提交
999

1000
	if (first_time) {
1001 1002 1003
		object_prefix = kstrndup(ondisk->object_prefix,
					 sizeof(ondisk->object_prefix),
					 GFP_KERNEL);
1004 1005 1006
		if (!object_prefix)
			return -ENOMEM;
	}
A
Alex Elder 已提交
1007

1008
	/* Allocate the snapshot context and fill it in */
A
Alex Elder 已提交
1009

1010 1011 1012 1013 1014
	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);
1015
	if (snap_count) {
1016
		struct rbd_image_snap_ondisk *snaps;
A
Alex Elder 已提交
1017 1018
		u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len);

1019
		/* We'll keep a copy of the snapshot names... */
1020

1021 1022 1023 1024
		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
1025 1026
			goto out_err;

1027
		/* ...as well as the array of their sizes. */
1028 1029 1030
		snap_sizes = kmalloc_array(snap_count,
					   sizeof(*header->snap_sizes),
					   GFP_KERNEL);
1031
		if (!snap_sizes)
A
Alex Elder 已提交
1032
			goto out_err;
1033

A
Alex Elder 已提交
1034
		/*
1035 1036 1037
		 * Copy the names, and fill in each snapshot's id
		 * and size.
		 *
1038
		 * Note that rbd_dev_v1_header_info() guarantees the
1039
		 * ondisk buffer we're working with has
A
Alex Elder 已提交
1040 1041 1042
		 * snap_names_len bytes beyond the end of the
		 * snapshot id array, this memcpy() is safe.
		 */
1043 1044 1045 1046 1047 1048
		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);
		}
1049
	}
A
Alex Elder 已提交
1050

1051
	/* We won't fail any more, fill in the header */
1052

1053 1054 1055
	if (first_time) {
		header->object_prefix = object_prefix;
		header->obj_order = ondisk->options.order;
1056
	} else {
A
Alex Elder 已提交
1057 1058 1059
		ceph_put_snap_context(header->snapc);
		kfree(header->snap_names);
		kfree(header->snap_sizes);
1060
	}
1061

1062
	/* The remaining fields always get updated (when we refresh) */
1063

A
Alex Elder 已提交
1064
	header->image_size = le64_to_cpu(ondisk->image_size);
1065 1066 1067
	header->snapc = snapc;
	header->snap_names = snap_names;
	header->snap_sizes = snap_sizes;
1068

1069
	return 0;
1070 1071
out_2big:
	ret = -EIO;
A
Alex Elder 已提交
1072
out_err:
1073 1074 1075 1076
	kfree(snap_sizes);
	kfree(snap_names);
	ceph_put_snap_context(snapc);
	kfree(object_prefix);
1077

1078
	return ret;
1079 1080
}

1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095
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);
}

1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119
/*
 * 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.
 */
1120 1121 1122
static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id)
{
	struct ceph_snap_context *snapc = rbd_dev->header.snapc;
1123
	u64 *found;
1124

1125 1126
	found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps,
				sizeof (snap_id), snapid_compare_reverse);
1127

1128
	return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX;
1129 1130
}

1131 1132
static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev,
					u64 snap_id)
1133
{
1134
	u32 which;
1135
	const char *snap_name;
1136

1137 1138
	which = rbd_dev_snap_index(rbd_dev, snap_id);
	if (which == BAD_SNAP_INDEX)
1139
		return ERR_PTR(-ENOENT);
1140

1141 1142
	snap_name = _rbd_dev_v1_snap_name(rbd_dev, which);
	return snap_name ? snap_name : ERR_PTR(-ENOMEM);
1143 1144 1145 1146
}

static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id)
{
1147 1148 1149
	if (snap_id == CEPH_NOSNAP)
		return RBD_SNAP_HEAD_NAME;

1150 1151 1152
	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);
1153

1154
	return rbd_dev_v2_snap_name(rbd_dev, snap_id);
1155 1156
}

1157 1158
static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
				u64 *snap_size)
1159
{
1160 1161 1162 1163 1164
	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;
1165

1166 1167 1168
		which = rbd_dev_snap_index(rbd_dev, snap_id);
		if (which == BAD_SNAP_INDEX)
			return -ENOENT;
1169

1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181
		*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;
1182 1183
}

1184 1185
static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
			u64 *snap_features)
1186
{
1187 1188 1189 1190 1191
	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 */
1192
	} else {
1193 1194
		u64 features = 0;
		int ret;
1195

1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206
		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)
{
1207
	u64 snap_id = rbd_dev->spec->snap_id;
1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221
	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;

1222
	return 0;
1223 1224
}

A
Alex Elder 已提交
1225 1226 1227 1228
static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev)
{
	rbd_dev->mapping.size = 0;
	rbd_dev->mapping.features = 0;
1229 1230
}

1231 1232 1233 1234 1235 1236 1237
static void rbd_segment_name_free(const char *name)
{
	/* The explicit cast here is needed to drop the const qualifier */

	kmem_cache_free(rbd_segment_name_cache, (void *)name);
}

A
Alex Elder 已提交
1238
static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset)
1239
{
A
Alex Elder 已提交
1240 1241 1242
	char *name;
	u64 segment;
	int ret;
1243
	char *name_format;
1244

1245
	name = kmem_cache_alloc(rbd_segment_name_cache, GFP_NOIO);
A
Alex Elder 已提交
1246 1247 1248
	if (!name)
		return NULL;
	segment = offset >> rbd_dev->header.obj_order;
1249 1250 1251
	name_format = "%s.%012llx";
	if (rbd_dev->image_format == 2)
		name_format = "%s.%016llx";
1252
	ret = snprintf(name, CEPH_MAX_OID_NAME_LEN + 1, name_format,
A
Alex Elder 已提交
1253
			rbd_dev->header.object_prefix, segment);
1254
	if (ret < 0 || ret > CEPH_MAX_OID_NAME_LEN) {
A
Alex Elder 已提交
1255 1256
		pr_err("error formatting segment name for #%llu (%d)\n",
			segment, ret);
1257
		rbd_segment_name_free(name);
A
Alex Elder 已提交
1258 1259
		name = NULL;
	}
1260

A
Alex Elder 已提交
1261 1262
	return name;
}
1263

A
Alex Elder 已提交
1264 1265
static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset)
{
I
Ilya Dryomov 已提交
1266
	u64 segment_size = rbd_obj_bytes(&rbd_dev->header);
1267

A
Alex Elder 已提交
1268 1269 1270 1271 1272 1273
	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
1274
	u64 segment_size = rbd_obj_bytes(&rbd_dev->header);
A
Alex Elder 已提交
1275 1276 1277

	offset &= segment_size - 1;

A
Alex Elder 已提交
1278
	rbd_assert(length <= U64_MAX - offset);
A
Alex Elder 已提交
1279 1280 1281 1282
	if (offset + length > segment_size)
		length = segment_size - offset;

	return length;
1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304
}

/*
 * 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)
{
1305 1306
	struct bio_vec bv;
	struct bvec_iter iter;
1307 1308 1309 1310 1311
	unsigned long flags;
	void *buf;
	int pos = 0;

	while (chain) {
1312 1313
		bio_for_each_segment(bv, chain, iter) {
			if (pos + bv.bv_len > start_ofs) {
1314
				int remainder = max(start_ofs - pos, 0);
1315
				buf = bvec_kmap_irq(&bv, &flags);
1316
				memset(buf + remainder, 0,
1317 1318
				       bv.bv_len - remainder);
				flush_dcache_page(bv.bv_page);
1319
				bvec_kunmap_irq(buf, &flags);
1320
			}
1321
			pos += bv.bv_len;
1322 1323 1324 1325 1326 1327
		}

		chain = chain->bi_next;
	}
}

A
Alex Elder 已提交
1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345
/*
 * 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;

1346 1347
		page_offset = offset & ~PAGE_MASK;
		length = min_t(size_t, PAGE_SIZE - page_offset, end - offset);
A
Alex Elder 已提交
1348 1349 1350
		local_irq_save(flags);
		kaddr = kmap_atomic(*page);
		memset(kaddr + page_offset, 0, length);
1351
		flush_dcache_page(*page);
A
Alex Elder 已提交
1352 1353 1354 1355 1356 1357 1358 1359
		kunmap_atomic(kaddr);
		local_irq_restore(flags);

		offset += length;
		page++;
	}
}

1360
/*
A
Alex Elder 已提交
1361 1362
 * Clone a portion of a bio, starting at the given byte offset
 * and continuing for the number of bytes indicated.
1363
 */
A
Alex Elder 已提交
1364 1365 1366 1367
static struct bio *bio_clone_range(struct bio *bio_src,
					unsigned int offset,
					unsigned int len,
					gfp_t gfpmask)
1368
{
A
Alex Elder 已提交
1369 1370
	struct bio *bio;

K
Kent Overstreet 已提交
1371
	bio = bio_clone(bio_src, gfpmask);
A
Alex Elder 已提交
1372 1373
	if (!bio)
		return NULL;	/* ENOMEM */
1374

K
Kent Overstreet 已提交
1375
	bio_advance(bio, offset);
1376
	bio->bi_iter.bi_size = len;
A
Alex Elder 已提交
1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406

	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 */

1407
	if (!bi || off >= bi->bi_iter.bi_size || !len)
A
Alex Elder 已提交
1408
		return NULL;		/* Nothing to clone */
1409

A
Alex Elder 已提交
1410 1411 1412 1413 1414
	end = &chain;
	while (len) {
		unsigned int bi_size;
		struct bio *bio;

1415 1416
		if (!bi) {
			rbd_warn(NULL, "bio_chain exhausted with %u left", len);
A
Alex Elder 已提交
1417
			goto out_err;	/* EINVAL; ran out of bio's */
1418
		}
1419
		bi_size = min_t(unsigned int, bi->bi_iter.bi_size - off, len);
A
Alex Elder 已提交
1420 1421 1422 1423 1424 1425
		bio = bio_clone_range(bi, off, bi_size, gfpmask);
		if (!bio)
			goto out_err;	/* ENOMEM */

		*end = bio;
		end = &bio->bi_next;
1426

A
Alex Elder 已提交
1427
		off += bi_size;
1428
		if (off == bi->bi_iter.bi_size) {
A
Alex Elder 已提交
1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439
			bi = bi->bi_next;
			off = 0;
		}
		len -= bi_size;
	}
	*bio_src = bi;
	*offset = off;

	return chain;
out_err:
	bio_chain_put(chain);
1440 1441 1442 1443

	return NULL;
}

1444 1445 1446 1447 1448
/*
 * 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
1449
static void obj_request_img_data_set(struct rbd_obj_request *obj_request)
1450
{
A
Alex Elder 已提交
1451
	if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) {
1452 1453
		struct rbd_device *rbd_dev;

A
Alex Elder 已提交
1454
		rbd_dev = obj_request->img_request->rbd_dev;
1455
		rbd_warn(rbd_dev, "obj_request %p already marked img_data",
1456 1457 1458 1459
			obj_request);
	}
}

A
Alex Elder 已提交
1460
static bool obj_request_img_data_test(struct rbd_obj_request *obj_request)
1461 1462
{
	smp_mb();
A
Alex Elder 已提交
1463
	return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0;
1464 1465
}

A
Alex Elder 已提交
1466
static void obj_request_done_set(struct rbd_obj_request *obj_request)
1467
{
A
Alex Elder 已提交
1468 1469
	if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) {
		struct rbd_device *rbd_dev = NULL;
1470

A
Alex Elder 已提交
1471 1472
		if (obj_request_img_data_test(obj_request))
			rbd_dev = obj_request->img_request->rbd_dev;
1473
		rbd_warn(rbd_dev, "obj_request %p already marked done",
1474 1475 1476 1477
			obj_request);
	}
}

A
Alex Elder 已提交
1478
static bool obj_request_done_test(struct rbd_obj_request *obj_request)
1479 1480
{
	smp_mb();
A
Alex Elder 已提交
1481
	return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0;
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 1509 1510 1511 1512 1513 1514
/*
 * 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;
}

1515 1516 1517 1518 1519 1520 1521 1522
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
1523 1524
static void rbd_obj_request_get(struct rbd_obj_request *obj_request)
{
A
Alex Elder 已提交
1525 1526
	dout("%s: obj %p (was %d)\n", __func__, obj_request,
		atomic_read(&obj_request->kref.refcount));
A
Alex Elder 已提交
1527 1528 1529 1530 1531 1532 1533
	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
1534 1535
	dout("%s: obj %p (was %d)\n", __func__, obj_request,
		atomic_read(&obj_request->kref.refcount));
A
Alex Elder 已提交
1536 1537 1538
	kref_put(&obj_request->kref, rbd_obj_request_destroy);
}

1539 1540 1541 1542 1543 1544 1545
static void rbd_img_request_get(struct rbd_img_request *img_request)
{
	dout("%s: img %p (was %d)\n", __func__, img_request,
	     atomic_read(&img_request->kref.refcount));
	kref_get(&img_request->kref);
}

1546 1547
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
1548 1549 1550 1551
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
1552 1553
	dout("%s: img %p (was %d)\n", __func__, img_request,
		atomic_read(&img_request->kref.refcount));
1554 1555 1556 1557
	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
1558 1559 1560 1561 1562
}

static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request,
					struct rbd_obj_request *obj_request)
{
1563 1564
	rbd_assert(obj_request->img_request == NULL);

1565
	/* Image request now owns object's original reference */
A
Alex Elder 已提交
1566
	obj_request->img_request = img_request;
1567
	obj_request->which = img_request->obj_request_count;
1568 1569
	rbd_assert(!obj_request_img_data_test(obj_request));
	obj_request_img_data_set(obj_request);
A
Alex Elder 已提交
1570
	rbd_assert(obj_request->which != BAD_WHICH);
1571 1572
	img_request->obj_request_count++;
	list_add_tail(&obj_request->links, &img_request->obj_requests);
A
Alex Elder 已提交
1573 1574
	dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
		obj_request->which);
A
Alex Elder 已提交
1575 1576 1577 1578 1579 1580
}

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);
1581

A
Alex Elder 已提交
1582 1583
	dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
		obj_request->which);
A
Alex Elder 已提交
1584
	list_del(&obj_request->links);
1585 1586 1587 1588
	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;
1589
	rbd_assert(obj_request_img_data_test(obj_request));
A
Alex Elder 已提交
1590 1591
	rbd_assert(obj_request->img_request == img_request);
	obj_request->img_request = NULL;
1592
	obj_request->callback = NULL;
A
Alex Elder 已提交
1593 1594 1595 1596 1597 1598
	rbd_obj_request_put(obj_request);
}

static bool obj_request_type_valid(enum obj_request_type type)
{
	switch (type) {
1599
	case OBJ_REQUEST_NODATA:
A
Alex Elder 已提交
1600
	case OBJ_REQUEST_BIO:
1601
	case OBJ_REQUEST_PAGES:
A
Alex Elder 已提交
1602 1603 1604 1605 1606 1607
		return true;
	default:
		return false;
	}
}

1608 1609
static void rbd_img_obj_callback(struct rbd_obj_request *obj_request);

1610
static void rbd_obj_request_submit(struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1611
{
1612 1613 1614
	struct ceph_osd_request *osd_req = obj_request->osd_req;

	dout("%s %p osd_req %p\n", __func__, obj_request, osd_req);
1615 1616 1617 1618
	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);
	}
1619
	ceph_osdc_start_request(osd_req->r_osdc, osd_req, false);
A
Alex Elder 已提交
1620 1621 1622 1623
}

static void rbd_img_request_complete(struct rbd_img_request *img_request)
{
1624

A
Alex Elder 已提交
1625
	dout("%s: img %p\n", __func__, img_request);
1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641

	/*
	 * 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
1642 1643 1644 1645 1646 1647
	if (img_request->callback)
		img_request->callback(img_request);
	else
		rbd_img_request_put(img_request);
}

A
Alex Elder 已提交
1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664
/*
 * 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;
}

1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679
/*
 * 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;
}

1680 1681 1682 1683 1684 1685
static void img_request_child_set(struct rbd_img_request *img_request)
{
	set_bit(IMG_REQ_CHILD, &img_request->flags);
	smp_mb();
}

1686 1687 1688 1689 1690 1691
static void img_request_child_clear(struct rbd_img_request *img_request)
{
	clear_bit(IMG_REQ_CHILD, &img_request->flags);
	smp_mb();
}

1692 1693 1694 1695 1696 1697
static bool img_request_child_test(struct rbd_img_request *img_request)
{
	smp_mb();
	return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0;
}

1698 1699 1700 1701 1702 1703
static void img_request_layered_set(struct rbd_img_request *img_request)
{
	set_bit(IMG_REQ_LAYERED, &img_request->flags);
	smp_mb();
}

1704 1705 1706 1707 1708 1709
static void img_request_layered_clear(struct rbd_img_request *img_request)
{
	clear_bit(IMG_REQ_LAYERED, &img_request->flags);
	smp_mb();
}

1710 1711 1712 1713 1714 1715
static bool img_request_layered_test(struct rbd_img_request *img_request)
{
	smp_mb();
	return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0;
}

1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726
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;
}

1727 1728 1729
static void
rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request)
{
A
Alex Elder 已提交
1730 1731 1732
	u64 xferred = obj_request->xferred;
	u64 length = obj_request->length;

1733 1734
	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
1735
		xferred, length);
1736
	/*
1737 1738 1739 1740 1741 1742
	 * 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.
1743
	 */
A
Alex Elder 已提交
1744
	rbd_assert(obj_request->type != OBJ_REQUEST_NODATA);
1745
	if (obj_request->result == -ENOENT) {
A
Alex Elder 已提交
1746 1747 1748 1749
		if (obj_request->type == OBJ_REQUEST_BIO)
			zero_bio_chain(obj_request->bio_list, 0);
		else
			zero_pages(obj_request->pages, 0, length);
1750
		obj_request->result = 0;
A
Alex Elder 已提交
1751 1752 1753 1754 1755
	} 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);
1756
	}
1757
	obj_request->xferred = length;
1758 1759 1760
	obj_request_done_set(obj_request);
}

A
Alex Elder 已提交
1761 1762
static void rbd_obj_request_complete(struct rbd_obj_request *obj_request)
{
A
Alex Elder 已提交
1763 1764
	dout("%s: obj %p cb %p\n", __func__, obj_request,
		obj_request->callback);
A
Alex Elder 已提交
1765 1766
	if (obj_request->callback)
		obj_request->callback(obj_request);
1767 1768
	else
		complete_all(&obj_request->completion);
A
Alex Elder 已提交
1769 1770
}

1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786
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);
}

1787
static void rbd_osd_read_callback(struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1788
{
A
Alex Elder 已提交
1789
	struct rbd_img_request *img_request = NULL;
A
Alex Elder 已提交
1790
	struct rbd_device *rbd_dev = NULL;
A
Alex Elder 已提交
1791 1792 1793 1794 1795
	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
1796
		rbd_dev = img_request->rbd_dev;
A
Alex Elder 已提交
1797
	}
A
Alex Elder 已提交
1798 1799 1800 1801

	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
1802 1803
	if (layered && obj_request->result == -ENOENT &&
			obj_request->img_offset < rbd_dev->parent_overlap)
A
Alex Elder 已提交
1804 1805
		rbd_img_parent_read(obj_request);
	else if (img_request)
1806 1807 1808
		rbd_img_obj_request_read_callback(obj_request);
	else
		obj_request_done_set(obj_request);
A
Alex Elder 已提交
1809 1810
}

1811
static void rbd_osd_write_callback(struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1812
{
1813 1814 1815
	dout("%s: obj %p result %d %llu\n", __func__, obj_request,
		obj_request->result, obj_request->length);
	/*
A
Alex Elder 已提交
1816 1817
	 * There is no such thing as a successful short write.  Set
	 * it to our originally-requested length.
1818 1819
	 */
	obj_request->xferred = obj_request->length;
1820
	obj_request_done_set(obj_request);
A
Alex Elder 已提交
1821 1822
}

1823 1824 1825 1826 1827 1828 1829 1830 1831
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;
1832 1833 1834
	/* discarding a non-existent object is not a problem */
	if (obj_request->result == -ENOENT)
		obj_request->result = 0;
1835 1836 1837
	obj_request_done_set(obj_request);
}

A
Alex Elder 已提交
1838 1839 1840 1841
/*
 * 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.
 */
1842
static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1843
{
A
Alex Elder 已提交
1844
	dout("%s: obj %p\n", __func__, obj_request);
A
Alex Elder 已提交
1845 1846 1847
	obj_request_done_set(obj_request);
}

I
Ilya Dryomov 已提交
1848 1849 1850 1851 1852 1853 1854 1855 1856 1857
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);
}

1858
static void rbd_osd_req_callback(struct ceph_osd_request *osd_req)
A
Alex Elder 已提交
1859 1860 1861 1862
{
	struct rbd_obj_request *obj_request = osd_req->r_priv;
	u16 opcode;

1863
	dout("%s: osd_req %p\n", __func__, osd_req);
A
Alex Elder 已提交
1864
	rbd_assert(osd_req == obj_request->osd_req);
A
Alex Elder 已提交
1865 1866 1867 1868 1869 1870
	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
1871

1872 1873
	if (osd_req->r_result < 0)
		obj_request->result = osd_req->r_result;
A
Alex Elder 已提交
1874

1875 1876
	/*
	 * We support a 64-bit length, but ultimately it has to be
C
Christoph Hellwig 已提交
1877 1878
	 * passed to the block layer, which just supports a 32-bit
	 * length field.
1879
	 */
1880
	obj_request->xferred = osd_req->r_ops[0].outdata_len;
A
Alex Elder 已提交
1881
	rbd_assert(obj_request->xferred < (u64)UINT_MAX);
1882

1883
	opcode = osd_req->r_ops[0].op;
A
Alex Elder 已提交
1884 1885
	switch (opcode) {
	case CEPH_OSD_OP_READ:
1886
		rbd_osd_read_callback(obj_request);
A
Alex Elder 已提交
1887
		break;
1888
	case CEPH_OSD_OP_SETALLOCHINT:
1889 1890
		rbd_assert(osd_req->r_ops[1].op == CEPH_OSD_OP_WRITE ||
			   osd_req->r_ops[1].op == CEPH_OSD_OP_WRITEFULL);
1891
		/* fall through */
A
Alex Elder 已提交
1892
	case CEPH_OSD_OP_WRITE:
1893
	case CEPH_OSD_OP_WRITEFULL:
1894
		rbd_osd_write_callback(obj_request);
A
Alex Elder 已提交
1895
		break;
A
Alex Elder 已提交
1896
	case CEPH_OSD_OP_STAT:
1897
		rbd_osd_stat_callback(obj_request);
A
Alex Elder 已提交
1898
		break;
1899 1900 1901 1902 1903
	case CEPH_OSD_OP_DELETE:
	case CEPH_OSD_OP_TRUNCATE:
	case CEPH_OSD_OP_ZERO:
		rbd_osd_discard_callback(obj_request);
		break;
1904
	case CEPH_OSD_OP_CALL:
I
Ilya Dryomov 已提交
1905 1906
		rbd_osd_call_callback(obj_request);
		break;
A
Alex Elder 已提交
1907
	default:
1908
		rbd_warn(NULL, "%s: unsupported op %hu",
A
Alex Elder 已提交
1909 1910 1911 1912
			obj_request->object_name, (unsigned short) opcode);
		break;
	}

1913
	if (obj_request_done_test(obj_request))
A
Alex Elder 已提交
1914 1915 1916
		rbd_obj_request_complete(obj_request);
}

1917
static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1918
{
1919
	struct ceph_osd_request *osd_req = obj_request->osd_req;
A
Alex Elder 已提交
1920

1921 1922
	rbd_assert(obj_request_img_data_test(obj_request));
	osd_req->r_snapid = obj_request->img_request->snap_id;
1923 1924 1925 1926 1927 1928
}

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

1929 1930
	osd_req->r_mtime = CURRENT_TIME;
	osd_req->r_data_offset = obj_request->offset;
A
Alex Elder 已提交
1931 1932
}

1933 1934 1935 1936 1937 1938
/*
 * 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
1939 1940
static struct ceph_osd_request *rbd_osd_req_create(
					struct rbd_device *rbd_dev,
G
Guangliang Zhao 已提交
1941
					enum obj_operation_type op_type,
1942
					unsigned int num_ops,
A
Alex Elder 已提交
1943
					struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1944 1945 1946 1947 1948
{
	struct ceph_snap_context *snapc = NULL;
	struct ceph_osd_client *osdc;
	struct ceph_osd_request *osd_req;

1949 1950
	if (obj_request_img_data_test(obj_request) &&
		(op_type == OBJ_OP_DISCARD || op_type == OBJ_OP_WRITE)) {
1951
		struct rbd_img_request *img_request = obj_request->img_request;
1952 1953 1954 1955 1956
		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
1957
		snapc = img_request->snapc;
A
Alex Elder 已提交
1958 1959
	}

G
Guangliang Zhao 已提交
1960
	rbd_assert(num_ops == 1 || ((op_type == OBJ_OP_WRITE) && num_ops == 2));
1961 1962

	/* Allocate and initialize the request, for the num_ops ops */
A
Alex Elder 已提交
1963 1964

	osdc = &rbd_dev->rbd_client->client->osdc;
1965
	osd_req = ceph_osdc_alloc_request(osdc, snapc, num_ops, false,
1966
					  GFP_NOIO);
A
Alex Elder 已提交
1967
	if (!osd_req)
1968
		goto fail;
A
Alex Elder 已提交
1969

1970
	if (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD)
A
Alex Elder 已提交
1971
		osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
A
Alex Elder 已提交
1972
	else
A
Alex Elder 已提交
1973 1974 1975 1976 1977
		osd_req->r_flags = CEPH_OSD_FLAG_READ;

	osd_req->r_callback = rbd_osd_req_callback;
	osd_req->r_priv = obj_request;

1978
	osd_req->r_base_oloc.pool = rbd_dev->layout.pool_id;
1979 1980 1981
	if (ceph_oid_aprintf(&osd_req->r_base_oid, GFP_NOIO, "%s",
			     obj_request->object_name))
		goto fail;
A
Alex Elder 已提交
1982

1983 1984 1985
	if (ceph_osdc_alloc_messages(osd_req, GFP_NOIO))
		goto fail;

A
Alex Elder 已提交
1986
	return osd_req;
1987 1988 1989 1990

fail:
	ceph_osdc_put_request(osd_req);
	return NULL;
A
Alex Elder 已提交
1991 1992
}

1993
/*
1994 1995 1996 1997
 * 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.
1998 1999 2000 2001 2002 2003 2004 2005 2006
 */
static struct ceph_osd_request *
rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request)
{
	struct rbd_img_request *img_request;
	struct ceph_snap_context *snapc;
	struct rbd_device *rbd_dev;
	struct ceph_osd_client *osdc;
	struct ceph_osd_request *osd_req;
2007
	int num_osd_ops = 3;
2008 2009 2010 2011

	rbd_assert(obj_request_img_data_test(obj_request));
	img_request = obj_request->img_request;
	rbd_assert(img_request);
2012 2013
	rbd_assert(img_request_write_test(img_request) ||
			img_request_discard_test(img_request));
2014

2015 2016 2017 2018
	if (img_request_discard_test(img_request))
		num_osd_ops = 2;

	/* Allocate and initialize the request, for all the ops */
2019 2020 2021 2022

	snapc = img_request->snapc;
	rbd_dev = img_request->rbd_dev;
	osdc = &rbd_dev->rbd_client->client->osdc;
2023
	osd_req = ceph_osdc_alloc_request(osdc, snapc, num_osd_ops,
2024
						false, GFP_NOIO);
2025
	if (!osd_req)
2026
		goto fail;
2027 2028 2029 2030 2031

	osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
	osd_req->r_callback = rbd_osd_req_callback;
	osd_req->r_priv = obj_request;

2032
	osd_req->r_base_oloc.pool = rbd_dev->layout.pool_id;
2033 2034 2035
	if (ceph_oid_aprintf(&osd_req->r_base_oid, GFP_NOIO, "%s",
			     obj_request->object_name))
		goto fail;
2036

2037 2038 2039
	if (ceph_osdc_alloc_messages(osd_req, GFP_NOIO))
		goto fail;

2040
	return osd_req;
2041 2042 2043 2044

fail:
	ceph_osdc_put_request(osd_req);
	return NULL;
2045 2046 2047
}


A
Alex Elder 已提交
2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065
static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req)
{
	ceph_osdc_put_request(osd_req);
}

/* object_name is assumed to be a non-null pointer and NUL-terminated */

static struct rbd_obj_request *rbd_obj_request_create(const char *object_name,
						u64 offset, u64 length,
						enum obj_request_type type)
{
	struct rbd_obj_request *obj_request;
	size_t size;
	char *name;

	rbd_assert(obj_request_type_valid(type));

	size = strlen(object_name) + 1;
2066
	name = kmalloc(size, GFP_NOIO);
2067
	if (!name)
A
Alex Elder 已提交
2068 2069
		return NULL;

2070
	obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_NOIO);
2071 2072 2073 2074 2075
	if (!obj_request) {
		kfree(name);
		return NULL;
	}

A
Alex Elder 已提交
2076 2077 2078
	obj_request->object_name = memcpy(name, object_name, size);
	obj_request->offset = offset;
	obj_request->length = length;
2079
	obj_request->flags = 0;
A
Alex Elder 已提交
2080 2081 2082
	obj_request->which = BAD_WHICH;
	obj_request->type = type;
	INIT_LIST_HEAD(&obj_request->links);
2083
	init_completion(&obj_request->completion);
A
Alex Elder 已提交
2084 2085
	kref_init(&obj_request->kref);

A
Alex Elder 已提交
2086 2087 2088
	dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name,
		offset, length, (int)type, obj_request);

A
Alex Elder 已提交
2089 2090 2091 2092 2093 2094 2095 2096 2097
	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
2098 2099
	dout("%s: obj %p\n", __func__, obj_request);

A
Alex Elder 已提交
2100 2101 2102 2103 2104 2105 2106 2107
	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) {
2108 2109
	case OBJ_REQUEST_NODATA:
		break;		/* Nothing to do */
A
Alex Elder 已提交
2110 2111 2112 2113
	case OBJ_REQUEST_BIO:
		if (obj_request->bio_list)
			bio_chain_put(obj_request->bio_list);
		break;
2114
	case OBJ_REQUEST_PAGES:
2115 2116 2117
		/* img_data requests don't own their page array */
		if (obj_request->pages &&
		    !obj_request_img_data_test(obj_request))
2118 2119 2120
			ceph_release_page_vector(obj_request->pages,
						obj_request->page_count);
		break;
A
Alex Elder 已提交
2121 2122
	}

2123
	kfree(obj_request->object_name);
2124 2125
	obj_request->object_name = NULL;
	kmem_cache_free(rbd_obj_request_cache, obj_request);
A
Alex Elder 已提交
2126 2127
}

A
Alex Elder 已提交
2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138
/* 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;
}

2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160
/*
 * 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
2161
		rbd_warn(rbd_dev, "parent reference underflow");
2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173
}

/*
 * 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)
{
2174
	int counter = 0;
2175 2176 2177 2178

	if (!rbd_dev->parent_spec)
		return false;

2179 2180 2181 2182
	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);
2183 2184

	if (counter < 0)
2185
		rbd_warn(rbd_dev, "parent reference overflow");
2186

2187
	return counter > 0;
2188 2189
}

A
Alex Elder 已提交
2190 2191 2192 2193 2194
/*
 * 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
2195 2196
static struct rbd_img_request *rbd_img_request_create(
					struct rbd_device *rbd_dev,
A
Alex Elder 已提交
2197
					u64 offset, u64 length,
G
Guangliang Zhao 已提交
2198
					enum obj_operation_type op_type,
2199
					struct ceph_snap_context *snapc)
A
Alex Elder 已提交
2200 2201 2202
{
	struct rbd_img_request *img_request;

2203
	img_request = kmem_cache_alloc(rbd_img_request_cache, GFP_NOIO);
A
Alex Elder 已提交
2204 2205 2206 2207 2208 2209 2210
	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
2211
	img_request->flags = 0;
2212 2213 2214 2215
	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
2216
		img_request_write_set(img_request);
2217
		img_request->snapc = snapc;
A
Alex Elder 已提交
2218
	} else {
A
Alex Elder 已提交
2219
		img_request->snap_id = rbd_dev->spec->snap_id;
A
Alex Elder 已提交
2220
	}
2221
	if (rbd_dev_parent_get(rbd_dev))
2222
		img_request_layered_set(img_request);
A
Alex Elder 已提交
2223 2224 2225
	spin_lock_init(&img_request->completion_lock);
	img_request->next_completion = 0;
	img_request->callback = NULL;
2226
	img_request->result = 0;
A
Alex Elder 已提交
2227 2228 2229 2230
	img_request->obj_request_count = 0;
	INIT_LIST_HEAD(&img_request->obj_requests);
	kref_init(&img_request->kref);

A
Alex Elder 已提交
2231
	dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev,
G
Guangliang Zhao 已提交
2232
		obj_op_name(op_type), offset, length, img_request);
A
Alex Elder 已提交
2233

A
Alex Elder 已提交
2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244
	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
2245 2246
	dout("%s: img %p\n", __func__, img_request);

A
Alex Elder 已提交
2247 2248
	for_each_obj_request_safe(img_request, obj_request, next_obj_request)
		rbd_img_obj_request_del(img_request, obj_request);
2249
	rbd_assert(img_request->obj_request_count == 0);
A
Alex Elder 已提交
2250

2251 2252 2253 2254 2255
	if (img_request_layered_test(img_request)) {
		img_request_layered_clear(img_request);
		rbd_dev_parent_put(img_request->rbd_dev);
	}

2256 2257
	if (img_request_write_test(img_request) ||
		img_request_discard_test(img_request))
2258
		ceph_put_snap_context(img_request->snapc);
A
Alex Elder 已提交
2259

2260
	kmem_cache_free(rbd_img_request_cache, img_request);
A
Alex Elder 已提交
2261 2262
}

2263 2264 2265 2266 2267 2268 2269 2270 2271 2272
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;

2273
	parent_request = rbd_img_request_create(rbd_dev->parent, img_offset,
G
Guangliang Zhao 已提交
2274
						length, OBJ_OP_READ, NULL);
2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299
	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);
}

2300 2301
static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request)
{
2302
	struct rbd_img_request *img_request;
2303 2304
	unsigned int xferred;
	int result;
A
Alex Elder 已提交
2305
	bool more;
2306

2307 2308 2309
	rbd_assert(obj_request_img_data_test(obj_request));
	img_request = obj_request->img_request;

2310 2311 2312 2313 2314
	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
2315 2316
		enum obj_operation_type op_type;

2317 2318 2319 2320 2321 2322
		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;
2323

2324
		rbd_warn(rbd_dev, "%s %llx at %llx (%llx)",
G
Guangliang Zhao 已提交
2325 2326
			obj_op_name(op_type), obj_request->length,
			obj_request->img_offset, obj_request->offset);
2327
		rbd_warn(rbd_dev, "  result %d xferred %x",
2328 2329 2330
			result, xferred);
		if (!img_request->result)
			img_request->result = result;
2331 2332 2333 2334 2335
		/*
		 * Need to end I/O on the entire obj_request worth of
		 * bytes in case of error.
		 */
		xferred = obj_request->length;
2336 2337
	}

A
Alex Elder 已提交
2338 2339 2340 2341 2342
	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
2343 2344 2345 2346

		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
2347 2348 2349
	}

	return more;
2350 2351
}

2352 2353 2354 2355 2356 2357
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;

2358
	rbd_assert(obj_request_img_data_test(obj_request));
2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376
	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;
2377
		more = rbd_img_obj_end_request(obj_request);
2378 2379 2380 2381 2382 2383 2384
		which++;
	}

	rbd_assert(more ^ (which == img_request->obj_request_count));
	img_request->next_completion = which;
out:
	spin_unlock_irq(&img_request->completion_lock);
2385
	rbd_img_request_put(img_request);
2386 2387 2388 2389 2390

	if (!more)
		rbd_img_request_complete(img_request);
}

2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409
/*
 * 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) {
2410 2411 2412
		if (!offset && length == object_size &&
		    (!img_request_layered_test(img_request) ||
		     !obj_request_overlaps_parent(obj_request))) {
2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426
			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) {
2427 2428 2429 2430
		if (!offset && length == object_size)
			opcode = CEPH_OSD_OP_WRITEFULL;
		else
			opcode = CEPH_OSD_OP_WRITE;
2431 2432 2433 2434 2435 2436 2437
		osd_req_op_alloc_hint_init(osd_request, num_ops,
					object_size, object_size);
		num_ops++;
	} else {
		opcode = CEPH_OSD_OP_READ;
	}

2438
	if (opcode == CEPH_OSD_OP_DELETE)
2439
		osd_req_op_init(osd_request, num_ops, opcode, 0);
2440 2441 2442 2443
	else
		osd_req_op_extent_init(osd_request, num_ops, opcode,
				       offset, length, 0, 0);

2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458
	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);
}

2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469
/*
 * 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
2470 2471 2472 2473
{
	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
2474
	struct bio *bio_list = NULL;
2475
	unsigned int bio_offset = 0;
J
Jingoo Han 已提交
2476
	struct page **pages = NULL;
G
Guangliang Zhao 已提交
2477
	enum obj_operation_type op_type;
2478
	u64 img_offset;
A
Alex Elder 已提交
2479 2480
	u64 resid;

2481 2482
	dout("%s: img %p type %d data_desc %p\n", __func__, img_request,
		(int)type, data_desc);
A
Alex Elder 已提交
2483

2484
	img_offset = img_request->offset;
A
Alex Elder 已提交
2485
	resid = img_request->length;
A
Alex Elder 已提交
2486
	rbd_assert(resid > 0);
2487
	op_type = rbd_img_request_op_type(img_request);
2488 2489 2490

	if (type == OBJ_REQUEST_BIO) {
		bio_list = data_desc;
2491 2492
		rbd_assert(img_offset ==
			   bio_list->bi_iter.bi_sector << SECTOR_SHIFT);
2493
	} else if (type == OBJ_REQUEST_PAGES) {
2494 2495 2496
		pages = data_desc;
	}

A
Alex Elder 已提交
2497
	while (resid) {
2498
		struct ceph_osd_request *osd_req;
A
Alex Elder 已提交
2499 2500 2501 2502
		const char *object_name;
		u64 offset;
		u64 length;

2503
		object_name = rbd_segment_name(rbd_dev, img_offset);
A
Alex Elder 已提交
2504 2505
		if (!object_name)
			goto out_unwind;
2506 2507
		offset = rbd_segment_offset(rbd_dev, img_offset);
		length = rbd_segment_length(rbd_dev, img_offset, resid);
A
Alex Elder 已提交
2508
		obj_request = rbd_obj_request_create(object_name,
2509
						offset, length, type);
2510 2511
		/* object request has its own copy of the object name */
		rbd_segment_name_free(object_name);
A
Alex Elder 已提交
2512 2513
		if (!obj_request)
			goto out_unwind;
2514

2515 2516 2517 2518 2519
		/*
		 * 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
2520

2521 2522 2523 2524 2525 2526 2527 2528 2529
		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,
2530
								GFP_NOIO);
2531
			if (!obj_request->bio_list)
2532
				goto out_unwind;
2533
		} else if (type == OBJ_REQUEST_PAGES) {
2534 2535 2536 2537 2538 2539 2540 2541 2542
			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
2543

G
Guangliang Zhao 已提交
2544 2545 2546
		osd_req = rbd_osd_req_create(rbd_dev, op_type,
					(op_type == OBJ_OP_WRITE) ? 2 : 1,
					obj_request);
2547
		if (!osd_req)
2548
			goto out_unwind;
2549

2550
		obj_request->osd_req = osd_req;
2551
		obj_request->callback = rbd_img_obj_callback;
2552
		obj_request->img_offset = img_offset;
2553

2554
		rbd_img_obj_request_fill(obj_request, osd_req, op_type, 0);
A
Alex Elder 已提交
2555

2556
		img_offset += length;
A
Alex Elder 已提交
2557 2558 2559 2560 2561 2562 2563
		resid -= length;
	}

	return 0;

out_unwind:
	for_each_obj_request_safe(img_request, obj_request, next_obj_request)
2564
		rbd_img_obj_request_del(img_request, obj_request);
A
Alex Elder 已提交
2565 2566 2567 2568

	return -ENOMEM;
}

2569
static void
I
Ilya Dryomov 已提交
2570
rbd_osd_copyup_callback(struct rbd_obj_request *obj_request)
2571 2572 2573
{
	struct rbd_img_request *img_request;
	struct rbd_device *rbd_dev;
2574
	struct page **pages;
2575 2576
	u32 page_count;

I
Ilya Dryomov 已提交
2577 2578
	dout("%s: obj %p\n", __func__, obj_request);

2579 2580
	rbd_assert(obj_request->type == OBJ_REQUEST_BIO ||
		obj_request->type == OBJ_REQUEST_NODATA);
2581 2582 2583 2584 2585 2586 2587
	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);

2588 2589
	pages = obj_request->copyup_pages;
	rbd_assert(pages != NULL);
2590
	obj_request->copyup_pages = NULL;
2591 2592 2593 2594
	page_count = obj_request->copyup_page_count;
	rbd_assert(page_count);
	obj_request->copyup_page_count = 0;
	ceph_release_page_vector(pages, page_count);
2595 2596 2597 2598 2599 2600 2601 2602 2603 2604

	/*
	 * 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
2605
	obj_request_done_set(obj_request);
2606 2607
}

2608 2609 2610 2611
static void
rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request)
{
	struct rbd_obj_request *orig_request;
2612 2613
	struct ceph_osd_request *osd_req;
	struct rbd_device *rbd_dev;
2614
	struct page **pages;
2615
	enum obj_operation_type op_type;
2616
	u32 page_count;
2617
	int img_result;
2618
	u64 parent_length;
2619 2620 2621 2622 2623 2624 2625 2626

	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;
2627 2628 2629
	page_count = img_request->copyup_page_count;
	rbd_assert(page_count);
	img_request->copyup_page_count = 0;
2630 2631 2632

	orig_request = img_request->obj_request;
	rbd_assert(orig_request != NULL);
2633
	rbd_assert(obj_request_type_valid(orig_request->type));
2634
	img_result = img_request->result;
2635
	parent_length = img_request->length;
2636
	rbd_assert(img_result || parent_length == img_request->xferred);
2637
	rbd_img_request_put(img_request);
2638

2639 2640
	rbd_assert(orig_request->img_request);
	rbd_dev = orig_request->img_request->rbd_dev;
2641 2642
	rbd_assert(rbd_dev);

2643 2644 2645 2646 2647 2648 2649
	/*
	 * 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);
2650 2651
		rbd_obj_request_submit(orig_request);
		return;
2652
	}
2653

2654
	if (img_result)
2655 2656
		goto out_err;

2657 2658
	/*
	 * The original osd request is of no use to use any more.
2659
	 * We need a new one that can hold the three ops in a copyup
2660 2661 2662
	 * request.  Allocate the new copyup osd request for the
	 * original request, and release the old one.
	 */
2663
	img_result = -ENOMEM;
2664 2665 2666
	osd_req = rbd_osd_req_create_copyup(orig_request);
	if (!osd_req)
		goto out_err;
2667
	rbd_osd_req_destroy(orig_request->osd_req);
2668 2669
	orig_request->osd_req = osd_req;
	orig_request->copyup_pages = pages;
2670
	orig_request->copyup_page_count = page_count;
2671

2672
	/* Initialize the copyup op */
2673

2674
	osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup");
2675
	osd_req_op_cls_request_data_pages(osd_req, 0, pages, parent_length, 0,
2676
						false, false);
2677

2678
	/* Add the other op(s) */
2679

2680 2681
	op_type = rbd_img_request_op_type(orig_request->img_request);
	rbd_img_obj_request_fill(orig_request, osd_req, op_type, 1);
2682 2683 2684

	/* All set, send it off. */

2685 2686
	rbd_obj_request_submit(orig_request);
	return;
2687 2688

out_err:
2689
	ceph_release_page_vector(pages, page_count);
2690
	rbd_obj_request_error(orig_request, img_result);
2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703
}

/*
 * 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.
 *
2704 2705
 * If an error occurs, it is recorded as the result of the original
 * object request in rbd_img_obj_exists_callback().
2706 2707 2708
 */
static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request)
{
2709
	struct rbd_device *rbd_dev = obj_request->img_request->rbd_dev;
2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723
	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
2724
	length = rbd_obj_bytes(&rbd_dev->header);
2725

A
Alex Elder 已提交
2726 2727 2728 2729 2730 2731 2732 2733 2734 2735
	/*
	 * 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;
	}

2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748
	/*
	 * 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;
2749 2750
	parent_request = rbd_parent_request_create(obj_request,
						img_offset, length);
2751 2752 2753 2754 2755 2756
	if (!parent_request)
		goto out_err;

	result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages);
	if (result)
		goto out_err;
2757

2758
	parent_request->copyup_pages = pages;
2759
	parent_request->copyup_page_count = page_count;
2760
	parent_request->callback = rbd_img_obj_parent_read_full_callback;
2761

2762 2763 2764 2765 2766
	result = rbd_img_request_submit(parent_request);
	if (!result)
		return 0;

	parent_request->copyup_pages = NULL;
2767
	parent_request->copyup_page_count = 0;
2768 2769 2770 2771 2772 2773 2774 2775 2776 2777
	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;
}

2778 2779 2780
static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request)
{
	struct rbd_obj_request *orig_request;
2781
	struct rbd_device *rbd_dev;
2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792
	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;
2793
	rbd_obj_request_put(orig_request);
2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804
	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);

2805 2806
	/*
	 * If the overlap has become 0 (most likely because the
2807 2808
	 * image has been flattened) we need to re-submit the
	 * original request.
2809 2810 2811
	 */
	rbd_dev = orig_request->img_request->rbd_dev;
	if (!rbd_dev->parent_overlap) {
2812 2813
		rbd_obj_request_submit(orig_request);
		return;
2814
	}
2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825

	/*
	 * 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);
2826 2827
	} else {
		goto fail_orig_request;
2828 2829 2830 2831 2832 2833
	}

	/*
	 * Resubmit the original request now that we have recorded
	 * whether the target object exists.
	 */
2834 2835 2836 2837 2838 2839 2840
	result = rbd_img_obj_request_submit(orig_request);
	if (result)
		goto fail_orig_request;

	return;

fail_orig_request:
2841
	rbd_obj_request_error(orig_request, result);
2842 2843 2844 2845
}

static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request)
{
2846
	struct rbd_device *rbd_dev = obj_request->img_request->rbd_dev;
2847
	struct rbd_obj_request *stat_request;
2848
	struct page **pages;
2849 2850 2851 2852
	u32 page_count;
	size_t size;
	int ret;

2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864
	stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0,
					      OBJ_REQUEST_PAGES);
	if (!stat_request)
		return -ENOMEM;

	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;
	}

2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875
	/*
	 * 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);
2876 2877 2878 2879
	if (IS_ERR(pages)) {
		ret = PTR_ERR(pages);
		goto fail_stat_request;
	}
2880

2881 2882 2883
	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);
2884 2885 2886 2887 2888 2889 2890

	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;

2891 2892
	rbd_obj_request_submit(stat_request);
	return 0;
2893

2894 2895
fail_stat_request:
	rbd_obj_request_put(stat_request);
2896 2897 2898
	return ret;
}

2899
static bool img_obj_request_simple(struct rbd_obj_request *obj_request)
2900
{
2901 2902
	struct rbd_img_request *img_request = obj_request->img_request;
	struct rbd_device *rbd_dev = img_request->rbd_dev;
2903

2904
	/* Reads */
2905 2906
	if (!img_request_write_test(img_request) &&
	    !img_request_discard_test(img_request))
2907 2908 2909 2910 2911 2912
		return true;

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

2913
	/*
2914 2915
	 * Layered writes outside of the parent overlap range don't
	 * share any data with the parent.
2916
	 */
2917 2918
	if (!obj_request_overlaps_parent(obj_request))
		return true;
2919

2920 2921 2922 2923 2924 2925 2926 2927
	/*
	 * 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;

2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940
	/*
	 * 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)
{
2941 2942 2943
	rbd_assert(obj_request_img_data_test(obj_request));
	rbd_assert(obj_request_type_valid(obj_request->type));
	rbd_assert(obj_request->img_request);
2944

2945
	if (img_obj_request_simple(obj_request)) {
2946 2947
		rbd_obj_request_submit(obj_request);
		return 0;
2948 2949 2950
	}

	/*
2951 2952 2953 2954
	 * 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.
2955
	 */
2956
	if (obj_request_known_test(obj_request))
2957 2958 2959
		return rbd_img_obj_parent_read_full(obj_request);

	/* We don't know whether the target exists.  Go find out. */
2960 2961 2962 2963

	return rbd_img_obj_exists_submit(obj_request);
}

A
Alex Elder 已提交
2964 2965 2966
static int rbd_img_request_submit(struct rbd_img_request *img_request)
{
	struct rbd_obj_request *obj_request;
2967
	struct rbd_obj_request *next_obj_request;
2968
	int ret = 0;
A
Alex Elder 已提交
2969

A
Alex Elder 已提交
2970
	dout("%s: img %p\n", __func__, img_request);
A
Alex Elder 已提交
2971

2972 2973
	rbd_img_request_get(img_request);
	for_each_obj_request_safe(img_request, obj_request, next_obj_request) {
2974
		ret = rbd_img_obj_request_submit(obj_request);
A
Alex Elder 已提交
2975
		if (ret)
2976
			goto out_put_ireq;
A
Alex Elder 已提交
2977 2978
	}

2979 2980 2981
out_put_ireq:
	rbd_img_request_put(img_request);
	return ret;
A
Alex Elder 已提交
2982
}
A
Alex Elder 已提交
2983 2984 2985 2986

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
2987 2988
	struct rbd_device *rbd_dev;
	u64 obj_end;
2989 2990
	u64 img_xferred;
	int img_result;
A
Alex Elder 已提交
2991 2992 2993

	rbd_assert(img_request_child_test(img_request));

2994 2995
	/* First get what we need from the image request and release it */

A
Alex Elder 已提交
2996
	obj_request = img_request->obj_request;
2997 2998 2999 3000 3001 3002 3003 3004 3005
	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
3006 3007
	rbd_assert(obj_request);
	rbd_assert(obj_request->img_request);
3008 3009
	rbd_dev = obj_request->img_request->rbd_dev;
	if (!rbd_dev->parent_overlap) {
3010 3011
		rbd_obj_request_submit(obj_request);
		return;
3012
	}
A
Alex Elder 已提交
3013

3014
	obj_request->result = img_result;
A
Alex Elder 已提交
3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032
	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
3033

3034
		obj_request->xferred = min(img_xferred, xferred);
A
Alex Elder 已提交
3035
	} else {
3036
		obj_request->xferred = img_xferred;
A
Alex Elder 已提交
3037 3038
	}
out:
A
Alex Elder 已提交
3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050
	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);
3051
	rbd_assert(obj_request_type_valid(obj_request->type));
A
Alex Elder 已提交
3052 3053

	/* rbd_read_finish(obj_request, obj_request->length); */
3054
	img_request = rbd_parent_request_create(obj_request,
A
Alex Elder 已提交
3055
						obj_request->img_offset,
3056
						obj_request->length);
A
Alex Elder 已提交
3057 3058 3059 3060
	result = -ENOMEM;
	if (!img_request)
		goto out_err;

3061 3062 3063 3064 3065 3066
	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
3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082
	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
3083

3084
static const struct rbd_client_id rbd_empty_cid;
A
Alex Elder 已提交
3085

3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125
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
3126
{
3127
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3128 3129
	struct rbd_client_id cid = rbd_get_cid(rbd_dev);
	char cookie[32];
3130
	int ret;
A
Alex Elder 已提交
3131

3132
	WARN_ON(__rbd_is_lock_owner(rbd_dev));
3133

3134 3135 3136 3137
	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);
3138
	if (ret)
3139
		return ret;
A
Alex Elder 已提交
3140

3141 3142 3143 3144
	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
3145 3146
}

3147 3148 3149 3150
/*
 * lock_rwsem must be held for write
 */
static int rbd_unlock(struct rbd_device *rbd_dev)
3151
{
3152
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3153
	char cookie[32];
3154 3155
	int ret;

3156
	WARN_ON(!__rbd_is_lock_owner(rbd_dev));
3157

3158
	rbd_dev->lock_state = RBD_LOCK_STATE_UNLOCKED;
3159

3160 3161 3162 3163 3164 3165
	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;
3166 3167
	}

3168 3169 3170
	rbd_set_owner_cid(rbd_dev, &rbd_empty_cid);
	queue_work(rbd_dev->task_wq, &rbd_dev->released_lock_work);
	return 0;
3171 3172
}

3173 3174 3175 3176
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)
3177 3178
{
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3179 3180 3181 3182
	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;
3183

3184
	dout("%s rbd_dev %p notify_op %d\n", __func__, rbd_dev, notify_op);
3185

3186 3187 3188 3189 3190
	/* 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);
3191

3192 3193 3194
	return ceph_osdc_notify(osdc, &rbd_dev->header_oid,
				&rbd_dev->header_oloc, buf, buf_size,
				RBD_NOTIFY_TIMEOUT, preply_pages, preply_len);
3195 3196
}

3197 3198
static void rbd_notify_op_lock(struct rbd_device *rbd_dev,
			       enum rbd_notify_op notify_op)
3199
{
3200 3201
	struct page **reply_pages;
	size_t reply_len;
3202

3203 3204 3205
	__rbd_notify_op_lock(rbd_dev, notify_op, &reply_pages, &reply_len);
	ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len));
}
3206

3207 3208 3209 3210
static void rbd_notify_acquired_lock(struct work_struct *work)
{
	struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
						  acquired_lock_work);
3211

3212
	rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_ACQUIRED_LOCK);
3213 3214
}

3215
static void rbd_notify_released_lock(struct work_struct *work)
3216
{
3217 3218
	struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
						  released_lock_work);
I
Ilya Dryomov 已提交
3219

3220
	rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_RELEASED_LOCK);
3221 3222
}

3223
static int rbd_request_lock(struct rbd_device *rbd_dev)
3224
{
3225 3226 3227
	struct page **reply_pages;
	size_t reply_len;
	bool lock_owner_responded = false;
3228 3229
	int ret;

3230
	dout("%s rbd_dev %p\n", __func__, rbd_dev);
3231

3232 3233 3234 3235
	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);
3236
		goto out;
3237
	}
3238

3239 3240 3241 3242
	if (reply_len > 0 && reply_len <= PAGE_SIZE) {
		void *p = page_address(reply_pages[0]);
		void *const end = p + reply_len;
		u32 n;
3243

3244 3245 3246 3247
		ceph_decode_32_safe(&p, end, n, e_inval); /* num_acks */
		while (n--) {
			u8 struct_v;
			u32 len;
3248

3249 3250
			ceph_decode_need(&p, end, 8 + 8, e_inval);
			p += 8 + 8; /* skip gid and cookie */
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 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527
			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);
3528
	/*
3529
	 * Ensure that all in-flight IO is flushed.
3530
	 *
3531 3532
	 * FIXME: ceph_osdc_sync() flushes the entire OSD client, which
	 * may be shared with other devices.
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 3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 3650 3651 3652 3653 3654 3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673 3674 3675 3676 3677 3678 3679 3680
	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
3681

3682 3683
	ret = ceph_osdc_notify_ack(osdc, &rbd_dev->header_oid,
				   &rbd_dev->header_oloc, notify_id, cookie,
3684
				   buf, buf_size);
3685
	if (ret)
3686 3687 3688 3689 3690 3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708
		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;
3709
	u8 struct_v = 0;
3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720 3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741 3742 3743 3744 3745 3746 3747 3748 3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766 3767
	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
3768 3769
}

3770 3771
static void __rbd_unregister_watch(struct rbd_device *rbd_dev);

3772
static void rbd_watch_errcb(void *arg, u64 cookie, int err)
3773
{
3774
	struct rbd_device *rbd_dev = arg;
3775

3776
	rbd_warn(rbd_dev, "encountered watch error: %d", err);
3777

3778 3779 3780 3781
	down_write(&rbd_dev->lock_rwsem);
	rbd_set_owner_cid(rbd_dev, &rbd_empty_cid);
	up_write(&rbd_dev->lock_rwsem);

3782 3783 3784 3785
	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;
3786

3787
		queue_delayed_work(rbd_dev->task_wq, &rbd_dev->watch_dwork, 0);
3788
	}
3789
	mutex_unlock(&rbd_dev->watch_mutex);
3790 3791
}

3792
/*
3793
 * watch_mutex must be locked
3794
 */
3795
static int __rbd_register_watch(struct rbd_device *rbd_dev)
3796 3797
{
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3798
	struct ceph_osd_linger_request *handle;
3799

3800
	rbd_assert(!rbd_dev->watch_handle);
3801
	dout("%s rbd_dev %p\n", __func__, rbd_dev);
3802

3803 3804 3805 3806 3807
	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);
3808

3809
	rbd_dev->watch_handle = handle;
3810 3811 3812
	return 0;
}

3813 3814 3815 3816
/*
 * watch_mutex must be locked
 */
static void __rbd_unregister_watch(struct rbd_device *rbd_dev)
3817
{
3818 3819
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
	int ret;
3820

3821 3822
	rbd_assert(rbd_dev->watch_handle);
	dout("%s rbd_dev %p\n", __func__, rbd_dev);
3823

3824 3825 3826
	ret = ceph_osdc_unwatch(osdc, rbd_dev->watch_handle);
	if (ret)
		rbd_warn(rbd_dev, "failed to unwatch: %d", ret);
3827

3828
	rbd_dev->watch_handle = NULL;
3829 3830
}

3831 3832 3833 3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844 3845 3846 3847 3848 3849
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)
3850
{
3851 3852 3853
	dout("%s rbd_dev %p\n", __func__, rbd_dev);

	cancel_delayed_work_sync(&rbd_dev->watch_dwork);
3854 3855 3856 3857
	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);
3858 3859 3860 3861
}

static void rbd_unregister_watch(struct rbd_device *rbd_dev)
{
3862
	WARN_ON(waitqueue_active(&rbd_dev->lock_waitq));
3863 3864 3865 3866 3867 3868 3869
	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
3870 3871

	ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc);
3872 3873
}

3874 3875 3876 3877
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);
3878
	bool was_lock_owner = false;
3879
	bool need_to_wake = false;
3880 3881 3882 3883
	int ret;

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

3884 3885 3886 3887
	down_write(&rbd_dev->lock_rwsem);
	if (rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED)
		was_lock_owner = rbd_release_lock(rbd_dev);

3888
	mutex_lock(&rbd_dev->watch_mutex);
3889 3890 3891 3892
	if (rbd_dev->watch_state != RBD_WATCH_STATE_ERROR) {
		mutex_unlock(&rbd_dev->watch_mutex);
		goto out;
	}
3893 3894 3895 3896

	ret = __rbd_register_watch(rbd_dev);
	if (ret) {
		rbd_warn(rbd_dev, "failed to reregister watch: %d", ret);
3897
		if (ret == -EBLACKLISTED || ret == -ENOENT) {
3898 3899 3900
			set_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags);
			need_to_wake = true;
		} else {
3901 3902 3903
			queue_delayed_work(rbd_dev->task_wq,
					   &rbd_dev->watch_dwork,
					   RBD_RETRY_DELAY);
3904 3905 3906
		}
		mutex_unlock(&rbd_dev->watch_mutex);
		goto out;
3907 3908
	}

3909
	need_to_wake = true;
3910 3911 3912 3913 3914 3915 3916 3917
	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);

3918 3919 3920 3921 3922 3923 3924
	if (was_lock_owner) {
		ret = rbd_try_lock(rbd_dev);
		if (ret)
			rbd_warn(rbd_dev, "reregisteration lock failed: %d",
				 ret);
	}

3925
out:
3926
	up_write(&rbd_dev->lock_rwsem);
3927 3928
	if (need_to_wake)
		wake_requests(rbd_dev, true);
3929 3930
}

3931
/*
3932 3933
 * Synchronous osd object method call.  Returns the number of bytes
 * returned in the outbound buffer, or a negative error code.
3934 3935
 */
static int rbd_obj_method_sync(struct rbd_device *rbd_dev,
3936 3937
			     struct ceph_object_id *oid,
			     struct ceph_object_locator *oloc,
3938
			     const char *method_name,
3939
			     const void *outbound,
3940
			     size_t outbound_size,
3941
			     void *inbound,
3942
			     size_t inbound_size)
3943
{
3944 3945 3946
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
	struct page *req_page = NULL;
	struct page *reply_page;
3947 3948 3949
	int ret;

	/*
3950 3951 3952 3953 3954
	 * 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.
3955
	 */
3956 3957 3958
	if (outbound) {
		if (outbound_size > PAGE_SIZE)
			return -E2BIG;
3959

3960 3961 3962
		req_page = alloc_page(GFP_KERNEL);
		if (!req_page)
			return -ENOMEM;
3963

3964
		memcpy(page_address(req_page), outbound, outbound_size);
3965
	}
A
Alex Elder 已提交
3966

3967 3968 3969 3970 3971 3972
	reply_page = alloc_page(GFP_KERNEL);
	if (!reply_page) {
		if (req_page)
			__free_page(req_page);
		return -ENOMEM;
	}
3973

3974 3975 3976 3977 3978 3979 3980
	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;
	}
3981

3982 3983 3984
	if (req_page)
		__free_page(req_page);
	__free_page(reply_page);
3985 3986 3987
	return ret;
}

3988 3989 3990 3991 3992 3993 3994 3995 3996 3997 3998 3999 4000 4001 4002 4003 4004 4005 4006
/*
 * 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);
4007 4008 4009
	} while (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED &&
		 !test_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags));

4010 4011 4012
	finish_wait(&rbd_dev->lock_waitq, &wait);
}

C
Christoph Hellwig 已提交
4013
static void rbd_queue_workfn(struct work_struct *work)
A
Alex Elder 已提交
4014
{
C
Christoph Hellwig 已提交
4015 4016
	struct request *rq = blk_mq_rq_from_pdu(work);
	struct rbd_device *rbd_dev = rq->q->queuedata;
I
Ilya Dryomov 已提交
4017
	struct rbd_img_request *img_request;
4018
	struct ceph_snap_context *snapc = NULL;
I
Ilya Dryomov 已提交
4019 4020
	u64 offset = (u64)blk_rq_pos(rq) << SECTOR_SHIFT;
	u64 length = blk_rq_bytes(rq);
G
Guangliang Zhao 已提交
4021
	enum obj_operation_type op_type;
4022
	u64 mapping_size;
I
Ilya Dryomov 已提交
4023
	bool must_be_locked;
A
Alex Elder 已提交
4024 4025
	int result;

C
Christoph Hellwig 已提交
4026 4027 4028 4029 4030 4031 4032
	if (rq->cmd_type != REQ_TYPE_FS) {
		dout("%s: non-fs request type %d\n", __func__,
			(int) rq->cmd_type);
		result = -EIO;
		goto err;
	}

M
Mike Christie 已提交
4033
	if (req_op(rq) == REQ_OP_DISCARD)
4034
		op_type = OBJ_OP_DISCARD;
M
Mike Christie 已提交
4035
	else if (req_op(rq) == REQ_OP_WRITE)
G
Guangliang Zhao 已提交
4036 4037 4038 4039
		op_type = OBJ_OP_WRITE;
	else
		op_type = OBJ_OP_READ;

I
Ilya Dryomov 已提交
4040
	/* Ignore/skip any zero-length requests */
A
Alex Elder 已提交
4041

I
Ilya Dryomov 已提交
4042 4043 4044 4045 4046
	if (!length) {
		dout("%s: zero-length request\n", __func__);
		result = 0;
		goto err_rq;
	}
A
Alex Elder 已提交
4047

G
Guangliang Zhao 已提交
4048
	/* Only reads are allowed to a read-only device */
I
Ilya Dryomov 已提交
4049

G
Guangliang Zhao 已提交
4050
	if (op_type != OBJ_OP_READ) {
I
Ilya Dryomov 已提交
4051 4052 4053
		if (rbd_dev->mapping.read_only) {
			result = -EROFS;
			goto err_rq;
A
Alex Elder 已提交
4054
		}
I
Ilya Dryomov 已提交
4055 4056
		rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP);
	}
A
Alex Elder 已提交
4057

I
Ilya Dryomov 已提交
4058 4059 4060 4061 4062 4063 4064 4065 4066 4067 4068 4069
	/*
	 * 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
4070

I
Ilya Dryomov 已提交
4071 4072 4073 4074 4075 4076
	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
4077

C
Christoph Hellwig 已提交
4078 4079
	blk_mq_start_request(rq);

4080 4081
	down_read(&rbd_dev->header_rwsem);
	mapping_size = rbd_dev->mapping.size;
G
Guangliang Zhao 已提交
4082
	if (op_type != OBJ_OP_READ) {
4083 4084
		snapc = rbd_dev->header.snapc;
		ceph_get_snap_context(snapc);
4085
		must_be_locked = rbd_is_lock_supported(rbd_dev);
I
Ilya Dryomov 已提交
4086 4087 4088
	} else {
		must_be_locked = rbd_dev->opts->lock_on_read &&
					rbd_is_lock_supported(rbd_dev);
4089 4090 4091 4092
	}
	up_read(&rbd_dev->header_rwsem);

	if (offset + length > mapping_size) {
I
Ilya Dryomov 已提交
4093
		rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)", offset,
4094
			 length, mapping_size);
I
Ilya Dryomov 已提交
4095 4096 4097
		result = -EIO;
		goto err_rq;
	}
A
Alex Elder 已提交
4098

4099 4100
	if (must_be_locked) {
		down_read(&rbd_dev->lock_rwsem);
4101 4102
		if (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED &&
		    !test_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags))
4103
			rbd_wait_state_locked(rbd_dev);
4104 4105 4106 4107 4108 4109 4110

		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;
		}
4111 4112
	}

G
Guangliang Zhao 已提交
4113
	img_request = rbd_img_request_create(rbd_dev, offset, length, op_type,
4114
					     snapc);
I
Ilya Dryomov 已提交
4115 4116
	if (!img_request) {
		result = -ENOMEM;
4117
		goto err_unlock;
I
Ilya Dryomov 已提交
4118 4119
	}
	img_request->rq = rq;
4120
	snapc = NULL; /* img_request consumes a ref */
A
Alex Elder 已提交
4121

4122 4123 4124 4125 4126 4127
	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
4128 4129
	if (result)
		goto err_img_request;
A
Alex Elder 已提交
4130

I
Ilya Dryomov 已提交
4131 4132 4133
	result = rbd_img_request_submit(img_request);
	if (result)
		goto err_img_request;
A
Alex Elder 已提交
4134

4135 4136
	if (must_be_locked)
		up_read(&rbd_dev->lock_rwsem);
I
Ilya Dryomov 已提交
4137
	return;
A
Alex Elder 已提交
4138

I
Ilya Dryomov 已提交
4139 4140
err_img_request:
	rbd_img_request_put(img_request);
4141 4142 4143
err_unlock:
	if (must_be_locked)
		up_read(&rbd_dev->lock_rwsem);
I
Ilya Dryomov 已提交
4144 4145 4146
err_rq:
	if (result)
		rbd_warn(rbd_dev, "%s %llx at %llx result %d",
G
Guangliang Zhao 已提交
4147
			 obj_op_name(op_type), length, offset, result);
4148
	ceph_put_snap_context(snapc);
C
Christoph Hellwig 已提交
4149 4150
err:
	blk_mq_end_request(rq, result);
I
Ilya Dryomov 已提交
4151
}
A
Alex Elder 已提交
4152

C
Christoph Hellwig 已提交
4153 4154
static int rbd_queue_rq(struct blk_mq_hw_ctx *hctx,
		const struct blk_mq_queue_data *bd)
I
Ilya Dryomov 已提交
4155
{
C
Christoph Hellwig 已提交
4156 4157
	struct request *rq = bd->rq;
	struct work_struct *work = blk_mq_rq_to_pdu(rq);
A
Alex Elder 已提交
4158

C
Christoph Hellwig 已提交
4159 4160
	queue_work(rbd_wq, work);
	return BLK_MQ_RQ_QUEUE_OK;
A
Alex Elder 已提交
4161 4162
}

4163 4164 4165 4166 4167 4168 4169
static void rbd_free_disk(struct rbd_device *rbd_dev)
{
	struct gendisk *disk = rbd_dev->disk;

	if (!disk)
		return;

4170 4171
	rbd_dev->disk = NULL;
	if (disk->flags & GENHD_FL_UP) {
4172
		del_gendisk(disk);
4173 4174
		if (disk->queue)
			blk_cleanup_queue(disk->queue);
C
Christoph Hellwig 已提交
4175
		blk_mq_free_tag_set(&rbd_dev->tag_set);
4176
	}
4177 4178 4179
	put_disk(disk);
}

4180
static int rbd_obj_read_sync(struct rbd_device *rbd_dev,
4181 4182 4183
			     struct ceph_object_id *oid,
			     struct ceph_object_locator *oloc,
			     void *buf, int buf_len)
4184 4185

{
4186 4187 4188 4189
	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);
4190 4191
	int ret;

4192 4193 4194
	req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_KERNEL);
	if (!req)
		return -ENOMEM;
4195

4196 4197 4198
	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
4199

4200
	ret = ceph_osdc_alloc_messages(req, GFP_KERNEL);
4201
	if (ret)
4202
		goto out_req;
4203

4204 4205 4206 4207 4208
	pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
	if (IS_ERR(pages)) {
		ret = PTR_ERR(pages);
		goto out_req;
	}
4209

4210 4211 4212 4213 4214 4215 4216 4217
	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);
4218

4219 4220
out_req:
	ceph_osdc_put_request(req);
4221 4222 4223
	return ret;
}

4224
/*
A
Alex Elder 已提交
4225 4226 4227
 * 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.
4228
 */
4229
static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev)
4230
{
4231
	struct rbd_image_header_ondisk *ondisk = NULL;
4232
	u32 snap_count = 0;
4233 4234 4235
	u64 names_size = 0;
	u32 want_count;
	int ret;
4236

A
Alex Elder 已提交
4237
	/*
4238 4239 4240 4241 4242
	 * 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
4243
	 */
4244 4245 4246 4247 4248 4249 4250 4251 4252 4253
	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
4254
			return -ENOMEM;
4255

4256 4257
		ret = rbd_obj_read_sync(rbd_dev, &rbd_dev->header_oid,
					&rbd_dev->header_oloc, ondisk, size);
4258
		if (ret < 0)
A
Alex Elder 已提交
4259
			goto out;
A
Alex Elder 已提交
4260
		if ((size_t)ret < size) {
4261
			ret = -ENXIO;
A
Alex Elder 已提交
4262 4263
			rbd_warn(rbd_dev, "short header read (want %zd got %d)",
				size, ret);
A
Alex Elder 已提交
4264
			goto out;
4265 4266 4267
		}
		if (!rbd_dev_ondisk_valid(ondisk)) {
			ret = -ENXIO;
A
Alex Elder 已提交
4268
			rbd_warn(rbd_dev, "invalid header");
A
Alex Elder 已提交
4269
			goto out;
4270
		}
4271

4272 4273 4274 4275
		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
4276

A
Alex Elder 已提交
4277 4278
	ret = rbd_header_from_disk(rbd_dev, ondisk);
out:
4279 4280 4281
	kfree(ondisk);

	return ret;
4282 4283
}

4284 4285 4286 4287 4288 4289 4290 4291 4292 4293 4294 4295 4296 4297 4298 4299 4300 4301 4302
/*
 * 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);
}

4303 4304 4305 4306 4307
static void rbd_dev_update_size(struct rbd_device *rbd_dev)
{
	sector_t size;

	/*
I
Ilya Dryomov 已提交
4308 4309 4310
	 * 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.
4311
	 */
I
Ilya Dryomov 已提交
4312 4313
	if (test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags) &&
	    !test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) {
4314 4315 4316 4317 4318 4319 4320
		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
4321
static int rbd_dev_refresh(struct rbd_device *rbd_dev)
A
Alex Elder 已提交
4322
{
4323
	u64 mapping_size;
A
Alex Elder 已提交
4324 4325
	int ret;

4326
	down_write(&rbd_dev->header_rwsem);
4327
	mapping_size = rbd_dev->mapping.size;
4328 4329

	ret = rbd_dev_header_info(rbd_dev);
4330
	if (ret)
4331
		goto out;
4332

4333 4334 4335 4336 4337 4338 4339
	/*
	 * 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)
4340
			goto out;
4341 4342
	}

4343
	if (rbd_dev->spec->snap_id == CEPH_NOSNAP) {
4344
		rbd_dev->mapping.size = rbd_dev->header.image_size;
4345 4346 4347 4348
	} else {
		/* validate mapped snapshot's EXISTS flag */
		rbd_exists_validate(rbd_dev);
	}
4349

4350
out:
4351
	up_write(&rbd_dev->header_rwsem);
4352
	if (!ret && mapping_size != rbd_dev->mapping.size)
4353
		rbd_dev_update_size(rbd_dev);
A
Alex Elder 已提交
4354

4355
	return ret;
A
Alex Elder 已提交
4356 4357
}

C
Christoph Hellwig 已提交
4358 4359 4360 4361 4362 4363 4364 4365 4366 4367 4368 4369 4370 4371 4372
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,
};

4373 4374 4375 4376
static int rbd_init_disk(struct rbd_device *rbd_dev)
{
	struct gendisk *disk;
	struct request_queue *q;
A
Alex Elder 已提交
4377
	u64 segment_size;
C
Christoph Hellwig 已提交
4378
	int err;
4379 4380

	/* create gendisk info */
4381 4382 4383
	disk = alloc_disk(single_major ?
			  (1 << RBD_SINGLE_MAJOR_PART_SHIFT) :
			  RBD_MINORS_PER_MAJOR);
4384
	if (!disk)
A
Alex Elder 已提交
4385
		return -ENOMEM;
4386

A
Alex Elder 已提交
4387
	snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
A
Alex Elder 已提交
4388
		 rbd_dev->dev_id);
4389
	disk->major = rbd_dev->major;
4390
	disk->first_minor = rbd_dev->minor;
4391 4392
	if (single_major)
		disk->flags |= GENHD_FL_EXT_DEVT;
4393 4394 4395
	disk->fops = &rbd_bd_ops;
	disk->private_data = rbd_dev;

C
Christoph Hellwig 已提交
4396 4397
	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
4398
	rbd_dev->tag_set.queue_depth = rbd_dev->opts->queue_depth;
C
Christoph Hellwig 已提交
4399
	rbd_dev->tag_set.numa_node = NUMA_NO_NODE;
I
Ilya Dryomov 已提交
4400
	rbd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
C
Christoph Hellwig 已提交
4401 4402 4403 4404 4405
	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)
4406
		goto out_disk;
4407

C
Christoph Hellwig 已提交
4408 4409 4410 4411 4412 4413
	q = blk_mq_init_queue(&rbd_dev->tag_set);
	if (IS_ERR(q)) {
		err = PTR_ERR(q);
		goto out_tag_set;
	}

4414 4415
	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
4416

4417
	/* set io sizes to object size */
A
Alex Elder 已提交
4418 4419
	segment_size = rbd_obj_bytes(&rbd_dev->header);
	blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE);
I
Ilya Dryomov 已提交
4420
	q->limits.max_sectors = queue_max_hw_sectors(q);
I
Ilya Dryomov 已提交
4421
	blk_queue_max_segments(q, segment_size / SECTOR_SIZE);
A
Alex Elder 已提交
4422 4423 4424
	blk_queue_max_segment_size(q, segment_size);
	blk_queue_io_min(q, segment_size);
	blk_queue_io_opt(q, segment_size);
4425

4426 4427 4428 4429
	/* enable the discard support */
	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
	q->limits.discard_granularity = segment_size;
	q->limits.discard_alignment = segment_size;
4430
	blk_queue_max_discard_sectors(q, segment_size / SECTOR_SIZE);
4431
	q->limits.discard_zeroes_data = 1;
4432

4433 4434 4435
	if (!ceph_test_opt(rbd_dev->rbd_client->client, NOCRC))
		q->backing_dev_info.capabilities |= BDI_CAP_STABLE_WRITES;

4436 4437 4438 4439 4440 4441 4442
	disk->queue = q;

	q->queuedata = rbd_dev;

	rbd_dev->disk = disk;

	return 0;
C
Christoph Hellwig 已提交
4443 4444
out_tag_set:
	blk_mq_free_tag_set(&rbd_dev->tag_set);
4445 4446
out_disk:
	put_disk(disk);
C
Christoph Hellwig 已提交
4447
	return err;
4448 4449
}

4450 4451 4452 4453
/*
  sysfs
*/

A
Alex Elder 已提交
4454 4455 4456 4457 4458
static struct rbd_device *dev_to_rbd_dev(struct device *dev)
{
	return container_of(dev, struct rbd_device, dev);
}

4459 4460 4461
static ssize_t rbd_size_show(struct device *dev,
			     struct device_attribute *attr, char *buf)
{
A
Alex Elder 已提交
4462
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4463

A
Alex Elder 已提交
4464 4465
	return sprintf(buf, "%llu\n",
		(unsigned long long)rbd_dev->mapping.size);
4466 4467
}

A
Alex Elder 已提交
4468 4469 4470 4471 4472 4473 4474 4475 4476 4477
/*
 * 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
4478
			(unsigned long long)rbd_dev->mapping.features);
A
Alex Elder 已提交
4479 4480
}

4481 4482 4483
static ssize_t rbd_major_show(struct device *dev,
			      struct device_attribute *attr, char *buf)
{
A
Alex Elder 已提交
4484
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4485

A
Alex Elder 已提交
4486 4487 4488 4489
	if (rbd_dev->major)
		return sprintf(buf, "%d\n", rbd_dev->major);

	return sprintf(buf, "(none)\n");
4490 4491 4492 4493 4494 4495
}

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
4496

4497
	return sprintf(buf, "%d\n", rbd_dev->minor);
4498 4499
}

4500 4501 4502 4503 4504 4505 4506 4507 4508 4509 4510
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));
}

4511 4512
static ssize_t rbd_client_id_show(struct device *dev,
				  struct device_attribute *attr, char *buf)
4513
{
A
Alex Elder 已提交
4514
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4515

4516
	return sprintf(buf, "client%lld\n",
4517
		       ceph_client_gid(rbd_dev->rbd_client->client));
4518 4519
}

4520 4521 4522 4523 4524 4525 4526 4527
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);
}

4528 4529 4530 4531 4532 4533
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);
4534 4535
}

4536 4537
static ssize_t rbd_pool_show(struct device *dev,
			     struct device_attribute *attr, char *buf)
4538
{
A
Alex Elder 已提交
4539
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4540

4541
	return sprintf(buf, "%s\n", rbd_dev->spec->pool_name);
4542 4543
}

4544 4545 4546 4547 4548
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);

4549
	return sprintf(buf, "%llu\n",
A
Alex Elder 已提交
4550
			(unsigned long long) rbd_dev->spec->pool_id);
4551 4552
}

4553 4554 4555
static ssize_t rbd_name_show(struct device *dev,
			     struct device_attribute *attr, char *buf)
{
A
Alex Elder 已提交
4556
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4557

A
Alex Elder 已提交
4558 4559 4560 4561
	if (rbd_dev->spec->image_name)
		return sprintf(buf, "%s\n", rbd_dev->spec->image_name);

	return sprintf(buf, "(unknown)\n");
4562 4563
}

A
Alex Elder 已提交
4564 4565 4566 4567 4568
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);

4569
	return sprintf(buf, "%s\n", rbd_dev->spec->image_id);
A
Alex Elder 已提交
4570 4571
}

A
Alex Elder 已提交
4572 4573 4574 4575
/*
 * Shows the name of the currently-mapped snapshot (or
 * RBD_SNAP_HEAD_NAME for the base image).
 */
4576 4577 4578 4579
static ssize_t rbd_snap_show(struct device *dev,
			     struct device_attribute *attr,
			     char *buf)
{
A
Alex Elder 已提交
4580
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4581

4582
	return sprintf(buf, "%s\n", rbd_dev->spec->snap_name);
4583 4584
}

4585 4586 4587 4588 4589 4590 4591 4592
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);
}

4593
/*
4594 4595 4596
 * 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)".
4597 4598
 */
static ssize_t rbd_parent_show(struct device *dev,
4599 4600
			       struct device_attribute *attr,
			       char *buf)
4601 4602
{
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4603
	ssize_t count = 0;
4604

4605
	if (!rbd_dev->parent)
4606 4607
		return sprintf(buf, "(no parent image)\n");

4608 4609 4610 4611 4612 4613 4614 4615 4616 4617 4618 4619 4620 4621 4622 4623
	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;
4624 4625
}

4626 4627 4628 4629 4630
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
4631
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4632
	int ret;
4633

A
Alex Elder 已提交
4634
	ret = rbd_dev_refresh(rbd_dev);
4635
	if (ret)
4636
		return ret;
4637

4638
	return size;
4639
}
4640

4641
static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
A
Alex Elder 已提交
4642
static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL);
4643
static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
4644
static DEVICE_ATTR(minor, S_IRUGO, rbd_minor_show, NULL);
4645
static DEVICE_ATTR(client_addr, S_IRUGO, rbd_client_addr_show, NULL);
4646
static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
4647
static DEVICE_ATTR(cluster_fsid, S_IRUGO, rbd_cluster_fsid_show, NULL);
4648
static DEVICE_ATTR(config_info, S_IRUSR, rbd_config_info_show, NULL);
4649
static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
4650
static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL);
4651
static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
A
Alex Elder 已提交
4652
static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL);
4653 4654
static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
4655
static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL);
4656
static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL);
4657 4658 4659

static struct attribute *rbd_attrs[] = {
	&dev_attr_size.attr,
A
Alex Elder 已提交
4660
	&dev_attr_features.attr,
4661
	&dev_attr_major.attr,
4662
	&dev_attr_minor.attr,
4663
	&dev_attr_client_addr.attr,
4664
	&dev_attr_client_id.attr,
4665
	&dev_attr_cluster_fsid.attr,
4666
	&dev_attr_config_info.attr,
4667
	&dev_attr_pool.attr,
4668
	&dev_attr_pool_id.attr,
4669
	&dev_attr_name.attr,
A
Alex Elder 已提交
4670
	&dev_attr_image_id.attr,
4671
	&dev_attr_current_snap.attr,
4672
	&dev_attr_snap_id.attr,
4673
	&dev_attr_parent.attr,
4674 4675 4676 4677 4678 4679 4680 4681 4682 4683 4684 4685 4686
	&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
};

4687
static void rbd_dev_release(struct device *dev);
4688 4689 4690 4691

static struct device_type rbd_device_type = {
	.name		= "rbd",
	.groups		= rbd_attr_groups,
4692
	.release	= rbd_dev_release,
4693 4694
};

4695 4696 4697 4698 4699 4700 4701 4702 4703 4704 4705 4706 4707 4708 4709 4710 4711 4712 4713 4714 4715
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;
4716 4717 4718

	spec->pool_id = CEPH_NOPOOL;
	spec->snap_id = CEPH_NOSNAP;
4719 4720 4721 4722 4723 4724 4725 4726 4727 4728 4729 4730 4731 4732 4733 4734
	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);
}

4735
static void rbd_dev_free(struct rbd_device *rbd_dev)
4736
{
4737
	WARN_ON(rbd_dev->watch_state != RBD_WATCH_STATE_UNREGISTERED);
4738
	WARN_ON(rbd_dev->lock_state != RBD_LOCK_STATE_UNLOCKED);
4739

4740
	ceph_oid_destroy(&rbd_dev->header_oid);
4741
	ceph_oloc_destroy(&rbd_dev->header_oloc);
4742
	kfree(rbd_dev->config_info);
4743

4744 4745 4746 4747
	rbd_put_client(rbd_dev->rbd_client);
	rbd_spec_put(rbd_dev->spec);
	kfree(rbd_dev->opts);
	kfree(rbd_dev);
4748 4749 4750 4751 4752 4753 4754 4755 4756 4757 4758 4759 4760
}

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);
4761 4762 4763 4764 4765 4766 4767 4768 4769 4770

	/*
	 * 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);
}

4771 4772
static struct rbd_device *__rbd_dev_create(struct rbd_client *rbdc,
					   struct rbd_spec *spec)
4773 4774 4775
{
	struct rbd_device *rbd_dev;

4776
	rbd_dev = kzalloc(sizeof(*rbd_dev), GFP_KERNEL);
4777 4778 4779 4780 4781 4782 4783
	if (!rbd_dev)
		return NULL;

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

4784
	ceph_oid_init(&rbd_dev->header_oid);
4785
	rbd_dev->header_oloc.pool = spec->pool_id;
4786

4787 4788 4789 4790
	mutex_init(&rbd_dev->watch_mutex);
	rbd_dev->watch_state = RBD_WATCH_STATE_UNREGISTERED;
	INIT_DELAYED_WORK(&rbd_dev->watch_dwork, rbd_reregister_watch);

4791 4792 4793 4794 4795 4796 4797 4798
	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);

4799 4800 4801 4802 4803
	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);

4804
	rbd_dev->rbd_client = rbdc;
4805
	rbd_dev->spec = spec;
4806

4807 4808 4809 4810
	rbd_dev->layout.stripe_unit = 1 << RBD_MAX_OBJ_ORDER;
	rbd_dev->layout.stripe_count = 1;
	rbd_dev->layout.object_size = 1 << RBD_MAX_OBJ_ORDER;
	rbd_dev->layout.pool_id = spec->pool_id;
4811
	RCU_INIT_POINTER(rbd_dev->layout.pool_ns, NULL);
4812

4813 4814 4815 4816 4817 4818 4819 4820 4821 4822 4823 4824 4825 4826 4827 4828 4829 4830 4831 4832 4833 4834 4835 4836 4837 4838 4839 4840 4841 4842
	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;
4843

4844 4845
	/* we have a ref from do_rbd_add() */
	__module_get(THIS_MODULE);
4846

4847
	dout("%s rbd_dev %p dev_id %d\n", __func__, rbd_dev, rbd_dev->dev_id);
4848
	return rbd_dev;
4849 4850 4851 4852 4853 4854

fail_dev_id:
	ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id);
fail_rbd_dev:
	rbd_dev_free(rbd_dev);
	return NULL;
4855 4856 4857 4858
}

static void rbd_dev_destroy(struct rbd_device *rbd_dev)
{
4859 4860
	if (rbd_dev)
		put_device(&rbd_dev->dev);
4861 4862
}

4863 4864 4865 4866 4867 4868 4869 4870 4871 4872 4873 4874 4875 4876 4877
/*
 * 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 };

4878 4879 4880 4881
	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));
4882
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4883 4884
	if (ret < 0)
		return ret;
4885 4886
	if (ret < sizeof (size_buf))
		return -ERANGE;
4887

J
Josh Durgin 已提交
4888
	if (order) {
4889
		*order = size_buf.order;
J
Josh Durgin 已提交
4890 4891
		dout("  order %u", (unsigned int)*order);
	}
4892 4893
	*snap_size = le64_to_cpu(size_buf.size);

J
Josh Durgin 已提交
4894 4895
	dout("  snap_id 0x%016llx snap_size = %llu\n",
		(unsigned long long)snap_id,
4896
		(unsigned long long)*snap_size);
4897 4898 4899 4900 4901 4902 4903 4904 4905 4906 4907

	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);
}

4908 4909 4910 4911 4912 4913 4914 4915 4916 4917
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;

4918 4919 4920
	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);
4921
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4922 4923 4924 4925 4926
	if (ret < 0)
		goto out;

	p = reply_buf;
	rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p,
4927 4928
						p + ret, NULL, GFP_NOIO);
	ret = 0;
4929 4930 4931 4932 4933 4934 4935 4936 4937 4938 4939 4940 4941

	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;
}

4942 4943 4944 4945 4946 4947 4948
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;
4949
	} __attribute__ ((packed)) features_buf = { 0 };
4950
	u64 unsup;
4951 4952
	int ret;

4953 4954 4955 4956
	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));
4957
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4958 4959
	if (ret < 0)
		return ret;
4960 4961
	if (ret < sizeof (features_buf))
		return -ERANGE;
A
Alex Elder 已提交
4962

4963 4964 4965 4966
	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
4967
		return -ENXIO;
4968
	}
A
Alex Elder 已提交
4969

4970 4971 4972
	*snap_features = le64_to_cpu(features_buf.features);

	dout("  snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
4973 4974 4975
		(unsigned long long)snap_id,
		(unsigned long long)*snap_features,
		(unsigned long long)le64_to_cpu(features_buf.incompat));
4976 4977 4978 4979 4980 4981 4982 4983 4984 4985

	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);
}

4986 4987 4988 4989 4990 4991 4992 4993
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
4994
	u64 pool_id;
4995
	char *image_id;
4996
	u64 snap_id;
4997 4998 4999 5000 5001 5002 5003 5004 5005 5006 5007 5008 5009 5010 5011 5012 5013
	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;
	}

5014
	snapid = cpu_to_le64(rbd_dev->spec->snap_id);
5015 5016 5017
	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
				  &rbd_dev->header_oloc, "get_parent",
				  &snapid, sizeof(snapid), reply_buf, size);
5018
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
5019 5020 5021 5022
	if (ret < 0)
		goto out_err;

	p = reply_buf;
5023 5024
	end = reply_buf + ret;
	ret = -ERANGE;
A
Alex Elder 已提交
5025
	ceph_decode_64_safe(&p, end, pool_id, out_err);
5026 5027 5028 5029 5030 5031 5032 5033 5034 5035 5036 5037 5038 5039 5040 5041 5042
	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);
		}

5043
		goto out;	/* No parent?  No problem. */
5044
	}
5045

5046 5047 5048
	/* The ceph file layout needs to fit pool id in 32 bits */

	ret = -EIO;
A
Alex Elder 已提交
5049
	if (pool_id > (u64)U32_MAX) {
5050
		rbd_warn(NULL, "parent pool id too large (%llu > %u)",
A
Alex Elder 已提交
5051
			(unsigned long long)pool_id, U32_MAX);
5052
		goto out_err;
A
Alex Elder 已提交
5053
	}
5054

A
Alex Elder 已提交
5055
	image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
5056 5057 5058 5059
	if (IS_ERR(image_id)) {
		ret = PTR_ERR(image_id);
		goto out_err;
	}
5060
	ceph_decode_64_safe(&p, end, snap_id, out_err);
5061 5062
	ceph_decode_64_safe(&p, end, overlap, out_err);

5063 5064 5065 5066 5067 5068 5069 5070 5071
	/*
	 * 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
5072 5073
		rbd_dev->parent_spec = parent_spec;
		parent_spec = NULL;	/* rbd_dev now owns this */
5074 5075
	} else {
		kfree(image_id);
5076 5077 5078
	}

	/*
5079 5080
	 * We always update the parent overlap.  If it's zero we issue
	 * a warning, as we will proceed as if there was no parent.
5081 5082 5083
	 */
	if (!overlap) {
		if (parent_spec) {
5084 5085 5086 5087
			/* refresh, careful to warn just once */
			if (rbd_dev->parent_overlap)
				rbd_warn(rbd_dev,
				    "clone now standalone (overlap became 0)");
5088
		} else {
5089 5090
			/* initial probe */
			rbd_warn(rbd_dev, "clone is standalone (overlap 0)");
5091
		}
A
Alex Elder 已提交
5092
	}
5093 5094
	rbd_dev->parent_overlap = overlap;

5095 5096 5097 5098 5099 5100 5101 5102 5103
out:
	ret = 0;
out_err:
	kfree(reply_buf);
	rbd_spec_put(parent_spec);

	return ret;
}

5104 5105 5106 5107 5108 5109 5110 5111 5112 5113 5114 5115 5116
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;

5117 5118 5119
	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);
5120 5121 5122 5123 5124 5125 5126 5127 5128 5129 5130 5131 5132
	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
5133
	obj_size = rbd_obj_bytes(&rbd_dev->header);
5134 5135 5136 5137 5138 5139 5140 5141 5142 5143 5144 5145 5146 5147
	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;
	}
5148 5149
	rbd_dev->header.stripe_unit = stripe_unit;
	rbd_dev->header.stripe_count = stripe_count;
5150 5151 5152 5153

	return 0;
}

5154 5155
static char *rbd_dev_image_name(struct rbd_device *rbd_dev)
{
5156
	CEPH_DEFINE_OID_ONSTACK(oid);
5157 5158 5159 5160 5161 5162 5163 5164 5165 5166 5167 5168
	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
5169 5170
	len = strlen(rbd_dev->spec->image_id);
	image_id_size = sizeof (__le32) + len;
5171 5172 5173 5174 5175
	image_id = kmalloc(image_id_size, GFP_KERNEL);
	if (!image_id)
		return NULL;

	p = image_id;
5176
	end = image_id + image_id_size;
5177
	ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len);
5178 5179 5180 5181 5182 5183

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

5184 5185 5186 5187
	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);
5188 5189 5190
	if (ret < 0)
		goto out;
	p = reply_buf;
5191 5192
	end = reply_buf + ret;

5193 5194 5195 5196 5197 5198 5199 5200 5201 5202 5203 5204
	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;
}

5205 5206 5207 5208 5209 5210 5211 5212 5213 5214 5215 5216 5217 5218 5219 5220 5221 5222 5223 5224 5225 5226 5227 5228 5229 5230 5231 5232 5233 5234
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);
5235 5236 5237 5238 5239 5240 5241
		if (IS_ERR(snap_name)) {
			/* ignore no-longer existing snapshots */
			if (PTR_ERR(snap_name) == -ENOENT)
				continue;
			else
				break;
		}
5242 5243 5244 5245 5246 5247 5248 5249 5250 5251 5252 5253 5254 5255 5256 5257 5258 5259
		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);
}

5260
/*
5261 5262 5263 5264 5265 5266 5267 5268 5269 5270 5271 5272 5273 5274 5275 5276 5277 5278 5279 5280 5281 5282 5283 5284 5285 5286 5287
 * 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.
5288
 *
5289 5290
 * All names in an rbd spec are dynamically allocated.  It's OK if we
 * can't figure out the name for an image id.
5291
 */
5292
static int rbd_spec_fill_names(struct rbd_device *rbd_dev)
5293
{
5294 5295 5296 5297 5298
	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;
5299 5300
	int ret;

5301 5302 5303
	rbd_assert(spec->pool_id != CEPH_NOPOOL);
	rbd_assert(spec->image_id);
	rbd_assert(spec->snap_id != CEPH_NOSNAP);
5304

5305
	/* Get the pool name; we have to make our own copy of this */
5306

5307 5308 5309
	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);
5310 5311
		return -EIO;
	}
5312 5313
	pool_name = kstrdup(pool_name, GFP_KERNEL);
	if (!pool_name)
5314 5315 5316 5317
		return -ENOMEM;

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

5318 5319
	image_name = rbd_dev_image_name(rbd_dev);
	if (!image_name)
A
Alex Elder 已提交
5320
		rbd_warn(rbd_dev, "unable to get image name");
5321

5322
	/* Fetch the snapshot name */
5323

5324
	snap_name = rbd_snap_name(rbd_dev, spec->snap_id);
5325 5326
	if (IS_ERR(snap_name)) {
		ret = PTR_ERR(snap_name);
5327
		goto out_err;
5328 5329 5330 5331 5332
	}

	spec->pool_name = pool_name;
	spec->image_name = image_name;
	spec->snap_name = snap_name;
5333 5334

	return 0;
5335

5336
out_err:
5337 5338
	kfree(image_name);
	kfree(pool_name);
5339 5340 5341
	return ret;
}

A
Alex Elder 已提交
5342
static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev)
5343 5344 5345 5346 5347 5348 5349 5350 5351 5352 5353 5354 5355 5356 5357 5358 5359 5360 5361 5362 5363 5364 5365
{
	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;

5366 5367 5368
	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
				  &rbd_dev->header_oloc, "get_snapcontext",
				  NULL, 0, reply_buf, size);
5369
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
5370 5371 5372 5373
	if (ret < 0)
		goto out;

	p = reply_buf;
5374 5375
	end = reply_buf + ret;
	ret = -ERANGE;
5376 5377 5378 5379 5380 5381 5382 5383 5384 5385 5386 5387 5388 5389 5390 5391
	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;
5392
	ret = 0;
5393

5394
	snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
5395 5396 5397 5398 5399 5400 5401 5402
	if (!snapc) {
		ret = -ENOMEM;
		goto out;
	}
	snapc->seq = seq;
	for (i = 0; i < snap_count; i++)
		snapc->snaps[i] = ceph_decode_64(&p);

5403
	ceph_put_snap_context(rbd_dev->header.snapc);
5404 5405 5406
	rbd_dev->header.snapc = snapc;

	dout("  snap context seq = %llu, snap_count = %u\n",
5407
		(unsigned long long)seq, (unsigned int)snap_count);
5408 5409 5410
out:
	kfree(reply_buf);

5411
	return ret;
5412 5413
}

5414 5415
static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
					u64 snap_id)
A
Alex Elder 已提交
5416 5417 5418
{
	size_t size;
	void *reply_buf;
5419
	__le64 snapid;
A
Alex Elder 已提交
5420 5421 5422 5423 5424 5425 5426 5427 5428 5429
	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);

5430
	snapid = cpu_to_le64(snap_id);
5431 5432 5433
	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
				  &rbd_dev->header_oloc, "get_snapshot_name",
				  &snapid, sizeof(snapid), reply_buf, size);
5434
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
5435 5436
	if (ret < 0) {
		snap_name = ERR_PTR(ret);
A
Alex Elder 已提交
5437
		goto out;
5438
	}
A
Alex Elder 已提交
5439 5440

	p = reply_buf;
5441
	end = reply_buf + ret;
A
Alex Elder 已提交
5442
	snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
5443
	if (IS_ERR(snap_name))
A
Alex Elder 已提交
5444 5445
		goto out;

5446
	dout("  snap_id 0x%016llx snap_name = %s\n",
5447
		(unsigned long long)snap_id, snap_name);
A
Alex Elder 已提交
5448 5449 5450
out:
	kfree(reply_buf);

5451
	return snap_name;
A
Alex Elder 已提交
5452 5453
}

5454
static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev)
A
Alex Elder 已提交
5455
{
5456
	bool first_time = rbd_dev->header.object_prefix == NULL;
A
Alex Elder 已提交
5457 5458
	int ret;

5459 5460
	ret = rbd_dev_v2_image_size(rbd_dev);
	if (ret)
5461
		return ret;
5462

5463 5464 5465
	if (first_time) {
		ret = rbd_dev_v2_header_onetime(rbd_dev);
		if (ret)
5466
			return ret;
5467 5468
	}

A
Alex Elder 已提交
5469
	ret = rbd_dev_v2_snap_context(rbd_dev);
5470 5471 5472 5473
	if (ret && first_time) {
		kfree(rbd_dev->header.object_prefix);
		rbd_dev->header.object_prefix = NULL;
	}
A
Alex Elder 已提交
5474 5475 5476 5477

	return ret;
}

5478 5479 5480 5481 5482 5483 5484 5485 5486 5487
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);
}

5488 5489 5490
/*
 * 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
5491 5492
 * the token (string of non-white space characters) found.  Note
 * that *buf must be terminated with '\0'.
5493 5494 5495 5496 5497 5498 5499 5500 5501 5502 5503 5504 5505 5506
 */
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
5507 5508 5509 5510 5511 5512 5513 5514 5515 5516 5517 5518 5519 5520 5521 5522 5523 5524 5525 5526 5527 5528
/*
 * 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
5529
	dup = kmemdup(*buf, len + 1, GFP_KERNEL);
A
Alex Elder 已提交
5530 5531 5532 5533 5534 5535 5536 5537 5538 5539 5540
	if (!dup)
		return NULL;
	*(dup + len) = '\0';
	*buf += len;

	if (lenp)
		*lenp = len;

	return dup;
}

5541
/*
5542 5543 5544 5545
 * 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
5546
 *
5547 5548 5549 5550 5551 5552 5553 5554 5555 5556 5557 5558 5559 5560 5561 5562 5563 5564 5565 5566 5567 5568 5569 5570 5571 5572 5573 5574 5575 5576 5577 5578 5579 5580
 * 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.
5581
 */
5582
static int rbd_add_parse_args(const char *buf,
5583
				struct ceph_options **ceph_opts,
5584 5585
				struct rbd_options **opts,
				struct rbd_spec **rbd_spec)
5586
{
A
Alex Elder 已提交
5587
	size_t len;
5588
	char *options;
5589
	const char *mon_addrs;
5590
	char *snap_name;
5591
	size_t mon_addrs_size;
5592
	struct rbd_spec *spec = NULL;
5593
	struct rbd_options *rbd_opts = NULL;
5594
	struct ceph_options *copts;
5595
	int ret;
5596 5597 5598

	/* The first four tokens are required */

5599
	len = next_token(&buf);
5600 5601 5602 5603
	if (!len) {
		rbd_warn(NULL, "no monitor address(es) provided");
		return -EINVAL;
	}
5604
	mon_addrs = buf;
5605
	mon_addrs_size = len + 1;
5606
	buf += len;
5607

5608
	ret = -EINVAL;
5609 5610
	options = dup_token(&buf, NULL);
	if (!options)
5611
		return -ENOMEM;
5612 5613 5614 5615
	if (!*options) {
		rbd_warn(NULL, "no options provided");
		goto out_err;
	}
5616

5617 5618
	spec = rbd_spec_alloc();
	if (!spec)
5619
		goto out_mem;
5620 5621 5622 5623

	spec->pool_name = dup_token(&buf, NULL);
	if (!spec->pool_name)
		goto out_mem;
5624 5625 5626 5627
	if (!*spec->pool_name) {
		rbd_warn(NULL, "no pool name provided");
		goto out_err;
	}
5628

A
Alex Elder 已提交
5629
	spec->image_name = dup_token(&buf, NULL);
5630
	if (!spec->image_name)
5631
		goto out_mem;
5632 5633 5634 5635
	if (!*spec->image_name) {
		rbd_warn(NULL, "no image name provided");
		goto out_err;
	}
5636

5637 5638 5639 5640
	/*
	 * Snapshot name is optional; default is to use "-"
	 * (indicating the head/no snapshot).
	 */
5641
	len = next_token(&buf);
5642
	if (!len) {
5643 5644
		buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */
		len = sizeof (RBD_SNAP_HEAD_NAME) - 1;
5645
	} else if (len > RBD_MAX_SNAP_NAME_LEN) {
5646
		ret = -ENAMETOOLONG;
5647
		goto out_err;
5648
	}
5649 5650
	snap_name = kmemdup(buf, len + 1, GFP_KERNEL);
	if (!snap_name)
5651
		goto out_mem;
5652 5653
	*(snap_name + len) = '\0';
	spec->snap_name = snap_name;
A
Alex Elder 已提交
5654

5655
	/* Initialize all rbd options to the defaults */
5656

5657 5658 5659 5660 5661
	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
5662
	rbd_opts->queue_depth = RBD_QUEUE_DEPTH_DEFAULT;
I
Ilya Dryomov 已提交
5663
	rbd_opts->lock_on_read = RBD_LOCK_ON_READ_DEFAULT;
A
Alex Elder 已提交
5664

5665
	copts = ceph_parse_options(options, mon_addrs,
5666
					mon_addrs + mon_addrs_size - 1,
5667
					parse_rbd_opts_token, rbd_opts);
5668 5669
	if (IS_ERR(copts)) {
		ret = PTR_ERR(copts);
5670 5671
		goto out_err;
	}
5672 5673 5674
	kfree(options);

	*ceph_opts = copts;
5675
	*opts = rbd_opts;
5676
	*rbd_spec = spec;
5677

5678
	return 0;
5679
out_mem:
5680
	ret = -ENOMEM;
A
Alex Elder 已提交
5681
out_err:
5682 5683
	kfree(rbd_opts);
	rbd_spec_put(spec);
5684
	kfree(options);
A
Alex Elder 已提交
5685

5686
	return ret;
5687 5688
}

5689 5690 5691 5692 5693
/*
 * Return pool id (>= 0) or a negative error code.
 */
static int rbd_add_get_pool_id(struct rbd_client *rbdc, const char *pool_name)
{
5694
	struct ceph_options *opts = rbdc->client->options;
5695 5696 5697 5698 5699 5700 5701
	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) {
5702 5703
		ret = ceph_monc_get_version(&rbdc->client->monc, "osdmap",
					    &newest_epoch);
5704 5705 5706 5707
		if (ret < 0)
			return ret;

		if (rbdc->client->osdc.osdmap->epoch < newest_epoch) {
5708
			ceph_osdc_maybe_request_map(&rbdc->client->osdc);
5709
			(void) ceph_monc_wait_osdmap(&rbdc->client->monc,
5710 5711
						     newest_epoch,
						     opts->mount_timeout);
5712 5713 5714 5715 5716 5717 5718 5719 5720 5721
			goto again;
		} else {
			/* the osdmap we have is new enough */
			return -ENOENT;
		}
	}

	return ret;
}

A
Alex Elder 已提交
5722 5723 5724 5725 5726 5727 5728 5729 5730 5731 5732 5733 5734 5735 5736 5737 5738 5739
/*
 * 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;
5740
	CEPH_DEFINE_OID_ONSTACK(oid);
A
Alex Elder 已提交
5741
	void *response;
5742
	char *image_id;
5743

A
Alex Elder 已提交
5744 5745 5746
	/*
	 * When probing a parent image, the image id is already
	 * known (and the image name likely is not).  There's no
5747 5748
	 * 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
5749
	 */
5750 5751 5752
	if (rbd_dev->spec->image_id) {
		rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1;

A
Alex Elder 已提交
5753
		return 0;
5754
	}
A
Alex Elder 已提交
5755

A
Alex Elder 已提交
5756 5757 5758 5759
	/*
	 * First, see if the format 2 image id file exists, and if
	 * so, get the image's persistent id from it.
	 */
5760 5761 5762 5763 5764 5765
	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
5766 5767 5768 5769 5770 5771 5772 5773 5774 5775

	/* 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;
	}

5776 5777
	/* If it doesn't exist we'll assume it's a format 1 image */

5778 5779 5780
	ret = rbd_obj_method_sync(rbd_dev, &oid, &rbd_dev->header_oloc,
				  "get_id", NULL, 0,
				  response, RBD_IMAGE_ID_LEN_MAX);
5781
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
5782 5783 5784 5785 5786
	if (ret == -ENOENT) {
		image_id = kstrdup("", GFP_KERNEL);
		ret = image_id ? 0 : -ENOMEM;
		if (!ret)
			rbd_dev->image_format = 1;
5787
	} else if (ret >= 0) {
5788 5789 5790
		void *p = response;

		image_id = ceph_extract_encoded_string(&p, p + ret,
A
Alex Elder 已提交
5791
						NULL, GFP_NOIO);
5792
		ret = PTR_ERR_OR_ZERO(image_id);
5793 5794 5795 5796 5797 5798 5799
		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
5800 5801 5802
	}
out:
	kfree(response);
5803
	ceph_oid_destroy(&oid);
A
Alex Elder 已提交
5804 5805 5806
	return ret;
}

A
Alex Elder 已提交
5807 5808 5809 5810
/*
 * Undo whatever state changes are made by v1 or v2 header info
 * call.
 */
A
Alex Elder 已提交
5811 5812 5813 5814
static void rbd_dev_unprobe(struct rbd_device *rbd_dev)
{
	struct rbd_image_header	*header;

5815
	rbd_dev_parent_put(rbd_dev);
A
Alex Elder 已提交
5816 5817 5818 5819

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

	header = &rbd_dev->header;
5820
	ceph_put_snap_context(header->snapc);
A
Alex Elder 已提交
5821 5822 5823 5824 5825 5826
	kfree(header->snap_sizes);
	kfree(header->snap_names);
	kfree(header->object_prefix);
	memset(header, 0, sizeof (*header));
}

5827
static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev)
5828 5829 5830
{
	int ret;

5831
	ret = rbd_dev_v2_object_prefix(rbd_dev);
5832
	if (ret)
5833 5834
		goto out_err;

5835 5836 5837 5838
	/*
	 * Get the and check features for the image.  Currently the
	 * features are assumed to never change.
	 */
5839
	ret = rbd_dev_v2_features(rbd_dev);
5840
	if (ret)
5841
		goto out_err;
5842

5843 5844 5845 5846 5847 5848 5849
	/* 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;
	}
5850

A
Alex Elder 已提交
5851
	return 0;
5852
out_err:
A
Alex Elder 已提交
5853
	rbd_dev->header.features = 0;
5854 5855
	kfree(rbd_dev->header.object_prefix);
	rbd_dev->header.object_prefix = NULL;
5856 5857

	return ret;
5858 5859
}

5860 5861 5862 5863 5864 5865
/*
 * @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
5866
{
5867
	struct rbd_device *parent = NULL;
5868 5869 5870 5871 5872
	int ret;

	if (!rbd_dev->parent_spec)
		return 0;

5873 5874 5875 5876 5877 5878
	if (++depth > RBD_MAX_PARENT_CHAIN_LEN) {
		pr_info("parent chain is too long (%d)\n", depth);
		ret = -EINVAL;
		goto out_err;
	}

5879
	parent = __rbd_dev_create(rbd_dev->rbd_client, rbd_dev->parent_spec);
5880 5881
	if (!parent) {
		ret = -ENOMEM;
5882
		goto out_err;
5883 5884 5885 5886 5887 5888 5889 5890
	}

	/*
	 * 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);
5891

5892
	ret = rbd_dev_image_probe(parent, depth);
5893 5894
	if (ret < 0)
		goto out_err;
5895

5896
	rbd_dev->parent = parent;
5897
	atomic_set(&rbd_dev->parent_ref, 1);
5898
	return 0;
5899

5900
out_err:
5901
	rbd_dev_unparent(rbd_dev);
5902
	rbd_dev_destroy(parent);
5903 5904 5905
	return ret;
}

I
Ilya Dryomov 已提交
5906 5907 5908 5909
/*
 * rbd_dev->header_rwsem must be locked for write and will be unlocked
 * upon return.
 */
5910
static int rbd_dev_device_setup(struct rbd_device *rbd_dev)
5911
{
A
Alex Elder 已提交
5912
	int ret;
A
Alex Elder 已提交
5913

5914
	/* Record our major and minor device numbers. */
A
Alex Elder 已提交
5915

5916 5917 5918
	if (!single_major) {
		ret = register_blkdev(0, rbd_dev->name);
		if (ret < 0)
5919
			goto err_out_unlock;
5920 5921 5922 5923 5924 5925 5926

		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
5927 5928 5929 5930 5931 5932 5933

	/* Set up the blkdev mapping. */

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

5934
	ret = rbd_dev_mapping_set(rbd_dev);
A
Alex Elder 已提交
5935 5936
	if (ret)
		goto err_out_disk;
I
Ilya Dryomov 已提交
5937

5938
	set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE);
5939
	set_disk_ro(rbd_dev->disk, rbd_dev->mapping.read_only);
5940

5941 5942
	dev_set_name(&rbd_dev->dev, "%d", rbd_dev->dev_id);
	ret = device_add(&rbd_dev->dev);
5943
	if (ret)
5944
		goto err_out_mapping;
A
Alex Elder 已提交
5945 5946 5947

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

5948
	set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
I
Ilya Dryomov 已提交
5949
	up_write(&rbd_dev->header_rwsem);
A
Alex Elder 已提交
5950

5951 5952 5953 5954
	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
5955
	add_disk(rbd_dev->disk);
5956 5957 5958
	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
5959 5960

	return ret;
5961

5962 5963
err_out_mapping:
	rbd_dev_mapping_clear(rbd_dev);
A
Alex Elder 已提交
5964 5965 5966
err_out_disk:
	rbd_free_disk(rbd_dev);
err_out_blkdev:
5967 5968
	if (!single_major)
		unregister_blkdev(rbd_dev->major, rbd_dev->name);
I
Ilya Dryomov 已提交
5969 5970
err_out_unlock:
	up_write(&rbd_dev->header_rwsem);
A
Alex Elder 已提交
5971 5972 5973
	return ret;
}

A
Alex Elder 已提交
5974 5975 5976
static int rbd_dev_header_name(struct rbd_device *rbd_dev)
{
	struct rbd_spec *spec = rbd_dev->spec;
5977
	int ret;
A
Alex Elder 已提交
5978 5979 5980 5981 5982

	/* 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)
5983 5984
		ret = ceph_oid_aprintf(&rbd_dev->header_oid, GFP_KERNEL, "%s%s",
				       spec->image_name, RBD_SUFFIX);
A
Alex Elder 已提交
5985
	else
5986 5987
		ret = ceph_oid_aprintf(&rbd_dev->header_oid, GFP_KERNEL, "%s%s",
				       RBD_HEADER_PREFIX, spec->image_id);
A
Alex Elder 已提交
5988

5989
	return ret;
A
Alex Elder 已提交
5990 5991
}

5992 5993
static void rbd_dev_image_release(struct rbd_device *rbd_dev)
{
A
Alex Elder 已提交
5994 5995 5996 5997 5998
	rbd_dev_unprobe(rbd_dev);
	rbd_dev->image_format = 0;
	kfree(rbd_dev->spec->image_id);
	rbd_dev->spec->image_id = NULL;

5999 6000 6001
	rbd_dev_destroy(rbd_dev);
}

6002 6003
/*
 * Probe for the existence of the header object for the given rbd
6004 6005 6006
 * 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.
6007
 */
6008
static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth)
6009 6010 6011 6012
{
	int ret;

	/*
A
Alex Elder 已提交
6013 6014 6015 6016
	 * 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.
6017 6018 6019
	 */
	ret = rbd_dev_image_id(rbd_dev);
	if (ret)
6020 6021
		return ret;

A
Alex Elder 已提交
6022 6023 6024 6025
	ret = rbd_dev_header_name(rbd_dev);
	if (ret)
		goto err_out_format;

6026
	if (!depth) {
6027
		ret = rbd_register_watch(rbd_dev);
6028 6029 6030 6031 6032
		if (ret) {
			if (ret == -ENOENT)
				pr_info("image %s/%s does not exist\n",
					rbd_dev->spec->pool_name,
					rbd_dev->spec->image_name);
6033
			goto err_out_format;
6034
		}
6035
	}
6036

6037
	ret = rbd_dev_header_info(rbd_dev);
6038
	if (ret)
6039
		goto err_out_watch;
A
Alex Elder 已提交
6040

6041 6042 6043 6044 6045 6046
	/*
	 * 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.
	 */
6047
	if (!depth)
6048 6049 6050
		ret = rbd_spec_fill_snap_id(rbd_dev);
	else
		ret = rbd_spec_fill_names(rbd_dev);
6051 6052 6053 6054 6055 6056
	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
6057
		goto err_out_probe;
6058
	}
6059

6060 6061 6062 6063 6064 6065 6066 6067 6068
	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.
		 */
6069
		if (!depth && rbd_dev->parent_spec)
6070 6071 6072 6073
			rbd_warn(rbd_dev,
				 "WARNING: kernel layering is EXPERIMENTAL!");
	}

6074
	ret = rbd_dev_probe_parent(rbd_dev, depth);
A
Alex Elder 已提交
6075 6076 6077 6078
	if (ret)
		goto err_out_probe;

	dout("discovered format %u image, header name is %s\n",
6079
		rbd_dev->image_format, rbd_dev->header_oid.name);
A
Alex Elder 已提交
6080
	return 0;
6081

A
Alex Elder 已提交
6082 6083
err_out_probe:
	rbd_dev_unprobe(rbd_dev);
6084
err_out_watch:
6085
	if (!depth)
6086
		rbd_unregister_watch(rbd_dev);
A
Alex Elder 已提交
6087 6088
err_out_format:
	rbd_dev->image_format = 0;
6089 6090
	kfree(rbd_dev->spec->image_id);
	rbd_dev->spec->image_id = NULL;
6091 6092 6093
	return ret;
}

6094 6095 6096
static ssize_t do_rbd_add(struct bus_type *bus,
			  const char *buf,
			  size_t count)
6097
{
6098
	struct rbd_device *rbd_dev = NULL;
6099
	struct ceph_options *ceph_opts = NULL;
6100
	struct rbd_options *rbd_opts = NULL;
6101
	struct rbd_spec *spec = NULL;
6102
	struct rbd_client *rbdc;
6103
	bool read_only;
6104
	int rc;
6105 6106 6107 6108 6109

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

	/* parse add command */
6110
	rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec);
6111
	if (rc < 0)
6112
		goto out;
6113

6114 6115 6116
	rbdc = rbd_get_client(ceph_opts);
	if (IS_ERR(rbdc)) {
		rc = PTR_ERR(rbdc);
6117
		goto err_out_args;
6118
	}
6119 6120

	/* pick the pool */
6121
	rc = rbd_add_get_pool_id(rbdc, spec->pool_name);
6122 6123 6124
	if (rc < 0) {
		if (rc == -ENOENT)
			pr_info("pool %s does not exist\n", spec->pool_name);
6125
		goto err_out_client;
6126
	}
A
Alex Elder 已提交
6127
	spec->pool_id = (u64)rc;
6128

6129
	rbd_dev = rbd_dev_create(rbdc, spec, rbd_opts);
6130 6131
	if (!rbd_dev) {
		rc = -ENOMEM;
6132
		goto err_out_client;
6133
	}
6134 6135
	rbdc = NULL;		/* rbd_dev now owns this */
	spec = NULL;		/* rbd_dev now owns this */
6136
	rbd_opts = NULL;	/* rbd_dev now owns this */
6137

6138 6139 6140 6141 6142 6143
	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
6144
	down_write(&rbd_dev->header_rwsem);
6145
	rc = rbd_dev_image_probe(rbd_dev, 0);
6146 6147
	if (rc < 0) {
		up_write(&rbd_dev->header_rwsem);
6148
		goto err_out_rbd_dev;
6149
	}
6150

6151 6152
	/* If we are mapping a snapshot it must be marked read-only */

6153
	read_only = rbd_dev->opts->read_only;
6154 6155 6156 6157
	if (rbd_dev->spec->snap_id != CEPH_NOSNAP)
		read_only = true;
	rbd_dev->mapping.read_only = read_only;

6158
	rc = rbd_dev_device_setup(rbd_dev);
A
Alex Elder 已提交
6159
	if (rc) {
6160
		/*
6161
		 * rbd_unregister_watch() can't be moved into
6162 6163 6164
		 * rbd_dev_image_release() without refactoring, see
		 * commit 1f3ef78861ac.
		 */
6165
		rbd_unregister_watch(rbd_dev);
A
Alex Elder 已提交
6166
		rbd_dev_image_release(rbd_dev);
6167
		goto out;
A
Alex Elder 已提交
6168 6169
	}

6170 6171 6172 6173
	rc = count;
out:
	module_put(THIS_MODULE);
	return rc;
6174

6175 6176
err_out_rbd_dev:
	rbd_dev_destroy(rbd_dev);
6177
err_out_client:
6178
	rbd_put_client(rbdc);
6179
err_out_args:
6180
	rbd_spec_put(spec);
6181
	kfree(rbd_opts);
6182
	goto out;
6183 6184
}

6185 6186 6187 6188 6189 6190 6191 6192 6193 6194 6195 6196 6197 6198 6199 6200 6201
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);
}

6202
static void rbd_dev_device_release(struct rbd_device *rbd_dev)
6203 6204
{
	rbd_free_disk(rbd_dev);
6205 6206 6207 6208 6209

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

6210
	clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
6211
	device_del(&rbd_dev->dev);
A
Alex Elder 已提交
6212
	rbd_dev_mapping_clear(rbd_dev);
6213 6214
	if (!single_major)
		unregister_blkdev(rbd_dev->major, rbd_dev->name);
6215 6216
}

6217 6218
static void rbd_dev_remove_parent(struct rbd_device *rbd_dev)
{
A
Alex Elder 已提交
6219
	while (rbd_dev->parent) {
6220 6221 6222 6223 6224 6225 6226 6227 6228 6229 6230 6231
		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
6232
		rbd_assert(second);
6233
		rbd_dev_image_release(second);
A
Alex Elder 已提交
6234 6235 6236 6237
		first->parent = NULL;
		first->parent_overlap = 0;

		rbd_assert(first->parent_spec);
6238 6239 6240 6241 6242
		rbd_spec_put(first->parent_spec);
		first->parent_spec = NULL;
	}
}

6243 6244 6245
static ssize_t do_rbd_remove(struct bus_type *bus,
			     const char *buf,
			     size_t count)
6246 6247
{
	struct rbd_device *rbd_dev = NULL;
6248 6249
	struct list_head *tmp;
	int dev_id;
M
Mike Christie 已提交
6250
	char opt_buf[6];
6251
	bool already = false;
M
Mike Christie 已提交
6252
	bool force = false;
6253
	int ret;
6254

M
Mike Christie 已提交
6255 6256 6257 6258 6259
	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");
6260
		return -EINVAL;
M
Mike Christie 已提交
6261 6262 6263 6264 6265 6266 6267 6268 6269
	}
	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;
		}
	}
6270

6271 6272 6273 6274 6275 6276 6277 6278
	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;
		}
6279
	}
6280 6281
	if (!ret) {
		spin_lock_irq(&rbd_dev->lock);
M
Mike Christie 已提交
6282
		if (rbd_dev->open_count && !force)
6283 6284
			ret = -EBUSY;
		else
6285 6286
			already = test_and_set_bit(RBD_DEV_FLAG_REMOVING,
							&rbd_dev->flags);
6287 6288 6289
		spin_unlock_irq(&rbd_dev->lock);
	}
	spin_unlock(&rbd_dev_list_lock);
6290
	if (ret < 0 || already)
6291
		return ret;
6292

M
Mike Christie 已提交
6293 6294 6295 6296 6297 6298 6299 6300 6301
	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);
	}

6302 6303 6304 6305
	down_write(&rbd_dev->lock_rwsem);
	if (__rbd_is_lock_owner(rbd_dev))
		rbd_unlock(rbd_dev);
	up_write(&rbd_dev->lock_rwsem);
6306
	rbd_unregister_watch(rbd_dev);
6307

6308 6309 6310 6311 6312 6313
	/*
	 * Don't free anything from rbd_dev->disk until after all
	 * notifies are completely processed. Otherwise
	 * rbd_bus_del_dev() will race with rbd_watch_cb(), resulting
	 * in a potential use after free of rbd_dev->disk or rbd_dev.
	 */
6314
	rbd_dev_device_release(rbd_dev);
6315
	rbd_dev_image_release(rbd_dev);
A
Alex Elder 已提交
6316

6317
	return count;
6318 6319
}

6320 6321 6322 6323 6324 6325 6326 6327 6328 6329 6330 6331 6332 6333 6334 6335 6336
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);
}

6337 6338
/*
 * create control files in sysfs
6339
 * /sys/bus/rbd/...
6340 6341 6342
 */
static int rbd_sysfs_init(void)
{
6343
	int ret;
6344

6345
	ret = device_register(&rbd_root_dev);
A
Alex Elder 已提交
6346
	if (ret < 0)
6347
		return ret;
6348

6349 6350 6351
	ret = bus_register(&rbd_bus_type);
	if (ret < 0)
		device_unregister(&rbd_root_dev);
6352 6353 6354 6355 6356 6357

	return ret;
}

static void rbd_sysfs_cleanup(void)
{
6358
	bus_unregister(&rbd_bus_type);
6359
	device_unregister(&rbd_root_dev);
6360 6361
}

6362 6363 6364
static int rbd_slab_init(void)
{
	rbd_assert(!rbd_img_request_cache);
G
Geliang Tang 已提交
6365
	rbd_img_request_cache = KMEM_CACHE(rbd_img_request, 0);
6366 6367 6368 6369
	if (!rbd_img_request_cache)
		return -ENOMEM;

	rbd_assert(!rbd_obj_request_cache);
G
Geliang Tang 已提交
6370
	rbd_obj_request_cache = KMEM_CACHE(rbd_obj_request, 0);
6371 6372 6373 6374 6375
	if (!rbd_obj_request_cache)
		goto out_err;

	rbd_assert(!rbd_segment_name_cache);
	rbd_segment_name_cache = kmem_cache_create("rbd_segment_name",
6376
					CEPH_MAX_OID_NAME_LEN + 1, 1, 0, NULL);
6377
	if (rbd_segment_name_cache)
6378
		return 0;
6379
out_err:
6380 6381
	kmem_cache_destroy(rbd_obj_request_cache);
	rbd_obj_request_cache = NULL;
6382

6383 6384 6385
	kmem_cache_destroy(rbd_img_request_cache);
	rbd_img_request_cache = NULL;

6386 6387 6388 6389 6390
	return -ENOMEM;
}

static void rbd_slab_exit(void)
{
6391 6392 6393 6394
	rbd_assert(rbd_segment_name_cache);
	kmem_cache_destroy(rbd_segment_name_cache);
	rbd_segment_name_cache = NULL;

6395 6396 6397 6398
	rbd_assert(rbd_obj_request_cache);
	kmem_cache_destroy(rbd_obj_request_cache);
	rbd_obj_request_cache = NULL;

6399 6400 6401 6402 6403
	rbd_assert(rbd_img_request_cache);
	kmem_cache_destroy(rbd_img_request_cache);
	rbd_img_request_cache = NULL;
}

A
Alex Elder 已提交
6404
static int __init rbd_init(void)
6405 6406 6407
{
	int rc;

6408 6409 6410 6411
	if (!libceph_compatible(NULL)) {
		rbd_warn(NULL, "libceph incompatibility (quitting)");
		return -EINVAL;
	}
I
Ilya Dryomov 已提交
6412

6413
	rc = rbd_slab_init();
6414 6415
	if (rc)
		return rc;
I
Ilya Dryomov 已提交
6416

6417 6418
	/*
	 * The number of active work items is limited by the number of
I
Ilya Dryomov 已提交
6419
	 * rbd devices * queue depth, so leave @max_active at default.
6420 6421 6422 6423 6424 6425 6426
	 */
	rbd_wq = alloc_workqueue(RBD_DRV_NAME, WQ_MEM_RECLAIM, 0);
	if (!rbd_wq) {
		rc = -ENOMEM;
		goto err_out_slab;
	}

6427 6428 6429 6430
	if (single_major) {
		rbd_major = register_blkdev(0, RBD_DRV_NAME);
		if (rbd_major < 0) {
			rc = rbd_major;
6431
			goto err_out_wq;
6432 6433 6434
		}
	}

6435 6436
	rc = rbd_sysfs_init();
	if (rc)
6437 6438 6439 6440 6441 6442
		goto err_out_blkdev;

	if (single_major)
		pr_info("loaded (major %d)\n", rbd_major);
	else
		pr_info("loaded\n");
6443

I
Ilya Dryomov 已提交
6444 6445
	return 0;

6446 6447 6448
err_out_blkdev:
	if (single_major)
		unregister_blkdev(rbd_major, RBD_DRV_NAME);
6449 6450
err_out_wq:
	destroy_workqueue(rbd_wq);
I
Ilya Dryomov 已提交
6451 6452
err_out_slab:
	rbd_slab_exit();
6453
	return rc;
6454 6455
}

A
Alex Elder 已提交
6456
static void __exit rbd_exit(void)
6457
{
I
Ilya Dryomov 已提交
6458
	ida_destroy(&rbd_dev_id_ida);
6459
	rbd_sysfs_cleanup();
6460 6461
	if (single_major)
		unregister_blkdev(rbd_major, RBD_DRV_NAME);
6462
	destroy_workqueue(rbd_wq);
6463
	rbd_slab_exit();
6464 6465 6466 6467 6468
}

module_init(rbd_init);
module_exit(rbd_exit);

A
Alex Elder 已提交
6469
MODULE_AUTHOR("Alex Elder <elder@inktank.com>");
6470 6471 6472 6473 6474
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>");

6475
MODULE_DESCRIPTION("RADOS Block Device (RBD) driver");
6476
MODULE_LICENSE("GPL");