rbd.c 165.7 KB
Newer Older
1

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


   based on drivers/block/osdblk.c:

   Copyright 2009 Red Hat, Inc.

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

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

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



25
   For usage instructions, please refer to:
26

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

 */

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

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

#include "rbd_types.h"

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

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

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

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

	atomic_dec(v);

	return -EINVAL;
}

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

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

	atomic_inc(v);

	return -EINVAL;
}

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

97 98
#define RBD_MINORS_PER_MAJOR		256
#define RBD_SINGLE_MAJOR_PART_SHIFT	4
99

100 101
#define RBD_MAX_PARENT_CHAIN_LEN	16

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

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

#define RBD_SNAP_HEAD_NAME	"-"

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

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

116
#define RBD_OBJ_PREFIX_LEN_MAX	64
A
Alex Elder 已提交
117

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

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

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;
146 147 148
	__u8 obj_order;
	__u8 crypt_type;
	__u8 comp_type;
149 150 151
	u64 stripe_unit;
	u64 stripe_count;
	u64 features;		/* Might be changeable someday? */
152

A
Alex Elder 已提交
153 154 155
	/* The remaining fields need to be updated occasionally */
	u64 image_size;
	struct ceph_snap_context *snapc;
156 157
	char *snap_names;	/* format 1 only */
	u64 *snap_sizes;	/* format 1 only */
158 159
};

160 161 162 163
/*
 * An rbd image specification.
 *
 * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely
A
Alex Elder 已提交
164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183
 * 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).
184 185 186
 */
struct rbd_spec {
	u64		pool_id;
187
	const char	*pool_name;
188

189 190
	const char	*image_id;
	const char	*image_name;
191 192

	u64		snap_id;
193
	const char	*snap_name;
194 195 196 197

	struct kref	kref;
};

198
/*
A
Alex Elder 已提交
199
 * an instance of the client.  multiple devices may share an rbd client.
200 201 202 203 204 205 206
 */
struct rbd_client {
	struct ceph_client	*client;
	struct kref		kref;
	struct list_head	node;
};

A
Alex Elder 已提交
207 208 209 210 211 212 213 214
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 *);

215 216 217
enum obj_request_type {
	OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES
};
A
Alex Elder 已提交
218

G
Guangliang Zhao 已提交
219 220 221
enum obj_operation_type {
	OBJ_OP_WRITE,
	OBJ_OP_READ,
222
	OBJ_OP_DISCARD,
G
Guangliang Zhao 已提交
223 224
};

225 226
enum obj_req_flags {
	OBJ_REQ_DONE,		/* completion flag: not done = 0, done = 1 */
227
	OBJ_REQ_IMG_DATA,	/* object usage: standalone = 0, image = 1 */
228 229
	OBJ_REQ_KNOWN,		/* EXISTS flag valid: no = 0, yes = 1 */
	OBJ_REQ_EXISTS,		/* target exists: no = 0, yes = 1 */
230 231
};

A
Alex Elder 已提交
232 233 234 235
struct rbd_obj_request {
	const char		*object_name;
	u64			offset;		/* object start byte */
	u64			length;		/* bytes from offset */
236
	unsigned long		flags;
A
Alex Elder 已提交
237

238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262
	/*
	 * 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
263 264 265
	u32			which;		/* posn image request list */

	enum obj_request_type	type;
266 267 268 269 270 271 272
	union {
		struct bio	*bio_list;
		struct {
			struct page	**pages;
			u32		page_count;
		};
	};
273
	struct page		**copyup_pages;
274
	u32			copyup_page_count;
A
Alex Elder 已提交
275 276 277 278

	struct ceph_osd_request	*osd_req;

	u64			xferred;	/* bytes transferred */
279
	int			result;
A
Alex Elder 已提交
280 281

	rbd_obj_callback_t	callback;
282
	struct completion	completion;
A
Alex Elder 已提交
283 284 285 286

	struct kref		kref;
};

A
Alex Elder 已提交
287
enum img_req_flags {
288 289
	IMG_REQ_WRITE,		/* I/O direction: read = 0, write = 1 */
	IMG_REQ_CHILD,		/* initiator: block = 0, child image = 1 */
290
	IMG_REQ_LAYERED,	/* ENOENT handling: normal = 0, layered = 1 */
291
	IMG_REQ_DISCARD,	/* discard: normal = 0, discard request = 1 */
A
Alex Elder 已提交
292 293
};

A
Alex Elder 已提交
294 295 296 297
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
298
	unsigned long		flags;
A
Alex Elder 已提交
299
	union {
300
		u64			snap_id;	/* for reads */
A
Alex Elder 已提交
301
		struct ceph_snap_context *snapc;	/* for writes */
302 303 304 305
	};
	union {
		struct request		*rq;		/* block request */
		struct rbd_obj_request	*obj_request;	/* obj req initiator */
A
Alex Elder 已提交
306
	};
307
	struct page		**copyup_pages;
308
	u32			copyup_page_count;
A
Alex Elder 已提交
309 310 311
	spinlock_t		completion_lock;/* protects next_completion */
	u32			next_completion;
	rbd_img_callback_t	callback;
312
	u64			xferred;/* aggregate bytes transferred */
313
	int			result;	/* first nonzero obj_request result */
A
Alex Elder 已提交
314 315 316 317 318 319 320 321

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

	struct kref		kref;
};

#define for_each_obj_request(ireq, oreq) \
322
	list_for_each_entry(oreq, &(ireq)->obj_requests, links)
A
Alex Elder 已提交
323
#define for_each_obj_request_from(ireq, oreq) \
324
	list_for_each_entry_from(oreq, &(ireq)->obj_requests, links)
A
Alex Elder 已提交
325
#define for_each_obj_request_safe(ireq, oreq, n) \
326
	list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links)
A
Alex Elder 已提交
327

328 329 330 331 332 333
enum rbd_watch_state {
	RBD_WATCH_STATE_UNREGISTERED,
	RBD_WATCH_STATE_REGISTERED,
	RBD_WATCH_STATE_ERROR,
};

334 335 336 337 338 339 340 341 342 343 344 345
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
346
struct rbd_mapping {
A
Alex Elder 已提交
347
	u64                     size;
A
Alex Elder 已提交
348
	u64                     features;
A
Alex Elder 已提交
349 350 351
	bool			read_only;
};

352 353 354 355
/*
 * a single device
 */
struct rbd_device {
A
Alex Elder 已提交
356
	int			dev_id;		/* blkdev unique id */
357 358

	int			major;		/* blkdev assigned major */
359
	int			minor;
360 361
	struct gendisk		*disk;		/* blkdev's gendisk and rq */

362
	u32			image_format;	/* Either 1 or 2 */
363 364 365 366
	struct rbd_client	*rbd_client;

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

367
	spinlock_t		lock;		/* queue, flags, open_count */
368 369

	struct rbd_image_header	header;
370
	unsigned long		flags;		/* possibly lock protected */
371
	struct rbd_spec		*spec;
372
	struct rbd_options	*opts;
373

374
	struct ceph_object_id	header_oid;
375
	struct ceph_object_locator header_oloc;
376

377
	struct ceph_file_layout	layout;		/* used for all rbd requests */
378

379 380
	struct mutex		watch_mutex;
	enum rbd_watch_state	watch_state;
381
	struct ceph_osd_linger_request *watch_handle;
382 383
	u64			watch_cookie;
	struct delayed_work	watch_dwork;
384

385 386 387 388 389 390 391 392 393
	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;

394 395
	struct workqueue_struct	*task_wq;

396 397
	struct rbd_spec		*parent_spec;
	u64			parent_overlap;
398
	atomic_t		parent_ref;
399
	struct rbd_device	*parent;
400

C
Christoph Hellwig 已提交
401 402 403
	/* Block layer tags. */
	struct blk_mq_tag_set	tag_set;

404 405
	/* protects updating the header */
	struct rw_semaphore     header_rwsem;
A
Alex Elder 已提交
406 407

	struct rbd_mapping	mapping;
408 409

	struct list_head	node;
410 411 412

	/* sysfs related */
	struct device		dev;
413
	unsigned long		open_count;	/* protected by lock */
414 415
};

416 417 418 419 420 421 422
/*
 * Flag bits for rbd_dev->flags.  If atomicity is required,
 * rbd_dev->lock is used to protect access.
 *
 * Currently, only the "removing" flag (which is coupled with the
 * "open_count" field) requires atomic access.
 */
423 424
enum rbd_dev_flags {
	RBD_DEV_FLAG_EXISTS,	/* mapped snapshot has not been deleted */
425
	RBD_DEV_FLAG_REMOVING,	/* this mapping is being removed */
426 427
};

428
static DEFINE_MUTEX(client_mutex);	/* Serialize client creation */
429

430
static LIST_HEAD(rbd_dev_list);    /* devices */
431 432
static DEFINE_SPINLOCK(rbd_dev_list_lock);

A
Alex Elder 已提交
433 434
static LIST_HEAD(rbd_client_list);		/* clients */
static DEFINE_SPINLOCK(rbd_client_list_lock);
435

436 437
/* Slab caches for frequently-allocated structures */

438
static struct kmem_cache	*rbd_img_request_cache;
439
static struct kmem_cache	*rbd_obj_request_cache;
440
static struct kmem_cache	*rbd_segment_name_cache;
441

442
static int rbd_major;
443 444
static DEFINE_IDA(rbd_dev_id_ida);

445 446
static struct workqueue_struct *rbd_wq;

447 448 449 450 451 452 453 454
/*
 * 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)");

455 456
static int rbd_img_request_submit(struct rbd_img_request *img_request);

A
Alex Elder 已提交
457 458 459 460
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);
461 462 463 464
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);
465
static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth);
466
static void rbd_spec_put(struct rbd_spec *spec);
A
Alex Elder 已提交
467

468 469
static int rbd_dev_id_to_minor(int dev_id)
{
470
	return dev_id << RBD_SINGLE_MAJOR_PART_SHIFT;
471 472 473 474
}

static int minor_to_rbd_dev_id(int minor)
{
475
	return minor >> RBD_SINGLE_MAJOR_PART_SHIFT;
476 477
}

478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500
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;
}

501 502
static BUS_ATTR(add, S_IWUSR, NULL, rbd_add);
static BUS_ATTR(remove, S_IWUSR, NULL, rbd_remove);
503 504
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);
505 506 507 508

static struct attribute *rbd_bus_attrs[] = {
	&bus_attr_add.attr,
	&bus_attr_remove.attr,
509 510
	&bus_attr_add_single_major.attr,
	&bus_attr_remove_single_major.attr,
511
	NULL,
A
Alex Elder 已提交
512
};
513 514 515 516

static umode_t rbd_bus_is_visible(struct kobject *kobj,
				  struct attribute *attr, int index)
{
517 518 519 520 521
	if (!single_major &&
	    (attr == &bus_attr_add_single_major.attr ||
	     attr == &bus_attr_remove_single_major.attr))
		return 0;

522 523 524 525 526 527 528 529
	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
530 531 532

static struct bus_type rbd_bus_type = {
	.name		= "rbd",
533
	.bus_groups	= rbd_bus_groups,
A
Alex Elder 已提交
534 535 536 537 538 539 540 541 542 543 544
};

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
545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571
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
572 573 574 575 576 577 578 579 580 581 582 583
#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 */
584

I
Ilya Dryomov 已提交
585
static void rbd_osd_copyup_callback(struct rbd_obj_request *obj_request);
586
static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request);
587 588
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
589

A
Alex Elder 已提交
590
static int rbd_dev_refresh(struct rbd_device *rbd_dev);
591
static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev);
592
static int rbd_dev_header_info(struct rbd_device *rbd_dev);
593
static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev);
594 595
static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
					u64 snap_id);
596 597 598 599
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);
600

601 602
static int rbd_open(struct block_device *bdev, fmode_t mode)
{
A
Alex Elder 已提交
603
	struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
604
	bool removing = false;
605

A
Alex Elder 已提交
606
	if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only)
607 608
		return -EROFS;

609
	spin_lock_irq(&rbd_dev->lock);
610 611 612 613
	if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags))
		removing = true;
	else
		rbd_dev->open_count++;
614
	spin_unlock_irq(&rbd_dev->lock);
615 616 617
	if (removing)
		return -ENOENT;

A
Alex Elder 已提交
618
	(void) get_device(&rbd_dev->dev);
619

620 621 622
	return 0;
}

623
static void rbd_release(struct gendisk *disk, fmode_t mode)
624 625
{
	struct rbd_device *rbd_dev = disk->private_data;
626 627
	unsigned long open_count_before;

628
	spin_lock_irq(&rbd_dev->lock);
629
	open_count_before = rbd_dev->open_count--;
630
	spin_unlock_irq(&rbd_dev->lock);
631
	rbd_assert(open_count_before > 0);
632

A
Alex Elder 已提交
633
	put_device(&rbd_dev->dev);
634 635
}

G
Guangliang Zhao 已提交
636 637
static int rbd_ioctl_set_ro(struct rbd_device *rbd_dev, unsigned long arg)
{
638
	int ret = 0;
G
Guangliang Zhao 已提交
639 640
	int val;
	bool ro;
641
	bool ro_changed = false;
G
Guangliang Zhao 已提交
642

643
	/* get_user() may sleep, so call it before taking rbd_dev->lock */
G
Guangliang Zhao 已提交
644 645 646 647 648 649 650 651
	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;

652 653 654 655 656 657 658
	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
659 660
	if (rbd_dev->mapping.read_only != ro) {
		rbd_dev->mapping.read_only = ro;
661
		ro_changed = true;
G
Guangliang Zhao 已提交
662 663
	}

664 665 666 667 668 669 670
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
671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697
}

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 */

698 699 700
static const struct block_device_operations rbd_bd_ops = {
	.owner			= THIS_MODULE,
	.open			= rbd_open,
701
	.release		= rbd_release,
G
Guangliang Zhao 已提交
702 703 704 705
	.ioctl			= rbd_ioctl,
#ifdef CONFIG_COMPAT
	.compat_ioctl		= rbd_compat_ioctl,
#endif
706 707 708
};

/*
709
 * Initialize an rbd client instance.  Success or not, this function
710
 * consumes ceph_opts.  Caller holds client_mutex.
711
 */
712
static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts)
713 714 715 716
{
	struct rbd_client *rbdc;
	int ret = -ENOMEM;

A
Alex Elder 已提交
717
	dout("%s:\n", __func__);
718 719 720 721 722 723 724
	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
725
	rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0);
726
	if (IS_ERR(rbdc->client))
727
		goto out_rbdc;
A
Alex Elder 已提交
728
	ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */
729 730 731

	ret = ceph_open_session(rbdc->client);
	if (ret < 0)
732
		goto out_client;
733

A
Alex Elder 已提交
734
	spin_lock(&rbd_client_list_lock);
735
	list_add_tail(&rbdc->node, &rbd_client_list);
A
Alex Elder 已提交
736
	spin_unlock(&rbd_client_list_lock);
737

A
Alex Elder 已提交
738
	dout("%s: rbdc %p\n", __func__, rbdc);
739

740
	return rbdc;
741
out_client:
742
	ceph_destroy_client(rbdc->client);
743
out_rbdc:
744 745
	kfree(rbdc);
out_opt:
A
Alex Elder 已提交
746 747
	if (ceph_opts)
		ceph_destroy_options(ceph_opts);
A
Alex Elder 已提交
748 749
	dout("%s: error %d\n", __func__, ret);

V
Vasiliy Kulikov 已提交
750
	return ERR_PTR(ret);
751 752
}

753 754 755 756 757 758 759
static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc)
{
	kref_get(&rbdc->kref);

	return rbdc;
}

760
/*
761 762
 * Find a ceph client with specific addr and configuration.  If
 * found, bump its reference count.
763
 */
764
static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts)
765 766
{
	struct rbd_client *client_node;
767
	bool found = false;
768

A
Alex Elder 已提交
769
	if (ceph_opts->flags & CEPH_OPT_NOSHARE)
770 771
		return NULL;

772 773 774
	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)) {
775 776
			__rbd_get_client(client_node);

777 778 779 780 781 782 783
			found = true;
			break;
		}
	}
	spin_unlock(&rbd_client_list_lock);

	return found ? client_node : NULL;
784 785
}

786
/*
787
 * (Per device) rbd map options
788 789
 */
enum {
I
Ilya Dryomov 已提交
790
	Opt_queue_depth,
791 792 793 794
	Opt_last_int,
	/* int args above */
	Opt_last_string,
	/* string args above */
A
Alex Elder 已提交
795 796
	Opt_read_only,
	Opt_read_write,
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 */
808
	{Opt_err, NULL}
809 810
};

A
Alex Elder 已提交
811
struct rbd_options {
I
Ilya Dryomov 已提交
812
	int	queue_depth;
A
Alex Elder 已提交
813 814 815
	bool	read_only;
};

I
Ilya Dryomov 已提交
816
#define RBD_QUEUE_DEPTH_DEFAULT	BLKDEV_MAX_RQ
A
Alex Elder 已提交
817 818
#define RBD_READ_ONLY_DEFAULT	false

819 820
static int parse_rbd_opts_token(char *c, void *private)
{
A
Alex Elder 已提交
821
	struct rbd_options *rbd_opts = private;
822 823 824
	substring_t argstr[MAX_OPT_ARGS];
	int token, intval, ret;

A
Alex Elder 已提交
825
	token = match_token(c, rbd_opts_tokens, argstr);
826 827 828
	if (token < Opt_last_int) {
		ret = match_int(&argstr[0], &intval);
		if (ret < 0) {
829
			pr_err("bad mount option arg (not int) at '%s'\n", c);
830 831 832 833
			return ret;
		}
		dout("got int token %d val %d\n", token, intval);
	} else if (token > Opt_last_int && token < Opt_last_string) {
834
		dout("got string token %d val %s\n", token, argstr[0].from);
835 836 837 838 839
	} else {
		dout("got token %d\n", token);
	}

	switch (token) {
I
Ilya Dryomov 已提交
840 841 842 843 844 845 846
	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
847 848 849 850 851 852
	case Opt_read_only:
		rbd_opts->read_only = true;
		break;
	case Opt_read_write:
		rbd_opts->read_only = false;
		break;
853
	default:
854 855
		/* libceph prints "bad option" msg */
		return -EINVAL;
856
	}
857

858 859 860
	return 0;
}

G
Guangliang Zhao 已提交
861 862 863 864 865 866 867
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";
868 869
	case OBJ_OP_DISCARD:
		return "discard";
G
Guangliang Zhao 已提交
870 871 872 873 874
	default:
		return "???";
	}
}

875 876
/*
 * Get a ceph client with specific addr and configuration, if one does
877 878
 * not exist create it.  Either way, ceph_opts is consumed by this
 * function.
879
 */
880
static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts)
881
{
882
	struct rbd_client *rbdc;
883

884
	mutex_lock_nested(&client_mutex, SINGLE_DEPTH_NESTING);
885
	rbdc = rbd_client_find(ceph_opts);
886
	if (rbdc)	/* using an existing client */
A
Alex Elder 已提交
887
		ceph_destroy_options(ceph_opts);
888
	else
889
		rbdc = rbd_client_create(ceph_opts);
890
	mutex_unlock(&client_mutex);
891

892
	return rbdc;
893 894 895 896
}

/*
 * Destroy ceph client
A
Alex Elder 已提交
897
 *
A
Alex Elder 已提交
898
 * Caller must hold rbd_client_list_lock.
899 900 901 902 903
 */
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
904
	dout("%s: rbdc %p\n", __func__, rbdc);
905
	spin_lock(&rbd_client_list_lock);
906
	list_del(&rbdc->node);
907
	spin_unlock(&rbd_client_list_lock);
908 909 910 911 912 913 914 915 916

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

/*
 * Drop reference to ceph client node. If it's not referenced anymore, release
 * it.
 */
917
static void rbd_put_client(struct rbd_client *rbdc)
918
{
919 920
	if (rbdc)
		kref_put(&rbdc->kref, rbd_client_release);
921 922
}

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

928 929
static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk)
{
930 931 932 933 934 935 936
	size_t size;
	u32 snap_count;

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

A
Alex Elder 已提交
937 938 939 940 941 942 943 944 945 946
	/* The bio layer requires at least sector-sized I/O */

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

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

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

947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964
	/*
	 * The size of a snapshot header has to fit in a size_t, and
	 * that limits the number of snapshots.
	 */
	snap_count = le32_to_cpu(ondisk->snap_count);
	size = SIZE_MAX - sizeof (struct ceph_snap_context);
	if (snap_count > size / sizeof (__le64))
		return false;

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

	return true;
965 966
}

967
/*
968 969
 * Fill an rbd image header with information from the given format 1
 * on-disk header.
970
 */
A
Alex Elder 已提交
971
static int rbd_header_from_disk(struct rbd_device *rbd_dev,
972
				 struct rbd_image_header_ondisk *ondisk)
973
{
A
Alex Elder 已提交
974
	struct rbd_image_header *header = &rbd_dev->header;
975 976 977 978 979
	bool first_time = header->object_prefix == NULL;
	struct ceph_snap_context *snapc;
	char *object_prefix = NULL;
	char *snap_names = NULL;
	u64 *snap_sizes = NULL;
980
	u32 snap_count;
981
	size_t size;
982
	int ret = -ENOMEM;
983
	u32 i;
984

985
	/* Allocate this now to avoid having to handle failure below */
A
Alex Elder 已提交
986

987 988
	if (first_time) {
		size_t len;
989

990 991 992 993 994 995 996 997
		len = strnlen(ondisk->object_prefix,
				sizeof (ondisk->object_prefix));
		object_prefix = kmalloc(len + 1, GFP_KERNEL);
		if (!object_prefix)
			return -ENOMEM;
		memcpy(object_prefix, ondisk->object_prefix, len);
		object_prefix[len] = '\0';
	}
A
Alex Elder 已提交
998

999
	/* Allocate the snapshot context and fill it in */
A
Alex Elder 已提交
1000

1001 1002 1003 1004 1005
	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);
1006
	if (snap_count) {
1007
		struct rbd_image_snap_ondisk *snaps;
A
Alex Elder 已提交
1008 1009
		u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len);

1010
		/* We'll keep a copy of the snapshot names... */
1011

1012 1013 1014 1015
		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
1016 1017
			goto out_err;

1018
		/* ...as well as the array of their sizes. */
1019

1020
		size = snap_count * sizeof (*header->snap_sizes);
1021 1022
		snap_sizes = kmalloc(size, GFP_KERNEL);
		if (!snap_sizes)
A
Alex Elder 已提交
1023
			goto out_err;
1024

A
Alex Elder 已提交
1025
		/*
1026 1027 1028
		 * Copy the names, and fill in each snapshot's id
		 * and size.
		 *
1029
		 * Note that rbd_dev_v1_header_info() guarantees the
1030
		 * ondisk buffer we're working with has
A
Alex Elder 已提交
1031 1032 1033
		 * snap_names_len bytes beyond the end of the
		 * snapshot id array, this memcpy() is safe.
		 */
1034 1035 1036 1037 1038 1039
		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);
		}
1040
	}
A
Alex Elder 已提交
1041

1042
	/* We won't fail any more, fill in the header */
1043

1044 1045 1046 1047 1048 1049 1050 1051 1052
	if (first_time) {
		header->object_prefix = object_prefix;
		header->obj_order = ondisk->options.order;
		header->crypt_type = ondisk->options.crypt_type;
		header->comp_type = ondisk->options.comp_type;
		/* The rest aren't used for format 1 images */
		header->stripe_unit = 0;
		header->stripe_count = 0;
		header->features = 0;
1053
	} else {
A
Alex Elder 已提交
1054 1055 1056
		ceph_put_snap_context(header->snapc);
		kfree(header->snap_names);
		kfree(header->snap_sizes);
1057
	}
1058

1059
	/* The remaining fields always get updated (when we refresh) */
1060

A
Alex Elder 已提交
1061
	header->image_size = le64_to_cpu(ondisk->image_size);
1062 1063 1064
	header->snapc = snapc;
	header->snap_names = snap_names;
	header->snap_sizes = snap_sizes;
1065

1066
	return 0;
1067 1068
out_2big:
	ret = -EIO;
A
Alex Elder 已提交
1069
out_err:
1070 1071 1072 1073
	kfree(snap_sizes);
	kfree(snap_names);
	ceph_put_snap_context(snapc);
	kfree(object_prefix);
1074

1075
	return ret;
1076 1077
}

1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092
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);
}

1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116
/*
 * 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.
 */
1117 1118 1119
static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id)
{
	struct ceph_snap_context *snapc = rbd_dev->header.snapc;
1120
	u64 *found;
1121

1122 1123
	found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps,
				sizeof (snap_id), snapid_compare_reverse);
1124

1125
	return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX;
1126 1127
}

1128 1129
static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev,
					u64 snap_id)
1130
{
1131
	u32 which;
1132
	const char *snap_name;
1133

1134 1135
	which = rbd_dev_snap_index(rbd_dev, snap_id);
	if (which == BAD_SNAP_INDEX)
1136
		return ERR_PTR(-ENOENT);
1137

1138 1139
	snap_name = _rbd_dev_v1_snap_name(rbd_dev, which);
	return snap_name ? snap_name : ERR_PTR(-ENOMEM);
1140 1141 1142 1143
}

static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id)
{
1144 1145 1146
	if (snap_id == CEPH_NOSNAP)
		return RBD_SNAP_HEAD_NAME;

1147 1148 1149
	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);
1150

1151
	return rbd_dev_v2_snap_name(rbd_dev, snap_id);
1152 1153
}

1154 1155
static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
				u64 *snap_size)
1156
{
1157 1158 1159 1160 1161
	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;
1162

1163 1164 1165
		which = rbd_dev_snap_index(rbd_dev, snap_id);
		if (which == BAD_SNAP_INDEX)
			return -ENOENT;
1166

1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178
		*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;
1179 1180
}

1181 1182
static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
			u64 *snap_features)
1183
{
1184 1185 1186 1187 1188
	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 */
1189
	} else {
1190 1191
		u64 features = 0;
		int ret;
1192

1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203
		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)
{
1204
	u64 snap_id = rbd_dev->spec->snap_id;
1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218
	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;

1219
	return 0;
1220 1221
}

A
Alex Elder 已提交
1222 1223 1224 1225
static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev)
{
	rbd_dev->mapping.size = 0;
	rbd_dev->mapping.features = 0;
1226 1227
}

1228 1229 1230 1231 1232 1233 1234
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
1235
static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset)
1236
{
A
Alex Elder 已提交
1237 1238 1239
	char *name;
	u64 segment;
	int ret;
1240
	char *name_format;
1241

1242
	name = kmem_cache_alloc(rbd_segment_name_cache, GFP_NOIO);
A
Alex Elder 已提交
1243 1244 1245
	if (!name)
		return NULL;
	segment = offset >> rbd_dev->header.obj_order;
1246 1247 1248
	name_format = "%s.%012llx";
	if (rbd_dev->image_format == 2)
		name_format = "%s.%016llx";
1249
	ret = snprintf(name, CEPH_MAX_OID_NAME_LEN + 1, name_format,
A
Alex Elder 已提交
1250
			rbd_dev->header.object_prefix, segment);
1251
	if (ret < 0 || ret > CEPH_MAX_OID_NAME_LEN) {
A
Alex Elder 已提交
1252 1253
		pr_err("error formatting segment name for #%llu (%d)\n",
			segment, ret);
1254
		rbd_segment_name_free(name);
A
Alex Elder 已提交
1255 1256
		name = NULL;
	}
1257

A
Alex Elder 已提交
1258 1259
	return name;
}
1260

A
Alex Elder 已提交
1261 1262 1263
static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset)
{
	u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
1264

A
Alex Elder 已提交
1265 1266 1267 1268 1269 1270 1271 1272 1273 1274
	return offset & (segment_size - 1);
}

static u64 rbd_segment_length(struct rbd_device *rbd_dev,
				u64 offset, u64 length)
{
	u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;

	offset &= segment_size - 1;

A
Alex Elder 已提交
1275
	rbd_assert(length <= U64_MAX - offset);
A
Alex Elder 已提交
1276 1277 1278 1279
	if (offset + length > segment_size)
		length = segment_size - offset;

	return length;
1280 1281
}

1282 1283 1284 1285 1286 1287 1288 1289
/*
 * returns the size of an object in the image
 */
static u64 rbd_obj_bytes(struct rbd_image_header *header)
{
	return 1 << header->obj_order;
}

1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309
/*
 * 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)
{
1310 1311
	struct bio_vec bv;
	struct bvec_iter iter;
1312 1313 1314 1315 1316
	unsigned long flags;
	void *buf;
	int pos = 0;

	while (chain) {
1317 1318
		bio_for_each_segment(bv, chain, iter) {
			if (pos + bv.bv_len > start_ofs) {
1319
				int remainder = max(start_ofs - pos, 0);
1320
				buf = bvec_kmap_irq(&bv, &flags);
1321
				memset(buf + remainder, 0,
1322 1323
				       bv.bv_len - remainder);
				flush_dcache_page(bv.bv_page);
1324
				bvec_kunmap_irq(buf, &flags);
1325
			}
1326
			pos += bv.bv_len;
1327 1328 1329 1330 1331 1332
		}

		chain = chain->bi_next;
	}
}

A
Alex Elder 已提交
1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350
/*
 * 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;

1351 1352
		page_offset = offset & ~PAGE_MASK;
		length = min_t(size_t, PAGE_SIZE - page_offset, end - offset);
A
Alex Elder 已提交
1353 1354 1355
		local_irq_save(flags);
		kaddr = kmap_atomic(*page);
		memset(kaddr + page_offset, 0, length);
1356
		flush_dcache_page(*page);
A
Alex Elder 已提交
1357 1358 1359 1360 1361 1362 1363 1364
		kunmap_atomic(kaddr);
		local_irq_restore(flags);

		offset += length;
		page++;
	}
}

1365
/*
A
Alex Elder 已提交
1366 1367
 * Clone a portion of a bio, starting at the given byte offset
 * and continuing for the number of bytes indicated.
1368
 */
A
Alex Elder 已提交
1369 1370 1371 1372
static struct bio *bio_clone_range(struct bio *bio_src,
					unsigned int offset,
					unsigned int len,
					gfp_t gfpmask)
1373
{
A
Alex Elder 已提交
1374 1375
	struct bio *bio;

K
Kent Overstreet 已提交
1376
	bio = bio_clone(bio_src, gfpmask);
A
Alex Elder 已提交
1377 1378
	if (!bio)
		return NULL;	/* ENOMEM */
1379

K
Kent Overstreet 已提交
1380
	bio_advance(bio, offset);
1381
	bio->bi_iter.bi_size = len;
A
Alex Elder 已提交
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 1407 1408 1409 1410 1411

	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 */

1412
	if (!bi || off >= bi->bi_iter.bi_size || !len)
A
Alex Elder 已提交
1413
		return NULL;		/* Nothing to clone */
1414

A
Alex Elder 已提交
1415 1416 1417 1418 1419
	end = &chain;
	while (len) {
		unsigned int bi_size;
		struct bio *bio;

1420 1421
		if (!bi) {
			rbd_warn(NULL, "bio_chain exhausted with %u left", len);
A
Alex Elder 已提交
1422
			goto out_err;	/* EINVAL; ran out of bio's */
1423
		}
1424
		bi_size = min_t(unsigned int, bi->bi_iter.bi_size - off, len);
A
Alex Elder 已提交
1425 1426 1427 1428 1429 1430
		bio = bio_clone_range(bi, off, bi_size, gfpmask);
		if (!bio)
			goto out_err;	/* ENOMEM */

		*end = bio;
		end = &bio->bi_next;
1431

A
Alex Elder 已提交
1432
		off += bi_size;
1433
		if (off == bi->bi_iter.bi_size) {
A
Alex Elder 已提交
1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444
			bi = bi->bi_next;
			off = 0;
		}
		len -= bi_size;
	}
	*bio_src = bi;
	*offset = off;

	return chain;
out_err:
	bio_chain_put(chain);
1445 1446 1447 1448

	return NULL;
}

1449 1450 1451 1452 1453
/*
 * 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
1454
static void obj_request_img_data_set(struct rbd_obj_request *obj_request)
1455
{
A
Alex Elder 已提交
1456
	if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) {
1457 1458
		struct rbd_device *rbd_dev;

A
Alex Elder 已提交
1459
		rbd_dev = obj_request->img_request->rbd_dev;
1460
		rbd_warn(rbd_dev, "obj_request %p already marked img_data",
1461 1462 1463 1464
			obj_request);
	}
}

A
Alex Elder 已提交
1465
static bool obj_request_img_data_test(struct rbd_obj_request *obj_request)
1466 1467
{
	smp_mb();
A
Alex Elder 已提交
1468
	return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0;
1469 1470
}

A
Alex Elder 已提交
1471
static void obj_request_done_set(struct rbd_obj_request *obj_request)
1472
{
A
Alex Elder 已提交
1473 1474
	if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) {
		struct rbd_device *rbd_dev = NULL;
1475

A
Alex Elder 已提交
1476 1477
		if (obj_request_img_data_test(obj_request))
			rbd_dev = obj_request->img_request->rbd_dev;
1478
		rbd_warn(rbd_dev, "obj_request %p already marked done",
1479 1480 1481 1482
			obj_request);
	}
}

A
Alex Elder 已提交
1483
static bool obj_request_done_test(struct rbd_obj_request *obj_request)
1484 1485
{
	smp_mb();
A
Alex Elder 已提交
1486
	return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0;
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 1515 1516 1517 1518 1519
/*
 * 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;
}

1520 1521 1522 1523 1524 1525 1526 1527
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
1528 1529
static void rbd_obj_request_get(struct rbd_obj_request *obj_request)
{
A
Alex Elder 已提交
1530 1531
	dout("%s: obj %p (was %d)\n", __func__, obj_request,
		atomic_read(&obj_request->kref.refcount));
A
Alex Elder 已提交
1532 1533 1534 1535 1536 1537 1538
	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
1539 1540
	dout("%s: obj %p (was %d)\n", __func__, obj_request,
		atomic_read(&obj_request->kref.refcount));
A
Alex Elder 已提交
1541 1542 1543
	kref_put(&obj_request->kref, rbd_obj_request_destroy);
}

1544 1545 1546 1547 1548 1549 1550
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);
}

1551 1552
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
1553 1554 1555 1556
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
1557 1558
	dout("%s: img %p (was %d)\n", __func__, img_request,
		atomic_read(&img_request->kref.refcount));
1559 1560 1561 1562
	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
1563 1564 1565 1566 1567
}

static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request,
					struct rbd_obj_request *obj_request)
{
1568 1569
	rbd_assert(obj_request->img_request == NULL);

1570
	/* Image request now owns object's original reference */
A
Alex Elder 已提交
1571
	obj_request->img_request = img_request;
1572
	obj_request->which = img_request->obj_request_count;
1573 1574
	rbd_assert(!obj_request_img_data_test(obj_request));
	obj_request_img_data_set(obj_request);
A
Alex Elder 已提交
1575
	rbd_assert(obj_request->which != BAD_WHICH);
1576 1577
	img_request->obj_request_count++;
	list_add_tail(&obj_request->links, &img_request->obj_requests);
A
Alex Elder 已提交
1578 1579
	dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
		obj_request->which);
A
Alex Elder 已提交
1580 1581 1582 1583 1584 1585
}

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);
1586

A
Alex Elder 已提交
1587 1588
	dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
		obj_request->which);
A
Alex Elder 已提交
1589
	list_del(&obj_request->links);
1590 1591 1592 1593
	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;
1594
	rbd_assert(obj_request_img_data_test(obj_request));
A
Alex Elder 已提交
1595 1596
	rbd_assert(obj_request->img_request == img_request);
	obj_request->img_request = NULL;
1597
	obj_request->callback = NULL;
A
Alex Elder 已提交
1598 1599 1600 1601 1602 1603
	rbd_obj_request_put(obj_request);
}

static bool obj_request_type_valid(enum obj_request_type type)
{
	switch (type) {
1604
	case OBJ_REQUEST_NODATA:
A
Alex Elder 已提交
1605
	case OBJ_REQUEST_BIO:
1606
	case OBJ_REQUEST_PAGES:
A
Alex Elder 已提交
1607 1608 1609 1610 1611 1612 1613 1614 1615
		return true;
	default:
		return false;
	}
}

static int rbd_obj_request_submit(struct ceph_osd_client *osdc,
				struct rbd_obj_request *obj_request)
{
1616
	dout("%s %p\n", __func__, obj_request);
A
Alex Elder 已提交
1617 1618 1619
	return ceph_osdc_start_request(osdc, obj_request->osd_req, false);
}

1620 1621 1622 1623 1624 1625 1626 1627 1628
static void rbd_obj_request_end(struct rbd_obj_request *obj_request)
{
	dout("%s %p\n", __func__, obj_request);
	ceph_osdc_cancel_request(obj_request->osd_req);
}

/*
 * Wait for an object request to complete.  If interrupted, cancel the
 * underlying osd request.
1629 1630
 *
 * @timeout: in jiffies, 0 means "wait forever"
1631
 */
1632 1633
static int __rbd_obj_request_wait(struct rbd_obj_request *obj_request,
				  unsigned long timeout)
1634
{
1635
	long ret;
1636 1637

	dout("%s %p\n", __func__, obj_request);
1638 1639 1640 1641 1642 1643
	ret = wait_for_completion_interruptible_timeout(
					&obj_request->completion,
					ceph_timeout_jiffies(timeout));
	if (ret <= 0) {
		if (ret == 0)
			ret = -ETIMEDOUT;
1644
		rbd_obj_request_end(obj_request);
1645 1646
	} else {
		ret = 0;
1647 1648
	}

1649 1650 1651 1652 1653 1654 1655 1656 1657
	dout("%s %p ret %d\n", __func__, obj_request, (int)ret);
	return ret;
}

static int rbd_obj_request_wait(struct rbd_obj_request *obj_request)
{
	return __rbd_obj_request_wait(obj_request, 0);
}

A
Alex Elder 已提交
1658 1659
static void rbd_img_request_complete(struct rbd_img_request *img_request)
{
1660

A
Alex Elder 已提交
1661
	dout("%s: img %p\n", __func__, img_request);
1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677

	/*
	 * 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
1678 1679 1680 1681 1682 1683
	if (img_request->callback)
		img_request->callback(img_request);
	else
		rbd_img_request_put(img_request);
}

A
Alex Elder 已提交
1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700
/*
 * 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;
}

1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715
/*
 * 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;
}

1716 1717 1718 1719 1720 1721
static void img_request_child_set(struct rbd_img_request *img_request)
{
	set_bit(IMG_REQ_CHILD, &img_request->flags);
	smp_mb();
}

1722 1723 1724 1725 1726 1727
static void img_request_child_clear(struct rbd_img_request *img_request)
{
	clear_bit(IMG_REQ_CHILD, &img_request->flags);
	smp_mb();
}

1728 1729 1730 1731 1732 1733
static bool img_request_child_test(struct rbd_img_request *img_request)
{
	smp_mb();
	return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0;
}

1734 1735 1736 1737 1738 1739
static void img_request_layered_set(struct rbd_img_request *img_request)
{
	set_bit(IMG_REQ_LAYERED, &img_request->flags);
	smp_mb();
}

1740 1741 1742 1743 1744 1745
static void img_request_layered_clear(struct rbd_img_request *img_request)
{
	clear_bit(IMG_REQ_LAYERED, &img_request->flags);
	smp_mb();
}

1746 1747 1748 1749 1750 1751
static bool img_request_layered_test(struct rbd_img_request *img_request)
{
	smp_mb();
	return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0;
}

1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762
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;
}

1763 1764 1765
static void
rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request)
{
A
Alex Elder 已提交
1766 1767 1768
	u64 xferred = obj_request->xferred;
	u64 length = obj_request->length;

1769 1770
	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
1771
		xferred, length);
1772
	/*
1773 1774 1775 1776 1777 1778
	 * 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.
1779
	 */
A
Alex Elder 已提交
1780
	rbd_assert(obj_request->type != OBJ_REQUEST_NODATA);
1781
	if (obj_request->result == -ENOENT) {
A
Alex Elder 已提交
1782 1783 1784 1785
		if (obj_request->type == OBJ_REQUEST_BIO)
			zero_bio_chain(obj_request->bio_list, 0);
		else
			zero_pages(obj_request->pages, 0, length);
1786
		obj_request->result = 0;
A
Alex Elder 已提交
1787 1788 1789 1790 1791
	} 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);
1792
	}
1793
	obj_request->xferred = length;
1794 1795 1796
	obj_request_done_set(obj_request);
}

A
Alex Elder 已提交
1797 1798
static void rbd_obj_request_complete(struct rbd_obj_request *obj_request)
{
A
Alex Elder 已提交
1799 1800
	dout("%s: obj %p cb %p\n", __func__, obj_request,
		obj_request->callback);
A
Alex Elder 已提交
1801 1802
	if (obj_request->callback)
		obj_request->callback(obj_request);
1803 1804
	else
		complete_all(&obj_request->completion);
A
Alex Elder 已提交
1805 1806
}

1807
static void rbd_osd_read_callback(struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1808
{
A
Alex Elder 已提交
1809
	struct rbd_img_request *img_request = NULL;
A
Alex Elder 已提交
1810
	struct rbd_device *rbd_dev = NULL;
A
Alex Elder 已提交
1811 1812 1813 1814 1815
	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
1816
		rbd_dev = img_request->rbd_dev;
A
Alex Elder 已提交
1817
	}
A
Alex Elder 已提交
1818 1819 1820 1821

	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
1822 1823
	if (layered && obj_request->result == -ENOENT &&
			obj_request->img_offset < rbd_dev->parent_overlap)
A
Alex Elder 已提交
1824 1825
		rbd_img_parent_read(obj_request);
	else if (img_request)
1826 1827 1828
		rbd_img_obj_request_read_callback(obj_request);
	else
		obj_request_done_set(obj_request);
A
Alex Elder 已提交
1829 1830
}

1831
static void rbd_osd_write_callback(struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1832
{
1833 1834 1835
	dout("%s: obj %p result %d %llu\n", __func__, obj_request,
		obj_request->result, obj_request->length);
	/*
A
Alex Elder 已提交
1836 1837
	 * There is no such thing as a successful short write.  Set
	 * it to our originally-requested length.
1838 1839
	 */
	obj_request->xferred = obj_request->length;
1840
	obj_request_done_set(obj_request);
A
Alex Elder 已提交
1841 1842
}

1843 1844 1845 1846 1847 1848 1849 1850 1851
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;
1852 1853 1854
	/* discarding a non-existent object is not a problem */
	if (obj_request->result == -ENOENT)
		obj_request->result = 0;
1855 1856 1857
	obj_request_done_set(obj_request);
}

A
Alex Elder 已提交
1858 1859 1860 1861
/*
 * 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.
 */
1862
static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1863
{
A
Alex Elder 已提交
1864
	dout("%s: obj %p\n", __func__, obj_request);
A
Alex Elder 已提交
1865 1866 1867
	obj_request_done_set(obj_request);
}

I
Ilya Dryomov 已提交
1868 1869 1870 1871 1872 1873 1874 1875 1876 1877
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);
}

1878
static void rbd_osd_req_callback(struct ceph_osd_request *osd_req)
A
Alex Elder 已提交
1879 1880 1881 1882
{
	struct rbd_obj_request *obj_request = osd_req->r_priv;
	u16 opcode;

1883
	dout("%s: osd_req %p\n", __func__, osd_req);
A
Alex Elder 已提交
1884
	rbd_assert(osd_req == obj_request->osd_req);
A
Alex Elder 已提交
1885 1886 1887 1888 1889 1890
	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
1891

1892 1893
	if (osd_req->r_result < 0)
		obj_request->result = osd_req->r_result;
A
Alex Elder 已提交
1894

1895 1896
	/*
	 * We support a 64-bit length, but ultimately it has to be
C
Christoph Hellwig 已提交
1897 1898
	 * passed to the block layer, which just supports a 32-bit
	 * length field.
1899
	 */
1900
	obj_request->xferred = osd_req->r_ops[0].outdata_len;
A
Alex Elder 已提交
1901
	rbd_assert(obj_request->xferred < (u64)UINT_MAX);
1902

1903
	opcode = osd_req->r_ops[0].op;
A
Alex Elder 已提交
1904 1905
	switch (opcode) {
	case CEPH_OSD_OP_READ:
1906
		rbd_osd_read_callback(obj_request);
A
Alex Elder 已提交
1907
		break;
1908
	case CEPH_OSD_OP_SETALLOCHINT:
1909 1910
		rbd_assert(osd_req->r_ops[1].op == CEPH_OSD_OP_WRITE ||
			   osd_req->r_ops[1].op == CEPH_OSD_OP_WRITEFULL);
1911
		/* fall through */
A
Alex Elder 已提交
1912
	case CEPH_OSD_OP_WRITE:
1913
	case CEPH_OSD_OP_WRITEFULL:
1914
		rbd_osd_write_callback(obj_request);
A
Alex Elder 已提交
1915
		break;
A
Alex Elder 已提交
1916
	case CEPH_OSD_OP_STAT:
1917
		rbd_osd_stat_callback(obj_request);
A
Alex Elder 已提交
1918
		break;
1919 1920 1921 1922 1923
	case CEPH_OSD_OP_DELETE:
	case CEPH_OSD_OP_TRUNCATE:
	case CEPH_OSD_OP_ZERO:
		rbd_osd_discard_callback(obj_request);
		break;
1924
	case CEPH_OSD_OP_CALL:
I
Ilya Dryomov 已提交
1925 1926
		rbd_osd_call_callback(obj_request);
		break;
A
Alex Elder 已提交
1927
	default:
1928
		rbd_warn(NULL, "%s: unsupported op %hu",
A
Alex Elder 已提交
1929 1930 1931 1932
			obj_request->object_name, (unsigned short) opcode);
		break;
	}

1933
	if (obj_request_done_test(obj_request))
A
Alex Elder 已提交
1934 1935 1936
		rbd_obj_request_complete(obj_request);
}

1937
static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1938 1939
{
	struct rbd_img_request *img_request = obj_request->img_request;
1940
	struct ceph_osd_request *osd_req = obj_request->osd_req;
A
Alex Elder 已提交
1941

1942 1943
	if (img_request)
		osd_req->r_snapid = img_request->snap_id;
1944 1945 1946 1947 1948 1949
}

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

1950 1951
	osd_req->r_mtime = CURRENT_TIME;
	osd_req->r_data_offset = obj_request->offset;
A
Alex Elder 已提交
1952 1953
}

1954 1955 1956 1957 1958 1959
/*
 * 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
1960 1961
static struct ceph_osd_request *rbd_osd_req_create(
					struct rbd_device *rbd_dev,
G
Guangliang Zhao 已提交
1962
					enum obj_operation_type op_type,
1963
					unsigned int num_ops,
A
Alex Elder 已提交
1964
					struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1965 1966 1967 1968 1969
{
	struct ceph_snap_context *snapc = NULL;
	struct ceph_osd_client *osdc;
	struct ceph_osd_request *osd_req;

1970 1971
	if (obj_request_img_data_test(obj_request) &&
		(op_type == OBJ_OP_DISCARD || op_type == OBJ_OP_WRITE)) {
1972
		struct rbd_img_request *img_request = obj_request->img_request;
1973 1974 1975 1976 1977
		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
1978
		snapc = img_request->snapc;
A
Alex Elder 已提交
1979 1980
	}

G
Guangliang Zhao 已提交
1981
	rbd_assert(num_ops == 1 || ((op_type == OBJ_OP_WRITE) && num_ops == 2));
1982 1983

	/* Allocate and initialize the request, for the num_ops ops */
A
Alex Elder 已提交
1984 1985

	osdc = &rbd_dev->rbd_client->client->osdc;
1986
	osd_req = ceph_osdc_alloc_request(osdc, snapc, num_ops, false,
1987
					  GFP_NOIO);
A
Alex Elder 已提交
1988
	if (!osd_req)
1989
		goto fail;
A
Alex Elder 已提交
1990

1991
	if (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD)
A
Alex Elder 已提交
1992
		osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
A
Alex Elder 已提交
1993
	else
A
Alex Elder 已提交
1994 1995 1996 1997 1998
		osd_req->r_flags = CEPH_OSD_FLAG_READ;

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

1999
	osd_req->r_base_oloc.pool = rbd_dev->layout.pool_id;
2000 2001 2002
	if (ceph_oid_aprintf(&osd_req->r_base_oid, GFP_NOIO, "%s",
			     obj_request->object_name))
		goto fail;
A
Alex Elder 已提交
2003

2004 2005 2006
	if (ceph_osdc_alloc_messages(osd_req, GFP_NOIO))
		goto fail;

A
Alex Elder 已提交
2007
	return osd_req;
2008 2009 2010 2011

fail:
	ceph_osdc_put_request(osd_req);
	return NULL;
A
Alex Elder 已提交
2012 2013
}

2014
/*
2015 2016 2017 2018
 * 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.
2019 2020 2021 2022 2023 2024 2025 2026 2027
 */
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;
2028
	int num_osd_ops = 3;
2029 2030 2031 2032

	rbd_assert(obj_request_img_data_test(obj_request));
	img_request = obj_request->img_request;
	rbd_assert(img_request);
2033 2034
	rbd_assert(img_request_write_test(img_request) ||
			img_request_discard_test(img_request));
2035

2036 2037 2038 2039
	if (img_request_discard_test(img_request))
		num_osd_ops = 2;

	/* Allocate and initialize the request, for all the ops */
2040 2041 2042 2043

	snapc = img_request->snapc;
	rbd_dev = img_request->rbd_dev;
	osdc = &rbd_dev->rbd_client->client->osdc;
2044
	osd_req = ceph_osdc_alloc_request(osdc, snapc, num_osd_ops,
2045
						false, GFP_NOIO);
2046
	if (!osd_req)
2047
		goto fail;
2048 2049 2050 2051 2052

	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;

2053
	osd_req->r_base_oloc.pool = rbd_dev->layout.pool_id;
2054 2055 2056
	if (ceph_oid_aprintf(&osd_req->r_base_oid, GFP_NOIO, "%s",
			     obj_request->object_name))
		goto fail;
2057

2058 2059 2060
	if (ceph_osdc_alloc_messages(osd_req, GFP_NOIO))
		goto fail;

2061
	return osd_req;
2062 2063 2064 2065

fail:
	ceph_osdc_put_request(osd_req);
	return NULL;
2066 2067 2068
}


A
Alex Elder 已提交
2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086
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;
2087
	name = kmalloc(size, GFP_NOIO);
2088
	if (!name)
A
Alex Elder 已提交
2089 2090
		return NULL;

2091
	obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_NOIO);
2092 2093 2094 2095 2096
	if (!obj_request) {
		kfree(name);
		return NULL;
	}

A
Alex Elder 已提交
2097 2098 2099
	obj_request->object_name = memcpy(name, object_name, size);
	obj_request->offset = offset;
	obj_request->length = length;
2100
	obj_request->flags = 0;
A
Alex Elder 已提交
2101 2102 2103
	obj_request->which = BAD_WHICH;
	obj_request->type = type;
	INIT_LIST_HEAD(&obj_request->links);
2104
	init_completion(&obj_request->completion);
A
Alex Elder 已提交
2105 2106
	kref_init(&obj_request->kref);

A
Alex Elder 已提交
2107 2108 2109
	dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name,
		offset, length, (int)type, obj_request);

A
Alex Elder 已提交
2110 2111 2112 2113 2114 2115 2116 2117 2118
	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
2119 2120
	dout("%s: obj %p\n", __func__, obj_request);

A
Alex Elder 已提交
2121 2122 2123 2124 2125 2126 2127 2128
	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) {
2129 2130
	case OBJ_REQUEST_NODATA:
		break;		/* Nothing to do */
A
Alex Elder 已提交
2131 2132 2133 2134
	case OBJ_REQUEST_BIO:
		if (obj_request->bio_list)
			bio_chain_put(obj_request->bio_list);
		break;
2135 2136 2137 2138 2139
	case OBJ_REQUEST_PAGES:
		if (obj_request->pages)
			ceph_release_page_vector(obj_request->pages,
						obj_request->page_count);
		break;
A
Alex Elder 已提交
2140 2141
	}

2142
	kfree(obj_request->object_name);
2143 2144
	obj_request->object_name = NULL;
	kmem_cache_free(rbd_obj_request_cache, obj_request);
A
Alex Elder 已提交
2145 2146
}

A
Alex Elder 已提交
2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157
/* 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;
}

2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179
/*
 * 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
2180
		rbd_warn(rbd_dev, "parent reference underflow");
2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192
}

/*
 * 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)
{
2193
	int counter = 0;
2194 2195 2196 2197

	if (!rbd_dev->parent_spec)
		return false;

2198 2199 2200 2201
	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);
2202 2203

	if (counter < 0)
2204
		rbd_warn(rbd_dev, "parent reference overflow");
2205

2206
	return counter > 0;
2207 2208
}

A
Alex Elder 已提交
2209 2210 2211 2212 2213
/*
 * 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
2214 2215
static struct rbd_img_request *rbd_img_request_create(
					struct rbd_device *rbd_dev,
A
Alex Elder 已提交
2216
					u64 offset, u64 length,
G
Guangliang Zhao 已提交
2217
					enum obj_operation_type op_type,
2218
					struct ceph_snap_context *snapc)
A
Alex Elder 已提交
2219 2220 2221
{
	struct rbd_img_request *img_request;

2222
	img_request = kmem_cache_alloc(rbd_img_request_cache, GFP_NOIO);
A
Alex Elder 已提交
2223 2224 2225 2226 2227 2228 2229
	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
2230
	img_request->flags = 0;
2231 2232 2233 2234
	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
2235
		img_request_write_set(img_request);
2236
		img_request->snapc = snapc;
A
Alex Elder 已提交
2237
	} else {
A
Alex Elder 已提交
2238
		img_request->snap_id = rbd_dev->spec->snap_id;
A
Alex Elder 已提交
2239
	}
2240
	if (rbd_dev_parent_get(rbd_dev))
2241
		img_request_layered_set(img_request);
A
Alex Elder 已提交
2242 2243 2244
	spin_lock_init(&img_request->completion_lock);
	img_request->next_completion = 0;
	img_request->callback = NULL;
2245
	img_request->result = 0;
A
Alex Elder 已提交
2246 2247 2248 2249
	img_request->obj_request_count = 0;
	INIT_LIST_HEAD(&img_request->obj_requests);
	kref_init(&img_request->kref);

A
Alex Elder 已提交
2250
	dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev,
G
Guangliang Zhao 已提交
2251
		obj_op_name(op_type), offset, length, img_request);
A
Alex Elder 已提交
2252

A
Alex Elder 已提交
2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263
	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
2264 2265
	dout("%s: img %p\n", __func__, img_request);

A
Alex Elder 已提交
2266 2267
	for_each_obj_request_safe(img_request, obj_request, next_obj_request)
		rbd_img_obj_request_del(img_request, obj_request);
2268
	rbd_assert(img_request->obj_request_count == 0);
A
Alex Elder 已提交
2269

2270 2271 2272 2273 2274
	if (img_request_layered_test(img_request)) {
		img_request_layered_clear(img_request);
		rbd_dev_parent_put(img_request->rbd_dev);
	}

2275 2276
	if (img_request_write_test(img_request) ||
		img_request_discard_test(img_request))
2277
		ceph_put_snap_context(img_request->snapc);
A
Alex Elder 已提交
2278

2279
	kmem_cache_free(rbd_img_request_cache, img_request);
A
Alex Elder 已提交
2280 2281
}

2282 2283 2284 2285 2286 2287 2288 2289 2290 2291
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;

2292
	parent_request = rbd_img_request_create(rbd_dev->parent, img_offset,
G
Guangliang Zhao 已提交
2293
						length, OBJ_OP_READ, NULL);
2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318
	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);
}

2319 2320
static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request)
{
2321
	struct rbd_img_request *img_request;
2322 2323
	unsigned int xferred;
	int result;
A
Alex Elder 已提交
2324
	bool more;
2325

2326 2327 2328
	rbd_assert(obj_request_img_data_test(obj_request));
	img_request = obj_request->img_request;

2329 2330 2331 2332 2333
	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
2334 2335
		enum obj_operation_type op_type;

2336 2337 2338 2339 2340 2341
		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;
2342

2343
		rbd_warn(rbd_dev, "%s %llx at %llx (%llx)",
G
Guangliang Zhao 已提交
2344 2345
			obj_op_name(op_type), obj_request->length,
			obj_request->img_offset, obj_request->offset);
2346
		rbd_warn(rbd_dev, "  result %d xferred %x",
2347 2348 2349
			result, xferred);
		if (!img_request->result)
			img_request->result = result;
2350 2351 2352 2353 2354
		/*
		 * Need to end I/O on the entire obj_request worth of
		 * bytes in case of error.
		 */
		xferred = obj_request->length;
2355 2356
	}

2357 2358 2359 2360 2361 2362 2363
	/* Image object requests don't own their page array */

	if (obj_request->type == OBJ_REQUEST_PAGES) {
		obj_request->pages = NULL;
		obj_request->page_count = 0;
	}

A
Alex Elder 已提交
2364 2365 2366 2367 2368
	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
2369 2370 2371 2372

		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
2373 2374 2375
	}

	return more;
2376 2377
}

2378 2379 2380 2381 2382 2383
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;

2384
	rbd_assert(obj_request_img_data_test(obj_request));
2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402
	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;
2403
		more = rbd_img_obj_end_request(obj_request);
2404 2405 2406 2407 2408 2409 2410
		which++;
	}

	rbd_assert(more ^ (which == img_request->obj_request_count));
	img_request->next_completion = which;
out:
	spin_unlock_irq(&img_request->completion_lock);
2411
	rbd_img_request_put(img_request);
2412 2413 2414 2415 2416

	if (!more)
		rbd_img_request_complete(img_request);
}

2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435
/*
 * 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) {
2436 2437 2438
		if (!offset && length == object_size &&
		    (!img_request_layered_test(img_request) ||
		     !obj_request_overlaps_parent(obj_request))) {
2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452
			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) {
2453 2454 2455 2456
		if (!offset && length == object_size)
			opcode = CEPH_OSD_OP_WRITEFULL;
		else
			opcode = CEPH_OSD_OP_WRITE;
2457 2458 2459 2460 2461 2462 2463
		osd_req_op_alloc_hint_init(osd_request, num_ops,
					object_size, object_size);
		num_ops++;
	} else {
		opcode = CEPH_OSD_OP_READ;
	}

2464
	if (opcode == CEPH_OSD_OP_DELETE)
2465
		osd_req_op_init(osd_request, num_ops, opcode, 0);
2466 2467 2468 2469
	else
		osd_req_op_extent_init(osd_request, num_ops, opcode,
				       offset, length, 0, 0);

2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484
	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);
}

2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495
/*
 * 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
2496 2497 2498 2499
{
	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
2500
	struct bio *bio_list = NULL;
2501
	unsigned int bio_offset = 0;
J
Jingoo Han 已提交
2502
	struct page **pages = NULL;
G
Guangliang Zhao 已提交
2503
	enum obj_operation_type op_type;
2504
	u64 img_offset;
A
Alex Elder 已提交
2505 2506
	u64 resid;

2507 2508
	dout("%s: img %p type %d data_desc %p\n", __func__, img_request,
		(int)type, data_desc);
A
Alex Elder 已提交
2509

2510
	img_offset = img_request->offset;
A
Alex Elder 已提交
2511
	resid = img_request->length;
A
Alex Elder 已提交
2512
	rbd_assert(resid > 0);
2513
	op_type = rbd_img_request_op_type(img_request);
2514 2515 2516

	if (type == OBJ_REQUEST_BIO) {
		bio_list = data_desc;
2517 2518
		rbd_assert(img_offset ==
			   bio_list->bi_iter.bi_sector << SECTOR_SHIFT);
2519
	} else if (type == OBJ_REQUEST_PAGES) {
2520 2521 2522
		pages = data_desc;
	}

A
Alex Elder 已提交
2523
	while (resid) {
2524
		struct ceph_osd_request *osd_req;
A
Alex Elder 已提交
2525 2526 2527 2528
		const char *object_name;
		u64 offset;
		u64 length;

2529
		object_name = rbd_segment_name(rbd_dev, img_offset);
A
Alex Elder 已提交
2530 2531
		if (!object_name)
			goto out_unwind;
2532 2533
		offset = rbd_segment_offset(rbd_dev, img_offset);
		length = rbd_segment_length(rbd_dev, img_offset, resid);
A
Alex Elder 已提交
2534
		obj_request = rbd_obj_request_create(object_name,
2535
						offset, length, type);
2536 2537
		/* object request has its own copy of the object name */
		rbd_segment_name_free(object_name);
A
Alex Elder 已提交
2538 2539
		if (!obj_request)
			goto out_unwind;
2540

2541 2542 2543 2544 2545
		/*
		 * 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
2546

2547 2548 2549 2550 2551 2552 2553 2554 2555
		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,
2556
								GFP_NOIO);
2557
			if (!obj_request->bio_list)
2558
				goto out_unwind;
2559
		} else if (type == OBJ_REQUEST_PAGES) {
2560 2561 2562 2563 2564 2565 2566 2567 2568
			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
2569

G
Guangliang Zhao 已提交
2570 2571 2572
		osd_req = rbd_osd_req_create(rbd_dev, op_type,
					(op_type == OBJ_OP_WRITE) ? 2 : 1,
					obj_request);
2573
		if (!osd_req)
2574
			goto out_unwind;
2575

2576
		obj_request->osd_req = osd_req;
2577
		obj_request->callback = rbd_img_obj_callback;
2578
		obj_request->img_offset = img_offset;
2579

2580
		rbd_img_obj_request_fill(obj_request, osd_req, op_type, 0);
A
Alex Elder 已提交
2581

2582
		rbd_img_request_get(img_request);
A
Alex Elder 已提交
2583

2584
		img_offset += length;
A
Alex Elder 已提交
2585 2586 2587 2588 2589 2590 2591
		resid -= length;
	}

	return 0;

out_unwind:
	for_each_obj_request_safe(img_request, obj_request, next_obj_request)
2592
		rbd_img_obj_request_del(img_request, obj_request);
A
Alex Elder 已提交
2593 2594 2595 2596

	return -ENOMEM;
}

2597
static void
I
Ilya Dryomov 已提交
2598
rbd_osd_copyup_callback(struct rbd_obj_request *obj_request)
2599 2600 2601
{
	struct rbd_img_request *img_request;
	struct rbd_device *rbd_dev;
2602
	struct page **pages;
2603 2604
	u32 page_count;

I
Ilya Dryomov 已提交
2605 2606
	dout("%s: obj %p\n", __func__, obj_request);

2607 2608
	rbd_assert(obj_request->type == OBJ_REQUEST_BIO ||
		obj_request->type == OBJ_REQUEST_NODATA);
2609 2610 2611 2612 2613 2614 2615
	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);

2616 2617
	pages = obj_request->copyup_pages;
	rbd_assert(pages != NULL);
2618
	obj_request->copyup_pages = NULL;
2619 2620 2621 2622
	page_count = obj_request->copyup_page_count;
	rbd_assert(page_count);
	obj_request->copyup_page_count = 0;
	ceph_release_page_vector(pages, page_count);
2623 2624 2625 2626 2627 2628 2629 2630 2631 2632

	/*
	 * 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
2633
	obj_request_done_set(obj_request);
2634 2635
}

2636 2637 2638 2639
static void
rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request)
{
	struct rbd_obj_request *orig_request;
2640 2641 2642
	struct ceph_osd_request *osd_req;
	struct ceph_osd_client *osdc;
	struct rbd_device *rbd_dev;
2643
	struct page **pages;
2644
	enum obj_operation_type op_type;
2645
	u32 page_count;
2646
	int img_result;
2647
	u64 parent_length;
2648 2649 2650 2651 2652 2653 2654 2655

	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;
2656 2657 2658
	page_count = img_request->copyup_page_count;
	rbd_assert(page_count);
	img_request->copyup_page_count = 0;
2659 2660 2661

	orig_request = img_request->obj_request;
	rbd_assert(orig_request != NULL);
2662
	rbd_assert(obj_request_type_valid(orig_request->type));
2663
	img_result = img_request->result;
2664 2665
	parent_length = img_request->length;
	rbd_assert(parent_length == img_request->xferred);
2666
	rbd_img_request_put(img_request);
2667

2668 2669
	rbd_assert(orig_request->img_request);
	rbd_dev = orig_request->img_request->rbd_dev;
2670 2671
	rbd_assert(rbd_dev);

2672 2673 2674 2675 2676 2677 2678
	/*
	 * 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) {
		struct ceph_osd_client *osdc;
2679

2680 2681 2682 2683 2684 2685
		ceph_release_page_vector(pages, page_count);
		osdc = &rbd_dev->rbd_client->client->osdc;
		img_result = rbd_obj_request_submit(osdc, orig_request);
		if (!img_result)
			return;
	}
2686

2687
	if (img_result)
2688 2689
		goto out_err;

2690 2691
	/*
	 * The original osd request is of no use to use any more.
2692
	 * We need a new one that can hold the three ops in a copyup
2693 2694 2695
	 * request.  Allocate the new copyup osd request for the
	 * original request, and release the old one.
	 */
2696
	img_result = -ENOMEM;
2697 2698 2699
	osd_req = rbd_osd_req_create_copyup(orig_request);
	if (!osd_req)
		goto out_err;
2700
	rbd_osd_req_destroy(orig_request->osd_req);
2701 2702
	orig_request->osd_req = osd_req;
	orig_request->copyup_pages = pages;
2703
	orig_request->copyup_page_count = page_count;
2704

2705
	/* Initialize the copyup op */
2706

2707
	osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup");
2708
	osd_req_op_cls_request_data_pages(osd_req, 0, pages, parent_length, 0,
2709
						false, false);
2710

2711
	/* Add the other op(s) */
2712

2713 2714
	op_type = rbd_img_request_op_type(orig_request->img_request);
	rbd_img_obj_request_fill(orig_request, osd_req, op_type, 1);
2715 2716 2717 2718

	/* All set, send it off. */

	osdc = &rbd_dev->rbd_client->client->osdc;
2719 2720
	img_result = rbd_obj_request_submit(osdc, orig_request);
	if (!img_result)
2721 2722 2723 2724
		return;
out_err:
	/* Record the error code and complete the request */

2725
	orig_request->result = img_result;
2726 2727 2728
	orig_request->xferred = 0;
	obj_request_done_set(orig_request);
	rbd_obj_request_complete(orig_request);
2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756
}

/*
 * 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.
 *
 * If an error occurs, record it as the result of the original
 * object request and mark it done so it gets completed.
 */
static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request)
{
	struct rbd_img_request *img_request = NULL;
	struct rbd_img_request *parent_request = NULL;
	struct rbd_device *rbd_dev;
	u64 img_offset;
	u64 length;
	struct page **pages = NULL;
	u32 page_count;
	int result;

	rbd_assert(obj_request_img_data_test(obj_request));
2757
	rbd_assert(obj_request_type_valid(obj_request->type));
2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770

	img_request = obj_request->img_request;
	rbd_assert(img_request != NULL);
	rbd_dev = img_request->rbd_dev;
	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;
	length = (u64)1 << rbd_dev->header.obj_order;

A
Alex Elder 已提交
2771 2772 2773 2774 2775 2776 2777 2778 2779 2780
	/*
	 * 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;
	}

2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793
	/*
	 * 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;
2794 2795
	parent_request = rbd_parent_request_create(obj_request,
						img_offset, length);
2796 2797 2798 2799 2800 2801 2802
	if (!parent_request)
		goto out_err;

	result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages);
	if (result)
		goto out_err;
	parent_request->copyup_pages = pages;
2803
	parent_request->copyup_page_count = page_count;
2804 2805 2806 2807 2808 2809 2810

	parent_request->callback = rbd_img_obj_parent_read_full_callback;
	result = rbd_img_request_submit(parent_request);
	if (!result)
		return 0;

	parent_request->copyup_pages = NULL;
2811
	parent_request->copyup_page_count = 0;
2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825
	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);
	obj_request->result = result;
	obj_request->xferred = 0;
	obj_request_done_set(obj_request);

	return result;
}

2826 2827 2828
static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request)
{
	struct rbd_obj_request *orig_request;
2829
	struct rbd_device *rbd_dev;
2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840
	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;
2841
	rbd_obj_request_put(orig_request);
2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852
	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);

2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866
	/*
	 * 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.
	 */
	rbd_dev = orig_request->img_request->rbd_dev;
	if (!rbd_dev->parent_overlap) {
		struct ceph_osd_client *osdc;

		osdc = &rbd_dev->rbd_client->client->osdc;
		result = rbd_obj_request_submit(osdc, orig_request);
		if (!result)
			return;
	}
2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879

	/*
	 * 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);
	} else if (result) {
		orig_request->result = result;
2880
		goto out;
2881 2882 2883 2884 2885 2886
	}

	/*
	 * Resubmit the original request now that we have recorded
	 * whether the target object exists.
	 */
2887
	orig_request->result = rbd_img_obj_request_submit(orig_request);
2888
out:
2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929
	if (orig_request->result)
		rbd_obj_request_complete(orig_request);
}

static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request)
{
	struct rbd_obj_request *stat_request;
	struct rbd_device *rbd_dev;
	struct ceph_osd_client *osdc;
	struct page **pages = NULL;
	u32 page_count;
	size_t size;
	int ret;

	/*
	 * 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);
	if (IS_ERR(pages))
		return PTR_ERR(pages);

	ret = -ENOMEM;
	stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0,
							OBJ_REQUEST_PAGES);
	if (!stat_request)
		goto out;

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

	rbd_assert(obj_request->img_request);
	rbd_dev = obj_request->img_request->rbd_dev;
G
Guangliang Zhao 已提交
2930
	stat_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1,
2931
						   stat_request);
2932 2933 2934 2935
	if (!stat_request->osd_req)
		goto out;
	stat_request->callback = rbd_img_obj_exists_callback;

2936
	osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT, 0);
2937 2938
	osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0,
					false, false);
2939
	rbd_osd_req_format_read(stat_request);
2940 2941 2942 2943 2944 2945 2946 2947 2948 2949

	osdc = &rbd_dev->rbd_client->client->osdc;
	ret = rbd_obj_request_submit(osdc, stat_request);
out:
	if (ret)
		rbd_obj_request_put(obj_request);

	return ret;
}

2950
static bool img_obj_request_simple(struct rbd_obj_request *obj_request)
2951 2952
{
	struct rbd_img_request *img_request;
A
Alex Elder 已提交
2953
	struct rbd_device *rbd_dev;
2954 2955 2956 2957 2958

	rbd_assert(obj_request_img_data_test(obj_request));

	img_request = obj_request->img_request;
	rbd_assert(img_request);
A
Alex Elder 已提交
2959
	rbd_dev = img_request->rbd_dev;
2960

2961
	/* Reads */
2962 2963
	if (!img_request_write_test(img_request) &&
	    !img_request_discard_test(img_request))
2964 2965 2966 2967 2968 2969
		return true;

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

2970
	/*
2971 2972
	 * Layered writes outside of the parent overlap range don't
	 * share any data with the parent.
2973
	 */
2974 2975
	if (!obj_request_overlaps_parent(obj_request))
		return true;
2976

2977 2978 2979 2980 2981 2982 2983 2984
	/*
	 * 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;

2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998
	/*
	 * 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)
{
	if (img_obj_request_simple(obj_request)) {
2999 3000 3001 3002 3003 3004 3005 3006 3007 3008
		struct rbd_device *rbd_dev;
		struct ceph_osd_client *osdc;

		rbd_dev = obj_request->img_request->rbd_dev;
		osdc = &rbd_dev->rbd_client->client->osdc;

		return rbd_obj_request_submit(osdc, obj_request);
	}

	/*
3009 3010 3011 3012
	 * 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.
3013
	 */
3014
	if (obj_request_known_test(obj_request))
3015 3016 3017
		return rbd_img_obj_parent_read_full(obj_request);

	/* We don't know whether the target exists.  Go find out. */
3018 3019 3020 3021

	return rbd_img_obj_exists_submit(obj_request);
}

A
Alex Elder 已提交
3022 3023 3024
static int rbd_img_request_submit(struct rbd_img_request *img_request)
{
	struct rbd_obj_request *obj_request;
3025
	struct rbd_obj_request *next_obj_request;
3026
	int ret = 0;
A
Alex Elder 已提交
3027

A
Alex Elder 已提交
3028
	dout("%s: img %p\n", __func__, img_request);
A
Alex Elder 已提交
3029

3030 3031
	rbd_img_request_get(img_request);
	for_each_obj_request_safe(img_request, obj_request, next_obj_request) {
3032
		ret = rbd_img_obj_request_submit(obj_request);
A
Alex Elder 已提交
3033
		if (ret)
3034
			goto out_put_ireq;
A
Alex Elder 已提交
3035 3036
	}

3037 3038 3039
out_put_ireq:
	rbd_img_request_put(img_request);
	return ret;
A
Alex Elder 已提交
3040
}
A
Alex Elder 已提交
3041 3042 3043 3044

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
3045 3046
	struct rbd_device *rbd_dev;
	u64 obj_end;
3047 3048
	u64 img_xferred;
	int img_result;
A
Alex Elder 已提交
3049 3050 3051

	rbd_assert(img_request_child_test(img_request));

3052 3053
	/* First get what we need from the image request and release it */

A
Alex Elder 已提交
3054
	obj_request = img_request->obj_request;
3055 3056 3057 3058 3059 3060 3061 3062 3063
	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
3064 3065
	rbd_assert(obj_request);
	rbd_assert(obj_request->img_request);
3066 3067 3068 3069 3070 3071 3072 3073 3074
	rbd_dev = obj_request->img_request->rbd_dev;
	if (!rbd_dev->parent_overlap) {
		struct ceph_osd_client *osdc;

		osdc = &rbd_dev->rbd_client->client->osdc;
		img_result = rbd_obj_request_submit(osdc, obj_request);
		if (!img_result)
			return;
	}
A
Alex Elder 已提交
3075

3076
	obj_request->result = img_result;
A
Alex Elder 已提交
3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094
	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
3095

3096
		obj_request->xferred = min(img_xferred, xferred);
A
Alex Elder 已提交
3097
	} else {
3098
		obj_request->xferred = img_xferred;
A
Alex Elder 已提交
3099 3100
	}
out:
A
Alex Elder 已提交
3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112
	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);
3113
	rbd_assert(obj_request_type_valid(obj_request->type));
A
Alex Elder 已提交
3114 3115

	/* rbd_read_finish(obj_request, obj_request->length); */
3116
	img_request = rbd_parent_request_create(obj_request,
A
Alex Elder 已提交
3117
						obj_request->img_offset,
3118
						obj_request->length);
A
Alex Elder 已提交
3119 3120 3121 3122
	result = -ENOMEM;
	if (!img_request)
		goto out_err;

3123 3124 3125 3126 3127 3128
	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
3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144
	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
3145

3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212
static const struct rbd_client_id rbd_empty_cid;

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)
{
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
	struct rbd_client_id cid = rbd_get_cid(rbd_dev);
	char cookie[32];
	int ret;

	WARN_ON(__rbd_is_lock_owner(rbd_dev));

	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);
	if (ret)
		return ret;

	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;
}

/*
 * lock_rwsem must be held for write
 */
static int rbd_unlock(struct rbd_device *rbd_dev)
A
Alex Elder 已提交
3213
{
3214
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3215
	char cookie[32];
3216
	int ret;
A
Alex Elder 已提交
3217

3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283 3284 3285 3286 3287 3288 3289 3290
	WARN_ON(!__rbd_is_lock_owner(rbd_dev));

	rbd_dev->lock_state = RBD_LOCK_STATE_UNLOCKED;

	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;
	}

	rbd_set_owner_cid(rbd_dev, &rbd_empty_cid);
	queue_work(rbd_dev->task_wq, &rbd_dev->released_lock_work);
	return 0;
}

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)
{
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
	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;

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

	/* 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);

	return ceph_osdc_notify(osdc, &rbd_dev->header_oid,
				&rbd_dev->header_oloc, buf, buf_size,
				RBD_NOTIFY_TIMEOUT, preply_pages, preply_len);
}

static void rbd_notify_op_lock(struct rbd_device *rbd_dev,
			       enum rbd_notify_op notify_op)
{
	struct page **reply_pages;
	size_t reply_len;

	__rbd_notify_op_lock(rbd_dev, notify_op, &reply_pages, &reply_len);
	ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len));
}

static void rbd_notify_acquired_lock(struct work_struct *work)
{
	struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
						  acquired_lock_work);

	rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_ACQUIRED_LOCK);
}

static void rbd_notify_released_lock(struct work_struct *work)
{
	struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
						  released_lock_work);

	rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_RELEASED_LOCK);
}

static int rbd_request_lock(struct rbd_device *rbd_dev)
{
	struct page **reply_pages;
	size_t reply_len;
	bool lock_owner_responded = false;
	int ret;
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 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585 3586 3587 3588 3589
	dout("%s rbd_dev %p\n", __func__, rbd_dev);

	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);
		goto out;
	}

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

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

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

			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);
3590
	/*
3591
	 * Ensure that all in-flight IO is flushed.
3592
	 *
3593 3594
	 * FIXME: ceph_osdc_sync() flushes the entire OSD client, which
	 * may be shared with other devices.
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 3681 3682 3683 3684 3685 3686 3687 3688 3689 3690 3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714 3715 3716 3717 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
	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
3743

3744 3745
	ret = ceph_osdc_notify_ack(osdc, &rbd_dev->header_oid,
				   &rbd_dev->header_oloc, notify_id, cookie,
3746
				   buf, buf_size);
3747
	if (ret)
3748 3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778 3779 3780 3781 3782 3783 3784 3785 3786 3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829
		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;
	u8 struct_v;
	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
3830 3831
}

3832 3833
static void __rbd_unregister_watch(struct rbd_device *rbd_dev);

3834
static void rbd_watch_errcb(void *arg, u64 cookie, int err)
3835
{
3836
	struct rbd_device *rbd_dev = arg;
3837

3838
	rbd_warn(rbd_dev, "encountered watch error: %d", err);
3839

3840 3841 3842 3843
	down_write(&rbd_dev->lock_rwsem);
	rbd_set_owner_cid(rbd_dev, &rbd_empty_cid);
	up_write(&rbd_dev->lock_rwsem);

3844 3845 3846 3847
	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;
3848

3849
		queue_delayed_work(rbd_dev->task_wq, &rbd_dev->watch_dwork, 0);
3850
	}
3851
	mutex_unlock(&rbd_dev->watch_mutex);
3852 3853
}

3854
/*
3855
 * watch_mutex must be locked
3856
 */
3857
static int __rbd_register_watch(struct rbd_device *rbd_dev)
3858 3859
{
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3860
	struct ceph_osd_linger_request *handle;
3861

3862
	rbd_assert(!rbd_dev->watch_handle);
3863
	dout("%s rbd_dev %p\n", __func__, rbd_dev);
3864

3865 3866 3867 3868 3869
	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);
3870

3871
	rbd_dev->watch_handle = handle;
3872 3873 3874
	return 0;
}

3875 3876 3877 3878
/*
 * watch_mutex must be locked
 */
static void __rbd_unregister_watch(struct rbd_device *rbd_dev)
3879
{
3880 3881
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
	int ret;
3882

3883 3884
	rbd_assert(rbd_dev->watch_handle);
	dout("%s rbd_dev %p\n", __func__, rbd_dev);
3885

3886 3887 3888
	ret = ceph_osdc_unwatch(osdc, rbd_dev->watch_handle);
	if (ret)
		rbd_warn(rbd_dev, "failed to unwatch: %d", ret);
3889

3890
	rbd_dev->watch_handle = NULL;
3891 3892
}

3893 3894 3895 3896 3897 3898 3899 3900 3901 3902 3903 3904 3905 3906 3907 3908 3909 3910 3911
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)
3912
{
3913 3914 3915
	dout("%s rbd_dev %p\n", __func__, rbd_dev);

	cancel_delayed_work_sync(&rbd_dev->watch_dwork);
3916 3917 3918 3919
	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);
3920 3921 3922 3923
}

static void rbd_unregister_watch(struct rbd_device *rbd_dev)
{
3924
	WARN_ON(waitqueue_active(&rbd_dev->lock_waitq));
3925 3926 3927 3928 3929 3930 3931
	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
3932 3933

	ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc);
3934 3935
}

3936 3937 3938 3939
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);
3940
	bool was_lock_owner = false;
3941 3942 3943 3944
	int ret;

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

3945 3946 3947 3948
	down_write(&rbd_dev->lock_rwsem);
	if (rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED)
		was_lock_owner = rbd_release_lock(rbd_dev);

3949 3950 3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966 3967 3968 3969 3970
	mutex_lock(&rbd_dev->watch_mutex);
	if (rbd_dev->watch_state != RBD_WATCH_STATE_ERROR)
		goto fail_unlock;

	ret = __rbd_register_watch(rbd_dev);
	if (ret) {
		rbd_warn(rbd_dev, "failed to reregister watch: %d", ret);
		if (ret != -EBLACKLISTED)
			queue_delayed_work(rbd_dev->task_wq,
					   &rbd_dev->watch_dwork,
					   RBD_RETRY_DELAY);
		goto fail_unlock;
	}

	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);

3971 3972 3973 3974 3975 3976 3977 3978 3979
	if (was_lock_owner) {
		ret = rbd_try_lock(rbd_dev);
		if (ret)
			rbd_warn(rbd_dev, "reregisteration lock failed: %d",
				 ret);
	}

	up_write(&rbd_dev->lock_rwsem);
	wake_requests(rbd_dev, true);
3980 3981 3982 3983
	return;

fail_unlock:
	mutex_unlock(&rbd_dev->watch_mutex);
3984
	up_write(&rbd_dev->lock_rwsem);
3985 3986
}

3987
/*
3988 3989
 * Synchronous osd object method call.  Returns the number of bytes
 * returned in the outbound buffer, or a negative error code.
3990 3991 3992 3993 3994
 */
static int rbd_obj_method_sync(struct rbd_device *rbd_dev,
			     const char *object_name,
			     const char *class_name,
			     const char *method_name,
3995
			     const void *outbound,
3996
			     size_t outbound_size,
3997
			     void *inbound,
3998
			     size_t inbound_size)
3999
{
4000
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
4001 4002 4003 4004 4005 4006
	struct rbd_obj_request *obj_request;
	struct page **pages;
	u32 page_count;
	int ret;

	/*
4007 4008 4009 4010 4011
	 * 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.
4012
	 */
4013
	page_count = (u32)calc_pages_for(0, inbound_size);
4014 4015 4016 4017 4018
	pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
	if (IS_ERR(pages))
		return PTR_ERR(pages);

	ret = -ENOMEM;
4019
	obj_request = rbd_obj_request_create(object_name, 0, inbound_size,
4020 4021 4022 4023 4024 4025 4026
							OBJ_REQUEST_PAGES);
	if (!obj_request)
		goto out;

	obj_request->pages = pages;
	obj_request->page_count = page_count;

G
Guangliang Zhao 已提交
4027
	obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1,
4028
						  obj_request);
4029 4030 4031
	if (!obj_request->osd_req)
		goto out;

4032
	osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL,
4033 4034 4035 4036 4037 4038 4039 4040 4041 4042 4043 4044 4045
					class_name, method_name);
	if (outbound_size) {
		struct ceph_pagelist *pagelist;

		pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
		if (!pagelist)
			goto out;

		ceph_pagelist_init(pagelist);
		ceph_pagelist_append(pagelist, outbound, outbound_size);
		osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0,
						pagelist);
	}
4046 4047
	osd_req_op_cls_response_data_pages(obj_request->osd_req, 0,
					obj_request->pages, inbound_size,
4048
					0, false, false);
4049
	rbd_osd_req_format_read(obj_request);
A
Alex Elder 已提交
4050

4051 4052 4053 4054 4055 4056 4057 4058 4059 4060
	ret = rbd_obj_request_submit(osdc, obj_request);
	if (ret)
		goto out;
	ret = rbd_obj_request_wait(obj_request);
	if (ret)
		goto out;

	ret = obj_request->result;
	if (ret < 0)
		goto out;
4061 4062 4063

	rbd_assert(obj_request->xferred < (u64)INT_MAX);
	ret = (int)obj_request->xferred;
4064
	ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred);
4065 4066 4067 4068 4069 4070 4071 4072 4073
out:
	if (obj_request)
		rbd_obj_request_put(obj_request);
	else
		ceph_release_page_vector(pages, page_count);

	return ret;
}

4074 4075 4076 4077 4078 4079 4080 4081 4082 4083 4084 4085 4086 4087 4088 4089 4090 4091 4092 4093 4094 4095 4096
/*
 * 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);
	} while (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED);
	finish_wait(&rbd_dev->lock_waitq, &wait);
}

C
Christoph Hellwig 已提交
4097
static void rbd_queue_workfn(struct work_struct *work)
A
Alex Elder 已提交
4098
{
C
Christoph Hellwig 已提交
4099 4100
	struct request *rq = blk_mq_rq_from_pdu(work);
	struct rbd_device *rbd_dev = rq->q->queuedata;
I
Ilya Dryomov 已提交
4101
	struct rbd_img_request *img_request;
4102
	struct ceph_snap_context *snapc = NULL;
I
Ilya Dryomov 已提交
4103 4104
	u64 offset = (u64)blk_rq_pos(rq) << SECTOR_SHIFT;
	u64 length = blk_rq_bytes(rq);
G
Guangliang Zhao 已提交
4105
	enum obj_operation_type op_type;
4106
	u64 mapping_size;
4107
	bool must_be_locked = false;
A
Alex Elder 已提交
4108 4109
	int result;

C
Christoph Hellwig 已提交
4110 4111 4112 4113 4114 4115 4116
	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
4117
	if (req_op(rq) == REQ_OP_DISCARD)
4118
		op_type = OBJ_OP_DISCARD;
M
Mike Christie 已提交
4119
	else if (req_op(rq) == REQ_OP_WRITE)
G
Guangliang Zhao 已提交
4120 4121 4122 4123
		op_type = OBJ_OP_WRITE;
	else
		op_type = OBJ_OP_READ;

I
Ilya Dryomov 已提交
4124
	/* Ignore/skip any zero-length requests */
A
Alex Elder 已提交
4125

I
Ilya Dryomov 已提交
4126 4127 4128 4129 4130
	if (!length) {
		dout("%s: zero-length request\n", __func__);
		result = 0;
		goto err_rq;
	}
A
Alex Elder 已提交
4131

G
Guangliang Zhao 已提交
4132
	/* Only reads are allowed to a read-only device */
I
Ilya Dryomov 已提交
4133

G
Guangliang Zhao 已提交
4134
	if (op_type != OBJ_OP_READ) {
I
Ilya Dryomov 已提交
4135 4136 4137
		if (rbd_dev->mapping.read_only) {
			result = -EROFS;
			goto err_rq;
A
Alex Elder 已提交
4138
		}
I
Ilya Dryomov 已提交
4139 4140
		rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP);
	}
A
Alex Elder 已提交
4141

I
Ilya Dryomov 已提交
4142 4143 4144 4145 4146 4147 4148 4149 4150 4151 4152 4153
	/*
	 * 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
4154

I
Ilya Dryomov 已提交
4155 4156 4157 4158 4159 4160
	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
4161

C
Christoph Hellwig 已提交
4162 4163
	blk_mq_start_request(rq);

4164 4165
	down_read(&rbd_dev->header_rwsem);
	mapping_size = rbd_dev->mapping.size;
G
Guangliang Zhao 已提交
4166
	if (op_type != OBJ_OP_READ) {
4167 4168
		snapc = rbd_dev->header.snapc;
		ceph_get_snap_context(snapc);
4169
		must_be_locked = rbd_is_lock_supported(rbd_dev);
4170 4171 4172 4173
	}
	up_read(&rbd_dev->header_rwsem);

	if (offset + length > mapping_size) {
I
Ilya Dryomov 已提交
4174
		rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)", offset,
4175
			 length, mapping_size);
I
Ilya Dryomov 已提交
4176 4177 4178
		result = -EIO;
		goto err_rq;
	}
A
Alex Elder 已提交
4179

4180 4181 4182 4183 4184 4185
	if (must_be_locked) {
		down_read(&rbd_dev->lock_rwsem);
		if (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED)
			rbd_wait_state_locked(rbd_dev);
	}

G
Guangliang Zhao 已提交
4186
	img_request = rbd_img_request_create(rbd_dev, offset, length, op_type,
4187
					     snapc);
I
Ilya Dryomov 已提交
4188 4189
	if (!img_request) {
		result = -ENOMEM;
4190
		goto err_unlock;
I
Ilya Dryomov 已提交
4191 4192
	}
	img_request->rq = rq;
4193
	snapc = NULL; /* img_request consumes a ref */
A
Alex Elder 已提交
4194

4195 4196 4197 4198 4199 4200
	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
4201 4202
	if (result)
		goto err_img_request;
A
Alex Elder 已提交
4203

I
Ilya Dryomov 已提交
4204 4205 4206
	result = rbd_img_request_submit(img_request);
	if (result)
		goto err_img_request;
A
Alex Elder 已提交
4207

4208 4209
	if (must_be_locked)
		up_read(&rbd_dev->lock_rwsem);
I
Ilya Dryomov 已提交
4210
	return;
A
Alex Elder 已提交
4211

I
Ilya Dryomov 已提交
4212 4213
err_img_request:
	rbd_img_request_put(img_request);
4214 4215 4216
err_unlock:
	if (must_be_locked)
		up_read(&rbd_dev->lock_rwsem);
I
Ilya Dryomov 已提交
4217 4218 4219
err_rq:
	if (result)
		rbd_warn(rbd_dev, "%s %llx at %llx result %d",
G
Guangliang Zhao 已提交
4220
			 obj_op_name(op_type), length, offset, result);
4221
	ceph_put_snap_context(snapc);
C
Christoph Hellwig 已提交
4222 4223
err:
	blk_mq_end_request(rq, result);
I
Ilya Dryomov 已提交
4224
}
A
Alex Elder 已提交
4225

C
Christoph Hellwig 已提交
4226 4227
static int rbd_queue_rq(struct blk_mq_hw_ctx *hctx,
		const struct blk_mq_queue_data *bd)
I
Ilya Dryomov 已提交
4228
{
C
Christoph Hellwig 已提交
4229 4230
	struct request *rq = bd->rq;
	struct work_struct *work = blk_mq_rq_to_pdu(rq);
A
Alex Elder 已提交
4231

C
Christoph Hellwig 已提交
4232 4233
	queue_work(rbd_wq, work);
	return BLK_MQ_RQ_QUEUE_OK;
A
Alex Elder 已提交
4234 4235
}

4236 4237 4238 4239 4240 4241 4242
static void rbd_free_disk(struct rbd_device *rbd_dev)
{
	struct gendisk *disk = rbd_dev->disk;

	if (!disk)
		return;

4243 4244
	rbd_dev->disk = NULL;
	if (disk->flags & GENHD_FL_UP) {
4245
		del_gendisk(disk);
4246 4247
		if (disk->queue)
			blk_cleanup_queue(disk->queue);
C
Christoph Hellwig 已提交
4248
		blk_mq_free_tag_set(&rbd_dev->tag_set);
4249
	}
4250 4251 4252
	put_disk(disk);
}

4253 4254
static int rbd_obj_read_sync(struct rbd_device *rbd_dev,
				const char *object_name,
4255
				u64 offset, u64 length, void *buf)
4256 4257

{
4258
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
4259 4260 4261
	struct rbd_obj_request *obj_request;
	struct page **pages = NULL;
	u32 page_count;
4262
	size_t size;
4263 4264 4265 4266 4267
	int ret;

	page_count = (u32) calc_pages_for(offset, length);
	pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
	if (IS_ERR(pages))
4268
		return PTR_ERR(pages);
4269 4270 4271

	ret = -ENOMEM;
	obj_request = rbd_obj_request_create(object_name, offset, length,
4272
							OBJ_REQUEST_PAGES);
4273 4274 4275 4276 4277 4278
	if (!obj_request)
		goto out;

	obj_request->pages = pages;
	obj_request->page_count = page_count;

G
Guangliang Zhao 已提交
4279
	obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1,
4280
						  obj_request);
4281 4282 4283
	if (!obj_request->osd_req)
		goto out;

4284 4285
	osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ,
					offset, length, 0, 0);
4286
	osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0,
4287
					obj_request->pages,
4288 4289 4290
					obj_request->length,
					obj_request->offset & ~PAGE_MASK,
					false, false);
4291
	rbd_osd_req_format_read(obj_request);
A
Alex Elder 已提交
4292

4293 4294 4295 4296 4297 4298 4299 4300 4301 4302
	ret = rbd_obj_request_submit(osdc, obj_request);
	if (ret)
		goto out;
	ret = rbd_obj_request_wait(obj_request);
	if (ret)
		goto out;

	ret = obj_request->result;
	if (ret < 0)
		goto out;
4303 4304 4305

	rbd_assert(obj_request->xferred <= (u64) SIZE_MAX);
	size = (size_t) obj_request->xferred;
4306
	ceph_copy_from_page_vector(pages, buf, 0, size);
4307 4308
	rbd_assert(size <= (size_t)INT_MAX);
	ret = (int)size;
4309 4310 4311 4312 4313 4314 4315 4316 4317
out:
	if (obj_request)
		rbd_obj_request_put(obj_request);
	else
		ceph_release_page_vector(pages, page_count);

	return ret;
}

4318
/*
A
Alex Elder 已提交
4319 4320 4321
 * 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.
4322
 */
4323
static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev)
4324
{
4325
	struct rbd_image_header_ondisk *ondisk = NULL;
4326
	u32 snap_count = 0;
4327 4328 4329
	u64 names_size = 0;
	u32 want_count;
	int ret;
4330

A
Alex Elder 已提交
4331
	/*
4332 4333 4334 4335 4336
	 * 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
4337
	 */
4338 4339 4340 4341 4342 4343 4344 4345 4346 4347
	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
4348
			return -ENOMEM;
4349

4350
		ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_oid.name,
4351
				       0, size, ondisk);
4352
		if (ret < 0)
A
Alex Elder 已提交
4353
			goto out;
A
Alex Elder 已提交
4354
		if ((size_t)ret < size) {
4355
			ret = -ENXIO;
A
Alex Elder 已提交
4356 4357
			rbd_warn(rbd_dev, "short header read (want %zd got %d)",
				size, ret);
A
Alex Elder 已提交
4358
			goto out;
4359 4360 4361
		}
		if (!rbd_dev_ondisk_valid(ondisk)) {
			ret = -ENXIO;
A
Alex Elder 已提交
4362
			rbd_warn(rbd_dev, "invalid header");
A
Alex Elder 已提交
4363
			goto out;
4364
		}
4365

4366 4367 4368 4369
		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
4370

A
Alex Elder 已提交
4371 4372
	ret = rbd_header_from_disk(rbd_dev, ondisk);
out:
4373 4374 4375
	kfree(ondisk);

	return ret;
4376 4377
}

4378 4379 4380 4381 4382 4383 4384 4385 4386 4387 4388 4389 4390 4391 4392 4393 4394 4395 4396
/*
 * 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);
}

4397 4398 4399 4400 4401
static void rbd_dev_update_size(struct rbd_device *rbd_dev)
{
	sector_t size;

	/*
I
Ilya Dryomov 已提交
4402 4403 4404
	 * 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.
4405
	 */
I
Ilya Dryomov 已提交
4406 4407
	if (test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags) &&
	    !test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) {
4408 4409 4410 4411 4412 4413 4414
		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
4415
static int rbd_dev_refresh(struct rbd_device *rbd_dev)
A
Alex Elder 已提交
4416
{
4417
	u64 mapping_size;
A
Alex Elder 已提交
4418 4419
	int ret;

4420
	down_write(&rbd_dev->header_rwsem);
4421
	mapping_size = rbd_dev->mapping.size;
4422 4423

	ret = rbd_dev_header_info(rbd_dev);
4424
	if (ret)
4425
		goto out;
4426

4427 4428 4429 4430 4431 4432 4433
	/*
	 * 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)
4434
			goto out;
4435 4436
	}

4437
	if (rbd_dev->spec->snap_id == CEPH_NOSNAP) {
4438
		rbd_dev->mapping.size = rbd_dev->header.image_size;
4439 4440 4441 4442
	} else {
		/* validate mapped snapshot's EXISTS flag */
		rbd_exists_validate(rbd_dev);
	}
4443

4444
out:
4445
	up_write(&rbd_dev->header_rwsem);
4446
	if (!ret && mapping_size != rbd_dev->mapping.size)
4447
		rbd_dev_update_size(rbd_dev);
A
Alex Elder 已提交
4448

4449
	return ret;
A
Alex Elder 已提交
4450 4451
}

C
Christoph Hellwig 已提交
4452 4453 4454 4455 4456 4457 4458 4459 4460 4461 4462 4463 4464 4465 4466 4467
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,
	.map_queue	= blk_mq_map_queue,
	.init_request	= rbd_init_request,
};

4468 4469 4470 4471
static int rbd_init_disk(struct rbd_device *rbd_dev)
{
	struct gendisk *disk;
	struct request_queue *q;
A
Alex Elder 已提交
4472
	u64 segment_size;
C
Christoph Hellwig 已提交
4473
	int err;
4474 4475

	/* create gendisk info */
4476 4477 4478
	disk = alloc_disk(single_major ?
			  (1 << RBD_SINGLE_MAJOR_PART_SHIFT) :
			  RBD_MINORS_PER_MAJOR);
4479
	if (!disk)
A
Alex Elder 已提交
4480
		return -ENOMEM;
4481

A
Alex Elder 已提交
4482
	snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
A
Alex Elder 已提交
4483
		 rbd_dev->dev_id);
4484
	disk->major = rbd_dev->major;
4485
	disk->first_minor = rbd_dev->minor;
4486 4487
	if (single_major)
		disk->flags |= GENHD_FL_EXT_DEVT;
4488 4489 4490
	disk->fops = &rbd_bd_ops;
	disk->private_data = rbd_dev;

C
Christoph Hellwig 已提交
4491 4492
	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
4493
	rbd_dev->tag_set.queue_depth = rbd_dev->opts->queue_depth;
C
Christoph Hellwig 已提交
4494
	rbd_dev->tag_set.numa_node = NUMA_NO_NODE;
I
Ilya Dryomov 已提交
4495
	rbd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
C
Christoph Hellwig 已提交
4496 4497 4498 4499 4500
	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)
4501
		goto out_disk;
4502

C
Christoph Hellwig 已提交
4503 4504 4505 4506 4507 4508
	q = blk_mq_init_queue(&rbd_dev->tag_set);
	if (IS_ERR(q)) {
		err = PTR_ERR(q);
		goto out_tag_set;
	}

4509 4510
	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
4511

4512
	/* set io sizes to object size */
A
Alex Elder 已提交
4513 4514
	segment_size = rbd_obj_bytes(&rbd_dev->header);
	blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE);
I
Ilya Dryomov 已提交
4515
	q->limits.max_sectors = queue_max_hw_sectors(q);
I
Ilya Dryomov 已提交
4516
	blk_queue_max_segments(q, segment_size / SECTOR_SIZE);
A
Alex Elder 已提交
4517 4518 4519
	blk_queue_max_segment_size(q, segment_size);
	blk_queue_io_min(q, segment_size);
	blk_queue_io_opt(q, segment_size);
4520

4521 4522 4523 4524
	/* enable the discard support */
	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
	q->limits.discard_granularity = segment_size;
	q->limits.discard_alignment = segment_size;
4525
	blk_queue_max_discard_sectors(q, segment_size / SECTOR_SIZE);
4526
	q->limits.discard_zeroes_data = 1;
4527

4528 4529 4530
	if (!ceph_test_opt(rbd_dev->rbd_client->client, NOCRC))
		q->backing_dev_info.capabilities |= BDI_CAP_STABLE_WRITES;

4531 4532 4533 4534 4535 4536 4537
	disk->queue = q;

	q->queuedata = rbd_dev;

	rbd_dev->disk = disk;

	return 0;
C
Christoph Hellwig 已提交
4538 4539
out_tag_set:
	blk_mq_free_tag_set(&rbd_dev->tag_set);
4540 4541
out_disk:
	put_disk(disk);
C
Christoph Hellwig 已提交
4542
	return err;
4543 4544
}

4545 4546 4547 4548
/*
  sysfs
*/

A
Alex Elder 已提交
4549 4550 4551 4552 4553
static struct rbd_device *dev_to_rbd_dev(struct device *dev)
{
	return container_of(dev, struct rbd_device, dev);
}

4554 4555 4556
static ssize_t rbd_size_show(struct device *dev,
			     struct device_attribute *attr, char *buf)
{
A
Alex Elder 已提交
4557
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4558

A
Alex Elder 已提交
4559 4560
	return sprintf(buf, "%llu\n",
		(unsigned long long)rbd_dev->mapping.size);
4561 4562
}

A
Alex Elder 已提交
4563 4564 4565 4566 4567 4568 4569 4570 4571 4572
/*
 * 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
4573
			(unsigned long long)rbd_dev->mapping.features);
A
Alex Elder 已提交
4574 4575
}

4576 4577 4578
static ssize_t rbd_major_show(struct device *dev,
			      struct device_attribute *attr, char *buf)
{
A
Alex Elder 已提交
4579
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4580

A
Alex Elder 已提交
4581 4582 4583 4584
	if (rbd_dev->major)
		return sprintf(buf, "%d\n", rbd_dev->major);

	return sprintf(buf, "(none)\n");
4585 4586 4587 4588 4589 4590
}

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
4591

4592
	return sprintf(buf, "%d\n", rbd_dev->minor);
4593 4594
}

4595 4596 4597 4598 4599 4600 4601 4602 4603 4604 4605
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));
}

4606 4607
static ssize_t rbd_client_id_show(struct device *dev,
				  struct device_attribute *attr, char *buf)
4608
{
A
Alex Elder 已提交
4609
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4610

4611
	return sprintf(buf, "client%lld\n",
4612
		       ceph_client_gid(rbd_dev->rbd_client->client));
4613 4614
}

4615 4616
static ssize_t rbd_pool_show(struct device *dev,
			     struct device_attribute *attr, char *buf)
4617
{
A
Alex Elder 已提交
4618
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4619

4620
	return sprintf(buf, "%s\n", rbd_dev->spec->pool_name);
4621 4622
}

4623 4624 4625 4626 4627
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);

4628
	return sprintf(buf, "%llu\n",
A
Alex Elder 已提交
4629
			(unsigned long long) rbd_dev->spec->pool_id);
4630 4631
}

4632 4633 4634
static ssize_t rbd_name_show(struct device *dev,
			     struct device_attribute *attr, char *buf)
{
A
Alex Elder 已提交
4635
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4636

A
Alex Elder 已提交
4637 4638 4639 4640
	if (rbd_dev->spec->image_name)
		return sprintf(buf, "%s\n", rbd_dev->spec->image_name);

	return sprintf(buf, "(unknown)\n");
4641 4642
}

A
Alex Elder 已提交
4643 4644 4645 4646 4647
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);

4648
	return sprintf(buf, "%s\n", rbd_dev->spec->image_id);
A
Alex Elder 已提交
4649 4650
}

A
Alex Elder 已提交
4651 4652 4653 4654
/*
 * Shows the name of the currently-mapped snapshot (or
 * RBD_SNAP_HEAD_NAME for the base image).
 */
4655 4656 4657 4658
static ssize_t rbd_snap_show(struct device *dev,
			     struct device_attribute *attr,
			     char *buf)
{
A
Alex Elder 已提交
4659
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4660

4661
	return sprintf(buf, "%s\n", rbd_dev->spec->snap_name);
4662 4663
}

4664
/*
4665 4666 4667
 * 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)".
4668 4669
 */
static ssize_t rbd_parent_show(struct device *dev,
4670 4671
			       struct device_attribute *attr,
			       char *buf)
4672 4673
{
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4674
	ssize_t count = 0;
4675

4676
	if (!rbd_dev->parent)
4677 4678
		return sprintf(buf, "(no parent image)\n");

4679 4680 4681 4682 4683 4684 4685 4686 4687 4688 4689 4690 4691 4692 4693 4694
	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;
4695 4696
}

4697 4698 4699 4700 4701
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
4702
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4703
	int ret;
4704

A
Alex Elder 已提交
4705
	ret = rbd_dev_refresh(rbd_dev);
4706
	if (ret)
4707
		return ret;
4708

4709
	return size;
4710
}
4711

4712
static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
A
Alex Elder 已提交
4713
static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL);
4714
static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
4715
static DEVICE_ATTR(minor, S_IRUGO, rbd_minor_show, NULL);
4716
static DEVICE_ATTR(client_addr, S_IRUGO, rbd_client_addr_show, NULL);
4717 4718
static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
4719
static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL);
4720
static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
A
Alex Elder 已提交
4721
static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL);
4722 4723
static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
4724
static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL);
4725 4726 4727

static struct attribute *rbd_attrs[] = {
	&dev_attr_size.attr,
A
Alex Elder 已提交
4728
	&dev_attr_features.attr,
4729
	&dev_attr_major.attr,
4730
	&dev_attr_minor.attr,
4731
	&dev_attr_client_addr.attr,
4732 4733
	&dev_attr_client_id.attr,
	&dev_attr_pool.attr,
4734
	&dev_attr_pool_id.attr,
4735
	&dev_attr_name.attr,
A
Alex Elder 已提交
4736
	&dev_attr_image_id.attr,
4737
	&dev_attr_current_snap.attr,
4738
	&dev_attr_parent.attr,
4739 4740 4741 4742 4743 4744 4745 4746 4747 4748 4749 4750 4751
	&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
};

4752
static void rbd_dev_release(struct device *dev);
4753 4754 4755 4756

static struct device_type rbd_device_type = {
	.name		= "rbd",
	.groups		= rbd_attr_groups,
4757
	.release	= rbd_dev_release,
4758 4759
};

4760 4761 4762 4763 4764 4765 4766 4767 4768 4769 4770 4771 4772 4773 4774 4775 4776 4777 4778 4779 4780
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;
4781 4782 4783

	spec->pool_id = CEPH_NOPOOL;
	spec->snap_id = CEPH_NOSNAP;
4784 4785 4786 4787 4788 4789 4790 4791 4792 4793 4794 4795 4796 4797 4798 4799
	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);
}

4800
static void rbd_dev_free(struct rbd_device *rbd_dev)
4801
{
4802
	WARN_ON(rbd_dev->watch_state != RBD_WATCH_STATE_UNREGISTERED);
4803
	WARN_ON(rbd_dev->lock_state != RBD_LOCK_STATE_UNLOCKED);
4804

4805
	ceph_oid_destroy(&rbd_dev->header_oid);
4806
	ceph_oloc_destroy(&rbd_dev->header_oloc);
4807

4808 4809 4810 4811
	rbd_put_client(rbd_dev->rbd_client);
	rbd_spec_put(rbd_dev->spec);
	kfree(rbd_dev->opts);
	kfree(rbd_dev);
4812 4813 4814 4815 4816 4817 4818 4819 4820 4821 4822 4823 4824
}

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);
4825 4826 4827 4828 4829 4830 4831 4832 4833 4834

	/*
	 * 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);
}

4835 4836
static struct rbd_device *__rbd_dev_create(struct rbd_client *rbdc,
					   struct rbd_spec *spec)
4837 4838 4839
{
	struct rbd_device *rbd_dev;

4840
	rbd_dev = kzalloc(sizeof(*rbd_dev), GFP_KERNEL);
4841 4842 4843 4844 4845 4846 4847
	if (!rbd_dev)
		return NULL;

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

4848
	ceph_oid_init(&rbd_dev->header_oid);
4849
	ceph_oloc_init(&rbd_dev->header_oloc);
4850

4851 4852 4853 4854
	mutex_init(&rbd_dev->watch_mutex);
	rbd_dev->watch_state = RBD_WATCH_STATE_UNREGISTERED;
	INIT_DELAYED_WORK(&rbd_dev->watch_dwork, rbd_reregister_watch);

4855 4856 4857 4858 4859 4860 4861 4862
	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);

4863 4864 4865 4866 4867
	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);

4868
	rbd_dev->rbd_client = rbdc;
4869
	rbd_dev->spec = spec;
4870

4871 4872 4873 4874
	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;
4875
	RCU_INIT_POINTER(rbd_dev->layout.pool_ns, NULL);
4876

4877 4878 4879 4880 4881 4882 4883 4884 4885 4886 4887 4888 4889 4890 4891 4892 4893 4894 4895 4896 4897 4898 4899 4900 4901 4902 4903 4904 4905 4906
	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;
4907

4908 4909 4910 4911
	/* we have a ref from do_rbd_add() */
	__module_get(THIS_MODULE);

	dout("%s rbd_dev %p dev_id %d\n", __func__, rbd_dev, rbd_dev->dev_id);
4912
	return rbd_dev;
4913 4914 4915 4916 4917 4918

fail_dev_id:
	ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id);
fail_rbd_dev:
	rbd_dev_free(rbd_dev);
	return NULL;
4919 4920 4921 4922
}

static void rbd_dev_destroy(struct rbd_device *rbd_dev)
{
4923 4924
	if (rbd_dev)
		put_device(&rbd_dev->dev);
4925 4926
}

4927 4928 4929 4930 4931 4932 4933 4934 4935 4936 4937 4938 4939 4940 4941
/*
 * 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 };

4942
	ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_oid.name,
4943
				"rbd", "get_size",
4944
				&snapid, sizeof (snapid),
4945
				&size_buf, sizeof (size_buf));
4946
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4947 4948
	if (ret < 0)
		return ret;
4949 4950
	if (ret < sizeof (size_buf))
		return -ERANGE;
4951

J
Josh Durgin 已提交
4952
	if (order) {
4953
		*order = size_buf.order;
J
Josh Durgin 已提交
4954 4955
		dout("  order %u", (unsigned int)*order);
	}
4956 4957
	*snap_size = le64_to_cpu(size_buf.size);

J
Josh Durgin 已提交
4958 4959
	dout("  snap_id 0x%016llx snap_size = %llu\n",
		(unsigned long long)snap_id,
4960
		(unsigned long long)*snap_size);
4961 4962 4963 4964 4965 4966 4967 4968 4969 4970 4971

	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);
}

4972 4973 4974 4975 4976 4977 4978 4979 4980 4981
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;

4982
	ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_oid.name,
4983
				"rbd", "get_object_prefix", NULL, 0,
4984
				reply_buf, RBD_OBJ_PREFIX_LEN_MAX);
4985
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4986 4987 4988 4989 4990
	if (ret < 0)
		goto out;

	p = reply_buf;
	rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p,
4991 4992
						p + ret, NULL, GFP_NOIO);
	ret = 0;
4993 4994 4995 4996 4997 4998 4999 5000 5001 5002 5003 5004 5005

	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;
}

5006 5007 5008 5009 5010 5011 5012
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;
5013
	} __attribute__ ((packed)) features_buf = { 0 };
5014
	u64 unsup;
5015 5016
	int ret;

5017
	ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_oid.name,
5018
				"rbd", "get_features",
5019
				&snapid, sizeof (snapid),
5020
				&features_buf, sizeof (features_buf));
5021
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
5022 5023
	if (ret < 0)
		return ret;
5024 5025
	if (ret < sizeof (features_buf))
		return -ERANGE;
A
Alex Elder 已提交
5026

5027 5028 5029 5030
	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
5031
		return -ENXIO;
5032
	}
A
Alex Elder 已提交
5033

5034 5035 5036
	*snap_features = le64_to_cpu(features_buf.features);

	dout("  snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
5037 5038 5039
		(unsigned long long)snap_id,
		(unsigned long long)*snap_features,
		(unsigned long long)le64_to_cpu(features_buf.incompat));
5040 5041 5042 5043 5044 5045 5046 5047 5048 5049

	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);
}

5050 5051 5052 5053 5054 5055 5056 5057
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
5058
	u64 pool_id;
5059
	char *image_id;
5060
	u64 snap_id;
5061 5062 5063 5064 5065 5066 5067 5068 5069 5070 5071 5072 5073 5074 5075 5076 5077
	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;
	}

5078
	snapid = cpu_to_le64(rbd_dev->spec->snap_id);
5079
	ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_oid.name,
5080
				"rbd", "get_parent",
5081
				&snapid, sizeof (snapid),
5082
				reply_buf, size);
5083
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
5084 5085 5086 5087
	if (ret < 0)
		goto out_err;

	p = reply_buf;
5088 5089
	end = reply_buf + ret;
	ret = -ERANGE;
A
Alex Elder 已提交
5090
	ceph_decode_64_safe(&p, end, pool_id, out_err);
5091 5092 5093 5094 5095 5096 5097 5098 5099 5100 5101 5102 5103 5104 5105 5106 5107
	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);
		}

5108
		goto out;	/* No parent?  No problem. */
5109
	}
5110

5111 5112 5113
	/* The ceph file layout needs to fit pool id in 32 bits */

	ret = -EIO;
A
Alex Elder 已提交
5114
	if (pool_id > (u64)U32_MAX) {
5115
		rbd_warn(NULL, "parent pool id too large (%llu > %u)",
A
Alex Elder 已提交
5116
			(unsigned long long)pool_id, U32_MAX);
5117
		goto out_err;
A
Alex Elder 已提交
5118
	}
5119

A
Alex Elder 已提交
5120
	image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
5121 5122 5123 5124
	if (IS_ERR(image_id)) {
		ret = PTR_ERR(image_id);
		goto out_err;
	}
5125
	ceph_decode_64_safe(&p, end, snap_id, out_err);
5126 5127
	ceph_decode_64_safe(&p, end, overlap, out_err);

5128 5129 5130 5131 5132 5133 5134 5135 5136
	/*
	 * 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
5137 5138
		rbd_dev->parent_spec = parent_spec;
		parent_spec = NULL;	/* rbd_dev now owns this */
5139 5140
	} else {
		kfree(image_id);
5141 5142 5143
	}

	/*
5144 5145
	 * We always update the parent overlap.  If it's zero we issue
	 * a warning, as we will proceed as if there was no parent.
5146 5147 5148
	 */
	if (!overlap) {
		if (parent_spec) {
5149 5150 5151 5152
			/* refresh, careful to warn just once */
			if (rbd_dev->parent_overlap)
				rbd_warn(rbd_dev,
				    "clone now standalone (overlap became 0)");
5153
		} else {
5154 5155
			/* initial probe */
			rbd_warn(rbd_dev, "clone is standalone (overlap 0)");
5156
		}
A
Alex Elder 已提交
5157
	}
5158 5159
	rbd_dev->parent_overlap = overlap;

5160 5161 5162 5163 5164 5165 5166 5167 5168
out:
	ret = 0;
out_err:
	kfree(reply_buf);
	rbd_spec_put(parent_spec);

	return ret;
}

5169 5170 5171 5172 5173 5174 5175 5176 5177 5178 5179 5180 5181
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;

5182
	ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_oid.name,
5183
				"rbd", "get_stripe_unit_count", NULL, 0,
5184
				(char *)&striping_info_buf, size);
5185 5186 5187 5188 5189 5190 5191 5192 5193 5194 5195 5196 5197 5198 5199 5200 5201 5202 5203 5204 5205 5206 5207 5208 5209 5210 5211 5212
	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;
	obj_size = (u64)1 << rbd_dev->header.obj_order;
	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;
	}
5213 5214
	rbd_dev->header.stripe_unit = stripe_unit;
	rbd_dev->header.stripe_count = stripe_count;
5215 5216 5217 5218

	return 0;
}

5219 5220 5221 5222 5223 5224 5225 5226 5227 5228 5229 5230 5231 5232
static char *rbd_dev_image_name(struct rbd_device *rbd_dev)
{
	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
5233 5234
	len = strlen(rbd_dev->spec->image_id);
	image_id_size = sizeof (__le32) + len;
5235 5236 5237 5238 5239
	image_id = kmalloc(image_id_size, GFP_KERNEL);
	if (!image_id)
		return NULL;

	p = image_id;
5240
	end = image_id + image_id_size;
5241
	ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len);
5242 5243 5244 5245 5246 5247

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

5248
	ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY,
5249 5250
				"rbd", "dir_get_name",
				image_id, image_id_size,
5251
				reply_buf, size);
5252 5253 5254
	if (ret < 0)
		goto out;
	p = reply_buf;
5255 5256
	end = reply_buf + ret;

5257 5258 5259 5260 5261 5262 5263 5264 5265 5266 5267 5268
	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;
}

5269 5270 5271 5272 5273 5274 5275 5276 5277 5278 5279 5280 5281 5282 5283 5284 5285 5286 5287 5288 5289 5290 5291 5292 5293 5294 5295 5296 5297 5298
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);
5299 5300 5301 5302 5303 5304 5305
		if (IS_ERR(snap_name)) {
			/* ignore no-longer existing snapshots */
			if (PTR_ERR(snap_name) == -ENOENT)
				continue;
			else
				break;
		}
5306 5307 5308 5309 5310 5311 5312 5313 5314 5315 5316 5317 5318 5319 5320 5321 5322 5323
		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);
}

5324
/*
5325 5326 5327 5328 5329 5330 5331 5332 5333 5334 5335 5336 5337 5338 5339 5340 5341 5342 5343 5344 5345 5346 5347 5348 5349 5350 5351
 * 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.
5352
 *
5353 5354
 * All names in an rbd spec are dynamically allocated.  It's OK if we
 * can't figure out the name for an image id.
5355
 */
5356
static int rbd_spec_fill_names(struct rbd_device *rbd_dev)
5357
{
5358 5359 5360 5361 5362
	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;
5363 5364
	int ret;

5365 5366 5367
	rbd_assert(spec->pool_id != CEPH_NOPOOL);
	rbd_assert(spec->image_id);
	rbd_assert(spec->snap_id != CEPH_NOSNAP);
5368

5369
	/* Get the pool name; we have to make our own copy of this */
5370

5371 5372 5373
	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);
5374 5375
		return -EIO;
	}
5376 5377
	pool_name = kstrdup(pool_name, GFP_KERNEL);
	if (!pool_name)
5378 5379 5380 5381
		return -ENOMEM;

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

5382 5383
	image_name = rbd_dev_image_name(rbd_dev);
	if (!image_name)
A
Alex Elder 已提交
5384
		rbd_warn(rbd_dev, "unable to get image name");
5385

5386
	/* Fetch the snapshot name */
5387

5388
	snap_name = rbd_snap_name(rbd_dev, spec->snap_id);
5389 5390
	if (IS_ERR(snap_name)) {
		ret = PTR_ERR(snap_name);
5391
		goto out_err;
5392 5393 5394 5395 5396
	}

	spec->pool_name = pool_name;
	spec->image_name = image_name;
	spec->snap_name = snap_name;
5397 5398

	return 0;
5399

5400
out_err:
5401 5402
	kfree(image_name);
	kfree(pool_name);
5403 5404 5405
	return ret;
}

A
Alex Elder 已提交
5406
static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev)
5407 5408 5409 5410 5411 5412 5413 5414 5415 5416 5417 5418 5419 5420 5421 5422 5423 5424 5425 5426 5427 5428 5429
{
	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;

5430
	ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_oid.name,
5431
				"rbd", "get_snapcontext", NULL, 0,
5432
				reply_buf, size);
5433
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
5434 5435 5436 5437
	if (ret < 0)
		goto out;

	p = reply_buf;
5438 5439
	end = reply_buf + ret;
	ret = -ERANGE;
5440 5441 5442 5443 5444 5445 5446 5447 5448 5449 5450 5451 5452 5453 5454 5455
	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;
5456
	ret = 0;
5457

5458
	snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
5459 5460 5461 5462 5463 5464 5465 5466
	if (!snapc) {
		ret = -ENOMEM;
		goto out;
	}
	snapc->seq = seq;
	for (i = 0; i < snap_count; i++)
		snapc->snaps[i] = ceph_decode_64(&p);

5467
	ceph_put_snap_context(rbd_dev->header.snapc);
5468 5469 5470
	rbd_dev->header.snapc = snapc;

	dout("  snap context seq = %llu, snap_count = %u\n",
5471
		(unsigned long long)seq, (unsigned int)snap_count);
5472 5473 5474
out:
	kfree(reply_buf);

5475
	return ret;
5476 5477
}

5478 5479
static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
					u64 snap_id)
A
Alex Elder 已提交
5480 5481 5482
{
	size_t size;
	void *reply_buf;
5483
	__le64 snapid;
A
Alex Elder 已提交
5484 5485 5486 5487 5488 5489 5490 5491 5492 5493
	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);

5494
	snapid = cpu_to_le64(snap_id);
5495
	ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_oid.name,
A
Alex Elder 已提交
5496
				"rbd", "get_snapshot_name",
5497
				&snapid, sizeof (snapid),
5498
				reply_buf, size);
5499
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
5500 5501
	if (ret < 0) {
		snap_name = ERR_PTR(ret);
A
Alex Elder 已提交
5502
		goto out;
5503
	}
A
Alex Elder 已提交
5504 5505

	p = reply_buf;
5506
	end = reply_buf + ret;
A
Alex Elder 已提交
5507
	snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
5508
	if (IS_ERR(snap_name))
A
Alex Elder 已提交
5509 5510
		goto out;

5511
	dout("  snap_id 0x%016llx snap_name = %s\n",
5512
		(unsigned long long)snap_id, snap_name);
A
Alex Elder 已提交
5513 5514 5515
out:
	kfree(reply_buf);

5516
	return snap_name;
A
Alex Elder 已提交
5517 5518
}

5519
static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev)
A
Alex Elder 已提交
5520
{
5521
	bool first_time = rbd_dev->header.object_prefix == NULL;
A
Alex Elder 已提交
5522 5523
	int ret;

5524 5525
	ret = rbd_dev_v2_image_size(rbd_dev);
	if (ret)
5526
		return ret;
5527

5528 5529 5530
	if (first_time) {
		ret = rbd_dev_v2_header_onetime(rbd_dev);
		if (ret)
5531
			return ret;
5532 5533
	}

A
Alex Elder 已提交
5534
	ret = rbd_dev_v2_snap_context(rbd_dev);
5535 5536 5537 5538
	if (ret && first_time) {
		kfree(rbd_dev->header.object_prefix);
		rbd_dev->header.object_prefix = NULL;
	}
A
Alex Elder 已提交
5539 5540 5541 5542

	return ret;
}

5543 5544 5545 5546 5547 5548 5549 5550 5551 5552
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);
}

5553 5554 5555
/*
 * 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
5556 5557
 * the token (string of non-white space characters) found.  Note
 * that *buf must be terminated with '\0'.
5558 5559 5560 5561 5562 5563 5564 5565 5566 5567 5568 5569 5570 5571
 */
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
5572 5573 5574 5575 5576 5577 5578 5579 5580 5581 5582 5583 5584 5585 5586 5587 5588 5589 5590 5591 5592 5593
/*
 * 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
5594
	dup = kmemdup(*buf, len + 1, GFP_KERNEL);
A
Alex Elder 已提交
5595 5596 5597 5598 5599 5600 5601 5602 5603 5604 5605
	if (!dup)
		return NULL;
	*(dup + len) = '\0';
	*buf += len;

	if (lenp)
		*lenp = len;

	return dup;
}

5606
/*
5607 5608 5609 5610
 * 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
5611
 *
5612 5613 5614 5615 5616 5617 5618 5619 5620 5621 5622 5623 5624 5625 5626 5627 5628 5629 5630 5631 5632 5633 5634 5635 5636 5637 5638 5639 5640 5641 5642 5643 5644 5645
 * 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.
5646
 */
5647
static int rbd_add_parse_args(const char *buf,
5648
				struct ceph_options **ceph_opts,
5649 5650
				struct rbd_options **opts,
				struct rbd_spec **rbd_spec)
5651
{
A
Alex Elder 已提交
5652
	size_t len;
5653
	char *options;
5654
	const char *mon_addrs;
5655
	char *snap_name;
5656
	size_t mon_addrs_size;
5657
	struct rbd_spec *spec = NULL;
5658
	struct rbd_options *rbd_opts = NULL;
5659
	struct ceph_options *copts;
5660
	int ret;
5661 5662 5663

	/* The first four tokens are required */

5664
	len = next_token(&buf);
5665 5666 5667 5668
	if (!len) {
		rbd_warn(NULL, "no monitor address(es) provided");
		return -EINVAL;
	}
5669
	mon_addrs = buf;
5670
	mon_addrs_size = len + 1;
5671
	buf += len;
5672

5673
	ret = -EINVAL;
5674 5675
	options = dup_token(&buf, NULL);
	if (!options)
5676
		return -ENOMEM;
5677 5678 5679 5680
	if (!*options) {
		rbd_warn(NULL, "no options provided");
		goto out_err;
	}
5681

5682 5683
	spec = rbd_spec_alloc();
	if (!spec)
5684
		goto out_mem;
5685 5686 5687 5688

	spec->pool_name = dup_token(&buf, NULL);
	if (!spec->pool_name)
		goto out_mem;
5689 5690 5691 5692
	if (!*spec->pool_name) {
		rbd_warn(NULL, "no pool name provided");
		goto out_err;
	}
5693

A
Alex Elder 已提交
5694
	spec->image_name = dup_token(&buf, NULL);
5695
	if (!spec->image_name)
5696
		goto out_mem;
5697 5698 5699 5700
	if (!*spec->image_name) {
		rbd_warn(NULL, "no image name provided");
		goto out_err;
	}
5701

5702 5703 5704 5705
	/*
	 * Snapshot name is optional; default is to use "-"
	 * (indicating the head/no snapshot).
	 */
5706
	len = next_token(&buf);
5707
	if (!len) {
5708 5709
		buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */
		len = sizeof (RBD_SNAP_HEAD_NAME) - 1;
5710
	} else if (len > RBD_MAX_SNAP_NAME_LEN) {
5711
		ret = -ENAMETOOLONG;
5712
		goto out_err;
5713
	}
5714 5715
	snap_name = kmemdup(buf, len + 1, GFP_KERNEL);
	if (!snap_name)
5716
		goto out_mem;
5717 5718
	*(snap_name + len) = '\0';
	spec->snap_name = snap_name;
A
Alex Elder 已提交
5719

5720
	/* Initialize all rbd options to the defaults */
5721

5722 5723 5724 5725 5726
	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
5727
	rbd_opts->queue_depth = RBD_QUEUE_DEPTH_DEFAULT;
A
Alex Elder 已提交
5728

5729
	copts = ceph_parse_options(options, mon_addrs,
5730
					mon_addrs + mon_addrs_size - 1,
5731
					parse_rbd_opts_token, rbd_opts);
5732 5733
	if (IS_ERR(copts)) {
		ret = PTR_ERR(copts);
5734 5735
		goto out_err;
	}
5736 5737 5738
	kfree(options);

	*ceph_opts = copts;
5739
	*opts = rbd_opts;
5740
	*rbd_spec = spec;
5741

5742
	return 0;
5743
out_mem:
5744
	ret = -ENOMEM;
A
Alex Elder 已提交
5745
out_err:
5746 5747
	kfree(rbd_opts);
	rbd_spec_put(spec);
5748
	kfree(options);
A
Alex Elder 已提交
5749

5750
	return ret;
5751 5752
}

5753 5754 5755 5756 5757
/*
 * Return pool id (>= 0) or a negative error code.
 */
static int rbd_add_get_pool_id(struct rbd_client *rbdc, const char *pool_name)
{
5758
	struct ceph_options *opts = rbdc->client->options;
5759 5760 5761 5762 5763 5764 5765
	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) {
5766 5767
		ret = ceph_monc_get_version(&rbdc->client->monc, "osdmap",
					    &newest_epoch);
5768 5769 5770 5771
		if (ret < 0)
			return ret;

		if (rbdc->client->osdc.osdmap->epoch < newest_epoch) {
5772
			ceph_osdc_maybe_request_map(&rbdc->client->osdc);
5773
			(void) ceph_monc_wait_osdmap(&rbdc->client->monc,
5774 5775
						     newest_epoch,
						     opts->mount_timeout);
5776 5777 5778 5779 5780 5781 5782 5783 5784 5785
			goto again;
		} else {
			/* the osdmap we have is new enough */
			return -ENOENT;
		}
	}

	return ret;
}

A
Alex Elder 已提交
5786 5787 5788 5789 5790 5791 5792 5793 5794 5795 5796 5797 5798 5799 5800 5801 5802 5803 5804 5805
/*
 * 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;
	char *object_name;
	void *response;
5806
	char *image_id;
5807

A
Alex Elder 已提交
5808 5809 5810
	/*
	 * When probing a parent image, the image id is already
	 * known (and the image name likely is not).  There's no
5811 5812
	 * 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
5813
	 */
5814 5815 5816
	if (rbd_dev->spec->image_id) {
		rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1;

A
Alex Elder 已提交
5817
		return 0;
5818
	}
A
Alex Elder 已提交
5819

A
Alex Elder 已提交
5820 5821 5822 5823
	/*
	 * First, see if the format 2 image id file exists, and if
	 * so, get the image's persistent id from it.
	 */
A
Alex Elder 已提交
5824
	size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name);
A
Alex Elder 已提交
5825 5826 5827
	object_name = kmalloc(size, GFP_NOIO);
	if (!object_name)
		return -ENOMEM;
5828
	sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name);
A
Alex Elder 已提交
5829 5830 5831 5832 5833 5834 5835 5836 5837 5838 5839
	dout("rbd id object name is %s\n", object_name);

	/* 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;
	}

5840 5841
	/* If it doesn't exist we'll assume it's a format 1 image */

5842
	ret = rbd_obj_method_sync(rbd_dev, object_name,
5843
				"rbd", "get_id", NULL, 0,
5844
				response, RBD_IMAGE_ID_LEN_MAX);
5845
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
5846 5847 5848 5849 5850
	if (ret == -ENOENT) {
		image_id = kstrdup("", GFP_KERNEL);
		ret = image_id ? 0 : -ENOMEM;
		if (!ret)
			rbd_dev->image_format = 1;
5851
	} else if (ret >= 0) {
5852 5853 5854
		void *p = response;

		image_id = ceph_extract_encoded_string(&p, p + ret,
A
Alex Elder 已提交
5855
						NULL, GFP_NOIO);
5856
		ret = PTR_ERR_OR_ZERO(image_id);
5857 5858 5859 5860 5861 5862 5863
		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
5864 5865 5866 5867 5868 5869 5870 5871
	}
out:
	kfree(response);
	kfree(object_name);

	return ret;
}

A
Alex Elder 已提交
5872 5873 5874 5875
/*
 * Undo whatever state changes are made by v1 or v2 header info
 * call.
 */
A
Alex Elder 已提交
5876 5877 5878 5879
static void rbd_dev_unprobe(struct rbd_device *rbd_dev)
{
	struct rbd_image_header	*header;

5880
	rbd_dev_parent_put(rbd_dev);
A
Alex Elder 已提交
5881 5882 5883 5884

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

	header = &rbd_dev->header;
5885
	ceph_put_snap_context(header->snapc);
A
Alex Elder 已提交
5886 5887 5888 5889 5890 5891
	kfree(header->snap_sizes);
	kfree(header->snap_names);
	kfree(header->object_prefix);
	memset(header, 0, sizeof (*header));
}

5892
static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev)
5893 5894 5895
{
	int ret;

5896
	ret = rbd_dev_v2_object_prefix(rbd_dev);
5897
	if (ret)
5898 5899
		goto out_err;

5900 5901 5902 5903
	/*
	 * Get the and check features for the image.  Currently the
	 * features are assumed to never change.
	 */
5904
	ret = rbd_dev_v2_features(rbd_dev);
5905
	if (ret)
5906
		goto out_err;
5907

5908 5909 5910 5911 5912 5913 5914
	/* 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;
	}
5915
	/* No support for crypto and compression type format 2 images */
5916

A
Alex Elder 已提交
5917
	return 0;
5918
out_err:
A
Alex Elder 已提交
5919
	rbd_dev->header.features = 0;
5920 5921
	kfree(rbd_dev->header.object_prefix);
	rbd_dev->header.object_prefix = NULL;
5922 5923

	return ret;
5924 5925
}

5926 5927 5928 5929 5930 5931
/*
 * @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
5932
{
5933
	struct rbd_device *parent = NULL;
5934 5935 5936 5937 5938
	int ret;

	if (!rbd_dev->parent_spec)
		return 0;

5939 5940 5941 5942 5943 5944
	if (++depth > RBD_MAX_PARENT_CHAIN_LEN) {
		pr_info("parent chain is too long (%d)\n", depth);
		ret = -EINVAL;
		goto out_err;
	}

5945
	parent = __rbd_dev_create(rbd_dev->rbd_client, rbd_dev->parent_spec);
5946 5947
	if (!parent) {
		ret = -ENOMEM;
5948
		goto out_err;
5949 5950 5951 5952 5953 5954 5955 5956
	}

	/*
	 * 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);
5957

5958
	ret = rbd_dev_image_probe(parent, depth);
5959 5960
	if (ret < 0)
		goto out_err;
5961

5962
	rbd_dev->parent = parent;
5963
	atomic_set(&rbd_dev->parent_ref, 1);
5964
	return 0;
5965

5966
out_err:
5967
	rbd_dev_unparent(rbd_dev);
5968
	rbd_dev_destroy(parent);
5969 5970 5971
	return ret;
}

I
Ilya Dryomov 已提交
5972 5973 5974 5975
/*
 * rbd_dev->header_rwsem must be locked for write and will be unlocked
 * upon return.
 */
5976
static int rbd_dev_device_setup(struct rbd_device *rbd_dev)
5977
{
A
Alex Elder 已提交
5978
	int ret;
A
Alex Elder 已提交
5979

5980
	/* Record our major and minor device numbers. */
A
Alex Elder 已提交
5981

5982 5983 5984
	if (!single_major) {
		ret = register_blkdev(0, rbd_dev->name);
		if (ret < 0)
5985
			goto err_out_unlock;
5986 5987 5988 5989 5990 5991 5992

		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
5993 5994 5995 5996 5997 5998 5999

	/* Set up the blkdev mapping. */

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

6000
	ret = rbd_dev_mapping_set(rbd_dev);
A
Alex Elder 已提交
6001 6002
	if (ret)
		goto err_out_disk;
I
Ilya Dryomov 已提交
6003

6004
	set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE);
6005
	set_disk_ro(rbd_dev->disk, rbd_dev->mapping.read_only);
6006

6007 6008
	dev_set_name(&rbd_dev->dev, "%d", rbd_dev->dev_id);
	ret = device_add(&rbd_dev->dev);
6009
	if (ret)
6010
		goto err_out_mapping;
A
Alex Elder 已提交
6011 6012 6013

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

6014
	set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
I
Ilya Dryomov 已提交
6015
	up_write(&rbd_dev->header_rwsem);
A
Alex Elder 已提交
6016

6017 6018 6019 6020
	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
6021
	add_disk(rbd_dev->disk);
6022 6023 6024
	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
6025 6026

	return ret;
6027

6028 6029
err_out_mapping:
	rbd_dev_mapping_clear(rbd_dev);
A
Alex Elder 已提交
6030 6031 6032
err_out_disk:
	rbd_free_disk(rbd_dev);
err_out_blkdev:
6033 6034
	if (!single_major)
		unregister_blkdev(rbd_dev->major, rbd_dev->name);
I
Ilya Dryomov 已提交
6035 6036
err_out_unlock:
	up_write(&rbd_dev->header_rwsem);
A
Alex Elder 已提交
6037 6038 6039
	return ret;
}

A
Alex Elder 已提交
6040 6041 6042
static int rbd_dev_header_name(struct rbd_device *rbd_dev)
{
	struct rbd_spec *spec = rbd_dev->spec;
6043
	int ret;
A
Alex Elder 已提交
6044 6045 6046 6047 6048

	/* Record the header object name for this rbd image. */

	rbd_assert(rbd_image_format_valid(rbd_dev->image_format));

6049
	rbd_dev->header_oloc.pool = rbd_dev->layout.pool_id;
A
Alex Elder 已提交
6050
	if (rbd_dev->image_format == 1)
6051 6052
		ret = ceph_oid_aprintf(&rbd_dev->header_oid, GFP_KERNEL, "%s%s",
				       spec->image_name, RBD_SUFFIX);
A
Alex Elder 已提交
6053
	else
6054 6055
		ret = ceph_oid_aprintf(&rbd_dev->header_oid, GFP_KERNEL, "%s%s",
				       RBD_HEADER_PREFIX, spec->image_id);
A
Alex Elder 已提交
6056

6057
	return ret;
A
Alex Elder 已提交
6058 6059
}

6060 6061
static void rbd_dev_image_release(struct rbd_device *rbd_dev)
{
A
Alex Elder 已提交
6062 6063 6064 6065 6066
	rbd_dev_unprobe(rbd_dev);
	rbd_dev->image_format = 0;
	kfree(rbd_dev->spec->image_id);
	rbd_dev->spec->image_id = NULL;

6067 6068 6069
	rbd_dev_destroy(rbd_dev);
}

6070 6071
/*
 * Probe for the existence of the header object for the given rbd
6072 6073 6074
 * 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.
6075
 */
6076
static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth)
6077 6078 6079 6080
{
	int ret;

	/*
A
Alex Elder 已提交
6081 6082 6083 6084
	 * 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.
6085 6086 6087
	 */
	ret = rbd_dev_image_id(rbd_dev);
	if (ret)
6088 6089
		return ret;

A
Alex Elder 已提交
6090 6091 6092 6093
	ret = rbd_dev_header_name(rbd_dev);
	if (ret)
		goto err_out_format;

6094
	if (!depth) {
6095
		ret = rbd_register_watch(rbd_dev);
6096 6097 6098 6099 6100
		if (ret) {
			if (ret == -ENOENT)
				pr_info("image %s/%s does not exist\n",
					rbd_dev->spec->pool_name,
					rbd_dev->spec->image_name);
6101
			goto err_out_format;
6102
		}
6103
	}
6104

6105
	ret = rbd_dev_header_info(rbd_dev);
6106
	if (ret)
6107
		goto err_out_watch;
A
Alex Elder 已提交
6108

6109 6110 6111 6112 6113 6114
	/*
	 * 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.
	 */
6115
	if (!depth)
6116 6117 6118
		ret = rbd_spec_fill_snap_id(rbd_dev);
	else
		ret = rbd_spec_fill_names(rbd_dev);
6119 6120 6121 6122 6123 6124
	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
6125
		goto err_out_probe;
6126
	}
6127

6128 6129 6130 6131 6132 6133 6134 6135 6136
	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.
		 */
6137
		if (!depth && rbd_dev->parent_spec)
6138 6139 6140 6141
			rbd_warn(rbd_dev,
				 "WARNING: kernel layering is EXPERIMENTAL!");
	}

6142
	ret = rbd_dev_probe_parent(rbd_dev, depth);
A
Alex Elder 已提交
6143 6144 6145 6146
	if (ret)
		goto err_out_probe;

	dout("discovered format %u image, header name is %s\n",
6147
		rbd_dev->image_format, rbd_dev->header_oid.name);
A
Alex Elder 已提交
6148
	return 0;
6149

A
Alex Elder 已提交
6150 6151
err_out_probe:
	rbd_dev_unprobe(rbd_dev);
6152
err_out_watch:
6153
	if (!depth)
6154
		rbd_unregister_watch(rbd_dev);
A
Alex Elder 已提交
6155 6156
err_out_format:
	rbd_dev->image_format = 0;
6157 6158
	kfree(rbd_dev->spec->image_id);
	rbd_dev->spec->image_id = NULL;
6159 6160 6161
	return ret;
}

6162 6163 6164
static ssize_t do_rbd_add(struct bus_type *bus,
			  const char *buf,
			  size_t count)
6165
{
6166
	struct rbd_device *rbd_dev = NULL;
6167
	struct ceph_options *ceph_opts = NULL;
6168
	struct rbd_options *rbd_opts = NULL;
6169
	struct rbd_spec *spec = NULL;
6170
	struct rbd_client *rbdc;
6171
	bool read_only;
6172
	int rc;
6173 6174 6175 6176 6177

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

	/* parse add command */
6178
	rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec);
6179
	if (rc < 0)
6180
		goto out;
6181

6182 6183 6184
	rbdc = rbd_get_client(ceph_opts);
	if (IS_ERR(rbdc)) {
		rc = PTR_ERR(rbdc);
6185
		goto err_out_args;
6186
	}
6187 6188

	/* pick the pool */
6189
	rc = rbd_add_get_pool_id(rbdc, spec->pool_name);
6190 6191 6192
	if (rc < 0) {
		if (rc == -ENOENT)
			pr_info("pool %s does not exist\n", spec->pool_name);
6193
		goto err_out_client;
6194
	}
A
Alex Elder 已提交
6195
	spec->pool_id = (u64)rc;
6196

6197
	rbd_dev = rbd_dev_create(rbdc, spec, rbd_opts);
6198 6199
	if (!rbd_dev) {
		rc = -ENOMEM;
6200
		goto err_out_client;
6201
	}
6202 6203
	rbdc = NULL;		/* rbd_dev now owns this */
	spec = NULL;		/* rbd_dev now owns this */
6204
	rbd_opts = NULL;	/* rbd_dev now owns this */
6205

I
Ilya Dryomov 已提交
6206
	down_write(&rbd_dev->header_rwsem);
6207
	rc = rbd_dev_image_probe(rbd_dev, 0);
6208
	if (rc < 0)
6209
		goto err_out_rbd_dev;
6210

6211 6212
	/* If we are mapping a snapshot it must be marked read-only */

6213
	read_only = rbd_dev->opts->read_only;
6214 6215 6216 6217
	if (rbd_dev->spec->snap_id != CEPH_NOSNAP)
		read_only = true;
	rbd_dev->mapping.read_only = read_only;

6218
	rc = rbd_dev_device_setup(rbd_dev);
A
Alex Elder 已提交
6219
	if (rc) {
6220
		/*
6221
		 * rbd_unregister_watch() can't be moved into
6222 6223 6224
		 * rbd_dev_image_release() without refactoring, see
		 * commit 1f3ef78861ac.
		 */
6225
		rbd_unregister_watch(rbd_dev);
A
Alex Elder 已提交
6226
		rbd_dev_image_release(rbd_dev);
6227
		goto out;
A
Alex Elder 已提交
6228 6229
	}

6230 6231 6232 6233
	rc = count;
out:
	module_put(THIS_MODULE);
	return rc;
6234

6235
err_out_rbd_dev:
I
Ilya Dryomov 已提交
6236
	up_write(&rbd_dev->header_rwsem);
6237
	rbd_dev_destroy(rbd_dev);
6238
err_out_client:
6239
	rbd_put_client(rbdc);
6240
err_out_args:
6241
	rbd_spec_put(spec);
6242
	kfree(rbd_opts);
6243
	goto out;
6244 6245
}

6246 6247 6248 6249 6250 6251 6252 6253 6254 6255 6256 6257 6258 6259 6260 6261 6262
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);
}

6263
static void rbd_dev_device_release(struct rbd_device *rbd_dev)
6264 6265
{
	rbd_free_disk(rbd_dev);
6266 6267 6268 6269 6270

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

6271
	clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
6272
	device_del(&rbd_dev->dev);
A
Alex Elder 已提交
6273
	rbd_dev_mapping_clear(rbd_dev);
6274 6275
	if (!single_major)
		unregister_blkdev(rbd_dev->major, rbd_dev->name);
6276 6277
}

6278 6279
static void rbd_dev_remove_parent(struct rbd_device *rbd_dev)
{
A
Alex Elder 已提交
6280
	while (rbd_dev->parent) {
6281 6282 6283 6284 6285 6286 6287 6288 6289 6290 6291 6292
		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
6293
		rbd_assert(second);
6294
		rbd_dev_image_release(second);
A
Alex Elder 已提交
6295 6296 6297 6298
		first->parent = NULL;
		first->parent_overlap = 0;

		rbd_assert(first->parent_spec);
6299 6300 6301 6302 6303
		rbd_spec_put(first->parent_spec);
		first->parent_spec = NULL;
	}
}

6304 6305 6306
static ssize_t do_rbd_remove(struct bus_type *bus,
			     const char *buf,
			     size_t count)
6307 6308
{
	struct rbd_device *rbd_dev = NULL;
6309 6310
	struct list_head *tmp;
	int dev_id;
6311
	unsigned long ul;
6312
	bool already = false;
6313
	int ret;
6314

6315
	ret = kstrtoul(buf, 10, &ul);
6316 6317
	if (ret)
		return ret;
6318 6319

	/* convert to int; abort if we lost anything in the conversion */
6320 6321
	dev_id = (int)ul;
	if (dev_id != ul)
6322 6323
		return -EINVAL;

6324 6325 6326 6327 6328 6329 6330 6331
	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;
		}
6332
	}
6333 6334 6335 6336 6337
	if (!ret) {
		spin_lock_irq(&rbd_dev->lock);
		if (rbd_dev->open_count)
			ret = -EBUSY;
		else
6338 6339
			already = test_and_set_bit(RBD_DEV_FLAG_REMOVING,
							&rbd_dev->flags);
6340 6341 6342
		spin_unlock_irq(&rbd_dev->lock);
	}
	spin_unlock(&rbd_dev_list_lock);
6343
	if (ret < 0 || already)
6344
		return ret;
6345

6346 6347 6348 6349
	down_write(&rbd_dev->lock_rwsem);
	if (__rbd_is_lock_owner(rbd_dev))
		rbd_unlock(rbd_dev);
	up_write(&rbd_dev->lock_rwsem);
6350
	rbd_unregister_watch(rbd_dev);
6351

6352 6353 6354 6355 6356 6357
	/*
	 * 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.
	 */
6358
	rbd_dev_device_release(rbd_dev);
6359
	rbd_dev_image_release(rbd_dev);
A
Alex Elder 已提交
6360

6361
	return count;
6362 6363
}

6364 6365 6366 6367 6368 6369 6370 6371 6372 6373 6374 6375 6376 6377 6378 6379 6380
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);
}

6381 6382
/*
 * create control files in sysfs
6383
 * /sys/bus/rbd/...
6384 6385 6386
 */
static int rbd_sysfs_init(void)
{
6387
	int ret;
6388

6389
	ret = device_register(&rbd_root_dev);
A
Alex Elder 已提交
6390
	if (ret < 0)
6391
		return ret;
6392

6393 6394 6395
	ret = bus_register(&rbd_bus_type);
	if (ret < 0)
		device_unregister(&rbd_root_dev);
6396 6397 6398 6399 6400 6401

	return ret;
}

static void rbd_sysfs_cleanup(void)
{
6402
	bus_unregister(&rbd_bus_type);
6403
	device_unregister(&rbd_root_dev);
6404 6405
}

6406 6407 6408
static int rbd_slab_init(void)
{
	rbd_assert(!rbd_img_request_cache);
G
Geliang Tang 已提交
6409
	rbd_img_request_cache = KMEM_CACHE(rbd_img_request, 0);
6410 6411 6412 6413
	if (!rbd_img_request_cache)
		return -ENOMEM;

	rbd_assert(!rbd_obj_request_cache);
G
Geliang Tang 已提交
6414
	rbd_obj_request_cache = KMEM_CACHE(rbd_obj_request, 0);
6415 6416 6417 6418 6419
	if (!rbd_obj_request_cache)
		goto out_err;

	rbd_assert(!rbd_segment_name_cache);
	rbd_segment_name_cache = kmem_cache_create("rbd_segment_name",
6420
					CEPH_MAX_OID_NAME_LEN + 1, 1, 0, NULL);
6421
	if (rbd_segment_name_cache)
6422
		return 0;
6423
out_err:
6424 6425
	kmem_cache_destroy(rbd_obj_request_cache);
	rbd_obj_request_cache = NULL;
6426

6427 6428 6429
	kmem_cache_destroy(rbd_img_request_cache);
	rbd_img_request_cache = NULL;

6430 6431 6432 6433 6434
	return -ENOMEM;
}

static void rbd_slab_exit(void)
{
6435 6436 6437 6438
	rbd_assert(rbd_segment_name_cache);
	kmem_cache_destroy(rbd_segment_name_cache);
	rbd_segment_name_cache = NULL;

6439 6440 6441 6442
	rbd_assert(rbd_obj_request_cache);
	kmem_cache_destroy(rbd_obj_request_cache);
	rbd_obj_request_cache = NULL;

6443 6444 6445 6446 6447
	rbd_assert(rbd_img_request_cache);
	kmem_cache_destroy(rbd_img_request_cache);
	rbd_img_request_cache = NULL;
}

A
Alex Elder 已提交
6448
static int __init rbd_init(void)
6449 6450 6451
{
	int rc;

6452 6453 6454 6455
	if (!libceph_compatible(NULL)) {
		rbd_warn(NULL, "libceph incompatibility (quitting)");
		return -EINVAL;
	}
I
Ilya Dryomov 已提交
6456

6457
	rc = rbd_slab_init();
6458 6459
	if (rc)
		return rc;
I
Ilya Dryomov 已提交
6460

6461 6462
	/*
	 * The number of active work items is limited by the number of
I
Ilya Dryomov 已提交
6463
	 * rbd devices * queue depth, so leave @max_active at default.
6464 6465 6466 6467 6468 6469 6470
	 */
	rbd_wq = alloc_workqueue(RBD_DRV_NAME, WQ_MEM_RECLAIM, 0);
	if (!rbd_wq) {
		rc = -ENOMEM;
		goto err_out_slab;
	}

6471 6472 6473 6474
	if (single_major) {
		rbd_major = register_blkdev(0, RBD_DRV_NAME);
		if (rbd_major < 0) {
			rc = rbd_major;
6475
			goto err_out_wq;
6476 6477 6478
		}
	}

6479 6480
	rc = rbd_sysfs_init();
	if (rc)
6481 6482 6483 6484 6485 6486
		goto err_out_blkdev;

	if (single_major)
		pr_info("loaded (major %d)\n", rbd_major);
	else
		pr_info("loaded\n");
6487

I
Ilya Dryomov 已提交
6488 6489
	return 0;

6490 6491 6492
err_out_blkdev:
	if (single_major)
		unregister_blkdev(rbd_major, RBD_DRV_NAME);
6493 6494
err_out_wq:
	destroy_workqueue(rbd_wq);
I
Ilya Dryomov 已提交
6495 6496
err_out_slab:
	rbd_slab_exit();
6497
	return rc;
6498 6499
}

A
Alex Elder 已提交
6500
static void __exit rbd_exit(void)
6501
{
I
Ilya Dryomov 已提交
6502
	ida_destroy(&rbd_dev_id_ida);
6503
	rbd_sysfs_cleanup();
6504 6505
	if (single_major)
		unregister_blkdev(rbd_major, RBD_DRV_NAME);
6506
	destroy_workqueue(rbd_wq);
6507
	rbd_slab_exit();
6508 6509 6510 6511 6512
}

module_init(rbd_init);
module_exit(rbd_exit);

A
Alex Elder 已提交
6513
MODULE_AUTHOR("Alex Elder <elder@inktank.com>");
6514 6515 6516 6517 6518
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>");

6519
MODULE_DESCRIPTION("RADOS Block Device (RBD) driver");
6520
MODULE_LICENSE("GPL");