rbd.c 146.2 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 34

 */

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

#include <linux/kernel.h>
#include <linux/device.h>
#include <linux/module.h>
C
Christoph Hellwig 已提交
41
#include <linux/blk-mq.h>
42 43
#include <linux/fs.h>
#include <linux/blkdev.h>
44
#include <linux/slab.h>
45
#include <linux/idr.h>
I
Ilya Dryomov 已提交
46
#include <linux/workqueue.h>
47 48 49

#include "rbd_types.h"

A
Alex Elder 已提交
50 51
#define RBD_DEBUG	/* Activate rbd_assert() calls */

A
Alex Elder 已提交
52 53 54 55 56 57 58 59 60
/*
 * 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)

61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93
/*
 * 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
94
#define RBD_DRV_NAME "rbd"
95

96 97
#define RBD_MINORS_PER_MAJOR		256
#define RBD_SINGLE_MAJOR_PART_SHIFT	4
98

99 100 101 102
#define RBD_SNAP_DEV_NAME_PREFIX	"snap_"
#define RBD_MAX_SNAP_NAME_LEN	\
			(NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1))

103
#define RBD_MAX_SNAP_COUNT	510	/* allows max snapc to fit in 4KB */
104 105 106

#define RBD_SNAP_HEAD_NAME	"-"

107 108
#define	BAD_SNAP_INDEX	U32_MAX		/* invalid index into snap array */

109 110
/* This allows a single page to hold an image name sent by OSD */
#define RBD_IMAGE_NAME_LEN_MAX	(PAGE_SIZE - sizeof (__le32) - 1)
111
#define RBD_IMAGE_ID_LEN_MAX	64
112

113
#define RBD_OBJ_PREFIX_LEN_MAX	64
A
Alex Elder 已提交
114

A
Alex Elder 已提交
115 116
/* Feature bits */

A
Alex Elder 已提交
117 118 119 120
#define RBD_FEATURE_LAYERING	(1<<0)
#define RBD_FEATURE_STRIPINGV2	(1<<1)
#define RBD_FEATURES_ALL \
	    (RBD_FEATURE_LAYERING | RBD_FEATURE_STRIPINGV2)
A
Alex Elder 已提交
121 122 123

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

124
#define RBD_FEATURES_SUPPORTED	(RBD_FEATURES_ALL)
A
Alex Elder 已提交
125

A
Alex Elder 已提交
126 127 128 129 130 131
/*
 * An RBD device name will be "rbd#", where the "rbd" comes from
 * RBD_DRV_NAME above, and # is a unique integer identifier.
 * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big
 * enough to hold all possible device names.
 */
132
#define DEV_NAME_LEN		32
A
Alex Elder 已提交
133
#define MAX_INT_FORMAT_WIDTH	((5 * sizeof (int)) / 2 + 1)
134 135 136 137 138

/*
 * block device image metadata (in-memory version)
 */
struct rbd_image_header {
139
	/* These six fields never change for a given rbd image */
140
	char *object_prefix;
141 142 143
	__u8 obj_order;
	__u8 crypt_type;
	__u8 comp_type;
144 145 146
	u64 stripe_unit;
	u64 stripe_count;
	u64 features;		/* Might be changeable someday? */
147

A
Alex Elder 已提交
148 149 150
	/* The remaining fields need to be updated occasionally */
	u64 image_size;
	struct ceph_snap_context *snapc;
151 152
	char *snap_names;	/* format 1 only */
	u64 *snap_sizes;	/* format 1 only */
153 154
};

155 156 157 158
/*
 * An rbd image specification.
 *
 * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely
A
Alex Elder 已提交
159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178
 * 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).
179 180 181
 */
struct rbd_spec {
	u64		pool_id;
182
	const char	*pool_name;
183

184 185
	const char	*image_id;
	const char	*image_name;
186 187

	u64		snap_id;
188
	const char	*snap_name;
189 190 191 192

	struct kref	kref;
};

193
/*
A
Alex Elder 已提交
194
 * an instance of the client.  multiple devices may share an rbd client.
195 196 197 198 199 200 201
 */
struct rbd_client {
	struct ceph_client	*client;
	struct kref		kref;
	struct list_head	node;
};

A
Alex Elder 已提交
202 203 204 205 206 207 208 209
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 *);

210 211 212
enum obj_request_type {
	OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES
};
A
Alex Elder 已提交
213

G
Guangliang Zhao 已提交
214 215 216
enum obj_operation_type {
	OBJ_OP_WRITE,
	OBJ_OP_READ,
217
	OBJ_OP_DISCARD,
G
Guangliang Zhao 已提交
218 219
};

220 221
enum obj_req_flags {
	OBJ_REQ_DONE,		/* completion flag: not done = 0, done = 1 */
222
	OBJ_REQ_IMG_DATA,	/* object usage: standalone = 0, image = 1 */
223 224
	OBJ_REQ_KNOWN,		/* EXISTS flag valid: no = 0, yes = 1 */
	OBJ_REQ_EXISTS,		/* target exists: no = 0, yes = 1 */
225 226
};

A
Alex Elder 已提交
227 228 229 230
struct rbd_obj_request {
	const char		*object_name;
	u64			offset;		/* object start byte */
	u64			length;		/* bytes from offset */
231
	unsigned long		flags;
A
Alex Elder 已提交
232

233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257
	/*
	 * 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
258 259 260
	u32			which;		/* posn image request list */

	enum obj_request_type	type;
261 262 263 264 265 266 267
	union {
		struct bio	*bio_list;
		struct {
			struct page	**pages;
			u32		page_count;
		};
	};
268
	struct page		**copyup_pages;
269
	u32			copyup_page_count;
A
Alex Elder 已提交
270 271 272 273

	struct ceph_osd_request	*osd_req;

	u64			xferred;	/* bytes transferred */
274
	int			result;
A
Alex Elder 已提交
275 276

	rbd_obj_callback_t	callback;
277
	struct completion	completion;
A
Alex Elder 已提交
278 279 280 281

	struct kref		kref;
};

A
Alex Elder 已提交
282
enum img_req_flags {
283 284
	IMG_REQ_WRITE,		/* I/O direction: read = 0, write = 1 */
	IMG_REQ_CHILD,		/* initiator: block = 0, child image = 1 */
285
	IMG_REQ_LAYERED,	/* ENOENT handling: normal = 0, layered = 1 */
286
	IMG_REQ_DISCARD,	/* discard: normal = 0, discard request = 1 */
A
Alex Elder 已提交
287 288
};

A
Alex Elder 已提交
289 290 291 292
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
293
	unsigned long		flags;
A
Alex Elder 已提交
294
	union {
295
		u64			snap_id;	/* for reads */
A
Alex Elder 已提交
296
		struct ceph_snap_context *snapc;	/* for writes */
297 298 299 300
	};
	union {
		struct request		*rq;		/* block request */
		struct rbd_obj_request	*obj_request;	/* obj req initiator */
A
Alex Elder 已提交
301
	};
302
	struct page		**copyup_pages;
303
	u32			copyup_page_count;
A
Alex Elder 已提交
304 305 306
	spinlock_t		completion_lock;/* protects next_completion */
	u32			next_completion;
	rbd_img_callback_t	callback;
307
	u64			xferred;/* aggregate bytes transferred */
308
	int			result;	/* first nonzero obj_request result */
A
Alex Elder 已提交
309 310 311 312 313 314 315 316

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

	struct kref		kref;
};

#define for_each_obj_request(ireq, oreq) \
317
	list_for_each_entry(oreq, &(ireq)->obj_requests, links)
A
Alex Elder 已提交
318
#define for_each_obj_request_from(ireq, oreq) \
319
	list_for_each_entry_from(oreq, &(ireq)->obj_requests, links)
A
Alex Elder 已提交
320
#define for_each_obj_request_safe(ireq, oreq, n) \
321
	list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links)
A
Alex Elder 已提交
322

A
Alex Elder 已提交
323
struct rbd_mapping {
A
Alex Elder 已提交
324
	u64                     size;
A
Alex Elder 已提交
325
	u64                     features;
A
Alex Elder 已提交
326 327 328
	bool			read_only;
};

329 330 331 332
/*
 * a single device
 */
struct rbd_device {
A
Alex Elder 已提交
333
	int			dev_id;		/* blkdev unique id */
334 335

	int			major;		/* blkdev assigned major */
336
	int			minor;
337 338
	struct gendisk		*disk;		/* blkdev's gendisk and rq */

339
	u32			image_format;	/* Either 1 or 2 */
340 341 342 343
	struct rbd_client	*rbd_client;

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

344
	spinlock_t		lock;		/* queue, flags, open_count */
345 346

	struct rbd_image_header	header;
347
	unsigned long		flags;		/* possibly lock protected */
348
	struct rbd_spec		*spec;
349

350
	char			*header_name;
351

352 353
	struct ceph_file_layout	layout;

354
	struct ceph_osd_event   *watch_event;
355
	struct rbd_obj_request	*watch_request;
356

357 358
	struct rbd_spec		*parent_spec;
	u64			parent_overlap;
359
	atomic_t		parent_ref;
360
	struct rbd_device	*parent;
361

C
Christoph Hellwig 已提交
362 363 364
	/* Block layer tags. */
	struct blk_mq_tag_set	tag_set;

365 366
	/* protects updating the header */
	struct rw_semaphore     header_rwsem;
A
Alex Elder 已提交
367 368

	struct rbd_mapping	mapping;
369 370

	struct list_head	node;
371 372 373

	/* sysfs related */
	struct device		dev;
374
	unsigned long		open_count;	/* protected by lock */
375 376
};

377 378 379 380 381 382 383
/*
 * 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.
 */
384 385
enum rbd_dev_flags {
	RBD_DEV_FLAG_EXISTS,	/* mapped snapshot has not been deleted */
386
	RBD_DEV_FLAG_REMOVING,	/* this mapping is being removed */
387 388
};

389
static DEFINE_MUTEX(client_mutex);	/* Serialize client creation */
390

391
static LIST_HEAD(rbd_dev_list);    /* devices */
392 393
static DEFINE_SPINLOCK(rbd_dev_list_lock);

A
Alex Elder 已提交
394 395
static LIST_HEAD(rbd_client_list);		/* clients */
static DEFINE_SPINLOCK(rbd_client_list_lock);
396

397 398
/* Slab caches for frequently-allocated structures */

399
static struct kmem_cache	*rbd_img_request_cache;
400
static struct kmem_cache	*rbd_obj_request_cache;
401
static struct kmem_cache	*rbd_segment_name_cache;
402

403
static int rbd_major;
404 405
static DEFINE_IDA(rbd_dev_id_ida);

406 407
static struct workqueue_struct *rbd_wq;

408 409 410 411 412 413 414 415
/*
 * 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)");

416 417
static int rbd_img_request_submit(struct rbd_img_request *img_request);

418
static void rbd_dev_device_release(struct device *dev);
419

A
Alex Elder 已提交
420 421 422 423
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);
424 425 426 427
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);
428
static int rbd_dev_image_probe(struct rbd_device *rbd_dev, bool mapping);
429
static void rbd_spec_put(struct rbd_spec *spec);
A
Alex Elder 已提交
430

431 432
static int rbd_dev_id_to_minor(int dev_id)
{
433
	return dev_id << RBD_SINGLE_MAJOR_PART_SHIFT;
434 435 436 437
}

static int minor_to_rbd_dev_id(int minor)
{
438
	return minor >> RBD_SINGLE_MAJOR_PART_SHIFT;
439 440
}

441 442
static BUS_ATTR(add, S_IWUSR, NULL, rbd_add);
static BUS_ATTR(remove, S_IWUSR, NULL, rbd_remove);
443 444
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);
445 446 447 448

static struct attribute *rbd_bus_attrs[] = {
	&bus_attr_add.attr,
	&bus_attr_remove.attr,
449 450
	&bus_attr_add_single_major.attr,
	&bus_attr_remove_single_major.attr,
451
	NULL,
A
Alex Elder 已提交
452
};
453 454 455 456

static umode_t rbd_bus_is_visible(struct kobject *kobj,
				  struct attribute *attr, int index)
{
457 458 459 460 461
	if (!single_major &&
	    (attr == &bus_attr_add_single_major.attr ||
	     attr == &bus_attr_remove_single_major.attr))
		return 0;

462 463 464 465 466 467 468 469
	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
470 471 472

static struct bus_type rbd_bus_type = {
	.name		= "rbd",
473
	.bus_groups	= rbd_bus_groups,
A
Alex Elder 已提交
474 475 476 477 478 479 480 481 482 483 484
};

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
485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511
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
512 513 514 515 516 517 518 519 520 521 522 523
#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 */
524

525
static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request);
526 527
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
528

A
Alex Elder 已提交
529
static int rbd_dev_refresh(struct rbd_device *rbd_dev);
530
static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev);
531
static int rbd_dev_header_info(struct rbd_device *rbd_dev);
532
static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev);
533 534
static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
					u64 snap_id);
535 536 537 538 539
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);
static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name);
540

541 542
static int rbd_open(struct block_device *bdev, fmode_t mode)
{
A
Alex Elder 已提交
543
	struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
544
	bool removing = false;
545

A
Alex Elder 已提交
546
	if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only)
547 548
		return -EROFS;

549
	spin_lock_irq(&rbd_dev->lock);
550 551 552 553
	if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags))
		removing = true;
	else
		rbd_dev->open_count++;
554
	spin_unlock_irq(&rbd_dev->lock);
555 556 557
	if (removing)
		return -ENOENT;

A
Alex Elder 已提交
558
	(void) get_device(&rbd_dev->dev);
559

560 561 562
	return 0;
}

563
static void rbd_release(struct gendisk *disk, fmode_t mode)
564 565
{
	struct rbd_device *rbd_dev = disk->private_data;
566 567
	unsigned long open_count_before;

568
	spin_lock_irq(&rbd_dev->lock);
569
	open_count_before = rbd_dev->open_count--;
570
	spin_unlock_irq(&rbd_dev->lock);
571
	rbd_assert(open_count_before > 0);
572

A
Alex Elder 已提交
573
	put_device(&rbd_dev->dev);
574 575
}

G
Guangliang Zhao 已提交
576 577
static int rbd_ioctl_set_ro(struct rbd_device *rbd_dev, unsigned long arg)
{
578
	int ret = 0;
G
Guangliang Zhao 已提交
579 580
	int val;
	bool ro;
581
	bool ro_changed = false;
G
Guangliang Zhao 已提交
582

583
	/* get_user() may sleep, so call it before taking rbd_dev->lock */
G
Guangliang Zhao 已提交
584 585 586 587 588 589 590 591
	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;

592 593 594 595 596 597 598
	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
599 600
	if (rbd_dev->mapping.read_only != ro) {
		rbd_dev->mapping.read_only = ro;
601
		ro_changed = true;
G
Guangliang Zhao 已提交
602 603
	}

604 605 606 607 608 609 610
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
611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637
}

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 */

638 639 640
static const struct block_device_operations rbd_bd_ops = {
	.owner			= THIS_MODULE,
	.open			= rbd_open,
641
	.release		= rbd_release,
G
Guangliang Zhao 已提交
642 643 644 645
	.ioctl			= rbd_ioctl,
#ifdef CONFIG_COMPAT
	.compat_ioctl		= rbd_compat_ioctl,
#endif
646 647 648
};

/*
649
 * Initialize an rbd client instance.  Success or not, this function
650
 * consumes ceph_opts.  Caller holds client_mutex.
651
 */
652
static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts)
653 654 655 656
{
	struct rbd_client *rbdc;
	int ret = -ENOMEM;

A
Alex Elder 已提交
657
	dout("%s:\n", __func__);
658 659 660 661 662 663 664
	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
665
	rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0);
666
	if (IS_ERR(rbdc->client))
667
		goto out_rbdc;
A
Alex Elder 已提交
668
	ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */
669 670 671

	ret = ceph_open_session(rbdc->client);
	if (ret < 0)
672
		goto out_client;
673

A
Alex Elder 已提交
674
	spin_lock(&rbd_client_list_lock);
675
	list_add_tail(&rbdc->node, &rbd_client_list);
A
Alex Elder 已提交
676
	spin_unlock(&rbd_client_list_lock);
677

A
Alex Elder 已提交
678
	dout("%s: rbdc %p\n", __func__, rbdc);
679

680
	return rbdc;
681
out_client:
682
	ceph_destroy_client(rbdc->client);
683
out_rbdc:
684 685
	kfree(rbdc);
out_opt:
A
Alex Elder 已提交
686 687
	if (ceph_opts)
		ceph_destroy_options(ceph_opts);
A
Alex Elder 已提交
688 689
	dout("%s: error %d\n", __func__, ret);

V
Vasiliy Kulikov 已提交
690
	return ERR_PTR(ret);
691 692
}

693 694 695 696 697 698 699
static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc)
{
	kref_get(&rbdc->kref);

	return rbdc;
}

700
/*
701 702
 * Find a ceph client with specific addr and configuration.  If
 * found, bump its reference count.
703
 */
704
static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts)
705 706
{
	struct rbd_client *client_node;
707
	bool found = false;
708

A
Alex Elder 已提交
709
	if (ceph_opts->flags & CEPH_OPT_NOSHARE)
710 711
		return NULL;

712 713 714
	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)) {
715 716
			__rbd_get_client(client_node);

717 718 719 720 721 722 723
			found = true;
			break;
		}
	}
	spin_unlock(&rbd_client_list_lock);

	return found ? client_node : NULL;
724 725
}

726
/*
727
 * (Per device) rbd map options
728 729 730 731 732 733
 */
enum {
	Opt_last_int,
	/* int args above */
	Opt_last_string,
	/* string args above */
A
Alex Elder 已提交
734 735
	Opt_read_only,
	Opt_read_write,
736
	Opt_err
737 738
};

A
Alex Elder 已提交
739
static match_table_t rbd_opts_tokens = {
740 741
	/* int args above */
	/* string args above */
A
Alex Elder 已提交
742
	{Opt_read_only, "read_only"},
A
Alex Elder 已提交
743 744 745
	{Opt_read_only, "ro"},		/* Alternate spelling */
	{Opt_read_write, "read_write"},
	{Opt_read_write, "rw"},		/* Alternate spelling */
746
	{Opt_err, NULL}
747 748
};

A
Alex Elder 已提交
749 750 751 752 753 754
struct rbd_options {
	bool	read_only;
};

#define RBD_READ_ONLY_DEFAULT	false

755 756
static int parse_rbd_opts_token(char *c, void *private)
{
A
Alex Elder 已提交
757
	struct rbd_options *rbd_opts = private;
758 759 760
	substring_t argstr[MAX_OPT_ARGS];
	int token, intval, ret;

A
Alex Elder 已提交
761
	token = match_token(c, rbd_opts_tokens, argstr);
762 763 764
	if (token < Opt_last_int) {
		ret = match_int(&argstr[0], &intval);
		if (ret < 0) {
765
			pr_err("bad mount option arg (not int) at '%s'\n", c);
766 767 768 769
			return ret;
		}
		dout("got int token %d val %d\n", token, intval);
	} else if (token > Opt_last_int && token < Opt_last_string) {
770
		dout("got string token %d val %s\n", token, argstr[0].from);
771 772 773 774 775
	} else {
		dout("got token %d\n", token);
	}

	switch (token) {
A
Alex Elder 已提交
776 777 778 779 780 781
	case Opt_read_only:
		rbd_opts->read_only = true;
		break;
	case Opt_read_write:
		rbd_opts->read_only = false;
		break;
782
	default:
783 784
		/* libceph prints "bad option" msg */
		return -EINVAL;
785
	}
786

787 788 789
	return 0;
}

G
Guangliang Zhao 已提交
790 791 792 793 794 795 796
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";
797 798
	case OBJ_OP_DISCARD:
		return "discard";
G
Guangliang Zhao 已提交
799 800 801 802 803
	default:
		return "???";
	}
}

804 805
/*
 * Get a ceph client with specific addr and configuration, if one does
806 807
 * not exist create it.  Either way, ceph_opts is consumed by this
 * function.
808
 */
809
static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts)
810
{
811
	struct rbd_client *rbdc;
812

813
	mutex_lock_nested(&client_mutex, SINGLE_DEPTH_NESTING);
814
	rbdc = rbd_client_find(ceph_opts);
815
	if (rbdc)	/* using an existing client */
A
Alex Elder 已提交
816
		ceph_destroy_options(ceph_opts);
817
	else
818
		rbdc = rbd_client_create(ceph_opts);
819
	mutex_unlock(&client_mutex);
820

821
	return rbdc;
822 823 824 825
}

/*
 * Destroy ceph client
A
Alex Elder 已提交
826
 *
A
Alex Elder 已提交
827
 * Caller must hold rbd_client_list_lock.
828 829 830 831 832
 */
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
833
	dout("%s: rbdc %p\n", __func__, rbdc);
834
	spin_lock(&rbd_client_list_lock);
835
	list_del(&rbdc->node);
836
	spin_unlock(&rbd_client_list_lock);
837 838 839 840 841 842 843 844 845

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

/*
 * Drop reference to ceph client node. If it's not referenced anymore, release
 * it.
 */
846
static void rbd_put_client(struct rbd_client *rbdc)
847
{
848 849
	if (rbdc)
		kref_put(&rbdc->kref, rbd_client_release);
850 851
}

852 853 854 855 856
static bool rbd_image_format_valid(u32 image_format)
{
	return image_format == 1 || image_format == 2;
}

857 858
static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk)
{
859 860 861 862 863 864 865
	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
866 867 868 869 870 871 872 873 874 875
	/* 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;

876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893
	/*
	 * 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;
894 895
}

896
/*
897 898
 * Fill an rbd image header with information from the given format 1
 * on-disk header.
899
 */
A
Alex Elder 已提交
900
static int rbd_header_from_disk(struct rbd_device *rbd_dev,
901
				 struct rbd_image_header_ondisk *ondisk)
902
{
A
Alex Elder 已提交
903
	struct rbd_image_header *header = &rbd_dev->header;
904 905 906 907 908
	bool first_time = header->object_prefix == NULL;
	struct ceph_snap_context *snapc;
	char *object_prefix = NULL;
	char *snap_names = NULL;
	u64 *snap_sizes = NULL;
909
	u32 snap_count;
910
	size_t size;
911
	int ret = -ENOMEM;
912
	u32 i;
913

914
	/* Allocate this now to avoid having to handle failure below */
A
Alex Elder 已提交
915

916 917
	if (first_time) {
		size_t len;
918

919 920 921 922 923 924 925 926
		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
927

928
	/* Allocate the snapshot context and fill it in */
A
Alex Elder 已提交
929

930 931 932 933 934
	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);
935
	if (snap_count) {
936
		struct rbd_image_snap_ondisk *snaps;
A
Alex Elder 已提交
937 938
		u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len);

939
		/* We'll keep a copy of the snapshot names... */
940

941 942 943 944
		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
945 946
			goto out_err;

947
		/* ...as well as the array of their sizes. */
948

949
		size = snap_count * sizeof (*header->snap_sizes);
950 951
		snap_sizes = kmalloc(size, GFP_KERNEL);
		if (!snap_sizes)
A
Alex Elder 已提交
952
			goto out_err;
953

A
Alex Elder 已提交
954
		/*
955 956 957
		 * Copy the names, and fill in each snapshot's id
		 * and size.
		 *
958
		 * Note that rbd_dev_v1_header_info() guarantees the
959
		 * ondisk buffer we're working with has
A
Alex Elder 已提交
960 961 962
		 * snap_names_len bytes beyond the end of the
		 * snapshot id array, this memcpy() is safe.
		 */
963 964 965 966 967 968
		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);
		}
969
	}
A
Alex Elder 已提交
970

971
	/* We won't fail any more, fill in the header */
972

973 974 975 976 977 978 979 980 981
	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;
982
	} else {
A
Alex Elder 已提交
983 984 985
		ceph_put_snap_context(header->snapc);
		kfree(header->snap_names);
		kfree(header->snap_sizes);
986
	}
987

988
	/* The remaining fields always get updated (when we refresh) */
989

A
Alex Elder 已提交
990
	header->image_size = le64_to_cpu(ondisk->image_size);
991 992 993
	header->snapc = snapc;
	header->snap_names = snap_names;
	header->snap_sizes = snap_sizes;
994

995
	return 0;
996 997
out_2big:
	ret = -EIO;
A
Alex Elder 已提交
998
out_err:
999 1000 1001 1002
	kfree(snap_sizes);
	kfree(snap_names);
	ceph_put_snap_context(snapc);
	kfree(object_prefix);
1003

1004
	return ret;
1005 1006
}

1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021
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);
}

1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045
/*
 * 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.
 */
1046 1047 1048
static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id)
{
	struct ceph_snap_context *snapc = rbd_dev->header.snapc;
1049
	u64 *found;
1050

1051 1052
	found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps,
				sizeof (snap_id), snapid_compare_reverse);
1053

1054
	return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX;
1055 1056
}

1057 1058
static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev,
					u64 snap_id)
1059
{
1060
	u32 which;
1061
	const char *snap_name;
1062

1063 1064
	which = rbd_dev_snap_index(rbd_dev, snap_id);
	if (which == BAD_SNAP_INDEX)
1065
		return ERR_PTR(-ENOENT);
1066

1067 1068
	snap_name = _rbd_dev_v1_snap_name(rbd_dev, which);
	return snap_name ? snap_name : ERR_PTR(-ENOMEM);
1069 1070 1071 1072
}

static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id)
{
1073 1074 1075
	if (snap_id == CEPH_NOSNAP)
		return RBD_SNAP_HEAD_NAME;

1076 1077 1078
	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);
1079

1080
	return rbd_dev_v2_snap_name(rbd_dev, snap_id);
1081 1082
}

1083 1084
static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
				u64 *snap_size)
1085
{
1086 1087 1088 1089 1090
	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;
1091

1092 1093 1094
		which = rbd_dev_snap_index(rbd_dev, snap_id);
		if (which == BAD_SNAP_INDEX)
			return -ENOENT;
1095

1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107
		*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;
1108 1109
}

1110 1111
static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
			u64 *snap_features)
1112
{
1113 1114 1115 1116 1117
	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 */
1118
	} else {
1119 1120
		u64 features = 0;
		int ret;
1121

1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132
		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)
{
1133
	u64 snap_id = rbd_dev->spec->snap_id;
1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147
	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;

1148
	return 0;
1149 1150
}

A
Alex Elder 已提交
1151 1152 1153 1154
static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev)
{
	rbd_dev->mapping.size = 0;
	rbd_dev->mapping.features = 0;
1155 1156
}

1157 1158 1159 1160 1161 1162 1163
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
1164
static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset)
1165
{
A
Alex Elder 已提交
1166 1167 1168
	char *name;
	u64 segment;
	int ret;
1169
	char *name_format;
1170

1171
	name = kmem_cache_alloc(rbd_segment_name_cache, GFP_NOIO);
A
Alex Elder 已提交
1172 1173 1174
	if (!name)
		return NULL;
	segment = offset >> rbd_dev->header.obj_order;
1175 1176 1177
	name_format = "%s.%012llx";
	if (rbd_dev->image_format == 2)
		name_format = "%s.%016llx";
1178
	ret = snprintf(name, CEPH_MAX_OID_NAME_LEN + 1, name_format,
A
Alex Elder 已提交
1179
			rbd_dev->header.object_prefix, segment);
1180
	if (ret < 0 || ret > CEPH_MAX_OID_NAME_LEN) {
A
Alex Elder 已提交
1181 1182
		pr_err("error formatting segment name for #%llu (%d)\n",
			segment, ret);
1183
		rbd_segment_name_free(name);
A
Alex Elder 已提交
1184 1185
		name = NULL;
	}
1186

A
Alex Elder 已提交
1187 1188
	return name;
}
1189

A
Alex Elder 已提交
1190 1191 1192
static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset)
{
	u64 segment_size = (u64) 1 << rbd_dev->header.obj_order;
1193

A
Alex Elder 已提交
1194 1195 1196 1197 1198 1199 1200 1201 1202 1203
	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
1204
	rbd_assert(length <= U64_MAX - offset);
A
Alex Elder 已提交
1205 1206 1207 1208
	if (offset + length > segment_size)
		length = segment_size - offset;

	return length;
1209 1210
}

1211 1212 1213 1214 1215 1216 1217 1218
/*
 * returns the size of an object in the image
 */
static u64 rbd_obj_bytes(struct rbd_image_header *header)
{
	return 1 << header->obj_order;
}

1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238
/*
 * 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)
{
1239 1240
	struct bio_vec bv;
	struct bvec_iter iter;
1241 1242 1243 1244 1245
	unsigned long flags;
	void *buf;
	int pos = 0;

	while (chain) {
1246 1247
		bio_for_each_segment(bv, chain, iter) {
			if (pos + bv.bv_len > start_ofs) {
1248
				int remainder = max(start_ofs - pos, 0);
1249
				buf = bvec_kmap_irq(&bv, &flags);
1250
				memset(buf + remainder, 0,
1251 1252
				       bv.bv_len - remainder);
				flush_dcache_page(bv.bv_page);
1253
				bvec_kunmap_irq(buf, &flags);
1254
			}
1255
			pos += bv.bv_len;
1256 1257 1258 1259 1260 1261
		}

		chain = chain->bi_next;
	}
}

A
Alex Elder 已提交
1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279
/*
 * 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;

1280 1281
		page_offset = offset & ~PAGE_MASK;
		length = min_t(size_t, PAGE_SIZE - page_offset, end - offset);
A
Alex Elder 已提交
1282 1283 1284
		local_irq_save(flags);
		kaddr = kmap_atomic(*page);
		memset(kaddr + page_offset, 0, length);
1285
		flush_dcache_page(*page);
A
Alex Elder 已提交
1286 1287 1288 1289 1290 1291 1292 1293
		kunmap_atomic(kaddr);
		local_irq_restore(flags);

		offset += length;
		page++;
	}
}

1294
/*
A
Alex Elder 已提交
1295 1296
 * Clone a portion of a bio, starting at the given byte offset
 * and continuing for the number of bytes indicated.
1297
 */
A
Alex Elder 已提交
1298 1299 1300 1301
static struct bio *bio_clone_range(struct bio *bio_src,
					unsigned int offset,
					unsigned int len,
					gfp_t gfpmask)
1302
{
A
Alex Elder 已提交
1303 1304
	struct bio *bio;

K
Kent Overstreet 已提交
1305
	bio = bio_clone(bio_src, gfpmask);
A
Alex Elder 已提交
1306 1307
	if (!bio)
		return NULL;	/* ENOMEM */
1308

K
Kent Overstreet 已提交
1309
	bio_advance(bio, offset);
1310
	bio->bi_iter.bi_size = len;
A
Alex Elder 已提交
1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340

	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 */

1341
	if (!bi || off >= bi->bi_iter.bi_size || !len)
A
Alex Elder 已提交
1342
		return NULL;		/* Nothing to clone */
1343

A
Alex Elder 已提交
1344 1345 1346 1347 1348
	end = &chain;
	while (len) {
		unsigned int bi_size;
		struct bio *bio;

1349 1350
		if (!bi) {
			rbd_warn(NULL, "bio_chain exhausted with %u left", len);
A
Alex Elder 已提交
1351
			goto out_err;	/* EINVAL; ran out of bio's */
1352
		}
1353
		bi_size = min_t(unsigned int, bi->bi_iter.bi_size - off, len);
A
Alex Elder 已提交
1354 1355 1356 1357 1358 1359
		bio = bio_clone_range(bi, off, bi_size, gfpmask);
		if (!bio)
			goto out_err;	/* ENOMEM */

		*end = bio;
		end = &bio->bi_next;
1360

A
Alex Elder 已提交
1361
		off += bi_size;
1362
		if (off == bi->bi_iter.bi_size) {
A
Alex Elder 已提交
1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373
			bi = bi->bi_next;
			off = 0;
		}
		len -= bi_size;
	}
	*bio_src = bi;
	*offset = off;

	return chain;
out_err:
	bio_chain_put(chain);
1374 1375 1376 1377

	return NULL;
}

1378 1379 1380 1381 1382
/*
 * 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
1383
static void obj_request_img_data_set(struct rbd_obj_request *obj_request)
1384
{
A
Alex Elder 已提交
1385
	if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) {
1386 1387
		struct rbd_device *rbd_dev;

A
Alex Elder 已提交
1388
		rbd_dev = obj_request->img_request->rbd_dev;
1389
		rbd_warn(rbd_dev, "obj_request %p already marked img_data",
1390 1391 1392 1393
			obj_request);
	}
}

A
Alex Elder 已提交
1394
static bool obj_request_img_data_test(struct rbd_obj_request *obj_request)
1395 1396
{
	smp_mb();
A
Alex Elder 已提交
1397
	return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0;
1398 1399
}

A
Alex Elder 已提交
1400
static void obj_request_done_set(struct rbd_obj_request *obj_request)
1401
{
A
Alex Elder 已提交
1402 1403
	if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) {
		struct rbd_device *rbd_dev = NULL;
1404

A
Alex Elder 已提交
1405 1406
		if (obj_request_img_data_test(obj_request))
			rbd_dev = obj_request->img_request->rbd_dev;
1407
		rbd_warn(rbd_dev, "obj_request %p already marked done",
1408 1409 1410 1411
			obj_request);
	}
}

A
Alex Elder 已提交
1412
static bool obj_request_done_test(struct rbd_obj_request *obj_request)
1413 1414
{
	smp_mb();
A
Alex Elder 已提交
1415
	return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0;
1416 1417
}

1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448
/*
 * 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;
}

1449 1450 1451 1452 1453 1454 1455 1456
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
1457 1458
static void rbd_obj_request_get(struct rbd_obj_request *obj_request)
{
A
Alex Elder 已提交
1459 1460
	dout("%s: obj %p (was %d)\n", __func__, obj_request,
		atomic_read(&obj_request->kref.refcount));
A
Alex Elder 已提交
1461 1462 1463 1464 1465 1466 1467
	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
1468 1469
	dout("%s: obj %p (was %d)\n", __func__, obj_request,
		atomic_read(&obj_request->kref.refcount));
A
Alex Elder 已提交
1470 1471 1472
	kref_put(&obj_request->kref, rbd_obj_request_destroy);
}

1473 1474 1475 1476 1477 1478 1479
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);
}

1480 1481
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
1482 1483 1484 1485
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
1486 1487
	dout("%s: img %p (was %d)\n", __func__, img_request,
		atomic_read(&img_request->kref.refcount));
1488 1489 1490 1491
	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
1492 1493 1494 1495 1496
}

static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request,
					struct rbd_obj_request *obj_request)
{
1497 1498
	rbd_assert(obj_request->img_request == NULL);

1499
	/* Image request now owns object's original reference */
A
Alex Elder 已提交
1500
	obj_request->img_request = img_request;
1501
	obj_request->which = img_request->obj_request_count;
1502 1503
	rbd_assert(!obj_request_img_data_test(obj_request));
	obj_request_img_data_set(obj_request);
A
Alex Elder 已提交
1504
	rbd_assert(obj_request->which != BAD_WHICH);
1505 1506
	img_request->obj_request_count++;
	list_add_tail(&obj_request->links, &img_request->obj_requests);
A
Alex Elder 已提交
1507 1508
	dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
		obj_request->which);
A
Alex Elder 已提交
1509 1510 1511 1512 1513 1514
}

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);
1515

A
Alex Elder 已提交
1516 1517
	dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request,
		obj_request->which);
A
Alex Elder 已提交
1518
	list_del(&obj_request->links);
1519 1520 1521 1522
	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;
1523
	rbd_assert(obj_request_img_data_test(obj_request));
A
Alex Elder 已提交
1524 1525
	rbd_assert(obj_request->img_request == img_request);
	obj_request->img_request = NULL;
1526
	obj_request->callback = NULL;
A
Alex Elder 已提交
1527 1528 1529 1530 1531 1532
	rbd_obj_request_put(obj_request);
}

static bool obj_request_type_valid(enum obj_request_type type)
{
	switch (type) {
1533
	case OBJ_REQUEST_NODATA:
A
Alex Elder 已提交
1534
	case OBJ_REQUEST_BIO:
1535
	case OBJ_REQUEST_PAGES:
A
Alex Elder 已提交
1536 1537 1538 1539 1540 1541 1542 1543 1544
		return true;
	default:
		return false;
	}
}

static int rbd_obj_request_submit(struct ceph_osd_client *osdc,
				struct rbd_obj_request *obj_request)
{
1545
	dout("%s %p\n", __func__, obj_request);
A
Alex Elder 已提交
1546 1547 1548
	return ceph_osdc_start_request(osdc, obj_request->osd_req, false);
}

1549 1550 1551 1552 1553 1554 1555 1556 1557
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.
1558 1559
 *
 * @timeout: in jiffies, 0 means "wait forever"
1560
 */
1561 1562
static int __rbd_obj_request_wait(struct rbd_obj_request *obj_request,
				  unsigned long timeout)
1563
{
1564
	long ret;
1565 1566

	dout("%s %p\n", __func__, obj_request);
1567 1568 1569 1570 1571 1572
	ret = wait_for_completion_interruptible_timeout(
					&obj_request->completion,
					ceph_timeout_jiffies(timeout));
	if (ret <= 0) {
		if (ret == 0)
			ret = -ETIMEDOUT;
1573
		rbd_obj_request_end(obj_request);
1574 1575
	} else {
		ret = 0;
1576 1577
	}

1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590
	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);
}

static int rbd_obj_request_wait_timeout(struct rbd_obj_request *obj_request,
					unsigned long timeout)
{
	return __rbd_obj_request_wait(obj_request, timeout);
1591 1592
}

A
Alex Elder 已提交
1593 1594
static void rbd_img_request_complete(struct rbd_img_request *img_request)
{
1595

A
Alex Elder 已提交
1596
	dout("%s: img %p\n", __func__, img_request);
1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612

	/*
	 * 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
1613 1614 1615 1616 1617 1618
	if (img_request->callback)
		img_request->callback(img_request);
	else
		rbd_img_request_put(img_request);
}

A
Alex Elder 已提交
1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635
/*
 * 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;
}

1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650
/*
 * 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;
}

1651 1652 1653 1654 1655 1656
static void img_request_child_set(struct rbd_img_request *img_request)
{
	set_bit(IMG_REQ_CHILD, &img_request->flags);
	smp_mb();
}

1657 1658 1659 1660 1661 1662
static void img_request_child_clear(struct rbd_img_request *img_request)
{
	clear_bit(IMG_REQ_CHILD, &img_request->flags);
	smp_mb();
}

1663 1664 1665 1666 1667 1668
static bool img_request_child_test(struct rbd_img_request *img_request)
{
	smp_mb();
	return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0;
}

1669 1670 1671 1672 1673 1674
static void img_request_layered_set(struct rbd_img_request *img_request)
{
	set_bit(IMG_REQ_LAYERED, &img_request->flags);
	smp_mb();
}

1675 1676 1677 1678 1679 1680
static void img_request_layered_clear(struct rbd_img_request *img_request)
{
	clear_bit(IMG_REQ_LAYERED, &img_request->flags);
	smp_mb();
}

1681 1682 1683 1684 1685 1686
static bool img_request_layered_test(struct rbd_img_request *img_request)
{
	smp_mb();
	return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0;
}

1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697
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;
}

1698 1699 1700
static void
rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request)
{
A
Alex Elder 已提交
1701 1702 1703
	u64 xferred = obj_request->xferred;
	u64 length = obj_request->length;

1704 1705
	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
1706
		xferred, length);
1707
	/*
1708 1709 1710 1711 1712 1713
	 * 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.
1714
	 */
A
Alex Elder 已提交
1715
	rbd_assert(obj_request->type != OBJ_REQUEST_NODATA);
1716
	if (obj_request->result == -ENOENT) {
A
Alex Elder 已提交
1717 1718 1719 1720
		if (obj_request->type == OBJ_REQUEST_BIO)
			zero_bio_chain(obj_request->bio_list, 0);
		else
			zero_pages(obj_request->pages, 0, length);
1721
		obj_request->result = 0;
A
Alex Elder 已提交
1722 1723 1724 1725 1726
	} 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);
1727
	}
1728
	obj_request->xferred = length;
1729 1730 1731
	obj_request_done_set(obj_request);
}

A
Alex Elder 已提交
1732 1733
static void rbd_obj_request_complete(struct rbd_obj_request *obj_request)
{
A
Alex Elder 已提交
1734 1735
	dout("%s: obj %p cb %p\n", __func__, obj_request,
		obj_request->callback);
A
Alex Elder 已提交
1736 1737
	if (obj_request->callback)
		obj_request->callback(obj_request);
1738 1739
	else
		complete_all(&obj_request->completion);
A
Alex Elder 已提交
1740 1741
}

1742
static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request)
1743 1744 1745 1746 1747
{
	dout("%s: obj %p\n", __func__, obj_request);
	obj_request_done_set(obj_request);
}

1748
static void rbd_osd_read_callback(struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1749
{
A
Alex Elder 已提交
1750
	struct rbd_img_request *img_request = NULL;
A
Alex Elder 已提交
1751
	struct rbd_device *rbd_dev = NULL;
A
Alex Elder 已提交
1752 1753 1754 1755 1756
	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
1757
		rbd_dev = img_request->rbd_dev;
A
Alex Elder 已提交
1758
	}
A
Alex Elder 已提交
1759 1760 1761 1762

	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
1763 1764
	if (layered && obj_request->result == -ENOENT &&
			obj_request->img_offset < rbd_dev->parent_overlap)
A
Alex Elder 已提交
1765 1766
		rbd_img_parent_read(obj_request);
	else if (img_request)
1767 1768 1769
		rbd_img_obj_request_read_callback(obj_request);
	else
		obj_request_done_set(obj_request);
A
Alex Elder 已提交
1770 1771
}

1772
static void rbd_osd_write_callback(struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1773
{
1774 1775 1776
	dout("%s: obj %p result %d %llu\n", __func__, obj_request,
		obj_request->result, obj_request->length);
	/*
A
Alex Elder 已提交
1777 1778
	 * There is no such thing as a successful short write.  Set
	 * it to our originally-requested length.
1779 1780
	 */
	obj_request->xferred = obj_request->length;
1781
	obj_request_done_set(obj_request);
A
Alex Elder 已提交
1782 1783
}

1784 1785 1786 1787 1788 1789 1790 1791 1792
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;
1793 1794 1795
	/* discarding a non-existent object is not a problem */
	if (obj_request->result == -ENOENT)
		obj_request->result = 0;
1796 1797 1798
	obj_request_done_set(obj_request);
}

A
Alex Elder 已提交
1799 1800 1801 1802
/*
 * 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.
 */
1803
static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1804
{
A
Alex Elder 已提交
1805
	dout("%s: obj %p\n", __func__, obj_request);
A
Alex Elder 已提交
1806 1807 1808
	obj_request_done_set(obj_request);
}

A
Alex Elder 已提交
1809 1810 1811 1812 1813 1814
static void rbd_osd_req_callback(struct ceph_osd_request *osd_req,
				struct ceph_msg *msg)
{
	struct rbd_obj_request *obj_request = osd_req->r_priv;
	u16 opcode;

A
Alex Elder 已提交
1815
	dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg);
A
Alex Elder 已提交
1816
	rbd_assert(osd_req == obj_request->osd_req);
A
Alex Elder 已提交
1817 1818 1819 1820 1821 1822
	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
1823

1824 1825
	if (osd_req->r_result < 0)
		obj_request->result = osd_req->r_result;
A
Alex Elder 已提交
1826

1827
	rbd_assert(osd_req->r_num_ops <= CEPH_OSD_MAX_OP);
A
Alex Elder 已提交
1828

1829 1830
	/*
	 * We support a 64-bit length, but ultimately it has to be
C
Christoph Hellwig 已提交
1831 1832
	 * passed to the block layer, which just supports a 32-bit
	 * length field.
1833
	 */
1834
	obj_request->xferred = osd_req->r_reply_op_len[0];
A
Alex Elder 已提交
1835
	rbd_assert(obj_request->xferred < (u64)UINT_MAX);
1836

1837
	opcode = osd_req->r_ops[0].op;
A
Alex Elder 已提交
1838 1839
	switch (opcode) {
	case CEPH_OSD_OP_READ:
1840
		rbd_osd_read_callback(obj_request);
A
Alex Elder 已提交
1841
		break;
1842 1843 1844
	case CEPH_OSD_OP_SETALLOCHINT:
		rbd_assert(osd_req->r_ops[1].op == CEPH_OSD_OP_WRITE);
		/* fall through */
A
Alex Elder 已提交
1845
	case CEPH_OSD_OP_WRITE:
1846
		rbd_osd_write_callback(obj_request);
A
Alex Elder 已提交
1847
		break;
A
Alex Elder 已提交
1848
	case CEPH_OSD_OP_STAT:
1849
		rbd_osd_stat_callback(obj_request);
A
Alex Elder 已提交
1850
		break;
1851 1852 1853 1854 1855
	case CEPH_OSD_OP_DELETE:
	case CEPH_OSD_OP_TRUNCATE:
	case CEPH_OSD_OP_ZERO:
		rbd_osd_discard_callback(obj_request);
		break;
1856
	case CEPH_OSD_OP_CALL:
A
Alex Elder 已提交
1857
	case CEPH_OSD_OP_NOTIFY_ACK:
1858
	case CEPH_OSD_OP_WATCH:
1859
		rbd_osd_trivial_callback(obj_request);
1860
		break;
A
Alex Elder 已提交
1861
	default:
1862
		rbd_warn(NULL, "%s: unsupported op %hu",
A
Alex Elder 已提交
1863 1864 1865 1866
			obj_request->object_name, (unsigned short) opcode);
		break;
	}

1867
	if (obj_request_done_test(obj_request))
A
Alex Elder 已提交
1868 1869 1870
		rbd_obj_request_complete(obj_request);
}

1871
static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1872 1873
{
	struct rbd_img_request *img_request = obj_request->img_request;
1874
	struct ceph_osd_request *osd_req = obj_request->osd_req;
1875
	u64 snap_id;
A
Alex Elder 已提交
1876

1877
	rbd_assert(osd_req != NULL);
A
Alex Elder 已提交
1878

1879
	snap_id = img_request ? img_request->snap_id : CEPH_NOSNAP;
1880
	ceph_osdc_build_request(osd_req, obj_request->offset,
1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895
			NULL, snap_id, NULL);
}

static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request)
{
	struct rbd_img_request *img_request = obj_request->img_request;
	struct ceph_osd_request *osd_req = obj_request->osd_req;
	struct ceph_snap_context *snapc;
	struct timespec mtime = CURRENT_TIME;

	rbd_assert(osd_req != NULL);

	snapc = img_request ? img_request->snapc : NULL;
	ceph_osdc_build_request(osd_req, obj_request->offset,
			snapc, CEPH_NOSNAP, &mtime);
A
Alex Elder 已提交
1896 1897
}

1898 1899 1900 1901 1902 1903
/*
 * 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
1904 1905
static struct ceph_osd_request *rbd_osd_req_create(
					struct rbd_device *rbd_dev,
G
Guangliang Zhao 已提交
1906
					enum obj_operation_type op_type,
1907
					unsigned int num_ops,
A
Alex Elder 已提交
1908
					struct rbd_obj_request *obj_request)
A
Alex Elder 已提交
1909 1910 1911 1912 1913
{
	struct ceph_snap_context *snapc = NULL;
	struct ceph_osd_client *osdc;
	struct ceph_osd_request *osd_req;

1914 1915
	if (obj_request_img_data_test(obj_request) &&
		(op_type == OBJ_OP_DISCARD || op_type == OBJ_OP_WRITE)) {
1916
		struct rbd_img_request *img_request = obj_request->img_request;
1917 1918 1919 1920 1921
		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
1922
		snapc = img_request->snapc;
A
Alex Elder 已提交
1923 1924
	}

G
Guangliang Zhao 已提交
1925
	rbd_assert(num_ops == 1 || ((op_type == OBJ_OP_WRITE) && num_ops == 2));
1926 1927

	/* Allocate and initialize the request, for the num_ops ops */
A
Alex Elder 已提交
1928 1929

	osdc = &rbd_dev->rbd_client->client->osdc;
1930 1931
	osd_req = ceph_osdc_alloc_request(osdc, snapc, num_ops, false,
					  GFP_ATOMIC);
A
Alex Elder 已提交
1932 1933 1934
	if (!osd_req)
		return NULL;	/* ENOMEM */

1935
	if (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD)
A
Alex Elder 已提交
1936
		osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK;
A
Alex Elder 已提交
1937
	else
A
Alex Elder 已提交
1938 1939 1940 1941 1942
		osd_req->r_flags = CEPH_OSD_FLAG_READ;

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

1943 1944
	osd_req->r_base_oloc.pool = ceph_file_layout_pg_pool(rbd_dev->layout);
	ceph_oid_set_name(&osd_req->r_base_oid, obj_request->object_name);
A
Alex Elder 已提交
1945 1946 1947 1948

	return osd_req;
}

1949
/*
1950 1951 1952 1953
 * 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.
1954 1955 1956 1957 1958 1959 1960 1961 1962
 */
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;
1963
	int num_osd_ops = 3;
1964 1965 1966 1967

	rbd_assert(obj_request_img_data_test(obj_request));
	img_request = obj_request->img_request;
	rbd_assert(img_request);
1968 1969
	rbd_assert(img_request_write_test(img_request) ||
			img_request_discard_test(img_request));
1970

1971 1972 1973 1974
	if (img_request_discard_test(img_request))
		num_osd_ops = 2;

	/* Allocate and initialize the request, for all the ops */
1975 1976 1977 1978

	snapc = img_request->snapc;
	rbd_dev = img_request->rbd_dev;
	osdc = &rbd_dev->rbd_client->client->osdc;
1979 1980
	osd_req = ceph_osdc_alloc_request(osdc, snapc, num_osd_ops,
						false, GFP_ATOMIC);
1981 1982 1983 1984 1985 1986 1987
	if (!osd_req)
		return NULL;	/* ENOMEM */

	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;

1988 1989
	osd_req->r_base_oloc.pool = ceph_file_layout_pg_pool(rbd_dev->layout);
	ceph_oid_set_name(&osd_req->r_base_oid, obj_request->object_name);
1990 1991 1992 1993 1994

	return osd_req;
}


A
Alex Elder 已提交
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
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;
2013 2014
	name = kmalloc(size, GFP_KERNEL);
	if (!name)
A
Alex Elder 已提交
2015 2016
		return NULL;

2017
	obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_KERNEL);
2018 2019 2020 2021 2022
	if (!obj_request) {
		kfree(name);
		return NULL;
	}

A
Alex Elder 已提交
2023 2024 2025
	obj_request->object_name = memcpy(name, object_name, size);
	obj_request->offset = offset;
	obj_request->length = length;
2026
	obj_request->flags = 0;
A
Alex Elder 已提交
2027 2028 2029
	obj_request->which = BAD_WHICH;
	obj_request->type = type;
	INIT_LIST_HEAD(&obj_request->links);
2030
	init_completion(&obj_request->completion);
A
Alex Elder 已提交
2031 2032
	kref_init(&obj_request->kref);

A
Alex Elder 已提交
2033 2034 2035
	dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name,
		offset, length, (int)type, obj_request);

A
Alex Elder 已提交
2036 2037 2038 2039 2040 2041 2042 2043 2044
	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
2045 2046
	dout("%s: obj %p\n", __func__, obj_request);

A
Alex Elder 已提交
2047 2048 2049 2050 2051 2052 2053 2054
	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) {
2055 2056
	case OBJ_REQUEST_NODATA:
		break;		/* Nothing to do */
A
Alex Elder 已提交
2057 2058 2059 2060
	case OBJ_REQUEST_BIO:
		if (obj_request->bio_list)
			bio_chain_put(obj_request->bio_list);
		break;
2061 2062 2063 2064 2065
	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
2066 2067
	}

2068
	kfree(obj_request->object_name);
2069 2070
	obj_request->object_name = NULL;
	kmem_cache_free(rbd_obj_request_cache, obj_request);
A
Alex Elder 已提交
2071 2072
}

A
Alex Elder 已提交
2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083
/* 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;
}

2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105
/*
 * 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
2106
		rbd_warn(rbd_dev, "parent reference underflow");
2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118
}

/*
 * 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)
{
2119
	int counter = 0;
2120 2121 2122 2123

	if (!rbd_dev->parent_spec)
		return false;

2124 2125 2126 2127
	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);
2128 2129

	if (counter < 0)
2130
		rbd_warn(rbd_dev, "parent reference overflow");
2131

2132
	return counter > 0;
2133 2134
}

A
Alex Elder 已提交
2135 2136 2137 2138 2139
/*
 * 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
2140 2141
static struct rbd_img_request *rbd_img_request_create(
					struct rbd_device *rbd_dev,
A
Alex Elder 已提交
2142
					u64 offset, u64 length,
G
Guangliang Zhao 已提交
2143
					enum obj_operation_type op_type,
2144
					struct ceph_snap_context *snapc)
A
Alex Elder 已提交
2145 2146 2147
{
	struct rbd_img_request *img_request;

2148
	img_request = kmem_cache_alloc(rbd_img_request_cache, GFP_NOIO);
A
Alex Elder 已提交
2149 2150 2151 2152 2153 2154 2155
	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
2156
	img_request->flags = 0;
2157 2158 2159 2160
	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
2161
		img_request_write_set(img_request);
2162
		img_request->snapc = snapc;
A
Alex Elder 已提交
2163
	} else {
A
Alex Elder 已提交
2164
		img_request->snap_id = rbd_dev->spec->snap_id;
A
Alex Elder 已提交
2165
	}
2166
	if (rbd_dev_parent_get(rbd_dev))
2167
		img_request_layered_set(img_request);
A
Alex Elder 已提交
2168 2169 2170
	spin_lock_init(&img_request->completion_lock);
	img_request->next_completion = 0;
	img_request->callback = NULL;
2171
	img_request->result = 0;
A
Alex Elder 已提交
2172 2173 2174 2175
	img_request->obj_request_count = 0;
	INIT_LIST_HEAD(&img_request->obj_requests);
	kref_init(&img_request->kref);

A
Alex Elder 已提交
2176
	dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev,
G
Guangliang Zhao 已提交
2177
		obj_op_name(op_type), offset, length, img_request);
A
Alex Elder 已提交
2178

A
Alex Elder 已提交
2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189
	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
2190 2191
	dout("%s: img %p\n", __func__, img_request);

A
Alex Elder 已提交
2192 2193
	for_each_obj_request_safe(img_request, obj_request, next_obj_request)
		rbd_img_obj_request_del(img_request, obj_request);
2194
	rbd_assert(img_request->obj_request_count == 0);
A
Alex Elder 已提交
2195

2196 2197 2198 2199 2200
	if (img_request_layered_test(img_request)) {
		img_request_layered_clear(img_request);
		rbd_dev_parent_put(img_request->rbd_dev);
	}

2201 2202
	if (img_request_write_test(img_request) ||
		img_request_discard_test(img_request))
2203
		ceph_put_snap_context(img_request->snapc);
A
Alex Elder 已提交
2204

2205
	kmem_cache_free(rbd_img_request_cache, img_request);
A
Alex Elder 已提交
2206 2207
}

2208 2209 2210 2211 2212 2213 2214 2215 2216 2217
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;

2218
	parent_request = rbd_img_request_create(rbd_dev->parent, img_offset,
G
Guangliang Zhao 已提交
2219
						length, OBJ_OP_READ, NULL);
2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244
	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);
}

2245 2246
static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request)
{
2247
	struct rbd_img_request *img_request;
2248 2249
	unsigned int xferred;
	int result;
A
Alex Elder 已提交
2250
	bool more;
2251

2252 2253 2254
	rbd_assert(obj_request_img_data_test(obj_request));
	img_request = obj_request->img_request;

2255 2256 2257 2258 2259
	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
2260 2261
		enum obj_operation_type op_type;

2262 2263 2264 2265 2266 2267
		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;
2268

2269
		rbd_warn(rbd_dev, "%s %llx at %llx (%llx)",
G
Guangliang Zhao 已提交
2270 2271
			obj_op_name(op_type), obj_request->length,
			obj_request->img_offset, obj_request->offset);
2272
		rbd_warn(rbd_dev, "  result %d xferred %x",
2273 2274 2275
			result, xferred);
		if (!img_request->result)
			img_request->result = result;
2276 2277 2278 2279 2280
		/*
		 * Need to end I/O on the entire obj_request worth of
		 * bytes in case of error.
		 */
		xferred = obj_request->length;
2281 2282
	}

2283 2284 2285 2286 2287 2288 2289
	/* 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
2290 2291 2292 2293 2294
	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
2295 2296 2297 2298

		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
2299 2300 2301
	}

	return more;
2302 2303
}

2304 2305 2306 2307 2308 2309
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;

2310
	rbd_assert(obj_request_img_data_test(obj_request));
2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328
	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;
2329
		more = rbd_img_obj_end_request(obj_request);
2330 2331 2332 2333 2334 2335 2336
		which++;
	}

	rbd_assert(more ^ (which == img_request->obj_request_count));
	img_request->next_completion = which;
out:
	spin_unlock_irq(&img_request->completion_lock);
2337
	rbd_img_request_put(img_request);
2338 2339 2340 2341 2342

	if (!more)
		rbd_img_request_complete(img_request);
}

2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361
/*
 * 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) {
2362 2363 2364
		if (!offset && length == object_size &&
		    (!img_request_layered_test(img_request) ||
		     !obj_request_overlaps_parent(obj_request))) {
2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386
			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) {
		opcode = CEPH_OSD_OP_WRITE;
		osd_req_op_alloc_hint_init(osd_request, num_ops,
					object_size, object_size);
		num_ops++;
	} else {
		opcode = CEPH_OSD_OP_READ;
	}

2387
	if (opcode == CEPH_OSD_OP_DELETE)
2388
		osd_req_op_init(osd_request, num_ops, opcode, 0);
2389 2390 2391 2392
	else
		osd_req_op_extent_init(osd_request, num_ops, opcode,
				       offset, length, 0, 0);

2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407
	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);
}

2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418
/*
 * 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
2419 2420 2421 2422
{
	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
2423
	struct bio *bio_list = NULL;
2424
	unsigned int bio_offset = 0;
J
Jingoo Han 已提交
2425
	struct page **pages = NULL;
G
Guangliang Zhao 已提交
2426
	enum obj_operation_type op_type;
2427
	u64 img_offset;
A
Alex Elder 已提交
2428 2429
	u64 resid;

2430 2431
	dout("%s: img %p type %d data_desc %p\n", __func__, img_request,
		(int)type, data_desc);
A
Alex Elder 已提交
2432

2433
	img_offset = img_request->offset;
A
Alex Elder 已提交
2434
	resid = img_request->length;
A
Alex Elder 已提交
2435
	rbd_assert(resid > 0);
2436
	op_type = rbd_img_request_op_type(img_request);
2437 2438 2439

	if (type == OBJ_REQUEST_BIO) {
		bio_list = data_desc;
2440 2441
		rbd_assert(img_offset ==
			   bio_list->bi_iter.bi_sector << SECTOR_SHIFT);
2442
	} else if (type == OBJ_REQUEST_PAGES) {
2443 2444 2445
		pages = data_desc;
	}

A
Alex Elder 已提交
2446
	while (resid) {
2447
		struct ceph_osd_request *osd_req;
A
Alex Elder 已提交
2448 2449 2450 2451
		const char *object_name;
		u64 offset;
		u64 length;

2452
		object_name = rbd_segment_name(rbd_dev, img_offset);
A
Alex Elder 已提交
2453 2454
		if (!object_name)
			goto out_unwind;
2455 2456
		offset = rbd_segment_offset(rbd_dev, img_offset);
		length = rbd_segment_length(rbd_dev, img_offset, resid);
A
Alex Elder 已提交
2457
		obj_request = rbd_obj_request_create(object_name,
2458
						offset, length, type);
2459 2460
		/* object request has its own copy of the object name */
		rbd_segment_name_free(object_name);
A
Alex Elder 已提交
2461 2462
		if (!obj_request)
			goto out_unwind;
2463

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

2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480
		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,
								GFP_ATOMIC);
			if (!obj_request->bio_list)
2481
				goto out_unwind;
2482
		} else if (type == OBJ_REQUEST_PAGES) {
2483 2484 2485 2486 2487 2488 2489 2490 2491
			unsigned int page_count;

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

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

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

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

2505
		rbd_img_request_get(img_request);
A
Alex Elder 已提交
2506

2507
		img_offset += length;
A
Alex Elder 已提交
2508 2509 2510 2511 2512 2513 2514
		resid -= length;
	}

	return 0;

out_unwind:
	for_each_obj_request_safe(img_request, obj_request, next_obj_request)
2515
		rbd_img_obj_request_del(img_request, obj_request);
A
Alex Elder 已提交
2516 2517 2518 2519

	return -ENOMEM;
}

2520 2521 2522 2523 2524
static void
rbd_img_obj_copyup_callback(struct rbd_obj_request *obj_request)
{
	struct rbd_img_request *img_request;
	struct rbd_device *rbd_dev;
2525
	struct page **pages;
2526 2527
	u32 page_count;

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

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

2537 2538
	pages = obj_request->copyup_pages;
	rbd_assert(pages != NULL);
2539
	obj_request->copyup_pages = NULL;
2540 2541 2542 2543
	page_count = obj_request->copyup_page_count;
	rbd_assert(page_count);
	obj_request->copyup_page_count = 0;
	ceph_release_page_vector(pages, page_count);
2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558

	/*
	 * 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;

	/* Finish up with the normal image object callback */

	rbd_img_obj_callback(obj_request);
}

2559 2560 2561 2562
static void
rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request)
{
	struct rbd_obj_request *orig_request;
2563 2564 2565
	struct ceph_osd_request *osd_req;
	struct ceph_osd_client *osdc;
	struct rbd_device *rbd_dev;
2566
	struct page **pages;
2567
	enum obj_operation_type op_type;
2568
	u32 page_count;
2569
	int img_result;
2570
	u64 parent_length;
2571 2572 2573 2574 2575 2576 2577 2578

	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;
2579 2580 2581
	page_count = img_request->copyup_page_count;
	rbd_assert(page_count);
	img_request->copyup_page_count = 0;
2582 2583 2584

	orig_request = img_request->obj_request;
	rbd_assert(orig_request != NULL);
2585
	rbd_assert(obj_request_type_valid(orig_request->type));
2586
	img_result = img_request->result;
2587 2588
	parent_length = img_request->length;
	rbd_assert(parent_length == img_request->xferred);
2589
	rbd_img_request_put(img_request);
2590

2591 2592
	rbd_assert(orig_request->img_request);
	rbd_dev = orig_request->img_request->rbd_dev;
2593 2594
	rbd_assert(rbd_dev);

2595 2596 2597 2598 2599 2600 2601
	/*
	 * 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;
2602

2603 2604 2605 2606 2607 2608
		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;
	}
2609

2610
	if (img_result)
2611 2612
		goto out_err;

2613 2614
	/*
	 * The original osd request is of no use to use any more.
2615
	 * We need a new one that can hold the three ops in a copyup
2616 2617 2618
	 * request.  Allocate the new copyup osd request for the
	 * original request, and release the old one.
	 */
2619
	img_result = -ENOMEM;
2620 2621 2622
	osd_req = rbd_osd_req_create_copyup(orig_request);
	if (!osd_req)
		goto out_err;
2623
	rbd_osd_req_destroy(orig_request->osd_req);
2624 2625
	orig_request->osd_req = osd_req;
	orig_request->copyup_pages = pages;
2626
	orig_request->copyup_page_count = page_count;
2627

2628
	/* Initialize the copyup op */
2629

2630
	osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup");
2631
	osd_req_op_cls_request_data_pages(osd_req, 0, pages, parent_length, 0,
2632
						false, false);
2633

2634
	/* Add the other op(s) */
2635

2636 2637
	op_type = rbd_img_request_op_type(orig_request->img_request);
	rbd_img_obj_request_fill(orig_request, osd_req, op_type, 1);
2638 2639 2640 2641 2642

	/* All set, send it off. */

	orig_request->callback = rbd_img_obj_copyup_callback;
	osdc = &rbd_dev->rbd_client->client->osdc;
2643 2644
	img_result = rbd_obj_request_submit(osdc, orig_request);
	if (!img_result)
2645 2646 2647 2648
		return;
out_err:
	/* Record the error code and complete the request */

2649
	orig_request->result = img_result;
2650 2651 2652
	orig_request->xferred = 0;
	obj_request_done_set(orig_request);
	rbd_obj_request_complete(orig_request);
2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680
}

/*
 * 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));
2681
	rbd_assert(obj_request_type_valid(obj_request->type));
2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694

	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
2695 2696 2697 2698 2699 2700 2701 2702 2703 2704
	/*
	 * 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;
	}

2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717
	/*
	 * 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;
2718 2719
	parent_request = rbd_parent_request_create(obj_request,
						img_offset, length);
2720 2721 2722 2723 2724 2725 2726
	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;
2727
	parent_request->copyup_page_count = page_count;
2728 2729 2730 2731 2732 2733 2734

	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;
2735
	parent_request->copyup_page_count = 0;
2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749
	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;
}

2750 2751 2752
static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request)
{
	struct rbd_obj_request *orig_request;
2753
	struct rbd_device *rbd_dev;
2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764
	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;
2765
	rbd_obj_request_put(orig_request);
2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776
	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);

2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790
	/*
	 * 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;
	}
2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803

	/*
	 * 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;
2804
		goto out;
2805 2806 2807 2808 2809 2810
	}

	/*
	 * Resubmit the original request now that we have recorded
	 * whether the target object exists.
	 */
2811
	orig_request->result = rbd_img_obj_request_submit(orig_request);
2812
out:
2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853
	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
2854
	stat_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1,
2855
						   stat_request);
2856 2857 2858 2859
	if (!stat_request->osd_req)
		goto out;
	stat_request->callback = rbd_img_obj_exists_callback;

2860
	osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT, 0);
2861 2862
	osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0,
					false, false);
2863
	rbd_osd_req_format_read(stat_request);
2864 2865 2866 2867 2868 2869 2870 2871 2872 2873

	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;
}

2874
static bool img_obj_request_simple(struct rbd_obj_request *obj_request)
2875 2876
{
	struct rbd_img_request *img_request;
A
Alex Elder 已提交
2877
	struct rbd_device *rbd_dev;
2878 2879 2880 2881 2882

	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
2883
	rbd_dev = img_request->rbd_dev;
2884

2885
	/* Reads */
2886 2887
	if (!img_request_write_test(img_request) &&
	    !img_request_discard_test(img_request))
2888 2889 2890 2891 2892 2893
		return true;

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

2894
	/*
2895 2896
	 * Layered writes outside of the parent overlap range don't
	 * share any data with the parent.
2897
	 */
2898 2899
	if (!obj_request_overlaps_parent(obj_request))
		return true;
2900

2901 2902 2903 2904 2905 2906 2907 2908
	/*
	 * 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;

2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922
	/*
	 * 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)) {
2923 2924 2925 2926 2927 2928 2929 2930 2931 2932
		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);
	}

	/*
2933 2934 2935 2936
	 * 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.
2937
	 */
2938
	if (obj_request_known_test(obj_request))
2939 2940 2941
		return rbd_img_obj_parent_read_full(obj_request);

	/* We don't know whether the target exists.  Go find out. */
2942 2943 2944 2945

	return rbd_img_obj_exists_submit(obj_request);
}

A
Alex Elder 已提交
2946 2947 2948
static int rbd_img_request_submit(struct rbd_img_request *img_request)
{
	struct rbd_obj_request *obj_request;
2949
	struct rbd_obj_request *next_obj_request;
A
Alex Elder 已提交
2950

A
Alex Elder 已提交
2951
	dout("%s: img %p\n", __func__, img_request);
2952
	for_each_obj_request_safe(img_request, obj_request, next_obj_request) {
A
Alex Elder 已提交
2953 2954
		int ret;

2955
		ret = rbd_img_obj_request_submit(obj_request);
A
Alex Elder 已提交
2956 2957 2958 2959 2960 2961
		if (ret)
			return ret;
	}

	return 0;
}
A
Alex Elder 已提交
2962 2963 2964 2965

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
2966 2967
	struct rbd_device *rbd_dev;
	u64 obj_end;
2968 2969
	u64 img_xferred;
	int img_result;
A
Alex Elder 已提交
2970 2971 2972

	rbd_assert(img_request_child_test(img_request));

2973 2974
	/* First get what we need from the image request and release it */

A
Alex Elder 已提交
2975
	obj_request = img_request->obj_request;
2976 2977 2978 2979 2980 2981 2982 2983 2984
	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
2985 2986
	rbd_assert(obj_request);
	rbd_assert(obj_request->img_request);
2987 2988 2989 2990 2991 2992 2993 2994 2995
	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
2996

2997
	obj_request->result = img_result;
A
Alex Elder 已提交
2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015
	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
3016

3017
		obj_request->xferred = min(img_xferred, xferred);
A
Alex Elder 已提交
3018
	} else {
3019
		obj_request->xferred = img_xferred;
A
Alex Elder 已提交
3020 3021
	}
out:
A
Alex Elder 已提交
3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033
	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);
3034
	rbd_assert(obj_request_type_valid(obj_request->type));
A
Alex Elder 已提交
3035 3036

	/* rbd_read_finish(obj_request, obj_request->length); */
3037
	img_request = rbd_parent_request_create(obj_request,
A
Alex Elder 已提交
3038
						obj_request->img_offset,
3039
						obj_request->length);
A
Alex Elder 已提交
3040 3041 3042 3043
	result = -ENOMEM;
	if (!img_request)
		goto out_err;

3044 3045 3046 3047 3048 3049
	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
3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065
	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
3066

3067
static int rbd_obj_notify_ack_sync(struct rbd_device *rbd_dev, u64 notify_id)
A
Alex Elder 已提交
3068 3069
{
	struct rbd_obj_request *obj_request;
3070
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
A
Alex Elder 已提交
3071 3072 3073 3074 3075 3076 3077 3078
	int ret;

	obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
							OBJ_REQUEST_NODATA);
	if (!obj_request)
		return -ENOMEM;

	ret = -ENOMEM;
G
Guangliang Zhao 已提交
3079
	obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1,
3080
						  obj_request);
A
Alex Elder 已提交
3081 3082 3083
	if (!obj_request->osd_req)
		goto out;

3084
	osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK,
A
Alex Elder 已提交
3085
					notify_id, 0, 0);
3086
	rbd_osd_req_format_read(obj_request);
A
Alex Elder 已提交
3087

A
Alex Elder 已提交
3088
	ret = rbd_obj_request_submit(osdc, obj_request);
A
Alex Elder 已提交
3089
	if (ret)
3090 3091 3092 3093
		goto out;
	ret = rbd_obj_request_wait(obj_request);
out:
	rbd_obj_request_put(obj_request);
A
Alex Elder 已提交
3094 3095 3096 3097 3098 3099 3100

	return ret;
}

static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
{
	struct rbd_device *rbd_dev = (struct rbd_device *)data;
3101
	int ret;
A
Alex Elder 已提交
3102 3103 3104 3105

	if (!rbd_dev)
		return;

A
Alex Elder 已提交
3106
	dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__,
A
Alex Elder 已提交
3107 3108
		rbd_dev->header_name, (unsigned long long)notify_id,
		(unsigned int)opcode);
3109 3110 3111 3112 3113 3114 3115

	/*
	 * Until adequate refresh error handling is in place, there is
	 * not much we can do here, except warn.
	 *
	 * See http://tracker.ceph.com/issues/5040
	 */
3116 3117
	ret = rbd_dev_refresh(rbd_dev);
	if (ret)
3118
		rbd_warn(rbd_dev, "refresh failed: %d", ret);
A
Alex Elder 已提交
3119

3120 3121
	ret = rbd_obj_notify_ack_sync(rbd_dev, notify_id);
	if (ret)
3122
		rbd_warn(rbd_dev, "notify_ack ret %d", ret);
A
Alex Elder 已提交
3123 3124
}

3125 3126 3127 3128 3129 3130 3131 3132 3133
/*
 * Send a (un)watch request and wait for the ack.  Return a request
 * with a ref held on success or error.
 */
static struct rbd_obj_request *rbd_obj_watch_request_helper(
						struct rbd_device *rbd_dev,
						bool watch)
{
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3134
	struct ceph_options *opts = osdc->client->options;
3135 3136 3137 3138 3139 3140 3141 3142
	struct rbd_obj_request *obj_request;
	int ret;

	obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0,
					     OBJ_REQUEST_NODATA);
	if (!obj_request)
		return ERR_PTR(-ENOMEM);

G
Guangliang Zhao 已提交
3143
	obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_WRITE, 1,
3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160
						  obj_request);
	if (!obj_request->osd_req) {
		ret = -ENOMEM;
		goto out;
	}

	osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH,
			      rbd_dev->watch_event->cookie, 0, watch);
	rbd_osd_req_format_write(obj_request);

	if (watch)
		ceph_osdc_set_request_linger(osdc, obj_request->osd_req);

	ret = rbd_obj_request_submit(osdc, obj_request);
	if (ret)
		goto out;

3161
	ret = rbd_obj_request_wait_timeout(obj_request, opts->mount_timeout);
3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178
	if (ret)
		goto out;

	ret = obj_request->result;
	if (ret) {
		if (watch)
			rbd_obj_request_end(obj_request);
		goto out;
	}

	return obj_request;

out:
	rbd_obj_request_put(obj_request);
	return ERR_PTR(ret);
}

3179
/*
3180
 * Initiate a watch request, synchronously.
3181
 */
3182
static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev)
3183 3184 3185 3186 3187
{
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
	struct rbd_obj_request *obj_request;
	int ret;

3188 3189
	rbd_assert(!rbd_dev->watch_event);
	rbd_assert(!rbd_dev->watch_request);
3190

3191 3192 3193 3194 3195
	ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev,
				     &rbd_dev->watch_event);
	if (ret < 0)
		return ret;

3196 3197 3198 3199 3200
	obj_request = rbd_obj_watch_request_helper(rbd_dev, true);
	if (IS_ERR(obj_request)) {
		ceph_osdc_cancel_event(rbd_dev->watch_event);
		rbd_dev->watch_event = NULL;
		return PTR_ERR(obj_request);
3201
	}
3202

3203 3204 3205 3206
	/*
	 * A watch request is set to linger, so the underlying osd
	 * request won't go away until we unregister it.  We retain
	 * a pointer to the object request during that time (in
3207 3208 3209
	 * rbd_dev->watch_request), so we'll keep a reference to it.
	 * We'll drop that reference after we've unregistered it in
	 * rbd_dev_header_unwatch_sync().
3210
	 */
3211
	rbd_dev->watch_request = obj_request;
3212

3213 3214 3215 3216 3217 3218
	return 0;
}

/*
 * Tear down a watch request, synchronously.
 */
3219
static void rbd_dev_header_unwatch_sync(struct rbd_device *rbd_dev)
3220 3221 3222 3223 3224 3225
{
	struct rbd_obj_request *obj_request;

	rbd_assert(rbd_dev->watch_event);
	rbd_assert(rbd_dev->watch_request);

3226
	rbd_obj_request_end(rbd_dev->watch_request);
3227 3228
	rbd_obj_request_put(rbd_dev->watch_request);
	rbd_dev->watch_request = NULL;
3229

3230 3231 3232 3233 3234 3235 3236
	obj_request = rbd_obj_watch_request_helper(rbd_dev, false);
	if (!IS_ERR(obj_request))
		rbd_obj_request_put(obj_request);
	else
		rbd_warn(rbd_dev, "unable to tear down watch request (%ld)",
			 PTR_ERR(obj_request));

3237 3238
	ceph_osdc_cancel_event(rbd_dev->watch_event);
	rbd_dev->watch_event = NULL;
3239 3240
}

3241
/*
3242 3243
 * Synchronous osd object method call.  Returns the number of bytes
 * returned in the outbound buffer, or a negative error code.
3244 3245 3246 3247 3248
 */
static int rbd_obj_method_sync(struct rbd_device *rbd_dev,
			     const char *object_name,
			     const char *class_name,
			     const char *method_name,
3249
			     const void *outbound,
3250
			     size_t outbound_size,
3251
			     void *inbound,
3252
			     size_t inbound_size)
3253
{
3254
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3255 3256 3257 3258 3259 3260
	struct rbd_obj_request *obj_request;
	struct page **pages;
	u32 page_count;
	int ret;

	/*
3261 3262 3263 3264 3265
	 * 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.
3266
	 */
3267
	page_count = (u32)calc_pages_for(0, inbound_size);
3268 3269 3270 3271 3272
	pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
	if (IS_ERR(pages))
		return PTR_ERR(pages);

	ret = -ENOMEM;
3273
	obj_request = rbd_obj_request_create(object_name, 0, inbound_size,
3274 3275 3276 3277 3278 3279 3280
							OBJ_REQUEST_PAGES);
	if (!obj_request)
		goto out;

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

G
Guangliang Zhao 已提交
3281
	obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1,
3282
						  obj_request);
3283 3284 3285
	if (!obj_request->osd_req)
		goto out;

3286
	osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL,
3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299
					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);
	}
3300 3301
	osd_req_op_cls_response_data_pages(obj_request->osd_req, 0,
					obj_request->pages, inbound_size,
3302
					0, false, false);
3303
	rbd_osd_req_format_read(obj_request);
A
Alex Elder 已提交
3304

3305 3306 3307 3308 3309 3310 3311 3312 3313 3314
	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;
3315 3316 3317

	rbd_assert(obj_request->xferred < (u64)INT_MAX);
	ret = (int)obj_request->xferred;
3318
	ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred);
3319 3320 3321 3322 3323 3324 3325 3326 3327
out:
	if (obj_request)
		rbd_obj_request_put(obj_request);
	else
		ceph_release_page_vector(pages, page_count);

	return ret;
}

C
Christoph Hellwig 已提交
3328
static void rbd_queue_workfn(struct work_struct *work)
A
Alex Elder 已提交
3329
{
C
Christoph Hellwig 已提交
3330 3331
	struct request *rq = blk_mq_rq_from_pdu(work);
	struct rbd_device *rbd_dev = rq->q->queuedata;
I
Ilya Dryomov 已提交
3332
	struct rbd_img_request *img_request;
3333
	struct ceph_snap_context *snapc = NULL;
I
Ilya Dryomov 已提交
3334 3335
	u64 offset = (u64)blk_rq_pos(rq) << SECTOR_SHIFT;
	u64 length = blk_rq_bytes(rq);
G
Guangliang Zhao 已提交
3336
	enum obj_operation_type op_type;
3337
	u64 mapping_size;
A
Alex Elder 已提交
3338 3339
	int result;

C
Christoph Hellwig 已提交
3340 3341 3342 3343 3344 3345 3346
	if (rq->cmd_type != REQ_TYPE_FS) {
		dout("%s: non-fs request type %d\n", __func__,
			(int) rq->cmd_type);
		result = -EIO;
		goto err;
	}

3347 3348 3349
	if (rq->cmd_flags & REQ_DISCARD)
		op_type = OBJ_OP_DISCARD;
	else if (rq->cmd_flags & REQ_WRITE)
G
Guangliang Zhao 已提交
3350 3351 3352 3353
		op_type = OBJ_OP_WRITE;
	else
		op_type = OBJ_OP_READ;

I
Ilya Dryomov 已提交
3354
	/* Ignore/skip any zero-length requests */
A
Alex Elder 已提交
3355

I
Ilya Dryomov 已提交
3356 3357 3358 3359 3360
	if (!length) {
		dout("%s: zero-length request\n", __func__);
		result = 0;
		goto err_rq;
	}
A
Alex Elder 已提交
3361

G
Guangliang Zhao 已提交
3362
	/* Only reads are allowed to a read-only device */
I
Ilya Dryomov 已提交
3363

G
Guangliang Zhao 已提交
3364
	if (op_type != OBJ_OP_READ) {
I
Ilya Dryomov 已提交
3365 3366 3367
		if (rbd_dev->mapping.read_only) {
			result = -EROFS;
			goto err_rq;
A
Alex Elder 已提交
3368
		}
I
Ilya Dryomov 已提交
3369 3370
		rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP);
	}
A
Alex Elder 已提交
3371

I
Ilya Dryomov 已提交
3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383
	/*
	 * 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
3384

I
Ilya Dryomov 已提交
3385 3386 3387 3388 3389 3390
	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
3391

C
Christoph Hellwig 已提交
3392 3393
	blk_mq_start_request(rq);

3394 3395
	down_read(&rbd_dev->header_rwsem);
	mapping_size = rbd_dev->mapping.size;
G
Guangliang Zhao 已提交
3396
	if (op_type != OBJ_OP_READ) {
3397 3398 3399 3400 3401 3402
		snapc = rbd_dev->header.snapc;
		ceph_get_snap_context(snapc);
	}
	up_read(&rbd_dev->header_rwsem);

	if (offset + length > mapping_size) {
I
Ilya Dryomov 已提交
3403
		rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)", offset,
3404
			 length, mapping_size);
I
Ilya Dryomov 已提交
3405 3406 3407
		result = -EIO;
		goto err_rq;
	}
A
Alex Elder 已提交
3408

G
Guangliang Zhao 已提交
3409
	img_request = rbd_img_request_create(rbd_dev, offset, length, op_type,
3410
					     snapc);
I
Ilya Dryomov 已提交
3411 3412 3413 3414 3415
	if (!img_request) {
		result = -ENOMEM;
		goto err_rq;
	}
	img_request->rq = rq;
A
Alex Elder 已提交
3416

3417 3418 3419 3420 3421 3422
	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
3423 3424
	if (result)
		goto err_img_request;
A
Alex Elder 已提交
3425

I
Ilya Dryomov 已提交
3426 3427 3428
	result = rbd_img_request_submit(img_request);
	if (result)
		goto err_img_request;
A
Alex Elder 已提交
3429

I
Ilya Dryomov 已提交
3430
	return;
A
Alex Elder 已提交
3431

I
Ilya Dryomov 已提交
3432 3433 3434 3435 3436
err_img_request:
	rbd_img_request_put(img_request);
err_rq:
	if (result)
		rbd_warn(rbd_dev, "%s %llx at %llx result %d",
G
Guangliang Zhao 已提交
3437
			 obj_op_name(op_type), length, offset, result);
3438
	ceph_put_snap_context(snapc);
C
Christoph Hellwig 已提交
3439 3440
err:
	blk_mq_end_request(rq, result);
I
Ilya Dryomov 已提交
3441
}
A
Alex Elder 已提交
3442

C
Christoph Hellwig 已提交
3443 3444
static int rbd_queue_rq(struct blk_mq_hw_ctx *hctx,
		const struct blk_mq_queue_data *bd)
I
Ilya Dryomov 已提交
3445
{
C
Christoph Hellwig 已提交
3446 3447
	struct request *rq = bd->rq;
	struct work_struct *work = blk_mq_rq_to_pdu(rq);
A
Alex Elder 已提交
3448

C
Christoph Hellwig 已提交
3449 3450
	queue_work(rbd_wq, work);
	return BLK_MQ_RQ_QUEUE_OK;
A
Alex Elder 已提交
3451 3452
}

3453 3454 3455
/*
 * a queue callback. Makes sure that we don't create a bio that spans across
 * multiple osd objects. One exception would be with a single page bios,
A
Alex Elder 已提交
3456
 * which we handle later at bio_chain_clone_range()
3457 3458 3459 3460 3461
 */
static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
			  struct bio_vec *bvec)
{
	struct rbd_device *rbd_dev = q->queuedata;
A
Alex Elder 已提交
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
	sector_t sector_offset;
	sector_t sectors_per_obj;
	sector_t obj_sector_offset;
	int ret;

	/*
	 * Find how far into its rbd object the partition-relative
	 * bio start sector is to offset relative to the enclosing
	 * device.
	 */
	sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector;
	sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT);
	obj_sector_offset = sector_offset & (sectors_per_obj - 1);

	/*
	 * Compute the number of bytes from that offset to the end
	 * of the object.  Account for what's already used by the bio.
	 */
	ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT;
	if (ret > bmd->bi_size)
		ret -= bmd->bi_size;
	else
		ret = 0;

	/*
	 * Don't send back more than was asked for.  And if the bio
	 * was empty, let the whole thing through because:  "Note
	 * that a block device *must* allow a single page to be
	 * added to an empty bio."
	 */
	rbd_assert(bvec->bv_len <= PAGE_SIZE);
	if (ret > (int) bvec->bv_len || !bmd->bi_size)
		ret = (int) bvec->bv_len;

	return ret;
3497 3498 3499 3500 3501 3502 3503 3504 3505
}

static void rbd_free_disk(struct rbd_device *rbd_dev)
{
	struct gendisk *disk = rbd_dev->disk;

	if (!disk)
		return;

3506 3507
	rbd_dev->disk = NULL;
	if (disk->flags & GENHD_FL_UP) {
3508
		del_gendisk(disk);
3509 3510
		if (disk->queue)
			blk_cleanup_queue(disk->queue);
C
Christoph Hellwig 已提交
3511
		blk_mq_free_tag_set(&rbd_dev->tag_set);
3512
	}
3513 3514 3515
	put_disk(disk);
}

3516 3517
static int rbd_obj_read_sync(struct rbd_device *rbd_dev,
				const char *object_name,
3518
				u64 offset, u64 length, void *buf)
3519 3520

{
3521
	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3522 3523 3524
	struct rbd_obj_request *obj_request;
	struct page **pages = NULL;
	u32 page_count;
3525
	size_t size;
3526 3527 3528 3529 3530
	int ret;

	page_count = (u32) calc_pages_for(offset, length);
	pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
	if (IS_ERR(pages))
3531
		return PTR_ERR(pages);
3532 3533 3534

	ret = -ENOMEM;
	obj_request = rbd_obj_request_create(object_name, offset, length,
3535
							OBJ_REQUEST_PAGES);
3536 3537 3538 3539 3540 3541
	if (!obj_request)
		goto out;

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

G
Guangliang Zhao 已提交
3542
	obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1,
3543
						  obj_request);
3544 3545 3546
	if (!obj_request->osd_req)
		goto out;

3547 3548
	osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ,
					offset, length, 0, 0);
3549
	osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0,
3550
					obj_request->pages,
3551 3552 3553
					obj_request->length,
					obj_request->offset & ~PAGE_MASK,
					false, false);
3554
	rbd_osd_req_format_read(obj_request);
A
Alex Elder 已提交
3555

3556 3557 3558 3559 3560 3561 3562 3563 3564 3565
	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;
3566 3567 3568

	rbd_assert(obj_request->xferred <= (u64) SIZE_MAX);
	size = (size_t) obj_request->xferred;
3569
	ceph_copy_from_page_vector(pages, buf, 0, size);
3570 3571
	rbd_assert(size <= (size_t)INT_MAX);
	ret = (int)size;
3572 3573 3574 3575 3576 3577 3578 3579 3580
out:
	if (obj_request)
		rbd_obj_request_put(obj_request);
	else
		ceph_release_page_vector(pages, page_count);

	return ret;
}

3581
/*
A
Alex Elder 已提交
3582 3583 3584
 * 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.
3585
 */
3586
static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev)
3587
{
3588
	struct rbd_image_header_ondisk *ondisk = NULL;
3589
	u32 snap_count = 0;
3590 3591 3592
	u64 names_size = 0;
	u32 want_count;
	int ret;
3593

A
Alex Elder 已提交
3594
	/*
3595 3596 3597 3598 3599
	 * 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
3600
	 */
3601 3602 3603 3604 3605 3606 3607 3608 3609 3610
	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
3611
			return -ENOMEM;
3612

3613
		ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name,
3614
				       0, size, ondisk);
3615
		if (ret < 0)
A
Alex Elder 已提交
3616
			goto out;
A
Alex Elder 已提交
3617
		if ((size_t)ret < size) {
3618
			ret = -ENXIO;
A
Alex Elder 已提交
3619 3620
			rbd_warn(rbd_dev, "short header read (want %zd got %d)",
				size, ret);
A
Alex Elder 已提交
3621
			goto out;
3622 3623 3624
		}
		if (!rbd_dev_ondisk_valid(ondisk)) {
			ret = -ENXIO;
A
Alex Elder 已提交
3625
			rbd_warn(rbd_dev, "invalid header");
A
Alex Elder 已提交
3626
			goto out;
3627
		}
3628

3629 3630 3631 3632
		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
3633

A
Alex Elder 已提交
3634 3635
	ret = rbd_header_from_disk(rbd_dev, ondisk);
out:
3636 3637 3638
	kfree(ondisk);

	return ret;
3639 3640
}

3641 3642 3643 3644 3645 3646 3647 3648 3649 3650 3651 3652 3653 3654 3655 3656 3657 3658 3659
/*
 * 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);
}

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
static void rbd_dev_update_size(struct rbd_device *rbd_dev)
{
	sector_t size;
	bool removing;

	/*
	 * Don't hold the lock while doing disk operations,
	 * or lock ordering will conflict with the bdev mutex via:
	 * rbd_add() -> blkdev_get() -> rbd_open()
	 */
	spin_lock_irq(&rbd_dev->lock);
	removing = test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags);
	spin_unlock_irq(&rbd_dev->lock);
	/*
	 * If the device is being removed, rbd_dev->disk has
	 * been destroyed, so don't try to update its size
	 */
	if (!removing) {
		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
3685
static int rbd_dev_refresh(struct rbd_device *rbd_dev)
A
Alex Elder 已提交
3686
{
3687
	u64 mapping_size;
A
Alex Elder 已提交
3688 3689
	int ret;

3690
	down_write(&rbd_dev->header_rwsem);
3691
	mapping_size = rbd_dev->mapping.size;
3692 3693

	ret = rbd_dev_header_info(rbd_dev);
3694
	if (ret)
3695
		goto out;
3696

3697 3698 3699 3700 3701 3702 3703
	/*
	 * 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)
3704
			goto out;
3705 3706
	}

3707
	if (rbd_dev->spec->snap_id == CEPH_NOSNAP) {
3708
		rbd_dev->mapping.size = rbd_dev->header.image_size;
3709 3710 3711 3712
	} else {
		/* validate mapped snapshot's EXISTS flag */
		rbd_exists_validate(rbd_dev);
	}
3713

3714
out:
3715
	up_write(&rbd_dev->header_rwsem);
3716
	if (!ret && mapping_size != rbd_dev->mapping.size)
3717
		rbd_dev_update_size(rbd_dev);
A
Alex Elder 已提交
3718

3719
	return ret;
A
Alex Elder 已提交
3720 3721
}

C
Christoph Hellwig 已提交
3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737
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,
};

3738 3739 3740 3741
static int rbd_init_disk(struct rbd_device *rbd_dev)
{
	struct gendisk *disk;
	struct request_queue *q;
A
Alex Elder 已提交
3742
	u64 segment_size;
C
Christoph Hellwig 已提交
3743
	int err;
3744 3745

	/* create gendisk info */
3746 3747 3748
	disk = alloc_disk(single_major ?
			  (1 << RBD_SINGLE_MAJOR_PART_SHIFT) :
			  RBD_MINORS_PER_MAJOR);
3749
	if (!disk)
A
Alex Elder 已提交
3750
		return -ENOMEM;
3751

A
Alex Elder 已提交
3752
	snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
A
Alex Elder 已提交
3753
		 rbd_dev->dev_id);
3754
	disk->major = rbd_dev->major;
3755
	disk->first_minor = rbd_dev->minor;
3756 3757
	if (single_major)
		disk->flags |= GENHD_FL_EXT_DEVT;
3758 3759 3760
	disk->fops = &rbd_bd_ops;
	disk->private_data = rbd_dev;

C
Christoph Hellwig 已提交
3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771
	memset(&rbd_dev->tag_set, 0, sizeof(rbd_dev->tag_set));
	rbd_dev->tag_set.ops = &rbd_mq_ops;
	rbd_dev->tag_set.queue_depth = BLKDEV_MAX_RQ;
	rbd_dev->tag_set.numa_node = NUMA_NO_NODE;
	rbd_dev->tag_set.flags =
		BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
	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)
3772
		goto out_disk;
3773

C
Christoph Hellwig 已提交
3774 3775 3776 3777 3778 3779
	q = blk_mq_init_queue(&rbd_dev->tag_set);
	if (IS_ERR(q)) {
		err = PTR_ERR(q);
		goto out_tag_set;
	}

3780 3781
	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
3782

3783
	/* set io sizes to object size */
A
Alex Elder 已提交
3784 3785
	segment_size = rbd_obj_bytes(&rbd_dev->header);
	blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE);
I
Ilya Dryomov 已提交
3786
	blk_queue_max_segments(q, segment_size / SECTOR_SIZE);
A
Alex Elder 已提交
3787 3788 3789
	blk_queue_max_segment_size(q, segment_size);
	blk_queue_io_min(q, segment_size);
	blk_queue_io_opt(q, segment_size);
3790

3791 3792 3793 3794
	/* enable the discard support */
	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
	q->limits.discard_granularity = segment_size;
	q->limits.discard_alignment = segment_size;
3795 3796
	q->limits.max_discard_sectors = segment_size / SECTOR_SIZE;
	q->limits.discard_zeroes_data = 1;
3797

3798 3799 3800 3801 3802 3803 3804 3805
	blk_queue_merge_bvec(q, rbd_merge_bvec);
	disk->queue = q;

	q->queuedata = rbd_dev;

	rbd_dev->disk = disk;

	return 0;
C
Christoph Hellwig 已提交
3806 3807
out_tag_set:
	blk_mq_free_tag_set(&rbd_dev->tag_set);
3808 3809
out_disk:
	put_disk(disk);
C
Christoph Hellwig 已提交
3810
	return err;
3811 3812
}

3813 3814 3815 3816
/*
  sysfs
*/

A
Alex Elder 已提交
3817 3818 3819 3820 3821
static struct rbd_device *dev_to_rbd_dev(struct device *dev)
{
	return container_of(dev, struct rbd_device, dev);
}

3822 3823 3824
static ssize_t rbd_size_show(struct device *dev,
			     struct device_attribute *attr, char *buf)
{
A
Alex Elder 已提交
3825
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3826

A
Alex Elder 已提交
3827 3828
	return sprintf(buf, "%llu\n",
		(unsigned long long)rbd_dev->mapping.size);
3829 3830
}

A
Alex Elder 已提交
3831 3832 3833 3834 3835 3836 3837 3838 3839 3840
/*
 * 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
3841
			(unsigned long long)rbd_dev->mapping.features);
A
Alex Elder 已提交
3842 3843
}

3844 3845 3846
static ssize_t rbd_major_show(struct device *dev,
			      struct device_attribute *attr, char *buf)
{
A
Alex Elder 已提交
3847
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3848

A
Alex Elder 已提交
3849 3850 3851 3852
	if (rbd_dev->major)
		return sprintf(buf, "%d\n", rbd_dev->major);

	return sprintf(buf, "(none)\n");
3853 3854 3855 3856 3857 3858
}

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
3859

3860
	return sprintf(buf, "%d\n", rbd_dev->minor);
3861 3862 3863 3864
}

static ssize_t rbd_client_id_show(struct device *dev,
				  struct device_attribute *attr, char *buf)
3865
{
A
Alex Elder 已提交
3866
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3867

3868 3869
	return sprintf(buf, "client%lld\n",
			ceph_client_id(rbd_dev->rbd_client->client));
3870 3871
}

3872 3873
static ssize_t rbd_pool_show(struct device *dev,
			     struct device_attribute *attr, char *buf)
3874
{
A
Alex Elder 已提交
3875
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3876

3877
	return sprintf(buf, "%s\n", rbd_dev->spec->pool_name);
3878 3879
}

3880 3881 3882 3883 3884
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);

3885
	return sprintf(buf, "%llu\n",
A
Alex Elder 已提交
3886
			(unsigned long long) rbd_dev->spec->pool_id);
3887 3888
}

3889 3890 3891
static ssize_t rbd_name_show(struct device *dev,
			     struct device_attribute *attr, char *buf)
{
A
Alex Elder 已提交
3892
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3893

A
Alex Elder 已提交
3894 3895 3896 3897
	if (rbd_dev->spec->image_name)
		return sprintf(buf, "%s\n", rbd_dev->spec->image_name);

	return sprintf(buf, "(unknown)\n");
3898 3899
}

A
Alex Elder 已提交
3900 3901 3902 3903 3904
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);

3905
	return sprintf(buf, "%s\n", rbd_dev->spec->image_id);
A
Alex Elder 已提交
3906 3907
}

A
Alex Elder 已提交
3908 3909 3910 3911
/*
 * Shows the name of the currently-mapped snapshot (or
 * RBD_SNAP_HEAD_NAME for the base image).
 */
3912 3913 3914 3915
static ssize_t rbd_snap_show(struct device *dev,
			     struct device_attribute *attr,
			     char *buf)
{
A
Alex Elder 已提交
3916
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3917

3918
	return sprintf(buf, "%s\n", rbd_dev->spec->snap_name);
3919 3920
}

3921
/*
3922 3923 3924
 * 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)".
3925 3926
 */
static ssize_t rbd_parent_show(struct device *dev,
3927 3928
			       struct device_attribute *attr,
			       char *buf)
3929 3930
{
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3931
	ssize_t count = 0;
3932

3933
	if (!rbd_dev->parent)
3934 3935
		return sprintf(buf, "(no parent image)\n");

3936 3937 3938 3939 3940 3941 3942 3943 3944 3945 3946 3947 3948 3949 3950 3951
	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;
3952 3953
}

3954 3955 3956 3957 3958
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
3959
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
3960
	int ret;
3961

A
Alex Elder 已提交
3962
	ret = rbd_dev_refresh(rbd_dev);
3963
	if (ret)
3964
		return ret;
3965

3966
	return size;
3967
}
3968

3969
static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
A
Alex Elder 已提交
3970
static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL);
3971
static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
3972
static DEVICE_ATTR(minor, S_IRUGO, rbd_minor_show, NULL);
3973 3974
static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
3975
static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL);
3976
static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
A
Alex Elder 已提交
3977
static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL);
3978 3979
static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
3980
static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL);
3981 3982 3983

static struct attribute *rbd_attrs[] = {
	&dev_attr_size.attr,
A
Alex Elder 已提交
3984
	&dev_attr_features.attr,
3985
	&dev_attr_major.attr,
3986
	&dev_attr_minor.attr,
3987 3988
	&dev_attr_client_id.attr,
	&dev_attr_pool.attr,
3989
	&dev_attr_pool_id.attr,
3990
	&dev_attr_name.attr,
A
Alex Elder 已提交
3991
	&dev_attr_image_id.attr,
3992
	&dev_attr_current_snap.attr,
3993
	&dev_attr_parent.attr,
3994 3995 3996 3997 3998 3999 4000 4001 4002 4003 4004 4005 4006 4007 4008 4009 4010 4011 4012 4013 4014 4015 4016
	&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
};

static void rbd_sysfs_dev_release(struct device *dev)
{
}

static struct device_type rbd_device_type = {
	.name		= "rbd",
	.groups		= rbd_attr_groups,
	.release	= rbd_sysfs_dev_release,
};

4017 4018 4019 4020 4021 4022 4023 4024 4025 4026 4027 4028 4029 4030 4031 4032 4033 4034 4035 4036 4037
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;
4038 4039 4040

	spec->pool_id = CEPH_NOPOOL;
	spec->snap_id = CEPH_NOSNAP;
4041 4042 4043 4044 4045 4046 4047 4048 4049 4050 4051 4052 4053 4054 4055 4056
	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);
}

A
Alex Elder 已提交
4057
static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc,
4058 4059 4060 4061 4062 4063 4064 4065 4066
				struct rbd_spec *spec)
{
	struct rbd_device *rbd_dev;

	rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL);
	if (!rbd_dev)
		return NULL;

	spin_lock_init(&rbd_dev->lock);
4067
	rbd_dev->flags = 0;
4068
	atomic_set(&rbd_dev->parent_ref, 0);
4069 4070 4071 4072 4073 4074
	INIT_LIST_HEAD(&rbd_dev->node);
	init_rwsem(&rbd_dev->header_rwsem);

	rbd_dev->spec = spec;
	rbd_dev->rbd_client = rbdc;

4075 4076 4077 4078 4079 4080 4081
	/* Initialize the layout used for all rbd requests */

	rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
	rbd_dev->layout.fl_stripe_count = cpu_to_le32(1);
	rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
	rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id);

4082 4083 4084 4085 4086 4087 4088 4089 4090 4091
	return rbd_dev;
}

static void rbd_dev_destroy(struct rbd_device *rbd_dev)
{
	rbd_put_client(rbd_dev->rbd_client);
	rbd_spec_put(rbd_dev->spec);
	kfree(rbd_dev);
}

4092 4093 4094 4095 4096 4097 4098 4099 4100 4101 4102 4103 4104 4105 4106
/*
 * 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 };

4107
	ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
4108
				"rbd", "get_size",
4109
				&snapid, sizeof (snapid),
4110
				&size_buf, sizeof (size_buf));
4111
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4112 4113
	if (ret < 0)
		return ret;
4114 4115
	if (ret < sizeof (size_buf))
		return -ERANGE;
4116

J
Josh Durgin 已提交
4117
	if (order) {
4118
		*order = size_buf.order;
J
Josh Durgin 已提交
4119 4120
		dout("  order %u", (unsigned int)*order);
	}
4121 4122
	*snap_size = le64_to_cpu(size_buf.size);

J
Josh Durgin 已提交
4123 4124
	dout("  snap_id 0x%016llx snap_size = %llu\n",
		(unsigned long long)snap_id,
4125
		(unsigned long long)*snap_size);
4126 4127 4128 4129 4130 4131 4132 4133 4134 4135 4136

	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);
}

4137 4138 4139 4140 4141 4142 4143 4144 4145 4146
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;

4147
	ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
4148
				"rbd", "get_object_prefix", NULL, 0,
4149
				reply_buf, RBD_OBJ_PREFIX_LEN_MAX);
4150
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4151 4152 4153 4154 4155
	if (ret < 0)
		goto out;

	p = reply_buf;
	rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p,
4156 4157
						p + ret, NULL, GFP_NOIO);
	ret = 0;
4158 4159 4160 4161 4162 4163 4164 4165 4166 4167 4168 4169 4170

	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;
}

4171 4172 4173 4174 4175 4176 4177
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;
4178
	} __attribute__ ((packed)) features_buf = { 0 };
A
Alex Elder 已提交
4179
	u64 incompat;
4180 4181
	int ret;

4182
	ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
4183
				"rbd", "get_features",
4184
				&snapid, sizeof (snapid),
4185
				&features_buf, sizeof (features_buf));
4186
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4187 4188
	if (ret < 0)
		return ret;
4189 4190
	if (ret < sizeof (features_buf))
		return -ERANGE;
A
Alex Elder 已提交
4191 4192

	incompat = le64_to_cpu(features_buf.incompat);
A
Alex Elder 已提交
4193
	if (incompat & ~RBD_FEATURES_SUPPORTED)
A
Alex Elder 已提交
4194
		return -ENXIO;
A
Alex Elder 已提交
4195

4196 4197 4198
	*snap_features = le64_to_cpu(features_buf.features);

	dout("  snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
4199 4200 4201
		(unsigned long long)snap_id,
		(unsigned long long)*snap_features,
		(unsigned long long)le64_to_cpu(features_buf.incompat));
4202 4203 4204 4205 4206 4207 4208 4209 4210 4211

	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);
}

4212 4213 4214 4215 4216 4217 4218 4219
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
4220
	u64 pool_id;
4221
	char *image_id;
4222
	u64 snap_id;
4223 4224 4225 4226 4227 4228 4229 4230 4231 4232 4233 4234 4235 4236 4237 4238 4239
	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;
	}

4240
	snapid = cpu_to_le64(rbd_dev->spec->snap_id);
4241
	ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
4242
				"rbd", "get_parent",
4243
				&snapid, sizeof (snapid),
4244
				reply_buf, size);
4245
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4246 4247 4248 4249
	if (ret < 0)
		goto out_err;

	p = reply_buf;
4250 4251
	end = reply_buf + ret;
	ret = -ERANGE;
A
Alex Elder 已提交
4252
	ceph_decode_64_safe(&p, end, pool_id, out_err);
4253 4254 4255 4256 4257 4258 4259 4260 4261 4262 4263 4264 4265 4266 4267 4268 4269
	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);
		}

4270
		goto out;	/* No parent?  No problem. */
4271
	}
4272

4273 4274 4275
	/* The ceph file layout needs to fit pool id in 32 bits */

	ret = -EIO;
A
Alex Elder 已提交
4276
	if (pool_id > (u64)U32_MAX) {
4277
		rbd_warn(NULL, "parent pool id too large (%llu > %u)",
A
Alex Elder 已提交
4278
			(unsigned long long)pool_id, U32_MAX);
4279
		goto out_err;
A
Alex Elder 已提交
4280
	}
4281

A
Alex Elder 已提交
4282
	image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
4283 4284 4285 4286
	if (IS_ERR(image_id)) {
		ret = PTR_ERR(image_id);
		goto out_err;
	}
4287
	ceph_decode_64_safe(&p, end, snap_id, out_err);
4288 4289
	ceph_decode_64_safe(&p, end, overlap, out_err);

4290 4291 4292 4293 4294 4295 4296 4297 4298
	/*
	 * 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
4299 4300
		rbd_dev->parent_spec = parent_spec;
		parent_spec = NULL;	/* rbd_dev now owns this */
4301 4302
	} else {
		kfree(image_id);
4303 4304 4305
	}

	/*
4306 4307
	 * We always update the parent overlap.  If it's zero we issue
	 * a warning, as we will proceed as if there was no parent.
4308 4309 4310
	 */
	if (!overlap) {
		if (parent_spec) {
4311 4312 4313 4314
			/* refresh, careful to warn just once */
			if (rbd_dev->parent_overlap)
				rbd_warn(rbd_dev,
				    "clone now standalone (overlap became 0)");
4315
		} else {
4316 4317
			/* initial probe */
			rbd_warn(rbd_dev, "clone is standalone (overlap 0)");
4318
		}
A
Alex Elder 已提交
4319
	}
4320 4321
	rbd_dev->parent_overlap = overlap;

4322 4323 4324 4325 4326 4327 4328 4329 4330
out:
	ret = 0;
out_err:
	kfree(reply_buf);
	rbd_spec_put(parent_spec);

	return ret;
}

4331 4332 4333 4334 4335 4336 4337 4338 4339 4340 4341 4342 4343 4344 4345
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;

	ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
				"rbd", "get_stripe_unit_count", NULL, 0,
4346
				(char *)&striping_info_buf, size);
4347 4348 4349 4350 4351 4352 4353 4354 4355 4356 4357 4358 4359 4360 4361 4362 4363 4364 4365 4366 4367 4368 4369 4370 4371 4372 4373 4374
	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;
	}
4375 4376
	rbd_dev->header.stripe_unit = stripe_unit;
	rbd_dev->header.stripe_count = stripe_count;
4377 4378 4379 4380

	return 0;
}

4381 4382 4383 4384 4385 4386 4387 4388 4389 4390 4391 4392 4393 4394
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
4395 4396
	len = strlen(rbd_dev->spec->image_id);
	image_id_size = sizeof (__le32) + len;
4397 4398 4399 4400 4401
	image_id = kmalloc(image_id_size, GFP_KERNEL);
	if (!image_id)
		return NULL;

	p = image_id;
4402
	end = image_id + image_id_size;
4403
	ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len);
4404 4405 4406 4407 4408 4409

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

4410
	ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY,
4411 4412
				"rbd", "dir_get_name",
				image_id, image_id_size,
4413
				reply_buf, size);
4414 4415 4416
	if (ret < 0)
		goto out;
	p = reply_buf;
4417 4418
	end = reply_buf + ret;

4419 4420 4421 4422 4423 4424 4425 4426 4427 4428 4429 4430
	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;
}

4431 4432 4433 4434 4435 4436 4437 4438 4439 4440 4441 4442 4443 4444 4445 4446 4447 4448 4449 4450 4451 4452 4453 4454 4455 4456 4457 4458 4459 4460
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);
4461 4462 4463 4464 4465 4466 4467
		if (IS_ERR(snap_name)) {
			/* ignore no-longer existing snapshots */
			if (PTR_ERR(snap_name) == -ENOENT)
				continue;
			else
				break;
		}
4468 4469 4470 4471 4472 4473 4474 4475 4476 4477 4478 4479 4480 4481 4482 4483 4484 4485
		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);
}

4486
/*
4487 4488 4489 4490 4491 4492 4493 4494 4495 4496 4497 4498 4499 4500 4501 4502 4503 4504 4505 4506 4507 4508 4509 4510 4511 4512 4513
 * 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.
4514
 *
4515 4516
 * All names in an rbd spec are dynamically allocated.  It's OK if we
 * can't figure out the name for an image id.
4517
 */
4518
static int rbd_spec_fill_names(struct rbd_device *rbd_dev)
4519
{
4520 4521 4522 4523 4524
	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;
4525 4526
	int ret;

4527 4528 4529
	rbd_assert(spec->pool_id != CEPH_NOPOOL);
	rbd_assert(spec->image_id);
	rbd_assert(spec->snap_id != CEPH_NOSNAP);
4530

4531
	/* Get the pool name; we have to make our own copy of this */
4532

4533 4534 4535
	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);
4536 4537
		return -EIO;
	}
4538 4539
	pool_name = kstrdup(pool_name, GFP_KERNEL);
	if (!pool_name)
4540 4541 4542 4543
		return -ENOMEM;

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

4544 4545
	image_name = rbd_dev_image_name(rbd_dev);
	if (!image_name)
A
Alex Elder 已提交
4546
		rbd_warn(rbd_dev, "unable to get image name");
4547

4548
	/* Fetch the snapshot name */
4549

4550
	snap_name = rbd_snap_name(rbd_dev, spec->snap_id);
4551 4552
	if (IS_ERR(snap_name)) {
		ret = PTR_ERR(snap_name);
4553
		goto out_err;
4554 4555 4556 4557 4558
	}

	spec->pool_name = pool_name;
	spec->image_name = image_name;
	spec->snap_name = snap_name;
4559 4560

	return 0;
4561

4562
out_err:
4563 4564
	kfree(image_name);
	kfree(pool_name);
4565 4566 4567
	return ret;
}

A
Alex Elder 已提交
4568
static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev)
4569 4570 4571 4572 4573 4574 4575 4576 4577 4578 4579 4580 4581 4582 4583 4584 4585 4586 4587 4588 4589 4590 4591
{
	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;

4592
	ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
4593
				"rbd", "get_snapcontext", NULL, 0,
4594
				reply_buf, size);
4595
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4596 4597 4598 4599
	if (ret < 0)
		goto out;

	p = reply_buf;
4600 4601
	end = reply_buf + ret;
	ret = -ERANGE;
4602 4603 4604 4605 4606 4607 4608 4609 4610 4611 4612 4613 4614 4615 4616 4617
	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;
4618
	ret = 0;
4619

4620
	snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
4621 4622 4623 4624 4625 4626 4627 4628
	if (!snapc) {
		ret = -ENOMEM;
		goto out;
	}
	snapc->seq = seq;
	for (i = 0; i < snap_count; i++)
		snapc->snaps[i] = ceph_decode_64(&p);

4629
	ceph_put_snap_context(rbd_dev->header.snapc);
4630 4631 4632
	rbd_dev->header.snapc = snapc;

	dout("  snap context seq = %llu, snap_count = %u\n",
4633
		(unsigned long long)seq, (unsigned int)snap_count);
4634 4635 4636
out:
	kfree(reply_buf);

4637
	return ret;
4638 4639
}

4640 4641
static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
					u64 snap_id)
A
Alex Elder 已提交
4642 4643 4644
{
	size_t size;
	void *reply_buf;
4645
	__le64 snapid;
A
Alex Elder 已提交
4646 4647 4648 4649 4650 4651 4652 4653 4654 4655
	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);

4656
	snapid = cpu_to_le64(snap_id);
4657
	ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name,
A
Alex Elder 已提交
4658
				"rbd", "get_snapshot_name",
4659
				&snapid, sizeof (snapid),
4660
				reply_buf, size);
4661
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
4662 4663
	if (ret < 0) {
		snap_name = ERR_PTR(ret);
A
Alex Elder 已提交
4664
		goto out;
4665
	}
A
Alex Elder 已提交
4666 4667

	p = reply_buf;
4668
	end = reply_buf + ret;
A
Alex Elder 已提交
4669
	snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
4670
	if (IS_ERR(snap_name))
A
Alex Elder 已提交
4671 4672
		goto out;

4673
	dout("  snap_id 0x%016llx snap_name = %s\n",
4674
		(unsigned long long)snap_id, snap_name);
A
Alex Elder 已提交
4675 4676 4677
out:
	kfree(reply_buf);

4678
	return snap_name;
A
Alex Elder 已提交
4679 4680
}

4681
static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev)
A
Alex Elder 已提交
4682
{
4683
	bool first_time = rbd_dev->header.object_prefix == NULL;
A
Alex Elder 已提交
4684 4685
	int ret;

4686 4687
	ret = rbd_dev_v2_image_size(rbd_dev);
	if (ret)
4688
		return ret;
4689

4690 4691 4692
	if (first_time) {
		ret = rbd_dev_v2_header_onetime(rbd_dev);
		if (ret)
4693
			return ret;
4694 4695
	}

A
Alex Elder 已提交
4696
	ret = rbd_dev_v2_snap_context(rbd_dev);
A
Alex Elder 已提交
4697 4698 4699 4700 4701
	dout("rbd_dev_v2_snap_context returned %d\n", ret);

	return ret;
}

4702 4703 4704 4705 4706 4707 4708 4709 4710 4711
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);
}

4712 4713 4714
static int rbd_bus_add_dev(struct rbd_device *rbd_dev)
{
	struct device *dev;
4715
	int ret;
4716

4717
	dev = &rbd_dev->dev;
4718 4719 4720
	dev->bus = &rbd_bus_type;
	dev->type = &rbd_device_type;
	dev->parent = &rbd_root_dev;
4721
	dev->release = rbd_dev_device_release;
A
Alex Elder 已提交
4722
	dev_set_name(dev, "%d", rbd_dev->dev_id);
4723 4724 4725
	ret = device_register(dev);

	return ret;
4726 4727
}

4728 4729 4730 4731 4732
static void rbd_bus_del_dev(struct rbd_device *rbd_dev)
{
	device_unregister(&rbd_dev->dev);
}

4733
/*
4734
 * Get a unique rbd identifier for the given new rbd_dev, and add
4735
 * the rbd_dev to the global list.
4736
 */
4737
static int rbd_dev_id_get(struct rbd_device *rbd_dev)
4738
{
4739 4740
	int new_dev_id;

4741 4742 4743
	new_dev_id = ida_simple_get(&rbd_dev_id_ida,
				    0, minor_to_rbd_dev_id(1 << MINORBITS),
				    GFP_KERNEL);
4744 4745 4746 4747
	if (new_dev_id < 0)
		return new_dev_id;

	rbd_dev->dev_id = new_dev_id;
4748 4749 4750 4751

	spin_lock(&rbd_dev_list_lock);
	list_add_tail(&rbd_dev->node, &rbd_dev_list);
	spin_unlock(&rbd_dev_list_lock);
4752

4753
	dout("rbd_dev %p given dev id %d\n", rbd_dev, rbd_dev->dev_id);
4754 4755

	return 0;
4756
}
4757

4758
/*
4759 4760
 * Remove an rbd_dev from the global list, and record that its
 * identifier is no longer in use.
4761
 */
A
Alex Elder 已提交
4762
static void rbd_dev_id_put(struct rbd_device *rbd_dev)
4763
{
4764 4765 4766
	spin_lock(&rbd_dev_list_lock);
	list_del_init(&rbd_dev->node);
	spin_unlock(&rbd_dev_list_lock);
4767

4768 4769 4770
	ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id);

	dout("rbd_dev %p released dev id %d\n", rbd_dev, rbd_dev->dev_id);
4771 4772
}

4773 4774 4775
/*
 * 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
4776 4777
 * the token (string of non-white space characters) found.  Note
 * that *buf must be terminated with '\0'.
4778 4779 4780 4781 4782 4783 4784 4785 4786 4787 4788 4789 4790 4791
 */
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
4792 4793 4794 4795 4796 4797 4798 4799 4800 4801 4802 4803 4804 4805 4806 4807 4808 4809 4810 4811 4812 4813
/*
 * 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
4814
	dup = kmemdup(*buf, len + 1, GFP_KERNEL);
A
Alex Elder 已提交
4815 4816 4817 4818 4819 4820 4821 4822 4823 4824 4825
	if (!dup)
		return NULL;
	*(dup + len) = '\0';
	*buf += len;

	if (lenp)
		*lenp = len;

	return dup;
}

4826
/*
4827 4828 4829 4830
 * 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
4831
 *
4832 4833 4834 4835 4836 4837 4838 4839 4840 4841 4842 4843 4844 4845 4846 4847 4848 4849 4850 4851 4852 4853 4854 4855 4856 4857 4858 4859 4860 4861 4862 4863 4864 4865
 * 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.
4866
 */
4867
static int rbd_add_parse_args(const char *buf,
4868
				struct ceph_options **ceph_opts,
4869 4870
				struct rbd_options **opts,
				struct rbd_spec **rbd_spec)
4871
{
A
Alex Elder 已提交
4872
	size_t len;
4873
	char *options;
4874
	const char *mon_addrs;
4875
	char *snap_name;
4876
	size_t mon_addrs_size;
4877
	struct rbd_spec *spec = NULL;
4878
	struct rbd_options *rbd_opts = NULL;
4879
	struct ceph_options *copts;
4880
	int ret;
4881 4882 4883

	/* The first four tokens are required */

4884
	len = next_token(&buf);
4885 4886 4887 4888
	if (!len) {
		rbd_warn(NULL, "no monitor address(es) provided");
		return -EINVAL;
	}
4889
	mon_addrs = buf;
4890
	mon_addrs_size = len + 1;
4891
	buf += len;
4892

4893
	ret = -EINVAL;
4894 4895
	options = dup_token(&buf, NULL);
	if (!options)
4896
		return -ENOMEM;
4897 4898 4899 4900
	if (!*options) {
		rbd_warn(NULL, "no options provided");
		goto out_err;
	}
4901

4902 4903
	spec = rbd_spec_alloc();
	if (!spec)
4904
		goto out_mem;
4905 4906 4907 4908

	spec->pool_name = dup_token(&buf, NULL);
	if (!spec->pool_name)
		goto out_mem;
4909 4910 4911 4912
	if (!*spec->pool_name) {
		rbd_warn(NULL, "no pool name provided");
		goto out_err;
	}
4913

A
Alex Elder 已提交
4914
	spec->image_name = dup_token(&buf, NULL);
4915
	if (!spec->image_name)
4916
		goto out_mem;
4917 4918 4919 4920
	if (!*spec->image_name) {
		rbd_warn(NULL, "no image name provided");
		goto out_err;
	}
4921

4922 4923 4924 4925
	/*
	 * Snapshot name is optional; default is to use "-"
	 * (indicating the head/no snapshot).
	 */
4926
	len = next_token(&buf);
4927
	if (!len) {
4928 4929
		buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */
		len = sizeof (RBD_SNAP_HEAD_NAME) - 1;
4930
	} else if (len > RBD_MAX_SNAP_NAME_LEN) {
4931
		ret = -ENAMETOOLONG;
4932
		goto out_err;
4933
	}
4934 4935
	snap_name = kmemdup(buf, len + 1, GFP_KERNEL);
	if (!snap_name)
4936
		goto out_mem;
4937 4938
	*(snap_name + len) = '\0';
	spec->snap_name = snap_name;
A
Alex Elder 已提交
4939

4940
	/* Initialize all rbd options to the defaults */
4941

4942 4943 4944 4945 4946
	rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL);
	if (!rbd_opts)
		goto out_mem;

	rbd_opts->read_only = RBD_READ_ONLY_DEFAULT;
A
Alex Elder 已提交
4947

4948
	copts = ceph_parse_options(options, mon_addrs,
4949
					mon_addrs + mon_addrs_size - 1,
4950
					parse_rbd_opts_token, rbd_opts);
4951 4952
	if (IS_ERR(copts)) {
		ret = PTR_ERR(copts);
4953 4954
		goto out_err;
	}
4955 4956 4957
	kfree(options);

	*ceph_opts = copts;
4958
	*opts = rbd_opts;
4959
	*rbd_spec = spec;
4960

4961
	return 0;
4962
out_mem:
4963
	ret = -ENOMEM;
A
Alex Elder 已提交
4964
out_err:
4965 4966
	kfree(rbd_opts);
	rbd_spec_put(spec);
4967
	kfree(options);
A
Alex Elder 已提交
4968

4969
	return ret;
4970 4971
}

4972 4973 4974 4975 4976
/*
 * Return pool id (>= 0) or a negative error code.
 */
static int rbd_add_get_pool_id(struct rbd_client *rbdc, const char *pool_name)
{
4977
	struct ceph_options *opts = rbdc->client->options;
4978 4979 4980 4981 4982 4983 4984 4985 4986 4987 4988 4989 4990 4991 4992
	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) {
		ret = ceph_monc_do_get_version(&rbdc->client->monc, "osdmap",
					       &newest_epoch);
		if (ret < 0)
			return ret;

		if (rbdc->client->osdc.osdmap->epoch < newest_epoch) {
			ceph_monc_request_next_osdmap(&rbdc->client->monc);
			(void) ceph_monc_wait_osdmap(&rbdc->client->monc,
4993 4994
						     newest_epoch,
						     opts->mount_timeout);
4995 4996 4997 4998 4999 5000 5001 5002 5003 5004
			goto again;
		} else {
			/* the osdmap we have is new enough */
			return -ENOENT;
		}
	}

	return ret;
}

A
Alex Elder 已提交
5005 5006 5007 5008 5009 5010 5011 5012 5013 5014 5015 5016 5017 5018 5019 5020 5021 5022 5023 5024
/*
 * 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;
5025
	char *image_id;
5026

A
Alex Elder 已提交
5027 5028 5029
	/*
	 * When probing a parent image, the image id is already
	 * known (and the image name likely is not).  There's no
5030 5031
	 * 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
5032
	 */
5033 5034 5035
	if (rbd_dev->spec->image_id) {
		rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1;

A
Alex Elder 已提交
5036
		return 0;
5037
	}
A
Alex Elder 已提交
5038

A
Alex Elder 已提交
5039 5040 5041 5042
	/*
	 * 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
5043
	size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name);
A
Alex Elder 已提交
5044 5045 5046
	object_name = kmalloc(size, GFP_NOIO);
	if (!object_name)
		return -ENOMEM;
5047
	sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name);
A
Alex Elder 已提交
5048 5049 5050 5051 5052 5053 5054 5055 5056 5057 5058
	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;
	}

5059 5060
	/* If it doesn't exist we'll assume it's a format 1 image */

5061
	ret = rbd_obj_method_sync(rbd_dev, object_name,
5062
				"rbd", "get_id", NULL, 0,
5063
				response, RBD_IMAGE_ID_LEN_MAX);
5064
	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
5065 5066 5067 5068 5069
	if (ret == -ENOENT) {
		image_id = kstrdup("", GFP_KERNEL);
		ret = image_id ? 0 : -ENOMEM;
		if (!ret)
			rbd_dev->image_format = 1;
5070
	} else if (ret >= 0) {
5071 5072 5073
		void *p = response;

		image_id = ceph_extract_encoded_string(&p, p + ret,
A
Alex Elder 已提交
5074
						NULL, GFP_NOIO);
5075
		ret = PTR_ERR_OR_ZERO(image_id);
5076 5077 5078 5079 5080 5081 5082
		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
5083 5084 5085 5086 5087 5088 5089 5090
	}
out:
	kfree(response);
	kfree(object_name);

	return ret;
}

A
Alex Elder 已提交
5091 5092 5093 5094
/*
 * Undo whatever state changes are made by v1 or v2 header info
 * call.
 */
A
Alex Elder 已提交
5095 5096 5097 5098
static void rbd_dev_unprobe(struct rbd_device *rbd_dev)
{
	struct rbd_image_header	*header;

5099
	rbd_dev_parent_put(rbd_dev);
A
Alex Elder 已提交
5100 5101 5102 5103

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

	header = &rbd_dev->header;
5104
	ceph_put_snap_context(header->snapc);
A
Alex Elder 已提交
5105 5106 5107 5108 5109 5110
	kfree(header->snap_sizes);
	kfree(header->snap_names);
	kfree(header->object_prefix);
	memset(header, 0, sizeof (*header));
}

5111
static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev)
5112 5113 5114
{
	int ret;

5115
	ret = rbd_dev_v2_object_prefix(rbd_dev);
5116
	if (ret)
5117 5118
		goto out_err;

5119 5120 5121 5122
	/*
	 * Get the and check features for the image.  Currently the
	 * features are assumed to never change.
	 */
5123
	ret = rbd_dev_v2_features(rbd_dev);
5124
	if (ret)
5125
		goto out_err;
5126

5127 5128 5129 5130 5131 5132 5133
	/* 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;
	}
5134
	/* No support for crypto and compression type format 2 images */
5135

A
Alex Elder 已提交
5136
	return 0;
5137
out_err:
A
Alex Elder 已提交
5138
	rbd_dev->header.features = 0;
5139 5140
	kfree(rbd_dev->header.object_prefix);
	rbd_dev->header.object_prefix = NULL;
5141 5142

	return ret;
5143 5144
}

5145
static int rbd_dev_probe_parent(struct rbd_device *rbd_dev)
A
Alex Elder 已提交
5146
{
5147
	struct rbd_device *parent = NULL;
5148 5149 5150 5151 5152 5153 5154 5155 5156 5157 5158 5159 5160 5161 5162 5163 5164 5165 5166
	struct rbd_spec *parent_spec;
	struct rbd_client *rbdc;
	int ret;

	if (!rbd_dev->parent_spec)
		return 0;
	/*
	 * We need to pass a reference to the client and the parent
	 * spec when creating the parent rbd_dev.  Images related by
	 * parent/child relationships always share both.
	 */
	parent_spec = rbd_spec_get(rbd_dev->parent_spec);
	rbdc = __rbd_get_client(rbd_dev->rbd_client);

	ret = -ENOMEM;
	parent = rbd_dev_create(rbdc, parent_spec);
	if (!parent)
		goto out_err;

5167
	ret = rbd_dev_image_probe(parent, false);
5168 5169 5170
	if (ret < 0)
		goto out_err;
	rbd_dev->parent = parent;
5171
	atomic_set(&rbd_dev->parent_ref, 1);
5172 5173 5174 5175

	return 0;
out_err:
	if (parent) {
A
Alex Elder 已提交
5176
		rbd_dev_unparent(rbd_dev);
5177 5178 5179 5180 5181 5182 5183 5184 5185 5186
		kfree(rbd_dev->header_name);
		rbd_dev_destroy(parent);
	} else {
		rbd_put_client(rbdc);
		rbd_spec_put(parent_spec);
	}

	return ret;
}

5187
static int rbd_dev_device_setup(struct rbd_device *rbd_dev)
5188
{
A
Alex Elder 已提交
5189
	int ret;
A
Alex Elder 已提交
5190

5191 5192 5193 5194 5195
	/* Get an id and fill in device name. */

	ret = rbd_dev_id_get(rbd_dev);
	if (ret)
		return ret;
A
Alex Elder 已提交
5196 5197 5198 5199 5200

	BUILD_BUG_ON(DEV_NAME_LEN
			< sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH);
	sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id);

5201
	/* Record our major and minor device numbers. */
A
Alex Elder 已提交
5202

5203 5204 5205 5206 5207 5208 5209 5210 5211 5212 5213
	if (!single_major) {
		ret = register_blkdev(0, rbd_dev->name);
		if (ret < 0)
			goto err_out_id;

		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
5214 5215 5216 5217 5218 5219 5220

	/* Set up the blkdev mapping. */

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

5221
	ret = rbd_dev_mapping_set(rbd_dev);
A
Alex Elder 已提交
5222 5223
	if (ret)
		goto err_out_disk;
I
Ilya Dryomov 已提交
5224

5225
	set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE);
5226
	set_disk_ro(rbd_dev->disk, rbd_dev->mapping.read_only);
5227 5228 5229

	ret = rbd_bus_add_dev(rbd_dev);
	if (ret)
5230
		goto err_out_mapping;
A
Alex Elder 已提交
5231 5232 5233

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

5234
	set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
A
Alex Elder 已提交
5235 5236 5237 5238 5239 5240
	add_disk(rbd_dev->disk);

	pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name,
		(unsigned long long) rbd_dev->mapping.size);

	return ret;
5241

5242 5243
err_out_mapping:
	rbd_dev_mapping_clear(rbd_dev);
A
Alex Elder 已提交
5244 5245 5246
err_out_disk:
	rbd_free_disk(rbd_dev);
err_out_blkdev:
5247 5248
	if (!single_major)
		unregister_blkdev(rbd_dev->major, rbd_dev->name);
A
Alex Elder 已提交
5249 5250
err_out_id:
	rbd_dev_id_put(rbd_dev);
A
Alex Elder 已提交
5251
	rbd_dev_mapping_clear(rbd_dev);
A
Alex Elder 已提交
5252 5253 5254 5255

	return ret;
}

A
Alex Elder 已提交
5256 5257 5258 5259 5260 5261 5262 5263 5264 5265 5266 5267 5268 5269 5270 5271 5272 5273 5274 5275 5276 5277 5278 5279 5280 5281 5282
static int rbd_dev_header_name(struct rbd_device *rbd_dev)
{
	struct rbd_spec *spec = rbd_dev->spec;
	size_t size;

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

	rbd_assert(rbd_image_format_valid(rbd_dev->image_format));

	if (rbd_dev->image_format == 1)
		size = strlen(spec->image_name) + sizeof (RBD_SUFFIX);
	else
		size = sizeof (RBD_HEADER_PREFIX) + strlen(spec->image_id);

	rbd_dev->header_name = kmalloc(size, GFP_KERNEL);
	if (!rbd_dev->header_name)
		return -ENOMEM;

	if (rbd_dev->image_format == 1)
		sprintf(rbd_dev->header_name, "%s%s",
			spec->image_name, RBD_SUFFIX);
	else
		sprintf(rbd_dev->header_name, "%s%s",
			RBD_HEADER_PREFIX, spec->image_id);
	return 0;
}

5283 5284
static void rbd_dev_image_release(struct rbd_device *rbd_dev)
{
A
Alex Elder 已提交
5285
	rbd_dev_unprobe(rbd_dev);
5286
	kfree(rbd_dev->header_name);
A
Alex Elder 已提交
5287 5288 5289 5290 5291
	rbd_dev->header_name = NULL;
	rbd_dev->image_format = 0;
	kfree(rbd_dev->spec->image_id);
	rbd_dev->spec->image_id = NULL;

5292 5293 5294
	rbd_dev_destroy(rbd_dev);
}

5295 5296
/*
 * Probe for the existence of the header object for the given rbd
5297 5298 5299
 * 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.
5300
 */
5301
static int rbd_dev_image_probe(struct rbd_device *rbd_dev, bool mapping)
5302 5303 5304 5305
{
	int ret;

	/*
A
Alex Elder 已提交
5306 5307 5308 5309
	 * 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.
5310 5311 5312
	 */
	ret = rbd_dev_image_id(rbd_dev);
	if (ret)
5313 5314
		return ret;

A
Alex Elder 已提交
5315 5316 5317 5318
	ret = rbd_dev_header_name(rbd_dev);
	if (ret)
		goto err_out_format;

5319
	if (mapping) {
5320
		ret = rbd_dev_header_watch_sync(rbd_dev);
5321 5322 5323 5324 5325
		if (ret) {
			if (ret == -ENOENT)
				pr_info("image %s/%s does not exist\n",
					rbd_dev->spec->pool_name,
					rbd_dev->spec->image_name);
5326
			goto out_header_name;
5327
		}
5328
	}
5329

5330
	ret = rbd_dev_header_info(rbd_dev);
5331
	if (ret)
5332
		goto err_out_watch;
A
Alex Elder 已提交
5333

5334 5335 5336 5337 5338 5339 5340 5341 5342 5343
	/*
	 * 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.
	 */
	if (mapping)
		ret = rbd_spec_fill_snap_id(rbd_dev);
	else
		ret = rbd_spec_fill_names(rbd_dev);
5344 5345 5346 5347 5348 5349
	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
5350
		goto err_out_probe;
5351
	}
5352

5353 5354 5355 5356 5357 5358 5359 5360 5361 5362 5363 5364 5365 5366
	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.
		 */
		if (mapping && rbd_dev->parent_spec)
			rbd_warn(rbd_dev,
				 "WARNING: kernel layering is EXPERIMENTAL!");
	}

5367
	ret = rbd_dev_probe_parent(rbd_dev);
A
Alex Elder 已提交
5368 5369 5370 5371 5372 5373
	if (ret)
		goto err_out_probe;

	dout("discovered format %u image, header name is %s\n",
		rbd_dev->image_format, rbd_dev->header_name);
	return 0;
5374

A
Alex Elder 已提交
5375 5376
err_out_probe:
	rbd_dev_unprobe(rbd_dev);
5377
err_out_watch:
5378 5379
	if (mapping)
		rbd_dev_header_unwatch_sync(rbd_dev);
A
Alex Elder 已提交
5380 5381 5382 5383 5384
out_header_name:
	kfree(rbd_dev->header_name);
	rbd_dev->header_name = NULL;
err_out_format:
	rbd_dev->image_format = 0;
5385 5386
	kfree(rbd_dev->spec->image_id);
	rbd_dev->spec->image_id = NULL;
5387 5388 5389
	return ret;
}

5390 5391 5392
static ssize_t do_rbd_add(struct bus_type *bus,
			  const char *buf,
			  size_t count)
5393
{
5394
	struct rbd_device *rbd_dev = NULL;
5395
	struct ceph_options *ceph_opts = NULL;
5396
	struct rbd_options *rbd_opts = NULL;
5397
	struct rbd_spec *spec = NULL;
5398
	struct rbd_client *rbdc;
5399
	bool read_only;
5400
	int rc = -ENOMEM;
5401 5402 5403 5404 5405

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

	/* parse add command */
5406
	rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec);
5407
	if (rc < 0)
5408
		goto err_out_module;
5409 5410 5411
	read_only = rbd_opts->read_only;
	kfree(rbd_opts);
	rbd_opts = NULL;	/* done with this */
5412

5413 5414 5415
	rbdc = rbd_get_client(ceph_opts);
	if (IS_ERR(rbdc)) {
		rc = PTR_ERR(rbdc);
5416
		goto err_out_args;
5417
	}
5418 5419

	/* pick the pool */
5420
	rc = rbd_add_get_pool_id(rbdc, spec->pool_name);
5421 5422 5423
	if (rc < 0) {
		if (rc == -ENOENT)
			pr_info("pool %s does not exist\n", spec->pool_name);
5424
		goto err_out_client;
5425
	}
A
Alex Elder 已提交
5426
	spec->pool_id = (u64)rc;
5427

5428 5429
	/* The ceph file layout needs to fit pool id in 32 bits */

A
Alex Elder 已提交
5430
	if (spec->pool_id > (u64)U32_MAX) {
5431
		rbd_warn(NULL, "pool id too large (%llu > %u)",
A
Alex Elder 已提交
5432
				(unsigned long long)spec->pool_id, U32_MAX);
5433 5434 5435 5436
		rc = -EIO;
		goto err_out_client;
	}

5437
	rbd_dev = rbd_dev_create(rbdc, spec);
5438 5439
	if (!rbd_dev)
		goto err_out_client;
5440 5441
	rbdc = NULL;		/* rbd_dev now owns this */
	spec = NULL;		/* rbd_dev now owns this */
5442

5443
	rc = rbd_dev_image_probe(rbd_dev, true);
5444
	if (rc < 0)
5445
		goto err_out_rbd_dev;
5446

5447 5448 5449 5450 5451 5452
	/* If we are mapping a snapshot it must be marked read-only */

	if (rbd_dev->spec->snap_id != CEPH_NOSNAP)
		read_only = true;
	rbd_dev->mapping.read_only = read_only;

5453
	rc = rbd_dev_device_setup(rbd_dev);
A
Alex Elder 已提交
5454
	if (rc) {
5455 5456 5457 5458 5459 5460
		/*
		 * rbd_dev_header_unwatch_sync() can't be moved into
		 * rbd_dev_image_release() without refactoring, see
		 * commit 1f3ef78861ac.
		 */
		rbd_dev_header_unwatch_sync(rbd_dev);
A
Alex Elder 已提交
5461 5462 5463 5464 5465
		rbd_dev_image_release(rbd_dev);
		goto err_out_module;
	}

	return count;
5466

5467 5468
err_out_rbd_dev:
	rbd_dev_destroy(rbd_dev);
5469
err_out_client:
5470
	rbd_put_client(rbdc);
5471
err_out_args:
5472
	rbd_spec_put(spec);
5473 5474
err_out_module:
	module_put(THIS_MODULE);
5475

5476
	dout("Error adding device %s\n", buf);
5477

A
Alex Elder 已提交
5478
	return (ssize_t)rc;
5479 5480
}

5481 5482 5483 5484 5485 5486 5487 5488 5489 5490 5491 5492 5493 5494 5495 5496 5497
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);
}

5498
static void rbd_dev_device_release(struct device *dev)
5499
{
A
Alex Elder 已提交
5500
	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5501 5502

	rbd_free_disk(rbd_dev);
5503
	clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
A
Alex Elder 已提交
5504
	rbd_dev_mapping_clear(rbd_dev);
5505 5506
	if (!single_major)
		unregister_blkdev(rbd_dev->major, rbd_dev->name);
A
Alex Elder 已提交
5507
	rbd_dev_id_put(rbd_dev);
A
Alex Elder 已提交
5508
	rbd_dev_mapping_clear(rbd_dev);
5509 5510
}

5511 5512
static void rbd_dev_remove_parent(struct rbd_device *rbd_dev)
{
A
Alex Elder 已提交
5513
	while (rbd_dev->parent) {
5514 5515 5516 5517 5518 5519 5520 5521 5522 5523 5524 5525
		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
5526
		rbd_assert(second);
5527
		rbd_dev_image_release(second);
A
Alex Elder 已提交
5528 5529 5530 5531
		first->parent = NULL;
		first->parent_overlap = 0;

		rbd_assert(first->parent_spec);
5532 5533 5534 5535 5536
		rbd_spec_put(first->parent_spec);
		first->parent_spec = NULL;
	}
}

5537 5538 5539
static ssize_t do_rbd_remove(struct bus_type *bus,
			     const char *buf,
			     size_t count)
5540 5541
{
	struct rbd_device *rbd_dev = NULL;
5542 5543
	struct list_head *tmp;
	int dev_id;
5544
	unsigned long ul;
5545
	bool already = false;
5546
	int ret;
5547

5548
	ret = kstrtoul(buf, 10, &ul);
5549 5550
	if (ret)
		return ret;
5551 5552

	/* convert to int; abort if we lost anything in the conversion */
5553 5554
	dev_id = (int)ul;
	if (dev_id != ul)
5555 5556
		return -EINVAL;

5557 5558 5559 5560 5561 5562 5563 5564
	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;
		}
5565
	}
5566 5567 5568 5569 5570
	if (!ret) {
		spin_lock_irq(&rbd_dev->lock);
		if (rbd_dev->open_count)
			ret = -EBUSY;
		else
5571 5572
			already = test_and_set_bit(RBD_DEV_FLAG_REMOVING,
							&rbd_dev->flags);
5573 5574 5575
		spin_unlock_irq(&rbd_dev->lock);
	}
	spin_unlock(&rbd_dev_list_lock);
5576
	if (ret < 0 || already)
5577
		return ret;
5578

5579
	rbd_dev_header_unwatch_sync(rbd_dev);
5580 5581 5582 5583 5584 5585
	/*
	 * flush remaining watch callbacks - these must be complete
	 * before the osd_client is shutdown
	 */
	dout("%s: flushing notifies", __func__);
	ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc);
5586

5587 5588 5589 5590 5591 5592 5593
	/*
	 * 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.
	 */
	rbd_bus_del_dev(rbd_dev);
5594
	rbd_dev_image_release(rbd_dev);
A
Alex Elder 已提交
5595
	module_put(THIS_MODULE);
A
Alex Elder 已提交
5596

5597
	return count;
5598 5599
}

5600 5601 5602 5603 5604 5605 5606 5607 5608 5609 5610 5611 5612 5613 5614 5615 5616
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);
}

5617 5618
/*
 * create control files in sysfs
5619
 * /sys/bus/rbd/...
5620 5621 5622
 */
static int rbd_sysfs_init(void)
{
5623
	int ret;
5624

5625
	ret = device_register(&rbd_root_dev);
A
Alex Elder 已提交
5626
	if (ret < 0)
5627
		return ret;
5628

5629 5630 5631
	ret = bus_register(&rbd_bus_type);
	if (ret < 0)
		device_unregister(&rbd_root_dev);
5632 5633 5634 5635 5636 5637

	return ret;
}

static void rbd_sysfs_cleanup(void)
{
5638
	bus_unregister(&rbd_bus_type);
5639
	device_unregister(&rbd_root_dev);
5640 5641
}

5642 5643 5644 5645 5646 5647 5648
static int rbd_slab_init(void)
{
	rbd_assert(!rbd_img_request_cache);
	rbd_img_request_cache = kmem_cache_create("rbd_img_request",
					sizeof (struct rbd_img_request),
					__alignof__(struct rbd_img_request),
					0, NULL);
5649 5650 5651 5652 5653 5654 5655 5656
	if (!rbd_img_request_cache)
		return -ENOMEM;

	rbd_assert(!rbd_obj_request_cache);
	rbd_obj_request_cache = kmem_cache_create("rbd_obj_request",
					sizeof (struct rbd_obj_request),
					__alignof__(struct rbd_obj_request),
					0, NULL);
5657 5658 5659 5660 5661
	if (!rbd_obj_request_cache)
		goto out_err;

	rbd_assert(!rbd_segment_name_cache);
	rbd_segment_name_cache = kmem_cache_create("rbd_segment_name",
5662
					CEPH_MAX_OID_NAME_LEN + 1, 1, 0, NULL);
5663
	if (rbd_segment_name_cache)
5664
		return 0;
5665 5666 5667 5668 5669
out_err:
	if (rbd_obj_request_cache) {
		kmem_cache_destroy(rbd_obj_request_cache);
		rbd_obj_request_cache = NULL;
	}
5670

5671 5672 5673
	kmem_cache_destroy(rbd_img_request_cache);
	rbd_img_request_cache = NULL;

5674 5675 5676 5677 5678
	return -ENOMEM;
}

static void rbd_slab_exit(void)
{
5679 5680 5681 5682
	rbd_assert(rbd_segment_name_cache);
	kmem_cache_destroy(rbd_segment_name_cache);
	rbd_segment_name_cache = NULL;

5683 5684 5685 5686
	rbd_assert(rbd_obj_request_cache);
	kmem_cache_destroy(rbd_obj_request_cache);
	rbd_obj_request_cache = NULL;

5687 5688 5689 5690 5691
	rbd_assert(rbd_img_request_cache);
	kmem_cache_destroy(rbd_img_request_cache);
	rbd_img_request_cache = NULL;
}

A
Alex Elder 已提交
5692
static int __init rbd_init(void)
5693 5694 5695
{
	int rc;

5696 5697 5698 5699
	if (!libceph_compatible(NULL)) {
		rbd_warn(NULL, "libceph incompatibility (quitting)");
		return -EINVAL;
	}
I
Ilya Dryomov 已提交
5700

5701
	rc = rbd_slab_init();
5702 5703
	if (rc)
		return rc;
I
Ilya Dryomov 已提交
5704

5705 5706
	/*
	 * The number of active work items is limited by the number of
I
Ilya Dryomov 已提交
5707
	 * rbd devices * queue depth, so leave @max_active at default.
5708 5709 5710 5711 5712 5713 5714
	 */
	rbd_wq = alloc_workqueue(RBD_DRV_NAME, WQ_MEM_RECLAIM, 0);
	if (!rbd_wq) {
		rc = -ENOMEM;
		goto err_out_slab;
	}

5715 5716 5717 5718
	if (single_major) {
		rbd_major = register_blkdev(0, RBD_DRV_NAME);
		if (rbd_major < 0) {
			rc = rbd_major;
5719
			goto err_out_wq;
5720 5721 5722
		}
	}

5723 5724
	rc = rbd_sysfs_init();
	if (rc)
5725 5726 5727 5728 5729 5730
		goto err_out_blkdev;

	if (single_major)
		pr_info("loaded (major %d)\n", rbd_major);
	else
		pr_info("loaded\n");
5731

I
Ilya Dryomov 已提交
5732 5733
	return 0;

5734 5735 5736
err_out_blkdev:
	if (single_major)
		unregister_blkdev(rbd_major, RBD_DRV_NAME);
5737 5738
err_out_wq:
	destroy_workqueue(rbd_wq);
I
Ilya Dryomov 已提交
5739 5740
err_out_slab:
	rbd_slab_exit();
5741
	return rc;
5742 5743
}

A
Alex Elder 已提交
5744
static void __exit rbd_exit(void)
5745
{
I
Ilya Dryomov 已提交
5746
	ida_destroy(&rbd_dev_id_ida);
5747
	rbd_sysfs_cleanup();
5748 5749
	if (single_major)
		unregister_blkdev(rbd_major, RBD_DRV_NAME);
5750
	destroy_workqueue(rbd_wq);
5751
	rbd_slab_exit();
5752 5753 5754 5755 5756
}

module_init(rbd_init);
module_exit(rbd_exit);

A
Alex Elder 已提交
5757
MODULE_AUTHOR("Alex Elder <elder@inktank.com>");
5758 5759 5760 5761 5762
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>");

5763
MODULE_DESCRIPTION("RADOS Block Device (RBD) driver");
5764
MODULE_LICENSE("GPL");