xref: /openbmc/linux/drivers/block/rbd.c (revision c9933d49)
1 
2 /*
3    rbd.c -- Export ceph rados objects as a Linux block device
4 
5 
6    based on drivers/block/osdblk.c:
7 
8    Copyright 2009 Red Hat, Inc.
9 
10    This program is free software; you can redistribute it and/or modify
11    it under the terms of the GNU General Public License as published by
12    the Free Software Foundation.
13 
14    This program is distributed in the hope that it will be useful,
15    but WITHOUT ANY WARRANTY; without even the implied warranty of
16    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
17    GNU General Public License for more details.
18 
19    You should have received a copy of the GNU General Public License
20    along with this program; see the file COPYING.  If not, write to
21    the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
22 
23 
24 
25    For usage instructions, please refer to:
26 
27                  Documentation/ABI/testing/sysfs-bus-rbd
28 
29  */
30 
31 #include <linux/ceph/libceph.h>
32 #include <linux/ceph/osd_client.h>
33 #include <linux/ceph/mon_client.h>
34 #include <linux/ceph/cls_lock_client.h>
35 #include <linux/ceph/striper.h>
36 #include <linux/ceph/decode.h>
37 #include <linux/fs_parser.h>
38 #include <linux/bsearch.h>
39 
40 #include <linux/kernel.h>
41 #include <linux/device.h>
42 #include <linux/module.h>
43 #include <linux/blk-mq.h>
44 #include <linux/fs.h>
45 #include <linux/blkdev.h>
46 #include <linux/slab.h>
47 #include <linux/idr.h>
48 #include <linux/workqueue.h>
49 
50 #include "rbd_types.h"
51 
52 #define RBD_DEBUG	/* Activate rbd_assert() calls */
53 
54 /*
55  * Increment the given counter and return its updated value.
56  * If the counter is already 0 it will not be incremented.
57  * If the counter is already at its maximum value returns
58  * -EINVAL without updating it.
59  */
60 static int atomic_inc_return_safe(atomic_t *v)
61 {
62 	unsigned int counter;
63 
64 	counter = (unsigned int)atomic_fetch_add_unless(v, 1, 0);
65 	if (counter <= (unsigned int)INT_MAX)
66 		return (int)counter;
67 
68 	atomic_dec(v);
69 
70 	return -EINVAL;
71 }
72 
73 /* Decrement the counter.  Return the resulting value, or -EINVAL */
74 static int atomic_dec_return_safe(atomic_t *v)
75 {
76 	int counter;
77 
78 	counter = atomic_dec_return(v);
79 	if (counter >= 0)
80 		return counter;
81 
82 	atomic_inc(v);
83 
84 	return -EINVAL;
85 }
86 
87 #define RBD_DRV_NAME "rbd"
88 
89 #define RBD_MINORS_PER_MAJOR		256
90 #define RBD_SINGLE_MAJOR_PART_SHIFT	4
91 
92 #define RBD_MAX_PARENT_CHAIN_LEN	16
93 
94 #define RBD_SNAP_DEV_NAME_PREFIX	"snap_"
95 #define RBD_MAX_SNAP_NAME_LEN	\
96 			(NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1))
97 
98 #define RBD_MAX_SNAP_COUNT	510	/* allows max snapc to fit in 4KB */
99 
100 #define RBD_SNAP_HEAD_NAME	"-"
101 
102 #define	BAD_SNAP_INDEX	U32_MAX		/* invalid index into snap array */
103 
104 /* This allows a single page to hold an image name sent by OSD */
105 #define RBD_IMAGE_NAME_LEN_MAX	(PAGE_SIZE - sizeof (__le32) - 1)
106 #define RBD_IMAGE_ID_LEN_MAX	64
107 
108 #define RBD_OBJ_PREFIX_LEN_MAX	64
109 
110 #define RBD_NOTIFY_TIMEOUT	5	/* seconds */
111 #define RBD_RETRY_DELAY		msecs_to_jiffies(1000)
112 
113 /* Feature bits */
114 
115 #define RBD_FEATURE_LAYERING		(1ULL<<0)
116 #define RBD_FEATURE_STRIPINGV2		(1ULL<<1)
117 #define RBD_FEATURE_EXCLUSIVE_LOCK	(1ULL<<2)
118 #define RBD_FEATURE_OBJECT_MAP		(1ULL<<3)
119 #define RBD_FEATURE_FAST_DIFF		(1ULL<<4)
120 #define RBD_FEATURE_DEEP_FLATTEN	(1ULL<<5)
121 #define RBD_FEATURE_DATA_POOL		(1ULL<<7)
122 #define RBD_FEATURE_OPERATIONS		(1ULL<<8)
123 
124 #define RBD_FEATURES_ALL	(RBD_FEATURE_LAYERING |		\
125 				 RBD_FEATURE_STRIPINGV2 |	\
126 				 RBD_FEATURE_EXCLUSIVE_LOCK |	\
127 				 RBD_FEATURE_OBJECT_MAP |	\
128 				 RBD_FEATURE_FAST_DIFF |	\
129 				 RBD_FEATURE_DEEP_FLATTEN |	\
130 				 RBD_FEATURE_DATA_POOL |	\
131 				 RBD_FEATURE_OPERATIONS)
132 
133 /* Features supported by this (client software) implementation. */
134 
135 #define RBD_FEATURES_SUPPORTED	(RBD_FEATURES_ALL)
136 
137 /*
138  * An RBD device name will be "rbd#", where the "rbd" comes from
139  * RBD_DRV_NAME above, and # is a unique integer identifier.
140  */
141 #define DEV_NAME_LEN		32
142 
143 /*
144  * block device image metadata (in-memory version)
145  */
146 struct rbd_image_header {
147 	/* These six fields never change for a given rbd image */
148 	char *object_prefix;
149 	__u8 obj_order;
150 	u64 stripe_unit;
151 	u64 stripe_count;
152 	s64 data_pool_id;
153 	u64 features;		/* Might be changeable someday? */
154 
155 	/* The remaining fields need to be updated occasionally */
156 	u64 image_size;
157 	struct ceph_snap_context *snapc;
158 	char *snap_names;	/* format 1 only */
159 	u64 *snap_sizes;	/* format 1 only */
160 };
161 
162 /*
163  * An rbd image specification.
164  *
165  * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely
166  * identify an image.  Each rbd_dev structure includes a pointer to
167  * an rbd_spec structure that encapsulates this identity.
168  *
169  * Each of the id's in an rbd_spec has an associated name.  For a
170  * user-mapped image, the names are supplied and the id's associated
171  * with them are looked up.  For a layered image, a parent image is
172  * defined by the tuple, and the names are looked up.
173  *
174  * An rbd_dev structure contains a parent_spec pointer which is
175  * non-null if the image it represents is a child in a layered
176  * image.  This pointer will refer to the rbd_spec structure used
177  * by the parent rbd_dev for its own identity (i.e., the structure
178  * is shared between the parent and child).
179  *
180  * Since these structures are populated once, during the discovery
181  * phase of image construction, they are effectively immutable so
182  * we make no effort to synchronize access to them.
183  *
184  * Note that code herein does not assume the image name is known (it
185  * could be a null pointer).
186  */
187 struct rbd_spec {
188 	u64		pool_id;
189 	const char	*pool_name;
190 	const char	*pool_ns;	/* NULL if default, never "" */
191 
192 	const char	*image_id;
193 	const char	*image_name;
194 
195 	u64		snap_id;
196 	const char	*snap_name;
197 
198 	struct kref	kref;
199 };
200 
201 /*
202  * an instance of the client.  multiple devices may share an rbd client.
203  */
204 struct rbd_client {
205 	struct ceph_client	*client;
206 	struct kref		kref;
207 	struct list_head	node;
208 };
209 
210 struct pending_result {
211 	int			result;		/* first nonzero result */
212 	int			num_pending;
213 };
214 
215 struct rbd_img_request;
216 
217 enum obj_request_type {
218 	OBJ_REQUEST_NODATA = 1,
219 	OBJ_REQUEST_BIO,	/* pointer into provided bio (list) */
220 	OBJ_REQUEST_BVECS,	/* pointer into provided bio_vec array */
221 	OBJ_REQUEST_OWN_BVECS,	/* private bio_vec array, doesn't own pages */
222 };
223 
224 enum obj_operation_type {
225 	OBJ_OP_READ = 1,
226 	OBJ_OP_WRITE,
227 	OBJ_OP_DISCARD,
228 	OBJ_OP_ZEROOUT,
229 };
230 
231 #define RBD_OBJ_FLAG_DELETION			(1U << 0)
232 #define RBD_OBJ_FLAG_COPYUP_ENABLED		(1U << 1)
233 #define RBD_OBJ_FLAG_COPYUP_ZEROS		(1U << 2)
234 #define RBD_OBJ_FLAG_MAY_EXIST			(1U << 3)
235 #define RBD_OBJ_FLAG_NOOP_FOR_NONEXISTENT	(1U << 4)
236 
237 enum rbd_obj_read_state {
238 	RBD_OBJ_READ_START = 1,
239 	RBD_OBJ_READ_OBJECT,
240 	RBD_OBJ_READ_PARENT,
241 };
242 
243 /*
244  * Writes go through the following state machine to deal with
245  * layering:
246  *
247  *            . . . . . RBD_OBJ_WRITE_GUARD. . . . . . . . . . . . . .
248  *            .                 |                                    .
249  *            .                 v                                    .
250  *            .    RBD_OBJ_WRITE_READ_FROM_PARENT. . .               .
251  *            .                 |                    .               .
252  *            .                 v                    v (deep-copyup  .
253  *    (image  .   RBD_OBJ_WRITE_COPYUP_EMPTY_SNAPC   .  not needed)  .
254  * flattened) v                 |                    .               .
255  *            .                 v                    .               .
256  *            . . . .RBD_OBJ_WRITE_COPYUP_OPS. . . . .      (copyup  .
257  *                              |                        not needed) v
258  *                              v                                    .
259  *                            done . . . . . . . . . . . . . . . . . .
260  *                              ^
261  *                              |
262  *                     RBD_OBJ_WRITE_FLAT
263  *
264  * Writes start in RBD_OBJ_WRITE_GUARD or _FLAT, depending on whether
265  * assert_exists guard is needed or not (in some cases it's not needed
266  * even if there is a parent).
267  */
268 enum rbd_obj_write_state {
269 	RBD_OBJ_WRITE_START = 1,
270 	RBD_OBJ_WRITE_PRE_OBJECT_MAP,
271 	RBD_OBJ_WRITE_OBJECT,
272 	__RBD_OBJ_WRITE_COPYUP,
273 	RBD_OBJ_WRITE_COPYUP,
274 	RBD_OBJ_WRITE_POST_OBJECT_MAP,
275 };
276 
277 enum rbd_obj_copyup_state {
278 	RBD_OBJ_COPYUP_START = 1,
279 	RBD_OBJ_COPYUP_READ_PARENT,
280 	__RBD_OBJ_COPYUP_OBJECT_MAPS,
281 	RBD_OBJ_COPYUP_OBJECT_MAPS,
282 	__RBD_OBJ_COPYUP_WRITE_OBJECT,
283 	RBD_OBJ_COPYUP_WRITE_OBJECT,
284 };
285 
286 struct rbd_obj_request {
287 	struct ceph_object_extent ex;
288 	unsigned int		flags;	/* RBD_OBJ_FLAG_* */
289 	union {
290 		enum rbd_obj_read_state	 read_state;	/* for reads */
291 		enum rbd_obj_write_state write_state;	/* for writes */
292 	};
293 
294 	struct rbd_img_request	*img_request;
295 	struct ceph_file_extent	*img_extents;
296 	u32			num_img_extents;
297 
298 	union {
299 		struct ceph_bio_iter	bio_pos;
300 		struct {
301 			struct ceph_bvec_iter	bvec_pos;
302 			u32			bvec_count;
303 			u32			bvec_idx;
304 		};
305 	};
306 
307 	enum rbd_obj_copyup_state copyup_state;
308 	struct bio_vec		*copyup_bvecs;
309 	u32			copyup_bvec_count;
310 
311 	struct list_head	osd_reqs;	/* w/ r_private_item */
312 
313 	struct mutex		state_mutex;
314 	struct pending_result	pending;
315 	struct kref		kref;
316 };
317 
318 enum img_req_flags {
319 	IMG_REQ_CHILD,		/* initiator: block = 0, child image = 1 */
320 	IMG_REQ_LAYERED,	/* ENOENT handling: normal = 0, layered = 1 */
321 };
322 
323 enum rbd_img_state {
324 	RBD_IMG_START = 1,
325 	RBD_IMG_EXCLUSIVE_LOCK,
326 	__RBD_IMG_OBJECT_REQUESTS,
327 	RBD_IMG_OBJECT_REQUESTS,
328 };
329 
330 struct rbd_img_request {
331 	struct rbd_device	*rbd_dev;
332 	enum obj_operation_type	op_type;
333 	enum obj_request_type	data_type;
334 	unsigned long		flags;
335 	enum rbd_img_state	state;
336 	union {
337 		u64			snap_id;	/* for reads */
338 		struct ceph_snap_context *snapc;	/* for writes */
339 	};
340 	struct rbd_obj_request	*obj_request;	/* obj req initiator */
341 
342 	struct list_head	lock_item;
343 	struct list_head	object_extents;	/* obj_req.ex structs */
344 
345 	struct mutex		state_mutex;
346 	struct pending_result	pending;
347 	struct work_struct	work;
348 	int			work_result;
349 };
350 
351 #define for_each_obj_request(ireq, oreq) \
352 	list_for_each_entry(oreq, &(ireq)->object_extents, ex.oe_item)
353 #define for_each_obj_request_safe(ireq, oreq, n) \
354 	list_for_each_entry_safe(oreq, n, &(ireq)->object_extents, ex.oe_item)
355 
356 enum rbd_watch_state {
357 	RBD_WATCH_STATE_UNREGISTERED,
358 	RBD_WATCH_STATE_REGISTERED,
359 	RBD_WATCH_STATE_ERROR,
360 };
361 
362 enum rbd_lock_state {
363 	RBD_LOCK_STATE_UNLOCKED,
364 	RBD_LOCK_STATE_LOCKED,
365 	RBD_LOCK_STATE_RELEASING,
366 };
367 
368 /* WatchNotify::ClientId */
369 struct rbd_client_id {
370 	u64 gid;
371 	u64 handle;
372 };
373 
374 struct rbd_mapping {
375 	u64                     size;
376 };
377 
378 /*
379  * a single device
380  */
381 struct rbd_device {
382 	int			dev_id;		/* blkdev unique id */
383 
384 	int			major;		/* blkdev assigned major */
385 	int			minor;
386 	struct gendisk		*disk;		/* blkdev's gendisk and rq */
387 
388 	u32			image_format;	/* Either 1 or 2 */
389 	struct rbd_client	*rbd_client;
390 
391 	char			name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */
392 
393 	spinlock_t		lock;		/* queue, flags, open_count */
394 
395 	struct rbd_image_header	header;
396 	unsigned long		flags;		/* possibly lock protected */
397 	struct rbd_spec		*spec;
398 	struct rbd_options	*opts;
399 	char			*config_info;	/* add{,_single_major} string */
400 
401 	struct ceph_object_id	header_oid;
402 	struct ceph_object_locator header_oloc;
403 
404 	struct ceph_file_layout	layout;		/* used for all rbd requests */
405 
406 	struct mutex		watch_mutex;
407 	enum rbd_watch_state	watch_state;
408 	struct ceph_osd_linger_request *watch_handle;
409 	u64			watch_cookie;
410 	struct delayed_work	watch_dwork;
411 
412 	struct rw_semaphore	lock_rwsem;
413 	enum rbd_lock_state	lock_state;
414 	char			lock_cookie[32];
415 	struct rbd_client_id	owner_cid;
416 	struct work_struct	acquired_lock_work;
417 	struct work_struct	released_lock_work;
418 	struct delayed_work	lock_dwork;
419 	struct work_struct	unlock_work;
420 	spinlock_t		lock_lists_lock;
421 	struct list_head	acquiring_list;
422 	struct list_head	running_list;
423 	struct completion	acquire_wait;
424 	int			acquire_err;
425 	struct completion	releasing_wait;
426 
427 	spinlock_t		object_map_lock;
428 	u8			*object_map;
429 	u64			object_map_size;	/* in objects */
430 	u64			object_map_flags;
431 
432 	struct workqueue_struct	*task_wq;
433 
434 	struct rbd_spec		*parent_spec;
435 	u64			parent_overlap;
436 	atomic_t		parent_ref;
437 	struct rbd_device	*parent;
438 
439 	/* Block layer tags. */
440 	struct blk_mq_tag_set	tag_set;
441 
442 	/* protects updating the header */
443 	struct rw_semaphore     header_rwsem;
444 
445 	struct rbd_mapping	mapping;
446 
447 	struct list_head	node;
448 
449 	/* sysfs related */
450 	struct device		dev;
451 	unsigned long		open_count;	/* protected by lock */
452 };
453 
454 /*
455  * Flag bits for rbd_dev->flags:
456  * - REMOVING (which is coupled with rbd_dev->open_count) is protected
457  *   by rbd_dev->lock
458  */
459 enum rbd_dev_flags {
460 	RBD_DEV_FLAG_EXISTS,	/* rbd_dev_device_setup() ran */
461 	RBD_DEV_FLAG_REMOVING,	/* this mapping is being removed */
462 	RBD_DEV_FLAG_READONLY,  /* -o ro or snapshot */
463 };
464 
465 static DEFINE_MUTEX(client_mutex);	/* Serialize client creation */
466 
467 static LIST_HEAD(rbd_dev_list);    /* devices */
468 static DEFINE_SPINLOCK(rbd_dev_list_lock);
469 
470 static LIST_HEAD(rbd_client_list);		/* clients */
471 static DEFINE_SPINLOCK(rbd_client_list_lock);
472 
473 /* Slab caches for frequently-allocated structures */
474 
475 static struct kmem_cache	*rbd_img_request_cache;
476 static struct kmem_cache	*rbd_obj_request_cache;
477 
478 static int rbd_major;
479 static DEFINE_IDA(rbd_dev_id_ida);
480 
481 static struct workqueue_struct *rbd_wq;
482 
483 static struct ceph_snap_context rbd_empty_snapc = {
484 	.nref = REFCOUNT_INIT(1),
485 };
486 
487 /*
488  * single-major requires >= 0.75 version of userspace rbd utility.
489  */
490 static bool single_major = true;
491 module_param(single_major, bool, 0444);
492 MODULE_PARM_DESC(single_major, "Use a single major number for all rbd devices (default: true)");
493 
494 static ssize_t add_store(struct bus_type *bus, const char *buf, size_t count);
495 static ssize_t remove_store(struct bus_type *bus, const char *buf,
496 			    size_t count);
497 static ssize_t add_single_major_store(struct bus_type *bus, const char *buf,
498 				      size_t count);
499 static ssize_t remove_single_major_store(struct bus_type *bus, const char *buf,
500 					 size_t count);
501 static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth);
502 
503 static int rbd_dev_id_to_minor(int dev_id)
504 {
505 	return dev_id << RBD_SINGLE_MAJOR_PART_SHIFT;
506 }
507 
508 static int minor_to_rbd_dev_id(int minor)
509 {
510 	return minor >> RBD_SINGLE_MAJOR_PART_SHIFT;
511 }
512 
513 static bool rbd_is_ro(struct rbd_device *rbd_dev)
514 {
515 	return test_bit(RBD_DEV_FLAG_READONLY, &rbd_dev->flags);
516 }
517 
518 static bool rbd_is_snap(struct rbd_device *rbd_dev)
519 {
520 	return rbd_dev->spec->snap_id != CEPH_NOSNAP;
521 }
522 
523 static bool __rbd_is_lock_owner(struct rbd_device *rbd_dev)
524 {
525 	lockdep_assert_held(&rbd_dev->lock_rwsem);
526 
527 	return rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED ||
528 	       rbd_dev->lock_state == RBD_LOCK_STATE_RELEASING;
529 }
530 
531 static bool rbd_is_lock_owner(struct rbd_device *rbd_dev)
532 {
533 	bool is_lock_owner;
534 
535 	down_read(&rbd_dev->lock_rwsem);
536 	is_lock_owner = __rbd_is_lock_owner(rbd_dev);
537 	up_read(&rbd_dev->lock_rwsem);
538 	return is_lock_owner;
539 }
540 
541 static ssize_t supported_features_show(struct bus_type *bus, char *buf)
542 {
543 	return sprintf(buf, "0x%llx\n", RBD_FEATURES_SUPPORTED);
544 }
545 
546 static BUS_ATTR_WO(add);
547 static BUS_ATTR_WO(remove);
548 static BUS_ATTR_WO(add_single_major);
549 static BUS_ATTR_WO(remove_single_major);
550 static BUS_ATTR_RO(supported_features);
551 
552 static struct attribute *rbd_bus_attrs[] = {
553 	&bus_attr_add.attr,
554 	&bus_attr_remove.attr,
555 	&bus_attr_add_single_major.attr,
556 	&bus_attr_remove_single_major.attr,
557 	&bus_attr_supported_features.attr,
558 	NULL,
559 };
560 
561 static umode_t rbd_bus_is_visible(struct kobject *kobj,
562 				  struct attribute *attr, int index)
563 {
564 	if (!single_major &&
565 	    (attr == &bus_attr_add_single_major.attr ||
566 	     attr == &bus_attr_remove_single_major.attr))
567 		return 0;
568 
569 	return attr->mode;
570 }
571 
572 static const struct attribute_group rbd_bus_group = {
573 	.attrs = rbd_bus_attrs,
574 	.is_visible = rbd_bus_is_visible,
575 };
576 __ATTRIBUTE_GROUPS(rbd_bus);
577 
578 static struct bus_type rbd_bus_type = {
579 	.name		= "rbd",
580 	.bus_groups	= rbd_bus_groups,
581 };
582 
583 static void rbd_root_dev_release(struct device *dev)
584 {
585 }
586 
587 static struct device rbd_root_dev = {
588 	.init_name =    "rbd",
589 	.release =      rbd_root_dev_release,
590 };
591 
592 static __printf(2, 3)
593 void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...)
594 {
595 	struct va_format vaf;
596 	va_list args;
597 
598 	va_start(args, fmt);
599 	vaf.fmt = fmt;
600 	vaf.va = &args;
601 
602 	if (!rbd_dev)
603 		printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf);
604 	else if (rbd_dev->disk)
605 		printk(KERN_WARNING "%s: %s: %pV\n",
606 			RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf);
607 	else if (rbd_dev->spec && rbd_dev->spec->image_name)
608 		printk(KERN_WARNING "%s: image %s: %pV\n",
609 			RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf);
610 	else if (rbd_dev->spec && rbd_dev->spec->image_id)
611 		printk(KERN_WARNING "%s: id %s: %pV\n",
612 			RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf);
613 	else	/* punt */
614 		printk(KERN_WARNING "%s: rbd_dev %p: %pV\n",
615 			RBD_DRV_NAME, rbd_dev, &vaf);
616 	va_end(args);
617 }
618 
619 #ifdef RBD_DEBUG
620 #define rbd_assert(expr)						\
621 		if (unlikely(!(expr))) {				\
622 			printk(KERN_ERR "\nAssertion failure in %s() "	\
623 						"at line %d:\n\n"	\
624 					"\trbd_assert(%s);\n\n",	\
625 					__func__, __LINE__, #expr);	\
626 			BUG();						\
627 		}
628 #else /* !RBD_DEBUG */
629 #  define rbd_assert(expr)	((void) 0)
630 #endif /* !RBD_DEBUG */
631 
632 static void rbd_dev_remove_parent(struct rbd_device *rbd_dev);
633 
634 static int rbd_dev_refresh(struct rbd_device *rbd_dev);
635 static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev);
636 static int rbd_dev_header_info(struct rbd_device *rbd_dev);
637 static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev);
638 static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
639 					u64 snap_id);
640 static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
641 				u8 *order, u64 *snap_size);
642 static int rbd_dev_v2_get_flags(struct rbd_device *rbd_dev);
643 
644 static void rbd_obj_handle_request(struct rbd_obj_request *obj_req, int result);
645 static void rbd_img_handle_request(struct rbd_img_request *img_req, int result);
646 
647 /*
648  * Return true if nothing else is pending.
649  */
650 static bool pending_result_dec(struct pending_result *pending, int *result)
651 {
652 	rbd_assert(pending->num_pending > 0);
653 
654 	if (*result && !pending->result)
655 		pending->result = *result;
656 	if (--pending->num_pending)
657 		return false;
658 
659 	*result = pending->result;
660 	return true;
661 }
662 
663 static int rbd_open(struct block_device *bdev, fmode_t mode)
664 {
665 	struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
666 	bool removing = false;
667 
668 	spin_lock_irq(&rbd_dev->lock);
669 	if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags))
670 		removing = true;
671 	else
672 		rbd_dev->open_count++;
673 	spin_unlock_irq(&rbd_dev->lock);
674 	if (removing)
675 		return -ENOENT;
676 
677 	(void) get_device(&rbd_dev->dev);
678 
679 	return 0;
680 }
681 
682 static void rbd_release(struct gendisk *disk, fmode_t mode)
683 {
684 	struct rbd_device *rbd_dev = disk->private_data;
685 	unsigned long open_count_before;
686 
687 	spin_lock_irq(&rbd_dev->lock);
688 	open_count_before = rbd_dev->open_count--;
689 	spin_unlock_irq(&rbd_dev->lock);
690 	rbd_assert(open_count_before > 0);
691 
692 	put_device(&rbd_dev->dev);
693 }
694 
695 static const struct block_device_operations rbd_bd_ops = {
696 	.owner			= THIS_MODULE,
697 	.open			= rbd_open,
698 	.release		= rbd_release,
699 };
700 
701 /*
702  * Initialize an rbd client instance.  Success or not, this function
703  * consumes ceph_opts.  Caller holds client_mutex.
704  */
705 static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts)
706 {
707 	struct rbd_client *rbdc;
708 	int ret = -ENOMEM;
709 
710 	dout("%s:\n", __func__);
711 	rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL);
712 	if (!rbdc)
713 		goto out_opt;
714 
715 	kref_init(&rbdc->kref);
716 	INIT_LIST_HEAD(&rbdc->node);
717 
718 	rbdc->client = ceph_create_client(ceph_opts, rbdc);
719 	if (IS_ERR(rbdc->client))
720 		goto out_rbdc;
721 	ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */
722 
723 	ret = ceph_open_session(rbdc->client);
724 	if (ret < 0)
725 		goto out_client;
726 
727 	spin_lock(&rbd_client_list_lock);
728 	list_add_tail(&rbdc->node, &rbd_client_list);
729 	spin_unlock(&rbd_client_list_lock);
730 
731 	dout("%s: rbdc %p\n", __func__, rbdc);
732 
733 	return rbdc;
734 out_client:
735 	ceph_destroy_client(rbdc->client);
736 out_rbdc:
737 	kfree(rbdc);
738 out_opt:
739 	if (ceph_opts)
740 		ceph_destroy_options(ceph_opts);
741 	dout("%s: error %d\n", __func__, ret);
742 
743 	return ERR_PTR(ret);
744 }
745 
746 static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc)
747 {
748 	kref_get(&rbdc->kref);
749 
750 	return rbdc;
751 }
752 
753 /*
754  * Find a ceph client with specific addr and configuration.  If
755  * found, bump its reference count.
756  */
757 static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts)
758 {
759 	struct rbd_client *client_node;
760 	bool found = false;
761 
762 	if (ceph_opts->flags & CEPH_OPT_NOSHARE)
763 		return NULL;
764 
765 	spin_lock(&rbd_client_list_lock);
766 	list_for_each_entry(client_node, &rbd_client_list, node) {
767 		if (!ceph_compare_options(ceph_opts, client_node->client)) {
768 			__rbd_get_client(client_node);
769 
770 			found = true;
771 			break;
772 		}
773 	}
774 	spin_unlock(&rbd_client_list_lock);
775 
776 	return found ? client_node : NULL;
777 }
778 
779 /*
780  * (Per device) rbd map options
781  */
782 enum {
783 	Opt_queue_depth,
784 	Opt_alloc_size,
785 	Opt_lock_timeout,
786 	/* int args above */
787 	Opt_pool_ns,
788 	Opt_compression_hint,
789 	/* string args above */
790 	Opt_read_only,
791 	Opt_read_write,
792 	Opt_lock_on_read,
793 	Opt_exclusive,
794 	Opt_notrim,
795 };
796 
797 enum {
798 	Opt_compression_hint_none,
799 	Opt_compression_hint_compressible,
800 	Opt_compression_hint_incompressible,
801 };
802 
803 static const struct constant_table rbd_param_compression_hint[] = {
804 	{"none",		Opt_compression_hint_none},
805 	{"compressible",	Opt_compression_hint_compressible},
806 	{"incompressible",	Opt_compression_hint_incompressible},
807 	{}
808 };
809 
810 static const struct fs_parameter_spec rbd_parameters[] = {
811 	fsparam_u32	("alloc_size",			Opt_alloc_size),
812 	fsparam_enum	("compression_hint",		Opt_compression_hint,
813 			 rbd_param_compression_hint),
814 	fsparam_flag	("exclusive",			Opt_exclusive),
815 	fsparam_flag	("lock_on_read",		Opt_lock_on_read),
816 	fsparam_u32	("lock_timeout",		Opt_lock_timeout),
817 	fsparam_flag	("notrim",			Opt_notrim),
818 	fsparam_string	("_pool_ns",			Opt_pool_ns),
819 	fsparam_u32	("queue_depth",			Opt_queue_depth),
820 	fsparam_flag	("read_only",			Opt_read_only),
821 	fsparam_flag	("read_write",			Opt_read_write),
822 	fsparam_flag	("ro",				Opt_read_only),
823 	fsparam_flag	("rw",				Opt_read_write),
824 	{}
825 };
826 
827 struct rbd_options {
828 	int	queue_depth;
829 	int	alloc_size;
830 	unsigned long	lock_timeout;
831 	bool	read_only;
832 	bool	lock_on_read;
833 	bool	exclusive;
834 	bool	trim;
835 
836 	u32 alloc_hint_flags;  /* CEPH_OSD_OP_ALLOC_HINT_FLAG_* */
837 };
838 
839 #define RBD_QUEUE_DEPTH_DEFAULT	BLKDEV_DEFAULT_RQ
840 #define RBD_ALLOC_SIZE_DEFAULT	(64 * 1024)
841 #define RBD_LOCK_TIMEOUT_DEFAULT 0  /* no timeout */
842 #define RBD_READ_ONLY_DEFAULT	false
843 #define RBD_LOCK_ON_READ_DEFAULT false
844 #define RBD_EXCLUSIVE_DEFAULT	false
845 #define RBD_TRIM_DEFAULT	true
846 
847 struct rbd_parse_opts_ctx {
848 	struct rbd_spec		*spec;
849 	struct ceph_options	*copts;
850 	struct rbd_options	*opts;
851 };
852 
853 static char* obj_op_name(enum obj_operation_type op_type)
854 {
855 	switch (op_type) {
856 	case OBJ_OP_READ:
857 		return "read";
858 	case OBJ_OP_WRITE:
859 		return "write";
860 	case OBJ_OP_DISCARD:
861 		return "discard";
862 	case OBJ_OP_ZEROOUT:
863 		return "zeroout";
864 	default:
865 		return "???";
866 	}
867 }
868 
869 /*
870  * Destroy ceph client
871  *
872  * Caller must hold rbd_client_list_lock.
873  */
874 static void rbd_client_release(struct kref *kref)
875 {
876 	struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref);
877 
878 	dout("%s: rbdc %p\n", __func__, rbdc);
879 	spin_lock(&rbd_client_list_lock);
880 	list_del(&rbdc->node);
881 	spin_unlock(&rbd_client_list_lock);
882 
883 	ceph_destroy_client(rbdc->client);
884 	kfree(rbdc);
885 }
886 
887 /*
888  * Drop reference to ceph client node. If it's not referenced anymore, release
889  * it.
890  */
891 static void rbd_put_client(struct rbd_client *rbdc)
892 {
893 	if (rbdc)
894 		kref_put(&rbdc->kref, rbd_client_release);
895 }
896 
897 /*
898  * Get a ceph client with specific addr and configuration, if one does
899  * not exist create it.  Either way, ceph_opts is consumed by this
900  * function.
901  */
902 static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts)
903 {
904 	struct rbd_client *rbdc;
905 	int ret;
906 
907 	mutex_lock(&client_mutex);
908 	rbdc = rbd_client_find(ceph_opts);
909 	if (rbdc) {
910 		ceph_destroy_options(ceph_opts);
911 
912 		/*
913 		 * Using an existing client.  Make sure ->pg_pools is up to
914 		 * date before we look up the pool id in do_rbd_add().
915 		 */
916 		ret = ceph_wait_for_latest_osdmap(rbdc->client,
917 					rbdc->client->options->mount_timeout);
918 		if (ret) {
919 			rbd_warn(NULL, "failed to get latest osdmap: %d", ret);
920 			rbd_put_client(rbdc);
921 			rbdc = ERR_PTR(ret);
922 		}
923 	} else {
924 		rbdc = rbd_client_create(ceph_opts);
925 	}
926 	mutex_unlock(&client_mutex);
927 
928 	return rbdc;
929 }
930 
931 static bool rbd_image_format_valid(u32 image_format)
932 {
933 	return image_format == 1 || image_format == 2;
934 }
935 
936 static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk)
937 {
938 	size_t size;
939 	u32 snap_count;
940 
941 	/* The header has to start with the magic rbd header text */
942 	if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT)))
943 		return false;
944 
945 	/* The bio layer requires at least sector-sized I/O */
946 
947 	if (ondisk->options.order < SECTOR_SHIFT)
948 		return false;
949 
950 	/* If we use u64 in a few spots we may be able to loosen this */
951 
952 	if (ondisk->options.order > 8 * sizeof (int) - 1)
953 		return false;
954 
955 	/*
956 	 * The size of a snapshot header has to fit in a size_t, and
957 	 * that limits the number of snapshots.
958 	 */
959 	snap_count = le32_to_cpu(ondisk->snap_count);
960 	size = SIZE_MAX - sizeof (struct ceph_snap_context);
961 	if (snap_count > size / sizeof (__le64))
962 		return false;
963 
964 	/*
965 	 * Not only that, but the size of the entire the snapshot
966 	 * header must also be representable in a size_t.
967 	 */
968 	size -= snap_count * sizeof (__le64);
969 	if ((u64) size < le64_to_cpu(ondisk->snap_names_len))
970 		return false;
971 
972 	return true;
973 }
974 
975 /*
976  * returns the size of an object in the image
977  */
978 static u32 rbd_obj_bytes(struct rbd_image_header *header)
979 {
980 	return 1U << header->obj_order;
981 }
982 
983 static void rbd_init_layout(struct rbd_device *rbd_dev)
984 {
985 	if (rbd_dev->header.stripe_unit == 0 ||
986 	    rbd_dev->header.stripe_count == 0) {
987 		rbd_dev->header.stripe_unit = rbd_obj_bytes(&rbd_dev->header);
988 		rbd_dev->header.stripe_count = 1;
989 	}
990 
991 	rbd_dev->layout.stripe_unit = rbd_dev->header.stripe_unit;
992 	rbd_dev->layout.stripe_count = rbd_dev->header.stripe_count;
993 	rbd_dev->layout.object_size = rbd_obj_bytes(&rbd_dev->header);
994 	rbd_dev->layout.pool_id = rbd_dev->header.data_pool_id == CEPH_NOPOOL ?
995 			  rbd_dev->spec->pool_id : rbd_dev->header.data_pool_id;
996 	RCU_INIT_POINTER(rbd_dev->layout.pool_ns, NULL);
997 }
998 
999 /*
1000  * Fill an rbd image header with information from the given format 1
1001  * on-disk header.
1002  */
1003 static int rbd_header_from_disk(struct rbd_device *rbd_dev,
1004 				 struct rbd_image_header_ondisk *ondisk)
1005 {
1006 	struct rbd_image_header *header = &rbd_dev->header;
1007 	bool first_time = header->object_prefix == NULL;
1008 	struct ceph_snap_context *snapc;
1009 	char *object_prefix = NULL;
1010 	char *snap_names = NULL;
1011 	u64 *snap_sizes = NULL;
1012 	u32 snap_count;
1013 	int ret = -ENOMEM;
1014 	u32 i;
1015 
1016 	/* Allocate this now to avoid having to handle failure below */
1017 
1018 	if (first_time) {
1019 		object_prefix = kstrndup(ondisk->object_prefix,
1020 					 sizeof(ondisk->object_prefix),
1021 					 GFP_KERNEL);
1022 		if (!object_prefix)
1023 			return -ENOMEM;
1024 	}
1025 
1026 	/* Allocate the snapshot context and fill it in */
1027 
1028 	snap_count = le32_to_cpu(ondisk->snap_count);
1029 	snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
1030 	if (!snapc)
1031 		goto out_err;
1032 	snapc->seq = le64_to_cpu(ondisk->snap_seq);
1033 	if (snap_count) {
1034 		struct rbd_image_snap_ondisk *snaps;
1035 		u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len);
1036 
1037 		/* We'll keep a copy of the snapshot names... */
1038 
1039 		if (snap_names_len > (u64)SIZE_MAX)
1040 			goto out_2big;
1041 		snap_names = kmalloc(snap_names_len, GFP_KERNEL);
1042 		if (!snap_names)
1043 			goto out_err;
1044 
1045 		/* ...as well as the array of their sizes. */
1046 		snap_sizes = kmalloc_array(snap_count,
1047 					   sizeof(*header->snap_sizes),
1048 					   GFP_KERNEL);
1049 		if (!snap_sizes)
1050 			goto out_err;
1051 
1052 		/*
1053 		 * Copy the names, and fill in each snapshot's id
1054 		 * and size.
1055 		 *
1056 		 * Note that rbd_dev_v1_header_info() guarantees the
1057 		 * ondisk buffer we're working with has
1058 		 * snap_names_len bytes beyond the end of the
1059 		 * snapshot id array, this memcpy() is safe.
1060 		 */
1061 		memcpy(snap_names, &ondisk->snaps[snap_count], snap_names_len);
1062 		snaps = ondisk->snaps;
1063 		for (i = 0; i < snap_count; i++) {
1064 			snapc->snaps[i] = le64_to_cpu(snaps[i].id);
1065 			snap_sizes[i] = le64_to_cpu(snaps[i].image_size);
1066 		}
1067 	}
1068 
1069 	/* We won't fail any more, fill in the header */
1070 
1071 	if (first_time) {
1072 		header->object_prefix = object_prefix;
1073 		header->obj_order = ondisk->options.order;
1074 		rbd_init_layout(rbd_dev);
1075 	} else {
1076 		ceph_put_snap_context(header->snapc);
1077 		kfree(header->snap_names);
1078 		kfree(header->snap_sizes);
1079 	}
1080 
1081 	/* The remaining fields always get updated (when we refresh) */
1082 
1083 	header->image_size = le64_to_cpu(ondisk->image_size);
1084 	header->snapc = snapc;
1085 	header->snap_names = snap_names;
1086 	header->snap_sizes = snap_sizes;
1087 
1088 	return 0;
1089 out_2big:
1090 	ret = -EIO;
1091 out_err:
1092 	kfree(snap_sizes);
1093 	kfree(snap_names);
1094 	ceph_put_snap_context(snapc);
1095 	kfree(object_prefix);
1096 
1097 	return ret;
1098 }
1099 
1100 static const char *_rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, u32 which)
1101 {
1102 	const char *snap_name;
1103 
1104 	rbd_assert(which < rbd_dev->header.snapc->num_snaps);
1105 
1106 	/* Skip over names until we find the one we are looking for */
1107 
1108 	snap_name = rbd_dev->header.snap_names;
1109 	while (which--)
1110 		snap_name += strlen(snap_name) + 1;
1111 
1112 	return kstrdup(snap_name, GFP_KERNEL);
1113 }
1114 
1115 /*
1116  * Snapshot id comparison function for use with qsort()/bsearch().
1117  * Note that result is for snapshots in *descending* order.
1118  */
1119 static int snapid_compare_reverse(const void *s1, const void *s2)
1120 {
1121 	u64 snap_id1 = *(u64 *)s1;
1122 	u64 snap_id2 = *(u64 *)s2;
1123 
1124 	if (snap_id1 < snap_id2)
1125 		return 1;
1126 	return snap_id1 == snap_id2 ? 0 : -1;
1127 }
1128 
1129 /*
1130  * Search a snapshot context to see if the given snapshot id is
1131  * present.
1132  *
1133  * Returns the position of the snapshot id in the array if it's found,
1134  * or BAD_SNAP_INDEX otherwise.
1135  *
1136  * Note: The snapshot array is in kept sorted (by the osd) in
1137  * reverse order, highest snapshot id first.
1138  */
1139 static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id)
1140 {
1141 	struct ceph_snap_context *snapc = rbd_dev->header.snapc;
1142 	u64 *found;
1143 
1144 	found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps,
1145 				sizeof (snap_id), snapid_compare_reverse);
1146 
1147 	return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX;
1148 }
1149 
1150 static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev,
1151 					u64 snap_id)
1152 {
1153 	u32 which;
1154 	const char *snap_name;
1155 
1156 	which = rbd_dev_snap_index(rbd_dev, snap_id);
1157 	if (which == BAD_SNAP_INDEX)
1158 		return ERR_PTR(-ENOENT);
1159 
1160 	snap_name = _rbd_dev_v1_snap_name(rbd_dev, which);
1161 	return snap_name ? snap_name : ERR_PTR(-ENOMEM);
1162 }
1163 
1164 static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id)
1165 {
1166 	if (snap_id == CEPH_NOSNAP)
1167 		return RBD_SNAP_HEAD_NAME;
1168 
1169 	rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
1170 	if (rbd_dev->image_format == 1)
1171 		return rbd_dev_v1_snap_name(rbd_dev, snap_id);
1172 
1173 	return rbd_dev_v2_snap_name(rbd_dev, snap_id);
1174 }
1175 
1176 static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
1177 				u64 *snap_size)
1178 {
1179 	rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
1180 	if (snap_id == CEPH_NOSNAP) {
1181 		*snap_size = rbd_dev->header.image_size;
1182 	} else if (rbd_dev->image_format == 1) {
1183 		u32 which;
1184 
1185 		which = rbd_dev_snap_index(rbd_dev, snap_id);
1186 		if (which == BAD_SNAP_INDEX)
1187 			return -ENOENT;
1188 
1189 		*snap_size = rbd_dev->header.snap_sizes[which];
1190 	} else {
1191 		u64 size = 0;
1192 		int ret;
1193 
1194 		ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size);
1195 		if (ret)
1196 			return ret;
1197 
1198 		*snap_size = size;
1199 	}
1200 	return 0;
1201 }
1202 
1203 static int rbd_dev_mapping_set(struct rbd_device *rbd_dev)
1204 {
1205 	u64 snap_id = rbd_dev->spec->snap_id;
1206 	u64 size = 0;
1207 	int ret;
1208 
1209 	ret = rbd_snap_size(rbd_dev, snap_id, &size);
1210 	if (ret)
1211 		return ret;
1212 
1213 	rbd_dev->mapping.size = size;
1214 	return 0;
1215 }
1216 
1217 static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev)
1218 {
1219 	rbd_dev->mapping.size = 0;
1220 }
1221 
1222 static void zero_bios(struct ceph_bio_iter *bio_pos, u32 off, u32 bytes)
1223 {
1224 	struct ceph_bio_iter it = *bio_pos;
1225 
1226 	ceph_bio_iter_advance(&it, off);
1227 	ceph_bio_iter_advance_step(&it, bytes, ({
1228 		memzero_bvec(&bv);
1229 	}));
1230 }
1231 
1232 static void zero_bvecs(struct ceph_bvec_iter *bvec_pos, u32 off, u32 bytes)
1233 {
1234 	struct ceph_bvec_iter it = *bvec_pos;
1235 
1236 	ceph_bvec_iter_advance(&it, off);
1237 	ceph_bvec_iter_advance_step(&it, bytes, ({
1238 		memzero_bvec(&bv);
1239 	}));
1240 }
1241 
1242 /*
1243  * Zero a range in @obj_req data buffer defined by a bio (list) or
1244  * (private) bio_vec array.
1245  *
1246  * @off is relative to the start of the data buffer.
1247  */
1248 static void rbd_obj_zero_range(struct rbd_obj_request *obj_req, u32 off,
1249 			       u32 bytes)
1250 {
1251 	dout("%s %p data buf %u~%u\n", __func__, obj_req, off, bytes);
1252 
1253 	switch (obj_req->img_request->data_type) {
1254 	case OBJ_REQUEST_BIO:
1255 		zero_bios(&obj_req->bio_pos, off, bytes);
1256 		break;
1257 	case OBJ_REQUEST_BVECS:
1258 	case OBJ_REQUEST_OWN_BVECS:
1259 		zero_bvecs(&obj_req->bvec_pos, off, bytes);
1260 		break;
1261 	default:
1262 		BUG();
1263 	}
1264 }
1265 
1266 static void rbd_obj_request_destroy(struct kref *kref);
1267 static void rbd_obj_request_put(struct rbd_obj_request *obj_request)
1268 {
1269 	rbd_assert(obj_request != NULL);
1270 	dout("%s: obj %p (was %d)\n", __func__, obj_request,
1271 		kref_read(&obj_request->kref));
1272 	kref_put(&obj_request->kref, rbd_obj_request_destroy);
1273 }
1274 
1275 static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request,
1276 					struct rbd_obj_request *obj_request)
1277 {
1278 	rbd_assert(obj_request->img_request == NULL);
1279 
1280 	/* Image request now owns object's original reference */
1281 	obj_request->img_request = img_request;
1282 	dout("%s: img %p obj %p\n", __func__, img_request, obj_request);
1283 }
1284 
1285 static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request,
1286 					struct rbd_obj_request *obj_request)
1287 {
1288 	dout("%s: img %p obj %p\n", __func__, img_request, obj_request);
1289 	list_del(&obj_request->ex.oe_item);
1290 	rbd_assert(obj_request->img_request == img_request);
1291 	rbd_obj_request_put(obj_request);
1292 }
1293 
1294 static void rbd_osd_submit(struct ceph_osd_request *osd_req)
1295 {
1296 	struct rbd_obj_request *obj_req = osd_req->r_priv;
1297 
1298 	dout("%s osd_req %p for obj_req %p objno %llu %llu~%llu\n",
1299 	     __func__, osd_req, obj_req, obj_req->ex.oe_objno,
1300 	     obj_req->ex.oe_off, obj_req->ex.oe_len);
1301 	ceph_osdc_start_request(osd_req->r_osdc, osd_req, false);
1302 }
1303 
1304 /*
1305  * The default/initial value for all image request flags is 0.  Each
1306  * is conditionally set to 1 at image request initialization time
1307  * and currently never change thereafter.
1308  */
1309 static void img_request_layered_set(struct rbd_img_request *img_request)
1310 {
1311 	set_bit(IMG_REQ_LAYERED, &img_request->flags);
1312 }
1313 
1314 static bool img_request_layered_test(struct rbd_img_request *img_request)
1315 {
1316 	return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0;
1317 }
1318 
1319 static bool rbd_obj_is_entire(struct rbd_obj_request *obj_req)
1320 {
1321 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
1322 
1323 	return !obj_req->ex.oe_off &&
1324 	       obj_req->ex.oe_len == rbd_dev->layout.object_size;
1325 }
1326 
1327 static bool rbd_obj_is_tail(struct rbd_obj_request *obj_req)
1328 {
1329 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
1330 
1331 	return obj_req->ex.oe_off + obj_req->ex.oe_len ==
1332 					rbd_dev->layout.object_size;
1333 }
1334 
1335 /*
1336  * Must be called after rbd_obj_calc_img_extents().
1337  */
1338 static bool rbd_obj_copyup_enabled(struct rbd_obj_request *obj_req)
1339 {
1340 	if (!obj_req->num_img_extents ||
1341 	    (rbd_obj_is_entire(obj_req) &&
1342 	     !obj_req->img_request->snapc->num_snaps))
1343 		return false;
1344 
1345 	return true;
1346 }
1347 
1348 static u64 rbd_obj_img_extents_bytes(struct rbd_obj_request *obj_req)
1349 {
1350 	return ceph_file_extents_bytes(obj_req->img_extents,
1351 				       obj_req->num_img_extents);
1352 }
1353 
1354 static bool rbd_img_is_write(struct rbd_img_request *img_req)
1355 {
1356 	switch (img_req->op_type) {
1357 	case OBJ_OP_READ:
1358 		return false;
1359 	case OBJ_OP_WRITE:
1360 	case OBJ_OP_DISCARD:
1361 	case OBJ_OP_ZEROOUT:
1362 		return true;
1363 	default:
1364 		BUG();
1365 	}
1366 }
1367 
1368 static void rbd_osd_req_callback(struct ceph_osd_request *osd_req)
1369 {
1370 	struct rbd_obj_request *obj_req = osd_req->r_priv;
1371 	int result;
1372 
1373 	dout("%s osd_req %p result %d for obj_req %p\n", __func__, osd_req,
1374 	     osd_req->r_result, obj_req);
1375 
1376 	/*
1377 	 * Writes aren't allowed to return a data payload.  In some
1378 	 * guarded write cases (e.g. stat + zero on an empty object)
1379 	 * a stat response makes it through, but we don't care.
1380 	 */
1381 	if (osd_req->r_result > 0 && rbd_img_is_write(obj_req->img_request))
1382 		result = 0;
1383 	else
1384 		result = osd_req->r_result;
1385 
1386 	rbd_obj_handle_request(obj_req, result);
1387 }
1388 
1389 static void rbd_osd_format_read(struct ceph_osd_request *osd_req)
1390 {
1391 	struct rbd_obj_request *obj_request = osd_req->r_priv;
1392 	struct rbd_device *rbd_dev = obj_request->img_request->rbd_dev;
1393 	struct ceph_options *opt = rbd_dev->rbd_client->client->options;
1394 
1395 	osd_req->r_flags = CEPH_OSD_FLAG_READ | opt->read_from_replica;
1396 	osd_req->r_snapid = obj_request->img_request->snap_id;
1397 }
1398 
1399 static void rbd_osd_format_write(struct ceph_osd_request *osd_req)
1400 {
1401 	struct rbd_obj_request *obj_request = osd_req->r_priv;
1402 
1403 	osd_req->r_flags = CEPH_OSD_FLAG_WRITE;
1404 	ktime_get_real_ts64(&osd_req->r_mtime);
1405 	osd_req->r_data_offset = obj_request->ex.oe_off;
1406 }
1407 
1408 static struct ceph_osd_request *
1409 __rbd_obj_add_osd_request(struct rbd_obj_request *obj_req,
1410 			  struct ceph_snap_context *snapc, int num_ops)
1411 {
1412 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
1413 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
1414 	struct ceph_osd_request *req;
1415 	const char *name_format = rbd_dev->image_format == 1 ?
1416 				      RBD_V1_DATA_FORMAT : RBD_V2_DATA_FORMAT;
1417 	int ret;
1418 
1419 	req = ceph_osdc_alloc_request(osdc, snapc, num_ops, false, GFP_NOIO);
1420 	if (!req)
1421 		return ERR_PTR(-ENOMEM);
1422 
1423 	list_add_tail(&req->r_private_item, &obj_req->osd_reqs);
1424 	req->r_callback = rbd_osd_req_callback;
1425 	req->r_priv = obj_req;
1426 
1427 	/*
1428 	 * Data objects may be stored in a separate pool, but always in
1429 	 * the same namespace in that pool as the header in its pool.
1430 	 */
1431 	ceph_oloc_copy(&req->r_base_oloc, &rbd_dev->header_oloc);
1432 	req->r_base_oloc.pool = rbd_dev->layout.pool_id;
1433 
1434 	ret = ceph_oid_aprintf(&req->r_base_oid, GFP_NOIO, name_format,
1435 			       rbd_dev->header.object_prefix,
1436 			       obj_req->ex.oe_objno);
1437 	if (ret)
1438 		return ERR_PTR(ret);
1439 
1440 	return req;
1441 }
1442 
1443 static struct ceph_osd_request *
1444 rbd_obj_add_osd_request(struct rbd_obj_request *obj_req, int num_ops)
1445 {
1446 	return __rbd_obj_add_osd_request(obj_req, obj_req->img_request->snapc,
1447 					 num_ops);
1448 }
1449 
1450 static struct rbd_obj_request *rbd_obj_request_create(void)
1451 {
1452 	struct rbd_obj_request *obj_request;
1453 
1454 	obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_NOIO);
1455 	if (!obj_request)
1456 		return NULL;
1457 
1458 	ceph_object_extent_init(&obj_request->ex);
1459 	INIT_LIST_HEAD(&obj_request->osd_reqs);
1460 	mutex_init(&obj_request->state_mutex);
1461 	kref_init(&obj_request->kref);
1462 
1463 	dout("%s %p\n", __func__, obj_request);
1464 	return obj_request;
1465 }
1466 
1467 static void rbd_obj_request_destroy(struct kref *kref)
1468 {
1469 	struct rbd_obj_request *obj_request;
1470 	struct ceph_osd_request *osd_req;
1471 	u32 i;
1472 
1473 	obj_request = container_of(kref, struct rbd_obj_request, kref);
1474 
1475 	dout("%s: obj %p\n", __func__, obj_request);
1476 
1477 	while (!list_empty(&obj_request->osd_reqs)) {
1478 		osd_req = list_first_entry(&obj_request->osd_reqs,
1479 				    struct ceph_osd_request, r_private_item);
1480 		list_del_init(&osd_req->r_private_item);
1481 		ceph_osdc_put_request(osd_req);
1482 	}
1483 
1484 	switch (obj_request->img_request->data_type) {
1485 	case OBJ_REQUEST_NODATA:
1486 	case OBJ_REQUEST_BIO:
1487 	case OBJ_REQUEST_BVECS:
1488 		break;		/* Nothing to do */
1489 	case OBJ_REQUEST_OWN_BVECS:
1490 		kfree(obj_request->bvec_pos.bvecs);
1491 		break;
1492 	default:
1493 		BUG();
1494 	}
1495 
1496 	kfree(obj_request->img_extents);
1497 	if (obj_request->copyup_bvecs) {
1498 		for (i = 0; i < obj_request->copyup_bvec_count; i++) {
1499 			if (obj_request->copyup_bvecs[i].bv_page)
1500 				__free_page(obj_request->copyup_bvecs[i].bv_page);
1501 		}
1502 		kfree(obj_request->copyup_bvecs);
1503 	}
1504 
1505 	kmem_cache_free(rbd_obj_request_cache, obj_request);
1506 }
1507 
1508 /* It's OK to call this for a device with no parent */
1509 
1510 static void rbd_spec_put(struct rbd_spec *spec);
1511 static void rbd_dev_unparent(struct rbd_device *rbd_dev)
1512 {
1513 	rbd_dev_remove_parent(rbd_dev);
1514 	rbd_spec_put(rbd_dev->parent_spec);
1515 	rbd_dev->parent_spec = NULL;
1516 	rbd_dev->parent_overlap = 0;
1517 }
1518 
1519 /*
1520  * Parent image reference counting is used to determine when an
1521  * image's parent fields can be safely torn down--after there are no
1522  * more in-flight requests to the parent image.  When the last
1523  * reference is dropped, cleaning them up is safe.
1524  */
1525 static void rbd_dev_parent_put(struct rbd_device *rbd_dev)
1526 {
1527 	int counter;
1528 
1529 	if (!rbd_dev->parent_spec)
1530 		return;
1531 
1532 	counter = atomic_dec_return_safe(&rbd_dev->parent_ref);
1533 	if (counter > 0)
1534 		return;
1535 
1536 	/* Last reference; clean up parent data structures */
1537 
1538 	if (!counter)
1539 		rbd_dev_unparent(rbd_dev);
1540 	else
1541 		rbd_warn(rbd_dev, "parent reference underflow");
1542 }
1543 
1544 /*
1545  * If an image has a non-zero parent overlap, get a reference to its
1546  * parent.
1547  *
1548  * Returns true if the rbd device has a parent with a non-zero
1549  * overlap and a reference for it was successfully taken, or
1550  * false otherwise.
1551  */
1552 static bool rbd_dev_parent_get(struct rbd_device *rbd_dev)
1553 {
1554 	int counter = 0;
1555 
1556 	if (!rbd_dev->parent_spec)
1557 		return false;
1558 
1559 	if (rbd_dev->parent_overlap)
1560 		counter = atomic_inc_return_safe(&rbd_dev->parent_ref);
1561 
1562 	if (counter < 0)
1563 		rbd_warn(rbd_dev, "parent reference overflow");
1564 
1565 	return counter > 0;
1566 }
1567 
1568 static void rbd_img_request_init(struct rbd_img_request *img_request,
1569 				 struct rbd_device *rbd_dev,
1570 				 enum obj_operation_type op_type)
1571 {
1572 	memset(img_request, 0, sizeof(*img_request));
1573 
1574 	img_request->rbd_dev = rbd_dev;
1575 	img_request->op_type = op_type;
1576 
1577 	INIT_LIST_HEAD(&img_request->lock_item);
1578 	INIT_LIST_HEAD(&img_request->object_extents);
1579 	mutex_init(&img_request->state_mutex);
1580 }
1581 
1582 static void rbd_img_capture_header(struct rbd_img_request *img_req)
1583 {
1584 	struct rbd_device *rbd_dev = img_req->rbd_dev;
1585 
1586 	lockdep_assert_held(&rbd_dev->header_rwsem);
1587 
1588 	if (rbd_img_is_write(img_req))
1589 		img_req->snapc = ceph_get_snap_context(rbd_dev->header.snapc);
1590 	else
1591 		img_req->snap_id = rbd_dev->spec->snap_id;
1592 
1593 	if (rbd_dev_parent_get(rbd_dev))
1594 		img_request_layered_set(img_req);
1595 }
1596 
1597 static void rbd_img_request_destroy(struct rbd_img_request *img_request)
1598 {
1599 	struct rbd_obj_request *obj_request;
1600 	struct rbd_obj_request *next_obj_request;
1601 
1602 	dout("%s: img %p\n", __func__, img_request);
1603 
1604 	WARN_ON(!list_empty(&img_request->lock_item));
1605 	for_each_obj_request_safe(img_request, obj_request, next_obj_request)
1606 		rbd_img_obj_request_del(img_request, obj_request);
1607 
1608 	if (img_request_layered_test(img_request))
1609 		rbd_dev_parent_put(img_request->rbd_dev);
1610 
1611 	if (rbd_img_is_write(img_request))
1612 		ceph_put_snap_context(img_request->snapc);
1613 
1614 	if (test_bit(IMG_REQ_CHILD, &img_request->flags))
1615 		kmem_cache_free(rbd_img_request_cache, img_request);
1616 }
1617 
1618 #define BITS_PER_OBJ	2
1619 #define OBJS_PER_BYTE	(BITS_PER_BYTE / BITS_PER_OBJ)
1620 #define OBJ_MASK	((1 << BITS_PER_OBJ) - 1)
1621 
1622 static void __rbd_object_map_index(struct rbd_device *rbd_dev, u64 objno,
1623 				   u64 *index, u8 *shift)
1624 {
1625 	u32 off;
1626 
1627 	rbd_assert(objno < rbd_dev->object_map_size);
1628 	*index = div_u64_rem(objno, OBJS_PER_BYTE, &off);
1629 	*shift = (OBJS_PER_BYTE - off - 1) * BITS_PER_OBJ;
1630 }
1631 
1632 static u8 __rbd_object_map_get(struct rbd_device *rbd_dev, u64 objno)
1633 {
1634 	u64 index;
1635 	u8 shift;
1636 
1637 	lockdep_assert_held(&rbd_dev->object_map_lock);
1638 	__rbd_object_map_index(rbd_dev, objno, &index, &shift);
1639 	return (rbd_dev->object_map[index] >> shift) & OBJ_MASK;
1640 }
1641 
1642 static void __rbd_object_map_set(struct rbd_device *rbd_dev, u64 objno, u8 val)
1643 {
1644 	u64 index;
1645 	u8 shift;
1646 	u8 *p;
1647 
1648 	lockdep_assert_held(&rbd_dev->object_map_lock);
1649 	rbd_assert(!(val & ~OBJ_MASK));
1650 
1651 	__rbd_object_map_index(rbd_dev, objno, &index, &shift);
1652 	p = &rbd_dev->object_map[index];
1653 	*p = (*p & ~(OBJ_MASK << shift)) | (val << shift);
1654 }
1655 
1656 static u8 rbd_object_map_get(struct rbd_device *rbd_dev, u64 objno)
1657 {
1658 	u8 state;
1659 
1660 	spin_lock(&rbd_dev->object_map_lock);
1661 	state = __rbd_object_map_get(rbd_dev, objno);
1662 	spin_unlock(&rbd_dev->object_map_lock);
1663 	return state;
1664 }
1665 
1666 static bool use_object_map(struct rbd_device *rbd_dev)
1667 {
1668 	/*
1669 	 * An image mapped read-only can't use the object map -- it isn't
1670 	 * loaded because the header lock isn't acquired.  Someone else can
1671 	 * write to the image and update the object map behind our back.
1672 	 *
1673 	 * A snapshot can't be written to, so using the object map is always
1674 	 * safe.
1675 	 */
1676 	if (!rbd_is_snap(rbd_dev) && rbd_is_ro(rbd_dev))
1677 		return false;
1678 
1679 	return ((rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP) &&
1680 		!(rbd_dev->object_map_flags & RBD_FLAG_OBJECT_MAP_INVALID));
1681 }
1682 
1683 static bool rbd_object_map_may_exist(struct rbd_device *rbd_dev, u64 objno)
1684 {
1685 	u8 state;
1686 
1687 	/* fall back to default logic if object map is disabled or invalid */
1688 	if (!use_object_map(rbd_dev))
1689 		return true;
1690 
1691 	state = rbd_object_map_get(rbd_dev, objno);
1692 	return state != OBJECT_NONEXISTENT;
1693 }
1694 
1695 static void rbd_object_map_name(struct rbd_device *rbd_dev, u64 snap_id,
1696 				struct ceph_object_id *oid)
1697 {
1698 	if (snap_id == CEPH_NOSNAP)
1699 		ceph_oid_printf(oid, "%s%s", RBD_OBJECT_MAP_PREFIX,
1700 				rbd_dev->spec->image_id);
1701 	else
1702 		ceph_oid_printf(oid, "%s%s.%016llx", RBD_OBJECT_MAP_PREFIX,
1703 				rbd_dev->spec->image_id, snap_id);
1704 }
1705 
1706 static int rbd_object_map_lock(struct rbd_device *rbd_dev)
1707 {
1708 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
1709 	CEPH_DEFINE_OID_ONSTACK(oid);
1710 	u8 lock_type;
1711 	char *lock_tag;
1712 	struct ceph_locker *lockers;
1713 	u32 num_lockers;
1714 	bool broke_lock = false;
1715 	int ret;
1716 
1717 	rbd_object_map_name(rbd_dev, CEPH_NOSNAP, &oid);
1718 
1719 again:
1720 	ret = ceph_cls_lock(osdc, &oid, &rbd_dev->header_oloc, RBD_LOCK_NAME,
1721 			    CEPH_CLS_LOCK_EXCLUSIVE, "", "", "", 0);
1722 	if (ret != -EBUSY || broke_lock) {
1723 		if (ret == -EEXIST)
1724 			ret = 0; /* already locked by myself */
1725 		if (ret)
1726 			rbd_warn(rbd_dev, "failed to lock object map: %d", ret);
1727 		return ret;
1728 	}
1729 
1730 	ret = ceph_cls_lock_info(osdc, &oid, &rbd_dev->header_oloc,
1731 				 RBD_LOCK_NAME, &lock_type, &lock_tag,
1732 				 &lockers, &num_lockers);
1733 	if (ret) {
1734 		if (ret == -ENOENT)
1735 			goto again;
1736 
1737 		rbd_warn(rbd_dev, "failed to get object map lockers: %d", ret);
1738 		return ret;
1739 	}
1740 
1741 	kfree(lock_tag);
1742 	if (num_lockers == 0)
1743 		goto again;
1744 
1745 	rbd_warn(rbd_dev, "breaking object map lock owned by %s%llu",
1746 		 ENTITY_NAME(lockers[0].id.name));
1747 
1748 	ret = ceph_cls_break_lock(osdc, &oid, &rbd_dev->header_oloc,
1749 				  RBD_LOCK_NAME, lockers[0].id.cookie,
1750 				  &lockers[0].id.name);
1751 	ceph_free_lockers(lockers, num_lockers);
1752 	if (ret) {
1753 		if (ret == -ENOENT)
1754 			goto again;
1755 
1756 		rbd_warn(rbd_dev, "failed to break object map lock: %d", ret);
1757 		return ret;
1758 	}
1759 
1760 	broke_lock = true;
1761 	goto again;
1762 }
1763 
1764 static void rbd_object_map_unlock(struct rbd_device *rbd_dev)
1765 {
1766 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
1767 	CEPH_DEFINE_OID_ONSTACK(oid);
1768 	int ret;
1769 
1770 	rbd_object_map_name(rbd_dev, CEPH_NOSNAP, &oid);
1771 
1772 	ret = ceph_cls_unlock(osdc, &oid, &rbd_dev->header_oloc, RBD_LOCK_NAME,
1773 			      "");
1774 	if (ret && ret != -ENOENT)
1775 		rbd_warn(rbd_dev, "failed to unlock object map: %d", ret);
1776 }
1777 
1778 static int decode_object_map_header(void **p, void *end, u64 *object_map_size)
1779 {
1780 	u8 struct_v;
1781 	u32 struct_len;
1782 	u32 header_len;
1783 	void *header_end;
1784 	int ret;
1785 
1786 	ceph_decode_32_safe(p, end, header_len, e_inval);
1787 	header_end = *p + header_len;
1788 
1789 	ret = ceph_start_decoding(p, end, 1, "BitVector header", &struct_v,
1790 				  &struct_len);
1791 	if (ret)
1792 		return ret;
1793 
1794 	ceph_decode_64_safe(p, end, *object_map_size, e_inval);
1795 
1796 	*p = header_end;
1797 	return 0;
1798 
1799 e_inval:
1800 	return -EINVAL;
1801 }
1802 
1803 static int __rbd_object_map_load(struct rbd_device *rbd_dev)
1804 {
1805 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
1806 	CEPH_DEFINE_OID_ONSTACK(oid);
1807 	struct page **pages;
1808 	void *p, *end;
1809 	size_t reply_len;
1810 	u64 num_objects;
1811 	u64 object_map_bytes;
1812 	u64 object_map_size;
1813 	int num_pages;
1814 	int ret;
1815 
1816 	rbd_assert(!rbd_dev->object_map && !rbd_dev->object_map_size);
1817 
1818 	num_objects = ceph_get_num_objects(&rbd_dev->layout,
1819 					   rbd_dev->mapping.size);
1820 	object_map_bytes = DIV_ROUND_UP_ULL(num_objects * BITS_PER_OBJ,
1821 					    BITS_PER_BYTE);
1822 	num_pages = calc_pages_for(0, object_map_bytes) + 1;
1823 	pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
1824 	if (IS_ERR(pages))
1825 		return PTR_ERR(pages);
1826 
1827 	reply_len = num_pages * PAGE_SIZE;
1828 	rbd_object_map_name(rbd_dev, rbd_dev->spec->snap_id, &oid);
1829 	ret = ceph_osdc_call(osdc, &oid, &rbd_dev->header_oloc,
1830 			     "rbd", "object_map_load", CEPH_OSD_FLAG_READ,
1831 			     NULL, 0, pages, &reply_len);
1832 	if (ret)
1833 		goto out;
1834 
1835 	p = page_address(pages[0]);
1836 	end = p + min(reply_len, (size_t)PAGE_SIZE);
1837 	ret = decode_object_map_header(&p, end, &object_map_size);
1838 	if (ret)
1839 		goto out;
1840 
1841 	if (object_map_size != num_objects) {
1842 		rbd_warn(rbd_dev, "object map size mismatch: %llu vs %llu",
1843 			 object_map_size, num_objects);
1844 		ret = -EINVAL;
1845 		goto out;
1846 	}
1847 
1848 	if (offset_in_page(p) + object_map_bytes > reply_len) {
1849 		ret = -EINVAL;
1850 		goto out;
1851 	}
1852 
1853 	rbd_dev->object_map = kvmalloc(object_map_bytes, GFP_KERNEL);
1854 	if (!rbd_dev->object_map) {
1855 		ret = -ENOMEM;
1856 		goto out;
1857 	}
1858 
1859 	rbd_dev->object_map_size = object_map_size;
1860 	ceph_copy_from_page_vector(pages, rbd_dev->object_map,
1861 				   offset_in_page(p), object_map_bytes);
1862 
1863 out:
1864 	ceph_release_page_vector(pages, num_pages);
1865 	return ret;
1866 }
1867 
1868 static void rbd_object_map_free(struct rbd_device *rbd_dev)
1869 {
1870 	kvfree(rbd_dev->object_map);
1871 	rbd_dev->object_map = NULL;
1872 	rbd_dev->object_map_size = 0;
1873 }
1874 
1875 static int rbd_object_map_load(struct rbd_device *rbd_dev)
1876 {
1877 	int ret;
1878 
1879 	ret = __rbd_object_map_load(rbd_dev);
1880 	if (ret)
1881 		return ret;
1882 
1883 	ret = rbd_dev_v2_get_flags(rbd_dev);
1884 	if (ret) {
1885 		rbd_object_map_free(rbd_dev);
1886 		return ret;
1887 	}
1888 
1889 	if (rbd_dev->object_map_flags & RBD_FLAG_OBJECT_MAP_INVALID)
1890 		rbd_warn(rbd_dev, "object map is invalid");
1891 
1892 	return 0;
1893 }
1894 
1895 static int rbd_object_map_open(struct rbd_device *rbd_dev)
1896 {
1897 	int ret;
1898 
1899 	ret = rbd_object_map_lock(rbd_dev);
1900 	if (ret)
1901 		return ret;
1902 
1903 	ret = rbd_object_map_load(rbd_dev);
1904 	if (ret) {
1905 		rbd_object_map_unlock(rbd_dev);
1906 		return ret;
1907 	}
1908 
1909 	return 0;
1910 }
1911 
1912 static void rbd_object_map_close(struct rbd_device *rbd_dev)
1913 {
1914 	rbd_object_map_free(rbd_dev);
1915 	rbd_object_map_unlock(rbd_dev);
1916 }
1917 
1918 /*
1919  * This function needs snap_id (or more precisely just something to
1920  * distinguish between HEAD and snapshot object maps), new_state and
1921  * current_state that were passed to rbd_object_map_update().
1922  *
1923  * To avoid allocating and stashing a context we piggyback on the OSD
1924  * request.  A HEAD update has two ops (assert_locked).  For new_state
1925  * and current_state we decode our own object_map_update op, encoded in
1926  * rbd_cls_object_map_update().
1927  */
1928 static int rbd_object_map_update_finish(struct rbd_obj_request *obj_req,
1929 					struct ceph_osd_request *osd_req)
1930 {
1931 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
1932 	struct ceph_osd_data *osd_data;
1933 	u64 objno;
1934 	u8 state, new_state, current_state;
1935 	bool has_current_state;
1936 	void *p;
1937 
1938 	if (osd_req->r_result)
1939 		return osd_req->r_result;
1940 
1941 	/*
1942 	 * Nothing to do for a snapshot object map.
1943 	 */
1944 	if (osd_req->r_num_ops == 1)
1945 		return 0;
1946 
1947 	/*
1948 	 * Update in-memory HEAD object map.
1949 	 */
1950 	rbd_assert(osd_req->r_num_ops == 2);
1951 	osd_data = osd_req_op_data(osd_req, 1, cls, request_data);
1952 	rbd_assert(osd_data->type == CEPH_OSD_DATA_TYPE_PAGES);
1953 
1954 	p = page_address(osd_data->pages[0]);
1955 	objno = ceph_decode_64(&p);
1956 	rbd_assert(objno == obj_req->ex.oe_objno);
1957 	rbd_assert(ceph_decode_64(&p) == objno + 1);
1958 	new_state = ceph_decode_8(&p);
1959 	has_current_state = ceph_decode_8(&p);
1960 	if (has_current_state)
1961 		current_state = ceph_decode_8(&p);
1962 
1963 	spin_lock(&rbd_dev->object_map_lock);
1964 	state = __rbd_object_map_get(rbd_dev, objno);
1965 	if (!has_current_state || current_state == state ||
1966 	    (current_state == OBJECT_EXISTS && state == OBJECT_EXISTS_CLEAN))
1967 		__rbd_object_map_set(rbd_dev, objno, new_state);
1968 	spin_unlock(&rbd_dev->object_map_lock);
1969 
1970 	return 0;
1971 }
1972 
1973 static void rbd_object_map_callback(struct ceph_osd_request *osd_req)
1974 {
1975 	struct rbd_obj_request *obj_req = osd_req->r_priv;
1976 	int result;
1977 
1978 	dout("%s osd_req %p result %d for obj_req %p\n", __func__, osd_req,
1979 	     osd_req->r_result, obj_req);
1980 
1981 	result = rbd_object_map_update_finish(obj_req, osd_req);
1982 	rbd_obj_handle_request(obj_req, result);
1983 }
1984 
1985 static bool update_needed(struct rbd_device *rbd_dev, u64 objno, u8 new_state)
1986 {
1987 	u8 state = rbd_object_map_get(rbd_dev, objno);
1988 
1989 	if (state == new_state ||
1990 	    (new_state == OBJECT_PENDING && state == OBJECT_NONEXISTENT) ||
1991 	    (new_state == OBJECT_NONEXISTENT && state != OBJECT_PENDING))
1992 		return false;
1993 
1994 	return true;
1995 }
1996 
1997 static int rbd_cls_object_map_update(struct ceph_osd_request *req,
1998 				     int which, u64 objno, u8 new_state,
1999 				     const u8 *current_state)
2000 {
2001 	struct page **pages;
2002 	void *p, *start;
2003 	int ret;
2004 
2005 	ret = osd_req_op_cls_init(req, which, "rbd", "object_map_update");
2006 	if (ret)
2007 		return ret;
2008 
2009 	pages = ceph_alloc_page_vector(1, GFP_NOIO);
2010 	if (IS_ERR(pages))
2011 		return PTR_ERR(pages);
2012 
2013 	p = start = page_address(pages[0]);
2014 	ceph_encode_64(&p, objno);
2015 	ceph_encode_64(&p, objno + 1);
2016 	ceph_encode_8(&p, new_state);
2017 	if (current_state) {
2018 		ceph_encode_8(&p, 1);
2019 		ceph_encode_8(&p, *current_state);
2020 	} else {
2021 		ceph_encode_8(&p, 0);
2022 	}
2023 
2024 	osd_req_op_cls_request_data_pages(req, which, pages, p - start, 0,
2025 					  false, true);
2026 	return 0;
2027 }
2028 
2029 /*
2030  * Return:
2031  *   0 - object map update sent
2032  *   1 - object map update isn't needed
2033  *  <0 - error
2034  */
2035 static int rbd_object_map_update(struct rbd_obj_request *obj_req, u64 snap_id,
2036 				 u8 new_state, const u8 *current_state)
2037 {
2038 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
2039 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2040 	struct ceph_osd_request *req;
2041 	int num_ops = 1;
2042 	int which = 0;
2043 	int ret;
2044 
2045 	if (snap_id == CEPH_NOSNAP) {
2046 		if (!update_needed(rbd_dev, obj_req->ex.oe_objno, new_state))
2047 			return 1;
2048 
2049 		num_ops++; /* assert_locked */
2050 	}
2051 
2052 	req = ceph_osdc_alloc_request(osdc, NULL, num_ops, false, GFP_NOIO);
2053 	if (!req)
2054 		return -ENOMEM;
2055 
2056 	list_add_tail(&req->r_private_item, &obj_req->osd_reqs);
2057 	req->r_callback = rbd_object_map_callback;
2058 	req->r_priv = obj_req;
2059 
2060 	rbd_object_map_name(rbd_dev, snap_id, &req->r_base_oid);
2061 	ceph_oloc_copy(&req->r_base_oloc, &rbd_dev->header_oloc);
2062 	req->r_flags = CEPH_OSD_FLAG_WRITE;
2063 	ktime_get_real_ts64(&req->r_mtime);
2064 
2065 	if (snap_id == CEPH_NOSNAP) {
2066 		/*
2067 		 * Protect against possible race conditions during lock
2068 		 * ownership transitions.
2069 		 */
2070 		ret = ceph_cls_assert_locked(req, which++, RBD_LOCK_NAME,
2071 					     CEPH_CLS_LOCK_EXCLUSIVE, "", "");
2072 		if (ret)
2073 			return ret;
2074 	}
2075 
2076 	ret = rbd_cls_object_map_update(req, which, obj_req->ex.oe_objno,
2077 					new_state, current_state);
2078 	if (ret)
2079 		return ret;
2080 
2081 	ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
2082 	if (ret)
2083 		return ret;
2084 
2085 	ceph_osdc_start_request(osdc, req, false);
2086 	return 0;
2087 }
2088 
2089 static void prune_extents(struct ceph_file_extent *img_extents,
2090 			  u32 *num_img_extents, u64 overlap)
2091 {
2092 	u32 cnt = *num_img_extents;
2093 
2094 	/* drop extents completely beyond the overlap */
2095 	while (cnt && img_extents[cnt - 1].fe_off >= overlap)
2096 		cnt--;
2097 
2098 	if (cnt) {
2099 		struct ceph_file_extent *ex = &img_extents[cnt - 1];
2100 
2101 		/* trim final overlapping extent */
2102 		if (ex->fe_off + ex->fe_len > overlap)
2103 			ex->fe_len = overlap - ex->fe_off;
2104 	}
2105 
2106 	*num_img_extents = cnt;
2107 }
2108 
2109 /*
2110  * Determine the byte range(s) covered by either just the object extent
2111  * or the entire object in the parent image.
2112  */
2113 static int rbd_obj_calc_img_extents(struct rbd_obj_request *obj_req,
2114 				    bool entire)
2115 {
2116 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
2117 	int ret;
2118 
2119 	if (!rbd_dev->parent_overlap)
2120 		return 0;
2121 
2122 	ret = ceph_extent_to_file(&rbd_dev->layout, obj_req->ex.oe_objno,
2123 				  entire ? 0 : obj_req->ex.oe_off,
2124 				  entire ? rbd_dev->layout.object_size :
2125 							obj_req->ex.oe_len,
2126 				  &obj_req->img_extents,
2127 				  &obj_req->num_img_extents);
2128 	if (ret)
2129 		return ret;
2130 
2131 	prune_extents(obj_req->img_extents, &obj_req->num_img_extents,
2132 		      rbd_dev->parent_overlap);
2133 	return 0;
2134 }
2135 
2136 static void rbd_osd_setup_data(struct ceph_osd_request *osd_req, int which)
2137 {
2138 	struct rbd_obj_request *obj_req = osd_req->r_priv;
2139 
2140 	switch (obj_req->img_request->data_type) {
2141 	case OBJ_REQUEST_BIO:
2142 		osd_req_op_extent_osd_data_bio(osd_req, which,
2143 					       &obj_req->bio_pos,
2144 					       obj_req->ex.oe_len);
2145 		break;
2146 	case OBJ_REQUEST_BVECS:
2147 	case OBJ_REQUEST_OWN_BVECS:
2148 		rbd_assert(obj_req->bvec_pos.iter.bi_size ==
2149 							obj_req->ex.oe_len);
2150 		rbd_assert(obj_req->bvec_idx == obj_req->bvec_count);
2151 		osd_req_op_extent_osd_data_bvec_pos(osd_req, which,
2152 						    &obj_req->bvec_pos);
2153 		break;
2154 	default:
2155 		BUG();
2156 	}
2157 }
2158 
2159 static int rbd_osd_setup_stat(struct ceph_osd_request *osd_req, int which)
2160 {
2161 	struct page **pages;
2162 
2163 	/*
2164 	 * The response data for a STAT call consists of:
2165 	 *     le64 length;
2166 	 *     struct {
2167 	 *         le32 tv_sec;
2168 	 *         le32 tv_nsec;
2169 	 *     } mtime;
2170 	 */
2171 	pages = ceph_alloc_page_vector(1, GFP_NOIO);
2172 	if (IS_ERR(pages))
2173 		return PTR_ERR(pages);
2174 
2175 	osd_req_op_init(osd_req, which, CEPH_OSD_OP_STAT, 0);
2176 	osd_req_op_raw_data_in_pages(osd_req, which, pages,
2177 				     8 + sizeof(struct ceph_timespec),
2178 				     0, false, true);
2179 	return 0;
2180 }
2181 
2182 static int rbd_osd_setup_copyup(struct ceph_osd_request *osd_req, int which,
2183 				u32 bytes)
2184 {
2185 	struct rbd_obj_request *obj_req = osd_req->r_priv;
2186 	int ret;
2187 
2188 	ret = osd_req_op_cls_init(osd_req, which, "rbd", "copyup");
2189 	if (ret)
2190 		return ret;
2191 
2192 	osd_req_op_cls_request_data_bvecs(osd_req, which, obj_req->copyup_bvecs,
2193 					  obj_req->copyup_bvec_count, bytes);
2194 	return 0;
2195 }
2196 
2197 static int rbd_obj_init_read(struct rbd_obj_request *obj_req)
2198 {
2199 	obj_req->read_state = RBD_OBJ_READ_START;
2200 	return 0;
2201 }
2202 
2203 static void __rbd_osd_setup_write_ops(struct ceph_osd_request *osd_req,
2204 				      int which)
2205 {
2206 	struct rbd_obj_request *obj_req = osd_req->r_priv;
2207 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
2208 	u16 opcode;
2209 
2210 	if (!use_object_map(rbd_dev) ||
2211 	    !(obj_req->flags & RBD_OBJ_FLAG_MAY_EXIST)) {
2212 		osd_req_op_alloc_hint_init(osd_req, which++,
2213 					   rbd_dev->layout.object_size,
2214 					   rbd_dev->layout.object_size,
2215 					   rbd_dev->opts->alloc_hint_flags);
2216 	}
2217 
2218 	if (rbd_obj_is_entire(obj_req))
2219 		opcode = CEPH_OSD_OP_WRITEFULL;
2220 	else
2221 		opcode = CEPH_OSD_OP_WRITE;
2222 
2223 	osd_req_op_extent_init(osd_req, which, opcode,
2224 			       obj_req->ex.oe_off, obj_req->ex.oe_len, 0, 0);
2225 	rbd_osd_setup_data(osd_req, which);
2226 }
2227 
2228 static int rbd_obj_init_write(struct rbd_obj_request *obj_req)
2229 {
2230 	int ret;
2231 
2232 	/* reverse map the entire object onto the parent */
2233 	ret = rbd_obj_calc_img_extents(obj_req, true);
2234 	if (ret)
2235 		return ret;
2236 
2237 	if (rbd_obj_copyup_enabled(obj_req))
2238 		obj_req->flags |= RBD_OBJ_FLAG_COPYUP_ENABLED;
2239 
2240 	obj_req->write_state = RBD_OBJ_WRITE_START;
2241 	return 0;
2242 }
2243 
2244 static u16 truncate_or_zero_opcode(struct rbd_obj_request *obj_req)
2245 {
2246 	return rbd_obj_is_tail(obj_req) ? CEPH_OSD_OP_TRUNCATE :
2247 					  CEPH_OSD_OP_ZERO;
2248 }
2249 
2250 static void __rbd_osd_setup_discard_ops(struct ceph_osd_request *osd_req,
2251 					int which)
2252 {
2253 	struct rbd_obj_request *obj_req = osd_req->r_priv;
2254 
2255 	if (rbd_obj_is_entire(obj_req) && !obj_req->num_img_extents) {
2256 		rbd_assert(obj_req->flags & RBD_OBJ_FLAG_DELETION);
2257 		osd_req_op_init(osd_req, which, CEPH_OSD_OP_DELETE, 0);
2258 	} else {
2259 		osd_req_op_extent_init(osd_req, which,
2260 				       truncate_or_zero_opcode(obj_req),
2261 				       obj_req->ex.oe_off, obj_req->ex.oe_len,
2262 				       0, 0);
2263 	}
2264 }
2265 
2266 static int rbd_obj_init_discard(struct rbd_obj_request *obj_req)
2267 {
2268 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
2269 	u64 off, next_off;
2270 	int ret;
2271 
2272 	/*
2273 	 * Align the range to alloc_size boundary and punt on discards
2274 	 * that are too small to free up any space.
2275 	 *
2276 	 * alloc_size == object_size && is_tail() is a special case for
2277 	 * filestore with filestore_punch_hole = false, needed to allow
2278 	 * truncate (in addition to delete).
2279 	 */
2280 	if (rbd_dev->opts->alloc_size != rbd_dev->layout.object_size ||
2281 	    !rbd_obj_is_tail(obj_req)) {
2282 		off = round_up(obj_req->ex.oe_off, rbd_dev->opts->alloc_size);
2283 		next_off = round_down(obj_req->ex.oe_off + obj_req->ex.oe_len,
2284 				      rbd_dev->opts->alloc_size);
2285 		if (off >= next_off)
2286 			return 1;
2287 
2288 		dout("%s %p %llu~%llu -> %llu~%llu\n", __func__,
2289 		     obj_req, obj_req->ex.oe_off, obj_req->ex.oe_len,
2290 		     off, next_off - off);
2291 		obj_req->ex.oe_off = off;
2292 		obj_req->ex.oe_len = next_off - off;
2293 	}
2294 
2295 	/* reverse map the entire object onto the parent */
2296 	ret = rbd_obj_calc_img_extents(obj_req, true);
2297 	if (ret)
2298 		return ret;
2299 
2300 	obj_req->flags |= RBD_OBJ_FLAG_NOOP_FOR_NONEXISTENT;
2301 	if (rbd_obj_is_entire(obj_req) && !obj_req->num_img_extents)
2302 		obj_req->flags |= RBD_OBJ_FLAG_DELETION;
2303 
2304 	obj_req->write_state = RBD_OBJ_WRITE_START;
2305 	return 0;
2306 }
2307 
2308 static void __rbd_osd_setup_zeroout_ops(struct ceph_osd_request *osd_req,
2309 					int which)
2310 {
2311 	struct rbd_obj_request *obj_req = osd_req->r_priv;
2312 	u16 opcode;
2313 
2314 	if (rbd_obj_is_entire(obj_req)) {
2315 		if (obj_req->num_img_extents) {
2316 			if (!(obj_req->flags & RBD_OBJ_FLAG_COPYUP_ENABLED))
2317 				osd_req_op_init(osd_req, which++,
2318 						CEPH_OSD_OP_CREATE, 0);
2319 			opcode = CEPH_OSD_OP_TRUNCATE;
2320 		} else {
2321 			rbd_assert(obj_req->flags & RBD_OBJ_FLAG_DELETION);
2322 			osd_req_op_init(osd_req, which++,
2323 					CEPH_OSD_OP_DELETE, 0);
2324 			opcode = 0;
2325 		}
2326 	} else {
2327 		opcode = truncate_or_zero_opcode(obj_req);
2328 	}
2329 
2330 	if (opcode)
2331 		osd_req_op_extent_init(osd_req, which, opcode,
2332 				       obj_req->ex.oe_off, obj_req->ex.oe_len,
2333 				       0, 0);
2334 }
2335 
2336 static int rbd_obj_init_zeroout(struct rbd_obj_request *obj_req)
2337 {
2338 	int ret;
2339 
2340 	/* reverse map the entire object onto the parent */
2341 	ret = rbd_obj_calc_img_extents(obj_req, true);
2342 	if (ret)
2343 		return ret;
2344 
2345 	if (rbd_obj_copyup_enabled(obj_req))
2346 		obj_req->flags |= RBD_OBJ_FLAG_COPYUP_ENABLED;
2347 	if (!obj_req->num_img_extents) {
2348 		obj_req->flags |= RBD_OBJ_FLAG_NOOP_FOR_NONEXISTENT;
2349 		if (rbd_obj_is_entire(obj_req))
2350 			obj_req->flags |= RBD_OBJ_FLAG_DELETION;
2351 	}
2352 
2353 	obj_req->write_state = RBD_OBJ_WRITE_START;
2354 	return 0;
2355 }
2356 
2357 static int count_write_ops(struct rbd_obj_request *obj_req)
2358 {
2359 	struct rbd_img_request *img_req = obj_req->img_request;
2360 
2361 	switch (img_req->op_type) {
2362 	case OBJ_OP_WRITE:
2363 		if (!use_object_map(img_req->rbd_dev) ||
2364 		    !(obj_req->flags & RBD_OBJ_FLAG_MAY_EXIST))
2365 			return 2; /* setallochint + write/writefull */
2366 
2367 		return 1; /* write/writefull */
2368 	case OBJ_OP_DISCARD:
2369 		return 1; /* delete/truncate/zero */
2370 	case OBJ_OP_ZEROOUT:
2371 		if (rbd_obj_is_entire(obj_req) && obj_req->num_img_extents &&
2372 		    !(obj_req->flags & RBD_OBJ_FLAG_COPYUP_ENABLED))
2373 			return 2; /* create + truncate */
2374 
2375 		return 1; /* delete/truncate/zero */
2376 	default:
2377 		BUG();
2378 	}
2379 }
2380 
2381 static void rbd_osd_setup_write_ops(struct ceph_osd_request *osd_req,
2382 				    int which)
2383 {
2384 	struct rbd_obj_request *obj_req = osd_req->r_priv;
2385 
2386 	switch (obj_req->img_request->op_type) {
2387 	case OBJ_OP_WRITE:
2388 		__rbd_osd_setup_write_ops(osd_req, which);
2389 		break;
2390 	case OBJ_OP_DISCARD:
2391 		__rbd_osd_setup_discard_ops(osd_req, which);
2392 		break;
2393 	case OBJ_OP_ZEROOUT:
2394 		__rbd_osd_setup_zeroout_ops(osd_req, which);
2395 		break;
2396 	default:
2397 		BUG();
2398 	}
2399 }
2400 
2401 /*
2402  * Prune the list of object requests (adjust offset and/or length, drop
2403  * redundant requests).  Prepare object request state machines and image
2404  * request state machine for execution.
2405  */
2406 static int __rbd_img_fill_request(struct rbd_img_request *img_req)
2407 {
2408 	struct rbd_obj_request *obj_req, *next_obj_req;
2409 	int ret;
2410 
2411 	for_each_obj_request_safe(img_req, obj_req, next_obj_req) {
2412 		switch (img_req->op_type) {
2413 		case OBJ_OP_READ:
2414 			ret = rbd_obj_init_read(obj_req);
2415 			break;
2416 		case OBJ_OP_WRITE:
2417 			ret = rbd_obj_init_write(obj_req);
2418 			break;
2419 		case OBJ_OP_DISCARD:
2420 			ret = rbd_obj_init_discard(obj_req);
2421 			break;
2422 		case OBJ_OP_ZEROOUT:
2423 			ret = rbd_obj_init_zeroout(obj_req);
2424 			break;
2425 		default:
2426 			BUG();
2427 		}
2428 		if (ret < 0)
2429 			return ret;
2430 		if (ret > 0) {
2431 			rbd_img_obj_request_del(img_req, obj_req);
2432 			continue;
2433 		}
2434 	}
2435 
2436 	img_req->state = RBD_IMG_START;
2437 	return 0;
2438 }
2439 
2440 union rbd_img_fill_iter {
2441 	struct ceph_bio_iter	bio_iter;
2442 	struct ceph_bvec_iter	bvec_iter;
2443 };
2444 
2445 struct rbd_img_fill_ctx {
2446 	enum obj_request_type	pos_type;
2447 	union rbd_img_fill_iter	*pos;
2448 	union rbd_img_fill_iter	iter;
2449 	ceph_object_extent_fn_t	set_pos_fn;
2450 	ceph_object_extent_fn_t	count_fn;
2451 	ceph_object_extent_fn_t	copy_fn;
2452 };
2453 
2454 static struct ceph_object_extent *alloc_object_extent(void *arg)
2455 {
2456 	struct rbd_img_request *img_req = arg;
2457 	struct rbd_obj_request *obj_req;
2458 
2459 	obj_req = rbd_obj_request_create();
2460 	if (!obj_req)
2461 		return NULL;
2462 
2463 	rbd_img_obj_request_add(img_req, obj_req);
2464 	return &obj_req->ex;
2465 }
2466 
2467 /*
2468  * While su != os && sc == 1 is technically not fancy (it's the same
2469  * layout as su == os && sc == 1), we can't use the nocopy path for it
2470  * because ->set_pos_fn() should be called only once per object.
2471  * ceph_file_to_extents() invokes action_fn once per stripe unit, so
2472  * treat su != os && sc == 1 as fancy.
2473  */
2474 static bool rbd_layout_is_fancy(struct ceph_file_layout *l)
2475 {
2476 	return l->stripe_unit != l->object_size;
2477 }
2478 
2479 static int rbd_img_fill_request_nocopy(struct rbd_img_request *img_req,
2480 				       struct ceph_file_extent *img_extents,
2481 				       u32 num_img_extents,
2482 				       struct rbd_img_fill_ctx *fctx)
2483 {
2484 	u32 i;
2485 	int ret;
2486 
2487 	img_req->data_type = fctx->pos_type;
2488 
2489 	/*
2490 	 * Create object requests and set each object request's starting
2491 	 * position in the provided bio (list) or bio_vec array.
2492 	 */
2493 	fctx->iter = *fctx->pos;
2494 	for (i = 0; i < num_img_extents; i++) {
2495 		ret = ceph_file_to_extents(&img_req->rbd_dev->layout,
2496 					   img_extents[i].fe_off,
2497 					   img_extents[i].fe_len,
2498 					   &img_req->object_extents,
2499 					   alloc_object_extent, img_req,
2500 					   fctx->set_pos_fn, &fctx->iter);
2501 		if (ret)
2502 			return ret;
2503 	}
2504 
2505 	return __rbd_img_fill_request(img_req);
2506 }
2507 
2508 /*
2509  * Map a list of image extents to a list of object extents, create the
2510  * corresponding object requests (normally each to a different object,
2511  * but not always) and add them to @img_req.  For each object request,
2512  * set up its data descriptor to point to the corresponding chunk(s) of
2513  * @fctx->pos data buffer.
2514  *
2515  * Because ceph_file_to_extents() will merge adjacent object extents
2516  * together, each object request's data descriptor may point to multiple
2517  * different chunks of @fctx->pos data buffer.
2518  *
2519  * @fctx->pos data buffer is assumed to be large enough.
2520  */
2521 static int rbd_img_fill_request(struct rbd_img_request *img_req,
2522 				struct ceph_file_extent *img_extents,
2523 				u32 num_img_extents,
2524 				struct rbd_img_fill_ctx *fctx)
2525 {
2526 	struct rbd_device *rbd_dev = img_req->rbd_dev;
2527 	struct rbd_obj_request *obj_req;
2528 	u32 i;
2529 	int ret;
2530 
2531 	if (fctx->pos_type == OBJ_REQUEST_NODATA ||
2532 	    !rbd_layout_is_fancy(&rbd_dev->layout))
2533 		return rbd_img_fill_request_nocopy(img_req, img_extents,
2534 						   num_img_extents, fctx);
2535 
2536 	img_req->data_type = OBJ_REQUEST_OWN_BVECS;
2537 
2538 	/*
2539 	 * Create object requests and determine ->bvec_count for each object
2540 	 * request.  Note that ->bvec_count sum over all object requests may
2541 	 * be greater than the number of bio_vecs in the provided bio (list)
2542 	 * or bio_vec array because when mapped, those bio_vecs can straddle
2543 	 * stripe unit boundaries.
2544 	 */
2545 	fctx->iter = *fctx->pos;
2546 	for (i = 0; i < num_img_extents; i++) {
2547 		ret = ceph_file_to_extents(&rbd_dev->layout,
2548 					   img_extents[i].fe_off,
2549 					   img_extents[i].fe_len,
2550 					   &img_req->object_extents,
2551 					   alloc_object_extent, img_req,
2552 					   fctx->count_fn, &fctx->iter);
2553 		if (ret)
2554 			return ret;
2555 	}
2556 
2557 	for_each_obj_request(img_req, obj_req) {
2558 		obj_req->bvec_pos.bvecs = kmalloc_array(obj_req->bvec_count,
2559 					      sizeof(*obj_req->bvec_pos.bvecs),
2560 					      GFP_NOIO);
2561 		if (!obj_req->bvec_pos.bvecs)
2562 			return -ENOMEM;
2563 	}
2564 
2565 	/*
2566 	 * Fill in each object request's private bio_vec array, splitting and
2567 	 * rearranging the provided bio_vecs in stripe unit chunks as needed.
2568 	 */
2569 	fctx->iter = *fctx->pos;
2570 	for (i = 0; i < num_img_extents; i++) {
2571 		ret = ceph_iterate_extents(&rbd_dev->layout,
2572 					   img_extents[i].fe_off,
2573 					   img_extents[i].fe_len,
2574 					   &img_req->object_extents,
2575 					   fctx->copy_fn, &fctx->iter);
2576 		if (ret)
2577 			return ret;
2578 	}
2579 
2580 	return __rbd_img_fill_request(img_req);
2581 }
2582 
2583 static int rbd_img_fill_nodata(struct rbd_img_request *img_req,
2584 			       u64 off, u64 len)
2585 {
2586 	struct ceph_file_extent ex = { off, len };
2587 	union rbd_img_fill_iter dummy = {};
2588 	struct rbd_img_fill_ctx fctx = {
2589 		.pos_type = OBJ_REQUEST_NODATA,
2590 		.pos = &dummy,
2591 	};
2592 
2593 	return rbd_img_fill_request(img_req, &ex, 1, &fctx);
2594 }
2595 
2596 static void set_bio_pos(struct ceph_object_extent *ex, u32 bytes, void *arg)
2597 {
2598 	struct rbd_obj_request *obj_req =
2599 	    container_of(ex, struct rbd_obj_request, ex);
2600 	struct ceph_bio_iter *it = arg;
2601 
2602 	dout("%s objno %llu bytes %u\n", __func__, ex->oe_objno, bytes);
2603 	obj_req->bio_pos = *it;
2604 	ceph_bio_iter_advance(it, bytes);
2605 }
2606 
2607 static void count_bio_bvecs(struct ceph_object_extent *ex, u32 bytes, void *arg)
2608 {
2609 	struct rbd_obj_request *obj_req =
2610 	    container_of(ex, struct rbd_obj_request, ex);
2611 	struct ceph_bio_iter *it = arg;
2612 
2613 	dout("%s objno %llu bytes %u\n", __func__, ex->oe_objno, bytes);
2614 	ceph_bio_iter_advance_step(it, bytes, ({
2615 		obj_req->bvec_count++;
2616 	}));
2617 
2618 }
2619 
2620 static void copy_bio_bvecs(struct ceph_object_extent *ex, u32 bytes, void *arg)
2621 {
2622 	struct rbd_obj_request *obj_req =
2623 	    container_of(ex, struct rbd_obj_request, ex);
2624 	struct ceph_bio_iter *it = arg;
2625 
2626 	dout("%s objno %llu bytes %u\n", __func__, ex->oe_objno, bytes);
2627 	ceph_bio_iter_advance_step(it, bytes, ({
2628 		obj_req->bvec_pos.bvecs[obj_req->bvec_idx++] = bv;
2629 		obj_req->bvec_pos.iter.bi_size += bv.bv_len;
2630 	}));
2631 }
2632 
2633 static int __rbd_img_fill_from_bio(struct rbd_img_request *img_req,
2634 				   struct ceph_file_extent *img_extents,
2635 				   u32 num_img_extents,
2636 				   struct ceph_bio_iter *bio_pos)
2637 {
2638 	struct rbd_img_fill_ctx fctx = {
2639 		.pos_type = OBJ_REQUEST_BIO,
2640 		.pos = (union rbd_img_fill_iter *)bio_pos,
2641 		.set_pos_fn = set_bio_pos,
2642 		.count_fn = count_bio_bvecs,
2643 		.copy_fn = copy_bio_bvecs,
2644 	};
2645 
2646 	return rbd_img_fill_request(img_req, img_extents, num_img_extents,
2647 				    &fctx);
2648 }
2649 
2650 static int rbd_img_fill_from_bio(struct rbd_img_request *img_req,
2651 				 u64 off, u64 len, struct bio *bio)
2652 {
2653 	struct ceph_file_extent ex = { off, len };
2654 	struct ceph_bio_iter it = { .bio = bio, .iter = bio->bi_iter };
2655 
2656 	return __rbd_img_fill_from_bio(img_req, &ex, 1, &it);
2657 }
2658 
2659 static void set_bvec_pos(struct ceph_object_extent *ex, u32 bytes, void *arg)
2660 {
2661 	struct rbd_obj_request *obj_req =
2662 	    container_of(ex, struct rbd_obj_request, ex);
2663 	struct ceph_bvec_iter *it = arg;
2664 
2665 	obj_req->bvec_pos = *it;
2666 	ceph_bvec_iter_shorten(&obj_req->bvec_pos, bytes);
2667 	ceph_bvec_iter_advance(it, bytes);
2668 }
2669 
2670 static void count_bvecs(struct ceph_object_extent *ex, u32 bytes, void *arg)
2671 {
2672 	struct rbd_obj_request *obj_req =
2673 	    container_of(ex, struct rbd_obj_request, ex);
2674 	struct ceph_bvec_iter *it = arg;
2675 
2676 	ceph_bvec_iter_advance_step(it, bytes, ({
2677 		obj_req->bvec_count++;
2678 	}));
2679 }
2680 
2681 static void copy_bvecs(struct ceph_object_extent *ex, u32 bytes, void *arg)
2682 {
2683 	struct rbd_obj_request *obj_req =
2684 	    container_of(ex, struct rbd_obj_request, ex);
2685 	struct ceph_bvec_iter *it = arg;
2686 
2687 	ceph_bvec_iter_advance_step(it, bytes, ({
2688 		obj_req->bvec_pos.bvecs[obj_req->bvec_idx++] = bv;
2689 		obj_req->bvec_pos.iter.bi_size += bv.bv_len;
2690 	}));
2691 }
2692 
2693 static int __rbd_img_fill_from_bvecs(struct rbd_img_request *img_req,
2694 				     struct ceph_file_extent *img_extents,
2695 				     u32 num_img_extents,
2696 				     struct ceph_bvec_iter *bvec_pos)
2697 {
2698 	struct rbd_img_fill_ctx fctx = {
2699 		.pos_type = OBJ_REQUEST_BVECS,
2700 		.pos = (union rbd_img_fill_iter *)bvec_pos,
2701 		.set_pos_fn = set_bvec_pos,
2702 		.count_fn = count_bvecs,
2703 		.copy_fn = copy_bvecs,
2704 	};
2705 
2706 	return rbd_img_fill_request(img_req, img_extents, num_img_extents,
2707 				    &fctx);
2708 }
2709 
2710 static int rbd_img_fill_from_bvecs(struct rbd_img_request *img_req,
2711 				   struct ceph_file_extent *img_extents,
2712 				   u32 num_img_extents,
2713 				   struct bio_vec *bvecs)
2714 {
2715 	struct ceph_bvec_iter it = {
2716 		.bvecs = bvecs,
2717 		.iter = { .bi_size = ceph_file_extents_bytes(img_extents,
2718 							     num_img_extents) },
2719 	};
2720 
2721 	return __rbd_img_fill_from_bvecs(img_req, img_extents, num_img_extents,
2722 					 &it);
2723 }
2724 
2725 static void rbd_img_handle_request_work(struct work_struct *work)
2726 {
2727 	struct rbd_img_request *img_req =
2728 	    container_of(work, struct rbd_img_request, work);
2729 
2730 	rbd_img_handle_request(img_req, img_req->work_result);
2731 }
2732 
2733 static void rbd_img_schedule(struct rbd_img_request *img_req, int result)
2734 {
2735 	INIT_WORK(&img_req->work, rbd_img_handle_request_work);
2736 	img_req->work_result = result;
2737 	queue_work(rbd_wq, &img_req->work);
2738 }
2739 
2740 static bool rbd_obj_may_exist(struct rbd_obj_request *obj_req)
2741 {
2742 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
2743 
2744 	if (rbd_object_map_may_exist(rbd_dev, obj_req->ex.oe_objno)) {
2745 		obj_req->flags |= RBD_OBJ_FLAG_MAY_EXIST;
2746 		return true;
2747 	}
2748 
2749 	dout("%s %p objno %llu assuming dne\n", __func__, obj_req,
2750 	     obj_req->ex.oe_objno);
2751 	return false;
2752 }
2753 
2754 static int rbd_obj_read_object(struct rbd_obj_request *obj_req)
2755 {
2756 	struct ceph_osd_request *osd_req;
2757 	int ret;
2758 
2759 	osd_req = __rbd_obj_add_osd_request(obj_req, NULL, 1);
2760 	if (IS_ERR(osd_req))
2761 		return PTR_ERR(osd_req);
2762 
2763 	osd_req_op_extent_init(osd_req, 0, CEPH_OSD_OP_READ,
2764 			       obj_req->ex.oe_off, obj_req->ex.oe_len, 0, 0);
2765 	rbd_osd_setup_data(osd_req, 0);
2766 	rbd_osd_format_read(osd_req);
2767 
2768 	ret = ceph_osdc_alloc_messages(osd_req, GFP_NOIO);
2769 	if (ret)
2770 		return ret;
2771 
2772 	rbd_osd_submit(osd_req);
2773 	return 0;
2774 }
2775 
2776 static int rbd_obj_read_from_parent(struct rbd_obj_request *obj_req)
2777 {
2778 	struct rbd_img_request *img_req = obj_req->img_request;
2779 	struct rbd_device *parent = img_req->rbd_dev->parent;
2780 	struct rbd_img_request *child_img_req;
2781 	int ret;
2782 
2783 	child_img_req = kmem_cache_alloc(rbd_img_request_cache, GFP_NOIO);
2784 	if (!child_img_req)
2785 		return -ENOMEM;
2786 
2787 	rbd_img_request_init(child_img_req, parent, OBJ_OP_READ);
2788 	__set_bit(IMG_REQ_CHILD, &child_img_req->flags);
2789 	child_img_req->obj_request = obj_req;
2790 
2791 	down_read(&parent->header_rwsem);
2792 	rbd_img_capture_header(child_img_req);
2793 	up_read(&parent->header_rwsem);
2794 
2795 	dout("%s child_img_req %p for obj_req %p\n", __func__, child_img_req,
2796 	     obj_req);
2797 
2798 	if (!rbd_img_is_write(img_req)) {
2799 		switch (img_req->data_type) {
2800 		case OBJ_REQUEST_BIO:
2801 			ret = __rbd_img_fill_from_bio(child_img_req,
2802 						      obj_req->img_extents,
2803 						      obj_req->num_img_extents,
2804 						      &obj_req->bio_pos);
2805 			break;
2806 		case OBJ_REQUEST_BVECS:
2807 		case OBJ_REQUEST_OWN_BVECS:
2808 			ret = __rbd_img_fill_from_bvecs(child_img_req,
2809 						      obj_req->img_extents,
2810 						      obj_req->num_img_extents,
2811 						      &obj_req->bvec_pos);
2812 			break;
2813 		default:
2814 			BUG();
2815 		}
2816 	} else {
2817 		ret = rbd_img_fill_from_bvecs(child_img_req,
2818 					      obj_req->img_extents,
2819 					      obj_req->num_img_extents,
2820 					      obj_req->copyup_bvecs);
2821 	}
2822 	if (ret) {
2823 		rbd_img_request_destroy(child_img_req);
2824 		return ret;
2825 	}
2826 
2827 	/* avoid parent chain recursion */
2828 	rbd_img_schedule(child_img_req, 0);
2829 	return 0;
2830 }
2831 
2832 static bool rbd_obj_advance_read(struct rbd_obj_request *obj_req, int *result)
2833 {
2834 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
2835 	int ret;
2836 
2837 again:
2838 	switch (obj_req->read_state) {
2839 	case RBD_OBJ_READ_START:
2840 		rbd_assert(!*result);
2841 
2842 		if (!rbd_obj_may_exist(obj_req)) {
2843 			*result = -ENOENT;
2844 			obj_req->read_state = RBD_OBJ_READ_OBJECT;
2845 			goto again;
2846 		}
2847 
2848 		ret = rbd_obj_read_object(obj_req);
2849 		if (ret) {
2850 			*result = ret;
2851 			return true;
2852 		}
2853 		obj_req->read_state = RBD_OBJ_READ_OBJECT;
2854 		return false;
2855 	case RBD_OBJ_READ_OBJECT:
2856 		if (*result == -ENOENT && rbd_dev->parent_overlap) {
2857 			/* reverse map this object extent onto the parent */
2858 			ret = rbd_obj_calc_img_extents(obj_req, false);
2859 			if (ret) {
2860 				*result = ret;
2861 				return true;
2862 			}
2863 			if (obj_req->num_img_extents) {
2864 				ret = rbd_obj_read_from_parent(obj_req);
2865 				if (ret) {
2866 					*result = ret;
2867 					return true;
2868 				}
2869 				obj_req->read_state = RBD_OBJ_READ_PARENT;
2870 				return false;
2871 			}
2872 		}
2873 
2874 		/*
2875 		 * -ENOENT means a hole in the image -- zero-fill the entire
2876 		 * length of the request.  A short read also implies zero-fill
2877 		 * to the end of the request.
2878 		 */
2879 		if (*result == -ENOENT) {
2880 			rbd_obj_zero_range(obj_req, 0, obj_req->ex.oe_len);
2881 			*result = 0;
2882 		} else if (*result >= 0) {
2883 			if (*result < obj_req->ex.oe_len)
2884 				rbd_obj_zero_range(obj_req, *result,
2885 						obj_req->ex.oe_len - *result);
2886 			else
2887 				rbd_assert(*result == obj_req->ex.oe_len);
2888 			*result = 0;
2889 		}
2890 		return true;
2891 	case RBD_OBJ_READ_PARENT:
2892 		/*
2893 		 * The parent image is read only up to the overlap -- zero-fill
2894 		 * from the overlap to the end of the request.
2895 		 */
2896 		if (!*result) {
2897 			u32 obj_overlap = rbd_obj_img_extents_bytes(obj_req);
2898 
2899 			if (obj_overlap < obj_req->ex.oe_len)
2900 				rbd_obj_zero_range(obj_req, obj_overlap,
2901 					    obj_req->ex.oe_len - obj_overlap);
2902 		}
2903 		return true;
2904 	default:
2905 		BUG();
2906 	}
2907 }
2908 
2909 static bool rbd_obj_write_is_noop(struct rbd_obj_request *obj_req)
2910 {
2911 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
2912 
2913 	if (rbd_object_map_may_exist(rbd_dev, obj_req->ex.oe_objno))
2914 		obj_req->flags |= RBD_OBJ_FLAG_MAY_EXIST;
2915 
2916 	if (!(obj_req->flags & RBD_OBJ_FLAG_MAY_EXIST) &&
2917 	    (obj_req->flags & RBD_OBJ_FLAG_NOOP_FOR_NONEXISTENT)) {
2918 		dout("%s %p noop for nonexistent\n", __func__, obj_req);
2919 		return true;
2920 	}
2921 
2922 	return false;
2923 }
2924 
2925 /*
2926  * Return:
2927  *   0 - object map update sent
2928  *   1 - object map update isn't needed
2929  *  <0 - error
2930  */
2931 static int rbd_obj_write_pre_object_map(struct rbd_obj_request *obj_req)
2932 {
2933 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
2934 	u8 new_state;
2935 
2936 	if (!(rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP))
2937 		return 1;
2938 
2939 	if (obj_req->flags & RBD_OBJ_FLAG_DELETION)
2940 		new_state = OBJECT_PENDING;
2941 	else
2942 		new_state = OBJECT_EXISTS;
2943 
2944 	return rbd_object_map_update(obj_req, CEPH_NOSNAP, new_state, NULL);
2945 }
2946 
2947 static int rbd_obj_write_object(struct rbd_obj_request *obj_req)
2948 {
2949 	struct ceph_osd_request *osd_req;
2950 	int num_ops = count_write_ops(obj_req);
2951 	int which = 0;
2952 	int ret;
2953 
2954 	if (obj_req->flags & RBD_OBJ_FLAG_COPYUP_ENABLED)
2955 		num_ops++; /* stat */
2956 
2957 	osd_req = rbd_obj_add_osd_request(obj_req, num_ops);
2958 	if (IS_ERR(osd_req))
2959 		return PTR_ERR(osd_req);
2960 
2961 	if (obj_req->flags & RBD_OBJ_FLAG_COPYUP_ENABLED) {
2962 		ret = rbd_osd_setup_stat(osd_req, which++);
2963 		if (ret)
2964 			return ret;
2965 	}
2966 
2967 	rbd_osd_setup_write_ops(osd_req, which);
2968 	rbd_osd_format_write(osd_req);
2969 
2970 	ret = ceph_osdc_alloc_messages(osd_req, GFP_NOIO);
2971 	if (ret)
2972 		return ret;
2973 
2974 	rbd_osd_submit(osd_req);
2975 	return 0;
2976 }
2977 
2978 /*
2979  * copyup_bvecs pages are never highmem pages
2980  */
2981 static bool is_zero_bvecs(struct bio_vec *bvecs, u32 bytes)
2982 {
2983 	struct ceph_bvec_iter it = {
2984 		.bvecs = bvecs,
2985 		.iter = { .bi_size = bytes },
2986 	};
2987 
2988 	ceph_bvec_iter_advance_step(&it, bytes, ({
2989 		if (memchr_inv(bvec_virt(&bv), 0, bv.bv_len))
2990 			return false;
2991 	}));
2992 	return true;
2993 }
2994 
2995 #define MODS_ONLY	U32_MAX
2996 
2997 static int rbd_obj_copyup_empty_snapc(struct rbd_obj_request *obj_req,
2998 				      u32 bytes)
2999 {
3000 	struct ceph_osd_request *osd_req;
3001 	int ret;
3002 
3003 	dout("%s obj_req %p bytes %u\n", __func__, obj_req, bytes);
3004 	rbd_assert(bytes > 0 && bytes != MODS_ONLY);
3005 
3006 	osd_req = __rbd_obj_add_osd_request(obj_req, &rbd_empty_snapc, 1);
3007 	if (IS_ERR(osd_req))
3008 		return PTR_ERR(osd_req);
3009 
3010 	ret = rbd_osd_setup_copyup(osd_req, 0, bytes);
3011 	if (ret)
3012 		return ret;
3013 
3014 	rbd_osd_format_write(osd_req);
3015 
3016 	ret = ceph_osdc_alloc_messages(osd_req, GFP_NOIO);
3017 	if (ret)
3018 		return ret;
3019 
3020 	rbd_osd_submit(osd_req);
3021 	return 0;
3022 }
3023 
3024 static int rbd_obj_copyup_current_snapc(struct rbd_obj_request *obj_req,
3025 					u32 bytes)
3026 {
3027 	struct ceph_osd_request *osd_req;
3028 	int num_ops = count_write_ops(obj_req);
3029 	int which = 0;
3030 	int ret;
3031 
3032 	dout("%s obj_req %p bytes %u\n", __func__, obj_req, bytes);
3033 
3034 	if (bytes != MODS_ONLY)
3035 		num_ops++; /* copyup */
3036 
3037 	osd_req = rbd_obj_add_osd_request(obj_req, num_ops);
3038 	if (IS_ERR(osd_req))
3039 		return PTR_ERR(osd_req);
3040 
3041 	if (bytes != MODS_ONLY) {
3042 		ret = rbd_osd_setup_copyup(osd_req, which++, bytes);
3043 		if (ret)
3044 			return ret;
3045 	}
3046 
3047 	rbd_osd_setup_write_ops(osd_req, which);
3048 	rbd_osd_format_write(osd_req);
3049 
3050 	ret = ceph_osdc_alloc_messages(osd_req, GFP_NOIO);
3051 	if (ret)
3052 		return ret;
3053 
3054 	rbd_osd_submit(osd_req);
3055 	return 0;
3056 }
3057 
3058 static int setup_copyup_bvecs(struct rbd_obj_request *obj_req, u64 obj_overlap)
3059 {
3060 	u32 i;
3061 
3062 	rbd_assert(!obj_req->copyup_bvecs);
3063 	obj_req->copyup_bvec_count = calc_pages_for(0, obj_overlap);
3064 	obj_req->copyup_bvecs = kcalloc(obj_req->copyup_bvec_count,
3065 					sizeof(*obj_req->copyup_bvecs),
3066 					GFP_NOIO);
3067 	if (!obj_req->copyup_bvecs)
3068 		return -ENOMEM;
3069 
3070 	for (i = 0; i < obj_req->copyup_bvec_count; i++) {
3071 		unsigned int len = min(obj_overlap, (u64)PAGE_SIZE);
3072 
3073 		obj_req->copyup_bvecs[i].bv_page = alloc_page(GFP_NOIO);
3074 		if (!obj_req->copyup_bvecs[i].bv_page)
3075 			return -ENOMEM;
3076 
3077 		obj_req->copyup_bvecs[i].bv_offset = 0;
3078 		obj_req->copyup_bvecs[i].bv_len = len;
3079 		obj_overlap -= len;
3080 	}
3081 
3082 	rbd_assert(!obj_overlap);
3083 	return 0;
3084 }
3085 
3086 /*
3087  * The target object doesn't exist.  Read the data for the entire
3088  * target object up to the overlap point (if any) from the parent,
3089  * so we can use it for a copyup.
3090  */
3091 static int rbd_obj_copyup_read_parent(struct rbd_obj_request *obj_req)
3092 {
3093 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
3094 	int ret;
3095 
3096 	rbd_assert(obj_req->num_img_extents);
3097 	prune_extents(obj_req->img_extents, &obj_req->num_img_extents,
3098 		      rbd_dev->parent_overlap);
3099 	if (!obj_req->num_img_extents) {
3100 		/*
3101 		 * The overlap has become 0 (most likely because the
3102 		 * image has been flattened).  Re-submit the original write
3103 		 * request -- pass MODS_ONLY since the copyup isn't needed
3104 		 * anymore.
3105 		 */
3106 		return rbd_obj_copyup_current_snapc(obj_req, MODS_ONLY);
3107 	}
3108 
3109 	ret = setup_copyup_bvecs(obj_req, rbd_obj_img_extents_bytes(obj_req));
3110 	if (ret)
3111 		return ret;
3112 
3113 	return rbd_obj_read_from_parent(obj_req);
3114 }
3115 
3116 static void rbd_obj_copyup_object_maps(struct rbd_obj_request *obj_req)
3117 {
3118 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
3119 	struct ceph_snap_context *snapc = obj_req->img_request->snapc;
3120 	u8 new_state;
3121 	u32 i;
3122 	int ret;
3123 
3124 	rbd_assert(!obj_req->pending.result && !obj_req->pending.num_pending);
3125 
3126 	if (!(rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP))
3127 		return;
3128 
3129 	if (obj_req->flags & RBD_OBJ_FLAG_COPYUP_ZEROS)
3130 		return;
3131 
3132 	for (i = 0; i < snapc->num_snaps; i++) {
3133 		if ((rbd_dev->header.features & RBD_FEATURE_FAST_DIFF) &&
3134 		    i + 1 < snapc->num_snaps)
3135 			new_state = OBJECT_EXISTS_CLEAN;
3136 		else
3137 			new_state = OBJECT_EXISTS;
3138 
3139 		ret = rbd_object_map_update(obj_req, snapc->snaps[i],
3140 					    new_state, NULL);
3141 		if (ret < 0) {
3142 			obj_req->pending.result = ret;
3143 			return;
3144 		}
3145 
3146 		rbd_assert(!ret);
3147 		obj_req->pending.num_pending++;
3148 	}
3149 }
3150 
3151 static void rbd_obj_copyup_write_object(struct rbd_obj_request *obj_req)
3152 {
3153 	u32 bytes = rbd_obj_img_extents_bytes(obj_req);
3154 	int ret;
3155 
3156 	rbd_assert(!obj_req->pending.result && !obj_req->pending.num_pending);
3157 
3158 	/*
3159 	 * Only send non-zero copyup data to save some I/O and network
3160 	 * bandwidth -- zero copyup data is equivalent to the object not
3161 	 * existing.
3162 	 */
3163 	if (obj_req->flags & RBD_OBJ_FLAG_COPYUP_ZEROS)
3164 		bytes = 0;
3165 
3166 	if (obj_req->img_request->snapc->num_snaps && bytes > 0) {
3167 		/*
3168 		 * Send a copyup request with an empty snapshot context to
3169 		 * deep-copyup the object through all existing snapshots.
3170 		 * A second request with the current snapshot context will be
3171 		 * sent for the actual modification.
3172 		 */
3173 		ret = rbd_obj_copyup_empty_snapc(obj_req, bytes);
3174 		if (ret) {
3175 			obj_req->pending.result = ret;
3176 			return;
3177 		}
3178 
3179 		obj_req->pending.num_pending++;
3180 		bytes = MODS_ONLY;
3181 	}
3182 
3183 	ret = rbd_obj_copyup_current_snapc(obj_req, bytes);
3184 	if (ret) {
3185 		obj_req->pending.result = ret;
3186 		return;
3187 	}
3188 
3189 	obj_req->pending.num_pending++;
3190 }
3191 
3192 static bool rbd_obj_advance_copyup(struct rbd_obj_request *obj_req, int *result)
3193 {
3194 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
3195 	int ret;
3196 
3197 again:
3198 	switch (obj_req->copyup_state) {
3199 	case RBD_OBJ_COPYUP_START:
3200 		rbd_assert(!*result);
3201 
3202 		ret = rbd_obj_copyup_read_parent(obj_req);
3203 		if (ret) {
3204 			*result = ret;
3205 			return true;
3206 		}
3207 		if (obj_req->num_img_extents)
3208 			obj_req->copyup_state = RBD_OBJ_COPYUP_READ_PARENT;
3209 		else
3210 			obj_req->copyup_state = RBD_OBJ_COPYUP_WRITE_OBJECT;
3211 		return false;
3212 	case RBD_OBJ_COPYUP_READ_PARENT:
3213 		if (*result)
3214 			return true;
3215 
3216 		if (is_zero_bvecs(obj_req->copyup_bvecs,
3217 				  rbd_obj_img_extents_bytes(obj_req))) {
3218 			dout("%s %p detected zeros\n", __func__, obj_req);
3219 			obj_req->flags |= RBD_OBJ_FLAG_COPYUP_ZEROS;
3220 		}
3221 
3222 		rbd_obj_copyup_object_maps(obj_req);
3223 		if (!obj_req->pending.num_pending) {
3224 			*result = obj_req->pending.result;
3225 			obj_req->copyup_state = RBD_OBJ_COPYUP_OBJECT_MAPS;
3226 			goto again;
3227 		}
3228 		obj_req->copyup_state = __RBD_OBJ_COPYUP_OBJECT_MAPS;
3229 		return false;
3230 	case __RBD_OBJ_COPYUP_OBJECT_MAPS:
3231 		if (!pending_result_dec(&obj_req->pending, result))
3232 			return false;
3233 		fallthrough;
3234 	case RBD_OBJ_COPYUP_OBJECT_MAPS:
3235 		if (*result) {
3236 			rbd_warn(rbd_dev, "snap object map update failed: %d",
3237 				 *result);
3238 			return true;
3239 		}
3240 
3241 		rbd_obj_copyup_write_object(obj_req);
3242 		if (!obj_req->pending.num_pending) {
3243 			*result = obj_req->pending.result;
3244 			obj_req->copyup_state = RBD_OBJ_COPYUP_WRITE_OBJECT;
3245 			goto again;
3246 		}
3247 		obj_req->copyup_state = __RBD_OBJ_COPYUP_WRITE_OBJECT;
3248 		return false;
3249 	case __RBD_OBJ_COPYUP_WRITE_OBJECT:
3250 		if (!pending_result_dec(&obj_req->pending, result))
3251 			return false;
3252 		fallthrough;
3253 	case RBD_OBJ_COPYUP_WRITE_OBJECT:
3254 		return true;
3255 	default:
3256 		BUG();
3257 	}
3258 }
3259 
3260 /*
3261  * Return:
3262  *   0 - object map update sent
3263  *   1 - object map update isn't needed
3264  *  <0 - error
3265  */
3266 static int rbd_obj_write_post_object_map(struct rbd_obj_request *obj_req)
3267 {
3268 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
3269 	u8 current_state = OBJECT_PENDING;
3270 
3271 	if (!(rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP))
3272 		return 1;
3273 
3274 	if (!(obj_req->flags & RBD_OBJ_FLAG_DELETION))
3275 		return 1;
3276 
3277 	return rbd_object_map_update(obj_req, CEPH_NOSNAP, OBJECT_NONEXISTENT,
3278 				     &current_state);
3279 }
3280 
3281 static bool rbd_obj_advance_write(struct rbd_obj_request *obj_req, int *result)
3282 {
3283 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
3284 	int ret;
3285 
3286 again:
3287 	switch (obj_req->write_state) {
3288 	case RBD_OBJ_WRITE_START:
3289 		rbd_assert(!*result);
3290 
3291 		if (rbd_obj_write_is_noop(obj_req))
3292 			return true;
3293 
3294 		ret = rbd_obj_write_pre_object_map(obj_req);
3295 		if (ret < 0) {
3296 			*result = ret;
3297 			return true;
3298 		}
3299 		obj_req->write_state = RBD_OBJ_WRITE_PRE_OBJECT_MAP;
3300 		if (ret > 0)
3301 			goto again;
3302 		return false;
3303 	case RBD_OBJ_WRITE_PRE_OBJECT_MAP:
3304 		if (*result) {
3305 			rbd_warn(rbd_dev, "pre object map update failed: %d",
3306 				 *result);
3307 			return true;
3308 		}
3309 		ret = rbd_obj_write_object(obj_req);
3310 		if (ret) {
3311 			*result = ret;
3312 			return true;
3313 		}
3314 		obj_req->write_state = RBD_OBJ_WRITE_OBJECT;
3315 		return false;
3316 	case RBD_OBJ_WRITE_OBJECT:
3317 		if (*result == -ENOENT) {
3318 			if (obj_req->flags & RBD_OBJ_FLAG_COPYUP_ENABLED) {
3319 				*result = 0;
3320 				obj_req->copyup_state = RBD_OBJ_COPYUP_START;
3321 				obj_req->write_state = __RBD_OBJ_WRITE_COPYUP;
3322 				goto again;
3323 			}
3324 			/*
3325 			 * On a non-existent object:
3326 			 *   delete - -ENOENT, truncate/zero - 0
3327 			 */
3328 			if (obj_req->flags & RBD_OBJ_FLAG_DELETION)
3329 				*result = 0;
3330 		}
3331 		if (*result)
3332 			return true;
3333 
3334 		obj_req->write_state = RBD_OBJ_WRITE_COPYUP;
3335 		goto again;
3336 	case __RBD_OBJ_WRITE_COPYUP:
3337 		if (!rbd_obj_advance_copyup(obj_req, result))
3338 			return false;
3339 		fallthrough;
3340 	case RBD_OBJ_WRITE_COPYUP:
3341 		if (*result) {
3342 			rbd_warn(rbd_dev, "copyup failed: %d", *result);
3343 			return true;
3344 		}
3345 		ret = rbd_obj_write_post_object_map(obj_req);
3346 		if (ret < 0) {
3347 			*result = ret;
3348 			return true;
3349 		}
3350 		obj_req->write_state = RBD_OBJ_WRITE_POST_OBJECT_MAP;
3351 		if (ret > 0)
3352 			goto again;
3353 		return false;
3354 	case RBD_OBJ_WRITE_POST_OBJECT_MAP:
3355 		if (*result)
3356 			rbd_warn(rbd_dev, "post object map update failed: %d",
3357 				 *result);
3358 		return true;
3359 	default:
3360 		BUG();
3361 	}
3362 }
3363 
3364 /*
3365  * Return true if @obj_req is completed.
3366  */
3367 static bool __rbd_obj_handle_request(struct rbd_obj_request *obj_req,
3368 				     int *result)
3369 {
3370 	struct rbd_img_request *img_req = obj_req->img_request;
3371 	struct rbd_device *rbd_dev = img_req->rbd_dev;
3372 	bool done;
3373 
3374 	mutex_lock(&obj_req->state_mutex);
3375 	if (!rbd_img_is_write(img_req))
3376 		done = rbd_obj_advance_read(obj_req, result);
3377 	else
3378 		done = rbd_obj_advance_write(obj_req, result);
3379 	mutex_unlock(&obj_req->state_mutex);
3380 
3381 	if (done && *result) {
3382 		rbd_assert(*result < 0);
3383 		rbd_warn(rbd_dev, "%s at objno %llu %llu~%llu result %d",
3384 			 obj_op_name(img_req->op_type), obj_req->ex.oe_objno,
3385 			 obj_req->ex.oe_off, obj_req->ex.oe_len, *result);
3386 	}
3387 	return done;
3388 }
3389 
3390 /*
3391  * This is open-coded in rbd_img_handle_request() to avoid parent chain
3392  * recursion.
3393  */
3394 static void rbd_obj_handle_request(struct rbd_obj_request *obj_req, int result)
3395 {
3396 	if (__rbd_obj_handle_request(obj_req, &result))
3397 		rbd_img_handle_request(obj_req->img_request, result);
3398 }
3399 
3400 static bool need_exclusive_lock(struct rbd_img_request *img_req)
3401 {
3402 	struct rbd_device *rbd_dev = img_req->rbd_dev;
3403 
3404 	if (!(rbd_dev->header.features & RBD_FEATURE_EXCLUSIVE_LOCK))
3405 		return false;
3406 
3407 	if (rbd_is_ro(rbd_dev))
3408 		return false;
3409 
3410 	rbd_assert(!test_bit(IMG_REQ_CHILD, &img_req->flags));
3411 	if (rbd_dev->opts->lock_on_read ||
3412 	    (rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP))
3413 		return true;
3414 
3415 	return rbd_img_is_write(img_req);
3416 }
3417 
3418 static bool rbd_lock_add_request(struct rbd_img_request *img_req)
3419 {
3420 	struct rbd_device *rbd_dev = img_req->rbd_dev;
3421 	bool locked;
3422 
3423 	lockdep_assert_held(&rbd_dev->lock_rwsem);
3424 	locked = rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED;
3425 	spin_lock(&rbd_dev->lock_lists_lock);
3426 	rbd_assert(list_empty(&img_req->lock_item));
3427 	if (!locked)
3428 		list_add_tail(&img_req->lock_item, &rbd_dev->acquiring_list);
3429 	else
3430 		list_add_tail(&img_req->lock_item, &rbd_dev->running_list);
3431 	spin_unlock(&rbd_dev->lock_lists_lock);
3432 	return locked;
3433 }
3434 
3435 static void rbd_lock_del_request(struct rbd_img_request *img_req)
3436 {
3437 	struct rbd_device *rbd_dev = img_req->rbd_dev;
3438 	bool need_wakeup;
3439 
3440 	lockdep_assert_held(&rbd_dev->lock_rwsem);
3441 	spin_lock(&rbd_dev->lock_lists_lock);
3442 	rbd_assert(!list_empty(&img_req->lock_item));
3443 	list_del_init(&img_req->lock_item);
3444 	need_wakeup = (rbd_dev->lock_state == RBD_LOCK_STATE_RELEASING &&
3445 		       list_empty(&rbd_dev->running_list));
3446 	spin_unlock(&rbd_dev->lock_lists_lock);
3447 	if (need_wakeup)
3448 		complete(&rbd_dev->releasing_wait);
3449 }
3450 
3451 static int rbd_img_exclusive_lock(struct rbd_img_request *img_req)
3452 {
3453 	struct rbd_device *rbd_dev = img_req->rbd_dev;
3454 
3455 	if (!need_exclusive_lock(img_req))
3456 		return 1;
3457 
3458 	if (rbd_lock_add_request(img_req))
3459 		return 1;
3460 
3461 	if (rbd_dev->opts->exclusive) {
3462 		WARN_ON(1); /* lock got released? */
3463 		return -EROFS;
3464 	}
3465 
3466 	/*
3467 	 * Note the use of mod_delayed_work() in rbd_acquire_lock()
3468 	 * and cancel_delayed_work() in wake_lock_waiters().
3469 	 */
3470 	dout("%s rbd_dev %p queueing lock_dwork\n", __func__, rbd_dev);
3471 	queue_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork, 0);
3472 	return 0;
3473 }
3474 
3475 static void rbd_img_object_requests(struct rbd_img_request *img_req)
3476 {
3477 	struct rbd_obj_request *obj_req;
3478 
3479 	rbd_assert(!img_req->pending.result && !img_req->pending.num_pending);
3480 
3481 	for_each_obj_request(img_req, obj_req) {
3482 		int result = 0;
3483 
3484 		if (__rbd_obj_handle_request(obj_req, &result)) {
3485 			if (result) {
3486 				img_req->pending.result = result;
3487 				return;
3488 			}
3489 		} else {
3490 			img_req->pending.num_pending++;
3491 		}
3492 	}
3493 }
3494 
3495 static bool rbd_img_advance(struct rbd_img_request *img_req, int *result)
3496 {
3497 	struct rbd_device *rbd_dev = img_req->rbd_dev;
3498 	int ret;
3499 
3500 again:
3501 	switch (img_req->state) {
3502 	case RBD_IMG_START:
3503 		rbd_assert(!*result);
3504 
3505 		ret = rbd_img_exclusive_lock(img_req);
3506 		if (ret < 0) {
3507 			*result = ret;
3508 			return true;
3509 		}
3510 		img_req->state = RBD_IMG_EXCLUSIVE_LOCK;
3511 		if (ret > 0)
3512 			goto again;
3513 		return false;
3514 	case RBD_IMG_EXCLUSIVE_LOCK:
3515 		if (*result)
3516 			return true;
3517 
3518 		rbd_assert(!need_exclusive_lock(img_req) ||
3519 			   __rbd_is_lock_owner(rbd_dev));
3520 
3521 		rbd_img_object_requests(img_req);
3522 		if (!img_req->pending.num_pending) {
3523 			*result = img_req->pending.result;
3524 			img_req->state = RBD_IMG_OBJECT_REQUESTS;
3525 			goto again;
3526 		}
3527 		img_req->state = __RBD_IMG_OBJECT_REQUESTS;
3528 		return false;
3529 	case __RBD_IMG_OBJECT_REQUESTS:
3530 		if (!pending_result_dec(&img_req->pending, result))
3531 			return false;
3532 		fallthrough;
3533 	case RBD_IMG_OBJECT_REQUESTS:
3534 		return true;
3535 	default:
3536 		BUG();
3537 	}
3538 }
3539 
3540 /*
3541  * Return true if @img_req is completed.
3542  */
3543 static bool __rbd_img_handle_request(struct rbd_img_request *img_req,
3544 				     int *result)
3545 {
3546 	struct rbd_device *rbd_dev = img_req->rbd_dev;
3547 	bool done;
3548 
3549 	if (need_exclusive_lock(img_req)) {
3550 		down_read(&rbd_dev->lock_rwsem);
3551 		mutex_lock(&img_req->state_mutex);
3552 		done = rbd_img_advance(img_req, result);
3553 		if (done)
3554 			rbd_lock_del_request(img_req);
3555 		mutex_unlock(&img_req->state_mutex);
3556 		up_read(&rbd_dev->lock_rwsem);
3557 	} else {
3558 		mutex_lock(&img_req->state_mutex);
3559 		done = rbd_img_advance(img_req, result);
3560 		mutex_unlock(&img_req->state_mutex);
3561 	}
3562 
3563 	if (done && *result) {
3564 		rbd_assert(*result < 0);
3565 		rbd_warn(rbd_dev, "%s%s result %d",
3566 		      test_bit(IMG_REQ_CHILD, &img_req->flags) ? "child " : "",
3567 		      obj_op_name(img_req->op_type), *result);
3568 	}
3569 	return done;
3570 }
3571 
3572 static void rbd_img_handle_request(struct rbd_img_request *img_req, int result)
3573 {
3574 again:
3575 	if (!__rbd_img_handle_request(img_req, &result))
3576 		return;
3577 
3578 	if (test_bit(IMG_REQ_CHILD, &img_req->flags)) {
3579 		struct rbd_obj_request *obj_req = img_req->obj_request;
3580 
3581 		rbd_img_request_destroy(img_req);
3582 		if (__rbd_obj_handle_request(obj_req, &result)) {
3583 			img_req = obj_req->img_request;
3584 			goto again;
3585 		}
3586 	} else {
3587 		struct request *rq = blk_mq_rq_from_pdu(img_req);
3588 
3589 		rbd_img_request_destroy(img_req);
3590 		blk_mq_end_request(rq, errno_to_blk_status(result));
3591 	}
3592 }
3593 
3594 static const struct rbd_client_id rbd_empty_cid;
3595 
3596 static bool rbd_cid_equal(const struct rbd_client_id *lhs,
3597 			  const struct rbd_client_id *rhs)
3598 {
3599 	return lhs->gid == rhs->gid && lhs->handle == rhs->handle;
3600 }
3601 
3602 static struct rbd_client_id rbd_get_cid(struct rbd_device *rbd_dev)
3603 {
3604 	struct rbd_client_id cid;
3605 
3606 	mutex_lock(&rbd_dev->watch_mutex);
3607 	cid.gid = ceph_client_gid(rbd_dev->rbd_client->client);
3608 	cid.handle = rbd_dev->watch_cookie;
3609 	mutex_unlock(&rbd_dev->watch_mutex);
3610 	return cid;
3611 }
3612 
3613 /*
3614  * lock_rwsem must be held for write
3615  */
3616 static void rbd_set_owner_cid(struct rbd_device *rbd_dev,
3617 			      const struct rbd_client_id *cid)
3618 {
3619 	dout("%s rbd_dev %p %llu-%llu -> %llu-%llu\n", __func__, rbd_dev,
3620 	     rbd_dev->owner_cid.gid, rbd_dev->owner_cid.handle,
3621 	     cid->gid, cid->handle);
3622 	rbd_dev->owner_cid = *cid; /* struct */
3623 }
3624 
3625 static void format_lock_cookie(struct rbd_device *rbd_dev, char *buf)
3626 {
3627 	mutex_lock(&rbd_dev->watch_mutex);
3628 	sprintf(buf, "%s %llu", RBD_LOCK_COOKIE_PREFIX, rbd_dev->watch_cookie);
3629 	mutex_unlock(&rbd_dev->watch_mutex);
3630 }
3631 
3632 static void __rbd_lock(struct rbd_device *rbd_dev, const char *cookie)
3633 {
3634 	struct rbd_client_id cid = rbd_get_cid(rbd_dev);
3635 
3636 	rbd_dev->lock_state = RBD_LOCK_STATE_LOCKED;
3637 	strcpy(rbd_dev->lock_cookie, cookie);
3638 	rbd_set_owner_cid(rbd_dev, &cid);
3639 	queue_work(rbd_dev->task_wq, &rbd_dev->acquired_lock_work);
3640 }
3641 
3642 /*
3643  * lock_rwsem must be held for write
3644  */
3645 static int rbd_lock(struct rbd_device *rbd_dev)
3646 {
3647 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3648 	char cookie[32];
3649 	int ret;
3650 
3651 	WARN_ON(__rbd_is_lock_owner(rbd_dev) ||
3652 		rbd_dev->lock_cookie[0] != '\0');
3653 
3654 	format_lock_cookie(rbd_dev, cookie);
3655 	ret = ceph_cls_lock(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc,
3656 			    RBD_LOCK_NAME, CEPH_CLS_LOCK_EXCLUSIVE, cookie,
3657 			    RBD_LOCK_TAG, "", 0);
3658 	if (ret)
3659 		return ret;
3660 
3661 	__rbd_lock(rbd_dev, cookie);
3662 	return 0;
3663 }
3664 
3665 /*
3666  * lock_rwsem must be held for write
3667  */
3668 static void rbd_unlock(struct rbd_device *rbd_dev)
3669 {
3670 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3671 	int ret;
3672 
3673 	WARN_ON(!__rbd_is_lock_owner(rbd_dev) ||
3674 		rbd_dev->lock_cookie[0] == '\0');
3675 
3676 	ret = ceph_cls_unlock(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc,
3677 			      RBD_LOCK_NAME, rbd_dev->lock_cookie);
3678 	if (ret && ret != -ENOENT)
3679 		rbd_warn(rbd_dev, "failed to unlock header: %d", ret);
3680 
3681 	/* treat errors as the image is unlocked */
3682 	rbd_dev->lock_state = RBD_LOCK_STATE_UNLOCKED;
3683 	rbd_dev->lock_cookie[0] = '\0';
3684 	rbd_set_owner_cid(rbd_dev, &rbd_empty_cid);
3685 	queue_work(rbd_dev->task_wq, &rbd_dev->released_lock_work);
3686 }
3687 
3688 static int __rbd_notify_op_lock(struct rbd_device *rbd_dev,
3689 				enum rbd_notify_op notify_op,
3690 				struct page ***preply_pages,
3691 				size_t *preply_len)
3692 {
3693 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3694 	struct rbd_client_id cid = rbd_get_cid(rbd_dev);
3695 	char buf[4 + 8 + 8 + CEPH_ENCODING_START_BLK_LEN];
3696 	int buf_size = sizeof(buf);
3697 	void *p = buf;
3698 
3699 	dout("%s rbd_dev %p notify_op %d\n", __func__, rbd_dev, notify_op);
3700 
3701 	/* encode *LockPayload NotifyMessage (op + ClientId) */
3702 	ceph_start_encoding(&p, 2, 1, buf_size - CEPH_ENCODING_START_BLK_LEN);
3703 	ceph_encode_32(&p, notify_op);
3704 	ceph_encode_64(&p, cid.gid);
3705 	ceph_encode_64(&p, cid.handle);
3706 
3707 	return ceph_osdc_notify(osdc, &rbd_dev->header_oid,
3708 				&rbd_dev->header_oloc, buf, buf_size,
3709 				RBD_NOTIFY_TIMEOUT, preply_pages, preply_len);
3710 }
3711 
3712 static void rbd_notify_op_lock(struct rbd_device *rbd_dev,
3713 			       enum rbd_notify_op notify_op)
3714 {
3715 	__rbd_notify_op_lock(rbd_dev, notify_op, NULL, NULL);
3716 }
3717 
3718 static void rbd_notify_acquired_lock(struct work_struct *work)
3719 {
3720 	struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
3721 						  acquired_lock_work);
3722 
3723 	rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_ACQUIRED_LOCK);
3724 }
3725 
3726 static void rbd_notify_released_lock(struct work_struct *work)
3727 {
3728 	struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
3729 						  released_lock_work);
3730 
3731 	rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_RELEASED_LOCK);
3732 }
3733 
3734 static int rbd_request_lock(struct rbd_device *rbd_dev)
3735 {
3736 	struct page **reply_pages;
3737 	size_t reply_len;
3738 	bool lock_owner_responded = false;
3739 	int ret;
3740 
3741 	dout("%s rbd_dev %p\n", __func__, rbd_dev);
3742 
3743 	ret = __rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_REQUEST_LOCK,
3744 				   &reply_pages, &reply_len);
3745 	if (ret && ret != -ETIMEDOUT) {
3746 		rbd_warn(rbd_dev, "failed to request lock: %d", ret);
3747 		goto out;
3748 	}
3749 
3750 	if (reply_len > 0 && reply_len <= PAGE_SIZE) {
3751 		void *p = page_address(reply_pages[0]);
3752 		void *const end = p + reply_len;
3753 		u32 n;
3754 
3755 		ceph_decode_32_safe(&p, end, n, e_inval); /* num_acks */
3756 		while (n--) {
3757 			u8 struct_v;
3758 			u32 len;
3759 
3760 			ceph_decode_need(&p, end, 8 + 8, e_inval);
3761 			p += 8 + 8; /* skip gid and cookie */
3762 
3763 			ceph_decode_32_safe(&p, end, len, e_inval);
3764 			if (!len)
3765 				continue;
3766 
3767 			if (lock_owner_responded) {
3768 				rbd_warn(rbd_dev,
3769 					 "duplicate lock owners detected");
3770 				ret = -EIO;
3771 				goto out;
3772 			}
3773 
3774 			lock_owner_responded = true;
3775 			ret = ceph_start_decoding(&p, end, 1, "ResponseMessage",
3776 						  &struct_v, &len);
3777 			if (ret) {
3778 				rbd_warn(rbd_dev,
3779 					 "failed to decode ResponseMessage: %d",
3780 					 ret);
3781 				goto e_inval;
3782 			}
3783 
3784 			ret = ceph_decode_32(&p);
3785 		}
3786 	}
3787 
3788 	if (!lock_owner_responded) {
3789 		rbd_warn(rbd_dev, "no lock owners detected");
3790 		ret = -ETIMEDOUT;
3791 	}
3792 
3793 out:
3794 	ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len));
3795 	return ret;
3796 
3797 e_inval:
3798 	ret = -EINVAL;
3799 	goto out;
3800 }
3801 
3802 /*
3803  * Either image request state machine(s) or rbd_add_acquire_lock()
3804  * (i.e. "rbd map").
3805  */
3806 static void wake_lock_waiters(struct rbd_device *rbd_dev, int result)
3807 {
3808 	struct rbd_img_request *img_req;
3809 
3810 	dout("%s rbd_dev %p result %d\n", __func__, rbd_dev, result);
3811 	lockdep_assert_held_write(&rbd_dev->lock_rwsem);
3812 
3813 	cancel_delayed_work(&rbd_dev->lock_dwork);
3814 	if (!completion_done(&rbd_dev->acquire_wait)) {
3815 		rbd_assert(list_empty(&rbd_dev->acquiring_list) &&
3816 			   list_empty(&rbd_dev->running_list));
3817 		rbd_dev->acquire_err = result;
3818 		complete_all(&rbd_dev->acquire_wait);
3819 		return;
3820 	}
3821 
3822 	list_for_each_entry(img_req, &rbd_dev->acquiring_list, lock_item) {
3823 		mutex_lock(&img_req->state_mutex);
3824 		rbd_assert(img_req->state == RBD_IMG_EXCLUSIVE_LOCK);
3825 		rbd_img_schedule(img_req, result);
3826 		mutex_unlock(&img_req->state_mutex);
3827 	}
3828 
3829 	list_splice_tail_init(&rbd_dev->acquiring_list, &rbd_dev->running_list);
3830 }
3831 
3832 static int get_lock_owner_info(struct rbd_device *rbd_dev,
3833 			       struct ceph_locker **lockers, u32 *num_lockers)
3834 {
3835 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3836 	u8 lock_type;
3837 	char *lock_tag;
3838 	int ret;
3839 
3840 	dout("%s rbd_dev %p\n", __func__, rbd_dev);
3841 
3842 	ret = ceph_cls_lock_info(osdc, &rbd_dev->header_oid,
3843 				 &rbd_dev->header_oloc, RBD_LOCK_NAME,
3844 				 &lock_type, &lock_tag, lockers, num_lockers);
3845 	if (ret)
3846 		return ret;
3847 
3848 	if (*num_lockers == 0) {
3849 		dout("%s rbd_dev %p no lockers detected\n", __func__, rbd_dev);
3850 		goto out;
3851 	}
3852 
3853 	if (strcmp(lock_tag, RBD_LOCK_TAG)) {
3854 		rbd_warn(rbd_dev, "locked by external mechanism, tag %s",
3855 			 lock_tag);
3856 		ret = -EBUSY;
3857 		goto out;
3858 	}
3859 
3860 	if (lock_type == CEPH_CLS_LOCK_SHARED) {
3861 		rbd_warn(rbd_dev, "shared lock type detected");
3862 		ret = -EBUSY;
3863 		goto out;
3864 	}
3865 
3866 	if (strncmp((*lockers)[0].id.cookie, RBD_LOCK_COOKIE_PREFIX,
3867 		    strlen(RBD_LOCK_COOKIE_PREFIX))) {
3868 		rbd_warn(rbd_dev, "locked by external mechanism, cookie %s",
3869 			 (*lockers)[0].id.cookie);
3870 		ret = -EBUSY;
3871 		goto out;
3872 	}
3873 
3874 out:
3875 	kfree(lock_tag);
3876 	return ret;
3877 }
3878 
3879 static int find_watcher(struct rbd_device *rbd_dev,
3880 			const struct ceph_locker *locker)
3881 {
3882 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3883 	struct ceph_watch_item *watchers;
3884 	u32 num_watchers;
3885 	u64 cookie;
3886 	int i;
3887 	int ret;
3888 
3889 	ret = ceph_osdc_list_watchers(osdc, &rbd_dev->header_oid,
3890 				      &rbd_dev->header_oloc, &watchers,
3891 				      &num_watchers);
3892 	if (ret)
3893 		return ret;
3894 
3895 	sscanf(locker->id.cookie, RBD_LOCK_COOKIE_PREFIX " %llu", &cookie);
3896 	for (i = 0; i < num_watchers; i++) {
3897 		/*
3898 		 * Ignore addr->type while comparing.  This mimics
3899 		 * entity_addr_t::get_legacy_str() + strcmp().
3900 		 */
3901 		if (ceph_addr_equal_no_type(&watchers[i].addr,
3902 					    &locker->info.addr) &&
3903 		    watchers[i].cookie == cookie) {
3904 			struct rbd_client_id cid = {
3905 				.gid = le64_to_cpu(watchers[i].name.num),
3906 				.handle = cookie,
3907 			};
3908 
3909 			dout("%s rbd_dev %p found cid %llu-%llu\n", __func__,
3910 			     rbd_dev, cid.gid, cid.handle);
3911 			rbd_set_owner_cid(rbd_dev, &cid);
3912 			ret = 1;
3913 			goto out;
3914 		}
3915 	}
3916 
3917 	dout("%s rbd_dev %p no watchers\n", __func__, rbd_dev);
3918 	ret = 0;
3919 out:
3920 	kfree(watchers);
3921 	return ret;
3922 }
3923 
3924 /*
3925  * lock_rwsem must be held for write
3926  */
3927 static int rbd_try_lock(struct rbd_device *rbd_dev)
3928 {
3929 	struct ceph_client *client = rbd_dev->rbd_client->client;
3930 	struct ceph_locker *lockers;
3931 	u32 num_lockers;
3932 	int ret;
3933 
3934 	for (;;) {
3935 		ret = rbd_lock(rbd_dev);
3936 		if (ret != -EBUSY)
3937 			return ret;
3938 
3939 		/* determine if the current lock holder is still alive */
3940 		ret = get_lock_owner_info(rbd_dev, &lockers, &num_lockers);
3941 		if (ret)
3942 			return ret;
3943 
3944 		if (num_lockers == 0)
3945 			goto again;
3946 
3947 		ret = find_watcher(rbd_dev, lockers);
3948 		if (ret)
3949 			goto out; /* request lock or error */
3950 
3951 		rbd_warn(rbd_dev, "breaking header lock owned by %s%llu",
3952 			 ENTITY_NAME(lockers[0].id.name));
3953 
3954 		ret = ceph_monc_blocklist_add(&client->monc,
3955 					      &lockers[0].info.addr);
3956 		if (ret) {
3957 			rbd_warn(rbd_dev, "blocklist of %s%llu failed: %d",
3958 				 ENTITY_NAME(lockers[0].id.name), ret);
3959 			goto out;
3960 		}
3961 
3962 		ret = ceph_cls_break_lock(&client->osdc, &rbd_dev->header_oid,
3963 					  &rbd_dev->header_oloc, RBD_LOCK_NAME,
3964 					  lockers[0].id.cookie,
3965 					  &lockers[0].id.name);
3966 		if (ret && ret != -ENOENT)
3967 			goto out;
3968 
3969 again:
3970 		ceph_free_lockers(lockers, num_lockers);
3971 	}
3972 
3973 out:
3974 	ceph_free_lockers(lockers, num_lockers);
3975 	return ret;
3976 }
3977 
3978 static int rbd_post_acquire_action(struct rbd_device *rbd_dev)
3979 {
3980 	int ret;
3981 
3982 	if (rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP) {
3983 		ret = rbd_object_map_open(rbd_dev);
3984 		if (ret)
3985 			return ret;
3986 	}
3987 
3988 	return 0;
3989 }
3990 
3991 /*
3992  * Return:
3993  *   0 - lock acquired
3994  *   1 - caller should call rbd_request_lock()
3995  *  <0 - error
3996  */
3997 static int rbd_try_acquire_lock(struct rbd_device *rbd_dev)
3998 {
3999 	int ret;
4000 
4001 	down_read(&rbd_dev->lock_rwsem);
4002 	dout("%s rbd_dev %p read lock_state %d\n", __func__, rbd_dev,
4003 	     rbd_dev->lock_state);
4004 	if (__rbd_is_lock_owner(rbd_dev)) {
4005 		up_read(&rbd_dev->lock_rwsem);
4006 		return 0;
4007 	}
4008 
4009 	up_read(&rbd_dev->lock_rwsem);
4010 	down_write(&rbd_dev->lock_rwsem);
4011 	dout("%s rbd_dev %p write lock_state %d\n", __func__, rbd_dev,
4012 	     rbd_dev->lock_state);
4013 	if (__rbd_is_lock_owner(rbd_dev)) {
4014 		up_write(&rbd_dev->lock_rwsem);
4015 		return 0;
4016 	}
4017 
4018 	ret = rbd_try_lock(rbd_dev);
4019 	if (ret < 0) {
4020 		rbd_warn(rbd_dev, "failed to lock header: %d", ret);
4021 		if (ret == -EBLOCKLISTED)
4022 			goto out;
4023 
4024 		ret = 1; /* request lock anyway */
4025 	}
4026 	if (ret > 0) {
4027 		up_write(&rbd_dev->lock_rwsem);
4028 		return ret;
4029 	}
4030 
4031 	rbd_assert(rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED);
4032 	rbd_assert(list_empty(&rbd_dev->running_list));
4033 
4034 	ret = rbd_post_acquire_action(rbd_dev);
4035 	if (ret) {
4036 		rbd_warn(rbd_dev, "post-acquire action failed: %d", ret);
4037 		/*
4038 		 * Can't stay in RBD_LOCK_STATE_LOCKED because
4039 		 * rbd_lock_add_request() would let the request through,
4040 		 * assuming that e.g. object map is locked and loaded.
4041 		 */
4042 		rbd_unlock(rbd_dev);
4043 	}
4044 
4045 out:
4046 	wake_lock_waiters(rbd_dev, ret);
4047 	up_write(&rbd_dev->lock_rwsem);
4048 	return ret;
4049 }
4050 
4051 static void rbd_acquire_lock(struct work_struct *work)
4052 {
4053 	struct rbd_device *rbd_dev = container_of(to_delayed_work(work),
4054 					    struct rbd_device, lock_dwork);
4055 	int ret;
4056 
4057 	dout("%s rbd_dev %p\n", __func__, rbd_dev);
4058 again:
4059 	ret = rbd_try_acquire_lock(rbd_dev);
4060 	if (ret <= 0) {
4061 		dout("%s rbd_dev %p ret %d - done\n", __func__, rbd_dev, ret);
4062 		return;
4063 	}
4064 
4065 	ret = rbd_request_lock(rbd_dev);
4066 	if (ret == -ETIMEDOUT) {
4067 		goto again; /* treat this as a dead client */
4068 	} else if (ret == -EROFS) {
4069 		rbd_warn(rbd_dev, "peer will not release lock");
4070 		down_write(&rbd_dev->lock_rwsem);
4071 		wake_lock_waiters(rbd_dev, ret);
4072 		up_write(&rbd_dev->lock_rwsem);
4073 	} else if (ret < 0) {
4074 		rbd_warn(rbd_dev, "error requesting lock: %d", ret);
4075 		mod_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork,
4076 				 RBD_RETRY_DELAY);
4077 	} else {
4078 		/*
4079 		 * lock owner acked, but resend if we don't see them
4080 		 * release the lock
4081 		 */
4082 		dout("%s rbd_dev %p requeuing lock_dwork\n", __func__,
4083 		     rbd_dev);
4084 		mod_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork,
4085 		    msecs_to_jiffies(2 * RBD_NOTIFY_TIMEOUT * MSEC_PER_SEC));
4086 	}
4087 }
4088 
4089 static bool rbd_quiesce_lock(struct rbd_device *rbd_dev)
4090 {
4091 	dout("%s rbd_dev %p\n", __func__, rbd_dev);
4092 	lockdep_assert_held_write(&rbd_dev->lock_rwsem);
4093 
4094 	if (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED)
4095 		return false;
4096 
4097 	/*
4098 	 * Ensure that all in-flight IO is flushed.
4099 	 */
4100 	rbd_dev->lock_state = RBD_LOCK_STATE_RELEASING;
4101 	rbd_assert(!completion_done(&rbd_dev->releasing_wait));
4102 	if (list_empty(&rbd_dev->running_list))
4103 		return true;
4104 
4105 	up_write(&rbd_dev->lock_rwsem);
4106 	wait_for_completion(&rbd_dev->releasing_wait);
4107 
4108 	down_write(&rbd_dev->lock_rwsem);
4109 	if (rbd_dev->lock_state != RBD_LOCK_STATE_RELEASING)
4110 		return false;
4111 
4112 	rbd_assert(list_empty(&rbd_dev->running_list));
4113 	return true;
4114 }
4115 
4116 static void rbd_pre_release_action(struct rbd_device *rbd_dev)
4117 {
4118 	if (rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP)
4119 		rbd_object_map_close(rbd_dev);
4120 }
4121 
4122 static void __rbd_release_lock(struct rbd_device *rbd_dev)
4123 {
4124 	rbd_assert(list_empty(&rbd_dev->running_list));
4125 
4126 	rbd_pre_release_action(rbd_dev);
4127 	rbd_unlock(rbd_dev);
4128 }
4129 
4130 /*
4131  * lock_rwsem must be held for write
4132  */
4133 static void rbd_release_lock(struct rbd_device *rbd_dev)
4134 {
4135 	if (!rbd_quiesce_lock(rbd_dev))
4136 		return;
4137 
4138 	__rbd_release_lock(rbd_dev);
4139 
4140 	/*
4141 	 * Give others a chance to grab the lock - we would re-acquire
4142 	 * almost immediately if we got new IO while draining the running
4143 	 * list otherwise.  We need to ack our own notifications, so this
4144 	 * lock_dwork will be requeued from rbd_handle_released_lock() by
4145 	 * way of maybe_kick_acquire().
4146 	 */
4147 	cancel_delayed_work(&rbd_dev->lock_dwork);
4148 }
4149 
4150 static void rbd_release_lock_work(struct work_struct *work)
4151 {
4152 	struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
4153 						  unlock_work);
4154 
4155 	down_write(&rbd_dev->lock_rwsem);
4156 	rbd_release_lock(rbd_dev);
4157 	up_write(&rbd_dev->lock_rwsem);
4158 }
4159 
4160 static void maybe_kick_acquire(struct rbd_device *rbd_dev)
4161 {
4162 	bool have_requests;
4163 
4164 	dout("%s rbd_dev %p\n", __func__, rbd_dev);
4165 	if (__rbd_is_lock_owner(rbd_dev))
4166 		return;
4167 
4168 	spin_lock(&rbd_dev->lock_lists_lock);
4169 	have_requests = !list_empty(&rbd_dev->acquiring_list);
4170 	spin_unlock(&rbd_dev->lock_lists_lock);
4171 	if (have_requests || delayed_work_pending(&rbd_dev->lock_dwork)) {
4172 		dout("%s rbd_dev %p kicking lock_dwork\n", __func__, rbd_dev);
4173 		mod_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork, 0);
4174 	}
4175 }
4176 
4177 static void rbd_handle_acquired_lock(struct rbd_device *rbd_dev, u8 struct_v,
4178 				     void **p)
4179 {
4180 	struct rbd_client_id cid = { 0 };
4181 
4182 	if (struct_v >= 2) {
4183 		cid.gid = ceph_decode_64(p);
4184 		cid.handle = ceph_decode_64(p);
4185 	}
4186 
4187 	dout("%s rbd_dev %p cid %llu-%llu\n", __func__, rbd_dev, cid.gid,
4188 	     cid.handle);
4189 	if (!rbd_cid_equal(&cid, &rbd_empty_cid)) {
4190 		down_write(&rbd_dev->lock_rwsem);
4191 		if (rbd_cid_equal(&cid, &rbd_dev->owner_cid)) {
4192 			dout("%s rbd_dev %p cid %llu-%llu == owner_cid\n",
4193 			     __func__, rbd_dev, cid.gid, cid.handle);
4194 		} else {
4195 			rbd_set_owner_cid(rbd_dev, &cid);
4196 		}
4197 		downgrade_write(&rbd_dev->lock_rwsem);
4198 	} else {
4199 		down_read(&rbd_dev->lock_rwsem);
4200 	}
4201 
4202 	maybe_kick_acquire(rbd_dev);
4203 	up_read(&rbd_dev->lock_rwsem);
4204 }
4205 
4206 static void rbd_handle_released_lock(struct rbd_device *rbd_dev, u8 struct_v,
4207 				     void **p)
4208 {
4209 	struct rbd_client_id cid = { 0 };
4210 
4211 	if (struct_v >= 2) {
4212 		cid.gid = ceph_decode_64(p);
4213 		cid.handle = ceph_decode_64(p);
4214 	}
4215 
4216 	dout("%s rbd_dev %p cid %llu-%llu\n", __func__, rbd_dev, cid.gid,
4217 	     cid.handle);
4218 	if (!rbd_cid_equal(&cid, &rbd_empty_cid)) {
4219 		down_write(&rbd_dev->lock_rwsem);
4220 		if (!rbd_cid_equal(&cid, &rbd_dev->owner_cid)) {
4221 			dout("%s rbd_dev %p cid %llu-%llu != owner_cid %llu-%llu\n",
4222 			     __func__, rbd_dev, cid.gid, cid.handle,
4223 			     rbd_dev->owner_cid.gid, rbd_dev->owner_cid.handle);
4224 		} else {
4225 			rbd_set_owner_cid(rbd_dev, &rbd_empty_cid);
4226 		}
4227 		downgrade_write(&rbd_dev->lock_rwsem);
4228 	} else {
4229 		down_read(&rbd_dev->lock_rwsem);
4230 	}
4231 
4232 	maybe_kick_acquire(rbd_dev);
4233 	up_read(&rbd_dev->lock_rwsem);
4234 }
4235 
4236 /*
4237  * Returns result for ResponseMessage to be encoded (<= 0), or 1 if no
4238  * ResponseMessage is needed.
4239  */
4240 static int rbd_handle_request_lock(struct rbd_device *rbd_dev, u8 struct_v,
4241 				   void **p)
4242 {
4243 	struct rbd_client_id my_cid = rbd_get_cid(rbd_dev);
4244 	struct rbd_client_id cid = { 0 };
4245 	int result = 1;
4246 
4247 	if (struct_v >= 2) {
4248 		cid.gid = ceph_decode_64(p);
4249 		cid.handle = ceph_decode_64(p);
4250 	}
4251 
4252 	dout("%s rbd_dev %p cid %llu-%llu\n", __func__, rbd_dev, cid.gid,
4253 	     cid.handle);
4254 	if (rbd_cid_equal(&cid, &my_cid))
4255 		return result;
4256 
4257 	down_read(&rbd_dev->lock_rwsem);
4258 	if (__rbd_is_lock_owner(rbd_dev)) {
4259 		if (rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED &&
4260 		    rbd_cid_equal(&rbd_dev->owner_cid, &rbd_empty_cid))
4261 			goto out_unlock;
4262 
4263 		/*
4264 		 * encode ResponseMessage(0) so the peer can detect
4265 		 * a missing owner
4266 		 */
4267 		result = 0;
4268 
4269 		if (rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED) {
4270 			if (!rbd_dev->opts->exclusive) {
4271 				dout("%s rbd_dev %p queueing unlock_work\n",
4272 				     __func__, rbd_dev);
4273 				queue_work(rbd_dev->task_wq,
4274 					   &rbd_dev->unlock_work);
4275 			} else {
4276 				/* refuse to release the lock */
4277 				result = -EROFS;
4278 			}
4279 		}
4280 	}
4281 
4282 out_unlock:
4283 	up_read(&rbd_dev->lock_rwsem);
4284 	return result;
4285 }
4286 
4287 static void __rbd_acknowledge_notify(struct rbd_device *rbd_dev,
4288 				     u64 notify_id, u64 cookie, s32 *result)
4289 {
4290 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
4291 	char buf[4 + CEPH_ENCODING_START_BLK_LEN];
4292 	int buf_size = sizeof(buf);
4293 	int ret;
4294 
4295 	if (result) {
4296 		void *p = buf;
4297 
4298 		/* encode ResponseMessage */
4299 		ceph_start_encoding(&p, 1, 1,
4300 				    buf_size - CEPH_ENCODING_START_BLK_LEN);
4301 		ceph_encode_32(&p, *result);
4302 	} else {
4303 		buf_size = 0;
4304 	}
4305 
4306 	ret = ceph_osdc_notify_ack(osdc, &rbd_dev->header_oid,
4307 				   &rbd_dev->header_oloc, notify_id, cookie,
4308 				   buf, buf_size);
4309 	if (ret)
4310 		rbd_warn(rbd_dev, "acknowledge_notify failed: %d", ret);
4311 }
4312 
4313 static void rbd_acknowledge_notify(struct rbd_device *rbd_dev, u64 notify_id,
4314 				   u64 cookie)
4315 {
4316 	dout("%s rbd_dev %p\n", __func__, rbd_dev);
4317 	__rbd_acknowledge_notify(rbd_dev, notify_id, cookie, NULL);
4318 }
4319 
4320 static void rbd_acknowledge_notify_result(struct rbd_device *rbd_dev,
4321 					  u64 notify_id, u64 cookie, s32 result)
4322 {
4323 	dout("%s rbd_dev %p result %d\n", __func__, rbd_dev, result);
4324 	__rbd_acknowledge_notify(rbd_dev, notify_id, cookie, &result);
4325 }
4326 
4327 static void rbd_watch_cb(void *arg, u64 notify_id, u64 cookie,
4328 			 u64 notifier_id, void *data, size_t data_len)
4329 {
4330 	struct rbd_device *rbd_dev = arg;
4331 	void *p = data;
4332 	void *const end = p + data_len;
4333 	u8 struct_v = 0;
4334 	u32 len;
4335 	u32 notify_op;
4336 	int ret;
4337 
4338 	dout("%s rbd_dev %p cookie %llu notify_id %llu data_len %zu\n",
4339 	     __func__, rbd_dev, cookie, notify_id, data_len);
4340 	if (data_len) {
4341 		ret = ceph_start_decoding(&p, end, 1, "NotifyMessage",
4342 					  &struct_v, &len);
4343 		if (ret) {
4344 			rbd_warn(rbd_dev, "failed to decode NotifyMessage: %d",
4345 				 ret);
4346 			return;
4347 		}
4348 
4349 		notify_op = ceph_decode_32(&p);
4350 	} else {
4351 		/* legacy notification for header updates */
4352 		notify_op = RBD_NOTIFY_OP_HEADER_UPDATE;
4353 		len = 0;
4354 	}
4355 
4356 	dout("%s rbd_dev %p notify_op %u\n", __func__, rbd_dev, notify_op);
4357 	switch (notify_op) {
4358 	case RBD_NOTIFY_OP_ACQUIRED_LOCK:
4359 		rbd_handle_acquired_lock(rbd_dev, struct_v, &p);
4360 		rbd_acknowledge_notify(rbd_dev, notify_id, cookie);
4361 		break;
4362 	case RBD_NOTIFY_OP_RELEASED_LOCK:
4363 		rbd_handle_released_lock(rbd_dev, struct_v, &p);
4364 		rbd_acknowledge_notify(rbd_dev, notify_id, cookie);
4365 		break;
4366 	case RBD_NOTIFY_OP_REQUEST_LOCK:
4367 		ret = rbd_handle_request_lock(rbd_dev, struct_v, &p);
4368 		if (ret <= 0)
4369 			rbd_acknowledge_notify_result(rbd_dev, notify_id,
4370 						      cookie, ret);
4371 		else
4372 			rbd_acknowledge_notify(rbd_dev, notify_id, cookie);
4373 		break;
4374 	case RBD_NOTIFY_OP_HEADER_UPDATE:
4375 		ret = rbd_dev_refresh(rbd_dev);
4376 		if (ret)
4377 			rbd_warn(rbd_dev, "refresh failed: %d", ret);
4378 
4379 		rbd_acknowledge_notify(rbd_dev, notify_id, cookie);
4380 		break;
4381 	default:
4382 		if (rbd_is_lock_owner(rbd_dev))
4383 			rbd_acknowledge_notify_result(rbd_dev, notify_id,
4384 						      cookie, -EOPNOTSUPP);
4385 		else
4386 			rbd_acknowledge_notify(rbd_dev, notify_id, cookie);
4387 		break;
4388 	}
4389 }
4390 
4391 static void __rbd_unregister_watch(struct rbd_device *rbd_dev);
4392 
4393 static void rbd_watch_errcb(void *arg, u64 cookie, int err)
4394 {
4395 	struct rbd_device *rbd_dev = arg;
4396 
4397 	rbd_warn(rbd_dev, "encountered watch error: %d", err);
4398 
4399 	down_write(&rbd_dev->lock_rwsem);
4400 	rbd_set_owner_cid(rbd_dev, &rbd_empty_cid);
4401 	up_write(&rbd_dev->lock_rwsem);
4402 
4403 	mutex_lock(&rbd_dev->watch_mutex);
4404 	if (rbd_dev->watch_state == RBD_WATCH_STATE_REGISTERED) {
4405 		__rbd_unregister_watch(rbd_dev);
4406 		rbd_dev->watch_state = RBD_WATCH_STATE_ERROR;
4407 
4408 		queue_delayed_work(rbd_dev->task_wq, &rbd_dev->watch_dwork, 0);
4409 	}
4410 	mutex_unlock(&rbd_dev->watch_mutex);
4411 }
4412 
4413 /*
4414  * watch_mutex must be locked
4415  */
4416 static int __rbd_register_watch(struct rbd_device *rbd_dev)
4417 {
4418 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
4419 	struct ceph_osd_linger_request *handle;
4420 
4421 	rbd_assert(!rbd_dev->watch_handle);
4422 	dout("%s rbd_dev %p\n", __func__, rbd_dev);
4423 
4424 	handle = ceph_osdc_watch(osdc, &rbd_dev->header_oid,
4425 				 &rbd_dev->header_oloc, rbd_watch_cb,
4426 				 rbd_watch_errcb, rbd_dev);
4427 	if (IS_ERR(handle))
4428 		return PTR_ERR(handle);
4429 
4430 	rbd_dev->watch_handle = handle;
4431 	return 0;
4432 }
4433 
4434 /*
4435  * watch_mutex must be locked
4436  */
4437 static void __rbd_unregister_watch(struct rbd_device *rbd_dev)
4438 {
4439 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
4440 	int ret;
4441 
4442 	rbd_assert(rbd_dev->watch_handle);
4443 	dout("%s rbd_dev %p\n", __func__, rbd_dev);
4444 
4445 	ret = ceph_osdc_unwatch(osdc, rbd_dev->watch_handle);
4446 	if (ret)
4447 		rbd_warn(rbd_dev, "failed to unwatch: %d", ret);
4448 
4449 	rbd_dev->watch_handle = NULL;
4450 }
4451 
4452 static int rbd_register_watch(struct rbd_device *rbd_dev)
4453 {
4454 	int ret;
4455 
4456 	mutex_lock(&rbd_dev->watch_mutex);
4457 	rbd_assert(rbd_dev->watch_state == RBD_WATCH_STATE_UNREGISTERED);
4458 	ret = __rbd_register_watch(rbd_dev);
4459 	if (ret)
4460 		goto out;
4461 
4462 	rbd_dev->watch_state = RBD_WATCH_STATE_REGISTERED;
4463 	rbd_dev->watch_cookie = rbd_dev->watch_handle->linger_id;
4464 
4465 out:
4466 	mutex_unlock(&rbd_dev->watch_mutex);
4467 	return ret;
4468 }
4469 
4470 static void cancel_tasks_sync(struct rbd_device *rbd_dev)
4471 {
4472 	dout("%s rbd_dev %p\n", __func__, rbd_dev);
4473 
4474 	cancel_work_sync(&rbd_dev->acquired_lock_work);
4475 	cancel_work_sync(&rbd_dev->released_lock_work);
4476 	cancel_delayed_work_sync(&rbd_dev->lock_dwork);
4477 	cancel_work_sync(&rbd_dev->unlock_work);
4478 }
4479 
4480 /*
4481  * header_rwsem must not be held to avoid a deadlock with
4482  * rbd_dev_refresh() when flushing notifies.
4483  */
4484 static void rbd_unregister_watch(struct rbd_device *rbd_dev)
4485 {
4486 	cancel_tasks_sync(rbd_dev);
4487 
4488 	mutex_lock(&rbd_dev->watch_mutex);
4489 	if (rbd_dev->watch_state == RBD_WATCH_STATE_REGISTERED)
4490 		__rbd_unregister_watch(rbd_dev);
4491 	rbd_dev->watch_state = RBD_WATCH_STATE_UNREGISTERED;
4492 	mutex_unlock(&rbd_dev->watch_mutex);
4493 
4494 	cancel_delayed_work_sync(&rbd_dev->watch_dwork);
4495 	ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc);
4496 }
4497 
4498 /*
4499  * lock_rwsem must be held for write
4500  */
4501 static void rbd_reacquire_lock(struct rbd_device *rbd_dev)
4502 {
4503 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
4504 	char cookie[32];
4505 	int ret;
4506 
4507 	if (!rbd_quiesce_lock(rbd_dev))
4508 		return;
4509 
4510 	format_lock_cookie(rbd_dev, cookie);
4511 	ret = ceph_cls_set_cookie(osdc, &rbd_dev->header_oid,
4512 				  &rbd_dev->header_oloc, RBD_LOCK_NAME,
4513 				  CEPH_CLS_LOCK_EXCLUSIVE, rbd_dev->lock_cookie,
4514 				  RBD_LOCK_TAG, cookie);
4515 	if (ret) {
4516 		if (ret != -EOPNOTSUPP)
4517 			rbd_warn(rbd_dev, "failed to update lock cookie: %d",
4518 				 ret);
4519 
4520 		/*
4521 		 * Lock cookie cannot be updated on older OSDs, so do
4522 		 * a manual release and queue an acquire.
4523 		 */
4524 		__rbd_release_lock(rbd_dev);
4525 		queue_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork, 0);
4526 	} else {
4527 		__rbd_lock(rbd_dev, cookie);
4528 		wake_lock_waiters(rbd_dev, 0);
4529 	}
4530 }
4531 
4532 static void rbd_reregister_watch(struct work_struct *work)
4533 {
4534 	struct rbd_device *rbd_dev = container_of(to_delayed_work(work),
4535 					    struct rbd_device, watch_dwork);
4536 	int ret;
4537 
4538 	dout("%s rbd_dev %p\n", __func__, rbd_dev);
4539 
4540 	mutex_lock(&rbd_dev->watch_mutex);
4541 	if (rbd_dev->watch_state != RBD_WATCH_STATE_ERROR) {
4542 		mutex_unlock(&rbd_dev->watch_mutex);
4543 		return;
4544 	}
4545 
4546 	ret = __rbd_register_watch(rbd_dev);
4547 	if (ret) {
4548 		rbd_warn(rbd_dev, "failed to reregister watch: %d", ret);
4549 		if (ret != -EBLOCKLISTED && ret != -ENOENT) {
4550 			queue_delayed_work(rbd_dev->task_wq,
4551 					   &rbd_dev->watch_dwork,
4552 					   RBD_RETRY_DELAY);
4553 			mutex_unlock(&rbd_dev->watch_mutex);
4554 			return;
4555 		}
4556 
4557 		mutex_unlock(&rbd_dev->watch_mutex);
4558 		down_write(&rbd_dev->lock_rwsem);
4559 		wake_lock_waiters(rbd_dev, ret);
4560 		up_write(&rbd_dev->lock_rwsem);
4561 		return;
4562 	}
4563 
4564 	rbd_dev->watch_state = RBD_WATCH_STATE_REGISTERED;
4565 	rbd_dev->watch_cookie = rbd_dev->watch_handle->linger_id;
4566 	mutex_unlock(&rbd_dev->watch_mutex);
4567 
4568 	down_write(&rbd_dev->lock_rwsem);
4569 	if (rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED)
4570 		rbd_reacquire_lock(rbd_dev);
4571 	up_write(&rbd_dev->lock_rwsem);
4572 
4573 	ret = rbd_dev_refresh(rbd_dev);
4574 	if (ret)
4575 		rbd_warn(rbd_dev, "reregistration refresh failed: %d", ret);
4576 }
4577 
4578 /*
4579  * Synchronous osd object method call.  Returns the number of bytes
4580  * returned in the outbound buffer, or a negative error code.
4581  */
4582 static int rbd_obj_method_sync(struct rbd_device *rbd_dev,
4583 			     struct ceph_object_id *oid,
4584 			     struct ceph_object_locator *oloc,
4585 			     const char *method_name,
4586 			     const void *outbound,
4587 			     size_t outbound_size,
4588 			     void *inbound,
4589 			     size_t inbound_size)
4590 {
4591 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
4592 	struct page *req_page = NULL;
4593 	struct page *reply_page;
4594 	int ret;
4595 
4596 	/*
4597 	 * Method calls are ultimately read operations.  The result
4598 	 * should placed into the inbound buffer provided.  They
4599 	 * also supply outbound data--parameters for the object
4600 	 * method.  Currently if this is present it will be a
4601 	 * snapshot id.
4602 	 */
4603 	if (outbound) {
4604 		if (outbound_size > PAGE_SIZE)
4605 			return -E2BIG;
4606 
4607 		req_page = alloc_page(GFP_KERNEL);
4608 		if (!req_page)
4609 			return -ENOMEM;
4610 
4611 		memcpy(page_address(req_page), outbound, outbound_size);
4612 	}
4613 
4614 	reply_page = alloc_page(GFP_KERNEL);
4615 	if (!reply_page) {
4616 		if (req_page)
4617 			__free_page(req_page);
4618 		return -ENOMEM;
4619 	}
4620 
4621 	ret = ceph_osdc_call(osdc, oid, oloc, RBD_DRV_NAME, method_name,
4622 			     CEPH_OSD_FLAG_READ, req_page, outbound_size,
4623 			     &reply_page, &inbound_size);
4624 	if (!ret) {
4625 		memcpy(inbound, page_address(reply_page), inbound_size);
4626 		ret = inbound_size;
4627 	}
4628 
4629 	if (req_page)
4630 		__free_page(req_page);
4631 	__free_page(reply_page);
4632 	return ret;
4633 }
4634 
4635 static void rbd_queue_workfn(struct work_struct *work)
4636 {
4637 	struct rbd_img_request *img_request =
4638 	    container_of(work, struct rbd_img_request, work);
4639 	struct rbd_device *rbd_dev = img_request->rbd_dev;
4640 	enum obj_operation_type op_type = img_request->op_type;
4641 	struct request *rq = blk_mq_rq_from_pdu(img_request);
4642 	u64 offset = (u64)blk_rq_pos(rq) << SECTOR_SHIFT;
4643 	u64 length = blk_rq_bytes(rq);
4644 	u64 mapping_size;
4645 	int result;
4646 
4647 	/* Ignore/skip any zero-length requests */
4648 	if (!length) {
4649 		dout("%s: zero-length request\n", __func__);
4650 		result = 0;
4651 		goto err_img_request;
4652 	}
4653 
4654 	blk_mq_start_request(rq);
4655 
4656 	down_read(&rbd_dev->header_rwsem);
4657 	mapping_size = rbd_dev->mapping.size;
4658 	rbd_img_capture_header(img_request);
4659 	up_read(&rbd_dev->header_rwsem);
4660 
4661 	if (offset + length > mapping_size) {
4662 		rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)", offset,
4663 			 length, mapping_size);
4664 		result = -EIO;
4665 		goto err_img_request;
4666 	}
4667 
4668 	dout("%s rbd_dev %p img_req %p %s %llu~%llu\n", __func__, rbd_dev,
4669 	     img_request, obj_op_name(op_type), offset, length);
4670 
4671 	if (op_type == OBJ_OP_DISCARD || op_type == OBJ_OP_ZEROOUT)
4672 		result = rbd_img_fill_nodata(img_request, offset, length);
4673 	else
4674 		result = rbd_img_fill_from_bio(img_request, offset, length,
4675 					       rq->bio);
4676 	if (result)
4677 		goto err_img_request;
4678 
4679 	rbd_img_handle_request(img_request, 0);
4680 	return;
4681 
4682 err_img_request:
4683 	rbd_img_request_destroy(img_request);
4684 	if (result)
4685 		rbd_warn(rbd_dev, "%s %llx at %llx result %d",
4686 			 obj_op_name(op_type), length, offset, result);
4687 	blk_mq_end_request(rq, errno_to_blk_status(result));
4688 }
4689 
4690 static blk_status_t rbd_queue_rq(struct blk_mq_hw_ctx *hctx,
4691 		const struct blk_mq_queue_data *bd)
4692 {
4693 	struct rbd_device *rbd_dev = hctx->queue->queuedata;
4694 	struct rbd_img_request *img_req = blk_mq_rq_to_pdu(bd->rq);
4695 	enum obj_operation_type op_type;
4696 
4697 	switch (req_op(bd->rq)) {
4698 	case REQ_OP_DISCARD:
4699 		op_type = OBJ_OP_DISCARD;
4700 		break;
4701 	case REQ_OP_WRITE_ZEROES:
4702 		op_type = OBJ_OP_ZEROOUT;
4703 		break;
4704 	case REQ_OP_WRITE:
4705 		op_type = OBJ_OP_WRITE;
4706 		break;
4707 	case REQ_OP_READ:
4708 		op_type = OBJ_OP_READ;
4709 		break;
4710 	default:
4711 		rbd_warn(rbd_dev, "unknown req_op %d", req_op(bd->rq));
4712 		return BLK_STS_IOERR;
4713 	}
4714 
4715 	rbd_img_request_init(img_req, rbd_dev, op_type);
4716 
4717 	if (rbd_img_is_write(img_req)) {
4718 		if (rbd_is_ro(rbd_dev)) {
4719 			rbd_warn(rbd_dev, "%s on read-only mapping",
4720 				 obj_op_name(img_req->op_type));
4721 			return BLK_STS_IOERR;
4722 		}
4723 		rbd_assert(!rbd_is_snap(rbd_dev));
4724 	}
4725 
4726 	INIT_WORK(&img_req->work, rbd_queue_workfn);
4727 	queue_work(rbd_wq, &img_req->work);
4728 	return BLK_STS_OK;
4729 }
4730 
4731 static void rbd_free_disk(struct rbd_device *rbd_dev)
4732 {
4733 	blk_cleanup_disk(rbd_dev->disk);
4734 	blk_mq_free_tag_set(&rbd_dev->tag_set);
4735 	rbd_dev->disk = NULL;
4736 }
4737 
4738 static int rbd_obj_read_sync(struct rbd_device *rbd_dev,
4739 			     struct ceph_object_id *oid,
4740 			     struct ceph_object_locator *oloc,
4741 			     void *buf, int buf_len)
4742 
4743 {
4744 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
4745 	struct ceph_osd_request *req;
4746 	struct page **pages;
4747 	int num_pages = calc_pages_for(0, buf_len);
4748 	int ret;
4749 
4750 	req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_KERNEL);
4751 	if (!req)
4752 		return -ENOMEM;
4753 
4754 	ceph_oid_copy(&req->r_base_oid, oid);
4755 	ceph_oloc_copy(&req->r_base_oloc, oloc);
4756 	req->r_flags = CEPH_OSD_FLAG_READ;
4757 
4758 	pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
4759 	if (IS_ERR(pages)) {
4760 		ret = PTR_ERR(pages);
4761 		goto out_req;
4762 	}
4763 
4764 	osd_req_op_extent_init(req, 0, CEPH_OSD_OP_READ, 0, buf_len, 0, 0);
4765 	osd_req_op_extent_osd_data_pages(req, 0, pages, buf_len, 0, false,
4766 					 true);
4767 
4768 	ret = ceph_osdc_alloc_messages(req, GFP_KERNEL);
4769 	if (ret)
4770 		goto out_req;
4771 
4772 	ceph_osdc_start_request(osdc, req, false);
4773 	ret = ceph_osdc_wait_request(osdc, req);
4774 	if (ret >= 0)
4775 		ceph_copy_from_page_vector(pages, buf, 0, ret);
4776 
4777 out_req:
4778 	ceph_osdc_put_request(req);
4779 	return ret;
4780 }
4781 
4782 /*
4783  * Read the complete header for the given rbd device.  On successful
4784  * return, the rbd_dev->header field will contain up-to-date
4785  * information about the image.
4786  */
4787 static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev)
4788 {
4789 	struct rbd_image_header_ondisk *ondisk = NULL;
4790 	u32 snap_count = 0;
4791 	u64 names_size = 0;
4792 	u32 want_count;
4793 	int ret;
4794 
4795 	/*
4796 	 * The complete header will include an array of its 64-bit
4797 	 * snapshot ids, followed by the names of those snapshots as
4798 	 * a contiguous block of NUL-terminated strings.  Note that
4799 	 * the number of snapshots could change by the time we read
4800 	 * it in, in which case we re-read it.
4801 	 */
4802 	do {
4803 		size_t size;
4804 
4805 		kfree(ondisk);
4806 
4807 		size = sizeof (*ondisk);
4808 		size += snap_count * sizeof (struct rbd_image_snap_ondisk);
4809 		size += names_size;
4810 		ondisk = kmalloc(size, GFP_KERNEL);
4811 		if (!ondisk)
4812 			return -ENOMEM;
4813 
4814 		ret = rbd_obj_read_sync(rbd_dev, &rbd_dev->header_oid,
4815 					&rbd_dev->header_oloc, ondisk, size);
4816 		if (ret < 0)
4817 			goto out;
4818 		if ((size_t)ret < size) {
4819 			ret = -ENXIO;
4820 			rbd_warn(rbd_dev, "short header read (want %zd got %d)",
4821 				size, ret);
4822 			goto out;
4823 		}
4824 		if (!rbd_dev_ondisk_valid(ondisk)) {
4825 			ret = -ENXIO;
4826 			rbd_warn(rbd_dev, "invalid header");
4827 			goto out;
4828 		}
4829 
4830 		names_size = le64_to_cpu(ondisk->snap_names_len);
4831 		want_count = snap_count;
4832 		snap_count = le32_to_cpu(ondisk->snap_count);
4833 	} while (snap_count != want_count);
4834 
4835 	ret = rbd_header_from_disk(rbd_dev, ondisk);
4836 out:
4837 	kfree(ondisk);
4838 
4839 	return ret;
4840 }
4841 
4842 static void rbd_dev_update_size(struct rbd_device *rbd_dev)
4843 {
4844 	sector_t size;
4845 
4846 	/*
4847 	 * If EXISTS is not set, rbd_dev->disk may be NULL, so don't
4848 	 * try to update its size.  If REMOVING is set, updating size
4849 	 * is just useless work since the device can't be opened.
4850 	 */
4851 	if (test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags) &&
4852 	    !test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) {
4853 		size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE;
4854 		dout("setting size to %llu sectors", (unsigned long long)size);
4855 		set_capacity_and_notify(rbd_dev->disk, size);
4856 	}
4857 }
4858 
4859 static int rbd_dev_refresh(struct rbd_device *rbd_dev)
4860 {
4861 	u64 mapping_size;
4862 	int ret;
4863 
4864 	down_write(&rbd_dev->header_rwsem);
4865 	mapping_size = rbd_dev->mapping.size;
4866 
4867 	ret = rbd_dev_header_info(rbd_dev);
4868 	if (ret)
4869 		goto out;
4870 
4871 	/*
4872 	 * If there is a parent, see if it has disappeared due to the
4873 	 * mapped image getting flattened.
4874 	 */
4875 	if (rbd_dev->parent) {
4876 		ret = rbd_dev_v2_parent_info(rbd_dev);
4877 		if (ret)
4878 			goto out;
4879 	}
4880 
4881 	rbd_assert(!rbd_is_snap(rbd_dev));
4882 	rbd_dev->mapping.size = rbd_dev->header.image_size;
4883 
4884 out:
4885 	up_write(&rbd_dev->header_rwsem);
4886 	if (!ret && mapping_size != rbd_dev->mapping.size)
4887 		rbd_dev_update_size(rbd_dev);
4888 
4889 	return ret;
4890 }
4891 
4892 static const struct blk_mq_ops rbd_mq_ops = {
4893 	.queue_rq	= rbd_queue_rq,
4894 };
4895 
4896 static int rbd_init_disk(struct rbd_device *rbd_dev)
4897 {
4898 	struct gendisk *disk;
4899 	struct request_queue *q;
4900 	unsigned int objset_bytes =
4901 	    rbd_dev->layout.object_size * rbd_dev->layout.stripe_count;
4902 	int err;
4903 
4904 	memset(&rbd_dev->tag_set, 0, sizeof(rbd_dev->tag_set));
4905 	rbd_dev->tag_set.ops = &rbd_mq_ops;
4906 	rbd_dev->tag_set.queue_depth = rbd_dev->opts->queue_depth;
4907 	rbd_dev->tag_set.numa_node = NUMA_NO_NODE;
4908 	rbd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
4909 	rbd_dev->tag_set.nr_hw_queues = num_present_cpus();
4910 	rbd_dev->tag_set.cmd_size = sizeof(struct rbd_img_request);
4911 
4912 	err = blk_mq_alloc_tag_set(&rbd_dev->tag_set);
4913 	if (err)
4914 		return err;
4915 
4916 	disk = blk_mq_alloc_disk(&rbd_dev->tag_set, rbd_dev);
4917 	if (IS_ERR(disk)) {
4918 		err = PTR_ERR(disk);
4919 		goto out_tag_set;
4920 	}
4921 	q = disk->queue;
4922 
4923 	snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
4924 		 rbd_dev->dev_id);
4925 	disk->major = rbd_dev->major;
4926 	disk->first_minor = rbd_dev->minor;
4927 	if (single_major)
4928 		disk->minors = (1 << RBD_SINGLE_MAJOR_PART_SHIFT);
4929 	else
4930 		disk->minors = RBD_MINORS_PER_MAJOR;
4931 	disk->fops = &rbd_bd_ops;
4932 	disk->private_data = rbd_dev;
4933 
4934 	blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
4935 	/* QUEUE_FLAG_ADD_RANDOM is off by default for blk-mq */
4936 
4937 	blk_queue_max_hw_sectors(q, objset_bytes >> SECTOR_SHIFT);
4938 	q->limits.max_sectors = queue_max_hw_sectors(q);
4939 	blk_queue_max_segments(q, USHRT_MAX);
4940 	blk_queue_max_segment_size(q, UINT_MAX);
4941 	blk_queue_io_min(q, rbd_dev->opts->alloc_size);
4942 	blk_queue_io_opt(q, rbd_dev->opts->alloc_size);
4943 
4944 	if (rbd_dev->opts->trim) {
4945 		blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
4946 		q->limits.discard_granularity = rbd_dev->opts->alloc_size;
4947 		blk_queue_max_discard_sectors(q, objset_bytes >> SECTOR_SHIFT);
4948 		blk_queue_max_write_zeroes_sectors(q, objset_bytes >> SECTOR_SHIFT);
4949 	}
4950 
4951 	if (!ceph_test_opt(rbd_dev->rbd_client->client, NOCRC))
4952 		blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, q);
4953 
4954 	rbd_dev->disk = disk;
4955 
4956 	return 0;
4957 out_tag_set:
4958 	blk_mq_free_tag_set(&rbd_dev->tag_set);
4959 	return err;
4960 }
4961 
4962 /*
4963   sysfs
4964 */
4965 
4966 static struct rbd_device *dev_to_rbd_dev(struct device *dev)
4967 {
4968 	return container_of(dev, struct rbd_device, dev);
4969 }
4970 
4971 static ssize_t rbd_size_show(struct device *dev,
4972 			     struct device_attribute *attr, char *buf)
4973 {
4974 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4975 
4976 	return sprintf(buf, "%llu\n",
4977 		(unsigned long long)rbd_dev->mapping.size);
4978 }
4979 
4980 static ssize_t rbd_features_show(struct device *dev,
4981 			     struct device_attribute *attr, char *buf)
4982 {
4983 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4984 
4985 	return sprintf(buf, "0x%016llx\n", rbd_dev->header.features);
4986 }
4987 
4988 static ssize_t rbd_major_show(struct device *dev,
4989 			      struct device_attribute *attr, char *buf)
4990 {
4991 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4992 
4993 	if (rbd_dev->major)
4994 		return sprintf(buf, "%d\n", rbd_dev->major);
4995 
4996 	return sprintf(buf, "(none)\n");
4997 }
4998 
4999 static ssize_t rbd_minor_show(struct device *dev,
5000 			      struct device_attribute *attr, char *buf)
5001 {
5002 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5003 
5004 	return sprintf(buf, "%d\n", rbd_dev->minor);
5005 }
5006 
5007 static ssize_t rbd_client_addr_show(struct device *dev,
5008 				    struct device_attribute *attr, char *buf)
5009 {
5010 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5011 	struct ceph_entity_addr *client_addr =
5012 	    ceph_client_addr(rbd_dev->rbd_client->client);
5013 
5014 	return sprintf(buf, "%pISpc/%u\n", &client_addr->in_addr,
5015 		       le32_to_cpu(client_addr->nonce));
5016 }
5017 
5018 static ssize_t rbd_client_id_show(struct device *dev,
5019 				  struct device_attribute *attr, char *buf)
5020 {
5021 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5022 
5023 	return sprintf(buf, "client%lld\n",
5024 		       ceph_client_gid(rbd_dev->rbd_client->client));
5025 }
5026 
5027 static ssize_t rbd_cluster_fsid_show(struct device *dev,
5028 				     struct device_attribute *attr, char *buf)
5029 {
5030 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5031 
5032 	return sprintf(buf, "%pU\n", &rbd_dev->rbd_client->client->fsid);
5033 }
5034 
5035 static ssize_t rbd_config_info_show(struct device *dev,
5036 				    struct device_attribute *attr, char *buf)
5037 {
5038 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5039 
5040 	if (!capable(CAP_SYS_ADMIN))
5041 		return -EPERM;
5042 
5043 	return sprintf(buf, "%s\n", rbd_dev->config_info);
5044 }
5045 
5046 static ssize_t rbd_pool_show(struct device *dev,
5047 			     struct device_attribute *attr, char *buf)
5048 {
5049 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5050 
5051 	return sprintf(buf, "%s\n", rbd_dev->spec->pool_name);
5052 }
5053 
5054 static ssize_t rbd_pool_id_show(struct device *dev,
5055 			     struct device_attribute *attr, char *buf)
5056 {
5057 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5058 
5059 	return sprintf(buf, "%llu\n",
5060 			(unsigned long long) rbd_dev->spec->pool_id);
5061 }
5062 
5063 static ssize_t rbd_pool_ns_show(struct device *dev,
5064 				struct device_attribute *attr, char *buf)
5065 {
5066 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5067 
5068 	return sprintf(buf, "%s\n", rbd_dev->spec->pool_ns ?: "");
5069 }
5070 
5071 static ssize_t rbd_name_show(struct device *dev,
5072 			     struct device_attribute *attr, char *buf)
5073 {
5074 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5075 
5076 	if (rbd_dev->spec->image_name)
5077 		return sprintf(buf, "%s\n", rbd_dev->spec->image_name);
5078 
5079 	return sprintf(buf, "(unknown)\n");
5080 }
5081 
5082 static ssize_t rbd_image_id_show(struct device *dev,
5083 			     struct device_attribute *attr, char *buf)
5084 {
5085 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5086 
5087 	return sprintf(buf, "%s\n", rbd_dev->spec->image_id);
5088 }
5089 
5090 /*
5091  * Shows the name of the currently-mapped snapshot (or
5092  * RBD_SNAP_HEAD_NAME for the base image).
5093  */
5094 static ssize_t rbd_snap_show(struct device *dev,
5095 			     struct device_attribute *attr,
5096 			     char *buf)
5097 {
5098 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5099 
5100 	return sprintf(buf, "%s\n", rbd_dev->spec->snap_name);
5101 }
5102 
5103 static ssize_t rbd_snap_id_show(struct device *dev,
5104 				struct device_attribute *attr, char *buf)
5105 {
5106 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5107 
5108 	return sprintf(buf, "%llu\n", rbd_dev->spec->snap_id);
5109 }
5110 
5111 /*
5112  * For a v2 image, shows the chain of parent images, separated by empty
5113  * lines.  For v1 images or if there is no parent, shows "(no parent
5114  * image)".
5115  */
5116 static ssize_t rbd_parent_show(struct device *dev,
5117 			       struct device_attribute *attr,
5118 			       char *buf)
5119 {
5120 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5121 	ssize_t count = 0;
5122 
5123 	if (!rbd_dev->parent)
5124 		return sprintf(buf, "(no parent image)\n");
5125 
5126 	for ( ; rbd_dev->parent; rbd_dev = rbd_dev->parent) {
5127 		struct rbd_spec *spec = rbd_dev->parent_spec;
5128 
5129 		count += sprintf(&buf[count], "%s"
5130 			    "pool_id %llu\npool_name %s\n"
5131 			    "pool_ns %s\n"
5132 			    "image_id %s\nimage_name %s\n"
5133 			    "snap_id %llu\nsnap_name %s\n"
5134 			    "overlap %llu\n",
5135 			    !count ? "" : "\n", /* first? */
5136 			    spec->pool_id, spec->pool_name,
5137 			    spec->pool_ns ?: "",
5138 			    spec->image_id, spec->image_name ?: "(unknown)",
5139 			    spec->snap_id, spec->snap_name,
5140 			    rbd_dev->parent_overlap);
5141 	}
5142 
5143 	return count;
5144 }
5145 
5146 static ssize_t rbd_image_refresh(struct device *dev,
5147 				 struct device_attribute *attr,
5148 				 const char *buf,
5149 				 size_t size)
5150 {
5151 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5152 	int ret;
5153 
5154 	if (!capable(CAP_SYS_ADMIN))
5155 		return -EPERM;
5156 
5157 	ret = rbd_dev_refresh(rbd_dev);
5158 	if (ret)
5159 		return ret;
5160 
5161 	return size;
5162 }
5163 
5164 static DEVICE_ATTR(size, 0444, rbd_size_show, NULL);
5165 static DEVICE_ATTR(features, 0444, rbd_features_show, NULL);
5166 static DEVICE_ATTR(major, 0444, rbd_major_show, NULL);
5167 static DEVICE_ATTR(minor, 0444, rbd_minor_show, NULL);
5168 static DEVICE_ATTR(client_addr, 0444, rbd_client_addr_show, NULL);
5169 static DEVICE_ATTR(client_id, 0444, rbd_client_id_show, NULL);
5170 static DEVICE_ATTR(cluster_fsid, 0444, rbd_cluster_fsid_show, NULL);
5171 static DEVICE_ATTR(config_info, 0400, rbd_config_info_show, NULL);
5172 static DEVICE_ATTR(pool, 0444, rbd_pool_show, NULL);
5173 static DEVICE_ATTR(pool_id, 0444, rbd_pool_id_show, NULL);
5174 static DEVICE_ATTR(pool_ns, 0444, rbd_pool_ns_show, NULL);
5175 static DEVICE_ATTR(name, 0444, rbd_name_show, NULL);
5176 static DEVICE_ATTR(image_id, 0444, rbd_image_id_show, NULL);
5177 static DEVICE_ATTR(refresh, 0200, NULL, rbd_image_refresh);
5178 static DEVICE_ATTR(current_snap, 0444, rbd_snap_show, NULL);
5179 static DEVICE_ATTR(snap_id, 0444, rbd_snap_id_show, NULL);
5180 static DEVICE_ATTR(parent, 0444, rbd_parent_show, NULL);
5181 
5182 static struct attribute *rbd_attrs[] = {
5183 	&dev_attr_size.attr,
5184 	&dev_attr_features.attr,
5185 	&dev_attr_major.attr,
5186 	&dev_attr_minor.attr,
5187 	&dev_attr_client_addr.attr,
5188 	&dev_attr_client_id.attr,
5189 	&dev_attr_cluster_fsid.attr,
5190 	&dev_attr_config_info.attr,
5191 	&dev_attr_pool.attr,
5192 	&dev_attr_pool_id.attr,
5193 	&dev_attr_pool_ns.attr,
5194 	&dev_attr_name.attr,
5195 	&dev_attr_image_id.attr,
5196 	&dev_attr_current_snap.attr,
5197 	&dev_attr_snap_id.attr,
5198 	&dev_attr_parent.attr,
5199 	&dev_attr_refresh.attr,
5200 	NULL
5201 };
5202 
5203 static struct attribute_group rbd_attr_group = {
5204 	.attrs = rbd_attrs,
5205 };
5206 
5207 static const struct attribute_group *rbd_attr_groups[] = {
5208 	&rbd_attr_group,
5209 	NULL
5210 };
5211 
5212 static void rbd_dev_release(struct device *dev);
5213 
5214 static const struct device_type rbd_device_type = {
5215 	.name		= "rbd",
5216 	.groups		= rbd_attr_groups,
5217 	.release	= rbd_dev_release,
5218 };
5219 
5220 static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec)
5221 {
5222 	kref_get(&spec->kref);
5223 
5224 	return spec;
5225 }
5226 
5227 static void rbd_spec_free(struct kref *kref);
5228 static void rbd_spec_put(struct rbd_spec *spec)
5229 {
5230 	if (spec)
5231 		kref_put(&spec->kref, rbd_spec_free);
5232 }
5233 
5234 static struct rbd_spec *rbd_spec_alloc(void)
5235 {
5236 	struct rbd_spec *spec;
5237 
5238 	spec = kzalloc(sizeof (*spec), GFP_KERNEL);
5239 	if (!spec)
5240 		return NULL;
5241 
5242 	spec->pool_id = CEPH_NOPOOL;
5243 	spec->snap_id = CEPH_NOSNAP;
5244 	kref_init(&spec->kref);
5245 
5246 	return spec;
5247 }
5248 
5249 static void rbd_spec_free(struct kref *kref)
5250 {
5251 	struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref);
5252 
5253 	kfree(spec->pool_name);
5254 	kfree(spec->pool_ns);
5255 	kfree(spec->image_id);
5256 	kfree(spec->image_name);
5257 	kfree(spec->snap_name);
5258 	kfree(spec);
5259 }
5260 
5261 static void rbd_dev_free(struct rbd_device *rbd_dev)
5262 {
5263 	WARN_ON(rbd_dev->watch_state != RBD_WATCH_STATE_UNREGISTERED);
5264 	WARN_ON(rbd_dev->lock_state != RBD_LOCK_STATE_UNLOCKED);
5265 
5266 	ceph_oid_destroy(&rbd_dev->header_oid);
5267 	ceph_oloc_destroy(&rbd_dev->header_oloc);
5268 	kfree(rbd_dev->config_info);
5269 
5270 	rbd_put_client(rbd_dev->rbd_client);
5271 	rbd_spec_put(rbd_dev->spec);
5272 	kfree(rbd_dev->opts);
5273 	kfree(rbd_dev);
5274 }
5275 
5276 static void rbd_dev_release(struct device *dev)
5277 {
5278 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5279 	bool need_put = !!rbd_dev->opts;
5280 
5281 	if (need_put) {
5282 		destroy_workqueue(rbd_dev->task_wq);
5283 		ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id);
5284 	}
5285 
5286 	rbd_dev_free(rbd_dev);
5287 
5288 	/*
5289 	 * This is racy, but way better than putting module outside of
5290 	 * the release callback.  The race window is pretty small, so
5291 	 * doing something similar to dm (dm-builtin.c) is overkill.
5292 	 */
5293 	if (need_put)
5294 		module_put(THIS_MODULE);
5295 }
5296 
5297 static struct rbd_device *__rbd_dev_create(struct rbd_client *rbdc,
5298 					   struct rbd_spec *spec)
5299 {
5300 	struct rbd_device *rbd_dev;
5301 
5302 	rbd_dev = kzalloc(sizeof(*rbd_dev), GFP_KERNEL);
5303 	if (!rbd_dev)
5304 		return NULL;
5305 
5306 	spin_lock_init(&rbd_dev->lock);
5307 	INIT_LIST_HEAD(&rbd_dev->node);
5308 	init_rwsem(&rbd_dev->header_rwsem);
5309 
5310 	rbd_dev->header.data_pool_id = CEPH_NOPOOL;
5311 	ceph_oid_init(&rbd_dev->header_oid);
5312 	rbd_dev->header_oloc.pool = spec->pool_id;
5313 	if (spec->pool_ns) {
5314 		WARN_ON(!*spec->pool_ns);
5315 		rbd_dev->header_oloc.pool_ns =
5316 		    ceph_find_or_create_string(spec->pool_ns,
5317 					       strlen(spec->pool_ns));
5318 	}
5319 
5320 	mutex_init(&rbd_dev->watch_mutex);
5321 	rbd_dev->watch_state = RBD_WATCH_STATE_UNREGISTERED;
5322 	INIT_DELAYED_WORK(&rbd_dev->watch_dwork, rbd_reregister_watch);
5323 
5324 	init_rwsem(&rbd_dev->lock_rwsem);
5325 	rbd_dev->lock_state = RBD_LOCK_STATE_UNLOCKED;
5326 	INIT_WORK(&rbd_dev->acquired_lock_work, rbd_notify_acquired_lock);
5327 	INIT_WORK(&rbd_dev->released_lock_work, rbd_notify_released_lock);
5328 	INIT_DELAYED_WORK(&rbd_dev->lock_dwork, rbd_acquire_lock);
5329 	INIT_WORK(&rbd_dev->unlock_work, rbd_release_lock_work);
5330 	spin_lock_init(&rbd_dev->lock_lists_lock);
5331 	INIT_LIST_HEAD(&rbd_dev->acquiring_list);
5332 	INIT_LIST_HEAD(&rbd_dev->running_list);
5333 	init_completion(&rbd_dev->acquire_wait);
5334 	init_completion(&rbd_dev->releasing_wait);
5335 
5336 	spin_lock_init(&rbd_dev->object_map_lock);
5337 
5338 	rbd_dev->dev.bus = &rbd_bus_type;
5339 	rbd_dev->dev.type = &rbd_device_type;
5340 	rbd_dev->dev.parent = &rbd_root_dev;
5341 	device_initialize(&rbd_dev->dev);
5342 
5343 	rbd_dev->rbd_client = rbdc;
5344 	rbd_dev->spec = spec;
5345 
5346 	return rbd_dev;
5347 }
5348 
5349 /*
5350  * Create a mapping rbd_dev.
5351  */
5352 static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc,
5353 					 struct rbd_spec *spec,
5354 					 struct rbd_options *opts)
5355 {
5356 	struct rbd_device *rbd_dev;
5357 
5358 	rbd_dev = __rbd_dev_create(rbdc, spec);
5359 	if (!rbd_dev)
5360 		return NULL;
5361 
5362 	rbd_dev->opts = opts;
5363 
5364 	/* get an id and fill in device name */
5365 	rbd_dev->dev_id = ida_simple_get(&rbd_dev_id_ida, 0,
5366 					 minor_to_rbd_dev_id(1 << MINORBITS),
5367 					 GFP_KERNEL);
5368 	if (rbd_dev->dev_id < 0)
5369 		goto fail_rbd_dev;
5370 
5371 	sprintf(rbd_dev->name, RBD_DRV_NAME "%d", rbd_dev->dev_id);
5372 	rbd_dev->task_wq = alloc_ordered_workqueue("%s-tasks", WQ_MEM_RECLAIM,
5373 						   rbd_dev->name);
5374 	if (!rbd_dev->task_wq)
5375 		goto fail_dev_id;
5376 
5377 	/* we have a ref from do_rbd_add() */
5378 	__module_get(THIS_MODULE);
5379 
5380 	dout("%s rbd_dev %p dev_id %d\n", __func__, rbd_dev, rbd_dev->dev_id);
5381 	return rbd_dev;
5382 
5383 fail_dev_id:
5384 	ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id);
5385 fail_rbd_dev:
5386 	rbd_dev_free(rbd_dev);
5387 	return NULL;
5388 }
5389 
5390 static void rbd_dev_destroy(struct rbd_device *rbd_dev)
5391 {
5392 	if (rbd_dev)
5393 		put_device(&rbd_dev->dev);
5394 }
5395 
5396 /*
5397  * Get the size and object order for an image snapshot, or if
5398  * snap_id is CEPH_NOSNAP, gets this information for the base
5399  * image.
5400  */
5401 static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
5402 				u8 *order, u64 *snap_size)
5403 {
5404 	__le64 snapid = cpu_to_le64(snap_id);
5405 	int ret;
5406 	struct {
5407 		u8 order;
5408 		__le64 size;
5409 	} __attribute__ ((packed)) size_buf = { 0 };
5410 
5411 	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
5412 				  &rbd_dev->header_oloc, "get_size",
5413 				  &snapid, sizeof(snapid),
5414 				  &size_buf, sizeof(size_buf));
5415 	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
5416 	if (ret < 0)
5417 		return ret;
5418 	if (ret < sizeof (size_buf))
5419 		return -ERANGE;
5420 
5421 	if (order) {
5422 		*order = size_buf.order;
5423 		dout("  order %u", (unsigned int)*order);
5424 	}
5425 	*snap_size = le64_to_cpu(size_buf.size);
5426 
5427 	dout("  snap_id 0x%016llx snap_size = %llu\n",
5428 		(unsigned long long)snap_id,
5429 		(unsigned long long)*snap_size);
5430 
5431 	return 0;
5432 }
5433 
5434 static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev)
5435 {
5436 	return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP,
5437 					&rbd_dev->header.obj_order,
5438 					&rbd_dev->header.image_size);
5439 }
5440 
5441 static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev)
5442 {
5443 	size_t size;
5444 	void *reply_buf;
5445 	int ret;
5446 	void *p;
5447 
5448 	/* Response will be an encoded string, which includes a length */
5449 	size = sizeof(__le32) + RBD_OBJ_PREFIX_LEN_MAX;
5450 	reply_buf = kzalloc(size, GFP_KERNEL);
5451 	if (!reply_buf)
5452 		return -ENOMEM;
5453 
5454 	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
5455 				  &rbd_dev->header_oloc, "get_object_prefix",
5456 				  NULL, 0, reply_buf, size);
5457 	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
5458 	if (ret < 0)
5459 		goto out;
5460 
5461 	p = reply_buf;
5462 	rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p,
5463 						p + ret, NULL, GFP_NOIO);
5464 	ret = 0;
5465 
5466 	if (IS_ERR(rbd_dev->header.object_prefix)) {
5467 		ret = PTR_ERR(rbd_dev->header.object_prefix);
5468 		rbd_dev->header.object_prefix = NULL;
5469 	} else {
5470 		dout("  object_prefix = %s\n", rbd_dev->header.object_prefix);
5471 	}
5472 out:
5473 	kfree(reply_buf);
5474 
5475 	return ret;
5476 }
5477 
5478 static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
5479 				     bool read_only, u64 *snap_features)
5480 {
5481 	struct {
5482 		__le64 snap_id;
5483 		u8 read_only;
5484 	} features_in;
5485 	struct {
5486 		__le64 features;
5487 		__le64 incompat;
5488 	} __attribute__ ((packed)) features_buf = { 0 };
5489 	u64 unsup;
5490 	int ret;
5491 
5492 	features_in.snap_id = cpu_to_le64(snap_id);
5493 	features_in.read_only = read_only;
5494 
5495 	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
5496 				  &rbd_dev->header_oloc, "get_features",
5497 				  &features_in, sizeof(features_in),
5498 				  &features_buf, sizeof(features_buf));
5499 	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
5500 	if (ret < 0)
5501 		return ret;
5502 	if (ret < sizeof (features_buf))
5503 		return -ERANGE;
5504 
5505 	unsup = le64_to_cpu(features_buf.incompat) & ~RBD_FEATURES_SUPPORTED;
5506 	if (unsup) {
5507 		rbd_warn(rbd_dev, "image uses unsupported features: 0x%llx",
5508 			 unsup);
5509 		return -ENXIO;
5510 	}
5511 
5512 	*snap_features = le64_to_cpu(features_buf.features);
5513 
5514 	dout("  snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
5515 		(unsigned long long)snap_id,
5516 		(unsigned long long)*snap_features,
5517 		(unsigned long long)le64_to_cpu(features_buf.incompat));
5518 
5519 	return 0;
5520 }
5521 
5522 static int rbd_dev_v2_features(struct rbd_device *rbd_dev)
5523 {
5524 	return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP,
5525 					 rbd_is_ro(rbd_dev),
5526 					 &rbd_dev->header.features);
5527 }
5528 
5529 /*
5530  * These are generic image flags, but since they are used only for
5531  * object map, store them in rbd_dev->object_map_flags.
5532  *
5533  * For the same reason, this function is called only on object map
5534  * (re)load and not on header refresh.
5535  */
5536 static int rbd_dev_v2_get_flags(struct rbd_device *rbd_dev)
5537 {
5538 	__le64 snapid = cpu_to_le64(rbd_dev->spec->snap_id);
5539 	__le64 flags;
5540 	int ret;
5541 
5542 	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
5543 				  &rbd_dev->header_oloc, "get_flags",
5544 				  &snapid, sizeof(snapid),
5545 				  &flags, sizeof(flags));
5546 	if (ret < 0)
5547 		return ret;
5548 	if (ret < sizeof(flags))
5549 		return -EBADMSG;
5550 
5551 	rbd_dev->object_map_flags = le64_to_cpu(flags);
5552 	return 0;
5553 }
5554 
5555 struct parent_image_info {
5556 	u64		pool_id;
5557 	const char	*pool_ns;
5558 	const char	*image_id;
5559 	u64		snap_id;
5560 
5561 	bool		has_overlap;
5562 	u64		overlap;
5563 };
5564 
5565 /*
5566  * The caller is responsible for @pii.
5567  */
5568 static int decode_parent_image_spec(void **p, void *end,
5569 				    struct parent_image_info *pii)
5570 {
5571 	u8 struct_v;
5572 	u32 struct_len;
5573 	int ret;
5574 
5575 	ret = ceph_start_decoding(p, end, 1, "ParentImageSpec",
5576 				  &struct_v, &struct_len);
5577 	if (ret)
5578 		return ret;
5579 
5580 	ceph_decode_64_safe(p, end, pii->pool_id, e_inval);
5581 	pii->pool_ns = ceph_extract_encoded_string(p, end, NULL, GFP_KERNEL);
5582 	if (IS_ERR(pii->pool_ns)) {
5583 		ret = PTR_ERR(pii->pool_ns);
5584 		pii->pool_ns = NULL;
5585 		return ret;
5586 	}
5587 	pii->image_id = ceph_extract_encoded_string(p, end, NULL, GFP_KERNEL);
5588 	if (IS_ERR(pii->image_id)) {
5589 		ret = PTR_ERR(pii->image_id);
5590 		pii->image_id = NULL;
5591 		return ret;
5592 	}
5593 	ceph_decode_64_safe(p, end, pii->snap_id, e_inval);
5594 	return 0;
5595 
5596 e_inval:
5597 	return -EINVAL;
5598 }
5599 
5600 static int __get_parent_info(struct rbd_device *rbd_dev,
5601 			     struct page *req_page,
5602 			     struct page *reply_page,
5603 			     struct parent_image_info *pii)
5604 {
5605 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
5606 	size_t reply_len = PAGE_SIZE;
5607 	void *p, *end;
5608 	int ret;
5609 
5610 	ret = ceph_osdc_call(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc,
5611 			     "rbd", "parent_get", CEPH_OSD_FLAG_READ,
5612 			     req_page, sizeof(u64), &reply_page, &reply_len);
5613 	if (ret)
5614 		return ret == -EOPNOTSUPP ? 1 : ret;
5615 
5616 	p = page_address(reply_page);
5617 	end = p + reply_len;
5618 	ret = decode_parent_image_spec(&p, end, pii);
5619 	if (ret)
5620 		return ret;
5621 
5622 	ret = ceph_osdc_call(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc,
5623 			     "rbd", "parent_overlap_get", CEPH_OSD_FLAG_READ,
5624 			     req_page, sizeof(u64), &reply_page, &reply_len);
5625 	if (ret)
5626 		return ret;
5627 
5628 	p = page_address(reply_page);
5629 	end = p + reply_len;
5630 	ceph_decode_8_safe(&p, end, pii->has_overlap, e_inval);
5631 	if (pii->has_overlap)
5632 		ceph_decode_64_safe(&p, end, pii->overlap, e_inval);
5633 
5634 	return 0;
5635 
5636 e_inval:
5637 	return -EINVAL;
5638 }
5639 
5640 /*
5641  * The caller is responsible for @pii.
5642  */
5643 static int __get_parent_info_legacy(struct rbd_device *rbd_dev,
5644 				    struct page *req_page,
5645 				    struct page *reply_page,
5646 				    struct parent_image_info *pii)
5647 {
5648 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
5649 	size_t reply_len = PAGE_SIZE;
5650 	void *p, *end;
5651 	int ret;
5652 
5653 	ret = ceph_osdc_call(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc,
5654 			     "rbd", "get_parent", CEPH_OSD_FLAG_READ,
5655 			     req_page, sizeof(u64), &reply_page, &reply_len);
5656 	if (ret)
5657 		return ret;
5658 
5659 	p = page_address(reply_page);
5660 	end = p + reply_len;
5661 	ceph_decode_64_safe(&p, end, pii->pool_id, e_inval);
5662 	pii->image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
5663 	if (IS_ERR(pii->image_id)) {
5664 		ret = PTR_ERR(pii->image_id);
5665 		pii->image_id = NULL;
5666 		return ret;
5667 	}
5668 	ceph_decode_64_safe(&p, end, pii->snap_id, e_inval);
5669 	pii->has_overlap = true;
5670 	ceph_decode_64_safe(&p, end, pii->overlap, e_inval);
5671 
5672 	return 0;
5673 
5674 e_inval:
5675 	return -EINVAL;
5676 }
5677 
5678 static int get_parent_info(struct rbd_device *rbd_dev,
5679 			   struct parent_image_info *pii)
5680 {
5681 	struct page *req_page, *reply_page;
5682 	void *p;
5683 	int ret;
5684 
5685 	req_page = alloc_page(GFP_KERNEL);
5686 	if (!req_page)
5687 		return -ENOMEM;
5688 
5689 	reply_page = alloc_page(GFP_KERNEL);
5690 	if (!reply_page) {
5691 		__free_page(req_page);
5692 		return -ENOMEM;
5693 	}
5694 
5695 	p = page_address(req_page);
5696 	ceph_encode_64(&p, rbd_dev->spec->snap_id);
5697 	ret = __get_parent_info(rbd_dev, req_page, reply_page, pii);
5698 	if (ret > 0)
5699 		ret = __get_parent_info_legacy(rbd_dev, req_page, reply_page,
5700 					       pii);
5701 
5702 	__free_page(req_page);
5703 	__free_page(reply_page);
5704 	return ret;
5705 }
5706 
5707 static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev)
5708 {
5709 	struct rbd_spec *parent_spec;
5710 	struct parent_image_info pii = { 0 };
5711 	int ret;
5712 
5713 	parent_spec = rbd_spec_alloc();
5714 	if (!parent_spec)
5715 		return -ENOMEM;
5716 
5717 	ret = get_parent_info(rbd_dev, &pii);
5718 	if (ret)
5719 		goto out_err;
5720 
5721 	dout("%s pool_id %llu pool_ns %s image_id %s snap_id %llu has_overlap %d overlap %llu\n",
5722 	     __func__, pii.pool_id, pii.pool_ns, pii.image_id, pii.snap_id,
5723 	     pii.has_overlap, pii.overlap);
5724 
5725 	if (pii.pool_id == CEPH_NOPOOL || !pii.has_overlap) {
5726 		/*
5727 		 * Either the parent never existed, or we have
5728 		 * record of it but the image got flattened so it no
5729 		 * longer has a parent.  When the parent of a
5730 		 * layered image disappears we immediately set the
5731 		 * overlap to 0.  The effect of this is that all new
5732 		 * requests will be treated as if the image had no
5733 		 * parent.
5734 		 *
5735 		 * If !pii.has_overlap, the parent image spec is not
5736 		 * applicable.  It's there to avoid duplication in each
5737 		 * snapshot record.
5738 		 */
5739 		if (rbd_dev->parent_overlap) {
5740 			rbd_dev->parent_overlap = 0;
5741 			rbd_dev_parent_put(rbd_dev);
5742 			pr_info("%s: clone image has been flattened\n",
5743 				rbd_dev->disk->disk_name);
5744 		}
5745 
5746 		goto out;	/* No parent?  No problem. */
5747 	}
5748 
5749 	/* The ceph file layout needs to fit pool id in 32 bits */
5750 
5751 	ret = -EIO;
5752 	if (pii.pool_id > (u64)U32_MAX) {
5753 		rbd_warn(NULL, "parent pool id too large (%llu > %u)",
5754 			(unsigned long long)pii.pool_id, U32_MAX);
5755 		goto out_err;
5756 	}
5757 
5758 	/*
5759 	 * The parent won't change (except when the clone is
5760 	 * flattened, already handled that).  So we only need to
5761 	 * record the parent spec we have not already done so.
5762 	 */
5763 	if (!rbd_dev->parent_spec) {
5764 		parent_spec->pool_id = pii.pool_id;
5765 		if (pii.pool_ns && *pii.pool_ns) {
5766 			parent_spec->pool_ns = pii.pool_ns;
5767 			pii.pool_ns = NULL;
5768 		}
5769 		parent_spec->image_id = pii.image_id;
5770 		pii.image_id = NULL;
5771 		parent_spec->snap_id = pii.snap_id;
5772 
5773 		rbd_dev->parent_spec = parent_spec;
5774 		parent_spec = NULL;	/* rbd_dev now owns this */
5775 	}
5776 
5777 	/*
5778 	 * We always update the parent overlap.  If it's zero we issue
5779 	 * a warning, as we will proceed as if there was no parent.
5780 	 */
5781 	if (!pii.overlap) {
5782 		if (parent_spec) {
5783 			/* refresh, careful to warn just once */
5784 			if (rbd_dev->parent_overlap)
5785 				rbd_warn(rbd_dev,
5786 				    "clone now standalone (overlap became 0)");
5787 		} else {
5788 			/* initial probe */
5789 			rbd_warn(rbd_dev, "clone is standalone (overlap 0)");
5790 		}
5791 	}
5792 	rbd_dev->parent_overlap = pii.overlap;
5793 
5794 out:
5795 	ret = 0;
5796 out_err:
5797 	kfree(pii.pool_ns);
5798 	kfree(pii.image_id);
5799 	rbd_spec_put(parent_spec);
5800 	return ret;
5801 }
5802 
5803 static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev)
5804 {
5805 	struct {
5806 		__le64 stripe_unit;
5807 		__le64 stripe_count;
5808 	} __attribute__ ((packed)) striping_info_buf = { 0 };
5809 	size_t size = sizeof (striping_info_buf);
5810 	void *p;
5811 	int ret;
5812 
5813 	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
5814 				&rbd_dev->header_oloc, "get_stripe_unit_count",
5815 				NULL, 0, &striping_info_buf, size);
5816 	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
5817 	if (ret < 0)
5818 		return ret;
5819 	if (ret < size)
5820 		return -ERANGE;
5821 
5822 	p = &striping_info_buf;
5823 	rbd_dev->header.stripe_unit = ceph_decode_64(&p);
5824 	rbd_dev->header.stripe_count = ceph_decode_64(&p);
5825 	return 0;
5826 }
5827 
5828 static int rbd_dev_v2_data_pool(struct rbd_device *rbd_dev)
5829 {
5830 	__le64 data_pool_id;
5831 	int ret;
5832 
5833 	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
5834 				  &rbd_dev->header_oloc, "get_data_pool",
5835 				  NULL, 0, &data_pool_id, sizeof(data_pool_id));
5836 	if (ret < 0)
5837 		return ret;
5838 	if (ret < sizeof(data_pool_id))
5839 		return -EBADMSG;
5840 
5841 	rbd_dev->header.data_pool_id = le64_to_cpu(data_pool_id);
5842 	WARN_ON(rbd_dev->header.data_pool_id == CEPH_NOPOOL);
5843 	return 0;
5844 }
5845 
5846 static char *rbd_dev_image_name(struct rbd_device *rbd_dev)
5847 {
5848 	CEPH_DEFINE_OID_ONSTACK(oid);
5849 	size_t image_id_size;
5850 	char *image_id;
5851 	void *p;
5852 	void *end;
5853 	size_t size;
5854 	void *reply_buf = NULL;
5855 	size_t len = 0;
5856 	char *image_name = NULL;
5857 	int ret;
5858 
5859 	rbd_assert(!rbd_dev->spec->image_name);
5860 
5861 	len = strlen(rbd_dev->spec->image_id);
5862 	image_id_size = sizeof (__le32) + len;
5863 	image_id = kmalloc(image_id_size, GFP_KERNEL);
5864 	if (!image_id)
5865 		return NULL;
5866 
5867 	p = image_id;
5868 	end = image_id + image_id_size;
5869 	ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len);
5870 
5871 	size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX;
5872 	reply_buf = kmalloc(size, GFP_KERNEL);
5873 	if (!reply_buf)
5874 		goto out;
5875 
5876 	ceph_oid_printf(&oid, "%s", RBD_DIRECTORY);
5877 	ret = rbd_obj_method_sync(rbd_dev, &oid, &rbd_dev->header_oloc,
5878 				  "dir_get_name", image_id, image_id_size,
5879 				  reply_buf, size);
5880 	if (ret < 0)
5881 		goto out;
5882 	p = reply_buf;
5883 	end = reply_buf + ret;
5884 
5885 	image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL);
5886 	if (IS_ERR(image_name))
5887 		image_name = NULL;
5888 	else
5889 		dout("%s: name is %s len is %zd\n", __func__, image_name, len);
5890 out:
5891 	kfree(reply_buf);
5892 	kfree(image_id);
5893 
5894 	return image_name;
5895 }
5896 
5897 static u64 rbd_v1_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
5898 {
5899 	struct ceph_snap_context *snapc = rbd_dev->header.snapc;
5900 	const char *snap_name;
5901 	u32 which = 0;
5902 
5903 	/* Skip over names until we find the one we are looking for */
5904 
5905 	snap_name = rbd_dev->header.snap_names;
5906 	while (which < snapc->num_snaps) {
5907 		if (!strcmp(name, snap_name))
5908 			return snapc->snaps[which];
5909 		snap_name += strlen(snap_name) + 1;
5910 		which++;
5911 	}
5912 	return CEPH_NOSNAP;
5913 }
5914 
5915 static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
5916 {
5917 	struct ceph_snap_context *snapc = rbd_dev->header.snapc;
5918 	u32 which;
5919 	bool found = false;
5920 	u64 snap_id;
5921 
5922 	for (which = 0; !found && which < snapc->num_snaps; which++) {
5923 		const char *snap_name;
5924 
5925 		snap_id = snapc->snaps[which];
5926 		snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id);
5927 		if (IS_ERR(snap_name)) {
5928 			/* ignore no-longer existing snapshots */
5929 			if (PTR_ERR(snap_name) == -ENOENT)
5930 				continue;
5931 			else
5932 				break;
5933 		}
5934 		found = !strcmp(name, snap_name);
5935 		kfree(snap_name);
5936 	}
5937 	return found ? snap_id : CEPH_NOSNAP;
5938 }
5939 
5940 /*
5941  * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if
5942  * no snapshot by that name is found, or if an error occurs.
5943  */
5944 static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
5945 {
5946 	if (rbd_dev->image_format == 1)
5947 		return rbd_v1_snap_id_by_name(rbd_dev, name);
5948 
5949 	return rbd_v2_snap_id_by_name(rbd_dev, name);
5950 }
5951 
5952 /*
5953  * An image being mapped will have everything but the snap id.
5954  */
5955 static int rbd_spec_fill_snap_id(struct rbd_device *rbd_dev)
5956 {
5957 	struct rbd_spec *spec = rbd_dev->spec;
5958 
5959 	rbd_assert(spec->pool_id != CEPH_NOPOOL && spec->pool_name);
5960 	rbd_assert(spec->image_id && spec->image_name);
5961 	rbd_assert(spec->snap_name);
5962 
5963 	if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) {
5964 		u64 snap_id;
5965 
5966 		snap_id = rbd_snap_id_by_name(rbd_dev, spec->snap_name);
5967 		if (snap_id == CEPH_NOSNAP)
5968 			return -ENOENT;
5969 
5970 		spec->snap_id = snap_id;
5971 	} else {
5972 		spec->snap_id = CEPH_NOSNAP;
5973 	}
5974 
5975 	return 0;
5976 }
5977 
5978 /*
5979  * A parent image will have all ids but none of the names.
5980  *
5981  * All names in an rbd spec are dynamically allocated.  It's OK if we
5982  * can't figure out the name for an image id.
5983  */
5984 static int rbd_spec_fill_names(struct rbd_device *rbd_dev)
5985 {
5986 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
5987 	struct rbd_spec *spec = rbd_dev->spec;
5988 	const char *pool_name;
5989 	const char *image_name;
5990 	const char *snap_name;
5991 	int ret;
5992 
5993 	rbd_assert(spec->pool_id != CEPH_NOPOOL);
5994 	rbd_assert(spec->image_id);
5995 	rbd_assert(spec->snap_id != CEPH_NOSNAP);
5996 
5997 	/* Get the pool name; we have to make our own copy of this */
5998 
5999 	pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id);
6000 	if (!pool_name) {
6001 		rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id);
6002 		return -EIO;
6003 	}
6004 	pool_name = kstrdup(pool_name, GFP_KERNEL);
6005 	if (!pool_name)
6006 		return -ENOMEM;
6007 
6008 	/* Fetch the image name; tolerate failure here */
6009 
6010 	image_name = rbd_dev_image_name(rbd_dev);
6011 	if (!image_name)
6012 		rbd_warn(rbd_dev, "unable to get image name");
6013 
6014 	/* Fetch the snapshot name */
6015 
6016 	snap_name = rbd_snap_name(rbd_dev, spec->snap_id);
6017 	if (IS_ERR(snap_name)) {
6018 		ret = PTR_ERR(snap_name);
6019 		goto out_err;
6020 	}
6021 
6022 	spec->pool_name = pool_name;
6023 	spec->image_name = image_name;
6024 	spec->snap_name = snap_name;
6025 
6026 	return 0;
6027 
6028 out_err:
6029 	kfree(image_name);
6030 	kfree(pool_name);
6031 	return ret;
6032 }
6033 
6034 static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev)
6035 {
6036 	size_t size;
6037 	int ret;
6038 	void *reply_buf;
6039 	void *p;
6040 	void *end;
6041 	u64 seq;
6042 	u32 snap_count;
6043 	struct ceph_snap_context *snapc;
6044 	u32 i;
6045 
6046 	/*
6047 	 * We'll need room for the seq value (maximum snapshot id),
6048 	 * snapshot count, and array of that many snapshot ids.
6049 	 * For now we have a fixed upper limit on the number we're
6050 	 * prepared to receive.
6051 	 */
6052 	size = sizeof (__le64) + sizeof (__le32) +
6053 			RBD_MAX_SNAP_COUNT * sizeof (__le64);
6054 	reply_buf = kzalloc(size, GFP_KERNEL);
6055 	if (!reply_buf)
6056 		return -ENOMEM;
6057 
6058 	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
6059 				  &rbd_dev->header_oloc, "get_snapcontext",
6060 				  NULL, 0, reply_buf, size);
6061 	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
6062 	if (ret < 0)
6063 		goto out;
6064 
6065 	p = reply_buf;
6066 	end = reply_buf + ret;
6067 	ret = -ERANGE;
6068 	ceph_decode_64_safe(&p, end, seq, out);
6069 	ceph_decode_32_safe(&p, end, snap_count, out);
6070 
6071 	/*
6072 	 * Make sure the reported number of snapshot ids wouldn't go
6073 	 * beyond the end of our buffer.  But before checking that,
6074 	 * make sure the computed size of the snapshot context we
6075 	 * allocate is representable in a size_t.
6076 	 */
6077 	if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context))
6078 				 / sizeof (u64)) {
6079 		ret = -EINVAL;
6080 		goto out;
6081 	}
6082 	if (!ceph_has_room(&p, end, snap_count * sizeof (__le64)))
6083 		goto out;
6084 	ret = 0;
6085 
6086 	snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
6087 	if (!snapc) {
6088 		ret = -ENOMEM;
6089 		goto out;
6090 	}
6091 	snapc->seq = seq;
6092 	for (i = 0; i < snap_count; i++)
6093 		snapc->snaps[i] = ceph_decode_64(&p);
6094 
6095 	ceph_put_snap_context(rbd_dev->header.snapc);
6096 	rbd_dev->header.snapc = snapc;
6097 
6098 	dout("  snap context seq = %llu, snap_count = %u\n",
6099 		(unsigned long long)seq, (unsigned int)snap_count);
6100 out:
6101 	kfree(reply_buf);
6102 
6103 	return ret;
6104 }
6105 
6106 static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
6107 					u64 snap_id)
6108 {
6109 	size_t size;
6110 	void *reply_buf;
6111 	__le64 snapid;
6112 	int ret;
6113 	void *p;
6114 	void *end;
6115 	char *snap_name;
6116 
6117 	size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN;
6118 	reply_buf = kmalloc(size, GFP_KERNEL);
6119 	if (!reply_buf)
6120 		return ERR_PTR(-ENOMEM);
6121 
6122 	snapid = cpu_to_le64(snap_id);
6123 	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
6124 				  &rbd_dev->header_oloc, "get_snapshot_name",
6125 				  &snapid, sizeof(snapid), reply_buf, size);
6126 	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
6127 	if (ret < 0) {
6128 		snap_name = ERR_PTR(ret);
6129 		goto out;
6130 	}
6131 
6132 	p = reply_buf;
6133 	end = reply_buf + ret;
6134 	snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
6135 	if (IS_ERR(snap_name))
6136 		goto out;
6137 
6138 	dout("  snap_id 0x%016llx snap_name = %s\n",
6139 		(unsigned long long)snap_id, snap_name);
6140 out:
6141 	kfree(reply_buf);
6142 
6143 	return snap_name;
6144 }
6145 
6146 static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev)
6147 {
6148 	bool first_time = rbd_dev->header.object_prefix == NULL;
6149 	int ret;
6150 
6151 	ret = rbd_dev_v2_image_size(rbd_dev);
6152 	if (ret)
6153 		return ret;
6154 
6155 	if (first_time) {
6156 		ret = rbd_dev_v2_header_onetime(rbd_dev);
6157 		if (ret)
6158 			return ret;
6159 	}
6160 
6161 	ret = rbd_dev_v2_snap_context(rbd_dev);
6162 	if (ret && first_time) {
6163 		kfree(rbd_dev->header.object_prefix);
6164 		rbd_dev->header.object_prefix = NULL;
6165 	}
6166 
6167 	return ret;
6168 }
6169 
6170 static int rbd_dev_header_info(struct rbd_device *rbd_dev)
6171 {
6172 	rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
6173 
6174 	if (rbd_dev->image_format == 1)
6175 		return rbd_dev_v1_header_info(rbd_dev);
6176 
6177 	return rbd_dev_v2_header_info(rbd_dev);
6178 }
6179 
6180 /*
6181  * Skips over white space at *buf, and updates *buf to point to the
6182  * first found non-space character (if any). Returns the length of
6183  * the token (string of non-white space characters) found.  Note
6184  * that *buf must be terminated with '\0'.
6185  */
6186 static inline size_t next_token(const char **buf)
6187 {
6188         /*
6189         * These are the characters that produce nonzero for
6190         * isspace() in the "C" and "POSIX" locales.
6191         */
6192 	static const char spaces[] = " \f\n\r\t\v";
6193 
6194         *buf += strspn(*buf, spaces);	/* Find start of token */
6195 
6196 	return strcspn(*buf, spaces);   /* Return token length */
6197 }
6198 
6199 /*
6200  * Finds the next token in *buf, dynamically allocates a buffer big
6201  * enough to hold a copy of it, and copies the token into the new
6202  * buffer.  The copy is guaranteed to be terminated with '\0'.  Note
6203  * that a duplicate buffer is created even for a zero-length token.
6204  *
6205  * Returns a pointer to the newly-allocated duplicate, or a null
6206  * pointer if memory for the duplicate was not available.  If
6207  * the lenp argument is a non-null pointer, the length of the token
6208  * (not including the '\0') is returned in *lenp.
6209  *
6210  * If successful, the *buf pointer will be updated to point beyond
6211  * the end of the found token.
6212  *
6213  * Note: uses GFP_KERNEL for allocation.
6214  */
6215 static inline char *dup_token(const char **buf, size_t *lenp)
6216 {
6217 	char *dup;
6218 	size_t len;
6219 
6220 	len = next_token(buf);
6221 	dup = kmemdup(*buf, len + 1, GFP_KERNEL);
6222 	if (!dup)
6223 		return NULL;
6224 	*(dup + len) = '\0';
6225 	*buf += len;
6226 
6227 	if (lenp)
6228 		*lenp = len;
6229 
6230 	return dup;
6231 }
6232 
6233 static int rbd_parse_param(struct fs_parameter *param,
6234 			    struct rbd_parse_opts_ctx *pctx)
6235 {
6236 	struct rbd_options *opt = pctx->opts;
6237 	struct fs_parse_result result;
6238 	struct p_log log = {.prefix = "rbd"};
6239 	int token, ret;
6240 
6241 	ret = ceph_parse_param(param, pctx->copts, NULL);
6242 	if (ret != -ENOPARAM)
6243 		return ret;
6244 
6245 	token = __fs_parse(&log, rbd_parameters, param, &result);
6246 	dout("%s fs_parse '%s' token %d\n", __func__, param->key, token);
6247 	if (token < 0) {
6248 		if (token == -ENOPARAM)
6249 			return inval_plog(&log, "Unknown parameter '%s'",
6250 					  param->key);
6251 		return token;
6252 	}
6253 
6254 	switch (token) {
6255 	case Opt_queue_depth:
6256 		if (result.uint_32 < 1)
6257 			goto out_of_range;
6258 		opt->queue_depth = result.uint_32;
6259 		break;
6260 	case Opt_alloc_size:
6261 		if (result.uint_32 < SECTOR_SIZE)
6262 			goto out_of_range;
6263 		if (!is_power_of_2(result.uint_32))
6264 			return inval_plog(&log, "alloc_size must be a power of 2");
6265 		opt->alloc_size = result.uint_32;
6266 		break;
6267 	case Opt_lock_timeout:
6268 		/* 0 is "wait forever" (i.e. infinite timeout) */
6269 		if (result.uint_32 > INT_MAX / 1000)
6270 			goto out_of_range;
6271 		opt->lock_timeout = msecs_to_jiffies(result.uint_32 * 1000);
6272 		break;
6273 	case Opt_pool_ns:
6274 		kfree(pctx->spec->pool_ns);
6275 		pctx->spec->pool_ns = param->string;
6276 		param->string = NULL;
6277 		break;
6278 	case Opt_compression_hint:
6279 		switch (result.uint_32) {
6280 		case Opt_compression_hint_none:
6281 			opt->alloc_hint_flags &=
6282 			    ~(CEPH_OSD_ALLOC_HINT_FLAG_COMPRESSIBLE |
6283 			      CEPH_OSD_ALLOC_HINT_FLAG_INCOMPRESSIBLE);
6284 			break;
6285 		case Opt_compression_hint_compressible:
6286 			opt->alloc_hint_flags |=
6287 			    CEPH_OSD_ALLOC_HINT_FLAG_COMPRESSIBLE;
6288 			opt->alloc_hint_flags &=
6289 			    ~CEPH_OSD_ALLOC_HINT_FLAG_INCOMPRESSIBLE;
6290 			break;
6291 		case Opt_compression_hint_incompressible:
6292 			opt->alloc_hint_flags |=
6293 			    CEPH_OSD_ALLOC_HINT_FLAG_INCOMPRESSIBLE;
6294 			opt->alloc_hint_flags &=
6295 			    ~CEPH_OSD_ALLOC_HINT_FLAG_COMPRESSIBLE;
6296 			break;
6297 		default:
6298 			BUG();
6299 		}
6300 		break;
6301 	case Opt_read_only:
6302 		opt->read_only = true;
6303 		break;
6304 	case Opt_read_write:
6305 		opt->read_only = false;
6306 		break;
6307 	case Opt_lock_on_read:
6308 		opt->lock_on_read = true;
6309 		break;
6310 	case Opt_exclusive:
6311 		opt->exclusive = true;
6312 		break;
6313 	case Opt_notrim:
6314 		opt->trim = false;
6315 		break;
6316 	default:
6317 		BUG();
6318 	}
6319 
6320 	return 0;
6321 
6322 out_of_range:
6323 	return inval_plog(&log, "%s out of range", param->key);
6324 }
6325 
6326 /*
6327  * This duplicates most of generic_parse_monolithic(), untying it from
6328  * fs_context and skipping standard superblock and security options.
6329  */
6330 static int rbd_parse_options(char *options, struct rbd_parse_opts_ctx *pctx)
6331 {
6332 	char *key;
6333 	int ret = 0;
6334 
6335 	dout("%s '%s'\n", __func__, options);
6336 	while ((key = strsep(&options, ",")) != NULL) {
6337 		if (*key) {
6338 			struct fs_parameter param = {
6339 				.key	= key,
6340 				.type	= fs_value_is_flag,
6341 			};
6342 			char *value = strchr(key, '=');
6343 			size_t v_len = 0;
6344 
6345 			if (value) {
6346 				if (value == key)
6347 					continue;
6348 				*value++ = 0;
6349 				v_len = strlen(value);
6350 				param.string = kmemdup_nul(value, v_len,
6351 							   GFP_KERNEL);
6352 				if (!param.string)
6353 					return -ENOMEM;
6354 				param.type = fs_value_is_string;
6355 			}
6356 			param.size = v_len;
6357 
6358 			ret = rbd_parse_param(&param, pctx);
6359 			kfree(param.string);
6360 			if (ret)
6361 				break;
6362 		}
6363 	}
6364 
6365 	return ret;
6366 }
6367 
6368 /*
6369  * Parse the options provided for an "rbd add" (i.e., rbd image
6370  * mapping) request.  These arrive via a write to /sys/bus/rbd/add,
6371  * and the data written is passed here via a NUL-terminated buffer.
6372  * Returns 0 if successful or an error code otherwise.
6373  *
6374  * The information extracted from these options is recorded in
6375  * the other parameters which return dynamically-allocated
6376  * structures:
6377  *  ceph_opts
6378  *      The address of a pointer that will refer to a ceph options
6379  *      structure.  Caller must release the returned pointer using
6380  *      ceph_destroy_options() when it is no longer needed.
6381  *  rbd_opts
6382  *	Address of an rbd options pointer.  Fully initialized by
6383  *	this function; caller must release with kfree().
6384  *  spec
6385  *	Address of an rbd image specification pointer.  Fully
6386  *	initialized by this function based on parsed options.
6387  *	Caller must release with rbd_spec_put().
6388  *
6389  * The options passed take this form:
6390  *  <mon_addrs> <options> <pool_name> <image_name> [<snap_id>]
6391  * where:
6392  *  <mon_addrs>
6393  *      A comma-separated list of one or more monitor addresses.
6394  *      A monitor address is an ip address, optionally followed
6395  *      by a port number (separated by a colon).
6396  *        I.e.:  ip1[:port1][,ip2[:port2]...]
6397  *  <options>
6398  *      A comma-separated list of ceph and/or rbd options.
6399  *  <pool_name>
6400  *      The name of the rados pool containing the rbd image.
6401  *  <image_name>
6402  *      The name of the image in that pool to map.
6403  *  <snap_id>
6404  *      An optional snapshot id.  If provided, the mapping will
6405  *      present data from the image at the time that snapshot was
6406  *      created.  The image head is used if no snapshot id is
6407  *      provided.  Snapshot mappings are always read-only.
6408  */
6409 static int rbd_add_parse_args(const char *buf,
6410 				struct ceph_options **ceph_opts,
6411 				struct rbd_options **opts,
6412 				struct rbd_spec **rbd_spec)
6413 {
6414 	size_t len;
6415 	char *options;
6416 	const char *mon_addrs;
6417 	char *snap_name;
6418 	size_t mon_addrs_size;
6419 	struct rbd_parse_opts_ctx pctx = { 0 };
6420 	int ret;
6421 
6422 	/* The first four tokens are required */
6423 
6424 	len = next_token(&buf);
6425 	if (!len) {
6426 		rbd_warn(NULL, "no monitor address(es) provided");
6427 		return -EINVAL;
6428 	}
6429 	mon_addrs = buf;
6430 	mon_addrs_size = len;
6431 	buf += len;
6432 
6433 	ret = -EINVAL;
6434 	options = dup_token(&buf, NULL);
6435 	if (!options)
6436 		return -ENOMEM;
6437 	if (!*options) {
6438 		rbd_warn(NULL, "no options provided");
6439 		goto out_err;
6440 	}
6441 
6442 	pctx.spec = rbd_spec_alloc();
6443 	if (!pctx.spec)
6444 		goto out_mem;
6445 
6446 	pctx.spec->pool_name = dup_token(&buf, NULL);
6447 	if (!pctx.spec->pool_name)
6448 		goto out_mem;
6449 	if (!*pctx.spec->pool_name) {
6450 		rbd_warn(NULL, "no pool name provided");
6451 		goto out_err;
6452 	}
6453 
6454 	pctx.spec->image_name = dup_token(&buf, NULL);
6455 	if (!pctx.spec->image_name)
6456 		goto out_mem;
6457 	if (!*pctx.spec->image_name) {
6458 		rbd_warn(NULL, "no image name provided");
6459 		goto out_err;
6460 	}
6461 
6462 	/*
6463 	 * Snapshot name is optional; default is to use "-"
6464 	 * (indicating the head/no snapshot).
6465 	 */
6466 	len = next_token(&buf);
6467 	if (!len) {
6468 		buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */
6469 		len = sizeof (RBD_SNAP_HEAD_NAME) - 1;
6470 	} else if (len > RBD_MAX_SNAP_NAME_LEN) {
6471 		ret = -ENAMETOOLONG;
6472 		goto out_err;
6473 	}
6474 	snap_name = kmemdup(buf, len + 1, GFP_KERNEL);
6475 	if (!snap_name)
6476 		goto out_mem;
6477 	*(snap_name + len) = '\0';
6478 	pctx.spec->snap_name = snap_name;
6479 
6480 	pctx.copts = ceph_alloc_options();
6481 	if (!pctx.copts)
6482 		goto out_mem;
6483 
6484 	/* Initialize all rbd options to the defaults */
6485 
6486 	pctx.opts = kzalloc(sizeof(*pctx.opts), GFP_KERNEL);
6487 	if (!pctx.opts)
6488 		goto out_mem;
6489 
6490 	pctx.opts->read_only = RBD_READ_ONLY_DEFAULT;
6491 	pctx.opts->queue_depth = RBD_QUEUE_DEPTH_DEFAULT;
6492 	pctx.opts->alloc_size = RBD_ALLOC_SIZE_DEFAULT;
6493 	pctx.opts->lock_timeout = RBD_LOCK_TIMEOUT_DEFAULT;
6494 	pctx.opts->lock_on_read = RBD_LOCK_ON_READ_DEFAULT;
6495 	pctx.opts->exclusive = RBD_EXCLUSIVE_DEFAULT;
6496 	pctx.opts->trim = RBD_TRIM_DEFAULT;
6497 
6498 	ret = ceph_parse_mon_ips(mon_addrs, mon_addrs_size, pctx.copts, NULL,
6499 				 ',');
6500 	if (ret)
6501 		goto out_err;
6502 
6503 	ret = rbd_parse_options(options, &pctx);
6504 	if (ret)
6505 		goto out_err;
6506 
6507 	*ceph_opts = pctx.copts;
6508 	*opts = pctx.opts;
6509 	*rbd_spec = pctx.spec;
6510 	kfree(options);
6511 	return 0;
6512 
6513 out_mem:
6514 	ret = -ENOMEM;
6515 out_err:
6516 	kfree(pctx.opts);
6517 	ceph_destroy_options(pctx.copts);
6518 	rbd_spec_put(pctx.spec);
6519 	kfree(options);
6520 	return ret;
6521 }
6522 
6523 static void rbd_dev_image_unlock(struct rbd_device *rbd_dev)
6524 {
6525 	down_write(&rbd_dev->lock_rwsem);
6526 	if (__rbd_is_lock_owner(rbd_dev))
6527 		__rbd_release_lock(rbd_dev);
6528 	up_write(&rbd_dev->lock_rwsem);
6529 }
6530 
6531 /*
6532  * If the wait is interrupted, an error is returned even if the lock
6533  * was successfully acquired.  rbd_dev_image_unlock() will release it
6534  * if needed.
6535  */
6536 static int rbd_add_acquire_lock(struct rbd_device *rbd_dev)
6537 {
6538 	long ret;
6539 
6540 	if (!(rbd_dev->header.features & RBD_FEATURE_EXCLUSIVE_LOCK)) {
6541 		if (!rbd_dev->opts->exclusive && !rbd_dev->opts->lock_on_read)
6542 			return 0;
6543 
6544 		rbd_warn(rbd_dev, "exclusive-lock feature is not enabled");
6545 		return -EINVAL;
6546 	}
6547 
6548 	if (rbd_is_ro(rbd_dev))
6549 		return 0;
6550 
6551 	rbd_assert(!rbd_is_lock_owner(rbd_dev));
6552 	queue_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork, 0);
6553 	ret = wait_for_completion_killable_timeout(&rbd_dev->acquire_wait,
6554 			    ceph_timeout_jiffies(rbd_dev->opts->lock_timeout));
6555 	if (ret > 0) {
6556 		ret = rbd_dev->acquire_err;
6557 	} else {
6558 		cancel_delayed_work_sync(&rbd_dev->lock_dwork);
6559 		if (!ret)
6560 			ret = -ETIMEDOUT;
6561 	}
6562 
6563 	if (ret) {
6564 		rbd_warn(rbd_dev, "failed to acquire exclusive lock: %ld", ret);
6565 		return ret;
6566 	}
6567 
6568 	/*
6569 	 * The lock may have been released by now, unless automatic lock
6570 	 * transitions are disabled.
6571 	 */
6572 	rbd_assert(!rbd_dev->opts->exclusive || rbd_is_lock_owner(rbd_dev));
6573 	return 0;
6574 }
6575 
6576 /*
6577  * An rbd format 2 image has a unique identifier, distinct from the
6578  * name given to it by the user.  Internally, that identifier is
6579  * what's used to specify the names of objects related to the image.
6580  *
6581  * A special "rbd id" object is used to map an rbd image name to its
6582  * id.  If that object doesn't exist, then there is no v2 rbd image
6583  * with the supplied name.
6584  *
6585  * This function will record the given rbd_dev's image_id field if
6586  * it can be determined, and in that case will return 0.  If any
6587  * errors occur a negative errno will be returned and the rbd_dev's
6588  * image_id field will be unchanged (and should be NULL).
6589  */
6590 static int rbd_dev_image_id(struct rbd_device *rbd_dev)
6591 {
6592 	int ret;
6593 	size_t size;
6594 	CEPH_DEFINE_OID_ONSTACK(oid);
6595 	void *response;
6596 	char *image_id;
6597 
6598 	/*
6599 	 * When probing a parent image, the image id is already
6600 	 * known (and the image name likely is not).  There's no
6601 	 * need to fetch the image id again in this case.  We
6602 	 * do still need to set the image format though.
6603 	 */
6604 	if (rbd_dev->spec->image_id) {
6605 		rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1;
6606 
6607 		return 0;
6608 	}
6609 
6610 	/*
6611 	 * First, see if the format 2 image id file exists, and if
6612 	 * so, get the image's persistent id from it.
6613 	 */
6614 	ret = ceph_oid_aprintf(&oid, GFP_KERNEL, "%s%s", RBD_ID_PREFIX,
6615 			       rbd_dev->spec->image_name);
6616 	if (ret)
6617 		return ret;
6618 
6619 	dout("rbd id object name is %s\n", oid.name);
6620 
6621 	/* Response will be an encoded string, which includes a length */
6622 	size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX;
6623 	response = kzalloc(size, GFP_NOIO);
6624 	if (!response) {
6625 		ret = -ENOMEM;
6626 		goto out;
6627 	}
6628 
6629 	/* If it doesn't exist we'll assume it's a format 1 image */
6630 
6631 	ret = rbd_obj_method_sync(rbd_dev, &oid, &rbd_dev->header_oloc,
6632 				  "get_id", NULL, 0,
6633 				  response, size);
6634 	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
6635 	if (ret == -ENOENT) {
6636 		image_id = kstrdup("", GFP_KERNEL);
6637 		ret = image_id ? 0 : -ENOMEM;
6638 		if (!ret)
6639 			rbd_dev->image_format = 1;
6640 	} else if (ret >= 0) {
6641 		void *p = response;
6642 
6643 		image_id = ceph_extract_encoded_string(&p, p + ret,
6644 						NULL, GFP_NOIO);
6645 		ret = PTR_ERR_OR_ZERO(image_id);
6646 		if (!ret)
6647 			rbd_dev->image_format = 2;
6648 	}
6649 
6650 	if (!ret) {
6651 		rbd_dev->spec->image_id = image_id;
6652 		dout("image_id is %s\n", image_id);
6653 	}
6654 out:
6655 	kfree(response);
6656 	ceph_oid_destroy(&oid);
6657 	return ret;
6658 }
6659 
6660 /*
6661  * Undo whatever state changes are made by v1 or v2 header info
6662  * call.
6663  */
6664 static void rbd_dev_unprobe(struct rbd_device *rbd_dev)
6665 {
6666 	struct rbd_image_header	*header;
6667 
6668 	rbd_dev_parent_put(rbd_dev);
6669 	rbd_object_map_free(rbd_dev);
6670 	rbd_dev_mapping_clear(rbd_dev);
6671 
6672 	/* Free dynamic fields from the header, then zero it out */
6673 
6674 	header = &rbd_dev->header;
6675 	ceph_put_snap_context(header->snapc);
6676 	kfree(header->snap_sizes);
6677 	kfree(header->snap_names);
6678 	kfree(header->object_prefix);
6679 	memset(header, 0, sizeof (*header));
6680 }
6681 
6682 static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev)
6683 {
6684 	int ret;
6685 
6686 	ret = rbd_dev_v2_object_prefix(rbd_dev);
6687 	if (ret)
6688 		goto out_err;
6689 
6690 	/*
6691 	 * Get the and check features for the image.  Currently the
6692 	 * features are assumed to never change.
6693 	 */
6694 	ret = rbd_dev_v2_features(rbd_dev);
6695 	if (ret)
6696 		goto out_err;
6697 
6698 	/* If the image supports fancy striping, get its parameters */
6699 
6700 	if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) {
6701 		ret = rbd_dev_v2_striping_info(rbd_dev);
6702 		if (ret < 0)
6703 			goto out_err;
6704 	}
6705 
6706 	if (rbd_dev->header.features & RBD_FEATURE_DATA_POOL) {
6707 		ret = rbd_dev_v2_data_pool(rbd_dev);
6708 		if (ret)
6709 			goto out_err;
6710 	}
6711 
6712 	rbd_init_layout(rbd_dev);
6713 	return 0;
6714 
6715 out_err:
6716 	rbd_dev->header.features = 0;
6717 	kfree(rbd_dev->header.object_prefix);
6718 	rbd_dev->header.object_prefix = NULL;
6719 	return ret;
6720 }
6721 
6722 /*
6723  * @depth is rbd_dev_image_probe() -> rbd_dev_probe_parent() ->
6724  * rbd_dev_image_probe() recursion depth, which means it's also the
6725  * length of the already discovered part of the parent chain.
6726  */
6727 static int rbd_dev_probe_parent(struct rbd_device *rbd_dev, int depth)
6728 {
6729 	struct rbd_device *parent = NULL;
6730 	int ret;
6731 
6732 	if (!rbd_dev->parent_spec)
6733 		return 0;
6734 
6735 	if (++depth > RBD_MAX_PARENT_CHAIN_LEN) {
6736 		pr_info("parent chain is too long (%d)\n", depth);
6737 		ret = -EINVAL;
6738 		goto out_err;
6739 	}
6740 
6741 	parent = __rbd_dev_create(rbd_dev->rbd_client, rbd_dev->parent_spec);
6742 	if (!parent) {
6743 		ret = -ENOMEM;
6744 		goto out_err;
6745 	}
6746 
6747 	/*
6748 	 * Images related by parent/child relationships always share
6749 	 * rbd_client and spec/parent_spec, so bump their refcounts.
6750 	 */
6751 	__rbd_get_client(rbd_dev->rbd_client);
6752 	rbd_spec_get(rbd_dev->parent_spec);
6753 
6754 	__set_bit(RBD_DEV_FLAG_READONLY, &parent->flags);
6755 
6756 	ret = rbd_dev_image_probe(parent, depth);
6757 	if (ret < 0)
6758 		goto out_err;
6759 
6760 	rbd_dev->parent = parent;
6761 	atomic_set(&rbd_dev->parent_ref, 1);
6762 	return 0;
6763 
6764 out_err:
6765 	rbd_dev_unparent(rbd_dev);
6766 	rbd_dev_destroy(parent);
6767 	return ret;
6768 }
6769 
6770 static void rbd_dev_device_release(struct rbd_device *rbd_dev)
6771 {
6772 	clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
6773 	rbd_free_disk(rbd_dev);
6774 	if (!single_major)
6775 		unregister_blkdev(rbd_dev->major, rbd_dev->name);
6776 }
6777 
6778 /*
6779  * rbd_dev->header_rwsem must be locked for write and will be unlocked
6780  * upon return.
6781  */
6782 static int rbd_dev_device_setup(struct rbd_device *rbd_dev)
6783 {
6784 	int ret;
6785 
6786 	/* Record our major and minor device numbers. */
6787 
6788 	if (!single_major) {
6789 		ret = register_blkdev(0, rbd_dev->name);
6790 		if (ret < 0)
6791 			goto err_out_unlock;
6792 
6793 		rbd_dev->major = ret;
6794 		rbd_dev->minor = 0;
6795 	} else {
6796 		rbd_dev->major = rbd_major;
6797 		rbd_dev->minor = rbd_dev_id_to_minor(rbd_dev->dev_id);
6798 	}
6799 
6800 	/* Set up the blkdev mapping. */
6801 
6802 	ret = rbd_init_disk(rbd_dev);
6803 	if (ret)
6804 		goto err_out_blkdev;
6805 
6806 	set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE);
6807 	set_disk_ro(rbd_dev->disk, rbd_is_ro(rbd_dev));
6808 
6809 	ret = dev_set_name(&rbd_dev->dev, "%d", rbd_dev->dev_id);
6810 	if (ret)
6811 		goto err_out_disk;
6812 
6813 	set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
6814 	up_write(&rbd_dev->header_rwsem);
6815 	return 0;
6816 
6817 err_out_disk:
6818 	rbd_free_disk(rbd_dev);
6819 err_out_blkdev:
6820 	if (!single_major)
6821 		unregister_blkdev(rbd_dev->major, rbd_dev->name);
6822 err_out_unlock:
6823 	up_write(&rbd_dev->header_rwsem);
6824 	return ret;
6825 }
6826 
6827 static int rbd_dev_header_name(struct rbd_device *rbd_dev)
6828 {
6829 	struct rbd_spec *spec = rbd_dev->spec;
6830 	int ret;
6831 
6832 	/* Record the header object name for this rbd image. */
6833 
6834 	rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
6835 	if (rbd_dev->image_format == 1)
6836 		ret = ceph_oid_aprintf(&rbd_dev->header_oid, GFP_KERNEL, "%s%s",
6837 				       spec->image_name, RBD_SUFFIX);
6838 	else
6839 		ret = ceph_oid_aprintf(&rbd_dev->header_oid, GFP_KERNEL, "%s%s",
6840 				       RBD_HEADER_PREFIX, spec->image_id);
6841 
6842 	return ret;
6843 }
6844 
6845 static void rbd_print_dne(struct rbd_device *rbd_dev, bool is_snap)
6846 {
6847 	if (!is_snap) {
6848 		pr_info("image %s/%s%s%s does not exist\n",
6849 			rbd_dev->spec->pool_name,
6850 			rbd_dev->spec->pool_ns ?: "",
6851 			rbd_dev->spec->pool_ns ? "/" : "",
6852 			rbd_dev->spec->image_name);
6853 	} else {
6854 		pr_info("snap %s/%s%s%s@%s does not exist\n",
6855 			rbd_dev->spec->pool_name,
6856 			rbd_dev->spec->pool_ns ?: "",
6857 			rbd_dev->spec->pool_ns ? "/" : "",
6858 			rbd_dev->spec->image_name,
6859 			rbd_dev->spec->snap_name);
6860 	}
6861 }
6862 
6863 static void rbd_dev_image_release(struct rbd_device *rbd_dev)
6864 {
6865 	if (!rbd_is_ro(rbd_dev))
6866 		rbd_unregister_watch(rbd_dev);
6867 
6868 	rbd_dev_unprobe(rbd_dev);
6869 	rbd_dev->image_format = 0;
6870 	kfree(rbd_dev->spec->image_id);
6871 	rbd_dev->spec->image_id = NULL;
6872 }
6873 
6874 /*
6875  * Probe for the existence of the header object for the given rbd
6876  * device.  If this image is the one being mapped (i.e., not a
6877  * parent), initiate a watch on its header object before using that
6878  * object to get detailed information about the rbd image.
6879  *
6880  * On success, returns with header_rwsem held for write if called
6881  * with @depth == 0.
6882  */
6883 static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth)
6884 {
6885 	bool need_watch = !rbd_is_ro(rbd_dev);
6886 	int ret;
6887 
6888 	/*
6889 	 * Get the id from the image id object.  Unless there's an
6890 	 * error, rbd_dev->spec->image_id will be filled in with
6891 	 * a dynamically-allocated string, and rbd_dev->image_format
6892 	 * will be set to either 1 or 2.
6893 	 */
6894 	ret = rbd_dev_image_id(rbd_dev);
6895 	if (ret)
6896 		return ret;
6897 
6898 	ret = rbd_dev_header_name(rbd_dev);
6899 	if (ret)
6900 		goto err_out_format;
6901 
6902 	if (need_watch) {
6903 		ret = rbd_register_watch(rbd_dev);
6904 		if (ret) {
6905 			if (ret == -ENOENT)
6906 				rbd_print_dne(rbd_dev, false);
6907 			goto err_out_format;
6908 		}
6909 	}
6910 
6911 	if (!depth)
6912 		down_write(&rbd_dev->header_rwsem);
6913 
6914 	ret = rbd_dev_header_info(rbd_dev);
6915 	if (ret) {
6916 		if (ret == -ENOENT && !need_watch)
6917 			rbd_print_dne(rbd_dev, false);
6918 		goto err_out_probe;
6919 	}
6920 
6921 	/*
6922 	 * If this image is the one being mapped, we have pool name and
6923 	 * id, image name and id, and snap name - need to fill snap id.
6924 	 * Otherwise this is a parent image, identified by pool, image
6925 	 * and snap ids - need to fill in names for those ids.
6926 	 */
6927 	if (!depth)
6928 		ret = rbd_spec_fill_snap_id(rbd_dev);
6929 	else
6930 		ret = rbd_spec_fill_names(rbd_dev);
6931 	if (ret) {
6932 		if (ret == -ENOENT)
6933 			rbd_print_dne(rbd_dev, true);
6934 		goto err_out_probe;
6935 	}
6936 
6937 	ret = rbd_dev_mapping_set(rbd_dev);
6938 	if (ret)
6939 		goto err_out_probe;
6940 
6941 	if (rbd_is_snap(rbd_dev) &&
6942 	    (rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP)) {
6943 		ret = rbd_object_map_load(rbd_dev);
6944 		if (ret)
6945 			goto err_out_probe;
6946 	}
6947 
6948 	if (rbd_dev->header.features & RBD_FEATURE_LAYERING) {
6949 		ret = rbd_dev_v2_parent_info(rbd_dev);
6950 		if (ret)
6951 			goto err_out_probe;
6952 	}
6953 
6954 	ret = rbd_dev_probe_parent(rbd_dev, depth);
6955 	if (ret)
6956 		goto err_out_probe;
6957 
6958 	dout("discovered format %u image, header name is %s\n",
6959 		rbd_dev->image_format, rbd_dev->header_oid.name);
6960 	return 0;
6961 
6962 err_out_probe:
6963 	if (!depth)
6964 		up_write(&rbd_dev->header_rwsem);
6965 	if (need_watch)
6966 		rbd_unregister_watch(rbd_dev);
6967 	rbd_dev_unprobe(rbd_dev);
6968 err_out_format:
6969 	rbd_dev->image_format = 0;
6970 	kfree(rbd_dev->spec->image_id);
6971 	rbd_dev->spec->image_id = NULL;
6972 	return ret;
6973 }
6974 
6975 static ssize_t do_rbd_add(struct bus_type *bus,
6976 			  const char *buf,
6977 			  size_t count)
6978 {
6979 	struct rbd_device *rbd_dev = NULL;
6980 	struct ceph_options *ceph_opts = NULL;
6981 	struct rbd_options *rbd_opts = NULL;
6982 	struct rbd_spec *spec = NULL;
6983 	struct rbd_client *rbdc;
6984 	int rc;
6985 
6986 	if (!capable(CAP_SYS_ADMIN))
6987 		return -EPERM;
6988 
6989 	if (!try_module_get(THIS_MODULE))
6990 		return -ENODEV;
6991 
6992 	/* parse add command */
6993 	rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec);
6994 	if (rc < 0)
6995 		goto out;
6996 
6997 	rbdc = rbd_get_client(ceph_opts);
6998 	if (IS_ERR(rbdc)) {
6999 		rc = PTR_ERR(rbdc);
7000 		goto err_out_args;
7001 	}
7002 
7003 	/* pick the pool */
7004 	rc = ceph_pg_poolid_by_name(rbdc->client->osdc.osdmap, spec->pool_name);
7005 	if (rc < 0) {
7006 		if (rc == -ENOENT)
7007 			pr_info("pool %s does not exist\n", spec->pool_name);
7008 		goto err_out_client;
7009 	}
7010 	spec->pool_id = (u64)rc;
7011 
7012 	rbd_dev = rbd_dev_create(rbdc, spec, rbd_opts);
7013 	if (!rbd_dev) {
7014 		rc = -ENOMEM;
7015 		goto err_out_client;
7016 	}
7017 	rbdc = NULL;		/* rbd_dev now owns this */
7018 	spec = NULL;		/* rbd_dev now owns this */
7019 	rbd_opts = NULL;	/* rbd_dev now owns this */
7020 
7021 	/* if we are mapping a snapshot it will be a read-only mapping */
7022 	if (rbd_dev->opts->read_only ||
7023 	    strcmp(rbd_dev->spec->snap_name, RBD_SNAP_HEAD_NAME))
7024 		__set_bit(RBD_DEV_FLAG_READONLY, &rbd_dev->flags);
7025 
7026 	rbd_dev->config_info = kstrdup(buf, GFP_KERNEL);
7027 	if (!rbd_dev->config_info) {
7028 		rc = -ENOMEM;
7029 		goto err_out_rbd_dev;
7030 	}
7031 
7032 	rc = rbd_dev_image_probe(rbd_dev, 0);
7033 	if (rc < 0)
7034 		goto err_out_rbd_dev;
7035 
7036 	if (rbd_dev->opts->alloc_size > rbd_dev->layout.object_size) {
7037 		rbd_warn(rbd_dev, "alloc_size adjusted to %u",
7038 			 rbd_dev->layout.object_size);
7039 		rbd_dev->opts->alloc_size = rbd_dev->layout.object_size;
7040 	}
7041 
7042 	rc = rbd_dev_device_setup(rbd_dev);
7043 	if (rc)
7044 		goto err_out_image_probe;
7045 
7046 	rc = rbd_add_acquire_lock(rbd_dev);
7047 	if (rc)
7048 		goto err_out_image_lock;
7049 
7050 	/* Everything's ready.  Announce the disk to the world. */
7051 
7052 	rc = device_add(&rbd_dev->dev);
7053 	if (rc)
7054 		goto err_out_image_lock;
7055 
7056 	rc = device_add_disk(&rbd_dev->dev, rbd_dev->disk, NULL);
7057 	if (rc)
7058 		goto err_out_cleanup_disk;
7059 
7060 	spin_lock(&rbd_dev_list_lock);
7061 	list_add_tail(&rbd_dev->node, &rbd_dev_list);
7062 	spin_unlock(&rbd_dev_list_lock);
7063 
7064 	pr_info("%s: capacity %llu features 0x%llx\n", rbd_dev->disk->disk_name,
7065 		(unsigned long long)get_capacity(rbd_dev->disk) << SECTOR_SHIFT,
7066 		rbd_dev->header.features);
7067 	rc = count;
7068 out:
7069 	module_put(THIS_MODULE);
7070 	return rc;
7071 
7072 err_out_cleanup_disk:
7073 	rbd_free_disk(rbd_dev);
7074 err_out_image_lock:
7075 	rbd_dev_image_unlock(rbd_dev);
7076 	rbd_dev_device_release(rbd_dev);
7077 err_out_image_probe:
7078 	rbd_dev_image_release(rbd_dev);
7079 err_out_rbd_dev:
7080 	rbd_dev_destroy(rbd_dev);
7081 err_out_client:
7082 	rbd_put_client(rbdc);
7083 err_out_args:
7084 	rbd_spec_put(spec);
7085 	kfree(rbd_opts);
7086 	goto out;
7087 }
7088 
7089 static ssize_t add_store(struct bus_type *bus, const char *buf, size_t count)
7090 {
7091 	if (single_major)
7092 		return -EINVAL;
7093 
7094 	return do_rbd_add(bus, buf, count);
7095 }
7096 
7097 static ssize_t add_single_major_store(struct bus_type *bus, const char *buf,
7098 				      size_t count)
7099 {
7100 	return do_rbd_add(bus, buf, count);
7101 }
7102 
7103 static void rbd_dev_remove_parent(struct rbd_device *rbd_dev)
7104 {
7105 	while (rbd_dev->parent) {
7106 		struct rbd_device *first = rbd_dev;
7107 		struct rbd_device *second = first->parent;
7108 		struct rbd_device *third;
7109 
7110 		/*
7111 		 * Follow to the parent with no grandparent and
7112 		 * remove it.
7113 		 */
7114 		while (second && (third = second->parent)) {
7115 			first = second;
7116 			second = third;
7117 		}
7118 		rbd_assert(second);
7119 		rbd_dev_image_release(second);
7120 		rbd_dev_destroy(second);
7121 		first->parent = NULL;
7122 		first->parent_overlap = 0;
7123 
7124 		rbd_assert(first->parent_spec);
7125 		rbd_spec_put(first->parent_spec);
7126 		first->parent_spec = NULL;
7127 	}
7128 }
7129 
7130 static ssize_t do_rbd_remove(struct bus_type *bus,
7131 			     const char *buf,
7132 			     size_t count)
7133 {
7134 	struct rbd_device *rbd_dev = NULL;
7135 	struct list_head *tmp;
7136 	int dev_id;
7137 	char opt_buf[6];
7138 	bool force = false;
7139 	int ret;
7140 
7141 	if (!capable(CAP_SYS_ADMIN))
7142 		return -EPERM;
7143 
7144 	dev_id = -1;
7145 	opt_buf[0] = '\0';
7146 	sscanf(buf, "%d %5s", &dev_id, opt_buf);
7147 	if (dev_id < 0) {
7148 		pr_err("dev_id out of range\n");
7149 		return -EINVAL;
7150 	}
7151 	if (opt_buf[0] != '\0') {
7152 		if (!strcmp(opt_buf, "force")) {
7153 			force = true;
7154 		} else {
7155 			pr_err("bad remove option at '%s'\n", opt_buf);
7156 			return -EINVAL;
7157 		}
7158 	}
7159 
7160 	ret = -ENOENT;
7161 	spin_lock(&rbd_dev_list_lock);
7162 	list_for_each(tmp, &rbd_dev_list) {
7163 		rbd_dev = list_entry(tmp, struct rbd_device, node);
7164 		if (rbd_dev->dev_id == dev_id) {
7165 			ret = 0;
7166 			break;
7167 		}
7168 	}
7169 	if (!ret) {
7170 		spin_lock_irq(&rbd_dev->lock);
7171 		if (rbd_dev->open_count && !force)
7172 			ret = -EBUSY;
7173 		else if (test_and_set_bit(RBD_DEV_FLAG_REMOVING,
7174 					  &rbd_dev->flags))
7175 			ret = -EINPROGRESS;
7176 		spin_unlock_irq(&rbd_dev->lock);
7177 	}
7178 	spin_unlock(&rbd_dev_list_lock);
7179 	if (ret)
7180 		return ret;
7181 
7182 	if (force) {
7183 		/*
7184 		 * Prevent new IO from being queued and wait for existing
7185 		 * IO to complete/fail.
7186 		 */
7187 		blk_mq_freeze_queue(rbd_dev->disk->queue);
7188 		blk_mark_disk_dead(rbd_dev->disk);
7189 	}
7190 
7191 	del_gendisk(rbd_dev->disk);
7192 	spin_lock(&rbd_dev_list_lock);
7193 	list_del_init(&rbd_dev->node);
7194 	spin_unlock(&rbd_dev_list_lock);
7195 	device_del(&rbd_dev->dev);
7196 
7197 	rbd_dev_image_unlock(rbd_dev);
7198 	rbd_dev_device_release(rbd_dev);
7199 	rbd_dev_image_release(rbd_dev);
7200 	rbd_dev_destroy(rbd_dev);
7201 	return count;
7202 }
7203 
7204 static ssize_t remove_store(struct bus_type *bus, const char *buf, size_t count)
7205 {
7206 	if (single_major)
7207 		return -EINVAL;
7208 
7209 	return do_rbd_remove(bus, buf, count);
7210 }
7211 
7212 static ssize_t remove_single_major_store(struct bus_type *bus, const char *buf,
7213 					 size_t count)
7214 {
7215 	return do_rbd_remove(bus, buf, count);
7216 }
7217 
7218 /*
7219  * create control files in sysfs
7220  * /sys/bus/rbd/...
7221  */
7222 static int __init rbd_sysfs_init(void)
7223 {
7224 	int ret;
7225 
7226 	ret = device_register(&rbd_root_dev);
7227 	if (ret < 0)
7228 		return ret;
7229 
7230 	ret = bus_register(&rbd_bus_type);
7231 	if (ret < 0)
7232 		device_unregister(&rbd_root_dev);
7233 
7234 	return ret;
7235 }
7236 
7237 static void __exit rbd_sysfs_cleanup(void)
7238 {
7239 	bus_unregister(&rbd_bus_type);
7240 	device_unregister(&rbd_root_dev);
7241 }
7242 
7243 static int __init rbd_slab_init(void)
7244 {
7245 	rbd_assert(!rbd_img_request_cache);
7246 	rbd_img_request_cache = KMEM_CACHE(rbd_img_request, 0);
7247 	if (!rbd_img_request_cache)
7248 		return -ENOMEM;
7249 
7250 	rbd_assert(!rbd_obj_request_cache);
7251 	rbd_obj_request_cache = KMEM_CACHE(rbd_obj_request, 0);
7252 	if (!rbd_obj_request_cache)
7253 		goto out_err;
7254 
7255 	return 0;
7256 
7257 out_err:
7258 	kmem_cache_destroy(rbd_img_request_cache);
7259 	rbd_img_request_cache = NULL;
7260 	return -ENOMEM;
7261 }
7262 
7263 static void rbd_slab_exit(void)
7264 {
7265 	rbd_assert(rbd_obj_request_cache);
7266 	kmem_cache_destroy(rbd_obj_request_cache);
7267 	rbd_obj_request_cache = NULL;
7268 
7269 	rbd_assert(rbd_img_request_cache);
7270 	kmem_cache_destroy(rbd_img_request_cache);
7271 	rbd_img_request_cache = NULL;
7272 }
7273 
7274 static int __init rbd_init(void)
7275 {
7276 	int rc;
7277 
7278 	if (!libceph_compatible(NULL)) {
7279 		rbd_warn(NULL, "libceph incompatibility (quitting)");
7280 		return -EINVAL;
7281 	}
7282 
7283 	rc = rbd_slab_init();
7284 	if (rc)
7285 		return rc;
7286 
7287 	/*
7288 	 * The number of active work items is limited by the number of
7289 	 * rbd devices * queue depth, so leave @max_active at default.
7290 	 */
7291 	rbd_wq = alloc_workqueue(RBD_DRV_NAME, WQ_MEM_RECLAIM, 0);
7292 	if (!rbd_wq) {
7293 		rc = -ENOMEM;
7294 		goto err_out_slab;
7295 	}
7296 
7297 	if (single_major) {
7298 		rbd_major = register_blkdev(0, RBD_DRV_NAME);
7299 		if (rbd_major < 0) {
7300 			rc = rbd_major;
7301 			goto err_out_wq;
7302 		}
7303 	}
7304 
7305 	rc = rbd_sysfs_init();
7306 	if (rc)
7307 		goto err_out_blkdev;
7308 
7309 	if (single_major)
7310 		pr_info("loaded (major %d)\n", rbd_major);
7311 	else
7312 		pr_info("loaded\n");
7313 
7314 	return 0;
7315 
7316 err_out_blkdev:
7317 	if (single_major)
7318 		unregister_blkdev(rbd_major, RBD_DRV_NAME);
7319 err_out_wq:
7320 	destroy_workqueue(rbd_wq);
7321 err_out_slab:
7322 	rbd_slab_exit();
7323 	return rc;
7324 }
7325 
7326 static void __exit rbd_exit(void)
7327 {
7328 	ida_destroy(&rbd_dev_id_ida);
7329 	rbd_sysfs_cleanup();
7330 	if (single_major)
7331 		unregister_blkdev(rbd_major, RBD_DRV_NAME);
7332 	destroy_workqueue(rbd_wq);
7333 	rbd_slab_exit();
7334 }
7335 
7336 module_init(rbd_init);
7337 module_exit(rbd_exit);
7338 
7339 MODULE_AUTHOR("Alex Elder <elder@inktank.com>");
7340 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
7341 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
7342 /* following authorship retained from original osdblk.c */
7343 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
7344 
7345 MODULE_DESCRIPTION("RADOS Block Device (RBD) driver");
7346 MODULE_LICENSE("GPL");
7347