xref: /openbmc/linux/drivers/block/rbd.c (revision 2bc7d3e0)
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 *rbdc = NULL, *iter;
760 
761 	if (ceph_opts->flags & CEPH_OPT_NOSHARE)
762 		return NULL;
763 
764 	spin_lock(&rbd_client_list_lock);
765 	list_for_each_entry(iter, &rbd_client_list, node) {
766 		if (!ceph_compare_options(ceph_opts, iter->client)) {
767 			__rbd_get_client(iter);
768 
769 			rbdc = iter;
770 			break;
771 		}
772 	}
773 	spin_unlock(&rbd_client_list_lock);
774 
775 	return rbdc;
776 }
777 
778 /*
779  * (Per device) rbd map options
780  */
781 enum {
782 	Opt_queue_depth,
783 	Opt_alloc_size,
784 	Opt_lock_timeout,
785 	/* int args above */
786 	Opt_pool_ns,
787 	Opt_compression_hint,
788 	/* string args above */
789 	Opt_read_only,
790 	Opt_read_write,
791 	Opt_lock_on_read,
792 	Opt_exclusive,
793 	Opt_notrim,
794 };
795 
796 enum {
797 	Opt_compression_hint_none,
798 	Opt_compression_hint_compressible,
799 	Opt_compression_hint_incompressible,
800 };
801 
802 static const struct constant_table rbd_param_compression_hint[] = {
803 	{"none",		Opt_compression_hint_none},
804 	{"compressible",	Opt_compression_hint_compressible},
805 	{"incompressible",	Opt_compression_hint_incompressible},
806 	{}
807 };
808 
809 static const struct fs_parameter_spec rbd_parameters[] = {
810 	fsparam_u32	("alloc_size",			Opt_alloc_size),
811 	fsparam_enum	("compression_hint",		Opt_compression_hint,
812 			 rbd_param_compression_hint),
813 	fsparam_flag	("exclusive",			Opt_exclusive),
814 	fsparam_flag	("lock_on_read",		Opt_lock_on_read),
815 	fsparam_u32	("lock_timeout",		Opt_lock_timeout),
816 	fsparam_flag	("notrim",			Opt_notrim),
817 	fsparam_string	("_pool_ns",			Opt_pool_ns),
818 	fsparam_u32	("queue_depth",			Opt_queue_depth),
819 	fsparam_flag	("read_only",			Opt_read_only),
820 	fsparam_flag	("read_write",			Opt_read_write),
821 	fsparam_flag	("ro",				Opt_read_only),
822 	fsparam_flag	("rw",				Opt_read_write),
823 	{}
824 };
825 
826 struct rbd_options {
827 	int	queue_depth;
828 	int	alloc_size;
829 	unsigned long	lock_timeout;
830 	bool	read_only;
831 	bool	lock_on_read;
832 	bool	exclusive;
833 	bool	trim;
834 
835 	u32 alloc_hint_flags;  /* CEPH_OSD_OP_ALLOC_HINT_FLAG_* */
836 };
837 
838 #define RBD_QUEUE_DEPTH_DEFAULT	BLKDEV_DEFAULT_RQ
839 #define RBD_ALLOC_SIZE_DEFAULT	(64 * 1024)
840 #define RBD_LOCK_TIMEOUT_DEFAULT 0  /* no timeout */
841 #define RBD_READ_ONLY_DEFAULT	false
842 #define RBD_LOCK_ON_READ_DEFAULT false
843 #define RBD_EXCLUSIVE_DEFAULT	false
844 #define RBD_TRIM_DEFAULT	true
845 
846 struct rbd_parse_opts_ctx {
847 	struct rbd_spec		*spec;
848 	struct ceph_options	*copts;
849 	struct rbd_options	*opts;
850 };
851 
852 static char* obj_op_name(enum obj_operation_type op_type)
853 {
854 	switch (op_type) {
855 	case OBJ_OP_READ:
856 		return "read";
857 	case OBJ_OP_WRITE:
858 		return "write";
859 	case OBJ_OP_DISCARD:
860 		return "discard";
861 	case OBJ_OP_ZEROOUT:
862 		return "zeroout";
863 	default:
864 		return "???";
865 	}
866 }
867 
868 /*
869  * Destroy ceph client
870  *
871  * Caller must hold rbd_client_list_lock.
872  */
873 static void rbd_client_release(struct kref *kref)
874 {
875 	struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref);
876 
877 	dout("%s: rbdc %p\n", __func__, rbdc);
878 	spin_lock(&rbd_client_list_lock);
879 	list_del(&rbdc->node);
880 	spin_unlock(&rbd_client_list_lock);
881 
882 	ceph_destroy_client(rbdc->client);
883 	kfree(rbdc);
884 }
885 
886 /*
887  * Drop reference to ceph client node. If it's not referenced anymore, release
888  * it.
889  */
890 static void rbd_put_client(struct rbd_client *rbdc)
891 {
892 	if (rbdc)
893 		kref_put(&rbdc->kref, rbd_client_release);
894 }
895 
896 /*
897  * Get a ceph client with specific addr and configuration, if one does
898  * not exist create it.  Either way, ceph_opts is consumed by this
899  * function.
900  */
901 static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts)
902 {
903 	struct rbd_client *rbdc;
904 	int ret;
905 
906 	mutex_lock(&client_mutex);
907 	rbdc = rbd_client_find(ceph_opts);
908 	if (rbdc) {
909 		ceph_destroy_options(ceph_opts);
910 
911 		/*
912 		 * Using an existing client.  Make sure ->pg_pools is up to
913 		 * date before we look up the pool id in do_rbd_add().
914 		 */
915 		ret = ceph_wait_for_latest_osdmap(rbdc->client,
916 					rbdc->client->options->mount_timeout);
917 		if (ret) {
918 			rbd_warn(NULL, "failed to get latest osdmap: %d", ret);
919 			rbd_put_client(rbdc);
920 			rbdc = ERR_PTR(ret);
921 		}
922 	} else {
923 		rbdc = rbd_client_create(ceph_opts);
924 	}
925 	mutex_unlock(&client_mutex);
926 
927 	return rbdc;
928 }
929 
930 static bool rbd_image_format_valid(u32 image_format)
931 {
932 	return image_format == 1 || image_format == 2;
933 }
934 
935 static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk)
936 {
937 	size_t size;
938 	u32 snap_count;
939 
940 	/* The header has to start with the magic rbd header text */
941 	if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT)))
942 		return false;
943 
944 	/* The bio layer requires at least sector-sized I/O */
945 
946 	if (ondisk->options.order < SECTOR_SHIFT)
947 		return false;
948 
949 	/* If we use u64 in a few spots we may be able to loosen this */
950 
951 	if (ondisk->options.order > 8 * sizeof (int) - 1)
952 		return false;
953 
954 	/*
955 	 * The size of a snapshot header has to fit in a size_t, and
956 	 * that limits the number of snapshots.
957 	 */
958 	snap_count = le32_to_cpu(ondisk->snap_count);
959 	size = SIZE_MAX - sizeof (struct ceph_snap_context);
960 	if (snap_count > size / sizeof (__le64))
961 		return false;
962 
963 	/*
964 	 * Not only that, but the size of the entire the snapshot
965 	 * header must also be representable in a size_t.
966 	 */
967 	size -= snap_count * sizeof (__le64);
968 	if ((u64) size < le64_to_cpu(ondisk->snap_names_len))
969 		return false;
970 
971 	return true;
972 }
973 
974 /*
975  * returns the size of an object in the image
976  */
977 static u32 rbd_obj_bytes(struct rbd_image_header *header)
978 {
979 	return 1U << header->obj_order;
980 }
981 
982 static void rbd_init_layout(struct rbd_device *rbd_dev)
983 {
984 	if (rbd_dev->header.stripe_unit == 0 ||
985 	    rbd_dev->header.stripe_count == 0) {
986 		rbd_dev->header.stripe_unit = rbd_obj_bytes(&rbd_dev->header);
987 		rbd_dev->header.stripe_count = 1;
988 	}
989 
990 	rbd_dev->layout.stripe_unit = rbd_dev->header.stripe_unit;
991 	rbd_dev->layout.stripe_count = rbd_dev->header.stripe_count;
992 	rbd_dev->layout.object_size = rbd_obj_bytes(&rbd_dev->header);
993 	rbd_dev->layout.pool_id = rbd_dev->header.data_pool_id == CEPH_NOPOOL ?
994 			  rbd_dev->spec->pool_id : rbd_dev->header.data_pool_id;
995 	RCU_INIT_POINTER(rbd_dev->layout.pool_ns, NULL);
996 }
997 
998 /*
999  * Fill an rbd image header with information from the given format 1
1000  * on-disk header.
1001  */
1002 static int rbd_header_from_disk(struct rbd_device *rbd_dev,
1003 				 struct rbd_image_header_ondisk *ondisk)
1004 {
1005 	struct rbd_image_header *header = &rbd_dev->header;
1006 	bool first_time = header->object_prefix == NULL;
1007 	struct ceph_snap_context *snapc;
1008 	char *object_prefix = NULL;
1009 	char *snap_names = NULL;
1010 	u64 *snap_sizes = NULL;
1011 	u32 snap_count;
1012 	int ret = -ENOMEM;
1013 	u32 i;
1014 
1015 	/* Allocate this now to avoid having to handle failure below */
1016 
1017 	if (first_time) {
1018 		object_prefix = kstrndup(ondisk->object_prefix,
1019 					 sizeof(ondisk->object_prefix),
1020 					 GFP_KERNEL);
1021 		if (!object_prefix)
1022 			return -ENOMEM;
1023 	}
1024 
1025 	/* Allocate the snapshot context and fill it in */
1026 
1027 	snap_count = le32_to_cpu(ondisk->snap_count);
1028 	snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
1029 	if (!snapc)
1030 		goto out_err;
1031 	snapc->seq = le64_to_cpu(ondisk->snap_seq);
1032 	if (snap_count) {
1033 		struct rbd_image_snap_ondisk *snaps;
1034 		u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len);
1035 
1036 		/* We'll keep a copy of the snapshot names... */
1037 
1038 		if (snap_names_len > (u64)SIZE_MAX)
1039 			goto out_2big;
1040 		snap_names = kmalloc(snap_names_len, GFP_KERNEL);
1041 		if (!snap_names)
1042 			goto out_err;
1043 
1044 		/* ...as well as the array of their sizes. */
1045 		snap_sizes = kmalloc_array(snap_count,
1046 					   sizeof(*header->snap_sizes),
1047 					   GFP_KERNEL);
1048 		if (!snap_sizes)
1049 			goto out_err;
1050 
1051 		/*
1052 		 * Copy the names, and fill in each snapshot's id
1053 		 * and size.
1054 		 *
1055 		 * Note that rbd_dev_v1_header_info() guarantees the
1056 		 * ondisk buffer we're working with has
1057 		 * snap_names_len bytes beyond the end of the
1058 		 * snapshot id array, this memcpy() is safe.
1059 		 */
1060 		memcpy(snap_names, &ondisk->snaps[snap_count], snap_names_len);
1061 		snaps = ondisk->snaps;
1062 		for (i = 0; i < snap_count; i++) {
1063 			snapc->snaps[i] = le64_to_cpu(snaps[i].id);
1064 			snap_sizes[i] = le64_to_cpu(snaps[i].image_size);
1065 		}
1066 	}
1067 
1068 	/* We won't fail any more, fill in the header */
1069 
1070 	if (first_time) {
1071 		header->object_prefix = object_prefix;
1072 		header->obj_order = ondisk->options.order;
1073 		rbd_init_layout(rbd_dev);
1074 	} else {
1075 		ceph_put_snap_context(header->snapc);
1076 		kfree(header->snap_names);
1077 		kfree(header->snap_sizes);
1078 	}
1079 
1080 	/* The remaining fields always get updated (when we refresh) */
1081 
1082 	header->image_size = le64_to_cpu(ondisk->image_size);
1083 	header->snapc = snapc;
1084 	header->snap_names = snap_names;
1085 	header->snap_sizes = snap_sizes;
1086 
1087 	return 0;
1088 out_2big:
1089 	ret = -EIO;
1090 out_err:
1091 	kfree(snap_sizes);
1092 	kfree(snap_names);
1093 	ceph_put_snap_context(snapc);
1094 	kfree(object_prefix);
1095 
1096 	return ret;
1097 }
1098 
1099 static const char *_rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, u32 which)
1100 {
1101 	const char *snap_name;
1102 
1103 	rbd_assert(which < rbd_dev->header.snapc->num_snaps);
1104 
1105 	/* Skip over names until we find the one we are looking for */
1106 
1107 	snap_name = rbd_dev->header.snap_names;
1108 	while (which--)
1109 		snap_name += strlen(snap_name) + 1;
1110 
1111 	return kstrdup(snap_name, GFP_KERNEL);
1112 }
1113 
1114 /*
1115  * Snapshot id comparison function for use with qsort()/bsearch().
1116  * Note that result is for snapshots in *descending* order.
1117  */
1118 static int snapid_compare_reverse(const void *s1, const void *s2)
1119 {
1120 	u64 snap_id1 = *(u64 *)s1;
1121 	u64 snap_id2 = *(u64 *)s2;
1122 
1123 	if (snap_id1 < snap_id2)
1124 		return 1;
1125 	return snap_id1 == snap_id2 ? 0 : -1;
1126 }
1127 
1128 /*
1129  * Search a snapshot context to see if the given snapshot id is
1130  * present.
1131  *
1132  * Returns the position of the snapshot id in the array if it's found,
1133  * or BAD_SNAP_INDEX otherwise.
1134  *
1135  * Note: The snapshot array is in kept sorted (by the osd) in
1136  * reverse order, highest snapshot id first.
1137  */
1138 static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id)
1139 {
1140 	struct ceph_snap_context *snapc = rbd_dev->header.snapc;
1141 	u64 *found;
1142 
1143 	found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps,
1144 				sizeof (snap_id), snapid_compare_reverse);
1145 
1146 	return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX;
1147 }
1148 
1149 static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev,
1150 					u64 snap_id)
1151 {
1152 	u32 which;
1153 	const char *snap_name;
1154 
1155 	which = rbd_dev_snap_index(rbd_dev, snap_id);
1156 	if (which == BAD_SNAP_INDEX)
1157 		return ERR_PTR(-ENOENT);
1158 
1159 	snap_name = _rbd_dev_v1_snap_name(rbd_dev, which);
1160 	return snap_name ? snap_name : ERR_PTR(-ENOMEM);
1161 }
1162 
1163 static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id)
1164 {
1165 	if (snap_id == CEPH_NOSNAP)
1166 		return RBD_SNAP_HEAD_NAME;
1167 
1168 	rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
1169 	if (rbd_dev->image_format == 1)
1170 		return rbd_dev_v1_snap_name(rbd_dev, snap_id);
1171 
1172 	return rbd_dev_v2_snap_name(rbd_dev, snap_id);
1173 }
1174 
1175 static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
1176 				u64 *snap_size)
1177 {
1178 	rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
1179 	if (snap_id == CEPH_NOSNAP) {
1180 		*snap_size = rbd_dev->header.image_size;
1181 	} else if (rbd_dev->image_format == 1) {
1182 		u32 which;
1183 
1184 		which = rbd_dev_snap_index(rbd_dev, snap_id);
1185 		if (which == BAD_SNAP_INDEX)
1186 			return -ENOENT;
1187 
1188 		*snap_size = rbd_dev->header.snap_sizes[which];
1189 	} else {
1190 		u64 size = 0;
1191 		int ret;
1192 
1193 		ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size);
1194 		if (ret)
1195 			return ret;
1196 
1197 		*snap_size = size;
1198 	}
1199 	return 0;
1200 }
1201 
1202 static int rbd_dev_mapping_set(struct rbd_device *rbd_dev)
1203 {
1204 	u64 snap_id = rbd_dev->spec->snap_id;
1205 	u64 size = 0;
1206 	int ret;
1207 
1208 	ret = rbd_snap_size(rbd_dev, snap_id, &size);
1209 	if (ret)
1210 		return ret;
1211 
1212 	rbd_dev->mapping.size = size;
1213 	return 0;
1214 }
1215 
1216 static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev)
1217 {
1218 	rbd_dev->mapping.size = 0;
1219 }
1220 
1221 static void zero_bios(struct ceph_bio_iter *bio_pos, u32 off, u32 bytes)
1222 {
1223 	struct ceph_bio_iter it = *bio_pos;
1224 
1225 	ceph_bio_iter_advance(&it, off);
1226 	ceph_bio_iter_advance_step(&it, bytes, ({
1227 		memzero_bvec(&bv);
1228 	}));
1229 }
1230 
1231 static void zero_bvecs(struct ceph_bvec_iter *bvec_pos, u32 off, u32 bytes)
1232 {
1233 	struct ceph_bvec_iter it = *bvec_pos;
1234 
1235 	ceph_bvec_iter_advance(&it, off);
1236 	ceph_bvec_iter_advance_step(&it, bytes, ({
1237 		memzero_bvec(&bv);
1238 	}));
1239 }
1240 
1241 /*
1242  * Zero a range in @obj_req data buffer defined by a bio (list) or
1243  * (private) bio_vec array.
1244  *
1245  * @off is relative to the start of the data buffer.
1246  */
1247 static void rbd_obj_zero_range(struct rbd_obj_request *obj_req, u32 off,
1248 			       u32 bytes)
1249 {
1250 	dout("%s %p data buf %u~%u\n", __func__, obj_req, off, bytes);
1251 
1252 	switch (obj_req->img_request->data_type) {
1253 	case OBJ_REQUEST_BIO:
1254 		zero_bios(&obj_req->bio_pos, off, bytes);
1255 		break;
1256 	case OBJ_REQUEST_BVECS:
1257 	case OBJ_REQUEST_OWN_BVECS:
1258 		zero_bvecs(&obj_req->bvec_pos, off, bytes);
1259 		break;
1260 	default:
1261 		BUG();
1262 	}
1263 }
1264 
1265 static void rbd_obj_request_destroy(struct kref *kref);
1266 static void rbd_obj_request_put(struct rbd_obj_request *obj_request)
1267 {
1268 	rbd_assert(obj_request != NULL);
1269 	dout("%s: obj %p (was %d)\n", __func__, obj_request,
1270 		kref_read(&obj_request->kref));
1271 	kref_put(&obj_request->kref, rbd_obj_request_destroy);
1272 }
1273 
1274 static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request,
1275 					struct rbd_obj_request *obj_request)
1276 {
1277 	rbd_assert(obj_request->img_request == NULL);
1278 
1279 	/* Image request now owns object's original reference */
1280 	obj_request->img_request = img_request;
1281 	dout("%s: img %p obj %p\n", __func__, img_request, obj_request);
1282 }
1283 
1284 static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request,
1285 					struct rbd_obj_request *obj_request)
1286 {
1287 	dout("%s: img %p obj %p\n", __func__, img_request, obj_request);
1288 	list_del(&obj_request->ex.oe_item);
1289 	rbd_assert(obj_request->img_request == img_request);
1290 	rbd_obj_request_put(obj_request);
1291 }
1292 
1293 static void rbd_osd_submit(struct ceph_osd_request *osd_req)
1294 {
1295 	struct rbd_obj_request *obj_req = osd_req->r_priv;
1296 
1297 	dout("%s osd_req %p for obj_req %p objno %llu %llu~%llu\n",
1298 	     __func__, osd_req, obj_req, obj_req->ex.oe_objno,
1299 	     obj_req->ex.oe_off, obj_req->ex.oe_len);
1300 	ceph_osdc_start_request(osd_req->r_osdc, osd_req, false);
1301 }
1302 
1303 /*
1304  * The default/initial value for all image request flags is 0.  Each
1305  * is conditionally set to 1 at image request initialization time
1306  * and currently never change thereafter.
1307  */
1308 static void img_request_layered_set(struct rbd_img_request *img_request)
1309 {
1310 	set_bit(IMG_REQ_LAYERED, &img_request->flags);
1311 }
1312 
1313 static bool img_request_layered_test(struct rbd_img_request *img_request)
1314 {
1315 	return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0;
1316 }
1317 
1318 static bool rbd_obj_is_entire(struct rbd_obj_request *obj_req)
1319 {
1320 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
1321 
1322 	return !obj_req->ex.oe_off &&
1323 	       obj_req->ex.oe_len == rbd_dev->layout.object_size;
1324 }
1325 
1326 static bool rbd_obj_is_tail(struct rbd_obj_request *obj_req)
1327 {
1328 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
1329 
1330 	return obj_req->ex.oe_off + obj_req->ex.oe_len ==
1331 					rbd_dev->layout.object_size;
1332 }
1333 
1334 /*
1335  * Must be called after rbd_obj_calc_img_extents().
1336  */
1337 static bool rbd_obj_copyup_enabled(struct rbd_obj_request *obj_req)
1338 {
1339 	if (!obj_req->num_img_extents ||
1340 	    (rbd_obj_is_entire(obj_req) &&
1341 	     !obj_req->img_request->snapc->num_snaps))
1342 		return false;
1343 
1344 	return true;
1345 }
1346 
1347 static u64 rbd_obj_img_extents_bytes(struct rbd_obj_request *obj_req)
1348 {
1349 	return ceph_file_extents_bytes(obj_req->img_extents,
1350 				       obj_req->num_img_extents);
1351 }
1352 
1353 static bool rbd_img_is_write(struct rbd_img_request *img_req)
1354 {
1355 	switch (img_req->op_type) {
1356 	case OBJ_OP_READ:
1357 		return false;
1358 	case OBJ_OP_WRITE:
1359 	case OBJ_OP_DISCARD:
1360 	case OBJ_OP_ZEROOUT:
1361 		return true;
1362 	default:
1363 		BUG();
1364 	}
1365 }
1366 
1367 static void rbd_osd_req_callback(struct ceph_osd_request *osd_req)
1368 {
1369 	struct rbd_obj_request *obj_req = osd_req->r_priv;
1370 	int result;
1371 
1372 	dout("%s osd_req %p result %d for obj_req %p\n", __func__, osd_req,
1373 	     osd_req->r_result, obj_req);
1374 
1375 	/*
1376 	 * Writes aren't allowed to return a data payload.  In some
1377 	 * guarded write cases (e.g. stat + zero on an empty object)
1378 	 * a stat response makes it through, but we don't care.
1379 	 */
1380 	if (osd_req->r_result > 0 && rbd_img_is_write(obj_req->img_request))
1381 		result = 0;
1382 	else
1383 		result = osd_req->r_result;
1384 
1385 	rbd_obj_handle_request(obj_req, result);
1386 }
1387 
1388 static void rbd_osd_format_read(struct ceph_osd_request *osd_req)
1389 {
1390 	struct rbd_obj_request *obj_request = osd_req->r_priv;
1391 	struct rbd_device *rbd_dev = obj_request->img_request->rbd_dev;
1392 	struct ceph_options *opt = rbd_dev->rbd_client->client->options;
1393 
1394 	osd_req->r_flags = CEPH_OSD_FLAG_READ | opt->read_from_replica;
1395 	osd_req->r_snapid = obj_request->img_request->snap_id;
1396 }
1397 
1398 static void rbd_osd_format_write(struct ceph_osd_request *osd_req)
1399 {
1400 	struct rbd_obj_request *obj_request = osd_req->r_priv;
1401 
1402 	osd_req->r_flags = CEPH_OSD_FLAG_WRITE;
1403 	ktime_get_real_ts64(&osd_req->r_mtime);
1404 	osd_req->r_data_offset = obj_request->ex.oe_off;
1405 }
1406 
1407 static struct ceph_osd_request *
1408 __rbd_obj_add_osd_request(struct rbd_obj_request *obj_req,
1409 			  struct ceph_snap_context *snapc, int num_ops)
1410 {
1411 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
1412 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
1413 	struct ceph_osd_request *req;
1414 	const char *name_format = rbd_dev->image_format == 1 ?
1415 				      RBD_V1_DATA_FORMAT : RBD_V2_DATA_FORMAT;
1416 	int ret;
1417 
1418 	req = ceph_osdc_alloc_request(osdc, snapc, num_ops, false, GFP_NOIO);
1419 	if (!req)
1420 		return ERR_PTR(-ENOMEM);
1421 
1422 	list_add_tail(&req->r_private_item, &obj_req->osd_reqs);
1423 	req->r_callback = rbd_osd_req_callback;
1424 	req->r_priv = obj_req;
1425 
1426 	/*
1427 	 * Data objects may be stored in a separate pool, but always in
1428 	 * the same namespace in that pool as the header in its pool.
1429 	 */
1430 	ceph_oloc_copy(&req->r_base_oloc, &rbd_dev->header_oloc);
1431 	req->r_base_oloc.pool = rbd_dev->layout.pool_id;
1432 
1433 	ret = ceph_oid_aprintf(&req->r_base_oid, GFP_NOIO, name_format,
1434 			       rbd_dev->header.object_prefix,
1435 			       obj_req->ex.oe_objno);
1436 	if (ret)
1437 		return ERR_PTR(ret);
1438 
1439 	return req;
1440 }
1441 
1442 static struct ceph_osd_request *
1443 rbd_obj_add_osd_request(struct rbd_obj_request *obj_req, int num_ops)
1444 {
1445 	return __rbd_obj_add_osd_request(obj_req, obj_req->img_request->snapc,
1446 					 num_ops);
1447 }
1448 
1449 static struct rbd_obj_request *rbd_obj_request_create(void)
1450 {
1451 	struct rbd_obj_request *obj_request;
1452 
1453 	obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_NOIO);
1454 	if (!obj_request)
1455 		return NULL;
1456 
1457 	ceph_object_extent_init(&obj_request->ex);
1458 	INIT_LIST_HEAD(&obj_request->osd_reqs);
1459 	mutex_init(&obj_request->state_mutex);
1460 	kref_init(&obj_request->kref);
1461 
1462 	dout("%s %p\n", __func__, obj_request);
1463 	return obj_request;
1464 }
1465 
1466 static void rbd_obj_request_destroy(struct kref *kref)
1467 {
1468 	struct rbd_obj_request *obj_request;
1469 	struct ceph_osd_request *osd_req;
1470 	u32 i;
1471 
1472 	obj_request = container_of(kref, struct rbd_obj_request, kref);
1473 
1474 	dout("%s: obj %p\n", __func__, obj_request);
1475 
1476 	while (!list_empty(&obj_request->osd_reqs)) {
1477 		osd_req = list_first_entry(&obj_request->osd_reqs,
1478 				    struct ceph_osd_request, r_private_item);
1479 		list_del_init(&osd_req->r_private_item);
1480 		ceph_osdc_put_request(osd_req);
1481 	}
1482 
1483 	switch (obj_request->img_request->data_type) {
1484 	case OBJ_REQUEST_NODATA:
1485 	case OBJ_REQUEST_BIO:
1486 	case OBJ_REQUEST_BVECS:
1487 		break;		/* Nothing to do */
1488 	case OBJ_REQUEST_OWN_BVECS:
1489 		kfree(obj_request->bvec_pos.bvecs);
1490 		break;
1491 	default:
1492 		BUG();
1493 	}
1494 
1495 	kfree(obj_request->img_extents);
1496 	if (obj_request->copyup_bvecs) {
1497 		for (i = 0; i < obj_request->copyup_bvec_count; i++) {
1498 			if (obj_request->copyup_bvecs[i].bv_page)
1499 				__free_page(obj_request->copyup_bvecs[i].bv_page);
1500 		}
1501 		kfree(obj_request->copyup_bvecs);
1502 	}
1503 
1504 	kmem_cache_free(rbd_obj_request_cache, obj_request);
1505 }
1506 
1507 /* It's OK to call this for a device with no parent */
1508 
1509 static void rbd_spec_put(struct rbd_spec *spec);
1510 static void rbd_dev_unparent(struct rbd_device *rbd_dev)
1511 {
1512 	rbd_dev_remove_parent(rbd_dev);
1513 	rbd_spec_put(rbd_dev->parent_spec);
1514 	rbd_dev->parent_spec = NULL;
1515 	rbd_dev->parent_overlap = 0;
1516 }
1517 
1518 /*
1519  * Parent image reference counting is used to determine when an
1520  * image's parent fields can be safely torn down--after there are no
1521  * more in-flight requests to the parent image.  When the last
1522  * reference is dropped, cleaning them up is safe.
1523  */
1524 static void rbd_dev_parent_put(struct rbd_device *rbd_dev)
1525 {
1526 	int counter;
1527 
1528 	if (!rbd_dev->parent_spec)
1529 		return;
1530 
1531 	counter = atomic_dec_return_safe(&rbd_dev->parent_ref);
1532 	if (counter > 0)
1533 		return;
1534 
1535 	/* Last reference; clean up parent data structures */
1536 
1537 	if (!counter)
1538 		rbd_dev_unparent(rbd_dev);
1539 	else
1540 		rbd_warn(rbd_dev, "parent reference underflow");
1541 }
1542 
1543 /*
1544  * If an image has a non-zero parent overlap, get a reference to its
1545  * parent.
1546  *
1547  * Returns true if the rbd device has a parent with a non-zero
1548  * overlap and a reference for it was successfully taken, or
1549  * false otherwise.
1550  */
1551 static bool rbd_dev_parent_get(struct rbd_device *rbd_dev)
1552 {
1553 	int counter = 0;
1554 
1555 	if (!rbd_dev->parent_spec)
1556 		return false;
1557 
1558 	if (rbd_dev->parent_overlap)
1559 		counter = atomic_inc_return_safe(&rbd_dev->parent_ref);
1560 
1561 	if (counter < 0)
1562 		rbd_warn(rbd_dev, "parent reference overflow");
1563 
1564 	return counter > 0;
1565 }
1566 
1567 static void rbd_img_request_init(struct rbd_img_request *img_request,
1568 				 struct rbd_device *rbd_dev,
1569 				 enum obj_operation_type op_type)
1570 {
1571 	memset(img_request, 0, sizeof(*img_request));
1572 
1573 	img_request->rbd_dev = rbd_dev;
1574 	img_request->op_type = op_type;
1575 
1576 	INIT_LIST_HEAD(&img_request->lock_item);
1577 	INIT_LIST_HEAD(&img_request->object_extents);
1578 	mutex_init(&img_request->state_mutex);
1579 }
1580 
1581 static void rbd_img_capture_header(struct rbd_img_request *img_req)
1582 {
1583 	struct rbd_device *rbd_dev = img_req->rbd_dev;
1584 
1585 	lockdep_assert_held(&rbd_dev->header_rwsem);
1586 
1587 	if (rbd_img_is_write(img_req))
1588 		img_req->snapc = ceph_get_snap_context(rbd_dev->header.snapc);
1589 	else
1590 		img_req->snap_id = rbd_dev->spec->snap_id;
1591 
1592 	if (rbd_dev_parent_get(rbd_dev))
1593 		img_request_layered_set(img_req);
1594 }
1595 
1596 static void rbd_img_request_destroy(struct rbd_img_request *img_request)
1597 {
1598 	struct rbd_obj_request *obj_request;
1599 	struct rbd_obj_request *next_obj_request;
1600 
1601 	dout("%s: img %p\n", __func__, img_request);
1602 
1603 	WARN_ON(!list_empty(&img_request->lock_item));
1604 	for_each_obj_request_safe(img_request, obj_request, next_obj_request)
1605 		rbd_img_obj_request_del(img_request, obj_request);
1606 
1607 	if (img_request_layered_test(img_request))
1608 		rbd_dev_parent_put(img_request->rbd_dev);
1609 
1610 	if (rbd_img_is_write(img_request))
1611 		ceph_put_snap_context(img_request->snapc);
1612 
1613 	if (test_bit(IMG_REQ_CHILD, &img_request->flags))
1614 		kmem_cache_free(rbd_img_request_cache, img_request);
1615 }
1616 
1617 #define BITS_PER_OBJ	2
1618 #define OBJS_PER_BYTE	(BITS_PER_BYTE / BITS_PER_OBJ)
1619 #define OBJ_MASK	((1 << BITS_PER_OBJ) - 1)
1620 
1621 static void __rbd_object_map_index(struct rbd_device *rbd_dev, u64 objno,
1622 				   u64 *index, u8 *shift)
1623 {
1624 	u32 off;
1625 
1626 	rbd_assert(objno < rbd_dev->object_map_size);
1627 	*index = div_u64_rem(objno, OBJS_PER_BYTE, &off);
1628 	*shift = (OBJS_PER_BYTE - off - 1) * BITS_PER_OBJ;
1629 }
1630 
1631 static u8 __rbd_object_map_get(struct rbd_device *rbd_dev, u64 objno)
1632 {
1633 	u64 index;
1634 	u8 shift;
1635 
1636 	lockdep_assert_held(&rbd_dev->object_map_lock);
1637 	__rbd_object_map_index(rbd_dev, objno, &index, &shift);
1638 	return (rbd_dev->object_map[index] >> shift) & OBJ_MASK;
1639 }
1640 
1641 static void __rbd_object_map_set(struct rbd_device *rbd_dev, u64 objno, u8 val)
1642 {
1643 	u64 index;
1644 	u8 shift;
1645 	u8 *p;
1646 
1647 	lockdep_assert_held(&rbd_dev->object_map_lock);
1648 	rbd_assert(!(val & ~OBJ_MASK));
1649 
1650 	__rbd_object_map_index(rbd_dev, objno, &index, &shift);
1651 	p = &rbd_dev->object_map[index];
1652 	*p = (*p & ~(OBJ_MASK << shift)) | (val << shift);
1653 }
1654 
1655 static u8 rbd_object_map_get(struct rbd_device *rbd_dev, u64 objno)
1656 {
1657 	u8 state;
1658 
1659 	spin_lock(&rbd_dev->object_map_lock);
1660 	state = __rbd_object_map_get(rbd_dev, objno);
1661 	spin_unlock(&rbd_dev->object_map_lock);
1662 	return state;
1663 }
1664 
1665 static bool use_object_map(struct rbd_device *rbd_dev)
1666 {
1667 	/*
1668 	 * An image mapped read-only can't use the object map -- it isn't
1669 	 * loaded because the header lock isn't acquired.  Someone else can
1670 	 * write to the image and update the object map behind our back.
1671 	 *
1672 	 * A snapshot can't be written to, so using the object map is always
1673 	 * safe.
1674 	 */
1675 	if (!rbd_is_snap(rbd_dev) && rbd_is_ro(rbd_dev))
1676 		return false;
1677 
1678 	return ((rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP) &&
1679 		!(rbd_dev->object_map_flags & RBD_FLAG_OBJECT_MAP_INVALID));
1680 }
1681 
1682 static bool rbd_object_map_may_exist(struct rbd_device *rbd_dev, u64 objno)
1683 {
1684 	u8 state;
1685 
1686 	/* fall back to default logic if object map is disabled or invalid */
1687 	if (!use_object_map(rbd_dev))
1688 		return true;
1689 
1690 	state = rbd_object_map_get(rbd_dev, objno);
1691 	return state != OBJECT_NONEXISTENT;
1692 }
1693 
1694 static void rbd_object_map_name(struct rbd_device *rbd_dev, u64 snap_id,
1695 				struct ceph_object_id *oid)
1696 {
1697 	if (snap_id == CEPH_NOSNAP)
1698 		ceph_oid_printf(oid, "%s%s", RBD_OBJECT_MAP_PREFIX,
1699 				rbd_dev->spec->image_id);
1700 	else
1701 		ceph_oid_printf(oid, "%s%s.%016llx", RBD_OBJECT_MAP_PREFIX,
1702 				rbd_dev->spec->image_id, snap_id);
1703 }
1704 
1705 static int rbd_object_map_lock(struct rbd_device *rbd_dev)
1706 {
1707 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
1708 	CEPH_DEFINE_OID_ONSTACK(oid);
1709 	u8 lock_type;
1710 	char *lock_tag;
1711 	struct ceph_locker *lockers;
1712 	u32 num_lockers;
1713 	bool broke_lock = false;
1714 	int ret;
1715 
1716 	rbd_object_map_name(rbd_dev, CEPH_NOSNAP, &oid);
1717 
1718 again:
1719 	ret = ceph_cls_lock(osdc, &oid, &rbd_dev->header_oloc, RBD_LOCK_NAME,
1720 			    CEPH_CLS_LOCK_EXCLUSIVE, "", "", "", 0);
1721 	if (ret != -EBUSY || broke_lock) {
1722 		if (ret == -EEXIST)
1723 			ret = 0; /* already locked by myself */
1724 		if (ret)
1725 			rbd_warn(rbd_dev, "failed to lock object map: %d", ret);
1726 		return ret;
1727 	}
1728 
1729 	ret = ceph_cls_lock_info(osdc, &oid, &rbd_dev->header_oloc,
1730 				 RBD_LOCK_NAME, &lock_type, &lock_tag,
1731 				 &lockers, &num_lockers);
1732 	if (ret) {
1733 		if (ret == -ENOENT)
1734 			goto again;
1735 
1736 		rbd_warn(rbd_dev, "failed to get object map lockers: %d", ret);
1737 		return ret;
1738 	}
1739 
1740 	kfree(lock_tag);
1741 	if (num_lockers == 0)
1742 		goto again;
1743 
1744 	rbd_warn(rbd_dev, "breaking object map lock owned by %s%llu",
1745 		 ENTITY_NAME(lockers[0].id.name));
1746 
1747 	ret = ceph_cls_break_lock(osdc, &oid, &rbd_dev->header_oloc,
1748 				  RBD_LOCK_NAME, lockers[0].id.cookie,
1749 				  &lockers[0].id.name);
1750 	ceph_free_lockers(lockers, num_lockers);
1751 	if (ret) {
1752 		if (ret == -ENOENT)
1753 			goto again;
1754 
1755 		rbd_warn(rbd_dev, "failed to break object map lock: %d", ret);
1756 		return ret;
1757 	}
1758 
1759 	broke_lock = true;
1760 	goto again;
1761 }
1762 
1763 static void rbd_object_map_unlock(struct rbd_device *rbd_dev)
1764 {
1765 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
1766 	CEPH_DEFINE_OID_ONSTACK(oid);
1767 	int ret;
1768 
1769 	rbd_object_map_name(rbd_dev, CEPH_NOSNAP, &oid);
1770 
1771 	ret = ceph_cls_unlock(osdc, &oid, &rbd_dev->header_oloc, RBD_LOCK_NAME,
1772 			      "");
1773 	if (ret && ret != -ENOENT)
1774 		rbd_warn(rbd_dev, "failed to unlock object map: %d", ret);
1775 }
1776 
1777 static int decode_object_map_header(void **p, void *end, u64 *object_map_size)
1778 {
1779 	u8 struct_v;
1780 	u32 struct_len;
1781 	u32 header_len;
1782 	void *header_end;
1783 	int ret;
1784 
1785 	ceph_decode_32_safe(p, end, header_len, e_inval);
1786 	header_end = *p + header_len;
1787 
1788 	ret = ceph_start_decoding(p, end, 1, "BitVector header", &struct_v,
1789 				  &struct_len);
1790 	if (ret)
1791 		return ret;
1792 
1793 	ceph_decode_64_safe(p, end, *object_map_size, e_inval);
1794 
1795 	*p = header_end;
1796 	return 0;
1797 
1798 e_inval:
1799 	return -EINVAL;
1800 }
1801 
1802 static int __rbd_object_map_load(struct rbd_device *rbd_dev)
1803 {
1804 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
1805 	CEPH_DEFINE_OID_ONSTACK(oid);
1806 	struct page **pages;
1807 	void *p, *end;
1808 	size_t reply_len;
1809 	u64 num_objects;
1810 	u64 object_map_bytes;
1811 	u64 object_map_size;
1812 	int num_pages;
1813 	int ret;
1814 
1815 	rbd_assert(!rbd_dev->object_map && !rbd_dev->object_map_size);
1816 
1817 	num_objects = ceph_get_num_objects(&rbd_dev->layout,
1818 					   rbd_dev->mapping.size);
1819 	object_map_bytes = DIV_ROUND_UP_ULL(num_objects * BITS_PER_OBJ,
1820 					    BITS_PER_BYTE);
1821 	num_pages = calc_pages_for(0, object_map_bytes) + 1;
1822 	pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
1823 	if (IS_ERR(pages))
1824 		return PTR_ERR(pages);
1825 
1826 	reply_len = num_pages * PAGE_SIZE;
1827 	rbd_object_map_name(rbd_dev, rbd_dev->spec->snap_id, &oid);
1828 	ret = ceph_osdc_call(osdc, &oid, &rbd_dev->header_oloc,
1829 			     "rbd", "object_map_load", CEPH_OSD_FLAG_READ,
1830 			     NULL, 0, pages, &reply_len);
1831 	if (ret)
1832 		goto out;
1833 
1834 	p = page_address(pages[0]);
1835 	end = p + min(reply_len, (size_t)PAGE_SIZE);
1836 	ret = decode_object_map_header(&p, end, &object_map_size);
1837 	if (ret)
1838 		goto out;
1839 
1840 	if (object_map_size != num_objects) {
1841 		rbd_warn(rbd_dev, "object map size mismatch: %llu vs %llu",
1842 			 object_map_size, num_objects);
1843 		ret = -EINVAL;
1844 		goto out;
1845 	}
1846 
1847 	if (offset_in_page(p) + object_map_bytes > reply_len) {
1848 		ret = -EINVAL;
1849 		goto out;
1850 	}
1851 
1852 	rbd_dev->object_map = kvmalloc(object_map_bytes, GFP_KERNEL);
1853 	if (!rbd_dev->object_map) {
1854 		ret = -ENOMEM;
1855 		goto out;
1856 	}
1857 
1858 	rbd_dev->object_map_size = object_map_size;
1859 	ceph_copy_from_page_vector(pages, rbd_dev->object_map,
1860 				   offset_in_page(p), object_map_bytes);
1861 
1862 out:
1863 	ceph_release_page_vector(pages, num_pages);
1864 	return ret;
1865 }
1866 
1867 static void rbd_object_map_free(struct rbd_device *rbd_dev)
1868 {
1869 	kvfree(rbd_dev->object_map);
1870 	rbd_dev->object_map = NULL;
1871 	rbd_dev->object_map_size = 0;
1872 }
1873 
1874 static int rbd_object_map_load(struct rbd_device *rbd_dev)
1875 {
1876 	int ret;
1877 
1878 	ret = __rbd_object_map_load(rbd_dev);
1879 	if (ret)
1880 		return ret;
1881 
1882 	ret = rbd_dev_v2_get_flags(rbd_dev);
1883 	if (ret) {
1884 		rbd_object_map_free(rbd_dev);
1885 		return ret;
1886 	}
1887 
1888 	if (rbd_dev->object_map_flags & RBD_FLAG_OBJECT_MAP_INVALID)
1889 		rbd_warn(rbd_dev, "object map is invalid");
1890 
1891 	return 0;
1892 }
1893 
1894 static int rbd_object_map_open(struct rbd_device *rbd_dev)
1895 {
1896 	int ret;
1897 
1898 	ret = rbd_object_map_lock(rbd_dev);
1899 	if (ret)
1900 		return ret;
1901 
1902 	ret = rbd_object_map_load(rbd_dev);
1903 	if (ret) {
1904 		rbd_object_map_unlock(rbd_dev);
1905 		return ret;
1906 	}
1907 
1908 	return 0;
1909 }
1910 
1911 static void rbd_object_map_close(struct rbd_device *rbd_dev)
1912 {
1913 	rbd_object_map_free(rbd_dev);
1914 	rbd_object_map_unlock(rbd_dev);
1915 }
1916 
1917 /*
1918  * This function needs snap_id (or more precisely just something to
1919  * distinguish between HEAD and snapshot object maps), new_state and
1920  * current_state that were passed to rbd_object_map_update().
1921  *
1922  * To avoid allocating and stashing a context we piggyback on the OSD
1923  * request.  A HEAD update has two ops (assert_locked).  For new_state
1924  * and current_state we decode our own object_map_update op, encoded in
1925  * rbd_cls_object_map_update().
1926  */
1927 static int rbd_object_map_update_finish(struct rbd_obj_request *obj_req,
1928 					struct ceph_osd_request *osd_req)
1929 {
1930 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
1931 	struct ceph_osd_data *osd_data;
1932 	u64 objno;
1933 	u8 state, new_state, current_state;
1934 	bool has_current_state;
1935 	void *p;
1936 
1937 	if (osd_req->r_result)
1938 		return osd_req->r_result;
1939 
1940 	/*
1941 	 * Nothing to do for a snapshot object map.
1942 	 */
1943 	if (osd_req->r_num_ops == 1)
1944 		return 0;
1945 
1946 	/*
1947 	 * Update in-memory HEAD object map.
1948 	 */
1949 	rbd_assert(osd_req->r_num_ops == 2);
1950 	osd_data = osd_req_op_data(osd_req, 1, cls, request_data);
1951 	rbd_assert(osd_data->type == CEPH_OSD_DATA_TYPE_PAGES);
1952 
1953 	p = page_address(osd_data->pages[0]);
1954 	objno = ceph_decode_64(&p);
1955 	rbd_assert(objno == obj_req->ex.oe_objno);
1956 	rbd_assert(ceph_decode_64(&p) == objno + 1);
1957 	new_state = ceph_decode_8(&p);
1958 	has_current_state = ceph_decode_8(&p);
1959 	if (has_current_state)
1960 		current_state = ceph_decode_8(&p);
1961 
1962 	spin_lock(&rbd_dev->object_map_lock);
1963 	state = __rbd_object_map_get(rbd_dev, objno);
1964 	if (!has_current_state || current_state == state ||
1965 	    (current_state == OBJECT_EXISTS && state == OBJECT_EXISTS_CLEAN))
1966 		__rbd_object_map_set(rbd_dev, objno, new_state);
1967 	spin_unlock(&rbd_dev->object_map_lock);
1968 
1969 	return 0;
1970 }
1971 
1972 static void rbd_object_map_callback(struct ceph_osd_request *osd_req)
1973 {
1974 	struct rbd_obj_request *obj_req = osd_req->r_priv;
1975 	int result;
1976 
1977 	dout("%s osd_req %p result %d for obj_req %p\n", __func__, osd_req,
1978 	     osd_req->r_result, obj_req);
1979 
1980 	result = rbd_object_map_update_finish(obj_req, osd_req);
1981 	rbd_obj_handle_request(obj_req, result);
1982 }
1983 
1984 static bool update_needed(struct rbd_device *rbd_dev, u64 objno, u8 new_state)
1985 {
1986 	u8 state = rbd_object_map_get(rbd_dev, objno);
1987 
1988 	if (state == new_state ||
1989 	    (new_state == OBJECT_PENDING && state == OBJECT_NONEXISTENT) ||
1990 	    (new_state == OBJECT_NONEXISTENT && state != OBJECT_PENDING))
1991 		return false;
1992 
1993 	return true;
1994 }
1995 
1996 static int rbd_cls_object_map_update(struct ceph_osd_request *req,
1997 				     int which, u64 objno, u8 new_state,
1998 				     const u8 *current_state)
1999 {
2000 	struct page **pages;
2001 	void *p, *start;
2002 	int ret;
2003 
2004 	ret = osd_req_op_cls_init(req, which, "rbd", "object_map_update");
2005 	if (ret)
2006 		return ret;
2007 
2008 	pages = ceph_alloc_page_vector(1, GFP_NOIO);
2009 	if (IS_ERR(pages))
2010 		return PTR_ERR(pages);
2011 
2012 	p = start = page_address(pages[0]);
2013 	ceph_encode_64(&p, objno);
2014 	ceph_encode_64(&p, objno + 1);
2015 	ceph_encode_8(&p, new_state);
2016 	if (current_state) {
2017 		ceph_encode_8(&p, 1);
2018 		ceph_encode_8(&p, *current_state);
2019 	} else {
2020 		ceph_encode_8(&p, 0);
2021 	}
2022 
2023 	osd_req_op_cls_request_data_pages(req, which, pages, p - start, 0,
2024 					  false, true);
2025 	return 0;
2026 }
2027 
2028 /*
2029  * Return:
2030  *   0 - object map update sent
2031  *   1 - object map update isn't needed
2032  *  <0 - error
2033  */
2034 static int rbd_object_map_update(struct rbd_obj_request *obj_req, u64 snap_id,
2035 				 u8 new_state, const u8 *current_state)
2036 {
2037 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
2038 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
2039 	struct ceph_osd_request *req;
2040 	int num_ops = 1;
2041 	int which = 0;
2042 	int ret;
2043 
2044 	if (snap_id == CEPH_NOSNAP) {
2045 		if (!update_needed(rbd_dev, obj_req->ex.oe_objno, new_state))
2046 			return 1;
2047 
2048 		num_ops++; /* assert_locked */
2049 	}
2050 
2051 	req = ceph_osdc_alloc_request(osdc, NULL, num_ops, false, GFP_NOIO);
2052 	if (!req)
2053 		return -ENOMEM;
2054 
2055 	list_add_tail(&req->r_private_item, &obj_req->osd_reqs);
2056 	req->r_callback = rbd_object_map_callback;
2057 	req->r_priv = obj_req;
2058 
2059 	rbd_object_map_name(rbd_dev, snap_id, &req->r_base_oid);
2060 	ceph_oloc_copy(&req->r_base_oloc, &rbd_dev->header_oloc);
2061 	req->r_flags = CEPH_OSD_FLAG_WRITE;
2062 	ktime_get_real_ts64(&req->r_mtime);
2063 
2064 	if (snap_id == CEPH_NOSNAP) {
2065 		/*
2066 		 * Protect against possible race conditions during lock
2067 		 * ownership transitions.
2068 		 */
2069 		ret = ceph_cls_assert_locked(req, which++, RBD_LOCK_NAME,
2070 					     CEPH_CLS_LOCK_EXCLUSIVE, "", "");
2071 		if (ret)
2072 			return ret;
2073 	}
2074 
2075 	ret = rbd_cls_object_map_update(req, which, obj_req->ex.oe_objno,
2076 					new_state, current_state);
2077 	if (ret)
2078 		return ret;
2079 
2080 	ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
2081 	if (ret)
2082 		return ret;
2083 
2084 	ceph_osdc_start_request(osdc, req, false);
2085 	return 0;
2086 }
2087 
2088 static void prune_extents(struct ceph_file_extent *img_extents,
2089 			  u32 *num_img_extents, u64 overlap)
2090 {
2091 	u32 cnt = *num_img_extents;
2092 
2093 	/* drop extents completely beyond the overlap */
2094 	while (cnt && img_extents[cnt - 1].fe_off >= overlap)
2095 		cnt--;
2096 
2097 	if (cnt) {
2098 		struct ceph_file_extent *ex = &img_extents[cnt - 1];
2099 
2100 		/* trim final overlapping extent */
2101 		if (ex->fe_off + ex->fe_len > overlap)
2102 			ex->fe_len = overlap - ex->fe_off;
2103 	}
2104 
2105 	*num_img_extents = cnt;
2106 }
2107 
2108 /*
2109  * Determine the byte range(s) covered by either just the object extent
2110  * or the entire object in the parent image.
2111  */
2112 static int rbd_obj_calc_img_extents(struct rbd_obj_request *obj_req,
2113 				    bool entire)
2114 {
2115 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
2116 	int ret;
2117 
2118 	if (!rbd_dev->parent_overlap)
2119 		return 0;
2120 
2121 	ret = ceph_extent_to_file(&rbd_dev->layout, obj_req->ex.oe_objno,
2122 				  entire ? 0 : obj_req->ex.oe_off,
2123 				  entire ? rbd_dev->layout.object_size :
2124 							obj_req->ex.oe_len,
2125 				  &obj_req->img_extents,
2126 				  &obj_req->num_img_extents);
2127 	if (ret)
2128 		return ret;
2129 
2130 	prune_extents(obj_req->img_extents, &obj_req->num_img_extents,
2131 		      rbd_dev->parent_overlap);
2132 	return 0;
2133 }
2134 
2135 static void rbd_osd_setup_data(struct ceph_osd_request *osd_req, int which)
2136 {
2137 	struct rbd_obj_request *obj_req = osd_req->r_priv;
2138 
2139 	switch (obj_req->img_request->data_type) {
2140 	case OBJ_REQUEST_BIO:
2141 		osd_req_op_extent_osd_data_bio(osd_req, which,
2142 					       &obj_req->bio_pos,
2143 					       obj_req->ex.oe_len);
2144 		break;
2145 	case OBJ_REQUEST_BVECS:
2146 	case OBJ_REQUEST_OWN_BVECS:
2147 		rbd_assert(obj_req->bvec_pos.iter.bi_size ==
2148 							obj_req->ex.oe_len);
2149 		rbd_assert(obj_req->bvec_idx == obj_req->bvec_count);
2150 		osd_req_op_extent_osd_data_bvec_pos(osd_req, which,
2151 						    &obj_req->bvec_pos);
2152 		break;
2153 	default:
2154 		BUG();
2155 	}
2156 }
2157 
2158 static int rbd_osd_setup_stat(struct ceph_osd_request *osd_req, int which)
2159 {
2160 	struct page **pages;
2161 
2162 	/*
2163 	 * The response data for a STAT call consists of:
2164 	 *     le64 length;
2165 	 *     struct {
2166 	 *         le32 tv_sec;
2167 	 *         le32 tv_nsec;
2168 	 *     } mtime;
2169 	 */
2170 	pages = ceph_alloc_page_vector(1, GFP_NOIO);
2171 	if (IS_ERR(pages))
2172 		return PTR_ERR(pages);
2173 
2174 	osd_req_op_init(osd_req, which, CEPH_OSD_OP_STAT, 0);
2175 	osd_req_op_raw_data_in_pages(osd_req, which, pages,
2176 				     8 + sizeof(struct ceph_timespec),
2177 				     0, false, true);
2178 	return 0;
2179 }
2180 
2181 static int rbd_osd_setup_copyup(struct ceph_osd_request *osd_req, int which,
2182 				u32 bytes)
2183 {
2184 	struct rbd_obj_request *obj_req = osd_req->r_priv;
2185 	int ret;
2186 
2187 	ret = osd_req_op_cls_init(osd_req, which, "rbd", "copyup");
2188 	if (ret)
2189 		return ret;
2190 
2191 	osd_req_op_cls_request_data_bvecs(osd_req, which, obj_req->copyup_bvecs,
2192 					  obj_req->copyup_bvec_count, bytes);
2193 	return 0;
2194 }
2195 
2196 static int rbd_obj_init_read(struct rbd_obj_request *obj_req)
2197 {
2198 	obj_req->read_state = RBD_OBJ_READ_START;
2199 	return 0;
2200 }
2201 
2202 static void __rbd_osd_setup_write_ops(struct ceph_osd_request *osd_req,
2203 				      int which)
2204 {
2205 	struct rbd_obj_request *obj_req = osd_req->r_priv;
2206 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
2207 	u16 opcode;
2208 
2209 	if (!use_object_map(rbd_dev) ||
2210 	    !(obj_req->flags & RBD_OBJ_FLAG_MAY_EXIST)) {
2211 		osd_req_op_alloc_hint_init(osd_req, which++,
2212 					   rbd_dev->layout.object_size,
2213 					   rbd_dev->layout.object_size,
2214 					   rbd_dev->opts->alloc_hint_flags);
2215 	}
2216 
2217 	if (rbd_obj_is_entire(obj_req))
2218 		opcode = CEPH_OSD_OP_WRITEFULL;
2219 	else
2220 		opcode = CEPH_OSD_OP_WRITE;
2221 
2222 	osd_req_op_extent_init(osd_req, which, opcode,
2223 			       obj_req->ex.oe_off, obj_req->ex.oe_len, 0, 0);
2224 	rbd_osd_setup_data(osd_req, which);
2225 }
2226 
2227 static int rbd_obj_init_write(struct rbd_obj_request *obj_req)
2228 {
2229 	int ret;
2230 
2231 	/* reverse map the entire object onto the parent */
2232 	ret = rbd_obj_calc_img_extents(obj_req, true);
2233 	if (ret)
2234 		return ret;
2235 
2236 	if (rbd_obj_copyup_enabled(obj_req))
2237 		obj_req->flags |= RBD_OBJ_FLAG_COPYUP_ENABLED;
2238 
2239 	obj_req->write_state = RBD_OBJ_WRITE_START;
2240 	return 0;
2241 }
2242 
2243 static u16 truncate_or_zero_opcode(struct rbd_obj_request *obj_req)
2244 {
2245 	return rbd_obj_is_tail(obj_req) ? CEPH_OSD_OP_TRUNCATE :
2246 					  CEPH_OSD_OP_ZERO;
2247 }
2248 
2249 static void __rbd_osd_setup_discard_ops(struct ceph_osd_request *osd_req,
2250 					int which)
2251 {
2252 	struct rbd_obj_request *obj_req = osd_req->r_priv;
2253 
2254 	if (rbd_obj_is_entire(obj_req) && !obj_req->num_img_extents) {
2255 		rbd_assert(obj_req->flags & RBD_OBJ_FLAG_DELETION);
2256 		osd_req_op_init(osd_req, which, CEPH_OSD_OP_DELETE, 0);
2257 	} else {
2258 		osd_req_op_extent_init(osd_req, which,
2259 				       truncate_or_zero_opcode(obj_req),
2260 				       obj_req->ex.oe_off, obj_req->ex.oe_len,
2261 				       0, 0);
2262 	}
2263 }
2264 
2265 static int rbd_obj_init_discard(struct rbd_obj_request *obj_req)
2266 {
2267 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
2268 	u64 off, next_off;
2269 	int ret;
2270 
2271 	/*
2272 	 * Align the range to alloc_size boundary and punt on discards
2273 	 * that are too small to free up any space.
2274 	 *
2275 	 * alloc_size == object_size && is_tail() is a special case for
2276 	 * filestore with filestore_punch_hole = false, needed to allow
2277 	 * truncate (in addition to delete).
2278 	 */
2279 	if (rbd_dev->opts->alloc_size != rbd_dev->layout.object_size ||
2280 	    !rbd_obj_is_tail(obj_req)) {
2281 		off = round_up(obj_req->ex.oe_off, rbd_dev->opts->alloc_size);
2282 		next_off = round_down(obj_req->ex.oe_off + obj_req->ex.oe_len,
2283 				      rbd_dev->opts->alloc_size);
2284 		if (off >= next_off)
2285 			return 1;
2286 
2287 		dout("%s %p %llu~%llu -> %llu~%llu\n", __func__,
2288 		     obj_req, obj_req->ex.oe_off, obj_req->ex.oe_len,
2289 		     off, next_off - off);
2290 		obj_req->ex.oe_off = off;
2291 		obj_req->ex.oe_len = next_off - off;
2292 	}
2293 
2294 	/* reverse map the entire object onto the parent */
2295 	ret = rbd_obj_calc_img_extents(obj_req, true);
2296 	if (ret)
2297 		return ret;
2298 
2299 	obj_req->flags |= RBD_OBJ_FLAG_NOOP_FOR_NONEXISTENT;
2300 	if (rbd_obj_is_entire(obj_req) && !obj_req->num_img_extents)
2301 		obj_req->flags |= RBD_OBJ_FLAG_DELETION;
2302 
2303 	obj_req->write_state = RBD_OBJ_WRITE_START;
2304 	return 0;
2305 }
2306 
2307 static void __rbd_osd_setup_zeroout_ops(struct ceph_osd_request *osd_req,
2308 					int which)
2309 {
2310 	struct rbd_obj_request *obj_req = osd_req->r_priv;
2311 	u16 opcode;
2312 
2313 	if (rbd_obj_is_entire(obj_req)) {
2314 		if (obj_req->num_img_extents) {
2315 			if (!(obj_req->flags & RBD_OBJ_FLAG_COPYUP_ENABLED))
2316 				osd_req_op_init(osd_req, which++,
2317 						CEPH_OSD_OP_CREATE, 0);
2318 			opcode = CEPH_OSD_OP_TRUNCATE;
2319 		} else {
2320 			rbd_assert(obj_req->flags & RBD_OBJ_FLAG_DELETION);
2321 			osd_req_op_init(osd_req, which++,
2322 					CEPH_OSD_OP_DELETE, 0);
2323 			opcode = 0;
2324 		}
2325 	} else {
2326 		opcode = truncate_or_zero_opcode(obj_req);
2327 	}
2328 
2329 	if (opcode)
2330 		osd_req_op_extent_init(osd_req, which, opcode,
2331 				       obj_req->ex.oe_off, obj_req->ex.oe_len,
2332 				       0, 0);
2333 }
2334 
2335 static int rbd_obj_init_zeroout(struct rbd_obj_request *obj_req)
2336 {
2337 	int ret;
2338 
2339 	/* reverse map the entire object onto the parent */
2340 	ret = rbd_obj_calc_img_extents(obj_req, true);
2341 	if (ret)
2342 		return ret;
2343 
2344 	if (rbd_obj_copyup_enabled(obj_req))
2345 		obj_req->flags |= RBD_OBJ_FLAG_COPYUP_ENABLED;
2346 	if (!obj_req->num_img_extents) {
2347 		obj_req->flags |= RBD_OBJ_FLAG_NOOP_FOR_NONEXISTENT;
2348 		if (rbd_obj_is_entire(obj_req))
2349 			obj_req->flags |= RBD_OBJ_FLAG_DELETION;
2350 	}
2351 
2352 	obj_req->write_state = RBD_OBJ_WRITE_START;
2353 	return 0;
2354 }
2355 
2356 static int count_write_ops(struct rbd_obj_request *obj_req)
2357 {
2358 	struct rbd_img_request *img_req = obj_req->img_request;
2359 
2360 	switch (img_req->op_type) {
2361 	case OBJ_OP_WRITE:
2362 		if (!use_object_map(img_req->rbd_dev) ||
2363 		    !(obj_req->flags & RBD_OBJ_FLAG_MAY_EXIST))
2364 			return 2; /* setallochint + write/writefull */
2365 
2366 		return 1; /* write/writefull */
2367 	case OBJ_OP_DISCARD:
2368 		return 1; /* delete/truncate/zero */
2369 	case OBJ_OP_ZEROOUT:
2370 		if (rbd_obj_is_entire(obj_req) && obj_req->num_img_extents &&
2371 		    !(obj_req->flags & RBD_OBJ_FLAG_COPYUP_ENABLED))
2372 			return 2; /* create + truncate */
2373 
2374 		return 1; /* delete/truncate/zero */
2375 	default:
2376 		BUG();
2377 	}
2378 }
2379 
2380 static void rbd_osd_setup_write_ops(struct ceph_osd_request *osd_req,
2381 				    int which)
2382 {
2383 	struct rbd_obj_request *obj_req = osd_req->r_priv;
2384 
2385 	switch (obj_req->img_request->op_type) {
2386 	case OBJ_OP_WRITE:
2387 		__rbd_osd_setup_write_ops(osd_req, which);
2388 		break;
2389 	case OBJ_OP_DISCARD:
2390 		__rbd_osd_setup_discard_ops(osd_req, which);
2391 		break;
2392 	case OBJ_OP_ZEROOUT:
2393 		__rbd_osd_setup_zeroout_ops(osd_req, which);
2394 		break;
2395 	default:
2396 		BUG();
2397 	}
2398 }
2399 
2400 /*
2401  * Prune the list of object requests (adjust offset and/or length, drop
2402  * redundant requests).  Prepare object request state machines and image
2403  * request state machine for execution.
2404  */
2405 static int __rbd_img_fill_request(struct rbd_img_request *img_req)
2406 {
2407 	struct rbd_obj_request *obj_req, *next_obj_req;
2408 	int ret;
2409 
2410 	for_each_obj_request_safe(img_req, obj_req, next_obj_req) {
2411 		switch (img_req->op_type) {
2412 		case OBJ_OP_READ:
2413 			ret = rbd_obj_init_read(obj_req);
2414 			break;
2415 		case OBJ_OP_WRITE:
2416 			ret = rbd_obj_init_write(obj_req);
2417 			break;
2418 		case OBJ_OP_DISCARD:
2419 			ret = rbd_obj_init_discard(obj_req);
2420 			break;
2421 		case OBJ_OP_ZEROOUT:
2422 			ret = rbd_obj_init_zeroout(obj_req);
2423 			break;
2424 		default:
2425 			BUG();
2426 		}
2427 		if (ret < 0)
2428 			return ret;
2429 		if (ret > 0) {
2430 			rbd_img_obj_request_del(img_req, obj_req);
2431 			continue;
2432 		}
2433 	}
2434 
2435 	img_req->state = RBD_IMG_START;
2436 	return 0;
2437 }
2438 
2439 union rbd_img_fill_iter {
2440 	struct ceph_bio_iter	bio_iter;
2441 	struct ceph_bvec_iter	bvec_iter;
2442 };
2443 
2444 struct rbd_img_fill_ctx {
2445 	enum obj_request_type	pos_type;
2446 	union rbd_img_fill_iter	*pos;
2447 	union rbd_img_fill_iter	iter;
2448 	ceph_object_extent_fn_t	set_pos_fn;
2449 	ceph_object_extent_fn_t	count_fn;
2450 	ceph_object_extent_fn_t	copy_fn;
2451 };
2452 
2453 static struct ceph_object_extent *alloc_object_extent(void *arg)
2454 {
2455 	struct rbd_img_request *img_req = arg;
2456 	struct rbd_obj_request *obj_req;
2457 
2458 	obj_req = rbd_obj_request_create();
2459 	if (!obj_req)
2460 		return NULL;
2461 
2462 	rbd_img_obj_request_add(img_req, obj_req);
2463 	return &obj_req->ex;
2464 }
2465 
2466 /*
2467  * While su != os && sc == 1 is technically not fancy (it's the same
2468  * layout as su == os && sc == 1), we can't use the nocopy path for it
2469  * because ->set_pos_fn() should be called only once per object.
2470  * ceph_file_to_extents() invokes action_fn once per stripe unit, so
2471  * treat su != os && sc == 1 as fancy.
2472  */
2473 static bool rbd_layout_is_fancy(struct ceph_file_layout *l)
2474 {
2475 	return l->stripe_unit != l->object_size;
2476 }
2477 
2478 static int rbd_img_fill_request_nocopy(struct rbd_img_request *img_req,
2479 				       struct ceph_file_extent *img_extents,
2480 				       u32 num_img_extents,
2481 				       struct rbd_img_fill_ctx *fctx)
2482 {
2483 	u32 i;
2484 	int ret;
2485 
2486 	img_req->data_type = fctx->pos_type;
2487 
2488 	/*
2489 	 * Create object requests and set each object request's starting
2490 	 * position in the provided bio (list) or bio_vec array.
2491 	 */
2492 	fctx->iter = *fctx->pos;
2493 	for (i = 0; i < num_img_extents; i++) {
2494 		ret = ceph_file_to_extents(&img_req->rbd_dev->layout,
2495 					   img_extents[i].fe_off,
2496 					   img_extents[i].fe_len,
2497 					   &img_req->object_extents,
2498 					   alloc_object_extent, img_req,
2499 					   fctx->set_pos_fn, &fctx->iter);
2500 		if (ret)
2501 			return ret;
2502 	}
2503 
2504 	return __rbd_img_fill_request(img_req);
2505 }
2506 
2507 /*
2508  * Map a list of image extents to a list of object extents, create the
2509  * corresponding object requests (normally each to a different object,
2510  * but not always) and add them to @img_req.  For each object request,
2511  * set up its data descriptor to point to the corresponding chunk(s) of
2512  * @fctx->pos data buffer.
2513  *
2514  * Because ceph_file_to_extents() will merge adjacent object extents
2515  * together, each object request's data descriptor may point to multiple
2516  * different chunks of @fctx->pos data buffer.
2517  *
2518  * @fctx->pos data buffer is assumed to be large enough.
2519  */
2520 static int rbd_img_fill_request(struct rbd_img_request *img_req,
2521 				struct ceph_file_extent *img_extents,
2522 				u32 num_img_extents,
2523 				struct rbd_img_fill_ctx *fctx)
2524 {
2525 	struct rbd_device *rbd_dev = img_req->rbd_dev;
2526 	struct rbd_obj_request *obj_req;
2527 	u32 i;
2528 	int ret;
2529 
2530 	if (fctx->pos_type == OBJ_REQUEST_NODATA ||
2531 	    !rbd_layout_is_fancy(&rbd_dev->layout))
2532 		return rbd_img_fill_request_nocopy(img_req, img_extents,
2533 						   num_img_extents, fctx);
2534 
2535 	img_req->data_type = OBJ_REQUEST_OWN_BVECS;
2536 
2537 	/*
2538 	 * Create object requests and determine ->bvec_count for each object
2539 	 * request.  Note that ->bvec_count sum over all object requests may
2540 	 * be greater than the number of bio_vecs in the provided bio (list)
2541 	 * or bio_vec array because when mapped, those bio_vecs can straddle
2542 	 * stripe unit boundaries.
2543 	 */
2544 	fctx->iter = *fctx->pos;
2545 	for (i = 0; i < num_img_extents; i++) {
2546 		ret = ceph_file_to_extents(&rbd_dev->layout,
2547 					   img_extents[i].fe_off,
2548 					   img_extents[i].fe_len,
2549 					   &img_req->object_extents,
2550 					   alloc_object_extent, img_req,
2551 					   fctx->count_fn, &fctx->iter);
2552 		if (ret)
2553 			return ret;
2554 	}
2555 
2556 	for_each_obj_request(img_req, obj_req) {
2557 		obj_req->bvec_pos.bvecs = kmalloc_array(obj_req->bvec_count,
2558 					      sizeof(*obj_req->bvec_pos.bvecs),
2559 					      GFP_NOIO);
2560 		if (!obj_req->bvec_pos.bvecs)
2561 			return -ENOMEM;
2562 	}
2563 
2564 	/*
2565 	 * Fill in each object request's private bio_vec array, splitting and
2566 	 * rearranging the provided bio_vecs in stripe unit chunks as needed.
2567 	 */
2568 	fctx->iter = *fctx->pos;
2569 	for (i = 0; i < num_img_extents; i++) {
2570 		ret = ceph_iterate_extents(&rbd_dev->layout,
2571 					   img_extents[i].fe_off,
2572 					   img_extents[i].fe_len,
2573 					   &img_req->object_extents,
2574 					   fctx->copy_fn, &fctx->iter);
2575 		if (ret)
2576 			return ret;
2577 	}
2578 
2579 	return __rbd_img_fill_request(img_req);
2580 }
2581 
2582 static int rbd_img_fill_nodata(struct rbd_img_request *img_req,
2583 			       u64 off, u64 len)
2584 {
2585 	struct ceph_file_extent ex = { off, len };
2586 	union rbd_img_fill_iter dummy = {};
2587 	struct rbd_img_fill_ctx fctx = {
2588 		.pos_type = OBJ_REQUEST_NODATA,
2589 		.pos = &dummy,
2590 	};
2591 
2592 	return rbd_img_fill_request(img_req, &ex, 1, &fctx);
2593 }
2594 
2595 static void set_bio_pos(struct ceph_object_extent *ex, u32 bytes, void *arg)
2596 {
2597 	struct rbd_obj_request *obj_req =
2598 	    container_of(ex, struct rbd_obj_request, ex);
2599 	struct ceph_bio_iter *it = arg;
2600 
2601 	dout("%s objno %llu bytes %u\n", __func__, ex->oe_objno, bytes);
2602 	obj_req->bio_pos = *it;
2603 	ceph_bio_iter_advance(it, bytes);
2604 }
2605 
2606 static void count_bio_bvecs(struct ceph_object_extent *ex, u32 bytes, void *arg)
2607 {
2608 	struct rbd_obj_request *obj_req =
2609 	    container_of(ex, struct rbd_obj_request, ex);
2610 	struct ceph_bio_iter *it = arg;
2611 
2612 	dout("%s objno %llu bytes %u\n", __func__, ex->oe_objno, bytes);
2613 	ceph_bio_iter_advance_step(it, bytes, ({
2614 		obj_req->bvec_count++;
2615 	}));
2616 
2617 }
2618 
2619 static void copy_bio_bvecs(struct ceph_object_extent *ex, u32 bytes, void *arg)
2620 {
2621 	struct rbd_obj_request *obj_req =
2622 	    container_of(ex, struct rbd_obj_request, ex);
2623 	struct ceph_bio_iter *it = arg;
2624 
2625 	dout("%s objno %llu bytes %u\n", __func__, ex->oe_objno, bytes);
2626 	ceph_bio_iter_advance_step(it, bytes, ({
2627 		obj_req->bvec_pos.bvecs[obj_req->bvec_idx++] = bv;
2628 		obj_req->bvec_pos.iter.bi_size += bv.bv_len;
2629 	}));
2630 }
2631 
2632 static int __rbd_img_fill_from_bio(struct rbd_img_request *img_req,
2633 				   struct ceph_file_extent *img_extents,
2634 				   u32 num_img_extents,
2635 				   struct ceph_bio_iter *bio_pos)
2636 {
2637 	struct rbd_img_fill_ctx fctx = {
2638 		.pos_type = OBJ_REQUEST_BIO,
2639 		.pos = (union rbd_img_fill_iter *)bio_pos,
2640 		.set_pos_fn = set_bio_pos,
2641 		.count_fn = count_bio_bvecs,
2642 		.copy_fn = copy_bio_bvecs,
2643 	};
2644 
2645 	return rbd_img_fill_request(img_req, img_extents, num_img_extents,
2646 				    &fctx);
2647 }
2648 
2649 static int rbd_img_fill_from_bio(struct rbd_img_request *img_req,
2650 				 u64 off, u64 len, struct bio *bio)
2651 {
2652 	struct ceph_file_extent ex = { off, len };
2653 	struct ceph_bio_iter it = { .bio = bio, .iter = bio->bi_iter };
2654 
2655 	return __rbd_img_fill_from_bio(img_req, &ex, 1, &it);
2656 }
2657 
2658 static void set_bvec_pos(struct ceph_object_extent *ex, u32 bytes, void *arg)
2659 {
2660 	struct rbd_obj_request *obj_req =
2661 	    container_of(ex, struct rbd_obj_request, ex);
2662 	struct ceph_bvec_iter *it = arg;
2663 
2664 	obj_req->bvec_pos = *it;
2665 	ceph_bvec_iter_shorten(&obj_req->bvec_pos, bytes);
2666 	ceph_bvec_iter_advance(it, bytes);
2667 }
2668 
2669 static void count_bvecs(struct ceph_object_extent *ex, u32 bytes, void *arg)
2670 {
2671 	struct rbd_obj_request *obj_req =
2672 	    container_of(ex, struct rbd_obj_request, ex);
2673 	struct ceph_bvec_iter *it = arg;
2674 
2675 	ceph_bvec_iter_advance_step(it, bytes, ({
2676 		obj_req->bvec_count++;
2677 	}));
2678 }
2679 
2680 static void copy_bvecs(struct ceph_object_extent *ex, u32 bytes, void *arg)
2681 {
2682 	struct rbd_obj_request *obj_req =
2683 	    container_of(ex, struct rbd_obj_request, ex);
2684 	struct ceph_bvec_iter *it = arg;
2685 
2686 	ceph_bvec_iter_advance_step(it, bytes, ({
2687 		obj_req->bvec_pos.bvecs[obj_req->bvec_idx++] = bv;
2688 		obj_req->bvec_pos.iter.bi_size += bv.bv_len;
2689 	}));
2690 }
2691 
2692 static int __rbd_img_fill_from_bvecs(struct rbd_img_request *img_req,
2693 				     struct ceph_file_extent *img_extents,
2694 				     u32 num_img_extents,
2695 				     struct ceph_bvec_iter *bvec_pos)
2696 {
2697 	struct rbd_img_fill_ctx fctx = {
2698 		.pos_type = OBJ_REQUEST_BVECS,
2699 		.pos = (union rbd_img_fill_iter *)bvec_pos,
2700 		.set_pos_fn = set_bvec_pos,
2701 		.count_fn = count_bvecs,
2702 		.copy_fn = copy_bvecs,
2703 	};
2704 
2705 	return rbd_img_fill_request(img_req, img_extents, num_img_extents,
2706 				    &fctx);
2707 }
2708 
2709 static int rbd_img_fill_from_bvecs(struct rbd_img_request *img_req,
2710 				   struct ceph_file_extent *img_extents,
2711 				   u32 num_img_extents,
2712 				   struct bio_vec *bvecs)
2713 {
2714 	struct ceph_bvec_iter it = {
2715 		.bvecs = bvecs,
2716 		.iter = { .bi_size = ceph_file_extents_bytes(img_extents,
2717 							     num_img_extents) },
2718 	};
2719 
2720 	return __rbd_img_fill_from_bvecs(img_req, img_extents, num_img_extents,
2721 					 &it);
2722 }
2723 
2724 static void rbd_img_handle_request_work(struct work_struct *work)
2725 {
2726 	struct rbd_img_request *img_req =
2727 	    container_of(work, struct rbd_img_request, work);
2728 
2729 	rbd_img_handle_request(img_req, img_req->work_result);
2730 }
2731 
2732 static void rbd_img_schedule(struct rbd_img_request *img_req, int result)
2733 {
2734 	INIT_WORK(&img_req->work, rbd_img_handle_request_work);
2735 	img_req->work_result = result;
2736 	queue_work(rbd_wq, &img_req->work);
2737 }
2738 
2739 static bool rbd_obj_may_exist(struct rbd_obj_request *obj_req)
2740 {
2741 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
2742 
2743 	if (rbd_object_map_may_exist(rbd_dev, obj_req->ex.oe_objno)) {
2744 		obj_req->flags |= RBD_OBJ_FLAG_MAY_EXIST;
2745 		return true;
2746 	}
2747 
2748 	dout("%s %p objno %llu assuming dne\n", __func__, obj_req,
2749 	     obj_req->ex.oe_objno);
2750 	return false;
2751 }
2752 
2753 static int rbd_obj_read_object(struct rbd_obj_request *obj_req)
2754 {
2755 	struct ceph_osd_request *osd_req;
2756 	int ret;
2757 
2758 	osd_req = __rbd_obj_add_osd_request(obj_req, NULL, 1);
2759 	if (IS_ERR(osd_req))
2760 		return PTR_ERR(osd_req);
2761 
2762 	osd_req_op_extent_init(osd_req, 0, CEPH_OSD_OP_READ,
2763 			       obj_req->ex.oe_off, obj_req->ex.oe_len, 0, 0);
2764 	rbd_osd_setup_data(osd_req, 0);
2765 	rbd_osd_format_read(osd_req);
2766 
2767 	ret = ceph_osdc_alloc_messages(osd_req, GFP_NOIO);
2768 	if (ret)
2769 		return ret;
2770 
2771 	rbd_osd_submit(osd_req);
2772 	return 0;
2773 }
2774 
2775 static int rbd_obj_read_from_parent(struct rbd_obj_request *obj_req)
2776 {
2777 	struct rbd_img_request *img_req = obj_req->img_request;
2778 	struct rbd_device *parent = img_req->rbd_dev->parent;
2779 	struct rbd_img_request *child_img_req;
2780 	int ret;
2781 
2782 	child_img_req = kmem_cache_alloc(rbd_img_request_cache, GFP_NOIO);
2783 	if (!child_img_req)
2784 		return -ENOMEM;
2785 
2786 	rbd_img_request_init(child_img_req, parent, OBJ_OP_READ);
2787 	__set_bit(IMG_REQ_CHILD, &child_img_req->flags);
2788 	child_img_req->obj_request = obj_req;
2789 
2790 	down_read(&parent->header_rwsem);
2791 	rbd_img_capture_header(child_img_req);
2792 	up_read(&parent->header_rwsem);
2793 
2794 	dout("%s child_img_req %p for obj_req %p\n", __func__, child_img_req,
2795 	     obj_req);
2796 
2797 	if (!rbd_img_is_write(img_req)) {
2798 		switch (img_req->data_type) {
2799 		case OBJ_REQUEST_BIO:
2800 			ret = __rbd_img_fill_from_bio(child_img_req,
2801 						      obj_req->img_extents,
2802 						      obj_req->num_img_extents,
2803 						      &obj_req->bio_pos);
2804 			break;
2805 		case OBJ_REQUEST_BVECS:
2806 		case OBJ_REQUEST_OWN_BVECS:
2807 			ret = __rbd_img_fill_from_bvecs(child_img_req,
2808 						      obj_req->img_extents,
2809 						      obj_req->num_img_extents,
2810 						      &obj_req->bvec_pos);
2811 			break;
2812 		default:
2813 			BUG();
2814 		}
2815 	} else {
2816 		ret = rbd_img_fill_from_bvecs(child_img_req,
2817 					      obj_req->img_extents,
2818 					      obj_req->num_img_extents,
2819 					      obj_req->copyup_bvecs);
2820 	}
2821 	if (ret) {
2822 		rbd_img_request_destroy(child_img_req);
2823 		return ret;
2824 	}
2825 
2826 	/* avoid parent chain recursion */
2827 	rbd_img_schedule(child_img_req, 0);
2828 	return 0;
2829 }
2830 
2831 static bool rbd_obj_advance_read(struct rbd_obj_request *obj_req, int *result)
2832 {
2833 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
2834 	int ret;
2835 
2836 again:
2837 	switch (obj_req->read_state) {
2838 	case RBD_OBJ_READ_START:
2839 		rbd_assert(!*result);
2840 
2841 		if (!rbd_obj_may_exist(obj_req)) {
2842 			*result = -ENOENT;
2843 			obj_req->read_state = RBD_OBJ_READ_OBJECT;
2844 			goto again;
2845 		}
2846 
2847 		ret = rbd_obj_read_object(obj_req);
2848 		if (ret) {
2849 			*result = ret;
2850 			return true;
2851 		}
2852 		obj_req->read_state = RBD_OBJ_READ_OBJECT;
2853 		return false;
2854 	case RBD_OBJ_READ_OBJECT:
2855 		if (*result == -ENOENT && rbd_dev->parent_overlap) {
2856 			/* reverse map this object extent onto the parent */
2857 			ret = rbd_obj_calc_img_extents(obj_req, false);
2858 			if (ret) {
2859 				*result = ret;
2860 				return true;
2861 			}
2862 			if (obj_req->num_img_extents) {
2863 				ret = rbd_obj_read_from_parent(obj_req);
2864 				if (ret) {
2865 					*result = ret;
2866 					return true;
2867 				}
2868 				obj_req->read_state = RBD_OBJ_READ_PARENT;
2869 				return false;
2870 			}
2871 		}
2872 
2873 		/*
2874 		 * -ENOENT means a hole in the image -- zero-fill the entire
2875 		 * length of the request.  A short read also implies zero-fill
2876 		 * to the end of the request.
2877 		 */
2878 		if (*result == -ENOENT) {
2879 			rbd_obj_zero_range(obj_req, 0, obj_req->ex.oe_len);
2880 			*result = 0;
2881 		} else if (*result >= 0) {
2882 			if (*result < obj_req->ex.oe_len)
2883 				rbd_obj_zero_range(obj_req, *result,
2884 						obj_req->ex.oe_len - *result);
2885 			else
2886 				rbd_assert(*result == obj_req->ex.oe_len);
2887 			*result = 0;
2888 		}
2889 		return true;
2890 	case RBD_OBJ_READ_PARENT:
2891 		/*
2892 		 * The parent image is read only up to the overlap -- zero-fill
2893 		 * from the overlap to the end of the request.
2894 		 */
2895 		if (!*result) {
2896 			u32 obj_overlap = rbd_obj_img_extents_bytes(obj_req);
2897 
2898 			if (obj_overlap < obj_req->ex.oe_len)
2899 				rbd_obj_zero_range(obj_req, obj_overlap,
2900 					    obj_req->ex.oe_len - obj_overlap);
2901 		}
2902 		return true;
2903 	default:
2904 		BUG();
2905 	}
2906 }
2907 
2908 static bool rbd_obj_write_is_noop(struct rbd_obj_request *obj_req)
2909 {
2910 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
2911 
2912 	if (rbd_object_map_may_exist(rbd_dev, obj_req->ex.oe_objno))
2913 		obj_req->flags |= RBD_OBJ_FLAG_MAY_EXIST;
2914 
2915 	if (!(obj_req->flags & RBD_OBJ_FLAG_MAY_EXIST) &&
2916 	    (obj_req->flags & RBD_OBJ_FLAG_NOOP_FOR_NONEXISTENT)) {
2917 		dout("%s %p noop for nonexistent\n", __func__, obj_req);
2918 		return true;
2919 	}
2920 
2921 	return false;
2922 }
2923 
2924 /*
2925  * Return:
2926  *   0 - object map update sent
2927  *   1 - object map update isn't needed
2928  *  <0 - error
2929  */
2930 static int rbd_obj_write_pre_object_map(struct rbd_obj_request *obj_req)
2931 {
2932 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
2933 	u8 new_state;
2934 
2935 	if (!(rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP))
2936 		return 1;
2937 
2938 	if (obj_req->flags & RBD_OBJ_FLAG_DELETION)
2939 		new_state = OBJECT_PENDING;
2940 	else
2941 		new_state = OBJECT_EXISTS;
2942 
2943 	return rbd_object_map_update(obj_req, CEPH_NOSNAP, new_state, NULL);
2944 }
2945 
2946 static int rbd_obj_write_object(struct rbd_obj_request *obj_req)
2947 {
2948 	struct ceph_osd_request *osd_req;
2949 	int num_ops = count_write_ops(obj_req);
2950 	int which = 0;
2951 	int ret;
2952 
2953 	if (obj_req->flags & RBD_OBJ_FLAG_COPYUP_ENABLED)
2954 		num_ops++; /* stat */
2955 
2956 	osd_req = rbd_obj_add_osd_request(obj_req, num_ops);
2957 	if (IS_ERR(osd_req))
2958 		return PTR_ERR(osd_req);
2959 
2960 	if (obj_req->flags & RBD_OBJ_FLAG_COPYUP_ENABLED) {
2961 		ret = rbd_osd_setup_stat(osd_req, which++);
2962 		if (ret)
2963 			return ret;
2964 	}
2965 
2966 	rbd_osd_setup_write_ops(osd_req, which);
2967 	rbd_osd_format_write(osd_req);
2968 
2969 	ret = ceph_osdc_alloc_messages(osd_req, GFP_NOIO);
2970 	if (ret)
2971 		return ret;
2972 
2973 	rbd_osd_submit(osd_req);
2974 	return 0;
2975 }
2976 
2977 /*
2978  * copyup_bvecs pages are never highmem pages
2979  */
2980 static bool is_zero_bvecs(struct bio_vec *bvecs, u32 bytes)
2981 {
2982 	struct ceph_bvec_iter it = {
2983 		.bvecs = bvecs,
2984 		.iter = { .bi_size = bytes },
2985 	};
2986 
2987 	ceph_bvec_iter_advance_step(&it, bytes, ({
2988 		if (memchr_inv(bvec_virt(&bv), 0, bv.bv_len))
2989 			return false;
2990 	}));
2991 	return true;
2992 }
2993 
2994 #define MODS_ONLY	U32_MAX
2995 
2996 static int rbd_obj_copyup_empty_snapc(struct rbd_obj_request *obj_req,
2997 				      u32 bytes)
2998 {
2999 	struct ceph_osd_request *osd_req;
3000 	int ret;
3001 
3002 	dout("%s obj_req %p bytes %u\n", __func__, obj_req, bytes);
3003 	rbd_assert(bytes > 0 && bytes != MODS_ONLY);
3004 
3005 	osd_req = __rbd_obj_add_osd_request(obj_req, &rbd_empty_snapc, 1);
3006 	if (IS_ERR(osd_req))
3007 		return PTR_ERR(osd_req);
3008 
3009 	ret = rbd_osd_setup_copyup(osd_req, 0, bytes);
3010 	if (ret)
3011 		return ret;
3012 
3013 	rbd_osd_format_write(osd_req);
3014 
3015 	ret = ceph_osdc_alloc_messages(osd_req, GFP_NOIO);
3016 	if (ret)
3017 		return ret;
3018 
3019 	rbd_osd_submit(osd_req);
3020 	return 0;
3021 }
3022 
3023 static int rbd_obj_copyup_current_snapc(struct rbd_obj_request *obj_req,
3024 					u32 bytes)
3025 {
3026 	struct ceph_osd_request *osd_req;
3027 	int num_ops = count_write_ops(obj_req);
3028 	int which = 0;
3029 	int ret;
3030 
3031 	dout("%s obj_req %p bytes %u\n", __func__, obj_req, bytes);
3032 
3033 	if (bytes != MODS_ONLY)
3034 		num_ops++; /* copyup */
3035 
3036 	osd_req = rbd_obj_add_osd_request(obj_req, num_ops);
3037 	if (IS_ERR(osd_req))
3038 		return PTR_ERR(osd_req);
3039 
3040 	if (bytes != MODS_ONLY) {
3041 		ret = rbd_osd_setup_copyup(osd_req, which++, bytes);
3042 		if (ret)
3043 			return ret;
3044 	}
3045 
3046 	rbd_osd_setup_write_ops(osd_req, which);
3047 	rbd_osd_format_write(osd_req);
3048 
3049 	ret = ceph_osdc_alloc_messages(osd_req, GFP_NOIO);
3050 	if (ret)
3051 		return ret;
3052 
3053 	rbd_osd_submit(osd_req);
3054 	return 0;
3055 }
3056 
3057 static int setup_copyup_bvecs(struct rbd_obj_request *obj_req, u64 obj_overlap)
3058 {
3059 	u32 i;
3060 
3061 	rbd_assert(!obj_req->copyup_bvecs);
3062 	obj_req->copyup_bvec_count = calc_pages_for(0, obj_overlap);
3063 	obj_req->copyup_bvecs = kcalloc(obj_req->copyup_bvec_count,
3064 					sizeof(*obj_req->copyup_bvecs),
3065 					GFP_NOIO);
3066 	if (!obj_req->copyup_bvecs)
3067 		return -ENOMEM;
3068 
3069 	for (i = 0; i < obj_req->copyup_bvec_count; i++) {
3070 		unsigned int len = min(obj_overlap, (u64)PAGE_SIZE);
3071 
3072 		obj_req->copyup_bvecs[i].bv_page = alloc_page(GFP_NOIO);
3073 		if (!obj_req->copyup_bvecs[i].bv_page)
3074 			return -ENOMEM;
3075 
3076 		obj_req->copyup_bvecs[i].bv_offset = 0;
3077 		obj_req->copyup_bvecs[i].bv_len = len;
3078 		obj_overlap -= len;
3079 	}
3080 
3081 	rbd_assert(!obj_overlap);
3082 	return 0;
3083 }
3084 
3085 /*
3086  * The target object doesn't exist.  Read the data for the entire
3087  * target object up to the overlap point (if any) from the parent,
3088  * so we can use it for a copyup.
3089  */
3090 static int rbd_obj_copyup_read_parent(struct rbd_obj_request *obj_req)
3091 {
3092 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
3093 	int ret;
3094 
3095 	rbd_assert(obj_req->num_img_extents);
3096 	prune_extents(obj_req->img_extents, &obj_req->num_img_extents,
3097 		      rbd_dev->parent_overlap);
3098 	if (!obj_req->num_img_extents) {
3099 		/*
3100 		 * The overlap has become 0 (most likely because the
3101 		 * image has been flattened).  Re-submit the original write
3102 		 * request -- pass MODS_ONLY since the copyup isn't needed
3103 		 * anymore.
3104 		 */
3105 		return rbd_obj_copyup_current_snapc(obj_req, MODS_ONLY);
3106 	}
3107 
3108 	ret = setup_copyup_bvecs(obj_req, rbd_obj_img_extents_bytes(obj_req));
3109 	if (ret)
3110 		return ret;
3111 
3112 	return rbd_obj_read_from_parent(obj_req);
3113 }
3114 
3115 static void rbd_obj_copyup_object_maps(struct rbd_obj_request *obj_req)
3116 {
3117 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
3118 	struct ceph_snap_context *snapc = obj_req->img_request->snapc;
3119 	u8 new_state;
3120 	u32 i;
3121 	int ret;
3122 
3123 	rbd_assert(!obj_req->pending.result && !obj_req->pending.num_pending);
3124 
3125 	if (!(rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP))
3126 		return;
3127 
3128 	if (obj_req->flags & RBD_OBJ_FLAG_COPYUP_ZEROS)
3129 		return;
3130 
3131 	for (i = 0; i < snapc->num_snaps; i++) {
3132 		if ((rbd_dev->header.features & RBD_FEATURE_FAST_DIFF) &&
3133 		    i + 1 < snapc->num_snaps)
3134 			new_state = OBJECT_EXISTS_CLEAN;
3135 		else
3136 			new_state = OBJECT_EXISTS;
3137 
3138 		ret = rbd_object_map_update(obj_req, snapc->snaps[i],
3139 					    new_state, NULL);
3140 		if (ret < 0) {
3141 			obj_req->pending.result = ret;
3142 			return;
3143 		}
3144 
3145 		rbd_assert(!ret);
3146 		obj_req->pending.num_pending++;
3147 	}
3148 }
3149 
3150 static void rbd_obj_copyup_write_object(struct rbd_obj_request *obj_req)
3151 {
3152 	u32 bytes = rbd_obj_img_extents_bytes(obj_req);
3153 	int ret;
3154 
3155 	rbd_assert(!obj_req->pending.result && !obj_req->pending.num_pending);
3156 
3157 	/*
3158 	 * Only send non-zero copyup data to save some I/O and network
3159 	 * bandwidth -- zero copyup data is equivalent to the object not
3160 	 * existing.
3161 	 */
3162 	if (obj_req->flags & RBD_OBJ_FLAG_COPYUP_ZEROS)
3163 		bytes = 0;
3164 
3165 	if (obj_req->img_request->snapc->num_snaps && bytes > 0) {
3166 		/*
3167 		 * Send a copyup request with an empty snapshot context to
3168 		 * deep-copyup the object through all existing snapshots.
3169 		 * A second request with the current snapshot context will be
3170 		 * sent for the actual modification.
3171 		 */
3172 		ret = rbd_obj_copyup_empty_snapc(obj_req, bytes);
3173 		if (ret) {
3174 			obj_req->pending.result = ret;
3175 			return;
3176 		}
3177 
3178 		obj_req->pending.num_pending++;
3179 		bytes = MODS_ONLY;
3180 	}
3181 
3182 	ret = rbd_obj_copyup_current_snapc(obj_req, bytes);
3183 	if (ret) {
3184 		obj_req->pending.result = ret;
3185 		return;
3186 	}
3187 
3188 	obj_req->pending.num_pending++;
3189 }
3190 
3191 static bool rbd_obj_advance_copyup(struct rbd_obj_request *obj_req, int *result)
3192 {
3193 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
3194 	int ret;
3195 
3196 again:
3197 	switch (obj_req->copyup_state) {
3198 	case RBD_OBJ_COPYUP_START:
3199 		rbd_assert(!*result);
3200 
3201 		ret = rbd_obj_copyup_read_parent(obj_req);
3202 		if (ret) {
3203 			*result = ret;
3204 			return true;
3205 		}
3206 		if (obj_req->num_img_extents)
3207 			obj_req->copyup_state = RBD_OBJ_COPYUP_READ_PARENT;
3208 		else
3209 			obj_req->copyup_state = RBD_OBJ_COPYUP_WRITE_OBJECT;
3210 		return false;
3211 	case RBD_OBJ_COPYUP_READ_PARENT:
3212 		if (*result)
3213 			return true;
3214 
3215 		if (is_zero_bvecs(obj_req->copyup_bvecs,
3216 				  rbd_obj_img_extents_bytes(obj_req))) {
3217 			dout("%s %p detected zeros\n", __func__, obj_req);
3218 			obj_req->flags |= RBD_OBJ_FLAG_COPYUP_ZEROS;
3219 		}
3220 
3221 		rbd_obj_copyup_object_maps(obj_req);
3222 		if (!obj_req->pending.num_pending) {
3223 			*result = obj_req->pending.result;
3224 			obj_req->copyup_state = RBD_OBJ_COPYUP_OBJECT_MAPS;
3225 			goto again;
3226 		}
3227 		obj_req->copyup_state = __RBD_OBJ_COPYUP_OBJECT_MAPS;
3228 		return false;
3229 	case __RBD_OBJ_COPYUP_OBJECT_MAPS:
3230 		if (!pending_result_dec(&obj_req->pending, result))
3231 			return false;
3232 		fallthrough;
3233 	case RBD_OBJ_COPYUP_OBJECT_MAPS:
3234 		if (*result) {
3235 			rbd_warn(rbd_dev, "snap object map update failed: %d",
3236 				 *result);
3237 			return true;
3238 		}
3239 
3240 		rbd_obj_copyup_write_object(obj_req);
3241 		if (!obj_req->pending.num_pending) {
3242 			*result = obj_req->pending.result;
3243 			obj_req->copyup_state = RBD_OBJ_COPYUP_WRITE_OBJECT;
3244 			goto again;
3245 		}
3246 		obj_req->copyup_state = __RBD_OBJ_COPYUP_WRITE_OBJECT;
3247 		return false;
3248 	case __RBD_OBJ_COPYUP_WRITE_OBJECT:
3249 		if (!pending_result_dec(&obj_req->pending, result))
3250 			return false;
3251 		fallthrough;
3252 	case RBD_OBJ_COPYUP_WRITE_OBJECT:
3253 		return true;
3254 	default:
3255 		BUG();
3256 	}
3257 }
3258 
3259 /*
3260  * Return:
3261  *   0 - object map update sent
3262  *   1 - object map update isn't needed
3263  *  <0 - error
3264  */
3265 static int rbd_obj_write_post_object_map(struct rbd_obj_request *obj_req)
3266 {
3267 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
3268 	u8 current_state = OBJECT_PENDING;
3269 
3270 	if (!(rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP))
3271 		return 1;
3272 
3273 	if (!(obj_req->flags & RBD_OBJ_FLAG_DELETION))
3274 		return 1;
3275 
3276 	return rbd_object_map_update(obj_req, CEPH_NOSNAP, OBJECT_NONEXISTENT,
3277 				     &current_state);
3278 }
3279 
3280 static bool rbd_obj_advance_write(struct rbd_obj_request *obj_req, int *result)
3281 {
3282 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
3283 	int ret;
3284 
3285 again:
3286 	switch (obj_req->write_state) {
3287 	case RBD_OBJ_WRITE_START:
3288 		rbd_assert(!*result);
3289 
3290 		if (rbd_obj_write_is_noop(obj_req))
3291 			return true;
3292 
3293 		ret = rbd_obj_write_pre_object_map(obj_req);
3294 		if (ret < 0) {
3295 			*result = ret;
3296 			return true;
3297 		}
3298 		obj_req->write_state = RBD_OBJ_WRITE_PRE_OBJECT_MAP;
3299 		if (ret > 0)
3300 			goto again;
3301 		return false;
3302 	case RBD_OBJ_WRITE_PRE_OBJECT_MAP:
3303 		if (*result) {
3304 			rbd_warn(rbd_dev, "pre object map update failed: %d",
3305 				 *result);
3306 			return true;
3307 		}
3308 		ret = rbd_obj_write_object(obj_req);
3309 		if (ret) {
3310 			*result = ret;
3311 			return true;
3312 		}
3313 		obj_req->write_state = RBD_OBJ_WRITE_OBJECT;
3314 		return false;
3315 	case RBD_OBJ_WRITE_OBJECT:
3316 		if (*result == -ENOENT) {
3317 			if (obj_req->flags & RBD_OBJ_FLAG_COPYUP_ENABLED) {
3318 				*result = 0;
3319 				obj_req->copyup_state = RBD_OBJ_COPYUP_START;
3320 				obj_req->write_state = __RBD_OBJ_WRITE_COPYUP;
3321 				goto again;
3322 			}
3323 			/*
3324 			 * On a non-existent object:
3325 			 *   delete - -ENOENT, truncate/zero - 0
3326 			 */
3327 			if (obj_req->flags & RBD_OBJ_FLAG_DELETION)
3328 				*result = 0;
3329 		}
3330 		if (*result)
3331 			return true;
3332 
3333 		obj_req->write_state = RBD_OBJ_WRITE_COPYUP;
3334 		goto again;
3335 	case __RBD_OBJ_WRITE_COPYUP:
3336 		if (!rbd_obj_advance_copyup(obj_req, result))
3337 			return false;
3338 		fallthrough;
3339 	case RBD_OBJ_WRITE_COPYUP:
3340 		if (*result) {
3341 			rbd_warn(rbd_dev, "copyup failed: %d", *result);
3342 			return true;
3343 		}
3344 		ret = rbd_obj_write_post_object_map(obj_req);
3345 		if (ret < 0) {
3346 			*result = ret;
3347 			return true;
3348 		}
3349 		obj_req->write_state = RBD_OBJ_WRITE_POST_OBJECT_MAP;
3350 		if (ret > 0)
3351 			goto again;
3352 		return false;
3353 	case RBD_OBJ_WRITE_POST_OBJECT_MAP:
3354 		if (*result)
3355 			rbd_warn(rbd_dev, "post object map update failed: %d",
3356 				 *result);
3357 		return true;
3358 	default:
3359 		BUG();
3360 	}
3361 }
3362 
3363 /*
3364  * Return true if @obj_req is completed.
3365  */
3366 static bool __rbd_obj_handle_request(struct rbd_obj_request *obj_req,
3367 				     int *result)
3368 {
3369 	struct rbd_img_request *img_req = obj_req->img_request;
3370 	struct rbd_device *rbd_dev = img_req->rbd_dev;
3371 	bool done;
3372 
3373 	mutex_lock(&obj_req->state_mutex);
3374 	if (!rbd_img_is_write(img_req))
3375 		done = rbd_obj_advance_read(obj_req, result);
3376 	else
3377 		done = rbd_obj_advance_write(obj_req, result);
3378 	mutex_unlock(&obj_req->state_mutex);
3379 
3380 	if (done && *result) {
3381 		rbd_assert(*result < 0);
3382 		rbd_warn(rbd_dev, "%s at objno %llu %llu~%llu result %d",
3383 			 obj_op_name(img_req->op_type), obj_req->ex.oe_objno,
3384 			 obj_req->ex.oe_off, obj_req->ex.oe_len, *result);
3385 	}
3386 	return done;
3387 }
3388 
3389 /*
3390  * This is open-coded in rbd_img_handle_request() to avoid parent chain
3391  * recursion.
3392  */
3393 static void rbd_obj_handle_request(struct rbd_obj_request *obj_req, int result)
3394 {
3395 	if (__rbd_obj_handle_request(obj_req, &result))
3396 		rbd_img_handle_request(obj_req->img_request, result);
3397 }
3398 
3399 static bool need_exclusive_lock(struct rbd_img_request *img_req)
3400 {
3401 	struct rbd_device *rbd_dev = img_req->rbd_dev;
3402 
3403 	if (!(rbd_dev->header.features & RBD_FEATURE_EXCLUSIVE_LOCK))
3404 		return false;
3405 
3406 	if (rbd_is_ro(rbd_dev))
3407 		return false;
3408 
3409 	rbd_assert(!test_bit(IMG_REQ_CHILD, &img_req->flags));
3410 	if (rbd_dev->opts->lock_on_read ||
3411 	    (rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP))
3412 		return true;
3413 
3414 	return rbd_img_is_write(img_req);
3415 }
3416 
3417 static bool rbd_lock_add_request(struct rbd_img_request *img_req)
3418 {
3419 	struct rbd_device *rbd_dev = img_req->rbd_dev;
3420 	bool locked;
3421 
3422 	lockdep_assert_held(&rbd_dev->lock_rwsem);
3423 	locked = rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED;
3424 	spin_lock(&rbd_dev->lock_lists_lock);
3425 	rbd_assert(list_empty(&img_req->lock_item));
3426 	if (!locked)
3427 		list_add_tail(&img_req->lock_item, &rbd_dev->acquiring_list);
3428 	else
3429 		list_add_tail(&img_req->lock_item, &rbd_dev->running_list);
3430 	spin_unlock(&rbd_dev->lock_lists_lock);
3431 	return locked;
3432 }
3433 
3434 static void rbd_lock_del_request(struct rbd_img_request *img_req)
3435 {
3436 	struct rbd_device *rbd_dev = img_req->rbd_dev;
3437 	bool need_wakeup;
3438 
3439 	lockdep_assert_held(&rbd_dev->lock_rwsem);
3440 	spin_lock(&rbd_dev->lock_lists_lock);
3441 	rbd_assert(!list_empty(&img_req->lock_item));
3442 	list_del_init(&img_req->lock_item);
3443 	need_wakeup = (rbd_dev->lock_state == RBD_LOCK_STATE_RELEASING &&
3444 		       list_empty(&rbd_dev->running_list));
3445 	spin_unlock(&rbd_dev->lock_lists_lock);
3446 	if (need_wakeup)
3447 		complete(&rbd_dev->releasing_wait);
3448 }
3449 
3450 static int rbd_img_exclusive_lock(struct rbd_img_request *img_req)
3451 {
3452 	struct rbd_device *rbd_dev = img_req->rbd_dev;
3453 
3454 	if (!need_exclusive_lock(img_req))
3455 		return 1;
3456 
3457 	if (rbd_lock_add_request(img_req))
3458 		return 1;
3459 
3460 	if (rbd_dev->opts->exclusive) {
3461 		WARN_ON(1); /* lock got released? */
3462 		return -EROFS;
3463 	}
3464 
3465 	/*
3466 	 * Note the use of mod_delayed_work() in rbd_acquire_lock()
3467 	 * and cancel_delayed_work() in wake_lock_waiters().
3468 	 */
3469 	dout("%s rbd_dev %p queueing lock_dwork\n", __func__, rbd_dev);
3470 	queue_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork, 0);
3471 	return 0;
3472 }
3473 
3474 static void rbd_img_object_requests(struct rbd_img_request *img_req)
3475 {
3476 	struct rbd_obj_request *obj_req;
3477 
3478 	rbd_assert(!img_req->pending.result && !img_req->pending.num_pending);
3479 
3480 	for_each_obj_request(img_req, obj_req) {
3481 		int result = 0;
3482 
3483 		if (__rbd_obj_handle_request(obj_req, &result)) {
3484 			if (result) {
3485 				img_req->pending.result = result;
3486 				return;
3487 			}
3488 		} else {
3489 			img_req->pending.num_pending++;
3490 		}
3491 	}
3492 }
3493 
3494 static bool rbd_img_advance(struct rbd_img_request *img_req, int *result)
3495 {
3496 	struct rbd_device *rbd_dev = img_req->rbd_dev;
3497 	int ret;
3498 
3499 again:
3500 	switch (img_req->state) {
3501 	case RBD_IMG_START:
3502 		rbd_assert(!*result);
3503 
3504 		ret = rbd_img_exclusive_lock(img_req);
3505 		if (ret < 0) {
3506 			*result = ret;
3507 			return true;
3508 		}
3509 		img_req->state = RBD_IMG_EXCLUSIVE_LOCK;
3510 		if (ret > 0)
3511 			goto again;
3512 		return false;
3513 	case RBD_IMG_EXCLUSIVE_LOCK:
3514 		if (*result)
3515 			return true;
3516 
3517 		rbd_assert(!need_exclusive_lock(img_req) ||
3518 			   __rbd_is_lock_owner(rbd_dev));
3519 
3520 		rbd_img_object_requests(img_req);
3521 		if (!img_req->pending.num_pending) {
3522 			*result = img_req->pending.result;
3523 			img_req->state = RBD_IMG_OBJECT_REQUESTS;
3524 			goto again;
3525 		}
3526 		img_req->state = __RBD_IMG_OBJECT_REQUESTS;
3527 		return false;
3528 	case __RBD_IMG_OBJECT_REQUESTS:
3529 		if (!pending_result_dec(&img_req->pending, result))
3530 			return false;
3531 		fallthrough;
3532 	case RBD_IMG_OBJECT_REQUESTS:
3533 		return true;
3534 	default:
3535 		BUG();
3536 	}
3537 }
3538 
3539 /*
3540  * Return true if @img_req is completed.
3541  */
3542 static bool __rbd_img_handle_request(struct rbd_img_request *img_req,
3543 				     int *result)
3544 {
3545 	struct rbd_device *rbd_dev = img_req->rbd_dev;
3546 	bool done;
3547 
3548 	if (need_exclusive_lock(img_req)) {
3549 		down_read(&rbd_dev->lock_rwsem);
3550 		mutex_lock(&img_req->state_mutex);
3551 		done = rbd_img_advance(img_req, result);
3552 		if (done)
3553 			rbd_lock_del_request(img_req);
3554 		mutex_unlock(&img_req->state_mutex);
3555 		up_read(&rbd_dev->lock_rwsem);
3556 	} else {
3557 		mutex_lock(&img_req->state_mutex);
3558 		done = rbd_img_advance(img_req, result);
3559 		mutex_unlock(&img_req->state_mutex);
3560 	}
3561 
3562 	if (done && *result) {
3563 		rbd_assert(*result < 0);
3564 		rbd_warn(rbd_dev, "%s%s result %d",
3565 		      test_bit(IMG_REQ_CHILD, &img_req->flags) ? "child " : "",
3566 		      obj_op_name(img_req->op_type), *result);
3567 	}
3568 	return done;
3569 }
3570 
3571 static void rbd_img_handle_request(struct rbd_img_request *img_req, int result)
3572 {
3573 again:
3574 	if (!__rbd_img_handle_request(img_req, &result))
3575 		return;
3576 
3577 	if (test_bit(IMG_REQ_CHILD, &img_req->flags)) {
3578 		struct rbd_obj_request *obj_req = img_req->obj_request;
3579 
3580 		rbd_img_request_destroy(img_req);
3581 		if (__rbd_obj_handle_request(obj_req, &result)) {
3582 			img_req = obj_req->img_request;
3583 			goto again;
3584 		}
3585 	} else {
3586 		struct request *rq = blk_mq_rq_from_pdu(img_req);
3587 
3588 		rbd_img_request_destroy(img_req);
3589 		blk_mq_end_request(rq, errno_to_blk_status(result));
3590 	}
3591 }
3592 
3593 static const struct rbd_client_id rbd_empty_cid;
3594 
3595 static bool rbd_cid_equal(const struct rbd_client_id *lhs,
3596 			  const struct rbd_client_id *rhs)
3597 {
3598 	return lhs->gid == rhs->gid && lhs->handle == rhs->handle;
3599 }
3600 
3601 static struct rbd_client_id rbd_get_cid(struct rbd_device *rbd_dev)
3602 {
3603 	struct rbd_client_id cid;
3604 
3605 	mutex_lock(&rbd_dev->watch_mutex);
3606 	cid.gid = ceph_client_gid(rbd_dev->rbd_client->client);
3607 	cid.handle = rbd_dev->watch_cookie;
3608 	mutex_unlock(&rbd_dev->watch_mutex);
3609 	return cid;
3610 }
3611 
3612 /*
3613  * lock_rwsem must be held for write
3614  */
3615 static void rbd_set_owner_cid(struct rbd_device *rbd_dev,
3616 			      const struct rbd_client_id *cid)
3617 {
3618 	dout("%s rbd_dev %p %llu-%llu -> %llu-%llu\n", __func__, rbd_dev,
3619 	     rbd_dev->owner_cid.gid, rbd_dev->owner_cid.handle,
3620 	     cid->gid, cid->handle);
3621 	rbd_dev->owner_cid = *cid; /* struct */
3622 }
3623 
3624 static void format_lock_cookie(struct rbd_device *rbd_dev, char *buf)
3625 {
3626 	mutex_lock(&rbd_dev->watch_mutex);
3627 	sprintf(buf, "%s %llu", RBD_LOCK_COOKIE_PREFIX, rbd_dev->watch_cookie);
3628 	mutex_unlock(&rbd_dev->watch_mutex);
3629 }
3630 
3631 static void __rbd_lock(struct rbd_device *rbd_dev, const char *cookie)
3632 {
3633 	struct rbd_client_id cid = rbd_get_cid(rbd_dev);
3634 
3635 	rbd_dev->lock_state = RBD_LOCK_STATE_LOCKED;
3636 	strcpy(rbd_dev->lock_cookie, cookie);
3637 	rbd_set_owner_cid(rbd_dev, &cid);
3638 	queue_work(rbd_dev->task_wq, &rbd_dev->acquired_lock_work);
3639 }
3640 
3641 /*
3642  * lock_rwsem must be held for write
3643  */
3644 static int rbd_lock(struct rbd_device *rbd_dev)
3645 {
3646 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3647 	char cookie[32];
3648 	int ret;
3649 
3650 	WARN_ON(__rbd_is_lock_owner(rbd_dev) ||
3651 		rbd_dev->lock_cookie[0] != '\0');
3652 
3653 	format_lock_cookie(rbd_dev, cookie);
3654 	ret = ceph_cls_lock(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc,
3655 			    RBD_LOCK_NAME, CEPH_CLS_LOCK_EXCLUSIVE, cookie,
3656 			    RBD_LOCK_TAG, "", 0);
3657 	if (ret)
3658 		return ret;
3659 
3660 	__rbd_lock(rbd_dev, cookie);
3661 	return 0;
3662 }
3663 
3664 /*
3665  * lock_rwsem must be held for write
3666  */
3667 static void rbd_unlock(struct rbd_device *rbd_dev)
3668 {
3669 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3670 	int ret;
3671 
3672 	WARN_ON(!__rbd_is_lock_owner(rbd_dev) ||
3673 		rbd_dev->lock_cookie[0] == '\0');
3674 
3675 	ret = ceph_cls_unlock(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc,
3676 			      RBD_LOCK_NAME, rbd_dev->lock_cookie);
3677 	if (ret && ret != -ENOENT)
3678 		rbd_warn(rbd_dev, "failed to unlock header: %d", ret);
3679 
3680 	/* treat errors as the image is unlocked */
3681 	rbd_dev->lock_state = RBD_LOCK_STATE_UNLOCKED;
3682 	rbd_dev->lock_cookie[0] = '\0';
3683 	rbd_set_owner_cid(rbd_dev, &rbd_empty_cid);
3684 	queue_work(rbd_dev->task_wq, &rbd_dev->released_lock_work);
3685 }
3686 
3687 static int __rbd_notify_op_lock(struct rbd_device *rbd_dev,
3688 				enum rbd_notify_op notify_op,
3689 				struct page ***preply_pages,
3690 				size_t *preply_len)
3691 {
3692 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3693 	struct rbd_client_id cid = rbd_get_cid(rbd_dev);
3694 	char buf[4 + 8 + 8 + CEPH_ENCODING_START_BLK_LEN];
3695 	int buf_size = sizeof(buf);
3696 	void *p = buf;
3697 
3698 	dout("%s rbd_dev %p notify_op %d\n", __func__, rbd_dev, notify_op);
3699 
3700 	/* encode *LockPayload NotifyMessage (op + ClientId) */
3701 	ceph_start_encoding(&p, 2, 1, buf_size - CEPH_ENCODING_START_BLK_LEN);
3702 	ceph_encode_32(&p, notify_op);
3703 	ceph_encode_64(&p, cid.gid);
3704 	ceph_encode_64(&p, cid.handle);
3705 
3706 	return ceph_osdc_notify(osdc, &rbd_dev->header_oid,
3707 				&rbd_dev->header_oloc, buf, buf_size,
3708 				RBD_NOTIFY_TIMEOUT, preply_pages, preply_len);
3709 }
3710 
3711 static void rbd_notify_op_lock(struct rbd_device *rbd_dev,
3712 			       enum rbd_notify_op notify_op)
3713 {
3714 	__rbd_notify_op_lock(rbd_dev, notify_op, NULL, NULL);
3715 }
3716 
3717 static void rbd_notify_acquired_lock(struct work_struct *work)
3718 {
3719 	struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
3720 						  acquired_lock_work);
3721 
3722 	rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_ACQUIRED_LOCK);
3723 }
3724 
3725 static void rbd_notify_released_lock(struct work_struct *work)
3726 {
3727 	struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
3728 						  released_lock_work);
3729 
3730 	rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_RELEASED_LOCK);
3731 }
3732 
3733 static int rbd_request_lock(struct rbd_device *rbd_dev)
3734 {
3735 	struct page **reply_pages;
3736 	size_t reply_len;
3737 	bool lock_owner_responded = false;
3738 	int ret;
3739 
3740 	dout("%s rbd_dev %p\n", __func__, rbd_dev);
3741 
3742 	ret = __rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_REQUEST_LOCK,
3743 				   &reply_pages, &reply_len);
3744 	if (ret && ret != -ETIMEDOUT) {
3745 		rbd_warn(rbd_dev, "failed to request lock: %d", ret);
3746 		goto out;
3747 	}
3748 
3749 	if (reply_len > 0 && reply_len <= PAGE_SIZE) {
3750 		void *p = page_address(reply_pages[0]);
3751 		void *const end = p + reply_len;
3752 		u32 n;
3753 
3754 		ceph_decode_32_safe(&p, end, n, e_inval); /* num_acks */
3755 		while (n--) {
3756 			u8 struct_v;
3757 			u32 len;
3758 
3759 			ceph_decode_need(&p, end, 8 + 8, e_inval);
3760 			p += 8 + 8; /* skip gid and cookie */
3761 
3762 			ceph_decode_32_safe(&p, end, len, e_inval);
3763 			if (!len)
3764 				continue;
3765 
3766 			if (lock_owner_responded) {
3767 				rbd_warn(rbd_dev,
3768 					 "duplicate lock owners detected");
3769 				ret = -EIO;
3770 				goto out;
3771 			}
3772 
3773 			lock_owner_responded = true;
3774 			ret = ceph_start_decoding(&p, end, 1, "ResponseMessage",
3775 						  &struct_v, &len);
3776 			if (ret) {
3777 				rbd_warn(rbd_dev,
3778 					 "failed to decode ResponseMessage: %d",
3779 					 ret);
3780 				goto e_inval;
3781 			}
3782 
3783 			ret = ceph_decode_32(&p);
3784 		}
3785 	}
3786 
3787 	if (!lock_owner_responded) {
3788 		rbd_warn(rbd_dev, "no lock owners detected");
3789 		ret = -ETIMEDOUT;
3790 	}
3791 
3792 out:
3793 	ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len));
3794 	return ret;
3795 
3796 e_inval:
3797 	ret = -EINVAL;
3798 	goto out;
3799 }
3800 
3801 /*
3802  * Either image request state machine(s) or rbd_add_acquire_lock()
3803  * (i.e. "rbd map").
3804  */
3805 static void wake_lock_waiters(struct rbd_device *rbd_dev, int result)
3806 {
3807 	struct rbd_img_request *img_req;
3808 
3809 	dout("%s rbd_dev %p result %d\n", __func__, rbd_dev, result);
3810 	lockdep_assert_held_write(&rbd_dev->lock_rwsem);
3811 
3812 	cancel_delayed_work(&rbd_dev->lock_dwork);
3813 	if (!completion_done(&rbd_dev->acquire_wait)) {
3814 		rbd_assert(list_empty(&rbd_dev->acquiring_list) &&
3815 			   list_empty(&rbd_dev->running_list));
3816 		rbd_dev->acquire_err = result;
3817 		complete_all(&rbd_dev->acquire_wait);
3818 		return;
3819 	}
3820 
3821 	list_for_each_entry(img_req, &rbd_dev->acquiring_list, lock_item) {
3822 		mutex_lock(&img_req->state_mutex);
3823 		rbd_assert(img_req->state == RBD_IMG_EXCLUSIVE_LOCK);
3824 		rbd_img_schedule(img_req, result);
3825 		mutex_unlock(&img_req->state_mutex);
3826 	}
3827 
3828 	list_splice_tail_init(&rbd_dev->acquiring_list, &rbd_dev->running_list);
3829 }
3830 
3831 static int get_lock_owner_info(struct rbd_device *rbd_dev,
3832 			       struct ceph_locker **lockers, u32 *num_lockers)
3833 {
3834 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3835 	u8 lock_type;
3836 	char *lock_tag;
3837 	int ret;
3838 
3839 	dout("%s rbd_dev %p\n", __func__, rbd_dev);
3840 
3841 	ret = ceph_cls_lock_info(osdc, &rbd_dev->header_oid,
3842 				 &rbd_dev->header_oloc, RBD_LOCK_NAME,
3843 				 &lock_type, &lock_tag, lockers, num_lockers);
3844 	if (ret)
3845 		return ret;
3846 
3847 	if (*num_lockers == 0) {
3848 		dout("%s rbd_dev %p no lockers detected\n", __func__, rbd_dev);
3849 		goto out;
3850 	}
3851 
3852 	if (strcmp(lock_tag, RBD_LOCK_TAG)) {
3853 		rbd_warn(rbd_dev, "locked by external mechanism, tag %s",
3854 			 lock_tag);
3855 		ret = -EBUSY;
3856 		goto out;
3857 	}
3858 
3859 	if (lock_type == CEPH_CLS_LOCK_SHARED) {
3860 		rbd_warn(rbd_dev, "shared lock type detected");
3861 		ret = -EBUSY;
3862 		goto out;
3863 	}
3864 
3865 	if (strncmp((*lockers)[0].id.cookie, RBD_LOCK_COOKIE_PREFIX,
3866 		    strlen(RBD_LOCK_COOKIE_PREFIX))) {
3867 		rbd_warn(rbd_dev, "locked by external mechanism, cookie %s",
3868 			 (*lockers)[0].id.cookie);
3869 		ret = -EBUSY;
3870 		goto out;
3871 	}
3872 
3873 out:
3874 	kfree(lock_tag);
3875 	return ret;
3876 }
3877 
3878 static int find_watcher(struct rbd_device *rbd_dev,
3879 			const struct ceph_locker *locker)
3880 {
3881 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
3882 	struct ceph_watch_item *watchers;
3883 	u32 num_watchers;
3884 	u64 cookie;
3885 	int i;
3886 	int ret;
3887 
3888 	ret = ceph_osdc_list_watchers(osdc, &rbd_dev->header_oid,
3889 				      &rbd_dev->header_oloc, &watchers,
3890 				      &num_watchers);
3891 	if (ret)
3892 		return ret;
3893 
3894 	sscanf(locker->id.cookie, RBD_LOCK_COOKIE_PREFIX " %llu", &cookie);
3895 	for (i = 0; i < num_watchers; i++) {
3896 		/*
3897 		 * Ignore addr->type while comparing.  This mimics
3898 		 * entity_addr_t::get_legacy_str() + strcmp().
3899 		 */
3900 		if (ceph_addr_equal_no_type(&watchers[i].addr,
3901 					    &locker->info.addr) &&
3902 		    watchers[i].cookie == cookie) {
3903 			struct rbd_client_id cid = {
3904 				.gid = le64_to_cpu(watchers[i].name.num),
3905 				.handle = cookie,
3906 			};
3907 
3908 			dout("%s rbd_dev %p found cid %llu-%llu\n", __func__,
3909 			     rbd_dev, cid.gid, cid.handle);
3910 			rbd_set_owner_cid(rbd_dev, &cid);
3911 			ret = 1;
3912 			goto out;
3913 		}
3914 	}
3915 
3916 	dout("%s rbd_dev %p no watchers\n", __func__, rbd_dev);
3917 	ret = 0;
3918 out:
3919 	kfree(watchers);
3920 	return ret;
3921 }
3922 
3923 /*
3924  * lock_rwsem must be held for write
3925  */
3926 static int rbd_try_lock(struct rbd_device *rbd_dev)
3927 {
3928 	struct ceph_client *client = rbd_dev->rbd_client->client;
3929 	struct ceph_locker *lockers;
3930 	u32 num_lockers;
3931 	int ret;
3932 
3933 	for (;;) {
3934 		ret = rbd_lock(rbd_dev);
3935 		if (ret != -EBUSY)
3936 			return ret;
3937 
3938 		/* determine if the current lock holder is still alive */
3939 		ret = get_lock_owner_info(rbd_dev, &lockers, &num_lockers);
3940 		if (ret)
3941 			return ret;
3942 
3943 		if (num_lockers == 0)
3944 			goto again;
3945 
3946 		ret = find_watcher(rbd_dev, lockers);
3947 		if (ret)
3948 			goto out; /* request lock or error */
3949 
3950 		rbd_warn(rbd_dev, "breaking header lock owned by %s%llu",
3951 			 ENTITY_NAME(lockers[0].id.name));
3952 
3953 		ret = ceph_monc_blocklist_add(&client->monc,
3954 					      &lockers[0].info.addr);
3955 		if (ret) {
3956 			rbd_warn(rbd_dev, "blocklist of %s%llu failed: %d",
3957 				 ENTITY_NAME(lockers[0].id.name), ret);
3958 			goto out;
3959 		}
3960 
3961 		ret = ceph_cls_break_lock(&client->osdc, &rbd_dev->header_oid,
3962 					  &rbd_dev->header_oloc, RBD_LOCK_NAME,
3963 					  lockers[0].id.cookie,
3964 					  &lockers[0].id.name);
3965 		if (ret && ret != -ENOENT)
3966 			goto out;
3967 
3968 again:
3969 		ceph_free_lockers(lockers, num_lockers);
3970 	}
3971 
3972 out:
3973 	ceph_free_lockers(lockers, num_lockers);
3974 	return ret;
3975 }
3976 
3977 static int rbd_post_acquire_action(struct rbd_device *rbd_dev)
3978 {
3979 	int ret;
3980 
3981 	if (rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP) {
3982 		ret = rbd_object_map_open(rbd_dev);
3983 		if (ret)
3984 			return ret;
3985 	}
3986 
3987 	return 0;
3988 }
3989 
3990 /*
3991  * Return:
3992  *   0 - lock acquired
3993  *   1 - caller should call rbd_request_lock()
3994  *  <0 - error
3995  */
3996 static int rbd_try_acquire_lock(struct rbd_device *rbd_dev)
3997 {
3998 	int ret;
3999 
4000 	down_read(&rbd_dev->lock_rwsem);
4001 	dout("%s rbd_dev %p read lock_state %d\n", __func__, rbd_dev,
4002 	     rbd_dev->lock_state);
4003 	if (__rbd_is_lock_owner(rbd_dev)) {
4004 		up_read(&rbd_dev->lock_rwsem);
4005 		return 0;
4006 	}
4007 
4008 	up_read(&rbd_dev->lock_rwsem);
4009 	down_write(&rbd_dev->lock_rwsem);
4010 	dout("%s rbd_dev %p write lock_state %d\n", __func__, rbd_dev,
4011 	     rbd_dev->lock_state);
4012 	if (__rbd_is_lock_owner(rbd_dev)) {
4013 		up_write(&rbd_dev->lock_rwsem);
4014 		return 0;
4015 	}
4016 
4017 	ret = rbd_try_lock(rbd_dev);
4018 	if (ret < 0) {
4019 		rbd_warn(rbd_dev, "failed to lock header: %d", ret);
4020 		if (ret == -EBLOCKLISTED)
4021 			goto out;
4022 
4023 		ret = 1; /* request lock anyway */
4024 	}
4025 	if (ret > 0) {
4026 		up_write(&rbd_dev->lock_rwsem);
4027 		return ret;
4028 	}
4029 
4030 	rbd_assert(rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED);
4031 	rbd_assert(list_empty(&rbd_dev->running_list));
4032 
4033 	ret = rbd_post_acquire_action(rbd_dev);
4034 	if (ret) {
4035 		rbd_warn(rbd_dev, "post-acquire action failed: %d", ret);
4036 		/*
4037 		 * Can't stay in RBD_LOCK_STATE_LOCKED because
4038 		 * rbd_lock_add_request() would let the request through,
4039 		 * assuming that e.g. object map is locked and loaded.
4040 		 */
4041 		rbd_unlock(rbd_dev);
4042 	}
4043 
4044 out:
4045 	wake_lock_waiters(rbd_dev, ret);
4046 	up_write(&rbd_dev->lock_rwsem);
4047 	return ret;
4048 }
4049 
4050 static void rbd_acquire_lock(struct work_struct *work)
4051 {
4052 	struct rbd_device *rbd_dev = container_of(to_delayed_work(work),
4053 					    struct rbd_device, lock_dwork);
4054 	int ret;
4055 
4056 	dout("%s rbd_dev %p\n", __func__, rbd_dev);
4057 again:
4058 	ret = rbd_try_acquire_lock(rbd_dev);
4059 	if (ret <= 0) {
4060 		dout("%s rbd_dev %p ret %d - done\n", __func__, rbd_dev, ret);
4061 		return;
4062 	}
4063 
4064 	ret = rbd_request_lock(rbd_dev);
4065 	if (ret == -ETIMEDOUT) {
4066 		goto again; /* treat this as a dead client */
4067 	} else if (ret == -EROFS) {
4068 		rbd_warn(rbd_dev, "peer will not release lock");
4069 		down_write(&rbd_dev->lock_rwsem);
4070 		wake_lock_waiters(rbd_dev, ret);
4071 		up_write(&rbd_dev->lock_rwsem);
4072 	} else if (ret < 0) {
4073 		rbd_warn(rbd_dev, "error requesting lock: %d", ret);
4074 		mod_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork,
4075 				 RBD_RETRY_DELAY);
4076 	} else {
4077 		/*
4078 		 * lock owner acked, but resend if we don't see them
4079 		 * release the lock
4080 		 */
4081 		dout("%s rbd_dev %p requeuing lock_dwork\n", __func__,
4082 		     rbd_dev);
4083 		mod_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork,
4084 		    msecs_to_jiffies(2 * RBD_NOTIFY_TIMEOUT * MSEC_PER_SEC));
4085 	}
4086 }
4087 
4088 static bool rbd_quiesce_lock(struct rbd_device *rbd_dev)
4089 {
4090 	dout("%s rbd_dev %p\n", __func__, rbd_dev);
4091 	lockdep_assert_held_write(&rbd_dev->lock_rwsem);
4092 
4093 	if (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED)
4094 		return false;
4095 
4096 	/*
4097 	 * Ensure that all in-flight IO is flushed.
4098 	 */
4099 	rbd_dev->lock_state = RBD_LOCK_STATE_RELEASING;
4100 	rbd_assert(!completion_done(&rbd_dev->releasing_wait));
4101 	if (list_empty(&rbd_dev->running_list))
4102 		return true;
4103 
4104 	up_write(&rbd_dev->lock_rwsem);
4105 	wait_for_completion(&rbd_dev->releasing_wait);
4106 
4107 	down_write(&rbd_dev->lock_rwsem);
4108 	if (rbd_dev->lock_state != RBD_LOCK_STATE_RELEASING)
4109 		return false;
4110 
4111 	rbd_assert(list_empty(&rbd_dev->running_list));
4112 	return true;
4113 }
4114 
4115 static void rbd_pre_release_action(struct rbd_device *rbd_dev)
4116 {
4117 	if (rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP)
4118 		rbd_object_map_close(rbd_dev);
4119 }
4120 
4121 static void __rbd_release_lock(struct rbd_device *rbd_dev)
4122 {
4123 	rbd_assert(list_empty(&rbd_dev->running_list));
4124 
4125 	rbd_pre_release_action(rbd_dev);
4126 	rbd_unlock(rbd_dev);
4127 }
4128 
4129 /*
4130  * lock_rwsem must be held for write
4131  */
4132 static void rbd_release_lock(struct rbd_device *rbd_dev)
4133 {
4134 	if (!rbd_quiesce_lock(rbd_dev))
4135 		return;
4136 
4137 	__rbd_release_lock(rbd_dev);
4138 
4139 	/*
4140 	 * Give others a chance to grab the lock - we would re-acquire
4141 	 * almost immediately if we got new IO while draining the running
4142 	 * list otherwise.  We need to ack our own notifications, so this
4143 	 * lock_dwork will be requeued from rbd_handle_released_lock() by
4144 	 * way of maybe_kick_acquire().
4145 	 */
4146 	cancel_delayed_work(&rbd_dev->lock_dwork);
4147 }
4148 
4149 static void rbd_release_lock_work(struct work_struct *work)
4150 {
4151 	struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
4152 						  unlock_work);
4153 
4154 	down_write(&rbd_dev->lock_rwsem);
4155 	rbd_release_lock(rbd_dev);
4156 	up_write(&rbd_dev->lock_rwsem);
4157 }
4158 
4159 static void maybe_kick_acquire(struct rbd_device *rbd_dev)
4160 {
4161 	bool have_requests;
4162 
4163 	dout("%s rbd_dev %p\n", __func__, rbd_dev);
4164 	if (__rbd_is_lock_owner(rbd_dev))
4165 		return;
4166 
4167 	spin_lock(&rbd_dev->lock_lists_lock);
4168 	have_requests = !list_empty(&rbd_dev->acquiring_list);
4169 	spin_unlock(&rbd_dev->lock_lists_lock);
4170 	if (have_requests || delayed_work_pending(&rbd_dev->lock_dwork)) {
4171 		dout("%s rbd_dev %p kicking lock_dwork\n", __func__, rbd_dev);
4172 		mod_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork, 0);
4173 	}
4174 }
4175 
4176 static void rbd_handle_acquired_lock(struct rbd_device *rbd_dev, u8 struct_v,
4177 				     void **p)
4178 {
4179 	struct rbd_client_id cid = { 0 };
4180 
4181 	if (struct_v >= 2) {
4182 		cid.gid = ceph_decode_64(p);
4183 		cid.handle = ceph_decode_64(p);
4184 	}
4185 
4186 	dout("%s rbd_dev %p cid %llu-%llu\n", __func__, rbd_dev, cid.gid,
4187 	     cid.handle);
4188 	if (!rbd_cid_equal(&cid, &rbd_empty_cid)) {
4189 		down_write(&rbd_dev->lock_rwsem);
4190 		if (rbd_cid_equal(&cid, &rbd_dev->owner_cid)) {
4191 			dout("%s rbd_dev %p cid %llu-%llu == owner_cid\n",
4192 			     __func__, rbd_dev, cid.gid, cid.handle);
4193 		} else {
4194 			rbd_set_owner_cid(rbd_dev, &cid);
4195 		}
4196 		downgrade_write(&rbd_dev->lock_rwsem);
4197 	} else {
4198 		down_read(&rbd_dev->lock_rwsem);
4199 	}
4200 
4201 	maybe_kick_acquire(rbd_dev);
4202 	up_read(&rbd_dev->lock_rwsem);
4203 }
4204 
4205 static void rbd_handle_released_lock(struct rbd_device *rbd_dev, u8 struct_v,
4206 				     void **p)
4207 {
4208 	struct rbd_client_id cid = { 0 };
4209 
4210 	if (struct_v >= 2) {
4211 		cid.gid = ceph_decode_64(p);
4212 		cid.handle = ceph_decode_64(p);
4213 	}
4214 
4215 	dout("%s rbd_dev %p cid %llu-%llu\n", __func__, rbd_dev, cid.gid,
4216 	     cid.handle);
4217 	if (!rbd_cid_equal(&cid, &rbd_empty_cid)) {
4218 		down_write(&rbd_dev->lock_rwsem);
4219 		if (!rbd_cid_equal(&cid, &rbd_dev->owner_cid)) {
4220 			dout("%s rbd_dev %p cid %llu-%llu != owner_cid %llu-%llu\n",
4221 			     __func__, rbd_dev, cid.gid, cid.handle,
4222 			     rbd_dev->owner_cid.gid, rbd_dev->owner_cid.handle);
4223 		} else {
4224 			rbd_set_owner_cid(rbd_dev, &rbd_empty_cid);
4225 		}
4226 		downgrade_write(&rbd_dev->lock_rwsem);
4227 	} else {
4228 		down_read(&rbd_dev->lock_rwsem);
4229 	}
4230 
4231 	maybe_kick_acquire(rbd_dev);
4232 	up_read(&rbd_dev->lock_rwsem);
4233 }
4234 
4235 /*
4236  * Returns result for ResponseMessage to be encoded (<= 0), or 1 if no
4237  * ResponseMessage is needed.
4238  */
4239 static int rbd_handle_request_lock(struct rbd_device *rbd_dev, u8 struct_v,
4240 				   void **p)
4241 {
4242 	struct rbd_client_id my_cid = rbd_get_cid(rbd_dev);
4243 	struct rbd_client_id cid = { 0 };
4244 	int result = 1;
4245 
4246 	if (struct_v >= 2) {
4247 		cid.gid = ceph_decode_64(p);
4248 		cid.handle = ceph_decode_64(p);
4249 	}
4250 
4251 	dout("%s rbd_dev %p cid %llu-%llu\n", __func__, rbd_dev, cid.gid,
4252 	     cid.handle);
4253 	if (rbd_cid_equal(&cid, &my_cid))
4254 		return result;
4255 
4256 	down_read(&rbd_dev->lock_rwsem);
4257 	if (__rbd_is_lock_owner(rbd_dev)) {
4258 		if (rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED &&
4259 		    rbd_cid_equal(&rbd_dev->owner_cid, &rbd_empty_cid))
4260 			goto out_unlock;
4261 
4262 		/*
4263 		 * encode ResponseMessage(0) so the peer can detect
4264 		 * a missing owner
4265 		 */
4266 		result = 0;
4267 
4268 		if (rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED) {
4269 			if (!rbd_dev->opts->exclusive) {
4270 				dout("%s rbd_dev %p queueing unlock_work\n",
4271 				     __func__, rbd_dev);
4272 				queue_work(rbd_dev->task_wq,
4273 					   &rbd_dev->unlock_work);
4274 			} else {
4275 				/* refuse to release the lock */
4276 				result = -EROFS;
4277 			}
4278 		}
4279 	}
4280 
4281 out_unlock:
4282 	up_read(&rbd_dev->lock_rwsem);
4283 	return result;
4284 }
4285 
4286 static void __rbd_acknowledge_notify(struct rbd_device *rbd_dev,
4287 				     u64 notify_id, u64 cookie, s32 *result)
4288 {
4289 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
4290 	char buf[4 + CEPH_ENCODING_START_BLK_LEN];
4291 	int buf_size = sizeof(buf);
4292 	int ret;
4293 
4294 	if (result) {
4295 		void *p = buf;
4296 
4297 		/* encode ResponseMessage */
4298 		ceph_start_encoding(&p, 1, 1,
4299 				    buf_size - CEPH_ENCODING_START_BLK_LEN);
4300 		ceph_encode_32(&p, *result);
4301 	} else {
4302 		buf_size = 0;
4303 	}
4304 
4305 	ret = ceph_osdc_notify_ack(osdc, &rbd_dev->header_oid,
4306 				   &rbd_dev->header_oloc, notify_id, cookie,
4307 				   buf, buf_size);
4308 	if (ret)
4309 		rbd_warn(rbd_dev, "acknowledge_notify failed: %d", ret);
4310 }
4311 
4312 static void rbd_acknowledge_notify(struct rbd_device *rbd_dev, u64 notify_id,
4313 				   u64 cookie)
4314 {
4315 	dout("%s rbd_dev %p\n", __func__, rbd_dev);
4316 	__rbd_acknowledge_notify(rbd_dev, notify_id, cookie, NULL);
4317 }
4318 
4319 static void rbd_acknowledge_notify_result(struct rbd_device *rbd_dev,
4320 					  u64 notify_id, u64 cookie, s32 result)
4321 {
4322 	dout("%s rbd_dev %p result %d\n", __func__, rbd_dev, result);
4323 	__rbd_acknowledge_notify(rbd_dev, notify_id, cookie, &result);
4324 }
4325 
4326 static void rbd_watch_cb(void *arg, u64 notify_id, u64 cookie,
4327 			 u64 notifier_id, void *data, size_t data_len)
4328 {
4329 	struct rbd_device *rbd_dev = arg;
4330 	void *p = data;
4331 	void *const end = p + data_len;
4332 	u8 struct_v = 0;
4333 	u32 len;
4334 	u32 notify_op;
4335 	int ret;
4336 
4337 	dout("%s rbd_dev %p cookie %llu notify_id %llu data_len %zu\n",
4338 	     __func__, rbd_dev, cookie, notify_id, data_len);
4339 	if (data_len) {
4340 		ret = ceph_start_decoding(&p, end, 1, "NotifyMessage",
4341 					  &struct_v, &len);
4342 		if (ret) {
4343 			rbd_warn(rbd_dev, "failed to decode NotifyMessage: %d",
4344 				 ret);
4345 			return;
4346 		}
4347 
4348 		notify_op = ceph_decode_32(&p);
4349 	} else {
4350 		/* legacy notification for header updates */
4351 		notify_op = RBD_NOTIFY_OP_HEADER_UPDATE;
4352 		len = 0;
4353 	}
4354 
4355 	dout("%s rbd_dev %p notify_op %u\n", __func__, rbd_dev, notify_op);
4356 	switch (notify_op) {
4357 	case RBD_NOTIFY_OP_ACQUIRED_LOCK:
4358 		rbd_handle_acquired_lock(rbd_dev, struct_v, &p);
4359 		rbd_acknowledge_notify(rbd_dev, notify_id, cookie);
4360 		break;
4361 	case RBD_NOTIFY_OP_RELEASED_LOCK:
4362 		rbd_handle_released_lock(rbd_dev, struct_v, &p);
4363 		rbd_acknowledge_notify(rbd_dev, notify_id, cookie);
4364 		break;
4365 	case RBD_NOTIFY_OP_REQUEST_LOCK:
4366 		ret = rbd_handle_request_lock(rbd_dev, struct_v, &p);
4367 		if (ret <= 0)
4368 			rbd_acknowledge_notify_result(rbd_dev, notify_id,
4369 						      cookie, ret);
4370 		else
4371 			rbd_acknowledge_notify(rbd_dev, notify_id, cookie);
4372 		break;
4373 	case RBD_NOTIFY_OP_HEADER_UPDATE:
4374 		ret = rbd_dev_refresh(rbd_dev);
4375 		if (ret)
4376 			rbd_warn(rbd_dev, "refresh failed: %d", ret);
4377 
4378 		rbd_acknowledge_notify(rbd_dev, notify_id, cookie);
4379 		break;
4380 	default:
4381 		if (rbd_is_lock_owner(rbd_dev))
4382 			rbd_acknowledge_notify_result(rbd_dev, notify_id,
4383 						      cookie, -EOPNOTSUPP);
4384 		else
4385 			rbd_acknowledge_notify(rbd_dev, notify_id, cookie);
4386 		break;
4387 	}
4388 }
4389 
4390 static void __rbd_unregister_watch(struct rbd_device *rbd_dev);
4391 
4392 static void rbd_watch_errcb(void *arg, u64 cookie, int err)
4393 {
4394 	struct rbd_device *rbd_dev = arg;
4395 
4396 	rbd_warn(rbd_dev, "encountered watch error: %d", err);
4397 
4398 	down_write(&rbd_dev->lock_rwsem);
4399 	rbd_set_owner_cid(rbd_dev, &rbd_empty_cid);
4400 	up_write(&rbd_dev->lock_rwsem);
4401 
4402 	mutex_lock(&rbd_dev->watch_mutex);
4403 	if (rbd_dev->watch_state == RBD_WATCH_STATE_REGISTERED) {
4404 		__rbd_unregister_watch(rbd_dev);
4405 		rbd_dev->watch_state = RBD_WATCH_STATE_ERROR;
4406 
4407 		queue_delayed_work(rbd_dev->task_wq, &rbd_dev->watch_dwork, 0);
4408 	}
4409 	mutex_unlock(&rbd_dev->watch_mutex);
4410 }
4411 
4412 /*
4413  * watch_mutex must be locked
4414  */
4415 static int __rbd_register_watch(struct rbd_device *rbd_dev)
4416 {
4417 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
4418 	struct ceph_osd_linger_request *handle;
4419 
4420 	rbd_assert(!rbd_dev->watch_handle);
4421 	dout("%s rbd_dev %p\n", __func__, rbd_dev);
4422 
4423 	handle = ceph_osdc_watch(osdc, &rbd_dev->header_oid,
4424 				 &rbd_dev->header_oloc, rbd_watch_cb,
4425 				 rbd_watch_errcb, rbd_dev);
4426 	if (IS_ERR(handle))
4427 		return PTR_ERR(handle);
4428 
4429 	rbd_dev->watch_handle = handle;
4430 	return 0;
4431 }
4432 
4433 /*
4434  * watch_mutex must be locked
4435  */
4436 static void __rbd_unregister_watch(struct rbd_device *rbd_dev)
4437 {
4438 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
4439 	int ret;
4440 
4441 	rbd_assert(rbd_dev->watch_handle);
4442 	dout("%s rbd_dev %p\n", __func__, rbd_dev);
4443 
4444 	ret = ceph_osdc_unwatch(osdc, rbd_dev->watch_handle);
4445 	if (ret)
4446 		rbd_warn(rbd_dev, "failed to unwatch: %d", ret);
4447 
4448 	rbd_dev->watch_handle = NULL;
4449 }
4450 
4451 static int rbd_register_watch(struct rbd_device *rbd_dev)
4452 {
4453 	int ret;
4454 
4455 	mutex_lock(&rbd_dev->watch_mutex);
4456 	rbd_assert(rbd_dev->watch_state == RBD_WATCH_STATE_UNREGISTERED);
4457 	ret = __rbd_register_watch(rbd_dev);
4458 	if (ret)
4459 		goto out;
4460 
4461 	rbd_dev->watch_state = RBD_WATCH_STATE_REGISTERED;
4462 	rbd_dev->watch_cookie = rbd_dev->watch_handle->linger_id;
4463 
4464 out:
4465 	mutex_unlock(&rbd_dev->watch_mutex);
4466 	return ret;
4467 }
4468 
4469 static void cancel_tasks_sync(struct rbd_device *rbd_dev)
4470 {
4471 	dout("%s rbd_dev %p\n", __func__, rbd_dev);
4472 
4473 	cancel_work_sync(&rbd_dev->acquired_lock_work);
4474 	cancel_work_sync(&rbd_dev->released_lock_work);
4475 	cancel_delayed_work_sync(&rbd_dev->lock_dwork);
4476 	cancel_work_sync(&rbd_dev->unlock_work);
4477 }
4478 
4479 /*
4480  * header_rwsem must not be held to avoid a deadlock with
4481  * rbd_dev_refresh() when flushing notifies.
4482  */
4483 static void rbd_unregister_watch(struct rbd_device *rbd_dev)
4484 {
4485 	cancel_tasks_sync(rbd_dev);
4486 
4487 	mutex_lock(&rbd_dev->watch_mutex);
4488 	if (rbd_dev->watch_state == RBD_WATCH_STATE_REGISTERED)
4489 		__rbd_unregister_watch(rbd_dev);
4490 	rbd_dev->watch_state = RBD_WATCH_STATE_UNREGISTERED;
4491 	mutex_unlock(&rbd_dev->watch_mutex);
4492 
4493 	cancel_delayed_work_sync(&rbd_dev->watch_dwork);
4494 	ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc);
4495 }
4496 
4497 /*
4498  * lock_rwsem must be held for write
4499  */
4500 static void rbd_reacquire_lock(struct rbd_device *rbd_dev)
4501 {
4502 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
4503 	char cookie[32];
4504 	int ret;
4505 
4506 	if (!rbd_quiesce_lock(rbd_dev))
4507 		return;
4508 
4509 	format_lock_cookie(rbd_dev, cookie);
4510 	ret = ceph_cls_set_cookie(osdc, &rbd_dev->header_oid,
4511 				  &rbd_dev->header_oloc, RBD_LOCK_NAME,
4512 				  CEPH_CLS_LOCK_EXCLUSIVE, rbd_dev->lock_cookie,
4513 				  RBD_LOCK_TAG, cookie);
4514 	if (ret) {
4515 		if (ret != -EOPNOTSUPP)
4516 			rbd_warn(rbd_dev, "failed to update lock cookie: %d",
4517 				 ret);
4518 
4519 		/*
4520 		 * Lock cookie cannot be updated on older OSDs, so do
4521 		 * a manual release and queue an acquire.
4522 		 */
4523 		__rbd_release_lock(rbd_dev);
4524 		queue_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork, 0);
4525 	} else {
4526 		__rbd_lock(rbd_dev, cookie);
4527 		wake_lock_waiters(rbd_dev, 0);
4528 	}
4529 }
4530 
4531 static void rbd_reregister_watch(struct work_struct *work)
4532 {
4533 	struct rbd_device *rbd_dev = container_of(to_delayed_work(work),
4534 					    struct rbd_device, watch_dwork);
4535 	int ret;
4536 
4537 	dout("%s rbd_dev %p\n", __func__, rbd_dev);
4538 
4539 	mutex_lock(&rbd_dev->watch_mutex);
4540 	if (rbd_dev->watch_state != RBD_WATCH_STATE_ERROR) {
4541 		mutex_unlock(&rbd_dev->watch_mutex);
4542 		return;
4543 	}
4544 
4545 	ret = __rbd_register_watch(rbd_dev);
4546 	if (ret) {
4547 		rbd_warn(rbd_dev, "failed to reregister watch: %d", ret);
4548 		if (ret != -EBLOCKLISTED && ret != -ENOENT) {
4549 			queue_delayed_work(rbd_dev->task_wq,
4550 					   &rbd_dev->watch_dwork,
4551 					   RBD_RETRY_DELAY);
4552 			mutex_unlock(&rbd_dev->watch_mutex);
4553 			return;
4554 		}
4555 
4556 		mutex_unlock(&rbd_dev->watch_mutex);
4557 		down_write(&rbd_dev->lock_rwsem);
4558 		wake_lock_waiters(rbd_dev, ret);
4559 		up_write(&rbd_dev->lock_rwsem);
4560 		return;
4561 	}
4562 
4563 	rbd_dev->watch_state = RBD_WATCH_STATE_REGISTERED;
4564 	rbd_dev->watch_cookie = rbd_dev->watch_handle->linger_id;
4565 	mutex_unlock(&rbd_dev->watch_mutex);
4566 
4567 	down_write(&rbd_dev->lock_rwsem);
4568 	if (rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED)
4569 		rbd_reacquire_lock(rbd_dev);
4570 	up_write(&rbd_dev->lock_rwsem);
4571 
4572 	ret = rbd_dev_refresh(rbd_dev);
4573 	if (ret)
4574 		rbd_warn(rbd_dev, "reregistration refresh failed: %d", ret);
4575 }
4576 
4577 /*
4578  * Synchronous osd object method call.  Returns the number of bytes
4579  * returned in the outbound buffer, or a negative error code.
4580  */
4581 static int rbd_obj_method_sync(struct rbd_device *rbd_dev,
4582 			     struct ceph_object_id *oid,
4583 			     struct ceph_object_locator *oloc,
4584 			     const char *method_name,
4585 			     const void *outbound,
4586 			     size_t outbound_size,
4587 			     void *inbound,
4588 			     size_t inbound_size)
4589 {
4590 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
4591 	struct page *req_page = NULL;
4592 	struct page *reply_page;
4593 	int ret;
4594 
4595 	/*
4596 	 * Method calls are ultimately read operations.  The result
4597 	 * should placed into the inbound buffer provided.  They
4598 	 * also supply outbound data--parameters for the object
4599 	 * method.  Currently if this is present it will be a
4600 	 * snapshot id.
4601 	 */
4602 	if (outbound) {
4603 		if (outbound_size > PAGE_SIZE)
4604 			return -E2BIG;
4605 
4606 		req_page = alloc_page(GFP_KERNEL);
4607 		if (!req_page)
4608 			return -ENOMEM;
4609 
4610 		memcpy(page_address(req_page), outbound, outbound_size);
4611 	}
4612 
4613 	reply_page = alloc_page(GFP_KERNEL);
4614 	if (!reply_page) {
4615 		if (req_page)
4616 			__free_page(req_page);
4617 		return -ENOMEM;
4618 	}
4619 
4620 	ret = ceph_osdc_call(osdc, oid, oloc, RBD_DRV_NAME, method_name,
4621 			     CEPH_OSD_FLAG_READ, req_page, outbound_size,
4622 			     &reply_page, &inbound_size);
4623 	if (!ret) {
4624 		memcpy(inbound, page_address(reply_page), inbound_size);
4625 		ret = inbound_size;
4626 	}
4627 
4628 	if (req_page)
4629 		__free_page(req_page);
4630 	__free_page(reply_page);
4631 	return ret;
4632 }
4633 
4634 static void rbd_queue_workfn(struct work_struct *work)
4635 {
4636 	struct rbd_img_request *img_request =
4637 	    container_of(work, struct rbd_img_request, work);
4638 	struct rbd_device *rbd_dev = img_request->rbd_dev;
4639 	enum obj_operation_type op_type = img_request->op_type;
4640 	struct request *rq = blk_mq_rq_from_pdu(img_request);
4641 	u64 offset = (u64)blk_rq_pos(rq) << SECTOR_SHIFT;
4642 	u64 length = blk_rq_bytes(rq);
4643 	u64 mapping_size;
4644 	int result;
4645 
4646 	/* Ignore/skip any zero-length requests */
4647 	if (!length) {
4648 		dout("%s: zero-length request\n", __func__);
4649 		result = 0;
4650 		goto err_img_request;
4651 	}
4652 
4653 	blk_mq_start_request(rq);
4654 
4655 	down_read(&rbd_dev->header_rwsem);
4656 	mapping_size = rbd_dev->mapping.size;
4657 	rbd_img_capture_header(img_request);
4658 	up_read(&rbd_dev->header_rwsem);
4659 
4660 	if (offset + length > mapping_size) {
4661 		rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)", offset,
4662 			 length, mapping_size);
4663 		result = -EIO;
4664 		goto err_img_request;
4665 	}
4666 
4667 	dout("%s rbd_dev %p img_req %p %s %llu~%llu\n", __func__, rbd_dev,
4668 	     img_request, obj_op_name(op_type), offset, length);
4669 
4670 	if (op_type == OBJ_OP_DISCARD || op_type == OBJ_OP_ZEROOUT)
4671 		result = rbd_img_fill_nodata(img_request, offset, length);
4672 	else
4673 		result = rbd_img_fill_from_bio(img_request, offset, length,
4674 					       rq->bio);
4675 	if (result)
4676 		goto err_img_request;
4677 
4678 	rbd_img_handle_request(img_request, 0);
4679 	return;
4680 
4681 err_img_request:
4682 	rbd_img_request_destroy(img_request);
4683 	if (result)
4684 		rbd_warn(rbd_dev, "%s %llx at %llx result %d",
4685 			 obj_op_name(op_type), length, offset, result);
4686 	blk_mq_end_request(rq, errno_to_blk_status(result));
4687 }
4688 
4689 static blk_status_t rbd_queue_rq(struct blk_mq_hw_ctx *hctx,
4690 		const struct blk_mq_queue_data *bd)
4691 {
4692 	struct rbd_device *rbd_dev = hctx->queue->queuedata;
4693 	struct rbd_img_request *img_req = blk_mq_rq_to_pdu(bd->rq);
4694 	enum obj_operation_type op_type;
4695 
4696 	switch (req_op(bd->rq)) {
4697 	case REQ_OP_DISCARD:
4698 		op_type = OBJ_OP_DISCARD;
4699 		break;
4700 	case REQ_OP_WRITE_ZEROES:
4701 		op_type = OBJ_OP_ZEROOUT;
4702 		break;
4703 	case REQ_OP_WRITE:
4704 		op_type = OBJ_OP_WRITE;
4705 		break;
4706 	case REQ_OP_READ:
4707 		op_type = OBJ_OP_READ;
4708 		break;
4709 	default:
4710 		rbd_warn(rbd_dev, "unknown req_op %d", req_op(bd->rq));
4711 		return BLK_STS_IOERR;
4712 	}
4713 
4714 	rbd_img_request_init(img_req, rbd_dev, op_type);
4715 
4716 	if (rbd_img_is_write(img_req)) {
4717 		if (rbd_is_ro(rbd_dev)) {
4718 			rbd_warn(rbd_dev, "%s on read-only mapping",
4719 				 obj_op_name(img_req->op_type));
4720 			return BLK_STS_IOERR;
4721 		}
4722 		rbd_assert(!rbd_is_snap(rbd_dev));
4723 	}
4724 
4725 	INIT_WORK(&img_req->work, rbd_queue_workfn);
4726 	queue_work(rbd_wq, &img_req->work);
4727 	return BLK_STS_OK;
4728 }
4729 
4730 static void rbd_free_disk(struct rbd_device *rbd_dev)
4731 {
4732 	blk_cleanup_disk(rbd_dev->disk);
4733 	blk_mq_free_tag_set(&rbd_dev->tag_set);
4734 	rbd_dev->disk = NULL;
4735 }
4736 
4737 static int rbd_obj_read_sync(struct rbd_device *rbd_dev,
4738 			     struct ceph_object_id *oid,
4739 			     struct ceph_object_locator *oloc,
4740 			     void *buf, int buf_len)
4741 
4742 {
4743 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
4744 	struct ceph_osd_request *req;
4745 	struct page **pages;
4746 	int num_pages = calc_pages_for(0, buf_len);
4747 	int ret;
4748 
4749 	req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_KERNEL);
4750 	if (!req)
4751 		return -ENOMEM;
4752 
4753 	ceph_oid_copy(&req->r_base_oid, oid);
4754 	ceph_oloc_copy(&req->r_base_oloc, oloc);
4755 	req->r_flags = CEPH_OSD_FLAG_READ;
4756 
4757 	pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
4758 	if (IS_ERR(pages)) {
4759 		ret = PTR_ERR(pages);
4760 		goto out_req;
4761 	}
4762 
4763 	osd_req_op_extent_init(req, 0, CEPH_OSD_OP_READ, 0, buf_len, 0, 0);
4764 	osd_req_op_extent_osd_data_pages(req, 0, pages, buf_len, 0, false,
4765 					 true);
4766 
4767 	ret = ceph_osdc_alloc_messages(req, GFP_KERNEL);
4768 	if (ret)
4769 		goto out_req;
4770 
4771 	ceph_osdc_start_request(osdc, req, false);
4772 	ret = ceph_osdc_wait_request(osdc, req);
4773 	if (ret >= 0)
4774 		ceph_copy_from_page_vector(pages, buf, 0, ret);
4775 
4776 out_req:
4777 	ceph_osdc_put_request(req);
4778 	return ret;
4779 }
4780 
4781 /*
4782  * Read the complete header for the given rbd device.  On successful
4783  * return, the rbd_dev->header field will contain up-to-date
4784  * information about the image.
4785  */
4786 static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev)
4787 {
4788 	struct rbd_image_header_ondisk *ondisk = NULL;
4789 	u32 snap_count = 0;
4790 	u64 names_size = 0;
4791 	u32 want_count;
4792 	int ret;
4793 
4794 	/*
4795 	 * The complete header will include an array of its 64-bit
4796 	 * snapshot ids, followed by the names of those snapshots as
4797 	 * a contiguous block of NUL-terminated strings.  Note that
4798 	 * the number of snapshots could change by the time we read
4799 	 * it in, in which case we re-read it.
4800 	 */
4801 	do {
4802 		size_t size;
4803 
4804 		kfree(ondisk);
4805 
4806 		size = sizeof (*ondisk);
4807 		size += snap_count * sizeof (struct rbd_image_snap_ondisk);
4808 		size += names_size;
4809 		ondisk = kmalloc(size, GFP_KERNEL);
4810 		if (!ondisk)
4811 			return -ENOMEM;
4812 
4813 		ret = rbd_obj_read_sync(rbd_dev, &rbd_dev->header_oid,
4814 					&rbd_dev->header_oloc, ondisk, size);
4815 		if (ret < 0)
4816 			goto out;
4817 		if ((size_t)ret < size) {
4818 			ret = -ENXIO;
4819 			rbd_warn(rbd_dev, "short header read (want %zd got %d)",
4820 				size, ret);
4821 			goto out;
4822 		}
4823 		if (!rbd_dev_ondisk_valid(ondisk)) {
4824 			ret = -ENXIO;
4825 			rbd_warn(rbd_dev, "invalid header");
4826 			goto out;
4827 		}
4828 
4829 		names_size = le64_to_cpu(ondisk->snap_names_len);
4830 		want_count = snap_count;
4831 		snap_count = le32_to_cpu(ondisk->snap_count);
4832 	} while (snap_count != want_count);
4833 
4834 	ret = rbd_header_from_disk(rbd_dev, ondisk);
4835 out:
4836 	kfree(ondisk);
4837 
4838 	return ret;
4839 }
4840 
4841 static void rbd_dev_update_size(struct rbd_device *rbd_dev)
4842 {
4843 	sector_t size;
4844 
4845 	/*
4846 	 * If EXISTS is not set, rbd_dev->disk may be NULL, so don't
4847 	 * try to update its size.  If REMOVING is set, updating size
4848 	 * is just useless work since the device can't be opened.
4849 	 */
4850 	if (test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags) &&
4851 	    !test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) {
4852 		size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE;
4853 		dout("setting size to %llu sectors", (unsigned long long)size);
4854 		set_capacity_and_notify(rbd_dev->disk, size);
4855 	}
4856 }
4857 
4858 static int rbd_dev_refresh(struct rbd_device *rbd_dev)
4859 {
4860 	u64 mapping_size;
4861 	int ret;
4862 
4863 	down_write(&rbd_dev->header_rwsem);
4864 	mapping_size = rbd_dev->mapping.size;
4865 
4866 	ret = rbd_dev_header_info(rbd_dev);
4867 	if (ret)
4868 		goto out;
4869 
4870 	/*
4871 	 * If there is a parent, see if it has disappeared due to the
4872 	 * mapped image getting flattened.
4873 	 */
4874 	if (rbd_dev->parent) {
4875 		ret = rbd_dev_v2_parent_info(rbd_dev);
4876 		if (ret)
4877 			goto out;
4878 	}
4879 
4880 	rbd_assert(!rbd_is_snap(rbd_dev));
4881 	rbd_dev->mapping.size = rbd_dev->header.image_size;
4882 
4883 out:
4884 	up_write(&rbd_dev->header_rwsem);
4885 	if (!ret && mapping_size != rbd_dev->mapping.size)
4886 		rbd_dev_update_size(rbd_dev);
4887 
4888 	return ret;
4889 }
4890 
4891 static const struct blk_mq_ops rbd_mq_ops = {
4892 	.queue_rq	= rbd_queue_rq,
4893 };
4894 
4895 static int rbd_init_disk(struct rbd_device *rbd_dev)
4896 {
4897 	struct gendisk *disk;
4898 	struct request_queue *q;
4899 	unsigned int objset_bytes =
4900 	    rbd_dev->layout.object_size * rbd_dev->layout.stripe_count;
4901 	int err;
4902 
4903 	memset(&rbd_dev->tag_set, 0, sizeof(rbd_dev->tag_set));
4904 	rbd_dev->tag_set.ops = &rbd_mq_ops;
4905 	rbd_dev->tag_set.queue_depth = rbd_dev->opts->queue_depth;
4906 	rbd_dev->tag_set.numa_node = NUMA_NO_NODE;
4907 	rbd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
4908 	rbd_dev->tag_set.nr_hw_queues = num_present_cpus();
4909 	rbd_dev->tag_set.cmd_size = sizeof(struct rbd_img_request);
4910 
4911 	err = blk_mq_alloc_tag_set(&rbd_dev->tag_set);
4912 	if (err)
4913 		return err;
4914 
4915 	disk = blk_mq_alloc_disk(&rbd_dev->tag_set, rbd_dev);
4916 	if (IS_ERR(disk)) {
4917 		err = PTR_ERR(disk);
4918 		goto out_tag_set;
4919 	}
4920 	q = disk->queue;
4921 
4922 	snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
4923 		 rbd_dev->dev_id);
4924 	disk->major = rbd_dev->major;
4925 	disk->first_minor = rbd_dev->minor;
4926 	if (single_major)
4927 		disk->minors = (1 << RBD_SINGLE_MAJOR_PART_SHIFT);
4928 	else
4929 		disk->minors = RBD_MINORS_PER_MAJOR;
4930 	disk->fops = &rbd_bd_ops;
4931 	disk->private_data = rbd_dev;
4932 
4933 	blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
4934 	/* QUEUE_FLAG_ADD_RANDOM is off by default for blk-mq */
4935 
4936 	blk_queue_max_hw_sectors(q, objset_bytes >> SECTOR_SHIFT);
4937 	q->limits.max_sectors = queue_max_hw_sectors(q);
4938 	blk_queue_max_segments(q, USHRT_MAX);
4939 	blk_queue_max_segment_size(q, UINT_MAX);
4940 	blk_queue_io_min(q, rbd_dev->opts->alloc_size);
4941 	blk_queue_io_opt(q, rbd_dev->opts->alloc_size);
4942 
4943 	if (rbd_dev->opts->trim) {
4944 		q->limits.discard_granularity = rbd_dev->opts->alloc_size;
4945 		blk_queue_max_discard_sectors(q, objset_bytes >> SECTOR_SHIFT);
4946 		blk_queue_max_write_zeroes_sectors(q, objset_bytes >> SECTOR_SHIFT);
4947 	}
4948 
4949 	if (!ceph_test_opt(rbd_dev->rbd_client->client, NOCRC))
4950 		blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, q);
4951 
4952 	rbd_dev->disk = disk;
4953 
4954 	return 0;
4955 out_tag_set:
4956 	blk_mq_free_tag_set(&rbd_dev->tag_set);
4957 	return err;
4958 }
4959 
4960 /*
4961   sysfs
4962 */
4963 
4964 static struct rbd_device *dev_to_rbd_dev(struct device *dev)
4965 {
4966 	return container_of(dev, struct rbd_device, dev);
4967 }
4968 
4969 static ssize_t rbd_size_show(struct device *dev,
4970 			     struct device_attribute *attr, char *buf)
4971 {
4972 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4973 
4974 	return sprintf(buf, "%llu\n",
4975 		(unsigned long long)rbd_dev->mapping.size);
4976 }
4977 
4978 static ssize_t rbd_features_show(struct device *dev,
4979 			     struct device_attribute *attr, char *buf)
4980 {
4981 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4982 
4983 	return sprintf(buf, "0x%016llx\n", rbd_dev->header.features);
4984 }
4985 
4986 static ssize_t rbd_major_show(struct device *dev,
4987 			      struct device_attribute *attr, char *buf)
4988 {
4989 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
4990 
4991 	if (rbd_dev->major)
4992 		return sprintf(buf, "%d\n", rbd_dev->major);
4993 
4994 	return sprintf(buf, "(none)\n");
4995 }
4996 
4997 static ssize_t rbd_minor_show(struct device *dev,
4998 			      struct device_attribute *attr, char *buf)
4999 {
5000 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5001 
5002 	return sprintf(buf, "%d\n", rbd_dev->minor);
5003 }
5004 
5005 static ssize_t rbd_client_addr_show(struct device *dev,
5006 				    struct device_attribute *attr, char *buf)
5007 {
5008 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5009 	struct ceph_entity_addr *client_addr =
5010 	    ceph_client_addr(rbd_dev->rbd_client->client);
5011 
5012 	return sprintf(buf, "%pISpc/%u\n", &client_addr->in_addr,
5013 		       le32_to_cpu(client_addr->nonce));
5014 }
5015 
5016 static ssize_t rbd_client_id_show(struct device *dev,
5017 				  struct device_attribute *attr, char *buf)
5018 {
5019 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5020 
5021 	return sprintf(buf, "client%lld\n",
5022 		       ceph_client_gid(rbd_dev->rbd_client->client));
5023 }
5024 
5025 static ssize_t rbd_cluster_fsid_show(struct device *dev,
5026 				     struct device_attribute *attr, char *buf)
5027 {
5028 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5029 
5030 	return sprintf(buf, "%pU\n", &rbd_dev->rbd_client->client->fsid);
5031 }
5032 
5033 static ssize_t rbd_config_info_show(struct device *dev,
5034 				    struct device_attribute *attr, char *buf)
5035 {
5036 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5037 
5038 	if (!capable(CAP_SYS_ADMIN))
5039 		return -EPERM;
5040 
5041 	return sprintf(buf, "%s\n", rbd_dev->config_info);
5042 }
5043 
5044 static ssize_t rbd_pool_show(struct device *dev,
5045 			     struct device_attribute *attr, char *buf)
5046 {
5047 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5048 
5049 	return sprintf(buf, "%s\n", rbd_dev->spec->pool_name);
5050 }
5051 
5052 static ssize_t rbd_pool_id_show(struct device *dev,
5053 			     struct device_attribute *attr, char *buf)
5054 {
5055 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5056 
5057 	return sprintf(buf, "%llu\n",
5058 			(unsigned long long) rbd_dev->spec->pool_id);
5059 }
5060 
5061 static ssize_t rbd_pool_ns_show(struct device *dev,
5062 				struct device_attribute *attr, char *buf)
5063 {
5064 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5065 
5066 	return sprintf(buf, "%s\n", rbd_dev->spec->pool_ns ?: "");
5067 }
5068 
5069 static ssize_t rbd_name_show(struct device *dev,
5070 			     struct device_attribute *attr, char *buf)
5071 {
5072 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5073 
5074 	if (rbd_dev->spec->image_name)
5075 		return sprintf(buf, "%s\n", rbd_dev->spec->image_name);
5076 
5077 	return sprintf(buf, "(unknown)\n");
5078 }
5079 
5080 static ssize_t rbd_image_id_show(struct device *dev,
5081 			     struct device_attribute *attr, char *buf)
5082 {
5083 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5084 
5085 	return sprintf(buf, "%s\n", rbd_dev->spec->image_id);
5086 }
5087 
5088 /*
5089  * Shows the name of the currently-mapped snapshot (or
5090  * RBD_SNAP_HEAD_NAME for the base image).
5091  */
5092 static ssize_t rbd_snap_show(struct device *dev,
5093 			     struct device_attribute *attr,
5094 			     char *buf)
5095 {
5096 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5097 
5098 	return sprintf(buf, "%s\n", rbd_dev->spec->snap_name);
5099 }
5100 
5101 static ssize_t rbd_snap_id_show(struct device *dev,
5102 				struct device_attribute *attr, char *buf)
5103 {
5104 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5105 
5106 	return sprintf(buf, "%llu\n", rbd_dev->spec->snap_id);
5107 }
5108 
5109 /*
5110  * For a v2 image, shows the chain of parent images, separated by empty
5111  * lines.  For v1 images or if there is no parent, shows "(no parent
5112  * image)".
5113  */
5114 static ssize_t rbd_parent_show(struct device *dev,
5115 			       struct device_attribute *attr,
5116 			       char *buf)
5117 {
5118 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5119 	ssize_t count = 0;
5120 
5121 	if (!rbd_dev->parent)
5122 		return sprintf(buf, "(no parent image)\n");
5123 
5124 	for ( ; rbd_dev->parent; rbd_dev = rbd_dev->parent) {
5125 		struct rbd_spec *spec = rbd_dev->parent_spec;
5126 
5127 		count += sprintf(&buf[count], "%s"
5128 			    "pool_id %llu\npool_name %s\n"
5129 			    "pool_ns %s\n"
5130 			    "image_id %s\nimage_name %s\n"
5131 			    "snap_id %llu\nsnap_name %s\n"
5132 			    "overlap %llu\n",
5133 			    !count ? "" : "\n", /* first? */
5134 			    spec->pool_id, spec->pool_name,
5135 			    spec->pool_ns ?: "",
5136 			    spec->image_id, spec->image_name ?: "(unknown)",
5137 			    spec->snap_id, spec->snap_name,
5138 			    rbd_dev->parent_overlap);
5139 	}
5140 
5141 	return count;
5142 }
5143 
5144 static ssize_t rbd_image_refresh(struct device *dev,
5145 				 struct device_attribute *attr,
5146 				 const char *buf,
5147 				 size_t size)
5148 {
5149 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5150 	int ret;
5151 
5152 	if (!capable(CAP_SYS_ADMIN))
5153 		return -EPERM;
5154 
5155 	ret = rbd_dev_refresh(rbd_dev);
5156 	if (ret)
5157 		return ret;
5158 
5159 	return size;
5160 }
5161 
5162 static DEVICE_ATTR(size, 0444, rbd_size_show, NULL);
5163 static DEVICE_ATTR(features, 0444, rbd_features_show, NULL);
5164 static DEVICE_ATTR(major, 0444, rbd_major_show, NULL);
5165 static DEVICE_ATTR(minor, 0444, rbd_minor_show, NULL);
5166 static DEVICE_ATTR(client_addr, 0444, rbd_client_addr_show, NULL);
5167 static DEVICE_ATTR(client_id, 0444, rbd_client_id_show, NULL);
5168 static DEVICE_ATTR(cluster_fsid, 0444, rbd_cluster_fsid_show, NULL);
5169 static DEVICE_ATTR(config_info, 0400, rbd_config_info_show, NULL);
5170 static DEVICE_ATTR(pool, 0444, rbd_pool_show, NULL);
5171 static DEVICE_ATTR(pool_id, 0444, rbd_pool_id_show, NULL);
5172 static DEVICE_ATTR(pool_ns, 0444, rbd_pool_ns_show, NULL);
5173 static DEVICE_ATTR(name, 0444, rbd_name_show, NULL);
5174 static DEVICE_ATTR(image_id, 0444, rbd_image_id_show, NULL);
5175 static DEVICE_ATTR(refresh, 0200, NULL, rbd_image_refresh);
5176 static DEVICE_ATTR(current_snap, 0444, rbd_snap_show, NULL);
5177 static DEVICE_ATTR(snap_id, 0444, rbd_snap_id_show, NULL);
5178 static DEVICE_ATTR(parent, 0444, rbd_parent_show, NULL);
5179 
5180 static struct attribute *rbd_attrs[] = {
5181 	&dev_attr_size.attr,
5182 	&dev_attr_features.attr,
5183 	&dev_attr_major.attr,
5184 	&dev_attr_minor.attr,
5185 	&dev_attr_client_addr.attr,
5186 	&dev_attr_client_id.attr,
5187 	&dev_attr_cluster_fsid.attr,
5188 	&dev_attr_config_info.attr,
5189 	&dev_attr_pool.attr,
5190 	&dev_attr_pool_id.attr,
5191 	&dev_attr_pool_ns.attr,
5192 	&dev_attr_name.attr,
5193 	&dev_attr_image_id.attr,
5194 	&dev_attr_current_snap.attr,
5195 	&dev_attr_snap_id.attr,
5196 	&dev_attr_parent.attr,
5197 	&dev_attr_refresh.attr,
5198 	NULL
5199 };
5200 
5201 static struct attribute_group rbd_attr_group = {
5202 	.attrs = rbd_attrs,
5203 };
5204 
5205 static const struct attribute_group *rbd_attr_groups[] = {
5206 	&rbd_attr_group,
5207 	NULL
5208 };
5209 
5210 static void rbd_dev_release(struct device *dev);
5211 
5212 static const struct device_type rbd_device_type = {
5213 	.name		= "rbd",
5214 	.groups		= rbd_attr_groups,
5215 	.release	= rbd_dev_release,
5216 };
5217 
5218 static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec)
5219 {
5220 	kref_get(&spec->kref);
5221 
5222 	return spec;
5223 }
5224 
5225 static void rbd_spec_free(struct kref *kref);
5226 static void rbd_spec_put(struct rbd_spec *spec)
5227 {
5228 	if (spec)
5229 		kref_put(&spec->kref, rbd_spec_free);
5230 }
5231 
5232 static struct rbd_spec *rbd_spec_alloc(void)
5233 {
5234 	struct rbd_spec *spec;
5235 
5236 	spec = kzalloc(sizeof (*spec), GFP_KERNEL);
5237 	if (!spec)
5238 		return NULL;
5239 
5240 	spec->pool_id = CEPH_NOPOOL;
5241 	spec->snap_id = CEPH_NOSNAP;
5242 	kref_init(&spec->kref);
5243 
5244 	return spec;
5245 }
5246 
5247 static void rbd_spec_free(struct kref *kref)
5248 {
5249 	struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref);
5250 
5251 	kfree(spec->pool_name);
5252 	kfree(spec->pool_ns);
5253 	kfree(spec->image_id);
5254 	kfree(spec->image_name);
5255 	kfree(spec->snap_name);
5256 	kfree(spec);
5257 }
5258 
5259 static void rbd_dev_free(struct rbd_device *rbd_dev)
5260 {
5261 	WARN_ON(rbd_dev->watch_state != RBD_WATCH_STATE_UNREGISTERED);
5262 	WARN_ON(rbd_dev->lock_state != RBD_LOCK_STATE_UNLOCKED);
5263 
5264 	ceph_oid_destroy(&rbd_dev->header_oid);
5265 	ceph_oloc_destroy(&rbd_dev->header_oloc);
5266 	kfree(rbd_dev->config_info);
5267 
5268 	rbd_put_client(rbd_dev->rbd_client);
5269 	rbd_spec_put(rbd_dev->spec);
5270 	kfree(rbd_dev->opts);
5271 	kfree(rbd_dev);
5272 }
5273 
5274 static void rbd_dev_release(struct device *dev)
5275 {
5276 	struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
5277 	bool need_put = !!rbd_dev->opts;
5278 
5279 	if (need_put) {
5280 		destroy_workqueue(rbd_dev->task_wq);
5281 		ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id);
5282 	}
5283 
5284 	rbd_dev_free(rbd_dev);
5285 
5286 	/*
5287 	 * This is racy, but way better than putting module outside of
5288 	 * the release callback.  The race window is pretty small, so
5289 	 * doing something similar to dm (dm-builtin.c) is overkill.
5290 	 */
5291 	if (need_put)
5292 		module_put(THIS_MODULE);
5293 }
5294 
5295 static struct rbd_device *__rbd_dev_create(struct rbd_client *rbdc,
5296 					   struct rbd_spec *spec)
5297 {
5298 	struct rbd_device *rbd_dev;
5299 
5300 	rbd_dev = kzalloc(sizeof(*rbd_dev), GFP_KERNEL);
5301 	if (!rbd_dev)
5302 		return NULL;
5303 
5304 	spin_lock_init(&rbd_dev->lock);
5305 	INIT_LIST_HEAD(&rbd_dev->node);
5306 	init_rwsem(&rbd_dev->header_rwsem);
5307 
5308 	rbd_dev->header.data_pool_id = CEPH_NOPOOL;
5309 	ceph_oid_init(&rbd_dev->header_oid);
5310 	rbd_dev->header_oloc.pool = spec->pool_id;
5311 	if (spec->pool_ns) {
5312 		WARN_ON(!*spec->pool_ns);
5313 		rbd_dev->header_oloc.pool_ns =
5314 		    ceph_find_or_create_string(spec->pool_ns,
5315 					       strlen(spec->pool_ns));
5316 	}
5317 
5318 	mutex_init(&rbd_dev->watch_mutex);
5319 	rbd_dev->watch_state = RBD_WATCH_STATE_UNREGISTERED;
5320 	INIT_DELAYED_WORK(&rbd_dev->watch_dwork, rbd_reregister_watch);
5321 
5322 	init_rwsem(&rbd_dev->lock_rwsem);
5323 	rbd_dev->lock_state = RBD_LOCK_STATE_UNLOCKED;
5324 	INIT_WORK(&rbd_dev->acquired_lock_work, rbd_notify_acquired_lock);
5325 	INIT_WORK(&rbd_dev->released_lock_work, rbd_notify_released_lock);
5326 	INIT_DELAYED_WORK(&rbd_dev->lock_dwork, rbd_acquire_lock);
5327 	INIT_WORK(&rbd_dev->unlock_work, rbd_release_lock_work);
5328 	spin_lock_init(&rbd_dev->lock_lists_lock);
5329 	INIT_LIST_HEAD(&rbd_dev->acquiring_list);
5330 	INIT_LIST_HEAD(&rbd_dev->running_list);
5331 	init_completion(&rbd_dev->acquire_wait);
5332 	init_completion(&rbd_dev->releasing_wait);
5333 
5334 	spin_lock_init(&rbd_dev->object_map_lock);
5335 
5336 	rbd_dev->dev.bus = &rbd_bus_type;
5337 	rbd_dev->dev.type = &rbd_device_type;
5338 	rbd_dev->dev.parent = &rbd_root_dev;
5339 	device_initialize(&rbd_dev->dev);
5340 
5341 	rbd_dev->rbd_client = rbdc;
5342 	rbd_dev->spec = spec;
5343 
5344 	return rbd_dev;
5345 }
5346 
5347 /*
5348  * Create a mapping rbd_dev.
5349  */
5350 static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc,
5351 					 struct rbd_spec *spec,
5352 					 struct rbd_options *opts)
5353 {
5354 	struct rbd_device *rbd_dev;
5355 
5356 	rbd_dev = __rbd_dev_create(rbdc, spec);
5357 	if (!rbd_dev)
5358 		return NULL;
5359 
5360 	rbd_dev->opts = opts;
5361 
5362 	/* get an id and fill in device name */
5363 	rbd_dev->dev_id = ida_simple_get(&rbd_dev_id_ida, 0,
5364 					 minor_to_rbd_dev_id(1 << MINORBITS),
5365 					 GFP_KERNEL);
5366 	if (rbd_dev->dev_id < 0)
5367 		goto fail_rbd_dev;
5368 
5369 	sprintf(rbd_dev->name, RBD_DRV_NAME "%d", rbd_dev->dev_id);
5370 	rbd_dev->task_wq = alloc_ordered_workqueue("%s-tasks", WQ_MEM_RECLAIM,
5371 						   rbd_dev->name);
5372 	if (!rbd_dev->task_wq)
5373 		goto fail_dev_id;
5374 
5375 	/* we have a ref from do_rbd_add() */
5376 	__module_get(THIS_MODULE);
5377 
5378 	dout("%s rbd_dev %p dev_id %d\n", __func__, rbd_dev, rbd_dev->dev_id);
5379 	return rbd_dev;
5380 
5381 fail_dev_id:
5382 	ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id);
5383 fail_rbd_dev:
5384 	rbd_dev_free(rbd_dev);
5385 	return NULL;
5386 }
5387 
5388 static void rbd_dev_destroy(struct rbd_device *rbd_dev)
5389 {
5390 	if (rbd_dev)
5391 		put_device(&rbd_dev->dev);
5392 }
5393 
5394 /*
5395  * Get the size and object order for an image snapshot, or if
5396  * snap_id is CEPH_NOSNAP, gets this information for the base
5397  * image.
5398  */
5399 static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
5400 				u8 *order, u64 *snap_size)
5401 {
5402 	__le64 snapid = cpu_to_le64(snap_id);
5403 	int ret;
5404 	struct {
5405 		u8 order;
5406 		__le64 size;
5407 	} __attribute__ ((packed)) size_buf = { 0 };
5408 
5409 	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
5410 				  &rbd_dev->header_oloc, "get_size",
5411 				  &snapid, sizeof(snapid),
5412 				  &size_buf, sizeof(size_buf));
5413 	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
5414 	if (ret < 0)
5415 		return ret;
5416 	if (ret < sizeof (size_buf))
5417 		return -ERANGE;
5418 
5419 	if (order) {
5420 		*order = size_buf.order;
5421 		dout("  order %u", (unsigned int)*order);
5422 	}
5423 	*snap_size = le64_to_cpu(size_buf.size);
5424 
5425 	dout("  snap_id 0x%016llx snap_size = %llu\n",
5426 		(unsigned long long)snap_id,
5427 		(unsigned long long)*snap_size);
5428 
5429 	return 0;
5430 }
5431 
5432 static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev)
5433 {
5434 	return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP,
5435 					&rbd_dev->header.obj_order,
5436 					&rbd_dev->header.image_size);
5437 }
5438 
5439 static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev)
5440 {
5441 	size_t size;
5442 	void *reply_buf;
5443 	int ret;
5444 	void *p;
5445 
5446 	/* Response will be an encoded string, which includes a length */
5447 	size = sizeof(__le32) + RBD_OBJ_PREFIX_LEN_MAX;
5448 	reply_buf = kzalloc(size, GFP_KERNEL);
5449 	if (!reply_buf)
5450 		return -ENOMEM;
5451 
5452 	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
5453 				  &rbd_dev->header_oloc, "get_object_prefix",
5454 				  NULL, 0, reply_buf, size);
5455 	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
5456 	if (ret < 0)
5457 		goto out;
5458 
5459 	p = reply_buf;
5460 	rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p,
5461 						p + ret, NULL, GFP_NOIO);
5462 	ret = 0;
5463 
5464 	if (IS_ERR(rbd_dev->header.object_prefix)) {
5465 		ret = PTR_ERR(rbd_dev->header.object_prefix);
5466 		rbd_dev->header.object_prefix = NULL;
5467 	} else {
5468 		dout("  object_prefix = %s\n", rbd_dev->header.object_prefix);
5469 	}
5470 out:
5471 	kfree(reply_buf);
5472 
5473 	return ret;
5474 }
5475 
5476 static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
5477 				     bool read_only, u64 *snap_features)
5478 {
5479 	struct {
5480 		__le64 snap_id;
5481 		u8 read_only;
5482 	} features_in;
5483 	struct {
5484 		__le64 features;
5485 		__le64 incompat;
5486 	} __attribute__ ((packed)) features_buf = { 0 };
5487 	u64 unsup;
5488 	int ret;
5489 
5490 	features_in.snap_id = cpu_to_le64(snap_id);
5491 	features_in.read_only = read_only;
5492 
5493 	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
5494 				  &rbd_dev->header_oloc, "get_features",
5495 				  &features_in, sizeof(features_in),
5496 				  &features_buf, sizeof(features_buf));
5497 	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
5498 	if (ret < 0)
5499 		return ret;
5500 	if (ret < sizeof (features_buf))
5501 		return -ERANGE;
5502 
5503 	unsup = le64_to_cpu(features_buf.incompat) & ~RBD_FEATURES_SUPPORTED;
5504 	if (unsup) {
5505 		rbd_warn(rbd_dev, "image uses unsupported features: 0x%llx",
5506 			 unsup);
5507 		return -ENXIO;
5508 	}
5509 
5510 	*snap_features = le64_to_cpu(features_buf.features);
5511 
5512 	dout("  snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
5513 		(unsigned long long)snap_id,
5514 		(unsigned long long)*snap_features,
5515 		(unsigned long long)le64_to_cpu(features_buf.incompat));
5516 
5517 	return 0;
5518 }
5519 
5520 static int rbd_dev_v2_features(struct rbd_device *rbd_dev)
5521 {
5522 	return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP,
5523 					 rbd_is_ro(rbd_dev),
5524 					 &rbd_dev->header.features);
5525 }
5526 
5527 /*
5528  * These are generic image flags, but since they are used only for
5529  * object map, store them in rbd_dev->object_map_flags.
5530  *
5531  * For the same reason, this function is called only on object map
5532  * (re)load and not on header refresh.
5533  */
5534 static int rbd_dev_v2_get_flags(struct rbd_device *rbd_dev)
5535 {
5536 	__le64 snapid = cpu_to_le64(rbd_dev->spec->snap_id);
5537 	__le64 flags;
5538 	int ret;
5539 
5540 	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
5541 				  &rbd_dev->header_oloc, "get_flags",
5542 				  &snapid, sizeof(snapid),
5543 				  &flags, sizeof(flags));
5544 	if (ret < 0)
5545 		return ret;
5546 	if (ret < sizeof(flags))
5547 		return -EBADMSG;
5548 
5549 	rbd_dev->object_map_flags = le64_to_cpu(flags);
5550 	return 0;
5551 }
5552 
5553 struct parent_image_info {
5554 	u64		pool_id;
5555 	const char	*pool_ns;
5556 	const char	*image_id;
5557 	u64		snap_id;
5558 
5559 	bool		has_overlap;
5560 	u64		overlap;
5561 };
5562 
5563 /*
5564  * The caller is responsible for @pii.
5565  */
5566 static int decode_parent_image_spec(void **p, void *end,
5567 				    struct parent_image_info *pii)
5568 {
5569 	u8 struct_v;
5570 	u32 struct_len;
5571 	int ret;
5572 
5573 	ret = ceph_start_decoding(p, end, 1, "ParentImageSpec",
5574 				  &struct_v, &struct_len);
5575 	if (ret)
5576 		return ret;
5577 
5578 	ceph_decode_64_safe(p, end, pii->pool_id, e_inval);
5579 	pii->pool_ns = ceph_extract_encoded_string(p, end, NULL, GFP_KERNEL);
5580 	if (IS_ERR(pii->pool_ns)) {
5581 		ret = PTR_ERR(pii->pool_ns);
5582 		pii->pool_ns = NULL;
5583 		return ret;
5584 	}
5585 	pii->image_id = ceph_extract_encoded_string(p, end, NULL, GFP_KERNEL);
5586 	if (IS_ERR(pii->image_id)) {
5587 		ret = PTR_ERR(pii->image_id);
5588 		pii->image_id = NULL;
5589 		return ret;
5590 	}
5591 	ceph_decode_64_safe(p, end, pii->snap_id, e_inval);
5592 	return 0;
5593 
5594 e_inval:
5595 	return -EINVAL;
5596 }
5597 
5598 static int __get_parent_info(struct rbd_device *rbd_dev,
5599 			     struct page *req_page,
5600 			     struct page *reply_page,
5601 			     struct parent_image_info *pii)
5602 {
5603 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
5604 	size_t reply_len = PAGE_SIZE;
5605 	void *p, *end;
5606 	int ret;
5607 
5608 	ret = ceph_osdc_call(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc,
5609 			     "rbd", "parent_get", CEPH_OSD_FLAG_READ,
5610 			     req_page, sizeof(u64), &reply_page, &reply_len);
5611 	if (ret)
5612 		return ret == -EOPNOTSUPP ? 1 : ret;
5613 
5614 	p = page_address(reply_page);
5615 	end = p + reply_len;
5616 	ret = decode_parent_image_spec(&p, end, pii);
5617 	if (ret)
5618 		return ret;
5619 
5620 	ret = ceph_osdc_call(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc,
5621 			     "rbd", "parent_overlap_get", CEPH_OSD_FLAG_READ,
5622 			     req_page, sizeof(u64), &reply_page, &reply_len);
5623 	if (ret)
5624 		return ret;
5625 
5626 	p = page_address(reply_page);
5627 	end = p + reply_len;
5628 	ceph_decode_8_safe(&p, end, pii->has_overlap, e_inval);
5629 	if (pii->has_overlap)
5630 		ceph_decode_64_safe(&p, end, pii->overlap, e_inval);
5631 
5632 	return 0;
5633 
5634 e_inval:
5635 	return -EINVAL;
5636 }
5637 
5638 /*
5639  * The caller is responsible for @pii.
5640  */
5641 static int __get_parent_info_legacy(struct rbd_device *rbd_dev,
5642 				    struct page *req_page,
5643 				    struct page *reply_page,
5644 				    struct parent_image_info *pii)
5645 {
5646 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
5647 	size_t reply_len = PAGE_SIZE;
5648 	void *p, *end;
5649 	int ret;
5650 
5651 	ret = ceph_osdc_call(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc,
5652 			     "rbd", "get_parent", CEPH_OSD_FLAG_READ,
5653 			     req_page, sizeof(u64), &reply_page, &reply_len);
5654 	if (ret)
5655 		return ret;
5656 
5657 	p = page_address(reply_page);
5658 	end = p + reply_len;
5659 	ceph_decode_64_safe(&p, end, pii->pool_id, e_inval);
5660 	pii->image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
5661 	if (IS_ERR(pii->image_id)) {
5662 		ret = PTR_ERR(pii->image_id);
5663 		pii->image_id = NULL;
5664 		return ret;
5665 	}
5666 	ceph_decode_64_safe(&p, end, pii->snap_id, e_inval);
5667 	pii->has_overlap = true;
5668 	ceph_decode_64_safe(&p, end, pii->overlap, e_inval);
5669 
5670 	return 0;
5671 
5672 e_inval:
5673 	return -EINVAL;
5674 }
5675 
5676 static int get_parent_info(struct rbd_device *rbd_dev,
5677 			   struct parent_image_info *pii)
5678 {
5679 	struct page *req_page, *reply_page;
5680 	void *p;
5681 	int ret;
5682 
5683 	req_page = alloc_page(GFP_KERNEL);
5684 	if (!req_page)
5685 		return -ENOMEM;
5686 
5687 	reply_page = alloc_page(GFP_KERNEL);
5688 	if (!reply_page) {
5689 		__free_page(req_page);
5690 		return -ENOMEM;
5691 	}
5692 
5693 	p = page_address(req_page);
5694 	ceph_encode_64(&p, rbd_dev->spec->snap_id);
5695 	ret = __get_parent_info(rbd_dev, req_page, reply_page, pii);
5696 	if (ret > 0)
5697 		ret = __get_parent_info_legacy(rbd_dev, req_page, reply_page,
5698 					       pii);
5699 
5700 	__free_page(req_page);
5701 	__free_page(reply_page);
5702 	return ret;
5703 }
5704 
5705 static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev)
5706 {
5707 	struct rbd_spec *parent_spec;
5708 	struct parent_image_info pii = { 0 };
5709 	int ret;
5710 
5711 	parent_spec = rbd_spec_alloc();
5712 	if (!parent_spec)
5713 		return -ENOMEM;
5714 
5715 	ret = get_parent_info(rbd_dev, &pii);
5716 	if (ret)
5717 		goto out_err;
5718 
5719 	dout("%s pool_id %llu pool_ns %s image_id %s snap_id %llu has_overlap %d overlap %llu\n",
5720 	     __func__, pii.pool_id, pii.pool_ns, pii.image_id, pii.snap_id,
5721 	     pii.has_overlap, pii.overlap);
5722 
5723 	if (pii.pool_id == CEPH_NOPOOL || !pii.has_overlap) {
5724 		/*
5725 		 * Either the parent never existed, or we have
5726 		 * record of it but the image got flattened so it no
5727 		 * longer has a parent.  When the parent of a
5728 		 * layered image disappears we immediately set the
5729 		 * overlap to 0.  The effect of this is that all new
5730 		 * requests will be treated as if the image had no
5731 		 * parent.
5732 		 *
5733 		 * If !pii.has_overlap, the parent image spec is not
5734 		 * applicable.  It's there to avoid duplication in each
5735 		 * snapshot record.
5736 		 */
5737 		if (rbd_dev->parent_overlap) {
5738 			rbd_dev->parent_overlap = 0;
5739 			rbd_dev_parent_put(rbd_dev);
5740 			pr_info("%s: clone image has been flattened\n",
5741 				rbd_dev->disk->disk_name);
5742 		}
5743 
5744 		goto out;	/* No parent?  No problem. */
5745 	}
5746 
5747 	/* The ceph file layout needs to fit pool id in 32 bits */
5748 
5749 	ret = -EIO;
5750 	if (pii.pool_id > (u64)U32_MAX) {
5751 		rbd_warn(NULL, "parent pool id too large (%llu > %u)",
5752 			(unsigned long long)pii.pool_id, U32_MAX);
5753 		goto out_err;
5754 	}
5755 
5756 	/*
5757 	 * The parent won't change (except when the clone is
5758 	 * flattened, already handled that).  So we only need to
5759 	 * record the parent spec we have not already done so.
5760 	 */
5761 	if (!rbd_dev->parent_spec) {
5762 		parent_spec->pool_id = pii.pool_id;
5763 		if (pii.pool_ns && *pii.pool_ns) {
5764 			parent_spec->pool_ns = pii.pool_ns;
5765 			pii.pool_ns = NULL;
5766 		}
5767 		parent_spec->image_id = pii.image_id;
5768 		pii.image_id = NULL;
5769 		parent_spec->snap_id = pii.snap_id;
5770 
5771 		rbd_dev->parent_spec = parent_spec;
5772 		parent_spec = NULL;	/* rbd_dev now owns this */
5773 	}
5774 
5775 	/*
5776 	 * We always update the parent overlap.  If it's zero we issue
5777 	 * a warning, as we will proceed as if there was no parent.
5778 	 */
5779 	if (!pii.overlap) {
5780 		if (parent_spec) {
5781 			/* refresh, careful to warn just once */
5782 			if (rbd_dev->parent_overlap)
5783 				rbd_warn(rbd_dev,
5784 				    "clone now standalone (overlap became 0)");
5785 		} else {
5786 			/* initial probe */
5787 			rbd_warn(rbd_dev, "clone is standalone (overlap 0)");
5788 		}
5789 	}
5790 	rbd_dev->parent_overlap = pii.overlap;
5791 
5792 out:
5793 	ret = 0;
5794 out_err:
5795 	kfree(pii.pool_ns);
5796 	kfree(pii.image_id);
5797 	rbd_spec_put(parent_spec);
5798 	return ret;
5799 }
5800 
5801 static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev)
5802 {
5803 	struct {
5804 		__le64 stripe_unit;
5805 		__le64 stripe_count;
5806 	} __attribute__ ((packed)) striping_info_buf = { 0 };
5807 	size_t size = sizeof (striping_info_buf);
5808 	void *p;
5809 	int ret;
5810 
5811 	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
5812 				&rbd_dev->header_oloc, "get_stripe_unit_count",
5813 				NULL, 0, &striping_info_buf, size);
5814 	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
5815 	if (ret < 0)
5816 		return ret;
5817 	if (ret < size)
5818 		return -ERANGE;
5819 
5820 	p = &striping_info_buf;
5821 	rbd_dev->header.stripe_unit = ceph_decode_64(&p);
5822 	rbd_dev->header.stripe_count = ceph_decode_64(&p);
5823 	return 0;
5824 }
5825 
5826 static int rbd_dev_v2_data_pool(struct rbd_device *rbd_dev)
5827 {
5828 	__le64 data_pool_id;
5829 	int ret;
5830 
5831 	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
5832 				  &rbd_dev->header_oloc, "get_data_pool",
5833 				  NULL, 0, &data_pool_id, sizeof(data_pool_id));
5834 	if (ret < 0)
5835 		return ret;
5836 	if (ret < sizeof(data_pool_id))
5837 		return -EBADMSG;
5838 
5839 	rbd_dev->header.data_pool_id = le64_to_cpu(data_pool_id);
5840 	WARN_ON(rbd_dev->header.data_pool_id == CEPH_NOPOOL);
5841 	return 0;
5842 }
5843 
5844 static char *rbd_dev_image_name(struct rbd_device *rbd_dev)
5845 {
5846 	CEPH_DEFINE_OID_ONSTACK(oid);
5847 	size_t image_id_size;
5848 	char *image_id;
5849 	void *p;
5850 	void *end;
5851 	size_t size;
5852 	void *reply_buf = NULL;
5853 	size_t len = 0;
5854 	char *image_name = NULL;
5855 	int ret;
5856 
5857 	rbd_assert(!rbd_dev->spec->image_name);
5858 
5859 	len = strlen(rbd_dev->spec->image_id);
5860 	image_id_size = sizeof (__le32) + len;
5861 	image_id = kmalloc(image_id_size, GFP_KERNEL);
5862 	if (!image_id)
5863 		return NULL;
5864 
5865 	p = image_id;
5866 	end = image_id + image_id_size;
5867 	ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len);
5868 
5869 	size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX;
5870 	reply_buf = kmalloc(size, GFP_KERNEL);
5871 	if (!reply_buf)
5872 		goto out;
5873 
5874 	ceph_oid_printf(&oid, "%s", RBD_DIRECTORY);
5875 	ret = rbd_obj_method_sync(rbd_dev, &oid, &rbd_dev->header_oloc,
5876 				  "dir_get_name", image_id, image_id_size,
5877 				  reply_buf, size);
5878 	if (ret < 0)
5879 		goto out;
5880 	p = reply_buf;
5881 	end = reply_buf + ret;
5882 
5883 	image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL);
5884 	if (IS_ERR(image_name))
5885 		image_name = NULL;
5886 	else
5887 		dout("%s: name is %s len is %zd\n", __func__, image_name, len);
5888 out:
5889 	kfree(reply_buf);
5890 	kfree(image_id);
5891 
5892 	return image_name;
5893 }
5894 
5895 static u64 rbd_v1_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
5896 {
5897 	struct ceph_snap_context *snapc = rbd_dev->header.snapc;
5898 	const char *snap_name;
5899 	u32 which = 0;
5900 
5901 	/* Skip over names until we find the one we are looking for */
5902 
5903 	snap_name = rbd_dev->header.snap_names;
5904 	while (which < snapc->num_snaps) {
5905 		if (!strcmp(name, snap_name))
5906 			return snapc->snaps[which];
5907 		snap_name += strlen(snap_name) + 1;
5908 		which++;
5909 	}
5910 	return CEPH_NOSNAP;
5911 }
5912 
5913 static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
5914 {
5915 	struct ceph_snap_context *snapc = rbd_dev->header.snapc;
5916 	u32 which;
5917 	bool found = false;
5918 	u64 snap_id;
5919 
5920 	for (which = 0; !found && which < snapc->num_snaps; which++) {
5921 		const char *snap_name;
5922 
5923 		snap_id = snapc->snaps[which];
5924 		snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id);
5925 		if (IS_ERR(snap_name)) {
5926 			/* ignore no-longer existing snapshots */
5927 			if (PTR_ERR(snap_name) == -ENOENT)
5928 				continue;
5929 			else
5930 				break;
5931 		}
5932 		found = !strcmp(name, snap_name);
5933 		kfree(snap_name);
5934 	}
5935 	return found ? snap_id : CEPH_NOSNAP;
5936 }
5937 
5938 /*
5939  * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if
5940  * no snapshot by that name is found, or if an error occurs.
5941  */
5942 static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
5943 {
5944 	if (rbd_dev->image_format == 1)
5945 		return rbd_v1_snap_id_by_name(rbd_dev, name);
5946 
5947 	return rbd_v2_snap_id_by_name(rbd_dev, name);
5948 }
5949 
5950 /*
5951  * An image being mapped will have everything but the snap id.
5952  */
5953 static int rbd_spec_fill_snap_id(struct rbd_device *rbd_dev)
5954 {
5955 	struct rbd_spec *spec = rbd_dev->spec;
5956 
5957 	rbd_assert(spec->pool_id != CEPH_NOPOOL && spec->pool_name);
5958 	rbd_assert(spec->image_id && spec->image_name);
5959 	rbd_assert(spec->snap_name);
5960 
5961 	if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) {
5962 		u64 snap_id;
5963 
5964 		snap_id = rbd_snap_id_by_name(rbd_dev, spec->snap_name);
5965 		if (snap_id == CEPH_NOSNAP)
5966 			return -ENOENT;
5967 
5968 		spec->snap_id = snap_id;
5969 	} else {
5970 		spec->snap_id = CEPH_NOSNAP;
5971 	}
5972 
5973 	return 0;
5974 }
5975 
5976 /*
5977  * A parent image will have all ids but none of the names.
5978  *
5979  * All names in an rbd spec are dynamically allocated.  It's OK if we
5980  * can't figure out the name for an image id.
5981  */
5982 static int rbd_spec_fill_names(struct rbd_device *rbd_dev)
5983 {
5984 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
5985 	struct rbd_spec *spec = rbd_dev->spec;
5986 	const char *pool_name;
5987 	const char *image_name;
5988 	const char *snap_name;
5989 	int ret;
5990 
5991 	rbd_assert(spec->pool_id != CEPH_NOPOOL);
5992 	rbd_assert(spec->image_id);
5993 	rbd_assert(spec->snap_id != CEPH_NOSNAP);
5994 
5995 	/* Get the pool name; we have to make our own copy of this */
5996 
5997 	pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id);
5998 	if (!pool_name) {
5999 		rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id);
6000 		return -EIO;
6001 	}
6002 	pool_name = kstrdup(pool_name, GFP_KERNEL);
6003 	if (!pool_name)
6004 		return -ENOMEM;
6005 
6006 	/* Fetch the image name; tolerate failure here */
6007 
6008 	image_name = rbd_dev_image_name(rbd_dev);
6009 	if (!image_name)
6010 		rbd_warn(rbd_dev, "unable to get image name");
6011 
6012 	/* Fetch the snapshot name */
6013 
6014 	snap_name = rbd_snap_name(rbd_dev, spec->snap_id);
6015 	if (IS_ERR(snap_name)) {
6016 		ret = PTR_ERR(snap_name);
6017 		goto out_err;
6018 	}
6019 
6020 	spec->pool_name = pool_name;
6021 	spec->image_name = image_name;
6022 	spec->snap_name = snap_name;
6023 
6024 	return 0;
6025 
6026 out_err:
6027 	kfree(image_name);
6028 	kfree(pool_name);
6029 	return ret;
6030 }
6031 
6032 static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev)
6033 {
6034 	size_t size;
6035 	int ret;
6036 	void *reply_buf;
6037 	void *p;
6038 	void *end;
6039 	u64 seq;
6040 	u32 snap_count;
6041 	struct ceph_snap_context *snapc;
6042 	u32 i;
6043 
6044 	/*
6045 	 * We'll need room for the seq value (maximum snapshot id),
6046 	 * snapshot count, and array of that many snapshot ids.
6047 	 * For now we have a fixed upper limit on the number we're
6048 	 * prepared to receive.
6049 	 */
6050 	size = sizeof (__le64) + sizeof (__le32) +
6051 			RBD_MAX_SNAP_COUNT * sizeof (__le64);
6052 	reply_buf = kzalloc(size, GFP_KERNEL);
6053 	if (!reply_buf)
6054 		return -ENOMEM;
6055 
6056 	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
6057 				  &rbd_dev->header_oloc, "get_snapcontext",
6058 				  NULL, 0, reply_buf, size);
6059 	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
6060 	if (ret < 0)
6061 		goto out;
6062 
6063 	p = reply_buf;
6064 	end = reply_buf + ret;
6065 	ret = -ERANGE;
6066 	ceph_decode_64_safe(&p, end, seq, out);
6067 	ceph_decode_32_safe(&p, end, snap_count, out);
6068 
6069 	/*
6070 	 * Make sure the reported number of snapshot ids wouldn't go
6071 	 * beyond the end of our buffer.  But before checking that,
6072 	 * make sure the computed size of the snapshot context we
6073 	 * allocate is representable in a size_t.
6074 	 */
6075 	if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context))
6076 				 / sizeof (u64)) {
6077 		ret = -EINVAL;
6078 		goto out;
6079 	}
6080 	if (!ceph_has_room(&p, end, snap_count * sizeof (__le64)))
6081 		goto out;
6082 	ret = 0;
6083 
6084 	snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
6085 	if (!snapc) {
6086 		ret = -ENOMEM;
6087 		goto out;
6088 	}
6089 	snapc->seq = seq;
6090 	for (i = 0; i < snap_count; i++)
6091 		snapc->snaps[i] = ceph_decode_64(&p);
6092 
6093 	ceph_put_snap_context(rbd_dev->header.snapc);
6094 	rbd_dev->header.snapc = snapc;
6095 
6096 	dout("  snap context seq = %llu, snap_count = %u\n",
6097 		(unsigned long long)seq, (unsigned int)snap_count);
6098 out:
6099 	kfree(reply_buf);
6100 
6101 	return ret;
6102 }
6103 
6104 static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
6105 					u64 snap_id)
6106 {
6107 	size_t size;
6108 	void *reply_buf;
6109 	__le64 snapid;
6110 	int ret;
6111 	void *p;
6112 	void *end;
6113 	char *snap_name;
6114 
6115 	size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN;
6116 	reply_buf = kmalloc(size, GFP_KERNEL);
6117 	if (!reply_buf)
6118 		return ERR_PTR(-ENOMEM);
6119 
6120 	snapid = cpu_to_le64(snap_id);
6121 	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
6122 				  &rbd_dev->header_oloc, "get_snapshot_name",
6123 				  &snapid, sizeof(snapid), reply_buf, size);
6124 	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
6125 	if (ret < 0) {
6126 		snap_name = ERR_PTR(ret);
6127 		goto out;
6128 	}
6129 
6130 	p = reply_buf;
6131 	end = reply_buf + ret;
6132 	snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
6133 	if (IS_ERR(snap_name))
6134 		goto out;
6135 
6136 	dout("  snap_id 0x%016llx snap_name = %s\n",
6137 		(unsigned long long)snap_id, snap_name);
6138 out:
6139 	kfree(reply_buf);
6140 
6141 	return snap_name;
6142 }
6143 
6144 static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev)
6145 {
6146 	bool first_time = rbd_dev->header.object_prefix == NULL;
6147 	int ret;
6148 
6149 	ret = rbd_dev_v2_image_size(rbd_dev);
6150 	if (ret)
6151 		return ret;
6152 
6153 	if (first_time) {
6154 		ret = rbd_dev_v2_header_onetime(rbd_dev);
6155 		if (ret)
6156 			return ret;
6157 	}
6158 
6159 	ret = rbd_dev_v2_snap_context(rbd_dev);
6160 	if (ret && first_time) {
6161 		kfree(rbd_dev->header.object_prefix);
6162 		rbd_dev->header.object_prefix = NULL;
6163 	}
6164 
6165 	return ret;
6166 }
6167 
6168 static int rbd_dev_header_info(struct rbd_device *rbd_dev)
6169 {
6170 	rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
6171 
6172 	if (rbd_dev->image_format == 1)
6173 		return rbd_dev_v1_header_info(rbd_dev);
6174 
6175 	return rbd_dev_v2_header_info(rbd_dev);
6176 }
6177 
6178 /*
6179  * Skips over white space at *buf, and updates *buf to point to the
6180  * first found non-space character (if any). Returns the length of
6181  * the token (string of non-white space characters) found.  Note
6182  * that *buf must be terminated with '\0'.
6183  */
6184 static inline size_t next_token(const char **buf)
6185 {
6186         /*
6187         * These are the characters that produce nonzero for
6188         * isspace() in the "C" and "POSIX" locales.
6189         */
6190 	static const char spaces[] = " \f\n\r\t\v";
6191 
6192         *buf += strspn(*buf, spaces);	/* Find start of token */
6193 
6194 	return strcspn(*buf, spaces);   /* Return token length */
6195 }
6196 
6197 /*
6198  * Finds the next token in *buf, dynamically allocates a buffer big
6199  * enough to hold a copy of it, and copies the token into the new
6200  * buffer.  The copy is guaranteed to be terminated with '\0'.  Note
6201  * that a duplicate buffer is created even for a zero-length token.
6202  *
6203  * Returns a pointer to the newly-allocated duplicate, or a null
6204  * pointer if memory for the duplicate was not available.  If
6205  * the lenp argument is a non-null pointer, the length of the token
6206  * (not including the '\0') is returned in *lenp.
6207  *
6208  * If successful, the *buf pointer will be updated to point beyond
6209  * the end of the found token.
6210  *
6211  * Note: uses GFP_KERNEL for allocation.
6212  */
6213 static inline char *dup_token(const char **buf, size_t *lenp)
6214 {
6215 	char *dup;
6216 	size_t len;
6217 
6218 	len = next_token(buf);
6219 	dup = kmemdup(*buf, len + 1, GFP_KERNEL);
6220 	if (!dup)
6221 		return NULL;
6222 	*(dup + len) = '\0';
6223 	*buf += len;
6224 
6225 	if (lenp)
6226 		*lenp = len;
6227 
6228 	return dup;
6229 }
6230 
6231 static int rbd_parse_param(struct fs_parameter *param,
6232 			    struct rbd_parse_opts_ctx *pctx)
6233 {
6234 	struct rbd_options *opt = pctx->opts;
6235 	struct fs_parse_result result;
6236 	struct p_log log = {.prefix = "rbd"};
6237 	int token, ret;
6238 
6239 	ret = ceph_parse_param(param, pctx->copts, NULL);
6240 	if (ret != -ENOPARAM)
6241 		return ret;
6242 
6243 	token = __fs_parse(&log, rbd_parameters, param, &result);
6244 	dout("%s fs_parse '%s' token %d\n", __func__, param->key, token);
6245 	if (token < 0) {
6246 		if (token == -ENOPARAM)
6247 			return inval_plog(&log, "Unknown parameter '%s'",
6248 					  param->key);
6249 		return token;
6250 	}
6251 
6252 	switch (token) {
6253 	case Opt_queue_depth:
6254 		if (result.uint_32 < 1)
6255 			goto out_of_range;
6256 		opt->queue_depth = result.uint_32;
6257 		break;
6258 	case Opt_alloc_size:
6259 		if (result.uint_32 < SECTOR_SIZE)
6260 			goto out_of_range;
6261 		if (!is_power_of_2(result.uint_32))
6262 			return inval_plog(&log, "alloc_size must be a power of 2");
6263 		opt->alloc_size = result.uint_32;
6264 		break;
6265 	case Opt_lock_timeout:
6266 		/* 0 is "wait forever" (i.e. infinite timeout) */
6267 		if (result.uint_32 > INT_MAX / 1000)
6268 			goto out_of_range;
6269 		opt->lock_timeout = msecs_to_jiffies(result.uint_32 * 1000);
6270 		break;
6271 	case Opt_pool_ns:
6272 		kfree(pctx->spec->pool_ns);
6273 		pctx->spec->pool_ns = param->string;
6274 		param->string = NULL;
6275 		break;
6276 	case Opt_compression_hint:
6277 		switch (result.uint_32) {
6278 		case Opt_compression_hint_none:
6279 			opt->alloc_hint_flags &=
6280 			    ~(CEPH_OSD_ALLOC_HINT_FLAG_COMPRESSIBLE |
6281 			      CEPH_OSD_ALLOC_HINT_FLAG_INCOMPRESSIBLE);
6282 			break;
6283 		case Opt_compression_hint_compressible:
6284 			opt->alloc_hint_flags |=
6285 			    CEPH_OSD_ALLOC_HINT_FLAG_COMPRESSIBLE;
6286 			opt->alloc_hint_flags &=
6287 			    ~CEPH_OSD_ALLOC_HINT_FLAG_INCOMPRESSIBLE;
6288 			break;
6289 		case Opt_compression_hint_incompressible:
6290 			opt->alloc_hint_flags |=
6291 			    CEPH_OSD_ALLOC_HINT_FLAG_INCOMPRESSIBLE;
6292 			opt->alloc_hint_flags &=
6293 			    ~CEPH_OSD_ALLOC_HINT_FLAG_COMPRESSIBLE;
6294 			break;
6295 		default:
6296 			BUG();
6297 		}
6298 		break;
6299 	case Opt_read_only:
6300 		opt->read_only = true;
6301 		break;
6302 	case Opt_read_write:
6303 		opt->read_only = false;
6304 		break;
6305 	case Opt_lock_on_read:
6306 		opt->lock_on_read = true;
6307 		break;
6308 	case Opt_exclusive:
6309 		opt->exclusive = true;
6310 		break;
6311 	case Opt_notrim:
6312 		opt->trim = false;
6313 		break;
6314 	default:
6315 		BUG();
6316 	}
6317 
6318 	return 0;
6319 
6320 out_of_range:
6321 	return inval_plog(&log, "%s out of range", param->key);
6322 }
6323 
6324 /*
6325  * This duplicates most of generic_parse_monolithic(), untying it from
6326  * fs_context and skipping standard superblock and security options.
6327  */
6328 static int rbd_parse_options(char *options, struct rbd_parse_opts_ctx *pctx)
6329 {
6330 	char *key;
6331 	int ret = 0;
6332 
6333 	dout("%s '%s'\n", __func__, options);
6334 	while ((key = strsep(&options, ",")) != NULL) {
6335 		if (*key) {
6336 			struct fs_parameter param = {
6337 				.key	= key,
6338 				.type	= fs_value_is_flag,
6339 			};
6340 			char *value = strchr(key, '=');
6341 			size_t v_len = 0;
6342 
6343 			if (value) {
6344 				if (value == key)
6345 					continue;
6346 				*value++ = 0;
6347 				v_len = strlen(value);
6348 				param.string = kmemdup_nul(value, v_len,
6349 							   GFP_KERNEL);
6350 				if (!param.string)
6351 					return -ENOMEM;
6352 				param.type = fs_value_is_string;
6353 			}
6354 			param.size = v_len;
6355 
6356 			ret = rbd_parse_param(&param, pctx);
6357 			kfree(param.string);
6358 			if (ret)
6359 				break;
6360 		}
6361 	}
6362 
6363 	return ret;
6364 }
6365 
6366 /*
6367  * Parse the options provided for an "rbd add" (i.e., rbd image
6368  * mapping) request.  These arrive via a write to /sys/bus/rbd/add,
6369  * and the data written is passed here via a NUL-terminated buffer.
6370  * Returns 0 if successful or an error code otherwise.
6371  *
6372  * The information extracted from these options is recorded in
6373  * the other parameters which return dynamically-allocated
6374  * structures:
6375  *  ceph_opts
6376  *      The address of a pointer that will refer to a ceph options
6377  *      structure.  Caller must release the returned pointer using
6378  *      ceph_destroy_options() when it is no longer needed.
6379  *  rbd_opts
6380  *	Address of an rbd options pointer.  Fully initialized by
6381  *	this function; caller must release with kfree().
6382  *  spec
6383  *	Address of an rbd image specification pointer.  Fully
6384  *	initialized by this function based on parsed options.
6385  *	Caller must release with rbd_spec_put().
6386  *
6387  * The options passed take this form:
6388  *  <mon_addrs> <options> <pool_name> <image_name> [<snap_id>]
6389  * where:
6390  *  <mon_addrs>
6391  *      A comma-separated list of one or more monitor addresses.
6392  *      A monitor address is an ip address, optionally followed
6393  *      by a port number (separated by a colon).
6394  *        I.e.:  ip1[:port1][,ip2[:port2]...]
6395  *  <options>
6396  *      A comma-separated list of ceph and/or rbd options.
6397  *  <pool_name>
6398  *      The name of the rados pool containing the rbd image.
6399  *  <image_name>
6400  *      The name of the image in that pool to map.
6401  *  <snap_id>
6402  *      An optional snapshot id.  If provided, the mapping will
6403  *      present data from the image at the time that snapshot was
6404  *      created.  The image head is used if no snapshot id is
6405  *      provided.  Snapshot mappings are always read-only.
6406  */
6407 static int rbd_add_parse_args(const char *buf,
6408 				struct ceph_options **ceph_opts,
6409 				struct rbd_options **opts,
6410 				struct rbd_spec **rbd_spec)
6411 {
6412 	size_t len;
6413 	char *options;
6414 	const char *mon_addrs;
6415 	char *snap_name;
6416 	size_t mon_addrs_size;
6417 	struct rbd_parse_opts_ctx pctx = { 0 };
6418 	int ret;
6419 
6420 	/* The first four tokens are required */
6421 
6422 	len = next_token(&buf);
6423 	if (!len) {
6424 		rbd_warn(NULL, "no monitor address(es) provided");
6425 		return -EINVAL;
6426 	}
6427 	mon_addrs = buf;
6428 	mon_addrs_size = len;
6429 	buf += len;
6430 
6431 	ret = -EINVAL;
6432 	options = dup_token(&buf, NULL);
6433 	if (!options)
6434 		return -ENOMEM;
6435 	if (!*options) {
6436 		rbd_warn(NULL, "no options provided");
6437 		goto out_err;
6438 	}
6439 
6440 	pctx.spec = rbd_spec_alloc();
6441 	if (!pctx.spec)
6442 		goto out_mem;
6443 
6444 	pctx.spec->pool_name = dup_token(&buf, NULL);
6445 	if (!pctx.spec->pool_name)
6446 		goto out_mem;
6447 	if (!*pctx.spec->pool_name) {
6448 		rbd_warn(NULL, "no pool name provided");
6449 		goto out_err;
6450 	}
6451 
6452 	pctx.spec->image_name = dup_token(&buf, NULL);
6453 	if (!pctx.spec->image_name)
6454 		goto out_mem;
6455 	if (!*pctx.spec->image_name) {
6456 		rbd_warn(NULL, "no image name provided");
6457 		goto out_err;
6458 	}
6459 
6460 	/*
6461 	 * Snapshot name is optional; default is to use "-"
6462 	 * (indicating the head/no snapshot).
6463 	 */
6464 	len = next_token(&buf);
6465 	if (!len) {
6466 		buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */
6467 		len = sizeof (RBD_SNAP_HEAD_NAME) - 1;
6468 	} else if (len > RBD_MAX_SNAP_NAME_LEN) {
6469 		ret = -ENAMETOOLONG;
6470 		goto out_err;
6471 	}
6472 	snap_name = kmemdup(buf, len + 1, GFP_KERNEL);
6473 	if (!snap_name)
6474 		goto out_mem;
6475 	*(snap_name + len) = '\0';
6476 	pctx.spec->snap_name = snap_name;
6477 
6478 	pctx.copts = ceph_alloc_options();
6479 	if (!pctx.copts)
6480 		goto out_mem;
6481 
6482 	/* Initialize all rbd options to the defaults */
6483 
6484 	pctx.opts = kzalloc(sizeof(*pctx.opts), GFP_KERNEL);
6485 	if (!pctx.opts)
6486 		goto out_mem;
6487 
6488 	pctx.opts->read_only = RBD_READ_ONLY_DEFAULT;
6489 	pctx.opts->queue_depth = RBD_QUEUE_DEPTH_DEFAULT;
6490 	pctx.opts->alloc_size = RBD_ALLOC_SIZE_DEFAULT;
6491 	pctx.opts->lock_timeout = RBD_LOCK_TIMEOUT_DEFAULT;
6492 	pctx.opts->lock_on_read = RBD_LOCK_ON_READ_DEFAULT;
6493 	pctx.opts->exclusive = RBD_EXCLUSIVE_DEFAULT;
6494 	pctx.opts->trim = RBD_TRIM_DEFAULT;
6495 
6496 	ret = ceph_parse_mon_ips(mon_addrs, mon_addrs_size, pctx.copts, NULL,
6497 				 ',');
6498 	if (ret)
6499 		goto out_err;
6500 
6501 	ret = rbd_parse_options(options, &pctx);
6502 	if (ret)
6503 		goto out_err;
6504 
6505 	*ceph_opts = pctx.copts;
6506 	*opts = pctx.opts;
6507 	*rbd_spec = pctx.spec;
6508 	kfree(options);
6509 	return 0;
6510 
6511 out_mem:
6512 	ret = -ENOMEM;
6513 out_err:
6514 	kfree(pctx.opts);
6515 	ceph_destroy_options(pctx.copts);
6516 	rbd_spec_put(pctx.spec);
6517 	kfree(options);
6518 	return ret;
6519 }
6520 
6521 static void rbd_dev_image_unlock(struct rbd_device *rbd_dev)
6522 {
6523 	down_write(&rbd_dev->lock_rwsem);
6524 	if (__rbd_is_lock_owner(rbd_dev))
6525 		__rbd_release_lock(rbd_dev);
6526 	up_write(&rbd_dev->lock_rwsem);
6527 }
6528 
6529 /*
6530  * If the wait is interrupted, an error is returned even if the lock
6531  * was successfully acquired.  rbd_dev_image_unlock() will release it
6532  * if needed.
6533  */
6534 static int rbd_add_acquire_lock(struct rbd_device *rbd_dev)
6535 {
6536 	long ret;
6537 
6538 	if (!(rbd_dev->header.features & RBD_FEATURE_EXCLUSIVE_LOCK)) {
6539 		if (!rbd_dev->opts->exclusive && !rbd_dev->opts->lock_on_read)
6540 			return 0;
6541 
6542 		rbd_warn(rbd_dev, "exclusive-lock feature is not enabled");
6543 		return -EINVAL;
6544 	}
6545 
6546 	if (rbd_is_ro(rbd_dev))
6547 		return 0;
6548 
6549 	rbd_assert(!rbd_is_lock_owner(rbd_dev));
6550 	queue_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork, 0);
6551 	ret = wait_for_completion_killable_timeout(&rbd_dev->acquire_wait,
6552 			    ceph_timeout_jiffies(rbd_dev->opts->lock_timeout));
6553 	if (ret > 0) {
6554 		ret = rbd_dev->acquire_err;
6555 	} else {
6556 		cancel_delayed_work_sync(&rbd_dev->lock_dwork);
6557 		if (!ret)
6558 			ret = -ETIMEDOUT;
6559 	}
6560 
6561 	if (ret) {
6562 		rbd_warn(rbd_dev, "failed to acquire exclusive lock: %ld", ret);
6563 		return ret;
6564 	}
6565 
6566 	/*
6567 	 * The lock may have been released by now, unless automatic lock
6568 	 * transitions are disabled.
6569 	 */
6570 	rbd_assert(!rbd_dev->opts->exclusive || rbd_is_lock_owner(rbd_dev));
6571 	return 0;
6572 }
6573 
6574 /*
6575  * An rbd format 2 image has a unique identifier, distinct from the
6576  * name given to it by the user.  Internally, that identifier is
6577  * what's used to specify the names of objects related to the image.
6578  *
6579  * A special "rbd id" object is used to map an rbd image name to its
6580  * id.  If that object doesn't exist, then there is no v2 rbd image
6581  * with the supplied name.
6582  *
6583  * This function will record the given rbd_dev's image_id field if
6584  * it can be determined, and in that case will return 0.  If any
6585  * errors occur a negative errno will be returned and the rbd_dev's
6586  * image_id field will be unchanged (and should be NULL).
6587  */
6588 static int rbd_dev_image_id(struct rbd_device *rbd_dev)
6589 {
6590 	int ret;
6591 	size_t size;
6592 	CEPH_DEFINE_OID_ONSTACK(oid);
6593 	void *response;
6594 	char *image_id;
6595 
6596 	/*
6597 	 * When probing a parent image, the image id is already
6598 	 * known (and the image name likely is not).  There's no
6599 	 * need to fetch the image id again in this case.  We
6600 	 * do still need to set the image format though.
6601 	 */
6602 	if (rbd_dev->spec->image_id) {
6603 		rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1;
6604 
6605 		return 0;
6606 	}
6607 
6608 	/*
6609 	 * First, see if the format 2 image id file exists, and if
6610 	 * so, get the image's persistent id from it.
6611 	 */
6612 	ret = ceph_oid_aprintf(&oid, GFP_KERNEL, "%s%s", RBD_ID_PREFIX,
6613 			       rbd_dev->spec->image_name);
6614 	if (ret)
6615 		return ret;
6616 
6617 	dout("rbd id object name is %s\n", oid.name);
6618 
6619 	/* Response will be an encoded string, which includes a length */
6620 	size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX;
6621 	response = kzalloc(size, GFP_NOIO);
6622 	if (!response) {
6623 		ret = -ENOMEM;
6624 		goto out;
6625 	}
6626 
6627 	/* If it doesn't exist we'll assume it's a format 1 image */
6628 
6629 	ret = rbd_obj_method_sync(rbd_dev, &oid, &rbd_dev->header_oloc,
6630 				  "get_id", NULL, 0,
6631 				  response, size);
6632 	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
6633 	if (ret == -ENOENT) {
6634 		image_id = kstrdup("", GFP_KERNEL);
6635 		ret = image_id ? 0 : -ENOMEM;
6636 		if (!ret)
6637 			rbd_dev->image_format = 1;
6638 	} else if (ret >= 0) {
6639 		void *p = response;
6640 
6641 		image_id = ceph_extract_encoded_string(&p, p + ret,
6642 						NULL, GFP_NOIO);
6643 		ret = PTR_ERR_OR_ZERO(image_id);
6644 		if (!ret)
6645 			rbd_dev->image_format = 2;
6646 	}
6647 
6648 	if (!ret) {
6649 		rbd_dev->spec->image_id = image_id;
6650 		dout("image_id is %s\n", image_id);
6651 	}
6652 out:
6653 	kfree(response);
6654 	ceph_oid_destroy(&oid);
6655 	return ret;
6656 }
6657 
6658 /*
6659  * Undo whatever state changes are made by v1 or v2 header info
6660  * call.
6661  */
6662 static void rbd_dev_unprobe(struct rbd_device *rbd_dev)
6663 {
6664 	struct rbd_image_header	*header;
6665 
6666 	rbd_dev_parent_put(rbd_dev);
6667 	rbd_object_map_free(rbd_dev);
6668 	rbd_dev_mapping_clear(rbd_dev);
6669 
6670 	/* Free dynamic fields from the header, then zero it out */
6671 
6672 	header = &rbd_dev->header;
6673 	ceph_put_snap_context(header->snapc);
6674 	kfree(header->snap_sizes);
6675 	kfree(header->snap_names);
6676 	kfree(header->object_prefix);
6677 	memset(header, 0, sizeof (*header));
6678 }
6679 
6680 static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev)
6681 {
6682 	int ret;
6683 
6684 	ret = rbd_dev_v2_object_prefix(rbd_dev);
6685 	if (ret)
6686 		goto out_err;
6687 
6688 	/*
6689 	 * Get the and check features for the image.  Currently the
6690 	 * features are assumed to never change.
6691 	 */
6692 	ret = rbd_dev_v2_features(rbd_dev);
6693 	if (ret)
6694 		goto out_err;
6695 
6696 	/* If the image supports fancy striping, get its parameters */
6697 
6698 	if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) {
6699 		ret = rbd_dev_v2_striping_info(rbd_dev);
6700 		if (ret < 0)
6701 			goto out_err;
6702 	}
6703 
6704 	if (rbd_dev->header.features & RBD_FEATURE_DATA_POOL) {
6705 		ret = rbd_dev_v2_data_pool(rbd_dev);
6706 		if (ret)
6707 			goto out_err;
6708 	}
6709 
6710 	rbd_init_layout(rbd_dev);
6711 	return 0;
6712 
6713 out_err:
6714 	rbd_dev->header.features = 0;
6715 	kfree(rbd_dev->header.object_prefix);
6716 	rbd_dev->header.object_prefix = NULL;
6717 	return ret;
6718 }
6719 
6720 /*
6721  * @depth is rbd_dev_image_probe() -> rbd_dev_probe_parent() ->
6722  * rbd_dev_image_probe() recursion depth, which means it's also the
6723  * length of the already discovered part of the parent chain.
6724  */
6725 static int rbd_dev_probe_parent(struct rbd_device *rbd_dev, int depth)
6726 {
6727 	struct rbd_device *parent = NULL;
6728 	int ret;
6729 
6730 	if (!rbd_dev->parent_spec)
6731 		return 0;
6732 
6733 	if (++depth > RBD_MAX_PARENT_CHAIN_LEN) {
6734 		pr_info("parent chain is too long (%d)\n", depth);
6735 		ret = -EINVAL;
6736 		goto out_err;
6737 	}
6738 
6739 	parent = __rbd_dev_create(rbd_dev->rbd_client, rbd_dev->parent_spec);
6740 	if (!parent) {
6741 		ret = -ENOMEM;
6742 		goto out_err;
6743 	}
6744 
6745 	/*
6746 	 * Images related by parent/child relationships always share
6747 	 * rbd_client and spec/parent_spec, so bump their refcounts.
6748 	 */
6749 	__rbd_get_client(rbd_dev->rbd_client);
6750 	rbd_spec_get(rbd_dev->parent_spec);
6751 
6752 	__set_bit(RBD_DEV_FLAG_READONLY, &parent->flags);
6753 
6754 	ret = rbd_dev_image_probe(parent, depth);
6755 	if (ret < 0)
6756 		goto out_err;
6757 
6758 	rbd_dev->parent = parent;
6759 	atomic_set(&rbd_dev->parent_ref, 1);
6760 	return 0;
6761 
6762 out_err:
6763 	rbd_dev_unparent(rbd_dev);
6764 	rbd_dev_destroy(parent);
6765 	return ret;
6766 }
6767 
6768 static void rbd_dev_device_release(struct rbd_device *rbd_dev)
6769 {
6770 	clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
6771 	rbd_free_disk(rbd_dev);
6772 	if (!single_major)
6773 		unregister_blkdev(rbd_dev->major, rbd_dev->name);
6774 }
6775 
6776 /*
6777  * rbd_dev->header_rwsem must be locked for write and will be unlocked
6778  * upon return.
6779  */
6780 static int rbd_dev_device_setup(struct rbd_device *rbd_dev)
6781 {
6782 	int ret;
6783 
6784 	/* Record our major and minor device numbers. */
6785 
6786 	if (!single_major) {
6787 		ret = register_blkdev(0, rbd_dev->name);
6788 		if (ret < 0)
6789 			goto err_out_unlock;
6790 
6791 		rbd_dev->major = ret;
6792 		rbd_dev->minor = 0;
6793 	} else {
6794 		rbd_dev->major = rbd_major;
6795 		rbd_dev->minor = rbd_dev_id_to_minor(rbd_dev->dev_id);
6796 	}
6797 
6798 	/* Set up the blkdev mapping. */
6799 
6800 	ret = rbd_init_disk(rbd_dev);
6801 	if (ret)
6802 		goto err_out_blkdev;
6803 
6804 	set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE);
6805 	set_disk_ro(rbd_dev->disk, rbd_is_ro(rbd_dev));
6806 
6807 	ret = dev_set_name(&rbd_dev->dev, "%d", rbd_dev->dev_id);
6808 	if (ret)
6809 		goto err_out_disk;
6810 
6811 	set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
6812 	up_write(&rbd_dev->header_rwsem);
6813 	return 0;
6814 
6815 err_out_disk:
6816 	rbd_free_disk(rbd_dev);
6817 err_out_blkdev:
6818 	if (!single_major)
6819 		unregister_blkdev(rbd_dev->major, rbd_dev->name);
6820 err_out_unlock:
6821 	up_write(&rbd_dev->header_rwsem);
6822 	return ret;
6823 }
6824 
6825 static int rbd_dev_header_name(struct rbd_device *rbd_dev)
6826 {
6827 	struct rbd_spec *spec = rbd_dev->spec;
6828 	int ret;
6829 
6830 	/* Record the header object name for this rbd image. */
6831 
6832 	rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
6833 	if (rbd_dev->image_format == 1)
6834 		ret = ceph_oid_aprintf(&rbd_dev->header_oid, GFP_KERNEL, "%s%s",
6835 				       spec->image_name, RBD_SUFFIX);
6836 	else
6837 		ret = ceph_oid_aprintf(&rbd_dev->header_oid, GFP_KERNEL, "%s%s",
6838 				       RBD_HEADER_PREFIX, spec->image_id);
6839 
6840 	return ret;
6841 }
6842 
6843 static void rbd_print_dne(struct rbd_device *rbd_dev, bool is_snap)
6844 {
6845 	if (!is_snap) {
6846 		pr_info("image %s/%s%s%s does not exist\n",
6847 			rbd_dev->spec->pool_name,
6848 			rbd_dev->spec->pool_ns ?: "",
6849 			rbd_dev->spec->pool_ns ? "/" : "",
6850 			rbd_dev->spec->image_name);
6851 	} else {
6852 		pr_info("snap %s/%s%s%s@%s does not exist\n",
6853 			rbd_dev->spec->pool_name,
6854 			rbd_dev->spec->pool_ns ?: "",
6855 			rbd_dev->spec->pool_ns ? "/" : "",
6856 			rbd_dev->spec->image_name,
6857 			rbd_dev->spec->snap_name);
6858 	}
6859 }
6860 
6861 static void rbd_dev_image_release(struct rbd_device *rbd_dev)
6862 {
6863 	if (!rbd_is_ro(rbd_dev))
6864 		rbd_unregister_watch(rbd_dev);
6865 
6866 	rbd_dev_unprobe(rbd_dev);
6867 	rbd_dev->image_format = 0;
6868 	kfree(rbd_dev->spec->image_id);
6869 	rbd_dev->spec->image_id = NULL;
6870 }
6871 
6872 /*
6873  * Probe for the existence of the header object for the given rbd
6874  * device.  If this image is the one being mapped (i.e., not a
6875  * parent), initiate a watch on its header object before using that
6876  * object to get detailed information about the rbd image.
6877  *
6878  * On success, returns with header_rwsem held for write if called
6879  * with @depth == 0.
6880  */
6881 static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth)
6882 {
6883 	bool need_watch = !rbd_is_ro(rbd_dev);
6884 	int ret;
6885 
6886 	/*
6887 	 * Get the id from the image id object.  Unless there's an
6888 	 * error, rbd_dev->spec->image_id will be filled in with
6889 	 * a dynamically-allocated string, and rbd_dev->image_format
6890 	 * will be set to either 1 or 2.
6891 	 */
6892 	ret = rbd_dev_image_id(rbd_dev);
6893 	if (ret)
6894 		return ret;
6895 
6896 	ret = rbd_dev_header_name(rbd_dev);
6897 	if (ret)
6898 		goto err_out_format;
6899 
6900 	if (need_watch) {
6901 		ret = rbd_register_watch(rbd_dev);
6902 		if (ret) {
6903 			if (ret == -ENOENT)
6904 				rbd_print_dne(rbd_dev, false);
6905 			goto err_out_format;
6906 		}
6907 	}
6908 
6909 	if (!depth)
6910 		down_write(&rbd_dev->header_rwsem);
6911 
6912 	ret = rbd_dev_header_info(rbd_dev);
6913 	if (ret) {
6914 		if (ret == -ENOENT && !need_watch)
6915 			rbd_print_dne(rbd_dev, false);
6916 		goto err_out_probe;
6917 	}
6918 
6919 	/*
6920 	 * If this image is the one being mapped, we have pool name and
6921 	 * id, image name and id, and snap name - need to fill snap id.
6922 	 * Otherwise this is a parent image, identified by pool, image
6923 	 * and snap ids - need to fill in names for those ids.
6924 	 */
6925 	if (!depth)
6926 		ret = rbd_spec_fill_snap_id(rbd_dev);
6927 	else
6928 		ret = rbd_spec_fill_names(rbd_dev);
6929 	if (ret) {
6930 		if (ret == -ENOENT)
6931 			rbd_print_dne(rbd_dev, true);
6932 		goto err_out_probe;
6933 	}
6934 
6935 	ret = rbd_dev_mapping_set(rbd_dev);
6936 	if (ret)
6937 		goto err_out_probe;
6938 
6939 	if (rbd_is_snap(rbd_dev) &&
6940 	    (rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP)) {
6941 		ret = rbd_object_map_load(rbd_dev);
6942 		if (ret)
6943 			goto err_out_probe;
6944 	}
6945 
6946 	if (rbd_dev->header.features & RBD_FEATURE_LAYERING) {
6947 		ret = rbd_dev_v2_parent_info(rbd_dev);
6948 		if (ret)
6949 			goto err_out_probe;
6950 	}
6951 
6952 	ret = rbd_dev_probe_parent(rbd_dev, depth);
6953 	if (ret)
6954 		goto err_out_probe;
6955 
6956 	dout("discovered format %u image, header name is %s\n",
6957 		rbd_dev->image_format, rbd_dev->header_oid.name);
6958 	return 0;
6959 
6960 err_out_probe:
6961 	if (!depth)
6962 		up_write(&rbd_dev->header_rwsem);
6963 	if (need_watch)
6964 		rbd_unregister_watch(rbd_dev);
6965 	rbd_dev_unprobe(rbd_dev);
6966 err_out_format:
6967 	rbd_dev->image_format = 0;
6968 	kfree(rbd_dev->spec->image_id);
6969 	rbd_dev->spec->image_id = NULL;
6970 	return ret;
6971 }
6972 
6973 static ssize_t do_rbd_add(struct bus_type *bus,
6974 			  const char *buf,
6975 			  size_t count)
6976 {
6977 	struct rbd_device *rbd_dev = NULL;
6978 	struct ceph_options *ceph_opts = NULL;
6979 	struct rbd_options *rbd_opts = NULL;
6980 	struct rbd_spec *spec = NULL;
6981 	struct rbd_client *rbdc;
6982 	int rc;
6983 
6984 	if (!capable(CAP_SYS_ADMIN))
6985 		return -EPERM;
6986 
6987 	if (!try_module_get(THIS_MODULE))
6988 		return -ENODEV;
6989 
6990 	/* parse add command */
6991 	rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec);
6992 	if (rc < 0)
6993 		goto out;
6994 
6995 	rbdc = rbd_get_client(ceph_opts);
6996 	if (IS_ERR(rbdc)) {
6997 		rc = PTR_ERR(rbdc);
6998 		goto err_out_args;
6999 	}
7000 
7001 	/* pick the pool */
7002 	rc = ceph_pg_poolid_by_name(rbdc->client->osdc.osdmap, spec->pool_name);
7003 	if (rc < 0) {
7004 		if (rc == -ENOENT)
7005 			pr_info("pool %s does not exist\n", spec->pool_name);
7006 		goto err_out_client;
7007 	}
7008 	spec->pool_id = (u64)rc;
7009 
7010 	rbd_dev = rbd_dev_create(rbdc, spec, rbd_opts);
7011 	if (!rbd_dev) {
7012 		rc = -ENOMEM;
7013 		goto err_out_client;
7014 	}
7015 	rbdc = NULL;		/* rbd_dev now owns this */
7016 	spec = NULL;		/* rbd_dev now owns this */
7017 	rbd_opts = NULL;	/* rbd_dev now owns this */
7018 
7019 	/* if we are mapping a snapshot it will be a read-only mapping */
7020 	if (rbd_dev->opts->read_only ||
7021 	    strcmp(rbd_dev->spec->snap_name, RBD_SNAP_HEAD_NAME))
7022 		__set_bit(RBD_DEV_FLAG_READONLY, &rbd_dev->flags);
7023 
7024 	rbd_dev->config_info = kstrdup(buf, GFP_KERNEL);
7025 	if (!rbd_dev->config_info) {
7026 		rc = -ENOMEM;
7027 		goto err_out_rbd_dev;
7028 	}
7029 
7030 	rc = rbd_dev_image_probe(rbd_dev, 0);
7031 	if (rc < 0)
7032 		goto err_out_rbd_dev;
7033 
7034 	if (rbd_dev->opts->alloc_size > rbd_dev->layout.object_size) {
7035 		rbd_warn(rbd_dev, "alloc_size adjusted to %u",
7036 			 rbd_dev->layout.object_size);
7037 		rbd_dev->opts->alloc_size = rbd_dev->layout.object_size;
7038 	}
7039 
7040 	rc = rbd_dev_device_setup(rbd_dev);
7041 	if (rc)
7042 		goto err_out_image_probe;
7043 
7044 	rc = rbd_add_acquire_lock(rbd_dev);
7045 	if (rc)
7046 		goto err_out_image_lock;
7047 
7048 	/* Everything's ready.  Announce the disk to the world. */
7049 
7050 	rc = device_add(&rbd_dev->dev);
7051 	if (rc)
7052 		goto err_out_image_lock;
7053 
7054 	rc = device_add_disk(&rbd_dev->dev, rbd_dev->disk, NULL);
7055 	if (rc)
7056 		goto err_out_cleanup_disk;
7057 
7058 	spin_lock(&rbd_dev_list_lock);
7059 	list_add_tail(&rbd_dev->node, &rbd_dev_list);
7060 	spin_unlock(&rbd_dev_list_lock);
7061 
7062 	pr_info("%s: capacity %llu features 0x%llx\n", rbd_dev->disk->disk_name,
7063 		(unsigned long long)get_capacity(rbd_dev->disk) << SECTOR_SHIFT,
7064 		rbd_dev->header.features);
7065 	rc = count;
7066 out:
7067 	module_put(THIS_MODULE);
7068 	return rc;
7069 
7070 err_out_cleanup_disk:
7071 	rbd_free_disk(rbd_dev);
7072 err_out_image_lock:
7073 	rbd_dev_image_unlock(rbd_dev);
7074 	rbd_dev_device_release(rbd_dev);
7075 err_out_image_probe:
7076 	rbd_dev_image_release(rbd_dev);
7077 err_out_rbd_dev:
7078 	rbd_dev_destroy(rbd_dev);
7079 err_out_client:
7080 	rbd_put_client(rbdc);
7081 err_out_args:
7082 	rbd_spec_put(spec);
7083 	kfree(rbd_opts);
7084 	goto out;
7085 }
7086 
7087 static ssize_t add_store(struct bus_type *bus, const char *buf, size_t count)
7088 {
7089 	if (single_major)
7090 		return -EINVAL;
7091 
7092 	return do_rbd_add(bus, buf, count);
7093 }
7094 
7095 static ssize_t add_single_major_store(struct bus_type *bus, const char *buf,
7096 				      size_t count)
7097 {
7098 	return do_rbd_add(bus, buf, count);
7099 }
7100 
7101 static void rbd_dev_remove_parent(struct rbd_device *rbd_dev)
7102 {
7103 	while (rbd_dev->parent) {
7104 		struct rbd_device *first = rbd_dev;
7105 		struct rbd_device *second = first->parent;
7106 		struct rbd_device *third;
7107 
7108 		/*
7109 		 * Follow to the parent with no grandparent and
7110 		 * remove it.
7111 		 */
7112 		while (second && (third = second->parent)) {
7113 			first = second;
7114 			second = third;
7115 		}
7116 		rbd_assert(second);
7117 		rbd_dev_image_release(second);
7118 		rbd_dev_destroy(second);
7119 		first->parent = NULL;
7120 		first->parent_overlap = 0;
7121 
7122 		rbd_assert(first->parent_spec);
7123 		rbd_spec_put(first->parent_spec);
7124 		first->parent_spec = NULL;
7125 	}
7126 }
7127 
7128 static ssize_t do_rbd_remove(struct bus_type *bus,
7129 			     const char *buf,
7130 			     size_t count)
7131 {
7132 	struct rbd_device *rbd_dev = NULL;
7133 	struct list_head *tmp;
7134 	int dev_id;
7135 	char opt_buf[6];
7136 	bool force = false;
7137 	int ret;
7138 
7139 	if (!capable(CAP_SYS_ADMIN))
7140 		return -EPERM;
7141 
7142 	dev_id = -1;
7143 	opt_buf[0] = '\0';
7144 	sscanf(buf, "%d %5s", &dev_id, opt_buf);
7145 	if (dev_id < 0) {
7146 		pr_err("dev_id out of range\n");
7147 		return -EINVAL;
7148 	}
7149 	if (opt_buf[0] != '\0') {
7150 		if (!strcmp(opt_buf, "force")) {
7151 			force = true;
7152 		} else {
7153 			pr_err("bad remove option at '%s'\n", opt_buf);
7154 			return -EINVAL;
7155 		}
7156 	}
7157 
7158 	ret = -ENOENT;
7159 	spin_lock(&rbd_dev_list_lock);
7160 	list_for_each(tmp, &rbd_dev_list) {
7161 		rbd_dev = list_entry(tmp, struct rbd_device, node);
7162 		if (rbd_dev->dev_id == dev_id) {
7163 			ret = 0;
7164 			break;
7165 		}
7166 	}
7167 	if (!ret) {
7168 		spin_lock_irq(&rbd_dev->lock);
7169 		if (rbd_dev->open_count && !force)
7170 			ret = -EBUSY;
7171 		else if (test_and_set_bit(RBD_DEV_FLAG_REMOVING,
7172 					  &rbd_dev->flags))
7173 			ret = -EINPROGRESS;
7174 		spin_unlock_irq(&rbd_dev->lock);
7175 	}
7176 	spin_unlock(&rbd_dev_list_lock);
7177 	if (ret)
7178 		return ret;
7179 
7180 	if (force) {
7181 		/*
7182 		 * Prevent new IO from being queued and wait for existing
7183 		 * IO to complete/fail.
7184 		 */
7185 		blk_mq_freeze_queue(rbd_dev->disk->queue);
7186 		blk_mark_disk_dead(rbd_dev->disk);
7187 	}
7188 
7189 	del_gendisk(rbd_dev->disk);
7190 	spin_lock(&rbd_dev_list_lock);
7191 	list_del_init(&rbd_dev->node);
7192 	spin_unlock(&rbd_dev_list_lock);
7193 	device_del(&rbd_dev->dev);
7194 
7195 	rbd_dev_image_unlock(rbd_dev);
7196 	rbd_dev_device_release(rbd_dev);
7197 	rbd_dev_image_release(rbd_dev);
7198 	rbd_dev_destroy(rbd_dev);
7199 	return count;
7200 }
7201 
7202 static ssize_t remove_store(struct bus_type *bus, const char *buf, size_t count)
7203 {
7204 	if (single_major)
7205 		return -EINVAL;
7206 
7207 	return do_rbd_remove(bus, buf, count);
7208 }
7209 
7210 static ssize_t remove_single_major_store(struct bus_type *bus, const char *buf,
7211 					 size_t count)
7212 {
7213 	return do_rbd_remove(bus, buf, count);
7214 }
7215 
7216 /*
7217  * create control files in sysfs
7218  * /sys/bus/rbd/...
7219  */
7220 static int __init rbd_sysfs_init(void)
7221 {
7222 	int ret;
7223 
7224 	ret = device_register(&rbd_root_dev);
7225 	if (ret < 0)
7226 		return ret;
7227 
7228 	ret = bus_register(&rbd_bus_type);
7229 	if (ret < 0)
7230 		device_unregister(&rbd_root_dev);
7231 
7232 	return ret;
7233 }
7234 
7235 static void __exit rbd_sysfs_cleanup(void)
7236 {
7237 	bus_unregister(&rbd_bus_type);
7238 	device_unregister(&rbd_root_dev);
7239 }
7240 
7241 static int __init rbd_slab_init(void)
7242 {
7243 	rbd_assert(!rbd_img_request_cache);
7244 	rbd_img_request_cache = KMEM_CACHE(rbd_img_request, 0);
7245 	if (!rbd_img_request_cache)
7246 		return -ENOMEM;
7247 
7248 	rbd_assert(!rbd_obj_request_cache);
7249 	rbd_obj_request_cache = KMEM_CACHE(rbd_obj_request, 0);
7250 	if (!rbd_obj_request_cache)
7251 		goto out_err;
7252 
7253 	return 0;
7254 
7255 out_err:
7256 	kmem_cache_destroy(rbd_img_request_cache);
7257 	rbd_img_request_cache = NULL;
7258 	return -ENOMEM;
7259 }
7260 
7261 static void rbd_slab_exit(void)
7262 {
7263 	rbd_assert(rbd_obj_request_cache);
7264 	kmem_cache_destroy(rbd_obj_request_cache);
7265 	rbd_obj_request_cache = NULL;
7266 
7267 	rbd_assert(rbd_img_request_cache);
7268 	kmem_cache_destroy(rbd_img_request_cache);
7269 	rbd_img_request_cache = NULL;
7270 }
7271 
7272 static int __init rbd_init(void)
7273 {
7274 	int rc;
7275 
7276 	if (!libceph_compatible(NULL)) {
7277 		rbd_warn(NULL, "libceph incompatibility (quitting)");
7278 		return -EINVAL;
7279 	}
7280 
7281 	rc = rbd_slab_init();
7282 	if (rc)
7283 		return rc;
7284 
7285 	/*
7286 	 * The number of active work items is limited by the number of
7287 	 * rbd devices * queue depth, so leave @max_active at default.
7288 	 */
7289 	rbd_wq = alloc_workqueue(RBD_DRV_NAME, WQ_MEM_RECLAIM, 0);
7290 	if (!rbd_wq) {
7291 		rc = -ENOMEM;
7292 		goto err_out_slab;
7293 	}
7294 
7295 	if (single_major) {
7296 		rbd_major = register_blkdev(0, RBD_DRV_NAME);
7297 		if (rbd_major < 0) {
7298 			rc = rbd_major;
7299 			goto err_out_wq;
7300 		}
7301 	}
7302 
7303 	rc = rbd_sysfs_init();
7304 	if (rc)
7305 		goto err_out_blkdev;
7306 
7307 	if (single_major)
7308 		pr_info("loaded (major %d)\n", rbd_major);
7309 	else
7310 		pr_info("loaded\n");
7311 
7312 	return 0;
7313 
7314 err_out_blkdev:
7315 	if (single_major)
7316 		unregister_blkdev(rbd_major, RBD_DRV_NAME);
7317 err_out_wq:
7318 	destroy_workqueue(rbd_wq);
7319 err_out_slab:
7320 	rbd_slab_exit();
7321 	return rc;
7322 }
7323 
7324 static void __exit rbd_exit(void)
7325 {
7326 	ida_destroy(&rbd_dev_id_ida);
7327 	rbd_sysfs_cleanup();
7328 	if (single_major)
7329 		unregister_blkdev(rbd_major, RBD_DRV_NAME);
7330 	destroy_workqueue(rbd_wq);
7331 	rbd_slab_exit();
7332 }
7333 
7334 module_init(rbd_init);
7335 module_exit(rbd_exit);
7336 
7337 MODULE_AUTHOR("Alex Elder <elder@inktank.com>");
7338 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
7339 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
7340 /* following authorship retained from original osdblk.c */
7341 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
7342 
7343 MODULE_DESCRIPTION("RADOS Block Device (RBD) driver");
7344 MODULE_LICENSE("GPL");
7345