xref: /openbmc/linux/fs/ceph/caps.c (revision 4fe59789)
13d14c5d2SYehuda Sadeh #include <linux/ceph/ceph_debug.h>
2a8599bd8SSage Weil 
3a8599bd8SSage Weil #include <linux/fs.h>
4a8599bd8SSage Weil #include <linux/kernel.h>
5a8599bd8SSage Weil #include <linux/sched.h>
65a0e3ad6STejun Heo #include <linux/slab.h>
7a8599bd8SSage Weil #include <linux/vmalloc.h>
8a8599bd8SSage Weil #include <linux/wait.h>
9f1a3d572SStephen Rothwell #include <linux/writeback.h>
10a8599bd8SSage Weil 
11a8599bd8SSage Weil #include "super.h"
123d14c5d2SYehuda Sadeh #include "mds_client.h"
1399ccbd22SMilosz Tanski #include "cache.h"
143d14c5d2SYehuda Sadeh #include <linux/ceph/decode.h>
153d14c5d2SYehuda Sadeh #include <linux/ceph/messenger.h>
16a8599bd8SSage Weil 
17a8599bd8SSage Weil /*
18a8599bd8SSage Weil  * Capability management
19a8599bd8SSage Weil  *
20a8599bd8SSage Weil  * The Ceph metadata servers control client access to inode metadata
21a8599bd8SSage Weil  * and file data by issuing capabilities, granting clients permission
22a8599bd8SSage Weil  * to read and/or write both inode field and file data to OSDs
23a8599bd8SSage Weil  * (storage nodes).  Each capability consists of a set of bits
24a8599bd8SSage Weil  * indicating which operations are allowed.
25a8599bd8SSage Weil  *
26a8599bd8SSage Weil  * If the client holds a *_SHARED cap, the client has a coherent value
27a8599bd8SSage Weil  * that can be safely read from the cached inode.
28a8599bd8SSage Weil  *
29a8599bd8SSage Weil  * In the case of a *_EXCL (exclusive) or FILE_WR capabilities, the
30a8599bd8SSage Weil  * client is allowed to change inode attributes (e.g., file size,
31a8599bd8SSage Weil  * mtime), note its dirty state in the ceph_cap, and asynchronously
32a8599bd8SSage Weil  * flush that metadata change to the MDS.
33a8599bd8SSage Weil  *
34a8599bd8SSage Weil  * In the event of a conflicting operation (perhaps by another
35a8599bd8SSage Weil  * client), the MDS will revoke the conflicting client capabilities.
36a8599bd8SSage Weil  *
37a8599bd8SSage Weil  * In order for a client to cache an inode, it must hold a capability
38a8599bd8SSage Weil  * with at least one MDS server.  When inodes are released, release
39a8599bd8SSage Weil  * notifications are batched and periodically sent en masse to the MDS
40a8599bd8SSage Weil  * cluster to release server state.
41a8599bd8SSage Weil  */
42a8599bd8SSage Weil 
43a8599bd8SSage Weil 
44a8599bd8SSage Weil /*
45a8599bd8SSage Weil  * Generate readable cap strings for debugging output.
46a8599bd8SSage Weil  */
47a8599bd8SSage Weil #define MAX_CAP_STR 20
48a8599bd8SSage Weil static char cap_str[MAX_CAP_STR][40];
49a8599bd8SSage Weil static DEFINE_SPINLOCK(cap_str_lock);
50a8599bd8SSage Weil static int last_cap_str;
51a8599bd8SSage Weil 
52a8599bd8SSage Weil static char *gcap_string(char *s, int c)
53a8599bd8SSage Weil {
54a8599bd8SSage Weil 	if (c & CEPH_CAP_GSHARED)
55a8599bd8SSage Weil 		*s++ = 's';
56a8599bd8SSage Weil 	if (c & CEPH_CAP_GEXCL)
57a8599bd8SSage Weil 		*s++ = 'x';
58a8599bd8SSage Weil 	if (c & CEPH_CAP_GCACHE)
59a8599bd8SSage Weil 		*s++ = 'c';
60a8599bd8SSage Weil 	if (c & CEPH_CAP_GRD)
61a8599bd8SSage Weil 		*s++ = 'r';
62a8599bd8SSage Weil 	if (c & CEPH_CAP_GWR)
63a8599bd8SSage Weil 		*s++ = 'w';
64a8599bd8SSage Weil 	if (c & CEPH_CAP_GBUFFER)
65a8599bd8SSage Weil 		*s++ = 'b';
66a8599bd8SSage Weil 	if (c & CEPH_CAP_GLAZYIO)
67a8599bd8SSage Weil 		*s++ = 'l';
68a8599bd8SSage Weil 	return s;
69a8599bd8SSage Weil }
70a8599bd8SSage Weil 
71a8599bd8SSage Weil const char *ceph_cap_string(int caps)
72a8599bd8SSage Weil {
73a8599bd8SSage Weil 	int i;
74a8599bd8SSage Weil 	char *s;
75a8599bd8SSage Weil 	int c;
76a8599bd8SSage Weil 
77a8599bd8SSage Weil 	spin_lock(&cap_str_lock);
78a8599bd8SSage Weil 	i = last_cap_str++;
79a8599bd8SSage Weil 	if (last_cap_str == MAX_CAP_STR)
80a8599bd8SSage Weil 		last_cap_str = 0;
81a8599bd8SSage Weil 	spin_unlock(&cap_str_lock);
82a8599bd8SSage Weil 
83a8599bd8SSage Weil 	s = cap_str[i];
84a8599bd8SSage Weil 
85a8599bd8SSage Weil 	if (caps & CEPH_CAP_PIN)
86a8599bd8SSage Weil 		*s++ = 'p';
87a8599bd8SSage Weil 
88a8599bd8SSage Weil 	c = (caps >> CEPH_CAP_SAUTH) & 3;
89a8599bd8SSage Weil 	if (c) {
90a8599bd8SSage Weil 		*s++ = 'A';
91a8599bd8SSage Weil 		s = gcap_string(s, c);
92a8599bd8SSage Weil 	}
93a8599bd8SSage Weil 
94a8599bd8SSage Weil 	c = (caps >> CEPH_CAP_SLINK) & 3;
95a8599bd8SSage Weil 	if (c) {
96a8599bd8SSage Weil 		*s++ = 'L';
97a8599bd8SSage Weil 		s = gcap_string(s, c);
98a8599bd8SSage Weil 	}
99a8599bd8SSage Weil 
100a8599bd8SSage Weil 	c = (caps >> CEPH_CAP_SXATTR) & 3;
101a8599bd8SSage Weil 	if (c) {
102a8599bd8SSage Weil 		*s++ = 'X';
103a8599bd8SSage Weil 		s = gcap_string(s, c);
104a8599bd8SSage Weil 	}
105a8599bd8SSage Weil 
106a8599bd8SSage Weil 	c = caps >> CEPH_CAP_SFILE;
107a8599bd8SSage Weil 	if (c) {
108a8599bd8SSage Weil 		*s++ = 'F';
109a8599bd8SSage Weil 		s = gcap_string(s, c);
110a8599bd8SSage Weil 	}
111a8599bd8SSage Weil 
112a8599bd8SSage Weil 	if (s == cap_str[i])
113a8599bd8SSage Weil 		*s++ = '-';
114a8599bd8SSage Weil 	*s = 0;
115a8599bd8SSage Weil 	return cap_str[i];
116a8599bd8SSage Weil }
117a8599bd8SSage Weil 
11837151668SYehuda Sadeh void ceph_caps_init(struct ceph_mds_client *mdsc)
119a8599bd8SSage Weil {
12037151668SYehuda Sadeh 	INIT_LIST_HEAD(&mdsc->caps_list);
12137151668SYehuda Sadeh 	spin_lock_init(&mdsc->caps_list_lock);
122a8599bd8SSage Weil }
123a8599bd8SSage Weil 
12437151668SYehuda Sadeh void ceph_caps_finalize(struct ceph_mds_client *mdsc)
125a8599bd8SSage Weil {
126a8599bd8SSage Weil 	struct ceph_cap *cap;
127a8599bd8SSage Weil 
12837151668SYehuda Sadeh 	spin_lock(&mdsc->caps_list_lock);
12937151668SYehuda Sadeh 	while (!list_empty(&mdsc->caps_list)) {
13037151668SYehuda Sadeh 		cap = list_first_entry(&mdsc->caps_list,
13137151668SYehuda Sadeh 				       struct ceph_cap, caps_item);
132a8599bd8SSage Weil 		list_del(&cap->caps_item);
133a8599bd8SSage Weil 		kmem_cache_free(ceph_cap_cachep, cap);
134a8599bd8SSage Weil 	}
13537151668SYehuda Sadeh 	mdsc->caps_total_count = 0;
13637151668SYehuda Sadeh 	mdsc->caps_avail_count = 0;
13737151668SYehuda Sadeh 	mdsc->caps_use_count = 0;
13837151668SYehuda Sadeh 	mdsc->caps_reserve_count = 0;
13937151668SYehuda Sadeh 	mdsc->caps_min_count = 0;
14037151668SYehuda Sadeh 	spin_unlock(&mdsc->caps_list_lock);
14185ccce43SSage Weil }
14285ccce43SSage Weil 
14337151668SYehuda Sadeh void ceph_adjust_min_caps(struct ceph_mds_client *mdsc, int delta)
14485ccce43SSage Weil {
14537151668SYehuda Sadeh 	spin_lock(&mdsc->caps_list_lock);
14637151668SYehuda Sadeh 	mdsc->caps_min_count += delta;
14737151668SYehuda Sadeh 	BUG_ON(mdsc->caps_min_count < 0);
14837151668SYehuda Sadeh 	spin_unlock(&mdsc->caps_list_lock);
149a8599bd8SSage Weil }
150a8599bd8SSage Weil 
15193faca6eSmajianpeng void ceph_reserve_caps(struct ceph_mds_client *mdsc,
15237151668SYehuda Sadeh 		      struct ceph_cap_reservation *ctx, int need)
153a8599bd8SSage Weil {
154a8599bd8SSage Weil 	int i;
155a8599bd8SSage Weil 	struct ceph_cap *cap;
156a8599bd8SSage Weil 	int have;
157a8599bd8SSage Weil 	int alloc = 0;
158a8599bd8SSage Weil 	LIST_HEAD(newcaps);
159a8599bd8SSage Weil 
160a8599bd8SSage Weil 	dout("reserve caps ctx=%p need=%d\n", ctx, need);
161a8599bd8SSage Weil 
162a8599bd8SSage Weil 	/* first reserve any caps that are already allocated */
16337151668SYehuda Sadeh 	spin_lock(&mdsc->caps_list_lock);
16437151668SYehuda Sadeh 	if (mdsc->caps_avail_count >= need)
165a8599bd8SSage Weil 		have = need;
166a8599bd8SSage Weil 	else
16737151668SYehuda Sadeh 		have = mdsc->caps_avail_count;
16837151668SYehuda Sadeh 	mdsc->caps_avail_count -= have;
16937151668SYehuda Sadeh 	mdsc->caps_reserve_count += have;
17037151668SYehuda Sadeh 	BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
17137151668SYehuda Sadeh 					 mdsc->caps_reserve_count +
17237151668SYehuda Sadeh 					 mdsc->caps_avail_count);
17337151668SYehuda Sadeh 	spin_unlock(&mdsc->caps_list_lock);
174a8599bd8SSage Weil 
175a8599bd8SSage Weil 	for (i = have; i < need; i++) {
176a8599bd8SSage Weil 		cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS);
17793faca6eSmajianpeng 		if (!cap)
17893faca6eSmajianpeng 			break;
179a8599bd8SSage Weil 		list_add(&cap->caps_item, &newcaps);
180a8599bd8SSage Weil 		alloc++;
181a8599bd8SSage Weil 	}
18293faca6eSmajianpeng 	/* we didn't manage to reserve as much as we needed */
18393faca6eSmajianpeng 	if (have + alloc != need)
18493faca6eSmajianpeng 		pr_warn("reserve caps ctx=%p ENOMEM need=%d got=%d\n",
18593faca6eSmajianpeng 			ctx, need, have + alloc);
186a8599bd8SSage Weil 
18737151668SYehuda Sadeh 	spin_lock(&mdsc->caps_list_lock);
18837151668SYehuda Sadeh 	mdsc->caps_total_count += alloc;
18937151668SYehuda Sadeh 	mdsc->caps_reserve_count += alloc;
19037151668SYehuda Sadeh 	list_splice(&newcaps, &mdsc->caps_list);
191a8599bd8SSage Weil 
19237151668SYehuda Sadeh 	BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
19337151668SYehuda Sadeh 					 mdsc->caps_reserve_count +
19437151668SYehuda Sadeh 					 mdsc->caps_avail_count);
19537151668SYehuda Sadeh 	spin_unlock(&mdsc->caps_list_lock);
196a8599bd8SSage Weil 
197a8599bd8SSage Weil 	ctx->count = need;
198a8599bd8SSage Weil 	dout("reserve caps ctx=%p %d = %d used + %d resv + %d avail\n",
19937151668SYehuda Sadeh 	     ctx, mdsc->caps_total_count, mdsc->caps_use_count,
20037151668SYehuda Sadeh 	     mdsc->caps_reserve_count, mdsc->caps_avail_count);
201a8599bd8SSage Weil }
202a8599bd8SSage Weil 
20337151668SYehuda Sadeh int ceph_unreserve_caps(struct ceph_mds_client *mdsc,
20437151668SYehuda Sadeh 			struct ceph_cap_reservation *ctx)
205a8599bd8SSage Weil {
206a8599bd8SSage Weil 	dout("unreserve caps ctx=%p count=%d\n", ctx, ctx->count);
207a8599bd8SSage Weil 	if (ctx->count) {
20837151668SYehuda Sadeh 		spin_lock(&mdsc->caps_list_lock);
20937151668SYehuda Sadeh 		BUG_ON(mdsc->caps_reserve_count < ctx->count);
21037151668SYehuda Sadeh 		mdsc->caps_reserve_count -= ctx->count;
21137151668SYehuda Sadeh 		mdsc->caps_avail_count += ctx->count;
212a8599bd8SSage Weil 		ctx->count = 0;
213a8599bd8SSage Weil 		dout("unreserve caps %d = %d used + %d resv + %d avail\n",
21437151668SYehuda Sadeh 		     mdsc->caps_total_count, mdsc->caps_use_count,
21537151668SYehuda Sadeh 		     mdsc->caps_reserve_count, mdsc->caps_avail_count);
21637151668SYehuda Sadeh 		BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
21737151668SYehuda Sadeh 						 mdsc->caps_reserve_count +
21837151668SYehuda Sadeh 						 mdsc->caps_avail_count);
21937151668SYehuda Sadeh 		spin_unlock(&mdsc->caps_list_lock);
220a8599bd8SSage Weil 	}
221a8599bd8SSage Weil 	return 0;
222a8599bd8SSage Weil }
223a8599bd8SSage Weil 
22437151668SYehuda Sadeh static struct ceph_cap *get_cap(struct ceph_mds_client *mdsc,
22537151668SYehuda Sadeh 				struct ceph_cap_reservation *ctx)
226a8599bd8SSage Weil {
227a8599bd8SSage Weil 	struct ceph_cap *cap = NULL;
228a8599bd8SSage Weil 
229a8599bd8SSage Weil 	/* temporary, until we do something about cap import/export */
230443b3760SSage Weil 	if (!ctx) {
231443b3760SSage Weil 		cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS);
232443b3760SSage Weil 		if (cap) {
2334d1d0534SYan, Zheng 			spin_lock(&mdsc->caps_list_lock);
23437151668SYehuda Sadeh 			mdsc->caps_use_count++;
23537151668SYehuda Sadeh 			mdsc->caps_total_count++;
2364d1d0534SYan, Zheng 			spin_unlock(&mdsc->caps_list_lock);
237443b3760SSage Weil 		}
238443b3760SSage Weil 		return cap;
239443b3760SSage Weil 	}
240a8599bd8SSage Weil 
24137151668SYehuda Sadeh 	spin_lock(&mdsc->caps_list_lock);
242a8599bd8SSage Weil 	dout("get_cap ctx=%p (%d) %d = %d used + %d resv + %d avail\n",
24337151668SYehuda Sadeh 	     ctx, ctx->count, mdsc->caps_total_count, mdsc->caps_use_count,
24437151668SYehuda Sadeh 	     mdsc->caps_reserve_count, mdsc->caps_avail_count);
245a8599bd8SSage Weil 	BUG_ON(!ctx->count);
24637151668SYehuda Sadeh 	BUG_ON(ctx->count > mdsc->caps_reserve_count);
24737151668SYehuda Sadeh 	BUG_ON(list_empty(&mdsc->caps_list));
248a8599bd8SSage Weil 
249a8599bd8SSage Weil 	ctx->count--;
25037151668SYehuda Sadeh 	mdsc->caps_reserve_count--;
25137151668SYehuda Sadeh 	mdsc->caps_use_count++;
252a8599bd8SSage Weil 
25337151668SYehuda Sadeh 	cap = list_first_entry(&mdsc->caps_list, struct ceph_cap, caps_item);
254a8599bd8SSage Weil 	list_del(&cap->caps_item);
255a8599bd8SSage Weil 
25637151668SYehuda Sadeh 	BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
25737151668SYehuda Sadeh 	       mdsc->caps_reserve_count + mdsc->caps_avail_count);
25837151668SYehuda Sadeh 	spin_unlock(&mdsc->caps_list_lock);
259a8599bd8SSage Weil 	return cap;
260a8599bd8SSage Weil }
261a8599bd8SSage Weil 
26237151668SYehuda Sadeh void ceph_put_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap)
263a8599bd8SSage Weil {
26437151668SYehuda Sadeh 	spin_lock(&mdsc->caps_list_lock);
2657c1332b8SSage Weil 	dout("put_cap %p %d = %d used + %d resv + %d avail\n",
26637151668SYehuda Sadeh 	     cap, mdsc->caps_total_count, mdsc->caps_use_count,
26737151668SYehuda Sadeh 	     mdsc->caps_reserve_count, mdsc->caps_avail_count);
26837151668SYehuda Sadeh 	mdsc->caps_use_count--;
269a8599bd8SSage Weil 	/*
27085ccce43SSage Weil 	 * Keep some preallocated caps around (ceph_min_count), to
27185ccce43SSage Weil 	 * avoid lots of free/alloc churn.
272a8599bd8SSage Weil 	 */
27337151668SYehuda Sadeh 	if (mdsc->caps_avail_count >= mdsc->caps_reserve_count +
27437151668SYehuda Sadeh 				      mdsc->caps_min_count) {
27537151668SYehuda Sadeh 		mdsc->caps_total_count--;
276a8599bd8SSage Weil 		kmem_cache_free(ceph_cap_cachep, cap);
277a8599bd8SSage Weil 	} else {
27837151668SYehuda Sadeh 		mdsc->caps_avail_count++;
27937151668SYehuda Sadeh 		list_add(&cap->caps_item, &mdsc->caps_list);
280a8599bd8SSage Weil 	}
281a8599bd8SSage Weil 
28237151668SYehuda Sadeh 	BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
28337151668SYehuda Sadeh 	       mdsc->caps_reserve_count + mdsc->caps_avail_count);
28437151668SYehuda Sadeh 	spin_unlock(&mdsc->caps_list_lock);
285a8599bd8SSage Weil }
286a8599bd8SSage Weil 
2873d14c5d2SYehuda Sadeh void ceph_reservation_status(struct ceph_fs_client *fsc,
28885ccce43SSage Weil 			     int *total, int *avail, int *used, int *reserved,
28985ccce43SSage Weil 			     int *min)
290a8599bd8SSage Weil {
2913d14c5d2SYehuda Sadeh 	struct ceph_mds_client *mdsc = fsc->mdsc;
29237151668SYehuda Sadeh 
293a8599bd8SSage Weil 	if (total)
29437151668SYehuda Sadeh 		*total = mdsc->caps_total_count;
295a8599bd8SSage Weil 	if (avail)
29637151668SYehuda Sadeh 		*avail = mdsc->caps_avail_count;
297a8599bd8SSage Weil 	if (used)
29837151668SYehuda Sadeh 		*used = mdsc->caps_use_count;
299a8599bd8SSage Weil 	if (reserved)
30037151668SYehuda Sadeh 		*reserved = mdsc->caps_reserve_count;
30185ccce43SSage Weil 	if (min)
30237151668SYehuda Sadeh 		*min = mdsc->caps_min_count;
303a8599bd8SSage Weil }
304a8599bd8SSage Weil 
305a8599bd8SSage Weil /*
306a8599bd8SSage Weil  * Find ceph_cap for given mds, if any.
307a8599bd8SSage Weil  *
308be655596SSage Weil  * Called with i_ceph_lock held.
309a8599bd8SSage Weil  */
310a8599bd8SSage Weil static struct ceph_cap *__get_cap_for_mds(struct ceph_inode_info *ci, int mds)
311a8599bd8SSage Weil {
312a8599bd8SSage Weil 	struct ceph_cap *cap;
313a8599bd8SSage Weil 	struct rb_node *n = ci->i_caps.rb_node;
314a8599bd8SSage Weil 
315a8599bd8SSage Weil 	while (n) {
316a8599bd8SSage Weil 		cap = rb_entry(n, struct ceph_cap, ci_node);
317a8599bd8SSage Weil 		if (mds < cap->mds)
318a8599bd8SSage Weil 			n = n->rb_left;
319a8599bd8SSage Weil 		else if (mds > cap->mds)
320a8599bd8SSage Weil 			n = n->rb_right;
321a8599bd8SSage Weil 		else
322a8599bd8SSage Weil 			return cap;
323a8599bd8SSage Weil 	}
324a8599bd8SSage Weil 	return NULL;
325a8599bd8SSage Weil }
326a8599bd8SSage Weil 
3272bc50259SGreg Farnum struct ceph_cap *ceph_get_cap_for_mds(struct ceph_inode_info *ci, int mds)
3282bc50259SGreg Farnum {
3292bc50259SGreg Farnum 	struct ceph_cap *cap;
3302bc50259SGreg Farnum 
331be655596SSage Weil 	spin_lock(&ci->i_ceph_lock);
3322bc50259SGreg Farnum 	cap = __get_cap_for_mds(ci, mds);
333be655596SSage Weil 	spin_unlock(&ci->i_ceph_lock);
3342bc50259SGreg Farnum 	return cap;
3352bc50259SGreg Farnum }
3362bc50259SGreg Farnum 
337a8599bd8SSage Weil /*
33833caad32SSage Weil  * Return id of any MDS with a cap, preferably FILE_WR|BUFFER|EXCL, else -1.
339a8599bd8SSage Weil  */
340ca81f3f6SSage Weil static int __ceph_get_cap_mds(struct ceph_inode_info *ci)
341a8599bd8SSage Weil {
342a8599bd8SSage Weil 	struct ceph_cap *cap;
343a8599bd8SSage Weil 	int mds = -1;
344a8599bd8SSage Weil 	struct rb_node *p;
345a8599bd8SSage Weil 
34633caad32SSage Weil 	/* prefer mds with WR|BUFFER|EXCL caps */
347a8599bd8SSage Weil 	for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
348a8599bd8SSage Weil 		cap = rb_entry(p, struct ceph_cap, ci_node);
349a8599bd8SSage Weil 		mds = cap->mds;
350a8599bd8SSage Weil 		if (cap->issued & (CEPH_CAP_FILE_WR |
351a8599bd8SSage Weil 				   CEPH_CAP_FILE_BUFFER |
352a8599bd8SSage Weil 				   CEPH_CAP_FILE_EXCL))
353a8599bd8SSage Weil 			break;
354a8599bd8SSage Weil 	}
355a8599bd8SSage Weil 	return mds;
356a8599bd8SSage Weil }
357a8599bd8SSage Weil 
358a8599bd8SSage Weil int ceph_get_cap_mds(struct inode *inode)
359a8599bd8SSage Weil {
360be655596SSage Weil 	struct ceph_inode_info *ci = ceph_inode(inode);
361a8599bd8SSage Weil 	int mds;
362be655596SSage Weil 	spin_lock(&ci->i_ceph_lock);
363ca81f3f6SSage Weil 	mds = __ceph_get_cap_mds(ceph_inode(inode));
364be655596SSage Weil 	spin_unlock(&ci->i_ceph_lock);
365a8599bd8SSage Weil 	return mds;
366a8599bd8SSage Weil }
367a8599bd8SSage Weil 
368a8599bd8SSage Weil /*
369be655596SSage Weil  * Called under i_ceph_lock.
370a8599bd8SSage Weil  */
371a8599bd8SSage Weil static void __insert_cap_node(struct ceph_inode_info *ci,
372a8599bd8SSage Weil 			      struct ceph_cap *new)
373a8599bd8SSage Weil {
374a8599bd8SSage Weil 	struct rb_node **p = &ci->i_caps.rb_node;
375a8599bd8SSage Weil 	struct rb_node *parent = NULL;
376a8599bd8SSage Weil 	struct ceph_cap *cap = NULL;
377a8599bd8SSage Weil 
378a8599bd8SSage Weil 	while (*p) {
379a8599bd8SSage Weil 		parent = *p;
380a8599bd8SSage Weil 		cap = rb_entry(parent, struct ceph_cap, ci_node);
381a8599bd8SSage Weil 		if (new->mds < cap->mds)
382a8599bd8SSage Weil 			p = &(*p)->rb_left;
383a8599bd8SSage Weil 		else if (new->mds > cap->mds)
384a8599bd8SSage Weil 			p = &(*p)->rb_right;
385a8599bd8SSage Weil 		else
386a8599bd8SSage Weil 			BUG();
387a8599bd8SSage Weil 	}
388a8599bd8SSage Weil 
389a8599bd8SSage Weil 	rb_link_node(&new->ci_node, parent, p);
390a8599bd8SSage Weil 	rb_insert_color(&new->ci_node, &ci->i_caps);
391a8599bd8SSage Weil }
392a8599bd8SSage Weil 
393a8599bd8SSage Weil /*
394a8599bd8SSage Weil  * (re)set cap hold timeouts, which control the delayed release
395a8599bd8SSage Weil  * of unused caps back to the MDS.  Should be called on cap use.
396a8599bd8SSage Weil  */
397a8599bd8SSage Weil static void __cap_set_timeouts(struct ceph_mds_client *mdsc,
398a8599bd8SSage Weil 			       struct ceph_inode_info *ci)
399a8599bd8SSage Weil {
4003d14c5d2SYehuda Sadeh 	struct ceph_mount_options *ma = mdsc->fsc->mount_options;
401a8599bd8SSage Weil 
402a8599bd8SSage Weil 	ci->i_hold_caps_min = round_jiffies(jiffies +
403a8599bd8SSage Weil 					    ma->caps_wanted_delay_min * HZ);
404a8599bd8SSage Weil 	ci->i_hold_caps_max = round_jiffies(jiffies +
405a8599bd8SSage Weil 					    ma->caps_wanted_delay_max * HZ);
406a8599bd8SSage Weil 	dout("__cap_set_timeouts %p min %lu max %lu\n", &ci->vfs_inode,
407a8599bd8SSage Weil 	     ci->i_hold_caps_min - jiffies, ci->i_hold_caps_max - jiffies);
408a8599bd8SSage Weil }
409a8599bd8SSage Weil 
410a8599bd8SSage Weil /*
411a8599bd8SSage Weil  * (Re)queue cap at the end of the delayed cap release list.
412a8599bd8SSage Weil  *
413a8599bd8SSage Weil  * If I_FLUSH is set, leave the inode at the front of the list.
414a8599bd8SSage Weil  *
415be655596SSage Weil  * Caller holds i_ceph_lock
416a8599bd8SSage Weil  *    -> we take mdsc->cap_delay_lock
417a8599bd8SSage Weil  */
418a8599bd8SSage Weil static void __cap_delay_requeue(struct ceph_mds_client *mdsc,
419a8599bd8SSage Weil 				struct ceph_inode_info *ci)
420a8599bd8SSage Weil {
421a8599bd8SSage Weil 	__cap_set_timeouts(mdsc, ci);
422a8599bd8SSage Weil 	dout("__cap_delay_requeue %p flags %d at %lu\n", &ci->vfs_inode,
423a8599bd8SSage Weil 	     ci->i_ceph_flags, ci->i_hold_caps_max);
424a8599bd8SSage Weil 	if (!mdsc->stopping) {
425a8599bd8SSage Weil 		spin_lock(&mdsc->cap_delay_lock);
426a8599bd8SSage Weil 		if (!list_empty(&ci->i_cap_delay_list)) {
427a8599bd8SSage Weil 			if (ci->i_ceph_flags & CEPH_I_FLUSH)
428a8599bd8SSage Weil 				goto no_change;
429a8599bd8SSage Weil 			list_del_init(&ci->i_cap_delay_list);
430a8599bd8SSage Weil 		}
431a8599bd8SSage Weil 		list_add_tail(&ci->i_cap_delay_list, &mdsc->cap_delay_list);
432a8599bd8SSage Weil no_change:
433a8599bd8SSage Weil 		spin_unlock(&mdsc->cap_delay_lock);
434a8599bd8SSage Weil 	}
435a8599bd8SSage Weil }
436a8599bd8SSage Weil 
437a8599bd8SSage Weil /*
438a8599bd8SSage Weil  * Queue an inode for immediate writeback.  Mark inode with I_FLUSH,
439a8599bd8SSage Weil  * indicating we should send a cap message to flush dirty metadata
440a8599bd8SSage Weil  * asap, and move to the front of the delayed cap list.
441a8599bd8SSage Weil  */
442a8599bd8SSage Weil static void __cap_delay_requeue_front(struct ceph_mds_client *mdsc,
443a8599bd8SSage Weil 				      struct ceph_inode_info *ci)
444a8599bd8SSage Weil {
445a8599bd8SSage Weil 	dout("__cap_delay_requeue_front %p\n", &ci->vfs_inode);
446a8599bd8SSage Weil 	spin_lock(&mdsc->cap_delay_lock);
447a8599bd8SSage Weil 	ci->i_ceph_flags |= CEPH_I_FLUSH;
448a8599bd8SSage Weil 	if (!list_empty(&ci->i_cap_delay_list))
449a8599bd8SSage Weil 		list_del_init(&ci->i_cap_delay_list);
450a8599bd8SSage Weil 	list_add(&ci->i_cap_delay_list, &mdsc->cap_delay_list);
451a8599bd8SSage Weil 	spin_unlock(&mdsc->cap_delay_lock);
452a8599bd8SSage Weil }
453a8599bd8SSage Weil 
454a8599bd8SSage Weil /*
455a8599bd8SSage Weil  * Cancel delayed work on cap.
456a8599bd8SSage Weil  *
457be655596SSage Weil  * Caller must hold i_ceph_lock.
458a8599bd8SSage Weil  */
459a8599bd8SSage Weil static void __cap_delay_cancel(struct ceph_mds_client *mdsc,
460a8599bd8SSage Weil 			       struct ceph_inode_info *ci)
461a8599bd8SSage Weil {
462a8599bd8SSage Weil 	dout("__cap_delay_cancel %p\n", &ci->vfs_inode);
463a8599bd8SSage Weil 	if (list_empty(&ci->i_cap_delay_list))
464a8599bd8SSage Weil 		return;
465a8599bd8SSage Weil 	spin_lock(&mdsc->cap_delay_lock);
466a8599bd8SSage Weil 	list_del_init(&ci->i_cap_delay_list);
467a8599bd8SSage Weil 	spin_unlock(&mdsc->cap_delay_lock);
468a8599bd8SSage Weil }
469a8599bd8SSage Weil 
470a8599bd8SSage Weil /*
471a8599bd8SSage Weil  * Common issue checks for add_cap, handle_cap_grant.
472a8599bd8SSage Weil  */
473a8599bd8SSage Weil static void __check_cap_issue(struct ceph_inode_info *ci, struct ceph_cap *cap,
474a8599bd8SSage Weil 			      unsigned issued)
475a8599bd8SSage Weil {
476a8599bd8SSage Weil 	unsigned had = __ceph_caps_issued(ci, NULL);
477a8599bd8SSage Weil 
478a8599bd8SSage Weil 	/*
479a8599bd8SSage Weil 	 * Each time we receive FILE_CACHE anew, we increment
480a8599bd8SSage Weil 	 * i_rdcache_gen.
481a8599bd8SSage Weil 	 */
4822962507cSSage Weil 	if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) &&
48399ccbd22SMilosz Tanski 	    (had & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0) {
484a8599bd8SSage Weil 		ci->i_rdcache_gen++;
48599ccbd22SMilosz Tanski 	}
486a8599bd8SSage Weil 
487a8599bd8SSage Weil 	/*
4882f276c51SYan, Zheng 	 * if we are newly issued FILE_SHARED, mark dir not complete; we
489a8599bd8SSage Weil 	 * don't know what happened to this directory while we didn't
490a8599bd8SSage Weil 	 * have the cap.
491a8599bd8SSage Weil 	 */
492a8599bd8SSage Weil 	if ((issued & CEPH_CAP_FILE_SHARED) &&
493a8599bd8SSage Weil 	    (had & CEPH_CAP_FILE_SHARED) == 0) {
494a8599bd8SSage Weil 		ci->i_shared_gen++;
495a8673d61SYan, Zheng 		if (S_ISDIR(ci->vfs_inode.i_mode)) {
496a8673d61SYan, Zheng 			dout(" marking %p NOT complete\n", &ci->vfs_inode);
4972f276c51SYan, Zheng 			__ceph_dir_clear_complete(ci);
498a8673d61SYan, Zheng 		}
499a8599bd8SSage Weil 	}
500a8599bd8SSage Weil }
501a8599bd8SSage Weil 
502a8599bd8SSage Weil /*
503a8599bd8SSage Weil  * Add a capability under the given MDS session.
504a8599bd8SSage Weil  *
505a8599bd8SSage Weil  * Caller should hold session snap_rwsem (read) and s_mutex.
506a8599bd8SSage Weil  *
507a8599bd8SSage Weil  * @fmode is the open file mode, if we are opening a file, otherwise
508a8599bd8SSage Weil  * it is < 0.  (This is so we can atomically add the cap and add an
509a8599bd8SSage Weil  * open file reference to it.)
510a8599bd8SSage Weil  */
511a8599bd8SSage Weil int ceph_add_cap(struct inode *inode,
512a8599bd8SSage Weil 		 struct ceph_mds_session *session, u64 cap_id,
513a8599bd8SSage Weil 		 int fmode, unsigned issued, unsigned wanted,
514a8599bd8SSage Weil 		 unsigned seq, unsigned mseq, u64 realmino, int flags,
515a8599bd8SSage Weil 		 struct ceph_cap_reservation *caps_reservation)
516a8599bd8SSage Weil {
5173d14c5d2SYehuda Sadeh 	struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
518a8599bd8SSage Weil 	struct ceph_inode_info *ci = ceph_inode(inode);
519a8599bd8SSage Weil 	struct ceph_cap *new_cap = NULL;
520a8599bd8SSage Weil 	struct ceph_cap *cap;
521a8599bd8SSage Weil 	int mds = session->s_mds;
522a8599bd8SSage Weil 	int actual_wanted;
523a8599bd8SSage Weil 
524a8599bd8SSage Weil 	dout("add_cap %p mds%d cap %llx %s seq %d\n", inode,
525a8599bd8SSage Weil 	     session->s_mds, cap_id, ceph_cap_string(issued), seq);
526a8599bd8SSage Weil 
527a8599bd8SSage Weil 	/*
528a8599bd8SSage Weil 	 * If we are opening the file, include file mode wanted bits
529a8599bd8SSage Weil 	 * in wanted.
530a8599bd8SSage Weil 	 */
531a8599bd8SSage Weil 	if (fmode >= 0)
532a8599bd8SSage Weil 		wanted |= ceph_caps_for_mode(fmode);
533a8599bd8SSage Weil 
534a8599bd8SSage Weil retry:
535be655596SSage Weil 	spin_lock(&ci->i_ceph_lock);
536a8599bd8SSage Weil 	cap = __get_cap_for_mds(ci, mds);
537a8599bd8SSage Weil 	if (!cap) {
538a8599bd8SSage Weil 		if (new_cap) {
539a8599bd8SSage Weil 			cap = new_cap;
540a8599bd8SSage Weil 			new_cap = NULL;
541a8599bd8SSage Weil 		} else {
542be655596SSage Weil 			spin_unlock(&ci->i_ceph_lock);
54337151668SYehuda Sadeh 			new_cap = get_cap(mdsc, caps_reservation);
544a8599bd8SSage Weil 			if (new_cap == NULL)
545a8599bd8SSage Weil 				return -ENOMEM;
546a8599bd8SSage Weil 			goto retry;
547a8599bd8SSage Weil 		}
548a8599bd8SSage Weil 
549a8599bd8SSage Weil 		cap->issued = 0;
550a8599bd8SSage Weil 		cap->implemented = 0;
551a8599bd8SSage Weil 		cap->mds = mds;
552a8599bd8SSage Weil 		cap->mds_wanted = 0;
553964266ccSYan, Zheng 		cap->mseq = 0;
554a8599bd8SSage Weil 
555a8599bd8SSage Weil 		cap->ci = ci;
556a8599bd8SSage Weil 		__insert_cap_node(ci, cap);
557a8599bd8SSage Weil 
558a8599bd8SSage Weil 		/* clear out old exporting info?  (i.e. on cap import) */
559a8599bd8SSage Weil 		if (ci->i_cap_exporting_mds == mds) {
560a8599bd8SSage Weil 			ci->i_cap_exporting_issued = 0;
561a8599bd8SSage Weil 			ci->i_cap_exporting_mseq = 0;
562a8599bd8SSage Weil 			ci->i_cap_exporting_mds = -1;
563a8599bd8SSage Weil 		}
564a8599bd8SSage Weil 
565a8599bd8SSage Weil 		/* add to session cap list */
566a8599bd8SSage Weil 		cap->session = session;
567a8599bd8SSage Weil 		spin_lock(&session->s_cap_lock);
568a8599bd8SSage Weil 		list_add_tail(&cap->session_caps, &session->s_caps);
569a8599bd8SSage Weil 		session->s_nr_caps++;
570a8599bd8SSage Weil 		spin_unlock(&session->s_cap_lock);
5713540303fSSage Weil 	} else if (new_cap)
5723540303fSSage Weil 		ceph_put_cap(mdsc, new_cap);
573a8599bd8SSage Weil 
574a8599bd8SSage Weil 	if (!ci->i_snap_realm) {
575a8599bd8SSage Weil 		/*
576a8599bd8SSage Weil 		 * add this inode to the appropriate snap realm
577a8599bd8SSage Weil 		 */
578a8599bd8SSage Weil 		struct ceph_snap_realm *realm = ceph_lookup_snap_realm(mdsc,
579a8599bd8SSage Weil 							       realmino);
580a8599bd8SSage Weil 		if (realm) {
581a8599bd8SSage Weil 			ceph_get_snap_realm(mdsc, realm);
582a8599bd8SSage Weil 			spin_lock(&realm->inodes_with_caps_lock);
583a8599bd8SSage Weil 			ci->i_snap_realm = realm;
584a8599bd8SSage Weil 			list_add(&ci->i_snap_realm_item,
585a8599bd8SSage Weil 				 &realm->inodes_with_caps);
586a8599bd8SSage Weil 			spin_unlock(&realm->inodes_with_caps_lock);
587a8599bd8SSage Weil 		} else {
588a8599bd8SSage Weil 			pr_err("ceph_add_cap: couldn't find snap realm %llx\n",
589a8599bd8SSage Weil 			       realmino);
590b8cd07e7SSage Weil 			WARN_ON(!realm);
591a8599bd8SSage Weil 		}
592a8599bd8SSage Weil 	}
593a8599bd8SSage Weil 
594a8599bd8SSage Weil 	__check_cap_issue(ci, cap, issued);
595a8599bd8SSage Weil 
596a8599bd8SSage Weil 	/*
597a8599bd8SSage Weil 	 * If we are issued caps we don't want, or the mds' wanted
598a8599bd8SSage Weil 	 * value appears to be off, queue a check so we'll release
599a8599bd8SSage Weil 	 * later and/or update the mds wanted value.
600a8599bd8SSage Weil 	 */
601a8599bd8SSage Weil 	actual_wanted = __ceph_caps_wanted(ci);
602a8599bd8SSage Weil 	if ((wanted & ~actual_wanted) ||
603a8599bd8SSage Weil 	    (issued & ~actual_wanted & CEPH_CAP_ANY_WR)) {
604a8599bd8SSage Weil 		dout(" issued %s, mds wanted %s, actual %s, queueing\n",
605a8599bd8SSage Weil 		     ceph_cap_string(issued), ceph_cap_string(wanted),
606a8599bd8SSage Weil 		     ceph_cap_string(actual_wanted));
607a8599bd8SSage Weil 		__cap_delay_requeue(mdsc, ci);
608a8599bd8SSage Weil 	}
609a8599bd8SSage Weil 
610b8c2f3aeSYan, Zheng 	if (flags & CEPH_CAP_FLAG_AUTH) {
611b8c2f3aeSYan, Zheng 		if (ci->i_auth_cap == NULL ||
612b8c2f3aeSYan, Zheng 		    ceph_seq_cmp(ci->i_auth_cap->mseq, mseq) < 0)
613a8599bd8SSage Weil 			ci->i_auth_cap = cap;
614b8c2f3aeSYan, Zheng 	} else if (ci->i_auth_cap == cap) {
615a8599bd8SSage Weil 		ci->i_auth_cap = NULL;
6168a92a119SYan, Zheng 		spin_lock(&mdsc->cap_dirty_lock);
6178a92a119SYan, Zheng 		if (!list_empty(&ci->i_dirty_item)) {
6188a92a119SYan, Zheng 			dout(" moving %p to cap_dirty_migrating\n", inode);
6198a92a119SYan, Zheng 			list_move(&ci->i_dirty_item,
6208a92a119SYan, Zheng 				  &mdsc->cap_dirty_migrating);
6218a92a119SYan, Zheng 		}
6228a92a119SYan, Zheng 		spin_unlock(&mdsc->cap_dirty_lock);
6238a92a119SYan, Zheng 	}
624a8599bd8SSage Weil 
625a8599bd8SSage Weil 	dout("add_cap inode %p (%llx.%llx) cap %p %s now %s seq %d mds%d\n",
626a8599bd8SSage Weil 	     inode, ceph_vinop(inode), cap, ceph_cap_string(issued),
627a8599bd8SSage Weil 	     ceph_cap_string(issued|cap->issued), seq, mds);
628a8599bd8SSage Weil 	cap->cap_id = cap_id;
629a8599bd8SSage Weil 	cap->issued = issued;
630a8599bd8SSage Weil 	cap->implemented |= issued;
631964266ccSYan, Zheng 	if (mseq > cap->mseq)
632964266ccSYan, Zheng 		cap->mds_wanted = wanted;
633964266ccSYan, Zheng 	else
634a8599bd8SSage Weil 		cap->mds_wanted |= wanted;
635a8599bd8SSage Weil 	cap->seq = seq;
636a8599bd8SSage Weil 	cap->issue_seq = seq;
637a8599bd8SSage Weil 	cap->mseq = mseq;
638685f9a5dSSage Weil 	cap->cap_gen = session->s_cap_gen;
639a8599bd8SSage Weil 
640a8599bd8SSage Weil 	if (fmode >= 0)
641a8599bd8SSage Weil 		__ceph_get_fmode(ci, fmode);
642be655596SSage Weil 	spin_unlock(&ci->i_ceph_lock);
64303066f23SYehuda Sadeh 	wake_up_all(&ci->i_cap_wq);
644a8599bd8SSage Weil 	return 0;
645a8599bd8SSage Weil }
646a8599bd8SSage Weil 
647a8599bd8SSage Weil /*
648a8599bd8SSage Weil  * Return true if cap has not timed out and belongs to the current
649a8599bd8SSage Weil  * generation of the MDS session (i.e. has not gone 'stale' due to
650a8599bd8SSage Weil  * us losing touch with the mds).
651a8599bd8SSage Weil  */
652a8599bd8SSage Weil static int __cap_is_valid(struct ceph_cap *cap)
653a8599bd8SSage Weil {
654a8599bd8SSage Weil 	unsigned long ttl;
655cdac8303SSage Weil 	u32 gen;
656a8599bd8SSage Weil 
657d8fb02abSAlex Elder 	spin_lock(&cap->session->s_gen_ttl_lock);
658a8599bd8SSage Weil 	gen = cap->session->s_cap_gen;
659a8599bd8SSage Weil 	ttl = cap->session->s_cap_ttl;
660d8fb02abSAlex Elder 	spin_unlock(&cap->session->s_gen_ttl_lock);
661a8599bd8SSage Weil 
662685f9a5dSSage Weil 	if (cap->cap_gen < gen || time_after_eq(jiffies, ttl)) {
663a8599bd8SSage Weil 		dout("__cap_is_valid %p cap %p issued %s "
664a8599bd8SSage Weil 		     "but STALE (gen %u vs %u)\n", &cap->ci->vfs_inode,
665685f9a5dSSage Weil 		     cap, ceph_cap_string(cap->issued), cap->cap_gen, gen);
666a8599bd8SSage Weil 		return 0;
667a8599bd8SSage Weil 	}
668a8599bd8SSage Weil 
669a8599bd8SSage Weil 	return 1;
670a8599bd8SSage Weil }
671a8599bd8SSage Weil 
672a8599bd8SSage Weil /*
673a8599bd8SSage Weil  * Return set of valid cap bits issued to us.  Note that caps time
674a8599bd8SSage Weil  * out, and may be invalidated in bulk if the client session times out
675a8599bd8SSage Weil  * and session->s_cap_gen is bumped.
676a8599bd8SSage Weil  */
677a8599bd8SSage Weil int __ceph_caps_issued(struct ceph_inode_info *ci, int *implemented)
678a8599bd8SSage Weil {
6797af8f1e4SSage Weil 	int have = ci->i_snap_caps | ci->i_cap_exporting_issued;
680a8599bd8SSage Weil 	struct ceph_cap *cap;
681a8599bd8SSage Weil 	struct rb_node *p;
682a8599bd8SSage Weil 
683a8599bd8SSage Weil 	if (implemented)
684a8599bd8SSage Weil 		*implemented = 0;
685a8599bd8SSage Weil 	for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
686a8599bd8SSage Weil 		cap = rb_entry(p, struct ceph_cap, ci_node);
687a8599bd8SSage Weil 		if (!__cap_is_valid(cap))
688a8599bd8SSage Weil 			continue;
689a8599bd8SSage Weil 		dout("__ceph_caps_issued %p cap %p issued %s\n",
690a8599bd8SSage Weil 		     &ci->vfs_inode, cap, ceph_cap_string(cap->issued));
691a8599bd8SSage Weil 		have |= cap->issued;
692a8599bd8SSage Weil 		if (implemented)
693a8599bd8SSage Weil 			*implemented |= cap->implemented;
694a8599bd8SSage Weil 	}
695b1530f57SYan, Zheng 	/*
696b1530f57SYan, Zheng 	 * exclude caps issued by non-auth MDS, but are been revoking
697b1530f57SYan, Zheng 	 * by the auth MDS. The non-auth MDS should be revoking/exporting
698b1530f57SYan, Zheng 	 * these caps, but the message is delayed.
699b1530f57SYan, Zheng 	 */
700b1530f57SYan, Zheng 	if (ci->i_auth_cap) {
701b1530f57SYan, Zheng 		cap = ci->i_auth_cap;
702b1530f57SYan, Zheng 		have &= ~cap->implemented | cap->issued;
703b1530f57SYan, Zheng 	}
704a8599bd8SSage Weil 	return have;
705a8599bd8SSage Weil }
706a8599bd8SSage Weil 
707a8599bd8SSage Weil /*
708a8599bd8SSage Weil  * Get cap bits issued by caps other than @ocap
709a8599bd8SSage Weil  */
710a8599bd8SSage Weil int __ceph_caps_issued_other(struct ceph_inode_info *ci, struct ceph_cap *ocap)
711a8599bd8SSage Weil {
712a8599bd8SSage Weil 	int have = ci->i_snap_caps;
713a8599bd8SSage Weil 	struct ceph_cap *cap;
714a8599bd8SSage Weil 	struct rb_node *p;
715a8599bd8SSage Weil 
716a8599bd8SSage Weil 	for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
717a8599bd8SSage Weil 		cap = rb_entry(p, struct ceph_cap, ci_node);
718a8599bd8SSage Weil 		if (cap == ocap)
719a8599bd8SSage Weil 			continue;
720a8599bd8SSage Weil 		if (!__cap_is_valid(cap))
721a8599bd8SSage Weil 			continue;
722a8599bd8SSage Weil 		have |= cap->issued;
723a8599bd8SSage Weil 	}
724a8599bd8SSage Weil 	return have;
725a8599bd8SSage Weil }
726a8599bd8SSage Weil 
727a8599bd8SSage Weil /*
728a8599bd8SSage Weil  * Move a cap to the end of the LRU (oldest caps at list head, newest
729a8599bd8SSage Weil  * at list tail).
730a8599bd8SSage Weil  */
731a8599bd8SSage Weil static void __touch_cap(struct ceph_cap *cap)
732a8599bd8SSage Weil {
733a8599bd8SSage Weil 	struct ceph_mds_session *s = cap->session;
734a8599bd8SSage Weil 
7355dacf091SSage Weil 	spin_lock(&s->s_cap_lock);
7367c1332b8SSage Weil 	if (s->s_cap_iterator == NULL) {
737a8599bd8SSage Weil 		dout("__touch_cap %p cap %p mds%d\n", &cap->ci->vfs_inode, cap,
738a8599bd8SSage Weil 		     s->s_mds);
739a8599bd8SSage Weil 		list_move_tail(&cap->session_caps, &s->s_caps);
7405dacf091SSage Weil 	} else {
7415dacf091SSage Weil 		dout("__touch_cap %p cap %p mds%d NOP, iterating over caps\n",
7425dacf091SSage Weil 		     &cap->ci->vfs_inode, cap, s->s_mds);
7435dacf091SSage Weil 	}
744a8599bd8SSage Weil 	spin_unlock(&s->s_cap_lock);
745a8599bd8SSage Weil }
746a8599bd8SSage Weil 
747a8599bd8SSage Weil /*
748a8599bd8SSage Weil  * Check if we hold the given mask.  If so, move the cap(s) to the
749a8599bd8SSage Weil  * front of their respective LRUs.  (This is the preferred way for
750a8599bd8SSage Weil  * callers to check for caps they want.)
751a8599bd8SSage Weil  */
752a8599bd8SSage Weil int __ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask, int touch)
753a8599bd8SSage Weil {
754a8599bd8SSage Weil 	struct ceph_cap *cap;
755a8599bd8SSage Weil 	struct rb_node *p;
756a8599bd8SSage Weil 	int have = ci->i_snap_caps;
757a8599bd8SSage Weil 
758a8599bd8SSage Weil 	if ((have & mask) == mask) {
759a8599bd8SSage Weil 		dout("__ceph_caps_issued_mask %p snap issued %s"
760a8599bd8SSage Weil 		     " (mask %s)\n", &ci->vfs_inode,
761a8599bd8SSage Weil 		     ceph_cap_string(have),
762a8599bd8SSage Weil 		     ceph_cap_string(mask));
763a8599bd8SSage Weil 		return 1;
764a8599bd8SSage Weil 	}
765a8599bd8SSage Weil 
766a8599bd8SSage Weil 	for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
767a8599bd8SSage Weil 		cap = rb_entry(p, struct ceph_cap, ci_node);
768a8599bd8SSage Weil 		if (!__cap_is_valid(cap))
769a8599bd8SSage Weil 			continue;
770a8599bd8SSage Weil 		if ((cap->issued & mask) == mask) {
771a8599bd8SSage Weil 			dout("__ceph_caps_issued_mask %p cap %p issued %s"
772a8599bd8SSage Weil 			     " (mask %s)\n", &ci->vfs_inode, cap,
773a8599bd8SSage Weil 			     ceph_cap_string(cap->issued),
774a8599bd8SSage Weil 			     ceph_cap_string(mask));
775a8599bd8SSage Weil 			if (touch)
776a8599bd8SSage Weil 				__touch_cap(cap);
777a8599bd8SSage Weil 			return 1;
778a8599bd8SSage Weil 		}
779a8599bd8SSage Weil 
780a8599bd8SSage Weil 		/* does a combination of caps satisfy mask? */
781a8599bd8SSage Weil 		have |= cap->issued;
782a8599bd8SSage Weil 		if ((have & mask) == mask) {
783a8599bd8SSage Weil 			dout("__ceph_caps_issued_mask %p combo issued %s"
784a8599bd8SSage Weil 			     " (mask %s)\n", &ci->vfs_inode,
785a8599bd8SSage Weil 			     ceph_cap_string(cap->issued),
786a8599bd8SSage Weil 			     ceph_cap_string(mask));
787a8599bd8SSage Weil 			if (touch) {
788a8599bd8SSage Weil 				struct rb_node *q;
789a8599bd8SSage Weil 
79025985edcSLucas De Marchi 				/* touch this + preceding caps */
791a8599bd8SSage Weil 				__touch_cap(cap);
792a8599bd8SSage Weil 				for (q = rb_first(&ci->i_caps); q != p;
793a8599bd8SSage Weil 				     q = rb_next(q)) {
794a8599bd8SSage Weil 					cap = rb_entry(q, struct ceph_cap,
795a8599bd8SSage Weil 						       ci_node);
796a8599bd8SSage Weil 					if (!__cap_is_valid(cap))
797a8599bd8SSage Weil 						continue;
798a8599bd8SSage Weil 					__touch_cap(cap);
799a8599bd8SSage Weil 				}
800a8599bd8SSage Weil 			}
801a8599bd8SSage Weil 			return 1;
802a8599bd8SSage Weil 		}
803a8599bd8SSage Weil 	}
804a8599bd8SSage Weil 
805a8599bd8SSage Weil 	return 0;
806a8599bd8SSage Weil }
807a8599bd8SSage Weil 
808a8599bd8SSage Weil /*
809a8599bd8SSage Weil  * Return true if mask caps are currently being revoked by an MDS.
810a8599bd8SSage Weil  */
8116ee6b953SYan, Zheng int __ceph_caps_revoking_other(struct ceph_inode_info *ci,
8126ee6b953SYan, Zheng 			       struct ceph_cap *ocap, int mask)
8136ee6b953SYan, Zheng {
8146ee6b953SYan, Zheng 	struct ceph_cap *cap;
8156ee6b953SYan, Zheng 	struct rb_node *p;
8166ee6b953SYan, Zheng 
8176ee6b953SYan, Zheng 	for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
8186ee6b953SYan, Zheng 		cap = rb_entry(p, struct ceph_cap, ci_node);
8196ee6b953SYan, Zheng 		if (cap != ocap && __cap_is_valid(cap) &&
8206ee6b953SYan, Zheng 		    (cap->implemented & ~cap->issued & mask))
8216ee6b953SYan, Zheng 			return 1;
8226ee6b953SYan, Zheng 	}
8236ee6b953SYan, Zheng 	return 0;
8246ee6b953SYan, Zheng }
8256ee6b953SYan, Zheng 
826a8599bd8SSage Weil int ceph_caps_revoking(struct ceph_inode_info *ci, int mask)
827a8599bd8SSage Weil {
828a8599bd8SSage Weil 	struct inode *inode = &ci->vfs_inode;
8296ee6b953SYan, Zheng 	int ret;
830a8599bd8SSage Weil 
831be655596SSage Weil 	spin_lock(&ci->i_ceph_lock);
8326ee6b953SYan, Zheng 	ret = __ceph_caps_revoking_other(ci, NULL, mask);
833be655596SSage Weil 	spin_unlock(&ci->i_ceph_lock);
834a8599bd8SSage Weil 	dout("ceph_caps_revoking %p %s = %d\n", inode,
835a8599bd8SSage Weil 	     ceph_cap_string(mask), ret);
836a8599bd8SSage Weil 	return ret;
837a8599bd8SSage Weil }
838a8599bd8SSage Weil 
839a8599bd8SSage Weil int __ceph_caps_used(struct ceph_inode_info *ci)
840a8599bd8SSage Weil {
841a8599bd8SSage Weil 	int used = 0;
842a8599bd8SSage Weil 	if (ci->i_pin_ref)
843a8599bd8SSage Weil 		used |= CEPH_CAP_PIN;
844a8599bd8SSage Weil 	if (ci->i_rd_ref)
845a8599bd8SSage Weil 		used |= CEPH_CAP_FILE_RD;
846a43fb731SSage Weil 	if (ci->i_rdcache_ref || ci->vfs_inode.i_data.nrpages)
847a8599bd8SSage Weil 		used |= CEPH_CAP_FILE_CACHE;
848a8599bd8SSage Weil 	if (ci->i_wr_ref)
849a8599bd8SSage Weil 		used |= CEPH_CAP_FILE_WR;
850d3d0720dSHenry C Chang 	if (ci->i_wb_ref || ci->i_wrbuffer_ref)
851a8599bd8SSage Weil 		used |= CEPH_CAP_FILE_BUFFER;
852a8599bd8SSage Weil 	return used;
853a8599bd8SSage Weil }
854a8599bd8SSage Weil 
855a8599bd8SSage Weil /*
856a8599bd8SSage Weil  * wanted, by virtue of open file modes
857a8599bd8SSage Weil  */
858a8599bd8SSage Weil int __ceph_caps_file_wanted(struct ceph_inode_info *ci)
859a8599bd8SSage Weil {
860a8599bd8SSage Weil 	int want = 0;
861a8599bd8SSage Weil 	int mode;
86233caad32SSage Weil 	for (mode = 0; mode < CEPH_FILE_MODE_NUM; mode++)
863a8599bd8SSage Weil 		if (ci->i_nr_by_mode[mode])
864a8599bd8SSage Weil 			want |= ceph_caps_for_mode(mode);
865a8599bd8SSage Weil 	return want;
866a8599bd8SSage Weil }
867a8599bd8SSage Weil 
868a8599bd8SSage Weil /*
869a8599bd8SSage Weil  * Return caps we have registered with the MDS(s) as 'wanted'.
870a8599bd8SSage Weil  */
871a8599bd8SSage Weil int __ceph_caps_mds_wanted(struct ceph_inode_info *ci)
872a8599bd8SSage Weil {
873a8599bd8SSage Weil 	struct ceph_cap *cap;
874a8599bd8SSage Weil 	struct rb_node *p;
875a8599bd8SSage Weil 	int mds_wanted = 0;
876a8599bd8SSage Weil 
877a8599bd8SSage Weil 	for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
878a8599bd8SSage Weil 		cap = rb_entry(p, struct ceph_cap, ci_node);
879a8599bd8SSage Weil 		if (!__cap_is_valid(cap))
880a8599bd8SSage Weil 			continue;
881a8599bd8SSage Weil 		mds_wanted |= cap->mds_wanted;
882a8599bd8SSage Weil 	}
883a8599bd8SSage Weil 	return mds_wanted;
884a8599bd8SSage Weil }
885a8599bd8SSage Weil 
886a8599bd8SSage Weil /*
887be655596SSage Weil  * called under i_ceph_lock
888a8599bd8SSage Weil  */
889a8599bd8SSage Weil static int __ceph_is_any_caps(struct ceph_inode_info *ci)
890a8599bd8SSage Weil {
891a8599bd8SSage Weil 	return !RB_EMPTY_ROOT(&ci->i_caps) || ci->i_cap_exporting_mds >= 0;
892a8599bd8SSage Weil }
893a8599bd8SSage Weil 
894a8599bd8SSage Weil /*
895f818a736SSage Weil  * Remove a cap.  Take steps to deal with a racing iterate_session_caps.
896f818a736SSage Weil  *
897be655596SSage Weil  * caller should hold i_ceph_lock.
898a6369741SSage Weil  * caller will not hold session s_mutex if called from destroy_inode.
899a8599bd8SSage Weil  */
900a096b09aSYan, Zheng void __ceph_remove_cap(struct ceph_cap *cap, bool queue_release)
901a8599bd8SSage Weil {
902a8599bd8SSage Weil 	struct ceph_mds_session *session = cap->session;
903a8599bd8SSage Weil 	struct ceph_inode_info *ci = cap->ci;
904640ef79dSCheng Renquan 	struct ceph_mds_client *mdsc =
9053d14c5d2SYehuda Sadeh 		ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
906f818a736SSage Weil 	int removed = 0;
907a8599bd8SSage Weil 
908a8599bd8SSage Weil 	dout("__ceph_remove_cap %p from %p\n", cap, &ci->vfs_inode);
909a8599bd8SSage Weil 
9107c1332b8SSage Weil 	/* remove from session list */
9117c1332b8SSage Weil 	spin_lock(&session->s_cap_lock);
91299a9c273SYan, Zheng 	/*
91399a9c273SYan, Zheng 	 * s_cap_reconnect is protected by s_cap_lock. no one changes
91499a9c273SYan, Zheng 	 * s_cap_gen while session is in the reconnect state.
91599a9c273SYan, Zheng 	 */
91699a9c273SYan, Zheng 	if (queue_release &&
91799a9c273SYan, Zheng 	    (!session->s_cap_reconnect ||
91899a9c273SYan, Zheng 	     cap->cap_gen == session->s_cap_gen))
919a096b09aSYan, Zheng 		__queue_cap_release(session, ci->i_vino.ino, cap->cap_id,
920a096b09aSYan, Zheng 				    cap->mseq, cap->issue_seq);
921a096b09aSYan, Zheng 
9227c1332b8SSage Weil 	if (session->s_cap_iterator == cap) {
9237c1332b8SSage Weil 		/* not yet, we are iterating over this very cap */
9247c1332b8SSage Weil 		dout("__ceph_remove_cap  delaying %p removal from session %p\n",
9257c1332b8SSage Weil 		     cap, cap->session);
9267c1332b8SSage Weil 	} else {
9277c1332b8SSage Weil 		list_del_init(&cap->session_caps);
9287c1332b8SSage Weil 		session->s_nr_caps--;
9297c1332b8SSage Weil 		cap->session = NULL;
930f818a736SSage Weil 		removed = 1;
9317c1332b8SSage Weil 	}
932f818a736SSage Weil 	/* protect backpointer with s_cap_lock: see iterate_session_caps */
933f818a736SSage Weil 	cap->ci = NULL;
9347c1332b8SSage Weil 	spin_unlock(&session->s_cap_lock);
9357c1332b8SSage Weil 
936f818a736SSage Weil 	/* remove from inode list */
937f818a736SSage Weil 	rb_erase(&cap->ci_node, &ci->i_caps);
938f818a736SSage Weil 	if (ci->i_auth_cap == cap)
939f818a736SSage Weil 		ci->i_auth_cap = NULL;
940f818a736SSage Weil 
941f818a736SSage Weil 	if (removed)
94237151668SYehuda Sadeh 		ceph_put_cap(mdsc, cap);
943a8599bd8SSage Weil 
944a8599bd8SSage Weil 	if (!__ceph_is_any_caps(ci) && ci->i_snap_realm) {
945a8599bd8SSage Weil 		struct ceph_snap_realm *realm = ci->i_snap_realm;
946a8599bd8SSage Weil 		spin_lock(&realm->inodes_with_caps_lock);
947a8599bd8SSage Weil 		list_del_init(&ci->i_snap_realm_item);
948a8599bd8SSage Weil 		ci->i_snap_realm_counter++;
949a8599bd8SSage Weil 		ci->i_snap_realm = NULL;
950a8599bd8SSage Weil 		spin_unlock(&realm->inodes_with_caps_lock);
951a8599bd8SSage Weil 		ceph_put_snap_realm(mdsc, realm);
952a8599bd8SSage Weil 	}
953a8599bd8SSage Weil 	if (!__ceph_is_any_real_caps(ci))
954a8599bd8SSage Weil 		__cap_delay_cancel(mdsc, ci);
955a8599bd8SSage Weil }
956a8599bd8SSage Weil 
957a8599bd8SSage Weil /*
958a8599bd8SSage Weil  * Build and send a cap message to the given MDS.
959a8599bd8SSage Weil  *
960a8599bd8SSage Weil  * Caller should be holding s_mutex.
961a8599bd8SSage Weil  */
962a8599bd8SSage Weil static int send_cap_msg(struct ceph_mds_session *session,
963a8599bd8SSage Weil 			u64 ino, u64 cid, int op,
964a8599bd8SSage Weil 			int caps, int wanted, int dirty,
965a8599bd8SSage Weil 			u32 seq, u64 flush_tid, u32 issue_seq, u32 mseq,
966a8599bd8SSage Weil 			u64 size, u64 max_size,
967a8599bd8SSage Weil 			struct timespec *mtime, struct timespec *atime,
968a8599bd8SSage Weil 			u64 time_warp_seq,
96905cb11c1SEric W. Biederman 			kuid_t uid, kgid_t gid, umode_t mode,
970a8599bd8SSage Weil 			u64 xattr_version,
971a8599bd8SSage Weil 			struct ceph_buffer *xattrs_buf,
972a8599bd8SSage Weil 			u64 follows)
973a8599bd8SSage Weil {
974a8599bd8SSage Weil 	struct ceph_mds_caps *fc;
975a8599bd8SSage Weil 	struct ceph_msg *msg;
976a8599bd8SSage Weil 
977a8599bd8SSage Weil 	dout("send_cap_msg %s %llx %llx caps %s wanted %s dirty %s"
978a8599bd8SSage Weil 	     " seq %u/%u mseq %u follows %lld size %llu/%llu"
979a8599bd8SSage Weil 	     " xattr_ver %llu xattr_len %d\n", ceph_cap_op_name(op),
980a8599bd8SSage Weil 	     cid, ino, ceph_cap_string(caps), ceph_cap_string(wanted),
981a8599bd8SSage Weil 	     ceph_cap_string(dirty),
982a8599bd8SSage Weil 	     seq, issue_seq, mseq, follows, size, max_size,
983a8599bd8SSage Weil 	     xattr_version, xattrs_buf ? (int)xattrs_buf->vec.iov_len : 0);
984a8599bd8SSage Weil 
985b61c2763SSage Weil 	msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPS, sizeof(*fc), GFP_NOFS, false);
986a79832f2SSage Weil 	if (!msg)
987a79832f2SSage Weil 		return -ENOMEM;
988a8599bd8SSage Weil 
9896df058c0SSage Weil 	msg->hdr.tid = cpu_to_le64(flush_tid);
990a8599bd8SSage Weil 
9916df058c0SSage Weil 	fc = msg->front.iov_base;
992a8599bd8SSage Weil 	memset(fc, 0, sizeof(*fc));
993a8599bd8SSage Weil 
994a8599bd8SSage Weil 	fc->cap_id = cpu_to_le64(cid);
995a8599bd8SSage Weil 	fc->op = cpu_to_le32(op);
996a8599bd8SSage Weil 	fc->seq = cpu_to_le32(seq);
997a8599bd8SSage Weil 	fc->issue_seq = cpu_to_le32(issue_seq);
998a8599bd8SSage Weil 	fc->migrate_seq = cpu_to_le32(mseq);
999a8599bd8SSage Weil 	fc->caps = cpu_to_le32(caps);
1000a8599bd8SSage Weil 	fc->wanted = cpu_to_le32(wanted);
1001a8599bd8SSage Weil 	fc->dirty = cpu_to_le32(dirty);
1002a8599bd8SSage Weil 	fc->ino = cpu_to_le64(ino);
1003a8599bd8SSage Weil 	fc->snap_follows = cpu_to_le64(follows);
1004a8599bd8SSage Weil 
1005a8599bd8SSage Weil 	fc->size = cpu_to_le64(size);
1006a8599bd8SSage Weil 	fc->max_size = cpu_to_le64(max_size);
1007a8599bd8SSage Weil 	if (mtime)
1008a8599bd8SSage Weil 		ceph_encode_timespec(&fc->mtime, mtime);
1009a8599bd8SSage Weil 	if (atime)
1010a8599bd8SSage Weil 		ceph_encode_timespec(&fc->atime, atime);
1011a8599bd8SSage Weil 	fc->time_warp_seq = cpu_to_le32(time_warp_seq);
1012a8599bd8SSage Weil 
101305cb11c1SEric W. Biederman 	fc->uid = cpu_to_le32(from_kuid(&init_user_ns, uid));
101405cb11c1SEric W. Biederman 	fc->gid = cpu_to_le32(from_kgid(&init_user_ns, gid));
1015a8599bd8SSage Weil 	fc->mode = cpu_to_le32(mode);
1016a8599bd8SSage Weil 
1017a8599bd8SSage Weil 	fc->xattr_version = cpu_to_le64(xattr_version);
1018a8599bd8SSage Weil 	if (xattrs_buf) {
1019a8599bd8SSage Weil 		msg->middle = ceph_buffer_get(xattrs_buf);
1020a8599bd8SSage Weil 		fc->xattr_len = cpu_to_le32(xattrs_buf->vec.iov_len);
1021a8599bd8SSage Weil 		msg->hdr.middle_len = cpu_to_le32(xattrs_buf->vec.iov_len);
1022a8599bd8SSage Weil 	}
1023a8599bd8SSage Weil 
1024a8599bd8SSage Weil 	ceph_con_send(&session->s_con, msg);
1025a8599bd8SSage Weil 	return 0;
1026a8599bd8SSage Weil }
1027a8599bd8SSage Weil 
1028d40ee0dcSYan, Zheng void __queue_cap_release(struct ceph_mds_session *session,
10293d7ded4dSSage Weil 			 u64 ino, u64 cap_id, u32 migrate_seq,
10303d7ded4dSSage Weil 			 u32 issue_seq)
10313d7ded4dSSage Weil {
10323d7ded4dSSage Weil 	struct ceph_msg *msg;
10333d7ded4dSSage Weil 	struct ceph_mds_cap_release *head;
10343d7ded4dSSage Weil 	struct ceph_mds_cap_item *item;
10353d7ded4dSSage Weil 
10363d7ded4dSSage Weil 	BUG_ON(!session->s_num_cap_releases);
10373d7ded4dSSage Weil 	msg = list_first_entry(&session->s_cap_releases,
10383d7ded4dSSage Weil 			       struct ceph_msg, list_head);
10393d7ded4dSSage Weil 
10403d7ded4dSSage Weil 	dout(" adding %llx release to mds%d msg %p (%d left)\n",
10413d7ded4dSSage Weil 	     ino, session->s_mds, msg, session->s_num_cap_releases);
10423d7ded4dSSage Weil 
10433d7ded4dSSage Weil 	BUG_ON(msg->front.iov_len + sizeof(*item) > PAGE_CACHE_SIZE);
10443d7ded4dSSage Weil 	head = msg->front.iov_base;
1045b905a7f8SWei Yongjun 	le32_add_cpu(&head->num, 1);
10463d7ded4dSSage Weil 	item = msg->front.iov_base + msg->front.iov_len;
10473d7ded4dSSage Weil 	item->ino = cpu_to_le64(ino);
10483d7ded4dSSage Weil 	item->cap_id = cpu_to_le64(cap_id);
10493d7ded4dSSage Weil 	item->migrate_seq = cpu_to_le32(migrate_seq);
10503d7ded4dSSage Weil 	item->seq = cpu_to_le32(issue_seq);
10513d7ded4dSSage Weil 
10523d7ded4dSSage Weil 	session->s_num_cap_releases--;
10533d7ded4dSSage Weil 
10543d7ded4dSSage Weil 	msg->front.iov_len += sizeof(*item);
10553d7ded4dSSage Weil 	if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
10563d7ded4dSSage Weil 		dout(" release msg %p full\n", msg);
10573d7ded4dSSage Weil 		list_move_tail(&msg->list_head, &session->s_cap_releases_done);
10583d7ded4dSSage Weil 	} else {
10593d7ded4dSSage Weil 		dout(" release msg %p at %d/%d (%d)\n", msg,
10603d7ded4dSSage Weil 		     (int)le32_to_cpu(head->num),
10613d7ded4dSSage Weil 		     (int)CEPH_CAPS_PER_RELEASE,
10623d7ded4dSSage Weil 		     (int)msg->front.iov_len);
10633d7ded4dSSage Weil 	}
10643d7ded4dSSage Weil }
10653d7ded4dSSage Weil 
1066a8599bd8SSage Weil /*
1067a6369741SSage Weil  * Queue cap releases when an inode is dropped from our cache.  Since
1068be655596SSage Weil  * inode is about to be destroyed, there is no need for i_ceph_lock.
1069a8599bd8SSage Weil  */
1070a8599bd8SSage Weil void ceph_queue_caps_release(struct inode *inode)
1071a8599bd8SSage Weil {
1072a8599bd8SSage Weil 	struct ceph_inode_info *ci = ceph_inode(inode);
1073a8599bd8SSage Weil 	struct rb_node *p;
1074a8599bd8SSage Weil 
1075a8599bd8SSage Weil 	p = rb_first(&ci->i_caps);
1076a8599bd8SSage Weil 	while (p) {
1077a8599bd8SSage Weil 		struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node);
1078a8599bd8SSage Weil 		p = rb_next(p);
1079a096b09aSYan, Zheng 		__ceph_remove_cap(cap, true);
1080a8599bd8SSage Weil 	}
1081a8599bd8SSage Weil }
1082a8599bd8SSage Weil 
1083a8599bd8SSage Weil /*
1084a8599bd8SSage Weil  * Send a cap msg on the given inode.  Update our caps state, then
1085be655596SSage Weil  * drop i_ceph_lock and send the message.
1086a8599bd8SSage Weil  *
1087a8599bd8SSage Weil  * Make note of max_size reported/requested from mds, revoked caps
1088a8599bd8SSage Weil  * that have now been implemented.
1089a8599bd8SSage Weil  *
1090a8599bd8SSage Weil  * Make half-hearted attempt ot to invalidate page cache if we are
1091a8599bd8SSage Weil  * dropping RDCACHE.  Note that this will leave behind locked pages
1092a8599bd8SSage Weil  * that we'll then need to deal with elsewhere.
1093a8599bd8SSage Weil  *
1094a8599bd8SSage Weil  * Return non-zero if delayed release, or we experienced an error
1095a8599bd8SSage Weil  * such that the caller should requeue + retry later.
1096a8599bd8SSage Weil  *
1097be655596SSage Weil  * called with i_ceph_lock, then drops it.
1098a8599bd8SSage Weil  * caller should hold snap_rwsem (read), s_mutex.
1099a8599bd8SSage Weil  */
1100a8599bd8SSage Weil static int __send_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap,
1101a8599bd8SSage Weil 		      int op, int used, int want, int retain, int flushing,
1102a8599bd8SSage Weil 		      unsigned *pflush_tid)
1103be655596SSage Weil 	__releases(cap->ci->i_ceph_lock)
1104a8599bd8SSage Weil {
1105a8599bd8SSage Weil 	struct ceph_inode_info *ci = cap->ci;
1106a8599bd8SSage Weil 	struct inode *inode = &ci->vfs_inode;
1107a8599bd8SSage Weil 	u64 cap_id = cap->cap_id;
110868c28323SSage Weil 	int held, revoking, dropping, keep;
1109a8599bd8SSage Weil 	u64 seq, issue_seq, mseq, time_warp_seq, follows;
1110a8599bd8SSage Weil 	u64 size, max_size;
1111a8599bd8SSage Weil 	struct timespec mtime, atime;
1112a8599bd8SSage Weil 	int wake = 0;
11135706b27dSAl Viro 	umode_t mode;
111405cb11c1SEric W. Biederman 	kuid_t uid;
111505cb11c1SEric W. Biederman 	kgid_t gid;
1116a8599bd8SSage Weil 	struct ceph_mds_session *session;
1117a8599bd8SSage Weil 	u64 xattr_version = 0;
1118082afec9SSage Weil 	struct ceph_buffer *xattr_blob = NULL;
1119a8599bd8SSage Weil 	int delayed = 0;
1120a8599bd8SSage Weil 	u64 flush_tid = 0;
1121a8599bd8SSage Weil 	int i;
1122a8599bd8SSage Weil 	int ret;
1123a8599bd8SSage Weil 
112468c28323SSage Weil 	held = cap->issued | cap->implemented;
112568c28323SSage Weil 	revoking = cap->implemented & ~cap->issued;
112668c28323SSage Weil 	retain &= ~revoking;
112768c28323SSage Weil 	dropping = cap->issued & ~retain;
112868c28323SSage Weil 
1129a8599bd8SSage Weil 	dout("__send_cap %p cap %p session %p %s -> %s (revoking %s)\n",
1130a8599bd8SSage Weil 	     inode, cap, cap->session,
1131a8599bd8SSage Weil 	     ceph_cap_string(held), ceph_cap_string(held & retain),
1132a8599bd8SSage Weil 	     ceph_cap_string(revoking));
1133a8599bd8SSage Weil 	BUG_ON((retain & CEPH_CAP_PIN) == 0);
1134a8599bd8SSage Weil 
1135a8599bd8SSage Weil 	session = cap->session;
1136a8599bd8SSage Weil 
1137a8599bd8SSage Weil 	/* don't release wanted unless we've waited a bit. */
1138a8599bd8SSage Weil 	if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 &&
1139a8599bd8SSage Weil 	    time_before(jiffies, ci->i_hold_caps_min)) {
1140a8599bd8SSage Weil 		dout(" delaying issued %s -> %s, wanted %s -> %s on send\n",
1141a8599bd8SSage Weil 		     ceph_cap_string(cap->issued),
1142a8599bd8SSage Weil 		     ceph_cap_string(cap->issued & retain),
1143a8599bd8SSage Weil 		     ceph_cap_string(cap->mds_wanted),
1144a8599bd8SSage Weil 		     ceph_cap_string(want));
1145a8599bd8SSage Weil 		want |= cap->mds_wanted;
1146a8599bd8SSage Weil 		retain |= cap->issued;
1147a8599bd8SSage Weil 		delayed = 1;
1148a8599bd8SSage Weil 	}
1149a8599bd8SSage Weil 	ci->i_ceph_flags &= ~(CEPH_I_NODELAY | CEPH_I_FLUSH);
1150a8599bd8SSage Weil 
1151a8599bd8SSage Weil 	cap->issued &= retain;  /* drop bits we don't want */
1152a8599bd8SSage Weil 	if (cap->implemented & ~cap->issued) {
1153a8599bd8SSage Weil 		/*
1154a8599bd8SSage Weil 		 * Wake up any waiters on wanted -> needed transition.
1155a8599bd8SSage Weil 		 * This is due to the weird transition from buffered
1156a8599bd8SSage Weil 		 * to sync IO... we need to flush dirty pages _before_
1157a8599bd8SSage Weil 		 * allowing sync writes to avoid reordering.
1158a8599bd8SSage Weil 		 */
1159a8599bd8SSage Weil 		wake = 1;
1160a8599bd8SSage Weil 	}
1161a8599bd8SSage Weil 	cap->implemented &= cap->issued | used;
1162a8599bd8SSage Weil 	cap->mds_wanted = want;
1163a8599bd8SSage Weil 
1164a8599bd8SSage Weil 	if (flushing) {
1165a8599bd8SSage Weil 		/*
1166a8599bd8SSage Weil 		 * assign a tid for flush operations so we can avoid
1167a8599bd8SSage Weil 		 * flush1 -> dirty1 -> flush2 -> flushack1 -> mark
1168a8599bd8SSage Weil 		 * clean type races.  track latest tid for every bit
1169a8599bd8SSage Weil 		 * so we can handle flush AxFw, flush Fw, and have the
1170a8599bd8SSage Weil 		 * first ack clean Ax.
1171a8599bd8SSage Weil 		 */
1172a8599bd8SSage Weil 		flush_tid = ++ci->i_cap_flush_last_tid;
1173a8599bd8SSage Weil 		if (pflush_tid)
1174a8599bd8SSage Weil 			*pflush_tid = flush_tid;
1175a8599bd8SSage Weil 		dout(" cap_flush_tid %d\n", (int)flush_tid);
1176a8599bd8SSage Weil 		for (i = 0; i < CEPH_CAP_BITS; i++)
1177a8599bd8SSage Weil 			if (flushing & (1 << i))
1178a8599bd8SSage Weil 				ci->i_cap_flush_tid[i] = flush_tid;
11797d8cb26dSSage Weil 
11807d8cb26dSSage Weil 		follows = ci->i_head_snapc->seq;
11817d8cb26dSSage Weil 	} else {
11827d8cb26dSSage Weil 		follows = 0;
1183a8599bd8SSage Weil 	}
1184a8599bd8SSage Weil 
1185a8599bd8SSage Weil 	keep = cap->implemented;
1186a8599bd8SSage Weil 	seq = cap->seq;
1187a8599bd8SSage Weil 	issue_seq = cap->issue_seq;
1188a8599bd8SSage Weil 	mseq = cap->mseq;
1189a8599bd8SSage Weil 	size = inode->i_size;
1190a8599bd8SSage Weil 	ci->i_reported_size = size;
1191a8599bd8SSage Weil 	max_size = ci->i_wanted_max_size;
1192a8599bd8SSage Weil 	ci->i_requested_max_size = max_size;
1193a8599bd8SSage Weil 	mtime = inode->i_mtime;
1194a8599bd8SSage Weil 	atime = inode->i_atime;
1195a8599bd8SSage Weil 	time_warp_seq = ci->i_time_warp_seq;
1196a8599bd8SSage Weil 	uid = inode->i_uid;
1197a8599bd8SSage Weil 	gid = inode->i_gid;
1198a8599bd8SSage Weil 	mode = inode->i_mode;
1199a8599bd8SSage Weil 
1200082afec9SSage Weil 	if (flushing & CEPH_CAP_XATTR_EXCL) {
1201a8599bd8SSage Weil 		__ceph_build_xattrs_blob(ci);
1202082afec9SSage Weil 		xattr_blob = ci->i_xattrs.blob;
1203082afec9SSage Weil 		xattr_version = ci->i_xattrs.version;
1204a8599bd8SSage Weil 	}
1205a8599bd8SSage Weil 
1206be655596SSage Weil 	spin_unlock(&ci->i_ceph_lock);
1207a8599bd8SSage Weil 
1208a8599bd8SSage Weil 	ret = send_cap_msg(session, ceph_vino(inode).ino, cap_id,
1209a8599bd8SSage Weil 		op, keep, want, flushing, seq, flush_tid, issue_seq, mseq,
1210a8599bd8SSage Weil 		size, max_size, &mtime, &atime, time_warp_seq,
1211082afec9SSage Weil 		uid, gid, mode, xattr_version, xattr_blob,
1212a8599bd8SSage Weil 		follows);
1213a8599bd8SSage Weil 	if (ret < 0) {
1214a8599bd8SSage Weil 		dout("error sending cap msg, must requeue %p\n", inode);
1215a8599bd8SSage Weil 		delayed = 1;
1216a8599bd8SSage Weil 	}
1217a8599bd8SSage Weil 
1218a8599bd8SSage Weil 	if (wake)
121903066f23SYehuda Sadeh 		wake_up_all(&ci->i_cap_wq);
1220a8599bd8SSage Weil 
1221a8599bd8SSage Weil 	return delayed;
1222a8599bd8SSage Weil }
1223a8599bd8SSage Weil 
1224a8599bd8SSage Weil /*
1225a8599bd8SSage Weil  * When a snapshot is taken, clients accumulate dirty metadata on
1226a8599bd8SSage Weil  * inodes with capabilities in ceph_cap_snaps to describe the file
1227a8599bd8SSage Weil  * state at the time the snapshot was taken.  This must be flushed
1228a8599bd8SSage Weil  * asynchronously back to the MDS once sync writes complete and dirty
1229a8599bd8SSage Weil  * data is written out.
1230a8599bd8SSage Weil  *
1231e835124cSSage Weil  * Unless @again is true, skip cap_snaps that were already sent to
1232e835124cSSage Weil  * the MDS (i.e., during this session).
1233e835124cSSage Weil  *
1234be655596SSage Weil  * Called under i_ceph_lock.  Takes s_mutex as needed.
1235a8599bd8SSage Weil  */
1236a8599bd8SSage Weil void __ceph_flush_snaps(struct ceph_inode_info *ci,
1237e835124cSSage Weil 			struct ceph_mds_session **psession,
1238e835124cSSage Weil 			int again)
1239be655596SSage Weil 		__releases(ci->i_ceph_lock)
1240be655596SSage Weil 		__acquires(ci->i_ceph_lock)
1241a8599bd8SSage Weil {
1242a8599bd8SSage Weil 	struct inode *inode = &ci->vfs_inode;
1243a8599bd8SSage Weil 	int mds;
1244a8599bd8SSage Weil 	struct ceph_cap_snap *capsnap;
1245a8599bd8SSage Weil 	u32 mseq;
12463d14c5d2SYehuda Sadeh 	struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
1247a8599bd8SSage Weil 	struct ceph_mds_session *session = NULL; /* if session != NULL, we hold
1248a8599bd8SSage Weil 						    session->s_mutex */
1249a8599bd8SSage Weil 	u64 next_follows = 0;  /* keep track of how far we've gotten through the
1250a8599bd8SSage Weil 			     i_cap_snaps list, and skip these entries next time
1251a8599bd8SSage Weil 			     around to avoid an infinite loop */
1252a8599bd8SSage Weil 
1253a8599bd8SSage Weil 	if (psession)
1254a8599bd8SSage Weil 		session = *psession;
1255a8599bd8SSage Weil 
1256a8599bd8SSage Weil 	dout("__flush_snaps %p\n", inode);
1257a8599bd8SSage Weil retry:
1258a8599bd8SSage Weil 	list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
1259a8599bd8SSage Weil 		/* avoid an infiniute loop after retry */
1260a8599bd8SSage Weil 		if (capsnap->follows < next_follows)
1261a8599bd8SSage Weil 			continue;
1262a8599bd8SSage Weil 		/*
1263a8599bd8SSage Weil 		 * we need to wait for sync writes to complete and for dirty
1264a8599bd8SSage Weil 		 * pages to be written out.
1265a8599bd8SSage Weil 		 */
1266a8599bd8SSage Weil 		if (capsnap->dirty_pages || capsnap->writing)
1267cfc0bf66SSage Weil 			break;
1268a8599bd8SSage Weil 
1269819ccbfaSSage Weil 		/*
1270819ccbfaSSage Weil 		 * if cap writeback already occurred, we should have dropped
1271819ccbfaSSage Weil 		 * the capsnap in ceph_put_wrbuffer_cap_refs.
1272819ccbfaSSage Weil 		 */
1273819ccbfaSSage Weil 		BUG_ON(capsnap->dirty == 0);
1274819ccbfaSSage Weil 
1275a8599bd8SSage Weil 		/* pick mds, take s_mutex */
1276ca81f3f6SSage Weil 		if (ci->i_auth_cap == NULL) {
1277ca81f3f6SSage Weil 			dout("no auth cap (migrating?), doing nothing\n");
1278ca81f3f6SSage Weil 			goto out;
1279ca81f3f6SSage Weil 		}
1280e835124cSSage Weil 
1281e835124cSSage Weil 		/* only flush each capsnap once */
1282e835124cSSage Weil 		if (!again && !list_empty(&capsnap->flushing_item)) {
1283e835124cSSage Weil 			dout("already flushed %p, skipping\n", capsnap);
1284e835124cSSage Weil 			continue;
1285e835124cSSage Weil 		}
1286e835124cSSage Weil 
1287ca81f3f6SSage Weil 		mds = ci->i_auth_cap->session->s_mds;
1288ca81f3f6SSage Weil 		mseq = ci->i_auth_cap->mseq;
1289ca81f3f6SSage Weil 
1290a8599bd8SSage Weil 		if (session && session->s_mds != mds) {
1291a8599bd8SSage Weil 			dout("oops, wrong session %p mutex\n", session);
1292a8599bd8SSage Weil 			mutex_unlock(&session->s_mutex);
1293a8599bd8SSage Weil 			ceph_put_mds_session(session);
1294a8599bd8SSage Weil 			session = NULL;
1295a8599bd8SSage Weil 		}
1296a8599bd8SSage Weil 		if (!session) {
1297be655596SSage Weil 			spin_unlock(&ci->i_ceph_lock);
1298a8599bd8SSage Weil 			mutex_lock(&mdsc->mutex);
1299a8599bd8SSage Weil 			session = __ceph_lookup_mds_session(mdsc, mds);
1300a8599bd8SSage Weil 			mutex_unlock(&mdsc->mutex);
1301a8599bd8SSage Weil 			if (session) {
1302a8599bd8SSage Weil 				dout("inverting session/ino locks on %p\n",
1303a8599bd8SSage Weil 				     session);
1304a8599bd8SSage Weil 				mutex_lock(&session->s_mutex);
1305a8599bd8SSage Weil 			}
1306a8599bd8SSage Weil 			/*
1307a8599bd8SSage Weil 			 * if session == NULL, we raced against a cap
1308ca81f3f6SSage Weil 			 * deletion or migration.  retry, and we'll
1309ca81f3f6SSage Weil 			 * get a better @mds value next time.
1310a8599bd8SSage Weil 			 */
1311be655596SSage Weil 			spin_lock(&ci->i_ceph_lock);
1312a8599bd8SSage Weil 			goto retry;
1313a8599bd8SSage Weil 		}
1314a8599bd8SSage Weil 
1315a8599bd8SSage Weil 		capsnap->flush_tid = ++ci->i_cap_flush_last_tid;
1316a8599bd8SSage Weil 		atomic_inc(&capsnap->nref);
1317a8599bd8SSage Weil 		if (!list_empty(&capsnap->flushing_item))
1318a8599bd8SSage Weil 			list_del_init(&capsnap->flushing_item);
1319a8599bd8SSage Weil 		list_add_tail(&capsnap->flushing_item,
1320a8599bd8SSage Weil 			      &session->s_cap_snaps_flushing);
1321be655596SSage Weil 		spin_unlock(&ci->i_ceph_lock);
1322a8599bd8SSage Weil 
1323cfc0bf66SSage Weil 		dout("flush_snaps %p cap_snap %p follows %lld tid %llu\n",
1324cfc0bf66SSage Weil 		     inode, capsnap, capsnap->follows, capsnap->flush_tid);
1325a8599bd8SSage Weil 		send_cap_msg(session, ceph_vino(inode).ino, 0,
1326a8599bd8SSage Weil 			     CEPH_CAP_OP_FLUSHSNAP, capsnap->issued, 0,
1327a8599bd8SSage Weil 			     capsnap->dirty, 0, capsnap->flush_tid, 0, mseq,
1328a8599bd8SSage Weil 			     capsnap->size, 0,
1329a8599bd8SSage Weil 			     &capsnap->mtime, &capsnap->atime,
1330a8599bd8SSage Weil 			     capsnap->time_warp_seq,
1331a8599bd8SSage Weil 			     capsnap->uid, capsnap->gid, capsnap->mode,
13324a625be4SSage Weil 			     capsnap->xattr_version, capsnap->xattr_blob,
1333a8599bd8SSage Weil 			     capsnap->follows);
1334a8599bd8SSage Weil 
1335a8599bd8SSage Weil 		next_follows = capsnap->follows + 1;
1336a8599bd8SSage Weil 		ceph_put_cap_snap(capsnap);
1337a8599bd8SSage Weil 
1338be655596SSage Weil 		spin_lock(&ci->i_ceph_lock);
1339a8599bd8SSage Weil 		goto retry;
1340a8599bd8SSage Weil 	}
1341a8599bd8SSage Weil 
1342a8599bd8SSage Weil 	/* we flushed them all; remove this inode from the queue */
1343a8599bd8SSage Weil 	spin_lock(&mdsc->snap_flush_lock);
1344a8599bd8SSage Weil 	list_del_init(&ci->i_snap_flush_item);
1345a8599bd8SSage Weil 	spin_unlock(&mdsc->snap_flush_lock);
1346a8599bd8SSage Weil 
1347ca81f3f6SSage Weil out:
1348a8599bd8SSage Weil 	if (psession)
1349a8599bd8SSage Weil 		*psession = session;
1350a8599bd8SSage Weil 	else if (session) {
1351a8599bd8SSage Weil 		mutex_unlock(&session->s_mutex);
1352a8599bd8SSage Weil 		ceph_put_mds_session(session);
1353a8599bd8SSage Weil 	}
1354a8599bd8SSage Weil }
1355a8599bd8SSage Weil 
1356a8599bd8SSage Weil static void ceph_flush_snaps(struct ceph_inode_info *ci)
1357a8599bd8SSage Weil {
1358be655596SSage Weil 	spin_lock(&ci->i_ceph_lock);
1359e835124cSSage Weil 	__ceph_flush_snaps(ci, NULL, 0);
1360be655596SSage Weil 	spin_unlock(&ci->i_ceph_lock);
1361a8599bd8SSage Weil }
1362a8599bd8SSage Weil 
1363a8599bd8SSage Weil /*
1364fca65b4aSSage Weil  * Mark caps dirty.  If inode is newly dirty, return the dirty flags.
1365fca65b4aSSage Weil  * Caller is then responsible for calling __mark_inode_dirty with the
1366fca65b4aSSage Weil  * returned flags value.
136776e3b390SSage Weil  */
1368fca65b4aSSage Weil int __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask)
136976e3b390SSage Weil {
1370640ef79dSCheng Renquan 	struct ceph_mds_client *mdsc =
13713d14c5d2SYehuda Sadeh 		ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
137276e3b390SSage Weil 	struct inode *inode = &ci->vfs_inode;
137376e3b390SSage Weil 	int was = ci->i_dirty_caps;
137476e3b390SSage Weil 	int dirty = 0;
137576e3b390SSage Weil 
137676e3b390SSage Weil 	dout("__mark_dirty_caps %p %s dirty %s -> %s\n", &ci->vfs_inode,
137776e3b390SSage Weil 	     ceph_cap_string(mask), ceph_cap_string(was),
137876e3b390SSage Weil 	     ceph_cap_string(was | mask));
137976e3b390SSage Weil 	ci->i_dirty_caps |= mask;
138076e3b390SSage Weil 	if (was == 0) {
13817d8cb26dSSage Weil 		if (!ci->i_head_snapc)
13827d8cb26dSSage Weil 			ci->i_head_snapc = ceph_get_snap_context(
13837d8cb26dSSage Weil 				ci->i_snap_realm->cached_context);
13840685235fSYan, Zheng 		dout(" inode %p now dirty snapc %p auth cap %p\n",
13850685235fSYan, Zheng 		     &ci->vfs_inode, ci->i_head_snapc, ci->i_auth_cap);
138676e3b390SSage Weil 		BUG_ON(!list_empty(&ci->i_dirty_item));
138776e3b390SSage Weil 		spin_lock(&mdsc->cap_dirty_lock);
13880685235fSYan, Zheng 		if (ci->i_auth_cap)
138976e3b390SSage Weil 			list_add(&ci->i_dirty_item, &mdsc->cap_dirty);
13900685235fSYan, Zheng 		else
13910685235fSYan, Zheng 			list_add(&ci->i_dirty_item,
13920685235fSYan, Zheng 				 &mdsc->cap_dirty_migrating);
139376e3b390SSage Weil 		spin_unlock(&mdsc->cap_dirty_lock);
139476e3b390SSage Weil 		if (ci->i_flushing_caps == 0) {
13953772d26dSSage Weil 			ihold(inode);
139676e3b390SSage Weil 			dirty |= I_DIRTY_SYNC;
139776e3b390SSage Weil 		}
139876e3b390SSage Weil 	}
139976e3b390SSage Weil 	BUG_ON(list_empty(&ci->i_dirty_item));
140076e3b390SSage Weil 	if (((was | ci->i_flushing_caps) & CEPH_CAP_FILE_BUFFER) &&
140176e3b390SSage Weil 	    (mask & CEPH_CAP_FILE_BUFFER))
140276e3b390SSage Weil 		dirty |= I_DIRTY_DATASYNC;
140376e3b390SSage Weil 	__cap_delay_requeue(mdsc, ci);
1404fca65b4aSSage Weil 	return dirty;
140576e3b390SSage Weil }
140676e3b390SSage Weil 
140776e3b390SSage Weil /*
1408a8599bd8SSage Weil  * Add dirty inode to the flushing list.  Assigned a seq number so we
1409a8599bd8SSage Weil  * can wait for caps to flush without starving.
1410cdc35f96SSage Weil  *
1411be655596SSage Weil  * Called under i_ceph_lock.
1412a8599bd8SSage Weil  */
1413cdc35f96SSage Weil static int __mark_caps_flushing(struct inode *inode,
1414a8599bd8SSage Weil 				 struct ceph_mds_session *session)
1415a8599bd8SSage Weil {
14163d14c5d2SYehuda Sadeh 	struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1417a8599bd8SSage Weil 	struct ceph_inode_info *ci = ceph_inode(inode);
1418cdc35f96SSage Weil 	int flushing;
1419a8599bd8SSage Weil 
1420cdc35f96SSage Weil 	BUG_ON(ci->i_dirty_caps == 0);
1421a8599bd8SSage Weil 	BUG_ON(list_empty(&ci->i_dirty_item));
1422cdc35f96SSage Weil 
1423cdc35f96SSage Weil 	flushing = ci->i_dirty_caps;
1424cdc35f96SSage Weil 	dout("__mark_caps_flushing flushing %s, flushing_caps %s -> %s\n",
1425cdc35f96SSage Weil 	     ceph_cap_string(flushing),
1426cdc35f96SSage Weil 	     ceph_cap_string(ci->i_flushing_caps),
1427cdc35f96SSage Weil 	     ceph_cap_string(ci->i_flushing_caps | flushing));
1428cdc35f96SSage Weil 	ci->i_flushing_caps |= flushing;
1429cdc35f96SSage Weil 	ci->i_dirty_caps = 0;
1430afcdaea3SSage Weil 	dout(" inode %p now !dirty\n", inode);
1431cdc35f96SSage Weil 
1432a8599bd8SSage Weil 	spin_lock(&mdsc->cap_dirty_lock);
1433cdc35f96SSage Weil 	list_del_init(&ci->i_dirty_item);
1434afcdaea3SSage Weil 
1435afcdaea3SSage Weil 	ci->i_cap_flush_seq = ++mdsc->cap_flush_seq;
1436afcdaea3SSage Weil 	if (list_empty(&ci->i_flushing_item)) {
1437a8599bd8SSage Weil 		list_add_tail(&ci->i_flushing_item, &session->s_cap_flushing);
1438a8599bd8SSage Weil 		mdsc->num_cap_flushing++;
1439afcdaea3SSage Weil 		dout(" inode %p now flushing seq %lld\n", inode,
1440afcdaea3SSage Weil 		     ci->i_cap_flush_seq);
1441afcdaea3SSage Weil 	} else {
1442afcdaea3SSage Weil 		list_move_tail(&ci->i_flushing_item, &session->s_cap_flushing);
1443afcdaea3SSage Weil 		dout(" inode %p now flushing (more) seq %lld\n", inode,
1444a8599bd8SSage Weil 		     ci->i_cap_flush_seq);
1445a8599bd8SSage Weil 	}
1446a8599bd8SSage Weil 	spin_unlock(&mdsc->cap_dirty_lock);
1447cdc35f96SSage Weil 
1448cdc35f96SSage Weil 	return flushing;
1449a8599bd8SSage Weil }
1450a8599bd8SSage Weil 
1451a8599bd8SSage Weil /*
14525ecad6fdSSage Weil  * try to invalidate mapping pages without blocking.
14535ecad6fdSSage Weil  */
14545ecad6fdSSage Weil static int try_nonblocking_invalidate(struct inode *inode)
14555ecad6fdSSage Weil {
14565ecad6fdSSage Weil 	struct ceph_inode_info *ci = ceph_inode(inode);
14575ecad6fdSSage Weil 	u32 invalidating_gen = ci->i_rdcache_gen;
14585ecad6fdSSage Weil 
1459be655596SSage Weil 	spin_unlock(&ci->i_ceph_lock);
14605ecad6fdSSage Weil 	invalidate_mapping_pages(&inode->i_data, 0, -1);
1461be655596SSage Weil 	spin_lock(&ci->i_ceph_lock);
14625ecad6fdSSage Weil 
146318a38193SSage Weil 	if (inode->i_data.nrpages == 0 &&
14645ecad6fdSSage Weil 	    invalidating_gen == ci->i_rdcache_gen) {
14655ecad6fdSSage Weil 		/* success. */
14665ecad6fdSSage Weil 		dout("try_nonblocking_invalidate %p success\n", inode);
1467cd045cb4SSage Weil 		/* save any racing async invalidate some trouble */
1468cd045cb4SSage Weil 		ci->i_rdcache_revoking = ci->i_rdcache_gen - 1;
14695ecad6fdSSage Weil 		return 0;
14705ecad6fdSSage Weil 	}
14715ecad6fdSSage Weil 	dout("try_nonblocking_invalidate %p failed\n", inode);
14725ecad6fdSSage Weil 	return -1;
14735ecad6fdSSage Weil }
14745ecad6fdSSage Weil 
14755ecad6fdSSage Weil /*
1476a8599bd8SSage Weil  * Swiss army knife function to examine currently used and wanted
1477a8599bd8SSage Weil  * versus held caps.  Release, flush, ack revoked caps to mds as
1478a8599bd8SSage Weil  * appropriate.
1479a8599bd8SSage Weil  *
1480a8599bd8SSage Weil  *  CHECK_CAPS_NODELAY - caller is delayed work and we should not delay
1481a8599bd8SSage Weil  *    cap release further.
1482a8599bd8SSage Weil  *  CHECK_CAPS_AUTHONLY - we should only check the auth cap
1483a8599bd8SSage Weil  *  CHECK_CAPS_FLUSH - we should flush any dirty caps immediately, without
1484a8599bd8SSage Weil  *    further delay.
1485a8599bd8SSage Weil  */
1486a8599bd8SSage Weil void ceph_check_caps(struct ceph_inode_info *ci, int flags,
1487a8599bd8SSage Weil 		     struct ceph_mds_session *session)
1488a8599bd8SSage Weil {
14893d14c5d2SYehuda Sadeh 	struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode);
14903d14c5d2SYehuda Sadeh 	struct ceph_mds_client *mdsc = fsc->mdsc;
1491a8599bd8SSage Weil 	struct inode *inode = &ci->vfs_inode;
1492a8599bd8SSage Weil 	struct ceph_cap *cap;
1493395c312bSYan, Zheng 	int file_wanted, used, cap_used;
1494a8599bd8SSage Weil 	int took_snap_rwsem = 0;             /* true if mdsc->snap_rwsem held */
1495cbd03635SSage Weil 	int issued, implemented, want, retain, revoking, flushing = 0;
1496a8599bd8SSage Weil 	int mds = -1;   /* keep track of how far we've gone through i_caps list
1497a8599bd8SSage Weil 			   to avoid an infinite loop on retry */
1498a8599bd8SSage Weil 	struct rb_node *p;
1499a8599bd8SSage Weil 	int tried_invalidate = 0;
1500a8599bd8SSage Weil 	int delayed = 0, sent = 0, force_requeue = 0, num;
1501cbd03635SSage Weil 	int queue_invalidate = 0;
1502a8599bd8SSage Weil 	int is_delayed = flags & CHECK_CAPS_NODELAY;
1503a8599bd8SSage Weil 
1504a8599bd8SSage Weil 	/* if we are unmounting, flush any unused caps immediately. */
1505a8599bd8SSage Weil 	if (mdsc->stopping)
1506a8599bd8SSage Weil 		is_delayed = 1;
1507a8599bd8SSage Weil 
1508be655596SSage Weil 	spin_lock(&ci->i_ceph_lock);
1509a8599bd8SSage Weil 
1510a8599bd8SSage Weil 	if (ci->i_ceph_flags & CEPH_I_FLUSH)
1511a8599bd8SSage Weil 		flags |= CHECK_CAPS_FLUSH;
1512a8599bd8SSage Weil 
1513a8599bd8SSage Weil 	/* flush snaps first time around only */
1514a8599bd8SSage Weil 	if (!list_empty(&ci->i_cap_snaps))
1515e835124cSSage Weil 		__ceph_flush_snaps(ci, &session, 0);
1516a8599bd8SSage Weil 	goto retry_locked;
1517a8599bd8SSage Weil retry:
1518be655596SSage Weil 	spin_lock(&ci->i_ceph_lock);
1519a8599bd8SSage Weil retry_locked:
1520a8599bd8SSage Weil 	file_wanted = __ceph_caps_file_wanted(ci);
1521a8599bd8SSage Weil 	used = __ceph_caps_used(ci);
1522a8599bd8SSage Weil 	want = file_wanted | used;
1523cbd03635SSage Weil 	issued = __ceph_caps_issued(ci, &implemented);
1524cbd03635SSage Weil 	revoking = implemented & ~issued;
1525a8599bd8SSage Weil 
1526a8599bd8SSage Weil 	retain = want | CEPH_CAP_PIN;
1527a8599bd8SSage Weil 	if (!mdsc->stopping && inode->i_nlink > 0) {
1528a8599bd8SSage Weil 		if (want) {
1529a8599bd8SSage Weil 			retain |= CEPH_CAP_ANY;       /* be greedy */
1530a8599bd8SSage Weil 		} else {
1531a8599bd8SSage Weil 			retain |= CEPH_CAP_ANY_SHARED;
1532a8599bd8SSage Weil 			/*
1533a8599bd8SSage Weil 			 * keep RD only if we didn't have the file open RW,
1534a8599bd8SSage Weil 			 * because then the mds would revoke it anyway to
1535a8599bd8SSage Weil 			 * journal max_size=0.
1536a8599bd8SSage Weil 			 */
1537a8599bd8SSage Weil 			if (ci->i_max_size == 0)
1538a8599bd8SSage Weil 				retain |= CEPH_CAP_ANY_RD;
1539a8599bd8SSage Weil 		}
1540a8599bd8SSage Weil 	}
1541a8599bd8SSage Weil 
1542a8599bd8SSage Weil 	dout("check_caps %p file_want %s used %s dirty %s flushing %s"
1543cbd03635SSage Weil 	     " issued %s revoking %s retain %s %s%s%s\n", inode,
1544a8599bd8SSage Weil 	     ceph_cap_string(file_wanted),
1545a8599bd8SSage Weil 	     ceph_cap_string(used), ceph_cap_string(ci->i_dirty_caps),
1546a8599bd8SSage Weil 	     ceph_cap_string(ci->i_flushing_caps),
1547cbd03635SSage Weil 	     ceph_cap_string(issued), ceph_cap_string(revoking),
1548a8599bd8SSage Weil 	     ceph_cap_string(retain),
1549a8599bd8SSage Weil 	     (flags & CHECK_CAPS_AUTHONLY) ? " AUTHONLY" : "",
1550a8599bd8SSage Weil 	     (flags & CHECK_CAPS_NODELAY) ? " NODELAY" : "",
1551a8599bd8SSage Weil 	     (flags & CHECK_CAPS_FLUSH) ? " FLUSH" : "");
1552a8599bd8SSage Weil 
1553a8599bd8SSage Weil 	/*
1554a8599bd8SSage Weil 	 * If we no longer need to hold onto old our caps, and we may
1555a8599bd8SSage Weil 	 * have cached pages, but don't want them, then try to invalidate.
1556a8599bd8SSage Weil 	 * If we fail, it's because pages are locked.... try again later.
1557a8599bd8SSage Weil 	 */
1558a8599bd8SSage Weil 	if ((!is_delayed || mdsc->stopping) &&
1559a8599bd8SSage Weil 	    ci->i_wrbuffer_ref == 0 &&               /* no dirty pages... */
156093afd449SSage Weil 	    inode->i_data.nrpages &&                 /* have cached pages */
1561cbd03635SSage Weil 	    (file_wanted == 0 ||                     /* no open files */
15622962507cSSage Weil 	     (revoking & (CEPH_CAP_FILE_CACHE|
15632962507cSSage Weil 			  CEPH_CAP_FILE_LAZYIO))) && /*  or revoking cache */
1564a8599bd8SSage Weil 	    !tried_invalidate) {
1565a8599bd8SSage Weil 		dout("check_caps trying to invalidate on %p\n", inode);
15665ecad6fdSSage Weil 		if (try_nonblocking_invalidate(inode) < 0) {
15672962507cSSage Weil 			if (revoking & (CEPH_CAP_FILE_CACHE|
15682962507cSSage Weil 					CEPH_CAP_FILE_LAZYIO)) {
1569cbd03635SSage Weil 				dout("check_caps queuing invalidate\n");
1570cbd03635SSage Weil 				queue_invalidate = 1;
1571cbd03635SSage Weil 				ci->i_rdcache_revoking = ci->i_rdcache_gen;
1572a8599bd8SSage Weil 			} else {
1573a8599bd8SSage Weil 				dout("check_caps failed to invalidate pages\n");
1574a8599bd8SSage Weil 				/* we failed to invalidate pages.  check these
1575a8599bd8SSage Weil 				   caps again later. */
1576a8599bd8SSage Weil 				force_requeue = 1;
1577a8599bd8SSage Weil 				__cap_set_timeouts(mdsc, ci);
1578a8599bd8SSage Weil 			}
15795ecad6fdSSage Weil 		}
1580a8599bd8SSage Weil 		tried_invalidate = 1;
1581a8599bd8SSage Weil 		goto retry_locked;
1582a8599bd8SSage Weil 	}
1583a8599bd8SSage Weil 
1584a8599bd8SSage Weil 	num = 0;
1585a8599bd8SSage Weil 	for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
1586a8599bd8SSage Weil 		cap = rb_entry(p, struct ceph_cap, ci_node);
1587a8599bd8SSage Weil 		num++;
1588a8599bd8SSage Weil 
1589a8599bd8SSage Weil 		/* avoid looping forever */
1590a8599bd8SSage Weil 		if (mds >= cap->mds ||
1591a8599bd8SSage Weil 		    ((flags & CHECK_CAPS_AUTHONLY) && cap != ci->i_auth_cap))
1592a8599bd8SSage Weil 			continue;
1593a8599bd8SSage Weil 
1594a8599bd8SSage Weil 		/* NOTE: no side-effects allowed, until we take s_mutex */
1595a8599bd8SSage Weil 
1596395c312bSYan, Zheng 		cap_used = used;
1597395c312bSYan, Zheng 		if (ci->i_auth_cap && cap != ci->i_auth_cap)
1598395c312bSYan, Zheng 			cap_used &= ~ci->i_auth_cap->issued;
1599395c312bSYan, Zheng 
1600a8599bd8SSage Weil 		revoking = cap->implemented & ~cap->issued;
1601395c312bSYan, Zheng 		dout(" mds%d cap %p used %s issued %s implemented %s revoking %s\n",
1602088b3f5eSSage Weil 		     cap->mds, cap, ceph_cap_string(cap->issued),
1603395c312bSYan, Zheng 		     ceph_cap_string(cap_used),
1604088b3f5eSSage Weil 		     ceph_cap_string(cap->implemented),
1605a8599bd8SSage Weil 		     ceph_cap_string(revoking));
1606a8599bd8SSage Weil 
1607a8599bd8SSage Weil 		if (cap == ci->i_auth_cap &&
1608a8599bd8SSage Weil 		    (cap->issued & CEPH_CAP_FILE_WR)) {
1609a8599bd8SSage Weil 			/* request larger max_size from MDS? */
1610a8599bd8SSage Weil 			if (ci->i_wanted_max_size > ci->i_max_size &&
1611a8599bd8SSage Weil 			    ci->i_wanted_max_size > ci->i_requested_max_size) {
1612a8599bd8SSage Weil 				dout("requesting new max_size\n");
1613a8599bd8SSage Weil 				goto ack;
1614a8599bd8SSage Weil 			}
1615a8599bd8SSage Weil 
1616a8599bd8SSage Weil 			/* approaching file_max? */
1617a8599bd8SSage Weil 			if ((inode->i_size << 1) >= ci->i_max_size &&
1618a8599bd8SSage Weil 			    (ci->i_reported_size << 1) < ci->i_max_size) {
1619a8599bd8SSage Weil 				dout("i_size approaching max_size\n");
1620a8599bd8SSage Weil 				goto ack;
1621a8599bd8SSage Weil 			}
1622a8599bd8SSage Weil 		}
1623a8599bd8SSage Weil 		/* flush anything dirty? */
1624a8599bd8SSage Weil 		if (cap == ci->i_auth_cap && (flags & CHECK_CAPS_FLUSH) &&
1625a8599bd8SSage Weil 		    ci->i_dirty_caps) {
1626a8599bd8SSage Weil 			dout("flushing dirty caps\n");
1627a8599bd8SSage Weil 			goto ack;
1628a8599bd8SSage Weil 		}
1629a8599bd8SSage Weil 
1630a8599bd8SSage Weil 		/* completed revocation? going down and there are no caps? */
1631395c312bSYan, Zheng 		if (revoking && (revoking & cap_used) == 0) {
1632a8599bd8SSage Weil 			dout("completed revocation of %s\n",
1633a8599bd8SSage Weil 			     ceph_cap_string(cap->implemented & ~cap->issued));
1634a8599bd8SSage Weil 			goto ack;
1635a8599bd8SSage Weil 		}
1636a8599bd8SSage Weil 
1637a8599bd8SSage Weil 		/* want more caps from mds? */
1638a8599bd8SSage Weil 		if (want & ~(cap->mds_wanted | cap->issued))
1639a8599bd8SSage Weil 			goto ack;
1640a8599bd8SSage Weil 
1641a8599bd8SSage Weil 		/* things we might delay */
1642a8599bd8SSage Weil 		if ((cap->issued & ~retain) == 0 &&
1643a8599bd8SSage Weil 		    cap->mds_wanted == want)
1644a8599bd8SSage Weil 			continue;     /* nope, all good */
1645a8599bd8SSage Weil 
1646a8599bd8SSage Weil 		if (is_delayed)
1647a8599bd8SSage Weil 			goto ack;
1648a8599bd8SSage Weil 
1649a8599bd8SSage Weil 		/* delay? */
1650a8599bd8SSage Weil 		if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 &&
1651a8599bd8SSage Weil 		    time_before(jiffies, ci->i_hold_caps_max)) {
1652a8599bd8SSage Weil 			dout(" delaying issued %s -> %s, wanted %s -> %s\n",
1653a8599bd8SSage Weil 			     ceph_cap_string(cap->issued),
1654a8599bd8SSage Weil 			     ceph_cap_string(cap->issued & retain),
1655a8599bd8SSage Weil 			     ceph_cap_string(cap->mds_wanted),
1656a8599bd8SSage Weil 			     ceph_cap_string(want));
1657a8599bd8SSage Weil 			delayed++;
1658a8599bd8SSage Weil 			continue;
1659a8599bd8SSage Weil 		}
1660a8599bd8SSage Weil 
1661a8599bd8SSage Weil ack:
1662e9964c10SSage Weil 		if (ci->i_ceph_flags & CEPH_I_NOFLUSH) {
1663e9964c10SSage Weil 			dout(" skipping %p I_NOFLUSH set\n", inode);
1664e9964c10SSage Weil 			continue;
1665e9964c10SSage Weil 		}
1666e9964c10SSage Weil 
1667a8599bd8SSage Weil 		if (session && session != cap->session) {
1668a8599bd8SSage Weil 			dout("oops, wrong session %p mutex\n", session);
1669a8599bd8SSage Weil 			mutex_unlock(&session->s_mutex);
1670a8599bd8SSage Weil 			session = NULL;
1671a8599bd8SSage Weil 		}
1672a8599bd8SSage Weil 		if (!session) {
1673a8599bd8SSage Weil 			session = cap->session;
1674a8599bd8SSage Weil 			if (mutex_trylock(&session->s_mutex) == 0) {
1675a8599bd8SSage Weil 				dout("inverting session/ino locks on %p\n",
1676a8599bd8SSage Weil 				     session);
1677be655596SSage Weil 				spin_unlock(&ci->i_ceph_lock);
1678a8599bd8SSage Weil 				if (took_snap_rwsem) {
1679a8599bd8SSage Weil 					up_read(&mdsc->snap_rwsem);
1680a8599bd8SSage Weil 					took_snap_rwsem = 0;
1681a8599bd8SSage Weil 				}
1682a8599bd8SSage Weil 				mutex_lock(&session->s_mutex);
1683a8599bd8SSage Weil 				goto retry;
1684a8599bd8SSage Weil 			}
1685a8599bd8SSage Weil 		}
1686a8599bd8SSage Weil 		/* take snap_rwsem after session mutex */
1687a8599bd8SSage Weil 		if (!took_snap_rwsem) {
1688a8599bd8SSage Weil 			if (down_read_trylock(&mdsc->snap_rwsem) == 0) {
1689a8599bd8SSage Weil 				dout("inverting snap/in locks on %p\n",
1690a8599bd8SSage Weil 				     inode);
1691be655596SSage Weil 				spin_unlock(&ci->i_ceph_lock);
1692a8599bd8SSage Weil 				down_read(&mdsc->snap_rwsem);
1693a8599bd8SSage Weil 				took_snap_rwsem = 1;
1694a8599bd8SSage Weil 				goto retry;
1695a8599bd8SSage Weil 			}
1696a8599bd8SSage Weil 			took_snap_rwsem = 1;
1697a8599bd8SSage Weil 		}
1698a8599bd8SSage Weil 
1699cdc35f96SSage Weil 		if (cap == ci->i_auth_cap && ci->i_dirty_caps)
1700cdc35f96SSage Weil 			flushing = __mark_caps_flushing(inode, session);
170124be0c48SSage Weil 		else
170224be0c48SSage Weil 			flushing = 0;
1703a8599bd8SSage Weil 
1704a8599bd8SSage Weil 		mds = cap->mds;  /* remember mds, so we don't repeat */
1705a8599bd8SSage Weil 		sent++;
1706a8599bd8SSage Weil 
1707be655596SSage Weil 		/* __send_cap drops i_ceph_lock */
1708395c312bSYan, Zheng 		delayed += __send_cap(mdsc, cap, CEPH_CAP_OP_UPDATE, cap_used,
1709395c312bSYan, Zheng 				      want, retain, flushing, NULL);
1710be655596SSage Weil 		goto retry; /* retake i_ceph_lock and restart our cap scan. */
1711a8599bd8SSage Weil 	}
1712a8599bd8SSage Weil 
1713a8599bd8SSage Weil 	/*
1714a8599bd8SSage Weil 	 * Reschedule delayed caps release if we delayed anything,
1715a8599bd8SSage Weil 	 * otherwise cancel.
1716a8599bd8SSage Weil 	 */
1717a8599bd8SSage Weil 	if (delayed && is_delayed)
1718a8599bd8SSage Weil 		force_requeue = 1;   /* __send_cap delayed release; requeue */
1719a8599bd8SSage Weil 	if (!delayed && !is_delayed)
1720a8599bd8SSage Weil 		__cap_delay_cancel(mdsc, ci);
1721a8599bd8SSage Weil 	else if (!is_delayed || force_requeue)
1722a8599bd8SSage Weil 		__cap_delay_requeue(mdsc, ci);
1723a8599bd8SSage Weil 
1724be655596SSage Weil 	spin_unlock(&ci->i_ceph_lock);
1725a8599bd8SSage Weil 
1726cbd03635SSage Weil 	if (queue_invalidate)
17273c6f6b79SSage Weil 		ceph_queue_invalidate(inode);
1728cbd03635SSage Weil 
1729cdc2ce05SSage Weil 	if (session)
1730a8599bd8SSage Weil 		mutex_unlock(&session->s_mutex);
1731a8599bd8SSage Weil 	if (took_snap_rwsem)
1732a8599bd8SSage Weil 		up_read(&mdsc->snap_rwsem);
1733a8599bd8SSage Weil }
1734a8599bd8SSage Weil 
1735a8599bd8SSage Weil /*
1736a8599bd8SSage Weil  * Try to flush dirty caps back to the auth mds.
1737a8599bd8SSage Weil  */
17384fe59789SYan, Zheng static int try_flush_caps(struct inode *inode, unsigned *flush_tid)
1739a8599bd8SSage Weil {
17403d14c5d2SYehuda Sadeh 	struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1741a8599bd8SSage Weil 	struct ceph_inode_info *ci = ceph_inode(inode);
1742a8599bd8SSage Weil 	int flushing = 0;
17434fe59789SYan, Zheng 	struct ceph_mds_session *session = NULL;
1744a8599bd8SSage Weil 
1745a8599bd8SSage Weil retry:
1746be655596SSage Weil 	spin_lock(&ci->i_ceph_lock);
1747e9964c10SSage Weil 	if (ci->i_ceph_flags & CEPH_I_NOFLUSH) {
1748e9964c10SSage Weil 		dout("try_flush_caps skipping %p I_NOFLUSH set\n", inode);
1749e9964c10SSage Weil 		goto out;
1750e9964c10SSage Weil 	}
1751a8599bd8SSage Weil 	if (ci->i_dirty_caps && ci->i_auth_cap) {
1752a8599bd8SSage Weil 		struct ceph_cap *cap = ci->i_auth_cap;
1753a8599bd8SSage Weil 		int used = __ceph_caps_used(ci);
1754a8599bd8SSage Weil 		int want = __ceph_caps_wanted(ci);
1755a8599bd8SSage Weil 		int delayed;
1756a8599bd8SSage Weil 
17574fe59789SYan, Zheng 		if (!session || session != cap->session) {
1758be655596SSage Weil 			spin_unlock(&ci->i_ceph_lock);
17594fe59789SYan, Zheng 			if (session)
17604fe59789SYan, Zheng 				mutex_unlock(&session->s_mutex);
1761a8599bd8SSage Weil 			session = cap->session;
1762a8599bd8SSage Weil 			mutex_lock(&session->s_mutex);
1763a8599bd8SSage Weil 			goto retry;
1764a8599bd8SSage Weil 		}
1765a8599bd8SSage Weil 		if (cap->session->s_state < CEPH_MDS_SESSION_OPEN)
1766a8599bd8SSage Weil 			goto out;
1767a8599bd8SSage Weil 
1768cdc35f96SSage Weil 		flushing = __mark_caps_flushing(inode, session);
1769a8599bd8SSage Weil 
1770be655596SSage Weil 		/* __send_cap drops i_ceph_lock */
1771a8599bd8SSage Weil 		delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, used, want,
1772a8599bd8SSage Weil 				     cap->issued | cap->implemented, flushing,
1773a8599bd8SSage Weil 				     flush_tid);
1774a8599bd8SSage Weil 		if (!delayed)
1775a8599bd8SSage Weil 			goto out_unlocked;
1776a8599bd8SSage Weil 
1777be655596SSage Weil 		spin_lock(&ci->i_ceph_lock);
1778a8599bd8SSage Weil 		__cap_delay_requeue(mdsc, ci);
1779a8599bd8SSage Weil 	}
1780a8599bd8SSage Weil out:
1781be655596SSage Weil 	spin_unlock(&ci->i_ceph_lock);
1782a8599bd8SSage Weil out_unlocked:
17834fe59789SYan, Zheng 	if (session)
1784a8599bd8SSage Weil 		mutex_unlock(&session->s_mutex);
1785a8599bd8SSage Weil 	return flushing;
1786a8599bd8SSage Weil }
1787a8599bd8SSage Weil 
1788a8599bd8SSage Weil /*
1789a8599bd8SSage Weil  * Return true if we've flushed caps through the given flush_tid.
1790a8599bd8SSage Weil  */
1791a8599bd8SSage Weil static int caps_are_flushed(struct inode *inode, unsigned tid)
1792a8599bd8SSage Weil {
1793a8599bd8SSage Weil 	struct ceph_inode_info *ci = ceph_inode(inode);
1794a5ee751cSDan Carpenter 	int i, ret = 1;
1795a8599bd8SSage Weil 
1796be655596SSage Weil 	spin_lock(&ci->i_ceph_lock);
1797a8599bd8SSage Weil 	for (i = 0; i < CEPH_CAP_BITS; i++)
1798a8599bd8SSage Weil 		if ((ci->i_flushing_caps & (1 << i)) &&
1799a8599bd8SSage Weil 		    ci->i_cap_flush_tid[i] <= tid) {
1800a8599bd8SSage Weil 			/* still flushing this bit */
1801a8599bd8SSage Weil 			ret = 0;
1802a8599bd8SSage Weil 			break;
1803a8599bd8SSage Weil 		}
1804be655596SSage Weil 	spin_unlock(&ci->i_ceph_lock);
1805a8599bd8SSage Weil 	return ret;
1806a8599bd8SSage Weil }
1807a8599bd8SSage Weil 
1808a8599bd8SSage Weil /*
1809a8599bd8SSage Weil  * Wait on any unsafe replies for the given inode.  First wait on the
1810a8599bd8SSage Weil  * newest request, and make that the upper bound.  Then, if there are
1811a8599bd8SSage Weil  * more requests, keep waiting on the oldest as long as it is still older
1812a8599bd8SSage Weil  * than the original request.
1813a8599bd8SSage Weil  */
1814a8599bd8SSage Weil static void sync_write_wait(struct inode *inode)
1815a8599bd8SSage Weil {
1816a8599bd8SSage Weil 	struct ceph_inode_info *ci = ceph_inode(inode);
1817a8599bd8SSage Weil 	struct list_head *head = &ci->i_unsafe_writes;
1818a8599bd8SSage Weil 	struct ceph_osd_request *req;
1819a8599bd8SSage Weil 	u64 last_tid;
1820a8599bd8SSage Weil 
1821a8599bd8SSage Weil 	spin_lock(&ci->i_unsafe_lock);
1822a8599bd8SSage Weil 	if (list_empty(head))
1823a8599bd8SSage Weil 		goto out;
1824a8599bd8SSage Weil 
1825a8599bd8SSage Weil 	/* set upper bound as _last_ entry in chain */
1826a8599bd8SSage Weil 	req = list_entry(head->prev, struct ceph_osd_request,
1827a8599bd8SSage Weil 			 r_unsafe_item);
1828a8599bd8SSage Weil 	last_tid = req->r_tid;
1829a8599bd8SSage Weil 
1830a8599bd8SSage Weil 	do {
1831a8599bd8SSage Weil 		ceph_osdc_get_request(req);
1832a8599bd8SSage Weil 		spin_unlock(&ci->i_unsafe_lock);
1833a8599bd8SSage Weil 		dout("sync_write_wait on tid %llu (until %llu)\n",
1834a8599bd8SSage Weil 		     req->r_tid, last_tid);
1835a8599bd8SSage Weil 		wait_for_completion(&req->r_safe_completion);
1836a8599bd8SSage Weil 		spin_lock(&ci->i_unsafe_lock);
1837a8599bd8SSage Weil 		ceph_osdc_put_request(req);
1838a8599bd8SSage Weil 
1839a8599bd8SSage Weil 		/*
1840a8599bd8SSage Weil 		 * from here on look at first entry in chain, since we
1841a8599bd8SSage Weil 		 * only want to wait for anything older than last_tid
1842a8599bd8SSage Weil 		 */
1843a8599bd8SSage Weil 		if (list_empty(head))
1844a8599bd8SSage Weil 			break;
1845a8599bd8SSage Weil 		req = list_entry(head->next, struct ceph_osd_request,
1846a8599bd8SSage Weil 				 r_unsafe_item);
1847a8599bd8SSage Weil 	} while (req->r_tid < last_tid);
1848a8599bd8SSage Weil out:
1849a8599bd8SSage Weil 	spin_unlock(&ci->i_unsafe_lock);
1850a8599bd8SSage Weil }
1851a8599bd8SSage Weil 
185202c24a82SJosef Bacik int ceph_fsync(struct file *file, loff_t start, loff_t end, int datasync)
1853a8599bd8SSage Weil {
18547ea80859SChristoph Hellwig 	struct inode *inode = file->f_mapping->host;
1855a8599bd8SSage Weil 	struct ceph_inode_info *ci = ceph_inode(inode);
1856a8599bd8SSage Weil 	unsigned flush_tid;
1857a8599bd8SSage Weil 	int ret;
1858a8599bd8SSage Weil 	int dirty;
1859a8599bd8SSage Weil 
1860a8599bd8SSage Weil 	dout("fsync %p%s\n", inode, datasync ? " datasync" : "");
1861a8599bd8SSage Weil 	sync_write_wait(inode);
1862a8599bd8SSage Weil 
186302c24a82SJosef Bacik 	ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
1864a8599bd8SSage Weil 	if (ret < 0)
1865a8599bd8SSage Weil 		return ret;
186602c24a82SJosef Bacik 	mutex_lock(&inode->i_mutex);
1867a8599bd8SSage Weil 
18684fe59789SYan, Zheng 	dirty = try_flush_caps(inode, &flush_tid);
1869a8599bd8SSage Weil 	dout("fsync dirty caps are %s\n", ceph_cap_string(dirty));
1870a8599bd8SSage Weil 
1871a8599bd8SSage Weil 	/*
1872a8599bd8SSage Weil 	 * only wait on non-file metadata writeback (the mds
1873a8599bd8SSage Weil 	 * can recover size and mtime, so we don't need to
1874a8599bd8SSage Weil 	 * wait for that)
1875a8599bd8SSage Weil 	 */
1876a8599bd8SSage Weil 	if (!datasync && (dirty & ~CEPH_CAP_ANY_FILE_WR)) {
1877a8599bd8SSage Weil 		dout("fsync waiting for flush_tid %u\n", flush_tid);
1878a8599bd8SSage Weil 		ret = wait_event_interruptible(ci->i_cap_wq,
1879a8599bd8SSage Weil 				       caps_are_flushed(inode, flush_tid));
1880a8599bd8SSage Weil 	}
1881a8599bd8SSage Weil 
1882a8599bd8SSage Weil 	dout("fsync %p%s done\n", inode, datasync ? " datasync" : "");
188302c24a82SJosef Bacik 	mutex_unlock(&inode->i_mutex);
1884a8599bd8SSage Weil 	return ret;
1885a8599bd8SSage Weil }
1886a8599bd8SSage Weil 
1887a8599bd8SSage Weil /*
1888a8599bd8SSage Weil  * Flush any dirty caps back to the mds.  If we aren't asked to wait,
1889a8599bd8SSage Weil  * queue inode for flush but don't do so immediately, because we can
1890a8599bd8SSage Weil  * get by with fewer MDS messages if we wait for data writeback to
1891a8599bd8SSage Weil  * complete first.
1892a8599bd8SSage Weil  */
1893f1a3d572SStephen Rothwell int ceph_write_inode(struct inode *inode, struct writeback_control *wbc)
1894a8599bd8SSage Weil {
1895a8599bd8SSage Weil 	struct ceph_inode_info *ci = ceph_inode(inode);
1896a8599bd8SSage Weil 	unsigned flush_tid;
1897a8599bd8SSage Weil 	int err = 0;
1898a8599bd8SSage Weil 	int dirty;
1899f1a3d572SStephen Rothwell 	int wait = wbc->sync_mode == WB_SYNC_ALL;
1900a8599bd8SSage Weil 
1901a8599bd8SSage Weil 	dout("write_inode %p wait=%d\n", inode, wait);
1902a8599bd8SSage Weil 	if (wait) {
19034fe59789SYan, Zheng 		dirty = try_flush_caps(inode, &flush_tid);
1904a8599bd8SSage Weil 		if (dirty)
1905a8599bd8SSage Weil 			err = wait_event_interruptible(ci->i_cap_wq,
1906a8599bd8SSage Weil 				       caps_are_flushed(inode, flush_tid));
1907a8599bd8SSage Weil 	} else {
1908640ef79dSCheng Renquan 		struct ceph_mds_client *mdsc =
19093d14c5d2SYehuda Sadeh 			ceph_sb_to_client(inode->i_sb)->mdsc;
1910a8599bd8SSage Weil 
1911be655596SSage Weil 		spin_lock(&ci->i_ceph_lock);
1912a8599bd8SSage Weil 		if (__ceph_caps_dirty(ci))
1913a8599bd8SSage Weil 			__cap_delay_requeue_front(mdsc, ci);
1914be655596SSage Weil 		spin_unlock(&ci->i_ceph_lock);
1915a8599bd8SSage Weil 	}
1916a8599bd8SSage Weil 	return err;
1917a8599bd8SSage Weil }
1918a8599bd8SSage Weil 
1919a8599bd8SSage Weil /*
1920a8599bd8SSage Weil  * After a recovering MDS goes active, we need to resend any caps
1921a8599bd8SSage Weil  * we were flushing.
1922a8599bd8SSage Weil  *
1923a8599bd8SSage Weil  * Caller holds session->s_mutex.
1924a8599bd8SSage Weil  */
1925a8599bd8SSage Weil static void kick_flushing_capsnaps(struct ceph_mds_client *mdsc,
1926a8599bd8SSage Weil 				   struct ceph_mds_session *session)
1927a8599bd8SSage Weil {
1928a8599bd8SSage Weil 	struct ceph_cap_snap *capsnap;
1929a8599bd8SSage Weil 
1930a8599bd8SSage Weil 	dout("kick_flushing_capsnaps mds%d\n", session->s_mds);
1931a8599bd8SSage Weil 	list_for_each_entry(capsnap, &session->s_cap_snaps_flushing,
1932a8599bd8SSage Weil 			    flushing_item) {
1933a8599bd8SSage Weil 		struct ceph_inode_info *ci = capsnap->ci;
1934a8599bd8SSage Weil 		struct inode *inode = &ci->vfs_inode;
1935a8599bd8SSage Weil 		struct ceph_cap *cap;
1936a8599bd8SSage Weil 
1937be655596SSage Weil 		spin_lock(&ci->i_ceph_lock);
1938a8599bd8SSage Weil 		cap = ci->i_auth_cap;
1939a8599bd8SSage Weil 		if (cap && cap->session == session) {
1940a8599bd8SSage Weil 			dout("kick_flushing_caps %p cap %p capsnap %p\n", inode,
1941a8599bd8SSage Weil 			     cap, capsnap);
1942e835124cSSage Weil 			__ceph_flush_snaps(ci, &session, 1);
1943a8599bd8SSage Weil 		} else {
1944a8599bd8SSage Weil 			pr_err("%p auth cap %p not mds%d ???\n", inode,
1945a8599bd8SSage Weil 			       cap, session->s_mds);
1946a8599bd8SSage Weil 		}
1947be655596SSage Weil 		spin_unlock(&ci->i_ceph_lock);
1948a8599bd8SSage Weil 	}
1949a8599bd8SSage Weil }
1950a8599bd8SSage Weil 
1951a8599bd8SSage Weil void ceph_kick_flushing_caps(struct ceph_mds_client *mdsc,
1952a8599bd8SSage Weil 			     struct ceph_mds_session *session)
1953a8599bd8SSage Weil {
1954a8599bd8SSage Weil 	struct ceph_inode_info *ci;
1955a8599bd8SSage Weil 
1956a8599bd8SSage Weil 	kick_flushing_capsnaps(mdsc, session);
1957a8599bd8SSage Weil 
1958a8599bd8SSage Weil 	dout("kick_flushing_caps mds%d\n", session->s_mds);
1959a8599bd8SSage Weil 	list_for_each_entry(ci, &session->s_cap_flushing, i_flushing_item) {
1960a8599bd8SSage Weil 		struct inode *inode = &ci->vfs_inode;
1961a8599bd8SSage Weil 		struct ceph_cap *cap;
1962a8599bd8SSage Weil 		int delayed = 0;
1963a8599bd8SSage Weil 
1964be655596SSage Weil 		spin_lock(&ci->i_ceph_lock);
1965a8599bd8SSage Weil 		cap = ci->i_auth_cap;
1966a8599bd8SSage Weil 		if (cap && cap->session == session) {
1967a8599bd8SSage Weil 			dout("kick_flushing_caps %p cap %p %s\n", inode,
1968a8599bd8SSage Weil 			     cap, ceph_cap_string(ci->i_flushing_caps));
1969a8599bd8SSage Weil 			delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH,
1970a8599bd8SSage Weil 					     __ceph_caps_used(ci),
1971a8599bd8SSage Weil 					     __ceph_caps_wanted(ci),
1972a8599bd8SSage Weil 					     cap->issued | cap->implemented,
1973a8599bd8SSage Weil 					     ci->i_flushing_caps, NULL);
1974a8599bd8SSage Weil 			if (delayed) {
1975be655596SSage Weil 				spin_lock(&ci->i_ceph_lock);
1976a8599bd8SSage Weil 				__cap_delay_requeue(mdsc, ci);
1977be655596SSage Weil 				spin_unlock(&ci->i_ceph_lock);
1978a8599bd8SSage Weil 			}
1979a8599bd8SSage Weil 		} else {
1980a8599bd8SSage Weil 			pr_err("%p auth cap %p not mds%d ???\n", inode,
1981a8599bd8SSage Weil 			       cap, session->s_mds);
1982be655596SSage Weil 			spin_unlock(&ci->i_ceph_lock);
1983a8599bd8SSage Weil 		}
1984a8599bd8SSage Weil 	}
1985a8599bd8SSage Weil }
1986a8599bd8SSage Weil 
1987088b3f5eSSage Weil static void kick_flushing_inode_caps(struct ceph_mds_client *mdsc,
1988088b3f5eSSage Weil 				     struct ceph_mds_session *session,
1989088b3f5eSSage Weil 				     struct inode *inode)
1990088b3f5eSSage Weil {
1991088b3f5eSSage Weil 	struct ceph_inode_info *ci = ceph_inode(inode);
1992088b3f5eSSage Weil 	struct ceph_cap *cap;
1993088b3f5eSSage Weil 	int delayed = 0;
1994088b3f5eSSage Weil 
1995be655596SSage Weil 	spin_lock(&ci->i_ceph_lock);
1996088b3f5eSSage Weil 	cap = ci->i_auth_cap;
1997088b3f5eSSage Weil 	dout("kick_flushing_inode_caps %p flushing %s flush_seq %lld\n", inode,
1998088b3f5eSSage Weil 	     ceph_cap_string(ci->i_flushing_caps), ci->i_cap_flush_seq);
1999005c4697SYan, Zheng 
2000088b3f5eSSage Weil 	__ceph_flush_snaps(ci, &session, 1);
2001005c4697SYan, Zheng 
2002088b3f5eSSage Weil 	if (ci->i_flushing_caps) {
2003005c4697SYan, Zheng 		spin_lock(&mdsc->cap_dirty_lock);
2004005c4697SYan, Zheng 		list_move_tail(&ci->i_flushing_item,
2005005c4697SYan, Zheng 			       &cap->session->s_cap_flushing);
2006005c4697SYan, Zheng 		spin_unlock(&mdsc->cap_dirty_lock);
2007005c4697SYan, Zheng 
2008088b3f5eSSage Weil 		delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH,
2009088b3f5eSSage Weil 				     __ceph_caps_used(ci),
2010088b3f5eSSage Weil 				     __ceph_caps_wanted(ci),
2011088b3f5eSSage Weil 				     cap->issued | cap->implemented,
2012088b3f5eSSage Weil 				     ci->i_flushing_caps, NULL);
2013088b3f5eSSage Weil 		if (delayed) {
2014be655596SSage Weil 			spin_lock(&ci->i_ceph_lock);
2015088b3f5eSSage Weil 			__cap_delay_requeue(mdsc, ci);
2016be655596SSage Weil 			spin_unlock(&ci->i_ceph_lock);
2017088b3f5eSSage Weil 		}
2018088b3f5eSSage Weil 	} else {
2019be655596SSage Weil 		spin_unlock(&ci->i_ceph_lock);
2020088b3f5eSSage Weil 	}
2021088b3f5eSSage Weil }
2022088b3f5eSSage Weil 
2023a8599bd8SSage Weil 
2024a8599bd8SSage Weil /*
2025a8599bd8SSage Weil  * Take references to capabilities we hold, so that we don't release
2026a8599bd8SSage Weil  * them to the MDS prematurely.
2027a8599bd8SSage Weil  *
2028be655596SSage Weil  * Protected by i_ceph_lock.
2029a8599bd8SSage Weil  */
2030a8599bd8SSage Weil static void __take_cap_refs(struct ceph_inode_info *ci, int got)
2031a8599bd8SSage Weil {
2032a8599bd8SSage Weil 	if (got & CEPH_CAP_PIN)
2033a8599bd8SSage Weil 		ci->i_pin_ref++;
2034a8599bd8SSage Weil 	if (got & CEPH_CAP_FILE_RD)
2035a8599bd8SSage Weil 		ci->i_rd_ref++;
2036a8599bd8SSage Weil 	if (got & CEPH_CAP_FILE_CACHE)
2037a8599bd8SSage Weil 		ci->i_rdcache_ref++;
2038a8599bd8SSage Weil 	if (got & CEPH_CAP_FILE_WR)
2039a8599bd8SSage Weil 		ci->i_wr_ref++;
2040a8599bd8SSage Weil 	if (got & CEPH_CAP_FILE_BUFFER) {
2041d3d0720dSHenry C Chang 		if (ci->i_wb_ref == 0)
20423772d26dSSage Weil 			ihold(&ci->vfs_inode);
2043d3d0720dSHenry C Chang 		ci->i_wb_ref++;
2044d3d0720dSHenry C Chang 		dout("__take_cap_refs %p wb %d -> %d (?)\n",
2045d3d0720dSHenry C Chang 		     &ci->vfs_inode, ci->i_wb_ref-1, ci->i_wb_ref);
2046a8599bd8SSage Weil 	}
2047a8599bd8SSage Weil }
2048a8599bd8SSage Weil 
2049a8599bd8SSage Weil /*
2050a8599bd8SSage Weil  * Try to grab cap references.  Specify those refs we @want, and the
2051a8599bd8SSage Weil  * minimal set we @need.  Also include the larger offset we are writing
2052a8599bd8SSage Weil  * to (when applicable), and check against max_size here as well.
2053a8599bd8SSage Weil  * Note that caller is responsible for ensuring max_size increases are
2054a8599bd8SSage Weil  * requested from the MDS.
2055a8599bd8SSage Weil  */
2056a8599bd8SSage Weil static int try_get_cap_refs(struct ceph_inode_info *ci, int need, int want,
2057a8599bd8SSage Weil 			    int *got, loff_t endoff, int *check_max, int *err)
2058a8599bd8SSage Weil {
2059a8599bd8SSage Weil 	struct inode *inode = &ci->vfs_inode;
2060a8599bd8SSage Weil 	int ret = 0;
2061a8599bd8SSage Weil 	int have, implemented;
2062195d3ce2SSage Weil 	int file_wanted;
2063a8599bd8SSage Weil 
2064a8599bd8SSage Weil 	dout("get_cap_refs %p need %s want %s\n", inode,
2065a8599bd8SSage Weil 	     ceph_cap_string(need), ceph_cap_string(want));
2066be655596SSage Weil 	spin_lock(&ci->i_ceph_lock);
2067a8599bd8SSage Weil 
2068195d3ce2SSage Weil 	/* make sure file is actually open */
2069195d3ce2SSage Weil 	file_wanted = __ceph_caps_file_wanted(ci);
2070195d3ce2SSage Weil 	if ((file_wanted & need) == 0) {
2071195d3ce2SSage Weil 		dout("try_get_cap_refs need %s file_wanted %s, EBADF\n",
2072195d3ce2SSage Weil 		     ceph_cap_string(need), ceph_cap_string(file_wanted));
2073a8599bd8SSage Weil 		*err = -EBADF;
2074a8599bd8SSage Weil 		ret = 1;
2075a8599bd8SSage Weil 		goto out;
2076a8599bd8SSage Weil 	}
2077a8599bd8SSage Weil 
207837505d57SYan, Zheng 	/* finish pending truncate */
207937505d57SYan, Zheng 	while (ci->i_truncate_pending) {
208037505d57SYan, Zheng 		spin_unlock(&ci->i_ceph_lock);
2081b415bf4fSYan, Zheng 		__ceph_do_pending_vmtruncate(inode);
208237505d57SYan, Zheng 		spin_lock(&ci->i_ceph_lock);
208337505d57SYan, Zheng 	}
208437505d57SYan, Zheng 
20853871cbb9SYan, Zheng 	have = __ceph_caps_issued(ci, &implemented);
20863871cbb9SYan, Zheng 
20873871cbb9SYan, Zheng 	if (have & need & CEPH_CAP_FILE_WR) {
2088a8599bd8SSage Weil 		if (endoff >= 0 && endoff > (loff_t)ci->i_max_size) {
2089a8599bd8SSage Weil 			dout("get_cap_refs %p endoff %llu > maxsize %llu\n",
2090a8599bd8SSage Weil 			     inode, endoff, ci->i_max_size);
20913871cbb9SYan, Zheng 			if (endoff > ci->i_requested_max_size) {
2092a8599bd8SSage Weil 				*check_max = 1;
2093a8599bd8SSage Weil 				ret = 1;
2094a8599bd8SSage Weil 			}
2095a8599bd8SSage Weil 			goto out;
2096a8599bd8SSage Weil 		}
2097a8599bd8SSage Weil 		/*
2098a8599bd8SSage Weil 		 * If a sync write is in progress, we must wait, so that we
2099a8599bd8SSage Weil 		 * can get a final snapshot value for size+mtime.
2100a8599bd8SSage Weil 		 */
2101a8599bd8SSage Weil 		if (__ceph_have_pending_cap_snap(ci)) {
2102a8599bd8SSage Weil 			dout("get_cap_refs %p cap_snap_pending\n", inode);
2103a8599bd8SSage Weil 			goto out;
2104a8599bd8SSage Weil 		}
2105a8599bd8SSage Weil 	}
2106a8599bd8SSage Weil 
2107a8599bd8SSage Weil 	if ((have & need) == need) {
2108a8599bd8SSage Weil 		/*
2109a8599bd8SSage Weil 		 * Look at (implemented & ~have & not) so that we keep waiting
2110a8599bd8SSage Weil 		 * on transition from wanted -> needed caps.  This is needed
2111a8599bd8SSage Weil 		 * for WRBUFFER|WR -> WR to avoid a new WR sync write from
2112a8599bd8SSage Weil 		 * going before a prior buffered writeback happens.
2113a8599bd8SSage Weil 		 */
2114a8599bd8SSage Weil 		int not = want & ~(have & need);
2115a8599bd8SSage Weil 		int revoking = implemented & ~have;
2116a8599bd8SSage Weil 		dout("get_cap_refs %p have %s but not %s (revoking %s)\n",
2117a8599bd8SSage Weil 		     inode, ceph_cap_string(have), ceph_cap_string(not),
2118a8599bd8SSage Weil 		     ceph_cap_string(revoking));
2119a8599bd8SSage Weil 		if ((revoking & not) == 0) {
2120a8599bd8SSage Weil 			*got = need | (have & want);
2121a8599bd8SSage Weil 			__take_cap_refs(ci, *got);
2122a8599bd8SSage Weil 			ret = 1;
2123a8599bd8SSage Weil 		}
2124a8599bd8SSage Weil 	} else {
2125a8599bd8SSage Weil 		dout("get_cap_refs %p have %s needed %s\n", inode,
2126a8599bd8SSage Weil 		     ceph_cap_string(have), ceph_cap_string(need));
2127a8599bd8SSage Weil 	}
2128a8599bd8SSage Weil out:
2129be655596SSage Weil 	spin_unlock(&ci->i_ceph_lock);
2130a8599bd8SSage Weil 	dout("get_cap_refs %p ret %d got %s\n", inode,
2131a8599bd8SSage Weil 	     ret, ceph_cap_string(*got));
2132a8599bd8SSage Weil 	return ret;
2133a8599bd8SSage Weil }
2134a8599bd8SSage Weil 
2135a8599bd8SSage Weil /*
2136a8599bd8SSage Weil  * Check the offset we are writing up to against our current
2137a8599bd8SSage Weil  * max_size.  If necessary, tell the MDS we want to write to
2138a8599bd8SSage Weil  * a larger offset.
2139a8599bd8SSage Weil  */
2140a8599bd8SSage Weil static void check_max_size(struct inode *inode, loff_t endoff)
2141a8599bd8SSage Weil {
2142a8599bd8SSage Weil 	struct ceph_inode_info *ci = ceph_inode(inode);
2143a8599bd8SSage Weil 	int check = 0;
2144a8599bd8SSage Weil 
2145a8599bd8SSage Weil 	/* do we need to explicitly request a larger max_size? */
2146be655596SSage Weil 	spin_lock(&ci->i_ceph_lock);
21473871cbb9SYan, Zheng 	if (endoff >= ci->i_max_size && endoff > ci->i_wanted_max_size) {
2148a8599bd8SSage Weil 		dout("write %p at large endoff %llu, req max_size\n",
2149a8599bd8SSage Weil 		     inode, endoff);
2150a8599bd8SSage Weil 		ci->i_wanted_max_size = endoff;
2151a8599bd8SSage Weil 	}
21523871cbb9SYan, Zheng 	/* duplicate ceph_check_caps()'s logic */
21533871cbb9SYan, Zheng 	if (ci->i_auth_cap &&
21543871cbb9SYan, Zheng 	    (ci->i_auth_cap->issued & CEPH_CAP_FILE_WR) &&
21553871cbb9SYan, Zheng 	    ci->i_wanted_max_size > ci->i_max_size &&
21563871cbb9SYan, Zheng 	    ci->i_wanted_max_size > ci->i_requested_max_size)
21573871cbb9SYan, Zheng 		check = 1;
2158be655596SSage Weil 	spin_unlock(&ci->i_ceph_lock);
2159a8599bd8SSage Weil 	if (check)
2160a8599bd8SSage Weil 		ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
2161a8599bd8SSage Weil }
2162a8599bd8SSage Weil 
2163a8599bd8SSage Weil /*
2164a8599bd8SSage Weil  * Wait for caps, and take cap references.  If we can't get a WR cap
2165a8599bd8SSage Weil  * due to a small max_size, make sure we check_max_size (and possibly
2166a8599bd8SSage Weil  * ask the mds) so we don't get hung up indefinitely.
2167a8599bd8SSage Weil  */
2168a8599bd8SSage Weil int ceph_get_caps(struct ceph_inode_info *ci, int need, int want, int *got,
2169a8599bd8SSage Weil 		  loff_t endoff)
2170a8599bd8SSage Weil {
2171a8599bd8SSage Weil 	int check_max, ret, err;
2172a8599bd8SSage Weil 
2173a8599bd8SSage Weil retry:
2174a8599bd8SSage Weil 	if (endoff > 0)
2175a8599bd8SSage Weil 		check_max_size(&ci->vfs_inode, endoff);
2176a8599bd8SSage Weil 	check_max = 0;
2177a8599bd8SSage Weil 	err = 0;
2178a8599bd8SSage Weil 	ret = wait_event_interruptible(ci->i_cap_wq,
2179a8599bd8SSage Weil 				       try_get_cap_refs(ci, need, want,
2180a8599bd8SSage Weil 							got, endoff,
2181a8599bd8SSage Weil 							&check_max, &err));
2182a8599bd8SSage Weil 	if (err)
2183a8599bd8SSage Weil 		ret = err;
2184a8599bd8SSage Weil 	if (check_max)
2185a8599bd8SSage Weil 		goto retry;
2186a8599bd8SSage Weil 	return ret;
2187a8599bd8SSage Weil }
2188a8599bd8SSage Weil 
2189a8599bd8SSage Weil /*
2190a8599bd8SSage Weil  * Take cap refs.  Caller must already know we hold at least one ref
2191a8599bd8SSage Weil  * on the caps in question or we don't know this is safe.
2192a8599bd8SSage Weil  */
2193a8599bd8SSage Weil void ceph_get_cap_refs(struct ceph_inode_info *ci, int caps)
2194a8599bd8SSage Weil {
2195be655596SSage Weil 	spin_lock(&ci->i_ceph_lock);
2196a8599bd8SSage Weil 	__take_cap_refs(ci, caps);
2197be655596SSage Weil 	spin_unlock(&ci->i_ceph_lock);
2198a8599bd8SSage Weil }
2199a8599bd8SSage Weil 
2200a8599bd8SSage Weil /*
2201a8599bd8SSage Weil  * Release cap refs.
2202a8599bd8SSage Weil  *
2203a8599bd8SSage Weil  * If we released the last ref on any given cap, call ceph_check_caps
2204a8599bd8SSage Weil  * to release (or schedule a release).
2205a8599bd8SSage Weil  *
2206a8599bd8SSage Weil  * If we are releasing a WR cap (from a sync write), finalize any affected
2207a8599bd8SSage Weil  * cap_snap, and wake up any waiters.
2208a8599bd8SSage Weil  */
2209a8599bd8SSage Weil void ceph_put_cap_refs(struct ceph_inode_info *ci, int had)
2210a8599bd8SSage Weil {
2211a8599bd8SSage Weil 	struct inode *inode = &ci->vfs_inode;
2212a8599bd8SSage Weil 	int last = 0, put = 0, flushsnaps = 0, wake = 0;
2213a8599bd8SSage Weil 	struct ceph_cap_snap *capsnap;
2214a8599bd8SSage Weil 
2215be655596SSage Weil 	spin_lock(&ci->i_ceph_lock);
2216a8599bd8SSage Weil 	if (had & CEPH_CAP_PIN)
2217a8599bd8SSage Weil 		--ci->i_pin_ref;
2218a8599bd8SSage Weil 	if (had & CEPH_CAP_FILE_RD)
2219a8599bd8SSage Weil 		if (--ci->i_rd_ref == 0)
2220a8599bd8SSage Weil 			last++;
2221a8599bd8SSage Weil 	if (had & CEPH_CAP_FILE_CACHE)
2222a8599bd8SSage Weil 		if (--ci->i_rdcache_ref == 0)
2223a8599bd8SSage Weil 			last++;
2224a8599bd8SSage Weil 	if (had & CEPH_CAP_FILE_BUFFER) {
2225d3d0720dSHenry C Chang 		if (--ci->i_wb_ref == 0) {
2226a8599bd8SSage Weil 			last++;
2227a8599bd8SSage Weil 			put++;
2228a8599bd8SSage Weil 		}
2229d3d0720dSHenry C Chang 		dout("put_cap_refs %p wb %d -> %d (?)\n",
2230d3d0720dSHenry C Chang 		     inode, ci->i_wb_ref+1, ci->i_wb_ref);
2231a8599bd8SSage Weil 	}
2232a8599bd8SSage Weil 	if (had & CEPH_CAP_FILE_WR)
2233a8599bd8SSage Weil 		if (--ci->i_wr_ref == 0) {
2234a8599bd8SSage Weil 			last++;
2235a8599bd8SSage Weil 			if (!list_empty(&ci->i_cap_snaps)) {
2236a8599bd8SSage Weil 				capsnap = list_first_entry(&ci->i_cap_snaps,
2237a8599bd8SSage Weil 						     struct ceph_cap_snap,
2238a8599bd8SSage Weil 						     ci_item);
2239a8599bd8SSage Weil 				if (capsnap->writing) {
2240a8599bd8SSage Weil 					capsnap->writing = 0;
2241a8599bd8SSage Weil 					flushsnaps =
2242a8599bd8SSage Weil 						__ceph_finish_cap_snap(ci,
2243a8599bd8SSage Weil 								       capsnap);
2244a8599bd8SSage Weil 					wake = 1;
2245a8599bd8SSage Weil 				}
2246a8599bd8SSage Weil 			}
2247a8599bd8SSage Weil 		}
2248be655596SSage Weil 	spin_unlock(&ci->i_ceph_lock);
2249a8599bd8SSage Weil 
2250819ccbfaSSage Weil 	dout("put_cap_refs %p had %s%s%s\n", inode, ceph_cap_string(had),
2251819ccbfaSSage Weil 	     last ? " last" : "", put ? " put" : "");
2252a8599bd8SSage Weil 
2253a8599bd8SSage Weil 	if (last && !flushsnaps)
2254a8599bd8SSage Weil 		ceph_check_caps(ci, 0, NULL);
2255a8599bd8SSage Weil 	else if (flushsnaps)
2256a8599bd8SSage Weil 		ceph_flush_snaps(ci);
2257a8599bd8SSage Weil 	if (wake)
225803066f23SYehuda Sadeh 		wake_up_all(&ci->i_cap_wq);
2259a8599bd8SSage Weil 	if (put)
2260a8599bd8SSage Weil 		iput(inode);
2261a8599bd8SSage Weil }
2262a8599bd8SSage Weil 
2263a8599bd8SSage Weil /*
2264a8599bd8SSage Weil  * Release @nr WRBUFFER refs on dirty pages for the given @snapc snap
2265a8599bd8SSage Weil  * context.  Adjust per-snap dirty page accounting as appropriate.
2266a8599bd8SSage Weil  * Once all dirty data for a cap_snap is flushed, flush snapped file
2267a8599bd8SSage Weil  * metadata back to the MDS.  If we dropped the last ref, call
2268a8599bd8SSage Weil  * ceph_check_caps.
2269a8599bd8SSage Weil  */
2270a8599bd8SSage Weil void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info *ci, int nr,
2271a8599bd8SSage Weil 				struct ceph_snap_context *snapc)
2272a8599bd8SSage Weil {
2273a8599bd8SSage Weil 	struct inode *inode = &ci->vfs_inode;
2274a8599bd8SSage Weil 	int last = 0;
2275819ccbfaSSage Weil 	int complete_capsnap = 0;
2276819ccbfaSSage Weil 	int drop_capsnap = 0;
2277a8599bd8SSage Weil 	int found = 0;
2278a8599bd8SSage Weil 	struct ceph_cap_snap *capsnap = NULL;
2279a8599bd8SSage Weil 
2280be655596SSage Weil 	spin_lock(&ci->i_ceph_lock);
2281a8599bd8SSage Weil 	ci->i_wrbuffer_ref -= nr;
2282a8599bd8SSage Weil 	last = !ci->i_wrbuffer_ref;
2283a8599bd8SSage Weil 
2284a8599bd8SSage Weil 	if (ci->i_head_snapc == snapc) {
2285a8599bd8SSage Weil 		ci->i_wrbuffer_ref_head -= nr;
22867d8cb26dSSage Weil 		if (ci->i_wrbuffer_ref_head == 0 &&
22877d8cb26dSSage Weil 		    ci->i_dirty_caps == 0 && ci->i_flushing_caps == 0) {
22887d8cb26dSSage Weil 			BUG_ON(!ci->i_head_snapc);
2289a8599bd8SSage Weil 			ceph_put_snap_context(ci->i_head_snapc);
2290a8599bd8SSage Weil 			ci->i_head_snapc = NULL;
2291a8599bd8SSage Weil 		}
2292a8599bd8SSage Weil 		dout("put_wrbuffer_cap_refs on %p head %d/%d -> %d/%d %s\n",
2293a8599bd8SSage Weil 		     inode,
2294a8599bd8SSage Weil 		     ci->i_wrbuffer_ref+nr, ci->i_wrbuffer_ref_head+nr,
2295a8599bd8SSage Weil 		     ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
2296a8599bd8SSage Weil 		     last ? " LAST" : "");
2297a8599bd8SSage Weil 	} else {
2298a8599bd8SSage Weil 		list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
2299a8599bd8SSage Weil 			if (capsnap->context == snapc) {
2300a8599bd8SSage Weil 				found = 1;
2301a8599bd8SSage Weil 				break;
2302a8599bd8SSage Weil 			}
2303a8599bd8SSage Weil 		}
2304a8599bd8SSage Weil 		BUG_ON(!found);
2305819ccbfaSSage Weil 		capsnap->dirty_pages -= nr;
2306819ccbfaSSage Weil 		if (capsnap->dirty_pages == 0) {
2307819ccbfaSSage Weil 			complete_capsnap = 1;
2308819ccbfaSSage Weil 			if (capsnap->dirty == 0)
2309819ccbfaSSage Weil 				/* cap writeback completed before we created
2310819ccbfaSSage Weil 				 * the cap_snap; no FLUSHSNAP is needed */
2311819ccbfaSSage Weil 				drop_capsnap = 1;
2312819ccbfaSSage Weil 		}
2313a8599bd8SSage Weil 		dout("put_wrbuffer_cap_refs on %p cap_snap %p "
2314819ccbfaSSage Weil 		     " snap %lld %d/%d -> %d/%d %s%s%s\n",
2315a8599bd8SSage Weil 		     inode, capsnap, capsnap->context->seq,
2316a8599bd8SSage Weil 		     ci->i_wrbuffer_ref+nr, capsnap->dirty_pages + nr,
2317a8599bd8SSage Weil 		     ci->i_wrbuffer_ref, capsnap->dirty_pages,
2318a8599bd8SSage Weil 		     last ? " (wrbuffer last)" : "",
2319819ccbfaSSage Weil 		     complete_capsnap ? " (complete capsnap)" : "",
2320819ccbfaSSage Weil 		     drop_capsnap ? " (drop capsnap)" : "");
2321819ccbfaSSage Weil 		if (drop_capsnap) {
2322819ccbfaSSage Weil 			ceph_put_snap_context(capsnap->context);
2323819ccbfaSSage Weil 			list_del(&capsnap->ci_item);
2324819ccbfaSSage Weil 			list_del(&capsnap->flushing_item);
2325819ccbfaSSage Weil 			ceph_put_cap_snap(capsnap);
2326819ccbfaSSage Weil 		}
2327a8599bd8SSage Weil 	}
2328a8599bd8SSage Weil 
2329be655596SSage Weil 	spin_unlock(&ci->i_ceph_lock);
2330a8599bd8SSage Weil 
2331a8599bd8SSage Weil 	if (last) {
2332a8599bd8SSage Weil 		ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
2333a8599bd8SSage Weil 		iput(inode);
2334819ccbfaSSage Weil 	} else if (complete_capsnap) {
2335a8599bd8SSage Weil 		ceph_flush_snaps(ci);
233603066f23SYehuda Sadeh 		wake_up_all(&ci->i_cap_wq);
2337a8599bd8SSage Weil 	}
2338819ccbfaSSage Weil 	if (drop_capsnap)
2339819ccbfaSSage Weil 		iput(inode);
2340a8599bd8SSage Weil }
2341a8599bd8SSage Weil 
2342a8599bd8SSage Weil /*
2343ca20c991SYan, Zheng  * Invalidate unlinked inode's aliases, so we can drop the inode ASAP.
2344ca20c991SYan, Zheng  */
2345ca20c991SYan, Zheng static void invalidate_aliases(struct inode *inode)
2346ca20c991SYan, Zheng {
2347ca20c991SYan, Zheng 	struct dentry *dn, *prev = NULL;
2348ca20c991SYan, Zheng 
2349ca20c991SYan, Zheng 	dout("invalidate_aliases inode %p\n", inode);
2350ca20c991SYan, Zheng 	d_prune_aliases(inode);
2351ca20c991SYan, Zheng 	/*
2352ca20c991SYan, Zheng 	 * For non-directory inode, d_find_alias() only returns
2353fc12c80aSJ. Bruce Fields 	 * hashed dentry. After calling d_invalidate(), the
2354fc12c80aSJ. Bruce Fields 	 * dentry becomes unhashed.
2355ca20c991SYan, Zheng 	 *
2356a8d436f0SYan, Zheng 	 * For directory inode, d_find_alias() can return
2357fc12c80aSJ. Bruce Fields 	 * unhashed dentry. But directory inode should have
2358ca20c991SYan, Zheng 	 * one alias at most.
2359ca20c991SYan, Zheng 	 */
2360ca20c991SYan, Zheng 	while ((dn = d_find_alias(inode))) {
2361ca20c991SYan, Zheng 		if (dn == prev) {
2362ca20c991SYan, Zheng 			dput(dn);
2363ca20c991SYan, Zheng 			break;
2364ca20c991SYan, Zheng 		}
2365a8d436f0SYan, Zheng 		d_invalidate(dn);
2366ca20c991SYan, Zheng 		if (prev)
2367ca20c991SYan, Zheng 			dput(prev);
2368ca20c991SYan, Zheng 		prev = dn;
2369ca20c991SYan, Zheng 	}
2370ca20c991SYan, Zheng 	if (prev)
2371ca20c991SYan, Zheng 		dput(prev);
2372ca20c991SYan, Zheng }
2373ca20c991SYan, Zheng 
2374ca20c991SYan, Zheng /*
2375a8599bd8SSage Weil  * Handle a cap GRANT message from the MDS.  (Note that a GRANT may
2376a8599bd8SSage Weil  * actually be a revocation if it specifies a smaller cap set.)
2377a8599bd8SSage Weil  *
2378be655596SSage Weil  * caller holds s_mutex and i_ceph_lock, we drop both.
237915637c8bSSage Weil  *
2380a8599bd8SSage Weil  * return value:
2381a8599bd8SSage Weil  *  0 - ok
2382a8599bd8SSage Weil  *  1 - check_caps on auth cap only (writeback)
2383a8599bd8SSage Weil  *  2 - check_caps (ack revoke)
2384a8599bd8SSage Weil  */
238515637c8bSSage Weil static void handle_cap_grant(struct inode *inode, struct ceph_mds_caps *grant,
2386a8599bd8SSage Weil 			     struct ceph_mds_session *session,
2387a8599bd8SSage Weil 			     struct ceph_cap *cap,
2388a8599bd8SSage Weil 			     struct ceph_buffer *xattr_buf)
2389be655596SSage Weil 		__releases(ci->i_ceph_lock)
2390a8599bd8SSage Weil {
2391a8599bd8SSage Weil 	struct ceph_inode_info *ci = ceph_inode(inode);
2392a8599bd8SSage Weil 	int mds = session->s_mds;
23932f56f56aSSage Weil 	int seq = le32_to_cpu(grant->seq);
2394a8599bd8SSage Weil 	int newcaps = le32_to_cpu(grant->caps);
2395a8599bd8SSage Weil 	int issued, implemented, used, wanted, dirty;
2396a8599bd8SSage Weil 	u64 size = le64_to_cpu(grant->size);
2397a8599bd8SSage Weil 	u64 max_size = le64_to_cpu(grant->max_size);
2398a8599bd8SSage Weil 	struct timespec mtime, atime, ctime;
239915637c8bSSage Weil 	int check_caps = 0;
2400a8599bd8SSage Weil 	int wake = 0;
2401a8599bd8SSage Weil 	int writeback = 0;
24023c6f6b79SSage Weil 	int queue_invalidate = 0;
2403ca20c991SYan, Zheng 	int deleted_inode = 0;
240499ccbd22SMilosz Tanski 	int queue_revalidate = 0;
2405a8599bd8SSage Weil 
24062f56f56aSSage Weil 	dout("handle_cap_grant inode %p cap %p mds%d seq %d %s\n",
24072f56f56aSSage Weil 	     inode, cap, mds, seq, ceph_cap_string(newcaps));
2408a8599bd8SSage Weil 	dout(" size %llu max_size %llu, i_size %llu\n", size, max_size,
2409a8599bd8SSage Weil 		inode->i_size);
2410a8599bd8SSage Weil 
2411a8599bd8SSage Weil 	/*
2412a8599bd8SSage Weil 	 * If CACHE is being revoked, and we have no dirty buffers,
2413a8599bd8SSage Weil 	 * try to invalidate (once).  (If there are dirty buffers, we
2414a8599bd8SSage Weil 	 * will invalidate _after_ writeback.)
2415a8599bd8SSage Weil 	 */
24163b454c49SSage Weil 	if (((cap->issued & ~newcaps) & CEPH_CAP_FILE_CACHE) &&
24173b454c49SSage Weil 	    (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 &&
2418bcd2cbd1SYehuda Sadeh 	    !ci->i_wrbuffer_ref) {
2419e9075743SLi Wang 		if (try_nonblocking_invalidate(inode)) {
2420a8599bd8SSage Weil 			/* there were locked pages.. invalidate later
2421a8599bd8SSage Weil 			   in a separate thread. */
2422a8599bd8SSage Weil 			if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
24233c6f6b79SSage Weil 				queue_invalidate = 1;
2424a8599bd8SSage Weil 				ci->i_rdcache_revoking = ci->i_rdcache_gen;
2425a8599bd8SSage Weil 			}
2426a8599bd8SSage Weil 		}
242799ccbd22SMilosz Tanski 
242899ccbd22SMilosz Tanski 		ceph_fscache_invalidate(inode);
2429a8599bd8SSage Weil 	}
2430a8599bd8SSage Weil 
2431a8599bd8SSage Weil 	/* side effects now are allowed */
2432a8599bd8SSage Weil 
2433a8599bd8SSage Weil 	issued = __ceph_caps_issued(ci, &implemented);
2434a8599bd8SSage Weil 	issued |= implemented | __ceph_caps_dirty(ci);
2435a8599bd8SSage Weil 
2436685f9a5dSSage Weil 	cap->cap_gen = session->s_cap_gen;
2437a8599bd8SSage Weil 
2438a8599bd8SSage Weil 	__check_cap_issue(ci, cap, newcaps);
2439a8599bd8SSage Weil 
2440a8599bd8SSage Weil 	if ((issued & CEPH_CAP_AUTH_EXCL) == 0) {
2441a8599bd8SSage Weil 		inode->i_mode = le32_to_cpu(grant->mode);
244205cb11c1SEric W. Biederman 		inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(grant->uid));
244305cb11c1SEric W. Biederman 		inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(grant->gid));
2444a8599bd8SSage Weil 		dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
2445bd2bae6aSEric W. Biederman 		     from_kuid(&init_user_ns, inode->i_uid),
2446bd2bae6aSEric W. Biederman 		     from_kgid(&init_user_ns, inode->i_gid));
2447a8599bd8SSage Weil 	}
2448a8599bd8SSage Weil 
2449ca20c991SYan, Zheng 	if ((issued & CEPH_CAP_LINK_EXCL) == 0) {
2450bfe86848SMiklos Szeredi 		set_nlink(inode, le32_to_cpu(grant->nlink));
2451ca20c991SYan, Zheng 		if (inode->i_nlink == 0 &&
2452ca20c991SYan, Zheng 		    (newcaps & (CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL)))
2453ca20c991SYan, Zheng 			deleted_inode = 1;
2454ca20c991SYan, Zheng 	}
2455a8599bd8SSage Weil 
2456a8599bd8SSage Weil 	if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && grant->xattr_len) {
2457a8599bd8SSage Weil 		int len = le32_to_cpu(grant->xattr_len);
2458a8599bd8SSage Weil 		u64 version = le64_to_cpu(grant->xattr_version);
2459a8599bd8SSage Weil 
2460a8599bd8SSage Weil 		if (version > ci->i_xattrs.version) {
2461a8599bd8SSage Weil 			dout(" got new xattrs v%llu on %p len %d\n",
2462a8599bd8SSage Weil 			     version, inode, len);
2463a8599bd8SSage Weil 			if (ci->i_xattrs.blob)
2464a8599bd8SSage Weil 				ceph_buffer_put(ci->i_xattrs.blob);
2465a8599bd8SSage Weil 			ci->i_xattrs.blob = ceph_buffer_get(xattr_buf);
2466a8599bd8SSage Weil 			ci->i_xattrs.version = version;
24677221fe4cSGuangliang Zhao 			ceph_forget_all_cached_acls(inode);
2468a8599bd8SSage Weil 		}
2469a8599bd8SSage Weil 	}
2470a8599bd8SSage Weil 
247199ccbd22SMilosz Tanski 	/* Do we need to revalidate our fscache cookie. Don't bother on the
247299ccbd22SMilosz Tanski 	 * first cache cap as we already validate at cookie creation time. */
247399ccbd22SMilosz Tanski 	if ((issued & CEPH_CAP_FILE_CACHE) && ci->i_rdcache_gen > 1)
247499ccbd22SMilosz Tanski 		queue_revalidate = 1;
247599ccbd22SMilosz Tanski 
2476a8599bd8SSage Weil 	/* size/ctime/mtime/atime? */
2477a8599bd8SSage Weil 	ceph_fill_file_size(inode, issued,
2478a8599bd8SSage Weil 			    le32_to_cpu(grant->truncate_seq),
2479a8599bd8SSage Weil 			    le64_to_cpu(grant->truncate_size), size);
2480a8599bd8SSage Weil 	ceph_decode_timespec(&mtime, &grant->mtime);
2481a8599bd8SSage Weil 	ceph_decode_timespec(&atime, &grant->atime);
2482a8599bd8SSage Weil 	ceph_decode_timespec(&ctime, &grant->ctime);
2483a8599bd8SSage Weil 	ceph_fill_file_time(inode, issued,
2484a8599bd8SSage Weil 			    le32_to_cpu(grant->time_warp_seq), &ctime, &mtime,
2485a8599bd8SSage Weil 			    &atime);
2486a8599bd8SSage Weil 
2487a8599bd8SSage Weil 	/* max size increase? */
24885e62ad30SYan, Zheng 	if (ci->i_auth_cap == cap && max_size != ci->i_max_size) {
2489a8599bd8SSage Weil 		dout("max_size %lld -> %llu\n", ci->i_max_size, max_size);
2490a8599bd8SSage Weil 		ci->i_max_size = max_size;
2491a8599bd8SSage Weil 		if (max_size >= ci->i_wanted_max_size) {
2492a8599bd8SSage Weil 			ci->i_wanted_max_size = 0;  /* reset */
2493a8599bd8SSage Weil 			ci->i_requested_max_size = 0;
2494a8599bd8SSage Weil 		}
2495a8599bd8SSage Weil 		wake = 1;
2496a8599bd8SSage Weil 	}
2497a8599bd8SSage Weil 
2498a8599bd8SSage Weil 	/* check cap bits */
2499a8599bd8SSage Weil 	wanted = __ceph_caps_wanted(ci);
2500a8599bd8SSage Weil 	used = __ceph_caps_used(ci);
2501a8599bd8SSage Weil 	dirty = __ceph_caps_dirty(ci);
2502a8599bd8SSage Weil 	dout(" my wanted = %s, used = %s, dirty %s\n",
2503a8599bd8SSage Weil 	     ceph_cap_string(wanted),
2504a8599bd8SSage Weil 	     ceph_cap_string(used),
2505a8599bd8SSage Weil 	     ceph_cap_string(dirty));
2506a8599bd8SSage Weil 	if (wanted != le32_to_cpu(grant->wanted)) {
2507a8599bd8SSage Weil 		dout("mds wanted %s -> %s\n",
2508a8599bd8SSage Weil 		     ceph_cap_string(le32_to_cpu(grant->wanted)),
2509a8599bd8SSage Weil 		     ceph_cap_string(wanted));
2510390306c3SYan, Zheng 		/* imported cap may not have correct mds_wanted */
2511390306c3SYan, Zheng 		if (le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT)
2512390306c3SYan, Zheng 			check_caps = 1;
2513a8599bd8SSage Weil 	}
2514a8599bd8SSage Weil 
2515a8599bd8SSage Weil 	cap->seq = seq;
2516a8599bd8SSage Weil 
2517a8599bd8SSage Weil 	/* file layout may have changed */
2518a8599bd8SSage Weil 	ci->i_layout = grant->layout;
2519a8599bd8SSage Weil 
2520a8599bd8SSage Weil 	/* revocation, grant, or no-op? */
2521a8599bd8SSage Weil 	if (cap->issued & ~newcaps) {
25223b454c49SSage Weil 		int revoking = cap->issued & ~newcaps;
25233b454c49SSage Weil 
25243b454c49SSage Weil 		dout("revocation: %s -> %s (revoking %s)\n",
25253b454c49SSage Weil 		     ceph_cap_string(cap->issued),
25263b454c49SSage Weil 		     ceph_cap_string(newcaps),
25273b454c49SSage Weil 		     ceph_cap_string(revoking));
25280eb6cd49SSage Weil 		if (revoking & used & CEPH_CAP_FILE_BUFFER)
25293b454c49SSage Weil 			writeback = 1;  /* initiate writeback; will delay ack */
25303b454c49SSage Weil 		else if (revoking == CEPH_CAP_FILE_CACHE &&
25313b454c49SSage Weil 			 (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 &&
25323b454c49SSage Weil 			 queue_invalidate)
25333b454c49SSage Weil 			; /* do nothing yet, invalidation will be queued */
25343b454c49SSage Weil 		else if (cap == ci->i_auth_cap)
25353b454c49SSage Weil 			check_caps = 1; /* check auth cap only */
25363b454c49SSage Weil 		else
25373b454c49SSage Weil 			check_caps = 2; /* check all caps */
2538a8599bd8SSage Weil 		cap->issued = newcaps;
2539978097c9SSage Weil 		cap->implemented |= newcaps;
2540a8599bd8SSage Weil 	} else if (cap->issued == newcaps) {
2541a8599bd8SSage Weil 		dout("caps unchanged: %s -> %s\n",
2542a8599bd8SSage Weil 		     ceph_cap_string(cap->issued), ceph_cap_string(newcaps));
2543a8599bd8SSage Weil 	} else {
2544a8599bd8SSage Weil 		dout("grant: %s -> %s\n", ceph_cap_string(cap->issued),
2545a8599bd8SSage Weil 		     ceph_cap_string(newcaps));
25466ee6b953SYan, Zheng 		/* non-auth MDS is revoking the newly grant caps ? */
25476ee6b953SYan, Zheng 		if (cap == ci->i_auth_cap &&
25486ee6b953SYan, Zheng 		    __ceph_caps_revoking_other(ci, cap, newcaps))
25496ee6b953SYan, Zheng 		    check_caps = 2;
25506ee6b953SYan, Zheng 
2551a8599bd8SSage Weil 		cap->issued = newcaps;
2552a8599bd8SSage Weil 		cap->implemented |= newcaps; /* add bits only, to
2553a8599bd8SSage Weil 					      * avoid stepping on a
2554a8599bd8SSage Weil 					      * pending revocation */
2555a8599bd8SSage Weil 		wake = 1;
2556a8599bd8SSage Weil 	}
2557978097c9SSage Weil 	BUG_ON(cap->issued & ~cap->implemented);
2558a8599bd8SSage Weil 
2559be655596SSage Weil 	spin_unlock(&ci->i_ceph_lock);
256099ccbd22SMilosz Tanski 
25613c6f6b79SSage Weil 	if (writeback)
2562a8599bd8SSage Weil 		/*
2563a8599bd8SSage Weil 		 * queue inode for writeback: we can't actually call
2564a8599bd8SSage Weil 		 * filemap_write_and_wait, etc. from message handler
2565a8599bd8SSage Weil 		 * context.
2566a8599bd8SSage Weil 		 */
25673c6f6b79SSage Weil 		ceph_queue_writeback(inode);
25683c6f6b79SSage Weil 	if (queue_invalidate)
25693c6f6b79SSage Weil 		ceph_queue_invalidate(inode);
2570ca20c991SYan, Zheng 	if (deleted_inode)
2571ca20c991SYan, Zheng 		invalidate_aliases(inode);
257299ccbd22SMilosz Tanski 	if (queue_revalidate)
257399ccbd22SMilosz Tanski 		ceph_queue_revalidate(inode);
2574a8599bd8SSage Weil 	if (wake)
257503066f23SYehuda Sadeh 		wake_up_all(&ci->i_cap_wq);
257615637c8bSSage Weil 
257715637c8bSSage Weil 	if (check_caps == 1)
257815637c8bSSage Weil 		ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_AUTHONLY,
257915637c8bSSage Weil 				session);
258015637c8bSSage Weil 	else if (check_caps == 2)
258115637c8bSSage Weil 		ceph_check_caps(ci, CHECK_CAPS_NODELAY, session);
258215637c8bSSage Weil 	else
258315637c8bSSage Weil 		mutex_unlock(&session->s_mutex);
2584a8599bd8SSage Weil }
2585a8599bd8SSage Weil 
2586a8599bd8SSage Weil /*
2587a8599bd8SSage Weil  * Handle FLUSH_ACK from MDS, indicating that metadata we sent to the
2588a8599bd8SSage Weil  * MDS has been safely committed.
2589a8599bd8SSage Weil  */
25906df058c0SSage Weil static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid,
2591a8599bd8SSage Weil 				 struct ceph_mds_caps *m,
2592a8599bd8SSage Weil 				 struct ceph_mds_session *session,
2593a8599bd8SSage Weil 				 struct ceph_cap *cap)
2594be655596SSage Weil 	__releases(ci->i_ceph_lock)
2595a8599bd8SSage Weil {
2596a8599bd8SSage Weil 	struct ceph_inode_info *ci = ceph_inode(inode);
25973d14c5d2SYehuda Sadeh 	struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
2598a8599bd8SSage Weil 	unsigned seq = le32_to_cpu(m->seq);
2599a8599bd8SSage Weil 	int dirty = le32_to_cpu(m->dirty);
2600a8599bd8SSage Weil 	int cleaned = 0;
2601afcdaea3SSage Weil 	int drop = 0;
2602a8599bd8SSage Weil 	int i;
2603a8599bd8SSage Weil 
2604a8599bd8SSage Weil 	for (i = 0; i < CEPH_CAP_BITS; i++)
2605a8599bd8SSage Weil 		if ((dirty & (1 << i)) &&
2606a8599bd8SSage Weil 		    flush_tid == ci->i_cap_flush_tid[i])
2607a8599bd8SSage Weil 			cleaned |= 1 << i;
2608a8599bd8SSage Weil 
2609a8599bd8SSage Weil 	dout("handle_cap_flush_ack inode %p mds%d seq %d on %s cleaned %s,"
2610a8599bd8SSage Weil 	     " flushing %s -> %s\n",
2611a8599bd8SSage Weil 	     inode, session->s_mds, seq, ceph_cap_string(dirty),
2612a8599bd8SSage Weil 	     ceph_cap_string(cleaned), ceph_cap_string(ci->i_flushing_caps),
2613a8599bd8SSage Weil 	     ceph_cap_string(ci->i_flushing_caps & ~cleaned));
2614a8599bd8SSage Weil 
2615a8599bd8SSage Weil 	if (ci->i_flushing_caps == (ci->i_flushing_caps & ~cleaned))
2616a8599bd8SSage Weil 		goto out;
2617a8599bd8SSage Weil 
2618a8599bd8SSage Weil 	ci->i_flushing_caps &= ~cleaned;
2619a8599bd8SSage Weil 
2620a8599bd8SSage Weil 	spin_lock(&mdsc->cap_dirty_lock);
2621a8599bd8SSage Weil 	if (ci->i_flushing_caps == 0) {
2622a8599bd8SSage Weil 		list_del_init(&ci->i_flushing_item);
2623a8599bd8SSage Weil 		if (!list_empty(&session->s_cap_flushing))
2624a8599bd8SSage Weil 			dout(" mds%d still flushing cap on %p\n",
2625a8599bd8SSage Weil 			     session->s_mds,
2626a8599bd8SSage Weil 			     &list_entry(session->s_cap_flushing.next,
2627a8599bd8SSage Weil 					 struct ceph_inode_info,
2628a8599bd8SSage Weil 					 i_flushing_item)->vfs_inode);
2629a8599bd8SSage Weil 		mdsc->num_cap_flushing--;
263003066f23SYehuda Sadeh 		wake_up_all(&mdsc->cap_flushing_wq);
2631a8599bd8SSage Weil 		dout(" inode %p now !flushing\n", inode);
2632afcdaea3SSage Weil 
2633afcdaea3SSage Weil 		if (ci->i_dirty_caps == 0) {
2634a8599bd8SSage Weil 			dout(" inode %p now clean\n", inode);
2635afcdaea3SSage Weil 			BUG_ON(!list_empty(&ci->i_dirty_item));
2636afcdaea3SSage Weil 			drop = 1;
26377d8cb26dSSage Weil 			if (ci->i_wrbuffer_ref_head == 0) {
26387d8cb26dSSage Weil 				BUG_ON(!ci->i_head_snapc);
26397d8cb26dSSage Weil 				ceph_put_snap_context(ci->i_head_snapc);
26407d8cb26dSSage Weil 				ci->i_head_snapc = NULL;
26417d8cb26dSSage Weil 			}
264276e3b390SSage Weil 		} else {
264376e3b390SSage Weil 			BUG_ON(list_empty(&ci->i_dirty_item));
2644afcdaea3SSage Weil 		}
2645a8599bd8SSage Weil 	}
2646a8599bd8SSage Weil 	spin_unlock(&mdsc->cap_dirty_lock);
264703066f23SYehuda Sadeh 	wake_up_all(&ci->i_cap_wq);
2648a8599bd8SSage Weil 
2649a8599bd8SSage Weil out:
2650be655596SSage Weil 	spin_unlock(&ci->i_ceph_lock);
2651afcdaea3SSage Weil 	if (drop)
2652a8599bd8SSage Weil 		iput(inode);
2653a8599bd8SSage Weil }
2654a8599bd8SSage Weil 
2655a8599bd8SSage Weil /*
2656a8599bd8SSage Weil  * Handle FLUSHSNAP_ACK.  MDS has flushed snap data to disk and we can
2657a8599bd8SSage Weil  * throw away our cap_snap.
2658a8599bd8SSage Weil  *
2659a8599bd8SSage Weil  * Caller hold s_mutex.
2660a8599bd8SSage Weil  */
26616df058c0SSage Weil static void handle_cap_flushsnap_ack(struct inode *inode, u64 flush_tid,
2662a8599bd8SSage Weil 				     struct ceph_mds_caps *m,
2663a8599bd8SSage Weil 				     struct ceph_mds_session *session)
2664a8599bd8SSage Weil {
2665a8599bd8SSage Weil 	struct ceph_inode_info *ci = ceph_inode(inode);
2666a8599bd8SSage Weil 	u64 follows = le64_to_cpu(m->snap_follows);
2667a8599bd8SSage Weil 	struct ceph_cap_snap *capsnap;
2668a8599bd8SSage Weil 	int drop = 0;
2669a8599bd8SSage Weil 
2670a8599bd8SSage Weil 	dout("handle_cap_flushsnap_ack inode %p ci %p mds%d follows %lld\n",
2671a8599bd8SSage Weil 	     inode, ci, session->s_mds, follows);
2672a8599bd8SSage Weil 
2673be655596SSage Weil 	spin_lock(&ci->i_ceph_lock);
2674a8599bd8SSage Weil 	list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
2675a8599bd8SSage Weil 		if (capsnap->follows == follows) {
2676a8599bd8SSage Weil 			if (capsnap->flush_tid != flush_tid) {
2677a8599bd8SSage Weil 				dout(" cap_snap %p follows %lld tid %lld !="
2678a8599bd8SSage Weil 				     " %lld\n", capsnap, follows,
2679a8599bd8SSage Weil 				     flush_tid, capsnap->flush_tid);
2680a8599bd8SSage Weil 				break;
2681a8599bd8SSage Weil 			}
2682a8599bd8SSage Weil 			WARN_ON(capsnap->dirty_pages || capsnap->writing);
2683819ccbfaSSage Weil 			dout(" removing %p cap_snap %p follows %lld\n",
2684819ccbfaSSage Weil 			     inode, capsnap, follows);
2685a8599bd8SSage Weil 			ceph_put_snap_context(capsnap->context);
2686a8599bd8SSage Weil 			list_del(&capsnap->ci_item);
2687a8599bd8SSage Weil 			list_del(&capsnap->flushing_item);
2688a8599bd8SSage Weil 			ceph_put_cap_snap(capsnap);
2689a8599bd8SSage Weil 			drop = 1;
2690a8599bd8SSage Weil 			break;
2691a8599bd8SSage Weil 		} else {
2692a8599bd8SSage Weil 			dout(" skipping cap_snap %p follows %lld\n",
2693a8599bd8SSage Weil 			     capsnap, capsnap->follows);
2694a8599bd8SSage Weil 		}
2695a8599bd8SSage Weil 	}
2696be655596SSage Weil 	spin_unlock(&ci->i_ceph_lock);
2697a8599bd8SSage Weil 	if (drop)
2698a8599bd8SSage Weil 		iput(inode);
2699a8599bd8SSage Weil }
2700a8599bd8SSage Weil 
2701a8599bd8SSage Weil /*
2702a8599bd8SSage Weil  * Handle TRUNC from MDS, indicating file truncation.
2703a8599bd8SSage Weil  *
2704a8599bd8SSage Weil  * caller hold s_mutex.
2705a8599bd8SSage Weil  */
2706a8599bd8SSage Weil static void handle_cap_trunc(struct inode *inode,
2707a8599bd8SSage Weil 			     struct ceph_mds_caps *trunc,
2708a8599bd8SSage Weil 			     struct ceph_mds_session *session)
2709be655596SSage Weil 	__releases(ci->i_ceph_lock)
2710a8599bd8SSage Weil {
2711a8599bd8SSage Weil 	struct ceph_inode_info *ci = ceph_inode(inode);
2712a8599bd8SSage Weil 	int mds = session->s_mds;
2713a8599bd8SSage Weil 	int seq = le32_to_cpu(trunc->seq);
2714a8599bd8SSage Weil 	u32 truncate_seq = le32_to_cpu(trunc->truncate_seq);
2715a8599bd8SSage Weil 	u64 truncate_size = le64_to_cpu(trunc->truncate_size);
2716a8599bd8SSage Weil 	u64 size = le64_to_cpu(trunc->size);
2717a8599bd8SSage Weil 	int implemented = 0;
2718a8599bd8SSage Weil 	int dirty = __ceph_caps_dirty(ci);
2719a8599bd8SSage Weil 	int issued = __ceph_caps_issued(ceph_inode(inode), &implemented);
2720a8599bd8SSage Weil 	int queue_trunc = 0;
2721a8599bd8SSage Weil 
2722a8599bd8SSage Weil 	issued |= implemented | dirty;
2723a8599bd8SSage Weil 
2724a8599bd8SSage Weil 	dout("handle_cap_trunc inode %p mds%d seq %d to %lld seq %d\n",
2725a8599bd8SSage Weil 	     inode, mds, seq, truncate_size, truncate_seq);
2726a8599bd8SSage Weil 	queue_trunc = ceph_fill_file_size(inode, issued,
2727a8599bd8SSage Weil 					  truncate_seq, truncate_size, size);
2728be655596SSage Weil 	spin_unlock(&ci->i_ceph_lock);
2729a8599bd8SSage Weil 
273099ccbd22SMilosz Tanski 	if (queue_trunc) {
27313c6f6b79SSage Weil 		ceph_queue_vmtruncate(inode);
273299ccbd22SMilosz Tanski 		ceph_fscache_invalidate(inode);
273399ccbd22SMilosz Tanski 	}
2734a8599bd8SSage Weil }
2735a8599bd8SSage Weil 
2736a8599bd8SSage Weil /*
2737a8599bd8SSage Weil  * Handle EXPORT from MDS.  Cap is being migrated _from_ this mds to a
2738a8599bd8SSage Weil  * different one.  If we are the most recent migration we've seen (as
2739a8599bd8SSage Weil  * indicated by mseq), make note of the migrating cap bits for the
2740a8599bd8SSage Weil  * duration (until we see the corresponding IMPORT).
2741a8599bd8SSage Weil  *
2742a8599bd8SSage Weil  * caller holds s_mutex
2743a8599bd8SSage Weil  */
2744a8599bd8SSage Weil static void handle_cap_export(struct inode *inode, struct ceph_mds_caps *ex,
2745154f42c2SSage Weil 			      struct ceph_mds_session *session,
2746154f42c2SSage Weil 			      int *open_target_sessions)
2747a8599bd8SSage Weil {
2748db354052SSage Weil 	struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
2749a8599bd8SSage Weil 	struct ceph_inode_info *ci = ceph_inode(inode);
2750a8599bd8SSage Weil 	int mds = session->s_mds;
2751a8599bd8SSage Weil 	unsigned mseq = le32_to_cpu(ex->migrate_seq);
2752a8599bd8SSage Weil 	struct ceph_cap *cap = NULL, *t;
2753a8599bd8SSage Weil 	struct rb_node *p;
2754a8599bd8SSage Weil 	int remember = 1;
2755a8599bd8SSage Weil 
2756a8599bd8SSage Weil 	dout("handle_cap_export inode %p ci %p mds%d mseq %d\n",
2757a8599bd8SSage Weil 	     inode, ci, mds, mseq);
2758a8599bd8SSage Weil 
2759be655596SSage Weil 	spin_lock(&ci->i_ceph_lock);
2760a8599bd8SSage Weil 
2761a8599bd8SSage Weil 	/* make sure we haven't seen a higher mseq */
2762a8599bd8SSage Weil 	for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
2763a8599bd8SSage Weil 		t = rb_entry(p, struct ceph_cap, ci_node);
2764a8599bd8SSage Weil 		if (ceph_seq_cmp(t->mseq, mseq) > 0) {
2765a8599bd8SSage Weil 			dout(" higher mseq on cap from mds%d\n",
2766a8599bd8SSage Weil 			     t->session->s_mds);
2767a8599bd8SSage Weil 			remember = 0;
2768a8599bd8SSage Weil 		}
2769a8599bd8SSage Weil 		if (t->session->s_mds == mds)
2770a8599bd8SSage Weil 			cap = t;
2771a8599bd8SSage Weil 	}
2772a8599bd8SSage Weil 
2773a8599bd8SSage Weil 	if (cap) {
2774a8599bd8SSage Weil 		if (remember) {
2775a8599bd8SSage Weil 			/* make note */
2776a8599bd8SSage Weil 			ci->i_cap_exporting_mds = mds;
2777a8599bd8SSage Weil 			ci->i_cap_exporting_mseq = mseq;
2778a8599bd8SSage Weil 			ci->i_cap_exporting_issued = cap->issued;
2779154f42c2SSage Weil 
2780154f42c2SSage Weil 			/*
2781154f42c2SSage Weil 			 * make sure we have open sessions with all possible
2782154f42c2SSage Weil 			 * export targets, so that we get the matching IMPORT
2783154f42c2SSage Weil 			 */
2784154f42c2SSage Weil 			*open_target_sessions = 1;
2785db354052SSage Weil 
2786db354052SSage Weil 			/*
2787db354052SSage Weil 			 * we can't flush dirty caps that we've seen the
2788db354052SSage Weil 			 * EXPORT but no IMPORT for
2789db354052SSage Weil 			 */
2790db354052SSage Weil 			spin_lock(&mdsc->cap_dirty_lock);
2791db354052SSage Weil 			if (!list_empty(&ci->i_dirty_item)) {
2792db354052SSage Weil 				dout(" moving %p to cap_dirty_migrating\n",
2793db354052SSage Weil 				     inode);
2794db354052SSage Weil 				list_move(&ci->i_dirty_item,
2795db354052SSage Weil 					  &mdsc->cap_dirty_migrating);
2796db354052SSage Weil 			}
2797db354052SSage Weil 			spin_unlock(&mdsc->cap_dirty_lock);
2798a8599bd8SSage Weil 		}
2799a096b09aSYan, Zheng 		__ceph_remove_cap(cap, false);
2800a8599bd8SSage Weil 	}
28014ea0043aSSage Weil 	/* else, we already released it */
2802a8599bd8SSage Weil 
2803be655596SSage Weil 	spin_unlock(&ci->i_ceph_lock);
2804a8599bd8SSage Weil }
2805a8599bd8SSage Weil 
2806a8599bd8SSage Weil /*
2807a8599bd8SSage Weil  * Handle cap IMPORT.  If there are temp bits from an older EXPORT,
2808a8599bd8SSage Weil  * clean them up.
2809a8599bd8SSage Weil  *
2810a8599bd8SSage Weil  * caller holds s_mutex.
2811a8599bd8SSage Weil  */
2812a8599bd8SSage Weil static void handle_cap_import(struct ceph_mds_client *mdsc,
2813a8599bd8SSage Weil 			      struct inode *inode, struct ceph_mds_caps *im,
2814a8599bd8SSage Weil 			      struct ceph_mds_session *session,
2815a8599bd8SSage Weil 			      void *snaptrace, int snaptrace_len)
2816a8599bd8SSage Weil {
2817a8599bd8SSage Weil 	struct ceph_inode_info *ci = ceph_inode(inode);
2818a8599bd8SSage Weil 	int mds = session->s_mds;
2819a8599bd8SSage Weil 	unsigned issued = le32_to_cpu(im->caps);
2820a8599bd8SSage Weil 	unsigned wanted = le32_to_cpu(im->wanted);
2821a8599bd8SSage Weil 	unsigned seq = le32_to_cpu(im->seq);
2822a8599bd8SSage Weil 	unsigned mseq = le32_to_cpu(im->migrate_seq);
2823a8599bd8SSage Weil 	u64 realmino = le64_to_cpu(im->realm);
2824a8599bd8SSage Weil 	u64 cap_id = le64_to_cpu(im->cap_id);
2825a8599bd8SSage Weil 
2826a8599bd8SSage Weil 	if (ci->i_cap_exporting_mds >= 0 &&
2827a8599bd8SSage Weil 	    ceph_seq_cmp(ci->i_cap_exporting_mseq, mseq) < 0) {
2828a8599bd8SSage Weil 		dout("handle_cap_import inode %p ci %p mds%d mseq %d"
2829a8599bd8SSage Weil 		     " - cleared exporting from mds%d\n",
2830a8599bd8SSage Weil 		     inode, ci, mds, mseq,
2831a8599bd8SSage Weil 		     ci->i_cap_exporting_mds);
2832a8599bd8SSage Weil 		ci->i_cap_exporting_issued = 0;
2833a8599bd8SSage Weil 		ci->i_cap_exporting_mseq = 0;
2834a8599bd8SSage Weil 		ci->i_cap_exporting_mds = -1;
2835db354052SSage Weil 
2836db354052SSage Weil 		spin_lock(&mdsc->cap_dirty_lock);
2837db354052SSage Weil 		if (!list_empty(&ci->i_dirty_item)) {
2838db354052SSage Weil 			dout(" moving %p back to cap_dirty\n", inode);
2839db354052SSage Weil 			list_move(&ci->i_dirty_item, &mdsc->cap_dirty);
2840db354052SSage Weil 		}
2841db354052SSage Weil 		spin_unlock(&mdsc->cap_dirty_lock);
2842a8599bd8SSage Weil 	} else {
2843a8599bd8SSage Weil 		dout("handle_cap_import inode %p ci %p mds%d mseq %d\n",
2844a8599bd8SSage Weil 		     inode, ci, mds, mseq);
2845a8599bd8SSage Weil 	}
2846a8599bd8SSage Weil 
2847a8599bd8SSage Weil 	down_write(&mdsc->snap_rwsem);
2848a8599bd8SSage Weil 	ceph_update_snap_trace(mdsc, snaptrace, snaptrace+snaptrace_len,
2849a8599bd8SSage Weil 			       false);
2850a8599bd8SSage Weil 	downgrade_write(&mdsc->snap_rwsem);
2851a8599bd8SSage Weil 	ceph_add_cap(inode, session, cap_id, -1,
2852a8599bd8SSage Weil 		     issued, wanted, seq, mseq, realmino, CEPH_CAP_FLAG_AUTH,
2853a8599bd8SSage Weil 		     NULL /* no caps context */);
2854088b3f5eSSage Weil 	kick_flushing_inode_caps(mdsc, session, inode);
2855a8599bd8SSage Weil 	up_read(&mdsc->snap_rwsem);
2856feb4cc9bSSage Weil 
2857feb4cc9bSSage Weil 	/* make sure we re-request max_size, if necessary */
2858be655596SSage Weil 	spin_lock(&ci->i_ceph_lock);
28590e5e1774SYan, Zheng 	ci->i_wanted_max_size = 0;  /* reset */
2860feb4cc9bSSage Weil 	ci->i_requested_max_size = 0;
2861be655596SSage Weil 	spin_unlock(&ci->i_ceph_lock);
2862a8599bd8SSage Weil }
2863a8599bd8SSage Weil 
2864a8599bd8SSage Weil /*
2865a8599bd8SSage Weil  * Handle a caps message from the MDS.
2866a8599bd8SSage Weil  *
2867a8599bd8SSage Weil  * Identify the appropriate session, inode, and call the right handler
2868a8599bd8SSage Weil  * based on the cap op.
2869a8599bd8SSage Weil  */
2870a8599bd8SSage Weil void ceph_handle_caps(struct ceph_mds_session *session,
2871a8599bd8SSage Weil 		      struct ceph_msg *msg)
2872a8599bd8SSage Weil {
2873a8599bd8SSage Weil 	struct ceph_mds_client *mdsc = session->s_mdsc;
28743d14c5d2SYehuda Sadeh 	struct super_block *sb = mdsc->fsc->sb;
2875a8599bd8SSage Weil 	struct inode *inode;
2876be655596SSage Weil 	struct ceph_inode_info *ci;
2877a8599bd8SSage Weil 	struct ceph_cap *cap;
2878a8599bd8SSage Weil 	struct ceph_mds_caps *h;
28792600d2ddSSage Weil 	int mds = session->s_mds;
2880a8599bd8SSage Weil 	int op;
28813d7ded4dSSage Weil 	u32 seq, mseq;
2882a8599bd8SSage Weil 	struct ceph_vino vino;
2883a8599bd8SSage Weil 	u64 cap_id;
2884a8599bd8SSage Weil 	u64 size, max_size;
28856df058c0SSage Weil 	u64 tid;
288670edb55bSSage Weil 	void *snaptrace;
2887ce1fbc8dSSage Weil 	size_t snaptrace_len;
2888ce1fbc8dSSage Weil 	void *flock;
2889ce1fbc8dSSage Weil 	u32 flock_len;
2890154f42c2SSage Weil 	int open_target_sessions = 0;
2891a8599bd8SSage Weil 
2892a8599bd8SSage Weil 	dout("handle_caps from mds%d\n", mds);
2893a8599bd8SSage Weil 
2894a8599bd8SSage Weil 	/* decode */
28956df058c0SSage Weil 	tid = le64_to_cpu(msg->hdr.tid);
2896a8599bd8SSage Weil 	if (msg->front.iov_len < sizeof(*h))
2897a8599bd8SSage Weil 		goto bad;
2898a8599bd8SSage Weil 	h = msg->front.iov_base;
2899a8599bd8SSage Weil 	op = le32_to_cpu(h->op);
2900a8599bd8SSage Weil 	vino.ino = le64_to_cpu(h->ino);
2901a8599bd8SSage Weil 	vino.snap = CEPH_NOSNAP;
2902a8599bd8SSage Weil 	cap_id = le64_to_cpu(h->cap_id);
2903a8599bd8SSage Weil 	seq = le32_to_cpu(h->seq);
29043d7ded4dSSage Weil 	mseq = le32_to_cpu(h->migrate_seq);
2905a8599bd8SSage Weil 	size = le64_to_cpu(h->size);
2906a8599bd8SSage Weil 	max_size = le64_to_cpu(h->max_size);
2907a8599bd8SSage Weil 
2908ce1fbc8dSSage Weil 	snaptrace = h + 1;
2909ce1fbc8dSSage Weil 	snaptrace_len = le32_to_cpu(h->snap_trace_len);
2910ce1fbc8dSSage Weil 
2911ce1fbc8dSSage Weil 	if (le16_to_cpu(msg->hdr.version) >= 2) {
2912ce1fbc8dSSage Weil 		void *p, *end;
2913ce1fbc8dSSage Weil 
2914ce1fbc8dSSage Weil 		p = snaptrace + snaptrace_len;
2915ce1fbc8dSSage Weil 		end = msg->front.iov_base + msg->front.iov_len;
2916ce1fbc8dSSage Weil 		ceph_decode_32_safe(&p, end, flock_len, bad);
2917ce1fbc8dSSage Weil 		flock = p;
2918ce1fbc8dSSage Weil 	} else {
2919ce1fbc8dSSage Weil 		flock = NULL;
2920ce1fbc8dSSage Weil 		flock_len = 0;
2921ce1fbc8dSSage Weil 	}
2922ce1fbc8dSSage Weil 
2923a8599bd8SSage Weil 	mutex_lock(&session->s_mutex);
2924a8599bd8SSage Weil 	session->s_seq++;
2925a8599bd8SSage Weil 	dout(" mds%d seq %lld cap seq %u\n", session->s_mds, session->s_seq,
2926a8599bd8SSage Weil 	     (unsigned)seq);
2927a8599bd8SSage Weil 
292866f58691SYan, Zheng 	if (op == CEPH_CAP_OP_IMPORT)
292966f58691SYan, Zheng 		ceph_add_cap_releases(mdsc, session);
293066f58691SYan, Zheng 
2931a8599bd8SSage Weil 	/* lookup ino */
2932a8599bd8SSage Weil 	inode = ceph_find_inode(sb, vino);
2933be655596SSage Weil 	ci = ceph_inode(inode);
2934a8599bd8SSage Weil 	dout(" op %s ino %llx.%llx inode %p\n", ceph_cap_op_name(op), vino.ino,
2935a8599bd8SSage Weil 	     vino.snap, inode);
2936a8599bd8SSage Weil 	if (!inode) {
2937a8599bd8SSage Weil 		dout(" i don't have ino %llx\n", vino.ino);
29383d7ded4dSSage Weil 
2939a096b09aSYan, Zheng 		if (op == CEPH_CAP_OP_IMPORT) {
2940a096b09aSYan, Zheng 			spin_lock(&session->s_cap_lock);
29413d7ded4dSSage Weil 			__queue_cap_release(session, vino.ino, cap_id,
29423d7ded4dSSage Weil 					    mseq, seq);
2943a096b09aSYan, Zheng 			spin_unlock(&session->s_cap_lock);
2944a096b09aSYan, Zheng 		}
294521b559deSGreg Farnum 		goto flush_cap_releases;
2946a8599bd8SSage Weil 	}
2947a8599bd8SSage Weil 
2948a8599bd8SSage Weil 	/* these will work even if we don't have a cap yet */
2949a8599bd8SSage Weil 	switch (op) {
2950a8599bd8SSage Weil 	case CEPH_CAP_OP_FLUSHSNAP_ACK:
29516df058c0SSage Weil 		handle_cap_flushsnap_ack(inode, tid, h, session);
2952a8599bd8SSage Weil 		goto done;
2953a8599bd8SSage Weil 
2954a8599bd8SSage Weil 	case CEPH_CAP_OP_EXPORT:
2955154f42c2SSage Weil 		handle_cap_export(inode, h, session, &open_target_sessions);
2956a8599bd8SSage Weil 		goto done;
2957a8599bd8SSage Weil 
2958a8599bd8SSage Weil 	case CEPH_CAP_OP_IMPORT:
2959a8599bd8SSage Weil 		handle_cap_import(mdsc, inode, h, session,
2960ce1fbc8dSSage Weil 				  snaptrace, snaptrace_len);
2961a8599bd8SSage Weil 	}
2962a8599bd8SSage Weil 
2963a8599bd8SSage Weil 	/* the rest require a cap */
2964be655596SSage Weil 	spin_lock(&ci->i_ceph_lock);
2965a8599bd8SSage Weil 	cap = __get_cap_for_mds(ceph_inode(inode), mds);
2966a8599bd8SSage Weil 	if (!cap) {
29679dbd412fSSage Weil 		dout(" no cap on %p ino %llx.%llx from mds%d\n",
2968a8599bd8SSage Weil 		     inode, ceph_ino(inode), ceph_snap(inode), mds);
2969be655596SSage Weil 		spin_unlock(&ci->i_ceph_lock);
297021b559deSGreg Farnum 		goto flush_cap_releases;
2971a8599bd8SSage Weil 	}
2972a8599bd8SSage Weil 
2973be655596SSage Weil 	/* note that each of these drops i_ceph_lock for us */
2974a8599bd8SSage Weil 	switch (op) {
2975a8599bd8SSage Weil 	case CEPH_CAP_OP_REVOKE:
2976a8599bd8SSage Weil 	case CEPH_CAP_OP_GRANT:
29770e5e1774SYan, Zheng 	case CEPH_CAP_OP_IMPORT:
297815637c8bSSage Weil 		handle_cap_grant(inode, h, session, cap, msg->middle);
297915637c8bSSage Weil 		goto done_unlocked;
2980a8599bd8SSage Weil 
2981a8599bd8SSage Weil 	case CEPH_CAP_OP_FLUSH_ACK:
29826df058c0SSage Weil 		handle_cap_flush_ack(inode, tid, h, session, cap);
2983a8599bd8SSage Weil 		break;
2984a8599bd8SSage Weil 
2985a8599bd8SSage Weil 	case CEPH_CAP_OP_TRUNC:
2986a8599bd8SSage Weil 		handle_cap_trunc(inode, h, session);
2987a8599bd8SSage Weil 		break;
2988a8599bd8SSage Weil 
2989a8599bd8SSage Weil 	default:
2990be655596SSage Weil 		spin_unlock(&ci->i_ceph_lock);
2991a8599bd8SSage Weil 		pr_err("ceph_handle_caps: unknown cap op %d %s\n", op,
2992a8599bd8SSage Weil 		       ceph_cap_op_name(op));
2993a8599bd8SSage Weil 	}
2994a8599bd8SSage Weil 
299521b559deSGreg Farnum 	goto done;
299621b559deSGreg Farnum 
299721b559deSGreg Farnum flush_cap_releases:
299821b559deSGreg Farnum 	/*
299921b559deSGreg Farnum 	 * send any full release message to try to move things
300021b559deSGreg Farnum 	 * along for the mds (who clearly thinks we still have this
300121b559deSGreg Farnum 	 * cap).
300221b559deSGreg Farnum 	 */
300321b559deSGreg Farnum 	ceph_add_cap_releases(mdsc, session);
300421b559deSGreg Farnum 	ceph_send_cap_releases(mdsc, session);
300521b559deSGreg Farnum 
3006a8599bd8SSage Weil done:
3007a8599bd8SSage Weil 	mutex_unlock(&session->s_mutex);
300815637c8bSSage Weil done_unlocked:
3009a8599bd8SSage Weil 	if (inode)
3010a8599bd8SSage Weil 		iput(inode);
3011154f42c2SSage Weil 	if (open_target_sessions)
3012154f42c2SSage Weil 		ceph_mdsc_open_export_target_sessions(mdsc, session);
3013a8599bd8SSage Weil 	return;
3014a8599bd8SSage Weil 
3015a8599bd8SSage Weil bad:
3016a8599bd8SSage Weil 	pr_err("ceph_handle_caps: corrupt message\n");
30179ec7cab1SSage Weil 	ceph_msg_dump(msg);
3018a8599bd8SSage Weil 	return;
3019a8599bd8SSage Weil }
3020a8599bd8SSage Weil 
3021a8599bd8SSage Weil /*
3022a8599bd8SSage Weil  * Delayed work handler to process end of delayed cap release LRU list.
3023a8599bd8SSage Weil  */
3024afcdaea3SSage Weil void ceph_check_delayed_caps(struct ceph_mds_client *mdsc)
3025a8599bd8SSage Weil {
3026a8599bd8SSage Weil 	struct ceph_inode_info *ci;
3027a8599bd8SSage Weil 	int flags = CHECK_CAPS_NODELAY;
3028a8599bd8SSage Weil 
3029a8599bd8SSage Weil 	dout("check_delayed_caps\n");
3030a8599bd8SSage Weil 	while (1) {
3031a8599bd8SSage Weil 		spin_lock(&mdsc->cap_delay_lock);
3032a8599bd8SSage Weil 		if (list_empty(&mdsc->cap_delay_list))
3033a8599bd8SSage Weil 			break;
3034a8599bd8SSage Weil 		ci = list_first_entry(&mdsc->cap_delay_list,
3035a8599bd8SSage Weil 				      struct ceph_inode_info,
3036a8599bd8SSage Weil 				      i_cap_delay_list);
3037a8599bd8SSage Weil 		if ((ci->i_ceph_flags & CEPH_I_FLUSH) == 0 &&
3038a8599bd8SSage Weil 		    time_before(jiffies, ci->i_hold_caps_max))
3039a8599bd8SSage Weil 			break;
3040a8599bd8SSage Weil 		list_del_init(&ci->i_cap_delay_list);
3041a8599bd8SSage Weil 		spin_unlock(&mdsc->cap_delay_lock);
3042a8599bd8SSage Weil 		dout("check_delayed_caps on %p\n", &ci->vfs_inode);
3043a8599bd8SSage Weil 		ceph_check_caps(ci, flags, NULL);
3044a8599bd8SSage Weil 	}
3045a8599bd8SSage Weil 	spin_unlock(&mdsc->cap_delay_lock);
3046a8599bd8SSage Weil }
3047a8599bd8SSage Weil 
3048a8599bd8SSage Weil /*
3049afcdaea3SSage Weil  * Flush all dirty caps to the mds
3050afcdaea3SSage Weil  */
3051afcdaea3SSage Weil void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc)
3052afcdaea3SSage Weil {
3053db354052SSage Weil 	struct ceph_inode_info *ci;
3054db354052SSage Weil 	struct inode *inode;
3055afcdaea3SSage Weil 
3056afcdaea3SSage Weil 	dout("flush_dirty_caps\n");
3057afcdaea3SSage Weil 	spin_lock(&mdsc->cap_dirty_lock);
3058db354052SSage Weil 	while (!list_empty(&mdsc->cap_dirty)) {
3059db354052SSage Weil 		ci = list_first_entry(&mdsc->cap_dirty, struct ceph_inode_info,
3060afcdaea3SSage Weil 				      i_dirty_item);
306170b666c3SSage Weil 		inode = &ci->vfs_inode;
306270b666c3SSage Weil 		ihold(inode);
3063db354052SSage Weil 		dout("flush_dirty_caps %p\n", inode);
3064afcdaea3SSage Weil 		spin_unlock(&mdsc->cap_dirty_lock);
306570b666c3SSage Weil 		ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_FLUSH, NULL);
3066afcdaea3SSage Weil 		iput(inode);
3067afcdaea3SSage Weil 		spin_lock(&mdsc->cap_dirty_lock);
3068afcdaea3SSage Weil 	}
3069afcdaea3SSage Weil 	spin_unlock(&mdsc->cap_dirty_lock);
3070db354052SSage Weil 	dout("flush_dirty_caps done\n");
3071afcdaea3SSage Weil }
3072afcdaea3SSage Weil 
3073afcdaea3SSage Weil /*
3074a8599bd8SSage Weil  * Drop open file reference.  If we were the last open file,
3075a8599bd8SSage Weil  * we may need to release capabilities to the MDS (or schedule
3076a8599bd8SSage Weil  * their delayed release).
3077a8599bd8SSage Weil  */
3078a8599bd8SSage Weil void ceph_put_fmode(struct ceph_inode_info *ci, int fmode)
3079a8599bd8SSage Weil {
3080a8599bd8SSage Weil 	struct inode *inode = &ci->vfs_inode;
3081a8599bd8SSage Weil 	int last = 0;
3082a8599bd8SSage Weil 
3083be655596SSage Weil 	spin_lock(&ci->i_ceph_lock);
3084a8599bd8SSage Weil 	dout("put_fmode %p fmode %d %d -> %d\n", inode, fmode,
3085a8599bd8SSage Weil 	     ci->i_nr_by_mode[fmode], ci->i_nr_by_mode[fmode]-1);
3086a8599bd8SSage Weil 	BUG_ON(ci->i_nr_by_mode[fmode] == 0);
3087a8599bd8SSage Weil 	if (--ci->i_nr_by_mode[fmode] == 0)
3088a8599bd8SSage Weil 		last++;
3089be655596SSage Weil 	spin_unlock(&ci->i_ceph_lock);
3090a8599bd8SSage Weil 
3091a8599bd8SSage Weil 	if (last && ci->i_vino.snap == CEPH_NOSNAP)
3092a8599bd8SSage Weil 		ceph_check_caps(ci, 0, NULL);
3093a8599bd8SSage Weil }
3094a8599bd8SSage Weil 
3095a8599bd8SSage Weil /*
3096a8599bd8SSage Weil  * Helpers for embedding cap and dentry lease releases into mds
3097a8599bd8SSage Weil  * requests.
3098a8599bd8SSage Weil  *
3099a8599bd8SSage Weil  * @force is used by dentry_release (below) to force inclusion of a
3100a8599bd8SSage Weil  * record for the directory inode, even when there aren't any caps to
3101a8599bd8SSage Weil  * drop.
3102a8599bd8SSage Weil  */
3103a8599bd8SSage Weil int ceph_encode_inode_release(void **p, struct inode *inode,
3104a8599bd8SSage Weil 			      int mds, int drop, int unless, int force)
3105a8599bd8SSage Weil {
3106a8599bd8SSage Weil 	struct ceph_inode_info *ci = ceph_inode(inode);
3107a8599bd8SSage Weil 	struct ceph_cap *cap;
3108a8599bd8SSage Weil 	struct ceph_mds_request_release *rel = *p;
3109ec97f88bSSage Weil 	int used, dirty;
3110a8599bd8SSage Weil 	int ret = 0;
3111a8599bd8SSage Weil 
3112be655596SSage Weil 	spin_lock(&ci->i_ceph_lock);
3113916623daSSage Weil 	used = __ceph_caps_used(ci);
3114ec97f88bSSage Weil 	dirty = __ceph_caps_dirty(ci);
3115916623daSSage Weil 
3116ec97f88bSSage Weil 	dout("encode_inode_release %p mds%d used|dirty %s drop %s unless %s\n",
3117ec97f88bSSage Weil 	     inode, mds, ceph_cap_string(used|dirty), ceph_cap_string(drop),
3118916623daSSage Weil 	     ceph_cap_string(unless));
3119916623daSSage Weil 
3120ec97f88bSSage Weil 	/* only drop unused, clean caps */
3121ec97f88bSSage Weil 	drop &= ~(used | dirty);
3122916623daSSage Weil 
3123a8599bd8SSage Weil 	cap = __get_cap_for_mds(ci, mds);
3124a8599bd8SSage Weil 	if (cap && __cap_is_valid(cap)) {
3125a8599bd8SSage Weil 		if (force ||
3126a8599bd8SSage Weil 		    ((cap->issued & drop) &&
3127a8599bd8SSage Weil 		     (cap->issued & unless) == 0)) {
3128a8599bd8SSage Weil 			if ((cap->issued & drop) &&
3129a8599bd8SSage Weil 			    (cap->issued & unless) == 0) {
3130bb137f84SYan, Zheng 				int wanted = __ceph_caps_wanted(ci);
3131bb137f84SYan, Zheng 				if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0)
3132bb137f84SYan, Zheng 					wanted |= cap->mds_wanted;
3133bb137f84SYan, Zheng 				dout("encode_inode_release %p cap %p "
3134bb137f84SYan, Zheng 				     "%s -> %s, wanted %s -> %s\n", inode, cap,
3135a8599bd8SSage Weil 				     ceph_cap_string(cap->issued),
3136bb137f84SYan, Zheng 				     ceph_cap_string(cap->issued & ~drop),
3137bb137f84SYan, Zheng 				     ceph_cap_string(cap->mds_wanted),
3138bb137f84SYan, Zheng 				     ceph_cap_string(wanted));
3139bb137f84SYan, Zheng 
3140a8599bd8SSage Weil 				cap->issued &= ~drop;
3141a8599bd8SSage Weil 				cap->implemented &= ~drop;
3142bb137f84SYan, Zheng 				cap->mds_wanted = wanted;
3143a8599bd8SSage Weil 			} else {
3144a8599bd8SSage Weil 				dout("encode_inode_release %p cap %p %s"
3145a8599bd8SSage Weil 				     " (force)\n", inode, cap,
3146a8599bd8SSage Weil 				     ceph_cap_string(cap->issued));
3147a8599bd8SSage Weil 			}
3148a8599bd8SSage Weil 
3149a8599bd8SSage Weil 			rel->ino = cpu_to_le64(ceph_ino(inode));
3150a8599bd8SSage Weil 			rel->cap_id = cpu_to_le64(cap->cap_id);
3151a8599bd8SSage Weil 			rel->seq = cpu_to_le32(cap->seq);
3152a8599bd8SSage Weil 			rel->issue_seq = cpu_to_le32(cap->issue_seq),
3153a8599bd8SSage Weil 			rel->mseq = cpu_to_le32(cap->mseq);
3154a8599bd8SSage Weil 			rel->caps = cpu_to_le32(cap->issued);
3155a8599bd8SSage Weil 			rel->wanted = cpu_to_le32(cap->mds_wanted);
3156a8599bd8SSage Weil 			rel->dname_len = 0;
3157a8599bd8SSage Weil 			rel->dname_seq = 0;
3158a8599bd8SSage Weil 			*p += sizeof(*rel);
3159a8599bd8SSage Weil 			ret = 1;
3160a8599bd8SSage Weil 		} else {
3161a8599bd8SSage Weil 			dout("encode_inode_release %p cap %p %s\n",
3162a8599bd8SSage Weil 			     inode, cap, ceph_cap_string(cap->issued));
3163a8599bd8SSage Weil 		}
3164a8599bd8SSage Weil 	}
3165be655596SSage Weil 	spin_unlock(&ci->i_ceph_lock);
3166a8599bd8SSage Weil 	return ret;
3167a8599bd8SSage Weil }
3168a8599bd8SSage Weil 
3169a8599bd8SSage Weil int ceph_encode_dentry_release(void **p, struct dentry *dentry,
3170a8599bd8SSage Weil 			       int mds, int drop, int unless)
3171a8599bd8SSage Weil {
3172a8599bd8SSage Weil 	struct inode *dir = dentry->d_parent->d_inode;
3173a8599bd8SSage Weil 	struct ceph_mds_request_release *rel = *p;
3174a8599bd8SSage Weil 	struct ceph_dentry_info *di = ceph_dentry(dentry);
3175a8599bd8SSage Weil 	int force = 0;
3176a8599bd8SSage Weil 	int ret;
3177a8599bd8SSage Weil 
3178a8599bd8SSage Weil 	/*
3179a8599bd8SSage Weil 	 * force an record for the directory caps if we have a dentry lease.
3180be655596SSage Weil 	 * this is racy (can't take i_ceph_lock and d_lock together), but it
3181a8599bd8SSage Weil 	 * doesn't have to be perfect; the mds will revoke anything we don't
3182a8599bd8SSage Weil 	 * release.
3183a8599bd8SSage Weil 	 */
3184a8599bd8SSage Weil 	spin_lock(&dentry->d_lock);
3185a8599bd8SSage Weil 	if (di->lease_session && di->lease_session->s_mds == mds)
3186a8599bd8SSage Weil 		force = 1;
3187a8599bd8SSage Weil 	spin_unlock(&dentry->d_lock);
3188a8599bd8SSage Weil 
3189a8599bd8SSage Weil 	ret = ceph_encode_inode_release(p, dir, mds, drop, unless, force);
3190a8599bd8SSage Weil 
3191a8599bd8SSage Weil 	spin_lock(&dentry->d_lock);
3192a8599bd8SSage Weil 	if (ret && di->lease_session && di->lease_session->s_mds == mds) {
3193a8599bd8SSage Weil 		dout("encode_dentry_release %p mds%d seq %d\n",
3194a8599bd8SSage Weil 		     dentry, mds, (int)di->lease_seq);
3195a8599bd8SSage Weil 		rel->dname_len = cpu_to_le32(dentry->d_name.len);
3196a8599bd8SSage Weil 		memcpy(*p, dentry->d_name.name, dentry->d_name.len);
3197a8599bd8SSage Weil 		*p += dentry->d_name.len;
3198a8599bd8SSage Weil 		rel->dname_seq = cpu_to_le32(di->lease_seq);
31991dadcce3SSage Weil 		__ceph_mdsc_drop_dentry_lease(dentry);
3200a8599bd8SSage Weil 	}
3201a8599bd8SSage Weil 	spin_unlock(&dentry->d_lock);
3202a8599bd8SSage Weil 	return ret;
3203a8599bd8SSage Weil }
3204