xref: /linux/fs/ceph/caps.c (revision 9d0ad045533ee37a208991ac5baaf6641e60a9ed)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
3 
4 #include <linux/fs.h>
5 #include <linux/kernel.h>
6 #include <linux/sched/signal.h>
7 #include <linux/slab.h>
8 #include <linux/vmalloc.h>
9 #include <linux/wait.h>
10 #include <linux/writeback.h>
11 #include <linux/iversion.h>
12 #include <linux/filelock.h>
13 #include <linux/jiffies.h>
14 
15 #include "super.h"
16 #include "mds_client.h"
17 #include "cache.h"
18 #include "crypto.h"
19 #include <linux/ceph/decode.h>
20 #include <linux/ceph/messenger.h>
21 
22 /*
23  * Capability management
24  *
25  * The Ceph metadata servers control client access to inode metadata
26  * and file data by issuing capabilities, granting clients permission
27  * to read and/or write both inode field and file data to OSDs
28  * (storage nodes).  Each capability consists of a set of bits
29  * indicating which operations are allowed.
30  *
31  * If the client holds a *_SHARED cap, the client has a coherent value
32  * that can be safely read from the cached inode.
33  *
34  * In the case of a *_EXCL (exclusive) or FILE_WR capabilities, the
35  * client is allowed to change inode attributes (e.g., file size,
36  * mtime), note its dirty state in the ceph_cap, and asynchronously
37  * flush that metadata change to the MDS.
38  *
39  * In the event of a conflicting operation (perhaps by another
40  * client), the MDS will revoke the conflicting client capabilities.
41  *
42  * In order for a client to cache an inode, it must hold a capability
43  * with at least one MDS server.  When inodes are released, release
44  * notifications are batched and periodically sent en masse to the MDS
45  * cluster to release server state.
46  */
47 
48 static u64 __get_oldest_flush_tid(struct ceph_mds_client *mdsc);
49 static void __kick_flushing_caps(struct ceph_mds_client *mdsc,
50 				 struct ceph_mds_session *session,
51 				 struct ceph_inode_info *ci,
52 				 u64 oldest_flush_tid);
53 
54 /*
55  * Generate readable cap strings for debugging output.
56  */
57 #define MAX_CAP_STR 20
58 static char cap_str[MAX_CAP_STR][40];
59 static DEFINE_SPINLOCK(cap_str_lock);
60 static int last_cap_str;
61 
gcap_string(char * s,int c)62 static char *gcap_string(char *s, int c)
63 {
64 	if (c & CEPH_CAP_GSHARED)
65 		*s++ = 's';
66 	if (c & CEPH_CAP_GEXCL)
67 		*s++ = 'x';
68 	if (c & CEPH_CAP_GCACHE)
69 		*s++ = 'c';
70 	if (c & CEPH_CAP_GRD)
71 		*s++ = 'r';
72 	if (c & CEPH_CAP_GWR)
73 		*s++ = 'w';
74 	if (c & CEPH_CAP_GBUFFER)
75 		*s++ = 'b';
76 	if (c & CEPH_CAP_GWREXTEND)
77 		*s++ = 'a';
78 	if (c & CEPH_CAP_GLAZYIO)
79 		*s++ = 'l';
80 	return s;
81 }
82 
ceph_cap_string(int caps)83 const char *ceph_cap_string(int caps)
84 {
85 	int i;
86 	char *s;
87 	int c;
88 
89 	spin_lock(&cap_str_lock);
90 	i = last_cap_str++;
91 	if (last_cap_str == MAX_CAP_STR)
92 		last_cap_str = 0;
93 	spin_unlock(&cap_str_lock);
94 
95 	s = cap_str[i];
96 
97 	if (caps & CEPH_CAP_PIN)
98 		*s++ = 'p';
99 
100 	c = (caps >> CEPH_CAP_SAUTH) & 3;
101 	if (c) {
102 		*s++ = 'A';
103 		s = gcap_string(s, c);
104 	}
105 
106 	c = (caps >> CEPH_CAP_SLINK) & 3;
107 	if (c) {
108 		*s++ = 'L';
109 		s = gcap_string(s, c);
110 	}
111 
112 	c = (caps >> CEPH_CAP_SXATTR) & 3;
113 	if (c) {
114 		*s++ = 'X';
115 		s = gcap_string(s, c);
116 	}
117 
118 	c = caps >> CEPH_CAP_SFILE;
119 	if (c) {
120 		*s++ = 'F';
121 		s = gcap_string(s, c);
122 	}
123 
124 	if (s == cap_str[i])
125 		*s++ = '-';
126 	*s = 0;
127 	return cap_str[i];
128 }
129 
ceph_caps_init(struct ceph_mds_client * mdsc)130 void ceph_caps_init(struct ceph_mds_client *mdsc)
131 {
132 	INIT_LIST_HEAD(&mdsc->caps_list);
133 	spin_lock_init(&mdsc->caps_list_lock);
134 }
135 
ceph_caps_finalize(struct ceph_mds_client * mdsc)136 void ceph_caps_finalize(struct ceph_mds_client *mdsc)
137 {
138 	struct ceph_cap *cap;
139 
140 	spin_lock(&mdsc->caps_list_lock);
141 	while (!list_empty(&mdsc->caps_list)) {
142 		cap = list_first_entry(&mdsc->caps_list,
143 				       struct ceph_cap, caps_item);
144 		list_del(&cap->caps_item);
145 		kmem_cache_free(ceph_cap_cachep, cap);
146 	}
147 	mdsc->caps_total_count = 0;
148 	mdsc->caps_avail_count = 0;
149 	mdsc->caps_use_count = 0;
150 	mdsc->caps_reserve_count = 0;
151 	mdsc->caps_min_count = 0;
152 	spin_unlock(&mdsc->caps_list_lock);
153 }
154 
ceph_adjust_caps_max_min(struct ceph_mds_client * mdsc,struct ceph_mount_options * fsopt)155 void ceph_adjust_caps_max_min(struct ceph_mds_client *mdsc,
156 			      struct ceph_mount_options *fsopt)
157 {
158 	spin_lock(&mdsc->caps_list_lock);
159 	mdsc->caps_min_count = fsopt->max_readdir;
160 	if (mdsc->caps_min_count < 1024)
161 		mdsc->caps_min_count = 1024;
162 	mdsc->caps_use_max = fsopt->caps_max;
163 	if (mdsc->caps_use_max > 0 &&
164 	    mdsc->caps_use_max < mdsc->caps_min_count)
165 		mdsc->caps_use_max = mdsc->caps_min_count;
166 	spin_unlock(&mdsc->caps_list_lock);
167 }
168 
__ceph_unreserve_caps(struct ceph_mds_client * mdsc,int nr_caps)169 static void __ceph_unreserve_caps(struct ceph_mds_client *mdsc, int nr_caps)
170 {
171 	struct ceph_cap *cap;
172 	int i;
173 
174 	if (nr_caps) {
175 		BUG_ON(mdsc->caps_reserve_count < nr_caps);
176 		mdsc->caps_reserve_count -= nr_caps;
177 		if (mdsc->caps_avail_count >=
178 		    mdsc->caps_reserve_count + mdsc->caps_min_count) {
179 			mdsc->caps_total_count -= nr_caps;
180 			for (i = 0; i < nr_caps; i++) {
181 				cap = list_first_entry(&mdsc->caps_list,
182 					struct ceph_cap, caps_item);
183 				list_del(&cap->caps_item);
184 				kmem_cache_free(ceph_cap_cachep, cap);
185 			}
186 		} else {
187 			mdsc->caps_avail_count += nr_caps;
188 		}
189 
190 		doutc(mdsc->fsc->client,
191 		      "caps %d = %d used + %d resv + %d avail\n",
192 		      mdsc->caps_total_count, mdsc->caps_use_count,
193 		      mdsc->caps_reserve_count, mdsc->caps_avail_count);
194 		BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
195 						 mdsc->caps_reserve_count +
196 						 mdsc->caps_avail_count);
197 	}
198 }
199 
200 /*
201  * Called under mdsc->mutex.
202  */
ceph_reserve_caps(struct ceph_mds_client * mdsc,struct ceph_cap_reservation * ctx,int need)203 int ceph_reserve_caps(struct ceph_mds_client *mdsc,
204 		      struct ceph_cap_reservation *ctx, int need)
205 {
206 	struct ceph_client *cl = mdsc->fsc->client;
207 	int i, j;
208 	struct ceph_cap *cap;
209 	int have;
210 	int alloc = 0;
211 	int max_caps;
212 	int err = 0;
213 	bool trimmed = false;
214 	struct ceph_mds_session *s;
215 	LIST_HEAD(newcaps);
216 
217 	doutc(cl, "ctx=%p need=%d\n", ctx, need);
218 
219 	/* first reserve any caps that are already allocated */
220 	spin_lock(&mdsc->caps_list_lock);
221 	if (mdsc->caps_avail_count >= need)
222 		have = need;
223 	else
224 		have = mdsc->caps_avail_count;
225 	mdsc->caps_avail_count -= have;
226 	mdsc->caps_reserve_count += have;
227 	BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
228 					 mdsc->caps_reserve_count +
229 					 mdsc->caps_avail_count);
230 	spin_unlock(&mdsc->caps_list_lock);
231 
232 	for (i = have; i < need; ) {
233 		cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS);
234 		if (cap) {
235 			list_add(&cap->caps_item, &newcaps);
236 			alloc++;
237 			i++;
238 			continue;
239 		}
240 
241 		if (!trimmed) {
242 			for (j = 0; j < mdsc->max_sessions; j++) {
243 				s = __ceph_lookup_mds_session(mdsc, j);
244 				if (!s)
245 					continue;
246 				mutex_unlock(&mdsc->mutex);
247 
248 				mutex_lock(&s->s_mutex);
249 				max_caps = s->s_nr_caps - (need - i);
250 				ceph_trim_caps(mdsc, s, max_caps);
251 				mutex_unlock(&s->s_mutex);
252 
253 				ceph_put_mds_session(s);
254 				mutex_lock(&mdsc->mutex);
255 			}
256 			trimmed = true;
257 
258 			spin_lock(&mdsc->caps_list_lock);
259 			if (mdsc->caps_avail_count) {
260 				int more_have;
261 				if (mdsc->caps_avail_count >= need - i)
262 					more_have = need - i;
263 				else
264 					more_have = mdsc->caps_avail_count;
265 
266 				i += more_have;
267 				have += more_have;
268 				mdsc->caps_avail_count -= more_have;
269 				mdsc->caps_reserve_count += more_have;
270 
271 			}
272 			spin_unlock(&mdsc->caps_list_lock);
273 
274 			continue;
275 		}
276 
277 		pr_warn_client(cl, "ctx=%p ENOMEM need=%d got=%d\n", ctx, need,
278 			       have + alloc);
279 		err = -ENOMEM;
280 		break;
281 	}
282 
283 	if (!err) {
284 		BUG_ON(have + alloc != need);
285 		ctx->count = need;
286 		ctx->used = 0;
287 	}
288 
289 	spin_lock(&mdsc->caps_list_lock);
290 	mdsc->caps_total_count += alloc;
291 	mdsc->caps_reserve_count += alloc;
292 	list_splice(&newcaps, &mdsc->caps_list);
293 
294 	BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
295 					 mdsc->caps_reserve_count +
296 					 mdsc->caps_avail_count);
297 
298 	if (err)
299 		__ceph_unreserve_caps(mdsc, have + alloc);
300 
301 	spin_unlock(&mdsc->caps_list_lock);
302 
303 	doutc(cl, "ctx=%p %d = %d used + %d resv + %d avail\n", ctx,
304 	      mdsc->caps_total_count, mdsc->caps_use_count,
305 	      mdsc->caps_reserve_count, mdsc->caps_avail_count);
306 	return err;
307 }
308 
ceph_unreserve_caps(struct ceph_mds_client * mdsc,struct ceph_cap_reservation * ctx)309 void ceph_unreserve_caps(struct ceph_mds_client *mdsc,
310 			 struct ceph_cap_reservation *ctx)
311 {
312 	struct ceph_client *cl = mdsc->fsc->client;
313 	bool reclaim = false;
314 	if (!ctx->count)
315 		return;
316 
317 	doutc(cl, "ctx=%p count=%d\n", ctx, ctx->count);
318 	spin_lock(&mdsc->caps_list_lock);
319 	__ceph_unreserve_caps(mdsc, ctx->count);
320 	ctx->count = 0;
321 
322 	if (mdsc->caps_use_max > 0 &&
323 	    mdsc->caps_use_count > mdsc->caps_use_max)
324 		reclaim = true;
325 	spin_unlock(&mdsc->caps_list_lock);
326 
327 	if (reclaim)
328 		ceph_reclaim_caps_nr(mdsc, ctx->used);
329 }
330 
ceph_get_cap(struct ceph_mds_client * mdsc,struct ceph_cap_reservation * ctx)331 struct ceph_cap *ceph_get_cap(struct ceph_mds_client *mdsc,
332 			      struct ceph_cap_reservation *ctx)
333 {
334 	struct ceph_client *cl = mdsc->fsc->client;
335 	struct ceph_cap *cap = NULL;
336 
337 	/* temporary, until we do something about cap import/export */
338 	if (!ctx) {
339 		cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS);
340 		if (cap) {
341 			spin_lock(&mdsc->caps_list_lock);
342 			mdsc->caps_use_count++;
343 			mdsc->caps_total_count++;
344 			spin_unlock(&mdsc->caps_list_lock);
345 		} else {
346 			spin_lock(&mdsc->caps_list_lock);
347 			if (mdsc->caps_avail_count) {
348 				BUG_ON(list_empty(&mdsc->caps_list));
349 
350 				mdsc->caps_avail_count--;
351 				mdsc->caps_use_count++;
352 				cap = list_first_entry(&mdsc->caps_list,
353 						struct ceph_cap, caps_item);
354 				list_del(&cap->caps_item);
355 
356 				BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
357 				       mdsc->caps_reserve_count + mdsc->caps_avail_count);
358 			}
359 			spin_unlock(&mdsc->caps_list_lock);
360 		}
361 
362 		return cap;
363 	}
364 
365 	spin_lock(&mdsc->caps_list_lock);
366 	doutc(cl, "ctx=%p (%d) %d = %d used + %d resv + %d avail\n", ctx,
367 	      ctx->count, mdsc->caps_total_count, mdsc->caps_use_count,
368 	      mdsc->caps_reserve_count, mdsc->caps_avail_count);
369 	BUG_ON(!ctx->count);
370 	BUG_ON(ctx->count > mdsc->caps_reserve_count);
371 	BUG_ON(list_empty(&mdsc->caps_list));
372 
373 	ctx->count--;
374 	ctx->used++;
375 	mdsc->caps_reserve_count--;
376 	mdsc->caps_use_count++;
377 
378 	cap = list_first_entry(&mdsc->caps_list, struct ceph_cap, caps_item);
379 	list_del(&cap->caps_item);
380 
381 	BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
382 	       mdsc->caps_reserve_count + mdsc->caps_avail_count);
383 	spin_unlock(&mdsc->caps_list_lock);
384 	return cap;
385 }
386 
ceph_put_cap(struct ceph_mds_client * mdsc,struct ceph_cap * cap)387 void ceph_put_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap)
388 {
389 	struct ceph_client *cl = mdsc->fsc->client;
390 
391 	spin_lock(&mdsc->caps_list_lock);
392 	doutc(cl, "%p %d = %d used + %d resv + %d avail\n", cap,
393 	      mdsc->caps_total_count, mdsc->caps_use_count,
394 	      mdsc->caps_reserve_count, mdsc->caps_avail_count);
395 	mdsc->caps_use_count--;
396 	/*
397 	 * Keep some preallocated caps around (ceph_min_count), to
398 	 * avoid lots of free/alloc churn.
399 	 */
400 	if (mdsc->caps_avail_count >= mdsc->caps_reserve_count +
401 				      mdsc->caps_min_count) {
402 		mdsc->caps_total_count--;
403 		kmem_cache_free(ceph_cap_cachep, cap);
404 	} else {
405 		mdsc->caps_avail_count++;
406 		list_add(&cap->caps_item, &mdsc->caps_list);
407 	}
408 
409 	BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
410 	       mdsc->caps_reserve_count + mdsc->caps_avail_count);
411 	spin_unlock(&mdsc->caps_list_lock);
412 }
413 
ceph_reservation_status(struct ceph_fs_client * fsc,int * total,int * avail,int * used,int * reserved,int * min)414 void ceph_reservation_status(struct ceph_fs_client *fsc,
415 			     int *total, int *avail, int *used, int *reserved,
416 			     int *min)
417 {
418 	struct ceph_mds_client *mdsc = fsc->mdsc;
419 
420 	spin_lock(&mdsc->caps_list_lock);
421 
422 	if (total)
423 		*total = mdsc->caps_total_count;
424 	if (avail)
425 		*avail = mdsc->caps_avail_count;
426 	if (used)
427 		*used = mdsc->caps_use_count;
428 	if (reserved)
429 		*reserved = mdsc->caps_reserve_count;
430 	if (min)
431 		*min = mdsc->caps_min_count;
432 
433 	spin_unlock(&mdsc->caps_list_lock);
434 }
435 
436 /*
437  * Find ceph_cap for given mds, if any.
438  *
439  * Called with i_ceph_lock held.
440  */
__get_cap_for_mds(struct ceph_inode_info * ci,int mds)441 struct ceph_cap *__get_cap_for_mds(struct ceph_inode_info *ci, int mds)
442 {
443 	struct ceph_cap *cap;
444 	struct rb_node *n = ci->i_caps.rb_node;
445 
446 	while (n) {
447 		cap = rb_entry(n, struct ceph_cap, ci_node);
448 		if (mds < cap->mds)
449 			n = n->rb_left;
450 		else if (mds > cap->mds)
451 			n = n->rb_right;
452 		else
453 			return cap;
454 	}
455 	return NULL;
456 }
457 
ceph_get_cap_for_mds(struct ceph_inode_info * ci,int mds)458 struct ceph_cap *ceph_get_cap_for_mds(struct ceph_inode_info *ci, int mds)
459 {
460 	struct ceph_cap *cap;
461 
462 	spin_lock(&ci->i_ceph_lock);
463 	cap = __get_cap_for_mds(ci, mds);
464 	spin_unlock(&ci->i_ceph_lock);
465 	return cap;
466 }
467 
468 /*
469  * Called under i_ceph_lock.
470  */
__insert_cap_node(struct ceph_inode_info * ci,struct ceph_cap * new)471 static void __insert_cap_node(struct ceph_inode_info *ci,
472 			      struct ceph_cap *new)
473 {
474 	struct rb_node **p = &ci->i_caps.rb_node;
475 	struct rb_node *parent = NULL;
476 	struct ceph_cap *cap = NULL;
477 
478 	while (*p) {
479 		parent = *p;
480 		cap = rb_entry(parent, struct ceph_cap, ci_node);
481 		if (new->mds < cap->mds)
482 			p = &(*p)->rb_left;
483 		else if (new->mds > cap->mds)
484 			p = &(*p)->rb_right;
485 		else
486 			BUG();
487 	}
488 
489 	rb_link_node(&new->ci_node, parent, p);
490 	rb_insert_color(&new->ci_node, &ci->i_caps);
491 }
492 
493 /*
494  * (re)set cap hold timeouts, which control the delayed release
495  * of unused caps back to the MDS.  Should be called on cap use.
496  */
__cap_set_timeouts(struct ceph_mds_client * mdsc,struct ceph_inode_info * ci)497 static void __cap_set_timeouts(struct ceph_mds_client *mdsc,
498 			       struct ceph_inode_info *ci)
499 {
500 	struct inode *inode = &ci->netfs.inode;
501 	struct ceph_mount_options *opt = mdsc->fsc->mount_options;
502 
503 	ci->i_hold_caps_max = round_jiffies(jiffies +
504 					    opt->caps_wanted_delay_max * HZ);
505 	doutc(mdsc->fsc->client, "%p %llx.%llx %lu\n", inode,
506 	      ceph_vinop(inode), ci->i_hold_caps_max - jiffies);
507 }
508 
509 /*
510  * (Re)queue cap at the end of the delayed cap release list.
511  *
512  * If I_FLUSH is set, leave the inode at the front of the list.
513  *
514  * Caller holds i_ceph_lock
515  *    -> we take mdsc->cap_delay_lock
516  */
__cap_delay_requeue(struct ceph_mds_client * mdsc,struct ceph_inode_info * ci)517 static void __cap_delay_requeue(struct ceph_mds_client *mdsc,
518 				struct ceph_inode_info *ci)
519 {
520 	struct inode *inode = &ci->netfs.inode;
521 
522 	doutc(mdsc->fsc->client, "%p %llx.%llx flags 0x%lx at %lu\n",
523 	      inode, ceph_vinop(inode), ci->i_ceph_flags,
524 	      ci->i_hold_caps_max);
525 	if (!mdsc->stopping) {
526 		spin_lock(&mdsc->cap_delay_lock);
527 		if (!list_empty(&ci->i_cap_delay_list)) {
528 			if (ci->i_ceph_flags & CEPH_I_FLUSH)
529 				goto no_change;
530 			list_del_init(&ci->i_cap_delay_list);
531 		}
532 		__cap_set_timeouts(mdsc, ci);
533 		list_add_tail(&ci->i_cap_delay_list, &mdsc->cap_delay_list);
534 no_change:
535 		spin_unlock(&mdsc->cap_delay_lock);
536 	}
537 }
538 
539 /*
540  * Queue an inode for immediate writeback.  Mark inode with I_FLUSH,
541  * indicating we should send a cap message to flush dirty metadata
542  * asap, and move to the front of the delayed cap list.
543  */
__cap_delay_requeue_front(struct ceph_mds_client * mdsc,struct ceph_inode_info * ci)544 static void __cap_delay_requeue_front(struct ceph_mds_client *mdsc,
545 				      struct ceph_inode_info *ci)
546 {
547 	struct inode *inode = &ci->netfs.inode;
548 
549 	doutc(mdsc->fsc->client, "%p %llx.%llx\n", inode, ceph_vinop(inode));
550 	spin_lock(&mdsc->cap_delay_lock);
551 	ci->i_ceph_flags |= CEPH_I_FLUSH;
552 	if (!list_empty(&ci->i_cap_delay_list))
553 		list_del_init(&ci->i_cap_delay_list);
554 	list_add(&ci->i_cap_delay_list, &mdsc->cap_delay_list);
555 	spin_unlock(&mdsc->cap_delay_lock);
556 }
557 
558 /*
559  * Cancel delayed work on cap.
560  *
561  * Caller must hold i_ceph_lock.
562  */
__cap_delay_cancel(struct ceph_mds_client * mdsc,struct ceph_inode_info * ci)563 static void __cap_delay_cancel(struct ceph_mds_client *mdsc,
564 			       struct ceph_inode_info *ci)
565 {
566 	struct inode *inode = &ci->netfs.inode;
567 
568 	doutc(mdsc->fsc->client, "%p %llx.%llx\n", inode, ceph_vinop(inode));
569 	if (list_empty(&ci->i_cap_delay_list))
570 		return;
571 	spin_lock(&mdsc->cap_delay_lock);
572 	list_del_init(&ci->i_cap_delay_list);
573 	spin_unlock(&mdsc->cap_delay_lock);
574 }
575 
576 /* Common issue checks for add_cap, handle_cap_grant. */
__check_cap_issue(struct ceph_inode_info * ci,struct ceph_cap * cap,unsigned issued)577 static void __check_cap_issue(struct ceph_inode_info *ci, struct ceph_cap *cap,
578 			      unsigned issued)
579 {
580 	struct inode *inode = &ci->netfs.inode;
581 	struct ceph_client *cl = ceph_inode_to_client(inode);
582 
583 	unsigned had = __ceph_caps_issued(ci, NULL);
584 
585 	lockdep_assert_held(&ci->i_ceph_lock);
586 
587 	/*
588 	 * Each time we receive FILE_CACHE anew, we increment
589 	 * i_rdcache_gen.
590 	 */
591 	if (S_ISREG(ci->netfs.inode.i_mode) &&
592 	    (issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) &&
593 	    (had & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0) {
594 		ci->i_rdcache_gen++;
595 	}
596 
597 	/*
598 	 * If FILE_SHARED is newly issued, mark dir not complete. We don't
599 	 * know what happened to this directory while we didn't have the cap.
600 	 * If FILE_SHARED is being revoked, also mark dir not complete. It
601 	 * stops on-going cached readdir.
602 	 */
603 	if ((issued & CEPH_CAP_FILE_SHARED) != (had & CEPH_CAP_FILE_SHARED)) {
604 		if (issued & CEPH_CAP_FILE_SHARED)
605 			atomic_inc(&ci->i_shared_gen);
606 		if (S_ISDIR(ci->netfs.inode.i_mode)) {
607 			doutc(cl, " marking %p NOT complete\n", inode);
608 			__ceph_dir_clear_complete(ci);
609 		}
610 	}
611 
612 	/* Wipe saved layout if we're losing DIR_CREATE caps */
613 	if (S_ISDIR(ci->netfs.inode.i_mode) && (had & CEPH_CAP_DIR_CREATE) &&
614 		!(issued & CEPH_CAP_DIR_CREATE)) {
615 	     ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns));
616 	     memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout));
617 	}
618 }
619 
620 /**
621  * change_auth_cap_ses - move inode to appropriate lists when auth caps change
622  * @ci: inode to be moved
623  * @session: new auth caps session
624  */
change_auth_cap_ses(struct ceph_inode_info * ci,struct ceph_mds_session * session)625 void change_auth_cap_ses(struct ceph_inode_info *ci,
626 			 struct ceph_mds_session *session)
627 {
628 	lockdep_assert_held(&ci->i_ceph_lock);
629 
630 	if (list_empty(&ci->i_dirty_item) && list_empty(&ci->i_flushing_item))
631 		return;
632 
633 	spin_lock(&session->s_mdsc->cap_dirty_lock);
634 	if (!list_empty(&ci->i_dirty_item))
635 		list_move(&ci->i_dirty_item, &session->s_cap_dirty);
636 	if (!list_empty(&ci->i_flushing_item))
637 		list_move_tail(&ci->i_flushing_item, &session->s_cap_flushing);
638 	spin_unlock(&session->s_mdsc->cap_dirty_lock);
639 }
640 
641 /*
642  * Add a capability under the given MDS session.
643  *
644  * Caller should hold session snap_rwsem (read) and ci->i_ceph_lock
645  *
646  * @fmode is the open file mode, if we are opening a file, otherwise
647  * it is < 0.  (This is so we can atomically add the cap and add an
648  * open file reference to it.)
649  */
ceph_add_cap(struct inode * inode,struct ceph_mds_session * session,u64 cap_id,unsigned issued,unsigned wanted,unsigned seq,unsigned mseq,u64 realmino,int flags,struct ceph_cap ** new_cap)650 void ceph_add_cap(struct inode *inode,
651 		  struct ceph_mds_session *session, u64 cap_id,
652 		  unsigned issued, unsigned wanted,
653 		  unsigned seq, unsigned mseq, u64 realmino, int flags,
654 		  struct ceph_cap **new_cap)
655 {
656 	struct ceph_mds_client *mdsc = ceph_inode_to_fs_client(inode)->mdsc;
657 	struct ceph_client *cl = ceph_inode_to_client(inode);
658 	struct ceph_inode_info *ci = ceph_inode(inode);
659 	struct ceph_cap *cap;
660 	int mds = session->s_mds;
661 	int actual_wanted;
662 	u32 gen;
663 
664 	lockdep_assert_held(&ci->i_ceph_lock);
665 
666 	doutc(cl, "%p %llx.%llx mds%d cap %llx %s seq %d\n", inode,
667 	      ceph_vinop(inode), session->s_mds, cap_id,
668 	      ceph_cap_string(issued), seq);
669 
670 	gen = atomic_read(&session->s_cap_gen);
671 
672 	cap = __get_cap_for_mds(ci, mds);
673 	if (!cap) {
674 		cap = *new_cap;
675 		*new_cap = NULL;
676 
677 		cap->issued = 0;
678 		cap->implemented = 0;
679 		cap->mds = mds;
680 		cap->mds_wanted = 0;
681 		cap->mseq = 0;
682 
683 		cap->ci = ci;
684 		__insert_cap_node(ci, cap);
685 
686 		/* add to session cap list */
687 		cap->session = session;
688 		spin_lock(&session->s_cap_lock);
689 		list_add_tail(&cap->session_caps, &session->s_caps);
690 		session->s_nr_caps++;
691 		atomic64_inc(&mdsc->metric.total_caps);
692 		spin_unlock(&session->s_cap_lock);
693 	} else {
694 		spin_lock(&session->s_cap_lock);
695 		list_move_tail(&cap->session_caps, &session->s_caps);
696 		spin_unlock(&session->s_cap_lock);
697 
698 		if (cap->cap_gen < gen)
699 			cap->issued = cap->implemented = CEPH_CAP_PIN;
700 
701 		/*
702 		 * auth mds of the inode changed. we received the cap export
703 		 * message, but still haven't received the cap import message.
704 		 * handle_cap_export() updated the new auth MDS' cap.
705 		 *
706 		 * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing
707 		 * a message that was send before the cap import message. So
708 		 * don't remove caps.
709 		 */
710 		if (ceph_seq_cmp(seq, cap->seq) <= 0) {
711 			WARN_ON(cap != ci->i_auth_cap);
712 			WARN_ON(cap->cap_id != cap_id);
713 			seq = cap->seq;
714 			mseq = cap->mseq;
715 			issued |= cap->issued;
716 			flags |= CEPH_CAP_FLAG_AUTH;
717 		}
718 	}
719 
720 	if (!ci->i_snap_realm ||
721 	    ((flags & CEPH_CAP_FLAG_AUTH) &&
722 	     realmino != (u64)-1 && ci->i_snap_realm->ino != realmino)) {
723 		/*
724 		 * add this inode to the appropriate snap realm
725 		 */
726 		struct ceph_snap_realm *realm = ceph_lookup_snap_realm(mdsc,
727 							       realmino);
728 		if (realm)
729 			ceph_change_snap_realm(inode, realm);
730 		else
731 			WARN(1, "%s: couldn't find snap realm 0x%llx (ino 0x%llx oldrealm 0x%llx)\n",
732 			     __func__, realmino, ci->i_vino.ino,
733 			     ci->i_snap_realm ? ci->i_snap_realm->ino : 0);
734 	}
735 
736 	__check_cap_issue(ci, cap, issued);
737 
738 	/*
739 	 * If we are issued caps we don't want, or the mds' wanted
740 	 * value appears to be off, queue a check so we'll release
741 	 * later and/or update the mds wanted value.
742 	 */
743 	actual_wanted = __ceph_caps_wanted(ci);
744 	if ((wanted & ~actual_wanted) ||
745 	    (issued & ~actual_wanted & CEPH_CAP_ANY_WR)) {
746 		doutc(cl, "issued %s, mds wanted %s, actual %s, queueing\n",
747 		      ceph_cap_string(issued), ceph_cap_string(wanted),
748 		      ceph_cap_string(actual_wanted));
749 		__cap_delay_requeue(mdsc, ci);
750 	}
751 
752 	if (flags & CEPH_CAP_FLAG_AUTH) {
753 		if (!ci->i_auth_cap ||
754 		    ceph_seq_cmp(ci->i_auth_cap->mseq, mseq) < 0) {
755 			if (ci->i_auth_cap &&
756 			    ci->i_auth_cap->session != cap->session)
757 				change_auth_cap_ses(ci, cap->session);
758 			ci->i_auth_cap = cap;
759 			cap->mds_wanted = wanted;
760 		}
761 	} else {
762 		WARN_ON(ci->i_auth_cap == cap);
763 	}
764 
765 	doutc(cl, "inode %p %llx.%llx cap %p %s now %s seq %d mds%d\n",
766 	      inode, ceph_vinop(inode), cap, ceph_cap_string(issued),
767 	      ceph_cap_string(issued|cap->issued), seq, mds);
768 	cap->cap_id = cap_id;
769 	cap->issued = issued;
770 	cap->implemented |= issued;
771 	if (ceph_seq_cmp(mseq, cap->mseq) > 0)
772 		cap->mds_wanted = wanted;
773 	else
774 		cap->mds_wanted |= wanted;
775 	cap->seq = seq;
776 	cap->issue_seq = seq;
777 	cap->mseq = mseq;
778 	cap->cap_gen = gen;
779 	wake_up_all(&ci->i_cap_wq);
780 }
781 
782 /*
783  * Return true if cap has not timed out and belongs to the current
784  * generation of the MDS session (i.e. has not gone 'stale' due to
785  * us losing touch with the mds).
786  */
__cap_is_valid(struct ceph_cap * cap)787 static int __cap_is_valid(struct ceph_cap *cap)
788 {
789 	struct inode *inode = &cap->ci->netfs.inode;
790 	struct ceph_client *cl = cap->session->s_mdsc->fsc->client;
791 	unsigned long ttl;
792 	u32 gen;
793 
794 	gen = atomic_read(&cap->session->s_cap_gen);
795 	ttl = cap->session->s_cap_ttl;
796 
797 	if (cap->cap_gen < gen || time_after_eq(jiffies, ttl)) {
798 		doutc(cl, "%p %llx.%llx cap %p issued %s but STALE (gen %u vs %u)\n",
799 		      inode, ceph_vinop(inode), cap,
800 		      ceph_cap_string(cap->issued), cap->cap_gen, gen);
801 		return 0;
802 	}
803 
804 	return 1;
805 }
806 
807 /*
808  * Return set of valid cap bits issued to us.  Note that caps time
809  * out, and may be invalidated in bulk if the client session times out
810  * and session->s_cap_gen is bumped.
811  */
__ceph_caps_issued(struct ceph_inode_info * ci,int * implemented)812 int __ceph_caps_issued(struct ceph_inode_info *ci, int *implemented)
813 {
814 	struct inode *inode = &ci->netfs.inode;
815 	struct ceph_client *cl = ceph_inode_to_client(inode);
816 	int have = ci->i_snap_caps;
817 	struct ceph_cap *cap;
818 	struct rb_node *p;
819 
820 	if (implemented)
821 		*implemented = 0;
822 	for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
823 		cap = rb_entry(p, struct ceph_cap, ci_node);
824 		if (!__cap_is_valid(cap))
825 			continue;
826 		doutc(cl, "%p %llx.%llx cap %p issued %s\n", inode,
827 		      ceph_vinop(inode), cap, ceph_cap_string(cap->issued));
828 		have |= cap->issued;
829 		if (implemented)
830 			*implemented |= cap->implemented;
831 	}
832 	/*
833 	 * exclude caps issued by non-auth MDS, but are been revoking
834 	 * by the auth MDS. The non-auth MDS should be revoking/exporting
835 	 * these caps, but the message is delayed.
836 	 */
837 	if (ci->i_auth_cap) {
838 		cap = ci->i_auth_cap;
839 		have &= ~cap->implemented | cap->issued;
840 	}
841 	return have;
842 }
843 
844 /*
845  * Get cap bits issued by caps other than @ocap
846  */
__ceph_caps_issued_other(struct ceph_inode_info * ci,struct ceph_cap * ocap)847 int __ceph_caps_issued_other(struct ceph_inode_info *ci, struct ceph_cap *ocap)
848 {
849 	int have = ci->i_snap_caps;
850 	struct ceph_cap *cap;
851 	struct rb_node *p;
852 
853 	for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
854 		cap = rb_entry(p, struct ceph_cap, ci_node);
855 		if (cap == ocap)
856 			continue;
857 		if (!__cap_is_valid(cap))
858 			continue;
859 		have |= cap->issued;
860 	}
861 	return have;
862 }
863 
864 /*
865  * Move a cap to the end of the LRU (oldest caps at list head, newest
866  * at list tail).
867  */
__touch_cap(struct ceph_cap * cap)868 static void __touch_cap(struct ceph_cap *cap)
869 {
870 	struct inode *inode = &cap->ci->netfs.inode;
871 	struct ceph_mds_session *s = cap->session;
872 	struct ceph_client *cl = s->s_mdsc->fsc->client;
873 
874 	spin_lock(&s->s_cap_lock);
875 	if (!s->s_cap_iterator) {
876 		doutc(cl, "%p %llx.%llx cap %p mds%d\n", inode,
877 		      ceph_vinop(inode), cap, s->s_mds);
878 		list_move_tail(&cap->session_caps, &s->s_caps);
879 	} else {
880 		doutc(cl, "%p %llx.%llx cap %p mds%d NOP, iterating over caps\n",
881 		      inode, ceph_vinop(inode), cap, s->s_mds);
882 	}
883 	spin_unlock(&s->s_cap_lock);
884 }
885 
886 /*
887  * Check if we hold the given mask.  If so, move the cap(s) to the
888  * front of their respective LRUs.  (This is the preferred way for
889  * callers to check for caps they want.)
890  */
__ceph_caps_issued_mask(struct ceph_inode_info * ci,int mask,int touch)891 int __ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask, int touch)
892 {
893 	struct inode *inode = &ci->netfs.inode;
894 	struct ceph_client *cl = ceph_inode_to_client(inode);
895 	struct ceph_cap *cap;
896 	struct rb_node *p;
897 	int have = ci->i_snap_caps;
898 
899 	if ((have & mask) == mask) {
900 		doutc(cl, "mask %p %llx.%llx snap issued %s (mask %s)\n",
901 		      inode, ceph_vinop(inode), ceph_cap_string(have),
902 		      ceph_cap_string(mask));
903 		return 1;
904 	}
905 
906 	for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
907 		cap = rb_entry(p, struct ceph_cap, ci_node);
908 		if (!__cap_is_valid(cap))
909 			continue;
910 		if ((cap->issued & mask) == mask) {
911 			doutc(cl, "mask %p %llx.%llx cap %p issued %s (mask %s)\n",
912 			      inode, ceph_vinop(inode), cap,
913 			      ceph_cap_string(cap->issued),
914 			      ceph_cap_string(mask));
915 			if (touch)
916 				__touch_cap(cap);
917 			return 1;
918 		}
919 
920 		/* does a combination of caps satisfy mask? */
921 		have |= cap->issued;
922 		if ((have & mask) == mask) {
923 			doutc(cl, "mask %p %llx.%llx combo issued %s (mask %s)\n",
924 			      inode, ceph_vinop(inode),
925 			      ceph_cap_string(cap->issued),
926 			      ceph_cap_string(mask));
927 			if (touch) {
928 				struct rb_node *q;
929 
930 				/* touch this + preceding caps */
931 				__touch_cap(cap);
932 				for (q = rb_first(&ci->i_caps); q != p;
933 				     q = rb_next(q)) {
934 					cap = rb_entry(q, struct ceph_cap,
935 						       ci_node);
936 					if (!__cap_is_valid(cap))
937 						continue;
938 					if (cap->issued & mask)
939 						__touch_cap(cap);
940 				}
941 			}
942 			return 1;
943 		}
944 	}
945 
946 	return 0;
947 }
948 
__ceph_caps_issued_mask_metric(struct ceph_inode_info * ci,int mask,int touch)949 int __ceph_caps_issued_mask_metric(struct ceph_inode_info *ci, int mask,
950 				   int touch)
951 {
952 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(ci->netfs.inode.i_sb);
953 	int r;
954 
955 	r = __ceph_caps_issued_mask(ci, mask, touch);
956 	if (r)
957 		ceph_update_cap_hit(&fsc->mdsc->metric);
958 	else
959 		ceph_update_cap_mis(&fsc->mdsc->metric);
960 	return r;
961 }
962 
963 /*
964  * Return true if mask caps are currently being revoked by an MDS.
965  */
__ceph_caps_revoking_other(struct ceph_inode_info * ci,struct ceph_cap * ocap,int mask)966 int __ceph_caps_revoking_other(struct ceph_inode_info *ci,
967 			       struct ceph_cap *ocap, int mask)
968 {
969 	struct ceph_cap *cap;
970 	struct rb_node *p;
971 
972 	for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
973 		cap = rb_entry(p, struct ceph_cap, ci_node);
974 		if (cap != ocap &&
975 		    (cap->implemented & ~cap->issued & mask))
976 			return 1;
977 	}
978 	return 0;
979 }
980 
__ceph_caps_used(struct ceph_inode_info * ci)981 int __ceph_caps_used(struct ceph_inode_info *ci)
982 {
983 	int used = 0;
984 	if (ci->i_pin_ref)
985 		used |= CEPH_CAP_PIN;
986 	if (ci->i_rd_ref)
987 		used |= CEPH_CAP_FILE_RD;
988 	if (ci->i_rdcache_ref ||
989 	    (S_ISREG(ci->netfs.inode.i_mode) &&
990 	     ci->netfs.inode.i_data.nrpages))
991 		used |= CEPH_CAP_FILE_CACHE;
992 	if (ci->i_wr_ref)
993 		used |= CEPH_CAP_FILE_WR;
994 	if (ci->i_wb_ref || ci->i_wrbuffer_ref)
995 		used |= CEPH_CAP_FILE_BUFFER;
996 	if (ci->i_fx_ref)
997 		used |= CEPH_CAP_FILE_EXCL;
998 	return used;
999 }
1000 
1001 #define FMODE_WAIT_BIAS 1000
1002 
1003 /*
1004  * wanted, by virtue of open file modes
1005  */
__ceph_caps_file_wanted(struct ceph_inode_info * ci)1006 int __ceph_caps_file_wanted(struct ceph_inode_info *ci)
1007 {
1008 	const int PIN_SHIFT = ffs(CEPH_FILE_MODE_PIN);
1009 	const int RD_SHIFT = ffs(CEPH_FILE_MODE_RD);
1010 	const int WR_SHIFT = ffs(CEPH_FILE_MODE_WR);
1011 	const int LAZY_SHIFT = ffs(CEPH_FILE_MODE_LAZY);
1012 	struct ceph_mount_options *opt =
1013 		ceph_inode_to_fs_client(&ci->netfs.inode)->mount_options;
1014 	unsigned long used_cutoff = jiffies - opt->caps_wanted_delay_max * HZ;
1015 	unsigned long idle_cutoff = jiffies - opt->caps_wanted_delay_min * HZ;
1016 
1017 	if (S_ISDIR(ci->netfs.inode.i_mode)) {
1018 		int want = 0;
1019 
1020 		/* use used_cutoff here, to keep dir's wanted caps longer */
1021 		if (ci->i_nr_by_mode[RD_SHIFT] > 0 ||
1022 		    time_after(ci->i_last_rd, used_cutoff))
1023 			want |= CEPH_CAP_ANY_SHARED;
1024 
1025 		if (ci->i_nr_by_mode[WR_SHIFT] > 0 ||
1026 		    time_after(ci->i_last_wr, used_cutoff)) {
1027 			want |= CEPH_CAP_ANY_SHARED | CEPH_CAP_FILE_EXCL;
1028 			if (opt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS)
1029 				want |= CEPH_CAP_ANY_DIR_OPS;
1030 		}
1031 
1032 		if (want || ci->i_nr_by_mode[PIN_SHIFT] > 0)
1033 			want |= CEPH_CAP_PIN;
1034 
1035 		return want;
1036 	} else {
1037 		int bits = 0;
1038 
1039 		if (ci->i_nr_by_mode[RD_SHIFT] > 0) {
1040 			if (ci->i_nr_by_mode[RD_SHIFT] >= FMODE_WAIT_BIAS ||
1041 			    time_after(ci->i_last_rd, used_cutoff))
1042 				bits |= 1 << RD_SHIFT;
1043 		} else if (time_after(ci->i_last_rd, idle_cutoff)) {
1044 			bits |= 1 << RD_SHIFT;
1045 		}
1046 
1047 		if (ci->i_nr_by_mode[WR_SHIFT] > 0) {
1048 			if (ci->i_nr_by_mode[WR_SHIFT] >= FMODE_WAIT_BIAS ||
1049 			    time_after(ci->i_last_wr, used_cutoff))
1050 				bits |= 1 << WR_SHIFT;
1051 		} else if (time_after(ci->i_last_wr, idle_cutoff)) {
1052 			bits |= 1 << WR_SHIFT;
1053 		}
1054 
1055 		/* check lazyio only when read/write is wanted */
1056 		if ((bits & (CEPH_FILE_MODE_RDWR << 1)) &&
1057 		    ci->i_nr_by_mode[LAZY_SHIFT] > 0)
1058 			bits |= 1 << LAZY_SHIFT;
1059 
1060 		return bits ? ceph_caps_for_mode(bits >> 1) : 0;
1061 	}
1062 }
1063 
1064 /*
1065  * wanted, by virtue of open file modes AND cap refs (buffered/cached data)
1066  */
__ceph_caps_wanted(struct ceph_inode_info * ci)1067 int __ceph_caps_wanted(struct ceph_inode_info *ci)
1068 {
1069 	int w = __ceph_caps_file_wanted(ci) | __ceph_caps_used(ci);
1070 	if (S_ISDIR(ci->netfs.inode.i_mode)) {
1071 		/* we want EXCL if holding caps of dir ops */
1072 		if (w & CEPH_CAP_ANY_DIR_OPS)
1073 			w |= CEPH_CAP_FILE_EXCL;
1074 	} else {
1075 		/* we want EXCL if dirty data */
1076 		if (w & CEPH_CAP_FILE_BUFFER)
1077 			w |= CEPH_CAP_FILE_EXCL;
1078 	}
1079 	return w;
1080 }
1081 
1082 /*
1083  * Return caps we have registered with the MDS(s) as 'wanted'.
1084  */
__ceph_caps_mds_wanted(struct ceph_inode_info * ci,bool check)1085 int __ceph_caps_mds_wanted(struct ceph_inode_info *ci, bool check)
1086 {
1087 	struct ceph_cap *cap;
1088 	struct rb_node *p;
1089 	int mds_wanted = 0;
1090 
1091 	for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
1092 		cap = rb_entry(p, struct ceph_cap, ci_node);
1093 		if (check && !__cap_is_valid(cap))
1094 			continue;
1095 		if (cap == ci->i_auth_cap)
1096 			mds_wanted |= cap->mds_wanted;
1097 		else
1098 			mds_wanted |= (cap->mds_wanted & ~CEPH_CAP_ANY_FILE_WR);
1099 	}
1100 	return mds_wanted;
1101 }
1102 
ceph_is_any_caps(struct inode * inode)1103 int ceph_is_any_caps(struct inode *inode)
1104 {
1105 	struct ceph_inode_info *ci = ceph_inode(inode);
1106 	int ret;
1107 
1108 	spin_lock(&ci->i_ceph_lock);
1109 	ret = __ceph_is_any_real_caps(ci);
1110 	spin_unlock(&ci->i_ceph_lock);
1111 
1112 	return ret;
1113 }
1114 
1115 /*
1116  * Remove a cap.  Take steps to deal with a racing iterate_session_caps.
1117  *
1118  * caller should hold i_ceph_lock.
1119  * caller will not hold session s_mutex if called from destroy_inode.
1120  */
__ceph_remove_cap(struct ceph_cap * cap,bool queue_release)1121 void __ceph_remove_cap(struct ceph_cap *cap, bool queue_release)
1122 {
1123 	struct ceph_mds_session *session = cap->session;
1124 	struct ceph_client *cl = session->s_mdsc->fsc->client;
1125 	struct ceph_inode_info *ci = cap->ci;
1126 	struct inode *inode = &ci->netfs.inode;
1127 	struct ceph_mds_client *mdsc;
1128 	int removed = 0;
1129 
1130 	/* 'ci' being NULL means the remove have already occurred */
1131 	if (!ci) {
1132 		doutc(cl, "inode is NULL\n");
1133 		return;
1134 	}
1135 
1136 	lockdep_assert_held(&ci->i_ceph_lock);
1137 
1138 	doutc(cl, "%p from %p %llx.%llx\n", cap, inode, ceph_vinop(inode));
1139 
1140 	mdsc = ceph_inode_to_fs_client(&ci->netfs.inode)->mdsc;
1141 
1142 	/* remove from inode's cap rbtree, and clear auth cap */
1143 	rb_erase(&cap->ci_node, &ci->i_caps);
1144 	if (ci->i_auth_cap == cap)
1145 		ci->i_auth_cap = NULL;
1146 
1147 	/* remove from session list */
1148 	spin_lock(&session->s_cap_lock);
1149 	if (session->s_cap_iterator == cap) {
1150 		/* not yet, we are iterating over this very cap */
1151 		doutc(cl, "delaying %p removal from session %p\n", cap,
1152 		      cap->session);
1153 	} else {
1154 		list_del_init(&cap->session_caps);
1155 		session->s_nr_caps--;
1156 		atomic64_dec(&mdsc->metric.total_caps);
1157 		cap->session = NULL;
1158 		removed = 1;
1159 	}
1160 	/* protect backpointer with s_cap_lock: see iterate_session_caps */
1161 	cap->ci = NULL;
1162 
1163 	/*
1164 	 * s_cap_reconnect is protected by s_cap_lock. no one changes
1165 	 * s_cap_gen while session is in the reconnect state.
1166 	 */
1167 	if (queue_release &&
1168 	    (!session->s_cap_reconnect ||
1169 	     cap->cap_gen == atomic_read(&session->s_cap_gen))) {
1170 		cap->queue_release = 1;
1171 		if (removed) {
1172 			__ceph_queue_cap_release(session, cap);
1173 			removed = 0;
1174 		}
1175 	} else {
1176 		cap->queue_release = 0;
1177 	}
1178 	cap->cap_ino = ci->i_vino.ino;
1179 
1180 	spin_unlock(&session->s_cap_lock);
1181 
1182 	if (removed)
1183 		ceph_put_cap(mdsc, cap);
1184 
1185 	if (!__ceph_is_any_real_caps(ci)) {
1186 		/* when reconnect denied, we remove session caps forcibly,
1187 		 * i_wr_ref can be non-zero. If there are ongoing write,
1188 		 * keep i_snap_realm.
1189 		 */
1190 		if (ci->i_wr_ref == 0 && ci->i_snap_realm)
1191 			ceph_change_snap_realm(&ci->netfs.inode, NULL);
1192 
1193 		__cap_delay_cancel(mdsc, ci);
1194 	}
1195 }
1196 
ceph_remove_cap(struct ceph_mds_client * mdsc,struct ceph_cap * cap,bool queue_release)1197 void ceph_remove_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap,
1198 		     bool queue_release)
1199 {
1200 	struct ceph_inode_info *ci = cap->ci;
1201 	struct ceph_fs_client *fsc;
1202 
1203 	/* 'ci' being NULL means the remove have already occurred */
1204 	if (!ci) {
1205 		doutc(mdsc->fsc->client, "inode is NULL\n");
1206 		return;
1207 	}
1208 
1209 	lockdep_assert_held(&ci->i_ceph_lock);
1210 
1211 	fsc = ceph_inode_to_fs_client(&ci->netfs.inode);
1212 	WARN_ON_ONCE(ci->i_auth_cap == cap &&
1213 		     !list_empty(&ci->i_dirty_item) &&
1214 		     !fsc->blocklisted &&
1215 		     !ceph_inode_is_shutdown(&ci->netfs.inode));
1216 
1217 	__ceph_remove_cap(cap, queue_release);
1218 }
1219 
1220 struct cap_msg_args {
1221 	struct ceph_mds_session	*session;
1222 	u64			ino, cid, follows;
1223 	u64			flush_tid, oldest_flush_tid, size, max_size;
1224 	u64			xattr_version;
1225 	u64			change_attr;
1226 	struct ceph_buffer	*xattr_buf;
1227 	struct ceph_buffer	*old_xattr_buf;
1228 	struct timespec64	atime, mtime, ctime, btime;
1229 	int			op, caps, wanted, dirty;
1230 	u32			seq, issue_seq, mseq, time_warp_seq;
1231 	u32			flags;
1232 	kuid_t			uid;
1233 	kgid_t			gid;
1234 	umode_t			mode;
1235 	bool			inline_data;
1236 	bool			wake;
1237 	bool			encrypted;
1238 	u32			fscrypt_auth_len;
1239 	u8			fscrypt_auth[sizeof(struct ceph_fscrypt_auth)]; // for context
1240 };
1241 
1242 /* Marshal up the cap msg to the MDS */
encode_cap_msg(struct ceph_msg * msg,struct cap_msg_args * arg)1243 static void encode_cap_msg(struct ceph_msg *msg, struct cap_msg_args *arg)
1244 {
1245 	struct ceph_mds_caps *fc;
1246 	void *p;
1247 	struct ceph_mds_client *mdsc = arg->session->s_mdsc;
1248 	struct ceph_osd_client *osdc = &mdsc->fsc->client->osdc;
1249 
1250 	doutc(mdsc->fsc->client,
1251 	      "%s %llx %llx caps %s wanted %s dirty %s seq %u/%u"
1252 	      " tid %llu/%llu mseq %u follows %lld size %llu/%llu"
1253 	      " xattr_ver %llu xattr_len %d\n",
1254 	      ceph_cap_op_name(arg->op), arg->cid, arg->ino,
1255 	      ceph_cap_string(arg->caps), ceph_cap_string(arg->wanted),
1256 	      ceph_cap_string(arg->dirty), arg->seq, arg->issue_seq,
1257 	      arg->flush_tid, arg->oldest_flush_tid, arg->mseq, arg->follows,
1258 	      arg->size, arg->max_size, arg->xattr_version,
1259 	      arg->xattr_buf ? (int)arg->xattr_buf->vec.iov_len : 0);
1260 
1261 	msg->hdr.version = cpu_to_le16(12);
1262 	msg->hdr.tid = cpu_to_le64(arg->flush_tid);
1263 
1264 	fc = msg->front.iov_base;
1265 	memset(fc, 0, sizeof(*fc));
1266 
1267 	fc->cap_id = cpu_to_le64(arg->cid);
1268 	fc->op = cpu_to_le32(arg->op);
1269 	fc->seq = cpu_to_le32(arg->seq);
1270 	fc->issue_seq = cpu_to_le32(arg->issue_seq);
1271 	fc->migrate_seq = cpu_to_le32(arg->mseq);
1272 	fc->caps = cpu_to_le32(arg->caps);
1273 	fc->wanted = cpu_to_le32(arg->wanted);
1274 	fc->dirty = cpu_to_le32(arg->dirty);
1275 	fc->ino = cpu_to_le64(arg->ino);
1276 	fc->snap_follows = cpu_to_le64(arg->follows);
1277 
1278 #if IS_ENABLED(CONFIG_FS_ENCRYPTION)
1279 	if (arg->encrypted)
1280 		fc->size = cpu_to_le64(round_up(arg->size,
1281 						CEPH_FSCRYPT_BLOCK_SIZE));
1282 	else
1283 #endif
1284 		fc->size = cpu_to_le64(arg->size);
1285 	fc->max_size = cpu_to_le64(arg->max_size);
1286 	ceph_encode_timespec64(&fc->mtime, &arg->mtime);
1287 	ceph_encode_timespec64(&fc->atime, &arg->atime);
1288 	ceph_encode_timespec64(&fc->ctime, &arg->ctime);
1289 	fc->time_warp_seq = cpu_to_le32(arg->time_warp_seq);
1290 
1291 	fc->uid = cpu_to_le32(from_kuid(&init_user_ns, arg->uid));
1292 	fc->gid = cpu_to_le32(from_kgid(&init_user_ns, arg->gid));
1293 	fc->mode = cpu_to_le32(arg->mode);
1294 
1295 	fc->xattr_version = cpu_to_le64(arg->xattr_version);
1296 	if (arg->xattr_buf) {
1297 		msg->middle = ceph_buffer_get(arg->xattr_buf);
1298 		fc->xattr_len = cpu_to_le32(arg->xattr_buf->vec.iov_len);
1299 		msg->hdr.middle_len = cpu_to_le32(arg->xattr_buf->vec.iov_len);
1300 	}
1301 
1302 	p = fc + 1;
1303 	/* flock buffer size (version 2) */
1304 	ceph_encode_32(&p, 0);
1305 	/* inline version (version 4) */
1306 	ceph_encode_64(&p, arg->inline_data ? 0 : CEPH_INLINE_NONE);
1307 	/* inline data size */
1308 	ceph_encode_32(&p, 0);
1309 	/*
1310 	 * osd_epoch_barrier (version 5)
1311 	 * The epoch_barrier is protected osdc->lock, so READ_ONCE here in
1312 	 * case it was recently changed
1313 	 */
1314 	ceph_encode_32(&p, READ_ONCE(osdc->epoch_barrier));
1315 	/* oldest_flush_tid (version 6) */
1316 	ceph_encode_64(&p, arg->oldest_flush_tid);
1317 
1318 	/*
1319 	 * caller_uid/caller_gid (version 7)
1320 	 *
1321 	 * Currently, we don't properly track which caller dirtied the caps
1322 	 * last, and force a flush of them when there is a conflict. For now,
1323 	 * just set this to 0:0, to emulate how the MDS has worked up to now.
1324 	 */
1325 	ceph_encode_32(&p, 0);
1326 	ceph_encode_32(&p, 0);
1327 
1328 	/* pool namespace (version 8) (mds always ignores this) */
1329 	ceph_encode_32(&p, 0);
1330 
1331 	/* btime and change_attr (version 9) */
1332 	ceph_encode_timespec64(p, &arg->btime);
1333 	p += sizeof(struct ceph_timespec);
1334 	ceph_encode_64(&p, arg->change_attr);
1335 
1336 	/* Advisory flags (version 10) */
1337 	ceph_encode_32(&p, arg->flags);
1338 
1339 	/* dirstats (version 11) - these are r/o on the client */
1340 	ceph_encode_64(&p, 0);
1341 	ceph_encode_64(&p, 0);
1342 
1343 #if IS_ENABLED(CONFIG_FS_ENCRYPTION)
1344 	/*
1345 	 * fscrypt_auth and fscrypt_file (version 12)
1346 	 *
1347 	 * fscrypt_auth holds the crypto context (if any). fscrypt_file
1348 	 * tracks the real i_size as an __le64 field (and we use a rounded-up
1349 	 * i_size in the traditional size field).
1350 	 */
1351 	ceph_encode_32(&p, arg->fscrypt_auth_len);
1352 	ceph_encode_copy(&p, arg->fscrypt_auth, arg->fscrypt_auth_len);
1353 	ceph_encode_32(&p, sizeof(__le64));
1354 	ceph_encode_64(&p, arg->size);
1355 #else /* CONFIG_FS_ENCRYPTION */
1356 	ceph_encode_32(&p, 0);
1357 	ceph_encode_32(&p, 0);
1358 #endif /* CONFIG_FS_ENCRYPTION */
1359 }
1360 
1361 /*
1362  * Queue cap releases when an inode is dropped from our cache.
1363  */
__ceph_remove_caps(struct ceph_inode_info * ci)1364 void __ceph_remove_caps(struct ceph_inode_info *ci)
1365 {
1366 	struct inode *inode = &ci->netfs.inode;
1367 	struct ceph_mds_client *mdsc = ceph_inode_to_fs_client(inode)->mdsc;
1368 	struct rb_node *p;
1369 
1370 	/* lock i_ceph_lock, because ceph_d_revalidate(..., LOOKUP_RCU)
1371 	 * may call __ceph_caps_issued_mask() on a freeing inode. */
1372 	spin_lock(&ci->i_ceph_lock);
1373 	p = rb_first(&ci->i_caps);
1374 	while (p) {
1375 		struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node);
1376 		p = rb_next(p);
1377 		ceph_remove_cap(mdsc, cap, true);
1378 	}
1379 	spin_unlock(&ci->i_ceph_lock);
1380 }
1381 
1382 /*
1383  * Prepare to send a cap message to an MDS. Update the cap state, and populate
1384  * the arg struct with the parameters that will need to be sent. This should
1385  * be done under the i_ceph_lock to guard against changes to cap state.
1386  *
1387  * Make note of max_size reported/requested from mds, revoked caps
1388  * that have now been implemented.
1389  */
__prep_cap(struct cap_msg_args * arg,struct ceph_cap * cap,int op,int flags,int used,int want,int retain,int flushing,u64 flush_tid,u64 oldest_flush_tid)1390 static void __prep_cap(struct cap_msg_args *arg, struct ceph_cap *cap,
1391 		       int op, int flags, int used, int want, int retain,
1392 		       int flushing, u64 flush_tid, u64 oldest_flush_tid)
1393 {
1394 	struct ceph_inode_info *ci = cap->ci;
1395 	struct inode *inode = &ci->netfs.inode;
1396 	struct ceph_client *cl = ceph_inode_to_client(inode);
1397 	int held, revoking;
1398 
1399 	lockdep_assert_held(&ci->i_ceph_lock);
1400 
1401 	held = cap->issued | cap->implemented;
1402 	revoking = cap->implemented & ~cap->issued;
1403 	retain &= ~revoking;
1404 
1405 	doutc(cl, "%p %llx.%llx cap %p session %p %s -> %s (revoking %s)\n",
1406 	      inode, ceph_vinop(inode), cap, cap->session,
1407 	      ceph_cap_string(held), ceph_cap_string(held & retain),
1408 	      ceph_cap_string(revoking));
1409 	BUG_ON((retain & CEPH_CAP_PIN) == 0);
1410 
1411 	ci->i_ceph_flags &= ~CEPH_I_FLUSH;
1412 
1413 	cap->issued &= retain;  /* drop bits we don't want */
1414 	/*
1415 	 * Wake up any waiters on wanted -> needed transition. This is due to
1416 	 * the weird transition from buffered to sync IO... we need to flush
1417 	 * dirty pages _before_ allowing sync writes to avoid reordering.
1418 	 */
1419 	arg->wake = cap->implemented & ~cap->issued;
1420 	cap->implemented &= cap->issued | used;
1421 	cap->mds_wanted = want;
1422 
1423 	arg->session = cap->session;
1424 	arg->ino = ceph_vino(inode).ino;
1425 	arg->cid = cap->cap_id;
1426 	arg->follows = flushing ? ci->i_head_snapc->seq : 0;
1427 	arg->flush_tid = flush_tid;
1428 	arg->oldest_flush_tid = oldest_flush_tid;
1429 	arg->size = i_size_read(inode);
1430 	ci->i_reported_size = arg->size;
1431 	arg->max_size = ci->i_wanted_max_size;
1432 	if (cap == ci->i_auth_cap) {
1433 		if (want & CEPH_CAP_ANY_FILE_WR)
1434 			ci->i_requested_max_size = arg->max_size;
1435 		else
1436 			ci->i_requested_max_size = 0;
1437 	}
1438 
1439 	if (flushing & CEPH_CAP_XATTR_EXCL) {
1440 		arg->old_xattr_buf = __ceph_build_xattrs_blob(ci);
1441 		arg->xattr_version = ci->i_xattrs.version;
1442 		arg->xattr_buf = ceph_buffer_get(ci->i_xattrs.blob);
1443 	} else {
1444 		arg->xattr_buf = NULL;
1445 		arg->old_xattr_buf = NULL;
1446 	}
1447 
1448 	arg->mtime = inode_get_mtime(inode);
1449 	arg->atime = inode_get_atime(inode);
1450 	arg->ctime = inode_get_ctime(inode);
1451 	arg->btime = ci->i_btime;
1452 	arg->change_attr = inode_peek_iversion_raw(inode);
1453 
1454 	arg->op = op;
1455 	arg->caps = cap->implemented;
1456 	arg->wanted = want;
1457 	arg->dirty = flushing;
1458 
1459 	arg->seq = cap->seq;
1460 	arg->issue_seq = cap->issue_seq;
1461 	arg->mseq = cap->mseq;
1462 	arg->time_warp_seq = ci->i_time_warp_seq;
1463 
1464 	arg->uid = inode->i_uid;
1465 	arg->gid = inode->i_gid;
1466 	arg->mode = inode->i_mode;
1467 
1468 	arg->inline_data = ci->i_inline_version != CEPH_INLINE_NONE;
1469 	if (!(flags & CEPH_CLIENT_CAPS_PENDING_CAPSNAP) &&
1470 	    !list_empty(&ci->i_cap_snaps)) {
1471 		struct ceph_cap_snap *capsnap;
1472 		list_for_each_entry_reverse(capsnap, &ci->i_cap_snaps, ci_item) {
1473 			if (capsnap->cap_flush.tid)
1474 				break;
1475 			if (capsnap->need_flush) {
1476 				flags |= CEPH_CLIENT_CAPS_PENDING_CAPSNAP;
1477 				break;
1478 			}
1479 		}
1480 	}
1481 	arg->flags = flags;
1482 	arg->encrypted = IS_ENCRYPTED(inode);
1483 #if IS_ENABLED(CONFIG_FS_ENCRYPTION)
1484 	if (ci->fscrypt_auth_len &&
1485 	    WARN_ON_ONCE(ci->fscrypt_auth_len > sizeof(struct ceph_fscrypt_auth))) {
1486 		/* Don't set this if it's too big */
1487 		arg->fscrypt_auth_len = 0;
1488 	} else {
1489 		arg->fscrypt_auth_len = ci->fscrypt_auth_len;
1490 		memcpy(arg->fscrypt_auth, ci->fscrypt_auth,
1491 		       min_t(size_t, ci->fscrypt_auth_len,
1492 			     sizeof(arg->fscrypt_auth)));
1493 	}
1494 #endif /* CONFIG_FS_ENCRYPTION */
1495 }
1496 
1497 #if IS_ENABLED(CONFIG_FS_ENCRYPTION)
1498 #define CAP_MSG_FIXED_FIELDS (sizeof(struct ceph_mds_caps) + \
1499 		      4 + 8 + 4 + 4 + 8 + 4 + 4 + 4 + 8 + 8 + 4 + 8 + 8 + 4 + 4 + 8)
1500 
cap_msg_size(struct cap_msg_args * arg)1501 static inline int cap_msg_size(struct cap_msg_args *arg)
1502 {
1503 	return CAP_MSG_FIXED_FIELDS + arg->fscrypt_auth_len;
1504 }
1505 #else
1506 #define CAP_MSG_FIXED_FIELDS (sizeof(struct ceph_mds_caps) + \
1507 		      4 + 8 + 4 + 4 + 8 + 4 + 4 + 4 + 8 + 8 + 4 + 8 + 8 + 4 + 4)
1508 
cap_msg_size(struct cap_msg_args * arg)1509 static inline int cap_msg_size(struct cap_msg_args *arg)
1510 {
1511 	return CAP_MSG_FIXED_FIELDS;
1512 }
1513 #endif /* CONFIG_FS_ENCRYPTION */
1514 
1515 /*
1516  * Send a cap msg on the given inode.
1517  *
1518  * Caller should hold snap_rwsem (read), s_mutex.
1519  */
__send_cap(struct cap_msg_args * arg,struct ceph_inode_info * ci)1520 static void __send_cap(struct cap_msg_args *arg, struct ceph_inode_info *ci)
1521 {
1522 	struct ceph_msg *msg;
1523 	struct inode *inode = &ci->netfs.inode;
1524 	struct ceph_client *cl = ceph_inode_to_client(inode);
1525 
1526 	msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPS, cap_msg_size(arg), GFP_NOFS,
1527 			   false);
1528 	if (!msg) {
1529 		pr_err_client(cl,
1530 			      "error allocating cap msg: ino (%llx.%llx)"
1531 			      " flushing %s tid %llu, requeuing cap.\n",
1532 			      ceph_vinop(inode), ceph_cap_string(arg->dirty),
1533 			      arg->flush_tid);
1534 		spin_lock(&ci->i_ceph_lock);
1535 		__cap_delay_requeue(arg->session->s_mdsc, ci);
1536 		spin_unlock(&ci->i_ceph_lock);
1537 		return;
1538 	}
1539 
1540 	encode_cap_msg(msg, arg);
1541 	ceph_con_send(&arg->session->s_con, msg);
1542 	ceph_buffer_put(arg->old_xattr_buf);
1543 	ceph_buffer_put(arg->xattr_buf);
1544 	if (arg->wake)
1545 		wake_up_all(&ci->i_cap_wq);
1546 }
1547 
__send_flush_snap(struct inode * inode,struct ceph_mds_session * session,struct ceph_cap_snap * capsnap,u32 mseq,u64 oldest_flush_tid)1548 static inline int __send_flush_snap(struct inode *inode,
1549 				    struct ceph_mds_session *session,
1550 				    struct ceph_cap_snap *capsnap,
1551 				    u32 mseq, u64 oldest_flush_tid)
1552 {
1553 	struct cap_msg_args	arg;
1554 	struct ceph_msg		*msg;
1555 
1556 	arg.session = session;
1557 	arg.ino = ceph_vino(inode).ino;
1558 	arg.cid = 0;
1559 	arg.follows = capsnap->follows;
1560 	arg.flush_tid = capsnap->cap_flush.tid;
1561 	arg.oldest_flush_tid = oldest_flush_tid;
1562 
1563 	arg.size = capsnap->size;
1564 	arg.max_size = 0;
1565 	arg.xattr_version = capsnap->xattr_version;
1566 	arg.xattr_buf = capsnap->xattr_blob;
1567 	arg.old_xattr_buf = NULL;
1568 
1569 	arg.atime = capsnap->atime;
1570 	arg.mtime = capsnap->mtime;
1571 	arg.ctime = capsnap->ctime;
1572 	arg.btime = capsnap->btime;
1573 	arg.change_attr = capsnap->change_attr;
1574 
1575 	arg.op = CEPH_CAP_OP_FLUSHSNAP;
1576 	arg.caps = capsnap->issued;
1577 	arg.wanted = 0;
1578 	arg.dirty = capsnap->dirty;
1579 
1580 	arg.seq = 0;
1581 	arg.issue_seq = 0;
1582 	arg.mseq = mseq;
1583 	arg.time_warp_seq = capsnap->time_warp_seq;
1584 
1585 	arg.uid = capsnap->uid;
1586 	arg.gid = capsnap->gid;
1587 	arg.mode = capsnap->mode;
1588 
1589 	arg.inline_data = capsnap->inline_data;
1590 	arg.flags = 0;
1591 	arg.wake = false;
1592 	arg.encrypted = IS_ENCRYPTED(inode);
1593 
1594 	/* No fscrypt_auth changes from a capsnap.*/
1595 	arg.fscrypt_auth_len = 0;
1596 
1597 	msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPS, cap_msg_size(&arg),
1598 			   GFP_NOFS, false);
1599 	if (!msg)
1600 		return -ENOMEM;
1601 
1602 	encode_cap_msg(msg, &arg);
1603 	ceph_con_send(&arg.session->s_con, msg);
1604 	return 0;
1605 }
1606 
1607 /*
1608  * When a snapshot is taken, clients accumulate dirty metadata on
1609  * inodes with capabilities in ceph_cap_snaps to describe the file
1610  * state at the time the snapshot was taken.  This must be flushed
1611  * asynchronously back to the MDS once sync writes complete and dirty
1612  * data is written out.
1613  *
1614  * Called under i_ceph_lock.
1615  */
__ceph_flush_snaps(struct ceph_inode_info * ci,struct ceph_mds_session * session)1616 static void __ceph_flush_snaps(struct ceph_inode_info *ci,
1617 			       struct ceph_mds_session *session)
1618 		__releases(ci->i_ceph_lock)
1619 		__acquires(ci->i_ceph_lock)
1620 {
1621 	struct inode *inode = &ci->netfs.inode;
1622 	struct ceph_mds_client *mdsc = session->s_mdsc;
1623 	struct ceph_client *cl = mdsc->fsc->client;
1624 	struct ceph_cap_snap *capsnap;
1625 	u64 oldest_flush_tid = 0;
1626 	u64 first_tid = 1, last_tid = 0;
1627 
1628 	doutc(cl, "%p %llx.%llx session %p\n", inode, ceph_vinop(inode),
1629 	      session);
1630 
1631 	list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
1632 		/*
1633 		 * we need to wait for sync writes to complete and for dirty
1634 		 * pages to be written out.
1635 		 */
1636 		if (capsnap->dirty_pages || capsnap->writing)
1637 			break;
1638 
1639 		/* should be removed by ceph_try_drop_cap_snap() */
1640 		BUG_ON(!capsnap->need_flush);
1641 
1642 		/* only flush each capsnap once */
1643 		if (capsnap->cap_flush.tid > 0) {
1644 			doutc(cl, "already flushed %p, skipping\n", capsnap);
1645 			continue;
1646 		}
1647 
1648 		spin_lock(&mdsc->cap_dirty_lock);
1649 		capsnap->cap_flush.tid = ++mdsc->last_cap_flush_tid;
1650 		list_add_tail(&capsnap->cap_flush.g_list,
1651 			      &mdsc->cap_flush_list);
1652 		if (oldest_flush_tid == 0)
1653 			oldest_flush_tid = __get_oldest_flush_tid(mdsc);
1654 		if (list_empty(&ci->i_flushing_item)) {
1655 			list_add_tail(&ci->i_flushing_item,
1656 				      &session->s_cap_flushing);
1657 		}
1658 		spin_unlock(&mdsc->cap_dirty_lock);
1659 
1660 		list_add_tail(&capsnap->cap_flush.i_list,
1661 			      &ci->i_cap_flush_list);
1662 
1663 		if (first_tid == 1)
1664 			first_tid = capsnap->cap_flush.tid;
1665 		last_tid = capsnap->cap_flush.tid;
1666 	}
1667 
1668 	ci->i_ceph_flags &= ~CEPH_I_FLUSH_SNAPS;
1669 
1670 	while (first_tid <= last_tid) {
1671 		struct ceph_cap *cap = ci->i_auth_cap;
1672 		struct ceph_cap_flush *cf = NULL, *iter;
1673 		int ret;
1674 
1675 		if (!(cap && cap->session == session)) {
1676 			doutc(cl, "%p %llx.%llx auth cap %p not mds%d, stop\n",
1677 			      inode, ceph_vinop(inode), cap, session->s_mds);
1678 			break;
1679 		}
1680 
1681 		ret = -ENOENT;
1682 		list_for_each_entry(iter, &ci->i_cap_flush_list, i_list) {
1683 			if (iter->tid >= first_tid) {
1684 				cf = iter;
1685 				ret = 0;
1686 				break;
1687 			}
1688 		}
1689 		if (ret < 0)
1690 			break;
1691 
1692 		first_tid = cf->tid + 1;
1693 
1694 		capsnap = container_of(cf, struct ceph_cap_snap, cap_flush);
1695 		refcount_inc(&capsnap->nref);
1696 		spin_unlock(&ci->i_ceph_lock);
1697 
1698 		doutc(cl, "%p %llx.%llx capsnap %p tid %llu %s\n", inode,
1699 		      ceph_vinop(inode), capsnap, cf->tid,
1700 		      ceph_cap_string(capsnap->dirty));
1701 
1702 		ret = __send_flush_snap(inode, session, capsnap, cap->mseq,
1703 					oldest_flush_tid);
1704 		if (ret < 0) {
1705 			pr_err_client(cl, "error sending cap flushsnap, "
1706 				      "ino (%llx.%llx) tid %llu follows %llu\n",
1707 				      ceph_vinop(inode), cf->tid,
1708 				      capsnap->follows);
1709 		}
1710 
1711 		ceph_put_cap_snap(capsnap);
1712 		spin_lock(&ci->i_ceph_lock);
1713 	}
1714 }
1715 
ceph_flush_snaps(struct ceph_inode_info * ci,struct ceph_mds_session ** psession)1716 void ceph_flush_snaps(struct ceph_inode_info *ci,
1717 		      struct ceph_mds_session **psession)
1718 {
1719 	struct inode *inode = &ci->netfs.inode;
1720 	struct ceph_mds_client *mdsc = ceph_inode_to_fs_client(inode)->mdsc;
1721 	struct ceph_client *cl = ceph_inode_to_client(inode);
1722 	struct ceph_mds_session *session = NULL;
1723 	bool need_put = false;
1724 	int mds;
1725 
1726 	doutc(cl, "%p %llx.%llx\n", inode, ceph_vinop(inode));
1727 	if (psession)
1728 		session = *psession;
1729 retry:
1730 	spin_lock(&ci->i_ceph_lock);
1731 	if (!(ci->i_ceph_flags & CEPH_I_FLUSH_SNAPS)) {
1732 		doutc(cl, " no capsnap needs flush, doing nothing\n");
1733 		goto out;
1734 	}
1735 	if (!ci->i_auth_cap) {
1736 		doutc(cl, " no auth cap (migrating?), doing nothing\n");
1737 		goto out;
1738 	}
1739 
1740 	mds = ci->i_auth_cap->session->s_mds;
1741 	if (session && session->s_mds != mds) {
1742 		doutc(cl, " oops, wrong session %p mutex\n", session);
1743 		ceph_put_mds_session(session);
1744 		session = NULL;
1745 	}
1746 	if (!session) {
1747 		spin_unlock(&ci->i_ceph_lock);
1748 		mutex_lock(&mdsc->mutex);
1749 		session = __ceph_lookup_mds_session(mdsc, mds);
1750 		mutex_unlock(&mdsc->mutex);
1751 		goto retry;
1752 	}
1753 
1754 	// make sure flushsnap messages are sent in proper order.
1755 	if (ci->i_ceph_flags & CEPH_I_KICK_FLUSH)
1756 		__kick_flushing_caps(mdsc, session, ci, 0);
1757 
1758 	__ceph_flush_snaps(ci, session);
1759 out:
1760 	spin_unlock(&ci->i_ceph_lock);
1761 
1762 	if (psession)
1763 		*psession = session;
1764 	else
1765 		ceph_put_mds_session(session);
1766 	/* we flushed them all; remove this inode from the queue */
1767 	spin_lock(&mdsc->snap_flush_lock);
1768 	if (!list_empty(&ci->i_snap_flush_item))
1769 		need_put = true;
1770 	list_del_init(&ci->i_snap_flush_item);
1771 	spin_unlock(&mdsc->snap_flush_lock);
1772 
1773 	if (need_put)
1774 		iput(inode);
1775 }
1776 
1777 /*
1778  * Mark caps dirty.  If inode is newly dirty, return the dirty flags.
1779  * Caller is then responsible for calling __mark_inode_dirty with the
1780  * returned flags value.
1781  */
__ceph_mark_dirty_caps(struct ceph_inode_info * ci,int mask,struct ceph_cap_flush ** pcf)1782 int __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask,
1783 			   struct ceph_cap_flush **pcf)
1784 {
1785 	struct ceph_mds_client *mdsc =
1786 		ceph_sb_to_fs_client(ci->netfs.inode.i_sb)->mdsc;
1787 	struct inode *inode = &ci->netfs.inode;
1788 	struct ceph_client *cl = ceph_inode_to_client(inode);
1789 	int was = ci->i_dirty_caps;
1790 	int dirty = 0;
1791 
1792 	lockdep_assert_held(&ci->i_ceph_lock);
1793 
1794 	if (!ci->i_auth_cap) {
1795 		pr_warn_client(cl, "%p %llx.%llx mask %s, "
1796 			       "but no auth cap (session was closed?)\n",
1797 				inode, ceph_vinop(inode),
1798 				ceph_cap_string(mask));
1799 		return 0;
1800 	}
1801 
1802 	doutc(cl, "%p %llx.%llx %s dirty %s -> %s\n", inode,
1803 	      ceph_vinop(inode), ceph_cap_string(mask),
1804 	      ceph_cap_string(was), ceph_cap_string(was | mask));
1805 	ci->i_dirty_caps |= mask;
1806 	if (was == 0) {
1807 		struct ceph_mds_session *session = ci->i_auth_cap->session;
1808 
1809 		WARN_ON_ONCE(ci->i_prealloc_cap_flush);
1810 		swap(ci->i_prealloc_cap_flush, *pcf);
1811 
1812 		if (!ci->i_head_snapc) {
1813 			WARN_ON_ONCE(!rwsem_is_locked(&mdsc->snap_rwsem));
1814 			ci->i_head_snapc = ceph_get_snap_context(
1815 				ci->i_snap_realm->cached_context);
1816 		}
1817 		doutc(cl, "%p %llx.%llx now dirty snapc %p auth cap %p\n",
1818 		      inode, ceph_vinop(inode), ci->i_head_snapc,
1819 		      ci->i_auth_cap);
1820 		BUG_ON(!list_empty(&ci->i_dirty_item));
1821 		spin_lock(&mdsc->cap_dirty_lock);
1822 		list_add(&ci->i_dirty_item, &session->s_cap_dirty);
1823 		spin_unlock(&mdsc->cap_dirty_lock);
1824 		if (ci->i_flushing_caps == 0) {
1825 			ihold(inode);
1826 			dirty |= I_DIRTY_SYNC;
1827 		}
1828 	} else {
1829 		WARN_ON_ONCE(!ci->i_prealloc_cap_flush);
1830 	}
1831 	BUG_ON(list_empty(&ci->i_dirty_item));
1832 	if (((was | ci->i_flushing_caps) & CEPH_CAP_FILE_BUFFER) &&
1833 	    (mask & CEPH_CAP_FILE_BUFFER))
1834 		dirty |= I_DIRTY_DATASYNC;
1835 	__cap_delay_requeue(mdsc, ci);
1836 	return dirty;
1837 }
1838 
ceph_alloc_cap_flush(void)1839 struct ceph_cap_flush *ceph_alloc_cap_flush(void)
1840 {
1841 	struct ceph_cap_flush *cf;
1842 
1843 	cf = kmem_cache_alloc(ceph_cap_flush_cachep, GFP_KERNEL);
1844 	if (!cf)
1845 		return NULL;
1846 
1847 	cf->is_capsnap = false;
1848 	return cf;
1849 }
1850 
ceph_free_cap_flush(struct ceph_cap_flush * cf)1851 void ceph_free_cap_flush(struct ceph_cap_flush *cf)
1852 {
1853 	if (cf)
1854 		kmem_cache_free(ceph_cap_flush_cachep, cf);
1855 }
1856 
__get_oldest_flush_tid(struct ceph_mds_client * mdsc)1857 static u64 __get_oldest_flush_tid(struct ceph_mds_client *mdsc)
1858 {
1859 	if (!list_empty(&mdsc->cap_flush_list)) {
1860 		struct ceph_cap_flush *cf =
1861 			list_first_entry(&mdsc->cap_flush_list,
1862 					 struct ceph_cap_flush, g_list);
1863 		return cf->tid;
1864 	}
1865 	return 0;
1866 }
1867 
1868 /*
1869  * Remove cap_flush from the mdsc's or inode's flushing cap list.
1870  * Return true if caller needs to wake up flush waiters.
1871  */
__detach_cap_flush_from_mdsc(struct ceph_mds_client * mdsc,struct ceph_cap_flush * cf)1872 static bool __detach_cap_flush_from_mdsc(struct ceph_mds_client *mdsc,
1873 					 struct ceph_cap_flush *cf)
1874 {
1875 	struct ceph_cap_flush *prev;
1876 	bool wake = cf->wake;
1877 
1878 	if (wake && cf->g_list.prev != &mdsc->cap_flush_list) {
1879 		prev = list_prev_entry(cf, g_list);
1880 		prev->wake = true;
1881 		wake = false;
1882 	}
1883 	list_del_init(&cf->g_list);
1884 	return wake;
1885 }
1886 
__detach_cap_flush_from_ci(struct ceph_inode_info * ci,struct ceph_cap_flush * cf)1887 static bool __detach_cap_flush_from_ci(struct ceph_inode_info *ci,
1888 				       struct ceph_cap_flush *cf)
1889 {
1890 	struct ceph_cap_flush *prev;
1891 	bool wake = cf->wake;
1892 
1893 	if (wake && cf->i_list.prev != &ci->i_cap_flush_list) {
1894 		prev = list_prev_entry(cf, i_list);
1895 		prev->wake = true;
1896 		wake = false;
1897 	}
1898 	list_del_init(&cf->i_list);
1899 	return wake;
1900 }
1901 
1902 /*
1903  * Add dirty inode to the flushing list.  Assigned a seq number so we
1904  * can wait for caps to flush without starving.
1905  *
1906  * Called under i_ceph_lock. Returns the flush tid.
1907  */
__mark_caps_flushing(struct inode * inode,struct ceph_mds_session * session,bool wake,u64 * oldest_flush_tid)1908 static u64 __mark_caps_flushing(struct inode *inode,
1909 				struct ceph_mds_session *session, bool wake,
1910 				u64 *oldest_flush_tid)
1911 {
1912 	struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(inode->i_sb)->mdsc;
1913 	struct ceph_client *cl = ceph_inode_to_client(inode);
1914 	struct ceph_inode_info *ci = ceph_inode(inode);
1915 	struct ceph_cap_flush *cf = NULL;
1916 	int flushing;
1917 
1918 	lockdep_assert_held(&ci->i_ceph_lock);
1919 	BUG_ON(ci->i_dirty_caps == 0);
1920 	BUG_ON(list_empty(&ci->i_dirty_item));
1921 	BUG_ON(!ci->i_prealloc_cap_flush);
1922 
1923 	flushing = ci->i_dirty_caps;
1924 	doutc(cl, "flushing %s, flushing_caps %s -> %s\n",
1925 	      ceph_cap_string(flushing),
1926 	      ceph_cap_string(ci->i_flushing_caps),
1927 	      ceph_cap_string(ci->i_flushing_caps | flushing));
1928 	ci->i_flushing_caps |= flushing;
1929 	ci->i_dirty_caps = 0;
1930 	doutc(cl, "%p %llx.%llx now !dirty\n", inode, ceph_vinop(inode));
1931 
1932 	swap(cf, ci->i_prealloc_cap_flush);
1933 	cf->caps = flushing;
1934 	cf->wake = wake;
1935 
1936 	spin_lock(&mdsc->cap_dirty_lock);
1937 	list_del_init(&ci->i_dirty_item);
1938 
1939 	cf->tid = ++mdsc->last_cap_flush_tid;
1940 	list_add_tail(&cf->g_list, &mdsc->cap_flush_list);
1941 	*oldest_flush_tid = __get_oldest_flush_tid(mdsc);
1942 
1943 	if (list_empty(&ci->i_flushing_item)) {
1944 		list_add_tail(&ci->i_flushing_item, &session->s_cap_flushing);
1945 		mdsc->num_cap_flushing++;
1946 	}
1947 	spin_unlock(&mdsc->cap_dirty_lock);
1948 
1949 	list_add_tail(&cf->i_list, &ci->i_cap_flush_list);
1950 
1951 	return cf->tid;
1952 }
1953 
1954 /*
1955  * try to invalidate mapping pages without blocking.
1956  */
try_nonblocking_invalidate(struct inode * inode)1957 static int try_nonblocking_invalidate(struct inode *inode)
1958 	__releases(ci->i_ceph_lock)
1959 	__acquires(ci->i_ceph_lock)
1960 {
1961 	struct ceph_client *cl = ceph_inode_to_client(inode);
1962 	struct ceph_inode_info *ci = ceph_inode(inode);
1963 	u32 invalidating_gen = ci->i_rdcache_gen;
1964 
1965 	spin_unlock(&ci->i_ceph_lock);
1966 	ceph_fscache_invalidate(inode, false);
1967 	invalidate_mapping_pages(&inode->i_data, 0, -1);
1968 	spin_lock(&ci->i_ceph_lock);
1969 
1970 	if (inode->i_data.nrpages == 0 &&
1971 	    invalidating_gen == ci->i_rdcache_gen) {
1972 		/* success. */
1973 		doutc(cl, "%p %llx.%llx success\n", inode,
1974 		      ceph_vinop(inode));
1975 		/* save any racing async invalidate some trouble */
1976 		ci->i_rdcache_revoking = ci->i_rdcache_gen - 1;
1977 		return 0;
1978 	}
1979 	doutc(cl, "%p %llx.%llx failed\n", inode, ceph_vinop(inode));
1980 	return -1;
1981 }
1982 
__ceph_should_report_size(struct ceph_inode_info * ci)1983 bool __ceph_should_report_size(struct ceph_inode_info *ci)
1984 {
1985 	loff_t size = i_size_read(&ci->netfs.inode);
1986 	/* mds will adjust max size according to the reported size */
1987 	if (ci->i_flushing_caps & CEPH_CAP_FILE_WR)
1988 		return false;
1989 	if (size >= ci->i_max_size)
1990 		return true;
1991 	/* half of previous max_size increment has been used */
1992 	if (ci->i_max_size > ci->i_reported_size &&
1993 	    (size << 1) >= ci->i_max_size + ci->i_reported_size)
1994 		return true;
1995 	return false;
1996 }
1997 
1998 /*
1999  * Swiss army knife function to examine currently used and wanted
2000  * versus held caps.  Release, flush, ack revoked caps to mds as
2001  * appropriate.
2002  *
2003  *  CHECK_CAPS_AUTHONLY - we should only check the auth cap
2004  *  CHECK_CAPS_FLUSH - we should flush any dirty caps immediately, without
2005  *    further delay.
2006  *  CHECK_CAPS_FLUSH_FORCE - we should flush any caps immediately, without
2007  *    further delay.
2008  */
ceph_check_caps(struct ceph_inode_info * ci,int flags)2009 void ceph_check_caps(struct ceph_inode_info *ci, int flags)
2010 {
2011 	struct inode *inode = &ci->netfs.inode;
2012 	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
2013 	struct ceph_client *cl = ceph_inode_to_client(inode);
2014 	struct ceph_cap *cap;
2015 	u64 flush_tid, oldest_flush_tid;
2016 	int file_wanted, used, cap_used;
2017 	int issued, implemented, want, retain, revoking, flushing = 0;
2018 	int mds = -1;   /* keep track of how far we've gone through i_caps list
2019 			   to avoid an infinite loop on retry */
2020 	struct rb_node *p;
2021 	bool queue_invalidate = false;
2022 	bool tried_invalidate = false;
2023 	bool queue_writeback = false;
2024 	struct ceph_mds_session *session = NULL;
2025 
2026 	spin_lock(&ci->i_ceph_lock);
2027 	if (ci->i_ceph_flags & CEPH_I_ASYNC_CREATE) {
2028 		ci->i_ceph_flags |= CEPH_I_ASYNC_CHECK_CAPS;
2029 
2030 		/* Don't send messages until we get async create reply */
2031 		spin_unlock(&ci->i_ceph_lock);
2032 		return;
2033 	}
2034 
2035 	if (ci->i_ceph_flags & CEPH_I_FLUSH)
2036 		flags |= CHECK_CAPS_FLUSH;
2037 retry:
2038 	/* Caps wanted by virtue of active open files. */
2039 	file_wanted = __ceph_caps_file_wanted(ci);
2040 
2041 	/* Caps which have active references against them */
2042 	used = __ceph_caps_used(ci);
2043 
2044 	/*
2045 	 * "issued" represents the current caps that the MDS wants us to have.
2046 	 * "implemented" is the set that we have been granted, and includes the
2047 	 * ones that have not yet been returned to the MDS (the "revoking" set,
2048 	 * usually because they have outstanding references).
2049 	 */
2050 	issued = __ceph_caps_issued(ci, &implemented);
2051 	revoking = implemented & ~issued;
2052 
2053 	want = file_wanted;
2054 
2055 	/* The ones we currently want to retain (may be adjusted below) */
2056 	retain = file_wanted | used | CEPH_CAP_PIN;
2057 	if (!mdsc->stopping && inode->i_nlink > 0) {
2058 		if (file_wanted) {
2059 			retain |= CEPH_CAP_ANY;       /* be greedy */
2060 		} else if (S_ISDIR(inode->i_mode) &&
2061 			   (issued & CEPH_CAP_FILE_SHARED) &&
2062 			   __ceph_dir_is_complete(ci)) {
2063 			/*
2064 			 * If a directory is complete, we want to keep
2065 			 * the exclusive cap. So that MDS does not end up
2066 			 * revoking the shared cap on every create/unlink
2067 			 * operation.
2068 			 */
2069 			if (IS_RDONLY(inode)) {
2070 				want = CEPH_CAP_ANY_SHARED;
2071 			} else {
2072 				want |= CEPH_CAP_ANY_SHARED | CEPH_CAP_FILE_EXCL;
2073 			}
2074 			retain |= want;
2075 		} else {
2076 
2077 			retain |= CEPH_CAP_ANY_SHARED;
2078 			/*
2079 			 * keep RD only if we didn't have the file open RW,
2080 			 * because then the mds would revoke it anyway to
2081 			 * journal max_size=0.
2082 			 */
2083 			if (ci->i_max_size == 0)
2084 				retain |= CEPH_CAP_ANY_RD;
2085 		}
2086 	}
2087 
2088 	doutc(cl, "%p %llx.%llx file_want %s used %s dirty %s "
2089 	      "flushing %s issued %s revoking %s retain %s %s%s%s%s\n",
2090 	     inode, ceph_vinop(inode), ceph_cap_string(file_wanted),
2091 	     ceph_cap_string(used), ceph_cap_string(ci->i_dirty_caps),
2092 	     ceph_cap_string(ci->i_flushing_caps),
2093 	     ceph_cap_string(issued), ceph_cap_string(revoking),
2094 	     ceph_cap_string(retain),
2095 	     (flags & CHECK_CAPS_AUTHONLY) ? " AUTHONLY" : "",
2096 	     (flags & CHECK_CAPS_FLUSH) ? " FLUSH" : "",
2097 	     (flags & CHECK_CAPS_NOINVAL) ? " NOINVAL" : "",
2098 	     (flags & CHECK_CAPS_FLUSH_FORCE) ? " FLUSH_FORCE" : "");
2099 
2100 	/*
2101 	 * If we no longer need to hold onto old our caps, and we may
2102 	 * have cached pages, but don't want them, then try to invalidate.
2103 	 * If we fail, it's because pages are locked.... try again later.
2104 	 */
2105 	if ((!(flags & CHECK_CAPS_NOINVAL) || mdsc->stopping) &&
2106 	    S_ISREG(inode->i_mode) &&
2107 	    !(ci->i_wb_ref || ci->i_wrbuffer_ref) &&   /* no dirty pages... */
2108 	    inode->i_data.nrpages &&		/* have cached pages */
2109 	    (revoking & (CEPH_CAP_FILE_CACHE|
2110 			 CEPH_CAP_FILE_LAZYIO)) && /*  or revoking cache */
2111 	    !tried_invalidate) {
2112 		doutc(cl, "trying to invalidate on %p %llx.%llx\n",
2113 		      inode, ceph_vinop(inode));
2114 		if (try_nonblocking_invalidate(inode) < 0) {
2115 			doutc(cl, "queuing invalidate\n");
2116 			queue_invalidate = true;
2117 			ci->i_rdcache_revoking = ci->i_rdcache_gen;
2118 		}
2119 		tried_invalidate = true;
2120 		goto retry;
2121 	}
2122 
2123 	for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
2124 		int mflags = 0;
2125 		struct cap_msg_args arg;
2126 
2127 		cap = rb_entry(p, struct ceph_cap, ci_node);
2128 
2129 		/* avoid looping forever */
2130 		if (mds >= cap->mds ||
2131 		    ((flags & CHECK_CAPS_AUTHONLY) && cap != ci->i_auth_cap))
2132 			continue;
2133 
2134 		/*
2135 		 * If we have an auth cap, we don't need to consider any
2136 		 * overlapping caps as used.
2137 		 */
2138 		cap_used = used;
2139 		if (ci->i_auth_cap && cap != ci->i_auth_cap)
2140 			cap_used &= ~ci->i_auth_cap->issued;
2141 
2142 		revoking = cap->implemented & ~cap->issued;
2143 		doutc(cl, " mds%d cap %p used %s issued %s implemented %s revoking %s\n",
2144 		      cap->mds, cap, ceph_cap_string(cap_used),
2145 		      ceph_cap_string(cap->issued),
2146 		      ceph_cap_string(cap->implemented),
2147 		      ceph_cap_string(revoking));
2148 
2149 		/* completed revocation? going down and there are no caps? */
2150 		if (revoking) {
2151 			if ((revoking & cap_used) == 0) {
2152 				doutc(cl, "completed revocation of %s\n",
2153 				      ceph_cap_string(cap->implemented & ~cap->issued));
2154 				goto ack;
2155 			}
2156 
2157 			/*
2158 			 * If the "i_wrbuffer_ref" was increased by mmap or generic
2159 			 * cache write just before the ceph_check_caps() is called,
2160 			 * the Fb capability revoking will fail this time. Then we
2161 			 * must wait for the BDI's delayed work to flush the dirty
2162 			 * pages and to release the "i_wrbuffer_ref", which will cost
2163 			 * at most 5 seconds. That means the MDS needs to wait at
2164 			 * most 5 seconds to finished the Fb capability's revocation.
2165 			 *
2166 			 * Let's queue a writeback for it.
2167 			 */
2168 			if (S_ISREG(inode->i_mode) && ci->i_wrbuffer_ref &&
2169 			    (revoking & CEPH_CAP_FILE_BUFFER))
2170 				queue_writeback = true;
2171 		}
2172 
2173 		if (flags & CHECK_CAPS_FLUSH_FORCE) {
2174 			doutc(cl, "force to flush caps\n");
2175 			goto ack;
2176 		}
2177 
2178 		if (cap == ci->i_auth_cap &&
2179 		    (cap->issued & CEPH_CAP_FILE_WR)) {
2180 			/* request larger max_size from MDS? */
2181 			if (ci->i_wanted_max_size > ci->i_max_size &&
2182 			    ci->i_wanted_max_size > ci->i_requested_max_size) {
2183 				doutc(cl, "requesting new max_size\n");
2184 				goto ack;
2185 			}
2186 
2187 			/* approaching file_max? */
2188 			if (__ceph_should_report_size(ci)) {
2189 				doutc(cl, "i_size approaching max_size\n");
2190 				goto ack;
2191 			}
2192 		}
2193 		/* flush anything dirty? */
2194 		if (cap == ci->i_auth_cap) {
2195 			if ((flags & CHECK_CAPS_FLUSH) && ci->i_dirty_caps) {
2196 				doutc(cl, "flushing dirty caps\n");
2197 				goto ack;
2198 			}
2199 			if (ci->i_ceph_flags & CEPH_I_FLUSH_SNAPS) {
2200 				doutc(cl, "flushing snap caps\n");
2201 				goto ack;
2202 			}
2203 		}
2204 
2205 		/* want more caps from mds? */
2206 		if (want & ~cap->mds_wanted) {
2207 			if (want & ~(cap->mds_wanted | cap->issued))
2208 				goto ack;
2209 			if (!__cap_is_valid(cap))
2210 				goto ack;
2211 		}
2212 
2213 		/* things we might delay */
2214 		if ((cap->issued & ~retain) == 0)
2215 			continue;     /* nope, all good */
2216 
2217 ack:
2218 		ceph_put_mds_session(session);
2219 		session = ceph_get_mds_session(cap->session);
2220 
2221 		/* kick flushing and flush snaps before sending normal
2222 		 * cap message */
2223 		if (cap == ci->i_auth_cap &&
2224 		    (ci->i_ceph_flags &
2225 		     (CEPH_I_KICK_FLUSH | CEPH_I_FLUSH_SNAPS))) {
2226 			if (ci->i_ceph_flags & CEPH_I_KICK_FLUSH)
2227 				__kick_flushing_caps(mdsc, session, ci, 0);
2228 			if (ci->i_ceph_flags & CEPH_I_FLUSH_SNAPS)
2229 				__ceph_flush_snaps(ci, session);
2230 
2231 			goto retry;
2232 		}
2233 
2234 		if (cap == ci->i_auth_cap && ci->i_dirty_caps) {
2235 			flushing = ci->i_dirty_caps;
2236 			flush_tid = __mark_caps_flushing(inode, session, false,
2237 							 &oldest_flush_tid);
2238 			if (flags & CHECK_CAPS_FLUSH &&
2239 			    list_empty(&session->s_cap_dirty))
2240 				mflags |= CEPH_CLIENT_CAPS_SYNC;
2241 		} else {
2242 			flushing = 0;
2243 			flush_tid = 0;
2244 			spin_lock(&mdsc->cap_dirty_lock);
2245 			oldest_flush_tid = __get_oldest_flush_tid(mdsc);
2246 			spin_unlock(&mdsc->cap_dirty_lock);
2247 		}
2248 
2249 		mds = cap->mds;  /* remember mds, so we don't repeat */
2250 
2251 		__prep_cap(&arg, cap, CEPH_CAP_OP_UPDATE, mflags, cap_used,
2252 			   want, retain, flushing, flush_tid, oldest_flush_tid);
2253 
2254 		spin_unlock(&ci->i_ceph_lock);
2255 		__send_cap(&arg, ci);
2256 		spin_lock(&ci->i_ceph_lock);
2257 
2258 		goto retry; /* retake i_ceph_lock and restart our cap scan. */
2259 	}
2260 
2261 	/* periodically re-calculate caps wanted by open files */
2262 	if (__ceph_is_any_real_caps(ci) &&
2263 	    list_empty(&ci->i_cap_delay_list) &&
2264 	    (file_wanted & ~CEPH_CAP_PIN) &&
2265 	    !(used & (CEPH_CAP_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
2266 		__cap_delay_requeue(mdsc, ci);
2267 	}
2268 
2269 	spin_unlock(&ci->i_ceph_lock);
2270 
2271 	ceph_put_mds_session(session);
2272 	if (queue_writeback)
2273 		ceph_queue_writeback(inode);
2274 	if (queue_invalidate)
2275 		ceph_queue_invalidate(inode);
2276 }
2277 
2278 /*
2279  * Try to flush dirty caps back to the auth mds.
2280  */
try_flush_caps(struct inode * inode,u64 * ptid)2281 static int try_flush_caps(struct inode *inode, u64 *ptid)
2282 {
2283 	struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(inode->i_sb)->mdsc;
2284 	struct ceph_inode_info *ci = ceph_inode(inode);
2285 	int flushing = 0;
2286 	u64 flush_tid = 0, oldest_flush_tid = 0;
2287 
2288 	spin_lock(&ci->i_ceph_lock);
2289 retry_locked:
2290 	if (ci->i_dirty_caps && ci->i_auth_cap) {
2291 		struct ceph_cap *cap = ci->i_auth_cap;
2292 		struct cap_msg_args arg;
2293 		struct ceph_mds_session *session = cap->session;
2294 
2295 		if (session->s_state < CEPH_MDS_SESSION_OPEN) {
2296 			spin_unlock(&ci->i_ceph_lock);
2297 			goto out;
2298 		}
2299 
2300 		if (ci->i_ceph_flags &
2301 		    (CEPH_I_KICK_FLUSH | CEPH_I_FLUSH_SNAPS)) {
2302 			if (ci->i_ceph_flags & CEPH_I_KICK_FLUSH)
2303 				__kick_flushing_caps(mdsc, session, ci, 0);
2304 			if (ci->i_ceph_flags & CEPH_I_FLUSH_SNAPS)
2305 				__ceph_flush_snaps(ci, session);
2306 			goto retry_locked;
2307 		}
2308 
2309 		flushing = ci->i_dirty_caps;
2310 		flush_tid = __mark_caps_flushing(inode, session, true,
2311 						 &oldest_flush_tid);
2312 
2313 		__prep_cap(&arg, cap, CEPH_CAP_OP_FLUSH, CEPH_CLIENT_CAPS_SYNC,
2314 			   __ceph_caps_used(ci), __ceph_caps_wanted(ci),
2315 			   (cap->issued | cap->implemented),
2316 			   flushing, flush_tid, oldest_flush_tid);
2317 		spin_unlock(&ci->i_ceph_lock);
2318 
2319 		__send_cap(&arg, ci);
2320 	} else {
2321 		if (!list_empty(&ci->i_cap_flush_list)) {
2322 			struct ceph_cap_flush *cf =
2323 				list_last_entry(&ci->i_cap_flush_list,
2324 						struct ceph_cap_flush, i_list);
2325 			cf->wake = true;
2326 			flush_tid = cf->tid;
2327 		}
2328 		flushing = ci->i_flushing_caps;
2329 		spin_unlock(&ci->i_ceph_lock);
2330 	}
2331 out:
2332 	*ptid = flush_tid;
2333 	return flushing;
2334 }
2335 
2336 /*
2337  * Return true if we've flushed caps through the given flush_tid.
2338  */
caps_are_flushed(struct inode * inode,u64 flush_tid)2339 static int caps_are_flushed(struct inode *inode, u64 flush_tid)
2340 {
2341 	struct ceph_inode_info *ci = ceph_inode(inode);
2342 	int ret = 1;
2343 
2344 	spin_lock(&ci->i_ceph_lock);
2345 	if (!list_empty(&ci->i_cap_flush_list)) {
2346 		struct ceph_cap_flush * cf =
2347 			list_first_entry(&ci->i_cap_flush_list,
2348 					 struct ceph_cap_flush, i_list);
2349 		if (cf->tid <= flush_tid)
2350 			ret = 0;
2351 	}
2352 	spin_unlock(&ci->i_ceph_lock);
2353 	return ret;
2354 }
2355 
2356 /*
2357  * flush the mdlog and wait for any unsafe requests to complete.
2358  */
flush_mdlog_and_wait_inode_unsafe_requests(struct inode * inode)2359 static int flush_mdlog_and_wait_inode_unsafe_requests(struct inode *inode)
2360 {
2361 	struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(inode->i_sb)->mdsc;
2362 	struct ceph_client *cl = ceph_inode_to_client(inode);
2363 	struct ceph_inode_info *ci = ceph_inode(inode);
2364 	struct ceph_mds_request *req1 = NULL, *req2 = NULL;
2365 	int ret, err = 0;
2366 
2367 	spin_lock(&ci->i_unsafe_lock);
2368 	if (S_ISDIR(inode->i_mode) && !list_empty(&ci->i_unsafe_dirops)) {
2369 		req1 = list_last_entry(&ci->i_unsafe_dirops,
2370 					struct ceph_mds_request,
2371 					r_unsafe_dir_item);
2372 		ceph_mdsc_get_request(req1);
2373 	}
2374 	if (!list_empty(&ci->i_unsafe_iops)) {
2375 		req2 = list_last_entry(&ci->i_unsafe_iops,
2376 					struct ceph_mds_request,
2377 					r_unsafe_target_item);
2378 		ceph_mdsc_get_request(req2);
2379 	}
2380 	spin_unlock(&ci->i_unsafe_lock);
2381 
2382 	/*
2383 	 * Trigger to flush the journal logs in all the relevant MDSes
2384 	 * manually, or in the worst case we must wait at most 5 seconds
2385 	 * to wait the journal logs to be flushed by the MDSes periodically.
2386 	 */
2387 	if (req1 || req2) {
2388 		struct ceph_mds_request *req;
2389 		struct ceph_mds_session **sessions;
2390 		struct ceph_mds_session *s;
2391 		unsigned int max_sessions;
2392 		int i;
2393 
2394 		mutex_lock(&mdsc->mutex);
2395 		max_sessions = mdsc->max_sessions;
2396 
2397 		sessions = kcalloc(max_sessions, sizeof(s), GFP_KERNEL);
2398 		if (!sessions) {
2399 			mutex_unlock(&mdsc->mutex);
2400 			err = -ENOMEM;
2401 			goto out;
2402 		}
2403 
2404 		spin_lock(&ci->i_unsafe_lock);
2405 		if (req1) {
2406 			list_for_each_entry(req, &ci->i_unsafe_dirops,
2407 					    r_unsafe_dir_item) {
2408 				s = req->r_session;
2409 				if (!s)
2410 					continue;
2411 				if (!sessions[s->s_mds]) {
2412 					s = ceph_get_mds_session(s);
2413 					sessions[s->s_mds] = s;
2414 				}
2415 			}
2416 		}
2417 		if (req2) {
2418 			list_for_each_entry(req, &ci->i_unsafe_iops,
2419 					    r_unsafe_target_item) {
2420 				s = req->r_session;
2421 				if (!s)
2422 					continue;
2423 				if (!sessions[s->s_mds]) {
2424 					s = ceph_get_mds_session(s);
2425 					sessions[s->s_mds] = s;
2426 				}
2427 			}
2428 		}
2429 		spin_unlock(&ci->i_unsafe_lock);
2430 
2431 		/* the auth MDS */
2432 		spin_lock(&ci->i_ceph_lock);
2433 		if (ci->i_auth_cap) {
2434 			s = ci->i_auth_cap->session;
2435 			if (!sessions[s->s_mds])
2436 				sessions[s->s_mds] = ceph_get_mds_session(s);
2437 		}
2438 		spin_unlock(&ci->i_ceph_lock);
2439 		mutex_unlock(&mdsc->mutex);
2440 
2441 		/* send flush mdlog request to MDSes */
2442 		for (i = 0; i < max_sessions; i++) {
2443 			s = sessions[i];
2444 			if (s) {
2445 				send_flush_mdlog(s);
2446 				ceph_put_mds_session(s);
2447 			}
2448 		}
2449 		kfree(sessions);
2450 	}
2451 
2452 	doutc(cl, "%p %llx.%llx wait on tid %llu %llu\n", inode,
2453 	      ceph_vinop(inode), req1 ? req1->r_tid : 0ULL,
2454 	      req2 ? req2->r_tid : 0ULL);
2455 	if (req1) {
2456 		ret = !wait_for_completion_timeout(&req1->r_safe_completion,
2457 					ceph_timeout_jiffies(req1->r_timeout));
2458 		if (ret)
2459 			err = -EIO;
2460 	}
2461 	if (req2) {
2462 		ret = !wait_for_completion_timeout(&req2->r_safe_completion,
2463 					ceph_timeout_jiffies(req2->r_timeout));
2464 		if (ret)
2465 			err = -EIO;
2466 	}
2467 
2468 out:
2469 	if (req1)
2470 		ceph_mdsc_put_request(req1);
2471 	if (req2)
2472 		ceph_mdsc_put_request(req2);
2473 	return err;
2474 }
2475 
ceph_fsync(struct file * file,loff_t start,loff_t end,int datasync)2476 int ceph_fsync(struct file *file, loff_t start, loff_t end, int datasync)
2477 {
2478 	struct inode *inode = file->f_mapping->host;
2479 	struct ceph_inode_info *ci = ceph_inode(inode);
2480 	struct ceph_client *cl = ceph_inode_to_client(inode);
2481 	u64 flush_tid;
2482 	int ret, err;
2483 	int dirty;
2484 
2485 	doutc(cl, "%p %llx.%llx%s\n", inode, ceph_vinop(inode),
2486 	      datasync ? " datasync" : "");
2487 
2488 	ret = file_write_and_wait_range(file, start, end);
2489 	if (datasync)
2490 		goto out;
2491 
2492 	ret = ceph_wait_on_async_create(inode);
2493 	if (ret)
2494 		goto out;
2495 
2496 	dirty = try_flush_caps(inode, &flush_tid);
2497 	doutc(cl, "dirty caps are %s\n", ceph_cap_string(dirty));
2498 
2499 	err = flush_mdlog_and_wait_inode_unsafe_requests(inode);
2500 
2501 	/*
2502 	 * only wait on non-file metadata writeback (the mds
2503 	 * can recover size and mtime, so we don't need to
2504 	 * wait for that)
2505 	 */
2506 	if (!err && (dirty & ~CEPH_CAP_ANY_FILE_WR)) {
2507 		err = wait_event_interruptible(ci->i_cap_wq,
2508 					caps_are_flushed(inode, flush_tid));
2509 	}
2510 
2511 	if (err < 0)
2512 		ret = err;
2513 
2514 	err = file_check_and_advance_wb_err(file);
2515 	if (err < 0)
2516 		ret = err;
2517 out:
2518 	doutc(cl, "%p %llx.%llx%s result=%d\n", inode, ceph_vinop(inode),
2519 	      datasync ? " datasync" : "", ret);
2520 	return ret;
2521 }
2522 
2523 /*
2524  * Flush any dirty caps back to the mds.  If we aren't asked to wait,
2525  * queue inode for flush but don't do so immediately, because we can
2526  * get by with fewer MDS messages if we wait for data writeback to
2527  * complete first.
2528  */
ceph_write_inode(struct inode * inode,struct writeback_control * wbc)2529 int ceph_write_inode(struct inode *inode, struct writeback_control *wbc)
2530 {
2531 	struct ceph_inode_info *ci = ceph_inode(inode);
2532 	struct ceph_client *cl = ceph_inode_to_client(inode);
2533 	u64 flush_tid;
2534 	int err = 0;
2535 	int dirty;
2536 	int wait = (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync);
2537 
2538 	doutc(cl, "%p %llx.%llx wait=%d\n", inode, ceph_vinop(inode), wait);
2539 	ceph_fscache_unpin_writeback(inode, wbc);
2540 	if (wait) {
2541 		err = ceph_wait_on_async_create(inode);
2542 		if (err)
2543 			return err;
2544 		dirty = try_flush_caps(inode, &flush_tid);
2545 		if (dirty)
2546 			err = wait_event_interruptible(ci->i_cap_wq,
2547 				       caps_are_flushed(inode, flush_tid));
2548 	} else {
2549 		struct ceph_mds_client *mdsc =
2550 			ceph_sb_to_fs_client(inode->i_sb)->mdsc;
2551 
2552 		spin_lock(&ci->i_ceph_lock);
2553 		if (__ceph_caps_dirty(ci))
2554 			__cap_delay_requeue_front(mdsc, ci);
2555 		spin_unlock(&ci->i_ceph_lock);
2556 	}
2557 	return err;
2558 }
2559 
__kick_flushing_caps(struct ceph_mds_client * mdsc,struct ceph_mds_session * session,struct ceph_inode_info * ci,u64 oldest_flush_tid)2560 static void __kick_flushing_caps(struct ceph_mds_client *mdsc,
2561 				 struct ceph_mds_session *session,
2562 				 struct ceph_inode_info *ci,
2563 				 u64 oldest_flush_tid)
2564 	__releases(ci->i_ceph_lock)
2565 	__acquires(ci->i_ceph_lock)
2566 {
2567 	struct inode *inode = &ci->netfs.inode;
2568 	struct ceph_client *cl = mdsc->fsc->client;
2569 	struct ceph_cap *cap;
2570 	struct ceph_cap_flush *cf;
2571 	int ret;
2572 	u64 first_tid = 0;
2573 	u64 last_snap_flush = 0;
2574 
2575 	/* Don't do anything until create reply comes in */
2576 	if (ci->i_ceph_flags & CEPH_I_ASYNC_CREATE)
2577 		return;
2578 
2579 	ci->i_ceph_flags &= ~CEPH_I_KICK_FLUSH;
2580 
2581 	list_for_each_entry_reverse(cf, &ci->i_cap_flush_list, i_list) {
2582 		if (cf->is_capsnap) {
2583 			last_snap_flush = cf->tid;
2584 			break;
2585 		}
2586 	}
2587 
2588 	list_for_each_entry(cf, &ci->i_cap_flush_list, i_list) {
2589 		if (cf->tid < first_tid)
2590 			continue;
2591 
2592 		cap = ci->i_auth_cap;
2593 		if (!(cap && cap->session == session)) {
2594 			pr_err_client(cl, "%p auth cap %p not mds%d ???\n",
2595 				      inode, cap, session->s_mds);
2596 			break;
2597 		}
2598 
2599 		first_tid = cf->tid + 1;
2600 
2601 		if (!cf->is_capsnap) {
2602 			struct cap_msg_args arg;
2603 
2604 			doutc(cl, "%p %llx.%llx cap %p tid %llu %s\n",
2605 			      inode, ceph_vinop(inode), cap, cf->tid,
2606 			      ceph_cap_string(cf->caps));
2607 			__prep_cap(&arg, cap, CEPH_CAP_OP_FLUSH,
2608 					 (cf->tid < last_snap_flush ?
2609 					  CEPH_CLIENT_CAPS_PENDING_CAPSNAP : 0),
2610 					  __ceph_caps_used(ci),
2611 					  __ceph_caps_wanted(ci),
2612 					  (cap->issued | cap->implemented),
2613 					  cf->caps, cf->tid, oldest_flush_tid);
2614 			spin_unlock(&ci->i_ceph_lock);
2615 			__send_cap(&arg, ci);
2616 		} else {
2617 			struct ceph_cap_snap *capsnap =
2618 					container_of(cf, struct ceph_cap_snap,
2619 						    cap_flush);
2620 			doutc(cl, "%p %llx.%llx capsnap %p tid %llu %s\n",
2621 			      inode, ceph_vinop(inode), capsnap, cf->tid,
2622 			      ceph_cap_string(capsnap->dirty));
2623 
2624 			refcount_inc(&capsnap->nref);
2625 			spin_unlock(&ci->i_ceph_lock);
2626 
2627 			ret = __send_flush_snap(inode, session, capsnap, cap->mseq,
2628 						oldest_flush_tid);
2629 			if (ret < 0) {
2630 				pr_err_client(cl, "error sending cap flushsnap,"
2631 					      " %p %llx.%llx tid %llu follows %llu\n",
2632 					      inode, ceph_vinop(inode), cf->tid,
2633 					      capsnap->follows);
2634 			}
2635 
2636 			ceph_put_cap_snap(capsnap);
2637 		}
2638 
2639 		spin_lock(&ci->i_ceph_lock);
2640 	}
2641 }
2642 
ceph_early_kick_flushing_caps(struct ceph_mds_client * mdsc,struct ceph_mds_session * session)2643 void ceph_early_kick_flushing_caps(struct ceph_mds_client *mdsc,
2644 				   struct ceph_mds_session *session)
2645 {
2646 	struct ceph_client *cl = mdsc->fsc->client;
2647 	struct ceph_inode_info *ci;
2648 	struct ceph_cap *cap;
2649 	u64 oldest_flush_tid;
2650 
2651 	doutc(cl, "mds%d\n", session->s_mds);
2652 
2653 	spin_lock(&mdsc->cap_dirty_lock);
2654 	oldest_flush_tid = __get_oldest_flush_tid(mdsc);
2655 	spin_unlock(&mdsc->cap_dirty_lock);
2656 
2657 	list_for_each_entry(ci, &session->s_cap_flushing, i_flushing_item) {
2658 		struct inode *inode = &ci->netfs.inode;
2659 
2660 		spin_lock(&ci->i_ceph_lock);
2661 		cap = ci->i_auth_cap;
2662 		if (!(cap && cap->session == session)) {
2663 			pr_err_client(cl, "%p %llx.%llx auth cap %p not mds%d ???\n",
2664 				      inode, ceph_vinop(inode), cap,
2665 				      session->s_mds);
2666 			spin_unlock(&ci->i_ceph_lock);
2667 			continue;
2668 		}
2669 
2670 
2671 		/*
2672 		 * if flushing caps were revoked, we re-send the cap flush
2673 		 * in client reconnect stage. This guarantees MDS * processes
2674 		 * the cap flush message before issuing the flushing caps to
2675 		 * other client.
2676 		 */
2677 		if ((cap->issued & ci->i_flushing_caps) !=
2678 		    ci->i_flushing_caps) {
2679 			/* encode_caps_cb() also will reset these sequence
2680 			 * numbers. make sure sequence numbers in cap flush
2681 			 * message match later reconnect message */
2682 			cap->seq = 0;
2683 			cap->issue_seq = 0;
2684 			cap->mseq = 0;
2685 			__kick_flushing_caps(mdsc, session, ci,
2686 					     oldest_flush_tid);
2687 		} else {
2688 			ci->i_ceph_flags |= CEPH_I_KICK_FLUSH;
2689 		}
2690 
2691 		spin_unlock(&ci->i_ceph_lock);
2692 	}
2693 }
2694 
ceph_kick_flushing_caps(struct ceph_mds_client * mdsc,struct ceph_mds_session * session)2695 void ceph_kick_flushing_caps(struct ceph_mds_client *mdsc,
2696 			     struct ceph_mds_session *session)
2697 {
2698 	struct ceph_client *cl = mdsc->fsc->client;
2699 	struct ceph_inode_info *ci;
2700 	struct ceph_cap *cap;
2701 	u64 oldest_flush_tid;
2702 
2703 	lockdep_assert_held(&session->s_mutex);
2704 
2705 	doutc(cl, "mds%d\n", session->s_mds);
2706 
2707 	spin_lock(&mdsc->cap_dirty_lock);
2708 	oldest_flush_tid = __get_oldest_flush_tid(mdsc);
2709 	spin_unlock(&mdsc->cap_dirty_lock);
2710 
2711 	list_for_each_entry(ci, &session->s_cap_flushing, i_flushing_item) {
2712 		struct inode *inode = &ci->netfs.inode;
2713 
2714 		spin_lock(&ci->i_ceph_lock);
2715 		cap = ci->i_auth_cap;
2716 		if (!(cap && cap->session == session)) {
2717 			pr_err_client(cl, "%p %llx.%llx auth cap %p not mds%d ???\n",
2718 				      inode, ceph_vinop(inode), cap,
2719 				      session->s_mds);
2720 			spin_unlock(&ci->i_ceph_lock);
2721 			continue;
2722 		}
2723 		if (ci->i_ceph_flags & CEPH_I_KICK_FLUSH) {
2724 			__kick_flushing_caps(mdsc, session, ci,
2725 					     oldest_flush_tid);
2726 		}
2727 		spin_unlock(&ci->i_ceph_lock);
2728 	}
2729 }
2730 
ceph_kick_flushing_inode_caps(struct ceph_mds_session * session,struct ceph_inode_info * ci)2731 void ceph_kick_flushing_inode_caps(struct ceph_mds_session *session,
2732 				   struct ceph_inode_info *ci)
2733 {
2734 	struct ceph_mds_client *mdsc = session->s_mdsc;
2735 	struct ceph_cap *cap = ci->i_auth_cap;
2736 	struct inode *inode = &ci->netfs.inode;
2737 
2738 	lockdep_assert_held(&ci->i_ceph_lock);
2739 
2740 	doutc(mdsc->fsc->client, "%p %llx.%llx flushing %s\n",
2741 	      inode, ceph_vinop(inode),
2742 	      ceph_cap_string(ci->i_flushing_caps));
2743 
2744 	if (!list_empty(&ci->i_cap_flush_list)) {
2745 		u64 oldest_flush_tid;
2746 		spin_lock(&mdsc->cap_dirty_lock);
2747 		list_move_tail(&ci->i_flushing_item,
2748 			       &cap->session->s_cap_flushing);
2749 		oldest_flush_tid = __get_oldest_flush_tid(mdsc);
2750 		spin_unlock(&mdsc->cap_dirty_lock);
2751 
2752 		__kick_flushing_caps(mdsc, session, ci, oldest_flush_tid);
2753 	}
2754 }
2755 
2756 
2757 /*
2758  * Take references to capabilities we hold, so that we don't release
2759  * them to the MDS prematurely.
2760  */
ceph_take_cap_refs(struct ceph_inode_info * ci,int got,bool snap_rwsem_locked)2761 void ceph_take_cap_refs(struct ceph_inode_info *ci, int got,
2762 			    bool snap_rwsem_locked)
2763 {
2764 	struct inode *inode = &ci->netfs.inode;
2765 	struct ceph_client *cl = ceph_inode_to_client(inode);
2766 
2767 	lockdep_assert_held(&ci->i_ceph_lock);
2768 
2769 	if (got & CEPH_CAP_PIN)
2770 		ci->i_pin_ref++;
2771 	if (got & CEPH_CAP_FILE_RD)
2772 		ci->i_rd_ref++;
2773 	if (got & CEPH_CAP_FILE_CACHE)
2774 		ci->i_rdcache_ref++;
2775 	if (got & CEPH_CAP_FILE_EXCL)
2776 		ci->i_fx_ref++;
2777 	if (got & CEPH_CAP_FILE_WR) {
2778 		if (ci->i_wr_ref == 0 && !ci->i_head_snapc) {
2779 			BUG_ON(!snap_rwsem_locked);
2780 			ci->i_head_snapc = ceph_get_snap_context(
2781 					ci->i_snap_realm->cached_context);
2782 		}
2783 		ci->i_wr_ref++;
2784 	}
2785 	if (got & CEPH_CAP_FILE_BUFFER) {
2786 		if (ci->i_wb_ref == 0)
2787 			ihold(inode);
2788 		ci->i_wb_ref++;
2789 		doutc(cl, "%p %llx.%llx wb %d -> %d (?)\n", inode,
2790 		      ceph_vinop(inode), ci->i_wb_ref-1, ci->i_wb_ref);
2791 	}
2792 }
2793 
2794 /*
2795  * Try to grab cap references.  Specify those refs we @want, and the
2796  * minimal set we @need.  Also include the larger offset we are writing
2797  * to (when applicable), and check against max_size here as well.
2798  * Note that caller is responsible for ensuring max_size increases are
2799  * requested from the MDS.
2800  *
2801  * Returns 0 if caps were not able to be acquired (yet), 1 if succeed,
2802  * or a negative error code. There are 3 special error codes:
2803  *  -EAGAIN:  need to sleep but non-blocking is specified
2804  *  -EFBIG:   ask caller to call check_max_size() and try again.
2805  *  -EUCLEAN: ask caller to call ceph_renew_caps() and try again.
2806  */
2807 enum {
2808 	/* first 8 bits are reserved for CEPH_FILE_MODE_FOO */
2809 	NON_BLOCKING	= (1 << 8),
2810 	CHECK_FILELOCK	= (1 << 9),
2811 };
2812 
try_get_cap_refs(struct inode * inode,int need,int want,loff_t endoff,int flags,int * got)2813 static int try_get_cap_refs(struct inode *inode, int need, int want,
2814 			    loff_t endoff, int flags, int *got)
2815 {
2816 	struct ceph_inode_info *ci = ceph_inode(inode);
2817 	struct ceph_mds_client *mdsc = ceph_inode_to_fs_client(inode)->mdsc;
2818 	struct ceph_client *cl = ceph_inode_to_client(inode);
2819 	int ret = 0;
2820 	int have, implemented;
2821 	bool snap_rwsem_locked = false;
2822 
2823 	doutc(cl, "%p %llx.%llx need %s want %s\n", inode,
2824 	      ceph_vinop(inode), ceph_cap_string(need),
2825 	      ceph_cap_string(want));
2826 
2827 again:
2828 	spin_lock(&ci->i_ceph_lock);
2829 
2830 	if ((flags & CHECK_FILELOCK) &&
2831 	    (ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK)) {
2832 		doutc(cl, "%p %llx.%llx error filelock\n", inode,
2833 		      ceph_vinop(inode));
2834 		ret = -EIO;
2835 		goto out_unlock;
2836 	}
2837 
2838 	/* finish pending truncate */
2839 	while (ci->i_truncate_pending) {
2840 		spin_unlock(&ci->i_ceph_lock);
2841 		if (snap_rwsem_locked) {
2842 			up_read(&mdsc->snap_rwsem);
2843 			snap_rwsem_locked = false;
2844 		}
2845 		__ceph_do_pending_vmtruncate(inode);
2846 		spin_lock(&ci->i_ceph_lock);
2847 	}
2848 
2849 	have = __ceph_caps_issued(ci, &implemented);
2850 
2851 	if (have & need & CEPH_CAP_FILE_WR) {
2852 		if (endoff >= 0 && endoff > (loff_t)ci->i_max_size) {
2853 			doutc(cl, "%p %llx.%llx endoff %llu > maxsize %llu\n",
2854 			      inode, ceph_vinop(inode), endoff, ci->i_max_size);
2855 			if (endoff > ci->i_requested_max_size)
2856 				ret = ci->i_auth_cap ? -EFBIG : -EUCLEAN;
2857 			goto out_unlock;
2858 		}
2859 		/*
2860 		 * If a sync write is in progress, we must wait, so that we
2861 		 * can get a final snapshot value for size+mtime.
2862 		 */
2863 		if (__ceph_have_pending_cap_snap(ci)) {
2864 			doutc(cl, "%p %llx.%llx cap_snap_pending\n", inode,
2865 			      ceph_vinop(inode));
2866 			goto out_unlock;
2867 		}
2868 	}
2869 
2870 	if ((have & need) == need) {
2871 		/*
2872 		 * Look at (implemented & ~have & not) so that we keep waiting
2873 		 * on transition from wanted -> needed caps.  This is needed
2874 		 * for WRBUFFER|WR -> WR to avoid a new WR sync write from
2875 		 * going before a prior buffered writeback happens.
2876 		 *
2877 		 * For RDCACHE|RD -> RD, there is not need to wait and we can
2878 		 * just exclude the revoking caps and force to sync read.
2879 		 */
2880 		int not = want & ~(have & need);
2881 		int revoking = implemented & ~have;
2882 		int exclude = revoking & not;
2883 		doutc(cl, "%p %llx.%llx have %s but not %s (revoking %s)\n",
2884 		      inode, ceph_vinop(inode), ceph_cap_string(have),
2885 		      ceph_cap_string(not), ceph_cap_string(revoking));
2886 		if (!exclude || !(exclude & CEPH_CAP_FILE_BUFFER)) {
2887 			if (!snap_rwsem_locked &&
2888 			    !ci->i_head_snapc &&
2889 			    (need & CEPH_CAP_FILE_WR)) {
2890 				if (!down_read_trylock(&mdsc->snap_rwsem)) {
2891 					/*
2892 					 * we can not call down_read() when
2893 					 * task isn't in TASK_RUNNING state
2894 					 */
2895 					if (flags & NON_BLOCKING) {
2896 						ret = -EAGAIN;
2897 						goto out_unlock;
2898 					}
2899 
2900 					spin_unlock(&ci->i_ceph_lock);
2901 					down_read(&mdsc->snap_rwsem);
2902 					snap_rwsem_locked = true;
2903 					goto again;
2904 				}
2905 				snap_rwsem_locked = true;
2906 			}
2907 			if ((have & want) == want)
2908 				*got = need | (want & ~exclude);
2909 			else
2910 				*got = need;
2911 			ceph_take_cap_refs(ci, *got, true);
2912 			ret = 1;
2913 		}
2914 	} else {
2915 		int session_readonly = false;
2916 		int mds_wanted;
2917 		if (ci->i_auth_cap &&
2918 		    (need & (CEPH_CAP_FILE_WR | CEPH_CAP_FILE_EXCL))) {
2919 			struct ceph_mds_session *s = ci->i_auth_cap->session;
2920 			spin_lock(&s->s_cap_lock);
2921 			session_readonly = s->s_readonly;
2922 			spin_unlock(&s->s_cap_lock);
2923 		}
2924 		if (session_readonly) {
2925 			doutc(cl, "%p %llx.%llx need %s but mds%d readonly\n",
2926 			      inode, ceph_vinop(inode), ceph_cap_string(need),
2927 			      ci->i_auth_cap->mds);
2928 			ret = -EROFS;
2929 			goto out_unlock;
2930 		}
2931 
2932 		if (ceph_inode_is_shutdown(inode)) {
2933 			doutc(cl, "%p %llx.%llx inode is shutdown\n",
2934 			      inode, ceph_vinop(inode));
2935 			ret = -ESTALE;
2936 			goto out_unlock;
2937 		}
2938 		mds_wanted = __ceph_caps_mds_wanted(ci, false);
2939 		if (need & ~mds_wanted) {
2940 			doutc(cl, "%p %llx.%llx need %s > mds_wanted %s\n",
2941 			      inode, ceph_vinop(inode), ceph_cap_string(need),
2942 			      ceph_cap_string(mds_wanted));
2943 			ret = -EUCLEAN;
2944 			goto out_unlock;
2945 		}
2946 
2947 		doutc(cl, "%p %llx.%llx have %s need %s\n", inode,
2948 		      ceph_vinop(inode), ceph_cap_string(have),
2949 		      ceph_cap_string(need));
2950 	}
2951 out_unlock:
2952 
2953 	__ceph_touch_fmode(ci, mdsc, flags);
2954 
2955 	spin_unlock(&ci->i_ceph_lock);
2956 	if (snap_rwsem_locked)
2957 		up_read(&mdsc->snap_rwsem);
2958 
2959 	if (!ret)
2960 		ceph_update_cap_mis(&mdsc->metric);
2961 	else if (ret == 1)
2962 		ceph_update_cap_hit(&mdsc->metric);
2963 
2964 	doutc(cl, "%p %llx.%llx ret %d got %s\n", inode,
2965 	      ceph_vinop(inode), ret, ceph_cap_string(*got));
2966 	return ret;
2967 }
2968 
2969 /*
2970  * Check the offset we are writing up to against our current
2971  * max_size.  If necessary, tell the MDS we want to write to
2972  * a larger offset.
2973  */
check_max_size(struct inode * inode,loff_t endoff)2974 static void check_max_size(struct inode *inode, loff_t endoff)
2975 {
2976 	struct ceph_inode_info *ci = ceph_inode(inode);
2977 	struct ceph_client *cl = ceph_inode_to_client(inode);
2978 	int check = 0;
2979 
2980 	/* do we need to explicitly request a larger max_size? */
2981 	spin_lock(&ci->i_ceph_lock);
2982 	if (endoff >= ci->i_max_size && endoff > ci->i_wanted_max_size) {
2983 		doutc(cl, "write %p %llx.%llx at large endoff %llu, req max_size\n",
2984 		      inode, ceph_vinop(inode), endoff);
2985 		ci->i_wanted_max_size = endoff;
2986 	}
2987 	/* duplicate ceph_check_caps()'s logic */
2988 	if (ci->i_auth_cap &&
2989 	    (ci->i_auth_cap->issued & CEPH_CAP_FILE_WR) &&
2990 	    ci->i_wanted_max_size > ci->i_max_size &&
2991 	    ci->i_wanted_max_size > ci->i_requested_max_size)
2992 		check = 1;
2993 	spin_unlock(&ci->i_ceph_lock);
2994 	if (check)
2995 		ceph_check_caps(ci, CHECK_CAPS_AUTHONLY);
2996 }
2997 
get_used_fmode(int caps)2998 static inline int get_used_fmode(int caps)
2999 {
3000 	int fmode = 0;
3001 	if (caps & CEPH_CAP_FILE_RD)
3002 		fmode |= CEPH_FILE_MODE_RD;
3003 	if (caps & CEPH_CAP_FILE_WR)
3004 		fmode |= CEPH_FILE_MODE_WR;
3005 	return fmode;
3006 }
3007 
ceph_try_get_caps(struct inode * inode,int need,int want,bool nonblock,int * got)3008 int ceph_try_get_caps(struct inode *inode, int need, int want,
3009 		      bool nonblock, int *got)
3010 {
3011 	int ret, flags;
3012 
3013 	BUG_ON(need & ~CEPH_CAP_FILE_RD);
3014 	BUG_ON(want & ~(CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO |
3015 			CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
3016 			CEPH_CAP_ANY_DIR_OPS));
3017 	if (need) {
3018 		ret = ceph_pool_perm_check(inode, need);
3019 		if (ret < 0)
3020 			return ret;
3021 	}
3022 
3023 	flags = get_used_fmode(need | want);
3024 	if (nonblock)
3025 		flags |= NON_BLOCKING;
3026 
3027 	ret = try_get_cap_refs(inode, need, want, 0, flags, got);
3028 	/* three special error codes */
3029 	if (ret == -EAGAIN || ret == -EFBIG || ret == -EUCLEAN)
3030 		ret = 0;
3031 	return ret;
3032 }
3033 
3034 /*
3035  * Wait for caps, and take cap references.  If we can't get a WR cap
3036  * due to a small max_size, make sure we check_max_size (and possibly
3037  * ask the mds) so we don't get hung up indefinitely.
3038  */
__ceph_get_caps(struct inode * inode,struct ceph_file_info * fi,int need,int want,loff_t endoff,int * got)3039 int __ceph_get_caps(struct inode *inode, struct ceph_file_info *fi, int need,
3040 		    int want, loff_t endoff, int *got)
3041 {
3042 	struct ceph_inode_info *ci = ceph_inode(inode);
3043 	struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
3044 	int ret, _got, flags;
3045 
3046 	ret = ceph_pool_perm_check(inode, need);
3047 	if (ret < 0)
3048 		return ret;
3049 
3050 	if (fi && (fi->fmode & CEPH_FILE_MODE_WR) &&
3051 	    fi->filp_gen != READ_ONCE(fsc->filp_gen))
3052 		return -EBADF;
3053 
3054 	flags = get_used_fmode(need | want);
3055 
3056 	while (true) {
3057 		flags &= CEPH_FILE_MODE_MASK;
3058 		if (vfs_inode_has_locks(inode))
3059 			flags |= CHECK_FILELOCK;
3060 		_got = 0;
3061 		ret = try_get_cap_refs(inode, need, want, endoff,
3062 				       flags, &_got);
3063 		WARN_ON_ONCE(ret == -EAGAIN);
3064 		if (!ret) {
3065 #ifdef CONFIG_DEBUG_FS
3066 			struct ceph_mds_client *mdsc = fsc->mdsc;
3067 			struct cap_wait cw;
3068 #endif
3069 			DEFINE_WAIT_FUNC(wait, woken_wake_function);
3070 
3071 #ifdef CONFIG_DEBUG_FS
3072 			cw.ino = ceph_ino(inode);
3073 			cw.tgid = current->tgid;
3074 			cw.need = need;
3075 			cw.want = want;
3076 
3077 			spin_lock(&mdsc->caps_list_lock);
3078 			list_add(&cw.list, &mdsc->cap_wait_list);
3079 			spin_unlock(&mdsc->caps_list_lock);
3080 #endif
3081 
3082 			/* make sure used fmode not timeout */
3083 			ceph_get_fmode(ci, flags, FMODE_WAIT_BIAS);
3084 			add_wait_queue(&ci->i_cap_wq, &wait);
3085 
3086 			flags |= NON_BLOCKING;
3087 			while (!(ret = try_get_cap_refs(inode, need, want,
3088 							endoff, flags, &_got))) {
3089 				if (signal_pending(current)) {
3090 					ret = -ERESTARTSYS;
3091 					break;
3092 				}
3093 				wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
3094 			}
3095 
3096 			remove_wait_queue(&ci->i_cap_wq, &wait);
3097 			ceph_put_fmode(ci, flags, FMODE_WAIT_BIAS);
3098 
3099 #ifdef CONFIG_DEBUG_FS
3100 			spin_lock(&mdsc->caps_list_lock);
3101 			list_del(&cw.list);
3102 			spin_unlock(&mdsc->caps_list_lock);
3103 #endif
3104 
3105 			if (ret == -EAGAIN)
3106 				continue;
3107 		}
3108 
3109 		if (fi && (fi->fmode & CEPH_FILE_MODE_WR) &&
3110 		    fi->filp_gen != READ_ONCE(fsc->filp_gen)) {
3111 			if (ret >= 0 && _got)
3112 				ceph_put_cap_refs(ci, _got);
3113 			return -EBADF;
3114 		}
3115 
3116 		if (ret < 0) {
3117 			if (ret == -EFBIG || ret == -EUCLEAN) {
3118 				int ret2 = ceph_wait_on_async_create(inode);
3119 				if (ret2 < 0)
3120 					return ret2;
3121 			}
3122 			if (ret == -EFBIG) {
3123 				check_max_size(inode, endoff);
3124 				continue;
3125 			}
3126 			if (ret == -EUCLEAN) {
3127 				/* session was killed, try renew caps */
3128 				ret = ceph_renew_caps(inode, flags);
3129 				if (ret == 0)
3130 					continue;
3131 			}
3132 			return ret;
3133 		}
3134 
3135 		if (S_ISREG(ci->netfs.inode.i_mode) &&
3136 		    ceph_has_inline_data(ci) &&
3137 		    (_got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) &&
3138 		    i_size_read(inode) > 0) {
3139 			struct page *page =
3140 				find_get_page(inode->i_mapping, 0);
3141 			if (page) {
3142 				bool uptodate = PageUptodate(page);
3143 
3144 				put_page(page);
3145 				if (uptodate)
3146 					break;
3147 			}
3148 			/*
3149 			 * drop cap refs first because getattr while
3150 			 * holding * caps refs can cause deadlock.
3151 			 */
3152 			ceph_put_cap_refs(ci, _got);
3153 			_got = 0;
3154 
3155 			/*
3156 			 * getattr request will bring inline data into
3157 			 * page cache
3158 			 */
3159 			ret = __ceph_do_getattr(inode, NULL,
3160 						CEPH_STAT_CAP_INLINE_DATA,
3161 						true);
3162 			if (ret < 0)
3163 				return ret;
3164 			continue;
3165 		}
3166 		break;
3167 	}
3168 	*got = _got;
3169 	return 0;
3170 }
3171 
ceph_get_caps(struct file * filp,int need,int want,loff_t endoff,int * got)3172 int ceph_get_caps(struct file *filp, int need, int want, loff_t endoff,
3173 		  int *got)
3174 {
3175 	struct ceph_file_info *fi = filp->private_data;
3176 	struct inode *inode = file_inode(filp);
3177 
3178 	return __ceph_get_caps(inode, fi, need, want, endoff, got);
3179 }
3180 
3181 /*
3182  * Take cap refs.  Caller must already know we hold at least one ref
3183  * on the caps in question or we don't know this is safe.
3184  */
ceph_get_cap_refs(struct ceph_inode_info * ci,int caps)3185 void ceph_get_cap_refs(struct ceph_inode_info *ci, int caps)
3186 {
3187 	spin_lock(&ci->i_ceph_lock);
3188 	ceph_take_cap_refs(ci, caps, false);
3189 	spin_unlock(&ci->i_ceph_lock);
3190 }
3191 
3192 
3193 /*
3194  * drop cap_snap that is not associated with any snapshot.
3195  * we don't need to send FLUSHSNAP message for it.
3196  */
ceph_try_drop_cap_snap(struct ceph_inode_info * ci,struct ceph_cap_snap * capsnap)3197 static int ceph_try_drop_cap_snap(struct ceph_inode_info *ci,
3198 				  struct ceph_cap_snap *capsnap)
3199 {
3200 	struct inode *inode = &ci->netfs.inode;
3201 	struct ceph_client *cl = ceph_inode_to_client(inode);
3202 
3203 	if (!capsnap->need_flush &&
3204 	    !capsnap->writing && !capsnap->dirty_pages) {
3205 		doutc(cl, "%p follows %llu\n", capsnap, capsnap->follows);
3206 		BUG_ON(capsnap->cap_flush.tid > 0);
3207 		ceph_put_snap_context(capsnap->context);
3208 		if (!list_is_last(&capsnap->ci_item, &ci->i_cap_snaps))
3209 			ci->i_ceph_flags |= CEPH_I_FLUSH_SNAPS;
3210 
3211 		list_del(&capsnap->ci_item);
3212 		ceph_put_cap_snap(capsnap);
3213 		return 1;
3214 	}
3215 	return 0;
3216 }
3217 
3218 enum put_cap_refs_mode {
3219 	PUT_CAP_REFS_SYNC = 0,
3220 	PUT_CAP_REFS_ASYNC,
3221 };
3222 
3223 /*
3224  * Release cap refs.
3225  *
3226  * If we released the last ref on any given cap, call ceph_check_caps
3227  * to release (or schedule a release).
3228  *
3229  * If we are releasing a WR cap (from a sync write), finalize any affected
3230  * cap_snap, and wake up any waiters.
3231  */
__ceph_put_cap_refs(struct ceph_inode_info * ci,int had,enum put_cap_refs_mode mode)3232 static void __ceph_put_cap_refs(struct ceph_inode_info *ci, int had,
3233 				enum put_cap_refs_mode mode)
3234 {
3235 	struct inode *inode = &ci->netfs.inode;
3236 	struct ceph_client *cl = ceph_inode_to_client(inode);
3237 	int last = 0, put = 0, flushsnaps = 0, wake = 0;
3238 	bool check_flushsnaps = false;
3239 
3240 	spin_lock(&ci->i_ceph_lock);
3241 	if (had & CEPH_CAP_PIN)
3242 		--ci->i_pin_ref;
3243 	if (had & CEPH_CAP_FILE_RD)
3244 		if (--ci->i_rd_ref == 0)
3245 			last++;
3246 	if (had & CEPH_CAP_FILE_CACHE)
3247 		if (--ci->i_rdcache_ref == 0)
3248 			last++;
3249 	if (had & CEPH_CAP_FILE_EXCL)
3250 		if (--ci->i_fx_ref == 0)
3251 			last++;
3252 	if (had & CEPH_CAP_FILE_BUFFER) {
3253 		if (--ci->i_wb_ref == 0) {
3254 			last++;
3255 			/* put the ref held by ceph_take_cap_refs() */
3256 			put++;
3257 			check_flushsnaps = true;
3258 		}
3259 		doutc(cl, "%p %llx.%llx wb %d -> %d (?)\n", inode,
3260 		      ceph_vinop(inode), ci->i_wb_ref+1, ci->i_wb_ref);
3261 	}
3262 	if (had & CEPH_CAP_FILE_WR) {
3263 		if (--ci->i_wr_ref == 0) {
3264 			/*
3265 			 * The Fb caps will always be took and released
3266 			 * together with the Fw caps.
3267 			 */
3268 			WARN_ON_ONCE(ci->i_wb_ref);
3269 
3270 			last++;
3271 			check_flushsnaps = true;
3272 			if (ci->i_wrbuffer_ref_head == 0 &&
3273 			    ci->i_dirty_caps == 0 &&
3274 			    ci->i_flushing_caps == 0) {
3275 				BUG_ON(!ci->i_head_snapc);
3276 				ceph_put_snap_context(ci->i_head_snapc);
3277 				ci->i_head_snapc = NULL;
3278 			}
3279 			/* see comment in __ceph_remove_cap() */
3280 			if (!__ceph_is_any_real_caps(ci) && ci->i_snap_realm)
3281 				ceph_change_snap_realm(inode, NULL);
3282 		}
3283 	}
3284 	if (check_flushsnaps && __ceph_have_pending_cap_snap(ci)) {
3285 		struct ceph_cap_snap *capsnap =
3286 			list_last_entry(&ci->i_cap_snaps,
3287 					struct ceph_cap_snap,
3288 					ci_item);
3289 
3290 		capsnap->writing = 0;
3291 		if (ceph_try_drop_cap_snap(ci, capsnap))
3292 			/* put the ref held by ceph_queue_cap_snap() */
3293 			put++;
3294 		else if (__ceph_finish_cap_snap(ci, capsnap))
3295 			flushsnaps = 1;
3296 		wake = 1;
3297 	}
3298 	spin_unlock(&ci->i_ceph_lock);
3299 
3300 	doutc(cl, "%p %llx.%llx had %s%s%s\n", inode, ceph_vinop(inode),
3301 	      ceph_cap_string(had), last ? " last" : "", put ? " put" : "");
3302 
3303 	switch (mode) {
3304 	case PUT_CAP_REFS_SYNC:
3305 		if (last)
3306 			ceph_check_caps(ci, 0);
3307 		else if (flushsnaps)
3308 			ceph_flush_snaps(ci, NULL);
3309 		break;
3310 	case PUT_CAP_REFS_ASYNC:
3311 		if (last)
3312 			ceph_queue_check_caps(inode);
3313 		else if (flushsnaps)
3314 			ceph_queue_flush_snaps(inode);
3315 		break;
3316 	default:
3317 		break;
3318 	}
3319 	if (wake)
3320 		wake_up_all(&ci->i_cap_wq);
3321 	while (put-- > 0)
3322 		iput(inode);
3323 }
3324 
ceph_put_cap_refs(struct ceph_inode_info * ci,int had)3325 void ceph_put_cap_refs(struct ceph_inode_info *ci, int had)
3326 {
3327 	__ceph_put_cap_refs(ci, had, PUT_CAP_REFS_SYNC);
3328 }
3329 
ceph_put_cap_refs_async(struct ceph_inode_info * ci,int had)3330 void ceph_put_cap_refs_async(struct ceph_inode_info *ci, int had)
3331 {
3332 	__ceph_put_cap_refs(ci, had, PUT_CAP_REFS_ASYNC);
3333 }
3334 
3335 /*
3336  * Release @nr WRBUFFER refs on dirty pages for the given @snapc snap
3337  * context.  Adjust per-snap dirty page accounting as appropriate.
3338  * Once all dirty data for a cap_snap is flushed, flush snapped file
3339  * metadata back to the MDS.  If we dropped the last ref, call
3340  * ceph_check_caps.
3341  */
ceph_put_wrbuffer_cap_refs(struct ceph_inode_info * ci,int nr,struct ceph_snap_context * snapc)3342 void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info *ci, int nr,
3343 				struct ceph_snap_context *snapc)
3344 {
3345 	struct inode *inode = &ci->netfs.inode;
3346 	struct ceph_client *cl = ceph_inode_to_client(inode);
3347 	struct ceph_cap_snap *capsnap = NULL, *iter;
3348 	int put = 0;
3349 	bool last = false;
3350 	bool flush_snaps = false;
3351 	bool complete_capsnap = false;
3352 
3353 	spin_lock(&ci->i_ceph_lock);
3354 	ci->i_wrbuffer_ref -= nr;
3355 	if (ci->i_wrbuffer_ref == 0) {
3356 		last = true;
3357 		put++;
3358 	}
3359 
3360 	if (ci->i_head_snapc == snapc) {
3361 		ci->i_wrbuffer_ref_head -= nr;
3362 		if (ci->i_wrbuffer_ref_head == 0 &&
3363 		    ci->i_wr_ref == 0 &&
3364 		    ci->i_dirty_caps == 0 &&
3365 		    ci->i_flushing_caps == 0) {
3366 			BUG_ON(!ci->i_head_snapc);
3367 			ceph_put_snap_context(ci->i_head_snapc);
3368 			ci->i_head_snapc = NULL;
3369 		}
3370 		doutc(cl, "on %p %llx.%llx head %d/%d -> %d/%d %s\n",
3371 		      inode, ceph_vinop(inode), ci->i_wrbuffer_ref+nr,
3372 		      ci->i_wrbuffer_ref_head+nr, ci->i_wrbuffer_ref,
3373 		      ci->i_wrbuffer_ref_head, last ? " LAST" : "");
3374 	} else {
3375 		list_for_each_entry(iter, &ci->i_cap_snaps, ci_item) {
3376 			if (iter->context == snapc) {
3377 				capsnap = iter;
3378 				break;
3379 			}
3380 		}
3381 
3382 		if (!capsnap) {
3383 			/*
3384 			 * The capsnap should already be removed when removing
3385 			 * auth cap in the case of a forced unmount.
3386 			 */
3387 			WARN_ON_ONCE(ci->i_auth_cap);
3388 			goto unlock;
3389 		}
3390 
3391 		capsnap->dirty_pages -= nr;
3392 		if (capsnap->dirty_pages == 0) {
3393 			complete_capsnap = true;
3394 			if (!capsnap->writing) {
3395 				if (ceph_try_drop_cap_snap(ci, capsnap)) {
3396 					put++;
3397 				} else {
3398 					ci->i_ceph_flags |= CEPH_I_FLUSH_SNAPS;
3399 					flush_snaps = true;
3400 				}
3401 			}
3402 		}
3403 		doutc(cl, "%p %llx.%llx cap_snap %p snap %lld %d/%d -> %d/%d %s%s\n",
3404 		      inode, ceph_vinop(inode), capsnap, capsnap->context->seq,
3405 		      ci->i_wrbuffer_ref+nr, capsnap->dirty_pages + nr,
3406 		      ci->i_wrbuffer_ref, capsnap->dirty_pages,
3407 		      last ? " (wrbuffer last)" : "",
3408 		      complete_capsnap ? " (complete capsnap)" : "");
3409 	}
3410 
3411 unlock:
3412 	spin_unlock(&ci->i_ceph_lock);
3413 
3414 	if (last) {
3415 		ceph_check_caps(ci, 0);
3416 	} else if (flush_snaps) {
3417 		ceph_flush_snaps(ci, NULL);
3418 	}
3419 	if (complete_capsnap)
3420 		wake_up_all(&ci->i_cap_wq);
3421 	while (put-- > 0) {
3422 		iput(inode);
3423 	}
3424 }
3425 
3426 /*
3427  * Invalidate unlinked inode's aliases, so we can drop the inode ASAP.
3428  */
invalidate_aliases(struct inode * inode)3429 static void invalidate_aliases(struct inode *inode)
3430 {
3431 	struct ceph_client *cl = ceph_inode_to_client(inode);
3432 	struct dentry *dn, *prev = NULL;
3433 
3434 	doutc(cl, "%p %llx.%llx\n", inode, ceph_vinop(inode));
3435 	d_prune_aliases(inode);
3436 	/*
3437 	 * For non-directory inode, d_find_alias() only returns
3438 	 * hashed dentry. After calling d_invalidate(), the
3439 	 * dentry becomes unhashed.
3440 	 *
3441 	 * For directory inode, d_find_alias() can return
3442 	 * unhashed dentry. But directory inode should have
3443 	 * one alias at most.
3444 	 */
3445 	while ((dn = d_find_alias(inode))) {
3446 		if (dn == prev) {
3447 			dput(dn);
3448 			break;
3449 		}
3450 		d_invalidate(dn);
3451 		if (prev)
3452 			dput(prev);
3453 		prev = dn;
3454 	}
3455 	if (prev)
3456 		dput(prev);
3457 }
3458 
3459 struct cap_extra_info {
3460 	struct ceph_string *pool_ns;
3461 	/* inline data */
3462 	u64 inline_version;
3463 	void *inline_data;
3464 	u32 inline_len;
3465 	/* dirstat */
3466 	bool dirstat_valid;
3467 	u64 nfiles;
3468 	u64 nsubdirs;
3469 	u64 change_attr;
3470 	/* currently issued */
3471 	int issued;
3472 	struct timespec64 btime;
3473 	u8 *fscrypt_auth;
3474 	u32 fscrypt_auth_len;
3475 	u64 fscrypt_file_size;
3476 };
3477 
3478 /*
3479  * Handle a cap GRANT message from the MDS.  (Note that a GRANT may
3480  * actually be a revocation if it specifies a smaller cap set.)
3481  *
3482  * caller holds s_mutex and i_ceph_lock, we drop both.
3483  */
handle_cap_grant(struct inode * inode,struct ceph_mds_session * session,struct ceph_cap * cap,struct ceph_mds_caps * grant,struct ceph_buffer * xattr_buf,struct cap_extra_info * extra_info)3484 static void handle_cap_grant(struct inode *inode,
3485 			     struct ceph_mds_session *session,
3486 			     struct ceph_cap *cap,
3487 			     struct ceph_mds_caps *grant,
3488 			     struct ceph_buffer *xattr_buf,
3489 			     struct cap_extra_info *extra_info)
3490 	__releases(ci->i_ceph_lock)
3491 	__releases(session->s_mdsc->snap_rwsem)
3492 {
3493 	struct ceph_client *cl = ceph_inode_to_client(inode);
3494 	struct ceph_inode_info *ci = ceph_inode(inode);
3495 	int seq = le32_to_cpu(grant->seq);
3496 	int newcaps = le32_to_cpu(grant->caps);
3497 	int used, wanted, dirty;
3498 	u64 size = le64_to_cpu(grant->size);
3499 	u64 max_size = le64_to_cpu(grant->max_size);
3500 	unsigned char check_caps = 0;
3501 	bool was_stale = cap->cap_gen < atomic_read(&session->s_cap_gen);
3502 	bool wake = false;
3503 	bool writeback = false;
3504 	bool queue_trunc = false;
3505 	bool queue_invalidate = false;
3506 	bool deleted_inode = false;
3507 	bool fill_inline = false;
3508 	bool revoke_wait = false;
3509 	int flags = 0;
3510 
3511 	/*
3512 	 * If there is at least one crypto block then we'll trust
3513 	 * fscrypt_file_size. If the real length of the file is 0, then
3514 	 * ignore it (it has probably been truncated down to 0 by the MDS).
3515 	 */
3516 	if (IS_ENCRYPTED(inode) && size)
3517 		size = extra_info->fscrypt_file_size;
3518 
3519 	doutc(cl, "%p %llx.%llx cap %p mds%d seq %d %s\n", inode,
3520 	      ceph_vinop(inode), cap, session->s_mds, seq,
3521 	      ceph_cap_string(newcaps));
3522 	doutc(cl, " size %llu max_size %llu, i_size %llu\n", size,
3523 	      max_size, i_size_read(inode));
3524 
3525 
3526 	/*
3527 	 * If CACHE is being revoked, and we have no dirty buffers,
3528 	 * try to invalidate (once).  (If there are dirty buffers, we
3529 	 * will invalidate _after_ writeback.)
3530 	 */
3531 	if (S_ISREG(inode->i_mode) && /* don't invalidate readdir cache */
3532 	    ((cap->issued & ~newcaps) & CEPH_CAP_FILE_CACHE) &&
3533 	    (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 &&
3534 	    !(ci->i_wrbuffer_ref || ci->i_wb_ref)) {
3535 		if (try_nonblocking_invalidate(inode)) {
3536 			/* there were locked pages.. invalidate later
3537 			   in a separate thread. */
3538 			if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
3539 				queue_invalidate = true;
3540 				ci->i_rdcache_revoking = ci->i_rdcache_gen;
3541 			}
3542 		}
3543 	}
3544 
3545 	if (was_stale)
3546 		cap->issued = cap->implemented = CEPH_CAP_PIN;
3547 
3548 	/*
3549 	 * auth mds of the inode changed. we received the cap export message,
3550 	 * but still haven't received the cap import message. handle_cap_export
3551 	 * updated the new auth MDS' cap.
3552 	 *
3553 	 * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing a message
3554 	 * that was sent before the cap import message. So don't remove caps.
3555 	 */
3556 	if (ceph_seq_cmp(seq, cap->seq) <= 0) {
3557 		WARN_ON(cap != ci->i_auth_cap);
3558 		WARN_ON(cap->cap_id != le64_to_cpu(grant->cap_id));
3559 		seq = cap->seq;
3560 		newcaps |= cap->issued;
3561 	}
3562 
3563 	/* side effects now are allowed */
3564 	cap->cap_gen = atomic_read(&session->s_cap_gen);
3565 	cap->seq = seq;
3566 
3567 	__check_cap_issue(ci, cap, newcaps);
3568 
3569 	inode_set_max_iversion_raw(inode, extra_info->change_attr);
3570 
3571 	if ((newcaps & CEPH_CAP_AUTH_SHARED) &&
3572 	    (extra_info->issued & CEPH_CAP_AUTH_EXCL) == 0) {
3573 		umode_t mode = le32_to_cpu(grant->mode);
3574 
3575 		if (inode_wrong_type(inode, mode))
3576 			pr_warn_once("inode type changed! (ino %llx.%llx is 0%o, mds says 0%o)\n",
3577 				     ceph_vinop(inode), inode->i_mode, mode);
3578 		else
3579 			inode->i_mode = mode;
3580 		inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(grant->uid));
3581 		inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(grant->gid));
3582 		ci->i_btime = extra_info->btime;
3583 		doutc(cl, "%p %llx.%llx mode 0%o uid.gid %d.%d\n", inode,
3584 		      ceph_vinop(inode), inode->i_mode,
3585 		      from_kuid(&init_user_ns, inode->i_uid),
3586 		      from_kgid(&init_user_ns, inode->i_gid));
3587 #if IS_ENABLED(CONFIG_FS_ENCRYPTION)
3588 		if (ci->fscrypt_auth_len != extra_info->fscrypt_auth_len ||
3589 		    memcmp(ci->fscrypt_auth, extra_info->fscrypt_auth,
3590 			   ci->fscrypt_auth_len))
3591 			pr_warn_ratelimited_client(cl,
3592 				"cap grant attempt to change fscrypt_auth on non-I_NEW inode (old len %d new len %d)\n",
3593 				ci->fscrypt_auth_len,
3594 				extra_info->fscrypt_auth_len);
3595 #endif
3596 	}
3597 
3598 	if ((newcaps & CEPH_CAP_LINK_SHARED) &&
3599 	    (extra_info->issued & CEPH_CAP_LINK_EXCL) == 0) {
3600 		set_nlink(inode, le32_to_cpu(grant->nlink));
3601 		if (inode->i_nlink == 0)
3602 			deleted_inode = true;
3603 	}
3604 
3605 	if ((extra_info->issued & CEPH_CAP_XATTR_EXCL) == 0 &&
3606 	    grant->xattr_len) {
3607 		int len = le32_to_cpu(grant->xattr_len);
3608 		u64 version = le64_to_cpu(grant->xattr_version);
3609 
3610 		if (version > ci->i_xattrs.version) {
3611 			doutc(cl, " got new xattrs v%llu on %p %llx.%llx len %d\n",
3612 			      version, inode, ceph_vinop(inode), len);
3613 			if (ci->i_xattrs.blob)
3614 				ceph_buffer_put(ci->i_xattrs.blob);
3615 			ci->i_xattrs.blob = ceph_buffer_get(xattr_buf);
3616 			ci->i_xattrs.version = version;
3617 			ceph_forget_all_cached_acls(inode);
3618 			ceph_security_invalidate_secctx(inode);
3619 		}
3620 	}
3621 
3622 	if (newcaps & CEPH_CAP_ANY_RD) {
3623 		struct timespec64 mtime, atime, ctime;
3624 		/* ctime/mtime/atime? */
3625 		ceph_decode_timespec64(&mtime, &grant->mtime);
3626 		ceph_decode_timespec64(&atime, &grant->atime);
3627 		ceph_decode_timespec64(&ctime, &grant->ctime);
3628 		ceph_fill_file_time(inode, extra_info->issued,
3629 				    le32_to_cpu(grant->time_warp_seq),
3630 				    &ctime, &mtime, &atime);
3631 	}
3632 
3633 	if ((newcaps & CEPH_CAP_FILE_SHARED) && extra_info->dirstat_valid) {
3634 		ci->i_files = extra_info->nfiles;
3635 		ci->i_subdirs = extra_info->nsubdirs;
3636 	}
3637 
3638 	if (newcaps & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR)) {
3639 		/* file layout may have changed */
3640 		s64 old_pool = ci->i_layout.pool_id;
3641 		struct ceph_string *old_ns;
3642 
3643 		ceph_file_layout_from_legacy(&ci->i_layout, &grant->layout);
3644 		old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
3645 					lockdep_is_held(&ci->i_ceph_lock));
3646 		rcu_assign_pointer(ci->i_layout.pool_ns, extra_info->pool_ns);
3647 
3648 		if (ci->i_layout.pool_id != old_pool ||
3649 		    extra_info->pool_ns != old_ns)
3650 			ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
3651 
3652 		extra_info->pool_ns = old_ns;
3653 
3654 		/* size/truncate_seq? */
3655 		queue_trunc = ceph_fill_file_size(inode, extra_info->issued,
3656 					le32_to_cpu(grant->truncate_seq),
3657 					le64_to_cpu(grant->truncate_size),
3658 					size);
3659 	}
3660 
3661 	if (ci->i_auth_cap == cap && (newcaps & CEPH_CAP_ANY_FILE_WR)) {
3662 		if (max_size != ci->i_max_size) {
3663 			doutc(cl, "max_size %lld -> %llu\n", ci->i_max_size,
3664 			      max_size);
3665 			ci->i_max_size = max_size;
3666 			if (max_size >= ci->i_wanted_max_size) {
3667 				ci->i_wanted_max_size = 0;  /* reset */
3668 				ci->i_requested_max_size = 0;
3669 			}
3670 			wake = true;
3671 		}
3672 	}
3673 
3674 	/* check cap bits */
3675 	wanted = __ceph_caps_wanted(ci);
3676 	used = __ceph_caps_used(ci);
3677 	dirty = __ceph_caps_dirty(ci);
3678 	doutc(cl, " my wanted = %s, used = %s, dirty %s\n",
3679 	      ceph_cap_string(wanted), ceph_cap_string(used),
3680 	      ceph_cap_string(dirty));
3681 
3682 	if ((was_stale || le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT) &&
3683 	    (wanted & ~(cap->mds_wanted | newcaps))) {
3684 		/*
3685 		 * If mds is importing cap, prior cap messages that update
3686 		 * 'wanted' may get dropped by mds (migrate seq mismatch).
3687 		 *
3688 		 * We don't send cap message to update 'wanted' if what we
3689 		 * want are already issued. If mds revokes caps, cap message
3690 		 * that releases caps also tells mds what we want. But if
3691 		 * caps got revoked by mds forcedly (session stale). We may
3692 		 * haven't told mds what we want.
3693 		 */
3694 		check_caps = 1;
3695 	}
3696 
3697 	/* revocation, grant, or no-op? */
3698 	if (cap->issued & ~newcaps) {
3699 		int revoking = cap->issued & ~newcaps;
3700 
3701 		doutc(cl, "revocation: %s -> %s (revoking %s)\n",
3702 		      ceph_cap_string(cap->issued), ceph_cap_string(newcaps),
3703 		      ceph_cap_string(revoking));
3704 		if (S_ISREG(inode->i_mode) &&
3705 		    (revoking & used & CEPH_CAP_FILE_BUFFER)) {
3706 			writeback = true;  /* initiate writeback; will delay ack */
3707 			revoke_wait = true;
3708 		} else if (queue_invalidate &&
3709 			 revoking == CEPH_CAP_FILE_CACHE &&
3710 			 (newcaps & CEPH_CAP_FILE_LAZYIO) == 0) {
3711 			revoke_wait = true; /* do nothing yet, invalidation will be queued */
3712 		} else if (cap == ci->i_auth_cap) {
3713 			check_caps = 1; /* check auth cap only */
3714 		} else {
3715 			check_caps = 2; /* check all caps */
3716 		}
3717 		/* If there is new caps, try to wake up the waiters */
3718 		if (~cap->issued & newcaps)
3719 			wake = true;
3720 		cap->issued = newcaps;
3721 		cap->implemented |= newcaps;
3722 	} else if (cap->issued == newcaps) {
3723 		doutc(cl, "caps unchanged: %s -> %s\n",
3724 		      ceph_cap_string(cap->issued),
3725 		      ceph_cap_string(newcaps));
3726 	} else {
3727 		doutc(cl, "grant: %s -> %s\n", ceph_cap_string(cap->issued),
3728 		      ceph_cap_string(newcaps));
3729 		/* non-auth MDS is revoking the newly grant caps ? */
3730 		if (cap == ci->i_auth_cap &&
3731 		    __ceph_caps_revoking_other(ci, cap, newcaps))
3732 		    check_caps = 2;
3733 
3734 		cap->issued = newcaps;
3735 		cap->implemented |= newcaps; /* add bits only, to
3736 					      * avoid stepping on a
3737 					      * pending revocation */
3738 		wake = true;
3739 	}
3740 	BUG_ON(cap->issued & ~cap->implemented);
3741 
3742 	/* don't let check_caps skip sending a response to MDS for revoke msgs */
3743 	if (!revoke_wait && le32_to_cpu(grant->op) == CEPH_CAP_OP_REVOKE) {
3744 		cap->mds_wanted = 0;
3745 		flags |= CHECK_CAPS_FLUSH_FORCE;
3746 		if (cap == ci->i_auth_cap)
3747 			check_caps = 1; /* check auth cap only */
3748 		else
3749 			check_caps = 2; /* check all caps */
3750 	}
3751 
3752 	if (extra_info->inline_version > 0 &&
3753 	    extra_info->inline_version >= ci->i_inline_version) {
3754 		ci->i_inline_version = extra_info->inline_version;
3755 		if (ci->i_inline_version != CEPH_INLINE_NONE &&
3756 		    (newcaps & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)))
3757 			fill_inline = true;
3758 	}
3759 
3760 	if (le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT) {
3761 		if (ci->i_auth_cap == cap) {
3762 			if (newcaps & ~extra_info->issued)
3763 				wake = true;
3764 
3765 			if (ci->i_requested_max_size > max_size ||
3766 			    !(le32_to_cpu(grant->wanted) & CEPH_CAP_ANY_FILE_WR)) {
3767 				/* re-request max_size if necessary */
3768 				ci->i_requested_max_size = 0;
3769 				wake = true;
3770 			}
3771 
3772 			ceph_kick_flushing_inode_caps(session, ci);
3773 		}
3774 		up_read(&session->s_mdsc->snap_rwsem);
3775 	}
3776 	spin_unlock(&ci->i_ceph_lock);
3777 
3778 	if (fill_inline)
3779 		ceph_fill_inline_data(inode, NULL, extra_info->inline_data,
3780 				      extra_info->inline_len);
3781 
3782 	if (queue_trunc)
3783 		ceph_queue_vmtruncate(inode);
3784 
3785 	if (writeback)
3786 		/*
3787 		 * queue inode for writeback: we can't actually call
3788 		 * filemap_write_and_wait, etc. from message handler
3789 		 * context.
3790 		 */
3791 		ceph_queue_writeback(inode);
3792 	if (queue_invalidate)
3793 		ceph_queue_invalidate(inode);
3794 	if (deleted_inode)
3795 		invalidate_aliases(inode);
3796 	if (wake)
3797 		wake_up_all(&ci->i_cap_wq);
3798 
3799 	mutex_unlock(&session->s_mutex);
3800 	if (check_caps == 1)
3801 		ceph_check_caps(ci, flags | CHECK_CAPS_AUTHONLY | CHECK_CAPS_NOINVAL);
3802 	else if (check_caps == 2)
3803 		ceph_check_caps(ci, flags | CHECK_CAPS_NOINVAL);
3804 }
3805 
3806 /*
3807  * Handle FLUSH_ACK from MDS, indicating that metadata we sent to the
3808  * MDS has been safely committed.
3809  */
handle_cap_flush_ack(struct inode * inode,u64 flush_tid,struct ceph_mds_caps * m,struct ceph_mds_session * session,struct ceph_cap * cap)3810 static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid,
3811 				 struct ceph_mds_caps *m,
3812 				 struct ceph_mds_session *session,
3813 				 struct ceph_cap *cap)
3814 	__releases(ci->i_ceph_lock)
3815 {
3816 	struct ceph_inode_info *ci = ceph_inode(inode);
3817 	struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(inode->i_sb)->mdsc;
3818 	struct ceph_client *cl = mdsc->fsc->client;
3819 	struct ceph_cap_flush *cf, *tmp_cf;
3820 	LIST_HEAD(to_remove);
3821 	unsigned seq = le32_to_cpu(m->seq);
3822 	int dirty = le32_to_cpu(m->dirty);
3823 	int cleaned = 0;
3824 	bool drop = false;
3825 	bool wake_ci = false;
3826 	bool wake_mdsc = false;
3827 
3828 	list_for_each_entry_safe(cf, tmp_cf, &ci->i_cap_flush_list, i_list) {
3829 		/* Is this the one that was flushed? */
3830 		if (cf->tid == flush_tid)
3831 			cleaned = cf->caps;
3832 
3833 		/* Is this a capsnap? */
3834 		if (cf->is_capsnap)
3835 			continue;
3836 
3837 		if (cf->tid <= flush_tid) {
3838 			/*
3839 			 * An earlier or current tid. The FLUSH_ACK should
3840 			 * represent a superset of this flush's caps.
3841 			 */
3842 			wake_ci |= __detach_cap_flush_from_ci(ci, cf);
3843 			list_add_tail(&cf->i_list, &to_remove);
3844 		} else {
3845 			/*
3846 			 * This is a later one. Any caps in it are still dirty
3847 			 * so don't count them as cleaned.
3848 			 */
3849 			cleaned &= ~cf->caps;
3850 			if (!cleaned)
3851 				break;
3852 		}
3853 	}
3854 
3855 	doutc(cl, "%p %llx.%llx mds%d seq %d on %s cleaned %s, flushing %s -> %s\n",
3856 	      inode, ceph_vinop(inode), session->s_mds, seq,
3857 	      ceph_cap_string(dirty), ceph_cap_string(cleaned),
3858 	      ceph_cap_string(ci->i_flushing_caps),
3859 	      ceph_cap_string(ci->i_flushing_caps & ~cleaned));
3860 
3861 	if (list_empty(&to_remove) && !cleaned)
3862 		goto out;
3863 
3864 	ci->i_flushing_caps &= ~cleaned;
3865 
3866 	spin_lock(&mdsc->cap_dirty_lock);
3867 
3868 	list_for_each_entry(cf, &to_remove, i_list)
3869 		wake_mdsc |= __detach_cap_flush_from_mdsc(mdsc, cf);
3870 
3871 	if (ci->i_flushing_caps == 0) {
3872 		if (list_empty(&ci->i_cap_flush_list)) {
3873 			list_del_init(&ci->i_flushing_item);
3874 			if (!list_empty(&session->s_cap_flushing)) {
3875 				struct inode *inode =
3876 					    &list_first_entry(&session->s_cap_flushing,
3877 							      struct ceph_inode_info,
3878 							      i_flushing_item)->netfs.inode;
3879 				doutc(cl, " mds%d still flushing cap on %p %llx.%llx\n",
3880 				      session->s_mds, inode, ceph_vinop(inode));
3881 			}
3882 		}
3883 		mdsc->num_cap_flushing--;
3884 		doutc(cl, " %p %llx.%llx now !flushing\n", inode,
3885 		      ceph_vinop(inode));
3886 
3887 		if (ci->i_dirty_caps == 0) {
3888 			doutc(cl, " %p %llx.%llx now clean\n", inode,
3889 			      ceph_vinop(inode));
3890 			BUG_ON(!list_empty(&ci->i_dirty_item));
3891 			drop = true;
3892 			if (ci->i_wr_ref == 0 &&
3893 			    ci->i_wrbuffer_ref_head == 0) {
3894 				BUG_ON(!ci->i_head_snapc);
3895 				ceph_put_snap_context(ci->i_head_snapc);
3896 				ci->i_head_snapc = NULL;
3897 			}
3898 		} else {
3899 			BUG_ON(list_empty(&ci->i_dirty_item));
3900 		}
3901 	}
3902 	spin_unlock(&mdsc->cap_dirty_lock);
3903 
3904 out:
3905 	spin_unlock(&ci->i_ceph_lock);
3906 
3907 	while (!list_empty(&to_remove)) {
3908 		cf = list_first_entry(&to_remove,
3909 				      struct ceph_cap_flush, i_list);
3910 		list_del_init(&cf->i_list);
3911 		if (!cf->is_capsnap)
3912 			ceph_free_cap_flush(cf);
3913 	}
3914 
3915 	if (wake_ci)
3916 		wake_up_all(&ci->i_cap_wq);
3917 	if (wake_mdsc)
3918 		wake_up_all(&mdsc->cap_flushing_wq);
3919 	if (drop)
3920 		iput(inode);
3921 }
3922 
__ceph_remove_capsnap(struct inode * inode,struct ceph_cap_snap * capsnap,bool * wake_ci,bool * wake_mdsc)3923 void __ceph_remove_capsnap(struct inode *inode, struct ceph_cap_snap *capsnap,
3924 			   bool *wake_ci, bool *wake_mdsc)
3925 {
3926 	struct ceph_inode_info *ci = ceph_inode(inode);
3927 	struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(inode->i_sb)->mdsc;
3928 	struct ceph_client *cl = mdsc->fsc->client;
3929 	bool ret;
3930 
3931 	lockdep_assert_held(&ci->i_ceph_lock);
3932 
3933 	doutc(cl, "removing capsnap %p, %p %llx.%llx ci %p\n", capsnap,
3934 	      inode, ceph_vinop(inode), ci);
3935 
3936 	list_del_init(&capsnap->ci_item);
3937 	ret = __detach_cap_flush_from_ci(ci, &capsnap->cap_flush);
3938 	if (wake_ci)
3939 		*wake_ci = ret;
3940 
3941 	spin_lock(&mdsc->cap_dirty_lock);
3942 	if (list_empty(&ci->i_cap_flush_list))
3943 		list_del_init(&ci->i_flushing_item);
3944 
3945 	ret = __detach_cap_flush_from_mdsc(mdsc, &capsnap->cap_flush);
3946 	if (wake_mdsc)
3947 		*wake_mdsc = ret;
3948 	spin_unlock(&mdsc->cap_dirty_lock);
3949 }
3950 
ceph_remove_capsnap(struct inode * inode,struct ceph_cap_snap * capsnap,bool * wake_ci,bool * wake_mdsc)3951 void ceph_remove_capsnap(struct inode *inode, struct ceph_cap_snap *capsnap,
3952 			 bool *wake_ci, bool *wake_mdsc)
3953 {
3954 	struct ceph_inode_info *ci = ceph_inode(inode);
3955 
3956 	lockdep_assert_held(&ci->i_ceph_lock);
3957 
3958 	WARN_ON_ONCE(capsnap->dirty_pages || capsnap->writing);
3959 	__ceph_remove_capsnap(inode, capsnap, wake_ci, wake_mdsc);
3960 }
3961 
3962 /*
3963  * Handle FLUSHSNAP_ACK.  MDS has flushed snap data to disk and we can
3964  * throw away our cap_snap.
3965  *
3966  * Caller hold s_mutex.
3967  */
handle_cap_flushsnap_ack(struct inode * inode,u64 flush_tid,struct ceph_mds_caps * m,struct ceph_mds_session * session)3968 static void handle_cap_flushsnap_ack(struct inode *inode, u64 flush_tid,
3969 				     struct ceph_mds_caps *m,
3970 				     struct ceph_mds_session *session)
3971 {
3972 	struct ceph_inode_info *ci = ceph_inode(inode);
3973 	struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(inode->i_sb)->mdsc;
3974 	struct ceph_client *cl = mdsc->fsc->client;
3975 	u64 follows = le64_to_cpu(m->snap_follows);
3976 	struct ceph_cap_snap *capsnap = NULL, *iter;
3977 	bool wake_ci = false;
3978 	bool wake_mdsc = false;
3979 
3980 	doutc(cl, "%p %llx.%llx ci %p mds%d follows %lld\n", inode,
3981 	      ceph_vinop(inode), ci, session->s_mds, follows);
3982 
3983 	spin_lock(&ci->i_ceph_lock);
3984 	list_for_each_entry(iter, &ci->i_cap_snaps, ci_item) {
3985 		if (iter->follows == follows) {
3986 			if (iter->cap_flush.tid != flush_tid) {
3987 				doutc(cl, " cap_snap %p follows %lld "
3988 				      "tid %lld != %lld\n", iter,
3989 				      follows, flush_tid,
3990 				      iter->cap_flush.tid);
3991 				break;
3992 			}
3993 			capsnap = iter;
3994 			break;
3995 		} else {
3996 			doutc(cl, " skipping cap_snap %p follows %lld\n",
3997 			      iter, iter->follows);
3998 		}
3999 	}
4000 	if (capsnap)
4001 		ceph_remove_capsnap(inode, capsnap, &wake_ci, &wake_mdsc);
4002 	spin_unlock(&ci->i_ceph_lock);
4003 
4004 	if (capsnap) {
4005 		ceph_put_snap_context(capsnap->context);
4006 		ceph_put_cap_snap(capsnap);
4007 		if (wake_ci)
4008 			wake_up_all(&ci->i_cap_wq);
4009 		if (wake_mdsc)
4010 			wake_up_all(&mdsc->cap_flushing_wq);
4011 		iput(inode);
4012 	}
4013 }
4014 
4015 /*
4016  * Handle TRUNC from MDS, indicating file truncation.
4017  *
4018  * caller hold s_mutex.
4019  */
handle_cap_trunc(struct inode * inode,struct ceph_mds_caps * trunc,struct ceph_mds_session * session,struct cap_extra_info * extra_info)4020 static bool handle_cap_trunc(struct inode *inode,
4021 			     struct ceph_mds_caps *trunc,
4022 			     struct ceph_mds_session *session,
4023 			     struct cap_extra_info *extra_info)
4024 {
4025 	struct ceph_inode_info *ci = ceph_inode(inode);
4026 	struct ceph_client *cl = ceph_inode_to_client(inode);
4027 	int mds = session->s_mds;
4028 	int seq = le32_to_cpu(trunc->seq);
4029 	u32 truncate_seq = le32_to_cpu(trunc->truncate_seq);
4030 	u64 truncate_size = le64_to_cpu(trunc->truncate_size);
4031 	u64 size = le64_to_cpu(trunc->size);
4032 	int implemented = 0;
4033 	int dirty = __ceph_caps_dirty(ci);
4034 	int issued = __ceph_caps_issued(ceph_inode(inode), &implemented);
4035 	bool queue_trunc = false;
4036 
4037 	lockdep_assert_held(&ci->i_ceph_lock);
4038 
4039 	issued |= implemented | dirty;
4040 
4041 	/*
4042 	 * If there is at least one crypto block then we'll trust
4043 	 * fscrypt_file_size. If the real length of the file is 0, then
4044 	 * ignore it (it has probably been truncated down to 0 by the MDS).
4045 	 */
4046 	if (IS_ENCRYPTED(inode) && size)
4047 		size = extra_info->fscrypt_file_size;
4048 
4049 	doutc(cl, "%p %llx.%llx mds%d seq %d to %lld truncate seq %d\n",
4050 	      inode, ceph_vinop(inode), mds, seq, truncate_size, truncate_seq);
4051 	queue_trunc = ceph_fill_file_size(inode, issued,
4052 					  truncate_seq, truncate_size, size);
4053 	return queue_trunc;
4054 }
4055 
4056 /*
4057  * Handle EXPORT from MDS.  Cap is being migrated _from_ this mds to a
4058  * different one.  If we are the most recent migration we've seen (as
4059  * indicated by mseq), make note of the migrating cap bits for the
4060  * duration (until we see the corresponding IMPORT).
4061  *
4062  * caller holds s_mutex
4063  */
handle_cap_export(struct inode * inode,struct ceph_mds_caps * ex,struct ceph_mds_cap_peer * ph,struct ceph_mds_session * session)4064 static void handle_cap_export(struct inode *inode, struct ceph_mds_caps *ex,
4065 			      struct ceph_mds_cap_peer *ph,
4066 			      struct ceph_mds_session *session)
4067 {
4068 	struct ceph_mds_client *mdsc = ceph_inode_to_fs_client(inode)->mdsc;
4069 	struct ceph_client *cl = mdsc->fsc->client;
4070 	struct ceph_mds_session *tsession = NULL;
4071 	struct ceph_cap *cap, *tcap, *new_cap = NULL;
4072 	struct ceph_inode_info *ci = ceph_inode(inode);
4073 	u64 t_cap_id;
4074 	u32 t_issue_seq, t_mseq;
4075 	int target, issued;
4076 	int mds = session->s_mds;
4077 
4078 	if (ph) {
4079 		t_cap_id = le64_to_cpu(ph->cap_id);
4080 		t_issue_seq = le32_to_cpu(ph->issue_seq);
4081 		t_mseq = le32_to_cpu(ph->mseq);
4082 		target = le32_to_cpu(ph->mds);
4083 	} else {
4084 		t_cap_id = t_issue_seq = t_mseq = 0;
4085 		target = -1;
4086 	}
4087 
4088 	doutc(cl, " cap %llx.%llx export to peer %d piseq %u pmseq %u\n",
4089 	      ceph_vinop(inode), target, t_issue_seq, t_mseq);
4090 retry:
4091 	down_read(&mdsc->snap_rwsem);
4092 	spin_lock(&ci->i_ceph_lock);
4093 	cap = __get_cap_for_mds(ci, mds);
4094 	if (!cap || cap->cap_id != le64_to_cpu(ex->cap_id))
4095 		goto out_unlock;
4096 
4097 	if (target < 0) {
4098 		ceph_remove_cap(mdsc, cap, false);
4099 		goto out_unlock;
4100 	}
4101 
4102 	/*
4103 	 * now we know we haven't received the cap import message yet
4104 	 * because the exported cap still exist.
4105 	 */
4106 
4107 	issued = cap->issued;
4108 	if (issued != cap->implemented)
4109 		pr_err_ratelimited_client(cl, "issued != implemented: "
4110 					  "%p %llx.%llx mds%d seq %d mseq %d"
4111 					  " issued %s implemented %s\n",
4112 					  inode, ceph_vinop(inode), mds,
4113 					  cap->seq, cap->mseq,
4114 					  ceph_cap_string(issued),
4115 					  ceph_cap_string(cap->implemented));
4116 
4117 
4118 	tcap = __get_cap_for_mds(ci, target);
4119 	if (tcap) {
4120 		/* already have caps from the target */
4121 		if (tcap->cap_id == t_cap_id &&
4122 		    ceph_seq_cmp(tcap->seq, t_issue_seq) < 0) {
4123 			doutc(cl, " updating import cap %p mds%d\n", tcap,
4124 			      target);
4125 			tcap->cap_id = t_cap_id;
4126 			tcap->seq = t_issue_seq - 1;
4127 			tcap->issue_seq = t_issue_seq - 1;
4128 			tcap->issued |= issued;
4129 			tcap->implemented |= issued;
4130 			if (cap == ci->i_auth_cap) {
4131 				ci->i_auth_cap = tcap;
4132 				change_auth_cap_ses(ci, tcap->session);
4133 			}
4134 		}
4135 		ceph_remove_cap(mdsc, cap, false);
4136 		goto out_unlock;
4137 	} else if (tsession) {
4138 		/* add placeholder for the export target */
4139 		int flag = (cap == ci->i_auth_cap) ? CEPH_CAP_FLAG_AUTH : 0;
4140 		tcap = new_cap;
4141 		ceph_add_cap(inode, tsession, t_cap_id, issued, 0,
4142 			     t_issue_seq - 1, t_mseq, (u64)-1, flag, &new_cap);
4143 
4144 		if (!list_empty(&ci->i_cap_flush_list) &&
4145 		    ci->i_auth_cap == tcap) {
4146 			spin_lock(&mdsc->cap_dirty_lock);
4147 			list_move_tail(&ci->i_flushing_item,
4148 				       &tcap->session->s_cap_flushing);
4149 			spin_unlock(&mdsc->cap_dirty_lock);
4150 		}
4151 
4152 		ceph_remove_cap(mdsc, cap, false);
4153 		goto out_unlock;
4154 	}
4155 
4156 	spin_unlock(&ci->i_ceph_lock);
4157 	up_read(&mdsc->snap_rwsem);
4158 	mutex_unlock(&session->s_mutex);
4159 
4160 	/* open target session */
4161 	tsession = ceph_mdsc_open_export_target_session(mdsc, target);
4162 	if (!IS_ERR(tsession)) {
4163 		if (mds > target) {
4164 			mutex_lock(&session->s_mutex);
4165 			mutex_lock_nested(&tsession->s_mutex,
4166 					  SINGLE_DEPTH_NESTING);
4167 		} else {
4168 			mutex_lock(&tsession->s_mutex);
4169 			mutex_lock_nested(&session->s_mutex,
4170 					  SINGLE_DEPTH_NESTING);
4171 		}
4172 		new_cap = ceph_get_cap(mdsc, NULL);
4173 	} else {
4174 		WARN_ON(1);
4175 		tsession = NULL;
4176 		target = -1;
4177 		mutex_lock(&session->s_mutex);
4178 	}
4179 	goto retry;
4180 
4181 out_unlock:
4182 	spin_unlock(&ci->i_ceph_lock);
4183 	up_read(&mdsc->snap_rwsem);
4184 	mutex_unlock(&session->s_mutex);
4185 	if (tsession) {
4186 		mutex_unlock(&tsession->s_mutex);
4187 		ceph_put_mds_session(tsession);
4188 	}
4189 	if (new_cap)
4190 		ceph_put_cap(mdsc, new_cap);
4191 }
4192 
4193 /*
4194  * Handle cap IMPORT.
4195  *
4196  * caller holds s_mutex. acquires i_ceph_lock
4197  */
handle_cap_import(struct ceph_mds_client * mdsc,struct inode * inode,struct ceph_mds_caps * im,struct ceph_mds_cap_peer * ph,struct ceph_mds_session * session,struct ceph_cap ** target_cap,int * old_issued)4198 static void handle_cap_import(struct ceph_mds_client *mdsc,
4199 			      struct inode *inode, struct ceph_mds_caps *im,
4200 			      struct ceph_mds_cap_peer *ph,
4201 			      struct ceph_mds_session *session,
4202 			      struct ceph_cap **target_cap, int *old_issued)
4203 {
4204 	struct ceph_inode_info *ci = ceph_inode(inode);
4205 	struct ceph_client *cl = mdsc->fsc->client;
4206 	struct ceph_cap *cap, *ocap, *new_cap = NULL;
4207 	int mds = session->s_mds;
4208 	int issued;
4209 	unsigned caps = le32_to_cpu(im->caps);
4210 	unsigned wanted = le32_to_cpu(im->wanted);
4211 	unsigned seq = le32_to_cpu(im->seq);
4212 	unsigned mseq = le32_to_cpu(im->migrate_seq);
4213 	u64 realmino = le64_to_cpu(im->realm);
4214 	u64 cap_id = le64_to_cpu(im->cap_id);
4215 	u64 p_cap_id;
4216 	u32 piseq = 0;
4217 	u32 pmseq = 0;
4218 	int peer;
4219 
4220 	if (ph) {
4221 		p_cap_id = le64_to_cpu(ph->cap_id);
4222 		peer = le32_to_cpu(ph->mds);
4223 		piseq = le32_to_cpu(ph->issue_seq);
4224 		pmseq = le32_to_cpu(ph->mseq);
4225 	} else {
4226 		p_cap_id = 0;
4227 		peer = -1;
4228 	}
4229 
4230 	doutc(cl, " cap %llx.%llx import from peer %d piseq %u pmseq %u\n",
4231 	      ceph_vinop(inode), peer, piseq, pmseq);
4232 retry:
4233 	cap = __get_cap_for_mds(ci, mds);
4234 	if (!cap) {
4235 		if (!new_cap) {
4236 			spin_unlock(&ci->i_ceph_lock);
4237 			new_cap = ceph_get_cap(mdsc, NULL);
4238 			spin_lock(&ci->i_ceph_lock);
4239 			goto retry;
4240 		}
4241 		cap = new_cap;
4242 	} else {
4243 		if (new_cap) {
4244 			ceph_put_cap(mdsc, new_cap);
4245 			new_cap = NULL;
4246 		}
4247 	}
4248 
4249 	__ceph_caps_issued(ci, &issued);
4250 	issued |= __ceph_caps_dirty(ci);
4251 
4252 	ceph_add_cap(inode, session, cap_id, caps, wanted, seq, mseq,
4253 		     realmino, CEPH_CAP_FLAG_AUTH, &new_cap);
4254 
4255 	ocap = peer >= 0 ? __get_cap_for_mds(ci, peer) : NULL;
4256 	if (ocap && ocap->cap_id == p_cap_id) {
4257 		doutc(cl, " remove export cap %p mds%d flags %d\n",
4258 		      ocap, peer, ph->flags);
4259 		if ((ph->flags & CEPH_CAP_FLAG_AUTH) &&
4260 		    (ocap->seq != piseq ||
4261 		     ocap->mseq != pmseq)) {
4262 			pr_err_ratelimited_client(cl, "mismatched seq/mseq: "
4263 					"%p %llx.%llx mds%d seq %d mseq %d"
4264 					" importer mds%d has peer seq %d mseq %d\n",
4265 					inode, ceph_vinop(inode), peer,
4266 					ocap->seq, ocap->mseq, mds, piseq, pmseq);
4267 		}
4268 		ceph_remove_cap(mdsc, ocap, (ph->flags & CEPH_CAP_FLAG_RELEASE));
4269 	}
4270 
4271 	*old_issued = issued;
4272 	*target_cap = cap;
4273 }
4274 
4275 #ifdef CONFIG_FS_ENCRYPTION
parse_fscrypt_fields(void ** p,void * end,struct cap_extra_info * extra)4276 static int parse_fscrypt_fields(void **p, void *end,
4277 				struct cap_extra_info *extra)
4278 {
4279 	u32 len;
4280 
4281 	ceph_decode_32_safe(p, end, extra->fscrypt_auth_len, bad);
4282 	if (extra->fscrypt_auth_len) {
4283 		ceph_decode_need(p, end, extra->fscrypt_auth_len, bad);
4284 		extra->fscrypt_auth = kmalloc(extra->fscrypt_auth_len,
4285 					      GFP_KERNEL);
4286 		if (!extra->fscrypt_auth)
4287 			return -ENOMEM;
4288 		ceph_decode_copy_safe(p, end, extra->fscrypt_auth,
4289 					extra->fscrypt_auth_len, bad);
4290 	}
4291 
4292 	ceph_decode_32_safe(p, end, len, bad);
4293 	if (len >= sizeof(u64)) {
4294 		ceph_decode_64_safe(p, end, extra->fscrypt_file_size, bad);
4295 		len -= sizeof(u64);
4296 	}
4297 	ceph_decode_skip_n(p, end, len, bad);
4298 	return 0;
4299 bad:
4300 	return -EIO;
4301 }
4302 #else
parse_fscrypt_fields(void ** p,void * end,struct cap_extra_info * extra)4303 static int parse_fscrypt_fields(void **p, void *end,
4304 				struct cap_extra_info *extra)
4305 {
4306 	u32 len;
4307 
4308 	/* Don't care about these fields unless we're encryption-capable */
4309 	ceph_decode_32_safe(p, end, len, bad);
4310 	if (len)
4311 		ceph_decode_skip_n(p, end, len, bad);
4312 	ceph_decode_32_safe(p, end, len, bad);
4313 	if (len)
4314 		ceph_decode_skip_n(p, end, len, bad);
4315 	return 0;
4316 bad:
4317 	return -EIO;
4318 }
4319 #endif
4320 
4321 /*
4322  * Handle a caps message from the MDS.
4323  *
4324  * Identify the appropriate session, inode, and call the right handler
4325  * based on the cap op.
4326  */
ceph_handle_caps(struct ceph_mds_session * session,struct ceph_msg * msg)4327 void ceph_handle_caps(struct ceph_mds_session *session,
4328 		      struct ceph_msg *msg)
4329 {
4330 	struct ceph_mds_client *mdsc = session->s_mdsc;
4331 	struct ceph_client *cl = mdsc->fsc->client;
4332 	struct inode *inode;
4333 	struct ceph_inode_info *ci;
4334 	struct ceph_cap *cap;
4335 	struct ceph_mds_caps *h;
4336 	struct ceph_mds_cap_peer *peer = NULL;
4337 	struct ceph_snap_realm *realm = NULL;
4338 	int op;
4339 	int msg_version = le16_to_cpu(msg->hdr.version);
4340 	u32 seq, mseq, issue_seq;
4341 	struct ceph_vino vino;
4342 	void *snaptrace;
4343 	size_t snaptrace_len;
4344 	void *p, *end;
4345 	struct cap_extra_info extra_info = {};
4346 	bool queue_trunc;
4347 	bool close_sessions = false;
4348 	bool do_cap_release = false;
4349 
4350 	if (!ceph_inc_mds_stopping_blocker(mdsc, session))
4351 		return;
4352 
4353 	/* decode */
4354 	end = msg->front.iov_base + msg->front.iov_len;
4355 	if (msg->front.iov_len < sizeof(*h))
4356 		goto bad;
4357 	h = msg->front.iov_base;
4358 	op = le32_to_cpu(h->op);
4359 	vino.ino = le64_to_cpu(h->ino);
4360 	vino.snap = CEPH_NOSNAP;
4361 	seq = le32_to_cpu(h->seq);
4362 	mseq = le32_to_cpu(h->migrate_seq);
4363 	issue_seq = le32_to_cpu(h->issue_seq);
4364 
4365 	snaptrace = h + 1;
4366 	snaptrace_len = le32_to_cpu(h->snap_trace_len);
4367 	p = snaptrace + snaptrace_len;
4368 
4369 	if (msg_version >= 2) {
4370 		u32 flock_len;
4371 		ceph_decode_32_safe(&p, end, flock_len, bad);
4372 		if (p + flock_len > end)
4373 			goto bad;
4374 		p += flock_len;
4375 	}
4376 
4377 	if (msg_version >= 3) {
4378 		if (op == CEPH_CAP_OP_IMPORT) {
4379 			if (p + sizeof(*peer) > end)
4380 				goto bad;
4381 			peer = p;
4382 			p += sizeof(*peer);
4383 		} else if (op == CEPH_CAP_OP_EXPORT) {
4384 			/* recorded in unused fields */
4385 			peer = (void *)&h->size;
4386 		}
4387 	}
4388 
4389 	if (msg_version >= 4) {
4390 		ceph_decode_64_safe(&p, end, extra_info.inline_version, bad);
4391 		ceph_decode_32_safe(&p, end, extra_info.inline_len, bad);
4392 		if (p + extra_info.inline_len > end)
4393 			goto bad;
4394 		extra_info.inline_data = p;
4395 		p += extra_info.inline_len;
4396 	}
4397 
4398 	if (msg_version >= 5) {
4399 		struct ceph_osd_client	*osdc = &mdsc->fsc->client->osdc;
4400 		u32			epoch_barrier;
4401 
4402 		ceph_decode_32_safe(&p, end, epoch_barrier, bad);
4403 		ceph_osdc_update_epoch_barrier(osdc, epoch_barrier);
4404 	}
4405 
4406 	if (msg_version >= 8) {
4407 		u32 pool_ns_len;
4408 
4409 		/* version >= 6 */
4410 		ceph_decode_skip_64(&p, end, bad);	// flush_tid
4411 		/* version >= 7 */
4412 		ceph_decode_skip_32(&p, end, bad);	// caller_uid
4413 		ceph_decode_skip_32(&p, end, bad);	// caller_gid
4414 		/* version >= 8 */
4415 		ceph_decode_32_safe(&p, end, pool_ns_len, bad);
4416 		if (pool_ns_len > 0) {
4417 			ceph_decode_need(&p, end, pool_ns_len, bad);
4418 			extra_info.pool_ns =
4419 				ceph_find_or_create_string(p, pool_ns_len);
4420 			p += pool_ns_len;
4421 		}
4422 	}
4423 
4424 	if (msg_version >= 9) {
4425 		struct ceph_timespec *btime;
4426 
4427 		if (p + sizeof(*btime) > end)
4428 			goto bad;
4429 		btime = p;
4430 		ceph_decode_timespec64(&extra_info.btime, btime);
4431 		p += sizeof(*btime);
4432 		ceph_decode_64_safe(&p, end, extra_info.change_attr, bad);
4433 	}
4434 
4435 	if (msg_version >= 11) {
4436 		/* version >= 10 */
4437 		ceph_decode_skip_32(&p, end, bad); // flags
4438 		/* version >= 11 */
4439 		extra_info.dirstat_valid = true;
4440 		ceph_decode_64_safe(&p, end, extra_info.nfiles, bad);
4441 		ceph_decode_64_safe(&p, end, extra_info.nsubdirs, bad);
4442 	}
4443 
4444 	if (msg_version >= 12) {
4445 		if (parse_fscrypt_fields(&p, end, &extra_info))
4446 			goto bad;
4447 	}
4448 
4449 	/* lookup ino */
4450 	inode = ceph_find_inode(mdsc->fsc->sb, vino);
4451 	doutc(cl, " caps mds%d op %s ino %llx.%llx inode %p seq %u iseq %u mseq %u\n",
4452 	      session->s_mds, ceph_cap_op_name(op), vino.ino, vino.snap, inode,
4453 	      seq, issue_seq, mseq);
4454 
4455 	mutex_lock(&session->s_mutex);
4456 
4457 	if (!inode) {
4458 		doutc(cl, " i don't have ino %llx\n", vino.ino);
4459 
4460 		switch (op) {
4461 		case CEPH_CAP_OP_IMPORT:
4462 		case CEPH_CAP_OP_REVOKE:
4463 		case CEPH_CAP_OP_GRANT:
4464 			do_cap_release = true;
4465 			break;
4466 		default:
4467 			break;
4468 		}
4469 		goto flush_cap_releases;
4470 	}
4471 	ci = ceph_inode(inode);
4472 
4473 	/* these will work even if we don't have a cap yet */
4474 	switch (op) {
4475 	case CEPH_CAP_OP_FLUSHSNAP_ACK:
4476 		handle_cap_flushsnap_ack(inode, le64_to_cpu(msg->hdr.tid),
4477 					 h, session);
4478 		goto done;
4479 
4480 	case CEPH_CAP_OP_EXPORT:
4481 		handle_cap_export(inode, h, peer, session);
4482 		goto done_unlocked;
4483 
4484 	case CEPH_CAP_OP_IMPORT:
4485 		realm = NULL;
4486 		if (snaptrace_len) {
4487 			down_write(&mdsc->snap_rwsem);
4488 			if (ceph_update_snap_trace(mdsc, snaptrace,
4489 						   snaptrace + snaptrace_len,
4490 						   false, &realm)) {
4491 				up_write(&mdsc->snap_rwsem);
4492 				close_sessions = true;
4493 				goto done;
4494 			}
4495 			downgrade_write(&mdsc->snap_rwsem);
4496 		} else {
4497 			down_read(&mdsc->snap_rwsem);
4498 		}
4499 		spin_lock(&ci->i_ceph_lock);
4500 		handle_cap_import(mdsc, inode, h, peer, session,
4501 				  &cap, &extra_info.issued);
4502 		handle_cap_grant(inode, session, cap,
4503 				 h, msg->middle, &extra_info);
4504 		if (realm)
4505 			ceph_put_snap_realm(mdsc, realm);
4506 		goto done_unlocked;
4507 	}
4508 
4509 	/* the rest require a cap */
4510 	spin_lock(&ci->i_ceph_lock);
4511 	cap = __get_cap_for_mds(ceph_inode(inode), session->s_mds);
4512 	if (!cap) {
4513 		doutc(cl, " no cap on %p ino %llx.%llx from mds%d\n",
4514 		      inode, ceph_ino(inode), ceph_snap(inode),
4515 		      session->s_mds);
4516 		spin_unlock(&ci->i_ceph_lock);
4517 		switch (op) {
4518 		case CEPH_CAP_OP_REVOKE:
4519 		case CEPH_CAP_OP_GRANT:
4520 			do_cap_release = true;
4521 			break;
4522 		default:
4523 			break;
4524 		}
4525 		goto flush_cap_releases;
4526 	}
4527 
4528 	/* note that each of these drops i_ceph_lock for us */
4529 	switch (op) {
4530 	case CEPH_CAP_OP_REVOKE:
4531 	case CEPH_CAP_OP_GRANT:
4532 		__ceph_caps_issued(ci, &extra_info.issued);
4533 		extra_info.issued |= __ceph_caps_dirty(ci);
4534 		handle_cap_grant(inode, session, cap,
4535 				 h, msg->middle, &extra_info);
4536 		goto done_unlocked;
4537 
4538 	case CEPH_CAP_OP_FLUSH_ACK:
4539 		handle_cap_flush_ack(inode, le64_to_cpu(msg->hdr.tid),
4540 				     h, session, cap);
4541 		break;
4542 
4543 	case CEPH_CAP_OP_TRUNC:
4544 		queue_trunc = handle_cap_trunc(inode, h, session,
4545 						&extra_info);
4546 		spin_unlock(&ci->i_ceph_lock);
4547 		if (queue_trunc)
4548 			ceph_queue_vmtruncate(inode);
4549 		break;
4550 
4551 	default:
4552 		spin_unlock(&ci->i_ceph_lock);
4553 		pr_err_client(cl, "unknown cap op %d %s\n", op,
4554 			      ceph_cap_op_name(op));
4555 	}
4556 
4557 done:
4558 	mutex_unlock(&session->s_mutex);
4559 done_unlocked:
4560 	iput(inode);
4561 out:
4562 	ceph_dec_mds_stopping_blocker(mdsc);
4563 
4564 	ceph_put_string(extra_info.pool_ns);
4565 
4566 	/* Defer closing the sessions after s_mutex lock being released */
4567 	if (close_sessions)
4568 		ceph_mdsc_close_sessions(mdsc);
4569 
4570 	kfree(extra_info.fscrypt_auth);
4571 	return;
4572 
4573 flush_cap_releases:
4574 	/*
4575 	 * send any cap release message to try to move things
4576 	 * along for the mds (who clearly thinks we still have this
4577 	 * cap).
4578 	 */
4579 	if (do_cap_release) {
4580 		cap = ceph_get_cap(mdsc, NULL);
4581 		cap->cap_ino = vino.ino;
4582 		cap->queue_release = 1;
4583 		cap->cap_id = le64_to_cpu(h->cap_id);
4584 		cap->mseq = mseq;
4585 		cap->seq = seq;
4586 		cap->issue_seq = seq;
4587 		spin_lock(&session->s_cap_lock);
4588 		__ceph_queue_cap_release(session, cap);
4589 		spin_unlock(&session->s_cap_lock);
4590 	}
4591 	ceph_flush_session_cap_releases(mdsc, session);
4592 	goto done;
4593 
4594 bad:
4595 	pr_err_client(cl, "corrupt message\n");
4596 	ceph_msg_dump(msg);
4597 	goto out;
4598 }
4599 
4600 /*
4601  * Delayed work handler to process end of delayed cap release LRU list.
4602  *
4603  * If new caps are added to the list while processing it, these won't get
4604  * processed in this run.  In this case, the ci->i_hold_caps_max will be
4605  * returned so that the work can be scheduled accordingly.
4606  */
ceph_check_delayed_caps(struct ceph_mds_client * mdsc)4607 unsigned long ceph_check_delayed_caps(struct ceph_mds_client *mdsc)
4608 {
4609 	struct ceph_client *cl = mdsc->fsc->client;
4610 	struct inode *inode;
4611 	struct ceph_inode_info *ci;
4612 	struct ceph_mount_options *opt = mdsc->fsc->mount_options;
4613 	unsigned long delay_max = opt->caps_wanted_delay_max * HZ;
4614 	unsigned long loop_start = jiffies;
4615 	unsigned long delay = 0;
4616 
4617 	doutc(cl, "begin\n");
4618 	spin_lock(&mdsc->cap_delay_lock);
4619 	while (!list_empty(&mdsc->cap_delay_list)) {
4620 		ci = list_first_entry(&mdsc->cap_delay_list,
4621 				      struct ceph_inode_info,
4622 				      i_cap_delay_list);
4623 		if (time_before(loop_start, ci->i_hold_caps_max - delay_max)) {
4624 			doutc(cl, "caps added recently.  Exiting loop");
4625 			delay = ci->i_hold_caps_max;
4626 			break;
4627 		}
4628 		if ((ci->i_ceph_flags & CEPH_I_FLUSH) == 0 &&
4629 		    time_before(jiffies, ci->i_hold_caps_max))
4630 			break;
4631 		list_del_init(&ci->i_cap_delay_list);
4632 
4633 		inode = igrab(&ci->netfs.inode);
4634 		if (inode) {
4635 			spin_unlock(&mdsc->cap_delay_lock);
4636 			doutc(cl, "on %p %llx.%llx\n", inode,
4637 			      ceph_vinop(inode));
4638 			ceph_check_caps(ci, 0);
4639 			iput(inode);
4640 			spin_lock(&mdsc->cap_delay_lock);
4641 		}
4642 
4643 		/*
4644 		 * Make sure too many dirty caps or general
4645 		 * slowness doesn't block mdsc delayed work,
4646 		 * preventing send_renew_caps() from running.
4647 		 */
4648 		if (time_after_eq(jiffies, loop_start + 5 * HZ))
4649 			break;
4650 	}
4651 	spin_unlock(&mdsc->cap_delay_lock);
4652 	doutc(cl, "done\n");
4653 
4654 	return delay;
4655 }
4656 
4657 /*
4658  * Flush all dirty caps to the mds
4659  */
flush_dirty_session_caps(struct ceph_mds_session * s)4660 static void flush_dirty_session_caps(struct ceph_mds_session *s)
4661 {
4662 	struct ceph_mds_client *mdsc = s->s_mdsc;
4663 	struct ceph_client *cl = mdsc->fsc->client;
4664 	struct ceph_inode_info *ci;
4665 	struct inode *inode;
4666 
4667 	doutc(cl, "begin\n");
4668 	spin_lock(&mdsc->cap_dirty_lock);
4669 	while (!list_empty(&s->s_cap_dirty)) {
4670 		ci = list_first_entry(&s->s_cap_dirty, struct ceph_inode_info,
4671 				      i_dirty_item);
4672 		inode = &ci->netfs.inode;
4673 		ihold(inode);
4674 		doutc(cl, "%p %llx.%llx\n", inode, ceph_vinop(inode));
4675 		spin_unlock(&mdsc->cap_dirty_lock);
4676 		ceph_wait_on_async_create(inode);
4677 		ceph_check_caps(ci, CHECK_CAPS_FLUSH);
4678 		iput(inode);
4679 		spin_lock(&mdsc->cap_dirty_lock);
4680 	}
4681 	spin_unlock(&mdsc->cap_dirty_lock);
4682 	doutc(cl, "done\n");
4683 }
4684 
ceph_flush_dirty_caps(struct ceph_mds_client * mdsc)4685 void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc)
4686 {
4687 	ceph_mdsc_iterate_sessions(mdsc, flush_dirty_session_caps, true);
4688 }
4689 
4690 /*
4691  * Flush all cap releases to the mds
4692  */
flush_cap_releases(struct ceph_mds_session * s)4693 static void flush_cap_releases(struct ceph_mds_session *s)
4694 {
4695 	struct ceph_mds_client *mdsc = s->s_mdsc;
4696 	struct ceph_client *cl = mdsc->fsc->client;
4697 
4698 	doutc(cl, "begin\n");
4699 	spin_lock(&s->s_cap_lock);
4700 	if (s->s_num_cap_releases)
4701 		ceph_flush_session_cap_releases(mdsc, s);
4702 	spin_unlock(&s->s_cap_lock);
4703 	doutc(cl, "done\n");
4704 
4705 }
4706 
ceph_flush_cap_releases(struct ceph_mds_client * mdsc)4707 void ceph_flush_cap_releases(struct ceph_mds_client *mdsc)
4708 {
4709 	ceph_mdsc_iterate_sessions(mdsc, flush_cap_releases, true);
4710 }
4711 
__ceph_touch_fmode(struct ceph_inode_info * ci,struct ceph_mds_client * mdsc,int fmode)4712 void __ceph_touch_fmode(struct ceph_inode_info *ci,
4713 			struct ceph_mds_client *mdsc, int fmode)
4714 {
4715 	unsigned long now = jiffies;
4716 	if (fmode & CEPH_FILE_MODE_RD)
4717 		ci->i_last_rd = now;
4718 	if (fmode & CEPH_FILE_MODE_WR)
4719 		ci->i_last_wr = now;
4720 	/* queue periodic check */
4721 	if (fmode &&
4722 	    __ceph_is_any_real_caps(ci) &&
4723 	    list_empty(&ci->i_cap_delay_list))
4724 		__cap_delay_requeue(mdsc, ci);
4725 }
4726 
ceph_get_fmode(struct ceph_inode_info * ci,int fmode,int count)4727 void ceph_get_fmode(struct ceph_inode_info *ci, int fmode, int count)
4728 {
4729 	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(ci->netfs.inode.i_sb);
4730 	int bits = (fmode << 1) | 1;
4731 	bool already_opened = false;
4732 	int i;
4733 
4734 	if (count == 1)
4735 		atomic64_inc(&mdsc->metric.opened_files);
4736 
4737 	spin_lock(&ci->i_ceph_lock);
4738 	for (i = 0; i < CEPH_FILE_MODE_BITS; i++) {
4739 		/*
4740 		 * If any of the mode ref is larger than 0,
4741 		 * that means it has been already opened by
4742 		 * others. Just skip checking the PIN ref.
4743 		 */
4744 		if (i && ci->i_nr_by_mode[i])
4745 			already_opened = true;
4746 
4747 		if (bits & (1 << i))
4748 			ci->i_nr_by_mode[i] += count;
4749 	}
4750 
4751 	if (!already_opened)
4752 		percpu_counter_inc(&mdsc->metric.opened_inodes);
4753 	spin_unlock(&ci->i_ceph_lock);
4754 }
4755 
4756 /*
4757  * Drop open file reference.  If we were the last open file,
4758  * we may need to release capabilities to the MDS (or schedule
4759  * their delayed release).
4760  */
ceph_put_fmode(struct ceph_inode_info * ci,int fmode,int count)4761 void ceph_put_fmode(struct ceph_inode_info *ci, int fmode, int count)
4762 {
4763 	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(ci->netfs.inode.i_sb);
4764 	int bits = (fmode << 1) | 1;
4765 	bool is_closed = true;
4766 	int i;
4767 
4768 	if (count == 1)
4769 		atomic64_dec(&mdsc->metric.opened_files);
4770 
4771 	spin_lock(&ci->i_ceph_lock);
4772 	for (i = 0; i < CEPH_FILE_MODE_BITS; i++) {
4773 		if (bits & (1 << i)) {
4774 			BUG_ON(ci->i_nr_by_mode[i] < count);
4775 			ci->i_nr_by_mode[i] -= count;
4776 		}
4777 
4778 		/*
4779 		 * If any of the mode ref is not 0 after
4780 		 * decreased, that means it is still opened
4781 		 * by others. Just skip checking the PIN ref.
4782 		 */
4783 		if (i && ci->i_nr_by_mode[i])
4784 			is_closed = false;
4785 	}
4786 
4787 	if (is_closed)
4788 		percpu_counter_dec(&mdsc->metric.opened_inodes);
4789 	spin_unlock(&ci->i_ceph_lock);
4790 }
4791 
4792 /*
4793  * For a soon-to-be unlinked file, drop the LINK caps. If it
4794  * looks like the link count will hit 0, drop any other caps (other
4795  * than PIN) we don't specifically want (due to the file still being
4796  * open).
4797  */
ceph_drop_caps_for_unlink(struct inode * inode)4798 int ceph_drop_caps_for_unlink(struct inode *inode)
4799 {
4800 	struct ceph_inode_info *ci = ceph_inode(inode);
4801 	int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
4802 
4803 	spin_lock(&ci->i_ceph_lock);
4804 	if (inode->i_nlink == 1) {
4805 		drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
4806 
4807 		if (__ceph_caps_dirty(ci)) {
4808 			struct ceph_mds_client *mdsc =
4809 				ceph_inode_to_fs_client(inode)->mdsc;
4810 
4811 			doutc(mdsc->fsc->client, "%p %llx.%llx\n", inode,
4812 			      ceph_vinop(inode));
4813 			spin_lock(&mdsc->cap_delay_lock);
4814 			ci->i_ceph_flags |= CEPH_I_FLUSH;
4815 			if (!list_empty(&ci->i_cap_delay_list))
4816 				list_del_init(&ci->i_cap_delay_list);
4817 			list_add_tail(&ci->i_cap_delay_list,
4818 				      &mdsc->cap_unlink_delay_list);
4819 			spin_unlock(&mdsc->cap_delay_lock);
4820 
4821 			/*
4822 			 * Fire the work immediately, because the MDS maybe
4823 			 * waiting for caps release.
4824 			 */
4825 			ceph_queue_cap_unlink_work(mdsc);
4826 		}
4827 	}
4828 	spin_unlock(&ci->i_ceph_lock);
4829 	return drop;
4830 }
4831 
4832 /*
4833  * Helpers for embedding cap and dentry lease releases into mds
4834  * requests.
4835  *
4836  * @force is used by dentry_release (below) to force inclusion of a
4837  * record for the directory inode, even when there aren't any caps to
4838  * drop.
4839  */
ceph_encode_inode_release(void ** p,struct inode * inode,int mds,int drop,int unless,int force)4840 int ceph_encode_inode_release(void **p, struct inode *inode,
4841 			      int mds, int drop, int unless, int force)
4842 {
4843 	struct ceph_inode_info *ci = ceph_inode(inode);
4844 	struct ceph_client *cl = ceph_inode_to_client(inode);
4845 	struct ceph_cap *cap;
4846 	struct ceph_mds_request_release *rel = *p;
4847 	int used, dirty;
4848 	int ret = 0;
4849 
4850 	spin_lock(&ci->i_ceph_lock);
4851 	used = __ceph_caps_used(ci);
4852 	dirty = __ceph_caps_dirty(ci);
4853 
4854 	doutc(cl, "%p %llx.%llx mds%d used|dirty %s drop %s unless %s\n",
4855 	      inode, ceph_vinop(inode), mds, ceph_cap_string(used|dirty),
4856 	      ceph_cap_string(drop), ceph_cap_string(unless));
4857 
4858 	/* only drop unused, clean caps */
4859 	drop &= ~(used | dirty);
4860 
4861 	cap = __get_cap_for_mds(ci, mds);
4862 	if (cap && __cap_is_valid(cap)) {
4863 		unless &= cap->issued;
4864 		if (unless) {
4865 			if (unless & CEPH_CAP_AUTH_EXCL)
4866 				drop &= ~CEPH_CAP_AUTH_SHARED;
4867 			if (unless & CEPH_CAP_LINK_EXCL)
4868 				drop &= ~CEPH_CAP_LINK_SHARED;
4869 			if (unless & CEPH_CAP_XATTR_EXCL)
4870 				drop &= ~CEPH_CAP_XATTR_SHARED;
4871 			if (unless & CEPH_CAP_FILE_EXCL)
4872 				drop &= ~CEPH_CAP_FILE_SHARED;
4873 		}
4874 
4875 		if (force || (cap->issued & drop)) {
4876 			if (cap->issued & drop) {
4877 				int wanted = __ceph_caps_wanted(ci);
4878 				doutc(cl, "%p %llx.%llx cap %p %s -> %s, "
4879 				      "wanted %s -> %s\n", inode,
4880 				      ceph_vinop(inode), cap,
4881 				      ceph_cap_string(cap->issued),
4882 				      ceph_cap_string(cap->issued & ~drop),
4883 				      ceph_cap_string(cap->mds_wanted),
4884 				      ceph_cap_string(wanted));
4885 
4886 				cap->issued &= ~drop;
4887 				cap->implemented &= ~drop;
4888 				cap->mds_wanted = wanted;
4889 				if (cap == ci->i_auth_cap &&
4890 				    !(wanted & CEPH_CAP_ANY_FILE_WR))
4891 					ci->i_requested_max_size = 0;
4892 			} else {
4893 				doutc(cl, "%p %llx.%llx cap %p %s (force)\n",
4894 				      inode, ceph_vinop(inode), cap,
4895 				      ceph_cap_string(cap->issued));
4896 			}
4897 
4898 			rel->ino = cpu_to_le64(ceph_ino(inode));
4899 			rel->cap_id = cpu_to_le64(cap->cap_id);
4900 			rel->seq = cpu_to_le32(cap->seq);
4901 			rel->issue_seq = cpu_to_le32(cap->issue_seq);
4902 			rel->mseq = cpu_to_le32(cap->mseq);
4903 			rel->caps = cpu_to_le32(cap->implemented);
4904 			rel->wanted = cpu_to_le32(cap->mds_wanted);
4905 			rel->dname_len = 0;
4906 			rel->dname_seq = 0;
4907 			*p += sizeof(*rel);
4908 			ret = 1;
4909 		} else {
4910 			doutc(cl, "%p %llx.%llx cap %p %s (noop)\n",
4911 			      inode, ceph_vinop(inode), cap,
4912 			      ceph_cap_string(cap->issued));
4913 		}
4914 	}
4915 	spin_unlock(&ci->i_ceph_lock);
4916 	return ret;
4917 }
4918 
4919 /**
4920  * ceph_encode_dentry_release - encode a dentry release into an outgoing request
4921  * @p: outgoing request buffer
4922  * @dentry: dentry to release
4923  * @dir: dir to release it from
4924  * @mds: mds that we're speaking to
4925  * @drop: caps being dropped
4926  * @unless: unless we have these caps
4927  *
4928  * Encode a dentry release into an outgoing request buffer. Returns 1 if the
4929  * thing was released, or a negative error code otherwise.
4930  */
ceph_encode_dentry_release(void ** p,struct dentry * dentry,struct inode * dir,int mds,int drop,int unless)4931 int ceph_encode_dentry_release(void **p, struct dentry *dentry,
4932 			       struct inode *dir,
4933 			       int mds, int drop, int unless)
4934 {
4935 	struct ceph_mds_request_release *rel = *p;
4936 	struct ceph_dentry_info *di = ceph_dentry(dentry);
4937 	struct ceph_client *cl;
4938 	int force = 0;
4939 	int ret;
4940 
4941 	/* This shouldn't happen */
4942 	BUG_ON(!dir);
4943 
4944 	/*
4945 	 * force an record for the directory caps if we have a dentry lease.
4946 	 * this is racy (can't take i_ceph_lock and d_lock together), but it
4947 	 * doesn't have to be perfect; the mds will revoke anything we don't
4948 	 * release.
4949 	 */
4950 	spin_lock(&dentry->d_lock);
4951 	if (di->lease_session && di->lease_session->s_mds == mds)
4952 		force = 1;
4953 	spin_unlock(&dentry->d_lock);
4954 
4955 	ret = ceph_encode_inode_release(p, dir, mds, drop, unless, force);
4956 
4957 	cl = ceph_inode_to_client(dir);
4958 	spin_lock(&dentry->d_lock);
4959 	if (ret && di->lease_session && di->lease_session->s_mds == mds) {
4960 		doutc(cl, "%p mds%d seq %d\n",  dentry, mds,
4961 		      (int)di->lease_seq);
4962 		rel->dname_seq = cpu_to_le32(di->lease_seq);
4963 		__ceph_mdsc_drop_dentry_lease(dentry);
4964 		spin_unlock(&dentry->d_lock);
4965 		if (IS_ENCRYPTED(dir) && fscrypt_has_encryption_key(dir)) {
4966 			int ret2 = ceph_encode_encrypted_fname(dir, dentry, *p);
4967 
4968 			if (ret2 < 0)
4969 				return ret2;
4970 
4971 			rel->dname_len = cpu_to_le32(ret2);
4972 			*p += ret2;
4973 		} else {
4974 			rel->dname_len = cpu_to_le32(dentry->d_name.len);
4975 			memcpy(*p, dentry->d_name.name, dentry->d_name.len);
4976 			*p += dentry->d_name.len;
4977 		}
4978 	} else {
4979 		spin_unlock(&dentry->d_lock);
4980 	}
4981 	return ret;
4982 }
4983 
remove_capsnaps(struct ceph_mds_client * mdsc,struct inode * inode)4984 static int remove_capsnaps(struct ceph_mds_client *mdsc, struct inode *inode)
4985 {
4986 	struct ceph_inode_info *ci = ceph_inode(inode);
4987 	struct ceph_client *cl = mdsc->fsc->client;
4988 	struct ceph_cap_snap *capsnap;
4989 	int capsnap_release = 0;
4990 
4991 	lockdep_assert_held(&ci->i_ceph_lock);
4992 
4993 	doutc(cl, "removing capsnaps, ci is %p, %p %llx.%llx\n",
4994 	      ci, inode, ceph_vinop(inode));
4995 
4996 	while (!list_empty(&ci->i_cap_snaps)) {
4997 		capsnap = list_first_entry(&ci->i_cap_snaps,
4998 					   struct ceph_cap_snap, ci_item);
4999 		__ceph_remove_capsnap(inode, capsnap, NULL, NULL);
5000 		ceph_put_snap_context(capsnap->context);
5001 		ceph_put_cap_snap(capsnap);
5002 		capsnap_release++;
5003 	}
5004 	wake_up_all(&ci->i_cap_wq);
5005 	wake_up_all(&mdsc->cap_flushing_wq);
5006 	return capsnap_release;
5007 }
5008 
ceph_purge_inode_cap(struct inode * inode,struct ceph_cap * cap,bool * invalidate)5009 int ceph_purge_inode_cap(struct inode *inode, struct ceph_cap *cap, bool *invalidate)
5010 {
5011 	struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
5012 	struct ceph_mds_client *mdsc = fsc->mdsc;
5013 	struct ceph_client *cl = fsc->client;
5014 	struct ceph_inode_info *ci = ceph_inode(inode);
5015 	bool is_auth;
5016 	bool dirty_dropped = false;
5017 	int iputs = 0;
5018 
5019 	lockdep_assert_held(&ci->i_ceph_lock);
5020 
5021 	doutc(cl, "removing cap %p, ci is %p, %p %llx.%llx\n",
5022 	      cap, ci, inode, ceph_vinop(inode));
5023 
5024 	is_auth = (cap == ci->i_auth_cap);
5025 	__ceph_remove_cap(cap, false);
5026 	if (is_auth) {
5027 		struct ceph_cap_flush *cf;
5028 
5029 		if (ceph_inode_is_shutdown(inode)) {
5030 			if (inode->i_data.nrpages > 0)
5031 				*invalidate = true;
5032 			if (ci->i_wrbuffer_ref > 0)
5033 				mapping_set_error(&inode->i_data, -EIO);
5034 		}
5035 
5036 		spin_lock(&mdsc->cap_dirty_lock);
5037 
5038 		/* trash all of the cap flushes for this inode */
5039 		while (!list_empty(&ci->i_cap_flush_list)) {
5040 			cf = list_first_entry(&ci->i_cap_flush_list,
5041 					      struct ceph_cap_flush, i_list);
5042 			list_del_init(&cf->g_list);
5043 			list_del_init(&cf->i_list);
5044 			if (!cf->is_capsnap)
5045 				ceph_free_cap_flush(cf);
5046 		}
5047 
5048 		if (!list_empty(&ci->i_dirty_item)) {
5049 			pr_warn_ratelimited_client(cl,
5050 				" dropping dirty %s state for %p %llx.%llx\n",
5051 				ceph_cap_string(ci->i_dirty_caps),
5052 				inode, ceph_vinop(inode));
5053 			ci->i_dirty_caps = 0;
5054 			list_del_init(&ci->i_dirty_item);
5055 			dirty_dropped = true;
5056 		}
5057 		if (!list_empty(&ci->i_flushing_item)) {
5058 			pr_warn_ratelimited_client(cl,
5059 				" dropping dirty+flushing %s state for %p %llx.%llx\n",
5060 				ceph_cap_string(ci->i_flushing_caps),
5061 				inode, ceph_vinop(inode));
5062 			ci->i_flushing_caps = 0;
5063 			list_del_init(&ci->i_flushing_item);
5064 			mdsc->num_cap_flushing--;
5065 			dirty_dropped = true;
5066 		}
5067 		spin_unlock(&mdsc->cap_dirty_lock);
5068 
5069 		if (dirty_dropped) {
5070 			mapping_set_error(inode->i_mapping, -EIO);
5071 
5072 			if (ci->i_wrbuffer_ref_head == 0 &&
5073 			    ci->i_wr_ref == 0 &&
5074 			    ci->i_dirty_caps == 0 &&
5075 			    ci->i_flushing_caps == 0) {
5076 				ceph_put_snap_context(ci->i_head_snapc);
5077 				ci->i_head_snapc = NULL;
5078 			}
5079 		}
5080 
5081 		if (atomic_read(&ci->i_filelock_ref) > 0) {
5082 			/* make further file lock syscall return -EIO */
5083 			ci->i_ceph_flags |= CEPH_I_ERROR_FILELOCK;
5084 			pr_warn_ratelimited_client(cl,
5085 				" dropping file locks for %p %llx.%llx\n",
5086 				inode, ceph_vinop(inode));
5087 		}
5088 
5089 		if (!ci->i_dirty_caps && ci->i_prealloc_cap_flush) {
5090 			cf = ci->i_prealloc_cap_flush;
5091 			ci->i_prealloc_cap_flush = NULL;
5092 			if (!cf->is_capsnap)
5093 				ceph_free_cap_flush(cf);
5094 		}
5095 
5096 		if (!list_empty(&ci->i_cap_snaps))
5097 			iputs = remove_capsnaps(mdsc, inode);
5098 	}
5099 	if (dirty_dropped)
5100 		++iputs;
5101 	return iputs;
5102 }
5103