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