xref: /openbmc/linux/fs/ceph/xattr.c (revision 901181b7) 
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
3 #include <linux/ceph/pagelist.h>
4 
5 #include "super.h"
6 #include "mds_client.h"
7 
8 #include <linux/ceph/decode.h>
9 
10 #include <linux/xattr.h>
11 #include <linux/security.h>
12 #include <linux/posix_acl_xattr.h>
13 #include <linux/slab.h>
14 
15 #define XATTR_CEPH_PREFIX "ceph."
16 #define XATTR_CEPH_PREFIX_LEN (sizeof (XATTR_CEPH_PREFIX) - 1)
17 
18 static int __remove_xattr(struct ceph_inode_info *ci,
19 			  struct ceph_inode_xattr *xattr);
20 
21 static bool ceph_is_valid_xattr(const char *name)
22 {
23 	return !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN) ||
24 	       !strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN) ||
25 	       !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
26 	       !strncmp(name, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN);
27 }
28 
29 /*
30  * These define virtual xattrs exposing the recursive directory
31  * statistics and layout metadata.
32  */
33 struct ceph_vxattr {
34 	char *name;
35 	size_t name_size;	/* strlen(name) + 1 (for '\0') */
36 	ssize_t (*getxattr_cb)(struct ceph_inode_info *ci, char *val,
37 			       size_t size);
38 	bool (*exists_cb)(struct ceph_inode_info *ci);
39 	unsigned int flags;
40 };
41 
42 #define VXATTR_FLAG_READONLY		(1<<0)
43 #define VXATTR_FLAG_HIDDEN		(1<<1)
44 #define VXATTR_FLAG_RSTAT		(1<<2)
45 #define VXATTR_FLAG_DIRSTAT		(1<<3)
46 
47 /* layouts */
48 
49 static bool ceph_vxattrcb_layout_exists(struct ceph_inode_info *ci)
50 {
51 	struct ceph_file_layout *fl = &ci->i_layout;
52 	return (fl->stripe_unit > 0 || fl->stripe_count > 0 ||
53 		fl->object_size > 0 || fl->pool_id >= 0 ||
54 		rcu_dereference_raw(fl->pool_ns) != NULL);
55 }
56 
57 static ssize_t ceph_vxattrcb_layout(struct ceph_inode_info *ci, char *val,
58 				    size_t size)
59 {
60 	struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
61 	struct ceph_osd_client *osdc = &fsc->client->osdc;
62 	struct ceph_string *pool_ns;
63 	s64 pool = ci->i_layout.pool_id;
64 	const char *pool_name;
65 	const char *ns_field = " pool_namespace=";
66 	char buf[128];
67 	size_t len, total_len = 0;
68 	ssize_t ret;
69 
70 	pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
71 
72 	dout("ceph_vxattrcb_layout %p\n", &ci->vfs_inode);
73 	down_read(&osdc->lock);
74 	pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
75 	if (pool_name) {
76 		len = snprintf(buf, sizeof(buf),
77 		"stripe_unit=%u stripe_count=%u object_size=%u pool=",
78 		ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
79 	        ci->i_layout.object_size);
80 		total_len = len + strlen(pool_name);
81 	} else {
82 		len = snprintf(buf, sizeof(buf),
83 		"stripe_unit=%u stripe_count=%u object_size=%u pool=%lld",
84 		ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
85 		ci->i_layout.object_size, pool);
86 		total_len = len;
87 	}
88 
89 	if (pool_ns)
90 		total_len += strlen(ns_field) + pool_ns->len;
91 
92 	ret = total_len;
93 	if (size >= total_len) {
94 		memcpy(val, buf, len);
95 		ret = len;
96 		if (pool_name) {
97 			len = strlen(pool_name);
98 			memcpy(val + ret, pool_name, len);
99 			ret += len;
100 		}
101 		if (pool_ns) {
102 			len = strlen(ns_field);
103 			memcpy(val + ret, ns_field, len);
104 			ret += len;
105 			memcpy(val + ret, pool_ns->str, pool_ns->len);
106 			ret += pool_ns->len;
107 		}
108 	}
109 	up_read(&osdc->lock);
110 	ceph_put_string(pool_ns);
111 	return ret;
112 }
113 
114 /*
115  * The convention with strings in xattrs is that they should not be NULL
116  * terminated, since we're returning the length with them. snprintf always
117  * NULL terminates however, so call it on a temporary buffer and then memcpy
118  * the result into place.
119  */
120 static __printf(3, 4)
121 int ceph_fmt_xattr(char *val, size_t size, const char *fmt, ...)
122 {
123 	int ret;
124 	va_list args;
125 	char buf[96]; /* NB: reevaluate size if new vxattrs are added */
126 
127 	va_start(args, fmt);
128 	ret = vsnprintf(buf, size ? sizeof(buf) : 0, fmt, args);
129 	va_end(args);
130 
131 	/* Sanity check */
132 	if (size && ret + 1 > sizeof(buf)) {
133 		WARN_ONCE(true, "Returned length too big (%d)", ret);
134 		return -E2BIG;
135 	}
136 
137 	if (ret <= size)
138 		memcpy(val, buf, ret);
139 	return ret;
140 }
141 
142 static ssize_t ceph_vxattrcb_layout_stripe_unit(struct ceph_inode_info *ci,
143 						char *val, size_t size)
144 {
145 	return ceph_fmt_xattr(val, size, "%u", ci->i_layout.stripe_unit);
146 }
147 
148 static ssize_t ceph_vxattrcb_layout_stripe_count(struct ceph_inode_info *ci,
149 						 char *val, size_t size)
150 {
151 	return ceph_fmt_xattr(val, size, "%u", ci->i_layout.stripe_count);
152 }
153 
154 static ssize_t ceph_vxattrcb_layout_object_size(struct ceph_inode_info *ci,
155 						char *val, size_t size)
156 {
157 	return ceph_fmt_xattr(val, size, "%u", ci->i_layout.object_size);
158 }
159 
160 static ssize_t ceph_vxattrcb_layout_pool(struct ceph_inode_info *ci,
161 					 char *val, size_t size)
162 {
163 	ssize_t ret;
164 	struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
165 	struct ceph_osd_client *osdc = &fsc->client->osdc;
166 	s64 pool = ci->i_layout.pool_id;
167 	const char *pool_name;
168 
169 	down_read(&osdc->lock);
170 	pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
171 	if (pool_name) {
172 		ret = strlen(pool_name);
173 		if (ret <= size)
174 			memcpy(val, pool_name, ret);
175 	} else {
176 		ret = ceph_fmt_xattr(val, size, "%lld", pool);
177 	}
178 	up_read(&osdc->lock);
179 	return ret;
180 }
181 
182 static ssize_t ceph_vxattrcb_layout_pool_namespace(struct ceph_inode_info *ci,
183 						   char *val, size_t size)
184 {
185 	ssize_t ret = 0;
186 	struct ceph_string *ns = ceph_try_get_string(ci->i_layout.pool_ns);
187 
188 	if (ns) {
189 		ret = ns->len;
190 		if (ret <= size)
191 			memcpy(val, ns->str, ret);
192 		ceph_put_string(ns);
193 	}
194 	return ret;
195 }
196 
197 /* directories */
198 
199 static ssize_t ceph_vxattrcb_dir_entries(struct ceph_inode_info *ci, char *val,
200 					 size_t size)
201 {
202 	return ceph_fmt_xattr(val, size, "%lld", ci->i_files + ci->i_subdirs);
203 }
204 
205 static ssize_t ceph_vxattrcb_dir_files(struct ceph_inode_info *ci, char *val,
206 				       size_t size)
207 {
208 	return ceph_fmt_xattr(val, size, "%lld", ci->i_files);
209 }
210 
211 static ssize_t ceph_vxattrcb_dir_subdirs(struct ceph_inode_info *ci, char *val,
212 					 size_t size)
213 {
214 	return ceph_fmt_xattr(val, size, "%lld", ci->i_subdirs);
215 }
216 
217 static ssize_t ceph_vxattrcb_dir_rentries(struct ceph_inode_info *ci, char *val,
218 					  size_t size)
219 {
220 	return ceph_fmt_xattr(val, size, "%lld",
221 				ci->i_rfiles + ci->i_rsubdirs);
222 }
223 
224 static ssize_t ceph_vxattrcb_dir_rfiles(struct ceph_inode_info *ci, char *val,
225 					size_t size)
226 {
227 	return ceph_fmt_xattr(val, size, "%lld", ci->i_rfiles);
228 }
229 
230 static ssize_t ceph_vxattrcb_dir_rsubdirs(struct ceph_inode_info *ci, char *val,
231 					  size_t size)
232 {
233 	return ceph_fmt_xattr(val, size, "%lld", ci->i_rsubdirs);
234 }
235 
236 static ssize_t ceph_vxattrcb_dir_rsnaps(struct ceph_inode_info *ci, char *val,
237 					  size_t size)
238 {
239 	return ceph_fmt_xattr(val, size, "%lld", ci->i_rsnaps);
240 }
241 
242 static ssize_t ceph_vxattrcb_dir_rbytes(struct ceph_inode_info *ci, char *val,
243 					size_t size)
244 {
245 	return ceph_fmt_xattr(val, size, "%lld", ci->i_rbytes);
246 }
247 
248 static ssize_t ceph_vxattrcb_dir_rctime(struct ceph_inode_info *ci, char *val,
249 					size_t size)
250 {
251 	return ceph_fmt_xattr(val, size, "%lld.%09ld", ci->i_rctime.tv_sec,
252 				ci->i_rctime.tv_nsec);
253 }
254 
255 /* dir pin */
256 static bool ceph_vxattrcb_dir_pin_exists(struct ceph_inode_info *ci)
257 {
258 	return ci->i_dir_pin != -ENODATA;
259 }
260 
261 static ssize_t ceph_vxattrcb_dir_pin(struct ceph_inode_info *ci, char *val,
262 				     size_t size)
263 {
264 	return ceph_fmt_xattr(val, size, "%d", (int)ci->i_dir_pin);
265 }
266 
267 /* quotas */
268 static bool ceph_vxattrcb_quota_exists(struct ceph_inode_info *ci)
269 {
270 	bool ret = false;
271 	spin_lock(&ci->i_ceph_lock);
272 	if ((ci->i_max_files || ci->i_max_bytes) &&
273 	    ci->i_vino.snap == CEPH_NOSNAP &&
274 	    ci->i_snap_realm &&
275 	    ci->i_snap_realm->ino == ci->i_vino.ino)
276 		ret = true;
277 	spin_unlock(&ci->i_ceph_lock);
278 	return ret;
279 }
280 
281 static ssize_t ceph_vxattrcb_quota(struct ceph_inode_info *ci, char *val,
282 				   size_t size)
283 {
284 	return ceph_fmt_xattr(val, size, "max_bytes=%llu max_files=%llu",
285 				ci->i_max_bytes, ci->i_max_files);
286 }
287 
288 static ssize_t ceph_vxattrcb_quota_max_bytes(struct ceph_inode_info *ci,
289 					     char *val, size_t size)
290 {
291 	return ceph_fmt_xattr(val, size, "%llu", ci->i_max_bytes);
292 }
293 
294 static ssize_t ceph_vxattrcb_quota_max_files(struct ceph_inode_info *ci,
295 					     char *val, size_t size)
296 {
297 	return ceph_fmt_xattr(val, size, "%llu", ci->i_max_files);
298 }
299 
300 /* snapshots */
301 static bool ceph_vxattrcb_snap_btime_exists(struct ceph_inode_info *ci)
302 {
303 	return (ci->i_snap_btime.tv_sec != 0 || ci->i_snap_btime.tv_nsec != 0);
304 }
305 
306 static ssize_t ceph_vxattrcb_snap_btime(struct ceph_inode_info *ci, char *val,
307 					size_t size)
308 {
309 	return ceph_fmt_xattr(val, size, "%lld.%09ld", ci->i_snap_btime.tv_sec,
310 				ci->i_snap_btime.tv_nsec);
311 }
312 
313 static ssize_t ceph_vxattrcb_cluster_fsid(struct ceph_inode_info *ci,
314 					  char *val, size_t size)
315 {
316 	struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
317 
318 	return ceph_fmt_xattr(val, size, "%pU", &fsc->client->fsid);
319 }
320 
321 static ssize_t ceph_vxattrcb_client_id(struct ceph_inode_info *ci,
322 				       char *val, size_t size)
323 {
324 	struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
325 
326 	return ceph_fmt_xattr(val, size, "client%lld",
327 			      ceph_client_gid(fsc->client));
328 }
329 
330 static ssize_t ceph_vxattrcb_caps(struct ceph_inode_info *ci, char *val,
331 					size_t size)
332 {
333 	int issued;
334 
335 	spin_lock(&ci->i_ceph_lock);
336 	issued = __ceph_caps_issued(ci, NULL);
337 	spin_unlock(&ci->i_ceph_lock);
338 
339 	return ceph_fmt_xattr(val, size, "%s/0x%x",
340 			      ceph_cap_string(issued), issued);
341 }
342 
343 static ssize_t ceph_vxattrcb_auth_mds(struct ceph_inode_info *ci,
344 				       char *val, size_t size)
345 {
346 	int ret;
347 
348 	spin_lock(&ci->i_ceph_lock);
349 	ret = ceph_fmt_xattr(val, size, "%d",
350 			     ci->i_auth_cap ? ci->i_auth_cap->session->s_mds : -1);
351 	spin_unlock(&ci->i_ceph_lock);
352 	return ret;
353 }
354 
355 #define CEPH_XATTR_NAME(_type, _name)	XATTR_CEPH_PREFIX #_type "." #_name
356 #define CEPH_XATTR_NAME2(_type, _name, _name2)	\
357 	XATTR_CEPH_PREFIX #_type "." #_name "." #_name2
358 
359 #define XATTR_NAME_CEPH(_type, _name, _flags)				\
360 	{								\
361 		.name = CEPH_XATTR_NAME(_type, _name),			\
362 		.name_size = sizeof (CEPH_XATTR_NAME(_type, _name)), \
363 		.getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name, \
364 		.exists_cb = NULL,					\
365 		.flags = (VXATTR_FLAG_READONLY | _flags),		\
366 	}
367 #define XATTR_RSTAT_FIELD(_type, _name)			\
368 	XATTR_NAME_CEPH(_type, _name, VXATTR_FLAG_RSTAT)
369 #define XATTR_LAYOUT_FIELD(_type, _name, _field)			\
370 	{								\
371 		.name = CEPH_XATTR_NAME2(_type, _name, _field),	\
372 		.name_size = sizeof (CEPH_XATTR_NAME2(_type, _name, _field)), \
373 		.getxattr_cb = ceph_vxattrcb_ ## _name ## _ ## _field, \
374 		.exists_cb = ceph_vxattrcb_layout_exists,	\
375 		.flags = VXATTR_FLAG_HIDDEN,			\
376 	}
377 #define XATTR_QUOTA_FIELD(_type, _name)					\
378 	{								\
379 		.name = CEPH_XATTR_NAME(_type, _name),			\
380 		.name_size = sizeof(CEPH_XATTR_NAME(_type, _name)),	\
381 		.getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name,	\
382 		.exists_cb = ceph_vxattrcb_quota_exists,		\
383 		.flags = VXATTR_FLAG_HIDDEN,				\
384 	}
385 
386 static struct ceph_vxattr ceph_dir_vxattrs[] = {
387 	{
388 		.name = "ceph.dir.layout",
389 		.name_size = sizeof("ceph.dir.layout"),
390 		.getxattr_cb = ceph_vxattrcb_layout,
391 		.exists_cb = ceph_vxattrcb_layout_exists,
392 		.flags = VXATTR_FLAG_HIDDEN,
393 	},
394 	XATTR_LAYOUT_FIELD(dir, layout, stripe_unit),
395 	XATTR_LAYOUT_FIELD(dir, layout, stripe_count),
396 	XATTR_LAYOUT_FIELD(dir, layout, object_size),
397 	XATTR_LAYOUT_FIELD(dir, layout, pool),
398 	XATTR_LAYOUT_FIELD(dir, layout, pool_namespace),
399 	XATTR_NAME_CEPH(dir, entries, VXATTR_FLAG_DIRSTAT),
400 	XATTR_NAME_CEPH(dir, files, VXATTR_FLAG_DIRSTAT),
401 	XATTR_NAME_CEPH(dir, subdirs, VXATTR_FLAG_DIRSTAT),
402 	XATTR_RSTAT_FIELD(dir, rentries),
403 	XATTR_RSTAT_FIELD(dir, rfiles),
404 	XATTR_RSTAT_FIELD(dir, rsubdirs),
405 	XATTR_RSTAT_FIELD(dir, rsnaps),
406 	XATTR_RSTAT_FIELD(dir, rbytes),
407 	XATTR_RSTAT_FIELD(dir, rctime),
408 	{
409 		.name = "ceph.dir.pin",
410 		.name_size = sizeof("ceph.dir.pin"),
411 		.getxattr_cb = ceph_vxattrcb_dir_pin,
412 		.exists_cb = ceph_vxattrcb_dir_pin_exists,
413 		.flags = VXATTR_FLAG_HIDDEN,
414 	},
415 	{
416 		.name = "ceph.quota",
417 		.name_size = sizeof("ceph.quota"),
418 		.getxattr_cb = ceph_vxattrcb_quota,
419 		.exists_cb = ceph_vxattrcb_quota_exists,
420 		.flags = VXATTR_FLAG_HIDDEN,
421 	},
422 	XATTR_QUOTA_FIELD(quota, max_bytes),
423 	XATTR_QUOTA_FIELD(quota, max_files),
424 	{
425 		.name = "ceph.snap.btime",
426 		.name_size = sizeof("ceph.snap.btime"),
427 		.getxattr_cb = ceph_vxattrcb_snap_btime,
428 		.exists_cb = ceph_vxattrcb_snap_btime_exists,
429 		.flags = VXATTR_FLAG_READONLY,
430 	},
431 	{
432 		.name = "ceph.caps",
433 		.name_size = sizeof("ceph.caps"),
434 		.getxattr_cb = ceph_vxattrcb_caps,
435 		.exists_cb = NULL,
436 		.flags = VXATTR_FLAG_HIDDEN,
437 	},
438 	{ .name = NULL, 0 }	/* Required table terminator */
439 };
440 
441 /* files */
442 
443 static struct ceph_vxattr ceph_file_vxattrs[] = {
444 	{
445 		.name = "ceph.file.layout",
446 		.name_size = sizeof("ceph.file.layout"),
447 		.getxattr_cb = ceph_vxattrcb_layout,
448 		.exists_cb = ceph_vxattrcb_layout_exists,
449 		.flags = VXATTR_FLAG_HIDDEN,
450 	},
451 	XATTR_LAYOUT_FIELD(file, layout, stripe_unit),
452 	XATTR_LAYOUT_FIELD(file, layout, stripe_count),
453 	XATTR_LAYOUT_FIELD(file, layout, object_size),
454 	XATTR_LAYOUT_FIELD(file, layout, pool),
455 	XATTR_LAYOUT_FIELD(file, layout, pool_namespace),
456 	{
457 		.name = "ceph.snap.btime",
458 		.name_size = sizeof("ceph.snap.btime"),
459 		.getxattr_cb = ceph_vxattrcb_snap_btime,
460 		.exists_cb = ceph_vxattrcb_snap_btime_exists,
461 		.flags = VXATTR_FLAG_READONLY,
462 	},
463 	{
464 		.name = "ceph.caps",
465 		.name_size = sizeof("ceph.caps"),
466 		.getxattr_cb = ceph_vxattrcb_caps,
467 		.exists_cb = NULL,
468 		.flags = VXATTR_FLAG_HIDDEN,
469 	},
470 	{ .name = NULL, 0 }	/* Required table terminator */
471 };
472 
473 static struct ceph_vxattr ceph_common_vxattrs[] = {
474 	{
475 		.name = "ceph.cluster_fsid",
476 		.name_size = sizeof("ceph.cluster_fsid"),
477 		.getxattr_cb = ceph_vxattrcb_cluster_fsid,
478 		.exists_cb = NULL,
479 		.flags = VXATTR_FLAG_READONLY,
480 	},
481 	{
482 		.name = "ceph.client_id",
483 		.name_size = sizeof("ceph.client_id"),
484 		.getxattr_cb = ceph_vxattrcb_client_id,
485 		.exists_cb = NULL,
486 		.flags = VXATTR_FLAG_READONLY,
487 	},
488 	{
489 		.name = "ceph.auth_mds",
490 		.name_size = sizeof("ceph.auth_mds"),
491 		.getxattr_cb = ceph_vxattrcb_auth_mds,
492 		.exists_cb = NULL,
493 		.flags = VXATTR_FLAG_READONLY,
494 	},
495 	{ .name = NULL, 0 }	/* Required table terminator */
496 };
497 
498 static struct ceph_vxattr *ceph_inode_vxattrs(struct inode *inode)
499 {
500 	if (S_ISDIR(inode->i_mode))
501 		return ceph_dir_vxattrs;
502 	else if (S_ISREG(inode->i_mode))
503 		return ceph_file_vxattrs;
504 	return NULL;
505 }
506 
507 static struct ceph_vxattr *ceph_match_vxattr(struct inode *inode,
508 						const char *name)
509 {
510 	struct ceph_vxattr *vxattr = ceph_inode_vxattrs(inode);
511 
512 	if (vxattr) {
513 		while (vxattr->name) {
514 			if (!strcmp(vxattr->name, name))
515 				return vxattr;
516 			vxattr++;
517 		}
518 	}
519 
520 	vxattr = ceph_common_vxattrs;
521 	while (vxattr->name) {
522 		if (!strcmp(vxattr->name, name))
523 			return vxattr;
524 		vxattr++;
525 	}
526 
527 	return NULL;
528 }
529 
530 static int __set_xattr(struct ceph_inode_info *ci,
531 			   const char *name, int name_len,
532 			   const char *val, int val_len,
533 			   int flags, int update_xattr,
534 			   struct ceph_inode_xattr **newxattr)
535 {
536 	struct rb_node **p;
537 	struct rb_node *parent = NULL;
538 	struct ceph_inode_xattr *xattr = NULL;
539 	int c;
540 	int new = 0;
541 
542 	p = &ci->i_xattrs.index.rb_node;
543 	while (*p) {
544 		parent = *p;
545 		xattr = rb_entry(parent, struct ceph_inode_xattr, node);
546 		c = strncmp(name, xattr->name, min(name_len, xattr->name_len));
547 		if (c < 0)
548 			p = &(*p)->rb_left;
549 		else if (c > 0)
550 			p = &(*p)->rb_right;
551 		else {
552 			if (name_len == xattr->name_len)
553 				break;
554 			else if (name_len < xattr->name_len)
555 				p = &(*p)->rb_left;
556 			else
557 				p = &(*p)->rb_right;
558 		}
559 		xattr = NULL;
560 	}
561 
562 	if (update_xattr) {
563 		int err = 0;
564 
565 		if (xattr && (flags & XATTR_CREATE))
566 			err = -EEXIST;
567 		else if (!xattr && (flags & XATTR_REPLACE))
568 			err = -ENODATA;
569 		if (err) {
570 			kfree(name);
571 			kfree(val);
572 			kfree(*newxattr);
573 			return err;
574 		}
575 		if (update_xattr < 0) {
576 			if (xattr)
577 				__remove_xattr(ci, xattr);
578 			kfree(name);
579 			kfree(*newxattr);
580 			return 0;
581 		}
582 	}
583 
584 	if (!xattr) {
585 		new = 1;
586 		xattr = *newxattr;
587 		xattr->name = name;
588 		xattr->name_len = name_len;
589 		xattr->should_free_name = update_xattr;
590 
591 		ci->i_xattrs.count++;
592 		dout("__set_xattr count=%d\n", ci->i_xattrs.count);
593 	} else {
594 		kfree(*newxattr);
595 		*newxattr = NULL;
596 		if (xattr->should_free_val)
597 			kfree(xattr->val);
598 
599 		if (update_xattr) {
600 			kfree(name);
601 			name = xattr->name;
602 		}
603 		ci->i_xattrs.names_size -= xattr->name_len;
604 		ci->i_xattrs.vals_size -= xattr->val_len;
605 	}
606 	ci->i_xattrs.names_size += name_len;
607 	ci->i_xattrs.vals_size += val_len;
608 	if (val)
609 		xattr->val = val;
610 	else
611 		xattr->val = "";
612 
613 	xattr->val_len = val_len;
614 	xattr->dirty = update_xattr;
615 	xattr->should_free_val = (val && update_xattr);
616 
617 	if (new) {
618 		rb_link_node(&xattr->node, parent, p);
619 		rb_insert_color(&xattr->node, &ci->i_xattrs.index);
620 		dout("__set_xattr_val p=%p\n", p);
621 	}
622 
623 	dout("__set_xattr_val added %llx.%llx xattr %p %.*s=%.*s\n",
624 	     ceph_vinop(&ci->vfs_inode), xattr, name_len, name, val_len, val);
625 
626 	return 0;
627 }
628 
629 static struct ceph_inode_xattr *__get_xattr(struct ceph_inode_info *ci,
630 			   const char *name)
631 {
632 	struct rb_node **p;
633 	struct rb_node *parent = NULL;
634 	struct ceph_inode_xattr *xattr = NULL;
635 	int name_len = strlen(name);
636 	int c;
637 
638 	p = &ci->i_xattrs.index.rb_node;
639 	while (*p) {
640 		parent = *p;
641 		xattr = rb_entry(parent, struct ceph_inode_xattr, node);
642 		c = strncmp(name, xattr->name, xattr->name_len);
643 		if (c == 0 && name_len > xattr->name_len)
644 			c = 1;
645 		if (c < 0)
646 			p = &(*p)->rb_left;
647 		else if (c > 0)
648 			p = &(*p)->rb_right;
649 		else {
650 			dout("__get_xattr %s: found %.*s\n", name,
651 			     xattr->val_len, xattr->val);
652 			return xattr;
653 		}
654 	}
655 
656 	dout("__get_xattr %s: not found\n", name);
657 
658 	return NULL;
659 }
660 
661 static void __free_xattr(struct ceph_inode_xattr *xattr)
662 {
663 	BUG_ON(!xattr);
664 
665 	if (xattr->should_free_name)
666 		kfree(xattr->name);
667 	if (xattr->should_free_val)
668 		kfree(xattr->val);
669 
670 	kfree(xattr);
671 }
672 
673 static int __remove_xattr(struct ceph_inode_info *ci,
674 			  struct ceph_inode_xattr *xattr)
675 {
676 	if (!xattr)
677 		return -ENODATA;
678 
679 	rb_erase(&xattr->node, &ci->i_xattrs.index);
680 
681 	if (xattr->should_free_name)
682 		kfree(xattr->name);
683 	if (xattr->should_free_val)
684 		kfree(xattr->val);
685 
686 	ci->i_xattrs.names_size -= xattr->name_len;
687 	ci->i_xattrs.vals_size -= xattr->val_len;
688 	ci->i_xattrs.count--;
689 	kfree(xattr);
690 
691 	return 0;
692 }
693 
694 static char *__copy_xattr_names(struct ceph_inode_info *ci,
695 				char *dest)
696 {
697 	struct rb_node *p;
698 	struct ceph_inode_xattr *xattr = NULL;
699 
700 	p = rb_first(&ci->i_xattrs.index);
701 	dout("__copy_xattr_names count=%d\n", ci->i_xattrs.count);
702 
703 	while (p) {
704 		xattr = rb_entry(p, struct ceph_inode_xattr, node);
705 		memcpy(dest, xattr->name, xattr->name_len);
706 		dest[xattr->name_len] = '\0';
707 
708 		dout("dest=%s %p (%s) (%d/%d)\n", dest, xattr, xattr->name,
709 		     xattr->name_len, ci->i_xattrs.names_size);
710 
711 		dest += xattr->name_len + 1;
712 		p = rb_next(p);
713 	}
714 
715 	return dest;
716 }
717 
718 void __ceph_destroy_xattrs(struct ceph_inode_info *ci)
719 {
720 	struct rb_node *p, *tmp;
721 	struct ceph_inode_xattr *xattr = NULL;
722 
723 	p = rb_first(&ci->i_xattrs.index);
724 
725 	dout("__ceph_destroy_xattrs p=%p\n", p);
726 
727 	while (p) {
728 		xattr = rb_entry(p, struct ceph_inode_xattr, node);
729 		tmp = p;
730 		p = rb_next(tmp);
731 		dout("__ceph_destroy_xattrs next p=%p (%.*s)\n", p,
732 		     xattr->name_len, xattr->name);
733 		rb_erase(tmp, &ci->i_xattrs.index);
734 
735 		__free_xattr(xattr);
736 	}
737 
738 	ci->i_xattrs.names_size = 0;
739 	ci->i_xattrs.vals_size = 0;
740 	ci->i_xattrs.index_version = 0;
741 	ci->i_xattrs.count = 0;
742 	ci->i_xattrs.index = RB_ROOT;
743 }
744 
745 static int __build_xattrs(struct inode *inode)
746 	__releases(ci->i_ceph_lock)
747 	__acquires(ci->i_ceph_lock)
748 {
749 	u32 namelen;
750 	u32 numattr = 0;
751 	void *p, *end;
752 	u32 len;
753 	const char *name, *val;
754 	struct ceph_inode_info *ci = ceph_inode(inode);
755 	u64 xattr_version;
756 	struct ceph_inode_xattr **xattrs = NULL;
757 	int err = 0;
758 	int i;
759 
760 	dout("__build_xattrs() len=%d\n",
761 	     ci->i_xattrs.blob ? (int)ci->i_xattrs.blob->vec.iov_len : 0);
762 
763 	if (ci->i_xattrs.index_version >= ci->i_xattrs.version)
764 		return 0; /* already built */
765 
766 	__ceph_destroy_xattrs(ci);
767 
768 start:
769 	/* updated internal xattr rb tree */
770 	if (ci->i_xattrs.blob && ci->i_xattrs.blob->vec.iov_len > 4) {
771 		p = ci->i_xattrs.blob->vec.iov_base;
772 		end = p + ci->i_xattrs.blob->vec.iov_len;
773 		ceph_decode_32_safe(&p, end, numattr, bad);
774 		xattr_version = ci->i_xattrs.version;
775 		spin_unlock(&ci->i_ceph_lock);
776 
777 		xattrs = kcalloc(numattr, sizeof(struct ceph_inode_xattr *),
778 				 GFP_NOFS);
779 		err = -ENOMEM;
780 		if (!xattrs)
781 			goto bad_lock;
782 
783 		for (i = 0; i < numattr; i++) {
784 			xattrs[i] = kmalloc(sizeof(struct ceph_inode_xattr),
785 					    GFP_NOFS);
786 			if (!xattrs[i])
787 				goto bad_lock;
788 		}
789 
790 		spin_lock(&ci->i_ceph_lock);
791 		if (ci->i_xattrs.version != xattr_version) {
792 			/* lost a race, retry */
793 			for (i = 0; i < numattr; i++)
794 				kfree(xattrs[i]);
795 			kfree(xattrs);
796 			xattrs = NULL;
797 			goto start;
798 		}
799 		err = -EIO;
800 		while (numattr--) {
801 			ceph_decode_32_safe(&p, end, len, bad);
802 			namelen = len;
803 			name = p;
804 			p += len;
805 			ceph_decode_32_safe(&p, end, len, bad);
806 			val = p;
807 			p += len;
808 
809 			err = __set_xattr(ci, name, namelen, val, len,
810 					  0, 0, &xattrs[numattr]);
811 
812 			if (err < 0)
813 				goto bad;
814 		}
815 		kfree(xattrs);
816 	}
817 	ci->i_xattrs.index_version = ci->i_xattrs.version;
818 	ci->i_xattrs.dirty = false;
819 
820 	return err;
821 bad_lock:
822 	spin_lock(&ci->i_ceph_lock);
823 bad:
824 	if (xattrs) {
825 		for (i = 0; i < numattr; i++)
826 			kfree(xattrs[i]);
827 		kfree(xattrs);
828 	}
829 	ci->i_xattrs.names_size = 0;
830 	return err;
831 }
832 
833 static int __get_required_blob_size(struct ceph_inode_info *ci, int name_size,
834 				    int val_size)
835 {
836 	/*
837 	 * 4 bytes for the length, and additional 4 bytes per each xattr name,
838 	 * 4 bytes per each value
839 	 */
840 	int size = 4 + ci->i_xattrs.count*(4 + 4) +
841 			     ci->i_xattrs.names_size +
842 			     ci->i_xattrs.vals_size;
843 	dout("__get_required_blob_size c=%d names.size=%d vals.size=%d\n",
844 	     ci->i_xattrs.count, ci->i_xattrs.names_size,
845 	     ci->i_xattrs.vals_size);
846 
847 	if (name_size)
848 		size += 4 + 4 + name_size + val_size;
849 
850 	return size;
851 }
852 
853 /*
854  * If there are dirty xattrs, reencode xattrs into the prealloc_blob
855  * and swap into place.  It returns the old i_xattrs.blob (or NULL) so
856  * that it can be freed by the caller as the i_ceph_lock is likely to be
857  * held.
858  */
859 struct ceph_buffer *__ceph_build_xattrs_blob(struct ceph_inode_info *ci)
860 {
861 	struct rb_node *p;
862 	struct ceph_inode_xattr *xattr = NULL;
863 	struct ceph_buffer *old_blob = NULL;
864 	void *dest;
865 
866 	dout("__build_xattrs_blob %p\n", &ci->vfs_inode);
867 	if (ci->i_xattrs.dirty) {
868 		int need = __get_required_blob_size(ci, 0, 0);
869 
870 		BUG_ON(need > ci->i_xattrs.prealloc_blob->alloc_len);
871 
872 		p = rb_first(&ci->i_xattrs.index);
873 		dest = ci->i_xattrs.prealloc_blob->vec.iov_base;
874 
875 		ceph_encode_32(&dest, ci->i_xattrs.count);
876 		while (p) {
877 			xattr = rb_entry(p, struct ceph_inode_xattr, node);
878 
879 			ceph_encode_32(&dest, xattr->name_len);
880 			memcpy(dest, xattr->name, xattr->name_len);
881 			dest += xattr->name_len;
882 			ceph_encode_32(&dest, xattr->val_len);
883 			memcpy(dest, xattr->val, xattr->val_len);
884 			dest += xattr->val_len;
885 
886 			p = rb_next(p);
887 		}
888 
889 		/* adjust buffer len; it may be larger than we need */
890 		ci->i_xattrs.prealloc_blob->vec.iov_len =
891 			dest - ci->i_xattrs.prealloc_blob->vec.iov_base;
892 
893 		if (ci->i_xattrs.blob)
894 			old_blob = ci->i_xattrs.blob;
895 		ci->i_xattrs.blob = ci->i_xattrs.prealloc_blob;
896 		ci->i_xattrs.prealloc_blob = NULL;
897 		ci->i_xattrs.dirty = false;
898 		ci->i_xattrs.version++;
899 	}
900 
901 	return old_blob;
902 }
903 
904 static inline int __get_request_mask(struct inode *in) {
905 	struct ceph_mds_request *req = current->journal_info;
906 	int mask = 0;
907 	if (req && req->r_target_inode == in) {
908 		if (req->r_op == CEPH_MDS_OP_LOOKUP ||
909 		    req->r_op == CEPH_MDS_OP_LOOKUPINO ||
910 		    req->r_op == CEPH_MDS_OP_LOOKUPPARENT ||
911 		    req->r_op == CEPH_MDS_OP_GETATTR) {
912 			mask = le32_to_cpu(req->r_args.getattr.mask);
913 		} else if (req->r_op == CEPH_MDS_OP_OPEN ||
914 			   req->r_op == CEPH_MDS_OP_CREATE) {
915 			mask = le32_to_cpu(req->r_args.open.mask);
916 		}
917 	}
918 	return mask;
919 }
920 
921 ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,
922 		      size_t size)
923 {
924 	struct ceph_inode_info *ci = ceph_inode(inode);
925 	struct ceph_inode_xattr *xattr;
926 	struct ceph_vxattr *vxattr = NULL;
927 	int req_mask;
928 	ssize_t err;
929 
930 	/* let's see if a virtual xattr was requested */
931 	vxattr = ceph_match_vxattr(inode, name);
932 	if (vxattr) {
933 		int mask = 0;
934 		if (vxattr->flags & VXATTR_FLAG_RSTAT)
935 			mask |= CEPH_STAT_RSTAT;
936 		if (vxattr->flags & VXATTR_FLAG_DIRSTAT)
937 			mask |= CEPH_CAP_FILE_SHARED;
938 		err = ceph_do_getattr(inode, mask, true);
939 		if (err)
940 			return err;
941 		err = -ENODATA;
942 		if (!(vxattr->exists_cb && !vxattr->exists_cb(ci))) {
943 			err = vxattr->getxattr_cb(ci, value, size);
944 			if (size && size < err)
945 				err = -ERANGE;
946 		}
947 		return err;
948 	}
949 
950 	req_mask = __get_request_mask(inode);
951 
952 	spin_lock(&ci->i_ceph_lock);
953 	dout("getxattr %p name '%s' ver=%lld index_ver=%lld\n", inode, name,
954 	     ci->i_xattrs.version, ci->i_xattrs.index_version);
955 
956 	if (ci->i_xattrs.version == 0 ||
957 	    !((req_mask & CEPH_CAP_XATTR_SHARED) ||
958 	      __ceph_caps_issued_mask_metric(ci, CEPH_CAP_XATTR_SHARED, 1))) {
959 		spin_unlock(&ci->i_ceph_lock);
960 
961 		/* security module gets xattr while filling trace */
962 		if (current->journal_info) {
963 			pr_warn_ratelimited("sync getxattr %p "
964 					    "during filling trace\n", inode);
965 			return -EBUSY;
966 		}
967 
968 		/* get xattrs from mds (if we don't already have them) */
969 		err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
970 		if (err)
971 			return err;
972 		spin_lock(&ci->i_ceph_lock);
973 	}
974 
975 	err = __build_xattrs(inode);
976 	if (err < 0)
977 		goto out;
978 
979 	err = -ENODATA;  /* == ENOATTR */
980 	xattr = __get_xattr(ci, name);
981 	if (!xattr)
982 		goto out;
983 
984 	err = -ERANGE;
985 	if (size && size < xattr->val_len)
986 		goto out;
987 
988 	err = xattr->val_len;
989 	if (size == 0)
990 		goto out;
991 
992 	memcpy(value, xattr->val, xattr->val_len);
993 
994 	if (current->journal_info &&
995 	    !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN) &&
996 	    security_ismaclabel(name + XATTR_SECURITY_PREFIX_LEN))
997 		ci->i_ceph_flags |= CEPH_I_SEC_INITED;
998 out:
999 	spin_unlock(&ci->i_ceph_lock);
1000 	return err;
1001 }
1002 
1003 ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)
1004 {
1005 	struct inode *inode = d_inode(dentry);
1006 	struct ceph_inode_info *ci = ceph_inode(inode);
1007 	bool len_only = (size == 0);
1008 	u32 namelen;
1009 	int err;
1010 
1011 	spin_lock(&ci->i_ceph_lock);
1012 	dout("listxattr %p ver=%lld index_ver=%lld\n", inode,
1013 	     ci->i_xattrs.version, ci->i_xattrs.index_version);
1014 
1015 	if (ci->i_xattrs.version == 0 ||
1016 	    !__ceph_caps_issued_mask_metric(ci, CEPH_CAP_XATTR_SHARED, 1)) {
1017 		spin_unlock(&ci->i_ceph_lock);
1018 		err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
1019 		if (err)
1020 			return err;
1021 		spin_lock(&ci->i_ceph_lock);
1022 	}
1023 
1024 	err = __build_xattrs(inode);
1025 	if (err < 0)
1026 		goto out;
1027 
1028 	/* add 1 byte for each xattr due to the null termination */
1029 	namelen = ci->i_xattrs.names_size + ci->i_xattrs.count;
1030 	if (!len_only) {
1031 		if (namelen > size) {
1032 			err = -ERANGE;
1033 			goto out;
1034 		}
1035 		names = __copy_xattr_names(ci, names);
1036 		size -= namelen;
1037 	}
1038 	err = namelen;
1039 out:
1040 	spin_unlock(&ci->i_ceph_lock);
1041 	return err;
1042 }
1043 
1044 static int ceph_sync_setxattr(struct inode *inode, const char *name,
1045 			      const char *value, size_t size, int flags)
1046 {
1047 	struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1048 	struct ceph_inode_info *ci = ceph_inode(inode);
1049 	struct ceph_mds_request *req;
1050 	struct ceph_mds_client *mdsc = fsc->mdsc;
1051 	struct ceph_osd_client *osdc = &fsc->client->osdc;
1052 	struct ceph_pagelist *pagelist = NULL;
1053 	int op = CEPH_MDS_OP_SETXATTR;
1054 	int err;
1055 
1056 	if (size > 0) {
1057 		/* copy value into pagelist */
1058 		pagelist = ceph_pagelist_alloc(GFP_NOFS);
1059 		if (!pagelist)
1060 			return -ENOMEM;
1061 
1062 		err = ceph_pagelist_append(pagelist, value, size);
1063 		if (err)
1064 			goto out;
1065 	} else if (!value) {
1066 		if (flags & CEPH_XATTR_REPLACE)
1067 			op = CEPH_MDS_OP_RMXATTR;
1068 		else
1069 			flags |= CEPH_XATTR_REMOVE;
1070 	}
1071 
1072 	dout("setxattr value=%.*s\n", (int)size, value);
1073 
1074 	/* do request */
1075 	req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1076 	if (IS_ERR(req)) {
1077 		err = PTR_ERR(req);
1078 		goto out;
1079 	}
1080 
1081 	req->r_path2 = kstrdup(name, GFP_NOFS);
1082 	if (!req->r_path2) {
1083 		ceph_mdsc_put_request(req);
1084 		err = -ENOMEM;
1085 		goto out;
1086 	}
1087 
1088 	if (op == CEPH_MDS_OP_SETXATTR) {
1089 		req->r_args.setxattr.flags = cpu_to_le32(flags);
1090 		req->r_args.setxattr.osdmap_epoch =
1091 			cpu_to_le32(osdc->osdmap->epoch);
1092 		req->r_pagelist = pagelist;
1093 		pagelist = NULL;
1094 	}
1095 
1096 	req->r_inode = inode;
1097 	ihold(inode);
1098 	req->r_num_caps = 1;
1099 	req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
1100 
1101 	dout("xattr.ver (before): %lld\n", ci->i_xattrs.version);
1102 	err = ceph_mdsc_do_request(mdsc, NULL, req);
1103 	ceph_mdsc_put_request(req);
1104 	dout("xattr.ver (after): %lld\n", ci->i_xattrs.version);
1105 
1106 out:
1107 	if (pagelist)
1108 		ceph_pagelist_release(pagelist);
1109 	return err;
1110 }
1111 
1112 int __ceph_setxattr(struct inode *inode, const char *name,
1113 			const void *value, size_t size, int flags)
1114 {
1115 	struct ceph_vxattr *vxattr;
1116 	struct ceph_inode_info *ci = ceph_inode(inode);
1117 	struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1118 	struct ceph_cap_flush *prealloc_cf = NULL;
1119 	struct ceph_buffer *old_blob = NULL;
1120 	int issued;
1121 	int err;
1122 	int dirty = 0;
1123 	int name_len = strlen(name);
1124 	int val_len = size;
1125 	char *newname = NULL;
1126 	char *newval = NULL;
1127 	struct ceph_inode_xattr *xattr = NULL;
1128 	int required_blob_size;
1129 	bool check_realm = false;
1130 	bool lock_snap_rwsem = false;
1131 
1132 	if (ceph_snap(inode) != CEPH_NOSNAP)
1133 		return -EROFS;
1134 
1135 	vxattr = ceph_match_vxattr(inode, name);
1136 	if (vxattr) {
1137 		if (vxattr->flags & VXATTR_FLAG_READONLY)
1138 			return -EOPNOTSUPP;
1139 		if (value && !strncmp(vxattr->name, "ceph.quota", 10))
1140 			check_realm = true;
1141 	}
1142 
1143 	/* pass any unhandled ceph.* xattrs through to the MDS */
1144 	if (!strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
1145 		goto do_sync_unlocked;
1146 
1147 	/* preallocate memory for xattr name, value, index node */
1148 	err = -ENOMEM;
1149 	newname = kmemdup(name, name_len + 1, GFP_NOFS);
1150 	if (!newname)
1151 		goto out;
1152 
1153 	if (val_len) {
1154 		newval = kmemdup(value, val_len, GFP_NOFS);
1155 		if (!newval)
1156 			goto out;
1157 	}
1158 
1159 	xattr = kmalloc(sizeof(struct ceph_inode_xattr), GFP_NOFS);
1160 	if (!xattr)
1161 		goto out;
1162 
1163 	prealloc_cf = ceph_alloc_cap_flush();
1164 	if (!prealloc_cf)
1165 		goto out;
1166 
1167 	spin_lock(&ci->i_ceph_lock);
1168 retry:
1169 	issued = __ceph_caps_issued(ci, NULL);
1170 	if (ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))
1171 		goto do_sync;
1172 
1173 	if (!lock_snap_rwsem && !ci->i_head_snapc) {
1174 		lock_snap_rwsem = true;
1175 		if (!down_read_trylock(&mdsc->snap_rwsem)) {
1176 			spin_unlock(&ci->i_ceph_lock);
1177 			down_read(&mdsc->snap_rwsem);
1178 			spin_lock(&ci->i_ceph_lock);
1179 			goto retry;
1180 		}
1181 	}
1182 
1183 	dout("setxattr %p name '%s' issued %s\n", inode, name,
1184 	     ceph_cap_string(issued));
1185 	__build_xattrs(inode);
1186 
1187 	required_blob_size = __get_required_blob_size(ci, name_len, val_len);
1188 
1189 	if (!ci->i_xattrs.prealloc_blob ||
1190 	    required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
1191 		struct ceph_buffer *blob;
1192 
1193 		spin_unlock(&ci->i_ceph_lock);
1194 		ceph_buffer_put(old_blob); /* Shouldn't be required */
1195 		dout(" pre-allocating new blob size=%d\n", required_blob_size);
1196 		blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
1197 		if (!blob)
1198 			goto do_sync_unlocked;
1199 		spin_lock(&ci->i_ceph_lock);
1200 		/* prealloc_blob can't be released while holding i_ceph_lock */
1201 		if (ci->i_xattrs.prealloc_blob)
1202 			old_blob = ci->i_xattrs.prealloc_blob;
1203 		ci->i_xattrs.prealloc_blob = blob;
1204 		goto retry;
1205 	}
1206 
1207 	err = __set_xattr(ci, newname, name_len, newval, val_len,
1208 			  flags, value ? 1 : -1, &xattr);
1209 
1210 	if (!err) {
1211 		dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL,
1212 					       &prealloc_cf);
1213 		ci->i_xattrs.dirty = true;
1214 		inode->i_ctime = current_time(inode);
1215 	}
1216 
1217 	spin_unlock(&ci->i_ceph_lock);
1218 	ceph_buffer_put(old_blob);
1219 	if (lock_snap_rwsem)
1220 		up_read(&mdsc->snap_rwsem);
1221 	if (dirty)
1222 		__mark_inode_dirty(inode, dirty);
1223 	ceph_free_cap_flush(prealloc_cf);
1224 	return err;
1225 
1226 do_sync:
1227 	spin_unlock(&ci->i_ceph_lock);
1228 do_sync_unlocked:
1229 	if (lock_snap_rwsem)
1230 		up_read(&mdsc->snap_rwsem);
1231 
1232 	/* security module set xattr while filling trace */
1233 	if (current->journal_info) {
1234 		pr_warn_ratelimited("sync setxattr %p "
1235 				    "during filling trace\n", inode);
1236 		err = -EBUSY;
1237 	} else {
1238 		err = ceph_sync_setxattr(inode, name, value, size, flags);
1239 		if (err >= 0 && check_realm) {
1240 			/* check if snaprealm was created for quota inode */
1241 			spin_lock(&ci->i_ceph_lock);
1242 			if ((ci->i_max_files || ci->i_max_bytes) &&
1243 			    !(ci->i_snap_realm &&
1244 			      ci->i_snap_realm->ino == ci->i_vino.ino))
1245 				err = -EOPNOTSUPP;
1246 			spin_unlock(&ci->i_ceph_lock);
1247 		}
1248 	}
1249 out:
1250 	ceph_free_cap_flush(prealloc_cf);
1251 	kfree(newname);
1252 	kfree(newval);
1253 	kfree(xattr);
1254 	return err;
1255 }
1256 
1257 static int ceph_get_xattr_handler(const struct xattr_handler *handler,
1258 				  struct dentry *dentry, struct inode *inode,
1259 				  const char *name, void *value, size_t size)
1260 {
1261 	if (!ceph_is_valid_xattr(name))
1262 		return -EOPNOTSUPP;
1263 	return __ceph_getxattr(inode, name, value, size);
1264 }
1265 
1266 static int ceph_set_xattr_handler(const struct xattr_handler *handler,
1267 				  struct user_namespace *mnt_userns,
1268 				  struct dentry *unused, struct inode *inode,
1269 				  const char *name, const void *value,
1270 				  size_t size, int flags)
1271 {
1272 	if (!ceph_is_valid_xattr(name))
1273 		return -EOPNOTSUPP;
1274 	return __ceph_setxattr(inode, name, value, size, flags);
1275 }
1276 
1277 static const struct xattr_handler ceph_other_xattr_handler = {
1278 	.prefix = "",  /* match any name => handlers called with full name */
1279 	.get = ceph_get_xattr_handler,
1280 	.set = ceph_set_xattr_handler,
1281 };
1282 
1283 #ifdef CONFIG_SECURITY
1284 bool ceph_security_xattr_wanted(struct inode *in)
1285 {
1286 	return in->i_security != NULL;
1287 }
1288 
1289 bool ceph_security_xattr_deadlock(struct inode *in)
1290 {
1291 	struct ceph_inode_info *ci;
1292 	bool ret;
1293 	if (!in->i_security)
1294 		return false;
1295 	ci = ceph_inode(in);
1296 	spin_lock(&ci->i_ceph_lock);
1297 	ret = !(ci->i_ceph_flags & CEPH_I_SEC_INITED) &&
1298 	      !(ci->i_xattrs.version > 0 &&
1299 		__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 0));
1300 	spin_unlock(&ci->i_ceph_lock);
1301 	return ret;
1302 }
1303 
1304 #ifdef CONFIG_CEPH_FS_SECURITY_LABEL
1305 int ceph_security_init_secctx(struct dentry *dentry, umode_t mode,
1306 			   struct ceph_acl_sec_ctx *as_ctx)
1307 {
1308 	struct ceph_pagelist *pagelist = as_ctx->pagelist;
1309 	const char *name;
1310 	size_t name_len;
1311 	int err;
1312 
1313 	err = security_dentry_init_security(dentry, mode, &dentry->d_name,
1314 					    &name, &as_ctx->sec_ctx,
1315 					    &as_ctx->sec_ctxlen);
1316 	if (err < 0) {
1317 		WARN_ON_ONCE(err != -EOPNOTSUPP);
1318 		err = 0; /* do nothing */
1319 		goto out;
1320 	}
1321 
1322 	err = -ENOMEM;
1323 	if (!pagelist) {
1324 		pagelist = ceph_pagelist_alloc(GFP_KERNEL);
1325 		if (!pagelist)
1326 			goto out;
1327 		err = ceph_pagelist_reserve(pagelist, PAGE_SIZE);
1328 		if (err)
1329 			goto out;
1330 		ceph_pagelist_encode_32(pagelist, 1);
1331 	}
1332 
1333 	/*
1334 	 * FIXME: Make security_dentry_init_security() generic. Currently
1335 	 * It only supports single security module and only selinux has
1336 	 * dentry_init_security hook.
1337 	 */
1338 	name_len = strlen(name);
1339 	err = ceph_pagelist_reserve(pagelist,
1340 				    4 * 2 + name_len + as_ctx->sec_ctxlen);
1341 	if (err)
1342 		goto out;
1343 
1344 	if (as_ctx->pagelist) {
1345 		/* update count of KV pairs */
1346 		BUG_ON(pagelist->length <= sizeof(__le32));
1347 		if (list_is_singular(&pagelist->head)) {
1348 			le32_add_cpu((__le32*)pagelist->mapped_tail, 1);
1349 		} else {
1350 			struct page *page = list_first_entry(&pagelist->head,
1351 							     struct page, lru);
1352 			void *addr = kmap_atomic(page);
1353 			le32_add_cpu((__le32*)addr, 1);
1354 			kunmap_atomic(addr);
1355 		}
1356 	} else {
1357 		as_ctx->pagelist = pagelist;
1358 	}
1359 
1360 	ceph_pagelist_encode_32(pagelist, name_len);
1361 	ceph_pagelist_append(pagelist, name, name_len);
1362 
1363 	ceph_pagelist_encode_32(pagelist, as_ctx->sec_ctxlen);
1364 	ceph_pagelist_append(pagelist, as_ctx->sec_ctx, as_ctx->sec_ctxlen);
1365 
1366 	err = 0;
1367 out:
1368 	if (pagelist && !as_ctx->pagelist)
1369 		ceph_pagelist_release(pagelist);
1370 	return err;
1371 }
1372 #endif /* CONFIG_CEPH_FS_SECURITY_LABEL */
1373 #endif /* CONFIG_SECURITY */
1374 
1375 void ceph_release_acl_sec_ctx(struct ceph_acl_sec_ctx *as_ctx)
1376 {
1377 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1378 	posix_acl_release(as_ctx->acl);
1379 	posix_acl_release(as_ctx->default_acl);
1380 #endif
1381 #ifdef CONFIG_CEPH_FS_SECURITY_LABEL
1382 	security_release_secctx(as_ctx->sec_ctx, as_ctx->sec_ctxlen);
1383 #endif
1384 	if (as_ctx->pagelist)
1385 		ceph_pagelist_release(as_ctx->pagelist);
1386 }
1387 
1388 /*
1389  * List of handlers for synthetic system.* attributes. Other
1390  * attributes are handled directly.
1391  */
1392 const struct xattr_handler *ceph_xattr_handlers[] = {
1393 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1394 	&posix_acl_access_xattr_handler,
1395 	&posix_acl_default_xattr_handler,
1396 #endif
1397 	&ceph_other_xattr_handler,
1398 	NULL,
1399 };
1400