xref: /openbmc/linux/fs/ceph/xattr.c (revision a325f174) 
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->netfs.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->netfs.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->netfs.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->netfs.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->netfs.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 #if IS_ENABLED(CONFIG_FS_ENCRYPTION)
356 static bool ceph_vxattrcb_fscrypt_auth_exists(struct ceph_inode_info *ci)
357 {
358 	return ci->fscrypt_auth_len;
359 }
360 
361 static ssize_t ceph_vxattrcb_fscrypt_auth(struct ceph_inode_info *ci,
362 					  char *val, size_t size)
363 {
364 	if (size) {
365 		if (size < ci->fscrypt_auth_len)
366 			return -ERANGE;
367 		memcpy(val, ci->fscrypt_auth, ci->fscrypt_auth_len);
368 	}
369 	return ci->fscrypt_auth_len;
370 }
371 #endif /* CONFIG_FS_ENCRYPTION */
372 
373 #define CEPH_XATTR_NAME(_type, _name)	XATTR_CEPH_PREFIX #_type "." #_name
374 #define CEPH_XATTR_NAME2(_type, _name, _name2)	\
375 	XATTR_CEPH_PREFIX #_type "." #_name "." #_name2
376 
377 #define XATTR_NAME_CEPH(_type, _name, _flags)				\
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 = NULL,					\
383 		.flags = (VXATTR_FLAG_READONLY | _flags),		\
384 	}
385 #define XATTR_RSTAT_FIELD(_type, _name)			\
386 	XATTR_NAME_CEPH(_type, _name, VXATTR_FLAG_RSTAT)
387 #define XATTR_RSTAT_FIELD_UPDATABLE(_type, _name)			\
388 	{								\
389 		.name = CEPH_XATTR_NAME(_type, _name),			\
390 		.name_size = sizeof (CEPH_XATTR_NAME(_type, _name)),	\
391 		.getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name,	\
392 		.exists_cb = NULL,					\
393 		.flags = VXATTR_FLAG_RSTAT,				\
394 	}
395 #define XATTR_LAYOUT_FIELD(_type, _name, _field)			\
396 	{								\
397 		.name = CEPH_XATTR_NAME2(_type, _name, _field),	\
398 		.name_size = sizeof (CEPH_XATTR_NAME2(_type, _name, _field)), \
399 		.getxattr_cb = ceph_vxattrcb_ ## _name ## _ ## _field, \
400 		.exists_cb = ceph_vxattrcb_layout_exists,	\
401 		.flags = VXATTR_FLAG_HIDDEN,			\
402 	}
403 #define XATTR_QUOTA_FIELD(_type, _name)					\
404 	{								\
405 		.name = CEPH_XATTR_NAME(_type, _name),			\
406 		.name_size = sizeof(CEPH_XATTR_NAME(_type, _name)),	\
407 		.getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name,	\
408 		.exists_cb = ceph_vxattrcb_quota_exists,		\
409 		.flags = VXATTR_FLAG_HIDDEN,				\
410 	}
411 
412 static struct ceph_vxattr ceph_dir_vxattrs[] = {
413 	{
414 		.name = "ceph.dir.layout",
415 		.name_size = sizeof("ceph.dir.layout"),
416 		.getxattr_cb = ceph_vxattrcb_layout,
417 		.exists_cb = ceph_vxattrcb_layout_exists,
418 		.flags = VXATTR_FLAG_HIDDEN,
419 	},
420 	XATTR_LAYOUT_FIELD(dir, layout, stripe_unit),
421 	XATTR_LAYOUT_FIELD(dir, layout, stripe_count),
422 	XATTR_LAYOUT_FIELD(dir, layout, object_size),
423 	XATTR_LAYOUT_FIELD(dir, layout, pool),
424 	XATTR_LAYOUT_FIELD(dir, layout, pool_namespace),
425 	XATTR_NAME_CEPH(dir, entries, VXATTR_FLAG_DIRSTAT),
426 	XATTR_NAME_CEPH(dir, files, VXATTR_FLAG_DIRSTAT),
427 	XATTR_NAME_CEPH(dir, subdirs, VXATTR_FLAG_DIRSTAT),
428 	XATTR_RSTAT_FIELD(dir, rentries),
429 	XATTR_RSTAT_FIELD(dir, rfiles),
430 	XATTR_RSTAT_FIELD(dir, rsubdirs),
431 	XATTR_RSTAT_FIELD(dir, rsnaps),
432 	XATTR_RSTAT_FIELD(dir, rbytes),
433 	XATTR_RSTAT_FIELD_UPDATABLE(dir, rctime),
434 	{
435 		.name = "ceph.dir.pin",
436 		.name_size = sizeof("ceph.dir.pin"),
437 		.getxattr_cb = ceph_vxattrcb_dir_pin,
438 		.exists_cb = ceph_vxattrcb_dir_pin_exists,
439 		.flags = VXATTR_FLAG_HIDDEN,
440 	},
441 	{
442 		.name = "ceph.quota",
443 		.name_size = sizeof("ceph.quota"),
444 		.getxattr_cb = ceph_vxattrcb_quota,
445 		.exists_cb = ceph_vxattrcb_quota_exists,
446 		.flags = VXATTR_FLAG_HIDDEN,
447 	},
448 	XATTR_QUOTA_FIELD(quota, max_bytes),
449 	XATTR_QUOTA_FIELD(quota, max_files),
450 	{
451 		.name = "ceph.snap.btime",
452 		.name_size = sizeof("ceph.snap.btime"),
453 		.getxattr_cb = ceph_vxattrcb_snap_btime,
454 		.exists_cb = ceph_vxattrcb_snap_btime_exists,
455 		.flags = VXATTR_FLAG_READONLY,
456 	},
457 	{
458 		.name = "ceph.caps",
459 		.name_size = sizeof("ceph.caps"),
460 		.getxattr_cb = ceph_vxattrcb_caps,
461 		.exists_cb = NULL,
462 		.flags = VXATTR_FLAG_HIDDEN,
463 	},
464 	{ .name = NULL, 0 }	/* Required table terminator */
465 };
466 
467 /* files */
468 
469 static struct ceph_vxattr ceph_file_vxattrs[] = {
470 	{
471 		.name = "ceph.file.layout",
472 		.name_size = sizeof("ceph.file.layout"),
473 		.getxattr_cb = ceph_vxattrcb_layout,
474 		.exists_cb = ceph_vxattrcb_layout_exists,
475 		.flags = VXATTR_FLAG_HIDDEN,
476 	},
477 	XATTR_LAYOUT_FIELD(file, layout, stripe_unit),
478 	XATTR_LAYOUT_FIELD(file, layout, stripe_count),
479 	XATTR_LAYOUT_FIELD(file, layout, object_size),
480 	XATTR_LAYOUT_FIELD(file, layout, pool),
481 	XATTR_LAYOUT_FIELD(file, layout, pool_namespace),
482 	{
483 		.name = "ceph.snap.btime",
484 		.name_size = sizeof("ceph.snap.btime"),
485 		.getxattr_cb = ceph_vxattrcb_snap_btime,
486 		.exists_cb = ceph_vxattrcb_snap_btime_exists,
487 		.flags = VXATTR_FLAG_READONLY,
488 	},
489 	{
490 		.name = "ceph.caps",
491 		.name_size = sizeof("ceph.caps"),
492 		.getxattr_cb = ceph_vxattrcb_caps,
493 		.exists_cb = NULL,
494 		.flags = VXATTR_FLAG_HIDDEN,
495 	},
496 	{ .name = NULL, 0 }	/* Required table terminator */
497 };
498 
499 static struct ceph_vxattr ceph_common_vxattrs[] = {
500 	{
501 		.name = "ceph.cluster_fsid",
502 		.name_size = sizeof("ceph.cluster_fsid"),
503 		.getxattr_cb = ceph_vxattrcb_cluster_fsid,
504 		.exists_cb = NULL,
505 		.flags = VXATTR_FLAG_READONLY,
506 	},
507 	{
508 		.name = "ceph.client_id",
509 		.name_size = sizeof("ceph.client_id"),
510 		.getxattr_cb = ceph_vxattrcb_client_id,
511 		.exists_cb = NULL,
512 		.flags = VXATTR_FLAG_READONLY,
513 	},
514 	{
515 		.name = "ceph.auth_mds",
516 		.name_size = sizeof("ceph.auth_mds"),
517 		.getxattr_cb = ceph_vxattrcb_auth_mds,
518 		.exists_cb = NULL,
519 		.flags = VXATTR_FLAG_READONLY,
520 	},
521 #if IS_ENABLED(CONFIG_FS_ENCRYPTION)
522 	{
523 		.name = "ceph.fscrypt.auth",
524 		.name_size = sizeof("ceph.fscrypt.auth"),
525 		.getxattr_cb = ceph_vxattrcb_fscrypt_auth,
526 		.exists_cb = ceph_vxattrcb_fscrypt_auth_exists,
527 		.flags = VXATTR_FLAG_READONLY,
528 	},
529 #endif /* CONFIG_FS_ENCRYPTION */
530 	{ .name = NULL, 0 }	/* Required table terminator */
531 };
532 
533 static struct ceph_vxattr *ceph_inode_vxattrs(struct inode *inode)
534 {
535 	if (S_ISDIR(inode->i_mode))
536 		return ceph_dir_vxattrs;
537 	else if (S_ISREG(inode->i_mode))
538 		return ceph_file_vxattrs;
539 	return NULL;
540 }
541 
542 static struct ceph_vxattr *ceph_match_vxattr(struct inode *inode,
543 						const char *name)
544 {
545 	struct ceph_vxattr *vxattr = ceph_inode_vxattrs(inode);
546 
547 	if (vxattr) {
548 		while (vxattr->name) {
549 			if (!strcmp(vxattr->name, name))
550 				return vxattr;
551 			vxattr++;
552 		}
553 	}
554 
555 	vxattr = ceph_common_vxattrs;
556 	while (vxattr->name) {
557 		if (!strcmp(vxattr->name, name))
558 			return vxattr;
559 		vxattr++;
560 	}
561 
562 	return NULL;
563 }
564 
565 #define MAX_XATTR_VAL_PRINT_LEN 256
566 
567 static int __set_xattr(struct ceph_inode_info *ci,
568 			   const char *name, int name_len,
569 			   const char *val, int val_len,
570 			   int flags, int update_xattr,
571 			   struct ceph_inode_xattr **newxattr)
572 {
573 	struct rb_node **p;
574 	struct rb_node *parent = NULL;
575 	struct ceph_inode_xattr *xattr = NULL;
576 	int c;
577 	int new = 0;
578 
579 	p = &ci->i_xattrs.index.rb_node;
580 	while (*p) {
581 		parent = *p;
582 		xattr = rb_entry(parent, struct ceph_inode_xattr, node);
583 		c = strncmp(name, xattr->name, min(name_len, xattr->name_len));
584 		if (c < 0)
585 			p = &(*p)->rb_left;
586 		else if (c > 0)
587 			p = &(*p)->rb_right;
588 		else {
589 			if (name_len == xattr->name_len)
590 				break;
591 			else if (name_len < xattr->name_len)
592 				p = &(*p)->rb_left;
593 			else
594 				p = &(*p)->rb_right;
595 		}
596 		xattr = NULL;
597 	}
598 
599 	if (update_xattr) {
600 		int err = 0;
601 
602 		if (xattr && (flags & XATTR_CREATE))
603 			err = -EEXIST;
604 		else if (!xattr && (flags & XATTR_REPLACE))
605 			err = -ENODATA;
606 		if (err) {
607 			kfree(name);
608 			kfree(val);
609 			kfree(*newxattr);
610 			return err;
611 		}
612 		if (update_xattr < 0) {
613 			if (xattr)
614 				__remove_xattr(ci, xattr);
615 			kfree(name);
616 			kfree(*newxattr);
617 			return 0;
618 		}
619 	}
620 
621 	if (!xattr) {
622 		new = 1;
623 		xattr = *newxattr;
624 		xattr->name = name;
625 		xattr->name_len = name_len;
626 		xattr->should_free_name = update_xattr;
627 
628 		ci->i_xattrs.count++;
629 		dout("%s count=%d\n", __func__, ci->i_xattrs.count);
630 	} else {
631 		kfree(*newxattr);
632 		*newxattr = NULL;
633 		if (xattr->should_free_val)
634 			kfree(xattr->val);
635 
636 		if (update_xattr) {
637 			kfree(name);
638 			name = xattr->name;
639 		}
640 		ci->i_xattrs.names_size -= xattr->name_len;
641 		ci->i_xattrs.vals_size -= xattr->val_len;
642 	}
643 	ci->i_xattrs.names_size += name_len;
644 	ci->i_xattrs.vals_size += val_len;
645 	if (val)
646 		xattr->val = val;
647 	else
648 		xattr->val = "";
649 
650 	xattr->val_len = val_len;
651 	xattr->dirty = update_xattr;
652 	xattr->should_free_val = (val && update_xattr);
653 
654 	if (new) {
655 		rb_link_node(&xattr->node, parent, p);
656 		rb_insert_color(&xattr->node, &ci->i_xattrs.index);
657 		dout("%s p=%p\n", __func__, p);
658 	}
659 
660 	dout("%s added %llx.%llx xattr %p %.*s=%.*s%s\n", __func__,
661 	     ceph_vinop(&ci->netfs.inode), xattr, name_len, name,
662 	     min(val_len, MAX_XATTR_VAL_PRINT_LEN), val,
663 	     val_len > MAX_XATTR_VAL_PRINT_LEN ? "..." : "");
664 
665 	return 0;
666 }
667 
668 static struct ceph_inode_xattr *__get_xattr(struct ceph_inode_info *ci,
669 			   const char *name)
670 {
671 	struct rb_node **p;
672 	struct rb_node *parent = NULL;
673 	struct ceph_inode_xattr *xattr = NULL;
674 	int name_len = strlen(name);
675 	int c;
676 
677 	p = &ci->i_xattrs.index.rb_node;
678 	while (*p) {
679 		parent = *p;
680 		xattr = rb_entry(parent, struct ceph_inode_xattr, node);
681 		c = strncmp(name, xattr->name, xattr->name_len);
682 		if (c == 0 && name_len > xattr->name_len)
683 			c = 1;
684 		if (c < 0)
685 			p = &(*p)->rb_left;
686 		else if (c > 0)
687 			p = &(*p)->rb_right;
688 		else {
689 			int len = min(xattr->val_len, MAX_XATTR_VAL_PRINT_LEN);
690 
691 			dout("%s %s: found %.*s%s\n", __func__, name, len,
692 			     xattr->val, xattr->val_len > len ? "..." : "");
693 			return xattr;
694 		}
695 	}
696 
697 	dout("%s %s: not found\n", __func__, name);
698 
699 	return NULL;
700 }
701 
702 static void __free_xattr(struct ceph_inode_xattr *xattr)
703 {
704 	BUG_ON(!xattr);
705 
706 	if (xattr->should_free_name)
707 		kfree(xattr->name);
708 	if (xattr->should_free_val)
709 		kfree(xattr->val);
710 
711 	kfree(xattr);
712 }
713 
714 static int __remove_xattr(struct ceph_inode_info *ci,
715 			  struct ceph_inode_xattr *xattr)
716 {
717 	if (!xattr)
718 		return -ENODATA;
719 
720 	rb_erase(&xattr->node, &ci->i_xattrs.index);
721 
722 	if (xattr->should_free_name)
723 		kfree(xattr->name);
724 	if (xattr->should_free_val)
725 		kfree(xattr->val);
726 
727 	ci->i_xattrs.names_size -= xattr->name_len;
728 	ci->i_xattrs.vals_size -= xattr->val_len;
729 	ci->i_xattrs.count--;
730 	kfree(xattr);
731 
732 	return 0;
733 }
734 
735 static char *__copy_xattr_names(struct ceph_inode_info *ci,
736 				char *dest)
737 {
738 	struct rb_node *p;
739 	struct ceph_inode_xattr *xattr = NULL;
740 
741 	p = rb_first(&ci->i_xattrs.index);
742 	dout("__copy_xattr_names count=%d\n", ci->i_xattrs.count);
743 
744 	while (p) {
745 		xattr = rb_entry(p, struct ceph_inode_xattr, node);
746 		memcpy(dest, xattr->name, xattr->name_len);
747 		dest[xattr->name_len] = '\0';
748 
749 		dout("dest=%s %p (%s) (%d/%d)\n", dest, xattr, xattr->name,
750 		     xattr->name_len, ci->i_xattrs.names_size);
751 
752 		dest += xattr->name_len + 1;
753 		p = rb_next(p);
754 	}
755 
756 	return dest;
757 }
758 
759 void __ceph_destroy_xattrs(struct ceph_inode_info *ci)
760 {
761 	struct rb_node *p, *tmp;
762 	struct ceph_inode_xattr *xattr = NULL;
763 
764 	p = rb_first(&ci->i_xattrs.index);
765 
766 	dout("__ceph_destroy_xattrs p=%p\n", p);
767 
768 	while (p) {
769 		xattr = rb_entry(p, struct ceph_inode_xattr, node);
770 		tmp = p;
771 		p = rb_next(tmp);
772 		dout("__ceph_destroy_xattrs next p=%p (%.*s)\n", p,
773 		     xattr->name_len, xattr->name);
774 		rb_erase(tmp, &ci->i_xattrs.index);
775 
776 		__free_xattr(xattr);
777 	}
778 
779 	ci->i_xattrs.names_size = 0;
780 	ci->i_xattrs.vals_size = 0;
781 	ci->i_xattrs.index_version = 0;
782 	ci->i_xattrs.count = 0;
783 	ci->i_xattrs.index = RB_ROOT;
784 }
785 
786 static int __build_xattrs(struct inode *inode)
787 	__releases(ci->i_ceph_lock)
788 	__acquires(ci->i_ceph_lock)
789 {
790 	u32 namelen;
791 	u32 numattr = 0;
792 	void *p, *end;
793 	u32 len;
794 	const char *name, *val;
795 	struct ceph_inode_info *ci = ceph_inode(inode);
796 	u64 xattr_version;
797 	struct ceph_inode_xattr **xattrs = NULL;
798 	int err = 0;
799 	int i;
800 
801 	dout("__build_xattrs() len=%d\n",
802 	     ci->i_xattrs.blob ? (int)ci->i_xattrs.blob->vec.iov_len : 0);
803 
804 	if (ci->i_xattrs.index_version >= ci->i_xattrs.version)
805 		return 0; /* already built */
806 
807 	__ceph_destroy_xattrs(ci);
808 
809 start:
810 	/* updated internal xattr rb tree */
811 	if (ci->i_xattrs.blob && ci->i_xattrs.blob->vec.iov_len > 4) {
812 		p = ci->i_xattrs.blob->vec.iov_base;
813 		end = p + ci->i_xattrs.blob->vec.iov_len;
814 		ceph_decode_32_safe(&p, end, numattr, bad);
815 		xattr_version = ci->i_xattrs.version;
816 		spin_unlock(&ci->i_ceph_lock);
817 
818 		xattrs = kcalloc(numattr, sizeof(struct ceph_inode_xattr *),
819 				 GFP_NOFS);
820 		err = -ENOMEM;
821 		if (!xattrs)
822 			goto bad_lock;
823 
824 		for (i = 0; i < numattr; i++) {
825 			xattrs[i] = kmalloc(sizeof(struct ceph_inode_xattr),
826 					    GFP_NOFS);
827 			if (!xattrs[i])
828 				goto bad_lock;
829 		}
830 
831 		spin_lock(&ci->i_ceph_lock);
832 		if (ci->i_xattrs.version != xattr_version) {
833 			/* lost a race, retry */
834 			for (i = 0; i < numattr; i++)
835 				kfree(xattrs[i]);
836 			kfree(xattrs);
837 			xattrs = NULL;
838 			goto start;
839 		}
840 		err = -EIO;
841 		while (numattr--) {
842 			ceph_decode_32_safe(&p, end, len, bad);
843 			namelen = len;
844 			name = p;
845 			p += len;
846 			ceph_decode_32_safe(&p, end, len, bad);
847 			val = p;
848 			p += len;
849 
850 			err = __set_xattr(ci, name, namelen, val, len,
851 					  0, 0, &xattrs[numattr]);
852 
853 			if (err < 0)
854 				goto bad;
855 		}
856 		kfree(xattrs);
857 	}
858 	ci->i_xattrs.index_version = ci->i_xattrs.version;
859 	ci->i_xattrs.dirty = false;
860 
861 	return err;
862 bad_lock:
863 	spin_lock(&ci->i_ceph_lock);
864 bad:
865 	if (xattrs) {
866 		for (i = 0; i < numattr; i++)
867 			kfree(xattrs[i]);
868 		kfree(xattrs);
869 	}
870 	ci->i_xattrs.names_size = 0;
871 	return err;
872 }
873 
874 static int __get_required_blob_size(struct ceph_inode_info *ci, int name_size,
875 				    int val_size)
876 {
877 	/*
878 	 * 4 bytes for the length, and additional 4 bytes per each xattr name,
879 	 * 4 bytes per each value
880 	 */
881 	int size = 4 + ci->i_xattrs.count*(4 + 4) +
882 			     ci->i_xattrs.names_size +
883 			     ci->i_xattrs.vals_size;
884 	dout("__get_required_blob_size c=%d names.size=%d vals.size=%d\n",
885 	     ci->i_xattrs.count, ci->i_xattrs.names_size,
886 	     ci->i_xattrs.vals_size);
887 
888 	if (name_size)
889 		size += 4 + 4 + name_size + val_size;
890 
891 	return size;
892 }
893 
894 /*
895  * If there are dirty xattrs, reencode xattrs into the prealloc_blob
896  * and swap into place.  It returns the old i_xattrs.blob (or NULL) so
897  * that it can be freed by the caller as the i_ceph_lock is likely to be
898  * held.
899  */
900 struct ceph_buffer *__ceph_build_xattrs_blob(struct ceph_inode_info *ci)
901 {
902 	struct rb_node *p;
903 	struct ceph_inode_xattr *xattr = NULL;
904 	struct ceph_buffer *old_blob = NULL;
905 	void *dest;
906 
907 	dout("__build_xattrs_blob %p\n", &ci->netfs.inode);
908 	if (ci->i_xattrs.dirty) {
909 		int need = __get_required_blob_size(ci, 0, 0);
910 
911 		BUG_ON(need > ci->i_xattrs.prealloc_blob->alloc_len);
912 
913 		p = rb_first(&ci->i_xattrs.index);
914 		dest = ci->i_xattrs.prealloc_blob->vec.iov_base;
915 
916 		ceph_encode_32(&dest, ci->i_xattrs.count);
917 		while (p) {
918 			xattr = rb_entry(p, struct ceph_inode_xattr, node);
919 
920 			ceph_encode_32(&dest, xattr->name_len);
921 			memcpy(dest, xattr->name, xattr->name_len);
922 			dest += xattr->name_len;
923 			ceph_encode_32(&dest, xattr->val_len);
924 			memcpy(dest, xattr->val, xattr->val_len);
925 			dest += xattr->val_len;
926 
927 			p = rb_next(p);
928 		}
929 
930 		/* adjust buffer len; it may be larger than we need */
931 		ci->i_xattrs.prealloc_blob->vec.iov_len =
932 			dest - ci->i_xattrs.prealloc_blob->vec.iov_base;
933 
934 		if (ci->i_xattrs.blob)
935 			old_blob = ci->i_xattrs.blob;
936 		ci->i_xattrs.blob = ci->i_xattrs.prealloc_blob;
937 		ci->i_xattrs.prealloc_blob = NULL;
938 		ci->i_xattrs.dirty = false;
939 		ci->i_xattrs.version++;
940 	}
941 
942 	return old_blob;
943 }
944 
945 static inline int __get_request_mask(struct inode *in) {
946 	struct ceph_mds_request *req = current->journal_info;
947 	int mask = 0;
948 	if (req && req->r_target_inode == in) {
949 		if (req->r_op == CEPH_MDS_OP_LOOKUP ||
950 		    req->r_op == CEPH_MDS_OP_LOOKUPINO ||
951 		    req->r_op == CEPH_MDS_OP_LOOKUPPARENT ||
952 		    req->r_op == CEPH_MDS_OP_GETATTR) {
953 			mask = le32_to_cpu(req->r_args.getattr.mask);
954 		} else if (req->r_op == CEPH_MDS_OP_OPEN ||
955 			   req->r_op == CEPH_MDS_OP_CREATE) {
956 			mask = le32_to_cpu(req->r_args.open.mask);
957 		}
958 	}
959 	return mask;
960 }
961 
962 ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,
963 		      size_t size)
964 {
965 	struct ceph_inode_info *ci = ceph_inode(inode);
966 	struct ceph_inode_xattr *xattr;
967 	struct ceph_vxattr *vxattr;
968 	int req_mask;
969 	ssize_t err;
970 
971 	if (strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
972 		goto handle_non_vxattrs;
973 
974 	/* let's see if a virtual xattr was requested */
975 	vxattr = ceph_match_vxattr(inode, name);
976 	if (vxattr) {
977 		int mask = 0;
978 		if (vxattr->flags & VXATTR_FLAG_RSTAT)
979 			mask |= CEPH_STAT_RSTAT;
980 		if (vxattr->flags & VXATTR_FLAG_DIRSTAT)
981 			mask |= CEPH_CAP_FILE_SHARED;
982 		err = ceph_do_getattr(inode, mask, true);
983 		if (err)
984 			return err;
985 		err = -ENODATA;
986 		if (!(vxattr->exists_cb && !vxattr->exists_cb(ci))) {
987 			err = vxattr->getxattr_cb(ci, value, size);
988 			if (size && size < err)
989 				err = -ERANGE;
990 		}
991 		return err;
992 	} else {
993 		err = ceph_do_getvxattr(inode, name, value, size);
994 		/* this would happen with a new client and old server combo */
995 		if (err == -EOPNOTSUPP)
996 			err = -ENODATA;
997 		return err;
998 	}
999 handle_non_vxattrs:
1000 	req_mask = __get_request_mask(inode);
1001 
1002 	spin_lock(&ci->i_ceph_lock);
1003 	dout("getxattr %p name '%s' ver=%lld index_ver=%lld\n", inode, name,
1004 	     ci->i_xattrs.version, ci->i_xattrs.index_version);
1005 
1006 	if (ci->i_xattrs.version == 0 ||
1007 	    !((req_mask & CEPH_CAP_XATTR_SHARED) ||
1008 	      __ceph_caps_issued_mask_metric(ci, CEPH_CAP_XATTR_SHARED, 1))) {
1009 		spin_unlock(&ci->i_ceph_lock);
1010 
1011 		/* security module gets xattr while filling trace */
1012 		if (current->journal_info) {
1013 			pr_warn_ratelimited("sync getxattr %p "
1014 					    "during filling trace\n", inode);
1015 			return -EBUSY;
1016 		}
1017 
1018 		/* get xattrs from mds (if we don't already have them) */
1019 		err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
1020 		if (err)
1021 			return err;
1022 		spin_lock(&ci->i_ceph_lock);
1023 	}
1024 
1025 	err = __build_xattrs(inode);
1026 	if (err < 0)
1027 		goto out;
1028 
1029 	err = -ENODATA;  /* == ENOATTR */
1030 	xattr = __get_xattr(ci, name);
1031 	if (!xattr)
1032 		goto out;
1033 
1034 	err = -ERANGE;
1035 	if (size && size < xattr->val_len)
1036 		goto out;
1037 
1038 	err = xattr->val_len;
1039 	if (size == 0)
1040 		goto out;
1041 
1042 	memcpy(value, xattr->val, xattr->val_len);
1043 
1044 	if (current->journal_info &&
1045 	    !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN) &&
1046 	    security_ismaclabel(name + XATTR_SECURITY_PREFIX_LEN))
1047 		ci->i_ceph_flags |= CEPH_I_SEC_INITED;
1048 out:
1049 	spin_unlock(&ci->i_ceph_lock);
1050 	return err;
1051 }
1052 
1053 ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)
1054 {
1055 	struct inode *inode = d_inode(dentry);
1056 	struct ceph_inode_info *ci = ceph_inode(inode);
1057 	bool len_only = (size == 0);
1058 	u32 namelen;
1059 	int err;
1060 
1061 	spin_lock(&ci->i_ceph_lock);
1062 	dout("listxattr %p ver=%lld index_ver=%lld\n", inode,
1063 	     ci->i_xattrs.version, ci->i_xattrs.index_version);
1064 
1065 	if (ci->i_xattrs.version == 0 ||
1066 	    !__ceph_caps_issued_mask_metric(ci, CEPH_CAP_XATTR_SHARED, 1)) {
1067 		spin_unlock(&ci->i_ceph_lock);
1068 		err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
1069 		if (err)
1070 			return err;
1071 		spin_lock(&ci->i_ceph_lock);
1072 	}
1073 
1074 	err = __build_xattrs(inode);
1075 	if (err < 0)
1076 		goto out;
1077 
1078 	/* add 1 byte for each xattr due to the null termination */
1079 	namelen = ci->i_xattrs.names_size + ci->i_xattrs.count;
1080 	if (!len_only) {
1081 		if (namelen > size) {
1082 			err = -ERANGE;
1083 			goto out;
1084 		}
1085 		names = __copy_xattr_names(ci, names);
1086 		size -= namelen;
1087 	}
1088 	err = namelen;
1089 out:
1090 	spin_unlock(&ci->i_ceph_lock);
1091 	return err;
1092 }
1093 
1094 static int ceph_sync_setxattr(struct inode *inode, const char *name,
1095 			      const char *value, size_t size, int flags)
1096 {
1097 	struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1098 	struct ceph_inode_info *ci = ceph_inode(inode);
1099 	struct ceph_mds_request *req;
1100 	struct ceph_mds_client *mdsc = fsc->mdsc;
1101 	struct ceph_osd_client *osdc = &fsc->client->osdc;
1102 	struct ceph_pagelist *pagelist = NULL;
1103 	int op = CEPH_MDS_OP_SETXATTR;
1104 	int err;
1105 
1106 	if (size > 0) {
1107 		/* copy value into pagelist */
1108 		pagelist = ceph_pagelist_alloc(GFP_NOFS);
1109 		if (!pagelist)
1110 			return -ENOMEM;
1111 
1112 		err = ceph_pagelist_append(pagelist, value, size);
1113 		if (err)
1114 			goto out;
1115 	} else if (!value) {
1116 		if (flags & CEPH_XATTR_REPLACE)
1117 			op = CEPH_MDS_OP_RMXATTR;
1118 		else
1119 			flags |= CEPH_XATTR_REMOVE;
1120 	}
1121 
1122 	dout("setxattr value size: %zu\n", size);
1123 
1124 	/* do request */
1125 	req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1126 	if (IS_ERR(req)) {
1127 		err = PTR_ERR(req);
1128 		goto out;
1129 	}
1130 
1131 	req->r_path2 = kstrdup(name, GFP_NOFS);
1132 	if (!req->r_path2) {
1133 		ceph_mdsc_put_request(req);
1134 		err = -ENOMEM;
1135 		goto out;
1136 	}
1137 
1138 	if (op == CEPH_MDS_OP_SETXATTR) {
1139 		req->r_args.setxattr.flags = cpu_to_le32(flags);
1140 		req->r_args.setxattr.osdmap_epoch =
1141 			cpu_to_le32(osdc->osdmap->epoch);
1142 		req->r_pagelist = pagelist;
1143 		pagelist = NULL;
1144 	}
1145 
1146 	req->r_inode = inode;
1147 	ihold(inode);
1148 	req->r_num_caps = 1;
1149 	req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
1150 
1151 	dout("xattr.ver (before): %lld\n", ci->i_xattrs.version);
1152 	err = ceph_mdsc_do_request(mdsc, NULL, req);
1153 	ceph_mdsc_put_request(req);
1154 	dout("xattr.ver (after): %lld\n", ci->i_xattrs.version);
1155 
1156 out:
1157 	if (pagelist)
1158 		ceph_pagelist_release(pagelist);
1159 	return err;
1160 }
1161 
1162 int __ceph_setxattr(struct inode *inode, const char *name,
1163 			const void *value, size_t size, int flags)
1164 {
1165 	struct ceph_vxattr *vxattr;
1166 	struct ceph_inode_info *ci = ceph_inode(inode);
1167 	struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1168 	struct ceph_cap_flush *prealloc_cf = NULL;
1169 	struct ceph_buffer *old_blob = NULL;
1170 	int issued;
1171 	int err;
1172 	int dirty = 0;
1173 	int name_len = strlen(name);
1174 	int val_len = size;
1175 	char *newname = NULL;
1176 	char *newval = NULL;
1177 	struct ceph_inode_xattr *xattr = NULL;
1178 	int required_blob_size;
1179 	bool check_realm = false;
1180 	bool lock_snap_rwsem = false;
1181 
1182 	if (ceph_snap(inode) != CEPH_NOSNAP)
1183 		return -EROFS;
1184 
1185 	vxattr = ceph_match_vxattr(inode, name);
1186 	if (vxattr) {
1187 		if (vxattr->flags & VXATTR_FLAG_READONLY)
1188 			return -EOPNOTSUPP;
1189 		if (value && !strncmp(vxattr->name, "ceph.quota", 10))
1190 			check_realm = true;
1191 	}
1192 
1193 	/* pass any unhandled ceph.* xattrs through to the MDS */
1194 	if (!strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
1195 		goto do_sync_unlocked;
1196 
1197 	/* preallocate memory for xattr name, value, index node */
1198 	err = -ENOMEM;
1199 	newname = kmemdup(name, name_len + 1, GFP_NOFS);
1200 	if (!newname)
1201 		goto out;
1202 
1203 	if (val_len) {
1204 		newval = kmemdup(value, val_len, GFP_NOFS);
1205 		if (!newval)
1206 			goto out;
1207 	}
1208 
1209 	xattr = kmalloc(sizeof(struct ceph_inode_xattr), GFP_NOFS);
1210 	if (!xattr)
1211 		goto out;
1212 
1213 	prealloc_cf = ceph_alloc_cap_flush();
1214 	if (!prealloc_cf)
1215 		goto out;
1216 
1217 	spin_lock(&ci->i_ceph_lock);
1218 retry:
1219 	issued = __ceph_caps_issued(ci, NULL);
1220 	required_blob_size = __get_required_blob_size(ci, name_len, val_len);
1221 	if ((ci->i_xattrs.version == 0) || !(issued & CEPH_CAP_XATTR_EXCL) ||
1222 	    (required_blob_size > mdsc->mdsmap->m_max_xattr_size)) {
1223 		dout("%s do sync setxattr: version: %llu size: %d max: %llu\n",
1224 		     __func__, ci->i_xattrs.version, required_blob_size,
1225 		     mdsc->mdsmap->m_max_xattr_size);
1226 		goto do_sync;
1227 	}
1228 
1229 	if (!lock_snap_rwsem && !ci->i_head_snapc) {
1230 		lock_snap_rwsem = true;
1231 		if (!down_read_trylock(&mdsc->snap_rwsem)) {
1232 			spin_unlock(&ci->i_ceph_lock);
1233 			down_read(&mdsc->snap_rwsem);
1234 			spin_lock(&ci->i_ceph_lock);
1235 			goto retry;
1236 		}
1237 	}
1238 
1239 	dout("setxattr %p name '%s' issued %s\n", inode, name,
1240 	     ceph_cap_string(issued));
1241 	__build_xattrs(inode);
1242 
1243 	if (!ci->i_xattrs.prealloc_blob ||
1244 	    required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
1245 		struct ceph_buffer *blob;
1246 
1247 		spin_unlock(&ci->i_ceph_lock);
1248 		ceph_buffer_put(old_blob); /* Shouldn't be required */
1249 		dout(" pre-allocating new blob size=%d\n", required_blob_size);
1250 		blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
1251 		if (!blob)
1252 			goto do_sync_unlocked;
1253 		spin_lock(&ci->i_ceph_lock);
1254 		/* prealloc_blob can't be released while holding i_ceph_lock */
1255 		if (ci->i_xattrs.prealloc_blob)
1256 			old_blob = ci->i_xattrs.prealloc_blob;
1257 		ci->i_xattrs.prealloc_blob = blob;
1258 		goto retry;
1259 	}
1260 
1261 	err = __set_xattr(ci, newname, name_len, newval, val_len,
1262 			  flags, value ? 1 : -1, &xattr);
1263 
1264 	if (!err) {
1265 		dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL,
1266 					       &prealloc_cf);
1267 		ci->i_xattrs.dirty = true;
1268 		inode_set_ctime_current(inode);
1269 	}
1270 
1271 	spin_unlock(&ci->i_ceph_lock);
1272 	ceph_buffer_put(old_blob);
1273 	if (lock_snap_rwsem)
1274 		up_read(&mdsc->snap_rwsem);
1275 	if (dirty)
1276 		__mark_inode_dirty(inode, dirty);
1277 	ceph_free_cap_flush(prealloc_cf);
1278 	return err;
1279 
1280 do_sync:
1281 	spin_unlock(&ci->i_ceph_lock);
1282 do_sync_unlocked:
1283 	if (lock_snap_rwsem)
1284 		up_read(&mdsc->snap_rwsem);
1285 
1286 	/* security module set xattr while filling trace */
1287 	if (current->journal_info) {
1288 		pr_warn_ratelimited("sync setxattr %p "
1289 				    "during filling trace\n", inode);
1290 		err = -EBUSY;
1291 	} else {
1292 		err = ceph_sync_setxattr(inode, name, value, size, flags);
1293 		if (err >= 0 && check_realm) {
1294 			/* check if snaprealm was created for quota inode */
1295 			spin_lock(&ci->i_ceph_lock);
1296 			if ((ci->i_max_files || ci->i_max_bytes) &&
1297 			    !(ci->i_snap_realm &&
1298 			      ci->i_snap_realm->ino == ci->i_vino.ino))
1299 				err = -EOPNOTSUPP;
1300 			spin_unlock(&ci->i_ceph_lock);
1301 		}
1302 	}
1303 out:
1304 	ceph_free_cap_flush(prealloc_cf);
1305 	kfree(newname);
1306 	kfree(newval);
1307 	kfree(xattr);
1308 	return err;
1309 }
1310 
1311 static int ceph_get_xattr_handler(const struct xattr_handler *handler,
1312 				  struct dentry *dentry, struct inode *inode,
1313 				  const char *name, void *value, size_t size)
1314 {
1315 	if (!ceph_is_valid_xattr(name))
1316 		return -EOPNOTSUPP;
1317 	return __ceph_getxattr(inode, name, value, size);
1318 }
1319 
1320 static int ceph_set_xattr_handler(const struct xattr_handler *handler,
1321 				  struct mnt_idmap *idmap,
1322 				  struct dentry *unused, struct inode *inode,
1323 				  const char *name, const void *value,
1324 				  size_t size, int flags)
1325 {
1326 	if (!ceph_is_valid_xattr(name))
1327 		return -EOPNOTSUPP;
1328 	return __ceph_setxattr(inode, name, value, size, flags);
1329 }
1330 
1331 static const struct xattr_handler ceph_other_xattr_handler = {
1332 	.prefix = "",  /* match any name => handlers called with full name */
1333 	.get = ceph_get_xattr_handler,
1334 	.set = ceph_set_xattr_handler,
1335 };
1336 
1337 #ifdef CONFIG_SECURITY
1338 bool ceph_security_xattr_wanted(struct inode *in)
1339 {
1340 	return in->i_security != NULL;
1341 }
1342 
1343 bool ceph_security_xattr_deadlock(struct inode *in)
1344 {
1345 	struct ceph_inode_info *ci;
1346 	bool ret;
1347 	if (!in->i_security)
1348 		return false;
1349 	ci = ceph_inode(in);
1350 	spin_lock(&ci->i_ceph_lock);
1351 	ret = !(ci->i_ceph_flags & CEPH_I_SEC_INITED) &&
1352 	      !(ci->i_xattrs.version > 0 &&
1353 		__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 0));
1354 	spin_unlock(&ci->i_ceph_lock);
1355 	return ret;
1356 }
1357 
1358 #ifdef CONFIG_CEPH_FS_SECURITY_LABEL
1359 int ceph_security_init_secctx(struct dentry *dentry, umode_t mode,
1360 			   struct ceph_acl_sec_ctx *as_ctx)
1361 {
1362 	struct ceph_pagelist *pagelist = as_ctx->pagelist;
1363 	const char *name;
1364 	size_t name_len;
1365 	int err;
1366 
1367 	err = security_dentry_init_security(dentry, mode, &dentry->d_name,
1368 					    &name, &as_ctx->sec_ctx,
1369 					    &as_ctx->sec_ctxlen);
1370 	if (err < 0) {
1371 		WARN_ON_ONCE(err != -EOPNOTSUPP);
1372 		err = 0; /* do nothing */
1373 		goto out;
1374 	}
1375 
1376 	err = -ENOMEM;
1377 	if (!pagelist) {
1378 		pagelist = ceph_pagelist_alloc(GFP_KERNEL);
1379 		if (!pagelist)
1380 			goto out;
1381 		err = ceph_pagelist_reserve(pagelist, PAGE_SIZE);
1382 		if (err)
1383 			goto out;
1384 		ceph_pagelist_encode_32(pagelist, 1);
1385 	}
1386 
1387 	/*
1388 	 * FIXME: Make security_dentry_init_security() generic. Currently
1389 	 * It only supports single security module and only selinux has
1390 	 * dentry_init_security hook.
1391 	 */
1392 	name_len = strlen(name);
1393 	err = ceph_pagelist_reserve(pagelist,
1394 				    4 * 2 + name_len + as_ctx->sec_ctxlen);
1395 	if (err)
1396 		goto out;
1397 
1398 	if (as_ctx->pagelist) {
1399 		/* update count of KV pairs */
1400 		BUG_ON(pagelist->length <= sizeof(__le32));
1401 		if (list_is_singular(&pagelist->head)) {
1402 			le32_add_cpu((__le32*)pagelist->mapped_tail, 1);
1403 		} else {
1404 			struct page *page = list_first_entry(&pagelist->head,
1405 							     struct page, lru);
1406 			void *addr = kmap_atomic(page);
1407 			le32_add_cpu((__le32*)addr, 1);
1408 			kunmap_atomic(addr);
1409 		}
1410 	} else {
1411 		as_ctx->pagelist = pagelist;
1412 	}
1413 
1414 	ceph_pagelist_encode_32(pagelist, name_len);
1415 	ceph_pagelist_append(pagelist, name, name_len);
1416 
1417 	ceph_pagelist_encode_32(pagelist, as_ctx->sec_ctxlen);
1418 	ceph_pagelist_append(pagelist, as_ctx->sec_ctx, as_ctx->sec_ctxlen);
1419 
1420 	err = 0;
1421 out:
1422 	if (pagelist && !as_ctx->pagelist)
1423 		ceph_pagelist_release(pagelist);
1424 	return err;
1425 }
1426 #endif /* CONFIG_CEPH_FS_SECURITY_LABEL */
1427 #endif /* CONFIG_SECURITY */
1428 
1429 void ceph_release_acl_sec_ctx(struct ceph_acl_sec_ctx *as_ctx)
1430 {
1431 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1432 	posix_acl_release(as_ctx->acl);
1433 	posix_acl_release(as_ctx->default_acl);
1434 #endif
1435 #ifdef CONFIG_CEPH_FS_SECURITY_LABEL
1436 	security_release_secctx(as_ctx->sec_ctx, as_ctx->sec_ctxlen);
1437 #endif
1438 #ifdef CONFIG_FS_ENCRYPTION
1439 	kfree(as_ctx->fscrypt_auth);
1440 #endif
1441 	if (as_ctx->pagelist)
1442 		ceph_pagelist_release(as_ctx->pagelist);
1443 }
1444 
1445 /*
1446  * List of handlers for synthetic system.* attributes. Other
1447  * attributes are handled directly.
1448  */
1449 const struct xattr_handler *ceph_xattr_handlers[] = {
1450 	&ceph_other_xattr_handler,
1451 	NULL,
1452 };
1453