xref: /openbmc/linux/fs/ceph/xattr.c (revision 9ad685db)
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 #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_RSTAT_FIELD_UPDATABLE(_type, _name)			\
370 	{								\
371 		.name = CEPH_XATTR_NAME(_type, _name),			\
372 		.name_size = sizeof (CEPH_XATTR_NAME(_type, _name)),	\
373 		.getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name,	\
374 		.exists_cb = NULL,					\
375 		.flags = VXATTR_FLAG_RSTAT,				\
376 	}
377 #define XATTR_LAYOUT_FIELD(_type, _name, _field)			\
378 	{								\
379 		.name = CEPH_XATTR_NAME2(_type, _name, _field),	\
380 		.name_size = sizeof (CEPH_XATTR_NAME2(_type, _name, _field)), \
381 		.getxattr_cb = ceph_vxattrcb_ ## _name ## _ ## _field, \
382 		.exists_cb = ceph_vxattrcb_layout_exists,	\
383 		.flags = VXATTR_FLAG_HIDDEN,			\
384 	}
385 #define XATTR_QUOTA_FIELD(_type, _name)					\
386 	{								\
387 		.name = CEPH_XATTR_NAME(_type, _name),			\
388 		.name_size = sizeof(CEPH_XATTR_NAME(_type, _name)),	\
389 		.getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name,	\
390 		.exists_cb = ceph_vxattrcb_quota_exists,		\
391 		.flags = VXATTR_FLAG_HIDDEN,				\
392 	}
393 
394 static struct ceph_vxattr ceph_dir_vxattrs[] = {
395 	{
396 		.name = "ceph.dir.layout",
397 		.name_size = sizeof("ceph.dir.layout"),
398 		.getxattr_cb = ceph_vxattrcb_layout,
399 		.exists_cb = ceph_vxattrcb_layout_exists,
400 		.flags = VXATTR_FLAG_HIDDEN,
401 	},
402 	XATTR_LAYOUT_FIELD(dir, layout, stripe_unit),
403 	XATTR_LAYOUT_FIELD(dir, layout, stripe_count),
404 	XATTR_LAYOUT_FIELD(dir, layout, object_size),
405 	XATTR_LAYOUT_FIELD(dir, layout, pool),
406 	XATTR_LAYOUT_FIELD(dir, layout, pool_namespace),
407 	XATTR_NAME_CEPH(dir, entries, VXATTR_FLAG_DIRSTAT),
408 	XATTR_NAME_CEPH(dir, files, VXATTR_FLAG_DIRSTAT),
409 	XATTR_NAME_CEPH(dir, subdirs, VXATTR_FLAG_DIRSTAT),
410 	XATTR_RSTAT_FIELD(dir, rentries),
411 	XATTR_RSTAT_FIELD(dir, rfiles),
412 	XATTR_RSTAT_FIELD(dir, rsubdirs),
413 	XATTR_RSTAT_FIELD(dir, rsnaps),
414 	XATTR_RSTAT_FIELD(dir, rbytes),
415 	XATTR_RSTAT_FIELD_UPDATABLE(dir, rctime),
416 	{
417 		.name = "ceph.dir.pin",
418 		.name_size = sizeof("ceph.dir.pin"),
419 		.getxattr_cb = ceph_vxattrcb_dir_pin,
420 		.exists_cb = ceph_vxattrcb_dir_pin_exists,
421 		.flags = VXATTR_FLAG_HIDDEN,
422 	},
423 	{
424 		.name = "ceph.quota",
425 		.name_size = sizeof("ceph.quota"),
426 		.getxattr_cb = ceph_vxattrcb_quota,
427 		.exists_cb = ceph_vxattrcb_quota_exists,
428 		.flags = VXATTR_FLAG_HIDDEN,
429 	},
430 	XATTR_QUOTA_FIELD(quota, max_bytes),
431 	XATTR_QUOTA_FIELD(quota, max_files),
432 	{
433 		.name = "ceph.snap.btime",
434 		.name_size = sizeof("ceph.snap.btime"),
435 		.getxattr_cb = ceph_vxattrcb_snap_btime,
436 		.exists_cb = ceph_vxattrcb_snap_btime_exists,
437 		.flags = VXATTR_FLAG_READONLY,
438 	},
439 	{
440 		.name = "ceph.caps",
441 		.name_size = sizeof("ceph.caps"),
442 		.getxattr_cb = ceph_vxattrcb_caps,
443 		.exists_cb = NULL,
444 		.flags = VXATTR_FLAG_HIDDEN,
445 	},
446 	{ .name = NULL, 0 }	/* Required table terminator */
447 };
448 
449 /* files */
450 
451 static struct ceph_vxattr ceph_file_vxattrs[] = {
452 	{
453 		.name = "ceph.file.layout",
454 		.name_size = sizeof("ceph.file.layout"),
455 		.getxattr_cb = ceph_vxattrcb_layout,
456 		.exists_cb = ceph_vxattrcb_layout_exists,
457 		.flags = VXATTR_FLAG_HIDDEN,
458 	},
459 	XATTR_LAYOUT_FIELD(file, layout, stripe_unit),
460 	XATTR_LAYOUT_FIELD(file, layout, stripe_count),
461 	XATTR_LAYOUT_FIELD(file, layout, object_size),
462 	XATTR_LAYOUT_FIELD(file, layout, pool),
463 	XATTR_LAYOUT_FIELD(file, layout, pool_namespace),
464 	{
465 		.name = "ceph.snap.btime",
466 		.name_size = sizeof("ceph.snap.btime"),
467 		.getxattr_cb = ceph_vxattrcb_snap_btime,
468 		.exists_cb = ceph_vxattrcb_snap_btime_exists,
469 		.flags = VXATTR_FLAG_READONLY,
470 	},
471 	{
472 		.name = "ceph.caps",
473 		.name_size = sizeof("ceph.caps"),
474 		.getxattr_cb = ceph_vxattrcb_caps,
475 		.exists_cb = NULL,
476 		.flags = VXATTR_FLAG_HIDDEN,
477 	},
478 	{ .name = NULL, 0 }	/* Required table terminator */
479 };
480 
481 static struct ceph_vxattr ceph_common_vxattrs[] = {
482 	{
483 		.name = "ceph.cluster_fsid",
484 		.name_size = sizeof("ceph.cluster_fsid"),
485 		.getxattr_cb = ceph_vxattrcb_cluster_fsid,
486 		.exists_cb = NULL,
487 		.flags = VXATTR_FLAG_READONLY,
488 	},
489 	{
490 		.name = "ceph.client_id",
491 		.name_size = sizeof("ceph.client_id"),
492 		.getxattr_cb = ceph_vxattrcb_client_id,
493 		.exists_cb = NULL,
494 		.flags = VXATTR_FLAG_READONLY,
495 	},
496 	{
497 		.name = "ceph.auth_mds",
498 		.name_size = sizeof("ceph.auth_mds"),
499 		.getxattr_cb = ceph_vxattrcb_auth_mds,
500 		.exists_cb = NULL,
501 		.flags = VXATTR_FLAG_READONLY,
502 	},
503 	{ .name = NULL, 0 }	/* Required table terminator */
504 };
505 
506 static struct ceph_vxattr *ceph_inode_vxattrs(struct inode *inode)
507 {
508 	if (S_ISDIR(inode->i_mode))
509 		return ceph_dir_vxattrs;
510 	else if (S_ISREG(inode->i_mode))
511 		return ceph_file_vxattrs;
512 	return NULL;
513 }
514 
515 static struct ceph_vxattr *ceph_match_vxattr(struct inode *inode,
516 						const char *name)
517 {
518 	struct ceph_vxattr *vxattr = ceph_inode_vxattrs(inode);
519 
520 	if (vxattr) {
521 		while (vxattr->name) {
522 			if (!strcmp(vxattr->name, name))
523 				return vxattr;
524 			vxattr++;
525 		}
526 	}
527 
528 	vxattr = ceph_common_vxattrs;
529 	while (vxattr->name) {
530 		if (!strcmp(vxattr->name, name))
531 			return vxattr;
532 		vxattr++;
533 	}
534 
535 	return NULL;
536 }
537 
538 static int __set_xattr(struct ceph_inode_info *ci,
539 			   const char *name, int name_len,
540 			   const char *val, int val_len,
541 			   int flags, int update_xattr,
542 			   struct ceph_inode_xattr **newxattr)
543 {
544 	struct rb_node **p;
545 	struct rb_node *parent = NULL;
546 	struct ceph_inode_xattr *xattr = NULL;
547 	int c;
548 	int new = 0;
549 
550 	p = &ci->i_xattrs.index.rb_node;
551 	while (*p) {
552 		parent = *p;
553 		xattr = rb_entry(parent, struct ceph_inode_xattr, node);
554 		c = strncmp(name, xattr->name, min(name_len, xattr->name_len));
555 		if (c < 0)
556 			p = &(*p)->rb_left;
557 		else if (c > 0)
558 			p = &(*p)->rb_right;
559 		else {
560 			if (name_len == xattr->name_len)
561 				break;
562 			else if (name_len < xattr->name_len)
563 				p = &(*p)->rb_left;
564 			else
565 				p = &(*p)->rb_right;
566 		}
567 		xattr = NULL;
568 	}
569 
570 	if (update_xattr) {
571 		int err = 0;
572 
573 		if (xattr && (flags & XATTR_CREATE))
574 			err = -EEXIST;
575 		else if (!xattr && (flags & XATTR_REPLACE))
576 			err = -ENODATA;
577 		if (err) {
578 			kfree(name);
579 			kfree(val);
580 			kfree(*newxattr);
581 			return err;
582 		}
583 		if (update_xattr < 0) {
584 			if (xattr)
585 				__remove_xattr(ci, xattr);
586 			kfree(name);
587 			kfree(*newxattr);
588 			return 0;
589 		}
590 	}
591 
592 	if (!xattr) {
593 		new = 1;
594 		xattr = *newxattr;
595 		xattr->name = name;
596 		xattr->name_len = name_len;
597 		xattr->should_free_name = update_xattr;
598 
599 		ci->i_xattrs.count++;
600 		dout("__set_xattr count=%d\n", ci->i_xattrs.count);
601 	} else {
602 		kfree(*newxattr);
603 		*newxattr = NULL;
604 		if (xattr->should_free_val)
605 			kfree(xattr->val);
606 
607 		if (update_xattr) {
608 			kfree(name);
609 			name = xattr->name;
610 		}
611 		ci->i_xattrs.names_size -= xattr->name_len;
612 		ci->i_xattrs.vals_size -= xattr->val_len;
613 	}
614 	ci->i_xattrs.names_size += name_len;
615 	ci->i_xattrs.vals_size += val_len;
616 	if (val)
617 		xattr->val = val;
618 	else
619 		xattr->val = "";
620 
621 	xattr->val_len = val_len;
622 	xattr->dirty = update_xattr;
623 	xattr->should_free_val = (val && update_xattr);
624 
625 	if (new) {
626 		rb_link_node(&xattr->node, parent, p);
627 		rb_insert_color(&xattr->node, &ci->i_xattrs.index);
628 		dout("__set_xattr_val p=%p\n", p);
629 	}
630 
631 	dout("__set_xattr_val added %llx.%llx xattr %p %.*s=%.*s\n",
632 	     ceph_vinop(&ci->netfs.inode), xattr, name_len, name, val_len, val);
633 
634 	return 0;
635 }
636 
637 static struct ceph_inode_xattr *__get_xattr(struct ceph_inode_info *ci,
638 			   const char *name)
639 {
640 	struct rb_node **p;
641 	struct rb_node *parent = NULL;
642 	struct ceph_inode_xattr *xattr = NULL;
643 	int name_len = strlen(name);
644 	int c;
645 
646 	p = &ci->i_xattrs.index.rb_node;
647 	while (*p) {
648 		parent = *p;
649 		xattr = rb_entry(parent, struct ceph_inode_xattr, node);
650 		c = strncmp(name, xattr->name, xattr->name_len);
651 		if (c == 0 && name_len > xattr->name_len)
652 			c = 1;
653 		if (c < 0)
654 			p = &(*p)->rb_left;
655 		else if (c > 0)
656 			p = &(*p)->rb_right;
657 		else {
658 			dout("__get_xattr %s: found %.*s\n", name,
659 			     xattr->val_len, xattr->val);
660 			return xattr;
661 		}
662 	}
663 
664 	dout("__get_xattr %s: not found\n", name);
665 
666 	return NULL;
667 }
668 
669 static void __free_xattr(struct ceph_inode_xattr *xattr)
670 {
671 	BUG_ON(!xattr);
672 
673 	if (xattr->should_free_name)
674 		kfree(xattr->name);
675 	if (xattr->should_free_val)
676 		kfree(xattr->val);
677 
678 	kfree(xattr);
679 }
680 
681 static int __remove_xattr(struct ceph_inode_info *ci,
682 			  struct ceph_inode_xattr *xattr)
683 {
684 	if (!xattr)
685 		return -ENODATA;
686 
687 	rb_erase(&xattr->node, &ci->i_xattrs.index);
688 
689 	if (xattr->should_free_name)
690 		kfree(xattr->name);
691 	if (xattr->should_free_val)
692 		kfree(xattr->val);
693 
694 	ci->i_xattrs.names_size -= xattr->name_len;
695 	ci->i_xattrs.vals_size -= xattr->val_len;
696 	ci->i_xattrs.count--;
697 	kfree(xattr);
698 
699 	return 0;
700 }
701 
702 static char *__copy_xattr_names(struct ceph_inode_info *ci,
703 				char *dest)
704 {
705 	struct rb_node *p;
706 	struct ceph_inode_xattr *xattr = NULL;
707 
708 	p = rb_first(&ci->i_xattrs.index);
709 	dout("__copy_xattr_names count=%d\n", ci->i_xattrs.count);
710 
711 	while (p) {
712 		xattr = rb_entry(p, struct ceph_inode_xattr, node);
713 		memcpy(dest, xattr->name, xattr->name_len);
714 		dest[xattr->name_len] = '\0';
715 
716 		dout("dest=%s %p (%s) (%d/%d)\n", dest, xattr, xattr->name,
717 		     xattr->name_len, ci->i_xattrs.names_size);
718 
719 		dest += xattr->name_len + 1;
720 		p = rb_next(p);
721 	}
722 
723 	return dest;
724 }
725 
726 void __ceph_destroy_xattrs(struct ceph_inode_info *ci)
727 {
728 	struct rb_node *p, *tmp;
729 	struct ceph_inode_xattr *xattr = NULL;
730 
731 	p = rb_first(&ci->i_xattrs.index);
732 
733 	dout("__ceph_destroy_xattrs p=%p\n", p);
734 
735 	while (p) {
736 		xattr = rb_entry(p, struct ceph_inode_xattr, node);
737 		tmp = p;
738 		p = rb_next(tmp);
739 		dout("__ceph_destroy_xattrs next p=%p (%.*s)\n", p,
740 		     xattr->name_len, xattr->name);
741 		rb_erase(tmp, &ci->i_xattrs.index);
742 
743 		__free_xattr(xattr);
744 	}
745 
746 	ci->i_xattrs.names_size = 0;
747 	ci->i_xattrs.vals_size = 0;
748 	ci->i_xattrs.index_version = 0;
749 	ci->i_xattrs.count = 0;
750 	ci->i_xattrs.index = RB_ROOT;
751 }
752 
753 static int __build_xattrs(struct inode *inode)
754 	__releases(ci->i_ceph_lock)
755 	__acquires(ci->i_ceph_lock)
756 {
757 	u32 namelen;
758 	u32 numattr = 0;
759 	void *p, *end;
760 	u32 len;
761 	const char *name, *val;
762 	struct ceph_inode_info *ci = ceph_inode(inode);
763 	u64 xattr_version;
764 	struct ceph_inode_xattr **xattrs = NULL;
765 	int err = 0;
766 	int i;
767 
768 	dout("__build_xattrs() len=%d\n",
769 	     ci->i_xattrs.blob ? (int)ci->i_xattrs.blob->vec.iov_len : 0);
770 
771 	if (ci->i_xattrs.index_version >= ci->i_xattrs.version)
772 		return 0; /* already built */
773 
774 	__ceph_destroy_xattrs(ci);
775 
776 start:
777 	/* updated internal xattr rb tree */
778 	if (ci->i_xattrs.blob && ci->i_xattrs.blob->vec.iov_len > 4) {
779 		p = ci->i_xattrs.blob->vec.iov_base;
780 		end = p + ci->i_xattrs.blob->vec.iov_len;
781 		ceph_decode_32_safe(&p, end, numattr, bad);
782 		xattr_version = ci->i_xattrs.version;
783 		spin_unlock(&ci->i_ceph_lock);
784 
785 		xattrs = kcalloc(numattr, sizeof(struct ceph_inode_xattr *),
786 				 GFP_NOFS);
787 		err = -ENOMEM;
788 		if (!xattrs)
789 			goto bad_lock;
790 
791 		for (i = 0; i < numattr; i++) {
792 			xattrs[i] = kmalloc(sizeof(struct ceph_inode_xattr),
793 					    GFP_NOFS);
794 			if (!xattrs[i])
795 				goto bad_lock;
796 		}
797 
798 		spin_lock(&ci->i_ceph_lock);
799 		if (ci->i_xattrs.version != xattr_version) {
800 			/* lost a race, retry */
801 			for (i = 0; i < numattr; i++)
802 				kfree(xattrs[i]);
803 			kfree(xattrs);
804 			xattrs = NULL;
805 			goto start;
806 		}
807 		err = -EIO;
808 		while (numattr--) {
809 			ceph_decode_32_safe(&p, end, len, bad);
810 			namelen = len;
811 			name = p;
812 			p += len;
813 			ceph_decode_32_safe(&p, end, len, bad);
814 			val = p;
815 			p += len;
816 
817 			err = __set_xattr(ci, name, namelen, val, len,
818 					  0, 0, &xattrs[numattr]);
819 
820 			if (err < 0)
821 				goto bad;
822 		}
823 		kfree(xattrs);
824 	}
825 	ci->i_xattrs.index_version = ci->i_xattrs.version;
826 	ci->i_xattrs.dirty = false;
827 
828 	return err;
829 bad_lock:
830 	spin_lock(&ci->i_ceph_lock);
831 bad:
832 	if (xattrs) {
833 		for (i = 0; i < numattr; i++)
834 			kfree(xattrs[i]);
835 		kfree(xattrs);
836 	}
837 	ci->i_xattrs.names_size = 0;
838 	return err;
839 }
840 
841 static int __get_required_blob_size(struct ceph_inode_info *ci, int name_size,
842 				    int val_size)
843 {
844 	/*
845 	 * 4 bytes for the length, and additional 4 bytes per each xattr name,
846 	 * 4 bytes per each value
847 	 */
848 	int size = 4 + ci->i_xattrs.count*(4 + 4) +
849 			     ci->i_xattrs.names_size +
850 			     ci->i_xattrs.vals_size;
851 	dout("__get_required_blob_size c=%d names.size=%d vals.size=%d\n",
852 	     ci->i_xattrs.count, ci->i_xattrs.names_size,
853 	     ci->i_xattrs.vals_size);
854 
855 	if (name_size)
856 		size += 4 + 4 + name_size + val_size;
857 
858 	return size;
859 }
860 
861 /*
862  * If there are dirty xattrs, reencode xattrs into the prealloc_blob
863  * and swap into place.  It returns the old i_xattrs.blob (or NULL) so
864  * that it can be freed by the caller as the i_ceph_lock is likely to be
865  * held.
866  */
867 struct ceph_buffer *__ceph_build_xattrs_blob(struct ceph_inode_info *ci)
868 {
869 	struct rb_node *p;
870 	struct ceph_inode_xattr *xattr = NULL;
871 	struct ceph_buffer *old_blob = NULL;
872 	void *dest;
873 
874 	dout("__build_xattrs_blob %p\n", &ci->netfs.inode);
875 	if (ci->i_xattrs.dirty) {
876 		int need = __get_required_blob_size(ci, 0, 0);
877 
878 		BUG_ON(need > ci->i_xattrs.prealloc_blob->alloc_len);
879 
880 		p = rb_first(&ci->i_xattrs.index);
881 		dest = ci->i_xattrs.prealloc_blob->vec.iov_base;
882 
883 		ceph_encode_32(&dest, ci->i_xattrs.count);
884 		while (p) {
885 			xattr = rb_entry(p, struct ceph_inode_xattr, node);
886 
887 			ceph_encode_32(&dest, xattr->name_len);
888 			memcpy(dest, xattr->name, xattr->name_len);
889 			dest += xattr->name_len;
890 			ceph_encode_32(&dest, xattr->val_len);
891 			memcpy(dest, xattr->val, xattr->val_len);
892 			dest += xattr->val_len;
893 
894 			p = rb_next(p);
895 		}
896 
897 		/* adjust buffer len; it may be larger than we need */
898 		ci->i_xattrs.prealloc_blob->vec.iov_len =
899 			dest - ci->i_xattrs.prealloc_blob->vec.iov_base;
900 
901 		if (ci->i_xattrs.blob)
902 			old_blob = ci->i_xattrs.blob;
903 		ci->i_xattrs.blob = ci->i_xattrs.prealloc_blob;
904 		ci->i_xattrs.prealloc_blob = NULL;
905 		ci->i_xattrs.dirty = false;
906 		ci->i_xattrs.version++;
907 	}
908 
909 	return old_blob;
910 }
911 
912 static inline int __get_request_mask(struct inode *in) {
913 	struct ceph_mds_request *req = current->journal_info;
914 	int mask = 0;
915 	if (req && req->r_target_inode == in) {
916 		if (req->r_op == CEPH_MDS_OP_LOOKUP ||
917 		    req->r_op == CEPH_MDS_OP_LOOKUPINO ||
918 		    req->r_op == CEPH_MDS_OP_LOOKUPPARENT ||
919 		    req->r_op == CEPH_MDS_OP_GETATTR) {
920 			mask = le32_to_cpu(req->r_args.getattr.mask);
921 		} else if (req->r_op == CEPH_MDS_OP_OPEN ||
922 			   req->r_op == CEPH_MDS_OP_CREATE) {
923 			mask = le32_to_cpu(req->r_args.open.mask);
924 		}
925 	}
926 	return mask;
927 }
928 
929 ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,
930 		      size_t size)
931 {
932 	struct ceph_inode_info *ci = ceph_inode(inode);
933 	struct ceph_inode_xattr *xattr;
934 	struct ceph_vxattr *vxattr;
935 	int req_mask;
936 	ssize_t err;
937 
938 	if (strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
939 		goto handle_non_vxattrs;
940 
941 	/* let's see if a virtual xattr was requested */
942 	vxattr = ceph_match_vxattr(inode, name);
943 	if (vxattr) {
944 		int mask = 0;
945 		if (vxattr->flags & VXATTR_FLAG_RSTAT)
946 			mask |= CEPH_STAT_RSTAT;
947 		if (vxattr->flags & VXATTR_FLAG_DIRSTAT)
948 			mask |= CEPH_CAP_FILE_SHARED;
949 		err = ceph_do_getattr(inode, mask, true);
950 		if (err)
951 			return err;
952 		err = -ENODATA;
953 		if (!(vxattr->exists_cb && !vxattr->exists_cb(ci))) {
954 			err = vxattr->getxattr_cb(ci, value, size);
955 			if (size && size < err)
956 				err = -ERANGE;
957 		}
958 		return err;
959 	} else {
960 		err = ceph_do_getvxattr(inode, name, value, size);
961 		/* this would happen with a new client and old server combo */
962 		if (err == -EOPNOTSUPP)
963 			err = -ENODATA;
964 		return err;
965 	}
966 handle_non_vxattrs:
967 	req_mask = __get_request_mask(inode);
968 
969 	spin_lock(&ci->i_ceph_lock);
970 	dout("getxattr %p name '%s' ver=%lld index_ver=%lld\n", inode, name,
971 	     ci->i_xattrs.version, ci->i_xattrs.index_version);
972 
973 	if (ci->i_xattrs.version == 0 ||
974 	    !((req_mask & CEPH_CAP_XATTR_SHARED) ||
975 	      __ceph_caps_issued_mask_metric(ci, CEPH_CAP_XATTR_SHARED, 1))) {
976 		spin_unlock(&ci->i_ceph_lock);
977 
978 		/* security module gets xattr while filling trace */
979 		if (current->journal_info) {
980 			pr_warn_ratelimited("sync getxattr %p "
981 					    "during filling trace\n", inode);
982 			return -EBUSY;
983 		}
984 
985 		/* get xattrs from mds (if we don't already have them) */
986 		err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
987 		if (err)
988 			return err;
989 		spin_lock(&ci->i_ceph_lock);
990 	}
991 
992 	err = __build_xattrs(inode);
993 	if (err < 0)
994 		goto out;
995 
996 	err = -ENODATA;  /* == ENOATTR */
997 	xattr = __get_xattr(ci, name);
998 	if (!xattr)
999 		goto out;
1000 
1001 	err = -ERANGE;
1002 	if (size && size < xattr->val_len)
1003 		goto out;
1004 
1005 	err = xattr->val_len;
1006 	if (size == 0)
1007 		goto out;
1008 
1009 	memcpy(value, xattr->val, xattr->val_len);
1010 
1011 	if (current->journal_info &&
1012 	    !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN) &&
1013 	    security_ismaclabel(name + XATTR_SECURITY_PREFIX_LEN))
1014 		ci->i_ceph_flags |= CEPH_I_SEC_INITED;
1015 out:
1016 	spin_unlock(&ci->i_ceph_lock);
1017 	return err;
1018 }
1019 
1020 ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)
1021 {
1022 	struct inode *inode = d_inode(dentry);
1023 	struct ceph_inode_info *ci = ceph_inode(inode);
1024 	bool len_only = (size == 0);
1025 	u32 namelen;
1026 	int err;
1027 
1028 	spin_lock(&ci->i_ceph_lock);
1029 	dout("listxattr %p ver=%lld index_ver=%lld\n", inode,
1030 	     ci->i_xattrs.version, ci->i_xattrs.index_version);
1031 
1032 	if (ci->i_xattrs.version == 0 ||
1033 	    !__ceph_caps_issued_mask_metric(ci, CEPH_CAP_XATTR_SHARED, 1)) {
1034 		spin_unlock(&ci->i_ceph_lock);
1035 		err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
1036 		if (err)
1037 			return err;
1038 		spin_lock(&ci->i_ceph_lock);
1039 	}
1040 
1041 	err = __build_xattrs(inode);
1042 	if (err < 0)
1043 		goto out;
1044 
1045 	/* add 1 byte for each xattr due to the null termination */
1046 	namelen = ci->i_xattrs.names_size + ci->i_xattrs.count;
1047 	if (!len_only) {
1048 		if (namelen > size) {
1049 			err = -ERANGE;
1050 			goto out;
1051 		}
1052 		names = __copy_xattr_names(ci, names);
1053 		size -= namelen;
1054 	}
1055 	err = namelen;
1056 out:
1057 	spin_unlock(&ci->i_ceph_lock);
1058 	return err;
1059 }
1060 
1061 static int ceph_sync_setxattr(struct inode *inode, const char *name,
1062 			      const char *value, size_t size, int flags)
1063 {
1064 	struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1065 	struct ceph_inode_info *ci = ceph_inode(inode);
1066 	struct ceph_mds_request *req;
1067 	struct ceph_mds_client *mdsc = fsc->mdsc;
1068 	struct ceph_osd_client *osdc = &fsc->client->osdc;
1069 	struct ceph_pagelist *pagelist = NULL;
1070 	int op = CEPH_MDS_OP_SETXATTR;
1071 	int err;
1072 
1073 	if (size > 0) {
1074 		/* copy value into pagelist */
1075 		pagelist = ceph_pagelist_alloc(GFP_NOFS);
1076 		if (!pagelist)
1077 			return -ENOMEM;
1078 
1079 		err = ceph_pagelist_append(pagelist, value, size);
1080 		if (err)
1081 			goto out;
1082 	} else if (!value) {
1083 		if (flags & CEPH_XATTR_REPLACE)
1084 			op = CEPH_MDS_OP_RMXATTR;
1085 		else
1086 			flags |= CEPH_XATTR_REMOVE;
1087 	}
1088 
1089 	dout("setxattr value size: %zu\n", size);
1090 
1091 	/* do request */
1092 	req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1093 	if (IS_ERR(req)) {
1094 		err = PTR_ERR(req);
1095 		goto out;
1096 	}
1097 
1098 	req->r_path2 = kstrdup(name, GFP_NOFS);
1099 	if (!req->r_path2) {
1100 		ceph_mdsc_put_request(req);
1101 		err = -ENOMEM;
1102 		goto out;
1103 	}
1104 
1105 	if (op == CEPH_MDS_OP_SETXATTR) {
1106 		req->r_args.setxattr.flags = cpu_to_le32(flags);
1107 		req->r_args.setxattr.osdmap_epoch =
1108 			cpu_to_le32(osdc->osdmap->epoch);
1109 		req->r_pagelist = pagelist;
1110 		pagelist = NULL;
1111 	}
1112 
1113 	req->r_inode = inode;
1114 	ihold(inode);
1115 	req->r_num_caps = 1;
1116 	req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
1117 
1118 	dout("xattr.ver (before): %lld\n", ci->i_xattrs.version);
1119 	err = ceph_mdsc_do_request(mdsc, NULL, req);
1120 	ceph_mdsc_put_request(req);
1121 	dout("xattr.ver (after): %lld\n", ci->i_xattrs.version);
1122 
1123 out:
1124 	if (pagelist)
1125 		ceph_pagelist_release(pagelist);
1126 	return err;
1127 }
1128 
1129 int __ceph_setxattr(struct inode *inode, const char *name,
1130 			const void *value, size_t size, int flags)
1131 {
1132 	struct ceph_vxattr *vxattr;
1133 	struct ceph_inode_info *ci = ceph_inode(inode);
1134 	struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1135 	struct ceph_cap_flush *prealloc_cf = NULL;
1136 	struct ceph_buffer *old_blob = NULL;
1137 	int issued;
1138 	int err;
1139 	int dirty = 0;
1140 	int name_len = strlen(name);
1141 	int val_len = size;
1142 	char *newname = NULL;
1143 	char *newval = NULL;
1144 	struct ceph_inode_xattr *xattr = NULL;
1145 	int required_blob_size;
1146 	bool check_realm = false;
1147 	bool lock_snap_rwsem = false;
1148 
1149 	if (ceph_snap(inode) != CEPH_NOSNAP)
1150 		return -EROFS;
1151 
1152 	vxattr = ceph_match_vxattr(inode, name);
1153 	if (vxattr) {
1154 		if (vxattr->flags & VXATTR_FLAG_READONLY)
1155 			return -EOPNOTSUPP;
1156 		if (value && !strncmp(vxattr->name, "ceph.quota", 10))
1157 			check_realm = true;
1158 	}
1159 
1160 	/* pass any unhandled ceph.* xattrs through to the MDS */
1161 	if (!strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
1162 		goto do_sync_unlocked;
1163 
1164 	/* preallocate memory for xattr name, value, index node */
1165 	err = -ENOMEM;
1166 	newname = kmemdup(name, name_len + 1, GFP_NOFS);
1167 	if (!newname)
1168 		goto out;
1169 
1170 	if (val_len) {
1171 		newval = kmemdup(value, val_len, GFP_NOFS);
1172 		if (!newval)
1173 			goto out;
1174 	}
1175 
1176 	xattr = kmalloc(sizeof(struct ceph_inode_xattr), GFP_NOFS);
1177 	if (!xattr)
1178 		goto out;
1179 
1180 	prealloc_cf = ceph_alloc_cap_flush();
1181 	if (!prealloc_cf)
1182 		goto out;
1183 
1184 	spin_lock(&ci->i_ceph_lock);
1185 retry:
1186 	issued = __ceph_caps_issued(ci, NULL);
1187 	required_blob_size = __get_required_blob_size(ci, name_len, val_len);
1188 	if ((ci->i_xattrs.version == 0) || !(issued & CEPH_CAP_XATTR_EXCL) ||
1189 	    (required_blob_size > mdsc->mdsmap->m_max_xattr_size)) {
1190 		dout("%s do sync setxattr: version: %llu size: %d max: %llu\n",
1191 		     __func__, ci->i_xattrs.version, required_blob_size,
1192 		     mdsc->mdsmap->m_max_xattr_size);
1193 		goto do_sync;
1194 	}
1195 
1196 	if (!lock_snap_rwsem && !ci->i_head_snapc) {
1197 		lock_snap_rwsem = true;
1198 		if (!down_read_trylock(&mdsc->snap_rwsem)) {
1199 			spin_unlock(&ci->i_ceph_lock);
1200 			down_read(&mdsc->snap_rwsem);
1201 			spin_lock(&ci->i_ceph_lock);
1202 			goto retry;
1203 		}
1204 	}
1205 
1206 	dout("setxattr %p name '%s' issued %s\n", inode, name,
1207 	     ceph_cap_string(issued));
1208 	__build_xattrs(inode);
1209 
1210 	if (!ci->i_xattrs.prealloc_blob ||
1211 	    required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
1212 		struct ceph_buffer *blob;
1213 
1214 		spin_unlock(&ci->i_ceph_lock);
1215 		ceph_buffer_put(old_blob); /* Shouldn't be required */
1216 		dout(" pre-allocating new blob size=%d\n", required_blob_size);
1217 		blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
1218 		if (!blob)
1219 			goto do_sync_unlocked;
1220 		spin_lock(&ci->i_ceph_lock);
1221 		/* prealloc_blob can't be released while holding i_ceph_lock */
1222 		if (ci->i_xattrs.prealloc_blob)
1223 			old_blob = ci->i_xattrs.prealloc_blob;
1224 		ci->i_xattrs.prealloc_blob = blob;
1225 		goto retry;
1226 	}
1227 
1228 	err = __set_xattr(ci, newname, name_len, newval, val_len,
1229 			  flags, value ? 1 : -1, &xattr);
1230 
1231 	if (!err) {
1232 		dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL,
1233 					       &prealloc_cf);
1234 		ci->i_xattrs.dirty = true;
1235 		inode->i_ctime = current_time(inode);
1236 	}
1237 
1238 	spin_unlock(&ci->i_ceph_lock);
1239 	ceph_buffer_put(old_blob);
1240 	if (lock_snap_rwsem)
1241 		up_read(&mdsc->snap_rwsem);
1242 	if (dirty)
1243 		__mark_inode_dirty(inode, dirty);
1244 	ceph_free_cap_flush(prealloc_cf);
1245 	return err;
1246 
1247 do_sync:
1248 	spin_unlock(&ci->i_ceph_lock);
1249 do_sync_unlocked:
1250 	if (lock_snap_rwsem)
1251 		up_read(&mdsc->snap_rwsem);
1252 
1253 	/* security module set xattr while filling trace */
1254 	if (current->journal_info) {
1255 		pr_warn_ratelimited("sync setxattr %p "
1256 				    "during filling trace\n", inode);
1257 		err = -EBUSY;
1258 	} else {
1259 		err = ceph_sync_setxattr(inode, name, value, size, flags);
1260 		if (err >= 0 && check_realm) {
1261 			/* check if snaprealm was created for quota inode */
1262 			spin_lock(&ci->i_ceph_lock);
1263 			if ((ci->i_max_files || ci->i_max_bytes) &&
1264 			    !(ci->i_snap_realm &&
1265 			      ci->i_snap_realm->ino == ci->i_vino.ino))
1266 				err = -EOPNOTSUPP;
1267 			spin_unlock(&ci->i_ceph_lock);
1268 		}
1269 	}
1270 out:
1271 	ceph_free_cap_flush(prealloc_cf);
1272 	kfree(newname);
1273 	kfree(newval);
1274 	kfree(xattr);
1275 	return err;
1276 }
1277 
1278 static int ceph_get_xattr_handler(const struct xattr_handler *handler,
1279 				  struct dentry *dentry, struct inode *inode,
1280 				  const char *name, void *value, size_t size)
1281 {
1282 	if (!ceph_is_valid_xattr(name))
1283 		return -EOPNOTSUPP;
1284 	return __ceph_getxattr(inode, name, value, size);
1285 }
1286 
1287 static int ceph_set_xattr_handler(const struct xattr_handler *handler,
1288 				  struct mnt_idmap *idmap,
1289 				  struct dentry *unused, struct inode *inode,
1290 				  const char *name, const void *value,
1291 				  size_t size, int flags)
1292 {
1293 	if (!ceph_is_valid_xattr(name))
1294 		return -EOPNOTSUPP;
1295 	return __ceph_setxattr(inode, name, value, size, flags);
1296 }
1297 
1298 static const struct xattr_handler ceph_other_xattr_handler = {
1299 	.prefix = "",  /* match any name => handlers called with full name */
1300 	.get = ceph_get_xattr_handler,
1301 	.set = ceph_set_xattr_handler,
1302 };
1303 
1304 #ifdef CONFIG_SECURITY
1305 bool ceph_security_xattr_wanted(struct inode *in)
1306 {
1307 	return in->i_security != NULL;
1308 }
1309 
1310 bool ceph_security_xattr_deadlock(struct inode *in)
1311 {
1312 	struct ceph_inode_info *ci;
1313 	bool ret;
1314 	if (!in->i_security)
1315 		return false;
1316 	ci = ceph_inode(in);
1317 	spin_lock(&ci->i_ceph_lock);
1318 	ret = !(ci->i_ceph_flags & CEPH_I_SEC_INITED) &&
1319 	      !(ci->i_xattrs.version > 0 &&
1320 		__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 0));
1321 	spin_unlock(&ci->i_ceph_lock);
1322 	return ret;
1323 }
1324 
1325 #ifdef CONFIG_CEPH_FS_SECURITY_LABEL
1326 int ceph_security_init_secctx(struct dentry *dentry, umode_t mode,
1327 			   struct ceph_acl_sec_ctx *as_ctx)
1328 {
1329 	struct ceph_pagelist *pagelist = as_ctx->pagelist;
1330 	const char *name;
1331 	size_t name_len;
1332 	int err;
1333 
1334 	err = security_dentry_init_security(dentry, mode, &dentry->d_name,
1335 					    &name, &as_ctx->sec_ctx,
1336 					    &as_ctx->sec_ctxlen);
1337 	if (err < 0) {
1338 		WARN_ON_ONCE(err != -EOPNOTSUPP);
1339 		err = 0; /* do nothing */
1340 		goto out;
1341 	}
1342 
1343 	err = -ENOMEM;
1344 	if (!pagelist) {
1345 		pagelist = ceph_pagelist_alloc(GFP_KERNEL);
1346 		if (!pagelist)
1347 			goto out;
1348 		err = ceph_pagelist_reserve(pagelist, PAGE_SIZE);
1349 		if (err)
1350 			goto out;
1351 		ceph_pagelist_encode_32(pagelist, 1);
1352 	}
1353 
1354 	/*
1355 	 * FIXME: Make security_dentry_init_security() generic. Currently
1356 	 * It only supports single security module and only selinux has
1357 	 * dentry_init_security hook.
1358 	 */
1359 	name_len = strlen(name);
1360 	err = ceph_pagelist_reserve(pagelist,
1361 				    4 * 2 + name_len + as_ctx->sec_ctxlen);
1362 	if (err)
1363 		goto out;
1364 
1365 	if (as_ctx->pagelist) {
1366 		/* update count of KV pairs */
1367 		BUG_ON(pagelist->length <= sizeof(__le32));
1368 		if (list_is_singular(&pagelist->head)) {
1369 			le32_add_cpu((__le32*)pagelist->mapped_tail, 1);
1370 		} else {
1371 			struct page *page = list_first_entry(&pagelist->head,
1372 							     struct page, lru);
1373 			void *addr = kmap_atomic(page);
1374 			le32_add_cpu((__le32*)addr, 1);
1375 			kunmap_atomic(addr);
1376 		}
1377 	} else {
1378 		as_ctx->pagelist = pagelist;
1379 	}
1380 
1381 	ceph_pagelist_encode_32(pagelist, name_len);
1382 	ceph_pagelist_append(pagelist, name, name_len);
1383 
1384 	ceph_pagelist_encode_32(pagelist, as_ctx->sec_ctxlen);
1385 	ceph_pagelist_append(pagelist, as_ctx->sec_ctx, as_ctx->sec_ctxlen);
1386 
1387 	err = 0;
1388 out:
1389 	if (pagelist && !as_ctx->pagelist)
1390 		ceph_pagelist_release(pagelist);
1391 	return err;
1392 }
1393 #endif /* CONFIG_CEPH_FS_SECURITY_LABEL */
1394 #endif /* CONFIG_SECURITY */
1395 
1396 void ceph_release_acl_sec_ctx(struct ceph_acl_sec_ctx *as_ctx)
1397 {
1398 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1399 	posix_acl_release(as_ctx->acl);
1400 	posix_acl_release(as_ctx->default_acl);
1401 #endif
1402 #ifdef CONFIG_CEPH_FS_SECURITY_LABEL
1403 	security_release_secctx(as_ctx->sec_ctx, as_ctx->sec_ctxlen);
1404 #endif
1405 	if (as_ctx->pagelist)
1406 		ceph_pagelist_release(as_ctx->pagelist);
1407 }
1408 
1409 /*
1410  * List of handlers for synthetic system.* attributes. Other
1411  * attributes are handled directly.
1412  */
1413 const struct xattr_handler *ceph_xattr_handlers[] = {
1414 	&ceph_other_xattr_handler,
1415 	NULL,
1416 };
1417