xref: /openbmc/linux/fs/ceph/xattr.c (revision 110e6f26) 
1 #include <linux/ceph/ceph_debug.h>
2 #include <linux/ceph/pagelist.h>
3 
4 #include "super.h"
5 #include "mds_client.h"
6 
7 #include <linux/ceph/decode.h>
8 
9 #include <linux/xattr.h>
10 #include <linux/posix_acl_xattr.h>
11 #include <linux/slab.h>
12 
13 #define XATTR_CEPH_PREFIX "ceph."
14 #define XATTR_CEPH_PREFIX_LEN (sizeof (XATTR_CEPH_PREFIX) - 1)
15 
16 static int __remove_xattr(struct ceph_inode_info *ci,
17 			  struct ceph_inode_xattr *xattr);
18 
19 /*
20  * List of handlers for synthetic system.* attributes. Other
21  * attributes are handled directly.
22  */
23 const struct xattr_handler *ceph_xattr_handlers[] = {
24 #ifdef CONFIG_CEPH_FS_POSIX_ACL
25 	&posix_acl_access_xattr_handler,
26 	&posix_acl_default_xattr_handler,
27 #endif
28 	NULL,
29 };
30 
31 static bool ceph_is_valid_xattr(const char *name)
32 {
33 	return !strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN) ||
34 	       !strncmp(name, XATTR_SECURITY_PREFIX,
35 			XATTR_SECURITY_PREFIX_LEN) ||
36 	       !strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN) ||
37 	       !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
38 	       !strncmp(name, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN);
39 }
40 
41 /*
42  * These define virtual xattrs exposing the recursive directory
43  * statistics and layout metadata.
44  */
45 struct ceph_vxattr {
46 	char *name;
47 	size_t name_size;	/* strlen(name) + 1 (for '\0') */
48 	size_t (*getxattr_cb)(struct ceph_inode_info *ci, char *val,
49 			      size_t size);
50 	bool readonly, hidden;
51 	bool (*exists_cb)(struct ceph_inode_info *ci);
52 };
53 
54 /* layouts */
55 
56 static bool ceph_vxattrcb_layout_exists(struct ceph_inode_info *ci)
57 {
58 	size_t s;
59 	char *p = (char *)&ci->i_layout;
60 
61 	for (s = 0; s < sizeof(ci->i_layout); s++, p++)
62 		if (*p)
63 			return true;
64 	return false;
65 }
66 
67 static size_t ceph_vxattrcb_layout(struct ceph_inode_info *ci, char *val,
68 				   size_t size)
69 {
70 	int ret;
71 	struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
72 	struct ceph_osd_client *osdc = &fsc->client->osdc;
73 	s64 pool = ceph_file_layout_pg_pool(ci->i_layout);
74 	const char *pool_name;
75 	char buf[128];
76 
77 	dout("ceph_vxattrcb_layout %p\n", &ci->vfs_inode);
78 	down_read(&osdc->map_sem);
79 	pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
80 	if (pool_name) {
81 		size_t len = strlen(pool_name);
82 		ret = snprintf(buf, sizeof(buf),
83 		"stripe_unit=%lld stripe_count=%lld object_size=%lld pool=",
84 		(unsigned long long)ceph_file_layout_su(ci->i_layout),
85 		(unsigned long long)ceph_file_layout_stripe_count(ci->i_layout),
86 	        (unsigned long long)ceph_file_layout_object_size(ci->i_layout));
87 		if (!size) {
88 			ret += len;
89 		} else if (ret + len > size) {
90 			ret = -ERANGE;
91 		} else {
92 			memcpy(val, buf, ret);
93 			memcpy(val + ret, pool_name, len);
94 			ret += len;
95 		}
96 	} else {
97 		ret = snprintf(buf, sizeof(buf),
98 		"stripe_unit=%lld stripe_count=%lld object_size=%lld pool=%lld",
99 		(unsigned long long)ceph_file_layout_su(ci->i_layout),
100 		(unsigned long long)ceph_file_layout_stripe_count(ci->i_layout),
101 	        (unsigned long long)ceph_file_layout_object_size(ci->i_layout),
102 		(unsigned long long)pool);
103 		if (size) {
104 			if (ret <= size)
105 				memcpy(val, buf, ret);
106 			else
107 				ret = -ERANGE;
108 		}
109 	}
110 	up_read(&osdc->map_sem);
111 	return ret;
112 }
113 
114 static size_t ceph_vxattrcb_layout_stripe_unit(struct ceph_inode_info *ci,
115 					       char *val, size_t size)
116 {
117 	return snprintf(val, size, "%lld",
118 			(unsigned long long)ceph_file_layout_su(ci->i_layout));
119 }
120 
121 static size_t ceph_vxattrcb_layout_stripe_count(struct ceph_inode_info *ci,
122 						char *val, size_t size)
123 {
124 	return snprintf(val, size, "%lld",
125 	       (unsigned long long)ceph_file_layout_stripe_count(ci->i_layout));
126 }
127 
128 static size_t ceph_vxattrcb_layout_object_size(struct ceph_inode_info *ci,
129 					       char *val, size_t size)
130 {
131 	return snprintf(val, size, "%lld",
132 	       (unsigned long long)ceph_file_layout_object_size(ci->i_layout));
133 }
134 
135 static size_t ceph_vxattrcb_layout_pool(struct ceph_inode_info *ci,
136 					char *val, size_t size)
137 {
138 	int ret;
139 	struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
140 	struct ceph_osd_client *osdc = &fsc->client->osdc;
141 	s64 pool = ceph_file_layout_pg_pool(ci->i_layout);
142 	const char *pool_name;
143 
144 	down_read(&osdc->map_sem);
145 	pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
146 	if (pool_name)
147 		ret = snprintf(val, size, "%s", pool_name);
148 	else
149 		ret = snprintf(val, size, "%lld", (unsigned long long)pool);
150 	up_read(&osdc->map_sem);
151 	return ret;
152 }
153 
154 /* directories */
155 
156 static size_t ceph_vxattrcb_dir_entries(struct ceph_inode_info *ci, char *val,
157 					size_t size)
158 {
159 	return snprintf(val, size, "%lld", ci->i_files + ci->i_subdirs);
160 }
161 
162 static size_t ceph_vxattrcb_dir_files(struct ceph_inode_info *ci, char *val,
163 				      size_t size)
164 {
165 	return snprintf(val, size, "%lld", ci->i_files);
166 }
167 
168 static size_t ceph_vxattrcb_dir_subdirs(struct ceph_inode_info *ci, char *val,
169 					size_t size)
170 {
171 	return snprintf(val, size, "%lld", ci->i_subdirs);
172 }
173 
174 static size_t ceph_vxattrcb_dir_rentries(struct ceph_inode_info *ci, char *val,
175 					 size_t size)
176 {
177 	return snprintf(val, size, "%lld", ci->i_rfiles + ci->i_rsubdirs);
178 }
179 
180 static size_t ceph_vxattrcb_dir_rfiles(struct ceph_inode_info *ci, char *val,
181 				       size_t size)
182 {
183 	return snprintf(val, size, "%lld", ci->i_rfiles);
184 }
185 
186 static size_t ceph_vxattrcb_dir_rsubdirs(struct ceph_inode_info *ci, char *val,
187 					 size_t size)
188 {
189 	return snprintf(val, size, "%lld", ci->i_rsubdirs);
190 }
191 
192 static size_t ceph_vxattrcb_dir_rbytes(struct ceph_inode_info *ci, char *val,
193 				       size_t size)
194 {
195 	return snprintf(val, size, "%lld", ci->i_rbytes);
196 }
197 
198 static size_t ceph_vxattrcb_dir_rctime(struct ceph_inode_info *ci, char *val,
199 				       size_t size)
200 {
201 	return snprintf(val, size, "%ld.09%ld", (long)ci->i_rctime.tv_sec,
202 			(long)ci->i_rctime.tv_nsec);
203 }
204 
205 
206 #define CEPH_XATTR_NAME(_type, _name)	XATTR_CEPH_PREFIX #_type "." #_name
207 #define CEPH_XATTR_NAME2(_type, _name, _name2)	\
208 	XATTR_CEPH_PREFIX #_type "." #_name "." #_name2
209 
210 #define XATTR_NAME_CEPH(_type, _name)					\
211 	{								\
212 		.name = CEPH_XATTR_NAME(_type, _name),			\
213 		.name_size = sizeof (CEPH_XATTR_NAME(_type, _name)), \
214 		.getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name, \
215 		.readonly = true,				\
216 		.hidden = false,				\
217 		.exists_cb = NULL,			\
218 	}
219 #define XATTR_LAYOUT_FIELD(_type, _name, _field)			\
220 	{								\
221 		.name = CEPH_XATTR_NAME2(_type, _name, _field),	\
222 		.name_size = sizeof (CEPH_XATTR_NAME2(_type, _name, _field)), \
223 		.getxattr_cb = ceph_vxattrcb_ ## _name ## _ ## _field, \
224 		.readonly = false,				\
225 		.hidden = true,			\
226 		.exists_cb = ceph_vxattrcb_layout_exists,	\
227 	}
228 
229 static struct ceph_vxattr ceph_dir_vxattrs[] = {
230 	{
231 		.name = "ceph.dir.layout",
232 		.name_size = sizeof("ceph.dir.layout"),
233 		.getxattr_cb = ceph_vxattrcb_layout,
234 		.readonly = false,
235 		.hidden = true,
236 		.exists_cb = ceph_vxattrcb_layout_exists,
237 	},
238 	XATTR_LAYOUT_FIELD(dir, layout, stripe_unit),
239 	XATTR_LAYOUT_FIELD(dir, layout, stripe_count),
240 	XATTR_LAYOUT_FIELD(dir, layout, object_size),
241 	XATTR_LAYOUT_FIELD(dir, layout, pool),
242 	XATTR_NAME_CEPH(dir, entries),
243 	XATTR_NAME_CEPH(dir, files),
244 	XATTR_NAME_CEPH(dir, subdirs),
245 	XATTR_NAME_CEPH(dir, rentries),
246 	XATTR_NAME_CEPH(dir, rfiles),
247 	XATTR_NAME_CEPH(dir, rsubdirs),
248 	XATTR_NAME_CEPH(dir, rbytes),
249 	XATTR_NAME_CEPH(dir, rctime),
250 	{ .name = NULL, 0 }	/* Required table terminator */
251 };
252 static size_t ceph_dir_vxattrs_name_size;	/* total size of all names */
253 
254 /* files */
255 
256 static struct ceph_vxattr ceph_file_vxattrs[] = {
257 	{
258 		.name = "ceph.file.layout",
259 		.name_size = sizeof("ceph.file.layout"),
260 		.getxattr_cb = ceph_vxattrcb_layout,
261 		.readonly = false,
262 		.hidden = true,
263 		.exists_cb = ceph_vxattrcb_layout_exists,
264 	},
265 	XATTR_LAYOUT_FIELD(file, layout, stripe_unit),
266 	XATTR_LAYOUT_FIELD(file, layout, stripe_count),
267 	XATTR_LAYOUT_FIELD(file, layout, object_size),
268 	XATTR_LAYOUT_FIELD(file, layout, pool),
269 	{ .name = NULL, 0 }	/* Required table terminator */
270 };
271 static size_t ceph_file_vxattrs_name_size;	/* total size of all names */
272 
273 static struct ceph_vxattr *ceph_inode_vxattrs(struct inode *inode)
274 {
275 	if (S_ISDIR(inode->i_mode))
276 		return ceph_dir_vxattrs;
277 	else if (S_ISREG(inode->i_mode))
278 		return ceph_file_vxattrs;
279 	return NULL;
280 }
281 
282 static size_t ceph_vxattrs_name_size(struct ceph_vxattr *vxattrs)
283 {
284 	if (vxattrs == ceph_dir_vxattrs)
285 		return ceph_dir_vxattrs_name_size;
286 	if (vxattrs == ceph_file_vxattrs)
287 		return ceph_file_vxattrs_name_size;
288 	BUG_ON(vxattrs);
289 	return 0;
290 }
291 
292 /*
293  * Compute the aggregate size (including terminating '\0') of all
294  * virtual extended attribute names in the given vxattr table.
295  */
296 static size_t __init vxattrs_name_size(struct ceph_vxattr *vxattrs)
297 {
298 	struct ceph_vxattr *vxattr;
299 	size_t size = 0;
300 
301 	for (vxattr = vxattrs; vxattr->name; vxattr++)
302 		if (!vxattr->hidden)
303 			size += vxattr->name_size;
304 
305 	return size;
306 }
307 
308 /* Routines called at initialization and exit time */
309 
310 void __init ceph_xattr_init(void)
311 {
312 	ceph_dir_vxattrs_name_size = vxattrs_name_size(ceph_dir_vxattrs);
313 	ceph_file_vxattrs_name_size = vxattrs_name_size(ceph_file_vxattrs);
314 }
315 
316 void ceph_xattr_exit(void)
317 {
318 	ceph_dir_vxattrs_name_size = 0;
319 	ceph_file_vxattrs_name_size = 0;
320 }
321 
322 static struct ceph_vxattr *ceph_match_vxattr(struct inode *inode,
323 						const char *name)
324 {
325 	struct ceph_vxattr *vxattr = ceph_inode_vxattrs(inode);
326 
327 	if (vxattr) {
328 		while (vxattr->name) {
329 			if (!strcmp(vxattr->name, name))
330 				return vxattr;
331 			vxattr++;
332 		}
333 	}
334 
335 	return NULL;
336 }
337 
338 static int __set_xattr(struct ceph_inode_info *ci,
339 			   const char *name, int name_len,
340 			   const char *val, int val_len,
341 			   int flags, int update_xattr,
342 			   struct ceph_inode_xattr **newxattr)
343 {
344 	struct rb_node **p;
345 	struct rb_node *parent = NULL;
346 	struct ceph_inode_xattr *xattr = NULL;
347 	int c;
348 	int new = 0;
349 
350 	p = &ci->i_xattrs.index.rb_node;
351 	while (*p) {
352 		parent = *p;
353 		xattr = rb_entry(parent, struct ceph_inode_xattr, node);
354 		c = strncmp(name, xattr->name, min(name_len, xattr->name_len));
355 		if (c < 0)
356 			p = &(*p)->rb_left;
357 		else if (c > 0)
358 			p = &(*p)->rb_right;
359 		else {
360 			if (name_len == xattr->name_len)
361 				break;
362 			else if (name_len < xattr->name_len)
363 				p = &(*p)->rb_left;
364 			else
365 				p = &(*p)->rb_right;
366 		}
367 		xattr = NULL;
368 	}
369 
370 	if (update_xattr) {
371 		int err = 0;
372 		if (xattr && (flags & XATTR_CREATE))
373 			err = -EEXIST;
374 		else if (!xattr && (flags & XATTR_REPLACE))
375 			err = -ENODATA;
376 		if (err) {
377 			kfree(name);
378 			kfree(val);
379 			return err;
380 		}
381 		if (update_xattr < 0) {
382 			if (xattr)
383 				__remove_xattr(ci, xattr);
384 			kfree(name);
385 			return 0;
386 		}
387 	}
388 
389 	if (!xattr) {
390 		new = 1;
391 		xattr = *newxattr;
392 		xattr->name = name;
393 		xattr->name_len = name_len;
394 		xattr->should_free_name = update_xattr;
395 
396 		ci->i_xattrs.count++;
397 		dout("__set_xattr count=%d\n", ci->i_xattrs.count);
398 	} else {
399 		kfree(*newxattr);
400 		*newxattr = NULL;
401 		if (xattr->should_free_val)
402 			kfree((void *)xattr->val);
403 
404 		if (update_xattr) {
405 			kfree((void *)name);
406 			name = xattr->name;
407 		}
408 		ci->i_xattrs.names_size -= xattr->name_len;
409 		ci->i_xattrs.vals_size -= xattr->val_len;
410 	}
411 	ci->i_xattrs.names_size += name_len;
412 	ci->i_xattrs.vals_size += val_len;
413 	if (val)
414 		xattr->val = val;
415 	else
416 		xattr->val = "";
417 
418 	xattr->val_len = val_len;
419 	xattr->dirty = update_xattr;
420 	xattr->should_free_val = (val && update_xattr);
421 
422 	if (new) {
423 		rb_link_node(&xattr->node, parent, p);
424 		rb_insert_color(&xattr->node, &ci->i_xattrs.index);
425 		dout("__set_xattr_val p=%p\n", p);
426 	}
427 
428 	dout("__set_xattr_val added %llx.%llx xattr %p %s=%.*s\n",
429 	     ceph_vinop(&ci->vfs_inode), xattr, name, val_len, val);
430 
431 	return 0;
432 }
433 
434 static struct ceph_inode_xattr *__get_xattr(struct ceph_inode_info *ci,
435 			   const char *name)
436 {
437 	struct rb_node **p;
438 	struct rb_node *parent = NULL;
439 	struct ceph_inode_xattr *xattr = NULL;
440 	int name_len = strlen(name);
441 	int c;
442 
443 	p = &ci->i_xattrs.index.rb_node;
444 	while (*p) {
445 		parent = *p;
446 		xattr = rb_entry(parent, struct ceph_inode_xattr, node);
447 		c = strncmp(name, xattr->name, xattr->name_len);
448 		if (c == 0 && name_len > xattr->name_len)
449 			c = 1;
450 		if (c < 0)
451 			p = &(*p)->rb_left;
452 		else if (c > 0)
453 			p = &(*p)->rb_right;
454 		else {
455 			dout("__get_xattr %s: found %.*s\n", name,
456 			     xattr->val_len, xattr->val);
457 			return xattr;
458 		}
459 	}
460 
461 	dout("__get_xattr %s: not found\n", name);
462 
463 	return NULL;
464 }
465 
466 static void __free_xattr(struct ceph_inode_xattr *xattr)
467 {
468 	BUG_ON(!xattr);
469 
470 	if (xattr->should_free_name)
471 		kfree((void *)xattr->name);
472 	if (xattr->should_free_val)
473 		kfree((void *)xattr->val);
474 
475 	kfree(xattr);
476 }
477 
478 static int __remove_xattr(struct ceph_inode_info *ci,
479 			  struct ceph_inode_xattr *xattr)
480 {
481 	if (!xattr)
482 		return -ENODATA;
483 
484 	rb_erase(&xattr->node, &ci->i_xattrs.index);
485 
486 	if (xattr->should_free_name)
487 		kfree((void *)xattr->name);
488 	if (xattr->should_free_val)
489 		kfree((void *)xattr->val);
490 
491 	ci->i_xattrs.names_size -= xattr->name_len;
492 	ci->i_xattrs.vals_size -= xattr->val_len;
493 	ci->i_xattrs.count--;
494 	kfree(xattr);
495 
496 	return 0;
497 }
498 
499 static int __remove_xattr_by_name(struct ceph_inode_info *ci,
500 			   const char *name)
501 {
502 	struct rb_node **p;
503 	struct ceph_inode_xattr *xattr;
504 	int err;
505 
506 	p = &ci->i_xattrs.index.rb_node;
507 	xattr = __get_xattr(ci, name);
508 	err = __remove_xattr(ci, xattr);
509 	return err;
510 }
511 
512 static char *__copy_xattr_names(struct ceph_inode_info *ci,
513 				char *dest)
514 {
515 	struct rb_node *p;
516 	struct ceph_inode_xattr *xattr = NULL;
517 
518 	p = rb_first(&ci->i_xattrs.index);
519 	dout("__copy_xattr_names count=%d\n", ci->i_xattrs.count);
520 
521 	while (p) {
522 		xattr = rb_entry(p, struct ceph_inode_xattr, node);
523 		memcpy(dest, xattr->name, xattr->name_len);
524 		dest[xattr->name_len] = '\0';
525 
526 		dout("dest=%s %p (%s) (%d/%d)\n", dest, xattr, xattr->name,
527 		     xattr->name_len, ci->i_xattrs.names_size);
528 
529 		dest += xattr->name_len + 1;
530 		p = rb_next(p);
531 	}
532 
533 	return dest;
534 }
535 
536 void __ceph_destroy_xattrs(struct ceph_inode_info *ci)
537 {
538 	struct rb_node *p, *tmp;
539 	struct ceph_inode_xattr *xattr = NULL;
540 
541 	p = rb_first(&ci->i_xattrs.index);
542 
543 	dout("__ceph_destroy_xattrs p=%p\n", p);
544 
545 	while (p) {
546 		xattr = rb_entry(p, struct ceph_inode_xattr, node);
547 		tmp = p;
548 		p = rb_next(tmp);
549 		dout("__ceph_destroy_xattrs next p=%p (%.*s)\n", p,
550 		     xattr->name_len, xattr->name);
551 		rb_erase(tmp, &ci->i_xattrs.index);
552 
553 		__free_xattr(xattr);
554 	}
555 
556 	ci->i_xattrs.names_size = 0;
557 	ci->i_xattrs.vals_size = 0;
558 	ci->i_xattrs.index_version = 0;
559 	ci->i_xattrs.count = 0;
560 	ci->i_xattrs.index = RB_ROOT;
561 }
562 
563 static int __build_xattrs(struct inode *inode)
564 	__releases(ci->i_ceph_lock)
565 	__acquires(ci->i_ceph_lock)
566 {
567 	u32 namelen;
568 	u32 numattr = 0;
569 	void *p, *end;
570 	u32 len;
571 	const char *name, *val;
572 	struct ceph_inode_info *ci = ceph_inode(inode);
573 	int xattr_version;
574 	struct ceph_inode_xattr **xattrs = NULL;
575 	int err = 0;
576 	int i;
577 
578 	dout("__build_xattrs() len=%d\n",
579 	     ci->i_xattrs.blob ? (int)ci->i_xattrs.blob->vec.iov_len : 0);
580 
581 	if (ci->i_xattrs.index_version >= ci->i_xattrs.version)
582 		return 0; /* already built */
583 
584 	__ceph_destroy_xattrs(ci);
585 
586 start:
587 	/* updated internal xattr rb tree */
588 	if (ci->i_xattrs.blob && ci->i_xattrs.blob->vec.iov_len > 4) {
589 		p = ci->i_xattrs.blob->vec.iov_base;
590 		end = p + ci->i_xattrs.blob->vec.iov_len;
591 		ceph_decode_32_safe(&p, end, numattr, bad);
592 		xattr_version = ci->i_xattrs.version;
593 		spin_unlock(&ci->i_ceph_lock);
594 
595 		xattrs = kcalloc(numattr, sizeof(struct ceph_inode_xattr *),
596 				 GFP_NOFS);
597 		err = -ENOMEM;
598 		if (!xattrs)
599 			goto bad_lock;
600 
601 		for (i = 0; i < numattr; i++) {
602 			xattrs[i] = kmalloc(sizeof(struct ceph_inode_xattr),
603 					    GFP_NOFS);
604 			if (!xattrs[i])
605 				goto bad_lock;
606 		}
607 
608 		spin_lock(&ci->i_ceph_lock);
609 		if (ci->i_xattrs.version != xattr_version) {
610 			/* lost a race, retry */
611 			for (i = 0; i < numattr; i++)
612 				kfree(xattrs[i]);
613 			kfree(xattrs);
614 			xattrs = NULL;
615 			goto start;
616 		}
617 		err = -EIO;
618 		while (numattr--) {
619 			ceph_decode_32_safe(&p, end, len, bad);
620 			namelen = len;
621 			name = p;
622 			p += len;
623 			ceph_decode_32_safe(&p, end, len, bad);
624 			val = p;
625 			p += len;
626 
627 			err = __set_xattr(ci, name, namelen, val, len,
628 					  0, 0, &xattrs[numattr]);
629 
630 			if (err < 0)
631 				goto bad;
632 		}
633 		kfree(xattrs);
634 	}
635 	ci->i_xattrs.index_version = ci->i_xattrs.version;
636 	ci->i_xattrs.dirty = false;
637 
638 	return err;
639 bad_lock:
640 	spin_lock(&ci->i_ceph_lock);
641 bad:
642 	if (xattrs) {
643 		for (i = 0; i < numattr; i++)
644 			kfree(xattrs[i]);
645 		kfree(xattrs);
646 	}
647 	ci->i_xattrs.names_size = 0;
648 	return err;
649 }
650 
651 static int __get_required_blob_size(struct ceph_inode_info *ci, int name_size,
652 				    int val_size)
653 {
654 	/*
655 	 * 4 bytes for the length, and additional 4 bytes per each xattr name,
656 	 * 4 bytes per each value
657 	 */
658 	int size = 4 + ci->i_xattrs.count*(4 + 4) +
659 			     ci->i_xattrs.names_size +
660 			     ci->i_xattrs.vals_size;
661 	dout("__get_required_blob_size c=%d names.size=%d vals.size=%d\n",
662 	     ci->i_xattrs.count, ci->i_xattrs.names_size,
663 	     ci->i_xattrs.vals_size);
664 
665 	if (name_size)
666 		size += 4 + 4 + name_size + val_size;
667 
668 	return size;
669 }
670 
671 /*
672  * If there are dirty xattrs, reencode xattrs into the prealloc_blob
673  * and swap into place.
674  */
675 void __ceph_build_xattrs_blob(struct ceph_inode_info *ci)
676 {
677 	struct rb_node *p;
678 	struct ceph_inode_xattr *xattr = NULL;
679 	void *dest;
680 
681 	dout("__build_xattrs_blob %p\n", &ci->vfs_inode);
682 	if (ci->i_xattrs.dirty) {
683 		int need = __get_required_blob_size(ci, 0, 0);
684 
685 		BUG_ON(need > ci->i_xattrs.prealloc_blob->alloc_len);
686 
687 		p = rb_first(&ci->i_xattrs.index);
688 		dest = ci->i_xattrs.prealloc_blob->vec.iov_base;
689 
690 		ceph_encode_32(&dest, ci->i_xattrs.count);
691 		while (p) {
692 			xattr = rb_entry(p, struct ceph_inode_xattr, node);
693 
694 			ceph_encode_32(&dest, xattr->name_len);
695 			memcpy(dest, xattr->name, xattr->name_len);
696 			dest += xattr->name_len;
697 			ceph_encode_32(&dest, xattr->val_len);
698 			memcpy(dest, xattr->val, xattr->val_len);
699 			dest += xattr->val_len;
700 
701 			p = rb_next(p);
702 		}
703 
704 		/* adjust buffer len; it may be larger than we need */
705 		ci->i_xattrs.prealloc_blob->vec.iov_len =
706 			dest - ci->i_xattrs.prealloc_blob->vec.iov_base;
707 
708 		if (ci->i_xattrs.blob)
709 			ceph_buffer_put(ci->i_xattrs.blob);
710 		ci->i_xattrs.blob = ci->i_xattrs.prealloc_blob;
711 		ci->i_xattrs.prealloc_blob = NULL;
712 		ci->i_xattrs.dirty = false;
713 		ci->i_xattrs.version++;
714 	}
715 }
716 
717 static inline int __get_request_mask(struct inode *in) {
718 	struct ceph_mds_request *req = current->journal_info;
719 	int mask = 0;
720 	if (req && req->r_target_inode == in) {
721 		if (req->r_op == CEPH_MDS_OP_LOOKUP ||
722 		    req->r_op == CEPH_MDS_OP_LOOKUPINO ||
723 		    req->r_op == CEPH_MDS_OP_LOOKUPPARENT ||
724 		    req->r_op == CEPH_MDS_OP_GETATTR) {
725 			mask = le32_to_cpu(req->r_args.getattr.mask);
726 		} else if (req->r_op == CEPH_MDS_OP_OPEN ||
727 			   req->r_op == CEPH_MDS_OP_CREATE) {
728 			mask = le32_to_cpu(req->r_args.open.mask);
729 		}
730 	}
731 	return mask;
732 }
733 
734 ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,
735 		      size_t size)
736 {
737 	struct ceph_inode_info *ci = ceph_inode(inode);
738 	struct ceph_inode_xattr *xattr;
739 	struct ceph_vxattr *vxattr = NULL;
740 	int req_mask;
741 	int err;
742 
743 	if (!ceph_is_valid_xattr(name))
744 		return -ENODATA;
745 
746 	/* let's see if a virtual xattr was requested */
747 	vxattr = ceph_match_vxattr(inode, name);
748 	if (vxattr) {
749 		err = -ENODATA;
750 		if (!(vxattr->exists_cb && !vxattr->exists_cb(ci)))
751 			err = vxattr->getxattr_cb(ci, value, size);
752 		return err;
753 	}
754 
755 	req_mask = __get_request_mask(inode);
756 
757 	spin_lock(&ci->i_ceph_lock);
758 	dout("getxattr %p ver=%lld index_ver=%lld\n", inode,
759 	     ci->i_xattrs.version, ci->i_xattrs.index_version);
760 
761 	if (ci->i_xattrs.version == 0 ||
762 	    !((req_mask & CEPH_CAP_XATTR_SHARED) ||
763 	      __ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1))) {
764 		spin_unlock(&ci->i_ceph_lock);
765 
766 		/* security module gets xattr while filling trace */
767 		if (current->journal_info != NULL) {
768 			pr_warn_ratelimited("sync getxattr %p "
769 					    "during filling trace\n", inode);
770 			return -EBUSY;
771 		}
772 
773 		/* get xattrs from mds (if we don't already have them) */
774 		err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
775 		if (err)
776 			return err;
777 		spin_lock(&ci->i_ceph_lock);
778 	}
779 
780 	err = __build_xattrs(inode);
781 	if (err < 0)
782 		goto out;
783 
784 	err = -ENODATA;  /* == ENOATTR */
785 	xattr = __get_xattr(ci, name);
786 	if (!xattr)
787 		goto out;
788 
789 	err = -ERANGE;
790 	if (size && size < xattr->val_len)
791 		goto out;
792 
793 	err = xattr->val_len;
794 	if (size == 0)
795 		goto out;
796 
797 	memcpy(value, xattr->val, xattr->val_len);
798 
799 	if (current->journal_info != NULL &&
800 	    !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN))
801 		ci->i_ceph_flags |= CEPH_I_SEC_INITED;
802 out:
803 	spin_unlock(&ci->i_ceph_lock);
804 	return err;
805 }
806 
807 ssize_t ceph_getxattr(struct dentry *dentry, const char *name, void *value,
808 		      size_t size)
809 {
810 	if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
811 		return generic_getxattr(dentry, name, value, size);
812 
813 	return __ceph_getxattr(d_inode(dentry), name, value, size);
814 }
815 
816 ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)
817 {
818 	struct inode *inode = d_inode(dentry);
819 	struct ceph_inode_info *ci = ceph_inode(inode);
820 	struct ceph_vxattr *vxattrs = ceph_inode_vxattrs(inode);
821 	u32 vir_namelen = 0;
822 	u32 namelen;
823 	int err;
824 	u32 len;
825 	int i;
826 
827 	spin_lock(&ci->i_ceph_lock);
828 	dout("listxattr %p ver=%lld index_ver=%lld\n", inode,
829 	     ci->i_xattrs.version, ci->i_xattrs.index_version);
830 
831 	if (ci->i_xattrs.version == 0 ||
832 	    !__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1)) {
833 		spin_unlock(&ci->i_ceph_lock);
834 		err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
835 		if (err)
836 			return err;
837 		spin_lock(&ci->i_ceph_lock);
838 	}
839 
840 	err = __build_xattrs(inode);
841 	if (err < 0)
842 		goto out;
843 	/*
844 	 * Start with virtual dir xattr names (if any) (including
845 	 * terminating '\0' characters for each).
846 	 */
847 	vir_namelen = ceph_vxattrs_name_size(vxattrs);
848 
849 	/* adding 1 byte per each variable due to the null termination */
850 	namelen = ci->i_xattrs.names_size + ci->i_xattrs.count;
851 	err = -ERANGE;
852 	if (size && vir_namelen + namelen > size)
853 		goto out;
854 
855 	err = namelen + vir_namelen;
856 	if (size == 0)
857 		goto out;
858 
859 	names = __copy_xattr_names(ci, names);
860 
861 	/* virtual xattr names, too */
862 	err = namelen;
863 	if (vxattrs) {
864 		for (i = 0; vxattrs[i].name; i++) {
865 			if (!vxattrs[i].hidden &&
866 			    !(vxattrs[i].exists_cb &&
867 			      !vxattrs[i].exists_cb(ci))) {
868 				len = sprintf(names, "%s", vxattrs[i].name);
869 				names += len + 1;
870 				err += len + 1;
871 			}
872 		}
873 	}
874 
875 out:
876 	spin_unlock(&ci->i_ceph_lock);
877 	return err;
878 }
879 
880 static int ceph_sync_setxattr(struct dentry *dentry, const char *name,
881 			      const char *value, size_t size, int flags)
882 {
883 	struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
884 	struct inode *inode = d_inode(dentry);
885 	struct ceph_inode_info *ci = ceph_inode(inode);
886 	struct ceph_mds_request *req;
887 	struct ceph_mds_client *mdsc = fsc->mdsc;
888 	struct ceph_pagelist *pagelist = NULL;
889 	int err;
890 
891 	if (size > 0) {
892 		/* copy value into pagelist */
893 		pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
894 		if (!pagelist)
895 			return -ENOMEM;
896 
897 		ceph_pagelist_init(pagelist);
898 		err = ceph_pagelist_append(pagelist, value, size);
899 		if (err)
900 			goto out;
901 	} else if (!value) {
902 		flags |= CEPH_XATTR_REMOVE;
903 	}
904 
905 	dout("setxattr value=%.*s\n", (int)size, value);
906 
907 	/* do request */
908 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETXATTR,
909 				       USE_AUTH_MDS);
910 	if (IS_ERR(req)) {
911 		err = PTR_ERR(req);
912 		goto out;
913 	}
914 
915 	req->r_args.setxattr.flags = cpu_to_le32(flags);
916 	req->r_path2 = kstrdup(name, GFP_NOFS);
917 	if (!req->r_path2) {
918 		ceph_mdsc_put_request(req);
919 		err = -ENOMEM;
920 		goto out;
921 	}
922 
923 	req->r_pagelist = pagelist;
924 	pagelist = NULL;
925 
926 	req->r_inode = inode;
927 	ihold(inode);
928 	req->r_num_caps = 1;
929 	req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
930 
931 	dout("xattr.ver (before): %lld\n", ci->i_xattrs.version);
932 	err = ceph_mdsc_do_request(mdsc, NULL, req);
933 	ceph_mdsc_put_request(req);
934 	dout("xattr.ver (after): %lld\n", ci->i_xattrs.version);
935 
936 out:
937 	if (pagelist)
938 		ceph_pagelist_release(pagelist);
939 	return err;
940 }
941 
942 int __ceph_setxattr(struct dentry *dentry, const char *name,
943 			const void *value, size_t size, int flags)
944 {
945 	struct inode *inode = d_inode(dentry);
946 	struct ceph_vxattr *vxattr;
947 	struct ceph_inode_info *ci = ceph_inode(inode);
948 	struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
949 	struct ceph_cap_flush *prealloc_cf = NULL;
950 	int issued;
951 	int err;
952 	int dirty = 0;
953 	int name_len = strlen(name);
954 	int val_len = size;
955 	char *newname = NULL;
956 	char *newval = NULL;
957 	struct ceph_inode_xattr *xattr = NULL;
958 	int required_blob_size;
959 	bool lock_snap_rwsem = false;
960 
961 	if (!ceph_is_valid_xattr(name))
962 		return -EOPNOTSUPP;
963 
964 	vxattr = ceph_match_vxattr(inode, name);
965 	if (vxattr && vxattr->readonly)
966 		return -EOPNOTSUPP;
967 
968 	/* pass any unhandled ceph.* xattrs through to the MDS */
969 	if (!strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
970 		goto do_sync_unlocked;
971 
972 	/* preallocate memory for xattr name, value, index node */
973 	err = -ENOMEM;
974 	newname = kmemdup(name, name_len + 1, GFP_NOFS);
975 	if (!newname)
976 		goto out;
977 
978 	if (val_len) {
979 		newval = kmemdup(value, val_len, GFP_NOFS);
980 		if (!newval)
981 			goto out;
982 	}
983 
984 	xattr = kmalloc(sizeof(struct ceph_inode_xattr), GFP_NOFS);
985 	if (!xattr)
986 		goto out;
987 
988 	prealloc_cf = ceph_alloc_cap_flush();
989 	if (!prealloc_cf)
990 		goto out;
991 
992 	spin_lock(&ci->i_ceph_lock);
993 retry:
994 	issued = __ceph_caps_issued(ci, NULL);
995 	if (ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))
996 		goto do_sync;
997 
998 	if (!lock_snap_rwsem && !ci->i_head_snapc) {
999 		lock_snap_rwsem = true;
1000 		if (!down_read_trylock(&mdsc->snap_rwsem)) {
1001 			spin_unlock(&ci->i_ceph_lock);
1002 			down_read(&mdsc->snap_rwsem);
1003 			spin_lock(&ci->i_ceph_lock);
1004 			goto retry;
1005 		}
1006 	}
1007 
1008 	dout("setxattr %p issued %s\n", inode, ceph_cap_string(issued));
1009 	__build_xattrs(inode);
1010 
1011 	required_blob_size = __get_required_blob_size(ci, name_len, val_len);
1012 
1013 	if (!ci->i_xattrs.prealloc_blob ||
1014 	    required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
1015 		struct ceph_buffer *blob;
1016 
1017 		spin_unlock(&ci->i_ceph_lock);
1018 		dout(" preaallocating new blob size=%d\n", required_blob_size);
1019 		blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
1020 		if (!blob)
1021 			goto do_sync_unlocked;
1022 		spin_lock(&ci->i_ceph_lock);
1023 		if (ci->i_xattrs.prealloc_blob)
1024 			ceph_buffer_put(ci->i_xattrs.prealloc_blob);
1025 		ci->i_xattrs.prealloc_blob = blob;
1026 		goto retry;
1027 	}
1028 
1029 	err = __set_xattr(ci, newname, name_len, newval, val_len,
1030 			  flags, value ? 1 : -1, &xattr);
1031 
1032 	if (!err) {
1033 		dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL,
1034 					       &prealloc_cf);
1035 		ci->i_xattrs.dirty = true;
1036 		inode->i_ctime = current_fs_time(inode->i_sb);
1037 	}
1038 
1039 	spin_unlock(&ci->i_ceph_lock);
1040 	if (lock_snap_rwsem)
1041 		up_read(&mdsc->snap_rwsem);
1042 	if (dirty)
1043 		__mark_inode_dirty(inode, dirty);
1044 	ceph_free_cap_flush(prealloc_cf);
1045 	return err;
1046 
1047 do_sync:
1048 	spin_unlock(&ci->i_ceph_lock);
1049 do_sync_unlocked:
1050 	if (lock_snap_rwsem)
1051 		up_read(&mdsc->snap_rwsem);
1052 
1053 	/* security module set xattr while filling trace */
1054 	if (current->journal_info != NULL) {
1055 		pr_warn_ratelimited("sync setxattr %p "
1056 				    "during filling trace\n", inode);
1057 		err = -EBUSY;
1058 	} else {
1059 		err = ceph_sync_setxattr(dentry, name, value, size, flags);
1060 	}
1061 out:
1062 	ceph_free_cap_flush(prealloc_cf);
1063 	kfree(newname);
1064 	kfree(newval);
1065 	kfree(xattr);
1066 	return err;
1067 }
1068 
1069 int ceph_setxattr(struct dentry *dentry, const char *name,
1070 		  const void *value, size_t size, int flags)
1071 {
1072 	if (ceph_snap(d_inode(dentry)) != CEPH_NOSNAP)
1073 		return -EROFS;
1074 
1075 	if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1076 		return generic_setxattr(dentry, name, value, size, flags);
1077 
1078 	if (size == 0)
1079 		value = "";  /* empty EA, do not remove */
1080 
1081 	return __ceph_setxattr(dentry, name, value, size, flags);
1082 }
1083 
1084 static int ceph_send_removexattr(struct dentry *dentry, const char *name)
1085 {
1086 	struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
1087 	struct ceph_mds_client *mdsc = fsc->mdsc;
1088 	struct inode *inode = d_inode(dentry);
1089 	struct ceph_mds_request *req;
1090 	int err;
1091 
1092 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RMXATTR,
1093 				       USE_AUTH_MDS);
1094 	if (IS_ERR(req))
1095 		return PTR_ERR(req);
1096 	req->r_path2 = kstrdup(name, GFP_NOFS);
1097 	if (!req->r_path2)
1098 		return -ENOMEM;
1099 
1100 	req->r_inode = inode;
1101 	ihold(inode);
1102 	req->r_num_caps = 1;
1103 	req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
1104 	err = ceph_mdsc_do_request(mdsc, NULL, req);
1105 	ceph_mdsc_put_request(req);
1106 	return err;
1107 }
1108 
1109 int __ceph_removexattr(struct dentry *dentry, const char *name)
1110 {
1111 	struct inode *inode = d_inode(dentry);
1112 	struct ceph_vxattr *vxattr;
1113 	struct ceph_inode_info *ci = ceph_inode(inode);
1114 	struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
1115 	struct ceph_cap_flush *prealloc_cf = NULL;
1116 	int issued;
1117 	int err;
1118 	int required_blob_size;
1119 	int dirty;
1120 	bool lock_snap_rwsem = false;
1121 
1122 	if (!ceph_is_valid_xattr(name))
1123 		return -EOPNOTSUPP;
1124 
1125 	vxattr = ceph_match_vxattr(inode, name);
1126 	if (vxattr && vxattr->readonly)
1127 		return -EOPNOTSUPP;
1128 
1129 	/* pass any unhandled ceph.* xattrs through to the MDS */
1130 	if (!strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
1131 		goto do_sync_unlocked;
1132 
1133 	prealloc_cf = ceph_alloc_cap_flush();
1134 	if (!prealloc_cf)
1135 		return -ENOMEM;
1136 
1137 	err = -ENOMEM;
1138 	spin_lock(&ci->i_ceph_lock);
1139 retry:
1140 	issued = __ceph_caps_issued(ci, NULL);
1141 	if (ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))
1142 		goto do_sync;
1143 
1144 	if (!lock_snap_rwsem && !ci->i_head_snapc) {
1145 		lock_snap_rwsem = true;
1146 		if (!down_read_trylock(&mdsc->snap_rwsem)) {
1147 			spin_unlock(&ci->i_ceph_lock);
1148 			down_read(&mdsc->snap_rwsem);
1149 			spin_lock(&ci->i_ceph_lock);
1150 			goto retry;
1151 		}
1152 	}
1153 
1154 	dout("removexattr %p issued %s\n", inode, ceph_cap_string(issued));
1155 
1156 	__build_xattrs(inode);
1157 
1158 	required_blob_size = __get_required_blob_size(ci, 0, 0);
1159 
1160 	if (!ci->i_xattrs.prealloc_blob ||
1161 	    required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
1162 		struct ceph_buffer *blob;
1163 
1164 		spin_unlock(&ci->i_ceph_lock);
1165 		dout(" preaallocating new blob size=%d\n", required_blob_size);
1166 		blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
1167 		if (!blob)
1168 			goto do_sync_unlocked;
1169 		spin_lock(&ci->i_ceph_lock);
1170 		if (ci->i_xattrs.prealloc_blob)
1171 			ceph_buffer_put(ci->i_xattrs.prealloc_blob);
1172 		ci->i_xattrs.prealloc_blob = blob;
1173 		goto retry;
1174 	}
1175 
1176 	err = __remove_xattr_by_name(ceph_inode(inode), name);
1177 
1178 	dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL,
1179 				       &prealloc_cf);
1180 	ci->i_xattrs.dirty = true;
1181 	inode->i_ctime = current_fs_time(inode->i_sb);
1182 	spin_unlock(&ci->i_ceph_lock);
1183 	if (lock_snap_rwsem)
1184 		up_read(&mdsc->snap_rwsem);
1185 	if (dirty)
1186 		__mark_inode_dirty(inode, dirty);
1187 	ceph_free_cap_flush(prealloc_cf);
1188 	return err;
1189 do_sync:
1190 	spin_unlock(&ci->i_ceph_lock);
1191 do_sync_unlocked:
1192 	if (lock_snap_rwsem)
1193 		up_read(&mdsc->snap_rwsem);
1194 	ceph_free_cap_flush(prealloc_cf);
1195 	err = ceph_send_removexattr(dentry, name);
1196 	return err;
1197 }
1198 
1199 int ceph_removexattr(struct dentry *dentry, const char *name)
1200 {
1201 	if (ceph_snap(d_inode(dentry)) != CEPH_NOSNAP)
1202 		return -EROFS;
1203 
1204 	if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1205 		return generic_removexattr(dentry, name);
1206 
1207 	return __ceph_removexattr(dentry, name);
1208 }
1209 
1210 #ifdef CONFIG_SECURITY
1211 bool ceph_security_xattr_wanted(struct inode *in)
1212 {
1213 	return in->i_security != NULL;
1214 }
1215 
1216 bool ceph_security_xattr_deadlock(struct inode *in)
1217 {
1218 	struct ceph_inode_info *ci;
1219 	bool ret;
1220 	if (in->i_security == NULL)
1221 		return false;
1222 	ci = ceph_inode(in);
1223 	spin_lock(&ci->i_ceph_lock);
1224 	ret = !(ci->i_ceph_flags & CEPH_I_SEC_INITED) &&
1225 	      !(ci->i_xattrs.version > 0 &&
1226 		__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 0));
1227 	spin_unlock(&ci->i_ceph_lock);
1228 	return ret;
1229 }
1230 #endif
1231