xref: /openbmc/linux/fs/ceph/xattr.c (revision 52fb57e7) 
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 ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,
718 		      size_t size)
719 {
720 	struct ceph_inode_info *ci = ceph_inode(inode);
721 	int err;
722 	struct ceph_inode_xattr *xattr;
723 	struct ceph_vxattr *vxattr = NULL;
724 
725 	if (!ceph_is_valid_xattr(name))
726 		return -ENODATA;
727 
728 	/* let's see if a virtual xattr was requested */
729 	vxattr = ceph_match_vxattr(inode, name);
730 	if (vxattr && !(vxattr->exists_cb && !vxattr->exists_cb(ci))) {
731 		err = vxattr->getxattr_cb(ci, value, size);
732 		return err;
733 	}
734 
735 	spin_lock(&ci->i_ceph_lock);
736 	dout("getxattr %p ver=%lld index_ver=%lld\n", inode,
737 	     ci->i_xattrs.version, ci->i_xattrs.index_version);
738 
739 	if (ci->i_xattrs.version == 0 ||
740 	    !__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1)) {
741 		spin_unlock(&ci->i_ceph_lock);
742 		/* get xattrs from mds (if we don't already have them) */
743 		err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
744 		if (err)
745 			return err;
746 		spin_lock(&ci->i_ceph_lock);
747 	}
748 
749 	err = __build_xattrs(inode);
750 	if (err < 0)
751 		goto out;
752 
753 	err = -ENODATA;  /* == ENOATTR */
754 	xattr = __get_xattr(ci, name);
755 	if (!xattr)
756 		goto out;
757 
758 	err = -ERANGE;
759 	if (size && size < xattr->val_len)
760 		goto out;
761 
762 	err = xattr->val_len;
763 	if (size == 0)
764 		goto out;
765 
766 	memcpy(value, xattr->val, xattr->val_len);
767 
768 out:
769 	spin_unlock(&ci->i_ceph_lock);
770 	return err;
771 }
772 
773 ssize_t ceph_getxattr(struct dentry *dentry, const char *name, void *value,
774 		      size_t size)
775 {
776 	if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
777 		return generic_getxattr(dentry, name, value, size);
778 
779 	return __ceph_getxattr(d_inode(dentry), name, value, size);
780 }
781 
782 ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)
783 {
784 	struct inode *inode = d_inode(dentry);
785 	struct ceph_inode_info *ci = ceph_inode(inode);
786 	struct ceph_vxattr *vxattrs = ceph_inode_vxattrs(inode);
787 	u32 vir_namelen = 0;
788 	u32 namelen;
789 	int err;
790 	u32 len;
791 	int i;
792 
793 	spin_lock(&ci->i_ceph_lock);
794 	dout("listxattr %p ver=%lld index_ver=%lld\n", inode,
795 	     ci->i_xattrs.version, ci->i_xattrs.index_version);
796 
797 	if (ci->i_xattrs.version == 0 ||
798 	    !__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1)) {
799 		spin_unlock(&ci->i_ceph_lock);
800 		err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
801 		if (err)
802 			return err;
803 		spin_lock(&ci->i_ceph_lock);
804 	}
805 
806 	err = __build_xattrs(inode);
807 	if (err < 0)
808 		goto out;
809 	/*
810 	 * Start with virtual dir xattr names (if any) (including
811 	 * terminating '\0' characters for each).
812 	 */
813 	vir_namelen = ceph_vxattrs_name_size(vxattrs);
814 
815 	/* adding 1 byte per each variable due to the null termination */
816 	namelen = ci->i_xattrs.names_size + ci->i_xattrs.count;
817 	err = -ERANGE;
818 	if (size && vir_namelen + namelen > size)
819 		goto out;
820 
821 	err = namelen + vir_namelen;
822 	if (size == 0)
823 		goto out;
824 
825 	names = __copy_xattr_names(ci, names);
826 
827 	/* virtual xattr names, too */
828 	err = namelen;
829 	if (vxattrs) {
830 		for (i = 0; vxattrs[i].name; i++) {
831 			if (!vxattrs[i].hidden &&
832 			    !(vxattrs[i].exists_cb &&
833 			      !vxattrs[i].exists_cb(ci))) {
834 				len = sprintf(names, "%s", vxattrs[i].name);
835 				names += len + 1;
836 				err += len + 1;
837 			}
838 		}
839 	}
840 
841 out:
842 	spin_unlock(&ci->i_ceph_lock);
843 	return err;
844 }
845 
846 static int ceph_sync_setxattr(struct dentry *dentry, const char *name,
847 			      const char *value, size_t size, int flags)
848 {
849 	struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
850 	struct inode *inode = d_inode(dentry);
851 	struct ceph_inode_info *ci = ceph_inode(inode);
852 	struct ceph_mds_request *req;
853 	struct ceph_mds_client *mdsc = fsc->mdsc;
854 	struct ceph_pagelist *pagelist = NULL;
855 	int err;
856 
857 	if (size > 0) {
858 		/* copy value into pagelist */
859 		pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
860 		if (!pagelist)
861 			return -ENOMEM;
862 
863 		ceph_pagelist_init(pagelist);
864 		err = ceph_pagelist_append(pagelist, value, size);
865 		if (err)
866 			goto out;
867 	} else if (!value) {
868 		flags |= CEPH_XATTR_REMOVE;
869 	}
870 
871 	dout("setxattr value=%.*s\n", (int)size, value);
872 
873 	/* do request */
874 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETXATTR,
875 				       USE_AUTH_MDS);
876 	if (IS_ERR(req)) {
877 		err = PTR_ERR(req);
878 		goto out;
879 	}
880 
881 	req->r_args.setxattr.flags = cpu_to_le32(flags);
882 	req->r_path2 = kstrdup(name, GFP_NOFS);
883 	if (!req->r_path2) {
884 		ceph_mdsc_put_request(req);
885 		err = -ENOMEM;
886 		goto out;
887 	}
888 
889 	req->r_pagelist = pagelist;
890 	pagelist = NULL;
891 
892 	req->r_inode = inode;
893 	ihold(inode);
894 	req->r_num_caps = 1;
895 	req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
896 
897 	dout("xattr.ver (before): %lld\n", ci->i_xattrs.version);
898 	err = ceph_mdsc_do_request(mdsc, NULL, req);
899 	ceph_mdsc_put_request(req);
900 	dout("xattr.ver (after): %lld\n", ci->i_xattrs.version);
901 
902 out:
903 	if (pagelist)
904 		ceph_pagelist_release(pagelist);
905 	return err;
906 }
907 
908 int __ceph_setxattr(struct dentry *dentry, const char *name,
909 			const void *value, size_t size, int flags)
910 {
911 	struct inode *inode = d_inode(dentry);
912 	struct ceph_vxattr *vxattr;
913 	struct ceph_inode_info *ci = ceph_inode(inode);
914 	int issued;
915 	int err;
916 	int dirty = 0;
917 	int name_len = strlen(name);
918 	int val_len = size;
919 	char *newname = NULL;
920 	char *newval = NULL;
921 	struct ceph_inode_xattr *xattr = NULL;
922 	int required_blob_size;
923 
924 	if (!ceph_is_valid_xattr(name))
925 		return -EOPNOTSUPP;
926 
927 	vxattr = ceph_match_vxattr(inode, name);
928 	if (vxattr && vxattr->readonly)
929 		return -EOPNOTSUPP;
930 
931 	/* pass any unhandled ceph.* xattrs through to the MDS */
932 	if (!strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
933 		goto do_sync_unlocked;
934 
935 	/* preallocate memory for xattr name, value, index node */
936 	err = -ENOMEM;
937 	newname = kmemdup(name, name_len + 1, GFP_NOFS);
938 	if (!newname)
939 		goto out;
940 
941 	if (val_len) {
942 		newval = kmemdup(value, val_len, GFP_NOFS);
943 		if (!newval)
944 			goto out;
945 	}
946 
947 	xattr = kmalloc(sizeof(struct ceph_inode_xattr), GFP_NOFS);
948 	if (!xattr)
949 		goto out;
950 
951 	spin_lock(&ci->i_ceph_lock);
952 retry:
953 	issued = __ceph_caps_issued(ci, NULL);
954 	dout("setxattr %p issued %s\n", inode, ceph_cap_string(issued));
955 	if (ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))
956 		goto do_sync;
957 	__build_xattrs(inode);
958 
959 	required_blob_size = __get_required_blob_size(ci, name_len, val_len);
960 
961 	if (!ci->i_xattrs.prealloc_blob ||
962 	    required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
963 		struct ceph_buffer *blob;
964 
965 		spin_unlock(&ci->i_ceph_lock);
966 		dout(" preaallocating new blob size=%d\n", required_blob_size);
967 		blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
968 		if (!blob)
969 			goto out;
970 		spin_lock(&ci->i_ceph_lock);
971 		if (ci->i_xattrs.prealloc_blob)
972 			ceph_buffer_put(ci->i_xattrs.prealloc_blob);
973 		ci->i_xattrs.prealloc_blob = blob;
974 		goto retry;
975 	}
976 
977 	err = __set_xattr(ci, newname, name_len, newval, val_len,
978 			  flags, value ? 1 : -1, &xattr);
979 
980 	if (!err) {
981 		dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL);
982 		ci->i_xattrs.dirty = true;
983 		inode->i_ctime = CURRENT_TIME;
984 	}
985 
986 	spin_unlock(&ci->i_ceph_lock);
987 	if (dirty)
988 		__mark_inode_dirty(inode, dirty);
989 	return err;
990 
991 do_sync:
992 	spin_unlock(&ci->i_ceph_lock);
993 do_sync_unlocked:
994 	err = ceph_sync_setxattr(dentry, name, value, size, flags);
995 out:
996 	kfree(newname);
997 	kfree(newval);
998 	kfree(xattr);
999 	return err;
1000 }
1001 
1002 int ceph_setxattr(struct dentry *dentry, const char *name,
1003 		  const void *value, size_t size, int flags)
1004 {
1005 	if (ceph_snap(d_inode(dentry)) != CEPH_NOSNAP)
1006 		return -EROFS;
1007 
1008 	if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1009 		return generic_setxattr(dentry, name, value, size, flags);
1010 
1011 	if (size == 0)
1012 		value = "";  /* empty EA, do not remove */
1013 
1014 	return __ceph_setxattr(dentry, name, value, size, flags);
1015 }
1016 
1017 static int ceph_send_removexattr(struct dentry *dentry, const char *name)
1018 {
1019 	struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
1020 	struct ceph_mds_client *mdsc = fsc->mdsc;
1021 	struct inode *inode = d_inode(dentry);
1022 	struct ceph_mds_request *req;
1023 	int err;
1024 
1025 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RMXATTR,
1026 				       USE_AUTH_MDS);
1027 	if (IS_ERR(req))
1028 		return PTR_ERR(req);
1029 	req->r_path2 = kstrdup(name, GFP_NOFS);
1030 	if (!req->r_path2)
1031 		return -ENOMEM;
1032 
1033 	req->r_inode = inode;
1034 	ihold(inode);
1035 	req->r_num_caps = 1;
1036 	req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
1037 	err = ceph_mdsc_do_request(mdsc, NULL, req);
1038 	ceph_mdsc_put_request(req);
1039 	return err;
1040 }
1041 
1042 int __ceph_removexattr(struct dentry *dentry, const char *name)
1043 {
1044 	struct inode *inode = d_inode(dentry);
1045 	struct ceph_vxattr *vxattr;
1046 	struct ceph_inode_info *ci = ceph_inode(inode);
1047 	int issued;
1048 	int err;
1049 	int required_blob_size;
1050 	int dirty;
1051 
1052 	if (!ceph_is_valid_xattr(name))
1053 		return -EOPNOTSUPP;
1054 
1055 	vxattr = ceph_match_vxattr(inode, name);
1056 	if (vxattr && vxattr->readonly)
1057 		return -EOPNOTSUPP;
1058 
1059 	/* pass any unhandled ceph.* xattrs through to the MDS */
1060 	if (!strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
1061 		goto do_sync_unlocked;
1062 
1063 	err = -ENOMEM;
1064 	spin_lock(&ci->i_ceph_lock);
1065 retry:
1066 	issued = __ceph_caps_issued(ci, NULL);
1067 	dout("removexattr %p issued %s\n", inode, ceph_cap_string(issued));
1068 
1069 	if (ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))
1070 		goto do_sync;
1071 	__build_xattrs(inode);
1072 
1073 	required_blob_size = __get_required_blob_size(ci, 0, 0);
1074 
1075 	if (!ci->i_xattrs.prealloc_blob ||
1076 	    required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
1077 		struct ceph_buffer *blob;
1078 
1079 		spin_unlock(&ci->i_ceph_lock);
1080 		dout(" preaallocating new blob size=%d\n", required_blob_size);
1081 		blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
1082 		if (!blob)
1083 			goto out;
1084 		spin_lock(&ci->i_ceph_lock);
1085 		if (ci->i_xattrs.prealloc_blob)
1086 			ceph_buffer_put(ci->i_xattrs.prealloc_blob);
1087 		ci->i_xattrs.prealloc_blob = blob;
1088 		goto retry;
1089 	}
1090 
1091 	err = __remove_xattr_by_name(ceph_inode(inode), name);
1092 
1093 	dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL);
1094 	ci->i_xattrs.dirty = true;
1095 	inode->i_ctime = CURRENT_TIME;
1096 	spin_unlock(&ci->i_ceph_lock);
1097 	if (dirty)
1098 		__mark_inode_dirty(inode, dirty);
1099 	return err;
1100 do_sync:
1101 	spin_unlock(&ci->i_ceph_lock);
1102 do_sync_unlocked:
1103 	err = ceph_send_removexattr(dentry, name);
1104 out:
1105 	return err;
1106 }
1107 
1108 int ceph_removexattr(struct dentry *dentry, const char *name)
1109 {
1110 	if (ceph_snap(d_inode(dentry)) != CEPH_NOSNAP)
1111 		return -EROFS;
1112 
1113 	if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1114 		return generic_removexattr(dentry, name);
1115 
1116 	return __ceph_removexattr(dentry, name);
1117 }
1118