xref: /openbmc/linux/fs/ceph/xattr.c (revision 5104d265)
1 #include <linux/ceph/ceph_debug.h>
2 
3 #include "super.h"
4 #include "mds_client.h"
5 
6 #include <linux/ceph/decode.h>
7 
8 #include <linux/xattr.h>
9 #include <linux/slab.h>
10 
11 #define XATTR_CEPH_PREFIX "ceph."
12 #define XATTR_CEPH_PREFIX_LEN (sizeof (XATTR_CEPH_PREFIX) - 1)
13 
14 static bool ceph_is_valid_xattr(const char *name)
15 {
16 	return !strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN) ||
17 	       !strncmp(name, XATTR_SECURITY_PREFIX,
18 			XATTR_SECURITY_PREFIX_LEN) ||
19 	       !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
20 	       !strncmp(name, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN);
21 }
22 
23 /*
24  * These define virtual xattrs exposing the recursive directory
25  * statistics and layout metadata.
26  */
27 struct ceph_vxattr {
28 	char *name;
29 	size_t name_size;	/* strlen(name) + 1 (for '\0') */
30 	size_t (*getxattr_cb)(struct ceph_inode_info *ci, char *val,
31 			      size_t size);
32 	bool readonly, hidden;
33 	bool (*exists_cb)(struct ceph_inode_info *ci);
34 };
35 
36 /* layouts */
37 
38 static bool ceph_vxattrcb_layout_exists(struct ceph_inode_info *ci)
39 {
40 	size_t s;
41 	char *p = (char *)&ci->i_layout;
42 
43 	for (s = 0; s < sizeof(ci->i_layout); s++, p++)
44 		if (*p)
45 			return true;
46 	return false;
47 }
48 
49 static size_t ceph_vxattrcb_layout(struct ceph_inode_info *ci, char *val,
50 					size_t size)
51 {
52 	int ret;
53 	struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
54 	struct ceph_osd_client *osdc = &fsc->client->osdc;
55 	s64 pool = ceph_file_layout_pg_pool(ci->i_layout);
56 	const char *pool_name;
57 
58 	dout("ceph_vxattrcb_layout %p\n", &ci->vfs_inode);
59 	down_read(&osdc->map_sem);
60 	pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
61 	if (pool_name)
62 		ret = snprintf(val, size,
63 		"stripe_unit=%lld stripe_count=%lld object_size=%lld pool=%s",
64 		(unsigned long long)ceph_file_layout_su(ci->i_layout),
65 		(unsigned long long)ceph_file_layout_stripe_count(ci->i_layout),
66 	        (unsigned long long)ceph_file_layout_object_size(ci->i_layout),
67 		pool_name);
68 	else
69 		ret = snprintf(val, size,
70 		"stripe_unit=%lld stripe_count=%lld object_size=%lld pool=%lld",
71 		(unsigned long long)ceph_file_layout_su(ci->i_layout),
72 		(unsigned long long)ceph_file_layout_stripe_count(ci->i_layout),
73 	        (unsigned long long)ceph_file_layout_object_size(ci->i_layout),
74 		(unsigned long long)pool);
75 
76 	up_read(&osdc->map_sem);
77 	return ret;
78 }
79 
80 static size_t ceph_vxattrcb_layout_stripe_unit(struct ceph_inode_info *ci,
81 					       char *val, size_t size)
82 {
83 	return snprintf(val, size, "%lld",
84 			(unsigned long long)ceph_file_layout_su(ci->i_layout));
85 }
86 
87 static size_t ceph_vxattrcb_layout_stripe_count(struct ceph_inode_info *ci,
88 						char *val, size_t size)
89 {
90 	return snprintf(val, size, "%lld",
91 	       (unsigned long long)ceph_file_layout_stripe_count(ci->i_layout));
92 }
93 
94 static size_t ceph_vxattrcb_layout_object_size(struct ceph_inode_info *ci,
95 					       char *val, size_t size)
96 {
97 	return snprintf(val, size, "%lld",
98 	       (unsigned long long)ceph_file_layout_object_size(ci->i_layout));
99 }
100 
101 static size_t ceph_vxattrcb_layout_pool(struct ceph_inode_info *ci,
102 					char *val, size_t size)
103 {
104 	int ret;
105 	struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
106 	struct ceph_osd_client *osdc = &fsc->client->osdc;
107 	s64 pool = ceph_file_layout_pg_pool(ci->i_layout);
108 	const char *pool_name;
109 
110 	down_read(&osdc->map_sem);
111 	pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
112 	if (pool_name)
113 		ret = snprintf(val, size, "%s", pool_name);
114 	else
115 		ret = snprintf(val, size, "%lld", (unsigned long long)pool);
116 	up_read(&osdc->map_sem);
117 	return ret;
118 }
119 
120 /* directories */
121 
122 static size_t ceph_vxattrcb_dir_entries(struct ceph_inode_info *ci, char *val,
123 					size_t size)
124 {
125 	return snprintf(val, size, "%lld", ci->i_files + ci->i_subdirs);
126 }
127 
128 static size_t ceph_vxattrcb_dir_files(struct ceph_inode_info *ci, char *val,
129 				      size_t size)
130 {
131 	return snprintf(val, size, "%lld", ci->i_files);
132 }
133 
134 static size_t ceph_vxattrcb_dir_subdirs(struct ceph_inode_info *ci, char *val,
135 					size_t size)
136 {
137 	return snprintf(val, size, "%lld", ci->i_subdirs);
138 }
139 
140 static size_t ceph_vxattrcb_dir_rentries(struct ceph_inode_info *ci, char *val,
141 					 size_t size)
142 {
143 	return snprintf(val, size, "%lld", ci->i_rfiles + ci->i_rsubdirs);
144 }
145 
146 static size_t ceph_vxattrcb_dir_rfiles(struct ceph_inode_info *ci, char *val,
147 				       size_t size)
148 {
149 	return snprintf(val, size, "%lld", ci->i_rfiles);
150 }
151 
152 static size_t ceph_vxattrcb_dir_rsubdirs(struct ceph_inode_info *ci, char *val,
153 					 size_t size)
154 {
155 	return snprintf(val, size, "%lld", ci->i_rsubdirs);
156 }
157 
158 static size_t ceph_vxattrcb_dir_rbytes(struct ceph_inode_info *ci, char *val,
159 				       size_t size)
160 {
161 	return snprintf(val, size, "%lld", ci->i_rbytes);
162 }
163 
164 static size_t ceph_vxattrcb_dir_rctime(struct ceph_inode_info *ci, char *val,
165 				       size_t size)
166 {
167 	return snprintf(val, size, "%ld.09%ld", (long)ci->i_rctime.tv_sec,
168 			(long)ci->i_rctime.tv_nsec);
169 }
170 
171 
172 #define CEPH_XATTR_NAME(_type, _name)	XATTR_CEPH_PREFIX #_type "." #_name
173 #define CEPH_XATTR_NAME2(_type, _name, _name2)	\
174 	XATTR_CEPH_PREFIX #_type "." #_name "." #_name2
175 
176 #define XATTR_NAME_CEPH(_type, _name)					\
177 	{								\
178 		.name = CEPH_XATTR_NAME(_type, _name),			\
179 		.name_size = sizeof (CEPH_XATTR_NAME(_type, _name)), \
180 		.getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name, \
181 		.readonly = true,				\
182 		.hidden = false,				\
183 		.exists_cb = NULL,			\
184 	}
185 #define XATTR_LAYOUT_FIELD(_type, _name, _field)			\
186 	{								\
187 		.name = CEPH_XATTR_NAME2(_type, _name, _field),	\
188 		.name_size = sizeof (CEPH_XATTR_NAME2(_type, _name, _field)), \
189 		.getxattr_cb = ceph_vxattrcb_ ## _name ## _ ## _field, \
190 		.readonly = false,				\
191 		.hidden = true,			\
192 		.exists_cb = ceph_vxattrcb_layout_exists,	\
193 	}
194 
195 static struct ceph_vxattr ceph_dir_vxattrs[] = {
196 	{
197 		.name = "ceph.dir.layout",
198 		.name_size = sizeof("ceph.dir.layout"),
199 		.getxattr_cb = ceph_vxattrcb_layout,
200 		.readonly = false,
201 		.hidden = false,
202 		.exists_cb = ceph_vxattrcb_layout_exists,
203 	},
204 	XATTR_LAYOUT_FIELD(dir, layout, stripe_unit),
205 	XATTR_LAYOUT_FIELD(dir, layout, stripe_count),
206 	XATTR_LAYOUT_FIELD(dir, layout, object_size),
207 	XATTR_LAYOUT_FIELD(dir, layout, pool),
208 	XATTR_NAME_CEPH(dir, entries),
209 	XATTR_NAME_CEPH(dir, files),
210 	XATTR_NAME_CEPH(dir, subdirs),
211 	XATTR_NAME_CEPH(dir, rentries),
212 	XATTR_NAME_CEPH(dir, rfiles),
213 	XATTR_NAME_CEPH(dir, rsubdirs),
214 	XATTR_NAME_CEPH(dir, rbytes),
215 	XATTR_NAME_CEPH(dir, rctime),
216 	{ .name = NULL, 0 }	/* Required table terminator */
217 };
218 static size_t ceph_dir_vxattrs_name_size;	/* total size of all names */
219 
220 /* files */
221 
222 static struct ceph_vxattr ceph_file_vxattrs[] = {
223 	{
224 		.name = "ceph.file.layout",
225 		.name_size = sizeof("ceph.file.layout"),
226 		.getxattr_cb = ceph_vxattrcb_layout,
227 		.readonly = false,
228 		.hidden = false,
229 		.exists_cb = ceph_vxattrcb_layout_exists,
230 	},
231 	XATTR_LAYOUT_FIELD(file, layout, stripe_unit),
232 	XATTR_LAYOUT_FIELD(file, layout, stripe_count),
233 	XATTR_LAYOUT_FIELD(file, layout, object_size),
234 	XATTR_LAYOUT_FIELD(file, layout, pool),
235 	{ .name = NULL, 0 }	/* Required table terminator */
236 };
237 static size_t ceph_file_vxattrs_name_size;	/* total size of all names */
238 
239 static struct ceph_vxattr *ceph_inode_vxattrs(struct inode *inode)
240 {
241 	if (S_ISDIR(inode->i_mode))
242 		return ceph_dir_vxattrs;
243 	else if (S_ISREG(inode->i_mode))
244 		return ceph_file_vxattrs;
245 	return NULL;
246 }
247 
248 static size_t ceph_vxattrs_name_size(struct ceph_vxattr *vxattrs)
249 {
250 	if (vxattrs == ceph_dir_vxattrs)
251 		return ceph_dir_vxattrs_name_size;
252 	if (vxattrs == ceph_file_vxattrs)
253 		return ceph_file_vxattrs_name_size;
254 	BUG();
255 
256 	return 0;
257 }
258 
259 /*
260  * Compute the aggregate size (including terminating '\0') of all
261  * virtual extended attribute names in the given vxattr table.
262  */
263 static size_t __init vxattrs_name_size(struct ceph_vxattr *vxattrs)
264 {
265 	struct ceph_vxattr *vxattr;
266 	size_t size = 0;
267 
268 	for (vxattr = vxattrs; vxattr->name; vxattr++)
269 		if (!vxattr->hidden)
270 			size += vxattr->name_size;
271 
272 	return size;
273 }
274 
275 /* Routines called at initialization and exit time */
276 
277 void __init ceph_xattr_init(void)
278 {
279 	ceph_dir_vxattrs_name_size = vxattrs_name_size(ceph_dir_vxattrs);
280 	ceph_file_vxattrs_name_size = vxattrs_name_size(ceph_file_vxattrs);
281 }
282 
283 void ceph_xattr_exit(void)
284 {
285 	ceph_dir_vxattrs_name_size = 0;
286 	ceph_file_vxattrs_name_size = 0;
287 }
288 
289 static struct ceph_vxattr *ceph_match_vxattr(struct inode *inode,
290 						const char *name)
291 {
292 	struct ceph_vxattr *vxattr = ceph_inode_vxattrs(inode);
293 
294 	if (vxattr) {
295 		while (vxattr->name) {
296 			if (!strcmp(vxattr->name, name))
297 				return vxattr;
298 			vxattr++;
299 		}
300 	}
301 
302 	return NULL;
303 }
304 
305 static int __set_xattr(struct ceph_inode_info *ci,
306 			   const char *name, int name_len,
307 			   const char *val, int val_len,
308 			   int dirty,
309 			   int should_free_name, int should_free_val,
310 			   struct ceph_inode_xattr **newxattr)
311 {
312 	struct rb_node **p;
313 	struct rb_node *parent = NULL;
314 	struct ceph_inode_xattr *xattr = NULL;
315 	int c;
316 	int new = 0;
317 
318 	p = &ci->i_xattrs.index.rb_node;
319 	while (*p) {
320 		parent = *p;
321 		xattr = rb_entry(parent, struct ceph_inode_xattr, node);
322 		c = strncmp(name, xattr->name, min(name_len, xattr->name_len));
323 		if (c < 0)
324 			p = &(*p)->rb_left;
325 		else if (c > 0)
326 			p = &(*p)->rb_right;
327 		else {
328 			if (name_len == xattr->name_len)
329 				break;
330 			else if (name_len < xattr->name_len)
331 				p = &(*p)->rb_left;
332 			else
333 				p = &(*p)->rb_right;
334 		}
335 		xattr = NULL;
336 	}
337 
338 	if (!xattr) {
339 		new = 1;
340 		xattr = *newxattr;
341 		xattr->name = name;
342 		xattr->name_len = name_len;
343 		xattr->should_free_name = should_free_name;
344 
345 		ci->i_xattrs.count++;
346 		dout("__set_xattr count=%d\n", ci->i_xattrs.count);
347 	} else {
348 		kfree(*newxattr);
349 		*newxattr = NULL;
350 		if (xattr->should_free_val)
351 			kfree((void *)xattr->val);
352 
353 		if (should_free_name) {
354 			kfree((void *)name);
355 			name = xattr->name;
356 		}
357 		ci->i_xattrs.names_size -= xattr->name_len;
358 		ci->i_xattrs.vals_size -= xattr->val_len;
359 	}
360 	ci->i_xattrs.names_size += name_len;
361 	ci->i_xattrs.vals_size += val_len;
362 	if (val)
363 		xattr->val = val;
364 	else
365 		xattr->val = "";
366 
367 	xattr->val_len = val_len;
368 	xattr->dirty = dirty;
369 	xattr->should_free_val = (val && should_free_val);
370 
371 	if (new) {
372 		rb_link_node(&xattr->node, parent, p);
373 		rb_insert_color(&xattr->node, &ci->i_xattrs.index);
374 		dout("__set_xattr_val p=%p\n", p);
375 	}
376 
377 	dout("__set_xattr_val added %llx.%llx xattr %p %s=%.*s\n",
378 	     ceph_vinop(&ci->vfs_inode), xattr, name, val_len, val);
379 
380 	return 0;
381 }
382 
383 static struct ceph_inode_xattr *__get_xattr(struct ceph_inode_info *ci,
384 			   const char *name)
385 {
386 	struct rb_node **p;
387 	struct rb_node *parent = NULL;
388 	struct ceph_inode_xattr *xattr = NULL;
389 	int name_len = strlen(name);
390 	int c;
391 
392 	p = &ci->i_xattrs.index.rb_node;
393 	while (*p) {
394 		parent = *p;
395 		xattr = rb_entry(parent, struct ceph_inode_xattr, node);
396 		c = strncmp(name, xattr->name, xattr->name_len);
397 		if (c == 0 && name_len > xattr->name_len)
398 			c = 1;
399 		if (c < 0)
400 			p = &(*p)->rb_left;
401 		else if (c > 0)
402 			p = &(*p)->rb_right;
403 		else {
404 			dout("__get_xattr %s: found %.*s\n", name,
405 			     xattr->val_len, xattr->val);
406 			return xattr;
407 		}
408 	}
409 
410 	dout("__get_xattr %s: not found\n", name);
411 
412 	return NULL;
413 }
414 
415 static void __free_xattr(struct ceph_inode_xattr *xattr)
416 {
417 	BUG_ON(!xattr);
418 
419 	if (xattr->should_free_name)
420 		kfree((void *)xattr->name);
421 	if (xattr->should_free_val)
422 		kfree((void *)xattr->val);
423 
424 	kfree(xattr);
425 }
426 
427 static int __remove_xattr(struct ceph_inode_info *ci,
428 			  struct ceph_inode_xattr *xattr)
429 {
430 	if (!xattr)
431 		return -EOPNOTSUPP;
432 
433 	rb_erase(&xattr->node, &ci->i_xattrs.index);
434 
435 	if (xattr->should_free_name)
436 		kfree((void *)xattr->name);
437 	if (xattr->should_free_val)
438 		kfree((void *)xattr->val);
439 
440 	ci->i_xattrs.names_size -= xattr->name_len;
441 	ci->i_xattrs.vals_size -= xattr->val_len;
442 	ci->i_xattrs.count--;
443 	kfree(xattr);
444 
445 	return 0;
446 }
447 
448 static int __remove_xattr_by_name(struct ceph_inode_info *ci,
449 			   const char *name)
450 {
451 	struct rb_node **p;
452 	struct ceph_inode_xattr *xattr;
453 	int err;
454 
455 	p = &ci->i_xattrs.index.rb_node;
456 	xattr = __get_xattr(ci, name);
457 	err = __remove_xattr(ci, xattr);
458 	return err;
459 }
460 
461 static char *__copy_xattr_names(struct ceph_inode_info *ci,
462 				char *dest)
463 {
464 	struct rb_node *p;
465 	struct ceph_inode_xattr *xattr = NULL;
466 
467 	p = rb_first(&ci->i_xattrs.index);
468 	dout("__copy_xattr_names count=%d\n", ci->i_xattrs.count);
469 
470 	while (p) {
471 		xattr = rb_entry(p, struct ceph_inode_xattr, node);
472 		memcpy(dest, xattr->name, xattr->name_len);
473 		dest[xattr->name_len] = '\0';
474 
475 		dout("dest=%s %p (%s) (%d/%d)\n", dest, xattr, xattr->name,
476 		     xattr->name_len, ci->i_xattrs.names_size);
477 
478 		dest += xattr->name_len + 1;
479 		p = rb_next(p);
480 	}
481 
482 	return dest;
483 }
484 
485 void __ceph_destroy_xattrs(struct ceph_inode_info *ci)
486 {
487 	struct rb_node *p, *tmp;
488 	struct ceph_inode_xattr *xattr = NULL;
489 
490 	p = rb_first(&ci->i_xattrs.index);
491 
492 	dout("__ceph_destroy_xattrs p=%p\n", p);
493 
494 	while (p) {
495 		xattr = rb_entry(p, struct ceph_inode_xattr, node);
496 		tmp = p;
497 		p = rb_next(tmp);
498 		dout("__ceph_destroy_xattrs next p=%p (%.*s)\n", p,
499 		     xattr->name_len, xattr->name);
500 		rb_erase(tmp, &ci->i_xattrs.index);
501 
502 		__free_xattr(xattr);
503 	}
504 
505 	ci->i_xattrs.names_size = 0;
506 	ci->i_xattrs.vals_size = 0;
507 	ci->i_xattrs.index_version = 0;
508 	ci->i_xattrs.count = 0;
509 	ci->i_xattrs.index = RB_ROOT;
510 }
511 
512 static int __build_xattrs(struct inode *inode)
513 	__releases(ci->i_ceph_lock)
514 	__acquires(ci->i_ceph_lock)
515 {
516 	u32 namelen;
517 	u32 numattr = 0;
518 	void *p, *end;
519 	u32 len;
520 	const char *name, *val;
521 	struct ceph_inode_info *ci = ceph_inode(inode);
522 	int xattr_version;
523 	struct ceph_inode_xattr **xattrs = NULL;
524 	int err = 0;
525 	int i;
526 
527 	dout("__build_xattrs() len=%d\n",
528 	     ci->i_xattrs.blob ? (int)ci->i_xattrs.blob->vec.iov_len : 0);
529 
530 	if (ci->i_xattrs.index_version >= ci->i_xattrs.version)
531 		return 0; /* already built */
532 
533 	__ceph_destroy_xattrs(ci);
534 
535 start:
536 	/* updated internal xattr rb tree */
537 	if (ci->i_xattrs.blob && ci->i_xattrs.blob->vec.iov_len > 4) {
538 		p = ci->i_xattrs.blob->vec.iov_base;
539 		end = p + ci->i_xattrs.blob->vec.iov_len;
540 		ceph_decode_32_safe(&p, end, numattr, bad);
541 		xattr_version = ci->i_xattrs.version;
542 		spin_unlock(&ci->i_ceph_lock);
543 
544 		xattrs = kcalloc(numattr, sizeof(struct ceph_xattr *),
545 				 GFP_NOFS);
546 		err = -ENOMEM;
547 		if (!xattrs)
548 			goto bad_lock;
549 		memset(xattrs, 0, numattr*sizeof(struct ceph_xattr *));
550 		for (i = 0; i < numattr; i++) {
551 			xattrs[i] = kmalloc(sizeof(struct ceph_inode_xattr),
552 					    GFP_NOFS);
553 			if (!xattrs[i])
554 				goto bad_lock;
555 		}
556 
557 		spin_lock(&ci->i_ceph_lock);
558 		if (ci->i_xattrs.version != xattr_version) {
559 			/* lost a race, retry */
560 			for (i = 0; i < numattr; i++)
561 				kfree(xattrs[i]);
562 			kfree(xattrs);
563 			xattrs = NULL;
564 			goto start;
565 		}
566 		err = -EIO;
567 		while (numattr--) {
568 			ceph_decode_32_safe(&p, end, len, bad);
569 			namelen = len;
570 			name = p;
571 			p += len;
572 			ceph_decode_32_safe(&p, end, len, bad);
573 			val = p;
574 			p += len;
575 
576 			err = __set_xattr(ci, name, namelen, val, len,
577 					  0, 0, 0, &xattrs[numattr]);
578 
579 			if (err < 0)
580 				goto bad;
581 		}
582 		kfree(xattrs);
583 	}
584 	ci->i_xattrs.index_version = ci->i_xattrs.version;
585 	ci->i_xattrs.dirty = false;
586 
587 	return err;
588 bad_lock:
589 	spin_lock(&ci->i_ceph_lock);
590 bad:
591 	if (xattrs) {
592 		for (i = 0; i < numattr; i++)
593 			kfree(xattrs[i]);
594 		kfree(xattrs);
595 	}
596 	ci->i_xattrs.names_size = 0;
597 	return err;
598 }
599 
600 static int __get_required_blob_size(struct ceph_inode_info *ci, int name_size,
601 				    int val_size)
602 {
603 	/*
604 	 * 4 bytes for the length, and additional 4 bytes per each xattr name,
605 	 * 4 bytes per each value
606 	 */
607 	int size = 4 + ci->i_xattrs.count*(4 + 4) +
608 			     ci->i_xattrs.names_size +
609 			     ci->i_xattrs.vals_size;
610 	dout("__get_required_blob_size c=%d names.size=%d vals.size=%d\n",
611 	     ci->i_xattrs.count, ci->i_xattrs.names_size,
612 	     ci->i_xattrs.vals_size);
613 
614 	if (name_size)
615 		size += 4 + 4 + name_size + val_size;
616 
617 	return size;
618 }
619 
620 /*
621  * If there are dirty xattrs, reencode xattrs into the prealloc_blob
622  * and swap into place.
623  */
624 void __ceph_build_xattrs_blob(struct ceph_inode_info *ci)
625 {
626 	struct rb_node *p;
627 	struct ceph_inode_xattr *xattr = NULL;
628 	void *dest;
629 
630 	dout("__build_xattrs_blob %p\n", &ci->vfs_inode);
631 	if (ci->i_xattrs.dirty) {
632 		int need = __get_required_blob_size(ci, 0, 0);
633 
634 		BUG_ON(need > ci->i_xattrs.prealloc_blob->alloc_len);
635 
636 		p = rb_first(&ci->i_xattrs.index);
637 		dest = ci->i_xattrs.prealloc_blob->vec.iov_base;
638 
639 		ceph_encode_32(&dest, ci->i_xattrs.count);
640 		while (p) {
641 			xattr = rb_entry(p, struct ceph_inode_xattr, node);
642 
643 			ceph_encode_32(&dest, xattr->name_len);
644 			memcpy(dest, xattr->name, xattr->name_len);
645 			dest += xattr->name_len;
646 			ceph_encode_32(&dest, xattr->val_len);
647 			memcpy(dest, xattr->val, xattr->val_len);
648 			dest += xattr->val_len;
649 
650 			p = rb_next(p);
651 		}
652 
653 		/* adjust buffer len; it may be larger than we need */
654 		ci->i_xattrs.prealloc_blob->vec.iov_len =
655 			dest - ci->i_xattrs.prealloc_blob->vec.iov_base;
656 
657 		if (ci->i_xattrs.blob)
658 			ceph_buffer_put(ci->i_xattrs.blob);
659 		ci->i_xattrs.blob = ci->i_xattrs.prealloc_blob;
660 		ci->i_xattrs.prealloc_blob = NULL;
661 		ci->i_xattrs.dirty = false;
662 		ci->i_xattrs.version++;
663 	}
664 }
665 
666 ssize_t ceph_getxattr(struct dentry *dentry, const char *name, void *value,
667 		      size_t size)
668 {
669 	struct inode *inode = dentry->d_inode;
670 	struct ceph_inode_info *ci = ceph_inode(inode);
671 	int err;
672 	struct ceph_inode_xattr *xattr;
673 	struct ceph_vxattr *vxattr = NULL;
674 
675 	if (!ceph_is_valid_xattr(name))
676 		return -ENODATA;
677 
678 
679 	/* let's see if a virtual xattr was requested */
680 	vxattr = ceph_match_vxattr(inode, name);
681 	if (vxattr && !(vxattr->exists_cb && !vxattr->exists_cb(ci))) {
682 		err = vxattr->getxattr_cb(ci, value, size);
683 		return err;
684 	}
685 
686 	spin_lock(&ci->i_ceph_lock);
687 	dout("getxattr %p ver=%lld index_ver=%lld\n", inode,
688 	     ci->i_xattrs.version, ci->i_xattrs.index_version);
689 
690 	if (__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1) &&
691 	    (ci->i_xattrs.index_version >= ci->i_xattrs.version)) {
692 		goto get_xattr;
693 	} else {
694 		spin_unlock(&ci->i_ceph_lock);
695 		/* get xattrs from mds (if we don't already have them) */
696 		err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR);
697 		if (err)
698 			return err;
699 	}
700 
701 	spin_lock(&ci->i_ceph_lock);
702 
703 	err = __build_xattrs(inode);
704 	if (err < 0)
705 		goto out;
706 
707 get_xattr:
708 	err = -ENODATA;  /* == ENOATTR */
709 	xattr = __get_xattr(ci, name);
710 	if (!xattr)
711 		goto out;
712 
713 	err = -ERANGE;
714 	if (size && size < xattr->val_len)
715 		goto out;
716 
717 	err = xattr->val_len;
718 	if (size == 0)
719 		goto out;
720 
721 	memcpy(value, xattr->val, xattr->val_len);
722 
723 out:
724 	spin_unlock(&ci->i_ceph_lock);
725 	return err;
726 }
727 
728 ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)
729 {
730 	struct inode *inode = dentry->d_inode;
731 	struct ceph_inode_info *ci = ceph_inode(inode);
732 	struct ceph_vxattr *vxattrs = ceph_inode_vxattrs(inode);
733 	u32 vir_namelen = 0;
734 	u32 namelen;
735 	int err;
736 	u32 len;
737 	int i;
738 
739 	spin_lock(&ci->i_ceph_lock);
740 	dout("listxattr %p ver=%lld index_ver=%lld\n", inode,
741 	     ci->i_xattrs.version, ci->i_xattrs.index_version);
742 
743 	if (__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1) &&
744 	    (ci->i_xattrs.index_version >= ci->i_xattrs.version)) {
745 		goto list_xattr;
746 	} else {
747 		spin_unlock(&ci->i_ceph_lock);
748 		err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR);
749 		if (err)
750 			return err;
751 	}
752 
753 	spin_lock(&ci->i_ceph_lock);
754 
755 	err = __build_xattrs(inode);
756 	if (err < 0)
757 		goto out;
758 
759 list_xattr:
760 	/*
761 	 * Start with virtual dir xattr names (if any) (including
762 	 * terminating '\0' characters for each).
763 	 */
764 	vir_namelen = ceph_vxattrs_name_size(vxattrs);
765 
766 	/* adding 1 byte per each variable due to the null termination */
767 	namelen = ci->i_xattrs.names_size + ci->i_xattrs.count;
768 	err = -ERANGE;
769 	if (size && vir_namelen + namelen > size)
770 		goto out;
771 
772 	err = namelen + vir_namelen;
773 	if (size == 0)
774 		goto out;
775 
776 	names = __copy_xattr_names(ci, names);
777 
778 	/* virtual xattr names, too */
779 	err = namelen;
780 	if (vxattrs) {
781 		for (i = 0; vxattrs[i].name; i++) {
782 			if (!vxattrs[i].hidden &&
783 			    !(vxattrs[i].exists_cb &&
784 			      !vxattrs[i].exists_cb(ci))) {
785 				len = sprintf(names, "%s", vxattrs[i].name);
786 				names += len + 1;
787 				err += len + 1;
788 			}
789 		}
790 	}
791 
792 out:
793 	spin_unlock(&ci->i_ceph_lock);
794 	return err;
795 }
796 
797 static int ceph_sync_setxattr(struct dentry *dentry, const char *name,
798 			      const char *value, size_t size, int flags)
799 {
800 	struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
801 	struct inode *inode = dentry->d_inode;
802 	struct ceph_inode_info *ci = ceph_inode(inode);
803 	struct inode *parent_inode;
804 	struct ceph_mds_request *req;
805 	struct ceph_mds_client *mdsc = fsc->mdsc;
806 	int err;
807 	int i, nr_pages;
808 	struct page **pages = NULL;
809 	void *kaddr;
810 
811 	/* copy value into some pages */
812 	nr_pages = calc_pages_for(0, size);
813 	if (nr_pages) {
814 		pages = kmalloc(sizeof(pages[0])*nr_pages, GFP_NOFS);
815 		if (!pages)
816 			return -ENOMEM;
817 		err = -ENOMEM;
818 		for (i = 0; i < nr_pages; i++) {
819 			pages[i] = __page_cache_alloc(GFP_NOFS);
820 			if (!pages[i]) {
821 				nr_pages = i;
822 				goto out;
823 			}
824 			kaddr = kmap(pages[i]);
825 			memcpy(kaddr, value + i*PAGE_CACHE_SIZE,
826 			       min(PAGE_CACHE_SIZE, size-i*PAGE_CACHE_SIZE));
827 		}
828 	}
829 
830 	dout("setxattr value=%.*s\n", (int)size, value);
831 
832 	/* do request */
833 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETXATTR,
834 				       USE_AUTH_MDS);
835 	if (IS_ERR(req)) {
836 		err = PTR_ERR(req);
837 		goto out;
838 	}
839 	req->r_inode = inode;
840 	ihold(inode);
841 	req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
842 	req->r_num_caps = 1;
843 	req->r_args.setxattr.flags = cpu_to_le32(flags);
844 	req->r_path2 = kstrdup(name, GFP_NOFS);
845 
846 	req->r_pages = pages;
847 	req->r_num_pages = nr_pages;
848 	req->r_data_len = size;
849 
850 	dout("xattr.ver (before): %lld\n", ci->i_xattrs.version);
851 	parent_inode = ceph_get_dentry_parent_inode(dentry);
852 	err = ceph_mdsc_do_request(mdsc, parent_inode, req);
853 	iput(parent_inode);
854 	ceph_mdsc_put_request(req);
855 	dout("xattr.ver (after): %lld\n", ci->i_xattrs.version);
856 
857 out:
858 	if (pages) {
859 		for (i = 0; i < nr_pages; i++)
860 			__free_page(pages[i]);
861 		kfree(pages);
862 	}
863 	return err;
864 }
865 
866 int ceph_setxattr(struct dentry *dentry, const char *name,
867 		  const void *value, size_t size, int flags)
868 {
869 	struct inode *inode = dentry->d_inode;
870 	struct ceph_vxattr *vxattr;
871 	struct ceph_inode_info *ci = ceph_inode(inode);
872 	int issued;
873 	int err;
874 	int dirty;
875 	int name_len = strlen(name);
876 	int val_len = size;
877 	char *newname = NULL;
878 	char *newval = NULL;
879 	struct ceph_inode_xattr *xattr = NULL;
880 	int required_blob_size;
881 
882 	if (ceph_snap(inode) != CEPH_NOSNAP)
883 		return -EROFS;
884 
885 	if (!ceph_is_valid_xattr(name))
886 		return -EOPNOTSUPP;
887 
888 	vxattr = ceph_match_vxattr(inode, name);
889 	if (vxattr && vxattr->readonly)
890 		return -EOPNOTSUPP;
891 
892 	/* pass any unhandled ceph.* xattrs through to the MDS */
893 	if (!strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
894 		goto do_sync_unlocked;
895 
896 	/* preallocate memory for xattr name, value, index node */
897 	err = -ENOMEM;
898 	newname = kmemdup(name, name_len + 1, GFP_NOFS);
899 	if (!newname)
900 		goto out;
901 
902 	if (val_len) {
903 		newval = kmemdup(value, val_len, GFP_NOFS);
904 		if (!newval)
905 			goto out;
906 	}
907 
908 	xattr = kmalloc(sizeof(struct ceph_inode_xattr), GFP_NOFS);
909 	if (!xattr)
910 		goto out;
911 
912 	spin_lock(&ci->i_ceph_lock);
913 retry:
914 	issued = __ceph_caps_issued(ci, NULL);
915 	dout("setxattr %p issued %s\n", inode, ceph_cap_string(issued));
916 	if (!(issued & CEPH_CAP_XATTR_EXCL))
917 		goto do_sync;
918 	__build_xattrs(inode);
919 
920 	required_blob_size = __get_required_blob_size(ci, name_len, val_len);
921 
922 	if (!ci->i_xattrs.prealloc_blob ||
923 	    required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
924 		struct ceph_buffer *blob;
925 
926 		spin_unlock(&ci->i_ceph_lock);
927 		dout(" preaallocating new blob size=%d\n", required_blob_size);
928 		blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
929 		if (!blob)
930 			goto out;
931 		spin_lock(&ci->i_ceph_lock);
932 		if (ci->i_xattrs.prealloc_blob)
933 			ceph_buffer_put(ci->i_xattrs.prealloc_blob);
934 		ci->i_xattrs.prealloc_blob = blob;
935 		goto retry;
936 	}
937 
938 	err = __set_xattr(ci, newname, name_len, newval,
939 			  val_len, 1, 1, 1, &xattr);
940 
941 	dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL);
942 	ci->i_xattrs.dirty = true;
943 	inode->i_ctime = CURRENT_TIME;
944 
945 	spin_unlock(&ci->i_ceph_lock);
946 	if (dirty)
947 		__mark_inode_dirty(inode, dirty);
948 	return err;
949 
950 do_sync:
951 	spin_unlock(&ci->i_ceph_lock);
952 do_sync_unlocked:
953 	err = ceph_sync_setxattr(dentry, name, value, size, flags);
954 out:
955 	kfree(newname);
956 	kfree(newval);
957 	kfree(xattr);
958 	return err;
959 }
960 
961 static int ceph_send_removexattr(struct dentry *dentry, const char *name)
962 {
963 	struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
964 	struct ceph_mds_client *mdsc = fsc->mdsc;
965 	struct inode *inode = dentry->d_inode;
966 	struct inode *parent_inode;
967 	struct ceph_mds_request *req;
968 	int err;
969 
970 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RMXATTR,
971 				       USE_AUTH_MDS);
972 	if (IS_ERR(req))
973 		return PTR_ERR(req);
974 	req->r_inode = inode;
975 	ihold(inode);
976 	req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
977 	req->r_num_caps = 1;
978 	req->r_path2 = kstrdup(name, GFP_NOFS);
979 
980 	parent_inode = ceph_get_dentry_parent_inode(dentry);
981 	err = ceph_mdsc_do_request(mdsc, parent_inode, req);
982 	iput(parent_inode);
983 	ceph_mdsc_put_request(req);
984 	return err;
985 }
986 
987 int ceph_removexattr(struct dentry *dentry, const char *name)
988 {
989 	struct inode *inode = dentry->d_inode;
990 	struct ceph_vxattr *vxattr;
991 	struct ceph_inode_info *ci = ceph_inode(inode);
992 	int issued;
993 	int err;
994 	int required_blob_size;
995 	int dirty;
996 
997 	if (ceph_snap(inode) != CEPH_NOSNAP)
998 		return -EROFS;
999 
1000 	if (!ceph_is_valid_xattr(name))
1001 		return -EOPNOTSUPP;
1002 
1003 	vxattr = ceph_match_vxattr(inode, name);
1004 	if (vxattr && vxattr->readonly)
1005 		return -EOPNOTSUPP;
1006 
1007 	/* pass any unhandled ceph.* xattrs through to the MDS */
1008 	if (!strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
1009 		goto do_sync_unlocked;
1010 
1011 	err = -ENOMEM;
1012 	spin_lock(&ci->i_ceph_lock);
1013 retry:
1014 	issued = __ceph_caps_issued(ci, NULL);
1015 	dout("removexattr %p issued %s\n", inode, ceph_cap_string(issued));
1016 
1017 	if (!(issued & CEPH_CAP_XATTR_EXCL))
1018 		goto do_sync;
1019 	__build_xattrs(inode);
1020 
1021 	required_blob_size = __get_required_blob_size(ci, 0, 0);
1022 
1023 	if (!ci->i_xattrs.prealloc_blob ||
1024 	    required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
1025 		struct ceph_buffer *blob;
1026 
1027 		spin_unlock(&ci->i_ceph_lock);
1028 		dout(" preaallocating new blob size=%d\n", required_blob_size);
1029 		blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
1030 		if (!blob)
1031 			goto out;
1032 		spin_lock(&ci->i_ceph_lock);
1033 		if (ci->i_xattrs.prealloc_blob)
1034 			ceph_buffer_put(ci->i_xattrs.prealloc_blob);
1035 		ci->i_xattrs.prealloc_blob = blob;
1036 		goto retry;
1037 	}
1038 
1039 	err = __remove_xattr_by_name(ceph_inode(inode), name);
1040 
1041 	dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL);
1042 	ci->i_xattrs.dirty = true;
1043 	inode->i_ctime = CURRENT_TIME;
1044 	spin_unlock(&ci->i_ceph_lock);
1045 	if (dirty)
1046 		__mark_inode_dirty(inode, dirty);
1047 	return err;
1048 do_sync:
1049 	spin_unlock(&ci->i_ceph_lock);
1050 do_sync_unlocked:
1051 	err = ceph_send_removexattr(dentry, name);
1052 out:
1053 	return err;
1054 }
1055 
1056