ceph: fix xattr dangling pointer / double free
[linux-block.git] / fs / ceph / xattr.c
CommitLineData
355da1eb
SW
1#include "ceph_debug.h"
2#include "super.h"
3#include "decode.h"
4
5#include <linux/xattr.h>
5a0e3ad6 6#include <linux/slab.h>
355da1eb
SW
7
8static bool ceph_is_valid_xattr(const char *name)
9{
10 return !strncmp(name, XATTR_SECURITY_PREFIX,
11 XATTR_SECURITY_PREFIX_LEN) ||
12 !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
13 !strncmp(name, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN);
14}
15
16/*
17 * These define virtual xattrs exposing the recursive directory
18 * statistics and layout metadata.
19 */
20struct ceph_vxattr_cb {
21 bool readonly;
22 char *name;
23 size_t (*getxattr_cb)(struct ceph_inode_info *ci, char *val,
24 size_t size);
25};
26
27/* directories */
28
29static size_t ceph_vxattrcb_entries(struct ceph_inode_info *ci, char *val,
30 size_t size)
31{
32 return snprintf(val, size, "%lld", ci->i_files + ci->i_subdirs);
33}
34
35static size_t ceph_vxattrcb_files(struct ceph_inode_info *ci, char *val,
36 size_t size)
37{
38 return snprintf(val, size, "%lld", ci->i_files);
39}
40
41static size_t ceph_vxattrcb_subdirs(struct ceph_inode_info *ci, char *val,
42 size_t size)
43{
44 return snprintf(val, size, "%lld", ci->i_subdirs);
45}
46
47static size_t ceph_vxattrcb_rentries(struct ceph_inode_info *ci, char *val,
48 size_t size)
49{
50 return snprintf(val, size, "%lld", ci->i_rfiles + ci->i_rsubdirs);
51}
52
53static size_t ceph_vxattrcb_rfiles(struct ceph_inode_info *ci, char *val,
54 size_t size)
55{
56 return snprintf(val, size, "%lld", ci->i_rfiles);
57}
58
59static size_t ceph_vxattrcb_rsubdirs(struct ceph_inode_info *ci, char *val,
60 size_t size)
61{
62 return snprintf(val, size, "%lld", ci->i_rsubdirs);
63}
64
65static size_t ceph_vxattrcb_rbytes(struct ceph_inode_info *ci, char *val,
66 size_t size)
67{
68 return snprintf(val, size, "%lld", ci->i_rbytes);
69}
70
71static size_t ceph_vxattrcb_rctime(struct ceph_inode_info *ci, char *val,
72 size_t size)
73{
74 return snprintf(val, size, "%ld.%ld", (long)ci->i_rctime.tv_sec,
75 (long)ci->i_rctime.tv_nsec);
76}
77
78static struct ceph_vxattr_cb ceph_dir_vxattrs[] = {
79 { true, "user.ceph.dir.entries", ceph_vxattrcb_entries},
80 { true, "user.ceph.dir.files", ceph_vxattrcb_files},
81 { true, "user.ceph.dir.subdirs", ceph_vxattrcb_subdirs},
82 { true, "user.ceph.dir.rentries", ceph_vxattrcb_rentries},
83 { true, "user.ceph.dir.rfiles", ceph_vxattrcb_rfiles},
84 { true, "user.ceph.dir.rsubdirs", ceph_vxattrcb_rsubdirs},
85 { true, "user.ceph.dir.rbytes", ceph_vxattrcb_rbytes},
86 { true, "user.ceph.dir.rctime", ceph_vxattrcb_rctime},
87 { true, NULL, NULL }
88};
89
90/* files */
91
92static size_t ceph_vxattrcb_layout(struct ceph_inode_info *ci, char *val,
93 size_t size)
94{
b195befd
SW
95 int ret;
96
97 ret = snprintf(val, size,
355da1eb
SW
98 "chunk_bytes=%lld\nstripe_count=%lld\nobject_size=%lld\n",
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));
b195befd
SW
102 if (ceph_file_layout_pg_preferred(ci->i_layout))
103 ret += snprintf(val + ret, size, "preferred_osd=%lld\n",
104 (unsigned long long)ceph_file_layout_pg_preferred(
105 ci->i_layout));
106 return ret;
355da1eb
SW
107}
108
109static struct ceph_vxattr_cb ceph_file_vxattrs[] = {
110 { true, "user.ceph.layout", ceph_vxattrcb_layout},
111 { NULL, NULL }
112};
113
114static struct ceph_vxattr_cb *ceph_inode_vxattrs(struct inode *inode)
115{
116 if (S_ISDIR(inode->i_mode))
117 return ceph_dir_vxattrs;
118 else if (S_ISREG(inode->i_mode))
119 return ceph_file_vxattrs;
120 return NULL;
121}
122
123static struct ceph_vxattr_cb *ceph_match_vxattr(struct ceph_vxattr_cb *vxattr,
124 const char *name)
125{
126 do {
127 if (strcmp(vxattr->name, name) == 0)
128 return vxattr;
129 vxattr++;
130 } while (vxattr->name);
131 return NULL;
132}
133
134static int __set_xattr(struct ceph_inode_info *ci,
135 const char *name, int name_len,
136 const char *val, int val_len,
137 int dirty,
138 int should_free_name, int should_free_val,
139 struct ceph_inode_xattr **newxattr)
140{
141 struct rb_node **p;
142 struct rb_node *parent = NULL;
143 struct ceph_inode_xattr *xattr = NULL;
144 int c;
145 int new = 0;
146
147 p = &ci->i_xattrs.index.rb_node;
148 while (*p) {
149 parent = *p;
150 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
151 c = strncmp(name, xattr->name, min(name_len, xattr->name_len));
152 if (c < 0)
153 p = &(*p)->rb_left;
154 else if (c > 0)
155 p = &(*p)->rb_right;
156 else {
157 if (name_len == xattr->name_len)
158 break;
159 else if (name_len < xattr->name_len)
160 p = &(*p)->rb_left;
161 else
162 p = &(*p)->rb_right;
163 }
164 xattr = NULL;
165 }
166
167 if (!xattr) {
168 new = 1;
169 xattr = *newxattr;
170 xattr->name = name;
171 xattr->name_len = name_len;
172 xattr->should_free_name = should_free_name;
173
174 ci->i_xattrs.count++;
175 dout("__set_xattr count=%d\n", ci->i_xattrs.count);
176 } else {
177 kfree(*newxattr);
178 *newxattr = NULL;
179 if (xattr->should_free_val)
180 kfree((void *)xattr->val);
181
182 if (should_free_name) {
183 kfree((void *)name);
184 name = xattr->name;
185 }
186 ci->i_xattrs.names_size -= xattr->name_len;
187 ci->i_xattrs.vals_size -= xattr->val_len;
188 }
355da1eb
SW
189 ci->i_xattrs.names_size += name_len;
190 ci->i_xattrs.vals_size += val_len;
191 if (val)
192 xattr->val = val;
193 else
194 xattr->val = "";
195
196 xattr->val_len = val_len;
197 xattr->dirty = dirty;
198 xattr->should_free_val = (val && should_free_val);
199
200 if (new) {
201 rb_link_node(&xattr->node, parent, p);
202 rb_insert_color(&xattr->node, &ci->i_xattrs.index);
203 dout("__set_xattr_val p=%p\n", p);
204 }
205
206 dout("__set_xattr_val added %llx.%llx xattr %p %s=%.*s\n",
207 ceph_vinop(&ci->vfs_inode), xattr, name, val_len, val);
208
209 return 0;
210}
211
212static struct ceph_inode_xattr *__get_xattr(struct ceph_inode_info *ci,
213 const char *name)
214{
215 struct rb_node **p;
216 struct rb_node *parent = NULL;
217 struct ceph_inode_xattr *xattr = NULL;
218 int c;
219
220 p = &ci->i_xattrs.index.rb_node;
221 while (*p) {
222 parent = *p;
223 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
224 c = strncmp(name, xattr->name, xattr->name_len);
225 if (c < 0)
226 p = &(*p)->rb_left;
227 else if (c > 0)
228 p = &(*p)->rb_right;
229 else {
230 dout("__get_xattr %s: found %.*s\n", name,
231 xattr->val_len, xattr->val);
232 return xattr;
233 }
234 }
235
236 dout("__get_xattr %s: not found\n", name);
237
238 return NULL;
239}
240
241static void __free_xattr(struct ceph_inode_xattr *xattr)
242{
243 BUG_ON(!xattr);
244
245 if (xattr->should_free_name)
246 kfree((void *)xattr->name);
247 if (xattr->should_free_val)
248 kfree((void *)xattr->val);
249
250 kfree(xattr);
251}
252
253static int __remove_xattr(struct ceph_inode_info *ci,
254 struct ceph_inode_xattr *xattr)
255{
256 if (!xattr)
257 return -EOPNOTSUPP;
258
259 rb_erase(&xattr->node, &ci->i_xattrs.index);
260
261 if (xattr->should_free_name)
262 kfree((void *)xattr->name);
263 if (xattr->should_free_val)
264 kfree((void *)xattr->val);
265
266 ci->i_xattrs.names_size -= xattr->name_len;
267 ci->i_xattrs.vals_size -= xattr->val_len;
268 ci->i_xattrs.count--;
269 kfree(xattr);
270
271 return 0;
272}
273
274static int __remove_xattr_by_name(struct ceph_inode_info *ci,
275 const char *name)
276{
277 struct rb_node **p;
278 struct ceph_inode_xattr *xattr;
279 int err;
280
281 p = &ci->i_xattrs.index.rb_node;
282 xattr = __get_xattr(ci, name);
283 err = __remove_xattr(ci, xattr);
284 return err;
285}
286
287static char *__copy_xattr_names(struct ceph_inode_info *ci,
288 char *dest)
289{
290 struct rb_node *p;
291 struct ceph_inode_xattr *xattr = NULL;
292
293 p = rb_first(&ci->i_xattrs.index);
294 dout("__copy_xattr_names count=%d\n", ci->i_xattrs.count);
295
296 while (p) {
297 xattr = rb_entry(p, struct ceph_inode_xattr, node);
298 memcpy(dest, xattr->name, xattr->name_len);
299 dest[xattr->name_len] = '\0';
300
301 dout("dest=%s %p (%s) (%d/%d)\n", dest, xattr, xattr->name,
302 xattr->name_len, ci->i_xattrs.names_size);
303
304 dest += xattr->name_len + 1;
305 p = rb_next(p);
306 }
307
308 return dest;
309}
310
311void __ceph_destroy_xattrs(struct ceph_inode_info *ci)
312{
313 struct rb_node *p, *tmp;
314 struct ceph_inode_xattr *xattr = NULL;
315
316 p = rb_first(&ci->i_xattrs.index);
317
318 dout("__ceph_destroy_xattrs p=%p\n", p);
319
320 while (p) {
321 xattr = rb_entry(p, struct ceph_inode_xattr, node);
322 tmp = p;
323 p = rb_next(tmp);
324 dout("__ceph_destroy_xattrs next p=%p (%.*s)\n", p,
325 xattr->name_len, xattr->name);
326 rb_erase(tmp, &ci->i_xattrs.index);
327
328 __free_xattr(xattr);
329 }
330
331 ci->i_xattrs.names_size = 0;
332 ci->i_xattrs.vals_size = 0;
333 ci->i_xattrs.index_version = 0;
334 ci->i_xattrs.count = 0;
335 ci->i_xattrs.index = RB_ROOT;
336}
337
338static int __build_xattrs(struct inode *inode)
339{
340 u32 namelen;
341 u32 numattr = 0;
342 void *p, *end;
343 u32 len;
344 const char *name, *val;
345 struct ceph_inode_info *ci = ceph_inode(inode);
346 int xattr_version;
347 struct ceph_inode_xattr **xattrs = NULL;
63ff78b2 348 int err = 0;
355da1eb
SW
349 int i;
350
351 dout("__build_xattrs() len=%d\n",
352 ci->i_xattrs.blob ? (int)ci->i_xattrs.blob->vec.iov_len : 0);
353
354 if (ci->i_xattrs.index_version >= ci->i_xattrs.version)
355 return 0; /* already built */
356
357 __ceph_destroy_xattrs(ci);
358
359start:
360 /* updated internal xattr rb tree */
361 if (ci->i_xattrs.blob && ci->i_xattrs.blob->vec.iov_len > 4) {
362 p = ci->i_xattrs.blob->vec.iov_base;
363 end = p + ci->i_xattrs.blob->vec.iov_len;
364 ceph_decode_32_safe(&p, end, numattr, bad);
365 xattr_version = ci->i_xattrs.version;
366 spin_unlock(&inode->i_lock);
367
368 xattrs = kcalloc(numattr, sizeof(struct ceph_xattr *),
369 GFP_NOFS);
370 err = -ENOMEM;
371 if (!xattrs)
372 goto bad_lock;
373 memset(xattrs, 0, numattr*sizeof(struct ceph_xattr *));
374 for (i = 0; i < numattr; i++) {
375 xattrs[i] = kmalloc(sizeof(struct ceph_inode_xattr),
376 GFP_NOFS);
377 if (!xattrs[i])
378 goto bad_lock;
379 }
380
381 spin_lock(&inode->i_lock);
382 if (ci->i_xattrs.version != xattr_version) {
383 /* lost a race, retry */
384 for (i = 0; i < numattr; i++)
385 kfree(xattrs[i]);
386 kfree(xattrs);
387 goto start;
388 }
389 err = -EIO;
390 while (numattr--) {
391 ceph_decode_32_safe(&p, end, len, bad);
392 namelen = len;
393 name = p;
394 p += len;
395 ceph_decode_32_safe(&p, end, len, bad);
396 val = p;
397 p += len;
398
399 err = __set_xattr(ci, name, namelen, val, len,
400 0, 0, 0, &xattrs[numattr]);
401
402 if (err < 0)
403 goto bad;
404 }
405 kfree(xattrs);
406 }
407 ci->i_xattrs.index_version = ci->i_xattrs.version;
408 ci->i_xattrs.dirty = false;
409
410 return err;
411bad_lock:
412 spin_lock(&inode->i_lock);
413bad:
414 if (xattrs) {
415 for (i = 0; i < numattr; i++)
416 kfree(xattrs[i]);
417 kfree(xattrs);
418 }
419 ci->i_xattrs.names_size = 0;
420 return err;
421}
422
423static int __get_required_blob_size(struct ceph_inode_info *ci, int name_size,
424 int val_size)
425{
426 /*
427 * 4 bytes for the length, and additional 4 bytes per each xattr name,
428 * 4 bytes per each value
429 */
430 int size = 4 + ci->i_xattrs.count*(4 + 4) +
431 ci->i_xattrs.names_size +
432 ci->i_xattrs.vals_size;
433 dout("__get_required_blob_size c=%d names.size=%d vals.size=%d\n",
434 ci->i_xattrs.count, ci->i_xattrs.names_size,
435 ci->i_xattrs.vals_size);
436
437 if (name_size)
438 size += 4 + 4 + name_size + val_size;
439
440 return size;
441}
442
443/*
444 * If there are dirty xattrs, reencode xattrs into the prealloc_blob
445 * and swap into place.
446 */
447void __ceph_build_xattrs_blob(struct ceph_inode_info *ci)
448{
449 struct rb_node *p;
450 struct ceph_inode_xattr *xattr = NULL;
451 void *dest;
452
453 dout("__build_xattrs_blob %p\n", &ci->vfs_inode);
454 if (ci->i_xattrs.dirty) {
455 int need = __get_required_blob_size(ci, 0, 0);
456
457 BUG_ON(need > ci->i_xattrs.prealloc_blob->alloc_len);
458
459 p = rb_first(&ci->i_xattrs.index);
460 dest = ci->i_xattrs.prealloc_blob->vec.iov_base;
461
462 ceph_encode_32(&dest, ci->i_xattrs.count);
463 while (p) {
464 xattr = rb_entry(p, struct ceph_inode_xattr, node);
465
466 ceph_encode_32(&dest, xattr->name_len);
467 memcpy(dest, xattr->name, xattr->name_len);
468 dest += xattr->name_len;
469 ceph_encode_32(&dest, xattr->val_len);
470 memcpy(dest, xattr->val, xattr->val_len);
471 dest += xattr->val_len;
472
473 p = rb_next(p);
474 }
475
476 /* adjust buffer len; it may be larger than we need */
477 ci->i_xattrs.prealloc_blob->vec.iov_len =
478 dest - ci->i_xattrs.prealloc_blob->vec.iov_base;
479
b6c1d5b8
SW
480 if (ci->i_xattrs.blob)
481 ceph_buffer_put(ci->i_xattrs.blob);
355da1eb
SW
482 ci->i_xattrs.blob = ci->i_xattrs.prealloc_blob;
483 ci->i_xattrs.prealloc_blob = NULL;
484 ci->i_xattrs.dirty = false;
485 }
486}
487
488ssize_t ceph_getxattr(struct dentry *dentry, const char *name, void *value,
489 size_t size)
490{
491 struct inode *inode = dentry->d_inode;
492 struct ceph_inode_info *ci = ceph_inode(inode);
493 struct ceph_vxattr_cb *vxattrs = ceph_inode_vxattrs(inode);
494 int err;
495 struct ceph_inode_xattr *xattr;
496 struct ceph_vxattr_cb *vxattr = NULL;
497
498 if (!ceph_is_valid_xattr(name))
499 return -ENODATA;
500
501 /* let's see if a virtual xattr was requested */
502 if (vxattrs)
503 vxattr = ceph_match_vxattr(vxattrs, name);
504
505 spin_lock(&inode->i_lock);
506 dout("getxattr %p ver=%lld index_ver=%lld\n", inode,
507 ci->i_xattrs.version, ci->i_xattrs.index_version);
508
509 if (__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1) &&
510 (ci->i_xattrs.index_version >= ci->i_xattrs.version)) {
511 goto get_xattr;
512 } else {
513 spin_unlock(&inode->i_lock);
514 /* get xattrs from mds (if we don't already have them) */
515 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR);
516 if (err)
517 return err;
518 }
519
520 spin_lock(&inode->i_lock);
521
522 if (vxattr && vxattr->readonly) {
523 err = vxattr->getxattr_cb(ci, value, size);
524 goto out;
525 }
526
527 err = __build_xattrs(inode);
528 if (err < 0)
529 goto out;
530
531get_xattr:
532 err = -ENODATA; /* == ENOATTR */
533 xattr = __get_xattr(ci, name);
534 if (!xattr) {
535 if (vxattr)
536 err = vxattr->getxattr_cb(ci, value, size);
537 goto out;
538 }
539
540 err = -ERANGE;
541 if (size && size < xattr->val_len)
542 goto out;
543
544 err = xattr->val_len;
545 if (size == 0)
546 goto out;
547
548 memcpy(value, xattr->val, xattr->val_len);
549
550out:
551 spin_unlock(&inode->i_lock);
552 return err;
553}
554
555ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)
556{
557 struct inode *inode = dentry->d_inode;
558 struct ceph_inode_info *ci = ceph_inode(inode);
559 struct ceph_vxattr_cb *vxattrs = ceph_inode_vxattrs(inode);
560 u32 vir_namelen = 0;
561 u32 namelen;
562 int err;
563 u32 len;
564 int i;
565
566 spin_lock(&inode->i_lock);
567 dout("listxattr %p ver=%lld index_ver=%lld\n", inode,
568 ci->i_xattrs.version, ci->i_xattrs.index_version);
569
570 if (__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1) &&
571 (ci->i_xattrs.index_version > ci->i_xattrs.version)) {
572 goto list_xattr;
573 } else {
574 spin_unlock(&inode->i_lock);
575 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR);
576 if (err)
577 return err;
578 }
579
580 spin_lock(&inode->i_lock);
581
582 err = __build_xattrs(inode);
583 if (err < 0)
584 goto out;
585
586list_xattr:
587 vir_namelen = 0;
588 /* include virtual dir xattrs */
589 if (vxattrs)
590 for (i = 0; vxattrs[i].name; i++)
591 vir_namelen += strlen(vxattrs[i].name) + 1;
592 /* adding 1 byte per each variable due to the null termination */
593 namelen = vir_namelen + ci->i_xattrs.names_size + ci->i_xattrs.count;
594 err = -ERANGE;
595 if (size && namelen > size)
596 goto out;
597
598 err = namelen;
599 if (size == 0)
600 goto out;
601
602 names = __copy_xattr_names(ci, names);
603
604 /* virtual xattr names, too */
605 if (vxattrs)
606 for (i = 0; vxattrs[i].name; i++) {
607 len = sprintf(names, "%s", vxattrs[i].name);
608 names += len + 1;
609 }
610
611out:
612 spin_unlock(&inode->i_lock);
613 return err;
614}
615
616static int ceph_sync_setxattr(struct dentry *dentry, const char *name,
617 const char *value, size_t size, int flags)
618{
640ef79d 619 struct ceph_client *client = ceph_sb_to_client(dentry->d_sb);
355da1eb
SW
620 struct inode *inode = dentry->d_inode;
621 struct ceph_inode_info *ci = ceph_inode(inode);
622 struct inode *parent_inode = dentry->d_parent->d_inode;
623 struct ceph_mds_request *req;
624 struct ceph_mds_client *mdsc = &client->mdsc;
625 int err;
626 int i, nr_pages;
627 struct page **pages = NULL;
628 void *kaddr;
629
630 /* copy value into some pages */
631 nr_pages = calc_pages_for(0, size);
632 if (nr_pages) {
633 pages = kmalloc(sizeof(pages[0])*nr_pages, GFP_NOFS);
634 if (!pages)
635 return -ENOMEM;
636 err = -ENOMEM;
637 for (i = 0; i < nr_pages; i++) {
31459fe4 638 pages[i] = __page_cache_alloc(GFP_NOFS);
355da1eb
SW
639 if (!pages[i]) {
640 nr_pages = i;
641 goto out;
642 }
643 kaddr = kmap(pages[i]);
644 memcpy(kaddr, value + i*PAGE_CACHE_SIZE,
645 min(PAGE_CACHE_SIZE, size-i*PAGE_CACHE_SIZE));
646 }
647 }
648
649 dout("setxattr value=%.*s\n", (int)size, value);
650
651 /* do request */
652 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETXATTR,
653 USE_AUTH_MDS);
60d87733
JL
654 if (IS_ERR(req)) {
655 err = PTR_ERR(req);
656 goto out;
657 }
355da1eb
SW
658 req->r_inode = igrab(inode);
659 req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
660 req->r_num_caps = 1;
661 req->r_args.setxattr.flags = cpu_to_le32(flags);
662 req->r_path2 = kstrdup(name, GFP_NOFS);
663
664 req->r_pages = pages;
665 req->r_num_pages = nr_pages;
666 req->r_data_len = size;
667
668 dout("xattr.ver (before): %lld\n", ci->i_xattrs.version);
669 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
670 ceph_mdsc_put_request(req);
671 dout("xattr.ver (after): %lld\n", ci->i_xattrs.version);
672
673out:
674 if (pages) {
675 for (i = 0; i < nr_pages; i++)
676 __free_page(pages[i]);
677 kfree(pages);
678 }
679 return err;
680}
681
682int ceph_setxattr(struct dentry *dentry, const char *name,
683 const void *value, size_t size, int flags)
684{
685 struct inode *inode = dentry->d_inode;
686 struct ceph_inode_info *ci = ceph_inode(inode);
687 struct ceph_vxattr_cb *vxattrs = ceph_inode_vxattrs(inode);
688 int err;
689 int name_len = strlen(name);
690 int val_len = size;
691 char *newname = NULL;
692 char *newval = NULL;
693 struct ceph_inode_xattr *xattr = NULL;
694 int issued;
695 int required_blob_size;
696
697 if (ceph_snap(inode) != CEPH_NOSNAP)
698 return -EROFS;
699
700 if (!ceph_is_valid_xattr(name))
701 return -EOPNOTSUPP;
702
703 if (vxattrs) {
704 struct ceph_vxattr_cb *vxattr =
705 ceph_match_vxattr(vxattrs, name);
706 if (vxattr && vxattr->readonly)
707 return -EOPNOTSUPP;
708 }
709
710 /* preallocate memory for xattr name, value, index node */
711 err = -ENOMEM;
712 newname = kmalloc(name_len + 1, GFP_NOFS);
713 if (!newname)
714 goto out;
715 memcpy(newname, name, name_len + 1);
716
717 if (val_len) {
718 newval = kmalloc(val_len + 1, GFP_NOFS);
719 if (!newval)
720 goto out;
721 memcpy(newval, value, val_len);
722 newval[val_len] = '\0';
723 }
724
725 xattr = kmalloc(sizeof(struct ceph_inode_xattr), GFP_NOFS);
726 if (!xattr)
727 goto out;
728
729 spin_lock(&inode->i_lock);
730retry:
731 issued = __ceph_caps_issued(ci, NULL);
732 if (!(issued & CEPH_CAP_XATTR_EXCL))
733 goto do_sync;
734 __build_xattrs(inode);
735
736 required_blob_size = __get_required_blob_size(ci, name_len, val_len);
737
738 if (!ci->i_xattrs.prealloc_blob ||
739 required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
740 struct ceph_buffer *blob = NULL;
741
742 spin_unlock(&inode->i_lock);
743 dout(" preaallocating new blob size=%d\n", required_blob_size);
b6c1d5b8 744 blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
355da1eb
SW
745 if (!blob)
746 goto out;
747 spin_lock(&inode->i_lock);
b6c1d5b8
SW
748 if (ci->i_xattrs.prealloc_blob)
749 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
355da1eb
SW
750 ci->i_xattrs.prealloc_blob = blob;
751 goto retry;
752 }
753
754 dout("setxattr %p issued %s\n", inode, ceph_cap_string(issued));
755 err = __set_xattr(ci, newname, name_len, newval,
756 val_len, 1, 1, 1, &xattr);
757 __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL);
758 ci->i_xattrs.dirty = true;
759 inode->i_ctime = CURRENT_TIME;
760 spin_unlock(&inode->i_lock);
761
762 return err;
763
764do_sync:
765 spin_unlock(&inode->i_lock);
766 err = ceph_sync_setxattr(dentry, name, value, size, flags);
767out:
768 kfree(newname);
769 kfree(newval);
770 kfree(xattr);
771 return err;
772}
773
774static int ceph_send_removexattr(struct dentry *dentry, const char *name)
775{
640ef79d 776 struct ceph_client *client = ceph_sb_to_client(dentry->d_sb);
355da1eb
SW
777 struct ceph_mds_client *mdsc = &client->mdsc;
778 struct inode *inode = dentry->d_inode;
779 struct inode *parent_inode = dentry->d_parent->d_inode;
780 struct ceph_mds_request *req;
781 int err;
782
783 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RMXATTR,
784 USE_AUTH_MDS);
785 if (IS_ERR(req))
786 return PTR_ERR(req);
787 req->r_inode = igrab(inode);
788 req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
789 req->r_num_caps = 1;
790 req->r_path2 = kstrdup(name, GFP_NOFS);
791
792 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
793 ceph_mdsc_put_request(req);
794 return err;
795}
796
797int ceph_removexattr(struct dentry *dentry, const char *name)
798{
799 struct inode *inode = dentry->d_inode;
800 struct ceph_inode_info *ci = ceph_inode(inode);
801 struct ceph_vxattr_cb *vxattrs = ceph_inode_vxattrs(inode);
802 int issued;
803 int err;
804
805 if (ceph_snap(inode) != CEPH_NOSNAP)
806 return -EROFS;
807
808 if (!ceph_is_valid_xattr(name))
809 return -EOPNOTSUPP;
810
811 if (vxattrs) {
812 struct ceph_vxattr_cb *vxattr =
813 ceph_match_vxattr(vxattrs, name);
814 if (vxattr && vxattr->readonly)
815 return -EOPNOTSUPP;
816 }
817
818 spin_lock(&inode->i_lock);
819 __build_xattrs(inode);
820 issued = __ceph_caps_issued(ci, NULL);
821 dout("removexattr %p issued %s\n", inode, ceph_cap_string(issued));
822
823 if (!(issued & CEPH_CAP_XATTR_EXCL))
824 goto do_sync;
825
826 err = __remove_xattr_by_name(ceph_inode(inode), name);
827 __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL);
828 ci->i_xattrs.dirty = true;
829 inode->i_ctime = CURRENT_TIME;
830
831 spin_unlock(&inode->i_lock);
832
833 return err;
834do_sync:
835 spin_unlock(&inode->i_lock);
836 err = ceph_send_removexattr(dentry, name);
837 return err;
838}
839