Loading...
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
11static bool ceph_is_valid_xattr(const char *name)
12{
13 return !strncmp(name, "ceph.", 5) ||
14 !strncmp(name, XATTR_SECURITY_PREFIX,
15 XATTR_SECURITY_PREFIX_LEN) ||
16 !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
17 !strncmp(name, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN);
18}
19
20/*
21 * These define virtual xattrs exposing the recursive directory
22 * statistics and layout metadata.
23 */
24struct ceph_vxattr_cb {
25 bool readonly;
26 char *name;
27 size_t (*getxattr_cb)(struct ceph_inode_info *ci, char *val,
28 size_t size);
29};
30
31/* directories */
32
33static size_t ceph_vxattrcb_entries(struct ceph_inode_info *ci, char *val,
34 size_t size)
35{
36 return snprintf(val, size, "%lld", ci->i_files + ci->i_subdirs);
37}
38
39static size_t ceph_vxattrcb_files(struct ceph_inode_info *ci, char *val,
40 size_t size)
41{
42 return snprintf(val, size, "%lld", ci->i_files);
43}
44
45static size_t ceph_vxattrcb_subdirs(struct ceph_inode_info *ci, char *val,
46 size_t size)
47{
48 return snprintf(val, size, "%lld", ci->i_subdirs);
49}
50
51static size_t ceph_vxattrcb_rentries(struct ceph_inode_info *ci, char *val,
52 size_t size)
53{
54 return snprintf(val, size, "%lld", ci->i_rfiles + ci->i_rsubdirs);
55}
56
57static size_t ceph_vxattrcb_rfiles(struct ceph_inode_info *ci, char *val,
58 size_t size)
59{
60 return snprintf(val, size, "%lld", ci->i_rfiles);
61}
62
63static size_t ceph_vxattrcb_rsubdirs(struct ceph_inode_info *ci, char *val,
64 size_t size)
65{
66 return snprintf(val, size, "%lld", ci->i_rsubdirs);
67}
68
69static size_t ceph_vxattrcb_rbytes(struct ceph_inode_info *ci, char *val,
70 size_t size)
71{
72 return snprintf(val, size, "%lld", ci->i_rbytes);
73}
74
75static size_t ceph_vxattrcb_rctime(struct ceph_inode_info *ci, char *val,
76 size_t size)
77{
78 return snprintf(val, size, "%ld.%ld", (long)ci->i_rctime.tv_sec,
79 (long)ci->i_rctime.tv_nsec);
80}
81
82static struct ceph_vxattr_cb ceph_dir_vxattrs[] = {
83 { true, "ceph.dir.entries", ceph_vxattrcb_entries},
84 { true, "ceph.dir.files", ceph_vxattrcb_files},
85 { true, "ceph.dir.subdirs", ceph_vxattrcb_subdirs},
86 { true, "ceph.dir.rentries", ceph_vxattrcb_rentries},
87 { true, "ceph.dir.rfiles", ceph_vxattrcb_rfiles},
88 { true, "ceph.dir.rsubdirs", ceph_vxattrcb_rsubdirs},
89 { true, "ceph.dir.rbytes", ceph_vxattrcb_rbytes},
90 { true, "ceph.dir.rctime", ceph_vxattrcb_rctime},
91 { true, NULL, NULL }
92};
93
94/* files */
95
96static size_t ceph_vxattrcb_layout(struct ceph_inode_info *ci, char *val,
97 size_t size)
98{
99 int ret;
100
101 ret = snprintf(val, size,
102 "chunk_bytes=%lld\nstripe_count=%lld\nobject_size=%lld\n",
103 (unsigned long long)ceph_file_layout_su(ci->i_layout),
104 (unsigned long long)ceph_file_layout_stripe_count(ci->i_layout),
105 (unsigned long long)ceph_file_layout_object_size(ci->i_layout));
106 if (ceph_file_layout_pg_preferred(ci->i_layout))
107 ret += snprintf(val + ret, size, "preferred_osd=%lld\n",
108 (unsigned long long)ceph_file_layout_pg_preferred(
109 ci->i_layout));
110 return ret;
111}
112
113static struct ceph_vxattr_cb ceph_file_vxattrs[] = {
114 { true, "ceph.layout", ceph_vxattrcb_layout},
115 { NULL, NULL }
116};
117
118static struct ceph_vxattr_cb *ceph_inode_vxattrs(struct inode *inode)
119{
120 if (S_ISDIR(inode->i_mode))
121 return ceph_dir_vxattrs;
122 else if (S_ISREG(inode->i_mode))
123 return ceph_file_vxattrs;
124 return NULL;
125}
126
127static struct ceph_vxattr_cb *ceph_match_vxattr(struct ceph_vxattr_cb *vxattr,
128 const char *name)
129{
130 do {
131 if (strcmp(vxattr->name, name) == 0)
132 return vxattr;
133 vxattr++;
134 } while (vxattr->name);
135 return NULL;
136}
137
138static int __set_xattr(struct ceph_inode_info *ci,
139 const char *name, int name_len,
140 const char *val, int val_len,
141 int dirty,
142 int should_free_name, int should_free_val,
143 struct ceph_inode_xattr **newxattr)
144{
145 struct rb_node **p;
146 struct rb_node *parent = NULL;
147 struct ceph_inode_xattr *xattr = NULL;
148 int c;
149 int new = 0;
150
151 p = &ci->i_xattrs.index.rb_node;
152 while (*p) {
153 parent = *p;
154 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
155 c = strncmp(name, xattr->name, min(name_len, xattr->name_len));
156 if (c < 0)
157 p = &(*p)->rb_left;
158 else if (c > 0)
159 p = &(*p)->rb_right;
160 else {
161 if (name_len == xattr->name_len)
162 break;
163 else if (name_len < xattr->name_len)
164 p = &(*p)->rb_left;
165 else
166 p = &(*p)->rb_right;
167 }
168 xattr = NULL;
169 }
170
171 if (!xattr) {
172 new = 1;
173 xattr = *newxattr;
174 xattr->name = name;
175 xattr->name_len = name_len;
176 xattr->should_free_name = should_free_name;
177
178 ci->i_xattrs.count++;
179 dout("__set_xattr count=%d\n", ci->i_xattrs.count);
180 } else {
181 kfree(*newxattr);
182 *newxattr = NULL;
183 if (xattr->should_free_val)
184 kfree((void *)xattr->val);
185
186 if (should_free_name) {
187 kfree((void *)name);
188 name = xattr->name;
189 }
190 ci->i_xattrs.names_size -= xattr->name_len;
191 ci->i_xattrs.vals_size -= xattr->val_len;
192 }
193 ci->i_xattrs.names_size += name_len;
194 ci->i_xattrs.vals_size += val_len;
195 if (val)
196 xattr->val = val;
197 else
198 xattr->val = "";
199
200 xattr->val_len = val_len;
201 xattr->dirty = dirty;
202 xattr->should_free_val = (val && should_free_val);
203
204 if (new) {
205 rb_link_node(&xattr->node, parent, p);
206 rb_insert_color(&xattr->node, &ci->i_xattrs.index);
207 dout("__set_xattr_val p=%p\n", p);
208 }
209
210 dout("__set_xattr_val added %llx.%llx xattr %p %s=%.*s\n",
211 ceph_vinop(&ci->vfs_inode), xattr, name, val_len, val);
212
213 return 0;
214}
215
216static struct ceph_inode_xattr *__get_xattr(struct ceph_inode_info *ci,
217 const char *name)
218{
219 struct rb_node **p;
220 struct rb_node *parent = NULL;
221 struct ceph_inode_xattr *xattr = NULL;
222 int name_len = strlen(name);
223 int c;
224
225 p = &ci->i_xattrs.index.rb_node;
226 while (*p) {
227 parent = *p;
228 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
229 c = strncmp(name, xattr->name, xattr->name_len);
230 if (c == 0 && name_len > xattr->name_len)
231 c = 1;
232 if (c < 0)
233 p = &(*p)->rb_left;
234 else if (c > 0)
235 p = &(*p)->rb_right;
236 else {
237 dout("__get_xattr %s: found %.*s\n", name,
238 xattr->val_len, xattr->val);
239 return xattr;
240 }
241 }
242
243 dout("__get_xattr %s: not found\n", name);
244
245 return NULL;
246}
247
248static void __free_xattr(struct ceph_inode_xattr *xattr)
249{
250 BUG_ON(!xattr);
251
252 if (xattr->should_free_name)
253 kfree((void *)xattr->name);
254 if (xattr->should_free_val)
255 kfree((void *)xattr->val);
256
257 kfree(xattr);
258}
259
260static int __remove_xattr(struct ceph_inode_info *ci,
261 struct ceph_inode_xattr *xattr)
262{
263 if (!xattr)
264 return -EOPNOTSUPP;
265
266 rb_erase(&xattr->node, &ci->i_xattrs.index);
267
268 if (xattr->should_free_name)
269 kfree((void *)xattr->name);
270 if (xattr->should_free_val)
271 kfree((void *)xattr->val);
272
273 ci->i_xattrs.names_size -= xattr->name_len;
274 ci->i_xattrs.vals_size -= xattr->val_len;
275 ci->i_xattrs.count--;
276 kfree(xattr);
277
278 return 0;
279}
280
281static int __remove_xattr_by_name(struct ceph_inode_info *ci,
282 const char *name)
283{
284 struct rb_node **p;
285 struct ceph_inode_xattr *xattr;
286 int err;
287
288 p = &ci->i_xattrs.index.rb_node;
289 xattr = __get_xattr(ci, name);
290 err = __remove_xattr(ci, xattr);
291 return err;
292}
293
294static char *__copy_xattr_names(struct ceph_inode_info *ci,
295 char *dest)
296{
297 struct rb_node *p;
298 struct ceph_inode_xattr *xattr = NULL;
299
300 p = rb_first(&ci->i_xattrs.index);
301 dout("__copy_xattr_names count=%d\n", ci->i_xattrs.count);
302
303 while (p) {
304 xattr = rb_entry(p, struct ceph_inode_xattr, node);
305 memcpy(dest, xattr->name, xattr->name_len);
306 dest[xattr->name_len] = '\0';
307
308 dout("dest=%s %p (%s) (%d/%d)\n", dest, xattr, xattr->name,
309 xattr->name_len, ci->i_xattrs.names_size);
310
311 dest += xattr->name_len + 1;
312 p = rb_next(p);
313 }
314
315 return dest;
316}
317
318void __ceph_destroy_xattrs(struct ceph_inode_info *ci)
319{
320 struct rb_node *p, *tmp;
321 struct ceph_inode_xattr *xattr = NULL;
322
323 p = rb_first(&ci->i_xattrs.index);
324
325 dout("__ceph_destroy_xattrs p=%p\n", p);
326
327 while (p) {
328 xattr = rb_entry(p, struct ceph_inode_xattr, node);
329 tmp = p;
330 p = rb_next(tmp);
331 dout("__ceph_destroy_xattrs next p=%p (%.*s)\n", p,
332 xattr->name_len, xattr->name);
333 rb_erase(tmp, &ci->i_xattrs.index);
334
335 __free_xattr(xattr);
336 }
337
338 ci->i_xattrs.names_size = 0;
339 ci->i_xattrs.vals_size = 0;
340 ci->i_xattrs.index_version = 0;
341 ci->i_xattrs.count = 0;
342 ci->i_xattrs.index = RB_ROOT;
343}
344
345static int __build_xattrs(struct inode *inode)
346 __releases(inode->i_lock)
347 __acquires(inode->i_lock)
348{
349 u32 namelen;
350 u32 numattr = 0;
351 void *p, *end;
352 u32 len;
353 const char *name, *val;
354 struct ceph_inode_info *ci = ceph_inode(inode);
355 int xattr_version;
356 struct ceph_inode_xattr **xattrs = NULL;
357 int err = 0;
358 int i;
359
360 dout("__build_xattrs() len=%d\n",
361 ci->i_xattrs.blob ? (int)ci->i_xattrs.blob->vec.iov_len : 0);
362
363 if (ci->i_xattrs.index_version >= ci->i_xattrs.version)
364 return 0; /* already built */
365
366 __ceph_destroy_xattrs(ci);
367
368start:
369 /* updated internal xattr rb tree */
370 if (ci->i_xattrs.blob && ci->i_xattrs.blob->vec.iov_len > 4) {
371 p = ci->i_xattrs.blob->vec.iov_base;
372 end = p + ci->i_xattrs.blob->vec.iov_len;
373 ceph_decode_32_safe(&p, end, numattr, bad);
374 xattr_version = ci->i_xattrs.version;
375 spin_unlock(&inode->i_lock);
376
377 xattrs = kcalloc(numattr, sizeof(struct ceph_xattr *),
378 GFP_NOFS);
379 err = -ENOMEM;
380 if (!xattrs)
381 goto bad_lock;
382 memset(xattrs, 0, numattr*sizeof(struct ceph_xattr *));
383 for (i = 0; i < numattr; i++) {
384 xattrs[i] = kmalloc(sizeof(struct ceph_inode_xattr),
385 GFP_NOFS);
386 if (!xattrs[i])
387 goto bad_lock;
388 }
389
390 spin_lock(&inode->i_lock);
391 if (ci->i_xattrs.version != xattr_version) {
392 /* lost a race, retry */
393 for (i = 0; i < numattr; i++)
394 kfree(xattrs[i]);
395 kfree(xattrs);
396 goto start;
397 }
398 err = -EIO;
399 while (numattr--) {
400 ceph_decode_32_safe(&p, end, len, bad);
401 namelen = len;
402 name = p;
403 p += len;
404 ceph_decode_32_safe(&p, end, len, bad);
405 val = p;
406 p += len;
407
408 err = __set_xattr(ci, name, namelen, val, len,
409 0, 0, 0, &xattrs[numattr]);
410
411 if (err < 0)
412 goto bad;
413 }
414 kfree(xattrs);
415 }
416 ci->i_xattrs.index_version = ci->i_xattrs.version;
417 ci->i_xattrs.dirty = false;
418
419 return err;
420bad_lock:
421 spin_lock(&inode->i_lock);
422bad:
423 if (xattrs) {
424 for (i = 0; i < numattr; i++)
425 kfree(xattrs[i]);
426 kfree(xattrs);
427 }
428 ci->i_xattrs.names_size = 0;
429 return err;
430}
431
432static int __get_required_blob_size(struct ceph_inode_info *ci, int name_size,
433 int val_size)
434{
435 /*
436 * 4 bytes for the length, and additional 4 bytes per each xattr name,
437 * 4 bytes per each value
438 */
439 int size = 4 + ci->i_xattrs.count*(4 + 4) +
440 ci->i_xattrs.names_size +
441 ci->i_xattrs.vals_size;
442 dout("__get_required_blob_size c=%d names.size=%d vals.size=%d\n",
443 ci->i_xattrs.count, ci->i_xattrs.names_size,
444 ci->i_xattrs.vals_size);
445
446 if (name_size)
447 size += 4 + 4 + name_size + val_size;
448
449 return size;
450}
451
452/*
453 * If there are dirty xattrs, reencode xattrs into the prealloc_blob
454 * and swap into place.
455 */
456void __ceph_build_xattrs_blob(struct ceph_inode_info *ci)
457{
458 struct rb_node *p;
459 struct ceph_inode_xattr *xattr = NULL;
460 void *dest;
461
462 dout("__build_xattrs_blob %p\n", &ci->vfs_inode);
463 if (ci->i_xattrs.dirty) {
464 int need = __get_required_blob_size(ci, 0, 0);
465
466 BUG_ON(need > ci->i_xattrs.prealloc_blob->alloc_len);
467
468 p = rb_first(&ci->i_xattrs.index);
469 dest = ci->i_xattrs.prealloc_blob->vec.iov_base;
470
471 ceph_encode_32(&dest, ci->i_xattrs.count);
472 while (p) {
473 xattr = rb_entry(p, struct ceph_inode_xattr, node);
474
475 ceph_encode_32(&dest, xattr->name_len);
476 memcpy(dest, xattr->name, xattr->name_len);
477 dest += xattr->name_len;
478 ceph_encode_32(&dest, xattr->val_len);
479 memcpy(dest, xattr->val, xattr->val_len);
480 dest += xattr->val_len;
481
482 p = rb_next(p);
483 }
484
485 /* adjust buffer len; it may be larger than we need */
486 ci->i_xattrs.prealloc_blob->vec.iov_len =
487 dest - ci->i_xattrs.prealloc_blob->vec.iov_base;
488
489 if (ci->i_xattrs.blob)
490 ceph_buffer_put(ci->i_xattrs.blob);
491 ci->i_xattrs.blob = ci->i_xattrs.prealloc_blob;
492 ci->i_xattrs.prealloc_blob = NULL;
493 ci->i_xattrs.dirty = false;
494 ci->i_xattrs.version++;
495 }
496}
497
498ssize_t ceph_getxattr(struct dentry *dentry, const char *name, void *value,
499 size_t size)
500{
501 struct inode *inode = dentry->d_inode;
502 struct ceph_inode_info *ci = ceph_inode(inode);
503 struct ceph_vxattr_cb *vxattrs = ceph_inode_vxattrs(inode);
504 int err;
505 struct ceph_inode_xattr *xattr;
506 struct ceph_vxattr_cb *vxattr = NULL;
507
508 if (!ceph_is_valid_xattr(name))
509 return -ENODATA;
510
511 /* let's see if a virtual xattr was requested */
512 if (vxattrs)
513 vxattr = ceph_match_vxattr(vxattrs, name);
514
515 spin_lock(&inode->i_lock);
516 dout("getxattr %p ver=%lld index_ver=%lld\n", inode,
517 ci->i_xattrs.version, ci->i_xattrs.index_version);
518
519 if (__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1) &&
520 (ci->i_xattrs.index_version >= ci->i_xattrs.version)) {
521 goto get_xattr;
522 } else {
523 spin_unlock(&inode->i_lock);
524 /* get xattrs from mds (if we don't already have them) */
525 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR);
526 if (err)
527 return err;
528 }
529
530 spin_lock(&inode->i_lock);
531
532 if (vxattr && vxattr->readonly) {
533 err = vxattr->getxattr_cb(ci, value, size);
534 goto out;
535 }
536
537 err = __build_xattrs(inode);
538 if (err < 0)
539 goto out;
540
541get_xattr:
542 err = -ENODATA; /* == ENOATTR */
543 xattr = __get_xattr(ci, name);
544 if (!xattr) {
545 if (vxattr)
546 err = vxattr->getxattr_cb(ci, value, size);
547 goto out;
548 }
549
550 err = -ERANGE;
551 if (size && size < xattr->val_len)
552 goto out;
553
554 err = xattr->val_len;
555 if (size == 0)
556 goto out;
557
558 memcpy(value, xattr->val, xattr->val_len);
559
560out:
561 spin_unlock(&inode->i_lock);
562 return err;
563}
564
565ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)
566{
567 struct inode *inode = dentry->d_inode;
568 struct ceph_inode_info *ci = ceph_inode(inode);
569 struct ceph_vxattr_cb *vxattrs = ceph_inode_vxattrs(inode);
570 u32 vir_namelen = 0;
571 u32 namelen;
572 int err;
573 u32 len;
574 int i;
575
576 spin_lock(&inode->i_lock);
577 dout("listxattr %p ver=%lld index_ver=%lld\n", inode,
578 ci->i_xattrs.version, ci->i_xattrs.index_version);
579
580 if (__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1) &&
581 (ci->i_xattrs.index_version >= ci->i_xattrs.version)) {
582 goto list_xattr;
583 } else {
584 spin_unlock(&inode->i_lock);
585 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR);
586 if (err)
587 return err;
588 }
589
590 spin_lock(&inode->i_lock);
591
592 err = __build_xattrs(inode);
593 if (err < 0)
594 goto out;
595
596list_xattr:
597 vir_namelen = 0;
598 /* include virtual dir xattrs */
599 if (vxattrs)
600 for (i = 0; vxattrs[i].name; i++)
601 vir_namelen += strlen(vxattrs[i].name) + 1;
602 /* adding 1 byte per each variable due to the null termination */
603 namelen = vir_namelen + ci->i_xattrs.names_size + ci->i_xattrs.count;
604 err = -ERANGE;
605 if (size && namelen > size)
606 goto out;
607
608 err = namelen;
609 if (size == 0)
610 goto out;
611
612 names = __copy_xattr_names(ci, names);
613
614 /* virtual xattr names, too */
615 if (vxattrs)
616 for (i = 0; vxattrs[i].name; i++) {
617 len = sprintf(names, "%s", vxattrs[i].name);
618 names += len + 1;
619 }
620
621out:
622 spin_unlock(&inode->i_lock);
623 return err;
624}
625
626static int ceph_sync_setxattr(struct dentry *dentry, const char *name,
627 const char *value, size_t size, int flags)
628{
629 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
630 struct inode *inode = dentry->d_inode;
631 struct ceph_inode_info *ci = ceph_inode(inode);
632 struct inode *parent_inode;
633 struct ceph_mds_request *req;
634 struct ceph_mds_client *mdsc = fsc->mdsc;
635 int err;
636 int i, nr_pages;
637 struct page **pages = NULL;
638 void *kaddr;
639
640 /* copy value into some pages */
641 nr_pages = calc_pages_for(0, size);
642 if (nr_pages) {
643 pages = kmalloc(sizeof(pages[0])*nr_pages, GFP_NOFS);
644 if (!pages)
645 return -ENOMEM;
646 err = -ENOMEM;
647 for (i = 0; i < nr_pages; i++) {
648 pages[i] = __page_cache_alloc(GFP_NOFS);
649 if (!pages[i]) {
650 nr_pages = i;
651 goto out;
652 }
653 kaddr = kmap(pages[i]);
654 memcpy(kaddr, value + i*PAGE_CACHE_SIZE,
655 min(PAGE_CACHE_SIZE, size-i*PAGE_CACHE_SIZE));
656 }
657 }
658
659 dout("setxattr value=%.*s\n", (int)size, value);
660
661 /* do request */
662 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETXATTR,
663 USE_AUTH_MDS);
664 if (IS_ERR(req)) {
665 err = PTR_ERR(req);
666 goto out;
667 }
668 req->r_inode = inode;
669 ihold(inode);
670 req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
671 req->r_num_caps = 1;
672 req->r_args.setxattr.flags = cpu_to_le32(flags);
673 req->r_path2 = kstrdup(name, GFP_NOFS);
674
675 req->r_pages = pages;
676 req->r_num_pages = nr_pages;
677 req->r_data_len = size;
678
679 dout("xattr.ver (before): %lld\n", ci->i_xattrs.version);
680 parent_inode = ceph_get_dentry_parent_inode(dentry);
681 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
682 iput(parent_inode);
683 ceph_mdsc_put_request(req);
684 dout("xattr.ver (after): %lld\n", ci->i_xattrs.version);
685
686out:
687 if (pages) {
688 for (i = 0; i < nr_pages; i++)
689 __free_page(pages[i]);
690 kfree(pages);
691 }
692 return err;
693}
694
695int ceph_setxattr(struct dentry *dentry, const char *name,
696 const void *value, size_t size, int flags)
697{
698 struct inode *inode = dentry->d_inode;
699 struct ceph_inode_info *ci = ceph_inode(inode);
700 struct ceph_vxattr_cb *vxattrs = ceph_inode_vxattrs(inode);
701 int err;
702 int name_len = strlen(name);
703 int val_len = size;
704 char *newname = NULL;
705 char *newval = NULL;
706 struct ceph_inode_xattr *xattr = NULL;
707 int issued;
708 int required_blob_size;
709 int dirty;
710
711 if (ceph_snap(inode) != CEPH_NOSNAP)
712 return -EROFS;
713
714 if (!ceph_is_valid_xattr(name))
715 return -EOPNOTSUPP;
716
717 if (vxattrs) {
718 struct ceph_vxattr_cb *vxattr =
719 ceph_match_vxattr(vxattrs, name);
720 if (vxattr && vxattr->readonly)
721 return -EOPNOTSUPP;
722 }
723
724 /* preallocate memory for xattr name, value, index node */
725 err = -ENOMEM;
726 newname = kmemdup(name, name_len + 1, GFP_NOFS);
727 if (!newname)
728 goto out;
729
730 if (val_len) {
731 newval = kmalloc(val_len + 1, GFP_NOFS);
732 if (!newval)
733 goto out;
734 memcpy(newval, value, val_len);
735 newval[val_len] = '\0';
736 }
737
738 xattr = kmalloc(sizeof(struct ceph_inode_xattr), GFP_NOFS);
739 if (!xattr)
740 goto out;
741
742 spin_lock(&inode->i_lock);
743retry:
744 issued = __ceph_caps_issued(ci, NULL);
745 if (!(issued & CEPH_CAP_XATTR_EXCL))
746 goto do_sync;
747 __build_xattrs(inode);
748
749 required_blob_size = __get_required_blob_size(ci, name_len, val_len);
750
751 if (!ci->i_xattrs.prealloc_blob ||
752 required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
753 struct ceph_buffer *blob = NULL;
754
755 spin_unlock(&inode->i_lock);
756 dout(" preaallocating new blob size=%d\n", required_blob_size);
757 blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
758 if (!blob)
759 goto out;
760 spin_lock(&inode->i_lock);
761 if (ci->i_xattrs.prealloc_blob)
762 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
763 ci->i_xattrs.prealloc_blob = blob;
764 goto retry;
765 }
766
767 dout("setxattr %p issued %s\n", inode, ceph_cap_string(issued));
768 err = __set_xattr(ci, newname, name_len, newval,
769 val_len, 1, 1, 1, &xattr);
770 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL);
771 ci->i_xattrs.dirty = true;
772 inode->i_ctime = CURRENT_TIME;
773 spin_unlock(&inode->i_lock);
774 if (dirty)
775 __mark_inode_dirty(inode, dirty);
776 return err;
777
778do_sync:
779 spin_unlock(&inode->i_lock);
780 err = ceph_sync_setxattr(dentry, name, value, size, flags);
781out:
782 kfree(newname);
783 kfree(newval);
784 kfree(xattr);
785 return err;
786}
787
788static int ceph_send_removexattr(struct dentry *dentry, const char *name)
789{
790 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
791 struct ceph_mds_client *mdsc = fsc->mdsc;
792 struct inode *inode = dentry->d_inode;
793 struct inode *parent_inode;
794 struct ceph_mds_request *req;
795 int err;
796
797 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RMXATTR,
798 USE_AUTH_MDS);
799 if (IS_ERR(req))
800 return PTR_ERR(req);
801 req->r_inode = inode;
802 ihold(inode);
803 req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
804 req->r_num_caps = 1;
805 req->r_path2 = kstrdup(name, GFP_NOFS);
806
807 parent_inode = ceph_get_dentry_parent_inode(dentry);
808 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
809 iput(parent_inode);
810 ceph_mdsc_put_request(req);
811 return err;
812}
813
814int ceph_removexattr(struct dentry *dentry, const char *name)
815{
816 struct inode *inode = dentry->d_inode;
817 struct ceph_inode_info *ci = ceph_inode(inode);
818 struct ceph_vxattr_cb *vxattrs = ceph_inode_vxattrs(inode);
819 int issued;
820 int err;
821 int dirty;
822
823 if (ceph_snap(inode) != CEPH_NOSNAP)
824 return -EROFS;
825
826 if (!ceph_is_valid_xattr(name))
827 return -EOPNOTSUPP;
828
829 if (vxattrs) {
830 struct ceph_vxattr_cb *vxattr =
831 ceph_match_vxattr(vxattrs, name);
832 if (vxattr && vxattr->readonly)
833 return -EOPNOTSUPP;
834 }
835
836 spin_lock(&inode->i_lock);
837 __build_xattrs(inode);
838 issued = __ceph_caps_issued(ci, NULL);
839 dout("removexattr %p issued %s\n", inode, ceph_cap_string(issued));
840
841 if (!(issued & CEPH_CAP_XATTR_EXCL))
842 goto do_sync;
843
844 err = __remove_xattr_by_name(ceph_inode(inode), name);
845 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL);
846 ci->i_xattrs.dirty = true;
847 inode->i_ctime = CURRENT_TIME;
848
849 spin_unlock(&inode->i_lock);
850 if (dirty)
851 __mark_inode_dirty(inode, dirty);
852 return err;
853do_sync:
854 spin_unlock(&inode->i_lock);
855 err = ceph_send_removexattr(dentry, name);
856 return err;
857}
858
1// SPDX-License-Identifier: GPL-2.0
2#include <linux/ceph/ceph_debug.h>
3#include <linux/ceph/pagelist.h>
4
5#include "super.h"
6#include "mds_client.h"
7
8#include <linux/ceph/decode.h>
9
10#include <linux/xattr.h>
11#include <linux/posix_acl_xattr.h>
12#include <linux/slab.h>
13
14#define XATTR_CEPH_PREFIX "ceph."
15#define XATTR_CEPH_PREFIX_LEN (sizeof (XATTR_CEPH_PREFIX) - 1)
16
17static int __remove_xattr(struct ceph_inode_info *ci,
18 struct ceph_inode_xattr *xattr);
19
20static const struct xattr_handler ceph_other_xattr_handler;
21
22/*
23 * List of handlers for synthetic system.* attributes. Other
24 * attributes are handled directly.
25 */
26const struct xattr_handler *ceph_xattr_handlers[] = {
27#ifdef CONFIG_CEPH_FS_POSIX_ACL
28 &posix_acl_access_xattr_handler,
29 &posix_acl_default_xattr_handler,
30#endif
31 &ceph_other_xattr_handler,
32 NULL,
33};
34
35static bool ceph_is_valid_xattr(const char *name)
36{
37 return !strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN) ||
38 !strncmp(name, XATTR_SECURITY_PREFIX,
39 XATTR_SECURITY_PREFIX_LEN) ||
40 !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
41 !strncmp(name, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN);
42}
43
44/*
45 * These define virtual xattrs exposing the recursive directory
46 * statistics and layout metadata.
47 */
48struct ceph_vxattr {
49 char *name;
50 size_t name_size; /* strlen(name) + 1 (for '\0') */
51 size_t (*getxattr_cb)(struct ceph_inode_info *ci, char *val,
52 size_t size);
53 bool readonly, hidden;
54 bool (*exists_cb)(struct ceph_inode_info *ci);
55};
56
57/* layouts */
58
59static bool ceph_vxattrcb_layout_exists(struct ceph_inode_info *ci)
60{
61 struct ceph_file_layout *fl = &ci->i_layout;
62 return (fl->stripe_unit > 0 || fl->stripe_count > 0 ||
63 fl->object_size > 0 || fl->pool_id >= 0 ||
64 rcu_dereference_raw(fl->pool_ns) != NULL);
65}
66
67static size_t ceph_vxattrcb_layout(struct ceph_inode_info *ci, char *val,
68 size_t size)
69{
70 struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
71 struct ceph_osd_client *osdc = &fsc->client->osdc;
72 struct ceph_string *pool_ns;
73 s64 pool = ci->i_layout.pool_id;
74 const char *pool_name;
75 const char *ns_field = " pool_namespace=";
76 char buf[128];
77 size_t len, total_len = 0;
78 int ret;
79
80 pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
81
82 dout("ceph_vxattrcb_layout %p\n", &ci->vfs_inode);
83 down_read(&osdc->lock);
84 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
85 if (pool_name) {
86 len = snprintf(buf, sizeof(buf),
87 "stripe_unit=%u stripe_count=%u object_size=%u pool=",
88 ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
89 ci->i_layout.object_size);
90 total_len = len + strlen(pool_name);
91 } else {
92 len = snprintf(buf, sizeof(buf),
93 "stripe_unit=%u stripe_count=%u object_size=%u pool=%lld",
94 ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
95 ci->i_layout.object_size, (unsigned long long)pool);
96 total_len = len;
97 }
98
99 if (pool_ns)
100 total_len += strlen(ns_field) + pool_ns->len;
101
102 if (!size) {
103 ret = total_len;
104 } else if (total_len > size) {
105 ret = -ERANGE;
106 } else {
107 memcpy(val, buf, len);
108 ret = len;
109 if (pool_name) {
110 len = strlen(pool_name);
111 memcpy(val + ret, pool_name, len);
112 ret += len;
113 }
114 if (pool_ns) {
115 len = strlen(ns_field);
116 memcpy(val + ret, ns_field, len);
117 ret += len;
118 memcpy(val + ret, pool_ns->str, pool_ns->len);
119 ret += pool_ns->len;
120 }
121 }
122 up_read(&osdc->lock);
123 ceph_put_string(pool_ns);
124 return ret;
125}
126
127static size_t ceph_vxattrcb_layout_stripe_unit(struct ceph_inode_info *ci,
128 char *val, size_t size)
129{
130 return snprintf(val, size, "%u", ci->i_layout.stripe_unit);
131}
132
133static size_t ceph_vxattrcb_layout_stripe_count(struct ceph_inode_info *ci,
134 char *val, size_t size)
135{
136 return snprintf(val, size, "%u", ci->i_layout.stripe_count);
137}
138
139static size_t ceph_vxattrcb_layout_object_size(struct ceph_inode_info *ci,
140 char *val, size_t size)
141{
142 return snprintf(val, size, "%u", ci->i_layout.object_size);
143}
144
145static size_t ceph_vxattrcb_layout_pool(struct ceph_inode_info *ci,
146 char *val, size_t size)
147{
148 int ret;
149 struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
150 struct ceph_osd_client *osdc = &fsc->client->osdc;
151 s64 pool = ci->i_layout.pool_id;
152 const char *pool_name;
153
154 down_read(&osdc->lock);
155 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
156 if (pool_name)
157 ret = snprintf(val, size, "%s", pool_name);
158 else
159 ret = snprintf(val, size, "%lld", (unsigned long long)pool);
160 up_read(&osdc->lock);
161 return ret;
162}
163
164static size_t ceph_vxattrcb_layout_pool_namespace(struct ceph_inode_info *ci,
165 char *val, size_t size)
166{
167 int ret = 0;
168 struct ceph_string *ns = ceph_try_get_string(ci->i_layout.pool_ns);
169 if (ns) {
170 ret = snprintf(val, size, "%.*s", (int)ns->len, ns->str);
171 ceph_put_string(ns);
172 }
173 return ret;
174}
175
176/* directories */
177
178static size_t ceph_vxattrcb_dir_entries(struct ceph_inode_info *ci, char *val,
179 size_t size)
180{
181 return snprintf(val, size, "%lld", ci->i_files + ci->i_subdirs);
182}
183
184static size_t ceph_vxattrcb_dir_files(struct ceph_inode_info *ci, char *val,
185 size_t size)
186{
187 return snprintf(val, size, "%lld", ci->i_files);
188}
189
190static size_t ceph_vxattrcb_dir_subdirs(struct ceph_inode_info *ci, char *val,
191 size_t size)
192{
193 return snprintf(val, size, "%lld", ci->i_subdirs);
194}
195
196static size_t ceph_vxattrcb_dir_rentries(struct ceph_inode_info *ci, char *val,
197 size_t size)
198{
199 return snprintf(val, size, "%lld", ci->i_rfiles + ci->i_rsubdirs);
200}
201
202static size_t ceph_vxattrcb_dir_rfiles(struct ceph_inode_info *ci, char *val,
203 size_t size)
204{
205 return snprintf(val, size, "%lld", ci->i_rfiles);
206}
207
208static size_t ceph_vxattrcb_dir_rsubdirs(struct ceph_inode_info *ci, char *val,
209 size_t size)
210{
211 return snprintf(val, size, "%lld", ci->i_rsubdirs);
212}
213
214static size_t ceph_vxattrcb_dir_rbytes(struct ceph_inode_info *ci, char *val,
215 size_t size)
216{
217 return snprintf(val, size, "%lld", ci->i_rbytes);
218}
219
220static size_t ceph_vxattrcb_dir_rctime(struct ceph_inode_info *ci, char *val,
221 size_t size)
222{
223 return snprintf(val, size, "%ld.09%ld", (long)ci->i_rctime.tv_sec,
224 (long)ci->i_rctime.tv_nsec);
225}
226
227/* quotas */
228
229static bool ceph_vxattrcb_quota_exists(struct ceph_inode_info *ci)
230{
231 bool ret = false;
232 spin_lock(&ci->i_ceph_lock);
233 if ((ci->i_max_files || ci->i_max_bytes) &&
234 ci->i_vino.snap == CEPH_NOSNAP &&
235 ci->i_snap_realm &&
236 ci->i_snap_realm->ino == ci->i_vino.ino)
237 ret = true;
238 spin_unlock(&ci->i_ceph_lock);
239 return ret;
240}
241
242static size_t ceph_vxattrcb_quota(struct ceph_inode_info *ci, char *val,
243 size_t size)
244{
245 return snprintf(val, size, "max_bytes=%llu max_files=%llu",
246 ci->i_max_bytes, ci->i_max_files);
247}
248
249static size_t ceph_vxattrcb_quota_max_bytes(struct ceph_inode_info *ci,
250 char *val, size_t size)
251{
252 return snprintf(val, size, "%llu", ci->i_max_bytes);
253}
254
255static size_t ceph_vxattrcb_quota_max_files(struct ceph_inode_info *ci,
256 char *val, size_t size)
257{
258 return snprintf(val, size, "%llu", ci->i_max_files);
259}
260
261#define CEPH_XATTR_NAME(_type, _name) XATTR_CEPH_PREFIX #_type "." #_name
262#define CEPH_XATTR_NAME2(_type, _name, _name2) \
263 XATTR_CEPH_PREFIX #_type "." #_name "." #_name2
264
265#define XATTR_NAME_CEPH(_type, _name) \
266 { \
267 .name = CEPH_XATTR_NAME(_type, _name), \
268 .name_size = sizeof (CEPH_XATTR_NAME(_type, _name)), \
269 .getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name, \
270 .readonly = true, \
271 .hidden = false, \
272 .exists_cb = NULL, \
273 }
274#define XATTR_LAYOUT_FIELD(_type, _name, _field) \
275 { \
276 .name = CEPH_XATTR_NAME2(_type, _name, _field), \
277 .name_size = sizeof (CEPH_XATTR_NAME2(_type, _name, _field)), \
278 .getxattr_cb = ceph_vxattrcb_ ## _name ## _ ## _field, \
279 .readonly = false, \
280 .hidden = true, \
281 .exists_cb = ceph_vxattrcb_layout_exists, \
282 }
283#define XATTR_QUOTA_FIELD(_type, _name) \
284 { \
285 .name = CEPH_XATTR_NAME(_type, _name), \
286 .name_size = sizeof(CEPH_XATTR_NAME(_type, _name)), \
287 .getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name, \
288 .readonly = false, \
289 .hidden = true, \
290 .exists_cb = ceph_vxattrcb_quota_exists, \
291 }
292
293static struct ceph_vxattr ceph_dir_vxattrs[] = {
294 {
295 .name = "ceph.dir.layout",
296 .name_size = sizeof("ceph.dir.layout"),
297 .getxattr_cb = ceph_vxattrcb_layout,
298 .readonly = false,
299 .hidden = true,
300 .exists_cb = ceph_vxattrcb_layout_exists,
301 },
302 XATTR_LAYOUT_FIELD(dir, layout, stripe_unit),
303 XATTR_LAYOUT_FIELD(dir, layout, stripe_count),
304 XATTR_LAYOUT_FIELD(dir, layout, object_size),
305 XATTR_LAYOUT_FIELD(dir, layout, pool),
306 XATTR_LAYOUT_FIELD(dir, layout, pool_namespace),
307 XATTR_NAME_CEPH(dir, entries),
308 XATTR_NAME_CEPH(dir, files),
309 XATTR_NAME_CEPH(dir, subdirs),
310 XATTR_NAME_CEPH(dir, rentries),
311 XATTR_NAME_CEPH(dir, rfiles),
312 XATTR_NAME_CEPH(dir, rsubdirs),
313 XATTR_NAME_CEPH(dir, rbytes),
314 XATTR_NAME_CEPH(dir, rctime),
315 {
316 .name = "ceph.quota",
317 .name_size = sizeof("ceph.quota"),
318 .getxattr_cb = ceph_vxattrcb_quota,
319 .readonly = false,
320 .hidden = true,
321 .exists_cb = ceph_vxattrcb_quota_exists,
322 },
323 XATTR_QUOTA_FIELD(quota, max_bytes),
324 XATTR_QUOTA_FIELD(quota, max_files),
325 { .name = NULL, 0 } /* Required table terminator */
326};
327static size_t ceph_dir_vxattrs_name_size; /* total size of all names */
328
329/* files */
330
331static struct ceph_vxattr ceph_file_vxattrs[] = {
332 {
333 .name = "ceph.file.layout",
334 .name_size = sizeof("ceph.file.layout"),
335 .getxattr_cb = ceph_vxattrcb_layout,
336 .readonly = false,
337 .hidden = true,
338 .exists_cb = ceph_vxattrcb_layout_exists,
339 },
340 XATTR_LAYOUT_FIELD(file, layout, stripe_unit),
341 XATTR_LAYOUT_FIELD(file, layout, stripe_count),
342 XATTR_LAYOUT_FIELD(file, layout, object_size),
343 XATTR_LAYOUT_FIELD(file, layout, pool),
344 XATTR_LAYOUT_FIELD(file, layout, pool_namespace),
345 { .name = NULL, 0 } /* Required table terminator */
346};
347static size_t ceph_file_vxattrs_name_size; /* total size of all names */
348
349static struct ceph_vxattr *ceph_inode_vxattrs(struct inode *inode)
350{
351 if (S_ISDIR(inode->i_mode))
352 return ceph_dir_vxattrs;
353 else if (S_ISREG(inode->i_mode))
354 return ceph_file_vxattrs;
355 return NULL;
356}
357
358static size_t ceph_vxattrs_name_size(struct ceph_vxattr *vxattrs)
359{
360 if (vxattrs == ceph_dir_vxattrs)
361 return ceph_dir_vxattrs_name_size;
362 if (vxattrs == ceph_file_vxattrs)
363 return ceph_file_vxattrs_name_size;
364 BUG_ON(vxattrs);
365 return 0;
366}
367
368/*
369 * Compute the aggregate size (including terminating '\0') of all
370 * virtual extended attribute names in the given vxattr table.
371 */
372static size_t __init vxattrs_name_size(struct ceph_vxattr *vxattrs)
373{
374 struct ceph_vxattr *vxattr;
375 size_t size = 0;
376
377 for (vxattr = vxattrs; vxattr->name; vxattr++)
378 if (!vxattr->hidden)
379 size += vxattr->name_size;
380
381 return size;
382}
383
384/* Routines called at initialization and exit time */
385
386void __init ceph_xattr_init(void)
387{
388 ceph_dir_vxattrs_name_size = vxattrs_name_size(ceph_dir_vxattrs);
389 ceph_file_vxattrs_name_size = vxattrs_name_size(ceph_file_vxattrs);
390}
391
392void ceph_xattr_exit(void)
393{
394 ceph_dir_vxattrs_name_size = 0;
395 ceph_file_vxattrs_name_size = 0;
396}
397
398static struct ceph_vxattr *ceph_match_vxattr(struct inode *inode,
399 const char *name)
400{
401 struct ceph_vxattr *vxattr = ceph_inode_vxattrs(inode);
402
403 if (vxattr) {
404 while (vxattr->name) {
405 if (!strcmp(vxattr->name, name))
406 return vxattr;
407 vxattr++;
408 }
409 }
410
411 return NULL;
412}
413
414static int __set_xattr(struct ceph_inode_info *ci,
415 const char *name, int name_len,
416 const char *val, int val_len,
417 int flags, int update_xattr,
418 struct ceph_inode_xattr **newxattr)
419{
420 struct rb_node **p;
421 struct rb_node *parent = NULL;
422 struct ceph_inode_xattr *xattr = NULL;
423 int c;
424 int new = 0;
425
426 p = &ci->i_xattrs.index.rb_node;
427 while (*p) {
428 parent = *p;
429 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
430 c = strncmp(name, xattr->name, min(name_len, xattr->name_len));
431 if (c < 0)
432 p = &(*p)->rb_left;
433 else if (c > 0)
434 p = &(*p)->rb_right;
435 else {
436 if (name_len == xattr->name_len)
437 break;
438 else if (name_len < xattr->name_len)
439 p = &(*p)->rb_left;
440 else
441 p = &(*p)->rb_right;
442 }
443 xattr = NULL;
444 }
445
446 if (update_xattr) {
447 int err = 0;
448
449 if (xattr && (flags & XATTR_CREATE))
450 err = -EEXIST;
451 else if (!xattr && (flags & XATTR_REPLACE))
452 err = -ENODATA;
453 if (err) {
454 kfree(name);
455 kfree(val);
456 kfree(*newxattr);
457 return err;
458 }
459 if (update_xattr < 0) {
460 if (xattr)
461 __remove_xattr(ci, xattr);
462 kfree(name);
463 kfree(*newxattr);
464 return 0;
465 }
466 }
467
468 if (!xattr) {
469 new = 1;
470 xattr = *newxattr;
471 xattr->name = name;
472 xattr->name_len = name_len;
473 xattr->should_free_name = update_xattr;
474
475 ci->i_xattrs.count++;
476 dout("__set_xattr count=%d\n", ci->i_xattrs.count);
477 } else {
478 kfree(*newxattr);
479 *newxattr = NULL;
480 if (xattr->should_free_val)
481 kfree((void *)xattr->val);
482
483 if (update_xattr) {
484 kfree((void *)name);
485 name = xattr->name;
486 }
487 ci->i_xattrs.names_size -= xattr->name_len;
488 ci->i_xattrs.vals_size -= xattr->val_len;
489 }
490 ci->i_xattrs.names_size += name_len;
491 ci->i_xattrs.vals_size += val_len;
492 if (val)
493 xattr->val = val;
494 else
495 xattr->val = "";
496
497 xattr->val_len = val_len;
498 xattr->dirty = update_xattr;
499 xattr->should_free_val = (val && update_xattr);
500
501 if (new) {
502 rb_link_node(&xattr->node, parent, p);
503 rb_insert_color(&xattr->node, &ci->i_xattrs.index);
504 dout("__set_xattr_val p=%p\n", p);
505 }
506
507 dout("__set_xattr_val added %llx.%llx xattr %p %s=%.*s\n",
508 ceph_vinop(&ci->vfs_inode), xattr, name, val_len, val);
509
510 return 0;
511}
512
513static struct ceph_inode_xattr *__get_xattr(struct ceph_inode_info *ci,
514 const char *name)
515{
516 struct rb_node **p;
517 struct rb_node *parent = NULL;
518 struct ceph_inode_xattr *xattr = NULL;
519 int name_len = strlen(name);
520 int c;
521
522 p = &ci->i_xattrs.index.rb_node;
523 while (*p) {
524 parent = *p;
525 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
526 c = strncmp(name, xattr->name, xattr->name_len);
527 if (c == 0 && name_len > xattr->name_len)
528 c = 1;
529 if (c < 0)
530 p = &(*p)->rb_left;
531 else if (c > 0)
532 p = &(*p)->rb_right;
533 else {
534 dout("__get_xattr %s: found %.*s\n", name,
535 xattr->val_len, xattr->val);
536 return xattr;
537 }
538 }
539
540 dout("__get_xattr %s: not found\n", name);
541
542 return NULL;
543}
544
545static void __free_xattr(struct ceph_inode_xattr *xattr)
546{
547 BUG_ON(!xattr);
548
549 if (xattr->should_free_name)
550 kfree((void *)xattr->name);
551 if (xattr->should_free_val)
552 kfree((void *)xattr->val);
553
554 kfree(xattr);
555}
556
557static int __remove_xattr(struct ceph_inode_info *ci,
558 struct ceph_inode_xattr *xattr)
559{
560 if (!xattr)
561 return -ENODATA;
562
563 rb_erase(&xattr->node, &ci->i_xattrs.index);
564
565 if (xattr->should_free_name)
566 kfree((void *)xattr->name);
567 if (xattr->should_free_val)
568 kfree((void *)xattr->val);
569
570 ci->i_xattrs.names_size -= xattr->name_len;
571 ci->i_xattrs.vals_size -= xattr->val_len;
572 ci->i_xattrs.count--;
573 kfree(xattr);
574
575 return 0;
576}
577
578static char *__copy_xattr_names(struct ceph_inode_info *ci,
579 char *dest)
580{
581 struct rb_node *p;
582 struct ceph_inode_xattr *xattr = NULL;
583
584 p = rb_first(&ci->i_xattrs.index);
585 dout("__copy_xattr_names count=%d\n", ci->i_xattrs.count);
586
587 while (p) {
588 xattr = rb_entry(p, struct ceph_inode_xattr, node);
589 memcpy(dest, xattr->name, xattr->name_len);
590 dest[xattr->name_len] = '\0';
591
592 dout("dest=%s %p (%s) (%d/%d)\n", dest, xattr, xattr->name,
593 xattr->name_len, ci->i_xattrs.names_size);
594
595 dest += xattr->name_len + 1;
596 p = rb_next(p);
597 }
598
599 return dest;
600}
601
602void __ceph_destroy_xattrs(struct ceph_inode_info *ci)
603{
604 struct rb_node *p, *tmp;
605 struct ceph_inode_xattr *xattr = NULL;
606
607 p = rb_first(&ci->i_xattrs.index);
608
609 dout("__ceph_destroy_xattrs p=%p\n", p);
610
611 while (p) {
612 xattr = rb_entry(p, struct ceph_inode_xattr, node);
613 tmp = p;
614 p = rb_next(tmp);
615 dout("__ceph_destroy_xattrs next p=%p (%.*s)\n", p,
616 xattr->name_len, xattr->name);
617 rb_erase(tmp, &ci->i_xattrs.index);
618
619 __free_xattr(xattr);
620 }
621
622 ci->i_xattrs.names_size = 0;
623 ci->i_xattrs.vals_size = 0;
624 ci->i_xattrs.index_version = 0;
625 ci->i_xattrs.count = 0;
626 ci->i_xattrs.index = RB_ROOT;
627}
628
629static int __build_xattrs(struct inode *inode)
630 __releases(ci->i_ceph_lock)
631 __acquires(ci->i_ceph_lock)
632{
633 u32 namelen;
634 u32 numattr = 0;
635 void *p, *end;
636 u32 len;
637 const char *name, *val;
638 struct ceph_inode_info *ci = ceph_inode(inode);
639 int xattr_version;
640 struct ceph_inode_xattr **xattrs = NULL;
641 int err = 0;
642 int i;
643
644 dout("__build_xattrs() len=%d\n",
645 ci->i_xattrs.blob ? (int)ci->i_xattrs.blob->vec.iov_len : 0);
646
647 if (ci->i_xattrs.index_version >= ci->i_xattrs.version)
648 return 0; /* already built */
649
650 __ceph_destroy_xattrs(ci);
651
652start:
653 /* updated internal xattr rb tree */
654 if (ci->i_xattrs.blob && ci->i_xattrs.blob->vec.iov_len > 4) {
655 p = ci->i_xattrs.blob->vec.iov_base;
656 end = p + ci->i_xattrs.blob->vec.iov_len;
657 ceph_decode_32_safe(&p, end, numattr, bad);
658 xattr_version = ci->i_xattrs.version;
659 spin_unlock(&ci->i_ceph_lock);
660
661 xattrs = kcalloc(numattr, sizeof(struct ceph_inode_xattr *),
662 GFP_NOFS);
663 err = -ENOMEM;
664 if (!xattrs)
665 goto bad_lock;
666
667 for (i = 0; i < numattr; i++) {
668 xattrs[i] = kmalloc(sizeof(struct ceph_inode_xattr),
669 GFP_NOFS);
670 if (!xattrs[i])
671 goto bad_lock;
672 }
673
674 spin_lock(&ci->i_ceph_lock);
675 if (ci->i_xattrs.version != xattr_version) {
676 /* lost a race, retry */
677 for (i = 0; i < numattr; i++)
678 kfree(xattrs[i]);
679 kfree(xattrs);
680 xattrs = NULL;
681 goto start;
682 }
683 err = -EIO;
684 while (numattr--) {
685 ceph_decode_32_safe(&p, end, len, bad);
686 namelen = len;
687 name = p;
688 p += len;
689 ceph_decode_32_safe(&p, end, len, bad);
690 val = p;
691 p += len;
692
693 err = __set_xattr(ci, name, namelen, val, len,
694 0, 0, &xattrs[numattr]);
695
696 if (err < 0)
697 goto bad;
698 }
699 kfree(xattrs);
700 }
701 ci->i_xattrs.index_version = ci->i_xattrs.version;
702 ci->i_xattrs.dirty = false;
703
704 return err;
705bad_lock:
706 spin_lock(&ci->i_ceph_lock);
707bad:
708 if (xattrs) {
709 for (i = 0; i < numattr; i++)
710 kfree(xattrs[i]);
711 kfree(xattrs);
712 }
713 ci->i_xattrs.names_size = 0;
714 return err;
715}
716
717static int __get_required_blob_size(struct ceph_inode_info *ci, int name_size,
718 int val_size)
719{
720 /*
721 * 4 bytes for the length, and additional 4 bytes per each xattr name,
722 * 4 bytes per each value
723 */
724 int size = 4 + ci->i_xattrs.count*(4 + 4) +
725 ci->i_xattrs.names_size +
726 ci->i_xattrs.vals_size;
727 dout("__get_required_blob_size c=%d names.size=%d vals.size=%d\n",
728 ci->i_xattrs.count, ci->i_xattrs.names_size,
729 ci->i_xattrs.vals_size);
730
731 if (name_size)
732 size += 4 + 4 + name_size + val_size;
733
734 return size;
735}
736
737/*
738 * If there are dirty xattrs, reencode xattrs into the prealloc_blob
739 * and swap into place.
740 */
741void __ceph_build_xattrs_blob(struct ceph_inode_info *ci)
742{
743 struct rb_node *p;
744 struct ceph_inode_xattr *xattr = NULL;
745 void *dest;
746
747 dout("__build_xattrs_blob %p\n", &ci->vfs_inode);
748 if (ci->i_xattrs.dirty) {
749 int need = __get_required_blob_size(ci, 0, 0);
750
751 BUG_ON(need > ci->i_xattrs.prealloc_blob->alloc_len);
752
753 p = rb_first(&ci->i_xattrs.index);
754 dest = ci->i_xattrs.prealloc_blob->vec.iov_base;
755
756 ceph_encode_32(&dest, ci->i_xattrs.count);
757 while (p) {
758 xattr = rb_entry(p, struct ceph_inode_xattr, node);
759
760 ceph_encode_32(&dest, xattr->name_len);
761 memcpy(dest, xattr->name, xattr->name_len);
762 dest += xattr->name_len;
763 ceph_encode_32(&dest, xattr->val_len);
764 memcpy(dest, xattr->val, xattr->val_len);
765 dest += xattr->val_len;
766
767 p = rb_next(p);
768 }
769
770 /* adjust buffer len; it may be larger than we need */
771 ci->i_xattrs.prealloc_blob->vec.iov_len =
772 dest - ci->i_xattrs.prealloc_blob->vec.iov_base;
773
774 if (ci->i_xattrs.blob)
775 ceph_buffer_put(ci->i_xattrs.blob);
776 ci->i_xattrs.blob = ci->i_xattrs.prealloc_blob;
777 ci->i_xattrs.prealloc_blob = NULL;
778 ci->i_xattrs.dirty = false;
779 ci->i_xattrs.version++;
780 }
781}
782
783static inline int __get_request_mask(struct inode *in) {
784 struct ceph_mds_request *req = current->journal_info;
785 int mask = 0;
786 if (req && req->r_target_inode == in) {
787 if (req->r_op == CEPH_MDS_OP_LOOKUP ||
788 req->r_op == CEPH_MDS_OP_LOOKUPINO ||
789 req->r_op == CEPH_MDS_OP_LOOKUPPARENT ||
790 req->r_op == CEPH_MDS_OP_GETATTR) {
791 mask = le32_to_cpu(req->r_args.getattr.mask);
792 } else if (req->r_op == CEPH_MDS_OP_OPEN ||
793 req->r_op == CEPH_MDS_OP_CREATE) {
794 mask = le32_to_cpu(req->r_args.open.mask);
795 }
796 }
797 return mask;
798}
799
800ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,
801 size_t size)
802{
803 struct ceph_inode_info *ci = ceph_inode(inode);
804 struct ceph_inode_xattr *xattr;
805 struct ceph_vxattr *vxattr = NULL;
806 int req_mask;
807 int err;
808
809 /* let's see if a virtual xattr was requested */
810 vxattr = ceph_match_vxattr(inode, name);
811 if (vxattr) {
812 err = ceph_do_getattr(inode, 0, true);
813 if (err)
814 return err;
815 err = -ENODATA;
816 if (!(vxattr->exists_cb && !vxattr->exists_cb(ci)))
817 err = vxattr->getxattr_cb(ci, value, size);
818 return err;
819 }
820
821 req_mask = __get_request_mask(inode);
822
823 spin_lock(&ci->i_ceph_lock);
824 dout("getxattr %p ver=%lld index_ver=%lld\n", inode,
825 ci->i_xattrs.version, ci->i_xattrs.index_version);
826
827 if (ci->i_xattrs.version == 0 ||
828 !((req_mask & CEPH_CAP_XATTR_SHARED) ||
829 __ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1))) {
830 spin_unlock(&ci->i_ceph_lock);
831
832 /* security module gets xattr while filling trace */
833 if (current->journal_info) {
834 pr_warn_ratelimited("sync getxattr %p "
835 "during filling trace\n", inode);
836 return -EBUSY;
837 }
838
839 /* get xattrs from mds (if we don't already have them) */
840 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
841 if (err)
842 return err;
843 spin_lock(&ci->i_ceph_lock);
844 }
845
846 err = __build_xattrs(inode);
847 if (err < 0)
848 goto out;
849
850 err = -ENODATA; /* == ENOATTR */
851 xattr = __get_xattr(ci, name);
852 if (!xattr)
853 goto out;
854
855 err = -ERANGE;
856 if (size && size < xattr->val_len)
857 goto out;
858
859 err = xattr->val_len;
860 if (size == 0)
861 goto out;
862
863 memcpy(value, xattr->val, xattr->val_len);
864
865 if (current->journal_info &&
866 !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN))
867 ci->i_ceph_flags |= CEPH_I_SEC_INITED;
868out:
869 spin_unlock(&ci->i_ceph_lock);
870 return err;
871}
872
873ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)
874{
875 struct inode *inode = d_inode(dentry);
876 struct ceph_inode_info *ci = ceph_inode(inode);
877 struct ceph_vxattr *vxattrs = ceph_inode_vxattrs(inode);
878 u32 vir_namelen = 0;
879 u32 namelen;
880 int err;
881 u32 len;
882 int i;
883
884 spin_lock(&ci->i_ceph_lock);
885 dout("listxattr %p ver=%lld index_ver=%lld\n", inode,
886 ci->i_xattrs.version, ci->i_xattrs.index_version);
887
888 if (ci->i_xattrs.version == 0 ||
889 !__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1)) {
890 spin_unlock(&ci->i_ceph_lock);
891 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
892 if (err)
893 return err;
894 spin_lock(&ci->i_ceph_lock);
895 }
896
897 err = __build_xattrs(inode);
898 if (err < 0)
899 goto out;
900 /*
901 * Start with virtual dir xattr names (if any) (including
902 * terminating '\0' characters for each).
903 */
904 vir_namelen = ceph_vxattrs_name_size(vxattrs);
905
906 /* adding 1 byte per each variable due to the null termination */
907 namelen = ci->i_xattrs.names_size + ci->i_xattrs.count;
908 err = -ERANGE;
909 if (size && vir_namelen + namelen > size)
910 goto out;
911
912 err = namelen + vir_namelen;
913 if (size == 0)
914 goto out;
915
916 names = __copy_xattr_names(ci, names);
917
918 /* virtual xattr names, too */
919 err = namelen;
920 if (vxattrs) {
921 for (i = 0; vxattrs[i].name; i++) {
922 if (!vxattrs[i].hidden &&
923 !(vxattrs[i].exists_cb &&
924 !vxattrs[i].exists_cb(ci))) {
925 len = sprintf(names, "%s", vxattrs[i].name);
926 names += len + 1;
927 err += len + 1;
928 }
929 }
930 }
931
932out:
933 spin_unlock(&ci->i_ceph_lock);
934 return err;
935}
936
937static int ceph_sync_setxattr(struct inode *inode, const char *name,
938 const char *value, size_t size, int flags)
939{
940 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
941 struct ceph_inode_info *ci = ceph_inode(inode);
942 struct ceph_mds_request *req;
943 struct ceph_mds_client *mdsc = fsc->mdsc;
944 struct ceph_pagelist *pagelist = NULL;
945 int op = CEPH_MDS_OP_SETXATTR;
946 int err;
947
948 if (size > 0) {
949 /* copy value into pagelist */
950 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
951 if (!pagelist)
952 return -ENOMEM;
953
954 ceph_pagelist_init(pagelist);
955 err = ceph_pagelist_append(pagelist, value, size);
956 if (err)
957 goto out;
958 } else if (!value) {
959 if (flags & CEPH_XATTR_REPLACE)
960 op = CEPH_MDS_OP_RMXATTR;
961 else
962 flags |= CEPH_XATTR_REMOVE;
963 }
964
965 dout("setxattr value=%.*s\n", (int)size, value);
966
967 /* do request */
968 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
969 if (IS_ERR(req)) {
970 err = PTR_ERR(req);
971 goto out;
972 }
973
974 req->r_path2 = kstrdup(name, GFP_NOFS);
975 if (!req->r_path2) {
976 ceph_mdsc_put_request(req);
977 err = -ENOMEM;
978 goto out;
979 }
980
981 if (op == CEPH_MDS_OP_SETXATTR) {
982 req->r_args.setxattr.flags = cpu_to_le32(flags);
983 req->r_pagelist = pagelist;
984 pagelist = NULL;
985 }
986
987 req->r_inode = inode;
988 ihold(inode);
989 req->r_num_caps = 1;
990 req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
991
992 dout("xattr.ver (before): %lld\n", ci->i_xattrs.version);
993 err = ceph_mdsc_do_request(mdsc, NULL, req);
994 ceph_mdsc_put_request(req);
995 dout("xattr.ver (after): %lld\n", ci->i_xattrs.version);
996
997out:
998 if (pagelist)
999 ceph_pagelist_release(pagelist);
1000 return err;
1001}
1002
1003int __ceph_setxattr(struct inode *inode, const char *name,
1004 const void *value, size_t size, int flags)
1005{
1006 struct ceph_vxattr *vxattr;
1007 struct ceph_inode_info *ci = ceph_inode(inode);
1008 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1009 struct ceph_cap_flush *prealloc_cf = NULL;
1010 int issued;
1011 int err;
1012 int dirty = 0;
1013 int name_len = strlen(name);
1014 int val_len = size;
1015 char *newname = NULL;
1016 char *newval = NULL;
1017 struct ceph_inode_xattr *xattr = NULL;
1018 int required_blob_size;
1019 bool check_realm = false;
1020 bool lock_snap_rwsem = false;
1021
1022 if (ceph_snap(inode) != CEPH_NOSNAP)
1023 return -EROFS;
1024
1025 vxattr = ceph_match_vxattr(inode, name);
1026 if (vxattr) {
1027 if (vxattr->readonly)
1028 return -EOPNOTSUPP;
1029 if (value && !strncmp(vxattr->name, "ceph.quota", 10))
1030 check_realm = true;
1031 }
1032
1033 /* pass any unhandled ceph.* xattrs through to the MDS */
1034 if (!strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
1035 goto do_sync_unlocked;
1036
1037 /* preallocate memory for xattr name, value, index node */
1038 err = -ENOMEM;
1039 newname = kmemdup(name, name_len + 1, GFP_NOFS);
1040 if (!newname)
1041 goto out;
1042
1043 if (val_len) {
1044 newval = kmemdup(value, val_len, GFP_NOFS);
1045 if (!newval)
1046 goto out;
1047 }
1048
1049 xattr = kmalloc(sizeof(struct ceph_inode_xattr), GFP_NOFS);
1050 if (!xattr)
1051 goto out;
1052
1053 prealloc_cf = ceph_alloc_cap_flush();
1054 if (!prealloc_cf)
1055 goto out;
1056
1057 spin_lock(&ci->i_ceph_lock);
1058retry:
1059 issued = __ceph_caps_issued(ci, NULL);
1060 if (ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))
1061 goto do_sync;
1062
1063 if (!lock_snap_rwsem && !ci->i_head_snapc) {
1064 lock_snap_rwsem = true;
1065 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1066 spin_unlock(&ci->i_ceph_lock);
1067 down_read(&mdsc->snap_rwsem);
1068 spin_lock(&ci->i_ceph_lock);
1069 goto retry;
1070 }
1071 }
1072
1073 dout("setxattr %p issued %s\n", inode, ceph_cap_string(issued));
1074 __build_xattrs(inode);
1075
1076 required_blob_size = __get_required_blob_size(ci, name_len, val_len);
1077
1078 if (!ci->i_xattrs.prealloc_blob ||
1079 required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
1080 struct ceph_buffer *blob;
1081
1082 spin_unlock(&ci->i_ceph_lock);
1083 dout(" preaallocating new blob size=%d\n", required_blob_size);
1084 blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
1085 if (!blob)
1086 goto do_sync_unlocked;
1087 spin_lock(&ci->i_ceph_lock);
1088 if (ci->i_xattrs.prealloc_blob)
1089 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
1090 ci->i_xattrs.prealloc_blob = blob;
1091 goto retry;
1092 }
1093
1094 err = __set_xattr(ci, newname, name_len, newval, val_len,
1095 flags, value ? 1 : -1, &xattr);
1096
1097 if (!err) {
1098 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL,
1099 &prealloc_cf);
1100 ci->i_xattrs.dirty = true;
1101 inode->i_ctime = current_time(inode);
1102 }
1103
1104 spin_unlock(&ci->i_ceph_lock);
1105 if (lock_snap_rwsem)
1106 up_read(&mdsc->snap_rwsem);
1107 if (dirty)
1108 __mark_inode_dirty(inode, dirty);
1109 ceph_free_cap_flush(prealloc_cf);
1110 return err;
1111
1112do_sync:
1113 spin_unlock(&ci->i_ceph_lock);
1114do_sync_unlocked:
1115 if (lock_snap_rwsem)
1116 up_read(&mdsc->snap_rwsem);
1117
1118 /* security module set xattr while filling trace */
1119 if (current->journal_info) {
1120 pr_warn_ratelimited("sync setxattr %p "
1121 "during filling trace\n", inode);
1122 err = -EBUSY;
1123 } else {
1124 err = ceph_sync_setxattr(inode, name, value, size, flags);
1125 if (err >= 0 && check_realm) {
1126 /* check if snaprealm was created for quota inode */
1127 spin_lock(&ci->i_ceph_lock);
1128 if ((ci->i_max_files || ci->i_max_bytes) &&
1129 !(ci->i_snap_realm &&
1130 ci->i_snap_realm->ino == ci->i_vino.ino))
1131 err = -EOPNOTSUPP;
1132 spin_unlock(&ci->i_ceph_lock);
1133 }
1134 }
1135out:
1136 ceph_free_cap_flush(prealloc_cf);
1137 kfree(newname);
1138 kfree(newval);
1139 kfree(xattr);
1140 return err;
1141}
1142
1143static int ceph_get_xattr_handler(const struct xattr_handler *handler,
1144 struct dentry *dentry, struct inode *inode,
1145 const char *name, void *value, size_t size)
1146{
1147 if (!ceph_is_valid_xattr(name))
1148 return -EOPNOTSUPP;
1149 return __ceph_getxattr(inode, name, value, size);
1150}
1151
1152static int ceph_set_xattr_handler(const struct xattr_handler *handler,
1153 struct dentry *unused, struct inode *inode,
1154 const char *name, const void *value,
1155 size_t size, int flags)
1156{
1157 if (!ceph_is_valid_xattr(name))
1158 return -EOPNOTSUPP;
1159 return __ceph_setxattr(inode, name, value, size, flags);
1160}
1161
1162static const struct xattr_handler ceph_other_xattr_handler = {
1163 .prefix = "", /* match any name => handlers called with full name */
1164 .get = ceph_get_xattr_handler,
1165 .set = ceph_set_xattr_handler,
1166};
1167
1168#ifdef CONFIG_SECURITY
1169bool ceph_security_xattr_wanted(struct inode *in)
1170{
1171 return in->i_security != NULL;
1172}
1173
1174bool ceph_security_xattr_deadlock(struct inode *in)
1175{
1176 struct ceph_inode_info *ci;
1177 bool ret;
1178 if (!in->i_security)
1179 return false;
1180 ci = ceph_inode(in);
1181 spin_lock(&ci->i_ceph_lock);
1182 ret = !(ci->i_ceph_flags & CEPH_I_SEC_INITED) &&
1183 !(ci->i_xattrs.version > 0 &&
1184 __ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 0));
1185 spin_unlock(&ci->i_ceph_lock);
1186 return ret;
1187}
1188#endif