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