ocfs2: fix locking for res->tracking and dlm->tracking_list
[linux/fpc-iii.git] / fs / ceph / xattr.c
blobb24275ef97f74d0d49f0f6f32ed4c32c9c3a6f8c
1 #include <linux/ceph/ceph_debug.h>
2 #include <linux/ceph/pagelist.h>
4 #include "super.h"
5 #include "mds_client.h"
7 #include <linux/ceph/decode.h>
9 #include <linux/xattr.h>
10 #include <linux/posix_acl_xattr.h>
11 #include <linux/slab.h>
13 #define XATTR_CEPH_PREFIX "ceph."
14 #define XATTR_CEPH_PREFIX_LEN (sizeof (XATTR_CEPH_PREFIX) - 1)
16 static int __remove_xattr(struct ceph_inode_info *ci,
17 struct ceph_inode_xattr *xattr);
20 * List of handlers for synthetic system.* attributes. Other
21 * attributes are handled directly.
23 const struct xattr_handler *ceph_xattr_handlers[] = {
24 #ifdef CONFIG_CEPH_FS_POSIX_ACL
25 &posix_acl_access_xattr_handler,
26 &posix_acl_default_xattr_handler,
27 #endif
28 NULL,
31 static bool ceph_is_valid_xattr(const char *name)
33 return !strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN) ||
34 !strncmp(name, XATTR_SECURITY_PREFIX,
35 XATTR_SECURITY_PREFIX_LEN) ||
36 !strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN) ||
37 !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
38 !strncmp(name, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN);
42 * These define virtual xattrs exposing the recursive directory
43 * statistics and layout metadata.
45 struct ceph_vxattr {
46 char *name;
47 size_t name_size; /* strlen(name) + 1 (for '\0') */
48 size_t (*getxattr_cb)(struct ceph_inode_info *ci, char *val,
49 size_t size);
50 bool readonly, hidden;
51 bool (*exists_cb)(struct ceph_inode_info *ci);
54 /* layouts */
56 static bool ceph_vxattrcb_layout_exists(struct ceph_inode_info *ci)
58 size_t s;
59 char *p = (char *)&ci->i_layout;
61 for (s = 0; s < sizeof(ci->i_layout); s++, p++)
62 if (*p)
63 return true;
64 return false;
67 static size_t ceph_vxattrcb_layout(struct ceph_inode_info *ci, char *val,
68 size_t size)
70 int ret;
71 struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
72 struct ceph_osd_client *osdc = &fsc->client->osdc;
73 s64 pool = ceph_file_layout_pg_pool(ci->i_layout);
74 const char *pool_name;
75 char buf[128];
77 dout("ceph_vxattrcb_layout %p\n", &ci->vfs_inode);
78 down_read(&osdc->map_sem);
79 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
80 if (pool_name) {
81 size_t len = strlen(pool_name);
82 ret = snprintf(buf, sizeof(buf),
83 "stripe_unit=%lld stripe_count=%lld object_size=%lld pool=",
84 (unsigned long long)ceph_file_layout_su(ci->i_layout),
85 (unsigned long long)ceph_file_layout_stripe_count(ci->i_layout),
86 (unsigned long long)ceph_file_layout_object_size(ci->i_layout));
87 if (!size) {
88 ret += len;
89 } else if (ret + len > size) {
90 ret = -ERANGE;
91 } else {
92 memcpy(val, buf, ret);
93 memcpy(val + ret, pool_name, len);
94 ret += len;
96 } else {
97 ret = snprintf(buf, sizeof(buf),
98 "stripe_unit=%lld stripe_count=%lld object_size=%lld pool=%lld",
99 (unsigned long long)ceph_file_layout_su(ci->i_layout),
100 (unsigned long long)ceph_file_layout_stripe_count(ci->i_layout),
101 (unsigned long long)ceph_file_layout_object_size(ci->i_layout),
102 (unsigned long long)pool);
103 if (size) {
104 if (ret <= size)
105 memcpy(val, buf, ret);
106 else
107 ret = -ERANGE;
110 up_read(&osdc->map_sem);
111 return ret;
114 static size_t ceph_vxattrcb_layout_stripe_unit(struct ceph_inode_info *ci,
115 char *val, size_t size)
117 return snprintf(val, size, "%lld",
118 (unsigned long long)ceph_file_layout_su(ci->i_layout));
121 static size_t ceph_vxattrcb_layout_stripe_count(struct ceph_inode_info *ci,
122 char *val, size_t size)
124 return snprintf(val, size, "%lld",
125 (unsigned long long)ceph_file_layout_stripe_count(ci->i_layout));
128 static size_t ceph_vxattrcb_layout_object_size(struct ceph_inode_info *ci,
129 char *val, size_t size)
131 return snprintf(val, size, "%lld",
132 (unsigned long long)ceph_file_layout_object_size(ci->i_layout));
135 static size_t ceph_vxattrcb_layout_pool(struct ceph_inode_info *ci,
136 char *val, size_t size)
138 int ret;
139 struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
140 struct ceph_osd_client *osdc = &fsc->client->osdc;
141 s64 pool = ceph_file_layout_pg_pool(ci->i_layout);
142 const char *pool_name;
144 down_read(&osdc->map_sem);
145 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
146 if (pool_name)
147 ret = snprintf(val, size, "%s", pool_name);
148 else
149 ret = snprintf(val, size, "%lld", (unsigned long long)pool);
150 up_read(&osdc->map_sem);
151 return ret;
154 /* directories */
156 static size_t ceph_vxattrcb_dir_entries(struct ceph_inode_info *ci, char *val,
157 size_t size)
159 return snprintf(val, size, "%lld", ci->i_files + ci->i_subdirs);
162 static size_t ceph_vxattrcb_dir_files(struct ceph_inode_info *ci, char *val,
163 size_t size)
165 return snprintf(val, size, "%lld", ci->i_files);
168 static size_t ceph_vxattrcb_dir_subdirs(struct ceph_inode_info *ci, char *val,
169 size_t size)
171 return snprintf(val, size, "%lld", ci->i_subdirs);
174 static size_t ceph_vxattrcb_dir_rentries(struct ceph_inode_info *ci, char *val,
175 size_t size)
177 return snprintf(val, size, "%lld", ci->i_rfiles + ci->i_rsubdirs);
180 static size_t ceph_vxattrcb_dir_rfiles(struct ceph_inode_info *ci, char *val,
181 size_t size)
183 return snprintf(val, size, "%lld", ci->i_rfiles);
186 static size_t ceph_vxattrcb_dir_rsubdirs(struct ceph_inode_info *ci, char *val,
187 size_t size)
189 return snprintf(val, size, "%lld", ci->i_rsubdirs);
192 static size_t ceph_vxattrcb_dir_rbytes(struct ceph_inode_info *ci, char *val,
193 size_t size)
195 return snprintf(val, size, "%lld", ci->i_rbytes);
198 static size_t ceph_vxattrcb_dir_rctime(struct ceph_inode_info *ci, char *val,
199 size_t size)
201 return snprintf(val, size, "%ld.09%ld", (long)ci->i_rctime.tv_sec,
202 (long)ci->i_rctime.tv_nsec);
206 #define CEPH_XATTR_NAME(_type, _name) XATTR_CEPH_PREFIX #_type "." #_name
207 #define CEPH_XATTR_NAME2(_type, _name, _name2) \
208 XATTR_CEPH_PREFIX #_type "." #_name "." #_name2
210 #define XATTR_NAME_CEPH(_type, _name) \
212 .name = CEPH_XATTR_NAME(_type, _name), \
213 .name_size = sizeof (CEPH_XATTR_NAME(_type, _name)), \
214 .getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name, \
215 .readonly = true, \
216 .hidden = false, \
217 .exists_cb = NULL, \
219 #define XATTR_LAYOUT_FIELD(_type, _name, _field) \
221 .name = CEPH_XATTR_NAME2(_type, _name, _field), \
222 .name_size = sizeof (CEPH_XATTR_NAME2(_type, _name, _field)), \
223 .getxattr_cb = ceph_vxattrcb_ ## _name ## _ ## _field, \
224 .readonly = false, \
225 .hidden = true, \
226 .exists_cb = ceph_vxattrcb_layout_exists, \
229 static struct ceph_vxattr ceph_dir_vxattrs[] = {
231 .name = "ceph.dir.layout",
232 .name_size = sizeof("ceph.dir.layout"),
233 .getxattr_cb = ceph_vxattrcb_layout,
234 .readonly = false,
235 .hidden = true,
236 .exists_cb = ceph_vxattrcb_layout_exists,
238 XATTR_LAYOUT_FIELD(dir, layout, stripe_unit),
239 XATTR_LAYOUT_FIELD(dir, layout, stripe_count),
240 XATTR_LAYOUT_FIELD(dir, layout, object_size),
241 XATTR_LAYOUT_FIELD(dir, layout, pool),
242 XATTR_NAME_CEPH(dir, entries),
243 XATTR_NAME_CEPH(dir, files),
244 XATTR_NAME_CEPH(dir, subdirs),
245 XATTR_NAME_CEPH(dir, rentries),
246 XATTR_NAME_CEPH(dir, rfiles),
247 XATTR_NAME_CEPH(dir, rsubdirs),
248 XATTR_NAME_CEPH(dir, rbytes),
249 XATTR_NAME_CEPH(dir, rctime),
250 { .name = NULL, 0 } /* Required table terminator */
252 static size_t ceph_dir_vxattrs_name_size; /* total size of all names */
254 /* files */
256 static struct ceph_vxattr ceph_file_vxattrs[] = {
258 .name = "ceph.file.layout",
259 .name_size = sizeof("ceph.file.layout"),
260 .getxattr_cb = ceph_vxattrcb_layout,
261 .readonly = false,
262 .hidden = true,
263 .exists_cb = ceph_vxattrcb_layout_exists,
265 XATTR_LAYOUT_FIELD(file, layout, stripe_unit),
266 XATTR_LAYOUT_FIELD(file, layout, stripe_count),
267 XATTR_LAYOUT_FIELD(file, layout, object_size),
268 XATTR_LAYOUT_FIELD(file, layout, pool),
269 { .name = NULL, 0 } /* Required table terminator */
271 static size_t ceph_file_vxattrs_name_size; /* total size of all names */
273 static struct ceph_vxattr *ceph_inode_vxattrs(struct inode *inode)
275 if (S_ISDIR(inode->i_mode))
276 return ceph_dir_vxattrs;
277 else if (S_ISREG(inode->i_mode))
278 return ceph_file_vxattrs;
279 return NULL;
282 static size_t ceph_vxattrs_name_size(struct ceph_vxattr *vxattrs)
284 if (vxattrs == ceph_dir_vxattrs)
285 return ceph_dir_vxattrs_name_size;
286 if (vxattrs == ceph_file_vxattrs)
287 return ceph_file_vxattrs_name_size;
288 BUG_ON(vxattrs);
289 return 0;
293 * Compute the aggregate size (including terminating '\0') of all
294 * virtual extended attribute names in the given vxattr table.
296 static size_t __init vxattrs_name_size(struct ceph_vxattr *vxattrs)
298 struct ceph_vxattr *vxattr;
299 size_t size = 0;
301 for (vxattr = vxattrs; vxattr->name; vxattr++)
302 if (!vxattr->hidden)
303 size += vxattr->name_size;
305 return size;
308 /* Routines called at initialization and exit time */
310 void __init ceph_xattr_init(void)
312 ceph_dir_vxattrs_name_size = vxattrs_name_size(ceph_dir_vxattrs);
313 ceph_file_vxattrs_name_size = vxattrs_name_size(ceph_file_vxattrs);
316 void ceph_xattr_exit(void)
318 ceph_dir_vxattrs_name_size = 0;
319 ceph_file_vxattrs_name_size = 0;
322 static struct ceph_vxattr *ceph_match_vxattr(struct inode *inode,
323 const char *name)
325 struct ceph_vxattr *vxattr = ceph_inode_vxattrs(inode);
327 if (vxattr) {
328 while (vxattr->name) {
329 if (!strcmp(vxattr->name, name))
330 return vxattr;
331 vxattr++;
335 return NULL;
338 static int __set_xattr(struct ceph_inode_info *ci,
339 const char *name, int name_len,
340 const char *val, int val_len,
341 int flags, int update_xattr,
342 struct ceph_inode_xattr **newxattr)
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;
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;
367 xattr = NULL;
370 if (update_xattr) {
371 int err = 0;
373 if (xattr && (flags & XATTR_CREATE))
374 err = -EEXIST;
375 else if (!xattr && (flags & XATTR_REPLACE))
376 err = -ENODATA;
377 if (err) {
378 kfree(name);
379 kfree(val);
380 kfree(*newxattr);
381 return err;
383 if (update_xattr < 0) {
384 if (xattr)
385 __remove_xattr(ci, xattr);
386 kfree(name);
387 kfree(*newxattr);
388 return 0;
392 if (!xattr) {
393 new = 1;
394 xattr = *newxattr;
395 xattr->name = name;
396 xattr->name_len = name_len;
397 xattr->should_free_name = update_xattr;
399 ci->i_xattrs.count++;
400 dout("__set_xattr count=%d\n", ci->i_xattrs.count);
401 } else {
402 kfree(*newxattr);
403 *newxattr = NULL;
404 if (xattr->should_free_val)
405 kfree((void *)xattr->val);
407 if (update_xattr) {
408 kfree((void *)name);
409 name = xattr->name;
411 ci->i_xattrs.names_size -= xattr->name_len;
412 ci->i_xattrs.vals_size -= xattr->val_len;
414 ci->i_xattrs.names_size += name_len;
415 ci->i_xattrs.vals_size += val_len;
416 if (val)
417 xattr->val = val;
418 else
419 xattr->val = "";
421 xattr->val_len = val_len;
422 xattr->dirty = update_xattr;
423 xattr->should_free_val = (val && update_xattr);
425 if (new) {
426 rb_link_node(&xattr->node, parent, p);
427 rb_insert_color(&xattr->node, &ci->i_xattrs.index);
428 dout("__set_xattr_val p=%p\n", p);
431 dout("__set_xattr_val added %llx.%llx xattr %p %s=%.*s\n",
432 ceph_vinop(&ci->vfs_inode), xattr, name, val_len, val);
434 return 0;
437 static struct ceph_inode_xattr *__get_xattr(struct ceph_inode_info *ci,
438 const char *name)
440 struct rb_node **p;
441 struct rb_node *parent = NULL;
442 struct ceph_inode_xattr *xattr = NULL;
443 int name_len = strlen(name);
444 int c;
446 p = &ci->i_xattrs.index.rb_node;
447 while (*p) {
448 parent = *p;
449 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
450 c = strncmp(name, xattr->name, xattr->name_len);
451 if (c == 0 && name_len > xattr->name_len)
452 c = 1;
453 if (c < 0)
454 p = &(*p)->rb_left;
455 else if (c > 0)
456 p = &(*p)->rb_right;
457 else {
458 dout("__get_xattr %s: found %.*s\n", name,
459 xattr->val_len, xattr->val);
460 return xattr;
464 dout("__get_xattr %s: not found\n", name);
466 return NULL;
469 static void __free_xattr(struct ceph_inode_xattr *xattr)
471 BUG_ON(!xattr);
473 if (xattr->should_free_name)
474 kfree((void *)xattr->name);
475 if (xattr->should_free_val)
476 kfree((void *)xattr->val);
478 kfree(xattr);
481 static int __remove_xattr(struct ceph_inode_info *ci,
482 struct ceph_inode_xattr *xattr)
484 if (!xattr)
485 return -ENODATA;
487 rb_erase(&xattr->node, &ci->i_xattrs.index);
489 if (xattr->should_free_name)
490 kfree((void *)xattr->name);
491 if (xattr->should_free_val)
492 kfree((void *)xattr->val);
494 ci->i_xattrs.names_size -= xattr->name_len;
495 ci->i_xattrs.vals_size -= xattr->val_len;
496 ci->i_xattrs.count--;
497 kfree(xattr);
499 return 0;
502 static int __remove_xattr_by_name(struct ceph_inode_info *ci,
503 const char *name)
505 struct rb_node **p;
506 struct ceph_inode_xattr *xattr;
507 int err;
509 p = &ci->i_xattrs.index.rb_node;
510 xattr = __get_xattr(ci, name);
511 err = __remove_xattr(ci, xattr);
512 return err;
515 static char *__copy_xattr_names(struct ceph_inode_info *ci,
516 char *dest)
518 struct rb_node *p;
519 struct ceph_inode_xattr *xattr = NULL;
521 p = rb_first(&ci->i_xattrs.index);
522 dout("__copy_xattr_names count=%d\n", ci->i_xattrs.count);
524 while (p) {
525 xattr = rb_entry(p, struct ceph_inode_xattr, node);
526 memcpy(dest, xattr->name, xattr->name_len);
527 dest[xattr->name_len] = '\0';
529 dout("dest=%s %p (%s) (%d/%d)\n", dest, xattr, xattr->name,
530 xattr->name_len, ci->i_xattrs.names_size);
532 dest += xattr->name_len + 1;
533 p = rb_next(p);
536 return dest;
539 void __ceph_destroy_xattrs(struct ceph_inode_info *ci)
541 struct rb_node *p, *tmp;
542 struct ceph_inode_xattr *xattr = NULL;
544 p = rb_first(&ci->i_xattrs.index);
546 dout("__ceph_destroy_xattrs p=%p\n", p);
548 while (p) {
549 xattr = rb_entry(p, struct ceph_inode_xattr, node);
550 tmp = p;
551 p = rb_next(tmp);
552 dout("__ceph_destroy_xattrs next p=%p (%.*s)\n", p,
553 xattr->name_len, xattr->name);
554 rb_erase(tmp, &ci->i_xattrs.index);
556 __free_xattr(xattr);
559 ci->i_xattrs.names_size = 0;
560 ci->i_xattrs.vals_size = 0;
561 ci->i_xattrs.index_version = 0;
562 ci->i_xattrs.count = 0;
563 ci->i_xattrs.index = RB_ROOT;
566 static int __build_xattrs(struct inode *inode)
567 __releases(ci->i_ceph_lock)
568 __acquires(ci->i_ceph_lock)
570 u32 namelen;
571 u32 numattr = 0;
572 void *p, *end;
573 u32 len;
574 const char *name, *val;
575 struct ceph_inode_info *ci = ceph_inode(inode);
576 int xattr_version;
577 struct ceph_inode_xattr **xattrs = NULL;
578 int err = 0;
579 int i;
581 dout("__build_xattrs() len=%d\n",
582 ci->i_xattrs.blob ? (int)ci->i_xattrs.blob->vec.iov_len : 0);
584 if (ci->i_xattrs.index_version >= ci->i_xattrs.version)
585 return 0; /* already built */
587 __ceph_destroy_xattrs(ci);
589 start:
590 /* updated internal xattr rb tree */
591 if (ci->i_xattrs.blob && ci->i_xattrs.blob->vec.iov_len > 4) {
592 p = ci->i_xattrs.blob->vec.iov_base;
593 end = p + ci->i_xattrs.blob->vec.iov_len;
594 ceph_decode_32_safe(&p, end, numattr, bad);
595 xattr_version = ci->i_xattrs.version;
596 spin_unlock(&ci->i_ceph_lock);
598 xattrs = kcalloc(numattr, sizeof(struct ceph_inode_xattr *),
599 GFP_NOFS);
600 err = -ENOMEM;
601 if (!xattrs)
602 goto bad_lock;
604 for (i = 0; i < numattr; i++) {
605 xattrs[i] = kmalloc(sizeof(struct ceph_inode_xattr),
606 GFP_NOFS);
607 if (!xattrs[i])
608 goto bad_lock;
611 spin_lock(&ci->i_ceph_lock);
612 if (ci->i_xattrs.version != xattr_version) {
613 /* lost a race, retry */
614 for (i = 0; i < numattr; i++)
615 kfree(xattrs[i]);
616 kfree(xattrs);
617 xattrs = NULL;
618 goto start;
620 err = -EIO;
621 while (numattr--) {
622 ceph_decode_32_safe(&p, end, len, bad);
623 namelen = len;
624 name = p;
625 p += len;
626 ceph_decode_32_safe(&p, end, len, bad);
627 val = p;
628 p += len;
630 err = __set_xattr(ci, name, namelen, val, len,
631 0, 0, &xattrs[numattr]);
633 if (err < 0)
634 goto bad;
636 kfree(xattrs);
638 ci->i_xattrs.index_version = ci->i_xattrs.version;
639 ci->i_xattrs.dirty = false;
641 return err;
642 bad_lock:
643 spin_lock(&ci->i_ceph_lock);
644 bad:
645 if (xattrs) {
646 for (i = 0; i < numattr; i++)
647 kfree(xattrs[i]);
648 kfree(xattrs);
650 ci->i_xattrs.names_size = 0;
651 return err;
654 static int __get_required_blob_size(struct ceph_inode_info *ci, int name_size,
655 int val_size)
658 * 4 bytes for the length, and additional 4 bytes per each xattr name,
659 * 4 bytes per each value
661 int size = 4 + ci->i_xattrs.count*(4 + 4) +
662 ci->i_xattrs.names_size +
663 ci->i_xattrs.vals_size;
664 dout("__get_required_blob_size c=%d names.size=%d vals.size=%d\n",
665 ci->i_xattrs.count, ci->i_xattrs.names_size,
666 ci->i_xattrs.vals_size);
668 if (name_size)
669 size += 4 + 4 + name_size + val_size;
671 return size;
675 * If there are dirty xattrs, reencode xattrs into the prealloc_blob
676 * and swap into place.
678 void __ceph_build_xattrs_blob(struct ceph_inode_info *ci)
680 struct rb_node *p;
681 struct ceph_inode_xattr *xattr = NULL;
682 void *dest;
684 dout("__build_xattrs_blob %p\n", &ci->vfs_inode);
685 if (ci->i_xattrs.dirty) {
686 int need = __get_required_blob_size(ci, 0, 0);
688 BUG_ON(need > ci->i_xattrs.prealloc_blob->alloc_len);
690 p = rb_first(&ci->i_xattrs.index);
691 dest = ci->i_xattrs.prealloc_blob->vec.iov_base;
693 ceph_encode_32(&dest, ci->i_xattrs.count);
694 while (p) {
695 xattr = rb_entry(p, struct ceph_inode_xattr, node);
697 ceph_encode_32(&dest, xattr->name_len);
698 memcpy(dest, xattr->name, xattr->name_len);
699 dest += xattr->name_len;
700 ceph_encode_32(&dest, xattr->val_len);
701 memcpy(dest, xattr->val, xattr->val_len);
702 dest += xattr->val_len;
704 p = rb_next(p);
707 /* adjust buffer len; it may be larger than we need */
708 ci->i_xattrs.prealloc_blob->vec.iov_len =
709 dest - ci->i_xattrs.prealloc_blob->vec.iov_base;
711 if (ci->i_xattrs.blob)
712 ceph_buffer_put(ci->i_xattrs.blob);
713 ci->i_xattrs.blob = ci->i_xattrs.prealloc_blob;
714 ci->i_xattrs.prealloc_blob = NULL;
715 ci->i_xattrs.dirty = false;
716 ci->i_xattrs.version++;
720 ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,
721 size_t size)
723 struct ceph_inode_info *ci = ceph_inode(inode);
724 int err;
725 struct ceph_inode_xattr *xattr;
726 struct ceph_vxattr *vxattr = NULL;
728 if (!ceph_is_valid_xattr(name))
729 return -ENODATA;
731 /* let's see if a virtual xattr was requested */
732 vxattr = ceph_match_vxattr(inode, name);
733 if (vxattr && !(vxattr->exists_cb && !vxattr->exists_cb(ci))) {
734 err = vxattr->getxattr_cb(ci, value, size);
735 return err;
738 spin_lock(&ci->i_ceph_lock);
739 dout("getxattr %p ver=%lld index_ver=%lld\n", inode,
740 ci->i_xattrs.version, ci->i_xattrs.index_version);
742 if (ci->i_xattrs.version == 0 ||
743 !__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1)) {
744 spin_unlock(&ci->i_ceph_lock);
745 /* get xattrs from mds (if we don't already have them) */
746 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
747 if (err)
748 return err;
749 spin_lock(&ci->i_ceph_lock);
752 err = __build_xattrs(inode);
753 if (err < 0)
754 goto out;
756 err = -ENODATA; /* == ENOATTR */
757 xattr = __get_xattr(ci, name);
758 if (!xattr)
759 goto out;
761 err = -ERANGE;
762 if (size && size < xattr->val_len)
763 goto out;
765 err = xattr->val_len;
766 if (size == 0)
767 goto out;
769 memcpy(value, xattr->val, xattr->val_len);
771 out:
772 spin_unlock(&ci->i_ceph_lock);
773 return err;
776 ssize_t ceph_getxattr(struct dentry *dentry, const char *name, void *value,
777 size_t size)
779 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
780 return generic_getxattr(dentry, name, value, size);
782 return __ceph_getxattr(d_inode(dentry), name, value, size);
785 ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)
787 struct inode *inode = d_inode(dentry);
788 struct ceph_inode_info *ci = ceph_inode(inode);
789 struct ceph_vxattr *vxattrs = ceph_inode_vxattrs(inode);
790 u32 vir_namelen = 0;
791 u32 namelen;
792 int err;
793 u32 len;
794 int i;
796 spin_lock(&ci->i_ceph_lock);
797 dout("listxattr %p ver=%lld index_ver=%lld\n", inode,
798 ci->i_xattrs.version, ci->i_xattrs.index_version);
800 if (ci->i_xattrs.version == 0 ||
801 !__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1)) {
802 spin_unlock(&ci->i_ceph_lock);
803 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
804 if (err)
805 return err;
806 spin_lock(&ci->i_ceph_lock);
809 err = __build_xattrs(inode);
810 if (err < 0)
811 goto out;
813 * Start with virtual dir xattr names (if any) (including
814 * terminating '\0' characters for each).
816 vir_namelen = ceph_vxattrs_name_size(vxattrs);
818 /* adding 1 byte per each variable due to the null termination */
819 namelen = ci->i_xattrs.names_size + ci->i_xattrs.count;
820 err = -ERANGE;
821 if (size && vir_namelen + namelen > size)
822 goto out;
824 err = namelen + vir_namelen;
825 if (size == 0)
826 goto out;
828 names = __copy_xattr_names(ci, names);
830 /* virtual xattr names, too */
831 err = namelen;
832 if (vxattrs) {
833 for (i = 0; vxattrs[i].name; i++) {
834 if (!vxattrs[i].hidden &&
835 !(vxattrs[i].exists_cb &&
836 !vxattrs[i].exists_cb(ci))) {
837 len = sprintf(names, "%s", vxattrs[i].name);
838 names += len + 1;
839 err += len + 1;
844 out:
845 spin_unlock(&ci->i_ceph_lock);
846 return err;
849 static int ceph_sync_setxattr(struct dentry *dentry, const char *name,
850 const char *value, size_t size, int flags)
852 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
853 struct inode *inode = d_inode(dentry);
854 struct ceph_inode_info *ci = ceph_inode(inode);
855 struct ceph_mds_request *req;
856 struct ceph_mds_client *mdsc = fsc->mdsc;
857 struct ceph_pagelist *pagelist = NULL;
858 int err;
860 if (size > 0) {
861 /* copy value into pagelist */
862 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
863 if (!pagelist)
864 return -ENOMEM;
866 ceph_pagelist_init(pagelist);
867 err = ceph_pagelist_append(pagelist, value, size);
868 if (err)
869 goto out;
870 } else if (!value) {
871 flags |= CEPH_XATTR_REMOVE;
874 dout("setxattr value=%.*s\n", (int)size, value);
876 /* do request */
877 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETXATTR,
878 USE_AUTH_MDS);
879 if (IS_ERR(req)) {
880 err = PTR_ERR(req);
881 goto out;
884 req->r_args.setxattr.flags = cpu_to_le32(flags);
885 req->r_path2 = kstrdup(name, GFP_NOFS);
886 if (!req->r_path2) {
887 ceph_mdsc_put_request(req);
888 err = -ENOMEM;
889 goto out;
892 req->r_pagelist = pagelist;
893 pagelist = NULL;
895 req->r_inode = inode;
896 ihold(inode);
897 req->r_num_caps = 1;
898 req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
900 dout("xattr.ver (before): %lld\n", ci->i_xattrs.version);
901 err = ceph_mdsc_do_request(mdsc, NULL, req);
902 ceph_mdsc_put_request(req);
903 dout("xattr.ver (after): %lld\n", ci->i_xattrs.version);
905 out:
906 if (pagelist)
907 ceph_pagelist_release(pagelist);
908 return err;
911 int __ceph_setxattr(struct dentry *dentry, const char *name,
912 const void *value, size_t size, int flags)
914 struct inode *inode = d_inode(dentry);
915 struct ceph_vxattr *vxattr;
916 struct ceph_inode_info *ci = ceph_inode(inode);
917 struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
918 struct ceph_cap_flush *prealloc_cf = NULL;
919 int issued;
920 int err;
921 int dirty = 0;
922 int name_len = strlen(name);
923 int val_len = size;
924 char *newname = NULL;
925 char *newval = NULL;
926 struct ceph_inode_xattr *xattr = NULL;
927 int required_blob_size;
928 bool lock_snap_rwsem = false;
930 if (!ceph_is_valid_xattr(name))
931 return -EOPNOTSUPP;
933 vxattr = ceph_match_vxattr(inode, name);
934 if (vxattr && vxattr->readonly)
935 return -EOPNOTSUPP;
937 /* pass any unhandled ceph.* xattrs through to the MDS */
938 if (!strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
939 goto do_sync_unlocked;
941 /* preallocate memory for xattr name, value, index node */
942 err = -ENOMEM;
943 newname = kmemdup(name, name_len + 1, GFP_NOFS);
944 if (!newname)
945 goto out;
947 if (val_len) {
948 newval = kmemdup(value, val_len, GFP_NOFS);
949 if (!newval)
950 goto out;
953 xattr = kmalloc(sizeof(struct ceph_inode_xattr), GFP_NOFS);
954 if (!xattr)
955 goto out;
957 prealloc_cf = ceph_alloc_cap_flush();
958 if (!prealloc_cf)
959 goto out;
961 spin_lock(&ci->i_ceph_lock);
962 retry:
963 issued = __ceph_caps_issued(ci, NULL);
964 if (ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))
965 goto do_sync;
967 if (!lock_snap_rwsem && !ci->i_head_snapc) {
968 lock_snap_rwsem = true;
969 if (!down_read_trylock(&mdsc->snap_rwsem)) {
970 spin_unlock(&ci->i_ceph_lock);
971 down_read(&mdsc->snap_rwsem);
972 spin_lock(&ci->i_ceph_lock);
973 goto retry;
977 dout("setxattr %p issued %s\n", inode, ceph_cap_string(issued));
978 __build_xattrs(inode);
980 required_blob_size = __get_required_blob_size(ci, name_len, val_len);
982 if (!ci->i_xattrs.prealloc_blob ||
983 required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
984 struct ceph_buffer *blob;
986 spin_unlock(&ci->i_ceph_lock);
987 dout(" preaallocating new blob size=%d\n", required_blob_size);
988 blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
989 if (!blob)
990 goto do_sync_unlocked;
991 spin_lock(&ci->i_ceph_lock);
992 if (ci->i_xattrs.prealloc_blob)
993 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
994 ci->i_xattrs.prealloc_blob = blob;
995 goto retry;
998 err = __set_xattr(ci, newname, name_len, newval, val_len,
999 flags, value ? 1 : -1, &xattr);
1001 if (!err) {
1002 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL,
1003 &prealloc_cf);
1004 ci->i_xattrs.dirty = true;
1005 inode->i_ctime = CURRENT_TIME;
1008 spin_unlock(&ci->i_ceph_lock);
1009 if (lock_snap_rwsem)
1010 up_read(&mdsc->snap_rwsem);
1011 if (dirty)
1012 __mark_inode_dirty(inode, dirty);
1013 ceph_free_cap_flush(prealloc_cf);
1014 return err;
1016 do_sync:
1017 spin_unlock(&ci->i_ceph_lock);
1018 do_sync_unlocked:
1019 if (lock_snap_rwsem)
1020 up_read(&mdsc->snap_rwsem);
1021 err = ceph_sync_setxattr(dentry, name, value, size, flags);
1022 out:
1023 ceph_free_cap_flush(prealloc_cf);
1024 kfree(newname);
1025 kfree(newval);
1026 kfree(xattr);
1027 return err;
1030 int ceph_setxattr(struct dentry *dentry, const char *name,
1031 const void *value, size_t size, int flags)
1033 if (ceph_snap(d_inode(dentry)) != CEPH_NOSNAP)
1034 return -EROFS;
1036 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1037 return generic_setxattr(dentry, name, value, size, flags);
1039 if (size == 0)
1040 value = ""; /* empty EA, do not remove */
1042 return __ceph_setxattr(dentry, name, value, size, flags);
1045 static int ceph_send_removexattr(struct dentry *dentry, const char *name)
1047 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
1048 struct ceph_mds_client *mdsc = fsc->mdsc;
1049 struct inode *inode = d_inode(dentry);
1050 struct ceph_mds_request *req;
1051 int err;
1053 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RMXATTR,
1054 USE_AUTH_MDS);
1055 if (IS_ERR(req))
1056 return PTR_ERR(req);
1057 req->r_path2 = kstrdup(name, GFP_NOFS);
1058 if (!req->r_path2)
1059 return -ENOMEM;
1061 req->r_inode = inode;
1062 ihold(inode);
1063 req->r_num_caps = 1;
1064 req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
1065 err = ceph_mdsc_do_request(mdsc, NULL, req);
1066 ceph_mdsc_put_request(req);
1067 return err;
1070 int __ceph_removexattr(struct dentry *dentry, const char *name)
1072 struct inode *inode = d_inode(dentry);
1073 struct ceph_vxattr *vxattr;
1074 struct ceph_inode_info *ci = ceph_inode(inode);
1075 struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
1076 struct ceph_cap_flush *prealloc_cf = NULL;
1077 int issued;
1078 int err;
1079 int required_blob_size;
1080 int dirty;
1081 bool lock_snap_rwsem = false;
1083 if (!ceph_is_valid_xattr(name))
1084 return -EOPNOTSUPP;
1086 vxattr = ceph_match_vxattr(inode, name);
1087 if (vxattr && vxattr->readonly)
1088 return -EOPNOTSUPP;
1090 /* pass any unhandled ceph.* xattrs through to the MDS */
1091 if (!strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
1092 goto do_sync_unlocked;
1094 prealloc_cf = ceph_alloc_cap_flush();
1095 if (!prealloc_cf)
1096 return -ENOMEM;
1098 err = -ENOMEM;
1099 spin_lock(&ci->i_ceph_lock);
1100 retry:
1101 issued = __ceph_caps_issued(ci, NULL);
1102 if (ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))
1103 goto do_sync;
1105 if (!lock_snap_rwsem && !ci->i_head_snapc) {
1106 lock_snap_rwsem = true;
1107 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1108 spin_unlock(&ci->i_ceph_lock);
1109 down_read(&mdsc->snap_rwsem);
1110 spin_lock(&ci->i_ceph_lock);
1111 goto retry;
1115 dout("removexattr %p issued %s\n", inode, ceph_cap_string(issued));
1117 __build_xattrs(inode);
1119 required_blob_size = __get_required_blob_size(ci, 0, 0);
1121 if (!ci->i_xattrs.prealloc_blob ||
1122 required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
1123 struct ceph_buffer *blob;
1125 spin_unlock(&ci->i_ceph_lock);
1126 dout(" preaallocating new blob size=%d\n", required_blob_size);
1127 blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
1128 if (!blob)
1129 goto do_sync_unlocked;
1130 spin_lock(&ci->i_ceph_lock);
1131 if (ci->i_xattrs.prealloc_blob)
1132 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
1133 ci->i_xattrs.prealloc_blob = blob;
1134 goto retry;
1137 err = __remove_xattr_by_name(ceph_inode(inode), name);
1139 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL,
1140 &prealloc_cf);
1141 ci->i_xattrs.dirty = true;
1142 inode->i_ctime = CURRENT_TIME;
1143 spin_unlock(&ci->i_ceph_lock);
1144 if (lock_snap_rwsem)
1145 up_read(&mdsc->snap_rwsem);
1146 if (dirty)
1147 __mark_inode_dirty(inode, dirty);
1148 ceph_free_cap_flush(prealloc_cf);
1149 return err;
1150 do_sync:
1151 spin_unlock(&ci->i_ceph_lock);
1152 do_sync_unlocked:
1153 if (lock_snap_rwsem)
1154 up_read(&mdsc->snap_rwsem);
1155 ceph_free_cap_flush(prealloc_cf);
1156 err = ceph_send_removexattr(dentry, name);
1157 return err;
1160 int ceph_removexattr(struct dentry *dentry, const char *name)
1162 if (ceph_snap(d_inode(dentry)) != CEPH_NOSNAP)
1163 return -EROFS;
1165 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1166 return generic_removexattr(dentry, name);
1168 return __ceph_removexattr(dentry, name);