1 #include <linux/ceph/ceph_debug.h>
2 #include <linux/ceph/pagelist.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);
19 static const struct xattr_handler ceph_other_xattr_handler;
22 * List of handlers for synthetic system.* attributes. Other
23 * attributes are handled directly.
25 const struct xattr_handler *ceph_xattr_handlers[] = {
26 #ifdef CONFIG_CEPH_FS_POSIX_ACL
27 &posix_acl_access_xattr_handler,
28 &posix_acl_default_xattr_handler,
30 &ceph_other_xattr_handler,
34 static bool ceph_is_valid_xattr(const char *name)
36 return !strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN) ||
37 !strncmp(name, XATTR_SECURITY_PREFIX,
38 XATTR_SECURITY_PREFIX_LEN) ||
39 !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
40 !strncmp(name, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN);
44 * These define virtual xattrs exposing the recursive directory
45 * statistics and layout metadata.
49 size_t name_size; /* strlen(name) + 1 (for '\0') */
50 size_t (*getxattr_cb)(struct ceph_inode_info *ci, char *val,
52 bool readonly, hidden;
53 bool (*exists_cb)(struct ceph_inode_info *ci);
58 static bool ceph_vxattrcb_layout_exists(struct ceph_inode_info *ci)
60 struct ceph_file_layout *fl = &ci->i_layout;
61 return (fl->stripe_unit > 0 || fl->stripe_count > 0 ||
62 fl->object_size > 0 || fl->pool_id >= 0 ||
63 rcu_dereference_raw(fl->pool_ns) != NULL);
66 static size_t ceph_vxattrcb_layout(struct ceph_inode_info *ci, char *val,
69 struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
70 struct ceph_osd_client *osdc = &fsc->client->osdc;
71 struct ceph_string *pool_ns;
72 s64 pool = ci->i_layout.pool_id;
73 const char *pool_name;
74 const char *ns_field = " pool_namespace=";
76 size_t len, total_len = 0;
79 pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
81 dout("ceph_vxattrcb_layout %p\n", &ci->vfs_inode);
82 down_read(&osdc->lock);
83 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
85 len = snprintf(buf, sizeof(buf),
86 "stripe_unit=%u stripe_count=%u object_size=%u pool=",
87 ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
88 ci->i_layout.object_size);
89 total_len = len + strlen(pool_name);
91 len = snprintf(buf, sizeof(buf),
92 "stripe_unit=%u stripe_count=%u object_size=%u pool=%lld",
93 ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
94 ci->i_layout.object_size, (unsigned long long)pool);
99 total_len += strlen(ns_field) + pool_ns->len;
102 if (size >= total_len) {
103 memcpy(val, buf, len);
106 len = strlen(pool_name);
107 memcpy(val + ret, pool_name, len);
111 len = strlen(ns_field);
112 memcpy(val + ret, ns_field, len);
114 memcpy(val + ret, pool_ns->str, pool_ns->len);
118 up_read(&osdc->lock);
119 ceph_put_string(pool_ns);
123 static size_t ceph_vxattrcb_layout_stripe_unit(struct ceph_inode_info *ci,
124 char *val, size_t size)
126 return snprintf(val, size, "%u", ci->i_layout.stripe_unit);
129 static size_t ceph_vxattrcb_layout_stripe_count(struct ceph_inode_info *ci,
130 char *val, size_t size)
132 return snprintf(val, size, "%u", ci->i_layout.stripe_count);
135 static size_t ceph_vxattrcb_layout_object_size(struct ceph_inode_info *ci,
136 char *val, size_t size)
138 return snprintf(val, size, "%u", ci->i_layout.object_size);
141 static size_t ceph_vxattrcb_layout_pool(struct ceph_inode_info *ci,
142 char *val, size_t size)
145 struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
146 struct ceph_osd_client *osdc = &fsc->client->osdc;
147 s64 pool = ci->i_layout.pool_id;
148 const char *pool_name;
150 down_read(&osdc->lock);
151 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
153 ret = snprintf(val, size, "%s", pool_name);
155 ret = snprintf(val, size, "%lld", (unsigned long long)pool);
156 up_read(&osdc->lock);
160 static size_t ceph_vxattrcb_layout_pool_namespace(struct ceph_inode_info *ci,
161 char *val, size_t size)
164 struct ceph_string *ns = ceph_try_get_string(ci->i_layout.pool_ns);
166 ret = snprintf(val, size, "%.*s", (int)ns->len, ns->str);
174 static size_t ceph_vxattrcb_dir_entries(struct ceph_inode_info *ci, char *val,
177 return snprintf(val, size, "%lld", ci->i_files + ci->i_subdirs);
180 static size_t ceph_vxattrcb_dir_files(struct ceph_inode_info *ci, char *val,
183 return snprintf(val, size, "%lld", ci->i_files);
186 static size_t ceph_vxattrcb_dir_subdirs(struct ceph_inode_info *ci, char *val,
189 return snprintf(val, size, "%lld", ci->i_subdirs);
192 static size_t ceph_vxattrcb_dir_rentries(struct ceph_inode_info *ci, char *val,
195 return snprintf(val, size, "%lld", ci->i_rfiles + ci->i_rsubdirs);
198 static size_t ceph_vxattrcb_dir_rfiles(struct ceph_inode_info *ci, char *val,
201 return snprintf(val, size, "%lld", ci->i_rfiles);
204 static size_t ceph_vxattrcb_dir_rsubdirs(struct ceph_inode_info *ci, char *val,
207 return snprintf(val, size, "%lld", ci->i_rsubdirs);
210 static size_t ceph_vxattrcb_dir_rbytes(struct ceph_inode_info *ci, char *val,
213 return snprintf(val, size, "%lld", ci->i_rbytes);
216 static size_t ceph_vxattrcb_dir_rctime(struct ceph_inode_info *ci, char *val,
219 return snprintf(val, size, "%ld.09%ld", (long)ci->i_rctime.tv_sec,
220 (long)ci->i_rctime.tv_nsec);
224 #define CEPH_XATTR_NAME(_type, _name) XATTR_CEPH_PREFIX #_type "." #_name
225 #define CEPH_XATTR_NAME2(_type, _name, _name2) \
226 XATTR_CEPH_PREFIX #_type "." #_name "." #_name2
228 #define XATTR_NAME_CEPH(_type, _name) \
230 .name = CEPH_XATTR_NAME(_type, _name), \
231 .name_size = sizeof (CEPH_XATTR_NAME(_type, _name)), \
232 .getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name, \
237 #define XATTR_LAYOUT_FIELD(_type, _name, _field) \
239 .name = CEPH_XATTR_NAME2(_type, _name, _field), \
240 .name_size = sizeof (CEPH_XATTR_NAME2(_type, _name, _field)), \
241 .getxattr_cb = ceph_vxattrcb_ ## _name ## _ ## _field, \
244 .exists_cb = ceph_vxattrcb_layout_exists, \
247 static struct ceph_vxattr ceph_dir_vxattrs[] = {
249 .name = "ceph.dir.layout",
250 .name_size = sizeof("ceph.dir.layout"),
251 .getxattr_cb = ceph_vxattrcb_layout,
254 .exists_cb = ceph_vxattrcb_layout_exists,
256 XATTR_LAYOUT_FIELD(dir, layout, stripe_unit),
257 XATTR_LAYOUT_FIELD(dir, layout, stripe_count),
258 XATTR_LAYOUT_FIELD(dir, layout, object_size),
259 XATTR_LAYOUT_FIELD(dir, layout, pool),
260 XATTR_LAYOUT_FIELD(dir, layout, pool_namespace),
261 XATTR_NAME_CEPH(dir, entries),
262 XATTR_NAME_CEPH(dir, files),
263 XATTR_NAME_CEPH(dir, subdirs),
264 XATTR_NAME_CEPH(dir, rentries),
265 XATTR_NAME_CEPH(dir, rfiles),
266 XATTR_NAME_CEPH(dir, rsubdirs),
267 XATTR_NAME_CEPH(dir, rbytes),
268 XATTR_NAME_CEPH(dir, rctime),
269 { .name = NULL, 0 } /* Required table terminator */
271 static size_t ceph_dir_vxattrs_name_size; /* total size of all names */
275 static struct ceph_vxattr ceph_file_vxattrs[] = {
277 .name = "ceph.file.layout",
278 .name_size = sizeof("ceph.file.layout"),
279 .getxattr_cb = ceph_vxattrcb_layout,
282 .exists_cb = ceph_vxattrcb_layout_exists,
284 XATTR_LAYOUT_FIELD(file, layout, stripe_unit),
285 XATTR_LAYOUT_FIELD(file, layout, stripe_count),
286 XATTR_LAYOUT_FIELD(file, layout, object_size),
287 XATTR_LAYOUT_FIELD(file, layout, pool),
288 XATTR_LAYOUT_FIELD(file, layout, pool_namespace),
289 { .name = NULL, 0 } /* Required table terminator */
291 static size_t ceph_file_vxattrs_name_size; /* total size of all names */
293 static struct ceph_vxattr *ceph_inode_vxattrs(struct inode *inode)
295 if (S_ISDIR(inode->i_mode))
296 return ceph_dir_vxattrs;
297 else if (S_ISREG(inode->i_mode))
298 return ceph_file_vxattrs;
302 static size_t ceph_vxattrs_name_size(struct ceph_vxattr *vxattrs)
304 if (vxattrs == ceph_dir_vxattrs)
305 return ceph_dir_vxattrs_name_size;
306 if (vxattrs == ceph_file_vxattrs)
307 return ceph_file_vxattrs_name_size;
313 * Compute the aggregate size (including terminating '\0') of all
314 * virtual extended attribute names in the given vxattr table.
316 static size_t __init vxattrs_name_size(struct ceph_vxattr *vxattrs)
318 struct ceph_vxattr *vxattr;
321 for (vxattr = vxattrs; vxattr->name; vxattr++)
323 size += vxattr->name_size;
328 /* Routines called at initialization and exit time */
330 void __init ceph_xattr_init(void)
332 ceph_dir_vxattrs_name_size = vxattrs_name_size(ceph_dir_vxattrs);
333 ceph_file_vxattrs_name_size = vxattrs_name_size(ceph_file_vxattrs);
336 void ceph_xattr_exit(void)
338 ceph_dir_vxattrs_name_size = 0;
339 ceph_file_vxattrs_name_size = 0;
342 static struct ceph_vxattr *ceph_match_vxattr(struct inode *inode,
345 struct ceph_vxattr *vxattr = ceph_inode_vxattrs(inode);
348 while (vxattr->name) {
349 if (!strcmp(vxattr->name, name))
358 static int __set_xattr(struct ceph_inode_info *ci,
359 const char *name, int name_len,
360 const char *val, int val_len,
361 int flags, int update_xattr,
362 struct ceph_inode_xattr **newxattr)
365 struct rb_node *parent = NULL;
366 struct ceph_inode_xattr *xattr = NULL;
370 p = &ci->i_xattrs.index.rb_node;
373 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
374 c = strncmp(name, xattr->name, min(name_len, xattr->name_len));
380 if (name_len == xattr->name_len)
382 else if (name_len < xattr->name_len)
393 if (xattr && (flags & XATTR_CREATE))
395 else if (!xattr && (flags & XATTR_REPLACE))
403 if (update_xattr < 0) {
405 __remove_xattr(ci, xattr);
416 xattr->name_len = name_len;
417 xattr->should_free_name = update_xattr;
419 ci->i_xattrs.count++;
420 dout("__set_xattr count=%d\n", ci->i_xattrs.count);
424 if (xattr->should_free_val)
425 kfree((void *)xattr->val);
431 ci->i_xattrs.names_size -= xattr->name_len;
432 ci->i_xattrs.vals_size -= xattr->val_len;
434 ci->i_xattrs.names_size += name_len;
435 ci->i_xattrs.vals_size += val_len;
441 xattr->val_len = val_len;
442 xattr->dirty = update_xattr;
443 xattr->should_free_val = (val && update_xattr);
446 rb_link_node(&xattr->node, parent, p);
447 rb_insert_color(&xattr->node, &ci->i_xattrs.index);
448 dout("__set_xattr_val p=%p\n", p);
451 dout("__set_xattr_val added %llx.%llx xattr %p %s=%.*s\n",
452 ceph_vinop(&ci->vfs_inode), xattr, name, val_len, val);
457 static struct ceph_inode_xattr *__get_xattr(struct ceph_inode_info *ci,
461 struct rb_node *parent = NULL;
462 struct ceph_inode_xattr *xattr = NULL;
463 int name_len = strlen(name);
466 p = &ci->i_xattrs.index.rb_node;
469 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
470 c = strncmp(name, xattr->name, xattr->name_len);
471 if (c == 0 && name_len > xattr->name_len)
478 dout("__get_xattr %s: found %.*s\n", name,
479 xattr->val_len, xattr->val);
484 dout("__get_xattr %s: not found\n", name);
489 static void __free_xattr(struct ceph_inode_xattr *xattr)
493 if (xattr->should_free_name)
494 kfree((void *)xattr->name);
495 if (xattr->should_free_val)
496 kfree((void *)xattr->val);
501 static int __remove_xattr(struct ceph_inode_info *ci,
502 struct ceph_inode_xattr *xattr)
507 rb_erase(&xattr->node, &ci->i_xattrs.index);
509 if (xattr->should_free_name)
510 kfree((void *)xattr->name);
511 if (xattr->should_free_val)
512 kfree((void *)xattr->val);
514 ci->i_xattrs.names_size -= xattr->name_len;
515 ci->i_xattrs.vals_size -= xattr->val_len;
516 ci->i_xattrs.count--;
522 static char *__copy_xattr_names(struct ceph_inode_info *ci,
526 struct ceph_inode_xattr *xattr = NULL;
528 p = rb_first(&ci->i_xattrs.index);
529 dout("__copy_xattr_names count=%d\n", ci->i_xattrs.count);
532 xattr = rb_entry(p, struct ceph_inode_xattr, node);
533 memcpy(dest, xattr->name, xattr->name_len);
534 dest[xattr->name_len] = '\0';
536 dout("dest=%s %p (%s) (%d/%d)\n", dest, xattr, xattr->name,
537 xattr->name_len, ci->i_xattrs.names_size);
539 dest += xattr->name_len + 1;
546 void __ceph_destroy_xattrs(struct ceph_inode_info *ci)
548 struct rb_node *p, *tmp;
549 struct ceph_inode_xattr *xattr = NULL;
551 p = rb_first(&ci->i_xattrs.index);
553 dout("__ceph_destroy_xattrs p=%p\n", p);
556 xattr = rb_entry(p, struct ceph_inode_xattr, node);
559 dout("__ceph_destroy_xattrs next p=%p (%.*s)\n", p,
560 xattr->name_len, xattr->name);
561 rb_erase(tmp, &ci->i_xattrs.index);
566 ci->i_xattrs.names_size = 0;
567 ci->i_xattrs.vals_size = 0;
568 ci->i_xattrs.index_version = 0;
569 ci->i_xattrs.count = 0;
570 ci->i_xattrs.index = RB_ROOT;
573 static int __build_xattrs(struct inode *inode)
574 __releases(ci->i_ceph_lock)
575 __acquires(ci->i_ceph_lock)
581 const char *name, *val;
582 struct ceph_inode_info *ci = ceph_inode(inode);
584 struct ceph_inode_xattr **xattrs = NULL;
588 dout("__build_xattrs() len=%d\n",
589 ci->i_xattrs.blob ? (int)ci->i_xattrs.blob->vec.iov_len : 0);
591 if (ci->i_xattrs.index_version >= ci->i_xattrs.version)
592 return 0; /* already built */
594 __ceph_destroy_xattrs(ci);
597 /* updated internal xattr rb tree */
598 if (ci->i_xattrs.blob && ci->i_xattrs.blob->vec.iov_len > 4) {
599 p = ci->i_xattrs.blob->vec.iov_base;
600 end = p + ci->i_xattrs.blob->vec.iov_len;
601 ceph_decode_32_safe(&p, end, numattr, bad);
602 xattr_version = ci->i_xattrs.version;
603 spin_unlock(&ci->i_ceph_lock);
605 xattrs = kcalloc(numattr, sizeof(struct ceph_inode_xattr *),
611 for (i = 0; i < numattr; i++) {
612 xattrs[i] = kmalloc(sizeof(struct ceph_inode_xattr),
618 spin_lock(&ci->i_ceph_lock);
619 if (ci->i_xattrs.version != xattr_version) {
620 /* lost a race, retry */
621 for (i = 0; i < numattr; i++)
629 ceph_decode_32_safe(&p, end, len, bad);
633 ceph_decode_32_safe(&p, end, len, bad);
637 err = __set_xattr(ci, name, namelen, val, len,
638 0, 0, &xattrs[numattr]);
645 ci->i_xattrs.index_version = ci->i_xattrs.version;
646 ci->i_xattrs.dirty = false;
650 spin_lock(&ci->i_ceph_lock);
653 for (i = 0; i < numattr; i++)
657 ci->i_xattrs.names_size = 0;
661 static int __get_required_blob_size(struct ceph_inode_info *ci, int name_size,
665 * 4 bytes for the length, and additional 4 bytes per each xattr name,
666 * 4 bytes per each value
668 int size = 4 + ci->i_xattrs.count*(4 + 4) +
669 ci->i_xattrs.names_size +
670 ci->i_xattrs.vals_size;
671 dout("__get_required_blob_size c=%d names.size=%d vals.size=%d\n",
672 ci->i_xattrs.count, ci->i_xattrs.names_size,
673 ci->i_xattrs.vals_size);
676 size += 4 + 4 + name_size + val_size;
682 * If there are dirty xattrs, reencode xattrs into the prealloc_blob
683 * and swap into place.
685 void __ceph_build_xattrs_blob(struct ceph_inode_info *ci)
688 struct ceph_inode_xattr *xattr = NULL;
691 dout("__build_xattrs_blob %p\n", &ci->vfs_inode);
692 if (ci->i_xattrs.dirty) {
693 int need = __get_required_blob_size(ci, 0, 0);
695 BUG_ON(need > ci->i_xattrs.prealloc_blob->alloc_len);
697 p = rb_first(&ci->i_xattrs.index);
698 dest = ci->i_xattrs.prealloc_blob->vec.iov_base;
700 ceph_encode_32(&dest, ci->i_xattrs.count);
702 xattr = rb_entry(p, struct ceph_inode_xattr, node);
704 ceph_encode_32(&dest, xattr->name_len);
705 memcpy(dest, xattr->name, xattr->name_len);
706 dest += xattr->name_len;
707 ceph_encode_32(&dest, xattr->val_len);
708 memcpy(dest, xattr->val, xattr->val_len);
709 dest += xattr->val_len;
714 /* adjust buffer len; it may be larger than we need */
715 ci->i_xattrs.prealloc_blob->vec.iov_len =
716 dest - ci->i_xattrs.prealloc_blob->vec.iov_base;
718 if (ci->i_xattrs.blob)
719 ceph_buffer_put(ci->i_xattrs.blob);
720 ci->i_xattrs.blob = ci->i_xattrs.prealloc_blob;
721 ci->i_xattrs.prealloc_blob = NULL;
722 ci->i_xattrs.dirty = false;
723 ci->i_xattrs.version++;
727 static inline int __get_request_mask(struct inode *in) {
728 struct ceph_mds_request *req = current->journal_info;
730 if (req && req->r_target_inode == in) {
731 if (req->r_op == CEPH_MDS_OP_LOOKUP ||
732 req->r_op == CEPH_MDS_OP_LOOKUPINO ||
733 req->r_op == CEPH_MDS_OP_LOOKUPPARENT ||
734 req->r_op == CEPH_MDS_OP_GETATTR) {
735 mask = le32_to_cpu(req->r_args.getattr.mask);
736 } else if (req->r_op == CEPH_MDS_OP_OPEN ||
737 req->r_op == CEPH_MDS_OP_CREATE) {
738 mask = le32_to_cpu(req->r_args.open.mask);
744 ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,
747 struct ceph_inode_info *ci = ceph_inode(inode);
748 struct ceph_inode_xattr *xattr;
749 struct ceph_vxattr *vxattr = NULL;
753 /* let's see if a virtual xattr was requested */
754 vxattr = ceph_match_vxattr(inode, name);
757 if (!(vxattr->exists_cb && !vxattr->exists_cb(ci))) {
758 err = vxattr->getxattr_cb(ci, value, size);
759 if (size && size < err)
765 req_mask = __get_request_mask(inode);
767 spin_lock(&ci->i_ceph_lock);
768 dout("getxattr %p ver=%lld index_ver=%lld\n", inode,
769 ci->i_xattrs.version, ci->i_xattrs.index_version);
771 if (ci->i_xattrs.version == 0 ||
772 !((req_mask & CEPH_CAP_XATTR_SHARED) ||
773 __ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1))) {
774 spin_unlock(&ci->i_ceph_lock);
776 /* security module gets xattr while filling trace */
777 if (current->journal_info != NULL) {
778 pr_warn_ratelimited("sync getxattr %p "
779 "during filling trace\n", inode);
783 /* get xattrs from mds (if we don't already have them) */
784 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
787 spin_lock(&ci->i_ceph_lock);
790 err = __build_xattrs(inode);
794 err = -ENODATA; /* == ENOATTR */
795 xattr = __get_xattr(ci, name);
800 if (size && size < xattr->val_len)
803 err = xattr->val_len;
807 memcpy(value, xattr->val, xattr->val_len);
809 if (current->journal_info != NULL &&
810 !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN))
811 ci->i_ceph_flags |= CEPH_I_SEC_INITED;
813 spin_unlock(&ci->i_ceph_lock);
817 ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)
819 struct inode *inode = d_inode(dentry);
820 struct ceph_inode_info *ci = ceph_inode(inode);
821 struct ceph_vxattr *vxattrs = ceph_inode_vxattrs(inode);
828 spin_lock(&ci->i_ceph_lock);
829 dout("listxattr %p ver=%lld index_ver=%lld\n", inode,
830 ci->i_xattrs.version, ci->i_xattrs.index_version);
832 if (ci->i_xattrs.version == 0 ||
833 !__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1)) {
834 spin_unlock(&ci->i_ceph_lock);
835 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
838 spin_lock(&ci->i_ceph_lock);
841 err = __build_xattrs(inode);
845 * Start with virtual dir xattr names (if any) (including
846 * terminating '\0' characters for each).
848 vir_namelen = ceph_vxattrs_name_size(vxattrs);
850 /* adding 1 byte per each variable due to the null termination */
851 namelen = ci->i_xattrs.names_size + ci->i_xattrs.count;
853 if (size && vir_namelen + namelen > size)
856 err = namelen + vir_namelen;
860 names = __copy_xattr_names(ci, names);
862 /* virtual xattr names, too */
865 for (i = 0; vxattrs[i].name; i++) {
866 if (!vxattrs[i].hidden &&
867 !(vxattrs[i].exists_cb &&
868 !vxattrs[i].exists_cb(ci))) {
869 len = sprintf(names, "%s", vxattrs[i].name);
877 spin_unlock(&ci->i_ceph_lock);
881 static int ceph_sync_setxattr(struct inode *inode, const char *name,
882 const char *value, size_t size, int flags)
884 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
885 struct ceph_inode_info *ci = ceph_inode(inode);
886 struct ceph_mds_request *req;
887 struct ceph_mds_client *mdsc = fsc->mdsc;
888 struct ceph_pagelist *pagelist = NULL;
889 int op = CEPH_MDS_OP_SETXATTR;
893 /* copy value into pagelist */
894 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
898 ceph_pagelist_init(pagelist);
899 err = ceph_pagelist_append(pagelist, value, size);
903 if (flags & CEPH_XATTR_REPLACE)
904 op = CEPH_MDS_OP_RMXATTR;
906 flags |= CEPH_XATTR_REMOVE;
909 dout("setxattr value=%.*s\n", (int)size, value);
912 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
918 req->r_path2 = kstrdup(name, GFP_NOFS);
920 ceph_mdsc_put_request(req);
925 if (op == CEPH_MDS_OP_SETXATTR) {
926 req->r_args.setxattr.flags = cpu_to_le32(flags);
927 req->r_pagelist = pagelist;
931 req->r_inode = inode;
934 req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
936 dout("xattr.ver (before): %lld\n", ci->i_xattrs.version);
937 err = ceph_mdsc_do_request(mdsc, NULL, req);
938 ceph_mdsc_put_request(req);
939 dout("xattr.ver (after): %lld\n", ci->i_xattrs.version);
943 ceph_pagelist_release(pagelist);
947 int __ceph_setxattr(struct inode *inode, const char *name,
948 const void *value, size_t size, int flags)
950 struct ceph_vxattr *vxattr;
951 struct ceph_inode_info *ci = ceph_inode(inode);
952 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
953 struct ceph_cap_flush *prealloc_cf = NULL;
954 struct ceph_buffer *old_blob = NULL;
958 int name_len = strlen(name);
960 char *newname = NULL;
962 struct ceph_inode_xattr *xattr = NULL;
963 int required_blob_size;
964 bool lock_snap_rwsem = false;
966 if (ceph_snap(inode) != CEPH_NOSNAP)
969 vxattr = ceph_match_vxattr(inode, name);
970 if (vxattr && vxattr->readonly)
973 /* pass any unhandled ceph.* xattrs through to the MDS */
974 if (!strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
975 goto do_sync_unlocked;
977 /* preallocate memory for xattr name, value, index node */
979 newname = kmemdup(name, name_len + 1, GFP_NOFS);
984 newval = kmemdup(value, val_len, GFP_NOFS);
989 xattr = kmalloc(sizeof(struct ceph_inode_xattr), GFP_NOFS);
993 prealloc_cf = ceph_alloc_cap_flush();
997 spin_lock(&ci->i_ceph_lock);
999 issued = __ceph_caps_issued(ci, NULL);
1000 if (ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))
1003 if (!lock_snap_rwsem && !ci->i_head_snapc) {
1004 lock_snap_rwsem = true;
1005 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1006 spin_unlock(&ci->i_ceph_lock);
1007 down_read(&mdsc->snap_rwsem);
1008 spin_lock(&ci->i_ceph_lock);
1013 dout("setxattr %p issued %s\n", inode, ceph_cap_string(issued));
1014 __build_xattrs(inode);
1016 required_blob_size = __get_required_blob_size(ci, name_len, val_len);
1018 if (!ci->i_xattrs.prealloc_blob ||
1019 required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
1020 struct ceph_buffer *blob;
1022 spin_unlock(&ci->i_ceph_lock);
1023 ceph_buffer_put(old_blob); /* Shouldn't be required */
1024 dout(" pre-allocating new blob size=%d\n", required_blob_size);
1025 blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
1027 goto do_sync_unlocked;
1028 spin_lock(&ci->i_ceph_lock);
1029 /* prealloc_blob can't be released while holding i_ceph_lock */
1030 if (ci->i_xattrs.prealloc_blob)
1031 old_blob = ci->i_xattrs.prealloc_blob;
1032 ci->i_xattrs.prealloc_blob = blob;
1036 err = __set_xattr(ci, newname, name_len, newval, val_len,
1037 flags, value ? 1 : -1, &xattr);
1040 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL,
1042 ci->i_xattrs.dirty = true;
1043 inode->i_ctime = current_time(inode);
1046 spin_unlock(&ci->i_ceph_lock);
1047 ceph_buffer_put(old_blob);
1048 if (lock_snap_rwsem)
1049 up_read(&mdsc->snap_rwsem);
1051 __mark_inode_dirty(inode, dirty);
1052 ceph_free_cap_flush(prealloc_cf);
1056 spin_unlock(&ci->i_ceph_lock);
1058 if (lock_snap_rwsem)
1059 up_read(&mdsc->snap_rwsem);
1061 /* security module set xattr while filling trace */
1062 if (current->journal_info != NULL) {
1063 pr_warn_ratelimited("sync setxattr %p "
1064 "during filling trace\n", inode);
1067 err = ceph_sync_setxattr(inode, name, value, size, flags);
1070 ceph_free_cap_flush(prealloc_cf);
1077 static int ceph_get_xattr_handler(const struct xattr_handler *handler,
1078 struct dentry *dentry, struct inode *inode,
1079 const char *name, void *value, size_t size)
1081 if (!ceph_is_valid_xattr(name))
1083 return __ceph_getxattr(inode, name, value, size);
1086 static int ceph_set_xattr_handler(const struct xattr_handler *handler,
1087 struct dentry *unused, struct inode *inode,
1088 const char *name, const void *value,
1089 size_t size, int flags)
1091 if (!ceph_is_valid_xattr(name))
1093 return __ceph_setxattr(inode, name, value, size, flags);
1096 static const struct xattr_handler ceph_other_xattr_handler = {
1097 .prefix = "", /* match any name => handlers called with full name */
1098 .get = ceph_get_xattr_handler,
1099 .set = ceph_set_xattr_handler,
1102 #ifdef CONFIG_SECURITY
1103 bool ceph_security_xattr_wanted(struct inode *in)
1105 return in->i_security != NULL;
1108 bool ceph_security_xattr_deadlock(struct inode *in)
1110 struct ceph_inode_info *ci;
1112 if (in->i_security == NULL)
1114 ci = ceph_inode(in);
1115 spin_lock(&ci->i_ceph_lock);
1116 ret = !(ci->i_ceph_flags & CEPH_I_SEC_INITED) &&
1117 !(ci->i_xattrs.version > 0 &&
1118 __ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 0));
1119 spin_unlock(&ci->i_ceph_lock);