1 #include <linux/ceph/ceph_debug.h>
4 #include "mds_client.h"
6 #include <linux/ceph/decode.h>
8 #include <linux/xattr.h>
9 #include <linux/slab.h>
11 static bool ceph_is_valid_xattr(const char *name)
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);
21 * These define virtual xattrs exposing the recursive directory
22 * statistics and layout metadata.
24 struct ceph_vxattr_cb {
27 size_t (*getxattr_cb)(struct ceph_inode_info *ci, char *val,
33 static size_t ceph_vxattrcb_entries(struct ceph_inode_info *ci, char *val,
36 return snprintf(val, size, "%lld", ci->i_files + ci->i_subdirs);
39 static size_t ceph_vxattrcb_files(struct ceph_inode_info *ci, char *val,
42 return snprintf(val, size, "%lld", ci->i_files);
45 static size_t ceph_vxattrcb_subdirs(struct ceph_inode_info *ci, char *val,
48 return snprintf(val, size, "%lld", ci->i_subdirs);
51 static size_t ceph_vxattrcb_rentries(struct ceph_inode_info *ci, char *val,
54 return snprintf(val, size, "%lld", ci->i_rfiles + ci->i_rsubdirs);
57 static size_t ceph_vxattrcb_rfiles(struct ceph_inode_info *ci, char *val,
60 return snprintf(val, size, "%lld", ci->i_rfiles);
63 static size_t ceph_vxattrcb_rsubdirs(struct ceph_inode_info *ci, char *val,
66 return snprintf(val, size, "%lld", ci->i_rsubdirs);
69 static size_t ceph_vxattrcb_rbytes(struct ceph_inode_info *ci, char *val,
72 return snprintf(val, size, "%lld", ci->i_rbytes);
75 static size_t ceph_vxattrcb_rctime(struct ceph_inode_info *ci, char *val,
78 return snprintf(val, size, "%ld.%ld", (long)ci->i_rctime.tv_sec,
79 (long)ci->i_rctime.tv_nsec);
82 static 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},
96 static size_t ceph_vxattrcb_layout(struct ceph_inode_info *ci, char *val,
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(
113 static struct ceph_vxattr_cb ceph_file_vxattrs[] = {
114 { true, "ceph.layout", ceph_vxattrcb_layout},
118 static struct ceph_vxattr_cb *ceph_inode_vxattrs(struct inode *inode)
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;
127 static struct ceph_vxattr_cb *ceph_match_vxattr(struct ceph_vxattr_cb *vxattr,
131 if (strcmp(vxattr->name, name) == 0)
134 } while (vxattr->name);
138 static int __set_xattr(struct ceph_inode_info *ci,
139 const char *name, int name_len,
140 const char *val, int val_len,
142 int should_free_name, int should_free_val,
143 struct ceph_inode_xattr **newxattr)
146 struct rb_node *parent = NULL;
147 struct ceph_inode_xattr *xattr = NULL;
151 p = &ci->i_xattrs.index.rb_node;
154 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
155 c = strncmp(name, xattr->name, min(name_len, xattr->name_len));
161 if (name_len == xattr->name_len)
163 else if (name_len < xattr->name_len)
175 xattr->name_len = name_len;
176 xattr->should_free_name = should_free_name;
178 ci->i_xattrs.count++;
179 dout("__set_xattr count=%d\n", ci->i_xattrs.count);
183 if (xattr->should_free_val)
184 kfree((void *)xattr->val);
186 if (should_free_name) {
190 ci->i_xattrs.names_size -= xattr->name_len;
191 ci->i_xattrs.vals_size -= xattr->val_len;
193 ci->i_xattrs.names_size += name_len;
194 ci->i_xattrs.vals_size += val_len;
200 xattr->val_len = val_len;
201 xattr->dirty = dirty;
202 xattr->should_free_val = (val && should_free_val);
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);
210 dout("__set_xattr_val added %llx.%llx xattr %p %s=%.*s\n",
211 ceph_vinop(&ci->vfs_inode), xattr, name, val_len, val);
216 static struct ceph_inode_xattr *__get_xattr(struct ceph_inode_info *ci,
220 struct rb_node *parent = NULL;
221 struct ceph_inode_xattr *xattr = NULL;
222 int name_len = strlen(name);
225 p = &ci->i_xattrs.index.rb_node;
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)
237 dout("__get_xattr %s: found %.*s\n", name,
238 xattr->val_len, xattr->val);
243 dout("__get_xattr %s: not found\n", name);
248 static void __free_xattr(struct ceph_inode_xattr *xattr)
252 if (xattr->should_free_name)
253 kfree((void *)xattr->name);
254 if (xattr->should_free_val)
255 kfree((void *)xattr->val);
260 static int __remove_xattr(struct ceph_inode_info *ci,
261 struct ceph_inode_xattr *xattr)
266 rb_erase(&xattr->node, &ci->i_xattrs.index);
268 if (xattr->should_free_name)
269 kfree((void *)xattr->name);
270 if (xattr->should_free_val)
271 kfree((void *)xattr->val);
273 ci->i_xattrs.names_size -= xattr->name_len;
274 ci->i_xattrs.vals_size -= xattr->val_len;
275 ci->i_xattrs.count--;
281 static int __remove_xattr_by_name(struct ceph_inode_info *ci,
285 struct ceph_inode_xattr *xattr;
288 p = &ci->i_xattrs.index.rb_node;
289 xattr = __get_xattr(ci, name);
290 err = __remove_xattr(ci, xattr);
294 static char *__copy_xattr_names(struct ceph_inode_info *ci,
298 struct ceph_inode_xattr *xattr = NULL;
300 p = rb_first(&ci->i_xattrs.index);
301 dout("__copy_xattr_names count=%d\n", ci->i_xattrs.count);
304 xattr = rb_entry(p, struct ceph_inode_xattr, node);
305 memcpy(dest, xattr->name, xattr->name_len);
306 dest[xattr->name_len] = '\0';
308 dout("dest=%s %p (%s) (%d/%d)\n", dest, xattr, xattr->name,
309 xattr->name_len, ci->i_xattrs.names_size);
311 dest += xattr->name_len + 1;
318 void __ceph_destroy_xattrs(struct ceph_inode_info *ci)
320 struct rb_node *p, *tmp;
321 struct ceph_inode_xattr *xattr = NULL;
323 p = rb_first(&ci->i_xattrs.index);
325 dout("__ceph_destroy_xattrs p=%p\n", p);
328 xattr = rb_entry(p, struct ceph_inode_xattr, node);
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);
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;
345 static int __build_xattrs(struct inode *inode)
346 __releases(ci->i_ceph_lock)
347 __acquires(ci->i_ceph_lock)
353 const char *name, *val;
354 struct ceph_inode_info *ci = ceph_inode(inode);
356 struct ceph_inode_xattr **xattrs = NULL;
360 dout("__build_xattrs() len=%d\n",
361 ci->i_xattrs.blob ? (int)ci->i_xattrs.blob->vec.iov_len : 0);
363 if (ci->i_xattrs.index_version >= ci->i_xattrs.version)
364 return 0; /* already built */
366 __ceph_destroy_xattrs(ci);
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(&ci->i_ceph_lock);
377 xattrs = kcalloc(numattr, sizeof(struct ceph_xattr *),
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),
390 spin_lock(&ci->i_ceph_lock);
391 if (ci->i_xattrs.version != xattr_version) {
392 /* lost a race, retry */
393 for (i = 0; i < numattr; i++)
400 ceph_decode_32_safe(&p, end, len, bad);
404 ceph_decode_32_safe(&p, end, len, bad);
408 err = __set_xattr(ci, name, namelen, val, len,
409 0, 0, 0, &xattrs[numattr]);
416 ci->i_xattrs.index_version = ci->i_xattrs.version;
417 ci->i_xattrs.dirty = false;
421 spin_lock(&ci->i_ceph_lock);
424 for (i = 0; i < numattr; i++)
428 ci->i_xattrs.names_size = 0;
432 static int __get_required_blob_size(struct ceph_inode_info *ci, int name_size,
436 * 4 bytes for the length, and additional 4 bytes per each xattr name,
437 * 4 bytes per each value
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);
447 size += 4 + 4 + name_size + val_size;
453 * If there are dirty xattrs, reencode xattrs into the prealloc_blob
454 * and swap into place.
456 void __ceph_build_xattrs_blob(struct ceph_inode_info *ci)
459 struct ceph_inode_xattr *xattr = NULL;
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);
466 BUG_ON(need > ci->i_xattrs.prealloc_blob->alloc_len);
468 p = rb_first(&ci->i_xattrs.index);
469 dest = ci->i_xattrs.prealloc_blob->vec.iov_base;
471 ceph_encode_32(&dest, ci->i_xattrs.count);
473 xattr = rb_entry(p, struct ceph_inode_xattr, node);
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;
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;
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++;
498 ssize_t ceph_getxattr(struct dentry *dentry, const char *name, void *value,
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);
505 struct ceph_inode_xattr *xattr;
506 struct ceph_vxattr_cb *vxattr = NULL;
508 if (!ceph_is_valid_xattr(name))
511 /* let's see if a virtual xattr was requested */
513 vxattr = ceph_match_vxattr(vxattrs, name);
515 spin_lock(&ci->i_ceph_lock);
516 dout("getxattr %p ver=%lld index_ver=%lld\n", inode,
517 ci->i_xattrs.version, ci->i_xattrs.index_version);
519 if (__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1) &&
520 (ci->i_xattrs.index_version >= ci->i_xattrs.version)) {
523 spin_unlock(&ci->i_ceph_lock);
524 /* get xattrs from mds (if we don't already have them) */
525 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR);
530 spin_lock(&ci->i_ceph_lock);
532 if (vxattr && vxattr->readonly) {
533 err = vxattr->getxattr_cb(ci, value, size);
537 err = __build_xattrs(inode);
542 err = -ENODATA; /* == ENOATTR */
543 xattr = __get_xattr(ci, name);
546 err = vxattr->getxattr_cb(ci, value, size);
551 if (size && size < xattr->val_len)
554 err = xattr->val_len;
558 memcpy(value, xattr->val, xattr->val_len);
561 spin_unlock(&ci->i_ceph_lock);
565 ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)
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);
576 spin_lock(&ci->i_ceph_lock);
577 dout("listxattr %p ver=%lld index_ver=%lld\n", inode,
578 ci->i_xattrs.version, ci->i_xattrs.index_version);
580 if (__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1) &&
581 (ci->i_xattrs.index_version >= ci->i_xattrs.version)) {
584 spin_unlock(&ci->i_ceph_lock);
585 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR);
590 spin_lock(&ci->i_ceph_lock);
592 err = __build_xattrs(inode);
598 /* include virtual dir xattrs */
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;
605 if (size && namelen > size)
612 names = __copy_xattr_names(ci, names);
614 /* virtual xattr names, too */
616 for (i = 0; vxattrs[i].name; i++) {
617 len = sprintf(names, "%s", vxattrs[i].name);
622 spin_unlock(&ci->i_ceph_lock);
626 static int ceph_sync_setxattr(struct dentry *dentry, const char *name,
627 const char *value, size_t size, int flags)
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;
637 struct page **pages = NULL;
640 /* copy value into some pages */
641 nr_pages = calc_pages_for(0, size);
643 pages = kmalloc(sizeof(pages[0])*nr_pages, GFP_NOFS);
647 for (i = 0; i < nr_pages; i++) {
648 pages[i] = __page_cache_alloc(GFP_NOFS);
653 kaddr = kmap(pages[i]);
654 memcpy(kaddr, value + i*PAGE_CACHE_SIZE,
655 min(PAGE_CACHE_SIZE, size-i*PAGE_CACHE_SIZE));
659 dout("setxattr value=%.*s\n", (int)size, value);
662 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETXATTR,
668 req->r_inode = inode;
670 req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
672 req->r_args.setxattr.flags = cpu_to_le32(flags);
673 req->r_path2 = kstrdup(name, GFP_NOFS);
675 req->r_pages = pages;
676 req->r_num_pages = nr_pages;
677 req->r_data_len = size;
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);
683 ceph_mdsc_put_request(req);
684 dout("xattr.ver (after): %lld\n", ci->i_xattrs.version);
688 for (i = 0; i < nr_pages; i++)
689 __free_page(pages[i]);
695 int ceph_setxattr(struct dentry *dentry, const char *name,
696 const void *value, size_t size, int flags)
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);
702 int name_len = strlen(name);
704 char *newname = NULL;
706 struct ceph_inode_xattr *xattr = NULL;
708 int required_blob_size;
711 if (ceph_snap(inode) != CEPH_NOSNAP)
714 if (!ceph_is_valid_xattr(name))
718 struct ceph_vxattr_cb *vxattr =
719 ceph_match_vxattr(vxattrs, name);
720 if (vxattr && vxattr->readonly)
724 /* preallocate memory for xattr name, value, index node */
726 newname = kmemdup(name, name_len + 1, GFP_NOFS);
731 newval = kmalloc(val_len + 1, GFP_NOFS);
734 memcpy(newval, value, val_len);
735 newval[val_len] = '\0';
738 xattr = kmalloc(sizeof(struct ceph_inode_xattr), GFP_NOFS);
742 spin_lock(&ci->i_ceph_lock);
744 issued = __ceph_caps_issued(ci, NULL);
745 if (!(issued & CEPH_CAP_XATTR_EXCL))
747 __build_xattrs(inode);
749 required_blob_size = __get_required_blob_size(ci, name_len, val_len);
751 if (!ci->i_xattrs.prealloc_blob ||
752 required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
753 struct ceph_buffer *blob = NULL;
755 spin_unlock(&ci->i_ceph_lock);
756 dout(" preaallocating new blob size=%d\n", required_blob_size);
757 blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
760 spin_lock(&ci->i_ceph_lock);
761 if (ci->i_xattrs.prealloc_blob)
762 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
763 ci->i_xattrs.prealloc_blob = blob;
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(&ci->i_ceph_lock);
775 __mark_inode_dirty(inode, dirty);
779 spin_unlock(&ci->i_ceph_lock);
780 err = ceph_sync_setxattr(dentry, name, value, size, flags);
788 static int ceph_send_removexattr(struct dentry *dentry, const char *name)
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;
797 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RMXATTR,
801 req->r_inode = inode;
803 req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
805 req->r_path2 = kstrdup(name, GFP_NOFS);
807 parent_inode = ceph_get_dentry_parent_inode(dentry);
808 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
810 ceph_mdsc_put_request(req);
814 int ceph_removexattr(struct dentry *dentry, const char *name)
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);
823 if (ceph_snap(inode) != CEPH_NOSNAP)
826 if (!ceph_is_valid_xattr(name))
830 struct ceph_vxattr_cb *vxattr =
831 ceph_match_vxattr(vxattrs, name);
832 if (vxattr && vxattr->readonly)
836 spin_lock(&ci->i_ceph_lock);
837 __build_xattrs(inode);
838 issued = __ceph_caps_issued(ci, NULL);
839 dout("removexattr %p issued %s\n", inode, ceph_cap_string(issued));
841 if (!(issued & CEPH_CAP_XATTR_EXCL))
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;
849 spin_unlock(&ci->i_ceph_lock);
851 __mark_inode_dirty(inode, dirty);
854 spin_unlock(&ci->i_ceph_lock);
855 err = ceph_send_removexattr(dentry, name);