2 * Copyright 2012 Xyratex Technology Limited
4 * Copyright (c) 2013, 2015, Intel Corporation.
6 * Author: Andrew Perepechko <Andrew_Perepechko@xyratex.com>
10 #define DEBUG_SUBSYSTEM S_LLITE
13 #include <linux/sched.h>
15 #include "../include/obd_support.h"
16 #include "../include/lustre_dlm.h"
17 #include "../include/lustre_ver.h"
18 #include "llite_internal.h"
20 /* If we ever have hundreds of extended attributes, we might want to consider
21 * using a hash or a tree structure instead of list for faster lookups.
23 struct ll_xattr_entry {
24 struct list_head xe_list; /* protected with
25 * lli_xattrs_list_rwsem
27 char *xe_name; /* xattr name, \0-terminated */
28 char *xe_value; /* xattr value */
29 unsigned xe_namelen; /* strlen(xe_name) + 1 */
30 unsigned xe_vallen; /* xattr value length */
33 static struct kmem_cache *xattr_kmem;
34 static struct lu_kmem_descr xattr_caches[] = {
36 .ckd_cache = &xattr_kmem,
37 .ckd_name = "xattr_kmem",
38 .ckd_size = sizeof(struct ll_xattr_entry)
45 int ll_xattr_init(void)
47 return lu_kmem_init(xattr_caches);
50 void ll_xattr_fini(void)
52 lu_kmem_fini(xattr_caches);
56 * Initializes xattr cache for an inode.
58 * This initializes the xattr list and marks cache presence.
60 static void ll_xattr_cache_init(struct ll_inode_info *lli)
62 INIT_LIST_HEAD(&lli->lli_xattrs);
63 lli->lli_flags |= LLIF_XATTR_CACHE;
67 * This looks for a specific extended attribute.
69 * Find in @cache and return @xattr_name attribute in @xattr,
70 * for the NULL @xattr_name return the first cached @xattr.
73 * \retval -ENODATA if not found
75 static int ll_xattr_cache_find(struct list_head *cache,
76 const char *xattr_name,
77 struct ll_xattr_entry **xattr)
79 struct ll_xattr_entry *entry;
81 list_for_each_entry(entry, cache, xe_list) {
82 /* xattr_name == NULL means look for any entry */
83 if (!xattr_name || strcmp(xattr_name, entry->xe_name) == 0) {
85 CDEBUG(D_CACHE, "find: [%s]=%.*s\n",
86 entry->xe_name, entry->xe_vallen,
98 * Add @xattr_name attr with @xattr_val value and @xattr_val_len length,
101 * \retval -ENOMEM if no memory could be allocated for the cached attr
102 * \retval -EPROTO if duplicate xattr is being added
104 static int ll_xattr_cache_add(struct list_head *cache,
105 const char *xattr_name,
106 const char *xattr_val,
107 unsigned xattr_val_len)
109 struct ll_xattr_entry *xattr;
111 if (ll_xattr_cache_find(cache, xattr_name, &xattr) == 0) {
112 CDEBUG(D_CACHE, "duplicate xattr: [%s]\n", xattr_name);
116 xattr = kmem_cache_zalloc(xattr_kmem, GFP_NOFS);
118 CDEBUG(D_CACHE, "failed to allocate xattr\n");
122 xattr->xe_name = kstrdup(xattr_name, GFP_NOFS);
123 if (!xattr->xe_name) {
124 CDEBUG(D_CACHE, "failed to alloc xattr name %u\n",
128 xattr->xe_value = kmemdup(xattr_val, xattr_val_len, GFP_NOFS);
129 if (!xattr->xe_value)
132 xattr->xe_vallen = xattr_val_len;
133 list_add(&xattr->xe_list, cache);
135 CDEBUG(D_CACHE, "set: [%s]=%.*s\n", xattr_name, xattr_val_len,
140 kfree(xattr->xe_name);
142 kmem_cache_free(xattr_kmem, xattr);
148 * This removes an extended attribute from cache.
150 * Remove @xattr_name attribute from @cache.
153 * \retval -ENODATA if @xattr_name is not cached
155 static int ll_xattr_cache_del(struct list_head *cache,
156 const char *xattr_name)
158 struct ll_xattr_entry *xattr;
160 CDEBUG(D_CACHE, "del xattr: %s\n", xattr_name);
162 if (ll_xattr_cache_find(cache, xattr_name, &xattr) == 0) {
163 list_del(&xattr->xe_list);
164 kfree(xattr->xe_name);
165 kfree(xattr->xe_value);
166 kmem_cache_free(xattr_kmem, xattr);
175 * This iterates cached extended attributes.
177 * Walk over cached attributes in @cache and
178 * fill in @xld_buffer or only calculate buffer
179 * size if @xld_buffer is NULL.
181 * \retval >= 0 buffer list size
182 * \retval -ENODATA if the list cannot fit @xld_size buffer
184 static int ll_xattr_cache_list(struct list_head *cache,
188 struct ll_xattr_entry *xattr, *tmp;
191 list_for_each_entry_safe(xattr, tmp, cache, xe_list) {
192 CDEBUG(D_CACHE, "list: buffer=%p[%d] name=%s\n",
193 xld_buffer, xld_tail, xattr->xe_name);
196 xld_size -= xattr->xe_namelen;
199 memcpy(&xld_buffer[xld_tail],
200 xattr->xe_name, xattr->xe_namelen);
202 xld_tail += xattr->xe_namelen;
212 * Check if the xattr cache is initialized (filled).
214 * \retval 0 @cache is not initialized
215 * \retval 1 @cache is initialized
217 static int ll_xattr_cache_valid(struct ll_inode_info *lli)
219 return !!(lli->lli_flags & LLIF_XATTR_CACHE);
223 * This finalizes the xattr cache.
225 * Free all xattr memory. @lli is the inode info pointer.
227 * \retval 0 no error occurred
229 static int ll_xattr_cache_destroy_locked(struct ll_inode_info *lli)
231 if (!ll_xattr_cache_valid(lli))
234 while (ll_xattr_cache_del(&lli->lli_xattrs, NULL) == 0)
236 lli->lli_flags &= ~LLIF_XATTR_CACHE;
241 int ll_xattr_cache_destroy(struct inode *inode)
243 struct ll_inode_info *lli = ll_i2info(inode);
246 down_write(&lli->lli_xattrs_list_rwsem);
247 rc = ll_xattr_cache_destroy_locked(lli);
248 up_write(&lli->lli_xattrs_list_rwsem);
254 * Match or enqueue a PR lock.
256 * Find or request an LDLM lock with xattr data.
257 * Since LDLM does not provide API for atomic match_or_enqueue,
258 * the function handles it with a separate enq lock.
259 * If successful, the function exits with the list lock held.
261 * \retval 0 no error occurred
262 * \retval -ENOMEM not enough memory
264 static int ll_xattr_find_get_lock(struct inode *inode,
265 struct lookup_intent *oit,
266 struct ptlrpc_request **req)
269 struct lustre_handle lockh = { 0 };
270 struct md_op_data *op_data;
271 struct ll_inode_info *lli = ll_i2info(inode);
272 struct ldlm_enqueue_info einfo = {
273 .ei_type = LDLM_IBITS,
274 .ei_mode = it_to_lock_mode(oit),
275 .ei_cb_bl = &ll_md_blocking_ast,
276 .ei_cb_cp = &ldlm_completion_ast,
278 struct ll_sb_info *sbi = ll_i2sbi(inode);
279 struct obd_export *exp = sbi->ll_md_exp;
282 mutex_lock(&lli->lli_xattrs_enq_lock);
283 /* inode may have been shrunk and recreated, so data is gone, match lock
284 * only when data exists.
286 if (ll_xattr_cache_valid(lli)) {
287 /* Try matching first. */
288 mode = ll_take_md_lock(inode, MDS_INODELOCK_XATTR, &lockh, 0,
291 /* fake oit in mdc_revalidate_lock() manner */
292 oit->it_lock_handle = lockh.cookie;
293 oit->it_lock_mode = mode;
298 /* Enqueue if the lock isn't cached locally. */
299 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
300 LUSTRE_OPC_ANY, NULL);
301 if (IS_ERR(op_data)) {
302 mutex_unlock(&lli->lli_xattrs_enq_lock);
303 return PTR_ERR(op_data);
306 op_data->op_valid = OBD_MD_FLXATTR | OBD_MD_FLXATTRLS;
308 rc = md_enqueue(exp, &einfo, NULL, oit, op_data, &lockh, 0);
309 ll_finish_md_op_data(op_data);
313 "md_intent_lock failed with %d for fid "DFID"\n",
314 rc, PFID(ll_inode2fid(inode)));
315 mutex_unlock(&lli->lli_xattrs_enq_lock);
319 *req = oit->it_request;
321 down_write(&lli->lli_xattrs_list_rwsem);
322 mutex_unlock(&lli->lli_xattrs_enq_lock);
328 * Refill the xattr cache.
330 * Fetch and cache the whole of xattrs for @inode, acquiring
331 * a read or a write xattr lock depending on operation in @oit.
332 * Intent is dropped on exit unless the operation is setxattr.
334 * \retval 0 no error occurred
335 * \retval -EPROTO network protocol error
336 * \retval -ENOMEM not enough memory for the cache
338 static int ll_xattr_cache_refill(struct inode *inode, struct lookup_intent *oit)
340 struct ll_sb_info *sbi = ll_i2sbi(inode);
341 struct ptlrpc_request *req = NULL;
342 const char *xdata, *xval, *xtail, *xvtail;
343 struct ll_inode_info *lli = ll_i2info(inode);
344 struct mdt_body *body;
348 rc = ll_xattr_find_get_lock(inode, oit, &req);
352 /* Do we have the data at this point? */
353 if (ll_xattr_cache_valid(lli)) {
354 ll_stats_ops_tally(sbi, LPROC_LL_GETXATTR_HITS, 1);
359 /* Matched but no cache? Cancelled on error by a parallel refill. */
360 if (unlikely(!req)) {
361 CDEBUG(D_CACHE, "cancelled by a parallel getxattr\n");
366 if (oit->it_status < 0) {
367 CDEBUG(D_CACHE, "getxattr intent returned %d for fid "DFID"\n",
368 oit->it_status, PFID(ll_inode2fid(inode)));
370 /* xattr data is so large that we don't want to cache it */
376 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
378 CERROR("no MDT BODY in the refill xattr reply\n");
382 /* do not need swab xattr data */
383 xdata = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA,
384 body->mbo_eadatasize);
385 xval = req_capsule_server_sized_get(&req->rq_pill, &RMF_EAVALS,
387 xsizes = req_capsule_server_sized_get(&req->rq_pill, &RMF_EAVALS_LENS,
388 body->mbo_max_mdsize * sizeof(__u32));
389 if (!xdata || !xval || !xsizes) {
390 CERROR("wrong setxattr reply\n");
395 xtail = xdata + body->mbo_eadatasize;
396 xvtail = xval + body->mbo_aclsize;
398 CDEBUG(D_CACHE, "caching: xdata=%p xtail=%p\n", xdata, xtail);
400 ll_xattr_cache_init(lli);
402 for (i = 0; i < body->mbo_max_mdsize; i++) {
403 CDEBUG(D_CACHE, "caching [%s]=%.*s\n", xdata, *xsizes, xval);
404 /* Perform consistency checks: attr names and vals in pill */
405 if (!memchr(xdata, 0, xtail - xdata)) {
406 CERROR("xattr protocol violation (names are broken)\n");
408 } else if (xval + *xsizes > xvtail) {
409 CERROR("xattr protocol violation (vals are broken)\n");
411 } else if (OBD_FAIL_CHECK(OBD_FAIL_LLITE_XATTR_ENOMEM)) {
413 } else if (!strcmp(xdata, XATTR_NAME_ACL_ACCESS)) {
414 /* Filter out ACL ACCESS since it's cached separately */
415 CDEBUG(D_CACHE, "not caching %s\n",
416 XATTR_NAME_ACL_ACCESS);
419 rc = ll_xattr_cache_add(&lli->lli_xattrs, xdata, xval,
423 ll_xattr_cache_destroy_locked(lli);
426 xdata += strlen(xdata) + 1;
431 if (xdata != xtail || xval != xvtail)
432 CERROR("a hole in xattr data\n");
434 ll_set_lock_data(sbi->ll_md_exp, inode, oit, NULL);
439 ll_intent_drop_lock(oit);
442 up_write(&lli->lli_xattrs_list_rwsem);
444 ptlrpc_req_finished(req);
449 up_write(&lli->lli_xattrs_list_rwsem);
451 ldlm_lock_decref_and_cancel((struct lustre_handle *)
452 &oit->it_lock_handle,
459 * Get an xattr value or list xattrs using the write-through cache.
461 * Get the xattr value (@valid has OBD_MD_FLXATTR set) of @name or
462 * list xattr names (@valid has OBD_MD_FLXATTRLS set) for @inode.
463 * The resulting value/list is stored in @buffer if the former
464 * is not larger than @size.
466 * \retval 0 no error occurred
467 * \retval -EPROTO network protocol error
468 * \retval -ENOMEM not enough memory for the cache
469 * \retval -ERANGE the buffer is not large enough
470 * \retval -ENODATA no such attr or the list is empty
472 int ll_xattr_cache_get(struct inode *inode, const char *name, char *buffer,
473 size_t size, __u64 valid)
475 struct lookup_intent oit = { .it_op = IT_GETXATTR };
476 struct ll_inode_info *lli = ll_i2info(inode);
479 LASSERT(!!(valid & OBD_MD_FLXATTR) ^ !!(valid & OBD_MD_FLXATTRLS));
481 down_read(&lli->lli_xattrs_list_rwsem);
482 if (!ll_xattr_cache_valid(lli)) {
483 up_read(&lli->lli_xattrs_list_rwsem);
484 rc = ll_xattr_cache_refill(inode, &oit);
487 downgrade_write(&lli->lli_xattrs_list_rwsem);
489 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_GETXATTR_HITS, 1);
492 if (valid & OBD_MD_FLXATTR) {
493 struct ll_xattr_entry *xattr;
495 rc = ll_xattr_cache_find(&lli->lli_xattrs, name, &xattr);
497 rc = xattr->xe_vallen;
498 /* zero size means we are only requested size in rc */
500 if (size >= xattr->xe_vallen)
501 memcpy(buffer, xattr->xe_value,
507 } else if (valid & OBD_MD_FLXATTRLS) {
508 rc = ll_xattr_cache_list(&lli->lli_xattrs,
509 size ? buffer : NULL, size);
514 up_read(&lli->lli_xattrs_list_rwsem);