[mirror_ubuntu-artful-kernel.git] / drivers / staging / lustre / lustre / llite / xattr_cache.c

/*
 * Copyright 2012 Xyratex Technology Limited
 *
 * Author: Andrew Perepechko <Andrew_Perepechko@xyratex.com>
 *
 */

#define DEBUG_SUBSYSTEM S_LLITE

#include <linux/fs.h>
#include <linux/sched.h>
#include <linux/mm.h>
#include <obd_support.h>
#include <lustre_lite.h>
#include <lustre_dlm.h>
#include <lustre_ver.h>
#include "llite_internal.h"

/* If we ever have hundreds of extended attributes, we might want to consider
 * using a hash or a tree structure instead of list for faster lookups.
 */
struct ll_xattr_entry {
	struct list_head	xe_list;    /* protected with
					     * lli_xattrs_list_rwsem */
	char			*xe_name;   /* xattr name, \0-terminated */
	char			*xe_value;  /* xattr value */
	unsigned		xe_namelen; /* strlen(xe_name) + 1 */
	unsigned		xe_vallen;  /* xattr value length */
};

static struct kmem_cache *xattr_kmem;
static struct lu_kmem_descr xattr_caches[] = {
	{
		.ckd_cache = &xattr_kmem,
		.ckd_name  = "xattr_kmem",
		.ckd_size  = sizeof(struct ll_xattr_entry)
	},
	{
		.ckd_cache = NULL
	}
};

int ll_xattr_init(void)
{
	return lu_kmem_init(xattr_caches);
}

void ll_xattr_fini(void)
{
	lu_kmem_fini(xattr_caches);
}

/**
 * Initializes xattr cache for an inode.
 *
 * This initializes the xattr list and marks cache presence.
 */
static void ll_xattr_cache_init(struct ll_inode_info *lli)
{


	LASSERT(lli != NULL);

	INIT_LIST_HEAD(&lli->lli_xattrs);
	lli->lli_flags |= LLIF_XATTR_CACHE;
}

/**
 *  This looks for a specific extended attribute.
 *
 *  Find in @cache and return @xattr_name attribute in @xattr,
 *  for the NULL @xattr_name return the first cached @xattr.
 *
 *  \retval 0        success
 *  \retval -ENODATA if not found
 */
static int ll_xattr_cache_find(struct list_head *cache,
			       const char *xattr_name,
			       struct ll_xattr_entry **xattr)
{
	struct ll_xattr_entry *entry;


	list_for_each_entry(entry, cache, xe_list) {
		/* xattr_name == NULL means look for any entry */
		if (xattr_name == NULL ||
		    strcmp(xattr_name, entry->xe_name) == 0) {
			*xattr = entry;
			CDEBUG(D_CACHE, "find: [%s]=%.*s\n",
			       entry->xe_name, entry->xe_vallen,
			       entry->xe_value);
			return 0;
		}
	}

	return -ENODATA;
}

/**
 * This adds an xattr.
 *
 * Add @xattr_name attr with @xattr_val value and @xattr_val_len length,
 *
 * \retval 0       success
 * \retval -ENOMEM if no memory could be allocated for the cached attr
 * \retval -EPROTO if duplicate xattr is being added
 */
static int ll_xattr_cache_add(struct list_head *cache,
			      const char *xattr_name,
			      const char *xattr_val,
			      unsigned xattr_val_len)
{
	struct ll_xattr_entry *xattr;


	if (ll_xattr_cache_find(cache, xattr_name, &xattr) == 0) {
		CDEBUG(D_CACHE, "duplicate xattr: [%s]\n", xattr_name);
		return -EPROTO;
	}

	OBD_SLAB_ALLOC_PTR_GFP(xattr, xattr_kmem, __GFP_IO);
	if (xattr == NULL) {
		CDEBUG(D_CACHE, "failed to allocate xattr\n");
		return -ENOMEM;
	}

	xattr->xe_namelen = strlen(xattr_name) + 1;

	OBD_ALLOC(xattr->xe_name, xattr->xe_namelen);
	if (!xattr->xe_name) {
		CDEBUG(D_CACHE, "failed to alloc xattr name %u\n",
		       xattr->xe_namelen);
		goto err_name;
	}
	OBD_ALLOC(xattr->xe_value, xattr_val_len);
	if (!xattr->xe_value) {
		CDEBUG(D_CACHE, "failed to alloc xattr value %d\n",
		       xattr_val_len);
		goto err_value;
	}

	memcpy(xattr->xe_name, xattr_name, xattr->xe_namelen);
	memcpy(xattr->xe_value, xattr_val, xattr_val_len);
	xattr->xe_vallen = xattr_val_len;
	list_add(&xattr->xe_list, cache);

	CDEBUG(D_CACHE, "set: [%s]=%.*s\n", xattr_name,
		xattr_val_len, xattr_val);

	return 0;
err_value:
	OBD_FREE(xattr->xe_name, xattr->xe_namelen);
err_name:
	OBD_SLAB_FREE_PTR(xattr, xattr_kmem);

	return -ENOMEM;
}

/**
 * This removes an extended attribute from cache.
 *
 * Remove @xattr_name attribute from @cache.
 *
 * \retval 0        success
 * \retval -ENODATA if @xattr_name is not cached
 */
static int ll_xattr_cache_del(struct list_head *cache,
			      const char *xattr_name)
{
	struct ll_xattr_entry *xattr;


	CDEBUG(D_CACHE, "del xattr: %s\n", xattr_name);

	if (ll_xattr_cache_find(cache, xattr_name, &xattr) == 0) {
		list_del(&xattr->xe_list);
		OBD_FREE(xattr->xe_name, xattr->xe_namelen);
		OBD_FREE(xattr->xe_value, xattr->xe_vallen);
		OBD_SLAB_FREE_PTR(xattr, xattr_kmem);

		return 0;
	}

	return -ENODATA;
}

/**
 * This iterates cached extended attributes.
 *
 * Walk over cached attributes in @cache and
 * fill in @xld_buffer or only calculate buffer
 * size if @xld_buffer is NULL.
 *
 * \retval >= 0     buffer list size
 * \retval -ENODATA if the list cannot fit @xld_size buffer
 */
static int ll_xattr_cache_list(struct list_head *cache,
			       char *xld_buffer,
			       int xld_size)
{
	struct ll_xattr_entry *xattr, *tmp;
	int xld_tail = 0;


	list_for_each_entry_safe(xattr, tmp, cache, xe_list) {
		CDEBUG(D_CACHE, "list: buffer=%p[%d] name=%s\n",
			xld_buffer, xld_tail, xattr->xe_name);

		if (xld_buffer) {
			xld_size -= xattr->xe_namelen;
			if (xld_size < 0)
				break;
			memcpy(&xld_buffer[xld_tail],
			       xattr->xe_name, xattr->xe_namelen);
		}
		xld_tail += xattr->xe_namelen;
	}

	if (xld_size < 0)
		return -ERANGE;

	return xld_tail;
}

/**
 * Check if the xattr cache is initialized (filled).
 *
 * \retval 0 @cache is not initialized
 * \retval 1 @cache is initialized
 */
int ll_xattr_cache_valid(struct ll_inode_info *lli)
{
	return !!(lli->lli_flags & LLIF_XATTR_CACHE);
}

/**
 * This finalizes the xattr cache.
 *
 * Free all xattr memory. @lli is the inode info pointer.
 *
 * \retval 0 no error occured
 */
static int ll_xattr_cache_destroy_locked(struct ll_inode_info *lli)
{


	if (!ll_xattr_cache_valid(lli))
		return 0;

	while (ll_xattr_cache_del(&lli->lli_xattrs, NULL) == 0)
		; /* empty loop */
	lli->lli_flags &= ~LLIF_XATTR_CACHE;

	return 0;
}

int ll_xattr_cache_destroy(struct inode *inode)
{
	struct ll_inode_info *lli = ll_i2info(inode);
	int rc;


	down_write(&lli->lli_xattrs_list_rwsem);
	rc = ll_xattr_cache_destroy_locked(lli);
	up_write(&lli->lli_xattrs_list_rwsem);

	return rc;
}

/**
 * Match or enqueue a PR lock.
 *
 * Find or request an LDLM lock with xattr data.
 * Since LDLM does not provide API for atomic match_or_enqueue,
 * the function handles it with a separate enq lock.
 * If successful, the function exits with the list lock held.
 *
 * \retval 0       no error occured
 * \retval -ENOMEM not enough memory
 */
static int ll_xattr_find_get_lock(struct inode *inode,
				  struct lookup_intent *oit,
				  struct ptlrpc_request **req)
{
	ldlm_mode_t mode;
	struct lustre_handle lockh = { 0 };
	struct md_op_data *op_data;
	struct ll_inode_info *lli = ll_i2info(inode);
	struct ldlm_enqueue_info einfo = { .ei_type = LDLM_IBITS,
					   .ei_mode = it_to_lock_mode(oit),
					   .ei_cb_bl = ll_md_blocking_ast,
					   .ei_cb_cp = ldlm_completion_ast };
	struct ll_sb_info *sbi = ll_i2sbi(inode);
	struct obd_export *exp = sbi->ll_md_exp;
	int rc;


	mutex_lock(&lli->lli_xattrs_enq_lock);
	/* Try matching first. */
	mode = ll_take_md_lock(inode, MDS_INODELOCK_XATTR, &lockh, 0, LCK_PR);
	if (mode != 0) {
		/* fake oit in mdc_revalidate_lock() manner */
		oit->d.lustre.it_lock_handle = lockh.cookie;
		oit->d.lustre.it_lock_mode = mode;
		goto out;
	}

	/* Enqueue if the lock isn't cached locally. */
	op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
				     LUSTRE_OPC_ANY, NULL);
	if (IS_ERR(op_data)) {
		mutex_unlock(&lli->lli_xattrs_enq_lock);
		return PTR_ERR(op_data);
	}

	op_data->op_valid = OBD_MD_FLXATTR | OBD_MD_FLXATTRLS;

	rc = md_enqueue(exp, &einfo, oit, op_data, &lockh, NULL, 0, NULL, 0);
	ll_finish_md_op_data(op_data);

	if (rc < 0) {
		CDEBUG(D_CACHE,
		       "md_intent_lock failed with %d for fid "DFID"\n",
		       rc, PFID(ll_inode2fid(inode)));
		mutex_unlock(&lli->lli_xattrs_enq_lock);
		return rc;
	}

	*req = (struct ptlrpc_request *)oit->d.lustre.it_data;
out:
	down_write(&lli->lli_xattrs_list_rwsem);
	mutex_unlock(&lli->lli_xattrs_enq_lock);

	return 0;
}

/**
 * Refill the xattr cache.
 *
 * Fetch and cache the whole of xattrs for @inode, acquiring
 * a read or a write xattr lock depending on operation in @oit.
 * Intent is dropped on exit unless the operation is setxattr.
 *
 * \retval 0       no error occured
 * \retval -EPROTO network protocol error
 * \retval -ENOMEM not enough memory for the cache
 */
static int ll_xattr_cache_refill(struct inode *inode, struct lookup_intent *oit)
{
	struct ll_sb_info *sbi = ll_i2sbi(inode);
	struct ptlrpc_request *req = NULL;
	const char *xdata, *xval, *xtail, *xvtail;
	struct ll_inode_info *lli = ll_i2info(inode);
	struct mdt_body *body;
	__u32 *xsizes;
	int rc = 0, i;


	rc = ll_xattr_find_get_lock(inode, oit, &req);
	if (rc)
		GOTO(out_no_unlock, rc);

	/* Do we have the data at this point? */
	if (ll_xattr_cache_valid(lli)) {
		ll_stats_ops_tally(sbi, LPROC_LL_GETXATTR_HITS, 1);
		GOTO(out_maybe_drop, rc = 0);
	}

	/* Matched but no cache? Cancelled on error by a parallel refill. */
	if (unlikely(req == NULL)) {
		CDEBUG(D_CACHE, "cancelled by a parallel getxattr\n");
		GOTO(out_maybe_drop, rc = -EIO);
	}

	if (oit->d.lustre.it_status < 0) {
		CDEBUG(D_CACHE, "getxattr intent returned %d for fid "DFID"\n",
		       oit->d.lustre.it_status, PFID(ll_inode2fid(inode)));
		rc = oit->d.lustre.it_status;
		/* xattr data is so large that we don't want to cache it */
		if (rc == -ERANGE)
			rc = -EAGAIN;
		GOTO(out_destroy, rc);
	}

	body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
	if (body == NULL) {
		CERROR("no MDT BODY in the refill xattr reply\n");
		GOTO(out_destroy, rc = -EPROTO);
	}
	/* do not need swab xattr data */
	xdata = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA,
						body->eadatasize);
	xval = req_capsule_server_sized_get(&req->rq_pill, &RMF_EAVALS,
						body->aclsize);
	xsizes = req_capsule_server_sized_get(&req->rq_pill, &RMF_EAVALS_LENS,
					      body->max_mdsize * sizeof(__u32));
	if (xdata == NULL || xval == NULL || xsizes == NULL) {
		CERROR("wrong setxattr reply\n");
		GOTO(out_destroy, rc = -EPROTO);
	}

	xtail = xdata + body->eadatasize;
	xvtail = xval + body->aclsize;

	CDEBUG(D_CACHE, "caching: xdata=%p xtail=%p\n", xdata, xtail);

	ll_xattr_cache_init(lli);

	for (i = 0; i < body->max_mdsize; i++) {
		CDEBUG(D_CACHE, "caching [%s]=%.*s\n", xdata, *xsizes, xval);
		/* Perform consistency checks: attr names and vals in pill */
		if (memchr(xdata, 0, xtail - xdata) == NULL) {
			CERROR("xattr protocol violation (names are broken)\n");
			rc = -EPROTO;
		} else if (xval + *xsizes > xvtail) {
			CERROR("xattr protocol violation (vals are broken)\n");
			rc = -EPROTO;
		} else if (OBD_FAIL_CHECK(OBD_FAIL_LLITE_XATTR_ENOMEM)) {
			rc = -ENOMEM;
		} else if (!strcmp(xdata, XATTR_NAME_ACL_ACCESS)) {
			/* Filter out ACL ACCESS since it's cached separately */
			CDEBUG(D_CACHE, "not caching %s\n",
			       XATTR_NAME_ACL_ACCESS);
			rc = 0;
		} else {
			rc = ll_xattr_cache_add(&lli->lli_xattrs, xdata, xval,
						*xsizes);
		}
		if (rc < 0) {
			ll_xattr_cache_destroy_locked(lli);
			GOTO(out_destroy, rc);
		}
		xdata += strlen(xdata) + 1;
		xval  += *xsizes;
		xsizes++;
	}

	if (xdata != xtail || xval != xvtail)
		CERROR("a hole in xattr data\n");

	ll_set_lock_data(sbi->ll_md_exp, inode, oit, NULL);

	GOTO(out_maybe_drop, rc);
out_maybe_drop:

		ll_intent_drop_lock(oit);

	if (rc != 0)
		up_write(&lli->lli_xattrs_list_rwsem);
out_no_unlock:
	ptlrpc_req_finished(req);

	return rc;

out_destroy:
	up_write(&lli->lli_xattrs_list_rwsem);

	ldlm_lock_decref_and_cancel((struct lustre_handle *)
					&oit->d.lustre.it_lock_handle,
					oit->d.lustre.it_lock_mode);

	goto out_no_unlock;
}

/**
 * Get an xattr value or list xattrs using the write-through cache.
 *
 * Get the xattr value (@valid has OBD_MD_FLXATTR set) of @name or
 * list xattr names (@valid has OBD_MD_FLXATTRLS set) for @inode.
 * The resulting value/list is stored in @buffer if the former
 * is not larger than @size.
 *
 * \retval 0        no error occured
 * \retval -EPROTO  network protocol error
 * \retval -ENOMEM  not enough memory for the cache
 * \retval -ERANGE  the buffer is not large enough
 * \retval -ENODATA no such attr or the list is empty
 */
int ll_xattr_cache_get(struct inode *inode,
			const char *name,
			char *buffer,
			size_t size,
			__u64 valid)
{
	struct lookup_intent oit = { .it_op = IT_GETXATTR };
	struct ll_inode_info *lli = ll_i2info(inode);
	int rc = 0;


	LASSERT(!!(valid & OBD_MD_FLXATTR) ^ !!(valid & OBD_MD_FLXATTRLS));

	down_read(&lli->lli_xattrs_list_rwsem);
	if (!ll_xattr_cache_valid(lli)) {
		up_read(&lli->lli_xattrs_list_rwsem);
		rc = ll_xattr_cache_refill(inode, &oit);
		if (rc)
			return rc;
		downgrade_write(&lli->lli_xattrs_list_rwsem);
	} else {
		ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_GETXATTR_HITS, 1);
	}

	if (valid & OBD_MD_FLXATTR) {
		struct ll_xattr_entry *xattr;

		rc = ll_xattr_cache_find(&lli->lli_xattrs, name, &xattr);
		if (rc == 0) {
			rc = xattr->xe_vallen;
			/* zero size means we are only requested size in rc */
			if (size != 0) {
				if (size >= xattr->xe_vallen)
					memcpy(buffer, xattr->xe_value,
						xattr->xe_vallen);
				else
					rc = -ERANGE;
			}
		}
	} else if (valid & OBD_MD_FLXATTRLS) {
		rc = ll_xattr_cache_list(&lli->lli_xattrs,
					 size ? buffer : NULL, size);
	}

	GOTO(out, rc);
out:
	up_read(&lli->lli_xattrs_list_rwsem);

	return rc;
}
Commit	Line	Data
7fc1f831 AP	1	/*
	2	* Copyright 2012 Xyratex Technology Limited
	3	*
	4	* Author: Andrew Perepechko <Andrew_Perepechko@xyratex.com>
	5	*
	6	*/
	7
	8	#define DEBUG_SUBSYSTEM S_LLITE
	9
	10	#include <linux/fs.h>
	11	#include <linux/sched.h>
	12	#include <linux/mm.h>
	13	#include <obd_support.h>
	14	#include <lustre_lite.h>
	15	#include <lustre_dlm.h>
	16	#include <lustre_ver.h>
	17	#include "llite_internal.h"
	18
	19	/* If we ever have hundreds of extended attributes, we might want to consider
	20	* using a hash or a tree structure instead of list for faster lookups.
	21	*/
	22	struct ll_xattr_entry {
	23	struct list_head xe_list; /* protected with
	24	* lli_xattrs_list_rwsem */
	25	char xe_name; / xattr name, \0-terminated */
	26	char xe_value; / xattr value */
	27	unsigned xe_namelen; /* strlen(xe_name) + 1 */
	28	unsigned xe_vallen; /* xattr value length */
	29	};
	30
	31	static struct kmem_cache *xattr_kmem;
	32	static struct lu_kmem_descr xattr_caches[] = {
	33	{
	34	.ckd_cache = &xattr_kmem,
	35	.ckd_name = "xattr_kmem",
	36	.ckd_size = sizeof(struct ll_xattr_entry)
	37	},
	38	{
	39	.ckd_cache = NULL
	40	}
	41	};
	42
	43	int ll_xattr_init(void)
	44	{
	45	return lu_kmem_init(xattr_caches);
	46	}
	47
	48	void ll_xattr_fini(void)
	49	{
	50	lu_kmem_fini(xattr_caches);
	51	}
	52
	53	/**
	54	* Initializes xattr cache for an inode.
	55	*
	56	* This initializes the xattr list and marks cache presence.
	57	*/
	58	static void ll_xattr_cache_init(struct ll_inode_info *lli)
	59	{
	60
	61
	62	LASSERT(lli != NULL);
	63
	64	INIT_LIST_HEAD(&lli->lli_xattrs);
65	lli->lli_flags \|= LLIF_XATTR_CACHE;
66	}
67
68	/**
69	* This looks for a specific extended attribute.
70	*
71	* Find in @cache and return @xattr_name attribute in @xattr,
72	* for the NULL @xattr_name return the first cached @xattr.
73	*
74	* \retval 0 success
75	* \retval -ENODATA if not found
76	*/
77	static int ll_xattr_cache_find(struct list_head *cache,
78	const char *xattr_name,
79	struct ll_xattr_entry **xattr)
80	{
81	struct ll_xattr_entry *entry;
82
83
84
85	list_for_each_entry(entry, cache, xe_list) {
86	/* xattr_name == NULL means look for any entry */
87	if (xattr_name == NULL \|\|
88	strcmp(xattr_name, entry->xe_name) == 0) {
89	*xattr = entry;
90	CDEBUG(D_CACHE, "find: [%s]=%.*s\n",
91	entry->xe_name, entry->xe_vallen,
92	entry->xe_value);
93	return 0;
94	}
95	}
96
97	return -ENODATA;
98	}
99
100	/**
e93a3082	101	* This adds an xattr.
7fc1f831 AP	102	*
7fc1f831 AP	103	* Add @xattr_name attr with @xattr_val value and @xattr_val_len length,
7fc1f831 AP	104	*
	105	* \retval 0 success
	106	* \retval -ENOMEM if no memory could be allocated for the cached attr
e93a3082	107	* \retval -EPROTO if duplicate xattr is being added
7fc1f831 AP	108	*/
	109	static int ll_xattr_cache_add(struct list_head *cache,
	110	const char *xattr_name,
	111	const char *xattr_val,
	112	unsigned xattr_val_len)
	113	{
	114	struct ll_xattr_entry *xattr;
	115
	116
	117
	118	if (ll_xattr_cache_find(cache, xattr_name, &xattr) == 0) {
e93a3082 AP	119	CDEBUG(D_CACHE, "duplicate xattr: [%s]\n", xattr_name);
e93a3082 AP	120	return -EPROTO;
7fc1f831 AP	121	}
	122
	123	OBD_SLAB_ALLOC_PTR_GFP(xattr, xattr_kmem, __GFP_IO);
	124	if (xattr == NULL) {
	125	CDEBUG(D_CACHE, "failed to allocate xattr\n");
	126	return -ENOMEM;
	127	}
	128
	129	xattr->xe_namelen = strlen(xattr_name) + 1;
	130
	131	OBD_ALLOC(xattr->xe_name, xattr->xe_namelen);
	132	if (!xattr->xe_name) {
	133	CDEBUG(D_CACHE, "failed to alloc xattr name %u\n",
	134	xattr->xe_namelen);
	135	goto err_name;
	136	}
	137	OBD_ALLOC(xattr->xe_value, xattr_val_len);
	138	if (!xattr->xe_value) {
	139	CDEBUG(D_CACHE, "failed to alloc xattr value %d\n",
	140	xattr_val_len);
	141	goto err_value;
	142	}
	143
	144	memcpy(xattr->xe_name, xattr_name, xattr->xe_namelen);
	145	memcpy(xattr->xe_value, xattr_val, xattr_val_len);
	146	xattr->xe_vallen = xattr_val_len;
	147	list_add(&xattr->xe_list, cache);
	148
	149	CDEBUG(D_CACHE, "set: [%s]=%.*s\n", xattr_name,
	150	xattr_val_len, xattr_val);
	151
	152	return 0;
	153	err_value:
	154	OBD_FREE(xattr->xe_name, xattr->xe_namelen);
	155	err_name:
	156	OBD_SLAB_FREE_PTR(xattr, xattr_kmem);
	157
	158	return -ENOMEM;
	159	}
	160
	161	/**
	162	* This removes an extended attribute from cache.
	163	*
	164	* Remove @xattr_name attribute from @cache.
	165	*
	166	* \retval 0 success
	167	* \retval -ENODATA if @xattr_name is not cached
	168	*/
	169	static int ll_xattr_cache_del(struct list_head *cache,
	170	const char *xattr_name)
	171	{
	172	struct ll_xattr_entry *xattr;
	173
	174
	175
	176	CDEBUG(D_CACHE, "del xattr: %s\n", xattr_name);
	177
	178	if (ll_xattr_cache_find(cache, xattr_name, &xattr) == 0) {
	179	list_del(&xattr->xe_list);
	180	OBD_FREE(xattr->xe_name, xattr->xe_namelen);
	181	OBD_FREE(xattr->xe_value, xattr->xe_vallen);
	182	OBD_SLAB_FREE_PTR(xattr, xattr_kmem);
	183
	184	return 0;
185	}
186
187	return -ENODATA;
188	}
189
190	/**
191	* This iterates cached extended attributes.
192	*
193	* Walk over cached attributes in @cache and
194	* fill in @xld_buffer or only calculate buffer
195	* size if @xld_buffer is NULL.
196	*
197	* \retval >= 0 buffer list size
198	* \retval -ENODATA if the list cannot fit @xld_size buffer
199	*/
200	static int ll_xattr_cache_list(struct list_head *cache,
201	char *xld_buffer,
202	int xld_size)
203	{
204	struct ll_xattr_entry xattr, tmp;
205	int xld_tail = 0;
206
207
208
209	list_for_each_entry_safe(xattr, tmp, cache, xe_list) {
210	CDEBUG(D_CACHE, "list: buffer=%p[%d] name=%s\n",
211	xld_buffer, xld_tail, xattr->xe_name);
212
213	if (xld_buffer) {
214	xld_size -= xattr->xe_namelen;
215	if (xld_size < 0)
216	break;
217	memcpy(&xld_buffer[xld_tail],
218	xattr->xe_name, xattr->xe_namelen);
219	}
220	xld_tail += xattr->xe_namelen;
221	}
222
223	if (xld_size < 0)
224	return -ERANGE;
225
226	return xld_tail;
227	}
228
229	/**
230	* Check if the xattr cache is initialized (filled).
231	*
232	* \retval 0 @cache is not initialized
233	* \retval 1 @cache is initialized
234	*/
235	int ll_xattr_cache_valid(struct ll_inode_info *lli)
236	{
237	return !!(lli->lli_flags & LLIF_XATTR_CACHE);
238	}
239
240	/**
241	* This finalizes the xattr cache.
242	*
243	* Free all xattr memory. @lli is the inode info pointer.
244	*
245	* \retval 0 no error occured
246	*/
247	static int ll_xattr_cache_destroy_locked(struct ll_inode_info *lli)
248	{
249
250
251	if (!ll_xattr_cache_valid(lli))
252	return 0;
253
254	while (ll_xattr_cache_del(&lli->lli_xattrs, NULL) == 0)
255	; /* empty loop */
256	lli->lli_flags &= ~LLIF_XATTR_CACHE;
257
258	return 0;
259	}
260
261	int ll_xattr_cache_destroy(struct inode *inode)
262	{
263	struct ll_inode_info *lli = ll_i2info(inode);
264	int rc;
265
266
267
268	down_write(&lli->lli_xattrs_list_rwsem);
269	rc = ll_xattr_cache_destroy_locked(lli);
270	up_write(&lli->lli_xattrs_list_rwsem);
271
272	return rc;
273	}
274
275	/**
e93a3082	276	* Match or enqueue a PR lock.
7fc1f831 AP	277	*
	278	* Find or request an LDLM lock with xattr data.
	279	* Since LDLM does not provide API for atomic match_or_enqueue,
	280	* the function handles it with a separate enq lock.
	281	* If successful, the function exits with the list lock held.
	282	*
	283	* \retval 0 no error occured
	284	* \retval -ENOMEM not enough memory
	285	*/
	286	static int ll_xattr_find_get_lock(struct inode *inode,
	287	struct lookup_intent *oit,
	288	struct ptlrpc_request **req)
	289	{
	290	ldlm_mode_t mode;
	291	struct lustre_handle lockh = { 0 };
	292	struct md_op_data *op_data;
	293	struct ll_inode_info *lli = ll_i2info(inode);
	294	struct ldlm_enqueue_info einfo = { .ei_type = LDLM_IBITS,
	295	.ei_mode = it_to_lock_mode(oit),
	296	.ei_cb_bl = ll_md_blocking_ast,
	297	.ei_cb_cp = ldlm_completion_ast };
	298	struct ll_sb_info *sbi = ll_i2sbi(inode);
	299	struct obd_export *exp = sbi->ll_md_exp;
	300	int rc;
	301
	302
	303
	304	mutex_lock(&lli->lli_xattrs_enq_lock);
	305	/* Try matching first. */
e93a3082	306	mode = ll_take_md_lock(inode, MDS_INODELOCK_XATTR, &lockh, 0, LCK_PR);
7fc1f831 AP	307	if (mode != 0) {
	308	/* fake oit in mdc_revalidate_lock() manner */
	309	oit->d.lustre.it_lock_handle = lockh.cookie;
	310	oit->d.lustre.it_lock_mode = mode;
	311	goto out;
	312	}
	313
	314	/* Enqueue if the lock isn't cached locally. */
	315	op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
	316	LUSTRE_OPC_ANY, NULL);
	317	if (IS_ERR(op_data)) {
	318	mutex_unlock(&lli->lli_xattrs_enq_lock);
	319	return PTR_ERR(op_data);
	320	}
	321
e93a3082	322	op_data->op_valid = OBD_MD_FLXATTR \| OBD_MD_FLXATTRLS;
7fc1f831 AP	323
	324	rc = md_enqueue(exp, &einfo, oit, op_data, &lockh, NULL, 0, NULL, 0);
	325	ll_finish_md_op_data(op_data);
	326
	327	if (rc < 0) {
	328	CDEBUG(D_CACHE,
	329	"md_intent_lock failed with %d for fid "DFID"\n",
	330	rc, PFID(ll_inode2fid(inode)));
	331	mutex_unlock(&lli->lli_xattrs_enq_lock);
	332	return rc;
	333	}
	334
	335	req = (struct ptlrpc_request )oit->d.lustre.it_data;
	336	out:
	337	down_write(&lli->lli_xattrs_list_rwsem);
	338	mutex_unlock(&lli->lli_xattrs_enq_lock);
	339
	340	return 0;
	341	}
	342
	343	/**
	344	* Refill the xattr cache.
	345	*
	346	* Fetch and cache the whole of xattrs for @inode, acquiring
	347	* a read or a write xattr lock depending on operation in @oit.
	348	* Intent is dropped on exit unless the operation is setxattr.
	349	*
	350	* \retval 0 no error occured
	351	* \retval -EPROTO network protocol error
	352	* \retval -ENOMEM not enough memory for the cache
	353	*/
	354	static int ll_xattr_cache_refill(struct inode inode, struct lookup_intent oit)
	355	{
	356	struct ll_sb_info *sbi = ll_i2sbi(inode);
	357	struct ptlrpc_request *req = NULL;
	358	const char xdata, xval, xtail, xvtail;
	359	struct ll_inode_info *lli = ll_i2info(inode);
	360	struct mdt_body *body;
	361	__u32 *xsizes;
	362	int rc = 0, i;
	363
	364
	365
	366	rc = ll_xattr_find_get_lock(inode, oit, &req);
	367	if (rc)
	368	GOTO(out_no_unlock, rc);
	369
	370	/* Do we have the data at this point? */
	371	if (ll_xattr_cache_valid(lli)) {
	372	ll_stats_ops_tally(sbi, LPROC_LL_GETXATTR_HITS, 1);
	373	GOTO(out_maybe_drop, rc = 0);
	374	}
	375
	376	/* Matched but no cache? Cancelled on error by a parallel refill. */
	377	if (unlikely(req == NULL)) {
	378	CDEBUG(D_CACHE, "cancelled by a parallel getxattr\n");
	379	GOTO(out_maybe_drop, rc = -EIO);
	380	}
	381
	382	if (oit->d.lustre.it_status < 0) {
	383	CDEBUG(D_CACHE, "getxattr intent returned %d for fid "DFID"\n",
	384	oit->d.lustre.it_status, PFID(ll_inode2fid(inode)));
e93a3082 AP	385	rc = oit->d.lustre.it_status;
	386	/* xattr data is so large that we don't want to cache it */
	387	if (rc == -ERANGE)
	388	rc = -EAGAIN;
	389	GOTO(out_destroy, rc);
7fc1f831 AP	390	}
	391
	392	body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
	393	if (body == NULL) {
	394	CERROR("no MDT BODY in the refill xattr reply\n");
	395	GOTO(out_destroy, rc = -EPROTO);
	396	}
	397	/* do not need swab xattr data */
	398	xdata = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA,
	399	body->eadatasize);
	400	xval = req_capsule_server_sized_get(&req->rq_pill, &RMF_EAVALS,
	401	body->aclsize);
	402	xsizes = req_capsule_server_sized_get(&req->rq_pill, &RMF_EAVALS_LENS,
	403	body->max_mdsize * sizeof(__u32));
	404	if (xdata == NULL \|\| xval == NULL \|\| xsizes == NULL) {
	405	CERROR("wrong setxattr reply\n");
	406	GOTO(out_destroy, rc = -EPROTO);
	407	}
	408
	409	xtail = xdata + body->eadatasize;
	410	xvtail = xval + body->aclsize;
	411
	412	CDEBUG(D_CACHE, "caching: xdata=%p xtail=%p\n", xdata, xtail);
	413
	414	ll_xattr_cache_init(lli);
	415
	416	for (i = 0; i < body->max_mdsize; i++) {
	417	CDEBUG(D_CACHE, "caching [%s]=%.s\n", xdata, xsizes, xval);
	418	/* Perform consistency checks: attr names and vals in pill */
	419	if (memchr(xdata, 0, xtail - xdata) == NULL) {
	420	CERROR("xattr protocol violation (names are broken)\n");
	421	rc = -EPROTO;
	422	} else if (xval + *xsizes > xvtail) {
	423	CERROR("xattr protocol violation (vals are broken)\n");
	424	rc = -EPROTO;
	425	} else if (OBD_FAIL_CHECK(OBD_FAIL_LLITE_XATTR_ENOMEM)) {
	426	rc = -ENOMEM;
e93a3082 AP	427	} else if (!strcmp(xdata, XATTR_NAME_ACL_ACCESS)) {
	428	/* Filter out ACL ACCESS since it's cached separately */
	429	CDEBUG(D_CACHE, "not caching %s\n",
	430	XATTR_NAME_ACL_ACCESS);
	431	rc = 0;
7fc1f831 AP	432	} else {
	433	rc = ll_xattr_cache_add(&lli->lli_xattrs, xdata, xval,
	434	*xsizes);
	435	}
	436	if (rc < 0) {
	437	ll_xattr_cache_destroy_locked(lli);
	438	GOTO(out_destroy, rc);
	439	}
	440	xdata += strlen(xdata) + 1;
	441	xval += *xsizes;
	442	xsizes++;
	443	}
	444
	445	if (xdata != xtail \|\| xval != xvtail)
	446	CERROR("a hole in xattr data\n");
	447
	448	ll_set_lock_data(sbi->ll_md_exp, inode, oit, NULL);
	449
	450	GOTO(out_maybe_drop, rc);
	451	out_maybe_drop:
e93a3082	452
7fc1f831 AP	453	ll_intent_drop_lock(oit);
	454
	455	if (rc != 0)
	456	up_write(&lli->lli_xattrs_list_rwsem);
	457	out_no_unlock:
	458	ptlrpc_req_finished(req);
	459
	460	return rc;
	461
	462	out_destroy:
	463	up_write(&lli->lli_xattrs_list_rwsem);
	464
	465	ldlm_lock_decref_and_cancel((struct lustre_handle *)
	466	&oit->d.lustre.it_lock_handle,
	467	oit->d.lustre.it_lock_mode);
	468
	469	goto out_no_unlock;
	470	}
	471
	472	/**
	473	* Get an xattr value or list xattrs using the write-through cache.
	474	*
	475	* Get the xattr value (@valid has OBD_MD_FLXATTR set) of @name or
	476	* list xattr names (@valid has OBD_MD_FLXATTRLS set) for @inode.
	477	* The resulting value/list is stored in @buffer if the former
	478	* is not larger than @size.
	479	*
	480	* \retval 0 no error occured
	481	* \retval -EPROTO network protocol error
	482	* \retval -ENOMEM not enough memory for the cache
	483	* \retval -ERANGE the buffer is not large enough
	484	* \retval -ENODATA no such attr or the list is empty
	485	*/
	486	int ll_xattr_cache_get(struct inode *inode,
	487	const char *name,
	488	char *buffer,
	489	size_t size,
	490	__u64 valid)
	491	{
	492	struct lookup_intent oit = { .it_op = IT_GETXATTR };
	493	struct ll_inode_info *lli = ll_i2info(inode);
	494	int rc = 0;
	495
	496
	497
	498	LASSERT(!!(valid & OBD_MD_FLXATTR) ^ !!(valid & OBD_MD_FLXATTRLS));
	499
	500	down_read(&lli->lli_xattrs_list_rwsem);
	501	if (!ll_xattr_cache_valid(lli)) {
	502	up_read(&lli->lli_xattrs_list_rwsem);
	503	rc = ll_xattr_cache_refill(inode, &oit);
	504	if (rc)
	505	return rc;
	506	downgrade_write(&lli->lli_xattrs_list_rwsem);
	507	} else {
	508	ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_GETXATTR_HITS, 1);
	509	}
	510
	511	if (valid & OBD_MD_FLXATTR) {
	512	struct ll_xattr_entry *xattr;
	513
	514	rc = ll_xattr_cache_find(&lli->lli_xattrs, name, &xattr);
	515	if (rc == 0) {
	516	rc = xattr->xe_vallen;
517	/* zero size means we are only requested size in rc */
518	if (size != 0) {
519	if (size >= xattr->xe_vallen)
520	memcpy(buffer, xattr->xe_value,
521	xattr->xe_vallen);
522	else
523	rc = -ERANGE;
524	}
525	}
526	} else if (valid & OBD_MD_FLXATTRLS) {
527	rc = ll_xattr_cache_list(&lli->lli_xattrs,
528	size ? buffer : NULL, size);
529	}
530
531	GOTO(out, rc);
532	out:
533	up_read(&lli->lli_xattrs_list_rwsem);
534
535	return rc;
536	}