[deliverable/linux.git] / drivers / staging / lustre / lustre / llite / xattr_cache.c

/*
 * Copyright 2012 Xyratex Technology Limited
 *
 * Copyright (c) 2013, 2015, Intel Corporation.
 *
 * 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 "../include/obd_support.h"
#include "../include/lustre_lite.h"
#include "../include/lustre_dlm.h"
#include "../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;
	}

	xattr = kmem_cache_alloc(xattr_kmem, GFP_NOFS | __GFP_ZERO);
	if (xattr == NULL) {
		CDEBUG(D_CACHE, "failed to allocate xattr\n");
		return -ENOMEM;
	}

	xattr->xe_name = kstrdup(xattr_name, GFP_NOFS);
	if (!xattr->xe_name) {
		CDEBUG(D_CACHE, "failed to alloc xattr name %u\n",
		       xattr->xe_namelen);
		goto err_name;
	}
	xattr->xe_value = kmemdup(xattr_val, xattr_val_len, GFP_NOFS);
	if (!xattr->xe_value)
		goto err_value;

	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:
	kfree(xattr->xe_name);
err_name:
	kmem_cache_free(xattr_kmem, xattr);

	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);
		kfree(xattr->xe_name);
		kfree(xattr->xe_value);
		kmem_cache_free(xattr_kmem, xattr);

		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
 */
static 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 occurred
 */
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 occurred
 * \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);
	/* inode may have been shrunk and recreated, so data is gone, match lock
	 * only when data exists. */
	if (ll_xattr_cache_valid(lli)) {
		/* 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 occurred
 * \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, i;

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

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

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

	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;
	}

	body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
	if (body == NULL) {
		CERROR("no MDT BODY in the refill xattr reply\n");
		rc = -EPROTO;
		goto out_destroy;
	}
	/* 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");
		rc = -EPROTO;
		goto out_destroy;
	}

	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;
		}
		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;
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 occurred
 * \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;
out:
	up_read(&lli->lli_xattrs_list_rwsem);

	return rc;
}
Commit	Line	Data
7fc1f831 AP	1	/*
	2	* Copyright 2012 Xyratex Technology Limited
	3	*
1dc563a6 AD	4	* Copyright (c) 2013, 2015, Intel Corporation.
1dc563a6 AD	5	*
7fc1f831 AP	6	* Author: Andrew Perepechko <Andrew_Perepechko@xyratex.com>
	7	*
	8	*/
	9
	10	#define DEBUG_SUBSYSTEM S_LLITE
	11
	12	#include <linux/fs.h>
	13	#include <linux/sched.h>
	14	#include <linux/mm.h>
67a235f5 GKH	15	#include "../include/obd_support.h"
	16	#include "../include/lustre_lite.h"
	17	#include "../include/lustre_dlm.h"
	18	#include "../include/lustre_ver.h"
7fc1f831 AP	19	#include "llite_internal.h"
	20
	21	/* If we ever have hundreds of extended attributes, we might want to consider
	22	* using a hash or a tree structure instead of list for faster lookups.
	23	*/
	24	struct ll_xattr_entry {
	25	struct list_head xe_list; /* protected with
	26	* 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 */
	31	};
	32
	33	static struct kmem_cache *xattr_kmem;
	34	static struct lu_kmem_descr xattr_caches[] = {
	35	{
	36	.ckd_cache = &xattr_kmem,
	37	.ckd_name = "xattr_kmem",
	38	.ckd_size = sizeof(struct ll_xattr_entry)
	39	},
	40	{
	41	.ckd_cache = NULL
	42	}
	43	};
	44
	45	int ll_xattr_init(void)
	46	{
	47	return lu_kmem_init(xattr_caches);
	48	}
	49
	50	void ll_xattr_fini(void)
	51	{
	52	lu_kmem_fini(xattr_caches);
	53	}
	54
	55	/**
	56	* Initializes xattr cache for an inode.
	57	*
	58	* This initializes the xattr list and marks cache presence.
	59	*/
	60	static void ll_xattr_cache_init(struct ll_inode_info *lli)
	61	{
	62
7fc1f831 AP	63	LASSERT(lli != NULL);
	64
	65	INIT_LIST_HEAD(&lli->lli_xattrs);
	66	lli->lli_flags \|= LLIF_XATTR_CACHE;
	67	}
	68
	69	/**
	70	* This looks for a specific extended attribute.
	71	*
	72	* Find in @cache and return @xattr_name attribute in @xattr,
	73	* for the NULL @xattr_name return the first cached @xattr.
	74	*
	75	* \retval 0 success
	76	* \retval -ENODATA if not found
	77	*/
	78	static int ll_xattr_cache_find(struct list_head *cache,
	79	const char *xattr_name,
	80	struct ll_xattr_entry **xattr)
	81	{
	82	struct ll_xattr_entry *entry;
	83
7fc1f831 AP	84	list_for_each_entry(entry, cache, xe_list) {
	85	/* xattr_name == NULL means look for any entry */
	86	if (xattr_name == NULL \|\|
	87	strcmp(xattr_name, entry->xe_name) == 0) {
	88	*xattr = entry;
	89	CDEBUG(D_CACHE, "find: [%s]=%.*s\n",
	90	entry->xe_name, entry->xe_vallen,
	91	entry->xe_value);
	92	return 0;
	93	}
	94	}
	95
	96	return -ENODATA;
	97	}
	98
	99	/**
e93a3082	100	* This adds an xattr.
7fc1f831 AP	101	*
7fc1f831 AP	102	* Add @xattr_name attr with @xattr_val value and @xattr_val_len length,
7fc1f831 AP	103	*
	104	* \retval 0 success
	105	* \retval -ENOMEM if no memory could be allocated for the cached attr
e93a3082	106	* \retval -EPROTO if duplicate xattr is being added
7fc1f831 AP	107	*/
	108	static int ll_xattr_cache_add(struct list_head *cache,
	109	const char *xattr_name,
	110	const char *xattr_val,
	111	unsigned xattr_val_len)
	112	{
	113	struct ll_xattr_entry *xattr;
	114
7fc1f831	115	if (ll_xattr_cache_find(cache, xattr_name, &xattr) == 0) {
e93a3082 AP	116	CDEBUG(D_CACHE, "duplicate xattr: [%s]\n", xattr_name);
e93a3082 AP	117	return -EPROTO;
7fc1f831 AP	118	}
7fc1f831 AP	119
ccaabce1	120	xattr = kmem_cache_alloc(xattr_kmem, GFP_NOFS \| __GFP_ZERO);
7fc1f831 AP	121	if (xattr == NULL) {
	122	CDEBUG(D_CACHE, "failed to allocate xattr\n");
	123	return -ENOMEM;
	124	}
	125
b3dd8957	126	xattr->xe_name = kstrdup(xattr_name, GFP_NOFS);
7fc1f831 AP	127	if (!xattr->xe_name) {
	128	CDEBUG(D_CACHE, "failed to alloc xattr name %u\n",
	129	xattr->xe_namelen);
	130	goto err_name;
	131	}
9cda6856	132	xattr->xe_value = kmemdup(xattr_val, xattr_val_len, GFP_NOFS);
695a0666	133	if (!xattr->xe_value)
7fc1f831	134	goto err_value;
7fc1f831	135
7fc1f831 AP	136	xattr->xe_vallen = xattr_val_len;
	137	list_add(&xattr->xe_list, cache);
	138
	139	CDEBUG(D_CACHE, "set: [%s]=%.*s\n", xattr_name,
	140	xattr_val_len, xattr_val);
	141
	142	return 0;
	143	err_value:
97903a26	144	kfree(xattr->xe_name);
7fc1f831	145	err_name:
50d30362	146	kmem_cache_free(xattr_kmem, xattr);
7fc1f831 AP	147
	148	return -ENOMEM;
	149	}
	150
	151	/**
	152	* This removes an extended attribute from cache.
	153	*
	154	* Remove @xattr_name attribute from @cache.
	155	*
	156	* \retval 0 success
	157	* \retval -ENODATA if @xattr_name is not cached
	158	*/
	159	static int ll_xattr_cache_del(struct list_head *cache,
	160	const char *xattr_name)
	161	{
	162	struct ll_xattr_entry *xattr;
	163
7fc1f831 AP	164	CDEBUG(D_CACHE, "del xattr: %s\n", xattr_name);
	165
	166	if (ll_xattr_cache_find(cache, xattr_name, &xattr) == 0) {
	167	list_del(&xattr->xe_list);
97903a26 JL	168	kfree(xattr->xe_name);
97903a26 JL	169	kfree(xattr->xe_value);
50d30362	170	kmem_cache_free(xattr_kmem, xattr);
7fc1f831 AP	171
	172	return 0;
	173	}
	174
	175	return -ENODATA;
	176	}
	177
	178	/**
	179	* This iterates cached extended attributes.
	180	*
	181	* Walk over cached attributes in @cache and
	182	* fill in @xld_buffer or only calculate buffer
	183	* size if @xld_buffer is NULL.
	184	*
	185	* \retval >= 0 buffer list size
	186	* \retval -ENODATA if the list cannot fit @xld_size buffer
	187	*/
	188	static int ll_xattr_cache_list(struct list_head *cache,
	189	char *xld_buffer,
	190	int xld_size)
	191	{
	192	struct ll_xattr_entry xattr, tmp;
	193	int xld_tail = 0;
	194
7fc1f831 AP	195	list_for_each_entry_safe(xattr, tmp, cache, xe_list) {
	196	CDEBUG(D_CACHE, "list: buffer=%p[%d] name=%s\n",
	197	xld_buffer, xld_tail, xattr->xe_name);
	198
	199	if (xld_buffer) {
	200	xld_size -= xattr->xe_namelen;
	201	if (xld_size < 0)
	202	break;
	203	memcpy(&xld_buffer[xld_tail],
	204	xattr->xe_name, xattr->xe_namelen);
	205	}
	206	xld_tail += xattr->xe_namelen;
	207	}
	208
	209	if (xld_size < 0)
	210	return -ERANGE;
	211
	212	return xld_tail;
	213	}
	214
	215	/**
	216	* Check if the xattr cache is initialized (filled).
	217	*
	218	* \retval 0 @cache is not initialized
	219	* \retval 1 @cache is initialized
	220	*/
2d95f10e	221	static int ll_xattr_cache_valid(struct ll_inode_info *lli)
7fc1f831 AP	222	{
	223	return !!(lli->lli_flags & LLIF_XATTR_CACHE);
	224	}
	225
	226	/**
	227	* This finalizes the xattr cache.
	228	*
	229	* Free all xattr memory. @lli is the inode info pointer.
	230	*
d0a0acc3	231	* \retval 0 no error occurred
7fc1f831 AP	232	*/
	233	static int ll_xattr_cache_destroy_locked(struct ll_inode_info *lli)
	234	{
	235
7fc1f831 AP	236	if (!ll_xattr_cache_valid(lli))
	237	return 0;
	238
	239	while (ll_xattr_cache_del(&lli->lli_xattrs, NULL) == 0)
	240	; /* empty loop */
	241	lli->lli_flags &= ~LLIF_XATTR_CACHE;
	242
	243	return 0;
	244	}
	245
	246	int ll_xattr_cache_destroy(struct inode *inode)
	247	{
	248	struct ll_inode_info *lli = ll_i2info(inode);
	249	int rc;
	250
7fc1f831 AP	251	down_write(&lli->lli_xattrs_list_rwsem);
	252	rc = ll_xattr_cache_destroy_locked(lli);
	253	up_write(&lli->lli_xattrs_list_rwsem);
	254
	255	return rc;
	256	}
	257
	258	/**
e93a3082	259	* Match or enqueue a PR lock.
7fc1f831 AP	260	*
	261	* Find or request an LDLM lock with xattr data.
	262	* Since LDLM does not provide API for atomic match_or_enqueue,
	263	* the function handles it with a separate enq lock.
	264	* If successful, the function exits with the list lock held.
	265	*
d0a0acc3	266	* \retval 0 no error occurred
7fc1f831 AP	267	* \retval -ENOMEM not enough memory
	268	*/
	269	static int ll_xattr_find_get_lock(struct inode *inode,
	270	struct lookup_intent *oit,
	271	struct ptlrpc_request **req)
	272	{
	273	ldlm_mode_t mode;
	274	struct lustre_handle lockh = { 0 };
	275	struct md_op_data *op_data;
	276	struct ll_inode_info *lli = ll_i2info(inode);
	277	struct ldlm_enqueue_info einfo = { .ei_type = LDLM_IBITS,
	278	.ei_mode = it_to_lock_mode(oit),
	279	.ei_cb_bl = ll_md_blocking_ast,
	280	.ei_cb_cp = ldlm_completion_ast };
	281	struct ll_sb_info *sbi = ll_i2sbi(inode);
	282	struct obd_export *exp = sbi->ll_md_exp;
	283	int rc;
	284
7fc1f831	285	mutex_lock(&lli->lli_xattrs_enq_lock);
d6abc59e LS	286	/* inode may have been shrunk and recreated, so data is gone, match lock
	287	* only when data exists. */
	288	if (ll_xattr_cache_valid(lli)) {
	289	/* Try matching first. */
	290	mode = ll_take_md_lock(inode, MDS_INODELOCK_XATTR, &lockh, 0,
	291	LCK_PR);
	292	if (mode != 0) {
	293	/* fake oit in mdc_revalidate_lock() manner */
	294	oit->d.lustre.it_lock_handle = lockh.cookie;
	295	oit->d.lustre.it_lock_mode = mode;
	296	goto out;
	297	}
7fc1f831 AP	298	}
	299
	300	/* Enqueue if the lock isn't cached locally. */
	301	op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
	302	LUSTRE_OPC_ANY, NULL);
	303	if (IS_ERR(op_data)) {
	304	mutex_unlock(&lli->lli_xattrs_enq_lock);
	305	return PTR_ERR(op_data);
	306	}
	307
e93a3082	308	op_data->op_valid = OBD_MD_FLXATTR \| OBD_MD_FLXATTRLS;
7fc1f831 AP	309
	310	rc = md_enqueue(exp, &einfo, oit, op_data, &lockh, NULL, 0, NULL, 0);
	311	ll_finish_md_op_data(op_data);
	312
	313	if (rc < 0) {
	314	CDEBUG(D_CACHE,
	315	"md_intent_lock failed with %d for fid "DFID"\n",
	316	rc, PFID(ll_inode2fid(inode)));
	317	mutex_unlock(&lli->lli_xattrs_enq_lock);
	318	return rc;
	319	}
	320
	321	req = (struct ptlrpc_request )oit->d.lustre.it_data;
	322	out:
	323	down_write(&lli->lli_xattrs_list_rwsem);
	324	mutex_unlock(&lli->lli_xattrs_enq_lock);
	325
	326	return 0;
	327	}
	328
	329	/**
	330	* Refill the xattr cache.
	331	*
	332	* Fetch and cache the whole of xattrs for @inode, acquiring
	333	* a read or a write xattr lock depending on operation in @oit.
	334	* Intent is dropped on exit unless the operation is setxattr.
	335	*
d0a0acc3	336	* \retval 0 no error occurred
7fc1f831 AP	337	* \retval -EPROTO network protocol error
	338	* \retval -ENOMEM not enough memory for the cache
	339	*/
	340	static int ll_xattr_cache_refill(struct inode inode, struct lookup_intent oit)
	341	{
	342	struct ll_sb_info *sbi = ll_i2sbi(inode);
	343	struct ptlrpc_request *req = NULL;
	344	const char xdata, xval, xtail, xvtail;
	345	struct ll_inode_info *lli = ll_i2info(inode);
	346	struct mdt_body *body;
	347	__u32 *xsizes;
f82ced5d	348	int rc, i;
7fc1f831	349
7fc1f831 AP	350	rc = ll_xattr_find_get_lock(inode, oit, &req);
7fc1f831 AP	351	if (rc)
34e1f2bb	352	goto out_no_unlock;
7fc1f831 AP	353
	354	/* Do we have the data at this point? */
	355	if (ll_xattr_cache_valid(lli)) {
	356	ll_stats_ops_tally(sbi, LPROC_LL_GETXATTR_HITS, 1);
34e1f2bb JL	357	rc = 0;
34e1f2bb JL	358	goto out_maybe_drop;
7fc1f831 AP	359	}
	360
	361	/* Matched but no cache? Cancelled on error by a parallel refill. */
	362	if (unlikely(req == NULL)) {
	363	CDEBUG(D_CACHE, "cancelled by a parallel getxattr\n");
34e1f2bb JL	364	rc = -EIO;
34e1f2bb JL	365	goto out_maybe_drop;
7fc1f831 AP	366	}
	367
	368	if (oit->d.lustre.it_status < 0) {
	369	CDEBUG(D_CACHE, "getxattr intent returned %d for fid "DFID"\n",
	370	oit->d.lustre.it_status, PFID(ll_inode2fid(inode)));
e93a3082 AP	371	rc = oit->d.lustre.it_status;
	372	/* xattr data is so large that we don't want to cache it */
	373	if (rc == -ERANGE)
	374	rc = -EAGAIN;
34e1f2bb	375	goto out_destroy;
7fc1f831 AP	376	}
	377
	378	body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
	379	if (body == NULL) {
	380	CERROR("no MDT BODY in the refill xattr reply\n");
34e1f2bb JL	381	rc = -EPROTO;
34e1f2bb JL	382	goto out_destroy;
7fc1f831 AP	383	}
	384	/* do not need swab xattr data */
	385	xdata = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA,
	386	body->eadatasize);
	387	xval = req_capsule_server_sized_get(&req->rq_pill, &RMF_EAVALS,
	388	body->aclsize);
	389	xsizes = req_capsule_server_sized_get(&req->rq_pill, &RMF_EAVALS_LENS,
	390	body->max_mdsize * sizeof(__u32));
	391	if (xdata == NULL \|\| xval == NULL \|\| xsizes == NULL) {
	392	CERROR("wrong setxattr reply\n");
34e1f2bb JL	393	rc = -EPROTO;
34e1f2bb JL	394	goto out_destroy;
7fc1f831 AP	395	}
	396
	397	xtail = xdata + body->eadatasize;
	398	xvtail = xval + body->aclsize;
	399
	400	CDEBUG(D_CACHE, "caching: xdata=%p xtail=%p\n", xdata, xtail);
	401
	402	ll_xattr_cache_init(lli);
	403
	404	for (i = 0; i < body->max_mdsize; i++) {
	405	CDEBUG(D_CACHE, "caching [%s]=%.s\n", xdata, xsizes, xval);
	406	/* Perform consistency checks: attr names and vals in pill */
	407	if (memchr(xdata, 0, xtail - xdata) == NULL) {
	408	CERROR("xattr protocol violation (names are broken)\n");
	409	rc = -EPROTO;
	410	} else if (xval + *xsizes > xvtail) {
	411	CERROR("xattr protocol violation (vals are broken)\n");
	412	rc = -EPROTO;
	413	} else if (OBD_FAIL_CHECK(OBD_FAIL_LLITE_XATTR_ENOMEM)) {
	414	rc = -ENOMEM;
e93a3082 AP	415	} else if (!strcmp(xdata, XATTR_NAME_ACL_ACCESS)) {
	416	/* Filter out ACL ACCESS since it's cached separately */
	417	CDEBUG(D_CACHE, "not caching %s\n",
	418	XATTR_NAME_ACL_ACCESS);
	419	rc = 0;
7fc1f831 AP	420	} else {
	421	rc = ll_xattr_cache_add(&lli->lli_xattrs, xdata, xval,
	422	*xsizes);
	423	}
	424	if (rc < 0) {
	425	ll_xattr_cache_destroy_locked(lli);
34e1f2bb	426	goto out_destroy;
7fc1f831 AP	427	}
	428	xdata += strlen(xdata) + 1;
	429	xval += *xsizes;
	430	xsizes++;
	431	}
	432
	433	if (xdata != xtail \|\| xval != xvtail)
	434	CERROR("a hole in xattr data\n");
	435
	436	ll_set_lock_data(sbi->ll_md_exp, inode, oit, NULL);
	437
34e1f2bb	438	goto out_maybe_drop;
7fc1f831	439	out_maybe_drop:
e93a3082	440
7fc1f831 AP	441	ll_intent_drop_lock(oit);
	442
	443	if (rc != 0)
	444	up_write(&lli->lli_xattrs_list_rwsem);
	445	out_no_unlock:
	446	ptlrpc_req_finished(req);
	447
	448	return rc;
	449
	450	out_destroy:
	451	up_write(&lli->lli_xattrs_list_rwsem);
	452
	453	ldlm_lock_decref_and_cancel((struct lustre_handle *)
	454	&oit->d.lustre.it_lock_handle,
	455	oit->d.lustre.it_lock_mode);
	456
	457	goto out_no_unlock;
	458	}
	459
	460	/**
	461	* Get an xattr value or list xattrs using the write-through cache.
	462	*
	463	* Get the xattr value (@valid has OBD_MD_FLXATTR set) of @name or
	464	* list xattr names (@valid has OBD_MD_FLXATTRLS set) for @inode.
	465	* The resulting value/list is stored in @buffer if the former
	466	* is not larger than @size.
	467	*
d0a0acc3	468	* \retval 0 no error occurred
7fc1f831 AP	469	* \retval -EPROTO network protocol error
	470	* \retval -ENOMEM not enough memory for the cache
	471	* \retval -ERANGE the buffer is not large enough
	472	* \retval -ENODATA no such attr or the list is empty
	473	*/
	474	int ll_xattr_cache_get(struct inode *inode,
	475	const char *name,
	476	char *buffer,
	477	size_t size,
	478	__u64 valid)
	479	{
	480	struct lookup_intent oit = { .it_op = IT_GETXATTR };
	481	struct ll_inode_info *lli = ll_i2info(inode);
	482	int rc = 0;
	483
7fc1f831 AP	484	LASSERT(!!(valid & OBD_MD_FLXATTR) ^ !!(valid & OBD_MD_FLXATTRLS));
	485
	486	down_read(&lli->lli_xattrs_list_rwsem);
	487	if (!ll_xattr_cache_valid(lli)) {
	488	up_read(&lli->lli_xattrs_list_rwsem);
	489	rc = ll_xattr_cache_refill(inode, &oit);
	490	if (rc)
	491	return rc;
	492	downgrade_write(&lli->lli_xattrs_list_rwsem);
	493	} else {
	494	ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_GETXATTR_HITS, 1);
	495	}
	496
	497	if (valid & OBD_MD_FLXATTR) {
	498	struct ll_xattr_entry *xattr;
	499
	500	rc = ll_xattr_cache_find(&lli->lli_xattrs, name, &xattr);
	501	if (rc == 0) {
	502	rc = xattr->xe_vallen;
	503	/* zero size means we are only requested size in rc */
	504	if (size != 0) {
	505	if (size >= xattr->xe_vallen)
	506	memcpy(buffer, xattr->xe_value,
	507	xattr->xe_vallen);
	508	else
	509	rc = -ERANGE;
	510	}
	511	}
	512	} else if (valid & OBD_MD_FLXATTRLS) {
	513	rc = ll_xattr_cache_list(&lli->lli_xattrs,
	514	size ? buffer : NULL, size);
	515	}
	516
34e1f2bb	517	goto out;
7fc1f831 AP	518	out:
	519	up_read(&lli->lli_xattrs_list_rwsem);
	520
	521	return rc;
	522	}