1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/llite/llite_lib.c
38 * Lustre Light Super operations
41 #define DEBUG_SUBSYSTEM S_LLITE
43 #include <linux/module.h>
44 #include <linux/types.h>
45 #include <linux/version.h>
48 #include <lustre_lite.h>
49 #include <lustre_ha.h>
50 #include <lustre_dlm.h>
51 #include <lprocfs_status.h>
52 #include <lustre_disk.h>
53 #include <lustre_param.h>
54 #include <lustre_log.h>
55 #include <cl_object.h>
56 #include <obd_cksum.h>
57 #include "llite_internal.h"
59 cfs_mem_cache_t *ll_file_data_slab;
61 CFS_LIST_HEAD(ll_super_blocks);
62 cfs_spinlock_t ll_sb_lock = CFS_SPIN_LOCK_UNLOCKED;
64 #ifndef MS_HAS_NEW_AOPS
65 extern struct address_space_operations ll_aops;
66 extern struct address_space_operations ll_dir_aops;
68 extern struct address_space_operations_ext ll_aops;
69 extern struct address_space_operations_ext ll_dir_aops;
73 #define log2(n) cfs_ffz(~(n))
76 static struct ll_sb_info *ll_init_sbi(void)
78 struct ll_sb_info *sbi = NULL;
85 OBD_ALLOC(sbi, sizeof(*sbi));
89 cfs_spin_lock_init(&sbi->ll_lock);
90 cfs_init_mutex(&sbi->ll_lco.lco_lock);
91 cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
92 cfs_spin_lock_init(&sbi->ll_process_lock);
93 sbi->ll_rw_stats_on = 0;
96 pages = si.totalram - si.totalhigh;
97 if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
98 #ifdef HAVE_BGL_SUPPORT
99 sbi->ll_async_page_max = pages / 4;
101 sbi->ll_async_page_max = pages / 2;
104 sbi->ll_async_page_max = (pages / 4) * 3;
107 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
108 SBI_DEFAULT_READAHEAD_MAX);
109 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
110 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
111 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
112 CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
113 CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
115 ll_generate_random_uuid(uuid);
116 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
117 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
119 cfs_spin_lock(&ll_sb_lock);
120 cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
121 cfs_spin_unlock(&ll_sb_lock);
123 #ifdef ENABLE_CHECKSUM
124 sbi->ll_flags |= LL_SBI_CHECKSUM;
127 #ifdef HAVE_LRU_RESIZE_SUPPORT
128 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
131 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
132 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
134 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
138 /* metadata statahead is enabled by default */
139 sbi->ll_sa_max = LL_SA_RPC_DEF;
140 atomic_set(&sbi->ll_sa_total, 0);
141 atomic_set(&sbi->ll_sa_wrong, 0);
146 void ll_free_sbi(struct super_block *sb)
148 struct ll_sb_info *sbi = ll_s2sbi(sb);
152 cfs_spin_lock(&ll_sb_lock);
153 cfs_list_del(&sbi->ll_list);
154 cfs_spin_unlock(&ll_sb_lock);
155 OBD_FREE(sbi, sizeof(*sbi));
160 static struct dentry_operations ll_d_root_ops = {
161 .d_compare = ll_dcompare,
162 .d_revalidate = ll_revalidate_nd,
165 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
166 struct vfsmount *mnt)
168 struct inode *root = 0;
169 struct ll_sb_info *sbi = ll_s2sbi(sb);
170 struct obd_device *obd;
171 struct obd_capa *oc = NULL;
172 struct obd_statfs osfs;
173 struct ptlrpc_request *request = NULL;
174 struct obd_connect_data *data = NULL;
175 struct obd_uuid *uuid;
176 struct md_op_data *op_data;
177 struct lustre_md lmd;
179 int size, err, checksum;
182 obd = class_name2obd(md);
184 CERROR("MD %s: not setup or attached\n", md);
192 if (proc_lustre_fs_root) {
193 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
196 CERROR("could not register mount in /proc/fs/lustre\n");
199 /* indicate the features supported by this client */
200 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
201 OBD_CONNECT_JOIN | OBD_CONNECT_ATTRFID |
202 OBD_CONNECT_VERSION | OBD_CONNECT_MDS_CAPA |
203 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
204 OBD_CONNECT_FID | OBD_CONNECT_AT |
205 OBD_CONNECT_LOV_V3 | OBD_CONNECT_RMT_CLIENT |
206 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
207 OBD_CONNECT_64BITHASH;
209 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
210 data->ocd_connect_flags |= OBD_CONNECT_SOM;
212 #ifdef HAVE_LRU_RESIZE_SUPPORT
213 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
214 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
216 #ifdef CONFIG_FS_POSIX_ACL
217 data->ocd_connect_flags |= OBD_CONNECT_ACL;
219 data->ocd_ibits_known = MDS_INODELOCK_FULL;
220 data->ocd_version = LUSTRE_VERSION_CODE;
222 if (sb->s_flags & MS_RDONLY)
223 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
224 if (sbi->ll_flags & LL_SBI_USER_XATTR)
225 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
227 #ifdef HAVE_MS_FLOCK_LOCK
228 /* force vfs to use lustre handler for flock() calls - bug 10743 */
229 sb->s_flags |= MS_FLOCK_LOCK;
231 #ifdef MS_HAS_NEW_AOPS
232 sb->s_flags |= MS_HAS_NEW_AOPS;
235 if (sbi->ll_flags & LL_SBI_FLOCK)
236 sbi->ll_fop = &ll_file_operations_flock;
237 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
238 sbi->ll_fop = &ll_file_operations;
240 sbi->ll_fop = &ll_file_operations_noflock;
243 data->ocd_connect_flags |= OBD_CONNECT_REAL;
244 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
245 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
247 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
249 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
250 "recovery, of which this client is not a "
251 "part. Please wait for recovery to complete,"
252 " abort, or time out.\n", md);
255 CERROR("cannot connect to %s: rc = %d\n", md, err);
259 err = obd_fid_init(sbi->ll_md_exp);
261 CERROR("Can't init metadata layer FID infrastructure, "
266 err = obd_statfs(obd, &osfs,
267 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
269 GOTO(out_md_fid, err);
271 size = sizeof(*data);
272 err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
273 KEY_CONN_DATA, &size, data, NULL);
275 CERROR("Get connect data failed: %d \n", err);
279 LASSERT(osfs.os_bsize);
280 sb->s_blocksize = osfs.os_bsize;
281 sb->s_blocksize_bits = log2(osfs.os_bsize);
282 sb->s_magic = LL_SUPER_MAGIC;
284 /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
285 * retval = in_file->f_op->sendfile(...);
287 * retval = -EOVERFLOW;
289 * it will check if *ppos is greater than max. However, max equals to
290 * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
291 * has been defined as a signed long long integer in linux kernel. */
292 #if BITS_PER_LONG == 64
293 sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
295 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
297 sbi->ll_namelen = osfs.os_namelen;
298 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
300 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
301 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
302 LCONSOLE_INFO("Disabling user_xattr feature because "
303 "it is not supported on the server\n");
304 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
307 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
309 sb->s_flags |= MS_POSIXACL;
311 sbi->ll_flags |= LL_SBI_ACL;
313 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
315 sb->s_flags &= ~MS_POSIXACL;
317 sbi->ll_flags &= ~LL_SBI_ACL;
320 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
321 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
322 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
323 LCONSOLE_INFO("client is set as remote by default.\n");
326 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
327 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
328 LCONSOLE_INFO("client claims to be remote, but server "
329 "rejected, forced to be local.\n");
333 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
334 LCONSOLE_INFO("client enabled MDS capability!\n");
335 sbi->ll_flags |= LL_SBI_MDS_CAPA;
338 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
339 LCONSOLE_INFO("client enabled OSS capability!\n");
340 sbi->ll_flags |= LL_SBI_OSS_CAPA;
343 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
344 sbi->ll_flags |= LL_SBI_64BIT_HASH;
346 obd = class_name2obd(dt);
348 CERROR("DT %s: not setup or attached\n", dt);
349 GOTO(out_md_fid, err = -ENODEV);
352 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
353 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
354 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
355 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
356 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
357 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
358 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH;
360 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
361 data->ocd_connect_flags |= OBD_CONNECT_SOM;
363 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
364 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
365 * disabled by default, because it can still be enabled on the
366 * fly via /proc. As a consequence, we still need to come to an
367 * agreement on the supported algorithms at connect time */
368 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
370 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
371 data->ocd_cksum_types = OBD_CKSUM_ADLER;
373 /* send the list of supported checksum types */
374 data->ocd_cksum_types = OBD_CKSUM_ALL;
377 #ifdef HAVE_LRU_RESIZE_SUPPORT
378 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
380 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
381 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
383 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
384 "ocd_grant: %d\n", data->ocd_connect_flags,
385 data->ocd_version, data->ocd_grant);
387 obd->obd_upcall.onu_owner = &sbi->ll_lco;
388 obd->obd_upcall.onu_upcall = cl_ocd_update;
389 data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
391 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
393 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
394 "recovery, of which this client is not a "
395 "part. Please wait for recovery to "
396 "complete, abort, or time out.\n", dt);
397 GOTO(out_md_fid, err);
399 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
400 GOTO(out_md_fid, err);
403 err = obd_fid_init(sbi->ll_dt_exp);
405 CERROR("Can't init data layer FID infrastructure, "
410 cfs_mutex_down(&sbi->ll_lco.lco_lock);
411 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
412 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
413 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
414 cfs_mutex_up(&sbi->ll_lco.lco_lock);
416 fid_zero(&sbi->ll_root_fid);
417 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
419 CERROR("cannot mds_connect: rc = %d\n", err);
420 GOTO(out_lock_cn_cb, err);
422 if (!fid_is_sane(&sbi->ll_root_fid)) {
423 CERROR("Invalid root fid during mount\n");
424 GOTO(out_lock_cn_cb, err = -EINVAL);
426 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
428 sb->s_op = &lustre_super_operations;
429 #if THREAD_SIZE >= 8192
430 sb->s_export_op = &lustre_export_operations;
434 * XXX: move this to after cbd setup? */
435 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
436 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
437 valid |= OBD_MD_FLRMTPERM;
438 else if (sbi->ll_flags & LL_SBI_ACL)
439 valid |= OBD_MD_FLACL;
441 OBD_ALLOC_PTR(op_data);
443 GOTO(out_lock_cn_cb, err = -ENOMEM);
445 op_data->op_fid1 = sbi->ll_root_fid;
446 op_data->op_mode = 0;
447 op_data->op_capa1 = oc;
448 op_data->op_valid = valid;
450 err = md_getattr(sbi->ll_md_exp, op_data, &request);
453 OBD_FREE_PTR(op_data);
455 CERROR("md_getattr failed for root: rc = %d\n", err);
456 GOTO(out_lock_cn_cb, err);
458 memset(&lmd, 0, sizeof(lmd));
459 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
460 sbi->ll_md_exp, &lmd);
462 CERROR("failed to understand root inode md: rc = %d\n", err);
463 ptlrpc_req_finished (request);
464 GOTO(out_lock_cn_cb, err);
467 LASSERT(fid_is_sane(&sbi->ll_root_fid));
468 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
469 md_free_lustre_md(sbi->ll_md_exp, &lmd);
470 ptlrpc_req_finished(request);
472 if (root == NULL || IS_ERR(root)) {
474 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
475 #ifdef CONFIG_FS_POSIX_ACL
477 posix_acl_release(lmd.posix_acl);
478 lmd.posix_acl = NULL;
481 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
483 CERROR("lustre_lite: bad iget4 for root\n");
487 err = ll_close_thread_start(&sbi->ll_lcq);
489 CERROR("cannot start close thread: rc %d\n", err);
493 #ifdef CONFIG_FS_POSIX_ACL
494 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
495 rct_init(&sbi->ll_rct);
496 et_init(&sbi->ll_et);
500 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
501 err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
502 KEY_CHECKSUM, sizeof(checksum), &checksum,
506 sb->s_root = d_alloc_root(root);
508 OBD_FREE(data, sizeof(*data));
510 sb->s_root->d_op = &ll_d_root_ops;
512 sbi->ll_sdev_orig = sb->s_dev;
514 /* We set sb->s_dev equal on all lustre clients in order to support
515 * NFS export clustering. NFSD requires that the FSID be the same
517 /* s_dev is also used in lt_compare() to compare two fs, but that is
518 * only a node-local comparison. */
519 uuid = obd_get_uuid(sbi->ll_md_exp);
521 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
529 obd_fid_fini(sbi->ll_dt_exp);
531 obd_disconnect(sbi->ll_dt_exp);
532 sbi->ll_dt_exp = NULL;
534 obd_fid_fini(sbi->ll_md_exp);
536 obd_disconnect(sbi->ll_md_exp);
537 sbi->ll_md_exp = NULL;
541 lprocfs_unregister_mountpoint(sbi);
545 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
549 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
551 rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
552 KEY_MAX_EASIZE, &size, lmmsize, NULL);
554 CERROR("Get max mdsize error rc %d \n", rc);
559 void ll_dump_inode(struct inode *inode)
561 struct list_head *tmp;
562 int dentry_count = 0;
564 LASSERT(inode != NULL);
566 list_for_each(tmp, &inode->i_dentry)
569 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
570 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
571 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
574 void lustre_dump_dentry(struct dentry *dentry, int recur)
576 struct list_head *tmp;
579 LASSERT(dentry != NULL);
581 list_for_each(tmp, &dentry->d_subdirs)
584 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
585 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
586 dentry->d_name.len, dentry->d_name.name,
587 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
588 dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
589 dentry->d_flags, dentry->d_fsdata, subdirs);
590 if (dentry->d_inode != NULL)
591 ll_dump_inode(dentry->d_inode);
596 list_for_each(tmp, &dentry->d_subdirs) {
597 struct dentry *d = list_entry(tmp, struct dentry, d_child);
598 lustre_dump_dentry(d, recur - 1);
602 void client_common_put_super(struct super_block *sb)
604 struct ll_sb_info *sbi = ll_s2sbi(sb);
607 #ifdef CONFIG_FS_POSIX_ACL
608 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
609 et_fini(&sbi->ll_et);
610 rct_fini(&sbi->ll_rct);
614 obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
616 ll_close_thread_shutdown(sbi->ll_lcq);
620 cfs_list_del(&sbi->ll_conn_chain);
622 obd_fid_fini(sbi->ll_dt_exp);
623 obd_disconnect(sbi->ll_dt_exp);
624 sbi->ll_dt_exp = NULL;
626 lprocfs_unregister_mountpoint(sbi);
628 obd_fid_fini(sbi->ll_md_exp);
629 obd_disconnect(sbi->ll_md_exp);
630 sbi->ll_md_exp = NULL;
635 void ll_kill_super(struct super_block *sb)
637 struct ll_sb_info *sbi;
642 if (!(sb->s_flags & MS_ACTIVE))
646 /* we need restore s_dev from changed for clustred NFS before put_super
647 * because new kernels have cached s_dev and change sb->s_dev in
648 * put_super not affected real removing devices */
650 sb->s_dev = sbi->ll_sdev_orig;
654 char *ll_read_opt(const char *opt, char *data)
660 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
661 if (strncmp(opt, data, strlen(opt)))
663 if ((value = strchr(data, '=')) == NULL)
667 OBD_ALLOC(retval, strlen(value) + 1);
669 CERROR("out of memory!\n");
673 memcpy(retval, value, strlen(value)+1);
674 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
678 static inline int ll_set_opt(const char *opt, char *data, int fl)
680 if (strncmp(opt, data, strlen(opt)) != 0)
686 /* non-client-specific mount options are parsed in lmd_parse */
687 static int ll_options(char *options, int *flags)
690 char *s1 = options, *s2;
696 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
699 CDEBUG(D_SUPER, "next opt=%s\n", s1);
700 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
705 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
710 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
715 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
720 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
725 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
730 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,5,50,0)
731 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
733 /* Ignore deprecated mount option. The client will
734 * always try to mount with ACL support, whether this
735 * is used depends on whether server supports it. */
736 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
737 "mount option 'acl'.\n");
740 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
742 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
743 "mount option 'noacl'.\n");
747 #warning "{no}acl options have been deprecated since 1.8, please remove them"
749 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
755 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
760 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
765 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
770 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
775 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
780 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
785 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
790 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
796 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
802 s2 = strchr(s1, ',');
810 void ll_lli_init(struct ll_inode_info *lli)
812 lli->lli_inode_magic = LLI_INODE_MAGIC;
813 cfs_sema_init(&lli->lli_size_sem, 1);
814 cfs_sema_init(&lli->lli_write_sem, 1);
815 cfs_init_rwsem(&lli->lli_trunc_sem);
817 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
818 cfs_spin_lock_init(&lli->lli_lock);
819 CFS_INIT_LIST_HEAD(&lli->lli_close_list);
820 lli->lli_inode_magic = LLI_INODE_MAGIC;
821 cfs_sema_init(&lli->lli_och_sem, 1);
822 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
823 lli->lli_mds_exec_och = NULL;
824 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
825 lli->lli_open_fd_exec_count = 0;
826 CFS_INIT_LIST_HEAD(&lli->lli_dead_list);
827 lli->lli_remote_perms = NULL;
828 lli->lli_rmtperm_utime = 0;
829 cfs_sema_init(&lli->lli_rmtperm_sem, 1);
830 CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
831 cfs_spin_lock_init(&lli->lli_sa_lock);
832 cfs_sema_init(&lli->lli_readdir_sem, 1);
833 fid_zero(&lli->lli_pfid);
836 static inline int ll_bdi_register(struct backing_dev_info *bdi)
838 #ifdef HAVE_BDI_REGISTER
839 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
841 bdi->name = "lustre";
842 return bdi_register(bdi, NULL, "lustre-%d",
843 atomic_inc_return(&ll_bdi_num));
849 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
851 struct lustre_profile *lprof;
852 struct lustre_sb_info *lsi = s2lsi(sb);
853 struct ll_sb_info *sbi;
854 char *dt = NULL, *md = NULL;
855 char *profilenm = get_profile_name(sb);
856 struct config_llog_instance cfg = {0, };
857 char ll_instance[sizeof(sb) * 2 + 1];
861 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
865 /* client additional sb info */
866 lsi->lsi_llsbi = sbi = ll_init_sbi();
868 cfs_module_put(THIS_MODULE);
872 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
876 err = ll_bdi_init(&lsi->lsi_bdi);
879 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
880 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
881 err = ll_bdi_register(&lsi->lsi_bdi);
886 sb->s_bdi = &lsi->lsi_bdi;
889 /* Generate a string unique to this super, in case some joker tries
890 to mount the same fs at two mount points.
891 Use the address of the super itself.*/
892 sprintf(ll_instance, "%p", sb);
893 cfg.cfg_instance = ll_instance;
894 cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
896 /* set up client obds */
897 err = lustre_process_log(sb, profilenm, &cfg);
899 CERROR("Unable to process log: %d\n", err);
903 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
904 lprof = class_get_profile(profilenm);
906 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
907 " read from the MGS. Does that filesystem "
908 "exist?\n", profilenm);
909 GOTO(out_free, err = -EINVAL);
911 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
912 lprof->lp_md, lprof->lp_dt);
914 OBD_ALLOC(dt, strlen(lprof->lp_dt) +
915 strlen(ll_instance) + 2);
917 GOTO(out_free, err = -ENOMEM);
918 sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
920 OBD_ALLOC(md, strlen(lprof->lp_md) +
921 strlen(ll_instance) + 2);
923 GOTO(out_free, err = -ENOMEM);
924 sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
926 /* connections, registrations, sb setup */
927 err = client_common_fill_super(sb, md, dt, mnt);
931 OBD_FREE(md, strlen(md) + 1);
933 OBD_FREE(dt, strlen(dt) + 1);
937 LCONSOLE_WARN("Client %s has started\n", profilenm);
940 } /* ll_fill_super */
943 void lu_context_keys_dump(void);
945 void ll_put_super(struct super_block *sb)
947 struct config_llog_instance cfg;
948 char ll_instance[sizeof(sb) * 2 + 1];
949 struct obd_device *obd;
950 struct lustre_sb_info *lsi = s2lsi(sb);
951 struct ll_sb_info *sbi = ll_s2sbi(sb);
952 char *profilenm = get_profile_name(sb);
956 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
958 ll_print_capa_stat(sbi);
960 sprintf(ll_instance, "%p", sb);
961 cfg.cfg_instance = ll_instance;
962 lustre_end_log(sb, NULL, &cfg);
964 if (sbi->ll_md_exp) {
965 obd = class_exp2obd(sbi->ll_md_exp);
967 force = obd->obd_force;
970 /* We need to set force before the lov_disconnect in
971 lustre_common_put_super, since l_d cleans up osc's as well. */
974 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
976 obd->obd_force = force;
981 /* Only if client_common_fill_super succeeded */
982 client_common_put_super(sb);
986 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
987 class_manual_cleanup(obd);
991 class_del_profile(profilenm);
993 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
994 ll_bdi_destroy(&lsi->lsi_bdi);
995 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
999 lsi->lsi_llsbi = NULL;
1001 lustre_common_put_super(sb);
1003 cl_env_cache_purge(~0);
1005 LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1007 cfs_module_put(THIS_MODULE);
1010 } /* client_put_super */
1012 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1014 struct inode *inode = NULL;
1015 /* NOTE: we depend on atomic igrab() -bzzz */
1016 lock_res_and_lock(lock);
1017 if (lock->l_ast_data) {
1018 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1019 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1020 inode = igrab(lock->l_ast_data);
1022 inode = lock->l_ast_data;
1023 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1024 D_WARNING, lock, "l_ast_data %p is "
1025 "bogus: magic %08x", lock->l_ast_data,
1026 lli->lli_inode_magic);
1030 unlock_res_and_lock(lock);
1034 static int null_if_equal(struct ldlm_lock *lock, void *data)
1036 if (data == lock->l_ast_data) {
1037 lock->l_ast_data = NULL;
1039 if (lock->l_req_mode != lock->l_granted_mode)
1040 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1043 return LDLM_ITER_CONTINUE;
1046 void ll_clear_inode(struct inode *inode)
1048 struct ll_inode_info *lli = ll_i2info(inode);
1049 struct ll_sb_info *sbi = ll_i2sbi(inode);
1052 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1053 inode->i_generation, inode);
1055 if (S_ISDIR(inode->i_mode)) {
1056 /* these should have been cleared in ll_file_release */
1057 LASSERT(lli->lli_sai == NULL);
1058 LASSERT(lli->lli_opendir_key == NULL);
1059 LASSERT(lli->lli_opendir_pid == 0);
1062 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1063 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1064 null_if_equal, inode);
1066 LASSERT(!lli->lli_open_fd_write_count);
1067 LASSERT(!lli->lli_open_fd_read_count);
1068 LASSERT(!lli->lli_open_fd_exec_count);
1070 if (lli->lli_mds_write_och)
1071 ll_md_real_close(inode, FMODE_WRITE);
1072 if (lli->lli_mds_exec_och)
1073 ll_md_real_close(inode, FMODE_EXEC);
1074 if (lli->lli_mds_read_och)
1075 ll_md_real_close(inode, FMODE_READ);
1077 if (lli->lli_symlink_name) {
1078 OBD_FREE(lli->lli_symlink_name,
1079 strlen(lli->lli_symlink_name) + 1);
1080 lli->lli_symlink_name = NULL;
1083 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1084 LASSERT(lli->lli_posix_acl == NULL);
1085 if (lli->lli_remote_perms) {
1086 free_rmtperm_hash(lli->lli_remote_perms);
1087 lli->lli_remote_perms = NULL;
1090 #ifdef CONFIG_FS_POSIX_ACL
1091 else if (lli->lli_posix_acl) {
1092 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1093 LASSERT(lli->lli_remote_perms == NULL);
1094 posix_acl_release(lli->lli_posix_acl);
1095 lli->lli_posix_acl = NULL;
1098 lli->lli_inode_magic = LLI_INODE_DEAD;
1100 ll_clear_inode_capas(inode);
1102 * XXX This has to be done before lsm is freed below, because
1103 * cl_object still uses inode lsm.
1105 cl_inode_fini(inode);
1108 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1109 lli->lli_smd = NULL;
1116 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1117 struct md_open_data **mod)
1119 struct lustre_md md;
1120 struct ll_sb_info *sbi = ll_i2sbi(inode);
1121 struct ptlrpc_request *request = NULL;
1125 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1126 LUSTRE_OPC_ANY, NULL);
1127 if (IS_ERR(op_data))
1128 RETURN(PTR_ERR(op_data));
1130 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1133 ptlrpc_req_finished(request);
1134 if (rc == -ENOENT) {
1136 /* Unlinked special device node? Or just a race?
1137 * Pretend we done everything. */
1138 if (!S_ISREG(inode->i_mode) &&
1139 !S_ISDIR(inode->i_mode))
1140 rc = inode_setattr(inode, &op_data->op_attr);
1141 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1142 CERROR("md_setattr fails: rc = %d\n", rc);
1147 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1148 sbi->ll_md_exp, &md);
1150 ptlrpc_req_finished(request);
1154 /* We call inode_setattr to adjust timestamps.
1155 * If there is at least some data in file, we cleared ATTR_SIZE
1156 * above to avoid invoking vmtruncate, otherwise it is important
1157 * to call vmtruncate in inode_setattr to update inode->i_size
1159 rc = inode_setattr(inode, &op_data->op_attr);
1161 /* Extract epoch data if obtained. */
1162 op_data->op_handle = md.body->handle;
1163 op_data->op_ioepoch = md.body->ioepoch;
1165 ll_update_inode(inode, &md);
1166 ptlrpc_req_finished(request);
1171 /* Close IO epoch and send Size-on-MDS attribute update. */
1172 static int ll_setattr_done_writing(struct inode *inode,
1173 struct md_op_data *op_data,
1174 struct md_open_data *mod)
1176 struct ll_inode_info *lli = ll_i2info(inode);
1180 LASSERT(op_data != NULL);
1181 if (!S_ISREG(inode->i_mode))
1184 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1185 op_data->op_ioepoch, PFID(&lli->lli_fid));
1187 op_data->op_flags = MF_EPOCH_CLOSE;
1188 ll_done_writing_attr(inode, op_data);
1189 ll_pack_inode2opdata(inode, op_data, NULL);
1191 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1192 if (rc == -EAGAIN) {
1193 /* MDS has instructed us to obtain Size-on-MDS attribute
1194 * from OSTs and send setattr to back to MDS. */
1195 rc = ll_som_update(inode, op_data);
1197 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1203 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1205 struct obd_capa *capa;
1208 if (attr->ia_valid & ATTR_SIZE)
1209 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1211 capa = ll_mdscapa_get(inode);
1213 rc = cl_setattr_ost(inode, attr, capa);
1215 if (attr->ia_valid & ATTR_SIZE)
1216 ll_truncate_free_capa(capa);
1223 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1224 * object(s) determine the file size and mtime. Otherwise, the MDS will
1225 * keep these values until such a time that objects are allocated for it.
1226 * We do the MDS operations first, as it is checking permissions for us.
1227 * We don't to the MDS RPC if there is nothing that we want to store there,
1228 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1229 * going to do an RPC anyways.
1231 * If we are doing a truncate, we will send the mtime and ctime updates
1232 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1233 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1236 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1238 struct ll_inode_info *lli = ll_i2info(inode);
1239 struct lov_stripe_md *lsm = lli->lli_smd;
1240 struct md_op_data *op_data = NULL;
1241 struct md_open_data *mod = NULL;
1242 int ia_valid = attr->ia_valid;
1243 int rc = 0, rc1 = 0;
1246 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1248 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1250 if (ia_valid & ATTR_SIZE) {
1251 if (attr->ia_size > ll_file_maxbytes(inode)) {
1252 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1253 attr->ia_size, ll_file_maxbytes(inode));
1257 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1260 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1261 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1262 if (cfs_curproc_fsuid() != inode->i_uid &&
1263 !cfs_capable(CFS_CAP_FOWNER))
1267 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1268 if (attr->ia_valid & ATTR_CTIME) {
1269 attr->ia_ctime = CFS_CURRENT_TIME;
1270 attr->ia_valid |= ATTR_CTIME_SET;
1272 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1273 attr->ia_atime = CFS_CURRENT_TIME;
1274 attr->ia_valid |= ATTR_ATIME_SET;
1276 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1277 attr->ia_mtime = CFS_CURRENT_TIME;
1278 attr->ia_valid |= ATTR_MTIME_SET;
1281 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1282 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1283 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1284 cfs_time_current_sec());
1286 /* NB: ATTR_SIZE will only be set after this point if the size
1287 * resides on the MDS, ie, this file has no objects. */
1289 attr->ia_valid &= ~ATTR_SIZE;
1291 /* We always do an MDS RPC, even if we're only changing the size;
1292 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1294 OBD_ALLOC_PTR(op_data);
1295 if (op_data == NULL)
1298 UNLOCK_INODE_MUTEX(inode);
1299 if (ia_valid & ATTR_SIZE)
1300 UP_WRITE_I_ALLOC_SEM(inode);
1301 cfs_down_write(&lli->lli_trunc_sem);
1302 LOCK_INODE_MUTEX(inode);
1303 if (ia_valid & ATTR_SIZE)
1304 DOWN_WRITE_I_ALLOC_SEM(inode);
1306 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1308 /* Open epoch for truncate. */
1309 if (exp_connect_som(ll_i2mdexp(inode)) &&
1310 (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1311 op_data->op_flags = MF_EPOCH_OPEN;
1313 rc = ll_md_setattr(inode, op_data, &mod);
1317 ll_ioepoch_open(lli, op_data->op_ioepoch);
1318 if (!lsm || !S_ISREG(inode->i_mode)) {
1319 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1323 if (ia_valid & ATTR_SIZE)
1324 attr->ia_valid |= ATTR_SIZE;
1325 if (ia_valid & (ATTR_SIZE |
1326 ATTR_ATIME | ATTR_ATIME_SET |
1327 ATTR_MTIME | ATTR_MTIME_SET))
1328 /* on truncate and utimes send attributes to osts, setting
1329 * mtime/atime to past will be performed under PW 0:EOF extent
1330 * lock (new_size:EOF for truncate)
1331 * it may seem excessive to send mtime/atime updates to osts
1332 * when not setting times to past, but it is necessary due to
1333 * possible time de-synchronization */
1334 rc = ll_setattr_ost(inode, attr);
1338 if (op_data->op_ioepoch)
1339 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1340 ll_finish_md_op_data(op_data);
1342 cfs_up_write(&lli->lli_trunc_sem);
1343 return rc ? rc : rc1;
1346 int ll_setattr(struct dentry *de, struct iattr *attr)
1348 int mode = de->d_inode->i_mode;
1350 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1351 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1352 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1354 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1355 (ATTR_SIZE|ATTR_MODE)) &&
1356 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1357 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1358 !(attr->ia_mode & S_ISGID))))
1359 attr->ia_valid |= ATTR_FORCE;
1361 if ((mode & S_ISUID) &&
1362 !(attr->ia_mode & S_ISUID) &&
1363 !(attr->ia_valid & ATTR_KILL_SUID))
1364 attr->ia_valid |= ATTR_KILL_SUID;
1366 if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1367 !(attr->ia_mode & S_ISGID) &&
1368 !(attr->ia_valid & ATTR_KILL_SGID))
1369 attr->ia_valid |= ATTR_KILL_SGID;
1371 return ll_setattr_raw(de->d_inode, attr);
1374 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1375 __u64 max_age, __u32 flags)
1377 struct ll_sb_info *sbi = ll_s2sbi(sb);
1378 struct obd_statfs obd_osfs;
1382 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1384 CERROR("md_statfs fails: rc = %d\n", rc);
1388 osfs->os_type = sb->s_magic;
1390 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1391 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1393 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1394 flags |= OBD_STATFS_NODELAY;
1396 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1397 &obd_osfs, max_age, flags);
1399 CERROR("obd_statfs fails: rc = %d\n", rc);
1403 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1404 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1407 osfs->os_bsize = obd_osfs.os_bsize;
1408 osfs->os_blocks = obd_osfs.os_blocks;
1409 osfs->os_bfree = obd_osfs.os_bfree;
1410 osfs->os_bavail = obd_osfs.os_bavail;
1412 /* If we don't have as many objects free on the OST as inodes
1413 * on the MDS, we reduce the total number of inodes to
1414 * compensate, so that the "inodes in use" number is correct.
1416 if (obd_osfs.os_ffree < osfs->os_ffree) {
1417 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1419 osfs->os_ffree = obd_osfs.os_ffree;
1424 #ifndef HAVE_STATFS_DENTRY_PARAM
1425 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1428 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1430 struct super_block *sb = de->d_sb;
1432 struct obd_statfs osfs;
1435 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1436 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1438 /* Some amount of caching on the client is allowed */
1439 rc = ll_statfs_internal(sb, &osfs,
1440 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1445 statfs_unpack(sfs, &osfs);
1447 /* We need to downshift for all 32-bit kernels, because we can't
1448 * tell if the kernel is being called via sys_statfs64() or not.
1449 * Stop before overflowing f_bsize - in which case it is better
1450 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1451 if (sizeof(long) < 8) {
1452 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1455 osfs.os_blocks >>= 1;
1456 osfs.os_bfree >>= 1;
1457 osfs.os_bavail >>= 1;
1461 sfs->f_blocks = osfs.os_blocks;
1462 sfs->f_bfree = osfs.os_bfree;
1463 sfs->f_bavail = osfs.os_bavail;
1468 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1470 struct ll_inode_info *lli;
1471 struct lov_stripe_md *lsm;
1473 lli = ll_i2info(inode);
1474 LASSERT(lli->lli_size_sem_owner != current);
1475 cfs_down(&lli->lli_size_sem);
1476 LASSERT(lli->lli_size_sem_owner == NULL);
1477 lli->lli_size_sem_owner = current;
1479 LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1482 lov_stripe_lock(lsm);
1485 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1487 struct ll_inode_info *lli;
1488 struct lov_stripe_md *lsm;
1490 lli = ll_i2info(inode);
1492 LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1495 lov_stripe_unlock(lsm);
1496 LASSERT(lli->lli_size_sem_owner == current);
1497 lli->lli_size_sem_owner = NULL;
1498 cfs_up(&lli->lli_size_sem);
1501 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1503 struct ll_inode_info *lli = ll_i2info(inode);
1504 struct mdt_body *body = md->body;
1505 struct lov_stripe_md *lsm = md->lsm;
1506 struct ll_sb_info *sbi = ll_i2sbi(inode);
1508 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1510 cfs_down(&lli->lli_och_sem);
1511 if (lli->lli_smd == NULL) {
1512 if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1513 lsm->lsm_magic != LOV_MAGIC_V3) {
1514 dump_lsm(D_ERROR, lsm);
1517 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1518 lsm, inode->i_ino, inode->i_generation, inode);
1519 /* cl_inode_init must go before lli_smd or a race is
1520 * possible where client thinks the file has stripes,
1521 * but lov raid0 is not setup yet and parallel e.g.
1522 * glimpse would try to use uninitialized lov */
1523 cl_inode_init(inode, md);
1524 cfs_spin_lock(&lli->lli_lock);
1526 cfs_spin_unlock(&lli->lli_lock);
1527 cfs_up(&lli->lli_och_sem);
1528 lli->lli_maxbytes = lsm->lsm_maxbytes;
1529 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1530 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1532 cfs_up(&lli->lli_och_sem);
1533 LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1534 lli->lli_smd->lsm_stripe_count ==
1535 lsm->lsm_stripe_count);
1536 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1537 CERROR("lsm mismatch for inode %ld\n",
1539 CERROR("lli_smd:\n");
1540 dump_lsm(D_ERROR, lli->lli_smd);
1542 dump_lsm(D_ERROR, lsm);
1546 if (lli->lli_smd != lsm)
1547 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1550 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1551 if (body->valid & OBD_MD_FLRMTPERM)
1552 ll_update_remote_perm(inode, md->remote_perm);
1554 #ifdef CONFIG_FS_POSIX_ACL
1555 else if (body->valid & OBD_MD_FLACL) {
1556 cfs_spin_lock(&lli->lli_lock);
1557 if (lli->lli_posix_acl)
1558 posix_acl_release(lli->lli_posix_acl);
1559 lli->lli_posix_acl = md->posix_acl;
1560 cfs_spin_unlock(&lli->lli_lock);
1563 inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1564 inode->i_generation = cl_fid_build_gen(&body->fid1);
1566 if (body->valid & OBD_MD_FLATIME) {
1567 if (body->atime > LTIME_S(inode->i_atime))
1568 LTIME_S(inode->i_atime) = body->atime;
1569 lli->lli_lvb.lvb_atime = body->atime;
1571 if (body->valid & OBD_MD_FLMTIME) {
1572 if (body->mtime > LTIME_S(inode->i_mtime)) {
1573 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1574 "to "LPU64"\n", inode->i_ino,
1575 LTIME_S(inode->i_mtime), body->mtime);
1576 LTIME_S(inode->i_mtime) = body->mtime;
1578 lli->lli_lvb.lvb_mtime = body->mtime;
1580 if (body->valid & OBD_MD_FLCTIME) {
1581 if (body->ctime > LTIME_S(inode->i_ctime))
1582 LTIME_S(inode->i_ctime) = body->ctime;
1583 lli->lli_lvb.lvb_ctime = body->ctime;
1585 if (body->valid & OBD_MD_FLMODE)
1586 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1587 if (body->valid & OBD_MD_FLTYPE)
1588 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1589 LASSERT(inode->i_mode != 0);
1590 if (S_ISREG(inode->i_mode)) {
1591 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1593 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1595 #ifdef HAVE_INODE_BLKSIZE
1596 inode->i_blksize = 1<<inode->i_blkbits;
1598 if (body->valid & OBD_MD_FLUID)
1599 inode->i_uid = body->uid;
1600 if (body->valid & OBD_MD_FLGID)
1601 inode->i_gid = body->gid;
1602 if (body->valid & OBD_MD_FLFLAGS)
1603 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1604 if (body->valid & OBD_MD_FLNLINK)
1605 inode->i_nlink = body->nlink;
1606 if (body->valid & OBD_MD_FLRDEV)
1607 inode->i_rdev = old_decode_dev(body->rdev);
1609 if (body->valid & OBD_MD_FLID) {
1610 /* FID shouldn't be changed! */
1611 if (fid_is_sane(&lli->lli_fid)) {
1612 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1613 "Trying to change FID "DFID
1614 " to the "DFID", inode %lu/%u(%p)\n",
1615 PFID(&lli->lli_fid), PFID(&body->fid1),
1616 inode->i_ino, inode->i_generation, inode);
1618 lli->lli_fid = body->fid1;
1621 LASSERT(fid_seq(&lli->lli_fid) != 0);
1623 if (body->valid & OBD_MD_FLSIZE) {
1624 if (exp_connect_som(ll_i2mdexp(inode)) &&
1625 S_ISREG(inode->i_mode) && lli->lli_smd) {
1626 struct lustre_handle lockh;
1629 /* As it is possible a blocking ast has been processed
1630 * by this time, we need to check there is an UPDATE
1631 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1633 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1636 if (lli->lli_flags & (LLIF_DONE_WRITING |
1637 LLIF_EPOCH_PENDING |
1639 CERROR("ino %lu flags %lu still has "
1640 "size authority! do not trust "
1641 "the size got from MDS\n",
1642 inode->i_ino, lli->lli_flags);
1644 /* Use old size assignment to avoid
1645 * deadlock bz14138 & bz14326 */
1646 inode->i_size = body->size;
1647 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1649 ldlm_lock_decref(&lockh, mode);
1652 /* Use old size assignment to avoid
1653 * deadlock bz14138 & bz14326 */
1654 inode->i_size = body->size;
1656 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1657 inode->i_ino, (unsigned long long)body->size);
1660 if (body->valid & OBD_MD_FLBLOCKS)
1661 inode->i_blocks = body->blocks;
1664 if (body->valid & OBD_MD_FLMDSCAPA) {
1665 LASSERT(md->mds_capa);
1666 ll_add_capa(inode, md->mds_capa);
1668 if (body->valid & OBD_MD_FLOSSCAPA) {
1669 LASSERT(md->oss_capa);
1670 ll_add_capa(inode, md->oss_capa);
1674 void ll_read_inode2(struct inode *inode, void *opaque)
1676 struct lustre_md *md = opaque;
1677 struct ll_inode_info *lli = ll_i2info(inode);
1680 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1681 inode->i_ino, inode->i_generation, inode);
1685 LASSERT(!lli->lli_smd);
1687 /* Core attributes from the MDS first. This is a new inode, and
1688 * the VFS doesn't zero times in the core inode so we have to do
1689 * it ourselves. They will be overwritten by either MDS or OST
1690 * attributes - we just need to make sure they aren't newer. */
1691 LTIME_S(inode->i_mtime) = 0;
1692 LTIME_S(inode->i_atime) = 0;
1693 LTIME_S(inode->i_ctime) = 0;
1695 ll_update_inode(inode, md);
1697 /* OIDEBUG(inode); */
1699 /* initializing backing dev info. */
1700 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1703 if (S_ISREG(inode->i_mode)) {
1704 struct ll_sb_info *sbi = ll_i2sbi(inode);
1705 inode->i_op = &ll_file_inode_operations;
1706 inode->i_fop = sbi->ll_fop;
1707 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1709 } else if (S_ISDIR(inode->i_mode)) {
1710 inode->i_op = &ll_dir_inode_operations;
1711 inode->i_fop = &ll_dir_operations;
1712 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_dir_aops;
1714 } else if (S_ISLNK(inode->i_mode)) {
1715 inode->i_op = &ll_fast_symlink_inode_operations;
1718 inode->i_op = &ll_special_inode_operations;
1720 init_special_inode(inode, inode->i_mode,
1721 kdev_t_to_nr(inode->i_rdev));
1727 void ll_delete_inode(struct inode *inode)
1729 struct ll_sb_info *sbi = ll_i2sbi(inode);
1733 rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1735 CERROR("fid_delete() failed, rc %d\n", rc);
1737 truncate_inode_pages(&inode->i_data, 0);
1739 /* Workaround for LU-118 */
1740 if (inode->i_data.nrpages) {
1741 TREE_READ_LOCK_IRQ(&inode->i_data);
1742 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1743 LASSERTF(inode->i_data.nrpages == 0,
1744 "inode=%lu/%u(%p) nrpages=%lu, see "
1745 "http://jira.whamcloud.com/browse/LU-118\n",
1746 inode->i_ino, inode->i_generation, inode,
1747 inode->i_data.nrpages);
1749 /* Workaround end */
1756 int ll_iocontrol(struct inode *inode, struct file *file,
1757 unsigned int cmd, unsigned long arg)
1759 struct ll_sb_info *sbi = ll_i2sbi(inode);
1760 struct ptlrpc_request *req = NULL;
1765 case FSFILT_IOC_GETFLAGS: {
1766 struct mdt_body *body;
1767 struct md_op_data *op_data;
1769 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1770 0, 0, LUSTRE_OPC_ANY,
1772 if (op_data == NULL)
1775 op_data->op_valid = OBD_MD_FLFLAGS;
1776 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1777 ll_finish_md_op_data(op_data);
1779 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1783 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1785 flags = body->flags;
1787 ptlrpc_req_finished(req);
1789 RETURN(put_user(flags, (int *)arg));
1791 case FSFILT_IOC_SETFLAGS: {
1792 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1793 struct obd_info oinfo = { { { 0 } } };
1794 struct md_op_data *op_data;
1796 if (get_user(flags, (int *)arg))
1800 OBDO_ALLOC(oinfo.oi_oa);
1804 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1805 LUSTRE_OPC_ANY, NULL);
1806 if (IS_ERR(op_data))
1807 RETURN(PTR_ERR(op_data));
1809 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1810 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1811 rc = md_setattr(sbi->ll_md_exp, op_data,
1812 NULL, 0, NULL, 0, &req, NULL);
1813 ll_finish_md_op_data(op_data);
1814 ptlrpc_req_finished(req);
1816 OBDO_FREE(oinfo.oi_oa);
1821 OBDO_FREE(oinfo.oi_oa);
1822 GOTO(update_cache, rc);
1825 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1826 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1827 oinfo.oi_oa->o_flags = flags;
1828 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1830 oinfo.oi_capa = ll_mdscapa_get(inode);
1831 obdo_from_inode(oinfo.oi_oa, inode,
1832 &ll_i2info(inode)->lli_fid, 0);
1833 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1834 capa_put(oinfo.oi_capa);
1835 OBDO_FREE(oinfo.oi_oa);
1837 if (rc != -EPERM && rc != -EACCES)
1838 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1844 inode->i_flags = ll_ext_to_inode_flags(flags);
1854 int ll_flush_ctx(struct inode *inode)
1856 struct ll_sb_info *sbi = ll_i2sbi(inode);
1858 CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1860 obd_set_info_async(sbi->ll_md_exp,
1861 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1863 obd_set_info_async(sbi->ll_dt_exp,
1864 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1869 /* umount -f client means force down, don't save state */
1870 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1871 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1873 struct super_block *sb = vfsmnt->mnt_sb;
1875 void ll_umount_begin(struct super_block *sb)
1878 struct lustre_sb_info *lsi = s2lsi(sb);
1879 struct ll_sb_info *sbi = ll_s2sbi(sb);
1880 struct obd_device *obd;
1881 struct obd_ioctl_data *ioc_data;
1884 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1885 if (!(flags & MNT_FORCE)) {
1891 /* Tell the MGC we got umount -f */
1892 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1894 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1895 sb->s_count, atomic_read(&sb->s_active));
1897 obd = class_exp2obd(sbi->ll_md_exp);
1899 CERROR("Invalid MDC connection handle "LPX64"\n",
1900 sbi->ll_md_exp->exp_handle.h_cookie);
1906 obd = class_exp2obd(sbi->ll_dt_exp);
1908 CERROR("Invalid LOV connection handle "LPX64"\n",
1909 sbi->ll_dt_exp->exp_handle.h_cookie);
1915 OBD_ALLOC_PTR(ioc_data);
1917 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1918 sizeof ioc_data, ioc_data, NULL);
1920 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1921 sizeof ioc_data, ioc_data, NULL);
1923 OBD_FREE_PTR(ioc_data);
1927 /* Really, we'd like to wait until there are no requests outstanding,
1928 * and then continue. For now, we just invalidate the requests,
1929 * schedule() and sleep one second if needed, and hope.
1932 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1933 if (atomic_read(&vfsmnt->mnt_count) > 2) {
1934 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
1935 cfs_time_seconds(1));
1936 if (atomic_read(&vfsmnt->mnt_count) > 2)
1937 LCONSOLE_WARN("Mount still busy with %d refs! You "
1938 "may try to umount it a bit later\n",
1939 atomic_read(&vfsmnt->mnt_count));
1946 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1948 struct ll_sb_info *sbi = ll_s2sbi(sb);
1952 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1953 read_only = *flags & MS_RDONLY;
1954 err = obd_set_info_async(sbi->ll_md_exp,
1955 sizeof(KEY_READ_ONLY),
1956 KEY_READ_ONLY, sizeof(read_only),
1959 CERROR("Failed to change the read-only flag during "
1960 "remount: %d\n", err);
1965 sb->s_flags |= MS_RDONLY;
1967 sb->s_flags &= ~MS_RDONLY;
1972 int ll_prep_inode(struct inode **inode,
1973 struct ptlrpc_request *req,
1974 struct super_block *sb)
1976 struct ll_sb_info *sbi = NULL;
1977 struct lustre_md md;
1981 LASSERT(*inode || sb);
1982 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
1983 memset(&md, 0, sizeof(struct lustre_md));
1985 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
1986 sbi->ll_md_exp, &md);
1991 ll_update_inode(*inode, &md);
1993 LASSERT(sb != NULL);
1996 * At this point server returns to client's same fid as client
1997 * generated for creating. So using ->fid1 is okay here.
1999 LASSERT(fid_is_sane(&md.body->fid1));
2001 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2002 if (*inode == NULL || IS_ERR(*inode)) {
2004 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2005 #ifdef CONFIG_FS_POSIX_ACL
2007 posix_acl_release(md.posix_acl);
2008 md.posix_acl = NULL;
2011 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2013 CERROR("new_inode -fatal: rc %d\n", rc);
2019 md_free_lustre_md(sbi->ll_md_exp, &md);
2023 int ll_obd_statfs(struct inode *inode, void *arg)
2025 struct ll_sb_info *sbi = NULL;
2026 struct obd_export *exp;
2028 struct obd_ioctl_data *data = NULL;
2032 if (!inode || !(sbi = ll_i2sbi(inode)))
2033 GOTO(out_statfs, rc = -EINVAL);
2035 rc = obd_ioctl_getdata(&buf, &len, arg);
2037 GOTO(out_statfs, rc);
2040 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2041 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2042 GOTO(out_statfs, rc = -EINVAL);
2044 if (data->ioc_inllen1 != sizeof(__u32) ||
2045 data->ioc_inllen2 != sizeof(__u32) ||
2046 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2047 data->ioc_plen2 != sizeof(struct obd_uuid))
2048 GOTO(out_statfs, rc = -EINVAL);
2050 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2051 if (type == LL_STATFS_MDC)
2052 exp = sbi->ll_md_exp;
2053 else if (type == LL_STATFS_LOV)
2054 exp = sbi->ll_dt_exp;
2056 GOTO(out_statfs, rc = -ENODEV);
2058 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2060 GOTO(out_statfs, rc);
2063 obd_ioctl_freedata(buf, len);
2067 int ll_process_config(struct lustre_cfg *lcfg)
2071 struct lprocfs_static_vars lvars;
2075 lprocfs_llite_init_vars(&lvars);
2077 /* The instance name contains the sb: lustre-client-aacfe000 */
2078 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2079 if (!ptr || !*(++ptr))
2081 if (sscanf(ptr, "%lx", &x) != 1)
2084 /* This better be a real Lustre superblock! */
2085 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2087 /* Note we have not called client_common_fill_super yet, so
2088 proc fns must be able to handle that! */
2089 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2096 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2097 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2098 struct inode *i1, struct inode *i2,
2099 const char *name, int namelen,
2100 int mode, __u32 opc, void *data)
2102 LASSERT(i1 != NULL);
2104 if (namelen > ll_i2sbi(i1)->ll_namelen)
2105 return ERR_PTR(-ENAMETOOLONG);
2107 if (op_data == NULL)
2108 OBD_ALLOC_PTR(op_data);
2110 if (op_data == NULL)
2111 return ERR_PTR(-ENOMEM);
2113 ll_i2gids(op_data->op_suppgids, i1, i2);
2114 op_data->op_fid1 = *ll_inode2fid(i1);
2115 op_data->op_capa1 = ll_mdscapa_get(i1);
2118 op_data->op_fid2 = *ll_inode2fid(i2);
2119 op_data->op_capa2 = ll_mdscapa_get(i2);
2121 fid_zero(&op_data->op_fid2);
2122 op_data->op_capa2 = NULL;
2125 op_data->op_name = name;
2126 op_data->op_namelen = namelen;
2127 op_data->op_mode = mode;
2128 op_data->op_mod_time = cfs_time_current_sec();
2129 op_data->op_fsuid = cfs_curproc_fsuid();
2130 op_data->op_fsgid = cfs_curproc_fsgid();
2131 op_data->op_cap = cfs_curproc_cap_pack();
2132 op_data->op_bias = MDS_CHECK_SPLIT;
2133 op_data->op_opc = opc;
2134 op_data->op_mds = 0;
2135 op_data->op_data = data;
2137 /* If the file is being opened after mknod() (normally due to NFS)
2138 * try to use the default stripe data from parent directory for
2139 * allocating OST objects. Try to pass the parent FID to MDS. */
2140 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2141 ll_i2info(i2)->lli_smd == NULL) {
2142 struct ll_inode_info *lli = ll_i2info(i2);
2144 cfs_spin_lock(&lli->lli_lock);
2145 if (likely(lli->lli_smd == NULL &&
2146 !fid_is_zero(&lli->lli_pfid)))
2147 op_data->op_fid1 = lli->lli_pfid;
2148 cfs_spin_unlock(&lli->lli_lock);
2149 /** We ignore parent's capability temporary. */
2155 void ll_finish_md_op_data(struct md_op_data *op_data)
2157 capa_put(op_data->op_capa1);
2158 capa_put(op_data->op_capa2);
2159 OBD_FREE_PTR(op_data);
2162 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2164 struct ll_sb_info *sbi;
2166 LASSERT((seq != NULL) && (vfs != NULL));
2167 sbi = ll_s2sbi(vfs->mnt_sb);
2169 if (sbi->ll_flags & LL_SBI_NOLCK)
2170 seq_puts(seq, ",nolock");
2172 if (sbi->ll_flags & LL_SBI_FLOCK)
2173 seq_puts(seq, ",flock");
2175 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2176 seq_puts(seq, ",localflock");
2178 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2179 seq_puts(seq, ",user_xattr");
2181 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2182 seq_puts(seq, ",lazystatfs");