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 * Copyright (c) 2011 Whamcloud, Inc.
36 * This file is part of Lustre, http://www.lustre.org/
37 * Lustre is a trademark of Sun Microsystems, Inc.
39 * lustre/llite/llite_lib.c
41 * Lustre Light Super operations
44 #define DEBUG_SUBSYSTEM S_LLITE
46 #include <linux/module.h>
47 #include <linux/types.h>
48 #include <linux/version.h>
51 #include <lustre_lite.h>
52 #include <lustre_ha.h>
53 #include <lustre_dlm.h>
54 #include <lprocfs_status.h>
55 #include <lustre_disk.h>
56 #include <lustre_param.h>
57 #include <lustre_log.h>
58 #include <cl_object.h>
59 #include <obd_cksum.h>
60 #include "llite_internal.h"
62 cfs_mem_cache_t *ll_file_data_slab;
64 CFS_LIST_HEAD(ll_super_blocks);
65 cfs_spinlock_t ll_sb_lock = CFS_SPIN_LOCK_UNLOCKED;
67 #ifndef MS_HAS_NEW_AOPS
68 extern struct address_space_operations ll_aops;
69 extern struct address_space_operations ll_dir_aops;
71 extern struct address_space_operations_ext ll_aops;
72 extern struct address_space_operations_ext ll_dir_aops;
76 #define log2(n) cfs_ffz(~(n))
79 static struct ll_sb_info *ll_init_sbi(void)
81 struct ll_sb_info *sbi = NULL;
88 OBD_ALLOC(sbi, sizeof(*sbi));
92 cfs_spin_lock_init(&sbi->ll_lock);
93 cfs_init_mutex(&sbi->ll_lco.lco_lock);
94 cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
95 cfs_spin_lock_init(&sbi->ll_process_lock);
96 sbi->ll_rw_stats_on = 0;
99 pages = si.totalram - si.totalhigh;
100 if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
101 #ifdef HAVE_BGL_SUPPORT
102 sbi->ll_async_page_max = pages / 4;
104 sbi->ll_async_page_max = pages / 2;
107 sbi->ll_async_page_max = (pages / 4) * 3;
110 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
111 SBI_DEFAULT_READAHEAD_MAX);
112 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
113 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
114 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
115 CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
116 CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
118 ll_generate_random_uuid(uuid);
119 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
120 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
122 cfs_spin_lock(&ll_sb_lock);
123 cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
124 cfs_spin_unlock(&ll_sb_lock);
126 #ifdef ENABLE_CHECKSUM
127 sbi->ll_flags |= LL_SBI_CHECKSUM;
130 #ifdef HAVE_LRU_RESIZE_SUPPORT
131 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
134 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
135 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
137 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
141 /* metadata statahead is enabled by default */
142 sbi->ll_sa_max = LL_SA_RPC_DEF;
143 cfs_atomic_set(&sbi->ll_sa_total, 0);
144 cfs_atomic_set(&sbi->ll_sa_wrong, 0);
145 cfs_atomic_set(&sbi->ll_agl_total, 0);
146 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
151 void ll_free_sbi(struct super_block *sb)
153 struct ll_sb_info *sbi = ll_s2sbi(sb);
157 cfs_spin_lock(&ll_sb_lock);
158 cfs_list_del(&sbi->ll_list);
159 cfs_spin_unlock(&ll_sb_lock);
160 OBD_FREE(sbi, sizeof(*sbi));
165 static struct dentry_operations ll_d_root_ops = {
166 .d_compare = ll_dcompare,
167 .d_revalidate = ll_revalidate_nd,
170 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
171 struct vfsmount *mnt)
173 struct inode *root = 0;
174 struct ll_sb_info *sbi = ll_s2sbi(sb);
175 struct obd_device *obd;
176 struct obd_capa *oc = NULL;
177 struct obd_statfs *osfs = NULL;
178 struct ptlrpc_request *request = NULL;
179 struct obd_connect_data *data = NULL;
180 struct obd_uuid *uuid;
181 struct md_op_data *op_data;
182 struct lustre_md lmd;
184 int size, err, checksum;
187 obd = class_name2obd(md);
189 CERROR("MD %s: not setup or attached\n", md);
203 if (proc_lustre_fs_root) {
204 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
207 CERROR("could not register mount in /proc/fs/lustre\n");
210 /* indicate the features supported by this client */
211 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
212 OBD_CONNECT_ATTRFID |
213 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
214 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
215 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
216 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
217 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
218 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH;
220 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
221 data->ocd_connect_flags |= OBD_CONNECT_SOM;
223 #ifdef HAVE_LRU_RESIZE_SUPPORT
224 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
225 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
227 #ifdef CONFIG_FS_POSIX_ACL
228 data->ocd_connect_flags |= OBD_CONNECT_ACL;
230 data->ocd_ibits_known = MDS_INODELOCK_FULL;
231 data->ocd_version = LUSTRE_VERSION_CODE;
233 if (sb->s_flags & MS_RDONLY)
234 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
235 if (sbi->ll_flags & LL_SBI_USER_XATTR)
236 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
238 #ifdef HAVE_MS_FLOCK_LOCK
239 /* force vfs to use lustre handler for flock() calls - bug 10743 */
240 sb->s_flags |= MS_FLOCK_LOCK;
242 #ifdef MS_HAS_NEW_AOPS
243 sb->s_flags |= MS_HAS_NEW_AOPS;
246 if (sbi->ll_flags & LL_SBI_FLOCK)
247 sbi->ll_fop = &ll_file_operations_flock;
248 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
249 sbi->ll_fop = &ll_file_operations;
251 sbi->ll_fop = &ll_file_operations_noflock;
254 data->ocd_connect_flags |= OBD_CONNECT_REAL;
255 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
256 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
258 data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
260 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
262 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
263 "recovery, of which this client is not a "
264 "part. Please wait for recovery to complete,"
265 " abort, or time out.\n", md);
268 CERROR("cannot connect to %s: rc = %d\n", md, err);
272 err = obd_fid_init(sbi->ll_md_exp);
274 CERROR("Can't init metadata layer FID infrastructure, "
279 err = obd_statfs(obd, osfs,
280 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
282 GOTO(out_md_fid, err);
284 /* This needs to be after statfs to ensure connect has finished.
285 * Note that "data" does NOT contain the valid connect reply.
286 * If connecting to a 1.8 server there will be no LMV device, so
287 * we can access the MDC export directly and exp_connect_flags will
288 * be non-zero, but if accessing an upgraded 2.1 server it will
289 * have the correct flags filled in.
290 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
291 valid = sbi->ll_md_exp->exp_connect_flags & CLIENT_CONNECT_MDT_REQD;
292 if (sbi->ll_md_exp->exp_connect_flags != 0 &&
293 valid != CLIENT_CONNECT_MDT_REQD) {
296 OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
297 obd_connect_flags2str(buf, CFS_PAGE_SIZE,
298 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
299 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
300 "feature(s) needed for correct operation "
301 "of this client (%s). Please upgrade "
302 "server or downgrade client.\n",
303 sbi->ll_md_exp->exp_obd->obd_name, buf);
304 OBD_FREE(buf, CFS_PAGE_SIZE);
305 GOTO(out_md, err = -EPROTO);
308 size = sizeof(*data);
309 err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
310 KEY_CONN_DATA, &size, data, NULL);
312 CERROR("Get connect data failed: %d \n", err);
316 LASSERT(osfs->os_bsize);
317 sb->s_blocksize = osfs->os_bsize;
318 sb->s_blocksize_bits = log2(osfs->os_bsize);
319 sb->s_magic = LL_SUPER_MAGIC;
321 /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
322 * retval = in_file->f_op->sendfile(...);
324 * retval = -EOVERFLOW;
326 * it will check if *ppos is greater than max. However, max equals to
327 * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
328 * has been defined as a signed long long integer in linux kernel. */
329 #if BITS_PER_LONG == 64
330 sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
332 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
334 sbi->ll_namelen = osfs->os_namelen;
335 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
337 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
338 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
339 LCONSOLE_INFO("Disabling user_xattr feature because "
340 "it is not supported on the server\n");
341 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
344 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
346 sb->s_flags |= MS_POSIXACL;
348 sbi->ll_flags |= LL_SBI_ACL;
350 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
352 sb->s_flags &= ~MS_POSIXACL;
354 sbi->ll_flags &= ~LL_SBI_ACL;
357 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
358 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
359 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
360 LCONSOLE_INFO("client is set as remote by default.\n");
363 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
364 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
365 LCONSOLE_INFO("client claims to be remote, but server "
366 "rejected, forced to be local.\n");
370 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
371 LCONSOLE_INFO("client enabled MDS capability!\n");
372 sbi->ll_flags |= LL_SBI_MDS_CAPA;
375 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
376 LCONSOLE_INFO("client enabled OSS capability!\n");
377 sbi->ll_flags |= LL_SBI_OSS_CAPA;
380 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
381 sbi->ll_flags |= LL_SBI_64BIT_HASH;
383 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
384 sbi->ll_md_brw_size = data->ocd_brw_size;
386 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
388 obd = class_name2obd(dt);
390 CERROR("DT %s: not setup or attached\n", dt);
391 GOTO(out_md_fid, err = -ENODEV);
394 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
395 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
396 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
397 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
398 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
399 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
400 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
401 OBD_CONNECT_MAXBYTES;
403 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
404 data->ocd_connect_flags |= OBD_CONNECT_SOM;
406 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
407 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
408 * disabled by default, because it can still be enabled on the
409 * fly via /proc. As a consequence, we still need to come to an
410 * agreement on the supported algorithms at connect time */
411 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
413 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
414 data->ocd_cksum_types = OBD_CKSUM_ADLER;
416 data->ocd_cksum_types = cksum_types_supported();
419 #ifdef HAVE_LRU_RESIZE_SUPPORT
420 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
422 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
423 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
425 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
426 "ocd_grant: %d\n", data->ocd_connect_flags,
427 data->ocd_version, data->ocd_grant);
429 obd->obd_upcall.onu_owner = &sbi->ll_lco;
430 obd->obd_upcall.onu_upcall = cl_ocd_update;
432 data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
434 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
436 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
437 "recovery, of which this client is not a "
438 "part. Please wait for recovery to "
439 "complete, abort, or time out.\n", dt);
440 GOTO(out_md_fid, err);
442 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
443 GOTO(out_md_fid, err);
446 err = obd_fid_init(sbi->ll_dt_exp);
448 CERROR("Can't init data layer FID infrastructure, "
453 cfs_mutex_down(&sbi->ll_lco.lco_lock);
454 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
455 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
456 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
457 cfs_mutex_up(&sbi->ll_lco.lco_lock);
459 fid_zero(&sbi->ll_root_fid);
460 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
462 CERROR("cannot mds_connect: rc = %d\n", err);
463 GOTO(out_lock_cn_cb, err);
465 if (!fid_is_sane(&sbi->ll_root_fid)) {
466 CERROR("Invalid root fid during mount\n");
467 GOTO(out_lock_cn_cb, err = -EINVAL);
469 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
471 sb->s_op = &lustre_super_operations;
472 #if THREAD_SIZE >= 8192
473 sb->s_export_op = &lustre_export_operations;
477 * XXX: move this to after cbd setup? */
478 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
479 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
480 valid |= OBD_MD_FLRMTPERM;
481 else if (sbi->ll_flags & LL_SBI_ACL)
482 valid |= OBD_MD_FLACL;
484 OBD_ALLOC_PTR(op_data);
486 GOTO(out_lock_cn_cb, err = -ENOMEM);
488 op_data->op_fid1 = sbi->ll_root_fid;
489 op_data->op_mode = 0;
490 op_data->op_capa1 = oc;
491 op_data->op_valid = valid;
493 err = md_getattr(sbi->ll_md_exp, op_data, &request);
496 OBD_FREE_PTR(op_data);
498 CERROR("md_getattr failed for root: rc = %d\n", err);
499 GOTO(out_lock_cn_cb, err);
501 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
502 sbi->ll_md_exp, &lmd);
504 CERROR("failed to understand root inode md: rc = %d\n", err);
505 ptlrpc_req_finished (request);
506 GOTO(out_lock_cn_cb, err);
509 LASSERT(fid_is_sane(&sbi->ll_root_fid));
510 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
511 md_free_lustre_md(sbi->ll_md_exp, &lmd);
512 ptlrpc_req_finished(request);
514 if (root == NULL || IS_ERR(root)) {
516 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
517 #ifdef CONFIG_FS_POSIX_ACL
519 posix_acl_release(lmd.posix_acl);
520 lmd.posix_acl = NULL;
523 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
525 CERROR("lustre_lite: bad iget4 for root\n");
529 err = ll_close_thread_start(&sbi->ll_lcq);
531 CERROR("cannot start close thread: rc %d\n", err);
535 #ifdef CONFIG_FS_POSIX_ACL
536 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
537 rct_init(&sbi->ll_rct);
538 et_init(&sbi->ll_et);
542 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
543 err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
544 KEY_CHECKSUM, sizeof(checksum), &checksum,
548 sb->s_root = d_alloc_root(root);
549 sb->s_root->d_op = &ll_d_root_ops;
551 sbi->ll_sdev_orig = sb->s_dev;
553 /* We set sb->s_dev equal on all lustre clients in order to support
554 * NFS export clustering. NFSD requires that the FSID be the same
556 /* s_dev is also used in lt_compare() to compare two fs, but that is
557 * only a node-local comparison. */
558 uuid = obd_get_uuid(sbi->ll_md_exp);
560 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
573 obd_fid_fini(sbi->ll_dt_exp);
575 obd_disconnect(sbi->ll_dt_exp);
576 sbi->ll_dt_exp = NULL;
578 obd_fid_fini(sbi->ll_md_exp);
580 obd_disconnect(sbi->ll_md_exp);
581 sbi->ll_md_exp = NULL;
587 lprocfs_unregister_mountpoint(sbi);
591 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
595 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
597 rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
598 KEY_MAX_EASIZE, &size, lmmsize, NULL);
600 CERROR("Get max mdsize error rc %d \n", rc);
605 void ll_dump_inode(struct inode *inode)
607 struct list_head *tmp;
608 int dentry_count = 0;
610 LASSERT(inode != NULL);
612 list_for_each(tmp, &inode->i_dentry)
615 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
616 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
617 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
620 void lustre_dump_dentry(struct dentry *dentry, int recur)
622 struct list_head *tmp;
625 LASSERT(dentry != NULL);
627 list_for_each(tmp, &dentry->d_subdirs)
630 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
631 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
632 dentry->d_name.len, dentry->d_name.name,
633 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
634 dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
635 dentry->d_flags, dentry->d_fsdata, subdirs);
636 if (dentry->d_inode != NULL)
637 ll_dump_inode(dentry->d_inode);
642 list_for_each(tmp, &dentry->d_subdirs) {
643 struct dentry *d = list_entry(tmp, struct dentry, d_child);
644 lustre_dump_dentry(d, recur - 1);
648 void client_common_put_super(struct super_block *sb)
650 struct ll_sb_info *sbi = ll_s2sbi(sb);
653 #ifdef CONFIG_FS_POSIX_ACL
654 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
655 et_fini(&sbi->ll_et);
656 rct_fini(&sbi->ll_rct);
660 obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
662 ll_close_thread_shutdown(sbi->ll_lcq);
666 cfs_list_del(&sbi->ll_conn_chain);
668 obd_fid_fini(sbi->ll_dt_exp);
669 obd_disconnect(sbi->ll_dt_exp);
670 sbi->ll_dt_exp = NULL;
672 lprocfs_unregister_mountpoint(sbi);
674 obd_fid_fini(sbi->ll_md_exp);
675 obd_disconnect(sbi->ll_md_exp);
676 sbi->ll_md_exp = NULL;
681 void ll_kill_super(struct super_block *sb)
683 struct ll_sb_info *sbi;
688 if (!(sb->s_flags & MS_ACTIVE))
692 /* we need restore s_dev from changed for clustred NFS before put_super
693 * because new kernels have cached s_dev and change sb->s_dev in
694 * put_super not affected real removing devices */
696 sb->s_dev = sbi->ll_sdev_orig;
700 char *ll_read_opt(const char *opt, char *data)
706 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
707 if (strncmp(opt, data, strlen(opt)))
709 if ((value = strchr(data, '=')) == NULL)
713 OBD_ALLOC(retval, strlen(value) + 1);
715 CERROR("out of memory!\n");
719 memcpy(retval, value, strlen(value)+1);
720 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
724 static inline int ll_set_opt(const char *opt, char *data, int fl)
726 if (strncmp(opt, data, strlen(opt)) != 0)
732 /* non-client-specific mount options are parsed in lmd_parse */
733 static int ll_options(char *options, int *flags)
736 char *s1 = options, *s2;
742 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
745 CDEBUG(D_SUPER, "next opt=%s\n", s1);
746 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
751 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
756 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
761 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
766 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
771 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
776 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,5,50,0)
777 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
779 /* Ignore deprecated mount option. The client will
780 * always try to mount with ACL support, whether this
781 * is used depends on whether server supports it. */
782 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
783 "mount option 'acl'.\n");
786 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
788 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
789 "mount option 'noacl'.\n");
793 #warning "{no}acl options have been deprecated since 1.8, please remove them"
795 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
801 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
806 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
811 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
816 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
821 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
826 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
831 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
836 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
842 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
848 s2 = strchr(s1, ',');
856 void ll_lli_init(struct ll_inode_info *lli)
858 lli->lli_inode_magic = LLI_INODE_MAGIC;
860 lli->lli_ioepoch = 0;
861 cfs_spin_lock_init(&lli->lli_lock);
862 lli->lli_posix_acl = NULL;
863 lli->lli_remote_perms = NULL;
864 cfs_sema_init(&lli->lli_rmtperm_sem, 1);
865 /* Do not set lli_fid, it has been initialized already. */
866 fid_zero(&lli->lli_pfid);
867 CFS_INIT_LIST_HEAD(&lli->lli_close_list);
868 CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
869 cfs_atomic_set(&lli->lli_open_count, 0);
870 lli->lli_mds_capa = NULL;
871 lli->lli_rmtperm_time = 0;
872 lli->lli_pending_och = NULL;
873 lli->lli_mds_read_och = NULL;
874 lli->lli_mds_write_och = NULL;
875 lli->lli_mds_exec_och = NULL;
876 lli->lli_open_fd_read_count = 0;
877 lli->lli_open_fd_write_count = 0;
878 lli->lli_open_fd_exec_count = 0;
879 cfs_sema_init(&lli->lli_och_sem, 1);
880 cfs_spin_lock_init(&lli->lli_agl_lock);
882 lli->lli_clob = NULL;
884 LASSERT(lli->lli_vfs_inode.i_mode != 0);
885 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
886 cfs_sema_init(&lli->lli_readdir_sem, 1);
887 lli->lli_opendir_key = NULL;
890 lli->lli_def_acl = NULL;
891 cfs_spin_lock_init(&lli->lli_sa_lock);
892 lli->lli_opendir_pid = 0;
894 cfs_sema_init(&lli->lli_size_sem, 1);
895 lli->lli_size_sem_owner = NULL;
896 lli->lli_symlink_name = NULL;
897 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
898 cfs_init_rwsem(&lli->lli_trunc_sem);
899 cfs_sema_init(&lli->lli_write_sem, 1);
900 lli->lli_async_rc = 0;
901 lli->lli_write_rc = 0;
902 cfs_init_rwsem(&lli->lli_glimpse_sem);
903 lli->lli_glimpse_time = 0;
904 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
905 lli->lli_agl_index = 0;
909 static inline int ll_bdi_register(struct backing_dev_info *bdi)
911 #ifdef HAVE_BDI_REGISTER
912 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
915 bdi->name = "lustre";
917 return bdi_register(bdi, NULL, "lustre-%d",
918 atomic_inc_return(&ll_bdi_num));
924 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
926 struct lustre_profile *lprof = NULL;
927 struct lustre_sb_info *lsi = s2lsi(sb);
928 struct ll_sb_info *sbi;
929 char *dt = NULL, *md = NULL;
930 char *profilenm = get_profile_name(sb);
931 struct config_llog_instance *cfg;
932 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
933 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
937 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
945 /* client additional sb info */
946 lsi->lsi_llsbi = sbi = ll_init_sbi();
948 cfs_module_put(THIS_MODULE);
953 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
957 err = ll_bdi_init(&lsi->lsi_bdi);
960 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
961 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
962 err = ll_bdi_register(&lsi->lsi_bdi);
967 sb->s_bdi = &lsi->lsi_bdi;
970 /* Generate a string unique to this super, in case some joker tries
971 to mount the same fs at two mount points.
972 Use the address of the super itself.*/
973 cfg->cfg_instance = sb;
974 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
976 /* set up client obds */
977 err = lustre_process_log(sb, profilenm, cfg);
979 CERROR("Unable to process log: %d\n", err);
983 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
984 lprof = class_get_profile(profilenm);
986 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
987 " read from the MGS. Does that filesystem "
988 "exist?\n", profilenm);
989 GOTO(out_free, err = -EINVAL);
991 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
992 lprof->lp_md, lprof->lp_dt);
994 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
996 GOTO(out_free, err = -ENOMEM);
997 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
999 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1001 GOTO(out_free, err = -ENOMEM);
1002 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1004 /* connections, registrations, sb setup */
1005 err = client_common_fill_super(sb, md, dt, mnt);
1009 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1011 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1015 LCONSOLE_WARN("Client %s has started\n", profilenm);
1019 } /* ll_fill_super */
1022 void lu_context_keys_dump(void);
1024 void ll_put_super(struct super_block *sb)
1026 struct config_llog_instance cfg;
1027 struct obd_device *obd;
1028 struct lustre_sb_info *lsi = s2lsi(sb);
1029 struct ll_sb_info *sbi = ll_s2sbi(sb);
1030 char *profilenm = get_profile_name(sb);
1031 int force = 1, next;
1034 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1036 ll_print_capa_stat(sbi);
1038 cfg.cfg_instance = sb;
1039 lustre_end_log(sb, profilenm, &cfg);
1041 if (sbi->ll_md_exp) {
1042 obd = class_exp2obd(sbi->ll_md_exp);
1044 force = obd->obd_force;
1047 /* We need to set force before the lov_disconnect in
1048 lustre_common_put_super, since l_d cleans up osc's as well. */
1051 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1053 obd->obd_force = force;
1058 /* Only if client_common_fill_super succeeded */
1059 client_common_put_super(sb);
1063 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1064 class_manual_cleanup(obd);
1068 class_del_profile(profilenm);
1070 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1071 ll_bdi_destroy(&lsi->lsi_bdi);
1072 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1076 lsi->lsi_llsbi = NULL;
1078 lustre_common_put_super(sb);
1080 cl_env_cache_purge(~0);
1082 LCONSOLE_WARN("client %p umount complete\n", cfg.cfg_instance);
1084 cfs_module_put(THIS_MODULE);
1087 } /* client_put_super */
1089 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1091 struct inode *inode = NULL;
1092 /* NOTE: we depend on atomic igrab() -bzzz */
1093 lock_res_and_lock(lock);
1094 if (lock->l_ast_data) {
1095 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1096 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1097 inode = igrab(lock->l_ast_data);
1099 inode = lock->l_ast_data;
1100 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1101 D_WARNING, lock, "l_ast_data %p is "
1102 "bogus: magic %08x", lock->l_ast_data,
1103 lli->lli_inode_magic);
1107 unlock_res_and_lock(lock);
1111 static int null_if_equal(struct ldlm_lock *lock, void *data)
1113 if (data == lock->l_ast_data) {
1114 lock->l_ast_data = NULL;
1116 if (lock->l_req_mode != lock->l_granted_mode)
1117 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1120 return LDLM_ITER_CONTINUE;
1123 void ll_clear_inode(struct inode *inode)
1125 struct ll_inode_info *lli = ll_i2info(inode);
1126 struct ll_sb_info *sbi = ll_i2sbi(inode);
1129 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1130 inode->i_generation, inode);
1132 if (S_ISDIR(inode->i_mode)) {
1133 /* these should have been cleared in ll_file_release */
1134 LASSERT(lli->lli_opendir_key == NULL);
1135 LASSERT(lli->lli_sai == NULL);
1136 LASSERT(lli->lli_opendir_pid == 0);
1139 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1140 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1141 null_if_equal, inode);
1143 LASSERT(!lli->lli_open_fd_write_count);
1144 LASSERT(!lli->lli_open_fd_read_count);
1145 LASSERT(!lli->lli_open_fd_exec_count);
1147 if (lli->lli_mds_write_och)
1148 ll_md_real_close(inode, FMODE_WRITE);
1149 if (lli->lli_mds_exec_och)
1150 ll_md_real_close(inode, FMODE_EXEC);
1151 if (lli->lli_mds_read_och)
1152 ll_md_real_close(inode, FMODE_READ);
1154 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1155 OBD_FREE(lli->lli_symlink_name,
1156 strlen(lli->lli_symlink_name) + 1);
1157 lli->lli_symlink_name = NULL;
1160 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1161 LASSERT(lli->lli_posix_acl == NULL);
1162 if (lli->lli_remote_perms) {
1163 free_rmtperm_hash(lli->lli_remote_perms);
1164 lli->lli_remote_perms = NULL;
1167 #ifdef CONFIG_FS_POSIX_ACL
1168 else if (lli->lli_posix_acl) {
1169 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1170 LASSERT(lli->lli_remote_perms == NULL);
1171 posix_acl_release(lli->lli_posix_acl);
1172 lli->lli_posix_acl = NULL;
1175 lli->lli_inode_magic = LLI_INODE_DEAD;
1177 ll_clear_inode_capas(inode);
1178 if (!S_ISDIR(inode->i_mode))
1179 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1182 * XXX This has to be done before lsm is freed below, because
1183 * cl_object still uses inode lsm.
1185 cl_inode_fini(inode);
1188 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1189 lli->lli_smd = NULL;
1196 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1197 struct md_open_data **mod)
1199 struct lustre_md md;
1200 struct inode *inode = dentry->d_inode;
1201 struct ll_sb_info *sbi = ll_i2sbi(inode);
1202 struct ptlrpc_request *request = NULL;
1206 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1207 LUSTRE_OPC_ANY, NULL);
1208 if (IS_ERR(op_data))
1209 RETURN(PTR_ERR(op_data));
1211 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1214 ptlrpc_req_finished(request);
1215 if (rc == -ENOENT) {
1217 /* Unlinked special device node? Or just a race?
1218 * Pretend we done everything. */
1219 if (!S_ISREG(inode->i_mode) &&
1220 !S_ISDIR(inode->i_mode)) {
1221 ia_valid = op_data->op_attr.ia_valid;
1222 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1223 rc = simple_setattr(dentry, &op_data->op_attr);
1224 op_data->op_attr.ia_valid = ia_valid;
1226 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1227 CERROR("md_setattr fails: rc = %d\n", rc);
1232 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1233 sbi->ll_md_exp, &md);
1235 ptlrpc_req_finished(request);
1239 /* We want to adjust timestamps.
1240 * If there is at least some data in file, we cleared ATTR_SIZE
1241 * to avoid update size, otherwise it is important to do.(SOM case)
1243 ia_valid = op_data->op_attr.ia_valid;
1244 /* Since we set ATTR_*_SET flags above, and already done permission
1245 * check, So don't let inode_change_ok() check it again. */
1246 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1247 rc = simple_setattr(dentry, &op_data->op_attr);
1248 op_data->op_attr.ia_valid = ia_valid;
1250 /* Extract epoch data if obtained. */
1251 op_data->op_handle = md.body->handle;
1252 op_data->op_ioepoch = md.body->ioepoch;
1254 ll_update_inode(inode, &md);
1255 ptlrpc_req_finished(request);
1260 /* Close IO epoch and send Size-on-MDS attribute update. */
1261 static int ll_setattr_done_writing(struct inode *inode,
1262 struct md_op_data *op_data,
1263 struct md_open_data *mod)
1265 struct ll_inode_info *lli = ll_i2info(inode);
1269 LASSERT(op_data != NULL);
1270 if (!S_ISREG(inode->i_mode))
1273 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1274 op_data->op_ioepoch, PFID(&lli->lli_fid));
1276 op_data->op_flags = MF_EPOCH_CLOSE;
1277 ll_done_writing_attr(inode, op_data);
1278 ll_pack_inode2opdata(inode, op_data, NULL);
1280 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1281 if (rc == -EAGAIN) {
1282 /* MDS has instructed us to obtain Size-on-MDS attribute
1283 * from OSTs and send setattr to back to MDS. */
1284 rc = ll_som_update(inode, op_data);
1286 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1292 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1294 struct obd_capa *capa;
1297 if (attr->ia_valid & ATTR_SIZE)
1298 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1300 capa = ll_mdscapa_get(inode);
1302 rc = cl_setattr_ost(inode, attr, capa);
1304 if (attr->ia_valid & ATTR_SIZE)
1305 ll_truncate_free_capa(capa);
1312 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1313 * object(s) determine the file size and mtime. Otherwise, the MDS will
1314 * keep these values until such a time that objects are allocated for it.
1315 * We do the MDS operations first, as it is checking permissions for us.
1316 * We don't to the MDS RPC if there is nothing that we want to store there,
1317 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1318 * going to do an RPC anyways.
1320 * If we are doing a truncate, we will send the mtime and ctime updates
1321 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1322 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1325 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1327 struct inode *inode = dentry->d_inode;
1328 struct ll_inode_info *lli = ll_i2info(inode);
1329 struct lov_stripe_md *lsm = lli->lli_smd;
1330 struct md_op_data *op_data = NULL;
1331 struct md_open_data *mod = NULL;
1332 int ia_valid = attr->ia_valid;
1333 int rc = 0, rc1 = 0;
1336 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1338 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1340 if (ia_valid & ATTR_SIZE) {
1341 if (attr->ia_size > ll_file_maxbytes(inode)) {
1342 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1343 attr->ia_size, ll_file_maxbytes(inode));
1347 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1350 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1351 if (ia_valid & TIMES_SET_FLAGS) {
1352 if (cfs_curproc_fsuid() != inode->i_uid &&
1353 !cfs_capable(CFS_CAP_FOWNER))
1357 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1358 if (attr->ia_valid & ATTR_CTIME) {
1359 attr->ia_ctime = CFS_CURRENT_TIME;
1360 attr->ia_valid |= ATTR_CTIME_SET;
1362 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1363 attr->ia_atime = CFS_CURRENT_TIME;
1364 attr->ia_valid |= ATTR_ATIME_SET;
1366 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1367 attr->ia_mtime = CFS_CURRENT_TIME;
1368 attr->ia_valid |= ATTR_MTIME_SET;
1371 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1372 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1373 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1374 cfs_time_current_sec());
1376 /* NB: ATTR_SIZE will only be set after this point if the size
1377 * resides on the MDS, ie, this file has no objects. */
1379 attr->ia_valid &= ~ATTR_SIZE;
1381 /* We always do an MDS RPC, even if we're only changing the size;
1382 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1384 OBD_ALLOC_PTR(op_data);
1385 if (op_data == NULL)
1388 if (!S_ISDIR(inode->i_mode)) {
1389 if (ia_valid & ATTR_SIZE)
1390 UP_WRITE_I_ALLOC_SEM(inode);
1391 UNLOCK_INODE_MUTEX(inode);
1392 cfs_down_write(&lli->lli_trunc_sem);
1393 LOCK_INODE_MUTEX(inode);
1394 if (ia_valid & ATTR_SIZE)
1395 DOWN_WRITE_I_ALLOC_SEM(inode);
1398 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1400 /* Open epoch for truncate. */
1401 if (exp_connect_som(ll_i2mdexp(inode)) &&
1402 (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1403 op_data->op_flags = MF_EPOCH_OPEN;
1405 rc = ll_md_setattr(dentry, op_data, &mod);
1409 ll_ioepoch_open(lli, op_data->op_ioepoch);
1410 if (!lsm || !S_ISREG(inode->i_mode)) {
1411 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1415 if (ia_valid & ATTR_SIZE)
1416 attr->ia_valid |= ATTR_SIZE;
1417 if (ia_valid & (ATTR_SIZE |
1418 ATTR_ATIME | ATTR_ATIME_SET |
1419 ATTR_MTIME | ATTR_MTIME_SET))
1420 /* on truncate and utimes send attributes to osts, setting
1421 * mtime/atime to past will be performed under PW 0:EOF extent
1422 * lock (new_size:EOF for truncate)
1423 * it may seem excessive to send mtime/atime updates to osts
1424 * when not setting times to past, but it is necessary due to
1425 * possible time de-synchronization */
1426 rc = ll_setattr_ost(inode, attr);
1430 if (op_data->op_ioepoch)
1431 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1432 ll_finish_md_op_data(op_data);
1434 if (!S_ISDIR(inode->i_mode))
1435 cfs_up_write(&lli->lli_trunc_sem);
1436 return rc ? rc : rc1;
1439 int ll_setattr(struct dentry *de, struct iattr *attr)
1441 int mode = de->d_inode->i_mode;
1443 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1444 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1445 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1447 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1448 (ATTR_SIZE|ATTR_MODE)) &&
1449 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1450 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1451 !(attr->ia_mode & S_ISGID))))
1452 attr->ia_valid |= ATTR_FORCE;
1454 if ((mode & S_ISUID) &&
1455 !(attr->ia_mode & S_ISUID) &&
1456 !(attr->ia_valid & ATTR_KILL_SUID))
1457 attr->ia_valid |= ATTR_KILL_SUID;
1459 if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1460 !(attr->ia_mode & S_ISGID) &&
1461 !(attr->ia_valid & ATTR_KILL_SGID))
1462 attr->ia_valid |= ATTR_KILL_SGID;
1464 return ll_setattr_raw(de, attr);
1467 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1468 __u64 max_age, __u32 flags)
1470 struct ll_sb_info *sbi = ll_s2sbi(sb);
1471 struct obd_statfs obd_osfs;
1475 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1477 CERROR("md_statfs fails: rc = %d\n", rc);
1481 osfs->os_type = sb->s_magic;
1483 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1484 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1486 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1487 flags |= OBD_STATFS_NODELAY;
1489 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1490 &obd_osfs, max_age, flags);
1492 CERROR("obd_statfs fails: rc = %d\n", rc);
1496 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1497 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1500 osfs->os_bsize = obd_osfs.os_bsize;
1501 osfs->os_blocks = obd_osfs.os_blocks;
1502 osfs->os_bfree = obd_osfs.os_bfree;
1503 osfs->os_bavail = obd_osfs.os_bavail;
1505 /* If we don't have as many objects free on the OST as inodes
1506 * on the MDS, we reduce the total number of inodes to
1507 * compensate, so that the "inodes in use" number is correct.
1509 if (obd_osfs.os_ffree < osfs->os_ffree) {
1510 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1512 osfs->os_ffree = obd_osfs.os_ffree;
1517 #ifndef HAVE_STATFS_DENTRY_PARAM
1518 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1521 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1523 struct super_block *sb = de->d_sb;
1525 struct obd_statfs osfs;
1528 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1529 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1531 /* Some amount of caching on the client is allowed */
1532 rc = ll_statfs_internal(sb, &osfs,
1533 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1538 statfs_unpack(sfs, &osfs);
1540 /* We need to downshift for all 32-bit kernels, because we can't
1541 * tell if the kernel is being called via sys_statfs64() or not.
1542 * Stop before overflowing f_bsize - in which case it is better
1543 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1544 if (sizeof(long) < 8) {
1545 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1548 osfs.os_blocks >>= 1;
1549 osfs.os_bfree >>= 1;
1550 osfs.os_bavail >>= 1;
1554 sfs->f_blocks = osfs.os_blocks;
1555 sfs->f_bfree = osfs.os_bfree;
1556 sfs->f_bavail = osfs.os_bavail;
1561 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1563 struct ll_inode_info *lli;
1564 struct lov_stripe_md *lsm;
1566 LASSERT(!S_ISDIR(inode->i_mode));
1568 lli = ll_i2info(inode);
1569 LASSERT(lli->lli_size_sem_owner != current);
1570 cfs_down(&lli->lli_size_sem);
1571 LASSERT(lli->lli_size_sem_owner == NULL);
1572 lli->lli_size_sem_owner = current;
1574 LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1577 lov_stripe_lock(lsm);
1580 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1582 struct ll_inode_info *lli;
1583 struct lov_stripe_md *lsm;
1585 lli = ll_i2info(inode);
1587 LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1590 lov_stripe_unlock(lsm);
1591 LASSERT(lli->lli_size_sem_owner == current);
1592 lli->lli_size_sem_owner = NULL;
1593 cfs_up(&lli->lli_size_sem);
1596 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1598 struct ll_inode_info *lli = ll_i2info(inode);
1599 struct mdt_body *body = md->body;
1600 struct lov_stripe_md *lsm = md->lsm;
1601 struct ll_sb_info *sbi = ll_i2sbi(inode);
1603 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1605 LASSERT(S_ISREG(inode->i_mode));
1607 cfs_down(&lli->lli_och_sem);
1608 if (lli->lli_smd == NULL) {
1609 if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1610 lsm->lsm_magic != LOV_MAGIC_V3) {
1611 dump_lsm(D_ERROR, lsm);
1614 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1615 lsm, inode->i_ino, inode->i_generation, inode);
1616 /* cl_file_inode_init must go before lli_smd or a race
1617 * is possible where client thinks the file has stripes,
1618 * but lov raid0 is not setup yet and parallel e.g.
1619 * glimpse would try to use uninitialized lov */
1620 cl_file_inode_init(inode, md);
1621 cfs_spin_lock(&lli->lli_lock);
1623 cfs_spin_unlock(&lli->lli_lock);
1624 cfs_up(&lli->lli_och_sem);
1625 lli->lli_maxbytes = lsm->lsm_maxbytes;
1626 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1627 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1629 cfs_up(&lli->lli_och_sem);
1630 LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1631 lli->lli_smd->lsm_stripe_count ==
1632 lsm->lsm_stripe_count);
1633 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1634 CERROR("lsm mismatch for inode %ld\n",
1636 CERROR("lli_smd:\n");
1637 dump_lsm(D_ERROR, lli->lli_smd);
1639 dump_lsm(D_ERROR, lsm);
1643 if (lli->lli_smd != lsm)
1644 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1647 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1648 if (body->valid & OBD_MD_FLRMTPERM)
1649 ll_update_remote_perm(inode, md->remote_perm);
1651 #ifdef CONFIG_FS_POSIX_ACL
1652 else if (body->valid & OBD_MD_FLACL) {
1653 cfs_spin_lock(&lli->lli_lock);
1654 if (lli->lli_posix_acl)
1655 posix_acl_release(lli->lli_posix_acl);
1656 lli->lli_posix_acl = md->posix_acl;
1657 cfs_spin_unlock(&lli->lli_lock);
1660 inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1661 inode->i_generation = cl_fid_build_gen(&body->fid1);
1663 if (body->valid & OBD_MD_FLATIME) {
1664 if (body->atime > LTIME_S(inode->i_atime))
1665 LTIME_S(inode->i_atime) = body->atime;
1666 lli->lli_lvb.lvb_atime = body->atime;
1668 if (body->valid & OBD_MD_FLMTIME) {
1669 if (body->mtime > LTIME_S(inode->i_mtime)) {
1670 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1671 "to "LPU64"\n", inode->i_ino,
1672 LTIME_S(inode->i_mtime), body->mtime);
1673 LTIME_S(inode->i_mtime) = body->mtime;
1675 lli->lli_lvb.lvb_mtime = body->mtime;
1677 if (body->valid & OBD_MD_FLCTIME) {
1678 if (body->ctime > LTIME_S(inode->i_ctime))
1679 LTIME_S(inode->i_ctime) = body->ctime;
1680 lli->lli_lvb.lvb_ctime = body->ctime;
1682 if (body->valid & OBD_MD_FLMODE)
1683 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1684 if (body->valid & OBD_MD_FLTYPE)
1685 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1686 LASSERT(inode->i_mode != 0);
1687 if (S_ISREG(inode->i_mode)) {
1688 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1690 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1692 #ifdef HAVE_INODE_BLKSIZE
1693 inode->i_blksize = 1<<inode->i_blkbits;
1695 if (body->valid & OBD_MD_FLUID)
1696 inode->i_uid = body->uid;
1697 if (body->valid & OBD_MD_FLGID)
1698 inode->i_gid = body->gid;
1699 if (body->valid & OBD_MD_FLFLAGS)
1700 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1701 if (body->valid & OBD_MD_FLNLINK)
1702 inode->i_nlink = body->nlink;
1703 if (body->valid & OBD_MD_FLRDEV)
1704 inode->i_rdev = old_decode_dev(body->rdev);
1706 if (body->valid & OBD_MD_FLID) {
1707 /* FID shouldn't be changed! */
1708 if (fid_is_sane(&lli->lli_fid)) {
1709 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1710 "Trying to change FID "DFID
1711 " to the "DFID", inode %lu/%u(%p)\n",
1712 PFID(&lli->lli_fid), PFID(&body->fid1),
1713 inode->i_ino, inode->i_generation, inode);
1715 lli->lli_fid = body->fid1;
1718 LASSERT(fid_seq(&lli->lli_fid) != 0);
1720 if (body->valid & OBD_MD_FLSIZE) {
1721 if (exp_connect_som(ll_i2mdexp(inode)) &&
1722 S_ISREG(inode->i_mode) && lli->lli_smd) {
1723 struct lustre_handle lockh;
1726 /* As it is possible a blocking ast has been processed
1727 * by this time, we need to check there is an UPDATE
1728 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1730 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1733 if (lli->lli_flags & (LLIF_DONE_WRITING |
1734 LLIF_EPOCH_PENDING |
1736 CERROR("ino %lu flags %u still has "
1737 "size authority! do not trust "
1738 "the size got from MDS\n",
1739 inode->i_ino, lli->lli_flags);
1741 /* Use old size assignment to avoid
1742 * deadlock bz14138 & bz14326 */
1743 i_size_write(inode, body->size);
1744 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1746 ldlm_lock_decref(&lockh, mode);
1749 /* Use old size assignment to avoid
1750 * deadlock bz14138 & bz14326 */
1751 i_size_write(inode, body->size);
1753 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1754 inode->i_ino, (unsigned long long)body->size);
1757 if (body->valid & OBD_MD_FLBLOCKS)
1758 inode->i_blocks = body->blocks;
1761 if (body->valid & OBD_MD_FLMDSCAPA) {
1762 LASSERT(md->mds_capa);
1763 ll_add_capa(inode, md->mds_capa);
1765 if (body->valid & OBD_MD_FLOSSCAPA) {
1766 LASSERT(md->oss_capa);
1767 ll_add_capa(inode, md->oss_capa);
1771 void ll_read_inode2(struct inode *inode, void *opaque)
1773 struct lustre_md *md = opaque;
1774 struct ll_inode_info *lli = ll_i2info(inode);
1777 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1778 PFID(&lli->lli_fid), inode);
1780 LASSERT(!lli->lli_smd);
1782 /* Core attributes from the MDS first. This is a new inode, and
1783 * the VFS doesn't zero times in the core inode so we have to do
1784 * it ourselves. They will be overwritten by either MDS or OST
1785 * attributes - we just need to make sure they aren't newer. */
1786 LTIME_S(inode->i_mtime) = 0;
1787 LTIME_S(inode->i_atime) = 0;
1788 LTIME_S(inode->i_ctime) = 0;
1790 ll_update_inode(inode, md);
1792 /* OIDEBUG(inode); */
1794 /* initializing backing dev info. */
1795 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1798 if (S_ISREG(inode->i_mode)) {
1799 struct ll_sb_info *sbi = ll_i2sbi(inode);
1800 inode->i_op = &ll_file_inode_operations;
1801 inode->i_fop = sbi->ll_fop;
1802 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1804 } else if (S_ISDIR(inode->i_mode)) {
1805 inode->i_op = &ll_dir_inode_operations;
1806 inode->i_fop = &ll_dir_operations;
1807 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_dir_aops;
1809 } else if (S_ISLNK(inode->i_mode)) {
1810 inode->i_op = &ll_fast_symlink_inode_operations;
1813 inode->i_op = &ll_special_inode_operations;
1815 init_special_inode(inode, inode->i_mode,
1816 kdev_t_to_nr(inode->i_rdev));
1822 void ll_delete_inode(struct inode *inode)
1824 struct ll_sb_info *sbi = ll_i2sbi(inode);
1828 rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1830 CERROR("fid_delete() failed, rc %d\n", rc);
1832 truncate_inode_pages(&inode->i_data, 0);
1834 /* Workaround for LU-118 */
1835 if (inode->i_data.nrpages) {
1836 TREE_READ_LOCK_IRQ(&inode->i_data);
1837 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1838 LASSERTF(inode->i_data.nrpages == 0,
1839 "inode=%lu/%u(%p) nrpages=%lu, see "
1840 "http://jira.whamcloud.com/browse/LU-118\n",
1841 inode->i_ino, inode->i_generation, inode,
1842 inode->i_data.nrpages);
1844 /* Workaround end */
1846 #ifdef HAVE_SBOPS_EVICT_INODE
1847 ll_clear_inode(inode);
1848 end_writeback(inode);
1856 int ll_iocontrol(struct inode *inode, struct file *file,
1857 unsigned int cmd, unsigned long arg)
1859 struct ll_sb_info *sbi = ll_i2sbi(inode);
1860 struct ptlrpc_request *req = NULL;
1865 case FSFILT_IOC_GETFLAGS: {
1866 struct mdt_body *body;
1867 struct md_op_data *op_data;
1869 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1870 0, 0, LUSTRE_OPC_ANY,
1872 if (IS_ERR(op_data))
1873 RETURN(PTR_ERR(op_data));
1875 op_data->op_valid = OBD_MD_FLFLAGS;
1876 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1877 ll_finish_md_op_data(op_data);
1879 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1883 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1885 flags = body->flags;
1887 ptlrpc_req_finished(req);
1889 RETURN(put_user(flags, (int *)arg));
1891 case FSFILT_IOC_SETFLAGS: {
1892 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1893 struct obd_info oinfo = { { { 0 } } };
1894 struct md_op_data *op_data;
1896 if (get_user(flags, (int *)arg))
1900 OBDO_ALLOC(oinfo.oi_oa);
1904 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1905 LUSTRE_OPC_ANY, NULL);
1906 if (IS_ERR(op_data))
1907 RETURN(PTR_ERR(op_data));
1909 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1910 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1911 rc = md_setattr(sbi->ll_md_exp, op_data,
1912 NULL, 0, NULL, 0, &req, NULL);
1913 ll_finish_md_op_data(op_data);
1914 ptlrpc_req_finished(req);
1916 OBDO_FREE(oinfo.oi_oa);
1921 OBDO_FREE(oinfo.oi_oa);
1922 GOTO(update_cache, rc);
1925 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1926 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1927 oinfo.oi_oa->o_flags = flags;
1928 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1930 oinfo.oi_capa = ll_mdscapa_get(inode);
1931 obdo_from_inode(oinfo.oi_oa, inode,
1932 &ll_i2info(inode)->lli_fid, 0);
1933 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1934 capa_put(oinfo.oi_capa);
1935 OBDO_FREE(oinfo.oi_oa);
1937 if (rc != -EPERM && rc != -EACCES)
1938 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1944 inode->i_flags = ll_ext_to_inode_flags(flags);
1954 int ll_flush_ctx(struct inode *inode)
1956 struct ll_sb_info *sbi = ll_i2sbi(inode);
1958 CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1960 obd_set_info_async(sbi->ll_md_exp,
1961 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1963 obd_set_info_async(sbi->ll_dt_exp,
1964 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1969 /* umount -f client means force down, don't save state */
1970 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1971 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1973 struct super_block *sb = vfsmnt->mnt_sb;
1975 void ll_umount_begin(struct super_block *sb)
1978 struct lustre_sb_info *lsi = s2lsi(sb);
1979 struct ll_sb_info *sbi = ll_s2sbi(sb);
1980 struct obd_device *obd;
1981 struct obd_ioctl_data *ioc_data;
1984 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1985 if (!(flags & MNT_FORCE)) {
1991 /* Tell the MGC we got umount -f */
1992 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1994 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1995 sb->s_count, atomic_read(&sb->s_active));
1997 obd = class_exp2obd(sbi->ll_md_exp);
1999 CERROR("Invalid MDC connection handle "LPX64"\n",
2000 sbi->ll_md_exp->exp_handle.h_cookie);
2006 obd = class_exp2obd(sbi->ll_dt_exp);
2008 CERROR("Invalid LOV connection handle "LPX64"\n",
2009 sbi->ll_dt_exp->exp_handle.h_cookie);
2015 OBD_ALLOC_PTR(ioc_data);
2017 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2018 sizeof ioc_data, ioc_data, NULL);
2020 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2021 sizeof ioc_data, ioc_data, NULL);
2023 OBD_FREE_PTR(ioc_data);
2027 /* Really, we'd like to wait until there are no requests outstanding,
2028 * and then continue. For now, we just invalidate the requests,
2029 * schedule() and sleep one second if needed, and hope.
2032 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2033 if (atomic_read(&vfsmnt->mnt_count) > 2) {
2034 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2035 cfs_time_seconds(1));
2036 if (atomic_read(&vfsmnt->mnt_count) > 2)
2037 LCONSOLE_WARN("Mount still busy with %d refs! You "
2038 "may try to umount it a bit later\n",
2039 atomic_read(&vfsmnt->mnt_count));
2046 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2048 struct ll_sb_info *sbi = ll_s2sbi(sb);
2052 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2053 read_only = *flags & MS_RDONLY;
2054 err = obd_set_info_async(sbi->ll_md_exp,
2055 sizeof(KEY_READ_ONLY),
2056 KEY_READ_ONLY, sizeof(read_only),
2059 CERROR("Failed to change the read-only flag during "
2060 "remount: %d\n", err);
2065 sb->s_flags |= MS_RDONLY;
2067 sb->s_flags &= ~MS_RDONLY;
2072 int ll_prep_inode(struct inode **inode,
2073 struct ptlrpc_request *req,
2074 struct super_block *sb)
2076 struct ll_sb_info *sbi = NULL;
2077 struct lustre_md md;
2081 LASSERT(*inode || sb);
2082 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2083 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2084 sbi->ll_md_exp, &md);
2089 ll_update_inode(*inode, &md);
2091 LASSERT(sb != NULL);
2094 * At this point server returns to client's same fid as client
2095 * generated for creating. So using ->fid1 is okay here.
2097 LASSERT(fid_is_sane(&md.body->fid1));
2099 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2100 if (*inode == NULL || IS_ERR(*inode)) {
2102 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2103 #ifdef CONFIG_FS_POSIX_ACL
2105 posix_acl_release(md.posix_acl);
2106 md.posix_acl = NULL;
2109 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2111 CERROR("new_inode -fatal: rc %d\n", rc);
2117 md_free_lustre_md(sbi->ll_md_exp, &md);
2121 int ll_obd_statfs(struct inode *inode, void *arg)
2123 struct ll_sb_info *sbi = NULL;
2124 struct obd_export *exp;
2126 struct obd_ioctl_data *data = NULL;
2130 if (!inode || !(sbi = ll_i2sbi(inode)))
2131 GOTO(out_statfs, rc = -EINVAL);
2133 rc = obd_ioctl_getdata(&buf, &len, arg);
2135 GOTO(out_statfs, rc);
2138 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2139 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2140 GOTO(out_statfs, rc = -EINVAL);
2142 if (data->ioc_inllen1 != sizeof(__u32) ||
2143 data->ioc_inllen2 != sizeof(__u32) ||
2144 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2145 data->ioc_plen2 != sizeof(struct obd_uuid))
2146 GOTO(out_statfs, rc = -EINVAL);
2148 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2149 if (type == LL_STATFS_LMV)
2150 exp = sbi->ll_md_exp;
2151 else if (type == LL_STATFS_LOV)
2152 exp = sbi->ll_dt_exp;
2154 GOTO(out_statfs, rc = -ENODEV);
2156 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2158 GOTO(out_statfs, rc);
2161 obd_ioctl_freedata(buf, len);
2165 int ll_process_config(struct lustre_cfg *lcfg)
2169 struct lprocfs_static_vars lvars;
2173 lprocfs_llite_init_vars(&lvars);
2175 /* The instance name contains the sb: lustre-client-aacfe000 */
2176 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2177 if (!ptr || !*(++ptr))
2179 if (sscanf(ptr, "%lx", &x) != 1)
2182 /* This better be a real Lustre superblock! */
2183 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2185 /* Note we have not called client_common_fill_super yet, so
2186 proc fns must be able to handle that! */
2187 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2194 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2195 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2196 struct inode *i1, struct inode *i2,
2197 const char *name, int namelen,
2198 int mode, __u32 opc, void *data)
2200 LASSERT(i1 != NULL);
2202 if (namelen > ll_i2sbi(i1)->ll_namelen)
2203 return ERR_PTR(-ENAMETOOLONG);
2205 if (op_data == NULL)
2206 OBD_ALLOC_PTR(op_data);
2208 if (op_data == NULL)
2209 return ERR_PTR(-ENOMEM);
2211 ll_i2gids(op_data->op_suppgids, i1, i2);
2212 op_data->op_fid1 = *ll_inode2fid(i1);
2213 op_data->op_capa1 = ll_mdscapa_get(i1);
2216 op_data->op_fid2 = *ll_inode2fid(i2);
2217 op_data->op_capa2 = ll_mdscapa_get(i2);
2219 fid_zero(&op_data->op_fid2);
2220 op_data->op_capa2 = NULL;
2223 op_data->op_name = name;
2224 op_data->op_namelen = namelen;
2225 op_data->op_mode = mode;
2226 op_data->op_mod_time = cfs_time_current_sec();
2227 op_data->op_fsuid = cfs_curproc_fsuid();
2228 op_data->op_fsgid = cfs_curproc_fsgid();
2229 op_data->op_cap = cfs_curproc_cap_pack();
2230 op_data->op_bias = MDS_CHECK_SPLIT;
2231 op_data->op_opc = opc;
2232 op_data->op_mds = 0;
2233 op_data->op_data = data;
2235 /* If the file is being opened after mknod() (normally due to NFS)
2236 * try to use the default stripe data from parent directory for
2237 * allocating OST objects. Try to pass the parent FID to MDS. */
2238 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2239 ll_i2info(i2)->lli_smd == NULL) {
2240 struct ll_inode_info *lli = ll_i2info(i2);
2242 cfs_spin_lock(&lli->lli_lock);
2243 if (likely(lli->lli_smd == NULL &&
2244 !fid_is_zero(&lli->lli_pfid)))
2245 op_data->op_fid1 = lli->lli_pfid;
2246 cfs_spin_unlock(&lli->lli_lock);
2247 /** We ignore parent's capability temporary. */
2253 void ll_finish_md_op_data(struct md_op_data *op_data)
2255 capa_put(op_data->op_capa1);
2256 capa_put(op_data->op_capa2);
2257 OBD_FREE_PTR(op_data);
2260 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2262 struct ll_sb_info *sbi;
2264 LASSERT((seq != NULL) && (vfs != NULL));
2265 sbi = ll_s2sbi(vfs->mnt_sb);
2267 if (sbi->ll_flags & LL_SBI_NOLCK)
2268 seq_puts(seq, ",nolock");
2270 if (sbi->ll_flags & LL_SBI_FLOCK)
2271 seq_puts(seq, ",flock");
2273 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2274 seq_puts(seq, ",localflock");
2276 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2277 seq_puts(seq, ",user_xattr");
2279 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2280 seq_puts(seq, ",lazystatfs");
2286 * Get obd name by cmd, and copy out to user space
2288 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2290 struct ll_sb_info *sbi = ll_i2sbi(inode);
2291 struct obd_device *obd;
2294 if (cmd == OBD_IOC_GETDTNAME)
2295 obd = class_exp2obd(sbi->ll_dt_exp);
2296 else if (cmd == OBD_IOC_GETMDNAME)
2297 obd = class_exp2obd(sbi->ll_md_exp);
2304 if (cfs_copy_to_user((void *)arg, obd->obd_name,
2305 strlen(obd->obd_name) + 1))