4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
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 DEFINE_SPINLOCK(ll_sb_lock);
64 #ifndef MS_HAS_NEW_AOPS
65 extern struct address_space_operations ll_aops;
67 extern struct address_space_operations_ext ll_aops;
71 #define log2(n) ffz(~(n))
74 static struct ll_sb_info *ll_init_sbi(void)
76 struct ll_sb_info *sbi = NULL;
78 unsigned long lru_page_max;
84 OBD_ALLOC(sbi, sizeof(*sbi));
88 spin_lock_init(&sbi->ll_lock);
89 mutex_init(&sbi->ll_lco.lco_lock);
90 spin_lock_init(&sbi->ll_pp_extent_lock);
91 spin_lock_init(&sbi->ll_process_lock);
92 sbi->ll_rw_stats_on = 0;
95 pages = si.totalram - si.totalhigh;
96 if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
97 #ifdef HAVE_BGL_SUPPORT
98 lru_page_max = pages / 4;
100 lru_page_max = pages / 2;
103 lru_page_max = (pages / 4) * 3;
106 /* initialize lru data */
107 cfs_atomic_set(&sbi->ll_cache.ccc_users, 0);
108 sbi->ll_cache.ccc_lru_max = lru_page_max;
109 cfs_atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
110 spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
111 CFS_INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
113 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
114 SBI_DEFAULT_READAHEAD_MAX);
115 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
116 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
117 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
118 CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
119 CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
121 ll_generate_random_uuid(uuid);
122 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
123 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
125 spin_lock(&ll_sb_lock);
126 cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
127 spin_unlock(&ll_sb_lock);
129 sbi->ll_flags |= LL_SBI_VERBOSE;
130 #ifdef ENABLE_CHECKSUM
131 sbi->ll_flags |= LL_SBI_CHECKSUM;
134 #ifdef HAVE_LRU_RESIZE_SUPPORT
135 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
138 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
139 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
141 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
145 /* metadata statahead is enabled by default */
146 sbi->ll_sa_max = LL_SA_RPC_DEF;
147 cfs_atomic_set(&sbi->ll_sa_total, 0);
148 cfs_atomic_set(&sbi->ll_sa_wrong, 0);
149 cfs_atomic_set(&sbi->ll_agl_total, 0);
150 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
155 void ll_free_sbi(struct super_block *sb)
157 struct ll_sb_info *sbi = ll_s2sbi(sb);
161 spin_lock(&ll_sb_lock);
162 cfs_list_del(&sbi->ll_list);
163 spin_unlock(&ll_sb_lock);
164 OBD_FREE(sbi, sizeof(*sbi));
169 static struct dentry_operations ll_d_root_ops = {
170 .d_compare = ll_dcompare,
171 .d_revalidate = ll_revalidate_nd,
174 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
175 struct vfsmount *mnt)
177 struct inode *root = 0;
178 struct ll_sb_info *sbi = ll_s2sbi(sb);
179 struct obd_device *obd;
180 struct obd_capa *oc = NULL;
181 struct obd_statfs *osfs = NULL;
182 struct ptlrpc_request *request = NULL;
183 struct obd_connect_data *data = NULL;
184 struct obd_uuid *uuid;
185 struct md_op_data *op_data;
186 struct lustre_md lmd;
188 int size, err, checksum;
191 obd = class_name2obd(md);
193 CERROR("MD %s: not setup or attached\n", md);
207 if (proc_lustre_fs_root) {
208 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
211 CERROR("could not register mount in /proc/fs/lustre\n");
214 /* indicate the features supported by this client */
215 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
216 OBD_CONNECT_ATTRFID |
217 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
218 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
219 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
220 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
221 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
222 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
223 OBD_CONNECT_EINPROGRESS |
224 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
225 OBD_CONNECT_LAYOUTLOCK;
227 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
228 data->ocd_connect_flags |= OBD_CONNECT_SOM;
230 #ifdef HAVE_LRU_RESIZE_SUPPORT
231 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
232 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
234 #ifdef CONFIG_FS_POSIX_ACL
235 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
238 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
239 /* flag mdc connection as lightweight, only used for test
240 * purpose, use with care */
241 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
243 data->ocd_ibits_known = MDS_INODELOCK_FULL;
244 data->ocd_version = LUSTRE_VERSION_CODE;
246 if (sb->s_flags & MS_RDONLY)
247 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
248 if (sbi->ll_flags & LL_SBI_USER_XATTR)
249 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
251 #ifdef HAVE_MS_FLOCK_LOCK
252 /* force vfs to use lustre handler for flock() calls - bug 10743 */
253 sb->s_flags |= MS_FLOCK_LOCK;
255 #ifdef MS_HAS_NEW_AOPS
256 sb->s_flags |= MS_HAS_NEW_AOPS;
259 if (sbi->ll_flags & LL_SBI_FLOCK)
260 sbi->ll_fop = &ll_file_operations_flock;
261 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
262 sbi->ll_fop = &ll_file_operations;
264 sbi->ll_fop = &ll_file_operations_noflock;
267 data->ocd_connect_flags |= OBD_CONNECT_REAL;
268 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
269 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
271 data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
273 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
275 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
276 "recovery, of which this client is not a "
277 "part. Please wait for recovery to complete,"
278 " abort, or time out.\n", md);
281 CERROR("cannot connect to %s: rc = %d\n", md, err);
285 err = obd_fid_init(sbi->ll_md_exp);
287 CERROR("Can't init metadata layer FID infrastructure, "
292 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
293 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
295 GOTO(out_md_fid, err);
297 /* This needs to be after statfs to ensure connect has finished.
298 * Note that "data" does NOT contain the valid connect reply.
299 * If connecting to a 1.8 server there will be no LMV device, so
300 * we can access the MDC export directly and exp_connect_flags will
301 * be non-zero, but if accessing an upgraded 2.1 server it will
302 * have the correct flags filled in.
303 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
304 valid = sbi->ll_md_exp->exp_connect_flags & CLIENT_CONNECT_MDT_REQD;
305 if (sbi->ll_md_exp->exp_connect_flags != 0 &&
306 valid != CLIENT_CONNECT_MDT_REQD) {
309 OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
310 obd_connect_flags2str(buf, CFS_PAGE_SIZE,
311 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
312 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
313 "feature(s) needed for correct operation "
314 "of this client (%s). Please upgrade "
315 "server or downgrade client.\n",
316 sbi->ll_md_exp->exp_obd->obd_name, buf);
317 OBD_FREE(buf, CFS_PAGE_SIZE);
318 GOTO(out_md, err = -EPROTO);
321 size = sizeof(*data);
322 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
323 KEY_CONN_DATA, &size, data, NULL);
325 CERROR("Get connect data failed: %d \n", err);
329 LASSERT(osfs->os_bsize);
330 sb->s_blocksize = osfs->os_bsize;
331 sb->s_blocksize_bits = log2(osfs->os_bsize);
332 sb->s_magic = LL_SUPER_MAGIC;
333 sb->s_maxbytes = MAX_LFS_FILESIZE;
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 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
389 LCONSOLE_INFO("Layout lock feature supported.\n");
390 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
393 obd = class_name2obd(dt);
395 CERROR("DT %s: not setup or attached\n", dt);
396 GOTO(out_md_fid, err = -ENODEV);
399 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
400 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
401 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
402 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
403 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
404 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
405 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
406 OBD_CONNECT_MAXBYTES |
407 OBD_CONNECT_EINPROGRESS |
408 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
409 OBD_CONNECT_LAYOUTLOCK;
411 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
412 data->ocd_connect_flags |= OBD_CONNECT_SOM;
414 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
415 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
416 * disabled by default, because it can still be enabled on the
417 * fly via /proc. As a consequence, we still need to come to an
418 * agreement on the supported algorithms at connect time */
419 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
421 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
422 data->ocd_cksum_types = OBD_CKSUM_ADLER;
424 data->ocd_cksum_types = cksum_types_supported_client();
427 #ifdef HAVE_LRU_RESIZE_SUPPORT
428 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
430 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
431 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
433 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
434 "ocd_grant: %d\n", data->ocd_connect_flags,
435 data->ocd_version, data->ocd_grant);
437 obd->obd_upcall.onu_owner = &sbi->ll_lco;
438 obd->obd_upcall.onu_upcall = cl_ocd_update;
440 data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
442 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
444 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
445 "recovery, of which this client is not a "
446 "part. Please wait for recovery to "
447 "complete, abort, or time out.\n", dt);
448 GOTO(out_md_fid, err);
450 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
451 GOTO(out_md_fid, err);
454 err = obd_fid_init(sbi->ll_dt_exp);
456 CERROR("Can't init data layer FID infrastructure, "
461 mutex_lock(&sbi->ll_lco.lco_lock);
462 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
463 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
464 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
465 mutex_unlock(&sbi->ll_lco.lco_lock);
467 fid_zero(&sbi->ll_root_fid);
468 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
470 CERROR("cannot mds_connect: rc = %d\n", err);
471 GOTO(out_lock_cn_cb, err);
473 if (!fid_is_sane(&sbi->ll_root_fid)) {
474 CERROR("Invalid root fid during mount\n");
475 GOTO(out_lock_cn_cb, err = -EINVAL);
477 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
479 sb->s_op = &lustre_super_operations;
480 #if THREAD_SIZE >= 8192 /*b=17630*/
481 sb->s_export_op = &lustre_export_operations;
485 * XXX: move this to after cbd setup? */
486 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
487 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
488 valid |= OBD_MD_FLRMTPERM;
489 else if (sbi->ll_flags & LL_SBI_ACL)
490 valid |= OBD_MD_FLACL;
492 OBD_ALLOC_PTR(op_data);
494 GOTO(out_lock_cn_cb, err = -ENOMEM);
496 op_data->op_fid1 = sbi->ll_root_fid;
497 op_data->op_mode = 0;
498 op_data->op_capa1 = oc;
499 op_data->op_valid = valid;
501 err = md_getattr(sbi->ll_md_exp, op_data, &request);
504 OBD_FREE_PTR(op_data);
506 CERROR("md_getattr failed for root: rc = %d\n", err);
507 GOTO(out_lock_cn_cb, err);
509 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
510 sbi->ll_md_exp, &lmd);
512 CERROR("failed to understand root inode md: rc = %d\n", err);
513 ptlrpc_req_finished (request);
514 GOTO(out_lock_cn_cb, err);
517 LASSERT(fid_is_sane(&sbi->ll_root_fid));
518 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
519 md_free_lustre_md(sbi->ll_md_exp, &lmd);
520 ptlrpc_req_finished(request);
522 if (root == NULL || IS_ERR(root)) {
524 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
525 #ifdef CONFIG_FS_POSIX_ACL
527 posix_acl_release(lmd.posix_acl);
528 lmd.posix_acl = NULL;
531 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
533 CERROR("lustre_lite: bad iget4 for root\n");
537 err = ll_close_thread_start(&sbi->ll_lcq);
539 CERROR("cannot start close thread: rc %d\n", err);
543 #ifdef CONFIG_FS_POSIX_ACL
544 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
545 rct_init(&sbi->ll_rct);
546 et_init(&sbi->ll_et);
550 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
551 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
552 KEY_CHECKSUM, sizeof(checksum), &checksum,
556 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
557 KEY_CACHE_SET, sizeof(sbi->ll_cache),
558 &sbi->ll_cache, NULL);
560 sb->s_root = d_make_root(root);
561 if (sb->s_root == NULL) {
562 CERROR("%s: can't make root dentry\n",
563 ll_get_fsname(sb, NULL, 0));
564 GOTO(out_lock_cn_cb, err = -ENOMEM);
567 #ifdef HAVE_DCACHE_LOCK
568 sb->s_root->d_op = &ll_d_root_ops;
570 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
571 d_set_d_op(sb->s_root, &ll_d_root_ops);
572 sb->s_d_op = &ll_d_ops;
575 sbi->ll_sdev_orig = sb->s_dev;
577 /* We set sb->s_dev equal on all lustre clients in order to support
578 * NFS export clustering. NFSD requires that the FSID be the same
580 /* s_dev is also used in lt_compare() to compare two fs, but that is
581 * only a node-local comparison. */
582 uuid = obd_get_uuid(sbi->ll_md_exp);
584 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
596 obd_fid_fini(sbi->ll_dt_exp);
598 obd_disconnect(sbi->ll_dt_exp);
599 sbi->ll_dt_exp = NULL;
600 /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
601 obd_zombie_barrier();
603 obd_fid_fini(sbi->ll_md_exp);
605 obd_disconnect(sbi->ll_md_exp);
606 sbi->ll_md_exp = NULL;
612 lprocfs_unregister_mountpoint(sbi);
616 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
620 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
622 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
623 KEY_MAX_EASIZE, &size, lmmsize, NULL);
625 CERROR("Get max mdsize error rc %d \n", rc);
630 void ll_dump_inode(struct inode *inode)
632 struct ll_d_hlist_node *tmp;
633 int dentry_count = 0;
635 LASSERT(inode != NULL);
637 ll_d_hlist_for_each(tmp, &inode->i_dentry)
640 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
641 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
642 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
645 void lustre_dump_dentry(struct dentry *dentry, int recur)
647 struct list_head *tmp;
650 LASSERT(dentry != NULL);
652 list_for_each(tmp, &dentry->d_subdirs)
655 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
656 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
657 dentry->d_name.len, dentry->d_name.name,
658 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
659 dentry->d_parent, dentry->d_inode, d_refcount(dentry),
660 dentry->d_flags, dentry->d_fsdata, subdirs);
661 if (dentry->d_inode != NULL)
662 ll_dump_inode(dentry->d_inode);
667 list_for_each(tmp, &dentry->d_subdirs) {
668 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
669 lustre_dump_dentry(d, recur - 1);
673 void client_common_put_super(struct super_block *sb)
675 struct ll_sb_info *sbi = ll_s2sbi(sb);
678 #ifdef CONFIG_FS_POSIX_ACL
679 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
680 et_fini(&sbi->ll_et);
681 rct_fini(&sbi->ll_rct);
685 ll_close_thread_shutdown(sbi->ll_lcq);
689 cfs_list_del(&sbi->ll_conn_chain);
691 obd_fid_fini(sbi->ll_dt_exp);
692 obd_disconnect(sbi->ll_dt_exp);
693 sbi->ll_dt_exp = NULL;
694 /* wait till all OSCs are gone, since cl_cache is accessing sbi.
696 obd_zombie_barrier();
698 lprocfs_unregister_mountpoint(sbi);
700 obd_fid_fini(sbi->ll_md_exp);
701 obd_disconnect(sbi->ll_md_exp);
702 sbi->ll_md_exp = NULL;
707 void ll_kill_super(struct super_block *sb)
709 struct ll_sb_info *sbi;
714 if (!(sb->s_flags & MS_ACTIVE))
718 /* we need restore s_dev from changed for clustred NFS before put_super
719 * because new kernels have cached s_dev and change sb->s_dev in
720 * put_super not affected real removing devices */
722 sb->s_dev = sbi->ll_sdev_orig;
726 char *ll_read_opt(const char *opt, char *data)
732 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
733 if (strncmp(opt, data, strlen(opt)))
735 if ((value = strchr(data, '=')) == NULL)
739 OBD_ALLOC(retval, strlen(value) + 1);
741 CERROR("out of memory!\n");
745 memcpy(retval, value, strlen(value)+1);
746 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
750 static inline int ll_set_opt(const char *opt, char *data, int fl)
752 if (strncmp(opt, data, strlen(opt)) != 0)
758 /* non-client-specific mount options are parsed in lmd_parse */
759 static int ll_options(char *options, int *flags)
762 char *s1 = options, *s2;
768 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
771 CDEBUG(D_SUPER, "next opt=%s\n", s1);
772 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
777 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
782 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
787 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
792 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
797 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
802 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
803 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
805 /* Ignore deprecated mount option. The client will
806 * always try to mount with ACL support, whether this
807 * is used depends on whether server supports it. */
808 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
809 "mount option 'acl'.\n");
812 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
814 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
815 "mount option 'noacl'.\n");
819 #warning "{no}acl options have been deprecated since 1.8, please remove them"
821 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
826 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
831 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
837 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
842 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
847 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
852 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
857 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
862 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
867 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
872 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
877 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
882 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
887 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
893 s2 = strchr(s1, ',');
901 void ll_lli_init(struct ll_inode_info *lli)
903 lli->lli_inode_magic = LLI_INODE_MAGIC;
905 lli->lli_ioepoch = 0;
906 lli->lli_maxbytes = MAX_LFS_FILESIZE;
907 spin_lock_init(&lli->lli_lock);
908 lli->lli_posix_acl = NULL;
909 lli->lli_remote_perms = NULL;
910 mutex_init(&lli->lli_rmtperm_mutex);
911 /* Do not set lli_fid, it has been initialized already. */
912 fid_zero(&lli->lli_pfid);
913 CFS_INIT_LIST_HEAD(&lli->lli_close_list);
914 CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
915 cfs_atomic_set(&lli->lli_open_count, 0);
916 lli->lli_mds_capa = NULL;
917 lli->lli_rmtperm_time = 0;
918 lli->lli_pending_och = NULL;
919 lli->lli_mds_read_och = NULL;
920 lli->lli_mds_write_och = NULL;
921 lli->lli_mds_exec_och = NULL;
922 lli->lli_open_fd_read_count = 0;
923 lli->lli_open_fd_write_count = 0;
924 lli->lli_open_fd_exec_count = 0;
925 mutex_init(&lli->lli_och_mutex);
926 spin_lock_init(&lli->lli_agl_lock);
927 lli->lli_has_smd = false;
928 lli->lli_layout_gen = LL_LAYOUT_GEN_ZERO;
929 lli->lli_clob = NULL;
931 LASSERT(lli->lli_vfs_inode.i_mode != 0);
932 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
933 mutex_init(&lli->lli_readdir_mutex);
934 lli->lli_opendir_key = NULL;
936 lli->lli_def_acl = NULL;
937 spin_lock_init(&lli->lli_sa_lock);
938 lli->lli_opendir_pid = 0;
940 sema_init(&lli->lli_size_sem, 1);
941 lli->lli_size_sem_owner = NULL;
942 lli->lli_symlink_name = NULL;
943 init_rwsem(&lli->lli_trunc_sem);
944 mutex_init(&lli->lli_write_mutex);
945 init_rwsem(&lli->lli_glimpse_sem);
946 lli->lli_glimpse_time = 0;
947 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
948 lli->lli_agl_index = 0;
949 lli->lli_async_rc = 0;
951 mutex_init(&lli->lli_layout_mutex);
954 static inline int ll_bdi_register(struct backing_dev_info *bdi)
956 #ifdef HAVE_BDI_REGISTER
957 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
960 bdi->name = "lustre";
962 return bdi_register(bdi, NULL, "lustre-%d",
963 atomic_inc_return(&ll_bdi_num));
969 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
971 struct lustre_profile *lprof = NULL;
972 struct lustre_sb_info *lsi = s2lsi(sb);
973 struct ll_sb_info *sbi;
974 char *dt = NULL, *md = NULL;
975 char *profilenm = get_profile_name(sb);
976 struct config_llog_instance *cfg;
977 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
978 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
982 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
990 /* client additional sb info */
991 lsi->lsi_llsbi = sbi = ll_init_sbi();
993 cfs_module_put(THIS_MODULE);
998 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1000 GOTO(out_free, err);
1002 err = ll_bdi_init(&lsi->lsi_bdi);
1004 GOTO(out_free, err);
1005 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1006 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1007 err = ll_bdi_register(&lsi->lsi_bdi);
1009 GOTO(out_free, err);
1012 sb->s_bdi = &lsi->lsi_bdi;
1015 /* Generate a string unique to this super, in case some joker tries
1016 to mount the same fs at two mount points.
1017 Use the address of the super itself.*/
1018 cfg->cfg_instance = sb;
1019 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1020 cfg->cfg_callback = class_config_llog_handler;
1021 /* set up client obds */
1022 err = lustre_process_log(sb, profilenm, cfg);
1024 CERROR("Unable to process log: %d\n", err);
1025 GOTO(out_free, err);
1028 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1029 lprof = class_get_profile(profilenm);
1030 if (lprof == NULL) {
1031 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1032 " read from the MGS. Does that filesystem "
1033 "exist?\n", profilenm);
1034 GOTO(out_free, err = -EINVAL);
1036 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1037 lprof->lp_md, lprof->lp_dt);
1039 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1041 GOTO(out_free, err = -ENOMEM);
1042 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1044 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1046 GOTO(out_free, err = -ENOMEM);
1047 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1049 /* connections, registrations, sb setup */
1050 err = client_common_fill_super(sb, md, dt, mnt);
1054 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1056 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1059 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1060 LCONSOLE_WARN("Mounted %s\n", profilenm);
1064 } /* ll_fill_super */
1067 void lu_context_keys_dump(void);
1069 void ll_put_super(struct super_block *sb)
1071 struct config_llog_instance cfg;
1072 struct obd_device *obd;
1073 struct lustre_sb_info *lsi = s2lsi(sb);
1074 struct ll_sb_info *sbi = ll_s2sbi(sb);
1075 char *profilenm = get_profile_name(sb);
1076 int force = 1, next;
1079 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1081 ll_print_capa_stat(sbi);
1083 cfg.cfg_instance = sb;
1084 lustre_end_log(sb, profilenm, &cfg);
1086 if (sbi->ll_md_exp) {
1087 obd = class_exp2obd(sbi->ll_md_exp);
1089 force = obd->obd_force;
1092 /* We need to set force before the lov_disconnect in
1093 lustre_common_put_super, since l_d cleans up osc's as well. */
1096 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1098 obd->obd_force = force;
1103 /* Only if client_common_fill_super succeeded */
1104 client_common_put_super(sb);
1108 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1109 class_manual_cleanup(obd);
1112 if (sbi->ll_flags & LL_SBI_VERBOSE)
1113 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1116 class_del_profile(profilenm);
1118 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1119 ll_bdi_destroy(&lsi->lsi_bdi);
1120 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1124 lsi->lsi_llsbi = NULL;
1126 lustre_common_put_super(sb);
1128 cl_env_cache_purge(~0);
1130 cfs_module_put(THIS_MODULE);
1133 } /* client_put_super */
1135 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1137 struct inode *inode = NULL;
1138 /* NOTE: we depend on atomic igrab() -bzzz */
1139 lock_res_and_lock(lock);
1140 if (lock->l_ast_data) {
1141 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1142 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1143 inode = igrab(lock->l_ast_data);
1145 inode = lock->l_ast_data;
1146 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1147 D_WARNING, lock, "l_ast_data %p is "
1148 "bogus: magic %08x", lock->l_ast_data,
1149 lli->lli_inode_magic);
1153 unlock_res_and_lock(lock);
1157 static int null_if_equal(struct ldlm_lock *lock, void *data)
1159 if (data == lock->l_ast_data) {
1160 lock->l_ast_data = NULL;
1162 if (lock->l_req_mode != lock->l_granted_mode)
1163 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1166 return LDLM_ITER_CONTINUE;
1169 void ll_clear_inode(struct inode *inode)
1171 struct ll_inode_info *lli = ll_i2info(inode);
1172 struct ll_sb_info *sbi = ll_i2sbi(inode);
1175 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1176 inode->i_generation, inode);
1178 if (S_ISDIR(inode->i_mode)) {
1179 /* these should have been cleared in ll_file_release */
1180 LASSERT(lli->lli_opendir_key == NULL);
1181 LASSERT(lli->lli_sai == NULL);
1182 LASSERT(lli->lli_opendir_pid == 0);
1185 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1186 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1187 null_if_equal, inode);
1189 LASSERT(!lli->lli_open_fd_write_count);
1190 LASSERT(!lli->lli_open_fd_read_count);
1191 LASSERT(!lli->lli_open_fd_exec_count);
1193 if (lli->lli_mds_write_och)
1194 ll_md_real_close(inode, FMODE_WRITE);
1195 if (lli->lli_mds_exec_och)
1196 ll_md_real_close(inode, FMODE_EXEC);
1197 if (lli->lli_mds_read_och)
1198 ll_md_real_close(inode, FMODE_READ);
1200 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1201 OBD_FREE(lli->lli_symlink_name,
1202 strlen(lli->lli_symlink_name) + 1);
1203 lli->lli_symlink_name = NULL;
1206 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1207 LASSERT(lli->lli_posix_acl == NULL);
1208 if (lli->lli_remote_perms) {
1209 free_rmtperm_hash(lli->lli_remote_perms);
1210 lli->lli_remote_perms = NULL;
1213 #ifdef CONFIG_FS_POSIX_ACL
1214 else if (lli->lli_posix_acl) {
1215 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1216 LASSERT(lli->lli_remote_perms == NULL);
1217 posix_acl_release(lli->lli_posix_acl);
1218 lli->lli_posix_acl = NULL;
1221 lli->lli_inode_magic = LLI_INODE_DEAD;
1223 ll_clear_inode_capas(inode);
1224 if (!S_ISDIR(inode->i_mode))
1225 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1228 * XXX This has to be done before lsm is freed below, because
1229 * cl_object still uses inode lsm.
1231 cl_inode_fini(inode);
1232 lli->lli_has_smd = false;
1237 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1238 struct md_open_data **mod)
1240 struct lustre_md md;
1241 struct inode *inode = dentry->d_inode;
1242 struct ll_sb_info *sbi = ll_i2sbi(inode);
1243 struct ptlrpc_request *request = NULL;
1247 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1248 LUSTRE_OPC_ANY, NULL);
1249 if (IS_ERR(op_data))
1250 RETURN(PTR_ERR(op_data));
1252 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1255 ptlrpc_req_finished(request);
1256 if (rc == -ENOENT) {
1258 /* Unlinked special device node? Or just a race?
1259 * Pretend we done everything. */
1260 if (!S_ISREG(inode->i_mode) &&
1261 !S_ISDIR(inode->i_mode)) {
1262 ia_valid = op_data->op_attr.ia_valid;
1263 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1264 rc = simple_setattr(dentry, &op_data->op_attr);
1265 op_data->op_attr.ia_valid = ia_valid;
1267 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1268 CERROR("md_setattr fails: rc = %d\n", rc);
1273 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1274 sbi->ll_md_exp, &md);
1276 ptlrpc_req_finished(request);
1280 /* We want to adjust timestamps.
1281 * If there is at least some data in file, we cleared ATTR_SIZE
1282 * to avoid update size, otherwise it is important to do.(SOM case)
1284 ia_valid = op_data->op_attr.ia_valid;
1285 /* Since we set ATTR_*_SET flags above, and already done permission
1286 * check, So don't let inode_change_ok() check it again. */
1287 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1288 rc = simple_setattr(dentry, &op_data->op_attr);
1289 op_data->op_attr.ia_valid = ia_valid;
1291 /* Extract epoch data if obtained. */
1292 op_data->op_handle = md.body->handle;
1293 op_data->op_ioepoch = md.body->ioepoch;
1295 ll_update_inode(inode, &md);
1296 ptlrpc_req_finished(request);
1301 /* Close IO epoch and send Size-on-MDS attribute update. */
1302 static int ll_setattr_done_writing(struct inode *inode,
1303 struct md_op_data *op_data,
1304 struct md_open_data *mod)
1306 struct ll_inode_info *lli = ll_i2info(inode);
1310 LASSERT(op_data != NULL);
1311 if (!S_ISREG(inode->i_mode))
1314 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1315 op_data->op_ioepoch, PFID(&lli->lli_fid));
1317 op_data->op_flags = MF_EPOCH_CLOSE;
1318 ll_done_writing_attr(inode, op_data);
1319 ll_pack_inode2opdata(inode, op_data, NULL);
1321 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1322 if (rc == -EAGAIN) {
1323 /* MDS has instructed us to obtain Size-on-MDS attribute
1324 * from OSTs and send setattr to back to MDS. */
1325 rc = ll_som_update(inode, op_data);
1327 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1333 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1335 struct obd_capa *capa;
1338 if (attr->ia_valid & ATTR_SIZE)
1339 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1341 capa = ll_mdscapa_get(inode);
1343 rc = cl_setattr_ost(inode, attr, capa);
1345 if (attr->ia_valid & ATTR_SIZE)
1346 ll_truncate_free_capa(capa);
1353 #ifndef HAVE_VFS_INODE_NEWSIZE_OK
1355 * inode_newsize_ok - may this inode be truncated to a given size
1356 * @inode: the inode to be truncated
1357 * @offset: the new size to assign to the inode
1358 * @Returns: 0 on success, -ve errno on failure
1360 * inode_newsize_ok will check filesystem limits and ulimits to check that the
1361 * new inode size is within limits. inode_newsize_ok will also send SIGXFSZ
1362 * when necessary. Caller must not proceed with inode size change if failure is
1363 * returned. @inode must be a file (not directory), with appropriate
1364 * permissions to allow truncate (inode_newsize_ok does NOT check these
1367 * inode_newsize_ok must be called with i_mutex held.
1369 int inode_newsize_ok(const struct inode *inode, loff_t offset)
1371 if (inode->i_size < offset) {
1372 unsigned long limit;
1374 limit = rlimit(RLIMIT_FSIZE);
1375 if (limit != RLIM_INFINITY && offset > limit)
1377 if (offset > inode->i_sb->s_maxbytes)
1381 * truncation of in-use swapfiles is disallowed - it would
1382 * cause subsequent swapout to scribble on the now-freed
1385 if (IS_SWAPFILE(inode))
1391 send_sig(SIGXFSZ, current, 0);
1397 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1398 * object(s) determine the file size and mtime. Otherwise, the MDS will
1399 * keep these values until such a time that objects are allocated for it.
1400 * We do the MDS operations first, as it is checking permissions for us.
1401 * We don't to the MDS RPC if there is nothing that we want to store there,
1402 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1403 * going to do an RPC anyways.
1405 * If we are doing a truncate, we will send the mtime and ctime updates
1406 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1407 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1410 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1412 struct lov_stripe_md *lsm;
1413 struct inode *inode = dentry->d_inode;
1414 struct ll_inode_info *lli = ll_i2info(inode);
1415 struct md_op_data *op_data = NULL;
1416 struct md_open_data *mod = NULL;
1417 int ia_valid = attr->ia_valid;
1418 int rc = 0, rc1 = 0;
1421 CDEBUG(D_VFSTRACE, "%s: setattr inode %p/fid:"DFID" from %llu to %llu, "
1422 "valid %x\n", ll_get_fsname(inode->i_sb, NULL, 0), inode,
1423 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1426 if (ia_valid & ATTR_SIZE) {
1427 /* Check new size against VFS/VM file size limit and rlimit */
1428 rc = inode_newsize_ok(inode, attr->ia_size);
1432 /* The maximum Lustre file size is variable, based on the
1433 * OST maximum object size and number of stripes. This
1434 * needs another check in addition to the VFS check above. */
1435 if (attr->ia_size > ll_file_maxbytes(inode)) {
1436 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1437 PFID(&lli->lli_fid), attr->ia_size,
1438 ll_file_maxbytes(inode));
1442 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1445 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1446 if (ia_valid & TIMES_SET_FLAGS) {
1447 if (cfs_curproc_fsuid() != inode->i_uid &&
1448 !cfs_capable(CFS_CAP_FOWNER))
1452 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1453 if (attr->ia_valid & ATTR_CTIME) {
1454 attr->ia_ctime = CFS_CURRENT_TIME;
1455 attr->ia_valid |= ATTR_CTIME_SET;
1457 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1458 attr->ia_atime = CFS_CURRENT_TIME;
1459 attr->ia_valid |= ATTR_ATIME_SET;
1461 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1462 attr->ia_mtime = CFS_CURRENT_TIME;
1463 attr->ia_valid |= ATTR_MTIME_SET;
1466 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1467 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1468 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1469 cfs_time_current_sec());
1471 /* If we are changing file size, file content is modified, flag it. */
1472 if (attr->ia_valid & ATTR_SIZE) {
1473 spin_lock(&lli->lli_lock);
1474 lli->lli_flags |= LLIF_DATA_MODIFIED;
1475 spin_unlock(&lli->lli_lock);
1478 /* We always do an MDS RPC, even if we're only changing the size;
1479 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1481 OBD_ALLOC_PTR(op_data);
1482 if (op_data == NULL)
1485 if (!S_ISDIR(inode->i_mode)) {
1486 if (ia_valid & ATTR_SIZE)
1487 inode_dio_write_done(inode);
1488 mutex_unlock(&inode->i_mutex);
1489 down_write(&lli->lli_trunc_sem);
1490 mutex_lock(&inode->i_mutex);
1491 if (ia_valid & ATTR_SIZE)
1492 inode_dio_wait(inode);
1495 /* We need a steady stripe configuration for setattr to avoid
1497 lsm = ccc_inode_lsm_get(inode);
1499 /* NB: ATTR_SIZE will only be set after this point if the size
1500 * resides on the MDS, ie, this file has no objects. */
1502 attr->ia_valid &= ~ATTR_SIZE;
1503 /* can't call ll_setattr_ost() while holding a refcount of lsm */
1504 ccc_inode_lsm_put(inode, lsm);
1506 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1508 /* Open epoch for truncate. */
1509 if (exp_connect_som(ll_i2mdexp(inode)) &&
1510 (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1511 op_data->op_flags = MF_EPOCH_OPEN;
1513 rc = ll_md_setattr(dentry, op_data, &mod);
1517 /* RPC to MDT is sent, cancel data modification flag */
1518 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1519 spin_lock(&lli->lli_lock);
1520 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1521 spin_unlock(&lli->lli_lock);
1524 ll_ioepoch_open(lli, op_data->op_ioepoch);
1525 if (!S_ISREG(inode->i_mode))
1528 if (ia_valid & ATTR_SIZE)
1529 attr->ia_valid |= ATTR_SIZE;
1530 if (ia_valid & (ATTR_SIZE |
1531 ATTR_ATIME | ATTR_ATIME_SET |
1532 ATTR_MTIME | ATTR_MTIME_SET))
1533 /* For truncate and utimes sending attributes to OSTs, setting
1534 * mtime/atime to the past will be performed under PW [0:EOF]
1535 * extent lock (new_size:EOF for truncate). It may seem
1536 * excessive to send mtime/atime updates to OSTs when not
1537 * setting times to past, but it is necessary due to possible
1538 * time de-synchronization between MDT inode and OST objects */
1539 rc = ll_setattr_ost(inode, attr);
1543 if (op_data->op_ioepoch) {
1544 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1548 ll_finish_md_op_data(op_data);
1550 if (!S_ISDIR(inode->i_mode))
1551 up_write(&lli->lli_trunc_sem);
1553 ll_stats_ops_tally(ll_i2sbi(inode), (ia_valid & ATTR_SIZE) ?
1554 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1559 int ll_setattr(struct dentry *de, struct iattr *attr)
1561 int mode = de->d_inode->i_mode;
1563 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1564 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1565 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1567 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1568 (ATTR_SIZE|ATTR_MODE)) &&
1569 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1570 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1571 !(attr->ia_mode & S_ISGID))))
1572 attr->ia_valid |= ATTR_FORCE;
1574 if ((mode & S_ISUID) &&
1575 !(attr->ia_mode & S_ISUID) &&
1576 !(attr->ia_valid & ATTR_KILL_SUID))
1577 attr->ia_valid |= ATTR_KILL_SUID;
1579 if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1580 !(attr->ia_mode & S_ISGID) &&
1581 !(attr->ia_valid & ATTR_KILL_SGID))
1582 attr->ia_valid |= ATTR_KILL_SGID;
1584 return ll_setattr_raw(de, attr);
1587 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1588 __u64 max_age, __u32 flags)
1590 struct ll_sb_info *sbi = ll_s2sbi(sb);
1591 struct obd_statfs obd_osfs;
1595 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1597 CERROR("md_statfs fails: rc = %d\n", rc);
1601 osfs->os_type = sb->s_magic;
1603 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1604 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1606 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1607 flags |= OBD_STATFS_NODELAY;
1609 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1611 CERROR("obd_statfs fails: rc = %d\n", rc);
1615 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1616 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1619 osfs->os_bsize = obd_osfs.os_bsize;
1620 osfs->os_blocks = obd_osfs.os_blocks;
1621 osfs->os_bfree = obd_osfs.os_bfree;
1622 osfs->os_bavail = obd_osfs.os_bavail;
1624 /* If we don't have as many objects free on the OST as inodes
1625 * on the MDS, we reduce the total number of inodes to
1626 * compensate, so that the "inodes in use" number is correct.
1628 if (obd_osfs.os_ffree < osfs->os_ffree) {
1629 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1631 osfs->os_ffree = obd_osfs.os_ffree;
1636 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1638 struct super_block *sb = de->d_sb;
1639 struct obd_statfs osfs;
1642 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1643 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1645 /* Some amount of caching on the client is allowed */
1646 rc = ll_statfs_internal(sb, &osfs,
1647 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1652 statfs_unpack(sfs, &osfs);
1654 /* We need to downshift for all 32-bit kernels, because we can't
1655 * tell if the kernel is being called via sys_statfs64() or not.
1656 * Stop before overflowing f_bsize - in which case it is better
1657 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1658 if (sizeof(long) < 8) {
1659 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1662 osfs.os_blocks >>= 1;
1663 osfs.os_bfree >>= 1;
1664 osfs.os_bavail >>= 1;
1668 sfs->f_blocks = osfs.os_blocks;
1669 sfs->f_bfree = osfs.os_bfree;
1670 sfs->f_bavail = osfs.os_bavail;
1675 void ll_inode_size_lock(struct inode *inode)
1677 struct ll_inode_info *lli;
1679 LASSERT(!S_ISDIR(inode->i_mode));
1681 lli = ll_i2info(inode);
1682 LASSERT(lli->lli_size_sem_owner != current);
1683 down(&lli->lli_size_sem);
1684 LASSERT(lli->lli_size_sem_owner == NULL);
1685 lli->lli_size_sem_owner = current;
1688 void ll_inode_size_unlock(struct inode *inode)
1690 struct ll_inode_info *lli;
1692 lli = ll_i2info(inode);
1693 LASSERT(lli->lli_size_sem_owner == current);
1694 lli->lli_size_sem_owner = NULL;
1695 up(&lli->lli_size_sem);
1698 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1700 struct ll_inode_info *lli = ll_i2info(inode);
1701 struct mdt_body *body = md->body;
1702 struct lov_stripe_md *lsm = md->lsm;
1703 struct ll_sb_info *sbi = ll_i2sbi(inode);
1705 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1707 lli->lli_maxbytes = lsm->lsm_maxbytes;
1708 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1709 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1712 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1713 if (body->valid & OBD_MD_FLRMTPERM)
1714 ll_update_remote_perm(inode, md->remote_perm);
1716 #ifdef CONFIG_FS_POSIX_ACL
1717 else if (body->valid & OBD_MD_FLACL) {
1718 spin_lock(&lli->lli_lock);
1719 if (lli->lli_posix_acl)
1720 posix_acl_release(lli->lli_posix_acl);
1721 lli->lli_posix_acl = md->posix_acl;
1722 spin_unlock(&lli->lli_lock);
1725 inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1726 inode->i_generation = cl_fid_build_gen(&body->fid1);
1728 if (body->valid & OBD_MD_FLATIME) {
1729 if (body->atime > LTIME_S(inode->i_atime))
1730 LTIME_S(inode->i_atime) = body->atime;
1731 lli->lli_lvb.lvb_atime = body->atime;
1733 if (body->valid & OBD_MD_FLMTIME) {
1734 if (body->mtime > LTIME_S(inode->i_mtime)) {
1735 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1736 "to "LPU64"\n", inode->i_ino,
1737 LTIME_S(inode->i_mtime), body->mtime);
1738 LTIME_S(inode->i_mtime) = body->mtime;
1740 lli->lli_lvb.lvb_mtime = body->mtime;
1742 if (body->valid & OBD_MD_FLCTIME) {
1743 if (body->ctime > LTIME_S(inode->i_ctime))
1744 LTIME_S(inode->i_ctime) = body->ctime;
1745 lli->lli_lvb.lvb_ctime = body->ctime;
1747 if (body->valid & OBD_MD_FLMODE)
1748 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1749 if (body->valid & OBD_MD_FLTYPE)
1750 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1751 LASSERT(inode->i_mode != 0);
1752 if (S_ISREG(inode->i_mode)) {
1753 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1755 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1757 if (body->valid & OBD_MD_FLUID)
1758 inode->i_uid = body->uid;
1759 if (body->valid & OBD_MD_FLGID)
1760 inode->i_gid = body->gid;
1761 if (body->valid & OBD_MD_FLFLAGS)
1762 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1763 if (body->valid & OBD_MD_FLNLINK)
1764 set_nlink(inode, body->nlink);
1765 if (body->valid & OBD_MD_FLRDEV)
1766 inode->i_rdev = old_decode_dev(body->rdev);
1768 if (body->valid & OBD_MD_FLID) {
1769 /* FID shouldn't be changed! */
1770 if (fid_is_sane(&lli->lli_fid)) {
1771 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1772 "Trying to change FID "DFID
1773 " to the "DFID", inode %lu/%u(%p)\n",
1774 PFID(&lli->lli_fid), PFID(&body->fid1),
1775 inode->i_ino, inode->i_generation, inode);
1777 lli->lli_fid = body->fid1;
1780 LASSERT(fid_seq(&lli->lli_fid) != 0);
1782 if (body->valid & OBD_MD_FLSIZE) {
1783 if (exp_connect_som(ll_i2mdexp(inode)) &&
1784 S_ISREG(inode->i_mode) && lli->lli_has_smd) {
1785 struct lustre_handle lockh;
1788 /* As it is possible a blocking ast has been processed
1789 * by this time, we need to check there is an UPDATE
1790 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1792 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1793 &lockh, LDLM_FL_CBPENDING);
1795 if (lli->lli_flags & (LLIF_DONE_WRITING |
1796 LLIF_EPOCH_PENDING |
1798 CERROR("ino %lu flags %u still has "
1799 "size authority! do not trust "
1800 "the size got from MDS\n",
1801 inode->i_ino, lli->lli_flags);
1803 /* Use old size assignment to avoid
1804 * deadlock bz14138 & bz14326 */
1805 i_size_write(inode, body->size);
1806 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1808 ldlm_lock_decref(&lockh, mode);
1811 /* Use old size assignment to avoid
1812 * deadlock bz14138 & bz14326 */
1813 i_size_write(inode, body->size);
1815 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1816 inode->i_ino, (unsigned long long)body->size);
1819 if (body->valid & OBD_MD_FLBLOCKS)
1820 inode->i_blocks = body->blocks;
1823 if (body->valid & OBD_MD_FLMDSCAPA) {
1824 LASSERT(md->mds_capa);
1825 ll_add_capa(inode, md->mds_capa);
1827 if (body->valid & OBD_MD_FLOSSCAPA) {
1828 LASSERT(md->oss_capa);
1829 ll_add_capa(inode, md->oss_capa);
1833 void ll_read_inode2(struct inode *inode, void *opaque)
1835 struct lustre_md *md = opaque;
1836 struct ll_inode_info *lli = ll_i2info(inode);
1839 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1840 PFID(&lli->lli_fid), inode);
1842 LASSERT(!lli->lli_has_smd);
1844 /* Core attributes from the MDS first. This is a new inode, and
1845 * the VFS doesn't zero times in the core inode so we have to do
1846 * it ourselves. They will be overwritten by either MDS or OST
1847 * attributes - we just need to make sure they aren't newer. */
1848 LTIME_S(inode->i_mtime) = 0;
1849 LTIME_S(inode->i_atime) = 0;
1850 LTIME_S(inode->i_ctime) = 0;
1852 ll_update_inode(inode, md);
1854 /* OIDEBUG(inode); */
1856 /* initializing backing dev info. */
1857 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1860 if (S_ISREG(inode->i_mode)) {
1861 struct ll_sb_info *sbi = ll_i2sbi(inode);
1862 inode->i_op = &ll_file_inode_operations;
1863 inode->i_fop = sbi->ll_fop;
1864 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1866 } else if (S_ISDIR(inode->i_mode)) {
1867 inode->i_op = &ll_dir_inode_operations;
1868 inode->i_fop = &ll_dir_operations;
1870 } else if (S_ISLNK(inode->i_mode)) {
1871 inode->i_op = &ll_fast_symlink_inode_operations;
1874 inode->i_op = &ll_special_inode_operations;
1876 init_special_inode(inode, inode->i_mode,
1883 void ll_delete_inode(struct inode *inode)
1885 struct cl_inode_info *lli = cl_i2info(inode);
1888 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1889 /* discard all dirty pages before truncating them, required by
1890 * osc_extent implementation at LU-1030. */
1891 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_DISCARD);
1893 truncate_inode_pages(&inode->i_data, 0);
1895 /* Workaround for LU-118 */
1896 if (inode->i_data.nrpages) {
1897 TREE_READ_LOCK_IRQ(&inode->i_data);
1898 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1899 LASSERTF(inode->i_data.nrpages == 0,
1900 "inode=%lu/%u(%p) nrpages=%lu, see "
1901 "http://jira.whamcloud.com/browse/LU-118\n",
1902 inode->i_ino, inode->i_generation, inode,
1903 inode->i_data.nrpages);
1905 /* Workaround end */
1907 #ifdef HAVE_SBOPS_EVICT_INODE
1908 ll_clear_inode(inode);
1915 int ll_iocontrol(struct inode *inode, struct file *file,
1916 unsigned int cmd, unsigned long arg)
1918 struct ll_sb_info *sbi = ll_i2sbi(inode);
1919 struct ptlrpc_request *req = NULL;
1924 case FSFILT_IOC_GETFLAGS: {
1925 struct mdt_body *body;
1926 struct md_op_data *op_data;
1928 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1929 0, 0, LUSTRE_OPC_ANY,
1931 if (IS_ERR(op_data))
1932 RETURN(PTR_ERR(op_data));
1934 op_data->op_valid = OBD_MD_FLFLAGS;
1935 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1936 ll_finish_md_op_data(op_data);
1938 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1942 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1944 flags = body->flags;
1946 ptlrpc_req_finished(req);
1948 RETURN(put_user(flags, (int *)arg));
1950 case FSFILT_IOC_SETFLAGS: {
1951 struct lov_stripe_md *lsm;
1952 struct obd_info oinfo = { { { 0 } } };
1953 struct md_op_data *op_data;
1955 if (get_user(flags, (int *)arg))
1958 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1959 LUSTRE_OPC_ANY, NULL);
1960 if (IS_ERR(op_data))
1961 RETURN(PTR_ERR(op_data));
1963 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1964 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1965 rc = md_setattr(sbi->ll_md_exp, op_data,
1966 NULL, 0, NULL, 0, &req, NULL);
1967 ll_finish_md_op_data(op_data);
1968 ptlrpc_req_finished(req);
1972 inode->i_flags = ll_ext_to_inode_flags(flags);
1974 lsm = ccc_inode_lsm_get(inode);
1978 OBDO_ALLOC(oinfo.oi_oa);
1980 ccc_inode_lsm_put(inode, lsm);
1984 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1985 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1986 oinfo.oi_oa->o_flags = flags;
1987 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1989 oinfo.oi_capa = ll_mdscapa_get(inode);
1990 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1991 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1992 capa_put(oinfo.oi_capa);
1993 OBDO_FREE(oinfo.oi_oa);
1994 ccc_inode_lsm_put(inode, lsm);
1996 if (rc && rc != -EPERM && rc != -EACCES)
1997 CERROR("osc_setattr_async fails: rc = %d\n", rc);
2008 int ll_flush_ctx(struct inode *inode)
2010 struct ll_sb_info *sbi = ll_i2sbi(inode);
2012 CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
2014 obd_set_info_async(NULL, sbi->ll_md_exp,
2015 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2017 obd_set_info_async(NULL, sbi->ll_dt_exp,
2018 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2023 /* umount -f client means force down, don't save state */
2024 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2025 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2027 struct super_block *sb = vfsmnt->mnt_sb;
2029 void ll_umount_begin(struct super_block *sb)
2032 struct ll_sb_info *sbi = ll_s2sbi(sb);
2033 struct obd_device *obd;
2034 struct obd_ioctl_data *ioc_data;
2037 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2038 if (!(flags & MNT_FORCE)) {
2044 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2045 sb->s_count, atomic_read(&sb->s_active));
2047 obd = class_exp2obd(sbi->ll_md_exp);
2049 CERROR("Invalid MDC connection handle "LPX64"\n",
2050 sbi->ll_md_exp->exp_handle.h_cookie);
2056 obd = class_exp2obd(sbi->ll_dt_exp);
2058 CERROR("Invalid LOV connection handle "LPX64"\n",
2059 sbi->ll_dt_exp->exp_handle.h_cookie);
2065 OBD_ALLOC_PTR(ioc_data);
2067 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2068 sizeof ioc_data, ioc_data, NULL);
2070 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2071 sizeof ioc_data, ioc_data, NULL);
2073 OBD_FREE_PTR(ioc_data);
2077 /* Really, we'd like to wait until there are no requests outstanding,
2078 * and then continue. For now, we just invalidate the requests,
2079 * schedule() and sleep one second if needed, and hope.
2082 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2083 if (atomic_read(&vfsmnt->mnt_count) > 2) {
2084 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2085 cfs_time_seconds(1));
2086 if (atomic_read(&vfsmnt->mnt_count) > 2)
2087 LCONSOLE_WARN("Mount still busy with %d refs! You "
2088 "may try to umount it a bit later\n",
2089 atomic_read(&vfsmnt->mnt_count));
2096 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2098 struct ll_sb_info *sbi = ll_s2sbi(sb);
2099 char *profilenm = get_profile_name(sb);
2103 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2104 read_only = *flags & MS_RDONLY;
2105 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2106 sizeof(KEY_READ_ONLY),
2107 KEY_READ_ONLY, sizeof(read_only),
2110 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2111 profilenm, read_only ?
2112 "read-only" : "read-write", err);
2117 sb->s_flags |= MS_RDONLY;
2119 sb->s_flags &= ~MS_RDONLY;
2121 if (sbi->ll_flags & LL_SBI_VERBOSE)
2122 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2123 read_only ? "read-only" : "read-write");
2128 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2129 struct super_block *sb, struct lookup_intent *it)
2131 struct ll_sb_info *sbi = NULL;
2132 struct lustre_md md;
2136 LASSERT(*inode || sb);
2137 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2138 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2139 sbi->ll_md_exp, &md);
2144 ll_update_inode(*inode, &md);
2146 LASSERT(sb != NULL);
2149 * At this point server returns to client's same fid as client
2150 * generated for creating. So using ->fid1 is okay here.
2152 LASSERT(fid_is_sane(&md.body->fid1));
2154 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2155 if (*inode == NULL || IS_ERR(*inode)) {
2156 #ifdef CONFIG_FS_POSIX_ACL
2158 posix_acl_release(md.posix_acl);
2159 md.posix_acl = NULL;
2162 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2164 CERROR("new_inode -fatal: rc %d\n", rc);
2169 /* Handling piggyback layout lock.
2170 * Layout lock can be piggybacked by getattr and open request.
2171 * The lsm can be applied to inode only if it comes with a layout lock
2172 * otherwise correct layout may be overwritten, for example:
2173 * 1. proc1: mdt returns a lsm but not granting layout
2174 * 2. layout was changed by another client
2175 * 3. proc2: refresh layout and layout lock granted
2176 * 4. proc1: to apply a stale layout */
2177 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2178 struct lustre_handle lockh;
2179 struct ldlm_lock *lock;
2181 lockh.cookie = it->d.lustre.it_lock_handle;
2182 lock = ldlm_handle2lock(&lockh);
2183 LASSERT(lock != NULL);
2184 if (ldlm_has_layout(lock)) {
2185 struct cl_object_conf conf;
2187 memset(&conf, 0, sizeof(conf));
2188 conf.coc_opc = OBJECT_CONF_SET;
2189 conf.coc_inode = *inode;
2190 conf.coc_lock = lock;
2191 conf.u.coc_md = &md;
2192 (void)ll_layout_conf(*inode, &conf);
2194 LDLM_LOCK_PUT(lock);
2199 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2200 md_free_lustre_md(sbi->ll_md_exp, &md);
2204 int ll_obd_statfs(struct inode *inode, void *arg)
2206 struct ll_sb_info *sbi = NULL;
2207 struct obd_export *exp;
2209 struct obd_ioctl_data *data = NULL;
2214 if (!inode || !(sbi = ll_i2sbi(inode)))
2215 GOTO(out_statfs, rc = -EINVAL);
2217 rc = obd_ioctl_getdata(&buf, &len, arg);
2219 GOTO(out_statfs, rc);
2222 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2223 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2224 GOTO(out_statfs, rc = -EINVAL);
2226 if (data->ioc_inllen1 != sizeof(__u32) ||
2227 data->ioc_inllen2 != sizeof(__u32) ||
2228 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2229 data->ioc_plen2 != sizeof(struct obd_uuid))
2230 GOTO(out_statfs, rc = -EINVAL);
2232 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2233 if (type & LL_STATFS_LMV)
2234 exp = sbi->ll_md_exp;
2235 else if (type & LL_STATFS_LOV)
2236 exp = sbi->ll_dt_exp;
2238 GOTO(out_statfs, rc = -ENODEV);
2240 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2241 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2243 GOTO(out_statfs, rc);
2246 obd_ioctl_freedata(buf, len);
2250 int ll_process_config(struct lustre_cfg *lcfg)
2254 struct lprocfs_static_vars lvars;
2258 lprocfs_llite_init_vars(&lvars);
2260 /* The instance name contains the sb: lustre-client-aacfe000 */
2261 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2262 if (!ptr || !*(++ptr))
2264 if (sscanf(ptr, "%lx", &x) != 1)
2267 /* This better be a real Lustre superblock! */
2268 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2270 /* Note we have not called client_common_fill_super yet, so
2271 proc fns must be able to handle that! */
2272 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2279 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2280 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2281 struct inode *i1, struct inode *i2,
2282 const char *name, int namelen,
2283 int mode, __u32 opc, void *data)
2285 LASSERT(i1 != NULL);
2287 if (namelen > ll_i2sbi(i1)->ll_namelen)
2288 return ERR_PTR(-ENAMETOOLONG);
2290 if (op_data == NULL)
2291 OBD_ALLOC_PTR(op_data);
2293 if (op_data == NULL)
2294 return ERR_PTR(-ENOMEM);
2296 ll_i2gids(op_data->op_suppgids, i1, i2);
2297 op_data->op_fid1 = *ll_inode2fid(i1);
2298 op_data->op_capa1 = ll_mdscapa_get(i1);
2301 op_data->op_fid2 = *ll_inode2fid(i2);
2302 op_data->op_capa2 = ll_mdscapa_get(i2);
2304 fid_zero(&op_data->op_fid2);
2305 op_data->op_capa2 = NULL;
2308 op_data->op_name = name;
2309 op_data->op_namelen = namelen;
2310 op_data->op_mode = mode;
2311 op_data->op_mod_time = cfs_time_current_sec();
2312 op_data->op_fsuid = cfs_curproc_fsuid();
2313 op_data->op_fsgid = cfs_curproc_fsgid();
2314 op_data->op_cap = cfs_curproc_cap_pack();
2315 op_data->op_bias = MDS_CHECK_SPLIT;
2316 op_data->op_opc = opc;
2317 op_data->op_mds = 0;
2318 op_data->op_data = data;
2320 /* If the file is being opened after mknod() (normally due to NFS)
2321 * try to use the default stripe data from parent directory for
2322 * allocating OST objects. Try to pass the parent FID to MDS. */
2323 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2324 !ll_i2info(i2)->lli_has_smd) {
2325 struct ll_inode_info *lli = ll_i2info(i2);
2327 spin_lock(&lli->lli_lock);
2328 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2329 op_data->op_fid1 = lli->lli_pfid;
2330 spin_unlock(&lli->lli_lock);
2331 /** We ignore parent's capability temporary. */
2334 /* When called by ll_setattr_raw, file is i1. */
2335 if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2336 op_data->op_bias |= MDS_DATA_MODIFIED;
2341 void ll_finish_md_op_data(struct md_op_data *op_data)
2343 capa_put(op_data->op_capa1);
2344 capa_put(op_data->op_capa2);
2345 OBD_FREE_PTR(op_data);
2348 #ifdef HAVE_SUPEROPS_USE_DENTRY
2349 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2351 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2354 struct ll_sb_info *sbi;
2356 #ifdef HAVE_SUPEROPS_USE_DENTRY
2357 LASSERT((seq != NULL) && (dentry != NULL));
2358 sbi = ll_s2sbi(dentry->d_sb);
2360 LASSERT((seq != NULL) && (vfs != NULL));
2361 sbi = ll_s2sbi(vfs->mnt_sb);
2364 if (sbi->ll_flags & LL_SBI_NOLCK)
2365 seq_puts(seq, ",nolock");
2367 if (sbi->ll_flags & LL_SBI_FLOCK)
2368 seq_puts(seq, ",flock");
2370 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2371 seq_puts(seq, ",localflock");
2373 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2374 seq_puts(seq, ",user_xattr");
2376 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2377 seq_puts(seq, ",lazystatfs");
2379 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2380 seq_puts(seq, ",user_fid2path");
2386 * Get obd name by cmd, and copy out to user space
2388 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2390 struct ll_sb_info *sbi = ll_i2sbi(inode);
2391 struct obd_device *obd;
2394 if (cmd == OBD_IOC_GETDTNAME)
2395 obd = class_exp2obd(sbi->ll_dt_exp);
2396 else if (cmd == OBD_IOC_GETMDNAME)
2397 obd = class_exp2obd(sbi->ll_md_exp);
2404 if (cfs_copy_to_user((void *)arg, obd->obd_name,
2405 strlen(obd->obd_name) + 1))
2412 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2413 * fsname will be returned in this buffer; otherwise, a static buffer will be
2414 * used to store the fsname and returned to caller.
2416 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2418 static char fsname_static[MTI_NAME_MAXLEN];
2419 struct lustre_sb_info *lsi = s2lsi(sb);
2424 /* this means the caller wants to use static buffer
2425 * and it doesn't care about race. Usually this is
2426 * in error reporting path */
2427 buf = fsname_static;
2428 buflen = sizeof(fsname_static);
2431 len = strlen(lsi->lsi_lmd->lmd_profile);
2432 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2433 if (ptr && (strcmp(ptr, "-client") == 0))
2436 if (unlikely(len >= buflen))
2438 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2444 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2448 #ifdef HAVE_FS_STRUCT_USE_PATH
2452 p.mnt = current->fs->root.mnt;
2454 path = d_path(&p, buf, bufsize);
2457 path = d_path(dentry, current->fs->rootmnt, buf, bufsize);
2463 void ll_dirty_page_discard_warn(cfs_page_t *page, int ioret)
2465 char *buf, *path = NULL;
2466 struct dentry *dentry = NULL;
2467 struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2469 /* this can be called inside spin lock so use GFP_ATOMIC. */
2470 buf = (char *)__get_free_page(GFP_ATOMIC);
2472 dentry = d_find_alias(page->mapping->host);
2474 path = ll_d_path(dentry, buf, PAGE_SIZE);
2477 CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2478 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2479 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2480 PFID(&obj->cob_header.coh_lu.loh_fid),
2481 (path && !IS_ERR(path)) ? path : "", ioret);
2487 free_page((unsigned long)buf);