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, Whamcloud, Inc.
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;
226 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
227 data->ocd_connect_flags |= OBD_CONNECT_SOM;
229 #ifdef HAVE_LRU_RESIZE_SUPPORT
230 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
231 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
233 #ifdef CONFIG_FS_POSIX_ACL
234 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
237 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
238 /* flag mdc connection as lightweight, only used for test
239 * purpose, use with care */
240 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
242 data->ocd_ibits_known = MDS_INODELOCK_FULL;
243 data->ocd_version = LUSTRE_VERSION_CODE;
245 if (sb->s_flags & MS_RDONLY)
246 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
247 if (sbi->ll_flags & LL_SBI_USER_XATTR)
248 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
250 #ifdef HAVE_MS_FLOCK_LOCK
251 /* force vfs to use lustre handler for flock() calls - bug 10743 */
252 sb->s_flags |= MS_FLOCK_LOCK;
254 #ifdef MS_HAS_NEW_AOPS
255 sb->s_flags |= MS_HAS_NEW_AOPS;
258 if (sbi->ll_flags & LL_SBI_FLOCK)
259 sbi->ll_fop = &ll_file_operations_flock;
260 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
261 sbi->ll_fop = &ll_file_operations;
263 sbi->ll_fop = &ll_file_operations_noflock;
266 data->ocd_connect_flags |= OBD_CONNECT_REAL;
267 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
268 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
270 data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
272 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
274 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
275 "recovery, of which this client is not a "
276 "part. Please wait for recovery to complete,"
277 " abort, or time out.\n", md);
280 CERROR("cannot connect to %s: rc = %d\n", md, err);
284 err = obd_fid_init(sbi->ll_md_exp);
286 CERROR("Can't init metadata layer FID infrastructure, "
291 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
292 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
294 GOTO(out_md_fid, err);
296 /* This needs to be after statfs to ensure connect has finished.
297 * Note that "data" does NOT contain the valid connect reply.
298 * If connecting to a 1.8 server there will be no LMV device, so
299 * we can access the MDC export directly and exp_connect_flags will
300 * be non-zero, but if accessing an upgraded 2.1 server it will
301 * have the correct flags filled in.
302 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
303 valid = sbi->ll_md_exp->exp_connect_flags & CLIENT_CONNECT_MDT_REQD;
304 if (sbi->ll_md_exp->exp_connect_flags != 0 &&
305 valid != CLIENT_CONNECT_MDT_REQD) {
308 OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
309 obd_connect_flags2str(buf, CFS_PAGE_SIZE,
310 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
311 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
312 "feature(s) needed for correct operation "
313 "of this client (%s). Please upgrade "
314 "server or downgrade client.\n",
315 sbi->ll_md_exp->exp_obd->obd_name, buf);
316 OBD_FREE(buf, CFS_PAGE_SIZE);
317 GOTO(out_md, err = -EPROTO);
320 size = sizeof(*data);
321 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
322 KEY_CONN_DATA, &size, data, NULL);
324 CERROR("Get connect data failed: %d \n", err);
328 LASSERT(osfs->os_bsize);
329 sb->s_blocksize = osfs->os_bsize;
330 sb->s_blocksize_bits = log2(osfs->os_bsize);
331 sb->s_magic = LL_SUPER_MAGIC;
332 sb->s_maxbytes = MAX_LFS_FILESIZE;
333 sbi->ll_namelen = osfs->os_namelen;
334 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
336 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
337 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
338 LCONSOLE_INFO("Disabling user_xattr feature because "
339 "it is not supported on the server\n");
340 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
343 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
345 sb->s_flags |= MS_POSIXACL;
347 sbi->ll_flags |= LL_SBI_ACL;
349 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
351 sb->s_flags &= ~MS_POSIXACL;
353 sbi->ll_flags &= ~LL_SBI_ACL;
356 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
357 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
358 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
359 LCONSOLE_INFO("client is set as remote by default.\n");
362 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
363 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
364 LCONSOLE_INFO("client claims to be remote, but server "
365 "rejected, forced to be local.\n");
369 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
370 LCONSOLE_INFO("client enabled MDS capability!\n");
371 sbi->ll_flags |= LL_SBI_MDS_CAPA;
374 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
375 LCONSOLE_INFO("client enabled OSS capability!\n");
376 sbi->ll_flags |= LL_SBI_OSS_CAPA;
379 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
380 sbi->ll_flags |= LL_SBI_64BIT_HASH;
382 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
383 sbi->ll_md_brw_size = data->ocd_brw_size;
385 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
387 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
388 LCONSOLE_INFO("Layout lock feature supported.\n");
389 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
392 obd = class_name2obd(dt);
394 CERROR("DT %s: not setup or attached\n", dt);
395 GOTO(out_md_fid, err = -ENODEV);
398 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
399 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
400 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
401 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
402 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
403 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
404 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
405 OBD_CONNECT_MAXBYTES |
406 OBD_CONNECT_EINPROGRESS |
407 OBD_CONNECT_JOBSTATS;
409 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
410 data->ocd_connect_flags |= OBD_CONNECT_SOM;
412 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
413 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
414 * disabled by default, because it can still be enabled on the
415 * fly via /proc. As a consequence, we still need to come to an
416 * agreement on the supported algorithms at connect time */
417 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
419 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
420 data->ocd_cksum_types = OBD_CKSUM_ADLER;
422 data->ocd_cksum_types = cksum_types_supported_client();
425 #ifdef HAVE_LRU_RESIZE_SUPPORT
426 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
428 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
429 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
431 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
432 "ocd_grant: %d\n", data->ocd_connect_flags,
433 data->ocd_version, data->ocd_grant);
435 obd->obd_upcall.onu_owner = &sbi->ll_lco;
436 obd->obd_upcall.onu_upcall = cl_ocd_update;
438 data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
440 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
442 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
443 "recovery, of which this client is not a "
444 "part. Please wait for recovery to "
445 "complete, abort, or time out.\n", dt);
446 GOTO(out_md_fid, err);
448 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
449 GOTO(out_md_fid, err);
452 err = obd_fid_init(sbi->ll_dt_exp);
454 CERROR("Can't init data layer FID infrastructure, "
459 mutex_lock(&sbi->ll_lco.lco_lock);
460 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
461 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
462 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
463 mutex_unlock(&sbi->ll_lco.lco_lock);
465 fid_zero(&sbi->ll_root_fid);
466 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
468 CERROR("cannot mds_connect: rc = %d\n", err);
469 GOTO(out_lock_cn_cb, err);
471 if (!fid_is_sane(&sbi->ll_root_fid)) {
472 CERROR("Invalid root fid during mount\n");
473 GOTO(out_lock_cn_cb, err = -EINVAL);
475 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
477 sb->s_op = &lustre_super_operations;
478 #if THREAD_SIZE >= 8192 /*b=17630*/
479 sb->s_export_op = &lustre_export_operations;
483 * XXX: move this to after cbd setup? */
484 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
485 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
486 valid |= OBD_MD_FLRMTPERM;
487 else if (sbi->ll_flags & LL_SBI_ACL)
488 valid |= OBD_MD_FLACL;
490 OBD_ALLOC_PTR(op_data);
492 GOTO(out_lock_cn_cb, err = -ENOMEM);
494 op_data->op_fid1 = sbi->ll_root_fid;
495 op_data->op_mode = 0;
496 op_data->op_capa1 = oc;
497 op_data->op_valid = valid;
499 err = md_getattr(sbi->ll_md_exp, op_data, &request);
502 OBD_FREE_PTR(op_data);
504 CERROR("md_getattr failed for root: rc = %d\n", err);
505 GOTO(out_lock_cn_cb, err);
507 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
508 sbi->ll_md_exp, &lmd);
510 CERROR("failed to understand root inode md: rc = %d\n", err);
511 ptlrpc_req_finished (request);
512 GOTO(out_lock_cn_cb, err);
515 LASSERT(fid_is_sane(&sbi->ll_root_fid));
516 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
517 md_free_lustre_md(sbi->ll_md_exp, &lmd);
518 ptlrpc_req_finished(request);
520 if (root == NULL || IS_ERR(root)) {
522 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
523 #ifdef CONFIG_FS_POSIX_ACL
525 posix_acl_release(lmd.posix_acl);
526 lmd.posix_acl = NULL;
529 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
531 CERROR("lustre_lite: bad iget4 for root\n");
535 err = ll_close_thread_start(&sbi->ll_lcq);
537 CERROR("cannot start close thread: rc %d\n", err);
541 #ifdef CONFIG_FS_POSIX_ACL
542 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
543 rct_init(&sbi->ll_rct);
544 et_init(&sbi->ll_et);
548 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
549 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
550 KEY_CHECKSUM, sizeof(checksum), &checksum,
554 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
555 KEY_CACHE_SET, sizeof(sbi->ll_cache),
556 &sbi->ll_cache, NULL);
558 sb->s_root = d_alloc_root(root);
559 #ifdef HAVE_DCACHE_LOCK
560 sb->s_root->d_op = &ll_d_root_ops;
562 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
563 d_set_d_op(sb->s_root, &ll_d_root_ops);
564 sb->s_d_op = &ll_d_ops;
567 sbi->ll_sdev_orig = sb->s_dev;
569 /* We set sb->s_dev equal on all lustre clients in order to support
570 * NFS export clustering. NFSD requires that the FSID be the same
572 /* s_dev is also used in lt_compare() to compare two fs, but that is
573 * only a node-local comparison. */
574 uuid = obd_get_uuid(sbi->ll_md_exp);
576 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
588 obd_fid_fini(sbi->ll_dt_exp);
590 obd_disconnect(sbi->ll_dt_exp);
591 sbi->ll_dt_exp = NULL;
593 obd_fid_fini(sbi->ll_md_exp);
595 obd_disconnect(sbi->ll_md_exp);
596 sbi->ll_md_exp = NULL;
602 lprocfs_unregister_mountpoint(sbi);
606 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
610 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
612 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
613 KEY_MAX_EASIZE, &size, lmmsize, NULL);
615 CERROR("Get max mdsize error rc %d \n", rc);
620 void ll_dump_inode(struct inode *inode)
622 struct list_head *tmp;
623 int dentry_count = 0;
625 LASSERT(inode != NULL);
627 list_for_each(tmp, &inode->i_dentry)
630 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
631 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
632 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
635 void lustre_dump_dentry(struct dentry *dentry, int recur)
637 struct list_head *tmp;
640 LASSERT(dentry != NULL);
642 list_for_each(tmp, &dentry->d_subdirs)
645 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
646 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
647 dentry->d_name.len, dentry->d_name.name,
648 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
649 dentry->d_parent, dentry->d_inode, d_refcount(dentry),
650 dentry->d_flags, dentry->d_fsdata, subdirs);
651 if (dentry->d_inode != NULL)
652 ll_dump_inode(dentry->d_inode);
657 list_for_each(tmp, &dentry->d_subdirs) {
658 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
659 lustre_dump_dentry(d, recur - 1);
663 void client_common_put_super(struct super_block *sb)
665 struct ll_sb_info *sbi = ll_s2sbi(sb);
668 #ifdef CONFIG_FS_POSIX_ACL
669 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
670 et_fini(&sbi->ll_et);
671 rct_fini(&sbi->ll_rct);
675 ll_close_thread_shutdown(sbi->ll_lcq);
679 cfs_list_del(&sbi->ll_conn_chain);
681 obd_fid_fini(sbi->ll_dt_exp);
682 obd_disconnect(sbi->ll_dt_exp);
683 sbi->ll_dt_exp = NULL;
685 lprocfs_unregister_mountpoint(sbi);
687 obd_fid_fini(sbi->ll_md_exp);
688 obd_disconnect(sbi->ll_md_exp);
689 sbi->ll_md_exp = NULL;
694 void ll_kill_super(struct super_block *sb)
696 struct ll_sb_info *sbi;
701 if (!(sb->s_flags & MS_ACTIVE))
705 /* we need restore s_dev from changed for clustred NFS before put_super
706 * because new kernels have cached s_dev and change sb->s_dev in
707 * put_super not affected real removing devices */
709 sb->s_dev = sbi->ll_sdev_orig;
713 char *ll_read_opt(const char *opt, char *data)
719 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
720 if (strncmp(opt, data, strlen(opt)))
722 if ((value = strchr(data, '=')) == NULL)
726 OBD_ALLOC(retval, strlen(value) + 1);
728 CERROR("out of memory!\n");
732 memcpy(retval, value, strlen(value)+1);
733 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
737 static inline int ll_set_opt(const char *opt, char *data, int fl)
739 if (strncmp(opt, data, strlen(opt)) != 0)
745 /* non-client-specific mount options are parsed in lmd_parse */
746 static int ll_options(char *options, int *flags)
749 char *s1 = options, *s2;
755 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
758 CDEBUG(D_SUPER, "next opt=%s\n", s1);
759 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
764 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
769 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
774 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
779 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
784 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
789 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
790 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
792 /* Ignore deprecated mount option. The client will
793 * always try to mount with ACL support, whether this
794 * is used depends on whether server supports it. */
795 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
796 "mount option 'acl'.\n");
799 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
801 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
802 "mount option 'noacl'.\n");
806 #warning "{no}acl options have been deprecated since 1.8, please remove them"
808 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
813 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
818 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
824 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
829 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
834 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
839 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
844 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
849 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
854 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
859 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
864 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
869 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
874 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
880 s2 = strchr(s1, ',');
888 void ll_lli_init(struct ll_inode_info *lli)
890 lli->lli_inode_magic = LLI_INODE_MAGIC;
892 lli->lli_ioepoch = 0;
893 lli->lli_maxbytes = MAX_LFS_FILESIZE;
894 spin_lock_init(&lli->lli_lock);
895 lli->lli_posix_acl = NULL;
896 lli->lli_remote_perms = NULL;
897 mutex_init(&lli->lli_rmtperm_mutex);
898 /* Do not set lli_fid, it has been initialized already. */
899 fid_zero(&lli->lli_pfid);
900 CFS_INIT_LIST_HEAD(&lli->lli_close_list);
901 CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
902 cfs_atomic_set(&lli->lli_open_count, 0);
903 lli->lli_mds_capa = NULL;
904 lli->lli_rmtperm_time = 0;
905 lli->lli_pending_och = NULL;
906 lli->lli_mds_read_och = NULL;
907 lli->lli_mds_write_och = NULL;
908 lli->lli_mds_exec_och = NULL;
909 lli->lli_open_fd_read_count = 0;
910 lli->lli_open_fd_write_count = 0;
911 lli->lli_open_fd_exec_count = 0;
912 mutex_init(&lli->lli_och_mutex);
913 spin_lock_init(&lli->lli_agl_lock);
914 lli->lli_has_smd = false;
915 lli->lli_clob = NULL;
917 LASSERT(lli->lli_vfs_inode.i_mode != 0);
918 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
919 mutex_init(&lli->lli_readdir_mutex);
920 lli->lli_opendir_key = NULL;
922 lli->lli_def_acl = NULL;
923 spin_lock_init(&lli->lli_sa_lock);
924 lli->lli_opendir_pid = 0;
926 sema_init(&lli->lli_size_sem, 1);
927 lli->lli_size_sem_owner = NULL;
928 lli->lli_symlink_name = NULL;
929 init_rwsem(&lli->lli_trunc_sem);
930 mutex_init(&lli->lli_write_mutex);
931 init_rwsem(&lli->lli_glimpse_sem);
932 lli->lli_glimpse_time = 0;
933 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
934 lli->lli_agl_index = 0;
935 lli->lli_async_rc = 0;
937 mutex_init(&lli->lli_layout_mutex);
940 static inline int ll_bdi_register(struct backing_dev_info *bdi)
942 #ifdef HAVE_BDI_REGISTER
943 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
946 bdi->name = "lustre";
948 return bdi_register(bdi, NULL, "lustre-%d",
949 atomic_inc_return(&ll_bdi_num));
955 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
957 struct lustre_profile *lprof = NULL;
958 struct lustre_sb_info *lsi = s2lsi(sb);
959 struct ll_sb_info *sbi;
960 char *dt = NULL, *md = NULL;
961 char *profilenm = get_profile_name(sb);
962 struct config_llog_instance *cfg;
963 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
964 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
968 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
976 /* client additional sb info */
977 lsi->lsi_llsbi = sbi = ll_init_sbi();
979 cfs_module_put(THIS_MODULE);
984 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
988 err = ll_bdi_init(&lsi->lsi_bdi);
991 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
992 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
993 err = ll_bdi_register(&lsi->lsi_bdi);
998 sb->s_bdi = &lsi->lsi_bdi;
1001 /* Generate a string unique to this super, in case some joker tries
1002 to mount the same fs at two mount points.
1003 Use the address of the super itself.*/
1004 cfg->cfg_instance = sb;
1005 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1006 cfg->cfg_callback = class_config_llog_handler;
1007 /* set up client obds */
1008 err = lustre_process_log(sb, profilenm, cfg);
1010 CERROR("Unable to process log: %d\n", err);
1011 GOTO(out_free, err);
1014 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1015 lprof = class_get_profile(profilenm);
1016 if (lprof == NULL) {
1017 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1018 " read from the MGS. Does that filesystem "
1019 "exist?\n", profilenm);
1020 GOTO(out_free, err = -EINVAL);
1022 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1023 lprof->lp_md, lprof->lp_dt);
1025 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1027 GOTO(out_free, err = -ENOMEM);
1028 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1030 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1032 GOTO(out_free, err = -ENOMEM);
1033 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1035 /* connections, registrations, sb setup */
1036 err = client_common_fill_super(sb, md, dt, mnt);
1040 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1042 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1045 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1046 LCONSOLE_WARN("Mounted %s\n", profilenm);
1050 } /* ll_fill_super */
1053 void lu_context_keys_dump(void);
1055 void ll_put_super(struct super_block *sb)
1057 struct config_llog_instance cfg;
1058 struct obd_device *obd;
1059 struct lustre_sb_info *lsi = s2lsi(sb);
1060 struct ll_sb_info *sbi = ll_s2sbi(sb);
1061 char *profilenm = get_profile_name(sb);
1062 int force = 1, next;
1065 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1067 ll_print_capa_stat(sbi);
1069 cfg.cfg_instance = sb;
1070 lustre_end_log(sb, profilenm, &cfg);
1072 if (sbi->ll_md_exp) {
1073 obd = class_exp2obd(sbi->ll_md_exp);
1075 force = obd->obd_force;
1078 /* We need to set force before the lov_disconnect in
1079 lustre_common_put_super, since l_d cleans up osc's as well. */
1082 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1084 obd->obd_force = force;
1089 /* Only if client_common_fill_super succeeded */
1090 client_common_put_super(sb);
1094 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1095 class_manual_cleanup(obd);
1098 if (sbi->ll_flags & LL_SBI_VERBOSE)
1099 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1102 class_del_profile(profilenm);
1104 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1105 ll_bdi_destroy(&lsi->lsi_bdi);
1106 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1110 lsi->lsi_llsbi = NULL;
1112 lustre_common_put_super(sb);
1114 cl_env_cache_purge(~0);
1116 cfs_module_put(THIS_MODULE);
1119 } /* client_put_super */
1121 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1123 struct inode *inode = NULL;
1124 /* NOTE: we depend on atomic igrab() -bzzz */
1125 lock_res_and_lock(lock);
1126 if (lock->l_ast_data) {
1127 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1128 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1129 inode = igrab(lock->l_ast_data);
1131 inode = lock->l_ast_data;
1132 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1133 D_WARNING, lock, "l_ast_data %p is "
1134 "bogus: magic %08x", lock->l_ast_data,
1135 lli->lli_inode_magic);
1139 unlock_res_and_lock(lock);
1143 static int null_if_equal(struct ldlm_lock *lock, void *data)
1145 if (data == lock->l_ast_data) {
1146 lock->l_ast_data = NULL;
1148 if (lock->l_req_mode != lock->l_granted_mode)
1149 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1152 return LDLM_ITER_CONTINUE;
1155 void ll_clear_inode(struct inode *inode)
1157 struct ll_inode_info *lli = ll_i2info(inode);
1158 struct ll_sb_info *sbi = ll_i2sbi(inode);
1161 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1162 inode->i_generation, inode);
1164 if (S_ISDIR(inode->i_mode)) {
1165 /* these should have been cleared in ll_file_release */
1166 LASSERT(lli->lli_opendir_key == NULL);
1167 LASSERT(lli->lli_sai == NULL);
1168 LASSERT(lli->lli_opendir_pid == 0);
1171 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1172 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1173 null_if_equal, inode);
1175 LASSERT(!lli->lli_open_fd_write_count);
1176 LASSERT(!lli->lli_open_fd_read_count);
1177 LASSERT(!lli->lli_open_fd_exec_count);
1179 if (lli->lli_mds_write_och)
1180 ll_md_real_close(inode, FMODE_WRITE);
1181 if (lli->lli_mds_exec_och)
1182 ll_md_real_close(inode, FMODE_EXEC);
1183 if (lli->lli_mds_read_och)
1184 ll_md_real_close(inode, FMODE_READ);
1186 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1187 OBD_FREE(lli->lli_symlink_name,
1188 strlen(lli->lli_symlink_name) + 1);
1189 lli->lli_symlink_name = NULL;
1192 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1193 LASSERT(lli->lli_posix_acl == NULL);
1194 if (lli->lli_remote_perms) {
1195 free_rmtperm_hash(lli->lli_remote_perms);
1196 lli->lli_remote_perms = NULL;
1199 #ifdef CONFIG_FS_POSIX_ACL
1200 else if (lli->lli_posix_acl) {
1201 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1202 LASSERT(lli->lli_remote_perms == NULL);
1203 posix_acl_release(lli->lli_posix_acl);
1204 lli->lli_posix_acl = NULL;
1207 lli->lli_inode_magic = LLI_INODE_DEAD;
1209 ll_clear_inode_capas(inode);
1210 if (!S_ISDIR(inode->i_mode))
1211 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1214 * XXX This has to be done before lsm is freed below, because
1215 * cl_object still uses inode lsm.
1217 cl_inode_fini(inode);
1218 lli->lli_has_smd = false;
1223 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1224 struct md_open_data **mod)
1226 struct lustre_md md;
1227 struct inode *inode = dentry->d_inode;
1228 struct ll_sb_info *sbi = ll_i2sbi(inode);
1229 struct ptlrpc_request *request = NULL;
1233 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1234 LUSTRE_OPC_ANY, NULL);
1235 if (IS_ERR(op_data))
1236 RETURN(PTR_ERR(op_data));
1238 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1241 ptlrpc_req_finished(request);
1242 if (rc == -ENOENT) {
1244 /* Unlinked special device node? Or just a race?
1245 * Pretend we done everything. */
1246 if (!S_ISREG(inode->i_mode) &&
1247 !S_ISDIR(inode->i_mode)) {
1248 ia_valid = op_data->op_attr.ia_valid;
1249 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1250 rc = simple_setattr(dentry, &op_data->op_attr);
1251 op_data->op_attr.ia_valid = ia_valid;
1253 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1254 CERROR("md_setattr fails: rc = %d\n", rc);
1259 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1260 sbi->ll_md_exp, &md);
1262 ptlrpc_req_finished(request);
1266 /* We want to adjust timestamps.
1267 * If there is at least some data in file, we cleared ATTR_SIZE
1268 * to avoid update size, otherwise it is important to do.(SOM case)
1270 ia_valid = op_data->op_attr.ia_valid;
1271 /* Since we set ATTR_*_SET flags above, and already done permission
1272 * check, So don't let inode_change_ok() check it again. */
1273 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1274 rc = simple_setattr(dentry, &op_data->op_attr);
1275 op_data->op_attr.ia_valid = ia_valid;
1277 /* Extract epoch data if obtained. */
1278 op_data->op_handle = md.body->handle;
1279 op_data->op_ioepoch = md.body->ioepoch;
1281 ll_update_inode(inode, &md);
1282 ptlrpc_req_finished(request);
1287 /* Close IO epoch and send Size-on-MDS attribute update. */
1288 static int ll_setattr_done_writing(struct inode *inode,
1289 struct md_op_data *op_data,
1290 struct md_open_data *mod)
1292 struct ll_inode_info *lli = ll_i2info(inode);
1296 LASSERT(op_data != NULL);
1297 if (!S_ISREG(inode->i_mode))
1300 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1301 op_data->op_ioepoch, PFID(&lli->lli_fid));
1303 op_data->op_flags = MF_EPOCH_CLOSE;
1304 ll_done_writing_attr(inode, op_data);
1305 ll_pack_inode2opdata(inode, op_data, NULL);
1307 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1308 if (rc == -EAGAIN) {
1309 /* MDS has instructed us to obtain Size-on-MDS attribute
1310 * from OSTs and send setattr to back to MDS. */
1311 rc = ll_som_update(inode, op_data);
1313 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1319 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1321 struct obd_capa *capa;
1324 if (attr->ia_valid & ATTR_SIZE)
1325 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1327 capa = ll_mdscapa_get(inode);
1329 rc = cl_setattr_ost(inode, attr, capa);
1331 if (attr->ia_valid & ATTR_SIZE)
1332 ll_truncate_free_capa(capa);
1339 #ifndef HAVE_VFS_INODE_NEWSIZE_OK
1341 * inode_newsize_ok - may this inode be truncated to a given size
1342 * @inode: the inode to be truncated
1343 * @offset: the new size to assign to the inode
1344 * @Returns: 0 on success, -ve errno on failure
1346 * inode_newsize_ok will check filesystem limits and ulimits to check that the
1347 * new inode size is within limits. inode_newsize_ok will also send SIGXFSZ
1348 * when necessary. Caller must not proceed with inode size change if failure is
1349 * returned. @inode must be a file (not directory), with appropriate
1350 * permissions to allow truncate (inode_newsize_ok does NOT check these
1353 * inode_newsize_ok must be called with i_mutex held.
1355 int inode_newsize_ok(const struct inode *inode, loff_t offset)
1357 if (inode->i_size < offset) {
1358 unsigned long limit;
1360 limit = rlimit(RLIMIT_FSIZE);
1361 if (limit != RLIM_INFINITY && offset > limit)
1363 if (offset > inode->i_sb->s_maxbytes)
1367 * truncation of in-use swapfiles is disallowed - it would
1368 * cause subsequent swapout to scribble on the now-freed
1371 if (IS_SWAPFILE(inode))
1377 send_sig(SIGXFSZ, current, 0);
1383 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1384 * object(s) determine the file size and mtime. Otherwise, the MDS will
1385 * keep these values until such a time that objects are allocated for it.
1386 * We do the MDS operations first, as it is checking permissions for us.
1387 * We don't to the MDS RPC if there is nothing that we want to store there,
1388 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1389 * going to do an RPC anyways.
1391 * If we are doing a truncate, we will send the mtime and ctime updates
1392 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1393 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1396 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1398 struct lov_stripe_md *lsm;
1399 struct inode *inode = dentry->d_inode;
1400 struct ll_inode_info *lli = ll_i2info(inode);
1401 struct md_op_data *op_data = NULL;
1402 struct md_open_data *mod = NULL;
1403 int ia_valid = attr->ia_valid;
1404 int rc = 0, rc1 = 0;
1407 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu (%p) to %llu, valid %x\n",
1409 inode,i_size_read(inode),
1412 if (ia_valid & ATTR_SIZE) {
1413 /* Check new size against VFS/VM file size limit and rlimit */
1414 rc = inode_newsize_ok(inode, attr->ia_size);
1418 /* The maximum Lustre file size is variable, based on the
1419 * OST maximum object size and number of stripes. This
1420 * needs another check in addition to the VFS check above. */
1421 if (attr->ia_size > ll_file_maxbytes(inode)) {
1422 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1423 PFID(&lli->lli_fid), attr->ia_size,
1424 ll_file_maxbytes(inode));
1428 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1431 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1432 if (ia_valid & TIMES_SET_FLAGS) {
1433 if (cfs_curproc_fsuid() != inode->i_uid &&
1434 !cfs_capable(CFS_CAP_FOWNER))
1438 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1439 if (attr->ia_valid & ATTR_CTIME) {
1440 attr->ia_ctime = CFS_CURRENT_TIME;
1441 attr->ia_valid |= ATTR_CTIME_SET;
1443 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1444 attr->ia_atime = CFS_CURRENT_TIME;
1445 attr->ia_valid |= ATTR_ATIME_SET;
1447 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1448 attr->ia_mtime = CFS_CURRENT_TIME;
1449 attr->ia_valid |= ATTR_MTIME_SET;
1452 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1453 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1454 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1455 cfs_time_current_sec());
1457 /* We always do an MDS RPC, even if we're only changing the size;
1458 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1460 OBD_ALLOC_PTR(op_data);
1461 if (op_data == NULL)
1464 if (!S_ISDIR(inode->i_mode)) {
1465 if (ia_valid & ATTR_SIZE)
1466 inode_dio_write_done(inode);
1467 mutex_unlock(&inode->i_mutex);
1468 down_write(&lli->lli_trunc_sem);
1469 mutex_lock(&inode->i_mutex);
1470 if (ia_valid & ATTR_SIZE)
1471 inode_dio_wait(inode);
1474 /* We need a steady stripe configuration for setattr to avoid
1476 lsm = ccc_inode_lsm_get(inode);
1478 /* NB: ATTR_SIZE will only be set after this point if the size
1479 * resides on the MDS, ie, this file has no objects. */
1481 attr->ia_valid &= ~ATTR_SIZE;
1482 /* can't call ll_setattr_ost() while holding a refcount of lsm */
1483 ccc_inode_lsm_put(inode, lsm);
1485 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1487 /* Open epoch for truncate. */
1488 if (exp_connect_som(ll_i2mdexp(inode)) &&
1489 (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1490 op_data->op_flags = MF_EPOCH_OPEN;
1492 rc = ll_md_setattr(dentry, op_data, &mod);
1496 ll_ioepoch_open(lli, op_data->op_ioepoch);
1497 if (!S_ISREG(inode->i_mode))
1500 if (ia_valid & ATTR_SIZE)
1501 attr->ia_valid |= ATTR_SIZE;
1502 if (ia_valid & (ATTR_SIZE |
1503 ATTR_ATIME | ATTR_ATIME_SET |
1504 ATTR_MTIME | ATTR_MTIME_SET))
1505 /* For truncate and utimes sending attributes to OSTs, setting
1506 * mtime/atime to the past will be performed under PW [0:EOF]
1507 * extent lock (new_size:EOF for truncate). It may seem
1508 * excessive to send mtime/atime updates to OSTs when not
1509 * setting times to past, but it is necessary due to possible
1510 * time de-synchronization between MDT inode and OST objects */
1511 rc = ll_setattr_ost(inode, attr);
1515 if (op_data->op_ioepoch) {
1516 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1520 ll_finish_md_op_data(op_data);
1522 if (!S_ISDIR(inode->i_mode))
1523 up_write(&lli->lli_trunc_sem);
1525 ll_stats_ops_tally(ll_i2sbi(inode), (ia_valid & ATTR_SIZE) ?
1526 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1531 int ll_setattr(struct dentry *de, struct iattr *attr)
1533 int mode = de->d_inode->i_mode;
1535 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1536 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1537 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1539 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1540 (ATTR_SIZE|ATTR_MODE)) &&
1541 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1542 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1543 !(attr->ia_mode & S_ISGID))))
1544 attr->ia_valid |= ATTR_FORCE;
1546 if ((mode & S_ISUID) &&
1547 !(attr->ia_mode & S_ISUID) &&
1548 !(attr->ia_valid & ATTR_KILL_SUID))
1549 attr->ia_valid |= ATTR_KILL_SUID;
1551 if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1552 !(attr->ia_mode & S_ISGID) &&
1553 !(attr->ia_valid & ATTR_KILL_SGID))
1554 attr->ia_valid |= ATTR_KILL_SGID;
1556 return ll_setattr_raw(de, attr);
1559 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1560 __u64 max_age, __u32 flags)
1562 struct ll_sb_info *sbi = ll_s2sbi(sb);
1563 struct obd_statfs obd_osfs;
1567 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1569 CERROR("md_statfs fails: rc = %d\n", rc);
1573 osfs->os_type = sb->s_magic;
1575 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1576 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1578 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1579 flags |= OBD_STATFS_NODELAY;
1581 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1583 CERROR("obd_statfs fails: rc = %d\n", rc);
1587 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1588 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1591 osfs->os_bsize = obd_osfs.os_bsize;
1592 osfs->os_blocks = obd_osfs.os_blocks;
1593 osfs->os_bfree = obd_osfs.os_bfree;
1594 osfs->os_bavail = obd_osfs.os_bavail;
1596 /* If we don't have as many objects free on the OST as inodes
1597 * on the MDS, we reduce the total number of inodes to
1598 * compensate, so that the "inodes in use" number is correct.
1600 if (obd_osfs.os_ffree < osfs->os_ffree) {
1601 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1603 osfs->os_ffree = obd_osfs.os_ffree;
1608 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1610 struct super_block *sb = de->d_sb;
1611 struct obd_statfs osfs;
1614 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1615 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1617 /* Some amount of caching on the client is allowed */
1618 rc = ll_statfs_internal(sb, &osfs,
1619 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1624 statfs_unpack(sfs, &osfs);
1626 /* We need to downshift for all 32-bit kernels, because we can't
1627 * tell if the kernel is being called via sys_statfs64() or not.
1628 * Stop before overflowing f_bsize - in which case it is better
1629 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1630 if (sizeof(long) < 8) {
1631 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1634 osfs.os_blocks >>= 1;
1635 osfs.os_bfree >>= 1;
1636 osfs.os_bavail >>= 1;
1640 sfs->f_blocks = osfs.os_blocks;
1641 sfs->f_bfree = osfs.os_bfree;
1642 sfs->f_bavail = osfs.os_bavail;
1647 void ll_inode_size_lock(struct inode *inode)
1649 struct ll_inode_info *lli;
1651 LASSERT(!S_ISDIR(inode->i_mode));
1653 lli = ll_i2info(inode);
1654 LASSERT(lli->lli_size_sem_owner != current);
1655 down(&lli->lli_size_sem);
1656 LASSERT(lli->lli_size_sem_owner == NULL);
1657 lli->lli_size_sem_owner = current;
1660 void ll_inode_size_unlock(struct inode *inode)
1662 struct ll_inode_info *lli;
1664 lli = ll_i2info(inode);
1665 LASSERT(lli->lli_size_sem_owner == current);
1666 lli->lli_size_sem_owner = NULL;
1667 up(&lli->lli_size_sem);
1670 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1672 struct ll_inode_info *lli = ll_i2info(inode);
1673 struct mdt_body *body = md->body;
1674 struct lov_stripe_md *lsm = md->lsm;
1675 struct ll_sb_info *sbi = ll_i2sbi(inode);
1677 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1679 LASSERT(S_ISREG(inode->i_mode));
1680 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1681 lsm, inode->i_ino, inode->i_generation, inode);
1682 /* cl_file_inode_init must go before lli_has_smd or a race
1683 * is possible where client thinks the file has stripes,
1684 * but lov raid0 is not setup yet and parallel e.g.
1685 * glimpse would try to use uninitialized lov */
1686 if (cl_file_inode_init(inode, md) == 0)
1687 lli->lli_has_smd = true;
1689 lli->lli_maxbytes = lsm->lsm_maxbytes;
1690 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1691 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1692 if (md->lsm != NULL)
1693 obd_free_memmd(ll_i2dtexp(inode), &md->lsm);
1696 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1697 if (body->valid & OBD_MD_FLRMTPERM)
1698 ll_update_remote_perm(inode, md->remote_perm);
1700 #ifdef CONFIG_FS_POSIX_ACL
1701 else if (body->valid & OBD_MD_FLACL) {
1702 spin_lock(&lli->lli_lock);
1703 if (lli->lli_posix_acl)
1704 posix_acl_release(lli->lli_posix_acl);
1705 lli->lli_posix_acl = md->posix_acl;
1706 spin_unlock(&lli->lli_lock);
1709 inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1710 inode->i_generation = cl_fid_build_gen(&body->fid1);
1712 if (body->valid & OBD_MD_FLATIME) {
1713 if (body->atime > LTIME_S(inode->i_atime))
1714 LTIME_S(inode->i_atime) = body->atime;
1715 lli->lli_lvb.lvb_atime = body->atime;
1717 if (body->valid & OBD_MD_FLMTIME) {
1718 if (body->mtime > LTIME_S(inode->i_mtime)) {
1719 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1720 "to "LPU64"\n", inode->i_ino,
1721 LTIME_S(inode->i_mtime), body->mtime);
1722 LTIME_S(inode->i_mtime) = body->mtime;
1724 lli->lli_lvb.lvb_mtime = body->mtime;
1726 if (body->valid & OBD_MD_FLCTIME) {
1727 if (body->ctime > LTIME_S(inode->i_ctime))
1728 LTIME_S(inode->i_ctime) = body->ctime;
1729 lli->lli_lvb.lvb_ctime = body->ctime;
1731 if (body->valid & OBD_MD_FLMODE)
1732 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1733 if (body->valid & OBD_MD_FLTYPE)
1734 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1735 LASSERT(inode->i_mode != 0);
1736 if (S_ISREG(inode->i_mode)) {
1737 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1739 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1741 if (body->valid & OBD_MD_FLUID)
1742 inode->i_uid = body->uid;
1743 if (body->valid & OBD_MD_FLGID)
1744 inode->i_gid = body->gid;
1745 if (body->valid & OBD_MD_FLFLAGS)
1746 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1747 if (body->valid & OBD_MD_FLNLINK)
1748 set_nlink(inode, body->nlink);
1749 if (body->valid & OBD_MD_FLRDEV)
1750 inode->i_rdev = old_decode_dev(body->rdev);
1752 if (body->valid & OBD_MD_FLID) {
1753 /* FID shouldn't be changed! */
1754 if (fid_is_sane(&lli->lli_fid)) {
1755 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1756 "Trying to change FID "DFID
1757 " to the "DFID", inode %lu/%u(%p)\n",
1758 PFID(&lli->lli_fid), PFID(&body->fid1),
1759 inode->i_ino, inode->i_generation, inode);
1761 lli->lli_fid = body->fid1;
1764 LASSERT(fid_seq(&lli->lli_fid) != 0);
1766 if (body->valid & OBD_MD_FLSIZE) {
1767 if (exp_connect_som(ll_i2mdexp(inode)) &&
1768 S_ISREG(inode->i_mode) && lli->lli_has_smd) {
1769 struct lustre_handle lockh;
1772 /* As it is possible a blocking ast has been processed
1773 * by this time, we need to check there is an UPDATE
1774 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1776 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1777 &lockh, LDLM_FL_CBPENDING);
1779 if (lli->lli_flags & (LLIF_DONE_WRITING |
1780 LLIF_EPOCH_PENDING |
1782 CERROR("ino %lu flags %u still has "
1783 "size authority! do not trust "
1784 "the size got from MDS\n",
1785 inode->i_ino, lli->lli_flags);
1787 /* Use old size assignment to avoid
1788 * deadlock bz14138 & bz14326 */
1789 i_size_write(inode, body->size);
1790 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1792 ldlm_lock_decref(&lockh, mode);
1795 /* Use old size assignment to avoid
1796 * deadlock bz14138 & bz14326 */
1797 i_size_write(inode, body->size);
1799 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1800 inode->i_ino, (unsigned long long)body->size);
1803 if (body->valid & OBD_MD_FLBLOCKS)
1804 inode->i_blocks = body->blocks;
1807 if (body->valid & OBD_MD_FLMDSCAPA) {
1808 LASSERT(md->mds_capa);
1809 ll_add_capa(inode, md->mds_capa);
1811 if (body->valid & OBD_MD_FLOSSCAPA) {
1812 LASSERT(md->oss_capa);
1813 ll_add_capa(inode, md->oss_capa);
1817 void ll_read_inode2(struct inode *inode, void *opaque)
1819 struct lustre_md *md = opaque;
1820 struct ll_inode_info *lli = ll_i2info(inode);
1823 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1824 PFID(&lli->lli_fid), inode);
1826 LASSERT(!lli->lli_has_smd);
1828 /* Core attributes from the MDS first. This is a new inode, and
1829 * the VFS doesn't zero times in the core inode so we have to do
1830 * it ourselves. They will be overwritten by either MDS or OST
1831 * attributes - we just need to make sure they aren't newer. */
1832 LTIME_S(inode->i_mtime) = 0;
1833 LTIME_S(inode->i_atime) = 0;
1834 LTIME_S(inode->i_ctime) = 0;
1836 ll_update_inode(inode, md);
1838 /* OIDEBUG(inode); */
1840 /* initializing backing dev info. */
1841 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1844 if (S_ISREG(inode->i_mode)) {
1845 struct ll_sb_info *sbi = ll_i2sbi(inode);
1846 inode->i_op = &ll_file_inode_operations;
1847 inode->i_fop = sbi->ll_fop;
1848 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1850 } else if (S_ISDIR(inode->i_mode)) {
1851 inode->i_op = &ll_dir_inode_operations;
1852 inode->i_fop = &ll_dir_operations;
1854 } else if (S_ISLNK(inode->i_mode)) {
1855 inode->i_op = &ll_fast_symlink_inode_operations;
1858 inode->i_op = &ll_special_inode_operations;
1860 init_special_inode(inode, inode->i_mode,
1867 void ll_delete_inode(struct inode *inode)
1869 struct cl_inode_info *lli = cl_i2info(inode);
1870 struct ll_sb_info *sbi = ll_i2sbi(inode);
1874 rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1876 CERROR("fid_delete() failed, rc %d\n", rc);
1878 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1879 /* discard all dirty pages before truncating them, required by
1880 * osc_extent implementation at LU-1030. */
1881 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_DISCARD);
1883 truncate_inode_pages(&inode->i_data, 0);
1885 /* Workaround for LU-118 */
1886 if (inode->i_data.nrpages) {
1887 TREE_READ_LOCK_IRQ(&inode->i_data);
1888 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1889 LASSERTF(inode->i_data.nrpages == 0,
1890 "inode=%lu/%u(%p) nrpages=%lu, see "
1891 "http://jira.whamcloud.com/browse/LU-118\n",
1892 inode->i_ino, inode->i_generation, inode,
1893 inode->i_data.nrpages);
1895 /* Workaround end */
1897 #ifdef HAVE_SBOPS_EVICT_INODE
1898 ll_clear_inode(inode);
1899 end_writeback(inode);
1907 int ll_iocontrol(struct inode *inode, struct file *file,
1908 unsigned int cmd, unsigned long arg)
1910 struct ll_sb_info *sbi = ll_i2sbi(inode);
1911 struct ptlrpc_request *req = NULL;
1916 case FSFILT_IOC_GETFLAGS: {
1917 struct mdt_body *body;
1918 struct md_op_data *op_data;
1920 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1921 0, 0, LUSTRE_OPC_ANY,
1923 if (IS_ERR(op_data))
1924 RETURN(PTR_ERR(op_data));
1926 op_data->op_valid = OBD_MD_FLFLAGS;
1927 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1928 ll_finish_md_op_data(op_data);
1930 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1934 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1936 flags = body->flags;
1938 ptlrpc_req_finished(req);
1940 RETURN(put_user(flags, (int *)arg));
1942 case FSFILT_IOC_SETFLAGS: {
1943 struct lov_stripe_md *lsm;
1944 struct obd_info oinfo = { { { 0 } } };
1945 struct md_op_data *op_data;
1947 if (get_user(flags, (int *)arg))
1950 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1951 LUSTRE_OPC_ANY, NULL);
1952 if (IS_ERR(op_data))
1953 RETURN(PTR_ERR(op_data));
1955 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1956 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1957 rc = md_setattr(sbi->ll_md_exp, op_data,
1958 NULL, 0, NULL, 0, &req, NULL);
1959 ll_finish_md_op_data(op_data);
1960 ptlrpc_req_finished(req);
1964 inode->i_flags = ll_ext_to_inode_flags(flags);
1966 lsm = ccc_inode_lsm_get(inode);
1970 OBDO_ALLOC(oinfo.oi_oa);
1972 ccc_inode_lsm_put(inode, lsm);
1976 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1977 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1978 oinfo.oi_oa->o_flags = flags;
1979 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1981 oinfo.oi_capa = ll_mdscapa_get(inode);
1982 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1983 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1984 capa_put(oinfo.oi_capa);
1985 OBDO_FREE(oinfo.oi_oa);
1986 ccc_inode_lsm_put(inode, lsm);
1988 if (rc && rc != -EPERM && rc != -EACCES)
1989 CERROR("osc_setattr_async fails: rc = %d\n", rc);
2000 int ll_flush_ctx(struct inode *inode)
2002 struct ll_sb_info *sbi = ll_i2sbi(inode);
2004 CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
2006 obd_set_info_async(NULL, sbi->ll_md_exp,
2007 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2009 obd_set_info_async(NULL, sbi->ll_dt_exp,
2010 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2015 /* umount -f client means force down, don't save state */
2016 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2017 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2019 struct super_block *sb = vfsmnt->mnt_sb;
2021 void ll_umount_begin(struct super_block *sb)
2024 struct ll_sb_info *sbi = ll_s2sbi(sb);
2025 struct obd_device *obd;
2026 struct obd_ioctl_data *ioc_data;
2029 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2030 if (!(flags & MNT_FORCE)) {
2036 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2037 sb->s_count, atomic_read(&sb->s_active));
2039 obd = class_exp2obd(sbi->ll_md_exp);
2041 CERROR("Invalid MDC connection handle "LPX64"\n",
2042 sbi->ll_md_exp->exp_handle.h_cookie);
2048 obd = class_exp2obd(sbi->ll_dt_exp);
2050 CERROR("Invalid LOV connection handle "LPX64"\n",
2051 sbi->ll_dt_exp->exp_handle.h_cookie);
2057 OBD_ALLOC_PTR(ioc_data);
2059 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2060 sizeof ioc_data, ioc_data, NULL);
2062 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2063 sizeof ioc_data, ioc_data, NULL);
2065 OBD_FREE_PTR(ioc_data);
2069 /* Really, we'd like to wait until there are no requests outstanding,
2070 * and then continue. For now, we just invalidate the requests,
2071 * schedule() and sleep one second if needed, and hope.
2074 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2075 if (atomic_read(&vfsmnt->mnt_count) > 2) {
2076 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2077 cfs_time_seconds(1));
2078 if (atomic_read(&vfsmnt->mnt_count) > 2)
2079 LCONSOLE_WARN("Mount still busy with %d refs! You "
2080 "may try to umount it a bit later\n",
2081 atomic_read(&vfsmnt->mnt_count));
2088 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2090 struct ll_sb_info *sbi = ll_s2sbi(sb);
2091 char *profilenm = get_profile_name(sb);
2095 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2096 read_only = *flags & MS_RDONLY;
2097 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2098 sizeof(KEY_READ_ONLY),
2099 KEY_READ_ONLY, sizeof(read_only),
2102 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2103 profilenm, read_only ?
2104 "read-only" : "read-write", err);
2109 sb->s_flags |= MS_RDONLY;
2111 sb->s_flags &= ~MS_RDONLY;
2113 if (sbi->ll_flags & LL_SBI_VERBOSE)
2114 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2115 read_only ? "read-only" : "read-write");
2120 int ll_prep_inode(struct inode **inode,
2121 struct ptlrpc_request *req,
2122 struct super_block *sb)
2124 struct ll_sb_info *sbi = NULL;
2125 struct lustre_md md;
2130 LASSERT(*inode || sb);
2131 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2132 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2133 sbi->ll_md_exp, &md);
2138 ll_update_inode(*inode, &md);
2140 LASSERT(sb != NULL);
2143 * At this point server returns to client's same fid as client
2144 * generated for creating. So using ->fid1 is okay here.
2146 LASSERT(fid_is_sane(&md.body->fid1));
2148 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2149 if (*inode == NULL || IS_ERR(*inode)) {
2151 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2152 #ifdef CONFIG_FS_POSIX_ACL
2154 posix_acl_release(md.posix_acl);
2155 md.posix_acl = NULL;
2158 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2160 CERROR("new_inode -fatal: rc %d\n", rc);
2165 /* sanity check for LAYOUT lock. */
2166 ibits = MDS_INODELOCK_LAYOUT;
2167 if (S_ISREG(md.body->mode) && sbi->ll_flags & LL_SBI_LAYOUT_LOCK &&
2168 md.lsm != NULL && !ll_have_md_lock(*inode, &ibits, LCK_MINMODE)) {
2169 CERROR("%s: inode "DFID" (%p) layout lock not granted.\n",
2170 ll_get_fsname(sb, NULL, 0),
2171 PFID(ll_inode2fid(*inode)), *inode);
2175 md_free_lustre_md(sbi->ll_md_exp, &md);
2179 int ll_obd_statfs(struct inode *inode, void *arg)
2181 struct ll_sb_info *sbi = NULL;
2182 struct obd_export *exp;
2184 struct obd_ioctl_data *data = NULL;
2189 if (!inode || !(sbi = ll_i2sbi(inode)))
2190 GOTO(out_statfs, rc = -EINVAL);
2192 rc = obd_ioctl_getdata(&buf, &len, arg);
2194 GOTO(out_statfs, rc);
2197 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2198 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2199 GOTO(out_statfs, rc = -EINVAL);
2201 if (data->ioc_inllen1 != sizeof(__u32) ||
2202 data->ioc_inllen2 != sizeof(__u32) ||
2203 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2204 data->ioc_plen2 != sizeof(struct obd_uuid))
2205 GOTO(out_statfs, rc = -EINVAL);
2207 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2208 if (type & LL_STATFS_LMV)
2209 exp = sbi->ll_md_exp;
2210 else if (type & LL_STATFS_LOV)
2211 exp = sbi->ll_dt_exp;
2213 GOTO(out_statfs, rc = -ENODEV);
2215 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2216 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2218 GOTO(out_statfs, rc);
2221 obd_ioctl_freedata(buf, len);
2225 int ll_process_config(struct lustre_cfg *lcfg)
2229 struct lprocfs_static_vars lvars;
2233 lprocfs_llite_init_vars(&lvars);
2235 /* The instance name contains the sb: lustre-client-aacfe000 */
2236 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2237 if (!ptr || !*(++ptr))
2239 if (sscanf(ptr, "%lx", &x) != 1)
2242 /* This better be a real Lustre superblock! */
2243 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2245 /* Note we have not called client_common_fill_super yet, so
2246 proc fns must be able to handle that! */
2247 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2254 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2255 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2256 struct inode *i1, struct inode *i2,
2257 const char *name, int namelen,
2258 int mode, __u32 opc, void *data)
2260 LASSERT(i1 != NULL);
2262 if (namelen > ll_i2sbi(i1)->ll_namelen)
2263 return ERR_PTR(-ENAMETOOLONG);
2265 if (op_data == NULL)
2266 OBD_ALLOC_PTR(op_data);
2268 if (op_data == NULL)
2269 return ERR_PTR(-ENOMEM);
2271 ll_i2gids(op_data->op_suppgids, i1, i2);
2272 op_data->op_fid1 = *ll_inode2fid(i1);
2273 op_data->op_capa1 = ll_mdscapa_get(i1);
2276 op_data->op_fid2 = *ll_inode2fid(i2);
2277 op_data->op_capa2 = ll_mdscapa_get(i2);
2279 fid_zero(&op_data->op_fid2);
2280 op_data->op_capa2 = NULL;
2283 op_data->op_name = name;
2284 op_data->op_namelen = namelen;
2285 op_data->op_mode = mode;
2286 op_data->op_mod_time = cfs_time_current_sec();
2287 op_data->op_fsuid = cfs_curproc_fsuid();
2288 op_data->op_fsgid = cfs_curproc_fsgid();
2289 op_data->op_cap = cfs_curproc_cap_pack();
2290 op_data->op_bias = MDS_CHECK_SPLIT;
2291 op_data->op_opc = opc;
2292 op_data->op_mds = 0;
2293 op_data->op_data = data;
2295 /* If the file is being opened after mknod() (normally due to NFS)
2296 * try to use the default stripe data from parent directory for
2297 * allocating OST objects. Try to pass the parent FID to MDS. */
2298 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2299 !ll_i2info(i2)->lli_has_smd) {
2300 struct ll_inode_info *lli = ll_i2info(i2);
2302 spin_lock(&lli->lli_lock);
2303 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2304 op_data->op_fid1 = lli->lli_pfid;
2305 spin_unlock(&lli->lli_lock);
2306 /** We ignore parent's capability temporary. */
2312 void ll_finish_md_op_data(struct md_op_data *op_data)
2314 capa_put(op_data->op_capa1);
2315 capa_put(op_data->op_capa2);
2316 OBD_FREE_PTR(op_data);
2319 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2321 struct ll_sb_info *sbi;
2323 LASSERT((seq != NULL) && (vfs != NULL));
2324 sbi = ll_s2sbi(vfs->mnt_sb);
2326 if (sbi->ll_flags & LL_SBI_NOLCK)
2327 seq_puts(seq, ",nolock");
2329 if (sbi->ll_flags & LL_SBI_FLOCK)
2330 seq_puts(seq, ",flock");
2332 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2333 seq_puts(seq, ",localflock");
2335 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2336 seq_puts(seq, ",user_xattr");
2338 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2339 seq_puts(seq, ",lazystatfs");
2341 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2342 seq_puts(seq, ",user_fid2path");
2348 * Get obd name by cmd, and copy out to user space
2350 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2352 struct ll_sb_info *sbi = ll_i2sbi(inode);
2353 struct obd_device *obd;
2356 if (cmd == OBD_IOC_GETDTNAME)
2357 obd = class_exp2obd(sbi->ll_dt_exp);
2358 else if (cmd == OBD_IOC_GETMDNAME)
2359 obd = class_exp2obd(sbi->ll_md_exp);
2366 if (cfs_copy_to_user((void *)arg, obd->obd_name,
2367 strlen(obd->obd_name) + 1))
2374 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2375 * fsname will be returned in this buffer; otherwise, a static buffer will be
2376 * used to store the fsname and returned to caller.
2378 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2380 static char fsname_static[MTI_NAME_MAXLEN];
2381 struct lustre_sb_info *lsi = s2lsi(sb);
2386 /* this means the caller wants to use static buffer
2387 * and it doesn't care about race. Usually this is
2388 * in error reporting path */
2389 buf = fsname_static;
2390 buflen = sizeof(fsname_static);
2393 len = strlen(lsi->lsi_lmd->lmd_profile);
2394 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2395 if (ptr && (strcmp(ptr, "-client") == 0))
2398 if (unlikely(len >= buflen))
2400 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2406 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2410 #ifdef HAVE_FS_STRUCT_USE_PATH
2414 p.mnt = current->fs->root.mnt;
2416 path = d_path(&p, buf, bufsize);
2419 path = d_path(dentry, current->fs->rootmnt, buf, bufsize);
2425 void ll_dirty_page_discard_warn(cfs_page_t *page, int ioret)
2427 char *buf, *path = NULL;
2428 struct dentry *dentry = NULL;
2429 struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2431 buf = (char *)__get_free_page(GFP_KERNEL);
2433 dentry = d_find_alias(page->mapping->host);
2435 path = ll_d_path(dentry, buf, PAGE_SIZE);
2438 CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2439 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2440 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2441 PFID(&obj->cob_header.coh_lu.loh_fid),
2442 (path && !IS_ERR(path)) ? path : "", ioret);
2448 free_page((unsigned long)buf);