1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/llite/llite_lib.c
38 * Lustre Light Super operations
41 #define DEBUG_SUBSYSTEM S_LLITE
43 #include <linux/module.h>
44 #include <linux/types.h>
45 #include <linux/version.h>
48 #include <lustre_lite.h>
49 #include <lustre_ha.h>
50 #include <lustre_dlm.h>
51 #include <lprocfs_status.h>
52 #include <lustre_disk.h>
53 #include <lustre_param.h>
54 #include <lustre_log.h>
55 #include <cl_object.h>
56 #include <obd_cksum.h>
57 #include "llite_internal.h"
59 cfs_mem_cache_t *ll_file_data_slab;
61 CFS_LIST_HEAD(ll_super_blocks);
62 cfs_spinlock_t ll_sb_lock = CFS_SPIN_LOCK_UNLOCKED;
64 extern struct address_space_operations ll_aops;
65 extern struct address_space_operations ll_dir_aops;
68 #define log2(n) cfs_ffz(~(n))
71 static struct ll_sb_info *ll_init_sbi(void)
73 struct ll_sb_info *sbi = NULL;
80 OBD_ALLOC(sbi, sizeof(*sbi));
84 cfs_spin_lock_init(&sbi->ll_lock);
85 cfs_init_mutex(&sbi->ll_lco.lco_lock);
86 cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
87 cfs_spin_lock_init(&sbi->ll_process_lock);
88 sbi->ll_rw_stats_on = 0;
91 pages = si.totalram - si.totalhigh;
92 if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
93 #ifdef HAVE_BGL_SUPPORT
94 sbi->ll_async_page_max = pages / 4;
96 sbi->ll_async_page_max = pages / 2;
99 sbi->ll_async_page_max = (pages / 4) * 3;
102 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
103 SBI_DEFAULT_READAHEAD_MAX);
104 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
105 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
106 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
107 CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
108 CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
110 ll_generate_random_uuid(uuid);
111 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
112 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
114 cfs_spin_lock(&ll_sb_lock);
115 cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
116 cfs_spin_unlock(&ll_sb_lock);
118 #ifdef ENABLE_LLITE_CHECKSUM
119 sbi->ll_flags |= LL_SBI_CHECKSUM;
122 #ifdef HAVE_LRU_RESIZE_SUPPORT
123 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
126 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
127 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
129 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
133 /* metadata statahead is enabled by default */
134 sbi->ll_sa_max = LL_SA_RPC_DEF;
135 atomic_set(&sbi->ll_sa_total, 0);
136 atomic_set(&sbi->ll_sa_wrong, 0);
141 void ll_free_sbi(struct super_block *sb)
143 struct ll_sb_info *sbi = ll_s2sbi(sb);
147 cfs_spin_lock(&ll_sb_lock);
148 cfs_list_del(&sbi->ll_list);
149 cfs_spin_unlock(&ll_sb_lock);
150 OBD_FREE(sbi, sizeof(*sbi));
155 static struct dentry_operations ll_d_root_ops = {
156 .d_compare = ll_dcompare,
157 .d_revalidate = ll_revalidate_nd,
160 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
162 struct inode *root = 0;
163 struct ll_sb_info *sbi = ll_s2sbi(sb);
164 struct obd_device *obd;
165 struct obd_capa *oc = NULL;
166 struct obd_statfs osfs;
167 struct ptlrpc_request *request = NULL;
168 struct obd_connect_data *data = NULL;
169 struct obd_uuid *uuid;
170 struct md_op_data *op_data;
171 struct lustre_md lmd;
173 int size, err, checksum;
176 obd = class_name2obd(md);
178 CERROR("MD %s: not setup or attached\n", md);
186 if (proc_lustre_fs_root) {
187 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
190 CERROR("could not register mount in /proc/fs/lustre\n");
193 /* indicate the features supported by this client */
194 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
195 OBD_CONNECT_JOIN | OBD_CONNECT_ATTRFID |
196 OBD_CONNECT_VERSION | OBD_CONNECT_MDS_CAPA |
197 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
198 OBD_CONNECT_FID | OBD_CONNECT_AT |
199 OBD_CONNECT_LOV_V3 | OBD_CONNECT_RMT_CLIENT |
200 OBD_CONNECT_VBR | OBD_CONNECT_FULL20;
202 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
203 data->ocd_connect_flags |= OBD_CONNECT_SOM;
205 #ifdef HAVE_LRU_RESIZE_SUPPORT
206 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
207 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
209 #ifdef CONFIG_FS_POSIX_ACL
210 data->ocd_connect_flags |= OBD_CONNECT_ACL;
212 data->ocd_ibits_known = MDS_INODELOCK_FULL;
213 data->ocd_version = LUSTRE_VERSION_CODE;
215 if (sb->s_flags & MS_RDONLY)
216 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
217 if (sbi->ll_flags & LL_SBI_USER_XATTR)
218 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
220 #ifdef HAVE_MS_FLOCK_LOCK
221 /* force vfs to use lustre handler for flock() calls - bug 10743 */
222 sb->s_flags |= MS_FLOCK_LOCK;
225 if (sbi->ll_flags & LL_SBI_FLOCK)
226 sbi->ll_fop = &ll_file_operations_flock;
227 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
228 sbi->ll_fop = &ll_file_operations;
230 sbi->ll_fop = &ll_file_operations_noflock;
233 data->ocd_connect_flags |= OBD_CONNECT_REAL;
234 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
235 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
237 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
239 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
240 "recovery, of which this client is not a "
241 "part. Please wait for recovery to complete,"
242 " abort, or time out.\n", md);
245 CERROR("cannot connect to %s: rc = %d\n", md, err);
249 err = obd_fid_init(sbi->ll_md_exp);
251 CERROR("Can't init metadata layer FID infrastructure, "
256 err = obd_statfs(obd, &osfs,
257 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
259 GOTO(out_md_fid, err);
261 size = sizeof(*data);
262 err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
263 KEY_CONN_DATA, &size, data, NULL);
265 CERROR("Get connect data failed: %d \n", err);
269 LASSERT(osfs.os_bsize);
270 sb->s_blocksize = osfs.os_bsize;
271 sb->s_blocksize_bits = log2(osfs.os_bsize);
272 sb->s_magic = LL_SUPER_MAGIC;
274 /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
275 * retval = in_file->f_op->sendfile(...);
277 * retval = -EOVERFLOW;
279 * it will check if *ppos is greater than max. However, max equals to
280 * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
281 * has been defined as a signed long long integer in linux kernel. */
282 #if BITS_PER_LONG == 64
283 sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
285 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
287 sbi->ll_namelen = osfs.os_namelen;
288 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
290 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
291 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
292 LCONSOLE_INFO("Disabling user_xattr feature because "
293 "it is not supported on the server\n");
294 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
297 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
299 sb->s_flags |= MS_POSIXACL;
301 sbi->ll_flags |= LL_SBI_ACL;
303 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
305 sb->s_flags &= ~MS_POSIXACL;
307 sbi->ll_flags &= ~LL_SBI_ACL;
310 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
311 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
312 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
313 LCONSOLE_INFO("client is set as remote by default.\n");
316 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
317 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
318 LCONSOLE_INFO("client claims to be remote, but server "
319 "rejected, forced to be local.\n");
323 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
324 LCONSOLE_INFO("client enabled MDS capability!\n");
325 sbi->ll_flags |= LL_SBI_MDS_CAPA;
328 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
329 LCONSOLE_INFO("client enabled OSS capability!\n");
330 sbi->ll_flags |= LL_SBI_OSS_CAPA;
333 obd = class_name2obd(dt);
335 CERROR("DT %s: not setup or attached\n", dt);
336 GOTO(out_md_fid, err = -ENODEV);
339 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
340 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
341 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
342 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
343 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
344 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
347 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
348 data->ocd_connect_flags |= OBD_CONNECT_SOM;
350 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
351 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
352 * disabled by default, because it can still be enabled on the
353 * fly via /proc. As a consequence, we still need to come to an
354 * agreement on the supported algorithms at connect time */
355 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
357 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
358 data->ocd_cksum_types = OBD_CKSUM_ADLER;
360 /* send the list of supported checksum types */
361 data->ocd_cksum_types = OBD_CKSUM_ALL;
364 #ifdef HAVE_LRU_RESIZE_SUPPORT
365 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
367 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
368 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
370 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
371 "ocd_grant: %d\n", data->ocd_connect_flags,
372 data->ocd_version, data->ocd_grant);
374 obd->obd_upcall.onu_owner = &sbi->ll_lco;
375 obd->obd_upcall.onu_upcall = cl_ocd_update;
376 data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
378 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
380 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
381 "recovery, of which this client is not a "
382 "part. Please wait for recovery to "
383 "complete, abort, or time out.\n", dt);
384 GOTO(out_md_fid, err);
386 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
387 GOTO(out_md_fid, err);
390 err = obd_fid_init(sbi->ll_dt_exp);
392 CERROR("Can't init data layer FID infrastructure, "
397 cfs_mutex_down(&sbi->ll_lco.lco_lock);
398 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
399 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
400 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
401 cfs_mutex_up(&sbi->ll_lco.lco_lock);
403 fid_zero(&sbi->ll_root_fid);
404 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
406 CERROR("cannot mds_connect: rc = %d\n", err);
407 GOTO(out_lock_cn_cb, err);
409 if (!fid_is_sane(&sbi->ll_root_fid)) {
410 CERROR("Invalid root fid during mount\n");
411 GOTO(out_lock_cn_cb, err = -EINVAL);
413 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
415 sb->s_op = &lustre_super_operations;
416 #if THREAD_SIZE >= 8192
417 sb->s_export_op = &lustre_export_operations;
421 * XXX: move this to after cbd setup? */
422 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
423 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
424 valid |= OBD_MD_FLRMTPERM;
425 else if (sbi->ll_flags & LL_SBI_ACL)
426 valid |= OBD_MD_FLACL;
428 OBD_ALLOC_PTR(op_data);
430 GOTO(out_lock_cn_cb, err = -ENOMEM);
432 op_data->op_fid1 = sbi->ll_root_fid;
433 op_data->op_mode = 0;
434 op_data->op_capa1 = oc;
435 op_data->op_valid = valid;
437 err = md_getattr(sbi->ll_md_exp, op_data, &request);
440 OBD_FREE_PTR(op_data);
442 CERROR("md_getattr failed for root: rc = %d\n", err);
443 GOTO(out_lock_cn_cb, err);
445 memset(&lmd, 0, sizeof(lmd));
446 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
447 sbi->ll_md_exp, &lmd);
449 CERROR("failed to understand root inode md: rc = %d\n", err);
450 ptlrpc_req_finished (request);
451 GOTO(out_lock_cn_cb, err);
454 LASSERT(fid_is_sane(&sbi->ll_root_fid));
455 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid), &lmd);
456 md_free_lustre_md(sbi->ll_md_exp, &lmd);
457 ptlrpc_req_finished(request);
459 if (root == NULL || IS_ERR(root)) {
461 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
462 #ifdef CONFIG_FS_POSIX_ACL
464 posix_acl_release(lmd.posix_acl);
465 lmd.posix_acl = NULL;
468 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
470 CERROR("lustre_lite: bad iget4 for root\n");
474 err = ll_close_thread_start(&sbi->ll_lcq);
476 CERROR("cannot start close thread: rc %d\n", err);
480 #ifdef CONFIG_FS_POSIX_ACL
481 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
482 rct_init(&sbi->ll_rct);
483 et_init(&sbi->ll_et);
487 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
488 err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
489 KEY_CHECKSUM, sizeof(checksum), &checksum,
493 sb->s_root = d_alloc_root(root);
495 OBD_FREE(data, sizeof(*data));
497 sb->s_root->d_op = &ll_d_root_ops;
499 sbi->ll_sdev_orig = sb->s_dev;
501 /* We set sb->s_dev equal on all lustre clients in order to support
502 * NFS export clustering. NFSD requires that the FSID be the same
504 /* s_dev is also used in lt_compare() to compare two fs, but that is
505 * only a node-local comparison. */
506 uuid = obd_get_uuid(sbi->ll_md_exp);
508 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
515 obd_fid_fini(sbi->ll_dt_exp);
517 obd_disconnect(sbi->ll_dt_exp);
518 sbi->ll_dt_exp = NULL;
520 obd_fid_fini(sbi->ll_md_exp);
522 obd_disconnect(sbi->ll_md_exp);
523 sbi->ll_md_exp = NULL;
527 lprocfs_unregister_mountpoint(sbi);
531 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
535 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
537 rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
538 KEY_MAX_EASIZE, &size, lmmsize, NULL);
540 CERROR("Get max mdsize error rc %d \n", rc);
545 void ll_dump_inode(struct inode *inode)
547 struct list_head *tmp;
548 int dentry_count = 0;
550 LASSERT(inode != NULL);
552 list_for_each(tmp, &inode->i_dentry)
555 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
556 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
557 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
560 void lustre_dump_dentry(struct dentry *dentry, int recur)
562 struct list_head *tmp;
565 LASSERT(dentry != NULL);
567 list_for_each(tmp, &dentry->d_subdirs)
570 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
571 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
572 dentry->d_name.len, dentry->d_name.name,
573 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
574 dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
575 dentry->d_flags, dentry->d_fsdata, subdirs);
576 if (dentry->d_inode != NULL)
577 ll_dump_inode(dentry->d_inode);
582 list_for_each(tmp, &dentry->d_subdirs) {
583 struct dentry *d = list_entry(tmp, struct dentry, d_child);
584 lustre_dump_dentry(d, recur - 1);
588 void client_common_put_super(struct super_block *sb)
590 struct ll_sb_info *sbi = ll_s2sbi(sb);
593 #ifdef CONFIG_FS_POSIX_ACL
594 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
595 et_fini(&sbi->ll_et);
596 rct_fini(&sbi->ll_rct);
600 obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
602 ll_close_thread_shutdown(sbi->ll_lcq);
606 cfs_list_del(&sbi->ll_conn_chain);
608 obd_fid_fini(sbi->ll_dt_exp);
609 obd_disconnect(sbi->ll_dt_exp);
610 sbi->ll_dt_exp = NULL;
612 lprocfs_unregister_mountpoint(sbi);
614 obd_fid_fini(sbi->ll_md_exp);
615 obd_disconnect(sbi->ll_md_exp);
616 sbi->ll_md_exp = NULL;
621 void ll_kill_super(struct super_block *sb)
623 struct ll_sb_info *sbi;
628 if (!(sb->s_flags & MS_ACTIVE))
632 /* we need restore s_dev from changed for clustred NFS before put_super
633 * because new kernels have cached s_dev and change sb->s_dev in
634 * put_super not affected real removing devices */
636 sb->s_dev = sbi->ll_sdev_orig;
640 char *ll_read_opt(const char *opt, char *data)
646 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
647 if (strncmp(opt, data, strlen(opt)))
649 if ((value = strchr(data, '=')) == NULL)
653 OBD_ALLOC(retval, strlen(value) + 1);
655 CERROR("out of memory!\n");
659 memcpy(retval, value, strlen(value)+1);
660 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
664 static inline int ll_set_opt(const char *opt, char *data, int fl)
666 if (strncmp(opt, data, strlen(opt)) != 0)
672 /* non-client-specific mount options are parsed in lmd_parse */
673 static int ll_options(char *options, int *flags)
676 char *s1 = options, *s2;
682 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
685 CDEBUG(D_SUPER, "next opt=%s\n", s1);
686 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
691 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
696 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
701 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
706 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
711 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
716 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,5,50,0)
717 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
719 /* Ignore deprecated mount option. The client will
720 * always try to mount with ACL support, whether this
721 * is used depends on whether server supports it. */
722 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
723 "mount option 'acl'.\n");
726 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
728 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
729 "mount option 'noacl'.\n");
733 #warning "{no}acl options have been deprecated since 1.8, please remove them"
735 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
741 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
746 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
751 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
756 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
761 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
766 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
771 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
776 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
782 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
788 s2 = strchr(s1, ',');
796 void ll_lli_init(struct ll_inode_info *lli)
798 lli->lli_inode_magic = LLI_INODE_MAGIC;
799 cfs_sema_init(&lli->lli_size_sem, 1);
800 cfs_sema_init(&lli->lli_write_sem, 1);
801 cfs_sema_init(&lli->lli_trunc_sem, 1);
803 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
804 cfs_spin_lock_init(&lli->lli_lock);
805 CFS_INIT_LIST_HEAD(&lli->lli_close_list);
806 lli->lli_inode_magic = LLI_INODE_MAGIC;
807 cfs_sema_init(&lli->lli_och_sem, 1);
808 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
809 lli->lli_mds_exec_och = NULL;
810 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
811 lli->lli_open_fd_exec_count = 0;
812 CFS_INIT_LIST_HEAD(&lli->lli_dead_list);
813 lli->lli_remote_perms = NULL;
814 lli->lli_rmtperm_utime = 0;
815 cfs_sema_init(&lli->lli_rmtperm_sem, 1);
816 CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
817 cfs_spin_lock_init(&lli->lli_sa_lock);
818 CFS_INIT_LIST_HEAD(&lli->lli_sa_dentry);
821 int ll_fill_super(struct super_block *sb)
823 struct lustre_profile *lprof;
824 struct lustre_sb_info *lsi = s2lsi(sb);
825 struct ll_sb_info *sbi;
826 char *dt = NULL, *md = NULL;
827 char *profilenm = get_profile_name(sb);
828 struct config_llog_instance cfg = {0, };
829 char ll_instance[sizeof(sb) * 2 + 1];
833 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
837 /* client additional sb info */
838 lsi->lsi_llsbi = sbi = ll_init_sbi();
840 cfs_module_put(THIS_MODULE);
844 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
848 /* Generate a string unique to this super, in case some joker tries
849 to mount the same fs at two mount points.
850 Use the address of the super itself.*/
851 sprintf(ll_instance, "%p", sb);
852 cfg.cfg_instance = ll_instance;
853 cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
855 /* set up client obds */
856 err = lustre_process_log(sb, profilenm, &cfg);
858 CERROR("Unable to process log: %d\n", err);
862 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
863 lprof = class_get_profile(profilenm);
865 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
866 " read from the MGS. Does that filesystem "
867 "exist?\n", profilenm);
868 GOTO(out_free, err = -EINVAL);
870 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
871 lprof->lp_md, lprof->lp_dt);
873 OBD_ALLOC(dt, strlen(lprof->lp_dt) +
874 strlen(ll_instance) + 2);
876 GOTO(out_free, err = -ENOMEM);
877 sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
879 OBD_ALLOC(md, strlen(lprof->lp_md) +
880 strlen(ll_instance) + 2);
882 GOTO(out_free, err = -ENOMEM);
883 sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
885 /* connections, registrations, sb setup */
886 err = client_common_fill_super(sb, md, dt);
890 OBD_FREE(md, strlen(md) + 1);
892 OBD_FREE(dt, strlen(dt) + 1);
896 LCONSOLE_WARN("Client %s has started\n", profilenm);
899 } /* ll_fill_super */
902 void lu_context_keys_dump(void);
904 void ll_put_super(struct super_block *sb)
906 struct config_llog_instance cfg;
907 char ll_instance[sizeof(sb) * 2 + 1];
908 struct obd_device *obd;
909 struct lustre_sb_info *lsi = s2lsi(sb);
910 struct ll_sb_info *sbi = ll_s2sbi(sb);
911 char *profilenm = get_profile_name(sb);
915 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
917 ll_print_capa_stat(sbi);
919 sprintf(ll_instance, "%p", sb);
920 cfg.cfg_instance = ll_instance;
921 lustre_end_log(sb, NULL, &cfg);
923 if (sbi->ll_md_exp) {
924 obd = class_exp2obd(sbi->ll_md_exp);
926 force = obd->obd_force;
929 /* We need to set force before the lov_disconnect in
930 lustre_common_put_super, since l_d cleans up osc's as well. */
933 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
935 obd->obd_force = force;
940 /* Only if client_common_fill_super succeeded */
941 client_common_put_super(sb);
945 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
946 class_manual_cleanup(obd);
950 class_del_profile(profilenm);
953 lsi->lsi_llsbi = NULL;
955 lustre_common_put_super(sb);
957 cl_env_cache_purge(~0);
959 LCONSOLE_WARN("client %s umount complete\n", ll_instance);
961 cfs_module_put(THIS_MODULE);
964 } /* client_put_super */
966 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
968 struct inode *inode = NULL;
969 /* NOTE: we depend on atomic igrab() -bzzz */
970 lock_res_and_lock(lock);
971 if (lock->l_ast_data) {
972 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
973 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
974 inode = igrab(lock->l_ast_data);
976 inode = lock->l_ast_data;
977 ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
979 lock, __FILE__, __func__, __LINE__,
980 "l_ast_data %p is bogus: magic %08x",
981 lock->l_ast_data, lli->lli_inode_magic);
985 unlock_res_and_lock(lock);
989 static int null_if_equal(struct ldlm_lock *lock, void *data)
991 if (data == lock->l_ast_data) {
992 lock->l_ast_data = NULL;
994 if (lock->l_req_mode != lock->l_granted_mode)
995 LDLM_ERROR(lock,"clearing inode with ungranted lock");
998 return LDLM_ITER_CONTINUE;
1001 void ll_clear_inode(struct inode *inode)
1003 struct ll_inode_info *lli = ll_i2info(inode);
1004 struct ll_sb_info *sbi = ll_i2sbi(inode);
1007 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1008 inode->i_generation, inode);
1010 if (S_ISDIR(inode->i_mode)) {
1011 /* these should have been cleared in ll_file_release */
1012 LASSERT(lli->lli_sai == NULL);
1013 LASSERT(lli->lli_opendir_key == NULL);
1014 LASSERT(lli->lli_opendir_pid == 0);
1017 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1018 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1019 null_if_equal, inode);
1021 LASSERT(!lli->lli_open_fd_write_count);
1022 LASSERT(!lli->lli_open_fd_read_count);
1023 LASSERT(!lli->lli_open_fd_exec_count);
1025 if (lli->lli_mds_write_och)
1026 ll_md_real_close(inode, FMODE_WRITE);
1027 if (lli->lli_mds_exec_och)
1028 ll_md_real_close(inode, FMODE_EXEC);
1029 if (lli->lli_mds_read_och)
1030 ll_md_real_close(inode, FMODE_READ);
1032 if (lli->lli_symlink_name) {
1033 OBD_FREE(lli->lli_symlink_name,
1034 strlen(lli->lli_symlink_name) + 1);
1035 lli->lli_symlink_name = NULL;
1038 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1039 LASSERT(lli->lli_posix_acl == NULL);
1040 if (lli->lli_remote_perms) {
1041 free_rmtperm_hash(lli->lli_remote_perms);
1042 lli->lli_remote_perms = NULL;
1045 #ifdef CONFIG_FS_POSIX_ACL
1046 else if (lli->lli_posix_acl) {
1047 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1048 LASSERT(lli->lli_remote_perms == NULL);
1049 posix_acl_release(lli->lli_posix_acl);
1050 lli->lli_posix_acl = NULL;
1053 lli->lli_inode_magic = LLI_INODE_DEAD;
1055 ll_clear_inode_capas(inode);
1057 * XXX This has to be done before lsm is freed below, because
1058 * cl_object still uses inode lsm.
1060 cl_inode_fini(inode);
1063 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1064 lli->lli_smd = NULL;
1071 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1072 struct md_open_data **mod)
1074 struct lustre_md md;
1075 struct ll_sb_info *sbi = ll_i2sbi(inode);
1076 struct ptlrpc_request *request = NULL;
1080 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1081 LUSTRE_OPC_ANY, NULL);
1082 if (IS_ERR(op_data))
1083 RETURN(PTR_ERR(op_data));
1085 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1088 ptlrpc_req_finished(request);
1089 if (rc == -ENOENT) {
1091 /* Unlinked special device node? Or just a race?
1092 * Pretend we done everything. */
1093 if (!S_ISREG(inode->i_mode) &&
1094 !S_ISDIR(inode->i_mode))
1095 rc = inode_setattr(inode, &op_data->op_attr);
1096 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1097 CERROR("md_setattr fails: rc = %d\n", rc);
1102 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1103 sbi->ll_md_exp, &md);
1105 ptlrpc_req_finished(request);
1109 /* We call inode_setattr to adjust timestamps.
1110 * If there is at least some data in file, we cleared ATTR_SIZE
1111 * above to avoid invoking vmtruncate, otherwise it is important
1112 * to call vmtruncate in inode_setattr to update inode->i_size
1114 rc = inode_setattr(inode, &op_data->op_attr);
1116 /* Extract epoch data if obtained. */
1117 op_data->op_handle = md.body->handle;
1118 op_data->op_ioepoch = md.body->ioepoch;
1120 ll_update_inode(inode, &md);
1121 ptlrpc_req_finished(request);
1126 /* Close IO epoch and send Size-on-MDS attribute update. */
1127 static int ll_setattr_done_writing(struct inode *inode,
1128 struct md_op_data *op_data,
1129 struct md_open_data *mod)
1131 struct ll_inode_info *lli = ll_i2info(inode);
1135 LASSERT(op_data != NULL);
1136 if (!S_ISREG(inode->i_mode))
1139 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1140 op_data->op_ioepoch, PFID(&lli->lli_fid));
1142 op_data->op_flags = MF_EPOCH_CLOSE;
1143 ll_done_writing_attr(inode, op_data);
1144 ll_pack_inode2opdata(inode, op_data, NULL);
1146 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1147 if (rc == -EAGAIN) {
1148 /* MDS has instructed us to obtain Size-on-MDS attribute
1149 * from OSTs and send setattr to back to MDS. */
1150 rc = ll_som_update(inode, op_data);
1152 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1158 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1160 struct obd_capa *capa;
1163 if (attr->ia_valid & ATTR_SIZE)
1164 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1166 capa = ll_mdscapa_get(inode);
1168 rc = cl_setattr_ost(inode, attr, capa);
1170 if (attr->ia_valid & ATTR_SIZE)
1171 ll_truncate_free_capa(capa);
1178 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1179 * object(s) determine the file size and mtime. Otherwise, the MDS will
1180 * keep these values until such a time that objects are allocated for it.
1181 * We do the MDS operations first, as it is checking permissions for us.
1182 * We don't to the MDS RPC if there is nothing that we want to store there,
1183 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1184 * going to do an RPC anyways.
1186 * If we are doing a truncate, we will send the mtime and ctime updates
1187 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1188 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1191 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1193 struct ll_inode_info *lli = ll_i2info(inode);
1194 struct lov_stripe_md *lsm = lli->lli_smd;
1195 struct md_op_data *op_data = NULL;
1196 struct md_open_data *mod = NULL;
1197 int ia_valid = attr->ia_valid;
1198 int rc = 0, rc1 = 0;
1201 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1203 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1205 if (ia_valid & ATTR_SIZE) {
1206 if (attr->ia_size > ll_file_maxbytes(inode)) {
1207 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1208 attr->ia_size, ll_file_maxbytes(inode));
1212 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1215 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1216 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1217 if (cfs_curproc_fsuid() != inode->i_uid &&
1218 !cfs_capable(CFS_CAP_FOWNER))
1222 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1223 if (attr->ia_valid & ATTR_CTIME) {
1224 attr->ia_ctime = CFS_CURRENT_TIME;
1225 attr->ia_valid |= ATTR_CTIME_SET;
1227 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1228 attr->ia_atime = CFS_CURRENT_TIME;
1229 attr->ia_valid |= ATTR_ATIME_SET;
1231 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1232 attr->ia_mtime = CFS_CURRENT_TIME;
1233 attr->ia_valid |= ATTR_MTIME_SET;
1236 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1237 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1238 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1239 cfs_time_current_sec());
1241 /* NB: ATTR_SIZE will only be set after this point if the size
1242 * resides on the MDS, ie, this file has no objects. */
1244 attr->ia_valid &= ~ATTR_SIZE;
1246 /* We always do an MDS RPC, even if we're only changing the size;
1247 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1249 OBD_ALLOC_PTR(op_data);
1250 if (op_data == NULL)
1253 UNLOCK_INODE_MUTEX(inode);
1254 if (ia_valid & ATTR_SIZE)
1255 UP_WRITE_I_ALLOC_SEM(inode);
1256 cfs_down(&lli->lli_trunc_sem);
1257 LOCK_INODE_MUTEX(inode);
1258 if (ia_valid & ATTR_SIZE)
1259 DOWN_WRITE_I_ALLOC_SEM(inode);
1261 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1263 /* Open epoch for truncate. */
1264 if (exp_connect_som(ll_i2mdexp(inode)) &&
1265 (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1266 op_data->op_flags = MF_EPOCH_OPEN;
1268 rc = ll_md_setattr(inode, op_data, &mod);
1272 ll_ioepoch_open(lli, op_data->op_ioepoch);
1273 if (!lsm || !S_ISREG(inode->i_mode)) {
1274 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1278 if (ia_valid & ATTR_SIZE)
1279 attr->ia_valid |= ATTR_SIZE;
1280 if ((ia_valid & ATTR_SIZE) |
1281 ((ia_valid | ATTR_ATIME | ATTR_ATIME_SET) &&
1282 LTIME_S(attr->ia_atime) < LTIME_S(attr->ia_ctime)) ||
1283 ((ia_valid | ATTR_MTIME | ATTR_MTIME_SET) &&
1284 LTIME_S(attr->ia_mtime) < LTIME_S(attr->ia_ctime)))
1285 /* perform truncate and setting mtime/atime to past under PW
1286 * 0:EOF extent lock (new_size:EOF for truncate) */
1287 rc = ll_setattr_ost(inode, attr);
1291 if (op_data->op_ioepoch)
1292 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1293 ll_finish_md_op_data(op_data);
1295 cfs_up(&lli->lli_trunc_sem);
1296 return rc ? rc : rc1;
1299 int ll_setattr(struct dentry *de, struct iattr *attr)
1301 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1302 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1303 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1305 if ((de->d_inode->i_mode & S_ISUID) &&
1306 !(attr->ia_mode & S_ISUID) &&
1307 !(attr->ia_valid & ATTR_KILL_SUID))
1308 attr->ia_valid |= ATTR_KILL_SUID;
1310 if (((de->d_inode->i_mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1311 !(attr->ia_mode & S_ISGID) &&
1312 !(attr->ia_valid & ATTR_KILL_SGID))
1313 attr->ia_valid |= ATTR_KILL_SGID;
1315 return ll_setattr_raw(de->d_inode, attr);
1318 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1319 __u64 max_age, __u32 flags)
1321 struct ll_sb_info *sbi = ll_s2sbi(sb);
1322 struct obd_statfs obd_osfs;
1326 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1328 CERROR("md_statfs fails: rc = %d\n", rc);
1332 osfs->os_type = sb->s_magic;
1334 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1335 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1337 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1338 flags |= OBD_STATFS_NODELAY;
1340 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1341 &obd_osfs, max_age, flags);
1343 CERROR("obd_statfs fails: rc = %d\n", rc);
1347 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1348 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1351 osfs->os_bsize = obd_osfs.os_bsize;
1352 osfs->os_blocks = obd_osfs.os_blocks;
1353 osfs->os_bfree = obd_osfs.os_bfree;
1354 osfs->os_bavail = obd_osfs.os_bavail;
1356 /* If we don't have as many objects free on the OST as inodes
1357 * on the MDS, we reduce the total number of inodes to
1358 * compensate, so that the "inodes in use" number is correct.
1360 if (obd_osfs.os_ffree < osfs->os_ffree) {
1361 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1363 osfs->os_ffree = obd_osfs.os_ffree;
1368 #ifndef HAVE_STATFS_DENTRY_PARAM
1369 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1372 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1374 struct super_block *sb = de->d_sb;
1376 struct obd_statfs osfs;
1379 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1380 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1382 /* Some amount of caching on the client is allowed */
1383 rc = ll_statfs_internal(sb, &osfs,
1384 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1389 statfs_unpack(sfs, &osfs);
1391 /* We need to downshift for all 32-bit kernels, because we can't
1392 * tell if the kernel is being called via sys_statfs64() or not.
1393 * Stop before overflowing f_bsize - in which case it is better
1394 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1395 if (sizeof(long) < 8) {
1396 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1399 osfs.os_blocks >>= 1;
1400 osfs.os_bfree >>= 1;
1401 osfs.os_bavail >>= 1;
1405 sfs->f_blocks = osfs.os_blocks;
1406 sfs->f_bfree = osfs.os_bfree;
1407 sfs->f_bavail = osfs.os_bavail;
1412 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1414 struct ll_inode_info *lli;
1415 struct lov_stripe_md *lsm;
1417 lli = ll_i2info(inode);
1418 LASSERT(lli->lli_size_sem_owner != current);
1419 cfs_down(&lli->lli_size_sem);
1420 LASSERT(lli->lli_size_sem_owner == NULL);
1421 lli->lli_size_sem_owner = current;
1423 LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1426 lov_stripe_lock(lsm);
1429 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1431 struct ll_inode_info *lli;
1432 struct lov_stripe_md *lsm;
1434 lli = ll_i2info(inode);
1436 LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1439 lov_stripe_unlock(lsm);
1440 LASSERT(lli->lli_size_sem_owner == current);
1441 lli->lli_size_sem_owner = NULL;
1442 cfs_up(&lli->lli_size_sem);
1445 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1447 struct ll_inode_info *lli = ll_i2info(inode);
1448 struct mdt_body *body = md->body;
1449 struct lov_stripe_md *lsm = md->lsm;
1450 struct ll_sb_info *sbi = ll_i2sbi(inode);
1452 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1454 cfs_down(&lli->lli_och_sem);
1455 if (lli->lli_smd == NULL) {
1456 if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1457 lsm->lsm_magic != LOV_MAGIC_V3) {
1458 dump_lsm(D_ERROR, lsm);
1461 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1462 lsm, inode->i_ino, inode->i_generation, inode);
1463 /* cl_inode_init must go before lli_smd or a race is
1464 * possible where client thinks the file has stripes,
1465 * but lov raid0 is not setup yet and parallel e.g.
1466 * glimpse would try to use uninitialized lov */
1467 cl_inode_init(inode, md);
1469 cfs_up(&lli->lli_och_sem);
1470 lli->lli_maxbytes = lsm->lsm_maxbytes;
1471 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1472 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1474 cfs_up(&lli->lli_och_sem);
1475 LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1476 lli->lli_smd->lsm_stripe_count ==
1477 lsm->lsm_stripe_count);
1478 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1479 CERROR("lsm mismatch for inode %ld\n",
1481 CERROR("lli_smd:\n");
1482 dump_lsm(D_ERROR, lli->lli_smd);
1484 dump_lsm(D_ERROR, lsm);
1488 if (lli->lli_smd != lsm)
1489 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1492 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1493 if (body->valid & OBD_MD_FLRMTPERM)
1494 ll_update_remote_perm(inode, md->remote_perm);
1496 #ifdef CONFIG_FS_POSIX_ACL
1497 else if (body->valid & OBD_MD_FLACL) {
1498 cfs_spin_lock(&lli->lli_lock);
1499 if (lli->lli_posix_acl)
1500 posix_acl_release(lli->lli_posix_acl);
1501 lli->lli_posix_acl = md->posix_acl;
1502 cfs_spin_unlock(&lli->lli_lock);
1505 inode->i_ino = cl_fid_build_ino(&body->fid1);
1506 inode->i_generation = cl_fid_build_gen(&body->fid1);
1508 if (body->valid & OBD_MD_FLATIME) {
1509 if (body->atime > LTIME_S(inode->i_atime))
1510 LTIME_S(inode->i_atime) = body->atime;
1511 lli->lli_lvb.lvb_atime = body->atime;
1513 if (body->valid & OBD_MD_FLMTIME) {
1514 if (body->mtime > LTIME_S(inode->i_mtime)) {
1515 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1516 "to "LPU64"\n", inode->i_ino,
1517 LTIME_S(inode->i_mtime), body->mtime);
1518 LTIME_S(inode->i_mtime) = body->mtime;
1520 lli->lli_lvb.lvb_mtime = body->mtime;
1522 if (body->valid & OBD_MD_FLCTIME) {
1523 if (body->ctime > LTIME_S(inode->i_ctime))
1524 LTIME_S(inode->i_ctime) = body->ctime;
1525 lli->lli_lvb.lvb_ctime = body->ctime;
1527 if (body->valid & OBD_MD_FLMODE)
1528 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1529 if (body->valid & OBD_MD_FLTYPE)
1530 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1531 LASSERT(inode->i_mode != 0);
1532 if (S_ISREG(inode->i_mode)) {
1533 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1535 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1537 #ifdef HAVE_INODE_BLKSIZE
1538 inode->i_blksize = 1<<inode->i_blkbits;
1540 if (body->valid & OBD_MD_FLUID)
1541 inode->i_uid = body->uid;
1542 if (body->valid & OBD_MD_FLGID)
1543 inode->i_gid = body->gid;
1544 if (body->valid & OBD_MD_FLFLAGS)
1545 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1546 if (body->valid & OBD_MD_FLNLINK)
1547 inode->i_nlink = body->nlink;
1548 if (body->valid & OBD_MD_FLRDEV)
1549 inode->i_rdev = old_decode_dev(body->rdev);
1551 if (body->valid & OBD_MD_FLID) {
1552 /* FID shouldn't be changed! */
1553 if (fid_is_sane(&lli->lli_fid)) {
1554 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1555 "Trying to change FID "DFID
1556 " to the "DFID", inode %lu/%u(%p)\n",
1557 PFID(&lli->lli_fid), PFID(&body->fid1),
1558 inode->i_ino, inode->i_generation, inode);
1560 lli->lli_fid = body->fid1;
1563 LASSERT(fid_seq(&lli->lli_fid) != 0);
1565 if (body->valid & OBD_MD_FLSIZE) {
1566 if (exp_connect_som(ll_i2mdexp(inode)) &&
1567 S_ISREG(inode->i_mode) && lli->lli_smd) {
1568 struct lustre_handle lockh;
1571 /* As it is possible a blocking ast has been processed
1572 * by this time, we need to check there is an UPDATE
1573 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1575 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1578 if (lli->lli_flags & (LLIF_DONE_WRITING |
1579 LLIF_EPOCH_PENDING |
1581 CERROR("ino %lu flags %lu still has "
1582 "size authority! do not trust "
1583 "the size got from MDS\n",
1584 inode->i_ino, lli->lli_flags);
1586 /* Use old size assignment to avoid
1587 * deadlock bz14138 & bz14326 */
1588 inode->i_size = body->size;
1589 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1591 ldlm_lock_decref(&lockh, mode);
1594 /* Use old size assignment to avoid
1595 * deadlock bz14138 & bz14326 */
1596 inode->i_size = body->size;
1598 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1599 inode->i_ino, (unsigned long long)body->size);
1602 if (body->valid & OBD_MD_FLBLOCKS)
1603 inode->i_blocks = body->blocks;
1606 if (body->valid & OBD_MD_FLMDSCAPA) {
1607 LASSERT(md->mds_capa);
1608 ll_add_capa(inode, md->mds_capa);
1610 if (body->valid & OBD_MD_FLOSSCAPA) {
1611 LASSERT(md->oss_capa);
1612 ll_add_capa(inode, md->oss_capa);
1616 static struct backing_dev_info ll_backing_dev_info = {
1617 .ra_pages = 0, /* No readahead */
1618 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1619 .capabilities = 0, /* Does contribute to dirty memory */
1621 .memory_backed = 0, /* Does contribute to dirty memory */
1625 void ll_read_inode2(struct inode *inode, void *opaque)
1627 struct lustre_md *md = opaque;
1628 struct ll_inode_info *lli = ll_i2info(inode);
1631 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1632 inode->i_ino, inode->i_generation, inode);
1636 LASSERT(!lli->lli_smd);
1638 /* Core attributes from the MDS first. This is a new inode, and
1639 * the VFS doesn't zero times in the core inode so we have to do
1640 * it ourselves. They will be overwritten by either MDS or OST
1641 * attributes - we just need to make sure they aren't newer. */
1642 LTIME_S(inode->i_mtime) = 0;
1643 LTIME_S(inode->i_atime) = 0;
1644 LTIME_S(inode->i_ctime) = 0;
1646 ll_update_inode(inode, md);
1648 /* OIDEBUG(inode); */
1650 if (S_ISREG(inode->i_mode)) {
1651 struct ll_sb_info *sbi = ll_i2sbi(inode);
1652 inode->i_op = &ll_file_inode_operations;
1653 inode->i_fop = sbi->ll_fop;
1654 inode->i_mapping->a_ops = &ll_aops;
1656 } else if (S_ISDIR(inode->i_mode)) {
1657 inode->i_op = &ll_dir_inode_operations;
1658 inode->i_fop = &ll_dir_operations;
1659 inode->i_mapping->a_ops = &ll_dir_aops;
1661 } else if (S_ISLNK(inode->i_mode)) {
1662 inode->i_op = &ll_fast_symlink_inode_operations;
1665 inode->i_op = &ll_special_inode_operations;
1667 init_special_inode(inode, inode->i_mode,
1668 kdev_t_to_nr(inode->i_rdev));
1670 /* initializing backing dev info. */
1671 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1677 void ll_delete_inode(struct inode *inode)
1679 struct ll_sb_info *sbi = ll_i2sbi(inode);
1683 rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1685 CERROR("fid_delete() failed, rc %d\n", rc);
1687 truncate_inode_pages(&inode->i_data, 0);
1693 int ll_iocontrol(struct inode *inode, struct file *file,
1694 unsigned int cmd, unsigned long arg)
1696 struct ll_sb_info *sbi = ll_i2sbi(inode);
1697 struct ptlrpc_request *req = NULL;
1702 case FSFILT_IOC_GETFLAGS: {
1703 struct mdt_body *body;
1704 struct md_op_data *op_data;
1706 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1707 0, 0, LUSTRE_OPC_ANY,
1709 if (op_data == NULL)
1712 op_data->op_valid = OBD_MD_FLFLAGS;
1713 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1714 ll_finish_md_op_data(op_data);
1716 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1720 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1722 flags = body->flags;
1724 ptlrpc_req_finished(req);
1726 RETURN(put_user(flags, (int *)arg));
1728 case FSFILT_IOC_SETFLAGS: {
1729 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1730 struct obd_info oinfo = { { { 0 } } };
1731 struct md_op_data *op_data;
1733 if (get_user(flags, (int *)arg))
1737 OBDO_ALLOC(oinfo.oi_oa);
1741 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1742 LUSTRE_OPC_ANY, NULL);
1743 if (IS_ERR(op_data))
1744 RETURN(PTR_ERR(op_data));
1746 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1747 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1748 rc = md_setattr(sbi->ll_md_exp, op_data,
1749 NULL, 0, NULL, 0, &req, NULL);
1750 ll_finish_md_op_data(op_data);
1751 ptlrpc_req_finished(req);
1753 OBDO_FREE(oinfo.oi_oa);
1758 OBDO_FREE(oinfo.oi_oa);
1759 GOTO(update_cache, rc);
1762 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1763 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1764 oinfo.oi_oa->o_flags = flags;
1765 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1767 oinfo.oi_capa = ll_mdscapa_get(inode);
1768 obdo_from_inode(oinfo.oi_oa, inode,
1769 &ll_i2info(inode)->lli_fid, 0);
1770 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1771 capa_put(oinfo.oi_capa);
1772 OBDO_FREE(oinfo.oi_oa);
1774 if (rc != -EPERM && rc != -EACCES)
1775 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1781 inode->i_flags = ll_ext_to_inode_flags(flags);
1791 int ll_flush_ctx(struct inode *inode)
1793 struct ll_sb_info *sbi = ll_i2sbi(inode);
1795 CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1797 obd_set_info_async(sbi->ll_md_exp,
1798 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1800 obd_set_info_async(sbi->ll_dt_exp,
1801 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1806 /* umount -f client means force down, don't save state */
1807 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1808 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1810 struct super_block *sb = vfsmnt->mnt_sb;
1812 void ll_umount_begin(struct super_block *sb)
1815 struct lustre_sb_info *lsi = s2lsi(sb);
1816 struct ll_sb_info *sbi = ll_s2sbi(sb);
1817 struct obd_device *obd;
1818 struct obd_ioctl_data *ioc_data;
1821 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1822 if (!(flags & MNT_FORCE)) {
1828 /* Tell the MGC we got umount -f */
1829 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1831 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1832 sb->s_count, atomic_read(&sb->s_active));
1834 obd = class_exp2obd(sbi->ll_md_exp);
1836 CERROR("Invalid MDC connection handle "LPX64"\n",
1837 sbi->ll_md_exp->exp_handle.h_cookie);
1843 obd = class_exp2obd(sbi->ll_dt_exp);
1845 CERROR("Invalid LOV connection handle "LPX64"\n",
1846 sbi->ll_dt_exp->exp_handle.h_cookie);
1852 OBD_ALLOC_PTR(ioc_data);
1854 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1855 sizeof ioc_data, ioc_data, NULL);
1857 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1858 sizeof ioc_data, ioc_data, NULL);
1860 OBD_FREE_PTR(ioc_data);
1864 /* Really, we'd like to wait until there are no requests outstanding,
1865 * and then continue. For now, we just invalidate the requests,
1866 * schedule() and sleep one second if needed, and hope.
1869 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1870 if (atomic_read(&vfsmnt->mnt_count) > 2) {
1871 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
1872 cfs_time_seconds(1));
1873 if (atomic_read(&vfsmnt->mnt_count) > 2)
1874 LCONSOLE_WARN("Mount still busy with %d refs! You "
1875 "may try to umount it a bit later\n",
1876 atomic_read(&vfsmnt->mnt_count));
1883 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1885 struct ll_sb_info *sbi = ll_s2sbi(sb);
1889 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1890 read_only = *flags & MS_RDONLY;
1891 err = obd_set_info_async(sbi->ll_md_exp,
1892 sizeof(KEY_READ_ONLY),
1893 KEY_READ_ONLY, sizeof(read_only),
1896 CERROR("Failed to change the read-only flag during "
1897 "remount: %d\n", err);
1902 sb->s_flags |= MS_RDONLY;
1904 sb->s_flags &= ~MS_RDONLY;
1909 int ll_prep_inode(struct inode **inode,
1910 struct ptlrpc_request *req,
1911 struct super_block *sb)
1913 struct ll_sb_info *sbi = NULL;
1914 struct lustre_md md;
1918 LASSERT(*inode || sb);
1919 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
1920 memset(&md, 0, sizeof(struct lustre_md));
1922 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
1923 sbi->ll_md_exp, &md);
1928 ll_update_inode(*inode, &md);
1930 LASSERT(sb != NULL);
1933 * At this point server returns to client's same fid as client
1934 * generated for creating. So using ->fid1 is okay here.
1936 LASSERT(fid_is_sane(&md.body->fid1));
1938 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1), &md);
1939 if (*inode == NULL || IS_ERR(*inode)) {
1941 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
1942 #ifdef CONFIG_FS_POSIX_ACL
1944 posix_acl_release(md.posix_acl);
1945 md.posix_acl = NULL;
1948 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
1950 CERROR("new_inode -fatal: rc %d\n", rc);
1956 md_free_lustre_md(sbi->ll_md_exp, &md);
1960 int ll_obd_statfs(struct inode *inode, void *arg)
1962 struct ll_sb_info *sbi = NULL;
1963 struct obd_export *exp;
1965 struct obd_ioctl_data *data = NULL;
1969 if (!inode || !(sbi = ll_i2sbi(inode)))
1970 GOTO(out_statfs, rc = -EINVAL);
1972 rc = obd_ioctl_getdata(&buf, &len, arg);
1974 GOTO(out_statfs, rc);
1977 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
1978 !data->ioc_pbuf1 || !data->ioc_pbuf2)
1979 GOTO(out_statfs, rc = -EINVAL);
1981 if (data->ioc_inllen1 != sizeof(__u32) ||
1982 data->ioc_inllen2 != sizeof(__u32) ||
1983 data->ioc_plen1 != sizeof(struct obd_statfs) ||
1984 data->ioc_plen2 != sizeof(struct obd_uuid))
1985 GOTO(out_statfs, rc = -EINVAL);
1987 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
1988 if (type == LL_STATFS_MDC)
1989 exp = sbi->ll_md_exp;
1990 else if (type == LL_STATFS_LOV)
1991 exp = sbi->ll_dt_exp;
1993 GOTO(out_statfs, rc = -ENODEV);
1995 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
1997 GOTO(out_statfs, rc);
2000 obd_ioctl_freedata(buf, len);
2004 int ll_process_config(struct lustre_cfg *lcfg)
2008 struct lprocfs_static_vars lvars;
2012 lprocfs_llite_init_vars(&lvars);
2014 /* The instance name contains the sb: lustre-client-aacfe000 */
2015 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2016 if (!ptr || !*(++ptr))
2018 if (sscanf(ptr, "%lx", &x) != 1)
2021 /* This better be a real Lustre superblock! */
2022 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2024 /* Note we have not called client_common_fill_super yet, so
2025 proc fns must be able to handle that! */
2026 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2033 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2034 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2035 struct inode *i1, struct inode *i2,
2036 const char *name, int namelen,
2037 int mode, __u32 opc, void *data)
2039 LASSERT(i1 != NULL);
2041 if (namelen > ll_i2sbi(i1)->ll_namelen)
2042 return ERR_PTR(-ENAMETOOLONG);
2044 if (op_data == NULL)
2045 OBD_ALLOC_PTR(op_data);
2047 if (op_data == NULL)
2048 return ERR_PTR(-ENOMEM);
2050 ll_i2gids(op_data->op_suppgids, i1, i2);
2051 op_data->op_fid1 = *ll_inode2fid(i1);
2052 op_data->op_capa1 = ll_mdscapa_get(i1);
2055 op_data->op_fid2 = *ll_inode2fid(i2);
2056 op_data->op_capa2 = ll_mdscapa_get(i2);
2058 fid_zero(&op_data->op_fid2);
2059 op_data->op_capa2 = NULL;
2062 op_data->op_name = name;
2063 op_data->op_namelen = namelen;
2064 op_data->op_mode = mode;
2065 op_data->op_mod_time = cfs_time_current_sec();
2066 op_data->op_fsuid = cfs_curproc_fsuid();
2067 op_data->op_fsgid = cfs_curproc_fsgid();
2068 op_data->op_cap = cfs_curproc_cap_pack();
2069 op_data->op_bias = MDS_CHECK_SPLIT;
2070 op_data->op_opc = opc;
2071 op_data->op_mds = 0;
2072 op_data->op_data = data;
2077 void ll_finish_md_op_data(struct md_op_data *op_data)
2079 capa_put(op_data->op_capa1);
2080 capa_put(op_data->op_capa2);
2081 OBD_FREE_PTR(op_data);
2084 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2086 struct ll_sb_info *sbi;
2088 LASSERT((seq != NULL) && (vfs != NULL));
2089 sbi = ll_s2sbi(vfs->mnt_sb);
2091 if (sbi->ll_flags & LL_SBI_NOLCK)
2092 seq_puts(seq, ",nolock");
2094 if (sbi->ll_flags & LL_SBI_FLOCK)
2095 seq_puts(seq, ",flock");
2097 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2098 seq_puts(seq, ",localflock");
2100 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2101 seq_puts(seq, ",user_xattr");
2103 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2104 seq_puts(seq, ",lazystatfs");