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 2008 Sun Microsystems, Inc. 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/random.h>
46 #include <linux/version.h>
49 #include <lustre_lite.h>
50 #include <lustre_ha.h>
51 #include <lustre_dlm.h>
52 #include <lprocfs_status.h>
53 #include <lustre_disk.h>
54 #include <lustre_param.h>
55 #include <lustre_log.h>
56 #include <cl_object.h>
57 #include <obd_cksum.h>
58 #include "llite_internal.h"
60 cfs_mem_cache_t *ll_file_data_slab;
62 LIST_HEAD(ll_super_blocks);
63 spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
65 extern struct address_space_operations ll_aops;
66 extern struct address_space_operations ll_dir_aops;
69 #define log2(n) ffz(~(n))
72 static struct ll_sb_info *ll_init_sbi(void)
74 struct ll_sb_info *sbi = NULL;
81 OBD_ALLOC(sbi, sizeof(*sbi));
85 spin_lock_init(&sbi->ll_lock);
86 init_mutex(&sbi->ll_lco.lco_lock);
87 spin_lock_init(&sbi->ll_pp_extent_lock);
88 spin_lock_init(&sbi->ll_process_lock);
89 sbi->ll_rw_stats_on = 0;
92 pages = si.totalram - si.totalhigh;
93 if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
94 #ifdef HAVE_BGL_SUPPORT
95 sbi->ll_async_page_max = pages / 4;
97 sbi->ll_async_page_max = pages / 2;
100 sbi->ll_async_page_max = (pages / 4) * 3;
103 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
104 SBI_DEFAULT_READAHEAD_MAX);
105 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
106 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
107 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
108 INIT_LIST_HEAD(&sbi->ll_conn_chain);
109 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
111 ll_generate_random_uuid(uuid);
112 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
113 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
115 spin_lock(&ll_sb_lock);
116 list_add_tail(&sbi->ll_list, &ll_super_blocks);
117 spin_unlock(&ll_sb_lock);
119 #ifdef ENABLE_LLITE_CHECKSUM
120 sbi->ll_flags |= LL_SBI_CHECKSUM;
123 #ifdef HAVE_LRU_RESIZE_SUPPORT
124 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
127 #ifdef HAVE_EXPORT___IGET
128 INIT_LIST_HEAD(&sbi->ll_deathrow);
129 spin_lock_init(&sbi->ll_deathrow_lock);
131 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
132 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
133 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
136 /* metadata statahead is enabled by default */
137 sbi->ll_sa_max = LL_SA_RPC_DEF;
142 void ll_free_sbi(struct super_block *sb)
144 struct ll_sb_info *sbi = ll_s2sbi(sb);
148 spin_lock(&ll_sb_lock);
149 list_del(&sbi->ll_list);
150 spin_unlock(&ll_sb_lock);
151 OBD_FREE(sbi, sizeof(*sbi));
156 static struct dentry_operations ll_d_root_ops = {
157 .d_compare = ll_dcompare,
158 .d_revalidate = ll_revalidate_nd,
161 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
163 struct inode *root = 0;
164 struct ll_sb_info *sbi = ll_s2sbi(sb);
165 struct obd_device *obd;
166 struct obd_capa *oc = NULL;
167 struct obd_statfs osfs;
168 struct ptlrpc_request *request = NULL;
169 struct obd_connect_data *data = NULL;
170 struct obd_uuid *uuid;
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_SOM;
202 #ifdef HAVE_LRU_RESIZE_SUPPORT
203 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
204 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
206 #ifdef CONFIG_FS_POSIX_ACL
207 data->ocd_connect_flags |= OBD_CONNECT_ACL;
209 data->ocd_ibits_known = MDS_INODELOCK_FULL;
210 data->ocd_version = LUSTRE_VERSION_CODE;
212 if (sb->s_flags & MS_RDONLY)
213 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
214 if (sbi->ll_flags & LL_SBI_USER_XATTR)
215 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
217 #ifdef HAVE_MS_FLOCK_LOCK
218 /* force vfs to use lustre handler for flock() calls - bug 10743 */
219 sb->s_flags |= MS_FLOCK_LOCK;
222 if (sbi->ll_flags & LL_SBI_FLOCK)
223 sbi->ll_fop = &ll_file_operations_flock;
224 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
225 sbi->ll_fop = &ll_file_operations;
227 sbi->ll_fop = &ll_file_operations_noflock;
230 data->ocd_connect_flags |= OBD_CONNECT_REAL;
231 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
232 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
234 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
236 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
237 "recovery, of which this client is not a "
238 "part. Please wait for recovery to complete,"
239 " abort, or time out.\n", md);
242 CERROR("cannot connect to %s: rc = %d\n", md, err);
246 err = obd_fid_init(sbi->ll_md_exp);
248 CERROR("Can't init metadata layer FID infrastructure, "
253 err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ, 0);
255 GOTO(out_md_fid, err);
257 size = sizeof(*data);
258 err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
259 KEY_CONN_DATA, &size, data, NULL);
261 CERROR("Get connect data failed: %d \n", err);
265 LASSERT(osfs.os_bsize);
266 sb->s_blocksize = osfs.os_bsize;
267 sb->s_blocksize_bits = log2(osfs.os_bsize);
268 sb->s_magic = LL_SUPER_MAGIC;
270 /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
271 * retval = in_file->f_op->sendfile(...);
273 * retval = -EOVERFLOW;
275 * it will check if *ppos is greater than max. However, max equals to
276 * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
277 * has been defined as a signed long long integer in linux kernel. */
278 #if BITS_PER_LONG == 64
279 sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
281 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
283 sbi->ll_namelen = osfs.os_namelen;
284 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
286 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
287 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
288 LCONSOLE_INFO("Disabling user_xattr feature because "
289 "it is not supported on the server\n");
290 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
293 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
295 sb->s_flags |= MS_POSIXACL;
297 sbi->ll_flags |= LL_SBI_ACL;
299 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
301 sb->s_flags &= ~MS_POSIXACL;
303 sbi->ll_flags &= ~LL_SBI_ACL;
306 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
307 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
308 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
309 LCONSOLE_INFO("client is set as remote by default.\n");
312 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
313 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
314 LCONSOLE_INFO("client claims to be remote, but server "
315 "rejected, forced to be local.\n");
319 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
320 LCONSOLE_INFO("client enabled MDS capability!\n");
321 sbi->ll_flags |= LL_SBI_MDS_CAPA;
324 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
325 LCONSOLE_INFO("client enabled OSS capability!\n");
326 sbi->ll_flags |= LL_SBI_OSS_CAPA;
329 obd = class_name2obd(dt);
331 CERROR("DT %s: not setup or attached\n", dt);
332 GOTO(out_md_fid, err = -ENODEV);
335 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
336 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
337 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
338 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
339 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
340 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
343 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
344 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
345 * disabled by default, because it can still be enabled on the
346 * fly via /proc. As a consequence, we still need to come to an
347 * agreement on the supported algorithms at connect time */
348 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
350 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
351 data->ocd_cksum_types = OBD_CKSUM_ADLER;
353 /* send the list of supported checksum types */
354 data->ocd_cksum_types = OBD_CKSUM_ALL;
357 #ifdef HAVE_LRU_RESIZE_SUPPORT
358 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
360 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
361 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
363 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
364 "ocd_grant: %d\n", data->ocd_connect_flags,
365 data->ocd_version, data->ocd_grant);
367 obd->obd_upcall.onu_owner = &sbi->ll_lco;
368 obd->obd_upcall.onu_upcall = cl_ocd_update;
369 data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
371 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
373 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
374 "recovery, of which this client is not a "
375 "part. Please wait for recovery to "
376 "complete, abort, or time out.\n", dt);
377 GOTO(out_md_fid, err);
379 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
380 GOTO(out_md_fid, err);
383 err = obd_fid_init(sbi->ll_dt_exp);
385 CERROR("Can't init data layer FID infrastructure, "
390 mutex_down(&sbi->ll_lco.lco_lock);
391 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
392 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
393 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
394 mutex_up(&sbi->ll_lco.lco_lock);
396 fid_zero(&sbi->ll_root_fid);
397 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
399 CERROR("cannot mds_connect: rc = %d\n", err);
400 GOTO(out_lock_cn_cb, err);
402 if (!fid_is_sane(&sbi->ll_root_fid)) {
403 CERROR("Invalid root fid during mount\n");
404 GOTO(out_lock_cn_cb, err = -EINVAL);
406 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
408 sb->s_op = &lustre_super_operations;
409 sb->s_export_op = &lustre_export_operations;
412 * XXX: move this to after cbd setup? */
413 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
414 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
415 valid |= OBD_MD_FLRMTPERM;
416 else if (sbi->ll_flags & LL_SBI_ACL)
417 valid |= OBD_MD_FLACL;
419 err = md_getattr(sbi->ll_md_exp, &sbi->ll_root_fid, oc, valid, 0,
424 CERROR("md_getattr failed for root: rc = %d\n", err);
425 GOTO(out_lock_cn_cb, err);
427 memset(&lmd, 0, sizeof(lmd));
428 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
429 sbi->ll_md_exp, &lmd);
431 CERROR("failed to understand root inode md: rc = %d\n", err);
432 ptlrpc_req_finished (request);
433 GOTO(out_lock_cn_cb, err);
436 LASSERT(fid_is_sane(&sbi->ll_root_fid));
437 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid), &lmd);
438 md_free_lustre_md(sbi->ll_md_exp, &lmd);
439 ptlrpc_req_finished(request);
441 if (root == NULL || IS_ERR(root)) {
443 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
444 #ifdef CONFIG_FS_POSIX_ACL
446 posix_acl_release(lmd.posix_acl);
447 lmd.posix_acl = NULL;
450 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
452 CERROR("lustre_lite: bad iget4 for root\n");
456 err = ll_close_thread_start(&sbi->ll_lcq);
458 CERROR("cannot start close thread: rc %d\n", err);
462 #ifdef CONFIG_FS_POSIX_ACL
463 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
464 rct_init(&sbi->ll_rct);
465 et_init(&sbi->ll_et);
469 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
470 err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
471 KEY_CHECKSUM, sizeof(checksum), &checksum,
475 sb->s_root = d_alloc_root(root);
477 OBD_FREE(data, sizeof(*data));
479 sb->s_root->d_op = &ll_d_root_ops;
481 sbi->ll_sdev_orig = sb->s_dev;
483 /* We set sb->s_dev equal on all lustre clients in order to support
484 * NFS export clustering. NFSD requires that the FSID be the same
486 /* s_dev is also used in lt_compare() to compare two fs, but that is
487 * only a node-local comparison. */
488 uuid = obd_get_uuid(sbi->ll_md_exp);
490 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
497 obd_fid_fini(sbi->ll_dt_exp);
499 obd_disconnect(sbi->ll_dt_exp);
500 sbi->ll_dt_exp = NULL;
502 obd_fid_fini(sbi->ll_md_exp);
504 obd_disconnect(sbi->ll_md_exp);
505 sbi->ll_md_exp = NULL;
509 lprocfs_unregister_mountpoint(sbi);
513 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
517 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
519 rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
520 KEY_MAX_EASIZE, &size, lmmsize, NULL);
522 CERROR("Get max mdsize error rc %d \n", rc);
527 void ll_dump_inode(struct inode *inode)
529 struct list_head *tmp;
530 int dentry_count = 0;
532 LASSERT(inode != NULL);
534 list_for_each(tmp, &inode->i_dentry)
537 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
538 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
539 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
542 void lustre_dump_dentry(struct dentry *dentry, int recur)
544 struct list_head *tmp;
547 LASSERT(dentry != NULL);
549 list_for_each(tmp, &dentry->d_subdirs)
552 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
553 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
554 dentry->d_name.len, dentry->d_name.name,
555 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
556 dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
557 dentry->d_flags, dentry->d_fsdata, subdirs);
558 if (dentry->d_inode != NULL)
559 ll_dump_inode(dentry->d_inode);
564 list_for_each(tmp, &dentry->d_subdirs) {
565 struct dentry *d = list_entry(tmp, struct dentry, d_child);
566 lustre_dump_dentry(d, recur - 1);
570 #ifdef HAVE_EXPORT___IGET
571 static void prune_dir_dentries(struct inode *inode)
573 struct dentry *dentry, *prev = NULL;
575 /* due to lustre specific logic, a directory
576 * can have few dentries - a bug from VFS POV */
578 spin_lock(&dcache_lock);
579 if (!list_empty(&inode->i_dentry)) {
580 dentry = list_entry(inode->i_dentry.prev,
581 struct dentry, d_alias);
582 /* in order to prevent infinite loops we
583 * break if previous dentry is busy */
584 if (dentry != prev) {
587 spin_unlock(&dcache_lock);
589 /* try to kill all child dentries */
591 shrink_dcache_parent(dentry);
592 unlock_dentry(dentry);
595 /* now try to get rid of current dentry */
596 d_prune_aliases(inode);
600 spin_unlock(&dcache_lock);
603 static void prune_deathrow_one(struct ll_inode_info *lli)
605 struct inode *inode = ll_info2i(lli);
607 /* first, try to drop any dentries - they hold a ref on the inode */
608 if (S_ISDIR(inode->i_mode))
609 prune_dir_dentries(inode);
611 d_prune_aliases(inode);
614 /* if somebody still uses it, leave it */
615 LASSERT(atomic_read(&inode->i_count) > 0);
616 if (atomic_read(&inode->i_count) > 1)
619 CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
620 inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
622 /* seems nobody uses it anymore */
630 static void prune_deathrow(struct ll_sb_info *sbi, int try)
632 struct ll_inode_info *lli;
636 if (need_resched() && try)
640 if (!spin_trylock(&sbi->ll_deathrow_lock))
643 spin_lock(&sbi->ll_deathrow_lock);
648 if (!list_empty(&sbi->ll_deathrow)) {
649 lli = list_entry(sbi->ll_deathrow.next,
650 struct ll_inode_info,
652 list_del_init(&lli->lli_dead_list);
653 if (!list_empty(&sbi->ll_deathrow))
656 spin_unlock(&sbi->ll_deathrow_lock);
659 prune_deathrow_one(lli);
661 } while (empty == 0);
663 #else /* !HAVE_EXPORT___IGET */
664 #define prune_deathrow(sbi, try) do {} while (0)
665 #endif /* HAVE_EXPORT___IGET */
667 void client_common_put_super(struct super_block *sb)
669 struct ll_sb_info *sbi = ll_s2sbi(sb);
672 #ifdef CONFIG_FS_POSIX_ACL
673 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
674 et_fini(&sbi->ll_et);
675 rct_fini(&sbi->ll_rct);
679 obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
681 ll_close_thread_shutdown(sbi->ll_lcq);
685 /* destroy inodes in deathrow */
686 prune_deathrow(sbi, 0);
688 list_del(&sbi->ll_conn_chain);
690 obd_fid_fini(sbi->ll_dt_exp);
691 obd_disconnect(sbi->ll_dt_exp);
692 sbi->ll_dt_exp = NULL;
694 lprocfs_unregister_mountpoint(sbi);
696 obd_fid_fini(sbi->ll_md_exp);
697 obd_disconnect(sbi->ll_md_exp);
698 sbi->ll_md_exp = NULL;
703 void ll_kill_super(struct super_block *sb)
705 struct ll_sb_info *sbi;
710 if (!(sb->s_flags & MS_ACTIVE))
714 /* we need restore s_dev from changed for clustred NFS before put_super
715 * because new kernels have cached s_dev and change sb->s_dev in
716 * put_super not affected real removing devices */
718 sb->s_dev = sbi->ll_sdev_orig;
722 char *ll_read_opt(const char *opt, char *data)
728 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
729 if (strncmp(opt, data, strlen(opt)))
731 if ((value = strchr(data, '=')) == NULL)
735 OBD_ALLOC(retval, strlen(value) + 1);
737 CERROR("out of memory!\n");
741 memcpy(retval, value, strlen(value)+1);
742 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
746 static inline int ll_set_opt(const char *opt, char *data, int fl)
748 if (strncmp(opt, data, strlen(opt)) != 0)
754 /* non-client-specific mount options are parsed in lmd_parse */
755 static int ll_options(char *options, int *flags)
758 char *s1 = options, *s2;
764 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
767 CDEBUG(D_SUPER, "next opt=%s\n", s1);
768 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
773 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
778 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
783 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
788 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
793 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
798 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
800 /* Ignore deprecated mount option. The client will
801 * always try to mount with ACL support, whether this
802 * is used depends on whether server supports it. */
805 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
809 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
815 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
820 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
825 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
830 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
835 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
840 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
846 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
852 s2 = strchr(s1, ',');
860 void ll_lli_init(struct ll_inode_info *lli)
862 lli->lli_inode_magic = LLI_INODE_MAGIC;
863 sema_init(&lli->lli_size_sem, 1);
864 sema_init(&lli->lli_write_sem, 1);
865 sema_init(&lli->lli_trunc_sem, 1);
867 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
868 spin_lock_init(&lli->lli_lock);
869 INIT_LIST_HEAD(&lli->lli_close_list);
870 lli->lli_inode_magic = LLI_INODE_MAGIC;
871 sema_init(&lli->lli_och_sem, 1);
872 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
873 lli->lli_mds_exec_och = NULL;
874 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
875 lli->lli_open_fd_exec_count = 0;
876 INIT_LIST_HEAD(&lli->lli_dead_list);
877 lli->lli_remote_perms = NULL;
878 lli->lli_rmtperm_utime = 0;
879 sema_init(&lli->lli_rmtperm_sem, 1);
880 INIT_LIST_HEAD(&lli->lli_oss_capas);
883 int ll_fill_super(struct super_block *sb)
885 struct lustre_profile *lprof;
886 struct lustre_sb_info *lsi = s2lsi(sb);
887 struct ll_sb_info *sbi;
888 char *dt = NULL, *md = NULL;
889 char *profilenm = get_profile_name(sb);
890 struct config_llog_instance cfg = {0, };
891 char ll_instance[sizeof(sb) * 2 + 1];
895 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
899 /* client additional sb info */
900 lsi->lsi_llsbi = sbi = ll_init_sbi();
906 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
910 /* Generate a string unique to this super, in case some joker tries
911 to mount the same fs at two mount points.
912 Use the address of the super itself.*/
913 sprintf(ll_instance, "%p", sb);
914 cfg.cfg_instance = ll_instance;
915 cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
917 /* set up client obds */
918 err = lustre_process_log(sb, profilenm, &cfg);
920 CERROR("Unable to process log: %d\n", err);
924 lprof = class_get_profile(profilenm);
926 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
927 " read from the MGS. Does that filesystem "
928 "exist?\n", profilenm);
929 GOTO(out_free, err = -EINVAL);
931 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
932 lprof->lp_md, lprof->lp_dt);
934 OBD_ALLOC(dt, strlen(lprof->lp_dt) +
935 strlen(ll_instance) + 2);
937 GOTO(out_free, err = -ENOMEM);
938 sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
940 OBD_ALLOC(md, strlen(lprof->lp_md) +
941 strlen(ll_instance) + 2);
943 GOTO(out_free, err = -ENOMEM);
944 sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
946 /* connections, registrations, sb setup */
947 err = client_common_fill_super(sb, md, dt);
951 OBD_FREE(md, strlen(md) + 1);
953 OBD_FREE(dt, strlen(dt) + 1);
957 LCONSOLE_WARN("Client %s has started\n", profilenm);
960 } /* ll_fill_super */
963 void lu_context_keys_dump(void);
965 void ll_put_super(struct super_block *sb)
967 struct config_llog_instance cfg;
968 char ll_instance[sizeof(sb) * 2 + 1];
969 struct obd_device *obd;
970 struct lustre_sb_info *lsi = s2lsi(sb);
971 struct ll_sb_info *sbi = ll_s2sbi(sb);
972 char *profilenm = get_profile_name(sb);
976 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
978 ll_print_capa_stat(sbi);
980 sprintf(ll_instance, "%p", sb);
981 cfg.cfg_instance = ll_instance;
982 lustre_end_log(sb, NULL, &cfg);
984 if (sbi->ll_md_exp) {
985 obd = class_exp2obd(sbi->ll_md_exp);
987 force = obd->obd_force;
990 /* We need to set force before the lov_disconnect in
991 lustre_common_put_super, since l_d cleans up osc's as well. */
994 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
996 obd->obd_force = force;
1001 /* Only if client_common_fill_super succeeded */
1002 client_common_put_super(sb);
1006 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1007 class_manual_cleanup(obd);
1011 class_del_profile(profilenm);
1014 lsi->lsi_llsbi = NULL;
1016 lustre_common_put_super(sb);
1018 cl_env_cache_purge(~0);
1020 LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1025 } /* client_put_super */
1027 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1029 struct inode *inode = NULL;
1030 /* NOTE: we depend on atomic igrab() -bzzz */
1031 lock_res_and_lock(lock);
1032 if (lock->l_ast_data) {
1033 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1034 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1035 inode = igrab(lock->l_ast_data);
1037 inode = lock->l_ast_data;
1038 ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1040 lock, __FILE__, __func__, __LINE__,
1041 "l_ast_data %p is bogus: magic %08x",
1042 lock->l_ast_data, lli->lli_inode_magic);
1046 unlock_res_and_lock(lock);
1050 static int null_if_equal(struct ldlm_lock *lock, void *data)
1052 if (data == lock->l_ast_data) {
1053 lock->l_ast_data = NULL;
1055 if (lock->l_req_mode != lock->l_granted_mode)
1056 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1059 return LDLM_ITER_CONTINUE;
1062 void ll_clear_inode(struct inode *inode)
1064 struct ll_inode_info *lli = ll_i2info(inode);
1065 struct ll_sb_info *sbi = ll_i2sbi(inode);
1068 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1069 inode->i_generation, inode);
1071 if (S_ISDIR(inode->i_mode)) {
1072 /* these should have been cleared in ll_file_release */
1073 LASSERT(lli->lli_sai == NULL);
1074 LASSERT(lli->lli_opendir_key == NULL);
1075 LASSERT(lli->lli_opendir_pid == 0);
1078 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1079 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1080 null_if_equal, inode);
1082 LASSERT(!lli->lli_open_fd_write_count);
1083 LASSERT(!lli->lli_open_fd_read_count);
1084 LASSERT(!lli->lli_open_fd_exec_count);
1086 if (lli->lli_mds_write_och)
1087 ll_md_real_close(inode, FMODE_WRITE);
1088 if (lli->lli_mds_exec_och)
1089 ll_md_real_close(inode, FMODE_EXEC);
1090 if (lli->lli_mds_read_och)
1091 ll_md_real_close(inode, FMODE_READ);
1093 if (lli->lli_symlink_name) {
1094 OBD_FREE(lli->lli_symlink_name,
1095 strlen(lli->lli_symlink_name) + 1);
1096 lli->lli_symlink_name = NULL;
1099 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1100 LASSERT(lli->lli_posix_acl == NULL);
1101 if (lli->lli_remote_perms) {
1102 free_rmtperm_hash(lli->lli_remote_perms);
1103 lli->lli_remote_perms = NULL;
1106 #ifdef CONFIG_FS_POSIX_ACL
1107 else if (lli->lli_posix_acl) {
1108 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1109 LASSERT(lli->lli_remote_perms == NULL);
1110 posix_acl_release(lli->lli_posix_acl);
1111 lli->lli_posix_acl = NULL;
1114 lli->lli_inode_magic = LLI_INODE_DEAD;
1116 #ifdef HAVE_EXPORT___IGET
1117 spin_lock(&sbi->ll_deathrow_lock);
1118 list_del_init(&lli->lli_dead_list);
1119 spin_unlock(&sbi->ll_deathrow_lock);
1121 ll_clear_inode_capas(inode);
1123 * XXX This has to be done before lsm is freed below, because
1124 * cl_object still uses inode lsm.
1126 cl_inode_fini(inode);
1129 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1130 lli->lli_smd = NULL;
1137 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1138 struct md_open_data **mod)
1140 struct lustre_md md;
1141 struct ll_sb_info *sbi = ll_i2sbi(inode);
1142 struct ptlrpc_request *request = NULL;
1146 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1147 LUSTRE_OPC_ANY, NULL);
1148 if (IS_ERR(op_data))
1149 RETURN(PTR_ERR(op_data));
1151 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1154 ptlrpc_req_finished(request);
1155 if (rc == -ENOENT) {
1157 /* Unlinked special device node? Or just a race?
1158 * Pretend we done everything. */
1159 if (!S_ISREG(inode->i_mode) &&
1160 !S_ISDIR(inode->i_mode))
1161 rc = inode_setattr(inode, &op_data->op_attr);
1162 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1163 CERROR("md_setattr fails: rc = %d\n", rc);
1168 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1169 sbi->ll_md_exp, &md);
1171 ptlrpc_req_finished(request);
1175 /* We call inode_setattr to adjust timestamps.
1176 * If there is at least some data in file, we cleared ATTR_SIZE
1177 * above to avoid invoking vmtruncate, otherwise it is important
1178 * to call vmtruncate in inode_setattr to update inode->i_size
1180 rc = inode_setattr(inode, &op_data->op_attr);
1182 /* Extract epoch data if obtained. */
1183 op_data->op_handle = md.body->handle;
1184 op_data->op_ioepoch = md.body->ioepoch;
1186 ll_update_inode(inode, &md);
1187 ptlrpc_req_finished(request);
1192 /* Close IO epoch and send Size-on-MDS attribute update. */
1193 static int ll_setattr_done_writing(struct inode *inode,
1194 struct md_op_data *op_data,
1195 struct md_open_data *mod)
1197 struct ll_inode_info *lli = ll_i2info(inode);
1201 LASSERT(op_data != NULL);
1202 if (!S_ISREG(inode->i_mode))
1205 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1206 op_data->op_ioepoch, PFID(&lli->lli_fid));
1208 op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1209 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1210 if (rc == -EAGAIN) {
1211 /* MDS has instructed us to obtain Size-on-MDS attribute
1212 * from OSTs and send setattr to back to MDS. */
1213 rc = ll_sizeonmds_update(inode, &op_data->op_handle,
1214 op_data->op_ioepoch);
1216 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1222 static int ll_setattr_do_truncate(struct inode *inode, loff_t size)
1224 struct obd_capa *capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1227 rc = cl_setattr_do_truncate(inode, size, capa);
1228 ll_truncate_free_capa(capa);
1232 static int ll_setattr_ost(struct inode *inode)
1234 struct obd_capa *capa = ll_mdscapa_get(inode);
1237 rc = cl_setattr_ost(inode, capa);
1243 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1244 * object(s) determine the file size and mtime. Otherwise, the MDS will
1245 * keep these values until such a time that objects are allocated for it.
1246 * We do the MDS operations first, as it is checking permissions for us.
1247 * We don't to the MDS RPC if there is nothing that we want to store there,
1248 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1249 * going to do an RPC anyways.
1251 * If we are doing a truncate, we will send the mtime and ctime updates
1252 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1253 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1256 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1258 struct ll_inode_info *lli = ll_i2info(inode);
1259 struct lov_stripe_md *lsm = lli->lli_smd;
1260 struct md_op_data *op_data = NULL;
1261 struct md_open_data *mod = NULL;
1262 int ia_valid = attr->ia_valid;
1263 int rc = 0, rc1 = 0;
1266 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1268 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1270 if (ia_valid & ATTR_SIZE) {
1271 if (attr->ia_size > ll_file_maxbytes(inode)) {
1272 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1273 attr->ia_size, ll_file_maxbytes(inode));
1277 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1280 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1281 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1282 if (current->fsuid != inode->i_uid &&
1283 !cfs_capable(CFS_CAP_FOWNER))
1287 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1288 if (attr->ia_valid & ATTR_CTIME) {
1289 attr->ia_ctime = CURRENT_TIME;
1290 attr->ia_valid |= ATTR_CTIME_SET;
1292 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1293 attr->ia_atime = CURRENT_TIME;
1294 attr->ia_valid |= ATTR_ATIME_SET;
1296 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1297 attr->ia_mtime = CURRENT_TIME;
1298 attr->ia_valid |= ATTR_MTIME_SET;
1300 if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1301 /* To avoid stale mtime on mds, obtain it from ost and send
1303 rc = cl_glimpse_size(inode);
1307 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1308 attr->ia_mtime = inode->i_mtime;
1311 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1312 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1313 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1314 cfs_time_current_sec());
1316 /* NB: ATTR_SIZE will only be set after this point if the size
1317 * resides on the MDS, ie, this file has no objects. */
1319 attr->ia_valid &= ~ATTR_SIZE;
1321 /* We always do an MDS RPC, even if we're only changing the size;
1322 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1324 OBD_ALLOC_PTR(op_data);
1325 if (op_data == NULL)
1328 UNLOCK_INODE_MUTEX(inode);
1329 if (ia_valid & ATTR_SIZE)
1330 UP_WRITE_I_ALLOC_SEM(inode);
1331 down(&lli->lli_trunc_sem);
1332 LOCK_INODE_MUTEX(inode);
1333 if (ia_valid & ATTR_SIZE)
1334 DOWN_WRITE_I_ALLOC_SEM(inode);
1336 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1338 /* Open epoch for truncate. */
1339 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1340 (ia_valid & ATTR_SIZE))
1341 op_data->op_flags = MF_EPOCH_OPEN;
1343 rc = ll_md_setattr(inode, op_data, &mod);
1347 ll_ioepoch_open(lli, op_data->op_ioepoch);
1348 if (!lsm || !S_ISREG(inode->i_mode)) {
1349 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1353 if (ia_valid & ATTR_SIZE)
1354 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1355 else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1356 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1357 inode->i_ino, LTIME_S(attr->ia_mtime));
1358 rc = ll_setattr_ost(inode);
1363 if (op_data->op_ioepoch)
1364 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1365 ll_finish_md_op_data(op_data);
1367 up(&lli->lli_trunc_sem);
1368 return rc ? rc : rc1;
1371 int ll_setattr(struct dentry *de, struct iattr *attr)
1373 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1374 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1375 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1377 if ((de->d_inode->i_mode & S_ISUID) &&
1378 !(attr->ia_mode & S_ISUID) &&
1379 !(attr->ia_valid & ATTR_KILL_SUID))
1380 attr->ia_valid |= ATTR_KILL_SUID;
1382 if (((de->d_inode->i_mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1383 !(attr->ia_mode & S_ISGID) &&
1384 !(attr->ia_valid & ATTR_KILL_SGID))
1385 attr->ia_valid |= ATTR_KILL_SGID;
1387 return ll_setattr_raw(de->d_inode, attr);
1390 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1391 __u64 max_age, __u32 flags)
1393 struct ll_sb_info *sbi = ll_s2sbi(sb);
1394 struct obd_statfs obd_osfs;
1398 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1400 CERROR("md_statfs fails: rc = %d\n", rc);
1404 osfs->os_type = sb->s_magic;
1406 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1407 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1409 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1410 flags |= OBD_STATFS_NODELAY;
1412 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1413 &obd_osfs, max_age, flags);
1415 CERROR("obd_statfs fails: rc = %d\n", rc);
1419 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1420 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1423 osfs->os_bsize = obd_osfs.os_bsize;
1424 osfs->os_blocks = obd_osfs.os_blocks;
1425 osfs->os_bfree = obd_osfs.os_bfree;
1426 osfs->os_bavail = obd_osfs.os_bavail;
1428 /* If we don't have as many objects free on the OST as inodes
1429 * on the MDS, we reduce the total number of inodes to
1430 * compensate, so that the "inodes in use" number is correct.
1432 if (obd_osfs.os_ffree < osfs->os_ffree) {
1433 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1435 osfs->os_ffree = obd_osfs.os_ffree;
1440 #ifndef HAVE_STATFS_DENTRY_PARAM
1441 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1444 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1446 struct super_block *sb = de->d_sb;
1448 struct obd_statfs osfs;
1451 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1452 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1454 /* For now we will always get up-to-date statfs values, but in the
1455 * future we may allow some amount of caching on the client (e.g.
1456 * from QOS or lprocfs updates). */
1457 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1461 statfs_unpack(sfs, &osfs);
1463 /* We need to downshift for all 32-bit kernels, because we can't
1464 * tell if the kernel is being called via sys_statfs64() or not.
1465 * Stop before overflowing f_bsize - in which case it is better
1466 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1467 if (sizeof(long) < 8) {
1468 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1471 osfs.os_blocks >>= 1;
1472 osfs.os_bfree >>= 1;
1473 osfs.os_bavail >>= 1;
1477 sfs->f_blocks = osfs.os_blocks;
1478 sfs->f_bfree = osfs.os_bfree;
1479 sfs->f_bavail = osfs.os_bavail;
1484 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1486 struct ll_inode_info *lli;
1487 struct lov_stripe_md *lsm;
1489 lli = ll_i2info(inode);
1490 LASSERT(lli->lli_size_sem_owner != current);
1491 down(&lli->lli_size_sem);
1492 LASSERT(lli->lli_size_sem_owner == NULL);
1493 lli->lli_size_sem_owner = current;
1495 LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1498 lov_stripe_lock(lsm);
1501 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1503 struct ll_inode_info *lli;
1504 struct lov_stripe_md *lsm;
1506 lli = ll_i2info(inode);
1508 LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1511 lov_stripe_unlock(lsm);
1512 LASSERT(lli->lli_size_sem_owner == current);
1513 lli->lli_size_sem_owner = NULL;
1514 up(&lli->lli_size_sem);
1517 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1519 struct ll_inode_info *lli = ll_i2info(inode);
1520 struct mdt_body *body = md->body;
1521 struct lov_stripe_md *lsm = md->lsm;
1522 struct ll_sb_info *sbi = ll_i2sbi(inode);
1524 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1526 if (lli->lli_smd == NULL) {
1527 if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1528 lsm->lsm_magic != LOV_MAGIC_V3) {
1529 dump_lsm(D_ERROR, lsm);
1532 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1533 lsm, inode->i_ino, inode->i_generation, inode);
1534 cl_inode_init(inode, md);
1535 /* ll_inode_size_lock() requires it is only
1536 * called with lli_smd != NULL or lock_lsm == 0
1537 * or we can race between lock/unlock.
1540 lli->lli_maxbytes = lsm->lsm_maxbytes;
1541 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1542 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1544 LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1545 lli->lli_smd->lsm_stripe_count ==
1546 lsm->lsm_stripe_count);
1547 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1548 CERROR("lsm mismatch for inode %ld\n",
1550 CERROR("lli_smd:\n");
1551 dump_lsm(D_ERROR, lli->lli_smd);
1553 dump_lsm(D_ERROR, lsm);
1557 if (lli->lli_smd != lsm)
1558 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1561 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1562 if (body->valid & OBD_MD_FLRMTPERM)
1563 ll_update_remote_perm(inode, md->remote_perm);
1565 #ifdef CONFIG_FS_POSIX_ACL
1566 else if (body->valid & OBD_MD_FLACL) {
1567 spin_lock(&lli->lli_lock);
1568 if (lli->lli_posix_acl)
1569 posix_acl_release(lli->lli_posix_acl);
1570 lli->lli_posix_acl = md->posix_acl;
1571 spin_unlock(&lli->lli_lock);
1574 inode->i_ino = cl_fid_build_ino(&body->fid1);
1575 inode->i_generation = cl_fid_build_gen(&body->fid1);
1577 if (body->valid & OBD_MD_FLATIME &&
1578 body->atime > LTIME_S(inode->i_atime))
1579 LTIME_S(inode->i_atime) = body->atime;
1581 /* mtime is always updated with ctime, but can be set in past.
1582 As write and utime(2) may happen within 1 second, and utime's
1583 mtime has a priority over write's one, so take mtime from mds
1584 for the same ctimes. */
1585 if (body->valid & OBD_MD_FLCTIME &&
1586 body->ctime >= LTIME_S(inode->i_ctime)) {
1587 LTIME_S(inode->i_ctime) = body->ctime;
1588 if (body->valid & OBD_MD_FLMTIME) {
1589 CDEBUG(D_INODE, "setting ino %lu mtime "
1590 "from %lu to "LPU64"\n", inode->i_ino,
1591 LTIME_S(inode->i_mtime), body->mtime);
1592 LTIME_S(inode->i_mtime) = body->mtime;
1595 if (body->valid & OBD_MD_FLMODE)
1596 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1597 if (body->valid & OBD_MD_FLTYPE)
1598 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1599 LASSERT(inode->i_mode != 0);
1600 if (S_ISREG(inode->i_mode)) {
1601 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1603 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1605 #ifdef HAVE_INODE_BLKSIZE
1606 inode->i_blksize = 1<<inode->i_blkbits;
1608 if (body->valid & OBD_MD_FLUID)
1609 inode->i_uid = body->uid;
1610 if (body->valid & OBD_MD_FLGID)
1611 inode->i_gid = body->gid;
1612 if (body->valid & OBD_MD_FLFLAGS)
1613 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1614 if (body->valid & OBD_MD_FLNLINK)
1615 inode->i_nlink = body->nlink;
1616 if (body->valid & OBD_MD_FLRDEV)
1617 inode->i_rdev = old_decode_dev(body->rdev);
1619 if (body->valid & OBD_MD_FLID) {
1620 /* FID shouldn't be changed! */
1621 if (fid_is_sane(&lli->lli_fid)) {
1622 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1623 "Trying to change FID "DFID
1624 " to the "DFID", inode %lu/%u(%p)\n",
1625 PFID(&lli->lli_fid), PFID(&body->fid1),
1626 inode->i_ino, inode->i_generation, inode);
1628 lli->lli_fid = body->fid1;
1631 LASSERT(fid_seq(&lli->lli_fid) != 0);
1633 if (body->valid & OBD_MD_FLSIZE) {
1634 if (exp_connect_som(ll_i2mdexp(inode)) &&
1635 S_ISREG(inode->i_mode) && lli->lli_smd) {
1636 struct lustre_handle lockh;
1639 /* As it is possible a blocking ast has been processed
1640 * by this time, we need to check there is an UPDATE
1641 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1643 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1646 if (lli->lli_flags & (LLIF_DONE_WRITING |
1647 LLIF_EPOCH_PENDING |
1649 CERROR("ino %lu flags %lu still has "
1650 "size authority! do not trust "
1651 "the size got from MDS\n",
1652 inode->i_ino, lli->lli_flags);
1654 /* Use old size assignment to avoid
1655 * deadlock bz14138 & bz14326 */
1656 inode->i_size = body->size;
1657 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1659 ldlm_lock_decref(&lockh, mode);
1662 /* Use old size assignment to avoid
1663 * deadlock bz14138 & bz14326 */
1664 inode->i_size = body->size;
1667 if (body->valid & OBD_MD_FLBLOCKS)
1668 inode->i_blocks = body->blocks;
1671 if (body->valid & OBD_MD_FLMDSCAPA) {
1672 LASSERT(md->mds_capa);
1673 ll_add_capa(inode, md->mds_capa);
1675 if (body->valid & OBD_MD_FLOSSCAPA) {
1676 LASSERT(md->oss_capa);
1677 ll_add_capa(inode, md->oss_capa);
1681 static struct backing_dev_info ll_backing_dev_info = {
1682 .ra_pages = 0, /* No readahead */
1683 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1684 .capabilities = 0, /* Does contribute to dirty memory */
1686 .memory_backed = 0, /* Does contribute to dirty memory */
1690 void ll_read_inode2(struct inode *inode, void *opaque)
1692 struct lustre_md *md = opaque;
1693 struct ll_inode_info *lli = ll_i2info(inode);
1696 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1697 inode->i_ino, inode->i_generation, inode);
1701 LASSERT(!lli->lli_smd);
1703 /* Core attributes from the MDS first. This is a new inode, and
1704 * the VFS doesn't zero times in the core inode so we have to do
1705 * it ourselves. They will be overwritten by either MDS or OST
1706 * attributes - we just need to make sure they aren't newer. */
1707 LTIME_S(inode->i_mtime) = 0;
1708 LTIME_S(inode->i_atime) = 0;
1709 LTIME_S(inode->i_ctime) = 0;
1711 ll_update_inode(inode, md);
1713 /* OIDEBUG(inode); */
1715 if (S_ISREG(inode->i_mode)) {
1716 struct ll_sb_info *sbi = ll_i2sbi(inode);
1717 inode->i_op = &ll_file_inode_operations;
1718 inode->i_fop = sbi->ll_fop;
1719 inode->i_mapping->a_ops = &ll_aops;
1721 } else if (S_ISDIR(inode->i_mode)) {
1722 inode->i_op = &ll_dir_inode_operations;
1723 inode->i_fop = &ll_dir_operations;
1724 inode->i_mapping->a_ops = &ll_dir_aops;
1726 } else if (S_ISLNK(inode->i_mode)) {
1727 inode->i_op = &ll_fast_symlink_inode_operations;
1730 inode->i_op = &ll_special_inode_operations;
1732 init_special_inode(inode, inode->i_mode,
1733 kdev_t_to_nr(inode->i_rdev));
1735 /* initializing backing dev info. */
1736 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1742 void ll_delete_inode(struct inode *inode)
1744 struct ll_sb_info *sbi = ll_i2sbi(inode);
1748 rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1750 CERROR("fid_delete() failed, rc %d\n", rc);
1752 truncate_inode_pages(&inode->i_data, 0);
1758 int ll_iocontrol(struct inode *inode, struct file *file,
1759 unsigned int cmd, unsigned long arg)
1761 struct ll_sb_info *sbi = ll_i2sbi(inode);
1762 struct ptlrpc_request *req = NULL;
1767 case FSFILT_IOC_GETFLAGS: {
1768 struct mdt_body *body;
1769 struct obd_capa *oc;
1771 oc = ll_mdscapa_get(inode);
1772 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1773 OBD_MD_FLFLAGS, 0, &req);
1776 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1780 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1782 flags = body->flags;
1784 ptlrpc_req_finished(req);
1786 RETURN(put_user(flags, (int *)arg));
1788 case FSFILT_IOC_SETFLAGS: {
1789 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1790 struct obd_info oinfo = { { { 0 } } };
1791 struct md_op_data *op_data;
1793 if (get_user(flags, (int *)arg))
1797 OBDO_ALLOC(oinfo.oi_oa);
1801 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1802 LUSTRE_OPC_ANY, NULL);
1803 if (IS_ERR(op_data))
1804 RETURN(PTR_ERR(op_data));
1806 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1807 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1808 rc = md_setattr(sbi->ll_md_exp, op_data,
1809 NULL, 0, NULL, 0, &req, NULL);
1810 ll_finish_md_op_data(op_data);
1811 ptlrpc_req_finished(req);
1813 OBDO_FREE(oinfo.oi_oa);
1818 OBDO_FREE(oinfo.oi_oa);
1819 GOTO(update_cache, rc);
1822 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1823 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1824 oinfo.oi_oa->o_flags = flags;
1825 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1827 oinfo.oi_capa = ll_mdscapa_get(inode);
1829 obdo_from_inode(oinfo.oi_oa, inode,
1830 OBD_MD_FLFID | OBD_MD_FLGENER);
1831 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1832 capa_put(oinfo.oi_capa);
1833 OBDO_FREE(oinfo.oi_oa);
1835 if (rc != -EPERM && rc != -EACCES)
1836 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1842 inode->i_flags = ll_ext_to_inode_flags(flags |
1843 MDS_BFLAG_EXT_FLAGS);
1853 int ll_flush_ctx(struct inode *inode)
1855 struct ll_sb_info *sbi = ll_i2sbi(inode);
1857 CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
1859 obd_set_info_async(sbi->ll_md_exp,
1860 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1862 obd_set_info_async(sbi->ll_dt_exp,
1863 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1868 /* umount -f client means force down, don't save state */
1869 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1870 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1872 struct super_block *sb = vfsmnt->mnt_sb;
1874 void ll_umount_begin(struct super_block *sb)
1877 struct lustre_sb_info *lsi = s2lsi(sb);
1878 struct ll_sb_info *sbi = ll_s2sbi(sb);
1879 struct obd_device *obd;
1880 struct obd_ioctl_data ioc_data = { 0 };
1883 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1884 if (!(flags & MNT_FORCE)) {
1890 /* Tell the MGC we got umount -f */
1891 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1893 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1894 sb->s_count, atomic_read(&sb->s_active));
1896 obd = class_exp2obd(sbi->ll_md_exp);
1898 CERROR("Invalid MDC connection handle "LPX64"\n",
1899 sbi->ll_md_exp->exp_handle.h_cookie);
1904 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
1907 obd = class_exp2obd(sbi->ll_dt_exp);
1909 CERROR("Invalid LOV connection handle "LPX64"\n",
1910 sbi->ll_dt_exp->exp_handle.h_cookie);
1916 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
1919 /* Really, we'd like to wait until there are no requests outstanding,
1920 * and then continue. For now, we just invalidate the requests,
1921 * schedule() and sleep one second if needed, and hope.
1924 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1925 if (atomic_read(&vfsmnt->mnt_count) > 2) {
1926 cfs_schedule_timeout(CFS_TASK_INTERRUPTIBLE,
1927 cfs_time_seconds(1));
1928 if (atomic_read(&vfsmnt->mnt_count) > 2)
1929 LCONSOLE_WARN("Mount still busy with %d refs! You "
1930 "may try to umount it a bit later\n",
1931 atomic_read(&vfsmnt->mnt_count));
1938 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1940 struct ll_sb_info *sbi = ll_s2sbi(sb);
1944 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1945 read_only = *flags & MS_RDONLY;
1946 err = obd_set_info_async(sbi->ll_md_exp,
1947 sizeof(KEY_READ_ONLY),
1948 KEY_READ_ONLY, sizeof(read_only),
1951 CERROR("Failed to change the read-only flag during "
1952 "remount: %d\n", err);
1957 sb->s_flags |= MS_RDONLY;
1959 sb->s_flags &= ~MS_RDONLY;
1964 int ll_prep_inode(struct inode **inode,
1965 struct ptlrpc_request *req,
1966 struct super_block *sb)
1968 struct ll_sb_info *sbi = NULL;
1969 struct lustre_md md;
1973 LASSERT(*inode || sb);
1974 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
1975 prune_deathrow(sbi, 1);
1976 memset(&md, 0, sizeof(struct lustre_md));
1978 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
1979 sbi->ll_md_exp, &md);
1984 ll_update_inode(*inode, &md);
1986 LASSERT(sb != NULL);
1989 * At this point server returns to client's same fid as client
1990 * generated for creating. So using ->fid1 is okay here.
1992 LASSERT(fid_is_sane(&md.body->fid1));
1994 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1), &md);
1995 if (*inode == NULL || IS_ERR(*inode)) {
1997 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
1998 #ifdef CONFIG_FS_POSIX_ACL
2000 posix_acl_release(md.posix_acl);
2001 md.posix_acl = NULL;
2004 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2006 CERROR("new_inode -fatal: rc %d\n", rc);
2012 md_free_lustre_md(sbi->ll_md_exp, &md);
2016 int ll_obd_statfs(struct inode *inode, void *arg)
2018 struct ll_sb_info *sbi = NULL;
2019 struct obd_export *exp;
2021 struct obd_ioctl_data *data = NULL;
2025 if (!inode || !(sbi = ll_i2sbi(inode)))
2026 GOTO(out_statfs, rc = -EINVAL);
2028 rc = obd_ioctl_getdata(&buf, &len, arg);
2030 GOTO(out_statfs, rc);
2033 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2034 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2035 GOTO(out_statfs, rc = -EINVAL);
2037 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2038 if (type == LL_STATFS_MDC)
2039 exp = sbi->ll_md_exp;
2040 else if (type == LL_STATFS_LOV)
2041 exp = sbi->ll_dt_exp;
2043 GOTO(out_statfs, rc = -ENODEV);
2045 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2047 GOTO(out_statfs, rc);
2050 obd_ioctl_freedata(buf, len);
2054 int ll_process_config(struct lustre_cfg *lcfg)
2058 struct lprocfs_static_vars lvars;
2062 lprocfs_llite_init_vars(&lvars);
2064 /* The instance name contains the sb: lustre-client-aacfe000 */
2065 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2066 if (!ptr || !*(++ptr))
2068 if (sscanf(ptr, "%lx", &x) != 1)
2071 /* This better be a real Lustre superblock! */
2072 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2074 /* Note we have not called client_common_fill_super yet, so
2075 proc fns must be able to handle that! */
2076 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2083 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2084 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2085 struct inode *i1, struct inode *i2,
2086 const char *name, int namelen,
2087 int mode, __u32 opc, void *data)
2089 LASSERT(i1 != NULL);
2091 if (namelen > ll_i2sbi(i1)->ll_namelen)
2092 return ERR_PTR(-ENAMETOOLONG);
2094 if (op_data == NULL)
2095 OBD_ALLOC_PTR(op_data);
2097 if (op_data == NULL)
2098 return ERR_PTR(-ENOMEM);
2100 ll_i2gids(op_data->op_suppgids, i1, i2);
2101 op_data->op_fid1 = *ll_inode2fid(i1);
2102 op_data->op_capa1 = ll_mdscapa_get(i1);
2105 op_data->op_fid2 = *ll_inode2fid(i2);
2106 op_data->op_capa2 = ll_mdscapa_get(i2);
2108 fid_zero(&op_data->op_fid2);
2109 op_data->op_capa2 = NULL;
2112 op_data->op_name = name;
2113 op_data->op_namelen = namelen;
2114 op_data->op_mode = mode;
2115 op_data->op_mod_time = cfs_time_current_sec();
2116 op_data->op_fsuid = current->fsuid;
2117 op_data->op_fsgid = current->fsgid;
2118 op_data->op_cap = cfs_curproc_cap_pack();
2119 op_data->op_bias = MDS_CHECK_SPLIT;
2120 op_data->op_opc = opc;
2121 op_data->op_mds = 0;
2122 op_data->op_data = data;
2127 void ll_finish_md_op_data(struct md_op_data *op_data)
2129 capa_put(op_data->op_capa1);
2130 capa_put(op_data->op_capa2);
2131 OBD_FREE_PTR(op_data);
2134 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2136 struct ll_sb_info *sbi;
2138 LASSERT((seq != NULL) && (vfs != NULL));
2139 sbi = ll_s2sbi(vfs->mnt_sb);
2141 if (sbi->ll_flags & LL_SBI_NOLCK)
2142 seq_puts(seq, ",nolock");
2144 if (sbi->ll_flags & LL_SBI_FLOCK)
2145 seq_puts(seq, ",flock");
2147 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2148 seq_puts(seq, ",localflock");
2150 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2151 seq_puts(seq, ",user_xattr");
2153 if (sbi->ll_flags & LL_SBI_ACL)
2154 seq_puts(seq, ",acl");
2156 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2157 seq_puts(seq, ",lazystatfs");