4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 * Directory code for lustre client.
42 #include <linux/pagemap.h>
44 #include <linux/version.h>
45 #include <asm/uaccess.h>
46 #include <linux/buffer_head.h> // for wait_on_buffer
47 #include <linux/pagevec.h>
49 #define DEBUG_SUBSYSTEM S_LLITE
51 #include <lustre/lustre_idl.h>
52 #include <obd_support.h>
53 #include <obd_class.h>
54 #include <lustre_ioctl.h>
55 #include <lustre_lib.h>
56 #include <lustre_lite.h>
57 #include <lustre_dlm.h>
58 #include <lustre_fid.h>
59 #include "llite_internal.h"
62 * (new) readdir implementation overview.
64 * Original lustre readdir implementation cached exact copy of raw directory
65 * pages on the client. These pages were indexed in client page cache by
66 * logical offset in the directory file. This design, while very simple and
67 * intuitive had some inherent problems:
69 * . it implies that byte offset to the directory entry serves as a
70 * telldir(3)/seekdir(3) cookie, but that offset is not stable: in
71 * ext3/htree directory entries may move due to splits, and more
74 * . it is incompatible with the design of split directories for cmd3,
75 * that assumes that names are distributed across nodes based on their
76 * hash, and so readdir should be done in hash order.
78 * New readdir implementation does readdir in hash order, and uses hash of a
79 * file name as a telldir/seekdir cookie. This led to number of complications:
81 * . hash is not unique, so it cannot be used to index cached directory
82 * pages on the client (note, that it requires a whole pageful of hash
83 * collided entries to cause two pages to have identical hashes);
85 * . hash is not unique, so it cannot, strictly speaking, be used as an
86 * entry cookie. ext3/htree has the same problem and lustre implementation
87 * mimics their solution: seekdir(hash) positions directory at the first
88 * entry with the given hash.
94 * Client caches directory pages using hash of the first entry as an index. As
95 * noted above hash is not unique, so this solution doesn't work as is:
96 * special processing is needed for "page hash chains" (i.e., sequences of
97 * pages filled with entries all having the same hash value).
99 * First, such chains have to be detected. To this end, server returns to the
100 * client the hash of the first entry on the page next to one returned. When
101 * client detects that this hash is the same as hash of the first entry on the
102 * returned page, page hash collision has to be handled. Pages in the
103 * hash chain, except first one, are termed "overflow pages".
105 * Solution to index uniqueness problem is to not cache overflow
106 * pages. Instead, when page hash collision is detected, all overflow pages
107 * from emerging chain are immediately requested from the server and placed in
108 * a special data structure (struct ll_dir_chain). This data structure is used
109 * by ll_readdir() to process entries from overflow pages. When readdir
110 * invocation finishes, overflow pages are discarded. If page hash collision
111 * chain weren't completely processed, next call to readdir will again detect
112 * page hash collision, again read overflow pages in, process next portion of
113 * entries and again discard the pages. This is not as wasteful as it looks,
114 * because, given reasonable hash, page hash collisions are extremely rare.
116 * 1. directory positioning
118 * When seekdir(hash) is called, original
129 * identification of and access to overflow pages
133 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
134 * a header lu_dirpage which describes the start/end hash, and whether this
135 * page is empty (contains no dir entry) or hash collide with next page.
136 * After client receives reply, several pages will be integrated into dir page
137 * in PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the
138 * lu_dirpage for this integrated page will be adjusted. See
139 * lmv_adjust_dirpages().
143 * The following three APIs will be used by llite to iterate directory
144 * entries from MDC dir page caches.
146 * ll_dir_entry_start(next) will lookup(return) entry by op_hash_offset.
147 * To avoid extra memory allocation, the @entry will be pointed to
148 * the dir entries in MDC page directly, so these pages can not be released
149 * until the entry has been accessed in ll_readdir(or statahead).
151 * The iterate process will be
153 * ll_dir_entry_start: locate the page in MDC, and return the first entry.
156 * ll_dir_entry_next: return the next entry in the current page, if it reaches
157 * to the end, release current page.
159 * ll_dir_entry_end: release the last page.
161 struct lu_dirent *ll_dir_entry_start(struct inode *dir,
162 struct md_op_data *op_data,
165 struct lu_dirent *entry = NULL;
166 struct md_callback cb_op;
170 LASSERT(*ppage == NULL);
171 cb_op.md_blocking_ast = ll_md_blocking_ast;
172 op_data->op_cli_flags &= ~CLI_NEXT_ENTRY;
173 rc = md_read_entry(ll_i2mdexp(dir), op_data, &cb_op, &entry, ppage);
179 struct lu_dirent *ll_dir_entry_next(struct inode *dir,
180 struct md_op_data *op_data,
181 struct lu_dirent *ent,
184 struct lu_dirent *entry = NULL;
185 struct md_callback cb_op;
189 op_data->op_hash_offset = le64_to_cpu(ent->lde_hash);
191 /* release last page */
192 LASSERT(*ppage != NULL);
194 page_cache_release(*ppage);
196 cb_op.md_blocking_ast = ll_md_blocking_ast;
197 op_data->op_cli_flags |= CLI_NEXT_ENTRY;
198 rc = md_read_entry(ll_i2mdexp(dir), op_data, &cb_op, &entry, ppage);
205 #ifdef HAVE_DIR_CONTEXT
206 int ll_dir_read(struct inode *inode, struct md_op_data *op_data,
207 struct dir_context *ctx)
210 int ll_dir_read(struct inode *inode, struct md_op_data *op_data,
211 void *cookie, filldir_t filldir)
214 struct ll_sb_info *sbi = ll_i2sbi(inode);
215 struct ll_dir_chain chain;
216 struct lu_dirent *ent;
217 int api32 = ll_need_32bit_api(sbi);
218 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
221 __u64 hash = MDS_DIR_END_OFF;
222 struct page *page = NULL;
225 ll_dir_chain_init(&chain);
226 for (ent = ll_dir_entry_start(inode, op_data, &page);
227 ent != NULL && !IS_ERR(ent) && !done;
228 ent = ll_dir_entry_next(inode, op_data, ent, &page)) {
235 hash = le64_to_cpu(ent->lde_hash);
236 if (hash < op_data->op_hash_offset)
238 * Skip until we find target hash
242 namelen = le16_to_cpu(ent->lde_namelen);
253 fid_le_to_cpu(&fid, &ent->lde_fid);
254 ino = cl_fid_build_ino(&fid, api32);
255 type = ll_dirent_type_get(ent);
257 #ifdef HAVE_DIR_CONTEXT
258 /* For 'll_nfs_get_name_filldir()', it will try
259 * to access the 'ent' through its 'lde_name',
260 * so the parameter 'name' for 'filldir()' must
261 * be part of the 'ent'. */
262 done = !dir_emit(ctx, ent->lde_name, namelen, ino, type);
264 done = filldir(cookie, ent->lde_name, namelen, lhash,
268 if (op_data->op_hash_offset != MDS_DIR_END_OFF)
269 op_data->op_hash_offset = hash;
276 else if (ent == NULL)
277 op_data->op_hash_offset = MDS_DIR_END_OFF;
281 page_cache_release(page);
284 ll_dir_chain_fini(&chain);
288 #ifdef HAVE_DIR_CONTEXT
289 static int ll_iterate(struct file *filp, struct dir_context *ctx)
291 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
294 struct inode *inode = filp->f_dentry->d_inode;
295 struct ll_file_data *lfd = LUSTRE_FPRIVATE(filp);
296 struct ll_sb_info *sbi = ll_i2sbi(inode);
297 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
298 int api32 = ll_need_32bit_api(sbi);
299 struct md_op_data *op_data;
309 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
310 "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
311 inode, (unsigned long)pos, i_size_read(inode), api32);
313 if (pos == MDS_DIR_END_OFF)
319 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
320 LUSTRE_OPC_ANY, inode);
322 GOTO(out, rc = PTR_ERR(op_data));
324 if (unlikely(op_data->op_mea1 != NULL)) {
325 /* This is only needed for striped dir to fill ..,
326 * see lmv_read_entry */
327 if (filp->f_dentry->d_parent != NULL &&
328 filp->f_dentry->d_parent->d_inode != NULL) {
329 __u64 ibits = MDS_INODELOCK_UPDATE;
330 struct inode *parent =
331 filp->f_dentry->d_parent->d_inode;
333 if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
334 op_data->op_fid3 = *ll_inode2fid(parent);
337 /* If it can not find in cache, do lookup .. on the master
339 if (fid_is_zero(&op_data->op_fid3)) {
340 rc = ll_dir_get_parent_fid(inode, &op_data->op_fid3);
345 op_data->op_hash_offset = pos;
346 op_data->op_max_pages = sbi->ll_md_brw_pages;
347 #ifdef HAVE_DIR_CONTEXT
349 rc = ll_dir_read(inode, op_data, ctx);
352 rc = ll_dir_read(inode, op_data, cookie, filldir);
355 lfd->lfd_pos = op_data->op_hash_offset;
357 if (pos == MDS_DIR_END_OFF) {
359 pos = LL_DIR_END_OFF_32BIT;
361 pos = LL_DIR_END_OFF;
364 pos = op_data->op_hash_offset >> 32;
366 pos = op_data->op_hash_offset;
368 #ifdef HAVE_DIR_CONTEXT
373 ll_finish_md_op_data(op_data);
374 filp->f_version = inode->i_version;
378 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
383 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
385 struct mgs_send_param *msp;
392 strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
393 rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
394 sizeof(struct mgs_send_param), msp, NULL);
396 CERROR("Failed to set parameter: %d\n", rc);
402 static int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump,
403 const char *filename)
405 struct ptlrpc_request *request = NULL;
406 struct md_op_data *op_data;
407 struct ll_sb_info *sbi = ll_i2sbi(dir);
412 if (unlikely(lump->lum_magic != LMV_USER_MAGIC))
415 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s"
416 "stripe_offset %d, stripe_count: %u\n",
417 PFID(ll_inode2fid(dir)), dir, filename,
418 (int)lump->lum_stripe_offset, lump->lum_stripe_count);
420 if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC))
421 lustre_swab_lmv_user_md(lump);
423 mode = (0755 & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
424 op_data = ll_prep_md_op_data(NULL, dir, NULL, filename,
425 strlen(filename), mode, LUSTRE_OPC_MKDIR,
428 GOTO(err_exit, err = PTR_ERR(op_data));
430 op_data->op_cli_flags |= CLI_SET_MEA;
431 err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
432 from_kuid(&init_user_ns, current_fsuid()),
433 from_kgid(&init_user_ns, current_fsgid()),
434 cfs_curproc_cap_pack(), 0, &request);
435 ll_finish_md_op_data(op_data);
439 ptlrpc_req_finished(request);
443 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
446 struct ll_sb_info *sbi = ll_i2sbi(inode);
447 struct md_op_data *op_data;
448 struct ptlrpc_request *req = NULL;
450 struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
451 struct obd_device *mgc = lsi->lsi_mgc;
457 * This is coming from userspace, so should be in
458 * local endian. But the MDS would like it in little
459 * endian, so we swab it before we send it.
461 switch (lump->lmm_magic) {
462 case LOV_USER_MAGIC_V1: {
463 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
464 lustre_swab_lov_user_md_v1(lump);
465 lum_size = sizeof(struct lov_user_md_v1);
468 case LOV_USER_MAGIC_V3: {
469 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
470 lustre_swab_lov_user_md_v3(
471 (struct lov_user_md_v3 *)lump);
472 lum_size = sizeof(struct lov_user_md_v3);
475 case LMV_USER_MAGIC: {
476 if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
477 lustre_swab_lmv_user_md(
478 (struct lmv_user_md *)lump);
479 lum_size = sizeof(struct lmv_user_md);
483 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
484 " %#08x != %#08x nor %#08x\n",
485 lump->lmm_magic, LOV_USER_MAGIC_V1,
491 lum_size = sizeof(struct lov_user_md_v1);
494 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
495 LUSTRE_OPC_ANY, NULL);
497 RETURN(PTR_ERR(op_data));
499 /* swabbing is done in lov_setstripe() on server side */
500 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
501 NULL, 0, &req, NULL);
502 ll_finish_md_op_data(op_data);
503 ptlrpc_req_finished(req);
505 if (rc != -EPERM && rc != -EACCES)
506 CERROR("mdc_setattr fails: rc = %d\n", rc);
509 /* In the following we use the fact that LOV_USER_MAGIC_V1 and
510 LOV_USER_MAGIC_V3 have the same initial fields so we do not
511 need the make the distiction between the 2 versions */
512 if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
516 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
518 GOTO(end, rc = -ENOMEM);
521 /* Get fsname and assume devname to be -MDT0000. */
522 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
523 strcat(buf, "-MDT0000.lov");
526 /* Set root stripesize */
527 sprintf(buf, ".stripesize=%u",
528 lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
529 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
533 /* Set root stripecount */
534 sprintf(buf, ".stripecount=%hd",
535 lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
536 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
540 /* Set root stripeoffset */
541 sprintf(buf, ".stripeoffset=%hd",
542 lump ? le16_to_cpu(lump->lmm_stripe_offset) :
543 (typeof(lump->lmm_stripe_offset))(-1));
544 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
548 OBD_FREE(param, MGS_PARAM_MAXLEN);
554 * This function will be used to get default LOV/LMV/Default LMV
555 * @valid will be used to indicate which stripe it will retrieve
556 * OBD_MD_MEA LMV stripe EA
557 * OBD_MD_DEFAULT_MEA Default LMV stripe EA
558 * otherwise Default LOV EA.
559 * Each time, it can only retrieve 1 stripe EA
561 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
562 struct ptlrpc_request **request, obd_valid valid)
564 struct ll_sb_info *sbi = ll_i2sbi(inode);
565 struct mdt_body *body;
566 struct lov_mds_md *lmm = NULL;
567 struct ptlrpc_request *req = NULL;
569 struct md_op_data *op_data;
572 rc = ll_get_default_mdsize(sbi, &lmm_size);
576 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
577 0, lmm_size, LUSTRE_OPC_ANY,
580 RETURN(PTR_ERR(op_data));
582 op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
583 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
584 ll_finish_md_op_data(op_data);
586 CDEBUG(D_INFO, "md_getattr failed on inode "
587 DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
591 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
592 LASSERT(body != NULL);
594 lmm_size = body->eadatasize;
596 if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
598 GOTO(out, rc = -ENODATA);
601 lmm = req_capsule_server_sized_get(&req->rq_pill,
602 &RMF_MDT_MD, lmm_size);
603 LASSERT(lmm != NULL);
606 * This is coming from the MDS, so is probably in
607 * little endian. We convert it to host endian before
608 * passing it to userspace.
610 /* We don't swab objects for directories */
611 switch (le32_to_cpu(lmm->lmm_magic)) {
613 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
614 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
617 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
618 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
621 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
622 lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
625 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
626 lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
629 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
634 *plmm_size = lmm_size;
639 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
641 struct md_op_data *op_data;
646 OBD_ALLOC_PTR(op_data);
650 op_data->op_flags |= MF_GET_MDT_IDX;
651 op_data->op_fid1 = *fid;
652 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
653 mdt_index = op_data->op_mds;
654 OBD_FREE_PTR(op_data);
662 * Get MDT index for the inode.
664 int ll_get_mdt_idx(struct inode *inode)
666 return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
670 * Generic handler to do any pre-copy work.
672 * It send a first hsm_progress (with extent length == 0) to coordinator as a
673 * first information for it that real work has started.
675 * Moreover, for a ARCHIVE request, it will sample the file data version and
676 * store it in \a copy.
678 * \return 0 on success.
680 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
682 struct ll_sb_info *sbi = ll_s2sbi(sb);
683 struct hsm_progress_kernel hpk;
687 /* Forge a hsm_progress based on data from copy. */
688 hpk.hpk_fid = copy->hc_hai.hai_fid;
689 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
690 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
691 hpk.hpk_extent.length = 0;
694 hpk.hpk_data_version = 0;
697 /* For archive request, we need to read the current file version. */
698 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
700 __u64 data_version = 0;
702 /* Get inode for this fid */
703 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
705 hpk.hpk_flags |= HP_FLAG_RETRY;
706 /* hpk_errval is >= 0 */
707 hpk.hpk_errval = -PTR_ERR(inode);
708 GOTO(progress, rc = PTR_ERR(inode));
711 /* Read current file data version */
712 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
715 CDEBUG(D_HSM, "Could not read file data version of "
716 DFID" (rc = %d). Archive request ("
717 LPX64") could not be done.\n",
718 PFID(©->hc_hai.hai_fid), rc,
719 copy->hc_hai.hai_cookie);
720 hpk.hpk_flags |= HP_FLAG_RETRY;
721 /* hpk_errval must be >= 0 */
722 hpk.hpk_errval = -rc;
726 /* Store it the hsm_copy for later copytool use.
727 * Always modified even if no lsm. */
728 copy->hc_data_version = data_version;
732 /* On error, the request should be considered as completed */
733 if (hpk.hpk_errval > 0)
734 hpk.hpk_flags |= HP_FLAG_COMPLETED;
735 rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
742 * Generic handler to do any post-copy work.
744 * It will send the last hsm_progress update to coordinator to inform it
745 * that copy is finished and whether it was successful or not.
748 * - for ARCHIVE request, it will sample the file data version and compare it
749 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
750 * will be considered as failed.
751 * - for RESTORE request, it will sample the file data version and send it to
752 * coordinator which is useful if the file was imported as 'released'.
754 * \return 0 on success.
756 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
758 struct ll_sb_info *sbi = ll_s2sbi(sb);
759 struct hsm_progress_kernel hpk;
763 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
764 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
765 * initialized if copy_end was called with copy == NULL.
768 /* Forge a hsm_progress based on data from copy. */
769 hpk.hpk_fid = copy->hc_hai.hai_fid;
770 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
771 hpk.hpk_extent = copy->hc_hai.hai_extent;
772 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
773 hpk.hpk_errval = copy->hc_errval;
774 hpk.hpk_data_version = 0;
776 /* For archive request, we need to check the file data was not changed.
778 * For restore request, we need to send the file data version, this is
779 * useful when the file was created using hsm_import.
781 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
782 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
783 (copy->hc_errval == 0)) {
785 __u64 data_version = 0;
787 /* Get lsm for this fid */
788 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
790 hpk.hpk_flags |= HP_FLAG_RETRY;
791 /* hpk_errval must be >= 0 */
792 hpk.hpk_errval = -PTR_ERR(inode);
793 GOTO(progress, rc = PTR_ERR(inode));
796 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
799 CDEBUG(D_HSM, "Could not read file data version. "
800 "Request could not be confirmed.\n");
801 if (hpk.hpk_errval == 0)
802 hpk.hpk_errval = -rc;
806 /* Store it the hsm_copy for later copytool use.
807 * Always modified even if no lsm. */
808 hpk.hpk_data_version = data_version;
810 /* File could have been stripped during archiving, so we need
811 * to check anyway. */
812 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
813 (copy->hc_data_version != data_version)) {
814 CDEBUG(D_HSM, "File data version mismatched. "
815 "File content was changed during archiving. "
816 DFID", start:"LPX64" current:"LPX64"\n",
817 PFID(©->hc_hai.hai_fid),
818 copy->hc_data_version, data_version);
819 /* File was changed, send error to cdt. Do not ask for
820 * retry because if a file is modified frequently,
821 * the cdt will loop on retried archive requests.
822 * The policy engine will ask for a new archive later
823 * when the file will not be modified for some tunable
825 /* we do not notify caller */
826 hpk.hpk_flags &= ~HP_FLAG_RETRY;
827 /* hpk_errval must be >= 0 */
828 hpk.hpk_errval = EBUSY;
834 rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
841 static int copy_and_ioctl(int cmd, struct obd_export *exp,
842 const void __user *data, size_t size)
847 OBD_ALLOC(copy, size);
851 if (copy_from_user(copy, data, size)) {
856 rc = obd_iocontrol(cmd, exp, size, copy, NULL);
858 OBD_FREE(copy, size);
863 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
865 int cmd = qctl->qc_cmd;
866 int type = qctl->qc_type;
867 int id = qctl->qc_id;
868 int valid = qctl->qc_valid;
873 case LUSTRE_Q_INVALIDATE:
874 case LUSTRE_Q_FINVALIDATE:
879 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
880 sbi->ll_flags & LL_SBI_RMT_CLIENT)
884 if (((type == USRQUOTA &&
885 !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
887 !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
888 (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
889 sbi->ll_flags & LL_SBI_RMT_CLIENT))
895 CERROR("unsupported quotactl op: %#x\n", cmd);
899 if (valid != QC_GENERAL) {
900 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
903 if (cmd == Q_GETINFO)
904 qctl->qc_cmd = Q_GETOINFO;
905 else if (cmd == Q_GETQUOTA)
906 qctl->qc_cmd = Q_GETOQUOTA;
912 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
913 sizeof(*qctl), qctl, NULL);
916 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
917 sizeof(*qctl), qctl, NULL);
920 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
921 sizeof(*qctl), qctl, NULL);
923 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
925 sizeof(*qctl), qctl, NULL);
937 struct obd_quotactl *oqctl;
939 OBD_ALLOC_PTR(oqctl);
943 QCTL_COPY(oqctl, qctl);
944 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
946 if (rc != -EALREADY && cmd == Q_QUOTAON) {
947 oqctl->qc_cmd = Q_QUOTAOFF;
948 obd_quotactl(sbi->ll_md_exp, oqctl);
953 /* If QIF_SPACE is not set, client should collect the
954 * space usage from OSSs by itself */
955 if (cmd == Q_GETQUOTA &&
956 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
957 !oqctl->qc_dqblk.dqb_curspace) {
958 struct obd_quotactl *oqctl_tmp;
960 OBD_ALLOC_PTR(oqctl_tmp);
961 if (oqctl_tmp == NULL)
962 GOTO(out, rc = -ENOMEM);
964 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
965 oqctl_tmp->qc_id = oqctl->qc_id;
966 oqctl_tmp->qc_type = oqctl->qc_type;
968 /* collect space usage from OSTs */
969 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
970 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
971 if (!rc || rc == -EREMOTEIO) {
972 oqctl->qc_dqblk.dqb_curspace =
973 oqctl_tmp->qc_dqblk.dqb_curspace;
974 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
977 /* collect space & inode usage from MDTs */
978 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
979 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
980 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
981 if (!rc || rc == -EREMOTEIO) {
982 oqctl->qc_dqblk.dqb_curspace +=
983 oqctl_tmp->qc_dqblk.dqb_curspace;
984 oqctl->qc_dqblk.dqb_curinodes =
985 oqctl_tmp->qc_dqblk.dqb_curinodes;
986 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
988 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
991 OBD_FREE_PTR(oqctl_tmp);
994 QCTL_COPY(qctl, oqctl);
1002 ll_getname(const char __user *filename)
1005 char *tmp = __getname();
1008 return ERR_PTR(-ENOMEM);
1010 len = strncpy_from_user(tmp, filename, PATH_MAX);
1013 else if (len > PATH_MAX)
1014 ret = -ENAMETOOLONG;
1023 #define ll_putname(filename) __putname(filename)
1025 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1027 struct inode *inode = file->f_dentry->d_inode;
1028 struct ll_sb_info *sbi = ll_i2sbi(inode);
1029 struct obd_ioctl_data *data;
1033 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1034 PFID(ll_inode2fid(inode)), inode, cmd);
1036 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1037 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1040 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1042 case FSFILT_IOC_GETFLAGS:
1043 case FSFILT_IOC_SETFLAGS:
1044 RETURN(ll_iocontrol(inode, file, cmd, arg));
1045 case FSFILT_IOC_GETVERSION_OLD:
1046 case FSFILT_IOC_GETVERSION:
1047 RETURN(put_user(inode->i_generation, (int *)arg));
1048 /* We need to special case any other ioctls we want to handle,
1049 * to send them to the MDS/OST as appropriate and to properly
1050 * network encode the arg field.
1051 case FSFILT_IOC_SETVERSION_OLD:
1052 case FSFILT_IOC_SETVERSION:
1054 case LL_IOC_GET_MDTIDX: {
1057 mdtidx = ll_get_mdt_idx(inode);
1061 if (put_user((int)mdtidx, (int*)arg))
1066 case IOC_MDC_LOOKUP: {
1067 int namelen, len = 0;
1071 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1076 filename = data->ioc_inlbuf1;
1077 namelen = strlen(filename);
1079 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1080 GOTO(out_free, rc = -EINVAL);
1083 rc = ll_get_fid_by_name(inode, filename, namelen, NULL);
1085 CERROR("%s: lookup %.*s failed: rc = %d\n",
1086 ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1091 obd_ioctl_freedata(buf, len);
1094 case LL_IOC_LMV_SETSTRIPE: {
1095 struct lmv_user_md *lum;
1103 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1108 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1109 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1110 GOTO(lmv_out_free, rc = -EINVAL);
1112 filename = data->ioc_inlbuf1;
1113 namelen = data->ioc_inllen1;
1116 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1117 GOTO(lmv_out_free, rc = -EINVAL);
1119 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1120 lumlen = data->ioc_inllen2;
1122 if (lum->lum_magic != LMV_USER_MAGIC ||
1123 lumlen != sizeof(*lum)) {
1124 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1125 filename, lum->lum_magic, lumlen, -EFAULT);
1126 GOTO(lmv_out_free, rc = -EINVAL);
1130 * ll_dir_setdirstripe will be used to set dir stripe
1131 * mdc_create--->mdt_reint_create (with dirstripe)
1133 rc = ll_dir_setdirstripe(inode, lum, filename);
1135 obd_ioctl_freedata(buf, len);
1139 case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1140 struct lmv_user_md lum;
1141 struct lmv_user_md __user *ulump =
1142 (struct lmv_user_md __user *)arg;
1145 if (copy_from_user(&lum, ulump, sizeof(lum)))
1148 if (lum.lum_magic != LMV_USER_MAGIC)
1151 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1155 case LL_IOC_LOV_SETSTRIPE: {
1156 struct lov_user_md_v3 lumv3;
1157 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1158 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1159 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1161 int set_default = 0;
1163 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1164 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1165 sizeof(lumv3p->lmm_objects[0]));
1166 /* first try with v1 which is smaller than v3 */
1167 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1170 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1171 if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1175 if (inode->i_sb->s_root == file->f_dentry)
1178 /* in v1 and v3 cases lumv1 points to data */
1179 rc = ll_dir_setstripe(inode, lumv1, set_default);
1183 case LL_IOC_LMV_GETSTRIPE: {
1184 struct lmv_user_md __user *ulmv =
1185 (struct lmv_user_md __user *)arg;
1186 struct lmv_user_md lum;
1187 struct ptlrpc_request *request = NULL;
1188 union lmv_mds_md *lmm = NULL;
1190 obd_valid valid = 0;
1191 struct lmv_user_md *tmp = NULL;
1198 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1201 /* lum_magic will indicate which stripe the ioctl will like
1202 * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1203 * is for default LMV stripe */
1204 if (lum.lum_magic == LMV_MAGIC_V1)
1205 valid |= OBD_MD_MEA;
1206 else if (lum.lum_magic == LMV_USER_MAGIC)
1207 valid |= OBD_MD_DEFAULT_MEA;
1211 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1214 GOTO(finish_req, rc);
1216 /* Get default LMV EA */
1217 if (lum.lum_magic == LMV_USER_MAGIC) {
1219 GOTO(finish_req, rc);
1221 if (lmmsize > sizeof(*ulmv))
1222 GOTO(finish_req, rc = -EINVAL);
1224 if (copy_to_user(ulmv, lmm, lmmsize))
1225 GOTO(finish_req, rc = -EFAULT);
1227 GOTO(finish_req, rc);
1230 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1231 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1232 OBD_ALLOC(tmp, lum_size);
1234 GOTO(finish_req, rc = -ENOMEM);
1236 mdt_index = ll_get_mdt_idx(inode);
1238 GOTO(out_tmp, rc = -ENOMEM);
1240 tmp->lum_magic = LMV_MAGIC_V1;
1241 tmp->lum_stripe_count = 0;
1242 tmp->lum_stripe_offset = mdt_index;
1243 for (i = 0; i < stripe_count; i++) {
1246 fid = &lmm->lmv_md_v1.lmv_stripe_fids[i];
1247 mdt_index = ll_get_mdt_idx_by_fid(sbi, fid);
1249 GOTO(out_tmp, rc = mdt_index);
1251 tmp->lum_objects[i].lum_mds = mdt_index;
1252 tmp->lum_objects[i].lum_fid = *fid;
1253 tmp->lum_stripe_count++;
1256 if (copy_to_user(ulmv, tmp, lum_size))
1257 GOTO(out_tmp, rc = -EFAULT);
1259 OBD_FREE(tmp, lum_size);
1261 ptlrpc_req_finished(request);
1265 case LL_IOC_REMOVE_ENTRY: {
1266 char *filename = NULL;
1270 /* Here is a little hack to avoid sending REINT_RMENTRY to
1271 * unsupported server, which might crash the server(LU-2730),
1272 * Because both LVB_TYPE and REINT_RMENTRY will be supported
1273 * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1274 * server will support REINT_RMENTRY XXX*/
1275 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1278 filename = ll_getname((const char *)arg);
1279 if (IS_ERR(filename))
1280 RETURN(PTR_ERR(filename));
1282 namelen = strlen(filename);
1284 GOTO(out_rmdir, rc = -EINVAL);
1286 rc = ll_rmdir_entry(inode, filename, namelen);
1289 ll_putname(filename);
1292 case LL_IOC_LOV_SWAP_LAYOUTS:
1294 case IOC_OBD_STATFS:
1295 RETURN(ll_obd_statfs(inode, (void *)arg));
1296 case LL_IOC_LOV_GETSTRIPE:
1297 case LL_IOC_MDC_GETINFO:
1298 case IOC_MDC_GETFILEINFO:
1299 case IOC_MDC_GETFILESTRIPE: {
1300 struct ptlrpc_request *request = NULL;
1301 struct lov_user_md *lump;
1302 struct lov_mds_md *lmm = NULL;
1303 struct mdt_body *body;
1304 char *filename = NULL;
1307 if (cmd == IOC_MDC_GETFILEINFO ||
1308 cmd == IOC_MDC_GETFILESTRIPE) {
1309 filename = ll_getname((const char *)arg);
1310 if (IS_ERR(filename))
1311 RETURN(PTR_ERR(filename));
1313 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1314 &lmmsize, &request);
1316 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1321 body = req_capsule_server_get(&request->rq_pill,
1323 LASSERT(body != NULL);
1329 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1330 cmd == LL_IOC_MDC_GETINFO))
1331 GOTO(skip_lmm, rc = 0);
1336 if (cmd == IOC_MDC_GETFILESTRIPE ||
1337 cmd == LL_IOC_LOV_GETSTRIPE) {
1338 lump = (struct lov_user_md *)arg;
1340 struct lov_user_mds_data *lmdp;
1341 lmdp = (struct lov_user_mds_data *)arg;
1342 lump = &lmdp->lmd_lmm;
1344 if (copy_to_user(lump, lmm, lmmsize)) {
1345 if (copy_to_user(lump, lmm, sizeof(*lump)))
1346 GOTO(out_req, rc = -EFAULT);
1350 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1351 struct lov_user_mds_data *lmdp;
1354 st.st_dev = inode->i_sb->s_dev;
1355 st.st_mode = body->mode;
1356 st.st_nlink = body->nlink;
1357 st.st_uid = body->uid;
1358 st.st_gid = body->gid;
1359 st.st_rdev = body->rdev;
1360 st.st_size = body->size;
1361 st.st_blksize = PAGE_CACHE_SIZE;
1362 st.st_blocks = body->blocks;
1363 st.st_atime = body->atime;
1364 st.st_mtime = body->mtime;
1365 st.st_ctime = body->ctime;
1366 st.st_ino = inode->i_ino;
1368 lmdp = (struct lov_user_mds_data *)arg;
1369 if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1370 GOTO(out_req, rc = -EFAULT);
1375 ptlrpc_req_finished(request);
1377 ll_putname(filename);
1380 case IOC_LOV_GETINFO: {
1381 struct lov_user_mds_data *lumd;
1382 struct lov_stripe_md *lsm;
1383 struct lov_user_md *lum;
1384 struct lov_mds_md *lmm;
1388 lumd = (struct lov_user_mds_data *)arg;
1389 lum = &lumd->lmd_lmm;
1391 rc = ll_get_max_mdsize(sbi, &lmmsize);
1395 OBD_ALLOC_LARGE(lmm, lmmsize);
1399 if (copy_from_user(lmm, lum, lmmsize))
1400 GOTO(free_lmm, rc = -EFAULT);
1402 switch (lmm->lmm_magic) {
1403 case LOV_USER_MAGIC_V1:
1404 if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1406 /* swab objects first so that stripes num will be sane */
1407 lustre_swab_lov_user_md_objects(
1408 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1409 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1410 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1412 case LOV_USER_MAGIC_V3:
1413 if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1415 /* swab objects first so that stripes num will be sane */
1416 lustre_swab_lov_user_md_objects(
1417 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1418 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1419 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1422 GOTO(free_lmm, rc = -EINVAL);
1425 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1427 GOTO(free_lmm, rc = -ENOMEM);
1429 /* Perform glimpse_size operation. */
1430 memset(&st, 0, sizeof(st));
1432 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1436 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1437 GOTO(free_lsm, rc = -EFAULT);
1441 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1443 OBD_FREE_LARGE(lmm, lmmsize);
1446 case OBD_IOC_QUOTACHECK: {
1447 struct obd_quotactl *oqctl;
1450 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1451 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1454 OBD_ALLOC_PTR(oqctl);
1457 oqctl->qc_type = arg;
1458 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1460 CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1464 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1466 CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1468 OBD_FREE_PTR(oqctl);
1471 case OBD_IOC_POLL_QUOTACHECK: {
1472 struct if_quotacheck *check;
1474 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1475 sbi->ll_flags & LL_SBI_RMT_CLIENT)
1478 OBD_ALLOC_PTR(check);
1482 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1485 CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1486 if (copy_to_user((void *)arg, check,
1488 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1492 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1495 CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1496 if (copy_to_user((void *)arg, check,
1498 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1502 OBD_FREE_PTR(check);
1505 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
1506 case LL_IOC_QUOTACTL_18: {
1507 /* copy the old 1.x quota struct for internal use, then copy
1508 * back into old format struct. For 1.8 compatibility. */
1509 struct if_quotactl_18 *qctl_18;
1510 struct if_quotactl *qctl_20;
1512 OBD_ALLOC_PTR(qctl_18);
1516 OBD_ALLOC_PTR(qctl_20);
1518 GOTO(out_quotactl_18, rc = -ENOMEM);
1520 if (copy_from_user(qctl_18, (void *)arg, sizeof(*qctl_18)))
1521 GOTO(out_quotactl_20, rc = -ENOMEM);
1523 QCTL_COPY(qctl_20, qctl_18);
1524 qctl_20->qc_idx = 0;
1526 /* XXX: dqb_valid was borrowed as a flag to mark that
1527 * only mds quota is wanted */
1528 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1529 qctl_18->qc_dqblk.dqb_valid) {
1530 qctl_20->qc_valid = QC_MDTIDX;
1531 qctl_20->qc_dqblk.dqb_valid = 0;
1532 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1533 qctl_20->qc_valid = QC_UUID;
1534 qctl_20->obd_uuid = qctl_18->obd_uuid;
1536 qctl_20->qc_valid = QC_GENERAL;
1539 rc = quotactl_ioctl(sbi, qctl_20);
1542 QCTL_COPY(qctl_18, qctl_20);
1543 qctl_18->obd_uuid = qctl_20->obd_uuid;
1545 if (copy_to_user((void *)arg, qctl_18,
1551 OBD_FREE_PTR(qctl_20);
1553 OBD_FREE_PTR(qctl_18);
1557 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
1558 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0) */
1559 case OBD_IOC_QUOTACTL: {
1560 struct if_quotactl *qctl;
1562 OBD_ALLOC_PTR(qctl);
1566 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1567 GOTO(out_quotactl, rc = -EFAULT);
1569 rc = quotactl_ioctl(sbi, qctl);
1571 if (rc == 0 && copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1578 case OBD_IOC_GETDTNAME:
1579 case OBD_IOC_GETMDNAME:
1580 RETURN(ll_get_obd_name(inode, cmd, arg));
1581 case LL_IOC_FLUSHCTX:
1582 RETURN(ll_flush_ctx(inode));
1583 #ifdef CONFIG_FS_POSIX_ACL
1584 case LL_IOC_RMTACL: {
1585 if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1586 inode == inode->i_sb->s_root->d_inode) {
1587 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1589 LASSERT(fd != NULL);
1590 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1592 fd->fd_flags |= LL_FILE_RMTACL;
1598 case LL_IOC_GETOBDCOUNT: {
1600 struct obd_export *exp;
1602 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1605 /* get ost count when count is zero, get mdt count otherwise */
1606 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1607 vallen = sizeof(count);
1608 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1609 KEY_TGT_COUNT, &vallen, &count, NULL);
1611 CERROR("get target count failed: %d\n", rc);
1615 if (copy_to_user((int *)arg, &count, sizeof(int)))
1620 case LL_IOC_PATH2FID:
1621 if (copy_to_user((void *)arg, ll_inode2fid(inode),
1622 sizeof(struct lu_fid)))
1625 case LL_IOC_GET_CONNECT_FLAGS: {
1626 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void*)arg));
1628 case OBD_IOC_CHANGELOG_SEND:
1629 case OBD_IOC_CHANGELOG_CLEAR:
1630 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1631 sizeof(struct ioc_changelog));
1633 case OBD_IOC_FID2PATH:
1634 RETURN(ll_fid2path(inode, (void *)arg));
1635 case LL_IOC_HSM_REQUEST: {
1636 struct hsm_user_request *hur;
1643 /* We don't know the true size yet; copy the fixed-size part */
1644 if (copy_from_user(hur, (void *)arg, sizeof(*hur))) {
1649 /* Compute the whole struct size */
1650 totalsize = hur_len(hur);
1653 /* Final size will be more than double totalsize */
1654 if (totalsize >= MDS_MAXREQSIZE / 3)
1657 OBD_ALLOC_LARGE(hur, totalsize);
1661 /* Copy the whole struct */
1662 if (copy_from_user(hur, (void *)arg, totalsize)) {
1663 OBD_FREE_LARGE(hur, totalsize);
1667 if (hur->hur_request.hr_action == HUA_RELEASE) {
1668 const struct lu_fid *fid;
1672 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1673 fid = &hur->hur_user_item[i].hui_fid;
1674 f = search_inode_for_lustre(inode->i_sb, fid);
1680 rc = ll_hsm_release(f);
1686 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1690 OBD_FREE_LARGE(hur, totalsize);
1694 case LL_IOC_HSM_PROGRESS: {
1695 struct hsm_progress_kernel hpk;
1696 struct hsm_progress hp;
1698 if (copy_from_user(&hp, (void *)arg, sizeof(hp)))
1701 hpk.hpk_fid = hp.hp_fid;
1702 hpk.hpk_cookie = hp.hp_cookie;
1703 hpk.hpk_extent = hp.hp_extent;
1704 hpk.hpk_flags = hp.hp_flags;
1705 hpk.hpk_errval = hp.hp_errval;
1706 hpk.hpk_data_version = 0;
1708 /* File may not exist in Lustre; all progress
1709 * reported to Lustre root */
1710 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1714 case LL_IOC_HSM_CT_START:
1715 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1718 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1719 sizeof(struct lustre_kernelcomm));
1722 case LL_IOC_HSM_COPY_START: {
1723 struct hsm_copy *copy;
1726 OBD_ALLOC_PTR(copy);
1729 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1734 rc = ll_ioc_copy_start(inode->i_sb, copy);
1735 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1741 case LL_IOC_HSM_COPY_END: {
1742 struct hsm_copy *copy;
1745 OBD_ALLOC_PTR(copy);
1748 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1753 rc = ll_ioc_copy_end(inode->i_sb, copy);
1754 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1760 case LL_IOC_MIGRATE: {
1762 const char *filename;
1768 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1772 data = (struct obd_ioctl_data *)buf;
1773 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1774 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1775 GOTO(migrate_free, rc = -EINVAL);
1777 filename = data->ioc_inlbuf1;
1778 namelen = data->ioc_inllen1;
1779 /* \0 is packed at the end of filename */
1780 if (namelen < 1 || namelen != strlen(filename) + 1)
1781 GOTO(migrate_free, rc = -EINVAL);
1783 if (data->ioc_inllen2 != sizeof(mdtidx))
1784 GOTO(migrate_free, rc = -EINVAL);
1785 mdtidx = *(int *)data->ioc_inlbuf2;
1787 rc = ll_migrate(inode, file, mdtidx, filename, namelen - 1);
1789 obd_ioctl_freedata(buf, len);
1794 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1799 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1801 struct inode *inode = file->f_mapping->host;
1802 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1803 struct ll_sb_info *sbi = ll_i2sbi(inode);
1804 int api32 = ll_need_32bit_api(sbi);
1805 loff_t ret = -EINVAL;
1808 mutex_lock(&inode->i_mutex);
1813 offset += file->f_pos;
1819 offset += LL_DIR_END_OFF_32BIT;
1821 offset += LL_DIR_END_OFF;
1828 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1829 (!api32 && offset <= LL_DIR_END_OFF))) {
1830 if (offset != file->f_pos) {
1831 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1832 (!api32 && offset == LL_DIR_END_OFF))
1833 fd->lfd_pos = MDS_DIR_END_OFF;
1834 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1835 fd->lfd_pos = offset << 32;
1837 fd->lfd_pos = offset;
1838 file->f_pos = offset;
1839 file->f_version = 0;
1846 mutex_unlock(&inode->i_mutex);
1850 static int ll_dir_open(struct inode *inode, struct file *file)
1853 RETURN(ll_file_open(inode, file));
1856 static int ll_dir_release(struct inode *inode, struct file *file)
1859 RETURN(ll_file_release(inode, file));
1862 const struct file_operations ll_dir_operations = {
1863 .llseek = ll_dir_seek,
1864 .open = ll_dir_open,
1865 .release = ll_dir_release,
1866 .read = generic_read_dir,
1867 #ifdef HAVE_DIR_CONTEXT
1868 .iterate = ll_iterate,
1870 .readdir = ll_readdir,
1872 .unlocked_ioctl = ll_dir_ioctl,