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, 2015, 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 <linux/user_namespace.h>
46 #ifdef HAVE_UIDGID_HEADER
47 # include <linux/uidgid.h>
49 #include <asm/uaccess.h>
50 #include <linux/buffer_head.h> // for wait_on_buffer
51 #include <linux/pagevec.h>
53 #define DEBUG_SUBSYSTEM S_LLITE
55 #include <lustre/lustre_idl.h>
57 #include <obd_support.h>
58 #include <obd_class.h>
59 #include <lustre_ioctl.h>
60 #include <lustre_lib.h>
61 #include <lustre_dlm.h>
62 #include <lustre_fid.h>
63 #include <lustre_kernelcomm.h>
64 #include <lustre_swab.h>
66 #include "llite_internal.h"
69 * (new) readdir implementation overview.
71 * Original lustre readdir implementation cached exact copy of raw directory
72 * pages on the client. These pages were indexed in client page cache by
73 * logical offset in the directory file. This design, while very simple and
74 * intuitive had some inherent problems:
76 * . it implies that byte offset to the directory entry serves as a
77 * telldir(3)/seekdir(3) cookie, but that offset is not stable: in
78 * ext3/htree directory entries may move due to splits, and more
81 * . it is incompatible with the design of split directories for cmd3,
82 * that assumes that names are distributed across nodes based on their
83 * hash, and so readdir should be done in hash order.
85 * New readdir implementation does readdir in hash order, and uses hash of a
86 * file name as a telldir/seekdir cookie. This led to number of complications:
88 * . hash is not unique, so it cannot be used to index cached directory
89 * pages on the client (note, that it requires a whole pageful of hash
90 * collided entries to cause two pages to have identical hashes);
92 * . hash is not unique, so it cannot, strictly speaking, be used as an
93 * entry cookie. ext3/htree has the same problem and lustre implementation
94 * mimics their solution: seekdir(hash) positions directory at the first
95 * entry with the given hash.
101 * Client caches directory pages using hash of the first entry as an index. As
102 * noted above hash is not unique, so this solution doesn't work as is:
103 * special processing is needed for "page hash chains" (i.e., sequences of
104 * pages filled with entries all having the same hash value).
106 * First, such chains have to be detected. To this end, server returns to the
107 * client the hash of the first entry on the page next to one returned. When
108 * client detects that this hash is the same as hash of the first entry on the
109 * returned page, page hash collision has to be handled. Pages in the
110 * hash chain, except first one, are termed "overflow pages".
112 * Solution to index uniqueness problem is to not cache overflow
113 * pages. Instead, when page hash collision is detected, all overflow pages
114 * from emerging chain are immediately requested from the server and placed in
115 * a special data structure (struct ll_dir_chain). This data structure is used
116 * by ll_readdir() to process entries from overflow pages. When readdir
117 * invocation finishes, overflow pages are discarded. If page hash collision
118 * chain weren't completely processed, next call to readdir will again detect
119 * page hash collision, again read overflow pages in, process next portion of
120 * entries and again discard the pages. This is not as wasteful as it looks,
121 * because, given reasonable hash, page hash collisions are extremely rare.
123 * 1. directory positioning
125 * When seekdir(hash) is called, original
136 * identification of and access to overflow pages
140 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
141 * a header lu_dirpage which describes the start/end hash, and whether this
142 * page is empty (contains no dir entry) or hash collide with next page.
143 * After client receives reply, several pages will be integrated into dir page
144 * in PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the
145 * lu_dirpage for this integrated page will be adjusted. See
146 * mdc_adjust_dirpages().
149 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
150 __u64 offset, struct ll_dir_chain *chain)
152 struct md_callback cb_op;
156 cb_op.md_blocking_ast = ll_md_blocking_ast;
157 rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
164 void ll_release_page(struct inode *inode, struct page *page,
169 /* Always remove the page for striped dir, because the page is
170 * built from temporarily in LMV layer */
171 if (inode != NULL && S_ISDIR(inode->i_mode) &&
172 ll_i2info(inode)->lli_lsm_md != NULL) {
179 if (likely(page->mapping != NULL))
180 truncate_complete_page(page->mapping, page);
183 page_cache_release(page);
187 * return IF_* type for given lu_dirent entry.
188 * IF_* flag shld be converted to particular OS file type in
189 * platform llite module.
191 static u16 ll_dirent_type_get(struct lu_dirent *ent)
194 struct luda_type *lt;
197 if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
198 const unsigned align = sizeof(struct luda_type) - 1;
200 len = le16_to_cpu(ent->lde_namelen);
201 len = (len + align) & ~align;
202 lt = (void *)ent->lde_name + len;
203 type = IFTODT(le16_to_cpu(lt->lt_type));
209 #ifdef HAVE_DIR_CONTEXT
210 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
211 struct dir_context *ctx)
214 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
215 void *cookie, filldir_t filldir)
218 struct ll_sb_info *sbi = ll_i2sbi(inode);
220 bool is_api32 = ll_need_32bit_api(sbi);
221 bool is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
223 struct ll_dir_chain chain;
228 ll_dir_chain_init(&chain);
230 page = ll_get_dir_page(inode, op_data, pos, &chain);
232 while (rc == 0 && !done) {
233 struct lu_dirpage *dp;
234 struct lu_dirent *ent;
243 hash = MDS_DIR_END_OFF;
244 dp = page_address(page);
245 for (ent = lu_dirent_start(dp); ent != NULL && !done;
246 ent = lu_dirent_next(ent)) {
253 hash = le64_to_cpu(ent->lde_hash);
256 * Skip until we find target hash
261 namelen = le16_to_cpu(ent->lde_namelen);
268 if (is_api32 && is_hash64)
272 fid_le_to_cpu(&fid, &ent->lde_fid);
273 ino = cl_fid_build_ino(&fid, is_api32);
274 type = ll_dirent_type_get(ent);
275 /* For 'll_nfs_get_name_filldir()', it will try
276 * to access the 'ent' through its 'lde_name',
277 * so the parameter 'name' for 'filldir()' must
278 * be part of the 'ent'. */
279 #ifdef HAVE_DIR_CONTEXT
281 done = !dir_emit(ctx, ent->lde_name, namelen, ino,
284 done = filldir(cookie, ent->lde_name, namelen, lhash,
291 ll_release_page(inode, page, false);
295 next = le64_to_cpu(dp->ldp_hash_end);
297 if (pos == MDS_DIR_END_OFF) {
299 * End of directory reached.
302 ll_release_page(inode, page, false);
305 * Normal case: continue to the next
308 ll_release_page(inode, page,
309 le32_to_cpu(dp->ldp_flags) &
312 page = ll_get_dir_page(inode, op_data, pos,
316 #ifdef HAVE_DIR_CONTEXT
321 ll_dir_chain_fini(&chain);
325 #ifdef HAVE_DIR_CONTEXT
326 static int ll_iterate(struct file *filp, struct dir_context *ctx)
328 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
331 struct inode *inode = filp->f_path.dentry->d_inode;
332 struct ll_file_data *lfd = LUSTRE_FPRIVATE(filp);
333 struct ll_sb_info *sbi = ll_i2sbi(inode);
334 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
335 int api32 = ll_need_32bit_api(sbi);
336 struct md_op_data *op_data;
346 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
347 "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
348 inode, (unsigned long)pos, i_size_read(inode), api32);
350 if (pos == MDS_DIR_END_OFF)
356 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
357 LUSTRE_OPC_ANY, inode);
359 GOTO(out, rc = PTR_ERR(op_data));
361 if (unlikely(op_data->op_mea1 != NULL)) {
362 /* This is only needed for striped dir to fill ..,
363 * see lmv_read_entry */
364 if (filp->f_path.dentry->d_parent != NULL &&
365 filp->f_path.dentry->d_parent->d_inode != NULL) {
366 __u64 ibits = MDS_INODELOCK_UPDATE;
367 struct inode *parent =
368 filp->f_path.dentry->d_parent->d_inode;
370 if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
371 op_data->op_fid3 = *ll_inode2fid(parent);
374 /* If it can not find in cache, do lookup .. on the master
376 if (fid_is_zero(&op_data->op_fid3)) {
377 rc = ll_dir_get_parent_fid(inode, &op_data->op_fid3);
379 ll_finish_md_op_data(op_data);
384 op_data->op_max_pages = sbi->ll_md_brw_pages;
385 #ifdef HAVE_DIR_CONTEXT
387 rc = ll_dir_read(inode, &pos, op_data, ctx);
390 rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
395 if (pos == MDS_DIR_END_OFF) {
397 pos = LL_DIR_END_OFF_32BIT;
399 pos = LL_DIR_END_OFF;
404 #ifdef HAVE_DIR_CONTEXT
409 ll_finish_md_op_data(op_data);
410 filp->f_version = inode->i_version;
414 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
419 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
421 struct mgs_send_param *msp;
428 strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
429 rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
430 sizeof(struct mgs_send_param), msp, NULL);
432 CERROR("Failed to set parameter: %d\n", rc);
439 * Create striped directory with specified stripe(@lump)
441 * param[in]parent the parent of the directory.
442 * param[in]lump the specified stripes.
443 * param[in]dirname the name of the directory.
444 * param[in]mode the specified mode of the directory.
446 * retval =0 if striped directory is being created successfully.
447 * <0 if the creation is failed.
449 static int ll_dir_setdirstripe(struct inode *parent, struct lmv_user_md *lump,
450 const char *dirname, umode_t mode)
452 struct ptlrpc_request *request = NULL;
453 struct md_op_data *op_data;
454 struct ll_sb_info *sbi = ll_i2sbi(parent);
455 struct inode *inode = NULL;
456 struct dentry dentry;
460 if (unlikely(lump->lum_magic != LMV_USER_MAGIC))
463 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s"
464 "stripe_offset %d, stripe_count: %u\n",
465 PFID(ll_inode2fid(parent)), parent, dirname,
466 (int)lump->lum_stripe_offset, lump->lum_stripe_count);
468 if (lump->lum_stripe_count > 1 &&
469 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
472 if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC))
473 lustre_swab_lmv_user_md(lump);
475 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
476 mode &= ~current_umask();
477 mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
478 op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
479 strlen(dirname), mode, LUSTRE_OPC_MKDIR,
482 GOTO(err_exit, err = PTR_ERR(op_data));
484 op_data->op_cli_flags |= CLI_SET_MEA;
485 err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
486 from_kuid(&init_user_ns, current_fsuid()),
487 from_kgid(&init_user_ns, current_fsgid()),
488 cfs_curproc_cap_pack(), 0, &request);
489 ll_finish_md_op_data(op_data);
493 err = ll_prep_inode(&inode, request, parent->i_sb, NULL);
497 memset(&dentry, 0, sizeof(dentry));
498 dentry.d_inode = inode;
500 err = ll_init_security(&dentry, inode, parent);
506 ptlrpc_req_finished(request);
510 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
513 struct ll_sb_info *sbi = ll_i2sbi(inode);
514 struct md_op_data *op_data;
515 struct ptlrpc_request *req = NULL;
517 struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
518 struct obd_device *mgc = lsi->lsi_mgc;
524 * This is coming from userspace, so should be in
525 * local endian. But the MDS would like it in little
526 * endian, so we swab it before we send it.
528 switch (lump->lmm_magic) {
529 case LOV_USER_MAGIC_V1: {
530 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
531 lustre_swab_lov_user_md_v1(lump);
532 lum_size = sizeof(struct lov_user_md_v1);
535 case LOV_USER_MAGIC_V3: {
536 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
537 lustre_swab_lov_user_md_v3(
538 (struct lov_user_md_v3 *)lump);
539 lum_size = sizeof(struct lov_user_md_v3);
542 case LMV_USER_MAGIC: {
543 if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
544 lustre_swab_lmv_user_md(
545 (struct lmv_user_md *)lump);
546 lum_size = sizeof(struct lmv_user_md);
550 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
551 " %#08x != %#08x nor %#08x\n",
552 lump->lmm_magic, LOV_USER_MAGIC_V1,
558 lum_size = sizeof(struct lov_user_md_v1);
561 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
562 LUSTRE_OPC_ANY, NULL);
564 RETURN(PTR_ERR(op_data));
566 /* swabbing is done in lov_setstripe() on server side */
567 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
568 ll_finish_md_op_data(op_data);
569 ptlrpc_req_finished(req);
571 if (rc != -EPERM && rc != -EACCES)
572 CERROR("mdc_setattr fails: rc = %d\n", rc);
575 /* In the following we use the fact that LOV_USER_MAGIC_V1 and
576 LOV_USER_MAGIC_V3 have the same initial fields so we do not
577 need the make the distiction between the 2 versions */
578 if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
582 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
584 GOTO(end, rc = -ENOMEM);
587 /* Get fsname and assume devname to be -MDT0000. */
588 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
589 strcat(buf, "-MDT0000.lov");
592 /* Set root stripesize */
593 sprintf(buf, ".stripesize=%u",
594 lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
595 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
599 /* Set root stripecount */
600 sprintf(buf, ".stripecount=%hd",
601 lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
602 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
606 /* Set root stripeoffset */
607 sprintf(buf, ".stripeoffset=%hd",
608 lump ? le16_to_cpu(lump->lmm_stripe_offset) :
609 (typeof(lump->lmm_stripe_offset))(-1));
610 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
614 OBD_FREE(param, MGS_PARAM_MAXLEN);
620 * This function will be used to get default LOV/LMV/Default LMV
621 * @valid will be used to indicate which stripe it will retrieve
622 * OBD_MD_MEA LMV stripe EA
623 * OBD_MD_DEFAULT_MEA Default LMV stripe EA
624 * otherwise Default LOV EA.
625 * Each time, it can only retrieve 1 stripe EA
627 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
628 struct ptlrpc_request **request, u64 valid)
630 struct ll_sb_info *sbi = ll_i2sbi(inode);
631 struct mdt_body *body;
632 struct lov_mds_md *lmm = NULL;
633 struct ptlrpc_request *req = NULL;
635 struct md_op_data *op_data;
638 rc = ll_get_default_mdsize(sbi, &lmm_size);
642 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
643 0, lmm_size, LUSTRE_OPC_ANY,
646 RETURN(PTR_ERR(op_data));
648 op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
649 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
650 ll_finish_md_op_data(op_data);
652 CDEBUG(D_INFO, "md_getattr failed on inode "
653 DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
657 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
658 LASSERT(body != NULL);
660 lmm_size = body->mbo_eadatasize;
662 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
664 GOTO(out, rc = -ENODATA);
667 lmm = req_capsule_server_sized_get(&req->rq_pill,
668 &RMF_MDT_MD, lmm_size);
669 LASSERT(lmm != NULL);
672 * This is coming from the MDS, so is probably in
673 * little endian. We convert it to host endian before
674 * passing it to userspace.
676 /* We don't swab objects for directories */
677 switch (le32_to_cpu(lmm->lmm_magic)) {
679 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
680 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
683 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
684 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
687 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
688 lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
691 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
692 lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
695 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
700 *plmm_size = lmm_size;
705 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
707 struct md_op_data *op_data;
712 OBD_ALLOC_PTR(op_data);
716 op_data->op_flags |= MF_GET_MDT_IDX;
717 op_data->op_fid1 = *fid;
718 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
719 mdt_index = op_data->op_mds;
720 OBD_FREE_PTR(op_data);
728 * Get MDT index for the inode.
730 int ll_get_mdt_idx(struct inode *inode)
732 return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
736 * Generic handler to do any pre-copy work.
738 * It sends a first hsm_progress (with extent length == 0) to coordinator as a
739 * first information for it that real work has started.
741 * Moreover, for a ARCHIVE request, it will sample the file data version and
742 * store it in \a copy.
744 * \return 0 on success.
746 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
748 struct ll_sb_info *sbi = ll_s2sbi(sb);
749 struct hsm_progress_kernel hpk;
754 /* Forge a hsm_progress based on data from copy. */
755 hpk.hpk_fid = copy->hc_hai.hai_fid;
756 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
757 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
758 hpk.hpk_extent.length = 0;
761 hpk.hpk_data_version = 0;
764 /* For archive request, we need to read the current file version. */
765 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
767 __u64 data_version = 0;
769 /* Get inode for this fid */
770 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
772 hpk.hpk_flags |= HP_FLAG_RETRY;
773 /* hpk_errval is >= 0 */
774 hpk.hpk_errval = -PTR_ERR(inode);
775 GOTO(progress, rc = PTR_ERR(inode));
778 /* Read current file data version */
779 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
782 CDEBUG(D_HSM, "Could not read file data version of "
783 DFID" (rc = %d). Archive request ("
784 LPX64") could not be done.\n",
785 PFID(©->hc_hai.hai_fid), rc,
786 copy->hc_hai.hai_cookie);
787 hpk.hpk_flags |= HP_FLAG_RETRY;
788 /* hpk_errval must be >= 0 */
789 hpk.hpk_errval = -rc;
793 /* Store in the hsm_copy for later copytool use.
794 * Always modified even if no lsm. */
795 copy->hc_data_version = data_version;
799 /* On error, the request should be considered as completed */
800 if (hpk.hpk_errval > 0)
801 hpk.hpk_flags |= HP_FLAG_COMPLETED;
803 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
806 /* Return first error */
807 RETURN(rc != 0 ? rc : rc2);
811 * Generic handler to do any post-copy work.
813 * It will send the last hsm_progress update to coordinator to inform it
814 * that copy is finished and whether it was successful or not.
817 * - for ARCHIVE request, it will sample the file data version and compare it
818 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
819 * will be considered as failed.
820 * - for RESTORE request, it will sample the file data version and send it to
821 * coordinator which is useful if the file was imported as 'released'.
823 * \return 0 on success.
825 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
827 struct ll_sb_info *sbi = ll_s2sbi(sb);
828 struct hsm_progress_kernel hpk;
833 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
834 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
835 * initialized if copy_end was called with copy == NULL.
838 /* Forge a hsm_progress based on data from copy. */
839 hpk.hpk_fid = copy->hc_hai.hai_fid;
840 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
841 hpk.hpk_extent = copy->hc_hai.hai_extent;
842 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
843 hpk.hpk_errval = copy->hc_errval;
844 hpk.hpk_data_version = 0;
846 /* For archive request, we need to check the file data was not changed.
848 * For restore request, we need to send the file data version, this is
849 * useful when the file was created using hsm_import.
851 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
852 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
853 (copy->hc_errval == 0)) {
855 __u64 data_version = 0;
857 /* Get lsm for this fid */
858 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
860 hpk.hpk_flags |= HP_FLAG_RETRY;
861 /* hpk_errval must be >= 0 */
862 hpk.hpk_errval = -PTR_ERR(inode);
863 GOTO(progress, rc = PTR_ERR(inode));
866 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
869 CDEBUG(D_HSM, "Could not read file data version. "
870 "Request could not be confirmed.\n");
871 if (hpk.hpk_errval == 0)
872 hpk.hpk_errval = -rc;
876 /* Store in the hsm_copy for later copytool use.
877 * Always modified even if no lsm. */
878 hpk.hpk_data_version = data_version;
880 /* File could have been stripped during archiving, so we need
881 * to check anyway. */
882 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
883 (copy->hc_data_version != data_version)) {
884 CDEBUG(D_HSM, "File data version mismatched. "
885 "File content was changed during archiving. "
886 DFID", start:"LPX64" current:"LPX64"\n",
887 PFID(©->hc_hai.hai_fid),
888 copy->hc_data_version, data_version);
889 /* File was changed, send error to cdt. Do not ask for
890 * retry because if a file is modified frequently,
891 * the cdt will loop on retried archive requests.
892 * The policy engine will ask for a new archive later
893 * when the file will not be modified for some tunable
895 hpk.hpk_flags &= ~HP_FLAG_RETRY;
897 /* hpk_errval must be >= 0 */
898 hpk.hpk_errval = -rc;
905 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
908 /* Return first error */
909 RETURN(rc != 0 ? rc : rc2);
913 static int copy_and_ioctl(int cmd, struct obd_export *exp,
914 const void __user *data, size_t size)
919 OBD_ALLOC(copy, size);
923 if (copy_from_user(copy, data, size)) {
928 rc = obd_iocontrol(cmd, exp, size, copy, NULL);
930 OBD_FREE(copy, size);
935 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
937 int cmd = qctl->qc_cmd;
938 int type = qctl->qc_type;
939 int id = qctl->qc_id;
940 int valid = qctl->qc_valid;
947 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
948 sbi->ll_flags & LL_SBI_RMT_CLIENT)
952 if (((type == USRQUOTA &&
953 !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
955 !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
956 (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
957 sbi->ll_flags & LL_SBI_RMT_CLIENT))
963 CERROR("unsupported quotactl op: %#x\n", cmd);
967 if (valid != QC_GENERAL) {
968 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
971 if (cmd == Q_GETINFO)
972 qctl->qc_cmd = Q_GETOINFO;
973 else if (cmd == Q_GETQUOTA)
974 qctl->qc_cmd = Q_GETOQUOTA;
980 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
981 sizeof(*qctl), qctl, NULL);
984 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
985 sizeof(*qctl), qctl, NULL);
988 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
989 sizeof(*qctl), qctl, NULL);
991 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
993 sizeof(*qctl), qctl, NULL);
1005 struct obd_quotactl *oqctl;
1007 OBD_ALLOC_PTR(oqctl);
1011 QCTL_COPY(oqctl, qctl);
1012 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1014 OBD_FREE_PTR(oqctl);
1017 /* If QIF_SPACE is not set, client should collect the
1018 * space usage from OSSs by itself */
1019 if (cmd == Q_GETQUOTA &&
1020 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1021 !oqctl->qc_dqblk.dqb_curspace) {
1022 struct obd_quotactl *oqctl_tmp;
1024 OBD_ALLOC_PTR(oqctl_tmp);
1025 if (oqctl_tmp == NULL)
1026 GOTO(out, rc = -ENOMEM);
1028 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1029 oqctl_tmp->qc_id = oqctl->qc_id;
1030 oqctl_tmp->qc_type = oqctl->qc_type;
1032 /* collect space usage from OSTs */
1033 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1034 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1035 if (!rc || rc == -EREMOTEIO) {
1036 oqctl->qc_dqblk.dqb_curspace =
1037 oqctl_tmp->qc_dqblk.dqb_curspace;
1038 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1041 /* collect space & inode usage from MDTs */
1042 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1043 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1044 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1045 if (!rc || rc == -EREMOTEIO) {
1046 oqctl->qc_dqblk.dqb_curspace +=
1047 oqctl_tmp->qc_dqblk.dqb_curspace;
1048 oqctl->qc_dqblk.dqb_curinodes =
1049 oqctl_tmp->qc_dqblk.dqb_curinodes;
1050 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1052 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1055 OBD_FREE_PTR(oqctl_tmp);
1058 QCTL_COPY(qctl, oqctl);
1059 OBD_FREE_PTR(oqctl);
1065 /* This function tries to get a single name component,
1066 * to send to the server. No actual path traversal involved,
1067 * so we limit to NAME_MAX */
1068 static char *ll_getname(const char __user *filename)
1073 OBD_ALLOC(tmp, NAME_MAX + 1);
1076 return ERR_PTR(-ENOMEM);
1078 len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1081 else if (len > NAME_MAX)
1082 ret = -ENAMETOOLONG;
1085 OBD_FREE(tmp, NAME_MAX + 1);
1091 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1093 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1095 struct inode *inode = file->f_path.dentry->d_inode;
1096 struct ll_sb_info *sbi = ll_i2sbi(inode);
1097 struct obd_ioctl_data *data;
1101 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1102 PFID(ll_inode2fid(inode)), inode, cmd);
1104 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1105 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1108 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1110 case FSFILT_IOC_GETFLAGS:
1111 case FSFILT_IOC_SETFLAGS:
1112 RETURN(ll_iocontrol(inode, file, cmd, arg));
1113 case FSFILT_IOC_GETVERSION_OLD:
1114 case FSFILT_IOC_GETVERSION:
1115 RETURN(put_user(inode->i_generation, (int __user *)arg));
1116 /* We need to special case any other ioctls we want to handle,
1117 * to send them to the MDS/OST as appropriate and to properly
1118 * network encode the arg field.
1119 case FSFILT_IOC_SETVERSION_OLD:
1120 case FSFILT_IOC_SETVERSION:
1122 case LL_IOC_GET_MDTIDX: {
1125 mdtidx = ll_get_mdt_idx(inode);
1129 if (put_user((int)mdtidx, (int __user *)arg))
1134 case IOC_MDC_LOOKUP: {
1135 int namelen, len = 0;
1139 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1144 filename = data->ioc_inlbuf1;
1145 namelen = strlen(filename);
1147 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1148 GOTO(out_free, rc = -EINVAL);
1151 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1153 CERROR("%s: lookup %.*s failed: rc = %d\n",
1154 ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1159 obd_ioctl_freedata(buf, len);
1162 case LL_IOC_LMV_SETSTRIPE: {
1163 struct lmv_user_md *lum;
1172 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1177 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1178 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1179 GOTO(lmv_out_free, rc = -EINVAL);
1181 filename = data->ioc_inlbuf1;
1182 namelen = data->ioc_inllen1;
1185 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1186 GOTO(lmv_out_free, rc = -EINVAL);
1188 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1189 lumlen = data->ioc_inllen2;
1191 if (lum->lum_magic != LMV_USER_MAGIC ||
1192 lumlen != sizeof(*lum)) {
1193 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1194 filename, lum->lum_magic, lumlen, -EFAULT);
1195 GOTO(lmv_out_free, rc = -EINVAL);
1198 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1199 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1201 mode = data->ioc_type;
1203 rc = ll_dir_setdirstripe(inode, lum, filename, mode);
1205 obd_ioctl_freedata(buf, len);
1209 case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1210 struct lmv_user_md lum;
1211 struct lmv_user_md __user *ulump =
1212 (struct lmv_user_md __user *)arg;
1215 if (copy_from_user(&lum, ulump, sizeof(lum)))
1218 if (lum.lum_magic != LMV_USER_MAGIC)
1221 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1225 case LL_IOC_LOV_SETSTRIPE: {
1226 struct lov_user_md_v3 lumv3;
1227 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1228 struct lov_user_md_v1 __user *lumv1p =
1229 (struct lov_user_md_v1 __user *)arg;
1230 struct lov_user_md_v3 __user *lumv3p =
1231 (struct lov_user_md_v3 __user *)arg;
1233 int set_default = 0;
1235 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1236 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1237 sizeof(lumv3p->lmm_objects[0]));
1238 /* first try with v1 which is smaller than v3 */
1239 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1242 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1243 if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1247 if (inode->i_sb->s_root == file->f_path.dentry)
1250 /* in v1 and v3 cases lumv1 points to data */
1251 rc = ll_dir_setstripe(inode, lumv1, set_default);
1255 case LL_IOC_LMV_GETSTRIPE: {
1256 struct lmv_user_md __user *ulmv =
1257 (struct lmv_user_md __user *)arg;
1258 struct lmv_user_md lum;
1259 struct ptlrpc_request *request = NULL;
1260 union lmv_mds_md *lmm = NULL;
1263 struct lmv_user_md *tmp = NULL;
1267 int max_stripe_count;
1271 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1274 max_stripe_count = lum.lum_stripe_count;
1275 /* lum_magic will indicate which stripe the ioctl will like
1276 * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1277 * is for default LMV stripe */
1278 if (lum.lum_magic == LMV_MAGIC_V1)
1279 valid |= OBD_MD_MEA;
1280 else if (lum.lum_magic == LMV_USER_MAGIC)
1281 valid |= OBD_MD_DEFAULT_MEA;
1285 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1288 GOTO(finish_req, rc);
1290 /* Get default LMV EA */
1291 if (lum.lum_magic == LMV_USER_MAGIC) {
1293 GOTO(finish_req, rc);
1295 if (lmmsize > sizeof(*ulmv))
1296 GOTO(finish_req, rc = -EINVAL);
1298 if (copy_to_user(ulmv, lmm, lmmsize))
1299 GOTO(finish_req, rc = -EFAULT);
1301 GOTO(finish_req, rc);
1304 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1305 if (max_stripe_count < stripe_count) {
1306 lum.lum_stripe_count = stripe_count;
1307 if (copy_to_user(ulmv, &lum, sizeof(lum)))
1308 GOTO(finish_req, rc = -EFAULT);
1309 GOTO(finish_req, rc = -E2BIG);
1312 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1313 OBD_ALLOC(tmp, lum_size);
1315 GOTO(finish_req, rc = -ENOMEM);
1317 mdt_index = ll_get_mdt_idx(inode);
1319 GOTO(out_tmp, rc = -ENOMEM);
1321 tmp->lum_magic = LMV_MAGIC_V1;
1322 tmp->lum_stripe_count = 0;
1323 tmp->lum_stripe_offset = mdt_index;
1324 for (i = 0; i < stripe_count; i++) {
1327 fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1328 mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1330 GOTO(out_tmp, rc = mdt_index);
1332 tmp->lum_objects[i].lum_mds = mdt_index;
1333 tmp->lum_objects[i].lum_fid = fid;
1334 tmp->lum_stripe_count++;
1337 if (copy_to_user(ulmv, tmp, lum_size))
1338 GOTO(out_tmp, rc = -EFAULT);
1340 OBD_FREE(tmp, lum_size);
1342 ptlrpc_req_finished(request);
1346 case LL_IOC_REMOVE_ENTRY: {
1347 char *filename = NULL;
1351 /* Here is a little hack to avoid sending REINT_RMENTRY to
1352 * unsupported server, which might crash the server(LU-2730),
1353 * Because both LVB_TYPE and REINT_RMENTRY will be supported
1354 * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1355 * server will support REINT_RMENTRY XXX*/
1356 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1359 filename = ll_getname((const char __user *)arg);
1360 if (IS_ERR(filename))
1361 RETURN(PTR_ERR(filename));
1363 namelen = strlen(filename);
1365 GOTO(out_rmdir, rc = -EINVAL);
1367 rc = ll_rmdir_entry(inode, filename, namelen);
1370 ll_putname(filename);
1373 case LL_IOC_LOV_SWAP_LAYOUTS:
1375 case IOC_OBD_STATFS:
1376 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1377 case LL_IOC_LOV_GETSTRIPE:
1378 case LL_IOC_MDC_GETINFO:
1379 case IOC_MDC_GETFILEINFO:
1380 case IOC_MDC_GETFILESTRIPE: {
1381 struct ptlrpc_request *request = NULL;
1382 struct lov_user_md __user *lump;
1383 struct lov_mds_md *lmm = NULL;
1384 struct mdt_body *body;
1385 char *filename = NULL;
1388 if (cmd == IOC_MDC_GETFILEINFO ||
1389 cmd == IOC_MDC_GETFILESTRIPE) {
1390 filename = ll_getname((const char __user *)arg);
1391 if (IS_ERR(filename))
1392 RETURN(PTR_ERR(filename));
1394 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1395 &lmmsize, &request);
1397 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1402 body = req_capsule_server_get(&request->rq_pill,
1404 LASSERT(body != NULL);
1410 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1411 cmd == LL_IOC_MDC_GETINFO))
1412 GOTO(skip_lmm, rc = 0);
1417 if (cmd == IOC_MDC_GETFILESTRIPE ||
1418 cmd == LL_IOC_LOV_GETSTRIPE) {
1419 lump = (struct lov_user_md __user *)arg;
1421 struct lov_user_mds_data __user *lmdp;
1422 lmdp = (struct lov_user_mds_data __user *)arg;
1423 lump = &lmdp->lmd_lmm;
1425 if (copy_to_user(lump, lmm, lmmsize)) {
1426 if (copy_to_user(lump, lmm, sizeof(*lump)))
1427 GOTO(out_req, rc = -EFAULT);
1431 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1432 struct lov_user_mds_data __user *lmdp;
1435 st.st_dev = inode->i_sb->s_dev;
1436 st.st_mode = body->mbo_mode;
1437 st.st_nlink = body->mbo_nlink;
1438 st.st_uid = body->mbo_uid;
1439 st.st_gid = body->mbo_gid;
1440 st.st_rdev = body->mbo_rdev;
1441 st.st_size = body->mbo_size;
1442 st.st_blksize = PAGE_CACHE_SIZE;
1443 st.st_blocks = body->mbo_blocks;
1444 st.st_atime = body->mbo_atime;
1445 st.st_mtime = body->mbo_mtime;
1446 st.st_ctime = body->mbo_ctime;
1447 st.st_ino = inode->i_ino;
1449 lmdp = (struct lov_user_mds_data __user *)arg;
1450 if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1451 GOTO(out_req, rc = -EFAULT);
1456 ptlrpc_req_finished(request);
1458 ll_putname(filename);
1461 case OBD_IOC_QUOTACTL: {
1462 struct if_quotactl *qctl;
1464 OBD_ALLOC_PTR(qctl);
1468 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1469 GOTO(out_quotactl, rc = -EFAULT);
1471 rc = quotactl_ioctl(sbi, qctl);
1474 copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1481 case OBD_IOC_GETDTNAME:
1482 case OBD_IOC_GETMDNAME:
1483 RETURN(ll_get_obd_name(inode, cmd, arg));
1484 case LL_IOC_FLUSHCTX:
1485 RETURN(ll_flush_ctx(inode));
1486 #ifdef CONFIG_FS_POSIX_ACL
1487 case LL_IOC_RMTACL: {
1488 if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1489 inode == inode->i_sb->s_root->d_inode) {
1490 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1492 LASSERT(fd != NULL);
1493 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1495 fd->fd_flags |= LL_FILE_RMTACL;
1501 case LL_IOC_GETOBDCOUNT: {
1503 struct obd_export *exp;
1505 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1508 /* get ost count when count is zero, get mdt count otherwise */
1509 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1510 vallen = sizeof(count);
1511 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1512 KEY_TGT_COUNT, &vallen, &count);
1514 CERROR("get target count failed: %d\n", rc);
1518 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1523 case LL_IOC_PATH2FID:
1524 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1525 sizeof(struct lu_fid)))
1528 case LL_IOC_GET_CONNECT_FLAGS: {
1529 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1530 (void __user *)arg));
1532 case OBD_IOC_CHANGELOG_SEND:
1533 case OBD_IOC_CHANGELOG_CLEAR:
1534 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1537 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1538 sizeof(struct ioc_changelog));
1540 case OBD_IOC_FID2PATH:
1541 RETURN(ll_fid2path(inode, (void __user *)arg));
1542 case LL_IOC_GETPARENT:
1543 RETURN(ll_getparent(file, (void __user *)arg));
1544 case LL_IOC_FID2MDTIDX: {
1545 struct obd_export *exp = ll_i2mdexp(inode);
1549 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1553 /* Call mdc_iocontrol */
1554 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1555 (__u32 __user *)&index);
1561 case LL_IOC_HSM_REQUEST: {
1562 struct hsm_user_request *hur;
1569 /* We don't know the true size yet; copy the fixed-size part */
1570 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1575 /* Compute the whole struct size */
1576 totalsize = hur_len(hur);
1581 /* Final size will be more than double totalsize */
1582 if (totalsize >= MDS_MAXREQSIZE / 3)
1585 OBD_ALLOC_LARGE(hur, totalsize);
1589 /* Copy the whole struct */
1590 if (copy_from_user(hur, (void __user *)arg, totalsize))
1591 GOTO(out_hur, rc = -EFAULT);
1593 if (hur->hur_request.hr_action == HUA_RELEASE) {
1594 const struct lu_fid *fid;
1598 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1599 fid = &hur->hur_user_item[i].hui_fid;
1600 f = search_inode_for_lustre(inode->i_sb, fid);
1606 rc = ll_hsm_release(f);
1612 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1617 OBD_FREE_LARGE(hur, totalsize);
1621 case LL_IOC_HSM_PROGRESS: {
1622 struct hsm_progress_kernel hpk;
1623 struct hsm_progress hp;
1625 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1628 hpk.hpk_fid = hp.hp_fid;
1629 hpk.hpk_cookie = hp.hp_cookie;
1630 hpk.hpk_extent = hp.hp_extent;
1631 hpk.hpk_flags = hp.hp_flags;
1632 hpk.hpk_errval = hp.hp_errval;
1633 hpk.hpk_data_version = 0;
1635 /* File may not exist in Lustre; all progress
1636 * reported to Lustre root */
1637 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1641 case LL_IOC_HSM_CT_START:
1642 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1645 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1646 sizeof(struct lustre_kernelcomm));
1649 case LL_IOC_HSM_COPY_START: {
1650 struct hsm_copy *copy;
1653 OBD_ALLOC_PTR(copy);
1656 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1661 rc = ll_ioc_copy_start(inode->i_sb, copy);
1662 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1668 case LL_IOC_HSM_COPY_END: {
1669 struct hsm_copy *copy;
1672 OBD_ALLOC_PTR(copy);
1675 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1680 rc = ll_ioc_copy_end(inode->i_sb, copy);
1681 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1687 case LL_IOC_MIGRATE: {
1689 const char *filename;
1695 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1699 data = (struct obd_ioctl_data *)buf;
1700 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1701 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1702 GOTO(migrate_free, rc = -EINVAL);
1704 filename = data->ioc_inlbuf1;
1705 namelen = data->ioc_inllen1;
1706 /* \0 is packed at the end of filename */
1707 if (namelen < 1 || namelen != strlen(filename) + 1)
1708 GOTO(migrate_free, rc = -EINVAL);
1710 if (data->ioc_inllen2 != sizeof(mdtidx))
1711 GOTO(migrate_free, rc = -EINVAL);
1712 mdtidx = *(int *)data->ioc_inlbuf2;
1714 rc = ll_migrate(inode, file, mdtidx, filename, namelen - 1);
1716 obd_ioctl_freedata(buf, len);
1721 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1722 (void __user *)arg));
1726 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1728 struct inode *inode = file->f_mapping->host;
1729 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1730 struct ll_sb_info *sbi = ll_i2sbi(inode);
1731 int api32 = ll_need_32bit_api(sbi);
1732 loff_t ret = -EINVAL;
1735 mutex_lock(&inode->i_mutex);
1740 offset += file->f_pos;
1746 offset += LL_DIR_END_OFF_32BIT;
1748 offset += LL_DIR_END_OFF;
1755 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1756 (!api32 && offset <= LL_DIR_END_OFF))) {
1757 if (offset != file->f_pos) {
1758 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1759 (!api32 && offset == LL_DIR_END_OFF))
1760 fd->lfd_pos = MDS_DIR_END_OFF;
1761 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1762 fd->lfd_pos = offset << 32;
1764 fd->lfd_pos = offset;
1765 file->f_pos = offset;
1766 file->f_version = 0;
1773 mutex_unlock(&inode->i_mutex);
1777 static int ll_dir_open(struct inode *inode, struct file *file)
1780 RETURN(ll_file_open(inode, file));
1783 static int ll_dir_release(struct inode *inode, struct file *file)
1786 RETURN(ll_file_release(inode, file));
1789 const struct file_operations ll_dir_operations = {
1790 .llseek = ll_dir_seek,
1791 .open = ll_dir_open,
1792 .release = ll_dir_release,
1793 .read = generic_read_dir,
1794 #ifdef HAVE_DIR_CONTEXT
1795 .iterate = ll_iterate,
1797 .readdir = ll_readdir,
1799 .unlocked_ioctl = ll_dir_ioctl,