1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/dir.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
16 * ext2 directory handling functions
18 * Big-endian to little-endian byte-swapping/bitmaps by
19 * David S. Miller (davem@caip.rutgers.edu), 1995
21 * All code that works with directory layout had been switched to pagecache
24 * Adapted for Lustre Light
25 * Copyright (C) 2002-2003, Cluster File Systems, Inc.
30 #include <linux/ext2_fs.h>
31 #include <linux/pagemap.h>
33 #include <linux/version.h>
34 #include <linux/smp_lock.h>
35 #include <asm/uaccess.h>
36 #include <linux/file.h>
37 #include <linux/kmod.h>
38 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
39 # include <linux/locks.h> // for wait_on_buffer
41 # include <linux/buffer_head.h> // for wait_on_buffer
44 #define DEBUG_SUBSYSTEM S_LLITE
46 #include <linux/obd_support.h>
47 #include <linux/obd_class.h>
48 #include <linux/lustre_lib.h>
49 #include <linux/lustre_idl.h>
50 #include <linux/lustre_mds.h>
51 #include <linux/lustre_lite.h>
52 #include <linux/lustre_dlm.h>
53 #include <linux/lustre_smfs.h>
54 #include <linux/lustre_snap.h>
55 #include "llite_internal.h"
57 typedef struct ext2_dir_entry_2 ext2_dirent;
59 #define PageChecked(page) test_bit(PG_checked, &(page)->flags)
60 #define SetPageChecked(page) set_bit(PG_checked, &(page)->flags)
62 /* returns the page unlocked, but with a reference */
63 static int ll_dir_readpage(struct file *file, struct page *page)
65 struct inode *inode = page->mapping->host;
66 struct ll_fid mdc_fid;
68 struct ptlrpc_request *request;
69 struct mds_body *body;
73 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
74 inode->i_generation, inode);
76 mdc_pack_fid(&mdc_fid, inode->i_ino, inode->i_generation, S_IFDIR);
77 mdc_fid.mds = ll_i2info(inode)->lli_mds;
78 offset = page->index << PAGE_SHIFT;
79 rc = md_readpage(ll_i2sbi(inode)->ll_mdc_exp, &mdc_fid,
80 offset, page, &request);
82 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof (*body));
83 LASSERT (body != NULL); /* checked by md_readpage() */
84 LASSERT_REPSWABBED (request, 0); /* swabbed by md_readpage() */
86 #warning "FIXME ASAP!"
87 //inode->i_size = body->size;
88 SetPageUptodate(page);
90 ptlrpc_req_finished(request);
97 struct address_space_operations ll_dir_aops = {
98 .readpage = ll_dir_readpage,
102 * ext2 uses block-sized chunks. Arguably, sector-sized ones would be
103 * more robust, but we have what we have
105 static inline unsigned ext2_chunk_size(struct inode *inode)
107 return inode->i_sb->s_blocksize;
110 static inline void ext2_put_page(struct page *page)
113 page_cache_release(page);
116 static inline unsigned long dir_pages(struct inode *inode)
118 return (inode->i_size+PAGE_CACHE_SIZE-1)>>PAGE_CACHE_SHIFT;
122 static void ext2_check_page(struct page *page)
124 struct inode *dir = page->mapping->host;
125 unsigned chunk_size = ext2_chunk_size(dir);
126 char *kaddr = page_address(page);
127 // u32 max_inumber = le32_to_cpu(sb->u.ext2_sb.s_es->s_inodes_count);
128 unsigned offs, rec_len;
129 unsigned limit = PAGE_CACHE_SIZE;
133 if ((dir->i_size >> PAGE_CACHE_SHIFT) == page->index) {
134 limit = dir->i_size & ~PAGE_CACHE_MASK;
135 if (limit & (chunk_size - 1)) {
136 CERROR("limit %d dir size %lld index %ld\n",
137 limit, dir->i_size, page->index);
140 for (offs = limit; offs<PAGE_CACHE_SIZE; offs += chunk_size) {
141 ext2_dirent *p = (ext2_dirent*)(kaddr + offs);
142 p->rec_len = cpu_to_le16(chunk_size);
149 for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
150 p = (ext2_dirent *)(kaddr + offs);
151 rec_len = le16_to_cpu(p->rec_len);
153 if (rec_len < EXT2_DIR_REC_LEN(1))
157 if (rec_len < EXT2_DIR_REC_LEN(p->name_len))
159 if (((offs + rec_len - 1) ^ offs) & ~(chunk_size-1))
161 // if (le32_to_cpu(p->inode) > max_inumber)
167 SetPageChecked(page);
170 /* Too bad, we had an error */
173 CERROR("ext2_check_page"
174 "size of directory #%lu is not a multiple of chunk size\n",
179 error = "rec_len is smaller than minimal";
182 error = "unaligned directory entry";
185 error = "rec_len is too small for name_len";
188 error = "directory entry across blocks";
191 // error = "inode out of bounds";
193 CERROR("ext2_check_page: bad entry in directory #%lu: %s - "
194 "offset=%lu+%u, inode=%lu, rec_len=%d, name_len=%d",
195 dir->i_ino, error, (page->index<<PAGE_CACHE_SHIFT), offs,
196 (unsigned long) le32_to_cpu(p->inode),
197 rec_len, p->name_len);
200 p = (ext2_dirent *)(kaddr + offs);
201 CERROR("ext2_check_page"
202 "entry in directory #%lu spans the page boundary"
203 "offset=%lu, inode=%lu",
204 dir->i_ino, (page->index<<PAGE_CACHE_SHIFT)+offs,
205 (unsigned long) le32_to_cpu(p->inode));
207 SetPageChecked(page);
211 static struct page *ll_get_dir_page(struct inode *dir, unsigned long n)
213 struct ldlm_res_id res_id =
214 { .name = { dir->i_ino, (__u64)dir->i_generation} };
215 struct lustre_handle lockh;
216 struct obd_device *obddev = class_exp2obd(ll_i2sbi(dir)->ll_mdc_exp);
217 struct address_space *mapping = dir->i_mapping;
219 ldlm_policy_data_t policy = { .l_inodebits = { MDS_INODELOCK_UPDATE } };
222 obddev = md_get_real_obd(ll_i2sbi(dir)->ll_mdc_exp, NULL, 0);
223 rc = ldlm_lock_match(obddev->obd_namespace, LDLM_FL_BLOCK_GRANTED,
224 &res_id, LDLM_IBITS, &policy, LCK_PR, &lockh);
226 struct lookup_intent it = { .it_op = IT_READDIR };
227 struct ptlrpc_request *request;
228 struct mdc_op_data data;
230 ll_prepare_mdc_op_data(&data, dir, NULL, NULL, 0, 0);
232 rc = md_enqueue(ll_i2sbi(dir)->ll_mdc_exp, LDLM_IBITS, &it,
233 LCK_PR, &data, &lockh, NULL, 0,
234 ldlm_completion_ast, ll_mdc_blocking_ast, dir);
236 request = (struct ptlrpc_request *)it.d.lustre.it_data;
238 ptlrpc_req_finished(request);
240 CERROR("lock enqueue: rc: %d\n", rc);
244 ldlm_lock_dump_handle(D_OTHER, &lockh);
246 page = read_cache_page(mapping, n,
247 (filler_t*)mapping->a_ops->readpage, NULL);
251 if (!PageUptodate(page))
253 if (!PageChecked(page))
254 ext2_check_page(page);
260 ldlm_lock_decref(&lockh, LCK_PR);
265 page = ERR_PTR(-EIO);
270 * p is at least 6 bytes before the end of page
272 static inline ext2_dirent *ext2_next_entry(ext2_dirent *p)
274 return (ext2_dirent *)((char*)p + le16_to_cpu(p->rec_len));
277 static inline unsigned
278 ext2_validate_entry(char *base, unsigned offset, unsigned mask)
280 ext2_dirent *de = (ext2_dirent*)(base + offset);
281 ext2_dirent *p = (ext2_dirent*)(base + (offset&mask));
282 while ((char*)p < (char*)de)
283 p = ext2_next_entry(p);
284 return (char *)p - base;
287 static unsigned char ext2_filetype_table[EXT2_FT_MAX] = {
288 [EXT2_FT_UNKNOWN] DT_UNKNOWN,
289 [EXT2_FT_REG_FILE] DT_REG,
290 [EXT2_FT_DIR] DT_DIR,
291 [EXT2_FT_CHRDEV] DT_CHR,
292 [EXT2_FT_BLKDEV] DT_BLK,
293 [EXT2_FT_FIFO] DT_FIFO,
294 [EXT2_FT_SOCK] DT_SOCK,
295 [EXT2_FT_SYMLINK] DT_LNK,
299 int ll_readdir(struct file * filp, void * dirent, filldir_t filldir)
301 struct inode *inode = filp->f_dentry->d_inode;
302 loff_t pos = filp->f_pos;
303 // XXX struct super_block *sb = inode->i_sb;
304 unsigned offset = pos & ~PAGE_CACHE_MASK;
305 unsigned long n = pos >> PAGE_CACHE_SHIFT;
306 unsigned long npages = dir_pages(inode);
307 unsigned chunk_mask = ~(ext2_chunk_size(inode)-1);
308 unsigned char *types = NULL;
309 int need_revalidate = (filp->f_version != inode->i_version);
313 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %llu/%llu\n",
314 inode->i_ino, inode->i_generation, inode, pos, inode->i_size);
316 if (pos > inode->i_size - EXT2_DIR_REC_LEN(1))
319 types = ext2_filetype_table;
321 for ( ; n < npages; n++, offset = 0) {
326 CDEBUG(D_EXT2,"read %lu of dir %lu/%u page %lu/%lu size %llu\n",
327 PAGE_CACHE_SIZE, inode->i_ino, inode->i_generation,
328 n, npages, inode->i_size);
329 page = ll_get_dir_page(inode, n);
331 /* size might have been updated by md_readpage() */
332 npages = dir_pages(inode);
336 CERROR("error reading dir %lu/%u page %lu: rc %d\n",
337 inode->i_ino, inode->i_generation, n, rc);
341 kaddr = page_address(page);
342 if (need_revalidate) {
343 offset = ext2_validate_entry(kaddr, offset, chunk_mask);
346 de = (ext2_dirent *)(kaddr+offset);
347 limit = kaddr + PAGE_CACHE_SIZE - EXT2_DIR_REC_LEN(1);
348 for ( ;(char*)de <= limit; de = ext2_next_entry(de)) {
351 unsigned char d_type = DT_UNKNOWN;
354 rc = 0; /* no error if we return something */
355 if (types && de->file_type < EXT2_FT_MAX)
356 d_type = types[de->file_type];
358 offset = (char *)de - kaddr;
360 /*FIXME-WANGDI will get dot info from MDS*/
361 if ((n << PAGE_CACHE_SHIFT | offset) == 0) {
362 int off =n << PAGE_CACHE_SHIFT | offset;
363 over = filldir(dirent, DOT_SNAP_NAME,
364 strlen(DOT_SNAP_NAME),
371 offset+=EXT2_DIR_REC_LEN(strlen(DOT_SNAP_NAME));
373 over = filldir(dirent, de->name, de->name_len,
374 (n<<PAGE_CACHE_SHIFT) | offset,
375 le32_to_cpu(de->inode), d_type);
386 filp->f_pos = (n << PAGE_CACHE_SHIFT) | offset;
387 filp->f_version = inode->i_version;
392 static int ll_mkdir_stripe(struct inode *inode, unsigned long arg)
395 struct ll_user_mkdir_stripe lums;
396 struct ptlrpc_request *request = NULL;
397 struct ll_sb_info *sbi = ll_i2sbi(inode);
398 struct mdc_op_data op_data;
404 if (copy_from_user(&lums, (void *)arg, sizeof(lums)))
407 if (lums.lums_namelen <= 0)
409 OBD_ALLOC(name, lums.lums_namelen);
413 if (copy_from_user(name, lums.lums_name, lums.lums_namelen))
414 GOTO(out, err=-EFAULT);
416 CDEBUG(D_VFSTRACE, "ioctl Op:name=%s,dir=%lu/%u(%p)\n",
417 name, inode->i_ino, inode->i_generation, inode);
418 nstripes = lums.lums_nstripes;
420 mode = lums.lums_mode;
421 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
422 ll_prepare_mdc_op_data(&op_data, inode, NULL, name,lums.lums_namelen,0);
423 err = md_create(sbi->ll_mdc_exp, &op_data, &nstripes, sizeof(nstripes),
424 mode, current->fsuid, current->fsgid, 0, &request);
425 ptlrpc_req_finished(request);
428 OBD_FREE(name, lums.lums_namelen);
432 static int ll_dir_ioctl(struct inode *inode, struct file *file,
433 unsigned int cmd, unsigned long arg)
435 struct ll_sb_info *sbi = ll_i2sbi(inode);
436 struct obd_ioctl_data *data;
439 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
440 inode->i_ino, inode->i_generation, inode, cmd);
442 if (_IOC_TYPE(cmd) == 'T') /* tty ioctls */
445 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_IOCTL);
447 case EXT3_IOC_GETFLAGS:
448 case EXT3_IOC_SETFLAGS:
449 RETURN(ll_iocontrol(inode, file, cmd, arg));
450 case IOC_MDC_LOOKUP: {
451 struct ptlrpc_request *request = NULL;
455 int namelen, rc, len = 0;
458 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
463 filename = data->ioc_inlbuf1;
464 namelen = data->ioc_inllen1;
467 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
468 GOTO(out, rc = -EINVAL);
472 ll_inode2fid(&fid, inode);
473 rc = md_getattr_name(sbi->ll_mdc_exp, &fid,
474 filename, namelen, valid, 0, &request);
476 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
480 ptlrpc_req_finished(request);
484 obd_ioctl_freedata(buf, len);
487 case LL_IOC_MDC_MKDIRSTRIPE:
488 RETURN(ll_mkdir_stripe(inode, arg));
489 case IOC_MDC_FINISH_GNS:
490 RETURN(ll_finish_gns(sbi));
491 case LL_IOC_LOV_SETSTRIPE: {
492 struct ptlrpc_request *request = NULL;
493 struct mdc_op_data op_data;
494 struct iattr attr = { 0 };
495 struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
498 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
500 LASSERT(sizeof(lum) == sizeof(*lump));
501 LASSERT(sizeof(lum.lmm_objects[0]) ==
502 sizeof(lump->lmm_objects[0]));
503 rc = copy_from_user(&lum, lump, sizeof(lum));
507 if (lum.lmm_magic != LOV_USER_MAGIC)
510 rc = md_setattr(sbi->ll_mdc_exp, &op_data,
511 &attr, &lum, sizeof(lum), NULL, 0, &request);
513 ptlrpc_req_finished(request);
514 if (rc != -EPERM && rc != -EACCES)
515 CERROR("md_setattr fails: rc = %d\n", rc);
518 ptlrpc_req_finished(request);
522 case LL_IOC_LOV_GETSTRIPE: {
523 struct ptlrpc_request *request = NULL;
524 struct lov_user_md *lump = (struct lov_user_md *)arg;
525 struct lov_mds_md *lmm;
527 struct mds_body *body;
528 unsigned long valid = 0;
531 valid |= OBD_MD_FLDIREA;
533 ll_inode2fid(&fid, inode);
534 rc = md_getattr(sbi->ll_mdc_exp, &fid, valid,
535 obd_size_diskmd(sbi->ll_osc_exp, NULL),
538 CDEBUG(D_INFO, "md_getattr failed: rc = %d\n", rc);
542 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
543 LASSERT(body != NULL); /* checked by md_getattr_name */
544 LASSERT_REPSWABBED(request, 0);/* swabbed by md_getattr_name */
546 lmmsize = body->eadatasize;
548 GOTO(out_get, rc = -ENODATA);
550 lmm = lustre_msg_buf(request->rq_repmsg, 1, lmmsize);
551 LASSERT(lmm != NULL);
552 LASSERT_REPSWABBED(request, 1);
553 rc = copy_to_user(lump, lmm, lmmsize);
555 GOTO(out_get, rc = -EFAULT);
559 ptlrpc_req_finished(request);
562 case IOC_MDC_GETSTRIPE: {
563 struct ptlrpc_request *request = NULL;
565 struct mds_body *body;
566 struct lov_user_md *lump = (struct lov_user_md *)arg;
567 struct lov_mds_md *lmm;
571 filename = getname((const char *)arg);
572 if (IS_ERR(filename))
573 RETURN(PTR_ERR(filename));
575 ll_inode2fid(&fid, inode);
576 rc = md_getattr_name(sbi->ll_mdc_exp, &fid, filename,
577 strlen(filename) + 1, OBD_MD_FLEASIZE,
578 obd_size_diskmd(sbi->ll_osc_exp, NULL),
581 CDEBUG(D_INFO, "md_getattr_name failed on %s: rc %d\n",
586 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof (*body));
587 LASSERT(body != NULL); /* checked by md_getattr_name */
588 LASSERT_REPSWABBED(request, 0);/* swabbed by md_getattr_name */
590 lmmsize = body->eadatasize;
592 if (!(body->valid & OBD_MD_FLEASIZE) || lmmsize == 0)
593 GOTO(out_req, rc = -ENODATA);
596 GOTO(out_req, rc = -EFBIG);
598 lmm = lustre_msg_buf(request->rq_repmsg, 1, lmmsize);
599 LASSERT(lmm != NULL);
600 LASSERT_REPSWABBED(request, 1);
602 rc = copy_to_user(lump, lmm, lmmsize);
604 GOTO(out_req, rc = -EFAULT);
608 ptlrpc_req_finished(request);
614 struct ptlrpc_request *req = NULL;
617 struct client_obd *cli;
618 struct obd_device *obd;
620 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
625 obd = class_name2obd(data->ioc_inlbuf1);
628 GOTO(out_ping, rc = -ENODEV);
630 if (!obd->obd_attached) {
631 CERROR("Device %d not attached\n", obd->obd_minor);
632 GOTO(out_ping, rc = -ENODEV);
634 if (!obd->obd_set_up) {
635 CERROR("Device %d still not setup\n", obd->obd_minor);
636 GOTO(out_ping, rc = -ENODEV);
639 req = ptlrpc_prep_req(cli->cl_import, OBD_PING, 0, NULL, NULL);
641 GOTO(out_ping, rc = -ENOMEM);
643 req->rq_replen = lustre_msg_size(0, NULL);
644 req->rq_send_state = LUSTRE_IMP_FULL;
646 rc = ptlrpc_queue_wait(req);
648 ptlrpc_req_finished(req);
650 obd_ioctl_freedata(buf, len);
653 case OBD_IOC_LLOG_CATINFO: {
654 struct ptlrpc_request *req = NULL;
660 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
665 if (!data->ioc_inlbuf1) {
666 obd_ioctl_freedata(buf, len);
670 lens[0] = data->ioc_inllen1;
671 bufs[0] = data->ioc_inlbuf1;
672 if (data->ioc_inllen2) {
673 lens[1] = data->ioc_inllen2;
674 bufs[1] = data->ioc_inlbuf2;
679 size = data->ioc_plen1;
680 req = ptlrpc_prep_req(sbi2mdc(sbi)->cl_import, LLOG_CATINFO,
683 GOTO(out_catinfo, rc = -ENOMEM);
684 req->rq_replen = lustre_msg_size(1, &size);
686 rc = ptlrpc_queue_wait(req);
687 str = lustre_msg_string(req->rq_repmsg, 0, data->ioc_plen1);
689 rc = copy_to_user(data->ioc_pbuf1, str,
691 ptlrpc_req_finished(req);
693 obd_ioctl_freedata(buf, len);
697 return obd_iocontrol(cmd, sbi->ll_osc_exp,0,NULL,(void *)arg);
701 int ll_dir_open(struct inode *inode, struct file *file)
704 RETURN(ll_file_open(inode, file));
707 int ll_dir_release(struct inode *inode, struct file *file)
710 RETURN(ll_file_release(inode, file));
713 struct file_operations ll_dir_operations = {
715 .release = ll_dir_release,
716 .read = generic_read_dir,
717 .readdir = ll_readdir,
718 .ioctl = ll_dir_ioctl