lustre/obdfs/rw.c

   1 /*
   2  * OBDFS Super operations
   3  *
   4  * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de>
   5  * Copryright (C) 1999 Stelias Computing Inc,
   6  *                (author Peter J. Braam <braam@stelias.com>)
   7  * Copryright (C) 1999 Seagate Technology Inc.
   8 */
   9
  10
  11 #include <linux/config.h>
  12 #include <linux/kernel.h>
  13 #include <linux/mm.h>
  14 #include <linux/string.h>
  15 #include <linux/stat.h>
  16 #include <linux/errno.h>
  17 #include <linux/locks.h>
  18 #include <linux/unistd.h>
  19
  20 #include <asm/system.h>
  21 #include <asm/uaccess.h>
  22
  23 #include <linux/fs.h>
  24 #include <linux/stat.h>
  25 #include <asm/uaccess.h>
  26 #include <linux/vmalloc.h>
  27 #include <asm/segment.h>
  28 #include <linux/mm.h>
  29 #include <linux/pagemap.h>
  30 #include <linux/smp_lock.h>
  31
  32 #include <linux/obd_support.h>
  33 #include <linux/obd_ext2.h>
  34 #include <linux/obdfs.h>
  35
  36
  37 /* SYNCHRONOUS I/O for an inode */
  38 static int obdfs_brw(int rw, struct inode *inode, struct page *page, int create)
  39 {
  40         obd_count        num_obdo = 1;
  41         obd_count        bufs_per_obdo = 1;
  42         struct obdo     *oa;
  43         char            *buf = (char *)page_address(page);
  44         obd_size         count = PAGE_SIZE;
  45         obd_off          offset = ((obd_off)page->index) << PAGE_SHIFT;
  46         obd_flag         flags = create ? OBD_BRW_CREATE : 0;
  47         int              err;
  48
  49         ENTRY;
  50         oa = obdo_fromid(IID(inode), inode->i_ino, OBD_MD_FLNOTOBD);
  51         if ( IS_ERR(oa) ) {
  52                 EXIT;
  53                 return PTR_ERR(oa);
  54         }
  55         obdfs_from_inode(oa, inode);
  56
  57         err = IOPS(inode, brw)(rw, IID(inode), num_obdo, &oa, &bufs_per_obdo,
  58                                &buf, &count, &offset, &flags);
  59
  60         if ( !err )
  61                 obdfs_to_inode(inode, oa); /* copy o_blocks to i_blocks */
  62
  63         obdo_free(oa);
  64
  65         EXIT;
  66         return err;
  67 } /* obdfs_brw */
  68
  69 /* returns the page unlocked, but with a reference */
  70 int obdfs_readpage(struct dentry *dentry, struct page *page)
  71 {
  72         struct inode *inode = dentry->d_inode;
  73         int rc;
  74
  75         ENTRY;
  76         PDEBUG(page, "READ");
  77         rc = obdfs_brw(READ, inode, page, 0);
  78         if ( !rc ) {
  79                 SetPageUptodate(page);
  80                 UnlockPage(page);
  81         }
  82         PDEBUG(page, "READ");
  83         EXIT;
  84         return rc;
  85 } /* obdfs_readpage */
  86
  87 static kmem_cache_t *obdfs_pgrq_cachep = NULL;
  88
  89 /* XXX should probably have one of these per superblock */
  90 static int obdfs_cache_count = 0;
  91
  92 int obdfs_init_pgrqcache(void)
  93 {
  94         ENTRY;
  95         if (obdfs_pgrq_cachep == NULL) {
  96                 CDEBUG(D_CACHE, "allocating obdfs_pgrq_cache\n");
  97                 obdfs_pgrq_cachep = kmem_cache_create("obdfs_pgrq",
  98                                                       sizeof(struct obdfs_pgrq),
  99                                                       0, SLAB_HWCACHE_ALIGN,
 100                                                       NULL, NULL);
 101                 if (obdfs_pgrq_cachep == NULL) {
 102                         EXIT;
 103                         return -ENOMEM;
 104                 } else {
 105                         CDEBUG(D_CACHE, "allocated cache at %p\n",
 106                                obdfs_pgrq_cachep);
 107                 }
 108         } else {
 109                 CDEBUG(D_CACHE, "using existing cache at %p\n",
 110                        obdfs_pgrq_cachep);
 111         }
 112         EXIT;
 113         return 0;
 114 } /* obdfs_init_wreqcache */
 115
 116 inline void obdfs_pgrq_del(struct obdfs_pgrq *pgrq)
 117 {
 118         obdfs_cache_count--;
 119         CDEBUG(D_INFO, "deleting page %p from list [count %d]\n",
 120                pgrq->rq_page, obdfs_cache_count);
 121         list_del(&pgrq->rq_plist);
 122         kmem_cache_free(obdfs_pgrq_cachep, pgrq);
 123 }
 124
 125 void obdfs_cleanup_pgrqcache(void)
 126 {
 127         ENTRY;
 128         if (obdfs_pgrq_cachep != NULL) {
 129                 CDEBUG(D_CACHE, "destroying obdfs_pgrqcache at %p, count %d\n",
 130                        obdfs_pgrq_cachep, obdfs_cache_count);
 131                 if (kmem_cache_destroy(obdfs_pgrq_cachep))
 132                         printk(KERN_INFO __FUNCTION__
 133                                ": unable to free all of cache\n");
 134                 obdfs_pgrq_cachep = NULL;
 135         } else
 136                 printk(KERN_INFO __FUNCTION__ ": called with NULL pointer\n");
 137
 138         EXIT;
 139 } /* obdfs_cleanup_wreqcache */
 140
 141
 142 /*
 143  * Find a specific page in the page cache.  If it is found, we return
 144  * the write request struct associated with it, if not found return NULL.
 145  * Called with the list lock held.
 146  */
 147 static struct obdfs_pgrq *
 148 obdfs_find_in_page_list(struct inode *inode, struct page *page)
 149 {
 150         struct list_head *page_list = obdfs_iplist(inode);
 151         struct list_head *tmp;
 152
 153         ENTRY;
 154
 155         CDEBUG(D_INFO, "looking for inode %ld page %p\n", inode->i_ino, page);
 156         OIDEBUG(inode);
 157
 158         if (list_empty(page_list)) {
 159                 CDEBUG(D_INFO, "empty list\n");
 160                 EXIT;
 161                 return NULL;
 162         }
 163         tmp = page_list;
 164         while ( (tmp = tmp->next) != page_list ) {
 165                 struct obdfs_pgrq *pgrq;
 166
 167                 pgrq = list_entry(tmp, struct obdfs_pgrq, rq_plist);
 168                 if (pgrq->rq_page == page) {
 169                         CDEBUG(D_INFO, "found page %p in list\n", page);
 170                         EXIT;
 171                         return pgrq;
 172                 }
 173         }
 174
 175         EXIT;
 176         return NULL;
 177 } /* obdfs_find_in_page_list */
 178
 179
 180 /* called with the list lock held */
 181 static struct page* obdfs_find_page_index(struct inode *inode,
 182                                           unsigned long index)
 183 {
 184         struct list_head *page_list = obdfs_iplist(inode);
 185         struct list_head *tmp;
 186         struct page *page;
 187
 188         ENTRY;
 189
 190         CDEBUG(D_INFO, "looking for inode %ld pageindex %ld\n",
 191                inode->i_ino, index);
 192         OIDEBUG(inode);
 193
 194         if (list_empty(page_list)) {
 195                 EXIT;
 196                 return NULL;
 197         }
 198         tmp = page_list;
 199         while ( (tmp = tmp->next) != page_list ) {
 200                 struct obdfs_pgrq *pgrq;
 201
 202                 pgrq = list_entry(tmp, struct obdfs_pgrq, rq_plist);
 203                 page = pgrq->rq_page;
 204                 if (index == page->index) {
 205                         CDEBUG(D_INFO,
 206                                "INDEX SEARCH found page %p, index %ld\n",
 207                                page, index);
 208                         EXIT;
 209                         return page;
 210                 }
 211         }
 212
 213         EXIT;
 214         return NULL;
 215 } /* obdfs_find_page_index */
 216
 217
 218 /* call and free pages from Linux page cache: called with io lock on inodes */
 219 int obdfs_do_vec_wr(struct inode **inodes, obd_count num_io,
 220                     obd_count num_obdos, struct obdo **obdos,
 221                     obd_count *oa_bufs, struct page **pages, char **bufs,
 222                     obd_size *counts, obd_off *offsets, obd_flag *flags)
 223 {
 224         struct super_block *sb = inodes[0]->i_sb;
 225         struct obdfs_sb_info *sbi = (struct obdfs_sb_info *)&sb->u.generic_sbp;
 226         int err;
 227
 228         ENTRY;
 229         CDEBUG(D_INFO, "writing %d page(s), %d obdo(s) in vector\n",
 230                num_io, num_obdos);
 231         err = OPS(sb, brw)(WRITE, &sbi->osi_conn, num_obdos, obdos, oa_bufs,
 232                                 bufs, counts, offsets, flags);
 233
 234         /* release the pages from the page cache */
 235         while ( num_io > 0 ) {
 236                 num_io--;
 237                 CDEBUG(D_INFO, "calling put_page for %p, index %ld\n",
 238                        pages[num_io], pages[num_io]->index);
 239                 put_page(pages[num_io]);
 240         }
 241
 242         while ( num_obdos > 0) {
 243                 num_obdos--;
 244                 CDEBUG(D_INFO, "copy/free obdo %ld\n",
 245                        (long)obdos[num_obdos]->o_id);
 246                 obdfs_to_inode(inodes[num_obdos], obdos[num_obdos]);
 247                 obdo_free(obdos[num_obdos]);
 248         }
 249         EXIT;
 250         return err;
 251 }
 252
 253
 254 /*
 255  * Add a page to the write request cache list for later writing
 256  * ASYNCHRONOUS write method.
 257  */
 258 static int obdfs_add_page_to_cache(struct inode *inode, struct page *page)
 259 {
 260         int res = 0;
 261
 262         ENTRY;
 263
 264         /* If this page isn't already in the inode page list, add it */
 265         obd_down(&obdfs_i2sbi(inode)->osi_list_mutex);
 266         if ( !obdfs_find_in_page_list(inode, page) ) {
 267                 struct obdfs_pgrq *pgrq;
 268                 pgrq = kmem_cache_alloc(obdfs_pgrq_cachep, SLAB_KERNEL);
 269                 CDEBUG(D_INFO,
 270                        "adding inode %ld page %p, pgrq: %p, cache count [%d]\n",
 271                        inode->i_ino, page, pgrq, obdfs_cache_count + 1);
 272                 if (!pgrq) {
 273                         EXIT;
 274                         obd_up(&obdfs_i2sbi(inode)->osi_list_mutex);
 275                         return -ENOMEM;
 276                 }
 277                 memset(pgrq, 0, sizeof(*pgrq));
 278
 279                 pgrq->rq_page = page;
 280                 pgrq->rq_jiffies = jiffies;
 281                 get_page(pgrq->rq_page);
 282                 list_add(&pgrq->rq_plist, obdfs_iplist(inode));
 283                 obdfs_cache_count++;
 284         }
 285
 286         /* If inode isn't already on the superblock inodes list, add it,
 287          * and increase ref count on inode so it doesn't disappear on us.
 288          *
 289          * We increment the reference count on the inode to keep it from
 290          * being freed from memory.  This _should_ be an iget() with an
 291          * iput() in both flush_reqs() and put_inode(), but since ut_inode()
 292          * is called from iput() we can't call iput() again there.  Instead
 293          * we just increment/decrement i_count, which is essentially what
 294          * iget/iput do for an inode already in memory.
 295          */
 296         if ( list_empty(obdfs_islist(inode)) ) {
 297                 inode->i_count++;
 298                 CDEBUG(D_INFO, "adding inode %ld to superblock list %p\n",
 299                        inode->i_ino, obdfs_slist(inode));
 300                 list_add(obdfs_islist(inode), obdfs_slist(inode));
 301         }
 302
 303         /* XXX For testing purposes, we write out the page here.
 304          *     In the future, a flush daemon will write out the page.
 305         res = obdfs_flush_reqs(obdfs_slist(inode), 0);
 306         obdfs_flush_dirty_pages(1);
 307          */
 308         obd_up(&obdfs_i2sbi(inode)->osi_list_mutex);
 309
 310         EXIT;
 311         return res;
 312 } /* obdfs_add_page_to_cache */
 313
 314
 315 /* select between SYNC and ASYNC I/O methods */
 316 int obdfs_do_writepage(struct inode *inode, struct page *page, int sync)
 317 {
 318         int err;
 319
 320         ENTRY;
 321         /* PDEBUG(page, "WRITEPAGE"); */
 322         if ( sync )
 323                 err = obdfs_brw(WRITE, inode, page, 1);
 324         else {
 325                 err = obdfs_add_page_to_cache(inode, page);
 326                 CDEBUG(D_INFO, "DO_WR ino: %ld, page %p, err %d, uptodate %d\n",
 327                        inode->i_ino, page, err, Page_Uptodate(page));
 328         }
 329
 330         if ( !err )
 331                 SetPageUptodate(page);
 332         /* PDEBUG(page,"WRITEPAGE"); */
 333         EXIT;
 334         return err;
 335 } /* obdfs_do_writepage */
 336
 337 /* returns the page unlocked, but with a reference */
 338 int obdfs_writepage(struct dentry *dentry, struct page *page)
 339 {
 340         return obdfs_do_writepage(dentry->d_inode, page, 0);
 341 }
 342
 343 /*
 344  * This does the "real" work of the write. The generic routine has
 345  * allocated the page, locked it, done all the page alignment stuff
 346  * calculations etc. Now we should just copy the data from user
 347  * space and write it back to the real medium..
 348  *
 349  * If the writer ends up delaying the write, the writer needs to
 350  * increment the page use counts until he is done with the page.
 351  *
 352  * Return value is the number of bytes written.
 353  */
 354 int obdfs_write_one_page(struct file *file, struct page *page,
 355                           unsigned long offset, unsigned long bytes,
 356                           const char * buf)
 357 {
 358         struct inode *inode = file->f_dentry->d_inode;
 359         int err;
 360
 361         ENTRY;
 362         if ( !Page_Uptodate(page) ) {
 363                 err = obdfs_brw(READ, inode, page, 1);
 364                 if ( !err )
 365                         SetPageUptodate(page);
 366                 else
 367                         return err;
 368         }
 369
 370         if (copy_from_user((u8*)page_address(page) + offset, buf, bytes))
 371                 return -EFAULT;
 372
 373         lock_kernel();
 374         err = obdfs_writepage(file->f_dentry, page);
 375         unlock_kernel();
 376
 377         return (err < 0 ? err : bytes);
 378 } /* obdfs_write_one_page */
 379
 380 /*
 381  * return an up to date page:
 382  *  - if locked is true then is returned locked
 383  *  - if create is true the corresponding disk blocks are created
 384  *  - page is held, i.e. caller must release the page
 385  *
 386  * modeled on NFS code.
 387  */
 388 struct page *obdfs_getpage(struct inode *inode, unsigned long offset,
 389                            int create, int locked)
 390 {
 391         struct page *page_cache;
 392         int index;
 393         struct page ** hash;
 394         struct page * page;
 395         int err;
 396
 397         ENTRY;
 398
 399         offset = offset & PAGE_CACHE_MASK;
 400         CDEBUG(D_INFO, "ino: %ld, offset %ld, create %d, locked %d\n",
 401                inode->i_ino, offset, create, locked);
 402         index = offset >> PAGE_CACHE_SHIFT;
 403
 404
 405         page = NULL;
 406         page_cache = page_cache_alloc();
 407         if ( ! page_cache ) {
 408                 EXIT;
 409                 return NULL;
 410         }
 411         CDEBUG(D_INFO, "page_cache %p\n", page_cache);
 412
 413         hash = page_hash(&inode->i_data, index);
 414         page = grab_cache_page(&inode->i_data, index);
 415
 416         /* Yuck, no page */
 417         if (! page) {
 418             printk(KERN_WARNING " grab_cache_page says no dice ...\n");
 419             EXIT;
 420             return 0;
 421         }
 422
 423         PDEBUG(page, "GETPAGE: got page - before reading\n");
 424         /* now check if the data in the page is up to date */
 425         if ( Page_Uptodate(page)) {
 426                 if (!locked)
 427                         UnlockPage(page);
 428                 EXIT;
 429                 return page;
 430         }
 431
 432
 433         if ( obdfs_find_page_index(inode, index) ) {
 434                 CDEBUG(D_INFO, "OVERWRITE: found dirty page %p, index %ld\n",
 435                        page, page->index);
 436         }
 437
 438         err = obdfs_brw(READ, inode, page, create);
 439
 440         if ( err ) {
 441                 SetPageError(page);
 442                 UnlockPage(page);
 443                 EXIT;
 444                 return page;
 445         }
 446
 447         if ( !locked )
 448                 UnlockPage(page);
 449         SetPageUptodate(page);
 450         PDEBUG(page,"GETPAGE - after reading");
 451         EXIT;
 452         return page;
 453 } /* obdfs_getpage */
 454
 455