fs/nfs/pagelist.c

   1 /*
   2  * linux/fs/nfs/pagelist.c
   3  *
   4  * A set of helper functions for managing NFS read and write requests.
   5  * The main purpose of these routines is to provide support for the
   6  * coalescing of several requests into a single RPC call.
   7  *
   8  * Copyright 2000, 2001 (c) Trond Myklebust <trond.myklebust@fys.uio.no>
   9  *
  10  */
  11
  12 #include <linux/slab.h>
  13 #include <linux/file.h>
  14 #include <linux/sched.h>
  15 #include <linux/sunrpc/clnt.h>
  16 #include <linux/nfs.h>
  17 #include <linux/nfs3.h>
  18 #include <linux/nfs4.h>
  19 #include <linux/nfs_page.h>
  20 #include <linux/nfs_fs.h>
  21 #include <linux/nfs_mount.h>
  22 #include <linux/export.h>
  23
  24 #include "internal.h"
  25 #include "pnfs.h"
  26
  27 static struct kmem_cache *nfs_page_cachep;
  28
  29 bool nfs_pgarray_set(struct nfs_page_array *p, unsigned int pagecount)
  30 {
  31         p->npages = pagecount;
  32         if (pagecount <= ARRAY_SIZE(p->page_array))
  33                 p->pagevec = p->page_array;
  34         else {
  35                 p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL);
  36                 if (!p->pagevec)
  37                         p->npages = 0;
  38         }
  39         return p->pagevec != NULL;
  40 }
  41
  42 void nfs_pgheader_init(struct nfs_pageio_descriptor *desc,
  43                        struct nfs_pgio_header *hdr,
  44                        void (*release)(struct nfs_pgio_header *hdr))
  45 {
  46         hdr->req = nfs_list_entry(desc->pg_list.next);
  47         hdr->inode = desc->pg_inode;
  48         hdr->cred = hdr->req->wb_context->cred;
  49         hdr->io_start = req_offset(hdr->req);
  50         hdr->good_bytes = desc->pg_count;
  51         hdr->dreq = desc->pg_dreq;
  52         hdr->layout_private = desc->pg_layout_private;
  53         hdr->release = release;
  54         hdr->completion_ops = desc->pg_completion_ops;
  55         if (hdr->completion_ops->init_hdr)
  56                 hdr->completion_ops->init_hdr(hdr);
  57 }
  58 EXPORT_SYMBOL_GPL(nfs_pgheader_init);
  59
  60 void nfs_set_pgio_error(struct nfs_pgio_header *hdr, int error, loff_t pos)
  61 {
  62         spin_lock(&hdr->lock);
  63         if (pos < hdr->io_start + hdr->good_bytes) {
  64                 set_bit(NFS_IOHDR_ERROR, &hdr->flags);
  65                 clear_bit(NFS_IOHDR_EOF, &hdr->flags);
  66                 hdr->good_bytes = pos - hdr->io_start;
  67                 hdr->error = error;
  68         }
  69         spin_unlock(&hdr->lock);
  70 }
  71
  72 static inline struct nfs_page *
  73 nfs_page_alloc(void)
  74 {
  75         struct nfs_page *p = kmem_cache_zalloc(nfs_page_cachep, GFP_NOIO);
  76         if (p)
  77                 INIT_LIST_HEAD(&p->wb_list);
  78         return p;
  79 }
  80
  81 static inline void
  82 nfs_page_free(struct nfs_page *p)
  83 {
  84         kmem_cache_free(nfs_page_cachep, p);
  85 }
  86
  87 /**
  88  * nfs_create_request - Create an NFS read/write request.
  89  * @ctx: open context to use
  90  * @inode: inode to which the request is attached
  91  * @page: page to write
  92  * @offset: starting offset within the page for the write
  93  * @count: number of bytes to read/write
  94  *
  95  * The page must be locked by the caller. This makes sure we never
  96  * create two different requests for the same page.
  97  * User should ensure it is safe to sleep in this function.
  98  */
  99 struct nfs_page *
 100 nfs_create_request(struct nfs_open_context *ctx, struct inode *inode,
 101                    struct page *page,
 102                    unsigned int offset, unsigned int count)
 103 {
 104         struct nfs_page         *req;
 105
 106         /* try to allocate the request struct */
 107         req = nfs_page_alloc();
 108         if (req == NULL)
 109                 return ERR_PTR(-ENOMEM);
 110
 111         /* get lock context early so we can deal with alloc failures */
 112         req->wb_lock_context = nfs_get_lock_context(ctx);
 113         if (req->wb_lock_context == NULL) {
 114                 nfs_page_free(req);
 115                 return ERR_PTR(-ENOMEM);
 116         }
 117
 118         /* Initialize the request struct. Initially, we assume a
 119          * long write-back delay. This will be adjusted in
 120          * update_nfs_request below if the region is not locked. */
 121         req->wb_page    = page;
 122         req->wb_index   = page_file_index(page);
 123         page_cache_get(page);
 124         req->wb_offset  = offset;
 125         req->wb_pgbase  = offset;
 126         req->wb_bytes   = count;
 127         req->wb_context = get_nfs_open_context(ctx);
 128         kref_init(&req->wb_kref);
 129         return req;
 130 }
 131
 132 /**
 133  * nfs_unlock_request - Unlock request and wake up sleepers.
 134  * @req:
 135  */
 136 void nfs_unlock_request(struct nfs_page *req)
 137 {
 138         if (!NFS_WBACK_BUSY(req)) {
 139                 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
 140                 BUG();
 141         }
 142         smp_mb__before_clear_bit();
 143         clear_bit(PG_BUSY, &req->wb_flags);
 144         smp_mb__after_clear_bit();
 145         wake_up_bit(&req->wb_flags, PG_BUSY);
 146 }
 147
 148 /**
 149  * nfs_unlock_and_release_request - Unlock request and release the nfs_page
 150  * @req:
 151  */
 152 void nfs_unlock_and_release_request(struct nfs_page *req)
 153 {
 154         nfs_unlock_request(req);
 155         nfs_release_request(req);
 156 }
 157
 158 /*
 159  * nfs_clear_request - Free up all resources allocated to the request
 160  * @req:
 161  *
 162  * Release page and open context resources associated with a read/write
 163  * request after it has completed.
 164  */
 165 static void nfs_clear_request(struct nfs_page *req)
 166 {
 167         struct page *page = req->wb_page;
 168         struct nfs_open_context *ctx = req->wb_context;
 169         struct nfs_lock_context *l_ctx = req->wb_lock_context;
 170
 171         if (page != NULL) {
 172                 page_cache_release(page);
 173                 req->wb_page = NULL;
 174         }
 175         if (l_ctx != NULL) {
 176                 nfs_put_lock_context(l_ctx);
 177                 req->wb_lock_context = NULL;
 178         }
 179         if (ctx != NULL) {
 180                 put_nfs_open_context(ctx);
 181                 req->wb_context = NULL;
 182         }
 183 }
 184
 185
 186 /**
 187  * nfs_release_request - Release the count on an NFS read/write request
 188  * @req: request to release
 189  *
 190  * Note: Should never be called with the spinlock held!
 191  */
 192 static void nfs_free_request(struct kref *kref)
 193 {
 194         struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
 195
 196         /* Release struct file and open context */
 197         nfs_clear_request(req);
 198         nfs_page_free(req);
 199 }
 200
 201 void nfs_release_request(struct nfs_page *req)
 202 {
 203         kref_put(&req->wb_kref, nfs_free_request);
 204 }
 205
 206 static int nfs_wait_bit_uninterruptible(void *word)
 207 {
 208         io_schedule();
 209         return 0;
 210 }
 211
 212 /**
 213  * nfs_wait_on_request - Wait for a request to complete.
 214  * @req: request to wait upon.
 215  *
 216  * Interruptible by fatal signals only.
 217  * The user is responsible for holding a count on the request.
 218  */
 219 int
 220 nfs_wait_on_request(struct nfs_page *req)
 221 {
 222         return wait_on_bit(&req->wb_flags, PG_BUSY,
 223                         nfs_wait_bit_uninterruptible,
 224                         TASK_UNINTERRUPTIBLE);
 225 }
 226
 227 bool nfs_generic_pg_test(struct nfs_pageio_descriptor *desc, struct nfs_page *prev, struct nfs_page *req)
 228 {
 229         /*
 230          * FIXME: ideally we should be able to coalesce all requests
 231          * that are not block boundary aligned, but currently this
 232          * is problematic for the case of bsize < PAGE_CACHE_SIZE,
 233          * since nfs_flush_multi and nfs_pagein_multi assume you
 234          * can have only one struct nfs_page.
 235          */
 236         if (desc->pg_bsize < PAGE_SIZE)
 237                 return 0;
 238
 239         return desc->pg_count + req->wb_bytes <= desc->pg_bsize;
 240 }
 241 EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
 242
 243 /**
 244  * nfs_pageio_init - initialise a page io descriptor
 245  * @desc: pointer to descriptor
 246  * @inode: pointer to inode
 247  * @doio: pointer to io function
 248  * @bsize: io block size
 249  * @io_flags: extra parameters for the io function
 250  */
 251 void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
 252                      struct inode *inode,
 253                      const struct nfs_pageio_ops *pg_ops,
 254                      const struct nfs_pgio_completion_ops *compl_ops,
 255                      size_t bsize,
 256                      int io_flags)
 257 {
 258         INIT_LIST_HEAD(&desc->pg_list);
 259         desc->pg_bytes_written = 0;
 260         desc->pg_count = 0;
 261         desc->pg_bsize = bsize;
 262         desc->pg_base = 0;
 263         desc->pg_moreio = 0;
 264         desc->pg_recoalesce = 0;
 265         desc->pg_inode = inode;
 266         desc->pg_ops = pg_ops;
 267         desc->pg_completion_ops = compl_ops;
 268         desc->pg_ioflags = io_flags;
 269         desc->pg_error = 0;
 270         desc->pg_lseg = NULL;
 271         desc->pg_dreq = NULL;
 272         desc->pg_layout_private = NULL;
 273 }
 274 EXPORT_SYMBOL_GPL(nfs_pageio_init);
 275
 276 /**
 277  * nfs_can_coalesce_requests - test two requests for compatibility
 278  * @prev: pointer to nfs_page
 279  * @req: pointer to nfs_page
 280  *
 281  * The nfs_page structures 'prev' and 'req' are compared to ensure that the
 282  * page data area they describe is contiguous, and that their RPC
 283  * credentials, NFSv4 open state, and lockowners are the same.
 284  *
 285  * Return 'true' if this is the case, else return 'false'.
 286  */
 287 static bool nfs_can_coalesce_requests(struct nfs_page *prev,
 288                                       struct nfs_page *req,
 289                                       struct nfs_pageio_descriptor *pgio)
 290 {
 291         if (req->wb_context->cred != prev->wb_context->cred)
 292                 return false;
 293         if (req->wb_lock_context->lockowner != prev->wb_lock_context->lockowner)
 294                 return false;
 295         if (req->wb_context->state != prev->wb_context->state)
 296                 return false;
 297         if (req->wb_pgbase != 0)
 298                 return false;
 299         if (prev->wb_pgbase + prev->wb_bytes != PAGE_CACHE_SIZE)
 300                 return false;
 301         if (req_offset(req) != req_offset(prev) + prev->wb_bytes)
 302                 return false;
 303         return pgio->pg_ops->pg_test(pgio, prev, req);
 304 }
 305
 306 /**
 307  * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
 308  * @desc: destination io descriptor
 309  * @req: request
 310  *
 311  * Returns true if the request 'req' was successfully coalesced into the
 312  * existing list of pages 'desc'.
 313  */
 314 static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
 315                                      struct nfs_page *req)
 316 {
 317         if (desc->pg_count != 0) {
 318                 struct nfs_page *prev;
 319
 320                 prev = nfs_list_entry(desc->pg_list.prev);
 321                 if (!nfs_can_coalesce_requests(prev, req, desc))
 322                         return 0;
 323         } else {
 324                 if (desc->pg_ops->pg_init)
 325                         desc->pg_ops->pg_init(desc, req);
 326                 desc->pg_base = req->wb_pgbase;
 327         }
 328         nfs_list_remove_request(req);
 329         nfs_list_add_request(req, &desc->pg_list);
 330         desc->pg_count += req->wb_bytes;
 331         return 1;
 332 }
 333
 334 /*
 335  * Helper for nfs_pageio_add_request and nfs_pageio_complete
 336  */
 337 static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
 338 {
 339         if (!list_empty(&desc->pg_list)) {
 340                 int error = desc->pg_ops->pg_doio(desc);
 341                 if (error < 0)
 342                         desc->pg_error = error;
 343                 else
 344                         desc->pg_bytes_written += desc->pg_count;
 345         }
 346         if (list_empty(&desc->pg_list)) {
 347                 desc->pg_count = 0;
 348                 desc->pg_base = 0;
 349         }
 350 }
 351
 352 /**
 353  * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
 354  * @desc: destination io descriptor
 355  * @req: request
 356  *
 357  * Returns true if the request 'req' was successfully coalesced into the
 358  * existing list of pages 'desc'.
 359  */
 360 static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
 361                            struct nfs_page *req)
 362 {
 363         while (!nfs_pageio_do_add_request(desc, req)) {
 364                 desc->pg_moreio = 1;
 365                 nfs_pageio_doio(desc);
 366                 if (desc->pg_error < 0)
 367                         return 0;
 368                 desc->pg_moreio = 0;
 369                 if (desc->pg_recoalesce)
 370                         return 0;
 371         }
 372         return 1;
 373 }
 374
 375 static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
 376 {
 377         LIST_HEAD(head);
 378
 379         do {
 380                 list_splice_init(&desc->pg_list, &head);
 381                 desc->pg_bytes_written -= desc->pg_count;
 382                 desc->pg_count = 0;
 383                 desc->pg_base = 0;
 384                 desc->pg_recoalesce = 0;
 385
 386                 while (!list_empty(&head)) {
 387                         struct nfs_page *req;
 388
 389                         req = list_first_entry(&head, struct nfs_page, wb_list);
 390                         nfs_list_remove_request(req);
 391                         if (__nfs_pageio_add_request(desc, req))
 392                                 continue;
 393                         if (desc->pg_error < 0)
 394                                 return 0;
 395                         break;
 396                 }
 397         } while (desc->pg_recoalesce);
 398         return 1;
 399 }
 400
 401 int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
 402                 struct nfs_page *req)
 403 {
 404         int ret;
 405
 406         do {
 407                 ret = __nfs_pageio_add_request(desc, req);
 408                 if (ret)
 409                         break;
 410                 if (desc->pg_error < 0)
 411                         break;
 412                 ret = nfs_do_recoalesce(desc);
 413         } while (ret);
 414         return ret;
 415 }
 416 EXPORT_SYMBOL_GPL(nfs_pageio_add_request);
 417
 418 /**
 419  * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor
 420  * @desc: pointer to io descriptor
 421  */
 422 void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
 423 {
 424         for (;;) {
 425                 nfs_pageio_doio(desc);
 426                 if (!desc->pg_recoalesce)
 427                         break;
 428                 if (!nfs_do_recoalesce(desc))
 429                         break;
 430         }
 431 }
 432 EXPORT_SYMBOL_GPL(nfs_pageio_complete);
 433
 434 /**
 435  * nfs_pageio_cond_complete - Conditional I/O completion
 436  * @desc: pointer to io descriptor
 437  * @index: page index
 438  *
 439  * It is important to ensure that processes don't try to take locks
 440  * on non-contiguous ranges of pages as that might deadlock. This
 441  * function should be called before attempting to wait on a locked
 442  * nfs_page. It will complete the I/O if the page index 'index'
 443  * is not contiguous with the existing list of pages in 'desc'.
 444  */
 445 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
 446 {
 447         if (!list_empty(&desc->pg_list)) {
 448                 struct nfs_page *prev = nfs_list_entry(desc->pg_list.prev);
 449                 if (index != prev->wb_index + 1)
 450                         nfs_pageio_complete(desc);
 451         }
 452 }
 453
 454 int __init nfs_init_nfspagecache(void)
 455 {
 456         nfs_page_cachep = kmem_cache_create("nfs_page",
 457                                             sizeof(struct nfs_page),
 458                                             0, SLAB_HWCACHE_ALIGN,
 459                                             NULL);
 460         if (nfs_page_cachep == NULL)
 461                 return -ENOMEM;
 462
 463         return 0;
 464 }
 465
 466 void nfs_destroy_nfspagecache(void)
 467 {
 468         kmem_cache_destroy(nfs_page_cachep);
 469 }
 470